mirror of https://github.com/vladmandic/human
8047 lines
1.5 MiB
8047 lines
1.5 MiB
/*
|
|
Human
|
|
homepage: <https://github.com/vladmandic/human>
|
|
author: <https://github.com/vladmandic>'
|
|
*/
|
|
|
|
var Human=(()=>{var Kg=Object.defineProperty;var LN=(e,t,n)=>t in e?Kg(e,t,{enumerable:!0,configurable:!0,writable:!0,value:n}):e[t]=n;var BN=e=>Kg(e,"__esModule",{value:!0});var va=(e=>typeof require!="undefined"?require:typeof Proxy!="undefined"?new Proxy(e,{get:(t,n)=>(typeof require!="undefined"?require:t)[n]}):e)(function(e){if(typeof require!="undefined")return require.apply(this,arguments);throw new Error('Dynamic require of "'+e+'" is not supported')});var Gc=(e,t)=>{BN(e);for(var n in t)Kg(e,n,{get:t[n],enumerable:!0})};var ce=(e,t,n)=>(LN(e,typeof t!="symbol"?t+"":t,n),n),m5=(e,t,n)=>{if(!t.has(e))throw TypeError("Cannot "+n)};var Hc=(e,t,n)=>(m5(e,t,"read from private field"),n?n.call(e):t.get(e)),jc=(e,t,n)=>{if(t.has(e))throw TypeError("Cannot add the same private member more than once");t instanceof WeakSet?t.add(e):t.set(e,n)},qc=(e,t,n,s)=>(m5(e,t,"write to private field"),s?s.call(e,n):t.set(e,n),n);var w2e={};Gc(w2e,{Human:()=>WT,default:()=>WT,defaults:()=>wa,env:()=>Ae});function nt(e,t){let n=e.endsWith("/")?"":"/",r=t.startsWith(".")||t.startsWith("/")||t.startsWith("http:")||t.startsWith("https:")||t.startsWith("file:")?`${t}`:`${e}${n}${t}`;if(!r.toLocaleLowerCase().includes(".json"))throw new Error(`modelpath error: ${r} expecting json file`);return r}function re(...e){let t=new Date,n=`${t.getHours().toString().padStart(2,"0")}:${t.getMinutes().toString().padStart(2,"0")}:${t.getSeconds().toString().padStart(2,"0")}.${t.getMilliseconds().toString().padStart(3,"0")}`;e&&console.log(n,"Human:",...e)}var he=()=>typeof performance!="undefined"?performance.now():parseInt((Number(process.hrtime.bigint())/1e3/1e3).toString());function Zg(e,t,n="config",s=[]){for(let r of Object.keys(t))if(typeof t[r]=="object")Zg(e[r],t[r],r,s);else{let a=e&&typeof e[r]!="undefined";a||s.push({reason:"unknown property",where:`${n}.${r} = ${t[r]}`});let o=e&&typeof e[r]==typeof t[r];a&&!o&&s.push({reason:"property type mismatch",where:`${n}.${r} = ${t[r]}`,expected:typeof e[r]})}return t.debug&&n==="config"&&s.length>0&&re("invalid configuration",s),s}function Rn(...e){let t=n=>n&&typeof n=="object";return e.reduce((n,s)=>(Object.keys(s||{}).forEach(r=>{let a=n[r],o=s[r];Array.isArray(a)&&Array.isArray(o)?n[r]=a.concat(...o):t(a)&&t(o)?n[r]=Rn(a,o):n[r]=o}),n),{})}var wa={backend:"",modelBasePath:"",wasmPath:"",debug:!0,async:!0,warmup:"full",cacheSensitivity:.7,skipAllowed:!1,deallocate:!1,filter:{enabled:!0,equalization:!1,width:0,height:0,flip:!1,return:!0,brightness:0,contrast:0,sharpness:0,blur:0,saturation:0,hue:0,negative:!1,sepia:!1,vintage:!1,kodachrome:!1,technicolor:!1,polaroid:!1,pixelate:0},gesture:{enabled:!0},face:{enabled:!0,detector:{modelPath:"blazeface.json",rotation:!0,maxDetected:1,skipFrames:99,skipTime:2500,minConfidence:.2,iouThreshold:.1,return:!1},mesh:{enabled:!0,modelPath:"facemesh.json"},iris:{enabled:!0,modelPath:"iris.json"},emotion:{enabled:!0,minConfidence:.1,skipFrames:99,skipTime:1500,modelPath:"emotion.json"},description:{enabled:!0,modelPath:"faceres.json",skipFrames:99,skipTime:3e3,minConfidence:.1},antispoof:{enabled:!1,skipFrames:99,skipTime:4e3,modelPath:"antispoof.json"}},body:{enabled:!0,modelPath:"movenet-lightning.json",detector:{modelPath:""},maxDetected:-1,minConfidence:.3,skipFrames:1,skipTime:200},hand:{enabled:!0,rotation:!0,skipFrames:99,skipTime:1e3,minConfidence:.5,iouThreshold:.2,maxDetected:-1,landmarks:!0,detector:{modelPath:"handtrack.json"},skeleton:{modelPath:"handlandmark-full.json"}},object:{enabled:!1,modelPath:"mb3-centernet.json",minConfidence:.2,iouThreshold:.4,maxDetected:10,skipFrames:99,skipTime:2e3},segmentation:{enabled:!1,modelPath:"selfie.json",blur:8}};var $l={};Gc($l,{Abs:()=>ci,Acos:()=>eu,Acosh:()=>tu,AdadeltaOptimizer:()=>Nf,AdagradOptimizer:()=>Ef,AdamOptimizer:()=>Rf,AdamaxOptimizer:()=>$f,Add:()=>Gr,AddN:()=>Ca,All:()=>nu,Any:()=>su,ArgMax:()=>Ta,ArgMin:()=>ru,Asin:()=>au,Asinh:()=>ou,Atan:()=>iu,Atan2:()=>uu,Atanh:()=>lu,AvgPool:()=>Na,AvgPool3D:()=>Yc,AvgPool3DGrad:()=>gh,AvgPoolGrad:()=>mh,BackendWasm:()=>D6,BatchMatMul:()=>Ea,BatchToSpaceND:()=>di,Bincount:()=>Ah,BroadcastArgs:()=>yh,BroadcastTo:()=>$5,Callback:()=>Vk,CallbackList:()=>Rw,Cast:()=>Ra,Ceil:()=>$a,ClipByValue:()=>Hr,Complex:()=>Jc,ComplexAbs:()=>Qc,Concat:()=>pi,Conv2D:()=>_a,Conv2DBackpropFilter:()=>xh,Conv2DBackpropInput:()=>Da,Conv3D:()=>ed,Conv3DBackpropFilterV2:()=>bh,Conv3DBackpropInputV2:()=>vh,Cos:()=>Pa,Cosh:()=>Fa,CropAndResize:()=>fi,Cumsum:()=>hi,CustomCallback:()=>_w,DataStorage:()=>Xc,DenseBincount:()=>wh,DepthToSpace:()=>mi,DepthwiseConv2dNative:()=>Oa,DepthwiseConv2dNativeBackpropFilter:()=>kh,DepthwiseConv2dNativeBackpropInput:()=>Sh,Diag:()=>Ih,Dilation2D:()=>td,Dilation2DBackpropFilter:()=>Th,Dilation2DBackpropInput:()=>Ch,ENV:()=>Sr,EarlyStopping:()=>Gk,Einsum:()=>nd,Elu:()=>za,EluGrad:()=>Nh,Environment:()=>E5,Equal:()=>gi,Erf:()=>cu,Exp:()=>La,ExpandDims:()=>Ai,Expm1:()=>yi,FFT:()=>Eh,Fill:()=>du,FlipLeftRight:()=>xi,Floor:()=>Ba,FloorDiv:()=>Wa,FromPixels:()=>dd,FusedBatchNorm:()=>Va,FusedConv2D:()=>xo,FusedDepthwiseConv2D:()=>bo,GPGPUContext:()=>Fm,GatherNd:()=>vi,GatherV2:()=>bi,GraphModel:()=>w7,Greater:()=>wi,GreaterEqual:()=>Ua,History:()=>$w,IFFT:()=>Rh,Identity:()=>Ga,Imag:()=>sd,InputSpec:()=>Kt,IsFinite:()=>pu,IsInf:()=>hu,IsNan:()=>fu,KernelBackend:()=>Yl,LRN:()=>ad,LRNGrad:()=>_h,LayerVariable:()=>Iw,LayersModel:()=>ea,LeakyRelu:()=>ki,Less:()=>Si,LessEqual:()=>Ii,LinSpace:()=>$h,Log:()=>Ha,Log1p:()=>mu,LogSoftmax:()=>_5,LogicalAnd:()=>Ci,LogicalNot:()=>gu,LogicalOr:()=>rd,MathBackendCPU:()=>_y,MathBackendWebGL:()=>pp,Max:()=>ja,MaxPool:()=>Xa,MaxPool3D:()=>od,MaxPool3DGrad:()=>Ph,MaxPoolGrad:()=>Dh,MaxPoolWithArgmax:()=>Fh,Maximum:()=>qa,Mean:()=>Ka,Min:()=>Za,Minimum:()=>Ya,MirrorPad:()=>Ja,Mod:()=>Au,MomentumOptimizer:()=>_f,Multinomial:()=>Oh,Multiply:()=>Qa,Neg:()=>Ti,NonMaxSuppressionV3:()=>Ei,NonMaxSuppressionV4:()=>yu,NonMaxSuppressionV5:()=>Ri,NotEqual:()=>Ni,OP_SCOPE_SUFFIX:()=>q5,OneHot:()=>_i,OnesLike:()=>$i,Optimizer:()=>Yr,OptimizerConstructors:()=>_o,Pack:()=>Di,PadV2:()=>eo,Pool:()=>NE,Pow:()=>to,Prelu:()=>no,Prod:()=>Pi,RMSPropOptimizer:()=>Df,RNN:()=>ta,Range:()=>xu,Rank:()=>i2,Real:()=>id,RealDiv:()=>Ma,Reciprocal:()=>bu,Reduction:()=>Wn,Relu:()=>so,Relu6:()=>ao,Reshape:()=>Fi,ResizeBilinear:()=>ro,ResizeBilinearGrad:()=>zh,ResizeNearestNeighbor:()=>vu,ResizeNearestNeighborGrad:()=>Mh,Reverse:()=>Oi,RotateWithOffset:()=>Zi,Round:()=>Mi,Rsqrt:()=>oo,SGDOptimizer:()=>Od,ScatterNd:()=>zi,Select:()=>Li,Selu:()=>wu,Sequential:()=>rm,Sigmoid:()=>lo,Sign:()=>ku,Sin:()=>io,Sinh:()=>Wi,Slice:()=>Bi,Softmax:()=>po,Softplus:()=>Su,SpaceToBatchND:()=>Vi,SparseFillEmptyRows:()=>Lh,SparseReshape:()=>Bh,SparseSegmentMean:()=>Wh,SparseSegmentSum:()=>Vh,SparseToDense:()=>ld,SplitV:()=>Ui,Sqrt:()=>uo,Square:()=>Iu,SquaredDifference:()=>ho,Step:()=>Ao,StridedSlice:()=>Gi,StringNGrams:()=>ud,StringSplit:()=>Uh,StringToHashBucketFast:()=>Gh,Sub:()=>fo,Sum:()=>co,SymbolicTensor:()=>hr,Tan:()=>Hi,Tanh:()=>mo,Tensor:()=>Ze,TensorBuffer:()=>en,Tile:()=>jr,TopK:()=>ji,Transform:()=>qi,Transpose:()=>go,Unique:()=>Hh,Unpack:()=>Xi,UnsortedSegmentSum:()=>cd,Variable:()=>xd,ZerosLike:()=>Ki,_FusedMatMul:()=>yo,abs:()=>tn,acos:()=>D3,acosh:()=>P3,add:()=>le,addN:()=>sf,all:()=>B2,any:()=>rf,argMax:()=>Hs,argMin:()=>F3,asin:()=>O3,asinh:()=>M3,atan:()=>z3,atan2:()=>L3,atanh:()=>B3,avgPool:()=>of,avgPool3d:()=>U2,backend:()=>Tr,backend_util:()=>E,basicLSTMCell:()=>m$,batchNorm:()=>Du,batchNorm2d:()=>G3,batchNorm3d:()=>H3,batchNorm4d:()=>j3,batchToSpaceND:()=>lf,bincount:()=>G2,booleanMaskAsync:()=>TP,broadcastArgs:()=>q3,broadcastTo:()=>Id,broadcast_util:()=>nl,browser:()=>Gs,buffer:()=>ze,callbacks:()=>VV,cast:()=>de,ceil:()=>X3,clipByValue:()=>ds,clone:()=>Ln,complex:()=>ko,concat:()=>vt,concat1d:()=>K3,concat2d:()=>Pu,concat3d:()=>Z3,concat4d:()=>Y3,constraints:()=>rw,conv1d:()=>H2,conv2d:()=>No,conv2dTranspose:()=>q2,conv3d:()=>X2,conv3dTranspose:()=>Q3,copyRegisteredKernels:()=>_E,cos:()=>uf,cosh:()=>K2,cosineWindow:()=>b1,cumsum:()=>Z2,customGrad:()=>Er,data:()=>k7,denseBincount:()=>ev,deprecationWarn:()=>z2,depthToSpace:()=>tv,depthwiseConv2d:()=>Cd,deregisterOp:()=>HV,device_util:()=>Nu,diag:()=>H$,dilation2d:()=>nv,disableDeprecationWarnings:()=>$R,dispose:()=>se,disposeVariables:()=>_R,div:()=>pe,divNoNan:()=>sv,dot:()=>J$,dropout:()=>_v,einsum:()=>rv,elu:()=>Td,enableDebugMode:()=>RR,enableProdMode:()=>$3,enclosingPowerOfTwo:()=>Dv,engine:()=>ns,env:()=>K,equal:()=>Cs,erf:()=>av,exp:()=>Ts,expandDims:()=>mn,expm1:()=>ov,eye:()=>Y2,fft:()=>bf,fill:()=>Fu,findBackend:()=>L2,findBackendFactory:()=>OR,floor:()=>Nd,floorDiv:()=>nf,forceHalfFloat:()=>QI,fused:()=>$o,gather:()=>Ou,gatherND:()=>$v,gather_util:()=>R2,getBackend:()=>Is,getGradient:()=>s2,getKernel:()=>jh,getKernelsForBackend:()=>qr,getThreadsCount:()=>yge,gpgpu_util:()=>CI,grad:()=>k_,grads:()=>S_,greater:()=>ps,greaterEqual:()=>il,ifft:()=>Dd,imag:()=>cf,image:()=>$e,inTopKAsync:()=>zP,initializers:()=>dw,input:()=>tk,io:()=>ts,irfft:()=>f1,isFinite:()=>h_,isInf:()=>m_,isNaN:()=>iv,keep:()=>fn,kernel_impls:()=>Xs,layers:()=>ww,leakyRelu:()=>df,less:()=>J2,lessEqual:()=>ll,linalg:()=>Uv,linspace:()=>lv,loadGraphModel:()=>et,loadLayersModel:()=>QB,localResponseNormalization:()=>uv,log:()=>Ns,log1p:()=>pf,logSigmoid:()=>R_,logSoftmax:()=>Q2,logSumExp:()=>fv,logicalAnd:()=>ir,logicalNot:()=>ff,logicalOr:()=>n1,logicalXor:()=>V_,losses:()=>wO,matMul:()=>Ue,math:()=>d3,max:()=>gn,maxPool:()=>mf,maxPool3d:()=>s1,maxPoolWithArgmax:()=>mv,maximum:()=>Kr,mean:()=>Bt,memory:()=>ef,meshgrid:()=>X_,metrics:()=>Lk,min:()=>Eo,minimum:()=>Ed,mirrorPad:()=>gv,mod:()=>Rd,model:()=>YB,models:()=>Bk,moments:()=>gf,movingAverage:()=>RP,mul:()=>L,multiRNNCell:()=>nD,multinomial:()=>Av,neg:()=>Ft,nextFrame:()=>Gv,norm:()=>y1,notEqual:()=>zu,oneHot:()=>kd,ones:()=>hs,onesLike:()=>Es,op:()=>V,outerProduct:()=>iD,pad:()=>qs,pad1d:()=>cD,pad2d:()=>pD,pad3d:()=>fD,pad4d:()=>gD,pool:()=>vD,pow:()=>Ro,prelu:()=>yf,print:()=>a3,prod:()=>r1,profile:()=>DR,rand:()=>CD,randomGamma:()=>RD,randomNormal:()=>yv,randomUniform:()=>Lu,range:()=>Bu,ready:()=>tf,real:()=>$d,reciprocal:()=>xv,registerBackend:()=>al,registerCallbackConstructor:()=>eW,registerGradient:()=>D5,registerKernel:()=>rr,registerOp:()=>GV,regularizers:()=>Wk,relu:()=>Rr,relu6:()=>i1,removeBackend:()=>FR,reshape:()=>G,reverse:()=>Rs,reverse1d:()=>LD,reverse2d:()=>WD,reverse3d:()=>UD,reverse4d:()=>HD,rfft:()=>vf,round:()=>l1,rsqrt:()=>u1,scalar:()=>Re,scatterND:()=>Rv,scatter_util:()=>$2,selu:()=>c1,separableConv2d:()=>bv,sequential:()=>JB,serialization:()=>ue,setBackend:()=>_3,setPlatform:()=>MR,setThreadsCount:()=>Age,setWasmPath:()=>gge,setWasmPaths:()=>F6,setWebGLContext:()=>Cm,setdiff1dAsync:()=>vv,shared:()=>bm,sigmoid:()=>cs,sign:()=>wv,signal:()=>vO,sin:()=>d1,sinh:()=>p1,slice:()=>De,slice1d:()=>xf,slice2d:()=>h1,slice3d:()=>Wu,slice4d:()=>_d,slice_util:()=>Pt,softmax:()=>Vu,softplus:()=>Mu,spaceToBatchND:()=>Af,sparse:()=>Fd,sparseToDense:()=>x1,spectral:()=>bO,split:()=>nn,sqrt:()=>_n,square:()=>gt,squaredDifference:()=>m1,squeeze:()=>ot,stack:()=>An,step:()=>Pd,stridedSlice:()=>kv,string:()=>Tf,sub:()=>me,sum:()=>Se,sumOutType:()=>bd,tan:()=>Sv,tanh:()=>_u,tensor:()=>Ut,tensor1d:()=>Xt,tensor2d:()=>lr,tensor3d:()=>h3,tensor4d:()=>AP,tensor5d:()=>yP,tensor6d:()=>xP,tensor_util:()=>ar,test_util:()=>N3,tidy:()=>q,tile:()=>js,time:()=>PR,topk:()=>Iv,train:()=>cl,transpose:()=>Ye,truncatedNormal:()=>wf,unique:()=>g1,unregisterGradient:()=>$E,unregisterKernel:()=>RE,unsortedSegmentSum:()=>Cv,unstack:()=>ss,upcastType:()=>zn,util:()=>v,valueAndGrad:()=>I_,valueAndGrads:()=>C_,variable:()=>Tv,variableGrads:()=>cv,version:()=>M6,version_converter:()=>XU,version_core:()=>Cp,version_cpu:()=>PH,version_layers:()=>nA,version_wasm:()=>xge,version_webgl:()=>iQ,webgl:()=>lQ,webgl_util:()=>ZS,webgpu:()=>FC,where:()=>Bn,whereAsync:()=>A1,zeros:()=>Gt,zerosLike:()=>Je});var Zl=(e=>typeof va!="undefined"?va:typeof Proxy!="undefined"?new Proxy(e,{get:(t,n)=>(typeof va!="undefined"?va:t)[n]}):e)(function(e){if(typeof va!="undefined")return va.apply(this,arguments);throw new Error('Dynamic require of "'+e+'" is not supported')}),WN=Object.create,uh=Object.defineProperty,VN=Object.getOwnPropertyDescriptor,UN=Object.getOwnPropertyNames,GN=Object.getPrototypeOf,HN=Object.prototype.hasOwnProperty,g5=e=>uh(e,"__esModule",{value:!0}),Ws=(e=>typeof Zl!="undefined"?Zl:typeof Proxy!="undefined"?new Proxy(e,{get:(t,n)=>(typeof Zl!="undefined"?Zl:t)[n]}):e)(function(e){if(typeof Zl!="undefined")return Zl.apply(this,arguments);throw new Error('Dynamic require of "'+e+'" is not supported')}),es=(e,t)=>function(){return t||(0,e[Object.keys(e)[0]])((t={exports:{}}).exports,t),t.exports},Oe=(e,t)=>{g5(e);for(var n in t)uh(e,n,{get:t[n],enumerable:!0})},jN=(e,t,n)=>{if(t&&typeof t=="object"||typeof t=="function")for(let s of UN(t))!HN.call(e,s)&&s!=="default"&&uh(e,s,{get:()=>t[s],enumerable:!(n=VN(t,s))||n.enumerable});return e},ii=e=>jN(g5(uh(e!=null?WN(GN(e)):{},"default",e&&e.__esModule&&"default"in e?{get:()=>e.default,enumerable:!0}:{value:e,enumerable:!0})),e),qN=es({"node_modules/.pnpm/long@4.0.0/node_modules/long/src/long.js"(e,t){t.exports=s;var n=null;try{n=new WebAssembly.Instance(new WebAssembly.Module(new Uint8Array([0,97,115,109,1,0,0,0,1,13,2,96,0,1,127,96,4,127,127,127,127,1,127,3,7,6,0,1,1,1,1,1,6,6,1,127,1,65,0,11,7,50,6,3,109,117,108,0,1,5,100,105,118,95,115,0,2,5,100,105,118,95,117,0,3,5,114,101,109,95,115,0,4,5,114,101,109,95,117,0,5,8,103,101,116,95,104,105,103,104,0,0,10,191,1,6,4,0,35,0,11,36,1,1,126,32,0,173,32,1,173,66,32,134,132,32,2,173,32,3,173,66,32,134,132,126,34,4,66,32,135,167,36,0,32,4,167,11,36,1,1,126,32,0,173,32,1,173,66,32,134,132,32,2,173,32,3,173,66,32,134,132,127,34,4,66,32,135,167,36,0,32,4,167,11,36,1,1,126,32,0,173,32,1,173,66,32,134,132,32,2,173,32,3,173,66,32,134,132,128,34,4,66,32,135,167,36,0,32,4,167,11,36,1,1,126,32,0,173,32,1,173,66,32,134,132,32,2,173,32,3,173,66,32,134,132,129,34,4,66,32,135,167,36,0,32,4,167,11,36,1,1,126,32,0,173,32,1,173,66,32,134,132,32,2,173,32,3,173,66,32,134,132,130,34,4,66,32,135,167,36,0,32,4,167,11])),{}).exports}catch(P){}function s(P,T,M){this.low=P|0,this.high=T|0,this.unsigned=!!M}s.prototype.__isLong__,Object.defineProperty(s.prototype,"__isLong__",{value:!0});function r(P){return(P&&P.__isLong__)===!0}s.isLong=r;var a={},o={};function i(P,T){var M,U,j;return T?(P>>>=0,(j=0<=P&&P<256)&&(U=o[P],U)?U:(M=c(P,(P|0)<0?-1:0,!0),j&&(o[P]=M),M)):(P|=0,(j=-128<=P&&P<128)&&(U=a[P],U)?U:(M=c(P,P<0?-1:0,!1),j&&(a[P]=M),M))}s.fromInt=i;function l(P,T){if(isNaN(P))return T?b:y;if(T){if(P<0)return b;if(P>=g)return R}else{if(P<=-A)return F;if(P+1>=A)return N}return P<0?l(-P,T).neg():c(P%m|0,P/m|0,T)}s.fromNumber=l;function c(P,T,M){return new s(P,T,M)}s.fromBits=c;var u=Math.pow;function d(P,T,M){if(P.length===0)throw Error("empty string");if(P==="NaN"||P==="Infinity"||P==="+Infinity"||P==="-Infinity")return y;if(typeof T=="number"?(M=T,T=!1):T=!!T,M=M||10,M<2||36<M)throw RangeError("radix");var U;if((U=P.indexOf("-"))>0)throw Error("interior hyphen");if(U===0)return d(P.substring(1),T,M).neg();for(var j=l(u(M,8)),z=y,X=0;X<P.length;X+=8){var Z=Math.min(8,P.length-X),J=parseInt(P.substring(X,X+Z),M);if(Z<8){var ee=l(u(M,Z));z=z.mul(ee).add(l(J))}else z=z.mul(j),z=z.add(l(J))}return z.unsigned=T,z}s.fromString=d;function p(P,T){return typeof P=="number"?l(P,T):typeof P=="string"?d(P,T):c(P.low,P.high,typeof T=="boolean"?T:P.unsigned)}s.fromValue=p;var h=1<<16,f=1<<24,m=h*h,g=m*m,A=g/2,x=i(f),y=i(0);s.ZERO=y;var b=i(0,!0);s.UZERO=b;var w=i(1);s.ONE=w;var k=i(1,!0);s.UONE=k;var C=i(-1);s.NEG_ONE=C;var N=c(4294967295|0,2147483647|0,!1);s.MAX_VALUE=N;var R=c(4294967295|0,4294967295|0,!0);s.MAX_UNSIGNED_VALUE=R;var F=c(0,2147483648|0,!1);s.MIN_VALUE=F;var _=s.prototype;_.toInt=function(){return this.unsigned?this.low>>>0:this.low},_.toNumber=function(){return this.unsigned?(this.high>>>0)*m+(this.low>>>0):this.high*m+(this.low>>>0)},_.toString=function(T){if(T=T||10,T<2||36<T)throw RangeError("radix");if(this.isZero())return"0";if(this.isNegative())if(this.eq(F)){var M=l(T),U=this.div(M),j=U.mul(M).sub(this);return U.toString(T)+j.toInt().toString(T)}else return"-"+this.neg().toString(T);for(var z=l(u(T,6),this.unsigned),X=this,Z="";;){var J=X.div(z),ee=X.sub(J.mul(z)).toInt()>>>0,ne=ee.toString(T);if(X=J,X.isZero())return ne+Z;for(;ne.length<6;)ne="0"+ne;Z=""+ne+Z}},_.getHighBits=function(){return this.high},_.getHighBitsUnsigned=function(){return this.high>>>0},_.getLowBits=function(){return this.low},_.getLowBitsUnsigned=function(){return this.low>>>0},_.getNumBitsAbs=function(){if(this.isNegative())return this.eq(F)?64:this.neg().getNumBitsAbs();for(var T=this.high!=0?this.high:this.low,M=31;M>0&&(T&1<<M)==0;M--);return this.high!=0?M+33:M+1},_.isZero=function(){return this.high===0&&this.low===0},_.eqz=_.isZero,_.isNegative=function(){return!this.unsigned&&this.high<0},_.isPositive=function(){return this.unsigned||this.high>=0},_.isOdd=function(){return(this.low&1)==1},_.isEven=function(){return(this.low&1)==0},_.equals=function(T){return r(T)||(T=p(T)),this.unsigned!==T.unsigned&&this.high>>>31==1&&T.high>>>31==1?!1:this.high===T.high&&this.low===T.low},_.eq=_.equals,_.notEquals=function(T){return!this.eq(T)},_.neq=_.notEquals,_.ne=_.notEquals,_.lessThan=function(T){return this.comp(T)<0},_.lt=_.lessThan,_.lessThanOrEqual=function(T){return this.comp(T)<=0},_.lte=_.lessThanOrEqual,_.le=_.lessThanOrEqual,_.greaterThan=function(T){return this.comp(T)>0},_.gt=_.greaterThan,_.greaterThanOrEqual=function(T){return this.comp(T)>=0},_.gte=_.greaterThanOrEqual,_.ge=_.greaterThanOrEqual,_.compare=function(T){if(r(T)||(T=p(T)),this.eq(T))return 0;var M=this.isNegative(),U=T.isNegative();return M&&!U?-1:!M&&U?1:this.unsigned?T.high>>>0>this.high>>>0||T.high===this.high&&T.low>>>0>this.low>>>0?-1:1:this.sub(T).isNegative()?-1:1},_.comp=_.compare,_.negate=function(){return!this.unsigned&&this.eq(F)?F:this.not().add(w)},_.neg=_.negate,_.add=function(T){r(T)||(T=p(T));var M=this.high>>>16,U=this.high&65535,j=this.low>>>16,z=this.low&65535,X=T.high>>>16,Z=T.high&65535,J=T.low>>>16,ee=T.low&65535,ne=0,Q=0,te=0,oe=0;return oe+=z+ee,te+=oe>>>16,oe&=65535,te+=j+J,Q+=te>>>16,te&=65535,Q+=U+Z,ne+=Q>>>16,Q&=65535,ne+=M+X,ne&=65535,c(te<<16|oe,ne<<16|Q,this.unsigned)},_.subtract=function(T){return r(T)||(T=p(T)),this.add(T.neg())},_.sub=_.subtract,_.multiply=function(T){if(this.isZero())return y;if(r(T)||(T=p(T)),n){var M=n.mul(this.low,this.high,T.low,T.high);return c(M,n.get_high(),this.unsigned)}if(T.isZero())return y;if(this.eq(F))return T.isOdd()?F:y;if(T.eq(F))return this.isOdd()?F:y;if(this.isNegative())return T.isNegative()?this.neg().mul(T.neg()):this.neg().mul(T).neg();if(T.isNegative())return this.mul(T.neg()).neg();if(this.lt(x)&&T.lt(x))return l(this.toNumber()*T.toNumber(),this.unsigned);var U=this.high>>>16,j=this.high&65535,z=this.low>>>16,X=this.low&65535,Z=T.high>>>16,J=T.high&65535,ee=T.low>>>16,ne=T.low&65535,Q=0,te=0,oe=0,fe=0;return fe+=X*ne,oe+=fe>>>16,fe&=65535,oe+=z*ne,te+=oe>>>16,oe&=65535,oe+=X*ee,te+=oe>>>16,oe&=65535,te+=j*ne,Q+=te>>>16,te&=65535,te+=z*ee,Q+=te>>>16,te&=65535,te+=X*J,Q+=te>>>16,te&=65535,Q+=U*ne+j*ee+z*J+X*Z,Q&=65535,c(oe<<16|fe,Q<<16|te,this.unsigned)},_.mul=_.multiply,_.divide=function(T){if(r(T)||(T=p(T)),T.isZero())throw Error("division by zero");if(n){if(!this.unsigned&&this.high===-2147483648&&T.low===-1&&T.high===-1)return this;var M=(this.unsigned?n.div_u:n.div_s)(this.low,this.high,T.low,T.high);return c(M,n.get_high(),this.unsigned)}if(this.isZero())return this.unsigned?b:y;var U,j,z;if(this.unsigned){if(T.unsigned||(T=T.toUnsigned()),T.gt(this))return b;if(T.gt(this.shru(1)))return k;z=b}else{if(this.eq(F)){if(T.eq(w)||T.eq(C))return F;if(T.eq(F))return w;var X=this.shr(1);return U=X.div(T).shl(1),U.eq(y)?T.isNegative()?w:C:(j=this.sub(T.mul(U)),z=U.add(j.div(T)),z)}else if(T.eq(F))return this.unsigned?b:y;if(this.isNegative())return T.isNegative()?this.neg().div(T.neg()):this.neg().div(T).neg();if(T.isNegative())return this.div(T.neg()).neg();z=y}for(j=this;j.gte(T);){U=Math.max(1,Math.floor(j.toNumber()/T.toNumber()));for(var Z=Math.ceil(Math.log(U)/Math.LN2),J=Z<=48?1:u(2,Z-48),ee=l(U),ne=ee.mul(T);ne.isNegative()||ne.gt(j);)U-=J,ee=l(U,this.unsigned),ne=ee.mul(T);ee.isZero()&&(ee=w),z=z.add(ee),j=j.sub(ne)}return z},_.div=_.divide,_.modulo=function(T){if(r(T)||(T=p(T)),n){var M=(this.unsigned?n.rem_u:n.rem_s)(this.low,this.high,T.low,T.high);return c(M,n.get_high(),this.unsigned)}return this.sub(this.div(T).mul(T))},_.mod=_.modulo,_.rem=_.modulo,_.not=function(){return c(~this.low,~this.high,this.unsigned)},_.and=function(T){return r(T)||(T=p(T)),c(this.low&T.low,this.high&T.high,this.unsigned)},_.or=function(T){return r(T)||(T=p(T)),c(this.low|T.low,this.high|T.high,this.unsigned)},_.xor=function(T){return r(T)||(T=p(T)),c(this.low^T.low,this.high^T.high,this.unsigned)},_.shiftLeft=function(T){return r(T)&&(T=T.toInt()),(T&=63)==0?this:T<32?c(this.low<<T,this.high<<T|this.low>>>32-T,this.unsigned):c(0,this.low<<T-32,this.unsigned)},_.shl=_.shiftLeft,_.shiftRight=function(T){return r(T)&&(T=T.toInt()),(T&=63)==0?this:T<32?c(this.low>>>T|this.high<<32-T,this.high>>T,this.unsigned):c(this.high>>T-32,this.high>=0?0:-1,this.unsigned)},_.shr=_.shiftRight,_.shiftRightUnsigned=function(T){if(r(T)&&(T=T.toInt()),T&=63,T===0)return this;var M=this.high;if(T<32){var U=this.low;return c(U>>>T|M<<32-T,M>>>T,this.unsigned)}else return T===32?c(M,0,this.unsigned):c(M>>>T-32,0,this.unsigned)},_.shru=_.shiftRightUnsigned,_.shr_u=_.shiftRightUnsigned,_.toSigned=function(){return this.unsigned?c(this.low,this.high,!1):this},_.toUnsigned=function(){return this.unsigned?this:c(this.low,this.high,!0)},_.toBytes=function(T){return T?this.toBytesLE():this.toBytesBE()},_.toBytesLE=function(){var T=this.high,M=this.low;return[M&255,M>>>8&255,M>>>16&255,M>>>24,T&255,T>>>8&255,T>>>16&255,T>>>24]},_.toBytesBE=function(){var T=this.high,M=this.low;return[T>>>24,T>>>16&255,T>>>8&255,T&255,M>>>24,M>>>16&255,M>>>8&255,M&255]},s.fromBytes=function(T,M,U){return U?s.fromBytesLE(T,M):s.fromBytesBE(T,M)},s.fromBytesLE=function(T,M){return new s(T[0]|T[1]<<8|T[2]<<16|T[3]<<24,T[4]|T[5]<<8|T[6]<<16|T[7]<<24,M)},s.fromBytesBE=function(T,M){return new s(T[4]<<24|T[5]<<16|T[6]<<8|T[7],T[0]<<24|T[1]<<16|T[2]<<8|T[3],M)}}}),XN=es({"(disabled):node-fetch"(){}}),KN=es({"(disabled):util"(){}}),ZN=es({"node_modules/.pnpm/seedrandom@3.0.5/node_modules/seedrandom/lib/alea.js"(e,t){(function(n,s,r){function a(c){var u=this,d=l();u.next=function(){var p=2091639*u.s0+u.c*23283064365386963e-26;return u.s0=u.s1,u.s1=u.s2,u.s2=p-(u.c=p|0)},u.c=1,u.s0=d(" "),u.s1=d(" "),u.s2=d(" "),u.s0-=d(c),u.s0<0&&(u.s0+=1),u.s1-=d(c),u.s1<0&&(u.s1+=1),u.s2-=d(c),u.s2<0&&(u.s2+=1),d=null}function o(c,u){return u.c=c.c,u.s0=c.s0,u.s1=c.s1,u.s2=c.s2,u}function i(c,u){var d=new a(c),p=u&&u.state,h=d.next;return h.int32=function(){return d.next()*4294967296|0},h.double=function(){return h()+(h()*2097152|0)*11102230246251565e-32},h.quick=h,p&&(typeof p=="object"&&o(p,d),h.state=function(){return o(d,{})}),h}function l(){var c=4022871197,u=function(d){d=String(d);for(var p=0;p<d.length;p++){c+=d.charCodeAt(p);var h=.02519603282416938*c;c=h>>>0,h-=c,h*=c,c=h>>>0,h-=c,c+=h*4294967296}return(c>>>0)*23283064365386963e-26};return u}s&&s.exports?s.exports=i:r&&r.amd?r(function(){return i}):this.alea=i})(e,typeof t=="object"&&t,typeof define=="function"&&define)}}),YN=es({"node_modules/.pnpm/seedrandom@3.0.5/node_modules/seedrandom/lib/xor128.js"(e,t){(function(n,s,r){function a(l){var c=this,u="";c.x=0,c.y=0,c.z=0,c.w=0,c.next=function(){var p=c.x^c.x<<11;return c.x=c.y,c.y=c.z,c.z=c.w,c.w^=c.w>>>19^p^p>>>8},l===(l|0)?c.x=l:u+=l;for(var d=0;d<u.length+64;d++)c.x^=u.charCodeAt(d)|0,c.next()}function o(l,c){return c.x=l.x,c.y=l.y,c.z=l.z,c.w=l.w,c}function i(l,c){var u=new a(l),d=c&&c.state,p=function(){return(u.next()>>>0)/4294967296};return p.double=function(){do var h=u.next()>>>11,f=(u.next()>>>0)/4294967296,m=(h+f)/(1<<21);while(m===0);return m},p.int32=u.next,p.quick=p,d&&(typeof d=="object"&&o(d,u),p.state=function(){return o(u,{})}),p}s&&s.exports?s.exports=i:r&&r.amd?r(function(){return i}):this.xor128=i})(e,typeof t=="object"&&t,typeof define=="function"&&define)}}),JN=es({"node_modules/.pnpm/seedrandom@3.0.5/node_modules/seedrandom/lib/xorwow.js"(e,t){(function(n,s,r){function a(l){var c=this,u="";c.next=function(){var p=c.x^c.x>>>2;return c.x=c.y,c.y=c.z,c.z=c.w,c.w=c.v,(c.d=c.d+362437|0)+(c.v=c.v^c.v<<4^(p^p<<1))|0},c.x=0,c.y=0,c.z=0,c.w=0,c.v=0,l===(l|0)?c.x=l:u+=l;for(var d=0;d<u.length+64;d++)c.x^=u.charCodeAt(d)|0,d==u.length&&(c.d=c.x<<10^c.x>>>4),c.next()}function o(l,c){return c.x=l.x,c.y=l.y,c.z=l.z,c.w=l.w,c.v=l.v,c.d=l.d,c}function i(l,c){var u=new a(l),d=c&&c.state,p=function(){return(u.next()>>>0)/4294967296};return p.double=function(){do var h=u.next()>>>11,f=(u.next()>>>0)/4294967296,m=(h+f)/(1<<21);while(m===0);return m},p.int32=u.next,p.quick=p,d&&(typeof d=="object"&&o(d,u),p.state=function(){return o(u,{})}),p}s&&s.exports?s.exports=i:r&&r.amd?r(function(){return i}):this.xorwow=i})(e,typeof t=="object"&&t,typeof define=="function"&&define)}}),QN=es({"node_modules/.pnpm/seedrandom@3.0.5/node_modules/seedrandom/lib/xorshift7.js"(e,t){(function(n,s,r){function a(l){var c=this;c.next=function(){var d=c.x,p=c.i,h,f,m;return h=d[p],h^=h>>>7,f=h^h<<24,h=d[p+1&7],f^=h^h>>>10,h=d[p+3&7],f^=h^h>>>3,h=d[p+4&7],f^=h^h<<7,h=d[p+7&7],h=h^h<<13,f^=h^h<<9,d[p]=f,c.i=p+1&7,f};function u(d,p){var h,f,m=[];if(p===(p|0))f=m[0]=p;else for(p=""+p,h=0;h<p.length;++h)m[h&7]=m[h&7]<<15^p.charCodeAt(h)+m[h+1&7]<<13;for(;m.length<8;)m.push(0);for(h=0;h<8&&m[h]===0;++h);for(h==8?f=m[7]=-1:f=m[h],d.x=m,d.i=0,h=256;h>0;--h)d.next()}u(c,l)}function o(l,c){return c.x=l.x.slice(),c.i=l.i,c}function i(l,c){l==null&&(l=+new Date);var u=new a(l),d=c&&c.state,p=function(){return(u.next()>>>0)/4294967296};return p.double=function(){do var h=u.next()>>>11,f=(u.next()>>>0)/4294967296,m=(h+f)/(1<<21);while(m===0);return m},p.int32=u.next,p.quick=p,d&&(d.x&&o(d,u),p.state=function(){return o(u,{})}),p}s&&s.exports?s.exports=i:r&&r.amd?r(function(){return i}):this.xorshift7=i})(e,typeof t=="object"&&t,typeof define=="function"&&define)}}),eE=es({"node_modules/.pnpm/seedrandom@3.0.5/node_modules/seedrandom/lib/xor4096.js"(e,t){(function(n,s,r){function a(l){var c=this;c.next=function(){var d=c.w,p=c.X,h=c.i,f,m;return c.w=d=d+1640531527|0,m=p[h+34&127],f=p[h=h+1&127],m^=m<<13,f^=f<<17,m^=m>>>15,f^=f>>>12,m=p[h]=m^f,c.i=h,m+(d^d>>>16)|0};function u(d,p){var h,f,m,g,A,x=[],y=128;for(p===(p|0)?(f=p,p=null):(p=p+"\0",f=0,y=Math.max(y,p.length)),m=0,g=-32;g<y;++g)p&&(f^=p.charCodeAt((g+32)%p.length)),g===0&&(A=f),f^=f<<10,f^=f>>>15,f^=f<<4,f^=f>>>13,g>=0&&(A=A+1640531527|0,h=x[g&127]^=f+A,m=h==0?m+1:0);for(m>=128&&(x[(p&&p.length||0)&127]=-1),m=127,g=4*128;g>0;--g)f=x[m+34&127],h=x[m=m+1&127],f^=f<<13,h^=h<<17,f^=f>>>15,h^=h>>>12,x[m]=f^h;d.w=A,d.X=x,d.i=m}u(c,l)}function o(l,c){return c.i=l.i,c.w=l.w,c.X=l.X.slice(),c}function i(l,c){l==null&&(l=+new Date);var u=new a(l),d=c&&c.state,p=function(){return(u.next()>>>0)/4294967296};return p.double=function(){do var h=u.next()>>>11,f=(u.next()>>>0)/4294967296,m=(h+f)/(1<<21);while(m===0);return m},p.int32=u.next,p.quick=p,d&&(d.X&&o(d,u),p.state=function(){return o(u,{})}),p}s&&s.exports?s.exports=i:r&&r.amd?r(function(){return i}):this.xor4096=i})(e,typeof t=="object"&&t,typeof define=="function"&&define)}}),tE=es({"node_modules/.pnpm/seedrandom@3.0.5/node_modules/seedrandom/lib/tychei.js"(e,t){(function(n,s,r){function a(l){var c=this,u="";c.next=function(){var p=c.b,h=c.c,f=c.d,m=c.a;return p=p<<25^p>>>7^h,h=h-f|0,f=f<<24^f>>>8^m,m=m-p|0,c.b=p=p<<20^p>>>12^h,c.c=h=h-f|0,c.d=f<<16^h>>>16^m,c.a=m-p|0},c.a=0,c.b=0,c.c=2654435769|0,c.d=1367130551,l===Math.floor(l)?(c.a=l/4294967296|0,c.b=l|0):u+=l;for(var d=0;d<u.length+20;d++)c.b^=u.charCodeAt(d)|0,c.next()}function o(l,c){return c.a=l.a,c.b=l.b,c.c=l.c,c.d=l.d,c}function i(l,c){var u=new a(l),d=c&&c.state,p=function(){return(u.next()>>>0)/4294967296};return p.double=function(){do var h=u.next()>>>11,f=(u.next()>>>0)/4294967296,m=(h+f)/(1<<21);while(m===0);return m},p.int32=u.next,p.quick=p,d&&(typeof d=="object"&&o(d,u),p.state=function(){return o(u,{})}),p}s&&s.exports?s.exports=i:r&&r.amd?r(function(){return i}):this.tychei=i})(e,typeof t=="object"&&t,typeof define=="function"&&define)}}),nE=es({"(disabled):crypto"(){}}),sE=es({"node_modules/.pnpm/seedrandom@3.0.5/node_modules/seedrandom/seedrandom.js"(e,t){(function(n,s,r){var a=256,o=6,i=52,l="random",c=r.pow(a,o),u=r.pow(2,i),d=u*2,p=a-1,h;function f(w,k,C){var N=[];k=k==!0?{entropy:!0}:k||{};var R=x(A(k.entropy?[w,b(s)]:w==null?y():w,3),N),F=new m(N),_=function(){for(var P=F.g(o),T=c,M=0;P<u;)P=(P+M)*a,T*=a,M=F.g(1);for(;P>=d;)P/=2,T/=2,M>>>=1;return(P+M)/T};return _.int32=function(){return F.g(4)|0},_.quick=function(){return F.g(4)/4294967296},_.double=_,x(b(F.S),s),(k.pass||C||function(P,T,M,U){return U&&(U.S&&g(U,F),P.state=function(){return g(F,{})}),M?(r[l]=P,T):P})(_,R,"global"in k?k.global:this==r,k.state)}function m(w){var k,C=w.length,N=this,R=0,F=N.i=N.j=0,_=N.S=[];for(C||(w=[C++]);R<a;)_[R]=R++;for(R=0;R<a;R++)_[R]=_[F=p&F+w[R%C]+(k=_[R])],_[F]=k;(N.g=function(P){for(var T,M=0,U=N.i,j=N.j,z=N.S;P--;)T=z[U=p&U+1],M=M*a+z[p&(z[U]=z[j=p&j+T])+(z[j]=T)];return N.i=U,N.j=j,M})(a)}function g(w,k){return k.i=w.i,k.j=w.j,k.S=w.S.slice(),k}function A(w,k){var C=[],N=typeof w,R;if(k&&N=="object")for(R in w)try{C.push(A(w[R],k-1))}catch(F){}return C.length?C:N=="string"?w:w+"\0"}function x(w,k){for(var C=w+"",N,R=0;R<C.length;)k[p&R]=p&(N^=k[p&R]*19)+C.charCodeAt(R++);return b(k)}function y(){try{var w;return h&&(w=h.randomBytes)?w=w(a):(w=new Uint8Array(a),(n.crypto||n.msCrypto).getRandomValues(w)),b(w)}catch(N){var k=n.navigator,C=k&&k.plugins;return[+new Date,n,C,n.screen,b(s)]}}function b(w){return String.fromCharCode.apply(0,w)}if(x(r.random(),s),typeof t=="object"&&t.exports){t.exports=f;try{h=nE()}catch(w){}}else typeof define=="function"&&define.amd?define(function(){return f}):r["seed"+l]=f})(typeof self!="undefined"?self:e,[],Math)}}),ch=es({"node_modules/.pnpm/seedrandom@3.0.5/node_modules/seedrandom/index.js"(e,t){var n=ZN(),s=YN(),r=JN(),a=QN(),o=eE(),i=tE(),l=sE();l.alea=n,l.xor128=s,l.xorwow=r,l.xorshift7=a,l.xor4096=o,l.tychei=i,t.exports=l}}),A5=es({"(disabled):src/node_modules/string_decoder/index.js"(){}}),rE=es({"src/tfjs-backend-wasm/wasm-out/tfjs-backend-wasm-threaded-simd.js"(e,t){var n=function(){var s=typeof document!="undefined"&&document.currentScript?document.currentScript.src:void 0;return typeof __filename!="undefined"&&(s=s||__filename),function(r){r=r||{};function a(){return te.buffer!=Ct&&kn(te.buffer),_t}function o(){return te.buffer!=Ct&&kn(te.buffer),ws}function i(){return te.buffer!=Ct&&kn(te.buffer),Pn}function l(){return te.buffer!=Ct&&kn(te.buffer),ls}function c(){return te.buffer!=Ct&&kn(te.buffer),ks}var u=typeof r!="undefined"?r:{},d,p;u.ready=new Promise(function(I,$){d=I,p=$});var h={},f;for(f in u)u.hasOwnProperty(f)&&(h[f]=u[f]);var m=[],g="./this.program",A=function(I,$){throw $},x=!1,y=!1,b=!1,w=!1;x=typeof window=="object",y=typeof importScripts=="function",b=typeof process=="object"&&typeof process.versions=="object"&&typeof process.versions.node=="string",w=!x&&!b&&!y;var k=u.ENVIRONMENT_IS_PTHREAD||!1;k&&(Ct=u.buffer);var C="";function N(I){return u.locateFile?u.locateFile(I,C):C+I}var R,F,_,P,T,M;if(b){y?C=Ws("path").dirname(C)+"/":C=__dirname+"/",R=function($,B){return T||(T=Ws("fs")),M||(M=Ws("path")),$=M.normalize($),T.readFileSync($,B?null:"utf8")},_=function($){var B=R($,!0);return B.buffer||(B=new Uint8Array(B)),we(B.buffer),B},process.argv.length>1&&(g=process.argv[1].replace(/\\/g,"/")),m=process.argv.slice(2),process.on("uncaughtException",function(I){if(!(I instanceof Uc))throw I}),process.on("unhandledRejection",Wr),A=function(I){process.exit(I)},u.inspect=function(){return"[Emscripten Module object]"};var U;try{U=Ws("worker_threads")}catch(I){throw console.error('The "worker_threads" module is not supported in this node.js build - perhaps a newer version is needed?'),I}global.Worker=U.Worker}else w?(typeof read!="undefined"&&(R=function($){return read($)}),_=function($){var B;return typeof readbuffer=="function"?new Uint8Array(readbuffer($)):(B=read($,"binary"),we(typeof B=="object"),B)},typeof scriptArgs!="undefined"?m=scriptArgs:typeof arguments!="undefined"&&(m=arguments),typeof quit=="function"&&(A=function(I){quit(I)}),typeof print!="undefined"&&(typeof console=="undefined"&&(console={}),console.log=print,console.warn=console.error=typeof printErr!="undefined"?printErr:print)):(x||y)&&(y?C=self.location.href:typeof document!="undefined"&&document.currentScript&&(C=document.currentScript.src),typeof s!="undefined"&&s&&(C=s),C.indexOf("blob:")!==0?C=C.substr(0,C.lastIndexOf("/")+1):C="",b?(R=function($,B){return T||(T=Ws("fs")),M||(M=Ws("path")),$=M.normalize($),T.readFileSync($,B?null:"utf8")},_=function($){var B=R($,!0);return B.buffer||(B=new Uint8Array(B)),we(B.buffer),B}):(R=function(I){var $=new XMLHttpRequest;return $.open("GET",I,!1),$.send(null),$.responseText},y&&(_=function(I){var $=new XMLHttpRequest;return $.open("GET",I,!1),$.responseType="arraybuffer",$.send(null),new Uint8Array($.response)}),F=function(I,$,B){var Y=new XMLHttpRequest;Y.open("GET",I,!0),Y.responseType="arraybuffer",Y.onload=function(){if(Y.status==200||Y.status==0&&Y.response){$(Y.response);return}B()},Y.onerror=B,Y.send(null)}),P=function(I){document.title=I});b&&typeof performance=="undefined"&&(global.performance=Ws("perf_hooks").performance);var j=u.print||console.log.bind(console),z=u.printErr||console.warn.bind(console);for(f in h)h.hasOwnProperty(f)&&(u[f]=h[f]);h=null,u.arguments&&(m=u.arguments),u.thisProgram&&(g=u.thisProgram),u.quit&&(A=u.quit);function X(I){X.shown||(X.shown={}),X.shown[I]||(X.shown[I]=1,z(I))}var Z=Atomics.load,J=Atomics.store,ee=Atomics.compareExchange,ne;u.wasmBinary&&(ne=u.wasmBinary);var Q=u.noExitRuntime||!0;typeof WebAssembly!="object"&&Wr("no native wasm support detected");var te,oe,fe=!1,be;function we(I,$){I||Wr("Assertion failed: "+$)}function Ce(I){var $=u["_"+I];return we($,"Cannot call unknown function "+I+", make sure it is exported"),$}function Me(I,$,B,Y,xe){var ge={string:function(Fn){var Kl=0;if(Fn!=null&&Fn!==0){var f5=(Fn.length<<2)+1;Kl=jl(f5),dt(Fn,Kl,f5)}return Kl},array:function(Fn){var Kl=jl(Fn.length);return wt(Fn,Kl),Kl}};function ye(Fn){return $==="string"?qe(Fn):$==="boolean"?Boolean(Fn):Fn}var Te=Ce(I),ut=[],dn=0;if(Y)for(var Qt=0;Qt<Y.length;Qt++){var ba=ge[B[Qt]];ba?(dn===0&&(dn=Vc()),ut[Qt]=ba(Y[Qt])):ut[Qt]=Y[Qt]}var Xl=Te.apply(null,ut);return Xl=ye(Xl),dn!==0&&Hl(dn),Xl}function We(I,$,B,Y){B=B||[];var xe=B.every(function(ye){return ye==="number"}),ge=$!=="string";return ge&&xe&&!Y?Ce(I):function(){return Me(I,$,B,arguments,Y)}}function He(I,$,B){for(var Y=$+B,xe="";!($>=Y);){var ge=I[$++];if(!ge)return xe;if(!(ge&128)){xe+=String.fromCharCode(ge);continue}var ye=I[$++]&63;if((ge&224)==192){xe+=String.fromCharCode((ge&31)<<6|ye);continue}var Te=I[$++]&63;if((ge&240)==224?ge=(ge&15)<<12|ye<<6|Te:ge=(ge&7)<<18|ye<<12|Te<<6|I[$++]&63,ge<65536)xe+=String.fromCharCode(ge);else{var ut=ge-65536;xe+=String.fromCharCode(55296|ut>>10,56320|ut&1023)}}return xe}function qe(I,$){return I?He(o(),I,$):""}function ct(I,$,B,Y){if(!(Y>0))return 0;for(var xe=B,ge=B+Y-1,ye=0;ye<I.length;++ye){var Te=I.charCodeAt(ye);if(Te>=55296&&Te<=57343){var ut=I.charCodeAt(++ye);Te=65536+((Te&1023)<<10)|ut&1023}if(Te<=127){if(B>=ge)break;$[B++]=Te}else if(Te<=2047){if(B+1>=ge)break;$[B++]=192|Te>>6,$[B++]=128|Te&63}else if(Te<=65535){if(B+2>=ge)break;$[B++]=224|Te>>12,$[B++]=128|Te>>6&63,$[B++]=128|Te&63}else{if(B+3>=ge)break;$[B++]=240|Te>>18,$[B++]=128|Te>>12&63,$[B++]=128|Te>>6&63,$[B++]=128|Te&63}}return $[B]=0,B-xe}function dt(I,$,B){return ct(I,o(),$,B)}function rt(I){for(var $=0,B=0;B<I.length;++B){var Y=I.charCodeAt(B);Y>=55296&&Y<=57343&&(Y=65536+((Y&1023)<<10)|I.charCodeAt(++B)&1023),Y<=127?++$:Y<=2047?$+=2:Y<=65535?$+=3:$+=4}return $}function wt(I,$){a().set(I,$)}function ft(I,$){return I%$>0&&(I+=$-I%$),I}var Ct,_t,ws,wn,nr,Pn,ls,Ls,ks;function kn(I){Ct=I,u.HEAP8=_t=new Int8Array(I),u.HEAP16=wn=new Int16Array(I),u.HEAP32=Pn=new Int32Array(I),u.HEAPU8=ws=new Uint8Array(I),u.HEAPU16=nr=new Uint16Array(I),u.HEAPU32=ls=new Uint32Array(I),u.HEAPF32=Ls=new Float32Array(I),u.HEAPF64=ks=new Float64Array(I)}var vr=u.INITIAL_MEMORY||16777216;if(k)te=u.wasmMemory,Ct=u.buffer;else if(u.wasmMemory)te=u.wasmMemory;else if(te=new WebAssembly.Memory({initial:vr/65536,maximum:2147483648/65536,shared:!0}),!(te.buffer instanceof SharedArrayBuffer))throw z("requested a shared WebAssembly.Memory but the returned buffer is not a SharedArrayBuffer, indicating that while the browser has SharedArrayBuffer it does not have WebAssembly threads support - you may need to set a flag"),b&&console.log("(on node you may need: --experimental-wasm-threads --experimental-wasm-bulk-memory and also use a recent version)"),Error("bad memory");te&&(Ct=te.buffer),vr=Ct.byteLength,kn(Ct);var En,wr=[],kr=[],fa=[],_c=[],sr=[],zp=!1,T0=!1;k||kr.push({func:function(){nh()}});function Lp(){if(!k){if(u.preRun)for(typeof u.preRun=="function"&&(u.preRun=[u.preRun]);u.preRun.length;)N0(u.preRun.shift());Vl(wr)}}function Bp(){zp=!0,!k&&Vl(kr)}function Wp(){k||Vl(fa)}function Jn(){k||(T0=!0)}function Vp(){if(!k){if(u.postRun)for(typeof u.postRun=="function"&&(u.postRun=[u.postRun]);u.postRun.length;)E0(u.postRun.shift());Vl(sr)}}function N0(I){wr.unshift(I)}function E0(I){sr.unshift(I)}var Bs=0,Dc=null,si=null;function R0(I){we(!k,"addRunDependency cannot be used in a pthread worker"),Bs++,u.monitorRunDependencies&&u.monitorRunDependencies(Bs)}function $0(I){if(Bs--,u.monitorRunDependencies&&u.monitorRunDependencies(Bs),Bs==0&&(Dc!==null&&(clearInterval(Dc),Dc=null),si)){var $=si;si=null,$()}}u.preloadedImages={},u.preloadedAudios={};function Wr(I){u.onAbort&&u.onAbort(I),k&&console.error("Pthread aborting at "+new Error().stack),I+="",z(I),fe=!0,be=1,I="abort("+I+"). Build with -s ASSERTIONS=1 for more info.";var $=new WebAssembly.RuntimeError(I);throw p($),$}function ri(I,$){return String.prototype.startsWith?I.startsWith($):I.indexOf($)===0}var _0="data:application/octet-stream;base64,";function Up(I){return ri(I,_0)}var D0="file://";function Gp(I){return ri(I,D0)}var Qn="tfjs-backend-wasm-threaded-simd.wasm";Up(Qn)||(Qn=N(Qn));function P0(I){try{if(I==Qn&&ne)return new Uint8Array(ne);if(_)return _(I);throw"both async and sync fetching of the wasm failed"}catch($){Wr($)}}function Hp(){if(!ne&&(x||y)){if(typeof fetch=="function"&&!Gp(Qn))return fetch(Qn,{credentials:"same-origin"}).then(function(I){if(!I.ok)throw"failed to load wasm binary file at '"+Qn+"'";return I.arrayBuffer()}).catch(function(){return P0(Qn)});if(F)return new Promise(function(I,$){F(Qn,function(B){I(new Uint8Array(B))},$)})}return Promise.resolve().then(function(){return P0(Qn)})}function F0(){var I={a:Tg};function $(ye,Te){var ut=ye.exports;if(u.asm=ut,En=u.asm.kb,oe=Te,!k){var dn=Ee.unusedWorkers.length;Ee.unusedWorkers.forEach(function(Qt){Ee.loadWasmModuleToWorker(Qt,function(){--dn||$0("wasm-instantiate")})})}}k||R0("wasm-instantiate");function B(ye){$(ye.instance,ye.module)}function Y(ye){return Hp().then(function(Te){return WebAssembly.instantiate(Te,I)}).then(ye,function(Te){z("failed to asynchronously prepare wasm: "+Te),Wr(Te)})}function xe(){return!ne&&typeof WebAssembly.instantiateStreaming=="function"&&!Up(Qn)&&!Gp(Qn)&&typeof fetch=="function"?fetch(Qn,{credentials:"same-origin"}).then(function(ye){var Te=WebAssembly.instantiateStreaming(ye,I);return Te.then(B,function(ut){return z("wasm streaming compile failed: "+ut),z("falling back to ArrayBuffer instantiation"),Y(B)})}):Y(B)}if(u.instantiateWasm)try{var ge=u.instantiateWasm(I,$);return ge}catch(ye){return z("Module.instantiateWasm callback failed with error: "+ye),!1}return xe().catch(p),{}}var jp={10072:function(){throw"Canceled!"},10090:function(I,$){setTimeout(function(){l5(I,$)},0)}};function O0(){Ee.initRuntime()}function Vl(I){for(;I.length>0;){var $=I.shift();if(typeof $=="function"){$(u);continue}var B=$.func;typeof B=="number"?$.arg===void 0?En.get(B)():En.get(B)($.arg):B($.arg===void 0?null:$.arg)}}var ma={EPERM:63,ENOENT:44,ESRCH:71,EINTR:27,EIO:29,ENXIO:60,E2BIG:1,ENOEXEC:45,EBADF:8,ECHILD:12,EAGAIN:6,EWOULDBLOCK:6,ENOMEM:48,EACCES:2,EFAULT:21,ENOTBLK:105,EBUSY:10,EEXIST:20,EXDEV:75,ENODEV:43,ENOTDIR:54,EISDIR:31,EINVAL:28,ENFILE:41,EMFILE:33,ENOTTY:59,ETXTBSY:74,EFBIG:22,ENOSPC:51,ESPIPE:70,EROFS:69,EMLINK:34,EPIPE:64,EDOM:18,ERANGE:68,ENOMSG:49,EIDRM:24,ECHRNG:106,EL2NSYNC:156,EL3HLT:107,EL3RST:108,ELNRNG:109,EUNATCH:110,ENOCSI:111,EL2HLT:112,EDEADLK:16,ENOLCK:46,EBADE:113,EBADR:114,EXFULL:115,ENOANO:104,EBADRQC:103,EBADSLT:102,EDEADLOCK:16,EBFONT:101,ENOSTR:100,ENODATA:116,ETIME:117,ENOSR:118,ENONET:119,ENOPKG:120,EREMOTE:121,ENOLINK:47,EADV:122,ESRMNT:123,ECOMM:124,EPROTO:65,EMULTIHOP:36,EDOTDOT:125,EBADMSG:9,ENOTUNIQ:126,EBADFD:127,EREMCHG:128,ELIBACC:129,ELIBBAD:130,ELIBSCN:131,ELIBMAX:132,ELIBEXEC:133,ENOSYS:52,ENOTEMPTY:55,ENAMETOOLONG:37,ELOOP:32,EOPNOTSUPP:138,EPFNOSUPPORT:139,ECONNRESET:15,ENOBUFS:42,EAFNOSUPPORT:5,EPROTOTYPE:67,ENOTSOCK:57,ENOPROTOOPT:50,ESHUTDOWN:140,ECONNREFUSED:14,EADDRINUSE:3,ECONNABORTED:13,ENETUNREACH:40,ENETDOWN:38,ETIMEDOUT:73,EHOSTDOWN:142,EHOSTUNREACH:23,EINPROGRESS:26,EALREADY:7,EDESTADDRREQ:17,EMSGSIZE:35,EPROTONOSUPPORT:66,ESOCKTNOSUPPORT:137,EADDRNOTAVAIL:4,ENETRESET:39,EISCONN:30,ENOTCONN:53,ETOOMANYREFS:141,EUSERS:136,EDQUOT:19,ESTALE:72,ENOTSUP:138,ENOMEDIUM:148,EILSEQ:25,EOVERFLOW:61,ECANCELED:11,ENOTRECOVERABLE:56,EOWNERDEAD:62,ESTRPIPE:135};function Pc(I,$){if(I<=0||I>a().length||I&!0||$<0)return-28;if($==0)return 0;$>=2147483647&&($=1/0);var B=Atomics.load(i(),ql>>2),Y=0;if(B==I){var xe=Atomics.compareExchange(i(),ql>>2,B,0);if(xe==B&&(--$,Y=1,$<=0))return 1}var ge=Atomics.notify(i(),I>>2,$);if(ge>=0)return ge+Y;throw"Atomics.notify returned an unexpected value "+ge}u._emscripten_futex_wake=Pc;function M0(I){if(k)throw"Internal Error! killThread() can only ever be called from main application thread!";if(!I)throw"Internal Error! Null pthread_ptr in killThread!";i()[I+12>>2]=0;var $=Ee.pthreads[I];$.worker.terminate(),Ee.freeThreadData($),Ee.runningWorkers.splice(Ee.runningWorkers.indexOf($.worker),1),$.worker.pthread=void 0}function z0(I){if(k)throw"Internal Error! cancelThread() can only ever be called from main application thread!";if(!I)throw"Internal Error! Null pthread_ptr in cancelThread!";var $=Ee.pthreads[I];$.worker.postMessage({cmd:"cancel"})}function qp(I){if(k)throw"Internal Error! cleanupThread() can only ever be called from main application thread!";if(!I)throw"Internal Error! Null pthread_ptr in cleanupThread!";var $=Ee.pthreads[I];if($){i()[I+12>>2]=0;var B=$.worker;Ee.returnWorkerToPool(B)}}var Ee={unusedWorkers:[],runningWorkers:[],initMainThreadBlock:function(){for(var I=8,$=0;$<I;++$)Ee.allocateUnusedWorker()},initRuntime:function(){for(var I=oi(228),$=0;$<228/4;++$)l()[I/4+$]=0;i()[I+12>>2]=I;var B=I+152;i()[B>>2]=B;for(var Y=oi(512),$=0;$<128;++$)l()[Y/4+$]=0;Atomics.store(l(),I+100>>2,Y),Atomics.store(l(),I+40>>2,I),qg(I,!y,1),o5(I)},initWorker:function(){},pthreads:{},threadExitHandlers:[],setThreadStatus:function(){},runExitHandlers:function(){for(;Ee.threadExitHandlers.length>0;)Ee.threadExitHandlers.pop()();k&&xa()&&a5()},runExitHandlersAndDeinitThread:function(I,$){Atomics.store(l(),I+56>>2,1),Atomics.store(l(),I+60>>2,0),Ee.runExitHandlers(),Atomics.store(l(),I+4>>2,$),Atomics.store(l(),I+0>>2,1),Pc(I+0,2147483647),qg(0,0,0)},threadExit:function(I){var $=xa();$&&(Ee.runExitHandlersAndDeinitThread($,I),k&&postMessage({cmd:"exit"}))},threadCancel:function(){Ee.runExitHandlersAndDeinitThread(xa(),-1),postMessage({cmd:"cancelDone"})},terminateAllThreads:function(){for(var I in Ee.pthreads){var $=Ee.pthreads[I];$&&$.worker&&Ee.returnWorkerToPool($.worker)}Ee.pthreads={};for(var B=0;B<Ee.unusedWorkers.length;++B){var Y=Ee.unusedWorkers[B];Y.terminate()}Ee.unusedWorkers=[];for(var B=0;B<Ee.runningWorkers.length;++B){var Y=Ee.runningWorkers[B],$=Y.pthread;Ee.freeThreadData($),Y.terminate()}Ee.runningWorkers=[]},freeThreadData:function(I){if(!!I){if(I.threadInfoStruct){var $=i()[I.threadInfoStruct+100>>2];i()[I.threadInfoStruct+100>>2]=0,Wc($),Wc(I.threadInfoStruct)}I.threadInfoStruct=0,I.allocatedOwnStack&&I.stackBase&&Wc(I.stackBase),I.stackBase=0,I.worker&&(I.worker.pthread=null)}},returnWorkerToPool:function(I){Ee.runWithoutMainThreadQueuedCalls(function(){delete Ee.pthreads[I.pthread.threadInfoStruct],Ee.unusedWorkers.push(I),Ee.runningWorkers.splice(Ee.runningWorkers.indexOf(I),1),Ee.freeThreadData(I.pthread),I.pthread=void 0})},runWithoutMainThreadQueuedCalls:function(I){i()[h5>>2]=0;try{I()}finally{i()[h5>>2]=1}},receiveObjectTransfer:function(I){},loadWasmModuleToWorker:function(I,$){I.onmessage=function(B){var Y=B.data,xe=Y.cmd;if(I.pthread&&(Ee.currentProxiedOperationCallerThread=I.pthread.threadInfoStruct),Y.targetThread&&Y.targetThread!=xa()){var ge=Ee.pthreads[Y.targetThread];ge?ge.worker.postMessage(B.data,Y.transferList):console.error('Internal error! Worker sent a message "'+xe+'" to target pthread '+Y.targetThread+", but that thread no longer exists!"),Ee.currentProxiedOperationCallerThread=void 0;return}if(xe==="processQueuedMainThreadWork")ih();else if(xe==="spawnThread")eh(B.data);else if(xe==="cleanupThread")qp(Y.thread);else if(xe==="killThread")M0(Y.thread);else if(xe==="cancelThread")z0(Y.thread);else if(xe==="loaded")I.loaded=!0,$&&$(I),I.runPthread&&(I.runPthread(),delete I.runPthread);else if(xe==="print")j("Thread "+Y.threadId+": "+Y.text);else if(xe==="printErr")z("Thread "+Y.threadId+": "+Y.text);else if(xe==="alert")alert("Thread "+Y.threadId+": "+Y.text);else if(xe==="exit"){var ye=I.pthread&&Atomics.load(l(),I.pthread.threadInfoStruct+64>>2);ye&&Ee.returnWorkerToPool(I)}else if(xe==="exitProcess")try{zN(Y.returnCode)}catch(Te){if(Te instanceof Uc)return;throw Te}else xe==="cancelDone"?Ee.returnWorkerToPool(I):xe==="objectTransfer"?Ee.receiveObjectTransfer(B.data):B.data.target==="setimmediate"?I.postMessage(B.data):z("worker sent an unknown command "+xe);Ee.currentProxiedOperationCallerThread=void 0},I.onerror=function(B){z("pthread sent an error! "+B.filename+":"+B.lineno+": "+B.message)},b&&(I.on("message",function(B){I.onmessage({data:B})}),I.on("error",function(B){I.onerror(B)}),I.on("exit",function(B){})),I.postMessage({cmd:"load",urlOrBlob:u.mainScriptUrlOrBlob||s,wasmMemory:te,wasmModule:oe})},allocateUnusedWorker:function(){var I=N("tfjs-backend-wasm-threaded-simd.worker.js");Ee.unusedWorkers.push(new Worker(I))},getNewWorker:function(){return Ee.unusedWorkers.length==0&&(Ee.allocateUnusedWorker(),Ee.loadWasmModuleToWorker(Ee.unusedWorkers[0])),Ee.unusedWorkers.length>0?Ee.unusedWorkers.pop():null},busySpinWait:function(I){for(var $=performance.now()+I;performance.now()<$;);}};function L0(I,$){d5(I,$),Hl(I)}u.establishStackSpace=L0;function B0(){return Q}u.getNoExitRuntime=B0;function W0(I,$){return En.get(I)($)}u.invokeEntryPoint=W0;function V0(I,$,B,Y){Wr("Assertion failed: "+qe(I)+", at: "+[$?qe($):"unknown filename",B,Y?qe(Y):"unknown function"])}function U0(I,$){var B=_main(I,$)}var ai;b?ai=function(){var I=process.hrtime();return I[0]*1e3+I[1]/1e6}:k?ai=function(){return performance.now()-u.__performance_now_clock_drift}:typeof dateNow!="undefined"?ai=dateNow:ai=function(){return performance.now()};function G0(I){return i()[s5()>>2]=I,I}function H0(I,$){if(k)return ga(1,1,I,$)}function j0(I,$){if(I==$)postMessage({cmd:"processQueuedMainThreadWork"});else if(k)postMessage({targetThread:I,cmd:"processThreadQueue"});else{var B=Ee.pthreads[I],Y=B&&B.worker;if(!Y)return;Y.postMessage({cmd:"processThreadQueue"})}return 1}function q0(){Wr()}function X0(I,$,B){var Y=J0($,B);return jp[I].apply(null,Y)}function K0(I,$){}function Xp(I,$,B){if(I<=0||I>a().length||I&!0)return-28;if(x){if(Atomics.load(i(),I>>2)!=$)return-6;for(var xe=performance.now(),ge=xe+B,ye=Atomics.exchange(i(),ql>>2,I);;){if(xe=performance.now(),xe>ge)return ye=Atomics.exchange(i(),ql>>2,0),-73;if(ye=Atomics.exchange(i(),ql>>2,0),ye==0)break;if(ih(),Atomics.load(i(),I>>2)!=$)return-6;ye=Atomics.exchange(i(),ql>>2,I)}return 0}else{var Y=Atomics.wait(i(),I>>2,$,B);if(Y==="timed-out")return-73;if(Y==="not-equal")return-6;if(Y==="ok")return 0;throw"Atomics.wait returned an unexpected value "+Y}}function Z0(I,$,B){o().copyWithin(I,$,$+B)}function Y0(){return b?Ws("os").cpus().length:navigator.hardwareConcurrency}function ga(I,$){for(var B=arguments.length-2,Y=Vc(),xe=B,ge=jl(xe*8),ye=ge>>3,Te=0;Te<B;Te++){var ut=arguments[2+Te];c()[ye+Te]=ut}var dn=c5(I,xe,ge,$);return Hl(Y),dn}var Fc=[],Oc=[];function J0(I,$){Oc.length=0;var B;for($>>=2;B=o()[I++];){var Y=B<105;Y&&$&1&&$++,Oc.push(Y?c()[$++>>1]:i()[$]),++$}return Oc}function Q0(I,$,B){Fc.length=$;for(var Y=B>>3,xe=0;xe<$;xe++)Fc[xe]=c()[Y+xe];var ge=I<0,ye=ge?jp[-I-1]:Cg[I];return ye.apply(null,Fc)}function eg(){return o().length}function tg(I){try{return te.grow(I-Ct.byteLength+65535>>>16),kn(te.buffer),1}catch($){}}function ng(I){var $=eg();if(I<=$)return!1;var B=2147483648;if(I>B)return!1;for(var Y=1;Y<=4;Y*=2){var xe=$*(1+.2/Y);xe=Math.min(xe,I+100663296);var ge=Math.min(B,ft(Math.max(I,xe),65536)),ye=tg(ge);if(ye)return!0}return!1}var Ve={inEventHandler:0,removeAllEventListeners:function(){for(var I=Ve.eventHandlers.length-1;I>=0;--I)Ve._removeHandler(I);Ve.eventHandlers=[],Ve.deferredCalls=[]},registerRemoveEventListeners:function(){Ve.removeEventListenersRegistered||(_c.push(Ve.removeAllEventListeners),Ve.removeEventListenersRegistered=!0)},deferredCalls:[],deferCall:function(I,$,B){function Y(ye,Te){if(ye.length!=Te.length)return!1;for(var ut in ye)if(ye[ut]!=Te[ut])return!1;return!0}for(var xe in Ve.deferredCalls){var ge=Ve.deferredCalls[xe];if(ge.targetFunction==I&&Y(ge.argsList,B))return}Ve.deferredCalls.push({targetFunction:I,precedence:$,argsList:B}),Ve.deferredCalls.sort(function(ye,Te){return ye.precedence<Te.precedence})},removeDeferredCalls:function(I){for(var $=0;$<Ve.deferredCalls.length;++$)Ve.deferredCalls[$].targetFunction==I&&(Ve.deferredCalls.splice($,1),--$)},canPerformEventHandlerRequests:function(){return Ve.inEventHandler&&Ve.currentEventHandler.allowsDeferredCalls},runDeferredCalls:function(){if(!!Ve.canPerformEventHandlerRequests())for(var I=0;I<Ve.deferredCalls.length;++I){var $=Ve.deferredCalls[I];Ve.deferredCalls.splice(I,1),--I,$.targetFunction.apply(null,$.argsList)}},eventHandlers:[],removeAllHandlersOnTarget:function(I,$){for(var B=0;B<Ve.eventHandlers.length;++B)Ve.eventHandlers[B].target==I&&(!$||$==Ve.eventHandlers[B].eventTypeString)&&Ve._removeHandler(B--)},_removeHandler:function(I){var $=Ve.eventHandlers[I];$.target.removeEventListener($.eventTypeString,$.eventListenerFunc,$.useCapture),Ve.eventHandlers.splice(I,1)},registerOrRemoveHandler:function(I){var $=function(xe){++Ve.inEventHandler,Ve.currentEventHandler=I,Ve.runDeferredCalls(),I.handlerFunc(xe),Ve.runDeferredCalls(),--Ve.inEventHandler};if(I.callbackfunc)I.eventListenerFunc=$,I.target.addEventListener(I.eventTypeString,$,I.useCapture),Ve.eventHandlers.push(I),Ve.registerRemoveEventListeners();else for(var B=0;B<Ve.eventHandlers.length;++B)Ve.eventHandlers[B].target==I.target&&Ve.eventHandlers[B].eventTypeString==I.eventTypeString&&Ve._removeHandler(B--)},queueEventHandlerOnThread_iiii:function(I,$,B,Y,xe){var ge=Vc(),ye=jl(12);i()[ye>>2]=B,i()[ye+4>>2]=Y,i()[ye+8>>2]=xe,jg(0,I,637534208,$,Y,ye),Hl(ge)},getTargetThreadForEventCallback:function(I){switch(I){case 1:return 0;case 2:return Ee.currentProxiedOperationCallerThread;default:return I}},getNodeNameForTarget:function(I){return I?I==window?"#window":I==screen?"#screen":I&&I.nodeName?I.nodeName:"":""},fullscreenEnabled:function(){return document.fullscreenEnabled||document.webkitFullscreenEnabled}};function sg(I){var $=rt(I)+1,B=oi($);return dt(I,B,$),B}function rg(I,$,B,Y){var xe=Vc(),ge=jl(12),ye=0;$&&(ye=sg($)),i()[ge>>2]=ye,i()[ge+4>>2]=B,i()[ge+8>>2]=Y,jg(0,I,657457152,0,ye,ge),Hl(xe)}function ag(I,$,B,Y){$=$?qe($):"",rg(I,$,B,Y)}function og(I){return I>2?qe(I):I}var ig=[0,typeof document!="undefined"?document:0,typeof window!="undefined"?window:0];function lg(I){I=og(I);var $=ig[I]||(typeof document!="undefined"?document.querySelector(I):void 0);return $}function Mc(I){return lg(I)}function Kp(I,$,B){var Y=Mc(I);if(!Y)return-4;if(Y.canvasSharedPtr&&(i()[Y.canvasSharedPtr>>2]=$,i()[Y.canvasSharedPtr+4>>2]=B),Y.offscreenCanvas||!Y.controlTransferredOffscreen){Y.offscreenCanvas&&(Y=Y.offscreenCanvas);var xe=!1;if(Y.GLctxObject&&Y.GLctxObject.GLctx){var ge=Y.GLctxObject.GLctx.getParameter(2978);xe=ge[0]===0&&ge[1]===0&&ge[2]===Y.width&&ge[3]===Y.height}Y.width=$,Y.height=B,xe&&Y.GLctxObject.GLctx.viewport(0,0,$,B)}else if(Y.canvasSharedPtr){var ye=i()[Y.canvasSharedPtr+8>>2];return ag(ye,I,$,B),1}else return-4;return 0}function Zp(I,$,B){return k?ga(2,1,I,$,B):Kp(I,$,B)}function ug(I,$,B){var Y=Mc(I);return Y?Kp(I,$,B):Zp(I,$,B)}function cg(I){}function dg(I,$){}function pg(I){var $=I.getExtension("ANGLE_instanced_arrays");if($)return I.vertexAttribDivisor=function(B,Y){$.vertexAttribDivisorANGLE(B,Y)},I.drawArraysInstanced=function(B,Y,xe,ge){$.drawArraysInstancedANGLE(B,Y,xe,ge)},I.drawElementsInstanced=function(B,Y,xe,ge,ye){$.drawElementsInstancedANGLE(B,Y,xe,ge,ye)},1}function hg(I){var $=I.getExtension("OES_vertex_array_object");if($)return I.createVertexArray=function(){return $.createVertexArrayOES()},I.deleteVertexArray=function(B){$.deleteVertexArrayOES(B)},I.bindVertexArray=function(B){$.bindVertexArrayOES(B)},I.isVertexArray=function(B){return $.isVertexArrayOES(B)},1}function fg(I){var $=I.getExtension("WEBGL_draw_buffers");if($)return I.drawBuffers=function(B,Y){$.drawBuffersWEBGL(B,Y)},1}function mg(I){return!!(I.multiDrawWebgl=I.getExtension("WEBGL_multi_draw"))}var lt={counter:1,buffers:[],programs:[],framebuffers:[],renderbuffers:[],textures:[],uniforms:[],shaders:[],vaos:[],contexts:{},offscreenCanvases:{},timerQueriesEXT:[],programInfos:{},stringCache:{},unpackAlignment:4,recordError:function($){lt.lastError||(lt.lastError=$)},getNewId:function(I){for(var $=lt.counter++,B=I.length;B<$;B++)I[B]=null;return $},getSource:function(I,$,B,Y){for(var xe="",ge=0;ge<$;++ge){var ye=Y?i()[Y+ge*4>>2]:-1;xe+=qe(i()[B+ge*4>>2],ye<0?void 0:ye)}return xe},createContext:function(I,$){var B=I.getContext("webgl",$);if(!B)return 0;var Y=lt.registerContext(B,$);return Y},registerContext:function(I,$){var B=oi(8);i()[B+4>>2]=xa();var Y={handle:B,attributes:$,version:$.majorVersion,GLctx:I};return I.canvas&&(I.canvas.GLctxObject=Y),lt.contexts[B]=Y,(typeof $.enableExtensionsByDefault=="undefined"||$.enableExtensionsByDefault)&<.initExtensions(Y),B},makeContextCurrent:function(I){return lt.currentContext=lt.contexts[I],u.ctx=Aa=lt.currentContext&<.currentContext.GLctx,!(I&&!Aa)},getContext:function(I){return lt.contexts[I]},deleteContext:function(I){lt.currentContext===lt.contexts[I]&&(lt.currentContext=null),typeof Ve=="object"&&Ve.removeAllHandlersOnTarget(lt.contexts[I].GLctx.canvas),lt.contexts[I]&<.contexts[I].GLctx.canvas&&(lt.contexts[I].GLctx.canvas.GLctxObject=void 0),Wc(lt.contexts[I].handle),lt.contexts[I]=null},initExtensions:function(I){if(I||(I=lt.currentContext),!I.initExtensionsDone){I.initExtensionsDone=!0;var $=I.GLctx;pg($),hg($),fg($),$.disjointTimerQueryExt=$.getExtension("EXT_disjoint_timer_query"),mg($);var B=$.getSupportedExtensions()||[];B.forEach(function(Y){Y.indexOf("lose_context")<0&&Y.indexOf("debug")<0&&$.getExtension(Y)})}},populateUniformTable:function(I){for(var $=lt.programs[I],B=lt.programInfos[I]={uniforms:{},maxUniformLength:0,maxAttributeLength:-1,maxUniformBlockNameLength:-1},Y=B.uniforms,xe=Aa.getProgramParameter($,35718),ge=0;ge<xe;++ge){var ye=Aa.getActiveUniform($,ge),Te=ye.name;B.maxUniformLength=Math.max(B.maxUniformLength,Te.length+1),Te.slice(-1)=="]"&&(Te=Te.slice(0,Te.lastIndexOf("[")));var ut=Aa.getUniformLocation($,Te);if(ut){var dn=lt.getNewId(lt.uniforms);Y[Te]=[ye.size,dn],lt.uniforms[dn]=ut;for(var Qt=1;Qt<ye.size;++Qt){var ba=Te+"["+Qt+"]";ut=Aa.getUniformLocation($,ba),dn=lt.getNewId(lt.uniforms),lt.uniforms[dn]=ut}}}}},gg=["default","low-power","high-performance"];function Ag(I,$){var B=$>>2,Y=i()[B+(24>>2)],xe={alpha:!!i()[B+(0>>2)],depth:!!i()[B+(4>>2)],stencil:!!i()[B+(8>>2)],antialias:!!i()[B+(12>>2)],premultipliedAlpha:!!i()[B+(16>>2)],preserveDrawingBuffer:!!i()[B+(20>>2)],powerPreference:gg[Y],failIfMajorPerformanceCaveat:!!i()[B+(28>>2)],majorVersion:i()[B+(32>>2)],minorVersion:i()[B+(36>>2)],enableExtensionsByDefault:i()[B+(40>>2)],explicitSwapControl:i()[B+(44>>2)],proxyContextToMainThread:i()[B+(48>>2)],renderViaOffscreenBackBuffer:i()[B+(52>>2)]},ge=Mc(I);if(!ge||xe.explicitSwapControl)return 0;var ye=lt.createContext(ge,xe);return ye}function yg(I,$){return Ag(I,$)}var Ul={mappings:{},buffers:[null,[],[]],printChar:function(I,$){var B=Ul.buffers[I];$===0||$===10?((I===1?j:z)(He(B,0)),B.length=0):B.push($)},varargs:void 0,get:function(){Ul.varargs+=4;var I=i()[Ul.varargs-4>>2];return I},getStr:function(I){var $=qe(I);return $},get64:function(I,$){return I}};function Yp(I){return k?ga(3,1,I):0}function Jp(I,$,B,Y,xe){if(k)return ga(4,1,I,$,B,Y,xe)}function Qp(I,$,B,Y){if(k)return ga(5,1,I,$,B,Y);for(var xe=0,ge=0;ge<B;ge++){for(var ye=i()[$+ge*8>>2],Te=i()[$+(ge*8+4)>>2],ut=0;ut<Te;ut++)Ul.printChar(I,o()[ye+ut]);xe+=Te}return i()[Y>>2]=xe,0}function xg(I){var $=Ee.threadExitHandlers.pop();I&&$()}function bg(I,$){Ee.threadExitHandlers.push(function(){En.get(I)($)})}function eh(I){if(k)throw"Internal Error! spawnThread() can only ever be called from main application thread!";var $=Ee.getNewWorker();if($.pthread!==void 0)throw"Internal error!";if(!I.pthread_ptr)throw"Internal error, no pthread ptr!";Ee.runningWorkers.push($);for(var B=oi(128*4),Y=0;Y<128;++Y)i()[B+Y*4>>2]=0;var xe=I.stackBase+I.stackSize,ge=Ee.pthreads[I.pthread_ptr]={worker:$,stackBase:I.stackBase,stackSize:I.stackSize,allocatedOwnStack:I.allocatedOwnStack,threadInfoStruct:I.pthread_ptr},ye=ge.threadInfoStruct>>2;Atomics.store(l(),ye+(64>>2),I.detached),Atomics.store(l(),ye+(100>>2),B),Atomics.store(l(),ye+(40>>2),ge.threadInfoStruct),Atomics.store(l(),ye+(80>>2),I.stackSize),Atomics.store(l(),ye+(76>>2),xe),Atomics.store(l(),ye+(104>>2),I.stackSize),Atomics.store(l(),ye+(104+8>>2),xe),Atomics.store(l(),ye+(104+12>>2),I.detached);var Te=r5(),ut=Te+40;Atomics.store(l(),ye+(172>>2),ut),$.pthread=ge;var dn={cmd:"run",start_routine:I.startRoutine,arg:I.arg,threadInfoStruct:I.pthread_ptr,stackBase:I.stackBase,stackSize:I.stackSize};$.runPthread=function(){dn.time=performance.now(),$.postMessage(dn,I.transferList)},$.loaded&&($.runPthread(),delete $.runPthread)}function vg(I,$,B,Y){if(typeof SharedArrayBuffer=="undefined")return z("Current environment does not support SharedArrayBuffer, pthreads are not available!"),6;if(!I)return z("pthread_create called with a null thread pointer!"),28;var xe=[],ge=0;if(k&&(xe.length===0||ge))return u5(687865856,I,$,B,Y);if(ge)return ge;var ye=0,Te=0,ut=0;$&&$!=-1?(ye=i()[$>>2],ye+=81920,Te=i()[$+8>>2],ut=i()[$+12>>2]!==0):ye=2097152;var dn=Te==0;dn?Te=p5(16,ye):(Te-=ye,we(Te>0));for(var Qt=oi(228),ba=0;ba<228>>2;++ba)l()[(Qt>>2)+ba]=0;i()[I>>2]=Qt,i()[Qt+12>>2]=Qt;var Xl=Qt+152;i()[Xl>>2]=Xl;var Fn={stackBase:Te,stackSize:ye,allocatedOwnStack:dn,detached:ut,startRoutine:B,pthread_ptr:Qt,arg:Y,transferList:xe};return k?(Fn.cmd="spawnThread",postMessage(Fn,xe)):eh(Fn),0}function wg(){if(!!k){var I=xa();if(!!I){var $=Atomics.load(l(),I+56>>2);if(!$){var B=Atomics.load(l(),I+0>>2);if(B==2)throw"Canceled!"}}}}function kg(){b||y||X("Blocking on the main thread is very dangerous, see https://emscripten.org/docs/porting/pthreads.html#blocking-on-the-main-browser-thread")}function Sg(I,$,B){if(!I)return z("pthread_join attempted on a null thread pointer!"),ma.ESRCH;if(k&&xa()==I)return z("PThread "+I+" is attempting to join to itself!"),ma.EDEADLK;if(!k&&i5()==I)return z("Main thread "+I+" is attempting to join to itself!"),ma.EDEADLK;var Y=i()[I+12>>2];if(Y!==I)return z("pthread_join attempted on thread "+I+", which does not point to a valid thread, or does not exist anymore!"),ma.ESRCH;var xe=Atomics.load(l(),I+64>>2);if(xe)return z("Attempted to join thread "+I+", which was already detached!"),ma.EINVAL;for(B&&kg();;){var ge=Atomics.load(l(),I+0>>2);if(ge==1){var ye=Atomics.load(l(),I+4>>2);return $&&(i()[$>>2]=ye),Atomics.store(l(),I+64>>2,1),k?postMessage({cmd:"cleanupThread",thread:I}):qp(I),0}if(!B)return ma.EBUSY;wg(),k||ih(),Xp(I+0,ge,k?100:1)}}function Ig(I,$){return Sg(I,$,!0)}function th(I){if(k)return ga(6,1,I);switch(I){case 30:return 16384;case 85:var $=2147483648;return $/16384;case 132:case 133:case 12:case 137:case 138:case 15:case 235:case 16:case 17:case 18:case 19:case 20:case 149:case 13:case 10:case 236:case 153:case 9:case 21:case 22:case 159:case 154:case 14:case 77:case 78:case 139:case 82:case 68:case 67:case 164:case 11:case 29:case 47:case 48:case 95:case 52:case 51:case 46:return 200809;case 27:case 246:case 127:case 128:case 23:case 24:case 160:case 161:case 181:case 182:case 242:case 183:case 184:case 243:case 244:case 245:case 165:case 178:case 179:case 49:case 50:case 168:case 169:case 175:case 170:case 171:case 172:case 97:case 76:case 32:case 173:case 35:case 80:case 81:case 79:return-1;case 176:case 177:case 7:case 155:case 8:case 157:case 125:case 126:case 92:case 93:case 129:case 130:case 131:case 94:case 91:return 1;case 74:case 60:case 69:case 70:case 4:return 1024;case 31:case 42:case 72:return 32;case 87:case 26:case 33:return 2147483647;case 34:case 1:return 47839;case 38:case 36:return 99;case 43:case 37:return 2048;case 0:return 2097152;case 3:return 65536;case 28:return 32768;case 44:return 32767;case 75:return 16384;case 39:return 1e3;case 89:return 700;case 71:return 256;case 40:return 255;case 2:return 100;case 180:return 64;case 25:return 20;case 5:return 16;case 6:return 6;case 73:return 4;case 84:return typeof navigator=="object"&&navigator.hardwareConcurrency||1}return G0(28),-1}k||Ee.initMainThreadBlock();var Aa,Cg=[null,H0,Zp,Yp,Jp,Qp,th],Tg={e:V0,r:U0,x:j0,b:q0,y:X0,j:K0,d:Xp,c:Pc,f:ai,p:Z0,A:Y0,u:Q0,q:ng,v:ug,i:cg,s:dg,w:yg,l:Yp,n:Jp,g:Qp,o:O0,a:te||u.wasmMemory,z:xg,k:bg,h:vg,m:Ig,t:th},n5=F0(),nh=u.___wasm_call_ctors=function(){return(nh=u.___wasm_call_ctors=u.asm.B).apply(null,arguments)},Ng=u._init=function(){return(Ng=u._init=u.asm.C).apply(null,arguments)},Eg=u._init_with_threads_count=function(){return(Eg=u._init_with_threads_count=u.asm.D).apply(null,arguments)},Rg=u._get_threads_count=function(){return(Rg=u._get_threads_count=u.asm.E).apply(null,arguments)},$g=u._register_tensor=function(){return($g=u._register_tensor=u.asm.F).apply(null,arguments)},_g=u._dispose_data=function(){return(_g=u._dispose_data=u.asm.G).apply(null,arguments)},Dg=u._dispose=function(){return(Dg=u._dispose=u.asm.H).apply(null,arguments)},Pg=u._Abs=function(){return(Pg=u._Abs=u.asm.I).apply(null,arguments)},Fg=u._Add=function(){return(Fg=u._Add=u.asm.J).apply(null,arguments)},Og=u._AddN=function(){return(Og=u._AddN=u.asm.K).apply(null,arguments)},Mg=u._All=function(){return(Mg=u._All=u.asm.L).apply(null,arguments)},zg=u._Any=function(){return(zg=u._Any=u.asm.M).apply(null,arguments)},Lg=u._ArgMax=function(){return(Lg=u._ArgMax=u.asm.N).apply(null,arguments)},Bg=u._AvgPool=function(){return(Bg=u._AvgPool=u.asm.O).apply(null,arguments)},Wg=u._BatchMatMul=function(){return(Wg=u._BatchMatMul=u.asm.P).apply(null,arguments)},Vg=u._Ceil=function(){return(Vg=u._Ceil=u.asm.Q).apply(null,arguments)},Ug=u._ClipByValue=function(){return(Ug=u._ClipByValue=u.asm.R).apply(null,arguments)},Gg=u._Conv2D=function(){return(Gg=u._Conv2D=u.asm.S).apply(null,arguments)},sh=u._Conv2DBackpropInput=function(){return(sh=u._Conv2DBackpropInput=u.asm.T).apply(null,arguments)},rh=u._Cos=function(){return(rh=u._Cos=u.asm.U).apply(null,arguments)},zc=u._Cosh=function(){return(zc=u._Cosh=u.asm.V).apply(null,arguments)},Gl=u._CropAndResize=function(){return(Gl=u._CropAndResize=u.asm.W).apply(null,arguments)},Hg=u._Cumsum=function(){return(Hg=u._Cumsum=u.asm.X).apply(null,arguments)},Lc=u._DepthToSpace=function(){return(Lc=u._DepthToSpace=u.asm.Y).apply(null,arguments)},ae=u._DepthwiseConv2dNative=function(){return(ae=u._DepthwiseConv2dNative=u.asm.Z).apply(null,arguments)},ie=u._Elu=function(){return(ie=u._Elu=u.asm._).apply(null,arguments)},ke=u._Equal=function(){return(ke=u._Equal=u.asm.$).apply(null,arguments)},at=u._Exp=function(){return(at=u._Exp=u.asm.aa).apply(null,arguments)},zt=u._FlipLeftRight=function(){return(zt=u._FlipLeftRight=u.asm.ba).apply(null,arguments)},Tt=u._Floor=function(){return(Tt=u._Floor=u.asm.ca).apply(null,arguments)},Xe=u._FloorDiv=function(){return(Xe=u._FloorDiv=u.asm.da).apply(null,arguments)},Ke=u._FusedBatchNorm=function(){return(Ke=u._FusedBatchNorm=u.asm.ea).apply(null,arguments)},Sn=u._FusedConv2D=function(){return(Sn=u._FusedConv2D=u.asm.fa).apply(null,arguments)},Vr=u._FusedDepthwiseConv2D=function(){return(Vr=u._FusedDepthwiseConv2D=u.asm.ga).apply(null,arguments)},Ur=u._Gather=function(){return(Ur=u._Gather=u.asm.ha).apply(null,arguments)},ah=u._GatherNd=function(){return(ah=u._GatherNd=u.asm.ia).apply(null,arguments)},Bc=u._Greater=function(){return(Bc=u._Greater=u.asm.ja).apply(null,arguments)},us=u._GreaterEqual=function(){return(us=u._GreaterEqual=u.asm.ka).apply(null,arguments)},ya=u._LeakyRelu=function(){return(ya=u._LeakyRelu=u.asm.la).apply(null,arguments)},oh=u._Less=function(){return(oh=u._Less=u.asm.ma).apply(null,arguments)},VT=u._LessEqual=function(){return(VT=u._LessEqual=u.asm.na).apply(null,arguments)},UT=u._Log=function(){return(UT=u._Log=u.asm.oa).apply(null,arguments)},GT=u._LogicalAnd=function(){return(GT=u._LogicalAnd=u.asm.pa).apply(null,arguments)},HT=u._Max=function(){return(HT=u._Max=u.asm.qa).apply(null,arguments)},jT=u._MaxPool=function(){return(jT=u._MaxPool=u.asm.ra).apply(null,arguments)},qT=u._Maximum=function(){return(qT=u._Maximum=u.asm.sa).apply(null,arguments)},XT=u._Mean=function(){return(XT=u._Mean=u.asm.ta).apply(null,arguments)},KT=u._Min=function(){return(KT=u._Min=u.asm.ua).apply(null,arguments)},ZT=u._Minimum=function(){return(ZT=u._Minimum=u.asm.va).apply(null,arguments)},YT=u._MirrorPad=function(){return(YT=u._MirrorPad=u.asm.wa).apply(null,arguments)},JT=u._Multiply=function(){return(JT=u._Multiply=u.asm.xa).apply(null,arguments)},QT=u._Neg=function(){return(QT=u._Neg=u.asm.ya).apply(null,arguments)},eN=u._NonMaxSuppressionV3=function(){return(eN=u._NonMaxSuppressionV3=u.asm.za).apply(null,arguments)},tN=u._NonMaxSuppressionV4=function(){return(tN=u._NonMaxSuppressionV4=u.asm.Aa).apply(null,arguments)},nN=u._NonMaxSuppressionV5=function(){return(nN=u._NonMaxSuppressionV5=u.asm.Ba).apply(null,arguments)},sN=u._NotEqual=function(){return(sN=u._NotEqual=u.asm.Ca).apply(null,arguments)},rN=u._OneHot=function(){return(rN=u._OneHot=u.asm.Da).apply(null,arguments)},aN=u._PadV2=function(){return(aN=u._PadV2=u.asm.Ea).apply(null,arguments)},oN=u._Pow=function(){return(oN=u._Pow=u.asm.Fa).apply(null,arguments)},iN=u._Prelu=function(){return(iN=u._Prelu=u.asm.Ga).apply(null,arguments)},lN=u._Prod=function(){return(lN=u._Prod=u.asm.Ha).apply(null,arguments)},uN=u._RealDiv=function(){return(uN=u._RealDiv=u.asm.Ia).apply(null,arguments)},cN=u._Relu=function(){return(cN=u._Relu=u.asm.Ja).apply(null,arguments)},dN=u._Relu6=function(){return(dN=u._Relu6=u.asm.Ka).apply(null,arguments)},pN=u._ResizeBilinear=function(){return(pN=u._ResizeBilinear=u.asm.La).apply(null,arguments)},hN=u._Reverse=function(){return(hN=u._Reverse=u.asm.Ma).apply(null,arguments)},fN=u._RotateWithOffset=function(){return(fN=u._RotateWithOffset=u.asm.Na).apply(null,arguments)},mN=u._Round=function(){return(mN=u._Round=u.asm.Oa).apply(null,arguments)},gN=u._Rsqrt=function(){return(gN=u._Rsqrt=u.asm.Pa).apply(null,arguments)},AN=u._ScatterNd=function(){return(AN=u._ScatterNd=u.asm.Qa).apply(null,arguments)},yN=u._SelectV2=function(){return(yN=u._SelectV2=u.asm.Ra).apply(null,arguments)},xN=u._Sigmoid=function(){return(xN=u._Sigmoid=u.asm.Sa).apply(null,arguments)},bN=u._Sin=function(){return(bN=u._Sin=u.asm.Ta).apply(null,arguments)},vN=u._Softmax=function(){return(vN=u._Softmax=u.asm.Ua).apply(null,arguments)},wN=u._Sqrt=function(){return(wN=u._Sqrt=u.asm.Va).apply(null,arguments)},kN=u._Square=function(){return(kN=u._Square=u.asm.Wa).apply(null,arguments)},SN=u._SquaredDifference=function(){return(SN=u._SquaredDifference=u.asm.Xa).apply(null,arguments)},IN=u._Step=function(){return(IN=u._Step=u.asm.Ya).apply(null,arguments)},CN=u._StridedSlice=function(){return(CN=u._StridedSlice=u.asm.Za).apply(null,arguments)},TN=u._Sub=function(){return(TN=u._Sub=u.asm._a).apply(null,arguments)},NN=u._Sum=function(){return(NN=u._Sum=u.asm.$a).apply(null,arguments)},EN=u._Tan=function(){return(EN=u._Tan=u.asm.ab).apply(null,arguments)},RN=u._Tanh=function(){return(RN=u._Tanh=u.asm.bb).apply(null,arguments)},$N=u._Tile=function(){return($N=u._Tile=u.asm.cb).apply(null,arguments)},_N=u._TopK=function(){return(_N=u._TopK=u.asm.db).apply(null,arguments)},DN=u._Transform=function(){return(DN=u._Transform=u.asm.eb).apply(null,arguments)},PN=u._Transpose=function(){return(PN=u._Transpose=u.asm.fb).apply(null,arguments)},FN=u.__FusedMatMul=function(){return(FN=u.__FusedMatMul=u.asm.gb).apply(null,arguments)},oi=u._malloc=function(){return(oi=u._malloc=u.asm.hb).apply(null,arguments)},Wc=u._free=function(){return(Wc=u._free=u.asm.ib).apply(null,arguments)},s5=u.___errno_location=function(){return(s5=u.___errno_location=u.asm.jb).apply(null,arguments)},r5=u._emscripten_get_global_libc=function(){return(r5=u._emscripten_get_global_libc=u.asm.lb).apply(null,arguments)},xa=u._pthread_self=function(){return(xa=u._pthread_self=u.asm.mb).apply(null,arguments)},a5=u.___pthread_tsd_run_dtors=function(){return(a5=u.___pthread_tsd_run_dtors=u.asm.nb).apply(null,arguments)},ih=u._emscripten_main_thread_process_queued_calls=function(){return(ih=u._emscripten_main_thread_process_queued_calls=u.asm.ob).apply(null,arguments)},ON=u._emscripten_current_thread_process_queued_calls=function(){return(ON=u._emscripten_current_thread_process_queued_calls=u.asm.pb).apply(null,arguments)},o5=u._emscripten_register_main_browser_thread_id=function(){return(o5=u._emscripten_register_main_browser_thread_id=u.asm.qb).apply(null,arguments)},i5=u._emscripten_main_browser_thread_id=function(){return(i5=u._emscripten_main_browser_thread_id=u.asm.rb).apply(null,arguments)},l5=u.__emscripten_do_dispatch_to_thread=function(){return(l5=u.__emscripten_do_dispatch_to_thread=u.asm.sb).apply(null,arguments)},u5=u._emscripten_sync_run_in_main_thread_4=function(){return(u5=u._emscripten_sync_run_in_main_thread_4=u.asm.tb).apply(null,arguments)},c5=u._emscripten_run_in_main_runtime_thread_js=function(){return(c5=u._emscripten_run_in_main_runtime_thread_js=u.asm.ub).apply(null,arguments)},jg=u.__emscripten_call_on_thread=function(){return(jg=u.__emscripten_call_on_thread=u.asm.vb).apply(null,arguments)},MN=u._emscripten_tls_init=function(){return(MN=u._emscripten_tls_init=u.asm.wb).apply(null,arguments)},qg=u.__emscripten_thread_init=function(){return(qg=u.__emscripten_thread_init=u.asm.xb).apply(null,arguments)},Vc=u.stackSave=function(){return(Vc=u.stackSave=u.asm.yb).apply(null,arguments)},Hl=u.stackRestore=function(){return(Hl=u.stackRestore=u.asm.zb).apply(null,arguments)},jl=u.stackAlloc=function(){return(jl=u.stackAlloc=u.asm.Ab).apply(null,arguments)},d5=u._emscripten_stack_set_limits=function(){return(d5=u._emscripten_stack_set_limits=u.asm.Bb).apply(null,arguments)},p5=u._memalign=function(){return(p5=u._memalign=u.asm.Cb).apply(null,arguments)},h5=u.__emscripten_allow_main_runtime_queued_calls=10064,ql=u.__emscripten_main_thread_futex=10268;u.cwrap=We,u.PThread=Ee,u.PThread=Ee,u.wasmMemory=te,u.ExitStatus=Uc;var lh;function Uc(I){this.name="ExitStatus",this.message="Program terminated with exit("+I+")",this.status=I}si=function I(){lh||Xg(),lh||(si=I)};function Xg(I){if(I=I||m,Bs>0)return;if(k){d(u),Bp(),postMessage({cmd:"loaded"});return}if(Lp(),Bs>0)return;function $(){lh||(lh=!0,u.calledRun=!0,!fe&&(Bp(),Wp(),d(u),u.onRuntimeInitialized&&u.onRuntimeInitialized(),Vp()))}u.setStatus?(u.setStatus("Running..."),setTimeout(function(){setTimeout(function(){u.setStatus("")},1),$()},1)):$()}u.run=Xg;function zN(I,$){if(!($&&Q&&I===0)){if(!$&&k)throw postMessage({cmd:"exitProcess",returnCode:I}),new Uc(I);Q||(Ee.terminateAllThreads(),be=I,Jn(),u.onExit&&u.onExit(I),fe=!0),A(I,new Uc(I))}}if(u.preInit)for(typeof u.preInit=="function"&&(u.preInit=[u.preInit]);u.preInit.length>0;)u.preInit.pop()();return k&&(Q=!1,Ee.initWorker()),Xg(),r.ready}}();typeof e=="object"&&typeof t=="object"?t.exports=n:typeof define=="function"&&define.amd?define([],function(){return n}):typeof e=="object"&&(e.WasmBackendModuleThreadedSimd=n)}}),aE=es({"src/tfjs-backend-wasm/wasm-out/tfjs-backend-wasm.js"(e,t){var n=function(){var s=typeof document!="undefined"&&document.currentScript?document.currentScript.src:void 0;return typeof __filename!="undefined"&&(s=s||__filename),function(r){r=r||{};var a=typeof r!="undefined"?r:{},o,i;a.ready=new Promise(function(ae,ie){o=ae,i=ie});var l={},c;for(c in a)a.hasOwnProperty(c)&&(l[c]=a[c]);var u=[],d="./this.program",p=function(ae,ie){throw ie},h=!1,f=!1,m=!1,g=!1;h=typeof window=="object",f=typeof importScripts=="function",m=typeof process=="object"&&typeof process.versions=="object"&&typeof process.versions.node=="string",g=!h&&!m&&!f;var A="";function x(ae){return a.locateFile?a.locateFile(ae,A):A+ae}var y,b,w,k,C,N;m?(f?A=Ws("path").dirname(A)+"/":A=__dirname+"/",y=function(ie,ke){return C||(C=Ws("fs")),N||(N=Ws("path")),ie=N.normalize(ie),C.readFileSync(ie,ke?null:"utf8")},w=function(ie){var ke=y(ie,!0);return ke.buffer||(ke=new Uint8Array(ke)),j(ke.buffer),ke},process.argv.length>1&&(d=process.argv[1].replace(/\\/g,"/")),u=process.argv.slice(2),process.on("uncaughtException",function(ae){if(!(ae instanceof Hg))throw ae}),process.on("unhandledRejection",sr),p=function(ae){process.exit(ae)},a.inspect=function(){return"[Emscripten Module object]"}):g?(typeof read!="undefined"&&(y=function(ie){return read(ie)}),w=function(ie){var ke;return typeof readbuffer=="function"?new Uint8Array(readbuffer(ie)):(ke=read(ie,"binary"),j(typeof ke=="object"),ke)},typeof scriptArgs!="undefined"?u=scriptArgs:typeof arguments!="undefined"&&(u=arguments),typeof quit=="function"&&(p=function(ae){quit(ae)}),typeof print!="undefined"&&(typeof console=="undefined"&&(console={}),console.log=print,console.warn=console.error=typeof printErr!="undefined"?printErr:print)):(h||f)&&(f?A=self.location.href:typeof document!="undefined"&&document.currentScript&&(A=document.currentScript.src),s&&(A=s),A.indexOf("blob:")!==0?A=A.substr(0,A.lastIndexOf("/")+1):A="",y=function(ae){var ie=new XMLHttpRequest;return ie.open("GET",ae,!1),ie.send(null),ie.responseText},f&&(w=function(ae){var ie=new XMLHttpRequest;return ie.open("GET",ae,!1),ie.responseType="arraybuffer",ie.send(null),new Uint8Array(ie.response)}),b=function(ae,ie,ke){var at=new XMLHttpRequest;at.open("GET",ae,!0),at.responseType="arraybuffer",at.onload=function(){if(at.status==200||at.status==0&&at.response){ie(at.response);return}ke()},at.onerror=ke,at.send(null)},k=function(ae){document.title=ae});var R=a.print||console.log.bind(console),F=a.printErr||console.warn.bind(console);for(c in l)l.hasOwnProperty(c)&&(a[c]=l[c]);l=null,a.arguments&&(u=a.arguments),a.thisProgram&&(d=a.thisProgram),a.quit&&(p=a.quit);var _;a.wasmBinary&&(_=a.wasmBinary);var P=a.noExitRuntime||!0;typeof WebAssembly!="object"&&sr("no native wasm support detected");var T,M=!1,U;function j(ae,ie){ae||sr("Assertion failed: "+ie)}function z(ae){var ie=a["_"+ae];return j(ie,"Cannot call unknown function "+ae+", make sure it is exported"),ie}function X(ae,ie,ke,at,zt){var Tt={string:function(us){var ya=0;if(us!=null&&us!==0){var oh=(us.length<<2)+1;ya=zc(oh),te(us,ya,oh)}return ya},array:function(us){var ya=zc(us.length);return oe(us,ya),ya}};function Xe(us){return ie==="string"?ne(us):ie==="boolean"?Boolean(us):us}var Ke=z(ae),Sn=[],Vr=0;if(at)for(var Ur=0;Ur<at.length;Ur++){var ah=Tt[ke[Ur]];ah?(Vr===0&&(Vr=sh()),Sn[Ur]=ah(at[Ur])):Sn[Ur]=at[Ur]}var Bc=Ke.apply(null,Sn);return Bc=Xe(Bc),Vr!==0&&rh(Vr),Bc}function Z(ae,ie,ke,at){ke=ke||[];var zt=ke.every(function(Xe){return Xe==="number"}),Tt=ie!=="string";return Tt&&zt&&!at?z(ae):function(){return X(ae,ie,ke,arguments,at)}}var J=typeof TextDecoder!="undefined"?new TextDecoder("utf8"):void 0;function ee(ae,ie,ke){for(var at=ie+ke,zt=ie;ae[zt]&&!(zt>=at);)++zt;if(zt-ie>16&&ae.subarray&&J)return J.decode(ae.subarray(ie,zt));for(var Tt="";ie<zt;){var Xe=ae[ie++];if(!(Xe&128)){Tt+=String.fromCharCode(Xe);continue}var Ke=ae[ie++]&63;if((Xe&224)==192){Tt+=String.fromCharCode((Xe&31)<<6|Ke);continue}var Sn=ae[ie++]&63;if((Xe&240)==224?Xe=(Xe&15)<<12|Ke<<6|Sn:Xe=(Xe&7)<<18|Ke<<12|Sn<<6|ae[ie++]&63,Xe<65536)Tt+=String.fromCharCode(Xe);else{var Vr=Xe-65536;Tt+=String.fromCharCode(55296|Vr>>10,56320|Vr&1023)}}return Tt}function ne(ae,ie){return ae?ee(Ce,ae,ie):""}function Q(ae,ie,ke,at){if(!(at>0))return 0;for(var zt=ke,Tt=ke+at-1,Xe=0;Xe<ae.length;++Xe){var Ke=ae.charCodeAt(Xe);if(Ke>=55296&&Ke<=57343){var Sn=ae.charCodeAt(++Xe);Ke=65536+((Ke&1023)<<10)|Sn&1023}if(Ke<=127){if(ke>=Tt)break;ie[ke++]=Ke}else if(Ke<=2047){if(ke+1>=Tt)break;ie[ke++]=192|Ke>>6,ie[ke++]=128|Ke&63}else if(Ke<=65535){if(ke+2>=Tt)break;ie[ke++]=224|Ke>>12,ie[ke++]=128|Ke>>6&63,ie[ke++]=128|Ke&63}else{if(ke+3>=Tt)break;ie[ke++]=240|Ke>>18,ie[ke++]=128|Ke>>12&63,ie[ke++]=128|Ke>>6&63,ie[ke++]=128|Ke&63}}return ie[ke]=0,ke-zt}function te(ae,ie,ke){return Q(ae,Ce,ie,ke)}function oe(ae,ie){we.set(ae,ie)}function fe(ae,ie){return ae%ie>0&&(ae+=ie-ae%ie),ae}var be,we,Ce,Me,We,He,qe,ct,dt;function rt(ae){be=ae,a.HEAP8=we=new Int8Array(ae),a.HEAP16=Me=new Int16Array(ae),a.HEAP32=He=new Int32Array(ae),a.HEAPU8=Ce=new Uint8Array(ae),a.HEAPU16=We=new Uint16Array(ae),a.HEAPU32=qe=new Uint32Array(ae),a.HEAPF32=ct=new Float32Array(ae),a.HEAPF64=dt=new Float64Array(ae)}var wt=a.INITIAL_MEMORY||16777216,ft,Ct=[],_t=[],ws=[],wn=[],nr=!1;_t.push({func:function(){Hp()}});function Pn(){if(a.preRun)for(typeof a.preRun=="function"&&(a.preRun=[a.preRun]);a.preRun.length;)kn(a.preRun.shift());Bs(Ct)}function ls(){nr=!0,Bs(_t)}function Ls(){Bs(ws)}function ks(){if(a.postRun)for(typeof a.postRun=="function"&&(a.postRun=[a.postRun]);a.postRun.length;)vr(a.postRun.shift());Bs(wn)}function kn(ae){Ct.unshift(ae)}function vr(ae){wn.unshift(ae)}var En=0,wr=null,kr=null;function fa(ae){En++,a.monitorRunDependencies&&a.monitorRunDependencies(En)}function _c(ae){if(En--,a.monitorRunDependencies&&a.monitorRunDependencies(En),En==0&&(wr!==null&&(clearInterval(wr),wr=null),kr)){var ie=kr;kr=null,ie()}}a.preloadedImages={},a.preloadedAudios={};function sr(ae){a.onAbort&&a.onAbort(ae),ae+="",F(ae),M=!0,U=1,ae="abort("+ae+"). Build with -s ASSERTIONS=1 for more info.";var ie=new WebAssembly.RuntimeError(ae);throw i(ie),ie}function zp(ae,ie){return String.prototype.startsWith?ae.startsWith(ie):ae.indexOf(ie)===0}var T0="data:application/octet-stream;base64,";function Lp(ae){return zp(ae,T0)}var Bp="file://";function Wp(ae){return zp(ae,Bp)}var Jn="tfjs-backend-wasm.wasm";Lp(Jn)||(Jn=x(Jn));function Vp(ae){try{if(ae==Jn&&_)return new Uint8Array(_);if(w)return w(ae);throw"both async and sync fetching of the wasm failed"}catch(ie){sr(ie)}}function N0(){if(!_&&(h||f)){if(typeof fetch=="function"&&!Wp(Jn))return fetch(Jn,{credentials:"same-origin"}).then(function(ae){if(!ae.ok)throw"failed to load wasm binary file at '"+Jn+"'";return ae.arrayBuffer()}).catch(function(){return Vp(Jn)});if(b)return new Promise(function(ae,ie){b(Jn,function(ke){ae(new Uint8Array(ke))},ie)})}return Promise.resolve().then(function(){return Vp(Jn)})}function E0(){var ae={a:Qn};function ie(Xe,Ke){var Sn=Xe.exports;a.asm=Sn,T=a.asm.h,rt(T.buffer),ft=a.asm.Sa,_c("wasm-instantiate")}fa("wasm-instantiate");function ke(Xe){ie(Xe.instance)}function at(Xe){return N0().then(function(Ke){return WebAssembly.instantiate(Ke,ae)}).then(Xe,function(Ke){F("failed to asynchronously prepare wasm: "+Ke),sr(Ke)})}function zt(){return!_&&typeof WebAssembly.instantiateStreaming=="function"&&!Lp(Jn)&&!Wp(Jn)&&typeof fetch=="function"?fetch(Jn,{credentials:"same-origin"}).then(function(Xe){var Ke=WebAssembly.instantiateStreaming(Xe,ae);return Ke.then(ke,function(Sn){return F("wasm streaming compile failed: "+Sn),F("falling back to ArrayBuffer instantiation"),at(ke)})}):at(ke)}if(a.instantiateWasm)try{var Tt=a.instantiateWasm(ae,ie);return Tt}catch(Xe){return F("Module.instantiateWasm callback failed with error: "+Xe),!1}return zt().catch(i),{}}function Bs(ae){for(;ae.length>0;){var ie=ae.shift();if(typeof ie=="function"){ie(a);continue}var ke=ie.func;typeof ke=="number"?ie.arg===void 0?ft.get(ke)():ft.get(ke)(ie.arg):ke(ie.arg===void 0?null:ie.arg)}}function Dc(){sr()}function si(ae,ie,ke){Ce.copyWithin(ae,ie,ie+ke)}function R0(){return Ce.length}function $0(ae){try{return T.grow(ae-be.byteLength+65535>>>16),rt(T.buffer),1}catch(ie){}}function Wr(ae){var ie=R0(),ke=2147483648;if(ae>ke)return!1;for(var at=1;at<=4;at*=2){var zt=ie*(1+.2/at);zt=Math.min(zt,ae+100663296);var Tt=Math.min(ke,fe(Math.max(ae,zt),65536)),Xe=$0(Tt);if(Xe)return!0}return!1}var ri={mappings:{},buffers:[null,[],[]],printChar:function(ae,ie){var ke=ri.buffers[ae];ie===0||ie===10?((ae===1?R:F)(ee(ke,0)),ke.length=0):ke.push(ie)},varargs:void 0,get:function(){ri.varargs+=4;var ae=He[ri.varargs-4>>2];return ae},getStr:function(ae){var ie=ne(ae);return ie},get64:function(ae,ie){return ae}};function _0(ae){return 0}function Up(ae,ie,ke,at,zt){}function D0(ae,ie,ke,at){for(var zt=0,Tt=0;Tt<ke;Tt++){for(var Xe=He[ie+Tt*8>>2],Ke=He[ie+(Tt*8+4)>>2],Sn=0;Sn<Ke;Sn++)ri.printChar(ae,Ce[Xe+Sn]);zt+=Ke}return He[at>>2]=zt,0}function Gp(){return 28}var Qn={a:Dc,d:si,e:Wr,f:_0,c:Up,b:D0,g:Gp},P0=E0(),Hp=a.___wasm_call_ctors=function(){return(Hp=a.___wasm_call_ctors=a.asm.i).apply(null,arguments)},F0=a._init=function(){return(F0=a._init=a.asm.j).apply(null,arguments)},jp=a._init_with_threads_count=function(){return(jp=a._init_with_threads_count=a.asm.k).apply(null,arguments)},O0=a._get_threads_count=function(){return(O0=a._get_threads_count=a.asm.l).apply(null,arguments)},Vl=a._register_tensor=function(){return(Vl=a._register_tensor=a.asm.m).apply(null,arguments)},ma=a._dispose_data=function(){return(ma=a._dispose_data=a.asm.n).apply(null,arguments)},Pc=a._dispose=function(){return(Pc=a._dispose=a.asm.o).apply(null,arguments)},M0=a._Abs=function(){return(M0=a._Abs=a.asm.p).apply(null,arguments)},z0=a._Add=function(){return(z0=a._Add=a.asm.q).apply(null,arguments)},qp=a._AddN=function(){return(qp=a._AddN=a.asm.r).apply(null,arguments)},Ee=a._All=function(){return(Ee=a._All=a.asm.s).apply(null,arguments)},L0=a._Any=function(){return(L0=a._Any=a.asm.t).apply(null,arguments)},B0=a._ArgMax=function(){return(B0=a._ArgMax=a.asm.u).apply(null,arguments)},W0=a._AvgPool=function(){return(W0=a._AvgPool=a.asm.v).apply(null,arguments)},V0=a._BatchMatMul=function(){return(V0=a._BatchMatMul=a.asm.w).apply(null,arguments)},U0=a._Ceil=function(){return(U0=a._Ceil=a.asm.x).apply(null,arguments)},ai=a._ClipByValue=function(){return(ai=a._ClipByValue=a.asm.y).apply(null,arguments)},G0=a._Conv2D=function(){return(G0=a._Conv2D=a.asm.z).apply(null,arguments)},H0=a._Conv2DBackpropInput=function(){return(H0=a._Conv2DBackpropInput=a.asm.A).apply(null,arguments)},j0=a._Cos=function(){return(j0=a._Cos=a.asm.B).apply(null,arguments)},q0=a._Cosh=function(){return(q0=a._Cosh=a.asm.C).apply(null,arguments)},X0=a._CropAndResize=function(){return(X0=a._CropAndResize=a.asm.D).apply(null,arguments)},K0=a._Cumsum=function(){return(K0=a._Cumsum=a.asm.E).apply(null,arguments)},Xp=a._DepthToSpace=function(){return(Xp=a._DepthToSpace=a.asm.F).apply(null,arguments)},Z0=a._DepthwiseConv2dNative=function(){return(Z0=a._DepthwiseConv2dNative=a.asm.G).apply(null,arguments)},Y0=a._Elu=function(){return(Y0=a._Elu=a.asm.H).apply(null,arguments)},ga=a._Equal=function(){return(ga=a._Equal=a.asm.I).apply(null,arguments)},Fc=a._Exp=function(){return(Fc=a._Exp=a.asm.J).apply(null,arguments)},Oc=a._FlipLeftRight=function(){return(Oc=a._FlipLeftRight=a.asm.K).apply(null,arguments)},J0=a._Floor=function(){return(J0=a._Floor=a.asm.L).apply(null,arguments)},Q0=a._FloorDiv=function(){return(Q0=a._FloorDiv=a.asm.M).apply(null,arguments)},eg=a._FusedBatchNorm=function(){return(eg=a._FusedBatchNorm=a.asm.N).apply(null,arguments)},tg=a._FusedConv2D=function(){return(tg=a._FusedConv2D=a.asm.O).apply(null,arguments)},ng=a._FusedDepthwiseConv2D=function(){return(ng=a._FusedDepthwiseConv2D=a.asm.P).apply(null,arguments)},Ve=a._Gather=function(){return(Ve=a._Gather=a.asm.Q).apply(null,arguments)},sg=a._GatherNd=function(){return(sg=a._GatherNd=a.asm.R).apply(null,arguments)},rg=a._Greater=function(){return(rg=a._Greater=a.asm.S).apply(null,arguments)},ag=a._GreaterEqual=function(){return(ag=a._GreaterEqual=a.asm.T).apply(null,arguments)},og=a._LeakyRelu=function(){return(og=a._LeakyRelu=a.asm.U).apply(null,arguments)},ig=a._Less=function(){return(ig=a._Less=a.asm.V).apply(null,arguments)},lg=a._LessEqual=function(){return(lg=a._LessEqual=a.asm.W).apply(null,arguments)},Mc=a._Log=function(){return(Mc=a._Log=a.asm.X).apply(null,arguments)},Kp=a._LogicalAnd=function(){return(Kp=a._LogicalAnd=a.asm.Y).apply(null,arguments)},Zp=a._Max=function(){return(Zp=a._Max=a.asm.Z).apply(null,arguments)},ug=a._MaxPool=function(){return(ug=a._MaxPool=a.asm._).apply(null,arguments)},cg=a._Maximum=function(){return(cg=a._Maximum=a.asm.$).apply(null,arguments)},dg=a._Mean=function(){return(dg=a._Mean=a.asm.aa).apply(null,arguments)},pg=a._Min=function(){return(pg=a._Min=a.asm.ba).apply(null,arguments)},hg=a._Minimum=function(){return(hg=a._Minimum=a.asm.ca).apply(null,arguments)},fg=a._MirrorPad=function(){return(fg=a._MirrorPad=a.asm.da).apply(null,arguments)},mg=a._Multiply=function(){return(mg=a._Multiply=a.asm.ea).apply(null,arguments)},lt=a._Neg=function(){return(lt=a._Neg=a.asm.fa).apply(null,arguments)},gg=a._NonMaxSuppressionV3=function(){return(gg=a._NonMaxSuppressionV3=a.asm.ga).apply(null,arguments)},Ag=a._NonMaxSuppressionV4=function(){return(Ag=a._NonMaxSuppressionV4=a.asm.ha).apply(null,arguments)},yg=a._NonMaxSuppressionV5=function(){return(yg=a._NonMaxSuppressionV5=a.asm.ia).apply(null,arguments)},Ul=a._NotEqual=function(){return(Ul=a._NotEqual=a.asm.ja).apply(null,arguments)},Yp=a._OneHot=function(){return(Yp=a._OneHot=a.asm.ka).apply(null,arguments)},Jp=a._PadV2=function(){return(Jp=a._PadV2=a.asm.la).apply(null,arguments)},Qp=a._Pow=function(){return(Qp=a._Pow=a.asm.ma).apply(null,arguments)},xg=a._Prelu=function(){return(xg=a._Prelu=a.asm.na).apply(null,arguments)},bg=a._Prod=function(){return(bg=a._Prod=a.asm.oa).apply(null,arguments)},eh=a._RealDiv=function(){return(eh=a._RealDiv=a.asm.pa).apply(null,arguments)},vg=a._Relu=function(){return(vg=a._Relu=a.asm.qa).apply(null,arguments)},wg=a._Relu6=function(){return(wg=a._Relu6=a.asm.ra).apply(null,arguments)},kg=a._ResizeBilinear=function(){return(kg=a._ResizeBilinear=a.asm.sa).apply(null,arguments)},Sg=a._Reverse=function(){return(Sg=a._Reverse=a.asm.ta).apply(null,arguments)},Ig=a._RotateWithOffset=function(){return(Ig=a._RotateWithOffset=a.asm.ua).apply(null,arguments)},th=a._Round=function(){return(th=a._Round=a.asm.va).apply(null,arguments)},Aa=a._Rsqrt=function(){return(Aa=a._Rsqrt=a.asm.wa).apply(null,arguments)},Cg=a._ScatterNd=function(){return(Cg=a._ScatterNd=a.asm.xa).apply(null,arguments)},Tg=a._SelectV2=function(){return(Tg=a._SelectV2=a.asm.ya).apply(null,arguments)},n5=a._Sigmoid=function(){return(n5=a._Sigmoid=a.asm.za).apply(null,arguments)},nh=a._Sin=function(){return(nh=a._Sin=a.asm.Aa).apply(null,arguments)},Ng=a._Softmax=function(){return(Ng=a._Softmax=a.asm.Ba).apply(null,arguments)},Eg=a._Sqrt=function(){return(Eg=a._Sqrt=a.asm.Ca).apply(null,arguments)},Rg=a._Square=function(){return(Rg=a._Square=a.asm.Da).apply(null,arguments)},$g=a._SquaredDifference=function(){return($g=a._SquaredDifference=a.asm.Ea).apply(null,arguments)},_g=a._Step=function(){return(_g=a._Step=a.asm.Fa).apply(null,arguments)},Dg=a._StridedSlice=function(){return(Dg=a._StridedSlice=a.asm.Ga).apply(null,arguments)},Pg=a._Sub=function(){return(Pg=a._Sub=a.asm.Ha).apply(null,arguments)},Fg=a._Sum=function(){return(Fg=a._Sum=a.asm.Ia).apply(null,arguments)},Og=a._Tan=function(){return(Og=a._Tan=a.asm.Ja).apply(null,arguments)},Mg=a._Tanh=function(){return(Mg=a._Tanh=a.asm.Ka).apply(null,arguments)},zg=a._Tile=function(){return(zg=a._Tile=a.asm.La).apply(null,arguments)},Lg=a._TopK=function(){return(Lg=a._TopK=a.asm.Ma).apply(null,arguments)},Bg=a._Transform=function(){return(Bg=a._Transform=a.asm.Na).apply(null,arguments)},Wg=a._Transpose=function(){return(Wg=a._Transpose=a.asm.Oa).apply(null,arguments)},Vg=a.__FusedMatMul=function(){return(Vg=a.__FusedMatMul=a.asm.Pa).apply(null,arguments)},Ug=a._malloc=function(){return(Ug=a._malloc=a.asm.Qa).apply(null,arguments)},Gg=a._free=function(){return(Gg=a._free=a.asm.Ra).apply(null,arguments)},sh=a.stackSave=function(){return(sh=a.stackSave=a.asm.Ta).apply(null,arguments)},rh=a.stackRestore=function(){return(rh=a.stackRestore=a.asm.Ua).apply(null,arguments)},zc=a.stackAlloc=function(){return(zc=a.stackAlloc=a.asm.Va).apply(null,arguments)};a.cwrap=Z;var Gl;function Hg(ae){this.name="ExitStatus",this.message="Program terminated with exit("+ae+")",this.status=ae}kr=function ae(){Gl||Lc(),Gl||(kr=ae)};function Lc(ae){if(ae=ae||u,En>0||(Pn(),En>0))return;function ie(){Gl||(Gl=!0,a.calledRun=!0,!M&&(ls(),Ls(),o(a),a.onRuntimeInitialized&&a.onRuntimeInitialized(),ks()))}a.setStatus?(a.setStatus("Running..."),setTimeout(function(){setTimeout(function(){a.setStatus("")},1),ie()},1)):ie()}if(a.run=Lc,a.preInit)for(typeof a.preInit=="function"&&(a.preInit=[a.preInit]);a.preInit.length>0;)a.preInit.pop()();return Lc(),r.ready}}();typeof e=="object"&&typeof t=="object"?t.exports=n:typeof define=="function"&&define.amd?define([],function(){return n}):typeof e=="object"&&(e.WasmBackendModule=n)}}),oE=1e-7,iE=1e-4,Xc=class{constructor(e,t){this.backend=e,this.dataMover=t,this.data=new WeakMap,this.dataIdsCount=0}get(e){return this.data.has(e)||this.dataMover.moveData(this.backend,e),this.data.get(e)}set(e,t){this.dataIdsCount++,this.data.set(e,t)}has(e){return this.data.has(e)}delete(e){return this.dataIdsCount--,this.data.delete(e)}numDataIds(){return this.dataIdsCount}},Yl=class{refCount(e){return Vs("refCount")}incRef(e){return Vs("incRef")}timerAvailable(){return!0}time(e){return Vs("time")}read(e){return Vs("read")}readSync(e){return Vs("readSync")}numDataIds(){return Vs("numDataIds")}disposeData(e,t){return Vs("disposeData")}write(e,t,n){return Vs("write")}move(e,t,n,s,r){return Vs("move")}memory(){return Vs("memory")}floatPrecision(){return Vs("floatPrecision")}epsilon(){return this.floatPrecision()===32?oE:iE}dispose(){return Vs("dispose")}};function Vs(e){throw new Error(`'${e}' not yet implemented or not found in the registry. This kernel may not be supported by the tfjs backend you have chosen`)}function y5(e){let t=e.length,n=0;for(;t>0;)n=Math.random()*t|0,t--,dh(e,t,n)}function lE(e,t){if(e.length!==t.length)throw new Error(`Array sizes must match to be shuffled together First array length was ${e.length}Second array length was ${t.length}`);let n=e.length,s=0;for(;n>0;)s=Math.random()*n|0,n--,dh(e,n,s),dh(t,n,s)}function Kc(e,t,n){return Math.max(e,Math.min(t,n))}function uE(e){return e%2==0?e:e+1}function dh(e,t,n){let s=e[t];e[t]=e[n],e[n]=s}function cE(e){let t=0;for(let n=0;n<e.length;n++)t+=e[n];return t}function dE(e,t){let n=Math.random();return t*n+(1-n)*e}function pE(e,t){let n=0;for(let s=0;s<e.length;s++){let r=Number(e[s])-Number(t[s]);n+=r*r}return n}function O(e,t){if(!e)throw new Error(typeof t=="string"?t:t())}function On(e,t,n=""){O(ka(e,t),()=>n+` Shapes ${e} and ${t} must match`)}function li(e){O(e!=null,()=>"The input to the tensor constructor must be a non-null value.")}function ui(e,t=[],n=!1){if(t==null&&(t=[]),Array.isArray(e)||$n(e)&&!n)for(let s=0;s<e.length;++s)ui(e[s],t,n);else t.push(e);return t}function Vt(e){if(e.length===0)return 1;let t=e[0];for(let n=1;n<e.length;n++)t*=e[n];return t}function hE(e){return e.length===0}function ka(e,t){if(e===t)return!0;if(e==null||t==null||e.length!==t.length)return!1;for(let n=0;n<e.length;n++)if(e[n]!==t[n])return!1;return!0}function pn(e){return e%1==0}function fE(e){if(Math.tanh!=null)return Math.tanh(e);if(e===1/0)return 1;if(e===-1/0)return-1;{let t=Math.exp(2*e);return(t-1)/(t+1)}}function mE(e){let t=Math.ceil(Math.sqrt(e));return[t,Math.ceil(e/t)]}function gE(e){let t=new Uint32Array(e);for(let n=0;n<e;++n)t[n]=n;return y5(t),t}function Zc(e,t){return t<=e.length?e:e+" ".repeat(t-e.length)}function AE(e,t=s=>0,n){return new Promise((s,r)=>{let a=0,o=()=>{if(e()){s();return}a++;let i=t(a);if(n!=null&&a>=n){r();return}setTimeout(o,i)};o()})}function yE(e,t){let n=1,s=-1;for(let a=0;a<e.length;++a)if(e[a]>=0)n*=e[a];else if(e[a]===-1){if(s!==-1)throw Error(`Shapes can only have 1 implicit size. Found -1 at dim ${s} and dim ${a}`);s=a}else if(e[a]<0)throw Error(`Shapes can not be < 0. Found ${e[a]} at dim ${a}`);if(s===-1){if(t>0&&t!==n)throw Error(`Size(${t}) must match the product of shape ${e}`);return e}if(n===0)throw Error(`Cannot infer the missing size in [${e}] when there are 0 elements`);if(t%n!=0)throw Error(`The implicit shape can't be a fractional number. Got ${t} / ${n}`);let r=e.slice();return r[s]=t/n,r}function Us(e,t){let n=t.length;return e=e==null?t.map((s,r)=>r):[].concat(e),O(e.every(s=>s>=-n&&s<n),()=>`All values in axis param must be in range [-${n}, ${n}) but got axis ${e}`),O(e.every(s=>pn(s)),()=>`All values in axis param must be integers but got axis ${e}`),e.map(s=>s<0?n+s:s)}function x5(e,t){let n=[],s=[],r=t!=null&&Array.isArray(t)&&t.length===0,a=t==null||r?null:Us(t,e).sort(),o=0;for(let i=0;i<e.length;++i){if(a!=null){if(a[o]===i&&e[i]!==1)throw new Error(`Can't squeeze axis ${i} since its dim '${e[i]}' is not 1`);(a[o]==null||a[o]>i)&&e[i]===1&&(n.push(e[i]),s.push(i)),a[o]<=i&&o++}e[i]!==1&&(n.push(e[i]),s.push(i))}return{newShape:n,keptDims:s}}function b5(e,t){let n=null;if(e==null||e==="float32")n=new Float32Array(t);else if(e==="int32")n=new Int32Array(t);else if(e==="bool")n=new Uint8Array(t);else throw new Error(`Unknown data type ${e}`);return n}function v5(e,t){let n=null;if(e==null||e==="float32")n=new Float32Array(t);else if(e==="int32")n=new Int32Array(t);else if(e==="bool")n=new Uint8Array(t);else if(e==="string")n=new Array(t);else throw new Error(`Unknown data type ${e}`);return n}function w5(e,t){for(let n=0;n<e.length;n++){let s=e[n];if(isNaN(s)||!isFinite(s))throw Error(`A tensor of type ${t} being uploaded contains ${s}.`)}}function k5(e){return e==="bool"||e==="complex64"||e==="float32"||e==="int32"||e==="string"}function xE(e,t){return!(t==="complex64"||t==="float32"&&e!=="complex64"||t==="int32"&&e!=="float32"&&e!=="complex64"||t==="bool"&&e==="bool")}function $n(e){return e instanceof Float32Array||e instanceof Int32Array||e instanceof Uint8Array||e instanceof Uint8ClampedArray}function Yg(e){if(e==="float32"||e==="int32")return 4;if(e==="complex64")return 8;if(e==="bool")return 1;throw new Error(`Unknown dtype ${e}`)}function S5(e){if(e==null)return 0;let t=0;return e.forEach(n=>t+=n.length),t}function Sa(e){return typeof e=="string"||e instanceof String}function I5(e){return typeof e=="boolean"}function C5(e){return typeof e=="number"}function ph(e){return Array.isArray(e)?ph(e[0]):e instanceof Float32Array?"float32":e instanceof Int32Array||e instanceof Uint8Array||e instanceof Uint8ClampedArray?"int32":C5(e)?"float32":Sa(e)?"string":I5(e)?"bool":"float32"}function Ia(e){return!!(e&&e.constructor&&e.call&&e.apply)}function hh(e,t){for(let n=t;n<e;++n)if(e%n==0)return n;return e}function Jl(e){let t=e.length;if(t<2)return[];let n=new Array(t-1);n[t-2]=e[t-1];for(let s=t-3;s>=0;--s)n[s]=n[s+1]*e[s+1];return n}function T5(e,t,n,s=!1){let r=new Array;if(t.length===1){let a=t[0]*(s?2:1);for(let o=0;o<a;o++)r[o]=n[e+o]}else{let a=t[0],o=t.slice(1),i=o.reduce((l,c)=>l*c)*(s?2:1);for(let l=0;l<a;l++)r[l]=T5(e+l*i,o,n,s)}return r}function Ql(e,t,n=!1){if(e.length===0)return t[0];let s=e.reduce((r,a)=>r*a)*(n?2:1);if(s===0)return[];if(s!==t.length)throw new Error(`[${e}] does not match the input size ${t.length}${n?" for a complex tensor":""}.`);return T5(0,e,t,n)}function Jg(e,t){let n=fh(e,t);for(let s=0;s<n.length;s++)n[s]=1;return n}function fh(e,t){if(t==null||t==="float32"||t==="complex64")return new Float32Array(e);if(t==="int32")return new Int32Array(e);if(t==="bool")return new Uint8Array(e);throw new Error(`Unknown data type ${t}`)}function bE(e,t){let n=e.reduce((s,r)=>s*r,1);if(t==null||t==="float32")return Ql(e,new Float32Array(n));if(t==="int32")return Ql(e,new Int32Array(n));if(t==="bool")return Ql(e,new Uint8Array(n));throw new Error(`Unknown data type ${t}`)}function Qg(e){e.forEach(t=>{O(Number.isInteger(t)&&t>=0,()=>`Tensor must have a shape comprised of positive integers but got shape [${e}].`)})}function vE(e,t,n){if(t===0)return 0;if(t===1)return e[0];let s=e[e.length-1];for(let r=0;r<e.length-1;++r)s+=n[r]*e[r];return s}function wE(e,t,n){if(t===0)return[];if(t===1)return[e];let s=new Array(t);for(let r=0;r<s.length-1;++r)s[r]=Math.floor(e/n[r]),e-=s[r]*n[r];return s[s.length-1]=e,s}function e2(e){return e&&e.then&&typeof e.then=="function"}var N5="tfjsflags",E5=class{constructor(e){this.global=e,this.flags={},this.flagRegistry={},this.urlFlags={},this.getQueryParams=kE,this.populateURLFlags()}setPlatform(e,t){this.platform!=null&&(K().getBool("IS_TEST")||K().getBool("PROD")||console.warn(`Platform ${this.platformName} has already been set. Overwriting the platform with ${t}.`)),this.platformName=e,this.platform=t}registerFlag(e,t,n){if(this.flagRegistry[e]={evaluationFn:t,setHook:n},this.urlFlags[e]!=null){let s=this.urlFlags[e];K().getBool("IS_TEST")||K().getBool("PROD")||console.warn(`Setting feature override from URL ${e}: ${s}.`),this.set(e,s)}}async getAsync(e){return e in this.flags?this.flags[e]:(this.flags[e]=await this.evaluateFlag(e),this.flags[e])}get(e){if(e in this.flags)return this.flags[e];let t=this.evaluateFlag(e);if(e2(t))throw new Error(`Flag ${e} cannot be synchronously evaluated. Please use getAsync() instead.`);return this.flags[e]=t,this.flags[e]}getNumber(e){return this.get(e)}getBool(e){return this.get(e)}getFlags(){return this.flags}get features(){return this.flags}set(e,t){if(this.flagRegistry[e]==null)throw new Error(`Cannot set flag ${e} as it has not been registered.`);this.flags[e]=t,this.flagRegistry[e].setHook!=null&&this.flagRegistry[e].setHook(t)}evaluateFlag(e){if(this.flagRegistry[e]==null)throw new Error(`Cannot evaluate flag '${e}': no evaluation function found.`);return this.flagRegistry[e].evaluationFn()}setFlags(e){this.flags=Object.assign({},e)}reset(){this.flags={},this.urlFlags={},this.populateURLFlags()}populateURLFlags(){if(typeof this.global=="undefined"||typeof this.global.location=="undefined"||typeof this.global.location.search=="undefined")return;let e=this.getQueryParams(this.global.location.search);N5 in e&&e[N5].split(",").forEach(n=>{let[s,r]=n.split(":");this.urlFlags[s]=IE(s,r)})}};function kE(e){let t={};return e.replace(/[?&]([^=?&]+)(?:=([^&]*))?/g,(n,...s)=>(SE(t,s[0],s[1]),s.join("="))),t}function SE(e,t,n){e[decodeURIComponent(t)]=decodeURIComponent(n||"")}function IE(e,t){if(t=t.toLowerCase(),t==="true"||t==="false")return t==="true";if(`${+t}`===t)return+t;throw new Error(`Could not parse value flag value ${t} for flag ${e}.`)}function K(){return Sr}var Sr=null;function CE(e){Sr=e}var t2;function R5(){if(t2==null){let e;if(typeof window!="undefined")e=window;else if(typeof global!="undefined")e=global;else if(typeof process!="undefined")e=process;else if(typeof self!="undefined")e=self;else throw new Error("Could not find a global object");t2=e}return t2}function TE(){let e=R5();return e._tfGlobals==null&&(e._tfGlobals=new Map),e._tfGlobals}function n2(e,t){let n=TE();if(n.has(e))return n.get(e);{let s=t();return n.set(e,s),n.get(e)}}var ci="Abs",eu="Acos",tu="Acosh",Gr="Add",Ca="AddN",nu="All",su="Any",Ta="ArgMax",ru="ArgMin",au="Asin",ou="Asinh",iu="Atan",lu="Atanh",uu="Atan2",Na="AvgPool",mh="AvgPoolGrad",Yc="AvgPool3D",gh="AvgPool3DGrad",Ea="BatchMatMul",di="BatchToSpaceND",Ah="Bincount",$5="BroadcastTo",yh="BroadcastArgs",Ra="Cast",$a="Ceil",Hr="ClipByValue",Jc="Complex",Qc="ComplexAbs",pi="Concat",_a="Conv2D",xh="Conv2DBackpropFilter",Da="Conv2DBackpropInput",ed="Conv3D",bh="Conv3DBackpropFilterV2",vh="Conv3DBackpropInputV2",Pa="Cos",Fa="Cosh",hi="Cumsum",fi="CropAndResize",wh="DenseBincount",mi="DepthToSpace",Oa="DepthwiseConv2dNative",kh="DepthwiseConv2dNativeBackpropFilter",Sh="DepthwiseConv2dNativeBackpropInput",Ih="Diag",td="Dilation2D",Ch="Dilation2DBackpropInput",Th="Dilation2DBackpropFilter",Ma="RealDiv",nd="Einsum",za="Elu",Nh="EluGrad",cu="Erf",gi="Equal",La="Exp",Ai="ExpandDims",yi="Expm1",Eh="FFT",du="Fill",xi="FlipLeftRight",Ba="Floor",Wa="FloorDiv",Va="FusedBatchNorm",bi="GatherV2",vi="GatherNd",wi="Greater",Ua="GreaterEqual",Ga="Identity",Rh="IFFT",sd="Imag",pu="IsFinite",hu="IsInf",fu="IsNan",ki="LeakyRelu",Si="Less",Ii="LessEqual",$h="LinSpace",Ha="Log",mu="Log1p",Ci="LogicalAnd",gu="LogicalNot",rd="LogicalOr",_5="LogSoftmax",ad="LRN",_h="LRNGrad",ja="Max",qa="Maximum",Xa="MaxPool",Dh="MaxPoolGrad",od="MaxPool3D",Ph="MaxPool3DGrad",Fh="MaxPoolWithArgmax",Ka="Mean",Za="Min",Ya="Minimum",Ja="MirrorPad",Au="Mod",Oh="Multinomial",Qa="Multiply",Ti="Neg",Ni="NotEqual",Ei="NonMaxSuppressionV3",yu="NonMaxSuppressionV4",Ri="NonMaxSuppressionV5",$i="OnesLike",_i="OneHot",Di="Pack",eo="PadV2",NE="Pool",to="Pow",no="Prelu",Pi="Prod",xu="Range",id="Real",bu="Reciprocal",so="Relu",Fi="Reshape",vu="ResizeNearestNeighbor",Mh="ResizeNearestNeighborGrad",ro="ResizeBilinear",zh="ResizeBilinearGrad",ao="Relu6",Oi="Reverse",Mi="Round",oo="Rsqrt",zi="ScatterNd",Li="Select",wu="Selu",Bi="Slice",io="Sin",Wi="Sinh",ku="Sign",lo="Sigmoid",Su="Softplus",uo="Sqrt",co="Sum",Vi="SpaceToBatchND",Ui="SplitV",po="Softmax",Lh="SparseFillEmptyRows",Bh="SparseReshape",Wh="SparseSegmentMean",Vh="SparseSegmentSum",ld="SparseToDense",ho="SquaredDifference",Iu="Square",Gi="StridedSlice",ud="StringNGrams",Uh="StringSplit",Gh="StringToHashBucketFast",fo="Sub",Hi="Tan",mo="Tanh",jr="Tile",ji="TopK",qi="Transform",go="Transpose",Hh="Unique",Xi="Unpack",cd="UnsortedSegmentSum",Ki="ZerosLike",Ao="Step",dd="FromPixels",Zi="RotateWithOffset",yo="_FusedMatMul",xo="FusedConv2D",bo="FusedDepthwiseConv2D";function vo(...e){K().getBool("IS_TEST")||K().getBool("PROD")||console.warn(...e)}function EE(...e){K().getBool("IS_TEST")||K().getBool("PROD")||console.log(...e)}var Cu=n2("kernelRegistry",()=>new Map),pd=n2("gradRegistry",()=>new Map);function jh(e,t){let n=r2(e,t);return Cu.get(n)}function s2(e){return pd.get(e)}function qr(e){let t=Cu.entries(),n=[];for(;;){let{done:s,value:r}=t.next();if(s)break;let[a,o]=r,[i]=a.split("_");i===e&&n.push(o)}return n}function rr(e){let{kernelName:t,backendName:n}=e,s=r2(t,n);Cu.has(s)&&vo(`The kernel '${t}' for backend '${n}' is already registered`),Cu.set(s,e)}function D5(e){let{kernelName:t}=e;pd.has(t)&&K().getBool("DEBUG")&&vo(`Overriding the gradient for '${t}'`),pd.set(t,e)}function RE(e,t){let n=r2(e,t);if(!Cu.has(n))throw new Error(`The kernel '${e}' for backend '${t}' is not registered`);Cu.delete(n)}function $E(e){if(!pd.has(e))throw new Error(`The gradient '${e}' for backend is not registered`);pd.delete(e)}function _E(e,t){qr(e).forEach(s=>{let r=Object.assign({},s,{backendName:t});rr(r)})}function r2(e,t){return`${t}_${e}`}var v={};Oe(v,{arraysEqual:()=>ka,assert:()=>O,assertNonNegativeIntegerDimensions:()=>Qg,assertNonNull:()=>li,assertShapesMatch:()=>On,bytesFromStringArray:()=>S5,bytesPerElement:()=>Yg,checkConversionForErrors:()=>w5,clamp:()=>Kc,computeStrides:()=>Jl,createScalarValue:()=>zE,createShuffledIndices:()=>gE,decodeString:()=>Kh,distSquared:()=>pE,encodeString:()=>md,fetch:()=>BE,fingerPrint64:()=>ME,flatten:()=>ui,getArrayFromDType:()=>v5,getTypedArrayFromDType:()=>b5,hasEncodingLoss:()=>xE,hexToLong:()=>hd,indexToLoc:()=>wE,inferDtype:()=>ph,inferFromImplicitShape:()=>yE,isBoolean:()=>I5,isFunction:()=>Ia,isInt:()=>pn,isNumber:()=>C5,isPromise:()=>e2,isScalarShape:()=>hE,isString:()=>Sa,isTypedArray:()=>$n,isValidDtype:()=>k5,locToIndex:()=>vE,makeOnesTypedArray:()=>Jg,makeZerosNestedTypedArray:()=>bE,makeZerosTypedArray:()=>fh,nearestDivisor:()=>hh,nearestLargerEven:()=>uE,now:()=>fd,parseAxisParam:()=>Us,randUniform:()=>dE,repeatedTry:()=>AE,rightPad:()=>Zc,shuffle:()=>y5,shuffleCombo:()=>lE,sizeFromShape:()=>Vt,sizeToSquarishShape:()=>mE,squeezeShape:()=>x5,sum:()=>cE,swap:()=>dh,tanh:()=>fE,toNestedArray:()=>Ql,toTypedArray:()=>Xh});var P5=ii(qN()),Yi=P5.default||P5;function hd(e){return Yi.fromString(e,!0,16)}var F5=hd("c3a5c85c97cb3127"),Ji=hd("b492b66fbe98f273"),Mn=hd("9ae16a3b2f90404f");function a2(e){return e.xor(e.shru(47))}function O5(e,t,n){let s=e.slice(t,t+n);return Yi.fromBytes(Array.from(s),!0,!0)}function bt(e,t){return O5(e,t,8)}function M5(e,t){return O5(e,t,4)}function hn(e,t){return t===0?e:e.shru(t).or(e.shl(64-t))}function wo(e,t,n=hd("9ddfea08eb382d69")){let s=e.xor(t).mul(n);s=s.xor(s.shru(47));let r=t.xor(s).mul(n);return r=r.xor(r.shru(47)),r=r.mul(n),r}function DE(e,t,n,s,r,a){r=r.add(e),a=hn(a.add(r).add(s),21);let o=r;return r=r.add(t),r=r.add(n),a=a.add(hn(r,44)),[r.add(s),a.add(o)]}function qh(e,t,n,s){return DE(bt(e,t),bt(e,t+8),bt(e,t+16),bt(e,t+24),n,s)}function PE(e,t=e.length){if(t>=8){let n=Mn.add(t*2),s=bt(e,0).add(Mn),r=bt(e,t-8),a=hn(r,37).mul(n).add(s),o=hn(s,25).add(r).mul(n);return wo(a,o,n)}if(t>=4){let n=Mn.add(t*2),s=M5(e,0);return wo(s.shl(3).add(t),M5(e,t-4),n)}if(t>0){let n=e[0],s=e[t>>1],r=e[t-1],a=n+(s<<8),o=t+(r<<2);return a2(Mn.mul(a).xor(F5.mul(o))).mul(Mn)}return Mn}function FE(e,t=e.length){let n=Mn.add(t*2),s=bt(e,0).mul(Ji),r=bt(e,8),a=bt(e,t-8).mul(n),o=bt(e,t-16).mul(Mn);return wo(hn(s.add(r),43).add(hn(a,30)).add(o),s.add(hn(r.add(Mn),18)).add(a),n)}function OE(e,t=e.length){let n=Mn.add(t*2),s=bt(e,0).mul(Mn),r=bt(e,8),a=bt(e,t-8).mul(n),o=bt(e,t-16).mul(Mn),i=hn(s.add(r),43).add(hn(a,30)).add(o),l=wo(i,s.add(hn(r.add(Mn),18)).add(a),n),c=bt(e,16).mul(n),u=bt(e,24),d=i.add(bt(e,t-32)).mul(n),p=l.add(bt(e,t-24)).mul(n);return wo(hn(c.add(u),43).add(hn(d,30)).add(p),c.add(hn(u.add(s),18)).add(d),n)}function ME(e,t=e.length){let n=Yi.fromNumber(81,!0);if(t<=32)return t<=16?PE(e,t):FE(e,t);if(t<=64)return OE(e,t);let s=n,r=n.mul(Ji).add(113),a=a2(r.mul(Mn).add(113)).mul(Mn),o=[Yi.UZERO,Yi.UZERO],i=[Yi.UZERO,Yi.UZERO];s=s.mul(Mn).add(bt(e,0));let l=0,c=(t-1>>6)*64,u=c+(t-1&63)-63;do s=hn(s.add(r).add(o[0]).add(bt(e,l+8)),37).mul(Ji),r=hn(r.add(o[1]).add(bt(e,l+48)),42).mul(Ji),s=s.xor(i[1]),r=r.add(o[0]).add(bt(e,l+40)),a=hn(a.add(i[0]),33).mul(Ji),o=qh(e,l,o[1].mul(Ji),s.add(i[0])),i=qh(e,l+32,a.add(i[1]),r.add(bt(e,l+16))),[a,s]=[s,a],l+=64;while(l!==c);let d=Ji.add(a.and(255).shl(1));return l=u,i[0]=i[0].add(t-1&63),o[0]=o[0].add(i[0]),i[0]=i[0].add(o[0]),s=hn(s.add(r).add(o[0]).add(bt(e,l+8)),37).mul(d),r=hn(r.add(o[1]).add(bt(e,l+48)),42).mul(d),s=s.xor(i[1].mul(9)),r=r.add(o[0].mul(9).add(bt(e,l+40))),a=hn(a.add(i[0]),33).mul(d),o=qh(e,l,o[1].mul(d),s.add(i[0])),i=qh(e,l+32,a.add(i[1]),r.add(bt(e,l+16))),[a,s]=[s,a],wo(wo(o[0],i[0],d).add(a2(r).mul(F5)).add(a),wo(o[1],i[1],d).add(s),d)}function zE(e,t){return t==="string"?md(e):Xh([e],t)}function LE(e,t){return e instanceof Float32Array&&t==="float32"||e instanceof Int32Array&&t==="int32"||e instanceof Uint8Array&&t==="bool"}function Xh(e,t){if(t==="string")throw new Error("Cannot convert a string[] to a TypedArray");if(Array.isArray(e)&&(e=ui(e)),K().getBool("DEBUG")&&w5(e,t),LE(e,t))return e;if(t==null||t==="float32"||t==="complex64")return new Float32Array(e);if(t==="int32")return new Int32Array(e);if(t==="bool"){let n=new Uint8Array(e.length);for(let s=0;s<n.length;++s)Math.round(e[s])!==0&&(n[s]=1);return n}else throw new Error(`Unknown data type ${t}`)}function fd(){return K().platform.now()}function BE(e,t){return K().platform.fetch(e,t)}function md(e,t="utf-8"){return t=t||"utf-8",K().platform.encode(e,t)}function Kh(e,t="utf-8"){return t=t||"utf-8",K().platform.decode(e,t)}var WE=class{constructor(e,t){this.backendTimer=e,this.logger=t,t==null&&(this.logger=new UE)}profileKernel(e,t,n){let s,r=()=>{s=n()},a,o=fd();if(this.backendTimer.timerAvailable())a=this.backendTimer.time(r);else{r();for(let l of s)l.dataSync();a=Promise.resolve({kernelMs:fd()-o})}if(K().getBool("CHECK_COMPUTATION_FOR_ERRORS"))for(let l=0;l<s.length;l++){let c=s[l];c.data().then(u=>{VE(u,c.dtype,e)})}return{kernelName:e,outputs:s,inputs:t,timeMs:a.then(l=>l.kernelMs),extraInfo:a.then(l=>l.getExtraProfileInfo!=null?l.getExtraProfileInfo():"")}}logKernelProfile(e){let{kernelName:t,outputs:n,timeMs:s,inputs:r,extraInfo:a}=e;n.forEach(o=>{Promise.all([o.data(),s,a]).then(i=>{this.logger.logKernelProfile(t,o,i[0],i[1],r,i[2])})})}};function VE(e,t,n){if(t!=="float32")return!1;for(let s=0;s<e.length;s++){let r=e[s];if(isNaN(r)||!isFinite(r))return console.warn(`Found ${r} in the result of '${n}'`),!0}return!1}var UE=class{logKernelProfile(e,t,n,s,r,a){let o=typeof s=="number"?Zc(`${s}ms`,9):s.error,i=Zc(e,25),l=t.rank,c=t.size,u=Zc(t.shape.toString(),14),d="";for(let p in r){let h=r[p];if(h!=null){let f=h.shape||t.shape,m=f.length;d+=`${p}: ${m}D ${m>0?f:""} `}}console.log(`%c${i} %c${o} %c${l}D ${u} %c${c} %c${d} %c${a}`,"font-weight:bold","color:red","color:blue","color: orange","color: green","color: steelblue")}};function GE(e,t,n){let s={},r={};for(let l=0;l<t.length;l++)s[t[l].id]=!0;for(let l=0;l<e.length;l++){let c=e[l],u=c.inputs;for(let d in u){let p=u[d],h=!1;for(let f=0;f<t.length;f++)if(s[p.id]){c.outputs.forEach(m=>s[m.id]=!0),h=!0,r[c.id]=!0;break}if(h)break}}let a={};a[n.id]=!0;let o={};for(let l=e.length-1;l>=0;l--){let c=e[l],u=c.inputs;for(let d=0;d<c.outputs.length;d++)if(a[c.outputs[d].id]){for(let p in u)a[u[p].id]=!0,o[c.id]=!0;break}}let i=[];for(let l=0;l<e.length;l++){let c=e[l];if(r[c.id]&&o[c.id]){let u={};for(let p in c.inputs){let h=c.inputs[p];s[h.id]&&(u[p]=h)}let d=Object.assign({},c);d.inputs=u,d.outputs=c.outputs,i.push(d)}}return i}function HE(e,t,n,s){for(let r=t.length-1;r>=0;r--){let a=t[r],o=[];if(a.outputs.forEach(l=>{let c=e[l.id];c!=null?o.push(c):o.push(null)}),a.gradient==null)throw new Error(`Cannot compute gradient: gradient function not found for ${a.kernelName}.`);let i=a.gradient(o);for(let l in a.inputs){if(!(l in i))throw new Error(`Cannot backprop through input ${l}. Available gradients found: ${Object.keys(i)}.`);let c=n(()=>i[l]());if(c.dtype!=="float32")throw new Error(`Error in gradient for op ${a.kernelName}. The gradient of input ${l} must have 'float32' dtype, but has '${c.dtype}'`);let u=a.inputs[l];if(!ka(c.shape,u.shape))throw new Error(`Error in gradient for op ${a.kernelName}. The gradient of input '${l}' has shape '${c.shape}', which does not match the shape of the input '${u.shape}'`);if(e[u.id]==null)e[u.id]=c;else{let d=e[u.id];e[u.id]=s(d,c),d.dispose()}}}}var z5=20,gd=3,o2=7;function jE(e,t,n,s){let r=Jl(t),a=qE(e,t,n,r),o=t.length,i=Zh(e,t,n,r,a),l=["Tensor"];return s&&(l.push(` dtype: ${n}`),l.push(` rank: ${o}`),l.push(` shape: [${t}]`),l.push(" values:")),l.push(i.map(c=>" "+c).join(`
|
|
`)),l.join(`
|
|
`)}function qE(e,t,n,s){let r=Vt(t),a=s[s.length-1],o=new Array(a).fill(0),i=t.length,l=n==="complex64"?yd(e):e;if(i>1)for(let c=0;c<r/a;c++){let u=c*a;for(let d=0;d<a;d++)o[d]=Math.max(o[d],Ad(l[u+d],0,n).length)}return o}function Ad(e,t,n){let s;return Array.isArray(e)?s=`${parseFloat(e[0].toFixed(o2))} + ${parseFloat(e[1].toFixed(o2))}j`:Sa(e)?s=`'${e}'`:n==="bool"?s=L5(e):s=parseFloat(e.toFixed(o2)).toString(),Zc(s,t)}function L5(e){return e===0?"false":"true"}function Zh(e,t,n,s,r,a=!0){let o=n==="complex64"?2:1,i=t[0],l=t.length;if(l===0){if(n==="complex64"){let m=yd(e);return[Ad(m[0],0,n)]}return n==="bool"?[L5(e[0])]:[e[0].toString()]}if(l===1){if(i>z5){let g=gd*o,A=Array.from(e.slice(0,g)),x=Array.from(e.slice((i-gd)*o,i*o));return n==="complex64"&&(A=yd(A),x=yd(x)),["["+A.map((y,b)=>Ad(y,r[b],n)).join(", ")+", ..., "+x.map((y,b)=>Ad(y,r[i-gd+b],n)).join(", ")+"]"]}let m=n==="complex64"?yd(e):Array.from(e);return["["+m.map((g,A)=>Ad(g,r[A],n)).join(", ")+"]"]}let c=t.slice(1),u=s.slice(1),d=s[0]*o,p=[];if(i>z5){for(let m=0;m<gd;m++){let g=m*d,A=g+d;p.push(...Zh(e.slice(g,A),c,n,u,r,!1))}p.push("...");for(let m=i-gd;m<i;m++){let g=m*d,A=g+d;p.push(...Zh(e.slice(g,A),c,n,u,r,m===i-1))}}else for(let m=0;m<i;m++){let g=m*d,A=g+d;p.push(...Zh(e.slice(g,A),c,n,u,r,m===i-1))}let h=l===2?",":"";p[0]="["+p[0]+h;for(let m=1;m<p.length-1;m++)p[m]=" "+p[m]+h;let f=`,
|
|
`;for(let m=2;m<l;m++)f+=`
|
|
`;return p[p.length-1]=" "+p[p.length-1]+"]"+(a?"":f),p}function yd(e){let t=[];for(let n=0;n<e.length;n+=2)t.push([e[n],e[n+1]]);return t}var en=class{constructor(e,t,n){if(this.dtype=t,this.shape=e.slice(),this.size=Vt(e),n!=null){let s=n.length;O(s===this.size,()=>`Length of values '${s}' does not match the size inferred by the shape '${this.size}'.`)}if(t==="complex64")throw new Error("complex64 dtype TensorBuffers are not supported. Please create a TensorBuffer for the real and imaginary parts separately and call tf.complex(real, imag).");this.values=n||v5(t,this.size),this.strides=Jl(e)}set(e,...t){t.length===0&&(t=[0]),O(t.length===this.rank,()=>`The number of provided coordinates (${t.length}) must match the rank (${this.rank})`);let n=this.locToIndex(t);this.values[n]=e}get(...e){e.length===0&&(e=[0]);let t=0;for(let s of e){if(s<0||s>=this.shape[t]){let r=`Requested out of range element at ${e}. Buffer shape=${this.shape}`;throw new Error(r)}t++}let n=e[e.length-1];for(let s=0;s<e.length-1;++s)n+=this.strides[s]*e[s];return this.values[n]}locToIndex(e){if(this.rank===0)return 0;if(this.rank===1)return e[0];let t=e[e.length-1];for(let n=0;n<e.length-1;++n)t+=this.strides[n]*e[n];return t}indexToLoc(e){if(this.rank===0)return[];if(this.rank===1)return[e];let t=new Array(this.shape.length);for(let n=0;n<t.length-1;++n)t[n]=Math.floor(e/this.strides[n]),e-=t[n]*this.strides[n];return t[t.length-1]=e,t}get rank(){return this.shape.length}toTensor(){return Ir().makeTensor(this.values,this.shape,this.dtype)}},Ir=null,Tu=null,XE=null;function KE(e){Ir=e}function ZE(e){Tu=e}function YE(e){XE=e}var Ze=class{constructor(e,t,n,s){this.kept=!1,this.isDisposedInternal=!1,this.shape=e.slice(),this.dtype=t||"float32",this.size=Vt(e),this.strides=Jl(e),this.dataId=n,this.id=s,this.rankType=this.rank<5?this.rank.toString():"higher"}get rank(){return this.shape.length}async buffer(){let e=await this.data();return Tu.buffer(this.shape,this.dtype,e)}bufferSync(){return Tu.buffer(this.shape,this.dtype,this.dataSync())}async array(){let e=await this.data();return Ql(this.shape,e,this.dtype==="complex64")}arraySync(){return Ql(this.shape,this.dataSync(),this.dtype==="complex64")}async data(){this.throwIfDisposed();let e=Ir().read(this.dataId);if(this.dtype==="string"){let t=await e;try{return t.map(n=>Kh(n))}catch(n){throw new Error("Failed to decode the string bytes into utf-8. To get the original bytes, call tensor.bytes().")}}return e}dataSync(){this.throwIfDisposed();let e=Ir().readSync(this.dataId);if(this.dtype==="string")try{return e.map(t=>Kh(t))}catch(t){throw new Error("Failed to decode the string bytes into utf-8. To get the original bytes, call tensor.bytes().")}return e}async bytes(){this.throwIfDisposed();let e=await Ir().read(this.dataId);return this.dtype==="string"?e:new Uint8Array(e.buffer)}dispose(){this.isDisposed||(Ir().disposeTensor(this),this.isDisposedInternal=!0)}get isDisposed(){return this.isDisposedInternal}throwIfDisposed(){if(this.isDisposed)throw new Error("Tensor is disposed.")}print(e=!1){return Tu.print(this,e)}clone(){return this.throwIfDisposed(),Tu.clone(this)}toString(e=!1){let t=this.dataSync();return jE(t,this.shape,this.dtype,e)}cast(e){return this.throwIfDisposed(),Tu.cast(this,e)}variable(e=!0,t,n){return this.throwIfDisposed(),Ir().makeVariable(this,e,t,n)}};Object.defineProperty(Ze,Symbol.hasInstance,{value:e=>!!e&&e.data!=null&&e.dataSync!=null&&e.throwIfDisposed!=null});function JE(){return n2("Tensor",()=>Ze)}JE();var xd=class extends Ze{constructor(e,t,n,s){super(e.shape,e.dtype,e.dataId,s);this.trainable=t,this.name=n}assign(e){if(e.dtype!==this.dtype)throw new Error(`dtype of the new value (${e.dtype}) and previous value (${this.dtype}) must match`);if(!ka(e.shape,this.shape))throw new Error(`shape of the new value (${e.shape}) and previous value (${this.shape}) must match`);Ir().disposeTensor(this),this.dataId=e.dataId,Ir().incRef(this,null)}dispose(){Ir().disposeVariable(this),this.isDisposedInternal=!0}};Object.defineProperty(xd,Symbol.hasInstance,{value:e=>e instanceof Ze&&e.assign!=null&&e.assign instanceof Function});var ar={};Oe(ar,{assertTypesMatch:()=>B5,getTensorsInContainer:()=>p2,isTensorInList:()=>e9,makeTypesMatch:()=>Dt});var i2;(function(e){e.R0="R0",e.R1="R1",e.R2="R2",e.R3="R3",e.R4="R4",e.R5="R5",e.R6="R6"})(i2||(i2={}));var l2;(function(e){e.float32="float32",e.int32="int32",e.bool="int32",e.complex64="complex64"})(l2||(l2={}));var u2;(function(e){e.float32="float32",e.int32="int32",e.bool="bool",e.complex64="complex64"})(u2||(u2={}));var c2;(function(e){e.float32="float32",e.int32="float32",e.bool="float32",e.complex64="complex64"})(c2||(c2={}));var d2;(function(e){e.float32="complex64",e.int32="complex64",e.bool="complex64",e.complex64="complex64"})(d2||(d2={}));var QE={float32:c2,int32:l2,bool:u2,complex64:d2};function zn(e,t){if(e==="string"||t==="string"){if(e==="string"&&t==="string")return"string";throw new Error(`Can not upcast ${e} with ${t}`)}return QE[e][t]}function bd(e){return zn(e,"int32")}function Dt(e,t){if(e.dtype===t.dtype)return[e,t];let n=zn(e.dtype,t.dtype);return[e.cast(n),t.cast(n)]}function B5(e,t){O(e.dtype===t.dtype,()=>`The dtypes of the first(${e.dtype}) and second(${t.dtype}) input must match`)}function e9(e,t){return t.some(n=>n.id===e.id)}function p2(e){let t=[],n=new Set;return W5(e,t,n),t}function W5(e,t,n){if(e==null)return;if(e instanceof Ze){t.push(e);return}if(!t9(e))return;let s=e;for(let r in s){let a=s[r];n.has(a)||(n.add(a),W5(a,t,n))}}function t9(e){return Array.isArray(e)||typeof e=="object"}function h2(e){return e.kernelName!=null}var V5=class{constructor(){this.registeredVariables={},this.nextTapeNodeId=0,this.numBytes=0,this.numTensors=0,this.numStringTensors=0,this.numDataBuffers=0,this.gradientDepth=0,this.kernelDepth=0,this.scopeStack=[],this.numDataMovesStack=[],this.nextScopeId=0,this.tensorInfo=new WeakMap,this.profiling=!1,this.activeProfile={newBytes:0,newTensors:0,peakBytes:0,kernels:[],result:null,get kernelNames(){return Array.from(new Set(this.kernels.map(e=>e.name)))}}}dispose(){for(let e in this.registeredVariables)this.registeredVariables[e].dispose()}},f2=class{constructor(e){this.ENV=e,this.registry={},this.registryFactory={},this.pendingBackendInitId=0,this.state=new V5}async ready(){if(this.pendingBackendInit!=null)return this.pendingBackendInit.then(()=>{});if(this.backendInstance!=null)return;let e=this.getSortedBackends();for(let t=0;t<e.length;t++){let n=e[t];if(await this.initializeBackend(n).success){await this.setBackend(n);return}}throw new Error("Could not initialize any backends, all backend initializations failed.")}get backend(){if(this.pendingBackendInit!=null)throw new Error(`Backend '${this.backendName}' has not yet been initialized. Make sure to await tf.ready() or await tf.setBackend() before calling other methods`);if(this.backendInstance==null){let{name:e,asyncInit:t}=this.initializeBackendsAndReturnBest();if(t)throw new Error(`The highest priority backend '${e}' has not yet been initialized. Make sure to await tf.ready() or await tf.setBackend() before calling other methods`);this.setBackend(e)}return this.backendInstance}backendNames(){return Object.keys(this.registryFactory)}findBackend(e){if(!(e in this.registry))if(e in this.registryFactory){let{asyncInit:t}=this.initializeBackend(e);if(t)return null}else return null;return this.registry[e]}findBackendFactory(e){return e in this.registryFactory?this.registryFactory[e].factory:null}registerBackend(e,t,n=1){return e in this.registryFactory?(vo(`${e} backend was already registered. Reusing existing backend factory.`),!1):(this.registryFactory[e]={factory:t,priority:n},!0)}async setBackend(e){if(this.registryFactory[e]==null)throw new Error(`Backend name '${e}' not found in registry`);if(this.backendName=e,this.registry[e]==null){this.backendInstance=null;let{success:t,asyncInit:n}=this.initializeBackend(e);if(!(n?await t:t))return!1}return this.backendInstance=this.registry[e],this.setupRegisteredKernels(),this.profiler=new WE(this.backendInstance),!0}setupRegisteredKernels(){qr(this.backendName).forEach(t=>{t.setupFunc!=null&&t.setupFunc(this.backendInstance)})}disposeRegisteredKernels(e){qr(e).forEach(n=>{n.disposeFunc!=null&&n.disposeFunc(this.registry[e])})}initializeBackend(e){let t=this.registryFactory[e];if(t==null)throw new Error(`Cannot initialize backend ${e}, no registration found.`);try{let n=t.factory();if(n&&!(n instanceof Yl)&&typeof n.then=="function"){let s=++this.pendingBackendInitId,r=n.then(a=>s<this.pendingBackendInitId?!1:(this.registry[e]=a,this.pendingBackendInit=null,!0)).catch(a=>(s<this.pendingBackendInitId||(this.pendingBackendInit=null,vo(`Initialization of backend ${e} failed`),vo(a.stack||a.message)),!1));return this.pendingBackendInit=r,{success:r,asyncInit:!0}}else return this.registry[e]=n,{success:!0,asyncInit:!1}}catch(n){return vo(`Initialization of backend ${e} failed`),vo(n.stack||n.message),{success:!1,asyncInit:!1}}}removeBackend(e){if(!(e in this.registryFactory))throw new Error(`${e} backend not found in registry`);this.backendName===e&&this.pendingBackendInit!=null&&this.pendingBackendInitId++,e in this.registry&&(this.disposeRegisteredKernels(e),this.registry[e].dispose(),delete this.registry[e]),delete this.registryFactory[e],this.backendName===e&&(this.pendingBackendInit=null,this.backendName=null,this.backendInstance=null)}getSortedBackends(){if(Object.keys(this.registryFactory).length===0)throw new Error("No backend found in registry.");return Object.keys(this.registryFactory).sort((e,t)=>this.registryFactory[t].priority-this.registryFactory[e].priority)}initializeBackendsAndReturnBest(){let e=this.getSortedBackends();for(let t=0;t<e.length;t++){let n=e[t],{success:s,asyncInit:r}=this.initializeBackend(n);if(r||s)return{name:n,asyncInit:r}}throw new Error("Could not initialize any backends, all backend initializations failed.")}moveData(e,t){let n=this.state.tensorInfo.get(t),s=n.backend,r=this.readSync(t),a=s.refCount(t);s.disposeData(t,!0),n.backend=e,e.move(t,r,n.shape,n.dtype,a),this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack[this.state.numDataMovesStack.length-1]++}tidy(e,t){let n=null;if(t==null){if(typeof e!="function")throw new Error("Please provide a function to tidy()");t=e}else{if(typeof e!="string"&&!(e instanceof String))throw new Error("When calling with two arguments, the first argument to tidy() must be a string");if(typeof t!="function")throw new Error("When calling with two arguments, the 2nd argument to tidy() must be a function");n=e}let s;return this.scopedRun(()=>this.startScope(n),()=>this.endScope(s),()=>(s=t(),s instanceof Promise&&console.error("Cannot return a Promise inside of tidy."),s))}scopedRun(e,t,n){e();try{let s=n();return t(),s}catch(s){throw t(),s}}nextTensorId(){return f2.nextTensorId++}nextVariableId(){return f2.nextVariableId++}clone(e){let t=W.runKernel(Ga,{x:e}),n={x:e},s=a=>({x:()=>{let o="float32",i={x:a},l={dtype:o};return W.runKernel(Ra,i,l)}}),r=[];return this.addTapeNode(this.state.activeScope.name,n,[t],s,r,{}),t}runKernel(e,t,n){if(this.backendName==null&&this.backend,!(jh(e,this.backendName)!=null))throw new Error(`Kernel '${e}' not registered for backend '${this.backendName}'`);return this.runKernelFunc({kernelName:e,inputs:t,attrs:n})}shouldCheckForMemLeaks(){return this.ENV.getBool("IS_TEST")}checkKernelForMemLeak(e,t,n){let s=this.backend.numDataIds(),r=0;n.forEach(i=>{r+=i.dtype==="complex64"?3:1});let a=this.state.numDataMovesStack[this.state.numDataMovesStack.length-1],o=s-t-r-a;if(o>0)throw new Error(`Backend '${this.backendName}' has an internal memory leak (${o} data ids) after running '${e}'`)}runKernelFunc(e){let t,n=[],s=this.isTapeOn(),r=this.state.numBytes,a=this.state.numTensors;this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack.push(0);let o;this.backendName==null&&this.backend;let i,l=h2(e)?e.kernelName:this.state.activeScope!=null?this.state.activeScope.name:"";if(h2(e)){let{kernelName:h,inputs:f,attrs:m}=e;this.backendName==null&&this.backend;let g=jh(h,this.backendName);O(g!=null,()=>`Cannot find registered kernel '${h}' for backend '${this.backendName}'`),o=()=>{let A=this.backend.numDataIds();i=g.kernelFunc({inputs:f,attrs:m,backend:this.backend});let x=Array.isArray(i)?i:[i];this.shouldCheckForMemLeaks()&&this.checkKernelForMemLeak(h,A,x);let y=x.map(b=>{if(b.rank!=null)return b;let{dataId:w,shape:k,dtype:C}=b;return this.makeTensorFromDataId(w,k,C)});if(s){let b=this.getTensorsForGradient(h,f,y);n=this.saveTensorsForBackwardMode(b)}return y}}else{let{forwardFunc:h}=e,f=m=>{!s||(n=m.map(g=>this.keep(this.clone(g))))};o=()=>{let m=this.backend.numDataIds();i=this.tidy(()=>h(this.backend,f));let g=Array.isArray(i)?i:[i];return this.shouldCheckForMemLeaks()&&this.checkKernelForMemLeak(l,m,g),g}}let{inputs:c,attrs:u}=e,d=h2(e)?null:e.backwardsFunc,p;return this.scopedRun(()=>this.state.kernelDepth++,()=>this.state.kernelDepth--,()=>{!this.ENV.getBool("DEBUG")&&!this.state.profiling?t=o():(p=this.profiler.profileKernel(l,c,()=>o()),this.ENV.getBool("DEBUG")&&this.profiler.logKernelProfile(p),t=p.outputs)}),s&&this.addTapeNode(l,c,t,d,n,u),this.state.profiling&&this.state.activeProfile.kernels.push({name:l,bytesAdded:this.state.numBytes-r,totalBytesSnapshot:this.state.numBytes,tensorsAdded:this.state.numTensors-a,totalTensorsSnapshot:this.state.numTensors,inputShapes:Object.keys(c).map(h=>c[h]!=null?c[h].shape:null),outputShapes:t.map(h=>h.shape),kernelTimeMs:p.timeMs,extraInfo:p.extraInfo}),Array.isArray(i)?t:t[0]}saveTensorsForBackwardMode(e){return e.map(n=>this.keep(this.clone(n)))}getTensorsForGradient(e,t,n){let s=s2(e);if(s!=null){let r=s.inputsToSave||[],a=s.outputsToSave||[],o;s.saveAllInputs?(O(Array.isArray(t),()=>"saveAllInputs is true, expected inputs to be an array."),o=Object.keys(t).map(l=>t[l])):o=r.map(l=>t[l]);let i=n.filter((l,c)=>a[c]);return o.concat(i)}return[]}makeTensor(e,t,n,s){if(e==null)throw new Error("Values passed to engine.makeTensor() are null");n=n||"float32",s=s||this.backend;let r=e;n==="string"&&Sa(e[0])&&(r=e.map(i=>md(i)));let a=s.write(r,t,n),o=new Ze(t,n,a,this.nextTensorId());if(this.trackTensor(o,s),n==="string"){let i=this.state.tensorInfo.get(a),l=S5(r);this.state.numBytes+=l-i.bytes,i.bytes=l}return o}makeTensorFromDataId(e,t,n,s){n=n||"float32";let r=new Ze(t,n,e,this.nextTensorId());return this.trackTensor(r,s),r}makeVariable(e,t=!0,n,s){n=n||this.nextVariableId().toString(),s!=null&&s!==e.dtype&&(e=e.cast(s));let r=new xd(e,t,n,this.nextTensorId());if(this.state.registeredVariables[r.name]!=null)throw new Error(`Variable with name ${r.name} was already registered`);return this.state.registeredVariables[r.name]=r,this.incRef(r,this.backend),r}trackTensor(e,t){this.state.numTensors++,e.dtype==="string"&&this.state.numStringTensors++;let n=0;e.dtype!=="complex64"&&e.dtype!=="string"&&(n=e.size*Yg(e.dtype)),this.state.numBytes+=n,this.state.tensorInfo.has(e.dataId)||(this.state.numDataBuffers++,this.state.tensorInfo.set(e.dataId,{backend:t||this.backend,dtype:e.dtype,shape:e.shape,bytes:n})),e instanceof xd||this.track(e)}incRef(e,t){this.trackTensor(e,t),this.backend.incRef(e.dataId)}removeDataId(e,t){this.state.tensorInfo.has(e)&&this.state.tensorInfo.get(e).backend===t&&(this.state.tensorInfo.delete(e),this.state.numDataBuffers--)}disposeTensor(e){if(!this.state.tensorInfo.has(e.dataId))return;let t=this.state.tensorInfo.get(e.dataId);if(this.state.numTensors--,e.dtype==="string"&&(this.state.numStringTensors--,this.state.numBytes-=t.bytes),e.dtype!=="complex64"&&e.dtype!=="string"){let n=e.size*Yg(e.dtype);this.state.numBytes-=n}t.backend.disposeData(e.dataId)&&this.removeDataId(e.dataId,t.backend)}disposeVariables(){for(let e in this.state.registeredVariables){let t=this.state.registeredVariables[e];this.disposeVariable(t)}}disposeVariable(e){this.disposeTensor(e),this.state.registeredVariables[e.name]!=null&&delete this.state.registeredVariables[e.name]}memory(){let e=this.backend.memory();return e.numTensors=this.state.numTensors,e.numDataBuffers=this.state.numDataBuffers,e.numBytes=this.state.numBytes,this.state.numStringTensors>0&&(e.unreliable=!0,e.reasons==null&&(e.reasons=[]),e.reasons.push("Memory usage by string tensors is approximate (2 bytes per character)")),e}async profile(e){this.state.profiling=!0;let t=this.state.numBytes,n=this.state.numTensors;this.state.activeProfile.kernels=[],this.state.activeProfile.result=await e(),this.state.profiling=!1,this.state.activeProfile.peakBytes=Math.max(...this.state.activeProfile.kernels.map(s=>s.totalBytesSnapshot)),this.state.activeProfile.newBytes=this.state.numBytes-t,this.state.activeProfile.newTensors=this.state.numTensors-n;for(let s of this.state.activeProfile.kernels)s.kernelTimeMs=await s.kernelTimeMs,s.extraInfo=await s.extraInfo;return this.state.activeProfile}isTapeOn(){return this.state.gradientDepth>0&&this.state.kernelDepth===0}addTapeNode(e,t,n,s,r,a){let o={id:this.state.nextTapeNodeId++,kernelName:e,inputs:t,outputs:n,saved:r},i=s2(e);i!=null&&(s=i.gradFunc),s!=null&&(o.gradient=l=>(l=l.map((c,u)=>{if(c==null){let d=n[u],p=fh(d.size,d.dtype);return this.makeTensor(p,d.shape,d.dtype)}return c}),s(l.length>1?l:l[0],r,a))),this.state.activeTape.push(o)}keep(e){return e.kept=!0,e}startTape(){this.state.gradientDepth===0&&(this.state.activeTape=[]),this.state.gradientDepth++}endTape(){this.state.gradientDepth--}startScope(e){let t={track:[],name:"unnamed scope",id:this.state.nextScopeId++};e&&(t.name=e),this.state.scopeStack.push(t),this.state.activeScope=t}endScope(e){let t=p2(e),n=new Set(t.map(r=>r.id));for(let r=0;r<this.state.activeScope.track.length;r++){let a=this.state.activeScope.track[r];!a.kept&&!n.has(a.id)&&a.dispose()}let s=this.state.scopeStack.pop();this.state.activeScope=this.state.scopeStack.length===0?null:this.state.scopeStack[this.state.scopeStack.length-1],t.forEach(r=>{!r.kept&&r.scopeId===s.id&&this.track(r)})}gradients(e,t,n,s=!1){if(O(t.length>0,()=>"gradients() received an empty list of xs."),n!=null&&n.dtype!=="float32")throw new Error(`dy must have 'float32' dtype, but has '${n.dtype}'`);let r=this.scopedRun(()=>this.startTape(),()=>this.endTape(),()=>this.tidy("forward",e));O(r instanceof Ze,()=>"The result y returned by f() must be a tensor.");let a=GE(this.state.activeTape,t,r);if(!s&&a.length===0&&t.length>0)throw new Error("Cannot compute gradient of y=f(x) with respect to x. Make sure that the f you passed encloses all operations that lead from x to y.");return this.tidy("backward",()=>{let o={};o[r.id]=n==null?n9(r.shape):n,HE(o,a,l=>this.tidy(l),s9);let i=t.map(l=>o[l.id]);return this.state.gradientDepth===0&&(this.state.activeTape.forEach(l=>{for(let c of l.saved)c.dispose()}),this.state.activeTape=null),{value:r,grads:i}})}customGrad(e){return O(Ia(e),()=>"The f passed in customGrad(f) must be a function."),(...t)=>{O(t.every(o=>o instanceof Ze),()=>"The args passed in customGrad(f)(x1, x2,...) must all be tensors");let n,s={};t.forEach((o,i)=>{s[i]=o});let r=(o,i)=>(n=e(...t,i),O(n.value instanceof Ze,()=>"The function f passed in customGrad(f) must return an object where `obj.value` is a tensor"),O(Ia(n.gradFunc),()=>"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function."),n.value),a=(o,i)=>{let l=n.gradFunc(o,i),c=Array.isArray(l)?l:[l];O(c.length===t.length,()=>"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function that returns the same number of tensors as inputs passed to f(...)."),O(c.every(d=>d instanceof Ze),()=>"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function that returns a list of only tensors.");let u={};return c.forEach((d,p)=>{u[p]=()=>d}),u};return this.runKernelFunc({forwardFunc:r,backwardsFunc:a,inputs:s})}}readSync(e){return this.state.tensorInfo.get(e).backend.readSync(e)}read(e){return this.state.tensorInfo.get(e).backend.read(e)}async time(e){let t=fd(),n=await this.backend.time(e);return n.wallMs=fd()-t,n}track(e){return this.state.activeScope!=null&&(e.scopeId=this.state.activeScope.id,this.state.activeScope.track.push(e)),e}get registeredVariables(){return this.state.registeredVariables}reset(){this.pendingBackendInitId++,this.state.dispose(),this.ENV.reset(),this.state=new V5;for(let e in this.registry)this.disposeRegisteredKernels(e),this.registry[e].dispose(),delete this.registry[e];this.backendName=null,this.backendInstance=null,this.pendingBackendInit=null}},m2=f2;m2.nextTensorId=0;m2.nextVariableId=0;function n9(e){let t=Jg(Vt(e),"float32");return W.makeTensor(t,e,"float32")}function U5(){let e=R5();if(e._tfengine==null){let t=new E5(e);e._tfengine=new m2(t)}return CE(e._tfengine.ENV),KE(()=>e._tfengine),e._tfengine}var W=U5();function s9(e,t){let n={a:e,b:t};return W.runKernel(Gr,n)}var Nu={};Oe(Nu,{isBrowser:()=>G5,isMobile:()=>o9,mockIsMobile:()=>a9});function r9(){return typeof navigator!="undefined"&&navigator!=null}var g2;function a9(e){g2=e}function o9(e){if(g2!==void 0)return g2;if(e||r9()){if(e||(e=navigator),e.product==="ReactNative")return!0;let t=e.userAgent||e.vendor||(typeof window!="undefined"?window.opera:"");if(!t){let n=e;return n.userAgentData&&n.userAgentData.mobile}return/(android|bb\d+|meego).+mobile|avantgo|bada\/|blackberry|blazer|compal|elaine|fennec|hiptop|iemobile|ip(hone|od)|iris|kindle|lge |maemo|midp|mmp|mobile.+firefox|netfront|opera m(ob|in)i|palm( os)?|phone|p(ixi|re)\/|plucker|pocket|psp|series(4|6)0|symbian|treo|up\.(browser|link)|vodafone|wap|windows ce|xda|xiino/i.test(t)||/1207|6310|6590|3gso|4thp|50[1-6]i|770s|802s|a wa|abac|ac(er|oo|s\-)|ai(ko|rn)|al(av|ca|co)|amoi|an(ex|ny|yw)|aptu|ar(ch|go)|as(te|us)|attw|au(di|\-m|r |s )|avan|be(ck|ll|nq)|bi(lb|rd)|bl(ac|az)|br(e|v)w|bumb|bw\-(n|u)|c55\/|capi|ccwa|cdm\-|cell|chtm|cldc|cmd\-|co(mp|nd)|craw|da(it|ll|ng)|dbte|dc\-s|devi|dica|dmob|do(c|p)o|ds(12|\-d)|el(49|ai)|em(l2|ul)|er(ic|k0)|esl8|ez([4-7]0|os|wa|ze)|fetc|fly(\-|_)|g1 u|g560|gene|gf\-5|g\-mo|go(\.w|od)|gr(ad|un)|haie|hcit|hd\-(m|p|t)|hei\-|hi(pt|ta)|hp( i|ip)|hs\-c|ht(c(\-| |_|a|g|p|s|t)|tp)|hu(aw|tc)|i\-(20|go|ma)|i230|iac( |\-|\/)|ibro|idea|ig01|ikom|im1k|inno|ipaq|iris|ja(t|v)a|jbro|jemu|jigs|kddi|keji|kgt( |\/)|klon|kpt |kwc\-|kyo(c|k)|le(no|xi)|lg( g|\/(k|l|u)|50|54|\-[a-w])|libw|lynx|m1\-w|m3ga|m50\/|ma(te|ui|xo)|mc(01|21|ca)|m\-cr|me(rc|ri)|mi(o8|oa|ts)|mmef|mo(01|02|bi|de|do|t(\-| |o|v)|zz)|mt(50|p1|v )|mwbp|mywa|n10[0-2]|n20[2-3]|n30(0|2)|n50(0|2|5)|n7(0(0|1)|10)|ne((c|m)\-|on|tf|wf|wg|wt)|nok(6|i)|nzph|o2im|op(ti|wv)|oran|owg1|p800|pan(a|d|t)|pdxg|pg(13|\-([1-8]|c))|phil|pire|pl(ay|uc)|pn\-2|po(ck|rt|se)|prox|psio|pt\-g|qa\-a|qc(07|12|21|32|60|\-[2-7]|i\-)|qtek|r380|r600|raks|rim9|ro(ve|zo)|s55\/|sa(ge|ma|mm|ms|ny|va)|sc(01|h\-|oo|p\-)|sdk\/|se(c(\-|0|1)|47|mc|nd|ri)|sgh\-|shar|sie(\-|m)|sk\-0|sl(45|id)|sm(al|ar|b3|it|t5)|so(ft|ny)|sp(01|h\-|v\-|v )|sy(01|mb)|t2(18|50)|t6(00|10|18)|ta(gt|lk)|tcl\-|tdg\-|tel(i|m)|tim\-|t\-mo|to(pl|sh)|ts(70|m\-|m3|m5)|tx\-9|up(\.b|g1|si)|utst|v400|v750|veri|vi(rg|te)|vk(40|5[0-3]|\-v)|vm40|voda|vulc|vx(52|53|60|61|70|80|81|83|85|98)|w3c(\-| )|webc|whit|wi(g |nc|nw)|wmlb|wonu|x700|yas\-|your|zeto|zte\-/i.test(t.substr(0,4))}return!1}function G5(){return typeof window!="undefined"&&window.document!=null||typeof WorkerGlobalScope!="undefined"}var or=K();or.registerFlag("DEBUG",()=>!1,e=>{e&&console.warn("Debugging mode is ON. The output of every math call will be downloaded to CPU and checked for NaNs. This significantly impacts performance.")});or.registerFlag("IS_BROWSER",()=>G5());or.registerFlag("IS_NODE",()=>typeof process!="undefined"&&typeof process.versions!="undefined"&&typeof process.versions.node!="undefined");or.registerFlag("IS_CHROME",()=>typeof navigator!="undefined"&&navigator!=null&&navigator.userAgent!=null&&/Chrome/.test(navigator.userAgent)&&/Google Inc/.test(navigator.vendor));or.registerFlag("PROD",()=>!1);or.registerFlag("TENSORLIKE_CHECK_SHAPE_CONSISTENCY",()=>or.getBool("DEBUG"));or.registerFlag("DEPRECATION_WARNINGS_ENABLED",()=>!0);or.registerFlag("IS_TEST",()=>!1);or.registerFlag("CHECK_COMPUTATION_FOR_ERRORS",()=>!0);or.registerFlag("WRAP_TO_IMAGEBITMAP",()=>!1);function Cr(e,t){let n=e;if($n(e))return t==="string"?[]:[e.length];if(!Array.isArray(e))return[];let s=[];for(;Array.isArray(n)||$n(n)&&t!=="string";)s.push(n.length),n=n[0];return Array.isArray(e)&&K().getBool("TENSORLIKE_CHECK_SHAPE_CONSISTENCY")&&H5(e,s,[]),s}function H5(e,t,n){if(n=n||[],!Array.isArray(e)&&!$n(e)){O(t.length===0,()=>`Element arr[${n.join("][")}] is a primitive, but should be an array/TypedArray of ${t[0]} elements`);return}O(t.length>0,()=>`Element arr[${n.join("][")}] should be a primitive, but is an array of ${e.length} elements`),O(e.length===t[0],()=>`Element arr[${n.join("][")}] should have ${t[0]} elements, but has ${e.length} elements`);let s=t.slice(1);for(let r=0;r<e.length;++r)H5(e[r],s,n.concat(r))}function j5(e,t,n,s){if(e!=="string_or_numeric"){if(e==null)throw new Error("Expected dtype cannot be null.");if(e!=="numeric"&&e!==t||e==="numeric"&&t==="string")throw new Error(`Argument '${n}' passed to '${s}' must be ${e} tensor, but got ${t} tensor`)}}function D(e,t,n,s="numeric"){if(e instanceof Ze)return j5(s,e.dtype,t,n),e;let r=ph(e);if(r!=="string"&&["bool","int32","float32"].indexOf(s)>=0&&(r=s),j5(s,r,t,n),e==null||!$n(e)&&!Array.isArray(e)&&typeof e!="number"&&typeof e!="boolean"&&typeof e!="string"){let l=e==null?"null":e.constructor.name;throw new Error(`Argument '${t}' passed to '${n}' must be a Tensor or TensorLike, but got '${l}'`)}let a=Cr(e,r);!$n(e)&&!Array.isArray(e)&&(e=[e]);let i=r!=="string"?Xh(e,r):ui(e,[],!0);return W.makeTensor(i,a,r)}function vd(e,t,n,s="numeric"){if(!Array.isArray(e))throw new Error(`Argument ${t} passed to ${n} must be a \`Tensor[]\` or \`TensorLike[]\``);return e.map((a,o)=>D(a,`${t}[${o}]`,n,s))}var q5="__op";function V(e){let t=Object.keys(e);if(t.length!==1)throw new Error(`Please provide an object with a single key (operation name) mapping to a function. Got an object with ${t.length} keys.`);let n=t[0],s=e[n];n.endsWith("_")&&(n=n.substring(0,n.length-1)),n=n+q5;let r=(...a)=>{W.startScope(n);try{let o=s(...a);return e2(o)&&console.error("Cannot return a Promise inside of tidy."),W.endScope(o),o}catch(o){throw W.endScope(null),o}};return Object.defineProperty(r,"name",{value:n,configurable:!0}),r}function i9(e,t){let n=D(e,"real","complex"),s=D(t,"imag","complex");On(n.shape,s.shape,`real and imag shapes, ${n.shape} and ${s.shape}, must match in call to tf.complex().`);let r={real:n,imag:s};return W.runKernel(Jc,r)}var ko=V({complex_:i9});function So(e,t,n,s){if(s==null&&(s=ph(e)),s==="complex64")throw new Error("Cannot construct a complex64 tensor directly. Please use tf.complex(real, imag).");if(!$n(e)&&!Array.isArray(e)&&typeof e!="number"&&typeof e!="boolean"&&typeof e!="string")throw new Error("values passed to tensor(values) must be a number/boolean/string or an array of numbers/booleans/strings, or a TypedArray");if(t!=null){Qg(t);let r=Vt(t),a=Vt(n);O(r===a,()=>`Based on the provided shape, [${t}], the tensor should have ${r} values but has ${a}`);for(let o=0;o<n.length;++o){let i=n[o],l=o===n.length-1?i!==Vt(t.slice(o)):!0;O(n[o]===t[o]||!l,()=>`Error creating a new Tensor. Inferred shape (${n}) does not match the provided shape (${t}). `)}}return!$n(e)&&!Array.isArray(e)&&(e=[e]),t=t||n,e=s!=="string"?Xh(e,s):ui(e,[],!0),W.makeTensor(e,t,s)}function Ut(e,t,n){let s=Cr(e,n);return So(e,t,s,n)}var A2={float32:4,float16:2,int32:4,uint16:2,uint8:1,bool:1,complex64:8},Yh=4;async function l9(e,t){let n=[],s=[],r=Array.isArray(e)?e.map(o=>o.name):Object.keys(e);for(let o=0;o<r.length;++o){let i=r[o],l=Array.isArray(e)?e[o].tensor:e[i];if(l.dtype!=="float32"&&l.dtype!=="int32"&&l.dtype!=="bool"&&l.dtype!=="string"&&l.dtype!=="complex64")throw new Error(`Unsupported dtype in weight '${i}': ${l.dtype}`);let c={name:i,shape:l.shape,dtype:l.dtype};if(l.dtype==="string"){let u=new Promise(async d=>{let p=await l.bytes(),h=p.reduce((g,A)=>g+A.length,0)+Yh*p.length,f=new Uint8Array(h),m=0;for(let g=0;g<p.length;g++){let A=p[g],x=new Uint8Array(new Uint32Array([A.length]).buffer);f.set(x,m),m+=Yh,f.set(A,m),m+=A.length}d(f)});s.push(u)}else s.push(l.data());t!=null&&(c.group=t),n.push(c)}let a=await Promise.all(s);return{data:u9(a),specs:n}}function X5(e,t){let n={},s,r=0;for(let a of t){let o=a.name,i=a.dtype,l=a.shape,c=Vt(l),u;if("quantization"in a){let d=a.quantization;if(d.dtype==="uint8"||d.dtype==="uint16"){if(!("min"in d&&"scale"in d))throw new Error(`Weight ${a.name} with quantization ${d.dtype} doesn't have corresponding metadata min and scale.`)}else if(d.dtype==="float16"){if(i!=="float32")throw new Error(`Weight ${a.name} is quantized with ${d.dtype} which only supports weights of type float32 not ${i}.`)}else throw new Error(`Weight ${a.name} has unknown quantization dtype ${d.dtype}. Supported quantization dtypes are: 'uint8', 'uint16', and 'float16'.`);let p=A2[d.dtype],h=e.slice(r,r+c*p),f=d.dtype==="uint8"?new Uint8Array(h):new Uint16Array(h);if(i==="float32")if(d.dtype==="uint8"||d.dtype==="uint16"){u=new Float32Array(f.length);for(let m=0;m<f.length;m++){let g=f[m];u[m]=g*d.scale+d.min}}else if(d.dtype==="float16")s===void 0&&(s=m9()),u=s(f);else throw new Error(`Unsupported quantization type ${d.dtype} for weight type float32.`);else if(i==="int32"){if(d.dtype!=="uint8"&&d.dtype!=="uint16")throw new Error(`Unsupported quantization type ${d.dtype} for weight type int32.`);u=new Int32Array(f.length);for(let m=0;m<f.length;m++){let g=f[m];u[m]=Math.round(g*d.scale+d.min)}}else throw new Error(`Unsupported dtype in weight '${o}': ${i}`);r+=c*p}else if(i==="string"){let d=Vt(a.shape);u=[];for(let p=0;p<d;p++){let h=new Uint32Array(e.slice(r,r+Yh))[0];r+=Yh;let f=new Uint8Array(e.slice(r,r+h));u.push(f),r+=h}}else{let d=A2[i],p=e.slice(r,r+c*d);if(i==="float32")u=new Float32Array(p);else if(i==="int32")u=new Int32Array(p);else if(i==="bool")u=new Uint8Array(p);else if(i==="complex64"){u=new Float32Array(p);let h=new Float32Array(u.length/2),f=new Float32Array(u.length/2);for(let A=0;A<h.length;A++)h[A]=u[A*2],f[A]=u[A*2+1];let m=Ut(h,l,"float32"),g=Ut(f,l,"float32");n[o]=ko(m,g),m.dispose(),g.dispose()}else throw new Error(`Unsupported dtype in weight '${o}': ${i}`);r+=c*d}i!=="complex64"&&(n[o]=Ut(u,l,i))}return n}function u9(e){if(e===null)throw new Error(`Invalid input value: ${JSON.stringify(e)}`);let t=0,n=[];e.forEach(a=>{if(t+=a.byteLength,n.push(a.byteLength===a.buffer.byteLength?a:new a.constructor(a)),!(a instanceof Float32Array||a instanceof Int32Array||a instanceof Uint8Array))throw new Error(`Unsupported TypedArray subtype: ${a.constructor.name}`)});let s=new Uint8Array(t),r=0;return n.forEach(a=>{s.set(new Uint8Array(a.buffer),r),r+=a.byteLength}),s.buffer}var y2=typeof Buffer!="undefined"&&(typeof Blob=="undefined"||typeof atob=="undefined"||typeof btoa=="undefined");function K5(e){return y2?Buffer.byteLength(e):new Blob([e]).size}function c9(e){if(y2)return Buffer.from(e).toString("base64");let t=new Uint8Array(e),n="";for(let s=0,r=t.length;s<r;s++)n+=String.fromCharCode(t[s]);return btoa(n)}function d9(e){if(y2){let s=Buffer.from(e,"base64");return s.buffer.slice(s.byteOffset,s.byteOffset+s.byteLength)}let t=atob(e),n=new Uint8Array(t.length);for(let s=0;s<t.length;++s)n.set([t.charCodeAt(s)],s);return n.buffer}function x2(e){if(e.length===1)return e[0];let t=0;e.forEach(r=>{t+=r.byteLength});let n=new Uint8Array(t),s=0;return e.forEach(r=>{n.set(new Uint8Array(r),s),s+=r.byteLength}),n.buffer}function Z5(e){let t="/";for(e=e.trim();e.endsWith(t);)e=e.slice(0,e.length-1);let n=e.split(t);return n[n.length-1]}function Y5(e,t){let n={modelTopology:e.modelTopology,format:e.format,generatedBy:e.generatedBy,convertedBy:e.convertedBy,weightsManifest:t};return e.signature!=null&&(n.signature=e.signature),e.userDefinedMetadata!=null&&(n.userDefinedMetadata=e.userDefinedMetadata),e.modelInitializer!=null&&(n.modelInitializer=e.modelInitializer),e.trainingConfig!=null&&(n.trainingConfig=e.trainingConfig),n}async function b2(e,t){let n={modelTopology:e.modelTopology,format:e.format,generatedBy:e.generatedBy,convertedBy:e.convertedBy};if(e.trainingConfig!=null&&(n.trainingConfig=e.trainingConfig),e.weightsManifest!=null){let[s,r]=await t(e.weightsManifest);n.weightSpecs=s,n.weightData=r}return e.signature!=null&&(n.signature=e.signature),e.userDefinedMetadata!=null&&(n.userDefinedMetadata=e.userDefinedMetadata),e.modelInitializer!=null&&(n.modelInitializer=e.modelInitializer),n}function wd(e){if(e.modelTopology instanceof ArrayBuffer)throw new Error("Expected JSON model topology, received ArrayBuffer.");return{dateSaved:new Date,modelTopologyType:"JSON",modelTopologyBytes:e.modelTopology==null?0:K5(JSON.stringify(e.modelTopology)),weightSpecsBytes:e.weightSpecs==null?0:K5(JSON.stringify(e.weightSpecs)),weightDataBytes:e.weightData==null?0:e.weightData.byteLength}}function p9(){let e=n=>{let s=n<<13,r=0;for(;(s&8388608)==0;)r-=8388608,s<<=1;return s&=~8388608,r+=947912704,s|r},t=new Uint32Array(2048);t[0]=0;for(let n=1;n<1024;n++)t[n]=e(n);for(let n=1024;n<2048;n++)t[n]=939524096+(n-1024<<13);return t}function h9(){let e=new Uint32Array(64);e[0]=0,e[31]=1199570944,e[32]=2147483648,e[63]=3347054592;for(let t=1;t<31;t++)e[t]=t<<23;for(let t=33;t<63;t++)e[t]=2147483648+(t-32<<23);return e}function f9(){let e=new Uint32Array(64);for(let t=0;t<64;t++)e[t]=1024;return e[0]=e[32]=0,e}function m9(){let e=p9(),t=h9(),n=f9();return s=>{let r=new ArrayBuffer(4*s.length),a=new Uint32Array(r);for(let o=0;o<s.length;o++){let i=s[o],l=e[n[i>>10]+(i&1023)]+t[i>>10];a[o]=l}return new Float32Array(r)}}var Lt=class{constructor(){this.saveRouters=[],this.loadRouters=[]}static getInstance(){return Lt.instance==null&&(Lt.instance=new Lt),Lt.instance}static registerSaveRouter(e){Lt.getInstance().saveRouters.push(e)}static registerLoadRouter(e){Lt.getInstance().loadRouters.push(e)}static getSaveHandlers(e){return Lt.getHandlers(e,"save")}static getLoadHandlers(e,t){return Lt.getHandlers(e,"load",t)}static getHandlers(e,t,n){let s=[];return(t==="load"?Lt.getInstance().loadRouters:Lt.getInstance().saveRouters).forEach(a=>{let o=a(e,n);o!==null&&s.push(o)}),s}},g9=e=>Lt.registerSaveRouter(e),A9=e=>Lt.registerLoadRouter(e),y9=e=>Lt.getSaveHandlers(e),x9=(e,t)=>Lt.getLoadHandlers(e,t),v2="tensorflowjs",w2=1,Qi="models_store",Io="model_info_store";function J5(){if(!K().getBool("IS_BROWSER"))throw new Error("Failed to obtain IndexedDB factory because the current environmentis not a web browser.");let e=typeof window=="undefined"?self:window,t=e.indexedDB||e.mozIndexedDB||e.webkitIndexedDB||e.msIndexedDB||e.shimIndexedDB;if(t==null)throw new Error("The current browser does not appear to support IndexedDB.");return t}function k2(e){let t=e.result;t.createObjectStore(Qi,{keyPath:"modelPath"}),t.createObjectStore(Io,{keyPath:"modelPath"})}var el=class{constructor(e){if(this.indexedDB=J5(),e==null||!e)throw new Error("For IndexedDB, modelPath must not be null, undefined or empty.");this.modelPath=e}async save(e){if(e.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");return this.databaseAction(this.modelPath,e)}async load(){return this.databaseAction(this.modelPath)}databaseAction(e,t){return new Promise((n,s)=>{let r=this.indexedDB.open(v2,w2);r.onupgradeneeded=()=>k2(r),r.onsuccess=()=>{let a=r.result;if(t==null){let o=a.transaction(Qi,"readonly"),l=o.objectStore(Qi).get(this.modelPath);l.onsuccess=()=>{if(l.result==null)return a.close(),s(new Error(`Cannot find model with path '${this.modelPath}' in IndexedDB.`));n(l.result.modelArtifacts)},l.onerror=c=>(a.close(),s(l.error)),o.oncomplete=()=>a.close()}else{let o=wd(t),i=a.transaction(Io,"readwrite"),l=i.objectStore(Io),c=l.put({modelPath:this.modelPath,modelArtifactsInfo:o}),u;c.onsuccess=()=>{u=a.transaction(Qi,"readwrite");let p=u.objectStore(Qi).put({modelPath:this.modelPath,modelArtifacts:t,modelArtifactsInfo:o});p.onsuccess=()=>n({modelArtifactsInfo:o}),p.onerror=h=>{l=i.objectStore(Io);let f=l.delete(this.modelPath);f.onsuccess=()=>(a.close(),s(p.error)),f.onerror=m=>(a.close(),s(p.error))}},c.onerror=d=>(a.close(),s(c.error)),i.oncomplete=()=>{u==null?a.close():u.oncomplete=()=>a.close()}}},r.onerror=a=>s(r.error)})}};el.URL_SCHEME="indexeddb://";var Q5=e=>K().getBool("IS_BROWSER")&&!Array.isArray(e)&&e.startsWith(el.URL_SCHEME)?b9(e.slice(el.URL_SCHEME.length)):null;Lt.registerSaveRouter(Q5);Lt.registerLoadRouter(Q5);function b9(e){return new el(e)}function v9(e){return e.startsWith(el.URL_SCHEME)?e.slice(el.URL_SCHEME.length):e}var w9=class{constructor(){this.indexedDB=J5()}async listModels(){return new Promise((e,t)=>{let n=this.indexedDB.open(v2,w2);n.onupgradeneeded=()=>k2(n),n.onsuccess=()=>{let s=n.result,r=s.transaction(Io,"readonly"),o=r.objectStore(Io).getAll();o.onsuccess=()=>{let i={};for(let l of o.result)i[l.modelPath]=l.modelArtifactsInfo;e(i)},o.onerror=i=>(s.close(),t(o.error)),r.oncomplete=()=>s.close()},n.onerror=s=>t(n.error)})}async removeModel(e){return e=v9(e),new Promise((t,n)=>{let s=this.indexedDB.open(v2,w2);s.onupgradeneeded=()=>k2(s),s.onsuccess=()=>{let r=s.result,a=r.transaction(Io,"readwrite"),o=a.objectStore(Io),i=o.get(e),l;i.onsuccess=()=>{if(i.result==null)return r.close(),n(new Error(`Cannot find model with path '${e}' in IndexedDB.`));{let c=o.delete(e),u=()=>{l=r.transaction(Qi,"readwrite");let p=l.objectStore(Qi).delete(e);p.onsuccess=()=>t(i.result.modelArtifactsInfo),p.onerror=h=>n(i.error)};c.onsuccess=u,c.onerror=d=>(u(),r.close(),n(i.error))}},i.onerror=c=>(r.close(),n(i.error)),a.oncomplete=()=>{l==null?r.close():l.oncomplete=()=>r.close()}},s.onerror=r=>n(s.error)})}},Xr="/",Eu="tensorflowjs_models",e3="info",k9="model_topology",S9="weight_specs",I9="weight_data",C9="model_metadata";function t3(e){return{info:[Eu,e,e3].join(Xr),topology:[Eu,e,k9].join(Xr),weightSpecs:[Eu,e,S9].join(Xr),weightData:[Eu,e,I9].join(Xr),modelMetadata:[Eu,e,C9].join(Xr)}}function n3(e){for(let t of Object.values(e))window.localStorage.removeItem(t)}function T9(e){let t=e.split(Xr);if(t.length<3)throw new Error(`Invalid key format: ${e}`);return t.slice(1,t.length-1).join(Xr)}function N9(e){return e.startsWith(tl.URL_SCHEME)?e.slice(tl.URL_SCHEME.length):e}var tl=class{constructor(e){if(!K().getBool("IS_BROWSER")||typeof window=="undefined"||typeof window.localStorage=="undefined")throw new Error("The current environment does not support local storage.");if(this.LS=window.localStorage,e==null||!e)throw new Error("For local storage, modelPath must not be null, undefined or empty.");this.modelPath=e,this.keys=t3(this.modelPath)}async save(e){if(e.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");{let t=JSON.stringify(e.modelTopology),n=JSON.stringify(e.weightSpecs),s=wd(e);try{this.LS.setItem(this.keys.info,JSON.stringify(s)),this.LS.setItem(this.keys.topology,t),this.LS.setItem(this.keys.weightSpecs,n),this.LS.setItem(this.keys.weightData,c9(e.weightData));let r={format:e.format,generatedBy:e.generatedBy,convertedBy:e.convertedBy,signature:e.signature!=null?e.signature:void 0,userDefinedMetadata:e.userDefinedMetadata!=null?e.userDefinedMetadata:void 0,modelInitializer:e.modelInitializer!=null?e.modelInitializer:void 0,trainingConfig:e.trainingConfig!=null?e.trainingConfig:void 0};return this.LS.setItem(this.keys.modelMetadata,JSON.stringify(r)),{modelArtifactsInfo:s}}catch(r){throw n3(this.keys),new Error(`Failed to save model '${this.modelPath}' to local storage: size quota being exceeded is a possible cause of this failure: modelTopologyBytes=${s.modelTopologyBytes}, weightSpecsBytes=${s.weightSpecsBytes}, weightDataBytes=${s.weightDataBytes}.`)}}}async load(){let e=JSON.parse(this.LS.getItem(this.keys.info));if(e==null)throw new Error(`In local storage, there is no model with name '${this.modelPath}'`);if(e.modelTopologyType!=="JSON")throw new Error("BrowserLocalStorage does not support loading non-JSON model topology yet.");let t={},n=JSON.parse(this.LS.getItem(this.keys.topology));if(n==null)throw new Error(`In local storage, the topology of model '${this.modelPath}' is missing.`);t.modelTopology=n;let s=JSON.parse(this.LS.getItem(this.keys.weightSpecs));if(s==null)throw new Error(`In local storage, the weight specs of model '${this.modelPath}' are missing.`);t.weightSpecs=s;let r=this.LS.getItem(this.keys.modelMetadata);if(r!=null){let o=JSON.parse(r);t.format=o.format,t.generatedBy=o.generatedBy,t.convertedBy=o.convertedBy,o.signature!=null&&(t.signature=o.signature),o.userDefinedMetadata!=null&&(t.userDefinedMetadata=o.userDefinedMetadata),o.modelInitializer!=null&&(t.modelInitializer=o.modelInitializer),o.trainingConfig!=null&&(t.trainingConfig=o.trainingConfig)}let a=this.LS.getItem(this.keys.weightData);if(a==null)throw new Error(`In local storage, the binary weight values of model '${this.modelPath}' are missing.`);return t.weightData=d9(a),t}};tl.URL_SCHEME="localstorage://";var s3=e=>K().getBool("IS_BROWSER")&&!Array.isArray(e)&&e.startsWith(tl.URL_SCHEME)?E9(e.slice(tl.URL_SCHEME.length)):null;Lt.registerSaveRouter(s3);Lt.registerLoadRouter(s3);function E9(e){return new tl(e)}var R9=class{constructor(){O(K().getBool("IS_BROWSER"),()=>"Current environment is not a web browser"),O(typeof window=="undefined"||typeof window.localStorage!="undefined",()=>"Current browser does not appear to support localStorage"),this.LS=window.localStorage}async listModels(){let e={},t=Eu+Xr,n=Xr+e3;for(let s=0;s<this.LS.length;++s){let r=this.LS.key(s);if(r.startsWith(t)&&r.endsWith(n)){let a=T9(r);e[a]=JSON.parse(this.LS.getItem(r))}}return e}async removeModel(e){e=N9(e);let t=t3(e);if(this.LS.getItem(t.info)==null)throw new Error(`Cannot find model at path '${e}'`);let n=JSON.parse(this.LS.getItem(t.info));return n3(t),n}},Ru="://",Ss=class{constructor(){this.managers={}}static getInstance(){return Ss.instance==null&&(Ss.instance=new Ss),Ss.instance}static registerManager(e,t){O(e!=null,()=>"scheme must not be undefined or null."),e.endsWith(Ru)&&(e=e.slice(0,e.indexOf(Ru))),O(e.length>0,()=>"scheme must not be an empty string.");let n=Ss.getInstance();O(n.managers[e]==null,()=>`A model store manager is already registered for scheme '${e}'.`),n.managers[e]=t}static getManager(e){let t=this.getInstance().managers[e];if(t==null)throw new Error(`Cannot find model manager for scheme '${e}'`);return t}static getSchemes(){return Object.keys(this.getInstance().managers)}};function Jh(e){if(e.indexOf(Ru)===-1)throw new Error(`The url string provided does not contain a scheme. Supported schemes are: ${Ss.getSchemes().join(",")}`);return{scheme:e.split(Ru)[0],path:e.split(Ru)[1]}}async function r3(e,t,n=!1){O(e!==t,()=>`Old path and new path are the same: '${e}'`);let s=Lt.getLoadHandlers(e);O(s.length>0,()=>`Copying failed because no load handler is found for source URL ${e}.`),O(s.length<2,()=>`Copying failed because more than one (${s.length}) load handlers for source URL ${e}.`);let r=s[0],a=Lt.getSaveHandlers(t);O(a.length>0,()=>`Copying failed because no save handler is found for destination URL ${t}.`),O(a.length<2,()=>`Copying failed because more than one (${s.length}) save handlers for destination URL ${t}.`);let o=a[0],i=Jh(e).scheme,l=Jh(e).path,c=i===Jh(e).scheme,u=await r.load();n&&c&&await Ss.getManager(i).removeModel(l);let d=await o.save(u);return n&&!c&&await Ss.getManager(i).removeModel(l),d.modelArtifactsInfo}async function $9(){let e=Ss.getSchemes(),t={};for(let n of e){let s=await Ss.getManager(n).listModels();for(let r in s){let a=n+Ru+r;t[a]=s[r]}}return t}async function _9(e){let t=Jh(e);return Ss.getManager(t.scheme).removeModel(t.path)}async function D9(e,t){return r3(e,t,!1)}async function P9(e,t){return r3(e,t,!0)}var F9=class{fetch(e,t){return fetch(e,t)}now(){return performance.now()}encode(e,t){if(t!=="utf-8"&&t!=="utf8")throw new Error(`Browser's encoder only supports utf-8, but got ${t}`);return this.textEncoder==null&&(this.textEncoder=new TextEncoder),this.textEncoder.encode(e)}decode(e,t){return new TextDecoder(t).decode(e)}};if(K().get("IS_BROWSER")){K().setPlatform("browser",new F9);try{Ss.registerManager(tl.URL_SCHEME,new R9)}catch(e){}try{Ss.registerManager(el.URL_SCHEME,new w9)}catch(e){}}var O9={importFetch:()=>XN()},S2,M9=class{constructor(){this.util=KN(),this.textEncoder=new this.util.TextEncoder}fetch(e,t){return K().global.fetch!=null?K().global.fetch(e,t):(S2==null&&(S2=O9.importFetch()),S2(e,t))}now(){let e=process.hrtime();return e[0]*1e3+e[1]/1e6}encode(e,t){if(t!=="utf-8"&&t!=="utf8")throw new Error(`Node built-in encoder only supports utf-8, but got ${t}`);return this.textEncoder.encode(e)}decode(e,t){return e.length===0?"":new this.util.TextDecoder(t).decode(e)}};K().get("IS_NODE")&&K().setPlatform("node",new M9);function ze(e,t="float32",n){return t=t||"float32",Qg(e),new en(e,t,n)}function z9(e,t){let n=D(e,"x","cast");if(!k5(t))throw new Error(`Failed to cast to unknown dtype ${t}`);if(t==="string"&&n.dtype!=="string"||t!=="string"&&n.dtype==="string")throw new Error("Only strings can be casted to strings");let s={x:n},r={dtype:t};return W.runKernel(Ra,s,r)}var de=V({cast_:z9});function L9(e){let n={x:D(e,"x","clone","string_or_numeric")};return W.runKernel(Ga,n)}var Ln=V({clone_:L9});function a3(e,t=!1){console.log(e.toString(t))}U5();var B9={buffer:ze,cast:de,clone:Ln,print:a3};ZE(B9);var ts={};Oe(ts,{browserFiles:()=>q9,browserHTTPRequest:()=>J9,concatenateArrayBuffers:()=>x2,copyModel:()=>D9,decodeWeights:()=>X5,encodeWeights:()=>l9,fromMemory:()=>eR,getLoadHandlers:()=>x9,getModelArtifactsForJSON:()=>b2,getModelArtifactsInfoForJSON:()=>wd,getSaveHandlers:()=>y9,http:()=>N2,isHTTPScheme:()=>T2,listModels:()=>$9,loadWeights:()=>X9,moveModel:()=>P9,registerLoadRouter:()=>A9,registerSaveRouter:()=>g9,removeModel:()=>_9,weightsLoaderFactory:()=>u3,withSaveHandler:()=>tR});var W9="model",V9=".json",U9=".weights.bin";function o3(e){return new Promise(t=>setTimeout(t)).then(e)}var I2=class{constructor(e){if(!K().getBool("IS_BROWSER"))throw new Error("browserDownloads() cannot proceed because the current environment is not a browser.");e.startsWith(I2.URL_SCHEME)&&(e=e.slice(I2.URL_SCHEME.length)),(e==null||e.length===0)&&(e=W9),this.modelJsonFileName=e+V9,this.weightDataFileName=e+U9}async save(e){if(typeof document=="undefined")throw new Error("Browser downloads are not supported in this environment since `document` is not present");let t=window.URL.createObjectURL(new Blob([e.weightData],{type:"application/octet-stream"}));if(e.modelTopology instanceof ArrayBuffer)throw new Error("BrowserDownloads.save() does not support saving model topology in binary formats yet.");{let n=[{paths:["./"+this.weightDataFileName],weights:e.weightSpecs}],s=Y5(e,n),r=window.URL.createObjectURL(new Blob([JSON.stringify(s)],{type:"application/json"})),a=this.modelJsonAnchor==null?document.createElement("a"):this.modelJsonAnchor;if(a.download=this.modelJsonFileName,a.href=r,await o3(()=>a.dispatchEvent(new MouseEvent("click"))),e.weightData!=null){let o=this.weightDataAnchor==null?document.createElement("a"):this.weightDataAnchor;o.download=this.weightDataFileName,o.href=t,await o3(()=>o.dispatchEvent(new MouseEvent("click")))}return{modelArtifactsInfo:wd(e)}}}},Qh=I2;Qh.URL_SCHEME="downloads://";var G9=class{constructor(e){if(e==null||e.length<1)throw new Error(`When calling browserFiles, at least 1 file is required, but received ${e}`);this.jsonFile=e[0],this.weightsFiles=e.slice(1)}async load(){return new Promise((e,t)=>{let n=new FileReader;n.onload=s=>{let r=JSON.parse(s.target.result),a=r.modelTopology;if(a==null){t(new Error(`modelTopology field is missing from file ${this.jsonFile.name}`));return}if(r.weightsManifest==null){t(new Error(`weightManifest field is missing from file ${this.jsonFile.name}`));return}if(this.weightsFiles.length===0){e({modelTopology:a});return}let i=b2(r,l=>this.loadWeights(l));e(i)},n.onerror=s=>t(`Failed to read model topology and weights manifest JSON from file '${this.jsonFile.name}'. BrowserFiles supports loading Keras-style tf.Model artifacts only.`),n.readAsText(this.jsonFile)})}loadWeights(e){let t=[],n=[];for(let a of e)t.push(...a.weights),n.push(...a.paths);let s=this.checkManifestAndWeightFiles(e),r=n.map(a=>this.loadWeightsFile(a,s[a]));return Promise.all(r).then(a=>[t,x2(a)])}loadWeightsFile(e,t){return new Promise((n,s)=>{let r=new FileReader;r.onload=a=>{let o=a.target.result;n(o)},r.onerror=a=>s(`Failed to weights data from file of path '${e}'.`),r.readAsArrayBuffer(t)})}checkManifestAndWeightFiles(e){let t=[],n=this.weightsFiles.map(r=>Z5(r.name)),s={};for(let r of e)r.paths.forEach(a=>{let o=Z5(a);if(t.indexOf(o)!==-1)throw new Error(`Duplicate file basename found in weights manifest: '${o}'`);if(t.push(o),n.indexOf(o)===-1)throw new Error(`Weight file with basename '${o}' is not provided.`);s[a]=this.weightsFiles[n.indexOf(o)]});if(t.length!==this.weightsFiles.length)throw new Error(`Mismatch in the number of files in weights manifest (${t.length}) and the number of weight files provided (${this.weightsFiles.length}).`);return s}},H9=e=>K().getBool("IS_BROWSER")&&!Array.isArray(e)&&e.startsWith(Qh.URL_SCHEME)?j9(e.slice(Qh.URL_SCHEME.length)):null;Lt.registerSaveRouter(H9);function j9(e="model"){return new Qh(e)}function q9(e){return new G9(e)}function i3(e,t,n,s){o(e),n=n==null?0:n,s=s==null?1:s,i(n,s);let r=0,a=l=>(l.then(c=>{let u=n+ ++r/e.length*(s-n);return t(u),c}),l);function o(l){O(l!=null&&Array.isArray(l)&&l.length>0,()=>"promises must be a none empty array")}function i(l,c){O(l>=0&&l<=1,()=>`Progress fraction must be in range [0, 1], but got startFraction ${l}`),O(c>=0&&c<=1,()=>`Progress fraction must be in range [0, 1], but got endFraction ${c}`),O(c>=l,()=>`startFraction must be no more than endFraction, but got startFraction ${l} and endFraction ${c}`)}return Promise.all(e.map(a))}async function l3(e,t){t==null&&(t={});let n=t.fetchFunc==null?K().platform.fetch:t.fetchFunc,s=e.map(d=>n(d,t.requestInit,{isBinary:!0})),r=0,a=.5,i=(t.onProgress==null?await Promise.all(s):await i3(s,t.onProgress,r,a)).map(d=>d.arrayBuffer()),l=.5,c=1;return t.onProgress==null?await Promise.all(i):await i3(i,t.onProgress,l,c)}async function X9(e,t="",n,s){return u3(o=>l3(o,{requestInit:s}))(e,t,n)}function u3(e){return async(t,n="",s)=>{let r=t.map(()=>!1),a={},o=s!=null?s.map(()=>!1):[],i=[];if(t.forEach((h,f)=>{let m=0;h.weights.forEach(g=>{let A="quantization"in g?g.quantization.dtype:g.dtype,x=A2[A]*Vt(g.shape),y=()=>{r[f]=!0,a[f]==null&&(a[f]=[]),a[f].push({manifestEntry:g,groupOffset:m,sizeBytes:x})};s!=null?s.forEach((b,w)=>{b===g.name&&(y(),o[w]=!0)}):y(),i.push(g.name),m+=x})}),!o.every(h=>h)){let h=s.filter((f,m)=>!o[m]);throw new Error(`Could not find weights in manifest with names: ${h.join(", ")}.
|
|
Manifest JSON has weights with names: ${i.join(", ")}.`)}let l=r.reduce((h,f,m)=>(f&&h.push(m),h),[]),c=[];l.forEach(h=>{t[h].paths.forEach(f=>{let m=n+(n.endsWith("/")?"":"/")+f;c.push(m)})});let u=await e(c),d={},p=0;return l.forEach(h=>{let f=t[h].paths.length,m=0;for(let b=0;b<f;b++)m+=u[p+b].byteLength;let g=new ArrayBuffer(m),A=new Uint8Array(g),x=0;for(let b=0;b<f;b++){let w=new Uint8Array(u[p+b]);A.set(w,x),x+=w.byteLength}a[h].forEach(b=>{let w=g.slice(b.groupOffset,b.groupOffset+b.sizeBytes),k=X5(w,[b.manifestEntry]);for(let C in k)d[C]=k[C]}),p+=f}),d}}var K9="application/octet-stream",Z9="application/json",C2=class{constructor(e,t){if(this.DEFAULT_METHOD="POST",t==null&&(t={}),this.weightPathPrefix=t.weightPathPrefix,this.onProgress=t.onProgress,this.weightUrlConverter=t.weightUrlConverter,t.fetchFunc!=null?(O(typeof t.fetchFunc=="function",()=>"Must pass a function that matches the signature of `fetch` (see https://developer.mozilla.org/en-US/docs/Web/API/Fetch_API)"),this.fetch=t.fetchFunc):this.fetch=K().platform.fetch,O(e!=null&&e.length>0,()=>"URL path for http must not be null, undefined or empty."),Array.isArray(e)&&O(e.length===2,()=>`URL paths for http must have a length of 2, (actual length is ${e.length}).`),this.path=e,t.requestInit!=null&&t.requestInit.body!=null)throw new Error("requestInit is expected to have no pre-existing body, but has one.");this.requestInit=t.requestInit||{}}async save(e){if(e.modelTopology instanceof ArrayBuffer)throw new Error("BrowserHTTPRequest.save() does not support saving model topology in binary formats yet.");let t=Object.assign({method:this.DEFAULT_METHOD},this.requestInit);t.body=new FormData;let n=[{paths:["./model.weights.bin"],weights:e.weightSpecs}],s=Y5(e,n);t.body.append("model.json",new Blob([JSON.stringify(s)],{type:Z9}),"model.json"),e.weightData!=null&&t.body.append("model.weights.bin",new Blob([e.weightData],{type:K9}),"model.weights.bin");let r=await this.fetch(this.path,t);if(r.ok)return{modelArtifactsInfo:wd(e),responses:[r]};throw new Error(`BrowserHTTPRequest.save() failed due to HTTP response status ${r.status}.`)}async load(){let e=await this.fetch(this.path,this.requestInit);if(!e.ok)throw new Error(`Request to ${this.path} failed with status code ${e.status}. Please verify this URL points to the model JSON of the model to load.`);let t;try{t=await e.json()}catch(r){let a=`Failed to parse model JSON of response from ${this.path}.`;throw this.path.endsWith(".pb")?a+=" Your path contains a .pb file extension. Support for .pb models have been removed in TensorFlow.js 1.0 in favor of .json models. You can re-convert your Python TensorFlow model using the TensorFlow.js 1.0 conversion scripts or you can convert your.pb models with the 'pb2json'NPM script in the tensorflow/tfjs-converter repository.":a+=" Please make sure the server is serving valid JSON for this request.",new Error(a)}let n=t.modelTopology,s=t.weightsManifest;if(n==null&&s==null)throw new Error(`The JSON from HTTP path ${this.path} contains neither model topology or manifest for weights.`);return b2(t,r=>this.loadWeights(r))}async loadWeights(e){let t=Array.isArray(this.path)?this.path[1]:this.path,[n,s]=Y9(t),r=this.weightPathPrefix||n,a=[];for(let c of e)a.push(...c.weights);let o=[],i=[];for(let c of e)for(let u of c.paths)this.weightUrlConverter!=null?i.push(this.weightUrlConverter(u)):o.push(r+u+s);this.weightUrlConverter&&o.push(...await Promise.all(i));let l=await l3(o,{requestInit:this.requestInit,fetchFunc:this.fetch,onProgress:this.onProgress});return[a,x2(l)]}};C2.URL_SCHEME_REGEX=/^https?:\/\//;function Y9(e){let t=e.lastIndexOf("/"),n=e.lastIndexOf("?"),s=e.substring(0,t),r=n>t?e.substring(n):"";return[s+"/",r]}function T2(e){return e.match(C2.URL_SCHEME_REGEX)!=null}var c3=(e,t)=>{if(typeof fetch=="undefined"&&(t==null||t.fetchFunc==null))return null;{let n=!0;if(Array.isArray(e)?n=e.every(s=>T2(s)):n=T2(e),n)return N2(e,t)}return null};Lt.registerSaveRouter(c3);Lt.registerLoadRouter(c3);function N2(e,t){return new C2(e,t)}function J9(e,t){return N2(e,t)}var E2=class{constructor(e){this.modelArtifacts=e}async load(){return this.modelArtifacts}},Q9=class{constructor(e){this.saveHandler=e}async save(e){return this.saveHandler(e)}};function eR(e,t,n,s){return arguments.length===1?e.modelTopology!=null||e.weightSpecs!=null?new E2(e):(console.warn("Please call tf.io.fromMemory() with only one argument. The argument should be of type ModelArtifacts. The multi-argument signature of tf.io.fromMemory() has been deprecated and will be removed in a future release."),new E2({modelTopology:e})):(console.warn("Please call tf.io.fromMemory() with only one argument. The argument should be of type ModelArtifacts. The multi-argument signature of tf.io.fromMemory() has been deprecated and will be removed in a future release."),new E2({modelTopology:e,weightSpecs:t,weightData:n,trainingConfig:s}))}function tR(e){return new Q9(e)}var d3={};Oe(d3,{confusionMatrix:()=>oR});function nR(e,t,n=!1,s=!1){let r=D(e,"a","matMul"),a=D(t,"b","matMul");[r,a]=Dt(r,a);let o={a:r,b:a},i={transposeA:n,transposeB:s};return W.runKernel(Ea,o,i)}var Ue=V({matMul_:nR});function sR(e,t,n=1,s=0){if(t<2)throw new Error(`Error in oneHot: depth must be >=2, but it is ${t}`);let a={indices:D(e,"indices","oneHot","int32")},o={depth:t,onValue:n,offValue:s};return W.runKernel(_i,a,o)}var kd=V({oneHot_:sR});function rR(e,t){let n=D(e,"x","transpose");if(t==null&&(t=n.shape.map((a,o)=>o).reverse()),O(n.rank===t.length,()=>`Error in transpose: rank of input ${n.rank} must match length of perm ${t}.`),t.forEach(a=>{O(a>=0&&a<n.rank,()=>`All entries in 'perm' must be between 0 and ${n.rank-1} but got ${t}`)}),n.rank<=1)return n.clone();let s={x:n},r={perm:t};return W.runKernel(go,s,r)}var Ye=V({transpose_:rR});function aR(e,t,n){let s=D(e,"labels","confusionMatrix"),r=D(t,"predictions","confusionMatrix");O(n==null||n>0&&Number.isInteger(n),()=>`If provided, numClasses must be a positive integer, but got ${n}`),O(s.rank===1,()=>`Expected the rank of labels to be 1, but got ${s.rank}`),O(r.rank===1,()=>`Expected the rank of predictions to be 1, but got ${r.rank}`),O(s.shape[0]===r.shape[0],()=>`Mismatch in the number of examples: ${s.shape[0]} vs. ${r.shape[0]}. Labels and predictions should have the same number of elements.`),O(n>0&&Number.isInteger(n),()=>`numClasses is required to be a positive integer, but got ${n}`);let a=kd(de(s,"int32"),n),o=kd(de(r,"int32"),n),i=Ye(a),l=Ue(i,o);return de(l,"int32")}var oR=V({confusionMatrix_:aR}),nl={};Oe(nl,{assertAndGetBroadcastShape:()=>mt,getBroadcastDims:()=>p3,getReductionAxes:()=>qt});function p3(e,t){let n=e.length,s=[];for(let r=0;r<n;r++){let a=n-1-r,o=e[a]||1;(t[t.length-1-r]||1)>1&&o===1&&s.unshift(a)}return s}function qt(e,t){let n=[];for(let s=0;s<t.length;s++){let r=e[e.length-s-1],a=t.length-s-1,o=t[a];(r==null||r===1&&o>1)&&n.unshift(a)}return n}function mt(e,t){let n=[],s=Math.max(e.length,t.length);for(let r=0;r<s;r++){let a=e[e.length-r-1];a==null&&(a=1);let o=t[t.length-r-1];if(o==null&&(o=1),a===1)n.unshift(o);else if(o===1)n.unshift(a);else if(a!==o){let i=`Operands could not be broadcast together with shapes ${e} and ${t}.`;throw Error(i)}else n.unshift(a)}return n}var Gs={};Oe(Gs,{fromPixels:()=>hR,fromPixelsAsync:()=>dR,toPixels:()=>pR});function h3(e,t,n){if(li(e),t!=null&&t.length!==3)throw new Error("tensor3d() requires shape to have three numbers");let s=Cr(e,n);if(s.length!==3&&s.length!==1)throw new Error("tensor3d() requires values to be number[][][] or flat/TypedArray");if(s.length===1&&t==null)throw new Error("tensor3d() requires shape to be provided when `values` are a flat array");return So(e,t,s,n)}var sl;function f3(e,t=3){if(t>4)throw new Error("Cannot construct Tensor with more than 4 channels from pixels.");if(e==null)throw new Error("pixels passed to tf.browser.fromPixels() can not be null");let n=!1,s=!1,r=!1,a=!1,o=!1,i=!1;if(e.data instanceof Uint8Array)n=!0;else if(typeof ImageData!="undefined"&&e instanceof ImageData)s=!0;else if(typeof HTMLVideoElement!="undefined"&&e instanceof HTMLVideoElement)r=!0;else if(typeof HTMLImageElement!="undefined"&&e instanceof HTMLImageElement)a=!0;else if(e.getContext!=null)o=!0;else if(typeof ImageBitmap!="undefined"&&e instanceof ImageBitmap)i=!0;else throw new Error(`pixels passed to tf.browser.fromPixels() must be either an HTMLVideoElement, HTMLImageElement, HTMLCanvasElement, ImageData in browser, or OffscreenCanvas, ImageData in webworker or {data: Uint32Array, width: number, height: number}, but was ${e.constructor.name}`);if(r){let f=2;if(r&&e.readyState<f)throw new Error("The video element has not loaded data yet. Please wait for `loadeddata` event on the <video> element.")}if(jh(dd,W.backendName)!=null){let f={pixels:e},m={numChannels:t};return W.runKernel(dd,f,m)}let[c,u]=r?[e.videoWidth,e.videoHeight]:[e.width,e.height],d;if(o)d=e.getContext("2d").getImageData(0,0,c,u).data;else if(s||n)d=e.data;else if(a||r||i){if(sl==null)if(typeof document=="undefined")if(typeof OffscreenCanvas!="undefined"&&typeof OffscreenCanvasRenderingContext2D!="undefined")sl=new OffscreenCanvas(1,1).getContext("2d");else throw new Error("Cannot parse input in current context. Reason: OffscreenCanvas Context2D rendering is not supported.");else sl=document.createElement("canvas").getContext("2d");sl.canvas.width=c,sl.canvas.height=u,sl.drawImage(e,0,0,c,u),d=sl.getImageData(0,0,c,u).data}let p;if(t===4)p=new Int32Array(d);else{let f=c*u;p=new Int32Array(f*t);for(let m=0;m<f;m++)for(let g=0;g<t;++g)p[m*t+g]=d[m*4+g]}return h3(p,[u,c,t],"int32")}function iR(e){return e!=null&&e.data instanceof Uint8Array}function lR(){return typeof window!="undefined"&&typeof ImageBitmap!="undefined"&&window.hasOwnProperty("createImageBitmap")}function uR(e){return e!=null&&e.width!==0&&e.height!==0}function cR(e){return lR()&&!(e instanceof ImageBitmap)&&uR(e)&&!iR(e)}async function dR(e,t=3){let n=null;if(K().getBool("WRAP_TO_IMAGEBITMAP")&&cR(e)){let s;try{s=await createImageBitmap(e,{premultiplyAlpha:"none"})}catch(r){s=null}s!=null&&s.width===e.width&&s.height===e.height?n=s:n=e}else n=e;return f3(n,t)}async function pR(e,t){let n=D(e,"img","toPixels");if(!(e instanceof Ze)){let c=n;n=de(c,"int32"),c.dispose()}if(n.rank!==2&&n.rank!==3)throw new Error(`toPixels only supports rank 2 or 3 tensors, got rank ${n.rank}.`);let[s,r]=n.shape.slice(0,2),a=n.rank===2?1:n.shape[2];if(a>4||a===2)throw new Error(`toPixels only supports depth of size 1, 3 or 4 but got ${a}`);if(n.dtype!=="float32"&&n.dtype!=="int32")throw new Error(`Unsupported type for toPixels: ${n.dtype}. Please use float32 or int32 tensors.`);let o=await n.data(),i=n.dtype==="float32"?255:1,l=new Uint8ClampedArray(r*s*4);for(let c=0;c<s*r;++c){let u=[0,0,0,255];for(let p=0;p<a;p++){let h=o[c*a+p];if(n.dtype==="float32"){if(h<0||h>1)throw new Error(`Tensor values for a float32 Tensor must be in the range [0 - 1] but encountered ${h}.`)}else if(n.dtype==="int32"&&(h<0||h>255))throw new Error(`Tensor values for a int32 Tensor must be in the range [0 - 255] but encountered ${h}.`);a===1?(u[0]=h*i,u[1]=h*i,u[2]=h*i):u[p]=h*i}let d=c*4;l[d+0]=Math.round(u[0]),l[d+1]=Math.round(u[1]),l[d+2]=Math.round(u[2]),l[d+3]=Math.round(u[3])}if(t!=null){t.width=r,t.height=s;let c=t.getContext("2d"),u=new ImageData(l,r,s);c.putImageData(u,0,0)}return n!==e&&n.dispose(),l}var hR=V({fromPixels_:f3}),R2={};Oe(R2,{prepareAndValidate:()=>m3});function m3(e,t){let n=e.shape.length,s=t.shape.length;if(n<1)throw new Error(`tf.gatherND() expects the input to be rank 1 or higher, but the rank was ${n}.`);if(s<1)throw new Error(`tf.gatherND() expects the indices to be rank 1 or higher, but the rank was ${s}.`);if(t.dtype!=="int32")throw new Error(`tf.gatherND() expects the indices to be int32 type, but the dtype was ${t.dtype}.`);if(t.shape[s-1]>n)throw new Error(`index innermost dimension length must be <= tensor rank; saw: ${t.shape[s-1]} vs. ${n}`);if(Vt(e.shape)===0)throw new Error(`Requested more than 0 entries, but input is empty. Input shape: ${e.shape}.`);let r=t.shape,a=r[r.length-1],o=1;for(let d=0;d<r.length-1;++d)o*=r[d];let i=e.shape,l=r.slice();l.pop();let c=1;for(let d=a;d<n;++d)c*=i[d],l.push(i[d]);let u=[...Jl(e.shape).map(d=>d/c),1].slice(0,a);return[l,o,c,u]}var $2={};Oe($2,{calculateShapes:()=>g3,validateInput:()=>D2,validateUpdateShape:()=>_2});function _2(e,t,n){let s=t.rank>1?t.shape[t.rank-1]:1,r=t.rank>1?t.rank-1:1,a=`Must have updates.shape = indices.shape[:batchDim] + shape[sliceDim:], got updates.shape: ${n.shape}, indices.shape: ${t.shape}, shape: ${e}, sliceDim: ${s}, and batchDim: ${r}.`;if(n.rank<r)throw new Error(a+` update.rank < ${r}. `);if(e.length<s+(n.rank-r))throw new Error(a+` Output shape length < ${s+(n.rank-r)}`);if(n.rank!==r+e.length-s)throw new Error(a+` update.rank != ${r+e.length-s}`);for(let o=0;o<r;++o)if(n.shape[o]!==t.shape[o])throw new Error(a+` updates.shape[${o}] (${n.shape[o]}) != indices.shape[${o}] (${t.shape[o]}).`);for(let o=0;o<n.rank-r;++o)if(n.shape[o+r]!==e[o+s])throw new Error(a+` updates.shape[${o+r}] (${n.shape[o+r]}) != shape[${o+r}] (${e[o+r]})`)}function D2(e,t,n){if(t.rank<1)throw new Error(`tf.scatterND() expects the indices to be rank 1 or higher, but the rank was ${t.rank}.`);if(e.rank<1)throw new Error(`tf.scatterND() expects the updates to be rank 1 or higher, but the rank was ${e.rank}.`);if(t.dtype!=="int32")throw new Error(`The dtype of 'indices' should be int32, but got dtype: ${t.dtype}`);if(n.length<1)throw new Error(`Output rank must be greater or equal to 1, but got shape: ${n}`);if(n.length===0){if(t.size===0)throw new Error(`Indices specified for empty output. indices shape: ${t.shape}`);if(e.size===0)throw new Error(`Updates specified for empty output. updates shape: ${e.shape}`)}_2(n,t,e)}function g3(e,t,n){let s=t.shape.length,r=s>1?t.shape[s-1]:1,a=n.length,o=1;for(let d=r;d<a;++d)o*=n[d];let i=r<1?1:r,l=Vt(t.shape)/i,c=[...Jl(n.slice(0,r)),1],u=Vt(n);return{sliceRank:r,numUpdates:l,sliceSize:o,strides:c,outputSize:u}}var Pt={};Oe(Pt,{assertParamsValid:()=>mR,computeFlatOffset:()=>bR,computeOutShape:()=>AR,getNormalizedAxes:()=>yR,isSliceContinous:()=>xR,maskToAxes:()=>gR,parseSliceParams:()=>I3,sliceInfo:()=>vR,startForAxis:()=>k3,startIndicesWithElidedDims:()=>b3,stopForAxis:()=>S3,stopIndicesWithElidedDims:()=>v3,stridesForAxis:()=>w3,stridesWithElidedDims:()=>A3});var P2=-2,fR=-1;function mR(e,t,n){let s=e.shape.length;O(s===t.length,()=>`Error in slice${s}D: Length of begin ${t} must match the rank of the array (${s}).`),O(s===n.length,()=>`Error in slice${s}D: Length of size ${n} must match the rank of the array (${s}).`);for(let r=0;r<s;++r)O(t[r]+n[r]<=e.shape[r],()=>`Error in slice${s}D: begin[${r}] + size[${r}] (${t[r]+n[r]}) would overflow input.shape[${r}] (${e.shape[r]})`)}function gR(e){let t=[],n=0;for(;e>0;)e&1&&t.push(n),e/=2,n++;return t}function AR(e,t,n){let s=[];for(let r=0;r<e.length;r++)s[r]=Math.ceil((t[r]-e[r])/n[r]);return s}function A3(e,t,n,s){let r=[...e];for(let a=r.length;a<s.length;a++)r.push(1);for(let a=0;a<n;a++)a===0?r[t]=1:(r.splice(t,0,1),r.pop());return r}function y3(e,t,n){return n<=e?n:n-(t-1)}function x3(e,t){let n=[];for(let s=0;s<e;s++)n.push(t+s);return n}function yR(e,t,n,s,r,a,o,i,l){let c=e.length,u=new Array(c),d=new Array(c),p=new Array(c);if(t.length&&n>0){let h=t[0],f=n+1;u=b3(o,h,f,s,e),d=v3(i,h,f,r,e),p=A3(a,h,f,e)}else for(let h=0;h<c;h++)u[h]=k3(o,s,a,e,h,l),d[h]=S3(i,r,a,e,h,l),p[h]=w3(a,h,l);return{begin:u,end:d,strides:p}}function b3(e,t,n,s,r){let a=[...r],o=x3(n,t);for(let i=0;i<a.length;i++)if(o.indexOf(i)>-1)a[i]=0;else{let l=y3(t,n,i),c=s[l];e&1<<l&&(c=0),a[i]=c}return a}function v3(e,t,n,s,r){let a=[...r],o=x3(n,t);for(let i=0;i<a.length;i++)if(o.indexOf(i)>-1)a[i]=Number.MAX_SAFE_INTEGER;else{let l=y3(t,n,i),c=s[l];e&1<<l&&(c=Number.MAX_SAFE_INTEGER),a[i]=c}for(let i=0;i<a.length;i++){let l=r[i];a[i]<0&&(a[i]+=l),a[i]=Kc(0,a[i],r[i])}return a}function w3(e,t,n){let s=e[t];return(n&1<<t||s==null)&&(s=1),s}function k3(e,t,n,s,r,a){let o=t[r],i=n[r]||1;(e&1<<r||a&1<<r||o==null)&&(i>0?o=Number.MIN_SAFE_INTEGER:o=Number.MAX_SAFE_INTEGER);let l=s[r];return o<0&&(o+=l),o=Kc(0,o,l-1),o}function S3(e,t,n,s,r,a){let o=t[r],i=n[r]||1;(e&1<<r||a&1<<r||o==null)&&(i>0?o=Number.MAX_SAFE_INTEGER:o=Number.MIN_SAFE_INTEGER);let l=s[r];return o<0&&(o+=l),i>0?o=Kc(0,o,l):o=Kc(-1,o,l-1),o}function xR(e,t,n){let s=n.length;for(let r=0;r<n.length;r++)if(n[r]>1){s=r;break}for(let r=s+1;r<n.length;r++)if(t[r]>0||n[r]!==e[r])return!1;return!0}function bR(e,t){let n=e.length>0?e[e.length-1]:1;for(let s=0;s<e.length-1;s++)n+=e[s]*t[s];return n}function I3(e,t,n){let s,r=e.shape.length;typeof t=="number"?s=[t,...new Array(r-1).fill(0)]:t.length<r?s=t.concat(new Array(r-t.length).fill(0)):s=t.slice(),s.forEach(o=>{O(o!==-1,()=>"slice() does not support negative begin indexing.")});let a;return n==null?a=new Array(r).fill(-1):typeof n=="number"?a=[n,...new Array(r-1).fill(-1)]:n.length<r?a=n.concat(new Array(r-n.length).fill(-1)):a=n,a=a.map((o,i)=>o>=0?o:(O(o===-1,()=>`Negative size values should be exactly -1 but got ${o} for the slice() size at index ${i}.`),e.shape[i]-s[i])),[s,a]}function vR(e,t,n,s,r,a,o,i,l){let c;if(s==null?(c=new Array(t.length),c.fill(1)):c=s,o!=null&&(o&o-1)!=0)throw new Error("Multiple ellipses in slice is not allowed.");let u=!1,d={dims:c.length,numAddAxisAfterEllipsis:0,begin:t.slice(),end:n.slice(),strides:c.slice(),beginMask:r,endMask:a,ellipsisMask:o,newAxisMask:i,shrinkAxisMask:l};for(let y=0;y<d.dims;y++)u&&(1<<y&i)!=0&&d.numAddAxisAfterEllipsis++,1<<y&o&&(u=!0);u||(d.ellipsisMask|=1<<d.dims,d.dims++);let p={dims:e.length,beginMask:0,endMask:0,beginValid:!1,endValid:!1};wR(d,p);let h=!0,f=!0,m=!0,g=[],A=[];for(let y=0;y<e.length;++y){if(p.strides[y]===0)throw Error(`strides[${y}] must be non-zero`);let b=!!(p.shrinkAxisMask&1<<y),w=e[y];if(w===-1){g.push(b?1:-1);continue}let k=[p.beginMask&1<<y,p.endMask&1<<y],C=[p.strides[y]>0?0:-1,p.strides[y]>0?w:w-1];if(b&&p.strides[y]<=0)throw Error("only stride 1 allowed on non-range indexing.");m=m&&p.strides[y]===1;let N=!!(p.beginMask&1<<y&&p.endMask&1<<y);if(p.beginValid&&p.endValid){if(b){let P=p.begin[y]<0?w+p.begin[y]:p.begin[y];if(p.begin[y]=P,p.end[y]=p.begin[y]+1,P<0||P>=w)throw Error(`slice index ${p.begin[y]} of dimension ${y} out of bounds.`)}else p.begin[y]=C3(p.begin[y],0,p.strides[y],w,k,C),p.end[y]=C3(p.end[y],1,p.strides[y],w,k,C);let _=p.strides[y]===1&&p.begin[y]===0&&p.end[y]===w;h=h&&_,f=f&&(y===0&&p.strides[y]===1||_)}else h=h&&p.strides[y]===1&&N,f=f&&(y===0&&p.strides[y]===1||N);let R,F=!1;if(p.beginValid&&p.endValid?(R=p.end[y]-p.begin[y],F=!0):b?(R=1,F=!0):N&&w>=0&&(p.strides[y]<0?R=-w:R=w,F=!0),F){let _;R===0||R<0!=p.strides[y]<0?_=0:_=Math.trunc(R/p.strides[y])+(R%p.strides[y]!=0?1:0),g.push(_)}else g.push(-1)}for(let y=0;y<p.finalShapeGatherIndices.length;++y){let b=p.finalShapeGatherIndices[y];b>=0?A.push(g[b]):b===P2&&A.push(1)}return{finalShapeSparse:A.filter((y,b)=>p.finalShapeGatherIndices[b]!==P2),finalShape:A,isIdentity:h,sliceDim0:f,isSimpleSlice:m,begin:p.begin,end:p.end,strides:p.strides}}function wR(e,t){t.beginMask=0,t.endMask=0,t.shrinkAxisMask=0;let n=0;t.beginValid=e.begin!=null,t.endValid=e.end!=null,t.begin=new Array(t.dims),t.end=new Array(t.dims),t.strides=new Array(t.dims),t.finalShapeGatherIndices=[],t.finalShapeGatherIndicesSparse=[],t.inputShapeGatherIndicesSparse=new Array(t.dims);for(let s=0;s<e.dims;s++)if(1<<s&e.ellipsisMask){let r=Math.min(t.dims-(e.dims-s)+1+e.numAddAxisAfterEllipsis,t.dims);for(;n<r;n++)t.begin[n]=0,t.end[n]=0,t.strides[n]=1,t.beginMask|=1<<n,t.endMask|=1<<n,t.finalShapeGatherIndices.push(n),t.finalShapeGatherIndicesSparse.push(-1),t.inputShapeGatherIndicesSparse[n]=s}else if(1<<s&e.newAxisMask)t.finalShapeGatherIndices.push(P2),t.finalShapeGatherIndicesSparse.push(-1);else{if(n===t.begin.length)throw Error(`Index out of range using input dim ${n}; input has only ${t.dims} dims, ${t.begin.length}.`);e.begin!=null&&(t.begin[n]=e.begin[s]),e.end!=null&&(t.end[n]=e.end[s]),t.strides[n]=e.strides[s],e.beginMask&1<<s&&(t.beginMask|=1<<n),e.endMask&1<<s&&(t.endMask|=1<<n),e.shrinkAxisMask&1<<s?(t.finalShapeGatherIndices.push(fR),t.finalShapeGatherIndicesSparse.push(-1),t.shrinkAxisMask|=1<<n):(t.finalShapeGatherIndices.push(n),t.finalShapeGatherIndicesSparse.push(s)),t.inputShapeGatherIndicesSparse[n]=s,n++}}function C3(e,t,n,s,r,a){if(r[t])return n>0?a[t]:a[t+1&1];{let o=e<0?s+e:e;return o<a[0]?a[0]:o>a[1]?a[1]:o}}var ue={};Oe(ue,{Serializable:()=>T3,SerializationMap:()=>rl,registerClass:()=>Co});var T3=class{getClassName(){return this.constructor.className}static fromConfig(e,t){return new e(t)}},rl=class{constructor(){this.classNameMap={}}static getMap(){return rl.instance==null&&(rl.instance=new rl),rl.instance}static register(e){rl.getMap().classNameMap[e.className]=[e,e.fromConfig]}};function Co(e){O(e.className!=null,()=>"Class being registered does not have the static className property defined."),O(typeof e.className=="string",()=>"className is required to be a string, but got type "+typeof e.className),O(e.className.length>0,()=>"Class being registered has an empty-string as its className, which is disallowed."),rl.register(e)}var N3={};Oe(N3,{TEST_EPSILON_FLOAT16:()=>E3,encodeStrings:()=>R3,expectArrayBuffersEqual:()=>ER,expectArraysClose:()=>SR,expectArraysEqual:()=>CR,expectNumbersClose:()=>TR,expectPromiseToFail:()=>IR,expectValuesInRange:()=>NR,testEpsilon:()=>F2});var kR=.001,E3=.1;function SR(e,t,n){return n==null&&(n=F2()),O2(e,t,(s,r)=>M2(s,r,n))}function F2(){return W.backend.floatPrecision()===32?kR:E3}function O2(e,t,n){let s=!0;if(($n(e)||$n(t))&&(s=!1),$n(e)&&$n(t)&&(s=!0),s){let o=e.constructor.name,i=t.constructor.name;if(o!==i)throw new Error(`Arrays are of different type. Actual: ${o}. Expected: ${i}`)}if(Array.isArray(e)&&Array.isArray(t)){let o=Cr(e),i=Cr(t);if(!ka(o,i))throw new Error(`Arrays have different shapes. Actual: [${o}]. Expected: [${i}]`)}let r=$n(e)?e:ui(e),a=$n(t)?t:ui(t);if(r.length!==a.length)throw new Error(`Arrays have different lengths actual: ${r.length} vs expected: ${a.length}.
|
|
Actual: ${r}.
|
|
Expected: ${a}.`);for(let o=0;o<a.length;++o){let i=r[o],l=a[o];if(!n(i,l))throw new Error(`Arrays differ: actual[${o}] = ${i}, expected[${o}] = ${l}.
|
|
Actual: ${r}.
|
|
Expected: ${a}.`)}}function IR(e,t){e().then(()=>t.fail(),()=>t())}function CR(e,t){let n=typeof t=="string"||typeof t=="number"||typeof t=="boolean"?[t]:t;return Sa(e)||Sa(e[0])||Sa(t)||Sa(t[0])?O2(e,n,(s,r)=>s==r):O2(e,t,(s,r)=>M2(s,r,0))}function TR(e,t,n){if(n==null&&(n=F2()),!M2(e,t,n))throw new Error(`Numbers differ: actual === ${e}, expected === ${t}`)}function M2(e,t,n){return!isFinite(e)&&!isFinite(t)?!0:!(isNaN(e)||isNaN(t)||Math.abs(e-t)>n)}function NR(e,t,n){for(let s=0;s<e.length;s++)if(e[s]<t||e[s]>n)throw new Error(`Value out of range:${e[s]} low: ${t}, high: ${n}`)}function ER(e,t){expect(new Float32Array(e)).toEqual(new Float32Array(t))}function R3(e){for(let t=0;t<e.length;t++){let n=e[t];Array.isArray(n)?R3(n):e[t]=md(n)}return e}function $3(){K().set("PROD",!0)}function RR(){K().set("DEBUG",!0)}function $R(){K().set("DEPRECATION_WARNINGS_ENABLED",!1),console.warn("TensorFlow.js deprecation warnings have been disabled.")}function z2(e){K().getBool("DEPRECATION_WARNINGS_ENABLED")&&console.warn(e+" You can disable deprecation warnings with tf.disableDeprecationWarnings().")}YE(z2);function _R(){W.disposeVariables()}function ns(){return W}function ef(){return W.memory()}function DR(e){return W.profile(e)}function q(e,t){return W.tidy(e,t)}function se(e){p2(e).forEach(n=>n.dispose())}function fn(e){return W.keep(e)}function PR(e){return W.time(e)}function _3(e){return W.setBackend(e)}function tf(){return W.ready()}function Is(){return W.backendName}function FR(e){W.removeBackend(e)}function L2(e){return W.findBackend(e)}function OR(e){return W.findBackendFactory(e)}function al(e,t,n=1){return W.registerBackend(e,t,n)}function Tr(){return W.backend}function MR(e,t){K().setPlatform(e,t)}function zR(e,t){let n=D(e,"a","add"),s=D(t,"b","add");[n,s]=Dt(n,s);let r={a:n,b:s};return W.runKernel(Gr,r)}var le=V({add_:zR});function LR(e,t){let n=D(e,"a","floorDiv"),s=D(t,"b","floorDiv");[n,s]=Dt(n,s);let r={a:n,b:s};return W.runKernel(Wa,r)}var nf=V({floorDiv_:LR});function BR(e,t){let n=D(e,"a","div"),s=D(t,"b","div");if([n,s]=Dt(n,s),n.dtype==="int32"&&s.dtype==="int32")return nf(n,s);let r={a:n,b:s},a={};return W.runKernel(Ma,r,a)}var pe=V({div_:BR});function WR(e,t){let n=D(e,"a","mul"),s=D(t,"b","mul");[n,s]=Dt(n,s);let r={a:n,b:s};return W.runKernel(Qa,r)}var L=V({mul_:WR});function VR(e){let t=D(e,"x","abs");if(t.dtype==="complex64"){let n={x:t};return W.runKernel(Qc,n)}else{let n={x:t};return W.runKernel(ci,n)}}var tn=V({abs_:VR});function UR(e){let n={x:D(e,"x","acos")};return W.runKernel(eu,n)}var D3=V({acos_:UR});function GR(e){let n={x:D(e,"x","acosh")};return W.runKernel(tu,n)}var P3=V({acosh_:GR});function HR(e){O(Array.isArray(e),()=>"The argument passed to tf.addN() must be a list of tensors"),O(e.length>=1,()=>`Must pass at least one tensor to tf.addN(), but got ${e.length}`);let t=e.map((r,a)=>D(r,`tensors${a}`,"addN")),n=t[0];t.forEach(r=>{if(r.dtype!==n.dtype)throw new Error("All tensors passed to tf.addN() must have the same dtype")}),t.forEach(r=>{if(!ka(r.shape,n.shape))throw new Error("All tensors passed to tf.addN() must have the same shape")});let s=t;return W.runKernel(Ca,s)}var sf=V({addN_:HR});function jR(e,t=null,n=!1){let r={x:D(e,"x","all","bool")},a={axis:t,keepDims:n};return W.runKernel(nu,r,a)}var B2=V({all_:jR});function qR(e,t=null,n=!1){let r={x:D(e,"x","any","bool")},a={axis:t,keepDims:n};return W.runKernel(su,r,a)}var rf=V({any_:qR});function XR(e,t=0){let s={x:D(e,"x","argMax")},r={axis:t};return W.runKernel(Ta,s,r)}var Hs=V({argMax_:XR});function KR(e,t=0){let s={x:D(e,"x","argMin")},r={axis:t};return W.runKernel(ru,s,r)}var F3=V({argMin_:KR});function ZR(e){let n={x:D(e,"x","asin")};return W.runKernel(au,n)}var O3=V({asin_:ZR});function YR(e){let n={x:D(e,"x","asinh")};return W.runKernel(ou,n)}var M3=V({asinh_:YR});function JR(e){let n={x:D(e,"x","atan")};return W.runKernel(iu,n)}var z3=V({atan_:JR});function QR(e,t){let n=D(e,"a","atan2"),s=D(t,"b","atan2");[n,s]=Dt(n,s);let r={a:n,b:s};return W.runKernel(uu,r)}var L3=V({atan2_:QR});function e$(e){let n={x:D(e,"x","atanh")};return W.runKernel(lu,n)}var B3=V({atanh_:e$});function t$(e,t,n,s,r="NHWC",a){let o=e[3],i=[...t,o],l=U3(r);return Sd(e,i,n,a,s,null,null,l)}function W3(e,t,n,s,r,a,o="channelsLast"){let[i,l]=af(t),c;if(o==="channelsLast")c=[i,l,e[3],e[3]];else if(o==="channelsFirst")c=[i,l,e[1],e[1]];else throw new Error(`Unknown dataFormat ${o}`);return Sd(e,c,n,s,r,a,!1,o)}function n$(e,t,n,s,r,a,o="NDHWC"){let[i,l,c]=V2(t),u,d;if(o==="NDHWC")d="channelsLast",u=[i,l,c,e[4],e[4]];else if(o==="NCDHW")d="channelsFirst",u=[i,l,c,e[1],e[1]];else throw new Error(`Unknown dataFormat ${o}`);return V3(e,u,n,s,r,!1,d,a)}function Sd(e,t,n,s,r,a,o=!1,i="channelsLast"){let[l,c,u,d]=[-1,-1,-1,-1];if(i==="channelsLast")[l,c,u,d]=e;else if(i==="channelsFirst")[l,d,c,u]=e;else throw new Error(`Unknown dataFormat ${i}`);let[p,h,,f]=t,[m,g]=af(n),[A,x]=af(s),y=$u(p,A),b=$u(h,x),{padInfo:w,outHeight:k,outWidth:C}=a$(r,c,u,m,g,y,b,a,i),N=o?f*d:f,R;return i==="channelsFirst"?R=[l,N,k,C]:i==="channelsLast"&&(R=[l,k,C,N]),{batchSize:l,dataFormat:i,inHeight:c,inWidth:u,inChannels:d,outHeight:k,outWidth:C,outChannels:N,padInfo:w,strideHeight:m,strideWidth:g,filterHeight:p,filterWidth:h,effectiveFilterHeight:y,effectiveFilterWidth:b,dilationHeight:A,dilationWidth:x,inShape:e,outShape:R,filterShape:t}}function V3(e,t,n,s,r,a=!1,o="channelsLast",i){let[l,c,u,d,p]=[-1,-1,-1,-1,-1];if(o==="channelsLast")[l,c,u,d,p]=e;else if(o==="channelsFirst")[l,p,c,u,d]=e;else throw new Error(`Unknown dataFormat ${o}`);let[h,f,m,,g]=t,[A,x,y]=V2(n),[b,w,k]=V2(s),C=$u(h,b),N=$u(f,w),R=$u(m,k),{padInfo:F,outDepth:_,outHeight:P,outWidth:T}=o$(r,c,u,d,A,x,y,C,N,R,i),M=a?g*p:g,U;return o==="channelsFirst"?U=[l,M,_,P,T]:o==="channelsLast"&&(U=[l,_,P,T,M]),{batchSize:l,dataFormat:o,inDepth:c,inHeight:u,inWidth:d,inChannels:p,outDepth:_,outHeight:P,outWidth:T,outChannels:M,padInfo:F,strideDepth:A,strideHeight:x,strideWidth:y,filterDepth:h,filterHeight:f,filterWidth:m,effectiveFilterDepth:C,effectiveFilterHeight:N,effectiveFilterWidth:R,dilationDepth:b,dilationHeight:w,dilationWidth:k,inShape:e,outShape:U,filterShape:t}}function s$(e,t,n,s,r){s==null&&(s=W2(e,t,n));let a=e[0],o=e[1],i=ol((a-t+2*s)/n+1,r),l=ol((o-t+2*s)/n+1,r);return[i,l]}function r$(e,t,n,s,r,a){r==null&&(r=W2(e,t,s));let o=e[0],i=e[1],l=e[2],c=ol((o-t+2*r)/s+1,a),u=ol((i-t+2*r)/s+1,a),d=ol((l-t+2*r)/s+1,a);return[c,u,d,n]}function W2(e,t,n,s=1){let r=$u(t,s);return Math.floor((e[0]*(n-1)-n+r)/2)}function af(e){return typeof e=="number"?[e,e,e]:e.length===2?[e[0],e[1],1]:e}function V2(e){return typeof e=="number"?[e,e,e]:e}function $u(e,t){return t<=1?e:e+(e-1)*(t-1)}function a$(e,t,n,s,r,a,o,i,l){let c,u,d;if(typeof e=="number"){c={top:e,bottom:e,left:e,right:e,type:e===0?"VALID":"NUMBER"};let h=s$([t,n],a,s,e,i);u=h[0],d=h[1]}else if(e==="same"){u=Math.ceil(t/s),d=Math.ceil(n/r);let p=Math.max(0,(u-1)*s+a-t),h=Math.max(0,(d-1)*r+o-n),f=Math.floor(p/2),m=p-f,g=Math.floor(h/2),A=h-g;c={top:f,bottom:m,left:g,right:A,type:"SAME"}}else if(e==="valid")c={top:0,bottom:0,left:0,right:0,type:"VALID"},u=Math.ceil((t-a+1)/s),d=Math.ceil((n-o+1)/r);else if(typeof e=="object"){let p=l==="channelsLast"?e[1][0]:e[2][0],h=l==="channelsLast"?e[1][1]:e[2][1],f=l==="channelsLast"?e[2][0]:e[3][0],m=l==="channelsLast"?e[2][1]:e[3][1];c={top:p,bottom:h,left:f,right:m,type:p===0&&h===0&&f===0&&m===0?"VALID":"EXPLICIT"},u=ol((t-a+p+h)/s+1,i),d=ol((n-o+f+m)/r+1,i)}else throw Error(`Unknown padding parameter: ${e}`);return{padInfo:c,outHeight:u,outWidth:d}}function o$(e,t,n,s,r,a,o,i,l,c,u){let d,p,h,f;if(typeof e=="number"){d={top:e,bottom:e,left:e,right:e,front:e,back:e,type:e===0?"VALID":"NUMBER"};let g=r$([t,n,s,1],i,1,r,e,u);p=g[0],h=g[1],f=g[2]}else if(e==="same"){p=Math.ceil(t/r),h=Math.ceil(n/a),f=Math.ceil(s/o);let m=(p-1)*r+i-t,g=(h-1)*a+l-n,A=(f-1)*o+c-s,x=Math.floor(m/2),y=m-x,b=Math.floor(g/2),w=g-b,k=Math.floor(A/2),C=A-k;d={top:b,bottom:w,left:k,right:C,front:x,back:y,type:"SAME"}}else if(e==="valid")d={top:0,bottom:0,left:0,right:0,front:0,back:0,type:"VALID"},p=Math.ceil((t-i+1)/r),h=Math.ceil((n-l+1)/a),f=Math.ceil((s-c+1)/o);else throw Error(`Unknown padding parameter: ${e}`);return{padInfo:d,outDepth:p,outHeight:h,outWidth:f}}function ol(e,t){if(!t)return Math.trunc(e);switch(t){case"round":return Math.round(e);case"ceil":return Math.ceil(e);case"floor":return Math.floor(e);default:throw new Error(`Unknown roundingMode ${t}`)}}function To(e){let[t,n,s]=af(e);return t===1&&n===1&&s===1}function Nr(e,t){return To(e)||To(t)}function U3(e){if(e==="NHWC")return"channelsLast";if(e==="NCHW")return"channelsFirst";throw new Error(`Unknown dataFormat ${e}`)}function i$(e,t){let s={x:D(e,"x","reshape","string_or_numeric")},r={shape:t};return W.runKernel(Fi,s,r)}var G=V({reshape_:i$});function l$(e,t,n,s,r){let a=D(e,"x","avgPool","float32"),o=1;O(Nr(n,o),()=>`Error in avgPool: Either strides or dilations must be 1. Got strides ${n} and dilations '${o}'`);let i=a,l=!1;a.rank===3&&(l=!0,i=G(a,[1,a.shape[0],a.shape[1],a.shape[2]])),O(i.rank===4,()=>`Error in avgPool: x must be rank 4 but got rank ${i.rank}.`),r!=null&&O(pn(s),()=>`Error in avgPool: pad must be an integer when using, dimRoundingMode ${r} but got pad ${s}.`);let c={x:i},u={filterSize:t,strides:n,pad:s,dimRoundingMode:r},d=W.runKernel(Na,c,u);return d=de(d,a.dtype),l?G(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var of=V({avgPool_:l$});function u$(e,t,n,s,r,a="NDHWC"){let o=D(e,"x","avgPool3d","float32"),i=o,l=!1;o.rank===4&&(l=!0,i=G(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]])),O(i.rank===5,()=>`Error in avgPool3d: x must be rank 5 but got rank ${i.rank}.`),O(a==="NDHWC",()=>`Error in avgPool3d: Only NDHWC is currently supported, but got dataFormat of ${a}`),r!=null&&O(pn(s),()=>`Error in avgPool3d: pad must be an integer when using, dimRoundingMode ${r} but got pad ${s}.`);let c={x:i},u={filterSize:t,strides:n,pad:s,dimRoundingMode:r,dataFormat:a},d=W.runKernel(Yc,c,u);return d=de(d,i.dtype),l?G(d,[d.shape[1],d.shape[2],d.shape[3],d.shape[4]]):d}var U2=V({avgPool3d_:u$});function c$(e,t=0){O(e.length>=1,()=>"Pass at least one tensor to concat");let n=vd(e,"tensors","concat","string_or_numeric");if(n[0].dtype==="complex64"&&n.forEach(a=>{if(a.dtype!=="complex64")throw new Error(`Cannot concatenate complex64 tensors with a tensor
|
|
with dtype ${a.dtype}. `)}),n.length===1)return Ln(n[0]);let s=n,r={axis:t};return W.runKernel(pi,s,r)}var vt=V({concat_:c$});function d$(e){let n={x:D(e,"x","sigmoid","float32")};return W.runKernel(lo,n)}var cs=V({sigmoid_:d$});function p$(e,t,n){let s=D(e,"x","slice","string_or_numeric");if(s.rank===0)throw new Error("Slicing scalar is not possible");let r={x:s},a={begin:t,size:n};return W.runKernel(Bi,r,a)}var De=V({slice_:p$});function h$(e){let n={x:D(e,"x","tanh","float32")};return W.runKernel(mo,n)}var _u=V({tanh_:h$});function f$(e,t,n,s,r,a){let o=D(e,"forgetBias","basicLSTMCell"),i=D(t,"lstmKernel","basicLSTMCell"),l=D(n,"lstmBias","basicLSTMCell"),c=D(s,"data","basicLSTMCell"),u=D(r,"c","basicLSTMCell"),d=D(a,"h","basicLSTMCell"),p=vt([c,d],1),h=Ue(p,i),f=le(h,l),m=f.shape[0],g=f.shape[1]/4,A=[m,g],x=De(f,[0,0],A),y=De(f,[0,g],A),b=De(f,[0,g*2],A),w=De(f,[0,g*3],A),k=le(L(cs(x),_u(y)),L(u,cs(le(o,b)))),C=L(_u(k),cs(w));return[k,C]}var m$=V({basicLSTMCell_:f$});function g$(e,t,n){let s=D(e,"x","batchToSpaceND"),r=t.reduce((i,l)=>i*l);O(s.rank>=1+t.length,()=>`input rank is ${s.rank} but should be > than blockShape.length ${t.length}`),O(n.length===t.length,()=>`crops.length is ${n.length} but should be equal to blockShape.length ${t.length}`),O(s.shape[0]%r==0,()=>`input tensor batch is ${s.shape[0]} but is not divisible by the product of the elements of blockShape ${t.join(" * ")} === ${r}`);let a={x:s},o={blockShape:t,crops:n};return W.runKernel(di,a,o)}var lf=V({batchToSpaceND_:g$});function A$(e){let t;return e.rank===0||e.rank===1?t=G(e,[1,1,1,e.size]):e.rank===2?t=G(e,[1,1,e.shape[0],e.shape[1]]):e.rank===3?t=G(e,[1,e.shape[0],e.shape[1],e.shape[2]]):t=e,t}function y$(e,t,n,s,r,a){a==null&&(a=.001);let o=D(e,"x","batchNorm"),i=D(t,"mean","batchNorm"),l=D(n,"variance","batchNorm"),c;r!=null&&(c=D(r,"scale","batchNorm"));let u;s!=null&&(u=D(s,"offset","batchNorm")),O(i.rank===l.rank,()=>"Batch normalization gradient requires mean and variance to have equal ranks."),O(u==null||i.rank===u.rank,()=>"Batch normalization gradient requires mean and offset to have equal ranks."),O(c==null||i.rank===c.rank,()=>"Batch normalization gradient requires mean and scale to have equal ranks.");let p={x:A$(o),scale:c,offset:u,mean:i,variance:l},h={varianceEpsilon:a},f=W.runKernel(Va,p,h);return G(f,o.shape)}var Du=V({batchNorm_:y$});function x$(e,t,n,s,r,a){let o=D(e,"x","batchNorm"),i=D(t,"mean","batchNorm"),l=D(n,"variance","batchNorm"),c;r!=null&&(c=D(r,"scale","batchNorm"));let u;return s!=null&&(u=D(s,"offset","batchNorm")),O(o.rank===2,()=>`Error in batchNorm2D: x must be rank 2 but got rank ${o.rank}.`),O(i.rank===2||i.rank===1,()=>`Error in batchNorm2D: mean must be rank 2 or rank 1 but got rank ${i.rank}.`),O(l.rank===2||l.rank===1,()=>`Error in batchNorm2D: variance must be rank 2 or rank 1 but got rank ${l.rank}.`),c!=null&&O(c.rank===2||c.rank===1,()=>`Error in batchNorm2D: scale must be rank 2 or rank 1 but got rank ${c.rank}.`),u!=null&&O(u.rank===2||u.rank===1,()=>`Error in batchNorm2D: offset must be rank 2 or rank 1 but got rank ${u.rank}.`),Du(o,i,l,u,c,a)}var G3=V({batchNorm2d_:x$});function b$(e,t,n,s,r,a){let o=D(e,"x","batchNorm"),i=D(t,"mean","batchNorm"),l=D(n,"variance","batchNorm"),c;r!=null&&(c=D(r,"scale","batchNorm"));let u;return s!=null&&(u=D(s,"offset","batchNorm")),O(o.rank===3,()=>`Error in batchNorm3D: x must be rank 3 but got rank ${o.rank}.`),O(i.rank===3||i.rank===1,()=>`Error in batchNorm3D: mean must be rank 3 or rank 1 but got rank ${i.rank}.`),O(l.rank===3||l.rank===1,()=>`Error in batchNorm3D: variance must be rank 3 or rank 1 but got rank ${l.rank}.`),c!=null&&O(c.rank===3||c.rank===1,()=>`Error in batchNorm3D: scale must be rank 3 or rank 1 but got rank ${c.rank}.`),u!=null&&O(u.rank===3||u.rank===1,()=>`Error in batchNorm3D: offset must be rank 3 or rank 1 but got rank ${u.rank}.`),Du(o,i,l,u,c,a)}var H3=V({batchNorm3d_:b$});function v$(e,t,n,s,r,a){let o=D(e,"x","batchNorm"),i=D(t,"mean","batchNorm"),l=D(n,"variance","batchNorm"),c;r!=null&&(c=D(r,"scale","batchNorm"));let u;return s!=null&&(u=D(s,"offset","batchNorm")),O(o.rank===4,()=>`Error in batchNorm4D: x must be rank 4 but got rank ${o.rank}.`),O(i.rank===4||i.rank===1,()=>`Error in batchNorm4D: mean must be rank 4 or rank 1 but got rank ${i.rank}.`),O(l.rank===4||l.rank===1,()=>`Error in batchNorm4D: variance must be rank 4 or rank 1 but got rank ${l.rank}.`),c!=null&&O(c.rank===4||c.rank===1,()=>`Error in batchNorm4D: scale must be rank 4 or rank 1 but got rank ${c.rank}.`),u!=null&&O(u.rank===4||u.rank===1,()=>`Error in batchNorm4D: offset must be rank 4 or rank 1 but got rank ${u.rank}.`),Du(o,i,l,u,c,a)}var j3=V({batchNorm4d_:v$});function w$(e,t,n){let s=D(e,"x","bincount"),r=D(t,"weights","bincount");O(s.dtype==="int32",()=>`Error in bincount: input dtype must be int32, but got ${s.dtype}`),O(n>=0,()=>`size must be non-negative, but got ${n}.`),O(r.size===s.size||r.size===0,()=>`Error in bincount: weights must have the same size as input or0-length, but got input shape: ${s.shape}, weights shape: ${r.shape}.`);let a={x:s,weights:r},o={size:n};return W.runKernel(Ah,a,o)}var G2=V({bincount_:w$});function k$(e,t){let n=D(e,"s0","broadcastArgs","int32"),s=D(t,"s1","broadcastArgs","int32");if(n.rank!==1)throw new Error(`broadcastArgs(): first input must be a vector (rank=1). Has rank ${n.rank}`);if(s.rank!==1)throw new Error(`broadcastArgs(): second input must be a vector (rank=1). Has rank ${s.rank}`);let r={s0:n,s1:s};return W.runKernel(yh,r)}var q3=V({broadcastArgs_:k$});function S$(e,t){let n=D(e,"broadcastTo","x"),s=n.shape;if(t.some(c=>!(c>0)||c%1!=0))throw new Error(`broadcastTo(): Invalid broadcast shape [${t}].`);if(t.length<n.rank)throw new Error(`broadcastTo(): shape.length=${t.length} < input.rank=${n.rank}.`);if(t.length>n.rank){let c=n.shape.slice();for(;c.length<t.length;)c.unshift(1);n=G(n,c)}let r=n.shape,a=Array.from(t);for(let c=t.length-1;c>=0;c--)if(r[c]===t[c])a[c]=1;else if(n.shape[c]!==1)throw new Error(`broadcastTo(): [${s}] cannot be broadcast to [${t}].`);if(a.map((c,u)=>c>1?u:-1).filter(c=>c>=0).length===0)return Ln(n);let i={x:n},l={reps:a};return W.runKernel(jr,i,l)}var Id=V({broadcastTo_:S$});function I$(e){let n={x:D(e,"x","ceil","float32")};return W.runKernel($a,n)}var X3=V({ceil_:I$});function C$(e,t,n){let s=D(e,"x","clipByValue");O(t<=n,()=>`Error in clip: min (${t}) must be less than or equal to max (${n}).`);let r={x:s},a={clipValueMin:t,clipValueMax:n};return W.runKernel(Hr,r,a)}var ds=V({clipByValue_:C$});function T$(e){return vt(e,0)}var K3=V({concat1d_:T$});function N$(e,t){return vt(e,t)}var Pu=V({concat2d_:N$});function E$(e,t){return vt(e,t)}var Z3=V({concat3d_:E$});function R$(e,t){return vt(e,t)}var Y3=V({concat4d_:R$});function $$(e,t,n,s,r="NHWC",a=[1,1],o){let i=D(e,"x","conv2d","float32"),l=D(t,"filter","conv2d","float32"),c=i,u=!1;i.rank===3&&(u=!0,c=G(i,[1,i.shape[0],i.shape[1],i.shape[2]])),O(c.rank===4,()=>`Error in conv2d: input must be rank 4, but got rank ${c.rank}.`),O(l.rank===4,()=>`Error in conv2d: filter must be rank 4, but got rank ${l.rank}.`),o!=null&&O(pn(s),()=>`Error in conv2d: pad must be an integer when using, dimRoundingMode ${o} but got pad ${s}.`);let d=r==="NHWC"?c.shape[3]:c.shape[1];O(d===l.shape[2],()=>`Error in conv2d: depth of input (${d}) must match input depth for filter ${l.shape[2]}.`),O(Nr(n,a),()=>`Error in conv2D: Either strides or dilations must be 1. Got strides ${n} and dilations '${a}'`);let p={x:c,filter:l},h={strides:n,pad:s,dataFormat:r,dilations:a,dimRoundingMode:o},f=W.runKernel(_a,p,h);return u?G(f,[f.shape[1],f.shape[2],f.shape[3]]):f}var No=V({conv2d_:$$});function _$(e,t,n,s,r="NWC",a=1,o){let i=D(e,"x","conv1d"),l=D(t,"filter","conv1d"),c=i,u=!1;i.rank===2&&(u=!0,c=G(i,[1,i.shape[0],i.shape[1]])),O(c.rank===3,()=>`Error in conv1d: input must be rank 3, but got rank ${c.rank}.`),O(l.rank===3,()=>`Error in conv1d: filter must be rank 3, but got rank ${l.rank}.`),o!=null&&O(pn(s),()=>`Error in conv1d: pad must be an integer when using, dimRoundingMode ${o} but got pad ${s}.`),O(c.shape[2]===l.shape[1],()=>`Error in conv1d: depth of input (${c.shape[2]}) must match input depth for filter ${l.shape[1]}.`),O(Nr(n,a),()=>`Error in conv1D: Either stride or dilation must be 1. Got stride ${n} and dilation '${a}'`),O(r==="NWC",()=>`Error in conv1d: got dataFormat of ${r} but only NWC is currently supported.`);let d=G(l,[1,l.shape[0],l.shape[1],l.shape[2]]),p=G(c,[c.shape[0],1,c.shape[1],c.shape[2]]),g=No(p,d,[1,n],s,"NHWC",[1,a],o);return u?G(g,[g.shape[2],g.shape[3]]):G(g,[g.shape[0],g.shape[2],g.shape[3]])}var H2=V({conv1d_:_$});function D$(e,t,n,s,r,a="NHWC",o){O(e.length===t.rank,()=>`Length of inShape (${e.length}) and rank of dy (${t.rank}) must match`);let i=e,l=t,c=!1;t.rank===3&&(c=!0,l=G(t,[1,t.shape[0],t.shape[1],t.shape[2]]),i=[1,e[0],e[1],e[2]]),O(i.length===4,()=>`Error in conv2dDerInput: inShape must be length 4, but got length ${i.length}.`),O(l.rank===4,()=>`Error in conv2dDerInput: dy must be rank 4, but got rank ${l.rank}`),O(n.rank===4,()=>`Error in conv2dDerInput: filter must be rank 4, but got rank ${n.rank}`);let u=a==="NHWC"?i[3]:i[1],d=a==="NHWC"?l.shape[3]:l.shape[1];O(u===n.shape[2],()=>`Error in conv2dDerInput: depth of input (${u}) must match input depth for filter ${n.shape[2]}.`),O(d===n.shape[3],()=>`Error in conv2dDerInput: depth of output (${d}) must match output depth for filter ${n.shape[3]}.`),o!=null&&O(pn(r),()=>`Error in conv2dDerInput: pad must be an integer when using, dimRoundingMode ${o} but got pad ${r}.`);let p={dy:l,filter:n},h={strides:s,pad:r,dataFormat:a,dimRoundingMode:o,inputShape:i},f=W.runKernel(Da,p,h);return c?G(f,[f.shape[1],f.shape[2],f.shape[3]]):f}var j2=V({conv2DBackpropInput_:D$});function P$(e,t,n,s,r,a){let o=D(e,"x","conv2dTranspose"),i=D(t,"filter","conv2dTranspose");return j2(n,o,i,s,r,"NHWC",a)}var q2=V({conv2dTranspose_:P$});function F$(e,t,n,s,r="NDHWC",a=[1,1,1]){let o=D(e,"x","conv3d"),i=D(t,"filter","conv3d"),l=o,c=!1;o.rank===4&&(c=!0,l=G(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]])),O(l.rank===5,()=>`Error in conv3d: input must be rank 5, but got rank ${l.rank}.`),O(i.rank===5,()=>`Error in conv3d: filter must be rank 5, but got rank ${i.rank}.`),O(l.shape[4]===i.shape[3],()=>`Error in conv3d: depth of input (${l.shape[4]}) must match input depth for filter ${i.shape[3]}.`),O(Nr(n,a),()=>`Error in conv3D: Either strides or dilations must be 1. Got strides ${n} and dilations '${a}'`),O(r==="NDHWC",()=>`Error in conv3d: got dataFormat of ${r} but only NDHWC is currently supported.`);let u={x:l,filter:i},d={strides:n,pad:s,dataFormat:r,dilations:a},p=W.runKernel(ed,u,d);return c?G(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var X2=V({conv3d_:F$});function O$(e,t,n,s,r){O(e.length===t.rank,()=>`Length of inShape (${e.length}) and rank of dy (${t.rank}) must match`);let a=e,o=t,i=!1;t.rank===4&&(i=!0,o=G(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]]),a=[1,e[0],e[1],e[2],e[3]]);let l=a[4],c=o.shape[4];O(a.length===5,()=>`Error in conv3dDerInput: inShape must be length 5, but got length ${a.length}.`),O(o.rank===5,()=>`Error in conv3dDerInput: dy must be rank 5, but got rank ${o.rank}`),O(n.rank===5,()=>`Error in conv3dDerInput: filter must be rank 5, but got rank ${n.rank}`),O(l===n.shape[3],()=>`Error in conv3dDerInput: depth of input (${l}) must match input depth for filter ${n.shape[3]}.`),O(c===n.shape[4],()=>`Error in conv3dDerInput: depth of output (${c}) must match output depth for filter ${n.shape[4]}.`);let u={dy:o,filter:n},d={pad:r,strides:s,inputShape:a},p=W.runKernel(vh,u,d);return i?G(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var J3=V({conv3DBackpropInput_:O$});function M$(e,t,n,s,r){let a=D(e,"x","conv3dTranspose"),o=D(t,"filter","conv3dTranspose");return J3(n,a,o,s,r)}var Q3=V({conv3dTranspose_:M$});function z$(e){let n={x:D(e,"x","cos","float32")};return W.runKernel(Pa,n)}var uf=V({cos_:z$});function L$(e){let n={x:D(e,"x","cosh","float32")};return W.runKernel(Fa,n)}var K2=V({cosh_:L$});function B$(e,t=0,n=!1,s=!1){let a={x:D(e,"x","cumsum")},o={axis:t,exclusive:n,reverse:s};return W.runKernel(hi,a,o)}var Z2=V({cumsum_:B$});function W$(e,t,n,s=!1){let r=D(e,"x","denseBincount"),a=D(t,"weights","denseBincount");O(r.dtype==="int32",()=>`Error in denseBincount: input dtype must be int32, but got ${r.dtype}`),O(r.rank<=2,()=>`Error in denseBincount: input must be at most rank 2, but got rank ${r.rank}.`),O(n>=0,()=>`size must be non-negative, but got ${n}.`),O(a.size===r.size||a.size===0,()=>`Error in denseBincount: weights must have the same shape as x or 0-length, but got x shape: ${r.shape}, weights shape: ${a.shape}.`);let o={x:r,weights:a},i={size:n,binaryOutput:s};return W.runKernel(wh,o,i)}var ev=V({denseBincount_:W$});function V$(e,t,n="NHWC"){let s=D(e,"x","depthToSpace","float32"),r=n==="NHWC"?s.shape[1]:s.shape[2],a=n==="NHWC"?s.shape[2]:s.shape[3],o=n==="NHWC"?s.shape[3]:s.shape[1];O(t>1,()=>`blockSize should be > 1 for depthToSpace, but was: ${t}`),O(r*t>=0,()=>`Negative dimension size caused by overflow when multiplying
|
|
${r} and ${t} for depthToSpace with input shape
|
|
${s.shape}`),O(a*t>=0,()=>`Negative dimension size caused by overflow when multiplying
|
|
${a} and ${t} for depthToSpace with input shape
|
|
${s.shape}`),O(o%(t*t)==0,()=>`Dimension size must be evenly divisible by ${t*t} but is ${o} for depthToSpace with input shape ${s.shape}`);let i={x:s},l={blockSize:t,dataFormat:n};return W.runKernel(mi,i,l)}var tv=V({depthToSpace_:V$});function U$(e,t,n,s,r="NHWC",a=[1,1],o){let i=D(e,"x","depthwiseConv2d","float32"),l=D(t,"filter","depthwiseConv2d","float32"),c=i,u=!1;i.rank===3&&(u=!0,c=G(i,[1,i.shape[0],i.shape[1],i.shape[2]])),O(c.rank===4,()=>`Error in depthwiseConv2d: input must be rank 4, but got rank ${c.rank}.`),O(l.rank===4,()=>`Error in depthwiseConv2d: filter must be rank 4, but got rank ${l.rank}.`),O(c.shape[3]===l.shape[2],()=>`Error in depthwiseConv2d: number of input channels (${c.shape[3]}) must match the inChannels dimension in filter ${l.shape[2]}.`),o!=null&&O(pn(s),()=>`Error in depthwiseConv2d: pad must be an integer when using, dimRoundingMode ${o} but got pad ${s}.`);let d={x:c,filter:l},p={strides:n,pad:s,dataFormat:r,dilations:a,dimRoundingMode:o},h=W.runKernel(Oa,d,p);return u?G(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var Cd=V({depthwiseConv2d_:U$});function G$(e){let n={x:D(e,"x","diag")};return W.runKernel(Ih,n)}var H$=V({diag_:G$});function j$(e,t,n,s,r=[1,1],a="NHWC"){let o=D(e,"x","dilation2d"),i=D(t,"filter","dilation2d");O(o.rank===3||o.rank===4,()=>`Error in dilation2d: input must be rank 3 or 4, but got rank ${o.rank}.`),O(i.rank===3,()=>`Error in dilation2d: filter must be rank 3, but got rank ${i.rank}.`),O(a==="NHWC",()=>`Error in dilation2d: Only NHWC is currently supported, but got dataFormat of ${a}`);let l=o,c=!1;o.rank===3&&(l=G(o,[1,o.shape[0],o.shape[1],o.shape[2]]),c=!0);let u={x:l,filter:i},d={strides:n,pad:s,dilations:r},p=W.runKernel(td,u,d);return c?G(p,[p.shape[1],p.shape[2],p.shape[3]]):p}var nv=V({dilation2d_:j$});function q$(e,t){let n=D(e,"a","equal","string_or_numeric"),s=D(t,"b","equal","string_or_numeric");[n,s]=Dt(n,s),mt(n.shape,s.shape);let r={a:n,b:s};return W.runKernel(gi,r)}var Cs=V({equal_:q$});function X$(e,t,n){let s=D(t,"a","where"),r=D(n,"b","where"),a=D(e,"condition","where","bool"),o=mt(mt(a.shape,s.shape),r.shape),i=Id(a,o),l=Id(s,o),c=Id(r,o),u={condition:i,t:l,e:c};return W.runKernel(Li,u)}var Bn=V({where_:X$});function K$(e){let n={x:D(e,"x","zerosLike")};return W.runKernel(Ki,n)}var Je=V({zerosLike_:K$});function Z$(e,t){let n=D(e,"a","div"),s=D(t,"b","div");[n,s]=Dt(n,s);let r=pe(n,s),a=Je(r),o=Cs(s,a);return Bn(o,a,r)}var sv=V({divNoNan_:Z$});function Y$(e,t){let n=D(e,"t1","dot"),s=D(t,"t2","dot");O((n.rank===1||n.rank===2)&&(s.rank===1||s.rank===2),()=>`Error in dot: inputs must all be rank 1 or 2, but got ranks ${n.rank} and ${s.rank}.`);let r=n.rank===1?n.size:n.shape[1],a=s.rank===1?s.size:s.shape[0];if(O(r===a,()=>`Error in dot: inner dimensions of inputs must match, but got ${r} and ${a}.`),n.rank===1&&s.rank===1){let o=G(n,[1,-1]),i=G(s,[-1,1]),l=Ue(o,i);return G(l,[])}else if(n.rank===1&&s.rank===2){let o=G(n,[1,-1]),i=G(s,[s.shape[0],s.shape[1]]),l=Ue(o,i);return G(l,[l.size])}else if(n.rank===2&&s.rank===1){let o=G(s,[-1,1]),i=Ue(n,o);return G(i,[i.size])}else{let o=G(s,[s.shape[0],s.shape[1]]);return Ue(n,o)}}var J$=V({dot_:Y$});function Q$(e,...t){let n=t.map((r,a)=>D(r,`tensors${a}`,"einsum")),s={equation:e};return W.runKernel(nd,n,s)}var rv=V({einsum_:Q$});function e_(e){let n={x:D(e,"x","elu","float32")};return W.runKernel(za,n)}var Td=V({elu_:e_});function t_(e){let t=D(e,"x","erf");O(t.dtype==="int32"||t.dtype==="float32",()=>"Input dtype must be `int32` or `float32`."),t.dtype==="int32"&&(t=de(t,"float32"));let n={x:t};return W.runKernel(cu,n)}var av=V({erf_:t_});function n_(e){let n={x:D(e,"x","exp")};return W.runKernel(La,n)}var Ts=V({exp_:n_});function s_(e,t=0){let n=D(e,"x","expandDims","string_or_numeric");O(t<=n.rank,()=>"Axis must be <= rank of the tensor");let s={input:n},r={dim:t};return W.runKernel(Ai,s,r)}var mn=V({expandDims_:s_});function r_(e){let n={x:D(e,"x","expm1")};return W.runKernel(yi,n)}var ov=V({expm1_:r_});function a_(e,t){let n=D(e,"x","tile","string_or_numeric");O(n.rank===t.length,()=>`Error in transpose: rank of input ${n.rank} must match length of reps ${t}.`);let s={x:n},r={reps:t};return W.runKernel(jr,s,r)}var js=V({tile_:a_});function o_(e,t,n,s="float32"){t==null&&(t=e);let r=ze([e,t],s),a=e<=t?e:t;for(let i=0;i<a;++i)r.set(1,i,i);let o=G(r.toTensor(),[e,t]);if(n==null)return o;if(n.length===1)return js(mn(o,0),[n[0],1,1]);if(n.length===2)return js(mn(mn(o,0),0),[n[0],n[1],1,1]);if(n.length===3)return js(mn(mn(mn(o,0),0),0),[n[0],n[1],n[2],1,1]);throw new Error(`eye() currently supports only 1D and 2D batchShapes, but received ${n.length}D.`)}var Y2=V({eye_:o_});function Fu(e,t,n){let s={shape:e,value:t,dtype:n};return W.runKernel(du,{},s)}function i_(e){let n={x:D(e,"x","floor","float32")};return W.runKernel(Ba,n)}var Nd=V({floor_:i_});function l_(e,t,n=0,s=0){let r=D(e,"x","gather"),a=D(t,"indices","gather","int32"),o={x:r,indices:a},i={axis:n,batchDims:s};return W.runKernel(bi,o,i)}var Ou=V({gather_:l_});function u_(e,t){let n=D(e,"a","greater","string_or_numeric"),s=D(t,"b","greater","string_or_numeric");[n,s]=Dt(n,s),mt(n.shape,s.shape);let r={a:n,b:s};return W.runKernel(wi,r)}var ps=V({greater_:u_});function c_(e,t){let n=D(e,"a","greaterEqual","string_or_numeric"),s=D(t,"b","greaterEqual","string_or_numeric");[n,s]=Dt(n,s),mt(n.shape,s.shape);let r={a:n,b:s};return W.runKernel(Ua,r)}var il=V({greaterEqual_:c_});function d_(e){let n={input:D(e,"input","imag")};return W.runKernel(sd,n)}var cf=V({imag_:d_});function p_(e){let n={x:D(e,"x","isFinite")};return W.runKernel(pu,n)}var h_=V({isFinite_:p_});function f_(e){let n={x:D(e,"x","isInf")};return W.runKernel(hu,n)}var m_=V({isInf_:f_});function g_(e){let n={x:D(e,"x","isNaN")};return W.runKernel(fu,n)}var iv=V({isNaN_:g_});function A_(e,t=.2){let s={x:D(e,"x","leakyRelu")},r={alpha:t};return W.runKernel(ki,s,r)}var df=V({leakyRelu_:A_});function y_(e,t){let n=D(e,"a","less","string_or_numeric"),s=D(t,"b","less","string_or_numeric");[n,s]=Dt(n,s),mt(n.shape,s.shape);let r={a:n,b:s};return W.runKernel(Si,r)}var J2=V({less_:y_});function x_(e,t){let n=D(e,"a","lessEqual","string_or_numeric"),s=D(t,"b","lessEqual","string_or_numeric");[n,s]=Dt(n,s),mt(n.shape,s.shape);let r={a:n,b:s};return W.runKernel(Ii,r)}var ll=V({lessEqual_:x_});function lv(e,t,n){if(n<=0)throw new Error("The number of values should be positive.");let s={start:e,stop:t,num:n};return W.runKernel($h,{},s)}function b_(e,t=5,n=1,s=1,r=.5){let a=D(e,"x","localResponseNormalization");O(a.rank===4||a.rank===3,()=>`Error in localResponseNormalization: x must be rank 3 or 4 but got
|
|
rank ${a.rank}.`),O(pn(t),()=>`Error in localResponseNormalization: depthRadius must be an integer but got depthRadius ${t}.`);let o=a,i=!1;a.rank===3&&(i=!0,o=G(a,[1,a.shape[0],a.shape[1],a.shape[2]]));let l={x:o},c={depthRadius:t,bias:n,alpha:s,beta:r},u=W.runKernel(ad,l,c);return i?G(u,[u.shape[1],u.shape[2],u.shape[3]]):u}var uv=V({localResponseNormalization_:b_});function v_(e){let n={x:D(e,"x","log","float32")};return W.runKernel(Ha,n)}var Ns=V({log_:v_});function w_(e){let n={x:D(e,"x","log1p")};return W.runKernel(mu,n)}var pf=V({log1p_:w_});function k_(e){return O(Ia(e),()=>"The f passed in grad(f) must be a function"),(t,n)=>{let s=D(t,"x","tf.grad","string_or_numeric"),r=n!=null?D(n,"dy","tf.grad"):null;return W.tidy(()=>{let{value:a,grads:o}=W.gradients(()=>e(s),[s],r);return r!=null&&On(a.shape,r.shape,"The shape of dy passed in grad(f)(x, dy) must match the shape returned by f(x)"),hf(o),o[0]})}}function S_(e){return O(Ia(e),()=>"The f passed in grads(f) must be a function"),(t,n)=>{O(Array.isArray(t),()=>"The args passed in grads(f)(args) must be an array of `Tensor`s or `TensorLike`s");let s=vd(t,"args","tf.grads","string_or_numeric"),r=n!=null?D(n,"dy","tf.grads"):null;return W.tidy(()=>{let{value:a,grads:o}=W.gradients(()=>e(...s),s,r);return r!=null&&On(a.shape,r.shape,"The shape of dy passed in grads(f)([x1,...], dy) must match the shape returned by f([x1,...])"),hf(o),o})}}function I_(e){return O(Ia(e),()=>"The f passed in valueAndGrad(f) must be a function"),(t,n)=>{O(t instanceof Ze,()=>"The x passed in valueAndGrad(f)(x) must be a tensor"),O(n==null||n instanceof Ze,()=>"The dy passed in valueAndGrad(f)(x, dy) must be a tensor");let{grads:s,value:r}=W.gradients(()=>e(t),[t],n);return hf(s),{grad:s[0],value:r}}}function C_(e){return O(Ia(e),()=>"The f passed in valueAndGrads(f) must be a function"),(t,n)=>{O(Array.isArray(t)&&t.every(r=>r instanceof Ze),()=>"The args passed in valueAndGrads(f)(args) must be array of tensors"),O(n==null||n instanceof Ze,()=>"The dy passed in valueAndGrads(f)(args, dy) must be a tensor");let s=W.gradients(()=>e(...t),t,n);return n!=null&&On(s.value.shape,n.shape,"The shape of dy passed in valueAndGrads(f)([x1,...], dy) must match the shape returned by f([x1,...])"),hf(s.grads),s}}function cv(e,t){O(Ia(e),()=>"The f passed in variableGrads(f) must be a function"),O(t==null||Array.isArray(t)&&t.every(c=>c instanceof xd),()=>"The varList passed in variableGrads(f, varList) must be an array of variables");let n=t!=null;if(!n){t=[];for(let c in W.registeredVariables)t.push(W.registeredVariables[c])}let s=n?t.filter(c=>!c.trainable):null,r=t.length;t=t.filter(c=>c.trainable),O(t.length>0,()=>`variableGrads() expects at least one of the input variables to be trainable, but none of the ${r} variables is trainable.`);let a=!0,{value:o,grads:i}=W.gradients(e,t,null,a);O(i.some(c=>c!=null),()=>"Cannot find a connection between any variable and the result of the loss function y=f(x). Please make sure the operations that use variables are inside the function f passed to minimize()."),O(o.rank===0,()=>`The f passed in variableGrads(f) must return a scalar, but it returned a rank-${o.rank} tensor`);let l={};return t.forEach((c,u)=>{i[u]!=null&&(l[c.name]=i[u])}),s!=null&&s.forEach(c=>l[c.name]=null),{value:o,grads:l}}function Er(e){return W.customGrad(e)}function hf(e){if(e.filter(n=>n==null).length>0)throw new Error(`Cannot compute gradient of y=f(x) with respect to x. Make sure that
|
|
the f you passed encloses all operations that lead from x to y.`)}function T_(e){let n={x:D(e,"x","neg")};return W.runKernel(Ti,n)}var Ft=V({neg_:T_});function N_(e){let n={x:D(e,"x","softplus")};return W.runKernel(Su,n)}var Mu=V({softplus_:N_});function E_(e){let t=D(e,"x","logSigmoid");return Er(s=>({value:Ft(Mu(Ft(s))),gradFunc:o=>L(o,cs(Ft(s)))}))(t)}var R_=V({logSigmoid_:E_});function $_(e,t=null,n=!1){let r={x:D(e,"x","max")},a={reductionIndices:t,keepDims:n};return W.runKernel(ja,r,a)}var gn=V({max_:$_});function __(e,t){let n=D(e,"a","sub"),s=D(t,"b","sub");[n,s]=Dt(n,s);let r={a:n,b:s};return W.runKernel(fo,r)}var me=V({sub_:__});function D_(e,t=null,n=!1){let s=D(e,"x","sum");s.dtype==="bool"&&(s=de(s,"int32"));let r={x:s},a={axis:t,keepDims:n};return W.runKernel(co,r,a)}var Se=V({sum_:D_});function P_(e,t=-1){let n=D(e,"logits","logSoftmax");if(t===-1&&(t=n.rank-1),t!==n.rank-1)throw Error(`Log Softmax along a non-last dimension is not yet supported. Logits was rank ${n.rank} and axis was ${t}`);return Er((r,a)=>{let o=!0,i=gn(r,t,!0),l=me(r,i),c=me(de(l,"float32"),Ns(Se(Ts(l),t,o)));return a([c]),{value:c,gradFunc:(d,p)=>{let[h]=p,f=!0,m=Ts(h);return me(d,L(Se(d,t,f),m))}}})(n)}var Q2=V({logSoftmax_:P_});function e1(e,t){for(let n=0;n<e.length;++n)if(e[e.length-n-1]!==t-1-n)return!1;return!0}function dv(e,t,n){let s=e.length+t.length,r=[],a=0,o=0;for(let i=0;i<s;i++)n.indexOf(i)===-1?r.push(e[a++]):r.push(t[o++]);return r}function pv(e,t){let n=[],s=e.length;for(let a=0;a<s;a++)t.indexOf(a)===-1&&n.push(e[a]);let r=t.map(a=>e[a]);return[n,r]}function ul(e,t){let n=t.map(s=>1);return dv(e,n,t)}function F_(e,t,n){O(e1(t,n),()=>`${e} supports only inner-most axes for now. Got axes ${t} and rank-${n} input.`)}function hv(e,t){if(e1(e,t))return null;let n=[];for(let s=0;s<t;++s)e.indexOf(s)===-1&&n.push(s);return e.forEach(s=>n.push(s)),n}function t1(e){return e.map((t,n)=>[n,t]).sort((t,n)=>t[1]-n[1]).map(t=>t[0])}function O_(e,t){let n=[];for(let s=t-e;s<t;++s)n.push(s);return n}function M_(e,t=null,n=!1){let s=D(e,"x","logSumExp"),r=Us(t,s.shape),a=gn(s,r,!0),o=me(s,a),i=Ts(o),l=Se(i,r),c=Ns(l),u=le(G(a,c.shape),c);if(n){let d=ul(u.shape,r);return G(u,d)}return u}var fv=V({logSumExp_:M_});function z_(e,t){let n=D(e,"a","logicalAnd","bool"),s=D(t,"b","logicalAnd","bool");mt(n.shape,s.shape);let r={a:n,b:s};return W.runKernel(Ci,r)}var ir=V({logicalAnd_:z_});function L_(e){let n={x:D(e,"x","logicalNot","bool")};return W.runKernel(gu,n)}var ff=V({logicalNot_:L_});function B_(e,t){let n=D(e,"a","logicalOr","bool"),s=D(t,"b","logicalOr","bool");mt(n.shape,s.shape);let r={a:n,b:s};return W.runKernel(rd,r)}var n1=V({logicalOr_:B_});function W_(e,t){let n=D(e,"a","logicalXor","bool"),s=D(t,"b","logicalXor","bool");return mt(n.shape,s.shape),ir(n1(e,t),ff(ir(e,t)))}var V_=V({logicalXor_:W_});function U_(e,t,n,s,r){let a=D(e,"x","maxPool"),o=1,i=a,l=!1;a.rank===3&&(l=!0,i=G(a,[1,a.shape[0],a.shape[1],a.shape[2]])),O(i.rank===4,()=>`Error in maxPool: input must be rank 4 but got rank ${i.rank}.`),O(Nr(n,o),()=>`Error in maxPool: Either strides or dilations must be 1. Got strides ${n} and dilations '${o}'`),r!=null&&O(pn(s),()=>`Error in maxPool: pad must be an integer when using, dimRoundingMode ${r} but got pad ${s}.`);let c={x:i},u={filterSize:t,strides:n,pad:s,dimRoundingMode:r},d=W.runKernel(Xa,c,u);return l?G(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var mf=V({maxPool_:U_});function G_(e,t=[1,1,1],n,s,r,a="NDHWC"){let o=D(e,"x","maxPool3d"),i=o,l=!1;o.rank===4&&(l=!0,i=G(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]])),O(i.rank===5,()=>`Error in maxPool3d: x must be rank 5 but got rank ${i.rank}.`),O(a==="NDHWC",()=>`Error in maxPool3d: Only NDHWC is currently supported, but got dataFormat of ${a}`),r!=null&&O(pn(s),()=>`Error in maxPool3d: pad must be an integer when using, dimRoundingMode ${r} but got pad ${s}.`);let c={x:i},u={filterSize:t,strides:n,pad:s,dimRoundingMode:r,dataFormat:a},d=W.runKernel(od,c,u);return l?G(d,[d.shape[1],d.shape[2],d.shape[3],d.shape[4]]):d}var s1=V({maxPool3d_:G_});function H_(e,t,n,s,r=!1){let o={x:D(e,"x","maxPoolWithArgmax")},i={filterSize:t,strides:n,pad:s,includeBatchInIndex:r},l=W.runKernel(Fh,o,i);return{result:l[0],indexes:l[1]}}var mv=V({maxPoolWithArgmax_:H_});function j_(e,t){let n=D(e,"a","maximum"),s=D(t,"b","maximum");[n,s]=Dt(n,s),n.dtype==="bool"&&(n=de(n,"int32"),s=de(s,"int32")),mt(n.shape,s.shape);let r={a:n,b:s};return W.runKernel(qa,r)}var Kr=V({maximum_:j_});function q_(e,t=null,n=!1){let r={x:D(e,"x","mean")},a={axis:t,keepDims:n};return W.runKernel(Ka,r,a)}var Bt=V({mean_:q_});function Gt(e,t="float32"){if(t==="complex64"){let s=Gt(e,"float32"),r=Gt(e,"float32");return ko(s,r)}let n=fh(Vt(e),t);return W.makeTensor(n,e,t)}function hs(e,t="float32"){if(t==="complex64"){let s=hs(e,"float32"),r=Gt(e,"float32");return ko(s,r)}let n=Jg(Vt(e),t);return W.makeTensor(n,e,t)}function X_(e,t,{indexing:n="xy"}={}){if(n!=="xy"&&n!=="ij")throw new TypeError(`${n} is not a valid third argument to meshgrid`);if(e===void 0)return[];let s=D(e,"x","meshgrid",e instanceof Ze?e.dtype:"float32");if(t===void 0)return[s];let r=D(t,"y","meshgrid",t instanceof Ze?t.dtype:"float32"),a=Vt(s.shape),o=Vt(r.shape);return n==="xy"?(s=G(s,[1,-1]),r=G(r,[-1,1]),[Ue(hs([o,1],s.dtype),s),Ue(r,hs([1,a],r.dtype))]):(s=G(s,[-1,1]),r=G(r,[1,-1]),[Ue(s,hs([1,o],s.dtype)),Ue(hs([a,1],r.dtype),r)])}function K_(e,t=null,n=!1){let r={x:D(e,"x","min")},a={axis:t,keepDims:n};return W.runKernel(Za,r,a)}var Eo=V({min_:K_});function Z_(e,t){let n=D(e,"a","minimum"),s=D(t,"b","minimum");[n,s]=Dt(n,s),n.dtype==="bool"&&(n=de(n,"int32"),s=de(s,"int32")),mt(n.shape,s.shape);let r={a:n,b:s};return W.runKernel(Ya,r)}var Ed=V({minimum_:Z_});function Y_(e,t,n){O(n==="reflect"||n==="symmetric",()=>`Invalid mode. Mode must be either reflect or symmetric. Got ${n}.`);let s=D(e,"x","mirrorPad");if(s.rank===0)throw new Error("mirrorPad(scalar) is not defined. Pass non-scalar to mirrorPad");O(t.length===s.rank,()=>`Padding doesn't match input. Must be ${s.rank}. Got ${t.length}.`);let r=n==="reflect"?1:0;for(let i=0;i<s.rank;i++)O(t[i].length===2,()=>"Invalid number of paddings. Must be length of 2 each."),O(t[i][0]>=0&&t[i][0]<=s.shape[i]-r&&t[i][1]>=0&&t[i][1]<=s.shape[i]-r,()=>`Padding in dimension ${i} cannot be greater than or equal to ${s.shape[i]-r} or less than 0 for input of shape ${s.shape}`);let a={paddings:t,mode:n},o={x:s};return W.runKernel(Ja,o,a)}var gv=V({mirrorPad_:Y_});function J_(e,t){let n=D(e,"a","mod"),s=D(t,"b","mod");[n,s]=Dt(n,s);let r={a:n,b:s};return W.runKernel(Au,r)}var Rd=V({mod_:J_});function Q_(e){let t=D(e,"x","square"),n={};return W.runKernel("Square",{x:t},n)}var gt=V({square_:Q_});function eD(e,t=null,n=!1){e=D(e,"x","moments");let s=Us(t,e.shape),r=Bt(e,s,n),a=r.shape;n||(a=ul(r.shape,s));let o=gt(me(de(e,"float32"),G(r,a))),i=Bt(o,s,n);return{mean:r,variance:i}}var gf=V({moments_:eD});function tD(e,t,n,s){let r=D(t,"data","multiRNNCell"),a=vd(n,"c","multiRNNCell"),o=vd(s,"h","multiRNNCell"),i=r,l=[];for(let d=0;d<e.length;d++){let p=e[d](i,a[d],o[d]);l.push(p[0]),l.push(p[1]),i=p[1]}let c=[],u=[];for(let d=0;d<l.length;d+=2)c.push(l[d]),u.push(l[d+1]);return[c,u]}var nD=V({multiRNNCell_:tD});function sD(e,t,n,s=!1){let r=D(e,"logits","multinomial"),a=r.size,o=r.rank;if(a<2)throw new Error(`Error in multinomial: you need at least 2 outcomes, but got ${a}.`);if(o>2)throw new Error(`Rank of probabilities must be 1 or 2, but is ${o}`);n=n||Math.random();let l={logits:o===1?G(r,[1,-1]):r},c={numSamples:t,seed:n,normalized:s},u=W.runKernel(Oh,l,c);return o===1?G(u,[u.size]):u}var Av=V({multinomial_:sD});function rD(e,t){let n=D(e,"a","notEqual","string_or_numeric"),s=D(t,"b","notEqual","string_or_numeric");[n,s]=Dt(n,s),mt(n.shape,s.shape);let r={a:n,b:s};return W.runKernel(Ni,r)}var zu=V({notEqual_:rD});function aD(e){let n={x:D(e,"x","onesLike")};return W.runKernel($i,n)}var Es=V({onesLike_:aD});function oD(e,t){let n=D(e,"v1","outerProduct"),s=D(t,"v2","outerProduct");O(n.rank===1&&s.rank===1,()=>`Error in outerProduct: inputs must be rank 1, but got ranks ${n.rank} and ${s.rank}.`);let r=G(n,[-1,1]),a=G(s,[1,-1]);return Ue(r,a)}var iD=V({outerProduct_:oD});function lD(e,t,n=0){let s=D(e,"x","pad");if(s.rank===0)throw new Error("pad(scalar) is not defined. Pass non-scalar to pad");let r={paddings:t,constantValue:n},a={x:s};return W.runKernel(eo,a,r)}var qs=V({pad_:lD});function uD(e,t,n=0){return O(t.length===2,()=>"Invalid number of paddings. Must be length of 2."),qs(e,[t],n)}var cD=V({pad1d_:uD});function dD(e,t,n=0){return O(t.length===2&&t[0].length===2&&t[1].length===2,()=>"Invalid number of paddings. Must be length of 2 each."),qs(e,t,n)}var pD=V({pad2d_:dD});function hD(e,t,n=0){return O(t.length===3&&t[0].length===2&&t[1].length===2&&t[2].length===2,()=>"Invalid number of paddings. Must be length of 2 each."),qs(e,t,n)}var fD=V({pad3d_:hD});function mD(e,t,n=0){return O(t.length===4&&t[0].length===2&&t[1].length===2&&t[2].length===2&&t[3].length===2,()=>"Invalid number of paddings. Must be length of 2 each."),qs(e,t,n)}var gD=V({pad4d_:mD});function AD(e,t,n){let s=D(e,"x","spaceToBatchND");O(s.rank>=1+t.length,()=>`input rank ${s.rank} should be > than [blockShape] ${t.length}`),O(n.length===t.length,()=>`paddings.shape[0] ${n.length} must be equal to [blockShape] ${t.length}`),O(s.shape.reduce((o,i,l)=>l>0&&l<=t.length?o&&(i+n[l-1][0]+n[l-1][1])%t[l-1]==0:o,!0),()=>`input spatial dimensions ${s.shape.slice(1)} with paddings ${n.toString()} must be divisible by blockShapes ${t.toString()}`);let r={x:s},a={blockShape:t,paddings:n};return W.runKernel(Vi,r,a)}var Af=V({spaceToBatchND_:AD});function yD(e,t,n,s,r,a){r==null&&(r=[1,1]),a==null&&(a=1),s===0&&(s="valid");let o=D(e,"x","maxPool"),i=o,l=!1;o.rank===3&&(l=!0,i=G(o,[1,o.shape[0],o.shape[1],o.shape[2]])),O(Nr(a,r),()=>`Error in pool: Either strides or dilations must be 1. Got strides ${a} and dilations '${r}'`);let c=W3(i.shape,t,a,r,s),u=[c.dilationHeight,c.dilationWidth],d;s==="same"?d=bD([c.filterHeight,c.filterWidth],u):d=[[0,0],[0,0]];let p=u[0]===1&&u[1]===1,[h,f]=xD([c.inHeight,c.inWidth],u,d),m=p?s:"valid",g=p?i:Af(i,u,h),x=(n==="avg"?()=>of(g,t,a,m):()=>mf(g,t,a,m))(),y=p?x:lf(x,u,f);return l?G(y,[y.shape[1],y.shape[2],y.shape[3]]):y}function xD(e,t,n){let s=n.map(u=>u[0]),r=n.map(u=>u[1]),a=e.concat(s,r),o=t.map((u,d)=>(u-a[d]%u)%u),i=r.map((u,d)=>u+o[d]),l=t.map((u,d)=>[s[d],i[d]]),c=t.map((u,d)=>[0,o[d]]);return[l,c]}function bD(e,t){let s=e.map((o,i)=>o+(o-1)*(t[i]-1)).map(o=>o-1),r=s.map(o=>Math.floor(o/2)),a=s.map((o,i)=>o-r[i]);return s.map((o,i)=>[r[i],a[i]])}var vD=V({pool_:yD});function wD(e,t){let n=D(e,"base","pow"),s=D(t,"exp","pow");[n,s]=Dt(n,s);let r={a:n,b:s};return W.runKernel(to,r)}var Ro=V({pow_:wD});function kD(e,t){let n=D(e,"x","prelu"),s=D(t,"alpha","prelu"),r={x:n,alpha:s};return W.runKernel(no,r)}var yf=V({prelu_:kD});function SD(e,t=null,n=!1){let s=D(e,"x","prod");s.dtype==="bool"&&(s=de(s,"int32"));let r={x:s},a={axis:t,keepDims:n};return W.runKernel(Pi,r,a)}var r1=V({prod_:SD});function ID(e,t,n){let s=Vt(e),r=null;if(n==null||n==="float32")r=new Float32Array(s);else if(n==="int32")r=new Int32Array(s);else if(n==="bool")r=new Uint8Array(s);else throw new Error(`Unknown data type ${n}`);for(let a=0;a<s;a++)r[a]=t();return W.makeTensor(r,e,n)}var CD=V({rand_:ID}),a1=ii(ch()),o1=class{constructor(e,t,n,s,r){this.mean=e,this.stdDev=t,this.dtype=n,this.nextVal=NaN,this.truncated=s,this.truncated&&(this.upper=this.mean+this.stdDev*2,this.lower=this.mean-this.stdDev*2);let a=r||Math.random();this.random=a1.alea(a.toString())}nextValue(){if(!isNaN(this.nextVal)){let s=this.nextVal;return this.nextVal=NaN,s}let e,t,n=!1;for(;!n;){let s,r,a;do s=2*this.random()-1,r=2*this.random()-1,a=s*s+r*r;while(a>=1||a===0);let o=Math.sqrt(-2*Math.log(a)/a);e=this.mean+this.stdDev*s*o,t=this.mean+this.stdDev*r*o,(!this.truncated||this.isValidTruncated(e))&&(n=!0)}return(!this.truncated||this.isValidTruncated(t))&&(this.nextVal=this.convertValue(t)),this.convertValue(e)}convertValue(e){return this.dtype==null||this.dtype==="float32"?e:Math.round(e)}isValidTruncated(e){return e<=this.upper&&e>=this.lower}},TD=class{constructor(e,t,n,s){this.alpha=e,this.beta=1/t,this.dtype=n;let r=s||Math.random();this.randu=a1.alea(r.toString()),this.randn=new o1(0,1,n,!1,this.randu()),e<1?this.d=e+2/3:this.d=e-1/3,this.c=1/Math.sqrt(9*this.d)}nextValue(){let e,t,n,s,r,a;for(;;){do s=this.randn.nextValue(),a=1+this.c*s;while(a<=0);if(a*=a*a,e=s*s,t=1-.331*e*e,n=.5*e+this.d*(1-a+Math.log(a)),r=this.randu(),r<t||Math.log(r)<n)break}return a=1/this.beta*this.d*a,this.alpha<1&&(a*=Math.pow(this.randu(),1/this.alpha)),this.convertValue(a)}convertValue(e){return this.dtype==="float32"?e:Math.round(e)}},ND=class{constructor(e=0,t=1,n,s){if(this.canReturnFloat=()=>this.dtype==null||this.dtype==="float32",this.min=e,this.range=t-e,this.dtype=n,s==null&&(s=Math.random()),typeof s=="number"&&(s=s.toString()),!this.canReturnFloat()&&this.range<=1)throw new Error(`The difference between ${e} - ${t} <= 1 and dtype is not float`);this.random=a1.alea(s)}convertValue(e){return this.canReturnFloat()?e:Math.round(e)}nextValue(){return this.convertValue(this.min+this.range*this.random())}};function ED(e,t,n=1,s="float32",r){if(n==null&&(n=1),s==null&&(s="float32"),s!=="float32"&&s!=="int32")throw new Error(`Unsupported data type ${s}`);let a=new TD(t,n,s,r),o=ze(e,s);for(let i=0;i<o.values.length;i++)o.values[i]=a.nextValue();return o.toTensor()}var RD=V({randomGamma_:ED});function $D(e,t=0,n=1,s,r){if(s!=null&&s==="bool")throw new Error(`Unsupported data type ${s}`);let a=new o1(t,n,s,!1,r),o=ze(e,s);for(let i=0;i<o.values.length;i++)o.values[i]=a.nextValue();return o.toTensor()}var yv=V({randomNormal_:$D});function _D(e,t=0,n=1,s="float32",r){let a=ze(e,s),o=new ND(t,n,null,r);for(let i=0;i<a.values.length;i++)a.values[i]=o.nextValue();return a.toTensor()}var Lu=V({randomUniform_:_D});function Bu(e,t,n=1,s="float32"){if(n===0)throw new Error("Cannot have a step of zero");let r={start:e,stop:t,step:n,dtype:s};return W.runKernel(xu,{},r)}function DD(e){let n={input:D(e,"input","real")};return W.runKernel(id,n)}var $d=V({real_:DD});function PD(e){let n={x:D(e,"x","reciprocal")};return W.runKernel(bu,n)}var xv=V({reciprocal_:PD});function FD(e){let n={x:D(e,"x","relu")};return W.runKernel(so,n)}var Rr=V({relu_:FD});function OD(e){let n={x:D(e,"x","relu6")};return W.runKernel(ao,n)}var i1=V({relu6_:OD});function MD(e,t){let s={x:D(e,"x","reverse")},r={dims:t};return W.runKernel(Oi,s,r)}var Rs=V({reverse_:MD});function zD(e){let t=D(e,"x","reverse");return O(t.rank===1,()=>`Error in reverse1D: x must be rank 1 but got rank ${t.rank}.`),Rs(t,0)}var LD=V({reverse1d_:zD});function BD(e,t){let n=D(e,"x","reverse");return O(n.rank===2,()=>`Error in reverse2D: x must be rank 2 but got rank ${n.rank}.`),Rs(n,t)}var WD=V({reverse2d_:BD});function VD(e,t){let n=D(e,"x","reverse");return O(n.rank===3,()=>`Error in reverse3D: x must be rank 3 but got rank ${n.rank}.`),Rs(n,t)}var UD=V({reverse3d_:VD});function GD(e,t){let n=D(e,"x","reverse");return O(n.rank===4,()=>`Error in reverse4D: x must be rank 4 but got rank ${n.rank}.`),Rs(n,t)}var HD=V({reverse4d_:GD});function jD(e){let n={x:D(e,"x","round")};return W.runKernel(Mi,n)}var l1=V({round_:jD});function qD(e){let n={x:D(e,"x","rsqrt","float32")};return W.runKernel(oo,n)}var u1=V({rsqrt_:qD});function Re(e,t){if(($n(e)&&t!=="string"||Array.isArray(e))&&t!=="complex64")throw new Error("Error creating a new Scalar: value must be a primitive (number|boolean|string)");if(t==="string"&&$n(e)&&!(e instanceof Uint8Array))throw new Error("When making a scalar from encoded string, the value must be `Uint8Array`.");return So(e,[],[],t)}function XD(e){let n={x:D(e,"x","selu")};return W.runKernel(wu,n)}var c1=V({selu_:XD});function KD(e,t,n,s,r,a=[1,1],o="NHWC"){let i=D(e,"x","separableConv2d"),l=D(t,"depthwiseFilter","separableConv2d"),c=D(n,"pointwiseFilter","separableConv2d"),u=i,d=!1;if(i.rank===3&&(d=!0,u=G(i,[1,i.shape[0],i.shape[1],i.shape[2]])),o==="NCHW")throw new Error("separableConv2d currently does not support dataFormat NCHW; only NHWC is supported");O(u.rank===4,()=>`Error in separableConv2d: input must be rank 4, but got rank ${u.rank}.`),O(l.rank===4,()=>`Error in separableConv2d: depthwise filter must be rank 4, but got rank ${l.rank}.`),O(c.rank===4,()=>`Error in separableConv2d: pointwise filter must be rank 4, but got rank ${l.rank}.`),O(c.shape[0]===1,()=>`Error in separableConv2d: the first dimension of pointwise filter must be 1, but got ${c.shape[0]}.`),O(c.shape[1]===1,()=>`Error in separableConv2d: the second dimension of pointwise filter must be 1, but got ${c.shape[1]}.`);let p=l.shape[2],h=l.shape[3];O(c.shape[2]===p*h,()=>`Error in separableConv2d: the third dimension of pointwise filter must be ${p*h}, but got ${c.shape[2]}.`);let f=Cd(u,l,s,r,o,a),g=No(f,c,1,"valid",o);return d?G(g,[g.shape[1],g.shape[2],g.shape[3]]):g}var bv=V({separableConv2d_:KD});async function ZD(e,t){let n=D(e,"x","setdiff1d"),s=D(t,"y","setdiff1d");O(n.dtype===s.dtype,()=>`x and y should have the same dtype, but got x (${n.dtype}) and y (${s.dtype}).`),O(n.rank===1,()=>`x should be 1D tensor, but got x (${n.shape}).`),O(s.rank===1,()=>`y should be 1D tensor, but got y (${s.shape}).`);let r=await n.data(),a=await s.data(),o=new Set(a),i=0;for(let u=0;u<r.length;u++)o.has(r[u])||i++;let l=new en([i],n.dtype),c=new en([i],"int32");for(let u=0,d=0;u<r.length;u++)o.has(r[u])||(l.values[d]=r[u],c.values[d]=u,d++);return[l.toTensor(),c.toTensor()]}var vv=ZD;function YD(e){let n={x:D(e,"x","sign")};return W.runKernel(ku,n)}var wv=V({sign_:YD});function JD(e){let n={x:D(e,"x","sin","float32")};return W.runKernel(io,n)}var d1=V({sin_:JD});function QD(e){let n={x:D(e,"x","sinh")};return W.runKernel(Wi,n)}var p1=V({sinh_:QD});function eP(e,t,n){let s=D(e,"x","slice1d");return O(s.rank===1,()=>`slice1d expects a rank-1 tensor, but got a rank-${s.rank} tensor`),De(s,[t],[n])}var xf=V({slice1d_:eP});function tP(e,t,n){let s=D(e,"x","slice2d");return O(s.rank===2,()=>`slice2d expects a rank-2 tensor, but got a rank-${s.rank} tensor`),De(s,t,n)}var h1=V({slice2d_:tP});function nP(e,t,n){let s=D(e,"x","slice3d");return O(s.rank===3,()=>`slice3d expects a rank-3 tensor, but got a rank-${s.rank} tensor`),De(s,t,n)}var Wu=V({slice3d_:nP});function sP(e,t,n){let s=D(e,"x","slice4d");return O(s.rank===4,()=>`slice4d expects a rank-4 tensor, but got a rank-${s.rank} tensor`),De(s,t,n)}var _d=V({slice4d_:sP});function rP(e,t=-1){let n=D(e,"logits","softmax","float32");if(t===-1&&(t=n.rank-1),t!==n.rank-1)throw Error(`Softmax along a non-last dimension is not yet supported. Logits was rank ${n.rank} and dim was ${t}`);let s={logits:n},r={dim:t};return W.runKernel(po,s,r)}var Vu=V({softmax_:rP});function aP(e){O(e.dtype==="complex64",()=>`The dtype for tf.spectral.fft() must be complex64 but got ${e.dtype}.`);let t={input:e};return W.runKernel(Eh,t)}var bf=V({fft_:aP});function oP(e){O(e.dtype==="complex64",()=>`The dtype for tf.spectral.ifft() must be complex64 but got ${e.dtype}.`);let t={input:e};return W.runKernel(Rh,t)}var Dd=V({ifft_:oP});function iP(e){let t=e.shape[e.shape.length-1],n=e.size/t,s;if(t<=2){let r=G(e,[n,t]);s=Dd(r)}else{let r=[n,2*(t-1)],a=G($d(e),[n,t]),o=G(cf(e),[n,t]),i=Rs(De(a,[0,1],[n,t-2]),1),l=L(Rs(De(o,[0,1],[n,t-2]),1),Re(-1)),c=vt([a,i],1),u=vt([o,l],1),d=G(ko(c,u),[r[0],r[1]]);s=Dd(d)}if(s=$d(s),e.rank===3&&e.shape[0]!==0){let r=s,a=e.shape[0];s=G(s,[a,s.shape[0]/a,s.shape[1]]),r.dispose()}return s}var f1=V({irfft_:iP});function lP(e,t,n=0){let r={x:D(e,"x","split")},a={numOrSizeSplits:t,axis:n};return W.runKernel(Ui,r,a)}var nn=V({split_:lP});function uP(e,t){O(e.dtype==="float32",()=>`The dtype for rfft() must be real value but got ${e.dtype}`);let n=e.shape[e.shape.length-1],s=e.size/n,r;if(t!=null&&t<n){let f=e.shape.map(g=>0),m=e.shape.map(g=>g);m[e.shape.length-1]=t,r=De(e,f,m),n=t}else if(t!=null&&t>n){let f=e.shape.map(m=>m);f[e.shape.length-1]=t-n,r=vt([e,Gt(f)],e.shape.length-1),n=t}else r=e;let a=Je(r),o=G(ko(r,a),[s,n]),i=bf(o),l=Math.floor(n/2)+1,c=$d(i),u=cf(i),d=nn(c,[l,n-l],c.shape.length-1),p=nn(u,[l,n-l],u.shape.length-1),h=r.shape.slice();return h[r.shape.length-1]=l,G(ko(d[0],p[0]),h)}var vf=V({rfft_:uP});function cP(e){let n={x:D(e,"x","sqrt","float32")};return W.runKernel(uo,n)}var _n=V({sqrt_:cP});function dP(e,t){let n=D(e,"a","squaredDifference"),s=D(t,"b","squaredDifference");[n,s]=Dt(n,s),mt(n.shape,s.shape);let r={a:n,b:s},a={};return W.runKernel(ho,r,a)}var m1=V({squaredDifference_:dP});function pP(e,t){let n=D(e,"x","squeeze");return G(n,x5(n.shape,t).newShape)}var ot=V({squeeze_:pP});function hP(e,t=0){let n=vd(e,"tensors","stack","string_or_numeric");O(n.length>=1,()=>"Pass at least one tensor to tf.stack"),n.length>0&&O(t<=n[0].rank,()=>"Axis must be <= rank of the tensor");let s=n,r={axis:t};return W.runKernel(Di,s,r)}var An=V({stack_:hP});function fP(e,t=0){let s={x:D(e,"x","step")},r={alpha:t};return W.runKernel(Ao,s,r)}var Pd=V({step_:fP});function mP(e,t,n,s,r=0,a=0,o=0,i=0,l=0){let u={x:D(e,"x","stridedSlice","string_or_numeric")},d={begin:t,end:n,strides:s,beginMask:r,endMask:a,ellipsisMask:o,newAxisMask:i,shrinkAxisMask:l};return W.runKernel(Gi,u,d)}var kv=V({stridedSlice_:mP});function gP(e){let n={x:D(e,"x","tan","float32")};return W.runKernel(Hi,n)}var Sv=V({tan_:gP});function Xt(e,t){li(e);let n=Cr(e,t);if(n.length!==1)throw new Error("tensor1d() requires values to be a flat/TypedArray");return So(e,null,n,t)}function lr(e,t,n){if(li(e),t!=null&&t.length!==2)throw new Error("tensor2d() requires shape to have two numbers");let s=Cr(e,n);if(s.length!==2&&s.length!==1)throw new Error("tensor2d() requires values to be number[][] or flat/TypedArray");if(s.length===1&&t==null)throw new Error("tensor2d() requires shape to be provided when `values` are a flat/TypedArray");return So(e,t,s,n)}function AP(e,t,n){if(li(e),t!=null&&t.length!==4)throw new Error("tensor4d() requires shape to have four numbers");let s=Cr(e,n);if(s.length!==4&&s.length!==1)throw new Error("tensor4d() requires values to be number[][][][] or flat/TypedArray");if(s.length===1&&t==null)throw new Error("tensor4d() requires shape to be provided when `values` are a flat array");return So(e,t,s,n)}function yP(e,t,n){if(li(e),t!=null&&t.length!==5)throw new Error("tensor5d() requires shape to have five numbers");let s=Cr(e,n);if(s.length!==5&&s.length!==1)throw new Error("tensor5d() requires values to be number[][][][][] or flat/TypedArray");if(s.length===1&&t==null)throw new Error("tensor5d() requires shape to be provided when `values` are a flat array");return So(e,t,s,n)}function xP(e,t,n){if(li(e),t!=null&&t.length!==6)throw new Error("tensor6d() requires shape to have six numbers");let s=Cr(e,n);if(s.length!==6&&s.length!==1)throw new Error("tensor6d() requires values to be number[][][][][][] or flat/TypedArray");if(s.length===1&&t==null)throw new Error("tensor6d() requires shape to be provided when `values` are a flat array");return t=t||s,So(e,t,s,n)}function bP(e,t=1,n=!0){let s=D(e,"x","topk");if(s.rank===0)throw new Error("topk() expects the input to be of rank 1 or higher");let r=s.shape[s.shape.length-1];if(t<0)throw new Error(`'k' passed to topk() must be >= 0 but got ${t}`);if(t>r)throw new Error(`'k' passed to topk() must be <= the last dimension (${r}) but got ${t}`);let a={x:s},o={k:t,sorted:n},[i,l]=W.runKernel(ji,a,o);return{values:i,indices:l}}var Iv=V({topk_:bP});function vP(e,t=0,n=1,s,r){if(s!=null&&s==="bool")throw new Error("Unsupported data type $ { dtype }");let a=new o1(t,n,s,!0,r),o=ze(e,s);for(let i=0;i<o.values.length;i++)o.values[i]=a.nextValue();return o.toTensor()}var wf=V({truncatedNormal_:vP});function wP(e,t=0){let n=D(e,"x","unique","string_or_numeric");O(n.rank>0,()=>"The input tensor must be at least 1D");let s={x:n},r={axis:t},[a,o]=W.runKernel(Hh,s,r);return{values:a,indices:o}}var g1=V({unique_:wP});function kP(e,t,n){let s=D(e,"x","unsortedSegmentSum"),r=D(t,"segmentIds","unsortedSegmentSum","int32");O(pn(n),()=>"numSegments must be of dtype int");let a={x:s,segmentIds:r},o={numSegments:n};return W.runKernel(cd,a,o)}var Cv=V({unsortedSegmentSum_:kP});function SP(e,t=0){let n=D(e,"x","unstack","string_or_numeric");O(t>=-n.shape.length&&t<n.shape.length,()=>`Axis = ${t} is not in [-${n.shape.length}, ${n.shape.length})`);let s={value:n},r={axis:t};return W.runKernel(Xi,s,r)}var ss=V({unstack_:SP});function Tv(e,t=!0,n,s){return W.makeVariable(e,t,n,s)}function Nv(e,t){let n=[];for(let a=0;a<t.length;a++)t[a]&&n.push(a);let s=ze(e,"int32"),r=ze([n.length,e.length],"int32");for(let a=0;a<n.length;a++){let o=s.indexToLoc(n[a]),i=a*e.length;r.values.set(o,i)}return r.toTensor()}async function IP(e){let t=D(e,"condition","whereAsync","bool"),n=await t.data(),s=Nv(t.shape,n);return e!==t&&t.dispose(),s}var A1=IP;async function CP(e,t,n){let s=D(e,"tensor","boolMask"),r=D(t,"mask","boolMask","bool"),a=n==null?0:n,o=r.rank,i=s.shape;O(o>0,()=>"mask cannot be scalar"),On(i.slice(a,a+o),r.shape,"mask's shape must match the first K dimensions of tensor's shape,");let l=1;for(let m=a;m<a+o;m++)l*=i[m];let c=i.slice(0,a).concat([l],i.slice(a+o)),u=G(s,c),d=G(r,[-1]),p=await A1(d),h=ot(p,[1]),f=Ou(u,h,a);return e!==s&&s.dispose(),t!==r&&r.dispose(),h.dispose(),u.dispose(),d.dispose(),p.dispose(),f}var TP=CP;function NP(e,t="euclidean",n=null,s=!1){e=D(e,"x","norm");let r=Ev(e,t,n),a=r.shape;if(s){let o=Us(n,e.shape);a=ul(r.shape,o)}return G(r,a)}function Ev(e,t,n=null){if(e.rank===0)return tn(e);if(e.rank!==1&&n===null)return Ev(G(e,[-1]),t,n);if(e.rank===1||typeof n=="number"||Array.isArray(n)&&n.length===1){if(t===1)return Se(tn(e),n);if(t===1/0)return gn(tn(e),n);if(t===-1/0)return Eo(tn(e),n);if(t==="euclidean"||t===2)return _n(Se(Ro(tn(e),Re(2,"int32")),n));throw new Error(`Error in norm: invalid ord value: ${t}`)}if(Array.isArray(n)&&n.length===2){if(t===1)return gn(Se(tn(e),n[0]),n[1]-1);if(t===1/0)return gn(Se(tn(e),n[1]),n[0]);if(t===-1/0)return Eo(Se(tn(e),n[1]),n[0]);if(t==="fro"||t==="euclidean")return _n(Se(gt(e),n));throw new Error(`Error in norm: invalid ord value: ${t}`)}throw new Error(`Error in norm: invalid axis: ${n}`)}var y1=V({norm_:NP});function EP(e,t,n,s,r=!0){let a=D(e,"v","movingAverage"),o=D(t,"x","movingAverage"),i=D(n,"decay","movingAverage");B5(a,o),O(ka(a.shape,o.shape),()=>"Shape mismatch in v and x");let l=Re(1),c=me(l,i),u=L(me(o,a),c);if(r){O(s!=null,()=>"When using zeroDebias: true, step is required.");let d=D(s,"step","movingAverage");u=pe(u,me(l,Ro(i,d)))}return le(a,u)}var RP=V({movingAverage_:EP});function $P(e,t,n){let s=D(e,"indices","scatterND","int32"),r=D(t,"updates","scatterND");D2(r,s,n);let a={indices:s,updates:r},o={shape:n};return W.runKernel(zi,a,o)}var Rv=V({scatterND_:$P});function _P(e,t,n,s){if(e.dtype!=="int32")throw new Error(`tf.sparseToDense() expects the indices to be int32 type, but the dtype was ${e.dtype}.`);if(e.rank>2)throw new Error(`sparseIndices should be a scalar, vector, or matrix, but got shape ${e.shape}.`);let r=e.rank>0?e.shape[0]:1,a=e.rank>1?e.shape[1]:1;if(n.length!==a)throw new Error(`outputShape has incorrect number of elements:, ${n.length}, should be: ${a}.`);let o=t.size;if(!(t.rank===0||t.rank===1&&o===r))throw new Error(`sparseValues has incorrect shape ${t.shape}, should be [] or [${r}]`);if(t.dtype!==s.dtype)throw new Error("sparseValues.dtype must match defaultValues.dtype")}function DP(e,t,n,s=0){let r=D(e,"sparseIndices","sparseToDense","int32"),a=D(t,"sparseValues","sparseToDense"),o=D(s,"defaultValue","sparseToDense",a.dtype);_P(r,a,n,o);let i={sparseIndices:r,sparseValues:a,defaultValue:o},l={outputShape:n};return W.runKernel(ld,i,l)}var x1=V({sparseToDense_:DP});function PP(e,t){let n=D(t,"indices","gatherND","int32"),r={params:D(e,"x","gatherND","string_or_numeric"),indices:n};return W.runKernel(vi,r)}var $v=V({gatherND_:PP});function FP(e,t){if(t==null)return e.shape.slice();if(ka(e.shape,t))return t;if(e.shape.length===t.length){let n=[];for(let s=0;s<e.shape.length;s++)t[s]==null&&e.shape[s]!=null?n.push(e.shape[s]):n.push(t[s]);return n}return t}function OP(e,t,n,s){let r=D(e,"x","dropout");if(O(r.dtype==="float32",()=>`x has to be a floating point tensor since it's going to be scaled, but got a ${r.dtype} tensor instead.`),O(t>=0&&t<1,()=>`rate must be a float in the range [0, 1), but got ${t}.`),t===0)return e instanceof Ze?r.clone():r;let a=FP(r,n),o=1-t,i=pe(Nd(le(Lu(a,0,1,"float32",s),o)),o);return L(r,i)}var _v=V({dropout_:OP});function Dv(e){return Math.floor(Math.pow(2,Math.ceil(Math.log(e)/Math.log(2))))}function b1(e,t,n){let s=1-e%2,r=new Float32Array(e);for(let a=0;a<e;++a){let o=2*Math.PI*a/(e+s-1);r[a]=t-n*Math.cos(o)}return Xt(r,"float32")}async function MP(e,t,n=1){let s=D(e,"predictions","inTopK"),r=D(t,"targets","inTopK");O(s.rank>1,()=>`inTopK() expects the predictions to be of rank 2 or higher, but got ${s.rank}`),O(s.rank-1===r.rank,()=>`predictions rank should be 1 larger than targets rank, but got predictions rank ${s.rank} and targets rank ${r.rank}`),On(s.shape.slice(0,s.shape.length-1),r.shape,"predictions's shape should be align with the targets' shape, except the last dimension.");let a=s.shape[s.shape.length-1];O(n>0&&n<=a,()=>`'k' passed to inTopK() must be > 0 && <= the predictions last dimension (${a}), but got ${n}`);let o=await s.data(),i=await r.data(),[l,c]=[o.length/a,a],u=b5("bool",l);for(let d=0;d<l;d++){let p=d*c,h=o.subarray(p,p+c),f=[];for(let m=0;m<h.length;m++)f.push({value:h[m],index:m});f.sort((m,g)=>g.value-m.value),u[d]=0;for(let m=0;m<n;m++)if(f[m].index===i[d]){u[d]=1;break}}return e!==s&&s.dispose(),t!==r&&r.dispose(),Ut(u,r.shape,"bool")}var zP=MP,$o={};Oe($o,{conv2d:()=>WP,depthwiseConv2d:()=>HP,matMul:()=>qP});function LP(e,t,n,s,r,a="NHWC",o){let i=e;e.rank===3&&(i=G(e,[1,e.shape[0],e.shape[1],e.shape[2]]));let l=t;l.rank===3&&(l=G(t,[1,t.shape[0],t.shape[1],t.shape[2]])),O(i.rank===4,()=>`Error in conv2dDerFilter: input must be rank 4, but got shape ${i.shape}.`),O(l.rank===4,()=>`Error in conv2dDerFilter: dy must be rank 4, but got shape ${l.shape}.`),O(n.length===4,()=>`Error in conv2dDerFilter: filterShape must be length 4, but got ${n}.`);let c=a==="NHWC"?i.shape[3]:i.shape[1],u=a==="NHWC"?l.shape[3]:l.shape[1];O(c===n[2],()=>`Error in conv2dDerFilter: depth of input ${c}) must match input depth in filter (${n[2]}.`),O(u===n[3],()=>`Error in conv2dDerFilter: depth of dy (${u}) must match output depth for filter (${n[3]}).`),o!=null&&O(pn(r),()=>`Error in conv2dDerFilter: pad must be an integer when using, dimRoundingMode ${o} but got pad ${r}.`);let d={x:i,dy:l},p={strides:s,pad:r,dataFormat:a,dimRoundingMode:o,filterShape:n};return W.runKernel(xh,d,p)}var v1=V({conv2DBackpropFilter_:LP});function kf(e,t,n){if(n==null||n==="linear")return e;if(n==="relu")return L(e,Pd(t));throw new Error(`Cannot compute gradient for fused activation ${n}.`)}function Sf(e,t){let n=t,s=qt(e.shape,t.shape);return s.length>0&&(n=Se(n,s)),G(n,e.shape)}function If(e,t,n,s){if(t==="linear")return e;if(t==="relu")return Rr(e);if(t==="elu")return Td(e);if(t==="relu6")return i1(e);if(t==="prelu")return yf(e,n);if(t==="leakyrelu")return df(e,s);if(t==="sigmoid")return cs(e);throw new Error(`Unknown fused activation ${t}.`)}var Cf=(e,t)=>!(e>0)||t==="linear";function BP({x:e,filter:t,strides:n,pad:s,dataFormat:r="NHWC",dilations:a=[1,1],dimRoundingMode:o,bias:i,activation:l="linear",preluActivationWeights:c,leakyreluAlpha:u}){if(l=l||"linear",Cf(W.state.gradientDepth,l)===!1){let w=No(e,t,n,s,r,a,o);return i!=null&&(w=le(w,i)),If(w,l,c,u)}let d=D(e,"x","conv2d","float32"),p=D(t,"filter","conv2d","float32"),h=d,f=!1;d.rank===3&&(f=!0,h=G(d,[1,d.shape[0],d.shape[1],d.shape[2]])),O(h.rank===4,()=>`Error in fused conv2d: input must be rank 4, but got rank ${h.rank}.`),O(p.rank===4,()=>`Error in fused conv2d: filter must be rank 4, but got rank ${p.rank}.`),o!=null&&O(pn(s),()=>`Error in fused conv2d: pad must be an integer when using, dimRoundingMode ${o} but got pad ${s}.`),O(h.shape[3]===p.shape[2],()=>`Error in conv2d: depth of input (${h.shape[3]}) must match input depth for filter ${p.shape[2]}.`),O(Nr(n,a),()=>`Error in conv2D: Either strides or dilations must be 1. Got strides ${n} and dilations '${a}'`),O(r==="NHWC",()=>`Error in conv2d: got dataFormat of ${r} but only NHWC is currently supported.`);let m=Sd(h.shape,p.shape,n,a,s,o),g;i!=null&&(g=D(i,"bias","fused conv2d"),[g]=Dt(g,d),mt(m.outShape,g.shape));let A;c!=null&&(A=D(c,"prelu weights","fused conv2d"));let x=(w,k)=>{let[C,N,R,F]=k,_=kf(w,R,l);O(To(a),()=>`Error in gradient of fused conv2D: dilation rates greater than 1 are not yet supported in gradients. Got dilations '${a}'`);let P=j2(N.shape,_,C,n,s),T=v1(N,_,C.shape,n,s),M=[P,T];if(F!=null){let U=Sf(F,_);M.push(U)}return M},y={x:h,filter:p,bias:g,preluActivationWeights:A},b={strides:n,pad:s,dataFormat:r,dilations:a,dimRoundingMode:o,activation:l,leakyreluAlpha:u};return i==null?Er((k,C,N)=>{let R=W.runKernel(xo,y,b);return N([C,k,R]),f&&(R=G(R,[R.shape[1],R.shape[2],R.shape[3]])),{value:R,gradFunc:x}})(h,p):Er((k,C,N,R)=>{let F=W.runKernel(xo,y,b);return R([C,k,F,N]),f&&(F=G(F,[F.shape[1],F.shape[2],F.shape[3]])),{value:F,gradFunc:x}})(h,p,g)}var WP=V({fusedConv2d_:BP});function VP(e,t,n,s,r,a=[1,1],o){let i=e;e.rank===3&&(i=G(e,[1,e.shape[0],e.shape[1],e.shape[2]]));let l=t;l.rank===3&&(l=G(t,[1,t.shape[0],t.shape[1],t.shape[2]]));let c={x:i,dy:l},u={strides:s,pad:r,dimRoundingMode:o,dilations:a,filterShape:n};return W.runKernel(kh,c,u)}var Pv=V({depthwiseConv2dNativeBackpropFilter_:VP});function UP(e,t,n,s,r,a=[1,1],o){let i=t,l=!1;t.rank===3&&(l=!0,i=G(t,[1,t.shape[0],t.shape[1],t.shape[2]]));let c={dy:i,filter:n},u={strides:s,pad:r,dimRoundingMode:o,dilations:a,inputShape:e},d=W.runKernel(Sh,c,u);return l?G(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var Fv=V({depthwiseConv2dNativeBackpropInput_:UP});function GP({x:e,filter:t,strides:n,pad:s,dataFormat:r="NHWC",dilations:a=[1,1],dimRoundingMode:o,bias:i,activation:l="linear",preluActivationWeights:c,leakyreluAlpha:u}){if(Cf(W.state.gradientDepth,l)===!1){let w=Cd(e,t,n,s,r,a,o);return i!=null&&(w=le(w,i)),If(w,l,c,u)}let d=D(e,"x","depthwiseConv2d","float32"),p=D(t,"filter","depthwiseConv2d","float32"),h=d,f=!1;d.rank===3&&(f=!0,h=G(d,[1,d.shape[0],d.shape[1],d.shape[2]])),O(h.rank===4,()=>`Error in fused depthwiseConv2d: input must be rank 4, but got rank ${h.rank}.`),O(p.rank===4,()=>`Error in fused depthwiseConv2d: filter must be rank 4, but got rank ${p.rank}.`),O(h.shape[3]===p.shape[2],()=>`Error in fused depthwiseConv2d: number of input channels (${h.shape[3]}) must match the inChannels dimension in filter ${p.shape[2]}.`),a==null&&(a=[1,1]),O(Nr(n,a),()=>`Error in fused depthwiseConv2d: Either strides or dilations must be 1. Got strides ${n} and dilations '${a}'`),o!=null&&O(pn(s),()=>`Error in fused depthwiseConv2d: pad must be an integer when using dimRoundingMode ${o} but got pad ${s}.`);let m=Sd(h.shape,p.shape,n,a,s,o,!0),g;i!=null&&(g=D(i,"bias","fused conv2d"),[g]=Dt(g,d),mt(m.outShape,g.shape));let A;c!=null&&(A=D(c,"prelu weights","fused depthwiseConv2d"));let x=(w,k)=>{O(To(a),()=>`Error in gradient of fused depthwiseConv2d: dilation rates greater than 1 are not yet supported. Got dilations '${a}'`);let[C,N,R,F]=k,_=kf(w,R,l),P=Fv(N.shape,_,C,n,s,a,o),T=Pv(N,_,C.shape,n,s,a,o);if(F!=null){let M=Sf(g,_);return[P,T,M]}return[P,T]},y={x:h,filter:p,bias:g,preluActivationWeights:A},b={strides:n,pad:s,dataFormat:r,dilations:a,dimRoundingMode:o,activation:l,leakyreluAlpha:u};return i==null?Er((k,C,N)=>{let R=W.runKernel(bo,y,b);return N([C,k,R]),f&&(R=G(R,[R.shape[1],R.shape[2],R.shape[3]])),{value:R,gradFunc:x}})(h,p):Er((k,C,N,R)=>{let F=W.runKernel(bo,y,b);return R([C,k,F,N]),f&&(F=G(F,[F.shape[1],F.shape[2],F.shape[3]])),{value:F,gradFunc:x}})(h,p,g)}var HP=V({fusedDepthwiseConv2d_:GP});function jP({a:e,b:t,transposeA:n=!1,transposeB:s=!1,bias:r,activation:a="linear",preluActivationWeights:o,leakyreluAlpha:i}){if(Cf(W.state.gradientDepth,a)===!1){let _=Ue(e,t,n,s);return r!=null&&(_=le(_,r)),If(_,a,o,i)}let l=D(e,"a","fused matMul"),c=D(t,"b","fused matMul");[l,c]=Dt(l,c);let u=n?l.shape[l.rank-2]:l.shape[l.rank-1],d=s?c.shape[c.rank-1]:c.shape[c.rank-2],p=n?l.shape[l.rank-1]:l.shape[l.rank-2],h=s?c.shape[c.rank-2]:c.shape[c.rank-1],f=l.shape.slice(0,-2),m=c.shape.slice(0,-2),g=Vt(f),A=Vt(m);O(u===d,()=>`Error in fused matMul: inner shapes (${u}) and (${d}) of Tensors with shapes ${l.shape} and ${c.shape} and transposeA=${n} and transposeB=${s} must match.`);let y=mt(l.shape.slice(0,-2),c.shape.slice(0,-2)).concat([p,h]),b=n?G(l,[g,u,p]):G(l,[g,p,u]),w=s?G(c,[A,h,d]):G(c,[A,d,h]),k;r!=null&&(k=D(r,"bias","fused matMul"),[k]=Dt(k,l),mt(y,k.shape));let C;o!=null&&(C=D(o,"prelu weights","fused matMul"));let N=(_,P)=>{let[T,M,U,j]=P,z=kf(G(_,U.shape),U,a),X,Z;if(!n&&!s?(X=Ue(z,M,!1,!0),Z=Ue(T,z,!0,!1)):!n&&s?(X=Ue(z,M,!1,!1),Z=Ue(z,T,!0,!1)):n&&!s?(X=Ue(M,z,!1,!0),Z=Ue(T,z,!1,!1)):(X=Ue(M,z,!0,!0),Z=Ue(z,T,!0,!0)),r!=null){let J=Sf(j,z);return[X,Z,J]}else return[X,Z]},R={a:b,b:w,bias:k,preluActivationWeights:C},F={transposeA:n,transposeB:s,activation:a,leakyreluAlpha:i};return r==null?Er((P,T,M)=>{let U=W.runKernel(yo,R,F);return M([P,T,U]),{value:G(U,y),gradFunc:N}})(b,w):Er((P,T,M,U)=>{let j=W.runKernel(yo,R,F);return U([P,T,j,M]),{value:G(j,y),gradFunc:N}})(b,w,k)}var qP=V({fusedMatMul_:jP});function XP(e){return b1(e,.54,.46)}var KP=V({hammingWindow_:XP});function ZP(e){return b1(e,.5,.5)}var Ov=V({hannWindow_:ZP});function YP(e,t,n,s=!1,r=0){let a=0,o=[];for(;a+t<=e.size;)o.push(De(e,a,t)),a+=n;if(s)for(;a<e.size;){let i=a+t-e.size,l=vt([De(e,a,t-i),Fu([i],r)]);o.push(l),a+=n}return o.length===0?lr([],[0,t]):G(vt(o),[o.length,t])}var Mv=V({frame_:YP});function JP(e,t,n,s,r=Ov){s==null&&(s=Dv(t));let a=Mv(e,t,n),o=L(a,r(t));return vf(o,s)}var QP=V({stft_:JP});function eF(e,t,n,s,r="bilinear",a=0){let o=D(e,"image","cropAndResize"),i=D(t,"boxes","cropAndResize","float32"),l=D(n,"boxInd","cropAndResize","int32"),c=i.shape[0];O(o.rank===4,()=>`Error in cropAndResize: image must be rank 4,but got rank ${o.rank}.`),O(i.rank===2&&i.shape[1]===4,()=>`Error in cropAndResize: boxes must be have size [${c},4] but had shape ${i.shape}.`),O(l.rank===1&&l.shape[0]===c,()=>`Error in cropAndResize: boxInd must be have size [${c}] but had shape ${i.shape}.`),O(s.length===2,()=>`Error in cropAndResize: cropSize must be of length 2, but got length ${s.length}.`),O(s[0]>=1&&s[1]>=1,()=>`cropSize must be atleast [1,1], but was ${s}`),O(r==="bilinear"||r==="nearest",()=>`method must be bilinear or nearest, but was ${r}`);let u={image:o,boxes:i,boxInd:l},d={method:r,extrapolationValue:a,cropSize:s};return W.runKernel(fi,u,d)}var tF=V({cropAndResize_:eF});function nF(e){let t=D(e,"image","flipLeftRight","float32");O(t.rank===4,()=>`Error in flipLeftRight: image must be rank 4,but got rank ${t.rank}.`);let n={image:t};return W.runKernel(xi,n,{})}var sF=V({flipLeftRight_:nF});function rF(e){let t=D(e,"image","grayscaleToRGB"),n=t.rank-1,s=t.shape[n];O(t.rank>=2,()=>`Error in grayscaleToRGB: images must be at least rank 2, but got rank ${t.rank}.`),O(s===1,()=>`Error in grayscaleToRGB: last dimension of a grayscale image should be size 1, but got size ${s}.`);let r=new Array(t.rank);return r.fill(1,0,n),r[n]=3,js(t,r)}var aF=V({grayscaleToRGB_:rF});function oF(e,t,n=0,s=.5){let r=D(e,"image","rotateWithOffset","float32");O(r.rank===4,()=>`Error in rotateWithOffset: image must be rank 4,but got rank ${r.rank}.`);let a={image:r},o={radians:t,fillValue:n,center:s};return W.runKernel(Zi,a,o)}var iF=V({rotateWithOffset_:oF});function Uu(e,t,n,s,r,a){s==null&&(s=.5),r==null&&(r=Number.NEGATIVE_INFINITY),a==null&&(a=0);let o=e.shape[0];return n=Math.min(n,o),O(0<=s&&s<=1,()=>`iouThreshold must be in [0, 1], but was '${s}'`),O(e.rank===2,()=>`boxes must be a 2D tensor, but was of rank '${e.rank}'`),O(e.shape[1]===4,()=>`boxes must have 4 columns, but 2nd dimension was ${e.shape[1]}`),O(t.rank===1,()=>"scores must be a 1D tensor"),O(t.shape[0]===o,()=>`scores has incompatible shape with boxes. Expected ${o}, but was ${t.shape[0]}`),O(0<=a&&a<=1,()=>`softNmsSigma must be in [0, 1], but was '${a}'`),{maxOutputSize:n,iouThreshold:s,scoreThreshold:r,softNmsSigma:a}}function lF(e,t,n,s=.5,r=Number.NEGATIVE_INFINITY){let a=D(e,"boxes","nonMaxSuppression","float32"),o=D(t,"scores","nonMaxSuppression","float32"),i=Uu(a,o,n,s,r);n=i.maxOutputSize,s=i.iouThreshold,r=i.scoreThreshold;let l={maxOutputSize:n,iouThreshold:s,scoreThreshold:r};return W.runKernel(Ei,{boxes:a,scores:o},l)}var uF=V({nonMaxSuppression_:lF});function cF(e,t,n){let s=dF(e,t,n),r=s<0?-(s+1):s;e.splice(r,0,t)}function dF(e,t,n){return hF(e,t,n||pF)}function pF(e,t){return e>t?1:e<t?-1:0}function hF(e,t,n){let s=0,r=e.length,a=0,o=!1;for(;s<r;){a=s+(r-s>>>1);let i=n(t,e[a]);i>0?s=a+1:(r=a,o=!i)}return o?s:-s-1}function zv(e,t,n,s,r){return w1(e,t,n,s,r,0)}function Lv(e,t,n,s,r,a){return w1(e,t,n,s,r,0,!1,a,!0)}function Bv(e,t,n,s,r,a){return w1(e,t,n,s,r,a,!0)}function w1(e,t,n,s,r,a,o=!1,i=!1,l=!1){let c=[];for(let g=0;g<t.length;g++)t[g]>r&&c.push({score:t[g],boxIndex:g,suppressBeginIndex:0});c.sort(Wv);let u=a>0?-.5/a:0,d=[],p=[];for(;d.length<n&&c.length>0;){let g=c.pop(),{score:A,boxIndex:x,suppressBeginIndex:y}=g;if(A<r)break;let b=!1;for(let w=d.length-1;w>=y;--w){let k=fF(e,x,d[w]);if(k>=s){b=!0;break}if(g.score=g.score*mF(s,u,k),g.score<=r)break}g.suppressBeginIndex=d.length,b||(g.score===A?(d.push(x),p.push(g.score)):g.score>r&&cF(c,g,Wv))}let h=d.length,f=n-h;i&&f>0&&(d.push(...new Array(f).fill(0)),p.push(...new Array(f).fill(0)));let m={selectedIndices:d};return o&&(m.selectedScores=p),l&&(m.validOutputs=h),m}function fF(e,t,n){let s=e.subarray(t*4,t*4+4),r=e.subarray(n*4,n*4+4),a=Math.min(s[0],s[2]),o=Math.min(s[1],s[3]),i=Math.max(s[0],s[2]),l=Math.max(s[1],s[3]),c=Math.min(r[0],r[2]),u=Math.min(r[1],r[3]),d=Math.max(r[0],r[2]),p=Math.max(r[1],r[3]),h=(i-a)*(l-o),f=(d-c)*(p-u);if(h<=0||f<=0)return 0;let m=Math.max(a,c),g=Math.max(o,u),A=Math.min(i,d),x=Math.min(l,p),y=Math.max(A-m,0)*Math.max(x-g,0);return y/(h+f-y)}function mF(e,t,n){let s=Math.exp(t*n*n);return n<=e?s:0}function Wv(e,t){return e.score-t.score||e.score===t.score&&t.boxIndex-e.boxIndex}async function gF(e,t,n,s=.5,r=Number.NEGATIVE_INFINITY){let a=D(e,"boxes","nonMaxSuppressionAsync"),o=D(t,"scores","nonMaxSuppressionAsync"),i=Uu(a,o,n,s,r);n=i.maxOutputSize,s=i.iouThreshold,r=i.scoreThreshold;let l=await Promise.all([a.data(),o.data()]),c=l[0],u=l[1],{selectedIndices:d}=zv(c,u,n,s,r);return a!==e&&a.dispose(),o!==t&&o.dispose(),Xt(d,"int32")}var AF=gF;function yF(e,t,n,s=.5,r=Number.NEGATIVE_INFINITY,a=0){let o=D(e,"boxes","nonMaxSuppression"),i=D(t,"scores","nonMaxSuppression"),l=Uu(o,i,n,s,r,a);n=l.maxOutputSize,s=l.iouThreshold,r=l.scoreThreshold,a=l.softNmsSigma;let c={boxes:o,scores:i},u={maxOutputSize:n,iouThreshold:s,scoreThreshold:r,softNmsSigma:a},d=W.runKernel(Ri,c,u);return{selectedIndices:d[0],selectedScores:d[1]}}var xF=V({nonMaxSuppressionWithScore_:yF});async function bF(e,t,n,s=.5,r=Number.NEGATIVE_INFINITY,a=0){let o=D(e,"boxes","nonMaxSuppressionAsync"),i=D(t,"scores","nonMaxSuppressionAsync"),l=Uu(o,i,n,s,r,a);n=l.maxOutputSize,s=l.iouThreshold,r=l.scoreThreshold,a=l.softNmsSigma;let c=await Promise.all([o.data(),i.data()]),u=c[0],d=c[1],{selectedIndices:p,selectedScores:h}=Bv(u,d,n,s,r,a);return o!==e&&o.dispose(),i!==t&&i.dispose(),{selectedIndices:Xt(p,"int32"),selectedScores:Xt(h)}}var vF=bF;function wF(e,t,n,s=.5,r=Number.NEGATIVE_INFINITY,a=!1){let o=D(e,"boxes","nonMaxSuppression"),i=D(t,"scores","nonMaxSuppression"),l=Uu(o,i,n,s,r,null),c=l.maxOutputSize,u=l.iouThreshold,d=l.scoreThreshold,p={boxes:o,scores:i},h={maxOutputSize:c,iouThreshold:u,scoreThreshold:d,padToMaxOutputSize:a},f=W.runKernel(yu,p,h);return{selectedIndices:f[0],validOutputs:f[1]}}var kF=V({nonMaxSuppressionPadded_:wF});async function SF(e,t,n,s=.5,r=Number.NEGATIVE_INFINITY,a=!1){let o=D(e,"boxes","nonMaxSuppressionAsync"),i=D(t,"scores","nonMaxSuppressionAsync"),l=Uu(o,i,n,s,r,null),c=l.maxOutputSize,u=l.iouThreshold,d=l.scoreThreshold,[p,h]=await Promise.all([o.data(),i.data()]),{selectedIndices:f,validOutputs:m}=Lv(p,h,c,u,d,a);return o!==e&&o.dispose(),i!==t&&i.dispose(),{selectedIndices:Xt(f,"int32"),validOutputs:Re(m,"int32")}}var IF=SF;function CF(e,t,n=!1,s=!1){let r=D(e,"images","resizeBilinear");O(r.rank===3||r.rank===4,()=>`Error in resizeBilinear: x must be rank 3 or 4, but got rank ${r.rank}.`),O(t.length===2,()=>`Error in resizeBilinear: new shape must 2D, but got shape ${t}.`),O(s===!1||n===!1,()=>"Error in resizeBilinear: If halfPixelCenters is true, alignCorners must be false.");let a=r,o=!1;r.rank===3&&(o=!0,a=G(r,[1,r.shape[0],r.shape[1],r.shape[2]]));let[]=t,i={images:a},l={alignCorners:n,halfPixelCenters:s,size:t},c=W.runKernel(ro,i,l);return o?G(c,[c.shape[1],c.shape[2],c.shape[3]]):c}var TF=V({resizeBilinear_:CF});function NF(e,t,n=!1,s=!1){let r=D(e,"images","resizeNearestNeighbor");O(r.rank===3||r.rank===4,()=>`Error in resizeNearestNeighbor: x must be rank 3 or 4, but got rank ${r.rank}.`),O(t.length===2,()=>`Error in resizeNearestNeighbor: new shape must 2D, but got shape ${t}.`),O(r.dtype==="float32"||r.dtype==="int32",()=>"`images` must have `int32` or `float32` as dtype"),O(s===!1||n===!1,()=>"Error in resizeNearestNeighbor: If halfPixelCenters is true, alignCorners must be false.");let a=r,o=!1;r.rank===3&&(o=!0,a=G(r,[1,r.shape[0],r.shape[1],r.shape[2]]));let[]=t,i={images:a},l={alignCorners:n,halfPixelCenters:s,size:t},c=W.runKernel(vu,i,l);return o?G(c,[c.shape[1],c.shape[2],c.shape[3]]):c}var EF=V({resizeNearestNeighbor_:NF});function RF(e,t="binary",n=!1,s=.5){let r=D(e,"image","threshold"),a=.2989,o=.587,i=.114,l=r.shape[0]*r.shape[1],c=L(Xt([s]),255),u,d,p,h;if(O(r.rank===3,()=>`Error in threshold: image must be rank 3,but got rank ${r.rank}.`),O(r.shape[2]===3||r.shape[2]===1,()=>`Error in threshold: image color channel must be equal to 3 or 1but got ${r.shape[2]}.`),O(r.dtype==="int32"||r.dtype==="float32",()=>`Error in dtype: image dtype must be int32 or float32,but got dtype ${r.dtype}.`),O(t==="otsu"||t==="binary",()=>`Method must be binary or otsu, but was ${t}`),r.shape[2]===3){[u,d,p]=nn(r,[1,1,1],-1);let g=L(u,a),A=L(d,o),x=L(p,i);h=le(le(g,A),x)}else h=e;if(t==="otsu"){let g=G2(de(l1(h),"int32"),Ut([]),256);c=$F(g,l)}let f=n?ll(h,c):ps(h,c);return de(L(f,255),"int32")}function $F(e,t){let n=Xt([-1]),s=Xt([0]),r=Xt([0]),a,o,i,l,c,u;for(let d=0;d<e.size-1;d++){a=De(e,0,d+1),o=De(e,d+1),c=pe(Se(a),t),u=pe(Se(o),t);let p=Se(L(a,Bu(0,a.size)));i=pe(p,Se(a));let h=Fu(o.shape,a.size),f=le(Bu(0,o.size),h),m=L(o,f);l=pe(Se(m),Se(o));let g=me(i,l),A=me(i,l),x=L(c,u);r=L(L(x,g),A);let y=ps(r,s);s=Bn(y,r,s),n=Bn(y,Xt([d]),n)}return n}var _F=V({threshold_:RF});function DF(e,t,n="nearest",s="constant",r=0,a){let o=D(e,"image","transform","float32"),i=D(t,"transforms","transform","float32");O(o.rank===4,()=>`Error in transform: image must be rank 4,but got rank ${o.rank}.`),O(i.rank===2&&(i.shape[0]===o.shape[0]||i.shape[0]===1)&&i.shape[1]===8,()=>"Error in transform: Input transform should be batch x 8 or 1 x 8"),O(a==null||a.length===2,()=>`Error in transform: outputShape must be [height, width] or null, but got ${a}.`);let l={image:o,transforms:i},c={interpolation:n,fillMode:s,fillValue:r,outputShape:a};return W.runKernel(qi,l,c)}var PF=V({transform_:DF});function FF(e,t,n){O(t%1==0,()=>`bandPart(): numLower must be an integer, got ${t}.`),O(n%1==0,()=>`bandPart(): numUpper must be an integer, got ${n}.`);let s=D(e,"a","bandPart");O(s.rank>=2,()=>`bandPart(): Rank must be at least 2, got ${s.rank}.`);let r=s.shape,[a,o]=s.shape.slice(-2);if(!(t<=a))throw new Error(`bandPart(): numLower (${t}) must not be greater than the number of rows (${a}).`);if(!(n<=o))throw new Error(`bandPart(): numUpper (${n}) must not be greater than the number of columns (${o}).`);t<0&&(t=a),n<0&&(n=o);let i=G(Bu(0,a,1,"int32"),[-1,1]),l=Bu(0,o,1,"int32"),c=me(i,l),u=ir(ll(c,Re(+t,"int32")),il(c,Re(-n,"int32"))),d=Gt([a,o],s.dtype);return G(An(ss(G(s,[-1,a,o])).map(p=>Bn(u,p,d))),r)}var OF=V({bandPart_:FF});function MF(e){let t;if(Array.isArray(e)){t=!1,O(e!=null&&e.length>0,()=>"Gram-Schmidt process: input must not be null, undefined, or empty");let r=e[0].shape[0];for(let a=1;a<e.length;++a)O(e[a].shape[0]===r,()=>`Gram-Schmidt: Non-unique lengths found in the input vectors: (${e[a].shape[0]} vs. ${r})`)}else t=!0,e=nn(e,e.shape[0],0).map(r=>ot(r,[0]));O(e.length<=e[0].shape[0],()=>`Gram-Schmidt: Number of vectors (${e.length}) exceeds number of dimensions (${e[0].shape[0]}).`);let n=[],s=e;for(let r=0;r<e.length;++r)n.push(W.tidy(()=>{let a=s[r];if(r>0)for(let o=0;o<r;++o){let i=L(Se(L(n[o],a)),n[o]);a=me(a,i)}return pe(a,y1(a,"euclidean"))}));return t?An(n,0):n}var zF=V({gramSchmidt_:MF});function LF(e,t=!1){if(O(e.rank>=2,()=>`qr() requires input tensor to have a rank >= 2, but got rank ${e.rank}`),e.rank===2)return Vv(e,t);{let n=e.shape.slice(0,e.shape.length-2).reduce((l,c)=>l*c),s=ss(G(e,[n,e.shape[e.shape.length-2],e.shape[e.shape.length-1]]),0),r=[],a=[];s.forEach(l=>{let[c,u]=Vv(l,t);r.push(c),a.push(u)});let o=G(An(r,0),e.shape),i=G(An(a,0),e.shape);return[o,i]}}function Vv(e,t=!1){return W.tidy(()=>{O(e.shape.length===2,()=>`qr2d() requires a 2D Tensor, but got a ${e.shape.length}D Tensor.`);let n=e.shape[0],s=e.shape[1],r=Y2(n),a=Ln(e),o=lr([[1]],[1,1]),i=Ln(o),l=n>=s?s:n;for(let c=0;c<l;++c){let u=a,d=i,p=r;[i,a,r]=W.tidy(()=>{let h=De(a,[c,c],[n-c,1]),f=y1(h),m=De(a,[c,c],[1,1]),g=Bn(ps(m,0),lr([[-1]]),lr([[1]])),A=me(m,L(g,f)),x=pe(h,A);x.shape[0]===1?i=Ln(o):i=vt([o,De(x,[1,0],[x.shape[0]-1,x.shape[1]])],0);let y=Ft(pe(Ue(g,A),f)),b=De(a,[c,0],[n-c,s]),w=L(y,i),k=Ye(i);if(c===0)a=me(b,Ue(w,Ue(k,b)));else{let R=me(b,Ue(w,Ue(k,b)));a=vt([De(a,[0,0],[c,s]),R],0)}let C=Ye(w),N=De(r,[0,c],[n,r.shape[1]-c]);if(c===0)r=me(N,Ue(Ue(N,i),C));else{let R=me(N,Ue(Ue(N,i),C));r=vt([De(r,[0,0],[n,c]),R],1)}return[i,a,r]}),se([u,d,p])}return!t&&n>s&&(r=De(r,[0,0],[n,s]),a=De(a,[0,0],[s,s])),[r,a]})}var BF=V({qr_:LF}),Wn;(function(e){e[e.NONE=0]="NONE",e[e.MEAN=1]="MEAN",e[e.SUM=2]="SUM",e[e.SUM_BY_NONZERO_WEIGHTS=3]="SUM_BY_NONZERO_WEIGHTS"})(Wn||(Wn={}));function WF(e,t,n=Wn.SUM_BY_NONZERO_WEIGHTS){let s=D(e,"losses","computeWeightedLoss"),r=null;t!=null&&(r=D(t,"weights","computeWeightedLoss"));let a=r==null?s:L(s,r);if(n===Wn.NONE)return a;if(n===Wn.SUM)return Se(a);if(n===Wn.MEAN){if(r==null)return Bt(a);{let o=s.size/r.size,i=pe(Se(a),Se(r));return o>1?pe(i,Re(o)):i}}if(n===Wn.SUM_BY_NONZERO_WEIGHTS){if(r==null)return pe(Se(a),Re(s.size));{let o=L(r,hs(s.shape)),i=de(Se(zu(o,Re(0))),"float32");return pe(Se(a),i)}}throw Error(`Unknown reduction: ${n}`)}var Zr=V({computeWeightedLoss_:WF});function VF(e,t,n,s=Wn.SUM_BY_NONZERO_WEIGHTS){let r=D(e,"labels","absoluteDifference"),a=D(t,"predictions","absoluteDifference"),o=null;n!=null&&(o=D(n,"weights","absoluteDifference")),On(r.shape,a.shape,"Error in absoluteDifference: ");let i=tn(me(r,a));return Zr(i,o,s)}var UF=V({absoluteDifference_:VF});function GF(e,t,n,s,r=Wn.SUM_BY_NONZERO_WEIGHTS){let a=D(e,"labels","cosineDistance"),o=D(t,"predictions","cosineDistance"),i=null;s!=null&&(i=D(s,"weights","cosineDistance")),On(a.shape,o.shape,"Error in cosineDistance: ");let l=Re(1),c=me(l,Se(L(a,o),n,!0));return Zr(c,i,r)}var HF=V({cosineDistance_:GF});function jF(e,t,n,s=Wn.SUM_BY_NONZERO_WEIGHTS){let r=D(e,"labels","hingeLoss"),a=D(t,"predictions","hingeLoss"),o=null;n!=null&&(o=D(n,"weights","hingeLoss")),On(r.shape,a.shape,"Error in hingeLoss: ");let i=Re(1);r=me(L(Re(2),r),i);let l=Rr(me(i,L(r,a)));return Zr(l,o,s)}var qF=V({hingeLoss_:jF});function XF(e,t,n,s=1,r=Wn.SUM_BY_NONZERO_WEIGHTS){let a=D(e,"labels","huberLoss"),o=D(t,"predictions","huberLoss"),i=null;n!=null&&(i=D(n,"weights","huberLoss")),On(a.shape,o.shape,"Error in huberLoss: ");let l=Re(s),c=tn(me(o,a)),u=Ed(c,l),d=me(c,u),p=le(L(Re(.5),gt(u)),L(l,d));return Zr(p,i,r)}var KF=V({huberLoss_:XF});function ZF(e,t,n,s=1e-7,r=Wn.SUM_BY_NONZERO_WEIGHTS){let a=D(e,"labels","logLoss"),o=D(t,"predictions","logLoss"),i=null;n!=null&&(i=D(n,"weights","logLoss")),On(a.shape,o.shape,"Error in logLoss: ");let l=Re(1),c=Re(s),u=Ft(L(a,Ns(le(o,c)))),d=L(me(l,a),Ns(le(me(l,o),c))),p=me(u,d);return Zr(p,i,r)}var YF=V({logLoss_:ZF});function JF(e,t,n,s=Wn.SUM_BY_NONZERO_WEIGHTS){let r=D(e,"labels","meanSquaredError"),a=D(t,"predictions","meanSquaredError"),o=null;n!=null&&(o=D(n,"weights","meanSquaredError")),On(r.shape,a.shape,"Error in meanSquaredError: ");let i=m1(r,a);return Zr(i,o,s)}var QF=V({meanSquaredError_:JF});function eO(e,t){let n=D(e,"labels","sigmoidCrossEntropyWithLogits"),s=D(t,"logits","sigmoidCrossEntropyWithLogits");On(n.shape,s.shape,"Error in sigmoidCrossEntropyWithLogits: ");let r=Rr(s),a=L(s,n),o=pf(Ts(Ft(tn(s))));return le(me(r,a),o)}function tO(e,t,n,s=0,r=Wn.SUM_BY_NONZERO_WEIGHTS){let a=D(e,"multiClassLabels","sigmoidCrossEntropy"),o=D(t,"logits","sigmoidCrossEntropy"),i=null;if(n!=null&&(i=D(n,"weights","sigmoidCrossEntropy")),On(a.shape,o.shape,"Error in sigmoidCrossEntropy: "),s>0){let c=Re(s),u=Re(1),d=Re(.5);a=le(L(a,me(u,c)),L(d,c))}let l=eO(a,o);return Zr(l,i,r)}var nO=V({sigmoidCrossEntropy_:tO});function sO(e,t,n=-1){if(n===-1&&(n=t.rank-1),n!==t.rank-1)throw Error(`Softmax cross entropy along a non-last dimension is not yet supported. Labels / logits was rank ${t.rank} and dim was ${n}`);return Er((r,a,o)=>{let l=fv(a,[n],!0),c=me(de(a,"float32"),l);o([r,c]);let u=Ft(L(c,r));return{value:Se(u,[n]),gradFunc:(h,f)=>{let[m,g]=f,A=ul(h.shape,[n]);return[L(G(h,A),me(de(m,"float32"),Ts(g))),L(G(h,A),me(Ts(g),de(m,"float32")))]}}})(e,t)}function rO(e,t,n,s=0,r=Wn.SUM_BY_NONZERO_WEIGHTS){let a=D(e,"onehotLabels","softmaxCrossEntropy"),o=D(t,"logits","softmaxCrossEntropy"),i=null;if(n!=null&&(i=D(n,"weights","softmaxCrossEntropy")),On(a.shape,o.shape,"Error in softmaxCrossEntropy: "),s>0){let c=Re(s),u=Re(1),d=Re(a.shape[1]);a=le(L(a,me(u,c)),pe(c,d))}let l=sO(a,o);return Zr(l,i,r)}var aO=V({softmaxCrossEntropy_:rO});function oO(e,t,n,s){let r=D(e,"indices","sparseFillEmptyRows"),a=D(t,"values","sparseFillEmptyRows"),o=D(n,"denseShape","sparseFillEmptyRows"),i=D(s,"defaultValue","sparseFillEmptyRows",a.dtype);if(r.rank!==2)throw new Error(`Indices should be Tensor2D but received shape
|
|
${r.shape}`);if(a.rank!==1)throw new Error(`Values should be Tensor1D but received shape ${a.shape}`);if(o.rank!==1)throw new Error(`Dense shape should be Tensor1D but received shape ${o.shape}`);if(i.rank!==0)throw new Error(`Default value should be a scalar but received shape ${i.shape}`);let l={indices:r,values:a,denseShape:o,defaultValue:i},c=W.runKernel(Lh,l);return{outputIndices:c[0],outputValues:c[1],emptyRowIndicator:c[2],reverseIndexMap:c[3]}}var iO=V({sparseFillEmptyRows_:oO});function lO(e,t,n){let s=D(e,"inputIndices","sparseReshape"),r=D(t,"inputShape","sparseReshape"),a=D(n,"newShape","sparseReshape");if(s.rank!==2)throw new Error(`Input indices should be Tensor2D but received shape
|
|
${s.shape}`);if(r.rank!==1)throw new Error(`Input shape should be Tensor1D but received shape ${r.shape}`);if(a.rank!==1)throw new Error(`New shape should be Tensor1D but received shape ${a.shape}`);let o={inputIndices:s,inputShape:r,newShape:a},i=W.runKernel(Bh,o);return{outputIndices:i[0],outputShape:i[1]}}var uO=V({sparseReshape_:lO});function cO(e,t,n){let s=D(e,"data","sparseSegmentMean"),r=D(t,"indices","sparseSegmentMean"),a=D(n,"segmentIds","sparseSegmentMean");if(s.rank<1)throw new Error("Data should be at least 1 dimensional but received scalar");if(r.rank!==1)throw new Error(`Indices should be Tensor1D but received shape
|
|
${r.shape}`);if(a.rank!==1)throw new Error(`Segment ids should be Tensor1D but received shape
|
|
${a.shape}`);let o={data:s,indices:r,segmentIds:a};return W.runKernel(Wh,o)}var dO=V({sparseSegmentMean_:cO});function pO(e,t,n){let s=D(e,"data","sparseSegmentSum"),r=D(t,"indices","sparseSegmentSum"),a=D(n,"segmentIds","sparseSegmentSum");if(s.rank<1)throw new Error("Data should be at least 1 dimensional but received scalar");if(r.rank!==1)throw new Error(`Indices should be Tensor1D but received shape
|
|
${r.shape}`);if(a.rank!==1)throw new Error(`Segment ids should be Tensor1D but received shape
|
|
${a.shape}`);let o={data:s,indices:r,segmentIds:a};return W.runKernel(Vh,o)}var hO=V({sparseSegmentSum_:pO});function fO(e,t,n,s,r,a,o,i){let l=D(e,"data","stringNGrams","string");if(l.dtype!=="string")throw new Error("Data must be of datatype string");if(l.shape.length!==1)throw new Error(`Data must be a vector, saw: ${l.shape}`);let c=D(t,"dataSplits","stringNGrams");if(c.dtype!=="int32")throw new Error("Data splits must be of datatype int32");let u={separator:n,nGramWidths:s,leftPad:r,rightPad:a,padWidth:o,preserveShortSequences:i},d={data:l,dataSplits:c},p=W.runKernel(ud,d,u);return{nGrams:p[0],nGramsSplits:p[1]}}var mO=V({stringNGrams_:fO});function gO(e,t,n=!0){let s=D(e,"input","stringSplit","string"),r=D(t,"delimiter","stringSplit","string");if(s.rank!==1)throw new Error(`Input should be Tensor1D but received shape ${s.shape}`);if(r.rank!==0)throw new Error(`Delimiter should be a scalar but received shape ${r.shape}`);let a={skipEmpty:n},o={input:s,delimiter:r},i=W.runKernel(Uh,o,a);return{indices:i[0],values:i[1],shape:i[2]}}var AO=V({stringSplit_:gO});function yO(e,t){let n=D(e,"input","stringToHashBucketFast","string"),s={numBuckets:t};if(t<=0)throw new Error("Number of buckets must be at least 1");let r={input:n};return W.runKernel(Gh,r,s)}var xO=V({stringToHashBucketFast_:yO}),bO={fft:bf,ifft:Dd,rfft:vf,irfft:f1},vO={hammingWindow:KP,hannWindow:Ov,frame:Mv,stft:QP},$e={flipLeftRight:sF,grayscaleToRGB:aF,resizeNearestNeighbor:EF,resizeBilinear:TF,rotateWithOffset:iF,cropAndResize:tF,nonMaxSuppression:uF,nonMaxSuppressionAsync:AF,nonMaxSuppressionWithScore:xF,nonMaxSuppressionWithScoreAsync:vF,nonMaxSuppressionPadded:kF,nonMaxSuppressionPaddedAsync:IF,threshold:_F,transform:PF},Uv={bandPart:OF,gramSchmidt:zF,qr:BF},wO={absoluteDifference:UF,computeWeightedLoss:Zr,cosineDistance:HF,hingeLoss:qF,huberLoss:KF,logLoss:YF,meanSquaredError:QF,sigmoidCrossEntropy:nO,softmaxCrossEntropy:aO},Fd={sparseFillEmptyRows:iO,sparseReshape:uO,sparseSegmentMean:dO,sparseSegmentSum:hO},Tf={stringNGrams:mO,stringSplit:AO,stringToHashBucketFast:xO},Yr=class extends T3{minimize(e,t=!1,n){let{value:s,grads:r}=this.computeGradients(e,n);if(n!=null){let a=n.map(o=>({name:o.name,tensor:r[o.name]}));this.applyGradients(a)}else this.applyGradients(r);return se(r),t?s:(s.dispose(),null)}get iterations(){return this.iterations_==null&&(this.iterations_=0),this.iterations_}incrementIterations(){this.iterations_=this.iterations+1}computeGradients(e,t){return cv(e,t)}dispose(){this.iterations_!=null&&se(this.iterations_)}async saveIterations(){return this.iterations_==null&&(this.iterations_=0),{name:"iter",tensor:Re(this.iterations_,"int32")}}async getWeights(){throw new Error("getWeights() is not implemented for this optimizer yet.")}async setWeights(e){throw new Error(`setWeights() is not implemented for this optimizer class ${this.getClassName()}`)}async extractIterations(e){return this.iterations_=(await e[0].tensor.data())[0],e.slice(1)}};Object.defineProperty(Yr,Symbol.hasInstance,{value:e=>e.minimize!=null&&e.computeGradients!=null&&e.applyGradients!=null});var Nf=class extends Yr{constructor(e,t,n=null){super();this.learningRate=e,this.rho=t,this.epsilon=n,this.accumulatedGrads=[],this.accumulatedUpdates=[],n==null&&(this.epsilon=W.backend.epsilon())}applyGradients(e){(Array.isArray(e)?e.map(n=>n.name):Object.keys(e)).forEach((n,s)=>{let r=W.registeredVariables[n],a=!1;this.accumulatedGrads[s]==null&&(this.accumulatedGrads[s]={originalName:`${n}/accum_grad`,variable:q(()=>Je(r).variable(a))}),this.accumulatedUpdates[s]==null&&(this.accumulatedUpdates[s]={originalName:`${n}/accum_var`,variable:q(()=>Je(r).variable(a))});let o=Array.isArray(e)?e[s].tensor:e[n];if(o==null)return;let i=this.accumulatedGrads[s].variable,l=this.accumulatedUpdates[s].variable;q(()=>{let c=le(L(i,this.rho),L(gt(o),1-this.rho)),u=L(pe(_n(le(l,this.epsilon)),_n(le(i,this.epsilon))),o),d=le(L(l,this.rho),L(gt(u),1-this.rho));i.assign(c),l.assign(d);let p=le(L(u,-this.learningRate),r);r.assign(p)})}),this.incrementIterations()}dispose(){this.accumulatedUpdates!=null&&(se(this.accumulatedGrads.map(e=>e.variable)),se(this.accumulatedUpdates.map(e=>e.variable)))}async getWeights(){let e=[...this.accumulatedGrads,...this.accumulatedUpdates];return[await this.saveIterations()].concat(e.map(t=>({name:t.originalName,tensor:t.variable})))}async setWeights(e){e=await this.extractIterations(e);let t=e.length/2,n=!1;this.accumulatedGrads=e.slice(0,t).map(s=>({originalName:s.name,variable:s.tensor.variable(n)})),this.accumulatedUpdates=e.slice(t,t*2).map(s=>({originalName:s.name,variable:s.tensor.variable(n)}))}getConfig(){return{learningRate:this.learningRate,rho:this.rho,epsilon:this.epsilon}}static fromConfig(e,t){return new e(t.learningRate,t.rho,t.epsilon)}};Nf.className="Adadelta";Co(Nf);var Ef=class extends Yr{constructor(e,t=.1){super();this.learningRate=e,this.initialAccumulatorValue=t,this.accumulatedGrads=[]}applyGradients(e){(Array.isArray(e)?e.map(n=>n.name):Object.keys(e)).forEach((n,s)=>{let r=W.registeredVariables[n];if(this.accumulatedGrads[s]==null){let i=!1;this.accumulatedGrads[s]={originalName:`${n}/accumulator`,variable:q(()=>Fu(r.shape,this.initialAccumulatorValue).variable(i))}}let a=Array.isArray(e)?e[s].tensor:e[n];if(a==null)return;let o=this.accumulatedGrads[s].variable;q(()=>{let i=le(o,gt(a));o.assign(i);let l=le(L(pe(a,_n(le(i,W.backend.epsilon()))),-this.learningRate),r);r.assign(l)})}),this.incrementIterations()}dispose(){this.accumulatedGrads!=null&&se(this.accumulatedGrads.map(e=>e.variable))}async getWeights(){return[await this.saveIterations()].concat(this.accumulatedGrads.map(e=>({name:e.originalName,tensor:e.variable})))}async setWeights(e){e=await this.extractIterations(e);let t=!1;this.accumulatedGrads=e.map(n=>({originalName:n.name,variable:n.tensor.variable(t)}))}getConfig(){return{learningRate:this.learningRate,initialAccumulatorValue:this.initialAccumulatorValue}}static fromConfig(e,t){return new e(t.learningRate,t.initialAccumulatorValue)}};Ef.className="Adagrad";Co(Ef);var Rf=class extends Yr{constructor(e,t,n,s=null){super();this.learningRate=e,this.beta1=t,this.beta2=n,this.epsilon=s,this.accumulatedFirstMoment=[],this.accumulatedSecondMoment=[],q(()=>{this.accBeta1=Re(t).variable(),this.accBeta2=Re(n).variable()}),s==null&&(this.epsilon=W.backend.epsilon())}applyGradients(e){let t=Array.isArray(e)?e.map(n=>n.name):Object.keys(e);q(()=>{let n=me(1,this.accBeta1),s=me(1,this.accBeta2);t.forEach((r,a)=>{let o=W.registeredVariables[r],i=!1;this.accumulatedFirstMoment[a]==null&&(this.accumulatedFirstMoment[a]={originalName:`${r}/m`,variable:q(()=>Je(o).variable(i))}),this.accumulatedSecondMoment[a]==null&&(this.accumulatedSecondMoment[a]={originalName:`${r}/v`,variable:q(()=>Je(o).variable(i))});let l=Array.isArray(e)?e[a].tensor:e[r];if(l==null)return;let c=this.accumulatedFirstMoment[a].variable,u=this.accumulatedSecondMoment[a].variable,d=le(L(c,this.beta1),L(l,1-this.beta1)),p=le(L(u,this.beta2),L(gt(l),1-this.beta2)),h=pe(d,n),f=pe(p,s);c.assign(d),u.assign(p);let m=le(L(pe(h,le(_n(f),this.epsilon)),-this.learningRate),o);o.assign(m)}),this.accBeta1.assign(L(this.accBeta1,this.beta1)),this.accBeta2.assign(L(this.accBeta2,this.beta2))}),this.incrementIterations()}dispose(){this.accBeta1.dispose(),this.accBeta2.dispose(),this.accumulatedFirstMoment!=null&&se(this.accumulatedFirstMoment.map(e=>e.variable)),this.accumulatedSecondMoment!=null&&se(this.accumulatedSecondMoment.map(e=>e.variable))}async getWeights(){let e=[...this.accumulatedFirstMoment,...this.accumulatedSecondMoment];return[await this.saveIterations()].concat(e.map(t=>({name:t.originalName,tensor:t.variable})))}async setWeights(e){e=await this.extractIterations(e),q(()=>{this.accBeta1.assign(Ro(this.beta1,this.iterations_+1)),this.accBeta2.assign(Ro(this.beta2,this.iterations_+1))});let t=e.length/2,n=!1;this.accumulatedFirstMoment=e.slice(0,t).map(s=>({originalName:s.name,variable:s.tensor.variable(n)})),this.accumulatedSecondMoment=e.slice(t,t*2).map(s=>({originalName:s.name,variable:s.tensor.variable(n)}))}getConfig(){return{learningRate:this.learningRate,beta1:this.beta1,beta2:this.beta2,epsilon:this.epsilon}}static fromConfig(e,t){return new e(t.learningRate,t.beta1,t.beta2,t.epsilon)}};Rf.className="Adam";Co(Rf);var $f=class extends Yr{constructor(e,t,n,s=null,r=0){super();this.learningRate=e,this.beta1=t,this.beta2=n,this.epsilon=s,this.decay=r,this.accumulatedFirstMoment=[],this.accumulatedWeightedInfNorm=[],q(()=>{this.iteration=Re(0).variable(),this.accBeta1=Re(t).variable()}),s==null&&(this.epsilon=W.backend.epsilon())}applyGradients(e){let t=Array.isArray(e)?e.map(n=>n.name):Object.keys(e);q(()=>{let n=me(1,this.accBeta1),s=pe(-this.learningRate,le(L(this.iteration,this.decay),1));t.forEach((r,a)=>{let o=W.registeredVariables[r],i=!1;this.accumulatedFirstMoment[a]==null&&(this.accumulatedFirstMoment[a]={originalName:`${r}/m`,variable:Je(o).variable(i)}),this.accumulatedWeightedInfNorm[a]==null&&(this.accumulatedWeightedInfNorm[a]={originalName:`${r}/v`,variable:Je(o).variable(i)});let l=Array.isArray(e)?e[a].tensor:e[r];if(l==null)return;let c=this.accumulatedFirstMoment[a].variable,u=this.accumulatedWeightedInfNorm[a].variable,d=le(L(c,this.beta1),L(l,1-this.beta1)),p=L(u,this.beta2),h=tn(l),f=Kr(p,h);c.assign(d),u.assign(f);let m=le(L(pe(s,n),pe(d,le(f,this.epsilon))),o);o.assign(m)}),this.iteration.assign(le(this.iteration,1)),this.accBeta1.assign(L(this.accBeta1,this.beta1))}),this.incrementIterations()}dispose(){this.accBeta1.dispose(),this.iteration.dispose(),this.accumulatedFirstMoment!=null&&se(this.accumulatedFirstMoment.map(e=>e.variable)),this.accumulatedWeightedInfNorm!=null&&se(this.accumulatedWeightedInfNorm.map(e=>e.variable))}async getWeights(){throw new Error("getWeights() is not implemented for Adamax yet.")}async setWeights(e){throw new Error("setWeights() is not implemented for Adamax yet.")}getConfig(){return{learningRate:this.learningRate,beta1:this.beta1,beta2:this.beta2,epsilon:this.epsilon,decay:this.decay}}static fromConfig(e,t){return new e(t.learningRate,t.beta1,t.beta2,t.epsilon,t.decay)}};$f.className="Adamax";Co($f);var Od=class extends Yr{constructor(e){super();this.learningRate=e,this.setLearningRate(e)}applyGradients(e){(Array.isArray(e)?e.map(n=>n.name):Object.keys(e)).forEach((n,s)=>{let r=Array.isArray(e)?e[s].tensor:e[n];if(r==null)return;let a=W.registeredVariables[n];q(()=>{let o=le(L(this.c,r),a);a.assign(o)})}),this.incrementIterations()}setLearningRate(e){this.learningRate=e,this.c!=null&&this.c.dispose(),this.c=fn(Re(-e))}dispose(){this.c.dispose()}async getWeights(){return[await this.saveIterations()]}async setWeights(e){if(e=await this.extractIterations(e),e.length!==0)throw new Error("SGD optimizer does not have settable weights.")}getConfig(){return{learningRate:this.learningRate}}static fromConfig(e,t){return new e(t.learningRate)}};Od.className="SGD";Co(Od);var _f=class extends Od{constructor(e,t,n=!1){super(e);this.learningRate=e,this.momentum=t,this.useNesterov=n,this.accumulations=[],this.m=Re(this.momentum)}applyGradients(e){(Array.isArray(e)?e.map(n=>n.name):Object.keys(e)).forEach((n,s)=>{let r=W.registeredVariables[n];if(this.accumulations[s]==null){let i=!1;this.accumulations[s]={originalName:`${n}/momentum`,variable:q(()=>Je(r).variable(i))}}let a=this.accumulations[s].variable,o=Array.isArray(e)?e[s].tensor:e[n];o!=null&&q(()=>{let i,l=le(L(this.m,a),o);this.useNesterov?i=le(L(this.c,le(o,L(l,this.m))),r):i=le(L(this.c,l),r),a.assign(l),r.assign(i)})}),this.incrementIterations()}dispose(){this.m.dispose(),this.accumulations!=null&&se(this.accumulations.map(e=>e.variable))}setMomentum(e){this.momentum=e}async getWeights(){return[await this.saveIterations()].concat(this.accumulations.map(e=>({name:e.originalName,tensor:e.variable})))}async setWeights(e){e=await this.extractIterations(e);let t=!1;this.accumulations=e.map(n=>({originalName:n.name,variable:n.tensor.variable(t)}))}getConfig(){return{learningRate:this.learningRate,momentum:this.momentum,useNesterov:this.useNesterov}}static fromConfig(e,t){return new e(t.learningRate,t.momentum,t.useNesterov)}};_f.className="Momentum";Co(_f);var Df=class extends Yr{constructor(e,t=.9,n=0,s=null,r=!1){super();if(this.learningRate=e,this.decay=t,this.momentum=n,this.epsilon=s,this.accumulatedMeanSquares=[],this.accumulatedMoments=[],this.accumulatedMeanGrads=[],this.centered=r,s==null&&(this.epsilon=W.backend.epsilon()),e==null)throw new Error("learningRate for RMSPropOptimizer must be defined.")}applyGradients(e){(Array.isArray(e)?e.map(n=>n.name):Object.keys(e)).forEach((n,s)=>{let r=W.registeredVariables[n],a=!1;this.accumulatedMeanSquares[s]==null&&(this.accumulatedMeanSquares[s]={originalName:`${n}/rms`,variable:q(()=>Je(r).variable(a))}),this.accumulatedMoments[s]==null&&(this.accumulatedMoments[s]={originalName:`${n}/momentum`,variable:q(()=>Je(r).variable(a))}),this.accumulatedMeanGrads[s]==null&&this.centered&&(this.accumulatedMeanGrads[s]={originalName:`${n}/mg`,variable:q(()=>Je(r).variable(a))});let o=Array.isArray(e)?e[s].tensor:e[n];if(o==null)return;let i=this.accumulatedMeanSquares[s].variable,l=this.accumulatedMoments[s].variable;q(()=>{let c=le(L(i,this.decay),L(gt(o),1-this.decay));if(this.centered){let u=this.accumulatedMeanGrads[s].variable,d=le(L(u,this.decay),L(o,1-this.decay)),p=pe(L(o,this.learningRate),_n(me(c,le(gt(d),this.epsilon)))),h=le(L(l,this.momentum),p);i.assign(c),u.assign(d),l.assign(h);let f=me(r,h);r.assign(f)}else{let u=le(L(i,this.decay),L(gt(o),1-this.decay)),d=le(L(l,this.momentum),pe(L(o,this.learningRate),_n(le(u,this.epsilon))));i.assign(u),l.assign(d);let p=me(r,d);r.assign(p)}})}),this.incrementIterations()}dispose(){this.accumulatedMeanSquares!=null&&se(this.accumulatedMeanSquares.map(e=>e.variable)),this.accumulatedMeanGrads!=null&&this.centered&&se(this.accumulatedMeanGrads.map(e=>e.variable)),this.accumulatedMoments!=null&&se(this.accumulatedMoments.map(e=>e.variable))}async getWeights(){let e=[...this.accumulatedMeanSquares,...this.accumulatedMoments];return this.centered&&e.push(...this.accumulatedMeanGrads),[await this.saveIterations()].concat(e.map(t=>({name:t.originalName,tensor:t.variable})))}async setWeights(e){e=await this.extractIterations(e);let t=this.centered?e.length/3:e.length/2,n=!1;this.accumulatedMeanSquares=e.slice(0,t).map(s=>({originalName:s.name,variable:s.tensor.variable(n)})),this.accumulatedMoments=e.slice(t,t*2).map(s=>({originalName:s.name,variable:s.tensor.variable(n)})),this.centered&&(this.accumulatedMeanGrads=e.slice(t*2,t*3).map(s=>({originalName:s.name,variable:s.tensor.variable(n)})))}getConfig(){return{learningRate:this.learningRate,decay:this.decay,momentum:this.momentum,epsilon:this.epsilon,centered:this.centered}}static fromConfig(e,t){return new e(t.learningRate,t.decay,t.momentum,t.epsilon,t.centered)}};Df.className="RMSProp";Co(Df);var _o=class{static sgd(e){return new Od(e)}static momentum(e,t,n=!1){return new _f(e,t,n)}static rmsprop(e,t=.9,n=0,s=null,r=!1){return new Df(e,t,n,s,r)}static adam(e=.001,t=.9,n=.999,s=null){return new Rf(e,t,n,s)}static adadelta(e=.001,t=.95,n=null){return new Nf(e,t,n)}static adamax(e=.002,t=.9,n=.999,s=null,r=0){return new $f(e,t,n,s,r)}static adagrad(e,t=.1){return new Ef(e,t)}},cl={sgd:_o.sgd,momentum:_o.momentum,adadelta:_o.adadelta,adagrad:_o.adagrad,rmsprop:_o.rmsprop,adamax:_o.adamax,adam:_o.adam},kO=(()=>typeof requestAnimationFrame!="undefined"?requestAnimationFrame:typeof setImmediate!="undefined"?setImmediate:e=>e())();function Gv(){return new Promise(e=>kO(()=>e()))}var E={};Oe(E,{ERF_A1:()=>PO,ERF_A2:()=>FO,ERF_A3:()=>OO,ERF_A4:()=>MO,ERF_A5:()=>zO,ERF_P:()=>DO,PARALLELIZE_THRESHOLD:()=>k1,SELU_SCALE:()=>jv,SELU_SCALEALPHA:()=>Hv,applyActivation:()=>If,assertAndGetBroadcastShape:()=>mt,assertAxesAreInnerMostDims:()=>F_,assertParamsConsistent:()=>SO,assignToTypedArray:()=>GO,axesAreInnerMostDims:()=>e1,calculateShapes:()=>g3,checkEinsumDimSizes:()=>ZO,combineLocations:()=>dv,complexWithEvenIndex:()=>WO,complexWithOddIndex:()=>VO,computeConv2DInfo:()=>Sd,computeConv3DInfo:()=>V3,computeDefaultPad:()=>W2,computeDilation2DInfo:()=>t$,computeOptimalWindowSize:()=>CO,computeOutAndReduceShapes:()=>pv,computeOutShape:()=>IO,computePool2DInfo:()=>W3,computePool3DInfo:()=>n$,convertConv2DDataFormat:()=>U3,decodeEinsumEquation:()=>XO,eitherStridesOrDilationsAreOne:()=>Nr,expandShapeToKeepDim:()=>ul,exponent:()=>jO,exponents:()=>HO,fromStringArrayToUint8:()=>aM,fromUint8ToStringArray:()=>rM,getAxesPermutation:()=>hv,getBroadcastDims:()=>p3,getComplexWithIndex:()=>UO,getEinsumComputePath:()=>YO,getEinsumPermutation:()=>KO,getFusedBiasGradient:()=>Sf,getFusedDyActivation:()=>kf,getImageCenter:()=>TO,getInnerMostAxes:()=>O_,getPermuted:()=>EO,getReductionAxes:()=>qt,getReshaped:()=>NO,getReshapedPermuted:()=>RO,getSliceBeginCoords:()=>$O,getSliceSize:()=>_O,getUndoAxesPermutation:()=>t1,isIdentityPermutation:()=>JO,log:()=>EE,mergeRealAndImagArrays:()=>LO,prepareAndValidate:()=>m3,prepareSplitSize:()=>eM,segment_util:()=>Kv,shouldFuse:()=>Cf,slice_util:()=>Pt,splitRealAndImagArrays:()=>BO,tupleValuesAreOne:()=>To,upcastType:()=>zn,validateInput:()=>D2,validateUpdateShape:()=>_2,warn:()=>vo});function SO(e,t){let n=e[0].length;e.forEach((r,a)=>{O(r.length===n,()=>`Error in concat${n}D: rank of tensors[${a}] must be the same as the rank of the rest (${n})`)}),O(t>=0&&t<n,()=>`Error in concat${n}D: axis must be between 0 and ${n-1}.`);let s=e[0];e.forEach((r,a)=>{for(let o=0;o<n;o++)O(o===t||r[o]===s[o],()=>`Error in concat${n}D: Shape of tensors[${a}] (${r}) does not match the shape of the rest (${s}) along the non-concatenated axis ${a}.`)})}function IO(e,t){let n=e[0].slice();for(let s=1;s<e.length;s++)n[t]+=e[s][t];return n}var k1=30;function CO(e){return e<=k1?e:hh(e,Math.floor(Math.sqrt(e)))}function TO(e,t,n){let s=n*(typeof e=="number"?e:e[0]),r=t*(typeof e=="number"?e:e[1]);return[s,r]}function NO(e,t,n,s=!0){let r=[];if(s)r=r.concat(t.slice(0)),r.push(e[0]/n),r=r.concat(e.slice(1));else{r=r.concat(e[0]);let a=t.length;for(let o=0;o<a;++o)r=r.concat([e[o+1]/t[o],t[o]]);r=r.concat(e.slice(a+1))}return r}function EO(e,t,n=!0){let s=[];if(n){s.push(t);for(let r=t+1;r<e;++r)r<=2*t?(s.push(r),s.push(r-(t+1))):s.push(r)}else{let r=[],a=[];for(let o=1;o<e;++o)o>=t*2+1||o%2==1?a.push(o):r.push(o);s.push(...r),s.push(0),s.push(...a)}return s}function RO(e,t,n,s=!0){let r=[];s?r.push(e[0]/n):r.push(e[0]*n);for(let a=1;a<e.length;++a)a<=t.length?s?r.push(t[a-1]*e[a]):r.push(e[a]/t[a-1]):r.push(e[a]);return r}function $O(e,t){let n=[0];for(let s=0;s<t;++s)n.push(e[s][0]);return n}function _O(e,t,n){let s=e.slice(0,1);for(let r=0;r<n;++r)s.push(e[r+1]-t[r][0]-t[r][1]);return s}var Hv=1.7580993408473768,jv=1.0507009873554805,DO=.3275911,PO=.254829592,FO=-.284496736,OO=1.421413741,MO=-1.453152027,zO=1.061405429;function LO(e,t){if(e.length!==t.length)throw new Error(`Cannot merge real and imag arrays of different lengths. real:${e.length}, imag: ${t.length}.`);let n=new Float32Array(e.length*2);for(let s=0;s<n.length;s+=2)n[s]=e[s/2],n[s+1]=t[s/2];return n}function BO(e){let t=new Float32Array(e.length/2),n=new Float32Array(e.length/2);for(let s=0;s<e.length;s+=2)t[s/2]=e[s],n[s/2]=e[s+1];return{real:t,imag:n}}function WO(e){let t=Math.ceil(e.length/4),n=new Float32Array(t),s=new Float32Array(t);for(let r=0;r<e.length;r+=4)n[Math.floor(r/4)]=e[r],s[Math.floor(r/4)]=e[r+1];return{real:n,imag:s}}function VO(e){let t=Math.floor(e.length/4),n=new Float32Array(t),s=new Float32Array(t);for(let r=2;r<e.length;r+=4)n[Math.floor(r/4)]=e[r],s[Math.floor(r/4)]=e[r+1];return{real:n,imag:s}}function UO(e,t){let n=e[t*2],s=e[t*2+1];return{real:n,imag:s}}function GO(e,t,n,s){e[s*2]=t,e[s*2+1]=n}function HO(e,t){let n=new Float32Array(e/2),s=new Float32Array(e/2);for(let r=0;r<Math.ceil(e/2);r++){let a=(t?2:-2)*Math.PI*(r/e);n[r]=Math.cos(a),s[r]=Math.sin(a)}return{real:n,imag:s}}function jO(e,t,n){let s=(n?2:-2)*Math.PI*(e/t),r=Math.cos(s),a=Math.sin(s);return{real:r,imag:a}}var S1="->",qO=/->/g,qv=",",Xv="...";function XO(e,t){e=e.replace(/\s/g,"");let n=(e.length-e.replace(qO,"").length)/S1.length;if(n<1)throw new Error("Equations without an arrow are not supported.");if(n>1)throw new Error(`Equation must contain exactly one arrow ("${S1}").`);let[s,r]=e.split(S1);O(s.indexOf(Xv)===-1,()=>`The ellipsis notation ("${Xv}") is not supported yet.`);let a=s.split(qv),o=a.length;if(t!==o)throw new Error(`Expected ${o} input tensors, received ${t}`);if(o>2)throw new Error("Support for more than 2 input tensors is not implemented yet.");let i=[];for(let p=0;p<r.length;++p){let h=r[p];if(!a.some(f=>f.indexOf(h)!==-1))throw new Error(`Output subscripts contain the label ${h} not present in the input subscripts.`);i.indexOf(h)===-1&&i.push(h)}for(let p=0;p<s.length;++p){let h=s[p];i.indexOf(h)===-1&&h!==qv&&i.push(h)}let l=new Array(a.length);for(let p=0;p<o;++p){if(new Set(a[p].split("")).size!==a[p].length)throw new Error(`Found duplicate axes in input component ${a[p]}. Support for duplicate axes in input is not implemented yet.`);l[p]=[];for(let h=0;h<a[p].length;++h)l[p].push(i.indexOf(a[p][h]))}let c=i.length,u=r.length,d=[];for(let p=u;p<c;++p)d.push(p);return{allDims:i,summedDims:d,idDims:l}}function KO(e,t){let n=new Array(e);n.fill(-1);for(let r=0;r<t.length;++r)n[t[r]]=r;let s=[];for(let r=0;r<e;++r)n[r]===-1&&s.push(r);return n=n.filter(r=>r!==-1),{permutationIndices:n,expandDims:s}}function ZO(e,t,n){let s=new Array(e);for(let r=0;r<n.length;++r){let a=n[r].shape;for(let o=0;o<t[r].length;++o)s[t[r][o]]===void 0?s[t[r][o]]=a[o]:O(s[t[r][o]]===a[o],()=>`Expected dimension ${s[t[r][o]]} at axis ${o} of input shaped ${JSON.stringify(a)}, but got dimension ${a[o]}`)}}function YO(e,t){let n=e,s=[],r=0;e.length===0&&n.push(-1),r=e.length+1;for(let o=0;o<r;++o)s.push([]);let a=[];for(let o=0;o<n.length;++o){let i=n[o],l=QO(t,i);for(let c of l)a.indexOf(c)===-1&&(s[o].push(c),a.push(c))}return{path:n,steps:s}}function JO(e){return e.every((t,n)=>t===n)}function QO(e,t){let n=[];for(let s=0;s<e.length;++s)(e[s].length===0||e[s].indexOf(t)!==-1||t===-1)&&n.push(s);return n}function eM(e,t,n=0){let s=[];if(typeof t=="number")O(e.shape[n]%t==0,()=>"Number of splits must evenly divide the axis."),s=new Array(t).fill(e.shape[n]/t);else{let r=t.reduce((o,i)=>(i===-1&&(o+=1),o),0);O(r<=1,()=>"There should be only one negative value in split array.");let a=t.indexOf(-1);if(a!==-1){let o=t.reduce((i,l)=>l>0?i+l:i);t[a]=e.shape[n]-o}O(e.shape[n]===t.reduce((o,i)=>o+i),()=>"The sum of sizes must match the size of the axis dimension."),s=t}return s}var Kv={};Oe(Kv,{collectGatherOpShapeInfo:()=>sM,computeOutShape:()=>nM,segOpComputeOptimalWindowSize:()=>tM});function tM(e,t){let n=!1,s;for(e<=k1?(s=e,n=!0):s=hh(e,Math.floor(Math.sqrt(e)));!n;)s>t||s===e?n=!0:s=hh(e,s+1);return s}function nM(e,t,n){let s=[],r=e.length;for(let a=0;a<r;a++)a!==t?s.push(e[a]):s.push(n);return s}function sM(e,t,n,s){let r=t.shape.length,a=e.shape.length;if(s!==0&&(s<-r||s>r))throw new Error(`Expect batchDims in the range of [-${r}, ${r}], but got ${s}`);if(s<0&&(s+=r),s>a)throw new Error(`batchDims (${s}) must be less than rank(x) (
|
|
${a}).`);if(n<s)throw new Error(`batchDims (${s}) must be less than or equal to axis (${n}).`);for(let d=0;d<s;++d)if(e.shape[d]!==t.shape[d])throw new Error(`x.shape[${d}]: ${e.shape[d]} should be equal to indices.shape[${d}]: ${t.shape[d]}.`);let o=e.shape[n],i=[],l=1,c=1,u=1;for(let d=0;d<s;++d)i.push(e.shape[d]),l*=e.shape[d];for(let d=s;d<n;d++)i.push(e.shape[d]),c*=e.shape[d];for(let d=s;d<r;d++)i.push(t.shape[d]);for(let d=n+1;d<a;d++)i.push(e.shape[d]),u*=e.shape[d];return{batchSize:l,sliceSize:u,outerSize:c,dimSize:o,outputShape:i}}function rM(e){try{return e.map(t=>Kh(t))}catch(t){throw new Error(`Failed to decode encoded string bytes into utf-8, error: ${t}`)}}function aM(e){return e.map(t=>md(t))}var Xs={};Oe(Xs,{nonMaxSuppressionV3Impl:()=>zv,nonMaxSuppressionV4Impl:()=>Lv,nonMaxSuppressionV5Impl:()=>Bv,whereImpl:()=>Nv});var Zv={kernelName:ci,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(e,Pd(de(n,"float32"),-1))}}},oM={kernelName:eu,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>{let s=gt(de(n,"float32")),r=_n(me(Re(1),s));return Ft(pe(e,r))}}}},iM={kernelName:tu,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>{let s=_n(me(gt(de(n,"float32")),1));return pe(e,s)}}}},lM={kernelName:Gr,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,s]=t,r=mt(n.shape,s.shape);return{a:()=>{let i=e,l=qt(n.shape,r);return l.length>0&&(i=Se(i,l)),G(i,n.shape)},b:()=>{let i=e,l=qt(s.shape,r);return l.length>0&&(i=Se(i,l)),G(i,s.shape)}}}},uM={kernelName:Ca,saveAllInputs:!0,gradFunc:(e,t)=>{let n={};return t.forEach((s,r)=>{n[r]=()=>e.clone()}),n}},cM={kernelName:Ta,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>Je(n)}}},dM={kernelName:ru,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>Je(n)}}},pM={kernelName:au,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>pe(e,_n(me(Re(1),gt(de(n,"float32")))))}}},hM={kernelName:ou,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>{let s=_n(le(Re(1),gt(de(n,"float32"))));return pe(e,s)}}}},fM={kernelName:uu,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,s]=t,r=mt(n.shape,s.shape);return{a:()=>{let i=le(gt(n),gt(s)),l=L(e,pe(s,i)),c=qt(n.shape,r);return c.length>0&&(l=Se(l,c)),G(l,n.shape)},b:()=>{let i=le(gt(n),gt(s)),l=Ft(L(e,pe(n,i))),c=qt(s.shape,r);return c.length>0&&(l=Se(l,c)),G(l,s.shape)}}}},mM={kernelName:iu,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>pe(e,le(gt(de(n,"float32")),1))}}},gM={kernelName:lu,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>pe(e,me(Re(1),gt(de(n,"float32"))))}}};function AM(e,t,n,s,r,a){let o=D(e,"dy","avgPool3dGrad"),i=D(t,"input","avgPool3dGrad"),l=o,c=i,u=!1;i.rank===4&&(u=!0,l=G(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]]),c=G(i,[1,i.shape[0],i.shape[1],i.shape[2],i.shape[3]])),O(l.rank===5,()=>`Error in avgPool3dGrad: dy must be rank 5 but got rank ${l.rank}.`),O(c.rank===5,()=>`Error in avgPool3dGrad: input must be rank 5 but got rank ${c.rank}.`),a!=null&&O(pn(r),()=>`Error in avgPool3dGrad: pad must be an integer when using, dimRoundingMode ${a} but got pad ${r}.`);let d={dy:l,input:c},p={filterSize:n,strides:s,pad:r,dimRoundingMode:a},h=W.runKernel(gh,d,p);return u?G(h,[h.shape[1],h.shape[2],h.shape[3],h.shape[4]]):h}var yM=V({avgPool3dGrad_:AM}),xM={kernelName:Yc,inputsToSave:["x"],gradFunc:(e,t,n)=>{let[s]=t,{filterSize:r,strides:a,pad:o,dimRoundingMode:i}=n;return{x:()=>yM(e,s,r,a,o,i)}}};function bM(e,t,n,s,r){let a=D(e,"dy","avgPoolGrad"),o=D(t,"input","avgPoolGrad");O(o.rank===a.rank,()=>`Rank of input (${o.rank}) does not match rank of dy (${a.rank})`);let i=o,l=a,c=!1;o.rank===3&&(c=!0,i=G(o,[1,o.shape[0],o.shape[1],o.shape[2]]),l=G(a,[1,a.shape[0],a.shape[1],a.shape[2]])),O(l.rank===4,()=>`Error in avgPoolGrad: dy must be rank 4 but got rank ${l.rank}.`),O(i.rank===4,()=>`Error in avgPoolGrad: input must be rank 4 but got rank ${i.rank}.`);let u={dy:l,input:i},d={filterSize:n,strides:s,pad:r},p=W.runKernel(mh,u,d);return c?G(p,[p.shape[1],p.shape[2],p.shape[3]]):p}var vM=V({avgPoolGrad_:bM}),wM={kernelName:Na,inputsToSave:["x"],gradFunc:(e,t,n)=>{let[s]=t,{filterSize:r,strides:a,pad:o}=n;return{x:()=>vM(e,s,r,a,o)}}},kM={kernelName:Ea,inputsToSave:["a","b"],gradFunc:(e,t,n)=>{let[s,r]=t,{transposeA:a,transposeB:o}=n;return!a&&!o?{a:()=>Ue(e,r,!1,!0),b:()=>Ue(s,e,!0,!1)}:!a&&o?{a:()=>Ue(e,r,!1,!1),b:()=>Ue(e,s,!0,!1)}:a&&!o?{a:()=>Ue(r,e,!1,!0),b:()=>Ue(s,e,!1,!1)}:{a:()=>Ue(r,e,!0,!0),b:()=>Ue(e,s,!0,!0)}}},SM={kernelName:di,gradFunc:(e,t,n)=>{let{blockShape:s,crops:r}=n;return{x:()=>Af(e,s,r)}}},IM={kernelName:$5,gradFunc:(e,t,n)=>{let s=n,r=s.inputShape,a=s.shape,o=Array.from(a);for(let l=r.length-1;l>=0;l--)if(r[l]===a[l])o[l]=1;else if(r[l]!==1)throw new Error(`broadcastTo(): [${r}] cannot be broadcast to [${a}].`);let i=[];for(let l=0;l<o.length;l++)o[l]>1&&i.push(l);return{x:()=>Se(e,i,!0)}}},CM={kernelName:Ra,gradFunc:e=>({x:()=>e.clone()})},TM={kernelName:$a,gradFunc:e=>({x:()=>Je(e)})},NM={kernelName:Hr,inputsToSave:["x"],gradFunc:(e,t,n)=>{let[s]=t,{clipValueMin:r,clipValueMax:a}=n;return{x:()=>Bn(ir(il(s,r),ll(s,a)),e,Je(e))}}},EM={kernelName:Qc,inputsToSave:["x"],gradFunc:Zv.gradFunc},RM={kernelName:pi,saveAllInputs:!0,gradFunc:(e,t,n)=>{let s=t.map(l=>l.shape),{axis:r}=n,a=Us(r,t[0].shape)[0],o=s.map(l=>l[a]);return nn(e,o,a).map(l=>()=>l)}},$M={kernelName:_a,inputsToSave:["x","filter"],gradFunc:(e,t,n)=>{let[s,r]=t,{dilations:a,strides:o,pad:i,dataFormat:l}=n;return O(To(a),()=>`Error in gradient of conv2D: dilation rates greater than 1 are not yet supported in gradients. Got dilations '${a}'`),{x:()=>j2(s.shape,e,r,o,i,l),filter:()=>v1(s,e,r.shape,o,i,l)}}},_M={kernelName:Da,inputsToSave:["dy","filter"],gradFunc:(e,t,n)=>{let[s,r]=t,{strides:a,pad:o,dataFormat:i,dimRoundingMode:l}=n;return{dy:()=>No(e,r,a,o,i,1,l),filter:()=>v1(e,s,r.shape,a,o,i,l)}}};function DM(e,t,n,s,r){let a=e;e.rank===4&&(a=G(e,[1,e.shape[0],e.shape[1],e.shape[2],e.shape[3]]));let o=t;o.rank===4&&(o=G(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]])),O(a.rank===5,()=>`Error in conv3dDerFilter: input must be rank 5, but got shape ${a.shape}.`),O(o.rank===5,()=>`Error in conv3dDerFilter: dy must be rank 5, but got shape ${o.shape}.`),O(n.length===5,()=>`Error in conv3dDerFilter: filterShape must be length 5, but got ${n}.`),O(a.shape[4]===n[3],()=>`Error in conv3dDerFilter: depth of input ${a.shape[4]}) must match input depth in filter (${n[3]}.`),O(o.shape[4]===n[4],()=>`Error in conv3dDerFilter: depth of dy (${o.shape[4]}) must match output depth for filter (${n[4]}).`);let i={x:a,dy:o},l={strides:s,pad:r,filterShape:n};return W.runKernel(bh,i,l)}var PM=V({conv3DBackpropFilter_:DM}),FM={kernelName:ed,inputsToSave:["x","filter"],gradFunc:(e,t,n)=>{let{dilations:s,strides:r,pad:a}=n;O(To(s),()=>`Error in gradient of conv3D: dilation rates greater than 1 are not yet supported in gradients. Got dilations '${s}'`);let[o,i]=t;return{x:()=>J3(o.shape,e,i,r,a),filter:()=>PM(o,e,i.shape,r,a)}}},OM={kernelName:Pa,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(Ft(d1(de(n,"float32"))),e)}}},MM={kernelName:Fa,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(p1(de(n,"float32")),e)}}},zM={kernelName:hi,inputsToSave:["x"],gradFunc:(e,t,n)=>{let[s]=t,{axis:r,exclusive:a,reverse:o}=n;return{x:()=>{let i=hv([r],s.rank),l=Z2(e,r,a,!o);return i!=null&&(l=Ye(l,i)),l}}}},LM={kernelName:Oa,inputsToSave:["x","filter"],gradFunc:(e,t,n)=>{let{dilations:s,strides:r,pad:a,dimRoundingMode:o}=n,i=s==null?[1,1]:s;O(To(i),()=>`Error in gradient of depthwiseConv2dNative: dilation rates greater than 1 are not yet supported. Got dilations '${i}'`);let[l,c]=t;return O(l.rank===4,()=>`Error in gradient of depthwiseConv2dNative: input must be rank 4, but got rank ${l.rank}.`),O(c.rank===4,()=>`Error in gradient of depthwiseConv2dNative: filter must be rank 4, but got rank ${c.rank}.`),O(l.shape[3]===c.shape[2],()=>`Error in gradient of depthwiseConv2d: number of input channels (${l.shape[3]}) must match the inChannels dimension in filter ${c.shape[2]}.`),O(Nr(r,i),()=>`Error in gradient of depthwiseConv2d: Either strides or dilations must be 1. Got strides ${r} and dilations '${i}'.`),o!=null&&O(pn(a),()=>`Error in depthwiseConv2d: pad must be an integer when using, dimRoundingMode ${o} but got pad ${a}.`),{x:()=>Fv(l.shape,e,c,r,a,i,o),filter:()=>Pv(l,e,c.shape,r,a,i,o)}}},BM={kernelName:td,inputsToSave:["x","filter"],gradFunc:(e,t,n)=>{let[s,r]=t,a={x:s,filter:r,dy:e},o={x:s,filter:r,dy:e};return{x:()=>W.runKernel(Ch,a,n),filter:()=>W.runKernel(Th,o,n)}}},WM={kernelName:za,outputsToSave:[!0],gradFunc:(e,t)=>{let[n]=t,s={dy:e,y:n};return{x:()=>W.runKernel(Nh,s)}}},VM={kernelName:cu,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t,s=L(Ts(Ft(gt(n))),2/Math.sqrt(Math.PI));return{x:()=>L(e,s)}}},UM={kernelName:La,outputsToSave:[!0],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(e,n)}}},GM={kernelName:Ai,inputsToSave:["input"],gradFunc:(e,t)=>{let[n]=t;return{input:()=>G(e,n.shape)}}},HM={kernelName:yi,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(e,Ts(n))}}},jM={kernelName:Ba,gradFunc:e=>({x:()=>Je(e)})},qM={kernelName:Wa,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,s]=t,r=mt(n.shape,s.shape);return{a:()=>{let i=pe(e,de(s,"float32")),l=qt(n.shape,r);return l.length>0?G(Se(i,l),n.shape):i},b:()=>{let i=L(e,de(n,"float32")),l=qt(s.shape,r);l.length>0&&(i=G(Se(i,l),s.shape));let c=gt(s);return Ft(pe(i,de(c,"float32")))}}}},XM={kernelName:Va,inputsToSave:["x","mean","variance","scale"],gradFunc:(e,t,n)=>{let{varianceEpsilon:s}=n,[r,a,o,i]=t,l=i==null?Re(1):i,c=qt(a.shape,r.shape),u=[];if(a.rank===1){for(let b=0;b<r.shape.length-1;++b)u.push(r.shape[b]);u.push(1)}let d=me(r,a),p=L(e,l),h=u1(le(o,Re(s))),f=L(L(L(h,h),h),Re(-.5));return{x:()=>a.rank===1?G(L(L(e,js(G(h,[1,1,1,a.shape[0]]),u)),l),r.shape):G(L(L(e,h),l),r.shape),mean:()=>{let b=L(L(h,Re(-1)),p);return a.rank===1&&(b=Se(b,c)),G(b,a.shape)},variance:()=>{let b=L(L(f,d),p);return a.rank===1&&(b=Se(b,c)),G(b,a.shape)},scale:()=>{let b=L(d,h),w=L(e,b);return a.rank===1&&(w=Se(w,c)),G(w,a.shape)},offset:()=>{let b=e;return a.rank===1&&(b=Se(b,c)),G(b,a.shape)}}}},KM={kernelName:bi,inputsToSave:["x","indices"],gradFunc:(e,t,n)=>{let[s,r]=t,{axis:a}=n,o=Us(a,s.shape)[0];return{x:()=>{let l=s.shape,c=r.size,u=l.slice(0,o),d=u.length,p=l.slice(a,l.length).slice(1),h=p.length,f=Yv(0,d),m=Yv(d+1,d+1+h),g=Jv([u,[c],p]),A=G(e,g),x=G(r,[c]),y=Jv([[d],f,m]),b=Ye(A,y),w=Cv(b,x,s.shape[o]),k=t1(y);return w=Ye(w,k),w},indices:()=>r}}};function Yv(e,t){let n=[];for(let s=e;s<t;++s)n.push(s);return n}function Jv(e){let t=[];for(let n=0;n<e.length;++n)for(let s=0;s<e[n].length;++s)t.push(e[n][s]);return t}var ZM={kernelName:Ua,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,s]=t;return{a:()=>Je(n),b:()=>Je(s)}}},YM={kernelName:Ga,gradFunc:e=>({x:()=>de(e,"float32")})},JM={kernelName:pu,gradFunc:e=>({x:()=>Je(e)})},QM={kernelName:hu,gradFunc:e=>({x:()=>Je(e)})},ez={kernelName:fu,gradFunc:e=>({x:()=>Je(e)})},tz={kernelName:ki,inputsToSave:["x"],gradFunc:(e,t,n)=>{let[s]=t,{alpha:r}=n,a=ps(s,0);return{x:()=>Bn(a,e,L(e,r))}}},nz={kernelName:mu,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>pe(e,le(n,1))}}},sz={kernelName:Ha,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>pe(e,de(n,"float32"))}}},rz={kernelName:_5,inputsToSave:[],outputsToSave:[!0],gradFunc:(e,t,n)=>{let[s]=t,{axis:r}=n;return{logits:()=>{let a=!0,o=Ts(s);return me(e,L(Se(e,r,a),o))}}}};function az(e,t,n,s=5,r=1,a=1,o=.5){let i={x:e,y:t,dy:n},l={depthRadius:s,bias:r,alpha:a,beta:o};return W.runKernel(_h,i,l)}var oz=V({localResponseNormalizationBackprop_:az}),iz={kernelName:ad,inputsToSave:["x"],outputsToSave:[!0],gradFunc:(e,t,n)=>{let[s,r]=t,{depthRadius:a,bias:o,alpha:i,beta:l}=n;return{x:()=>oz(s,r,e,a,o,i,l)}}};function Qv(e,t,n,s){return t.rank<n.rank&&(t=G(t,ul(t.shape,s))),e.rank<n.rank&&(e=G(e,ul(e.shape,s))),{x:()=>L(e,de(Cs(n,t),e.dtype))}}var ew={kernelName:ja,inputsToSave:["x"],outputsToSave:[!0],gradFunc:(e,t,n)=>{let s=n,{reductionIndices:r}=s,a=t[0],o=t[1],i=Us(r,a.shape),l=Qv(e,o,a,i);return{x:()=>l.x()}}},lz={kernelName:qa,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,s]=t;return{a:()=>L(e,de(il(n,s),"float32")),b:()=>L(e,de(J2(n,s),"float32"))}}};function uz(e,t,n,s,r,a,o){let i=D(e,"dy","maxPool3dGrad"),l=D(t,"input","maxPool3dGrad"),c=D(n,"output","maxPool3dGrad"),u=i,d=l,p=c,h=!1;l.rank===4&&(h=!0,u=G(i,[1,i.shape[0],i.shape[1],i.shape[2],i.shape[3]]),d=G(l,[1,l.shape[0],l.shape[1],l.shape[2],l.shape[3]]),p=G(c,[1,c.shape[0],c.shape[1],c.shape[2],c.shape[3]])),O(u.rank===5,()=>`Error in maxPool3dGrad: dy must be rank 5 but got rank ${u.rank}.`),O(d.rank===5,()=>`Error in maxPool3dGrad: input must be rank 5 but got rank ${d.rank}.`),O(p.rank===5,()=>`Error in maxPool3dGrad: output must be rank 5 but got rank ${p.rank}.`),o!=null&&O(pn(a),()=>`Error in maxPool3dGrad: pad must be an integer when using, dimRoundingMode ${o} but got pad ${a}.`);let f={dy:u,input:d,output:p},m={filterSize:s,strides:r,pad:a,dimRoundingMode:o},g=W.runKernel(Ph,f,m);return h?G(g,[g.shape[1],g.shape[2],g.shape[3],g.shape[4]]):g}var cz=V({maxPool3dGrad_:uz}),dz={kernelName:od,inputsToSave:["x"],outputsToSave:[!0],gradFunc:(e,t,n)=>{let[s,r]=t,{filterSize:a,strides:o,pad:i,dimRoundingMode:l}=n;return{x:()=>cz(e,s,r,a,o,i,l)}}};function pz(e,t,n,s,r,a,o){let i=D(e,"dy","maxPoolGrad"),l=D(t,"input","maxPoolGrad"),c=D(n,"output","maxPoolGrad");O(l.rank===i.rank,()=>`Rank of input (${l.rank}) does not match rank of dy (${i.rank})`),O(i.rank===4,()=>`Error in maxPoolGrad: dy must be rank 4 but got rank ${i.rank}.`),O(l.rank===4,()=>`Error in maxPoolGrad: input must be rank 4 but got rank ${l.rank}.`),o!=null&&O(pn(a),()=>`Error in maxPoolGrad: pad must be an integer when using, dimRoundingMode ${o} but got pad ${a}.`);let u={dy:i,input:l,output:c},d={filterSize:s,strides:r,pad:a,dimRoundingMode:o};return W.runKernel(Dh,u,d)}var hz=V({maxPoolGrad_:pz}),fz={kernelName:Xa,inputsToSave:["x"],outputsToSave:[!0],gradFunc:(e,t,n)=>{let[s,r]=t,{filterSize:a,strides:o,pad:i}=n;return{x:()=>hz(e,s,r,a,o,i)}}},mz={kernelName:Ka,inputsToSave:["x"],gradFunc:(e,t,n)=>{let[s]=t,{axis:r}=n,a=Us(r,s.shape),i=pv(s.shape,a)[1],l=Vt(i);return{x:()=>{let u=s.shape.slice();a.forEach(h=>{u[h]=1});let d=G(e,u);return pe(L(d,hs(s.shape,"float32")),l)}}}},gz={kernelName:Za,inputsToSave:["x"],outputsToSave:[!0],gradFunc:(e,t,n)=>{let s=n,{axis:r}=s,[a,o]=t,i=Us(r,a.shape),l=Qv(e,o,a,i);return{x:()=>l.x()}}},Az={kernelName:Ya,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,s]=t;return{a:()=>L(e,de(ll(n,s),"float32")),b:()=>L(e,de(ps(n,s),"float32"))}}},yz={kernelName:Ja,inputsToSave:["x"],gradFunc:(e,t,n)=>{let s=t[0],{paddings:r}=n,a=r.map(o=>o[0]);return{x:()=>De(e,a,s.shape)}}},xz={kernelName:Au,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,s]=t,r=mt(n.shape,s.shape);return{a:()=>{let i=qt(n.shape,r);return i.length>0?G(Se(e,i),n.shape):e},b:()=>{let i=L(e,Ft(Nd(pe(n,s)))),l=qt(s.shape,r);return l.length>0?G(Se(i,l),s.shape):i}}}},bz={kernelName:Qa,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,s]=t,r=mt(n.shape,s.shape);return{a:()=>{let i=L(e,de(s,"float32")),l=qt(n.shape,r);return l.length>0?G(Se(i,l),n.shape):i},b:()=>{let i=L(e,de(n,"float32")),l=qt(s.shape,r);return l.length>0?G(Se(i,l),s.shape):i}}}},vz={kernelName:Ti,gradFunc:e=>({x:()=>Ft(e)})},wz={kernelName:_i,inputsToSave:["indices"],gradFunc:(e,t)=>{let n=t[0];return{indices:()=>Gt(n.shape,"float32")}}},kz={kernelName:$i,gradFunc:e=>({x:()=>Je(e)})},Sz={kernelName:Di,saveAllInputs:!0,gradFunc:(e,t,n)=>{let{axis:s}=n;return ss(e,s).map(a=>()=>a)}},tw={kernelName:eo,inputsToSave:["x"],gradFunc:(e,t,n)=>{let s=t[0],{paddings:r}=n,a=r.map(o=>o[0]);return{x:()=>De(e,a,s.shape)}}},Iz={kernelName:to,inputsToSave:["a","b"],outputsToSave:[!0],gradFunc:(e,t)=>{let[n,s,r]=t,a=n,o=s,i=mt(a.shape,o.shape);return{a:()=>{let u=de(o,"float32"),d=L(e,L(u,Ro(a,me(u,Re(1))))),p=qt(a.shape,i);return p.length>0&&(d=Se(d,p)),G(d,a.shape)},b:()=>{let u=ps(a,0),d=Bn(u,Ns(a),Je(a)),p=L(e,L(r,d)),h=qt(o.shape,i);return h.length>0&&(p=Se(p,h)),G(p,o.shape)}}}},Cz={kernelName:no,inputsToSave:["x","alpha"],gradFunc:(e,t)=>{let[n,s]=t,r=ps(n,0);return{x:()=>Bn(r,e,L(e,s)),alpha:()=>{let a=Bn(r,Je(e),L(e,n)),o=qt(s.shape,e.shape);return o.length>0&&(a=Se(a,o)),G(a,s.shape)}}}},Tz={kernelName:Ma,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,s]=t,r=mt(n.shape,s.shape);return{a:()=>{let i=pe(e,de(s,"float32")),l=qt(n.shape,r);return l.length>0?G(Se(i,l),n.shape):i},b:()=>{let i=L(e,de(n,"float32")),l=qt(s.shape,r);l.length>0&&(i=G(Se(i,l),s.shape));let c=gt(s);return Ft(pe(i,de(c,"float32")))}}}},Nz={kernelName:bu,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>pe(e,Ft(gt(n)))}}},Ez={kernelName:ao,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t,s=L(ll(n,6),Pd(n));return{x:()=>L(e,de(s,"float32"))}}},Rz={kernelName:so,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(e,de(Pd(n),"float32"))}}},$z={kernelName:Fi,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>G(e,n.shape)}}},_z={kernelName:ro,inputsToSave:["images"],gradFunc:(e,t,n)=>{let[s]=t,r={dy:e,images:s};return{images:()=>W.runKernel(zh,r,n)}}},Dz={kernelName:vu,inputsToSave:["images"],gradFunc:(e,t,n)=>{let[s]=t,r={dy:e,images:s};return{images:()=>W.runKernel(Mh,r,n)}}},Pz={kernelName:Oi,gradFunc:(e,t,n)=>{let{dims:s}=n,r=Us(s,e.shape);return{x:()=>Rs(e,r)}}},Fz={kernelName:Mi,gradFunc:e=>({x:()=>Je(e)})},Oz={kernelName:oo,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>Ft(pe(e,L(Ro(n,1.5),2)))}}},Mz={kernelName:Li,inputsToSave:["condition"],gradFunc:(e,t)=>{let[n]=t;return{condition:()=>de(Je(n),"float32"),t:()=>L(e,de(n,e.dtype)),e:()=>L(e,de(ff(n),e.dtype))}}},zz={kernelName:wu,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>{let s=ps(n,Re(0)),r=Re(Hv),a=Re(jv),o=L(e,a),i=L(L(e,r),Ts(de(n,"float32")));return Bn(s,o,i)}}}},Lz={kernelName:lo,outputsToSave:[!0],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(e,L(n,me(Re(1),n)))}}},Bz={kernelName:ku,gradFunc:e=>({x:()=>Je(e)})},Wz={kernelName:io,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(uf(de(n,"float32")),e)}}},Vz={kernelName:Wi,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(K2(de(n,"float32")),e)}}},Uz={kernelName:Bi,inputsToSave:["x"],gradFunc:(e,t,n)=>{let[s]=t,{begin:r,size:a}=n,o=s.shape,[i,l]=I3(s,r,a),c=[];for(let u=0;u<e.rank;u++)c.push([i[u],o[u]-i[u]-l[u]]);return{x:()=>qs(e,c)}}},Gz={kernelName:po,outputsToSave:[!0],gradFunc:(e,t,n)=>{let[s]=t,{dim:r}=n,a=!0,o=L(e,s);return{logits:()=>me(o,L(Se(o,[r],a),s))}}},Hz={kernelName:Su,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(e,cs(n))}}},nw={kernelName:Vi,gradFunc:(e,t,n)=>{let{blockShape:s,paddings:r}=n;return{x:()=>lf(e,s,r)}}},sw={kernelName:Ui,gradFunc:(e,t,n)=>{let{axis:s}=n;return{x:()=>vt(e,s)}}},jz={kernelName:uo,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>pe(e,L(_n(de(n,"float32")),2))}}},qz={kernelName:Iu,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(e,L(de(n,"float32"),2))}}},Xz={kernelName:ho,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,s]=t,r=Re(2);return{a:()=>L(e,L(r,me(n,s))),b:()=>L(e,L(r,me(s,n)))}}},Kz={kernelName:Ao,gradFunc:e=>({x:()=>Je(e)})},Zz={kernelName:fo,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,s]=t,r=mt(n.shape,s.shape);return{a:()=>{let i=e,l=qt(n.shape,r);return l.length>0&&(i=Se(i,l)),G(i,n.shape)},b:()=>{let i=e,l=qt(s.shape,r);return l.length>0&&(i=Se(i,l)),G(Ft(i),s.shape)}}}},Yz={kernelName:co,inputsToSave:["x"],gradFunc:(e,t,n)=>{let[s]=t,r=s.shape.slice(),{axis:a}=n;Us(a,s.shape).forEach(c=>{r[c]=1});let i=G(e,r),l=L(i,hs(s.shape,"float32"));return{x:()=>l}}},Jz={kernelName:Hi,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>pe(e,gt(uf(n)))}}},Qz={kernelName:mo,outputsToSave:[!0],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(me(Re(1),gt(n)),e)}}},eL={kernelName:jr,inputsToSave:["x"],gradFunc:(e,t,n)=>{let[s]=t,{reps:r}=n;return{x:()=>{let o=Je(s);if(s.rank===1)for(let i=0;i<r[0];++i)o=le(o,De(e,[i*s.shape[0]],[s.shape[0]]));else if(s.rank===2)for(let i=0;i<r[0];++i)for(let l=0;l<r[1];++l)o=le(o,De(e,[i*s.shape[0],l*s.shape[1]],[s.shape[0],s.shape[1]]));else if(s.rank===3)for(let i=0;i<r[0];++i)for(let l=0;l<r[1];++l)for(let c=0;c<r[2];++c)o=le(o,De(e,[i*s.shape[0],l*s.shape[1],c*s.shape[2]],[s.shape[0],s.shape[1],s.shape[2]]));else if(s.rank===4)for(let i=0;i<r[0];++i)for(let l=0;l<r[1];++l)for(let c=0;c<r[2];++c)for(let u=0;u<r[3];++u)o=le(o,De(e,[i*s.shape[0],l*s.shape[1],c*s.shape[2],u*s.shape[3]],[s.shape[0],s.shape[1],s.shape[2],s.shape[3]]));else throw new Error(`Gradient for tile operation is not implemented for rank-${s.rank} tensors yet.`);return o}}}},tL={kernelName:go,gradFunc:(e,t,n)=>{let s=n,{perm:r}=s,a=t1(r);return{x:()=>Ye(e,a)}}},nL={kernelName:Xi,gradFunc:(e,t,n)=>{let s=n,{axis:r}=s;return{value:()=>An(e,r)}}},sL={kernelName:cd,inputsToSave:["segmentIds"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>rL(e,n)}}};function rL(e,t){let n=Kr(t,Je(t)),s=Ou(e,n),r=il(t,Re(0,"int32")),a=s.rank-r.rank;for(let i=0;i<a;++i)r=mn(r,i+1);r=ir(r,hs(s.shape,"bool"));let o=Je(s);return Bn(r,s,o)}var aL={kernelName:Ki,gradFunc:e=>({x:()=>Je(e)})},oL=[Zv,oM,iM,lM,uM,cM,dM,pM,hM,fM,mM,gM,xM,wM,kM,SM,IM,CM,TM,NM,EM,RM,_M,$M,FM,OM,MM,zM,LM,BM,Tz,WM,VM,UM,GM,HM,qM,jM,XM,KM,ZM,YM,JM,QM,ez,tz,nz,sz,rz,iz,ew,ew,lz,dz,fz,mz,gz,Az,yz,xz,bz,vz,wz,kz,Sz,tw,tw,Iz,Cz,Nz,Ez,Rz,$z,_z,Dz,Pz,Fz,Oz,Mz,zz,Lz,Bz,Wz,Vz,Uz,Gz,Hz,nw,nw,sw,sw,jz,Xz,qz,Kz,Zz,Yz,Jz,Qz,eL,tL,nL,sL,aL];for(let e of oL)D5(e);var rw={};Oe(rw,{maxNorm:()=>cL,minMaxNorm:()=>hL,nonNeg:()=>pL,unitNorm:()=>dL});var I1;function sn(){return I1==null&&(I1=Tr().epsilon()),I1}function ur(){return"channelsLast"}var Jr=class extends Error{constructor(e){super(e);Object.setPrototypeOf(this,Jr.prototype)}},cr=class extends Error{constructor(e){super(e);Object.setPrototypeOf(this,cr.prototype)}},H=class extends Error{constructor(e){super(e);Object.setPrototypeOf(this,H.prototype)}},Le=class extends Error{constructor(e){super(e);Object.setPrototypeOf(this,Le.prototype)}},aw=class extends Error{constructor(e){super(e);Object.setPrototypeOf(this,aw.prototype)}};function dl(e,t){if(Array.isArray(e)){let n=[];for(let s=0;s<t;s++)n=n.concat(e);return n}else{let n=new Array(t);return n.fill(e),n}}function $r(e,t){if(!e)throw new aw(t)}function ow(e,t){let n=0;for(let s of e)s===t&&n++;return n}function rs(e){return e.length===1?e[0]:e}function kt(e){return Array.isArray(e)?e:[e]}function Qr(e){let n=e.replace(/(.)([A-Z][a-z0-9]+)/g,"$1_$2").replace(/([a-z])([A-Z])/g,"$1_$2").toLowerCase();return n[0]!=="_"?n:"private"+n}function pl(e){return e.length<=1||e.indexOf("_")===-1?e:e.replace(/[_]+(\w|$)/g,(t,n)=>n.toUpperCase())}var Ks={};function C1(e){if(e==null)return null;let t={};return t.className=e.getClassName(),t.config=e.getConfig(),t}function T1(e){if(!(e==null||typeof e!="object"))if(Array.isArray(e))e.forEach(t=>T1(t));else{let t=Object.keys(e);for(let n of t){let s=e[n];s!=null&&typeof s=="object"&&(!Array.isArray(s)&&s.type==="ndarray"&&typeof s.value=="number"?e[n]=s.value:T1(s))}}}function Md(e,t={},n={},s="object",r=!1){if(typeof e=="string"){let a=e,o;if(a in n)o=n[a];else if(a in Ks)o=Ks[a];else if(o=t[a],o==null)throw new H(`Unknown ${s}: ${e}. This may be due to one of the following reasons:
|
|
1. The ${s} is defined in Python, in which case it needs to be ported to TensorFlow.js or your JavaScript code.
|
|
2. The custom ${s} is defined in JavaScript, but is not registered properly with tf.serialization.registerClass().`);return o}else{let a=e;if(a.className==null||a.config==null)throw new H(`${s}: Improper config format: ${JSON.stringify(a)}.
|
|
'className' and 'config' must set.`);let o=a.className,i,l;if(o in n?[i,l]=n[o]:o in Ks?[i,l]=Ks.className:o in t&&([i,l]=t[o]),i==null)throw new H(`Unknown ${s}: ${o}. This may be due to one of the following reasons:
|
|
1. The ${s} is defined in Python, in which case it needs to be ported to TensorFlow.js or your JavaScript code.
|
|
2. The custom ${s} is defined in JavaScript, but is not registered properly with tf.serialization.registerClass().`);if(l!=null){let c={};for(let h of Object.keys(Ks))c[h]=Ks[h];for(let h of Object.keys(n))c[h]=n[h];let u=a.config;u.customObjects=c;let d={...Ks};for(let h of Object.keys(n))Ks[h]=n[h];T1(a.config);let p=l(i,a.config,n,r);return Ks={...d},p}else{let c={...Ks};for(let d of Object.keys(n))Ks[d]=n[d];let u=new i(a.config);return Ks={...c},u}}}function iL(e,t){return e<t?-1:e>t?1:0}function Pf(e,t){return-1*iL(e,t)}function Do(e){if(e==null)return e;let t=[];for(let n of e)t.indexOf(n)===-1&&t.push(n);return t}function lL(e){if(e==null)throw new H(`Invalid value in obj: ${JSON.stringify(e)}`);for(let t in e)if(e.hasOwnProperty(t))return!1;return!0}function hl(e,t,n){if(n!=null&&e.indexOf(n)<0)throw new H(`${n} is not a valid ${t}. Valid values are ${e} or null/undefined.`)}function N1(e,t,n=0,s=1/0){return $r(n>=0),$r(s>=n),Array.isArray(e)&&e.length>=n&&e.length<=s&&e.every(r=>typeof r===t)}function yn(e,t){Array.isArray(e)?(v.assert(e.length>0,()=>`${t} is unexpectedly an empty array.`),e.forEach((n,s)=>yn(n,`element ${s+1} of ${t}`))):v.assert(Number.isInteger(e)&&e>0,()=>`Expected ${t} to be a positive integer, but got ${iw(e)}.`)}function iw(e){return e===null?"null":Array.isArray(e)?"["+e.map(t=>iw(t)).join(",")+"]":typeof e=="string"?`"${e}"`:`${e}`}function uL(e,t,n){let s=n!=null?n():v.now(),r;return(...o)=>{let i=n!=null?n():v.now();return i-s<t||(s=i,r=e(...o)),r}}function lw(e){return e==="relu"?"relu":e==="linear"?"linear":e==="elu"?"elu":null}function E1(e,t){return q(()=>_n(Se(L(e,e),t,!0)))}var zd=class extends ue.Serializable{getConfig(){return{}}},R1=class extends zd{constructor(e){super();this.defaultMaxValue=2,this.defaultAxis=0,this.maxValue=e.maxValue!=null?e.maxValue:this.defaultMaxValue,this.axis=e.axis!=null?e.axis:this.defaultAxis}apply(e){return q(()=>{let t=E1(e,this.axis),n=ds(t,0,this.maxValue);return L(e,pe(n,le(sn(),t)))})}getConfig(){return{maxValue:this.maxValue,axis:this.axis}}};R1.className="MaxNorm";ue.registerClass(R1);var $1=class extends zd{constructor(e){super();this.defaultAxis=0,this.axis=e.axis!=null?e.axis:this.defaultAxis}apply(e){return q(()=>pe(e,le(sn(),E1(e,this.axis))))}getConfig(){return{axis:this.axis}}};$1.className="UnitNorm";ue.registerClass($1);var _1=class extends zd{apply(e){return Rr(e)}};_1.className="NonNeg";ue.registerClass(_1);var D1=class extends zd{constructor(e){super();this.defaultMinValue=0,this.defaultMaxValue=1,this.defaultRate=1,this.defaultAxis=0,this.minValue=e.minValue!=null?e.minValue:this.defaultMinValue,this.maxValue=e.maxValue!=null?e.maxValue:this.defaultMaxValue,this.rate=e.rate!=null?e.rate:this.defaultRate,this.axis=e.axis!=null?e.axis:this.defaultAxis}apply(e){return q(()=>{let t=E1(e,this.axis),n=le(L(this.rate,ds(t,this.minValue,this.maxValue)),L(1-this.rate,t));return L(e,pe(n,le(sn(),t)))})}getConfig(){return{minValue:this.minValue,maxValue:this.maxValue,rate:this.rate,axis:this.axis}}};D1.className="MinMaxNorm";ue.registerClass(D1);var uw={maxNorm:"MaxNorm",minMaxNorm:"MinMaxNorm",nonNeg:"NonNeg",unitNorm:"UnitNorm"};function rn(e){return C1(e)}function cw(e,t={}){return Md(e,ue.SerializationMap.getMap().classNameMap,t,"constraint")}function an(e){if(e==null)return null;if(typeof e=="string"){let n={className:e in uw?uw[e]:e,config:{}};return cw(n)}else return e instanceof zd?e:cw(e)}function cL(e){return new R1(e)}function dL(e){return new $1(e)}function pL(){return new _1}function hL(e){return new D1(e)}var dw={};Oe(dw,{constant:()=>OL,glorotNormal:()=>UL,glorotUniform:()=>VL,heNormal:()=>GL,heUniform:()=>HL,identity:()=>BL,leCunNormal:()=>jL,leCunUniform:()=>qL,ones:()=>FL,orthogonal:()=>XL,randomNormal:()=>zL,randomUniform:()=>ML,truncatedNormal:()=>LL,varianceScaling:()=>WL,zeros:()=>PL});var fL=["channelsFirst","channelsLast"],mL=["nearest","bilinear"],gL=["valid","same","causal"],AL=["max","avg"],yL=["sum","mul","concat","ave"],Gu=new Map;function Ht(e){hl(fL,"DataFormat",e)}function xL(e){hl(mL,"InterpolationFormat",e)}function $s(e){hl(gL,"PaddingMode",e)}function pw(e){hl(AL,"PoolMode",e)}var Ld=[],hw="/";function fl(e,t){Ld.push(e);try{let n=t();return Ld.pop(),n}catch(n){throw Ld.pop(),n}}function bL(){return Ld.length===0?"":Ld.join(hw)+hw}function fw(e){if(!gw(e))throw new Error("Not a valid tensor name: '"+e+"'");return bL()+e}function mw(e){if(!gw(e))throw new Error("Not a valid tensor name: '"+e+"'");Gu.has(e)||Gu.set(e,0);let t=Gu.get(e);if(Gu.set(e,Gu.get(e)+1),t>0){let n=`${e}_${t}`;return Gu.set(n,1),n}else return e}var vL=new RegExp(/^[A-Za-z0-9][-A-Za-z0-9\._\/]*$/);function gw(e){return!!e.match(vL)}function wL(e){return e===parseInt(e.toString(),10)}function Po(e,t,n){t==null&&(t=0),n==null&&(n=e.length);let s=1;for(let r=t;r<n;++r)s*=e[r];return s}function Hu(e){if(e.length===0)return Number.NaN;let t=Number.POSITIVE_INFINITY;for(let n=0;n<e.length;n++){let s=e[n];s<t&&(t=s)}return t}function Fo(e){if(e.length===0)return Number.NaN;let t=Number.NEGATIVE_INFINITY;for(let n=0;n<e.length;n++){let s=e[n];s>t&&(t=s)}return t}function dr(e,t){if(t<e)throw new H(`end (${t}) < begin (${e}) is forbidden.`);let n=[];for(let s=e;s<t;++s)n.push(s);return n}function Ff(e,t){return de(e,t)}function Bd(e,t=-1){let n=e.shape.slice();return t<0&&(t=n.length+t+1),n.splice(t,0,1),G(e,n)}function kL(e,t){return q(()=>{if(e.shape.length!==2)throw new H(`repeat() expects a rank-2 tensor, but received a rank-${e.shape.length} tensor.`);let n=Bd(e,1);return O1(n,[1,t,1])})}function SL(e){let t=[Po(e.shape)];return G(e,t)}function IL(e){if(e.rank<=1)throw new H(`batchFlatten requires a minimum rank of 2. Got rank: ${e.rank}.`);let t=[e.shape[0],Po(e.shape,1)];return G(e,t)}function ml(e,t,n){return q(()=>{switch(e.rank){case 1:return xf(e,t,n);case 2:return h1(e,[t,0],[n,e.shape[1]]);case 3:return Wu(e,[t,0,0],[n,e.shape[1],e.shape[2]]);case 4:return _d(e,[t,0,0,0],[n,e.shape[1],e.shape[2],e.shape[3]]);case 5:return De(e,[t,0,0,0,0],[n,e.shape[1],e.shape[2],e.shape[3],e.shape[4]]);case 6:return De(e,[t,0,0,0,0,0],[n,e.shape[1],e.shape[2],e.shape[3],e.shape[4],e.shape[5]]);default:throw new H(`sliceAlongFirstAxis() received an unsupported tensor rank: ${e.rank}`)}})}function P1(e,t,n){return q(()=>{switch(e.rank){case 1:return xf(e,t,n);case 2:return h1(e,[0,t],[e.shape[0],n]);case 3:return Wu(e,[0,0,t],[e.shape[0],e.shape[1],n]);case 4:return _d(e,[0,0,0,t],[e.shape[0],e.shape[1],e.shape[2],n]);default:throw new H(`sliceAlongLastAxis() received an unsupported tensor rank: ${e.rank}`)}})}function Of(e,t,n,s){return q(()=>{switch(e.rank){case 1:return xf(e,t,n);case 2:switch(s){case 1:return ml(e,t,n);case 2:return P1(e,t,n);default:throw new H(`The axis is not within the rank of the tensor ${s}`)}case 3:switch(s){case 1:return ml(e,t,n);case 2:return Wu(e,[0,t,0],[e.shape[0],n,e.shape[2]]);case 3:return P1(e,t,n);default:throw new H(`The axis is not within the rank of the tensor ${s}`)}case 4:switch(s){case 1:return ml(e,t,n);case 2:return _d(e,[0,t,0,0],[e.shape[0],n,e.shape[2],e.shape[3]]);case 3:return _d(e,[0,0,t,0],[e.shape[0],e.shape[1],n,e.shape[3]]);case 4:return P1(e,t,n);default:throw new H(`The axis is not within the rank of the tensor ${s}`)}default:throw new H(`sliceAlongLastAxis() received an unsupported tensor rank: ${e.rank}`)}})}function F1(e,t=-1){let n;return t<0&&(n=e[0].rank,n!==0?t=n:t=0),t===e[0].rank&&(t=-1),vt(e,t)}function Aw(e,t){switch(e.rank){case 1:return K3([e,t]);case 2:return Pu([e,t],0);case 3:return Z3([e,t],0);case 4:return Y3([e,t],0);default:throw new H(`concatAlongFirstAxis() received an unsupported tensor rank: ${e.rank}`)}}function O1(e,t){if(Array.isArray(t)||(t=[t]),e.rank!==t.length)throw new H(`The length of input n (${t.length}) does not match the number of dimensions in input x (${e.rank})`);return js(e,t)}function Mf(e,t=0,n=1,s,r){return yv(e,t,n,s,r)}function _r(e,t,n,s){if(e.rank<2||t.rank<2)throw new Le(`dot requires both inputs to be rank >= 2 but got x shape = ${e.shape} and y shape = ${t.shape}`);if(t.rank>=3){let r=e.shape.slice(-1)[0],a=t.shape.slice(-2)[0];if(r!==a)throw new Le(`If rank y >= 3, then the second last dim of y must equal the last dim of x but got x shape = ${e.shape} and y shape = ${t.shape}`)}if(e.rank===2&&t.rank===2){let r=!1,a=!1;return $o.matMul({a:e,b:t,transposeA:r,transposeB:a,bias:s?M1(e.rank,s,ur()):null,activation:n})}else{let r=e.shape.slice(),a=r.pop();e=G(e,[-1,a]);let o=t.shape.slice(),i=o.pop(),l=o.pop(),c=[...o,i],u=Array.from({length:t.rank},(f,m)=>m===0?t.rank-2:m<=t.rank-2?m-1:m);t=G(Ye(t,u),[l,-1]);let d=[...r,...c],p=!1,h=!1;return G($o.matMul({a:e,b:t,transposeA:p,transposeB:h,bias:s?M1(e.rank,s,ur()):null,activation:n}),d)}}function yw(e,t,n){return q(()=>(Array.isArray(t)?t=Xt(t,"int32"):t=de(t,"int32"),Ou(e,t,n)))}function Wd(e){return L(e,e)}function M1(e,t,n){let s=t.shape;if(t.rank!==1&&t.rank!==e)throw new H(`Unexpected bias dimensions: ${t.rank}; expected it to be 1 or ${e}`);if(e===5){if(n==="channelsFirst")return s.length===1?G(t,[1,s[0],1,1,1]):G(t,[1,s[3],s[0],s[1],s[2]]);if(n==="channelsLast")return s.length===1?G(t,[1,1,1,1,s[0]]):G(t,[1].concat(s))}else if(e===4){if(n==="channelsFirst")return s.length===1?G(t,[1,s[0],1,1]):G(t,[1,s[2],s[0],s[1]]);if(n==="channelsLast")return s.length===1?G(t,[1,1,1,s[0]]):G(t,[1].concat(s))}else if(e===3){if(n==="channelsFirst")return s.length===1?G(t,[1,s[0],1]):G(t,[1,s[1],s[0]]);if(n==="channelsLast")return s.length===1?G(t,[1,1,s[0]]):G(t,[1].concat(s))}else if(e<3)return t;throw new H(`Unsupported input rank by biasAdd: ${t.rank}`)}function pr(e,t,n){return q(()=>(n==null&&(n=ur()),Ht(n),le(e,M1(e.rank,t,n))))}function CL(e,t=1){if(t!==1)throw new Le(`Support for alpha values other than 1 (${t}) is not implemented yet.`);return Td(e)}function TL(e){return q(()=>pe(e,le(tn(e),1)))}function xw(e,t,n,s){return q(()=>_v(e,t,n,s))}function NL(e){return q(()=>{let t=le(.5,L(.2,e));return ds(t,0,1)})}function Vd(e,t,n=!1){return n?e():t()}var EL=["fanIn","fanOut","fanAvg"],RL=["normal","uniform","truncatedNormal"];function $L(e){hl(EL,"FanMode",e)}function _L(e){hl(RL,"Distribution",e)}var Zs=class extends ue.Serializable{fromConfigUsesCustomObjects(){return!1}getConfig(){return{}}},z1=class extends Zs{apply(e,t){return Gt(e,t)}};z1.className="Zeros";ue.registerClass(z1);var zf=class extends Zs{apply(e,t){return hs(e,t)}};zf.className="Ones";ue.registerClass(zf);var L1=class extends Zs{constructor(e){super();if(typeof e!="object")throw new H(`Expected argument of type ConstantConfig but got ${e}`);if(e.value===void 0)throw new H(`config must have value set but got ${e}`);this.value=e.value}apply(e,t){return q(()=>L(Re(this.value),hs(e,t)))}getConfig(){return{value:this.value}}};L1.className="Constant";ue.registerClass(L1);var B1=class extends Zs{constructor(e){super();this.DEFAULT_MINVAL=-.05,this.DEFAULT_MAXVAL=.05,this.minval=e.minval||this.DEFAULT_MINVAL,this.maxval=e.maxval||this.DEFAULT_MAXVAL,this.seed=e.seed}apply(e,t){return Lu(e,this.minval,this.maxval,t)}getConfig(){return{minval:this.minval,maxval:this.maxval,seed:this.seed}}};B1.className="RandomUniform";ue.registerClass(B1);var W1=class extends Zs{constructor(e){super();this.DEFAULT_MEAN=0,this.DEFAULT_STDDEV=.05,this.mean=e.mean||this.DEFAULT_MEAN,this.stddev=e.stddev||this.DEFAULT_STDDEV,this.seed=e.seed}apply(e,t){if(t=t||"float32",t!=="float32"&&t!=="int32")throw new Le(`randomNormal does not support dType ${t}.`);return Mf(e,this.mean,this.stddev,t,this.seed)}getConfig(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}}};W1.className="RandomNormal";ue.registerClass(W1);var V1=class extends Zs{constructor(e){super();this.DEFAULT_MEAN=0,this.DEFAULT_STDDEV=.05,this.mean=e.mean||this.DEFAULT_MEAN,this.stddev=e.stddev||this.DEFAULT_STDDEV,this.seed=e.seed}apply(e,t){if(t=t||"float32",t!=="float32"&&t!=="int32")throw new Le(`truncatedNormal does not support dType ${t}.`);return wf(e,this.mean,this.stddev,t,this.seed)}getConfig(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}}};V1.className="TruncatedNormal";ue.registerClass(V1);var U1=class extends Zs{constructor(e){super();this.gain=e.gain!=null?e.gain:1}apply(e,t){return q(()=>{if(e.length!==2||e[0]!==e[1])throw new H("Identity matrix initializer can only be used for 2D square matrices.");return L(this.gain,Y2(e[0]))})}getConfig(){return{gain:this.gain}}};U1.className="Identity";ue.registerClass(U1);function DL(e,t="channelsLast"){let n,s;if(Ht(t),e.length===2)n=e[0],s=e[1];else if([3,4,5].indexOf(e.length)!==-1){if(t==="channelsFirst"){let r=Po(e,2);n=e[1]*r,s=e[0]*r}else if(t==="channelsLast"){let r=Po(e,0,e.length-2);n=e[e.length-2]*r,s=e[e.length-1]*r}}else{let r=Po(e);n=Math.sqrt(r),s=Math.sqrt(r)}return[n,s]}var as=class extends Zs{constructor(e){super();if(e.scale<0)throw new H(`scale must be a positive float. Got: ${e.scale}`);this.scale=e.scale==null?1:e.scale,this.mode=e.mode==null?"fanIn":e.mode,$L(this.mode),this.distribution=e.distribution==null?"normal":e.distribution,_L(this.distribution),this.seed=e.seed}apply(e,t){let n=DL(e),s=n[0],r=n[1],a=this.scale;if(this.mode==="fanIn"?a/=Math.max(1,s):this.mode==="fanOut"?a/=Math.max(1,r):a/=Math.max(1,(s+r)/2),this.distribution==="normal"){let o=Math.sqrt(a);if(t=t||"float32",t!=="float32"&&t!=="int32")throw new Le(`${this.getClassName()} does not support dType ${t}.`);return wf(e,0,o,t,this.seed)}else{let o=Math.sqrt(3*a);return Lu(e,-o,o,t)}}getConfig(){return{scale:this.scale,mode:this.mode,distribution:this.distribution,seed:this.seed}}};as.className="VarianceScaling";ue.registerClass(as);var Lf=class extends as{constructor(e){super({scale:1,mode:"fanAvg",distribution:"uniform",seed:e==null?null:e.seed})}getClassName(){return as.className}};Lf.className="GlorotUniform";ue.registerClass(Lf);var Bf=class extends as{constructor(e){super({scale:1,mode:"fanAvg",distribution:"normal",seed:e==null?null:e.seed})}getClassName(){return as.className}};Bf.className="GlorotNormal";ue.registerClass(Bf);var Wf=class extends as{constructor(e){super({scale:2,mode:"fanIn",distribution:"normal",seed:e==null?null:e.seed})}getClassName(){return as.className}};Wf.className="HeNormal";ue.registerClass(Wf);var Vf=class extends as{constructor(e){super({scale:2,mode:"fanIn",distribution:"uniform",seed:e==null?null:e.seed})}getClassName(){return as.className}};Vf.className="HeUniform";ue.registerClass(Vf);var Uf=class extends as{constructor(e){super({scale:1,mode:"fanIn",distribution:"normal",seed:e==null?null:e.seed})}getClassName(){return as.className}};Uf.className="LeCunNormal";ue.registerClass(Uf);var Gf=class extends as{constructor(e){super({scale:1,mode:"fanIn",distribution:"uniform",seed:e==null?null:e.seed})}getClassName(){return as.className}};Gf.className="LeCunNormal";ue.registerClass(Gf);var G1=class extends Zs{constructor(e){super();if(this.DEFAULT_GAIN=1,this.gain=e.gain==null?this.DEFAULT_GAIN:e.gain,this.seed=e.seed,this.seed!=null)throw new Le("Random seed is not implemented for Orthogonal Initializer yet.")}apply(e,t){return q(()=>{if(e.length<2)throw new Le("Shape must be at least 2D.");e[0]*e[1]>2e3&&console.warn(`Orthogonal initializer is being called on a matrix with more than 2000 (${e[0]*e[1]}) elements: Slowness may result.`);let n=e[0]>e[1]?[e[1],e[0]]:e,s=Mf(n,0,1,"float32"),r=Uv.gramSchmidt(s);return e[0]>e[1]&&(r=Ye(r)),L(this.gain,r)})}getConfig(){return{gain:this.gain,seed:this.seed}}};G1.className="Orthogonal";ue.registerClass(G1);var bw={constant:"Constant",glorotNormal:"GlorotNormal",glorotUniform:"GlorotUniform",heNormal:"HeNormal",heUniform:"HeUniform",identity:"Identity",leCunNormal:"LeCunNormal",leCunUniform:"LeCunUniform",ones:"Ones",orthogonal:"Orthogonal",randomNormal:"RandomNormal",randomUniform:"RandomUniform",truncatedNormal:"TruncatedNormal",varianceScaling:"VarianceScaling",zeros:"Zeros"};function vw(e,t={}){return Md(e,ue.SerializationMap.getMap().classNameMap,t,"initializer")}function Ot(e){return C1(e)}function Nt(e){if(typeof e=="string"){let t=e in bw?bw[e]:e;if(t==="GlorotNormal")return new Bf;if(t==="GlorotUniform")return new Lf;if(t==="HeNormal")return new Wf;if(t==="HeUniform")return new Vf;if(t==="LeCunNormal")return new Uf;if(t==="LeCunUniform")return new Gf;{let n={};return n.className=t,n.config={},vw(n)}}else return e instanceof Zs?e:vw(e)}function PL(){return new z1}function FL(){return new zf}function OL(e){return new L1(e)}function ML(e){return new B1(e)}function zL(e){return new W1(e)}function LL(e){return new V1(e)}function BL(e){return new U1(e)}function WL(e){return new as(e)}function VL(e){return new Lf(e)}function UL(e){return new Bf(e)}function GL(e){return new Wf(e)}function HL(e){return new Vf(e)}function jL(e){return new Uf(e)}function qL(e){return new Gf(e)}function XL(e){return new G1(e)}var ww={};Oe(ww,{Layer:()=>Qe,RNN:()=>ta,RNNCell:()=>Kd,activation:()=>EW,add:()=>zW,alphaDropout:()=>vV,average:()=>LW,averagePooling1d:()=>dy,averagePooling2d:()=>py,averagePooling3d:()=>hy,avgPool1d:()=>XW,avgPool2d:()=>ZW,avgPool3d:()=>JW,avgPooling1d:()=>KW,avgPooling2d:()=>YW,avgPooling3d:()=>QW,batchNormalization:()=>HW,bidirectional:()=>hV,concatenate:()=>BW,conv1d:()=>bW,conv2d:()=>vW,conv2dTranspose:()=>wW,conv3d:()=>kW,conv3dTranspose:()=>SW,convLstm2d:()=>uV,convLstm2dCell:()=>cV,cropping2D:()=>CW,dense:()=>RW,depthwiseConv2d:()=>NW,dot:()=>GW,dropout:()=>$W,elu:()=>fW,embedding:()=>MW,flatten:()=>DW,gaussianDropout:()=>bV,gaussianNoise:()=>xV,globalAveragePooling1d:()=>eV,globalAveragePooling2d:()=>tV,globalMaxPool1d:()=>mV,globalMaxPool2d:()=>gV,globalMaxPooling1d:()=>Fk,globalMaxPooling2d:()=>Ok,gru:()=>sV,gruCell:()=>rV,input:()=>tk,inputLayer:()=>hW,layerNormalization:()=>jW,leakyReLU:()=>gW,lstm:()=>aV,lstmCell:()=>oV,masking:()=>wV,maxPool1d:()=>AV,maxPool2d:()=>yV,maxPooling1d:()=>Mk,maxPooling2d:()=>zk,maxPooling3d:()=>nV,maximum:()=>WW,minimum:()=>VW,multiply:()=>UW,permute:()=>OW,prelu:()=>AW,reLU:()=>mW,repeatVector:()=>PW,reshape:()=>FW,rnn:()=>dV,separableConv2d:()=>IW,simpleRNN:()=>iV,simpleRNNCell:()=>lV,softmax:()=>yW,spatialDropout1d:()=>_W,stackedRNNCells:()=>pV,thresholdedReLU:()=>xW,timeDistributed:()=>fV,upSampling2d:()=>TW,zeroPadding2d:()=>qW});var KL=0;function kw(){return KL++}var Hf={};function jf(e=""){return e in Hf||(Hf[e]=0),Hf[e]+=1,e+Hf[e].toString()}function H1(e){return Array.isArray(e)&&Array.isArray(e[0])}function qf(e){return e.length===0?[]:Array.isArray(e[0])?e:[e]}function Be(e){let t;if(Array.isArray(e)){if(e.length!==1)throw new H(`Expected Tensor length to be 1; got ${e.length}`);t=e[0]}else t=e;return t}function pt(e){if(Array.isArray(e)&&Array.isArray(e[0])){if(e.length===1)return e=e,e[0];throw new H(`Expected exactly 1 Shape; got ${e.length}`)}else return e}function Xf(e){let t=0;for(let n of e)n.shape.length===0?t+=1:t+=n.shape.reduce((s,r)=>s*r);return t}var Sw="Variable",Iw=class{constructor(e,t="float32",n=Sw,s=!0,r=null){this.dtype=t==null?"float32":t,this.shape=e.shape,this.id=kw(),n=n==null?Sw:n,this.originalName=fw(n),this.name=mw(this.originalName),this.trainable_=s,this.constraint=r,this.val=Tv(e,this.trainable_,this.name,this.dtype)}read(){return this.assertNotDisposed(),this.val}write(e){return this.assertNotDisposed(),ZL(this.val,e),this.val.id!==e.id&&(this.val.assign(e),this.constraint!=null&&this.val.assign(this.constraint.apply(this.val))),this}dispose(){this.assertNotDisposed(),this.val.dispose()}assertNotDisposed(){if(this.val.isDisposed)throw new Error(`LayersVariable ${this.name} is already disposed.`)}get trainable(){return this.trainable_}set trainable(e){this.trainable_=e,this.val.trainable=e}};function ZL(e,t){if(e.shape.toString()!==t.shape.toString())throw new Error("Shape mismatch: "+JSON.stringify(e.shape)+" vs. "+JSON.stringify(t.shape))}function j1(e){return e.map(t=>t.read())}function q1(e){e.forEach(t=>{t[0].write(t[1])})}var Kt=class{constructor(e){this.dtype=e.dtype,this.shape=e.shape,e.shape!=null?this.ndim=e.shape.length:this.ndim=e.ndim,this.maxNDim=e.maxNDim,this.minNDim=e.minNDim,this.axes=e.axes||{}}},hr=class{constructor(e,t,n,s,r,a,o){this.dtype=e,this.shape=t,this.sourceLayer=n,this.inputs=s,this.callArgs=r,this.outputTensorIndex=o,this.id=kw(),a!=null&&(this.originalName=fw(a),this.name=mw(this.originalName)),this.rank=t.length}},YL=0,Kf=class{constructor(e,t){this.callArgs=t,this.id=YL++,this.outboundLayer=e.outboundLayer,this.inboundLayers=e.inboundLayers,this.nodeIndices=e.nodeIndices,this.tensorIndices=e.tensorIndices,this.inputTensors=e.inputTensors,this.outputTensors=e.outputTensors,this.inputMasks=e.inputMasks,this.outputMasks=e.outputMasks,this.inputShapes=e.inputShapes,this.outputShapes=e.outputShapes;for(let n of e.inboundLayers)n!=null&&n.outboundNodes.push(this);e.outboundLayer.inboundNodes.push(this)}getConfig(){let e=[];for(let t of this.inboundLayers)t!=null?e.push(t.name):e.push(null);return{outboundLayer:this.outboundLayer?this.outboundLayer.name:null,inboundLayers:e,nodeIndices:this.nodeIndices,tensorIndices:this.tensorIndices}}},JL=0,Qe=class extends ue.Serializable{constructor(e={}){super();this._callHook=null,this._addedWeightNames=[],this._stateful=!1,this.id=JL++,this.activityRegularizer=null,this.inputSpec=null,this.supportsMasking=!1,this._trainableWeights=[],this._nonTrainableWeights=[],this._losses=[],this._updates=[],this._built=!1,this.inboundNodes=[],this.outboundNodes=[];let t=e.name;if(!t){let n=this.getClassName();t=Qr(n)+"_"+jf(n)}if(this.name=t,this.trainable_=e.trainable==null?!0:e.trainable,e.inputShape!=null||e.batchInputShape!=null){let n;if(e.batchInputShape!=null)n=e.batchInputShape;else if(e.inputShape!=null){let r=null;e.batchSize!=null&&(r=e.batchSize),n=[r].concat(e.inputShape)}this.batchInputShape=n;let s=e.dtype;s==null&&(s=e.inputDType),s==null&&(s="float32"),this.dtype=s}e.weights!=null?this.initialWeights=e.weights:this.initialWeights=null,this._refCount=null,this.fastWeightInitDuringBuild=!1}static nodeKey(e,t){return e.name+"_ib-"+t.toString()}getNodeAtIndex(e,t){if(this.inboundNodes.length===0)throw new cr(`The layer has never been called and thus has no defined ${t}.`);if(this.inboundNodes.length<=e)throw new H(`Asked to get ${t} at node ${e}, but the layer has only ${this.inboundNodes.length} inbound nodes.`);return this.inboundNodes[e]}getInputAt(e){return rs(this.getNodeAtIndex(e,"input").inputTensors)}getOutputAt(e){return rs(this.getNodeAtIndex(e,"output").outputTensors)}get input(){if(this.inboundNodes.length>1)throw new Jr(`Layer ${this.name} has multiple inbound nodes, hence the notion of "layer input" is ill-defined. Use \`getInputAt(nodeIndex)\` instead.`);if(this.inboundNodes.length===0)throw new Jr(`Layer ${this.name} is not connected, no input to return.`);return rs(this.getNodeAtIndex(0,"input").inputTensors)}get output(){if(this.inboundNodes.length===0)throw new Jr(`Layer ${this.name} has no inbound nodes.`);if(this.inboundNodes.length>1)throw new Jr(`Layer ${this.name} has multiple inbound nodes, hence the notion of "layer output" is ill-defined. Use \`getOutputAt(nodeIndex)\` instead.`);return rs(this.getNodeAtIndex(0,"output").outputTensors)}get losses(){return this._losses}calculateLosses(){return this.losses.map(e=>e())}get updates(){return this._updates}get built(){return this._built}set built(e){this._built=e}get trainable(){return this.trainable_}set trainable(e){this._trainableWeights.forEach(t=>t.trainable=e),this.trainable_=e}get trainableWeights(){return this.trainable_?this._trainableWeights.filter(e=>e.trainable):[]}set trainableWeights(e){this._trainableWeights=e}get nonTrainableWeights(){return this.trainable?this._trainableWeights.filter(e=>!e.trainable).concat(this._nonTrainableWeights):this._trainableWeights.concat(this._nonTrainableWeights)}set nonTrainableWeights(e){this._nonTrainableWeights=e}get weights(){return this.trainableWeights.concat(this.nonTrainableWeights)}get stateful(){return this._stateful}resetStates(){if(!this.stateful)throw new Error("Cannot call the resetStates() method of a non-stateful Layer object.")}assertInputCompatibility(e){if(e=kt(e),this.inputSpec==null||this.inputSpec.length===0)return;let t=kt(this.inputSpec);if(e.length!==t.length)throw new H(`Layer ${this.name} expects ${t.length} inputs, but it received ${e.length} input tensors. Input received: ${e}`);for(let n=0;n<e.length;n++){let s=e[n],r=t[n];if(r==null)continue;let a=s.rank;if(r.ndim!=null&&a!==r.ndim)throw new H(`Input ${n} is incompatible with layer ${this.name}: expected ndim=${r.ndim}, found ndim=${a}`);if(r.maxNDim!=null&&a>r.maxNDim)throw new H(`Input ${n} is incompatible with layer ${this.name}: expected max_ndim=${r.maxNDim}, found ndim=${a}`);if(r.minNDim!=null&&a<r.minNDim)throw new H(`Input ${n} is incompatible with layer ${this.name}: expected min_ndim=${r.minNDim}, found ndim=${a}.`);if(r.dtype!=null&&s.dtype!==r.dtype)throw new H(`Input ${n} is incompatible with layer ${this.name} : expected dtype=${r.dtype}, found dtype=${s.dtype}.`);if(r.axes){let o=s.shape;for(let i in r.axes){let l=Number(i),c=r.axes[i],u=l>=0?o[l]:o[o.length+l];if(c!=null&&[c,null].indexOf(u)===-1)throw new H(`Input ${n} is incompatible with layer ${this.name}: expected axis ${l} of input shape to have value ${c} but got shape ${o}.`)}}if(r.shape!=null)for(let o=0;o<r.shape.length;++o){let i=r.shape[o],l=s.shape[o];if(i!=null&&l!=null&&i!==l)throw new H(`Input ${n} is incompatible with layer ${this.name}: expected shape=${r.shape}, found shape=${s.shape}.`)}}}call(e,t){return e}invokeCallHook(e,t){this._callHook!=null&&this._callHook(e,t)}setCallHook(e){this._callHook=e}clearCallHook(){this._callHook=null}apply(e,t){t=t||{},this.assertNotDisposed();let n=kt(e),s=!0;for(let a of n)if(!(a instanceof hr)){s=!1;break}let r=!0;for(let a of n)if(a instanceof hr){r=!1;break}if(s===r)throw new H("Arguments to apply() must be all SymbolicTensors or all Tensors");return fl(this.name,()=>{if(!this.built){this.assertInputCompatibility(e);let a=[];for(let o of kt(e))a.push(o.shape);this.build(rs(a)),this.built=!0,this.initialWeights&&this.setWeights(this.initialWeights),this._refCount===null&&r&&(this._refCount=1)}if(this.assertInputCompatibility(e),r){let a=this.call(e,t),o=kt(a),i=[];for(let l of o)n.indexOf(l)!==-1&&(l=l.clone()),i.push(l);if(a=rs(i),this.activityRegularizer!=null)throw new Le("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return a}else{let a=QL(e),o=this.computeOutputShape(a),i,l=eB(e);if(this.warnOnIncompatibleInputShape(Array.isArray(e)?a[0]:a),o!=null&&o.length>0&&Array.isArray(o[0])?i=o.map((c,u)=>new hr(l,c,this,kt(e),t,this.name,u)):i=new hr(l,o,this,kt(e),t,this.name),this.addInboundNode(e,i,null,null,a,o,t),this._refCount++,this.activityRegularizer!=null)throw new Le("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return i}})}warnOnIncompatibleInputShape(e){if(this.batchInputShape!=null)if(e.length!==this.batchInputShape.length)console.warn(`The rank of the input tensor provided (shape: ${JSON.stringify(e)}) does not match that of the batchInputShape (${JSON.stringify(this.batchInputShape)}) of the layer ${this.name}`);else{let t=!1;this.batchInputShape.forEach((n,s)=>{n!=null&&e[s]!=null&&e[s]!==n&&(t=!0)}),t&&console.warn(`The shape of the input tensor (${JSON.stringify(e)}) does not match the expectation of layer ${this.name}: ${JSON.stringify(this.batchInputShape)}`)}}get outputShape(){if(this.inboundNodes==null||this.inboundNodes.length===0)throw new Jr(`The layer ${this.name} has never been called and thus has no defined output shape.`);let e=[];for(let t of this.inboundNodes){let n=JSON.stringify(t.outputShapes);e.indexOf(n)===-1&&e.push(n)}if(e.length===1){let t=this.inboundNodes[0].outputShapes;return Array.isArray(t)&&Array.isArray(t[0])&&t.length===1?t[0]:t}else throw new Jr(`The layer ${this.name} has multiple inbound nodes with different output shapes. Hence the notion of "output shape" is ill-defined for the layer.`)}countParams(){if(!this.built)throw new cr(`You tried to call countParams() on ${this.name}, but the layer is not built yet. Build it first by calling build(batchInputShape).`);return Xf(this.weights)}build(e){this.built=!0}getWeights(e=!1){return j1(e?this.trainableWeights:this.weights)}setWeights(e){q(()=>{let t=this.weights;if(t.length!==e.length)throw new H(`You called setWeights(weights) on layer "${this.name}" with a weight list of length ${e.length}, but the layer was expecting ${t.length} weights. Provided weights: ${e}...`);if(t.length===0)return;let n=[],s=j1(t);for(let r=0;r<s.length;++r){let a=s[r],o=t[r],i=e[r];if(!v.arraysEqual(a.shape,i.shape))throw new H(`Layer weight shape ${a.shape} not compatible with provided weight shape ${i.shape}`);n.push([o,i])}q1(n)})}addWeight(e,t,n,s,r,a,o,i){if(this._addedWeightNames.indexOf(e)!==-1)throw new H(`Duplicate weight name ${e} for layer ${this.name}`);this._addedWeightNames.push(e),n==null&&(n="float32"),this.fastWeightInitDuringBuild&&(s=i!=null?i():Nt("zeros"));let l=s.apply(t,n),c=new Iw(l,n,e,a,o);return l.dispose(),r!=null&&this.addLoss(()=>r.apply(c.read())),a==null&&(a=!0),a?this._trainableWeights.push(c):this._nonTrainableWeights.push(c),c}setFastWeightInitDuringBuild(e){this.fastWeightInitDuringBuild=e}addLoss(e){e==null||Array.isArray(e)&&e.length===0||(e=kt(e),this._losses!==void 0&&this._losses!==null&&this.losses.push(...e))}computeOutputShape(e){return e}computeMask(e,t){if(!this.supportsMasking){if(t!=null)if(Array.isArray(t))t.forEach(n=>{if(n!=null)throw new TypeError(`Layer ${this.name} does not support masking, but was passed an inputMask.`)});else throw new TypeError(`Layer ${this.name} does not support masking, but was passed an inputMask.`);return null}return t}addInboundNode(e,t,n,s,r,a,o=null){let i=kt(e);t=kt(t),n=kt(n),s=kt(s),r=qf(r),a=qf(a);let l=[],c=[],u=[];for(let d of i)l.push(d.sourceLayer),c.push(d.nodeIndex),u.push(d.tensorIndex);new Kf({outboundLayer:this,inboundLayers:l,nodeIndices:c,tensorIndices:u,inputTensors:i,outputTensors:t,inputMasks:n,outputMasks:s,inputShapes:r,outputShapes:a},o);for(let d=0;d<t.length;d++)t[d].sourceLayer=this,t[d].nodeIndex=this.inboundNodes.length-1,t[d].tensorIndex=d}getConfig(){let e={name:this.name,trainable:this.trainable};return this.batchInputShape!=null&&(e.batchInputShape=this.batchInputShape),this.dtype!=null&&(e.dtype=this.dtype),e}disposeWeights(){return this.weights.forEach(e=>e.dispose()),this.weights.length}assertNotDisposed(){if(this._refCount===0)throw new Error(`Layer '${this.name}' is already disposed.`)}dispose(){if(!this.built)throw new Error(`Cannot dispose Layer ${this.name} because it has not been built yet.`);if(this._refCount===null)throw new Error(`Cannot dispose Layer ${this.name} because it has not been used yet.`);this.assertNotDisposed();let e=0;return--this._refCount==0&&(e=this.disposeWeights()),{refCountAfterDispose:this._refCount,numDisposedVariables:e}}};function QL(e){e=kt(e);let t=[];for(let n of e)t.push(n.shape);return rs(t)}function eB(e){return"float32"}function Cw(e,t,n){if((t==null||n!=null&&n>0)&&(t=e.sourceLayer,n=e.nodeIndex),t.inboundNodes.length===0)return[e];{let s=t.inboundNodes[n];if(s.inboundLayers.length===0)return s.inputTensors;{let r=[];for(let a=0;a<s.inboundLayers.length;a++){let o=s.inputTensors[a],i=s.inboundLayers[a],l=s.nodeIndices[a],c=Cw(o,i,l);for(let u of c)r.indexOf(u)===-1&&r.push(u)}return r}}}var ju=class extends Qe{constructor(e){super({dtype:e.dtype,name:e.name!=null?e.name:jf("input").toString()});if(e.batchSize==null&&(e.batchSize=null),e.sparse==null&&(e.sparse=!1),this.trainable=!1,this.built=!0,this.sparse=e.sparse,e.inputShape!=null&&e.batchInputShape!=null)throw new H("Only provide the inputShape OR batchInputShape argument to inputLayer, not both at the same time.");let t=e.batchInputShape;if(t==null){if(e.inputShape==null)throw new H("An InputLayer should be passed either a `batchInputShape` or an `inputShape`.");t=[e.batchSize].concat(e.inputShape)}else if(e.batchSize!=null)throw new H("Cannot specify batchSize if batchInputShape is specified when creating an InputLayer.");let n=e.dtype||"float32";this.batchInputShape=t,this.dtype=n,this.inputSpec=[{shape:t}];let s=new hr(this.dtype,this.batchInputShape,this,[],{},this.name);s.nodeIndex=0,s.tensorIndex=0,new Kf({outboundLayer:this,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:[s],outputTensors:[s],inputMasks:[null],outputMasks:[null],inputShapes:[t],outputShapes:[t]})}apply(e,t){throw new H(`Cannot pass any input to an InputLayer's apply() method. InputLayer name: ${this.name}`)}dispose(){return{refCountAfterDispose:this._refCount,numDisposedVariables:0}}getConfig(){return{batchInputShape:this.batchInputShape,dtype:this.dtype,sparse:this.sparse,name:this.name}}};ju.className="InputLayer";ue.registerClass(ju);function Tw(e){if(e.batchShape==null&&e.shape==null)throw new Error("Please provide to Input either a `shape` or a `batchShape` argument. Note that `shape` does not include the batch dimension.");if(e.batchShape!=null&&e.shape!=null)throw new H("Please provide either a `shape` or `batchShape` argument to Input, but not both.");let t=e.batchShape;e.shape!=null&&t==null&&(t=[null].concat(e.shape));let n=e.dtype;return n==null&&(n="float32"),new ju({batchInputShape:t,name:e.name,dtype:n,sparse:e.sparse}).inboundNodes[0].outputTensors[0]}async function Oo(e){if(e==null)return;let t=[],n=[],s=[];for(let r in e){let a=e[r];if(typeof a!="number"){let o=a;t.push(o.data()),n.push(r),s.push(o)}}if(t.length>0){let r=await Promise.all(t);for(let a=0;a<r.length;++a)e[n[a]]=r[a][0];se(s)}}function Nw(e){if(e!=null)for(let t in e){let n=e[t];typeof n!="number"&&n.dispose()}}var Ew;(function(e){e[e.SILENT=0]="SILENT",e[e.VERBOSE=1]="VERBOSE"})(Ew||(Ew={}));var tB=125,qu=class{constructor(){this.validationData=null}setParams(e){this.params=e}async onEpochBegin(e,t){}async onEpochEnd(e,t){}async onBatchBegin(e,t){}async onBatchEnd(e,t){}async onTrainBegin(e){}async onTrainEnd(e){}setModel(e){}},Rw=class{constructor(e,t=10){e==null&&(e=[]),this.callbacks=e,this.queueLength=t}append(e){this.callbacks.push(e)}setParams(e){for(let t of this.callbacks)t.setParams(e)}setModel(e){for(let t of this.callbacks)t.setModel(e)}async onEpochBegin(e,t){t==null&&(t={});for(let n of this.callbacks)await n.onEpochBegin(e,t)}async onEpochEnd(e,t){t==null&&(t={});for(let n of this.callbacks)await n.onEpochEnd(e,t)}async onBatchBegin(e,t){t==null&&(t={});for(let n of this.callbacks)await n.onBatchBegin(e,t)}async onBatchEnd(e,t){t==null&&(t={});for(let n of this.callbacks)await n.onBatchEnd(e,t)}async onTrainBegin(e){e==null&&(e={});for(let t of this.callbacks)await t.onTrainBegin(e)}async onTrainEnd(e){e==null&&(e={});for(let t of this.callbacks)await t.onTrainEnd(e)}},nB=class extends qu{constructor(){super()}async onEpochBegin(e){this.seen=0,this.totals={}}async onBatchEnd(e,t){t==null&&(t={});let n=t.size==null?0:t.size;this.seen+=n;for(let s in t){let r=t[s];if(typeof r=="number")this.totals.hasOwnProperty(s)||(this.totals[s]=0),this.totals[s]=this.totals[s]+r*n;else{let a;s in this.totals?a=this.totals[s]:this.totals[s]=0;let o=q(()=>le(this.totals[s],L(r,n)));this.totals[s]=o,a!=null&&a.dispose()}}}async onEpochEnd(e,t){if(t!=null)for(let n of this.params.metrics)this.totals[n]!=null&&(typeof this.totals[n]=="number"?t[n]=this.totals[n]/this.seen:q(()=>{let s=L(pe(1,this.seen),this.totals[n]);t[n]=s,this.totals[n].dispose(),fn(t[n])}))}},$w=class extends qu{async onTrainBegin(e){this.epoch=[],this.history={}}async onEpochEnd(e,t){t==null&&(t={}),this.epoch.push(e);for(let n in t)this.history[n]==null&&(this.history[n]=[]),this.history[n].push(t[n])}async syncData(){let e=[],t=[],n=[];for(let r in this.history){let a=this.history[r];for(let o=0;o<a.length;++o)if(typeof a[o]!="number"){let i=a[o];e.push(i.data()),t.push(r),n.push(o)}}let s=await Promise.all(e);for(let r=0;r<s.length;++r)this.history[t[r]][n[r]].dispose(),this.history[t[r]][n[r]]=s[r][0]}},_w=class extends qu{constructor(e,t){super();if(this.currentEpoch=0,this.nowFunc=e.nowFunc,this.nextFrameFunc=e.nextFrameFunc||Gv,this.yieldEvery=t||"auto",this.yieldEvery==="auto"&&(this.yieldEvery=tB),this.yieldEvery==="never"&&e.onYield!=null)throw new Error("yieldEvery is `never` but you provided an `onYield` callback. Either change `yieldEvery` or remove the callback");v.isNumber(this.yieldEvery)&&(this.maybeWait=uL(this.maybeWait.bind(this),this.yieldEvery,this.nowFunc)),this.trainBegin=e.onTrainBegin,this.trainEnd=e.onTrainEnd,this.epochBegin=e.onEpochBegin,this.epochEnd=e.onEpochEnd,this.batchBegin=e.onBatchBegin,this.batchEnd=e.onBatchEnd,this.yield=e.onYield}async maybeWait(e,t,n){let s=[];this.yield!=null&&(await Oo(n),s.push(this.yield(e,t,n))),s.push(this.nextFrameFunc()),await Promise.all(s)}async onEpochBegin(e,t){this.currentEpoch=e,this.epochBegin!=null&&(await Oo(t),await this.epochBegin(e,t))}async onEpochEnd(e,t){let n=[];this.epochEnd!=null&&(await Oo(t),n.push(this.epochEnd(e,t))),this.yieldEvery==="epoch"&&n.push(this.nextFrameFunc()),await Promise.all(n)}async onBatchBegin(e,t){this.batchBegin!=null&&(await Oo(t),await this.batchBegin(e,t))}async onBatchEnd(e,t){let n=[];this.batchEnd!=null&&(await Oo(t),n.push(this.batchEnd(e,t))),this.yieldEvery==="batch"?n.push(this.nextFrameFunc()):v.isNumber(this.yieldEvery)&&n.push(this.maybeWait(this.currentEpoch,e,t)),await Promise.all(n)}async onTrainBegin(e){this.trainBegin!=null&&(await Oo(e),await this.trainBegin(e))}async onTrainEnd(e){this.trainEnd!=null&&(await Oo(e),await this.trainEnd(e))}};function Dw(e,t){return e==null&&(e={}),e instanceof qu?[e]:Array.isArray(e)&&e[0]instanceof qu?e:kt(e).map(s=>new _w(s,t))}var Dr=class{constructor(){}static registerCallbackConstructor(e,t){v.assert(e>=0&&Number.isInteger(e),()=>`Verbosity level is expected to be an integer >= 0, but got ${e}`),Dr.checkForDuplicate(t),Dr.constructors[e]==null&&(Dr.constructors[e]=[]),Dr.constructors[e].push(t)}static checkForDuplicate(e){for(let t in Dr.constructors)Dr.constructors[+t].forEach(s=>{if(s===e)throw new H("Duplicate callback constructor.")})}static clear(){Dr.constructors={}}static createCallbacks(e){let t=[];for(let n in Dr.constructors){let s=+n;e>=s&&t.push(...Dr.constructors[s])}return t.map(n=>new n)}},X1=Dr;X1.constructors={};function Pw(e,t,n,s,r,a,o,i,l){let c=new $w,u=[new nB,...X1.createCallbacks(t)];e!=null&&u.push(...e),u.push(c);let d=new Rw(u);return d.setParams({epochs:n,initialEpoch:s,samples:r,steps:a,batchSize:o,verbose:t,doValidation:i,metrics:l}),{callbackList:d,history:c}}function fr(e,t={},n=!1){return Md(e,ue.SerializationMap.getMap().classNameMap,t,"layer",n)}function Zf(e,t){return q(()=>{e.dtype!=="float32"&&(e=de(e,"float32"));let n=Se(Wd(e),t,!0),s=Fu(n.shape,sn()),r=_n(Kr(n,s));return pe(e,r)})}function gl(e,t){return q(()=>Bt(Wd(me(t,e)),-1))}function Yf(e,t){return q(()=>Bt(tn(me(t,e)),-1))}function Xu(e,t){return q(()=>{let n=me(e,t),s=ds(tn(e),sn(),Number.MAX_VALUE),r=tn(pe(n,s));return L(100,Bt(r,-1))})}function sB(e,t){return q(()=>{let n=ds(t,sn(),Number.MAX_VALUE),s=Ns(le(1,n)),r=ds(e,sn(),Number.MAX_VALUE),a=Ns(le(1,r));return Bt(Wd(me(s,a)),-1)})}function rB(e,t){return q(()=>{let n=Kr(0,me(1,L(e,t)));return Bt(Wd(n),-1)})}function aB(e,t){return q(()=>{let n=Kr(0,me(1,L(e,t)));return Bt(n,-1)})}function oB(e,t){return q(()=>{let n=Se(L(e,t),-1),s=gn(L(me(1,e),t),-1);return Kr(0,le(1,me(s,n)))})}function iB(e,t){return q(()=>{let n=Math.log(2),s=me(t,e),r=me(le(s,Mu(L(-2,s))),n);return Bt(r,-1)})}function Ud(e,t,n=!1){return q(()=>{if(n)t=Vu(t);else{let s=Se(t,t.shape.length-1,!0);t=pe(t,s)}return t=ds(t,sn(),1-sn()),Ft(Se(L(de(e,"float32"),Ns(t)),t.shape.length-1))})}function Jf(e,t,n=!1){return q(()=>{let s=de(Nd(SL(e)),"int32");t=ds(t,sn(),1-sn());let r=t.shape,a=G(kd(s,r[r.length-1]),r);return Ud(a,t,n)})}function lB(e,t){if(!v.arraysEqual(e.shape,t.shape))throw new H(`logits and labels must have the same shape, but got shapes ${JSON.stringify(e.shape)} and ${JSON.stringify(t.shape)}`);return q(()=>{let n=Rr(t),s=Ft(tn(t));return le(me(n,L(t,e)),pf(Ts(s)))})}function Qf(e,t){return q(()=>{let n;return n=ds(t,sn(),1-sn()),n=Ns(pe(n,me(1,n))),Bt(lB(e,n),-1)})}function uB(e,t){return q(()=>{let n=ds(e,sn(),1),s=ds(t,sn(),1);return Se(L(e,Ns(pe(n,s))),-1)})}function cB(e,t){return q(()=>{let n=Ns(le(sn(),t));return Bt(me(t,L(e,n)),-1)})}function K1(e,t){return q(()=>{let n=Zf(e,-1),s=Zf(t,-1),r=L(n,s);return Ft(Se(r,-1))})}var em={meanSquaredError:gl,meanAbsoluteError:Yf,meanAbsolutePercentageError:Xu,meanSquaredLogarithmicError:sB,squaredHinge:rB,hinge:aB,categoricalHinge:oB,logcosh:iB,categoricalCrossentropy:Ud,sparseCategoricalCrossentropy:Jf,binaryCrossentropy:Qf,kullbackLeiblerDivergence:uB,poisson:cB,cosineProximity:K1};function Z1(e){if(typeof e=="string"){if(e in em)return em[e];let t=`Unknown loss ${e}`;throw e.toLowerCase().includes("softmaxcrossentropy")&&(t=`Unknown loss ${e}. Use "categoricalCrossentropy" as the string name for tf.losses.softmaxCrossEntropy`),new H(t)}else return e}function Y1(e,t){return q(()=>{let n=L(.5,Es(t)),s=Ff(ps(t,n),e.dtype);return Bt(Cs(e,s),-1)})}function J1(e,t){return q(()=>Ff(Cs(Hs(e,-1),Hs(t,-1)),"float32"))}function Fw(e,t){return q(()=>de(Se(ir(Cs(e,1),Cs(t,1))),"float32"))}function dB(e,t){return q(()=>de(Se(ir(Cs(e,1),Cs(t,0))),"float32"))}function pB(e,t){return q(()=>de(Se(ir(Cs(e,0),Cs(t,1))),"float32"))}function Ow(e,t){return q(()=>{let n=Fw(e,t),s=pB(e,t),r=le(n,s);return de(Bn(ps(r,0),pe(n,r),0),"float32")})}function hB(e,t){return q(()=>{let n=Fw(e,t),s=dB(e,t),r=le(n,s);return de(Bn(ps(r,0),pe(n,r),0),"float32")})}function Mw(e,t){return Qf(e,t)}function zw(e,t){return e.rank===t.rank&&(e=ot(e,[e.rank-1])),t=Hs(t,-1),t.dtype!==e.dtype&&(t=de(t,e.dtype)),de(Cs(e,t),"float32")}var fB=gl,mB=gl,gB=Yf,AB=Yf,yB=Xu,xB=Xu,Q1=Ud,bB=K1,Lw=Jf,tm={binaryAccuracy:Y1,categoricalAccuracy:J1,precision:Ow,categoricalCrossentropy:Q1,sparseCategoricalCrossentropy:Lw,mse:fB,MSE:mB,mae:gB,MAE:AB,mape:yB,MAPE:xB,cosine:bB};function vB(e){if(typeof e=="string"&&e in tm)return tm[e];if(typeof e!="string"&&e!=null)return e;throw new H(`Unknown metric ${e}`)}function nm(e){if($r(e!==null,`Unknown LossOrMetricFn ${e}`),typeof e=="string")return e;{let t;for(let n of Object.keys(em))if(em[n]===e){t=n;break}if(t!==void 0)return t;for(let n of Object.keys(tm))if(tm[n]===e){t=n;break}return t!==void 0?t:e.name}}function wB(e){let t={Adagrad:()=>cl.adagrad(.01),Adadelta:()=>cl.adadelta(1,.95,sn()),Adam:()=>cl.adam(.001,.9,.999,sn()),Adamax:()=>cl.adamax(.002,.9,.999,sn(),0),RMSProp:()=>cl.rmsprop(.001,.9,0,sn()),SGD:()=>cl.sgd(.01)};if(t.adagrad=t.Adagrad,t.adadelta=t.Adadelta,t.adam=t.Adam,t.adamax=t.Adamax,t.rmsprop=t.RMSProp,t.sgd=t.SGD,e in t)return t[e]();throw new H(`Unknown Optimizer ${e}`)}var Bw=1*1024*1024;function Ww(e,t,n=!1){if(e==null||typeof e!="object"||Object.getPrototypeOf(e)!==Object.prototype||!eA(e))throw new Error("User-defined metadata is expected to be a JSON object, but is not.");if(n){let s=JSON.stringify(e);s.length>Bw&&console.warn(`User-defined metadata of model "${t}" is too large in size (length=${s.length} when serialized). It is not recommended to store such large objects in user-defined metadata. Please make sure its serialized length is <= ${Bw}.`)}}function eA(e){if(e===null)return!0;if(typeof e=="object")if(Object.getPrototypeOf(e)===Object.prototype){let t=Object.keys(e);for(let n of t)if(typeof n!="string"||!eA(e[n]))return!1;return!0}else if(Array.isArray(e)){for(let t of e)if(!eA(t))return!1;return!0}else return!1;else{let t=typeof e;return t==="string"||t==="number"||t==="boolean"}}function kB(e,t,n,s=console.log){let r=IB(e),a=["Layer (type)","Output shape","Param #"];r?(t=t||65,n=n||[.45,.85,1]):(t=t||98,n=n||[.33,.55,.67,1]),n[n.length-1]<=1&&(n=n.map(u=>Math.floor(t*u)));let o;if(!r){a.push("Receives inputs"),o=[];for(let u in e.nodesByDepth)o.push(...e.nodesByDepth[u])}s("_".repeat(t)),sm(a,n,s),s("=".repeat(t));let i=e.layers;for(let u=0;u<i.length;++u)r?CB(i[u],n,s):TB(i[u],n,o,s),s((u===i.length-1?"=":"_").repeat(t));e.checkTrainableWeightsConsistency();let l=SB(e),c=Xf(e.nonTrainableWeights);s(`Total params: ${l+c}`),s(`Trainable params: ${l}`),s(`Non-trainable params: ${c}`),s("_".repeat(t))}function SB(e){let t;return e.collectedTrainableWeights!=null?t=Xf(e.collectedTrainableWeights):t=Xf(e.trainableWeights),t}function IB(e){let t=!0,n=[],s=[];for(let r in e.nodesByDepth)n.push(e.nodesByDepth[r]);for(let r of n){if(r.length>1||r.length===1&&r[0].inboundLayers.length>1){t=!1;break}s.push(...r)}if(t)for(let r of e.layers){let a=!1;for(let o of r.inboundNodes)if(s.indexOf(o)!==-1)if(a){t=!1;break}else a=!0;if(!t)break}return t}function sm(e,t,n=console.log){let s="";for(let r=0;r<e.length;++r)r>0&&(s=s.slice(0,s.length-1)+" "),s+=e[r],s=s.slice(0,t[r]),s+=" ".repeat(t[r]-s.length);n(s)}function CB(e,t,n){let s;try{s=JSON.stringify(e.outputShape)}catch(i){s="multiple"}let r=e.name,a=e.getClassName(),o=[`${r} (${a})`,s,e.countParams().toString()];sm(o,t,n)}function TB(e,t,n,s){let r;try{r=JSON.stringify(e.outputShape)}catch(u){r="multiple"}let a=[];for(let u of e.inboundNodes)if(!(n!=null&&n.length>0&&n.indexOf(u)===-1))for(let d=0;d<u.inboundLayers.length;++d){let p=u.inboundLayers[d].name,h=u.nodeIndices[d],f=u.tensorIndices[d];a.push(`${p}[${h}][${f}]`)}let o=e.name,i=e.getClassName(),l=a.length===0?"":a[0],c=[`${o} (${i})`,r,e.countParams().toString(),l];sm(c,t,s);for(let u=1;u<a.length;++u)sm(["","","",a[u]],t,s)}function Vw(e,t,n){return(e==="inboundNodes"||e==="outputLayers"||e==="inputLayers")&&t===0&&typeof n=="string"}function Gd(e,t){if(e===null)return null;if(typeof e=="string")return pl(e);if(typeof e=="number"||typeof e=="boolean")return e;if(e instanceof Array){let n=[],s=e.length;for(let r=0;r<s;++r){let a=e[r];Vw(t,r,a)?n.push(a):n.push(Gd(a,t))}return n}else{let n={};for(let s of Object.keys(e)){let r=e[s];if(s==="name"&&typeof r=="string")n[s]=r;else{let a=pl(s);n[a]=Gd(r,a)}}return n}}function tA(e,t){if(e==null)return null;if(typeof e=="string")return Qr(e);if(typeof e=="number"||typeof e=="boolean")return e;if(e instanceof Array){let n=[],s=e.length;for(let r=0;r<s;++r){let a=e[r];Vw(t,r,a)?n.push(a):n.push(tA(a,t))}return n}else{let n={};for(let s of Object.keys(e)){let r=e[s],a=Qr(s);(s==="name"||s==="className")&&typeof r=="string"?n[a]=r:n[a]=tA(r,s)}return n}}var nA="0.0.0";function NB(e,t){if(e.dtype==null||e.dtype===t.dtype)return t;try{return de(t,e.dtype)}catch(n){throw new H(`The dtype of the feed (${t.dtype}) can not be cast to the dtype of the key '${e.name}' (${e.dtype}).`)}}var Al=class{constructor(e){if(this.id2Value={},this.id2Mask={},this.name2Id={},e instanceof Al)for(let t in e.id2Value)this.id2Value[t]=e.id2Value[t],t in e.id2Mask&&(this.id2Mask[t]=e.id2Mask[t]);else{if(e==null)return;for(let t of e)this.add(t.key,t.value)}}add(e,t,n){if(this.id2Value[e.id]==null)this.id2Value[e.id]=NB(e,t),this.name2Id[e.name]=e.id,n!=null&&(this.id2Mask[e.id]=n);else throw new H(`Duplicate key: name=${e.name}, id=${e.id}`);return this}addFeed(e){this.add(e.key,e.value)}hasKey(e){return this.id2Value[e.id]!=null}names(){return Object.keys(this.name2Id)}getValue(e){if(e instanceof hr){if(this.id2Value[e.id]==null)throw new H(`Nonexistent key: ${e.name}`);return this.id2Value[e.id]}else{let t=this.name2Id[e];if(t==null)throw new H(`Feed dict has no SymbolicTensor name: ${e}`);return this.id2Value[t]}}getMask(e){if(e instanceof hr){if(this.id2Value[e.id]==null)throw new H(`Nonexistent key: ${e.name}`);return this.id2Mask[e.id]}else{let t=this.name2Id[e];if(t==null)throw new H(`Feed dict has no SymbolicTensor name: ${e}`);return this.id2Mask[t]}}disposeMasks(){this.id2Mask!=null&&se(this.id2Mask)}},sA={},Uw={};function Hd(e,t,n,s){let r=n==null?!1:n.training,a=Array.isArray(e),o=a?e:[e],i=o.map(f=>f.name),l=[],c=t.names();for(let f of i)c.indexOf(f)!==-1?l.push(t.getValue(f)):l.push(null);s!=null&&(s.maxNumTensors=-1/0,s.minNumTensors=1/0);let u=i.join(",")+"|"+t.names().join(","),d,p;if(sA[u]==null){let f=EB(o,t);d=f.sorted,p=f.recipientCounts,sA[u]=d,Uw[u]=p}d=sA[u],p={},r||Object.assign(p,Uw[u]);let h=new Al(t);for(let f=0;f<d.length;++f){if(s!=null){let R=ef().numTensors;R>s.maxNumTensors&&(s.maxNumTensors=R),R<s.minNumTensors&&(s.minNumTensors=R)}let m=d[f],g=m.sourceLayer;if(g instanceof ju)continue;let A=[],x=[],y=[],b=!1;for(let R of m.inputs){let F=h.getValue(R),_=h.getMask(R);A.push(F),x.push(_),_!=null&&(b=!0),r||(p[R.name]--,p[R.name]===0&&!t.hasKey(R)&&i.indexOf(R.name)===-1&&!F.isDisposed&&R.sourceLayer.stateful!==!0&&y.push(F))}b&&(n=n||{},n.mask=x[0]);let w=kt(g.apply(A,n)),k=null;g.supportsMasking&&(k=g.computeMask(A,x));let C=$B(m),N=Array.isArray(C)?C:[C];for(let R=0;R<N.length;++R){h.hasKey(N[R])||h.add(N[R],w[R],Array.isArray(k)?k[0]:k);let F=i.indexOf(N[R].name);F!==-1&&(l[F]=w[R])}r||se(y)}return h.disposeMasks(),a?l:l[0]}function EB(e,t){v.assert(e!=null&&e.length>0,()=>"Expected at least one fetch, got none");let n=[],s={};if(e.length===1){let r=Gw(e[0],t);n=r.sorted,s=r.recipientMap}else{let r=new Set;for(let a of e){let{sorted:o,recipientMap:i}=Gw(a,t);for(let l of o)r.has(l.name)||(n.push(l),r.add(l.name));for(let l in i)s[l]==null&&(s[l]=new Set),i[l].forEach(c=>s[l].add(c))}}return{sorted:n,recipientCounts:RB(s)}}function RB(e){let t={};for(let n in e)t[n]=e[n].size;return t}function Gw(e,t){let n=new Set,s=[],r={};for(let i of t.names())n.add(i);let a=[],o=[];for(a.push(e);a.length>0;){let i=a[a.length-1];if(n.has(i.name)){a.pop();continue}let l=o[o.length-1]===a.length-1;if(i.inputs.length===0||l)a.pop(),s.push(i),n.add(i.name),l&&o.pop();else{o.push(a.length-1);for(let c of i.inputs)r[c.name]==null&&(r[c.name]=new Set),r[c.name].add(i.name),!n.has(c.name)&&a.push(c)}}return{sorted:s,recipientMap:r}}function $B(e){let t;if(e.sourceLayer.inboundNodes.length===1)t=e.sourceLayer.output;else{let n=null;for(let s=0;s<e.sourceLayer.inboundNodes.length;++s)for(let r of e.sourceLayer.inboundNodes[s].outputTensors)if(r.id===e.id){n=s;break}t=e.sourceLayer.getOutputAt(n)}return t}var Pr=class extends Qe{constructor(e){super({});if(this.containerNodes=new Set,this.name=e.name,this.name==null){let A=this.getClassName().toLowerCase();this.name=jf(A)}if(this.supportsMasking=!1,this.trainable_=!0,Array.isArray(e.inputs)?this.inputs=e.inputs.slice():this.inputs=[e.inputs],Array.isArray(e.outputs)?this.outputs=e.outputs.slice():this.outputs=[e.outputs],Do(this.inputs).length!==this.inputs.length)throw new H(`The list of inputs passed to the model is redundant. All inputs should only appear once. Found: ${this.inputs.map(A=>A.name)}`);Do(this.outputs).length!==this.outputs.length&&console.warn(`The list of outputs passed to the model is redundant. All outputs should only appear once. Found: ${this.outputs.map(A=>A.name)}`),this.inputLayers=[],this.inputLayersNodeIndices=[],this.inputLayersTensorIndices=[],this.outputLayers=[],this.outputLayersNodeIndices=[],this.outputLayersTensorIndices=[],this.layers=[],this.internalContainerRefs=[];for(let A of this.outputs){let x=A.sourceLayer,y=A.nodeIndex,b=A.tensorIndex;this.outputLayers.push(x),this.outputLayersNodeIndices.push(y),this.outputLayersTensorIndices.push(b)}for(let A of this.inputs){let x=A.sourceLayer,y=A.nodeIndex,b=A.tensorIndex;$r(y===0,"input layer has >1 nodes"),$r(b===0,"input layer has >1 tensors"),this.inputLayers.push(x),this.inputLayersNodeIndices.push(y),this.inputLayersTensorIndices.push(b)}this.inputNames=[],this.outputNames=[],this.feedInputShapes=[],this.feedInputNames=[],this.feedOutputNames=[];for(let A=0;A<this.inputLayers.length;A++){let x=this.inputLayers[A];if(!(x instanceof ju))throw new TypeError(`Input layers to a LayersModel must be InputLayer objects. Received inputs: ${e.inputs}. Input ${A} (0-based) originates from layer type ${x.getClassName()}.`);this.inputNames.push(x.name),this.feedInputShapes.push(x.batchInputShape),this.feedInputNames.push(x.name)}for(let A of this.outputLayers)this.outputNames.push(A.name);this.internalInputShapes=this.inputs.map(A=>A.shape),this.internalOutputShapes=this.outputs.map(A=>A.shape);let t={},n={},s={},r={},a={},o=[],i=(A,x,y,b,w,k)=>{(b==null||w==null||k==null)&&(b=A.sourceLayer,w=A.nodeIndex,k=A.tensorIndex);let C=b.inboundNodes[w];if(y.indexOf(C)!==-1)throw new cr(`The tensor ${A.name} at layer "${b.name}" is part of a cycle.`);if(x.indexOf(C)!==-1)return;this.containerNodes.add(Pr.nodeKey(b,w)),b.id in a||(a[b.id]=Object.keys(a).length),y.indexOf(C)===-1&&y.push(C);let N=C.inboundLayers.length;for(let R=0;R<N;R++){let F=C.inputTensors[R],_=C.inboundLayers[R],P=C.nodeIndices[R],T=C.tensorIndices[R];i(F,x,y,_,P,T)}for(x.push(C);y.indexOf(C)>=0;)y.splice(y.indexOf(C),1);o.push(C)},l=[],c=[];for(let A of this.outputs)i(A,l,c);let u=o.slice().reverse();for(let A of u){n[A.id]=A,A.id in t||(t[A.id]=0);let x=t[A.id],y=s[A.outboundLayer.id]==null?0:s[A.outboundLayer.id];x=Math.max(x,y),s[A.outboundLayer.id]=x,r[A.outboundLayer.id]=A.outboundLayer,t[A.id]=x;for(let b=0;b<A.inboundLayers.length;b++){let w=A.inboundLayers[b],k=A.nodeIndices[b],C=w.inboundNodes[k],N=t[C.id]==null?0:t[C.id];t[C.id]=Math.max(x+1,N),n[C.id]=C}}let d={};for(let A in t){let x=t[A];x in d||(d[x]=[]),d[x].push(n[A])}let p={};for(let A in s){let x=s[A];x in p||(p[x]=[]),p[x].push(r[A])}let h=Object.keys(p).map(A=>parseInt(A,10)).sort(Pf);this.layers=[];for(let A of h){let x=p[A];x.sort((y,b)=>{let w=a[y.id],k=a[b.id];return w<k?-1:w>k?1:0});for(let y of x)y instanceof Pr&&this.internalContainerRefs.push(y),this.layers.push(y)}this.layersByDepth=p,h=Object.keys(d).map(A=>parseInt(A,10)).sort(Pf);let f=this.inputs.slice(),m=[];for(let A of h)for(let x of d[A]){let y=x.outboundLayer;if(y!=null){for(let b of x.inputTensors)if(f.indexOf(b)===-1)throw new cr(`Graph disconnected: cannot obtain value for tensor ${b} at layer "${y.name}". The following previous layers were accessed without issue: ${m}`);for(let b of x.outputTensors)f.push(b);m.push(y.name)}}this.nodesByDepth=d;let g=this.layers.map(A=>A.name);for(let A of g){let x=g.filter(y=>y===A).length;if(x!==1)throw new cr(`The name "${A}" is used ${x} times in the model. All layer names should be unique. Layer names: `+JSON.stringify(g))}this.outboundNodes=[],this.inboundNodes=[],new Kf({outboundLayer:this,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:this.inputs,outputTensors:this.outputs,inputMasks:this.inputs.map(A=>null),outputMasks:this.outputs.map(A=>null),inputShapes:this.inputs.map(A=>A.shape),outputShapes:this.outputs.map(A=>A.shape)}),this.built=!0,this._refCount=1}assertNotDisposed(){if(this._refCount===0)throw new Error(`Container '${this.name}' is already disposed.`)}dispose(){this.assertNotDisposed();let e={refCountAfterDispose:null,numDisposedVariables:0};if(--this._refCount==0){for(let t of this.layers)e.numDisposedVariables+=t.dispose().numDisposedVariables;for(let t of this.internalContainerRefs)e.numDisposedVariables+=t.dispose().numDisposedVariables}return e.refCountAfterDispose=this._refCount,e}get trainable(){return this.trainable_}set trainable(e){this.layers.forEach(t=>{t._trainableWeights.forEach(n=>n.trainable=e)}),this.trainable_=e}get trainableWeights(){if(this._trainableWeights.length>0)throw new H("Container instance unexpectedly contains _trainableWeights.The trainable weights of a Container are a union of the trainable weights of its consituent Layers. Its own _trainableWeights must remain an empty Array.");if(!this.trainable)return[];let e=[];for(let t of this.layers)e=e.concat(t.trainableWeights);return e}get nonTrainableWeights(){let e=[];for(let t of this.layers)e.push(...t.nonTrainableWeights);if(!this.trainable){let t=[];for(let n of this.layers)t.push(...n.trainableWeights);return t.concat(e)}return e}get weights(){return this.trainableWeights.concat(this.nonTrainableWeights)}loadWeights(e,t=!0){let n={},s=0;for(let a of this.layers)for(let o of a.weights){if(n[o.originalName]!=null)throw new H(`Duplicate weight name: ${o.originalName}`);n[o.originalName]=o,s++}let r=[];for(let a in e){let o=a;if(n[a]==null){let i=a.split("/");o=i.slice(0,-2).concat([i[i.length-1]]).join("/")}if(n[o]!=null)r.push([n[o],e[a]]);else if(t)throw new H(`Provided weight data has no target variable: ${a}`);delete n[o]}if(t){let a=[];for(let o in n)a.push(o);if(a.length>0)throw new H(`${a.length} of ${s} weights are not set: ${a}`)}q1(r)}updatedConfig(){let e=this.getConfig(),t={};return t.className=this.getClassName(),t.config=e,t.kerasVersion=`tfjs-layers ${nA}`,t.backend="TensorFlow.js",t}toJSON(e,t=!0){let n=tA(this.updatedConfig());return t?JSON.stringify(n):n}call(e,t){return q(()=>{e=kt(e);let n=new Al;for(let s=0;s<this.inputs.length;++s)n.add(this.inputs[s],e[s]);return Hd(this.outputs,n,t)})}computeMask(e,t){return q(()=>{e=kt(e);let n;return t==null?n=dl(null,e.length):n=kt(t),this.runInternalGraph(e,n)[1]})}computeOutputShape(e){let t=qf(e);if(t.length!==this.inputLayers.length)throw new H(`Invalid inputShape argument ${e}: model has ${this.inputLayers.length} tensor inputs.`);let n={};for(let o=0;o<t.length;o++){let i=this.inputLayers[o],l=t[o],c=i.name+"_0_0";n[c]=l}let s=Object.keys(this.nodesByDepth).map(o=>parseInt(o,10)).sort(Pf);if(s.length>1)for(let o of s){let i=this.nodesByDepth[o];for(let l of i){let c=l.outboundLayer;if(this.inputLayers.map(f=>f.id).indexOf(c.id)!==-1)continue;let u=[];for(let f=0;f<l.inboundLayers.length;f++){let m=l.inboundLayers[f],g=l.nodeIndices[f],A=l.tensorIndices[f],x=`${m.name}_${g}_${A}`,y=n[x];u.push(y)}let d=c.computeOutputShape(rs(u)),p=qf(d),h=c.inboundNodes.indexOf(l);for(let f=0;f<p.length;f++){let m=`${c.name}_${h}_${f}`;n[m]=p[f]}}}let r=[],a=[];for(let o=0;o<this.outputLayers.length;o++){let i=this.outputLayers[o],l=this.outputLayersNodeIndices[o],c=this.outputLayersTensorIndices[o],u=`${i.name}_${l}_${c}`;a.push(u)}for(let o=0;o<a.length;o++){let i=a[o];$r(i in n),r.push(n[i])}return rs(r)}runInternalGraph(e,t){t==null&&(t=dl(null,e.length));let n={};for(let i=0;i<this.inputs.length;++i){let l=this.inputs[i],c=e[i],u=t[i];n[l.id]=[c,u]}let s=Object.keys(this.nodesByDepth).map(i=>parseInt(i,10)).sort(Pf);for(let i of s){let l=this.nodesByDepth[i];for(let c of l){let u=c.outboundLayer,d=c.inputTensors,p=c.outputTensors,h=new Array;for(let f of d)f.id in n&&h.push(n[f.id]);if(h.length===d.length){let f={},m,g,A,x;if(c.callArgs!=null&&(f=c.callArgs),h.length===1){let[y,b]=h[0];f.mask==null&&(f.mask=b),A=kt(u.call(y,f)),x=kt(u.computeMask(y,b)),m=[y],g=[b]}else m=h.map(y=>y[0]),g=h.map(y=>y[1]),f.mask==null&&(f.mask=g),A=kt(u.call(m,f)),x=kt(u.computeMask(m,g));if(u.activityRegularizer)throw new Le("LayersModel invocation with concrete Tensor value(s) in the presence of activity regularizer(s) is not supported yet.");for(let y=0;y<p.length;++y){let b=p[y],w=A[y],k=x[y];n[b.id]=[w,k]}}}}let r=[],a=[],o=[];for(let i of this.outputs){$r(i.id in n,`Could not compute output ${i.name} : ${i.id}`);let[l,c]=n[i.id];o.push(l.shape),r.push(l),a.push(c)}return[r,a,o]}buildNodeConversionMap(e){let t={},n;for(let s of this.layers){n=s instanceof Pr?1:0;for(let r=0;r<s.inboundNodes.length;r++){let a=Pr.nodeKey(s,r);this.containerNodes.has(a)&&(t[a]=n,n+=1)}}return t}getLayer(e,t){if(t!=null){if(this.layers.length<=t)throw new H(`Was asked to retrieve layer at index ${t}, but model only has ${this.layers.length} layer(s).`);return this.layers[t]}else if(e==null)throw new H("Provide either a layer name or layer index");for(let n of this.layers)if(n.name===e)return n;throw new H(`No such layer: ${e}`)}calculateLosses(){return q(()=>{let e=[];for(let t of this.layers)for(let n=0;n<t.inboundNodes.length;++n){let s=Pr.nodeKey(t,n);this.containerNodes.has(s)&&e.push(...t.calculateLosses())}return e})}getConfig(){let e={name:this.name},t=this.buildNodeConversionMap(this.layers),n=[];for(let a of this.layers){let o=a.getClassName(),i=a.getConfig(),l=[];for(let u=0;u<a.inboundNodes.length;u++){let d=a.inboundNodes[u],p=Pr.nodeKey(a,u),h={};if(this.containerNodes.has(p)){if(d.callArgs)try{JSON.stringify(d.callArgs),h=d.callArgs}catch(f){console.warn(`Layer ${a.name} was passed non-serializable keyword arguments: ${d.callArgs}. They will not be included in the serialized model (and thus will be missing at deserialization time).`),h={}}if(d.inboundLayers.length>0){let f=[];for(let m=0;m<d.inboundLayers.length;m++){let g=d.inboundLayers[m],A=d.nodeIndices[m],x=d.tensorIndices[m],y=Pr.nodeKey(g,A),b=t[y];b==null&&(b=0),f.push([g.name,b,x,h])}l.push(f)}}}let c={};c.name=a.name,c.className=o,c.config=i,c.inboundNodes=l,n.push(c)}e.layers=n;let s=[];for(let a=0;a<this.inputLayers.length;a++){let o=this.inputLayers[a],i=this.inputLayersNodeIndices[a],l=Pr.nodeKey(o,i);if(!this.containerNodes.has(l))continue;let c=t[l];c==null&&(c=0);let u=this.inputLayersTensorIndices[a];s.push([o.name,c,u])}e.inputLayers=s;let r=[];for(let a=0;a<this.outputLayers.length;a++){let o=this.outputLayers[a],i=this.outputLayersNodeIndices[a],l=Pr.nodeKey(o,i);if(!this.containerNodes.has(l))continue;let c=t[l];c==null&&(c=0);let u=this.outputLayersTensorIndices[a];r.push([o.name,c,u])}return e.outputLayers=r,e}static fromConfig(e,t,n={},s=!1){let r={},a={};function o(m,g){m.name in a?a[m.name].push(g):a[m.name]=[g]}function i(m,g){let A=[],x;for(let y of g){let b=y[0],w=y[1],k=y[2];if(x=y[3]==null?{}:y[3],!(b in r)){o(m,g);return}let C=r[b];if(C.inboundNodes.length<=w){o(m,g);return}let N=C.inboundNodes[w];A.push(N.outputTensors[k])}A.length>0&&m.apply(rs(A),x)}function l(m){let g=m.name,A=fr(m,t.customObjects!=null?t.customObjects:{});A.setFastWeightInitDuringBuild(s),r[g]=A,m.inboundNodes.forEach(y=>{if(!(y instanceof Array))throw new H(`Corrupted configuration, expected array for nodeData: ${y}`);o(A,y)})}let c=t.name,u=t.layers;for(let m of u)l(m);for(;!lL(a);)for(let m of u){let g=r[m.name];if(g.name in a){let A=a[g.name];delete a[g.name];for(let x of A)i(g,x)}}let d=[],p=[],h=t.inputLayers;for(let m of h){let g=m[0],A=m[1],x=m[2];$r(g in r);let b=r[g].inboundNodes[A].outputTensors;d.push(b[x])}let f=t.outputLayers;for(let m of f){let g=m[0],A=m[1],x=m[2];$r(g in r);let b=r[g].inboundNodes[A].outputTensors;p.push(b[x])}return new e({inputs:d,outputs:p,name:c})}get stateful(){if(this._stateful)throw new H("Container instance unexpectedly has _stateful = true. The statefulness of a Container is determined by the Layers it contains. Its _stateful property must remain the default false.");for(let e of this.layers)if(e.stateful)return!0;return!1}resetStates(){q(()=>{this.layers.forEach(e=>{e.stateful&&e.resetStates()})})}};function _B(e,t,n){let s=t.length;if(e==null||Array.isArray(e)&&e.length===0)return t.map(r=>null);if(s===1)return Array.isArray(e)&&e.length===1?e:typeof e=="object"&&t[0]in e?[e[t[0]]]:[e];if(Array.isArray(e)){if(e.length!==s)throw new Error(`Provided ${n} is an array of ${e.length} element(s), but the model has ${s} outputs. Make sure a set of weights is provided for each model output.`);return e}else if(typeof e=="object"&&Object.keys(e).length>0&&typeof e[Object.keys(e)[0]]=="object"){let r=[];return t.forEach(a=>{a in e?r.push(e[a]):r.push(null)}),r}else throw new Error(`The model has multiple (${s}) outputs, so ${n} must be either an array with ${s} elements or an object with ${t} keys. Provided ${n} not understood: ${JSON.stringify(e)}`)}function Hw(e,t){return _B(e,t,"classWeight")}async function jw(e,t,n,s){if(t!=null||s!=null)throw new Error("Support sampleWeight is not implemented yet");if(n!=null){let r=q(()=>{if(e.shape.length===1)return Ln(e);if(e.shape.length===2){if(e.shape[1]>1)return Hs(e,1);if(e.shape[1]===1)return G(e,[e.shape[0]]);throw new Error(`Encountered unexpected last-dimension size (${e.shape[1]}) during handling of class weights. The size is expected to be >= 1.`)}else throw new Error(`Unexpected rank of target (y) tensor (${e.rank}) during handling of class weights. The rank is expected to be 1 or 2.`)}),a=Array.from(await r.data());se(r);let o=[];return a.forEach(i=>{if(n[i]==null)throw new Error(`classWeight must contain all classes in the training data. The class ${i} exists in the data but not in classWeight`);o.push(n[i])}),Xt(o,"float32")}else return null}function DB(e,t){return L(e,t)}var PB=32;function qw(e,t){let n,s,r=t;n=r.xs,s=r.ys,v.assert(n!=null&&s!=null,()=>`A Dataset iterator for fitDataset() is expected to generate objects of the form \`{xs: xVal, ys: yVal}\`, where the two values may be \`tf.Tensor\`, an array of Tensors, or a map of string to Tensor. The provided Dataset instead generates ${t}`);let a=Xw("input",e.inputNames,n),o=Xw("output",e.outputNames,s),i=a[0].shape[0];v.assert(a.length===e.inputs.length,()=>`LayersModel has ${e.inputs.length} inputs, but the dataset provides ${a.length} inputs. (Expected input keys: ${JSON.stringify(e.inputNames)})`),v.assert(o.length===e.outputs.length,()=>`LayersModel has ${e.outputs.length} outputs, but the dataset provides ${o.length} outputs. (Expected output keys: ${JSON.stringify(e.outputNames)})`);for(let l=0;l<a.length;l++)v.assert(a[l].shape[0]===i,()=>`Batch size mismatch: input ${e.inputNames[l]} has ${a[l].shape[0]}; expected ${i} based on input ${e.inputNames[0]}.`);for(let l=0;l<o.length;l++)v.assert(o[l].shape[0]===i,()=>`Batch size mismatch: output ${e.outputNames[l]} has ${o[l].shape[0]}; expected ${i} based on input ${e.inputNames[0]}.`);return{xs:a,ys:o}}function Xw(e,t,n){if(n instanceof Ze)return[n];if(Array.isArray(n))return v.assert(n.length===t.length,()=>`Received an array of ${n.length} Tensors, but expected ${t.length} to match the ${e} keys ${t}.`),n;{let s=[];for(let r of t){if(n[r]==null)throw new H(`The feature data generated by the dataset lacks the required ${e} key '${r}'.`);s.push(n[r])}return s}}function FB(e){if(e.length===3)throw new Le("Validation with sample weights is not implemented yet.");return{xs:e[0],ys:e[1]}}async function OB(e,t,n){let s=n.batchesPerEpoch!=null;if(v.assert(e.optimizer!=null,()=>"You must compile a model before training/testing. Use LayersModel.compile(modelCompileConfig)."),v.assert(n!=null,()=>"For fitDataset(), the 2nd argument (config) is required, but it is not provided in this call."),v.assert(n.epochs!=null&&n.epochs>0&&Number.isInteger(n.epochs),()=>`For fitDataset(), config.epochs is expected to be a positive integer, but got ${n.epochs}`),v.assert(!s||n.batchesPerEpoch>0&&Number.isInteger(n.batchesPerEpoch),()=>`For fitDataset(), config.batchesPerEpoch is expected to be a positive integer if specified, but got ${n.batchesPerEpoch}`),v.assert(n.validationSplit==null,()=>"`validationSplit` is not supported by `fitDataset()`. Use validationData instead."),e.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");e.isTraining=!0;try{let r=n.validationData!=null,a,o;if(r)if(Kw(n.validationData))v.assert(n.validationBatches==null||n.validationBatches>0&&Number.isInteger(n.validationBatches),()=>`For fitDataset() with dataset-based validation, config.validationBatches is expected not to be provided, or to be a positive integer, but got ${n.validationBatches}`);else{let g=FB(n.validationData);a=g.xs,o=g.ys}let i=e.makeTrainFunction(),l=e.getDedupedMetricsNames(),c;r?c=l.slice().concat(l.map(g=>"val_"+g)):c=l.slice();let u=Dw(n.callbacks,n.yieldEvery),d=n.verbose==null?1:n.verbose,{callbackList:p,history:h}=Pw(u,d,n.epochs,null,null,MB(t,n),null,r,c);p.setModel(e),e.history=h,await p.onTrainBegin(),e.stopTraining_=!1;let f=n.initialEpoch==null?0:n.initialEpoch,m=await t.iterator();for(;f<n.epochs;){let g={};await p.onEpochBegin(f);let A=0,x=0;for(s||(m=await t.iterator());s?A<n.batchesPerEpoch:!0;){let y=await m.next();if(s&&y.done){console.warn(`You provided \`batchesPerEpoch\` as ${n.batchesPerEpoch}, but your dataset iterator ran out of data after ${A} batches; interrupting training. Make sure that your dataset can generate at least \`batchesPerEpoch * epochs\` batches (in this case, ${n.batchesPerEpoch*n.epochs} batches). You may need to use the repeat() function when building your dataset.`);break}if(y.value!=null){let{xs:b,ys:w}=qw(e,y.value),k={};k.batch=x,k.size=b[0].shape[0],await p.onBatchBegin(x,k);let C=[];if(n.classWeight!=null){let F=Hw(n.classWeight,e.outputNames);for(let _=0;_<F.length;++_)C.push(await jw(w[_],null,F[_]))}let N=b.concat(w).concat(C),R=i(N);se(N);for(let F=0;F<l.length;++F){let _=l[F],P=R[F];k[_]=P,fn(P)}await p.onBatchEnd(x,k),Nw(k),x++,A++}if(s?A>=n.batchesPerEpoch:y.done){if(r){let b;Kw(n.validationData)?b=kt(await e.evaluateDataset(n.validationData,{batches:n.validationBatches})):b=kt(e.evaluate(a,o,{batchSize:n.validationBatchSize==null?PB:n.validationBatchSize,verbose:0}));for(let w=0;w<e.metricsNames.length;++w)g[`val_${e.metricsNames[w]}`]=b[w]}break}if(e.stopTraining_)break}if(await p.onEpochEnd(f,g),f++,e.stopTraining_)break}return await p.onTrainEnd(),await e.history.syncData(),e.history}finally{e.isTraining=!1}}function MB(e,t){let n=null;return t.batchesPerEpoch!=null?n=t.batchesPerEpoch:Number.isFinite(e.size)&&(n=e.size),n}function Kw(e){return typeof e.iterator=="function"}function zB(e){return typeof e.next=="function"}async function LB(e,t,n){n=n||{};let s=n.batches!=null,r=e.testFunction,a=[];if(n.verbose>0)throw new Le("Verbose mode is not implemented yet.");v.assert(!s||n.batches>0&&Number.isInteger(n.batches),()=>`Test loop expects \`batches\` to be a positive integer, but received ${JSON.stringify(n.batches)}`);let o=zB(t)?t:await t.iterator(),i=0,l=0;for(;s?l<n.batches:!0;){let c=await o.next();if(a=q(()=>{if(c.value){let{xs:u,ys:d}=qw(e,c.value),p=u.concat(d),h=q(()=>r(p));if(se(p),l===0)for(let m=0;m<h.length;++m)a.push(Re(0));let f=p[0].shape[0];for(let m=0;m<h.length;++m){let g=h[m],A=a[m];a[m]=q(()=>le(a[m],L(f,g))),l>0&&se(A)}se(h),i+=f,++l}return a}),c.done){s&&console.warn(`Your dataset iterator ran out of data during evaluateDataset(). Interrupting evalution. Make sure that your dataset can generate at least \`batches\` batches (in this case, ${n.batches} batches). You may need to use the repeat() function when building your dataset.`);break}}for(let c=0;c<a.length;++c){let u=a[c];a[c]=pe(a[c],i),se(u)}return rs(a)}function rA(e){v.assert(e>0&&Number.isInteger(e),()=>`batchSize is required to be a positive integer, but got ${e}`)}function jd(e,t,n){return e==null?[null]:Array.isArray(e)?e.map(s=>ml(s,t,n-t)):ml(e,t,n-t)}function aA(e,t){return q(()=>e==null?null:Array.isArray(e)?e.map(n=>aA(n,t)):yw(e,t.dtype==="int32"?t:de(t,"int32")))}function oA(e,t){let n=[],s=0,r=null;for(;s<e;)r=s+t,r>=e&&(r=e),n.push([s,r]),s=r;return n}async function BB(e,t,n,s,r,a,o,i,l,c,u,d,p,h,f){r==null&&(r=32),a==null&&(a=1),u==null&&(u=!0),p==null&&(p=0);let m=!1;if(l!=null&&c!=null&&(m=!0),f!=null&&(m=!0,h==null))throw new H("Can only use `validationSteps` when doing step-wise training, i.e., `stepsPerEpoch` must be set.");let g=e.checkNumSamples(n,r,h,"steps_per_epoch"),A;g!=null&&(A=dr(0,g)),o==null&&(o=1);let{callbackList:x,history:y}=Pw(i,o,a,p,g,h,r,m,d);x.setModel(e),e.history=y,await x.onTrainBegin(),e.stopTraining_=!1;for(let b=p;b<a;++b){await x.onEpochBegin(b);let w={};if(h!=null)throw new Le("stepsPerEpoch mode is not implemented yet.");{if(u==="batch")throw new Le("batch shuffling is not implemneted yet");u&&v.shuffle(A);let k=Xt(A),C=oA(g,r);for(let N=0;N<C.length;++N){let R={};if(await x.onBatchBegin(N,R),q(()=>{let F=C[N][0],_=C[N][1],P=ml(k,F,_-F);R.batch=N,R.size=_-F;let T=aA(n,P),M=t(T);for(let U=0;U<s.length;++U){let j=s[U],z=M[U];R[j]=z,fn(z)}if(N===C.length-1&&m){let U=e.testLoop(l,c,r);for(let j=0;j<s.length;++j){let z=s[j],X=U[j];fn(X),w["val_"+z]=X}}}),await x.onBatchEnd(N,R),Nw(R),e.stopTraining_)break}k.dispose()}if(await x.onEpochEnd(b,w),e.stopTraining_)break}return await x.onTrainEnd(),await e.history.syncData(),e.history}async function WB(e,t,n,s={}){if(e.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");e.isTraining=!0;let r,a,o,i,l,c,u;try{let d=s.batchSize==null?32:s.batchSize;rA(d);let p=!1,h=await e.standardizeUserData(t,n,s.sampleWeight,s.classWeight,p,d);r=h[0],a=h[1],u=h[2];let f=!1,m;if(s.validationData!=null&&s.validationData.length>0){if(f=!0,s.validationData.length===2)o=s.validationData[0],i=s.validationData[1];else throw s.validationData.length===3?new Le("validationData including sample weights is not supported yet."):new H(`When passing validation data, it must contain 2 (valX, valY) or 3 (valX, valY, valSampleWeight) items; ${s.validationData} is invalid.`);let C=!0,N=await e.standardizeUserData(o,i,null,null,C,d);l=N[0],c=N[1],m=l.concat(c)}else if(s.validationSplit!=null&&s.validationSplit>0&&s.validationSplit<1){f=!0;let C=Math.floor(r[0].shape[0]*(1-s.validationSplit)),N=r[0].shape[0];l=jd(r,C,N),r=jd(r,0,C),c=jd(a,C,N),a=jd(a,0,C),m=l.concat(c)}else s.validationSteps!=null&&(f=!0);let g=r.concat(a).concat(u);e.checkTrainableWeightsConsistency();let A=e.makeTrainFunction(),x=e.getDedupedMetricsNames(),y,b;f?(e.makeTestFunction(),y=e.testFunction,b=x.slice().concat(x.map(C=>"val_"+C))):(y=null,m=[],b=x.slice());let w=Dw(s.callbacks,s.yieldEvery);return await BB(e,A,g,x,d,s.epochs,s.verbose,w,y,m,s.shuffle,b,s.initialEpoch,null,null)}finally{e.isTraining=!1,yl(r,t),yl(a,n),yl(l,o),yl(c,i),u!=null&&se(u)}}function Zw(e){let t=[];e instanceof Ze&&(e=[e]);for(let n=0;n<e.length;++n){let s=e[n];if(s.rank===1)t.push(Bd(s,1));else{if(s.rank===0)throw new Error("Expected tensor to be at least 1D, but received a 0D tensor (scalar).");t.push(s)}}return t}function yl(e,t){if(e==null)return;let n=[];if(t instanceof Ze)n.push(t.id);else if(Array.isArray(t))t.forEach(r=>n.push(r.id));else if(t!=null)for(let r in t){let a=t[r];n.push(a.id)}let s=[];if(e instanceof Ze)n.indexOf(e.id)===-1&&s.push(e);else if(Array.isArray(e))e.forEach(r=>{n.indexOf(r.id)===-1&&s.push(r)});else if(e!=null)for(let r in e){let a=e[r];n.indexOf(a.id)===-1&&s.push(a)}s.forEach(r=>{r.isDisposed||r.dispose()})}function VB(e){return e instanceof Ze}function iA(e){return Array.isArray(e)}function Yw(e){return!VB(e)&&!iA(e)}function Jw(e,t,n,s=!0,r=""){if(t==null||t.length===0){if(e!=null){let o=!1;if(iA(e)&&e.length>0)o=!0;else if(Yw(e)){for(let i in e)if(e.hasOwnProperty(i)){o=!0;break}}else o=!0;if(o)throw new H(`Error when checking model ${r} expected no data, but got ${e}`)}return[]}if(e==null)return t.map(o=>null);let a;if(Yw(e)){e=e,a=[];for(let o of t){if(e[o]==null)throw new H(`No data provided for "${o}". Need data for each key in: ${t}`);a.push(e[o])}}else if(iA(e)){if(e=e,e.length!==t.length)throw new H(`Error when checking model ${r}: the Array of Tensors that you are passing to your model is not the size the model expected. Expected to see ${t.length} Tensor(s), but instead got the following list of Tensor(s): ${e}`);a=e}else{if(e=e,t.length>1)throw new H(`The model ${r} expects ${t.length} Tensor(s), but only received one Tensor. Found: Tensor with shape ${e.shape}`);a=[e]}if(a=Zw(a),n!=null)for(let o=0;o<t.length;++o){if(n[o]==null)continue;let i=a[o];if(i.shape.length!==n[o].length)throw new H(`Error when checking ${r}: expected ${t[o]} to have ${n[o].length} dimension(s). but got array with shape ${i.shape}`);for(let l=0;l<n[o].length;++l){if(l===0&&!s)continue;let c=i.shape[l],u=n[o][l];if(u!=null&&u>=0&&c!==u)throw new H(`${r} expected a batch of elements where each example has shape [${n[o].slice(1,n[o].length)}] (i.e.,tensor shape [*,${n[o].slice(1,n[o].length)}]) but the ${r} received an input with ${i.shape[0]} examples, each with shape [${i.shape.slice(1,i.shape.length)}] (tensor shape [${i.shape}])`)}}return a}function UB(e,t,n){let s=Do(e.map(a=>a.shape[0]));s.sort();let r=Do(t.map(a=>a.shape[0]));if(r.sort(),s.length>1)throw new H(`All input Tensors (x) should have the same number of samples. Got array shapes: ${JSON.stringify(e.map(a=>a.shape))}`);if(r.length>1)throw new H(`All target Tensors (y) should have the same number of samples. Got array shapes: ${JSON.stringify(t.map(a=>a.shape))}`);if(s.length>0&&r.length>0&&!v.arraysEqual(s,r))throw new H(`Input Tensors should have the same number of samples as target Tensors. Found ${s[0]} input sample(s) and ${r[0]} target sample(s).`)}function GB(e,t,n){let s=[gl,Qf,Ud];for(let r=0;r<e.length;++r){let a=e[r],o=t[r],i=n[r];if(o!=null){if(o===Ud&&a.shape[a.shape.length-1]===1)throw new H(`You are passing a target array of shape ${a.shape} while using a loss 'categorical_crossentropy'. 'categorical_crossentropy'expects targets to be binary matrices (1s and 0s) of shape [samples, classes].`);if(s.indexOf(o)!==-1){let l=a.shape.slice(1),c=i.slice(1);for(let u=0;u<l.length;++u){let d=l[u],p=c[u];if(p!=null&&d!==p)throw new H(`A target Tensor with shape ${a.shape} was passed for an output of shape ${i}, while using a loss function that expects targets to have the same shape as the output.`)}}}}}function Qw(e,t,n,s=!0,r=""){let a;if(Array.isArray(e)){if(e.length!==t.length)throw new H(`Error when checking model ${r}: the Array of Tensors that you are passing to your model is not the size the the model expected. Expected to see ${t.length} Tensor(s), but instead got ${e.length} Tensors(s).`);a=e}else{if(t.length>1)throw new H(`The model expects ${t.length} ${r} Tensors, but only received one Tensor. Found: array with shape ${JSON.stringify(e.shape)}.`);a=[e]}if(n!=null)for(let o=0;o<t.length;++o){if(n[o]==null)continue;let i=a[o];if(i.shape.length!==n[o].length)throw new H(`Error when checking ${r}: expected ${t[o]} to have ${n[o].length} dimension(s), but got array with shape ${JSON.stringify(i.shape)}`);for(let l=0;l<n[o].length;++l){if(l===0&&!s)continue;let c=i.shape[l],u=n[o][l];if(u!=null&&u!==c)throw new H(`Error when checking ${r}: expected ${t[o]} to have shape ${JSON.stringify(n[o])} but got array with shape ${JSON.stringify(i.shape)}.`)}}}function HB(e,t){if(e==null||Array.isArray(e)&&e.length===0)return t.map(s=>[]);let n;if(typeof e=="string"||typeof e=="function")n=[e];else if(Array.isArray(e)||typeof e=="object")n=e;else throw new TypeError(`Type of metrics argument not understood. Expected an string,function, Array, or Object, found: ${e}`);if(Array.isArray(n))return t.map(s=>n);{let s=[];for(let r of t){let a=n.hasOwnProperty(r)?n[r]:[];Array.isArray(a)||(a=[a]),s.push(a)}return s}}var jB="layers-model",ea=class extends Pr{constructor(e){super(e);this.isTraining=!1}summary(e,t,n=console.log){if(!this.built)throw new H("This model has never been called, thus its weights have not been created yet. So no summary can be displayed. Build the model first (e.g., by calling it on some test data).");kB(this,e,t,n)}compile(e){if(e.loss==null&&(e.loss=[]),this.loss=e.loss,typeof e.optimizer=="string")this.optimizer_=wB(e.optimizer),this.isOptimizerOwned=!0;else{if(!(e.optimizer instanceof Yr))throw new H("User-defined optimizer must be an instance of tf.Optimizer.");this.optimizer_=e.optimizer,this.isOptimizerOwned=!1}let t=[];if(!Array.isArray(e.loss)&&typeof e.loss!="string"&&typeof e.loss!="function"){e.loss=e.loss;for(let a in e.loss)if(this.outputNames.indexOf(a)===-1)throw new H(`Unknown entry in loss dictionary: "${a}". Only expected the following keys: ${this.outputNames}`);for(let a of this.outputNames)e.loss[a]==null&&console.warn(`Output "${a}" is missing from loss dictionary. We assume this was done on purpose, and we will not be expecting data to be passed to ${a} during training`),t.push(Z1(e.loss[a]))}else if(Array.isArray(e.loss)){if(e.loss.length!==this.outputs.length)throw new H(`When passing an Array as loss, it should have one entry per model output. The model has ${this.outputs.length} output(s), but you passed loss=${e.loss}.`);t=e.loss.map(o=>Z1(o))}else{let a=Z1(e.loss);this.outputs.forEach(o=>{t.push(a)})}this.lossFunctions=t,this.feedOutputNames=[],this.feedOutputShapes=[],this.feedLossFns=[];for(let a=0;a<this.outputs.length;++a){let o=this.internalOutputShapes[a],i=this.outputNames[a];this.feedOutputNames.push(i),this.feedOutputShapes.push(o),this.feedLossFns.push(this.lossFunctions[a])}let n=[];this.metrics=e.metrics,this.metricsNames=["loss"],this.metricsTensors=[],fl("loss",()=>{for(let a=0;a<this.outputs.length;++a){if(n.indexOf(a)!==-1)continue;let o=this.lossFunctions[a];this.outputs.length>1&&(this.metricsTensors.push([o,a]),this.metricsNames.push(this.outputNames[a]+"_loss"))}});let s=HB(e.metrics,this.outputNames),r=(a,o,i)=>{this.outputNames.length>1&&(o=this.outputNames[a]+"_"+o),this.metricsNames.push(o),this.metricsTensors.push([i,a])};fl("metric",()=>{for(let a=0;a<this.outputs.length;++a){if(n.indexOf(a)!==-1)continue;let o=s[a];(l=>{let c="",u,d,p;for(let h of l){if(typeof h=="string"&&["accuracy","acc","crossentropy","ce"].indexOf(h)!==-1){let m=this.internalOutputShapes[a];m[m.length-1]===1||this.lossFunctions[a]===Qf?["accuracy","acc"].indexOf(h)!==-1?d=Y1:["crossentropy","ce"].indexOf(h)!==-1&&(d=Mw):this.lossFunctions[a]===Jf?["accuracy","acc"].indexOf(h)!==-1?d=zw:["crossentropy","ce"].indexOf(h)!==-1&&(d=Lw):["accuracy","acc"].indexOf(h)!==-1?d=J1:["crossentropy","ce"].indexOf(h)!==-1&&(d=Q1);let g;["accuracy","acc"].indexOf(h)!==-1?g="acc":["crossentropy","ce"].indexOf(h)!==-1&&(g="ce"),p=d,u=c+g}else p=vB(h),u=c+nm(h);let f;fl(u,()=>{f=p}),r(a,u,f)}})(o)}}),this.collectedTrainableWeights=this.trainableWeights}checkTrainableWeightsConsistency(){this.collectedTrainableWeights!=null&&this.trainableWeights.length!==this.collectedTrainableWeights.length&&console.warn("Discrepancy between trainableweights and collected trainable weights. Did you set `model.trainable` without calling `model.compile()` afterwards?")}evaluate(e,t,n={}){let s=n.batchSize==null?32:n.batchSize;rA(s);let r=!0,a=this.standardizeUserDataXY(e,t,r,s);try{let o=a[0].concat(a[1]);this.makeTestFunction();let i=this.testFunction,l=this.testLoop(i,o,s,n.verbose,n.steps);return rs(l)}finally{yl(a[0],e),yl(a[1],t)}}async evaluateDataset(e,t){return this.makeTestFunction(),LB(this,e,t)}checkNumSamples(e,t,n,s="steps"){let r;if(n!=null){if(r=null,t!=null)throw new H(`If ${s} is set, batchSize must be null or undefined.Got batchSize = ${t}`)}else if(e!=null)Array.isArray(e)?r=e[0].shape[0]:r=e.shape[0];else throw new H(`Either the input data should have a defined shape, or ${s} shoud be specified.`);return r}execute(e,t){if(Array.isArray(t)&&t.length===0)throw new H("`outputs` is an empty Array, which is not allowed.");let n=Array.isArray(t),s=n?t:[t],r=this.retrieveSymbolicTensors(s),a=new Al;if(e instanceof Ze&&(e=[e]),Array.isArray(e)){if(e.length!==this.inputs.length)throw new H(`The number of inputs provided (${e.length}) does not match the number of inputs of this model (${this.inputs.length}).`);for(let i=0;i<this.inputs.length;++i)a.add(this.inputs[i],e[i])}else for(let i of this.inputs){let l=e[i.name];if(l==null)throw new H(`No value is provided for the model's input ${i.name}`);a.add(i,l)}let o=Hd(r,a);return n?o:o[0]}retrieveSymbolicTensors(e){let t=dl(null,e.length),n=e.length;for(let s of this.layers){let r=Array.isArray(s.output)?s.output:[s.output],a=r.map(o=>o.name);for(let o=0;o<e.length;++o){let i=a.indexOf(e[o]);if(i!==-1&&(t[o]=r[i],n--),n===0)break}if(n===0)break}if(n>0){let s=[];throw t.forEach((r,a)=>{r==null&&s.push(e[a])}),new H(`Cannot find SymbolicTensors for output name(s): ${JSON.stringify(s)}`)}return t}predictLoop(e,t=32,n=!1){return q(()=>{let s=this.checkNumSamples(e);if(n)throw new Le("Verbose predictLoop() is not implemented yet.");let r=oA(s,t),a=this.outputs.map(o=>[]);for(let o=0;o<r.length;++o)q(()=>{let l=r[o][0],c=r[o][1],u=jd(e,l,c),d=[];if(Array.isArray(u))for(let h=0;h<u.length;++h)d.push({key:this.inputs[h],value:u[h]});else d.push({key:this.inputs[0],value:u});let p=new Al(d);return Hd(this.outputs,p)}).forEach((l,c)=>a[c].push(l));return rs(a.map(o=>vt(o,0)))})}predict(e,t={}){let n=Zw(e);Qw(n,this.inputNames,this.feedInputShapes,!1);try{let s=t.batchSize==null?32:t.batchSize;return rA(s),this.predictLoop(n,s)}finally{yl(n,e)}}predictOnBatch(e){Qw(e,this.inputNames,this.feedInputShapes,!0);let t=(Array.isArray(e)?e[0]:e).shape[0];return this.predictLoop(e,t)}standardizeUserDataXY(e,t,n=!0,s){if(this.optimizer_==null)throw new cr("You must compile a model before training/testing. Use LayersModel.compile(modelCompileArgs).");let r=[];for(let a=0;a<this.feedOutputShapes.length;++a){let o=this.feedOutputShapes[a];this.feedLossFns[a]===Jf?r.push(o.slice(0,o.length-1).concat([1])):r.push(o)}if(e=Jw(e,this.feedInputNames,this.feedInputShapes,!1,"input"),t=Jw(t,this.feedOutputNames,r,!1,"target"),UB(e,t,null),GB(t,this.feedLossFns,this.feedOutputShapes),this.stateful&&s!=null&&s>0&&e[0].shape[0]%s!=0)throw new H(`In a stateful network, you should only pass inputs with a number of samples that is divisible by the batch size ${s}. Found: ${e[0].shape[0]} sample(s).`);return[e,t]}async standardizeUserData(e,t,n,s,r=!0,a){let[o,i]=this.standardizeUserDataXY(e,t,r,a);if(n!=null)throw new Error("sample weight is not supported yet.");let l=null;if(s!=null){let c=Hw(s,this.outputNames);l=[];for(let u=0;u<c.length;++u)l.push(await jw(i[u],null,c[u]))}return[o,i,l]}testLoop(e,t,n,s=0,r){return q(()=>{let a=this.checkNumSamples(t,n,r,"steps"),o=[];if(s>0)throw new Le("Verbose mode is not implemented yet.");if(r!=null)throw new Le("steps mode in testLoop() is not implemented yet");{let i=oA(a,n),l=Xt(dr(0,a));for(let c=0;c<i.length;++c){let u=i[c][0],d=i[c][1],p=ml(l,u,d-u),h=aA(t,p),f=e(h);if(c===0)for(let m=0;m<f.length;++m)o.push(Re(0));for(let m=0;m<f.length;++m){let g=f[m];o[m]=le(o[m],L(d-u,g))}}for(let c=0;c<o.length;++c)o[c]=pe(o[c],a)}return o})}getDedupedMetricsNames(){let e=this.metricsNames,t=[];for(let n=0;n<e.length;++n){let s=e[n],r=s;ow(e,s)>1&&(r+=`_${ow(e.slice(0,n),s)}`),t.push(r)}return t}makeTrainFunction(){return e=>{let t=[],n=e.slice(0,this.inputs.length),s=e.slice(this.inputs.length,this.inputs.length+this.outputs.length),r=e.slice(this.inputs.length+this.outputs.length,this.inputs.length+this.outputs.length*2),a=[],o=()=>{let u=[];for(let f=0;f<this.inputs.length;++f)u.push({key:this.inputs[f],value:n[f]});let d=new Al(u),p=Hd(this.outputs,d,{training:!0}),h;for(let f=0;f<this.lossFunctions.length;++f){let g=this.lossFunctions[f](s[f],p[f]);r[f]!=null&&(g=DB(g,r[f]));let A=Bt(g);t.push(A),f===0?h=g:h=le(h,g)}for(let f=0;f<this.metricsTensors.length;++f){let m;if(this.outputs.length>1&&f<this.outputs.length)m=t[f];else{let g=this.metricsTensors[f][0],A=this.metricsTensors[f][1];m=Bt(g(s[A],p[A]))}fn(m),a.push(m)}return h=Bt(h),this.calculateLosses().forEach(f=>{h=le(h,f)}),h},i=this.collectedTrainableWeights.map(u=>u.read()),l=!0;return[this.optimizer_.minimize(o,l,i)].concat(a)}}makeTestFunction(){this.testFunction=e=>q(()=>{let t=[],n,s=e.slice(0,this.inputs.length),r=e.slice(this.inputs.length,this.inputs.length+this.outputs.length),a=[];for(let l=0;l<this.inputs.length;++l)a.push({key:this.inputs[l],value:s[l]});let o=new Al(a),i=Hd(this.outputs,o);for(let l=0;l<this.lossFunctions.length;++l){let c=this.lossFunctions[l],u=Bt(c(r[l],i[l]));l===0?n=u:n=le(n,u),t.push(n)}for(let l=0;l<this.metricsTensors.length;++l){let c=this.metricsTensors[l][0],u=this.metricsTensors[l][1],d=Bt(c(r[u],i[u]));t.push(d)}return t})}async fit(e,t,n={}){return WB(this,e,t,n)}async fitDataset(e,t){return OB(this,e,t)}async trainOnBatch(e,t){let n=await this.standardizeUserData(e,t),s=n[0],r=n[1],o=this.makeTrainFunction()(s.concat(r)),i=[];for(let l of o){let c=await l.data();i.push(c[0])}return se(o),rs(i)}getNamedWeights(e){let t=[],n=e!=null&&e.trainableOnly,s=n?this.trainableWeights:this.weights,r=this.getWeights(n);for(let a=0;a<s.length;++a)n&&!s[a].trainable||t.push({name:s[a].originalName,tensor:r[a]});return t}set stopTraining(e){this.stopTraining_=e}get stopTraining(){return this.stopTraining_}get optimizer(){return this.optimizer_}set optimizer(e){this.optimizer_!==e&&(this.optimizer_=e,this.isOptimizerOwned=!1)}dispose(){let e=super.dispose();if(e.refCountAfterDispose===0&&this.optimizer!=null&&this.isOptimizerOwned){let t=ef().numTensors;this.optimizer_.dispose(),e.numDisposedVariables+=t-ef().numTensors}return e}getLossIdentifiers(){let e;if(typeof this.loss=="string")e=Qr(this.loss);else if(Array.isArray(this.loss)){for(let t of this.loss)if(typeof t!="string")throw new Error("Serialization of non-string loss is not supported.");e=this.loss.map(t=>Qr(t))}else{let t=Object.keys(this.loss);e={};let n=this.loss;for(let s of t)if(typeof n[s]=="string")e[s]=Qr(n[s]);else throw new Error("Serialization of non-string loss is not supported.")}return e}getMetricIdentifiers(){if(typeof this.metrics=="string"||typeof this.metrics=="function")return[Qr(nm(this.metrics))];if(Array.isArray(this.metrics))return this.metrics.map(e=>Qr(nm(e)));{let e={};for(let t in this.metrics)e[t]=Qr(nm(this.metrics[t]));return e}}getTrainingConfig(){return{loss:this.getLossIdentifiers(),metrics:this.getMetricIdentifiers(),optimizer_config:{class_name:this.optimizer.getClassName(),config:this.optimizer.getConfig()}}}loadTrainingConfig(e){if(e.weighted_metrics!=null)throw new Error("Loading weight_metrics is not supported yet.");if(e.loss_weights!=null)throw new Error("Loading loss_weights is not supported yet.");if(e.sample_weight_mode!=null)throw new Error("Loading sample_weight_mode is not supported yet.");let t=Gd(e.optimizer_config),n=fr(t),s;if(typeof e.loss=="string")s=pl(e.loss);else if(Array.isArray(e.loss))s=e.loss.map(a=>pl(a));else if(e.loss!=null){s={};for(let a in e.loss)s[a]=pl(e.loss[a])}let r;if(Array.isArray(e.metrics))r=e.metrics.map(a=>pl(a));else if(e.metrics!=null){r={};for(let a in e.metrics)r[a]=pl(e.metrics[a])}this.compile({loss:s,metrics:r,optimizer:n})}async save(e,t){if(typeof e=="string"){let l=ts.getSaveHandlers(e);if(l.length===0)throw new H(`Cannot find any save handlers for URL '${e}'`);if(l.length>1)throw new H(`Found more than one (${l.length}) save handlers for URL '${e}'`);e=l[0]}if(e.save==null)throw new H("LayersModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");let n=await ts.encodeWeights(this.getNamedWeights(t)),s=!1,r=null,o={modelTopology:this.toJSON(r,s),format:jB,generatedBy:`TensorFlow.js tfjs-layers v${nA}`,convertedBy:null};if((t==null?!1:t.includeOptimizer)&&this.optimizer!=null){o.trainingConfig=this.getTrainingConfig();let l="optimizer",{data:c,specs:u}=await ts.encodeWeights(await this.optimizer.getWeights(),l);n.specs.push(...u),n.data=ts.concatenateArrayBuffers([n.data,c])}if(this.userDefinedMetadata!=null){let l=!0;Ww(this.userDefinedMetadata,this.name,l),o.userDefinedMetadata=this.userDefinedMetadata}return o.weightData=n.data,o.weightSpecs=n.specs,e.save(o)}setUserDefinedMetadata(e){Ww(e,this.name),this.userDefinedMetadata=e}getUserDefinedMetadata(){return this.userDefinedMetadata}};ea.className="Model";ue.registerClass(ea);var ek=class extends ea{};ek.className="Functional";ue.registerClass(ek);async function qB(e,t){"modelTopology"in e||(e={modelTopology:e}),e=e;let n=e.modelTopology;n.model_config!=null&&(n=n.model_config);let s=Gd(n),r=fr(s,t);if(e.weightsManifest!=null){let a=await ts.loadWeights(e.weightsManifest,e.pathPrefix,r.weights.map(i=>i.originalName)),o={};for(let i of r.weights)o[i.originalName]=a[i.originalName];r.loadWeights(o),se(a)}return r}async function XB(e,t){if(t==null&&(t={}),typeof e=="string"){let n=ts.getLoadHandlers(e,t);if(n.length===0)n.push(ts.browserHTTPRequest(e,t));else if(n.length>1)throw new H(`Found more than one (${n.length}) load handlers for URL '${e}'`);e=n[0]}return KB(e,void 0,t)}async function KB(e,t,n){if(n==null&&(n={}),e.load==null)throw new H("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");let s=await e.load(),r=s.modelTopology;r.model_config!=null&&(r=r.model_config);let a=n.strict==null?!0:n.strict,o=s.weightData!=null&&s.weightSpecs!=null&&a,i=fr(Gd(r),t,o),l=s.trainingConfig;if(l!=null&&i.loadTrainingConfig(l),s.userDefinedMetadata!=null&&i.setUserDefinedMetadata(s.userDefinedMetadata),s.weightData!=null){if(s.weightSpecs==null)throw new H("LayersModel artifacts contains weight data, but not weight specs. Therefore loading of weights cannot proceed.");let{modelWeights:c,optimizerWeights:u}=ZB(s.weightData,s.weightSpecs);i.loadWeights(c,a),i.optimizer!=null&&u.length>0&&await i.optimizer.setWeights(u),se(c),se(u.map(d=>d.tensor))}return i}function ZB(e,t){let n=ts.decodeWeights(e,t),s={},r=[];return t.forEach(a=>{a.group==="optimizer"?r.push({name:a.name,tensor:n[a.name]}):s[a.name]=n[a.name]}),{modelWeights:s,optimizerWeights:r}}var lA=class extends ea{constructor(e){super({inputs:[],outputs:[]});if(e=e||{},this.trainable=!0,this.built=!1,this.name=e.name!=null?e.name:jf("sequential_"),e.layers!=null)for(let t of e.layers)this.add(t)}checkShape(e){if(e.inboundNodes[0].outputTensors[0].shape.some(n=>n<0))throw new H(`Negative dimension size caused by adding layer ${e.name} with input shape [${e.inboundNodes[0].inputTensors[0].shape}]`)}add(e){let t=e instanceof lA||e instanceof ea,n;if(t){if(n=e,n.outputs.length!==1)throw new H("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");if(n.inputs.length!==1)throw new H("All layers in a Sequential model should have a single input tensor. For multi-input layers, use the functional API.")}if(this.outputs.length===0){if(e.inboundNodes.length===0){if(e.batchInputShape==null)throw new H("The first layer in a Sequential model must get an `inputShape` or `batchInputShape` argument.");let s=Tw({batchShape:e.batchInputShape,dtype:e.dtype,name:e.name+"_input"});e.apply(s)}if(t)this.outputs=n.outputs,this.inputs=n.inputs;else{if(e.inboundNodes.length!==1)throw new H(`A layer added to a Sequential model must not already be connected somewhere else. LayersModel received layer ${e.name} which has ${e.inboundNodes.length} pre-existing inbound connections.`);if(e.inboundNodes[0].outputTensors.length!==1)throw new H("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");this.checkShape(e),this.outputs=[e.inboundNodes[0].outputTensors[0]],this.inputs=Cw(this.outputs[0])}this.inboundNodes=[],new Kf({outboundLayer:this,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:this.inputs,outputTensors:this.outputs,inputMasks:dl(null,this.inputs.length),outputMasks:[null],inputShapes:this.inputs.map(s=>s.shape),outputShapes:this.outputs[0].shape})}else{let s=e.apply(this.outputs[0]);if(Array.isArray(s))throw new TypeError("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");this.checkShape(e),this.outputs=[s],this.inboundNodes[0].outputTensors=this.outputs,this.inboundNodes[0].outputShapes=[this.outputs[0].shape]}this.layers.push(e),this.built=!1}pop(){if(this.layers.length===0)throw new TypeError("There are no layers in the model.");if(this.layers.pop(),this.layers.length===0)this.outputs=[],this.inboundNodes=[],this.outboundNodes=[];else{let e=this.layers.length-1;this.layers[e].outboundNodes=[],this.outputs=[this.layers[e].output],this.inboundNodes[0].outputTensors=this.outputs,this.inboundNodes[0].outputShapes=[this.outputs[0].shape]}}call(e,t){return this.model==null&&this.build(),this.model.call(e,t)}build(e){if(pt(e),this.inputs.length===0||this.outputs.length===0)throw new TypeError("Sequential model cannot be built: model is empty. Add some layers first.");this.model=new ea({inputs:this.inputs,outputs:this.outputs[0],name:this.name+"_model"}),this.model.trainable=this.trainable,this.supportsMasking=this.model.supportsMasking,this.inputLayers=this.model.inputLayers,this.inputLayersNodeIndices=this.model.inputLayersNodeIndices,this.inputLayersTensorIndices=this.model.inputLayersTensorIndices,this.outputLayers=this.model.outputLayers,this.outputLayersNodeIndices=this.model.outputLayersNodeIndices,this.outputLayersTensorIndices=this.model.outputLayersTensorIndices,this.nodesByDepth=this.model.nodesByDepth,this.containerNodes=this.model.containerNodes,this.outputNames=this.model.outputNames,this.inputNames=this.model.inputNames,this.built=!0}countParams(){return this.built||this.build(),super.countParams()}summary(e,t,n=console.log){this.built||this.build(),super.summary(e,t,n)}setWeights(e){this.model==null&&this.build(),this.model.setWeights(e)}evaluate(e,t,n={}){if(!this.built)throw new cr("The model needs to be compiled before being used.");return this.model.evaluate(e,t,n)}async evaluateDataset(e,t){if(!this.built)throw new cr("The model needs to be compiled before being used.");return this.model.evaluateDataset(e,t)}predict(e,t={}){return this.model==null&&this.build(),this.model.predict(e,t)}predictOnBatch(e){return this.model==null&&this.build(),this.model.predictOnBatch(e)}compile(e){this.build(),this.model.compile(e),this.optimizer_=this.model.optimizer,this.isOptimizerOwned=this.model.isOptimizerOwned,this.loss=this.model.loss,this.metrics=this.model.metrics,this.metricsTensors=this.model.metricsTensors,this.metricsNames=this.model.metricsNames}get optimizer(){return this.model==null?void 0:this.model.optimizer}set optimizer(e){this.model.optimizer=e}async fit(e,t,n={}){if(!this.built)throw new cr("The model needs to be compiled before being used.");return this.model.fit(e,t,n)}async fitDataset(e,t){if(!this.built)throw new cr("The model needs to be compiled before being used.");return this.model.fitDataset(e,t)}async trainOnBatch(e,t){return this.model.trainOnBatch(e,t)}static fromConfig(e,t,n={},s=!1){let r,a={};if(t instanceof Array){if(t[0].className==null||t[0].className==="Merge")throw new H("Legacy serialization format not supported yet.");r=t}else v.assert(t.layers!=null,()=>"When the config data for a Sequential model is not an Array, it must be an Object that contains the 'layers' field."),r=t.layers,delete t.layers,a=t;let o=new e(a);if(!(o instanceof lA))throw new Le(`Sequential.fromConfig called on non-Sequential input: ${o}`);for(let i of r){let c=fr(i,void 0,s);s&&c.setFastWeightInitDuringBuild(!0),o.add(c)}return o}set stopTraining(e){if(this.model==null)throw new H("Cannot set the stopTraining property of a sequential model before it is compiled.");this.model.stopTraining=e}get stopTraining(){if(this.model==null)throw new H("Cannot get the stopTraining property of a sequential model before it is compiled.");return this.model.stopTraining}getConfig(){let e=[];for(let t of this.layers){let n={};n.className=t.getClassName(),n.config=t.getConfig(),e.push(n)}return{name:this.name,layers:e}}},rm=lA;rm.className="Sequential";ue.registerClass(rm);function YB(e){return new ea(e)}function JB(e){return new rm(e)}function QB(e,t){return t==null&&(t={}),XB(e,t)}function tk(e){return Tw(e)}function eW(e,t){X1.registerCallbackConstructor(e,t)}var os=class extends ue.Serializable{getConfig(){return{}}},nk=class extends os{apply(e,t=1){return CL(e,t)}};nk.className="elu";ue.registerClass(nk);var sk=class extends os{apply(e){return c1(e)}};sk.className="selu";ue.registerClass(sk);var rk=class extends os{apply(e){return Rr(e)}};rk.className="relu";ue.registerClass(rk);var ak=class extends os{apply(e){return q(()=>Ed(6,Rr(e)))}};ak.className="relu6";ue.registerClass(ak);var ok=class extends os{apply(e){return e}};ok.className="linear";ue.registerClass(ok);var ik=class extends os{apply(e){return cs(e)}};ik.className="sigmoid";ue.registerClass(ik);var lk=class extends os{apply(e){return NL(e)}};lk.className="hardSigmoid";ue.registerClass(lk);var uk=class extends os{apply(e){return Mu(e)}};uk.className="softplus";ue.registerClass(uk);var ck=class extends os{apply(e){return TL(e)}};ck.className="softsign";ue.registerClass(ck);var dk=class extends os{apply(e){return _u(e)}};dk.className="tanh";ue.registerClass(dk);var uA=class extends os{apply(e,t=-1){return Vu(e,t)}};uA.className="softmax";ue.registerClass(uA);var pk=class extends os{apply(e,t=-1){return Q2(e,t)}};pk.className="logSoftmax";ue.registerClass(pk);var hk=class extends os{apply(e,t=1){return q(()=>L(cs(L(e,t)),e))}};hk.className="swish";ue.registerClass(hk);var fk=class extends os{apply(e){return q(()=>L(e,_u(Mu(e))))}};fk.className="mish";ue.registerClass(fk);function Mo(e){return e.getClassName()}function cA(e,t={}){return Md(e,ue.SerializationMap.getMap().classNameMap,t,"activation")}function zo(e){if(e==null){let t={};return t.className="linear",t.config={},cA(t)}if(typeof e=="string"){let t={};return t.className=e,t.config={},cA(t)}else return e instanceof os?e:cA(e)}function dA(e){if(e!=null&&typeof e!="object")throw new Error(`Argument to L1L2 regularizer's constructor is expected to be an object, but received: ${e}`)}var mk=class extends ue.Serializable{},qd=class extends mk{constructor(e){super();dA(e),this.l1=e==null||e.l1==null?.01:e.l1,this.l2=e==null||e.l2==null?.01:e.l2,this.hasL1=this.l1!==0,this.hasL2=this.l2!==0}apply(e){return q(()=>{let t=Gt([1]);return this.hasL1&&(t=le(t,Se(L(this.l1,tn(e))))),this.hasL2&&(t=le(t,Se(L(this.l2,Wd(e))))),G(t,[])})}getConfig(){return{l1:this.l1,l2:this.l2}}static fromConfig(e,t){return new e({l1:t.l1,l2:t.l2})}};qd.className="L1L2";ue.registerClass(qd);function tW(e){return dA(e),new qd({l1:e!=null?e.l1:null,l2:0})}function nW(e){return dA(e),new qd({l2:e!=null?e.l2:null,l1:0})}var gk={l1l2:"L1L2"};function At(e){return C1(e)}function Ak(e,t={}){return Md(e,ue.SerializationMap.getMap().classNameMap,t,"regularizer")}function Et(e){if(e==null)return null;if(typeof e=="string"){let n={className:e in gk?gk[e]:e,config:{}};return Ak(n)}else return e instanceof mk?e:Ak(e)}var pA=class extends Qe{constructor(e){super(e==null?{}:e);this.supportsMasking=!0,e!=null&&(this.maxValue=e.maxValue)}call(e,t){e=Be(e);let n=Rr(e);return this.maxValue!=null&&(n=ds(n,0,this.maxValue)),n}computeOutputShape(e){return e}getConfig(){let e={maxValue:this.maxValue},t=super.getConfig();return Object.assign(e,t),e}};pA.className="ReLU";ue.registerClass(pA);var hA=class extends Qe{constructor(e){super(e==null?{}:e);this.DEFAULT_ALPHA=.3,e==null&&(e={}),this.alpha=e.alpha==null?this.DEFAULT_ALPHA:e.alpha}call(e,t){let n=Be(e);return df(n,this.alpha)}computeOutputShape(e){return e}getConfig(){let e={alpha:this.alpha},t=super.getConfig();return Object.assign(e,t),e}};hA.className="LeakyReLU";ue.registerClass(hA);var fA=class extends Qe{constructor(e){super(e==null?{}:e);if(this.DEFAULT_ALPHA_INITIALIZER="zeros",e==null&&(e={}),this.supportsMasking=!0,this.alphaInitializer=Nt(e.alphaInitializer||this.DEFAULT_ALPHA_INITIALIZER),this.alphaRegularizer=Et(e.alphaRegularizer),this.alphaConstraint=an(e.alphaConstraint),e.sharedAxes==null)this.sharedAxes=null;else if(Array.isArray(e.sharedAxes))this.sharedAxes=e.sharedAxes;else if(typeof e.sharedAxes=="number")this.sharedAxes=[e.sharedAxes];else throw new H(`Expected sharedAxes to be a number or an array of numbers, but got ${e.sharedAxes}`)}build(e){e=pt(e);let t=e.slice(1);if(this.sharedAxes!=null)for(let s of this.sharedAxes)t[s-1]=1;this.alpha=this.addWeight("alpha",t,"float32",this.alphaInitializer,this.alphaRegularizer,!0,this.alphaConstraint);let n={};if(this.sharedAxes!=null)for(let s=1;s<e.length;++s)n[s]=e[s];this.inputSpec=[new Kt({ndim:e.length,axes:n})],this.built=!0}call(e,t){return e=Be(e),yf(e,this.alpha.read())}getConfig(){let e={alphaInitializer:Ot(this.alphaInitializer),alphaRegularizer:At(this.alphaRegularizer),alphaConstraint:rn(this.alphaConstraint),sharedAxes:this.sharedAxes},t=super.getConfig();return Object.assign(e,t),e}};fA.className="PReLU";ue.registerClass(fA);var mA=class extends Qe{constructor(e){super(e==null?{}:e);if(this.DEFAULT_ALPHA=1,e==null&&(e={}),e.alpha!=null&&e.alpha!==this.DEFAULT_ALPHA)throw new Le(`Non-default alpha value (${e.alpha}) is not supported by the ELU layer yet.`);this.alpha=e.alpha==null?this.DEFAULT_ALPHA:e.alpha}call(e,t){let n=Be(e);return Td(n)}computeOutputShape(e){return e}getConfig(){let e={alpha:this.alpha},t=super.getConfig();return Object.assign(e,t),e}};mA.className="ELU";ue.registerClass(mA);var gA=class extends Qe{constructor(e){super(e==null?{}:e);this.DEFAULT_THETA=1,e==null&&(e={}),this.theta=e.theta==null?this.DEFAULT_THETA:e.theta}call(e,t){let n=Be(e);return L(n,de(ps(n,this.theta),"float32"))}computeOutputShape(e){return e}getConfig(){let e={theta:this.theta},t=super.getConfig();return Object.assign(e,t),e}};gA.className="ThresholdedReLU";ue.registerClass(gA);var AA=class extends Qe{constructor(e){super(e==null?{}:e);this.DEFAULT_AXIS=1,e==null&&(e={}),this.softmax=new uA().apply,this.axis=e.axis==null?this.DEFAULT_AXIS:e.axis}call(e,t){let n=Be(e);return this.softmax(n,this.axis)}computeOutputShape(e){return e}getConfig(){let e={axis:this.axis},t=super.getConfig();return Object.assign(e,t),e}};AA.className="Softmax";ue.registerClass(AA);function Ku(e,t,n){if(typeof e=="number")return dl(e,t);if(e.length!==t)throw new H(`The ${n} argument must be an integer or tuple of ${t} integers. Received: ${e.length} elements.`);for(let s=0;s<t;++s){let r=e[s];if(!wL(r))throw new H(`The ${n} argument must be an integer or tuple of ${t} integers. Received: ${JSON.stringify(e)} including a non-integer number ${r}`)}return e}function mr(e,t,n,s,r=1){if(e==null)return e;let a=t+(t-1)*(r-1),o;return n==="same"?o=e:o=e-a+1,Math.floor((o+s-1)/s)}function Fr(e,t,n,s){if(e==null)return null;if(s==="valid")e=e*t+Fo([n-t,0]);else if(s==="same")e=e*t;else throw new H(`Unsupport padding mode: ${s}.`);return e}function yA(e,t){return q(()=>(Ht(t),t==="channelsFirst"?Ye(e,[0,2,3,1]):e))}function yk(e,t){return q(()=>(Ht(t),t==="channelsFirst"?Ye(e,[0,2,3,4,1]):e))}function sW(e,t,n,s=1,r="valid",a,o=1){return q(()=>{if(a==null&&(a=ur()),Ht(a),e.shape.length!==3)throw new H(`The input of a conv1dWithBias operation should be 3, but is ${e.shape.length} instead.`);if(t.shape.length!==3)throw new H(`The kernel for a conv1dWithBias operation should be 3, but is ${t.shape.length} instead`);if(n!=null&&n.shape.length!==1)throw new H(`The bias for a conv1dWithBias operation should be 1, but is ${t.shape.length} instead`);if(a==="channelsFirst"&&(e=Ye(e,[0,2,1])),r==="causal")throw new Le("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");let i=H2(e,t,s,r==="same"?"same":"valid","NWC",o);return n!=null&&(i=pr(i,n)),i})}function xk(e,t,n,s=[1,1],r="valid",a,o,i=null){return q(()=>{if(a==null&&(a=ur()),Ht(a),e.rank!==3&&e.rank!==4)throw new H(`conv2dWithBiasActivation expects input to be of rank 3 or 4, but received ${e.rank}.`);if(t.rank!==3&&t.rank!==4)throw new H(`conv2dWithBiasActivation expects kernel to be of rank 3 or 4, but received ${e.rank}.`);let l=yA(e,a);if(r==="causal")throw new Le("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");return l=$o.conv2d({x:l,filter:t,strides:s,pad:r==="same"?"same":"valid",dilations:o,dataFormat:"NHWC",bias:n,activation:i}),a==="channelsFirst"&&(l=Ye(l,[0,3,1,2])),l})}function rW(e,t,n,s=[1,1,1],r="valid",a,o){return q(()=>{if(a==null&&(a=ur()),Ht(a),e.rank!==4&&e.rank!==5)throw new H(`conv3dWithBias expects input to be of rank 4 or 5, but received ${e.rank}.`);if(t.rank!==4&&t.rank!==5)throw new H(`conv3dWithBias expects kernel to be of rank 4 or 5, but received ${e.rank}.`);let i=yk(e,a);if(r==="causal")throw new Le("The support for CAUSAL padding mode in conv3dWithBias is not implemented yet.");return i=X2(i,t,s,r==="same"?"same":"valid","NDHWC",o),n!=null&&(i=pr(i,n)),a==="channelsFirst"&&(i=Ye(i,[0,4,1,2,3])),i})}var xA=class extends Qe{constructor(e,t){super(t);if(this.bias=null,this.DEFAULT_KERNEL_INITIALIZER="glorotNormal",this.DEFAULT_BIAS_INITIALIZER="zeros",xA.verifyArgs(t),this.rank=e,yn(this.rank,"rank"),this.rank!==1&&this.rank!==2&&this.rank!==3)throw new Le(`Convolution layer for rank other than 1, 2, or 3 (${this.rank}) is not implemented yet.`);if(this.kernelSize=Ku(t.kernelSize,e,"kernelSize"),this.strides=Ku(t.strides==null?1:t.strides,e,"strides"),this.padding=t.padding==null?"valid":t.padding,$s(this.padding),this.dataFormat=t.dataFormat==null?"channelsLast":t.dataFormat,Ht(this.dataFormat),this.activation=zo(t.activation),this.useBias=t.useBias==null?!0:t.useBias,this.biasInitializer=Nt(t.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.biasConstraint=an(t.biasConstraint),this.biasRegularizer=Et(t.biasRegularizer),this.activityRegularizer=Et(t.activityRegularizer),this.dilationRate=Ku(t.dilationRate==null?1:t.dilationRate,e,"dilationRate"),this.rank===1&&Array.isArray(this.dilationRate)&&this.dilationRate.length!==1)throw new H(`dilationRate must be a number or an array of a single number for 1D convolution, but received ${JSON.stringify(this.dilationRate)}`);if(this.rank===2){if(typeof this.dilationRate=="number")this.dilationRate=[this.dilationRate,this.dilationRate];else if(this.dilationRate.length!==2)throw new H(`dilationRate must be a number or array of two numbers for 2D convolution, but received ${JSON.stringify(this.dilationRate)}`)}else if(this.rank===3){if(typeof this.dilationRate=="number")this.dilationRate=[this.dilationRate,this.dilationRate,this.dilationRate];else if(this.dilationRate.length!==3)throw new H(`dilationRate must be a number or array of three numbers for 3D convolution, but received ${JSON.stringify(this.dilationRate)}`)}}static verifyArgs(e){if($r("kernelSize"in e,"required key 'kernelSize' not in config"),typeof e.kernelSize!="number"&&!N1(e.kernelSize,"number",1,3))throw new H(`BaseConv expects config.kernelSize to be number or number[] with length 1, 2, or 3, but received ${JSON.stringify(e.kernelSize)}.`)}getConfig(){let e={kernelSize:this.kernelSize,strides:this.strides,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,activation:Mo(this.activation),useBias:this.useBias,biasInitializer:Ot(this.biasInitializer),biasRegularizer:At(this.biasRegularizer),activityRegularizer:At(this.activityRegularizer),biasConstraint:rn(this.biasConstraint)},t=super.getConfig();return Object.assign(e,t),e}},Xd=class extends xA{constructor(e,t){super(e,t);this.kernel=null,Xd.verifyArgs(t),this.filters=t.filters,yn(this.filters,"filters"),this.kernelInitializer=Nt(t.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.kernelConstraint=an(t.kernelConstraint),this.kernelRegularizer=Et(t.kernelRegularizer)}build(e){e=pt(e);let t=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[t]==null)throw new H(`The channel dimension of the input should be defined. Found ${e[t]}`);let n=e[t],s=this.kernelSize.concat([n,this.filters]);this.kernel=this.addWeight("kernel",s,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[{ndim:this.rank+2,axes:{[t]:n}}],this.built=!0}call(e,t){return q(()=>{e=Be(e);let n,s=this.bias==null?null:this.bias.read(),r=lw(this.activation.getClassName());if(r!=null&&this.rank===2)n=xk(e,this.kernel.read(),s,this.strides,this.padding,this.dataFormat,this.dilationRate,r);else{if(this.rank===1)n=sW(e,this.kernel.read(),s,this.strides[0],this.padding,this.dataFormat,this.dilationRate[0]);else if(this.rank===2)n=xk(e,this.kernel.read(),s,this.strides,this.padding,this.dataFormat,this.dilationRate);else if(this.rank===3)n=rW(e,this.kernel.read(),s,this.strides,this.padding,this.dataFormat,this.dilationRate);else throw new Le("convolutions greater than 3D are not implemented yet.");this.activation!=null&&(n=this.activation.apply(n))}return n})}computeOutputShape(e){e=pt(e);let t=[],n=this.dataFormat==="channelsLast"?e.slice(1,e.length-1):e.slice(2);for(let r=0;r<n.length;++r){let a=mr(n[r],this.kernelSize[r],this.padding,this.strides[r],typeof this.dilationRate=="number"?this.dilationRate:this.dilationRate[r]);t.push(a)}let s=[e[0]];return this.dataFormat==="channelsLast"?(s=s.concat(t),s.push(this.filters)):(s.push(this.filters),s=s.concat(t)),s}getConfig(){let e={filters:this.filters,kernelInitializer:Ot(this.kernelInitializer),kernelRegularizer:At(this.kernelRegularizer),kernelConstraint:rn(this.kernelConstraint)},t=super.getConfig();return Object.assign(e,t),e}static verifyArgs(e){if(!("filters"in e)||typeof e.filters!="number"||e.filters<1)throw new H(`Convolution layer expected config.filters to be a 'number' > 0 but got ${JSON.stringify(e.filters)}`)}},bk=class extends Xd{constructor(e){super(2,e);bk.verifyArgs(e)}getConfig(){let e=super.getConfig();return delete e.rank,e}static verifyArgs(e){if(typeof e.kernelSize!="number"&&!N1(e.kernelSize,"number",1,2))throw new H(`Conv2D expects config.kernelSize to be number or number[] with length 1 or 2, but received ${JSON.stringify(e.kernelSize)}.`)}},am=bk;am.className="Conv2D";ue.registerClass(am);var vk=class extends Xd{constructor(e){super(3,e);vk.verifyArgs(e)}getConfig(){let e=super.getConfig();return delete e.rank,e}static verifyArgs(e){if(typeof e.kernelSize!="number"&&!(Array.isArray(e.kernelSize)&&(e.kernelSize.length===1||e.kernelSize.length===3)))throw new H(`Conv3D expects config.kernelSize to be number or [number, number, number], but received ${JSON.stringify(e.kernelSize)}.`)}},om=vk;om.className="Conv3D";ue.registerClass(om);var bA=class extends am{constructor(e){super(e);if(this.inputSpec=[new Kt({ndim:4})],this.padding!=="same"&&this.padding!=="valid")throw new H(`Conv2DTranspose currently supports only padding modes 'same' and 'valid', but received padding mode ${this.padding}`)}build(e){if(e=pt(e),e.length!==4)throw new H("Input should have rank 4; Received input shape: "+JSON.stringify(e));let t=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[t]==null)throw new H("The channel dimension of the inputs should be defined. Found `None`.");let n=e[t],s=this.kernelSize.concat([this.filters,n]);this.kernel=this.addWeight("kernel",s,"float32",this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[new Kt({ndim:4,axes:{[t]:n}})],this.built=!0}call(e,t){return q(()=>{let n=Be(e);if(n.shape.length!==4)throw new H(`Conv2DTranspose.call() expects input tensor to be rank-4, but received a tensor of rank-${n.shape.length}`);let s=n.shape,r=s[0],a,o;this.dataFormat==="channelsFirst"?(a=2,o=3):(a=1,o=2);let i=s[a],l=s[o],c=this.kernelSize[0],u=this.kernelSize[1],d=this.strides[0],p=this.strides[1],h=Fr(i,d,c,this.padding),f=Fr(l,p,u,this.padding),m=[r,h,f,this.filters];this.dataFormat!=="channelsLast"&&(n=Ye(n,[0,2,3,1]));let g=q2(n,this.kernel.read(),m,this.strides,this.padding);return this.dataFormat!=="channelsLast"&&(g=Ye(g,[0,3,1,2])),this.bias!=null&&(g=pr(g,this.bias.read(),this.dataFormat)),this.activation!=null&&(g=this.activation.apply(g)),g})}computeOutputShape(e){e=pt(e);let t=e.slice(),n,s,r;this.dataFormat==="channelsFirst"?(n=1,s=2,r=3):(n=3,s=1,r=2);let a=this.kernelSize[0],o=this.kernelSize[1],i=this.strides[0],l=this.strides[1];return t[n]=this.filters,t[s]=Fr(t[s],i,a,this.padding),t[r]=Fr(t[r],l,o,this.padding),t}getConfig(){let e=super.getConfig();return delete e.dilationRate,e}};bA.className="Conv2DTranspose";ue.registerClass(bA);var vA=class extends om{constructor(e){super(e);if(this.inputSpec=[new Kt({ndim:5})],this.padding!=="same"&&this.padding!=="valid")throw new H(`Conv3DTranspose currently supports only padding modes 'same' and 'valid', but received padding mode ${this.padding}`)}build(e){if(e=pt(e),e.length!==5)throw new H("Input should have rank 5; Received input shape: "+JSON.stringify(e));let t=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[t]==null)throw new H("The channel dimension of the inputs should be defined. Found `None`.");let n=e[t],s=this.kernelSize.concat([this.filters,n]);this.kernel=this.addWeight("kernel",s,"float32",this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[new Kt({ndim:5,axes:{[t]:n}})],this.built=!0}call(e,t){return q(()=>{let n=Be(e);if(n.shape.length!==5)throw new H(`Conv3DTranspose.call() expects input tensor to be rank-4, but received a tensor of rank-${n.shape.length}`);let s=n.shape,r=s[0],a,o,i;this.dataFormat==="channelsFirst"?(i=2,a=3,o=4):(i=1,a=2,o=3);let l=s[i],c=s[a],u=s[o],d=this.kernelSize[0],p=this.kernelSize[1],h=this.kernelSize[2],f=this.strides[0],m=this.strides[1],g=this.strides[2],A=Fr(l,f,d,this.padding),x=Fr(c,m,p,this.padding),y=Fr(u,g,h,this.padding),b=[r,A,x,y,this.filters];this.dataFormat!=="channelsLast"&&(n=Ye(n,[0,2,3,4,1]));let w=Q3(n,this.kernel.read(),b,this.strides,this.padding);return this.dataFormat!=="channelsLast"&&(w=Ye(w,[0,4,1,2,3])),this.bias!==null&&(w=pr(w,this.bias.read(),this.dataFormat)),this.activation!==null&&(w=this.activation.apply(w)),w})}computeOutputShape(e){e=pt(e);let t=e.slice(),n,s,r,a;this.dataFormat==="channelsFirst"?(n=1,s=2,r=3,a=4):(n=4,s=1,r=2,a=3);let o=this.kernelSize[0],i=this.kernelSize[1],l=this.kernelSize[2],c=this.strides[0],u=this.strides[1],d=this.strides[2];return t[n]=this.filters,t[s]=Fr(t[s],c,o,this.padding),t[r]=Fr(t[r],u,i,this.padding),t[a]=Fr(t[a],d,l,this.padding),t}getConfig(){let e=super.getConfig();return delete e.dilationRate,e}};vA.className="Conv3DTranspose";ue.registerClass(vA);var wk=class extends Xd{constructor(e,t){super(e,t);if(this.DEFAULT_DEPTHWISE_INITIALIZER="glorotUniform",this.DEFAULT_POINTWISE_INITIALIZER="glorotUniform",this.depthwiseKernel=null,this.pointwiseKernel=null,t.filters==null)throw new H("The `filters` configuration field is required by SeparableConv, but is unspecified.");if(t.kernelInitializer!=null||t.kernelRegularizer!=null||t.kernelConstraint!=null)throw new H("Fields kernelInitializer, kernelRegularizer and kernelConstraint are invalid for SeparableConv2D. Use depthwiseInitializer, depthwiseRegularizer, depthwiseConstraint, pointwiseInitializer, pointwiseRegularizer and pointwiseConstraint instead.");if(t.padding!=null&&t.padding!=="same"&&t.padding!=="valid")throw new H(`SeparableConv${this.rank}D supports only padding modes: 'same' and 'valid', but received ${JSON.stringify(t.padding)}`);this.depthMultiplier=t.depthMultiplier==null?1:t.depthMultiplier,this.depthwiseInitializer=Nt(t.depthwiseInitializer||this.DEFAULT_DEPTHWISE_INITIALIZER),this.depthwiseRegularizer=Et(t.depthwiseRegularizer),this.depthwiseConstraint=an(t.depthwiseConstraint),this.pointwiseInitializer=Nt(t.depthwiseInitializer||this.DEFAULT_POINTWISE_INITIALIZER),this.pointwiseRegularizer=Et(t.pointwiseRegularizer),this.pointwiseConstraint=an(t.pointwiseConstraint)}build(e){if(e=pt(e),e.length<this.rank+2)throw new H(`Inputs to SeparableConv${this.rank}D should have rank ${this.rank+2}, but received input shape: ${JSON.stringify(e)}`);let t=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[t]==null||e[t]<0)throw new H(`The channel dimension of the inputs should be defined, but found ${JSON.stringify(e[t])}`);let n=e[t],s=this.kernelSize.concat([n,this.depthMultiplier]),r=[];for(let o=0;o<this.rank;++o)r.push(1);r.push(n*this.depthMultiplier,this.filters);let a=!0;this.depthwiseKernel=this.addWeight("depthwise_kernel",s,"float32",this.depthwiseInitializer,this.depthwiseRegularizer,a,this.depthwiseConstraint),this.pointwiseKernel=this.addWeight("pointwise_kernel",r,"float32",this.pointwiseInitializer,this.pointwiseRegularizer,a,this.pointwiseConstraint),this.useBias?this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,a,this.biasConstraint):this.bias=null,this.inputSpec=[new Kt({ndim:this.rank+2,axes:{[t]:n}})],this.built=!0}call(e,t){return q(()=>{e=Be(e);let n;if(this.rank===1)throw new Le("1D separable convolution is not implemented yet.");return this.rank===2&&(this.dataFormat==="channelsFirst"&&(e=Ye(e,[0,2,3,1])),n=bv(e,this.depthwiseKernel.read(),this.pointwiseKernel.read(),this.strides,this.padding,this.dilationRate,"NHWC")),this.useBias&&(n=pr(n,this.bias.read(),this.dataFormat)),this.activation!=null&&(n=this.activation.apply(n)),this.dataFormat==="channelsFirst"&&(n=Ye(n,[0,3,1,2])),n})}getConfig(){let e=super.getConfig();return delete e.rank,delete e.kernelInitializer,delete e.kernelRegularizer,delete e.kernelConstraint,e.depthwiseInitializer=Ot(this.depthwiseInitializer),e.pointwiseInitializer=Ot(this.pointwiseInitializer),e.depthwiseRegularizer=At(this.depthwiseRegularizer),e.pointwiseRegularizer=At(this.pointwiseRegularizer),e.depthwiseConstraint=rn(this.depthwiseConstraint),e.pointwiseConstraint=rn(this.pointwiseConstraint),e}};wk.className="SeparableConv";var wA=class extends wk{constructor(e){super(2,e)}};wA.className="SeparableConv2D";ue.registerClass(wA);var kk=class extends Xd{constructor(e){super(1,e);kk.verifyArgs(e),this.inputSpec=[{ndim:3}]}getConfig(){let e=super.getConfig();return delete e.rank,delete e.dataFormat,e}static verifyArgs(e){if(typeof e.kernelSize!="number"&&!N1(e.kernelSize,"number",1,1))throw new H(`Conv1D expects config.kernelSize to be number or number[] with length 1, but received ${JSON.stringify(e.kernelSize)}.`)}},kA=kk;kA.className="Conv1D";ue.registerClass(kA);var SA=class extends Qe{constructor(e){super(e);typeof e.cropping=="number"?this.cropping=[[e.cropping,e.cropping],[e.cropping,e.cropping]]:typeof e.cropping[0]=="number"?this.cropping=[[e.cropping[0],e.cropping[0]],[e.cropping[1],e.cropping[1]]]:this.cropping=e.cropping,this.dataFormat=e.dataFormat===void 0?"channelsLast":e.dataFormat,this.inputSpec=[{ndim:4}]}computeOutputShape(e){return this.dataFormat==="channelsFirst"?[e[0],e[1],e[2]-this.cropping[0][0]-this.cropping[0][1],e[3]-this.cropping[1][0]-this.cropping[1][1]]:[e[0],e[1]-this.cropping[0][0]-this.cropping[0][1],e[2]-this.cropping[1][0]-this.cropping[1][1],e[3]]}call(e,t){return q(()=>{if(e=Be(e),this.dataFormat==="channelsLast"){let n=Of(e,this.cropping[0][0],e.shape[1]-this.cropping[0][0]-this.cropping[0][1],2);return Of(n,this.cropping[1][0],e.shape[2]-this.cropping[1][1]-this.cropping[1][0],3)}else{let n=Of(e,this.cropping[0][0],e.shape[2]-this.cropping[0][0]-this.cropping[0][1],3);return Of(n,this.cropping[1][0],e.shape[3]-this.cropping[1][1]-this.cropping[1][0],4)}})}getConfig(){let e={cropping:this.cropping,dataFormat:this.dataFormat},t=super.getConfig();return Object.assign(e,t),e}};SA.className="Cropping2D";ue.registerClass(SA);var IA=class extends Qe{constructor(e){super(e);this.DEFAULT_SIZE=[2,2],this.inputSpec=[{ndim:4}],this.size=e.size==null?this.DEFAULT_SIZE:e.size,this.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,Ht(this.dataFormat),this.interpolation=e.interpolation==null?"nearest":e.interpolation,xL(this.interpolation)}computeOutputShape(e){if(this.dataFormat==="channelsFirst"){let t=e[2]==null?null:this.size[0]*e[2],n=e[3]==null?null:this.size[1]*e[3];return[e[0],e[1],t,n]}else{let t=e[1]==null?null:this.size[0]*e[1],n=e[2]==null?null:this.size[1]*e[2];return[e[0],t,n,e[3]]}}call(e,t){return q(()=>{let n=Be(e),s=n.shape;if(this.dataFormat==="channelsFirst"){n=Ye(n,[0,2,3,1]);let r=this.size[0]*s[2],a=this.size[1]*s[3],o=this.interpolation==="nearest"?$e.resizeNearestNeighbor(n,[r,a]):$e.resizeBilinear(n,[r,a]);return Ye(o,[0,3,1,2])}else{let r=this.size[0]*s[1],a=this.size[1]*s[2];return this.interpolation==="nearest"?$e.resizeNearestNeighbor(n,[r,a]):$e.resizeBilinear(n,[r,a])}})}getConfig(){let e={size:this.size,dataFormat:this.dataFormat},t=super.getConfig();return Object.assign(e,t),e}};IA.className="UpSampling2D";ue.registerClass(IA);function aW(e,t,n=[1,1],s="valid",r,a){return q(()=>{r==null&&(r=ur()),Ht(r);let o=yA(e,r);if(e.rank!==4)throw new H(`Input for depthwiseConv2d is required to be 4-D, but is instead ${e.rank}-D`);if(t.rank!==4)throw new H(`depthwiseKernel is required to be 4-D, but is instead ${t.rank}-D`);return o=Cd(o,t,n,s==="same"?"same":"valid","NHWC",a),r==="channelsFirst"&&(o=Ye(o,[0,3,1,2])),o})}var CA=class extends xA{constructor(e){super(2,e);this.depthwiseKernel=null,this.depthMultiplier=e.depthMultiplier==null?1:e.depthMultiplier,this.depthwiseInitializer=Nt(e.depthwiseInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.depthwiseConstraint=an(e.depthwiseConstraint),this.depthwiseRegularizer=Et(e.depthwiseRegularizer)}build(e){if(e=pt(e),e.length<4)throw new H(`Inputs to DepthwiseConv2D should have rank 4. Received input shape: ${JSON.stringify(e)}.`);let t=this.dataFormat==="channelsFirst"?1:3;if(e[t]==null||e[t]<0)throw new H(`The channel dimension of the inputs to DepthwiseConv2D should be defined, but is not (${e[t]}).`);let n=e[t],s=[this.kernelSize[0],this.kernelSize[1],n,this.depthMultiplier];this.depthwiseKernel=this.addWeight("depthwise_kernel",s,null,this.depthwiseInitializer,this.depthwiseRegularizer,!0,this.depthwiseConstraint),this.useBias?this.bias=this.addWeight("bias",[n*this.depthMultiplier],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0}call(e,t){return q(()=>{e=Be(e);let n=aW(e,this.depthwiseKernel.read(),this.strides,this.padding,this.dataFormat,null);return this.useBias&&(n=pr(n,this.bias.read(),this.dataFormat)),this.activation!=null&&(n=this.activation.apply(n)),n})}computeOutputShape(e){e=pt(e);let t=this.dataFormat==="channelsFirst"?e[2]:e[1],n=this.dataFormat==="channelsFirst"?e[3]:e[2],s=this.dataFormat==="channelsFirst"?e[1]*this.depthMultiplier:e[3]*this.depthMultiplier,r=mr(t,this.kernelSize[0],this.padding,this.strides[0]),a=mr(n,this.kernelSize[1],this.padding,this.strides[1]);return this.dataFormat==="channelsFirst"?[e[0],s,r,a]:[e[0],r,a,s]}getConfig(){let e=super.getConfig();return e.depthMultiplier=this.depthMultiplier,e.depthwiseInitializer=Ot(this.depthwiseInitializer),e.depthwiseRegularizer=At(this.depthwiseRegularizer),e.depthwiseConstraint=rn(this.depthwiseRegularizer),e}};CA.className="DepthwiseConv2D";ue.registerClass(CA);function Sk(e,t,n,s){if(Array.isArray(e)){if(t!=null||n!=null)throw new H("When inputs is an array, neither initialState or constants should be provided");s!=null&&(n=e.slice(e.length-s,e.length),e=e.slice(0,e.length-s)),e.length>1&&(t=e.slice(1,e.length)),e=e[0]}function r(a){return a==null||Array.isArray(a)?a:[a]}return t=r(t),n=r(n),{inputs:e,initialState:t,constants:n}}function Ik(e,t,n,s=!1,r,a,o=!1,i=!1){return q(()=>{let l=t.shape.length;if(l<3)throw new H(`Input should be at least 3D, but is ${l}D.`);let c=[1,0].concat(dr(2,l));if(t=Ye(t,c),a!=null)throw new Le("The rnn() functoin of the deeplearn.js backend does not support constants yet.");o&&console.warn("Backend rnn(): the unroll = true option is not applicable to the imperative deeplearn.js backend."),r!=null&&(r=de(de(r,"bool"),"float32"),r.rank===l-1&&(r=mn(r,-1)),r=Ye(r,c)),s&&(t=Rs(t,0),r!=null&&(r=Rs(r,0)));let u=[],d,p=n,h=t.shape[0],f=ss(t),m;r!=null&&(m=ss(r));for(let A=0;A<h;++A){let x=f[A],y=q(()=>e(x,p));if(r==null)d=y[0],p=y[1];else{let b=q(()=>{let w=m[A],k=me(Es(w),w),C=le(L(y[0],w),L(p[0],k)),N=p.map((R,F)=>le(L(y[1][F],w),L(R,k)));return{output:C,newStates:N}});d=b.output,p=b.newStates}i&&u.push(d)}let g;return i&&(g=An(u,1)),[d,g,p]})}var Ck=class extends Qe{constructor(e){super(e);let t;if(e.cell==null)throw new H("cell property is missing for the constructor of RNN.");if(Array.isArray(e.cell)?t=new um({cells:e.cell}):t=e.cell,t.stateSize==null)throw new H("The RNN cell should have an attribute `stateSize` (tuple of integers, one integer per RNN state).");this.cell=t,this.returnSequences=e.returnSequences==null?!1:e.returnSequences,this.returnState=e.returnState==null?!1:e.returnState,this.goBackwards=e.goBackwards==null?!1:e.goBackwards,this._stateful=e.stateful==null?!1:e.stateful,this.unroll=e.unroll==null?!1:e.unroll,this.supportsMasking=!0,this.inputSpec=[new Kt({ndim:3})],this.stateSpec=null,this.states_=null,this.numConstants=null,this.keptStates=[]}getStates(){if(this.states_==null){let e=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1;return dr(0,e).map(t=>null)}else return this.states_}setStates(e){this.states_=e}computeOutputShape(e){H1(e)&&(e=e[0]),e=e;let t=this.cell.stateSize;Array.isArray(t)||(t=[t]);let n=t[0],s;if(this.returnSequences?s=[e[0],e[1],n]:s=[e[0],n],this.returnState){let r=[];for(let a of t)r.push([e[0],a]);return[s].concat(r)}else return s}computeMask(e,t){return q(()=>{Array.isArray(t)&&(t=t[0]);let n=this.returnSequences?t:null;if(this.returnState){let s=this.states.map(r=>null);return[n].concat(s)}else return n})}get states(){if(this.states_==null){let e=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1,t=[];for(let n=0;n<e;++n)t.push(null);return t}else return this.states_}set states(e){this.states_=e}build(e){let t=null;if(this.numConstants!=null)throw new Le("Constants support is not implemented in RNN yet.");H1(e)&&(e=e[0]),e=e;let n=this.stateful?e[0]:null,s=e.slice(2);this.inputSpec[0]=new Kt({shape:[n,null,...s]});let r=[e[0]].concat(e.slice(2));if(t!=null)throw new Le("Constants support is not implemented in RNN yet.");this.cell.build(r);let a;if(Array.isArray(this.cell.stateSize)?a=this.cell.stateSize:a=[this.cell.stateSize],this.stateSpec!=null){if(!v.arraysEqual(this.stateSpec.map(o=>o.shape[o.shape.length-1]),a))throw new H(`An initialState was passed that is not compatible with cell.stateSize. Received stateSpec=${this.stateSpec}; However cell.stateSize is ${this.cell.stateSize}`)}else this.stateSpec=a.map(o=>new Kt({shape:[null,o]}));this.stateful&&this.resetStates()}resetStates(e,t=!1){q(()=>{if(!this.stateful)throw new Jr("Cannot call resetStates() on an RNN Layer that is not stateful.");let n=this.inputSpec[0].shape[0];if(n==null)throw new H("If an RNN is stateful, it needs to know its batch size. Specify the batch size of your input tensors: \n- If using a Sequential model, specify the batch size by passing a `batchInputShape` option to your first layer.\n- If using the functional API, specify the batch size by passing a `batchShape` option to your Input layer.");if(this.states_==null)Array.isArray(this.cell.stateSize)?this.states_=this.cell.stateSize.map(s=>Gt([n,s])):this.states_=[Gt([n,this.cell.stateSize])];else if(e==null)se(this.states_),this.keptStates!=null&&(se(this.keptStates),this.keptStates=[]),Array.isArray(this.cell.stateSize)?this.states_=this.cell.stateSize.map(s=>Gt([n,s])):this.states_[0]=Gt([n,this.cell.stateSize]);else{if(Array.isArray(e)||(e=[e]),e.length!==this.states_.length)throw new H(`Layer ${this.name} expects ${this.states_.length} state(s), but it received ${e.length} state value(s). Input received: ${e}`);t===!0?this.keptStates.push(this.states_.slice()):se(this.states_);for(let s=0;s<this.states_.length;++s){let r=e[s],a=Array.isArray(this.cell.stateSize)?this.cell.stateSize[s]:this.cell.stateSize,o=[n,a];if(!v.arraysEqual(r.shape,o))throw new H(`State ${s} is incompatible with layer ${this.name}: expected shape=${o}, received shape=${r.shape}`);this.states_[s]=r}}this.states_=this.states_.map(s=>fn(s.clone()))})}apply(e,t){let n=t==null?null:t.initialState,s=t==null?null:t.constants;t==null&&(t={});let r=Sk(e,n,s,this.numConstants);e=r.inputs,n=r.initialState,s=r.constants;let a=[],o=[];if(n!=null){t.initialState=n,a=a.concat(n),this.stateSpec=[];for(let l of n)this.stateSpec.push(new Kt({shape:l.shape}));o=o.concat(this.stateSpec)}if(s!=null&&(t.constants=s,a=a.concat(s),this.numConstants=s.length),a[0]instanceof hr){let l=[e].concat(a),c=this.inputSpec.concat(o),u=this.inputSpec;this.inputSpec=c;let d=super.apply(l,t);return this.inputSpec=u,d}else return super.apply(e,t)}call(e,t){return q(()=>{let n=t==null?null:t.mask,s=t==null?null:t.training,r=t==null?null:t.initialState;e=Be(e),r==null&&(this.stateful?r=this.states_:r=this.getInitialState(e));let a=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1;if(r.length!==a)throw new H(`RNN Layer has ${a} state(s) but was passed ${r.length} initial state(s).`);this.unroll&&console.warn("Ignoring unroll = true for RNN layer, due to imperative backend.");let o={training:s},l=Ik((h,f)=>{let m=this.cell.call([h].concat(f),o);return[m[0],m.slice(1)]},e,r,this.goBackwards,n,null,this.unroll,this.returnSequences),c=l[0],u=l[1],d=l[2];this.stateful&&this.resetStates(d,s);let p=this.returnSequences?u:c;return this.returnState?[p].concat(d):p})}getInitialState(e){return q(()=>{let t=Gt(e.shape);return t=Se(t,[1,2]),t=Bd(t),Array.isArray(this.cell.stateSize)?this.cell.stateSize.map(n=>n>1?O1(t,[1,n]):t):this.cell.stateSize>1?[O1(t,[1,this.cell.stateSize])]:[t]})}get trainableWeights(){return this.trainable?this.cell.trainableWeights:[]}get nonTrainableWeights(){return this.trainable?this.cell.nonTrainableWeights:this.cell.weights}setFastWeightInitDuringBuild(e){super.setFastWeightInitDuringBuild(e),this.cell!=null&&this.cell.setFastWeightInitDuringBuild(e)}getConfig(){let e=super.getConfig(),t={returnSequences:this.returnSequences,returnState:this.returnState,goBackwards:this.goBackwards,stateful:this.stateful,unroll:this.unroll};this.numConstants!=null&&(t.numConstants=this.numConstants);let n=this.cell.getConfig();return this.getClassName()===Ck.className&&(t.cell={className:this.cell.getClassName(),config:n}),{...n,...e,...t}}static fromConfig(e,t,n={}){let s=t.cell,r=fr(s,n);return new e(Object.assign(t,{cell:r}))}},ta=Ck;ta.className="RNN";ue.registerClass(ta);var Kd=class extends Qe{},im=class extends Kd{constructor(e){super(e);this.DEFAULT_ACTIVATION="tanh",this.DEFAULT_KERNEL_INITIALIZER="glorotNormal",this.DEFAULT_RECURRENT_INITIALIZER="orthogonal",this.DEFAULT_BIAS_INITIALIZER="zeros",this.units=e.units,yn(this.units,"units"),this.activation=zo(e.activation==null?this.DEFAULT_ACTIVATION:e.activation),this.useBias=e.useBias==null?!0:e.useBias,this.kernelInitializer=Nt(e.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.recurrentInitializer=Nt(e.recurrentInitializer||this.DEFAULT_RECURRENT_INITIALIZER),this.biasInitializer=Nt(e.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.kernelRegularizer=Et(e.kernelRegularizer),this.recurrentRegularizer=Et(e.recurrentRegularizer),this.biasRegularizer=Et(e.biasRegularizer),this.kernelConstraint=an(e.kernelConstraint),this.recurrentConstraint=an(e.recurrentConstraint),this.biasConstraint=an(e.biasConstraint),this.dropout=Hu([1,Fo([0,e.dropout==null?0:e.dropout])]),this.recurrentDropout=Hu([1,Fo([0,e.recurrentDropout==null?0:e.recurrentDropout])]),this.dropoutFunc=e.dropoutFunc,this.stateSize=this.units,this.dropoutMask=null,this.recurrentDropoutMask=null}build(e){e=pt(e),this.kernel=this.addWeight("kernel",[e[e.length-1],this.units],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias?this.bias=this.addWeight("bias",[this.units],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0}call(e,t){return q(()=>{if(e=e,e.length!==2)throw new H(`SimpleRNNCell expects 2 input Tensors, got ${e.length}.`);let n=e[1];e=e[0];let s=t.training==null?!1:t.training;0<this.dropout&&this.dropout<1&&this.dropoutMask==null&&(this.dropoutMask=Lo({ones:()=>Es(e),rate:this.dropout,training:s,dropoutFunc:this.dropoutFunc})),0<this.recurrentDropout&&this.recurrentDropout<1&&this.recurrentDropoutMask==null&&(this.recurrentDropoutMask=Lo({ones:()=>Es(n),rate:this.recurrentDropout,training:s,dropoutFunc:this.dropoutFunc}));let r,a=this.dropoutMask,o=this.recurrentDropoutMask;a!=null?r=_r(L(e,a),this.kernel.read()):r=_r(e,this.kernel.read()),this.bias!=null&&(r=pr(r,this.bias.read())),o!=null&&(n=L(n,o));let i=le(r,_r(n,this.recurrentKernel.read()));return this.activation!=null&&(i=this.activation.apply(i)),[i,i]})}getConfig(){let e=super.getConfig(),t={units:this.units,activation:Mo(this.activation),useBias:this.useBias,kernelInitializer:Ot(this.kernelInitializer),recurrentInitializer:Ot(this.recurrentInitializer),biasInitializer:Ot(this.biasInitializer),kernelRegularizer:At(this.kernelRegularizer),recurrentRegularizer:At(this.recurrentRegularizer),biasRegularizer:At(this.biasRegularizer),activityRegularizer:At(this.activityRegularizer),kernelConstraint:rn(this.kernelConstraint),recurrentConstraint:rn(this.recurrentConstraint),biasConstraint:rn(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout};return{...e,...t}}};im.className="SimpleRNNCell";ue.registerClass(im);var TA=class extends ta{constructor(e){e.cell=new im(e);super(e)}call(e,t){return q(()=>{this.cell.dropoutMask!=null&&(se(this.cell.dropoutMask),this.cell.dropoutMask=null),this.cell.recurrentDropoutMask!=null&&(se(this.cell.recurrentDropoutMask),this.cell.recurrentDropoutMask=null);let n=t==null?null:t.mask,s=t==null?null:t.training,r=t==null?null:t.initialState;return super.call(e,{mask:n,training:s,initialState:r})})}static fromConfig(e,t){return new e(t)}};TA.className="SimpleRNN";ue.registerClass(TA);var lm=class extends Kd{constructor(e){super(e);if(this.DEFAULT_ACTIVATION="tanh",this.DEFAULT_RECURRENT_ACTIVATION="hardSigmoid",this.DEFAULT_KERNEL_INITIALIZER="glorotNormal",this.DEFAULT_RECURRENT_INITIALIZER="orthogonal",this.DEFAULT_BIAS_INITIALIZER="zeros",e.resetAfter)throw new H("GRUCell does not support reset_after parameter set to true.");this.units=e.units,yn(this.units,"units"),this.activation=zo(e.activation===void 0?this.DEFAULT_ACTIVATION:e.activation),this.recurrentActivation=zo(e.recurrentActivation===void 0?this.DEFAULT_RECURRENT_ACTIVATION:e.recurrentActivation),this.useBias=e.useBias==null?!0:e.useBias,this.kernelInitializer=Nt(e.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.recurrentInitializer=Nt(e.recurrentInitializer||this.DEFAULT_RECURRENT_INITIALIZER),this.biasInitializer=Nt(e.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.kernelRegularizer=Et(e.kernelRegularizer),this.recurrentRegularizer=Et(e.recurrentRegularizer),this.biasRegularizer=Et(e.biasRegularizer),this.kernelConstraint=an(e.kernelConstraint),this.recurrentConstraint=an(e.recurrentConstraint),this.biasConstraint=an(e.biasConstraint),this.dropout=Hu([1,Fo([0,e.dropout==null?0:e.dropout])]),this.recurrentDropout=Hu([1,Fo([0,e.recurrentDropout==null?0:e.recurrentDropout])]),this.dropoutFunc=e.dropoutFunc,this.implementation=e.implementation,this.stateSize=this.units,this.dropoutMask=null,this.recurrentDropoutMask=null}build(e){e=pt(e);let t=e[e.length-1];this.kernel=this.addWeight("kernel",[t,this.units*3],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units*3],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias?this.bias=this.addWeight("bias",[this.units*3],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint):this.bias=null,this.built=!0}call(e,t){return q(()=>{if(e=e,e.length!==2)throw new H(`GRUCell expects 2 input Tensors (inputs, h, c), got ${e.length}.`);let n=t.training==null?!1:t.training,s=e[1];e=e[0],0<this.dropout&&this.dropout<1&&this.dropoutMask==null&&(this.dropoutMask=Lo({ones:()=>Es(e),rate:this.dropout,training:n,count:3,dropoutFunc:this.dropoutFunc})),0<this.recurrentDropout&&this.recurrentDropout<1&&this.recurrentDropoutMask==null&&(this.recurrentDropoutMask=Lo({ones:()=>Es(s),rate:this.recurrentDropout,training:n,count:3,dropoutFunc:this.dropoutFunc}));let r=this.dropoutMask,a=this.recurrentDropoutMask,o,i,l;0<this.dropout&&this.dropout<1&&(e=L(e,r[0]));let c=_r(e,this.kernel.read());this.useBias&&(c=pr(c,this.bias.read())),0<this.recurrentDropout&&this.recurrentDropout<1&&(s=L(s,a[0]));let u=this.recurrentKernel.read(),[d,p]=nn(u,[2*this.units,this.units],u.rank-1),h=_r(s,d),[f,m,g]=nn(c,3,c.rank-1),[A,x]=nn(h,2,h.rank-1);o=this.recurrentActivation.apply(le(f,A)),i=this.recurrentActivation.apply(le(m,x));let y=_r(L(i,s),p);l=this.activation.apply(le(g,y));let b=le(L(o,s),L(le(1,Ft(o)),l));return[b,b]})}getConfig(){let e=super.getConfig(),t={units:this.units,activation:Mo(this.activation),recurrentActivation:Mo(this.recurrentActivation),useBias:this.useBias,kernelInitializer:Ot(this.kernelInitializer),recurrentInitializer:Ot(this.recurrentInitializer),biasInitializer:Ot(this.biasInitializer),kernelRegularizer:At(this.kernelRegularizer),recurrentRegularizer:At(this.recurrentRegularizer),biasRegularizer:At(this.biasRegularizer),activityRegularizer:At(this.activityRegularizer),kernelConstraint:rn(this.kernelConstraint),recurrentConstraint:rn(this.recurrentConstraint),biasConstraint:rn(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation,resetAfter:!1};return{...e,...t}}};lm.className="GRUCell";ue.registerClass(lm);var NA=class extends ta{constructor(e){e.implementation===0&&console.warn("`implementation=0` has been deprecated, and now defaults to `implementation=1`. Please update your layer call."),e.cell=new lm(e);super(e)}call(e,t){return q(()=>{this.cell.dropoutMask!=null&&(se(this.cell.dropoutMask),this.cell.dropoutMask=null),this.cell.recurrentDropoutMask!=null&&(se(this.cell.recurrentDropoutMask),this.cell.recurrentDropoutMask=null);let n=t==null?null:t.mask,s=t==null?null:t.training,r=t==null?null:t.initialState;return super.call(e,{mask:n,training:s,initialState:r})})}static fromConfig(e,t){return t.implmentation===0&&(t.implementation=1),new e(t)}};NA.className="GRU";ue.registerClass(NA);var Zd=class extends Kd{constructor(e){super(e);this.DEFAULT_ACTIVATION="tanh",this.DEFAULT_RECURRENT_ACTIVATION="hardSigmoid",this.DEFAULT_KERNEL_INITIALIZER="glorotNormal",this.DEFAULT_RECURRENT_INITIALIZER="orthogonal",this.DEFAULT_BIAS_INITIALIZER="zeros",this.units=e.units,yn(this.units,"units"),this.activation=zo(e.activation===void 0?this.DEFAULT_ACTIVATION:e.activation),this.recurrentActivation=zo(e.recurrentActivation===void 0?this.DEFAULT_RECURRENT_ACTIVATION:e.recurrentActivation),this.useBias=e.useBias==null?!0:e.useBias,this.kernelInitializer=Nt(e.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.recurrentInitializer=Nt(e.recurrentInitializer||this.DEFAULT_RECURRENT_INITIALIZER),this.biasInitializer=Nt(e.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.unitForgetBias=e.unitForgetBias,this.kernelRegularizer=Et(e.kernelRegularizer),this.recurrentRegularizer=Et(e.recurrentRegularizer),this.biasRegularizer=Et(e.biasRegularizer),this.kernelConstraint=an(e.kernelConstraint),this.recurrentConstraint=an(e.recurrentConstraint),this.biasConstraint=an(e.biasConstraint),this.dropout=Hu([1,Fo([0,e.dropout==null?0:e.dropout])]),this.recurrentDropout=Hu([1,Fo([0,e.recurrentDropout==null?0:e.recurrentDropout])]),this.dropoutFunc=e.dropoutFunc,this.implementation=e.implementation,this.stateSize=[this.units,this.units],this.dropoutMask=null,this.recurrentDropoutMask=null}build(e){var t;e=pt(e);let n=e[e.length-1];this.kernel=this.addWeight("kernel",[n,this.units*4],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.recurrentKernel=this.addWeight("recurrent_kernel",[this.units,this.units*4],null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint);let s;if(this.useBias){if(this.unitForgetBias){let r=this.biasInitializer,a=this.units;s=new(t=class extends Zs{apply(o,i){let l=r.apply([a]),c=new zf().apply([a]),u=r.apply([a*2]);return Aw(Aw(l,c),u)}},t.className="CustomInit",t)}else s=this.biasInitializer;this.bias=this.addWeight("bias",[this.units*4],null,s,this.biasRegularizer,!0,this.biasConstraint)}else this.bias=null;this.built=!0}call(e,t){return q(()=>{let n=t.training==null?!1:t.training;if(e=e,e.length!==3)throw new H(`LSTMCell expects 3 input Tensors (inputs, h, c), got ${e.length}.`);let s=e[1],r=e[2];e=e[0],0<this.dropout&&this.dropout<1&&this.dropoutMask==null&&(this.dropoutMask=Lo({ones:()=>Es(e),rate:this.dropout,training:n,count:4,dropoutFunc:this.dropoutFunc})),0<this.recurrentDropout&&this.recurrentDropout<1&&this.recurrentDropoutMask==null&&(this.recurrentDropoutMask=Lo({ones:()=>Es(s),rate:this.recurrentDropout,training:n,count:4,dropoutFunc:this.dropoutFunc}));let a=this.dropoutMask,o=this.recurrentDropoutMask,i,l,c,u;0<this.dropout&&this.dropout<1&&(e=L(e,a[0]));let d=_r(e,this.kernel.read());0<this.recurrentDropout&&this.recurrentDropout<1&&(s=L(s,o[0])),d=le(d,_r(s,this.recurrentKernel.read())),this.useBias&&(d=pr(d,this.bias.read()));let[p,h,f,m]=nn(d,4,d.rank-1);i=this.recurrentActivation.apply(p),l=this.recurrentActivation.apply(h),c=le(L(l,r),L(i,this.activation.apply(f))),u=this.recurrentActivation.apply(m);let g=L(u,this.activation.apply(c));return[g,g,c]})}getConfig(){let e=super.getConfig(),t={units:this.units,activation:Mo(this.activation),recurrentActivation:Mo(this.recurrentActivation),useBias:this.useBias,kernelInitializer:Ot(this.kernelInitializer),recurrentInitializer:Ot(this.recurrentInitializer),biasInitializer:Ot(this.biasInitializer),unitForgetBias:this.unitForgetBias,kernelRegularizer:At(this.kernelRegularizer),recurrentRegularizer:At(this.recurrentRegularizer),biasRegularizer:At(this.biasRegularizer),activityRegularizer:At(this.activityRegularizer),kernelConstraint:rn(this.kernelConstraint),recurrentConstraint:rn(this.recurrentConstraint),biasConstraint:rn(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation};return{...e,...t}}};Zd.className="LSTMCell";ue.registerClass(Zd);var EA=class extends ta{constructor(e){e.implementation===0&&console.warn("`implementation=0` has been deprecated, and now defaults to `implementation=1`. Please update your layer call."),e.cell=new Zd(e);super(e)}call(e,t){return q(()=>{this.cell.dropoutMask!=null&&(se(this.cell.dropoutMask),this.cell.dropoutMask=null),this.cell.recurrentDropoutMask!=null&&(se(this.cell.recurrentDropoutMask),this.cell.recurrentDropoutMask=null);let n=t==null?null:t.mask,s=t==null?null:t.training,r=t==null?null:t.initialState;return super.call(e,{mask:n,training:s,initialState:r})})}static fromConfig(e,t){return t.implmentation===0&&(t.implementation=1),new e(t)}};EA.className="LSTM";ue.registerClass(EA);var um=class extends Kd{constructor(e){super(e);this.cells=e.cells}get stateSize(){let e=[];for(let t of this.cells.slice().reverse())Array.isArray(t.stateSize)?e.push(...t.stateSize):e.push(t.stateSize);return e}call(e,t){return q(()=>{e=e;let n=e.slice(1),s=[];for(let o of this.cells.slice().reverse())Array.isArray(o.stateSize)?s.push(n.splice(0,o.stateSize.length)):s.push(n.splice(0,1));s.reverse();let r=[],a;for(let o=0;o<this.cells.length;++o){let i=this.cells[o];n=s[o],o===0?a=[e[0]].concat(n):a=[a[0]].concat(n),a=i.call(a,t),r.push(a.slice(1))}n=[];for(let o of r.slice().reverse())n.push(...o);return[a[0]].concat(n)})}build(e){H1(e)&&(e=e[0]),e=e;let t;this.cells.forEach((n,s)=>{fl(`RNNCell_${s}`,()=>{n.build(e),Array.isArray(n.stateSize)?t=n.stateSize[0]:t=n.stateSize,e=[e[0],t]})}),this.built=!0}getConfig(){let e=super.getConfig(),t=r=>({className:r.getClassName(),config:r.getConfig()}),s={cells:this.cells.map(t)};return{...e,...s}}static fromConfig(e,t,n={}){let s=[];for(let r of t.cells)s.push(fr(r,n));return new e({cells:s})}get trainableWeights(){if(!this.trainable)return[];let e=[];for(let t of this.cells)e.push(...t.trainableWeights);return e}get nonTrainableWeights(){let e=[];for(let t of this.cells)e.push(...t.nonTrainableWeights);if(!this.trainable){let t=[];for(let n of this.cells)t.push(...n.trainableWeights);return t.concat(e)}return e}getWeights(){let e=[];for(let t of this.cells)e.push(...t.weights);return j1(e)}setWeights(e){let t=[];for(let n of this.cells){let s=n.weights.length,r=e.splice(s);for(let a=0;a<n.weights.length;++a)t.push([n.weights[a],r[a]])}q1(t)}};um.className="StackedRNNCells";ue.registerClass(um);function Lo(e){let{ones:t,rate:n,training:s=!1,count:r=1,dropoutFunc:a}=e,o=()=>a!=null?a(t(),n):xw(t(),n),i=()=>Vd(o,t,s);return!r||r<=1?fn(i().clone()):Array(r).fill(void 0).map(i).map(c=>fn(c.clone()))}var Tk=class extends ta{constructor(e){if(e.unroll)throw new Le("Unrolling is not possible with convolutional RNNs.");if(Array.isArray(e.cell))throw new Le("It is not possible at the moment to stack convolutional cells.");super(e);this.inputSpec=[new Kt({ndim:5})]}call(e,t){return q(()=>{if(this.cell.dropoutMask!=null&&(se(this.cell.dropoutMask),this.cell.dropoutMask=null),this.cell.recurrentDropoutMask!=null&&(se(this.cell.recurrentDropoutMask),this.cell.recurrentDropoutMask=null),t&&t.constants)throw new H("ConvRNN2D cell does not support constants");let n=t==null?null:t.mask,s=t==null?null:t.training,r=t==null?null:t.initialState;return super.call(e,{mask:n,training:s,initialState:r})})}computeOutputShape(e){let t=this.computeSingleOutputShape(e);return this.returnSequences||(t=[t[0],...t.slice(2)]),this.returnState&&(t=[t,...Array(2).fill([e[0],...t.slice(-3)])]),t}getInitialState(e){return q(()=>{let{stateSize:t}=this.cell,n=e.shape,s=this.computeSingleOutputShape(n),r=[s[0],...s.slice(2)],a=Gt(r);return Array.isArray(t)?Array(t.length).fill(a):[a]})}resetStates(e,t=!1){q(()=>{if(!this.stateful)throw new Jr("Cannot call resetStates() on an RNN Layer that is not stateful.");let n=this.inputSpec[0].shape,s=this.computeSingleOutputShape(n),r=[s[0],...s.slice(2)];if(n[0]==null)throw new H("If an RNN is stateful, it needs to know its batch size. Specify the batch size of your input tensors: \n- If using a Sequential model, specify the batch size by passing a `batchInputShape` option to your first layer.\n- If using the functional API, specify the batch size by passing a `batchShape` option to your Input layer.");if(this.getStates()==null)Array.isArray(this.cell.stateSize)?this.states_=this.cell.stateSize.map(()=>Gt(r)):this.states_=[Gt(r)];else if(e==null)se(this.states_),this.keptStates!=null&&(se(this.keptStates),this.keptStates=[]),Array.isArray(this.cell.stateSize)?this.states_=this.cell.stateSize.map(()=>Gt(r)):this.states_[0]=Gt(r);else{if(Array.isArray(e)||(e=[e]),e.length!==this.states_.length)throw new H(`Layer ${this.name} expects ${this.states_.length} state(s), but it received ${e.length} state value(s). Input received: ${e}`);t?this.keptStates.push(this.states_.slice()):se(this.states_);for(let o=0;o<this.states_.length;++o){let i=e[o],l=r;if(!v.arraysEqual(i.shape,l))throw new H(`State ${o} is incompatible with layer ${this.name}: expected shape=${l}, received shape=${i.shape}`);this.states_[o]=i}}this.states_=this.states_.map(o=>fn(o.clone()))})}computeSingleOutputShape(e){let{dataFormat:t,filters:n,kernelSize:s,padding:r,strides:a,dilationRate:o}=this.cell,i=t==="channelsFirst",l=e[i?3:2],c=e[i?4:3],u=mr(l,s[0],r,a[0],o[0]),d=mr(c,s[1],r,a[1],o[1]);return[...e.slice(0,2),...i?[n,u,d]:[u,d,n]]}};Tk.className="ConvRNN2D";var cm=class extends Zd{constructor(e){let{filters:t,kernelSize:n,strides:s,padding:r,dataFormat:a,dilationRate:o}=e;super({...e,units:t});this.filters=t,yn(this.filters,"filters"),this.kernelSize=Ku(n,2,"kernelSize"),this.kernelSize.forEach(i=>yn(i,"kernelSize")),this.strides=Ku(s||1,2,"strides"),this.strides.forEach(i=>yn(i,"strides")),this.padding=r||"valid",$s(this.padding),this.dataFormat=a||"channelsLast",Ht(this.dataFormat),this.dilationRate=Ku(o||1,2,"dilationRate"),this.dilationRate.forEach(i=>yn(i,"dilationRate"))}build(e){var t;e=pt(e);let n=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[n]==null)throw new H(`The channel dimension of the input should be defined. Found ${e[n]}`);let s=e[n],r=4,a=this.kernelSize.concat([s,this.filters*r]);this.kernel=this.addWeight("kernel",a,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint);let o=this.kernelSize.concat([this.filters,this.filters*r]);if(this.recurrentKernel=this.addWeight("recurrent_kernel",o,null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias){let i;if(this.unitForgetBias){let l=this.biasInitializer,c=this.filters;i=new(t=class extends Zs{apply(u,d){let p=l.apply([c]),h=hs([c]),f=l.apply([c*2]);return F1([p,h,f])}},t.className="CustomInit",t)}else i=this.biasInitializer;this.bias=this.addWeight("bias",[this.filters*r],null,i,this.biasRegularizer,!0,this.biasConstraint)}this.built=!0}call(e,t){return q(()=>{if(e.length!==3)throw new H(`ConvLSTM2DCell expects 3 input Tensors (inputs, h, c), got ${e.length}.`);let n=t.training||!1,s=e[0],r=e[1],a=e[2],o=4;0<this.dropout&&this.dropout<1&&this.dropoutMask==null&&(this.dropoutMask=Lo({ones:()=>Es(s),rate:this.dropout,training:n,count:o,dropoutFunc:this.dropoutFunc}));let i=this.dropoutMask,l=(Z,J,ee)=>!J||!J[ee]?Z:L(J[ee],Z),c=l(s,i,0),u=l(s,i,1),d=l(s,i,2),p=l(s,i,3);0<this.recurrentDropout&&this.recurrentDropout<1&&this.recurrentDropoutMask==null&&(this.recurrentDropoutMask=Lo({ones:()=>Es(r),rate:this.recurrentDropout,training:n,count:o,dropoutFunc:this.dropoutFunc}));let h=this.recurrentDropoutMask,f=l(r,h,0),m=l(r,h,1),g=l(r,h,2),A=l(r,h,3),x=3,[y,b,w,k]=nn(this.kernel.read(),o,x),[C,N,R,F]=this.useBias?nn(this.bias.read(),o):[null,null,null,null];c=this.inputConv(c,y,C,this.padding),u=this.inputConv(u,b,N,this.padding),d=this.inputConv(d,w,R,this.padding),p=this.inputConv(p,k,F,this.padding);let[_,P,T,M]=nn(this.recurrentKernel.read(),o,x);f=this.recurrentConv(f,_),m=this.recurrentConv(m,P),g=this.recurrentConv(g,T),A=this.recurrentConv(A,M);let U=this.recurrentActivation.apply(le(c,f)),j=this.recurrentActivation.apply(le(u,m)),z=le(L(j,a),L(U,this.activation.apply(le(d,g)))),X=L(this.recurrentActivation.apply(le(p,A)),this.activation.apply(z));return[X,X,z]})}getConfig(){let{units:e,...t}=super.getConfig(),n={filters:this.filters,kernelSize:this.kernelSize,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,strides:this.strides};return{...t,...n}}inputConv(e,t,n,s){let r=No(e,t,this.strides,s||"valid",this.dataFormat==="channelsFirst"?"NCHW":"NHWC",this.dilationRate);return n?pr(r,n,this.dataFormat):r}recurrentConv(e,t){return No(e,t,1,"same",this.dataFormat==="channelsFirst"?"NCHW":"NHWC")}};cm.className="ConvLSTM2DCell";ue.registerClass(cm);var RA=class extends Tk{constructor(e){let t=new cm(e);super({...e,cell:t})}static fromConfig(e,t){return new e(t)}};RA.className="ConvLSTM2D";ue.registerClass(RA);var dm=class extends Qe{constructor(e){super(e);this.rate=Math.max(Math.min(e.rate,1),0),this.noiseShape=e.noiseShape,this.seed=e.seed,this.supportsMasking=!0}getNoiseShape(e){if(this.noiseShape==null)return this.noiseShape;let t=e.shape,n=[];for(let s=0;s<this.noiseShape.length;++s)n.push(this.noiseShape[s]==null?t[s]:this.noiseShape[s]);return n}call(e,t){return q(()=>{this.invokeCallHook(e,t);let n=Be(e);if(0<this.rate&&this.rate<1){let s=t.training==null?!1:t.training,r=this.getNoiseShape(n);return Vd(()=>xw(n,this.rate,r,this.seed),()=>n,s)}return e})}getConfig(){let e={rate:this.rate,noiseShape:this.noiseShape,seed:this.seed},t=super.getConfig();return Object.assign(e,t),e}dispose(){return super.dispose()}};dm.className="Dropout";ue.registerClass(dm);var $A=class extends dm{constructor(e){super(e);this.inputSpec=[{ndim:3}]}getNoiseShape(e){let t=e.shape;return[t[0],1,t[2]]}};$A.className="SpatialDropout1D";ue.registerClass($A);var _A=class extends Qe{constructor(e){super(e);if(this.activation=null,this.useBias=!0,this.kernel=null,this.bias=null,this.DEFAULT_KERNEL_INITIALIZER="glorotNormal",this.DEFAULT_BIAS_INITIALIZER="zeros",e.batchInputShape==null&&e.inputShape==null&&e.inputDim!=null){let t=null;e.batchSize!=null&&(t=e.batchSize),this.batchInputShape=[t,e.inputDim]}this.units=e.units,yn(this.units,"units"),this.activation=zo(e.activation),e.useBias!=null&&(this.useBias=e.useBias),this.kernelInitializer=Nt(e.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.biasInitializer=Nt(e.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.kernelConstraint=an(e.kernelConstraint),this.biasConstraint=an(e.biasConstraint),this.kernelRegularizer=Et(e.kernelRegularizer),this.biasRegularizer=Et(e.biasRegularizer),this.activityRegularizer=Et(e.activityRegularizer),this.supportsMasking=!0,this.inputSpec=[{minNDim:2}]}build(e){e=pt(e);let t=e[e.length-1];this.kernel==null&&(this.kernel=this.addWeight("kernel",[t,this.units],null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.units],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint))),this.inputSpec=[{minNDim:2,axes:{[-1]:t}}],this.built=!0}computeOutputShape(e){e=pt(e);let t=e.slice();return t[t.length-1]=this.units,t}call(e,t){return q(()=>{this.invokeCallHook(e,t);let n=Be(e),s=lw(this.activation.getClassName()),r;return s!=null?r=_r(n,this.kernel.read(),s,this.bias?this.bias.read():null):(r=_r(n,this.kernel.read()),this.bias!=null&&(r=pr(r,this.bias.read())),this.activation!=null&&(r=this.activation.apply(r))),r})}getConfig(){let e={units:this.units,activation:Mo(this.activation),useBias:this.useBias,kernelInitializer:Ot(this.kernelInitializer),biasInitializer:Ot(this.biasInitializer),kernelRegularizer:At(this.kernelRegularizer),biasRegularizer:At(this.biasRegularizer),activityRegularizer:At(this.activityRegularizer),kernelConstraint:rn(this.kernelConstraint),biasConstraint:rn(this.biasConstraint)},t=super.getConfig();return Object.assign(e,t),e}};_A.className="Dense";ue.registerClass(_A);var DA=class extends Qe{constructor(e){e=e||{};super(e);this.inputSpec=[{minNDim:3}],this.dataFormat=e.dataFormat}computeOutputShape(e){e=pt(e);for(let t of e.slice(1))if(t==null)throw new H(`The shape of the input to "Flatten" is not fully defined (got ${e.slice(1)}). Make sure to pass a complete "input_shape" or "batch_input_shape" argument to the first layer in your model.`);return[e[0],Po(e,1)]}call(e,t){return q(()=>{this.invokeCallHook(e,t);let n=Be(e);if(this.dataFormat==="channelsFirst"&&n.rank>1){let s=[0];for(let r=2;r<n.rank;++r)s.push(r);s.push(1),n=Ye(n,s)}return IL(n)})}getConfig(){let e={};this.dataFormat!=null&&(e.dataFormat=this.dataFormat);let t=super.getConfig();return Object.assign(e,t),e}};DA.className="Flatten";ue.registerClass(DA);var PA=class extends Qe{constructor(e){super(e);this.supportsMasking=!0,this.activation=zo(e.activation)}call(e,t){return q(()=>{this.invokeCallHook(e,t);let n=Be(e);return this.activation.apply(n)})}getConfig(){let e={activation:Mo(this.activation)},t=super.getConfig();return Object.assign(e,t),e}};PA.className="Activation";ue.registerClass(PA);var FA=class extends Qe{constructor(e){super(e);this.n=e.n,this.inputSpec=[{ndim:2}]}computeOutputShape(e){return[e[0],this.n,e[1]]}call(e,t){return q(()=>(e=Be(e),kL(e,this.n)))}getConfig(){let e={n:this.n},t=super.getConfig();return Object.assign(e,t),e}};FA.className="RepeatVector";ue.registerClass(FA);var OA=class extends Qe{constructor(e){super(e);this.targetShape=e.targetShape;for(let t=0;t<this.targetShape.length;++t)this.isUnknown(this.targetShape[t])&&(this.targetShape[t]=null)}isUnknown(e){return e<0||e==null}fixUnknownDimension(e,t){let n="Total size of new array must be unchanged.",s=t.slice(),r=1,a=null;for(let i=0;i<s.length;++i){let l=s[i];if(this.isUnknown(l))if(a===null)a=i;else throw new H("Can only specifiy one unknown dimension.");else r*=l}let o=Po(e);if(a!==null){if(r===0||o%r!=0)throw new H(n);s[a]=o/r}else if(o!==r)throw new H(n);return s}computeOutputShape(e){let t=!1;for(let n=0;n<e.length;++n)if(this.isUnknown(e[n])){t=!0;break}return t?e.slice(0,1).concat(this.targetShape):e.slice(0,1).concat(this.fixUnknownDimension(e.slice(1),this.targetShape))}call(e,t){return q(()=>{this.invokeCallHook(e,t);let n=Be(e),s=n.shape,r=s.slice(0,1).concat(this.fixUnknownDimension(s.slice(1),this.targetShape));return G(n,r)})}getConfig(){let e={targetShape:this.targetShape},t=super.getConfig();return Object.assign(e,t),e}};OA.className="Reshape";ue.registerClass(OA);var MA=class extends Qe{constructor(e){super(e);if(e.dims==null)throw new Error("Required configuration field `dims` is missing during Permute constructor call.");if(!Array.isArray(e.dims))throw new Error(`Permute constructor requires \`dims\` to be an Array, but received ${e.dims} instead.`);let t=dr(1,e.dims.length+1);if(!v.arraysEqual(e.dims.slice().sort(),t))throw new Error("Invalid permutation `dims`: "+JSON.stringify(e.dims)+" `dims` must contain consecutive integers starting from 1.");this.dims=e.dims,this.dimsIncludingBatch=[0].concat(this.dims),this.inputSpec=[new Kt({ndim:this.dims.length+1})]}computeOutputShape(e){e=pt(e);let t=e.slice();return this.dims.forEach((n,s)=>{t[s+1]=e[n]}),t}call(e,t){return Ye(Be(e),this.dimsIncludingBatch)}getConfig(){let e={dims:this.dims},t=super.getConfig();return Object.assign(e,t),e}};MA.className="Permute";ue.registerClass(MA);var zA=class extends Qe{constructor(e){super(e==null?{}:e);this.supportsMasking=!0,e!=null?this.maskValue=e.maskValue==null?0:e.maskValue:this.maskValue=0}computeOutputShape(e){return e}getConfig(){let e=super.getConfig(),t={maskValue:this.maskValue};return Object.assign(t,e),t}computeMask(e,t){let n=Be(e),s=-1;return rf(zu(n,this.maskValue),s)}call(e,t){return q(()=>{this.invokeCallHook(e,t);let n=Be(e),s=-1,r=!0,a=rf(zu(n,this.maskValue),s,r);return L(n,de(a,n.dtype))})}};zA.className="Masking";ue.registerClass(zA);var LA=class extends Qe{constructor(e){super(e);if(this.embeddings=null,this.DEFAULT_EMBEDDINGS_INITIALIZER="randomUniform",e.batchInputShape==null&&e.inputShape==null){let t=null;e.batchSize!=null&&(t=e.batchSize),e.inputLength==null?this.batchInputShape=[t,null]:this.batchInputShape=[t].concat(kt(e.inputLength))}this.inputDim=e.inputDim,yn(this.inputDim,"inputDim"),this.outputDim=e.outputDim,yn(this.outputDim,"outputDim"),this.embeddingsInitializer=Nt(e.embeddingsInitializer||this.DEFAULT_EMBEDDINGS_INITIALIZER),this.embeddingsRegularizer=Et(e.embeddingsRegularizer),this.activityRegularizer=Et(e.activityRegularizer),this.embeddingsConstraint=an(e.embeddingsConstraint),this.maskZero=e.maskZero,this.supportsMasking=e.maskZero,this.inputLength=e.inputLength}build(e){this.embeddings=this.addWeight("embeddings",[this.inputDim,this.outputDim],this.dtype,this.embeddingsInitializer,this.embeddingsRegularizer,!0,this.embeddingsConstraint),this.built=!0}warnOnIncompatibleInputShape(e){}computeMask(e,t){return q(()=>this.maskZero?(e=Be(e),zu(e,Je(e))):null)}computeOutputShape(e){if(e=pt(e),this.inputLength==null)return[...e,this.outputDim];let t=kt(this.inputLength);if(t.length!==e.length-1)throw new H(`"inputLength" is ${this.inputLength}, but received input shape has shape ${e}`);{let n=0;for(let s=0;s<t.length;++s){let r=t[s],a=e[s+1];if(r!=null&&a!=null&&r!==a)throw new H(`"inputLength" is ${this.inputLength}, but received input shape has shape ${e}`);r==null&&(t[n]=a),n++}}return[e[0],...t,this.outputDim]}call(e,t){return q(()=>{this.invokeCallHook(e,t);let n=Be(e);n.dtype!=="int32"&&(n=Ff(n,"int32"));let s=yw(this.embeddings.read(),G(n,[n.size]));return G(s,pt(this.computeOutputShape(n.shape)))})}getConfig(){let e={inputDim:this.inputDim,outputDim:this.outputDim,embeddingsInitializer:Ot(this.embeddingsInitializer),embeddingsRegularizer:At(this.embeddingsRegularizer),activityRegularizer:At(this.activityRegularizer),embeddingsConstraint:rn(this.embeddingsConstraint),maskZero:this.maskZero,inputLength:this.inputLength},t=super.getConfig();return Object.assign(e,t),e}};LA.className="Embedding";ue.registerClass(LA);var xl=class extends Qe{constructor(e){super(e||{});this.supportsMasking=!0}mergeFunction(e){throw new Le}computeElementwiseOpOutputShape(e,t){if(e==null||t==null)return null;if(e.length<t.length)return this.computeElementwiseOpOutputShape(t,e);if(t.length===0)return e;let n=e.slice(0,e.length-t.length);for(let s=0;s<t.length;++s){let r=e[e.length-t.length+s],a=t[s];if(r==null||a==null||r<0||a<0)n.push(null);else if(r===1)n.push(a);else if(a===1)n.push(r);else{if(r!==a)throw new H("Operands could not be broadcast together with shapes "+JSON.stringify(e)+" "+JSON.stringify(t));n.push(r)}}return n}build(e){if(Array.isArray(e)&&!Array.isArray(e[0])&&(e=[pt(e)]),e=e,e.length<2)throw new H(`A merge layer should be called on an Array of at least 2 inputs. Got ${e.length} input(s).`);let t=[];for(let r of e)r!=null&&r[0]!==null&&t.push(r[0]);if(t=Do(t),t.length>1)throw new H(`Can not merge tensors with different batch sizes. Got tensors with shapes: ${JSON.stringify(e)}.`);let n=e[0]==null?null:e[0].slice(1);for(let r=1;r<e.length;++r){let a=e[r]==null?null:e[r].slice(1);n=this.computeElementwiseOpOutputShape(n,a)}let s=e.map(r=>r.length);e.indexOf(null)===-1&&Do(s).length===1?this.reshapeRequired=!1:this.reshapeRequired=!0}call(e,t){return q(()=>{if(e=e,this.reshapeRequired){let n=[],s=e.map(r=>r.rank);if(s.indexOf(null)===-1){let r=Fo(s);for(let a of e){let o=a.rank;for(let i=0;i<r-o;++i)a=Bd(a,1);n.push(a)}return this.mergeFunction(n)}else{let r=!1;for(let i of e){let l=i.rank;if(l==null){let c=i.shape,u=c[0],d=c.slice(1).concat([u]),p=G(i,[u].concat(Po(c.slice(1))));p=Ye(p,[1,0]),p=G(p,d),n.push(p),r=!0}else if(l>1){let c=dr(1,l).concat([0]);n.push(Ye(i,c)),r=!0}else n.push(i)}let a=this.mergeFunction(n),o=a.rank;if(r){if(o==null){let i=a.shape,l=i.length,c=i[l-1],u=[c].concat(i.slice(0,i.length-1));a=G(Ye(G(a,[-1,c]),[1,0]),u)}else if(o>1){let i=[o-1].concat(dr(0,o-1));a=Ye(a,i)}}return a}}else return this.mergeFunction(e)})}computeOutputShape(e){e=e;let t;e[0]==null?t=null:t=e[0].slice(1);for(let s=1;s<e.length;++s){let r=e[s]==null?null:e[s].slice(1);t=this.computeElementwiseOpOutputShape(t,r)}let n=[];for(let s of e)s!=null&&s[0]!==null&&n.push(s[0]);return n=Do(n),n.length===1?t=n.concat(t):t=[null].concat(t),t}computeMask(e,t){return q(()=>{if(t==null)return null;if(!Array.isArray(t))throw new H("`mask` should be an Array");if(!Array.isArray(e))throw new H("`inputs` should be an Array");if(t.length!==e.length)throw new H(`The Array 'inputs' and 'mask' are expected to have the same length, but have different lengths (${e.length} vs ${t.length})`);if(t.every(s=>s==null))return null;t=t.map(s=>s==null?s:mn(s,0));let n=t[0];for(let s=1;s<t.length-1;++s)n=ir(n,t[s]);return n})}},BA=class extends xl{constructor(e){super(e)}mergeFunction(e){return q(()=>{let t=e[0].clone();for(let n=1;n<e.length;++n)t=le(t,e[n]);return t})}};BA.className="Add";ue.registerClass(BA);var WA=class extends xl{constructor(e){super(e)}mergeFunction(e){return q(()=>{let t=e[0].clone();for(let n=1;n<e.length;++n)t=L(t,e[n]);return t})}};WA.className="Multiply";ue.registerClass(WA);var VA=class extends xl{constructor(e){super(e)}mergeFunction(e){return q(()=>{let t=e[0].clone();for(let n=1;n<e.length;++n)t=le(t,e[n]);return L(1/e.length,t)})}};VA.className="Average";ue.registerClass(VA);var UA=class extends xl{constructor(e){super(e)}mergeFunction(e){return q(()=>{let t=e[0];for(let n=1;n<e.length;++n)t=Kr(t,e[n]);return t})}};UA.className="Maximum";ue.registerClass(UA);var GA=class extends xl{constructor(e){super(e)}mergeFunction(e){return q(()=>{let t=e[0];for(let n=1;n<e.length;++n)t=Ed(t,e[n]);return t})}};GA.className="Minimum";ue.registerClass(GA);var HA=class extends xl{constructor(e){super(e);this.DEFAULT_AXIS=-1,e==null&&(e={}),this.axis=e.axis==null?this.DEFAULT_AXIS:e.axis,this.supportsMasking=!0,this.reshapeRequired=!1}build(e){if(!(Array.isArray(e)&&Array.isArray(e[0]))||e.length===1)throw new H("A `Concatenate` layer should be called on a list of at least 2 inputs");e=e;let t=!0;for(let s of e)if(s!=null){t=!1;break}if(t)return;let n=[];for(let s=0;s<e.length;++s){let r=e[s].slice();r.splice(this.axis,1);let a=!1;for(let o of n)if(v.arraysEqual(o,r)){a=!0;break}a||n.push(r)}if(n.length>1)throw new H("A `Concatenate` layer requires inputs with matching shapes except for the concat axis. Got input shapes: "+JSON.stringify(e))}mergeFunction(e){return q(()=>F1(e,this.axis))}computeOutputShape(e){if(!(Array.isArray(e)&&Array.isArray(e[0])))throw new H("A `Concatenate` layer should be called on a list of inputs.");let t=e,n=t[0].slice(),s=this.axis<0?n.length+this.axis:this.axis;for(let r of t.slice(1)){if(n[s]==null||r[s]==null){n[s]=null;break}n[s]+=r[s]}return n}computeMask(e,t){if(t==null)return null;if(!Array.isArray(t))throw new H("`mask` should be an array for Concatenate");if(!Array.isArray(e))throw new H("`inputs` should be an array for Concatenate");if(t.length!==e.length)throw new H(`Mismatch in the length of mask (${t.length}) and the legnth of inputs (${e.length})`);return q(()=>{let n=!0;if(t.forEach(a=>{if(a!=null){n=!1;return}}),n)return null;let s=[];for(let a=0;a<e.length;++a)t[a]==null?s.push(de(Es(e[a]),"bool")):t[a].rank<e[a].rank?s.push(mn(t[a],-1)):s.push(t[a]);let r=vt(s,this.axis);return B2(r,-1,!1)})}getConfig(){let e={axis:this.axis},t=super.getConfig();return Object.assign(e,t),e}};HA.className="Concatenate";ue.registerClass(HA);function Yd(e,t){for(;e<0;)e+=t;return e}function oW(e,t,n){if(e.shape.length>3||t.shape.length>3)throw new Le("batchDot is not implemented for tensors of 4D or higher rank yet");if(v.assert(e.shape.length>=2,()=>`batchDot requires the rank of x to be >= 2, but got ${e.shape.length}`),v.assert(e.shape.length>=2,()=>`batchDot requires the rank of y to be >= 2, but got ${t.shape.length}`),typeof n=="number"&&(n=[n,n]),e.dtype==="complex64"||t.dtype==="complex64")throw new Le("batchDot is not implemented for complex64-type Tensors yet.");let s=e.shape.length,r=t.shape.length;n==null&&(n=[s-1,r-2]);let a=n;return q(()=>{let o;if(s>r){o=s-r;let l=[];for(let c=0;c<o;++c)l.push(1);t=G(t,t.shape.concat(l))}else if(r>s){o=r-s;let l=[];for(let c=0;c<o;++c)l.push(1);e=G(e,e.shape.concat(l))}else o=0;let i;if(e.shape.length===2&&t.shape.length===2)a[0]===a[1]?i=Se(L(e,t),a[0]):i=Se(L(Ye(e,[1,0]),t),a[1]);else{let l=a[0]!==e.shape.length-1,c=a[1]===t.shape.length-1;i=Ue(e,t,l,c)}if(o>0){let l;s>r?l=s+r-3:l=s-1;let c=[];for(let u=l;u<l+o;++u)c.push(u);i=ot(i,c)}return i.shape.length===1&&(i=mn(i,1)),i})}var jA=class extends xl{constructor(e){super(e);this.axes=e.axes,this.normalize=e.normalize==null?!1:e.normalize,this.supportsMasking=!0,this.reshapeRequired=!1}build(e){v.assert(Array.isArray(e)&&e.length===2&&Array.isArray(e[0])&&Array.isArray(e[1]),()=>"A `Dot` layer should be called on a list of exactly 2 inputs.");let t=e[0],n=e[1];if(t.length>3||n.length>3)throw new Le("Dot layer does not support tensors of 4D or higher rank yet.");let s=this.interpretAxes(t,n);if(t[s[0]]!==n[s[1]])throw new H(`Dimension incompatibility: ${t[s[0]]} !== ${n[s[1]]}`)}mergeFunction(e){if(e.length!==2)throw new H(`A \`Dot\` layer must be called on exactly 2 inputs, but received ${e.length} input(s).`);let t=e[0],n=e[1],s;return Array.isArray(this.axes)?s=this.axes.map((r,a)=>Yd(r,e[a].shape.length)):s=[Yd(this.axes,t.shape.length),Yd(this.axes,n.shape.length)],this.normalize&&(t=Zf(t,s[0]),n=Zf(n,s[1])),oW(t,n,s)}interpretAxes(e,t){let n;return Array.isArray(this.axes)?n=this.axes:n=[Yd(this.axes,e.length),Yd(this.axes,t.length)],n}computeOutputShape(e){v.assert(Array.isArray(e)&&e.length===2&&Array.isArray(e[0])&&Array.isArray(e[1]),()=>"A `Dot` layer should be called on a list of exactly 2 inputs.");let t=e[0].slice(),n=e[1].slice();if(t.length>3||n.length>3)throw new Le("Dot layer does not support tensors of 4D or higher rank yet.");let s=this.interpretAxes(t,n);t.splice(s[0],1),n.splice(s[1],1),n.splice(0,1);let r=t.concat(n);return r.length===1&&r.push(1),r}computeMask(e,t){return null}getConfig(){let e={axes:this.axes,normalize:this.normalize},t=super.getConfig();return Object.assign(e,t),e}};jA.className="Dot";ue.registerClass(jA);var qA=class extends Qe{constructor(e){super(e);this.supportsMasking=!0,this.stddev=e.stddev}computeOutputShape(e){return e}getConfig(){let e=super.getConfig(),t={stddev:this.stddev};return Object.assign(t,e),t}call(e,t){return q(()=>{this.invokeCallHook(e,t);let n=Be(e);return Vd(()=>le(Mf(n.shape,0,this.stddev),n),()=>n,t.training||!1)})}};qA.className="GaussianNoise";ue.registerClass(qA);var XA=class extends Qe{constructor(e){super(e);this.supportsMasking=!0,this.rate=e.rate}computeOutputShape(e){return e}getConfig(){let e=super.getConfig(),t={rate:this.rate};return Object.assign(t,e),t}call(e,t){return q(()=>{this.invokeCallHook(e,t);let n=Be(e);return this.rate>0&&this.rate<1?Vd(()=>{let r=Math.sqrt(this.rate/(1-this.rate));return L(n,Mf(n.shape,1,r))},()=>n,t.training||!1):n})}};XA.className="GaussianDropout";ue.registerClass(XA);var KA=class extends Qe{constructor(e){super(e);this.supportsMasking=!0,this.rate=e.rate,this.noiseShape=e.noiseShape}_getNoiseShape(e){return this.noiseShape||Be(e).shape}computeOutputShape(e){return e}getConfig(){let e=super.getConfig(),t={rate:this.rate};return Object.assign(t,e),t}call(e,t){return q(()=>{if(this.rate<1&&this.rate>0){let n=this._getNoiseShape(e);return Vd(()=>{let r=Be(e),a=1.6732632423543772,o=1.0507009873554805,i=-a*o,l=il(Lu(n),this.rate);l=Ff(l,"float32");let c=((1-this.rate)*(1+this.rate*i**2))**-.5,u=-c*i*this.rate,d=le(L(r,l),L(le(l,-1),i));return le(L(d,c),u)},()=>Be(e),t.training||!1)}return e})}};KA.className="AlphaDropout";ue.registerClass(KA);function Jd(e,t,n,s,r,a=.001){let o;if(e.rank===2)o=G3(e,t,n,s,r,a);else if(e.rank===3)o=H3(e,t,n,s,r,a);else if(e.rank===4)o=j3(e,t,n,s,r,a);else throw new Le(`batchNormalization is not implemented for array of rank ${e.rank} yet`);return o}function iW(e,t,n,s,r=.001){return q(()=>{let a=gf(e,s),o=a.mean,i=a.variance;return[Jd(e,o,i,n,t,r),o,i]})}function lW(e,t,n,s,r=.001){return q(()=>{let a=gf(e,s),o=a.mean,i=a.variance,l=[];for(let f of dr(0,e.rank))s.indexOf(f)!==-1?l.push(1):l.push(e.shape[f]);let c=G(o,l),u=G(i,l),d=t==null?null:G(t,l),p=n==null?null:G(n,l);return[Jd(e,c,u,p,d,r),o,i]})}function uW(e,t,n,s,r=.001){return v.arraysEqual(s.slice().sort(),dr(0,e.rank-1))?iW(e,t,n,s,r):lW(e,t,n,s,r)}var ZA=class extends Qe{constructor(e){e==null&&(e={});super(e);this.supportsMasking=!0,this.axis=e.axis==null?-1:e.axis,this.momentum=e.momentum==null?.99:e.momentum,this.epsilon=e.epsilon==null?.001:e.epsilon,this.center=e.center==null?!0:e.center,this.scale=e.scale==null?!0:e.scale,this.betaInitializer=Nt(e.betaInitializer||"zeros"),this.gammaInitializer=Nt(e.gammaInitializer||"ones"),this.movingMeanInitializer=Nt(e.movingMeanInitializer||"zeros"),this.movingVarianceInitializer=Nt(e.movingVarianceInitializer||"ones"),this.betaConstraint=an(e.betaConstraint),this.gammaConstraint=an(e.gammaConstraint),this.betaRegularizer=Et(e.betaRegularizer),this.gammaRegularizer=Et(e.gammaRegularizer)}build(e){e=pt(e);let t=this.axis>=0?this.axis:this.axis+e.length,n=e[t];if(n==null)throw new H(`Axis ${t} of input tensor should have a defined dimension but the layer received an input with shape ${JSON.stringify(e)}.`);this.inputSpec=[new Kt({ndim:e.length,axes:{[t]:n}})];let s=[n];this.scale&&(this.gamma=this.addWeight("gamma",s,null,this.gammaInitializer,this.gammaRegularizer,!0,this.gammaConstraint)),this.center&&(this.beta=this.addWeight("beta",s,null,this.betaInitializer,this.betaRegularizer,!0,this.betaConstraint)),this.movingMean=this.addWeight("moving_mean",s,null,this.movingMeanInitializer,null,!1),this.movingVariance=this.addWeight("moving_variance",s,null,this.movingVarianceInitializer,null,!1),this.built=!0}call(e,t){return q(()=>{let n=t.training==null?!1:t.training,s=Be(e),r=s.shape,a=r.length,o=dr(0,a),i=this.axis>=0?this.axis:this.axis+a;o.splice(i,1);let l=dl(1,a);l[i]=r[i];let c=o.slice();c.sort();let u=!v.arraysEqual(c,dr(0,a).slice(0,a-1)),d=()=>{if(u){let A=G(this.movingMean.read(),l),x=G(this.movingVariance.read(),l),y=this.center?G(this.beta.read(),l):null,b=this.scale?G(this.gamma.read(),l):null;return Jd(s,A,x,y,b,this.epsilon)}else return Jd(s,this.movingMean.read(),this.movingVariance.read(),this.beta==null?null:this.beta.read(),this.gamma==null?null:this.gamma.read(),this.epsilon)};if(!n)return d();let[p,h,f]=uW(s,this.gamma.read(),this.beta.read(),o,this.epsilon),m=(A,x,y)=>{q(()=>{let b=1-y,w=A.read(),k=L(me(w,x),b);A.write(me(w,k))})};return(()=>{m(this.movingMean,h,this.momentum),m(this.movingVariance,f,this.momentum)})(),p})}getConfig(){let e={axis:this.axis,momentum:this.momentum,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:Ot(this.betaInitializer),gammaInitializer:Ot(this.gammaInitializer),movingMeanInitializer:Ot(this.movingMeanInitializer),movingVarianceInitializer:Ot(this.movingVarianceInitializer),betaRegularizer:At(this.betaRegularizer),gammaRegularizer:At(this.gammaRegularizer),betaConstraint:rn(this.betaConstraint),gammaConstraint:rn(this.gammaConstraint)},t=super.getConfig();return Object.assign(e,t),e}};ZA.className="BatchNormalization";ue.registerClass(ZA);var YA=class extends Qe{constructor(e){e==null&&(e={});super(e);if(this.axis=e.axis==null?-1:e.axis,typeof this.axis=="number"){if(!Number.isInteger(this.axis))throw new Error(`Expected axis to be an integer, but received ${this.axis}`)}else if(Array.isArray(this.axis)){for(let t of this.axis)if(!Number.isInteger(t))throw new Error(`Expected axis to be an array of integers, but received ${JSON.stringify(this.axis)}`)}else throw new Error(`Expected axis to be an integer or an array of integers, but received ${JSON.stringify(this.axis)}`);this.epsilon=e.epsilon==null?.001:e.epsilon,this.center=e.center==null?!0:e.center,this.scale=e.scale==null?!0:e.scale,this.betaInitializer=Nt(e.betaInitializer||"zeros"),this.gammaInitializer=Nt(e.gammaInitializer||"ones"),this.betaRegularizer=Et(e.betaRegularizer),this.gammaRegularizer=Et(e.gammaRegularizer),this.supportsMasking=!0}build(e){e=pt(e);let t=e.length;typeof this.axis=="number"&&(this.axis=[this.axis]);for(let r=0;r<this.axis.length;++r)this.axis[r]<0&&(this.axis[r]+=t);for(let r of this.axis)if(r<0||r>=t)throw new Error(`Invalid axis: ${r}`);if(this.axis.length!==Do(this.axis).length)throw new Error(`Found duplicate axes in: ${this.axis}`);let n=this.axis.map(r=>e[r]),s=!0;this.scale?this.gamma=this.addWeight("gamma",n,"float32",this.gammaInitializer,this.gammaRegularizer,s):this.gamma=null,this.center?this.beta=this.addWeight("beta",n,"float32",this.betaInitializer,this.betaRegularizer,s):this.beta=null,this.built=!0}call(e,t){let n=Be(e),s=n.shape,r=s.length;return q(()=>{let a=!0,{mean:o,variance:i}=gf(n,this.axis,a),l=dl(1,r);for(let f of this.axis)l[f]=s[f];let c=f=>f!=null&&f.shape.length!==r?G(f,l):f,u=c(this.gamma.read()),d=c(this.beta.read()),p=[],h=[];for(let f=0;f<r;++f)this.axis.indexOf(f)!==-1?(p.push(s[f]),h.push(1)):(p.push(1),h.push(s[f]));return o=js(o,p),i=js(i,p),u=js(u,h),d=js(d,h),Jd(n,o,i,d,u,this.epsilon)})}getConfig(){let e={axis:this.axis,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:Ot(this.betaInitializer),gammaInitializer:Ot(this.gammaInitializer),betaRegularizer:At(this.betaRegularizer),gammaRegularizer:At(this.gammaRegularizer)},t=super.getConfig();return Object.assign(e,t),e}};YA.className="LayerNormalization";ue.registerClass(YA);function cW(e,t,n){return q(()=>{if(e.rank!==4)throw new H(`temporalPadding expects input tensor to be 4-D, but received a ${e.rank}-D tensor.`);if(t==null&&(t=[[1,1],[1,1]]),t.length!==2||t[0].length!==2||t[1].length!==2)throw new H("spatial2dPadding expects `padding` to be an Array of two Arrays, each of which is an Array of two integers.");if(n==null&&(n=ur()),n!=="channelsLast"&&n!=="channelsFirst")throw new H(`Unknown data format: ${n}. Supported data formats are 'channelsLast' and 'channelsFirst.`);let s;return n==="channelsFirst"?s=[[0,0],[0,0],t[0],t[1]]:s=[[0,0],t[0],t[1],[0,0]],qs(e,s)})}var JA=class extends Qe{constructor(e){e==null&&(e={});super(e);if(this.dataFormat=e.dataFormat==null?ur():e.dataFormat,e.padding==null)this.padding=[[1,1],[1,1]];else if(typeof e.padding=="number")this.padding=[[e.padding,e.padding],[e.padding,e.padding]];else{if(e.padding=e.padding,e.padding.length!==2)throw new H(`ZeroPadding2D expects padding to be a length-2 array, but received a length-${e.padding.length} array.`);let t,n;if(typeof e.padding[0]=="number")t=[e.padding[0],e.padding[0]],n=[e.padding[1],e.padding[1]];else{if(e.padding=e.padding,e.padding[0].length!==2)throw new H(`ZeroPadding2D expects height padding to be a length-2 array, but received a length-${e.padding[0].length} array.`);if(t=e.padding[0],e.padding[1].length!==2)throw new H(`ZeroPadding2D expects width padding to be a length-2 array, but received a length-${e.padding[1].length} array.`);n=e.padding[1]}this.padding=[t,n]}this.inputSpec=[new Kt({ndim:4})]}computeOutputShape(e){e=pt(e);let t,n;return this.dataFormat==="channelsFirst"?(e[2]!=null&&e[2]>=0?t=e[2]+this.padding[0][0]+this.padding[0][1]:t=null,e[3]!=null&&e[3]>=0?n=e[3]+this.padding[1][0]+this.padding[1][1]:n=null,[e[0],e[1],t,n]):(e[1]!=null&&e[1]>=0?t=e[1]+this.padding[0][0]+this.padding[0][1]:t=null,e[2]!=null&&e[2]>=0?n=e[2]+this.padding[1][0]+this.padding[1][1]:n=null,[e[0],t,n,e[3]])}call(e,t){return q(()=>cW(Be(e),this.padding,this.dataFormat))}getConfig(){let e={padding:this.padding,dataFormat:this.dataFormat},t=super.getConfig();return Object.assign(e,t),e}};JA.className="ZeroPadding2D";ue.registerClass(JA);function pm(e,t,n,s,r,a){return q(()=>{Ht(r),pw(a),$s(s),n==null&&(n=[1,1]),s==null&&(s="valid"),r==null&&(r=ur()),a==null&&(a="max"),e=yA(e,r);let o,i=s==="same"?"same":"valid";return a==="max"?o=mf(e,t,n,i):o=of(e,t,n,i),r==="channelsFirst"&&(o=Ye(o,[0,3,1,2])),o})}function Nk(e,t,n,s,r,a){return q(()=>{Ht(r),pw(a),$s(s),n==null&&(n=[1,1,1]),s==null&&(s="valid"),r==null&&(r=ur()),a==null&&(a="max"),e=yk(e,r);let o,i=s==="same"?"same":"valid";return a==="max"?o=s1(e,t,n,i):o=U2(e,t,n,i),r==="channelsFirst"&&(o=Ye(o,[0,4,1,2,3])),o})}var Ek=class extends Qe{constructor(e){e.poolSize==null&&(e.poolSize=2);super(e);if(typeof e.poolSize=="number")this.poolSize=[e.poolSize];else if(Array.isArray(e.poolSize)&&e.poolSize.length===1&&typeof e.poolSize[0]=="number")this.poolSize=e.poolSize;else throw new H(`poolSize for 1D convolutional layer must be a number or an Array of a single number, but received ${JSON.stringify(e.poolSize)}`);if(yn(this.poolSize,"poolSize"),e.strides==null)this.strides=this.poolSize;else if(typeof e.strides=="number")this.strides=[e.strides];else if(Array.isArray(e.strides)&&e.strides.length===1&&typeof e.strides[0]=="number")this.strides=e.strides;else throw new H(`strides for 1D convolutional layer must be a number or an Array of a single number, but received ${JSON.stringify(e.strides)}`);yn(this.strides,"strides"),this.padding=e.padding==null?"valid":e.padding,$s(this.padding),this.inputSpec=[new Kt({ndim:3})]}computeOutputShape(e){e=pt(e);let t=mr(e[1],this.poolSize[0],this.padding,this.strides[0]);return[e[0],t,e[2]]}call(e,t){return q(()=>{this.invokeCallHook(e,t),e=Bd(Be(e),2);let n=this.poolingFunction(Be(e),[this.poolSize[0],1],[this.strides[0],1],this.padding,"channelsLast");return ot(n,[2])})}getConfig(){let e={poolSize:this.poolSize,padding:this.padding,strides:this.strides},t=super.getConfig();return Object.assign(e,t),e}},QA=class extends Ek{constructor(e){super(e)}poolingFunction(e,t,n,s,r){return Ht(r),$s(s),pm(e,t,n,s,r,"max")}};QA.className="MaxPooling1D";ue.registerClass(QA);var ey=class extends Ek{constructor(e){super(e)}poolingFunction(e,t,n,s,r){return Ht(r),$s(s),pm(e,t,n,s,r,"avg")}};ey.className="AveragePooling1D";ue.registerClass(ey);var Rk=class extends Qe{constructor(e){e.poolSize==null&&(e.poolSize=[2,2]);super(e);if(this.poolSize=Array.isArray(e.poolSize)?e.poolSize:[e.poolSize,e.poolSize],e.strides==null)this.strides=this.poolSize;else if(Array.isArray(e.strides)){if(e.strides.length!==2)throw new H(`If the strides property of a 2D pooling layer is an Array, it is expected to have a length of 2, but received length ${e.strides.length}.`);this.strides=e.strides}else this.strides=[e.strides,e.strides];yn(this.poolSize,"poolSize"),yn(this.strides,"strides"),this.padding=e.padding==null?"valid":e.padding,this.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,Ht(this.dataFormat),$s(this.padding),this.inputSpec=[new Kt({ndim:4})]}computeOutputShape(e){e=pt(e);let t=this.dataFormat==="channelsFirst"?e[2]:e[1],n=this.dataFormat==="channelsFirst"?e[3]:e[2];return t=mr(t,this.poolSize[0],this.padding,this.strides[0]),n=mr(n,this.poolSize[1],this.padding,this.strides[1]),this.dataFormat==="channelsFirst"?[e[0],e[1],t,n]:[e[0],t,n,e[3]]}call(e,t){return q(()=>(this.invokeCallHook(e,t),this.poolingFunction(Be(e),this.poolSize,this.strides,this.padding,this.dataFormat)))}getConfig(){let e={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},t=super.getConfig();return Object.assign(e,t),e}},ty=class extends Rk{constructor(e){super(e)}poolingFunction(e,t,n,s,r){return Ht(r),$s(s),pm(e,t,n,s,r,"max")}};ty.className="MaxPooling2D";ue.registerClass(ty);var ny=class extends Rk{constructor(e){super(e)}poolingFunction(e,t,n,s,r){return Ht(r),$s(s),pm(e,t,n,s,r,"avg")}};ny.className="AveragePooling2D";ue.registerClass(ny);var $k=class extends Qe{constructor(e){e.poolSize==null&&(e.poolSize=[2,2,2]);super(e);if(this.poolSize=Array.isArray(e.poolSize)?e.poolSize:[e.poolSize,e.poolSize,e.poolSize],e.strides==null)this.strides=this.poolSize;else if(Array.isArray(e.strides)){if(e.strides.length!==3)throw new H(`If the strides property of a 3D pooling layer is an Array, it is expected to have a length of 3, but received length ${e.strides.length}.`);this.strides=e.strides}else this.strides=[e.strides,e.strides,e.strides];yn(this.poolSize,"poolSize"),yn(this.strides,"strides"),this.padding=e.padding==null?"valid":e.padding,this.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,Ht(this.dataFormat),$s(this.padding),this.inputSpec=[new Kt({ndim:5})]}computeOutputShape(e){e=pt(e);let t=this.dataFormat==="channelsFirst"?e[2]:e[1],n=this.dataFormat==="channelsFirst"?e[3]:e[2],s=this.dataFormat==="channelsFirst"?e[4]:e[3];return t=mr(t,this.poolSize[0],this.padding,this.strides[0]),n=mr(n,this.poolSize[1],this.padding,this.strides[1]),s=mr(s,this.poolSize[2],this.padding,this.strides[2]),this.dataFormat==="channelsFirst"?[e[0],e[1],t,n,s]:[e[0],t,n,s,e[4]]}call(e,t){return q(()=>(this.invokeCallHook(e,t),this.poolingFunction(Be(e),this.poolSize,this.strides,this.padding,this.dataFormat)))}getConfig(){let e={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},t=super.getConfig();return Object.assign(e,t),e}},sy=class extends $k{constructor(e){super(e)}poolingFunction(e,t,n,s,r){return Ht(r),$s(s),Nk(e,t,n,s,r,"max")}};sy.className="MaxPooling3D";ue.registerClass(sy);var ry=class extends $k{constructor(e){super(e)}poolingFunction(e,t,n,s,r){return Ht(r),$s(s),Nk(e,t,n,s,r,"avg")}};ry.className="AveragePooling3D";ue.registerClass(ry);var _k=class extends Qe{constructor(e){super(e);this.inputSpec=[new Kt({ndim:3})]}computeOutputShape(e){return[e[0],e[2]]}call(e,t){throw new Le}},ay=class extends _k{constructor(e){super(e||{})}call(e,t){return q(()=>{let n=Be(e);return Bt(n,1)})}};ay.className="GlobalAveragePooling1D";ue.registerClass(ay);var oy=class extends _k{constructor(e){super(e||{})}call(e,t){return q(()=>{let n=Be(e);return gn(n,1)})}};oy.className="GlobalMaxPooling1D";ue.registerClass(oy);var Dk=class extends Qe{constructor(e){super(e);this.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,Ht(this.dataFormat),this.inputSpec=[new Kt({ndim:4})]}computeOutputShape(e){return e=e,this.dataFormat==="channelsLast"?[e[0],e[3]]:[e[0],e[1]]}call(e,t){throw new Le}getConfig(){let e={dataFormat:this.dataFormat},t=super.getConfig();return Object.assign(e,t),e}},iy=class extends Dk{call(e,t){return q(()=>{let n=Be(e);return this.dataFormat==="channelsLast"?Bt(n,[1,2]):Bt(n,[2,3])})}};iy.className="GlobalAveragePooling2D";ue.registerClass(iy);var ly=class extends Dk{call(e,t){return q(()=>{let n=Be(e);return this.dataFormat==="channelsLast"?gn(n,[1,2]):gn(n,[2,3])})}};ly.className="GlobalMaxPooling2D";ue.registerClass(ly);var Pk=class extends Qe{constructor(e){super(e);this.layer=e.layer}build(e){this.built=!0}get trainable(){return this.layer!=null?this.layer.trainable:!1}set trainable(e){this.layer!=null&&(this.layer.trainable=e)}get trainableWeights(){return this.layer.trainableWeights}get nonTrainableWeights(){return this.layer.nonTrainableWeights}get updates(){return this.layer._updates}get losses(){return this.layer.losses}getWeights(){return this.layer.getWeights()}setWeights(e){this.layer.setWeights(e)}getConfig(){let e={layer:{className:this.layer.getClassName(),config:this.layer.getConfig()}},t=super.getConfig();return Object.assign(e,t),e}setFastWeightInitDuringBuild(e){super.setFastWeightInitDuringBuild(e),this.layer!=null&&this.layer.setFastWeightInitDuringBuild(e)}static fromConfig(e,t,n={}){let s=t.layer,r=fr(s,n);delete t.layer;let a={layer:r};return Object.assign(a,t),new e(a)}},uy=class extends Pk{constructor(e){super(e);this.supportsMasking=!0}build(e){if(e=pt(e),e.length<3)throw new H(`TimeDistributed layer expects an input shape >= 3D, but received input shape ${JSON.stringify(e)}`);this.inputSpec=[{shape:e}];let t=[e[0]].concat(e.slice(2));this.layer.built||(this.layer.build(t),this.layer.built=!0),super.build(e)}computeOutputShape(e){e=pt(e);let t=[e[0]].concat(e.slice(2)),n=this.layer.computeOutputShape(t),s=e[1];return[n[0],s].concat(n.slice(1))}call(e,t){return q(()=>(e=Be(e),Ik((a,o)=>[Be(this.layer.call(a,t)),[]],e,[],!1,null,null,!1,!0)[1]))}};uy.className="TimeDistributed";ue.registerClass(uy);function dW(e){hl(yL,"BidirectionalMergeMode",e)}var pW="concat",cy=class extends Pk{constructor(e){super(e);let t=e.layer.getConfig(),n={};n.className=e.layer.getClassName(),n.config=t,this.forwardLayer=fr(n),t.goBackwards=t.goBackwards!==!0;let s={};if(s.className=e.layer.getClassName(),s.config=t,this.backwardLayer=fr(s),this.forwardLayer.name="forward_"+this.forwardLayer.name,this.backwardLayer.name="backward_"+this.backwardLayer.name,this.mergeMode=e.mergeMode===void 0?pW:e.mergeMode,dW(this.mergeMode),e.weights)throw new Le("weights support is not implemented for Bidirectional layer yet.");this._stateful=e.layer.stateful,this.returnSequences=e.layer.returnSequences,this.returnState=e.layer.returnState,this.supportsMasking=!0,this._trainable=!0,this.inputSpec=e.layer.inputSpec,this.numConstants=null}get trainable(){return this._trainable}set trainable(e){this._trainable=e,this.forwardLayer!=null&&(this.forwardLayer.trainable=e),this.backwardLayer!=null&&(this.backwardLayer.trainable=e)}getWeights(){return this.forwardLayer.getWeights().concat(this.backwardLayer.getWeights())}setWeights(e){let t=e.length,n=Math.floor(t/2);this.forwardLayer.setWeights(e.slice(0,n)),this.backwardLayer.setWeights(e.slice(n))}computeOutputShape(e){let t=this.forwardLayer.computeOutputShape(e);Array.isArray(t)&&Array.isArray(t[0])||(t=[t]),t=t;let n,s,r;return this.returnState&&(r=t.slice(1)),n=t[0],n=n,this.mergeMode==="concat"?(n[n.length-1]*=2,s=[n]):this.mergeMode==null?s=[n,n.slice()]:s=[n],this.returnState?this.mergeMode==null?s.concat(r).concat(r.slice()):[n].concat(r).concat(r.slice()):rs(s)}apply(e,t){let n=t==null?null:t.initialState,s=t==null?null:t.constants;t==null&&(t={});let r=Sk(e,n,s,this.numConstants);if(e=r.inputs,n=r.initialState,s=r.constants,Array.isArray(e)&&(n=e.slice(1),e=e[0]),(n==null||n.length===0)&&s==null)return super.apply(e,t);let a=[],o=[];if(n!=null){let l=n.length;if(l%2>0)throw new H("When passing `initialState` to a Bidrectional RNN, the state should be an Array containing the states of the underlying RNNs.");t.initialState=n,a.push(...n);let c=n.map(u=>new Kt({shape:u.shape}));this.forwardLayer.stateSpec=c.slice(0,l/2),this.backwardLayer.stateSpec=c.slice(l/2),o.push(...c)}if(s!=null)throw new Le("Support for constants in Bidirectional layers is not implemented yet.");let i=a[0]instanceof hr;for(let l of a)if(l instanceof hr!==i)throw new H("The initial state of a Bidirectional layer cannot be specified as a mix of symbolic and non-symbolic tensors");if(i){let l=[e].concat(a),c=this.inputSpec.concat(o),u=this.inputSpec;this.inputSpec=c;let d=super.apply(l,t);return this.inputSpec=u,d}else return super.apply(e,t)}call(e,t){return q(()=>{let n=t.initialState,s,r;if(n==null)s=this.forwardLayer.call(e,t),r=this.backwardLayer.call(e,t);else{let i=n.slice(0,n.length/2),l=n.slice(n.length/2);s=this.forwardLayer.call(e,Object.assign(t,{initialState:i})),r=this.backwardLayer.call(e,Object.assign(t,{initialState:l}))}let a;this.returnState&&(Array.isArray(s)&&(a=s.slice(1).concat(r.slice(1))),s=s[0],r=r[0]),this.returnSequences&&(r=Rs(r,1));let o;return this.mergeMode==="concat"?o=F1([s,r]):this.mergeMode==="sum"?o=le(s,r):this.mergeMode==="ave"?o=L(.5,le(s,r)):this.mergeMode==="mul"?o=L(s,r):this.mergeMode==null&&(o=[s,r]),this.returnState?this.mergeMode==null?o.concat(a):[o].concat(a):o})}resetStates(e){this.forwardLayer.resetStates(),this.backwardLayer.resetStates()}build(e){fl(this.forwardLayer.name,()=>{this.forwardLayer.build(e)}),fl(this.backwardLayer.name,()=>{this.backwardLayer.build(e)}),this.built=!0}computeMask(e,t){Array.isArray(t)&&(t=t[0]);let n;if(this.returnSequences?this.mergeMode==null?n=[t,t]:n=t:this.mergeMode==null?n=[null,null]:n=null,this.returnState){let r=this.forwardLayer.states.map(a=>null);return Array.isArray(n)?n.concat(r).concat(r):[n].concat(r).concat(r)}else return n}get trainableWeights(){return this.forwardLayer.trainableWeights.concat(this.backwardLayer.trainableWeights)}get nonTrainableWeights(){return this.forwardLayer.nonTrainableWeights.concat(this.backwardLayer.nonTrainableWeights)}setFastWeightInitDuringBuild(e){super.setFastWeightInitDuringBuild(e),this.forwardLayer!=null&&this.forwardLayer.setFastWeightInitDuringBuild(e),this.backwardLayer!=null&&this.backwardLayer.setFastWeightInitDuringBuild(e)}getConfig(){let e={mergeMode:this.mergeMode},t=super.getConfig();return Object.assign(e,t),e}static fromConfig(e,t){let n=fr(t.layer);if(delete t.layer,t.numConstants!=null)throw new Le("Deserialization of a Bidirectional layer with numConstants present is not supported yet.");let s=t;return s.layer=n,new e(s)}};cy.className="Bidirectional";ue.registerClass(cy);function hW(e){return new ju(e)}function fW(e){return new mA(e)}function mW(e){return new pA(e)}function gW(e){return new hA(e)}function AW(e){return new fA(e)}function yW(e){return new AA(e)}function xW(e){return new gA(e)}function bW(e){return new kA(e)}function vW(e){return new am(e)}function wW(e){return new bA(e)}function kW(e){return new om(e)}function SW(e){return new vA(e)}function IW(e){return new wA(e)}function CW(e){return new SA(e)}function TW(e){return new IA(e)}function NW(e){return new CA(e)}function EW(e){return new PA(e)}function RW(e){return new _A(e)}function $W(e){return new dm(e)}function _W(e){return new $A(e)}function DW(e){return new DA(e)}function PW(e){return new FA(e)}function FW(e){return new OA(e)}function OW(e){return new MA(e)}function MW(e){return new LA(e)}function zW(e){return new BA(e)}function LW(e){return new VA(e)}function BW(e){return new HA(e)}function WW(e){return new UA(e)}function VW(e){return new GA(e)}function UW(e){return new WA(e)}function GW(e){return new jA(e)}function HW(e){return new ZA(e)}function jW(e){return new YA(e)}function qW(e){return new JA(e)}function dy(e){return new ey(e)}function XW(e){return dy(e)}function KW(e){return dy(e)}function py(e){return new ny(e)}function ZW(e){return py(e)}function YW(e){return py(e)}function hy(e){return new ry(e)}function JW(e){return hy(e)}function QW(e){return hy(e)}function eV(e){return new ay(e)}function tV(e){return new iy(e)}function Fk(e){return new oy(e)}function Ok(e){return new ly(e)}function Mk(e){return new QA(e)}function zk(e){return new ty(e)}function nV(e){return new sy(e)}function sV(e){return new NA(e)}function rV(e){return new lm(e)}function aV(e){return new EA(e)}function oV(e){return new Zd(e)}function iV(e){return new TA(e)}function lV(e){return new im(e)}function uV(e){return new RA(e)}function cV(e){return new cm(e)}function dV(e){return new ta(e)}function pV(e){return new um(e)}function hV(e){return new cy(e)}function fV(e){return new uy(e)}var mV=Fk,gV=Ok,AV=Mk,yV=zk;function xV(e){return new qA(e)}function bV(e){return new XA(e)}function vV(e){return new KA(e)}function wV(e){return new zA(e)}var Lk={};Oe(Lk,{MAPE:()=>DV,MSE:()=>OV,binaryAccuracy:()=>kV,binaryCrossentropy:()=>SV,categoricalAccuracy:()=>CV,categoricalCrossentropy:()=>TV,cosineProximity:()=>RV,mape:()=>PV,meanAbsoluteError:()=>$V,meanAbsolutePercentageError:()=>_V,meanSquaredError:()=>FV,mse:()=>MV,precision:()=>NV,recall:()=>EV,sparseCategoricalAccuracy:()=>IV});function kV(e,t){return Y1(e,t)}function SV(e,t){return Mw(e,t)}function IV(e,t){return zw(e,t)}function CV(e,t){return J1(e,t)}function TV(e,t){return Q1(e,t)}function NV(e,t){return Ow(e,t)}function EV(e,t){return hB(e,t)}function RV(e,t){return K1(e,t)}function $V(e,t){return Yf(e,t)}function _V(e,t){return Xu(e,t)}function DV(e,t){return Xu(e,t)}function PV(e,t){return Xu(e,t)}function FV(e,t){return gl(e,t)}function OV(e,t){return gl(e,t)}function MV(e,t){return gl(e,t)}var Bk={};Oe(Bk,{modelFromJSON:()=>qB});var Wk={};Oe(Wk,{l1:()=>LV,l1l2:()=>zV,l2:()=>BV});function zV(e){return new qd(e)}function LV(e){return tW(e)}function BV(e){return nW(e)}var Vk=class extends qu{constructor(){super(...arguments);this.model=null}setModel(e){if(!(e instanceof ea))throw new Error("model must be a LayersModel, not some other Container");this.model=e}};function hm(e,t){return e<t}function Uk(e,t){return e>t}var Gk=class extends Vk{constructor(e){super();if(e==null&&(e={}),e.restoreBestWeights)throw new Le("restoreBestWeights = True is not implemented in EarlyStopping yet.");this.monitor=e.monitor||"val_loss",this.minDelta=Math.abs(e.minDelta||0),this.patience=e.patience||0,this.verbose=e.verbose||0,this.mode=e.mode||"auto",this.baseline=e.baseline,["auto","min","max"].indexOf(this.mode)===-1&&(console.warn(`EarlyStopping mode '${this.mode}' is invalid. Falling back to mode 'auto'.`),this.mode="auto"),this.mode==="min"?this.monitorFunc=hm:this.mode==="max"?this.monitorFunc=Uk:this.monitor.indexOf("acc")!==-1?this.monitorFunc=Uk:this.monitorFunc=hm,this.monitorFunc===hm&&(this.minDelta*=-1)}async onTrainBegin(e){this.wait=0,this.stoppedEpoch=0,this.baseline!=null?this.best=this.baseline:this.best=this.monitorFunc===hm?1/0:-1/0}async onEpochEnd(e,t){await Oo(t);let n=this.getMonitorValue(t);n!=null&&(this.monitorFunc(n-this.minDelta,this.best)?(this.best=n,this.wait=0):(this.wait++,this.wait>=this.patience&&(this.stoppedEpoch=e,this.model.stopTraining=!0)))}async onTrainEnd(e){this.stoppedEpoch>0&&this.verbose&&console.log(`Epoch ${this.stoppedEpoch}: early stopping.`)}getMonitorValue(e){e==null&&(e={});let t=e[this.monitor];return t==null&&console.warn(`Metric for EarlyStopping ${this.monitor} is not available. Available metrics are: ${Object.keys(e)}`),t}};function WV(e){return new Gk(e)}var VV={earlyStopping:WV},UV=K();UV.registerFlag("KEEP_INTERMEDIATE_TENSORS",()=>!1,e=>{e&&console.warn("Keep intermediate tensors is ON. This will print the values of all intermediate tensors during model inference. Not all models support this mode. For details, check e2e/benchmarks/ model_config.js. This significantly impacts performance.")});var gr;(function(e){e[e.DT_INVALID=0]="DT_INVALID",e[e.DT_FLOAT=1]="DT_FLOAT",e[e.DT_DOUBLE=2]="DT_DOUBLE",e[e.DT_INT32=3]="DT_INT32",e[e.DT_UINT8=4]="DT_UINT8",e[e.DT_INT16=5]="DT_INT16",e[e.DT_INT8=6]="DT_INT8",e[e.DT_STRING=7]="DT_STRING",e[e.DT_COMPLEX64=8]="DT_COMPLEX64",e[e.DT_INT64=9]="DT_INT64",e[e.DT_BOOL=10]="DT_BOOL",e[e.DT_QINT8=11]="DT_QINT8",e[e.DT_QUINT8=12]="DT_QUINT8",e[e.DT_QINT32=13]="DT_QINT32",e[e.DT_BFLOAT16=14]="DT_BFLOAT16",e[e.DT_FLOAT_REF=101]="DT_FLOAT_REF",e[e.DT_DOUBLE_REF=102]="DT_DOUBLE_REF",e[e.DT_INT32_REF=103]="DT_INT32_REF",e[e.DT_UINT8_REF=104]="DT_UINT8_REF",e[e.DT_INT16_REF=105]="DT_INT16_REF",e[e.DT_INT8_REF=106]="DT_INT8_REF",e[e.DT_STRING_REF=107]="DT_STRING_REF",e[e.DT_COMPLEX64_REF=108]="DT_COMPLEX64_REF",e[e.DT_INT64_REF=109]="DT_INT64_REF",e[e.DT_BOOL_REF=110]="DT_BOOL_REF",e[e.DT_QINT8_REF=111]="DT_QINT8_REF",e[e.DT_QUINT8_REF=112]="DT_QUINT8_REF",e[e.DT_QINT32_REF=113]="DT_QINT32_REF",e[e.DT_BFLOAT16_REF=114]="DT_BFLOAT16_REF"})(gr||(gr={}));var Hk;(function(e){let t;(function(n){n[n.LEGACY=0]="LEGACY",n[n.V1=1]="V1",n[n.V2=2]="V2"})(t=e.CheckpointFormatVersion||(e.CheckpointFormatVersion={}))})(Hk||(Hk={}));var fy={};function GV(e,t){let n={tfOpName:e,category:"custom",inputs:[],attrs:[],customExecutor:t};fy[e]=n}function jk(e){return fy[e]}function HV(e){delete fy[e]}function S(e,t,n,s,r){let a=t.inputParams[e];if(a&&a.inputIndexStart!==void 0){let i=a.inputIndexStart,l=a.inputIndexEnd===0?void 0:a.inputIndexEnd===void 0?i+1:a.inputIndexEnd;if(a.type==="tensor")return Vn(t.inputNames[a.inputIndexStart],n,s,r);if(a.type==="tensors")return t.inputNames.slice(i,l).map(p=>Vn(p,n,s,r));let c=Vn(t.inputNames.slice(i)[0],n,s,r),u=c.dataSync();return a.type==="number"?u[0]:v.toNestedArray(c.shape,u)}let o=t.attrParams[e];return o&&o.value}function Vn(e,t,n,s){let[r,a]=fs(e);if(s!=null){let i=s.getHashTableHandleByName(r);if(i!=null)return i}let o=n.currentContextIds.find(i=>!!t[fm(r,i)]);return o!==void 0?t[fm(r,o)][a]:void 0}function jV(e,t,n){return t[fm(e,n.currentContextId)]}function Or(e,t){let[n,s,r]=fs(e);return[fm(n,t&&t.currentContextId),s,r]}function fm(e,t){return t?`${e}-${t}`:e}function fs(e){let t=e.split(":");if(t.length===1)return[e,0,void 0];let n=t[0],s=t.length===3?t[1]:void 0,r=Number(t[t.length-1]);return[n,r,s]}function mm(e,t,n){let s=S("pad",e,t,n);if(s==="explicit"){s=S("explicitPaddings",e,t,n);let r=[[0,0],[0,0],[0,0],[0,0]];for(let a=0;a<4;a++)r[a][0]=s[a*2],r[a][1]=s[a*2+1];return r}return s}function na(e){return e.kept?e:Ln(e)}var qk={};Oe(qk,{json:()=>qV});var qV=[{tfOpName:"Add",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AddV2",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AddN",category:"arithmetic",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}]},{tfOpName:"BiasAdd",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"Sub",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"RealDiv",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Div",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"DivNoNan",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"FloorDiv",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Mul",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Maximum",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Minimum",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Pow",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"SquaredDifference",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Mod",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"FloorMod",category:"arithmetic",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],Xk={};Oe(Xk,{json:()=>XV});var XV=[{tfOpName:"Abs",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Acos",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Asin",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atan",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atan2",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"y",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Ceil",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ClipByValue",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"clipValueMin",type:"number"},{start:2,name:"clipValueMax",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Complex",category:"basic_math",inputs:[{start:0,name:"real",type:"tensor"},{start:1,name:"imag",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ComplexAbs",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Cos",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Cosh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Elu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Exp",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Floor",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Log",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Imag",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"Tout",name:"outputType",type:"dtype",notSupported:!0}]},{tfOpName:"Neg",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Real",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"Tout",name:"outputType",type:"dtype",notSupported:!0}]},{tfOpName:"Prelu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"alpha",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Relu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Relu6",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Selu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sigmoid",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sin",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sinh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sqrt",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Rsqrt",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Square",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Tan",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Tanh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sign",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Round",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Expm1",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Log1p",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Reciprocal",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Softplus",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Asinh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Acosh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atanh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Erf",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Prod",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axes",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool",notSupported:!0},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LeakyRelu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"alpha",name:"alpha",type:"number",defaultValue:.2},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"IsNan",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],Kk={};Oe(Kk,{json:()=>KV});var KV=[{tfOpName:"EmptyTensorList",category:"control",inputs:[{start:0,name:"elementShape",type:"shape"},{start:1,name:"maxNumElements",type:"number"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"LoopCond",category:"control",inputs:[{start:0,name:"pred",type:"tensor"}]},{tfOpName:"Switch",category:"control",inputs:[{start:0,name:"data",type:"tensor"},{start:1,name:"pred",type:"tensor"}]},{tfOpName:"Merge",category:"control",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}]},{tfOpName:"Enter",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"frame_name",name:"frameName",type:"string"},{tfName:"is_constant",name:"isConstant",type:"bool"}]},{tfOpName:"Exit",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"NextIteration",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayV3",category:"control",inputs:[{start:0,name:"size",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape",name:"elementShape",type:"shape"},{tfName:"dynamic_size",name:"dynamicSize",type:"bool"},{tfName:"clear_after_read",name:"clearAfterRead",type:"bool"},{tfName:"identical_element_shapes",name:"identicalElementShapes",type:"bool"},{tfName:"tensor_array_name",name:"name",type:"string"}]},{tfOpName:"TensorArrayWriteV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"tensor",type:"tensor"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayReadV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"TensorArrayGatherV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape",name:"elementShape",type:"shape"}]},{tfOpName:"TensorArrayScatterV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"tensor",type:"tensor"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"TensorArrayConcatV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"flowIn",type:"number"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"element_shape_except0",name:"elementShapeExcept0",type:"shape",notSupported:!0}]},{tfOpName:"TensorArraySplitV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"tensor",type:"tensor"},{start:2,name:"lengths",type:"number[]"},{start:3,name:"flowIn",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"TensorArraySizeV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"},{start:1,name:"flowIn",type:"number"}]},{tfOpName:"TensorArrayCloseV3",category:"control",inputs:[{start:0,name:"tensorArrayId",type:"tensor"}]},{tfOpName:"StatelessIf",category:"control",inputs:[{start:0,name:"cond",type:"tensor"},{start:1,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"then_branch",name:"thenBranch",type:"func"},{tfName:"else_branch",name:"elseBranch",type:"func"}]},{tfOpName:"If",category:"control",inputs:[{start:0,name:"cond",type:"tensor"},{start:1,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"then_branch",name:"thenBranch",type:"func"},{tfName:"else_branch",name:"elseBranch",type:"func"}]},{tfOpName:"StatelessWhile",category:"control",inputs:[{start:0,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"cond",name:"cond",type:"func"},{tfName:"body",name:"body",type:"func"}]},{tfOpName:"While",category:"control",inputs:[{start:0,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"cond",name:"cond",type:"func"},{tfName:"body",name:"body",type:"func"}]},{tfOpName:"TensorListScatter",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListScatterV2",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"},{start:3,name:"numElements",type:"number"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListGather",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"indices",type:"number[]"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListGetItem",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListSetItem",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"index",type:"number"},{start:2,name:"tensor",type:"tensor"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListReserve",category:"control",inputs:[{start:0,name:"elementShape",type:"shape"},{start:1,name:"numElements",type:"number"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListFromTensor",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListStack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"},{tfName:"num_elements",name:"numElements",type:"dtype"}]},{tfOpName:"TensorListSplit",category:"control",inputs:[{start:0,name:"tensor",type:"tensor"},{start:1,name:"elementShape",type:"shape"},{start:2,name:"lengths",type:"number[]"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListConcat",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"}],attrs:[{tfName:"element_shape",name:"elementShape",type:"shape"},{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListPopBack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"elementShape",type:"shape"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]},{tfOpName:"TensorListPushBack",category:"control",inputs:[{start:0,name:"tensorListId",type:"tensor"},{start:1,name:"tensor",type:"tensor"}],attrs:[{tfName:"element_dtype",name:"elementDType",type:"dtype"}]}],Zk={};Oe(Zk,{json:()=>ZV});var ZV=[{tfOpName:"AvgPool",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPool",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[],notSupported:!0},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPoolWithArgmax",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"include_batch_in_index",name:"includeBatchInIndex",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"AvgPool3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MaxPool3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"ksize",name:"kernelSize",type:"number[]"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Conv1D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"stride",name:"stride",type:"number"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NWC"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"dilation",name:"dilation",type:"number",defaultValue:1}]},{tfOpName:"Conv2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"useCudnnOnGpu",name:"useCudnnOnGpu",type:"bool"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"_FusedConv2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"use_cudnn_on_gpu",name:"useCudnnOnGpu",type:"bool",defaultValue:!0},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]",defaultValue:[1,1,1,1]},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]},{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:1e-4},{tfName:"leakyrelu_alpha",name:"leakyreluAlpha",type:"number"}]},{tfOpName:"Conv2DBackpropInput",category:"convolution",inputs:[{start:2,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:0,name:"outputShape",type:"number[]"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]",notSupported:!0}]},{tfOpName:"DepthwiseConv2d",category:"convolution",inputs:[{start:0,name:"input",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"DepthwiseConv2dNative",category:"convolution",inputs:[{start:0,name:"input",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"FusedDepthwiseConv2dNative",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]",defaultValue:[1,1,1,1]},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]},{tfName:"explicit_paddings",name:"explicitPaddings",type:"number[]",defaultValue:[]}]},{tfOpName:"Conv3D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"padding",name:"pad",type:"string"},{tfName:"data_format",name:"dataFormat",type:"string",defaultValue:"NHWC"},{tfName:"dilations",name:"dilations",type:"number[]"}]},{tfOpName:"Dilation2D",category:"convolution",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"filter",type:"tensor"}],attrs:[{tfName:"strides",name:"strides",type:"number[]"},{tfName:"rates",name:"dilations",type:"number[]"},{tfName:"padding",name:"pad",type:"string"}]}],Yk={};Oe(Yk,{json:()=>YV});var YV=[{tfOpName:"Fill",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"},{start:1,name:"value",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"LinSpace",category:"creation",inputs:[{start:0,name:"start",type:"number"},{start:1,name:"stop",type:"number"},{start:2,name:"num",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"OneHot",category:"creation",inputs:[{start:0,name:"indices",type:"tensor"},{start:1,name:"depth",type:"number"},{start:2,name:"onValue",type:"number",defaultValue:1},{start:3,name:"offValue",type:"number",defaultValue:0}],attrs:[{tfName:"axis",name:"axis",type:"number",notSupported:!0},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Ones",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"OnesLike",category:"creation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"RandomUniform",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"minval",name:"minval",type:"number",defaultValue:0},{tfName:"maxval",name:"maxval",type:"number",defaultValue:1},{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"seed",name:"seed",type:"number",defaultValue:0},{tfName:"seed2",name:"seed2",type:"number",defaultValue:0,notSupported:!0},{tfName:"T",name:"T",type:"number",notSupported:!0}]},{tfOpName:"Range",category:"creation",inputs:[{start:0,name:"start",type:"number"},{start:1,name:"stop",type:"number"},{start:2,name:"step",type:"number",defaultValue:0}],attrs:[{tfName:"Tidx",name:"dtype",type:"dtype"}]},{tfOpName:"TruncatedNormal",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"means",name:"mean",type:"number",defaultValue:0},{tfName:"stddev",name:"stdDev",type:"number",defaultValue:1},{tfName:"seed",name:"seed",type:"number"},{tfName:"seed2",name:"seed2",type:"number",defaultValue:0,notSupported:!0},{tfName:"dtype",name:"dtype",type:"dtype"},{tfName:"T",name:"T",type:"number",notSupported:!0}]},{tfOpName:"Zeros",category:"creation",inputs:[{start:0,name:"shape",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"ZerosLike",category:"creation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype"}]},{tfOpName:"Multinomial",category:"creation",inputs:[{start:0,name:"logits",type:"tensor"},{start:1,name:"numSamples",type:"number"}],attrs:[{tfName:"seed",name:"seed",type:"number"},{tfName:"seed2",name:"seed2",type:"number"},{tfName:"T",name:"dtype",type:"dtype"},{tfName:"output_dtype",name:"output_dtype",type:"dtype"}]}],Jk={};Oe(Jk,{json:()=>JV});var JV=[{tfOpName:"NonMaxSuppressionV2",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"}]},{tfOpName:"NonMaxSuppressionV3",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"}]},{tfOpName:"NonMaxSuppressionV4",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"T_threshold",name:"threshold",type:"dtype",notSupported:!0},{tfName:"pad_to_max_output_size",name:"padToMaxOutputSize",type:"bool"}]},{tfOpName:"NonMaxSuppressionV5",category:"dynamic",inputs:[{start:0,name:"boxes",type:"tensor"},{start:1,name:"scores",type:"tensor"},{start:2,name:"maxOutputSize",type:"number"},{start:3,name:"iouThreshold",type:"number"},{start:4,name:"scoreThreshold",type:"number"},{start:5,name:"softNmsSigma",type:"number"}]},{tfOpName:"Where",category:"dynamic",inputs:[{start:0,name:"condition",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ListDiff",category:"dynamic",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"y",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],Qk={};Oe(Qk,{json:()=>QV});var QV=[{tfOpName:"TopKV2",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"k",type:"number"}],attrs:[{tfName:"sorted",name:"sorted",type:"bool"}]},{tfOpName:"Unique",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"UniqueV2",category:"evaluation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]}],e7={};Oe(e7,{json:()=>eU});var eU=[{tfOpName:"PlaceholderWithDefault",category:"graph",inputs:[{start:0,name:"default",type:"tensor"}],attrs:[{tfName:"shape",name:"shape",type:"shape"},{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"Placeholder",category:"graph",attrs:[{tfName:"shape",name:"shape",type:"shape"},{tfName:"dtype",name:"dtype",type:"dtype"}]},{tfOpName:"Const",category:"graph"},{tfOpName:"Identity",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"IdentityN",category:"graph",inputs:[{start:0,end:0,name:"x",type:"tensors"}]},{tfOpName:"Snapshot",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Rank",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Size",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"Shape",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"ShapeN",category:"graph",inputs:[{start:0,end:0,name:"x",type:"tensors"}]},{tfOpName:"Print",category:"graph",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"data",type:"tensors"}],attrs:[{tfName:"message",name:"message",type:"string"},{tfName:"first_n",name:"firstN",type:"number",notSupported:!0},{tfName:"summarize",name:"summarize",type:"number",defaultValue:3}]},{tfOpName:"NoOp",category:"graph",inputs:[]},{tfOpName:"StopGradient",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"FakeQuantWithMinMaxVars",category:"graph",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"min",name:"min",type:"number"},{tfName:"max",name:"max",type:"number"}]}],t7={};Oe(t7,{json:()=>tU});var tU=[{tfOpName:"HashTable",category:"hash_table",inputs:[],attrs:[{tfName:"shared_name",name:"sharedName",type:"string"},{tfName:"use_node_name_sharing",name:"useNodeNameSharing",type:"bool"},{tfName:"key_dtype",name:"keyDType",type:"dtype"},{tfName:"value_dtype",name:"valueDType",type:"dtype"}]},{tfOpName:"HashTableV2",category:"hash_table",inputs:[],attrs:[{tfName:"shared_name",name:"sharedName",type:"string"},{tfName:"use_node_name_sharing",name:"useNodeNameSharing",type:"bool"},{tfName:"key_dtype",name:"keyDType",type:"dtype"},{tfName:"value_dtype",name:"valueDType",type:"dtype"}]},{tfOpName:"LookupTableImport",category:"hash_table",inputs:[{start:0,name:"tableHandle",type:"tensor"},{start:1,name:"keys",type:"tensor"},{start:2,name:"values",type:"tensor"}],attrs:[{tfName:"Tin",name:"tIn",type:"dtype",notSupported:!0},{tfName:"Tout",name:"tOut",type:"dtype",notSupported:!0}]},{tfOpName:"LookupTableImportV2",category:"hash_table",inputs:[{start:0,name:"tableHandle",type:"tensor"},{start:1,name:"keys",type:"tensor"},{start:2,name:"values",type:"tensor"}],attrs:[{tfName:"Tin",name:"tIn",type:"dtype",notSupported:!0},{tfName:"Tout",name:"tOut",type:"dtype",notSupported:!0}]},{tfOpName:"LookupTableFind",category:"hash_table",inputs:[{start:0,name:"tableHandle",type:"tensor"},{start:1,name:"keys",type:"tensor"},{start:2,name:"defaultValue",type:"tensor"}],attrs:[{tfName:"Tin",name:"tIn",type:"dtype",notSupported:!0},{tfName:"Tout",name:"tOut",type:"dtype",notSupported:!0}]},{tfOpName:"LookupTableFindV2",category:"hash_table",inputs:[{start:0,name:"tableHandle",type:"tensor"},{start:1,name:"keys",type:"tensor"},{start:2,name:"defaultValue",type:"tensor"}],attrs:[{tfName:"Tin",name:"tIn",type:"dtype",notSupported:!0},{tfName:"Tout",name:"tOut",type:"dtype",notSupported:!0}]},{tfOpName:"LookupTableSize",category:"hash_table",inputs:[{start:0,name:"tableHandle",type:"tensor"}]},{tfOpName:"LookupTableSizeV2",category:"hash_table",inputs:[{start:0,name:"tableHandle",type:"tensor"}]}],n7={};Oe(n7,{json:()=>nU});var nU=[{tfOpName:"ResizeBilinear",category:"image",inputs:[{start:0,name:"images",type:"tensor"},{start:1,name:"size",type:"number[]"}],attrs:[{tfName:"align_corners",name:"alignCorners",type:"bool"},{tfName:"half_pixel_centers",name:"halfPixelCenters",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ResizeNearestNeighbor",category:"image",inputs:[{start:0,name:"images",type:"tensor"},{start:1,name:"size",type:"number[]"}],attrs:[{tfName:"align_corners",name:"alignCorners",type:"bool"},{tfName:"half_pixel_centers",name:"halfPixelCenters",type:"bool"},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"CropAndResize",category:"image",inputs:[{start:0,name:"image",type:"tensor"},{start:1,name:"boxes",type:"tensor"},{start:2,name:"boxInd",type:"tensor"},{start:3,name:"cropSize",type:"number[]"}],attrs:[{tfName:"method",name:"method",type:"string"},{tfName:"extrapolation_value",name:"extrapolationValue",type:"number"}]}],s7={};Oe(s7,{json:()=>sU});var sU=[{tfOpName:"Equal",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"NotEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Greater",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"GreaterEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Less",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LessEqual",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalAnd",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalNot",category:"logical",inputs:[{start:0,name:"a",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LogicalOr",category:"logical",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Select",category:"logical",inputs:[{start:0,name:"condition",type:"tensor"},{start:1,name:"a",type:"tensor"},{start:2,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"SelectV2",category:"logical",inputs:[{start:0,name:"condition",type:"tensor"},{start:1,name:"a",type:"tensor"},{start:2,name:"b",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}],r7={};Oe(r7,{json:()=>rU});var rU=[{tfOpName:"_FusedMatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]},{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:1e-4},{tfName:"transpose_a",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"transpose_b",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"transpose_a",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"transpose_b",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"BatchMatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"adj_x",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"adj_y",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"BatchMatMulV2",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"adj_x",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"adj_y",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Transpose",category:"matrices",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"perm",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Einsum",category:"matrices",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}],attrs:[{tfName:"equation",name:"equation",type:"string"},{tfName:"N",name:"n",type:"number",defaultValue:2},{tfName:"T",name:"dtype",type:"dtype"}]}],a7={};Oe(a7,{json:()=>aU});var aU=[{tfOpName:"FusedBatchNorm",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"FusedBatchNormV2",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"FusedBatchNormV3",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"scale",type:"tensor"},{start:2,name:"offset",type:"tensor"},{start:3,name:"mean",type:"tensor"},{start:4,name:"variance",type:"tensor"}],attrs:[{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:.001},{tfName:"data_format",name:"dataFormat",type:"string",notSupported:!0}]},{tfOpName:"LRN",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"depth_radius",name:"radius",type:"number",defaultValue:5},{tfName:"bias",name:"bias",type:"number",defaultValue:1},{tfName:"alpha",name:"alpha",type:"number",defaultValue:1},{tfName:"beta",name:"beta",type:"number",defaultValue:.5}]},{tfOpName:"Softmax",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"LogSoftmax",category:"normalization",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"SparseToDense",category:"normalization",inputs:[{start:0,name:"sparseIndices",type:"tensor"},{start:1,name:"outputShape",type:"number[]"},{start:2,name:"sparseValues",type:"tensor"},{start:3,name:"defaultValue",type:"tensor"}],attrs:[{tfName:"validate_indices",name:"validateIndices",type:"bool",defaultValue:!0,notSupported:!0}]}],o7={};Oe(o7,{json:()=>oU});var oU=[{tfOpName:"Bincount",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"size",type:"number"},{start:2,name:"weights",type:"tensor"}]},{tfOpName:"DenseBincount",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"size",type:"number"},{start:2,name:"weights",type:"tensor"}],attrs:[{tfName:"binary_output",name:"binaryOutput",type:"bool"}]},{tfOpName:"Max",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Mean",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Min",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Sum",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"All",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Any",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"ArgMax",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"ArgMin",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"Prod",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool"}]},{tfOpName:"Cumsum",category:"reduction",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}],attrs:[{tfName:"exclusive",name:"exclusive",type:"bool"},{tfName:"reverse",name:"reverse",type:"bool"}]}],i7={};Oe(i7,{json:()=>iU});var iU=[{tfOpName:"ConcatV2",category:"slice_join",inputs:[{start:0,end:-1,name:"tensors",type:"tensors"},{start:-1,name:"axis",type:"number"}],attrs:[{tfName:"N",name:"n",type:"number",defaultValue:2}]},{tfOpName:"Concat",category:"slice_join",inputs:[{start:1,end:0,name:"tensors",type:"tensors"},{start:0,name:"axis",type:"number"}],attrs:[{tfName:"N",name:"n",type:"number",defaultValue:2}]},{tfOpName:"GatherV2",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"},{start:2,name:"axis",type:"number",defaultValue:0}],attrs:[{tfName:"batch_dims",name:"batchDims",type:"number",defaultValue:0}]},{tfOpName:"Gather",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"}],attrs:[{tfName:"validate_indices",name:"validateIndices",type:"bool",notSupported:!0}]},{tfOpName:"Reverse",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"dims",type:"bool[]"}]},{tfOpName:"ReverseV2",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number[]"}]},{tfOpName:"Slice",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"begin",type:"number[]"},{start:2,name:"size",type:"number[]"}]},{tfOpName:"StridedSlice",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"begin",type:"number[]"},{start:2,name:"end",type:"number[]"},{start:3,name:"strides",type:"number[]"}],attrs:[{tfName:"begin_mask",name:"beginMask",type:"number",defaultValue:0},{tfName:"end_mask",name:"endMask",type:"number",defaultValue:0},{tfName:"new_axis_mask",name:"newAxisMask",type:"number",defaultValue:0},{tfName:"ellipsis_mask",name:"ellipsisMask",type:"number",defaultValue:0},{tfName:"shrink_axis_mask",name:"shrinkAxisMask",type:"number",defaultValue:0}]},{tfOpName:"Pack",category:"slice_join",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0}]},{tfOpName:"Unpack",category:"slice_join",inputs:[{start:0,name:"tensor",type:"tensor"}],attrs:[{tfName:"axis",name:"axis",type:"number",defaultValue:0},{tfName:"num",name:"num",type:"number",defaultValue:0,notSupported:!0}]},{tfOpName:"Tile",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"reps",type:"number[]"}]},{tfOpName:"Split",category:"slice_join",inputs:[{start:0,name:"axis",type:"number",defaultValue:0},{start:1,name:"x",type:"tensor"}],attrs:[{tfName:"num_split",name:"numOrSizeSplits",type:"number",defaultValue:1}]},{tfOpName:"SplitV",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"numOrSizeSplits",type:"number[]"},{start:2,name:"axis",type:"number",defaultValue:0}]},{tfOpName:"ScatterNd",category:"slice_join",inputs:[{start:0,name:"indices",type:"tensor"},{start:1,name:"values",type:"tensor"},{start:2,name:"shape",type:"number[]"}]},{tfOpName:"GatherNd",category:"slice_join",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"indices",type:"tensor"}]},{tfOpName:"SparseToDense",category:"slice_join",inputs:[{start:0,name:"sparseIndices",type:"tensor"},{start:1,name:"outputShape",type:"number[]"},{start:2,name:"sparseValues",type:"tensor"},{start:3,name:"defaultValue",type:"tensor"}],attrs:[{tfName:"validate_indices",name:"validateIndices",type:"bool",defaultValue:!1,notSupported:!0}]}],l7={};Oe(l7,{json:()=>lU});var lU=[{tfOpName:"SparseFillEmptyRows",category:"sparse",inputs:[{start:0,name:"indices",type:"tensor"},{start:1,name:"values",type:"tensor"},{start:2,name:"denseShape",type:"tensor"},{start:3,name:"defaultValue",type:"tensor"}]},{tfOpName:"SparseReshape",category:"sparse",inputs:[{start:0,name:"inputIndices",type:"tensor"},{start:1,name:"inputShape",type:"tensor"},{start:2,name:"newShape",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"SparseSegmentMean",category:"sparse",inputs:[{start:0,name:"data",type:"tensor"},{start:1,name:"indices",type:"tensor"},{start:2,name:"segmentIds",type:"tensor"}]},{tfOpName:"SparseSegmentSum",category:"sparse",inputs:[{start:0,name:"data",type:"tensor"},{start:1,name:"indices",type:"tensor"},{start:2,name:"segmentIds",type:"tensor"}]}],u7={};Oe(u7,{json:()=>uU});var uU=[{tfOpName:"FFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"IFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"}]},{tfOpName:"RFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"fft_length",type:"number",notSupported:!0}]},{tfOpName:"IRFFT",category:"spectral",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"fft_length",type:"number",notSupported:!0}]}],c7={};Oe(c7,{json:()=>cU});var cU=[{tfOpName:"StringNGrams",category:"string",inputs:[{start:0,name:"data",type:"tensor"},{start:1,name:"dataSplits",type:"tensor"}],attrs:[{tfName:"separator",name:"separator",type:"string"},{tfName:"ngram_widths",name:"nGramWidths",type:"number[]"},{tfName:"left_pad",name:"leftPad",type:"string"},{tfName:"right_pad",name:"rightPad",type:"string"},{tfName:"pad_width",name:"padWidth",type:"number"},{tfName:"preserve_short_sequences",name:"preserveShortSequences",type:"bool"}],outputs:["ngrams","ngrams_splits"]},{tfOpName:"StringSplit",category:"string",inputs:[{start:0,name:"input",type:"tensor"},{start:1,name:"delimiter",type:"tensor"}],attrs:[{tfName:"skip_empty",name:"skipEmpty",type:"bool"}],outputs:["indices","values","shape"]},{tfOpName:"StringToHashBucketFast",category:"string",inputs:[{start:0,name:"input",type:"tensor"}],attrs:[{tfName:"num_buckets",name:"numBuckets",type:"number"}]}],d7={};Oe(d7,{json:()=>dU});var dU=[{tfOpName:"Cast",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"SrcT",name:"sdtype",type:"dtype",notSupported:!0},{tfName:"DstT",name:"dtype",type:"dtype"}]},{tfOpName:"ExpandDims",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"MirrorPad",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"padding",type:"number[]"}],attrs:[{tfName:"mode",name:"mode",type:"string"}]},{tfOpName:"Pad",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"padding",type:"number[]"}],attrs:[{tfName:"constant_value",name:"constantValue",type:"number",defaultValue:0}]},{tfOpName:"PadV2",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"padding",type:"number[]"},{start:2,name:"constantValue",type:"number",defaultValue:0}]},{tfOpName:"Reshape",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"shape",type:"number[]"}]},{tfOpName:"Squeeze",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"axis",tfDeprecatedName:"squeeze_dims",name:"axis",type:"number[]"}]},{tfOpName:"SpaceToBatchND",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"blockShape",type:"number[]"},{start:2,name:"paddings",type:"number[]"}]},{tfOpName:"BatchToSpaceND",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"blockShape",type:"number[]"},{start:2,name:"crops",type:"number[]"}]},{tfOpName:"DepthToSpace",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"block_size",name:"blockSize",type:"number"},{tfName:"data_format",name:"dataFormat",type:"string"}]},{tfOpName:"BroadcastTo",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"shape",type:"number[]"}],attrs:[]},{tfOpName:"BroadcastArgs",category:"transformation",inputs:[{start:0,name:"s0",type:"tensor"},{start:1,name:"s1",type:"tensor"}],attrs:[]}],p7=class{static get Instance(){return this._instance||(this._instance=new this)}constructor(){let e=[qk,Xk,Kk,Zk,Yk,Jk,Qk,e7,t7,n7,s7,r7,a7,o7,i7,l7,u7,c7,d7],t=[].concat(...e.map(n=>n.json));this.opMappers=t.reduce((n,s)=>(n[s.tfOpName]=s,n),{})}transformGraph(e,t={}){let n=e.node,s=[],r=[],a=[],o=n.reduce((f,m)=>(f[m.name]=this.mapNode(m),m.op.startsWith("Placeholder")?s.push(f[m.name]):m.op==="Const"?r.push(f[m.name]):(m.input==null||m.input.length===0)&&a.push(f[m.name]),f),{}),i=[],l=[],c={},u={};t!=null&&(c=this.mapSignatureEntries(t.inputs),u=this.mapSignatureEntries(t.outputs));let d=Object.keys(o);d.forEach(f=>{let m=o[f];m.inputNames.forEach((g,A)=>{let[x,,y]=Or(g),b=o[x];if(b.outputs!=null){let w=b.outputs.indexOf(y);if(w!==-1){let k=`${x}:${w}`;m.inputNames[A]=k}}m.inputs.push(b),b.children.push(m)})}),Object.keys(u).length===0?d.forEach(f=>{let m=o[f];m.children.length===0&&l.push(m)}):Object.keys(u).forEach(f=>{let[m]=Or(f),g=o[m];g!=null&&(g.signatureKey=u[f],l.push(g))}),Object.keys(c).length>0?Object.keys(c).forEach(f=>{let[m]=Or(f),g=o[m];g&&(g.signatureKey=c[f],i.push(g))}):i=s;let p={};e.library!=null&&e.library.function!=null&&(p=e.library.function.reduce((f,m)=>(f[m.signature.name]=this.mapFunction(m),f),{}));let h={nodes:o,inputs:i,outputs:l,weights:r,placeholders:s,signature:t,functions:p};return a.length>0&&(h.initNodes=a),h}mapSignatureEntries(e){return Object.keys(e||{}).reduce((t,n)=>(t[e[n].name]=n,t),{})}mapNode(e){let t=jk(e.op)||this.opMappers[e.op]||{};e.attr==null&&(e.attr={});let n={name:e.name,op:e.op,category:t.category,inputNames:(e.input||[]).map(s=>s.startsWith("^")?s.substr(1):s),inputs:[],children:[],inputParams:{},attrParams:{},rawAttrs:e.attr,outputs:t.outputs};return t.inputs!=null&&(n.inputParams=t.inputs.reduce((s,r)=>(s[r.name]={type:r.type,inputIndexStart:r.start,inputIndexEnd:r.end},s),{})),t.attrs!=null&&(n.attrParams=t.attrs.reduce((s,r)=>{let a=r.type,o;switch(r.type){case"string":o=my(e.attr,r.tfName,r.defaultValue),o===void 0&&!!r.tfDeprecatedName&&(o=my(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"string[]":o=ky(e.attr,r.tfName,r.defaultValue),o===void 0&&!!r.tfDeprecatedName&&(o=ky(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"number":o=Ay(e.attr,r.tfName,r.defaultValue||0),o===void 0&&!!r.tfDeprecatedName&&(o=Ay(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"number[]":o=wy(e.attr,r.tfName,r.defaultValue),o===void 0&&!!r.tfDeprecatedName&&(o=wy(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"bool":o=gy(e.attr,r.tfName,r.defaultValue),o===void 0&&!!r.tfDeprecatedName&&(o=gy(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"bool[]":o=Iy(e.attr,r.tfName,r.defaultValue),o===void 0&&!!r.tfDeprecatedName&&(o=Iy(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"shape":o=vy(e.attr,r.tfName,r.defaultValue),o===void 0&&!!r.tfDeprecatedName&&(o=vy(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"shape[]":o=Sy(e.attr,r.tfName,r.defaultValue),o===void 0&&!!r.tfDeprecatedName&&(o=Sy(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"dtype":o=xy(e.attr,r.tfName,r.defaultValue),o===void 0&&!!r.tfDeprecatedName&&(o=xy(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"dtype[]":o=by(e.attr,r.tfName,r.defaultValue),o===void 0&&!!r.tfDeprecatedName&&(o=by(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"func":o=f7(e.attr,r.tfName,r.defaultValue),o===void 0&&!!r.tfDeprecatedName&&(o=f7(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"tensor":case"tensors":break;default:throw new Error(`Unsupported param type: ${r.type} for op: ${e.op}`)}return s[r.name]={value:o,type:a},s},{})),n}mapFunction(e){let t=e.nodeDef,n=[],s=[],r={};t!=null&&(r=t.reduce((u,d)=>(u[d.name]=this.mapNode(d),d.op==="Const"&&s.push(u[d.name]),u),{}));let a=[],o=[];e.signature.inputArg.forEach(u=>{let[d]=Or(u.name),p={name:d,op:"Placeholder",inputs:[],inputNames:[],category:"graph",inputParams:{},attrParams:{dtype:{value:yy(u.type),type:"dtype"}},children:[]};p.signatureKey=u.name,a.push(p),r[d]=p}),Object.keys(r).forEach(u=>{let d=r[u];d.inputNames.forEach((p,h)=>{let[f,,m]=Or(p),g=r[f];if(g.outputs!=null){let A=g.outputs.indexOf(m);if(A!==-1){let x=`${f}:${A}`;d.inputNames[h]=x}}d.inputs.push(g),g.children.push(d)})});let l=e.ret;e.signature.outputArg.forEach(u=>{let[d,p]=Or(l[u.name]),h=r[d];h!=null&&(h.defaultOutput=p,o.push(h))});let c=this.mapArgsToSignature(e);return{nodes:r,inputs:a,outputs:o,weights:s,placeholders:n,signature:c}}mapArgsToSignature(e){return{methodName:e.signature.name,inputs:e.signature.inputArg.reduce((t,n)=>(t[n.name]=this.mapArgToTensorInfo(n),t),{}),outputs:e.signature.outputArg.reduce((t,n)=>(t[n.name]=this.mapArgToTensorInfo(n,e.ret),t),{})}}mapArgToTensorInfo(e,t){let n=e.name;return t!=null&&(n=t[n]),{name:n,dtype:e.type}}};function pU(e){let t=K().global;if(typeof t.atob!="undefined")return t.atob(e);if(typeof Buffer!="undefined")return new Buffer(e,"base64").toString();throw new Error("Unable to decode base64 in this environment. Missing built-in atob() or Buffer()")}function h7(e,t){let n=Array.isArray(e)?String.fromCharCode.apply(null,e):pU(e);return t?n:n.toLowerCase()}function my(e,t,n,s=!1){let r=e[t];return r!=null?h7(r.s,s):n}function gy(e,t,n){let s=e[t];return s?s.b:n}function Ay(e,t,n){let s=e[t]||{},r=s.i!=null?s.i:s.f!=null?s.f:n;return typeof r=="number"?r:parseInt(r,10)}function yy(e){switch(typeof e=="string"&&(e=gr[e]),e){case gr.DT_FLOAT:return"float32";case gr.DT_INT32:case gr.DT_INT64:case gr.DT_INT8:case gr.DT_UINT8:return"int32";case gr.DT_BOOL:return"bool";case gr.DT_DOUBLE:return"float32";case gr.DT_STRING:return"string";default:return null}}function f7(e,t,n){let s=e[t];return s&&s.func?s.func.name:n}function xy(e,t,n){let s=e[t];return s&&s.type?yy(s.type):n}function by(e,t,n){let s=e[t];return s&&s.list&&s.list.type?s.list.type.map(r=>yy(r)):n}function m7(e){if(!e.unknownRank)return e.dim!=null?e.dim.map(t=>typeof t.size=="number"?t.size:parseInt(t.size,10)):[]}function vy(e,t,n){let s=e[t];return s&&s.shape?m7(s.shape):n}function wy(e,t,n){let s=e[t];return s?((s.list.f&&s.list.f.length?s.list.f:s.list.i)||[]).map(r=>typeof r=="number"?r:parseInt(r,10)):n}function ky(e,t,n,s=!1){let r=e[t];return r&&r.list&&r.list.s?r.list.s.map(a=>h7(a,s)):n}function Sy(e,t,n){let s=e[t];return s&&s.list&&s.list.shape?s.list.shape.map(r=>m7(r)):n}function Iy(e,t,n){let s=e[t];return s&&s.list&&s.list.b?s.list.b:n}var hU=class{constructor(e,t,n){this.node=e,this.tensorMap=t,this.context=n,this.inputs=[],this.attrs={},this.inputs=e.inputNames.map(s=>this.getInput(s)),e.rawAttrs!=null&&(this.attrs=Object.keys(e.rawAttrs).reduce((s,r)=>(s[r]=this.getAttr(r),s),{}))}getInput(e){return Vn(e,this.tensorMap,this.context)}getAttr(e,t){let n=this.node.rawAttrs[e];if(n.tensor!=null)return Vn(e,this.tensorMap,this.context);if(n.i!=null||n.f!=null)return Ay(this.node.rawAttrs,e,t);if(n.s!=null)return my(this.node.rawAttrs,e,t);if(n.b!=null)return gy(this.node.rawAttrs,e,t);if(n.shape!=null)return vy(this.node.rawAttrs,e,t);if(n.type!=null)return xy(this.node.rawAttrs,e,t);if(n.list!=null){if(n.list.i!=null||n.list.f!=null)return wy(this.node.rawAttrs,e,t);if(n.list.s!=null)return ky(this.node.rawAttrs,e,t);if(n.list.shape!=null)return Sy(this.node.rawAttrs,e,t);if(n.list.b!=null)return Iy(this.node.rawAttrs,e,t);if(n.list.type!=null)return by(this.node.rawAttrs,e,t)}return t}},fU=(e,t,n)=>{switch(e.op){case"BiasAdd":case"AddV2":case"Add":return[le(S("a",e,t,n),S("b",e,t,n))];case"AddN":return[sf(S("tensors",e,t,n))];case"FloorMod":case"Mod":return[Rd(S("a",e,t,n),S("b",e,t,n))];case"Mul":return[L(S("a",e,t,n),S("b",e,t,n))];case"RealDiv":case"Div":return[pe(S("a",e,t,n),S("b",e,t,n))];case"DivNoNan":return[sv(S("a",e,t,n),S("b",e,t,n))];case"FloorDiv":return[nf(S("a",e,t,n),S("b",e,t,n))];case"Sub":return[me(S("a",e,t,n),S("b",e,t,n))];case"Minimum":return[Ed(S("a",e,t,n),S("b",e,t,n))];case"Maximum":return[Kr(S("a",e,t,n),S("b",e,t,n))];case"Pow":return[Ro(S("a",e,t,n),S("b",e,t,n))];case"SquaredDifference":return[m1(S("a",e,t,n),S("b",e,t,n))];default:throw TypeError(`Node type ${e.op} is not implemented`)}},mU=(e,t,n)=>{switch(e.op){case"Abs":case"ComplexAbs":return[tn(S("x",e,t,n))];case"Acos":return[D3(S("x",e,t,n))];case"Acosh":return[P3(S("x",e,t,n))];case"Asin":return[O3(S("x",e,t,n))];case"Asinh":return[M3(S("x",e,t,n))];case"Atan":return[z3(S("x",e,t,n))];case"Atan2":return[L3(S("x",e,t,n),S("y",e,t,n))];case"Atanh":return[B3(S("x",e,t,n))];case"Ceil":return[X3(S("x",e,t,n))];case"Complex":return[ko(S("real",e,t,n),S("imag",e,t,n))];case"Cos":return[uf(S("x",e,t,n))];case"Cosh":return[K2(S("x",e,t,n))];case"Elu":return[Td(S("x",e,t,n))];case"Erf":return[av(S("x",e,t,n))];case"Exp":return[Ts(S("x",e,t,n))];case"Expm1":return[ov(S("x",e,t,n))];case"Floor":return[Nd(S("x",e,t,n))];case"Log":return[Ns(S("x",e,t,n))];case"Log1p":return[pf(S("x",e,t,n))];case"Imag":return[cf(S("x",e,t,n))];case"Neg":return[Ft(S("x",e,t,n))];case"Reciprocal":return[xv(S("x",e,t,n))];case"Real":return[$d(S("x",e,t,n))];case"Relu":return[Rr(S("x",e,t,n))];case"Round":return[l1(S("x",e,t,n))];case"Selu":return[c1(S("x",e,t,n))];case"Sigmoid":return[cs(S("x",e,t,n))];case"Sin":return[d1(S("x",e,t,n))];case"Sign":return[wv(S("x",e,t,n))];case"Sinh":return[p1(S("x",e,t,n))];case"Softplus":return[Mu(S("x",e,t,n))];case"Sqrt":return[_n(S("x",e,t,n))];case"Square":return[gt(S("x",e,t,n))];case"Tanh":return[_u(S("x",e,t,n))];case"Tan":return[Sv(S("x",e,t,n))];case"ClipByValue":return[ds(S("x",e,t,n),S("clipValueMin",e,t,n),S("clipValueMax",e,t,n))];case"Relu6":return[i1(S("x",e,t,n))];case"Rsqrt":return[u1(Vn(e.inputNames[0],t,n))];case"Prod":return[r1(S("x",e,t,n),S("axes",e,t,n))];case"LeakyRelu":return[df(S("x",e,t,n),S("alpha",e,t,n))];case"Prelu":return[yf(S("x",e,t,n),S("alpha",e,t,n))];case"IsNan":return[iv(Vn(e.inputNames[0],t,n))];default:throw TypeError(`Node type ${e.op} is not implemented`)}};function Ys(e,t,n=""){if(!(typeof e=="number"||typeof t=="number")){v.assert(e.length===t.length,()=>n+` Shapes ${e} and ${t} must match`);for(let s=0;s<e.length;s++){let r=e[s],a=t[s];v.assert(r<0||a<0||r===a,()=>n+` Shapes ${e} and ${t} must match`)}}}function g7(e){return!(typeof e=="number"||e.some(t=>t<0))}function Qd(e,t,n){let s=Cy(e,n),r=!g7(s);if(r&&t.length===0)throw new Error(`Tried to calculate elements of an empty list with non-fully-defined elementShape: ${s}`);if(r&&t.forEach(a=>{s=Cy(a.shape,s)}),!g7(s))throw new Error(`Non-fully-defined elementShape: ${s}`);return s}function Cy(e,t){if(typeof e=="number")return t;if(typeof t=="number")return e;if(e.length!==t.length)throw new Error(`Incompatible ranks during merge: ${e} vs. ${t}`);let n=[];for(let s=0;s<e.length;++s){let r=e[s],a=t[s];if(r>=0&&a>=0&&r!==a)throw new Error(`Incompatible shape during merge: ${e} vs. ${t}`);n[s]=r>=0?r:a}return n}var gU=class{constructor(e,t,n,s,r,a,o){this.name=e,this.dtype=t,this.maxSize=n,this.elementShape=s,this.identicalElementShapes=r,this.dynamicSize=a,this.clearAfterRead=o,this.tensors=[],this.closed_=!1,this.idTensor=Re(0),fn(this.idTensor)}get id(){return this.idTensor.id}get closed(){return this.closed_}clearAndClose(e){this.tensors.forEach(t=>{(e==null||!e.has(t.tensor.id))&&t.tensor.dispose()}),this.tensors=[],this.closed_=!0,this.idTensor.dispose()}size(){return this.tensors.length}read(e){if(this.closed_)throw new Error(`TensorArray ${this.name} has already been closed.`);if(e<0||e>=this.size())throw new Error(`Tried to read from index ${e}, but array size is: ${this.size()}`);let t=this.tensors[e];if(t.cleared)throw new Error(`TensorArray ${this.name}: Could not read index ${e} twice because it was cleared after a previous read (perhaps try setting clear_after_read = false?).`);return this.clearAfterRead&&(t.cleared=!0),t.read=!0,t.tensor}readMany(e){return e.map(t=>this.read(t))}write(e,t){if(this.closed_)throw new Error(`TensorArray ${this.name} has already been closed.`);if(e<0||!this.dynamicSize&&e>=this.maxSize)throw new Error(`Tried to write to index ${e}, but array is not resizeable and size is: ${this.maxSize}`);let n=this.tensors[e]||{};if(t.dtype!==this.dtype)throw new Error(`TensorArray ${this.name}: Could not write to TensorArray index ${e},
|
|
because the value dtype is ${t.dtype}, but TensorArray dtype is ${this.dtype}.`);if(this.size()===0&&(this.elementShape==null||this.elementShape.length===0)&&(this.elementShape=t.shape),Ys(this.elementShape,t.shape,`TensorArray ${this.name}: Could not write to TensorArray index ${e}.`),n.read)throw new Error(`TensorArray ${this.name}: Could not write to TensorArray index ${e}, because it has already been read.`);if(n.written)throw new Error(`TensorArray ${this.name}: Could not write to TensorArray index ${e}, because it has already been written.`);n.tensor=t,fn(t),n.written=!0,this.tensors[e]=n}writeMany(e,t){if(e.length!==t.length)throw new Error(`TensorArray ${this.name}: could not write multiple tensors,because the index size: ${e.length} is not the same as tensors size: ${t.length}.`);e.forEach((n,s)=>this.write(n,t[s]))}gather(e,t){if(!!t&&t!==this.dtype)throw new Error(`TensorArray dtype is ${this.dtype} but gather requested dtype ${t}`);if(e)e=e.slice(0,this.size());else{e=[];for(let s=0;s<this.size();s++)e.push(s)}if(e.length===0)return Ut([],[0].concat(this.elementShape));let n=this.readMany(e);return Ys(this.elementShape,n[0].shape,"TensorArray shape mismatch: "),An(n,0)}concat(e){if(!!e&&e!==this.dtype)throw new Error(`TensorArray dtype is ${this.dtype} but concat requested dtype ${e}`);if(this.size()===0)return Ut([],[0].concat(this.elementShape));let t=[];for(let s=0;s<this.size();s++)t.push(s);let n=this.readMany(t);return Ys(this.elementShape,n[0].shape,`TensorArray shape mismatch: tensor array shape (${this.elementShape}) vs first tensor shape (${n[0].shape})`),vt(n,0)}scatter(e,t){if(t.dtype!==this.dtype)throw new Error(`TensorArray dtype is ${this.dtype} but tensor has dtype ${t.dtype}`);if(e.length!==t.shape[0])throw new Error(`Expected len(indices) == tensor.shape[0], but saw: ${e.length} vs. ${t.shape[0]}`);let n=Math.max(...e);if(!this.dynamicSize&&n>=this.maxSize)throw new Error(`Max index must be < array size (${n} vs. ${this.maxSize})`);this.writeMany(e,ss(t,0))}split(e,t){if(t.dtype!==this.dtype)throw new Error(`TensorArray dtype is ${this.dtype} but tensor has dtype ${t.dtype}`);let n=0,s=e.map(i=>(n+=i,n));if(n!==t.shape[0])throw new Error(`Expected sum of lengths to be equal to
|
|
tensor.shape[0], but sum of lengths is
|
|
${n}, and tensor's shape is: ${t.shape}`);if(!this.dynamicSize&&e.length!==this.maxSize)throw new Error(`TensorArray's size is not equal to the size of lengths (${this.maxSize} vs. ${e.length}), and the TensorArray is not marked as dynamically resizeable`);let r=n===0?0:t.size/n,a=[];q(()=>{t=G(t,[1,n,r]);for(let i=0;i<e.length;++i){let l=i===0?0:s[i-1],c=[0,l,0],u=[1,e[i],r];a[i]=G(De(t,c,u),this.elementShape)}return a});let o=[];for(let i=0;i<e.length;i++)o[i]=i;this.writeMany(o,a)}},ep=class{constructor(e,t,n,s=-1){this.tensors=e,this.elementShape=t,this.elementDtype=n,e!=null&&e.forEach(r=>{if(n!==r.dtype)throw new Error(`Invalid data types; op elements ${n}, but list elements ${r.dtype}`);Ys(t,r.shape,"TensorList shape mismatch: "),fn(r)}),this.idTensor=Re(0),this.maxNumElements=s,fn(this.idTensor)}get id(){return this.idTensor.id}copy(){return new ep([...this.tensors],this.elementShape,this.elementDtype)}clearAndClose(e){this.tensors.forEach(t=>{(e==null||!e.has(t.id))&&t.dispose()}),this.tensors.length=0,this.idTensor.dispose()}size(){return this.tensors.length}stack(e,t,n=-1){if(t!==this.elementDtype)throw new Error(`Invalid data types; op elements ${t}, but list elements ${this.elementDtype}`);if(n!==-1&&this.tensors.length!==n)throw new Error(`Operation expected a list with ${n} elements but got a list with ${this.tensors.length} elements.`);Ys(e,this.elementShape,"TensorList shape mismatch: ");let s=Qd(this.elementShape,this.tensors,e);return q(()=>{let r=this.tensors.map(a=>G(a,s));return An(r,0)})}popBack(e,t){if(t!==this.elementDtype)throw new Error(`Invalid data types; op elements ${t}, but list elements ${this.elementDtype}`);if(this.size()===0)throw new Error("Trying to pop from an empty list.");let n=Qd(this.elementShape,this.tensors,e),s=this.tensors.pop();return Ys(s.shape,e,"TensorList shape mismatch: "),G(s,n)}pushBack(e){if(e.dtype!==this.elementDtype)throw new Error(`Invalid data types; op elements ${e.dtype}, but list elements ${this.elementDtype}`);if(Ys(e.shape,this.elementShape,"TensorList shape mismatch: "),this.maxNumElements===this.size())throw new Error("Trying to push element into a full list.");fn(e),this.tensors.push(e)}resize(e){if(e<0)throw new Error(`TensorListResize expects size to be non-negative. Got: ${e}`);if(this.maxNumElements!==-1&&e>this.maxNumElements)throw new Error(`TensorListResize input size ${e} is greater maxNumElement ${this.maxNumElements}.`);this.tensors.length=e}getItem(e,t,n){if(n!==this.elementDtype)throw new Error(`Invalid data types; op elements ${n}, but list elements ${this.elementDtype}`);if(e<0||e>this.tensors.length)throw new Error(`Trying to access element ${e} in a list with ${this.tensors.length} elements.`);if(this.tensors[e]==null)throw new Error(`element at index ${e} is null.`);Ys(this.tensors[e].shape,t,"TensorList shape mismatch: ");let s=Qd(this.elementShape,this.tensors,t);return G(this.tensors[e],s)}setItem(e,t){if(t.dtype!==this.elementDtype)throw new Error(`Invalid data types; op elements ${t.dtype}, but list elements ${this.elementDtype}`);if(e<0||this.maxNumElements!==-1&&e>=this.maxNumElements)throw new Error(`Trying to set element ${e} in a list with max ${this.maxNumElements} elements.`);Ys(this.elementShape,t.shape,"TensorList shape mismatch: "),fn(t),this.tensors[e]=t}gather(e,t,n){if(t!==this.elementDtype)throw new Error(`Invalid data types; op elements ${t}, but list elements ${this.elementDtype}`);Ys(this.elementShape,n,"TensorList shape mismatch: "),e=e.slice(0,this.size());let s=Qd(this.elementShape,this.tensors,n);return e.length===0?Ut([],[0].concat(s)):q(()=>{let r=e.map(a=>G(this.tensors[a],s));return An(r,0)})}concat(e,t){if(!!e&&e!==this.elementDtype)throw new Error(`TensorList dtype is ${this.elementDtype} but concat requested dtype ${e}`);Ys(this.elementShape,t,"TensorList shape mismatch: ");let n=Qd(this.elementShape,this.tensors,t);return this.size()===0?Ut([],[0].concat(n)):q(()=>{let s=this.tensors.map(r=>G(r,n));return vt(s,0)})}};function AU(e,t,n){let s=e.dtype;if(e.shape.length<1)throw new Error(`Tensor must be at least a vector, but saw shape: ${e.shape}`);if(e.dtype!==n)throw new Error(`Invalid data types; op elements ${e.dtype}, but list elements ${n}`);let r=e.shape.slice(1);Ys(r,t,"TensorList shape mismatch: ");let a=ss(e);return new ep(a,t,s)}function yU(e,t,n){return new ep([],e,t,n)}function xU(e,t,n,s){if(t.length!==e.shape[0])throw new Error(`Expected len(indices) == tensor.shape[0], but saw: ${t.length} vs. ${e.shape[0]}`);let r=Math.max(...t);if(s!=null&&s!==-1&&r>=s)throw new Error(`Max index must be < array size (${r} vs. ${s})`);let a=new ep([],n,e.dtype,s),o=ss(e,0);return t.forEach((i,l)=>{a.setItem(i,o[l])}),a}function bU(e,t,n){let s=0,r=t.map(u=>(s+=u,s));if(s!==e.shape[0])throw new Error(`Expected sum of lengths to be equal to
|
|
tensor.shape[0], but sum of lengths is
|
|
${s}, and tensor's shape is: ${e.shape}`);let a=e.shape.slice(1),o=Cy(a,n),i=s===0?0:e.size/s,l=q(()=>{let u=[];e=G(e,[1,s,i]);for(let d=0;d<t.length;++d){let p=d===0?0:r[d-1],h=[0,p,0],f=[1,t[d],i];u[d]=G(De(e,h,f),o)}return e.dispose(),u}),c=new ep([],n,e.dtype,t.length);for(let u=0;u<l.length;u++)c.setItem(u,l[u]);return c}var vU=async(e,t,n)=>{switch(e.op){case"If":case"StatelessIf":{let s=S("thenBranch",e,t,n),r=S("elseBranch",e,t,n),a=S("cond",e,t,n),o=S("args",e,t,n);return(await a.data())[0]?n.functionMap[s].executeFunctionAsync(o,n.tensorArrayMap,n.tensorListMap):n.functionMap[r].executeFunctionAsync(o,n.tensorArrayMap,n.tensorListMap)}case"While":case"StatelessWhile":{let s=S("body",e,t,n),r=S("cond",e,t,n),a=S("args",e,t,n),o=await n.functionMap[r].executeFunctionAsync(a,n.tensorArrayMap,n.tensorListMap),i=a.map(u=>u.id),l=await o[0].data();o.forEach(u=>{!u.kept&&i.indexOf(u.id)===-1&&u.dispose()});let c=a;for(;l[0];){let u=c;c=await n.functionMap[s].executeFunctionAsync(c,n.tensorArrayMap,n.tensorListMap);let d=c.map(h=>h.id);u.forEach(h=>{!h.kept&&i.indexOf(h.id)===-1&&d.indexOf(h.id)===-1&&h.dispose()});let p=await n.functionMap[r].executeFunctionAsync(c,n.tensorArrayMap,n.tensorListMap);l=await p[0].data(),p.forEach(h=>{!h.kept&&i.indexOf(h.id)===-1&&d.indexOf(h.id)===-1&&h.dispose()})}return c}case"LoopCond":{let s=S("pred",e,t,n);return[na(s)]}case"Switch":{let s=S("pred",e,t,n),r=S("data",e,t,n);return r.kept||(r=na(r)),(await s.data())[0]?[void 0,r]:[r,void 0]}case"Merge":{let s=e.inputNames.find(r=>Vn(r,t,n)!==void 0);if(s){let r=Vn(s,t,n);return[na(r)]}return}case"Enter":{let s=S("frameName",e,t,n),r=S("tensor",e,t,n);return n.enterFrame(s),[na(r)]}case"Exit":{let s=S("tensor",e,t,n);return n.exitFrame(),[na(s)]}case"NextIteration":{let s=S("tensor",e,t,n);return n.nextIteration(),[na(s)]}case"TensorArrayV3":{let s=S("size",e,t,n),r=S("dtype",e,t,n),a=S("elementShape",e,t,n),o=S("dynamicSize",e,t,n),i=S("clearAfterRead",e,t,n),l=S("identicalElementShapes",e,t,n),c=S("name",e,t,n),u=new gU(c,r,s,a,l,o,i);return n.addTensorArray(u),[u.idTensor,Re(1)]}case"TensorArrayWriteV3":{let s=S("tensorArrayId",e,t,n),r=S("index",e,t,n),a=S("tensor",e,t,n),o=n.getTensorArray(s.id);return o.write(r,a),[o.idTensor]}case"TensorArrayReadV3":{let s=S("tensorArrayId",e,t,n),r=S("index",e,t,n);return[n.getTensorArray(s.id).read(r)]}case"TensorArrayGatherV3":{let s=S("tensorArrayId",e,t,n),r=S("indices",e,t,n),a=S("dtype",e,t,n);return[n.getTensorArray(s.id).gather(r,a)]}case"TensorArrayScatterV3":{let s=S("tensorArrayId",e,t,n),r=S("indices",e,t,n),a=S("tensor",e,t,n),o=n.getTensorArray(s.id);return o.scatter(r,a),[o.idTensor]}case"TensorArrayConcatV3":{let s=S("tensorArrayId",e,t,n),r=n.getTensorArray(s.id),a=S("dtype",e,t,n);return[r.concat(a)]}case"TensorArraySplitV3":{let s=S("tensorArrayId",e,t,n),r=S("tensor",e,t,n),a=S("lengths",e,t,n),o=n.getTensorArray(s.id);return o.split(a,r),[o.idTensor]}case"TensorArraySizeV3":{let s=S("tensorArrayId",e,t,n),r=n.getTensorArray(s.id);return[Re(r.size(),"int32")]}case"TensorArrayCloseV3":{let s=S("tensorArrayId",e,t,n),r=n.getTensorArray(s.id);return r.clearAndClose(),[r.idTensor]}case"TensorListSetItem":{let s=S("tensorListId",e,t,n),r=S("index",e,t,n),a=S("tensor",e,t,n),o=n.getTensorList(s.id);return o.setItem(r,a),[o.idTensor]}case"TensorListGetItem":{let s=S("tensorListId",e,t,n),r=S("index",e,t,n),a=S("elementShape",e,t,n),o=S("elementDType",e,t,n);return[n.getTensorList(s.id).getItem(r,a,o)]}case"TensorListScatterV2":case"TensorListScatter":{let s=S("indices",e,t,n),r=S("tensor",e,t,n),a=S("elementShape",e,t,n),o=S("numElements",e,t,n),i=xU(r,s,a,o);return n.addTensorList(i),[i.idTensor]}case"TensorListReserve":case"EmptyTensorList":{let s=S("elementShape",e,t,n),r=S("elementDType",e,t,n),a;e.op==="TensorListReserve"?a="numElements":a="maxNumElements";let o=S(a,e,t,n),i=yU(s,r,o);return n.addTensorList(i),[i.idTensor]}case"TensorListGather":{let s=S("tensorListId",e,t,n),r=S("indices",e,t,n),a=S("elementShape",e,t,n),o=S("elementDType",e,t,n);return[n.getTensorList(s.id).gather(r,o,a)]}case"TensorListStack":{let s=S("tensorListId",e,t,n),r=S("elementShape",e,t,n),a=S("elementDType",e,t,n),o=S("numElements",e,t,n);return[n.getTensorList(s.id).stack(r,a,o)]}case"TensorListFromTensor":{let s=S("tensor",e,t,n),r=S("elementShape",e,t,n),a=S("elementDType",e,t,n),o=AU(s,r,a);return n.addTensorList(o),[o.idTensor]}case"TensorListConcat":{let s=S("tensorListId",e,t,n),r=n.getTensorList(s.id),a=S("dtype",e,t,n),o=S("elementShape",e,t,n);return[r.concat(a,o)]}case"TensorListPushBack":{let s=S("tensorListId",e,t,n),r=S("tensor",e,t,n),a=n.getTensorList(s.id);return a.pushBack(r),[a.idTensor]}case"TensorListPopBack":{let s=S("tensorListId",e,t,n),r=S("elementShape",e,t,n),a=S("elementDType",e,t,n);return[n.getTensorList(s.id).popBack(r,a)]}case"TensorListSplit":{let s=S("tensor",e,t,n),r=S("elementShape",e,t,n),a=S("lengths",e,t,n),o=bU(s,a,r);return n.addTensorList(o),[o.idTensor]}default:throw TypeError(`Node type ${e.op} is not implemented`)}};function A7(e,t,n){let[s,r]=S("fusedOps",e,t,n),a=s==="biasadd",o=!a,i=r==="prelu",l=s==="fusedbatchnorm",c=S("numArgs",e,t,n);if(a){if(i&&c!==2)throw new Error("FusedConv2d and DepthwiseConv2d with BiasAdd and Prelu must have two extra arguments: bias and alpha.");if(!i&&a&&c!==1)throw new Error("FusedConv2d and DepthwiseConv2d with BiasAdd must have one extra argument: bias.")}if(l)throw new Error("FusedConv2d and DepthwiseConv2d with FusedBatchNorm is not supported");let u=S("strides",e,t,n),d=mm(e,t,n),p=S("dataFormat",e,t,n).toUpperCase(),h=S("dilations",e,t,n),[f,m]=S("args",e,t,n);o&&(m=f,f=void 0);let g=S("leakyreluAlpha",e,t,n);return{stride:u,pad:d,dataFormat:p,dilations:h,biasArg:f,preluArg:m,activationFunc:r,leakyreluAlpha:g}}var wU=(e,t,n)=>{switch(e.op){case"Conv1D":{let s=S("stride",e,t,n),r=S("pad",e,t,n),a=S("dataFormat",e,t,n).toUpperCase(),o=S("dilation",e,t,n);return[H2(S("x",e,t,n),S("filter",e,t,n),s,r,a,o)]}case"Conv2D":{let s=S("strides",e,t,n),r=mm(e,t,n),a=S("dataFormat",e,t,n).toUpperCase(),o=S("dilations",e,t,n);return[No(S("x",e,t,n),S("filter",e,t,n),[s[1],s[2]],r,a,[o[1],o[2]])]}case"_FusedConv2D":{let{stride:s,pad:r,dataFormat:a,dilations:o,biasArg:i,preluArg:l,activationFunc:c,leakyreluAlpha:u}=A7(e,t,n);return[$o.conv2d({x:S("x",e,t,n),filter:S("filter",e,t,n),strides:[s[1],s[2]],pad:r,dataFormat:a,dilations:[o[1],o[2]],bias:i,activation:c,preluActivationWeights:l,leakyreluAlpha:u})]}case"FusedDepthwiseConv2dNative":{let{stride:s,pad:r,dataFormat:a,dilations:o,biasArg:i,preluArg:l,activationFunc:c,leakyreluAlpha:u}=A7(e,t,n);return[$o.depthwiseConv2d({x:S("x",e,t,n),filter:S("filter",e,t,n),strides:[s[1],s[2]],pad:r,dataFormat:a,dilations:[o[1],o[2]],bias:i,activation:c,preluActivationWeights:l,leakyreluAlpha:u})]}case"Conv2DBackpropInput":case"Conv2dTranspose":{let s=S("outputShape",e,t,n),r=S("strides",e,t,n),a=mm(e,t,n);return[q2(S("x",e,t,n),S("filter",e,t,n),s,[r[1],r[2]],a)]}case"DepthwiseConv2dNative":case"DepthwiseConv2d":{let s=S("strides",e,t,n),r=mm(e,t,n),a=S("dilations",e,t,n),o=S("dataFormat",e,t,n).toUpperCase();return[Cd(S("input",e,t,n),S("filter",e,t,n),[s[1],s[2]],r,o,[a[1],a[2]])]}case"Conv3D":{let s=S("strides",e,t,n),r=S("pad",e,t,n),a=S("dataFormat",e,t,n).toUpperCase(),o=S("dilations",e,t,n);return[X2(S("x",e,t,n),S("filter",e,t,n),[s[1],s[2],s[3]],r,a,[o[1],o[2],o[3]])]}case"AvgPool":{let s=S("strides",e,t,n),r=S("pad",e,t,n),a=S("kernelSize",e,t,n);return[of(S("x",e,t,n),[a[1],a[2]],[s[1],s[2]],r)]}case"MaxPool":{let s=S("strides",e,t,n),r=S("pad",e,t,n),a=S("kernelSize",e,t,n);return[mf(S("x",e,t,n),[a[1],a[2]],[s[1],s[2]],r)]}case"MaxPoolWithArgmax":{let s=S("strides",e,t,n),r=S("pad",e,t,n),a=S("kernelSize",e,t,n),o=S("includeBatchInIndex",e,t,n),{result:i,indexes:l}=mv(S("x",e,t,n),[a[1],a[2]],[s[1],s[2]],r,o);return[i,l]}case"AvgPool3D":{let s=S("strides",e,t,n),r=S("pad",e,t,n),a=S("kernelSize",e,t,n);return[U2(S("x",e,t,n),[a[1],a[2],a[3]],[s[1],s[2],s[3]],r)]}case"MaxPool3D":{let s=S("strides",e,t,n),r=S("pad",e,t,n),a=S("kernelSize",e,t,n);return[s1(S("x",e,t,n),[a[1],a[2],a[3]],[s[1],s[2],s[3]],r)]}case"Dilation2D":{let s=S("strides",e,t,n),r=S("pad",e,t,n),a=S("dilations",e,t,n),o=s[1],i=s[2],l=a[1],c=a[2];return[nv(S("x",e,t,n),S("filter",e,t,n),[o,i],r,[l,c],"NHWC")]}default:throw TypeError(`Node type ${e.op} is not implemented`)}},kU=(e,t,n)=>{switch(e.op){case"Fill":{let s=S("shape",e,t,n),r=S("dtype",e,t,n),a=S("value",e,t,n);return[Fu(s,a,r)]}case"LinSpace":{let s=S("start",e,t,n),r=S("stop",e,t,n),a=S("num",e,t,n);return[lv(s,r,a)]}case"Multinomial":{let s=S("logits",e,t,n),r=S("numSamples",e,t,n),a=S("seed",e,t,n);return[Av(s,r,a)]}case"OneHot":{let s=S("indices",e,t,n),r=S("depth",e,t,n),a=S("onValue",e,t,n),o=S("offValue",e,t,n);return[kd(s,r,a,o)]}case"Ones":return[hs(S("shape",e,t,n),S("dtype",e,t,n))];case"OnesLike":return[Es(S("x",e,t,n))];case"RandomUniform":return[Lu(S("shape",e,t,n),S("minval",e,t,n),S("maxval",e,t,n),S("dtype",e,t,n))];case"Range":{let s=S("start",e,t,n),r=S("stop",e,t,n),a=S("step",e,t,n);return[Bu(s,r,a,S("dtype",e,t,n))]}case"TruncatedNormal":{let s=S("shape",e,t,n),r=S("mean",e,t,n),a=S("stdDev",e,t,n),o=S("seed",e,t,n);return[wf(s,r,a,S("dtype",e,t,n),o)]}case"Zeros":return[Gt(S("shape",e,t,n),S("dtype",e,t,n))];case"ZerosLike":return[Je(S("x",e,t,n))];default:throw TypeError(`Node type ${e.op} is not implemented`)}};function Ty(e,t,n){let s=S("boxes",e,t,n),r=S("scores",e,t,n),a=S("maxOutputSize",e,t,n),o=S("iouThreshold",e,t,n),i=S("scoreThreshold",e,t,n),l=S("softNmsSigma",e,t,n);return{boxes:s,scores:r,maxOutputSize:a,iouThreshold:o,scoreThreshold:i,softNmsSigma:l}}var SU=async(e,t,n)=>{switch(e.op){case"NonMaxSuppressionV5":{let{boxes:s,scores:r,maxOutputSize:a,iouThreshold:o,scoreThreshold:i,softNmsSigma:l}=Ty(e,t,n),c=await $e.nonMaxSuppressionWithScoreAsync(s,r,a,o,i,l);return[c.selectedIndices,c.selectedScores]}case"NonMaxSuppressionV4":{let{boxes:s,scores:r,maxOutputSize:a,iouThreshold:o,scoreThreshold:i}=Ty(e,t,n),l=S("padToMaxOutputSize",e,t,n),c=await $e.nonMaxSuppressionPaddedAsync(s,r,a,o,i,l);return[c.selectedIndices,c.validOutputs]}case"NonMaxSuppressionV3":case"NonMaxSuppressionV2":{let{boxes:s,scores:r,maxOutputSize:a,iouThreshold:o,scoreThreshold:i}=Ty(e,t,n);return[await $e.nonMaxSuppressionAsync(s,r,a,o,i)]}case"Where":{let s=de(S("condition",e,t,n),"bool"),r=[await A1(s)];return s.dispose(),r}case"ListDiff":return vv(S("x",e,t,n),S("y",e,t,n));default:throw TypeError(`Node type ${e.op} is not implemented`)}},IU=(e,t,n)=>{switch(e.op){case"TopKV2":{let s=S("x",e,t,n),r=S("k",e,t,n),a=S("sorted",e,t,n),o=Iv(s,r,a);return[o.values,o.indices]}case"Unique":{let s=S("x",e,t,n),r=g1(s);return[r.values,r.indices]}case"UniqueV2":{let s=S("x",e,t,n),r=S("axis",e,t,n),a=g1(s,r);return[a.values,a.indices]}default:throw TypeError(`Node type ${e.op} is not implemented`)}},CU=(e,t,n)=>{switch(e.op){case"Const":return t[e.name];case"PlaceholderWithDefault":let s=S("default",e,t,n);return[Vn(e.name,t,n)||s];case"Placeholder":return[Vn(e.name,t,n)];case"Identity":case"StopGradient":case"FakeQuantWithMinMaxVars":{let c=S("x",e,t,n);return[na(c)]}case"IdentityN":return S("x",e,t,n).map(c=>na(c));case"Snapshot":let r=S("x",e,t,n);return[na(r)];case"Shape":return[Xt(S("x",e,t,n).shape,"int32")];case"ShapeN":return S("x",e,t,n).map(c=>Xt(c.shape));case"Size":return[Re(S("x",e,t,n).size,"int32")];case"Rank":return[Re(S("x",e,t,n).rank,"int32")];case"NoOp":return[Re(1)];case"Print":let a=S("x",e,t,n),o=S("data",e,t,n),i=S("message",e,t,n),l=S("summarize",e,t,n);console.warn("The graph has a tf.print() operation,usually used for debugging, which slows down performance."),console.log(i);for(let c=0;c<o.length;c++)console.log(Array.prototype.slice.call(o[c].dataSync()).slice(0,l));return[a];default:throw TypeError(`Node type ${e.op} is not implemented`)}},TU=class{constructor(e,t){this.keyDType=e,this.valueDType=t,this.handle=Re(0),this.tensorMap=new Map,fn(this.handle)}get id(){return this.handle.id}clearAndClose(){this.tensorMap.forEach(e=>e.dispose()),this.tensorMap.clear(),this.handle.dispose()}size(){return this.tensorMap.size}tensorSize(){return Re(this.size(),"int32")}async import(e,t){this.checkKeyAndValueTensor(e,t);let n=await e.data();return this.tensorMap.forEach(s=>s.dispose()),this.tensorMap.clear(),q(()=>{let s=ss(t),r=n.length,a=s.length;v.assert(r===a,()=>`The number of elements doesn't match, keys has ${r} elements, the values has ${a} elements.`);for(let o=0;o<r;o++){let i=n[o],l=s[o];fn(l),this.tensorMap.set(i,l)}return this.handle})}async find(e,t){this.checkKeyAndValueTensor(e,t);let n=await e.data();return q(()=>{let s=[];for(let r=0;r<n.length;r++){let a=n[r],o=this.findWithDefault(a,t);s.push(o)}return An(s)})}findWithDefault(e,t){let n=this.tensorMap.get(e);return n!=null?n:t}checkKeyAndValueTensor(e,t){if(e.dtype!==this.keyDType)throw new Error(`Expect key dtype ${this.keyDType}, but got ${e.dtype}`);if(t.dtype!==this.valueDType)throw new Error(`Expect value dtype ${this.valueDType}, but got ${t.dtype}`)}},NU=async(e,t,n,s)=>{switch(e.op){case"HashTable":case"HashTableV2":{let r=S("keyDType",e,t,n),a=S("valueDType",e,t,n),o=new TU(r,a);return s.addHashTable(e.name,o),[o.handle]}case"LookupTableImport":case"LookupTableImportV2":{let r=S("tableHandle",e,t,n,s),a=S("keys",e,t,n),o=S("values",e,t,n);return[await s.getHashTableById(r.id).import(a,o)]}case"LookupTableFind":case"LookupTableFindV2":{let r=S("tableHandle",e,t,n,s),a=S("keys",e,t,n),o=S("defaultValue",e,t,n);return[await s.getHashTableById(r.id).find(a,o)]}case"LookupTableSize":case"LookupTableSizeV2":{let r=S("tableHandle",e,t,n,s);return[s.getHashTableById(r.id).tensorSize()]}default:throw TypeError(`Node type ${e.op} is not implemented`)}},EU=(e,t,n)=>{switch(e.op){case"ResizeBilinear":{let s=S("images",e,t,n),r=S("size",e,t,n),a=S("alignCorners",e,t,n),o=S("halfPixelCenters",e,t,n);return[$e.resizeBilinear(s,[r[0],r[1]],a,o)]}case"ResizeNearestNeighbor":{let s=S("images",e,t,n),r=S("size",e,t,n),a=S("alignCorners",e,t,n),o=S("halfPixelCenters",e,t,n);return[$e.resizeNearestNeighbor(s,[r[0],r[1]],a,o)]}case"CropAndResize":{let s=S("image",e,t,n),r=S("boxes",e,t,n),a=S("boxInd",e,t,n),o=S("cropSize",e,t,n),i=S("method",e,t,n),l=S("extrapolationValue",e,t,n);return[$e.cropAndResize(s,r,a,o,i,l)]}default:throw TypeError(`Node type ${e.op} is not implemented`)}},RU=(e,t,n)=>{switch(e.op){case"Equal":return[Cs(S("a",e,t,n),S("b",e,t,n))];case"NotEqual":return[zu(S("a",e,t,n),S("b",e,t,n))];case"Greater":return[ps(S("a",e,t,n),S("b",e,t,n))];case"GreaterEqual":return[il(S("a",e,t,n),S("b",e,t,n))];case"Less":return[J2(S("a",e,t,n),S("b",e,t,n))];case"LessEqual":return[ll(S("a",e,t,n),S("b",e,t,n))];case"LogicalAnd":return[ir(S("a",e,t,n),S("b",e,t,n))];case"LogicalNot":return[ff(S("a",e,t,n))];case"LogicalOr":return[n1(S("a",e,t,n),S("b",e,t,n))];case"Select":case"SelectV2":return[Bn(S("condition",e,t,n),S("a",e,t,n),S("b",e,t,n))];default:throw TypeError(`Node type ${e.op} is not implemented`)}},$U=(e,t,n)=>{switch(e.op){case"BatchMatMul":case"BatchMatMulV2":case"MatMul":return[Ue(S("a",e,t,n),S("b",e,t,n),S("transposeA",e,t,n),S("transposeB",e,t,n))];case"Einsum":return[rv(S("equation",e,t,n),...S("tensors",e,t,n))];case"Transpose":return[Ye(S("x",e,t,n),S("perm",e,t,n))];case"_FusedMatMul":let[s,r]=S("fusedOps",e,t,n),a=s==="biasadd",o=r==="prelu",i=S("numArgs",e,t,n),l=S("leakyreluAlpha",e,t,n);if(a){if(o&&i!==2)throw new Error("Fused MatMul with BiasAdd and Prelu must have two extra arguments: bias and alpha.");if(!o&&i!==1)throw new Error("Fused MatMul with BiasAdd must have one extra argument: bias.")}let[c,u]=S("args",e,t,n);return[$o.matMul({a:S("a",e,t,n),b:S("b",e,t,n),transposeA:S("transposeA",e,t,n),transposeB:S("transposeB",e,t,n),bias:c,activation:r,preluActivationWeights:u,leakyreluAlpha:l})];default:throw TypeError(`Node type ${e.op} is not implemented`)}},_U=(e,t,n)=>{switch(e.op){case"FusedBatchNorm":case"FusedBatchNormV2":return[Du(S("x",e,t,n),S("mean",e,t,n),S("variance",e,t,n),S("offset",e,t,n),S("scale",e,t,n),S("epsilon",e,t,n))];case"FusedBatchNormV3":return[Du(S("x",e,t,n),S("mean",e,t,n),S("variance",e,t,n),S("offset",e,t,n),S("scale",e,t,n),S("epsilon",e,t,n))];case"LRN":return[uv(S("x",e,t,n),S("radius",e,t,n),S("bias",e,t,n),S("alpha",e,t,n),S("beta",e,t,n))];case"Softmax":return[Vu(S("x",e,t,n))];case"LogSoftmax":return[Q2(S("x",e,t,n))];case"SparseToDense":return[x1(S("sparseIndices",e,t,n),S("outputShape",e,t,n),S("sparseValues",e,t,n),S("defaultValue",e,t,n))];default:throw TypeError(`Node type ${e.op} is not implemented`)}},DU=(e,t,n)=>{switch(e.op){case"Max":{let o=S("axis",e,t,n),i=S("keepDims",e,t,n);return[gn(S("x",e,t,n),o,i)]}case"Mean":{let o=S("axis",e,t,n),i=S("keepDims",e,t,n);return[Bt(S("x",e,t,n),o,i)]}case"Min":{let o=S("axis",e,t,n),i=S("keepDims",e,t,n);return[Eo(S("x",e,t,n),o,i)]}case"Sum":{let o=S("axis",e,t,n),i=S("keepDims",e,t,n);return[Se(S("x",e,t,n),o,i)]}case"All":{let o=S("axis",e,t,n),i=S("keepDims",e,t,n);return[B2(S("x",e,t,n),o,i)]}case"Any":{let o=S("axis",e,t,n),i=S("keepDims",e,t,n);return[rf(S("x",e,t,n),o,i)]}case"ArgMax":{let o=S("axis",e,t,n);return[Hs(S("x",e,t,n),o)]}case"ArgMin":{let o=S("axis",e,t,n);return[F3(S("x",e,t,n),o)]}case"Prod":{let o=S("axis",e,t,n),i=S("keepDims",e,t,n);return[r1(S("x",e,t,n),o,i)]}case"Cumsum":{let o=S("axis",e,t,n),i=S("exclusive",e,t,n),l=S("reverse",e,t,n);return[Z2(S("x",e,t,n),o,i,l)]}case"Bincount":let s=S("x",e,t,n),r=S("weights",e,t,n),a=S("size",e,t,n);return[G2(s,r,a)];case"DenseBincount":{let o=S("x",e,t,n),i=S("weights",e,t,n),l=S("size",e,t,n),c=S("binaryOutput",e,t,n);return[ev(o,i,l,c)]}default:throw TypeError(`Node type ${e.op} is not implemented`)}},PU=(e,t,n)=>{switch(e.op){case"ConcatV2":case"Concat":{let s=S("n",e,t,n),r=S("axis",e,t,n),a=S("tensors",e,t,n);return a=a.slice(0,s),[vt(a,r)]}case"Gather":{let s=S("x",e,t,n),r=S("indices",e,t,n);return[Ou(s,de(r,"int32"),0)]}case"GatherV2":{let s=S("axis",e,t,n),r=S("batchDims",e,t,n),a=S("x",e,t,n),o=S("indices",e,t,n);return[Ou(a,de(o,"int32"),s,r)]}case"Reverse":{let s=S("dims",e,t,n),r=[];for(let o=0;o<s.length;o++)s[o]&&r.push(o);let a=S("x",e,t,n);return[Rs(a,r)]}case"ReverseV2":{let s=S("axis",e,t,n),r=S("x",e,t,n);return[Rs(r,s)]}case"Slice":{let s=S("begin",e,t,n),r=S("size",e,t,n);return[De(S("x",e,t,n),s,r)]}case"StridedSlice":{let s=S("begin",e,t,n),r=S("end",e,t,n),a=S("strides",e,t,n),o=S("beginMask",e,t,n),i=S("endMask",e,t,n),l=S("ellipsisMask",e,t,n),c=S("newAxisMask",e,t,n),u=S("shrinkAxisMask",e,t,n),d=S("x",e,t,n);return[kv(d,s,r,a,o,i,l,c,u)]}case"Pack":return q(()=>{let s=S("axis",e,t,n),r=S("tensors",e,t,n),a=r[0].shape,o=ot(r[0]).shape,i=r.map(l=>{let c=v.arraysEqual(l.shape,a);if(!c&&!v.arraysEqual(ot(l).shape,o))throw new Error("the input tensors shape does not match");return c?l:G(l,a)});return[An(i,s)]});case"Unpack":{let s=S("axis",e,t,n),r=S("tensor",e,t,n);return ss(r,s)}case"Tile":{let s=S("reps",e,t,n);return[js(S("x",e,t,n),s)]}case"Split":case"SplitV":{let s=S("axis",e,t,n),r=S("numOrSizeSplits",e,t,n),a=S("x",e,t,n);return nn(a,r,s)}case"ScatterNd":{let s=S("indices",e,t,n),r=S("values",e,t,n),a=S("shape",e,t,n);return[Rv(s,r,a)]}case"GatherNd":{let s=S("x",e,t,n),r=S("indices",e,t,n);return[$v(s,r)]}case"SparseToDense":{let s=S("sparseIndices",e,t,n),r=S("outputShape",e,t,n),a=S("sparseValues",e,t,n),o=S("defaultValue",e,t,n);return[x1(s,a,r,a.dtype===o.dtype?o:de(o,a.dtype))]}default:throw TypeError(`Node type ${e.op} is not implemented`)}},FU=(e,t,n)=>{switch(e.op){case"SparseFillEmptyRows":{let{outputIndices:s,outputValues:r,emptyRowIndicator:a,reverseIndexMap:o}=Fd.sparseFillEmptyRows(S("indices",e,t,n),S("values",e,t,n),S("denseShape",e,t,n),S("defaultValue",e,t,n));return[s,r,a,o]}case"SparseReshape":{let{outputIndices:s,outputShape:r}=Fd.sparseReshape(S("inputIndices",e,t,n),S("inputShape",e,t,n),S("newShape",e,t,n));return[s,r]}case"SparseSegmentMean":return[Fd.sparseSegmentMean(S("data",e,t,n),S("indices",e,t,n),S("segmentIds",e,t,n))];case"SparseSegmentSum":return[Fd.sparseSegmentSum(S("data",e,t,n),S("indices",e,t,n),S("segmentIds",e,t,n))];default:throw TypeError(`Node type ${e.op} is not implemented`)}},OU=(e,t,n)=>{switch(e.op){case"FFT":return[bf(S("x",e,t,n))];case"IFFT":return[Dd(S("x",e,t,n))];case"RFFT":return[vf(S("x",e,t,n))];case"IRFFT":return[f1(S("x",e,t,n))];default:throw TypeError(`Node type ${e.op} is not implemented`)}},MU=(e,t,n)=>{switch(e.op){case"StringNGrams":{let{nGrams:s,nGramsSplits:r}=Tf.stringNGrams(S("data",e,t,n),S("dataSplits",e,t,n),S("separator",e,t,n),S("nGramWidths",e,t,n),S("leftPad",e,t,n),S("rightPad",e,t,n),S("padWidth",e,t,n),S("preserveShortSequences",e,t,n));return[s,r]}case"StringSplit":{let{indices:s,values:r,shape:a}=Tf.stringSplit(S("input",e,t,n),S("delimiter",e,t,n),S("skipEmpty",e,t,n));return[s,r,a]}case"StringToHashBucketFast":return[Tf.stringToHashBucketFast(S("input",e,t,n),S("numBuckets",e,t,n))];default:throw TypeError(`Node type ${e.op} is not implemented`)}},zU=(e,t,n)=>{switch(e.op){case"Cast":return[de(S("x",e,t,n),S("dtype",e,t,n))];case"ExpandDims":{let s=S("axis",e,t,n);return[mn(S("x",e,t,n),s)]}case"Squeeze":{let s=S("axis",e,t,n);return[ot(S("x",e,t,n),s)]}case"Reshape":return[G(S("x",e,t,n),S("shape",e,t,n))];case"MirrorPad":return[gv(S("x",e,t,n),S("padding",e,t,n),S("mode",e,t,n))];case"PadV2":case"Pad":return[qs(S("x",e,t,n),S("padding",e,t,n),S("constantValue",e,t,n))];case"SpaceToBatchND":{let s=S("blockShape",e,t,n),r=S("paddings",e,t,n);return[Af(S("x",e,t,n),s,r)]}case"BatchToSpaceND":{let s=S("blockShape",e,t,n),r=S("crops",e,t,n);return[lf(S("x",e,t,n),s,r)]}case"DepthToSpace":{let s=S("blockSize",e,t,n),r=S("dataFormat",e,t,n).toUpperCase();return[tv(S("x",e,t,n),s,r)]}case"BroadcastTo":return[Id(S("x",e,t,n),S("shape",e,t,n))];case"BroadcastArgs":return[q3(S("s0",e,t,n),S("s1",e,t,n))];default:throw TypeError(`Node type ${e.op} is not implemented`)}};function y7(e,t,n,s){let r=((a,o,i)=>{switch(a.category){case"arithmetic":return q(()=>fU(a,o,i));case"basic_math":return q(()=>mU(a,o,i));case"control":return vU(a,o,i);case"convolution":return q(()=>wU(a,o,i));case"creation":return q(()=>kU(a,o,i));case"dynamic":return SU(a,o,i);case"evaluation":return q(()=>IU(a,o,i));case"image":return q(()=>EU(a,o,i));case"graph":return q(()=>CU(a,o,i));case"logical":return q(()=>RU(a,o,i));case"matrices":return q(()=>$U(a,o,i));case"normalization":return q(()=>_U(a,o,i));case"reduction":return q(()=>DU(a,o,i));case"slice_join":return q(()=>PU(a,o,i));case"sparse":return q(()=>FU(a,o,i));case"spectral":return q(()=>OU(a,o,i));case"string":return q(()=>MU(a,o,i));case"transformation":return q(()=>zU(a,o,i));case"hash_table":return NU(a,o,i,s);case"custom":let l=jk(a.op);if(l&&l.customExecutor)return l.customExecutor(new hU(a,o,i));throw TypeError(`Custom op ${a.op} is not registered.`);default:throw TypeError(`Unknown op '${a.op}'. File an issue at https://github.com/tensorflow/tfjs/issues so we can add it, or register a custom execution with tf.registerOp()`)}})(e,t,n);return v.isPromise(r)?r.then(a=>[].concat(a)):[].concat(r)}var x7=class{constructor(e={},t={},n={},s={}){this.weightMap=e,this.tensorArrayMap=t,this.tensorListMap=n,this.functionMap=s,this.rootContext={id:0,frameName:"",iterationId:0},this.contexts=[this.rootContext],this.lastId=0,this.generateCurrentContextIds()}newFrame(e,t){return{id:e,frameName:t,iterationId:0}}set currentContext(e){this.contexts!==e&&(this.contexts=e,this.generateCurrentContextIds())}get currentContext(){return this.contexts}get currentContextId(){return this._currentContextIds[0]}get currentContextIds(){return this._currentContextIds}generateCurrentContextIds(){let e=[];for(let t=0;t<this.contexts.length-1;t++){let n=this.contexts.slice(0,this.contexts.length-t);e.push(this.contextIdforContexts(n))}e.push(""),this._currentContextIds=e}contextIdforContexts(e){return e?e.map(t=>t.id===0&&t.iterationId===0?"":`${t.frameName}-${t.iterationId}`).join("/"):""}enterFrame(e){this.contexts&&(this.lastId++,this.contexts=this.contexts.slice(),this.contexts.push(this.newFrame(this.lastId,e)),this._currentContextIds.unshift(this.contextIdforContexts(this.contexts)))}exitFrame(){if(this.contexts&&this.contexts.length>1)this.contexts=this.contexts.slice(),this.contexts.splice(-1),this.currentContextIds.shift();else throw new Error("Cannot exit frame, the context is empty")}nextIteration(){if(this.contexts&&this.contexts.length>0){this.contexts=this.contexts.slice(),this.lastId++;let e=Object.assign({},this.contexts[this.contexts.length-1]);e.iterationId+=1,e.id=this.lastId,this.contexts.splice(-1,1,e),this._currentContextIds.splice(0,1,this.contextIdforContexts(this.contexts))}else throw new Error("Cannot increase frame iteration, the context is empty")}getWeight(e){return this.weightMap[e]}addTensorArray(e){this.tensorArrayMap[e.id]=e}getTensorArray(e){return this.tensorArrayMap[e]}addTensorList(e){this.tensorListMap[e.id]=e}getTensorList(e){return this.tensorListMap[e]}dispose(e){for(let t in this.tensorArrayMap)this.tensorArrayMap[t].clearAndClose(e);for(let t in this.tensorListMap)this.tensorListMap[t].clearAndClose(e)}};function b7(e,t,n,s){let r=new Set,a=[],o=null,i=null,l=new Set,c=Object.keys(e).map(p=>fs(p)[0]),u=[];s!=null&&(u=s.map(p=>fs(p.name)[0]));let d=[...t];for(;d.length>0;){let p=d.pop();if((v7(p)||UU(p)||GU(p))&&o==null&&(o=p,i=o.children.map(h=>h.name).filter(h=>r.has(h))),r.add(p.name),n[p.name]==null&&c.indexOf(p.name)===-1&&u.indexOf(p.name)===-1){if(p.inputs.length===0){a.push(p.name);continue}p.inputs.forEach(h=>{l.has(h.name)||(l.add(h.name),d.push(h))})}}return{inputs:e,outputs:t,usedNodes:r,missingInputs:a,dynamicNode:o,syncInputs:i}}function LU(e,t,n){let{usedNodes:s,inputs:r}=n,a=[],o=Object.keys(r).map(u=>fs(u)[0]).map(u=>e.nodes[u]),i=e.initNodes;o.forEach(u=>{s.has(u.name)&&a.push(u)}),e.weights.forEach(u=>{s.has(u.name)&&a.push(u)}),i!=null&&i.forEach(u=>{s.has(u.name)&&a.push(u)});let l=new Set,c=[];for(;a.length>0;){let u=a.pop();l.add(u.name),t[u.name]||c.push(u),u.children.forEach(d=>{!l.has(d.name)&&s.has(d.name)&&d.inputs.every(p=>l.has(p.name))&&a.push(d)})}return c}var BU=["Switch","Merge","Enter","Exit","NextIteration","StatelessIf","StatelessWhile","if","While"],WU=["NonMaxSuppressionV2","NonMaxSuppressionV3","NonMaxSuppressionV5","Where"],VU=["HashTable","HashTableV2","LookupTableImport","LookupTableImportV2","LookupTableFind","LookupTableFindV2","LookupTableSize","LookupTableSizeV2"];function v7(e){return BU.indexOf(e.op)>=0}function UU(e){return WU.indexOf(e.op)>=0}function GU(e){return VU.indexOf(e.op)>=0}var Ny=class{constructor(e,t){this.graph=e,this.parent=t,this.compiledMap=new Map,this._weightMap={},this.SEPERATOR=",",this._functions={},this._functionExecutorMap={},this.intermediateTensors={},this.keepTensorForDebug=!1,this._outputs=e.outputs,this._inputs=e.inputs,this._initNodes=e.initNodes,this._signature=e.signature,this._functions=e.functions,e.functions!=null&&Object.keys(e.functions).forEach(n=>{this._functionExecutorMap[n]=new Ny(e.functions[n],this)})}get weightIds(){return this.parent?this.parent.weightIds:this._weightIds}get functionExecutorMap(){return this.parent?this.parent.functionExecutorMap:this._functionExecutorMap}get weightMap(){return this.parent?this.parent.weightMap:this._weightMap}set weightMap(e){let t=Object.keys(e).map(n=>e[n].map(s=>s.id));this._weightIds=[].concat(...t),this._weightMap=e}set resourceManager(e){this._resourceManager=e}get inputs(){return this._inputs.map(e=>({name:e.name,shape:e.attrParams.shape?e.attrParams.shape.value:void 0,dtype:e.attrParams.dtype?e.attrParams.dtype.value:void 0}))}get outputs(){return this._outputs.map(e=>({name:e.name,shape:e.attrParams.shape?e.attrParams.shape.value:void 0,dtype:e.attrParams.dtype?e.attrParams.dtype.value:void 0}))}get inputNodes(){return this._inputs.map(e=>e.signatureKey||e.name)}get outputNodes(){return this._outputs.map(e=>{let t=e.signatureKey||e.name;return e.defaultOutput?`${t}:${e.defaultOutput}`:t})}get functions(){return Object.keys(this._functions).reduce((e,t)=>(e[t]=this._functions[t].signature,e),{})}getCompilationKey(e,t){let n=e.map(r=>r.name).sort(),s=t.map(r=>r.name).sort();return n.join(this.SEPERATOR)+"--"+s.join(this.SEPERATOR)}compile(e,t){let n=b7(e,t,this.weightMap,this._initNodes),{missingInputs:s,dynamicNode:r,syncInputs:a}=n;if(r!=null)throw new Error(`This execution contains the node '${r.name}', which has the dynamic op '${r.op}'. Please use model.executeAsync() instead. Alternatively, to avoid the dynamic ops, specify the inputs [${a}]`);if(s.length>0){let o=t.map(l=>l.name),i=Object.keys(e);throw new Error(`Cannot compute the outputs [${o}] from the provided inputs [${i}]. Missing the following inputs: [${s}]`)}return LU(this.graph,this.weightMap,n)}execute(e,t){e=this.mapInputs(e);let n=Object.keys(e).sort();this.checkInputs(e),this.checkInputShapeAndType(e),t=this.mapOutputs(t),this.checkOutputs(t);let s=n.map(u=>this.graph.nodes[fs(u)[0]]),r=t.map(u=>fs(u)[0]),a=r.map(u=>this.graph.nodes[u]);this.resetIntermediateTensors(),a.length===0&&(a=this._outputs);let o=this.getCompilationKey(s,a),i=this.compiledMap.get(o);i==null&&(i=this.compile(e,a),this.compiledMap.set(o,i));let l={},c={};return q(()=>{let u=new x7(this.weightMap,l,c,this.functionExecutorMap),d={...this.weightMap};Object.keys(e).forEach(f=>{let[m,g]=fs(f),A=[];A[g]=e[f],d[m]=A});let p=this.getFrozenTensorIds(d),h={};for(let f=0;f<i.length;f++){let m=i[f];if(!d[m.name]){let g=y7(m,d,u,this._resourceManager);if(v.isPromise(g))throw new Error(`The execution of the op '${m.op}' returned a promise. Please use model.executeAsync() instead.`);d[m.name]=g,this.checkTensorForDisposal(m.name,m,d,u,p,r,h)}}return this.parent==null&&u.dispose(p),t.map(f=>Vn(f,d,u))})}getFrozenTensorIds(e){let t=[].concat.apply([],Object.keys(e).map(n=>e[n]).map(n=>n.map(s=>s.id)));return new Set(t)}checkTensorForDisposal(e,t,n,s,r,a,o){t.category==="control"||a.indexOf(e)!==-1||(n[e].forEach(i=>{i!=null&&(o[i.id]=(o[i.id]||0)+t.children.length)}),t.inputs.forEach(i=>{if(i.category!=="control"){let l=jV(i.name,n,s);l!=null&&l.forEach(c=>{if(c&&!c.kept&&!r.has(c.id)){let u=o[c.id];if(u===1){if(!this.keepTensorForDebug)c.dispose();else{let[d,p]=Or(t.name,s);this.intermediateTensors[d]?this.intermediateTensors[d][p]=c:(this.intermediateTensors[d]=[],this.intermediateTensors[d][p]=c)}delete o[c.id]}else u!=null&&o[c.id]--}})}}))}async executeAsync(e,t){return this._executeAsync(e,t)}disposeIntermediateTensors(){!this.intermediateTensors||(Object.keys(this.intermediateTensors).forEach(e=>this.intermediateTensors[e].forEach(t=>t.dispose())),this.disposeTensorsMap())}disposeTensorsMap(){!this.tensorsMap||Object.keys(this.tensorsMap).forEach(e=>{this.tensorsMap[e].forEach(n=>{n&&!n.kept&&!n.isDisposed&&!this.keepIds.has(n.id)&&n.dispose()})})}getIntermediateTensors(){return this.tensorsMap}resetIntermediateTensors(){for(let e in this.intermediateTensors)this.intermediateTensors[e].forEach(t=>t.dispose()),delete this.intermediateTensors[e]}async _executeAsync(e,t,n=!1,s={},r={}){n||(e=this.mapInputs(e),this.checkInputs(e),this.checkInputShapeAndType(e),t=this.mapOutputs(t),this.checkOutputs(t));try{this.keepTensorForDebug=K().getBool("KEEP_INTERMEDIATE_TENSORS")}catch(c){console.warn(c.message)}this.resetIntermediateTensors();let a=new x7(this.weightMap,s,r,this.functionExecutorMap);this.tensorsMap=await this.executeWithControlFlow(e,a,t,n);let o=t.map(c=>Vn(c,this.tensorsMap,a)),i=o.map(c=>c.id),l=Object.keys(e).map(c=>e[c].id);return this.keepIds=new Set([...i,...l,...this.weightIds]),this.keepTensorForDebug||this.disposeTensorsMap(),this.parent==null&&a.dispose(this.keepIds),o}async executeFunctionAsync(e,t,n){let s=e.reduce((r,a,o)=>(r[this.inputs[o].name]=a,r),{});return this._executeAsync(s,this.outputNodes,!0,t,n)}async executeWithControlFlow(e,t,n,s){let r=Object.keys(e),a=r.map(x=>this.graph.nodes[fs(x)[0]]),o=n.map(x=>fs(x)[0]),i=o.map(x=>this.graph.nodes[x]);i.length===0&&(i=this._outputs);let{usedNodes:l,missingInputs:c,dynamicNode:u,syncInputs:d}=b7(e,i,this.weightMap,this._initNodes),p=[...a,...this.graph.weights,...this._initNodes||[]].map(x=>({node:x,contexts:t.currentContext})),h={...this.weightMap};Object.keys(e).forEach(x=>{let[y,b]=fs(x),w=[];w[b]=e[x],h[y]=w});let f={},m=this.getFrozenTensorIds(h),g={};for(;p.length>0;){let x=this.processStack(a,p,t,h,g,m,o,f,l);await Promise.all(x)}u==null&&!s&&console.warn("This model execution did not contain any nodes with control flow or dynamic output shapes. You can use model.execute() instead.");let A=i.filter(x=>!v7(x)&&!Vn(x.name,h,t)).map(x=>x.name);if(A.length>0){let x="";throw u!=null&&(x=`Alternatively, to avoid the dynamic ops, use model.execute() and specify the inputs [${d}]`),new Error(`Cannot compute the outputs [${A}] from the provided inputs [${r}]. Consider providing the following inputs: [${c}]. ${x}`)}return h}processStack(e,t,n,s,r,a,o,i,l){let c=[];for(;t.length>0;){let u=t.pop();n.currentContext=u.contexts;let d="";if(u.node.op==="Enter"&&S("isConstant",u.node,s,n)&&([d]=Or(u.node.name,n)),s[u.node.name]==null){let p=y7(u.node,s,n,this._resourceManager);d||([d]=Or(u.node.name,n));let h=n.currentContext;v.isPromise(p)?c.push(p.then(f=>(s[d]=f,n.currentContext=h,this.checkTensorForDisposal(d,u.node,s,n,a,o,i),this.processChildNodes(u.node,t,n,s,r,l),f))):(s[d]=p,this.checkTensorForDisposal(d,u.node,s,n,a,o,i),this.processChildNodes(u.node,t,n,s,r,l))}else this.processChildNodes(u.node,t,n,s,r,l)}return c}processChildNodes(e,t,n,s,r,a){e.children.forEach(o=>{let[i]=Or(o.name,n);r[i]||!a.has(o.name)||(o.op==="Merge"?o.inputNames.some(l=>!!Vn(l,s,n))&&(r[i]=!0,t.push({contexts:n.currentContext,node:o})):o.inputNames.every(l=>!!Vn(l,s,n))&&(r[i]=!0,t.push({contexts:n.currentContext,node:o})))})}dispose(){Object.keys(this.weightMap).forEach(e=>this.weightMap[e].forEach(t=>t.dispose()))}checkInputShapeAndType(e){Object.keys(e).forEach(t=>{let n=e[t],[s]=fs(t),r=this.graph.nodes[s];if(r.attrParams.shape&&r.attrParams.shape.value){let a=r.attrParams.shape.value,o=a.length===n.shape.length&&n.shape.every((i,l)=>a[l]===-1||a[l]===i);v.assert(o,()=>`The shape of dict['${r.name}'] provided in model.execute(dict) must be [${a}], but was [${n.shape}]`)}r.attrParams.dtype&&r.attrParams.dtype.value&&v.assert(n.dtype===r.attrParams.dtype.value,()=>`The dtype of dict['${r.name}'] provided in model.execute(dict) must be ${r.attrParams.dtype.value}, but was ${n.dtype}`)})}mapInputs(e){let t={};for(let n in e)if(this._signature!=null&&this._signature.inputs!=null&&this._signature.inputs[n]!=null){let s=this._signature.inputs[n];t[s.name]=e[n]}else t[n]=e[n];return t}checkInputs(e){let t=Object.keys(e).filter(n=>{let[s]=fs(n);return this.graph.nodes[s]==null});if(t.length>0)throw new Error(`The dict provided in model.execute(dict) has keys: [${t}] that are not part of graph`)}mapOutputs(e){return e.map(t=>this._signature!=null&&this._signature.outputs!=null&&this._signature.outputs[t]!=null?this._signature.outputs[t].name:t,{})}checkOutputs(e){e.forEach(t=>{let[n]=fs(t);if(!this.graph.nodes[n])throw new Error(`The output '${t}' is not found in the graph`)})}},HU=class{constructor(e={},t={}){this.hashTableNameToHandle=e,this.hashTableMap=t}addHashTable(e,t){this.hashTableNameToHandle[e]=t.handle,this.hashTableMap[t.id]=t}getHashTableHandleByName(e){return this.hashTableNameToHandle[e]}getHashTableById(e){return this.hashTableMap[e]}dispose(){for(let e in this.hashTableMap)this.hashTableMap[e].clearAndClose(),delete this.hashTableMap[e];for(let e in this.hashTableNameToHandle)this.hashTableNameToHandle[e].dispose(),delete this.hashTableNameToHandle[e]}},jU="?tfjs-format=file",qU="model.json",w7=class{constructor(e,t={}){this.modelUrl=e,this.loadOptions=t,this.version="n/a",t==null&&(this.loadOptions={}),this.resourceManager=new HU}get modelVersion(){return this.version}get inputNodes(){return this.executor.inputNodes}get outputNodes(){return this.executor.outputNodes}get inputs(){return this.executor.inputs}get outputs(){return this.executor.outputs}get weights(){return this.executor.weightMap}get metadata(){return this.artifacts.userDefinedMetadata}get modelSignature(){return this.signature}findIOHandler(){let e=this.modelUrl;if(e.load!=null)this.handler=e;else if(this.loadOptions.requestInit!=null)this.handler=ts.browserHTTPRequest(e,this.loadOptions);else{let t=ts.getLoadHandlers(e,this.loadOptions);if(t.length===0)t.push(ts.browserHTTPRequest(e,this.loadOptions));else if(t.length>1)throw new Error(`Found more than one (${t.length}) load handlers for URL '${[e]}'`);this.handler=t[0]}}async load(){if(this.findIOHandler(),this.handler.load==null)throw new Error("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");let e=await this.handler.load();return this.loadSync(e)}loadSync(e){this.artifacts=e;let t=this.artifacts.modelTopology,n;this.artifacts.userDefinedMetadata!=null&&this.artifacts.userDefinedMetadata.signature!=null?n=this.artifacts.userDefinedMetadata.signature:n=this.artifacts.signature,this.signature=n,this.version=`${t.versions.producer}.${t.versions.minConsumer}`;let s=ts.decodeWeights(this.artifacts.weightData,this.artifacts.weightSpecs);if(this.executor=new Ny(p7.Instance.transformGraph(t,this.signature)),this.executor.weightMap=this.convertTensorMapToTensorsMap(s),this.executor.resourceManager=this.resourceManager,e.modelInitializer!=null&&e.modelInitializer.node!=null){let r=p7.Instance.transformGraph(e.modelInitializer);this.initializer=new Ny(r),this.initializer.weightMap=this.executor.weightMap,this.initializer.resourceManager=this.resourceManager,this.initializer.executeAsync({},[])}return!0}async save(e,t){if(typeof e=="string"){let n=ts.getSaveHandlers(e);if(n.length===0)throw new Error(`Cannot find any save handlers for URL '${e}'`);if(n.length>1)throw new Error(`Found more than one (${n.length}) save handlers for URL '${e}'`);e=n[0]}if(e.save==null)throw new Error("GraphModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");return e.save(this.artifacts)}predict(e,t){return this.execute(e,this.outputNodes)}normalizeInputs(e){if(!(e instanceof Ze)&&!Array.isArray(e))return e;if(e=Array.isArray(e)?e:[e],e.length!==this.inputNodes.length)throw new Error(`Input tensor count mismatch,the graph model has ${this.inputNodes.length} placeholders, while there are ${e.length} input tensors.`);return this.inputNodes.reduce((t,n,s)=>(t[n]=e[s],t),{})}normalizeOutputs(e){return e=e||this.outputNodes,Array.isArray(e)?e:[e]}execute(e,t){e=this.normalizeInputs(e),t=this.normalizeOutputs(t);let n=this.executor.execute(e,t);return n.length>1?n:n[0]}async executeAsync(e,t){e=this.normalizeInputs(e),t=this.normalizeOutputs(t);let n=await this.executor.executeAsync(e,t);return n.length>1?n:n[0]}getIntermediateTensors(){return this.executor.getIntermediateTensors()}disposeIntermediateTensors(){this.executor.disposeIntermediateTensors()}convertTensorMapToTensorsMap(e){return Object.keys(e).reduce((t,n)=>(t[n]=[e[n]],t),{})}dispose(){this.executor.dispose(),this.initializer&&this.initializer.dispose(),this.resourceManager.dispose()}};async function et(e,t={}){if(e==null)throw new Error("modelUrl in loadGraphModel() cannot be null. Please provide a url or an IOHandler that loads the model");t==null&&(t={}),t.fromTFHub&&e.load==null&&(e.endsWith("/")||(e=e+"/"),e=`${e}${qU}${jU}`);let n=new w7(e,t);return await n.load(),n}var XU="0.0.0",k7={};Oe(k7,{CSVDataset:()=>M7,Dataset:()=>Yu,FileDataSource:()=>G7,TextLineDataset:()=>P7,URLDataSource:()=>H7,array:()=>AG,csv:()=>NG,func:()=>EG,generator:()=>RG,microphone:()=>_G,version_data:()=>DG,webcam:()=>$G,zip:()=>yG});var KU=ii(ch()),ZU=ii(ch());function YU(e,t){return gm(e,t)}function gm(e,t,n=new Map,s=new Set){if(e==null)return null;if(typeof Blob=="function"&&e instanceof Blob)return e.slice();if(s.has(e))throw new Error("Circular references are not supported.");if(n.has(e))return n.get(e);let r=t(e);if(r.recurse&&r.value!==null)throw new Error("A deep map function may not return both a value and recurse=true.");if(r.recurse)if(Zu(e)){let a=Array.isArray(e)?[]:{};s.add(e);for(let o in e){let i=e[o],l=gm(i,t,n,s);a[o]=l}return s.delete(e),e.__proto__&&(a.__proto__=e.__proto__),a}else throw new Error(`Can't recurse into non-iterable type: ${e}`);else return n.set(e,r.value),r.value}function JU(e,t=I7){return S7(e,t)}function S7(e,t,n=new Set){let s=e[0];if(n.has(s))throw new Error("Circular references are not supported.");let r=t(e);if(r.recurse&&r.value!==null)throw new Error("A deep zip function may not return both a value and recurse=true.");if(r.recurse)if(Zu(s)){let a=Array.isArray(s)?[]:{};n.add(s);for(let o in s){let i=e.map(c=>c[o]),l=S7(i,t,n);a[o]=l}return n.delete(s),a}else throw new Error(`Can't recurse into non-iterable type: ${s}`);else return r.value}function I7(e){return e===null?null:Zu(e[0])?{value:null,recurse:!0}:{value:e,recurse:!1}}async function C7(e,t){let n=new Map;gm(e,t,n);for(let r of Array.from(n.keys())){let a=n.get(r);if(v.isPromise(a)){let o=await a;n.set(r,o)}}return gm(e,t,n)}function Zu(e){let t=!1;if(K().get("IS_BROWSER"))t=e instanceof TextDecoder;else{let{StringDecoder:n}=A5();t=e instanceof n}return e!=null&&!ArrayBuffer.isView(e)&&(Array.isArray(e)||typeof e=="object"&&!(e instanceof Ze)&&!(e instanceof Promise)&&!t)}function QU(e){return e==null||eG(e)||Array.isArray(e)||typeof e=="object"&&e instanceof Ze||v.isTypedArray(e)}function eG(e){return e===null||typeof e!="object"&&typeof e!="function"}function tG(e){return YU(e,nG)}function nG(e){return e instanceof Ze?{value:e.clone(),recurse:!1}:Zu(e)?{value:null,recurse:!0}:{value:e,recurse:!1}}var T7=class{constructor(e){if(this.capacity=e,this.begin=0,this.end=0,e==null)throw new RangeError("Can't create a ring buffer of unknown capacity.");if(e<1)throw new RangeError("Can't create ring buffer of capacity < 1.");this.data=new Array(e),this.doubledCapacity=2*e}wrap(e){for(;e<0;)e+=this.doubledCapacity;return e%this.doubledCapacity}get(e){if(e<0)throw new RangeError("Can't get item at a negative index.");return this.data[e%this.capacity]}set(e,t){if(e<0)throw new RangeError("Can't set item at a negative index.");this.data[e%this.capacity]=t}length(){let e=this.end-this.begin;return e<0&&(e=this.doubledCapacity+e),e}isFull(){return this.length()===this.capacity}isEmpty(){return this.length()===0}push(e){if(this.isFull())throw new RangeError("Ring buffer is full.");this.set(this.end,e),this.end=this.wrap(this.end+1)}pushAll(e){for(let t of e)this.push(t)}pop(){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");this.end=this.wrap(this.end-1);let e=this.get(this.end);return this.set(this.end,void 0),e}unshift(e){if(this.isFull())throw new RangeError("Ring buffer is full.");this.begin=this.wrap(this.begin-1),this.set(this.begin,e)}shift(){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");let e=this.get(this.begin);return this.set(this.begin,void 0),this.begin=this.wrap(this.begin+1),e}shuffleExcise(e){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");let t=this.wrap(this.begin+e),n=this.get(t);return this.set(t,this.pop()),n}},N7=class extends T7{constructor(){super(N7.INITIAL_CAPACITY)}isFull(){return!1}push(e){super.isFull()&&this.expand(),super.push(e)}unshift(e){super.isFull()&&this.expand(),super.unshift(e)}expand(){let e=this.capacity*2,t=new Array(e),n=this.length();for(let s=0;s<n;s++)t[s]=this.get(this.wrap(this.begin+s));this.data=t,this.capacity=e,this.doubledCapacity=2*this.capacity,this.begin=0,this.end=n}},E7=N7;E7.INITIAL_CAPACITY=32;function R7(e){return new aG(e)}function Ey(e){return new oG(e)}function sG(e,t){return new _7(e,t)}function rG(e,t=Am.FAIL){return new mG(e,t)}var xn=class{async toArray(){let e=[],t=await this.next();for(;!t.done;)e.push(t.value),t=await this.next();return e}async toArrayForTest(){let e=this.prefetch(100),t=[],n=await e.next();for(;!n.done;)t.push(n.value),n=await e.next();return t}async resolveFully(){let e=await this.next();for(;!e.done;)e=await this.next()}async resolveWhile(e){let t=await this.next(),n=e(t.value);for(;!t.done&&n;)t=await this.next(),n=e(t.value)}handleErrors(e){return new hG(this,e)}filter(e){return new dG(this,e)}map(e){return new pG(this,e)}mapAsync(e){return new $7(this,e)}serialMapAsync(e){return new $7(this,e).serial()}flatmap(e){return new fG(this,e)}async forEachAsync(e){return this.map(e).resolveFully()}async serialForEach(e){return this.serialMapAsync(e).resolveWhile(t=>t===!0)}rowMajorBatch(e,t=!0){return new cG(this,e,t)}columnMajorBatch(e,t=!0,n=I7){return this.rowMajorBatch(e,t).map(r=>JU(r,n))}concatenate(e,t){return new _7(R7([this,e]),t)}take(e){return e<0||e==null?this:new uG(this,e)}skip(e){return e<0||e==null?this:new lG(this,e)}prefetch(e){return new D7(this,e)}shuffle(e,t){return new gG(this,e,t)}serial(){return new iG(this)}},aG=class extends xn{constructor(e){super();this.items=e,this.trav=0}summary(){return`Array of ${this.items.length} items`}async next(){if(this.trav>=this.items.length)return{value:null,done:!0};let e=this.items[this.trav];return this.trav++,{value:tG(e),done:!1}}},oG=class extends xn{constructor(e){super();this.nextFn=e}summary(){return"Function call"}async next(){try{return this.nextFn()}catch(e){throw e.message=`Error thrown while iterating through a dataset: ${e.message}`,e}}},iG=class extends xn{constructor(e){super();this.upstream=e,this.lastRead=Promise.resolve({value:null,done:!1})}summary(){return`${this.upstream.summary()} -> Serial`}async next(){return this.lastRead=this.lastRead.then(()=>this.serialNext()),this.lastRead}async serialNext(){return this.upstream.next()}},lG=class extends xn{constructor(e,t){super();this.upstream=e,this.maxCount=t,this.count=0,this.lastRead=Promise.resolve({value:null,done:!1})}summary(){return`${this.upstream.summary()} -> Skip`}async next(){return this.lastRead=this.lastRead.then(()=>this.serialNext()),this.lastRead}async serialNext(){for(;this.count++<this.maxCount;){let e=await this.upstream.next();if(e.done)return e;se(e.value)}return this.upstream.next()}},uG=class extends xn{constructor(e,t){super();this.upstream=e,this.maxCount=t,this.count=0}summary(){return`${this.upstream.summary()} -> Take`}async next(){return this.count++>=this.maxCount?{value:null,done:!0}:this.upstream.next()}},cG=class extends xn{constructor(e,t,n=!0){super();this.upstream=e,this.batchSize=t,this.enableSmallLastBatch=n,this.lastRead=Promise.resolve({value:null,done:!1})}summary(){return`${this.upstream.summary()} -> RowMajorBatch`}async next(){return this.lastRead=this.lastRead.then(()=>this.serialNext()),this.lastRead}async serialNext(){let e=[];for(;e.length<this.batchSize;){let t=await this.upstream.next();if(t.done)return this.enableSmallLastBatch&&e.length>0?{value:e,done:!1}:{value:null,done:!0};e.push(t.value)}return{value:e,done:!1}}},dG=class extends xn{constructor(e,t){super();this.upstream=e,this.predicate=t,this.lastRead=Promise.resolve({value:null,done:!1})}summary(){return`${this.upstream.summary()} -> Filter`}async next(){return this.lastRead=this.lastRead.then(()=>this.serialNext()),this.lastRead}async serialNext(){for(;;){let e=await this.upstream.next();if(e.done||this.predicate(e.value))return e;se(e.value)}}},pG=class extends xn{constructor(e,t){super();this.upstream=e,this.transform=t}summary(){return`${this.upstream.summary()} -> Map`}async next(){let e=await this.upstream.next();if(e.done)return{value:null,done:!0};let t=ar.getTensorsInContainer(e.value),n=this.transform(e.value),s=ar.getTensorsInContainer(n);for(let r of t)ar.isTensorInList(r,s)||r.dispose();return{value:n,done:!1}}},hG=class extends xn{constructor(e,t){super();this.upstream=e,this.handler=t,this.count=0,this.lastRead=Promise.resolve({value:null,done:!1})}summary(){return`${this.upstream.summary()} -> handleErrors`}async next(){return this.lastRead=this.lastRead.then(()=>this.serialNext()),this.lastRead}async serialNext(){for(;;)try{return await this.upstream.next()}catch(e){if(!this.handler(e))return{value:null,done:!0}}}},$7=class extends xn{constructor(e,t){super();this.upstream=e,this.transform=t}summary(){return`${this.upstream.summary()} -> AsyncMap`}async next(){let e=await this.upstream.next();if(e.done)return{value:null,done:!0};let t=ar.getTensorsInContainer(e.value),n=await this.transform(e.value),s=ar.getTensorsInContainer(n);for(let r of t)ar.isTensorInList(r,s)||r.dispose();return{value:n,done:!1}}},Ry=class extends xn{constructor(){super();this.outputQueue=new E7,this.lastRead=Promise.resolve({value:null,done:!1})}async next(){return this.lastRead=this.lastRead.then(()=>this.serialNext()),this.lastRead}async serialNext(){for(;this.outputQueue.length()===0;)if(!await this.pump())return{value:null,done:!0};return{value:this.outputQueue.shift(),done:!1}}},fG=class extends Ry{constructor(e,t){super();this.upstream=e,this.transform=t}summary(){return`${this.upstream.summary()} -> Flatmap`}async pump(){let e=await this.upstream.next();if(e.done)return!1;let t=ar.getTensorsInContainer(e.value),n=this.transform(e.value),s=ar.getTensorsInContainer(n);this.outputQueue.pushAll(n);for(let r of t)ar.isTensorInList(r,s)||r.dispose();return!0}},_7=class extends xn{constructor(e,t){super();this.baseErrorHandler=t,this.lastRead=null,this.iterator=null,this.moreIterators=e}summary(){return"TODO: fill in upstream of chained summaries -> Chained"}async next(){return this.lastRead=this.readFromChain(this.lastRead),this.lastRead}async readFromChain(e){if(await e,this.iterator==null){let n=await this.moreIterators.next();if(n.done)return{value:null,done:!0};this.iterator=n.value,this.baseErrorHandler!=null&&(this.iterator=this.iterator.handleErrors(this.baseErrorHandler))}let t=await this.iterator.next();return t.done?(this.iterator=null,this.readFromChain(e)):t}},Am;(function(e){e[e.FAIL=0]="FAIL",e[e.SHORTEST=1]="SHORTEST",e[e.LONGEST=2]="LONGEST"})(Am||(Am={}));var mG=class extends xn{constructor(e,t=0){super();this.iterators=e,this.mismatchMode=t,this.count=0,this.currentPromise=null}summary(){return"{TODO: fill in upstream of zip summaries} -> Zip"}async nextState(e){await e;let t=0,n=0;function s(a){return a instanceof xn?{value:a.next().then(i=>(t++,i.done&&n++,i.value)),recurse:!1}:{value:null,recurse:!0}}let r=await C7(this.iterators,s);if(t===n)return{value:null,done:!0};if(n>0)switch(this.mismatchMode){case 0:throw new Error(`Zipped streams should have the same length. Mismatched at element ${this.count}.`);case 1:return{value:null,done:!0};case 2:default:}return this.count++,{value:r,done:!1}}async next(){return this.currentPromise=this.nextState(this.currentPromise),this.currentPromise}},D7=class extends xn{constructor(e,t){super();this.upstream=e,this.bufferSize=t,this.buffer=new T7(t)}summary(){return`${this.upstream.summary()} -> Prefetch`}refill(){for(;!this.buffer.isFull();){let e=this.upstream.next();this.buffer.push(e)}}next(){return this.refill(),this.buffer.shift()}},gG=class extends D7{constructor(e,t,n){super(e,t);this.upstream=e,this.windowSize=t,this.upstreamExhausted=!1,this.random=ZU.alea(n||v.now().toString()),this.lastRead=Promise.resolve({value:null,done:!1})}async next(){return this.lastRead=this.lastRead.then(()=>this.serialNext()),this.lastRead}randomInt(e){return Math.floor(this.random()*e)}chooseIndex(){return this.randomInt(this.buffer.length())}async serialNext(){for(this.upstreamExhausted||this.refill();!this.buffer.isEmpty();){let e=this.chooseIndex(),t=await this.buffer.shuffleExcise(e);if(t.done)this.upstreamExhausted=!0;else return this.refill(),t}return{value:null,done:!0}}},Yu=class{constructor(){this.size=null}batch(e,t=!0){let n=this;v.assert(e>0,()=>`batchSize needs to be positive, but it is
|
|
${e}`);let s;return this.size===1/0||this.size==null?s=this.size:t?s=Math.ceil(this.size/e):s=Math.floor(this.size/e),ms(async()=>(await n.iterator()).columnMajorBatch(e,t,xG),s)}concatenate(e){let t=this,n;return this.size===1/0||e.size===1/0?n=1/0:this.size!=null&&e.size!=null?n=this.size+e.size:n=null,ms(async()=>(await t.iterator()).concatenate(await e.iterator()),n)}filter(e){let t=this,n;return this.size===1/0?n=1/0:n=null,ms(async()=>(await t.iterator()).filter(s=>q(()=>e(s))),n)}async forEachAsync(e){return(await this.iterator()).forEachAsync(e)}map(e){let t=this;return ms(async()=>(await t.iterator()).map(n=>q(()=>e(n))),this.size)}mapAsync(e){let t=this;return ms(async()=>(await t.iterator()).mapAsync(e),this.size)}prefetch(e){if(e==null)throw new RangeError("`Dataset.prefetch()` requires bufferSize to be specified.");let t=this;return ms(async()=>(await t.iterator()).prefetch(e),this.size)}repeat(e){let t=this,n;return this.size!=null&&e>0?n=this.size*e:e===0?n=0:this.size!=null&&(e===void 0||e<0)?n=1/0:n=null,ms(async()=>{let s=Ey(async()=>({value:await t.iterator(),done:!1}));return sG(s.take(e))},n)}skip(e){let t=this,n;return this.size!=null&&e>=0&&this.size>=e?n=this.size-e:this.size!=null&&(this.size<e||e===void 0||e<0)?n=0:n=null,ms(async()=>(await t.iterator()).skip(e),n)}shuffle(e,t,n=!0){if(e==null||e<0)throw this.size==null?new RangeError("`Dataset.shuffle()` requires bufferSize to be specified."):new RangeError(`\`Dataset.shuffle()\` requires bufferSize to be specified. If your data fits in main memory (for regular JS objects), and/or GPU memory (for \`tf.Tensor\`s), consider setting bufferSize to the dataset size (${this.size} elements)`);let s=this,r=KU.alea(t||v.now().toString());return ms(async()=>{let a=r.int32();return n&&(a+=r.int32()),(await s.iterator()).shuffle(e,a.toString())},this.size)}take(e){let t=this,n;return this.size!=null&&this.size>e?n=e:this.size!=null&&this.size<=e?n=this.size:n=null,ms(async()=>(await t.iterator()).take(e),n)}async toArray(){if(this.size===1/0)throw new Error("Can not convert infinite data stream to array.");return(await this.iterator()).toArray()}async toArrayForTest(){if(this.size===1/0)throw new Error("Can not convert infinite data stream to array.");return(await this.iterator()).toArrayForTest()}};Yu.MAX_BUFFER_SIZE=1e4;function ms(e,t=null){return new class extends Yu{constructor(){super(...arguments);this.size=t}async iterator(){return e()}}}function AG(e){return ms(async()=>R7(e),e.length)}function yG(e){if(!Zu(e))throw new Error("The argument to zip() must be an object or array.");let t;if(Array.isArray(e))for(let n=0;n<e.length;n++)t=t==null?e[n].size:Math.min(t,e[n].size);else if(e instanceof Object)for(let n in e)t=t==null?e[n].size:Math.min(t,e[n].size);return ms(async()=>{let n=await C7(e,s=>{if(s instanceof Yu)return{value:s.iterator(),recurse:!1};if(Zu(s))return{value:null,recurse:!0};throw new Error("Leaves of the structure passed to zip() must be Datasets, not primitives.")});return rG(n,Am.SHORTEST)},t)}function xG(e){if(e===null)return null;let t=e[0];return QU(t)?{value:bG(e),recurse:!1}:{value:null,recurse:!0}}function bG(e){if(e.length===0)throw new Error("Can't make a batch of zero elements.");return e[0]instanceof Ze?An(e):Ut(e)}var P7=class extends Yu{constructor(e){super();this.input=e}async iterator(){return(await this.input.iterator()).decodeUTF8().split(`
|
|
`).map(s=>(s.endsWith("\r")&&(s=s.slice(0,-1)),s))}},ym='"',tp=Symbol("out"),F7=Symbol("field"),xm=Symbol("quote"),$y=Symbol("quoteafterquote"),O7=Symbol("quoteinquote"),M7=class extends Yu{constructor(e,t){super();this.input=e,this.hasHeader=!0,this.fullColumnNames=null,this.columnNamesValidated=!1,this.columnConfigs=null,this.configuredColumnsOnly=!1,this.delimiter=",",this.delimWhitespace=!1,this.base=new P7(e),t||(t={}),this.hasHeader=t.hasHeader!==!1,this.fullColumnNames=t.columnNames,this.columnConfigs=t.columnConfigs,this.configuredColumnsOnly=t.configuredColumnsOnly,t.delimWhitespace?(v.assert(t.delimiter==null,()=>"Delimiter should not be provided when delimWhitespace is true."),this.delimWhitespace=!0,this.delimiter=" "):this.delimiter=t.delimiter?t.delimiter:","}async columnNames(){return this.columnNamesValidated||await this.setColumnNames(),this.configuredColumnsOnly?Object.keys(this.columnConfigs):this.fullColumnNames}async setColumnNames(){let e=await this.maybeReadHeaderLine();if(!this.fullColumnNames&&!e)throw new Error("Column names must be provided if there is no header line.");this.fullColumnNames&&e&&v.assert(e.length===this.fullColumnNames.length,()=>"The length of provided columnNames ("+this.fullColumnNames.length.toString()+") does not match the length of the header line read from file ("+e.length.toString()+")."),this.fullColumnNames||(this.fullColumnNames=e);let t=this.fullColumnNames.reduce((s,r)=>(s[r]=s[r]+1||1,s),{}),n=Object.keys(t).filter(s=>t[s]>1);if(v.assert(n.length===0,()=>"Duplicate column names found: "+n.toString()),this.columnConfigs){for(let s of Object.keys(this.columnConfigs))if(this.fullColumnNames.indexOf(s)===-1)throw new Error('The key "'+s+'" provided in columnConfigs does not match any of the column names ('+this.fullColumnNames.toString()+").")}this.columnNamesValidated=!0}async maybeReadHeaderLine(){if(this.hasHeader){let t=await(await this.base.iterator()).next();if(t.done)throw new Error("No data was found for CSV parsing.");let n=t.value;return this.parseRow(n,!1)}else return null}async iterator(){this.columnNamesValidated||await this.setColumnNames();let e=await this.base.iterator();return this.hasHeader&&(e=e.skip(1)),e.map(t=>this.makeDataElement(t))}makeDataElement(e){let t=this.parseRow(e),n={},s={};for(let r=0;r<this.fullColumnNames.length;r++){let a=this.fullColumnNames[r],o=this.columnConfigs?this.columnConfigs[a]:null;if(!(this.configuredColumnsOnly&&!o)){let i=t[r],l=null;if(i==="")if(o&&o.default!==void 0)l=o.default;else{if(o&&(o.required||o.isLabel))throw new Error(`Required column ${a} is empty in this line: ${e}`);l=void 0}else{let c=Number(i);if(isNaN(c))o&&o.dtype==="bool"?l=this.getBoolean(i):l=i;else if(!o||!o.dtype)l=c;else switch(o.dtype){case"float32":l=c;break;case"int32":l=Math.floor(c);break;case"bool":l=this.getBoolean(i);break;default:l=c}}o&&o.isLabel?s[a]=l:n[a]=l}}return Object.keys(s).length===0?n:{xs:n,ys:s}}getBoolean(e){return e==="1"||e.toLowerCase()==="true"?1:0}parseRow(e,t=!0){let n=[],s=0,r=e.length,a=tp;for(let o=0;o<r;o++)switch(a){case tp:switch(e.charAt(o)){case ym:s=o+1,a=xm;break;case this.delimiter:if(s=o+1,this.delimiter===" "&&this.delimWhitespace)break;n.push(""),a=tp;break;default:a=F7,s=o;break}break;case F7:switch(e.charAt(o)){case this.delimiter:n.push(e.substring(s,o)),a=tp,s=o+1;break;default:}break;case xm:switch(e.charAt(o)){case ym:a=$y;break;default:}break;case $y:switch(e.charAt(o)){case this.delimiter:n.push(e.substring(s,o-1)),a=tp,s=o+1;break;case ym:a=xm;break;default:a=O7;break}break;case O7:switch(e.charAt(o)){case ym:a=xm;break;default:}break;default:}if(a===$y?n.push(e.substring(s,r-1)):n.push(e.substring(s)),t&&n.length!==this.fullColumnNames.length)throw new Error(`Invalid row in csv file. Should have ${this.fullColumnNames.length} elements in a row, but got ${n}`);return n}},z7=class extends xn{constructor(e){super();this.microphoneConfig=e,this.isClosed=!1,this.fftSize=e.fftSize||1024;let t=Math.log2(this.fftSize);if(this.fftSize<0||t<4||t>14||!Number.isInteger(t))throw new Error(`Invalid fftSize: it must be a power of 2 between 2 to 4 and 2 to 14, but got ${this.fftSize}`);if(this.numFrames=e.numFramesPerSpectrogram||43,this.sampleRateHz=e.sampleRateHz,this.columnTruncateLength=e.columnTruncateLength||this.fftSize,this.audioTrackConstraints=e.audioTrackConstraints,this.smoothingTimeConstant=e.smoothingTimeConstant||0,this.includeSpectrogram=e.includeSpectrogram!==!1,this.includeWaveform=e.includeWaveform===!0,!this.includeSpectrogram&&!this.includeWaveform)throw new Error("Both includeSpectrogram and includeWaveform are false. At least one type of data should be returned.")}summary(){return"microphone"}static async create(e={}){if(K().get("IS_NODE"))throw new Error("microphone API is only supported in browser environment.");let t=new z7(e);return await t.start(),t}async start(){try{this.stream=await navigator.mediaDevices.getUserMedia({audio:this.audioTrackConstraints==null?!0:this.audioTrackConstraints,video:!1})}catch(n){throw new Error(`Error thrown while initializing video stream: ${n.message}`)}if(!this.stream)throw new Error("Could not obtain audio from microphone.");let e=window.AudioContext||window.webkitAudioContext;if(this.audioContext=new e,!this.sampleRateHz)this.sampleRateHz=this.audioContext.sampleRate;else if(this.audioContext.sampleRate!==this.sampleRateHz)throw new Error(`Mismatch in sampling rate: Expected: ${this.sampleRateHz}; Actual: ${this.audioContext.sampleRate}`);let t=this.audioContext.createMediaStreamSource(this.stream);this.analyser=this.audioContext.createAnalyser(),this.analyser.fftSize=this.fftSize*2,this.analyser.smoothingTimeConstant=this.smoothingTimeConstant,t.connect(this.analyser),this.freqData=new Float32Array(this.fftSize),this.timeData=new Float32Array(this.fftSize)}async next(){if(this.isClosed)return{value:null,done:!0};let e,t,n=await this.getAudioData();if(this.includeSpectrogram){let s=this.flattenQueue(n.freqDataQueue);e=this.getTensorFromAudioDataArray(s,[this.numFrames,this.columnTruncateLength,1])}if(this.includeWaveform){let s=this.flattenQueue(n.timeDataQueue);t=this.getTensorFromAudioDataArray(s,[this.numFrames*this.fftSize,1])}return{value:{spectrogram:e,waveform:t},done:!1}}async capture(){return(await this.next()).value}async getAudioData(){let e=[],t=[],n=0;return new Promise(s=>{let r=setInterval(()=>{this.includeSpectrogram&&(this.analyser.getFloatFrequencyData(this.freqData),this.freqData[0]===-1/0&&s({freqDataQueue:e,timeDataQueue:t}),e.push(this.freqData.slice(0,this.columnTruncateLength))),this.includeWaveform&&(this.analyser.getFloatTimeDomainData(this.timeData),t.push(this.timeData.slice())),++n===this.numFrames&&(clearInterval(r),s({freqDataQueue:e,timeDataQueue:t}))},this.fftSize/this.sampleRateHz*1e3)})}stop(){this.isClosed||(this.isClosed=!0,this.analyser.disconnect(),this.audioContext.close(),this.stream!=null&&this.stream.getTracks().length>0&&this.stream.getTracks()[0].stop())}toArray(){throw new Error("Can not convert infinite audio stream to array.")}getSampleRate(){return this.sampleRateHz}flattenQueue(e){let t=e[0].length,n=new Float32Array(e.length*t);return e.forEach((s,r)=>n.set(s,r*t)),n}getTensorFromAudioDataArray(e,t){let n=new Float32Array(v.sizeFromShape(t));return n.set(e,n.length-e.length),Ut(n,t)}},L7=class extends xn{constructor(e,t){super();if(this.webcamVideoElement=e,this.webcamConfig=t,this.isClosed=!0,this.resize=!1,this.needToResize())if(this.resize=!0,this.cropSize=[this.webcamConfig.resizeHeight,this.webcamConfig.resizeWidth],this.cropBoxInd=Xt([0],"int32"),this.webcamConfig.centerCrop){let n=this.webcamConfig.resizeWidth*1/this.webcamVideoElement.width,s=this.webcamConfig.resizeHeight*1/this.webcamVideoElement.height,r=(1-n)/2,a=(1-s)/2,o=r+n,i=s+a;this.cropBox=lr([a,r,i,o],[1,4])}else this.cropBox=lr([0,0,1,1],[1,4])}summary(){return"webcam"}static async create(e,t={}){if(K().get("IS_NODE"))throw new Error("tf.data.webcam is only supported in browser environment.");if(!e){if(e=document.createElement("video"),!t.resizeWidth||!t.resizeHeight)throw new Error("Please provide webcam video element, or resizeWidth and resizeHeight to create a hidden video element.");e.width=t.resizeWidth,e.height=t.resizeHeight}let n=new L7(e,t);return await n.start(),n}async start(){this.webcamConfig.facingMode&&v.assert(this.webcamConfig.facingMode==="user"||this.webcamConfig.facingMode==="environment",()=>`Invalid webcam facing mode: ${this.webcamConfig.facingMode}. Please provide 'user' or 'environment'`);try{this.stream=await navigator.mediaDevices.getUserMedia({video:{deviceId:this.webcamConfig.deviceId,facingMode:this.webcamConfig.facingMode?this.webcamConfig.facingMode:"user",width:this.webcamVideoElement.width,height:this.webcamVideoElement.height}})}catch(e){throw e.message=`Error thrown while initializing video stream: ${e.message}`,e}if(!this.stream)throw new Error("Could not obtain video from webcam.");try{this.webcamVideoElement.srcObject=this.stream}catch(e){console.log(e),this.webcamVideoElement.src=window.URL.createObjectURL(this.stream)}return this.webcamVideoElement.play(),this.isClosed=!1,new Promise(e=>{this.webcamVideoElement.onloadedmetadata=()=>{e()}})}async next(){if(this.isClosed)return{value:null,done:!0};let e;try{e=Gs.fromPixels(this.webcamVideoElement)}catch(t){throw new Error(`Error thrown converting video to pixels: ${JSON.stringify(t)}`)}if(this.resize)try{return{value:this.cropAndResizeFrame(e),done:!1}}catch(t){throw new Error(`Error thrown cropping the video: ${t.message}`)}finally{e.dispose()}else return{value:e,done:!1}}needToResize(){return!!(this.webcamConfig.resizeWidth&&this.webcamConfig.resizeHeight&&(this.webcamVideoElement.width!==this.webcamConfig.resizeWidth||this.webcamVideoElement.height!==this.webcamConfig.resizeHeight))}cropAndResizeFrame(e){return q(()=>{let t=mn(de(e,"float32"),0),n;n=$e.cropAndResize(t,this.cropBox,this.cropBoxInd,this.cropSize,"bilinear");let s=n.shape;return G(n,s.slice(1))})}async capture(){return(await this.next()).value}stop(){this.stream.getTracks().forEach(t=>t.stop());try{this.webcamVideoElement.srcObject=null}catch(t){console.log(t),this.webcamVideoElement.src=null}this.isClosed=!0}toArray(){throw new Error("Can not convert infinite video stream to array.")}},B7=class{},W7=class extends xn{split(e){return new vG(this,e)}},vG=class extends W7{constructor(e,t){super();this.upstream=e,this.impl=new wG(e,t)}summary(){return this.impl.summary()}async next(){return this.impl.next()}},wG=class extends Ry{constructor(e,t){super();this.upstream=e,this.separator=t,this.carryover=""}summary(){return`${this.upstream.summary()} -> Split('${this.separator}')`}async pump(){let e=await this.upstream.next();if(e.done)return this.carryover===""?!1:(this.outputQueue.push(this.carryover),this.carryover="",!0);let t=e.value.split(this.separator);t[0]=this.carryover+t[0];for(let n of t.slice(0,-1))this.outputQueue.push(n);return this.carryover=t[t.length-1],!0}},kG=class extends xn{decodeUTF8(){return new SG(this)}},SG=class extends W7{constructor(e){super();this.upstream=e,this.impl=new IG(e)}summary(){return this.impl.summary()}async next(){return this.impl.next()}},IG=class extends Ry{constructor(e){super();if(this.upstream=e,K().get("IS_BROWSER"))this.decoder=new TextDecoder("utf-8");else{let{StringDecoder:t}=A5();this.decoder=new t("utf8")}}summary(){return`${this.upstream.summary()} -> Utf8`}async pump(){let e=await this.upstream.next(),t;if(e.done)return!1;t=e.value;let n;return K().get("IS_BROWSER")?n=this.decoder.decode(t,{stream:!0}):n=this.decoder.write(Buffer.from(t.buffer)),this.outputQueue.push(n),!0}},V7=class extends kG{constructor(e,t={}){super();this.file=e,this.options=t,v.assert(e instanceof Uint8Array||(K().get("IS_BROWSER")?e instanceof File||e instanceof Blob:!1),()=>"FileChunkIterator only supports File, Blob and Uint8Array right now."),this.offset=t.offset||0,this.chunkSize=t.chunkSize||1024*1024}summary(){return`FileChunks ${this.file}`}async next(){return this.offset>=(this.file instanceof Uint8Array?this.file.byteLength:this.file.size)?{value:null,done:!0}:{value:await new Promise((t,n)=>{let s=this.offset+this.chunkSize;if(this.file instanceof Uint8Array)t(new Uint8Array(this.file.slice(this.offset,s)));else{let r=new FileReader;r.onload=o=>{let i=r.result;if(i instanceof ArrayBuffer&&(i=new Uint8Array(i)),!(i instanceof Uint8Array))return n(new TypeError("FileReader returned unknown type."));t(i)},r.onabort=o=>n(new Error("Aborted")),r.onerror=o=>n(new Error(o.type));let a=this.file.slice(this.offset,s);r.readAsArrayBuffer(a)}this.offset=s}),done:!1}}};async function CG(e,t={},n){let s,r;typeof e=="string"?s=e:(s=e.url,r=TG(e));let a=await(n||v.fetch)(s,r);if(a.ok){let o=new Uint8Array(await a.arrayBuffer());return new V7(o,t)}else throw new Error(a.statusText)}var TG=e=>({method:e.method,headers:e.headers,body:e.body,mode:e.mode,credentials:e.credentials,cache:e.cache,redirect:e.redirect,referrer:e.referrer,integrity:e.integrity});function U7(e){return typeof e=="string"&&e.substr(0,7)==="file://"}var G7=class extends B7{constructor(e,t={}){super();this.input=e,this.options=t}async iterator(){if(U7(this.input)&&K().get("IS_NODE")){let e=Ws("fs");this.input=e.readFileSync(this.input.substr(7))}return new V7(this.input,this.options)}},H7=class extends B7{constructor(e,t={}){super();this.url=e,this.fileOptions=t}async iterator(){return U7(this.url)?new G7(this.url,this.fileOptions).iterator():CG(this.url,this.fileOptions)}};function NG(e,t={}){return new M7(new H7(e),t)}function EG(e){let t=Ey(e);return ms(async()=>t)}function RG(e){return ms(async()=>{let t=await e();return Ey(()=>t.next())})}async function $G(e,t){return L7.create(e,t)}async function _G(e){return z7.create(e)}var DG="0.0.0";function Ne(e,t){Array.isArray(e)||(e=[e]),e.forEach(n=>{n!=null&&v.assert(n.dtype!=="complex64",()=>`${t} does not support complex64 tensors in the CPU backend.`)})}var PG=Xs.whereImpl,j7=class extends Yl{constructor(){super();this.blockSize=48,this.firstUse=!0,this.data=new Xc(this,ns())}nextDataId(){return j7.nextDataId++}write(e,t,n){this.firstUse&&(this.firstUse=!1,K().get("IS_NODE")&&E.warn(`
|
|
============================
|
|
Hi there \u{1F44B}. Looks like you are running TensorFlow.js in Node.js. To speed things up dramatically, install our node backend, which binds to TensorFlow C++, by running npm i @tensorflow/tfjs-node, or npm i @tensorflow/tfjs-node-gpu if you have CUDA. Then call require('@tensorflow/tfjs-node'); (-gpu suffix for CUDA) at the start of your program. Visit https://github.com/tensorflow/tfjs-node for more details.
|
|
============================`));let s={id:this.nextDataId()};return this.data.set(s,{values:e,dtype:n,refCount:1}),s}makeTensorInfo(e,t,n){let s;if(t==="string"&&n!=null&&n.length>0&&v.isString(n[0])){let r=n.map(a=>v.encodeString(a));s=this.write(r,e,t)}else s=this.write(n,e,t);return{dataId:s,shape:e,dtype:t}}refCount(e){return this.data.has(e)?this.data.get(e).refCount:0}incRef(e){let t=this.data.get(e);t.refCount++}decRef(e){if(this.data.has(e)){let t=this.data.get(e);t.refCount--}}move(e,t,n,s,r){this.data.set(e,{values:t,dtype:s,refCount:r})}numDataIds(){return this.data.numDataIds()}async read(e){return this.readSync(e)}readSync(e){let{dtype:t,complexTensorInfos:n}=this.data.get(e);if(t==="complex64"){let s=this.readSync(n.real.dataId),r=this.readSync(n.imag.dataId);return E.mergeRealAndImagArrays(s,r)}return this.data.get(e).values}bufferSync(e){let t=this.readSync(e.dataId),n=t;if(e.dtype==="string")try{n=t.map(s=>v.decodeString(s))}catch(s){throw new Error("Failed to decode encoded string bytes into utf-8")}return ze(e.shape,e.dtype,n)}makeOutput(e,t,n){let s=this.write(e,t,n);return ns().makeTensorFromDataId(s,t,n,this)}disposeData(e,t=!1){if(this.data.has(e)){if(this.data.get(e).refCount--,!t&&this.data.get(e).refCount>0)return!1;let{complexTensorInfos:n}=this.data.get(e);n!=null&&(this.disposeData(n.real.dataId,!0),this.disposeData(n.imag.dataId,!0)),this.data.delete(e)}return!0}disposeIntermediateTensorInfo(e){this.disposeData(e.dataId)}async time(e){let t=v.now();return e(),{kernelMs:v.now()-t}}memory(){return{unreliable:!0,reasons:["The reported memory is an upper bound. Due to automatic garbage collection, the true allocated memory may be less."]}}where(e){Ne([e],"where");let t=this.readSync(e.dataId);return PG(e.shape,t)}dispose(){}floatPrecision(){return 32}epsilon(){return super.epsilon()}},_y=j7;_y.nextDataId=0;var bm={};Oe(bm,{addImpl:()=>X7,bincountImpl:()=>Py,bincountReduceImpl:()=>K7,ceilImpl:()=>Z7,concatImpl:()=>Fy,equalImpl:()=>Y7,expImpl:()=>Q7,expm1Impl:()=>tS,floorImpl:()=>nS,gatherNdImpl:()=>sS,gatherV2Impl:()=>rS,greaterEqualImpl:()=>oS,greaterImpl:()=>aS,lessEqualImpl:()=>lS,lessImpl:()=>iS,linSpaceImpl:()=>uS,logImpl:()=>cS,maxImpl:()=>dS,maximumImpl:()=>pS,minimumImpl:()=>hS,multiplyImpl:()=>Oy,negImpl:()=>fS,notEqualImpl:()=>mS,prodImpl:()=>gS,rangeImpl:()=>zy,rsqrtImpl:()=>AS,sigmoidImpl:()=>wH,simpleAbsImpl:()=>q7,sliceImpl:()=>km,sparseFillEmptyRowsImpl:()=>xS,sparseReshapeImpl:()=>bS,sparseSegmentReductionImpl:()=>Ly,sqrtImpl:()=>IH,squaredDifferenceImpl:()=>vS,stridedSliceImpl:()=>wS,stringNGramsImpl:()=>kS,stringSplitImpl:()=>SS,stringToHashBucketFastImpl:()=>IS,subImpl:()=>CS,tileImpl:()=>TS,topKImpl:()=>ES,transposeImpl:()=>My,uniqueImpl:()=>RS});function q7(e){let t=new Float32Array(e.length);for(let n=0;n<e.length;++n)t[n]=Math.abs(e[n]);return t}var FG=e=>{let{x:t}=e.inputs,n=e.backend;Ne(t,"abs");let s=new Float32Array(v.sizeFromShape(t.shape)),r=n.data.get(t.dataId).values;return s=q7(r),n.makeOutput(s,t.shape,t.dtype)},OG={kernelName:ci,backendName:"cpu",kernelFunc:FG};function Zt(e){return(t,n,s,r,a)=>{let o=E.assertAndGetBroadcastShape(t,n),i=o.length,l=v.computeStrides(o),c=v.sizeFromShape(o),u=v.getTypedArrayFromDType(a,c),d=t.length,p=n.length,h=v.computeStrides(t),f=v.computeStrides(n),m=E.getBroadcastDims(t,o),g=E.getBroadcastDims(n,o);if(m.length+g.length===0)for(let A=0;A<u.length;++A)u[A]=e(s[A%s.length],r[A%r.length]);else for(let A=0;A<u.length;++A){let x=v.indexToLoc(A,i,l),y=x.slice(-d);m.forEach(C=>y[C]=0);let b=v.locToIndex(y,d,h),w=x.slice(-p);g.forEach(C=>w[C]=0);let k=v.locToIndex(w,p,f);u[A]=e(s[b],r[k])}return[u,o]}}function gs(e){let{inputs:t,backend:n}=e,{real:s,imag:r}=t,a=n.data.get(s.dataId).values,o=n.data.get(r.dataId).values,i=n.makeTensorInfo(s.shape,"complex64"),l=n.data.get(i.dataId);return l.complexTensorInfos={real:n.makeTensorInfo(s.shape,"float32",a),imag:n.makeTensorInfo(r.shape,"float32",o)},i}var MG={kernelName:Jc,backendName:"cpu",kernelFunc:gs};function vm(e,t,n="float32"){if(n==="complex64"){let r=vm(e,t,"float32"),a=vm(e,t,"float32");return gs({inputs:{real:r,imag:a},backend:e})}let s=v.makeZerosTypedArray(v.sizeFromShape(t),n);return e.makeTensorInfo(t,n,s)}function Mr(e){let{inputs:t,backend:n}=e,{x:s}=t;return n.incRef(s.dataId),{dataId:s.dataId,shape:s.shape,dtype:s.dtype}}var zG={kernelName:Ga,backendName:"cpu",kernelFunc:Mr};function bl(e){let{inputs:t,backend:n}=e,{input:s}=t,r=n.data.get(s.dataId).complexTensorInfos.real,a=n.data.get(r.dataId).values;return n.makeTensorInfo(r.shape,r.dtype,a)}var LG={kernelName:id,backendName:"cpu",kernelFunc:bl};function Bo(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{dtype:a}=s;if(a==="complex64"){if(r.dtype==="complex64")return Mr({inputs:{x:r},backend:n});let o=vm(n,r.shape,r.dtype),i=Bo({inputs:{x:r},backend:n,attrs:{dtype:"float32"}}),l=gs({inputs:{real:i,imag:o},backend:n});return n.disposeIntermediateTensorInfo(o),n.disposeIntermediateTensorInfo(i),l}if(r.dtype==="complex64"){let o=bl({inputs:{input:r},backend:n}),i=Bo({inputs:{x:o},backend:n,attrs:{dtype:a}});return n.disposeIntermediateTensorInfo(o),i}if(!v.hasEncodingLoss(r.dtype,a)){let o=Mr({inputs:{x:r},backend:n});return{dataId:o.dataId,shape:o.shape,dtype:a}}if(a==="int32"){let o=n.data.get(r.dataId).values,i=Int32Array.from(o);return n.makeTensorInfo(r.shape,"int32",i)}if(a==="bool"){let o=n.data.get(r.dataId).values,i=v.toTypedArray([0],r.dtype),[l,c]=Zt((u,d)=>u!==d?1:0)(r.shape,[],o,i,"bool");return n.makeTensorInfo(c,"bool",l)}throw new Error(`Error in Cast: failed to cast ${r.dtype} to ${a}`)}var BG={kernelName:Ra,backendName:"cpu",kernelFunc:Bo};function bn(e,t,n,s){return n==null?({inputs:r,backend:a})=>{let{a:o,b:i}=r,l=a;Ne([o,i],e);let c=l.data.get(o.dataId).values,u=l.data.get(i.dataId).values,d=o.dtype==="string"?E.fromUint8ToStringArray(c):c,p=o.dtype==="string"?E.fromUint8ToStringArray(u):u,h=s||o.dtype,[f,m]=t(o.shape,i.shape,d,p,h);return l.makeTensorInfo(m,h,f)}:({inputs:r,backend:a})=>{let{a:o,b:i}=r,l=a;if(o.dtype==="complex64"||i.dtype==="complex64"){let c=Bo({inputs:{x:o},backend:l,attrs:{dtype:"complex64"}}),u=l.data.get(c.dataId),d=u.complexTensorInfos.real,p=u.complexTensorInfos.imag,h=l.data.get(d.dataId).values,f=l.data.get(p.dataId).values,m=Bo({inputs:{x:i},backend:l,attrs:{dtype:"complex64"}}),g=l.data.get(m.dataId),A=g.complexTensorInfos.real,x=g.complexTensorInfos.imag,y=l.data.get(A.dataId).values,b=l.data.get(x.dataId).values,[w,k,C]=n(o.shape,i.shape,h,f,y,b),N=l.makeTensorInfo(C,"float32",w),R=l.makeTensorInfo(C,"float32",k),F=gs({inputs:{real:N,imag:R},backend:l});return l.disposeIntermediateTensorInfo(c),l.disposeIntermediateTensorInfo(m),l.disposeIntermediateTensorInfo(N),l.disposeIntermediateTensorInfo(R),F}else{let c=l.data.get(o.dataId).values,u=l.data.get(i.dataId).values,d=s||o.dtype,[p,h]=t(o.shape,i.shape,c,u,d);return l.makeTensorInfo(h,d,p)}}}function Dy(e){return(t,n,s,r,a,o)=>{let i=E.assertAndGetBroadcastShape(t,n),l=v.sizeFromShape(i),c=i.length,u=v.computeStrides(i),d=v.getTypedArrayFromDType("float32",l),p=v.getTypedArrayFromDType("float32",l),h=E.getBroadcastDims(t,i),f=E.getBroadcastDims(n,i),m=E.mergeRealAndImagArrays(s,r),g=E.mergeRealAndImagArrays(a,o),A=t.length,x=v.computeStrides(t),y=n.length,b=v.computeStrides(n);if(h.length+f.length===0)for(let w=0;w<d.length;w++){let k=w%m.length,C=w%g.length,N=e(m[k*2],m[k*2+1],g[C*2],g[C*2+1]);d[w]=N.real,p[w]=N.imag}else for(let w=0;w<d.length;w++){let k=v.indexToLoc(w,c,u),C=k.slice(-A);h.forEach(P=>C[P]=0);let N=v.locToIndex(C,A,x),R=k.slice(-y);f.forEach(P=>R[P]=0);let F=v.locToIndex(R,y,b),_=e(m[N*2],m[N*2+1],g[F*2],g[F*2+1]);d[w]=_.real,p[w]=_.imag}return[d,p,i]}}var X7=Zt((e,t)=>e+t),WG=Dy((e,t,n,s)=>({real:e+n,imag:t+s})),np=bn(Gr,X7,WG),VG={kernelName:Gr,backendName:"cpu",kernelFunc:np};function Py(e,t,n,s,r){let a=v.sizeFromShape(s),o=v.makeZerosTypedArray(r,n);for(let i=0;i<e.length;i++){let l=e[i];if(l<0)throw new Error("Input x must be non-negative!");l>=r||(a>0?o[l]+=t[i]:o[l]+=1)}return o}function K7(e,t,n,s=!1){let r=e.shape[0],a=e.shape[1],o=ze([r,n],t.dtype);for(let i=0;i<r;i++)for(let l=0;l<a;l++){let c=e.get(i,l);if(c<0)throw new Error("Input x must be non-negative!");c>=n||(s?o.set(1,i,c):t.size>0?o.set(o.get(i,c)+t.get(i,l),i,c):o.set(o.get(i,c)+1,i,c))}return o}function Wo(e){return(t,n,s)=>{let r=v.getTypedArrayFromDType(n,t.length);for(let a=0;a<t.length;++a)r[a]=e(t[a],s);return r}}function ht(e,t,n){return({inputs:s,attrs:r,backend:a})=>{let{x:o}=s;if(Ne(o,e),o.dtype==="string"||n==="string")throw new Error("unaryKernelFunc does not support string input/output");let i=a,l=i.data.get(o.dataId).values,c=v.sizeFromShape(o.shape),u=n||o.dtype,d=v.getArrayFromDType(u,c);for(let p=0;p<c;++p)d[p]=t(l[p],r);return i.makeTensorInfo(o.shape,u,d)}}function Ju(e,t,n){return({inputs:s,attrs:r,backend:a})=>{let{x:o}=s;if(Ne(o,e),o.dtype==="string"||n==="string")throw new Error("unaryKernelFunc does not support string input/output");let i=a,l=i.data.get(o.dataId).values,c=n||o.dtype,u=t(l,c,r);return i.makeTensorInfo(o.shape,c,u)}}var Z7=Wo(e=>Math.ceil(e)),UG=Ju($a,Z7),GG={kernelName:$a,backendName:"cpu",kernelFunc:UG};function Fy(e,t,n,s){let r=v.getArrayFromDType(n,v.sizeFromShape(t));if(s&&n!=="string"){let a=0;e.forEach(o=>{let i=v.sizeFromShape(o.shape);r.set(o.vals,a),a+=i})}else{let a=0;e.forEach(o=>{let i=n==="string"?E.fromUint8ToStringArray(o.vals):o.vals,l=0;for(let c=0;c<o.shape[0];++c){let u=c*t[1]+a;for(let d=0;d<o.shape[1];++d)r[u+d]=i[l++]}a+=o.shape[1]})}return r}var Y7=Zt((e,t)=>e===t?1:0),J7=bn(gi,Y7,null,"bool"),HG={kernelName:gi,backendName:"cpu",kernelFunc:J7},Q7=Wo(e=>Math.exp(e)),eS=Ju(La,Q7,"float32"),jG={kernelName:La,backendName:"cpu",kernelFunc:eS},tS=Wo(e=>Math.expm1(e)),qG=Ju(yi,tS),XG={kernelName:yi,backendName:"cpu",kernelFunc:qG},nS=Wo(e=>Math.floor(e)),KG=Ju(Ba,nS),ZG={kernelName:Ba,backendName:"cpu",kernelFunc:KG};function sS(e,t,n,s,r,a,o,i,l){let c=ze([s,a],n);for(let u=0;u<s;u++){let d=[],p=0;for(let h=0;h<r;h++){let f=e[u*r+h];p+=f*o[h],d.push(f)}if(p<0||p>=l/a)throw new Error(`Invalid indices: ${d} does not index into ${i}`);for(let h=0;h<a;h++)c.values[u*a+h]=t.get(...t.indexToLoc(p*a+h))}return c}function rS(e,t,n){let s=ze(n,e.dtype);for(let r=0;r<s.size;++r){let o=s.indexToLoc(r).slice(),i=o[0],l=o[2],c=t.locToIndex([i,l]);o[2]=t.values[c];let u=e.locToIndex(o);s.values[r]=e.values[u]}return s}var aS=Zt((e,t)=>e>t?1:0),YG=bn(wi,aS,null,"bool"),JG={kernelName:wi,backendName:"cpu",kernelFunc:YG},oS=Zt((e,t)=>e>=t?1:0),QG=bn(Ua,oS,null,"bool"),eH={kernelName:Ua,backendName:"cpu",kernelFunc:QG},iS=Zt((e,t)=>e<t?1:0),tH=bn(Si,iS,null,"bool"),nH={kernelName:Si,backendName:"cpu",kernelFunc:tH},lS=Zt((e,t)=>e<=t?1:0),sH=bn(Ii,lS,null,"bool"),rH={kernelName:Ii,backendName:"cpu",kernelFunc:sH};function uS(e,t,n){let s=(t-e)/(n-1),r=v.makeZerosTypedArray(n,"float32");r[0]=e;for(let a=1;a<r.length;a++)r[a]=r[a-1]+s;return r}var cS=Wo(e=>Math.log(e)),aH=Ju(Ha,cS),oH={kernelName:Ha,backendName:"cpu",kernelFunc:aH};function dS(e,t,n,s){let r=v.getTypedArrayFromDType(s,v.sizeFromShape(n));for(let a=0;a<r.length;++a){let o=a*t,i=e[o];for(let l=0;l<t;++l){let c=e[o+l];(Number.isNaN(c)||c>i)&&(i=c)}r[a]=i}return r}var pS=Zt((e,t)=>Math.max(e,t)),iH=bn(qa,pS),lH={kernelName:qa,backendName:"cpu",kernelFunc:iH},hS=Zt((e,t)=>Math.min(e,t)),uH=bn(Ya,hS),cH={kernelName:Ya,backendName:"cpu",kernelFunc:uH},Oy=Zt((e,t)=>e*t),dH=Dy((e,t,n,s)=>({real:e*n-t*s,imag:e*s+t*n})),wm=bn(Qa,Oy,dH),pH={kernelName:Qa,backendName:"cpu",kernelFunc:wm};function fS(e,t,n){let s=v.createScalarValue(-1,n);return Oy([],t,s,e,n)}function hH(e){let{inputs:t,backend:n}=e,{x:s}=t;Ne(s,"neg");let r=n.data.get(s.dataId).values,[a,o]=fS(r,s.shape,s.dtype);return n.makeTensorInfo(o,s.dtype,a)}var fH={kernelName:Ti,backendName:"cpu",kernelFunc:hH},mS=Zt((e,t)=>e!==t?1:0),mH=bn(Ni,mS,null,"bool"),gH={kernelName:Ni,backendName:"cpu",kernelFunc:mH};function My(e,t,n,s,r){let a=t.length,o=v.sizeFromShape(t),i=v.computeStrides(t),l=v.computeStrides(r),c=v.getTypedArrayFromDType(n,v.sizeFromShape(r));for(let u=0;u<o;++u){let d=v.indexToLoc(u,a,i),p=new Array(d.length);for(let f=0;f<p.length;f++)p[f]=d[s[f]];let h=v.locToIndex(p,a,l);c[h]=e[u]}return c}function _s(e){let{inputs:t,attrs:n,backend:s}=e,{x:r}=t,{perm:a}=n;Ne(r,"transpose");let o=r.shape.length,i=new Array(o);for(let d=0;d<i.length;d++)i[d]=r.shape[a[d]];let l=s.data.get(r.dataId).values,c=My(l,r.shape,r.dtype,a,i);return{dataId:s.write(c,i,r.dtype),shape:i,dtype:r.dtype}}var AH={kernelName:go,backendName:"cpu",kernelFunc:_s};function gS(e,t,n,s){let[r,a]=E.computeOutAndReduceShapes(e,s),o=zn(t,"int32"),i=v.makeZerosTypedArray(v.sizeFromShape(r),o),l=v.sizeFromShape(a);for(let c=0;c<i.length;++c){let u=c*l,d=1;for(let p=0;p<l;++p)d*=n[u+p];i[c]=d}return{outVals:i,outShape:r,outDtype:o}}function yH(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a,keepDims:o}=s;Ne(r,"prod");let i=r.shape.length,l=v.parseAxisParam(a,r.shape),c=E.getAxesPermutation(l,i),u=l,d=r,p=[];c!=null&&(d=_s({inputs:{x:r},backend:n,attrs:{perm:c}}),p.push(d),u=E.getInnerMostAxes(u.length,i));let h=n.data.get(d.dataId).values,{outVals:f,outShape:m,outDtype:g}=gS(d.shape,d.dtype,h,u),A=m;return o&&(A=E.expandShapeToKeepDim(m,l)),p.forEach(x=>n.disposeIntermediateTensorInfo(x)),n.makeTensorInfo(A,g,f)}var xH={kernelName:Pi,backendName:"cpu",kernelFunc:yH};function zy(e,t,n,s){let r=e===t,a=e<t&&n<0,o=t<e&&n>1;if(r||a||o)return v.makeZerosTypedArray(0,s);let i=Math.abs(Math.ceil((t-e)/n)),l=v.makeZerosTypedArray(i,s);t<e&&n===1&&(n=-1),l[0]=e;for(let c=1;c<l.length;c++)l[c]=l[c-1]+n;return l}var AS=Wo(e=>1/Math.sqrt(e)),bH=Ju(oo,AS),vH={kernelName:oo,backendName:"cpu",kernelFunc:bH},wH=Wo(e=>1/(1+Math.exp(-e))),yS=ht(lo,e=>1/(1+Math.exp(-e))),kH={kernelName:lo,backendName:"cpu",kernelFunc:yS};function km(e,t,n,s,r){let a=Pt.isSliceContinous(s,t,n),o=v.sizeFromShape(n),i=v.computeStrides(s);if(a){let d=Pt.computeFlatOffset(t,i);return r==="string"?e.slice(d,d+o):e.subarray(d,d+o)}let l=r==="string"?E.fromUint8ToStringArray(e):e,c=ze(s,r,l),u=ze(n,r);for(let d=0;d<u.size;++d){let p=u.indexToLoc(d),h=p.map((f,m)=>f+t[m]);u.set(c.get(...h),...p)}return r==="string"?E.fromStringArrayToUint8(u.values):u.values}function vl(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{begin:a,size:o}=s;Ne(r,"slice");let[i,l]=Pt.parseSliceParams(r,a,o);Pt.assertParamsValid(r,i,l);let c=n.data.get(r.dataId).values,u=km(c,i,l,r.shape,r.dtype);return n.makeTensorInfo(l,r.dtype,u)}var SH={kernelName:Bi,backendName:"cpu",kernelFunc:vl};function xS(e,t,n,s,r,a,o){let i=t[0],l=a[0],c=new Array(l),u=new Array(i),d=t[1];if(l===0){if(i!==0)throw new Error(`Received SparseTensor with denseShape[0] = 0 but
|
|
indices.shape[0] = ${i}`);let g=v.getArrayFromDType(n,0),A=v.getArrayFromDType(r,0);return[g,[0,d],A,c,u]}let p=!0,h=0,f=new Array(l).fill(0);for(let g=0;g<i;++g){let A=e[g*d];if(A<0)throw new Error(`indices(${g}, 0) is invalid: ${A} < 0`);if(A>=l)throw new Error(`indices(${g}, 0) is invalid: ${A} >= ${l}`);++f[A],p=p&&A>=h,h=A}let m=!0;for(let g=0;g<l;++g){let A=f[g]===0;c[g]=A,m=m&&!A,f[g]=Math.max(f[g],1),g>0&&(f[g]+=f[g-1])}if(m&&p){let g=e,A=s;for(let x=0;x<i;++x)u[x]=x;return[g,[i,d],A,c,u]}else{let g=f[l-1],A=v.getArrayFromDType(n,g*d),x=v.getArrayFromDType(r,g),y=new Array(l).fill(0);for(let b=0;b<i;++b){let w=e[b*d],k=y[w],C=(w===0?0:f[w-1])+k;y[w]++;for(let N=0;N<d;++N)A[C*d+N]=e[b*d+N];x[C]=s[b],u[b]=C}for(let b=0;b<l;++b)if(y[b]===0){let k=b===0?0:f[b-1];A[k*d+0]=b;for(let C=1;C<d;++C)A[k*d+C]=0;x[k]=o}return[A,[g,d],x,c,u]}}function bS(e,t,n,s,r){let a=v.sizeFromShape(s),o=t[0],i=r.length,l=[],c=1,u=-1;for(let g=0;g<i;++g){let A=r[g];if(A===-1){if(u!==-1)throw new Error(`only one output dimension may be -1, not both ${u} and ${g}`);u=g,l.push(1)}else{if(A<0)throw new Error(`size ${g} must be non-negative, not ${A}`);c*=A,l.push(A)}}if(u!==-1){if(c<=0)throw new Error("reshape cannot infer the missing input size for an empty tensor unless all specified input sizes are non-zero");let g=Math.trunc(a/c);if(c*g!==a)throw new Error(`Input to reshape is a SparseTensor with ${a}
|
|
dense values, but the requested shape requires a multiple of ${c}. inputShape=${s} outputShape= ${l}`);l[u]=g}let d=v.sizeFromShape(l);if(d!==a)throw new Error(`Input to reshape is a tensor with ${a} dense values, but the requested shape has ${d}. inputShape=${s} outputShape=${l}`);let p=s.length,h=[];if(p>0){h[p-1]=1;for(let g=p-2;g>=0;--g)h[g]=h[g+1]*s[g+1]}let f=[];if(i>0){f[i-1]=1;for(let g=i-2;g>=0;--g)f[g]=f[g+1]*l[g+1]}let m=v.getArrayFromDType(n,o*i);for(let g=0;g<o;++g){let A=0;for(let x=0;x<p;++x)A+=e[g*p+x]*h[x];for(let x=0;x<i;++x)m[g*i+x]=Math.trunc(A/f[x]),A%=f[x]}return[m,[o,i],l]}function Ly(e,t,n,s,r,a=!1,o=0){let i=s.length;if(i!==r.length)throw new Error("segmentIds and indices should have same size.");let l=[t[0],e.length/t[0]],c=l[1],d=i>0?r[i-1]+1:0;if(d<0)throw new Error("segment ids must be >= 0");let p=t.slice();p[0]=d;let h=p.reduce((y,b)=>y*b,1),f=v.getArrayFromDType(n,h);if(i===0)return d>0&&f.fill(o),[f,p];if(d<=0)throw new Error("segment ids must be >= 0");let m=0,g=1,A=0,x=r[m];for(;;){let y=0;if(g<i){if(y=r[g],x===y){++g;continue}if(x>=y)throw new Error("segment ids are not increasing")}if(x<0||x>=d)throw new Error(`Segment id ${x} out of range [0, ${d}), possibly because segmentIds input is not sorted.`);x>A&&f.fill(o,A*c,x*c);for(let b=m;b<g;++b){let w=s[b];if(w<0||w>=l[0])throw new Error(`Bad: indices[${b}] == ${s[b]} out of range [0, ${l[0]})`);for(let k=0;k<c;k++)f[x*c+k]+=e[w*c+k]}if(a)for(let b=0;b<c;b++)f[x*c+b]/=g-m;if(m=g,++g,A=x+1,x=y,g>i)break}return A<d&&f.fill(o,A*c,d*c),[f,p]}var IH=Wo(e=>Math.sqrt(e)),CH=ht(uo,e=>Math.sqrt(e)),TH={kernelName:uo,backendName:"cpu",kernelFunc:CH},vS=Zt((e,t)=>{let n=e-t;return n*n}),NH=bn(ho,vS),EH={kernelName:ho,backendName:"cpu",kernelFunc:NH};function wS(e,t,n,s){let r=ze(e,t.dtype);for(let a=0;a<r.size;a++){let o=r.indexToLoc(a),i=new Array(o.length);for(let l=0;l<i.length;l++)i[l]=o[l]*n[l]+s[l];r.set(t.get(...i),...o)}return r}var RH=class{constructor(e,t,n,s,r,a){this.separator=v.encodeString(e),this.nGramWidths=t,this.leftPad=v.encodeString(n),this.rightPad=v.encodeString(s),this.padWidth=r,this.preserveShort=a}getPadWidth(e){return Math.min(this.padWidth<0?e-1:this.padWidth,e-1)}getNumNGrams(e,t){let n=this.getPadWidth(t);return Math.max(0,e+2*n-t+1)}createNGrams(e,t,n,s,r,a){for(let o=0;o<r;++o){let i=this.getPadWidth(a),l=Math.max(0,i-o),c=Math.max(0,i-(r-(o+1))),u=a-(l+c),d=t+(l>0?0:o-i),p=0;p+=l*this.leftPad.length;for(let A=0;A<u;++A)p+=e[d+A].length;p+=c*this.rightPad.length,p+=(l+c+u-1)*this.separator.length,n[s+o]=new Uint8Array(p);let f=n[s+o],m=0,g=A=>A.forEach(x=>f[m++]=x);for(let A=0;A<l;++A)g(this.leftPad),g(this.separator);for(let A=0;A<u-1;++A)g(e[d+A]),g(this.separator);if(u>0){g(e[d+u-1]);for(let A=0;A<c;++A)g(this.separator),g(this.rightPad)}else{for(let A=0;A<c-1;++A)g(this.rightPad),g(this.separator);g(this.rightPad)}}}compute(e,t){let n=e.length,s=t.length;if(s>0){let i=t[0];if(i!==0)throw new Error(`First split value must be 0, got ${i}`);for(let l=1;l<s;++l){let c=t[l]>=i;if(c=c&&t[l]<=n,!c)throw new Error(`Invalid split value ${t[l]}, must be in [${i}, ${n}]`);i=t[l]}if(i!==n)throw new Error(`Last split value must be data size. Expected ${n}, got ${i}`)}let r=s-1,a=v.getArrayFromDType("int32",s);if(n===0||s===0){let i=new Array(n);for(let l=0;l<=r;++l)a[l]=0;return[i,a]}a[0]=0;for(let i=1;i<=r;++i){let l=t[i]-t[i-1],c=0;this.nGramWidths.forEach(u=>{c+=this.getNumNGrams(l,u)}),this.preserveShort&&l>0&&c===0&&(c=1),a[i]=a[i-1]+c}let o=new Array(a[r]);for(let i=0;i<r;++i){let l=t[i],c=a[i];if(this.nGramWidths.forEach(u=>{let d=t[i+1]-t[i],p=this.getNumNGrams(d,u);this.createNGrams(e,l,o,c,p,u),c+=p}),this.preserveShort&&c===a[i]){let u=t[i+1]-t[i];if(u===0)continue;let d=u+2*this.padWidth,p=1;this.createNGrams(e,l,o,c,p,d)}}return[o,a]}};function kS(e,t,n,s,r,a,o,i){return new RH(n,s,r,a,o,i).compute(e,t)}function $H(e,t,n,s){if(!e.length)return;if(t.length===0){for(let a=0;a<e.length;++a)s.push(e.subarray(a,a+1));return}if(t.length===1){let a=t[0],o=e.indexOf(a);for(;o!==-1;){let i=e.subarray(0,o);(!n||i.length!==0)&&s.push(i),e=e.subarray(o+1),o=e.indexOf(a)}(!n||e.length!==0)&&s.push(e);return}let r=0;for(let a=0;a<e.length+1;a++)if(a===e.length||t.indexOf(e[a])!==-1){let o=e.subarray(r,a);(!n||o.length!==0)&&s.push(o),r=a+1}}function SS(e,t,n){let s=e.length,r=[],a=0,o=0,i=new Array(s);for(let p=0;p<s;++p){let h=r.length;$H(e[p],t,n,r);let f=r.length-h;i[p]=f,a+=f,o=Math.max(o,f)}let l=v.getArrayFromDType("int32",a*2),c=new Array(a),u=[s,o],d=0;for(let p=0;p<s;++p)for(let h=0;h<i[p];++h)l[d*2]=p,l[d*2+1]=h,c[d]=r[d],++d;return[l,c,u]}function IS(e,t){let n=v.getArrayFromDType("int32",e.length);for(let s=0;s<e.length;++s)n[s]=v.fingerPrint64(e[s]).modulo(t).getLowBitsUnsigned();return n}var CS=Zt((e,t)=>e-t),_H=Dy((e,t,n,s)=>({real:e-n,imag:t-s})),By=bn(fo,CS,_H),DH={kernelName:fo,backendName:"cpu",kernelFunc:By};function TS(e,t){let n=new Array(e.rank);for(let r=0;r<n.length;r++)n[r]=e.shape[r]*t[r];let s=ze(n,e.dtype);for(let r=0;r<s.values.length;++r){let a=s.indexToLoc(r),o=new Array(e.rank);for(let l=0;l<o.length;l++)o[l]=a[l]%e.shape[l];let i=e.locToIndex(o);s.values[r]=e.values[i]}return s}var sp=(e,t)=>{let n=t.value-e.value;return n===0?e.index-t.index:n};function NS(e,t,n=0,s=e.length-1){for(;s>n;){if(s-n>600){let i=s-n+1,l=t-n+1,c=Math.log(i),u=.5*Math.exp(2*c/3),d=.5*Math.sqrt(c*u*(i-u)/i)*Math.sign(l-i/2),p=Math.max(n,Math.floor(t-l*u/i+d)),h=Math.min(s,Math.floor(t+(i-l)*u/i+d));NS(e,t,p,h)}let r=e[t],a=n,o=s;for(v.swap(e,n,t),sp(e[s],r)>0&&v.swap(e,n,s);a<o;){for(v.swap(e,a,o),a++,o--;sp(e[a],r)<0;)a=a+1;for(;sp(e[o],r)>0;)o=o-1}sp(e[n],r)===0?v.swap(e,n,o):(o=o+1,v.swap(e,o,s)),o<=t&&(n=o+1),t<=o&&(s=o-1)}}function ES(e,t,n,s,r){let a=t[t.length-1],[o,i]=[e.length/a,a],l=v.getTypedArrayFromDType(n,o*s),c=v.getTypedArrayFromDType("int32",o*s);for(let d=0;d<o;d++){let p=d*i,h=e.subarray(p,p+i),f=new Array(h.length);h.forEach((x,y)=>f[y]={value:x,index:y}),s<f.length&&(NS(f,s),f=f.slice(0,s)),r&&f.sort(sp);let m=d*s,g=l.subarray(m,m+s),A=c.subarray(m,m+s);for(let x=0;x<s;x++)g[x]=f[x].value,A[x]=f[x].index}let u=t.slice();return u[u.length-1]=s,[ze(u,n,l),ze(u,"int32",c)]}function RS(e,t,n,s){let r=v.parseAxisParam(t,n)[0],a=[1,n[0],1];for(let f=0;f<r;f++)a[0]*=n[f];a[1]=n[r];for(let f=r+1;f<n.length;f++)a[2]*=n[f];let o={},i=new Int32Array(n[r]),l=new en(a,s,e),c=[],u=a[0]===1&&a[2]===1;for(let f=0;f<n[r];f++){let m;if(u)m=e[f].toString();else{let g=[];for(let A=0;A<a[0];A++)for(let x=0;x<a[2];x++)g.push(l.get(A,f,x));m=g.join(",")}if(o[m]!==void 0)i[f]=o[m];else{let g=Object.keys(o).length;o[m]=g,i[f]=g,c.push(f)}}let d=a.slice();d[1]=Object.keys(o).length;let p=new en(d,s);c.forEach((f,m)=>{for(let g=0;g<a[0];g++)for(let A=0;A<a[2];A++)p.set(l.get(g,f,A),g,m,A)});let h=n.slice();return h[r]=d[1],{outputValues:p.values,outputShape:h,indices:i}}var PH="0.0.0";al("cpu",()=>new _y,1);var $S=ht(za,e=>e>=0?e:Math.exp(e)-1),FH={kernelName:za,backendName:"cpu",kernelFunc:$S};function _S(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{alpha:a}=s;Ne([r],"leakyRelu");let o=v.sizeFromShape(r.shape),i=n.data.get(r.dataId).values,l=v.getTypedArrayFromDType("float32",o);for(let c=0;c<i.length;c++)l[c]=i[c]<0?a*i[c]:i[c];return n.makeTensorInfo(r.shape,"float32",l)}var OH={kernelName:ki,backendName:"cpu",kernelFunc:_S},MH=Zt((e,t)=>e<0?t*e:e);function DS(e){let{inputs:t,backend:n}=e,{x:s,alpha:r}=t;Ne([s,r],"prelu");let a=n.data.get(s.dataId).values,o=n.data.get(r.dataId).values,[i,l]=MH(s.shape,r.shape,a,o,"float32");return n.makeTensorInfo(l,"float32",i)}var zH={kernelName:no,backendName:"cpu",kernelFunc:DS},PS=ht(so,e=>Math.max(0,e)),LH={kernelName:so,backendName:"cpu",kernelFunc:PS},FS=ht(ao,e=>Math.min(Math.max(0,e),6)),BH={kernelName:ao,backendName:"cpu",kernelFunc:FS};function Wy(e,t,n,s,r){if(n==="linear")return Mr({inputs:{x:t},backend:e});if(n==="relu")return PS({inputs:{x:t},backend:e});if(n==="elu")return $S({inputs:{x:t},backend:e});if(n==="relu6")return FS({inputs:{x:t},backend:e});if(n==="prelu")return DS({inputs:{x:t,alpha:s},backend:e});if(n==="leakyrelu")return _S({inputs:{x:t},backend:e,attrs:{alpha:r}});if(n==="sigmoid")return yS({inputs:{x:t},backend:e});throw new Error(`Activation ${n} has not been implemented for the CPU backend.`)}function Rt(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{shape:a}=s,o=v.sizeFromShape(r.shape),i=v.inferFromImplicitShape(a,o),l=v.sizeFromShape(i);v.assert(o===l,()=>`The new shape (${i}) has ${l} elements and the old shape (${r.shape}) has ${o} elements. The new shape and old shape must have the same number of elements.`),n.incRef(r.dataId);let c=n.data.get(r.dataId);if(c.complexTensorInfos!=null){let u=c.complexTensorInfos.real,d=c.complexTensorInfos.imag;u.shape=i,d.shape=i}return{dataId:r.dataId,shape:i,dtype:r.dtype}}var WH={kernelName:Fi,backendName:"cpu",kernelFunc:Rt};function OS(e){let{inputs:t,backend:n,attrs:s}=e,{a:r,b:a}=t,{transposeA:o,transposeB:i}=s;Ne([r,a],"matMul");let l=r.shape.length,c=a.shape.length,u=o?r.shape[l-2]:r.shape[l-1],d=i?a.shape[c-1]:a.shape[c-2],p=o?r.shape[l-1]:r.shape[l-2],h=i?a.shape[c-2]:a.shape[c-1],f=r.shape.slice(0,-2),m=a.shape.slice(0,-2),g=v.sizeFromShape(f),A=v.sizeFromShape(m),y=nl.assertAndGetBroadcastShape(r.shape.slice(0,-2),a.shape.slice(0,-2)).concat([p,h]);v.assert(u===d,()=>`Error in matMul: inner shapes (${u}) and (${d}) of Tensors with shapes ${r.shape} and ${a.shape} and transposeA=${o} and transposeB=${i} must match.`);let b=o?[g,u,p]:[g,p,u],w=i?[A,h,d]:[A,d,h],k=Rt({inputs:{x:r},backend:n,attrs:{shape:b}}),C=Rt({inputs:{x:a},backend:n,attrs:{shape:w}}),N=o?k.shape[1]:k.shape[2],R=o?k.shape[2]:k.shape[1],F=i?C.shape[1]:C.shape[2],_=Math.max(g,A),P=n.data.get(k.dataId).values,T=n.data.get(C.dataId).values,M=v.computeStrides(k.shape),U=v.computeStrides(C.shape),[j,z,X]=o?[M[0],1,M[1]]:[M[0],M[1],1],[Z,J,ee]=i?[1,U[1],U[0]]:[U[1],1,U[0]],ne=R*F,Q=ze([_,R,F],k.dtype),te=Q.values,oe=n.blockSize;for(let fe=0;fe<_;fe++)for(let be=0;be<R;be+=oe)for(let we=0;we<F;we+=oe)for(let Ce=0;Ce<N;Ce+=oe){let Me=Math.min(be+oe,R),We=Math.min(we+oe,F),He=Math.min(Ce+oe,N);for(let qe=be;qe<Me;qe++)for(let ct=we;ct<We;ct++){let dt=0;for(let rt=Ce;rt<He;rt++){let wt=Math.min(fe,g-1)*j,ft=Math.min(fe,A-1)*ee,Ct=P[wt+qe*z+rt*X],_t=T[rt*Z+ct*J+ft];dt+=Ct*_t}te[fe*ne+(qe*F+ct)]+=dt}}return n.disposeIntermediateTensorInfo(k),n.disposeIntermediateTensorInfo(C),n.makeTensorInfo(y,Q.dtype,Q.values)}var VH={kernelName:Ea,backendName:"cpu",kernelFunc:OS};function UH(e){let{inputs:t,backend:n,attrs:s}=e,{a:r,b:a,bias:o,preluActivationWeights:i}=t,{transposeA:l,transposeB:c,activation:u,leakyreluAlpha:d}=s,p,h,f,m=[];p=OS({inputs:{a:r,b:a},attrs:{transposeA:l,transposeB:c},backend:n}),o&&(h=np({inputs:{a:p,b:o},backend:n}),m.push(p),p=h),u&&(f=Wy(n,p,u,i,d),m.push(p),p=f);for(let A of m)n.disposeIntermediateTensorInfo(A);return p}var GH={kernelName:yo,backendName:"cpu",kernelFunc:UH},HH=ht(eu,e=>Math.acos(e)),jH={kernelName:eu,backendName:"cpu",kernelFunc:HH},qH=ht(tu,e=>Math.acosh(e)),XH={kernelName:tu,backendName:"cpu",kernelFunc:qH};function KH(e){let{inputs:t,backend:n}=e,s=t;Ne(t,"addN");let r=s.map(i=>n.data.get(i.dataId).values),a=ze(s[0].shape,s[0].dtype),o=a.values;for(let i=0;i<s.length;i++){let l=r[i];for(let c=0;c<o.length;c++)o[c]+=l[c]}return n.makeTensorInfo(a.shape,a.dtype,a.values)}var ZH={kernelName:Ca,backendName:"cpu",kernelFunc:KH};function YH(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a,keepDims:o}=s;Ne(r,"all");let i=v.parseAxisParam(a,r.shape),l=i,c=E.getAxesPermutation(l,r.shape.length),u=r;c!=null&&(u=_s({inputs:{x:r},backend:n,attrs:{perm:c}}),l=E.getInnerMostAxes(l.length,r.shape.length)),E.assertAxesAreInnerMostDims("all",l,u.shape.length);let[d,p]=E.computeOutAndReduceShapes(u.shape,l),h=v.sizeFromShape(p),f=v.makeZerosTypedArray(v.sizeFromShape(d),u.dtype),m=n.data.get(u.dataId).values;for(let A=0;A<f.length;++A){let x=A*h,y=m[x];for(let b=0;b<h;++b){let w=m[x+b];y=y&&w}f[A]=y}c!=null&&n.disposeIntermediateTensorInfo(u);let g=n.makeTensorInfo(d,u.dtype,f);if(o){let A=E.expandShapeToKeepDim(d,i),x=Rt({inputs:{x:g},backend:n,attrs:{shape:A}});return n.disposeIntermediateTensorInfo(g),x}return g}var JH={kernelName:nu,backendName:"cpu",kernelFunc:YH};function QH(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a,keepDims:o}=s;Ne(r,"any");let i=v.parseAxisParam(a,r.shape),l=i,c=E.getAxesPermutation(l,r.shape.length),u=r;c!=null&&(u=_s({inputs:{x:r},backend:n,attrs:{perm:c}}),l=E.getInnerMostAxes(l.length,r.shape.length)),E.assertAxesAreInnerMostDims("any",l,u.shape.length);let[d,p]=E.computeOutAndReduceShapes(u.shape,l),h=v.sizeFromShape(p),f=v.makeZerosTypedArray(v.sizeFromShape(d),u.dtype),m=n.data.get(u.dataId).values;for(let A=0;A<f.length;++A){let x=A*h,y=m[x];for(let b=0;b<h;++b){let w=m[x+b];y=y||w}f[A]=y}c!=null&&n.disposeIntermediateTensorInfo(u);let g=n.makeTensorInfo(d,u.dtype,f);if(o){let A=E.expandShapeToKeepDim(d,i),x=Rt({inputs:{x:g},backend:n,attrs:{shape:A}});return n.disposeIntermediateTensorInfo(g),x}return g}var ej={kernelName:su,backendName:"cpu",kernelFunc:QH};function tj(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a}=s;Ne(r,"argMax");let o=v.parseAxisParam(a,r.shape),i=E.getAxesPermutation(o,r.shape.length),l=r,c=[];i!=null&&(l=_s({inputs:{x:r},backend:n,attrs:{perm:i}}),c.push(l),o=E.getInnerMostAxes(o.length,l.shape.length)),o=[o[0]],E.assertAxesAreInnerMostDims("argMax",o,l.shape.length);let[u,d]=E.computeOutAndReduceShapes(l.shape,o),p=v.sizeFromShape(u),h=v.makeZerosTypedArray(p,"int32"),f=v.sizeFromShape(d),m=n.data.get(l.dataId).values;for(let g=0;g<h.length;++g){let A=g*f,x=m[A],y=0;for(let b=0;b<f;++b){let w=m[A+b];w>x&&(x=w,y=b)}h[g]=y}return c.forEach(g=>n.disposeIntermediateTensorInfo(g)),n.makeTensorInfo(u,"int32",h)}var nj={kernelName:Ta,backendName:"cpu",kernelFunc:tj};function sj(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a}=s;Ne(r,"argMin");let o=v.parseAxisParam(a,r.shape),i=E.getAxesPermutation(o,r.shape.length),l=r,c=[];i!=null&&(l=_s({inputs:{x:r},backend:n,attrs:{perm:i}}),c.push(l),o=E.getInnerMostAxes(o.length,l.shape.length)),o=[o[0]],E.assertAxesAreInnerMostDims("argMin",o,l.shape.length);let[u,d]=E.computeOutAndReduceShapes(l.shape,o),p=v.sizeFromShape(u),h=v.makeZerosTypedArray(p,"int32"),f=v.sizeFromShape(d),m=n.data.get(l.dataId).values;for(let g=0;g<h.length;++g){let A=g*f,x=m[A],y=0;for(let b=0;b<f;++b){let w=m[A+b];w<x&&(x=w,y=b)}h[g]=y}return c.forEach(g=>n.disposeIntermediateTensorInfo(g)),n.makeTensorInfo(u,"int32",h)}var rj={kernelName:ru,backendName:"cpu",kernelFunc:sj},aj=ht(au,e=>Math.asin(e)),oj={kernelName:au,backendName:"cpu",kernelFunc:aj},ij=ht(ou,e=>Math.asinh(e)),lj={kernelName:ou,backendName:"cpu",kernelFunc:ij},uj=ht(iu,e=>Math.atan(e)),cj={kernelName:iu,backendName:"cpu",kernelFunc:uj},dj=Zt((e,t)=>Math.atan2(e,t)),pj=bn(uu,dj),hj={kernelName:uu,backendName:"cpu",kernelFunc:pj},fj=ht(lu,e=>Math.atanh(e)),mj={kernelName:lu,backendName:"cpu",kernelFunc:fj};function Vy(e,t,n,s,r,a){let o=r.strideHeight,i=r.strideWidth,l=r.dilationHeight,c=r.dilationWidth,u=r.effectiveFilterHeight,d=r.effectiveFilterWidth,p=r.padInfo.top,h=r.padInfo.left,f=a==="max"?Number.NEGATIVE_INFINITY:Number.POSITIVE_INFINITY,m=ze(r.outShape,n),g=m.values,A=r.outShape[1]*r.outShape[2]*r.outShape[3],x=r.outShape[2]*r.outShape[3],y=r.outShape[3];for(let b=0;b<r.batchSize;++b){let w=b*A,k=b*s[0];for(let C=0;C<r.inChannels;++C)for(let N=0;N<r.outHeight;++N){let R=N*o-p,F=Math.max(0,R),_=Math.min(r.inHeight,u+R),P=w+N*x;for(let T=0;T<r.outWidth;++T){let M=T*i-h,U=Math.max(0,M),j=Math.min(r.inWidth,d+M),z=f,X=0,Z=0;for(let ee=F;ee<_;ee+=l){let ne=k+ee*s[1];for(let Q=U;Q<j;Q+=c){let te=ne+Q*s[2],oe=e[te+C];a==="max"&&oe>z?z=oe:a==="avg"&&(X+=oe,Z++)}if(isNaN(z))break}let J=P+T*y+C;g[J]=a==="avg"?X/Z:z}}}return m}function MS(e,t,n,s,r=!1,a=!1){let o=ze(s.outShape,"int32"),i=s.strideHeight,l=s.strideWidth,c=s.dilationHeight,u=s.dilationWidth,d=s.effectiveFilterHeight,p=s.effectiveFilterWidth,h=s.padInfo.top,f=s.padInfo.left,m=ze(t,n,e);for(let g=0;g<s.batchSize;++g)for(let A=0;A<s.inChannels;++A)for(let x=0;x<s.outHeight;++x){let y=x*i-h,b=y;for(;b<0;)b+=c;let w=Math.min(s.inHeight,d+y);for(let k=0;k<s.outWidth;++k){let C=k*l-f,N=C;for(;N<0;)N+=u;let R=Math.min(s.inWidth,p+C),F=Number.NEGATIVE_INFINITY,_=-1;for(let P=b;P<w;P+=c){let T=P-y;for(let M=N;M<R;M+=u){let U=M-C,j=m.get(g,P,M,A);j>F&&(F=j,r?_=a?((g*s.inHeight+P)*s.inWidth+M)*s.inChannels+A:(P*s.inWidth+M)*s.inChannels+A:_=T*p+U)}}o.set(_,g,x,k,A)}}return o}function zS(e,t,n,s,r,a){let o=r.strideDepth,i=r.strideHeight,l=r.strideWidth,c=r.dilationDepth,u=r.dilationHeight,d=r.dilationWidth,p=r.effectiveFilterDepth,h=r.effectiveFilterHeight,f=r.effectiveFilterWidth,m=r.padInfo.front,g=r.padInfo.top,A=r.padInfo.left,x=a==="max"?Number.NEGATIVE_INFINITY:Number.POSITIVE_INFINITY,y=ze(r.outShape,n),b=y.values,w=r.outShape[1]*r.outShape[2]*r.outShape[3]*r.outShape[4],k=r.outShape[2]*r.outShape[3]*r.outShape[4],C=r.outShape[3]*r.outShape[4],N=r.outShape[4];for(let R=0;R<r.batchSize;++R){let F=R*w,_=R*s[0];for(let P=0;P<r.inChannels;++P)for(let T=0;T<r.outDepth;++T){let M=T*o-m,U=M;for(;U<0;)U+=c;let j=Math.min(r.inDepth,p+M),z=F+T*k;for(let X=0;X<r.outHeight;++X){let Z=X*i-g,J=Z;for(;J<0;)J+=u;let ee=Math.min(r.inHeight,h+Z),ne=z+X*C;for(let Q=0;Q<r.outWidth;++Q){let te=Q*l-A,oe=te;for(;oe<0;)oe+=d;let fe=Math.min(r.inWidth,f+te),be=ne+Q*N,we=x,Ce=0,Me=0;for(let He=U;He<j;He+=c){let qe=_+He*s[1];for(let ct=J;ct<ee;ct+=u){let dt=qe+ct*s[2];for(let rt=oe;rt<fe;rt+=d){let wt=dt+rt*s[3],ft=e[wt+P];if(a==="max"&&ft>we?we=ft:a==="avg"&&(Ce+=ft,Me++),isNaN(we))break}if(isNaN(we))break}if(isNaN(we))break}let We=be+P;b[We]=a==="avg"?Ce/Me:we}}}}return y}function gj(e,t){let n=ze(t.outShape,"int32"),s=t.strideDepth,r=t.strideHeight,a=t.strideWidth,o=t.dilationDepth,i=t.dilationHeight,l=t.dilationWidth,c=t.effectiveFilterDepth,u=t.effectiveFilterHeight,d=t.effectiveFilterWidth,p=t.padInfo.front,h=t.padInfo.top,f=t.padInfo.left;for(let m=0;m<t.batchSize;++m)for(let g=0;g<t.inChannels;++g)for(let A=0;A<t.outDepth;++A){let x=A*s-p,y=x;for(;y<0;)y+=o;let b=Math.min(t.inDepth,c+x);for(let w=0;w<t.outHeight;++w){let k=w*r-h,C=k;for(;C<0;)C+=i;let N=Math.min(t.inHeight,u+k);for(let R=0;R<t.outWidth;++R){let F=R*a-f,_=F;for(;_<0;)_+=l;let P=Math.min(t.inWidth,d+F),T=Number.NEGATIVE_INFINITY,M=-1;for(let U=y;U<b;U+=o){let j=U-x;for(let z=C;z<N;z+=i){let X=z-k;for(let Z=_;Z<P;Z+=l){let J=Z-F,ee=e.get(m,U,z,Z,g);ee>=T&&(T=ee,M=j*u*d+X*u+J)}}}n.set(M,m,A,w,R,g)}}}return n}function Aj(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t;Ne(r,"avgPool");let{filterSize:a,strides:o,pad:i,dimRoundingMode:l}=s,c=1;v.assert(E.eitherStridesOrDilationsAreOne(o,c),()=>`Error in avgPool: Either strides or dilations must be 1. Got strides ${o} and dilations '${c}'`);let u=E.computePool2DInfo(r.shape,a,o,c,i,l),d;if(u.filterWidth===1&&u.filterHeight===1&&v.arraysEqual(u.inShape,u.outShape))d=Mr({inputs:{x:r},backend:n});else{let p=n.data.get(r.dataId).values,h=v.computeStrides(r.shape),f=Vy(p,r.shape,r.dtype,h,u,"avg");d=n.makeTensorInfo(u.outShape,r.dtype,f.values)}return d}var yj={kernelName:Na,backendName:"cpu",kernelFunc:Aj};function xj(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{filterSize:a,strides:o,pad:i,dimRoundingMode:l,dataFormat:c}=s;Ne(r,"avgPool3d");let u=E.computePool3DInfo(r.shape,a,o,1,i,l,c),d=n.data.get(r.dataId).values,p=zS(d,r.shape,r.dtype,v.computeStrides(r.shape),u,"avg");return n.makeTensorInfo(p.shape,"float32",p.values)}var bj={kernelName:Yc,backendName:"cpu",kernelFunc:xj};function vj(e){let{inputs:t,backend:n,attrs:s}=e,{dy:r,input:a}=t,{filterSize:o,strides:i,pad:l,dimRoundingMode:c}=s;Ne([r,a],"avgPool3DGrad");let u=E.computePool3DInfo(a.shape,o,i,1,l,c),d=u.strideDepth,p=u.strideHeight,h=u.strideWidth,f=u.filterDepth,m=u.filterHeight,g=u.filterWidth,A=u.dilationDepth,x=u.dilationHeight,y=u.dilationWidth,b=u.effectiveFilterDepth,w=u.effectiveFilterHeight,k=u.effectiveFilterWidth,C=b-1-u.padInfo.front,N=k-1-u.padInfo.left,R=w-1-u.padInfo.top,F=ze(a.shape,"float32"),_=1/(f*m*g),P=n.bufferSync(r);for(let T=0;T<u.batchSize;++T)for(let M=0;M<u.inChannels;++M)for(let U=0;U<u.inDepth;++U)for(let j=0;j<u.inHeight;++j)for(let z=0;z<u.inWidth;++z){let X=U-C,Z=j-R,J=z-N,ee=0;for(let ne=0;ne<b;ne+=A){let Q=(X+ne)/d;if(!(Q<0||Q>=u.outDepth||Math.floor(Q)!==Q))for(let te=0;te<w;te+=x){let oe=(Z+te)/p;if(!(oe<0||oe>=u.outHeight||Math.floor(oe)!==oe))for(let fe=0;fe<k;fe+=y){let be=(J+fe)/h;if(be<0||be>=u.outWidth||Math.floor(be)!==be)continue;ee+=P.get(T,Q,oe,be,M)}}}F.set(ee*_,T,U,j,z,M)}return n.makeTensorInfo(F.shape,F.dtype,F.values)}var wj={kernelName:gh,backendName:"cpu",kernelFunc:vj};function kj(e){let{inputs:t,backend:n,attrs:s}=e,{dy:r,input:a}=t,o=a;Ne([r,a],"avgPoolGrad");let{filterSize:i,strides:l,pad:c}=s,u=E.computePool2DInfo(o.shape,i,l,1,c),d=u.strideHeight,p=u.strideWidth,h=u.filterHeight,f=u.filterWidth,m=u.dilationHeight,g=u.dilationWidth,A=u.effectiveFilterHeight,x=u.effectiveFilterWidth,y=x-1-u.padInfo.left,b=A-1-u.padInfo.top,w=ze(o.shape,"float32"),k=1/(h*f),C=n.data.get(r.dataId).values,N=ze(r.shape,"float32",C);for(let R=0;R<u.batchSize;++R)for(let F=0;F<u.inChannels;++F)for(let _=0;_<u.inHeight;++_)for(let P=0;P<u.inWidth;++P){let T=_-b,M=P-y,U=0;for(let j=0;j<A;j+=m){let z=(T+j)/d;if(!(z<0||z>=u.outHeight||Math.floor(z)!==z))for(let X=0;X<x;X+=g){let Z=(M+X)/p;if(Z<0||Z>=u.outWidth||Math.floor(Z)!==Z)continue;U+=N.get(R,z,Z,F)}}w.set(U*k,R,_,P,F)}return n.makeTensorInfo(w.shape,w.dtype,w.values)}var Sj={kernelName:mh,backendName:"cpu",kernelFunc:kj};function Ij(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,scale:a,offset:o,mean:i,variance:l}=t;v.assert(i.shape.length===l.shape.length,()=>"Batch normalization gradient requires mean and variance to have equal ranks."),v.assert(o==null||i.shape.length===o.shape.length,()=>"Batch normalization gradient requires mean and offset to have equal ranks."),v.assert(a==null||i.shape.length===a.shape.length,()=>"Batch normalization gradient requires mean and scale to have equal ranks."),Ne([r,i,l,a,o],"batchNorm");let{varianceEpsilon:c}=s;c==null&&(c=.001);let u=n.data.get(r.dataId).values,d=n.data.get(i.dataId).values,p=n.data.get(l.dataId).values,h=a?n.data.get(a.dataId).values:new Float32Array([1]),f=o?n.data.get(o.dataId).values:new Float32Array([0]),m=new Float32Array(u.length),g=f.length,A=h.length,x=p.length,y=d.length,b=0,w=0,k=0,C=0;for(let N=0;N<u.length;++N)m[N]=f[b++]+(u[N]-d[w++])*h[k++]/Math.sqrt(p[C++]+c),b>=g&&(b=0),w>=y&&(w=0),k>=A&&(k=0),C>=x&&(C=0);return n.makeTensorInfo(r.shape,r.dtype,m)}var Cj={kernelName:Va,backendName:"cpu",kernelFunc:Ij};function Tj(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{blockShape:a,crops:o}=s;Ne([r],"batchToSpaceND");let i=a.reduce((A,x)=>A*x),l=E.getReshaped(r.shape,a,i),c=E.getPermuted(l.length,a.length),u=E.getReshapedPermuted(r.shape,a,i),d=E.getSliceBeginCoords(o,a.length),p=E.getSliceSize(u,o,a.length),h=Rt({inputs:{x:r},backend:n,attrs:{shape:l}}),f=_s({inputs:{x:h},backend:n,attrs:{perm:c}}),m=Rt({inputs:{x:f},backend:n,attrs:{shape:u}}),g=vl({inputs:{x:m},backend:n,attrs:{begin:d,size:p}});return n.disposeIntermediateTensorInfo(h),n.disposeIntermediateTensorInfo(f),n.disposeIntermediateTensorInfo(m),g}var Nj={kernelName:di,backendName:"cpu",kernelFunc:Tj};function Ej(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,weights:a}=t,{size:o}=s,i=n.data.get(r.dataId).values,l=n.data.get(a.dataId).values,c=Py(i,l,a.dtype,a.shape,o);return n.makeTensorInfo([o],a.dtype,c)}var Rj={kernelName:Ah,backendName:"cpu",kernelFunc:Ej};function $j(e){let{inputs:t,backend:n}=e,{s0:s,s1:r}=t,a=n.data.get(s.dataId).values,o=n.data.get(r.dataId).values,i=E.assertAndGetBroadcastShape(Array.from(a),Array.from(o));return n.makeTensorInfo([i.length],"int32",Int32Array.from(i))}var _j={kernelName:yh,backendName:"cpu",kernelFunc:$j},Dj=ht(Hr,(e,t)=>{let n=t;return e>n.clipValueMax?n.clipValueMax:e<n.clipValueMin?n.clipValueMin:e}),Pj={kernelName:Hr,backendName:"cpu",kernelFunc:Dj},Fj=e=>{let{x:t}=e.inputs,n=e.backend,s=new Float32Array(v.sizeFromShape(t.shape)),r=n.data.get(t.dataId),a=r.complexTensorInfos.real,o=r.complexTensorInfos.imag,i=n.data.get(a.dataId).values,l=n.data.get(o.dataId).values;for(let c=0;c<i.length;c++){let u=i[c],d=l[c];s[c]=Math.hypot(u,d)}return n.makeOutput(s,t.shape,"float32")},Oj={kernelName:Qc,backendName:"cpu",kernelFunc:Fj};function Qu(e){let{inputs:t,backend:n}=e,{input:s}=t,r=n.data.get(s.dataId).complexTensorInfos.imag,a=n.data.get(r.dataId).values;return n.makeTensorInfo(r.shape,r.dtype,a)}var Mj={kernelName:sd,backendName:"cpu",kernelFunc:Qu};function ec(e){let{inputs:t,backend:n,attrs:s}=e,{axis:r}=s,a=v.parseAxisParam(r,t[0].shape)[0],o=E.computeOutShape(t.map(m=>m.shape),a);if(v.sizeFromShape(o)===0)return n.makeTensorInfo(o,t[0].dtype,[]);let i=t.filter(m=>v.sizeFromShape(m.shape)>0);if(i.length===1)return Mr({inputs:{x:i[0]},backend:n});let l=i.map(m=>m.shape);if(E.assertParamsConsistent(l,a),i[0].dtype==="complex64"){let m=i.map(b=>bl({inputs:{input:b},backend:n})),g=i.map(b=>Qu({inputs:{input:b},backend:n})),A=ec({inputs:m,backend:n,attrs:{axis:a}}),x=ec({inputs:g,backend:n,attrs:{axis:a}}),y=gs({inputs:{real:A,imag:x},backend:n});return m.forEach(b=>n.disposeIntermediateTensorInfo(b)),g.forEach(b=>n.disposeIntermediateTensorInfo(b)),n.disposeIntermediateTensorInfo(A),n.disposeIntermediateTensorInfo(x),y}let c=i.map(m=>{let g=v.sizeFromShape(m.shape.slice(a));return Rt({inputs:{x:m},backend:n,attrs:{shape:[-1,g]}})}),u=c.map(m=>({vals:n.data.get(m.dataId).values,shape:m.shape}));o=E.computeOutShape(c.map(m=>m.shape),1);let d=c[0].shape[0]===1,p=Fy(u,o,t[0].dtype,d),h=E.computeOutShape(i.map(m=>m.shape),a),f=n.makeTensorInfo(h,t[0].dtype,p);return c.forEach(m=>n.disposeIntermediateTensorInfo(m)),f}var zj={kernelName:pi,backendName:"cpu",kernelFunc:ec};function LS(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,filter:a}=t,{strides:o,pad:i,dataFormat:l,dilations:c,dimRoundingMode:u}=s;Ne([r,a],"conv2d");let d=E.convertConv2DDataFormat(l),p=E.computeConv2DInfo(r.shape,a.shape,o,c,i,u,!1,d),h=p.filterHeight,f=p.filterWidth,m=p.dilationHeight,g=p.dilationWidth,A=p.padInfo.left,x=p.padInfo.top,y=p.dataFormat==="channelsLast",b=new en(p.outShape,r.dtype),w=v.computeStrides(r.shape),k=v.computeStrides(a.shape),C=w[0],N=y?w[1]:w[2],R=y?w[2]:1,F=y?1:w[1],_=b.strides[0],P=y?b.strides[1]:b.strides[2],T=y?b.strides[2]:1,M=y?1:b.strides[1],U=n.data.get(r.dataId).values,j=n.data.get(a.dataId).values,z=b.values;for(let X=0;X<p.batchSize;++X){let Z=X*C,J=X*_;for(let ee=0;ee<p.outHeight;++ee){let ne=J+ee*P,Q=ee*p.strideHeight-x;for(let te=0;te<h;++te){let oe=Q+te*m;if(oe<0||oe>=p.inHeight)continue;let fe=te*k[0],be=Z+oe*N;for(let we=0;we<p.outWidth;++we){let Ce=ne+we*T,Me=we*p.strideWidth-A;for(let We=0;We<f;++We){let He=Me+We*g;if(He<0||He>=p.inWidth)continue;let qe=fe+We*k[1],ct=be+He*R,dt=qe;for(let rt=0;rt<p.inChannels;++rt){let wt=U[ct+rt*F];for(let ft=0;ft<p.outChannels;++ft)z[Ce+ft*M]+=wt*j[dt+ft];dt+=p.outChannels}}}}}}return n.makeTensorInfo(b.shape,b.dtype,z)}var Lj={kernelName:_a,backendName:"cpu",kernelFunc:LS};function Bj(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,dy:a}=t,{strides:o,pad:i,dataFormat:l,dimRoundingMode:c,filterShape:u}=s;Ne([r,a],"conv2dBackpropFilter");let d=E.convertConv2DDataFormat(l),p=E.computeConv2DInfo(r.shape,u,o,1,i,c,!1,d),{strideHeight:h,strideWidth:f,filterHeight:m,filterWidth:g}=p,A=p.dataFormat==="channelsLast",x=new en(p.filterShape,"float32"),y=p.padInfo.left,b=p.padInfo.top,w=n.data.get(r.dataId).values,k=n.data.get(a.dataId).values,C=new en(r.shape,r.dtype,w),N=new en(a.shape,a.dtype,k);for(let R=0;R<m;++R){let F=Math.max(0,Math.ceil((b-R)/h)),_=Math.min(p.outHeight,(p.inHeight+b-R)/h);for(let P=0;P<g;++P){let T=Math.max(0,Math.ceil((y-P)/f)),M=Math.min(p.outWidth,(p.inWidth+y-P)/f);for(let U=0;U<p.inChannels;++U)for(let j=0;j<p.outChannels;++j){let z=0;for(let X=0;X<p.batchSize;++X)for(let Z=F;Z<_;++Z){let J=R+Z*h-b;for(let ee=T;ee<M;++ee){let ne=P+ee*f-y;A?z+=C.get(X,J,ne,U)*N.get(X,Z,ee,j):z+=C.get(X,U,J,ne)*N.get(X,j,Z,ee)}}x.set(z,R,P,U,j)}}}return n.makeTensorInfo(x.shape,x.dtype,x.values)}var Wj={kernelName:xh,backendName:"cpu",kernelFunc:Bj};function Vj(e){let{inputs:t,backend:n,attrs:s}=e,{dy:r,filter:a}=t,{inputShape:o,strides:i,pad:l,dataFormat:c,dimRoundingMode:u}=s;Ne([r,a],"conv2dBackpropInput");let d=v.computeStrides(a.shape),p=v.computeStrides(r.shape),h=E.convertConv2DDataFormat(c),f=E.computeConv2DInfo(o,a.shape,i,1,l,u,!1,h),m=new en(f.inShape,"float32"),g=m.values,A=n.data.get(r.dataId).values,x=n.data.get(a.dataId).values,[y,b,w]=d,{batchSize:k,filterHeight:C,filterWidth:N,inChannels:R,inHeight:F,inWidth:_,outChannels:P,outHeight:T,outWidth:M,strideHeight:U,strideWidth:j}=f;h=f.dataFormat;let z=C-1-f.padInfo.top,X=N-1-f.padInfo.left,Z=h==="channelsLast",J=m.strides[0],ee=Z?m.strides[1]:m.strides[2],ne=Z?m.strides[2]:1,Q=Z?1:m.strides[1],te=p[0],oe=Z?p[1]:p[2],fe=Z?p[2]:1,be=Z?1:p[1];for(let we=0;we<k;++we)for(let Ce=0;Ce<R;++Ce)for(let Me=0;Me<F;++Me){let We=Me-z,He=Math.max(0,Math.ceil(We/U)),qe=Math.min(T,(C+We)/U);for(let ct=0;ct<_;++ct){let dt=ct-X,rt=Math.max(0,Math.ceil(dt/j)),wt=Math.min(M,(N+dt)/j),ft=0;for(let _t=He;_t<qe;++_t){let ws=_t*U-We;for(let wn=rt;wn<wt;++wn){let nr=wn*j-dt,Pn=te*we+oe*_t+fe*wn,ls=y*(C-1-ws)+b*(N-1-nr)+w*Ce;for(let Ls=0;Ls<P;++Ls){let ks=A[Pn+be*Ls],kn=x[ls+Ls];ft+=ks*kn}}}let Ct=J*we+ee*Me+ne*ct+Q*Ce;g[Ct]=ft}}return n.makeTensorInfo(m.shape,m.dtype,m.values)}var Uj={kernelName:Da,backendName:"cpu",kernelFunc:Vj};function Gj(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,filter:a}=t,{strides:o,pad:i,dilations:l}=s;Ne([r,a],"conv3d");let c=E.computeConv3DInfo(r.shape,a.shape,o,l,i),{filterDepth:u,filterHeight:d,filterWidth:p,dilationDepth:h,dilationHeight:f,dilationWidth:m,padInfo:g}=c,A=g.front,x=g.left,y=g.top,b=new en(c.outShape,r.dtype),w=n.data.get(r.dataId).values,k=n.data.get(a.dataId).values,C=b.values,N=v.computeStrides(r.shape),R=v.computeStrides(a.shape);for(let F=0;F<c.batchSize;++F){let _=F*N[0],P=F*b.strides[0];for(let T=0;T<c.outDepth;++T){let M=P+T*b.strides[1],U=T*c.strideDepth-A;for(let j=0;j<u;++j){let z=U+j*h;if(z<0||z>=c.inDepth)continue;let X=j*R[0],Z=_+z*N[1];for(let J=0;J<c.outHeight;++J){let ee=M+J*b.strides[2],ne=J*c.strideHeight-y;for(let Q=0;Q<d;++Q){let te=ne+Q*f;if(te<0||te>=c.inHeight)continue;let oe=X+Q*R[1],fe=Z+te*N[2];for(let be=0;be<c.outWidth;++be){let we=ee+be*c.outChannels,Ce=be*c.strideWidth-x;for(let Me=0;Me<p;++Me){let We=Ce+Me*m;if(We<0||We>=c.inWidth)continue;let He=oe+Me*R[2],qe=fe+We*c.inChannels,ct=He;for(let dt=0;dt<c.inChannels;++dt){let rt=w[qe+dt];for(let wt=0;wt<c.outChannels;++wt)C[we+wt]+=rt*k[ct+wt];ct+=c.outChannels}}}}}}}}return n.makeTensorInfo(b.shape,b.dtype,b.values)}var Hj={kernelName:ed,backendName:"cpu",kernelFunc:Gj};function jj(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,dy:a}=t,{strides:o,pad:i,filterShape:l}=s;Ne([r,a],"conv3dBackpropFilterV2");let c=v.computeStrides(r.shape),u=v.computeStrides(a.shape),d=E.computeConv3DInfo(r.shape,l,o,1,i),p=d.strideDepth,h=d.strideHeight,f=d.strideWidth,m=d.filterDepth,g=d.filterHeight,A=d.filterWidth,x=new en(d.filterShape,"float32"),y=x.values,[b,w,k,C]=x.strides,N=n.data.get(a.dataId).values,[R,F,_,P]=u,T=n.data.get(r.dataId).values,[M,U,j,z]=c,X=d.padInfo.front,Z=d.padInfo.left,J=d.padInfo.top;for(let ee=0;ee<m;++ee){let ne=Math.max(0,Math.ceil((X-ee)/p)),Q=Math.min(d.outDepth,(d.inDepth+X-ee)/p),te=ee*b;for(let oe=0;oe<g;++oe){let fe=Math.max(0,Math.ceil((J-oe)/h)),be=Math.min(d.outHeight,(d.inHeight+J-oe)/h),we=oe*w+te;for(let Ce=0;Ce<A;++Ce){let Me=Math.max(0,Math.ceil((Z-Ce)/f)),We=Math.min(d.outWidth,(d.inWidth+Z-Ce)/f),He=Ce*k+we;for(let qe=0;qe<d.inChannels;++qe){let ct=qe*C+He;for(let dt=0;dt<d.outChannels;++dt){let rt=0;for(let wt=0;wt<d.batchSize;++wt){let ft=wt*M,Ct=wt*R;for(let _t=ne;_t<Q;++_t){let wn=(ee+_t*p-X)*U+ft,nr=_t*F+Ct;for(let Pn=fe;Pn<be;++Pn){let Ls=(oe+Pn*h-J)*j+wn,ks=Pn*_+nr;for(let kn=Me;kn<We;++kn){let En=(Ce+kn*f-Z)*z+Ls,wr=kn*P+ks;rt+=T[En+qe]*N[wr+dt]}}}}y[ct+dt]=rt}}}}}return n.makeTensorInfo(x.shape,x.dtype,x.values)}var qj={kernelName:bh,backendName:"cpu",kernelFunc:jj};function Xj(e){let{inputs:t,backend:n,attrs:s}=e,{dy:r,filter:a}=t,{pad:o,strides:i,inputShape:l}=s;Ne([r],"conv3dBackpropInputV2");let c=v.computeStrides(r.shape),u=v.computeStrides(a.shape),d=E.computeConv3DInfo(l,a.shape,i,1,o),p=new en(d.inShape,"float32"),h=p.values,[f,m,g,A]=p.strides,x=n.data.get(r.dataId).values,[y,b,w,k]=c,C=n.data.get(a.dataId).values,[N,R,F,_]=u,{batchSize:P,filterDepth:T,filterHeight:M,filterWidth:U,inChannels:j,inDepth:z,inHeight:X,inWidth:Z,outChannels:J,outDepth:ee,outHeight:ne,outWidth:Q,strideDepth:te,strideHeight:oe,strideWidth:fe}=d,be=T-1-d.padInfo.front,we=M-1-d.padInfo.top,Ce=U-1-d.padInfo.left;for(let Me=0;Me<P;++Me)for(let We=0;We<j;++We)for(let He=0;He<z;++He){let qe=He-be,ct=Math.max(0,Math.ceil(qe/te)),dt=Math.min(ee,(T+qe)/te);for(let rt=0;rt<X;++rt){let wt=rt-we,ft=Math.max(0,Math.ceil(wt/oe)),Ct=Math.min(ne,(M+wt)/oe);for(let _t=0;_t<Z;++_t){let ws=_t-Ce,wn=Math.max(0,Math.ceil(ws/fe)),nr=Math.min(Q,(U+ws)/fe),Pn=0;for(let ls=ct;ls<dt;++ls){let Ls=ls*te-qe;for(let ks=ft;ks<Ct;++ks){let kn=ks*oe-wt;for(let vr=wn;vr<nr;++vr){let En=vr*fe-ws,wr=y*Me+b*ls+w*ks+k*vr,kr=N*(T-1-Ls)+R*(M-1-kn)+F*(U-1-En)+_*We;for(let fa=0;fa<J;++fa){let _c=x[wr+fa],sr=C[kr+fa];Pn+=_c*sr}}}}h[f*Me+m*He+g*rt+A*_t+We]=Pn}}}return n.makeTensorInfo(p.shape,p.dtype,p.values)}var Kj={kernelName:vh,backendName:"cpu",kernelFunc:Xj},Zj=ht(Pa,e=>Math.cos(e)),Yj={kernelName:Pa,backendName:"cpu",kernelFunc:Zj},Jj=ht(Fa,e=>Math.cosh(e)),Qj={kernelName:Fa,backendName:"cpu",kernelFunc:Jj};function eq(e){let{inputs:t,backend:n,attrs:s}=e,{image:r,boxes:a,boxInd:o}=t,{cropSize:i,method:l,extrapolationValue:c}=s,[u,d,p,h]=r.shape,f=a.shape[0],[m,g]=i,A=ze([f,m,g,h],"float32"),x=n.data.get(a.dataId).values,y=n.data.get(o.dataId).values,b=n.data.get(r.dataId).values,w=v.computeStrides(r.shape),k=v.computeStrides(A.shape);for(let C=0;C<f;C++){let N=C*4,R=x[N],F=x[N+1],_=x[N+2],P=x[N+3],T=y[C];if(T>=u)continue;let M=m>1?(_-R)*(d-1)/(m-1):0,U=g>1?(P-F)*(p-1)/(g-1):0;for(let j=0;j<m;j++){let z=m>1?R*(d-1)+j*M:.5*(R+_)*(d-1);if(z<0||z>d-1){for(let X=0;X<g;X++)for(let Z=0;Z<h;Z++){let J=Z+X*k[2]+j*k[1]+C*k[0];A.values[J]=c}continue}if(l==="bilinear"){let X=Math.floor(z),Z=Math.ceil(z),J=z-X;for(let ee=0;ee<g;ee++){let ne=g>1?F*(p-1)+ee*U:.5*(F+P)*(p-1);if(ne<0||ne>p-1){for(let fe=0;fe<h;fe++){let be=fe+ee*k[2]+j*k[1]+C*k[0];A.values[be]=c}continue}let Q=Math.floor(ne),te=Math.ceil(ne),oe=ne-Q;for(let fe=0;fe<h;fe++){let be=fe+Q*w[2]+X*w[1]+T*w[0],we=b[be];be=fe+te*w[2]+X*w[1]+T*w[0];let Ce=b[be];be=fe+Q*w[2]+Z*w[1]+T*w[0];let Me=b[be];be=fe+te*w[2]+Z*w[1]+T*w[0];let We=b[be],He=we+(Ce-we)*oe,qe=Me+(We-Me)*oe;be=fe+ee*k[2]+j*k[1]+C*k[0],A.values[be]=He+(qe-He)*J}}}else for(let X=0;X<g;++X){let Z=g>1?F*(p-1)+X*U:.5*(F+P)*(p-1);if(Z<0||Z>p-1){for(let ne=0;ne<h;ne++){let Q=ne+X*k[2]+j*k[1]+C*k[0];A.values[Q]=c}continue}let J=Math.round(Z),ee=Math.round(z);for(let ne=0;ne<h;ne++){let Q=ne+J*w[2]+ee*w[1]+T*w[0],te=ne+X*k[2]+j*k[1]+C*k[0];A.values[te]=b[Q]}}}}return n.makeTensorInfo(A.shape,A.dtype,A.values)}var tq={kernelName:fi,backendName:"cpu",kernelFunc:eq};function nq(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a,exclusive:o,reverse:i}=s;Ne(r,"cumsum");let l=E.getAxesPermutation([a],r.shape.length),c=r;l!=null&&(c=_s({inputs:{x:r},backend:n,attrs:{perm:l}}));let u=E.getInnerMostAxes(1,r.shape.length)[0];if(u!==c.shape.length-1)throw new Error(`backend.cumsum in CPU expects an inner-most axis=${c.shape.length-1} but got axis=${u}`);let d=zn(c.dtype,"int32"),p=v.makeZerosTypedArray(v.sizeFromShape(c.shape),d),h=n.data.get(c.dataId).values,f=c.shape[c.shape.length-1],m=i?(A,x)=>A+f-x-1:(A,x)=>A+x;for(let A=0;A<h.length;A+=f)for(let x=0;x<f;x++){let y=m(A,x);if(x===0)p[y]=o?0:h[y];else{let b=m(A,x-1);p[y]=o?h[b]+p[b]:h[y]+p[b]}}let g=n.makeTensorInfo(c.shape,d,p);if(l!=null){let A=E.getUndoAxesPermutation(l),x=_s({inputs:{x:g},backend:n,attrs:{perm:A}});return n.disposeIntermediateTensorInfo(g),n.disposeIntermediateTensorInfo(c),x}return g}var sq={kernelName:hi,backendName:"cpu",kernelFunc:nq};function rq(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,weights:a}=t,{size:o,binaryOutput:i}=s;if(r.shape.length===1){let l=n.data.get(r.dataId).values,c=n.data.get(a.dataId).values,u=Py(l,c,a.dtype,a.shape,o);return n.makeTensorInfo([o],a.dtype,u)}else if(r.shape.length===2){let l=n.bufferSync(r),c=n.bufferSync(a),u=K7(l,c,o,i);return n.makeTensorInfo(u.shape,a.dtype,u.values)}throw new Error(`Error in denseBincount: input must be at most rank 2, but got rank${r.shape.length}.`)}var aq={kernelName:wh,backendName:"cpu",kernelFunc:rq};function oq(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{blockSize:a,dataFormat:o}=s;v.assert(o==="NHWC",()=>`Only NHWC dataFormat supported on CPU for depthToSpace. Got ${o}`);let i=r.shape[0],l=r.shape[1],c=r.shape[2],u=r.shape[3],d=l*a,p=c*a,h=u/(a*a),f=n.data.get(r.dataId).values,m=new Float32Array(i*d*p*h),g=0;for(let A=0;A<i;++A)for(let x=0;x<d;++x){let y=Math.floor(x/a),b=x%a;for(let w=0;w<p;++w){let k=Math.floor(w/a),C=w%a,N=(b*a+C)*h;for(let R=0;R<h;++R){let _=R+N+u*(k+c*(y+l*A));m[g++]=f[_]}}}return n.makeTensorInfo([i,d,p,h],r.dtype,m)}var iq={kernelName:mi,backendName:"cpu",kernelFunc:oq};function BS(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,filter:a}=t,{strides:o,pad:i,dilations:l,dimRoundingMode:c}=s;Ne([r,a],"depthwiseConv2DNative");let u=v.computeStrides(r.shape),d=v.computeStrides(a.shape),p=l;p==null&&(p=[1,1]),v.assert(E.eitherStridesOrDilationsAreOne(o,p),()=>`Error in depthwiseConv2d: Either strides or dilations must be 1. Got strides ${o} and dilations '${p}'`);let h=E.computeConv2DInfo(r.shape,a.shape,o,p,i,c,!0),{filterHeight:f,filterWidth:m,dilationHeight:g,dilationWidth:A,padInfo:x}=h,y=x.left,b=x.top,w=h.outChannels/h.inChannels,k=new en(h.outShape,r.dtype),C=n.data.get(r.dataId).values,N=n.data.get(a.dataId).values,R=k.values;for(let F=0;F<h.batchSize;++F){let _=F*u[0],P=F*k.strides[0];for(let T=0;T<h.outHeight;++T){let M=P+T*k.strides[1],U=T*h.strideHeight-b;for(let j=0;j<f;++j){let z=U+j*g;if(z<0||z>=h.inHeight)continue;let X=j*d[0],Z=_+z*u[1];for(let J=0;J<h.outWidth;++J){let ee=M+J*k.strides[2],ne=J*h.strideWidth-y;for(let Q=0;Q<m;++Q){let te=ne+Q*A;if(te<0||te>=h.inWidth)continue;let oe=X+Q*d[1],fe=Z+te*h.inChannels,be=ee,we=oe;for(let Ce=0;Ce<h.inChannels;++Ce){let Me=C[fe+Ce];for(let We=0;We<w;++We)R[be+We]+=Me*N[we+We];be+=w,we+=w}}}}}}return n.makeTensorInfo(k.shape,k.dtype,k.values)}var lq={kernelName:Oa,backendName:"cpu",kernelFunc:BS};function uq(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,dy:a}=t,{strides:o,dilations:i,pad:l,dimRoundingMode:c,filterShape:u}=s;Ne([r,a],"depthwiseConv2dNativeBackpropFilter");let d=E.computeConv2DInfo(r.shape,u,o,i,l,c,!0),{strideHeight:p,strideWidth:h,filterHeight:f,filterWidth:m}=d,g=new en(d.filterShape,"float32"),A=d.padInfo.left,x=d.padInfo.top,y=d.outChannels/d.inChannels,b=n.data.get(r.dataId).values,w=new en(r.shape,r.dtype,b),k=n.data.get(a.dataId).values,C=new en(a.shape,a.dtype,k);for(let N=0;N<f;++N){let R=Math.max(0,Math.ceil((x-N)/p)),F=Math.min(d.outHeight,(d.inHeight+x-N)/p);for(let _=0;_<m;++_){let P=Math.max(0,Math.ceil((A-_)/h)),T=Math.min(d.outWidth,(d.inWidth+A-_)/h);for(let M=0;M<d.outChannels;++M){let U=Math.trunc(M/y),j=M%y,z=0;for(let X=0;X<d.batchSize;++X)for(let Z=R;Z<F;++Z){let J=N+Z*p-x;for(let ee=P;ee<T;++ee){let ne=_+ee*h-A;z+=w.get(X,J,ne,U)*C.get(X,Z,ee,M)}}g.set(z,N,_,U,j)}}}return n.makeTensorInfo(g.shape,g.dtype,g.values)}var cq={kernelName:kh,backendName:"cpu",kernelFunc:uq};function dq(e){let{inputs:t,backend:n,attrs:s}=e,{dy:r,filter:a}=t,{strides:o,dilations:i,pad:l,dimRoundingMode:c,inputShape:u}=s;Ne([r,a],"depthwiseConv2DNativeBackpropInput");let d=v.computeStrides(r.shape),p=v.computeStrides(a.shape),h=E.computeConv2DInfo(u,a.shape,o,i,l,c,!0),f=new en(h.inShape,"float32"),m=f.values,[g,A,x]=f.strides,y=n.data.get(r.dataId).values,[b,w,k]=d,C=n.data.get(a.dataId).values,[N,R,F]=p,{batchSize:_,filterHeight:P,filterWidth:T,inChannels:M,inHeight:U,inWidth:j,outChannels:z,outHeight:X,outWidth:Z,strideHeight:J,strideWidth:ee}=h,ne=P-1-h.padInfo.top,Q=T-1-h.padInfo.left,te=z/M;for(let oe=0;oe<_;++oe)for(let fe=0;fe<M;++fe)for(let be=0;be<U;++be){let we=be-ne,Ce=Math.max(0,Math.ceil(we/J)),Me=Math.min(X,(P+we)/J);for(let We=0;We<j;++We){let He=We-Q,qe=Math.max(0,Math.ceil(He/ee)),ct=Math.min(Z,(T+He)/ee),dt=0;for(let rt=Ce;rt<Me;++rt){let wt=rt*J-we;for(let ft=qe;ft<ct;++ft){let Ct=ft*ee-He,_t=b*oe+w*rt+k*ft,ws=N*(P-1-wt)+R*(T-1-Ct)+F*fe;for(let wn=0;wn<te;++wn){let nr=fe*te+wn,Pn=y[_t+nr],ls=C[ws+wn];dt+=Pn*ls}}}m[g*oe+A*be+x*We+fe]=dt}}return n.makeTensorInfo(f.shape,f.dtype,f.values)}var pq={kernelName:Sh,backendName:"cpu",kernelFunc:dq};function hq(e){let{inputs:t,backend:n}=e,{x:s}=t,r=v.sizeFromShape(s.shape),a=n.data.get(s.dataId).values,o=ze([r,r],s.dtype),i=o.values;for(let c=0;c<a.length;c++)i[c*r+c]=a[c];let l=[...s.shape,...s.shape];return n.makeTensorInfo(l,o.dtype,o.values)}var fq={kernelName:Ih,backendName:"cpu",kernelFunc:hq},mq={kernelName:td,backendName:"cpu",kernelFunc:({inputs:e,backend:t,attrs:n})=>{let{x:s,filter:r}=e,{strides:a,pad:o,dilations:i}=n,l=t,c=l.data.get(s.dataId).values,u=s.shape.length,d=l.data.get(r.dataId).values,p=r.shape.length,{batchSize:h,inHeight:f,inWidth:m,inChannels:g,outHeight:A,outWidth:x,padInfo:y,strideHeight:b,strideWidth:w,filterHeight:k,filterWidth:C,dilationHeight:N,dilationWidth:R,outShape:F}=E.computeDilation2DInfo(s.shape,r.shape,a,o,"NHWC",i),_=v.sizeFromShape(F),P=F.length,T=v.getArrayFromDType(s.dtype,_);for(let U=0;U<h;++U)for(let j=0;j<A;++j){let z=j*b-y.top;for(let X=0;X<x;++X){let Z=X*w-y.left;for(let J=0;J<g;++J){let ee=Number.MIN_SAFE_INTEGER;for(let Q=0;Q<k;++Q){let te=z+Q*N;if(te>=0&&te<f)for(let oe=0;oe<C;++oe){let fe=Z+oe*R;if(fe>=0&&fe<m){let be=v.locToIndex([U,te,fe,J],u,v.computeStrides(s.shape)),we=v.locToIndex([Q,oe,J],p,v.computeStrides(r.shape)),Ce=c[be]+d[we];Ce>ee&&(ee=Ce)}}}let ne=v.locToIndex([U,j,X,J],P,v.computeStrides(F));T[ne]=ee}}}return{dataId:l.write(v.toTypedArray(T,s.dtype),F,s.dtype),shape:F,dtype:s.dtype}}},gq={kernelName:Th,backendName:"cpu",kernelFunc:({inputs:e,backend:t,attrs:n})=>{let{x:s,filter:r,dy:a}=e,{strides:o,pad:i,dilations:l}=n,c=t,u=v.toNestedArray(s.shape,c.data.get(s.dataId).values),d=v.toNestedArray(r.shape,c.data.get(r.dataId).values),{batchSize:p,inHeight:h,inWidth:f,inChannels:m,outHeight:g,outWidth:A,padInfo:x,strideHeight:y,strideWidth:b,filterHeight:w,filterWidth:k,dilationHeight:C,dilationWidth:N,outShape:R}=E.computeDilation2DInfo(s.shape,r.shape,o,i,"NHWC",l);v.assert(a.rank===R.length,()=>`Error in ${Th}, dy must have the same rank as output ${R.length}, but got ${a.rank}`);let F=v.toNestedArray(R,c.data.get(a.dataId).values),_=v.makeZerosNestedTypedArray(r.shape,r.dtype);for(let T=0;T<p;++T)for(let M=0;M<g;++M){let U=M*y-x.top;for(let j=0;j<A;++j){let z=j*b-x.left;for(let X=0;X<m;++X){let Z=Number.MIN_SAFE_INTEGER,J=0,ee=0;for(let ne=0;ne<w;++ne){let Q=U+ne*C;if(Q>=0&&Q<h)for(let te=0;te<k;++te){let oe=z+te*N;if(oe>=0&&oe<f){let fe=u[T][Q][oe][X]+d[ne][te][X];fe>Z&&(Z=fe,J=ne,ee=te)}}}_[J][ee][X]+=F[T][M][j][X]}}}return{dataId:c.write(v.toTypedArray(_,s.dtype),r.shape,r.dtype),shape:r.shape,dtype:r.dtype}}},Aq={kernelName:Ch,backendName:"cpu",kernelFunc:({inputs:e,backend:t,attrs:n})=>{let{x:s,filter:r,dy:a}=e,{strides:o,pad:i,dilations:l}=n,c=t,u=v.toNestedArray(s.shape,c.data.get(s.dataId).values),d=v.toNestedArray(r.shape,c.data.get(r.dataId).values),{batchSize:p,inHeight:h,inWidth:f,inChannels:m,outHeight:g,outWidth:A,padInfo:x,strideHeight:y,strideWidth:b,filterHeight:w,filterWidth:k,dilationHeight:C,dilationWidth:N,outShape:R}=E.computeDilation2DInfo(s.shape,r.shape,o,i,"NHWC",l);v.assert(a.rank===R.length,()=>`Error in ${Ch}, dy must have the same rank as output ${R.length}, but got ${a.rank}`);let F=v.toNestedArray(R,c.data.get(a.dataId).values),_=v.makeZerosNestedTypedArray(s.shape,s.dtype);for(let T=0;T<p;++T)for(let M=0;M<g;++M){let U=M*y-x.top;for(let j=0;j<A;++j){let z=j*b-x.left;for(let X=0;X<m;++X){let Z=Number.MIN_SAFE_INTEGER,J=U<0?0:U,ee=z<0?0:z;for(let ne=0;ne<w;++ne){let Q=U+ne*C;if(Q>=0&&Q<h)for(let te=0;te<k;++te){let oe=z+te*N;if(oe>=0&&oe<f){let fe=u[T][Q][oe][X]+d[ne][te][X];fe>Z&&(Z=fe,J=Q,ee=oe)}}}_[T][J][ee][X]+=F[T][M][j][X]}}}return{dataId:c.write(v.toTypedArray(_,s.dtype),s.shape,s.dtype),shape:s.shape,dtype:s.dtype}}};function rp(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a,keepDims:o}=s;Ne(r,"sum");let i;r.dtype==="bool"?i=Bo({inputs:{x:r},backend:n,attrs:{dtype:"int32"}}):i=Mr({inputs:{x:r},backend:n});let l=i.shape.length,c=v.parseAxisParam(a,i.shape),u=E.getAxesPermutation(c,l),d=c,p=i;u!=null&&(p=_s({inputs:{x:i},backend:n,attrs:{perm:u}}),d=E.getInnerMostAxes(d.length,l)),E.assertAxesAreInnerMostDims("sum",d,p.shape.length);let[h,f]=E.computeOutAndReduceShapes(p.shape,d),m=E.upcastType(p.dtype,"int32"),g=vm(n,h,m),A=v.sizeFromShape(f),x=n.data.get(g.dataId).values,y=n.data.get(p.dataId).values;for(let b=0;b<x.length;++b){let w=b*A,k=0;for(let C=0;C<A;++C)k+=y[w+C];x[b]=k}if(o){let b=E.expandShapeToKeepDim(g.shape,c),w=g;g=Rt({inputs:{x:g},backend:n,attrs:{shape:b}}),n.disposeIntermediateTensorInfo(w)}return n.disposeIntermediateTensorInfo(i),u!=null&&n.disposeIntermediateTensorInfo(p),g}var yq={kernelName:co,backendName:"cpu",kernelFunc:rp};function xq(e){let{inputs:t,backend:n,attrs:s}=e,{equation:r}=s,a=t,{allDims:o,summedDims:i,idDims:l}=E.decodeEinsumEquation(r,a.length);E.checkEinsumDimSizes(o.length,l,a);let{path:c,steps:u}=E.getEinsumComputePath(i,l),d=u.length,p=null,h=o.length,f=[];for(let m=0;m<d;++m){for(let g of u[m]){let{permutationIndices:A,expandDims:x}=E.getEinsumPermutation(h,l[g]),y;E.isIdentityPermutation(A)?y=a[g]:(y=_s({inputs:{x:a[g]},backend:n,attrs:{perm:A}}),f.push(y));let b=y.shape.slice();for(let w=0;w<x.length;++w)b.splice(x[w],0,1);v.arraysEqual(y.shape,b)||(y=Rt({inputs:{x:y},backend:n,attrs:{shape:b}}),f.push(y)),p===null?p=y:(p=wm({inputs:{a:y,b:p},backend:n}),f.push(p))}m<d-1&&(c[m]>=0&&(p=rp({inputs:{x:p},backend:n,attrs:{axis:c[m]-(o.length-h),keepDims:!1}}),f.push(p)),h--)}for(let m of f)m!==p&&n.disposeIntermediateTensorInfo(m);return p}var bq={kernelName:nd,backendName:"cpu",kernelFunc:xq};function vq(e){let{inputs:t,backend:n}=e,{dy:s,y:r}=t;Ne([s,r],"eluGrad");let a=new Float32Array(v.sizeFromShape(r.shape)),o=n.data.get(r.dataId).values,i=n.data.get(s.dataId).values;for(let l=0;l<o.length;++l){let c=o[l];c>=1?a[l]=i[l]:a[l]=i[l]*(c+1)}return n.makeTensorInfo(r.shape,"float32",a)}var wq={kernelName:Nh,backendName:"cpu",kernelFunc:vq},kq=E.ERF_P,Sq=E.ERF_A1,Iq=E.ERF_A2,Cq=E.ERF_A3,Tq=E.ERF_A4,Nq=E.ERF_A5,Eq=ht(cu,e=>{let t=Math.sign(e),n=Math.abs(e),s=1/(1+kq*n);return t*(1-((((Nq*s+Tq)*s+Cq)*s+Iq)*s+Sq)*s*Math.exp(-n*n))}),Rq={kernelName:cu,backendName:"cpu",kernelFunc:Eq};function Sm(e){let{inputs:t,backend:n,attrs:s}=e,{input:r}=t,{dim:a}=s,o=r.shape.length,i=r.shape.slice(),l=a;return a<0&&(v.assert(-(o+1)<=a,()=>`Axis must be in the interval [${-(o+1)}, ${o}]`),l=o+a+1),i.splice(l,0,1),Rt({inputs:{x:r},backend:n,attrs:{shape:i}})}var $q={kernelName:Ai,backendName:"cpu",kernelFunc:Sm},_q=Zt((e,t)=>e/t),Uy=bn(Ma,_q),Gy={kernelName:Ma,backendName:"cpu",kernelFunc:Uy};function WS(e,t,n){let s=e.shape,r=s[0],a=s[1],o=n.data.get(e.dataId),i=o.complexTensorInfos.real,l=o.complexTensorInfos.imag,c=[r,a],u=v.sizeFromShape(c),d=v.getTypedArrayFromDType("float32",u),p=v.getTypedArrayFromDType("float32",u);for(let g=0;g<r;g++){let A=vl({inputs:{x:i},backend:n,attrs:{begin:[g,0],size:[1,a]}}),x=vl({inputs:{x:l},backend:n,attrs:{begin:[g,0],size:[1,a]}}),y=gs({inputs:{real:A,imag:x},backend:n}),{real:b,imag:w}=Dq(y,t,n),k=E.mergeRealAndImagArrays(b,w);for(let C=0;C<a;C++){let N=E.getComplexWithIndex(k,C);d[g*a+C]=N.real,p[g*a+C]=N.imag}n.disposeIntermediateTensorInfo(A),n.disposeIntermediateTensorInfo(x),n.disposeIntermediateTensorInfo(y)}let h=n.makeTensorInfo(c,"float32",d),f=n.makeTensorInfo(c,"float32",p),m=gs({inputs:{real:h,imag:f},backend:n});return n.disposeIntermediateTensorInfo(h),n.disposeIntermediateTensorInfo(f),m}function Dq(e,t,n){let s=v.sizeFromShape(e.shape),r=n.data.get(e.dataId),a=n.data.get(r.complexTensorInfos.real.dataId).values,o=n.data.get(r.complexTensorInfos.imag.dataId).values;if(Pq(s)){let i=Hy(a,o,s,t,n),l=[e.shape[0],e.shape[1]];if(t){let c=n.makeTensorInfo(l,"float32",i.real),u=n.makeTensorInfo(l,"float32",i.imag),d=n.makeTensorInfo([],"float32",v.createScalarValue(s,"float32")),p=Mr({inputs:{x:d},backend:n}),h=Gy.kernelFunc({inputs:{a:c,b:d},backend:n}),f=Gy.kernelFunc({inputs:{a:u,b:p},backend:n}),m=n.data.get(h.dataId).values,g=n.data.get(f.dataId).values;return n.disposeIntermediateTensorInfo(c),n.disposeIntermediateTensorInfo(u),n.disposeIntermediateTensorInfo(d),n.disposeIntermediateTensorInfo(p),n.disposeIntermediateTensorInfo(h),n.disposeIntermediateTensorInfo(f),{real:m,imag:g}}return i}else{let i=E.mergeRealAndImagArrays(a,o),l=Fq(i,s,t);return E.splitRealAndImagArrays(l)}}function Pq(e){return(e&e-1)==0}function Hy(e,t,n,s,r){if(n===1)return{real:e,imag:t};let a=E.mergeRealAndImagArrays(e,t),o=n/2,i=E.complexWithEvenIndex(a),l=i.real,c=i.imag,u=[l.length],d=r.makeTensorInfo(u,"float32",l),p=r.makeTensorInfo(u,"float32",c),h=gs({inputs:{real:d,imag:p},backend:r}),f=E.complexWithOddIndex(a),m=f.real,g=f.imag,A=[m.length],x=r.makeTensorInfo(A,"float32",m),y=r.makeTensorInfo(A,"float32",g),b=gs({inputs:{real:x,imag:y},backend:r}),w=Hy(l,c,o,s,r),k=w.real,C=w.imag,N=[k.length],R=r.makeTensorInfo(N,"float32",k),F=r.makeTensorInfo(N,"float32",C),_=gs({inputs:{real:R,imag:F},backend:r}),P=Hy(m,g,o,s,r),T=P.real,M=P.imag,U=[T.length],j=r.makeTensorInfo(U,"float32",T),z=r.makeTensorInfo(U,"float32",M),X=gs({inputs:{real:j,imag:z},backend:r}),Z=E.exponents(n,s),J=[Z.real.length],ee=r.makeTensorInfo(J,"float32",Z.real),ne=r.makeTensorInfo(J,"float32",Z.imag),Q=gs({inputs:{real:ee,imag:ne},backend:r}),te=wm({inputs:{a:Q,b:X},backend:r}),oe=np({inputs:{a:_,b:te},backend:r}),fe=By({inputs:{a:_,b:te},backend:r}),be=bl({inputs:{input:oe},backend:r}),we=bl({inputs:{input:fe},backend:r}),Ce=Qu({inputs:{input:oe},backend:r}),Me=Qu({inputs:{input:fe},backend:r}),We=ec({inputs:[be,we],backend:r,attrs:{axis:0}}),He=ec({inputs:[Ce,Me],backend:r,attrs:{axis:0}}),qe=r.data.get(We.dataId).values,ct=r.data.get(He.dataId).values;return r.disposeIntermediateTensorInfo(d),r.disposeIntermediateTensorInfo(p),r.disposeIntermediateTensorInfo(h),r.disposeIntermediateTensorInfo(x),r.disposeIntermediateTensorInfo(y),r.disposeIntermediateTensorInfo(b),r.disposeIntermediateTensorInfo(R),r.disposeIntermediateTensorInfo(F),r.disposeIntermediateTensorInfo(_),r.disposeIntermediateTensorInfo(j),r.disposeIntermediateTensorInfo(z),r.disposeIntermediateTensorInfo(X),r.disposeIntermediateTensorInfo(ee),r.disposeIntermediateTensorInfo(ne),r.disposeIntermediateTensorInfo(Q),r.disposeIntermediateTensorInfo(te),r.disposeIntermediateTensorInfo(oe),r.disposeIntermediateTensorInfo(fe),r.disposeIntermediateTensorInfo(be),r.disposeIntermediateTensorInfo(Ce),r.disposeIntermediateTensorInfo(we),r.disposeIntermediateTensorInfo(Me),r.disposeIntermediateTensorInfo(We),r.disposeIntermediateTensorInfo(He),{real:qe,imag:ct}}function Fq(e,t,n){let s=new Float32Array(t*2);for(let r=0;r<t;r++){let a=0,o=0;for(let i=0;i<t;i++){let l=E.exponent(r*i,t,n),c=E.getComplexWithIndex(e,i);a+=c.real*l.real-c.imag*l.imag,o+=c.real*l.imag+c.imag*l.real}n&&(a/=t,o/=t),E.assignToTypedArray(s,a,o,r)}return s}function Oq(e){let{inputs:t,backend:n}=e,{input:s}=t,r=v.sizeFromShape(s.shape),a=s.shape[s.shape.length-1],o=r/a,i=Rt({inputs:{x:s},backend:n,attrs:{shape:[o,a]}}),l=WS(i,!1,n),c=Rt({inputs:{x:l},backend:n,attrs:{shape:s.shape}});return n.disposeIntermediateTensorInfo(i),n.disposeIntermediateTensorInfo(l),c}var Mq={kernelName:Eh,backendName:"cpu",kernelFunc:Oq};function jy(e){let{backend:t,attrs:n}=e,{shape:s,value:r,dtype:a}=n,o=a||v.inferDtype(r),i=v.getArrayFromDType(o,v.sizeFromShape(s));return Lq(i,r,o),t.makeTensorInfo(s,o,i)}var zq={kernelName:du,backendName:"cpu",kernelFunc:jy};function Lq(e,t,n){e.fill(t)}var Bq={kernelName:xi,backendName:"cpu",kernelFunc:({inputs:e,attrs:t,backend:n})=>{let{image:s}=e,r=n,a=v.getTypedArrayFromDType(s.dtype,v.sizeFromShape(s.shape)),[o,i,l,c]=s.shape,u=r.data.get(s.dataId).values;for(let p=0;p<o;p++){let h=p*l*i*c;for(let f=0;f<i;f++){let m=f*(l*c);for(let g=0;g<l;g++){let A=g*c;for(let x=0;x<c;x++){let y=Math.round(l-g-1),b=h+m+A+x,w=u[b];if(y>=0&&y<l){let k=y*c,C=h+m+k+x;w=u[C]}a[b]=w}}}}return{dataId:r.write(a,s.shape,s.dtype),shape:s.shape,dtype:s.dtype}}},Wq=Zt((e,t)=>Math.floor(e/t)),Vq=bn(Wa,Wq,null,"int32"),Uq={kernelName:Wa,backendName:"cpu",kernelFunc:Vq};function Gq(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,filter:a,bias:o,preluActivationWeights:i}=t,{strides:l,pad:c,dataFormat:u,dilations:d,dimRoundingMode:p,activation:h,leakyreluAlpha:f}=s,m=LS({inputs:{x:r,filter:a},backend:n,attrs:{strides:l,pad:c,dataFormat:u,dilations:d,dimRoundingMode:p}});if(o){let g=m;m=np({inputs:{a:m,b:o},backend:n}),n.disposeIntermediateTensorInfo(g)}if(h){let g=m;m=Wy(n,m,h,i,f),n.disposeIntermediateTensorInfo(g)}return m}var Hq={kernelName:xo,backendName:"cpu",kernelFunc:Gq};function jq(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,filter:a,bias:o,preluActivationWeights:i}=t,{strides:l,pad:c,dataFormat:u,dilations:d,dimRoundingMode:p,activation:h,leakyreluAlpha:f}=s,m=BS({inputs:{x:r,filter:a},backend:n,attrs:{strides:l,pad:c,dataFormat:u,dilations:d,dimRoundingMode:p}});if(o){let g=m;m=np({inputs:{a:m,b:o},backend:n}),n.disposeIntermediateTensorInfo(g)}if(h){let g=m;m=Wy(n,m,h,i,f),n.disposeIntermediateTensorInfo(g)}return m}var qq={kernelName:bo,backendName:"cpu",kernelFunc:jq};function Xq(e){let{inputs:t,backend:n}=e,{params:s,indices:r}=t,a=v.sizeFromShape(s.shape),o=r.shape,i=o[o.length-1],[l,c,u,d]=E.prepareAndValidate(s,r);if(c===0)return n.makeTensorInfo(l,s.dtype,[]);let p=n.data.get(r.dataId).values,h=n.bufferSync(s),f=sS(p,h,s.dtype,c,i,u,d,s.shape,a);return n.makeTensorInfo(l,s.dtype,f.values)}var Kq={kernelName:vi,backendName:"cpu",kernelFunc:Xq};function Zq(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,indices:a}=t,{axis:o,batchDims:i}=s;Ne([r,a],"gatherV2");let l=v.parseAxisParam(o,r.shape)[0],c=n.data.get(a.dataId).values,u=r.shape[l];for(let b=0;b<c.length;++b){let w=c[b];v.assert(w<=u-1&&w>=0,()=>`GatherV2: the index value ${w} is not in [0, ${u-1}]`)}let d=i;i==null&&(d=0);let p=v.sizeFromShape(a.shape),h=E.segment_util.collectGatherOpShapeInfo(r,a,l,d),f=Rt({inputs:{x:r},backend:n,attrs:{shape:[h.batchSize,h.outerSize,h.dimSize,h.sliceSize]}}),m=Rt({inputs:{x:a},backend:n,attrs:{shape:[h.batchSize,p/h.batchSize]}}),g=[h.batchSize,h.outerSize,p/h.batchSize,h.sliceSize],A=n.bufferSync(m),x=n.bufferSync(f),y=rS(x,A,g);return n.disposeIntermediateTensorInfo(f),n.disposeIntermediateTensorInfo(m),n.makeTensorInfo(h.outputShape,y.dtype,y.values)}var Yq={kernelName:bi,backendName:"cpu",kernelFunc:Zq};function Jq(e){let{inputs:t,backend:n}=e,{input:s}=t,r=v.sizeFromShape(s.shape),a=s.shape[s.shape.length-1],o=r/a,i=Rt({inputs:{x:s},backend:n,attrs:{shape:[o,a]}}),l=WS(i,!0,n),c=Rt({inputs:{x:l},backend:n,attrs:{shape:s.shape}});return n.disposeIntermediateTensorInfo(i),n.disposeIntermediateTensorInfo(l),c}var Qq={kernelName:Rh,backendName:"cpu",kernelFunc:Jq},eX=ht(pu,e=>Number.isFinite(e)?1:0,"bool"),tX={kernelName:pu,backendName:"cpu",kernelFunc:eX},nX=ht(hu,e=>Math.abs(e)===1/0?1:0,"bool"),sX={kernelName:hu,backendName:"cpu",kernelFunc:nX},rX=ht(fu,e=>Number.isNaN(e)?1:0,"bool"),aX={kernelName:fu,backendName:"cpu",kernelFunc:rX};function oX(e){let{backend:t,attrs:n}=e,{start:s,stop:r,num:a}=n,o=uS(s,r,a);return t.makeTensorInfo([o.length],"float32",o)}var iX={kernelName:$h,backendName:"cpu",kernelFunc:oX},lX=ht(mu,e=>Math.log1p(e)),uX={kernelName:mu,backendName:"cpu",kernelFunc:lX},cX=Zt((e,t)=>e&&t),dX=bn(Ci,cX,null,"bool"),pX={kernelName:Ci,backendName:"cpu",kernelFunc:dX},hX=ht(gu,e=>e?0:1,"bool"),fX={kernelName:gu,backendName:"cpu",kernelFunc:hX},mX=Zt((e,t)=>e||t),gX=bn(rd,mX,null,"bool"),AX={kernelName:rd,backendName:"cpu",kernelFunc:gX};function yX(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{depthRadius:a,bias:o,alpha:i,beta:l}=s;Ne(r,"LRN");let c=r.shape[3],u=c-1,d=n.data.get(r.dataId).values,p=v.sizeFromShape(r.shape),h=new Float32Array(p);function f(m){let g=m%c,A=m-g+Math.max(0,g-a),x=m-g+Math.min(g+a,u),y=0;for(;A<=x;A++){let b=d[A];y+=b*b}return y}for(let m=0;m<p;m++){let g=f(m),A=d[m]*Math.pow(o+i*g,-l);h[m]=A}return n.makeTensorInfo(r.shape,r.dtype,h)}var xX={kernelName:ad,backendName:"cpu",kernelFunc:yX};function bX(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,y:a,dy:o}=t,{depthRadius:i,bias:l,alpha:c,beta:u}=s;Ne(o,"LRNGrad");let d=v.sizeFromShape(o.shape),p=o.shape[3],h=n.data.get(o.dataId).values,f=n.data.get(r.dataId).values,m=n.data.get(a.dataId).values,g=new Float32Array(d),A=d;for(let x=0;x<A;x++){let y=x%p,b=x-y+Math.max(0,y-i),w=x-y+Math.min(p,y+i+1),k=0;for(let C=b;C<w;C++)k+=Math.pow(f[C],2);k=c*k+l;for(let C=b;C<w;C++){let N=-2*c*u*f[C]*m[x]/k;x===C&&(N+=Math.pow(k,-u)),N*=h[x],g[C]+=N}}return n.makeTensorInfo(o.shape,r.dtype,g)}var vX={kernelName:_h,backendName:"cpu",kernelFunc:bX};function VS(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{reductionIndices:a,keepDims:o}=s,i=n,l=r.shape,c=l.length,u=v.parseAxisParam(a,l),d=u,p=E.getAxesPermutation(d,c),h=i.data.get(r.dataId).values;if(p!=null){let b=new Array(c);for(let w=0;w<b.length;w++)b[w]=l[p[w]];h=My(h,l,r.dtype,p,b),d=E.getInnerMostAxes(d.length,c),l=b}Ne(r,"max"),E.assertAxesAreInnerMostDims("max",d,c);let[f,m]=E.computeOutAndReduceShapes(l,d),g=v.sizeFromShape(m),A=dS(h,g,f,r.dtype),x=i.write(A,f,r.dtype),y=f;return o&&(y=E.expandShapeToKeepDim(f,u)),{dataId:x,shape:y,dtype:r.dtype}}var wX={kernelName:ja,backendName:"cpu",kernelFunc:VS};function kX(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t;Ne(r,"maxPool");let{filterSize:a,strides:o,pad:i,dimRoundingMode:l}=s,c=1;v.assert(E.eitherStridesOrDilationsAreOne(o,c),()=>`Error in maxPool: Either strides or dilations must be 1. Got strides ${o} and dilations '${c}'`);let u=E.computePool2DInfo(r.shape,a,o,c,i,l),d;if(u.filterWidth===1&&u.filterHeight===1&&v.arraysEqual(u.inShape,u.outShape))d=Mr({inputs:{x:r},backend:n});else{let p=n.data.get(r.dataId).values,h=v.computeStrides(r.shape),f=Vy(p,r.shape,r.dtype,h,u,"max");d=n.makeTensorInfo(u.outShape,r.dtype,f.values)}return d}var SX={kernelName:Xa,backendName:"cpu",kernelFunc:kX};function IX(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{filterSize:a,strides:o,pad:i,dimRoundingMode:l,dataFormat:c}=s;Ne(r,"maxPool3d");let u=E.computePool3DInfo(r.shape,a,o,1,i,l,c),d=n.data.get(r.dataId).values,p=zS(d,r.shape,r.dtype,v.computeStrides(r.shape),u,"max");return n.makeTensorInfo(p.shape,"float32",p.values)}var CX={kernelName:od,backendName:"cpu",kernelFunc:IX};function TX(e){let{inputs:t,backend:n,attrs:s}=e,{dy:r,input:a}=t,{filterSize:o,strides:i,pad:l,dimRoundingMode:c}=s;Ne([r,a],"maxPool3DGrad");let u=E.computePool3DInfo(a.shape,o,i,1,l,c),d=n.bufferSync(a),p=gj(d,u),h=u.strideDepth,f=u.strideHeight,m=u.strideWidth,g=u.dilationDepth,A=u.dilationHeight,x=u.dilationWidth,y=u.effectiveFilterDepth,b=u.effectiveFilterHeight,w=u.effectiveFilterWidth,k=y-1-u.padInfo.front,C=w-1-u.padInfo.left,N=b-1-u.padInfo.top,R=ze(a.shape,"float32"),F=n.bufferSync(r);for(let _=0;_<u.batchSize;++_)for(let P=0;P<u.inChannels;++P)for(let T=0;T<u.inDepth;++T)for(let M=0;M<u.inHeight;++M)for(let U=0;U<u.inWidth;++U){let j=T-k,z=M-N,X=U-C,Z=0;for(let J=0;J<y;J+=g){let ee=(j+J)/h;if(!(ee<0||ee>=u.outDepth||Math.floor(ee)!==ee))for(let ne=0;ne<b;ne+=A){let Q=(z+ne)/f;if(!(Q<0||Q>=u.outHeight||Math.floor(Q)!==Q))for(let te=0;te<w;te+=x){let oe=(X+te)/m;if(oe<0||oe>=u.outWidth||Math.floor(oe)!==oe)continue;let fe=y*b*w-1-p.get(_,ee,Q,oe,P),be=J*b*w+ne*w+te,we=fe===be?1:0;if(we===0)continue;Z+=F.get(_,ee,Q,oe,P)*we}}}R.set(Z,_,T,M,U,P)}return n.makeTensorInfo(R.shape,R.dtype,R.values)}var NX={kernelName:Ph,backendName:"cpu",kernelFunc:TX};function EX(e){let{inputs:t,backend:n,attrs:s}=e,{dy:r,input:a,output:o}=t,i=a;Ne([a,o],"maxPoolGrad");let{filterSize:l,strides:c,pad:u,dimRoundingMode:d}=s,p=E.computePool2DInfo(i.shape,l,c,1,u,d),h=n.data.get(i.dataId).values,f=ze(p.outShape,i.dtype,MS(h,i.shape,i.dtype,p).values),m=p.strideHeight,g=p.strideWidth,A=p.dilationHeight,x=p.dilationWidth,y=p.effectiveFilterHeight,b=p.effectiveFilterWidth,w=b-1-p.padInfo.left,k=y-1-p.padInfo.top,C=ze(i.shape,"float32"),N=n.data.get(r.dataId).values,R=ze(r.shape,"float32",N);for(let F=0;F<p.batchSize;++F)for(let _=0;_<p.inChannels;++_)for(let P=0;P<p.inHeight;++P)for(let T=0;T<p.inWidth;++T){let M=P-k,U=T-w,j=0;for(let z=0;z<y;z+=A){let X=(M+z)/m;if(!(X<0||X>=p.outHeight||Math.floor(X)!==X))for(let Z=0;Z<b;Z+=x){let J=(U+Z)/g;if(J<0||J>=p.outWidth||Math.floor(J)!==J)continue;let ee=y*b-1-f.get(F,X,J,_),ne=z*b+Z,Q=ee===ne?1:0;if(Q===0)continue;j+=R.get(F,X,J,_)*Q}}C.set(j,F,P,T,_)}return n.makeTensorInfo(C.shape,C.dtype,C.values)}var RX={kernelName:Dh,backendName:"cpu",kernelFunc:EX};function $X(e,t,n,s,r){let a=v.computeStrides(t),o=Vy(e,t,n,a,r,"max"),i=MS(e,t,n,r,!0,s);return[o.values,i.values]}var _X={kernelName:Fh,backendName:"cpu",kernelFunc:({inputs:e,attrs:t,backend:n})=>{let{x:s}=e,{filterSize:r,strides:a,pad:o,includeBatchInIndex:i}=t,l=n;Ne(s,"MaxPoolWithArgmax");let c=l.data.get(s.dataId).values,u=E.computePool2DInfo(s.shape,r,a,[1,1],o),[d,p]=$X(c,s.shape,s.dtype,i,u),h=l.write(d,u.outShape,s.dtype),f=l.write(p,u.outShape,s.dtype);return[{dataId:h,shape:u.outShape,dtype:s.dtype},{dataId:f,shape:u.outShape,dtype:"int32"}]}};function DX(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a,keepDims:o}=s,i=v.parseAxisParam(a,r.shape),c=E.computeOutAndReduceShapes(r.shape,i)[1],u=v.sizeFromShape(c),d=[],p=n.makeTensorInfo([],"float32",new Float32Array([u]));d.push(p);let h=Bo({inputs:{x:r},backend:n,attrs:{dtype:"float32"}});d.push(h);let f=Uy({inputs:{a:h,b:p},backend:n});d.push(f);let m=rp({inputs:{x:f},backend:n,attrs:{axis:a,keepDims:o}});return d.forEach(g=>n.disposeIntermediateTensorInfo(g)),m}var PX={kernelName:Ka,backendName:"cpu",kernelFunc:DX};function FX(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a,keepDims:o}=s;Ne(r,"min");let i=v.parseAxisParam(a,r.shape),l=i,c=E.getAxesPermutation(l,r.shape.length),u=r;c!=null&&(u=_s({inputs:{x:r},backend:n,attrs:{perm:c}}),l=E.getInnerMostAxes(l.length,r.shape.length)),E.assertAxesAreInnerMostDims("min",l,u.shape.length);let[d,p]=E.computeOutAndReduceShapes(u.shape,l),h=v.sizeFromShape(p),f=v.makeZerosTypedArray(v.sizeFromShape(d),u.dtype),m=n.data.get(u.dataId).values;for(let A=0;A<f.length;++A){let x=A*h,y=m[x];for(let b=0;b<h;++b){let w=m[x+b];(Number.isNaN(w)||w<y)&&(y=w)}f[A]=y}c!=null&&n.disposeIntermediateTensorInfo(u);let g=n.makeTensorInfo(d,u.dtype,f);if(o){let A=E.expandShapeToKeepDim(d,i),x=Rt({inputs:{x:g},backend:n,attrs:{shape:A}});return n.disposeIntermediateTensorInfo(g),x}return g}var OX={kernelName:Za,backendName:"cpu",kernelFunc:FX};function MX(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{paddings:a,mode:o}=s;Ne(r,"mirrorPad");let i=a.map((y,b)=>y[0]+r.shape[b]+y[1]),l=a.map(y=>y[0]),c=a.map((y,b)=>y[0]+r.shape[b]),u=o==="reflect"?0:1,d=n.data.get(r.dataId).values,p=r.shape.length,h=v.computeStrides(r.shape),f=v.sizeFromShape(i),m=i.length,g=v.computeStrides(i),A=v.getTypedArrayFromDType(r.dtype,f);for(let y=0;y<f;y++){let b=v.indexToLoc(y,m,g);for(let k=0;k<m;k++)b[k]<l[k]?b[k]=l[k]*2-b[k]-u:b[k]>=c[k]&&(b[k]=(c[k]-1)*2-b[k]+u);b=b.map((k,C)=>k-l[C]);let w=v.locToIndex(b,p,h);A[y]=d[w]}return{dataId:n.write(A,i,r.dtype),shape:i,dtype:r.dtype}}var zX={kernelName:Ja,backendName:"cpu",kernelFunc:MX},LX=Zt((e,t)=>{let n=e%t;return e<0&&t<0||e>=0&&t>=0?n:(n+t)%t}),BX=bn(Au,LX),WX={kernelName:Au,backendName:"cpu",kernelFunc:BX},VX=ii(ch());function US(e){let{inputs:t,backend:n,attrs:s}=e,{logits:r}=t,{dim:a}=s,o=r.shape.length,i=a;if(i===-1&&(i=o-1),i!==o-1)throw Error(`Softmax along a non-last dimension is not yet supported. Logits was rank ${o} and dim was ${i}`);let l=v.parseAxisParam([i],r.shape),c=VS({inputs:{x:r},backend:n,attrs:{reductionIndices:l,keepDims:!1}}),u=E.expandShapeToKeepDim(c.shape,l),d=Rt({inputs:{x:c},backend:n,attrs:{shape:u}}),p=By({inputs:{a:r,b:d},backend:n}),h=eS({inputs:{x:p},backend:n}),f=rp({inputs:{x:h},backend:n,attrs:{axis:l,keepDims:!1}}),m=Rt({inputs:{x:f},backend:n,attrs:{shape:u}}),g=Uy({inputs:{a:h,b:m},backend:n});return n.disposeIntermediateTensorInfo(c),n.disposeIntermediateTensorInfo(d),n.disposeIntermediateTensorInfo(p),n.disposeIntermediateTensorInfo(h),n.disposeIntermediateTensorInfo(f),n.disposeIntermediateTensorInfo(m),g}var UX={kernelName:po,backendName:"cpu",kernelFunc:US};function GX(e){let{inputs:t,backend:n,attrs:s}=e,{logits:r}=t,{numSamples:a,seed:o,normalized:i}=s;Ne(r,"multinomial");let l=i?r:US({inputs:{logits:r},backend:n,attrs:{dim:-1}}),c=l.shape[0],u=l.shape[1],d=n.data.get(l.dataId).values,p=[c,a],h=v.makeZerosTypedArray(v.sizeFromShape(p),"int32");for(let f=0;f<c;++f){let m=f*u,g=new Float32Array(u-1);g[0]=d[m];for(let y=1;y<g.length;++y)g[y]=g[y-1]+d[m+y];let A=VX.alea(o.toString()),x=f*a;for(let y=0;y<a;++y){let b=A();h[x+y]=g.length;for(let w=0;w<g.length;w++)if(b<g[w]){h[x+y]=w;break}}}return i||n.disposeIntermediateTensorInfo(l),n.makeTensorInfo(p,"int32",h)}var HX={kernelName:Oh,backendName:"cpu",kernelFunc:GX},jX=Xs.nonMaxSuppressionV3Impl;function qX(e){let{inputs:t,backend:n,attrs:s}=e,{boxes:r,scores:a}=t,{maxOutputSize:o,iouThreshold:i,scoreThreshold:l}=s;Ne(r,"NonMaxSuppression");let c=n.data.get(r.dataId).values,u=n.data.get(a.dataId).values,{selectedIndices:d}=jX(c,u,o,i,l);return n.makeTensorInfo([d.length],"int32",new Int32Array(d))}var XX={kernelName:Ei,backendName:"cpu",kernelFunc:qX},KX=Xs.nonMaxSuppressionV4Impl;function ZX(e){let{inputs:t,backend:n,attrs:s}=e,{boxes:r,scores:a}=t,{maxOutputSize:o,iouThreshold:i,scoreThreshold:l,padToMaxOutputSize:c}=s;Ne(r,"NonMaxSuppressionPadded");let u=n.data.get(r.dataId).values,d=n.data.get(a.dataId).values,{selectedIndices:p,validOutputs:h}=KX(u,d,o,i,l,c);return[n.makeTensorInfo([p.length],"int32",new Int32Array(p)),n.makeTensorInfo([],"int32",new Int32Array([h]))]}var YX={kernelName:yu,backendName:"cpu",kernelFunc:ZX},JX=Xs.nonMaxSuppressionV5Impl;function QX(e){let{inputs:t,backend:n,attrs:s}=e,{boxes:r,scores:a}=t,{maxOutputSize:o,iouThreshold:i,scoreThreshold:l,softNmsSigma:c}=s;Ne(r,"NonMaxSuppressionWithScore");let u=n.data.get(r.dataId).values,d=n.data.get(a.dataId).values,p=o,h=i,f=l,m=c,{selectedIndices:g,selectedScores:A}=JX(u,d,p,h,f,m);return[n.makeTensorInfo([g.length],"int32",new Int32Array(g)),n.makeTensorInfo([A.length],"float32",new Float32Array(A))]}var eK={kernelName:Ri,backendName:"cpu",kernelFunc:QX};function tK(e){let{inputs:t,backend:n,attrs:s}=e,{indices:r}=t,{depth:a,onValue:o,offValue:i}=s;Ne(r,"oneHot");let l=v.sizeFromShape(r.shape),c=new Float32Array(l*a);c.fill(i);let u=n.data.get(r.dataId).values;for(let d=0;d<l;++d)u[d]>=0&&u[d]<a&&(c[d*a+u[d]]=o);return n.makeTensorInfo([...r.shape,a],"int32",c)}var nK={kernelName:_i,backendName:"cpu",kernelFunc:tK};function Im(e){let{inputs:t,backend:n}=e,{x:s}=t;if(s.dtype==="string")throw new Error("zerosLike is not supported for string tensors");if(s.dtype==="complex64"){let r=bl({inputs:{input:s},backend:n}),a=Im({inputs:{x:r},backend:n}),o=Qu({inputs:{input:s},backend:n}),i=Im({inputs:{x:o},backend:n}),l=gs({inputs:{real:a,imag:i},backend:n});return n.disposeIntermediateTensorInfo(r),n.disposeIntermediateTensorInfo(a),n.disposeIntermediateTensorInfo(o),n.disposeIntermediateTensorInfo(i),l}else return jy({backend:n,attrs:{shape:s.shape,value:0,dtype:s.dtype}})}var sK={kernelName:Ki,backendName:"cpu",kernelFunc:Im};function GS(e){let{inputs:t,backend:n}=e,{x:s}=t;if(s.dtype==="string")throw new Error("onesLike is not supported for string tensors");if(s.dtype==="complex64"){let r=bl({inputs:{input:s},backend:n}),a=GS({inputs:{x:r},backend:n}),o=Qu({inputs:{input:s},backend:n}),i=Im({inputs:{x:o},backend:n}),l=gs({inputs:{real:a,imag:i},backend:n});return n.disposeIntermediateTensorInfo(r),n.disposeIntermediateTensorInfo(a),n.disposeIntermediateTensorInfo(o),n.disposeIntermediateTensorInfo(i),l}else return jy({backend:n,attrs:{shape:s.shape,value:1,dtype:s.dtype}})}var rK={kernelName:$i,backendName:"cpu",kernelFunc:GS};function HS(e){let{inputs:t,backend:n,attrs:s}=e,{axis:r}=s;if(t.length===1)return Sm({inputs:{input:t[0]},backend:n,attrs:{dim:r}});let a=t[0].shape,o=t[0].dtype;t.forEach(u=>{v.assertShapesMatch(a,u.shape,"All tensors passed to stack must have matching shapes"),v.assert(o===u.dtype,()=>"All tensors passed to stack must have matching dtypes")});let i=[],l=t.map(u=>{let d=Sm({inputs:{input:u},backend:n,attrs:{dim:r}});return i.push(d),d}),c=ec({inputs:l,backend:n,attrs:{axis:r}});return i.forEach(u=>n.disposeIntermediateTensorInfo(u)),c}var aK={kernelName:Di,backendName:"cpu",kernelFunc:HS};function oK(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{paddings:a,constantValue:o}=s;Ne(r,"pad");let i=a.map((x,y)=>x[0]+r.shape[y]+x[1]),l=a.map(x=>x[0]),c=n.data.get(r.dataId).values,u=v.sizeFromShape(r.shape),d=r.shape.length,p=v.computeStrides(r.shape),h=v.sizeFromShape(i),f=i.length,m=v.computeStrides(i),g=v.getTypedArrayFromDType(r.dtype,h);o!==0&&g.fill(o);for(let x=0;x<u;x++){let b=v.indexToLoc(x,d,p).map((k,C)=>k+l[C]),w=v.locToIndex(b,f,m);g[w]=c[x]}return{dataId:n.write(g,i,r.dtype),shape:i,dtype:r.dtype}}var jS={kernelName:eo,backendName:"cpu",kernelFunc:oK},iK=Zt((e,t)=>Math.pow(e,t)),lK=bn(to,iK),uK={kernelName:to,backendName:"cpu",kernelFunc:lK};function cK(e){let{backend:t,attrs:n}=e,{start:s,stop:r,dtype:a,step:o}=n,i=zy(s,r,o,a);return t.makeTensorInfo([i.length],a,i)}var dK={kernelName:xu,backendName:"cpu",kernelFunc:cK},pK=ht(bu,e=>1/e),hK={kernelName:bu,backendName:"cpu",kernelFunc:pK};function fK(e){let{inputs:t,backend:n,attrs:s}=e,{images:r}=t,{alignCorners:a,halfPixelCenters:o,size:i}=s;Ne(r,"resizeBilinear");let l=v.computeStrides(r.shape),[c,u]=i,[d,p,h,f]=r.shape,m=n.data.get(r.dataId).values,g=new Float32Array(v.sizeFromShape([d,c,u,f])),A=[a&&c>1?p-1:p,a&&u>1?h-1:h],x=[a&&c>1?c-1:c,a&&u>1?u-1:u],y=0,b=A[0]/x[0],w=A[1]/x[1];for(let k=0;k<d;k++)for(let C=0;C<c;C++){let N;o?N=b*(C+.5)-.5:N=b*C;let R=Math.max(0,Math.floor(N)),F=N-R,_=Math.min(p-1,Math.ceil(N)),P=k*l[0]+R*l[1],T=k*l[0]+_*l[1];for(let M=0;M<u;M++){let U;o?U=w*(M+.5)-.5:U=w*M;let j=Math.max(0,Math.floor(U)),z=U-j,X=Math.min(h-1,Math.ceil(U)),Z=P+j*l[2],J=T+j*l[2],ee=P+X*l[2],ne=T+X*l[2];for(let Q=0;Q<f;Q++){let te=m[Z+Q],oe=m[J+Q],fe=m[ee+Q],be=m[ne+Q],we=te+(fe-te)*z,Ce=oe+(be-oe)*z,Me=we+(Ce-we)*F;g[y++]=Me}}}return n.makeTensorInfo([d,c,u,f],"float32",g)}var mK={kernelName:ro,backendName:"cpu",kernelFunc:fK};function gK(e){let{inputs:t,backend:n,attrs:s}=e,{images:r,dy:a}=t,{alignCorners:o}=s;Ne([a,r],"resizeBilinearGrad");let i=v.computeStrides(r.shape),[l,c,u,d]=r.shape,[,p,h]=a.shape,f=new Float32Array(l*c*u*d),m=[o&&p>1?c-1:c,o&&h>1?u-1:u],g=[o&&p>1?p-1:p,o&&h>1?h-1:h],A=m[0]/g[0],x=m[1]/g[1],y=n.data.get(a.dataId).values,b=0;for(let w=0;w<l;w++){let k=w*i[0];for(let C=0;C<p;C++){let N=C*A,R=Math.floor(N),F=Math.min(Math.ceil(N),c-1),_=k+R*i[1],P=k+F*i[1],T=N-R,M=1-T;for(let U=0;U<h;U++){let j=U*x,z=Math.floor(j),X=Math.min(Math.ceil(j),u-1),Z=j-z,J=1-Z,ee=_+z*i[2],ne=_+X*i[2],Q=P+z*i[2],te=P+X*i[2],oe=M*J,fe=M*Z,be=T*J,we=T*Z;for(let Ce=0;Ce<d;Ce++){let Me=y[b++];f[ee+Ce]+=Me*oe,f[ne+Ce]+=Me*fe,f[Q+Ce]+=Me*be,f[te+Ce]+=Me*we}}}}return n.makeTensorInfo([l,u,c,d],"float32",f)}var AK={kernelName:zh,backendName:"cpu",kernelFunc:gK};function yK(e){let{inputs:t,backend:n,attrs:s}=e,{images:r}=t,{alignCorners:a,halfPixelCenters:o,size:i}=s;Ne(r,"resizeNearestNeighbor");let l=v.computeStrides(r.shape),[c,u]=i,[d,p,h,f]=r.shape,m=n.data.get(r.dataId).values,g=new Float32Array(d*c*u*f),A=[a&&c>1?p-1:p,a&&u>1?h-1:h],x=[a&&c>1?c-1:c,a&&u>1?u-1:u],y=A[0]/x[0],b=A[1]/x[1],w=0;for(let k=0;k<d;k++){let C=k*l[0];for(let N=0;N<c;N++){let R=o?y*(N+.5):y*N,F=Math.min(p-1,a?Math.round(R):Math.floor(R));o&&(F=Math.max(0,F));let _=C+F*l[1];for(let P=0;P<u;P++){let T=o?b*(P+.5):b*P,M=Math.min(h-1,a?Math.round(T):Math.floor(T));o&&(M=Math.max(0,M));let U=_+M*l[2];for(let j=0;j<f;j++){let z=m[U+j];g[w++]=z}}}}return n.makeTensorInfo([d,c,u,f],r.dtype,g)}var xK={kernelName:vu,backendName:"cpu",kernelFunc:yK};function bK(e){let{inputs:t,backend:n,attrs:s}=e,{images:r,dy:a}=t,{alignCorners:o}=s;Ne([a,r],"resizeNearestNeighborGrad");let i=v.computeStrides(r.shape),l=v.computeStrides(a.shape),[c,u,d,p]=r.shape,[,h,f]=a.shape,m=new Float32Array(c*u*d*p),g=n.data.get(a.dataId).values,A=[o&&h>1?u-1:u,o&&f>1?d-1:d],x=[o&&h>1?h-1:h,o&&f>1?f-1:f],y=A[0]/x[0],b=A[1]/x[1],w=1/y,k=1/b,C=Math.ceil(w)*2+2,N=Math.ceil(k)*2+2;for(let R=0;R<c;R++){let F=R*i[0];for(let _=0;_<u;_++){let P=F+_*i[1],T=Math.floor(_*w),M=Math.floor(T-C/2);for(let U=0;U<d;U++){let j=P+U*i[2],z=Math.floor(U*k),X=Math.floor(z-N/2);for(let Z=0;Z<p;Z++){let J=0;for(let ee=0;ee<C;ee++){let ne=ee+M;if(ne<0||ne>=h)continue;let Q=F+ne*l[1],te=ne*y,oe=Math.min(u-1,o?Math.round(te):Math.floor(te));if(_===oe)for(let fe=0;fe<N;fe++){let be=fe+X;if(be<0||be>=f)continue;let we=Q+be*l[2],Ce=be*b,Me=Math.min(d-1,o?Math.round(Ce):Math.floor(Ce));U===Me&&(J+=g[we+Z])}}m[j+Z]=J}}}}return n.makeTensorInfo(r.shape,r.dtype,m)}var vK={kernelName:Mh,backendName:"cpu",kernelFunc:bK};function wK(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{dims:a}=s;Ne(r,"reverse");let o=r.shape.length,i=v.parseAxisParam(a,r.shape);if(o===0)return Mr({inputs:{x:r},backend:n});let l=new en(r.shape,r.dtype),c=n.bufferSync(r);for(let u=0;u<l.size;u++){let d=l.indexToLoc(u),p=d.slice();i.forEach(h=>p[h]=r.shape[h]-1-p[h]),l.set(c.get(...p),...d)}return n.makeTensorInfo(l.shape,l.dtype,l.values)}var kK={kernelName:Oi,backendName:"cpu",kernelFunc:wK},SK={kernelName:Zi,backendName:"cpu",kernelFunc:({inputs:e,attrs:t,backend:n})=>{let{image:s}=e,{radians:r,fillValue:a,center:o}=t,i=n,l=v.getTypedArrayFromDType(s.dtype,v.sizeFromShape(s.shape)),[c,u,d,p]=s.shape,[h,f]=E.getImageCenter(o,u,d),m=255,g=Math.sin(r),A=Math.cos(r),x=i.data.get(s.dataId).values;for(let b=0;b<c;b++){let w=b*d*u*p;for(let k=0;k<u;k++){let C=k*(d*p);for(let N=0;N<d;N++){let R=N*p;for(let F=0;F<p;F++){let _=[c,k,N,F],P=_[2],T=_[1],M=(P-h)*A-(T-f)*g,U=(P-h)*g+(T-f)*A;M=Math.round(M+h),U=Math.round(U+f);let j=a;if(typeof a!="number"&&(F===3?j=m:j=a[F]),M>=0&&M<d&&U>=0&&U<u){let X=U*(d*p),Z=M*p,J=w+X+Z+F;j=x[J]}let z=w+C+R+F;l[z]=j}}}}return{dataId:i.write(l,s.shape,s.dtype),shape:s.shape,dtype:s.dtype}}},IK=ht(Mi,e=>{let t=Math.floor(e);return e-t<.5?Math.floor(e):e-t>.5?Math.ceil(e):t%2==0?t:t+1}),CK={kernelName:Mi,backendName:"cpu",kernelFunc:IK};function qS(e,t,n,s,r,a,o,i,l,c){let u=[s/r,r],d=e.values,p=t.values;if(s===0)return ze(n,t.dtype);let h=ze(u,t.dtype);h.values.fill(l);for(let f=0;f<a;f++){let m=[],g=0;for(let A=0;A<o;A++){let x=d[f*o+A];m.push(x),g+=x*i[A]}if(g<0||g>=s/r)throw new Error(`Invalid indices: ${m} does not index into ${n}`);for(let A=0;A<r;A++)c?h.values[g*r+A]+=p[f*r+A]:h.values[g*r+A]=t.rank===0?p[0]:p[f*r+A]}return h}function TK(e){let{inputs:t,backend:n,attrs:s}=e,{indices:r,updates:a}=t,{shape:o}=s,{sliceRank:i,numUpdates:l,sliceSize:c,strides:u,outputSize:d}=E.calculateShapes(a,r,o),p=!0,h=n.bufferSync(r),f=n.bufferSync(a),m=qS(h,f,o,d,c,l,i,u,0,p);return n.makeTensorInfo(o,m.dtype,m.values)}var NK={kernelName:zi,backendName:"cpu",kernelFunc:TK};function EK(e){let{inputs:t,backend:n}=e,{condition:s,t:r,e:a}=t;Ne([s,r,a],"select");let o=s.shape.length,i=n.data.get(s.dataId).values,l=n.data.get(r.dataId).values,c=n.data.get(a.dataId).values,u=zn(r.dtype,a.dtype),d=v.makeZerosTypedArray(v.sizeFromShape(r.shape),u),p=0,h=o===0||o>1||r.shape.length===1?1:v.sizeFromShape(r.shape.slice(1));for(let f=0;f<i.length;f++)for(let m=0;m<h;m++)i[f]===1?d[p++]=l[f]:d[p++]=c[f];return n.makeTensorInfo(r.shape,u,d)}var RK={kernelName:Li,backendName:"cpu",kernelFunc:EK},$K=E.SELU_SCALEALPHA,_K=E.SELU_SCALE,DK=ht(wu,e=>e>=0?_K*e:$K*(Math.exp(e)-1)),PK={kernelName:wu,backendName:"cpu",kernelFunc:DK},FK=ht(ku,e=>e<0?-1:e>0?1:0),OK={kernelName:ku,backendName:"cpu",kernelFunc:FK},MK=ht(io,e=>Math.sin(e)),zK={kernelName:io,backendName:"cpu",kernelFunc:MK},LK=ht(Wi,e=>Math.sinh(e)),BK={kernelName:Wi,backendName:"cpu",kernelFunc:LK},WK=11920928955078125e-23,XS=Math.log(WK)+2,VK=ht(Su,e=>{let t=e>-XS,n=e<XS,s=Math.exp(e),r;return n?r=s:t?r=e:r=Math.log(1+s),r}),UK={kernelName:Su,backendName:"cpu",kernelFunc:VK};function GK(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{blockShape:a,paddings:o}=s;Ne([r],"spaceToBatchND");let i=v.sizeFromShape(a),l=[[0,0]];l.push(...o);for(let k=1+a.length;k<r.shape.length;++k)l.push([0,0]);let c=jS.kernelFunc({inputs:{x:r},backend:n,attrs:{paddings:l,constantValue:0}}),u=E.getReshaped(c.shape,a,i,!1),d=E.getPermuted(u.length,a.length,!1),p=E.getReshapedPermuted(c.shape,a,i,!1),m=Rt({inputs:{x:c},backend:n,attrs:{shape:u}}),x=_s({inputs:{x:m},backend:n,attrs:{perm:d}}),w=Rt({inputs:{x},backend:n,attrs:{shape:p}});return n.disposeIntermediateTensorInfo(c),n.disposeIntermediateTensorInfo(m),n.disposeIntermediateTensorInfo(x),w}var HK={kernelName:Vi,backendName:"cpu",kernelFunc:GK};function jK(e){let{inputs:t,backend:n}=e,{indices:s,values:r,denseShape:a,defaultValue:o}=t;if(a.shape.length!==1)throw new Error(`Dense shape must be a vector, saw:
|
|
${a.shape}`);if(s.shape.length!==2)throw new Error(`Indices must be a matrix, saw:
|
|
${s.shape}`);if(r.shape.length!==1)throw new Error(`Values must be a vector, saw:
|
|
${r.shape}`);if(o.shape.length!==0)throw new Error(`Default value must be a scalar, saw:
|
|
${o.shape}`);let i=n.data.get(s.dataId).values,l=n.data.get(r.dataId).values,c=n.data.get(a.dataId).values,u=n.data.get(o.dataId).values[0],[d,p,h,f,m]=xS(i,s.shape,s.dtype,l,r.dtype,c,u);return[n.makeTensorInfo(p,s.dtype,d),n.makeTensorInfo([p[0]],r.dtype,h),n.makeTensorInfo([f.length],"bool",new Uint8Array(f.map(g=>Number(g)))),n.makeTensorInfo([m.length],s.dtype,new Int32Array(m))]}var qK={kernelName:Lh,backendName:"cpu",kernelFunc:jK};function XK(e){let{inputs:t,backend:n}=e,{inputIndices:s,inputShape:r,newShape:a}=t;if(s.shape.length!==2)throw new Error(`Input indices should be a matrix but received shape
|
|
${s.shape}`);if(r.shape.length!==1)throw new Error(`Input shape should be a vector but received shape
|
|
${r.shape}`);if(a.shape.length!==1)throw new Error(`Target shape should be a vector but received shape ${a.shape}`);let o=Array.from(n.data.get(r.dataId).values),i=n.data.get(s.dataId).values,l=Array.from(n.data.get(a.dataId).values),[c,u,d]=bS(i,s.shape,s.dtype,o,l);return[n.makeTensorInfo(u,s.dtype,c),n.makeTensorInfo([d.length],a.dtype,new Int32Array(d))]}var KK={kernelName:Bh,backendName:"cpu",kernelFunc:XK};function ZK(e){let{inputs:t,backend:n}=e,{data:s,indices:r,segmentIds:a}=t;if(s.shape.length<1)throw new Error("Data should be at least 1 dimensional but received scalar");if(r.shape.length!==1)throw new Error(`Indices should be a vector but received shape
|
|
${r.shape}`);if(a.shape.length!==1)throw new Error(`Segment ids should be a vector but received shape
|
|
${a.shape}`);let o=n.data.get(s.dataId).values,i=n.data.get(r.dataId).values,l=n.data.get(a.dataId).values,[c,u]=Ly(o,s.shape,s.dtype,i,l,!0);return n.makeTensorInfo(u,s.dtype,c)}var YK={kernelName:Wh,backendName:"cpu",kernelFunc:ZK};function JK(e){let{inputs:t,backend:n}=e,{data:s,indices:r,segmentIds:a}=t;if(s.shape.length<1)throw new Error("Data should be at least 1 dimensional but received scalar");if(r.shape.length!==1)throw new Error(`Indices should be a vector but received shape
|
|
${r.shape}`);if(a.shape.length!==1)throw new Error(`Segment ids should be a vector but received shape
|
|
${a.shape}`);let o=n.data.get(s.dataId).values,i=n.data.get(r.dataId).values,l=n.data.get(a.dataId).values,[c,u]=Ly(o,s.shape,s.dtype,i,l);return n.makeTensorInfo(u,s.dtype,c)}var QK={kernelName:Vh,backendName:"cpu",kernelFunc:JK};function eZ(e){let{inputs:t,backend:n,attrs:s}=e,{sparseIndices:r,sparseValues:a,defaultValue:o}=t,{outputShape:i}=s,{sliceRank:l,numUpdates:c,sliceSize:u,strides:d,outputSize:p}=E.calculateShapes(a,r,i),h=!1,f=n.bufferSync(r),m=n.bufferSync(a),g=n.data.get(o.dataId).values[0],A=qS(f,m,i,p,u,c,l,d,g,h);return n.makeTensorInfo(i,A.dtype,A.values)}var tZ={kernelName:ld,backendName:"cpu",kernelFunc:eZ};function nZ(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{numOrSizeSplits:a,axis:o}=s,i=v.parseAxisParam(o,r.shape)[0],l=E.prepareSplitSize(r,a,i),c=new Array(r.shape.length).fill(0),u=r.shape.slice();return l.map(d=>{let p=[...u];p[i]=d;let h=vl({inputs:{x:r},backend:n,attrs:{begin:c,size:p}});return c[i]+=d,h})}var sZ={kernelName:Ui,backendName:"cpu",kernelFunc:nZ},rZ={kernelName:Iu,backendName:"cpu",kernelFunc:({inputs:e,backend:t})=>{let{x:n}=e,s=t;Ne(n,"square");let r=s.data.get(n.dataId).values,a=new Float32Array(r.length);for(let i=0;i<r.length;++i){let l=r[i];a[i]=l*l}return{dataId:s.write(a,n.shape,n.dtype),shape:n.shape,dtype:n.dtype}}},aZ=ht(Ao,(e,t)=>{let n=t;return isNaN(e)?NaN:e>0?1:n.alpha}),oZ={kernelName:Ao,backendName:"cpu",kernelFunc:aZ};function iZ(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{begin:a,end:o,strides:i,beginMask:l,endMask:c,ellipsisMask:u,newAxisMask:d,shrinkAxisMask:p}=s;Ne(r,"stridedSlice");let{finalShapeSparse:h,finalShape:f,isIdentity:m,sliceDim0:g,isSimpleSlice:A,begin:x,end:y,strides:b}=Pt.sliceInfo(r.shape,a,o,i,l,c,u,d,p),w;if(m)w=Rt({inputs:{x:r},backend:n,attrs:{shape:f}});else if(g||A){v.assert(r.shape.length>=1,()=>`Input must have rank at least 1, got: ${r.shape.length}`);let k=Pt.computeOutShape(x,y,b),C=vl({inputs:{x:r},backend:n,attrs:{begin:x,size:k}});w=Rt({inputs:{x:C},backend:n,attrs:{shape:f}}),n.disposeIntermediateTensorInfo(C)}else{let k=n.bufferSync(r),C=wS(h,k,b,x);w=n.makeTensorInfo(f,C.dtype,C.values)}return w}var lZ={kernelName:Gi,backendName:"cpu",kernelFunc:iZ};function uZ(e){let{inputs:t,backend:n,attrs:s}=e,{separator:r,nGramWidths:a,leftPad:o,rightPad:i,padWidth:l,preserveShortSequences:c}=s,{data:u,dataSplits:d}=t,p=n.data.get(u.dataId).values,h=n.data.get(d.dataId).values,[f,m]=kS(p,h,r,a,o,i,l,c);return[n.makeTensorInfo([f.length],"string",f),n.makeTensorInfo(d.shape,"int32",m)]}var cZ={kernelName:ud,backendName:"cpu",kernelFunc:uZ};function dZ(e){let{inputs:t,backend:n,attrs:s}=e,{skipEmpty:r}=s,{input:a,delimiter:o}=t;if(a.dtype!=="string")throw new Error("Input must be of datatype string");if(a.shape.length!==1)throw new Error(`Input must be a vector, got shape: ${a.shape}`);if(o.shape.length!==0)throw new Error(`Delimiter must be a scalar, got shape: ${o.shape}`);let i=n.data.get(a.dataId).values,l=n.data.get(o.dataId).values[0],[c,u,d]=SS(i,l,r),p=u.length;return[n.makeTensorInfo([p,2],"int32",c),n.makeTensorInfo([p],"string",u),n.makeTensorInfo([2],"int32",new Int32Array(d))]}var pZ={kernelName:Uh,backendName:"cpu",kernelFunc:dZ};function hZ(e){let{inputs:t,backend:n,attrs:s}=e,{numBuckets:r}=s,{input:a}=t;if(a.dtype!=="string")throw new Error("Input must be of datatype string");if(r<=0)throw new Error("Number of buckets must be at least 1");let o=n.data.get(a.dataId).values,i=IS(o,r);return n.makeTensorInfo(a.shape,"int32",i)}var fZ={kernelName:Gh,backendName:"cpu",kernelFunc:hZ},mZ=ht(Hi,e=>Math.tan(e)),gZ={kernelName:Hi,backendName:"cpu",kernelFunc:mZ},AZ=ht(mo,e=>Math.tanh(e)),yZ={kernelName:mo,backendName:"cpu",kernelFunc:AZ};function xZ(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{reps:a}=s;Ne(r,"tile");let o=TS(n.bufferSync(r),a);return n.makeTensorInfo(o.shape,o.dtype,o.values)}var bZ={kernelName:jr,backendName:"cpu",kernelFunc:xZ};function vZ(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{k:a,sorted:o}=s;Ne(r,"topk");let i=n.data.get(r.dataId).values,[l,c]=ES(i,r.shape,r.dtype,a,o);return[n.makeTensorInfo(l.shape,l.dtype,l.values),n.makeTensorInfo(c.shape,c.dtype,c.values)]}var wZ={kernelName:ji,backendName:"cpu",kernelFunc:vZ};function kZ(e){let{inputs:t,attrs:n,backend:s}=e,{image:r,transforms:a}=t,{interpolation:o,fillMode:i,fillValue:l,outputShape:c}=n,[u,d,p,h]=r.shape,[f,m]=c!=null?c:[d,p],g=[u,f,m,h],A=v.computeStrides(r.shape),x=A[0],y=A[1],b=A[2],w=v.getTypedArrayFromDType(r.dtype,v.sizeFromShape(g));w.fill(l);let k=s.data.get(r.dataId).values,C=s.data.get(a.dataId).values;for(let R=0;R<u;++R){let F=a.shape[0]===1?C:C.subarray(R*8,R*8+8);for(let _=0;_<f;++_)for(let P=0;P<m;++P)for(let T=0;T<h;++T){let M,U=F[6]*P+F[7]*_+1;if(U===0)continue;let j=(F[0]*P+F[1]*_+F[2])/U,z=(F[3]*P+F[4]*_+F[5])/U,X=KS(j,p,i),Z=KS(z,d,i);switch(o){case"nearest":M=EZ(k,d,p,x,y,b,R,Z,X,T,l);break;case"bilinear":M=RZ(k,d,p,x,y,b,R,Z,X,T,l);break;default:throw new Error(`Error in Transform: Expect 'nearest' or 'bilinear', but got ${o}`)}let J=R*x+_*y+P*b+T;w[J]=M}return s.makeTensorInfo(g,r.dtype,w)}return{dataId:s.write(w,g,r.dtype),shape:r.shape,dtype:r.dtype}}var SZ={kernelName:qi,backendName:"cpu",kernelFunc:kZ};function KS(e,t,n){switch(n){case"reflect":return IZ(e,t);case"wrap":return CZ(e,t);case"nearest":return NZ(e,t);case"constant":default:return TZ(e,t)}}function IZ(e,t){let n=e;if(n<0)if(t<=1)n=0;else{let s=2*t;n<s&&(n=s*Math.trunc(-n/s)+n),n=n<-t?n+s:-n-1}else if(n>t-1)if(t<=1)n=0;else{let s=2*t;n-=s*Math.trunc(n/s),n>=t&&(n=s-n-1)}return v.clamp(0,n,t-1)}function CZ(e,t){let n=e;if(n<0)if(t<=1)n=0;else{let s=t-1;n+=t*(Math.trunc(-n/s)+1)}else if(n>t-1)if(t<=1)n=0;else{let s=t-1;n-=t*Math.trunc(n/s)}return v.clamp(0,n,t-1)}function TZ(e,t){return e}function NZ(e,t){return v.clamp(0,e,t-1)}function ap(e,t,n,s,r,a,o,i,l,c,u){let d=o*s+i*r+l*a+c;return 0<=i&&i<t&&0<=l&&l<n?e[d]:u}function EZ(e,t,n,s,r,a,o,i,l,c,u){let d=Math.round(i),p=Math.round(l);return ap(e,t,n,s,r,a,o,d,p,c,u)}function RZ(e,t,n,s,r,a,o,i,l,c,u){let d=Math.floor(i),p=Math.floor(l),h=d+1,f=p+1,m=(f-l)*ap(e,t,n,s,r,a,o,d,p,c,u)+(l-p)*ap(e,t,n,s,r,a,o,d,f,c,u),g=(f-l)*ap(e,t,n,s,r,a,o,h,p,c,u)+(l-p)*ap(e,t,n,s,r,a,o,h,f,c,u);return(h-i)*m+(i-d)*g}function $Z(e){let{inputs:t,attrs:n,backend:s}=e,{axis:r}=n,{x:a}=t;Ne(a,"unique");let o=s.data.get(a.dataId).values,{outputValues:i,outputShape:l,indices:c}=RS(o,r,a.shape,a.dtype);return[s.makeTensorInfo(l,a.dtype,i),s.makeTensorInfo([c.length],"int32",c)]}var _Z={kernelName:Hh,backendName:"cpu",kernelFunc:$Z};function DZ(e){let{inputs:t,backend:n,attrs:s}=e,{value:r}=t,{axis:a}=s;a<0&&(a+=r.shape.length);let o=r.shape.length,i=r.shape[a],l=new Array(o-1),c=0;for(let h=0;h<o;h++)h!==a&&(l[c++]=r.shape[h]);let u=new Array(o).fill(0),d=r.shape.slice();d[a]=1;let p=new Array(i);for(let h=0;h<p.length;h++){u[a]=h;let f=vl({inputs:{x:r},backend:n,attrs:{begin:u,size:d}});p[h]=Rt({inputs:{x:f},backend:n,attrs:{shape:l}}),n.disposeIntermediateTensorInfo(f)}return p}var PZ={kernelName:Xi,backendName:"cpu",kernelFunc:DZ};function FZ(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,segmentIds:a}=t,{numSegments:o}=s;Ne(r,"unsortedSegmentSum");let i=r.shape.length,l=a.shape.length,c=[],u=[],d=i-l,p=a;for(let f=0;f<d;++f){let m=Sm({inputs:{input:p},backend:n,attrs:{dim:f+1}});p=m,u.push(m)}for(let f=0;f<o;++f){let m=v.createScalarValue(f,"int32"),g=n.makeTensorInfo([],"int32",m),A=J7({inputs:{a:g,b:p},backend:n}),x=Bo({inputs:{x:A},backend:n,attrs:{dtype:"float32"}}),y=wm({inputs:{a:x,b:r},backend:n}),b=rp({inputs:{x:y},backend:n,attrs:{axis:0,keepDims:!1}});c.push(b),u.push(g),u.push(A),u.push(x),u.push(y),u.push(b)}let h=HS({inputs:c,backend:n,attrs:{axis:0}});return u.forEach(f=>n.disposeIntermediateTensorInfo(f)),h}var OZ={kernelName:cd,backendName:"cpu",kernelFunc:FZ},MZ=[GH,OG,jH,XH,VG,ZH,JH,ej,nj,rj,oj,lj,cj,hj,mj,yj,bj,wj,Sj,VH,Cj,Nj,Rj,_j,BG,GG,Pj,MG,Oj,zj,Wj,Uj,Lj,qj,Kj,Hj,Yj,Qj,tq,sq,aq,iq,lq,cq,pq,fq,mq,Aq,gq,Gy,bq,FH,wq,HG,Rq,jG,$q,XG,Mq,zq,Bq,ZG,Uq,Hq,qq,Kq,Yq,JG,eH,zG,Qq,Mj,tX,sX,aX,OH,nH,rH,iX,oH,uX,pX,fX,AX,xX,vX,lH,SX,CX,NX,RX,_X,wX,PX,OX,cH,zX,WX,HX,pH,fH,XX,YX,eK,gH,nK,rK,aK,jS,uK,zH,xH,dK,LG,hK,LH,BH,WH,mK,AK,xK,vK,kK,SK,CK,vH,NK,RK,PK,kH,OK,zK,BK,SH,UX,UK,HK,qK,KK,YK,QK,tZ,sZ,TH,rZ,EH,oZ,lZ,cZ,pZ,fZ,DH,yq,gZ,yZ,bZ,wZ,AH,SZ,_Z,PZ,OZ,sK];for(let e of MZ)rr(e);var ZS={};Oe(ZS,{assertNotComplex:()=>nc,bindCanvasToFramebuffer:()=>KZ,bindColorTextureToFramebuffer:()=>Em,bindTextureToProgramUniformSampler:()=>dI,bindTextureUnit:()=>lI,bindVertexBufferToProgramAttribute:()=>Ky,callAndCheck:()=>Ie,canBeRepresented:()=>YS,createFragmentShader:()=>eI,createFramebuffer:()=>iI,createProgram:()=>tI,createStaticIndexBuffer:()=>rI,createStaticVertexBuffer:()=>sI,createTexture:()=>aI,createVertexShader:()=>QS,getBatchDim:()=>kl,getExtensionOrThrow:()=>lp,getFramebufferErrorMessage:()=>pI,getMaxTexturesInShader:()=>gI,getNumChannels:()=>qZ,getProgramUniformLocation:()=>cI,getProgramUniformLocationOrThrow:()=>uI,getRowsCols:()=>Sl,getShapeAs3D:()=>Rm,getTextureShapeFromLogicalShape:()=>fI,getWebGLDisjointQueryTimerVersion:()=>AI,getWebGLErrorMessage:()=>JS,getWebGLMaxTextureSize:()=>mI,hasExtension:()=>Ps,isCapableOfRenderingToFloatTexture:()=>yI,isDownloadFloatTextureEnabled:()=>xI,isReshapeFree:()=>cp,isWebGLFenceEnabled:()=>bI,isWebGLVersionEnabled:()=>Yy,linkProgram:()=>nI,resetMaxTextureSize:()=>ZZ,resetMaxTexturesInShader:()=>YZ,unbindColorTextureFromFramebuffer:()=>Zy,unbindTextureUnit:()=>XZ,validateFramebuffer:()=>up,validateProgram:()=>Nm,validateTextureSize:()=>oI});var wl={},qy={alpha:!1,antialias:!1,premultipliedAlpha:!1,preserveDrawingBuffer:!1,depth:!1,stencil:!1,failIfMajorPerformanceCaveat:!0};function Cm(e,t){wl[e]=t}function zr(e){if(!(e in wl)){let n=LZ(e);if(n!==null)wl[e]=n;else return console.log("Could not get context for WebGL version",e),null}let t=wl[e];return t==null||t.isContextLost()?(delete wl[e],zr(e)):(t.disable(t.DEPTH_TEST),t.disable(t.STENCIL_TEST),t.disable(t.BLEND),t.disable(t.DITHER),t.disable(t.POLYGON_OFFSET_FILL),t.disable(t.SAMPLE_COVERAGE),t.enable(t.SCISSOR_TEST),t.enable(t.CULL_FACE),t.cullFace(t.BACK),wl[e])}function zZ(e){if(typeof OffscreenCanvas!="undefined"&&e===2)return new OffscreenCanvas(300,150);if(typeof document!="undefined")return document.createElement("canvas");throw new Error("Cannot create a canvas in this context")}function LZ(e){if(e!==1&&e!==2)throw new Error("Cannot get WebGL rendering context, WebGL is disabled.");let t=zZ(e);return t.addEventListener("webglcontextlost",n=>{n.preventDefault(),delete wl[e]},!1),e===1?t.getContext("webgl",qy)||t.getContext("experimental-webgl",qy):t.getContext("webgl2",qy)}var op;(function(e){e[e.DENSE=0]="DENSE",e[e.SHARED_BATCH=1]="SHARED_BATCH"})(op||(op={}));var Ds;(function(e){e[e.RENDER=0]="RENDER",e[e.UPLOAD=1]="UPLOAD",e[e.PIXELS=2]="PIXELS",e[e.DOWNLOAD=3]="DOWNLOAD"})(Ds||(Ds={}));var In;(function(e){e[e.UNPACKED_FLOAT16=0]="UNPACKED_FLOAT16",e[e.UNPACKED_FLOAT32=1]="UNPACKED_FLOAT32",e[e.PACKED_4X1_UNSIGNED_BYTE=2]="PACKED_4X1_UNSIGNED_BYTE",e[e.PACKED_2X2_FLOAT32=3]="PACKED_2X2_FLOAT32",e[e.PACKED_2X2_FLOAT16=4]="PACKED_2X2_FLOAT16"})(In||(In={}));function ip(e,t){return[t,e]}function BZ(e,t){return e*t}function Tm(e){let t=v.sizeFromShape(e),n=Math.ceil(t/4);return v.sizeToSquarishShape(n)}function tc(e,t){return[Math.max(1,Math.ceil(t/2)),Math.max(1,Math.ceil(e/2))]}function WZ(e,t){let[n,s]=tc(e,t);return n*s*4}function Xy(e,t){let n=e,s,r,a,o,i,l,c,u,d,p;return K().getNumber("WEBGL_VERSION")===2?(s=n.R32F,r=n.R16F,a=n.RGBA16F,o=n.RGBA32F,i=n.RED,c=4,u=1,d=n.HALF_FLOAT,p=n.FLOAT):(s=e.RGBA,r=e.RGBA,a=e.RGBA,o=n.RGBA,i=e.RGBA,c=4,u=4,d=t!=null?t.HALF_FLOAT_OES:null,p=e.FLOAT),l=e.RGBA,{internalFormatFloat:s,internalFormatHalfFloat:r,internalFormatPackedHalfFloat:a,internalFormatPackedFloat:o,textureFormatFloat:i,downloadTextureFormat:l,downloadUnpackNumChannels:c,defaultNumChannels:u,textureTypeHalfFloat:d,textureTypeFloat:p}}function Ie(e,t){let n=t();return K().getBool("DEBUG")&&VZ(e),n}function VZ(e){let t=e.getError();if(t!==e.NO_ERROR)throw new Error("WebGL Error: "+JS(e,t))}var UZ=596e-10,GZ=65504;function YS(e){return!!(K().getBool("WEBGL_RENDER_FLOAT32_ENABLED")||e===0||UZ<Math.abs(e)&&Math.abs(e)<GZ)}function JS(e,t){switch(t){case e.NO_ERROR:return"NO_ERROR";case e.INVALID_ENUM:return"INVALID_ENUM";case e.INVALID_VALUE:return"INVALID_VALUE";case e.INVALID_OPERATION:return"INVALID_OPERATION";case e.INVALID_FRAMEBUFFER_OPERATION:return"INVALID_FRAMEBUFFER_OPERATION";case e.OUT_OF_MEMORY:return"OUT_OF_MEMORY";case e.CONTEXT_LOST_WEBGL:return"CONTEXT_LOST_WEBGL";default:return`Unknown error code ${t}`}}function lp(e,t){return sa(e,()=>e.getExtension(t),'Extension "'+t+'" not supported on this browser.')}function QS(e,t){let n=sa(e,()=>e.createShader(e.VERTEX_SHADER),"Unable to create vertex WebGLShader.");if(Ie(e,()=>e.shaderSource(n,t)),Ie(e,()=>e.compileShader(n)),e.getShaderParameter(n,e.COMPILE_STATUS)===!1)throw console.log(e.getShaderInfoLog(n)),new Error("Failed to compile vertex shader.");return n}function eI(e,t){let n=sa(e,()=>e.createShader(e.FRAGMENT_SHADER),"Unable to create fragment WebGLShader.");if(Ie(e,()=>e.shaderSource(n,t)),Ie(e,()=>e.compileShader(n)),e.getShaderParameter(n,e.COMPILE_STATUS)===!1)throw jZ(t,e.getShaderInfoLog(n)),new Error("Failed to compile fragment shader.");return n}var HZ=/ERROR: [0-9]+:([0-9]+):/g;function jZ(e,t){let n=HZ.exec(t);if(n==null){console.log(`Couldn't parse line number in error: ${t}`),console.log(e);return}let s=+n[1],r=e.split(`
|
|
`),a=r.length.toString().length+2,o=r.map((d,p)=>v.rightPad((p+1).toString(),a)+d),i=0;for(let d=0;d<o.length;d++)i=Math.max(o[d].length,i);let l=o.slice(0,s-1),c=o.slice(s-1,s),u=o.slice(s);console.log(l.join(`
|
|
`)),console.log(t.split(`
|
|
`)[0]),console.log(`%c ${v.rightPad(c[0],i)}`,"border:1px solid red; background-color:#e3d2d2; color:#a61717"),console.log(u.join(`
|
|
`))}function tI(e){return sa(e,()=>e.createProgram(),"Unable to create WebGLProgram.")}function nI(e,t){if(Ie(e,()=>e.linkProgram(t)),e.getProgramParameter(t,e.LINK_STATUS)===!1)throw console.log(e.getProgramInfoLog(t)),new Error("Failed to link vertex and fragment shaders.")}function Nm(e,t){if(Ie(e,()=>e.validateProgram(t)),e.getProgramParameter(t,e.VALIDATE_STATUS)===!1)throw console.log(e.getProgramInfoLog(t)),new Error("Shader program validation failed.")}function sI(e,t){let n=sa(e,()=>e.createBuffer(),"Unable to create WebGLBuffer");return Ie(e,()=>e.bindBuffer(e.ARRAY_BUFFER,n)),Ie(e,()=>e.bufferData(e.ARRAY_BUFFER,t,e.STATIC_DRAW)),n}function rI(e,t){let n=sa(e,()=>e.createBuffer(),"Unable to create WebGLBuffer");return Ie(e,()=>e.bindBuffer(e.ELEMENT_ARRAY_BUFFER,n)),Ie(e,()=>e.bufferData(e.ELEMENT_ARRAY_BUFFER,t,e.STATIC_DRAW)),n}function qZ(){return K().getNumber("WEBGL_VERSION")===2?1:4}function aI(e){return sa(e,()=>e.createTexture(),"Unable to create WebGLTexture.")}function oI(e,t){let n=K().getNumber("WEBGL_MAX_TEXTURE_SIZE");if(e<=0||t<=0){let s=`[${e}x${t}]`;throw new Error("Requested texture size "+s+" is invalid.")}if(e>n||t>n){let s=`[${e}x${t}]`,r=`[${n}x${n}]`;throw new Error("Requested texture size "+s+" greater than WebGL maximum on this browser / GPU "+r+".")}}function iI(e){return sa(e,()=>e.createFramebuffer(),"Unable to create WebGLFramebuffer.")}function Ky(e,t,n,s,r,a,o){let i=e.getAttribLocation(t,n);return i===-1?!1:(Ie(e,()=>e.bindBuffer(e.ARRAY_BUFFER,s)),Ie(e,()=>e.vertexAttribPointer(i,r,e.FLOAT,!1,a,o)),Ie(e,()=>e.enableVertexAttribArray(i)),!0)}function lI(e,t,n){hI(e,n),Ie(e,()=>e.activeTexture(e.TEXTURE0+n)),Ie(e,()=>e.bindTexture(e.TEXTURE_2D,t))}function XZ(e,t){hI(e,t),Ie(e,()=>e.activeTexture(e.TEXTURE0+t)),Ie(e,()=>e.bindTexture(e.TEXTURE_2D,null))}function uI(e,t,n){return sa(e,()=>e.getUniformLocation(t,n),'uniform "'+n+'" not present in program.')}function cI(e,t,n){return e.getUniformLocation(t,n)}function dI(e,t,n,s){Ie(e,()=>lI(e,t,s)),Ie(e,()=>e.uniform1i(n,s))}function KZ(e){Ie(e,()=>e.bindFramebuffer(e.FRAMEBUFFER,null)),Ie(e,()=>e.viewport(0,0,e.canvas.width,e.canvas.height)),Ie(e,()=>e.scissor(0,0,e.canvas.width,e.canvas.height))}function Em(e,t,n){Ie(e,()=>e.bindFramebuffer(e.FRAMEBUFFER,n)),Ie(e,()=>e.framebufferTexture2D(e.FRAMEBUFFER,e.COLOR_ATTACHMENT0,e.TEXTURE_2D,t,0))}function Zy(e,t){Ie(e,()=>e.bindFramebuffer(e.FRAMEBUFFER,t)),Ie(e,()=>e.framebufferTexture2D(e.FRAMEBUFFER,e.COLOR_ATTACHMENT0,e.TEXTURE_2D,null,0))}function up(e){let t=e.checkFramebufferStatus(e.FRAMEBUFFER);if(t!==e.FRAMEBUFFER_COMPLETE)throw new Error("Error binding framebuffer: "+pI(e,t))}function pI(e,t){switch(t){case e.FRAMEBUFFER_INCOMPLETE_ATTACHMENT:return"FRAMEBUFFER_INCOMPLETE_ATTACHMENT";case e.FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT:return"FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT";case e.FRAMEBUFFER_INCOMPLETE_DIMENSIONS:return"FRAMEBUFFER_INCOMPLETE_DIMENSIONS";case e.FRAMEBUFFER_UNSUPPORTED:return"FRAMEBUFFER_UNSUPPORTED";default:return`unknown error ${t}`}}function sa(e,t,n){let s=Ie(e,()=>t());if(s==null)throw new Error(n);return s}function hI(e,t){let n=e.MAX_COMBINED_TEXTURE_IMAGE_UNITS-1,s=t+e.TEXTURE0;if(s<e.TEXTURE0||s>n){let r=`[gl.TEXTURE0, gl.TEXTURE${n}]`;throw new Error(`textureUnit must be in ${r}.`)}}function kl(e,t=2){return v.sizeFromShape(e.slice(0,e.length-t))}function Sl(e){if(e.length===0)throw Error("Cannot get rows and columns of an empty shape array.");return[e.length>1?e[e.length-2]:1,e[e.length-1]]}function Rm(e){let t=[1,1,1];return e.length===0||e.length===1&&e[0]===1||(t=[kl(e),...Sl(e)]),t}function fI(e,t=!1){let n=K().getNumber("WEBGL_MAX_TEXTURE_SIZE");t&&(n=n*2,e=e.map((r,a)=>a>=e.length-2?v.nearestLargerEven(e[a]):e[a]),e.length===1&&(e=[2,e[0]])),e.length!==2&&(e=v.squeezeShape(e).newShape);let s=v.sizeFromShape(e);if(e.length<=1&&s<=n)return[1,s];if(e.length===2&&e[0]<=n&&e[1]<=n)return e;if(e.length===3&&e[0]*e[1]<=n&&e[2]<=n)return[e[0]*e[1],e[2]];if(e.length===3&&e[0]<=n&&e[1]*e[2]<=n)return[e[0],e[1]*e[2]];if(e.length===4&&e[0]*e[1]*e[2]<=n&&e[3]<=n)return[e[0]*e[1]*e[2],e[3]];if(e.length===4&&e[0]<=n&&e[1]*e[2]*e[3]<=n)return[e[0],e[1]*e[2]*e[3]];if(t){let r=kl(e),a=2,o=2;return e.length&&([a,o]=Sl(e)),s=r*(a/2)*(o/2),v.sizeToSquarishShape(s).map(i=>i*2)}return v.sizeToSquarishShape(s)}function $m(e){return e%2==0}function cp(e,t){if(e=e.slice(-2),t=t.slice(-2),v.arraysEqual(e,t)||!e.length||!t.length||e[0]===0||e[1]===0||t[0]===0||t[1]===0)return!0;if(e.length!==t.length){let n=e.slice(-1)[0],s=t.slice(-1)[0];if(n===s||$m(n)&&$m(s)&&(e[0]===1||t[0]===1))return!0}return e[1]===t[1]&&$m(e[0])&&$m(t[0])}var _m,Dm;function mI(e){if(_m==null){let t=zr(e);_m=t.getParameter(t.MAX_TEXTURE_SIZE)}return _m}function ZZ(){_m=null}function YZ(){Dm=null}function gI(e){if(Dm==null){let t=zr(e);Dm=t.getParameter(t.MAX_TEXTURE_IMAGE_UNITS)}return Math.min(16,Dm)}function AI(e){if(e===0)return 0;let t,n=zr(e);return Ps(n,"EXT_disjoint_timer_query_webgl2")&&e===2?t=2:Ps(n,"EXT_disjoint_timer_query")?t=1:t=0,t}function Ps(e,t){return e.getExtension(t)!=null}function Yy(e){try{if(zr(e)!=null)return!0}catch(t){return console.log("Error when getting WebGL context: ",t),!1}return!1}function yI(e){if(e===0)return!1;let t=zr(e);if(e===1){if(!Ps(t,"OES_texture_float"))return!1}else if(!Ps(t,"EXT_color_buffer_float"))return!1;return Jy(t)}function xI(e){if(e===0)return!1;let t=zr(e);if(e===1){if(!Ps(t,"OES_texture_float")||!Ps(t,"WEBGL_color_buffer_float"))return!1}else{if(Ps(t,"EXT_color_buffer_float"))return Jy(t);let s="EXT_color_buffer_half_float";if(Ps(t,s)){let r=t.getExtension(s);return JZ(t,r)}return!1}return Jy(t)}function Jy(e){let t=Xy(e),n=e.createTexture();e.bindTexture(e.TEXTURE_2D,n);let s=1,r=1;e.texImage2D(e.TEXTURE_2D,0,t.internalFormatFloat,s,r,0,t.textureFormatFloat,t.textureTypeFloat,null);let a=e.createFramebuffer();e.bindFramebuffer(e.FRAMEBUFFER,a),e.framebufferTexture2D(e.FRAMEBUFFER,e.COLOR_ATTACHMENT0,e.TEXTURE_2D,n,0);let o=e.checkFramebufferStatus(e.FRAMEBUFFER)===e.FRAMEBUFFER_COMPLETE;return e.bindTexture(e.TEXTURE_2D,null),e.bindFramebuffer(e.FRAMEBUFFER,null),e.deleteTexture(n),e.deleteFramebuffer(a),o}function JZ(e,t){let n=Xy(e,t),s=e.createTexture();e.bindTexture(e.TEXTURE_2D,s);let r=1,a=1;e.texImage2D(e.TEXTURE_2D,0,n.internalFormatHalfFloat,r,a,0,n.textureFormatFloat,n.textureTypeHalfFloat,null);let o=e.createFramebuffer();e.bindFramebuffer(e.FRAMEBUFFER,o),e.framebufferTexture2D(e.FRAMEBUFFER,e.COLOR_ATTACHMENT0,e.TEXTURE_2D,s,0);let i=e.checkFramebufferStatus(e.FRAMEBUFFER)===e.FRAMEBUFFER_COMPLETE;return e.bindTexture(e.TEXTURE_2D,null),e.bindFramebuffer(e.FRAMEBUFFER,null),e.deleteTexture(s),e.deleteFramebuffer(o),i}function bI(e){return e!==2?!1:zr(e).fenceSync!=null}function nc(e,t){Array.isArray(e)||(e=[e]),e.forEach(n=>{n!=null&&v.assert(n.dtype!=="complex64",()=>`${t} does not support complex64 tensors in the WebGL backend.`)})}var _e=K();_e.registerFlag("HAS_WEBGL",()=>_e.getNumber("WEBGL_VERSION")>0);_e.registerFlag("WEBGL_VERSION",()=>Yy(2)?2:Yy(1)?1:0);_e.registerFlag("WEBGL_CHECK_NUMERICAL_PROBLEMS",()=>!1);_e.registerFlag("WEBGL_BUFFER_SUPPORTED",()=>_e.get("WEBGL_VERSION")===2);_e.registerFlag("WEBGL_CPU_FORWARD",()=>!0);_e.registerFlag("WEBGL_FORCE_F16_TEXTURES",()=>!1);_e.registerFlag("WEBGL_PACK",()=>_e.getBool("HAS_WEBGL"));_e.registerFlag("WEBGL_PACK_NORMALIZATION",()=>_e.getBool("WEBGL_PACK"));_e.registerFlag("WEBGL_PACK_CLIP",()=>_e.getBool("WEBGL_PACK"));_e.registerFlag("WEBGL_PACK_DEPTHWISECONV",()=>_e.getBool("WEBGL_PACK"));_e.registerFlag("WEBGL_PACK_BINARY_OPERATIONS",()=>_e.getBool("WEBGL_PACK"));_e.registerFlag("WEBGL_PACK_UNARY_OPERATIONS",()=>_e.getBool("WEBGL_PACK"));_e.registerFlag("WEBGL_PACK_ARRAY_OPERATIONS",()=>_e.getBool("WEBGL_PACK"));_e.registerFlag("WEBGL_PACK_IMAGE_OPERATIONS",()=>_e.getBool("WEBGL_PACK"));_e.registerFlag("WEBGL_PACK_REDUCE",()=>_e.getBool("WEBGL_PACK"));_e.registerFlag("WEBGL_LAZILY_UNPACK",()=>_e.getBool("WEBGL_PACK"));_e.registerFlag("WEBGL_CONV_IM2COL",()=>_e.getBool("WEBGL_PACK"));_e.registerFlag("WEBGL_MAX_TEXTURE_SIZE",()=>mI(_e.getNumber("WEBGL_VERSION")));_e.registerFlag("WEBGL_MAX_TEXTURES_IN_SHADER",()=>gI(_e.getNumber("WEBGL_VERSION")));_e.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION",()=>{let e=_e.getNumber("WEBGL_VERSION");return e===0?0:AI(e)});_e.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE",()=>_e.getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0&&!Nu.isMobile());_e.registerFlag("WEBGL_RENDER_FLOAT32_CAPABLE",()=>yI(_e.getNumber("WEBGL_VERSION")));_e.registerFlag("WEBGL_RENDER_FLOAT32_ENABLED",()=>_e.getBool("WEBGL_FORCE_F16_TEXTURES")?!1:_e.getBool("WEBGL_RENDER_FLOAT32_CAPABLE"));_e.registerFlag("WEBGL_DOWNLOAD_FLOAT_ENABLED",()=>xI(_e.getNumber("WEBGL_VERSION")));_e.registerFlag("WEBGL_FENCE_API_ENABLED",()=>bI(_e.getNumber("WEBGL_VERSION")));_e.registerFlag("WEBGL_SIZE_UPLOAD_UNIFORM",()=>_e.getBool("WEBGL_RENDER_FLOAT32_ENABLED")?4:0);_e.registerFlag("WEBGL_DELETE_TEXTURE_THRESHOLD",()=>-1,e=>{if(e<0&&e!==-1)throw new Error(`WEBGL_DELETE_TEXTURE_THRESHOLD must be -1 (indicating never delete) or at least 0, but got ${e}.`)});_e.registerFlag("WEBGL_FLUSH_THRESHOLD",()=>Nu.isMobile()?1:-1,e=>{if(e<0&&e!==-1)throw new Error(`WEBGL_FLUSH_THRESHOLD must be -1 (indicating never manual flush) or at least 0, but got ${e}.`)});_e.registerFlag("CPU_HANDOFF_SIZE_THRESHOLD",()=>128);_e.registerFlag("WEBGL_USE_SHAPES_UNIFORMS",()=>!1);_e.registerFlag("TOPK_LAST_DIM_CPU_HANDOFF_SIZE_THRESHOLD",()=>1e5);_e.registerFlag("TOPK_K_CPU_HANDOFF_THRESHOLD",()=>128);function Un(){let e,t,n,s,r,a,o,i,l,c;return K().getNumber("WEBGL_VERSION")===2?(e="#version 300 es",t="in",n="out",s="in",r="texture",a="outputColor",o="out vec4 outputColor;",i=`
|
|
bool isnan_custom(float val) {
|
|
return (val > 0.0 || val < 0.0) ? false : val != 0.0;
|
|
}
|
|
|
|
bvec4 isnan_custom(vec4 val) {
|
|
return bvec4(isnan_custom(val.x),
|
|
isnan_custom(val.y), isnan_custom(val.z), isnan_custom(val.w));
|
|
}
|
|
|
|
#define isnan(value) isnan_custom(value)
|
|
`,l="",c=`
|
|
#define round(value) newRound(value)
|
|
int newRound(float value) {
|
|
return int(floor(value + 0.5));
|
|
}
|
|
|
|
ivec4 newRound(vec4 value) {
|
|
return ivec4(floor(value + vec4(0.5)));
|
|
}
|
|
`):(e="",t="attribute",n="varying",s="varying",r="texture2D",a="gl_FragColor",o="",i=`
|
|
#define isnan(value) isnan_custom(value)
|
|
bool isnan_custom(float val) {
|
|
return (val > 0. || val < 1. || val == 0.) ? false : true;
|
|
}
|
|
bvec4 isnan_custom(vec4 val) {
|
|
return bvec4(isnan(val.x), isnan(val.y), isnan(val.z), isnan(val.w));
|
|
}
|
|
`,l=`
|
|
uniform float INFINITY;
|
|
|
|
bool isinf(float val) {
|
|
return abs(val) == INFINITY;
|
|
}
|
|
bvec4 isinf(vec4 val) {
|
|
return equal(abs(val), vec4(INFINITY));
|
|
}
|
|
`,c=`
|
|
int round(float value) {
|
|
return int(floor(value + 0.5));
|
|
}
|
|
|
|
ivec4 round(vec4 value) {
|
|
return ivec4(floor(value + vec4(0.5)));
|
|
}
|
|
`),{version:e,attribute:t,varyingVs:n,varyingFs:s,texture2D:r,output:a,defineOutput:o,defineSpecialNaN:i,defineSpecialInf:l,defineRound:c}}function Il(e,t,n="index"){let s=v.computeStrides(t);return s.map((r,a)=>{let o=`int ${e[a]} = ${n} / ${r}`,i=a===s.length-1?`int ${e[a+1]} = ${n} - ${e[a]} * ${r}`:`index -= ${e[a]} * ${r}`;return`${o}; ${i};`}).join("")}function Pm(e,t,n="index"){let s=v.computeStrides(t);return s.map((r,a)=>{let o=`int ${e[a]} = ${n} / outShapeStrides[${a}]`,i=a===s.length-1?`int ${e[a+1]} = ${n} - ${e[a]} * outShapeStrides[${a}]`:`index -= ${e[a]} * outShapeStrides[${a}]`;return`${o}; ${i};`}).join("")}function QZ(e,t){let n=e.length,s=e.map(a=>`${t}[${a}]`),r=new Array(n-1);r[n-2]=s[n-1];for(let a=n-3;a>=0;--a)r[a]=`(${r[a+1]} * ${s[a+1]})`;return r}function eY(e,t,n="index"){let s=e.map((a,o)=>o),r=QZ(s,t);return r.map((a,o)=>{let i=`int ${e[o]} = ${n} / ${r[o]}`,l=o===r.length-1?`int ${e[o+1]} = ${n} - ${e[o]} * ${r[o]}`:`index -= ${e[o]} * ${r[o]}`;return`${i}; ${l};`}).join("")}function Qy(e){let t=v.computeStrides(e).map(n=>n.toString());return`
|
|
int getFlatIndex(ivec3 coords) {
|
|
return coords.x * ${t[0]} + coords.y * ${t[1]} + coords.z;
|
|
}
|
|
`}function ex(){return`
|
|
int getFlatIndex(ivec3 coords) {
|
|
return coords.x * outShapeStrides[0] + coords.y * outShapeStrides[1] + coords.z;
|
|
}
|
|
`}var vI=`
|
|
const float FLOAT_MAX = 1.70141184e38;
|
|
const float FLOAT_MIN = 1.17549435e-38;
|
|
|
|
lowp vec4 encode_float(highp float v) {
|
|
if (isnan(v)) {
|
|
return vec4(255, 255, 255, 255);
|
|
}
|
|
|
|
highp float av = abs(v);
|
|
|
|
if(av < FLOAT_MIN) {
|
|
return vec4(0.0, 0.0, 0.0, 0.0);
|
|
} else if(v > FLOAT_MAX) {
|
|
return vec4(0.0, 0.0, 128.0, 127.0) / 255.0;
|
|
} else if(v < -FLOAT_MAX) {
|
|
return vec4(0.0, 0.0, 128.0, 255.0) / 255.0;
|
|
}
|
|
|
|
highp vec4 c = vec4(0,0,0,0);
|
|
|
|
highp float e = floor(log2(av));
|
|
highp float m = exp2(fract(log2(av))) - 1.0;
|
|
|
|
c[2] = floor(128.0 * m);
|
|
m -= c[2] / 128.0;
|
|
c[1] = floor(32768.0 * m);
|
|
m -= c[1] / 32768.0;
|
|
c[0] = floor(8388608.0 * m);
|
|
|
|
highp float ebias = e + 127.0;
|
|
c[3] = floor(ebias / 2.0);
|
|
ebias -= c[3] * 2.0;
|
|
c[2] += floor(ebias) * 128.0;
|
|
|
|
c[3] += 128.0 * step(0.0, -v);
|
|
|
|
return c / 255.0;
|
|
}
|
|
`,{getBroadcastDims:wI}=E;function tY(e,t,n){let s=[];if(e.forEach(h=>{let f=v.sizeFromShape(h.shapeInfo.logicalShape);if(h.shapeInfo.isUniform?s.push(`uniform float ${h.name}${f>1?`[${f}]`:""};`):(s.push(`uniform sampler2D ${h.name};`),s.push(`uniform int offset${h.name};`)),n.enableShapeUniforms){let{uniformShape:m}=tx(n.packedInputs,h.shapeInfo.logicalShape,h.shapeInfo.texShape);switch(m.length){case 1:s.push(`uniform int ${h.name}Shape;`);break;case 2:s.push(`uniform ivec2 ${h.name}Shape;`);break;case 3:s.push(`uniform ivec3 ${h.name}Shape;`);break;case 4:s.push(`uniform ivec4 ${h.name}Shape;`);break;default:break}s.push(`uniform ivec2 ${h.name}TexShape;`)}}),n.enableShapeUniforms){switch(t.logicalShape.length){case 1:s.push("uniform int outShape;");break;case 2:s.push("uniform ivec2 outShape;"),s.push("uniform int outShapeStrides;");break;case 3:s.push("uniform ivec3 outShape;"),s.push("uniform ivec2 outShapeStrides;");break;case 4:s.push("uniform ivec4 outShape;"),s.push("uniform ivec3 outShapeStrides;");break;default:break}s.push("uniform ivec2 outTexShape;")}n.customUniforms&&n.customUniforms.forEach(h=>{s.push(`uniform ${h.type} ${h.name}${h.arrayIndex?`[${h.arrayIndex}]`:""};`)});let r=s.join(`
|
|
`),a=e.map(h=>nY(h,t,n.packedInputs,n.enableShapeUniforms)).join(`
|
|
`),o=t.texShape,i=Un(),l=aY(i),c,u,d=lY(i);return t.isPacked?(c=sY(t.logicalShape,o,n.enableShapeUniforms),u=iY(i)):(c=rY(t.logicalShape,o,n.enableShapeUniforms),u=oY(i)),n.packedInputs&&(d+=pY),[d,l,u,r,c,a,n.userCode].join(`
|
|
`)}function sc(e,t=!1){let n=e.shapeInfo.logicalShape;switch(n.length){case 0:return SY(e,t);case 1:return CY(e,t);case 2:return NY(e,t);case 3:return RY(e,t);case 4:return _Y(e,t);case 5:return DY(e);case 6:return PY(e);default:throw new Error(`${n.length}-D input sampling is not yet supported`)}}function kI(e,t){switch(e.shapeInfo.logicalShape.length){case 0:return kY(e);case 1:return IY(e,t);case 2:return TY(e,t);case 3:return EY(e,t);default:return $Y(e,t)}}function nY(e,t,n=!1,s){let r="";n?r+=kI(e,s):r+=sc(e,s);let a=e.shapeInfo.logicalShape,o=t.logicalShape;return a.length<=o.length&&(n?r+=FY(e,t):r+=OY(e,t)),r}function sY(e,t,n){switch(e.length){case 0:return SI();case 1:return hY(e,t,n);case 2:return vY(e,t,n);case 3:return mY(e,t,n);default:return AY(e,t,n)}}function rY(e,t,n){switch(e.length){case 0:return SI();case 1:return fY(e,t,n);case 2:return wY(e,t,n);case 3:return gY(e,t,n);case 4:return yY(e,t,n);case 5:return xY(e,t);case 6:return bY(e,t);default:throw new Error(`${e.length}-D output sampling is not yet supported`)}}function aY(e){return`
|
|
float sampleTexture(sampler2D textureSampler, vec2 uv) {
|
|
return ${e.texture2D}(textureSampler, uv).r;
|
|
}
|
|
`}function oY(e){return`
|
|
void setOutput(float val) {
|
|
${e.output} = vec4(val, 0, 0, 0);
|
|
}
|
|
`}function iY(e){return`
|
|
void setOutput(vec4 val) {
|
|
${e.output} = val;
|
|
}
|
|
`}function lY(e){return`${e.version}
|
|
precision highp float;
|
|
precision highp int;
|
|
precision highp sampler2D;
|
|
${e.varyingFs} vec2 resultUV;
|
|
${e.defineOutput}
|
|
const vec2 halfCR = vec2(0.5, 0.5);
|
|
|
|
struct ivec5
|
|
{
|
|
int x;
|
|
int y;
|
|
int z;
|
|
int w;
|
|
int u;
|
|
};
|
|
|
|
struct ivec6
|
|
{
|
|
int x;
|
|
int y;
|
|
int z;
|
|
int w;
|
|
int u;
|
|
int v;
|
|
};
|
|
|
|
uniform float NAN;
|
|
${e.defineSpecialNaN}
|
|
${e.defineSpecialInf}
|
|
${e.defineRound}
|
|
|
|
int imod(int x, int y) {
|
|
return x - y * (x / y);
|
|
}
|
|
|
|
int idiv(int a, int b, float sign) {
|
|
int res = a / b;
|
|
int mod = imod(a, b);
|
|
if (sign < 0. && mod != 0) {
|
|
res -= 1;
|
|
}
|
|
return res;
|
|
}
|
|
|
|
//Based on the work of Dave Hoskins
|
|
//https://www.shadertoy.com/view/4djSRW
|
|
#define HASHSCALE1 443.8975
|
|
float random(float seed){
|
|
vec2 p = resultUV * seed;
|
|
vec3 p3 = fract(vec3(p.xyx) * HASHSCALE1);
|
|
p3 += dot(p3, p3.yzx + 19.19);
|
|
return fract((p3.x + p3.y) * p3.z);
|
|
}
|
|
|
|
${uY}
|
|
${cY}
|
|
${dY}
|
|
`}var uY=`
|
|
vec2 uvFromFlat(int texNumR, int texNumC, int index) {
|
|
int texR = index / texNumC;
|
|
int texC = index - texR * texNumC;
|
|
return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);
|
|
}
|
|
vec2 packedUVfrom1D(int texNumR, int texNumC, int index) {
|
|
int texelIndex = index / 2;
|
|
int texR = texelIndex / texNumC;
|
|
int texC = texelIndex - texR * texNumC;
|
|
return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);
|
|
}
|
|
`,cY=`
|
|
vec2 packedUVfrom2D(int texelsInLogicalRow, int texNumR,
|
|
int texNumC, int row, int col) {
|
|
int texelIndex = (row / 2) * texelsInLogicalRow + (col / 2);
|
|
int texR = texelIndex / texNumC;
|
|
int texC = texelIndex - texR * texNumC;
|
|
return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);
|
|
}
|
|
`,dY=`
|
|
vec2 packedUVfrom3D(int texNumR, int texNumC,
|
|
int texelsInBatch, int texelsInLogicalRow, int b,
|
|
int row, int col) {
|
|
int index = b * texelsInBatch + (row / 2) * texelsInLogicalRow + (col / 2);
|
|
int texR = index / texNumC;
|
|
int texC = index - texR * texNumC;
|
|
return (vec2(texC, texR) + halfCR) / vec2(texNumC, texNumR);
|
|
}
|
|
`,pY=`
|
|
float getChannel(vec4 frag, vec2 innerDims) {
|
|
vec2 modCoord = mod(innerDims, 2.);
|
|
return modCoord.x == 0. ?
|
|
(modCoord.y == 0. ? frag.r : frag.g) :
|
|
(modCoord.y == 0. ? frag.b : frag.a);
|
|
}
|
|
float getChannel(vec4 frag, int dim) {
|
|
float modCoord = mod(float(dim), 2.);
|
|
return modCoord == 0. ? frag.r : frag.g;
|
|
}
|
|
`;function SI(){return`
|
|
int getOutputCoords() {
|
|
return 0;
|
|
}
|
|
`}function hY(e,t,n){let s=[Math.ceil(t[0]/2),Math.ceil(t[1]/2)];return s[0]===1?n?`
|
|
int getOutputCoords() {
|
|
return 2 * int(resultUV.x * ceil(float(outTexShape[1]) / 2.0));
|
|
}
|
|
`:`
|
|
int getOutputCoords() {
|
|
return 2 * int(resultUV.x * ${s[1]}.0);
|
|
}
|
|
`:s[1]===1?n?`
|
|
int getOutputCoords() {
|
|
return 2 * int(resultUV.y * ceil(float(outTexShape[0]) / 2.0));
|
|
}
|
|
`:`
|
|
int getOutputCoords() {
|
|
return 2 * int(resultUV.y * ${s[0]}.0);
|
|
}
|
|
`:n?`
|
|
int getOutputCoords() {
|
|
ivec2 packedTexShape = ivec2(ceil(float(outTexShape[0]) / 2.0), ceil(float(outTexShape[1]) / 2.0));
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(packedTexShape[0], packedTexShape[1]));
|
|
return 2 * (resTexRC.x * packedTexShape[1] + resTexRC.y);
|
|
}
|
|
`:`
|
|
int getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${s[0]}, ${s[1]}));
|
|
return 2 * (resTexRC.x * ${s[1]} + resTexRC.y);
|
|
}
|
|
`}function fY(e,t,n){return t[0]===1?n?`
|
|
int getOutputCoords() {
|
|
return int(resultUV.x * float(outTexShape[1]));
|
|
}
|
|
`:`
|
|
int getOutputCoords() {
|
|
return int(resultUV.x * ${t[1]}.0);
|
|
}
|
|
`:t[1]===1?n?`
|
|
int getOutputCoords() {
|
|
return int(resultUV.y * float(outTexShape[0]));
|
|
}
|
|
`:`
|
|
int getOutputCoords() {
|
|
return int(resultUV.y * ${t[0]}.0);
|
|
}
|
|
`:n?`
|
|
int getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(outTexShape[0], outTexShape[1]));
|
|
return resTexRC.x * outTexShape[1] + resTexRC.y;
|
|
}
|
|
`:`
|
|
int getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${t[0]}, ${t[1]}));
|
|
return resTexRC.x * ${t[1]} + resTexRC.y;
|
|
}
|
|
`}function mY(e,t,n){if(n)return`
|
|
ivec3 getOutputCoords() {
|
|
ivec2 packedTexShape = ivec2(ceil(float(outTexShape[0]) / 2.0), ceil(float(outTexShape[1]) / 2.0));
|
|
int texelsInLogicalRow = int(ceil(float(outShape[2]) / 2.0));
|
|
int texelsInBatch = texelsInLogicalRow * int(ceil(float(outShape[1]) / 2.0));
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(packedTexShape[0], packedTexShape[1]));
|
|
int index = resTexRC.x * packedTexShape[1] + resTexRC.y;
|
|
|
|
int b = index / texelsInBatch;
|
|
index -= b * texelsInBatch;
|
|
|
|
int r = 2 * (index / texelsInLogicalRow);
|
|
int c = imod(index, texelsInLogicalRow) * 2;
|
|
|
|
return ivec3(b, r, c);
|
|
}
|
|
`;let s=[Math.ceil(t[0]/2),Math.ceil(t[1]/2)],r=Math.ceil(e[2]/2),a=r*Math.ceil(e[1]/2);return`
|
|
ivec3 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${s[0]}, ${s[1]}));
|
|
int index = resTexRC.x * ${s[1]} + resTexRC.y;
|
|
|
|
int b = index / ${a};
|
|
index -= b * ${a};
|
|
|
|
int r = 2 * (index / ${r});
|
|
int c = imod(index, ${r}) * 2;
|
|
|
|
return ivec3(b, r, c);
|
|
}
|
|
`}function gY(e,t,n){if(n)return`
|
|
ivec3 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(outTexShape[0], outTexShape[1]));
|
|
int index = resTexRC.x * outTexShape[1] + resTexRC.y;
|
|
${Pm(["r","c","d"],e)}
|
|
return ivec3(r, c, d);
|
|
}
|
|
`;let s=Il(["r","c","d"],e);return`
|
|
ivec3 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${t[0]}, ${t[1]}));
|
|
int index = resTexRC.x * ${t[1]} + resTexRC.y;
|
|
${s}
|
|
return ivec3(r, c, d);
|
|
}
|
|
`}function AY(e,t,n){if(n)return`
|
|
ivec4 getOutputCoords() {
|
|
ivec2 packedTexShape = ivec2(ceil(float(outTexShape[0]) / 2.0), ceil(float(outTexShape[1]) / 2.0));
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(packedTexShape[0], packedTexShape[1]));
|
|
int index = resTexRC.x * packedTexShape[1] + resTexRC.y;
|
|
|
|
int texelsInLogicalRow = int(ceil(float(outShape[3]) / 2.0));
|
|
int texelsInBatch = texelsInLogicalRow * int(ceil(float(outShape[2]) / 2.0));
|
|
int texelsInBatchN = texelsInBatch * outShape[1];
|
|
|
|
int b2 = index / texelsInBatchN;
|
|
index -= b2 * texelsInBatchN;
|
|
|
|
int b = index / texelsInBatch;
|
|
index -= b * texelsInBatch;
|
|
|
|
int r = 2 * (index / texelsInLogicalRow);
|
|
int c = imod(index, texelsInLogicalRow) * 2;
|
|
|
|
return ivec4(b2, b, r, c);
|
|
}
|
|
`;let s=[Math.ceil(t[0]/2),Math.ceil(t[1]/2)],r=Math.ceil(e[e.length-1]/2),a=r*Math.ceil(e[e.length-2]/2),o=a,i="",l="b, r, c";for(let c=2;c<e.length-1;c++)o*=e[e.length-c-1],i=`
|
|
int b${c} = index / ${o};
|
|
index -= b${c} * ${o};
|
|
`+i,l=`b${c}, `+l;return`
|
|
ivec${e.length} getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${s[0]}, ${s[1]}));
|
|
int index = resTexRC.x * ${s[1]} + resTexRC.y;
|
|
|
|
${i}
|
|
|
|
int b = index / ${a};
|
|
index -= b * ${a};
|
|
|
|
int r = 2 * (index / ${r});
|
|
int c = imod(index, ${r}) * 2;
|
|
|
|
return ivec${e.length}(${l});
|
|
}
|
|
`}function yY(e,t,n){if(n)return`
|
|
ivec4 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(outTexShape[0], outTexShape[1]));
|
|
int index = resTexRC.x * outTexShape[1] + resTexRC.y;
|
|
${Pm(["r","c","d","d2"],e)}
|
|
return ivec4(r, c, d, d2);
|
|
}
|
|
`;let s=Il(["r","c","d","d2"],e);return`
|
|
ivec4 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${t[0]}, ${t[1]}));
|
|
int index = resTexRC.x * ${t[1]} + resTexRC.y;
|
|
${s}
|
|
return ivec4(r, c, d, d2);
|
|
}
|
|
`}function xY(e,t){let n=Il(["r","c","d","d2","d3"],e);return`
|
|
ivec5 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx * vec2(${t[0]},
|
|
${t[1]}));
|
|
|
|
int index = resTexRC.x * ${t[1]} + resTexRC.y;
|
|
|
|
${n}
|
|
|
|
ivec5 outShape = ivec5(r, c, d, d2, d3);
|
|
return outShape;
|
|
}
|
|
`}function bY(e,t){let n=Il(["r","c","d","d2","d3","d4"],e);return`
|
|
ivec6 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${t[0]}, ${t[1]}));
|
|
int index = resTexRC.x * ${t[1]} + resTexRC.y;
|
|
|
|
${n}
|
|
|
|
ivec6 result = ivec6(r, c, d, d2, d3, d4);
|
|
return result;
|
|
}
|
|
`}function vY(e,t,n){let s=[Math.ceil(t[0]/2),Math.ceil(t[1]/2)];if(v.arraysEqual(e,t))return n?`
|
|
ivec2 getOutputCoords() {
|
|
ivec2 packedTexShape = ivec2(ceil(float(outTexShape[0]) / 2.0), ceil(float(outTexShape[1]) / 2.0));
|
|
return 2 * ivec2(resultUV.yx * vec2(packedTexShape[0], packedTexShape[1]));
|
|
}
|
|
`:`
|
|
ivec2 getOutputCoords() {
|
|
return 2 * ivec2(resultUV.yx * vec2(${s[0]}, ${s[1]}));
|
|
}
|
|
`;let r=Math.ceil(e[1]/2);return n?`
|
|
ivec2 getOutputCoords() {
|
|
ivec2 packedTexShape = ivec2(ceil(float(outTexShape[0]) / 2.0), ceil(float(outTexShape[1]) / 2.0));
|
|
int texelsInLogicalRow = int(ceil(float(outShape[1]) / 2.0));
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(packedTexShape[0], packedTexShape[1]));
|
|
|
|
int index = resTexRC.x * packedTexShape[1] + resTexRC.y;
|
|
int r = 2 * (index / texelsInLogicalRow);
|
|
int c = imod(index, texelsInLogicalRow) * 2;
|
|
|
|
return ivec2(r, c);
|
|
}
|
|
`:`
|
|
ivec2 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${s[0]}, ${s[1]}));
|
|
|
|
int index = resTexRC.x * ${s[1]} + resTexRC.y;
|
|
int r = 2 * (index / ${r});
|
|
int c = imod(index, ${r}) * 2;
|
|
|
|
return ivec2(r, c);
|
|
}
|
|
`}function wY(e,t,n){return v.arraysEqual(e,t)?n?`
|
|
ivec2 getOutputCoords() {
|
|
return ivec2(resultUV.yx * vec2(outTexShape[0], outTexShape[1]));
|
|
}
|
|
`:`
|
|
ivec2 getOutputCoords() {
|
|
return ivec2(resultUV.yx * vec2(${t[0]}, ${t[1]}));
|
|
}
|
|
`:e[1]===1?n?`
|
|
ivec2 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(outTexShape[0], outTexShape[1]));
|
|
int index = resTexRC.x * outTexShape[1] + resTexRC.y;
|
|
return ivec2(index, 0);
|
|
}
|
|
`:`
|
|
ivec2 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${t[0]}, ${t[1]}));
|
|
int index = resTexRC.x * ${t[1]} + resTexRC.y;
|
|
return ivec2(index, 0);
|
|
}
|
|
`:e[0]===1?n?`
|
|
ivec2 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(outTexShape[0], outTexShape[1]));
|
|
int index = resTexRC.x * outTexShape[1] + resTexRC.y;
|
|
return ivec2(0, index);
|
|
}
|
|
`:`
|
|
ivec2 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${t[0]}, ${t[1]}));
|
|
int index = resTexRC.x * ${t[1]} + resTexRC.y;
|
|
return ivec2(0, index);
|
|
}
|
|
`:n?`
|
|
ivec2 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(outTexShape[0], outTexShape[1]));
|
|
int index = resTexRC.x * outTexShape[1] + resTexRC.y;
|
|
int r = index / outShape[1];
|
|
int c = index - r * outShape[1];
|
|
return ivec2(r, c);
|
|
}
|
|
`:`
|
|
ivec2 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${t[0]}, ${t[1]}));
|
|
int index = resTexRC.x * ${t[1]} + resTexRC.y;
|
|
int r = index / ${e[1]};
|
|
int c = index - r * ${e[1]};
|
|
return ivec2(r, c);
|
|
}
|
|
`}function Cl(e){return`offset${e}`}function kY(e){let t=e.name,n="get"+t.charAt(0).toUpperCase()+t.slice(1),s=Un();return`
|
|
vec4 ${n}() {
|
|
return ${s.texture2D}(${t}, halfCR);
|
|
}
|
|
`}function SY(e,t){let n=e.name,s="get"+n.charAt(0).toUpperCase()+n.slice(1);if(e.shapeInfo.isUniform)return`float ${s}() {return ${n};}`;let[r,a]=e.shapeInfo.texShape;if(r===1&&a===1)return`
|
|
float ${s}() {
|
|
return sampleTexture(${n}, halfCR);
|
|
}
|
|
`;let o=Cl(n);if(t)return`
|
|
float ${s}() {
|
|
vec2 uv = uvFromFlat(${n}TexShape[0], ${n}TexShape[1], ${o});
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`;let[i,l]=e.shapeInfo.texShape;return`
|
|
float ${s}() {
|
|
vec2 uv = uvFromFlat(${i}, ${l}, ${o});
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`}function IY(e,t){let n=e.name,s="get"+n.charAt(0).toUpperCase()+n.slice(1),r=e.shapeInfo.texShape,a=Un();if(t)return`
|
|
vec4 ${s}(int index) {
|
|
ivec2 packedTexShape = ivec2(ceil(float(${n}TexShape[0]) / 2.0), ceil(float(${n}TexShape[1]) / 2.0));
|
|
vec2 uv = packedUVfrom1D(
|
|
packedTexShape[0], packedTexShape[1], index);
|
|
return ${a.texture2D}(${n}, uv);
|
|
}
|
|
`;let o=[Math.ceil(r[0]/2),Math.ceil(r[1]/2)];return`
|
|
vec4 ${s}(int index) {
|
|
vec2 uv = packedUVfrom1D(
|
|
${o[0]}, ${o[1]}, index);
|
|
return ${a.texture2D}(${n}, uv);
|
|
}
|
|
`}function CY(e,t){let n=e.name,s="get"+n.charAt(0).toUpperCase()+n.slice(1);if(e.shapeInfo.isUniform)return`
|
|
float ${s}(int index) {
|
|
${rc(e)}
|
|
}
|
|
`;let r=e.shapeInfo.texShape,a=r[0],o=r[1];if(o===1&&a===1)return`
|
|
float ${s}(int index) {
|
|
return sampleTexture(${n}, halfCR);
|
|
}
|
|
`;let i=Cl(n);return o===1?t?`
|
|
float ${s}(int index) {
|
|
vec2 uv = vec2(0.5, (float(index + ${i}) + 0.5) / float(${n}TexShape[0]));
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`:`
|
|
float ${s}(int index) {
|
|
vec2 uv = vec2(0.5, (float(index + ${i}) + 0.5) / ${a}.0);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`:a===1?t?`
|
|
float ${s}(int index) {
|
|
vec2 uv = vec2((float(index + ${i}) + 0.5) / float(${n}TexShape[1]), 0.5);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`:`
|
|
float ${s}(int index) {
|
|
vec2 uv = vec2((float(index + ${i}) + 0.5) / ${o}.0, 0.5);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`:t?`
|
|
float ${s}(int index) {
|
|
vec2 uv = uvFromFlat(${n}TexShape[0], ${n}TexShape[1], index + ${i});
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`:`
|
|
float ${s}(int index) {
|
|
vec2 uv = uvFromFlat(${a}, ${o}, index + ${i});
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`}function TY(e,t){let n=e.shapeInfo.logicalShape,s=e.name,r="get"+s.charAt(0).toUpperCase()+s.slice(1),a=e.shapeInfo.texShape,o=a[0],i=a[1],l=Un();if(a!=null&&v.arraysEqual(n,a))return t?`
|
|
vec4 ${r}(int row, int col) {
|
|
vec2 uv = (vec2(col, row) + halfCR) / vec2(${s}TexShape[1], ${s}TexShape[0]);
|
|
|
|
return ${l.texture2D}(${s}, uv);
|
|
}
|
|
`:`
|
|
vec4 ${r}(int row, int col) {
|
|
vec2 uv = (vec2(col, row) + halfCR) / vec2(${i}.0, ${o}.0);
|
|
|
|
return ${l.texture2D}(${s}, uv);
|
|
}
|
|
`;if(t)return`
|
|
vec4 ${r}(int row, int col) {
|
|
ivec2 packedTexShape = ivec2(ceil(float(${s}TexShape[0]) / 2.0), ceil(float(${s}TexShape[1]) / 2.0));
|
|
int valuesPerRow = int(ceil(float(${s}Shape[1]) / 2.0));
|
|
vec2 uv = packedUVfrom2D(valuesPerRow, packedTexShape[0], packedTexShape[1], row, col);
|
|
return ${l.texture2D}(${s}, uv);
|
|
}
|
|
`;let c=[Math.ceil(a[0]/2),Math.ceil(a[1]/2)],u=Math.ceil(n[1]/2);return`
|
|
vec4 ${r}(int row, int col) {
|
|
vec2 uv = packedUVfrom2D(${u}, ${c[0]}, ${c[1]}, row, col);
|
|
return ${l.texture2D}(${s}, uv);
|
|
}
|
|
`}function NY(e,t){let n=e.shapeInfo.logicalShape,s=e.name,r="get"+s.charAt(0).toUpperCase()+s.slice(1),a=e.shapeInfo.texShape;if(a!=null&&v.arraysEqual(n,a)){if(t)return`
|
|
float ${r}(int row, int col) {
|
|
vec2 uv = (vec2(col, row) + halfCR) / vec2(${s}TexShape[1], ${s}TexShape[0]);
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`;let p=a[0],h=a[1];return`
|
|
float ${r}(int row, int col) {
|
|
vec2 uv = (vec2(col, row) + halfCR) / vec2(${h}.0, ${p}.0);
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`}let{newShape:o,keptDims:i}=v.squeezeShape(n),l=o;if(l.length<n.length){let p=ac(e,l),h=["row","col"];return`
|
|
${sc(p,t)}
|
|
float ${r}(int row, int col) {
|
|
return ${r}(${oc(h,i)});
|
|
}
|
|
`}if(e.shapeInfo.isUniform)return`
|
|
float ${r}(int row, int col) {
|
|
int index = round(dot(vec2(row, col), vec2(${n[1]}, 1)));
|
|
${rc(e)}
|
|
}
|
|
`;let c=a[0],u=a[1],d=Cl(s);return u===1?t?`
|
|
float ${r}(int row, int col) {
|
|
float index = dot(vec3(row, col, ${d}), vec3(${s}Shape[1], 1, 1));
|
|
vec2 uv = vec2(0.5, (index + 0.5) / float(${s}TexShape[0]));
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`:`
|
|
float ${r}(int row, int col) {
|
|
float index = dot(vec3(row, col, ${d}), vec3(${n[1]}, 1, 1));
|
|
vec2 uv = vec2(0.5, (index + 0.5) / ${c}.0);
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`:c===1?t?`
|
|
float ${r}(int row, int col) {
|
|
float index = dot(vec3(row, col, ${d}), vec3(${s}Shape[1], 1, 1));
|
|
vec2 uv = vec2((index + 0.5) / float(${s}TexShape[1]), 0.5);
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`:`
|
|
float ${r}(int row, int col) {
|
|
float index = dot(vec3(row, col, ${d}), vec3(${n[1]}, 1, 1));
|
|
vec2 uv = vec2((index + 0.5) / ${u}.0, 0.5);
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`:t?`
|
|
float ${r}(int row, int col) {
|
|
// Explicitly use integer operations as dot() only works on floats.
|
|
int index = row * ${s}Shape[1] + col + ${d};
|
|
vec2 uv = uvFromFlat(${s}TexShape[0], ${s}TexShape[1], index);
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`:`
|
|
float ${r}(int row, int col) {
|
|
// Explicitly use integer operations as dot() only works on floats.
|
|
int index = row * ${n[1]} + col + ${d};
|
|
vec2 uv = uvFromFlat(${c}, ${u}, index);
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`}function EY(e,t){let n=e.shapeInfo.logicalShape,s=e.name,r="get"+s.charAt(0).toUpperCase()+s.slice(1),a=e.shapeInfo.texShape,o=[Math.ceil(a[0]/2),Math.ceil(a[1]/2)];if(n[0]===1){let p=n.slice(1),h=[1,2],f=ac(e,p),m=["b","row","col"];return`
|
|
${kI(f,t)}
|
|
vec4 ${r}(int b, int row, int col) {
|
|
return ${r}(${oc(m,h)});
|
|
}
|
|
`}let i=Un();if(t)return`
|
|
vec4 ${r}(int b, int row, int col) {
|
|
ivec2 packedTexShape = ivec2(ceil(float(${s}TexShape[0]) / 2.0), ceil(float(${s}TexShape[1]) / 2.0));
|
|
int valuesPerRow = int(ceil(float(${s}Shape[2]) / 2.0));
|
|
int texelsInBatch = valuesPerRow * int(ceil(float(${s}Shape[1]) / 2.0));
|
|
vec2 uv = packedUVfrom3D(
|
|
packedTexShape[0], packedTexShape[1], texelsInBatch, valuesPerRow, b, row, col);
|
|
return ${i.texture2D}(${s}, uv);
|
|
}
|
|
`;let l=o[0],c=o[1],u=Math.ceil(n[2]/2),d=u*Math.ceil(n[1]/2);return`
|
|
vec4 ${r}(int b, int row, int col) {
|
|
vec2 uv = packedUVfrom3D(
|
|
${l}, ${c}, ${d}, ${u}, b, row, col);
|
|
return ${i.texture2D}(${s}, uv);
|
|
}
|
|
`}function RY(e,t){let n=e.shapeInfo.logicalShape,s=e.name,r="get"+s.charAt(0).toUpperCase()+s.slice(1),a=n[1]*n[2],o=n[2],{newShape:i,keptDims:l}=v.squeezeShape(n),c=i;if(c.length<n.length){let m=ac(e,c),g=["row","col","depth"];return`
|
|
${sc(m,t)}
|
|
float ${r}(int row, int col, int depth) {
|
|
return ${r}(${oc(g,l)});
|
|
}
|
|
`}if(e.shapeInfo.isUniform)return`
|
|
float ${r}(int row, int col, int depth) {
|
|
int index = round(dot(vec3(row, col, depth),
|
|
vec3(${a}, ${o}, 1)));
|
|
${rc(e)}
|
|
}
|
|
`;let u=e.shapeInfo.texShape,d=u[0],p=u[1],h=e.shapeInfo.flatOffset;if(p===a&&h==null)return t?`
|
|
float ${r}(int row, int col, int depth) {
|
|
int stride1 = ${s}Shape[2];
|
|
float texR = float(row);
|
|
float texC = dot(vec2(col, depth), vec2(stride1, 1));
|
|
vec2 uv = (vec2(texC, texR) + halfCR) /
|
|
vec2(${s}TexShape[1], ${s}TexShape[0]);
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`:`
|
|
float ${r}(int row, int col, int depth) {
|
|
float texR = float(row);
|
|
float texC = dot(vec2(col, depth), vec2(${o}, 1));
|
|
vec2 uv = (vec2(texC, texR) + halfCR) /
|
|
vec2(${p}.0, ${d}.0);
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`;if(p===o&&h==null)return t?`
|
|
float ${r}(int row, int col, int depth) {
|
|
float texR = dot(vec2(row, col), vec2(${s}Shape[1], 1));
|
|
float texC = float(depth);
|
|
vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${s}TexShape[1], ${s}TexShape[0]);
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`:`
|
|
float ${r}(int row, int col, int depth) {
|
|
float texR = dot(vec2(row, col), vec2(${n[1]}, 1));
|
|
float texC = float(depth);
|
|
vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${p}.0, ${d}.0);
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`;let f=Cl(s);return t?`
|
|
float ${r}(int row, int col, int depth) {
|
|
// Explicitly use integer operations as dot() only works on floats.
|
|
int stride0 = ${s}Shape[1] * ${s}Shape[2];
|
|
int stride1 = ${s}Shape[2];
|
|
int index = row * ${a} + col * ${o} + depth + ${f};
|
|
vec2 uv = uvFromFlat(${s}TexShape[0], ${s}TexShape[1], index);
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`:`
|
|
float ${r}(int row, int col, int depth) {
|
|
// Explicitly use integer operations as dot() only works on floats.
|
|
int index = row * ${a} + col * ${o} + depth + ${f};
|
|
vec2 uv = uvFromFlat(${d}, ${p}, index);
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`}function $Y(e,t){let n=e.name,s="get"+n.charAt(0).toUpperCase()+n.slice(1),r=Un();if(t)return`
|
|
vec4 ${s}(int b2, int b, int row, int col) {
|
|
int valuesPerRow = int(ceil(float(${n}Shape[3]) / 2.0));
|
|
int texelsInBatch = valuesPerRow * int(ceil(float(${n}Shape[2]) / 2.0));
|
|
int index = b * texelsInBatch + (row / 2) * valuesPerRow + (col / 2);
|
|
texelsInBatch *= ${n}Shape[1];
|
|
index = b2 * texelsInBatch + index;
|
|
ivec2 packedTexShape = ivec2(ceil(float(${n}TexShape[0]) / 2.0), ceil(float(${n}TexShape[1]) / 2.0));
|
|
int texR = index / packedTexShape[1];
|
|
int texC = index - texR * packedTexShape[1];
|
|
vec2 uv = (vec2(texC, texR) + halfCR) / vec2(packedTexShape[1], packedTexShape[0]); return ${r.texture2D}(${n}, uv);
|
|
}
|
|
`;let a=e.shapeInfo.logicalShape,o=a.length,i=e.shapeInfo.texShape,l=[Math.ceil(i[0]/2),Math.ceil(i[1]/2)],c=l[0],u=l[1],d=Math.ceil(a[o-1]/2),p=d*Math.ceil(a[o-2]/2),h="int b, int row, int col",f=`b * ${p} + (row / 2) * ${d} + (col / 2)`;for(let m=2;m<o-1;m++)h=`int b${m}, `+h,p*=a[o-m-1],f=`b${m} * ${p} + `+f;return`
|
|
vec4 ${s}(${h}) {
|
|
int index = ${f};
|
|
int texR = index / ${u};
|
|
int texC = index - texR * ${u};
|
|
vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${u}, ${c});
|
|
return ${r.texture2D}(${n}, uv);
|
|
}
|
|
`}function _Y(e,t){let n=e.shapeInfo.logicalShape,s=e.name,r="get"+s.charAt(0).toUpperCase()+s.slice(1),a=n[3],o=n[2]*a,i=n[1]*o,{newShape:l,keptDims:c}=v.squeezeShape(n);if(l.length<n.length){let x=ac(e,l),y=["row","col","depth","depth2"];return`
|
|
${sc(x,t)}
|
|
float ${r}(int row, int col, int depth, int depth2) {
|
|
return ${r}(${oc(y,c)});
|
|
}
|
|
`}if(e.shapeInfo.isUniform)return`
|
|
float ${r}(int row, int col, int depth, int depth2) {
|
|
int index = round(dot(vec4(row, col, depth, depth2),
|
|
vec4(${i}, ${o}, ${a}, 1)));
|
|
${rc(e)}
|
|
}
|
|
`;let u=e.shapeInfo.flatOffset,d=e.shapeInfo.texShape,p=d[0],h=d[1],f=`int stride2 = ${s}Shape[3];`,m=`int stride1 = ${s}Shape[2] * stride2;`,g=`int stride0 = ${s}Shape[1] * stride1;`;if(h===i&&u==null)return t?`
|
|
float ${r}(int row, int col, int depth, int depth2) {
|
|
${f}
|
|
${m}
|
|
float texR = float(row);
|
|
float texC =
|
|
dot(vec3(col, depth, depth2),
|
|
vec3(stride1, stride2, 1));
|
|
vec2 uv = (vec2(texC, texR) + halfCR) /
|
|
vec2(${s}TexShape[1], ${s}TexShape[0]);
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`:`
|
|
float ${r}(int row, int col, int depth, int depth2) {
|
|
float texR = float(row);
|
|
float texC =
|
|
dot(vec3(col, depth, depth2),
|
|
vec3(${o}, ${a}, 1));
|
|
vec2 uv = (vec2(texC, texR) + halfCR) /
|
|
vec2(${h}.0, ${p}.0);
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`;if(h===a&&u==null)return t?`
|
|
float ${r}(int row, int col, int depth, int depth2) {
|
|
float texR = dot(vec3(row, col, depth),
|
|
vec3(${s}Shape[1] * ${s}Shape[2], ${s}Shape[2], 1));
|
|
float texC = float(depth2);
|
|
vec2 uv = (vec2(texC, texR) + halfCR) /
|
|
vec2(${s}TexShape[1], ${s}TexShape[0]);
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`:`
|
|
float ${r}(int row, int col, int depth, int depth2) {
|
|
float texR = dot(vec3(row, col, depth),
|
|
vec3(${n[1]*n[2]}, ${n[2]}, 1));
|
|
float texC = float(depth2);
|
|
vec2 uv = (vec2(texC, texR) + halfCR) /
|
|
vec2(${h}.0, ${p}.0);
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`;let A=Cl(s);return t?`
|
|
float ${r}(int row, int col, int depth, int depth2) {
|
|
// Explicitly use integer operations as dot() only works on floats.
|
|
${f}
|
|
${m}
|
|
${g}
|
|
int index = row * stride0 + col * stride1 +
|
|
depth * stride2 + depth2;
|
|
vec2 uv = uvFromFlat(${s}TexShape[0], ${s}TexShape[1], index + ${A});
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`:`
|
|
float ${r}(int row, int col, int depth, int depth2) {
|
|
// Explicitly use integer operations as dot() only works on floats.
|
|
int index = row * ${i} + col * ${o} +
|
|
depth * ${a} + depth2;
|
|
vec2 uv = uvFromFlat(${p}, ${h}, index + ${A});
|
|
return sampleTexture(${s}, uv);
|
|
}
|
|
`}function DY(e){let t=e.shapeInfo.logicalShape,n=e.name,s="get"+n.charAt(0).toUpperCase()+n.slice(1),r=t[4],a=t[3]*r,o=t[2]*a,i=t[1]*o,{newShape:l,keptDims:c}=v.squeezeShape(t);if(l.length<t.length){let m=ac(e,l),g=["row","col","depth","depth2","depth3"];return`
|
|
${sc(m)}
|
|
float ${s}(int row, int col, int depth, int depth2, int depth3) {
|
|
return ${s}(${oc(g,c)});
|
|
}
|
|
`}if(e.shapeInfo.isUniform)return`
|
|
float ${s}(int row, int col, int depth, int depth2, int depth3) {
|
|
float index = dot(
|
|
vec4(row, col, depth, depth2),
|
|
vec4(${i}, ${o}, ${a}, ${r})) +
|
|
depth3;
|
|
${rc(e)}
|
|
}
|
|
`;let u=e.shapeInfo.flatOffset,d=e.shapeInfo.texShape,p=d[0],h=d[1];if(h===i&&u==null)return`
|
|
float ${s}(int row, int col, int depth, int depth2, int depth3) {
|
|
int texR = row;
|
|
float texC = dot(vec4(col, depth, depth2, depth3),
|
|
vec4(${o}, ${a}, ${r}, 1));
|
|
vec2 uv = (vec2(texC, texR) + halfCR) /
|
|
vec2(${h}.0, ${p}.0);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`;if(h===r&&u==null)return`
|
|
float ${s}(int row, int col, int depth, int depth2, int depth3) {
|
|
float texR = dot(
|
|
vec4(row, col, depth, depth2),
|
|
vec4(${t[1]*t[2]*t[3]},
|
|
${t[2]*t[3]}, ${t[3]}, 1));
|
|
int texC = depth3;
|
|
vec2 uv = (vec2(texC, texR) + halfCR) /
|
|
vec2(${h}.0, ${p}.0);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`;let f=Cl(n);return`
|
|
float ${s}(int row, int col, int depth, int depth2, int depth3) {
|
|
// Explicitly use integer operations as dot() only works on floats.
|
|
int index = row * ${i} + col * ${o} + depth * ${a} +
|
|
depth2 * ${r} + depth3 + ${f};
|
|
vec2 uv = uvFromFlat(${p}, ${h}, index);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`}function PY(e){let t=e.shapeInfo.logicalShape,n=e.name,s="get"+n.charAt(0).toUpperCase()+n.slice(1),{newShape:r,keptDims:a}=v.squeezeShape(t);if(r.length<t.length){let g=ac(e,r),A=["row","col","depth","depth2","depth3","depth4"];return`
|
|
${sc(g)}
|
|
float ${s}(int row, int col, int depth,
|
|
int depth2, int depth3, int depth4) {
|
|
return ${s}(${oc(A,a)});
|
|
}
|
|
`}let o=t[5],i=t[4]*o,l=t[3]*i,c=t[2]*l,u=t[1]*c;if(e.shapeInfo.isUniform)return`
|
|
float ${s}(int row, int col, int depth,
|
|
int depth2, int depth3, int depth4) {
|
|
int index = round(dot(
|
|
vec4(row, col, depth, depth2),
|
|
vec4(${u}, ${c}, ${l}, ${i})) +
|
|
dot(
|
|
vec2(depth3, depth4),
|
|
vec2(${o}, 1)));
|
|
${rc(e)}
|
|
}
|
|
`;let d=e.shapeInfo.flatOffset,p=e.shapeInfo.texShape,h=p[0],f=p[1];if(f===u&&d==null)return`
|
|
float ${s}(int row, int col, int depth,
|
|
int depth2, int depth3, int depth4) {
|
|
int texR = row;
|
|
float texC = dot(vec4(col, depth, depth2, depth3),
|
|
vec4(${c}, ${l}, ${i}, ${o})) +
|
|
float(depth4);
|
|
vec2 uv = (vec2(texC, texR) + halfCR) /
|
|
vec2(${f}.0, ${h}.0);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`;if(f===o&&d==null)return`
|
|
float ${s}(int row, int col, int depth,
|
|
int depth2, int depth3, int depth4) {
|
|
float texR = dot(vec4(row, col, depth, depth2),
|
|
vec4(${t[1]*t[2]*t[3]*t[4]},
|
|
${t[2]*t[3]*t[4]},
|
|
${t[3]*t[4]},
|
|
${t[4]})) + float(depth3);
|
|
int texC = depth4;
|
|
vec2 uv = (vec2(texC, texR) + halfCR) /
|
|
vec2(${f}.0, ${h}.0);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`;let m=Cl(n);return`
|
|
float ${s}(int row, int col, int depth,
|
|
int depth2, int depth3, int depth4) {
|
|
// Explicitly use integer operations as dot() only works on floats.
|
|
int index = row * ${u} + col * ${c} + depth * ${l} +
|
|
depth2 * ${i} + depth3 * ${o} + depth4 + ${m};
|
|
vec2 uv = uvFromFlat(${h}, ${f}, index);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`}function rc(e){let t=e.name,n=v.sizeFromShape(e.shapeInfo.logicalShape);return n<2?`return ${t};`:`
|
|
for (int i = 0; i < ${n}; i++) {
|
|
if (i == index) {
|
|
return ${t}[i];
|
|
}
|
|
}
|
|
`}function FY(e,t){let n=e.name,s=n.charAt(0).toUpperCase()+n.slice(1),r="get"+s+"AtOutCoords",a=e.shapeInfo.logicalShape.length,o=t.logicalShape.length,i=wI(e.shapeInfo.logicalShape,t.logicalShape),l=yt(o),c=o-a,u,d=["x","y","z","w","u","v"];a===0?u="":o<2&&i.length>=1?u="coords = 0;":u=i.map(x=>`coords.${d[x+c]} = 0;`).join(`
|
|
`);let p="";o<2&&a>0?p="coords":p=e.shapeInfo.logicalShape.map((x,y)=>`coords.${d[y+c]}`).join(", ");let h="return outputValue;",m=v.sizeFromShape(e.shapeInfo.logicalShape)===1,A=v.sizeFromShape(t.logicalShape)===1;if(a===1&&!m&&!A)h=`
|
|
return vec4(outputValue.xy, outputValue.xy);
|
|
`;else if(m&&!A)o===1?h=`
|
|
return vec4(outputValue.x, outputValue.x, 0., 0.);
|
|
`:h=`
|
|
return vec4(outputValue.x);
|
|
`;else if(i.length){let x=a-2,y=a-1;i.indexOf(x)>-1&&i.indexOf(y)>-1?h="return vec4(outputValue.x);":i.indexOf(x)>-1?h="return vec4(outputValue.x, outputValue.y, outputValue.x, outputValue.y);":i.indexOf(y)>-1&&(h="return vec4(outputValue.xx, outputValue.zz);")}return`
|
|
vec4 ${r}() {
|
|
${l} coords = getOutputCoords();
|
|
${u}
|
|
vec4 outputValue = get${s}(${p});
|
|
${h}
|
|
}
|
|
`}function OY(e,t){let n=e.name,s=n.charAt(0).toUpperCase()+n.slice(1),r="get"+s+"AtOutCoords",a=t.texShape,o=e.shapeInfo.texShape,i=e.shapeInfo.logicalShape.length,l=t.logicalShape.length;if(!e.shapeInfo.isUniform&&i===l&&e.shapeInfo.flatOffset==null&&v.arraysEqual(o,a))return`
|
|
float ${r}() {
|
|
return sampleTexture(${n}, resultUV);
|
|
}
|
|
`;let c=yt(l),u=wI(e.shapeInfo.logicalShape,t.logicalShape),d=l-i,p,h=["x","y","z","w","u","v"];i===0?p="":l<2&&u.length>=1?p="coords = 0;":p=u.map(m=>`coords.${h[m+d]} = 0;`).join(`
|
|
`);let f="";return l<2&&i>0?f="coords":f=e.shapeInfo.logicalShape.map((m,g)=>`coords.${h[g+d]}`).join(", "),`
|
|
float ${r}() {
|
|
${c} coords = getOutputCoords();
|
|
${p}
|
|
return get${s}(${f});
|
|
}
|
|
`}function yt(e){if(e<=1)return"int";if(e===2)return"ivec2";if(e===3)return"ivec3";if(e===4)return"ivec4";if(e===5)return"ivec5";if(e===6)return"ivec6";throw Error(`GPU for rank ${e} is not yet supported`)}function tx(e,t,n){let{newShape:s,keptDims:r}=v.squeezeShape(t),a=t.length,o=e&&a===3&&t[0]===1,i=o?t.slice(1):s,l=!e&&a>1&&!v.arraysEqual(t,n)&&s.length<a||o;return{useSqueezeShape:l,uniformShape:l?i:t,keptDims:r}}function ac(e,t){let n=JSON.parse(JSON.stringify(e));return n.shapeInfo.logicalShape=t,n}function oc(e,t){return t.map(n=>e[n]).join(", ")}function MY(e,t,n,s){let r=n.map((b,w)=>{let k={logicalShape:b.shape,texShape:b.isUniform?null:b.texData.texShape,isUniform:b.isUniform,isPacked:b.isUniform?!1:b.texData.isPacked,flatOffset:null};return b.texData!=null&&b.texData.slice!=null&&b.texData.slice.flatOffset>0&&(k.flatOffset=b.texData.slice.flatOffset),{name:t.variableNames[w],shapeInfo:k}}),a=r.map(b=>b.shapeInfo),o={logicalShape:s.shape,texShape:s.texData.texShape,isUniform:!1,isPacked:s.texData.isPacked,flatOffset:null},i=tY(r,o,t),l=eI(e.gl,i),c=e.createProgram(l),u=null,d=e.getUniformLocation(c,"NAN",!1);K().getNumber("WEBGL_VERSION")===1&&(u=e.getUniformLocation(c,"INFINITY",!1));let p=!1,h={},f={},m={};for(let b=0;b<t.variableNames.length;b++){let w=t.variableNames[b];h[w]=e.getUniformLocation(c,w,p),h[`offset${w}`]=e.getUniformLocation(c,`offset${w}`,p),t.enableShapeUniforms&&(f[`${w}Shape`]=e.getUniformLocation(c,`${w}Shape`,p),m[`${w}TexShape`]=e.getUniformLocation(c,`${w}TexShape`,p))}let g,A,x;t.enableShapeUniforms&&(g=e.getUniformLocation(c,"outShape",p),x=e.getUniformLocation(c,"outShapeStrides",p),A=e.getUniformLocation(c,"outTexShape",p));let y=[];return t.customUniforms&&t.customUniforms.forEach((b,w)=>{y[w]=e.getUniformLocation(c,b.name,p)}),{program:t,fragmentShader:l,source:i,webGLProgram:c,uniformLocations:h,customUniformLocations:y,inShapeInfos:a,outShapeInfo:o,infLoc:u,nanLoc:d,inShapesLocations:f,inTexShapesLocations:m,outShapeLocation:g,outShapeStridesLocation:x,outTexShapeLocation:A}}function II(e,t){if(e.length!==t.length)throw Error(`Binary was compiled with ${e.length} inputs, but was executed with ${t.length} inputs`);e.forEach((n,s)=>{let r=n.logicalShape,a=t[s],o=a.shape;if(!v.arraysEqual(r,o))throw Error(`Binary was compiled with different shapes than the current args. Shapes ${r} and ${o} must match`);if(n.isUniform&&a.isUniform)return;let i=n.texShape,l=a.isUniform?null:a.texData.texShape;if(!v.arraysEqual(i,l))throw Error(`Binary was compiled with different texture shapes than the current args. Shape ${i} and ${l} must match`)})}function zY(e,t,n,s,r){t.program.enableShapeUniforms||(II(t.inShapeInfos,n),II([t.outShapeInfo],[s]));let a=s.texData.texture,o=s.texData.texShape;s.texData.isPacked?e.setOutputPackedMatrixTexture(a,o[0],o[1]):e.setOutputMatrixTexture(a,o[0],o[1]),e.setProgram(t.webGLProgram),K().getNumber("WEBGL_VERSION")===1&&t.infLoc!==null&&e.gl.uniform1f(t.infLoc,1/0),t.nanLoc!==null&&e.gl.uniform1f(t.nanLoc,NaN),n.forEach((l,c)=>{let u=t.program.variableNames[c],d=t.uniformLocations[u],p=t.uniformLocations[`offset${u}`],h=t.inShapesLocations[`${u}Shape`],f=t.inTexShapesLocations[`${u}TexShape`];if(h){let{uniformShape:m}=tx(t.program.packedInputs,l.shape,l.texData.texShape);switch(m.length){case 1:e.gl.uniform1iv(h,new Int32Array(m));break;case 2:e.gl.uniform2iv(h,new Int32Array(m));break;case 3:e.gl.uniform3iv(h,new Int32Array(m));break;case 4:e.gl.uniform4iv(h,new Int32Array(m));break;default:break}}if(f&&e.gl.uniform2i(f,l.texData.texShape[0],l.texData.texShape[1]),d!=null){if(l.isUniform){if(v.sizeFromShape(l.shape)<2)e.gl.uniform1f(d,l.uniformValues[0]);else{let m=l.uniformValues;m instanceof Float32Array||(m=new Float32Array(m)),e.gl.uniform1fv(d,m)}return}l.texData.slice!=null&&p!=null&&e.gl.uniform1i(p,l.texData.slice.flatOffset),e.setInputMatrixTexture(l.texData.texture,d,c)}});let i=t.outShapeLocation;if(i)switch(s.shape.length){case 1:e.gl.uniform1iv(i,new Int32Array(s.shape));break;case 2:e.gl.uniform2iv(i,new Int32Array(s.shape));break;case 3:e.gl.uniform3iv(i,new Int32Array(s.shape));break;case 4:e.gl.uniform4iv(i,new Int32Array(s.shape));break;default:break}if(t.outShapeStridesLocation){let l=v.computeStrides(s.shape);switch(s.shape.length){case 2:e.gl.uniform1iv(t.outShapeStridesLocation,new Int32Array(l));break;case 3:e.gl.uniform2iv(t.outShapeStridesLocation,new Int32Array(l));break;case 4:e.gl.uniform3iv(t.outShapeStridesLocation,new Int32Array(l));break;default:break}}t.outTexShapeLocation&&e.gl.uniform2i(t.outTexShapeLocation,s.texData.texShape[0],s.texData.texShape[1]),t.program.customUniforms&&r&&t.program.customUniforms.forEach((l,c)=>{let u=t.customUniformLocations[c],d=r[c];if(l.type==="float")e.gl.uniform1fv(u,d);else if(l.type==="vec2")e.gl.uniform2fv(u,d);else if(l.type==="vec3")e.gl.uniform3fv(u,d);else if(l.type==="vec4")e.gl.uniform4fv(u,d);else if(l.type==="int")e.gl.uniform1iv(u,d);else if(l.type==="ivec2")e.gl.uniform2iv(u,d);else if(l.type==="ivec3")e.gl.uniform3iv(u,d);else if(l.type==="ivec4")e.gl.uniform4iv(u,d);else throw Error(`uniform type ${l.type} is not supported yet.`)}),e.executeProgram()}function LY(e,t,n){let s="";t.concat(n).forEach(o=>{let i=o.texData!=null&&o.texData.slice!=null&&o.texData.slice.flatOffset>0;if(e.enableShapeUniforms&&!o.isUniform){let l=o.texData.texShape,{useSqueezeShape:c,uniformShape:u,keptDims:d}=tx(e.packedInputs,o.shape,l),p="",h="",f="";if(u.length===1&&e.packedInputs){let w=[Math.ceil(l[0]/2),Math.ceil(l[1]/2)];p=`${w[0]>1}_${w[1]>1}`}else if(u.length===2&&!e.packedInputs)h=`${u[0]>1}_${u[1]>1}`;else if(u.length>2&&!e.packedInputs){let w=v.computeStrides(u);f=`${w[0]===l[1]}_${w[w.length-1]===l[1]}`}let m=o.shape.length,g=u.length===2&&v.arraysEqual(o.shape,l),A=v.sizeFromShape(o.shape)===1,x=E.getBroadcastDims(o.shape,n.shape),y=!e.packedInputs&&m===n.shape.length&&v.arraysEqual(l,n.texData.texShape),b=e.packedInputs||u.length>2?"":`${l[0]>1}_${l[1]>1}`;s+=`${m}_${y}_${c?d:""}_${u.length}_${A}_${x}_${g}_${p}_${h}_${f}_${b}_${i}`}else{let l=o.isUniform?"uniform":o.texData.texShape;s+=`${o.shape}_${l}_${i}`}});let r=e.userCode,a=e.constructor.name;return a+="_"+s+"_"+r+`${K().getNumber("WEBGL_VERSION")}`,a}function Fs(e){return K().getBool("WEBGL_USE_SHAPES_UNIFORMS")&&e<=4}var BY=class{constructor(e){this.variableNames=["A"],this.packedInputs=!1,this.packedOutput=!0,this.outPackingScheme=op.DENSE,this.customUniforms=[{name:"texShape",type:"ivec2"}];let t=Un();this.outputShape=e,this.enableShapeUniforms=Fs(this.outputShape.length),this.userCode=`
|
|
ivec3 outCoordsFromFlatIndex(int index) {
|
|
${this.enableShapeUniforms?Pm(["r","c","d"],e):Il(["r","c","d"],e)}
|
|
return ivec3(r, c, d);
|
|
}
|
|
|
|
void main() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx * vec2(texShape[0], texShape[1]));
|
|
int index = 4 * (resTexRC.x * texShape[1] + resTexRC.y);
|
|
|
|
vec4 result = vec4(0.);
|
|
|
|
for (int i=0; i<4; i++) {
|
|
int flatIndex = index + i;
|
|
ivec3 rc = outCoordsFromFlatIndex(flatIndex);
|
|
result[i] = getA(rc.x, rc.y, rc.z);
|
|
}
|
|
|
|
${t.output} = result;
|
|
}
|
|
`}},WY=class{constructor(e){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outPackingScheme=op.DENSE,this.customUniforms=[{name:"texShape",type:"ivec2"}];let t=Un();this.outputShape=e,this.enableShapeUniforms=Fs(this.outputShape.length),this.userCode=`
|
|
ivec3 outCoordsFromFlatIndex(int index) {
|
|
${this.enableShapeUniforms?Pm(["r","c","d"],e):Il(["r","c","d"],e)}
|
|
return ivec3(r, c, d);
|
|
}
|
|
|
|
void main() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx * vec2(texShape[0], texShape[1]));
|
|
int index = 4 * (resTexRC.x * texShape[1] + resTexRC.y);
|
|
|
|
vec4 result = vec4(0.);
|
|
|
|
for (int i=0; i<4; i++) {
|
|
int flatIndex = index + i;
|
|
ivec3 rc = outCoordsFromFlatIndex(flatIndex);
|
|
result[i] = getChannel(getA(rc.x, rc.y, rc.z), vec2(rc.y, rc.z));
|
|
}
|
|
|
|
${t.output} = result;
|
|
}
|
|
`}},VY=class{constructor(e){this.variableNames=["A"],this.outTexUsage=Ds.DOWNLOAD;let t=Un();this.outputShape=e,this.userCode=`
|
|
${vI}
|
|
|
|
void main() {
|
|
float x = getAAtOutCoords();
|
|
${t.output} = encode_float(x);
|
|
}
|
|
`}},UY=class{constructor(e){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outTexUsage=Ds.DOWNLOAD;let t=Un();this.outputShape=e,this.userCode=`
|
|
${vI}
|
|
|
|
void main() {
|
|
ivec3 coords = getOutputCoords();
|
|
float x = getChannel(getAAtOutCoords(), vec2(coords.y, coords.z));
|
|
${t.output} = encode_float(x);
|
|
}
|
|
`}},GY=class{constructor(e,t=!1){this.variableNames=["A"],this.customUniforms=[{name:"texShape",type:"ivec2"}];let n=Un();this.outputShape=e,this.enableShapeUniforms=Fs(this.outputShape.length);let s="result";t&&(s="floor(result * 255. + 0.5)"),this.userCode=`
|
|
${this.enableShapeUniforms?ex():Qy(e)}
|
|
|
|
void main() {
|
|
ivec3 coords = getOutputCoords();
|
|
|
|
int flatIndex = getFlatIndex(coords);
|
|
int offset = imod(flatIndex, 4);
|
|
|
|
flatIndex = idiv(flatIndex, 4, 1.);
|
|
|
|
int r = flatIndex / texShape[1];
|
|
int c = imod(flatIndex, texShape[1]);
|
|
vec2 uv = (vec2(c, r) + halfCR) / vec2(texShape[1], texShape[0]);
|
|
vec4 values = ${n.texture2D}(A, uv);
|
|
|
|
float result;
|
|
|
|
if(offset == 0) {
|
|
result = values[0];
|
|
} else if(offset == 1) {
|
|
result = values[1];
|
|
} else if(offset == 2) {
|
|
result = values[2];
|
|
} else {
|
|
result = values[3];
|
|
}
|
|
|
|
${n.output} = vec4(${s}, 0., 0., 0.);
|
|
}
|
|
`}},HY=class{constructor(e,t=!1){this.variableNames=["A"],this.packedInputs=!1,this.packedOutput=!0,this.customUniforms=[{name:"texShape",type:"ivec2"}];let n=Un();this.outputShape=e,this.enableShapeUniforms=Fs(this.outputShape.length);let s="",r="result";t&&(r="floor(result * 255. + 0.5)");for(let a=0;a<=1;a++)for(let o=0;o<=1;o++){let i=a*2+o;s+=`
|
|
localCoords = coords;
|
|
if(localCoords[2] + ${o} < ${this.enableShapeUniforms?"outShape[2]":`${e[2]}`}) {
|
|
localCoords[2] += ${o};
|
|
if (localCoords[1] + ${a} < ${this.enableShapeUniforms?"outShape[1]":`${e[1]}`}) {
|
|
localCoords[1] += ${a};
|
|
|
|
flatIndex = getFlatIndex(localCoords);
|
|
offset = imod(flatIndex, 4);
|
|
|
|
flatIndex = idiv(flatIndex, 4, 1.);
|
|
|
|
int r = flatIndex / texShape[1];
|
|
int c = imod(flatIndex, texShape[1]);
|
|
vec2 uv = (vec2(c, r) + halfCR) / vec2(texShape[1], texShape[0]);
|
|
values = ${n.texture2D}(A, uv);
|
|
|
|
if (offset == 0) {
|
|
result[${i}] = values[0];
|
|
} else if (offset == 1) {
|
|
result[${i}] = values[1];
|
|
} else if (offset == 2) {
|
|
result[${i}] = values[2];
|
|
} else {
|
|
result[${i}] = values[3];
|
|
}
|
|
}
|
|
}
|
|
`}this.userCode=`
|
|
${this.enableShapeUniforms?ex():Qy(e)}
|
|
|
|
void main() {
|
|
ivec3 coords = getOutputCoords();
|
|
|
|
vec4 result = vec4(0.);
|
|
int flatIndex, r, c, offset;
|
|
ivec3 localCoords;
|
|
vec2 uv;
|
|
vec4 values;
|
|
|
|
${s}
|
|
|
|
${n.output} = ${r};
|
|
}
|
|
`}},CI={};Oe(CI,{bindVertexProgramAttributeStreams:()=>FI,createBufferFromOutputTexture:()=>zI,createFloat16MatrixTexture:()=>$I,createFloat16PackedMatrixTexture:()=>PI,createFloat32MatrixTexture:()=>RI,createIndexBuffer:()=>EI,createPackedMatrixTexture:()=>DI,createUnsignedBytesMatrixTexture:()=>_I,createVertexBuffer:()=>NI,createVertexShader:()=>TI,downloadByteEncodedFloatMatrixFromOutputTexture:()=>BI,downloadFloat32MatrixFromBuffer:()=>LI,downloadMatrixFromPackedOutputTexture:()=>VI,downloadPackedMatrixFromBuffer:()=>WI,getInternalFormatForFloat16MatrixTexture:()=>sx,getInternalFormatForFloat16PackedMatrixTexture:()=>ox,getInternalFormatForFloat32MatrixTexture:()=>nx,getInternalFormatForPackedMatrixTexture:()=>ax,getInternalFormatForUnsignedBytesMatrixTexture:()=>rx,uploadDenseMatrixToTexture:()=>OI,uploadPixelDataToTexture:()=>MI});function TI(e){let t=Un(),n=`${t.version}
|
|
precision highp float;
|
|
${t.attribute} vec3 clipSpacePos;
|
|
${t.attribute} vec2 uv;
|
|
${t.varyingVs} vec2 resultUV;
|
|
|
|
void main() {
|
|
gl_Position = vec4(clipSpacePos, 1);
|
|
resultUV = uv;
|
|
}`;return QS(e,n)}function NI(e){let t=new Float32Array([-1,1,0,0,1,-1,-1,0,0,0,1,1,0,1,1,1,-1,0,1,0]);return sI(e,t)}function EI(e){let t=new Uint16Array([0,1,2,2,1,3]);return rI(e,t)}function dp(e,t,n,s,r,a){oI(t,n);let o=aI(e),i=e.TEXTURE_2D;return Ie(e,()=>e.bindTexture(i,o)),Ie(e,()=>e.texParameteri(i,e.TEXTURE_WRAP_S,e.CLAMP_TO_EDGE)),Ie(e,()=>e.texParameteri(i,e.TEXTURE_WRAP_T,e.CLAMP_TO_EDGE)),Ie(e,()=>e.texParameteri(i,e.TEXTURE_MIN_FILTER,e.NEAREST)),Ie(e,()=>e.texParameteri(i,e.TEXTURE_MAG_FILTER,e.NEAREST)),Ie(e,()=>e.texImage2D(i,0,s,t,n,0,r,a,null)),Ie(e,()=>e.bindTexture(e.TEXTURE_2D,null)),o}function nx(e){return e.internalFormatFloat}function RI(e,t,n,s){let[r,a]=ip(t,n);return dp(e,r,a,nx(s),s.textureFormatFloat,e.FLOAT)}function sx(e){return e.internalFormatHalfFloat}function $I(e,t,n,s){let[r,a]=ip(t,n);return dp(e,r,a,sx(s),s.textureFormatFloat,s.textureTypeHalfFloat)}function rx(e){return e.downloadTextureFormat}function _I(e,t,n,s){let[r,a]=ip(t,n);return dp(e,r,a,rx(s),e.RGBA,e.UNSIGNED_BYTE)}function ax(e){return e.internalFormatPackedFloat}function DI(e,t,n,s){let[r,a]=tc(t,n);return dp(e,r,a,ax(s),e.RGBA,e.FLOAT)}function ox(e){return e.internalFormatPackedHalfFloat}function PI(e,t,n,s){let[r,a]=tc(t,n);return dp(e,r,a,ox(s),e.RGBA,s.textureTypeHalfFloat)}function FI(e,t,n){let s=0,r=3*4,a=3*4+2*4;return Ie(e,()=>e.bindBuffer(e.ARRAY_BUFFER,n)),Ky(e,t,"clipSpacePos",n,3,a,s)&&Ky(e,t,"uv",n,2,a,r)}function OI(e,t,n,s,r,a){Ie(e,()=>e.bindTexture(e.TEXTURE_2D,t));let o,i,l;r instanceof Uint8Array?(o=new Uint8Array(n*s*4),i=e.UNSIGNED_BYTE,l=e.RGBA):(o=new Float32Array(n*s*4),i=e.FLOAT,l=a.internalFormatPackedFloat),o.set(r),Ie(e,()=>e.texImage2D(e.TEXTURE_2D,0,l,n,s,0,e.RGBA,i,o)),Ie(e,()=>e.bindTexture(e.TEXTURE_2D,null))}function MI(e,t,n){Ie(e,()=>e.bindTexture(e.TEXTURE_2D,t)),n.data instanceof Uint8Array?Ie(e,()=>e.texImage2D(e.TEXTURE_2D,0,e.RGBA,n.width,n.height,0,e.RGBA,e.UNSIGNED_BYTE,n.data)):Ie(e,()=>e.texImage2D(e.TEXTURE_2D,0,e.RGBA,e.RGBA,e.UNSIGNED_BYTE,n)),Ie(e,()=>e.bindTexture(e.TEXTURE_2D,null))}function zI(e,t,n,s){let r=e.createBuffer();Ie(e,()=>e.bindBuffer(e.PIXEL_PACK_BUFFER,r));let i=4*4*t*n;return Ie(e,()=>e.bufferData(e.PIXEL_PACK_BUFFER,i,e.STREAM_READ)),Ie(e,()=>e.readPixels(0,0,n,t,e.RGBA,e.FLOAT,0)),Ie(e,()=>e.bindBuffer(e.PIXEL_PACK_BUFFER,null)),r}function LI(e,t,n){let s=e,r=new Float32Array(n);return s.bindBuffer(s.PIXEL_PACK_BUFFER,t),s.getBufferSubData(s.PIXEL_PACK_BUFFER,0,r),s.bindBuffer(s.PIXEL_PACK_BUFFER,null),r}function BI(e,t,n,s){let[r,a]=ip(t,n),o=4,i=new Uint8Array(BZ(t*n,o));return Ie(e,()=>e.readPixels(0,0,r,a,s.downloadTextureFormat,e.UNSIGNED_BYTE,i)),new Float32Array(i.buffer)}function WI(e,t,n,s,r,a,o,i){let l=e,c=new Float32Array(WZ(a,o));return l.bindBuffer(l.PIXEL_PACK_BUFFER,t),l.getBufferSubData(l.PIXEL_PACK_BUFFER,0,c),l.bindBuffer(l.PIXEL_PACK_BUFFER,null),c}function VI(e,t,n){let s=new Float32Array(t*n*4);return Ie(e,()=>e.readPixels(0,0,n,t,e.RGBA,e.FLOAT,s)),s}var Fm=class{constructor(e){this.outputTexture=null,this.program=null,this.disposed=!1,this.vertexAttrsAreBound=!1,this.itemsToPoll=[];let t=K().getNumber("WEBGL_VERSION");e!=null?(this.gl=e,Cm(t,e)):this.gl=zr(t);let n="WEBGL_color_buffer_float",s="EXT_color_buffer_half_float";if(K().getNumber("WEBGL_VERSION")===1){let r="OES_texture_float",a="OES_texture_half_float";if(this.textureFloatExtension=lp(this.gl,r),Ps(this.gl,a))this.textureHalfFloatExtension=lp(this.gl,a);else if(K().get("WEBGL_FORCE_F16_TEXTURES"))throw new Error("GL context does not support half float textures, yet the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.");if(this.colorBufferFloatExtension=this.gl.getExtension(n),Ps(this.gl,s))this.colorBufferHalfFloatExtension=lp(this.gl,s);else if(K().get("WEBGL_FORCE_F16_TEXTURES"))throw new Error("GL context does not support color renderable half floats, yet the environment flag WEBGL_FORCE_F16_TEXTURES is set to true.")}else if(n="EXT_color_buffer_float",Ps(this.gl,n))this.colorBufferFloatExtension=this.gl.getExtension(n);else if(Ps(this.gl,s))this.colorBufferHalfFloatExtension=this.gl.getExtension(s);else throw new Error("GL context does not support color renderable floats");this.vertexBuffer=NI(this.gl),this.indexBuffer=EI(this.gl),this.framebuffer=iI(this.gl),this.textureConfig=Xy(this.gl,this.textureHalfFloatExtension)}get debug(){return K().getBool("DEBUG")}dispose(){if(this.disposed)return;this.program!=null&&console.warn("Disposing a GPGPUContext that still has a bound WebGLProgram. This is probably a resource leak, delete the program with GPGPUContext.deleteProgram before disposing."),this.outputTexture!=null&&console.warn("Disposing a GPGPUContext that still has a bound output matrix texture. This is probably a resource leak, delete the output matrix texture with GPGPUContext.deleteMatrixTexture before disposing.");let e=this.gl;Ie(e,()=>e.finish()),Ie(e,()=>e.bindFramebuffer(e.FRAMEBUFFER,null)),Ie(e,()=>e.deleteFramebuffer(this.framebuffer)),Ie(e,()=>e.bindBuffer(e.ARRAY_BUFFER,null)),Ie(e,()=>e.bindBuffer(e.ELEMENT_ARRAY_BUFFER,null)),Ie(e,()=>e.deleteBuffer(this.indexBuffer)),this.disposed=!0}createFloat32MatrixTexture(e,t){return this.throwIfDisposed(),RI(this.gl,e,t,this.textureConfig)}createFloat16MatrixTexture(e,t){return this.throwIfDisposed(),$I(this.gl,e,t,this.textureConfig)}createUnsignedBytesMatrixTexture(e,t){return this.throwIfDisposed(),_I(this.gl,e,t,this.textureConfig)}uploadPixelDataToTexture(e,t){this.throwIfDisposed(),MI(this.gl,e,t)}uploadDenseMatrixToTexture(e,t,n,s){this.throwIfDisposed(),OI(this.gl,e,t,n,s,this.textureConfig)}createFloat16PackedMatrixTexture(e,t){return this.throwIfDisposed(),PI(this.gl,e,t,this.textureConfig)}createPackedMatrixTexture(e,t){return this.throwIfDisposed(),DI(this.gl,e,t,this.textureConfig)}deleteMatrixTexture(e){this.throwIfDisposed(),this.outputTexture===e&&(Zy(this.gl,this.framebuffer),this.outputTexture=null),Ie(this.gl,()=>this.gl.deleteTexture(e))}downloadByteEncodedFloatMatrixFromOutputTexture(e,t,n){return this.downloadMatrixDriver(e,()=>BI(this.gl,t,n,this.textureConfig))}downloadPackedMatrixFromBuffer(e,t,n,s,r,a){return WI(this.gl,e,t,n,s,r,a,this.textureConfig)}downloadFloat32MatrixFromBuffer(e,t){return LI(this.gl,e,t)}createBufferFromTexture(e,t,n){this.bindTextureToFrameBuffer(e);let s=zI(this.gl,t,n,this.textureConfig);return this.unbindTextureToFrameBuffer(),s}createAndWaitForFence(){let e=this.createFence(this.gl);return this.pollFence(e)}createFence(e){let t,n;if(K().getBool("WEBGL_FENCE_API_ENABLED")){let s=e,r=s.fenceSync(s.SYNC_GPU_COMMANDS_COMPLETE,0);e.flush(),n=()=>{let a=s.clientWaitSync(r,0,0);return a===s.ALREADY_SIGNALED||a===s.CONDITION_SATISFIED},t=r}else K().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0?(t=this.beginQuery(),this.endQuery(),n=()=>this.isQueryAvailable(t,K().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))):n=()=>!0;return{query:t,isFencePassed:n}}downloadMatrixFromPackedTexture(e,t,n){return this.downloadMatrixDriver(e,()=>VI(this.gl,t,n))}createProgram(e){this.throwIfDisposed();let t=this.gl;this.vertexShader==null&&(this.vertexShader=TI(t));let n=tI(t);return Ie(t,()=>t.attachShader(n,this.vertexShader)),Ie(t,()=>t.attachShader(n,e)),nI(t,n),this.debug&&Nm(t,n),this.vertexAttrsAreBound||(this.setProgram(n),this.vertexAttrsAreBound=FI(t,this.program,this.vertexBuffer)),n}deleteProgram(e){this.throwIfDisposed(),e===this.program&&(this.program=null),e!=null&&Ie(this.gl,()=>this.gl.deleteProgram(e))}setProgram(e){this.throwIfDisposed(),this.program=e,this.program!=null&&this.debug&&Nm(this.gl,this.program),Ie(this.gl,()=>this.gl.useProgram(e))}getUniformLocation(e,t,n=!0){return this.throwIfDisposed(),n?uI(this.gl,e,t):cI(this.gl,e,t)}getAttributeLocation(e,t){return this.throwIfDisposed(),Ie(this.gl,()=>this.gl.getAttribLocation(e,t))}getUniformLocationNoThrow(e,t){return this.throwIfDisposed(),this.gl.getUniformLocation(e,t)}setInputMatrixTexture(e,t,n){this.throwIfDisposed(),this.throwIfNoProgram(),dI(this.gl,e,t,n)}setOutputMatrixTexture(e,t,n){this.setOutputMatrixTextureDriver(e,n,t)}setOutputPackedMatrixTexture(e,t,n){this.throwIfDisposed();let[s,r]=tc(t,n);this.setOutputMatrixTextureDriver(e,s,r)}setOutputMatrixWriteRegion(e,t,n,s){this.setOutputMatrixWriteRegionDriver(n,e,s,t)}setOutputPackedMatrixWriteRegion(e,t,n,s){throw new Error("setOutputPackedMatrixWriteRegion not implemented.")}debugValidate(){this.program!=null&&Nm(this.gl,this.program),up(this.gl)}executeProgram(){this.throwIfDisposed(),this.throwIfNoProgram();let e=this.gl;this.debug&&this.debugValidate(),Ie(e,()=>e.drawElements(e.TRIANGLES,6,e.UNSIGNED_SHORT,0))}blockUntilAllProgramsCompleted(){this.throwIfDisposed(),Ie(this.gl,()=>this.gl.finish())}getQueryTimerExtension(){return this.disjointQueryTimerExtension==null&&(this.disjointQueryTimerExtension=lp(this.gl,K().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2?"EXT_disjoint_timer_query_webgl2":"EXT_disjoint_timer_query")),this.disjointQueryTimerExtension}getQueryTimerExtensionWebGL2(){return this.getQueryTimerExtension()}getQueryTimerExtensionWebGL1(){return this.getQueryTimerExtension()}beginQuery(){if(K().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){let n=this.gl,s=this.getQueryTimerExtensionWebGL2(),r=n.createQuery();return n.beginQuery(s.TIME_ELAPSED_EXT,r),r}let e=this.getQueryTimerExtensionWebGL1(),t=e.createQueryEXT();return e.beginQueryEXT(e.TIME_ELAPSED_EXT,t),t}endQuery(){if(K().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){let t=this.gl,n=this.getQueryTimerExtensionWebGL2();t.endQuery(n.TIME_ELAPSED_EXT);return}let e=this.getQueryTimerExtensionWebGL1();e.endQueryEXT(e.TIME_ELAPSED_EXT)}async waitForQueryAndGetTime(e){return await v.repeatedTry(()=>this.disposed||this.isQueryAvailable(e,K().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))),this.getQueryTime(e,K().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))}getQueryTime(e,t){if(t===0)return null;if(t===2){let n=this.gl;return n.getQueryParameter(e,n.QUERY_RESULT)/1e6}else{let n=this.getQueryTimerExtensionWebGL1();return n.getQueryObjectEXT(e,n.QUERY_RESULT_EXT)/1e6}}isQueryAvailable(e,t){if(t===0)return!0;if(t===2){let n=this.gl,s=this.getQueryTimerExtensionWebGL2(),r=n.getQueryParameter(e,n.QUERY_RESULT_AVAILABLE);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(s.GPU_DISJOINT_EXT)),r&&!this.disjoint}else{let n=this.getQueryTimerExtensionWebGL1(),s=n.getQueryObjectEXT(e,n.QUERY_RESULT_AVAILABLE_EXT);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(n.GPU_DISJOINT_EXT)),s&&!this.disjoint}}pollFence(e){return new Promise(t=>{this.addItemToPoll(()=>e.isFencePassed(),()=>t())})}pollItems(){let e=jY(this.itemsToPoll.map(t=>t.isDoneFn));for(let t=0;t<=e;++t){let{resolveFn:n}=this.itemsToPoll[t];n()}this.itemsToPoll=this.itemsToPoll.slice(e+1)}addItemToPoll(e,t){this.itemsToPoll.push({isDoneFn:e,resolveFn:t}),!(this.itemsToPoll.length>1)&&v.repeatedTry(()=>(this.pollItems(),this.itemsToPoll.length===0))}bindTextureToFrameBuffer(e){this.throwIfDisposed(),Em(this.gl,e,this.framebuffer),this.debug&&up(this.gl)}unbindTextureToFrameBuffer(){this.outputTexture!=null?(Em(this.gl,this.outputTexture,this.framebuffer),this.debug&&up(this.gl)):Zy(this.gl,this.framebuffer)}downloadMatrixDriver(e,t){this.bindTextureToFrameBuffer(e);let n=t();return this.unbindTextureToFrameBuffer(),n}setOutputMatrixTextureDriver(e,t,n){this.throwIfDisposed();let s=this.gl;Em(s,e,this.framebuffer),this.debug&&up(s),this.outputTexture=e,Ie(s,()=>s.viewport(0,0,t,n)),Ie(s,()=>s.scissor(0,0,t,n))}setOutputMatrixWriteRegionDriver(e,t,n,s){this.throwIfDisposed(),Ie(this.gl,()=>this.gl.scissor(e,t,n,s))}throwIfDisposed(){if(this.disposed)throw new Error("Attempted to use disposed GPGPUContext.")}throwIfNoProgram(){if(this.program==null)throw new Error("No GPU program is currently set.")}};function jY(e){let t=0;for(;t<e.length&&e[t]();++t);return t-1}var{addImpl:qY,bincountImpl:UI,bincountReduceImpl:XY,ceilImpl:KY,concatImpl:ZY,equalImpl:YY,expImpl:JY,expm1Impl:QY,floorImpl:eJ,gatherNdImpl:tJ,gatherV2Impl:nJ,greaterImpl:sJ,greaterEqualImpl:rJ,lessImpl:aJ,lessEqualImpl:oJ,linSpaceImpl:iJ,logImpl:lJ,maxImpl:uJ,maximumImpl:cJ,minimumImpl:dJ,multiplyImpl:pJ,negImpl:hJ,notEqualImpl:fJ,prodImpl:mJ,rangeImpl:gJ,rsqrtImpl:AJ,sigmoidImpl:yJ,simpleAbsImpl:GI,sliceImpl:xJ,sparseFillEmptyRowsImpl:bJ,sparseReshapeImpl:vJ,sparseSegmentReductionImpl:HI,sqrtImpl:wJ,stridedSliceImpl:kJ,stringNGramsImpl:SJ,stringSplitImpl:IJ,stringToHashBucketFastImpl:CJ,subImpl:TJ,tileImpl:NJ,topKImpl:EJ,transposeImpl:ix,uniqueImpl:RJ}=bm;function jI(e,t){return["x","y","z","w","u","v"].slice(0,t).map(n=>`${e}.${n}`)}function Gn(e,t){return t===1?[e]:jI(e,t)}function $J(e,t){if(e===1)return"rc";let n="";for(let s=0;s<e;s++)n+=t[s],s<e-1&&(n+=",");return n}var _J=class{constructor(e){this.variableNames=["A"],this.packedInputs=!1,this.packedOutput=!0,this.outputShape=e;let t=e.length;if(t===0)this.userCode=`
|
|
void main() {
|
|
setOutput(vec4(getA(), 0., 0., 0.));
|
|
}
|
|
`;else{let n=Gn("rc",t),s=yt(t),r=PJ(t,e,n),a=FJ(t,e[e.length-1],e[e.length-2],n),o=OJ(e,n);this.userCode=`
|
|
void main() {
|
|
${s} rc = getOutputCoords();
|
|
|
|
if(${r}) {
|
|
setOutput(vec4(0));
|
|
} else {
|
|
${a}
|
|
|
|
setOutput(vec4(${o}));
|
|
}
|
|
}
|
|
`}}};function DJ(e,t){let n=[];for(let s=0;s<=1;s++)for(let r=0;r<=1;r++){let a=`${s===0?"r":"rp1"}, ${r===0?"c":"cp1"}`;for(let o=2;o<e;o++)a=`${t[t.length-1-o]},`+a;n.push(a)}return n}function PJ(e,t,n){if(e===1)return`rc > ${t[0]}`;let s="";for(let r=e-2;r<e;r++)s+=`${n[r]} >= ${t[r]}`,r<e-1&&(s+="||");return s}function FJ(e,t,n,s){if(e===1)return"";let r=s.slice(-2);return`
|
|
int r = ${r[0]};
|
|
int c = ${r[1]};
|
|
int rp1 = r + 1;
|
|
int cp1 = c + 1;
|
|
|
|
bool cEdge = cp1 >= ${t};
|
|
bool rEdge = rp1 >= ${n};
|
|
`}function OJ(e,t){let n=e.length,s=DJ(n,t);return n===1?`getA(rc),
|
|
rc + 1 >= ${e[0]} ? 0. : getA(rc + 1),
|
|
0, 0`:`getA(${s[0]}),
|
|
cEdge ? 0. : getA(${s[1]}),
|
|
rEdge ? 0. : getA(${s[2]}),
|
|
rEdge || cEdge ? 0. : getA(${s[3]})`}var qI=class{constructor(e,t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.customUniforms=[{name:"inputShape",type:"ivec3"}],this.outputShape=e,this.enableShapeUniforms=Fs(this.outputShape.length);let n="";for(let s=0;s<4;s++){let r="thisRC = rc;";s%2==1&&(r+="thisRC.z += 1;"),s>1&&(r+="thisRC.y += 1;"),n+=`
|
|
${r}
|
|
${s>0?"if(thisRC.y < rows && thisRC.z < cols){":""}
|
|
int flatIndex = getFlatIndex(thisRC);
|
|
|
|
ivec3 inputRC = inputCoordsFromReshapedOutCoords(flatIndex);
|
|
vec2 inputRCInnerDims = vec2(float(inputRC.y),float(inputRC.z));
|
|
|
|
result[${s}] =
|
|
getChannel(getA(inputRC.x, inputRC.y, inputRC.z), inputRCInnerDims);
|
|
${s>0?"}":""}
|
|
`}this.userCode=`
|
|
${MJ(t,this.enableShapeUniforms)}
|
|
${this.enableShapeUniforms?ex():Qy(e)}
|
|
|
|
void main() {
|
|
ivec3 rc = getOutputCoords();
|
|
|
|
vec4 result = vec4(0.);
|
|
|
|
ivec3 thisRC;
|
|
int rows = ${this.enableShapeUniforms?"outShape[1]":e[1]};
|
|
int cols = ${this.enableShapeUniforms?"outShape[2]":e[2]};
|
|
|
|
${n}
|
|
|
|
setOutput(result);
|
|
}
|
|
`}};function MJ(e,t){return`
|
|
ivec3 inputCoordsFromReshapedOutCoords(int index) {
|
|
${t?eY(["r","c","d"],"inputShape"):Il(["r","c","d"],e)}
|
|
return ivec3(r, c, d);
|
|
}
|
|
`}var zJ=class{constructor(e){this.gpgpu=e,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0,this.freeTextures={},this.logEnabled=!1,this.usedTextures={}}acquireTexture(e,t,n){let s=KI(t,n),r=ZI(e,s,n);r in this.freeTextures||(this.freeTextures[r]=[]),r in this.usedTextures||(this.usedTextures[r]=[]);let a=XI(e,s,this.gpgpu.gl,this.gpgpu.textureConfig,n);if(this.freeTextures[r].length>0){this.numFreeTextures--,this.numUsedTextures++,this._numBytesFree-=a,this.log();let i=this.freeTextures[r].shift();return this.usedTextures[r].push(i),i}let o;return s===In.PACKED_2X2_FLOAT32?o=this.gpgpu.createPackedMatrixTexture(e[0],e[1]):s===In.PACKED_2X2_FLOAT16?o=this.gpgpu.createFloat16PackedMatrixTexture(e[0],e[1]):s===In.UNPACKED_FLOAT32?o=this.gpgpu.createFloat32MatrixTexture(e[0],e[1]):s===In.UNPACKED_FLOAT16?o=this.gpgpu.createFloat16MatrixTexture(e[0],e[1]):s===In.PACKED_4X1_UNSIGNED_BYTE&&(o=this.gpgpu.createUnsignedBytesMatrixTexture(e[0],e[1])),this.usedTextures[r].push(o),this.numUsedTextures++,this._numBytesAllocated+=a,this.log(),o}releaseTexture(e,t,n,s){if(this.freeTextures==null)return;let r=KI(n,s),a=ZI(t,r,s);a in this.freeTextures||(this.freeTextures[a]=[]);let o=XI(t,r,this.gpgpu.gl,this.gpgpu.textureConfig,s),i=K().get("WEBGL_DELETE_TEXTURE_THRESHOLD");i!==-1&&this._numBytesAllocated>i?(this.gpgpu.deleteMatrixTexture(e),this._numBytesAllocated-=o):(this.freeTextures[a].push(e),this.numFreeTextures++,this._numBytesFree+=o),this.numUsedTextures--;let l=this.usedTextures[a],c=l.indexOf(e);if(c<0)throw new Error("Cannot release a texture that was never provided by this texture manager");l.splice(c,1),this.log()}log(){if(!this.logEnabled)return;let e=this.numFreeTextures+this.numUsedTextures;console.log("Free/Used",`${this.numFreeTextures} / ${this.numUsedTextures}`,`(${e})`);let t=this._numBytesFree/this._numBytesAllocated;console.log(`Bytes allocated: ${this._numBytesAllocated}`),console.log(`Bytes unused: ${this._numBytesFree} (${Math.round(100*t)}%)`)}get numBytesAllocated(){return this._numBytesAllocated}get numBytesFree(){return this._numBytesFree}getNumUsedTextures(){return this.numUsedTextures}getNumFreeTextures(){return this.numFreeTextures}dispose(){if(this.freeTextures!=null){for(let e in this.freeTextures)this.freeTextures[e].forEach(t=>{this.gpgpu.deleteMatrixTexture(t)});for(let e in this.usedTextures)this.usedTextures[e].forEach(t=>{this.gpgpu.deleteMatrixTexture(t)});this.freeTextures=null,this.usedTextures=null,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0}}};function LJ(e,t){let n=e;if(t===n.R32F)return 4;if(t===n.R16F)return 2;if(t===n.RGBA32F)return 16;if(t===e.RGBA)return 16;if(t===n.RGBA16F)return 8;throw new Error(`Unknown internal format ${t}`)}function XI(e,t,n,s,r){let a=BJ(t,s),o;if(r){let[l,c]=tc(e[0],e[1]);o=l*c}else{let[l,c]=ip(e[0],e[1]);o=l*c}let i=LJ(n,a);return o*i}function BJ(e,t){switch(e){case In.PACKED_2X2_FLOAT32:return ax(t);case In.PACKED_2X2_FLOAT16:return ox(t);case In.UNPACKED_FLOAT32:return nx(t);case In.UNPACKED_FLOAT16:return sx(t);case In.PACKED_4X1_UNSIGNED_BYTE:return rx(t);default:throw new Error(`Unknown physical texture type ${e}`)}}function WJ(e){return K().getBool("WEBGL_RENDER_FLOAT32_ENABLED")?e?In.PACKED_2X2_FLOAT32:In.UNPACKED_FLOAT32:e?In.PACKED_2X2_FLOAT16:In.UNPACKED_FLOAT16}function KI(e,t){if(e===Ds.UPLOAD)return In.PACKED_2X2_FLOAT32;if(e===Ds.RENDER||e==null)return WJ(t);if(e===Ds.DOWNLOAD||e===Ds.PIXELS)return In.PACKED_4X1_UNSIGNED_BYTE;throw new Error(`Unknown logical texture type ${e}`)}function ZI(e,t,n){return`${e[0]}_${e[1]}_${t}_${n}`}var Vo=class{constructor(e,t){this.variableNames=["A"],this.outputShape=e,this.enableShapeUniforms=Fs(this.outputShape.length),this.userCode=`
|
|
float unaryOperation(float x) {
|
|
${t}
|
|
}
|
|
|
|
void main() {
|
|
float x = getAAtOutCoords();
|
|
float y = unaryOperation(x);
|
|
|
|
setOutput(y);
|
|
}
|
|
`}},Ar="if (isnan(x)) return x;",VJ="return x;",YI="return abs(x);",UJ="return (x >= 0.0) ? x : (exp(x) - 1.0);",GJ=Ar+`
|
|
return (x < 0.0) ? 0.0 : x;
|
|
`,HJ=Ar+`
|
|
return (x < 0.0) ? 0.0 : min(6.0, x);
|
|
`,Om="return x;",jJ="return 1.0 / (1.0 + exp(-1.0 * x));",qJ="return x;",XJ=`
|
|
vec4 result;
|
|
|
|
result.r = (x.r >= 0.0) ? x.r : (exp(x.r) - 1.0);
|
|
result.g = (x.g >= 0.0) ? x.g : (exp(x.g) - 1.0);
|
|
result.b = (x.b >= 0.0) ? x.b : (exp(x.b) - 1.0);
|
|
result.a = (x.a >= 0.0) ? x.a : (exp(x.a) - 1.0);
|
|
|
|
return result;
|
|
`,KJ=`
|
|
vec4 result = x * vec4(greaterThanEqual(x, vec4(0.0)));
|
|
bvec4 isNaN = isnan(x);
|
|
|
|
result.r = isNaN.r ? x.r : result.r;
|
|
result.g = isNaN.g ? x.g : result.g;
|
|
result.b = isNaN.b ? x.b : result.b;
|
|
result.a = isNaN.a ? x.a : result.a;
|
|
|
|
return result;
|
|
`,ZJ=`
|
|
vec4 result = min(x, vec4(6.)) * vec4(greaterThanEqual(x, vec4(0.0)));
|
|
bvec4 isNaN = isnan(x);
|
|
|
|
result.r = isNaN.r ? x.r : result.r;
|
|
result.g = isNaN.g ? x.g : result.g;
|
|
result.b = isNaN.b ? x.b : result.b;
|
|
result.a = isNaN.a ? x.a : result.a;
|
|
|
|
return result;
|
|
`,YJ="return 1.0 / (1.0 + exp(-1.0 * x));",ic=class{constructor(e,t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e,this.enableShapeUniforms=Fs(this.outputShape.length),this.userCode=`
|
|
vec4 unaryOperation(vec4 x) {
|
|
${t}
|
|
}
|
|
|
|
void main() {
|
|
vec4 x = getAAtOutCoords();
|
|
vec4 y = unaryOperation(x);
|
|
|
|
setOutput(y);
|
|
}
|
|
`}},JJ=class{constructor(e){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outputShape=e;let t=e.length,n=Gn("rc",t),s=yt(t),r=$J(t,n),a=n.slice(-2),o=t<=1?"rc":`vec2(${a.join(",")})`;this.userCode=`
|
|
void main() {
|
|
${s} rc = getOutputCoords();
|
|
vec4 packedInput = getA(${r});
|
|
|
|
setOutput(getChannel(packedInput, ${o}));
|
|
}
|
|
`}},QJ=Xs.whereImpl,eQ=1e-7,tQ=1e-4,Mm={};function nQ(e){return e in Mm||(Mm[e]={}),Mm[e]}var sQ=K().getNumber("CPU_HANDOFF_SIZE_THRESHOLD"),rQ=600;function aQ(){return K().global.screen==null?1024:K().global.screen.height*K().global.screen.width*window.devicePixelRatio*rQ/1024/1024}var JI=class extends Yl{constructor(e){super();if(this.pendingRead=new WeakMap,this.pendingDisposal=new WeakSet,this.dataRefCount=new WeakMap,this.numBytesInGPU=0,this.uploadWaitMs=0,this.downloadWaitMs=0,this.lastGlFlushTime=0,this.warnedAboutMemory=!1,this.pendingDeletes=0,this.disposed=!1,!K().getBool("HAS_WEBGL"))throw new Error("WebGL is not supported on this device");if(e==null){let t=zr(K().getNumber("WEBGL_VERSION"));this.binaryCache=nQ(K().getNumber("WEBGL_VERSION")),this.gpgpu=new Fm(t),this.canvas=t.canvas,this.gpgpuCreatedLocally=!0}else this.gpgpu=e,this.binaryCache={},this.gpgpuCreatedLocally=!1,this.canvas=e.gl.canvas;this.textureManager=new zJ(this.gpgpu),this.numMBBeforeWarning=aQ(),this.texData=new Xc(this,ns())}nextDataId(){return JI.nextDataId++}numDataIds(){return this.texData.numDataIds()-this.pendingDeletes}write(e,t,n){if((K().getBool("WEBGL_CHECK_NUMERICAL_PROBLEMS")||K().getBool("DEBUG"))&&this.checkNumericalProblems(e),n==="complex64"&&e!=null)throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");let s={id:this.nextDataId()};return this.texData.set(s,{shape:t,dtype:n,values:e,usage:Ds.UPLOAD,refCount:1}),s}refCount(e){return this.texData.has(e)?this.texData.get(e).refCount:0}incRef(e){let t=this.texData.get(e);t.refCount++}decRef(e){if(this.texData.has(e)){let t=this.texData.get(e);t.refCount--}}move(e,t,n,s,r){if(K().getBool("DEBUG")&&this.checkNumericalProblems(t),s==="complex64")throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");this.texData.set(e,{shape:n,dtype:s,values:t,usage:Ds.UPLOAD,refCount:r})}disposeIntermediateTensorInfo(e){this.disposeData(e.dataId)}readSync(e){let t=this.texData.get(e),{values:n,dtype:s,complexTensorInfos:r,slice:a,shape:o,isPacked:i}=t;if(a!=null){let d;i?d=new ic(o,Om):d=new Vo(o,Om);let p=this.runWebGLProgram(d,[{dataId:e,shape:o,dtype:s}],s),h=this.readSync(p.dataId);return this.disposeIntermediateTensorInfo(p),h}if(n!=null)return this.convertAndCacheOnCPU(e);if(s==="string")return n;let l=this.activeTimers!=null,c;l&&(c=v.now());let u;if(s==="complex64"){let d=this.readSync(r.real.dataId),p=this.readSync(r.imag.dataId);u=E.mergeRealAndImagArrays(d,p)}else u=this.getValuesFromTexture(e);return l&&(this.downloadWaitMs+=v.now()-c),this.convertAndCacheOnCPU(e,u)}async read(e){if(this.pendingRead.has(e)){let h=this.pendingRead.get(e);return new Promise(f=>h.push(f))}let t=this.texData.get(e),{values:n,shape:s,slice:r,dtype:a,complexTensorInfos:o,isPacked:i}=t;if(r!=null){let h;i?h=new ic(s,Om):h=new Vo(s,Om);let f=this.runWebGLProgram(h,[{dataId:e,shape:s,dtype:a}],a),m=this.read(f.dataId);return this.disposeIntermediateTensorInfo(f),m}if(n!=null)return this.convertAndCacheOnCPU(e);if(K().getBool("DEBUG")&&!K().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")&&K().getNumber("WEBGL_VERSION")===2)throw new Error("tensor.data() with WEBGL_DOWNLOAD_FLOAT_ENABLED=false and WEBGL_VERSION=2 not yet supported.");let l=null,c;if(a!=="complex64"&&K().get("WEBGL_BUFFER_SUPPORTED")){c=this.decode(e);let h=this.texData.get(c.dataId);l=this.gpgpu.createBufferFromTexture(h.texture,...Tm(s))}this.pendingRead.set(e,[]),a!=="complex64"&&await this.gpgpu.createAndWaitForFence();let u;if(a==="complex64"){let h=await Promise.all([this.read(o.real.dataId),this.read(o.imag.dataId)]),f=h[0],m=h[1];u=E.mergeRealAndImagArrays(f,m)}else if(l==null)u=this.getValuesFromTexture(e);else{let h=v.sizeFromShape(s);u=this.gpgpu.downloadFloat32MatrixFromBuffer(l,h)}if(c!=null&&this.disposeIntermediateTensorInfo(c),l!=null){let h=this.gpgpu.gl;Ie(h,()=>h.deleteBuffer(l))}let d=this.convertAndCacheOnCPU(e,u),p=this.pendingRead.get(e);return this.pendingRead.delete(e),p.forEach(h=>h(d)),this.pendingDisposal.has(e)&&(this.pendingDisposal.delete(e),this.disposeData(e)&&ns().removeDataId(e,this),this.pendingDeletes--),d}bufferSync(e){let t=this.readSync(e.dataId),n=t;if(e.dtype==="string")try{n=t.map(s=>v.decodeString(s))}catch(s){throw new Error("Failed to decode encoded string bytes into utf-8")}return ze(e.shape,e.dtype,n)}checkNumericalProblems(e){if(e!=null)for(let t=0;t<e.length;t++){let n=e[t];if(!YS(n))throw K().getBool("WEBGL_RENDER_FLOAT32_CAPABLE")?Error(`The value ${n} cannot be represented with your current settings. Consider enabling float32 rendering: 'tf.env().set('WEBGL_RENDER_FLOAT32_ENABLED', true);'`):Error(`The value ${n} cannot be represented on this device.`)}}getValuesFromTexture(e){let{shape:t,dtype:n,isPacked:s}=this.texData.get(e),r=v.sizeFromShape(t);if(K().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")){let d=this.decode(e),p=this.texData.get(d.dataId),h=this.gpgpu.downloadMatrixFromPackedTexture(p.texture,...Tm(t)).subarray(0,r);return this.disposeIntermediateTensorInfo(d),h}let a=K().getBool("WEBGL_PACK")&&s===!0,o=a?Rm(t):t,i=a?new UY(o):new VY(o),l=this.runWebGLProgram(i,[{shape:o,dtype:n,dataId:e}],"float32"),c=this.texData.get(l.dataId),u=this.gpgpu.downloadByteEncodedFloatMatrixFromOutputTexture(c.texture,c.texShape[0],c.texShape[1]).subarray(0,r);return this.disposeIntermediateTensorInfo(l),u}timerAvailable(){return K().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0}async time(e){let t=this.activeTimers,n=[],s=!1;this.programTimersStack==null?(this.programTimersStack=n,s=!0):this.activeTimers.push(n),this.activeTimers=n,e();let r=v.flatten(this.activeTimers.map(i=>i.query)).filter(i=>i!=null),a=v.flatten(this.activeTimers.map(i=>i.name)).filter(i=>i!=null);this.activeTimers=t,s&&(this.programTimersStack=null);let o={uploadWaitMs:this.uploadWaitMs,downloadWaitMs:this.downloadWaitMs,kernelMs:null,wallMs:null};if(K().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0){let i=await Promise.all(r);o.kernelMs=v.sum(i),o.getExtraProfileInfo=()=>i.map((l,c)=>({name:a[c],ms:l})).map(l=>`${l.name}: ${l.ms}`).join(", ")}else o.kernelMs={error:"WebGL query timers are not supported in this environment."};return this.uploadWaitMs=0,this.downloadWaitMs=0,o}memory(){return{unreliable:!1,numBytesInGPU:this.numBytesInGPU,numBytesInGPUAllocated:this.textureManager.numBytesAllocated,numBytesInGPUFree:this.textureManager.numBytesFree}}startTimer(){return K().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?this.gpgpu.beginQuery():{startMs:v.now(),endMs:null}}endTimer(e){return K().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?(this.gpgpu.endQuery(),e):(e.endMs=v.now(),e)}async getQueryTime(e){if(K().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0)return this.gpgpu.waitForQueryAndGetTime(e);let t=e;return t.endMs-t.startMs}disposeData(e,t=!1){if(this.pendingDisposal.has(e))return!1;if(!this.texData.has(e))return!0;if(t?this.texData.get(e).refCount=0:this.texData.get(e).refCount--,!t&&this.texData.get(e).refCount>0)return!1;if(this.pendingRead.has(e))return this.pendingDisposal.add(e),this.pendingDeletes++,!1;this.releaseGPUData(e);let{complexTensorInfos:n}=this.texData.get(e);return n!=null&&(this.disposeData(n.real.dataId,t),this.disposeData(n.imag.dataId,t)),this.texData.delete(e),!0}releaseGPUData(e){let{texture:t,dtype:n,texShape:s,usage:r,isPacked:a,slice:o}=this.texData.get(e),i=o&&o.origDataId||e,l=this.dataRefCount.get(i);l>1?this.dataRefCount.set(i,l-1):(this.dataRefCount.delete(i),t!=null&&(this.numBytesInGPU-=this.computeBytes(s,n),this.textureManager.releaseTexture(t,s,r,a)));let c=this.texData.get(e);c.texture=null,c.texShape=null,c.isPacked=!1,c.slice=null}getTexture(e){return this.uploadToGPU(e),this.texData.get(e).texture}getDataInfo(e){return this.texData.get(e)}shouldExecuteOnCPU(e,t=sQ){return K().getBool("WEBGL_CPU_FORWARD")&&e.every(n=>this.texData.get(n.dataId).texture==null&&v.sizeFromShape(n.shape)<t)}getGPGPUContext(){return this.gpgpu}where(e){E.warn("tf.where() in webgl locks the UI thread. Call tf.whereAsync() instead");let t=e.dataSync();return QJ(e.shape,t)}packedUnaryOp(e,t,n){let s=new ic(e.shape,t),r=this.compileAndRun(s,[e],n);return ns().makeTensorFromDataId(r.dataId,r.shape,r.dtype)}abs(e){if(this.shouldExecuteOnCPU([e])&&e.dtype!=="complex64"){let s=GI(this.texData.get(e.dataId).values);return this.makeOutput(e.shape,e.dtype,s)}if(K().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,YI,e.dtype);let t=new Vo(e.shape,YI),n=this.compileAndRun(t,[e]);return ns().makeTensorFromDataId(n.dataId,n.shape,n.dtype)}makeTensorInfo(e,t,n){let s;if(t==="string"&&n!=null&&n.length>0&&v.isString(n[0])){let r=n.map(a=>v.encodeString(a));s=this.write(r,e,t)}else s=this.write(n,e,t);return this.texData.get(s).usage=null,{dataId:s,shape:e,dtype:t}}makeOutput(e,t,n){let{dataId:s}=this.makeTensorInfo(e,t,n);return ns().makeTensorFromDataId(s,e,t,this)}unpackTensor(e){let t=new JJ(e.shape);return this.runWebGLProgram(t,[e],e.dtype)}packTensor(e){let t=new _J(e.shape),n=!0;return this.runWebGLProgram(t,[e],e.dtype,null,n)}packedReshape(e,t){let n=[kl(e.shape),...Sl(e.shape)],s={dtype:e.dtype,shape:n,dataId:e.dataId},r=[kl(t),...Sl(t)],a=new qI(r,n),o=!0,i=[n],l=this.runWebGLProgram(a,[s],e.dtype,i,o);return{dataId:l.dataId,shape:t,dtype:l.dtype}}decode(e){let t=this.texData.get(e),{isPacked:n,shape:s,dtype:r}=t,a=Rm(s),o,i=Tm(a);n?o=new WY(a):o=new BY(a);let l=!0,c=[i],u=this.runWebGLProgram(o,[{shape:a,dtype:r,dataId:e}],r,c,l);return{dtype:r,shape:s,dataId:u.dataId}}runWebGLProgram(e,t,n,s,r=!1){let a=this.makeTensorInfo(e.outputShape,n),o=this.texData.get(a.dataId);if(e.packedOutput&&(o.isPacked=!0),e.outPackingScheme===op.DENSE){let m=Tm(e.outputShape);o.texShape=m.map(g=>g*2)}if(e.outTexUsage!=null&&(o.usage=e.outTexUsage),v.sizeFromShape(a.shape)===0)return o.values=v.getTypedArrayFromDType(a.dtype,0),a;let i=[],l=t.map(m=>{if(m.dtype==="complex64")throw new Error("GPGPUProgram does not support complex64 input. For complex64 dtypes, please separate the program into real and imaginary parts.");let g=this.texData.get(m.dataId);if(g.texture==null){if(!e.packedInputs&&v.sizeFromShape(m.shape)<=K().getNumber("WEBGL_SIZE_UPLOAD_UNIFORM"))return{shape:m.shape,texData:null,isUniform:!0,uniformValues:g.values};e.packedInputs&&(g.isPacked=!0,g.shape=m.shape)}else if(!!g.isPacked!=!!e.packedInputs)m=g.isPacked?this.unpackTensor(m):this.packTensor(m),i.push(m),g=this.texData.get(m.dataId);else if(g.isPacked&&!cp(g.shape,m.shape)){let A=m,x=m.shape;m.shape=g.shape,m=this.packedReshape(m,x),i.push(m),g=this.texData.get(m.dataId),A.shape=x}return this.uploadToGPU(m.dataId),{shape:m.shape,texData:g,isUniform:!1}});this.uploadToGPU(a.dataId);let c={shape:a.shape,texData:o,isUniform:!1},u=LY(e,l,c),d=this.getAndSaveBinary(u,()=>MY(this.gpgpu,e,l,c)),p=this.activeTimers!=null,h;p&&(h=this.startTimer()),zY(this.gpgpu,d,l,c,s),i.forEach(m=>this.disposeIntermediateTensorInfo(m)),p&&(h=this.endTimer(h),this.activeTimers.push({name:e.constructor.name,query:this.getQueryTime(h)}));let f=K().get("WEBGL_FLUSH_THRESHOLD");if(f>0){let m=v.now();m-this.lastGlFlushTime>f&&(this.gpgpu.gl.flush(),this.lastGlFlushTime=m)}if(!K().getBool("WEBGL_LAZILY_UNPACK")&&o.isPacked&&r===!1){let m=this.unpackTensor(a);return this.disposeIntermediateTensorInfo(a),m}return a}compileAndRun(e,t,n,s,r=!1){return n=n||t[0].dtype,this.runWebGLProgram(e,t,n,s,r)}getAndSaveBinary(e,t){return e in this.binaryCache||(this.binaryCache[e]=t()),this.binaryCache[e]}getTextureManager(){return this.textureManager}dispose(){this.disposed||(K().getBool("IS_TEST")||Object.keys(this.binaryCache).forEach(t=>{this.gpgpu.deleteProgram(this.binaryCache[t].webGLProgram),delete this.binaryCache[t]}),this.textureManager.dispose(),this.canvas!=null&&typeof HTMLCanvasElement!="undefined"&&this.canvas instanceof HTMLCanvasElement?this.canvas.remove():this.canvas=null,this.gpgpuCreatedLocally&&(this.gpgpu.program=null,this.gpgpu.dispose()),this.disposed=!0)}floatPrecision(){return this.floatPrecisionValue==null&&(this.floatPrecisionValue=q(()=>{if(!K().get("WEBGL_RENDER_FLOAT32_ENABLED")){let e=K().getBool("DEBUG");K().set("DEBUG",!1);let t=this.abs(Re(1e-8)).dataSync()[0];if(K().set("DEBUG",e),t>0)return 32}return 16})),this.floatPrecisionValue}epsilon(){return this.floatPrecision()===32?eQ:tQ}uploadToGPU(e){let t=this.texData.get(e),{shape:n,dtype:s,values:r,texture:a,usage:o,isPacked:i}=t;if(a!=null)return;let l=this.activeTimers!=null,c;l&&(c=v.now());let u=t.texShape;if(u==null&&(u=fI(n,i),t.texShape=u),r!=null){let d=Rm(n),p,h=u[1],f=u[0],m=r instanceof Uint8Array||r instanceof Uint8ClampedArray;i?([h,f]=tc(u[0],u[1]),p=new HY(d,m)):p=new GY(d,m);let g=this.makeTensorInfo([f,h],s);m?this.texData.get(g.dataId).usage=Ds.PIXELS:this.texData.get(g.dataId).usage=Ds.UPLOAD,this.gpgpu.uploadDenseMatrixToTexture(this.getTexture(g.dataId),h,f,r);let A=[[f,h]],x=!0,y=this.runWebGLProgram(p,[g],s,A,x),b=this.texData.get(y.dataId);t.texture=b.texture,t.texShape=b.texShape,t.isPacked=b.isPacked,t.usage=b.usage,this.disposeIntermediateTensorInfo(g),this.texData.delete(y.dataId),t.values=null,l&&(this.uploadWaitMs+=v.now()-c)}else{let d=this.acquireTexture(u,o,s,i);t.texture=d}}convertAndCacheOnCPU(e,t){let n=this.texData.get(e),{dtype:s}=n;return this.releaseGPUData(e),t!=null&&(n.values=oQ(t,s)),n.values}acquireTexture(e,t,n,s){if(this.numBytesInGPU+=this.computeBytes(e,n),!this.warnedAboutMemory&&this.numBytesInGPU>this.numMBBeforeWarning*1024*1024){let r=(this.numBytesInGPU/1024/1024).toFixed(2);this.warnedAboutMemory=!0,console.warn(`High memory usage in GPU: ${r} MB, most likely due to a memory leak`)}return this.textureManager.acquireTexture(e,t,s)}computeBytes(e,t){return e[0]*e[1]*v.bytesPerElement(t)}},pp=JI;pp.nextDataId=0;function oQ(e,t){if(t==="float32"||t==="complex64")return e;if(t==="int32"||t==="bool"){let n=t==="int32"?new Int32Array(e.length):new Uint8Array(e.length);for(let s=0;s<n.length;++s)n[s]=Math.round(e[s]);return n}else throw new Error(`Unknown dtype ${t}`)}var iQ="0.0.0";function QI(){K().set("WEBGL_FORCE_F16_TEXTURES",!0)}Nu.isBrowser()&&al("webgl",()=>new pp,2);var lQ={forceHalfFloat:QI},e4=`
|
|
if (isnan(a)) return a;
|
|
if (isnan(b)) return b;
|
|
`,lc=class{constructor(e,t,n){this.variableNames=["A","B"],this.outputShape=E.assertAndGetBroadcastShape(t,n),this.enableShapeUniforms=Fs(this.outputShape.length),this.userCode=`
|
|
float binaryOperation(float a, float b) {
|
|
${e}
|
|
}
|
|
|
|
void main() {
|
|
float a = getAAtOutCoords();
|
|
float b = getBAtOutCoords();
|
|
setOutput(binaryOperation(a, b));
|
|
}
|
|
`}},zm=`
|
|
result.r = isNaN.r > 0. ? NAN : result.r;
|
|
result.g = isNaN.g > 0. ? NAN : result.g;
|
|
result.b = isNaN.b > 0. ? NAN : result.b;
|
|
result.a = isNaN.a > 0. ? NAN : result.a;
|
|
`,hp=class{constructor(e,t,n,s=!1){this.variableNames=["A","B"],this.supportsBroadcasting=!0,this.packedInputs=!0,this.packedOutput=!0,this.outputShape=E.assertAndGetBroadcastShape(t,n);let r=this.outputShape.length;this.enableShapeUniforms=Fs(r);let a="";if(s)if(r===0||v.sizeFromShape(this.outputShape)===1)a=`
|
|
result.y = 0.;
|
|
result.z = 0.;
|
|
result.w = 0.;
|
|
`;else if(a=`
|
|
${yt(r)} coords = getOutputCoords();
|
|
`,r===1)this.enableShapeUniforms?a+=`
|
|
result.y = (coords + 1) >= outShape ? 0. : result.y;
|
|
result.z = 0.;
|
|
result.w = 0.;
|
|
`:a+=`
|
|
result.y = (coords + 1) >= ${this.outputShape[0]} ? 0. : result.y;
|
|
result.z = 0.;
|
|
result.w = 0.;
|
|
`;else{let i=Gn("coords",r);this.enableShapeUniforms?a+=`
|
|
bool nextRowOutOfBounds =
|
|
(${i[r-2]} + 1) >= outShape[${r} - 2];
|
|
bool nextColOutOfBounds =
|
|
(${i[r-1]} + 1) >= outShape[${r} - 1];
|
|
result.y = nextColOutOfBounds ? 0. : result.y;
|
|
result.z = nextRowOutOfBounds ? 0. : result.z;
|
|
result.w = nextColOutOfBounds || nextRowOutOfBounds ? 0. : result.w;
|
|
`:a+=`
|
|
bool nextRowOutOfBounds =
|
|
(${i[r-2]} + 1) >= ${this.outputShape[r-2]};
|
|
bool nextColOutOfBounds =
|
|
(${i[r-1]} + 1) >= ${this.outputShape[r-1]};
|
|
result.y = nextColOutOfBounds ? 0. : result.y;
|
|
result.z = nextRowOutOfBounds ? 0. : result.z;
|
|
result.w = nextColOutOfBounds || nextRowOutOfBounds ? 0. : result.w;
|
|
`}this.userCode=`
|
|
vec4 binaryOperation(vec4 a, vec4 b) {
|
|
${e}
|
|
}
|
|
|
|
void main() {
|
|
vec4 a = getAAtOutCoords();
|
|
vec4 b = getBAtOutCoords();
|
|
|
|
vec4 result = binaryOperation(a, b);
|
|
${a}
|
|
|
|
setOutput(result);
|
|
}
|
|
`}};function As(e){let{inputs:t,backend:n}=e,{x:s}=t;return n.incRef(s.dataId),{dataId:s.dataId,shape:s.shape,dtype:s.dtype}}var uQ={kernelName:Ga,backendName:"webgl",kernelFunc:As};function Uo(e){let{inputs:t,backend:n}=e,{real:s,imag:r}=t,a=n.makeTensorInfo(s.shape,"complex64"),o=n.texData.get(a.dataId),i=As({inputs:{x:s},backend:n}),l=As({inputs:{x:r},backend:n});return o.complexTensorInfos={real:i,imag:l},a}var cQ={kernelName:Jc,backendName:"webgl",kernelFunc:Uo},t4="return (a < 0.) ? b * a : a;",n4=`
|
|
vec4 aLessThanZero = vec4(lessThan(a, vec4(0.)));
|
|
return (aLessThanZero * (b * a)) + ((vec4(1.0) - aLessThanZero) * a);
|
|
`;function dQ(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{alpha:a}=s,o=n.makeTensorInfo([],"float32",v.createScalarValue(a,"float32")),i=K().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new hp(n4,r.shape,o.shape):new lc(t4,r.shape,o.shape),l=n.runWebGLProgram(i,[r,o],"float32");return n.disposeIntermediateTensorInfo(o),l}var pQ={kernelName:ki,backendName:"webgl",kernelFunc:dQ},s4="return (a < 0.) ? b * a : a;",r4=`
|
|
vec4 aLessThanZero = vec4(lessThan(a, vec4(0.)));
|
|
return (aLessThanZero * (b * a)) + ((vec4(1.0) - aLessThanZero) * a);
|
|
`;function hQ(e){let{inputs:t,backend:n}=e,{x:s,alpha:r}=t,a=K().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new hp(r4,s.shape,r.shape):new lc(s4,s.shape,r.shape);return n.runWebGLProgram(a,[s,r],"float32")}var fQ={kernelName:no,backendName:"webgl",kernelFunc:hQ},a4="if (isnan(x)) return x;",mQ=`
|
|
if (isnan(a)) return a;
|
|
if (isnan(b)) return b;
|
|
`,gQ=`
|
|
result.r = isNaN.r > 0. ? NAN : result.r;
|
|
result.g = isNaN.g > 0. ? NAN : result.g;
|
|
result.b = isNaN.b > 0. ? NAN : result.b;
|
|
result.a = isNaN.a > 0. ? NAN : result.a;
|
|
`;function st({opSnippet:e,packedOpSnippet:t,cpuKernelImpl:n,dtype:s}){return({inputs:r,backend:a})=>{let{x:o}=r,i=a,l=s||o.dtype;if(i.shouldExecuteOnCPU([o])&&n!=null){let d=i.texData.get(o.dataId),p=n(d.values,l);return i.makeTensorInfo(o.shape,l,p)}let c=K().getBool("WEBGL_PACK_UNARY_OPERATIONS")&&t!=null,u;return c?u=new ic(o.shape,t):u=new Vo(o.shape,e),i.runWebGLProgram(u,[o],l)}}function Cn({opSnippet:e,packedOpSnippet:t,checkOutOfBounds:n=!1,supportsComplex:s=!1,cpuKernelImpl:r,dtype:a}){return({inputs:o,backend:i})=>{let{a:l,b:c}=o,u=i;if(s&&l.dtype==="complex64"){let f=u.texData.get(l.dataId),m=u.texData.get(c.dataId),[g,A]=[[f.complexTensorInfos.real,m.complexTensorInfos.real],[f.complexTensorInfos.imag,m.complexTensorInfos.imag]].map(y=>{let[b,w]=y,k={dataId:b.dataId,dtype:b.dtype,shape:l.shape},C={dataId:w.dataId,dtype:w.dtype,shape:c.shape},N=new lc(e,l.shape,c.shape);return u.runWebGLProgram(N,[k,C],zn(b.dtype,w.dtype))}),x=Uo({inputs:{real:g,imag:A},backend:u});return u.disposeIntermediateTensorInfo(g),u.disposeIntermediateTensorInfo(A),x}let d=a||zn(l.dtype,c.dtype);if((l.dtype==="string"||c.dtype==="string"||u.shouldExecuteOnCPU([l,c]))&&r!=null){let f=u.texData.get(l.dataId).values,m=u.texData.get(c.dataId).values,g=l.dtype==="string"?E.fromUint8ToStringArray(f):f,A=l.dtype==="string"?E.fromUint8ToStringArray(m):m,[x,y]=r(l.shape,c.shape,g,A,d),b=u.makeTensorInfo(y,d),w=u.texData.get(b.dataId);return w.values=x,b}let p=K().getBool("WEBGL_PACK_BINARY_OPERATIONS")&&t!=null,h;return p?h=new hp(t,l.shape,c.shape,n):h=new lc(e,l.shape,c.shape),u.runWebGLProgram(h,[l,c],d)}}function Lm(e,t=!1){if(e==="linear")return t?qJ:VJ;if(e==="relu")return t?KJ:GJ;if(e==="elu")return t?XJ:UJ;if(e==="relu6")return t?ZJ:HJ;if(e==="prelu")return t?r4:s4;if(e==="leakyrelu")return t?n4:t4;if(e==="sigmoid")return t?YJ:jJ;throw new Error(`Activation ${e} has not been implemented for the WebGL backend.`)}var o4=class{constructor(e,t,n,s=!1,r=!1,a=!1,o=null,i=!1,l=!1){this.variableNames=["matrixA","matrixB"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=n,this.enableShapeUniforms=Fs(this.outputShape.length);let c=s?e[1]:e[2],u=Math.ceil(c/2),d=s?"i * 2, rc.y":"rc.y, i * 2",p=r?"rc.z, i * 2":"i * 2, rc.z",h=s?["a.xxyy","a.zzww"]:["a.xxzz","a.yyww"],f=r?["b.xzxz","b.ywyw"]:["b.xyxy","b.zwzw"],m="",g="";o&&(i?m=`vec4 activation(vec4 a) {
|
|
vec4 b = getPreluActivationWeightsAtOutCoords();
|
|
${o}
|
|
}`:l?m=`vec4 activation(vec4 a) {
|
|
vec4 b = getLeakyreluAlphaAtOutCoords();
|
|
${o}
|
|
}`:m=`vec4 activation(vec4 x) {
|
|
${o}
|
|
}`,g="result = activation(result);");let A=a?"result += getBiasAtOutCoords();":"";a&&this.variableNames.push("bias"),i&&this.variableNames.push("preluActivationWeights"),l&&this.variableNames.push("leakyreluAlpha");let x="rc.x",y="rc.x";e[0]<t[0]?x=`int(min(float(rc.x), ${e[0]-1}.))`:t[0]<e[0]&&(y=`int(min(float(rc.x), ${t[0]-1}.))`),this.userCode=`
|
|
${m}
|
|
// Don't use uniform for sharedDimensionPacked for performance.
|
|
const float sharedDimension = ${u}.0;
|
|
|
|
vec4 dot2x2ARowBCol(ivec3 rc) {
|
|
vec4 result = vec4(0);
|
|
for (int i = 0; i < ${u}; i++) {
|
|
int batchA = ${x};
|
|
int batchB = ${y};
|
|
vec4 a = getMatrixA(batchA, ${d});
|
|
vec4 b = getMatrixB(batchB, ${p});
|
|
|
|
// These swizzled products need to be separately added.
|
|
// See: https://github.com/tensorflow/tfjs/issues/1735
|
|
result += (${h[0]} * ${f[0]});
|
|
result += (${h[1]} * ${f[1]});
|
|
}
|
|
return result;
|
|
}
|
|
|
|
void main() {
|
|
ivec3 rc = getOutputCoords();
|
|
vec4 result = dot2x2ARowBCol(rc);
|
|
|
|
${A}
|
|
|
|
${g}
|
|
|
|
setOutput(result);
|
|
}
|
|
`}},i4={REAL:"return areal * breal - aimag * bimag;",IMAG:"return areal * bimag + aimag * breal;"},l4=class{constructor(e,t,n){this.variableNames=["AReal","AImag","BReal","BImag"],this.outputShape=E.assertAndGetBroadcastShape(t,n),this.userCode=`
|
|
float binaryOpComplex(
|
|
float areal, float aimag, float breal, float bimag) {
|
|
${e}
|
|
}
|
|
|
|
void main() {
|
|
float areal = getARealAtOutCoords();
|
|
float aimag = getAImagAtOutCoords();
|
|
float breal = getBRealAtOutCoords();
|
|
float bimag = getBImagAtOutCoords();
|
|
setOutput(binaryOpComplex(areal, aimag, breal, bimag));
|
|
}
|
|
`}},u4="return a * b;";function lx(e){let{inputs:t,backend:n}=e,{a:s,b:r}=t,a=E.upcastType(s.dtype,r.dtype);if(s.dtype==="complex64"){let i=n.texData.get(s.dataId),l=n.texData.get(r.dataId),c=new l4(i4.REAL,s.shape,r.shape),u=new l4(i4.IMAG,s.shape,r.shape),d=[{dataId:i.complexTensorInfos.real.dataId,dtype:i.complexTensorInfos.real.dtype,shape:s.shape},{dataId:i.complexTensorInfos.imag.dataId,dtype:i.complexTensorInfos.imag.dtype,shape:s.shape},{dataId:l.complexTensorInfos.real.dataId,dtype:l.complexTensorInfos.real.dtype,shape:r.shape},{dataId:l.complexTensorInfos.imag.dataId,dtype:l.complexTensorInfos.imag.dtype,shape:r.shape}],p=n.runWebGLProgram(c,d,"float32"),h=n.runWebGLProgram(u,d,"float32"),f=Uo({inputs:{real:p,imag:h},backend:n});return n.disposeIntermediateTensorInfo(p),n.disposeIntermediateTensorInfo(h),f}if(n.shouldExecuteOnCPU([s,r])){let i=n.texData.get(s.dataId),l=n.texData.get(r.dataId),[c,u]=pJ(s.shape,r.shape,i.values,l.values,a),d=n.makeTensorInfo(u,a),p=n.texData.get(d.dataId);return p.values=c,d}let o;return K().getBool("WEBGL_PACK_BINARY_OPERATIONS")?o=new hp(u4,s.shape,r.shape):o=new lc(u4,s.shape,r.shape),n.runWebGLProgram(o,[s,r],a)}var AQ={kernelName:Qa,backendName:"webgl",kernelFunc:lx};function yQ(e,t,n){let s=[kl(e.shape),...Sl(e.shape)],r={dtype:e.dtype,shape:s,dataId:e.dataId},a=[kl(t),...Sl(t)],o=new qI(a,s),i=!0,l=[s],c=n.runWebGLProgram(o,[r],e.dtype,l,i);return{dataId:c.dataId,shape:t,dtype:c.dtype}}function ve(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{shape:a}=s,o=n,i=v.sizeFromShape(r.shape),l=v.inferFromImplicitShape(a,i),c=v.sizeFromShape(l);v.assert(i===c,()=>`The new shape (${l}) has ${c} elements and the old shape (${r.shape}) has ${i} elements. The new shape and old shape must have the same number of elements.`);let u=o.texData.get(r.dataId);return u.isPacked&&!cp(r.shape,l)&&!(u.texture!==null&&cp(u.shape,l))?yQ(r,l,o):(o.incRef(r.dataId),{dataId:r.dataId,shape:l,dtype:r.dtype})}var xQ={kernelName:Fi,backendName:"webgl",kernelFunc:ve},c4=class{constructor(e,t){this.variableNames=["x"];let{windowSize:n,batchSize:s,inSize:r,outSize:a}=e;this.outputShape=[s,a];let o=Math.floor(n/4)*4,i=n%4,l="sumValue += dot(values, ones);";if(t!=null){let u=1/t;l=`sumValue += dot(values * ${v.isInt(u)?u.toPrecision(2):u}, ones);`}let c="";r%n>0&&(c=`
|
|
if (inIdx < 0 || inIdx >= ${r}) {
|
|
return 0.0;
|
|
}
|
|
`),this.userCode=`
|
|
const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);
|
|
|
|
float getValue(int batch, int inIdx) {
|
|
${c}
|
|
return getX(batch, inIdx);
|
|
}
|
|
|
|
void main() {
|
|
ivec2 coords = getOutputCoords();
|
|
int batch = coords[0];
|
|
int outIdx = coords[1];
|
|
int inOffset = outIdx * ${n};
|
|
|
|
float sumValue = 0.0;
|
|
|
|
for (int i = 0; i < ${o}; i += 4) {
|
|
int inIdx = inOffset + i;
|
|
vec4 values = vec4(
|
|
getValue(batch, inIdx),
|
|
getValue(batch, inIdx + 1),
|
|
getValue(batch, inIdx + 2),
|
|
getValue(batch, inIdx + 3)
|
|
);
|
|
|
|
${l}
|
|
}
|
|
|
|
int inIdx = inOffset + ${o};
|
|
if (${i===1}) {
|
|
vec4 values = vec4(getValue(batch, inIdx), 0.0, 0.0, 0.0);
|
|
|
|
${l}
|
|
} else if (${i===2}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, inIdx),
|
|
getValue(batch, inIdx + 1), 0.0, 0.0);
|
|
|
|
${l}
|
|
} else if (${i===3}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, inIdx),
|
|
getValue(batch, inIdx + 1),
|
|
getValue(batch, inIdx + 2), 0.0);
|
|
|
|
${l}
|
|
}
|
|
setOutput(sumValue);
|
|
}
|
|
`}},bQ=class{constructor(e,t){this.variableNames=["x"];let{windowSize:n,batchSize:s,inSize:r,outSize:a}=e;this.outputShape=[s,a];let o="0.0",i="";t==="prod"?o="1.0":t==="min"?(o="1.0 / 1e-20",i="min"):t==="max"&&(o="-1.0 / 1e-20",i="max");let l=`${t}(${t}(${t}(minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])`;t==="sum"?l="sumValue":t==="prod"?l="prodValue":t==="all"?l="allValue":t==="any"&&(l="anyValue");let c=Math.floor(n/4)*4,u=n%4,d=`
|
|
if (${t==="sum"}) {
|
|
sumValue += dot(values, ones);
|
|
} else if (${t==="prod"}) {
|
|
vec2 tmp = vec2(values[0], values[1]) * vec2(values[2], values[3]);
|
|
prodValue *= tmp[0] * tmp[1];
|
|
} else {
|
|
minMaxValue = ${i}(values, minMaxValue);
|
|
if (${t==="min"} || ${t==="max"}) {
|
|
minMaxValue = ${i}(values, minMaxValue);
|
|
bvec4 isNaN = isnan(values);
|
|
if (isNaN.r || isNaN.g || isNaN.b || isNaN.a) {
|
|
minMaxValue = vec4(NAN);
|
|
}
|
|
}
|
|
}
|
|
`,p="vec4";t==="all"?(o="1.0",d=`
|
|
bool reducedAllValue = all(values);
|
|
float floatedReducedAllValue = float(reducedAllValue);
|
|
allValue = float(allValue >= 1.0 && floatedReducedAllValue >= 1.0);
|
|
`,p="bvec4"):t==="any"&&(o="0.0",d=`
|
|
bool reducedAnyValue = any(values);
|
|
float floatedReducedAnyValue = float(reducedAnyValue);
|
|
anyValue = float(anyValue >= 1.0 || floatedReducedAnyValue >= 1.0);
|
|
`,p="bvec4");let h="";r%n>0&&(h=`
|
|
if (inIdx < 0 || inIdx >= ${r}) {
|
|
return initializationValue;
|
|
}
|
|
`),this.userCode=`
|
|
const float initializationValue = ${o};
|
|
const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);
|
|
|
|
float getValue(int batch, int inIdx) {
|
|
${h}
|
|
return getX(batch, inIdx);
|
|
}
|
|
|
|
void main() {
|
|
ivec2 coords = getOutputCoords();
|
|
int batch = coords[0];
|
|
int outIdx = coords[1];
|
|
int inOffset = outIdx * ${n};
|
|
|
|
vec4 minMaxValue = vec4(${o});
|
|
float prodValue = 1.0;
|
|
float sumValue = 0.0;
|
|
float allValue = 1.0;
|
|
float anyValue = 0.0;
|
|
|
|
for (int i = 0; i < ${c}; i += 4) {
|
|
int inIdx = inOffset + i;
|
|
${p} values = ${p}(
|
|
getValue(batch, inIdx),
|
|
getValue(batch, inIdx + 1),
|
|
getValue(batch, inIdx + 2),
|
|
getValue(batch, inIdx + 3)
|
|
);
|
|
|
|
${d}
|
|
}
|
|
|
|
int inIdx = inOffset + ${c};
|
|
if (${u===1}) {
|
|
${p} values = ${p}(
|
|
getValue(batch, inIdx),
|
|
initializationValue,
|
|
initializationValue,
|
|
initializationValue
|
|
);
|
|
|
|
${d}
|
|
} else if (${u===2}) {
|
|
${p} values = ${p}(
|
|
getValue(batch, inIdx),
|
|
getValue(batch, inIdx + 1),
|
|
initializationValue,
|
|
initializationValue
|
|
);
|
|
|
|
${d}
|
|
} else if (${u===3}) {
|
|
${p} values = ${p}(
|
|
getValue(batch, inIdx),
|
|
getValue(batch, inIdx + 1),
|
|
getValue(batch, inIdx + 2),
|
|
initializationValue
|
|
);
|
|
|
|
${d}
|
|
}
|
|
setOutput(${l});
|
|
}
|
|
`}};function vQ(e){let t=[];for(;t.length===0||t[t.length-1].outSize!==1;){let n=t.length?t[t.length-1].outSize:e[1],s=E.computeOptimalWindowSize(n);t.push({inSize:n,windowSize:s,outSize:Math.ceil(n/s)})}return t}function Tl(e,t,n,s){let r=vQ(e.shape),a=e;for(let o=0;o<r.length;o++){let{inSize:i,windowSize:l,outSize:c}=r[o],u,d;n==="mean"?u=o===0?new c4({windowSize:l,inSize:i,batchSize:e.shape[0],outSize:c},i):new c4({windowSize:l,inSize:i,batchSize:e.shape[0],outSize:c}):u=new bQ({windowSize:l,inSize:i,batchSize:e.shape[0],outSize:c},n),d=a,a=s.runWebGLProgram(u,[a],t),d.dataId!==e.dataId&&s.disposeIntermediateTensorInfo(d)}return a}var wQ=class{constructor(e,t){this.variableNames=["A"];let n=new Array(e.length);for(let a=0;a<n.length;a++)n[a]=e[t[a]];this.outputShape=n,this.rank=n.length;let s=yt(this.rank),r=kQ(t);this.userCode=`
|
|
void main() {
|
|
${s} resRC = getOutputCoords();
|
|
setOutput(getA(${r}));
|
|
}
|
|
`}};function kQ(e){let t=e.length;if(t>6)throw Error(`Transpose for rank ${t} is not yet supported`);let n=["resRC.x","resRC.y","resRC.z","resRC.w","resRC.u","resRC.v"],s=new Array(t);for(let r=0;r<e.length;r++)s[e[r]]=n[r];return s.join()}var SQ=class{constructor(e,t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0;let n=new Array(e.length);for(let c=0;c<n.length;c++)n[c]=e[t[c]];if(this.outputShape=n,this.rank=n.length,this.rank>6)throw Error(`Packed transpose for rank ${this.rank} is not yet supported.`);let s=yt(this.rank),r=jI("rc",this.rank),a=new Array(this.rank);for(let c=0;c<t.length;c++)a[t[c]]=r[c];let o=`vec2(${a.slice(-2).join()})`,i=`++${r[this.rank-1]} < ${n[this.rank-1]}`,l=`getChannel(getA(${a.join()}), ${o})`;this.userCode=`
|
|
void main() {
|
|
${s} rc = getOutputCoords();
|
|
vec4 result = vec4(0.);
|
|
result[0] = ${l};
|
|
if(${i}) {
|
|
result[1] = ${l};
|
|
}
|
|
--${r[this.rank-1]};
|
|
if(++${r[this.rank-2]} < ${n[this.rank-2]}) {
|
|
result[2] = ${l};
|
|
if(${i}) {
|
|
result[3] = ${l};
|
|
}
|
|
}
|
|
setOutput(result);
|
|
}
|
|
`}};function Bm(e,t,n){let s=K().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new SQ(e.shape,t):new wQ(e.shape,t);return n.runWebGLProgram(s,[e],e.dtype)}function IQ(e,t,n,s){let r=t,a=e.shape.length,o=v.parseAxisParam(r,e.shape),i=o,l=E.getAxesPermutation(i,a),c=l!=null,u=e;c&&(u=Bm(e,l,s),i=E.getInnerMostAxes(i.length,a)),E.assertAxesAreInnerMostDims("sum",i,a);let[d,p]=E.computeOutAndReduceShapes(u.shape,i),h=d;n&&(h=E.expandShapeToKeepDim(d,o));let f=v.sizeFromShape(p),g=v.sizeFromShape(e.shape)/f,A=ve({inputs:{x:u},attrs:{shape:[g,f]},backend:s}),x=bd(e.dtype),y=Tl(A,x,"sum",s),b=ve({inputs:{x:y},attrs:{shape:h},backend:s});return s.disposeIntermediateTensorInfo(A),s.disposeIntermediateTensorInfo(y),c&&s.disposeIntermediateTensorInfo(u),b}function Wm(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a,keepDims:o}=s;return IQ(r,a,o,n)}var CQ={kernelName:co,backendName:"webgl",kernelFunc:Wm};function Hn(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{perm:a}=s,o=n,i=r.shape.length,l=new Array(i);for(let u=0;u<l.length;u++)l[u]=r.shape[a[u]];let c;if(o.shouldExecuteOnCPU([r])){let d=o.texData.get(r.dataId).values,p=ix(d,r.shape,r.dtype,a,l);c=o.makeTensorInfo(l,r.dtype);let h=o.texData.get(c.dataId);h.values=p}else c=Bm(r,a,o);return c}var TQ={kernelName:go,backendName:"webgl",kernelFunc:Hn},d4=1e3;function Vm({a:e,b:t,transposeA:n,transposeB:s,backend:r,bias:a=null,preluActivationWeights:o=null,leakyreluAlpha:i=0,activation:l=null}){let c=e.shape.length,u=t.shape.length,d=n?e.shape[c-2]:e.shape[c-1],p=s?t.shape[u-1]:t.shape[u-2],h=n?e.shape[c-1]:e.shape[c-2],f=s?t.shape[u-2]:t.shape[u-1],m=e.shape.slice(0,-2),g=t.shape.slice(0,-2),A=v.sizeFromShape(m),x=v.sizeFromShape(g),b=nl.assertAndGetBroadcastShape(e.shape.slice(0,-2),t.shape.slice(0,-2)).concat([h,f]);v.assert(d===p,()=>`Error in matMul: inner shapes (${d}) and (${p}) of Tensors with shapes ${e.shape} and ${t.shape} and transposeA=${n} and transposeB=${s} must match.`);let w=n?[A,d,h]:[A,h,d],k=s?[x,f,p]:[x,p,f],C=ve({inputs:{x:e},backend:r,attrs:{shape:w}}),N=ve({inputs:{x:t},backend:r,attrs:{shape:k}}),R=[C,N],F=Math.max(A,x),_=n?C.shape[1]:C.shape[2],P=a!=null,T=o!=null,M=l==="leakyrelu",U=l!=null?Lm(l,!0):null,j=P||T||M||U!=null,z;if((h===1||f===1)&&_>d4&&j===!1){let Z=C,J=N;n&&(Z=Hn({inputs:{x:C},backend:r,attrs:{perm:[0,2,1]}}),R.push(Z)),s&&(J=Hn({inputs:{x:N},backend:r,attrs:{perm:[0,2,1]}}),R.push(J));let ee=f!==1,ne=f===1,Q=Z;ee&&(Q=ve({inputs:{x:Z},backend:r,attrs:{shape:[F,_,1]}}),R.push(Q));let te=f===1?2:1,oe=J;ne&&(oe=ve({inputs:{x:J},backend:r,attrs:{shape:[F,1,_]}}),R.push(oe));let fe=lx({inputs:{a:Q,b:oe},backend:r});z=Wm({inputs:{x:fe},backend:r,attrs:{axis:te,keepDims:!0}}),R.push(fe)}else{let Z=zn(e.dtype,t.dtype),J=new o4(w,k,[F,h,f],n,s,P,U,T,M),ee=[C,N];if(a!=null&&ee.push(a),T&&ee.push(o),M){let ne=r.makeTensorInfo([],"float32",v.createScalarValue(i,"float32"));ee.push(ne),R.push(ne)}z=r.runWebGLProgram(J,ee,Z)}let X=ve({inputs:{x:z},backend:r,attrs:{shape:b}});R.push(z);for(let Z of R)r.disposeIntermediateTensorInfo(Z);return X}function NQ(e){let{inputs:t,backend:n,attrs:s}=e,{a:r,b:a,bias:o,preluActivationWeights:i}=t,{transposeA:l,transposeB:c,activation:u,leakyreluAlpha:d}=s;return Vm({a:r,b:a,transposeA:l,transposeB:c,backend:n,bias:o,preluActivationWeights:i,leakyreluAlpha:d,activation:u})}var EQ={kernelName:yo,backendName:"webgl",kernelFunc:NQ},p4="return abs(x);";function RQ(e){let{inputs:t,backend:n}=e,{x:s}=t;if(n.shouldExecuteOnCPU([s])&&s.dtype!=="complex64"){let a=n.texData.get(s.dataId),o=GI(a.values);return n.makeTensorInfo(s.shape,s.dtype,o)}let r;return K().getBool("WEBGL_PACK_UNARY_OPERATIONS")?r=new ic(s.shape,p4):r=new Vo(s.shape,p4),n.runWebGLProgram(r,[s],s.dtype)}var $Q={kernelName:ci,backendName:"webgl",kernelFunc:RQ},_Q=Ar+`
|
|
if (abs(x) > 1.) {
|
|
return NAN;
|
|
}
|
|
return acos(x);
|
|
`,DQ=st({opSnippet:_Q}),PQ={kernelName:eu,backendName:"webgl",kernelFunc:DQ},FQ=Ar+`
|
|
if (x < 1.0) return NAN;
|
|
return log(x + sqrt(x * x - 1.0));`,OQ=st({opSnippet:FQ}),MQ={kernelName:tu,backendName:"webgl",kernelFunc:OQ},h4="return a + b;",zQ=Cn({opSnippet:h4,packedOpSnippet:h4,supportsComplex:!0,cpuKernelImpl:qY}),LQ={kernelName:Gr,backendName:"webgl",kernelFunc:zQ},BQ=class{constructor(e,t){this.outputShape=[],this.outputShape=e,this.variableNames=t.map((r,a)=>`T${a}`);let n=[];this.variableNames.forEach(r=>{n.push(`float v${r} = get${r}AtOutCoords();`)});let s=this.variableNames.map(r=>`v${r}`).join(" + ");this.userCode=`
|
|
void main() {
|
|
${n.join(`
|
|
`)}
|
|
|
|
float result = ${s};
|
|
setOutput(result);
|
|
}
|
|
`}},WQ=class{constructor(e,t){this.outputShape=[],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e,this.variableNames=t.map((r,a)=>`T${a}`);let n=[];this.variableNames.forEach(r=>{n.push(`vec4 v${r} = get${r}AtOutCoords();`)});let s=this.variableNames.map(r=>`v${r}`).join(" + ");this.userCode=`
|
|
void main() {
|
|
${n.join(`
|
|
`)}
|
|
|
|
vec4 result = ${s};
|
|
setOutput(result);
|
|
}
|
|
`}};function Um(e){let{inputs:t,backend:n}=e,s=t;if(s.length===1)return As({inputs:{x:s[0]},backend:n});if(s.length>K().get("WEBGL_MAX_TEXTURES_IN_SHADER")){let l=Math.floor(s.length/2),c=Um({inputs:s.slice(0,l),backend:n}),u=Um({inputs:s.slice(l),backend:n});return Um({inputs:[c,u],backend:n})}let r=s.map(l=>l.dtype).reduce((l,c)=>zn(l,c)),a=s.map(l=>l.shape),i=K().getBool("WEBGL_PACK")?new WQ(s[0].shape,a):new BQ(s[0].shape,a);return n.runWebGLProgram(i,s,r)}var VQ={kernelName:Ca,backendName:"webgl",kernelFunc:Um};function UQ(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a,keepDims:o}=s,i=r.shape.length,l=v.parseAxisParam(a,r.shape),c=l,u=E.getAxesPermutation(c,i),d=r;u!=null&&(d=Hn({inputs:{x:r},backend:n,attrs:{perm:u}}),c=E.getInnerMostAxes(c.length,i)),E.assertAxesAreInnerMostDims("all",c,i);let[p,h]=E.computeOutAndReduceShapes(d.shape,c),f=v.sizeFromShape(h),m=ve({inputs:{x:d},backend:n,attrs:{shape:[-1,f]}}),g=Tl(m,m.dtype,"all",n),A;if(o){let x=E.expandShapeToKeepDim(p,l);A=ve({inputs:{x:g},backend:n,attrs:{shape:x}})}else A=ve({inputs:{x:g},backend:n,attrs:{shape:p}});return n.disposeIntermediateTensorInfo(m),n.disposeIntermediateTensorInfo(g),u!=null&&n.disposeIntermediateTensorInfo(d),A}var GQ={kernelName:nu,backendName:"webgl",kernelFunc:UQ};function HQ(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a,keepDims:o}=s,i=r.shape.length,l=v.parseAxisParam(a,r.shape),c=l,u=E.getAxesPermutation(c,i),d=r;u!=null&&(d=Hn({inputs:{x:r},backend:n,attrs:{perm:u}}),c=E.getInnerMostAxes(c.length,i)),E.assertAxesAreInnerMostDims("any",c,i);let[p,h]=E.computeOutAndReduceShapes(d.shape,c),f=v.sizeFromShape(h),m=ve({inputs:{x:d},backend:n,attrs:{shape:[-1,f]}}),g=Tl(m,m.dtype,"any",n),A;if(o){let x=E.expandShapeToKeepDim(p,l);A=ve({inputs:{x:g},backend:n,attrs:{shape:x}})}else A=ve({inputs:{x:g},backend:n,attrs:{shape:p}});return n.disposeIntermediateTensorInfo(m),n.disposeIntermediateTensorInfo(g),u!=null&&n.disposeIntermediateTensorInfo(d),A}var jQ={kernelName:su,backendName:"webgl",kernelFunc:HQ},qQ=class{constructor(e,t,n){this.variableNames=["A"];let{windowSize:s,batchSize:r,outSize:a}=e;n||this.variableNames.push("bestIndicesA"),this.outputShape=[r,a];let o=t==="max"?">":"<",i=n?"inOffset + i;":"round(getBestIndicesA(batch, inOffset + i));";this.userCode=`
|
|
void main() {
|
|
ivec2 coords = getOutputCoords();
|
|
int batch = coords[0];
|
|
int outIdx = coords[1];
|
|
int inOffset = outIdx * ${s};
|
|
|
|
int bestIndex = inOffset;
|
|
float bestValue = getA(batch, bestIndex);
|
|
|
|
for (int i = 0; i < ${s}; i++) {
|
|
int inIdx = ${i};
|
|
float candidate = getA(batch, inIdx);
|
|
if (candidate ${o} bestValue) {
|
|
bestValue = candidate;
|
|
bestIndex = inIdx;
|
|
}
|
|
}
|
|
setOutput(float(bestIndex));
|
|
}
|
|
`}},XQ=class{constructor(e,t,n,s){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,v.assert(e.length>2,()=>`Packed arg${n.charAt(0).toUpperCase()+n.slice(1)} supports only inputs with rank above 2.`);let r=e[e.length-1],a=Math.ceil(r/t);this.outputShape=e.slice(0,-1),a>1&&this.outputShape.push(a),s||this.variableNames.push("bestIndicesA");let o=this.outputShape,i=o.length,l=yt(i),c=Gn("coords",i),u,d;if(a===1){d=i+1;let C=yt(d);u=`
|
|
${C} sourceLocR = ${C}(${c.join()}, 0);
|
|
++${c[i-1]};
|
|
${C} sourceLocG = ${C}(${c.join()}, 0);
|
|
++${c[i-2]};
|
|
${C} sourceLocA = ${C}(${c.join()}, 0);
|
|
--${c[i-1]};
|
|
${C} sourceLocB = ${C}(${c.join()}, 0);
|
|
--${c[i-2]};`}else d=i,u=`
|
|
${l} sourceLocR = coords;
|
|
++${c[i-1]};
|
|
${l} sourceLocG = coords;
|
|
++${c[i-2]};
|
|
${l} sourceLocA = coords;
|
|
--${c[i-1]};
|
|
${l} sourceLocB = coords;
|
|
--${c[i-2]};`;let p=["x","y","z","w","u","v"].slice(0,d),h="."+p[d-1],f=p.map(C=>"int "+C),m=Gn("sourceLocR",d-1).concat("inIdx.r"),g=Gn("sourceLocG",d-1).concat("inIdx.g"),A=Gn("sourceLocB",d-1).concat("inIdx.b"),x=Gn("sourceLocA",d-1).concat("inIdx.a"),y=n==="max"?"greaterThan":"lessThan",b=s?"":`
|
|
inIdx = round(vec4(getBestIndicesAChannel(${m.join()}),
|
|
getBestIndicesAChannel(${g.join()}),
|
|
getBestIndicesAChannel(${A.join()}),
|
|
getBestIndicesAChannel(${x.join()})));`,w=`vec4(
|
|
getAChannel(${m.join()}),
|
|
hasNextCol ? getAChannel(${g.join()}) : 0.,
|
|
hasNextRow ? getAChannel(${A.join()}) : 0.,
|
|
hasNextRow && hasNextCol ? getAChannel(${x.join()}) : 0.)`,k=s?"":`
|
|
float getBestIndicesAChannel(${f.join()}) {
|
|
return getChannel(getBestIndicesA(${p.join()}),
|
|
vec2(${p.slice(-2).join()}));
|
|
}`;this.userCode=`
|
|
float getAChannel(${f.join()}) {
|
|
return getChannel(getA(${p.join()}),
|
|
vec2(${p.slice(-2).join()}));
|
|
}
|
|
${k}
|
|
void main() {
|
|
${l} coords = getOutputCoords();
|
|
bool hasNextCol = ${c[i-1]} < ${o[i-1]-1};
|
|
bool hasNextRow = ${c[i-2]} < ${o[i-2]-1};
|
|
${u}
|
|
ivec4 srcIdx = ivec4(sourceLocR${h}, sourceLocG${h},
|
|
sourceLocB${h}, sourceLocA${h}) * ${t};
|
|
ivec4 inIdx = srcIdx;
|
|
vec4 bestIndex = vec4(inIdx);
|
|
vec4 bestValue = ${w};
|
|
|
|
for (int i = 0; i < ${t}; i++) {
|
|
inIdx = srcIdx;
|
|
${b}
|
|
vec4 candidate = ${w};
|
|
bvec4 nan = isnan(candidate);
|
|
bvec4 replace = bvec4(
|
|
vec4(${y}(candidate, bestValue)) * (vec4(1.0) - vec4(nan)));
|
|
|
|
bestValue = vec4(replace.x ? candidate.x : bestValue.x,
|
|
replace.y ? candidate.y : bestValue.y,
|
|
replace.z ? candidate.z : bestValue.z,
|
|
replace.w ? candidate.w : bestValue.w);
|
|
bestIndex = mix(bestIndex, vec4(inIdx), vec4(replace));
|
|
srcIdx++;
|
|
}
|
|
setOutput(bestIndex);
|
|
}
|
|
`}};function f4(e,t,n,s=null){let r=t.shape[0],a=t.shape[1];s!=null&&(r=s.shape[0],a=s.shape[1]);let o=E.computeOptimalWindowSize(a),i={windowSize:o,inSize:a,batchSize:r,outSize:Math.ceil(a/o)},l=new qQ(i,n,s==null),c=[t];s!=null&&c.push(s);let u=e.runWebGLProgram(l,c,"int32");if(u.shape[1]===1)return u;let d=f4(e,t,n,u);return e.disposeIntermediateTensorInfo(u),d}function m4(e,t,n,s=null){let r=s!=null?s.shape:t.shape,a=r[r.length-1],o=E.computeOptimalWindowSize(a),i=new XQ(r,o,n,s==null),l=s==null?[t]:[t,s],c=e.runWebGLProgram(i,l,"int32");if(c.shape.length===t.shape.length){let u=m4(e,t,n,c);return e.disposeIntermediateTensorInfo(c),u}return c}function g4(e,t,n,s){let r=[n];if(E.assertAxesAreInnerMostDims("arg"+s.charAt(0).toUpperCase()+s.slice(1),r,t.shape.length),!K().getBool("WEBGL_PACK_REDUCE")||t.shape.length<=2){let a=[],o=e.texData.get(t.dataId),i=o!==null&&o.isPacked,l=t;i&&(l=e.unpackTensor(t),a.push(l));let[c,u]=E.computeOutAndReduceShapes(l.shape,r),d=v.sizeFromShape(u),p=ve({inputs:{x:l},backend:e,attrs:{shape:[-1,d]}});a.push(p);let h=f4(e,p,s);a.push(h);let f=ve({inputs:{x:h},backend:e,attrs:{shape:c}});return a.forEach(m=>e.disposeIntermediateTensorInfo(m)),f}return m4(e,t,s)}function KQ(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a}=s,o=v.parseAxisParam(a,r.shape),i=E.getAxesPermutation(o,r.shape.length),l=r,c=[];i!=null&&(l=Hn({inputs:{x:r},backend:n,attrs:{perm:i}}),c.push(l),o=E.getInnerMostAxes(o.length,l.shape.length)),E.assertAxesAreInnerMostDims("argMax",[o[0]],l.shape.length);let u=g4(n,l,o[0],"max");return c.forEach(d=>n.disposeIntermediateTensorInfo(d)),u}var ZQ={kernelName:Ta,backendName:"webgl",kernelFunc:KQ};function YQ(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a}=s,o=v.parseAxisParam(a,r.shape),i=E.getAxesPermutation(o,r.shape.length),l=r,c=[];i!=null&&(l=Hn({inputs:{x:r},backend:n,attrs:{perm:i}}),c.push(l),o=E.getInnerMostAxes(o.length,l.shape.length)),E.assertAxesAreInnerMostDims("argMin",[o[0]],l.shape.length);let u=g4(n,l,o[0],"min");return c.forEach(d=>n.disposeIntermediateTensorInfo(d)),u}var JQ={kernelName:ru,backendName:"webgl",kernelFunc:YQ},QQ=Ar+`
|
|
if (abs(x) > 1.) {
|
|
return NAN;
|
|
}
|
|
return asin(x);
|
|
`,eee=st({opSnippet:QQ}),tee={kernelName:au,backendName:"webgl",kernelFunc:eee},nee=Ar+"return log(x + sqrt(x * x + 1.0));",see=st({opSnippet:nee}),ree={kernelName:ou,backendName:"webgl",kernelFunc:see},aee=Ar+`
|
|
return atan(x);
|
|
`,oee=st({opSnippet:aee}),iee={kernelName:iu,backendName:"webgl",kernelFunc:oee},lee=mQ+`
|
|
return atan(a, b);
|
|
`,uee=`
|
|
vec4 result = atan(a, b);
|
|
vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));
|
|
`+gQ+`
|
|
return result;
|
|
`,cee=Cn({opSnippet:lee,packedOpSnippet:uee}),dee={kernelName:uu,backendName:"webgl",kernelFunc:cee},pee=Ar+`
|
|
if ((x < -1.0) || (x > 1.0)) return NAN;
|
|
return (log(1.0 + x) - log(1.0 - x)) / 2.0;`,hee=st({opSnippet:pee}),fee={kernelName:lu,backendName:"webgl",kernelFunc:hee},fp=class{constructor(e,t,n,s=!1,r=!1){if(this.variableNames=["x"],t==="avg"&&n)throw new Error("Cannot compute positions for average pool.");let a=e.filterWidth,o=e.strideHeight,i=e.strideWidth,l=e.dilationHeight,c=e.dilationWidth,u=e.effectiveFilterHeight,d=e.effectiveFilterWidth,p=e.padInfo.top,h=e.padInfo.left;this.outputShape=e.outShape;let f=t==="avg",m=`((batch * ${e.inHeight} + xR) * ${e.inWidth} + xC) * ${e.inChannels} + d`,g=`(xR * ${e.inWidth} + xC) * ${e.inChannels} + d`,A="0.0";if(f||(A="-1.0 / 1e-20"),n){let C=">=";this.userCode=`
|
|
const ivec2 strides = ivec2(${o}, ${i});
|
|
const ivec2 pads = ivec2(${p}, ${h});
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int batch = coords[0];
|
|
int d = coords[3];
|
|
|
|
ivec2 xRCCorner = coords.yz * strides - pads;
|
|
int xRCorner = xRCCorner.x;
|
|
int xCCorner = xRCCorner.y;
|
|
|
|
// max/min x(?, ?, d) to get y(yR, yC, d).
|
|
// ? = to be determined
|
|
float minMaxValue = 0.0;
|
|
float minMaxValueFound = 0.0;
|
|
int minMaxPosition = 0;
|
|
float avgValue = 0.0;
|
|
|
|
for (int wR = 0; wR < ${u};
|
|
wR += ${l}) {
|
|
int xR = xRCorner + wR;
|
|
|
|
if (xR < 0 || xR >= ${e.inHeight}) {
|
|
continue;
|
|
}
|
|
|
|
for (int wC = 0; wC < ${d};
|
|
wC += ${c}) {
|
|
int xC = xCCorner + wC;
|
|
|
|
if (xC < 0 || xC >= ${e.inWidth}) {
|
|
continue;
|
|
}
|
|
|
|
float value = getX(batch, xR, xC, d);
|
|
|
|
// If a min / max value has already been found, use it. If not,
|
|
// use the current value.
|
|
float currMinMaxValue = mix(
|
|
value, minMaxValue, minMaxValueFound);
|
|
if (value ${C} currMinMaxValue) {
|
|
minMaxValue = value;
|
|
minMaxValueFound = 1.0;
|
|
minMaxPosition = ${s?r?m:g:`wR * ${d} + wC`};
|
|
}
|
|
}
|
|
}
|
|
setOutput(float(minMaxPosition));
|
|
}
|
|
`;return}let x="max",y=`${t}(${t}(${t}(minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])`;t==="avg"&&(y="avgValue / count");let b=Math.floor(a/4)*4,w=a%4,k=`
|
|
if (${f}) {
|
|
avgValue += dot(values, ones);
|
|
} else {
|
|
minMaxValue = ${x}(values, minMaxValue);
|
|
}
|
|
`;this.userCode=`
|
|
const ivec2 strides = ivec2(${o}, ${i});
|
|
const ivec2 pads = ivec2(${p}, ${h});
|
|
const float initializationValue = ${A};
|
|
const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);
|
|
|
|
float count = 0.0;
|
|
|
|
float getValue(int batch, int xR, int xC, int d) {
|
|
if (xC < 0 || xC >= ${e.inWidth}) {
|
|
return initializationValue;
|
|
}
|
|
count += 1.0;
|
|
return getX(batch, xR, xC, d);
|
|
}
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int batch = coords[0];
|
|
int d = coords[3];
|
|
|
|
ivec2 xRCCorner = coords.yz * strides - pads;
|
|
int xRCorner = xRCCorner.x;
|
|
int xCCorner = xRCCorner.y;
|
|
|
|
// max/min x(?, ?, d) to get y(yR, yC, d).
|
|
// ? = to be determined
|
|
vec4 minMaxValue = vec4(${A});
|
|
float avgValue = 0.0;
|
|
count = 0.0;
|
|
|
|
for (int wR = 0; wR < ${u};
|
|
wR += ${l}) {
|
|
int xR = xRCorner + wR;
|
|
|
|
if (xR < 0 || xR >= ${e.inHeight}) {
|
|
continue;
|
|
}
|
|
|
|
for (int wC = 0; wC < ${b}; wC += 4) {
|
|
int xC = xCCorner + wC * ${c};
|
|
|
|
vec4 values = vec4(
|
|
getValue(batch, xR, xC, d),
|
|
getValue(batch, xR, xC + ${c}, d),
|
|
getValue(batch, xR, xC + 2 * ${c}, d),
|
|
getValue(batch, xR, xC + 3 * ${c}, d)
|
|
);
|
|
|
|
${k}
|
|
}
|
|
|
|
int xC = xCCorner + ${b};
|
|
if (${w===1}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, xR, xC, d),
|
|
initializationValue,
|
|
initializationValue,
|
|
initializationValue
|
|
);
|
|
|
|
${k}
|
|
} else if (${w===2}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, xR, xC, d),
|
|
getValue(batch, xR, xC + ${c}, d),
|
|
initializationValue,
|
|
initializationValue
|
|
);
|
|
|
|
${k}
|
|
} else if (${w===3}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, xR, xC, d),
|
|
getValue(batch, xR, xC + ${c}, d),
|
|
getValue(batch, xR, xC + 2 * ${c}, d),
|
|
initializationValue
|
|
);
|
|
|
|
${k}
|
|
}
|
|
}
|
|
setOutput(${y});
|
|
}
|
|
`}},ux=class{constructor(e,t,n,s=!1,r=!1){if(this.variableNames=["x"],t==="avg"&&n)throw new Error("Cannot compute positions for average pool.");let a=e.filterWidth,o=e.strideDepth,i=e.strideHeight,l=e.strideWidth,c=e.dilationDepth,u=e.dilationHeight,d=e.dilationWidth,p=e.effectiveFilterDepth,h=e.effectiveFilterHeight,f=e.effectiveFilterWidth,m=e.padInfo.front,g=e.padInfo.top,A=e.padInfo.left;this.outputShape=e.outShape;let x=t==="avg",y="0.0";if(x||(y="-1.0 / 1e-20"),n){let R=">=";this.userCode=`
|
|
const ivec3 strides =
|
|
ivec3(${o}, ${i}, ${l});
|
|
const ivec3 pads = ivec3(${m}, ${g}, ${A});
|
|
|
|
void main() {
|
|
ivec5 coords = getOutputCoords();
|
|
int batch = coords.x;
|
|
int ch = coords.u;
|
|
|
|
ivec3 xCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;
|
|
int xDCorner = xCorner.x;
|
|
int xRCorner = xCorner.y;
|
|
int xCCorner = xCorner.z;
|
|
|
|
// max/min x(?, ?, ?, ch) to get y(yD, yR, yC, ch).
|
|
// ? = to be determined
|
|
float minMaxValue = 0.0;
|
|
float minMaxValueFound = 0.0;
|
|
int minMaxPosition = 0;
|
|
|
|
for (int wD = 0; wD < ${p};
|
|
wD += ${c}) {
|
|
int xD = xDCorner + wD;
|
|
|
|
if (xD < 0 || xD >= ${e.inDepth}) {
|
|
continue;
|
|
}
|
|
|
|
for (int wR = 0; wR < ${h};
|
|
wR += ${u}) {
|
|
int xR = xRCorner + wR;
|
|
|
|
if (xR < 0 || xR >= ${e.inHeight}) {
|
|
continue;
|
|
}
|
|
|
|
for (int wC = 0; wC < ${f};
|
|
wC += ${d}) {
|
|
int xC = xCCorner + wC;
|
|
|
|
if (xC < 0 || xC >= ${e.inWidth}) {
|
|
continue;
|
|
}
|
|
|
|
float value = getX(batch, xD, xR, xC, ch);
|
|
|
|
// If a min / max value has already been found, use it. If not,
|
|
// use the current value.
|
|
float currMinMaxValue = mix(
|
|
value, minMaxValue, minMaxValueFound);
|
|
if (value ${R} currMinMaxValue) {
|
|
minMaxValue = value;
|
|
minMaxValueFound = 1.0;
|
|
minMaxPosition = ${s?r?`(((batch * ${e.inDepth} + xD) * ${e.inHeight} + xR) * ${e.inWidth} + xC) * ${e.inChannels} + ch`:`((xD * ${e.inHeight} + xR) * ${e.inWidth} + xC) * ${e.inChannels} + ch`:`wD * ${h} * ${f} +
|
|
wR * ${f} + wC`};
|
|
}
|
|
}
|
|
}
|
|
}
|
|
setOutput(float(minMaxPosition));
|
|
}
|
|
`;return}let b="max",w=`${t}(${t}(${t}(minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])`;t==="avg"&&(w="avgValue / count");let k=Math.floor(a/4)*4,C=a%4,N=`
|
|
if (${x}) {
|
|
avgValue += dot(values, ones);
|
|
} else {
|
|
minMaxValue = ${b}(values, minMaxValue);
|
|
}
|
|
`;this.userCode=`
|
|
const ivec3 strides =
|
|
ivec3(${o}, ${i}, ${l});
|
|
const ivec3 pads = ivec3(${m}, ${g}, ${A});
|
|
const float initializationValue = ${y};
|
|
const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);
|
|
|
|
float count = 0.0;
|
|
|
|
float getValue(int batch, int xD, int xR, int xC, int ch) {
|
|
if (xC < 0 || xC >= ${e.inWidth}) {
|
|
return initializationValue;
|
|
}
|
|
count += 1.0;
|
|
return getX(batch, xD, xR, xC, ch);
|
|
}
|
|
|
|
void main() {
|
|
ivec5 coords = getOutputCoords();
|
|
int batch = coords.x;
|
|
int ch = coords.u;
|
|
|
|
ivec3 xCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;
|
|
int xDCorner = xCorner.x;
|
|
int xRCorner = xCorner.y;
|
|
int xCCorner = xCorner.z;
|
|
|
|
// max/min x(?, ?, ?, d) to get y(yD, yR, yC, ch).
|
|
// ? = to be determined
|
|
vec4 minMaxValue = vec4(${y});
|
|
float avgValue = 0.0;
|
|
count = 0.0;
|
|
|
|
for (int wD = 0; wD < ${p};
|
|
wD += ${c}) {
|
|
int xD = xDCorner + wD;
|
|
|
|
if (xD < 0 || xD >= ${e.inDepth}) {
|
|
continue;
|
|
}
|
|
|
|
for (int wR = 0; wR < ${h};
|
|
wR += ${u}) {
|
|
int xR = xRCorner + wR;
|
|
|
|
if (xR < 0 || xR >= ${e.inHeight}) {
|
|
continue;
|
|
}
|
|
|
|
for (int wC = 0; wC < ${k}; wC += 4) {
|
|
int xC = xCCorner + wC * ${d};
|
|
|
|
vec4 values = vec4(
|
|
getValue(batch, xD, xR, xC, ch),
|
|
getValue(batch, xD, xR, xC + ${d}, ch),
|
|
getValue(batch, xD, xR, xC + 2 * ${d}, ch),
|
|
getValue(batch, xD, xR, xC + 3 * ${d}, ch)
|
|
);
|
|
|
|
${N}
|
|
}
|
|
|
|
int xC = xCCorner + ${k};
|
|
if (${C===1}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, xD, xR, xC, ch),
|
|
initializationValue,
|
|
initializationValue,
|
|
initializationValue
|
|
);
|
|
|
|
${N}
|
|
} else if (${C===2}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, xD, xR, xC, ch),
|
|
getValue(batch, xD, xR, xC + ${d}, ch),
|
|
initializationValue,
|
|
initializationValue
|
|
);
|
|
|
|
${N}
|
|
} else if (${C===3}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, xD, xR, xC, ch),
|
|
getValue(batch, xD, xR, xC + ${d}, ch),
|
|
getValue(batch, xD, xR, xC + 2 * ${d}, ch),
|
|
initializationValue
|
|
);
|
|
|
|
${N}
|
|
}
|
|
}
|
|
setOutput(${w});
|
|
}
|
|
}
|
|
`}};function mee(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t;nc(r,"avgPool");let{filterSize:a,strides:o,pad:i,dimRoundingMode:l}=s,c=1;v.assert(E.eitherStridesOrDilationsAreOne(o,c),()=>`Error in avgPool: Either strides or dilations must be 1. Got strides ${o} and dilations '${c}'`);let u=E.computePool2DInfo(r.shape,a,o,c,i,l);if(u.filterWidth===1&&u.filterHeight===1&&v.arraysEqual(u.inShape,u.outShape))return As({inputs:{x:r},backend:n});let d=new fp(u,"avg",!1);return n.runWebGLProgram(d,[r],"float32")}var gee={kernelName:Na,backendName:"webgl",kernelFunc:mee};function Aee(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{filterSize:a,strides:o,pad:i,dimRoundingMode:l,dataFormat:c}=s,u=[1,1,1],d=E.computePool3DInfo(r.shape,a,o,u,i,l,c),p=new ux(d,"avg",!1);return n.runWebGLProgram(p,[r],"float32")}var yee={kernelName:Yc,backendName:"webgl",kernelFunc:Aee},xee=class{constructor(e){this.variableNames=["dy"],this.outputShape=e.inShape;let t=e.filterHeight,n=e.filterWidth,s=e.strideHeight,r=e.strideWidth,a=e.dilationHeight,o=e.dilationWidth,i=e.effectiveFilterHeight,l=e.effectiveFilterWidth,c=i-1-e.padInfo.top,u=l-1-e.padInfo.left,d=1/(t*n);this.userCode=`
|
|
const ivec2 pads = ivec2(${c}, ${u});
|
|
const float avgMultiplier = float(${d});
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int b = coords[0];
|
|
int d = coords[3];
|
|
|
|
ivec2 dyRCCorner = coords.yz - pads;
|
|
int dyRCorner = dyRCCorner.x;
|
|
int dyCCorner = dyRCCorner.y;
|
|
|
|
// Convolve dy(?, ?, d) with pos mask(:, :, d) to get dx(xR, xC, d).
|
|
// ? = to be determined. : = across all values in that axis.
|
|
float dotProd = 0.0;
|
|
for (int wR = 0; wR < ${i};
|
|
wR += ${a}) {
|
|
float dyR = float(dyRCorner + wR) / ${s}.0;
|
|
|
|
if (dyR < 0.0 || dyR >= ${e.outHeight}.0 || fract(dyR) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyR = int(dyR);
|
|
|
|
for (int wC = 0; wC < ${l};
|
|
wC+= ${o}) {
|
|
float dyC = float(dyCCorner + wC) / ${r}.0;
|
|
|
|
if (dyC < 0.0 || dyC >= ${e.outWidth}.0 ||
|
|
fract(dyC) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyC = int(dyC);
|
|
|
|
float dyValue = getDy(b, idyR, idyC, d);
|
|
|
|
dotProd += dyValue * avgMultiplier;
|
|
}
|
|
}
|
|
setOutput(dotProd);
|
|
}
|
|
`}},bee=class{constructor(e){this.variableNames=["dy"],this.outputShape=e.inShape;let t=e.filterDepth,n=e.filterHeight,s=e.filterWidth,r=e.strideDepth,a=e.strideHeight,o=e.strideWidth,i=e.dilationDepth,l=e.dilationHeight,c=e.dilationWidth,u=e.effectiveFilterDepth,d=e.effectiveFilterHeight,p=e.effectiveFilterWidth,h=u-1-e.padInfo.front,f=d-1-e.padInfo.top,m=p-1-e.padInfo.left,g=1/(t*n*s);this.userCode=`
|
|
const ivec3 pads = ivec3(${h}, ${f}, ${m});
|
|
const float avgMultiplier = float(${g});
|
|
|
|
void main() {
|
|
ivec5 coords = getOutputCoords();
|
|
int batch = coords.x;
|
|
int ch = coords.u;
|
|
|
|
ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;
|
|
int dyDCorner = dyCorner.x;
|
|
int dyRCorner = dyCorner.y;
|
|
int dyCCorner = dyCorner.z;
|
|
|
|
// Convolve dy(?, ?, ?, d) with pos mask(:, :, :, ch) to get
|
|
// dx(xD, xR, xC, ch).
|
|
// ? = to be determined. : = across all values in that axis.
|
|
float dotProd = 0.0;
|
|
|
|
for (int wD = 0; wD < ${u};
|
|
wD += ${i}) {
|
|
float dyD = float(dyDCorner + wD) / ${r}.0;
|
|
|
|
if (dyD < 0.0 || dyD >= ${e.outDepth}.0 || fract(dyD) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyD = int(dyD);
|
|
|
|
for (int wR = 0; wR < ${d};
|
|
wR += ${l}) {
|
|
float dyR = float(dyRCorner + wR) / ${a}.0;
|
|
|
|
if (dyR < 0.0 || dyR >= ${e.outHeight}.0 ||
|
|
fract(dyR) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyR = int(dyR);
|
|
|
|
for (int wC = 0; wC < ${p};
|
|
wC += ${c}) {
|
|
float dyC = float(dyCCorner + wC) / ${o}.0;
|
|
|
|
if (dyC < 0.0 || dyC >= ${e.outWidth}.0 ||
|
|
fract(dyC) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyC = int(dyC);
|
|
|
|
float dyValue = getDy(batch, idyD, idyR, idyC, ch);
|
|
|
|
dotProd += dyValue * avgMultiplier;
|
|
}
|
|
}
|
|
}
|
|
setOutput(dotProd);
|
|
}
|
|
`}};function vee(e){let{inputs:t,backend:n,attrs:s}=e,{dy:r,input:a}=t,o=a,{filterSize:i,strides:l,pad:c,dimRoundingMode:u}=s,d=[1,1,1],p=E.computePool3DInfo(o.shape,i,l,d,c,u),h=new bee(p);return n.runWebGLProgram(h,[r],o.dtype)}var wee={kernelName:gh,backendName:"webgl",kernelFunc:vee};function kee(e){let{inputs:t,backend:n,attrs:s}=e,{dy:r,input:a}=t,o=a;nc([r,a],"avgPoolGrad");let{filterSize:i,strides:l,pad:c}=s,u=E.computePool2DInfo(o.shape,i,l,1,c),d=new xee(u);return n.runWebGLProgram(d,[r],o.dtype)}var See={kernelName:mh,backendName:"webgl",kernelFunc:kee};function Iee(e){let{inputs:t,backend:n,attrs:s}=e,{a:r,b:a}=t,{transposeA:o,transposeB:i}=s;return Vm({a:r,b:a,transposeA:o,transposeB:i,backend:n})}var Cee={kernelName:Ea,backendName:"webgl",kernelFunc:Iee},Tee=class{constructor(e,t,n,s,r,a){this.outputShape=[],this.variableNames=["x","mean","variance"],E.assertAndGetBroadcastShape(e,t),E.assertAndGetBroadcastShape(e,n);let o="0.0";s!=null&&(E.assertAndGetBroadcastShape(e,s),this.variableNames.push("offset"),o="getOffsetAtOutCoords()");let i="1.0";r!=null&&(E.assertAndGetBroadcastShape(e,r),this.variableNames.push("scale"),i="getScaleAtOutCoords()"),this.outputShape=e,this.userCode=`
|
|
void main() {
|
|
float x = getXAtOutCoords();
|
|
float mean = getMeanAtOutCoords();
|
|
float variance = getVarianceAtOutCoords();
|
|
float offset = ${o};
|
|
float scale = ${i};
|
|
float inv = scale * inversesqrt(variance + float(${a}));
|
|
setOutput(dot(vec3(x, -mean, offset), vec3(inv, inv, 1)));
|
|
}
|
|
`}},Nee=class{constructor(e,t,n,s,r,a){this.packedInputs=!0,this.packedOutput=!0,this.variableNames=["x","mean","variance"],E.assertAndGetBroadcastShape(e,t),E.assertAndGetBroadcastShape(e,n);let o="vec4(0.0)";s!=null&&(E.assertAndGetBroadcastShape(e,s),this.variableNames.push("offset"),o="getOffsetAtOutCoords()");let i="vec4(1.0)";r!=null&&(E.assertAndGetBroadcastShape(e,r),this.variableNames.push("scale"),i="getScaleAtOutCoords()"),this.outputShape=e,this.userCode=`
|
|
void main() {
|
|
vec4 offset = ${o};
|
|
vec4 scale = ${i};
|
|
|
|
vec4 x = getXAtOutCoords();
|
|
vec4 mean = getMeanAtOutCoords();
|
|
vec4 variance = getVarianceAtOutCoords();
|
|
|
|
vec4 inv = scale * inversesqrt(variance + vec4(${a}));
|
|
|
|
setOutput((x - mean) * inv + offset);
|
|
}
|
|
`}},Eee=({inputs:e,backend:t,attrs:n})=>{let{x:s,mean:r,variance:a,offset:o,scale:i}=e;v.assert(r.shape.length===a.shape.length,()=>"Batch normalization gradient requires mean and variance to have equal ranks."),v.assert(o==null||r.shape.length===o.shape.length,()=>"Batch normalization gradient requires mean and offset to have equal ranks."),v.assert(i==null||r.shape.length===i.shape.length,()=>"Batch normalization gradient requires mean and scale to have equal ranks.");let{varianceEpsilon:l}=n;l==null&&(l=.001);let c=[s,r,a],u=null;o!=null&&(u=o.shape,c.push(o));let d=null;i!=null&&(d=i.shape,c.push(i));let p=K().getBool("WEBGL_PACK_NORMALIZATION")?new Nee(s.shape,r.shape,a.shape,u,d,l):new Tee(s.shape,r.shape,a.shape,u,d,l);return t.runWebGLProgram(p,c,c[0].dtype)},Ree={kernelName:Va,backendName:"webgl",kernelFunc:Eee},$ee=class{constructor(e){this.variableNames=["source"],this.outputShape=e,this.rank=e.length;let t=yt(this.rank);this.customUniforms=[{name:"start",arrayIndex:this.rank,type:"int"}];let n=_ee(this.rank),s,r=e.map((a,o)=>`sourceLoc.${cx[o]} = start[${o}] + coords.${cx[o]};`);s=`
|
|
${t} sourceLoc;
|
|
${t} coords = getOutputCoords();
|
|
${r.join(`
|
|
`)}
|
|
`,this.userCode=`
|
|
void main() {
|
|
${s}
|
|
setOutput(getSource(${n}));
|
|
}
|
|
`}},cx=["x","y","z","w","u","v"];function _ee(e){if(e===1)return"sourceLoc";if(e<=6)return cx.slice(0,e).map(t=>"sourceLoc."+t).join(",");throw Error(`Slicing for rank ${e} is not yet supported`)}var Dee=class{constructor(e){this.variableNames=["source"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e,this.rank=e.length,this.customUniforms=[{name:"start",arrayIndex:this.rank,type:"int"}];let t=yt(this.rank),n=Gn("coords",this.rank),s=Gn("sourceLoc",this.rank),r=this.rank===1?"sourceLoc":`vec2(${s.slice(-2).join()})`,a=`getChannel(getSource(${s.join()}), ${r})`,o=`
|
|
result.x = ${a};
|
|
if (++${n[this.rank-1]} < ${e[this.rank-1]}) {
|
|
++${s[this.rank-1]};
|
|
result.y = ${a};
|
|
--${s[this.rank-1]};
|
|
}
|
|
`,i=this.rank===1?"":`
|
|
--${n[this.rank-1]};
|
|
if (++${n[this.rank-2]} < ${e[this.rank-2]}) {
|
|
++${s[this.rank-2]};
|
|
result.z = ${a};
|
|
if (++${n[this.rank-1]} < ${e[this.rank-1]}) {
|
|
++${s[this.rank-1]};
|
|
result.w = ${a};
|
|
}
|
|
}
|
|
`,l=this.rank<=4?`sourceLoc = coords +
|
|
${t}(${e.map((c,u)=>`start[${u}]`).join()});`:e.map((c,u)=>`${s[u]} = ${n[u]} + start[${u}];`).join(`
|
|
`);this.userCode=`
|
|
void main() {
|
|
${t} coords = getOutputCoords();
|
|
${t} sourceLoc;
|
|
${l}
|
|
vec4 result = vec4(0.);
|
|
${o}
|
|
${i}
|
|
setOutput(result);
|
|
}
|
|
`}};function Pee(e,t,n,s){let r=s.texData.get(e.dataId),a=s.makeTensorInfo(n,e.dtype),o=s.texData.get(a.dataId);Object.assign(o,r),o.refCount=1,o.shape=n,o.dtype=e.dtype;let i=Pt.computeFlatOffset(t,v.computeStrides(e.shape));r.slice&&(i+=r.slice.flatOffset),o.slice={flatOffset:i,origDataId:r.slice&&r.slice.origDataId||e.dataId};let l=s.dataRefCount.get(o.slice.origDataId)||1;return s.dataRefCount.set(o.slice.origDataId,l+1),a}function uc(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{begin:a,size:o}=s,[i,l]=Pt.parseSliceParams(r,a,o);if(Pt.assertParamsValid(r,i,l),v.sizeFromShape(l)===0)return n.makeTensorInfo(l,r.dtype,[]);if(n.shouldExecuteOnCPU([r])||r.dtype==="string"){let d=n.texData.get(r.dataId),p=xJ(d.values,i,l,r.shape,r.dtype);return n.makeTensorInfo(l,r.dtype,p)}let{isPacked:c}=n.texData.get(r.dataId),u=Pt.isSliceContinous(r.shape,i,l);if(c||!u){let d=K().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new Dee(l):new $ee(l),p=[i];return n.runWebGLProgram(d,[r],r.dtype,p)}return n.uploadToGPU(r.dataId),Pee(r,i,l,n)}var Fee={kernelName:Bi,backendName:"webgl",kernelFunc:uc},Oee=e=>{let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{blockShape:a,crops:o}=s;v.assert(r.shape.length<=4,()=>"batchToSpaceND for rank > 4 with a WebGL backend not implemented yet");let i=a.reduce((x,y)=>x*y),l=E.getReshaped(r.shape,a,i),c=E.getPermuted(l.length,a.length),u=E.getReshapedPermuted(r.shape,a,i),d=E.getSliceBeginCoords(o,a.length),p=E.getSliceSize(u,o,a.length),h=[],f=ve({inputs:{x:r},backend:n,attrs:{shape:l}}),m=Hn({inputs:{x:f},backend:n,attrs:{perm:c}}),g=ve({inputs:{x:m},backend:n,attrs:{shape:u}}),A=uc({inputs:{x:g},backend:n,attrs:{begin:d,size:p}});return h.push(f),h.push(m),h.push(g),h.forEach(x=>n.disposeIntermediateTensorInfo(x)),A},Mee={kernelName:di,backendName:"webgl",kernelFunc:Oee};function zee(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,weights:a}=t,{size:o}=s,i=n.readSync(r.dataId),l=n.readSync(a.dataId),c=UI(i,l,a.dtype,a.shape,o);return n.makeTensorInfo([o],a.dtype,c)}var Lee={kernelName:Ah,backendName:"webgl",kernelFunc:zee};function Bee(e){let{inputs:t,backend:n}=e,{s0:s,s1:r}=t,a=n.readSync(s.dataId),o=n.readSync(r.dataId),i=E.assertAndGetBroadcastShape(Array.from(a),Array.from(o));return n.makeTensorInfo([i.length],"int32",Int32Array.from(i))}var Wee={kernelName:yh,backendName:"webgl",kernelFunc:Bee},Vee="return float(a != b);",A4=Cn({opSnippet:Vee,cpuKernelImpl:fJ,dtype:"bool"}),Uee={kernelName:Ni,backendName:"webgl",kernelFunc:A4};function mp(e){let{inputs:t,backend:n}=e,{input:s}=t,r=n.texData.get(s.dataId);return As({inputs:{x:r.complexTensorInfos.real},backend:n})}var Gee={kernelName:id,backendName:"webgl",kernelFunc:mp},Hee="return float(int(x));";function jee(e,t){let n=new Vo(e.shape,Hee),s=t.runWebGLProgram(n,[e],"int32");return{dataId:s.dataId,shape:s.shape,dtype:s.dtype}}function dx(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{dtype:a}=s;if(a==="complex64"){if(r.dtype==="complex64")return As({inputs:{x:r},backend:n});let o=Gt(r.shape),i=dx({inputs:{x:r},backend:n,attrs:{dtype:"float32"}}),l=Uo({inputs:{real:i,imag:o},backend:n});return o.dispose(),n.disposeIntermediateTensorInfo(i),l}if(r.dtype==="complex64"){let o=mp({inputs:{input:r},backend:n}),i=dx({inputs:{x:o},backend:n,attrs:{dtype:a}});return n.disposeIntermediateTensorInfo(o),i}if(!v.hasEncodingLoss(r.dtype,a)){let o=As({inputs:{x:r},backend:n});return{dataId:o.dataId,shape:o.shape,dtype:a}}if(a==="int32")return jee(r,n);if(a==="bool"){let o=n.makeTensorInfo([],"bool",v.getTypedArrayFromDType("bool",1)),l=A4({inputs:{a:r,b:o},backend:n});return n.disposeIntermediateTensorInfo(o),l}throw new Error(`Error in Cast: failed to cast ${r.dtype} to ${a}`)}var qee={kernelName:Ra,backendName:"webgl",kernelFunc:dx},y4="return ceil(x);",Xee=st({opSnippet:y4,packedOpSnippet:y4,cpuKernelImpl:KY}),Kee={kernelName:$a,backendName:"webgl",kernelFunc:Xee},Zee=class{constructor(e){this.variableNames=["A"],this.customUniforms=[{name:"minVal",type:"float"},{name:"maxVal",type:"float"}],this.outputShape=e,this.userCode=`
|
|
|
|
void main() {
|
|
float value = getAAtOutCoords();
|
|
if (isnan(value)) {
|
|
setOutput(value);
|
|
return;
|
|
}
|
|
|
|
setOutput(clamp(value, minVal, maxVal));
|
|
}
|
|
`}},Yee=class{constructor(e){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.customUniforms=[{name:"minVal",type:"float"},{name:"maxVal",type:"float"}],this.outputShape=e,this.userCode=`
|
|
void main() {
|
|
vec4 value = getAAtOutCoords();
|
|
|
|
if (any(isnan(value))) {
|
|
setOutput(value);
|
|
return;
|
|
}
|
|
|
|
setOutput(clamp(value, vec4(minVal), vec4(maxVal)));
|
|
}
|
|
`}};function Jee(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{clipValueMin:a,clipValueMax:o}=s,i;K().getBool("WEBGL_PACK_CLIP")?i=new Yee(r.shape):i=new Zee(r.shape);let l=[[a],[o]];return n.runWebGLProgram(i,[r],r.dtype,l)}var Qee={kernelName:Hr,backendName:"webgl",kernelFunc:Jee},ete=class{constructor(e){this.variableNames=["real","imag"],this.outputShape=e,this.userCode=`
|
|
void main() {
|
|
float re = abs(getRealAtOutCoords());
|
|
float im = abs(getImagAtOutCoords());
|
|
float mx = max(re, im);
|
|
|
|
// sadly the length function in glsl is not underflow-safe
|
|
// (at least not on Intel GPUs). So the safe solution is
|
|
// to ensure underflow-safety in all cases.
|
|
setOutput(
|
|
mx == 0.0 ? 0.0 : mx * length(vec2(1, min(re, im)/mx))
|
|
);
|
|
}
|
|
`}};function x4(e,t){return{dataId:t.dataId,dtype:t.dtype,shape:e.shape}}function tte(e){let{inputs:t,backend:n}=e,{x:s}=t,r=n.texData.get(s.dataId),a=new ete(s.shape),o=[x4(s,r.complexTensorInfos.real),x4(s,r.complexTensorInfos.imag)];return n.runWebGLProgram(a,o,o[0].dtype)}var nte={kernelName:Qc,backendName:"webgl",kernelFunc:tte},ste=class{constructor(e){this.outputShape=[],this.outputShape=E.computeOutShape(e,1),this.variableNames=e.map((a,o)=>`T${o}`);let t=new Array(e.length-1);t[0]=e[0][1];for(let a=1;a<t.length;a++)t[a]=t[a-1]+e[a][1];let n=[`if (yC < ${t[0]}) setOutput(getT0(yR, yC));`];for(let a=1;a<t.length;a++){let o=t[a-1];n.push(`else if (yC < ${t[a]}) setOutput(getT${a}(yR, yC-${o}));`)}let s=t.length,r=t[t.length-1];n.push(`else setOutput(getT${s}(yR, yC-${r}));`),this.userCode=`
|
|
void main() {
|
|
ivec2 coords = getOutputCoords();
|
|
int yR = coords.x;
|
|
int yC = coords.y;
|
|
|
|
${n.join(`
|
|
`)}
|
|
}
|
|
`}},rte=class{constructor(e,t){this.packedInputs=!0,this.packedOutput=!0,this.outputShape=[],this.outputShape=E.computeOutShape(e,t);let n=this.outputShape,s=n.length,r=yt(s),a=Gn("coords",s),o=["x","y","z","w","u","v"].slice(0,s);this.variableNames=e.map((f,m)=>`T${m}`);let i=new Array(e.length-1);i[0]=e[0][t];for(let f=1;f<i.length;f++)i[f]=i[f-1]+e[f][t];let l=o[t],c=o.slice(-2),u=o.join(),d=`if (${l} < ${i[0]}) {
|
|
return getChannel(
|
|
getT0(${u}), vec2(${c.join()}));
|
|
}`;for(let f=1;f<i.length;f++){let m=i[f-1];d+=`
|
|
if (${l} < ${i[f]} && ${l} >= ${i[f-1]}) {
|
|
return getChannel(
|
|
getT${f}(${Gm(o,l,m)}),
|
|
vec2(${Gm(c,l,m)}));
|
|
}`}let p=i.length,h=i[i.length-1];d+=`
|
|
return getChannel(
|
|
getT${p}(${Gm(o,l,h)}),
|
|
vec2(${Gm(c,l,h)}));`,this.userCode=`
|
|
float getValue(${o.map(f=>"int "+f)}) {
|
|
${d}
|
|
}
|
|
|
|
void main() {
|
|
${r} coords = getOutputCoords();
|
|
vec4 result = vec4(getValue(${a}), 0., 0., 0.);
|
|
|
|
${a[s-1]} = ${a[s-1]} + 1;
|
|
if (${a[s-1]} < ${n[s-1]}) {
|
|
result.g = getValue(${a});
|
|
}
|
|
|
|
${a[s-2]} = ${a[s-2]} + 1;
|
|
if (${a[s-2]} < ${n[s-2]}) {
|
|
result.a = getValue(${a});
|
|
}
|
|
|
|
${a[s-1]} = ${a[s-1]} - 1;
|
|
if (${a[s-2]} < ${n[s-2]} &&
|
|
${a[s-1]} < ${n[s-1]}) {
|
|
result.b = getValue(${a});
|
|
}
|
|
setOutput(result);
|
|
}
|
|
`}};function Gm(e,t,n){let s=e.indexOf(t);return e.map((a,o)=>o===s?`${a} - ${n}`:a).join()}function Hm(e){let{inputs:t,backend:n}=e,{input:s}=t,r=n.texData.get(s.dataId);return As({inputs:{x:r.complexTensorInfos.imag},backend:n})}var ate={kernelName:sd,backendName:"webgl",kernelFunc:Hm};function cc(e,t,n){let s=e[0].dtype;if(s==="complex64"){let u=e.map(m=>mp({inputs:{input:m},backend:n})),d=e.map(m=>Hm({inputs:{input:m},backend:n})),p=cc(u,t,n),h=cc(d,t,n),f=Uo({inputs:{real:p,imag:h},backend:n});return u.forEach(m=>n.disposeIntermediateTensorInfo(m)),d.forEach(m=>n.disposeIntermediateTensorInfo(m)),n.disposeIntermediateTensorInfo(p),n.disposeIntermediateTensorInfo(h),f}let r=n.shouldExecuteOnCPU(e);if(s==="string"&&(r=!0),r){let u=e.map(A=>{let x=v.sizeFromShape(A.shape.slice(t));return ve({inputs:{x:A},backend:n,attrs:{shape:[-1,x]}})}),d=u.map(A=>({vals:n.readSync(A.dataId),shape:A.shape})),p=E.computeOutShape(u.map(A=>A.shape),1),h=u[0].shape[0]===1,f=ZY(d,p,s,h),m=E.computeOutShape(e.map(A=>A.shape),t),g=n.makeTensorInfo(m,s,f);return u.forEach(A=>n.disposeIntermediateTensorInfo(A)),g}if(e.length>K().getNumber("WEBGL_MAX_TEXTURES_IN_SHADER")){let u=Math.floor(e.length/2),d=cc(e.slice(0,u),t,n),p=cc(e.slice(u),t,n),h=cc([d,p],t,n);return n.disposeIntermediateTensorInfo(d),n.disposeIntermediateTensorInfo(p),h}if(K().getBool("WEBGL_PACK_ARRAY_OPERATIONS")&&e[0].shape.length>1){let u=new rte(e.map(d=>d.shape),t);return n.runWebGLProgram(u,e,s)}let{tensors2D:a,outShape:o}=ote(e,t,n),i=new ste(a.map(u=>u.shape)),l=n.runWebGLProgram(i,a,s);a.forEach(u=>n.disposeIntermediateTensorInfo(u));let c=ve({inputs:{x:l},attrs:{shape:o},backend:n});return n.disposeIntermediateTensorInfo(l),c}function ote(e,t,n){let s=E.computeOutShape(e.map(a=>a.shape),t);return{tensors2D:e.map(a=>ve({inputs:{x:a},attrs:{shape:[-1,v.sizeFromShape(a.shape.slice(t))]},backend:n})),outShape:s}}function b4(e){let{inputs:t,backend:n,attrs:s}=e,{axis:r}=s,a=v.parseAxisParam(r,t[0].shape)[0],o=E.computeOutShape(t.map(c=>c.shape),a);if(v.sizeFromShape(o)===0)return n.makeTensorInfo(o,t[0].dtype,[]);let i=t.filter(c=>v.sizeFromShape(c.shape)>0);if(i.length===1)return As({inputs:{x:i[0]},backend:n});let l=i.map(c=>c.shape);return E.assertParamsConsistent(l,a),cc(i,a,n)}var ite={kernelName:pi,backendName:"webgl",kernelFunc:b4},v4=class{constructor(e,t=!1,n=null,s=!1,r=!1){this.variableNames=["x","W"],this.outputShape=e.outShape;let a=e.padInfo.top,o=e.padInfo.left,i=e.strideHeight,l=e.strideWidth,c=e.dilationHeight,u=e.dilationWidth,d=e.filterHeight,p=e.filterWidth,h=Math.floor(e.inChannels/4)*4,f=e.inChannels%4,m=e.dataFormat==="channelsLast",g=m?1:2,A=m?2:3,x=m?3:1,y="",b="";n&&(s?y=`float activation(float a) {
|
|
float b = getPreluActivationWeightsAtOutCoords();
|
|
${n}
|
|
}`:r?y=`float activation(float a) {
|
|
float b = getLeakyreluAlphaAtOutCoords();
|
|
${n}
|
|
}`:y=`
|
|
float activation(float x) {
|
|
${n}
|
|
}
|
|
`,b="result = activation(result);");let w=t?"result += getBiasAtOutCoords();":"";t&&this.variableNames.push("bias"),s&&this.variableNames.push("preluActivationWeights"),r&&this.variableNames.push("leakyreluAlpha"),this.userCode=`
|
|
${y}
|
|
|
|
const ivec2 strides = ivec2(${i}, ${l});
|
|
const ivec2 pads = ivec2(${a}, ${o});
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int batch = coords[0];
|
|
int d2 = coords[${x}];
|
|
|
|
ivec2 xRCCorner =
|
|
ivec2(coords[${g}], coords[${A}]) * strides - pads;
|
|
int xRCorner = xRCCorner.x;
|
|
int xCCorner = xRCCorner.y;
|
|
|
|
// Convolve x(?, ?, d1) with w(:, :, d1, d2) to get y(yR, yC, d2).
|
|
// ? = to be determined. : = across all values in that axis.
|
|
float dotProd = 0.0;
|
|
for (int wR = 0; wR < ${d}; wR++) {
|
|
int xR = xRCorner + wR * ${c};
|
|
|
|
if (xR < 0 || xR >= ${e.inHeight}) {
|
|
continue;
|
|
}
|
|
|
|
for (int wC = 0; wC < ${p}; wC++) {
|
|
int xC = xCCorner + wC * ${u};
|
|
|
|
if (xC < 0 || xC >= ${e.inWidth}) {
|
|
continue;
|
|
}
|
|
|
|
for (int d1 = 0; d1 < ${h}; d1 += 4) {
|
|
vec4 wValues = vec4(
|
|
getW(wR, wC, d1, d2),
|
|
getW(wR, wC, d1 + 1, d2),
|
|
getW(wR, wC, d1 + 2, d2),
|
|
getW(wR, wC, d1 + 3, d2)
|
|
);
|
|
|
|
if (${m}) {
|
|
vec4 xValues = vec4(
|
|
getX(batch, xR, xC, d1),
|
|
getX(batch, xR, xC, d1 + 1),
|
|
getX(batch, xR, xC, d1 + 2),
|
|
getX(batch, xR, xC, d1 + 3)
|
|
);
|
|
dotProd += dot(xValues, wValues);
|
|
} else {
|
|
vec4 xValues = vec4(
|
|
getX(batch, d1, xR, xC),
|
|
getX(batch, d1 + 1, xR, xC),
|
|
getX(batch, d1 + 2, xR, xC),
|
|
getX(batch, d1 + 3, xR, xC)
|
|
);
|
|
dotProd += dot(xValues, wValues);
|
|
}
|
|
}
|
|
|
|
if (${f===1}) {
|
|
|
|
if (${m}) {
|
|
dotProd +=
|
|
getX(batch, xR, xC, ${h}) *
|
|
getW(wR, wC, ${h}, d2);
|
|
} else {
|
|
dotProd +=
|
|
getX(batch, ${h}, xR, xC) *
|
|
getW(wR, wC, ${h}, d2);
|
|
}
|
|
|
|
} else if (${f===2}) {
|
|
vec2 wValues = vec2(
|
|
getW(wR, wC, ${h}, d2),
|
|
getW(wR, wC, ${h} + 1, d2)
|
|
);
|
|
|
|
if (${m}) {
|
|
vec2 xValues = vec2(
|
|
getX(batch, xR, xC, ${h}),
|
|
getX(batch, xR, xC, ${h} + 1)
|
|
);
|
|
dotProd += dot(xValues, wValues);
|
|
} else {
|
|
vec2 xValues = vec2(
|
|
getX(batch, ${h}, xR, xC),
|
|
getX(batch, ${h} + 1, xR, xC)
|
|
);
|
|
dotProd += dot(xValues, wValues);
|
|
}
|
|
|
|
} else if (${f===3}) {
|
|
vec3 wValues = vec3(
|
|
getW(wR, wC, ${h}, d2),
|
|
getW(wR, wC, ${h} + 1, d2),
|
|
getW(wR, wC, ${h} + 2, d2)
|
|
);
|
|
|
|
if (${m}) {
|
|
vec3 xValues = vec3(
|
|
getX(batch, xR, xC, ${h}),
|
|
getX(batch, xR, xC, ${h} + 1),
|
|
getX(batch, xR, xC, ${h} + 2)
|
|
);
|
|
dotProd += dot(xValues, wValues);
|
|
} else {
|
|
vec3 xValues = vec3(
|
|
getX(batch, ${h}, xR, xC),
|
|
getX(batch, ${h} + 1, xR, xC),
|
|
getX(batch, ${h} + 2, xR, xC)
|
|
);
|
|
dotProd += dot(xValues, wValues);
|
|
}
|
|
|
|
}
|
|
}
|
|
}
|
|
|
|
float result = dotProd;
|
|
${w}
|
|
${b}
|
|
setOutput(result);
|
|
}
|
|
`}},lte=class{constructor(e){this.variableNames=["x","W"],this.outputShape=e.outShape;let t=e.padInfo.front,n=e.padInfo.top,s=e.padInfo.left,r=e.strideDepth,a=e.strideHeight,o=e.strideWidth,i=e.dilationDepth,l=e.dilationHeight,c=e.dilationWidth,u=e.filterDepth,d=e.filterHeight,p=e.filterWidth,h=Math.floor(e.inChannels/4)*4,f=e.inChannels%4;this.userCode=`
|
|
const ivec3 strides = ivec3(${r}, ${a}, ${o});
|
|
const ivec3 pads = ivec3(${t}, ${n}, ${s});
|
|
|
|
void main() {
|
|
ivec5 coords = getOutputCoords();
|
|
int batch = coords.x;
|
|
int d2 = coords.u;
|
|
|
|
ivec3 xFRCCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;
|
|
int xFCorner = xFRCCorner.x;
|
|
int xRCorner = xFRCCorner.y;
|
|
int xCCorner = xFRCCorner.z;
|
|
|
|
// Convolve x(?, ?, ?, d1) with w(:, :, :, d1, d2) to get
|
|
// y(yF, yR, yC, d2). ? = to be determined. : = across all
|
|
// values in that axis.
|
|
float dotProd = 0.0;
|
|
for (int wF = 0; wF < ${u}; wF++) {
|
|
int xF = xFCorner + wF * ${i};
|
|
|
|
if (xF < 0 || xF >= ${e.inDepth}) {
|
|
continue;
|
|
}
|
|
|
|
for (int wR = 0; wR < ${d}; wR++) {
|
|
int xR = xRCorner + wR * ${l};
|
|
|
|
if (xR < 0 || xR >= ${e.inHeight}) {
|
|
continue;
|
|
}
|
|
|
|
for (int wC = 0; wC < ${p}; wC++) {
|
|
int xC = xCCorner + wC * ${c};
|
|
|
|
if (xC < 0 || xC >= ${e.inWidth}) {
|
|
continue;
|
|
}
|
|
|
|
for (int d1 = 0; d1 < ${h}; d1 += 4) {
|
|
vec4 xValues = vec4(
|
|
getX(batch, xF, xR, xC, d1),
|
|
getX(batch, xF, xR, xC, d1 + 1),
|
|
getX(batch, xF, xR, xC, d1 + 2),
|
|
getX(batch, xF, xR, xC, d1 + 3)
|
|
);
|
|
vec4 wValues = vec4(
|
|
getW(wF, wR, wC, d1, d2),
|
|
getW(wF, wR, wC, d1 + 1, d2),
|
|
getW(wF, wR, wC, d1 + 2, d2),
|
|
getW(wF, wR, wC, d1 + 3, d2)
|
|
);
|
|
|
|
dotProd += dot(xValues, wValues);
|
|
}
|
|
|
|
if (${f===1}) {
|
|
dotProd +=
|
|
getX(batch, xF, xR, xC, ${h}) *
|
|
getW(wF, wR, wC, ${h}, d2);
|
|
} else if (${f===2}) {
|
|
vec2 xValues = vec2(
|
|
getX(batch, xF, xR, xC, ${h}),
|
|
getX(batch, xF, xR, xC, ${h} + 1)
|
|
);
|
|
vec2 wValues = vec2(
|
|
getW(wF, wR, wC, ${h}, d2),
|
|
getW(wF, wR, wC, ${h} + 1, d2)
|
|
);
|
|
dotProd += dot(xValues, wValues);
|
|
} else if (${f===3}) {
|
|
vec3 xValues = vec3(
|
|
getX(batch, xF, xR, xC, ${h}),
|
|
getX(batch, xF, xR, xC, ${h} + 1),
|
|
getX(batch, xF, xR, xC, ${h} + 2)
|
|
);
|
|
vec3 wValues = vec3(
|
|
getW(wF, wR, wC, ${h}, d2),
|
|
getW(wF, wR, wC, ${h} + 1, d2),
|
|
getW(wF, wR, wC, ${h} + 2, d2)
|
|
);
|
|
dotProd += dot(xValues, wValues);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
setOutput(dotProd);
|
|
}
|
|
`}},ute=class{constructor(e,t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.customUniforms=[{name:"inputShape",type:"ivec3"},{name:"pad",type:"ivec2"},{name:"stride",type:"ivec2"},{name:"dilation",type:"ivec2"},{name:"inChannels",type:"int"},{name:"itemsPerBlockRow",type:"int"},{name:"outWidth",type:"int"}],this.outputShape=e,this.enableShapeUniforms=Fs(this.outputShape.length);let{dataFormat:n}=t,s=Un(),r=n==="channelsLast",a=r?0:1,o=r?1:2,i=this.enableShapeUniforms?"if(blockIndex < outShape[1] && pos < outShape[0]) {":`if(blockIndex < ${e[1]} && pos < ${e[0]}) {`,l="";for(let c=0;c<=1;c++)for(let u=0;u<=1;u++)l+=`
|
|
blockIndex = rc.y + ${u};
|
|
pos = rc.x + ${c};
|
|
|
|
${i}
|
|
offsetY = int(blockIndex / outWidth) * stride[0] - pad[0];
|
|
d0 = offsetY + dilation[0] * (pos / itemsPerBlockRow);
|
|
|
|
if(d0 < inputShape[${a}] && d0 >= 0) {
|
|
// Use custom imod instead mod. On Intel GPU, mod may generate
|
|
// unexpected value.
|
|
// https://github.com/tensorflow/tfjs/issues/5447
|
|
offsetX = imod(blockIndex, outWidth) * stride[1] - pad[1];
|
|
d1 = offsetX + dilation[1] * (imod(pos, itemsPerBlockRow) /
|
|
inChannels);
|
|
|
|
if(d1 < inputShape[${o}] && d1 >= 0) {
|
|
|
|
ch = imod(pos, inChannels);
|
|
|
|
if (${r}) {
|
|
innerDims = vec2(d1, ch);
|
|
result[${c*2+u}] = getChannel(
|
|
getA(d0, int(innerDims.x),
|
|
int(innerDims.y)), innerDims);
|
|
} else {
|
|
innerDims = vec2(d0, d1);
|
|
result[${c*2+u}] = getChannel(
|
|
getA(ch, int(innerDims.x),
|
|
int(innerDims.y)), innerDims);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
`;this.userCode=`
|
|
void main() {
|
|
ivec2 rc = getOutputCoords();
|
|
|
|
vec4 result = vec4(0);
|
|
|
|
int blockIndex, pos, offsetY, d0, offsetX, d1, ch;
|
|
vec2 innerDims;
|
|
|
|
${l}
|
|
|
|
${s.output} = result;
|
|
}
|
|
`}};function w4({x:e,filter:t,convInfo:n,backend:s,bias:r=null,preluActivationWeights:a=null,leakyreluAlpha:o=0,activation:i=null}){let l=e.shape,c=s.texData.get(e.dataId),u=n.inChannels,d=l[0]*l[1]*l[2],p=n.outChannels,h=n.dataFormat==="channelsLast",f=!1,m=!1,g,A=[];if(!((d===1||p===1)&&u>d4)&&c.isPacked&&h&&c.texture!=null&&l[2]%2!=0&&v.arraysEqual(c.shape.slice(-3),l.slice(-3))){let b=l[0]*l[1]*(l[2]+1),w={dataId:e.dataId,shape:[1,b,n.inChannels],dtype:e.dtype},k=c.shape;c.shape=c.shape.slice(),c.shape[c.shape.length-2]++,v.assert(cp(c.shape,w.shape),()=>`packed reshape ${c.shape} to ${w.shape} isn't free`);let C=ve({inputs:{x:t},backend:s,attrs:{shape:[1,n.inChannels,n.outChannels]}});A.push(C);let N=Vm({a:w,b:C,backend:s,transposeA:f,transposeB:m,bias:r,activation:i,preluActivationWeights:a,leakyreluAlpha:o}),R=s.texData.get(N.dataId);v.assert(R.isPacked,()=>"batchMatMul result is expected to be packed"),c.shape=k,R.shape=n.outShape,g=As({inputs:{x:N},backend:s}),g.shape=n.outShape,A.push(N)}else{let b=h?l[0]*l[1]*l[2]:l[0]*l[2]*l[3],w=ve({inputs:{x:e},backend:s,attrs:{shape:[1,b,n.inChannels]}}),k=ve({inputs:{x:t},backend:s,attrs:{shape:[1,n.inChannels,n.outChannels]}}),C=Vm({a:w,b:k,transposeA:f,transposeB:m,backend:s,bias:r,activation:i,preluActivationWeights:a,leakyreluAlpha:o});g=ve({inputs:{x:C},backend:s,attrs:{shape:n.outShape}}),A.push(w),A.push(k),A.push(C)}for(let b of A)s.disposeIntermediateTensorInfo(b);return g}function k4({x:e,filter:t,convInfo:n,backend:s,bias:r=null,preluActivationWeights:a=null,leakyreluAlpha:o=0,activation:i=null}){let{filterWidth:l,filterHeight:c,inChannels:u,outWidth:d,outHeight:p,dataFormat:h}=n,f=h==="channelsLast",m=l*c*u,g=p*d,A=[m,g],x=!0,y=!1,b=[],w=ve({inputs:{x:e},backend:s,attrs:{shape:e.shape.slice(1)}}),k=ve({inputs:{x:t},backend:s,attrs:{shape:[1,m,v.sizeFromShape(t.shape)/m]}});b.push(w),b.push(k);let C=new ute(A,n),N=[w.shape,[n.padInfo.top,n.padInfo.left],[n.strideHeight,n.strideWidth],[n.dilationHeight,n.dilationWidth],[n.inChannels],[n.filterWidth*n.inChannels],[n.outWidth]],R=s.runWebGLProgram(C,[w],"float32",N),F=ve({inputs:{x:R},backend:s,attrs:{shape:[1,A[0],A[1]]}});b.push(R),b.push(F);let _=r!=null,P=a!=null,T=i==="leakyrelu",M=i?Lm(i,!0):null,U=new o4(F.shape,k.shape,[1,g,n.outChannels],x,y,_,M,P,T),j=[F,k];if(r&&j.push(r),P&&j.push(a),T){let J=s.makeTensorInfo([],"float32",v.createScalarValue(o,"float32"));j.push(J),b.push(J)}let z=s.runWebGLProgram(U,j,"float32"),X=f?[1,p,d,n.outChannels]:[1,n.outChannels,p,d],Z=ve({inputs:{x:z},backend:s,attrs:{shape:X}});b.push(z);for(let J of b)s.disposeIntermediateTensorInfo(J);return Z}function cte(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,filter:a}=t,{strides:o,pad:i,dataFormat:l,dilations:c,dimRoundingMode:u}=s,d=E.convertConv2DDataFormat(l),p=E.computeConv2DInfo(r.shape,a.shape,o,c,i,u,!1,d),h;if(p.filterHeight===1&&p.filterWidth===1&&p.dilationHeight===1&&p.dilationWidth===1&&p.strideHeight===1&&p.strideWidth===1&&(p.padInfo.type==="SAME"||p.padInfo.type==="VALID"))h=w4({x:r,filter:a,convInfo:p,backend:n});else if(K().getBool("WEBGL_CONV_IM2COL")&&r.shape[0]===1)h=k4({x:r,filter:a,convInfo:p,backend:n});else{let m=new v4(p);h=n.runWebGLProgram(m,[r,a],"float32")}let f=ve({inputs:{x:h},backend:n,attrs:{shape:p.outShape}});return n.disposeIntermediateTensorInfo(h),f}var dte={kernelName:_a,backendName:"webgl",kernelFunc:cte},pte=class{constructor(e){this.variableNames=["x","dy"],this.outputShape=e.filterShape;let t=e.strideHeight,n=e.strideWidth,s=e.padInfo.top,r=e.padInfo.left,a=e.dataFormat==="channelsLast";this.userCode=`
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int wR = coords.x;
|
|
int wC = coords.y;
|
|
int d1 = coords.z;
|
|
int d2 = coords.w;
|
|
|
|
// Convolve x(?, ?, d1) with dy(:, :, d2) to get dw(wR, wC, d1, d2).
|
|
// ? = to be determined. : = across all values in that axis.
|
|
float dotProd = 0.0;
|
|
|
|
for (int b = 0; b < ${e.batchSize}; b++) {
|
|
for (int yR = 0; yR < ${e.outHeight}; yR++) {
|
|
int xR = wR + yR * ${t} - ${s};
|
|
|
|
if (xR < 0 || xR >= ${e.inHeight}) {
|
|
continue;
|
|
}
|
|
|
|
for (int yC = 0; yC < ${e.outWidth}; yC++) {
|
|
int xC = wC + yC * ${n} - ${r};
|
|
|
|
if (xC < 0 || xC >= ${e.inWidth}) {
|
|
continue;
|
|
}
|
|
|
|
if (${a}) {
|
|
float dyValue = getDy(b, yR, yC, d2);
|
|
float xValue = getX(b, xR, xC, d1);
|
|
dotProd += (xValue * dyValue);
|
|
} else {
|
|
float dyValue = getDy(b, d2, yR, yC);
|
|
float xValue = getX(b, d1, xR, xC);
|
|
dotProd += (xValue * dyValue);
|
|
}
|
|
|
|
}
|
|
}
|
|
}
|
|
setOutput(dotProd);
|
|
}
|
|
`}},hte=class{constructor(e){this.variableNames=["dy","W"],this.outputShape=e.inShape;let t=e.filterHeight,n=e.filterWidth,s=e.strideHeight,r=e.strideWidth,a=e.dataFormat==="channelsLast",o=t-1-e.padInfo.top,i=n-1-e.padInfo.left,l=a?1:2,c=a?2:3,u=a?3:1;this.userCode=`
|
|
const ivec2 pads = ivec2(${o}, ${i});
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int batch = coords[0];
|
|
int d1 = coords[${u}];
|
|
|
|
ivec2 dyCorner = ivec2(coords[${l}], coords[${c}]) - pads;
|
|
int dyRCorner = dyCorner.x;
|
|
int dyCCorner = dyCorner.y;
|
|
|
|
// Convolve dy(?, ?, d2) with w(:, :, d1, d2) to compute dx(xR, xC, d1).
|
|
// ? = to be determined. : = across all values in that axis.
|
|
float dotProd = 0.0;
|
|
for (int wR = 0; wR < ${t}; wR++) {
|
|
float dyR = float(dyRCorner + wR) / ${s}.0;
|
|
|
|
if (dyR < 0.0 || dyR >= ${e.outHeight}.0 || fract(dyR) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyR = int(dyR);
|
|
|
|
int wRPerm = ${t} - 1 - wR;
|
|
|
|
for (int wC = 0; wC < ${n}; wC++) {
|
|
float dyC = float(dyCCorner + wC) / ${r}.0;
|
|
|
|
if (dyC < 0.0 || dyC >= ${e.outWidth}.0 ||
|
|
fract(dyC) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyC = int(dyC);
|
|
|
|
int wCPerm = ${n} - 1 - wC;
|
|
|
|
for (int d2 = 0; d2 < ${e.outChannels}; d2++) {
|
|
|
|
if (${a}) {
|
|
float xValue = getDy(batch, idyR, idyC, d2);
|
|
float wValue = getW(wRPerm, wCPerm, d1, d2);
|
|
dotProd += xValue * wValue;
|
|
} else {
|
|
float xValue = getDy(batch, d2, idyR, idyC);
|
|
float wValue = getW(wRPerm, wCPerm, d1, d2);
|
|
dotProd += xValue * wValue;
|
|
}
|
|
|
|
}
|
|
}
|
|
}
|
|
setOutput(dotProd);
|
|
}
|
|
`}},fte=class{constructor(e){this.variableNames=["x","dy"],this.outputShape=e.filterShape;let t=e.strideDepth,n=e.strideHeight,s=e.strideWidth,r=e.padInfo.front,a=e.padInfo.top,o=e.padInfo.left;this.userCode=`
|
|
void main() {
|
|
ivec5 coords = getOutputCoords();
|
|
int wF = coords.x;
|
|
int wR = coords.y;
|
|
int wC = coords.z;
|
|
int d1 = coords.w;
|
|
int d2 = coords.u;
|
|
|
|
float dotProd = 0.0;
|
|
|
|
for (int b = 0; b < ${e.batchSize}; b++) {
|
|
for (int yF = 0; yF < ${e.outDepth}; yF++) {
|
|
int xF = wF + yF * ${t} - ${r};
|
|
|
|
if (xF < 0 || xF >= ${e.inDepth}) {
|
|
continue;
|
|
}
|
|
|
|
for (int yR = 0; yR < ${e.outHeight}; yR++) {
|
|
int xR = wR + yR * ${n} - ${a};
|
|
|
|
if (xR < 0 || xR >= ${e.inHeight}) {
|
|
continue;
|
|
}
|
|
|
|
for (int yC = 0; yC < ${e.outWidth}; yC++) {
|
|
int xC = wC + yC * ${s} - ${o};
|
|
|
|
if (xC < 0 || xC >= ${e.inWidth}) {
|
|
continue;
|
|
}
|
|
|
|
float dyValue = getDy(b, yF, yR, yC, d2);
|
|
float xValue = getX(b, xF, xR, xC, d1);
|
|
dotProd += (xValue * dyValue);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
setOutput(dotProd);
|
|
}
|
|
`}},mte=class{constructor(e){this.variableNames=["dy","W"],this.outputShape=e.inShape;let t=e.filterDepth,n=e.filterHeight,s=e.filterWidth,r=e.strideDepth,a=e.strideHeight,o=e.strideWidth,i=t-1-e.padInfo.front,l=n-1-e.padInfo.top,c=s-1-e.padInfo.left;this.userCode=`
|
|
const ivec3 pads = ivec3(${i}, ${l}, ${c});
|
|
|
|
void main() {
|
|
ivec5 coords = getOutputCoords();
|
|
int batch = coords.x;
|
|
int d1 = coords.u;
|
|
|
|
|
|
ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;
|
|
int dyFCorner = dyCorner.x;
|
|
int dyRCorner = dyCorner.y;
|
|
int dyCCorner = dyCorner.z;
|
|
|
|
float dotProd = 0.0;
|
|
for (int wF = 0; wF < ${t}; wF++) {
|
|
float dyF = float(dyFCorner + wF) / ${r}.0;
|
|
|
|
if (dyF < 0.0 || dyF >= ${e.outDepth}.0 || fract(dyF) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyF = int(dyF);
|
|
|
|
int wFPerm = ${t} - 1 - wF;
|
|
|
|
for (int wR = 0; wR < ${n}; wR++) {
|
|
float dyR = float(dyRCorner + wR) / ${a}.0;
|
|
|
|
if (dyR < 0.0 || dyR >= ${e.outHeight}.0 ||
|
|
fract(dyR) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyR = int(dyR);
|
|
|
|
int wRPerm = ${n} - 1 - wR;
|
|
|
|
for (int wC = 0; wC < ${s}; wC++) {
|
|
float dyC = float(dyCCorner + wC) / ${o}.0;
|
|
|
|
if (dyC < 0.0 || dyC >= ${e.outWidth}.0 ||
|
|
fract(dyC) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyC = int(dyC);
|
|
|
|
int wCPerm = ${s} - 1 - wC;
|
|
|
|
for (int d2 = 0; d2 < ${e.outChannels}; d2++) {
|
|
float xValue = getDy(batch, idyF, idyR, idyC, d2);
|
|
float wValue = getW(wFPerm, wRPerm, wCPerm, d1, d2);
|
|
dotProd += xValue * wValue;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
setOutput(dotProd);
|
|
}
|
|
`}};function gte(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,dy:a}=t,{strides:o,pad:i,dataFormat:l,dimRoundingMode:c,filterShape:u}=s,d=E.convertConv2DDataFormat(l),p=E.computeConv2DInfo(r.shape,u,o,1,i,c,!1,d),h=new pte(p);return n.runWebGLProgram(h,[r,a],"float32")}var Ate={kernelName:xh,backendName:"webgl",kernelFunc:gte};function yte(e){let{inputs:t,backend:n,attrs:s}=e,{dy:r,filter:a}=t,{inputShape:o,strides:i,pad:l,dataFormat:c,dimRoundingMode:u}=s,d=E.convertConv2DDataFormat(c),p=E.computeConv2DInfo(o,a.shape,i,1,l,u,!1,d),h=new hte(p);return n.runWebGLProgram(h,[r,a],"float32")}var xte={kernelName:Da,backendName:"webgl",kernelFunc:yte};function bte(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,filter:a}=t,{strides:o,pad:i,dilations:l}=s,c=E.computeConv3DInfo(r.shape,a.shape,o,l,i),u=new lte(c);return n.runWebGLProgram(u,[r,a],"float32")}var vte={kernelName:ed,backendName:"webgl",kernelFunc:bte};function wte(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,dy:a}=t,{strides:o,pad:i,filterShape:l}=s,c=E.computeConv3DInfo(r.shape,l,o,1,i),u=new fte(c);return n.runWebGLProgram(u,[r,a],"float32")}var kte={kernelName:bh,backendName:"webgl",kernelFunc:wte};function Ste(e){let{inputs:t,backend:n,attrs:s}=e,{dy:r,filter:a}=t,{pad:o,strides:i,inputShape:l}=s,c=E.computeConv3DInfo(l,a.shape,i,1,o),u=new mte(c);return n.runWebGLProgram(u,[r,a],"float32")}var Ite={kernelName:vh,backendName:"webgl",kernelFunc:Ste},Cte=a4+`
|
|
return cos(x);
|
|
`,Tte=st({opSnippet:Cte}),Nte={kernelName:Pa,backendName:"webgl",kernelFunc:Tte},Ete=`
|
|
float e2x = exp(-x);
|
|
return (e2x + 1.0 / e2x) / 2.0;
|
|
`,Rte=st({opSnippet:Ete}),$te={kernelName:Fa,backendName:"webgl",kernelFunc:Rte},_te=class{constructor(e,t,n,s,r){this.variableNames=["Image","Boxes","BoxInd"],this.outputShape=[];let[a,o,i,l]=e,[c]=t,[u,d]=n;this.outputShape=[c,u,d,l];let p=s==="bilinear"?1:0,[h,f]=[`${o-1}.0`,`${i-1}.0`],[m,g,A]=u>1?[`${(o-1)/(u-1)}`,"(y2-y1) * height_ratio",`y1*${h} + float(y)*(height_scale)`]:["0.0","0.0",`0.5 * (y1+y2) * ${h}`],[x,y,b]=d>1?[`${(i-1)/(d-1)}`,"(x2-x1) * width_ratio",`x1*${f} + float(x)*(width_scale)`]:["0.0","0.0",`0.5 * (x1+x2) * ${f}`];this.userCode=`
|
|
const float height_ratio = float(${m});
|
|
const float width_ratio = float(${x});
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int b = coords[0];
|
|
int y = coords[1];
|
|
int x = coords[2];
|
|
int d = coords[3];
|
|
|
|
// get box vals
|
|
float y1 = getBoxes(b,0);
|
|
float x1 = getBoxes(b,1);
|
|
float y2 = getBoxes(b,2);
|
|
float x2 = getBoxes(b,3);
|
|
|
|
// get image in batch index
|
|
int bInd = round(getBoxInd(b));
|
|
if(bInd < 0 || bInd >= ${a}) {
|
|
return;
|
|
}
|
|
|
|
float height_scale = ${g};
|
|
float width_scale = ${y};
|
|
|
|
float in_y = ${A};
|
|
if( in_y < 0.0 || in_y > ${h} ) {
|
|
setOutput(float(${r}));
|
|
return;
|
|
}
|
|
float in_x = ${b};
|
|
if( in_x < 0.0 || in_x > ${f} ) {
|
|
setOutput(float(${r}));
|
|
return;
|
|
}
|
|
|
|
vec2 sourceFracIndexCR = vec2(in_x,in_y);
|
|
if(${p} == 1) {
|
|
// Compute the four integer indices.
|
|
ivec2 sourceFloorCR = ivec2(sourceFracIndexCR);
|
|
ivec2 sourceCeilCR = ivec2(ceil(sourceFracIndexCR));
|
|
|
|
float topLeft = getImage(b, sourceFloorCR.y, sourceFloorCR.x, d);
|
|
float bottomLeft = getImage(b, sourceCeilCR.y, sourceFloorCR.x, d);
|
|
float topRight = getImage(b, sourceFloorCR.y, sourceCeilCR.x, d);
|
|
float bottomRight = getImage(b, sourceCeilCR.y, sourceCeilCR.x, d);
|
|
|
|
vec2 fracCR = sourceFracIndexCR - vec2(sourceFloorCR);
|
|
|
|
float top = topLeft + (topRight - topLeft) * fracCR.x;
|
|
float bottom = bottomLeft + (bottomRight - bottomLeft) * fracCR.x;
|
|
float newValue = top + (bottom - top) * fracCR.y;
|
|
setOutput(newValue);
|
|
} else {
|
|
// Compute the coordinators of nearest neighbor point.
|
|
ivec2 sourceNearestCR = ivec2(floor(
|
|
sourceFracIndexCR + vec2(0.5,0.5)));
|
|
float newValue = getImage(b, sourceNearestCR.y, sourceNearestCR.x, d);
|
|
setOutput(newValue);
|
|
}
|
|
}
|
|
`}},Dte=e=>{let{inputs:t,backend:n,attrs:s}=e,{image:r,boxes:a,boxInd:o}=t,{cropSize:i,method:l,extrapolationValue:c}=s,u=new _te(r.shape,a.shape,i,l,c);return n.runWebGLProgram(u,[r,a,o],"float32")},Pte={kernelName:fi,backendName:"webgl",kernelFunc:Dte},S4=class{constructor(e,t,n){this.variableNames=["x"],this.customUniforms=[{name:"index",type:"float"}],this.outputShape=e;let s=e.length,r=t?"0.0":`getX(${I4(s,"coords")})`,a=e[e.length-1],o="",i="";t?(o=n?`end != ${a-1}`:"end != 0",i=n?"end + 1":"end - 1"):(o=n?`end + pow2 < ${a}`:"end >= pow2",i=n?"end + pow2":"end - pow2"),this.userCode=`
|
|
void main() {
|
|
${yt(s)} coords = getOutputCoords();
|
|
int end = ${C4(s,"coords")};
|
|
float val = ${r};
|
|
int pow2 = int(pow(2.0, index));
|
|
if (${o}) {
|
|
int idx = ${i};
|
|
${C4(s,"coords")} = idx;
|
|
val += getX(${I4(s,"coords")});
|
|
}
|
|
setOutput(val);
|
|
}
|
|
`}};function I4(e,t){if(e===1)return`${t}`;if(e===2)return`${t}.x, ${t}.y`;if(e===3)return`${t}.x, ${t}.y, ${t}.z`;if(e===4)return`${t}.x, ${t}.y, ${t}.z, ${t}.w`;throw Error(`Cumulative sum for rank ${e} is not yet supported`)}function C4(e,t){if(e===1)return`${t}`;if(e===2)return`${t}.y`;if(e===3)return`${t}.z`;if(e===4)return`${t}.w`;throw Error(`Cumulative sum for rank ${e} is not yet supported`)}function Fte(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a,exclusive:o,reverse:i}=s,l=r.shape.length,c=E.getAxesPermutation([a],l),u=r;c!=null&&(u=Hn({inputs:{x:r},backend:n,attrs:{perm:c}}));let d=E.getInnerMostAxes(1,l)[0];if(d!==l-1)throw new Error(`WebGL cumsum shader expects an inner-most axis=${r.shape.length-1} but got axis=${a}`);let p=u.shape[d],h=As({inputs:{x:u},backend:n});for(let f=0;f<=Math.ceil(Math.log2(p))-1;f++){let m=new S4(u.shape,!1,i),g=[[f]],A=h;h=n.runWebGLProgram(m,[h],h.dtype,g),n.disposeIntermediateTensorInfo(A)}if(o){let f=new S4(u.shape,o,i),m=h;h=n.runWebGLProgram(f,[h],h.dtype),n.disposeIntermediateTensorInfo(m)}if(c!=null){let f=E.getUndoAxesPermutation(c),m=Hn({inputs:{x:h},backend:n,attrs:{perm:f}});return n.disposeIntermediateTensorInfo(h),n.disposeIntermediateTensorInfo(u),m}return h}var Ote={kernelName:hi,backendName:"webgl",kernelFunc:Fte};function Mte(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,weights:a}=t,{size:o,binaryOutput:i}=s;if(r.shape.length===1){let l=n.readSync(r.dataId),c=n.readSync(a.dataId),u=UI(l,c,a.dtype,a.shape,o);return n.makeTensorInfo([o],a.dtype,u)}else if(r.shape.length===2){let l=n.bufferSync(r),c=n.bufferSync(a),u=XY(l,c,o,i);return n.makeTensorInfo(u.shape,a.dtype,u.values)}throw new Error(`Error in denseBincount: input must be at most rank 2, but got rank${r.shape.length}.`)}var zte={kernelName:wh,backendName:"webgl",kernelFunc:Mte},Lte=class{constructor(e,t,n){this.variableNames=["x"],this.outputShape=[],this.outputShape=e,this.blockSize=t,this.dataFormat=n,this.userCode=`
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int b = coords[0];
|
|
int h = ${this.getHeightCoordString()};
|
|
int w = ${this.getWidthCoordString()};
|
|
int d = ${this.getDepthCoordString()};
|
|
|
|
int in_h = h / ${t};
|
|
int offset_h = imod(h, ${t});
|
|
int in_w = w / ${t};
|
|
int offset_w = imod(w, ${t});
|
|
int offset_d = (offset_h * ${t} + offset_w) *
|
|
${this.getOutputDepthSize()};
|
|
int in_d = d + offset_d;
|
|
|
|
float result = ${this.getInputSamplingString()};
|
|
setOutput(result);
|
|
}
|
|
`}getHeightCoordString(){return this.dataFormat==="NHWC"?"coords[1]":"coords[2]"}getWidthCoordString(){return this.dataFormat==="NHWC"?"coords[2]":"coords[3]"}getDepthCoordString(){return this.dataFormat==="NHWC"?"coords[3]":"coords[1]"}getOutputDepthSize(){return this.dataFormat==="NHWC"?this.outputShape[3]:this.outputShape[1]}getInputSamplingString(){return this.dataFormat==="NHWC"?"getX(b, in_h, in_w, in_d)":"getX(b, in_d, in_h, in_w)"}};function Bte(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{blockSize:a,dataFormat:o}=s,i=r.shape[0],l=o==="NHWC"?r.shape[1]:r.shape[2],c=o==="NHWC"?r.shape[2]:r.shape[3],u=o==="NHWC"?r.shape[3]:r.shape[1],d=l*a,p=c*a,h=u/(a*a),f=o==="NHWC"?[i,d,p,h]:[i,h,d,p],m=new Lte(f,a,o);return n.runWebGLProgram(m,[r],r.dtype)}var Wte={kernelName:mi,backendName:"webgl",kernelFunc:Bte},T4=class{constructor(e,t=!1,n=null,s=!1,r=!1){this.variableNames=["x","W"],this.customUniforms=[{name:"pads",type:"ivec2"},{name:"strides",type:"ivec2"},{name:"dilations",type:"ivec2"},{name:"inDims",type:"ivec2"}],this.outputShape=e.outShape,this.enableShapeUniforms=Fs(this.outputShape.length);let a=e.filterHeight,o=e.filterWidth,i=e.outChannels/e.inChannels,l="",c="";n&&(s?l=`float activation(float a) {
|
|
float b = getPreluActivationWeightsAtOutCoords();
|
|
${n}
|
|
}`:r?l=`float activation(float a) {
|
|
float b = getLeakyreluAlphaAtOutCoords();
|
|
${n}
|
|
}`:l=`
|
|
float activation(float x) {
|
|
${n}
|
|
}
|
|
`,c="result = activation(result);");let u=t?"result += getBiasAtOutCoords();":"";t&&this.variableNames.push("bias"),s&&this.variableNames.push("preluActivationWeights"),r&&this.variableNames.push("leakyreluAlpha"),this.userCode=`
|
|
${l}
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int batch = coords.x;
|
|
ivec2 xRCCorner = coords.yz * strides - pads;
|
|
int d2 = coords.w;
|
|
int d1 = d2 / ${i};
|
|
int q = d2 - d1 * ${i};
|
|
|
|
int xRCorner = xRCCorner.x;
|
|
int xCCorner = xRCCorner.y;
|
|
|
|
// Convolve x(?, ?, d1) with w(:, :, d1, q) to get y(yR, yC, d2).
|
|
// ? = to be determined. : = across all values in that axis.
|
|
float dotProd = 0.0;
|
|
// TO DO(dsmilkov): Flatten the two for loops and vec4 the operations.
|
|
for (int wR = 0; wR < ${a}; wR++) {
|
|
int xR = xRCorner + wR * dilations[0];
|
|
|
|
if (xR < 0 || xR >= inDims[0]) {
|
|
continue;
|
|
}
|
|
|
|
for (int wC = 0; wC < ${o}; wC++) {
|
|
int xC = xCCorner + wC * dilations[1];
|
|
|
|
if (xC < 0 || xC >= inDims[1]) {
|
|
continue;
|
|
}
|
|
|
|
float xVal = getX(batch, xR, xC, d1);
|
|
float wVal = getW(wR, wC, d1, q);
|
|
dotProd += xVal * wVal;
|
|
}
|
|
}
|
|
|
|
float result = dotProd;
|
|
${u}
|
|
${c}
|
|
setOutput(result);
|
|
}
|
|
`}},N4=class{constructor(e,t=!1,n=null,s=!1,r=!1){this.variableNames=["x","W"],this.packedInputs=!0,this.packedOutput=!0,this.customUniforms=[{name:"pads",type:"ivec2"},{name:"strides",type:"ivec2"},{name:"dilations",type:"ivec2"},{name:"inDims",type:"ivec2"}],this.outputShape=e.outShape,this.enableShapeUniforms=Fs(this.outputShape.length);let a=e.outChannels/e.inChannels,o=e.padInfo.left,i=e.strideWidth,l=e.dilationWidth,c=e.filterHeight,u=e.filterWidth,d=u,p=`
|
|
int xR; int xC; int xCOffset;
|
|
vec4 wTexel; vec4 previous; vec4 final;`;for(let g=0;g<u;g++)p+=`
|
|
vec4 xTexelC${g*2};
|
|
int xTexelC${g*2}Ready;
|
|
vec4 xTexelC${g*2+1};
|
|
int xTexelC${g*2+1}Ready;
|
|
vec4 xC${g};`;p+=`
|
|
for (int r = 0; r < ${c}; r++) {
|
|
`;for(let g=0;g<u;g++)p+=`
|
|
xTexelC${g*2} = vec4(0.0);
|
|
xTexelC${g*2}Ready = 0;
|
|
xTexelC${g*2+1} = vec4(0.0);
|
|
xTexelC${g*2+1}Ready = 0;
|
|
xC${g} = vec4(0.0);`;p+=`
|
|
xR = xRCorner + r * dilations[0];
|
|
if (xR >=0 && xR < inDims[0]) {
|
|
`;for(let g=0;g<(d+1)/2;g++){let A=g*2;if(p+=`
|
|
xC = xCCorner + ${A*l};
|
|
`,i===1){if(A<u&&(o%2==1?(p+=`
|
|
xCOffset = xC + 1;
|
|
if (xCOffset >= 0 && xCOffset < inDims[1] && xTexelC${A}Ready == 0) {
|
|
xTexelC${A} = getX(batch, xR, xCOffset, d1);
|
|
|
|
// Need to manually clear unused channels in case
|
|
// we're reading from recycled texture.
|
|
if (xCOffset + 1 >= inDims[1]) {
|
|
xTexelC${A}.zw = vec2(0.0);
|
|
}
|
|
xTexelC${A}Ready = 1;
|
|
}
|
|
`,l===1&&A>0?p+=`
|
|
xC${A} = vec4(xTexelC${A-2}.zw, xTexelC${A}.xy);
|
|
`:p+=`
|
|
xCOffset = xC + 1 - 2;
|
|
|
|
if (xCOffset >= 0 && xCOffset < inDims[1]) {
|
|
previous = getX(batch, xR, xCOffset, d1);
|
|
|
|
// Need to manually clear unused channels in case
|
|
// we're reading from recycled texture.
|
|
if (xCOffset + 1 >= inDims[1]) {
|
|
previous.zw = vec2(0.0);
|
|
}
|
|
|
|
xC${A} = vec4(previous.zw, xTexelC${A}.xy);
|
|
} else {
|
|
xC${A} = vec4(0.0, 0.0, xTexelC${A}.xy);
|
|
}
|
|
`):p+=`
|
|
if (xC >= 0 && xC < inDims[1] && xTexelC${A}Ready == 0) {
|
|
xTexelC${A} = getX(batch, xR, xC, d1);
|
|
if (xC + 1 >= inDims[1]) {
|
|
xTexelC${A}.zw = vec2(0.0);
|
|
}
|
|
xTexelC${A}Ready = 1;
|
|
}
|
|
|
|
xC${A} = xTexelC${A};
|
|
`,A+1<u)){let x=o%2==0?v.nearestLargerEven(l):l;l%2==0&&o%2==1||l%2!=0&&o%2!=1?(p+=`
|
|
xCOffset = xC + imod(pads[1], 2) + ${x};
|
|
|
|
if (xCOffset >= 0 && xCOffset < inDims[1] && xTexelC${A+1}Ready == 0) {
|
|
xTexelC${A+1} = getX(batch, xR, xCOffset, d1);
|
|
|
|
// Need to manually clear unused channels in case
|
|
// we're reading from recycled texture.
|
|
if (xCOffset + 1 >= inDims[1]) {
|
|
xTexelC${A+1}.zw = vec2(0.0);
|
|
}
|
|
xTexelC${A+1}Ready = 1;
|
|
}
|
|
`,l>1&&(p+=`
|
|
xCOffset -= 2;
|
|
if (xCOffset >= 0 && xCOffset < inDims[1] && xTexelC${A}Ready == 0) {
|
|
xTexelC${A} = getX(batch, xR, xCOffset, d1);
|
|
xTexelC${A}Ready = 1;
|
|
}
|
|
`),p+=`
|
|
xC${A+1} = vec4(xTexelC${A}.zw, xTexelC${A+1}.xy);
|
|
`):x===1?p+=`
|
|
xC${A+1} = xTexelC${A};
|
|
`:p+=`
|
|
xCOffset = xC + ${x};
|
|
|
|
if (xCOffset >= 0 && xCOffset < inDims[1] && xTexelC${A+1}Ready == 0) {
|
|
xTexelC${A+1} = getX(batch, xR, xCOffset, d1);
|
|
if (xCOffset + 1 >= inDims[1]) {
|
|
xTexelC${A+1}.zw = vec2(0.0);
|
|
}
|
|
xTexelC${A+1}Ready = 1;
|
|
}
|
|
|
|
xC${A+1} = xTexelC${A+1};
|
|
`}}else A<u&&(o%2==1?(p+=`
|
|
xCOffset = xC + 1 - strides[1];
|
|
if(xCOffset >= 0 && xCOffset < inDims[1] && xTexelC${A}Ready == 0) {
|
|
xTexelC${A} = getX(batch, xR, xCOffset, d1);
|
|
// Need to manually clear unused channels in case
|
|
// we're reading from recycled texture.
|
|
if (xCOffset + 1 >= inDims[1]) {
|
|
xTexelC${A}.zw = vec2(0.0);
|
|
}
|
|
xTexelC${A}Ready = 1;
|
|
}
|
|
|
|
if(xC + 1 >= 0 && xC + 1 < inDims[1] && xTexelC${A+1}Ready == 0) {
|
|
xTexelC${A+1} = getX(batch, xR, xC + 1, d1);
|
|
// Need to manually clear unused channels in case
|
|
// we're reading from recycled texture.
|
|
if (xC + 2 >= inDims[1]) {
|
|
xTexelC${A+1}.zw = vec2(0.0);
|
|
}
|
|
xTexelC${A+1}Ready = 1;
|
|
}
|
|
|
|
xC${A} = vec4(xTexelC${A}.zw, xTexelC${A+1}.zw);
|
|
`,A+1<u&&(p+=`
|
|
final = vec4(0.0);
|
|
xCOffset = xC + 1 + strides[1];
|
|
if(xCOffset >= 0 && xCOffset < inDims[1]) {
|
|
final = getX(batch, xR, xCOffset, d1);
|
|
}
|
|
xC${A+1} = vec4(xTexelC${A+1}.xy, final.xy);
|
|
`)):(p+=`
|
|
if(xC >= 0 && xC < inDims[1] && xTexelC${A}Ready == 0) {
|
|
xTexelC${A} = getX(batch, xR, xC, d1);
|
|
if (xC + 1 >= inDims[1]) {
|
|
xTexelC${A}.zw = vec2(0.0);
|
|
}
|
|
xTexelC${A}Ready = 1;
|
|
}
|
|
|
|
xCOffset = xC + strides[1];
|
|
if(xCOffset >= 0 && xCOffset < inDims[1] && xTexelC${A+1}Ready == 0) {
|
|
xTexelC${A+1} = getX(batch, xR, xCOffset, d1);
|
|
if (xCOffset + 1 >= inDims[1]) {
|
|
xTexelC${A+1}.zw = vec2(0.);
|
|
}
|
|
xTexelC${A+1}Ready = 1;
|
|
}
|
|
|
|
xC${A} = vec4(
|
|
xTexelC${A}.xy, xTexelC${A+1}.xy);
|
|
`,A+1<u&&(p+=`
|
|
xC${A+1} = vec4(xTexelC${A}.zw, xTexelC${A+1}.zw);
|
|
`)));A<u&&(p+=`
|
|
wTexel = getW(r, ${A}, d1, q);
|
|
dotProd += xC${A} * vec4(wTexel.xz, wTexel.xz);
|
|
`,A+1<u&&(p+=`
|
|
wTexel = getW(r, ${A+1}, d1, q);
|
|
dotProd += xC${A+1} * vec4(wTexel.xz, wTexel.xz);
|
|
`))}p+=`
|
|
}
|
|
`,p+=`
|
|
}
|
|
`;let h="",f="";n&&(s?h=`vec4 activation(vec4 a) {
|
|
vec4 b = getPreluActivationWeightsAtOutCoords();
|
|
${n}
|
|
}`:r?h=`vec4 activation(vec4 a) {
|
|
vec4 b = getLeakyreluAlphaAtOutCoords();
|
|
${n}
|
|
}`:h=`vec4 activation(vec4 x) {
|
|
${n}
|
|
}`,f="result = activation(result);");let m=t?"result += getBiasAtOutCoords();":"";t&&this.variableNames.push("bias"),s&&this.variableNames.push("preluActivationWeights"),r&&this.variableNames.push("leakyreluAlpha"),this.userCode=`
|
|
${h}
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int batch = coords.x;
|
|
ivec2 xRCCorner = coords.yz * strides - pads;
|
|
int d2 = coords.w;
|
|
int d1 = d2 / ${a};
|
|
int q = d2 - d1 * ${a};
|
|
int xRCorner = xRCCorner.x;
|
|
int xCCorner = xRCCorner.y;
|
|
|
|
//intialize dotProd with a small epsilon seems to reduce GPU accuracy loss.
|
|
vec4 dotProd = vec4(0.000000000000001);
|
|
|
|
${p}
|
|
|
|
vec4 result = dotProd - vec4(0.000000000000001);
|
|
${m}
|
|
${f}
|
|
setOutput(result);
|
|
}
|
|
`}};function Vte(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,filter:a}=t,{strides:o,pad:i,dilations:l,dimRoundingMode:c}=s,u=l;u==null&&(u=[1,1]),v.assert(E.eitherStridesOrDilationsAreOne(o,u),()=>`Error in depthwiseConv2d: Either strides or dilations must be 1. Got strides ${o} and dilations '${u}'`);let d=E.computeConv2DInfo(r.shape,a.shape,o,u,i,c,!0),p;K().getBool("WEBGL_PACK_DEPTHWISECONV")&&d.strideWidth<=2&&d.outChannels/d.inChannels==1?p=new N4(d):p=new T4(d);let h=[[d.padInfo.top,d.padInfo.left],[d.strideHeight,d.strideWidth],[d.dilationHeight,d.dilationWidth],[d.inHeight,d.inWidth]];return n.runWebGLProgram(p,[r,a],"float32",h)}var Ute={kernelName:Oa,backendName:"webgl",kernelFunc:Vte},Gte=class{constructor(e){this.variableNames=["x","dy"],this.outputShape=e.filterShape;let t=e.strideHeight,n=e.strideWidth,s=e.padInfo.top,r=e.padInfo.left,a=e.outChannels/e.inChannels;this.userCode=`
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int wR = coords.x;
|
|
int wC = coords.y;
|
|
int d1 = coords.z;
|
|
int dm = coords.w;
|
|
int d2 = d1 * ${a} + dm;
|
|
|
|
float dotProd = 0.0;
|
|
|
|
// TO DO: Vec4 over the batch size
|
|
for (int b = 0; b < ${e.batchSize}; b++) {
|
|
for (int yR = 0; yR < ${e.outHeight}; yR++) {
|
|
int xR = wR + yR * ${t} - ${s};
|
|
|
|
if (xR < 0 || xR >= ${e.inHeight}) {
|
|
continue;
|
|
}
|
|
|
|
for (int yC = 0; yC < ${e.outWidth}; yC++) {
|
|
int xC = wC + yC * ${n} - ${r};
|
|
|
|
if (xC < 0 || xC >= ${e.inWidth}) {
|
|
continue;
|
|
}
|
|
|
|
float dyValue = getDy(b, yR, yC, d2);
|
|
float xValue = getX(b, xR, xC, d1);
|
|
dotProd += (xValue * dyValue);
|
|
}
|
|
}
|
|
}
|
|
setOutput(dotProd);
|
|
}
|
|
`}},Hte=class{constructor(e){this.variableNames=["dy","W"],this.outputShape=e.inShape;let t=e.filterHeight,n=e.filterWidth,s=e.strideHeight,r=e.strideWidth,a=t-1-e.padInfo.top,o=n-1-e.padInfo.left,i=e.outChannels/e.inChannels;this.userCode=`
|
|
const ivec2 pads = ivec2(${a}, ${o});
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int batch = coords[0];
|
|
int d1 = coords[3];
|
|
ivec2 dyCorner = coords.yz - pads;
|
|
int dyRCorner = dyCorner.x;
|
|
int dyCCorner = dyCorner.y;
|
|
|
|
float dotProd = 0.0;
|
|
|
|
for (int wR = 0; wR < ${t}; wR++) {
|
|
float dyR = float(dyRCorner + wR) / ${s}.0;
|
|
|
|
if (dyR < 0.0 || dyR >= ${e.outHeight}.0 || fract(dyR) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyR = int(dyR);
|
|
|
|
int wRPerm = ${t} - 1 - wR;
|
|
|
|
for (int wC = 0; wC < ${n}; wC++) {
|
|
float dyC = float(dyCCorner + wC) / ${r}.0;
|
|
|
|
if (dyC < 0.0 || dyC >= ${e.outWidth}.0 ||
|
|
fract(dyC) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyC = int(dyC);
|
|
|
|
int wCPerm = ${n} - 1 - wC;
|
|
|
|
// TO DO: Vec4 over the channelMul
|
|
for (int dm = 0; dm < ${i}; dm++) {
|
|
int d2 = d1 * ${i} + dm;
|
|
float xValue = getDy(batch, idyR, idyC, d2);
|
|
float wValue = getW(wRPerm, wCPerm, d1, dm);
|
|
dotProd += xValue * wValue;
|
|
}
|
|
}
|
|
}
|
|
setOutput(dotProd);
|
|
}
|
|
`}};function jte(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,dy:a}=t,{strides:o,dilations:i,pad:l,dimRoundingMode:c,filterShape:u}=s,d=E.computeConv2DInfo(r.shape,u,o,i,l,c,!0),p=new Gte(d);return n.runWebGLProgram(p,[r,a],"float32")}var qte={kernelName:kh,backendName:"webgl",kernelFunc:jte};function Xte(e){let{inputs:t,backend:n,attrs:s}=e,{dy:r,filter:a}=t,{strides:o,dilations:i,pad:l,dimRoundingMode:c,inputShape:u}=s,d=E.computeConv2DInfo(u,a.shape,o,i,l,c,!0),p=new Hte(d);return n.runWebGLProgram(p,[r,a],"float32")}var Kte={kernelName:Sh,backendName:"webgl",kernelFunc:Xte},Zte=class{constructor(e){this.variableNames=["X"],this.outputShape=[e,e],this.userCode=`
|
|
void main() {
|
|
ivec2 coords = getOutputCoords();
|
|
float val = coords[0] == coords[1] ? getX(coords[0]) : 0.0;
|
|
setOutput(val);
|
|
}
|
|
`}};function Yte(e){let{inputs:t,backend:n}=e,{x:s}=t,r=[...s.shape,...s.shape],a=v.sizeFromShape(s.shape),o=ve({inputs:{x:s},backend:n,attrs:{shape:[a]}}),i=new Zte(a),l=n.runWebGLProgram(i,[o],o.dtype),c=ve({inputs:{x:l},backend:n,attrs:{shape:r}});return n.disposeIntermediateTensorInfo(o),n.disposeIntermediateTensorInfo(l),c}var Jte={kernelName:Ih,backendName:"webgl",kernelFunc:Yte},Qte=class{constructor(e){this.variableNames=["x","W"],this.outputShape=e.outShape;let{inHeight:t,inWidth:n,padInfo:s,strideHeight:r,strideWidth:a,filterHeight:o,filterWidth:i,dilationHeight:l,dilationWidth:c}=e,{top:u,left:d}=s;this.userCode=`
|
|
const ivec2 strides = ivec2(${r}, ${a});
|
|
const ivec2 pads = ivec2(${u}, ${d});
|
|
const float neg_infinity = -3.4e38;
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int batch = coords.x;
|
|
int d1 = coords.w;
|
|
ivec2 outTopLeftCorner =
|
|
coords.yz * strides - pads;
|
|
int hBeg = outTopLeftCorner.x;
|
|
int wBeg = outTopLeftCorner.y;
|
|
|
|
float curVal = neg_infinity;
|
|
for (int h = 0; h < ${o}; h++) {
|
|
int hIn = hBeg + h * ${l};
|
|
|
|
if (hIn >= 0 && hIn < ${t}) {
|
|
for (int w = 0; w < ${i}; w++) {
|
|
int wIn = wBeg + w * ${c};
|
|
|
|
if (wIn >= 0 && wIn < ${n}) {
|
|
float xVal = getX(batch, hIn, wIn, d1);
|
|
float wVal = getW(h, w, d1);
|
|
|
|
float val = xVal + wVal;
|
|
if (val > curVal) {
|
|
curVal = val;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
float result = curVal;
|
|
setOutput(result);
|
|
}
|
|
`}};function ene(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,filter:a}=t,{strides:o,pad:i,dilations:l}=s,c=E.computeDilation2DInfo(r.shape,a.shape,o,i,"NHWC",l),u,d=new Qte(c);u=n.runWebGLProgram(d,[r,a],"float32");let p=ve({inputs:{x:u},backend:n,attrs:{shape:c.outShape}});return n.disposeIntermediateTensorInfo(u),p}var tne={kernelName:td,backendName:"webgl",kernelFunc:ene};function nne(e){let{inputs:t,backend:n,attrs:s}=e,{equation:r}=s,a=t,{allDims:o,summedDims:i,idDims:l}=E.decodeEinsumEquation(r,a.length);E.checkEinsumDimSizes(o.length,l,a);let{path:c,steps:u}=E.getEinsumComputePath(i,l),d=u.length,p=null,h=o.length,f=[];for(let m=0;m<d;++m){for(let g of u[m]){let{permutationIndices:A,expandDims:x}=E.getEinsumPermutation(h,l[g]),y;E.isIdentityPermutation(A)?y=a[g]:(y=Hn({inputs:{x:a[g]},backend:n,attrs:{perm:A}}),f.push(y));let b=y.shape.slice();for(let w=0;w<x.length;++w)b.splice(x[w],0,1);v.arraysEqual(y.shape,b)||(y=ve({inputs:{x:y},backend:n,attrs:{shape:b}}),f.push(y)),p===null?p=y:(p=lx({inputs:{a:y,b:p},backend:n}),f.push(p))}m<d-1&&(c[m]>=0&&(p=Wm({inputs:{x:p},backend:n,attrs:{axis:c[m]-(o.length-h),keepDims:!1}}),f.push(p)),h--)}for(let m of f)m!==p&&n.disposeIntermediateTensorInfo(m);return p}var sne={kernelName:nd,backendName:"webgl",kernelFunc:nne},rne="return (x >= 0.0) ? x : (exp(x) - 1.0);",ane=`
|
|
vec4 result;
|
|
|
|
result.r = (x.r >= 0.0) ? x.r : (exp(x.r) - 1.0);
|
|
result.g = (x.g >= 0.0) ? x.g : (exp(x.g) - 1.0);
|
|
result.b = (x.b >= 0.0) ? x.b : (exp(x.b) - 1.0);
|
|
result.a = (x.a >= 0.0) ? x.a : (exp(x.a) - 1.0);
|
|
|
|
return result;
|
|
`,one=st({opSnippet:rne,packedOpSnippet:ane}),ine={kernelName:za,backendName:"webgl",kernelFunc:one},lne="return (b >= 1.0) ? a : a * (b + 1.0);",une=`
|
|
vec4 bGTEZero = vec4(greaterThanEqual(b, vec4(0.)));
|
|
return (bGTEZero * a) + ((vec4(1.0) - bGTEZero) * (a * (b + vec4(1.0))));
|
|
`,cne=e=>{let{inputs:t,backend:n}=e,{dy:s,y:r}=t,a=K().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new hp(une,s.shape,r.shape):new lc(lne,s.shape,r.shape);return n.runWebGLProgram(a,[s,r],s.dtype)},dne={kernelName:Nh,backendName:"webgl",kernelFunc:cne},pne=`
|
|
return vec4(equal(a, b));
|
|
`,hne="return float(a == b);",fne=Cn({opSnippet:hne,packedOpSnippet:pne,dtype:"bool",cpuKernelImpl:YY}),mne={kernelName:gi,backendName:"webgl",kernelFunc:fne},gne=`
|
|
// Error function is calculated approximately with elementary function.
|
|
// See "Handbook of Mathematical Functions with Formulas,
|
|
// Graphs, and Mathematical Tables", Abramowitz and Stegun.
|
|
float p = ${E.ERF_P};
|
|
float a1 = ${E.ERF_A1};
|
|
float a2 = ${E.ERF_A2};
|
|
float a3 = ${E.ERF_A3};
|
|
float a4 = ${E.ERF_A4};
|
|
float a5 = ${E.ERF_A5};
|
|
|
|
float sign = sign(x);
|
|
x = abs(x);
|
|
float t = 1.0 / (1.0 + p * x);
|
|
return sign * (1.0 - (((((a5*t + a4)*t) + a3)*t + a2)*t + a1)*t*exp(-x*x));
|
|
`,Ane=st({opSnippet:gne}),yne={kernelName:cu,backendName:"webgl",kernelFunc:Ane},E4="return exp(x);",R4=st({opSnippet:E4,packedOpSnippet:E4,cpuKernelImpl:JY,dtype:"float32"}),xne={kernelName:La,backendName:"webgl",kernelFunc:R4};function px(e){let{inputs:t,attrs:n,backend:s}=e,{dim:r}=n,{input:a}=t,o=a.shape.length,i=a.shape.slice(),l=r;return r<0&&(v.assert(-(o+1)<=r,()=>`Axis must be in the interval [${-(o+1)}, ${o}]`),l=o+r+1),i.splice(l,0,1),ve({inputs:{x:a},backend:s,attrs:{shape:i}})}var bne={kernelName:Ai,backendName:"webgl",kernelFunc:px},$4="return exp(x) - 1.0;",vne=st({opSnippet:$4,packedOpSnippet:$4,cpuKernelImpl:QY}),wne={kernelName:yi,backendName:"webgl",kernelFunc:vne},_4=class{constructor(e,t,n){this.variableNames=["real","imag"];let s=t[1];this.outputShape=t;let r=n?`2.0 * ${Math.PI}`:`-2.0 * ${Math.PI}`,a=n?`${s}.0`:"1.0",o;if(e==="real")o="return real * expR - imag * expI;";else if(e==="imag")o="return real * expI + imag * expR;";else throw new Error(`FFT component must be either "real" or "imag", got ${e}.`);this.userCode=`
|
|
const float exponentMultiplier = ${r};
|
|
|
|
float unaryOpComplex(float real, float expR, float imag, float expI) {
|
|
${o}
|
|
}
|
|
|
|
float mulMatDFT(int batch, int index) {
|
|
float indexRatio = float(index) / float(${s});
|
|
float exponentMultiplierTimesIndexRatio =
|
|
exponentMultiplier * indexRatio;
|
|
|
|
float result = 0.0;
|
|
|
|
for (int i = 0; i < ${s}; i++) {
|
|
// x = (-2|2 * PI / N) * index * i;
|
|
float x = exponentMultiplierTimesIndexRatio * float(i);
|
|
float expR = cos(x);
|
|
float expI = sin(x);
|
|
float real = getReal(batch, i);
|
|
float imag = getImag(batch, i);
|
|
|
|
result +=
|
|
unaryOpComplex(real, expR, imag, expI) / ${a};
|
|
}
|
|
|
|
return result;
|
|
}
|
|
|
|
void main() {
|
|
ivec2 coords = getOutputCoords();
|
|
setOutput(mulMatDFT(coords[0], coords[1]));
|
|
}
|
|
`}};function D4(e,t,n){let s=n.texData.get(e.dataId),r=v.sizeFromShape(e.shape),a=e.shape[e.shape.length-1],o=r/a,i=ve({inputs:{x:e},backend:n,attrs:{shape:[o,a]}}),l=i.shape,c=new _4("real",l,t),u=new _4("imag",l,t),d=[{dataId:s.complexTensorInfos.real.dataId,dtype:s.complexTensorInfos.real.dtype,shape:l},{dataId:s.complexTensorInfos.imag.dataId,dtype:s.complexTensorInfos.imag.dtype,shape:l}],p=n.runWebGLProgram(c,d,"float32"),h=n.runWebGLProgram(u,d,"float32"),f=Uo({inputs:{real:p,imag:h},backend:n});n.disposeIntermediateTensorInfo(p),n.disposeIntermediateTensorInfo(h);let m=ve({inputs:{x:f},backend:n,attrs:{shape:e.shape}});return n.disposeIntermediateTensorInfo(i),n.disposeIntermediateTensorInfo(f),m}function kne(e){let{inputs:t,backend:n}=e,{input:s}=t;return D4(s,!1,n)}var Sne={kernelName:Eh,backendName:"webgl",kernelFunc:kne},Ine=class{constructor(e,t){this.outputShape=[],this.customUniforms=[{name:"value",type:"float"}],this.variableNames=["x"],this.outputShape=e,this.userCode=`
|
|
void main() {
|
|
// Input can be obtained from uniform value.
|
|
setOutput(value);
|
|
}
|
|
`}};function gp(e){let{backend:t,attrs:n}=e,{shape:s,value:r}=n,{dtype:a}=n;if(a=a||v.inferDtype(r),a==="string"){let o=v.getArrayFromDType(a,v.sizeFromShape(s));return o.fill(r),t.makeTensorInfo(s,a,o)}else{let o=new Ine(s,r),i=[[r]];return t.runWebGLProgram(o,[],a,i)}}var Cne={kernelName:du,backendName:"webgl",kernelFunc:gp},Tne=class{constructor(e){this.variableNames=["Image"],this.outputShape=[];let t=e[2];this.outputShape=e,this.userCode=`
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int x = coords[2];
|
|
|
|
int coordX = ${t} - x - 1;
|
|
float outputValue;
|
|
if(coordX >= 0 && coordX < ${t}) {
|
|
outputValue = getImage(coords[0], coords[1], coordX, coords[3]);
|
|
} else {
|
|
outputValue = getImage(coords[0], coords[1], coords[2], coords[3]);
|
|
}
|
|
setOutput(outputValue);
|
|
}
|
|
`}},Nne={kernelName:xi,backendName:"webgl",kernelFunc:({inputs:e,backend:t})=>{let{image:n}=e,s=t,r=new Tne(n.shape);return s.runWebGLProgram(r,[n],n.dtype)}},P4="return floor(x);",Ene=st({opSnippet:P4,packedOpSnippet:P4,cpuKernelImpl:eJ}),Rne={kernelName:Ba,backendName:"webgl",kernelFunc:Ene},$ne=`
|
|
float s = sign(a) * sign(b);
|
|
int ia = round(a);
|
|
int ib = round(b);
|
|
if (ib != 0) {
|
|
// Windows (D3D) wants guaranteed non-zero int division at compile-time.
|
|
return float(idiv(ia, ib, s));
|
|
} else {
|
|
return NAN;
|
|
}
|
|
`,_ne=`
|
|
ivec4 ia = round(a);
|
|
ivec4 ib = round(b);
|
|
bvec4 cond = notEqual(ib, ivec4(0));
|
|
ivec4 result = ivec4(0);
|
|
vec4 s = sign(a) * sign(b);
|
|
|
|
// Windows (D3D) wants guaranteed non-zero int division at compile-time.
|
|
if (cond[0]) {
|
|
result[0] = idiv(ia[0], ib[0], s[0]);
|
|
}
|
|
if (cond[1]) {
|
|
result[1] = idiv(ia[1], ib[1], s[1]);
|
|
}
|
|
if (cond[2]) {
|
|
result[2] = idiv(ia[2], ib[2], s[2]);
|
|
}
|
|
if (cond[3]) {
|
|
result[3] = idiv(ia[3], ib[3], s[3]);
|
|
}
|
|
return vec4(result);
|
|
`,Dne=Cn({opSnippet:$ne,packedOpSnippet:_ne,dtype:"int32"}),Pne={kernelName:Wa,backendName:"webgl",kernelFunc:Dne},Fne=class{constructor(e){this.variableNames=["A"];let t=Un(),[n,s]=e;this.outputShape=e,this.userCode=`
|
|
void main() {
|
|
ivec3 coords = getOutputCoords();
|
|
int texR = coords[0];
|
|
int texC = coords[1];
|
|
int depth = coords[2];
|
|
vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${s}.0, ${n}.0);
|
|
|
|
vec4 values = ${t.texture2D}(A, uv);
|
|
float value;
|
|
if (depth == 0) {
|
|
value = values.r;
|
|
} else if (depth == 1) {
|
|
value = values.g;
|
|
} else if (depth == 2) {
|
|
value = values.b;
|
|
} else if (depth == 3) {
|
|
value = values.a;
|
|
}
|
|
|
|
setOutput(floor(value * 255.0 + 0.5));
|
|
}
|
|
`}},One=class{constructor(e){this.variableNames=["A"],this.packedInputs=!1,this.packedOutput=!0;let t=Un(),[n,s]=e;this.outputShape=e,this.userCode=`
|
|
void main() {
|
|
ivec3 coords = getOutputCoords();
|
|
int texR = coords[0];
|
|
int texC = coords[1];
|
|
int depth = coords[2];
|
|
|
|
vec4 result = vec4(0.);
|
|
|
|
for(int row=0; row<=1; row++) {
|
|
for(int col=0; col<=1; col++) {
|
|
texC = coords[1] + row;
|
|
depth = coords[2] + col;
|
|
|
|
vec2 uv = (vec2(texC, texR) + halfCR) /
|
|
vec2(${s}.0, ${n}.0);
|
|
vec4 values = ${t.texture2D}(A, uv);
|
|
float value;
|
|
if (depth == 0) {
|
|
value = values.r;
|
|
} else if (depth == 1) {
|
|
value = values.g;
|
|
} else if (depth == 2) {
|
|
value = values.b;
|
|
} else if (depth == 3) {
|
|
value = values.a;
|
|
}
|
|
|
|
result[row * 2 + col] = floor(value * 255.0 + 0.5);
|
|
}
|
|
}
|
|
|
|
${t.output} = result;
|
|
}
|
|
`}},Mne={kernelName:dd,backendName:"webgl",kernelFunc:zne},dc;function zne(e){let{inputs:t,backend:n,attrs:s}=e,{pixels:r}=t,{numChannels:a}=s,o=typeof HTMLVideoElement!="undefined"&&r instanceof HTMLVideoElement,i=typeof HTMLImageElement!="undefined"&&r instanceof HTMLImageElement,[l,c]=o?[r.videoWidth,r.videoHeight]:[r.width,r.height],u=[c,l],d=[c,l,a];(i||o)&&(dc==null&&(dc=document.createElement("canvas").getContext("2d")),dc.canvas.width=l,dc.canvas.height=c,dc.drawImage(r,0,0,l,c),r=dc.canvas);let p=n.makeTensorInfo(u,"int32");n.texData.get(p.dataId).usage=Ds.PIXELS,n.gpgpu.uploadPixelDataToTexture(n.getTexture(p.dataId),r);let h=K().getBool("WEBGL_PACK")?new One(d):new Fne(d),f=n.runWebGLProgram(h,[p],"int32");return n.disposeData(p.dataId),f}function Lne(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,filter:a,bias:o,preluActivationWeights:i}=t,{strides:l,pad:c,dataFormat:u,dilations:d,dimRoundingMode:p,activation:h,leakyreluAlpha:f}=s,m=E.convertConv2DDataFormat(u),g=E.computeConv2DInfo(r.shape,a.shape,l,d,c,p,!1,m),A,x=[];if(g.filterHeight===1&&g.filterWidth===1&&g.dilationHeight===1&&g.dilationWidth===1&&g.strideHeight===1&&g.strideWidth===1&&(g.padInfo.type==="SAME"||g.padInfo.type==="VALID"))A=w4({x:r,filter:a,convInfo:g,backend:n,bias:o,activation:h,preluActivationWeights:i,leakyreluAlpha:f});else if(K().getBool("WEBGL_CONV_IM2COL")&&r.shape[0]===1)A=k4({x:r,filter:a,convInfo:g,backend:n,bias:o,activation:h,preluActivationWeights:i,leakyreluAlpha:f});else{let b=o!=null,w=i!=null,k=h==="leakyrelu",C=h?Lm(h,!1):null,N=new v4(g,b,C,w,k),R=[r,a];if(o&&R.push(o),i&&R.push(i),k){let F=n.makeTensorInfo([],"float32",v.createScalarValue(f,"float32"));R.push(F),x.push(F)}A=n.runWebGLProgram(N,R,"float32")}let y=ve({inputs:{x:A},backend:n,attrs:{shape:g.outShape}});return x.push(A),x.forEach(b=>n.disposeIntermediateTensorInfo(b)),y}var Bne={kernelName:xo,backendName:"webgl",kernelFunc:Lne};function Wne(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,filter:a,bias:o,preluActivationWeights:i}=t,{strides:l,pad:c,dilations:u,dimRoundingMode:d,activation:p,leakyreluAlpha:h}=s,f=[],m=u;m==null&&(m=[1,1]),v.assert(E.eitherStridesOrDilationsAreOne(l,m),()=>`Error in depthwiseConv2d: Either strides or dilations must be 1. Got strides ${l} and dilations '${m}'`);let g=E.computeConv2DInfo(r.shape,a.shape,l,m,c,d,!0),A=K().getBool("WEBGL_PACK_DEPTHWISECONV")&&g.strideWidth<=2&&g.outChannels/g.inChannels==1,x=p?Lm(p,A):null,y=[r,a],b=o!=null,w=i!=null,k=p==="leakyrelu";if(b&&y.push(o),w&&y.push(i),k){let F=n.makeTensorInfo([],"float32",v.createScalarValue(h,"float32"));y.push(F),f.push(F)}let C;A?C=new N4(g,b,x,w,k):C=new T4(g,b,x,w,k);let N=[[g.padInfo.top,g.padInfo.left],[g.strideHeight,g.strideWidth],[g.dilationHeight,g.dilationWidth],[g.inHeight,g.inWidth]],R=n.runWebGLProgram(C,y,"float32",N);return f.forEach(F=>n.disposeIntermediateTensorInfo(F)),R}var Vne={kernelName:bo,backendName:"webgl",kernelFunc:Wne},Une=class{constructor(e,t,n){this.sliceDim=e,this.strides=t,this.variableNames=["x","indices"],this.outputShape=n;let s=yt(t.length),r=yt(n.length),a=this.sliceDim>1?"strides[j]":"strides";this.userCode=`
|
|
${s} strides = ${s}(${this.strides});
|
|
void main() {
|
|
${r} coords = getOutputCoords();
|
|
int flattenIndex = 0;
|
|
for (int j = 0; j < ${this.sliceDim}; j++) {
|
|
int index = round(getIndices(coords[0], j));
|
|
flattenIndex += index * ${a};
|
|
}
|
|
setOutput(getX(flattenIndex, coords[1]));
|
|
}
|
|
`}};function Gne(e){let{inputs:t,backend:n}=e,{params:s,indices:r}=t,a=r.shape,o=a[a.length-1],i=v.sizeFromShape(s.shape),[l,c,u,d]=E.prepareAndValidate(s,r),p=ve({inputs:{x:r},backend:n,attrs:{shape:[c,o]}}),h=ve({inputs:{x:s},backend:n,attrs:{shape:[v.sizeFromShape(s.shape)/u,u]}});if(n.shouldExecuteOnCPU([s,r])||s.dtype==="string"){let A=n.readSync(r.dataId),x=n.bufferSync(s),y=tJ(A,x,s.dtype,c,o,u,d,s.shape,i);return n.makeTensorInfo(l,s.dtype,y.values)}let f=new Une(o,d,[c,u]),m=n.runWebGLProgram(f,[h,p],h.dtype),g=ve({inputs:{x:m},backend:n,attrs:{shape:l}});return n.disposeIntermediateTensorInfo(p),n.disposeIntermediateTensorInfo(h),n.disposeIntermediateTensorInfo(m),g}var Hne={kernelName:vi,backendName:"webgl",kernelFunc:Gne},jne=class{constructor(e,t){this.variableNames=["A","indices"],this.outputShape=t,this.rank=t.length;let n=yt(this.rank),s=qne(e,2);this.userCode=`
|
|
void main() {
|
|
${n} resRC = getOutputCoords();
|
|
setOutput(getA(${s}));
|
|
}
|
|
`}};function qne(e,t){let n=["resRC.x","resRC.y","resRC.z","resRC.w"],s=[];for(let r=0;r<e.length;r++)r===2?s.push("int(getIndices(resRC.x, resRC.z))"):s.push(`${n[r]}`);return s.join()}function F4(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,indices:a}=t,{axis:o,batchDims:i}=s,l=v.parseAxisParam(o,r.shape)[0],c=n.readSync(a.dataId),u=r.shape[l];for(let b=0;b<c.length;++b){let w=c[b];v.assert(w<=u-1&&w>=0,()=>`GatherV2: the index value ${w} is not in [0, ${u-1}]`)}let d=E.segment_util.collectGatherOpShapeInfo(r,a,l,i),p=v.sizeFromShape(a.shape),h=[],f=ve({inputs:{x:r},backend:n,attrs:{shape:[d.batchSize,d.outerSize,d.dimSize,d.sliceSize]}}),m=ve({inputs:{x:a},backend:n,attrs:{shape:[d.batchSize,p/d.batchSize]}});h.push(f),h.push(m);let g=[d.batchSize,d.outerSize,p/d.batchSize,d.sliceSize];if(n.shouldExecuteOnCPU([r,a])||r.dtype==="string"){let b=n.bufferSync(m),w=n.bufferSync(f),k=nJ(w,b,g);return h.forEach(C=>n.disposeIntermediateTensorInfo(C)),n.makeTensorInfo(d.outputShape,k.dtype,k.values)}let A=new jne(f.shape,g),x=n.runWebGLProgram(A,[f,m],f.dtype);h.push(x);let y=ve({inputs:{x},backend:n,attrs:{shape:d.outputShape}});return h.forEach(b=>n.disposeIntermediateTensorInfo(b)),y}var Xne={kernelName:bi,backendName:"webgl",kernelFunc:F4},Kne="return float(a > b);",Zne=`
|
|
return vec4(greaterThan(a, b));
|
|
`,Yne=Cn({opSnippet:Kne,packedOpSnippet:Zne,cpuKernelImpl:sJ,dtype:"bool"}),Jne={kernelName:wi,backendName:"webgl",kernelFunc:Yne},Qne="return float(a >= b);",ese=`
|
|
return vec4(greaterThanEqual(a, b));
|
|
`,tse=Cn({opSnippet:Qne,packedOpSnippet:ese,dtype:"bool",cpuKernelImpl:rJ}),nse={kernelName:Ua,backendName:"webgl",kernelFunc:tse};function sse(e){let{inputs:t,backend:n}=e,{input:s}=t;return D4(s,!0,n)}var rse={kernelName:Rh,backendName:"webgl",kernelFunc:sse},ase="return float(!isnan(x) && !isinf(x));",ose=st({opSnippet:ase,dtype:"bool"}),ise={kernelName:pu,backendName:"webgl",kernelFunc:ose},lse="return float(isinf(x));",use=st({opSnippet:lse,dtype:"bool"}),cse={kernelName:hu,backendName:"webgl",kernelFunc:use},dse="return float(isnan(x));",pse=st({opSnippet:dse,dtype:"bool"}),hse={kernelName:fu,backendName:"webgl",kernelFunc:pse},fse="return float(a < b);",mse=`
|
|
return vec4(lessThan(a, b));
|
|
`,gse=Cn({opSnippet:fse,packedOpSnippet:mse,cpuKernelImpl:aJ,dtype:"bool"}),Ase={kernelName:Si,backendName:"webgl",kernelFunc:gse},yse="return float(a <= b);",xse=`
|
|
return vec4(lessThanEqual(a, b));
|
|
`,bse=Cn({opSnippet:yse,packedOpSnippet:xse,cpuKernelImpl:oJ,dtype:"bool"}),vse={kernelName:Ii,backendName:"webgl",kernelFunc:bse};function wse(e){let{backend:t,attrs:n}=e,{start:s,stop:r,num:a}=n,o=iJ(s,r,a);return t.makeTensorInfo([o.length],"float32",o)}var kse={kernelName:$h,backendName:"webgl",kernelFunc:wse},Sse=`if (x < 0.0) return NAN;
|
|
return log(x);`,Ise=`
|
|
vec4 result = log(x);
|
|
vec4 isNaN = vec4(lessThan(x, vec4(0.0)));
|
|
result.r = isNaN.r == 1.0 ? NAN : result.r;
|
|
result.g = isNaN.g == 1.0 ? NAN : result.g;
|
|
result.b = isNaN.b == 1.0 ? NAN : result.b;
|
|
result.a = isNaN.a == 1.0 ? NAN : result.a;
|
|
|
|
return result;
|
|
`,Cse=st({opSnippet:Sse,packedOpSnippet:Ise,cpuKernelImpl:lJ}),Tse={kernelName:Ha,backendName:"webgl",kernelFunc:Cse},Nse="return log(1.0 + x);",Ese=st({opSnippet:Nse}),Rse={kernelName:mu,backendName:"webgl",kernelFunc:Ese},$se="return float(a >= 1.0 && b >= 1.0);",_se=`
|
|
return vec4(
|
|
vec4(greaterThanEqual(a, vec4(1.0))) *
|
|
vec4(greaterThanEqual(b, vec4(1.0))));
|
|
`,Dse=Cn({opSnippet:$se,packedOpSnippet:_se,dtype:"bool"}),Pse={kernelName:Ci,backendName:"webgl",kernelFunc:Dse},Fse="return float(!(x >= 1.0));",Ose=st({opSnippet:Fse}),Mse={kernelName:gu,backendName:"webgl",kernelFunc:Ose},zse="return float(a >= 1.0 || b >= 1.0);",Lse=`
|
|
return min(
|
|
vec4(greaterThanEqual(a, vec4(1.0))) +
|
|
vec4(greaterThanEqual(b, vec4(1.0))),
|
|
vec4(1.0));
|
|
`,Bse=Cn({opSnippet:zse,packedOpSnippet:Lse,dtype:"bool"}),Wse={kernelName:rd,backendName:"webgl",kernelFunc:Bse},Vse=class{constructor(e,t,n,s,r){this.variableNames=["x"],this.outputShape=[];let a=t,o=e[3]-1;this.outputShape=e;let i,l=`float(${n}) + float(${s}) * sum`;r===.5?i=`inversesqrt(${l})`:r===1?i=`1.0/(${l})`:i=`exp(log(${l}) * float(-${r}));`,this.userCode=`
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int b = coords[0];
|
|
int r = coords[1];
|
|
int c = coords[2];
|
|
int d = coords[3];
|
|
float x = getX(b, r, c, d);
|
|
float sum = 0.0;
|
|
for (int j = -${a}; j <= ${a}; j++) {
|
|
int idx = d + j;
|
|
if (idx >= 0 && idx <= ${o}) {
|
|
float z = getX(b, r, c, idx);
|
|
sum += z * z;
|
|
}
|
|
}
|
|
float val = x * ${i};
|
|
setOutput(val);
|
|
}
|
|
`}},Use=class{constructor(e,t,n,s,r){this.variableNames=["x"],this.outputShape=[],this.packedInputs=!0,this.packedOutput=!0;let a=t,o=e[3]-1;this.outputShape=e;let i,l=`float(${n}) + float(${s}) * sum`;r===.5?i=`inversesqrt(${l})`:r===1?i=`1.0/(${l})`:i=`exp(log(${l}) * float(-${r}));`,this.userCode=`
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int b = coords.x;
|
|
int r = coords.y;
|
|
int c = coords.z;
|
|
int d = coords.w;
|
|
|
|
bool hasNextCol = d < ${this.outputShape[3]};
|
|
bool hasNextRow = c < ${this.outputShape[2]};
|
|
|
|
vec4 sum = vec4(0.);
|
|
vec4 xFragAtOutputCoords = getX(b, r, c, d);
|
|
|
|
vec4 xAtOutputCoords = vec4(
|
|
getChannel(xFragAtOutputCoords, vec2(c, d)),
|
|
hasNextCol ?
|
|
getChannel(xFragAtOutputCoords, vec2(c, d + 1)) : 0.0,
|
|
hasNextRow ?
|
|
getChannel(xFragAtOutputCoords , vec2(c + 1, d)) : 0.0,
|
|
(hasNextRow && hasNextCol) ?
|
|
getChannel(xFragAtOutputCoords, vec2(c + 1, d + 1)) : 0.0
|
|
);
|
|
|
|
int firstChannel = d - ${a};
|
|
vec2 cache = vec2(0.);
|
|
if(firstChannel >= 0){
|
|
vec4 firstChannelFrag = getX(b, r, c, firstChannel);
|
|
cache.x = getChannel(firstChannelFrag, vec2(c, firstChannel));
|
|
if(hasNextRow){
|
|
cache.y = getChannel(firstChannelFrag, vec2(c + 1, firstChannel));
|
|
}
|
|
}
|
|
|
|
ivec2 depth = ivec2(d, d + 1);
|
|
for (int j = - ${a}; j <= ${a}; j++) {
|
|
ivec2 idx = depth + j;
|
|
bvec2 aboveLowerBound = greaterThanEqual(idx, ivec2(0));
|
|
bvec2 belowUpperBound = lessThanEqual(idx, ivec2(${o}));
|
|
|
|
bool depthInRange = aboveLowerBound.x && belowUpperBound.x;
|
|
bool depthPlusOneInRange = aboveLowerBound.y && belowUpperBound.y;
|
|
|
|
if(depthInRange || depthPlusOneInRange){
|
|
vec4 z = vec4(0.);
|
|
vec4 xFragAtCurrentDepth;
|
|
z.xz = cache.xy;
|
|
if(depthPlusOneInRange && hasNextCol){
|
|
xFragAtCurrentDepth = idx.y != d ?
|
|
getX(b, r, c, idx.y) : xFragAtOutputCoords;
|
|
z.y = getChannel(xFragAtCurrentDepth, vec2(c, idx.y));
|
|
if(hasNextRow){
|
|
z.w = getChannel(xFragAtCurrentDepth, vec2(c + 1, idx.y));
|
|
}
|
|
}
|
|
cache.xy = z.yw;
|
|
sum += z * z;
|
|
}
|
|
}
|
|
vec4 result = xAtOutputCoords * ${i};
|
|
setOutput(result);
|
|
}
|
|
`}},Gse=e=>{let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{depthRadius:a,bias:o,alpha:i,beta:l}=s,c=K().getBool("WEBGL_PACK_NORMALIZATION")?new Use(r.shape,a,o,i,l):new Vse(r.shape,a,o,i,l);return n.runWebGLProgram(c,[r],r.dtype)},Hse={kernelName:ad,backendName:"webgl",kernelFunc:Gse},jse=class{constructor(e,t,n,s,r){this.variableNames=["inputImage","outputImage","dy"],this.outputShape=[],this.outputShape=e,this.depth=e[3],this.depthRadius=t,this.bias=n,this.alpha=s,this.beta=r,this.userCode=`
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int b = coords[0];
|
|
int r = coords[1];
|
|
int c = coords[2];
|
|
|
|
float result = 0.0;
|
|
for (int d = 0; d < ${this.depth}; ++d) {
|
|
int depthBegin = int(max(0.0, float(d - ${t})));
|
|
int depthEnd = int(min(float(${this.depth}),
|
|
float(d + ${t} + 1)));
|
|
|
|
const int MIN_DEPTH_BEGIN = 0;
|
|
const int MAX_DEPTH_END = ${this.depth};
|
|
|
|
float norm = 0.0;
|
|
for (int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k) {
|
|
if (k < depthBegin){
|
|
continue;
|
|
}
|
|
else if (k >= depthBegin && k < depthEnd) {
|
|
norm += getInputImage(b, r, c, k) * getInputImage(b, r, c, k);
|
|
}
|
|
else {
|
|
break;
|
|
}
|
|
}
|
|
|
|
norm = float(${s}) * norm + float(${n});
|
|
|
|
for(int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k){
|
|
if (k < depthBegin){
|
|
continue;
|
|
}
|
|
else if (k >= depthBegin && k < depthEnd){
|
|
float dyi = -2.0 * float(${s})
|
|
* float(${r})
|
|
* getInputImage(b ,r ,c, k) * getOutputImage(b, r, c, d)
|
|
/ norm;
|
|
if (k == d) {
|
|
dyi += pow(norm, -1.0 * ${r});
|
|
}
|
|
if (k == coords[3]) {
|
|
dyi *= getDy(b, r, c, d);
|
|
result += dyi;
|
|
}
|
|
}
|
|
else {
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
setOutput(result);
|
|
}
|
|
`}},qse=e=>{let{inputs:t,backend:n,attrs:s}=e,{x:r,y:a,dy:o}=t,{depthRadius:i,bias:l,alpha:c,beta:u}=s,d=new jse(r.shape,i,l,c,u);return n.runWebGLProgram(d,[r,a,o],r.dtype)},Xse={kernelName:_h,backendName:"webgl",kernelFunc:qse};function Kse(e,t,n,s){let r=v.sizeFromShape(t),o=v.sizeFromShape(e.shape)/r,i=ve({inputs:{x:e},attrs:{shape:[o,r]},backend:s}),l=Tl(i,e.dtype,"max",s),c=ve({inputs:{x:l},attrs:{shape:n},backend:s});return s.disposeIntermediateTensorInfo(i),s.disposeIntermediateTensorInfo(l),c}function O4(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{reductionIndices:a,keepDims:o}=s,i=r.shape.length,l=v.parseAxisParam(a,r.shape),c=l,u=E.getAxesPermutation(c,i),d=u!=null,p=n.shouldExecuteOnCPU([r]),h=r;if(d){if(p){let y=n.texData.get(h.dataId).values,b=new Array(i);for(let C=0;C<b.length;C++)b[C]=r.shape[u[C]];let w=ix(y,r.shape,r.dtype,u,b);h=n.makeTensorInfo(b,r.dtype);let k=n.texData.get(h.dataId);k.values=w}else h=Bm(r,u,n);c=E.getInnerMostAxes(c.length,i)}E.assertAxesAreInnerMostDims("max",c,i);let[f,m]=E.computeOutAndReduceShapes(h.shape,c),g=f;o&&(g=E.expandShapeToKeepDim(f,l));let A;if(p){let y=n.texData.get(h.dataId).values,b=uJ(y,v.sizeFromShape(m),g,r.dtype);A=n.makeTensorInfo(g,r.dtype);let w=n.texData.get(A.dataId);w.values=b}else A=Kse(h,m,g,n);return d&&n.disposeIntermediateTensorInfo(h),A}var Zse={kernelName:ja,backendName:"webgl",kernelFunc:O4},Yse=e4+`
|
|
return max(a, b);
|
|
`,Jse=`
|
|
vec4 result = vec4(max(a, b));
|
|
vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));
|
|
`+zm+`
|
|
return result;
|
|
`,Qse=Cn({opSnippet:Yse,packedOpSnippet:Jse,cpuKernelImpl:cJ}),ere={kernelName:qa,backendName:"webgl",kernelFunc:Qse};function tre(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t;nc(r,"maxPool");let{filterSize:a,strides:o,pad:i,dimRoundingMode:l}=s,c=1;v.assert(E.eitherStridesOrDilationsAreOne(o,c),()=>`Error in maxPool: Either strides or dilations must be 1. Got strides ${o} and dilations '${c}'`);let u=E.computePool2DInfo(r.shape,a,o,c,i,l);if(u.filterWidth===1&&u.filterHeight===1&&v.arraysEqual(u.inShape,u.outShape))return As({inputs:{x:r},backend:n});let d=new fp(u,"max",!1);return n.runWebGLProgram(d,[r],r.dtype)}var nre={kernelName:Xa,backendName:"webgl",kernelFunc:tre};function sre(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{filterSize:a,strides:o,pad:i,dataFormat:l,dimRoundingMode:c}=s,u=[1,1,1],d=E.computePool3DInfo(r.shape,a,o,u,i,c,l),p=new ux(d,"max",!1);return n.runWebGLProgram(p,[r],r.dtype)}var rre={kernelName:od,backendName:"webgl",kernelFunc:sre},are=class{constructor(e){this.variableNames=["dy","maxPos"],this.outputShape=e.inShape;let t=e.strideHeight,n=e.strideWidth,s=e.dilationHeight,r=e.effectiveFilterHeight,a=e.effectiveFilterWidth,o=r-1-e.padInfo.top,i=a-1-e.padInfo.left,l=r*a-1;this.userCode=`
|
|
const ivec2 pads = ivec2(${o}, ${i});
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int b = coords[0];
|
|
int d = coords[3];
|
|
|
|
ivec2 dyRCCorner = coords.yz - pads;
|
|
int dyRCorner = dyRCCorner.x;
|
|
int dyCCorner = dyRCCorner.y;
|
|
|
|
// Convolve dy(?, ?, d) with pos mask(:, :, d) to get dx(xR, xC, d).
|
|
// ? = to be determined. : = across all values in that axis.
|
|
float dotProd = 0.0;
|
|
for (int wR = 0; wR < ${r};
|
|
wR += ${s}) {
|
|
float dyR = float(dyRCorner + wR) / ${t}.0;
|
|
|
|
if (dyR < 0.0 || dyR >= ${e.outHeight}.0 || fract(dyR) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyR = int(dyR);
|
|
|
|
for (int wC = 0; wC < ${a}; wC++) {
|
|
float dyC = float(dyCCorner + wC) / ${n}.0;
|
|
|
|
if (dyC < 0.0 || dyC >= ${e.outWidth}.0 ||
|
|
fract(dyC) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyC = int(dyC);
|
|
|
|
float dyValue = getDy(b, idyR, idyC, d);
|
|
int maxPosValue = ${l} - int(getMaxPos(b, idyR, idyC, d));
|
|
|
|
// Get the current value, check it against the value from the
|
|
// position matrix.
|
|
int curPosValue = wR * ${a} + wC;
|
|
float mask = float(maxPosValue == curPosValue ? 1.0 : 0.0);
|
|
|
|
dotProd += dyValue * mask;
|
|
}
|
|
}
|
|
setOutput(dotProd);
|
|
}
|
|
`}},ore=class{constructor(e){this.variableNames=["dy","maxPos"],this.outputShape=e.inShape;let t=e.strideDepth,n=e.strideHeight,s=e.strideWidth,r=e.dilationDepth,a=e.dilationHeight,o=e.dilationWidth,i=e.effectiveFilterDepth,l=e.effectiveFilterHeight,c=e.effectiveFilterWidth,u=i-1-e.padInfo.front,d=l-1-e.padInfo.top,p=c-1-e.padInfo.left,h=i*l*c-1;this.userCode=`
|
|
const ivec3 pads = ivec3(${u}, ${d}, ${p});
|
|
|
|
void main() {
|
|
ivec5 coords = getOutputCoords();
|
|
int batch = coords.x;
|
|
int ch = coords.u;
|
|
|
|
ivec3 dyCorner = ivec3(coords.y, coords.z, coords.w) - pads;
|
|
int dyDCorner = dyCorner.x;
|
|
int dyRCorner = dyCorner.y;
|
|
int dyCCorner = dyCorner.z;
|
|
|
|
// Convolve dy(?, ?, ?, ch) with pos mask(:, :, :, d) to get
|
|
// dx(xD, xR, xC, ch).
|
|
// ? = to be determined. : = across all values in that axis.
|
|
float dotProd = 0.0;
|
|
|
|
for (int wD = 0; wD < ${i};
|
|
wD += ${r}) {
|
|
float dyD = float(dyDCorner + wD) / ${t}.0;
|
|
|
|
if (dyD < 0.0 || dyD >= ${e.outDepth}.0 || fract(dyD) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyD = int(dyD);
|
|
|
|
for (int wR = 0; wR < ${l};
|
|
wR += ${a}) {
|
|
float dyR = float(dyRCorner + wR) / ${n}.0;
|
|
|
|
if (dyR < 0.0 || dyR >= ${e.outHeight}.0 ||
|
|
fract(dyR) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyR = int(dyR);
|
|
|
|
for (int wC = 0; wC < ${c};
|
|
wC += ${o}) {
|
|
float dyC = float(dyCCorner + wC) / ${s}.0;
|
|
|
|
if (dyC < 0.0 || dyC >= ${e.outWidth}.0 ||
|
|
fract(dyC) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyC = int(dyC);
|
|
|
|
float dyValue = getDy(batch, idyD, idyR, idyC, ch);
|
|
int maxPosValue = ${h} -
|
|
int(getMaxPos(batch, idyD, idyR, idyC, ch));
|
|
|
|
// Get the current value, check it against the value from the
|
|
// position matrix.
|
|
int curPosValue =
|
|
wD * ${l} * ${c} +
|
|
wR * ${c} + wC;
|
|
float mask = float(maxPosValue == curPosValue ? 1.0 : 0.0);
|
|
|
|
dotProd += dyValue * mask;
|
|
}
|
|
}
|
|
}
|
|
setOutput(dotProd);
|
|
}
|
|
`}};function ire(e){let{inputs:t,backend:n,attrs:s}=e,{dy:r,input:a}=t,o=a,{filterSize:i,strides:l,pad:c,dimRoundingMode:u}=s,d=[1,1,1],p=E.computePool3DInfo(o.shape,i,l,d,c,u),h=new ux(p,"max",!0),f=n.runWebGLProgram(h,[o],o.dtype),m=new ore(p),g=n.runWebGLProgram(m,[r,f],o.dtype);return n.disposeIntermediateTensorInfo(f),g}var lre={kernelName:Ph,backendName:"webgl",kernelFunc:ire};function ure(e){let{inputs:t,backend:n,attrs:s}=e,{dy:r,input:a,output:o}=t,i=a;nc([a,o],"maxPoolGrad");let{filterSize:l,strides:c,pad:u,dimRoundingMode:d}=s,p=E.computePool2DInfo(i.shape,l,c,1,u,d),h=!0,f=new fp(p,"max",h),m=n.runWebGLProgram(f,[i],i.dtype),g=new are(p),A=n.runWebGLProgram(g,[r,m],i.dtype);return n.disposeIntermediateTensorInfo(m),A}var cre={kernelName:Dh,backendName:"webgl",kernelFunc:ure};function dre(e,t,n,s){let r=new fp(n,"max",!1),a=s.runWebGLProgram(r,[e],"float32");r=new fp(n,"max",!0,!0,t);let o=s.runWebGLProgram(r,[e],"float32");return[a,o]}var pre={kernelName:Fh,backendName:"webgl",kernelFunc:({inputs:e,attrs:t,backend:n})=>{let{x:s}=e,{filterSize:r,strides:a,pad:o,includeBatchInIndex:i}=t,l=n;v.assert(s.shape.length===4,()=>`Error in maxPool: input must be rank 4 but got rank ${s.shape.length}.`);let c=[1,1];v.assert(E.eitherStridesOrDilationsAreOne(a,c),()=>`Error in maxPool: Either strides or dilations must be 1. Got strides ${a} and dilations '${c}'`);let u=E.computePool2DInfo(s.shape,r,a,c,o),[d,p]=dre(s,i,u,l);return[d,p]}};function hre(e,t,n,s){let r=v.sizeFromShape(t),o=v.sizeFromShape(e.shape)/r,i=ve({inputs:{x:e},attrs:{shape:[o,r]},backend:s}),l=Tl(i,"float32","mean",s),c=ve({inputs:{x:l},attrs:{shape:n},backend:s});return s.disposeIntermediateTensorInfo(i),s.disposeIntermediateTensorInfo(l),c}var fre={kernelName:Ka,backendName:"webgl",kernelFunc:({inputs:e,attrs:t,backend:n})=>{let{x:s}=e,{keepDims:r,axis:a}=t,o=n,i=s.shape.length,l=v.parseAxisParam(a,s.shape),c=l,u=E.getAxesPermutation(c,i),d=u!=null,p=o.shouldExecuteOnCPU([s]),h=[],f=s;if(d){if(p){let b=o.texData.get(f.dataId).values,w=new Array(i);for(let N=0;N<w.length;N++)w[N]=s.shape[u[N]];let k=ix(b,s.shape,s.dtype,u,w);f=o.makeTensorInfo(w,s.dtype);let C=o.texData.get(f.dataId);C.values=k}else f=Bm(s,u,o);h.push(f),c=E.getInnerMostAxes(c.length,i)}E.assertAxesAreInnerMostDims("sum",c,i);let[m,g]=E.computeOutAndReduceShapes(f.shape,c),A=m;r&&(A=E.expandShapeToKeepDim(m,l));let x=hre(f,g,A,o);for(let y of h)o.disposeIntermediateTensorInfo(y);return x}};function mre(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a,keepDims:o}=s,i=r.shape.length,l=v.parseAxisParam(a,r.shape),c=l,u=E.getAxesPermutation(c,i),d=r;u!=null&&(d=Hn({inputs:{x:r},backend:n,attrs:{perm:u}}),c=E.getInnerMostAxes(c.length,r.shape.length)),E.assertAxesAreInnerMostDims("min",c,i);let[p,h]=E.computeOutAndReduceShapes(d.shape,c),f=v.sizeFromShape(h),m=ve({inputs:{x:d},backend:n,attrs:{shape:[-1,f]}}),g=Tl(m,m.dtype,"min",n),A;if(o){let x=E.expandShapeToKeepDim(p,l);A=ve({inputs:{x:g},backend:n,attrs:{shape:x}})}else A=ve({inputs:{x:g},backend:n,attrs:{shape:p}});return n.disposeIntermediateTensorInfo(m),n.disposeIntermediateTensorInfo(g),u!=null&&n.disposeIntermediateTensorInfo(d),A}var gre={kernelName:Za,backendName:"webgl",kernelFunc:mre},Are=e4+`
|
|
return min(a, b);
|
|
`,yre=`
|
|
vec4 result = vec4(min(a, b));
|
|
vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));
|
|
`+zm+`
|
|
return result;
|
|
`,xre=Cn({opSnippet:Are,packedOpSnippet:yre,cpuKernelImpl:dJ}),bre={kernelName:Ya,backendName:"webgl",kernelFunc:xre},vre=class{constructor(e,t,n){this.variableNames=["x"],this.outputShape=t.map((c,u)=>c[0]+e[u]+c[1]);let s=e.length,r=yt(s),a=t.map(c=>c[0]).join(","),o=t.map((c,u)=>c[0]+e[u]).join(","),i=["coords[0]","coords[1]","coords[2]","coords[3]"].slice(0,s),l=n==="reflect"?0:1;if(s===1){this.userCode=`
|
|
int start = ${a};
|
|
int end = ${o};
|
|
|
|
void main() {
|
|
int outC = getOutputCoords();
|
|
if (outC < start) {
|
|
outC = start * 2 - outC - ${l};
|
|
} else if(outC >= end) {
|
|
outC = (end - 1) * 2 - outC + ${l};
|
|
}
|
|
setOutput(getX(outC - start));
|
|
}
|
|
`;return}this.userCode=`
|
|
${r} start = ${r}(${a});
|
|
${r} end = ${r}(${o});
|
|
|
|
void main() {
|
|
${r} outC = getOutputCoords();
|
|
for (int i = 0; i < ${s}; i++) {
|
|
if (outC[i] < start[i]) {
|
|
outC[i] = start[i] * 2 - outC[i] - ${l};
|
|
} else if(outC[i] >= end[i]) {
|
|
outC[i] = (end[i] - 1) * 2 - outC[i] + ${l};
|
|
}
|
|
}
|
|
${r} coords = outC - start;
|
|
setOutput(getX(${i}));
|
|
}
|
|
`}},wre=class{constructor(e,t,n){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t.map((h,f)=>h[0]+e[f]+h[1]);let s=e.length,r=yt(s),a=t.map(h=>h[0]).join(","),o=t.map((h,f)=>h[0]+e[f]).join(","),i=Gn("rc",s),l=Gn("source",s),c=`${i[s-1]} < ${this.outputShape[s-1]}`,u=s===1?"source":`vec2(${l.slice(-2).join()})`,d=n==="reflect"?0:1,p="";if(s===1){let h=`
|
|
${r} source = rc;
|
|
if (source < start) {
|
|
source = start * 2 - source - ${d};
|
|
} else if (source >= end) {
|
|
source = (end - 1) * 2 - source + ${d};
|
|
}
|
|
source -= start;
|
|
`;p=`
|
|
${r} rc = outputLoc;
|
|
${h}
|
|
result[0] = getChannel(getX(${l.join()}), ${u});
|
|
${i[s-1]} += 1;
|
|
if(${c}) {
|
|
${h}
|
|
result[1] = getChannel(getX(${l.join()}), ${u});
|
|
}
|
|
`}else{let h=`
|
|
${r} source = rc;
|
|
${r} lt = ${r}(lessThan(source, start));
|
|
${r} gte = ${r}(greaterThanEqual(source, end));
|
|
${r} orig = 1 - (lt + gte);
|
|
source = orig * source +
|
|
lt * (start * 2 - source - ${d}) +
|
|
gte * ((end - 1) * 2 - source + ${d});
|
|
source -= start;
|
|
`;p=`
|
|
${r} rc = outputLoc;
|
|
${h}
|
|
result[0] = getChannel(getX(${l.join()}), ${u});
|
|
${i[s-1]} += 1;
|
|
if(${c}) {
|
|
${h}
|
|
result[1] = getChannel(getX(${l.join()}), ${u});
|
|
}
|
|
rc = outputLoc;
|
|
${i[s-2]} += 1;
|
|
if(${i[s-2]} < ${this.outputShape[s-2]}) {
|
|
${h}
|
|
result[2] = getChannel(getX(${l.join()}), ${u});
|
|
${i[s-1]} += 1;
|
|
if(${c}) {
|
|
${h}
|
|
result[3] = getChannel(getX(${l.join()}), ${u});
|
|
}
|
|
}
|
|
`}this.userCode=`
|
|
const ${r} start = ${r}(${a});
|
|
const ${r} end = ${r}(${o});
|
|
|
|
void main() {
|
|
${r} outputLoc = getOutputCoords();
|
|
vec4 result = vec4(0.);
|
|
${p}
|
|
setOutput(result);
|
|
}
|
|
`}},kre=({inputs:e,backend:t,attrs:n})=>{let{x:s}=e,{paddings:r,mode:a}=n,o=K().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new wre(s.shape,r,a):new vre(s.shape,r,a);return t.runWebGLProgram(o,[s],s.dtype)},Sre={kernelName:Ja,backendName:"webgl",kernelFunc:kre},Ire=`if (b == 0.0) return NAN;
|
|
return mod(a, b);`,Cre=`
|
|
vec4 result = mod(a, b);
|
|
vec4 isNaN = vec4(equal(b, vec4(0.0)));
|
|
`+zm+`
|
|
return result;
|
|
`,Tre=Cn({opSnippet:Ire,packedOpSnippet:Cre}),Nre={kernelName:Au,backendName:"webgl",kernelFunc:Tre},Ere=class{constructor(e,t,n){this.variableNames=["probs"],this.customUniforms=[{name:"seed",type:"float"}],this.outputShape=[e,n],this.userCode=`
|
|
void main() {
|
|
ivec2 coords = getOutputCoords();
|
|
int batch = coords[0];
|
|
|
|
float r = random(seed);
|
|
float cdf = 0.0;
|
|
|
|
for (int i = 0; i < ${t-1}; i++) {
|
|
cdf += getProbs(batch, i);
|
|
|
|
if (r < cdf) {
|
|
setOutput(float(i));
|
|
return;
|
|
}
|
|
}
|
|
|
|
// If no other event happened, last event happened.
|
|
setOutput(float(${t-1}));
|
|
}
|
|
`}},Rre=`
|
|
if (a == b) {
|
|
return 1.0;
|
|
};
|
|
return a / b;`,$re=`
|
|
// vec4 one = vec4(equal(a, b));
|
|
// return one + (vec4(1.0) - one) * a / b;
|
|
vec4 result = a / b;
|
|
if(a.x == b.x) {
|
|
result.x = 1.;
|
|
}
|
|
if(a.y == b.y) {
|
|
result.y = 1.;
|
|
}
|
|
if(a.z == b.z) {
|
|
result.z = 1.;
|
|
}
|
|
if(a.w == b.w) {
|
|
result.w = 1.;
|
|
}
|
|
|
|
return result;
|
|
`,M4=Cn({opSnippet:Rre,packedOpSnippet:$re,checkOutOfBounds:!0}),_re={kernelName:Ma,backendName:"webgl",kernelFunc:M4},z4="return a - b;",L4=Cn({opSnippet:z4,packedOpSnippet:z4,supportsComplex:!0,cpuKernelImpl:TJ}),Dre={kernelName:fo,backendName:"webgl",kernelFunc:L4};function B4(e){let{inputs:t,backend:n,attrs:s}=e,{logits:r}=t,{dim:a}=s,o=v.parseAxisParam([a],r.shape),i=O4({inputs:{x:r},backend:n,attrs:{reductionIndices:o,keepDims:!1}}),l=E.expandShapeToKeepDim(i.shape,o),c=ve({inputs:{x:i},backend:n,attrs:{shape:l}}),u=L4({inputs:{a:r,b:c},backend:n}),d=R4({inputs:{x:u},backend:n}),p=Wm({inputs:{x:d},backend:n,attrs:{axis:o,keepDims:!1}}),h=ve({inputs:{x:p},backend:n,attrs:{shape:l}}),f=M4({inputs:{a:d,b:h},backend:n});return n.disposeIntermediateTensorInfo(i),n.disposeIntermediateTensorInfo(c),n.disposeIntermediateTensorInfo(u),n.disposeIntermediateTensorInfo(d),n.disposeIntermediateTensorInfo(p),n.disposeIntermediateTensorInfo(h),f}var Pre={kernelName:po,backendName:"webgl",kernelFunc:B4};function Fre(e){let{inputs:t,backend:n,attrs:s}=e,{logits:r}=t,{numSamples:a,seed:o,normalized:i}=s,l=i?r:B4({inputs:{logits:r},backend:n,attrs:{dim:r.shape.length-1}}),c=l.shape[0],u=l.shape[1],d=new Ere(c,u,a),p=[[o]],h=n.runWebGLProgram(d,[l],"int32",p);return i||n.disposeIntermediateTensorInfo(l),h}var Ore={kernelName:Oh,backendName:"webgl",kernelFunc:Fre},W4="return -x;";function Mre(e){let{inputs:t,backend:n}=e,{x:s}=t;if(n.shouldExecuteOnCPU([s])){let a=n.texData.get(s.dataId),[o,i]=hJ(a.values,s.shape,s.dtype);return n.makeTensorInfo(i,s.dtype,o)}let r;return K().getBool("WEBGL_PACK_UNARY_OPERATIONS")?r=new ic(s.shape,W4):r=new Vo(s.shape,W4),n.runWebGLProgram(r,[s],s.dtype)}var zre={kernelName:Ti,backendName:"webgl",kernelFunc:Mre},Lre=Xs.nonMaxSuppressionV3Impl;function Bre(e){E.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");let{inputs:t,backend:n,attrs:s}=e,{boxes:r,scores:a}=t,{maxOutputSize:o,iouThreshold:i,scoreThreshold:l}=s,c=n.readSync(r.dataId),u=n.readSync(a.dataId),{selectedIndices:d}=Lre(c,u,o,i,l);return n.makeTensorInfo([d.length],"int32",new Int32Array(d))}var Wre={kernelName:Ei,backendName:"webgl",kernelFunc:Bre},Vre=Xs.nonMaxSuppressionV4Impl;function Ure(e){E.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");let{inputs:t,backend:n,attrs:s}=e,{boxes:r,scores:a}=t,{maxOutputSize:o,iouThreshold:i,scoreThreshold:l,padToMaxOutputSize:c}=s,u=n.readSync(r.dataId),d=n.readSync(a.dataId),{selectedIndices:p,validOutputs:h}=Vre(u,d,o,i,l,c);return[n.makeTensorInfo([p.length],"int32",new Int32Array(p)),n.makeTensorInfo([],"int32",new Int32Array([h]))]}var Gre={kernelName:yu,backendName:"webgl",kernelFunc:Ure},Hre=Xs.nonMaxSuppressionV5Impl;function jre(e){E.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");let{inputs:t,backend:n,attrs:s}=e,{boxes:r,scores:a}=t,{maxOutputSize:o,iouThreshold:i,scoreThreshold:l,softNmsSigma:c}=s,u=n.readSync(r.dataId),d=n.readSync(a.dataId),p=o,h=i,f=l,m=c,{selectedIndices:g,selectedScores:A}=Hre(u,d,p,h,f,m);return[n.makeTensorInfo([g.length],"int32",new Int32Array(g)),n.makeTensorInfo([A.length],"float32",new Float32Array(A))]}var qre={kernelName:Ri,backendName:"webgl",kernelFunc:jre},Xre=class{constructor(e,t,n,s){this.variableNames=["indices"],this.outputShape=[e,t],this.userCode=`
|
|
void main() {
|
|
ivec2 coords = getOutputCoords();
|
|
int index = round(getIndices(coords.x));
|
|
setOutput(mix(float(${s}), float(${n}),
|
|
float(index == coords.y)));
|
|
}
|
|
`}},Kre=e=>{let{inputs:t,backend:n,attrs:s}=e,{indices:r}=t,{depth:a,onValue:o,offValue:i}=s,l=v.sizeFromShape(r.shape),c=new Xre(l,a,o,i),u=ve({inputs:{x:r},backend:n,attrs:{shape:[l]}}),d=n.runWebGLProgram(c,[u],r.dtype);n.disposeIntermediateTensorInfo(u);let p=[...r.shape,a],h=ve({inputs:{x:d},backend:n,attrs:{shape:p}});return n.disposeIntermediateTensorInfo(d),h},Zre={kernelName:_i,backendName:"webgl",kernelFunc:Kre};function jm(e){let{inputs:t,backend:n}=e,{x:s}=t;if(s.dtype==="complex64"){let r=mp({inputs:{input:s},backend:n}),a=jm({inputs:{x:r},backend:n}),o=Hm({inputs:{input:s},backend:n}),i=jm({inputs:{x:o},backend:n}),l=Uo({inputs:{real:a,imag:i},backend:n});return n.disposeIntermediateTensorInfo(r),n.disposeIntermediateTensorInfo(a),n.disposeIntermediateTensorInfo(o),n.disposeIntermediateTensorInfo(i),l}else return gp({attrs:{shape:s.shape,dtype:s.dtype,value:s.dtype==="string"?"":0},backend:n})}var Yre={kernelName:Ki,backendName:"webgl",kernelFunc:jm};function V4(e){let{inputs:t,backend:n}=e,{x:s}=t;if(s.dtype==="string")throw new Error("onesLike is not supported under string dtype");if(s.dtype==="complex64"){let r=mp({inputs:{input:s},backend:n}),a=V4({inputs:{x:r},backend:n}),o=Hm({inputs:{input:s},backend:n}),i=jm({inputs:{x:o},backend:n}),l=Uo({inputs:{real:a,imag:i},backend:n});return n.disposeIntermediateTensorInfo(r),n.disposeIntermediateTensorInfo(a),n.disposeIntermediateTensorInfo(o),n.disposeIntermediateTensorInfo(i),l}else return gp({attrs:{shape:s.shape,dtype:s.dtype,value:1},backend:n})}var Jre={kernelName:$i,backendName:"webgl",kernelFunc:V4};function Qre(e){let{inputs:t,backend:n,attrs:s}=e,{axis:r}=s;if(t.length===1)return px({inputs:{input:t[0]},backend:n,attrs:{dim:r}});let a=t[0].shape,o=t[0].dtype;t.forEach(u=>{v.assertShapesMatch(a,u.shape,"All tensors passed to stack must have matching shapes"),v.assert(o===u.dtype,()=>"All tensors passed to stack must have matching dtypes")});let i=[],l=t.map(u=>{let d=px({inputs:{input:u},backend:n,attrs:{dim:r}});return i.push(d),d}),c=b4({inputs:l,backend:n,attrs:{axis:r}});return i.forEach(u=>n.disposeIntermediateTensorInfo(u)),c}var eae={kernelName:Di,backendName:"webgl",kernelFunc:Qre},tae=class{constructor(e,t,n){this.variableNames=["x"],this.customUniforms=[{name:"value",type:"float"}],this.outputShape=t.map((l,c)=>l[0]+e[c]+l[1]);let s=e.length,r=yt(s),a=t.map(l=>l[0]).join(","),o=t.map((l,c)=>l[0]+e[c]).join(","),i=["coords[0]","coords[1]","coords[2]","coords[3]"].slice(0,s);if(s===1){this.userCode=`
|
|
int start = ${a};
|
|
int end = ${o};
|
|
|
|
void main() {
|
|
int outC = getOutputCoords();
|
|
if (outC < start || outC >= end) {
|
|
setOutput(value);
|
|
} else {
|
|
setOutput(getX(outC - start));
|
|
}
|
|
}
|
|
`;return}this.userCode=`
|
|
${r} start = ${r}(${a});
|
|
${r} end = ${r}(${o});
|
|
|
|
void main() {
|
|
${r} outC = getOutputCoords();
|
|
if (any(lessThan(outC, start)) || any(greaterThanEqual(outC, end))) {
|
|
setOutput(value);
|
|
} else {
|
|
${r} coords = outC - start;
|
|
setOutput(getX(${i}));
|
|
}
|
|
}
|
|
`}},nae=class{constructor(e,t,n){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0,this.customUniforms=[{name:"value",type:"float"}],this.outputShape=t.map((f,m)=>f[0]+e[m]+f[1]);let s=e.length,r=yt(s),a=t.map(f=>f[0]).join(","),o=t.map((f,m)=>f[0]+e[m]).join(","),i=Gn("rc",s),l=Gn("source",s),c=`${i[s-1]} < ${this.outputShape[s-1]}`,u=s===1?"source":`vec2(${l.slice(-2).join()})`,d=[`${r} rc = outputLoc;`,`${i[s-1]} += 1;
|
|
if(${c}) {
|
|
`,s===1?"":`}
|
|
rc = outputLoc;
|
|
${i[s-2]} += 1;
|
|
if(${i[s-2]} < ${this.outputShape[s-2]}) {`,s===1?"":` ${i[s-1]} += 1;
|
|
if(${c}) {`],p=s===1?"rc < start || rc >= end":"any(lessThan(rc, start)) || any(greaterThanEqual(rc, end))",h="";for(let f=0,m=s===1?2:4;f<m;f++)h+=`
|
|
${d[f]}
|
|
if (${p}) {
|
|
result[${f}] = float(value);
|
|
} else {
|
|
${r} source = rc - start;
|
|
result[${f}] = getChannel(getX(${l.join()}), ${u});
|
|
}
|
|
`;h+=s===1?"} ":"}}",this.userCode=`
|
|
const ${r} start = ${r}(${a});
|
|
const ${r} end = ${r}(${o});
|
|
|
|
void main() {
|
|
${r} outputLoc = getOutputCoords();
|
|
vec4 result = vec4(0.);
|
|
${h}
|
|
setOutput(result);
|
|
}
|
|
`}},U4=e=>{let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{paddings:a,constantValue:o}=s;if(v.sizeFromShape(r.shape)===0){let c=a.map((u,d)=>u[0]+r.shape[d]+u[1]);return gp({backend:n,attrs:{shape:c,value:o,dtype:r.dtype}})}let i=K().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new nae(r.shape,a,o):new tae(r.shape,a,o),l=[[o]];return n.runWebGLProgram(i,[r],r.dtype,l)},sae={kernelName:eo,backendName:"webgl",kernelFunc:U4},rae=`
|
|
if(a < 0.0 && floor(b) < b){
|
|
return NAN;
|
|
}
|
|
if (b == 0.0) {
|
|
return 1.0;
|
|
}
|
|
return (round(mod(b, 2.0)) != 1) ?
|
|
pow(abs(a), b) : sign(a) * pow(abs(a), b);
|
|
`,aae=`
|
|
// isModRound1 has 1 for components with round(mod(b, 2.0)) == 1, 0 otherwise.
|
|
vec4 isModRound1 = vec4(equal(round(mod(b, 2.0)), ivec4(1)));
|
|
vec4 multiplier = sign(a) * isModRound1 + (vec4(1.0) - isModRound1);
|
|
vec4 result = multiplier * pow(abs(a), b);
|
|
|
|
// Ensure that a^0 = 1, including 0^0 = 1 as this correspond to TF and JS
|
|
bvec4 isExpZero = equal(b, vec4(0.0));
|
|
result.r = isExpZero.r ? 1.0 : result.r;
|
|
result.g = isExpZero.g ? 1.0 : result.g;
|
|
result.b = isExpZero.b ? 1.0 : result.b;
|
|
result.a = isExpZero.a ? 1.0 : result.a;
|
|
|
|
vec4 isNaN = vec4(lessThan(a, vec4(0.0))) * vec4(lessThan(floor(b), b));
|
|
`+zm+`
|
|
return result;
|
|
`,oae=Cn({opSnippet:rae,packedOpSnippet:aae}),iae={kernelName:to,backendName:"webgl",kernelFunc:oae};function lae(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a,keepDims:o}=s,i=r.shape.length,l=[],c=v.parseAxisParam(a,r.shape),u=c,d=E.getAxesPermutation(u,i),p=r;d!=null&&(p=Hn({inputs:{x:r},backend:n,attrs:{perm:d}}),u=E.getInnerMostAxes(u.length,i),l.push(p)),E.assertAxesAreInnerMostDims("prod",u,i);let h;if(n.shouldExecuteOnCPU([p])){let f=n.texData.get(p.dataId).values,{outVals:m,outShape:g,outDtype:A}=mJ(p.shape,p.dtype,f,u);h=n.makeTensorInfo(g,A,m)}else{let[f,m]=E.computeOutAndReduceShapes(p.shape,u),g=v.sizeFromShape(m),A=ve({inputs:{x:p},backend:n,attrs:{shape:[-1,g]}}),x=bd(r.dtype),y=Tl(A,x,"prod",n);h=ve({inputs:{x:y},backend:n,attrs:{shape:f}}),l.push(A),l.push(y)}if(o){l.push(h);let f=E.expandShapeToKeepDim(h.shape,c);h=ve({inputs:{x:h},backend:n,attrs:{shape:f}})}return l.forEach(f=>n.disposeIntermediateTensorInfo(f)),h}var uae={kernelName:Pi,backendName:"webgl",kernelFunc:lae},G4=e=>{let{backend:t,attrs:n}=e,{start:s,stop:r,step:a,dtype:o}=n,i=gJ(s,r,a,o);return t.makeTensorInfo([i.length],o,i)},cae={kernelName:xu,backendName:"webgl",kernelFunc:G4},dae="return 1.0 / x;",pae=st({opSnippet:dae}),hae={kernelName:bu,backendName:"webgl",kernelFunc:pae},fae=Ar+`
|
|
return (x < 0.0) ? 0.0 : x;
|
|
`,mae=`
|
|
vec4 result = x * vec4(greaterThanEqual(x, vec4(0.0)));
|
|
bvec4 isNaN = isnan(x);
|
|
|
|
result.r = isNaN.r ? x.r : result.r;
|
|
result.g = isNaN.g ? x.g : result.g;
|
|
result.b = isNaN.b ? x.b : result.b;
|
|
result.a = isNaN.a ? x.a : result.a;
|
|
|
|
return result;
|
|
`,gae=st({opSnippet:fae,packedOpSnippet:mae}),Aae={kernelName:so,backendName:"webgl",kernelFunc:gae},yae=Ar+`
|
|
return (x < 0.0) ? 0.0 : min(6.0, x);
|
|
`,xae=`
|
|
vec4 result = min(x, vec4(6.)) * vec4(greaterThanEqual(x, vec4(0.0)));
|
|
bvec4 isNaN = isnan(x);
|
|
|
|
result.r = isNaN.r ? x.r : result.r;
|
|
result.g = isNaN.g ? x.g : result.g;
|
|
result.b = isNaN.b ? x.b : result.b;
|
|
result.a = isNaN.a ? x.a : result.a;
|
|
|
|
return result;
|
|
`,bae=st({opSnippet:yae,packedOpSnippet:xae}),vae={kernelName:ao,backendName:"webgl",kernelFunc:bae},wae=class{constructor(e,t,n,s,r){this.variableNames=["A"],this.outputShape=[];let[a,o,i,l]=e;this.outputShape=[a,t,n,l];let c=[s&&t>1?o-1:o,s&&n>1?i-1:i],u=[s&&t>1?t-1:t,s&&n>1?n-1:n],d;r?d="(vec2(yRC) + vec2(0.5)) * effectiveInputOverOutputRatioRC - vec2(0.5)":d="vec2(yRC) * effectiveInputOverOutputRatioRC",this.userCode=`
|
|
const vec2 effectiveInputOverOutputRatioRC = vec2(
|
|
${c[0]/u[0]},
|
|
${c[1]/u[1]});
|
|
const vec2 inputShapeRC = vec2(${o}.0, ${i}.0);
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int b = coords[0];
|
|
int d = coords[3];
|
|
ivec2 yRC = coords.yz;
|
|
|
|
// Fractional source index.
|
|
vec2 sourceFracIndexRC = ${d};
|
|
|
|
// Compute the four integer indices.
|
|
ivec2 sourceFloorRC = ivec2(max(sourceFracIndexRC, vec2(0.0)));
|
|
ivec2 sourceCeilRC = ivec2(
|
|
min(inputShapeRC - 1.0, ceil(sourceFracIndexRC)));
|
|
|
|
float topLeft = getA(b, sourceFloorRC.x, sourceFloorRC.y, d);
|
|
float bottomLeft = getA(b, sourceCeilRC.x, sourceFloorRC.y, d);
|
|
float topRight = getA(b, sourceFloorRC.x, sourceCeilRC.y, d);
|
|
float bottomRight = getA(b, sourceCeilRC.x, sourceCeilRC.y, d);
|
|
|
|
vec2 fracRC = sourceFracIndexRC - vec2(sourceFloorRC);
|
|
|
|
float top = topLeft + (topRight - topLeft) * fracRC.y;
|
|
float bottom = bottomLeft + (bottomRight - bottomLeft) * fracRC.y;
|
|
float newValue = top + (bottom - top) * fracRC.x;
|
|
|
|
setOutput(newValue);
|
|
}
|
|
`}},kae=class{constructor(e,t,n,s,r){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=[];let[a,o,i,l]=e;this.outputShape=[a,t,n,l];let c=[s&&t>1?o-1:o,s&&n>1?i-1:i],u=[s&&t>1?t-1:t,s&&n>1?n-1:n],d;r?d="(vec3(yRC) + vec3(0.5)) * effectiveInputOverOutputRatioRC - vec3(0.5)":d="vec3(yRC) * effectiveInputOverOutputRatioRC",this.userCode=`
|
|
const vec3 effectiveInputOverOutputRatioRC = vec3(
|
|
${c[0]/u[0]},
|
|
${c[1]/u[1]},
|
|
${c[1]/u[1]});
|
|
const vec3 inputShapeRC = vec3(${o}.0, ${i}.0,
|
|
${i}.0);
|
|
|
|
float getAValue(int b, int r, int c, int d) {
|
|
return getChannel(getA(b, r, c, d), vec2(c, d));
|
|
}
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int b = coords[0];
|
|
int d = coords[3];
|
|
// Calculate values for next column in yRC.z.
|
|
ivec3 yRC = coords.yzz + ivec3(0, 0, 1);
|
|
|
|
// Fractional source index.
|
|
vec3 sourceFracIndexRC = ${d};
|
|
|
|
// Compute the four integer indices.
|
|
ivec3 sourceFloorRC = ivec3(max(sourceFracIndexRC, vec3(0.0)));
|
|
ivec3 sourceCeilRC = ivec3(
|
|
min(inputShapeRC - 1.0, ceil(sourceFracIndexRC)));
|
|
|
|
// Should we calculate next column and row elements in 2x2 packed cell.
|
|
bool hasNextCol = d < ${l-1};
|
|
bool hasNextRow = coords.z < ${n-1};
|
|
|
|
// In parallel, construct four corners for all four components in
|
|
// packed 2x2 cell.
|
|
vec4 topLeft = vec4(
|
|
getAValue(b, sourceFloorRC.x, sourceFloorRC.y, d),
|
|
hasNextCol ? getAValue(b, sourceFloorRC.x, sourceFloorRC.y, d + 1)
|
|
: 0.0,
|
|
hasNextRow ? getAValue(b, sourceFloorRC.x, sourceFloorRC.z, d)
|
|
: 0.0,
|
|
(hasNextRow && hasNextCol) ?
|
|
getAValue(b, sourceFloorRC.x, sourceFloorRC.z, d + 1) : 0.0);
|
|
|
|
vec4 bottomLeft = vec4(
|
|
getAValue(b, sourceCeilRC.x, sourceFloorRC.y, d),
|
|
hasNextCol ? getAValue(b, sourceCeilRC.x, sourceFloorRC.y, d + 1)
|
|
: 0.0,
|
|
hasNextRow ? getAValue(b, sourceCeilRC.x, sourceFloorRC.z, d)
|
|
: 0.0,
|
|
(hasNextRow && hasNextCol) ?
|
|
getAValue(b, sourceCeilRC.x, sourceFloorRC.z, d + 1) : 0.0);
|
|
|
|
vec4 topRight = vec4(
|
|
getAValue(b, sourceFloorRC.x, sourceCeilRC.y, d),
|
|
hasNextCol ? getAValue(b, sourceFloorRC.x, sourceCeilRC.y, d + 1)
|
|
: 0.0,
|
|
hasNextRow ? getAValue(b, sourceFloorRC.x, sourceCeilRC.z, d)
|
|
: 0.0,
|
|
(hasNextRow && hasNextCol) ?
|
|
getAValue(b, sourceFloorRC.x, sourceCeilRC.z, d + 1) : 0.0);
|
|
|
|
vec4 bottomRight = vec4(
|
|
getAValue(b, sourceCeilRC.x, sourceCeilRC.y, d),
|
|
hasNextCol ? getAValue(b, sourceCeilRC.x, sourceCeilRC.y, d + 1)
|
|
: 0.0,
|
|
hasNextRow ? getAValue(b, sourceCeilRC.x, sourceCeilRC.z, d)
|
|
: 0.0,
|
|
(hasNextRow && hasNextCol) ?
|
|
getAValue(b, sourceCeilRC.x, sourceCeilRC.z, d + 1) : 0.0);
|
|
|
|
vec3 fracRC = sourceFracIndexRC - vec3(sourceFloorRC);
|
|
|
|
vec4 top = mix(topLeft, topRight, fracRC.yyzz);
|
|
vec4 bottom = mix(bottomLeft, bottomRight, fracRC.yyzz);
|
|
vec4 newValue = mix(top, bottom, fracRC.x);
|
|
|
|
setOutput(newValue);
|
|
}
|
|
`}};function Sae(e){let{inputs:t,backend:n,attrs:s}=e,{images:r}=t,{alignCorners:a,halfPixelCenters:o,size:i}=s,[l,c]=i,u=K().getBool("WEBGL_PACK_IMAGE_OPERATIONS")?new kae(r.shape,l,c,a,o):new wae(r.shape,l,c,a,o);return n.runWebGLProgram(u,[r],"float32")}var Iae={kernelName:ro,backendName:"webgl",kernelFunc:Sae},Cae=class{constructor(e,t,n){this.variableNames=["dy"],this.outputShape=[],this.outputShape=t;let[,s,r]=t,[,a,o]=e,i=[n&&a>1?s-1:s,n&&o>1?r-1:r],l=[n&&a>1?a-1:a,n&&o>1?o-1:o],c=i[0]/l[0],u=i[1]/l[1],d=1/c,p=1/u,h=Math.ceil(d)*2+2,f=Math.ceil(p)*2+2;this.userCode=`
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int b = coords[0];
|
|
int d = coords[3];
|
|
int r = coords[1];
|
|
int c = coords[2];
|
|
|
|
float accumulator = 0.0;
|
|
|
|
const float heightScale = float(${c});
|
|
const float widthScale = float(${u});
|
|
|
|
const float invHeightScale = float(${d});
|
|
const float invWidthScale = float(${p});
|
|
|
|
const int winHeight = int(${h});
|
|
const int winWidth = int(${f});
|
|
|
|
// Compute bounds for where in dy we will look
|
|
float startRLerp = floor(float(r) * invHeightScale);
|
|
int startDyR = int(startRLerp - float(winHeight / 2));
|
|
|
|
float startCLerp = floor(float(c) * invWidthScale);
|
|
int startDyC = int(startCLerp - float(winWidth / 2));
|
|
|
|
// Loop over dy
|
|
for (int dyROffset = 0; dyROffset < winHeight; dyROffset++) {
|
|
int dyR = dyROffset + startDyR;
|
|
|
|
// Guard against the window exceeding the bounds of dy
|
|
if (dyR < 0 || dyR >= ${a}) {
|
|
continue;
|
|
}
|
|
|
|
for (int dyCOffset = 0; dyCOffset < winWidth; dyCOffset++) {
|
|
int dyC = dyCOffset + startDyC;
|
|
|
|
// Guard against the window exceeding the bounds of dy
|
|
if (dyC < 0 || dyC >= ${o}) {
|
|
continue;
|
|
}
|
|
|
|
float dxR = float(dyR) * heightScale;
|
|
int topDxRIndex = int(floor(dxR));
|
|
int bottomDxRIndex = int(min(ceil(dxR), ${s-1}.0));
|
|
float dxRLerp = dxR - float(topDxRIndex);
|
|
float inverseDxRLerp = 1.0 - dxRLerp;
|
|
|
|
float dxC = float(dyC) * widthScale;
|
|
int leftDxCIndex = int(floor(dxC));
|
|
int rightDxCIndex = int(min(ceil(dxC), ${r-1}.0));
|
|
float dxCLerp = dxC - float(leftDxCIndex);
|
|
float inverseDxCLerp = 1.0 - dxCLerp;
|
|
|
|
if (r == topDxRIndex && c == leftDxCIndex) {
|
|
// topLeft
|
|
accumulator +=
|
|
getDy(b, dyR, dyC, d) * inverseDxRLerp * inverseDxCLerp;
|
|
}
|
|
|
|
if (r == topDxRIndex && c == rightDxCIndex) {
|
|
// topRight
|
|
accumulator += getDy(b, dyR, dyC, d) * inverseDxRLerp * dxCLerp;
|
|
}
|
|
|
|
if (r == bottomDxRIndex && c == leftDxCIndex) {
|
|
// bottomLeft
|
|
accumulator += getDy(b, dyR, dyC, d) * dxRLerp * inverseDxCLerp;
|
|
}
|
|
|
|
if (r == bottomDxRIndex && c == rightDxCIndex) {
|
|
// bottomRight
|
|
accumulator += getDy(b, dyR, dyC, d) * dxRLerp * dxCLerp;
|
|
}
|
|
}
|
|
}
|
|
// End loop over dy
|
|
|
|
setOutput(accumulator);
|
|
}
|
|
`}};function Tae(e){let{inputs:t,backend:n,attrs:s}=e,{images:r,dy:a}=t,{alignCorners:o}=s,i=new Cae(a.shape,r.shape,o);return n.runWebGLProgram(i,[a],a.dtype)}var Nae={kernelName:zh,backendName:"webgl",kernelFunc:Tae},Eae=class{constructor(e,t,n,s,r){this.variableNames=["A"],this.outputShape=[];let[a,o,i,l]=e;this.outputShape=[a,t,n,l];let c=[s&&t>1?o-1:o,s&&n>1?i-1:i],u=[s&&t>1?t-1:t,s&&n>1?n-1:n],d=s?"0.5":"0.0",p;r?p="max((vec2(yRC) + vec2(0.5)) * effectiveInputOverOutputRatioRC, vec2(0.0))":p="vec2(yRC) * effectiveInputOverOutputRatioRC",this.userCode=`
|
|
const vec2 effectiveInputOverOutputRatioRC = vec2(
|
|
${c[0]/u[0]},
|
|
${c[1]/u[1]});
|
|
const vec2 inputShapeRC = vec2(${o}.0, ${i}.0);
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int b = coords[0];
|
|
int d = coords[3];
|
|
ivec2 yRC = coords.yz;
|
|
|
|
// Fractional source index.
|
|
vec2 sourceFracIndexRC = ${p};
|
|
|
|
// Compute the coordinators of nearest neighbor point.
|
|
ivec2 sourceNearestRC = ivec2(
|
|
min(inputShapeRC - 1.0, floor(sourceFracIndexRC + ${d})));
|
|
float newValue = getA(b, sourceNearestRC.x, sourceNearestRC.y, d);
|
|
|
|
setOutput(newValue);
|
|
}
|
|
`}},Rae=class{constructor(e,t,n,s,r){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=[];let[a,o,i,l]=e;this.outputShape=[a,t,n,l];let c=[s&&t>1?o-1:o,s&&n>1?i-1:i],u=[s&&t>1?t-1:t,s&&n>1?n-1:n],d=s?"0.5":"0.0",p;r?p="max((vec3(yRC) + vec3(0.5)) * effectiveInputOverOutputRatioRC, vec3(0.0))":p="vec3(yRC) * effectiveInputOverOutputRatioRC",this.userCode=`
|
|
const vec3 effectiveInputOverOutputRatioRC = vec3(
|
|
${c[0]/u[0]},
|
|
${c[1]/u[1]},
|
|
${c[1]/u[1]});
|
|
const vec3 inputShapeRC = vec3(${o}.0, ${i}.0,
|
|
${i}.0);
|
|
|
|
float getAValue(int b, int r, int c, int d) {
|
|
return getChannel(getA(b, r, c, d), vec2(c, d));
|
|
}
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int b = coords[0];
|
|
int d = coords[3];
|
|
// Calculate values for next column in yRC.z.
|
|
ivec3 yRC = coords.yzz + ivec3(0, 0, 1);
|
|
|
|
// Fractional source index.
|
|
vec3 sourceFracIndexRC = ${p};
|
|
|
|
// Compute the coordinators of nearest neighbor point.
|
|
ivec3 sourceNearestRC = ivec3(
|
|
min(inputShapeRC - 1.0, floor(sourceFracIndexRC + ${d})));
|
|
|
|
// Should we calculate next column and row elements in 2x2 packed cell.
|
|
bool hasNextCol = d < ${l-1};
|
|
bool hasNextRow = coords.z < ${n-1};
|
|
|
|
vec4 newValue = vec4(
|
|
getAValue(b, sourceNearestRC.x, sourceNearestRC.y, d),
|
|
hasNextCol ? getAValue(b, sourceNearestRC.x, sourceNearestRC.y, d + 1)
|
|
: 0.0,
|
|
hasNextRow ? getAValue(b, sourceNearestRC.x, sourceNearestRC.z, d)
|
|
: 0.0,
|
|
(hasNextRow && hasNextCol) ?
|
|
getAValue(b, sourceNearestRC.x, sourceNearestRC.z, d + 1) : 0.0);
|
|
|
|
setOutput(newValue);
|
|
}
|
|
`}};function $ae(e){let{inputs:t,backend:n,attrs:s}=e,{images:r}=t,{alignCorners:a,halfPixelCenters:o,size:i}=s,[l,c]=i,u=K().getBool("WEBGL_PACK_IMAGE_OPERATIONS")?new Rae(r.shape,l,c,a,o):new Eae(r.shape,l,c,a,o);return n.runWebGLProgram(u,[r],r.dtype)}var _ae={kernelName:vu,backendName:"webgl",kernelFunc:$ae},Dae=class{constructor(e,t,n){this.variableNames=["dy"],this.outputShape=[],this.outputShape=t;let[,s,r]=t,[,a,o]=e,i=[n&&a>1?s-1:s,n&&o>1?r-1:r],l=[n&&a>1?a-1:a,n&&o>1?o-1:o],c=i[0]/l[0],u=i[1]/l[1],d=1/c,p=1/u,h=Math.ceil(d)*2+2,f=Math.ceil(p)*2+2;this.userCode=`
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int b = coords[0];
|
|
int d = coords[3];
|
|
int r = coords[1];
|
|
int c = coords[2];
|
|
|
|
float accumulator = 0.0;
|
|
|
|
const float heightScale = float(${c});
|
|
const float widthScale = float(${u});
|
|
|
|
const float invHeightScale = float(${d});
|
|
const float invWidthScale = float(${p});
|
|
|
|
const int winHeight = int(${h});
|
|
const int winWidth = int(${f});
|
|
|
|
// Compute bounds for where in dy we will look
|
|
float startRLerp = floor(float(r) * invHeightScale);
|
|
int startDyR = int(floor(startRLerp - float(winHeight / 2)));
|
|
|
|
float startCLerp = floor(float(c) * invWidthScale);
|
|
int startDyC = int(floor(startCLerp - float(winWidth / 2)));
|
|
|
|
// Loop over dy
|
|
for (int dyROffset = 0; dyROffset < winHeight; dyROffset++) {
|
|
int dyR = dyROffset + startDyR;
|
|
|
|
// Guard against the window exceeding the bounds of dy
|
|
if (dyR < 0 || dyR >= ${a}) {
|
|
continue;
|
|
}
|
|
|
|
for (int dyCOffset = 0; dyCOffset < winWidth; dyCOffset++) {
|
|
int dyC = dyCOffset + startDyC;
|
|
|
|
// Guard against the window exceeding the bounds of dy
|
|
if (dyC < 0 || dyC >= ${o}) {
|
|
continue;
|
|
}
|
|
|
|
float sourceFracRow =
|
|
float(${i[0]}) *
|
|
(float(dyR) / float(${l[0]}));
|
|
|
|
float sourceFracCol =
|
|
float(${i[1]}) *
|
|
(float(dyC) / float(${l[1]}));
|
|
|
|
int sourceNearestRow = int(min(
|
|
float(int(${s}) - 1),
|
|
${n} ? float(round(sourceFracRow)) :
|
|
float(floor(sourceFracRow))));
|
|
|
|
int sourceNearestCol = int(min(
|
|
float(int(${r}) - 1),
|
|
${n} ? float(round(sourceFracCol)) :
|
|
float(floor(sourceFracCol))));
|
|
|
|
if (r == sourceNearestRow && c == sourceNearestCol) {
|
|
accumulator += getDy(b, dyR, dyC, d);
|
|
}
|
|
}
|
|
}
|
|
// End loop over dy
|
|
|
|
setOutput(accumulator);
|
|
}
|
|
`}};function Pae(e){let{inputs:t,backend:n,attrs:s}=e,{images:r,dy:a}=t,{alignCorners:o}=s,i=new Dae(a.shape,r.shape,o);return n.runWebGLProgram(i,[a],a.dtype)}var Fae={kernelName:Mh,backendName:"webgl",kernelFunc:Pae},Oae=class{constructor(e,t){this.variableNames=["x"];let n=e.length;if(n>4)throw new Error(`WebGL backend: Reverse of rank-${n} tensor is not yet supported`);if(this.outputShape=e,n===1){this.userCode=`
|
|
void main() {
|
|
int coord = getOutputCoords();
|
|
setOutput(getX(${e[0]} - coord - 1));
|
|
}
|
|
`;return}let s=o=>t.indexOf(o)!==-1&&e[o]!==1?`${e[o]} - coords[${o}] - 1`:`coords[${o}]`,r=e.map((o,i)=>s(i)).join(","),a=yt(n);this.userCode=`
|
|
void main() {
|
|
${a} coords = getOutputCoords();
|
|
setOutput(getX(${r}));
|
|
}
|
|
`}},Mae=class{constructor(e,t){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0;let n=e.length;if(n>4)throw new Error(`WebGL backend: Reverse of rank-${n} tensor is not yet supported`);this.outputShape=e;let s=Gn("rc",n),r=`${s[n-1]} + 1 < ${this.outputShape[n-1]}`,a=`${s[n-2]} + 1 < ${this.outputShape[n-2]}`,o=yt(n);n===1?this.userCode=`
|
|
void main(){
|
|
int rc = getOutputCoords();
|
|
vec4 result = vec4(0.);
|
|
result.r = getChannel(getX(${e[0]} - rc - 1),
|
|
${e[0]} - rc - 1);
|
|
if(${r}){
|
|
result.g = getChannel(getX(${e[0]} - (rc + 1) - 1),
|
|
${e[0]} - (rc + 1) - 1);
|
|
}
|
|
setOutput(result);
|
|
}
|
|
`:this.userCode=`
|
|
void main() {
|
|
${o} rc = getOutputCoords();
|
|
vec4 result = vec4(0.);
|
|
result.r = ${i(s.slice())};
|
|
if(${r}){
|
|
result.g = ${l(s.slice())};
|
|
}
|
|
if(${a}) {
|
|
result.b = ${c(s.slice())};
|
|
if(${r}) {
|
|
result.a = ${u(s.slice())};
|
|
}
|
|
}
|
|
setOutput(result);
|
|
}
|
|
`;function i(h){return d(h)}function l(h){return h[n-1]="("+h[n-1]+" + 1)",d(h)}function c(h){return h[n-2]="("+h[n-2]+" + 1)",d(h)}function u(h){return h[n-1]="("+h[n-1]+" + 1)",h[n-2]="("+h[n-2]+" + 1)",d(h)}function d(h){let f=e.map((A,x)=>p(x,h)),m=f.join(","),g=f.slice(-2).join(",");return`getChannel(getX(${m}), vec2(${g}))`}function p(h,f){return t.indexOf(h)!==-1&&e[h]!==1?`${e[h]} - ${f[h]} - 1`:`${f[h]}`}}};function zae(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{dims:a}=s,o=r.shape.length,i=v.parseAxisParam(a,r.shape);if(o===0)return As({inputs:{x:r},backend:n});let l=K().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new Mae(r.shape,i):new Oae(r.shape,i);return n.runWebGLProgram(l,[r],r.dtype)}var Lae={kernelName:Oi,backendName:"webgl",kernelFunc:zae},Bae=class{constructor(e,t){this.variableNames=["Image"],this.outputShape=[],this.customUniforms=[{name:"params",type:"vec4"}];let n=e[1],s=e[2];this.outputShape=e;let r="";typeof t=="number"?r=`float outputValue = ${t.toFixed(2)};`:r=`
|
|
vec3 fill = vec3(${t.join(",")});
|
|
float outputValue = fill[coords[3]];`,this.userCode=`
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int x = coords[2];
|
|
int y = coords[1];
|
|
float coordXFloat = (float(x) - params[0]) * params[3] -
|
|
(float(y) - params[1]) * params[2];
|
|
float coordYFloat = (float(x) - params[0]) * params[2] +
|
|
(float(y) - params[1]) * params[3];
|
|
int coordX = int(round(coordXFloat + params[0]));
|
|
int coordY = int(round(coordYFloat + params[1]));
|
|
${r}
|
|
if(coordX >= 0 && coordX < ${s} && coordY >= 0 && coordY < ${n}) {
|
|
outputValue = getImage(coords[0], coordY, coordX, coords[3]);
|
|
}
|
|
setOutput(outputValue);
|
|
}
|
|
`}},Wae={kernelName:Zi,backendName:"webgl",kernelFunc:({inputs:e,attrs:t,backend:n})=>{let{image:s}=e,{radians:r,fillValue:a,center:o}=t,i=n,l=new Bae(s.shape,a),[c,u]=E.getImageCenter(o,s.shape[1],s.shape[2]),d=[[c,u,Math.sin(r),Math.cos(r)]];return i.runWebGLProgram(l,[s],s.dtype,d)}},Vae=`
|
|
// OpenGL ES does not support round function.
|
|
// The algorithm is based on banker's rounding.
|
|
float base = floor(x);
|
|
if ((x - base) < 0.5) {
|
|
return floor(x);
|
|
} else if ((x - base) > 0.5) {
|
|
return ceil(x);
|
|
} else {
|
|
if (mod(base, 2.0) == 0.0) {
|
|
return base;
|
|
} else {
|
|
return base + 1.0;
|
|
}
|
|
}
|
|
`,Uae=st({opSnippet:Vae}),Gae={kernelName:Mi,backendName:"webgl",kernelFunc:Uae},Hae="return inversesqrt(x);",jae=st({opSnippet:Hae,cpuKernelImpl:AJ}),qae={kernelName:oo,backendName:"webgl",kernelFunc:jae},H4=class{constructor(e,t,n,s,r,a,o=!0){this.variableNames=["updates","indices","defaultValue"],this.outputShape=a;let i=yt(r.length),l=yt(a.length),c="";n===1?c="i":n===2&&(c="i, j");let u=`getIndices(${c})`,d="";s===1?d="i":s===2&&(d="i, coords[1]");let p=`getUpdates(${d})`,h=t>1?"strides[j]":"strides";this.userCode=`
|
|
${i} strides = ${i}(${r});
|
|
|
|
void main() {
|
|
${l} coords = getOutputCoords();
|
|
float sum = 0.0;
|
|
bool found = false;
|
|
for (int i = 0; i < ${e}; i++) {
|
|
int flattenedIndex = 0;
|
|
for (int j = 0; j < ${t}; j++) {
|
|
int index = round(${u});
|
|
flattenedIndex += index * ${h};
|
|
}
|
|
if (flattenedIndex == coords[0]) {
|
|
sum += ${p};
|
|
found = true;
|
|
}
|
|
}
|
|
setOutput(mix(getDefaultValue(), sum, float(found)));
|
|
}
|
|
`}};function Xae(e){let{inputs:t,backend:n,attrs:s}=e,{indices:r,updates:a}=t,{shape:o}=s,{sliceRank:i,numUpdates:l,sliceSize:c,strides:u,outputSize:d}=E.calculateShapes(a,r,o),p=[d/c,c];if(d===0)return n.makeTensorInfo(o,r.dtype);let h=ve({inputs:{x:r},backend:n,attrs:{shape:[l,i]}}),f=ve({inputs:{x:a},backend:n,attrs:{shape:[l,c]}}),m=n.makeTensorInfo([],"float32",new Float32Array([0])),g=new H4(l,i,h.shape.length,f.shape.length,u,p),A=n.runWebGLProgram(g,[f,h,m],f.dtype),x=ve({inputs:{x:A},backend:n,attrs:{shape:o}});return n.disposeIntermediateTensorInfo(h),n.disposeIntermediateTensorInfo(f),n.disposeIntermediateTensorInfo(A),n.disposeIntermediateTensorInfo(m),x}var Kae={kernelName:zi,backendName:"webgl",kernelFunc:Xae},Zae=class{constructor(e,t,n){this.variableNames=["c","a","b"],this.outputShape=t;let s,r;if(n>4)throw Error(`Where for rank ${n} is not yet supported`);if(n===1)r="resRC",s="resRC";else{let o=["resRC.x","resRC.y","resRC.z","resRC.w"],i=[],l=[];for(let c=0;c<t.length;c++)l.push(`${o[c]}`),c<e&&i.push(`${o[c]}`);s=i.join(),r=l.join()}let a=yt(n);this.userCode=`
|
|
void main() {
|
|
${a} resRC = getOutputCoords();
|
|
float cVal = getC(${s});
|
|
if (cVal >= 1.0) {
|
|
setOutput(getA(${r}));
|
|
} else {
|
|
setOutput(getB(${r}));
|
|
}
|
|
}
|
|
`}};function Yae(e){let{inputs:t,backend:n}=e,{condition:s,t:r,e:a}=t,o=new Zae(s.shape.length,r.shape,r.shape.length);return n.runWebGLProgram(o,[s,r,a],zn(r.dtype,a.dtype))}var Jae={kernelName:Li,backendName:"webgl",kernelFunc:Yae},Qae=`
|
|
// Stable and Attracting Fixed Point (0, 1) for Normalized Weights.
|
|
// see: https://arxiv.org/abs/1706.02515
|
|
float scaleAlpha = ${E.SELU_SCALEALPHA};
|
|
float scale = ${E.SELU_SCALE};
|
|
return (x >= 0.0) ? scale * x : scaleAlpha * (exp(x) - 1.0);
|
|
`,eoe=st({opSnippet:Qae}),toe={kernelName:wu,backendName:"webgl",kernelFunc:eoe},j4="return 1.0 / (1.0 + exp(-1.0 * x));",noe=st({opSnippet:j4,packedOpSnippet:j4,cpuKernelImpl:yJ}),soe={kernelName:lo,backendName:"webgl",kernelFunc:noe},roe=`
|
|
if (isnan(x)) { return 0.0; }
|
|
return sign(x);
|
|
`,aoe=st({opSnippet:roe}),ooe={kernelName:ku,backendName:"webgl",kernelFunc:aoe},ioe=a4+`
|
|
return sin(x);
|
|
`,loe=st({opSnippet:ioe}),uoe={kernelName:io,backendName:"webgl",kernelFunc:loe},coe=`
|
|
float e2x = exp(x);
|
|
return (e2x - 1.0 / e2x) / 2.0;
|
|
`,doe=st({opSnippet:coe}),poe={kernelName:Wi,backendName:"webgl",kernelFunc:doe},hoe=`
|
|
float epsilon = 1.1920928955078125e-7;
|
|
float threshold = log(epsilon) + 2.0;
|
|
|
|
bool too_large = x > -threshold;
|
|
bool too_small = x < threshold;
|
|
|
|
float result;
|
|
float exp_x = exp(x);
|
|
|
|
if (too_large){
|
|
result = x;
|
|
}
|
|
else if (too_small){
|
|
result = exp_x;
|
|
}
|
|
else{
|
|
result = log(exp_x + 1.0);
|
|
}
|
|
return result;
|
|
`,foe=st({opSnippet:hoe}),moe={kernelName:Su,backendName:"webgl",kernelFunc:foe},goe=e=>{let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{blockShape:a,paddings:o}=s;v.assert(r.shape.length<=4,()=>"spaceToBatchND for rank > 4 with a WebGL backend not implemented yet");let i=a.reduce((A,x)=>A*x),l=[[0,0]];l.push(...o);for(let A=1+a.length;A<r.shape.length;++A)l.push([0,0]);let c=[],u=U4({inputs:{x:r},backend:n,attrs:{paddings:l,constantValue:0}}),d=E.getReshaped(u.shape,a,i,!1),p=E.getPermuted(d.length,a.length,!1),h=E.getReshapedPermuted(u.shape,a,i,!1),f=ve({inputs:{x:u},backend:n,attrs:{shape:d}}),m=Hn({inputs:{x:f},backend:n,attrs:{perm:p}}),g=ve({inputs:{x:m},backend:n,attrs:{shape:h}});return c.push(u),c.push(f),c.push(m),c.forEach(A=>n.disposeIntermediateTensorInfo(A)),g},Aoe={kernelName:Vi,backendName:"webgl",kernelFunc:goe};function yoe(e){let{inputs:t,backend:n}=e,{indices:s,values:r,denseShape:a,defaultValue:o}=t;if(a.shape.length!==1)throw new Error(`Dense shape must be a vector, saw:
|
|
${a.shape}`);if(s.shape.length!==2)throw new Error(`Indices must be a matrix, saw:
|
|
${s.shape}`);if(r.shape.length!==1)throw new Error(`Values must be a vector, saw:
|
|
${r.shape}`);if(o.shape.length!==0)throw new Error(`Default value must be a scalar, saw:
|
|
${o.shape}`);let i=n.readSync(s.dataId),l=n.readSync(r.dataId),c=n.readSync(a.dataId),u=n.readSync(o.dataId)[0],[d,p,h,f,m]=bJ(i,s.shape,s.dtype,l,r.dtype,c,u);return[n.makeTensorInfo(p,s.dtype,d),n.makeTensorInfo([p[0]],r.dtype,h),n.makeTensorInfo([f.length],"bool",new Uint8Array(f.map(g=>Number(g)))),n.makeTensorInfo([m.length],s.dtype,new Int32Array(m))]}var xoe={kernelName:Lh,backendName:"webgl",kernelFunc:yoe};function boe(e){let{inputs:t,backend:n}=e,{inputIndices:s,inputShape:r,newShape:a}=t;if(s.shape.length!==2)throw new Error(`Input indices should be a matrix but received shape ${s.shape}`);if(r.shape.length!==1)throw new Error(`Input shape should be a vector but received shape ${r.shape}`);if(a.shape.length!==1)throw new Error(`Target shape should be a vector but received shape ${a.shape}`);let o=Array.from(n.readSync(r.dataId)),i=n.readSync(s.dataId),l=Array.from(n.readSync(a.dataId)),[c,u,d]=vJ(i,s.shape,s.dtype,o,l);return[n.makeTensorInfo(u,s.dtype,c),n.makeTensorInfo([d.length],a.dtype,new Int32Array(d))]}var voe={kernelName:Bh,backendName:"webgl",kernelFunc:boe};function woe(e){let{inputs:t,backend:n}=e,{data:s,indices:r,segmentIds:a}=t;if(s.shape.length<1)throw new Error("Data should be at least 1 dimensional but received scalar");if(r.shape.length!==1)throw new Error(`Indices should be a vector but received shape
|
|
${r.shape}`);if(a.shape.length!==1)throw new Error(`Segment ids should be a vector but received shape
|
|
${a.shape}`);let o=n.readSync(s.dataId),i=n.readSync(r.dataId),l=n.readSync(a.dataId),[c,u]=HI(o,s.shape,s.dtype,i,l,!0);return n.makeTensorInfo(u,s.dtype,c)}var koe={kernelName:Wh,backendName:"webgl",kernelFunc:woe};function Soe(e){let{inputs:t,backend:n}=e,{data:s,indices:r,segmentIds:a}=t;if(s.shape.length<1)throw new Error("Data should be at least 1 dimensional but received scalar");if(r.shape.length!==1)throw new Error(`Indices should be a vector but received shape
|
|
${r.shape}`);if(a.shape.length!==1)throw new Error(`Segment ids should be a vector but received shape
|
|
${a.shape}`);let o=n.readSync(s.dataId),i=n.readSync(r.dataId),l=n.readSync(a.dataId),[c,u]=HI(o,s.shape,s.dtype,i,l);return n.makeTensorInfo(u,s.dtype,c)}var Ioe={kernelName:Vh,backendName:"webgl",kernelFunc:Soe};function Coe(e){let{inputs:t,backend:n,attrs:s}=e,{sparseIndices:r,sparseValues:a,defaultValue:o}=t,{outputShape:i}=s,{sliceRank:l,numUpdates:c,strides:u,outputSize:d}=E.calculateShapes(a,r,i),p=!1,h=new H4(c,l,r.shape.length,a.shape.length,u,[d,1],p),f=n.runWebGLProgram(h,[a,r,o],a.dtype),m=ve({inputs:{x:f},backend:n,attrs:{shape:i}});return n.disposeIntermediateTensorInfo(f),m}var Toe={kernelName:ld,backendName:"webgl",kernelFunc:Coe};function Noe(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{numOrSizeSplits:a,axis:o}=s,i=v.parseAxisParam(o,r.shape)[0],l=E.prepareSplitSize(r,a,i),c=r.shape.length,u=new Array(c).fill(0),d=r.shape.slice();return l.map(p=>{let h=[...d];h[i]=p;let f=uc({inputs:{x:r},backend:n,attrs:{begin:u,size:h}});return u[i]+=p,f})}var Eoe={kernelName:Ui,backendName:"webgl",kernelFunc:Noe},q4="return sqrt(x);",Roe=st({opSnippet:q4,packedOpSnippet:q4,cpuKernelImpl:wJ}),$oe={kernelName:uo,backendName:"webgl",kernelFunc:Roe},_oe="return x * x;",Doe=st({opSnippet:_oe}),Poe={kernelName:Iu,backendName:"webgl",kernelFunc:Doe},X4="return (a - b) * (a - b);",Foe=Cn({opSnippet:X4,packedOpSnippet:X4}),Ooe={kernelName:ho,backendName:"webgl",kernelFunc:Foe};function Moe({inputs:e,attrs:t,backend:n}){let{x:s}=e,r=Ar+`
|
|
return x > 0.0 ? 1.0 : float(${t.alpha});
|
|
`,a=new Vo(s.shape,r);return n.runWebGLProgram(a,[s],s.dtype)}var zoe={kernelName:Ao,backendName:"webgl",kernelFunc:Moe},Loe=class{constructor(e,t,n){this.variableNames=["x"],this.outputShape=n;let s=n.length,r=yt(n.length),a=yt(n.length),o="";if(s===1)o="coords * strides + begin";else{let i=0;o=n.map((l,c)=>(i++,n.length===1?`coords * strides[${c}] + begin[${c}]`:`coords[${i-1}] * strides[${c}] + begin[${c}]`)).join(",")}this.userCode=`
|
|
${r} begin = ${r}(${e});
|
|
${r} strides = ${r}(${t});
|
|
|
|
void main() {
|
|
${a} coords = getOutputCoords();
|
|
setOutput(getX(${o}));
|
|
}
|
|
`}};function Boe(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{begin:a,end:o,strides:i,beginMask:l,endMask:c,ellipsisMask:u,newAxisMask:d,shrinkAxisMask:p}=s,{finalShapeSparse:h,finalShape:f,isIdentity:m,sliceDim0:g,isSimpleSlice:A,begin:x,end:y,strides:b}=Pt.sliceInfo(r.shape,a,o,i,l,c,u,d,p),w;if(m)w=ve({inputs:{x:r},backend:n,attrs:{shape:f}});else if(g||A){v.assert(r.shape.length>=1,()=>`Input must have rank at least 1, got: ${r.shape.length}`);let C=Pt.computeOutShape(x,y,b),N=uc({inputs:{x:r},backend:n,attrs:{begin:x,size:C}});w=ve({inputs:{x:N},backend:n,attrs:{shape:f}}),n.disposeIntermediateTensorInfo(N)}else if(n.shouldExecuteOnCPU([r])){let N=n.readSync(r.dataId),R=ze(r.shape,r.dtype,N),F=kJ(h,R,b,x);w=n.makeTensorInfo(f,r.dtype,F.values)}else{let N=new Loe(x,b,h);w=n.runWebGLProgram(N,[r],r.dtype)}let k=ve({inputs:{x:w},backend:n,attrs:{shape:f}});return n.disposeIntermediateTensorInfo(w),k}var Woe={kernelName:Gi,backendName:"webgl",kernelFunc:Boe};function Voe(e){let{inputs:t,backend:n,attrs:s}=e,{separator:r,nGramWidths:a,leftPad:o,rightPad:i,padWidth:l,preserveShortSequences:c}=s,{data:u,dataSplits:d}=t,p=n.readSync(u.dataId),h=n.readSync(d.dataId),[f,m]=SJ(p,h,r,a,o,i,l,c);return[n.makeTensorInfo([f.length],"string",f),n.makeTensorInfo(d.shape,"int32",m)]}var Uoe={kernelName:ud,backendName:"webgl",kernelFunc:Voe};function Goe(e){let{inputs:t,backend:n,attrs:s}=e,{skipEmpty:r}=s,{input:a,delimiter:o}=t;if(a.dtype!=="string")throw new Error("Input must be of datatype string");if(a.shape.length!==1)throw new Error(`Input must be a vector, got shape: ${a.shape}`);if(o.shape.length!==0)throw new Error(`Delimiter must be a scalar, got shape: ${o.shape}`);let i=n.readSync(a.dataId),l=n.readSync(o.dataId)[0],[c,u,d]=IJ(i,l,r),p=u.length;return[n.makeTensorInfo([p,2],"int32",c),n.makeTensorInfo([p],"string",u),n.makeTensorInfo([2],"int32",new Int32Array(d))]}var Hoe={kernelName:Uh,backendName:"webgl",kernelFunc:Goe};function joe(e){let{inputs:t,backend:n,attrs:s}=e,{numBuckets:r}=s,{input:a}=t;if(a.dtype!=="string")throw new Error("Input must be of datatype string");if(r<=0)throw new Error("Number of buckets must be at least 1");let o=n.readSync(a.dataId),i=CJ(o,r);return n.makeTensorInfo(a.shape,"int32",i)}var qoe={kernelName:Gh,backendName:"webgl",kernelFunc:joe},Xoe="return tan(x);",Koe=st({opSnippet:Xoe}),Zoe={kernelName:Hi,backendName:"webgl",kernelFunc:Koe},Yoe=`
|
|
float e2x = exp(-2.0 * abs(x));
|
|
return sign(x) * (1.0 - e2x) / (1.0 + e2x);
|
|
`,Joe=st({opSnippet:Yoe}),Qoe={kernelName:mo,backendName:"webgl",kernelFunc:Joe},eie=class{constructor(e,t){this.variableNames=["A"];let n=new Array(e.length);for(let a=0;a<n.length;a++)n[a]=e[a]*t[a];this.outputShape=n,this.rank=n.length;let s=yt(this.rank),r=tie(e);this.userCode=`
|
|
void main() {
|
|
${s} resRC = getOutputCoords();
|
|
setOutput(getA(${r}));
|
|
}
|
|
`}};function tie(e){let t=e.length;if(t>5)throw Error(`Tile for rank ${t} is not yet supported`);if(t===1)return`imod(resRC, ${e[0]})`;let n=["resRC.x","resRC.y","resRC.z","resRC.w","resRC.u"],s=[];for(let r=0;r<e.length;r++)s.push(`imod(${n[r]}, ${e[r]})`);return s.join()}function K4(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{reps:a}=s;if(r.dtype==="string"||r.shape.length>5){let l=n.readSync(r.dataId),c=r.dtype==="string"?l.map(p=>v.decodeString(p)):l,u=ze(r.shape,r.dtype,c),d=NJ(u,a);return n.makeTensorInfo(d.shape,d.dtype,d.values)}let o=new eie(r.shape,a);return n.runWebGLProgram(o,[r],r.dtype)}var nie={kernelName:jr,backendName:"webgl",kernelFunc:K4},sie=class{constructor(e){this.variableNames=["x","indices"],this.customUniforms=[{name:"n",type:"int"},{name:"firstPass",type:"int"},{name:"negativeInf",type:"float"},{name:"dir",type:"int"},{name:"inc",type:"int"}],this.outputShape=e,this.userCode=`
|
|
void main() {
|
|
ivec2 coords = getOutputCoords();
|
|
int batch = coords[0];
|
|
int elemIdx = coords[1];
|
|
|
|
// We compare elements pair-wise within a group of size 2 * inc.
|
|
// The comparing rule for each group alternates between ascending
|
|
// and descending. Within each group, we compare each pair at
|
|
// positions i and i+inc. To decide whether an element at position i
|
|
// is x0 or x1, we mod it by 2 * inc, if the result is smaller than
|
|
// inc, it is in the first half of the group, we denote it as x0,
|
|
// otherwise we denote it as x1.
|
|
// For example, as shown in the Bitonic top K paper referenced above,
|
|
// Figure5(a) shows that element[1] is in the
|
|
// second half of the group when group size is 2, but it is in the
|
|
// first half of the group when group size is 4.
|
|
|
|
bool isFirstInPair = imod(elemIdx, 2 * inc) < inc;
|
|
int i = isFirstInPair ? elemIdx : elemIdx - inc;
|
|
|
|
int i0 = firstPass == 1 ? i : int(getIndices(batch, i));
|
|
int i1 = firstPass == 1 ? i + inc : int(getIndices(batch, i + inc));
|
|
float x0 = i0 < n ? getX(batch, i0) : negativeInf;
|
|
float x1 = i1 < n ? getX(batch, i1) : negativeInf;
|
|
|
|
// Denotes which direction indices are in (ascending or descending).
|
|
bool reverse = imod(elemIdx, 2 * dir) >= dir;
|
|
bool isGreater = x0 > x1 || (x0 == x1 && i1 > i0);
|
|
if (reverse == isGreater) { // Elements in opposite order of direction
|
|
int iTemp = i0;
|
|
i0 = i1;
|
|
i1 = iTemp;
|
|
}
|
|
if (isFirstInPair) {
|
|
setOutput(float(i0));
|
|
} else {
|
|
setOutput(float(i1));
|
|
}
|
|
}
|
|
`}},rie=class{constructor(e){this.variableNames=["x","indices"],this.customUniforms=[{name:"n",type:"int"},{name:"firstPass",type:"int"},{name:"k",type:"int"}],this.outputShape=e,this.userCode=`
|
|
void main() {
|
|
// Takes max of indices (0, k), (1, k + 1), (2, k + 2) ...
|
|
ivec2 coords = getOutputCoords();
|
|
int batch = coords[0];
|
|
int elemIdx = coords[1];
|
|
|
|
// The output size is half of the previous size.
|
|
// If the previous sequence is | | | | _ _ _ _ | | | | _ _ _ _ (k=4),
|
|
// we only need to output the indices at positions |, the indices at
|
|
// positions _ can be thrown away, see Figure5(b) After Phase 2
|
|
// (Merge phase) in the Bitonic Top K paper referenced above.
|
|
// For example, the paper shows we only need to output the orange bars.
|
|
// The output sequence should look like this | | | | | | | |.
|
|
// Because the sequence is halved, to map the output index back
|
|
// to the previous sequence to find the corresponding value,
|
|
// we need to double the index. When we double the index,
|
|
// we basically interpolate a position, so 2i looks like
|
|
// | _ | _ | _ | _ | _ | _ | _. We move the | to the first k position
|
|
// of each 2k positions by - elemIdx % k. E.g. for output at
|
|
// index 4,5,6,7, we want to get the corresponding element at
|
|
// original index 8,9,10,11, for output at index 8,9,10,11,
|
|
// we want to get the corresponding element at original index
|
|
// 16,17,18,19, so on and so forth.
|
|
|
|
int i = elemIdx < k ? elemIdx : (elemIdx * 2 - imod(elemIdx, k));
|
|
int i0 = firstPass == 1 ? i : int(getIndices(batch, i));
|
|
int i1 = firstPass == 1 ? i + k : int(getIndices(batch, i + k));
|
|
|
|
float x0 = getX(batch, i0);
|
|
float x1 = i1 < n ? getX(batch, i1) : x0;
|
|
|
|
setOutput(x0 >= x1 ? float(i0) : float(i1));
|
|
}
|
|
`}};function Nl(e,t){t!==null&&e.disposeIntermediateTensorInfo(t)}function Z4(e){let t=1;for(;t<e;)t*=2;return t}function aie(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{k:a,sorted:o}=s,i=K().getNumber("TOPK_LAST_DIM_CPU_HANDOFF_SIZE_THRESHOLD"),l=K().getNumber("TOPK_K_CPU_HANDOFF_THRESHOLD"),c=r.shape,u=c[c.length-1];if(n.shouldExecuteOnCPU([r])||u<i||a>l){let F=n.readSync(r.dataId),[_,P]=EJ(F,c,r.dtype,a,o);return[n.makeTensorInfo(_.shape,_.dtype,_.values),n.makeTensorInfo(P.shape,P.dtype,P.values)]}if(a===0)return c[c.length-1]=0,[n.makeTensorInfo(c,r.dtype,[]),n.makeTensorInfo(c,"int32",[])];if(u===1)return[r,gp({attrs:{shape:c,dtype:"int32",value:0},backend:n})];let d=n.texData.get(r.dataId),p=d!==null&&d.isPacked,h=p?n.unpackTensor(r):r,m=v.sizeFromShape(c)/u,g=ve({inputs:{x:h},attrs:{shape:[m,u]},backend:n});p&&Nl(n,h);let A=Z4(a),x=Z4(u),y=null,b=()=>y===null?[g,g]:[g,y],w=(F,_,P)=>{let T=b(),M=new sie(P),j=[[u],[y===null?1:0],[Number.NEGATIVE_INFINITY],[F],[_]],z=y;y=n.runWebGLProgram(M,T,"int32",j),Nl(n,z)};for(let F=1;F<A;F*=2){let _=F*2;for(let P=F;P>=1;P/=2)w(_,P,[m,x])}for(let F=x;F>A;F/=2){let _=b(),P=new rie([m,F/2]),M=[[u],[y===null?1:0],[A]],U=y;y=n.runWebGLProgram(P,_,"int32",M),Nl(n,U);let j=A/2,z=j*2;for(let X=j;X>=1;X/=2)w(z,X,y.shape)}let k=y;y=uc({inputs:{x:y},backend:n,attrs:{begin:0,size:[m,a]}}),Nl(n,k);let C=F4({inputs:{x:g,indices:y},backend:n,attrs:{axis:1,batchDims:1}});Nl(n,g);let N=c.slice(0,-1);N.push(a),k=y,y=ve({inputs:{x:y},attrs:{shape:N},backend:n}),Nl(n,k);let R=C;return C=ve({inputs:{x:C},attrs:{shape:N},backend:n}),Nl(n,R),[C,y]}var oie={kernelName:ji,backendName:"webgl",kernelFunc:aie},iie=class{constructor(e,t,n,s,r,a){this.variableNames=["Image","Transforms"],this.outputShape=a;let o=n==="nearest"?1:2,i;switch(s){case"constant":i=1;break;case"reflect":i=2;break;case"wrap":i=3;break;case"nearest":i=4;break;default:i=1;break}this.userCode=`
|
|
float mapCoord(float outCoord, float len) {
|
|
float inCoord = outCoord;
|
|
if(${i} == 2) {
|
|
if (inCoord < 0.0) {
|
|
if (len <= 1.0) {
|
|
inCoord = 0.0;
|
|
} else {
|
|
float sz2 = 2.0 * len;
|
|
if (inCoord < sz2) {
|
|
inCoord = sz2 * float(int(float(-inCoord / sz2))) +
|
|
inCoord;
|
|
}
|
|
inCoord = inCoord < -len ? inCoord + sz2 : -inCoord - 1.0;
|
|
}
|
|
} else if (inCoord > len - 1.0) {
|
|
if (len <= 1.0) {
|
|
inCoord = 0.0;
|
|
} else {
|
|
float sz2 = 2.0 * len;
|
|
inCoord -= sz2 * float(int(float(inCoord / sz2)));
|
|
if (inCoord >= len) {
|
|
inCoord = sz2 - inCoord - 1.0;
|
|
}
|
|
}
|
|
}
|
|
return clamp(inCoord, 0.0, len - 1.0);
|
|
} else if (${i} == 3) {
|
|
if (inCoord < 0.0) {
|
|
if (len <= 1.0) {
|
|
inCoord = 0.0;
|
|
} else {
|
|
float sz = len - 1.0;
|
|
inCoord += len * (float(int(float(-inCoord / sz))) + 1.0);
|
|
}
|
|
} else if (inCoord > len - 1.0) {
|
|
if (len <= 1.0) {
|
|
inCoord = 0.0;
|
|
} else {
|
|
float sz = len - 1.0;
|
|
inCoord -= len * float(int(float(inCoord / sz)));
|
|
}
|
|
}
|
|
return clamp(inCoord, 0.0, len - 1.0);
|
|
} else if (${i} == 4) {
|
|
return clamp(outCoord, 0.0, len - 1.0);
|
|
} else {
|
|
return outCoord;
|
|
}
|
|
}
|
|
|
|
float readWithFillValue(int batch, int coordY, int coordX,
|
|
int channel) {
|
|
float outputValue;
|
|
if (0 <= coordY && coordY < ${e} && 0 <= coordX && coordX < ${t}) {
|
|
outputValue = getImage(batch, coordY, coordX, channel);
|
|
} else {
|
|
outputValue = float(${r});
|
|
}
|
|
return outputValue;
|
|
}
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
float outputValue;
|
|
int batch = coords[0];
|
|
int x = coords[2];
|
|
int y = coords[1];
|
|
int channel = coords[3];
|
|
float xf = float(x);
|
|
float yf = float(y);
|
|
float a1 = getTransforms(batch, 0);
|
|
float a2 = getTransforms(batch, 1);
|
|
float a3 = getTransforms(batch, 2);
|
|
float b1 = getTransforms(batch, 3);
|
|
float b2 = getTransforms(batch, 4);
|
|
float b3 = getTransforms(batch, 5);
|
|
float c1 = getTransforms(batch, 6);
|
|
float c2 = getTransforms(batch, 7);
|
|
float projection = c1 * xf + c2 * yf + 1.0;
|
|
if (projection == 0.0) {
|
|
outputValue = float(${r});
|
|
} else {
|
|
float inX = (a1 * xf + a2 * yf + a3) / projection;
|
|
float inY = (b1 * xf + b2 * yf + b3) / projection;
|
|
float mapX = mapCoord(inX, float(${t}));
|
|
float mapY = mapCoord(inY, float(${e}));
|
|
|
|
if (${o} == 1) {
|
|
int coordY = int(round(mapY));
|
|
int coordX = int(round(mapX));
|
|
outputValue = readWithFillValue(batch, coordY, coordX,
|
|
channel);
|
|
} else {
|
|
float yFloor = floor(mapY);
|
|
float xFloor = floor(mapX);
|
|
float yCeil = yFloor + 1.0;
|
|
float xCeil = xFloor + 1.0;
|
|
float valueYFloor = (xCeil - mapX) *
|
|
readWithFillValue(batch, int(yFloor), int(xFloor), channel) +
|
|
(mapX - xFloor) *
|
|
readWithFillValue(batch, int(yFloor), int(xCeil), channel);
|
|
float valueYCeil = (xCeil - mapX) *
|
|
readWithFillValue(batch, int(yCeil), int(xFloor), channel) +
|
|
(mapX - xFloor) *
|
|
readWithFillValue(batch, int(yCeil), int(xCeil), channel);
|
|
outputValue = (yCeil - mapY) * valueYFloor +
|
|
(mapY - yFloor) * valueYCeil;
|
|
}
|
|
}
|
|
setOutput(outputValue);
|
|
}
|
|
`}};function lie(e){let{inputs:t,backend:n,attrs:s}=e,{image:r,transforms:a}=t,{interpolation:o,fillMode:i,fillValue:l,outputShape:c}=s,[u,d,p,h]=r.shape,[f,m]=c!=null?c:[d,p],g=[u,f,m,h],A=new iie(d,p,o,i,l,g);return n.runWebGLProgram(A,[r,a],"float32")}var uie={kernelName:qi,backendName:"webgl",kernelFunc:lie};function cie(e){let{inputs:t,attrs:n,backend:s}=e,{axis:r}=n,{x:a}=t;nc(a,"unique"),console.warn("WARNING: ","UI might be locked temporarily as data is being downloaded");let o=s.readSync(a.dataId),{outputValues:i,outputShape:l,indices:c}=RJ(o,r,a.shape,a.dtype);return[s.makeTensorInfo(l,a.dtype,i),s.makeTensorInfo([c.length],"int32",c)]}var die={kernelName:Hh,backendName:"webgl",kernelFunc:cie};function pie(e){let{inputs:t,backend:n,attrs:s}=e,{value:r}=t,{axis:a}=s;a<0&&(a+=r.shape.length);let o=r,i=o.shape.length,l=r.shape[a],c=new Array(i-1),u=0;for(let m=0;m<i;m++)m!==a&&(c[u++]=o.shape[m]);let d=[],p=new Array(i).fill(0),h=o.shape.slice();h[a]=1;let f=new Array(l);for(let m=0;m<f.length;m++){p[a]=m;let g=uc({inputs:{x:o},backend:n,attrs:{begin:p,size:h}}),A=ve({inputs:{x:g},backend:n,attrs:{shape:c}});f[m]=A,d.push(g)}return d.forEach(m=>n.disposeIntermediateTensorInfo(m)),f}var hie={kernelName:Xi,backendName:"webgl",kernelFunc:pie},fie=class{constructor(e,t){this.variableNames=["x","segmentIds"];let n=e.windowSize,s=e.batchSize,r=e.inSize,a=e.numSegments,o=a*Math.ceil(r/n);this.outputShape=[s,o];let i="0.0",l="sumValue",c=Math.floor(n/4)*4,u=n%4,d=`
|
|
sumValue += dot(values, segFilter);
|
|
`,p="";r%n>0&&(p=`
|
|
if (inIdx < 0 || inIdx >= ${r}) {
|
|
return initializationValue;
|
|
}
|
|
`);let h="";r%n>0&&(h=`
|
|
if (inIdx < 0 || inIdx >= ${r}) {
|
|
return -1.0;
|
|
}
|
|
`),this.userCode=`
|
|
const float initializationValue = ${i};
|
|
|
|
float getValue(int batch, int inIdx) {
|
|
${p}
|
|
return getX(batch, inIdx);
|
|
}
|
|
|
|
float getSegmentIdAtIndex(int inIdx) {
|
|
${h}
|
|
return getSegmentIds(inIdx);
|
|
}
|
|
|
|
void main() {
|
|
ivec2 coords = getOutputCoords();
|
|
int batch = coords[0];
|
|
int outIdx = coords[1];
|
|
int inOffset = int(floor(float(outIdx) / float(
|
|
${a})) * float(${n}));
|
|
int currentSeg = int(mod(float(outIdx), float(${a})));
|
|
|
|
float sumValue = 0.0;
|
|
|
|
for (int i = 0; i < ${c}; i += 4) {
|
|
int inIdx = inOffset + i;
|
|
vec4 values = vec4(
|
|
getValue(batch, inIdx),
|
|
getValue(batch, inIdx + 1),
|
|
getValue(batch, inIdx + 2),
|
|
getValue(batch, inIdx + 3)
|
|
);
|
|
|
|
vec4 segFilter = vec4(
|
|
int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,
|
|
int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,
|
|
int(getSegmentIdAtIndex(inIdx + 2)) == currentSeg ? 1 : 0,
|
|
int(getSegmentIdAtIndex(inIdx + 3)) == currentSeg ? 1 : 0
|
|
);
|
|
|
|
${d}
|
|
}
|
|
|
|
int inIdx = inOffset + ${c};
|
|
if (${u===1}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, inIdx),
|
|
initializationValue,
|
|
initializationValue,
|
|
initializationValue
|
|
);
|
|
|
|
int inIdxSeg = int(getSegmentIdAtIndex(inIdx));
|
|
|
|
vec4 segFilter = vec4(
|
|
int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,
|
|
0,
|
|
0,
|
|
0
|
|
);
|
|
|
|
${d}
|
|
} else if (${u===2}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, inIdx),
|
|
getValue(batch, inIdx + 1),
|
|
initializationValue,
|
|
initializationValue
|
|
);
|
|
|
|
vec4 segFilter = vec4(
|
|
int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,
|
|
int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,
|
|
0,
|
|
0
|
|
);
|
|
|
|
${d}
|
|
} else if (${u===3}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, inIdx),
|
|
getValue(batch, inIdx + 1),
|
|
getValue(batch, inIdx + 2),
|
|
initializationValue
|
|
);
|
|
|
|
vec4 segFilter = vec4(
|
|
int(getSegmentIdAtIndex(inIdx)) == currentSeg ? 1 : 0,
|
|
int(getSegmentIdAtIndex(inIdx + 1)) == currentSeg ? 1 : 0,
|
|
int(getSegmentIdAtIndex(inIdx + 2)) == currentSeg ? 1 : 0,
|
|
0
|
|
);
|
|
|
|
${d}
|
|
}
|
|
setOutput(${l});
|
|
}
|
|
`}};function mie(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,segmentIds:a}=t,{numSegments:o}=s,i=r.shape.length,l=[],c=0,u=E.getAxesPermutation([c],i),d=r;u!=null&&(d=Hn({inputs:{x:r},backend:n,attrs:{perm:u}}),l.push(d),c=E.getInnerMostAxes(1,i)[0]);let p=E.segment_util.computeOutShape(d.shape,c,o),h=v.sizeFromShape([d.shape[c]]),f=ve({inputs:{x:d},backend:n,attrs:{shape:[-1,h]}});l.push(f);let m=bd(r.dtype),g=(b,w,k,C,N)=>{let R=b.shape[0],F=b.shape[1],_=E.segment_util.segOpComputeOptimalWindowSize(F,N),P={windowSize:_,inSize:F,batchSize:R,numSegments:N},T=new fie(P,w),M=n.compileAndRun(T,[b,k],C);if(l.push(M),M.shape[1]===N)return M;let U=G4({backend:n,attrs:{start:0,stop:N,step:1,dtype:"float32"}}),j=K4({inputs:{x:U},backend:n,attrs:{reps:[F/_]}});return l.push(U),l.push(j),g(M,w,j,C,N)},A=g(f,"unsortedSegmentSum",a,m,o),x=ve({inputs:{x:A},backend:n,attrs:{shape:p}}),y=x;if(u!=null){l.push(x);let b=E.getUndoAxesPermutation(u);y=Hn({inputs:{x:y},backend:n,attrs:{perm:b}})}return l.forEach(b=>n.disposeIntermediateTensorInfo(b)),y}var gie={kernelName:cd,backendName:"webgl",kernelFunc:mie},Aie=[Hse,Xse,EQ,$Q,PQ,MQ,LQ,VQ,GQ,jQ,ZQ,JQ,tee,ree,dee,iee,fee,yee,gee,wee,See,Cee,Ree,Mee,Lee,Wee,qee,Kee,Qee,nte,cQ,ite,Ate,xte,dte,kte,Ite,vte,Nte,$te,Pte,Ote,zte,Wte,qte,Kte,Ute,Jte,tne,sne,ine,dne,mne,yne,xne,bne,wne,Sne,Cne,Nne,Rne,Pne,Mne,Bne,Vne,Hne,Xne,Jne,nse,uQ,rse,ate,ise,cse,hse,pQ,Ase,vse,kse,Rse,Tse,Pse,Mse,Wse,Zse,rre,nre,lre,cre,pre,ere,fre,gre,bre,Sre,Nre,Ore,AQ,zre,Wre,Gre,qre,Uee,Zre,Jre,eae,sae,iae,fQ,uae,cae,Gee,_re,hae,vae,Aae,xQ,Iae,Nae,_ae,Fae,Lae,Wae,Gae,qae,Kae,Jae,toe,soe,ooe,uoe,poe,Fee,Pre,moe,Aoe,xoe,voe,koe,Ioe,Toe,Eoe,$oe,Poe,Ooe,zoe,Woe,Uoe,Hoe,qoe,Dre,CQ,Zoe,Qoe,nie,oie,uie,TQ,die,hie,gie,Yre];for(let e of Aie)rr(e);var Lr=K();Lr.registerFlag("WEBGPU_DEFERRED_SUBMIT_BATCH_SIZE",()=>15);Lr.registerFlag("WEBGPU_CPU_FORWARD",()=>!0);Lr.registerFlag("WEBGPU_MATMUL_WORK_PER_THREAD",()=>4);Lr.registerFlag("WEBGPU_USE_NAIVE_CONV2D",()=>!1);Lr.registerFlag("WEBGPU_USE_NAIVE_CONV2D_TRANSPOSE",()=>!1);Lr.registerFlag("WEBGPU_CONV_SEPARATE_IM2COL_SHADER",()=>!1);Lr.registerFlag("WEBGPU_USE_LOW_POWER_GPU",()=>!1);Lr.registerFlag("WEBGPU_CPU_HANDOFF_SIZE_THRESHOLD",()=>1e3);Lr.registerFlag("WEBGPU_USE_PROFILE_TOOL",()=>!1);Lr.registerFlag("WEBGPU_USE_IMPORT",()=>!1);function yie(e,t){if(Math.max(...e)>3)throw new Error("Cannot symbolically compute strides for rank > 4 tensor.");let n=e.length,s=e.map(a=>`${t}[${a}]`),r=new Array(n-1);r[n-2]=s[n-1];for(let a=n-3;a>=0;--a)r[a]=`(${r[a+1]} * ${s[a+1]})`;return r}function vn(e){if(e<=1)return"i32";if(e===2)return"vec2<i32>";if(e===3)return"vec3<i32>";if(e===4)return"vec4<i32>";throw Error(`GPU for rank ${e} is not yet supported`)}function qm(e,t){return e==="float32"?t?"vec4<f32>":"f32":e==="int32"||e==="bool"?t?"vec4<i32>":"i32":e}function Xm(){return`
|
|
[[stage(compute), workgroup_size(workGroupSizeX, workGroupSizeY, workGroupSizeZ)]]
|
|
`}function hx(){return`
|
|
${Xm()}
|
|
fn main([[builtin(local_invocation_id)]] localId : vec3<u32>,
|
|
[[builtin(global_invocation_id)]] globalId : vec3<u32>,
|
|
[[builtin(num_workgroups)]] numWorkgroups: vec3<u32>)
|
|
`}function Go(){return`
|
|
${Xm()}
|
|
fn main([[builtin(local_invocation_id)]] localId : vec3<u32>,
|
|
[[builtin(global_invocation_id)]] globalId : vec3<u32>)
|
|
`}function tt(){return`
|
|
${hx()} {
|
|
let index = getGlobalIndex(globalId, localId, numWorkgroups);
|
|
`}function xie(e,t,n,s=!1){let r=`
|
|
let workGroupSizeX = ${n.workGroupSize[0]}u;
|
|
let workGroupSizeY = ${n.workGroupSize[1]}u;
|
|
let workGroupSizeZ = ${n.workGroupSize[2]}u;`;if(s===!0){let h=Q4(t.shape),f=`
|
|
[[block]] struct Matrix0 {
|
|
numbers: array<${qm(t.dtype,n.isVec4)}>;
|
|
};
|
|
[[block]] struct Uniform {
|
|
size : i32;
|
|
numChannels : i32;
|
|
outShapeStrides : vec2<i32>;
|
|
dispatchSize : vec3<u32>;
|
|
};
|
|
|
|
[[group(0), binding(0)]] var<storage, write> result : Matrix0;
|
|
[[group(0), binding(2)]] var<uniform> uniforms: Uniform;
|
|
`;return[Y4,f,r,J4,h,n.getUserCode()].join(`
|
|
`)}let a=[],o="[[block]] struct Uniforms { NAN : f32; ";n.variableNames.forEach((h,f)=>{o+=`${h.charAt(0).toLowerCase()+h.slice(1)}Shape : ${vn(e[f].shape.length)}; `}),o+=`outShape : ${vn(t.shape.length)} ; `;let i=t.shape.length-1;o+=`
|
|
outShapeStrides: ${vn(i)}; `,n.size&&(o+="size : i32; "),n.uniforms&&(o+=n.uniforms),o+="};",a.push(o),n.atomic?a.push(`
|
|
[[block]] struct Matrix0 {
|
|
numbers: array<atomic<i32>>;
|
|
};
|
|
|
|
[[group(0), binding(0)]] var<storage, read_write> result : Matrix0;
|
|
`):a.push(`
|
|
[[block]] struct Matrix0 {
|
|
numbers: array<${qm(t.dtype,n.isVec4)}>;
|
|
};
|
|
|
|
[[group(0), binding(0)]] var<storage, write> result : Matrix0;
|
|
`),n.variableNames.forEach((h,f)=>{a.push(`
|
|
[[block]] struct Matrix${1+f} {
|
|
numbers: array<${qm(e[f].dtype,n.isVec4)}>;
|
|
};
|
|
[[group(0), binding(${1+f})]] var<storage, read> ${h} : Matrix${1+f};
|
|
`)}),o!==""&&a.push(`
|
|
[[group(0), binding(${1+n.variableNames.length})]] var<uniform> uniforms : Uniforms;
|
|
`),a.push(r);let[l,c]=Iie(t.shape,n.dispatchLayout),u=Q4(t.shape),d=[Y4,a.join(`
|
|
`),J4,u,l,bie(t.shape.length)];if(n.atomic||d.push(vie(t.shape,t.dtype,n.isVec4)),c===t.shape.length){let h=e.map(f=>wie(f,t.shape,n.isVec4,n.dispatchLayout.x.length===t.shape.length)).join(`
|
|
`);d.push(h)}return d.push(n.getUserCode()),d.join(`
|
|
`)}var Y4=`
|
|
fn idiv(a: i32, b: i32, sign: f32) -> i32 {
|
|
var res: i32 = a / b;
|
|
let mod: i32 = a % b;
|
|
if (sign < 0. && mod != 0) {
|
|
res = res - 1;
|
|
}
|
|
return res;
|
|
}
|
|
|
|
fn isNanCustom(val : f32) -> bool {
|
|
if (val > 0.0) {
|
|
return false;
|
|
}
|
|
if (val < 0.0) {
|
|
return false;
|
|
}
|
|
if (val == 0.0) {
|
|
return false;
|
|
}
|
|
return true;
|
|
}
|
|
|
|
fn isNanCustomVec4F32(val : vec4<f32>) -> vec4<f32> {
|
|
var res = vec4<f32> (0.0);
|
|
for (var i = 0u; i < 4u; i = i + 1u) {
|
|
if (isNanCustom(val[i])) {
|
|
res[i] = 1.0;
|
|
} else {
|
|
res[i] = 0.0;
|
|
}
|
|
}
|
|
return res;
|
|
}
|
|
|
|
// Checks whether coordinates lie within the bounds of the shape.
|
|
fn coordsInBounds4D(coord : vec4<i32>, shape : vec4<i32>) -> bool {
|
|
return all(coord >= vec4<i32>(0)) &&
|
|
all(coord < shape);
|
|
}
|
|
|
|
fn coordsInBounds3D(coord : vec3<i32>, shape : vec3<i32>) -> bool {
|
|
return all(coord >= vec3<i32>(0)) &&
|
|
all(coord < shape);
|
|
}
|
|
|
|
fn coordsInBounds2D(coord : vec2<i32>, shape : vec2<i32>) -> bool {
|
|
return all(coord >= vec2<i32>(0)) &&
|
|
all(coord < shape);
|
|
}
|
|
`,J4=`
|
|
fn getFlatIndex1D(coord : i32, shape : i32) -> i32 {
|
|
return coord;
|
|
}
|
|
|
|
fn getFlatIndex2D(coords : vec2<i32>, shape : vec2<i32>) -> i32 {
|
|
return i32(dot(vec2<f32>(coords), vec2<f32>(f32(shape.y), 1.0)));
|
|
}
|
|
|
|
fn getFlatIndex3D(coords : vec3<i32>, shape : vec3<i32>) -> i32 {
|
|
return i32(dot(vec3<f32>(coords), vec3<f32>(f32(shape.y) * f32(shape.z), f32(shape.z), 1.0)));
|
|
}
|
|
|
|
fn getFlatIndex4D(coords : vec4<i32>, shape : vec4<i32>) -> i32 {
|
|
return i32(dot(vec4<f32>(coords), vec4<f32>(
|
|
f32(shape.y) * f32(shape.z) * f32(shape.w), f32(shape.z) * f32(shape.w), f32(shape.w), 1.0)));
|
|
}
|
|
|
|
// Only used when the y/z dimension of workgroup size is 1.
|
|
fn getGlobalIndex(globalId : vec3<u32>, localId : vec3<u32>, numWorkgroups: vec3<u32>) -> i32 {
|
|
if (numWorkgroups.y == 1u && numWorkgroups.z == 1u) {
|
|
return i32(globalId.x);
|
|
}
|
|
|
|
let localInvocationIndex = localId.z * workGroupSizeX * workGroupSizeY +
|
|
localId.y * workGroupSizeX + localId.x;
|
|
let workGroupID = (globalId - localId)/vec3<u32>(
|
|
workGroupSizeX, workGroupSizeY, workGroupSizeZ);
|
|
|
|
return i32((workGroupID.z * numWorkgroups.x * numWorkgroups.y +
|
|
workGroupID.y * numWorkgroups.x + workGroupID.x) *
|
|
(workGroupSizeX * workGroupSizeY * workGroupSizeZ) +
|
|
localInvocationIndex);
|
|
}
|
|
`;function bie(e){let t="";switch(e){case 0:case 1:t+=`
|
|
fn getOutputFlatIndex(coords : i32) -> i32 {
|
|
return coords;
|
|
}
|
|
`;break;case 2:t+=`
|
|
fn getOutputFlatIndex(coords : vec2<i32>) -> i32 {
|
|
return i32(dot(vec2<f32>(coords), vec2<f32>(f32(uniforms.outShapeStrides), 1.0)));
|
|
}
|
|
`;break;case 3:t+=`
|
|
fn getOutputFlatIndex(coords : vec3<i32>) -> i32 {
|
|
return i32(dot(vec3<f32>(coords), vec3<f32>(f32(uniforms.outShapeStrides.x), f32(uniforms.outShapeStrides.y), 1.0)));
|
|
}
|
|
`;break;case 4:t+=`
|
|
fn getOutputFlatIndex(coords : vec4<i32>) -> i32 {
|
|
return i32(dot(vec4<f32>(coords), vec4<f32>(
|
|
f32(uniforms.outShapeStrides.x), f32(uniforms.outShapeStrides.y), f32(uniforms.outShapeStrides.z), 1.0)));
|
|
}
|
|
`;break;default:v.assert(!1,()=>`Unsupported ${e}D shape`);break}return t}function vie(e,t,n){let s=e.length,r=qm(t,n),a;if(n?a=`fn setOutputFlat(flatIndex : i32, value : vec4<f32>) {
|
|
result.numbers[flatIndex] = ${r}(value);
|
|
}
|
|
fn setOutputFlatI32(flatIndex : i32, value : vec4<i32>) {
|
|
result.numbers[flatIndex] = ${r}(value);
|
|
}`:a=`fn setOutputFlat(flatIndex : i32, value : f32) {
|
|
result.numbers[flatIndex] = ${r}(value);
|
|
}
|
|
fn setOutputFlatI32(flatIndex : i32, value : i32) {
|
|
result.numbers[flatIndex] = ${r}(value);
|
|
}`,s>=2){let o=["d0","d1","d2","d3"].slice(0,s),i=vn(s);n?a+=`
|
|
fn setOutput(${o.map(l=>`${l} : i32`).join(", ")}, value : vec4<f32>) {
|
|
let flatIndex = getOutputFlatIndex(${i}(${o.join(", ")}));
|
|
setOutputFlat(flatIndex / 4, value);
|
|
}
|
|
fn setOutputI32(${o.map(l=>`${l} : i32`).join(", ")}, value : vec4<i32>) {
|
|
let flatIndex = getOutputFlatIndex(${i}(${o.join(", ")}));
|
|
setOutputFlatI32(flatIndex / 4, value);
|
|
}
|
|
`:a+=`
|
|
fn setOutput(${o.map(l=>`${l} : i32`).join(", ")}, value : f32) {
|
|
let flatIndex = getOutputFlatIndex(${i}(${o.join(", ")}));
|
|
setOutputFlat(flatIndex, value);
|
|
}
|
|
fn setOutputI32(${o.map(l=>`${l} : i32`).join(", ")}, value : i32) {
|
|
let flatIndex = getOutputFlatIndex(${i}(${o.join(", ")}));
|
|
setOutputFlatI32(flatIndex, value);
|
|
}
|
|
`}return a}function wie(e,t,n,s){let r=kie(e,n);return e.shape.length<=t.length&&(r+=Sie(e,t,n,s)),r}function kie(e,t){let n=e.name,s=e.shape.length,r=vn(s),a="get"+n.charAt(0).toUpperCase()+n.slice(1),o=["d0","d1","d2","d3"].slice(0,s),i=o.map(u=>`${u} : i32`).join(", ");if(s<1)return t?`
|
|
fn ${a}() -> vec4<f32> {
|
|
return vec4<f32>(${n}.numbers[0]);
|
|
}
|
|
`:`
|
|
fn ${a}() ->f32 {
|
|
return f32(${n}.numbers[0]);
|
|
}
|
|
`;let l=`uniforms.${n.charAt(0).toLowerCase()+n.slice(1)}Shape`,c=`${s}D`;return s===0&&(c="1D"),t?`
|
|
fn ${a}(${i}) -> vec4<f32> {
|
|
return vec4<f32>(${n}.numbers[getFlatIndex${c}(${r}(${o.join(",")}),
|
|
${l}) / 4]);
|
|
}
|
|
`:`
|
|
fn ${a}(${i}) -> f32 {
|
|
return f32(${n}.numbers[getFlatIndex${c}(${r}(${o.join(",")}),
|
|
${l})]);
|
|
}
|
|
`}function Sie(e,t,n,s){let r=e.name,a=r.charAt(0).toUpperCase()+r.slice(1),o="get"+a+"AtOutCoords",i=e.shape.length,l=t.length,c=vn(l);if(v.arraysEqual(e.shape,t)&&s)return n?`
|
|
fn ${o}ByGlobalIndex(globalIndex : i32) -> vec4<f32> {
|
|
return vec4<f32>(${r}.numbers[globalIndex]);
|
|
}
|
|
|
|
fn ${o}ByCoords(coords : ${c}) -> vec4<f32> {
|
|
return vec4<f32>(${r}.numbers[${l>1?"getOutputFlatIndex(coords)":"coords"} / 4]);
|
|
}
|
|
`:`
|
|
fn ${o}ByGlobalIndex(globalIndex : i32) -> f32 {
|
|
return f32(${r}.numbers[globalIndex]);
|
|
}
|
|
|
|
fn ${o}ByCoords(coords : ${c}) -> f32 {
|
|
return f32(${r}.numbers[${l>1?"getOutputFlatIndex(coords)":"coords"}]);
|
|
}
|
|
`;let u=E.getBroadcastDims(e.shape,t),d=l-i,p="";if(i===0)return n?`
|
|
fn ${o}ByGlobalIndex(globalIndex : i32) -> vec4<f32> {
|
|
return get${a}();
|
|
}
|
|
|
|
fn ${o}ByCoords(coords : ${c}) -> vec4<f32> {
|
|
return get${a}();
|
|
}
|
|
`:`
|
|
fn ${o}ByGlobalIndex(globalIndex : i32) -> f32{
|
|
return get${a}();
|
|
}
|
|
|
|
fn ${o}ByCoords(coords : ${c}) -> f32{
|
|
return get${a}();
|
|
}
|
|
`;l<2&&u.length>=1?p="coords = 0;":p=u.map(g=>`coords[${g+d}] = 0;`).join(`
|
|
`);let h="";if(l<2&&i>0)h="coords";else if(l>1){let g=vn(i),A=e.shape.map((x,y)=>`coords[${y+d}]`).join(", ");h=`${g}(${A})`}else h="coords";let f=`uniforms.${r.charAt(0).toLowerCase()+r.slice(1)}Shape`,m=`${i}D`;return n?`
|
|
fn ${o}ByGlobalIndex(globalIndex : i32) -> vec4<f32> {
|
|
var coords = getCoordsFromFlatIndex(globalIndex);
|
|
${p}
|
|
return ${r}.numbers[getFlatIndex${m}(${h}, ${f}) / 4];
|
|
}
|
|
|
|
fn ${o}ByCoords(coordsIn : ${c}) -> vec4<f32> {
|
|
var coords = coordsIn;
|
|
${p}
|
|
return ${r}.numbers[getFlatIndex${m}(${h}, ${f}) / 4];
|
|
}
|
|
`:`
|
|
fn ${o}ByGlobalIndex(globalIndex : i32) -> f32 {
|
|
var coords = getCoordsFromFlatIndex(globalIndex);
|
|
${p}
|
|
return f32(${r}.numbers[getFlatIndex${m}(${h}, ${f})]);
|
|
}
|
|
|
|
fn ${o}ByCoords(coordsIn : ${c}) -> f32 {
|
|
var coords = coordsIn;
|
|
${p}
|
|
return f32(${r}.numbers[getFlatIndex${m}(${h}, ${f})]);
|
|
}
|
|
`}function Iie(e,t){let{x:n,y:s=[],z:r=[]}=t,a=e.length;if(n.length===a)return[`fn getOutputCoordsWithFlatDispatchLayout(globalId : vec3<u32>, localId : vec3<u32>, numWorkgroups: vec3<u32>) -> ${vn(a)}{
|
|
let globalIndex = getGlobalIndex(globalId, localId, numWorkgroups);
|
|
return getCoordsFromFlatIndex(globalIndex);
|
|
}
|
|
`,a];let o="",i=[n,s,r],l=0;for(let p=0;p<i.length;p++){let h=i[p];if(h.length!==0)if(l+=h.length,h.length===1)o+=`let d${h[0]} = i32(globalId[${p}]);`;else{let f=yie(h,"uniforms.outShape");o+=`var index${p} = i32(globalId[${p}]);`;for(let m=0;m<f.length;m++)o+=`let d${h[m]} = index${p} / ${f[m]};`,m===f.length-1?o+=`let d${h[m+1]} = index${p} - d${h[m]} * ${f[m]};`:o+=`index${p} = index${p} - d${h[m]} * ${f[m]};`}}let c=[];for(let p=0;p<l;p++)c.push(`d${p}`);let u=vn(l),d=`fn getOutputCoordsWithNonFlatDispatchLayout(globalId : vec3<u32>) -> ${u} {
|
|
${o}
|
|
`;return c.length===0?d+=`return ${u}(0); }`:d+=`return ${u}(${c.join(",")}); }`,[d,l]}function Q4(e){let t=e.length;if(t<=1)return"fn getCoordsFromFlatIndex(index : i32) -> i32 { return index; }";let n=v.computeStrides(e),s=vn(t),r=[];for(let o=0;o<t;o++)r.push(`d${o}`);if(n.length===1)return` fn getCoordsFromFlatIndex(index : i32) -> vec2<i32> {
|
|
let d0 = index / uniforms.outShapeStrides; let d1 = index - d0 * uniforms.outShapeStrides;
|
|
return vec2<i32>(d0, d1);
|
|
}`;let a="var index2 = index;"+n.map((o,i)=>{let l=`let ${r[i]} = index2 / uniforms.outShapeStrides[${i}]`,c=i===n.length-1?`let ${r[i+1]} = index2 - ${r[i]} * uniforms.outShapeStrides[${i}]`:`index2 = index2 - ${r[i]} * uniforms.outShapeStrides[${i}]`;return`${l}; ${c};`}).join("");return`
|
|
fn getCoordsFromFlatIndex(index : i32) -> ${s} {
|
|
${a}
|
|
return ${s}(${r.join(",")});
|
|
}
|
|
`}var eC={};Oe(eC,{ArrayBufferToTypedArray:()=>tC,GPUBytesPerElement:()=>Ax,computeDispatch:()=>Fe,computeWorkGroupSizeForConv2d:()=>fx,computeWorkGroupSizeForMatMul:()=>mx,computeWorkPerThreadForConv2d:()=>gx,flatDispatchLayout:()=>je,isWebGPUSupported:()=>yx,tilesFitEvenlyIntoShape:()=>ra});var pc=65535,El=e=>{let t=1;for(let n=0;n<e.length;n++)t*=e[n];return t};function ra(e,t){if(e.length!==t.length)throw new Error(`Cannot compute whether rank ${e.length} tiles fit evenly into rank ${t.length} shape - ranks must match.`);return t.every((n,s)=>n%e[s]==0)}function Fe(e,t,n=[1,1,1],s=[1,1,1]){let[r,a,o]=[Math.ceil(El(e.x.map(l=>t[l]))/(n[0]*s[0])),e.y?Math.ceil(El(e.y.map(l=>t[l]))/(n[1]*s[1])):1,e.z?Math.ceil(El(e.z.map(l=>t[l]))/(n[2]*s[2])):1];if(r<=pc&&a<=pc&&o<=pc)return[r,a,o];v.assert(r>pc&&e.y===void 0&&e.z===void 0,()=>"Dispatch size exceeds WebGPU limits in Y or Z dimension.");let i=Math.ceil(Math.sqrt(r));return i>pc?(i=Math.ceil(Math.cbrt(r)),v.assert(i<=pc,()=>"Total dispatch size exceeds WebGPU maximum."),[i,i,i]):[i,i,1]}function fx(e,t){let n=El(e.x.map(r=>t[r])),s=El(e.y.map(r=>t[r]));return n<=4?[4,16,1]:s<=4?[16,4,1]:[16,16,1]}function mx(e,t,n){return e===1?[32,1,1]:n===1?[1,32,1]:[8,8,1]}function gx(e,t){let n=El(e.x.map(r=>t[r])),s=El(e.y.map(r=>t[r]));return n<=4?[1,2,1]:s<=4?[2,1,1]:[2,2,1]}function je(e){return{x:e.map((t,n)=>n)}}function Ax(e){if(e==="float32"||e==="int32"||e==="bool"||e==="string")return 4;if(e==="complex64")return 8;throw new Error(`Unknown dtype ${e}`)}function tC(e,t){if(t==="float32")return new Float32Array(e);if(t==="int32")return new Int32Array(e);if(t==="bool"||t==="string"){let n=new Int32Array(e),s=new ArrayBuffer(n.length),r=new Uint8Array(s);for(let a=0;a<n.length;a++)r[a]=n[a];return r}else throw new Error(`Unknown dtype ${t}`)}function yx(){return!!navigator.gpu}var Wt;(function(e){e[e.MUL=0]="MUL",e[e.ADD=1]="ADD",e[e.SUB=2]="SUB",e[e.DIV=3]="DIV",e[e.EQUAL=4]="EQUAL",e[e.GREATER=5]="GREATER",e[e.GREATER_EQUAL=6]="GREATER_EQUAL",e[e.LESS=7]="LESS",e[e.LESS_EQUAL=8]="LESS_EQUAL",e[e.LOGICAL_AND=9]="LOGICAL_AND",e[e.NOT_EQUAL=10]="NOT_EQUAL",e[e.SQUARED_DIFFERENCE=11]="SQUARED_DIFFERENCE",e[e.INT_DIV=12]="INT_DIV",e[e.POW=13]="POW",e[e.PRELU=14]="PRELU",e[e.MAX=15]="MAX",e[e.MIN=16]="MIN",e[e.COMPLEX_MULTIPLY_REAL=17]="COMPLEX_MULTIPLY_REAL",e[e.COMPLEX_MULTIPLY_IMAG=18]="COMPLEX_MULTIPLY_IMAG"})(Wt||(Wt={}));var Cie="return a + b;",Tie="return areal * breal - aimag * bimag;",Nie="return areal * bimag + aimag * breal;",Eie="return a / b;",Rie="return a * b;",$ie="return (a - b) * (a - b);",_ie="return a - b;",Die="return f32(a == b);",Pie="return vec4<f32>(a == b);",Fie="return f32(a > b);",Oie="return vec4<f32>(a > b);",Mie="return f32(a >= b);",zie="return vec4<f32>(a >= b);",Lie="return f32(a < b);",Bie="return vec4<f32>(a < b);",Wie="return f32(a <= b);",Vie="return vec4<f32>(a <= b);",Uie="return f32(f32(a) >= 1.0 && f32(b) >= 1.0);",Gie=`return (vec4<f32>(a >= vec4<f32>(1.0)) *
|
|
vec4<f32>(b >= vec4<f32>(1.0)));`,Hie=`
|
|
if (isNanCustom(a)) { return a; }
|
|
if (isNanCustom(b)) { return b; }
|
|
`,nC=`
|
|
if (isNaN.r > 0.) {
|
|
resultTemp.r = uniforms.NAN;
|
|
}
|
|
if (isNaN.g > 0.) {
|
|
resultTemp.g = uniforms.NAN;
|
|
}
|
|
if (isNaN.b > 0.) {
|
|
resultTemp.b = uniforms.NAN;
|
|
}
|
|
if (isNaN.a > 0.) {
|
|
resultTemp.a = uniforms.NAN;
|
|
}
|
|
`,jie=`
|
|
let s = sign(a) * sign(b);
|
|
let ia = i32(round(a));
|
|
let ib = i32(round(b));
|
|
return f32(idiv(ia, ib, s));
|
|
`,qie=`
|
|
let ia = vec4<i32>(round(a));
|
|
let ib = vec4<i32>(round(b));
|
|
let cond = ib != vec4<i32>(0);
|
|
var resultTemp = vec4<i32>(0);
|
|
let s = sign(a) * sign(b);
|
|
|
|
// Windows (D3D) wants guaranteed non-zero int division at compile-time.
|
|
if (cond[0]) {
|
|
resultTemp[0] = idiv(ia[0], ib[0], s[0]);
|
|
}
|
|
if (cond[1]) {
|
|
resultTemp[1] = idiv(ia[1], ib[1], s[1]);
|
|
}
|
|
if (cond[2]) {
|
|
resultTemp[2] = idiv(ia[2], ib[2], s[2]);
|
|
}
|
|
if (cond[3]) {
|
|
resultTemp[3] = idiv(ia[3], ib[3], s[3]);
|
|
}
|
|
return vec4<f32>(resultTemp);
|
|
`,Xie="return f32(a != b);",Kie="return vec4<f32>(a != b);",Zie=`
|
|
if(a < 0.0 && floor(b) < b) {
|
|
return uniforms.NAN;
|
|
}
|
|
if (b == 0.0) {
|
|
return 1.0;
|
|
}
|
|
if (round(abs(b) % 2.0) != 1.0) {
|
|
return pow(abs(a), b);
|
|
}
|
|
return sign(a) * pow(abs(a), b);
|
|
`,Yie=`
|
|
let isModRound1Bool = vec4<i32>(round(abs(b) % vec4<f32>(2.0))) == vec4<i32>(1);
|
|
let isModRound1 = vec4<f32>(isModRound1Bool);
|
|
let multiplier = sign(a) * isModRound1 + (vec4<f32>(1.0) - isModRound1);
|
|
var resultTemp = multiplier * pow(abs(a), b);
|
|
|
|
// Ensure that a^0 = 1, including 0^0 = 1 as this correspond to TF and JS
|
|
let isExpZero = b == vec4<f32>(0.0);
|
|
if (isExpZero.r) {
|
|
resultTemp.r = 1.0;
|
|
}
|
|
if (isExpZero.g) {
|
|
resultTemp.g = 1.0;
|
|
}
|
|
if (isExpZero.b) {
|
|
resultTemp.b = 1.0;
|
|
}
|
|
if (isExpZero.a) {
|
|
resultTemp.a = 1.0;
|
|
}
|
|
let isNaN = vec4<f32>(a < vec4<f32>(0.0)) * vec4<f32>(floor(b) < b);
|
|
${nC}
|
|
return resultTemp;
|
|
`,Jie="if (a < 0.0) { return b * a; } return a;",Qie=`
|
|
let aLessThanZero = vec4<f32>(a < vec4<f32>(0.0));
|
|
return (aLessThanZero * (b * a)) + ((vec4<f32>(1.0) - aLessThanZero) * a);
|
|
`;function sC(e,t){let n=t?nC:Hie;return t?`
|
|
var resultTemp = vec4<f32>(${e}(a, b));
|
|
let isNaN = min(vec4<f32>(isNanCustomVec4F32(a)) + vec4<f32>(isNanCustomVec4F32(b)), vec4<f32>(1.0));
|
|
`+n+`
|
|
return resultTemp;
|
|
`:n+`
|
|
return ${e}(a, b);
|
|
`}function Ap(e,t){switch(e){case 0:return Rie;case 1:return Cie;case 2:return _ie;case 3:return Eie;case 4:return t?Pie:Die;case 5:return t?Oie:Fie;case 6:return t?zie:Mie;case 7:return t?Bie:Lie;case 8:return t?Vie:Wie;case 9:return t?Gie:Uie;case 10:return t?Kie:Xie;case 11:return $ie;case 12:return t?qie:jie;case 14:return t?Qie:Jie;case 15:return sC("max",t);case 16:return sC("min",t);case 13:return t?Yie:Zie;case 17:return Tie;case 18:return Nie;default:throw new Error(`BinaryType ${e} is not implemented!`)}}var xt;(function(e){e[e.ABS=0]="ABS",e[e.CEIL=1]="CEIL",e[e.COS=2]="COS",e[e.COSH=3]="COSH",e[e.ELU=4]="ELU",e[e.EXP=5]="EXP",e[e.EXPM1=6]="EXPM1",e[e.FLOOR=7]="FLOOR",e[e.LINEAR=8]="LINEAR",e[e.LOG=9]="LOG",e[e.LOGICAL_NOT=10]="LOGICAL_NOT",e[e.NEG=11]="NEG",e[e.PRELU=12]="PRELU",e[e.RELU=13]="RELU",e[e.RELU6=14]="RELU6",e[e.RSQRT=15]="RSQRT",e[e.SIN=16]="SIN",e[e.SINH=17]="SINH",e[e.SIGMOID=18]="SIGMOID",e[e.SQRT=19]="SQRT",e[e.SQUARE=20]="SQUARE",e[e.TANH=21]="TANH",e[e.TO_INT=22]="TO_INT"})(xt||(xt={}));var ele="return abs(a);",tle="return ceil(a);",nle="return cos(a);",sle=`
|
|
let e2x = exp(-a);
|
|
return (e2x + 1.0 / e2x) / 2.0;
|
|
`,rle="return exp(a) - 1.0;",ale="if (a >= 0.0) { return a; } return (exp(a) - 1.0);",ole=`
|
|
var resFloat = exp(a) - vec4<f32>(1.0);
|
|
if (a.r >= 0.0) {
|
|
resFloat.r = a.r;
|
|
}
|
|
if (a.g >= 0.0) {
|
|
resFloat.g = a.g;
|
|
}
|
|
if (a.b >= 0.0) {
|
|
resFloat.b = a.b;
|
|
}
|
|
if (a.a >= 0.0) {
|
|
resFloat.a = a.a;
|
|
}
|
|
return resFloat;
|
|
`,ile="return exp(a);",lle="return floor(a);",ule="return a;",cle=`if (a < 0.0) { return 1.0/0.0; }
|
|
return log(a);`,dle="return f32(!(a >= 1.0));",ple="return -a;",hle="return (a < 0.0) ? b * a : a;",fle="return max(a, 0.0);",mle="return clamp(a, 0.0, 6.0);",gle="return clamp(a, vec4<f32>(0.0, 0.0, 0.0, 0.0), vec4<f32>(6.0, 6.0, 6.0, 6.0));",Ale=`
|
|
var resFloat = a * vec4<f32>(a >= vec4<f32>(0.0));
|
|
let isNaN = isNan(a);
|
|
|
|
if (isNaN.r) {
|
|
resFloat.r = a.r;
|
|
}
|
|
if (isNaN.g) {
|
|
resFloat.g = a.g;
|
|
}
|
|
if (isNaN.b) {
|
|
resFloat.b = a.b;
|
|
}
|
|
if (isNaN.a) {
|
|
resFloat.a = a.a;
|
|
}
|
|
return resFloat;
|
|
`,yle="return 1.0/sqrt(a);",xle="return 1.0 / (1.0 + exp(-1.0 * a));",ble="return sin(a);",vle=`
|
|
let e2x = exp(a);
|
|
return (e2x - 1.0 / e2x) / 2.0;
|
|
`,wle="return sqrt(a);",kle="return a * a;",Sle=`
|
|
let e2x = exp(-2.0 * abs(a));
|
|
return sign(a) * (1.0 - e2x) / (1.0 + e2x);
|
|
`,Ile="return f32(i32((a)));";function hc(e,t){switch(e){case 0:return ele;case 2:return nle;case 3:return sle;case 1:return tle;case 4:return t?ole:ale;case 5:return ile;case 6:return rle;case 7:return lle;case 8:return ule;case 9:return cle;case 10:return dle;case 11:return ple;case 12:return hle;case 13:return t?Ale:fle;case 14:return t?gle:mle;case 15:return yle;case 18:return xle;case 16:return ble;case 17:return vle;case 19:return wle;case 20:return kle;case 21:return Sle;case 22:return Ile;default:throw new Error(`BinaryType ${e} is not implemented!`)}}function aa(e,t=!1){if(e===null)return null;if(e==="linear")return hc(xt.LINEAR);if(e==="relu")return hc(xt.RELU,t);if(e==="elu")return hc(xt.ELU,t);if(e==="relu6")return hc(xt.RELU6,t);if(e==="prelu")return Ap(Wt.PRELU,t);if(e==="sigmoid")return hc(xt.SIGMOID);throw new Error(`Activation ${e} has not been implemented for the WebGPU backend.`)}function rC(e,t){let n={RowPerThread:e[1],ColPerThread:e[0],TileAOuter:t[1]*e[1],TileBOuter:t[0]*e[0],TileInner:t[0]*e[0]};return`
|
|
var<workgroup> mm_Asub : array<array<vec4<f32>, ${n.TileInner/n.ColPerThread}>, ${n.TileAOuter}>;
|
|
var<workgroup> mm_Bsub : array<array<vec4<f32>, ${n.TileBOuter/n.ColPerThread}>, ${n.TileInner}>;
|
|
|
|
let RowPerThread = ${n.RowPerThread};
|
|
let ColPerThread = ${n.ColPerThread}; // only support ColPerThread = 4
|
|
let TileAOuter = ${n.TileAOuter};
|
|
let TileBOuter = ${n.TileBOuter};
|
|
let TileInner = ${n.TileInner};
|
|
|
|
${Go()} {
|
|
|
|
let tileRow = i32(localId.y) * RowPerThread;
|
|
let tileCol = i32(localId.x);
|
|
|
|
let globalRow = i32(globalId.y) * RowPerThread;
|
|
let globalCol = i32(globalId.x);
|
|
let numTiles = (uniforms.dimInner - 1) / TileInner + 1;
|
|
|
|
var acc: array<vec4<f32>, ${n.RowPerThread}>;
|
|
var ACached : vec4<f32>;
|
|
var BCached : array<vec4<f32>, 4>;
|
|
|
|
// Loop over shared dimension.
|
|
var globalColA = tileCol;
|
|
let RowPerThreadB = TileInner / ${t[1]};
|
|
let tileRowB = i32(localId.y) * RowPerThreadB;
|
|
for (var t = 0; t < numTiles; t = t + 1) {
|
|
// Load one tile of A into local memory.
|
|
for (var innerRow = 0; innerRow < RowPerThread; innerRow = innerRow + 1) {
|
|
let inputRow = tileRow + innerRow;
|
|
let inputCol = tileCol;
|
|
mm_Asub[inputRow][inputCol] = mm_readA(globalRow + innerRow, globalColA, globalId);
|
|
}
|
|
globalColA = globalColA + TileInner / ColPerThread;
|
|
|
|
// Load one tile of B into local memory.
|
|
for (var innerRow = 0; innerRow < RowPerThreadB; innerRow = innerRow + 1) {
|
|
let inputRow = tileRowB + innerRow;
|
|
let inputCol = tileCol;
|
|
mm_Bsub[inputRow][inputCol] = mm_readB(t * TileInner + inputRow, globalCol, globalId);
|
|
}
|
|
|
|
workgroupBarrier();
|
|
|
|
// Compute acc values for a single thread.
|
|
for (var k = 0; k < TileInner / ColPerThread; k = k + 1) {
|
|
BCached[0] = mm_Bsub[k * ColPerThread][tileCol];
|
|
BCached[1] = mm_Bsub[k * ColPerThread + 1][tileCol];
|
|
BCached[2] = mm_Bsub[k * ColPerThread + 2][tileCol];
|
|
BCached[3] = mm_Bsub[k * ColPerThread + 3][tileCol];
|
|
|
|
for (var i = 0; i < RowPerThread; i = i + 1) {
|
|
ACached = mm_Asub[tileRow + i][k];
|
|
acc[i] = BCached[0] * ACached.x + acc[i];
|
|
acc[i] = BCached[1] * ACached.y + acc[i];
|
|
acc[i] = BCached[2] * ACached.z + acc[i];
|
|
acc[i] = BCached[3] * ACached.w + acc[i];
|
|
}
|
|
}
|
|
|
|
workgroupBarrier();
|
|
}
|
|
|
|
for (var innerRow = 0; innerRow < RowPerThread; innerRow = innerRow + 1) {
|
|
mm_write(globalRow + innerRow,
|
|
globalCol,
|
|
acc[innerRow], globalId);
|
|
}
|
|
}`}function Cle(e){return`
|
|
var<workgroup> mm_Asub : array<vec4<f32>, ${e[0]}>;
|
|
let tileSize = ${e[0]*4};
|
|
${Go()} {
|
|
let tileCol = i32(localId.x);
|
|
let globalCol = i32(globalId.x);
|
|
let globalRow = i32(globalId.y);
|
|
|
|
let numTiles = (uniforms.dimInner - 1) / tileSize + 1;
|
|
|
|
// Without this initialization strange values show up in acc.
|
|
var acc = vec4<f32>(0.0);
|
|
|
|
// Loop over shared dimension.
|
|
for (var t = 0; t < numTiles; t = t + 1) {
|
|
// Load one tile of A into local memory.
|
|
let colA = t * tileSize / 4 + tileCol;
|
|
mm_Asub[tileCol] = mm_readA(globalRow, colA, globalId);
|
|
|
|
workgroupBarrier();
|
|
|
|
// Compute acc values for a single thread.
|
|
for (var k = 0; k < tileSize / 4; k = k + 1) {
|
|
let rowB = t * tileSize + k * 4;
|
|
let BCached0 = mm_readB(rowB, globalCol, globalId);
|
|
let BCached1 = mm_readB(rowB + 1, globalCol, globalId);
|
|
let BCached2 = mm_readB(rowB + 2, globalCol, globalId);
|
|
let BCached3 = mm_readB(rowB + 3, globalCol, globalId);
|
|
|
|
let ACached = mm_Asub[k];
|
|
acc = acc + BCached0 * ACached.x;
|
|
acc = acc + BCached1 * ACached.y;
|
|
acc = acc + BCached2 * ACached.z;
|
|
acc = acc + BCached3 * ACached.w;
|
|
}
|
|
|
|
workgroupBarrier();
|
|
}
|
|
|
|
if (globalRow < uniforms.dimAOuter && globalCol < uniforms.dimBOuter) {
|
|
mm_write(globalRow, globalCol, acc, globalId);
|
|
}
|
|
}
|
|
`}var Tle=class{constructor(e,t,n,s=null,r=null,a=null){this.variableNames=["A","B"],this.uniforms="dimAOuter : i32; dimBOuter : i32; dimInner : i32;",this.workGroupSize=[16,16,1],this.isVec4=!0,this.vecSize=4,this.outputShape=t,this.workGroupSize=mx(t[1],e[2],t[2]),this.dispatchLayout={x:[2],y:[1],z:[0]},t[1]===1&&(n=1),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,[this.vecSize,n,1]);let o=s!=null,i=a!=null;o&&this.variableNames.push("bias"),i&&this.variableNames.push("preluActivationWeights"),this.workPerThread=n,this.aShape=e,this.addBias=o,this.activation=r,this.hasPreluActivationWeights=i,[this.fitA,this.fitB]=this.getShapeFit(),this.shaderKey=`matMulPackedVec4_${n}_${this.activation}_${this.fitA}_${this.fitB}_${this.outputShape[1]>1}`}getShapeFit(){let e=this.aShape[2],t=this.outputShape[2],n=[this.outputShape[0],e,t],s=this.workGroupSize[1]*this.workPerThread,r=this.workGroupSize[0]*this.vecSize,a=r,o=[s,a],i=[a,r];return[ra(o,this.aShape.slice(1)),ra(i,n.slice(1))]}getUserCode(){let e=this.fitA?"return A.numbers[batch * batchASize + row * uniforms.dimInner / 4 + col]":`if (coordsInBounds2D(vec2<i32>(row, col * 4), vec2<i32>(uniforms.dimAOuter, uniforms.dimInner))) {
|
|
return A.numbers[batch * batchASize + row * uniforms.dimInner / 4 + col];
|
|
}
|
|
return vec4<f32>(0.0)`,t=this.fitB?"return B.numbers[batch * batchBSize + row * uniforms.dimBOuter / 4 + col]":`if(coordsInBounds2D(vec2<i32>(row, col * 4), vec2<i32>(uniforms.dimInner, uniforms.dimBOuter))) {
|
|
return B.numbers[batch * batchBSize + row * uniforms.dimBOuter / 4 + col];
|
|
}
|
|
return vec4<f32>(0.0)`,n="",s="";if(this.activation){let o=aa(this.activation,this.isVec4);this.hasPreluActivationWeights?n=`fn activation(a : vec4<f32>, outCoord : vec3<i32>) -> vec4<f32> {
|
|
let b = getPreluActivationWeightsAtOutCoordsByCoords(outCoord);
|
|
${o}
|
|
}`:n=`
|
|
fn activation(a : vec4<f32>, outCoord : vec3<i32>) -> vec4<f32> {
|
|
${o}
|
|
}`,s="value = activation(value, outCoord);"}let r=this.addBias?"value = value + getBiasAtOutCoordsByCoords(outCoord);":"";return`
|
|
${n}
|
|
fn mm_readA(row : i32, col : i32, globalId : vec3<u32>) -> vec4<f32> {
|
|
let batchASize = uniforms.aShape[1] * uniforms.aShape[2] / ${this.vecSize};
|
|
let batch = i32(globalId.z);
|
|
${e};
|
|
}
|
|
|
|
fn mm_readB(row : i32, col : i32, globalId : vec3<u32>) -> vec4<f32> {
|
|
let batchBSize = uniforms.bShape[1] * uniforms.bShape[2] / ${this.vecSize};
|
|
let batch = i32(globalId.z);
|
|
${t};
|
|
}
|
|
|
|
fn mm_write(row : i32, col : i32, valueIn : vec4<f32>, globalId : vec3<u32>) {
|
|
if (row < uniforms.aShape[1] && col * 4 < uniforms.bShape[2])
|
|
{
|
|
var value = valueIn;
|
|
let batch = i32(globalId.z);
|
|
let outCoord = vec3<i32>(batch, row, col * 4);
|
|
${r}
|
|
${s}
|
|
setOutput(outCoord[0], outCoord[1], outCoord[2], value);
|
|
}
|
|
}
|
|
${this.outputShape[1]>1?rC([this.vecSize,this.workPerThread,1],this.workGroupSize):Cle(this.workGroupSize)}
|
|
|
|
`}};function xx(e,t){let n=t[1]*e[1],s=t[0]*e[0],r=n>s?n:s;return`
|
|
var<workgroup> mm_Asub : array<array<f32, ${r}>, ${n}>;
|
|
var<workgroup> mm_Bsub : array<array<f32, ${s}>, ${r}>;
|
|
${Go()} {
|
|
let tileRow = i32(localId.y) * ${e[1]};
|
|
let tileCol = i32(localId.x) * ${e[0]};
|
|
|
|
let globalRow = i32(globalId.y) * ${e[1]};
|
|
let globalCol = i32(globalId.x) * ${e[0]};
|
|
|
|
let numTiles = (uniforms.dimInner - 1) / ${r} + 1;
|
|
|
|
var acc : array<array<f32, ${e[0]}>, ${e[1]}>;
|
|
var ACached : f32;
|
|
var BCached : array<f32, ${e[0]}>;
|
|
|
|
// Without this initialization strange values show up in acc.
|
|
for (var innerRow = 0; innerRow < ${e[1]}; innerRow = innerRow + 1) {
|
|
for (var innerCol = 0; innerCol < ${e[0]}; innerCol = innerCol + 1) {
|
|
acc[innerRow][innerCol] = 0.0;
|
|
}
|
|
}
|
|
|
|
let ColPerThreadA = ${r} / ${t[0]};
|
|
let tileColA = i32(localId.x) * ColPerThreadA;
|
|
let RowPerThreadB = ${r} / ${t[1]};
|
|
let tileRowB = i32(localId.y) * RowPerThreadB;
|
|
|
|
// Loop over shared dimension.
|
|
for (var t = 0; t < numTiles; t = t + 1) {
|
|
// Load one tile of A into local memory.
|
|
for (var innerRow = 0; innerRow < ${e[1]}; innerRow = innerRow + 1) {
|
|
for (var innerCol = 0; innerCol < ColPerThreadA; innerCol = innerCol + 1) {
|
|
let inputRow = tileRow + innerRow;
|
|
let inputCol = tileColA + innerCol;
|
|
|
|
mm_Asub[inputRow][inputCol] = mm_readA(
|
|
globalRow + innerRow,
|
|
t * ${r} + inputCol, globalId);
|
|
}
|
|
}
|
|
// Load one tile of B into local memory.
|
|
for (var innerRow = 0; innerRow < RowPerThreadB; innerRow = innerRow + 1) {
|
|
for (var innerCol = 0; innerCol < ${e[0]}; innerCol = innerCol + 1) {
|
|
let inputRow = tileRowB + innerRow;
|
|
let inputCol = tileCol + innerCol;
|
|
|
|
mm_Bsub[inputRow][inputCol] = mm_readB(
|
|
t * ${r} + inputRow,
|
|
globalCol + innerCol, globalId);
|
|
}
|
|
}
|
|
|
|
workgroupBarrier();
|
|
|
|
// Compute acc values for a single thread.
|
|
for (var k = 0; k < ${r}; k = k + 1) {
|
|
for (var inner = 0; inner < ${e[0]}; inner = inner + 1) {
|
|
BCached[inner] = mm_Bsub[k][tileCol + inner];
|
|
}
|
|
|
|
for (var innerRow = 0; innerRow < ${e[1]}; innerRow = innerRow + 1) {
|
|
ACached = mm_Asub[tileRow + innerRow][k];
|
|
for (var innerCol = 0; innerCol < ${e[0]}; innerCol = innerCol + 1) {
|
|
acc[innerRow][innerCol] = acc[innerRow][innerCol] + ACached * BCached[innerCol];
|
|
}
|
|
}
|
|
}
|
|
|
|
workgroupBarrier();
|
|
}
|
|
|
|
for (var innerRow = 0; innerRow < ${e[1]}; innerRow = innerRow + 1) {
|
|
for (var innerCol = 0; innerCol < ${e[0]}; innerCol = innerCol + 1) {
|
|
|
|
if ((globalCol + innerCol) < uniforms.dimBOuter &&
|
|
(globalRow + innerRow) < uniforms.dimAOuter) {
|
|
mm_write(globalRow + innerRow,
|
|
globalCol + innerCol,
|
|
acc[innerRow][innerCol], globalId);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
`}function Nle(e){return`
|
|
let TileSize = ${e[0]*4};
|
|
var<workgroup> mm_Asub : array<vec4<f32>, ${e[0]}>;
|
|
|
|
${Go()} {
|
|
let tileCol = i32(localId.x);
|
|
let globalCol = i32(globalId.x);
|
|
let globalRow = i32(globalId.y);
|
|
|
|
let numTiles = (uniforms.dimInner - 1) / TileSize + 1;
|
|
|
|
// Without this initialization strange values show up in acc.
|
|
var acc = 0.0;
|
|
|
|
// Loop over shared dimension.
|
|
for (var t = 0; t < numTiles; t = t + 1) {
|
|
// Load one tile of A into local memory.
|
|
let colA = t * TileSize + tileCol * 4;
|
|
mm_Asub[tileCol] = vec4<f32>(mm_readA(globalRow, colA, globalId),
|
|
mm_readA(globalRow, colA + 1, globalId),
|
|
mm_readA(globalRow, colA + 2, globalId),
|
|
mm_readA(globalRow, colA + 3, globalId));
|
|
workgroupBarrier();
|
|
|
|
// Compute acc values for a single thread.
|
|
for (var k = 0; k < TileSize / 4; k = k + 1) {
|
|
let rowB = t * TileSize + k * 4;
|
|
let BCached = vec4<f32>(mm_readB(rowB, globalCol, globalId),
|
|
mm_readB(rowB + 1, globalCol, globalId),
|
|
mm_readB(rowB + 2, globalCol, globalId),
|
|
mm_readB(rowB + 3, globalCol, globalId));
|
|
|
|
let ACached = mm_Asub[k];
|
|
acc = acc + dot(ACached, BCached);
|
|
}
|
|
|
|
workgroupBarrier();
|
|
}
|
|
|
|
if (globalRow < uniforms.dimAOuter && globalCol < uniforms.dimBOuter) {
|
|
mm_write(globalRow, globalCol, acc, globalId);
|
|
}
|
|
}
|
|
`}var aC=class{constructor(e,t,n,s=!1,r=!1,a=null,o=null,i=null){this.variableNames=["A","B"],this.uniforms="dimAOuter : i32; dimBOuter : i32; dimInner : i32;",this.workGroupSize=[16,16,1],this.outputShape=t,this.dispatchLayout={x:[2],y:[1],z:[0]};let l=s?e[1]:e[2];this.workGroupSize=mx(t[1],l,t[2]),(t[1]===1||t[2]===1)&&(n=1),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,[n,n,1]),v.arraysEqual(this.dispatch,[1,1,1])&&(n=1,this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,[n,n,1]));let c=a!=null,u=i!=null;c&&this.variableNames.push("bias"),u&&this.variableNames.push("preluActivationWeights"),this.workPerThread=n,this.aShape=e,this.transposeA=s,this.transposeB=r,this.addBias=c,this.activation=o,this.hasPreluActivationWeights=u;let d=this.outputShape[2],p=this.transposeB?[this.outputShape[0],d,l]:[this.outputShape[0],l,d];[this.fitA,this.fitB]=this.getShapeFit(p),this.shaderKey=`matMulPacked_${this.workPerThread}_${s}_${r}_${this.activation}_${this.fitA}_${this.fitB}_${this.outputShape[1]>1}`}getShapeFit(e){let t=this.workGroupSize[1]*this.workPerThread,n=this.workGroupSize[0]*this.workPerThread,s=t>n?t:n;this.outputShape[1]===1&&(s*=4),v.assert(s%this.workGroupSize[0]==0&&s%this.workGroupSize[1]==0,()=>"tileInner must be multiple of workgroupsize.x and workgroupsize.y");let r=[t,s],a=[s,n];return[ra(r,this.aShape.slice(1)),ra(a,e.slice(1))]}getUserCode(){let e;this.transposeA===!1?e=this.fitA?"return A.numbers[batch * batchASize + row * uniforms.dimInner + col];":`if(coordsInBounds2D(vec2<i32>(row, col), vec2<i32>(uniforms.dimAOuter, uniforms.dimInner))) {
|
|
return A.numbers[batch * batchASize + row * uniforms.dimInner + col];
|
|
}
|
|
return 0.0;`:e=this.fitA?"return A.numbers[batch * batchASize + col * uniforms.dimAOuter + row];":`if(coordsInBounds2D(vec2<i32>(row, col), vec2<i32>(uniforms.dimAOuter, uniforms.dimInner))) {
|
|
return A.numbers[batch* batchASize + col * uniforms.dimAOuter + row];
|
|
}
|
|
return 0.0;`;let t;this.transposeB===!1?t=this.fitB?"return B.numbers[batch * batchBSize + row * uniforms.dimBOuter + col];":`if(coordsInBounds2D(vec2<i32>(row, col), vec2<i32>(uniforms.dimInner, uniforms.dimBOuter))) {
|
|
return B.numbers[batch * batchBSize + row * uniforms.dimBOuter + col];
|
|
}
|
|
return 0.0;`:t=this.fitB?"return B.numbers[batch * batchBSize + col * uniforms.dimInner + row];":`if(coordsInBounds2D(vec2<i32>(row, col), vec2<i32>(uniforms.dimInner, uniforms.dimBOuter))) {
|
|
return B.numbers[batch * batchBSize + col * uniforms.dimInner + row];
|
|
}
|
|
return 0.0;`;let n="",s="";if(this.activation){let o=aa(this.activation,!1);this.hasPreluActivationWeights?n=`fn activation(a : f32, outCoord : vec3<i32>) -> f32 {
|
|
let b = getPreluActivationWeightsAtOutCoordsByCoords(outCoord);
|
|
${o}
|
|
}`:n=`
|
|
fn activation(a : f32, outCoord : vec3<i32>) -> f32 {
|
|
${o}
|
|
}
|
|
`,s="value = activation(value, outCoord);"}let r=this.addBias?"value = value + getBiasAtOutCoordsByCoords(outCoord);":"";return`
|
|
${n}
|
|
|
|
fn mm_readA(row : i32, col : i32, globalId : vec3<u32>) -> f32 {
|
|
let batchASize = uniforms.aShape[1] * uniforms.aShape[2];
|
|
let batch = i32(globalId.z);
|
|
${e}
|
|
}
|
|
|
|
fn mm_readB(row : i32, col : i32, globalId : vec3<u32>) -> f32 {
|
|
let batch = i32(globalId.z);
|
|
let batchBSize = uniforms.bShape[1] * uniforms.bShape[2];
|
|
${t}
|
|
}
|
|
|
|
fn mm_write(row : i32, col : i32, valueIn : f32, globalId : vec3<u32>) {
|
|
var value = valueIn;
|
|
let batch = i32(globalId.z);
|
|
let outCoord = vec3<i32>(batch, row, col);
|
|
${r}
|
|
${s}
|
|
setOutput(batch, row, col, value);
|
|
}
|
|
${this.outputShape[1]>1?xx([this.workPerThread,this.workPerThread,1],this.workGroupSize):Nle(this.workGroupSize)}
|
|
`}};function Ele(){return`
|
|
var<workgroup> sumValues : array<f32, workGroupSizeX>;
|
|
${Go()} {
|
|
let coords = getOutputCoordsWithNonFlatDispatchLayout(globalId);
|
|
let batch = coords[0];
|
|
let row = coords[1];
|
|
let col = coords[2];
|
|
var sum = 0.0;
|
|
let Length = uniforms.dimInner;
|
|
for (var k = i32(localId.x); k < Length; k = k + i32(workGroupSizeX)) {
|
|
let dataA = mm_readA(batch, row, k);
|
|
let dataB = mm_readB(batch, k, col);
|
|
sum = sum + dataA * dataB;
|
|
}
|
|
sumValues[localId.x] = sum;
|
|
workgroupBarrier();
|
|
|
|
for(var currentSize = workGroupSizeX / 2u; currentSize > 1u;
|
|
currentSize = currentSize / 2u) {
|
|
if (localId.x < currentSize)
|
|
{
|
|
sumValues[localId.x] = sumValues[localId.x] + sumValues[localId.x + currentSize];
|
|
}
|
|
workgroupBarrier();
|
|
}
|
|
|
|
if (localId.x == 0u) {
|
|
sum = sumValues[0] + sumValues[1];
|
|
mm_write(batch, row, col, sum);
|
|
}
|
|
}
|
|
`}var Rle=class{constructor(e,t=!1,n=!1,s=null,r=null,a=null){this.variableNames=["A","B"],this.uniforms="dimAOuter : i32; dimBOuter : i32; dimInner : i32;",this.workGroupSize=[256,1,1],this.outputShape=e,this.dispatchLayout={x:[],y:[1,2],z:[0]},this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize);let o=s!=null,i=a!=null;o&&this.variableNames.push("bias"),i&&this.variableNames.push("preluActivationWeights"),this.transposeA=t,this.transposeB=n,this.addBias=o,this.activation=r,this.hasPreluActivationWeights=i,this.shaderKey=`matMulReduce_${this.activation}_${t}_${n}`}getUserCode(){let e;this.transposeA===!1?e="return A.numbers[batch * batchASize + row * uniforms.dimInner + col];":e="return A.numbers[batch * batchASize + col * uniforms.dimAOuter + row];";let t;this.transposeB===!1?t="return B.numbers[batch * batchBSize + row * uniforms.dimBOuter + col];":t="return B.numbers[batch * batchBSize + col * uniforms.dimInner + row];";let n="",s="";if(this.activation){let o=aa(this.activation,!1);this.hasPreluActivationWeights?n=`fn activation(a : f32, outCoord : vec3<i32>) -> f32 {
|
|
let b = getPreluActivationWeightsAtOutCoordsByCoords(outCoord);
|
|
${o}
|
|
}`:n=`
|
|
fn activation(a : f32, outCoord : vec3<i32>) -> f32 {
|
|
${o}
|
|
}
|
|
`,s="value = activation(value, outCoord);"}let r=this.addBias?"value = value + getBiasAtOutCoordsByCoords(outCoord);":"";return`
|
|
${n}
|
|
|
|
fn mm_readA(batch: i32, row : i32, col : i32) -> f32 {
|
|
let batchASize = uniforms.aShape[1] * uniforms.aShape[2];
|
|
${e}
|
|
}
|
|
|
|
fn mm_readB(batch: i32, row : i32, col : i32) -> f32 {
|
|
let batchBSize = uniforms.bShape[1] * uniforms.bShape[2];
|
|
${t}
|
|
}
|
|
|
|
fn mm_write(batch: i32, row : i32, col : i32, valueIn : f32) {
|
|
var value = valueIn;
|
|
let outCoord = vec3<i32>(batch, row, col);
|
|
${r}
|
|
${s}
|
|
setOutput(batch, row, col, value);
|
|
}
|
|
${Ele()}
|
|
`}};function $le(e){let t=e[1]/2,n=e[0],s=t>n?t:n;return`
|
|
var<workgroup> mm_Asub1 : array<array<f32, ${s}>, ${t}>;
|
|
var<workgroup> mm_Bsub1 : array<array<f32, ${n}>, ${s}>;
|
|
var<workgroup> mm_Asub2 : array<array<f32, ${s}>, ${t}>;
|
|
var<workgroup> mm_Bsub2 : array<array<f32, ${n}>, ${s}>;
|
|
|
|
// If the output size is small for matrix multiplication, avoid to use vec4
|
|
// and handle some elements per thread to optimally utilize the ALU.
|
|
// Introduces two shared memory buffers, some logical threads could handle
|
|
// arithmetic operations and others handle IO operations between barrier api,
|
|
// makes ALUs and load/store units work simultaneously, could improves
|
|
// the performance.
|
|
${Go()} {
|
|
let tileRow = i32(localId.y);
|
|
let tileCol = i32(localId.x);
|
|
let globalRow = i32(globalId.y);
|
|
let globalCol = i32(globalId.x);
|
|
|
|
// uniforms.dimInner should be greater than 0.
|
|
let numTiles = (uniforms.dimInner - 1) / ${s} + 1;
|
|
var acc = 0.0;
|
|
|
|
var globalColA = tileCol;
|
|
var globalRowB = tileRow;
|
|
for (var t = 0; t < numTiles; t = t + 1) {
|
|
if (t == 0) {
|
|
if (tileRow < ${t}) {
|
|
// Load one tile of A and B into local memory.
|
|
// globalRow is always greater than or equal tileRow.
|
|
mm_Asub1[tileRow][tileCol] =
|
|
mm_readA((globalRow - tileRow) / 2 + tileRow, globalColA, globalId);
|
|
globalColA = globalColA + ${s};
|
|
mm_Bsub1[tileRow][tileCol] = mm_readB(globalRowB, globalCol, globalId);
|
|
globalRowB = globalRowB + ${s};
|
|
}
|
|
} else {
|
|
if (tileRow < ${t}) {
|
|
// Load one tile of A and B into local memory.
|
|
// globalRow is always greater than or equal tileRow.
|
|
mm_Asub1[tileRow][tileCol] =
|
|
mm_readA((globalRow - tileRow) / 2 + tileRow, globalColA, globalId);
|
|
globalColA = globalColA + ${s};
|
|
mm_Bsub1[tileRow][tileCol] = mm_readB(globalRowB, globalCol, globalId);
|
|
globalRowB = globalRowB + ${s};
|
|
} else {
|
|
// Compute acc values for a single thread.
|
|
for (var k = 0; k < ${s}; k = k + 1) {
|
|
let subRow = tileRow - ${t};
|
|
if (subRow < 0) {
|
|
continue;
|
|
}
|
|
acc = acc + mm_Asub2[subRow][k] * mm_Bsub2[k][tileCol];
|
|
}
|
|
}
|
|
}
|
|
workgroupBarrier();
|
|
if (t != 0) {
|
|
t = t + 1;
|
|
}
|
|
|
|
if (t < numTiles) {
|
|
if (tileRow < ${t}) {
|
|
// Load one tile of A and B into local memory.
|
|
// globalRow is always greater than or equal tileRow.
|
|
mm_Asub2[tileRow][tileCol] =
|
|
mm_readA((globalRow - tileRow) / 2 + tileRow, globalColA, globalId);
|
|
globalColA = globalColA + ${s};
|
|
mm_Bsub2[tileRow][tileCol] = mm_readB(globalRowB, globalCol, globalId);
|
|
globalRowB = globalRowB + ${s};
|
|
} else {
|
|
// Compute acc values for a single thread.
|
|
for (var k = 0; k < ${s}; k = k + 1) {
|
|
let subRow = tileRow - ${t};
|
|
if (subRow < 0) {
|
|
continue;
|
|
}
|
|
acc = acc + mm_Asub1[subRow][k] * mm_Bsub1[k][tileCol];
|
|
}
|
|
}
|
|
}
|
|
workgroupBarrier();
|
|
}
|
|
let writeCol = (globalRow - tileRow) / 2 + tileRow - ${t};
|
|
if (tileRow >= ${t} && writeCol >= 0) {
|
|
mm_write(writeCol, globalCol, acc, globalId);
|
|
}
|
|
}
|
|
`}var _le=class{constructor(e,t,n,s=null,r=null,a=null){this.variableNames=["A","B"],this.uniforms="dimAOuter : i32; dimBOuter : i32; dimInner : i32;",this.workGroupSize=[8,16,1],v.assert(e[1]<=16||t[2]<=16,()=>"This program can be only used when A width or B Height are small"),this.outputShape=n,this.dispatchLayout={x:[2],y:[1],z:[0]},this.dispatch=[Math.ceil(n[2]/this.workGroupSize[0]),Math.ceil(n[1]*2/this.workGroupSize[1]),n[0]];let o=s!=null;o&&this.variableNames.push("bias");let i=a!=null;i&&this.variableNames.push("preluActivationWeights"),this.addBias=o,this.activation=r,this.hasPreluActivationWeights=i,this.shaderKey=`matMulSmallOutputSize_${this.activation}`}getUserCode(){let e=`if (coordsInBounds2D(vec2<i32>(row, col), vec2<i32>(uniforms.dimAOuter, uniforms.dimInner))) {
|
|
return A.numbers[batch * batchASize + row * uniforms.dimInner + col];
|
|
}
|
|
return 0.0;`,t=`if (coordsInBounds2D(vec2<i32>(row, col), vec2<i32>(uniforms.dimInner, uniforms.dimBOuter))) {
|
|
return B.numbers[batch * batchBSize + row * uniforms.dimBOuter + col];
|
|
}
|
|
return 0.0;`,n="",s="";if(this.activation){let o=aa(this.activation,!1);this.hasPreluActivationWeights?n=`fn activation(a : f32, outCoord : vec3<i32>) -> f32 {
|
|
let b = getPreluActivationWeightsAtOutCoordsByCoords(outCoord);
|
|
${o}
|
|
}`:n=`fn activation(a : f32, outCoord : vec3<i32>) -> f32 {
|
|
${o}
|
|
}`,s="value = activation(value, outCoord);"}let r=this.addBias?"value = value + getBiasAtOutCoordsByCoords(outCoord);":"";return`
|
|
${n}
|
|
|
|
fn mm_readA(row : i32, col : i32, globalId : vec3<u32>) -> f32 {
|
|
let batchASize = uniforms.aShape[1] * uniforms.aShape[2];
|
|
let batch = i32(globalId.z);
|
|
${e}
|
|
}
|
|
fn mm_readB(row : i32, col : i32, globalId : vec3<u32>) -> f32 {
|
|
let batch = i32(globalId.z);
|
|
let batchBSize = uniforms.bShape[1] * uniforms.bShape[2];
|
|
${t}
|
|
}
|
|
fn mm_write(row : i32, col : i32, valueIn : f32, globalId : vec3<u32>) {
|
|
if (coordsInBounds2D(vec2<i32>(row, col), vec2<i32>(uniforms.dimAOuter, uniforms.dimBOuter))) {
|
|
let batch = i32(globalId.z);
|
|
let outCoord = vec3<i32>(batch, row, col);
|
|
var value = valueIn;
|
|
${r}
|
|
${s}
|
|
setOutput(batch, row, col, value);
|
|
}
|
|
}
|
|
${$le(this.workGroupSize)}
|
|
`}};function Ge(e){let{inputs:t,attrs:n}=e,{x:s}=t,{shape:r}=n,a=v.sizeFromShape(s.shape),o=v.inferFromImplicitShape(r,a),i=v.sizeFromShape(o);return v.assert(a===i,()=>`The new shape (${o}) has ${i} elements and the old shape (${s.shape}) has ${a} elements. The new shape and old shape must have the same number of elements.`),e.backend.incRef(s.dataId),{dataId:s.dataId,shape:o,dtype:s.dtype}}var Dle={kernelName:Fi,backendName:"webgpu",kernelFunc:Ge};function bx({a:e,b:t,transposeA:n,transposeB:s,backend:r,bias:a=null,preluActivationWeights:o=null,leakyreluAlpha:i=0,activation:l=null}){let c=e.shape.length,u=t.shape.length,d=n?e.shape[c-2]:e.shape[c-1],p=s?t.shape[u-1]:t.shape[u-2],h=n?e.shape[c-1]:e.shape[c-2],f=s?t.shape[u-2]:t.shape[u-1],m=e.shape.slice(0,-2),g=t.shape.slice(0,-2),A=v.sizeFromShape(m),x=v.sizeFromShape(g),b=nl.assertAndGetBroadcastShape(e.shape.slice(0,-2),t.shape.slice(0,-2)).concat([h,f]);v.assert(d===p,()=>`Error in matMul: inner shapes (${d}) and (${p}) of Tensors with shapes ${e.shape} and ${t.shape} and transposeA=${n} and transposeB=${s} must match.`);let w=n?[A,d,h]:[A,h,d],k=s?[x,f,p]:[x,p,f],C=Ge({inputs:{x:e},backend:r,attrs:{shape:w}}),N=Ge({inputs:{x:t},backend:r,attrs:{shape:k}}),R=[C,N],F=Math.max(A,x),_=d%4==0&&f%4==0&&!n&&!s&&f>=32,P;h*f<=32?P=new Rle([F,h,f],n,s,a,l,o):!n&&!s&&(h<=16&&(f<=512||p>=2*f)||f<=16&&(h<=512||d>=2*h))?P=new _le(w,k,[F,h,f],a,l,o):_?P=new Tle(w,[F,h,f],K().get("WEBGPU_MATMUL_WORK_PER_THREAD"),a,l,o):P=new aC(w,[F,h,f],K().get("WEBGPU_MATMUL_WORK_PER_THREAD"),n,s,a,l,o);let T=[C,N];a&&T.push(a),o&&T.push(o);let M=[{type:"int32",data:[h]},{type:"int32",data:[f]},{type:"int32",data:[d]}],U=r.runWebGPUProgram(P,T,e.dtype,M),j=Ge({inputs:{x:U},backend:r,attrs:{shape:b}});R.push(U);for(let z of R)r.disposeData(z.dataId);return j}function Ple(e){let{inputs:t,backend:n,attrs:s}=e,{a:r,b:a,bias:o,preluActivationWeights:i}=t,{transposeA:l,transposeB:c,activation:u,leakyreluAlpha:d}=s;return bx({a:r,b:a,transposeA:l,transposeB:c,backend:n,bias:o,preluActivationWeights:i,leakyreluAlpha:d,activation:u})}var Fle={kernelName:yo,backendName:"webgpu",kernelFunc:Ple},oC=class{constructor(e,t,n){this.variableNames=["AReal","AImag","BReal","BImag"],this.workGroupSize=[128,1,1],this.size=!0,this.outputShape=E.assertAndGetBroadcastShape(t,n),this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.shaderKey=`binaryOpComplex_${e}`,this.op=e}getUserCode(){return`
|
|
fn binaryOpComplex(
|
|
areal : f32, aimag : f32, breal : f32, bimag : f32) -> f32 {
|
|
${Ap(this.op,!1)}
|
|
}
|
|
|
|
${tt()}
|
|
if(index < uniforms.size) {
|
|
let areal = getARealAtOutCoordsByGlobalIndex(index);
|
|
let aimag = getAImagAtOutCoordsByGlobalIndex(index);
|
|
let breal = getBRealAtOutCoordsByGlobalIndex(index);
|
|
let bimag = getBImagAtOutCoordsByGlobalIndex(index);
|
|
setOutputFlat(index, binaryOpComplex(areal, aimag, breal, bimag));
|
|
}
|
|
}
|
|
`}},Ole=class{constructor(e,t,n,s){this.variableNames=["A","B"],this.size=!0;let r=256;this.workGroupSize=[r,1,1],this.outputShape=E.assertAndGetBroadcastShape(t,n),this.dispatchLayout=je(this.outputShape),this.lastDimensionSize=s?n[0]:t[0],this.lastDimensionSize<256?this.workPerThread=1:this.lastDimensionSize<512?this.workPerThread=2:this.workPerThread=4,this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,[this.workPerThread,1,1]),this.useSharedMemoryWithB=s,this.op=e,this.shaderKey=`binaryShared_${e}_${this.lastDimensionSize}_${this.useSharedMemoryWithB}`}getUserCode(){let e=this.lastDimensionSize>1?`coords[${this.outputShape.length-1}]`:"0",t=this.useSharedMemoryWithB?`let a = getAAtOutCoordsByCoords(coords);
|
|
let b = sharedBuf[${e}];`:`let a = sharedBuf[${e}];
|
|
let b = getBAtOutCoordsByCoords(coords);`;return`
|
|
fn binaryOperation(a : f32, b : f32) -> f32 {
|
|
${Ap(this.op,!1)}
|
|
}
|
|
var<workgroup> sharedBuf : array<f32, ${this.lastDimensionSize}>;
|
|
${tt()}
|
|
|
|
// Fill in the shared memory buffer. Here we need a loop to make sure
|
|
// that all data in A|B are uploaded when |sharedMemorySize| is larger
|
|
// than work group size.
|
|
for(var localIndex = i32(localId.x); localIndex < ${this.lastDimensionSize}; localIndex = localIndex + ${this.workGroupSize[0]}) {
|
|
sharedBuf[localIndex] = f32(${this.useSharedMemoryWithB?"B":"A"}.numbers[localIndex]);
|
|
}
|
|
workgroupBarrier();
|
|
|
|
for(var i = 0; i < ${this.workPerThread}; i = i + 1) {
|
|
let flatIndex = index * ${this.workPerThread} + i;
|
|
if(flatIndex < uniforms.size) {
|
|
let coords = getCoordsFromFlatIndex(flatIndex);
|
|
|
|
${t}
|
|
setOutputFlat(flatIndex, binaryOperation(a, b));
|
|
}
|
|
}
|
|
}
|
|
`}},Mle=class{constructor(e,t,n){this.variableNames=["A","B"],this.workPerThread=4,this.isVec4=!0,this.size=!0;let s=128;this.workGroupSize=[s,1,1],this.outputShape=E.assertAndGetBroadcastShape(t,n),this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,[this.workPerThread,1,1]),this.op=e,this.shaderKey=`binaryVec4_${e}`}getUserCode(){return`
|
|
fn binaryOperation(a : vec4<f32>, b : vec4<f32>) -> vec4<f32> {
|
|
${Ap(this.op,this.isVec4)}
|
|
}
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let a = getAAtOutCoordsByGlobalIndex(index);
|
|
let b = getBAtOutCoordsByGlobalIndex(index);
|
|
setOutputFlat(index, binaryOperation(a, b));
|
|
}
|
|
}
|
|
`}},iC=class{constructor(e,t,n){this.variableNames=["A","B"],this.size=!0;let s=128;this.workGroupSize=[s,1,1],this.outputShape=E.assertAndGetBroadcastShape(t,n),this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.shaderKey=`binary_${e}`,this.op=e}getUserCode(){return`
|
|
fn binaryOperation(a : f32, b : f32) -> f32 {
|
|
${Ap(this.op,!1)}
|
|
}
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let a = getAAtOutCoordsByGlobalIndex(index);
|
|
let b = getBAtOutCoordsByGlobalIndex(index);
|
|
setOutputFlat(index, binaryOperation(a, b));
|
|
}
|
|
}
|
|
`}};function lC(e,t,n){if(v.arraysEqual(t,n)&&v.sizeFromShape(t)%4==0)return new Mle(e,t,n);let r=t.length===1&&n.length>1&&t[0]<1024,a=n.length===1&&t.length>1&&n[0]<1024;return r||a?new Ole(e,t,n,a):new iC(e,t,n)}function Js(e){let{inputs:t}=e,{x:n}=t;return e.backend.incRef(n.dataId),{dataId:n.dataId,shape:n.shape,dtype:n.dtype}}var zle={kernelName:Ga,backendName:"webgpu",kernelFunc:Js};function fc(e){let{inputs:t,backend:n}=e,{real:s,imag:r}=t,a=n.makeTensorInfo(s.shape,"complex64"),o=n.tensorMap.get(a.dataId),i=Js({inputs:{x:s},backend:n}),l=Js({inputs:{x:r},backend:n});return o.complexTensorInfos={real:i,imag:l},a}var Lle={kernelName:Jc,backendName:"webgpu",kernelFunc:fc},Km=class{constructor(e,t){this.variableNames=["A"],this.size=!0;let n=128;this.workGroupSize=[n,1,1],this.outputShape=e,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.op=t,this.shaderKey=`unary_${t}`}getUserCode(){return`
|
|
fn unaryOperation(a : f32) -> f32 {
|
|
${hc(this.op,!1)}
|
|
}
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let a = getAAtOutCoordsByGlobalIndex(index);
|
|
setOutputFlat(index, unaryOperation(a));
|
|
}
|
|
}
|
|
`}};function Tn({opType:e,cpuKernelImpl:t,dtype:n}){return({inputs:s,backend:r})=>{let{x:a}=s,o=r,i=n||a.dtype;if(o.shouldExecuteOnCPU([a])&&t!=null){let c=o.tensorMap.get(a.dataId),u=t(c.values,i);return o.makeTensorInfo(a.shape,i,u)}let l=new Km(a.shape,e);return o.runWebGPUProgram(l,[a],i)}}function jn({opSnippet:e,cpuKernelImpl:t,supportsComplex:n=!1,dtype:s}){return({inputs:r,backend:a})=>{let{a:o,b:i}=r,l=a;if(n&&o.dtype==="complex64"){let d=l.tensorMap.get(o.dataId),p=l.tensorMap.get(i.dataId),h,f;if(e!==Wt.MUL)[h,f]=[[d.complexTensorInfos.real,p.complexTensorInfos.real],[d.complexTensorInfos.imag,p.complexTensorInfos.imag]].map(g=>{let[A,x]=g,y={dataId:A.dataId,dtype:A.dtype,shape:o.shape},b={dataId:x.dataId,dtype:x.dtype,shape:i.shape},w=lC(e,o.shape,i.shape);return l.runWebGPUProgram(w,[y,b],zn(A.dtype,x.dtype))});else{let g=new oC(Wt.COMPLEX_MULTIPLY_REAL,o.shape,i.shape),A=new oC(Wt.COMPLEX_MULTIPLY_IMAG,o.shape,i.shape),x=[{dataId:d.complexTensorInfos.real.dataId,dtype:d.complexTensorInfos.real.dtype,shape:o.shape},{dataId:d.complexTensorInfos.imag.dataId,dtype:d.complexTensorInfos.imag.dtype,shape:o.shape},{dataId:p.complexTensorInfos.real.dataId,dtype:p.complexTensorInfos.real.dtype,shape:i.shape},{dataId:p.complexTensorInfos.imag.dataId,dtype:p.complexTensorInfos.imag.dtype,shape:i.shape}];h=l.runWebGPUProgram(g,x,"float32"),f=l.runWebGPUProgram(A,x,"float32")}let m=fc({inputs:{real:h,imag:f},backend:l});return l.disposeData(h.dataId),l.disposeData(f.dataId),m}let c=s||zn(o.dtype,i.dtype);if((o.dtype==="string"||i.dtype==="string"||l.shouldExecuteOnCPU([o,i]))&&t!=null){let d=l.tensorMap.get(o.dataId).values,p=l.tensorMap.get(i.dataId).values,h=o.dtype==="string"?E.fromUint8ToStringArray(d):d,f=o.dtype==="string"?E.fromUint8ToStringArray(p):p,[m,g]=t(o.shape,i.shape,h,f,c);return l.makeTensorInfo(g,c,m)}let u=lC(e,o.shape,i.shape);return l.runWebGPUProgram(u,[o,i],c)}}var{addImpl:Ble,ceilImpl:Wle,concatImpl:Vle,equalImpl:Ule,expImpl:Gle,expm1Impl:Hle,floorImpl:jle,gatherNdImpl:qle,gatherV2Impl:Xle,greaterEqualImpl:Kle,greaterImpl:Zle,lessEqualImpl:Yle,lessImpl:Jle,logImpl:Qle,maxImpl:eue,maximumImpl:tue,minimumImpl:nue,multiplyImpl:sue,negImpl:rue,notEqualImpl:aue,prodImpl:oue,rangeImpl:iue,rsqrtImpl:lue,simpleAbsImpl:uue,sliceImpl:cue,stridedSliceImpl:due,stringNGramsImpl:pue,subImpl:hue,tileImpl:fue,topKImpl:mue,transposeImpl:gue,uniqueImpl:C2e}=bm,Aue=Tn({opType:xt.ABS,cpuKernelImpl:uue}),yue={kernelName:ci,backendName:"webgpu",kernelFunc:Aue},xue=jn({opSnippet:Wt.ADD,cpuKernelImpl:Ble,supportsComplex:!0}),bue={kernelName:Gr,backendName:"webgpu",kernelFunc:xue},vue=class{constructor(e){this.workPerThread=4,this.workGroupSize=[64,1,1],this.size=!0,this.outputShape=e[0],this.variableNames=e.map((t,n)=>`T${n}`),this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,[this.workPerThread,1,1]),this.shaderKey="addN"}getUserCode(){let e=[];this.variableNames.forEach(s=>{e.push(`let v${s} = get${s}AtOutCoordsByCoords(coords);`)});let t=this.variableNames.map(s=>`v${s}`).join(" + ");return`
|
|
${tt()}
|
|
for (var i = 0; i < ${this.workPerThread}; i = i + 1) {
|
|
let flatIndex = index * ${this.workPerThread} + i;
|
|
if (flatIndex < uniforms.size) {
|
|
let coords = getCoordsFromFlatIndex(flatIndex);
|
|
${e.join(`
|
|
`)}
|
|
setOutputFlat(flatIndex, ${t});
|
|
}
|
|
}
|
|
}
|
|
`}};function wue(e){let{inputs:t,backend:n}=e,s=t;if(s.length===1)return Js({inputs:{x:s[0]},backend:n});let r=s.map(i=>i.dtype).reduce((i,l)=>zn(i,l)),a=s.map(i=>i.shape),o=new vue(a);return n.runWebGPUProgram(o,s,r)}var kue={kernelName:Ca,backendName:"webgpu",kernelFunc:wue},uC=class{constructor(e,t,n){this.variableNames=["x"],this.uniforms="axis : i32;";let s=[t];E.assertAxesAreInnerMostDims("arg"+n.charAt(0).toUpperCase()+n.slice(1),s,e.length),this.op=n==="min"?"<":">";let[r,a]=E.computeOutAndReduceShapes(e,s);this.outputShape=r.length===0?[1]:r;let o=v.sizeFromShape(a);this.reductionFactor=2;let i=256,l=Math.min(Math.ceil(o/this.reductionFactor),i);this.workGroupSize=[l,1,1],this.dispatchLayout={x:[],y:this.outputShape.map((c,u)=>u)},this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.inputShape=e,this.shaderKey=`argMinMax${this.op}`}getUserCode(){let e=this.workGroupSize[0]>1,t=`
|
|
var<workgroup> xBestIndices : array<i32, ${this.workGroupSize[0]}>;
|
|
var<workgroup> xBestValues : array<f32, ${this.workGroupSize[0]}>;
|
|
`,n=`
|
|
xBestIndices[localId.x] = bestIndex;
|
|
xBestValues[localId.x] = bestValue;
|
|
|
|
for(var currentSize = WorkGroupSize; currentSize > 1; currentSize = DIV_CEIL(currentSize, ${this.reductionFactor})) {
|
|
workgroupBarrier();
|
|
|
|
for (var w = 0; w < ${this.reductionFactor}; w = w + 1) {
|
|
let i = i32(localId.x) * ${this.reductionFactor} + w;
|
|
if (i < currentSize) {
|
|
let candidateIndex = xBestIndices[i];
|
|
let candidate = xBestValues[i];
|
|
if(candidate ${this.op} bestValue && !isNanCustom(candidate)) {
|
|
bestValue = candidate;
|
|
bestIndex = candidateIndex;
|
|
}
|
|
}
|
|
}
|
|
|
|
xBestIndices[localId.x] = bestIndex;
|
|
xBestValues[localId.x] = bestValue;
|
|
}
|
|
|
|
if (localId.x == 0u) {
|
|
setOutputFlatI32(flatOutputIndex, i32(bestIndex));
|
|
}
|
|
`,s=(o,i)=>this.outputShape.length===1?o:`${o}[${i}]`,r=o=>this.inputShape.length===1?"uniforms.xShape":`uniforms.xShape[${o}]`;return`
|
|
fn DIV_CEIL(a : i32, b : i32) -> i32 {
|
|
return ((a - 1) / b + 1);
|
|
}
|
|
|
|
let WorkGroupSize = ${this.workGroupSize[0]};
|
|
|
|
${e?t:""}
|
|
|
|
// In order to get a flattened index into the input tensor, we need to
|
|
// add back the index along the reduced dimension to |outputCoords|.
|
|
// This function outputs the offset to the first value along
|
|
// |axis| and the stride to get the next value of the input along |axis|.
|
|
fn getInputCoordInfo(globalId : vec3<u32>) -> vec2<i32>{
|
|
let outputCoords = getOutputCoordsWithNonFlatDispatchLayout(globalId);
|
|
var i = ${this.outputShape.length-1};
|
|
|
|
var stride = 1;
|
|
var inputStride = 1;
|
|
var offset = 0;
|
|
|
|
for (var r = 1; r <= ${this.inputShape.length}; r = r + 1) {
|
|
let length = ${r(`${this.inputShape.length} - r`)};
|
|
if (${this.inputShape.length} - r == uniforms.axis) {
|
|
inputStride = stride;
|
|
} else {
|
|
offset = offset + ${s("outputCoords","i")} * stride;
|
|
i = i - 1;
|
|
}
|
|
stride = stride * length;
|
|
}
|
|
|
|
return vec2<i32>(offset, inputStride);
|
|
}
|
|
|
|
fn getInputIndex(coordInfo : vec2<i32>, index : i32) -> i32{
|
|
return coordInfo[0] + coordInfo[1] * index;
|
|
}
|
|
|
|
${Go()} {
|
|
let coordInfo = getInputCoordInfo(globalId);
|
|
|
|
var bestIndex = 0;
|
|
var bestValue = f32(x.numbers[getInputIndex(coordInfo, bestIndex)]);
|
|
|
|
let Length = ${r("uniforms.axis")};
|
|
let WorkPerThread = DIV_CEIL(Length, WorkGroupSize);
|
|
|
|
for (var w = 0; w < WorkPerThread; w = w + 1) {
|
|
let i = i32(globalId.x) * WorkPerThread + w;
|
|
if (i < Length) {
|
|
let candidate = f32(x.numbers[getInputIndex(coordInfo, i)]);
|
|
if (candidate ${this.op} bestValue && !isNanCustom(f32(candidate))) {
|
|
bestValue = candidate;
|
|
bestIndex = i;
|
|
}
|
|
}
|
|
}
|
|
|
|
let flatOutputIndex = i32(globalId.y);
|
|
${e?n:"setOutputFlatI32(flatOutputIndex, bestIndex);"}
|
|
}
|
|
`}},Sue=class{constructor(e,t){this.variableNames=["A"],this.workGroupSize=[16,16,1];let n=new Array(e.length);for(let s=0;s<n.length;s++)n[s]=e[t[s]];this.outputShape=n,this.dispatchLayout={x:[0],y:[1]},this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,[1,1,1]),this.shaderKey="transposeShared"}getUserCode(){return`
|
|
let TILE_DIM = ${this.workGroupSize[0]};
|
|
var<workgroup> tile : array<array<f32, ${this.workGroupSize[0]+1}>, ${this.workGroupSize[0]}>;
|
|
${Xm()}
|
|
fn main([[builtin(local_invocation_id)]] localId : vec3<u32>,
|
|
[[builtin(workgroup_id)]] workgroupId : vec3<u32>) {
|
|
var x = i32(workgroupId.x) * TILE_DIM + i32(localId.x);
|
|
var y = i32(workgroupId.y) * TILE_DIM + i32(localId.y);
|
|
let width = uniforms.outShape[0];
|
|
let height = uniforms.outShape[1];
|
|
if (x < width && y < height) {
|
|
tile[localId.y][localId.x] =
|
|
A.numbers[y * width + x];
|
|
}
|
|
workgroupBarrier();
|
|
|
|
x = i32(workgroupId.y) * TILE_DIM + i32(localId.x);
|
|
y = i32(workgroupId.x) * TILE_DIM + i32(localId.y);
|
|
if (x < height && y < width) {
|
|
setOutputFlat((y * height + x), tile[localId.x]
|
|
[localId.y]);
|
|
}
|
|
}
|
|
`}},Iue=class{constructor(e,t){this.variableNames=["A"],this.workPerThread=4,this.workGroupSize=[64,1,1],this.size=!0;let n=new Array(e.length);for(let s=0;s<n.length;s++)n[s]=e[t[s]];this.outputShape=n,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,[this.workPerThread,1,1]),this.newDim=t,this.shaderKey=`transpose_${t}`}getUserCode(){let e=vn(this.outputShape.length),t=Cue(this.newDim);return`
|
|
${tt()}
|
|
|
|
for(var i = 0; i < ${this.workPerThread}; i = i + 1) {
|
|
let flatIndex = index * ${this.workPerThread} + i;
|
|
if(flatIndex < uniforms.size) {
|
|
let resRC = getCoordsFromFlatIndex(flatIndex);
|
|
setOutputFlat(flatIndex, A.numbers[getFlatIndex${this.outputShape.length}D(
|
|
${e}(${t}), uniforms.aShape)]);
|
|
}
|
|
}
|
|
}
|
|
`}};function Cue(e){let t=e.length;if(t>4)throw Error(`Transpose for rank ${t} is not yet supported`);let n=new Array(t);for(let s=0;s<e.length;s++)n[e[s]]=`resRC[${s}]`;return n.join()}function Rl(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{perm:a}=s,o=n,i=r.shape.length,l=new Array(i);for(let u=0;u<l.length;u++)l[u]=r.shape[a[u]];if(n.shouldExecuteOnCPU([r])){let d=o.tensorMap.get(r.dataId).values,p=gue(d,r.shape,r.dtype,a,l);return n.makeTensorInfo(l,r.dtype,p)}if(r.shape.length===2&&v.arraysEqual(a,[1,0])){let u=new Sue(r.shape,a);return o.runWebGPUProgram(u,[r],r.dtype)}let c=new Iue(r.shape,a);return o.runWebGPUProgram(c,[r],r.dtype)}var Tue={kernelName:go,backendName:"webgpu",kernelFunc:Rl};function Nue(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a}=s,o=v.parseAxisParam(a,r.shape),i=E.getAxesPermutation(o,r.shape.length),l=r,c=[];i!=null&&(l=Rl({inputs:{x:r},backend:n,attrs:{perm:i}}),c.push(l),o=E.getInnerMostAxes(o.length,l.shape.length)),E.assertAxesAreInnerMostDims("argMax",[o[0]],l.shape.length);let u=new uC(l.shape,o[0],"max"),d=[{type:"int32",data:[o[0]]}],p=n.runWebGPUProgram(u,[l],"int32",d);return c.forEach(h=>n.disposeData(h.dataId)),p}var Eue={kernelName:Ta,backendName:"webgpu",kernelFunc:Nue};function Rue(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a}=s,o=v.parseAxisParam(a,r.shape),i=E.getAxesPermutation(o,r.shape.length),l=r,c=[];i!=null&&(l=Rl({inputs:{x:r},backend:n,attrs:{perm:i}}),c.push(l),o=E.getInnerMostAxes(o.length,l.shape.length)),E.assertAxesAreInnerMostDims("argMin",[o[0]],l.shape.length);let u=new uC(l.shape,o[0],"min"),d=[{type:"int32",data:[o[0]]}],p=n.runWebGPUProgram(u,[l],"int32",d);return c.forEach(h=>n.disposeData(h.dataId)),p}var $ue={kernelName:ru,backendName:"webgpu",kernelFunc:Rue},cC=class{constructor(e,t){this.variableNames=["x"],this.uniforms="stride : vec2<i32>; pad : vec2<i32>; dilation : vec2<i32>; convDims : vec2<i32>; filterDims : vec2<i32>;",this.workGroupSize=[128,1,1],this.size=!0,this.outputShape=e.outShape,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.shaderKey=`pool2D_${t}`,this.poolType=t}getUserCode(){let e="resultValue = max(value, resultValue);";this.poolType==="avg"&&(e="resultValue = resultValue + value; count = count + 1.0;");let t="resultValue";return this.poolType==="avg"&&(t="resultValue / count"),`
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let coords = getCoordsFromFlatIndex(index);
|
|
let batch = coords[0];
|
|
let xRCCorner = vec2<i32>(coords.yz) * uniforms.stride - uniforms.pad;
|
|
let xRCorner = xRCCorner.x;
|
|
let xCCorner = xRCCorner.y;
|
|
|
|
var resultValue = ${this.poolType==="avg"?"0.0":"-1.0 / pow(10.0, -20.0)"};
|
|
var count = 0.0;
|
|
|
|
for (var wR = 0; wR < uniforms.filterDims.x; wR = wR + uniforms.dilation.x) {
|
|
let xR = xRCorner + wR;
|
|
|
|
if (xR < 0 || xR >= uniforms.convDims.x) {
|
|
continue;
|
|
}
|
|
|
|
for (var wC = 0; wC < uniforms.filterDims.y; wC = wC + uniforms.dilation.y) {
|
|
let xC = xCCorner + wC;
|
|
if (xC < 0 || xC >= uniforms.convDims.y) {
|
|
continue;
|
|
}
|
|
|
|
let value = getX(batch, xR, xC, coords[3]);
|
|
${e}
|
|
}
|
|
}
|
|
|
|
setOutputFlat(index, ${t});
|
|
}
|
|
}
|
|
`}},dC=class{constructor(e){this.variableNames=["x"],this.uniforms="stride : vec2<i32>;",this.workGroupSize=[256,1,1],this.size=!0,this.outputShape=e.outShape,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.shaderKey="poolWithFilterSizeEqualsOne"}getUserCode(){return`
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let coords = getCoordsFromFlatIndex(index);
|
|
let batch = coords[0];
|
|
let d = coords[3];
|
|
|
|
let xRCCorner = coords.yz * uniforms.stride;
|
|
let xRCorner = xRCCorner.x;
|
|
let xCCorner = xRCCorner.y;
|
|
|
|
let value = getX(batch, xRCorner, xCCorner, d);
|
|
setOutputFlat(index, value);
|
|
}
|
|
}
|
|
`}};function _ue(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{filterSize:a,strides:o,pad:i,dimRoundingMode:l}=s,c=1,u=E.computePool2DInfo(r.shape,a,o,c,i,l);if(u.filterWidth===1&&u.filterHeight===1&&v.arraysEqual(u.inShape,u.outShape))return Js({inputs:{x:r},backend:n});let d,p=[{type:"int32",data:[u.strideHeight,u.strideWidth]}];return u.filterHeight===1&&u.filterWidth===1?d=new dC(u):(d=new cC(u,"avg"),p.push({type:"int32",data:[u.padInfo.top,u.padInfo.left]},{type:"int32",data:[u.dilationHeight,u.dilationWidth]},{type:"int32",data:[u.inHeight,u.inWidth]},{type:"int32",data:[u.effectiveFilterHeight,u.effectiveFilterWidth]})),n.runWebGPUProgram(d,[r],r.dtype,p)}var Due={kernelName:Na,backendName:"webgpu",kernelFunc:_ue};function Pue(e){let{inputs:t,backend:n,attrs:s}=e,{a:r,b:a}=t,{transposeA:o,transposeB:i}=s;return bx({a:r,b:a,transposeA:o,transposeB:i,backend:n})}var Fue={kernelName:Ea,backendName:"webgpu",kernelFunc:Pue},Oue=class{constructor(e,t){this.variableNames=["source"],this.workPerThread=1,this.workGroupSize=[64,1,1],this.size=!0,this.outputShape=t,this.rank=t.length,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,[this.workPerThread,1,1]),this.start=e,this.uniforms=`start : ${vn(e.length)}; `,this.shaderKey="slice"}getUserCode(){let e=vn(this.rank),t=Mue(this.rank),n;return this.start.length===1?n=this.outputShape.map((r,a)=>"sourceLoc = uniforms.start + coords;"):n=this.outputShape.map((r,a)=>`sourceLoc.${vx[a]} = uniforms.start[${a}] + coords.${vx[a]};`),`
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
var sourceLoc : ${e};
|
|
let coords = getCoordsFromFlatIndex(index);
|
|
${n.join(`
|
|
`)}
|
|
setOutputFlat(index, getSource(${t}));
|
|
}
|
|
}
|
|
`}},vx=["x","y","z","w","u","v"];function Mue(e){if(e===1)return"sourceLoc";if(e<=6)return vx.slice(0,e).map(t=>`sourceLoc.${t}`).join(",");throw Error(`Slicing for rank ${e} is not yet supported`)}function mc(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{begin:a,size:o}=s,[i,l]=Pt.parseSliceParams(r,a,o);if(Pt.assertParamsValid(r,i,l),n.shouldExecuteOnCPU([r])||r.dtype==="string"){let d=n.tensorMap.get(r.dataId),p=cue(d.values,i,l,r.shape,r.dtype);return n.makeTensorInfo(l,r.dtype,p)}if(v.sizeFromShape(l)===0)return n.makeTensorInfo(l,r.dtype,[]);let c=new Oue(i,l),u=[{type:"int32",data:i}];return n.runWebGPUProgram(c,[r],r.dtype,u)}var zue={kernelName:Bi,backendName:"webgpu",kernelFunc:mc},Lue=e=>{let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{blockShape:a,crops:o}=s;v.assert(r.shape.length<=4,()=>"batchToSpaceND for rank > 4 with a WebGPU backend not implemented yet");let i=a.reduce((x,y)=>x*y),l=E.getReshaped(r.shape,a,i),c=E.getPermuted(l.length,a.length),u=E.getReshapedPermuted(r.shape,a,i),d=E.getSliceBeginCoords(o,a.length),p=E.getSliceSize(u,o,a.length),h=[],f=Ge({inputs:{x:r},backend:n,attrs:{shape:l}}),m=Rl({inputs:{x:f},backend:n,attrs:{perm:c}}),g=Ge({inputs:{x:m},backend:n,attrs:{shape:u}}),A=mc({inputs:{x:g},backend:n,attrs:{begin:d,size:p}});return h.push(f),h.push(m),h.push(g),h.forEach(x=>n.disposeData(x.dataId)),A},Bue={kernelName:di,backendName:"webgpu",kernelFunc:Lue},pC=jn({opSnippet:Wt.NOT_EQUAL,dtype:"bool",cpuKernelImpl:aue}),Wue={kernelName:Ni,backendName:"webgpu",kernelFunc:pC};function yp(e){let{inputs:t,backend:n}=e,{input:s}=t,r=n.tensorMap.get(s.dataId);return Js({inputs:{x:r.complexTensorInfos.real},backend:n})}var Vue={kernelName:id,backendName:"webgpu",kernelFunc:yp};function Uue(e,t){let n=new Km(e.shape,xt.TO_INT),s=t.runWebGPUProgram(n,[e],"int32");return{dataId:s.dataId,shape:s.shape,dtype:s.dtype}}function wx(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{dtype:a}=s;if(a==="complex64"){if(r.dtype==="complex64")return Js({inputs:{x:r},backend:n});let o=Gt(r.shape),i=wx({inputs:{x:r},backend:n,attrs:{dtype:"float32"}}),l=fc({inputs:{real:i,imag:o},backend:n});return o.dispose(),n.disposeData(i.dataId),l}if(r.dtype==="complex64"){let o=yp({inputs:{input:r},backend:n}),i=wx({inputs:{x:o},backend:n,attrs:{dtype:a}});return n.disposeData(o.dataId),i}if(!v.hasEncodingLoss(r.dtype,a)){let o=Js({inputs:{x:r},backend:n});return{dataId:o.dataId,shape:o.shape,dtype:a}}if(a==="int32")return Uue(r,n);if(a==="bool"){let o=n.makeTensorInfo([],"bool",v.getTypedArrayFromDType("bool",1)),l=pC({inputs:{a:r,b:o},backend:n});return n.disposeData(o.dataId),l}throw new Error(`Error in Cast: failed to cast ${r.dtype} to ${a}`)}var Gue={kernelName:Ra,backendName:"webgpu",kernelFunc:wx},Hue=Tn({opType:xt.CEIL,cpuKernelImpl:Wle}),jue={kernelName:$a,backendName:"webgpu",kernelFunc:Hue},que=class{constructor(e){this.variableNames=["A"],this.uniforms="minVal : f32; maxVal : f32;",this.workPerThread=4,this.workGroupSize=[64,1,1],this.isVec4=!0,this.size=!0,this.outputShape=e,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,[this.workPerThread,1,1]),this.shaderKey="clipVec4"}getUserCode(){return`
|
|
${tt()}
|
|
if(index < uniforms.size) {
|
|
let value = getAAtOutCoordsByGlobalIndex(index);
|
|
var clampedValue : vec4<f32>;
|
|
for (var i = 0; i < 4; i = i + 1) {
|
|
if (isNanCustom(value[i])) {
|
|
clampedValue[i] = value[i];
|
|
} else {
|
|
clampedValue[i] = clamp(value[i], uniforms.minVal, uniforms.maxVal);
|
|
}
|
|
}
|
|
|
|
setOutputFlat(index, clampedValue);
|
|
}
|
|
}
|
|
`}},Xue=class{constructor(e){this.variableNames=["A"],this.uniforms="minVal : f32; maxVal : f32;",this.workGroupSize=[64,1,1],this.size=!0,this.outputShape=e,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.shaderKey="clip"}getUserCode(){return`
|
|
${tt()}
|
|
if(index < uniforms.size) {
|
|
let value = getAAtOutCoordsByGlobalIndex(index);
|
|
if (isNanCustom(value)) {
|
|
setOutputFlat(index, value);
|
|
return;
|
|
}
|
|
setOutputFlat(index, clamp(value, uniforms.minVal, uniforms.maxVal));
|
|
}
|
|
}
|
|
`}};function Kue(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{clipValueMin:a,clipValueMax:o}=s,i,l=[{type:"float32",data:[a]},{type:"float32",data:[o]}];return v.sizeFromShape(r.shape)%4==0?i=new que(r.shape):i=new Xue(r.shape),n.runWebGPUProgram(i,[r],r.dtype,l)}var Zue={kernelName:Hr,backendName:"webgpu",kernelFunc:Kue},Yue=class{constructor(e){this.workPerThread=4,this.workGroupSize=[64,1,1],this.size=!0,this.outputShape=E.computeOutShape(e,1),this.variableNames=e.map((t,n)=>`T${n}`),this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,[this.workPerThread,1,1]),this.shapes=e,this.shaderKey=`concat${e}`}getUserCode(){let e=new Array(this.shapes.length-1),t=[];if(e.length>0){e[0]=this.shapes[0][1];for(let a=1;a<e.length;a++)e[a]=e[a-1]+this.shapes[a][1];t.push(`if (yC < ${e[0]}){ setOutput(coords.x, coords.y, getT0(yR, yC)); }`);for(let a=1;a<e.length;a++){let o=e[a-1];t.push(`elseif (yC < ${e[a]}){ setOutput(coords.x, coords.y, getT${a}(yR, yC - ${o})); }`)}let s=e.length,r=e[e.length-1];t.push(`else { setOutput(coords.x, coords.y, getT${s}(yR, yC - ${r})); }`)}else t.push("setOutput(coords.x, coords.y, getT0(yR, yC));");return`
|
|
${tt()}
|
|
for(var i = 0; i < ${this.workPerThread}; i = i + 1) {
|
|
let flatIndex = index * ${this.workPerThread} + i;
|
|
if(flatIndex < uniforms.size) {
|
|
let coords = getCoordsFromFlatIndex(flatIndex);
|
|
let yR = coords.x;
|
|
let yC = coords.y;
|
|
|
|
${t.join(`
|
|
`)}
|
|
}
|
|
}
|
|
}
|
|
`}};function Zm(e){let{inputs:t,backend:n}=e,{input:s}=t,r=n.tensorMap.get(s.dataId);return Js({inputs:{x:r.complexTensorInfos.imag},backend:n})}var Jue={kernelName:sd,backendName:"webgpu",kernelFunc:Zm};function kx(e,t,n){let s=e[0].dtype;if(s==="complex64"){let u=e.map(m=>yp({inputs:{input:m},backend:n})),d=e.map(m=>Zm({inputs:{input:m},backend:n})),p=kx(u,t,n),h=kx(d,t,n),f=fc({inputs:{real:p,imag:h},backend:n});return u.forEach(m=>n.disposeData(m.dataId)),d.forEach(m=>n.disposeData(m.dataId)),n.disposeData(p.dataId),n.disposeData(h.dataId),f}let r=n.shouldExecuteOnCPU(e);if(s==="string"&&(r=!0),r){let u=e.map(A=>{let x=v.sizeFromShape(A.shape.slice(t));return Ge({inputs:{x:A},backend:n,attrs:{shape:[-1,x]}})}),d=u.map(A=>({vals:n.readSync(A.dataId),shape:A.shape})),p=E.computeOutShape(u.map(A=>A.shape),1),h=u[0].shape[0]===1,f=Vle(d,p,s,h),m=E.computeOutShape(e.map(A=>A.shape),t),g=n.makeTensorInfo(m,s,f);return u.forEach(A=>n.disposeData(A.dataId)),g}let{tensors2D:a,outShape:o}=Que(e,t,n),i=new Yue(a.map(u=>u.shape)),l=n.runWebGPUProgram(i,a,a[0].dtype);a.forEach(u=>n.disposeData(u.dataId));let c=Ge({inputs:{x:l},backend:n,attrs:{shape:o}});return n.disposeData(l.dataId),c}function Que(e,t,n){let s=E.computeOutShape(e.map(a=>a.shape),t);return{tensors2D:e.map(a=>Ge({inputs:{x:a},backend:n,attrs:{shape:[v.sizeFromShape(a.shape.slice(0,t)),v.sizeFromShape(a.shape.slice(t))]}})),outShape:s}}function hC(e){let{inputs:t,backend:n,attrs:s}=e,{axis:r}=s,a=v.parseAxisParam(r,t[0].shape)[0],o=E.computeOutShape(t.map(c=>c.shape),a);if(v.sizeFromShape(o)===0)return n.makeTensorInfo(o,t[0].dtype,[]);let i=t.filter(c=>v.sizeFromShape(c.shape)>0);if(i.length===1)return Js({inputs:{x:i[0]},backend:n});let l=i.map(c=>c.shape);return E.assertParamsConsistent(l,a),kx(i,a,n)}var ece={kernelName:pi,backendName:"webgpu",kernelFunc:hC},tce=class{constructor(e,t){this.variableNames=["A"],this.uniforms=`pad : vec2<i32>; stride : vec2<i32>; dilation : vec2<i32>; outWidth : i32; itemsPerBlockRow : i32;
|
|
inChannels : i32;`,this.workPerThread=4,this.workGroupSize=[64,1,1],this.size=!0,this.outputShape=e,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,[this.workPerThread,1,1]),this.isChannelsLast=t,this.shaderKey=`im2col_${this.isChannelsLast}`}getUserCode(){let e=this.isChannelsLast?0:1,t=this.isChannelsLast?1:2;return`
|
|
${tt()}
|
|
|
|
for(var i = 0; i<${this.workPerThread}; i = i + 1) {
|
|
let flatIndex = index * ${this.workPerThread} + i;
|
|
|
|
let rc = getCoordsFromFlatIndex(flatIndex);
|
|
|
|
if(flatIndex < uniforms.size) {
|
|
let blockIndex = rc[0];
|
|
let pos = rc[1];
|
|
|
|
let offsetY = blockIndex / uniforms.outWidth * uniforms.stride[1] - uniforms.pad[1];
|
|
let d0 = offsetY + uniforms.dilation[1] * pos / uniforms.itemsPerBlockRow;
|
|
var value = 0.0;
|
|
if(d0 < uniforms.aShape[${e}] && d0 >= 0) {
|
|
let offsetX = (blockIndex % uniforms.outWidth) * uniforms.stride[0] -
|
|
uniforms.pad[0];
|
|
let d1 = offsetX + uniforms.dilation[0] * ((pos %
|
|
uniforms.itemsPerBlockRow) / uniforms.inChannels);
|
|
let ch = pos % uniforms.inChannels;
|
|
if(d1 < uniforms.aShape[${t}] && d1 >= 0) {
|
|
value = getA(d0, d1, ch);
|
|
}
|
|
}
|
|
setOutputFlat(flatIndex, value);
|
|
}
|
|
}
|
|
}
|
|
`}};function fC({x:e,filter:t,convInfo:n,backend:s,bias:r=null,preluActivationWeights:a=null,leakyreluAlpha:o=0,activation:i=null}){let l=e.shape,c=n.dataFormat==="channelsLast",u=!1,d=!1,p=c?l[0]*l[1]*l[2]:l[0]*l[2]*l[3],h=Ge({inputs:{x:e},backend:s,attrs:{shape:[1,p,n.inChannels]}}),f=Ge({inputs:{x:t},backend:s,attrs:{shape:[1,n.inChannels,n.outChannels]}}),m=bx({a:h,b:f,transposeA:u,transposeB:d,backend:s,bias:r,activation:i,preluActivationWeights:a,leakyreluAlpha:o}),g=Ge({inputs:{x:m},backend:s,attrs:{shape:n.outShape}});return s.disposeData(h.dataId),s.disposeData(f.dataId),s.disposeData(m.dataId),g}function nce({x:e,filter:t,convInfo:n,backend:s,bias:r=null,preluActivationWeights:a=null,leakyreluAlpha:o=0,activation:i=null}){let{filterWidth:l,filterHeight:c,inChannels:u,strideWidth:d,strideHeight:p,padInfo:h,outWidth:f,outHeight:m,dilationWidth:g,dilationHeight:A,dataFormat:x}=n,y=x==="channelsLast",b=l*c*u,w=m*f,k=[w,b],C=!1,N=!1,R=[],F=Ge({inputs:{x:e},backend:s,attrs:{shape:e.shape.slice(1)}}),_=Ge({inputs:{x:t},backend:s,attrs:{shape:[1,b,-1]}});R.push(F),R.push(_);let P=new tce(k,y),T=[{type:"int32",data:[h.left,h.top]},{type:"int32",data:[d,p]},{type:"int32",data:[g,A]},{type:"int32",data:[f]},{type:"int32",data:[u*l]},{type:"int32",data:[u]}],M=s.runWebGPUProgram(P,[F],F.dtype,T),U=Ge({inputs:{x:M},backend:s,attrs:{shape:[1,k[0],k[1]]}});R.push(M),R.push(U);let j=[1,k[0],k[1]],z=new aC(j,[1,w,n.outChannels],K().get("WEBGPU_MATMUL_WORK_PER_THREAD"),C,N),X=j[1],Z=j[2],J=n.outChannels,ee=[{type:"int32",data:[X]},{type:"int32",data:[J]},{type:"int32",data:[Z]}],ne=s.runWebGPUProgram(z,[U,_],U.dtype,ee),Q=y?[1,m,f,n.outChannels]:[1,n.outChannels,m,f],te=Ge({inputs:{x:ne},backend:s,attrs:{shape:Q}});R.push(ne);for(let oe of R)s.disposeData(oe.dataId);return te}var mC=class{constructor(e,t=!1,n=null,s=!1,r=!1){this.variableNames=["x","W"],this.uniforms=`filterDims : vec2<i32>; pad : vec2<i32>; stride : vec2<i32>; dilation : vec2<i32>;
|
|
dimAOuter : i32; dimBOuter : i32; dimInner : i32;`,this.isVec4=!0,this.outputShape=e.outShape,v.assert(e.dataFormat==="channelsLast",()=>"TODO: NCHW is unimplemented"),this.dispatchLayout={x:[3],y:[1,2],z:[0]},this.workGroupSize=[8,8,1];let a=[4,4,1];this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,a),this.convInfo=e,this.addBias=t,this.activation=n,this.hasPreluActivationWeights=s,this.hasLeakyreluAlpha=r,this.addBias&&this.variableNames.push("bias"),this.hasPreluActivationWeights&&this.variableNames.push("preluActivationWeights"),this.hasLeakyreluAlpha&&this.variableNames.push("leakyreluAlpha"),[this.fitA,this.fitB]=this.getShapeFit(a),this.shaderKey=`conv2DMMVec4_${this.activation}_${this.fitA}_${this.fitB}`}getShapeFit(e){let t=this.workGroupSize[1]*e[1],n=this.workGroupSize[0]*e[0],s=n,r=[t,s],a=[s,n],o=this.outputShape[1]*this.outputShape[2],i=this.outputShape[3],l=this.convInfo.filterHeight*this.convInfo.filterWidth*this.convInfo.inChannels;return[ra(r,[o,l]),ra(a,[l,i])]}getSampleAWithRemainder(e){return`let flatIndex${e} = getFlatIndex4D(coord, uniforms.xShape);
|
|
let divBy4Remainder${e} = flatIndex${e} % 4;
|
|
let divBy4Index${e} = flatIndex${e} / 4;
|
|
let curData${e} = x.numbers[divBy4Index${e}];
|
|
if (divBy4Remainder${e} == 0) {
|
|
temp = curData${e};
|
|
} else {
|
|
// TODO: This could end up being a redundant load with another one in
|
|
// the same shader invocation. Perhaps there's an opportunity for
|
|
// optimization
|
|
let nextData${e} = x.numbers[divBy4Index${e} + 1];
|
|
if (divBy4Remainder${e} == 1) {
|
|
temp = vec4<f32>(curData${e}.yzw, nextData${e}.x);
|
|
} elseif (divBy4Remainder${e} == 2) {
|
|
temp = vec4<f32>(curData${e}.zw, nextData${e}.xy);
|
|
} elseif (divBy4Remainder${e} == 3) {
|
|
temp = vec4<f32>(curData${e}.w, nextData${e}.xyz);
|
|
}
|
|
}
|
|
`}getUserCode(){let t=rC([4,4,1],this.workGroupSize),r=`let outRow = r / uniforms.outShape[2];
|
|
let outCol = r % uniforms.outShape[2];
|
|
let WRow = c / (uniforms.filterDims[1] * uniforms.xShape[3]);
|
|
let WCol = c / uniforms.xShape[3] % uniforms.filterDims[1];
|
|
let inChCoord = c % uniforms.xShape[3];
|
|
var coord = vec4<i32>(
|
|
batch,
|
|
outRow * uniforms.stride[0] + uniforms.dilation[0] * WRow - uniforms.pad[0],
|
|
outCol * uniforms.stride[1] + uniforms.dilation[1] * WCol - uniforms.pad[1],
|
|
inChCoord);
|
|
var resData = vec4<f32>(0.0);
|
|
${this.convInfo.inChannels%4===0?`// The bounds checking is always needed since we use it to pad zero for
|
|
// the 'same' padding type.
|
|
if (coordsInBounds4D(coord, uniforms.xShape)) {
|
|
resData = x.numbers[getFlatIndex4D(coord, uniforms.xShape) / 4];
|
|
} else {
|
|
resData = vec4<f32>(0.0); }`:`var temp = vec4<f32>(0.0);
|
|
${this.getSampleAWithRemainder(1)}
|
|
resData = temp;
|
|
if (WCol == (uniforms.filterDims[1] - 1)) {
|
|
coord = vec4<i32>(
|
|
coord.x, coord.y + 1, coord.z + 1 - uniforms.filterDims[1], 0);
|
|
${this.getSampleAWithRemainder(2)}
|
|
if (inChCoord == 0) {
|
|
resData = vec4<f32>(resData.xyz, temp.x);
|
|
} elseif (inChCoord == 1) {
|
|
resData = vec4<f32>(resData.xy, temp.xy);
|
|
} else {
|
|
resData = vec4<f32>(resData.x, temp.xyz);
|
|
}
|
|
}
|
|
`}
|
|
return resData;`,a=this.fitA?`${r}`:`if (r < uniforms.dimAOuter && c < uniforms.dimInner) {
|
|
${r}
|
|
}
|
|
return vec4<f32>(0.0);
|
|
`,o=this.fitB?"return W.numbers[row * uniforms.dimBOuter / 4 + col];":`if(coordsInBounds2D(vec2<i32>(row, col * 4), vec2<i32>(uniforms.dimInner, uniforms.dimBOuter))) {
|
|
return W.numbers[row * uniforms.dimBOuter / 4 + col];
|
|
}
|
|
return vec4<f32>(0.0);
|
|
`,i="",l="";if(this.activation){let d=aa(this.activation,this.isVec4);if(this.hasPreluActivationWeights)i=`fn activation(a : vec4<f32>, outCoord : vec4<i32>) -> vec4<f32> {
|
|
let b = getPreluActivationWeightsAtOutCoordsByCoords(outCoord);
|
|
${d}
|
|
}`;else{if(this.hasLeakyreluAlpha)throw i=`fn activation(a: vec4<f32>) -> vec4<f32> {
|
|
let b = getLeakyreluAlphaAtOutCoords();
|
|
${d}
|
|
}`,new Error("Leakyrelu is not supported.");i=`
|
|
fn activation(a : vec4<f32>, outCoord : vec4<i32>) -> vec4<f32> {
|
|
${d}
|
|
}`}l="value = activation(value, outCoord);"}let c=this.addBias?"value = value + getBiasAtOutCoordsByCoords(outCoord);":"";return`
|
|
${i}
|
|
fn mm_readA(row : i32, col : i32, globalId : vec3<u32>) -> vec4<f32> {
|
|
let r = row;
|
|
let c = col * 4;
|
|
var batch = i32(globalId.z);
|
|
${a}
|
|
}
|
|
|
|
fn mm_readB(row : i32, col : i32, globalId : vec3<u32>) -> vec4<f32> {
|
|
${o}
|
|
}
|
|
|
|
fn mm_write(row : i32, col : i32, valueInput : vec4<f32>, globalId : vec3<u32>) {
|
|
var batch = i32(globalId.z);
|
|
var value = valueInput;
|
|
if (row < uniforms.dimAOuter && col * 4 < uniforms.dimBOuter)
|
|
{
|
|
let outCoord = vec4<i32>(
|
|
batch,
|
|
row / uniforms.outShape[2],
|
|
row % uniforms.outShape[2],
|
|
col * 4);
|
|
${c}
|
|
${l}
|
|
setOutput(outCoord[0], outCoord[1], outCoord[2], outCoord[3],
|
|
value);
|
|
}
|
|
}
|
|
${t}
|
|
`}},gC=class{constructor(e,t=!1,n=null,s=!1){this.variableNames=["x","W"],this.uniforms="filterDims : vec2<i32>; pad : vec2<i32>; stride : vec2<i32>; dilation : vec2<i32>; dimAOuter : i32; dimBOuter : i32; dimInner : i32;",this.outputShape=e.outShape,v.assert(e.dataFormat==="channelsLast",()=>"TODO: NCHW is unimplemented"),this.dispatchLayout={x:[3],y:[1,2],z:[0]},this.workGroupSize=fx(this.dispatchLayout,this.outputShape),this.elementsPerThread=gx(this.dispatchLayout,this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,this.elementsPerThread),t&&this.variableNames.push("bias"),s&&this.variableNames.push("preluActivationWeights"),this.convInfo=e,this.addBias=t,this.activation=n,this.hasPreluActivationWeights=s,[this.fitA,this.fitB]=this.getShapeFit(),this.shaderKey=`conv2DMM_${this.elementsPerThread}_${this.activation}_${this.fitA}_${this.fitB}`}getShapeFit(){let e=this.workGroupSize[1]*this.elementsPerThread[1],t=this.workGroupSize[0]*this.elementsPerThread[0],n=e>t?e:t;v.assert(n%this.workGroupSize[0]==0&&n%this.workGroupSize[1]==0,()=>"tileInner must be multiple of workgroupsize.x and workgroupsize.y");let s=[e,n],r=[n,t],a=this.outputShape[1]*this.outputShape[2],o=this.outputShape[3],i=this.convInfo.filterHeight*this.convInfo.filterWidth*this.convInfo.inChannels;return[ra(s,[a,i]),ra(r,[i,o])]}getUserCode(){let e=xx(this.elementsPerThread,this.workGroupSize),t=`
|
|
let outRow = row / uniforms.outShape[2];
|
|
let outCol = row % uniforms.outShape[2];
|
|
|
|
let WRow = col / (uniforms.filterDims[1] * uniforms.xShape[3]);
|
|
let WCol = col / uniforms.xShape[3] % uniforms.filterDims[1];
|
|
let coord = vec4<i32>(
|
|
batch,
|
|
outRow * uniforms.stride[0] + uniforms.dilation[0] * WRow - uniforms.pad[0],
|
|
outCol * uniforms.stride[1] + uniforms.dilation[1] * WCol - uniforms.pad[1],
|
|
col % uniforms.xShape[3]);
|
|
// The bounds checking is always needed since we use it to pad zero for the
|
|
// 'same' padding type.
|
|
if(coordsInBounds4D(coord, uniforms.xShape)) {
|
|
return x.numbers[getFlatIndex4D(coord, uniforms.xShape)];
|
|
}
|
|
return 0.0;`,n=this.fitA?`${t}`:`if (row < uniforms.dimAOuter && col < uniforms.dimInner) {
|
|
${t}
|
|
}
|
|
return 0.0;
|
|
`,s=this.fitB?"return W.numbers[row * uniforms.dimBOuter + col];":`if(coordsInBounds2D(vec2<i32>(row, col), vec2<i32>(uniforms.dimInner, uniforms.dimBOuter))) {
|
|
return W.numbers[row * uniforms.dimBOuter + col];
|
|
}
|
|
return 0.0;
|
|
`,r="",a="";if(this.activation){let l=aa(this.activation,!1);this.hasPreluActivationWeights?r=`fn activation(a: f32, outCoord : vec4<i32>) -> f32 {
|
|
let b = getPreluActivationWeightsAtOutCoordsByCoords(outCoord);
|
|
${l}
|
|
}`:r=`
|
|
fn activation(a : f32, outCoord : vec4<i32>) -> f32 {
|
|
${l}
|
|
}
|
|
`,a="value = activation(value, outCoord);"}let o=this.addBias?"value = value + getBiasAtOutCoordsByCoords(outCoord);":"";return`
|
|
${r}
|
|
fn mm_readA(row : i32, col : i32, globalId : vec3<u32>) -> f32 {
|
|
var batch = i32(globalId.z);
|
|
${n}
|
|
}
|
|
|
|
fn mm_readB(row : i32, col : i32, globalId : vec3<u32>) -> f32 {
|
|
${s}
|
|
}
|
|
|
|
fn mm_write(row : i32, col : i32, valueInput : f32, globalId : vec3<u32>) {
|
|
var batch = i32(globalId.z);
|
|
var value = valueInput;
|
|
let outCoord = vec4<i32>(
|
|
batch,
|
|
row / uniforms.outShape[2],
|
|
row % uniforms.outShape[2],
|
|
col);
|
|
${o}
|
|
${a}
|
|
result.numbers[getFlatIndex4D(outCoord, uniforms.outShape)] = value;
|
|
}
|
|
${e}
|
|
`}},AC=class{constructor(e,t=!1,n=null,s=!1){this.variableNames=["x","W"],this.uniforms="filterDims : vec2<i32>; pad : vec2<i32>; stride : vec2<i32>; dilation : vec2<i32>;",this.workGroupSize=[128,1,1],this.outputShape=e.outShape,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),v.assert(e.dataFormat==="channelsLast",()=>"TODO: NCHW is unimplemented"),t&&this.variableNames.push("bias"),s&&this.variableNames.push("preluActivationWeights"),this.convInfo=e,this.addBias=t,this.activation=n,this.hasPreluActivationWeights=s,this.shaderKey=`conv2DNaive_${this.activation}`}getUserCode(){let e="",t="";if(this.activation){let r=aa(this.activation);this.hasPreluActivationWeights?e=`fn activation(a : f32, outCoord : vec4<i32>) -> f32{
|
|
let b = getPreluActivationWeightsAtOutCoordsByCoords(outCoord);
|
|
${r}
|
|
}`:e=`
|
|
fn activation(a : f32, outCoord : vec4<i32>) -> f32{
|
|
${r}
|
|
}
|
|
`,t="value = activation(value, outCoord);"}let n=this.addBias?"value = value + getBiasAtOutCoordsByCoords(outCoord);":"";return`
|
|
${e}
|
|
fn readInp(batch : i32, row : i32, col : i32, chan : i32) -> f32 {
|
|
let coord = vec4<i32>(batch, row, col, chan);
|
|
if(coordsInBounds4D(coord, uniforms.xShape)) {
|
|
return getX(batch, row, col, chan);
|
|
}
|
|
return 0.0;
|
|
}
|
|
|
|
fn readFilt(row : i32, col : i32, xChannel : i32, outChannel : i32) -> f32{
|
|
let coord = vec4<i32>(row, col, xChannel, outChannel);
|
|
if(coordsInBounds4D(coord, uniforms.wShape)) {
|
|
return getW(row, col, xChannel, outChannel);
|
|
}
|
|
return 0.0;
|
|
}
|
|
|
|
fn writeResult(batch : i32, row : i32, col : i32, chan : i32, value : f32) {
|
|
let coord = vec4<i32>(batch, row, col, chan);
|
|
if (coordsInBounds4D(coord, uniforms.outShape)) {
|
|
${n}
|
|
${t}
|
|
setOutput(batch, row, col, chan, value);
|
|
}
|
|
}
|
|
|
|
${hx()} {
|
|
let coords = getOutputCoordsWithFlatDispatchLayout(globalId, localId, numWorkgroups);
|
|
let batch = coords[0];
|
|
let outChannel = coords[3];
|
|
|
|
var acc = 0.0;
|
|
|
|
for (var row = 0; row < uniforms.filterDims[0]; row = row + 1) {
|
|
for (var col = 0; col < uniforms.filterDims[1]; col = col + 1) {
|
|
for (var xChannel = 0; xChannel < uniforms.xShape[3]; xChannel = xChannel + 1) {
|
|
let coordRow = coords[1] * uniforms.stride[0] + uniforms.dilation[0] * row - uniforms.pad[0];
|
|
let coordCol = coords[2] * uniforms.stride[1] + uniforms.dilation[1] * col - uniforms.pad[1];
|
|
let v = readInp(batch, coordRow, coordCol, xChannel);
|
|
let f = readFilt(row, col, xChannel, outChannel);
|
|
acc = acc + v * f;
|
|
}
|
|
}
|
|
}
|
|
|
|
writeResult(batch, coords[1], coords[2], outChannel, acc);
|
|
}
|
|
`}};function sce(e){let{inputs:t,attrs:n,backend:s}=e,{x:r,filter:a}=t,{strides:o,pad:i,dataFormat:l,dilations:c,dimRoundingMode:u}=n,d=E.convertConv2DDataFormat(l),p=E.computeConv2DInfo(r.shape,a.shape,o,c,i,u,!1,d);if(p.filterHeight===1&&p.filterWidth===1&&p.dilationHeight===1&&p.dilationWidth===1&&p.strideHeight===1&&p.strideWidth===1&&(p.padInfo.type==="SAME"||p.padInfo.type==="VALID"))return fC({x:r,filter:a,convInfo:p,backend:s});if(K().getBool("WEBGPU_CONV_SEPARATE_IM2COL_SHADER")&&r.shape[0]===1)return nce({x:r,filter:a,convInfo:p,backend:s});let h,f=[p.padInfo.top,p.padInfo.left],m=[{type:"int32",data:[p.filterHeight,p.filterWidth]},{type:"int32",data:[...f]},{type:"int32",data:[p.strideHeight,p.strideWidth]},{type:"int32",data:[p.dilationHeight,p.dilationWidth]}],g=K().getBool("WEBGPU_USE_NAIVE_CONV2D");if(g?h=new AC(p):(p.inChannels%4==0||p.inChannels===3&&p.padInfo.type==="VALID")&&p.outChannels%4==0&&p.outChannels>=64?h=new mC(p):h=new gC(p),!g){let A=p.outShape[1]*p.outShape[2],x=p.outShape[3],y=p.filterHeight*p.filterWidth*p.inShape[3];m.push({type:"int32",data:[A]},{type:"int32",data:[x]},{type:"int32",data:[y]})}return s.runWebGPUProgram(h,[r,a],r.dtype,m)}var rce={kernelName:_a,backendName:"webgpu",kernelFunc:sce},ace=class{constructor(e){this.variableNames=["x","W"],this.uniforms="filterDims : vec2<i32>; pads : vec2<i32>; stride : vec2<i32>; outBackprop : vec4<i32>; dimAOuter : i32; dimBOuter : i32; dimInner : i32;",this.outputShape=e.inShape,v.assert(e.dataFormat==="channelsLast",()=>"TODO: NCHW is unimplemented"),this.dispatchLayout={x:[3],y:[1,2],z:[0]},this.workGroupSize=fx(this.dispatchLayout,this.outputShape),this.elementsPerThread=gx(this.dispatchLayout,this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,this.elementsPerThread),this.shaderKey=`conv2DDerInputMM_${this.elementsPerThread}`}getUserCode(){return`
|
|
fn mm_readA(row : i32, col : i32, globalId : vec3<u32>) -> f32 {
|
|
var batch = i32(globalId.z);
|
|
if (row < uniforms.dimAOuter && col < uniforms.dimInner) {
|
|
|
|
let outRow = row / uniforms.outShape[2];
|
|
let outCol = row % uniforms.outShape[2];
|
|
|
|
let WRow = col / (uniforms.filterDims[1] * uniforms.outBackprop[3]);
|
|
let WCol = col / uniforms.outBackprop[3] % uniforms.filterDims[1];
|
|
let xR = f32(outRow - uniforms.pads[0] + WRow) / f32(uniforms.stride[0]);
|
|
let xC = f32(outCol - uniforms.pads[1] + WCol) / f32(uniforms.stride[1]);
|
|
if (xR < 0.0 || xR >= f32(uniforms.outBackprop[1]) || fract(xR) > 0.0) {
|
|
return 0.0;
|
|
}
|
|
if (xC < 0.0 || xC >= f32(uniforms.outBackprop[2]) || fract(xC) > 0.0) {
|
|
return 0.0;
|
|
}
|
|
let coord = vec4<i32>(
|
|
batch,
|
|
i32(xR),
|
|
i32(xC),
|
|
col % uniforms.outBackprop[3]);
|
|
return x.numbers[getFlatIndex4D(coord, uniforms.xShape)];
|
|
}
|
|
return 0.0;
|
|
}
|
|
|
|
fn mm_readB(row : i32, col : i32, globalId : vec3<u32>) -> f32 {
|
|
let coordX = uniforms.filterDims.x - 1 -
|
|
row / (uniforms.filterDims[1] * uniforms.outBackprop[3]);
|
|
let coordY = uniforms.filterDims.y - 1 -
|
|
(row / uniforms.outBackprop[3]) % uniforms.filterDims[1];
|
|
if (row < uniforms.dimInner && col < uniforms.dimBOuter &&
|
|
coordX >= 0 && coordY >= 0) {
|
|
let coord = vec4<i32>(coordX, coordY, col,
|
|
row % uniforms.outBackprop[3]);
|
|
return W.numbers[getFlatIndex4D(coord, uniforms.wShape)];
|
|
}
|
|
return 0.0;
|
|
}
|
|
|
|
fn mm_write(row : i32, col : i32, valueInput : f32, globalId : vec3<u32>) {
|
|
var batch = i32(globalId.z);
|
|
var value = valueInput;
|
|
let outCoord = vec4<i32>(
|
|
batch,
|
|
row / uniforms.outShape[2],
|
|
row % uniforms.outShape[2],
|
|
col);
|
|
result.numbers[getFlatIndex4D(outCoord, uniforms.outShape)] = value;
|
|
}
|
|
|
|
${xx(this.elementsPerThread,this.workGroupSize)}
|
|
`}},oce=class{constructor(e){this.variableNames=["dy","W"],this.uniforms="filterDims : vec2<i32>; pads : vec2<i32>; stride : vec2<i32>; outBackprop : vec4<i32>;",this.workGroupSize=[64,1,1],this.size=!0,this.outputShape=e.inShape,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.isChannelsLast=e.dataFormat==="channelsLast",this.shaderKey=`conv2DDerInput_${this.isChannelsLast}`}getUserCode(){let e=this.isChannelsLast?1:2,t=this.isChannelsLast?2:3,n=this.isChannelsLast?3:1;return`
|
|
${tt()} {
|
|
if(index < uniforms.size) {
|
|
let coords = getCoordsFromFlatIndex(index);
|
|
let batch = coords[0];
|
|
let d1 = coords[${n}];
|
|
|
|
let dyCorner = vec2<i32>(coords[${e}]), coords[${t}]) - uniforms.pads;
|
|
let dyRCorner = dyCorner.x;
|
|
let dyCCorner = dyCorner.y;
|
|
|
|
// Convolve dy(?, ?, d2) with w(:, :, d1, d2) to compute dx(xR, xC, d1).
|
|
// ? = to be determined. : = across all values in that axis.
|
|
var dotProd = 0.0;
|
|
for (var wR = 0; wR < uniforms.filterDims.x; wR = wR + 1) {
|
|
let dyR = (f32(dyRCorner) + f32(wR)) / f32(uniforms.stride.x);
|
|
let wRPerm = uniforms.filterDims.x - 1 - wR;
|
|
if (dyR < 0.0 || dyR >= f32(uniforms.outBackprop[1]) || fract(dyR) > 0.0 ||
|
|
wRPerm < 0) {
|
|
continue;
|
|
}
|
|
let idyR = dyR;
|
|
|
|
for (var wC = 0; wC < uniforms.filterDims.y; wC = wC + 1) {
|
|
let dyC = (f32(dyCCorner) + f32(wC)) / f32(uniforms.stride.y);
|
|
let wCPerm = uniforms.filterDims.y - 1 - wC;
|
|
if (dyC < 0.0 || dyC >= f32(uniforms.outBackprop[2]) ||
|
|
fract(dyC) > 0.0 || wCPerm < 0) {
|
|
continue;
|
|
}
|
|
let idyC = dyC;
|
|
|
|
for (var d2 = 0; d2 < uniforms.outBackprop[3]; d2 = d2 + 1) {
|
|
if (${this.isChannelsLast}) {
|
|
let xValue = getDy(batch, idyR, idyC, d2);
|
|
let wValue = getW(wRPerm, wCPerm, d1, d2);
|
|
dotProd = dotProd + xValue * wValue;
|
|
} else {
|
|
let xValue = getDy(batch, d2, idyR, idyC);
|
|
let wValue = getW(wRPerm, wCPerm, d1, d2);
|
|
dotProd = dotProd + xValue * wValue;
|
|
}
|
|
|
|
}
|
|
}
|
|
}
|
|
setOutputFlat(index, dotProd);
|
|
}
|
|
}
|
|
`}};function ice(e){let{inputs:t,backend:n,attrs:s}=e,{dy:r,filter:a}=t,{inputShape:o,strides:i,pad:l,dataFormat:c,dimRoundingMode:u}=s,d=E.convertConv2DDataFormat(c),p=E.computeConv2DInfo(o,a.shape,i,1,l,u,!1,d),h=[{type:"int32",data:[p.filterHeight,p.filterWidth]},{type:"int32",data:[p.filterHeight-1-p.padInfo.top,p.filterWidth-1-p.padInfo.left]},{type:"int32",data:[p.strideHeight,p.strideWidth]},{type:"int32",data:[p.batchSize,p.outHeight,p.outWidth,p.outChannels]}],f;if(K().getBool("WEBGPU_USE_NAIVE_CONV2D_TRANSPOSE"))f=new oce(p);else{f=new ace(p);let m=p.inShape[1]*p.inShape[2],g=p.inShape[3],A=p.filterHeight*p.filterWidth*p.outChannels;h.push({type:"uint32",data:[m]},{type:"uint32",data:[g]},{type:"uint32",data:[A]})}return n.runWebGPUProgram(f,[r,a],"float32",h)}var lce={kernelName:Da,backendName:"webgpu",kernelFunc:ice},uce=Tn({opType:xt.COS}),cce={kernelName:Pa,backendName:"webgpu",kernelFunc:uce},dce=Tn({opType:xt.COSH}),pce={kernelName:Fa,backendName:"webgpu",kernelFunc:dce},hce=class{constructor(e,t,n,s){this.variableNames=["Image","Boxes","BoxInd"],this.uniforms="extrapolationValue : f32;",this.workGroupSize=[64,1,1],this.size=!0;let[r]=t;this.outputShape=[r,n[0],n[1],e],this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.methodId=s==="bilinear"?1:0,this.cropHeightBiggerThan1=this.outputShape[1]>1,this.cropWidthBiggerThan1=this.outputShape[2]>1,this.shaderKey=`cropAndResize_${this.methodId}_${this.cropHeightBiggerThan1}_${this.cropWidthBiggerThan1}`}getUserCode(){let[e,t]=["f32(uniforms.imageShape[1] - 1)","f32(uniforms.imageShape[2] - 1)"],[n,s,r]=this.cropHeightBiggerThan1?[`(${e} / f32(uniforms.outShape[1] - 1))`,"(y2-y1) * height_ratio",`y1*${e} + f32(y)*(height_scale)`]:["0.0","0.0",`0.5 * (y1+y2) * ${e}`],[a,o,i]=this.cropWidthBiggerThan1?[`(${t} / f32(uniforms.outShape[2] - 1))`,"(x2-x1) * width_ratio",`x1*${t} + f32(x)*(width_scale)`]:["0.0","0.0",`0.5 * (x1+x2) * ${t}`];return`
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let coords = getCoordsFromFlatIndex(index);
|
|
let height_ratio = f32(${n});
|
|
let width_ratio = f32(${a});
|
|
let b = coords[0];
|
|
let y = coords[1];
|
|
let x = coords[2];
|
|
let d = coords[3];
|
|
// get box vals
|
|
let y1 = getBoxes(b, 0);
|
|
let x1 = getBoxes(b, 1);
|
|
let y2 = getBoxes(b, 2);
|
|
let x2 = getBoxes(b, 3);
|
|
// get image in batch index
|
|
let bInd = i32(round(getBoxInd(b)));
|
|
if(bInd < 0 || bInd >= uniforms.outShape[0]) {
|
|
return;
|
|
}
|
|
let height_scale = ${s};
|
|
let width_scale = ${o};
|
|
let in_y = ${r};
|
|
if( in_y < 0.0 || in_y > ${e} ) {
|
|
setOutputFlat(index, uniforms.extrapolationValue);
|
|
return;
|
|
}
|
|
let in_x = ${i};
|
|
if( in_x < 0.0 || in_x > ${t} ) {
|
|
setOutputFlat(index, uniforms.extrapolationValue);
|
|
return;
|
|
}
|
|
let sourceFracIndexCR = vec2<f32>(in_x,in_y);
|
|
if(${this.methodId} == 1) {
|
|
// Compute the four integer indices.
|
|
let sourceFloorCR = vec2<i32>(sourceFracIndexCR);
|
|
let sourceCeilCR = vec2<i32>(ceil(sourceFracIndexCR));
|
|
let topLeft = getImage(bInd, sourceFloorCR.y, sourceFloorCR.x, d);
|
|
let bottomLeft = getImage(bInd, sourceCeilCR.y, sourceFloorCR.x, d);
|
|
let topRight = getImage(bInd, sourceFloorCR.y, sourceCeilCR.x, d);
|
|
let bottomRight = getImage(bInd, sourceCeilCR.y, sourceCeilCR.x, d);
|
|
let fracCR = sourceFracIndexCR - vec2<f32>(sourceFloorCR);
|
|
let top = topLeft + (topRight - topLeft) * fracCR.x;
|
|
let bottom = bottomLeft + (bottomRight - bottomLeft) * fracCR.x;
|
|
let newValue = top + (bottom - top) * fracCR.y;
|
|
setOutputFlat(index, newValue);
|
|
} else {
|
|
// Compute the coordinators of nearest neighbor point.
|
|
let sourceNearestCR = vec2<i32>(floor(
|
|
sourceFracIndexCR + vec2<f32>(0.5,0.5)));
|
|
let newValue = getImage(
|
|
bInd, sourceNearestCR.y, sourceNearestCR.x, d);
|
|
setOutputFlat(index, newValue);
|
|
}
|
|
}
|
|
}
|
|
`}},fce=e=>{let{inputs:t,backend:n,attrs:s}=e,{image:r,boxes:a,boxInd:o}=t,{cropSize:i,method:l,extrapolationValue:c}=s,u=new hce(r.shape[3],a.shape,i,l),d=[{type:"float32",data:[c]}];return n.runWebGPUProgram(u,[r,a,o],"float32",d)},mce={kernelName:fi,backendName:"webgpu",kernelFunc:fce},gce=class{constructor(e,t){this.variableNames=["x"],this.workGroupSize=[64,1,1],this.size=!0,this.uniforms="blockSize : i32;",this.outputShape=e,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.shaderKey=`depthToSpace_${t}`,this.dataFormat=t}getUserCode(){return`
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let coords = getCoordsFromFlatIndex(index);
|
|
let b = coords[0];
|
|
let h = ${this.getHeightCoordString()};
|
|
let w = ${this.getWidthCoordString()};
|
|
let d = ${this.getDepthCoordString()};
|
|
|
|
let in_h = h / uniforms.blockSize;
|
|
let offset_h = h % uniforms.blockSize;
|
|
let in_w = w / uniforms.blockSize;
|
|
let offset_w = w % uniforms.blockSize;
|
|
let offset_d = (offset_h * uniforms.blockSize + offset_w) *
|
|
${this.getOutputDepthSize()};
|
|
let in_d = d + offset_d;
|
|
|
|
let rlt = ${this.getInputSamplingString()};
|
|
setOutputFlat(index, rlt);
|
|
}
|
|
}`}getHeightCoordString(){return this.dataFormat==="NHWC"?"coords[1]":"coords[2]"}getWidthCoordString(){return this.dataFormat==="NHWC"?"coords[2]":"coords[3]"}getDepthCoordString(){return this.dataFormat==="NHWC"?"coords[3]":"coords[1]"}getOutputDepthSize(){return this.dataFormat==="NHWC"?"uniforms.outShape[3]":"uniforms.outShape[1]"}getInputSamplingString(){return this.dataFormat==="NHWC"?"getX(b, in_h, in_w, in_d)":"getX(b, in_d, in_h, in_w)"}};function Ace(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{blockSize:a,dataFormat:o}=s,i=r.shape[0],l=o==="NHWC"?r.shape[1]:r.shape[2],c=o==="NHWC"?r.shape[2]:r.shape[3],u=o==="NHWC"?r.shape[3]:r.shape[1],d=l*a,p=c*a,h=u/(a*a),f=o==="NHWC"?[i,d,p,h]:[i,h,d,p],m=[{type:"int32",data:[a]}],g=new gce(f,o);return n.runWebGPUProgram(g,[r],r.dtype,m)}var yce={kernelName:mi,backendName:"webgpu",kernelFunc:Ace},yC=class{constructor(e,t=!1,n=null,s=!1){this.variableNames=["x","W"],this.uniforms="pad : vec2<i32>; stride : vec2<i32>; dilation : vec2<i32>; inDims : vec2<i32>;",this.workGroupSize=[4,4,4],this.isVec4=!0,this.outputShape=e.outShape,this.dispatchLayout={x:[0,1],y:[2],z:[3]},this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,[1,4,4]),v.assert(e.dataFormat==="channelsLast",()=>"TODO: NCHW is unimplemented"),t&&this.variableNames.push("bias"),s&&this.variableNames.push("preluActivationWeights"),this.convInfo=e,this.addBias=t,this.activation=n,this.hasPreluActivation=s,this.shaderKey=`depthwise3x3_${n}`}getUserCode(){let e="",t="";if(this.activation){let r=aa(this.activation,this.isVec4);this.hasPreluActivation?e=`fn activation(a : vec4<f32>, outCoord : vec4<i32>) -> vec4<f32> {
|
|
let b = getPreluActivationWeightsAtOutCoordsByCoords(outCoord);
|
|
${r}
|
|
}`:e=`
|
|
fn activation(a : vec4<f32>, outCoord : vec4<i32>) -> vec4<f32> {
|
|
${r}
|
|
}
|
|
`,t="dotProd[i] = activation(dotProd[i], coords);"}let n=this.addBias?"dotProd[i] = dotProd[i] + getBiasAtOutCoordsByCoords(coords);":"";return`
|
|
${e}
|
|
|
|
${Xm()}
|
|
fn main([[builtin(global_invocation_id)]] globalId: vec3<u32>) {
|
|
let batch = 0;
|
|
let r = i32(globalId.x);
|
|
let c = i32(globalId.y) * 4;
|
|
let d2 = i32(globalId.z) * 4;
|
|
let xRCCorner = vec2<i32>(r, c) * uniforms.stride - uniforms.pad;
|
|
let d1 = d2;
|
|
let q = 0;
|
|
|
|
let xRCorner = xRCCorner.x;
|
|
let xCCorner = xRCCorner.y;
|
|
|
|
var wVals : array<vec4<f32>, 9>;
|
|
wVals[0] = getW(0, 0, d1, q);
|
|
wVals[1] = getW(0, 1, d1, q);
|
|
wVals[2] = getW(0, 2, d1, q);
|
|
wVals[3] = getW(1, 0, d1, q);
|
|
wVals[4] = getW(1, 1, d1, q);
|
|
wVals[5] = getW(1, 2, d1, q);
|
|
wVals[6] = getW(2, 0, d1, q);
|
|
wVals[7] = getW(2, 1, d1, q);
|
|
wVals[8] = getW(2, 2, d1, q);
|
|
|
|
var xVals : array<array<vec4<f32>, 6>, 3>;
|
|
for (var wR = 0; wR < 3; wR = wR + 1) {
|
|
let xR = xRCorner + wR * uniforms.dilation[0];
|
|
for (var wC = 0; wC < 6; wC = wC + 1) {
|
|
let xC = xCCorner + wC * uniforms.dilation[1];
|
|
if (xR < 0 || xR >= uniforms.inDims[0] || xC < 0 || xC >= uniforms.inDims[1]) {
|
|
xVals[wR][wC] = vec4<f32>(0.0);
|
|
} else {
|
|
xVals[wR][wC] = getX(batch, xR, xC, d1);
|
|
}
|
|
}
|
|
}
|
|
|
|
var dotProd : array<vec4<f32>, 4>;
|
|
dotProd[0] = vec4<f32>(0.0);
|
|
dotProd[1] = vec4<f32>(0.0);
|
|
dotProd[2] = vec4<f32>(0.0);
|
|
dotProd[3] = vec4<f32>(0.0);
|
|
|
|
for (var wR = 0; wR < 3; wR = wR + 1) {
|
|
for (var wC = 0; wC < 3; wC = wC + 1) {
|
|
let indexW = wR * 3 + wC;
|
|
dotProd[0] = dotProd[0] + xVals[wR][0 + wC] * wVals[indexW];
|
|
dotProd[1] = dotProd[1] + xVals[wR][1 + wC] * wVals[indexW];
|
|
dotProd[2] = dotProd[2] + xVals[wR][2 + wC] * wVals[indexW];
|
|
dotProd[3] = dotProd[3] + xVals[wR][3 + wC] * wVals[indexW];
|
|
}
|
|
}
|
|
|
|
for (var i = 0; i < 4; i = i + 1) {
|
|
let coords = vec4<i32>(batch, r, c + i, d2);
|
|
if (coordsInBounds4D(coords, uniforms.outShape)) {
|
|
${n}
|
|
${t}
|
|
setOutput(coords[0], coords[1], coords[2], coords[3], dotProd[i]);
|
|
}
|
|
}
|
|
}
|
|
`}},xC=class{constructor(e,t=!1,n=null,s=!1){this.variableNames=["x","W"],this.uniforms="pad : vec2<i32>; stride : vec2<i32>; dilation : vec2<i32>; inDims : vec2<i32>;",this.workGroupSize=[256,1,1],this.outputShape=e.outShape,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),v.assert(e.dataFormat==="channelsLast",()=>"TODO: NCHW is unimplemented"),t&&this.variableNames.push("bias"),s&&this.variableNames.push("preluActivationWeights"),this.convInfo=e,this.addBias=t,this.activation=n,this.hasPreluActivation=s,this.shaderKey=`depthwise_${this.convInfo.filterHeight}_${this.convInfo.filterWidth}_${this.activation}_${this.convInfo.outChannels/this.convInfo.inChannels}`}getUserCode(){let e=this.convInfo.outChannels/this.convInfo.inChannels,t="",n="";if(this.activation){let a=aa(this.activation,!1);this.hasPreluActivation?t=`fn activation(a : f32, outCoord : vec4<i32>) -> f32 {
|
|
let b = getPreluActivationWeightsAtOutCoordsByCoords(outCoord);
|
|
${a}
|
|
}`:t=`
|
|
fn activation(a : f32, outCoord : vec4<i32>) -> f32 {
|
|
${a}
|
|
}
|
|
`,n="dotProd = activation(dotProd, coords);"}let s=this.addBias?"dotProd = dotProd + getBiasAtOutCoordsByCoords(coords);":"";return`
|
|
${t}
|
|
|
|
fn writeResult(batch : i32, row : i32, col : i32, chan : i32, value : f32) {
|
|
let coord = vec4<i32>(batch, row, col, chan);
|
|
if (coordsInBounds4D(coord, uniforms.outShape)) {
|
|
setOutput(batch, row, col, chan, value);
|
|
}
|
|
}
|
|
|
|
${hx()} {
|
|
let coords = getOutputCoordsWithFlatDispatchLayout(globalId, localId, numWorkgroups);
|
|
let batch = coords[0];
|
|
let xRCCorner = vec2<i32>(coords.yz) * uniforms.stride - uniforms.pad;
|
|
let d2 = coords[3];
|
|
let d1 = d2 / ${e};
|
|
let q = d2 - d1 * ${e};
|
|
|
|
let inputRowStart = xRCCorner.x;
|
|
let inputColStart = xRCCorner.y;
|
|
let inputRowEnd = inputRowStart + ${this.convInfo.filterHeight} * uniforms.dilation[0];
|
|
let inputColEnd = inputColStart + ${this.convInfo.filterWidth} * uniforms.dilation[1];
|
|
|
|
// Convolve x(?, ?, d1) with w(:, :, d1, q) to get y(yR, yC, d2).
|
|
// ? = to be determined. : = across all values in that axis.
|
|
var dotProd = 0.0;
|
|
|
|
// Extract if checking out of for loop for performance.
|
|
if (inputRowStart >= 0 && inputColStart >= 0 &&
|
|
inputRowEnd < uniforms.inDims[0] && inputColEnd < uniforms.inDims[1]) {
|
|
// Here using a constant value |this.convInfo.filterHeight| instead
|
|
// of uniform value is in order to loop unrolling.
|
|
for (var wR = 0; wR < ${this.convInfo.filterHeight}; wR = wR + 1) {
|
|
let xR = inputRowStart + wR * uniforms.dilation[0];
|
|
|
|
for (var wC = 0; wC < ${this.convInfo.filterWidth}; wC = wC + 1) {
|
|
let xC = inputColStart + wC * uniforms.dilation[1];
|
|
|
|
let xVal = getX(batch, xR, xC, d1);
|
|
let wVal = getW(wR, wC, d1, q);
|
|
dotProd = dotProd + xVal * wVal;
|
|
}
|
|
}
|
|
} else {
|
|
for (var wR = 0; wR < ${this.convInfo.filterHeight}; wR = wR + 1) {
|
|
let xR = inputRowStart + wR * uniforms.dilation[0];
|
|
|
|
if (xR < 0 || xR >= uniforms.inDims[0]) {
|
|
continue;
|
|
}
|
|
|
|
for (var wC = 0; wC < ${this.convInfo.filterWidth}; wC = wC + 1) {
|
|
let xC = inputColStart + wC * uniforms.dilation[1];
|
|
|
|
if (xC < 0 || xC >= uniforms.inDims[1]) {
|
|
continue;
|
|
}
|
|
|
|
let xVal = getX(batch, xR, xC, d1);
|
|
let wVal = getW(wR, wC, d1, q);
|
|
dotProd = dotProd + xVal * wVal;
|
|
}
|
|
}
|
|
}
|
|
|
|
${s}
|
|
${n}
|
|
writeResult(batch, coords[1], coords[2], d2, dotProd);
|
|
}
|
|
`}};function xce(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,filter:a}=t,{strides:o,pad:i,dilations:l,dimRoundingMode:c}=s,u=l;u==null&&(u=[1,1]);let d=E.computeConv2DInfo(r.shape,a.shape,o,u,i,c,!0),p;d.batchSize===1&&d.inHeight===d.outHeight&&d.inWidth===d.outWidth&&d.strideHeight===1&&d.strideWidth===1&&d.filterHeight===d.filterWidth&&d.inChannels===d.outChannels&&d.filterHeight===3&&d.inChannels%4==0?p=new yC(d):p=new xC(d);let h=[{type:"int32",data:[d.padInfo.top,d.padInfo.left]},{type:"int32",data:[d.strideHeight,d.strideWidth]},{type:"int32",data:[d.dilationHeight,d.dilationWidth]},{type:"int32",data:[d.inHeight,d.inWidth]}];return n.runWebGPUProgram(p,[r,a],r.dtype,h)}var bce={kernelName:Oa,backendName:"webgpu",kernelFunc:xce},bC=jn({opSnippet:Wt.MUL,cpuKernelImpl:sue,supportsComplex:!0}),vce={kernelName:Qa,backendName:"webgpu",kernelFunc:bC},wce=class{constructor(e,t,n){this.variableNames=["x"],this.uniforms="reduceSize : i32;",this.inputShape=[e.batchSize,e.inSize];let[s]=E.computeOutAndReduceShapes(this.inputShape,[1]);this.outputShape=s.length===0?[1]:s,this.reductionFactor=2;let r=256,a=Math.min(Math.ceil(e.inSize/this.reductionFactor),r);this.workGroupSize=[a,1,1],this.dispatchLayout={x:[],y:this.outputShape.map((o,i)=>i)},this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.reduceType=t,this.shaderKey=`reduce_${t}_${n}`}getUserCode(){let e=this.workGroupSize[0]>1,t="",n="0.0";this.reduceType==="min"||this.reduceType==="max"?(t=`
|
|
if (isNanCustom(candidate)) {
|
|
bestValue = uniforms.NAN;
|
|
} elseif (candidate ${this.reduceType==="min"?"<":">"}
|
|
bestValue)
|
|
{ bestValue = candidate; }`,n="f32(x.numbers[offset])"):this.reduceType==="sum"||this.reduceType==="mean"?t=" bestValue = bestValue + candidate; ":this.reduceType==="prod"&&(t=" bestValue = bestValue * candidate; ",n="1.0");let s=this.reduceType==="mean"?"setOutputFlat(flatOutputIndex, bestValue / f32(uniforms.reduceSize));":"setOutputFlat(flatOutputIndex, bestValue);",r=`
|
|
var<workgroup> xBestValues : array<f32, ${this.workGroupSize[0]}>;
|
|
`,a=`
|
|
xBestValues[localId.x] = bestValue;
|
|
${this.reduceType==="sum"||this.reduceType==="mean"||this.reduceType==="prod"?`bestValue = ${n};`:" "}
|
|
var currentSize = WorkGroupSize;
|
|
for(; currentSize > 1;) {
|
|
workgroupBarrier();
|
|
for (var w = 0; w < ${this.reductionFactor}; w = w + 1) {
|
|
let i = i32(localId.x) * ${this.reductionFactor} + w;
|
|
if (i < currentSize) {
|
|
let candidate = xBestValues[i];
|
|
${t}
|
|
}
|
|
}
|
|
workgroupBarrier();
|
|
xBestValues[localId.x] = bestValue;
|
|
currentSize = DIV_CEIL(currentSize, ${this.reductionFactor});
|
|
${this.reduceType==="sum"||this.reduceType==="mean"||this.reduceType==="prod"?`if(currentSize > 1) { bestValue = ${n}; }`:""}
|
|
}
|
|
if (localId.x == 0u) {
|
|
${s}
|
|
}
|
|
`;return`
|
|
fn DIV_CEIL(a : i32, b : i32) -> i32 {
|
|
return ((a - 1) / b + 1);
|
|
}
|
|
let WorkGroupSize = ${this.workGroupSize[0]};
|
|
${e?r:""}
|
|
fn getOffset(globalId : vec3<u32>) -> i32 {
|
|
let outputCoords = getOutputCoordsWithNonFlatDispatchLayout(globalId);
|
|
let offset = ${this.outputShape.length===1?"outputCoords":"outputCoords[0]"} * uniforms.reduceSize;
|
|
return offset;
|
|
}
|
|
${Go()} {
|
|
let offset = getOffset(globalId);
|
|
var bestValue = ${n};
|
|
let Length = uniforms.reduceSize;
|
|
let WorkPerThread = DIV_CEIL(Length, WorkGroupSize);
|
|
for (var w = 0; w < WorkPerThread; w = w + 1) {
|
|
let i = i32(globalId.x) * WorkPerThread + w;
|
|
if (i < Length) {
|
|
let candidate = f32(x.numbers[offset + i]);
|
|
${t}
|
|
}
|
|
}
|
|
let flatOutputIndex = i32(globalId.y);
|
|
${e?a:s}
|
|
}
|
|
`}};function xp(e,t,n,s,r){let a=e.shape.length,o=[],i=v.parseAxisParam(t,e.shape),l=i,c=E.getAxesPermutation(l,a),u=e;c!=null&&(u=Rl({inputs:{x:e},attrs:{perm:c},backend:r}),l=E.getInnerMostAxes(l.length,a),o.push(u)),E.assertAxesAreInnerMostDims(s,l,a);let[d,p]=E.computeOutAndReduceShapes(u.shape,l),h=d;n&&(h=E.expandShapeToKeepDim(d,i));let f;if((s==="max"||s==="prod")&&r.shouldExecuteOnCPU([u])){let m=r.tensorMap.get(u.dataId).values;switch(s){case"max":let g=eue(m,v.sizeFromShape(p),h,e.dtype);f=r.makeTensorInfo(h,e.dtype,g);break;case"prod":let{outVals:A,outShape:x,outDtype:y}=oue(u.shape,u.dtype,m,l);f=r.makeTensorInfo(x,y,A);break;default:throw new Error(`${s} CPU implementation is not yet supported.`)}}else{let m=v.sizeFromShape(p),A=v.sizeFromShape(u.shape)/m,x={windowSize:m,inSize:m,batchSize:A,outSize:1},y=s==="mean"?"float32":bd(e.dtype),b=[{type:"int32",data:[m]}],w=new wce(x,s,y),k=r.runWebGPUProgram(w,[u],y,b);o.push(k),f=Ge({inputs:{x:k},attrs:{shape:h},backend:r})}return o.forEach(m=>r.disposeData(m.dataId)),f}function Sx(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a,keepDims:o}=s;return xp(r,a,o,"sum",n)}var kce={kernelName:co,backendName:"webgpu",kernelFunc:Sx};function Sce(e){let{inputs:t,backend:n,attrs:s}=e,{equation:r}=s,a=t,{allDims:o,summedDims:i,idDims:l}=E.decodeEinsumEquation(r,a.length);E.checkEinsumDimSizes(o.length,l,a);let{path:c,steps:u}=E.getEinsumComputePath(i,l),d=u.length,p=null,h=o.length,f=[];for(let m=0;m<d;++m){for(let g of u[m]){let{permutationIndices:A,expandDims:x}=E.getEinsumPermutation(h,l[g]),y;E.isIdentityPermutation(A)?y=a[g]:(y=Rl({inputs:{x:a[g]},backend:n,attrs:{perm:A}}),f.push(y));let b=y.shape.slice();for(let w=0;w<x.length;++w)b.splice(x[w],0,1);v.arraysEqual(y.shape,b)||(y=Ge({inputs:{x:y},backend:n,attrs:{shape:b}}),f.push(y)),p===null?p=y:(p=bC({inputs:{a:y,b:p},backend:n}),f.push(p))}m<d-1&&(c[m]>=0&&(p=Sx({inputs:{x:p},backend:n,attrs:{axis:c[m]-(o.length-h),keepDims:!1}}),f.push(p)),h--)}for(let m of f)m!==p&&n.disposeData(m.dataId);return p}var Ice={kernelName:nd,backendName:"webgpu",kernelFunc:Sce},Cce=Tn({opType:xt.ELU}),Tce={kernelName:za,backendName:"webgpu",kernelFunc:Cce},Nce=jn({opSnippet:Wt.EQUAL,dtype:"bool",cpuKernelImpl:Ule}),Ece={kernelName:gi,backendName:"webgpu",kernelFunc:Nce},vC=Tn({opType:xt.EXP,cpuKernelImpl:Gle,dtype:"float32"}),Rce={kernelName:La,backendName:"webgpu",kernelFunc:vC};function Ix(e){let{inputs:t,attrs:n,backend:s}=e,{dim:r}=n,{input:a}=t,o=a.shape.length,i=a.shape.slice(),l=r;return r<0&&(v.assert(-(o+1)<=r,()=>`Axis must be in the interval [${-(o+1)}, ${o}]`),l=o+r+1),i.splice(l,0,1),Ge({inputs:{x:a},backend:s,attrs:{shape:i}})}var $ce={kernelName:Ai,backendName:"webgpu",kernelFunc:Ix},_ce=Tn({opType:xt.EXPM1,cpuKernelImpl:Hle}),Dce={kernelName:yi,backendName:"webgpu",kernelFunc:_ce},Pce=class{constructor(e){this.variableNames=[],this.outputShape=[],this.uniforms="value : f32;",this.workGroupSize=[64,1,1],this.size=!0,this.outputShape=e,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.shaderKey="fill"}getUserCode(){return`
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
setOutputFlat(index, uniforms.value);
|
|
}
|
|
}
|
|
`}};function gc(e){let{backend:t,attrs:n}=e,{shape:s,value:r}=n,{dtype:a}=n;if(a=a||v.inferDtype(r),a==="string"){let o=v.getArrayFromDType(a,v.sizeFromShape(s));return o.fill(r),t.makeTensorInfo(s,a,o)}else{let o=new Pce(s),i=[{type:"float32",data:[r]}];return t.runWebGPUProgram(o,[],a,i)}}var Fce={kernelName:du,backendName:"webgpu",kernelFunc:gc},Oce=class{constructor(e){this.outputShape=[],this.variableNames=["x"],this.workGroupSize=[64,1,1],this.size=!0,this.outputShape=e,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.shaderKey="flipLeftRight"}getUserCode(){return`
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let coords = getCoordsFromFlatIndex(index);
|
|
let coordX = uniforms.xShape[2] - coords[2] - 1;
|
|
let outputValue = getX(coords[0], coords[1], coordX, coords[3]);
|
|
setOutputFlat(index, outputValue);
|
|
}
|
|
}
|
|
`}},Mce={kernelName:xi,backendName:"webgpu",kernelFunc:({inputs:e,backend:t})=>{let{image:n}=e,s=t,r=new Oce(n.shape);return s.runWebGPUProgram(r,[n],n.dtype)}},zce=Tn({opType:xt.FLOOR,cpuKernelImpl:jle}),Lce={kernelName:Ba,backendName:"webgpu",kernelFunc:zce},Bce=jn({opSnippet:Wt.INT_DIV,dtype:"int32"}),Wce={kernelName:Wa,backendName:"webgpu",kernelFunc:Bce},Vce=(e,t,n,s,r)=>{let a=[s,...n];return r&&a.push(r),e.createBindGroup({layout:t,entries:a.map((o,i)=>({binding:i,resource:o}))})},wC=(e,t,n,s,r,a=!1)=>{let o={dtype:r.dtype,shape:r.shape},i=xie(s,o,t,a),l=e.createShaderModule({code:i});return e.createComputePipeline({layout:n,compute:{module:l,entryPoint:"main"}})};function kC(e,t,n,s="",r=""){return e.shaderKey+"_"+(e.workGroupSize?e.workGroupSize.join(","):"")+t.map(o=>o.length).join(",")+n.join(",")+e.variableNames.join(",")+s+r}function SC(e){let{externalImage:t,backend:n,attrs:s,outShape:r,useImport:a}=e,{numChannels:o}=s,i=v.sizeFromShape(r),l=v.computeStrides(r),c=n.makeTensorInfo(r,"int32"),u=n.getFromPixelsProgram(a?"import":"copyExternal");u.updateOutputShape(r);let d=[c.shape],p=[c.dtype,a?"import":"copyExternal"],h=kC(u,d,p),f=u.getLayout(n.device),m=n.getAndSavePipeline(h,()=>wC(n.device,u,f.pipelineLayout,[],c,!0));u.setPipeline(m),a||n.queue.copyExternalImageToTexture({source:t,origin:{x:0,y:0}},{texture:u.makeInputTexture(n.device,r[1],r[0])},[r[1],r[0]]);let g=n.tensorMap.get(c.dataId);g.bufferInfo.buffer=n.acquireBuffer(g.bufferInfo.byteSize);let A=[i,o,...l,...u.dispatch];u.setUniform(n.device,A);let x;if(a){let y={source:t};x=n.device.importExternalTexture(y)}else x=u.inputTexture.createView();return n.runFromPixelsProgram(u,g.bufferInfo.buffer,f,x,c.dataId),c}var Uce={kernelName:dd,backendName:"webgpu",kernelFunc:Gce},Ac;function Gce(e){let{inputs:t,backend:n,attrs:s}=e,{pixels:r}=t,{numChannels:a}=s;if(r==null)throw new Error("pixels passed to tf.browser.fromPixels() can not be null");let o=typeof HTMLVideoElement!="undefined"&&r instanceof HTMLVideoElement,i=typeof HTMLImageElement!="undefined"&&r instanceof HTMLImageElement,l=typeof HTMLCanvasElement!="undefined"&&r instanceof HTMLCanvasElement||typeof OffscreenCanvas!="undefined"&&r instanceof OffscreenCanvas,c=typeof ImageBitmap!="undefined"&&r instanceof ImageBitmap,[u,d]=o?[r.videoWidth,r.videoHeight]:[r.width,r.height],p=[d,u,a];if(K().getBool("WEBGPU_USE_IMPORT")&&o)return SC({externalImage:r,backend:n,attrs:s,outShape:p,useImport:!0});if((o||i)&&(Ac==null&&(Ac=document.createElement("canvas").getContext("2d")),Ac.canvas.width=u,Ac.canvas.height=d,Ac.drawImage(r,0,0,u,d),r=Ac.canvas),c||l||o||i)return SC({externalImage:r,backend:n,attrs:s,outShape:p,useImport:!1});let h=r.data,f=h;if(a!=null&&a!==4){f=new Uint8Array(r.width*r.height*a);let A=h.length,x=0;for(let y=0;y<A;y++)y%4<a&&(f[x++]=h[y])}let m=n.makeTensorInfo(p,"int32"),g=n.tensorMap.get(m.dataId);return g.values=new Int32Array(f),n.maybeReleaseBuffer(m.dataId),n.uploadToGPU(m.dataId),m}var Hce=class{constructor(e,t,n,s,r){this.uniforms="varianceEpsilon : f32;",this.workGroupSize=[128,1,1],this.size=!0,this.variableNames=["x","mean","variance"],E.assertAndGetBroadcastShape(e,t),E.assertAndGetBroadcastShape(e,n),this.outputShape=e,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),s!=null&&(E.assertAndGetBroadcastShape(e,s),this.variableNames.push("offset")),r!=null&&(E.assertAndGetBroadcastShape(e,r),this.variableNames.push("scale")),this.offsetShape=s,this.scaleShape=r,this.shaderKey="batchNorm"}getUserCode(){let e="0.0";this.offsetShape!=null&&(e="getOffsetAtOutCoordsByGlobalIndex(index)");let t="1.0";return this.scaleShape!=null&&(t="getScaleAtOutCoordsByGlobalIndex(index)"),`
|
|
${tt()}
|
|
if (index < uniforms.size)
|
|
{
|
|
let xValue = getXAtOutCoordsByGlobalIndex(index);
|
|
let meanValue = getMeanAtOutCoordsByGlobalIndex(index);
|
|
let varianValue = getVarianceAtOutCoordsByGlobalIndex(index);
|
|
let offsetValue = ${e};
|
|
let scaleValue = ${t};
|
|
let inv = scaleValue * inverseSqrt(varianValue + f32(uniforms.varianceEpsilon));
|
|
setOutputFlat(index,dot(vec3<f32>(xValue, -meanValue, offsetValue), vec3<f32>(inv, inv, 1.0)));
|
|
}
|
|
}
|
|
`}},jce={kernelName:Va,backendName:"webgpu",kernelFunc:({inputs:e,attrs:t,backend:n})=>{let{x:s,scale:r,offset:a,mean:o,variance:i}=e,{varianceEpsilon:l}=t,c=n,u=[s,o,i],d=null;a!=null&&(d=a.shape,u.push(a));let p=null;r!=null&&(p=r.shape,u.push(r));let h=new Hce(s.shape,o.shape,i.shape,d,p),f=[{type:"float32",data:[l]}];return c.runWebGPUProgram(h,u,s.dtype,f)}};function qce(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,filter:a,bias:o,preluActivationWeights:i}=t,{strides:l,pad:c,dataFormat:u,dilations:d,dimRoundingMode:p,activation:h,leakyreluAlpha:f}=s,m=E.convertConv2DDataFormat(u),g=E.computeConv2DInfo(r.shape,a.shape,l,d,c,p,!1,m),A=o!=null,x=i!=null,y;if(g.filterHeight===1&&g.filterWidth===1&&g.dilationHeight===1&&g.dilationWidth===1&&g.strideHeight===1&&g.strideWidth===1&&(g.padInfo.type==="SAME"||g.padInfo.type==="VALID"))return fC({x:r,filter:a,convInfo:g,backend:n,bias:o,activation:h,preluActivationWeights:i,leakyreluAlpha:f});let b=K().getBool("WEBGPU_USE_NAIVE_CONV2D"),w=g.inChannels%4==0&&g.outChannels%4==0,k=[g.padInfo.top,g.padInfo.left],C=[{type:"int32",data:[g.filterHeight,g.filterWidth]},{type:"int32",data:[...k]},{type:"int32",data:[g.strideHeight,g.strideWidth]},{type:"int32",data:[g.dilationHeight,g.dilationWidth]}];if(b)y=new AC(g,A,h,x);else{w?y=new mC(g,A,h,x):y=new gC(g,A,h,x);let R=g.outShape[1]*g.outShape[2],F=g.outShape[3],_=g.filterHeight*g.filterWidth*g.inShape[3];C.push({type:"int32",data:[R]},{type:"int32",data:[F]},{type:"int32",data:[_]})}let N=[r,a];return A&&N.push(o),x&&N.push(i),n.runWebGPUProgram(y,N,r.dtype,C)}var Xce={kernelName:xo,backendName:"webgpu",kernelFunc:qce};function Kce(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,filter:a,bias:o,preluActivationWeights:i}=t,{strides:l,pad:c,dilations:u,dimRoundingMode:d,activation:p}=s,h=u;h==null&&(h=[1,1]),v.assert(E.eitherStridesOrDilationsAreOne(l,h),()=>`Error in depthwiseConv2d: Either strides or dilations must be 1. Got strides ${l} and dilations '${h}'`);let f=E.computeConv2DInfo(r.shape,a.shape,l,h,c,d,!0),m=[r,a],g=o!=null,A=i!=null;g&&m.push(o),A&&m.push(i);let x;f.batchSize===1&&f.inHeight===f.outHeight&&f.inWidth===f.outWidth&&f.strideHeight===1&&f.strideWidth===1&&f.filterHeight===f.filterWidth&&f.inChannels===f.outChannels&&f.filterHeight===3&&f.inChannels%4==0?x=new yC(f,g,p,A):x=new xC(f,g,p,A);let y=[{type:"int32",data:[f.padInfo.top,f.padInfo.left]},{type:"int32",data:[f.strideHeight,f.strideWidth]},{type:"int32",data:[f.dilationHeight,f.dilationWidth]},{type:"int32",data:[f.inHeight,f.inWidth]}];return n.runWebGPUProgram(x,m,"float32",y)}var Zce={kernelName:bo,backendName:"webgpu",kernelFunc:Kce},Yce=class{constructor(e,t){this.variableNames=["A","indices"],this.workGroupSize=[64,1,1],this.size=!0,this.outputShape=t,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.shaderKey=`gathernd_${e}`,this.sliceDim=e,this.uniforms=`sliceDim : i32; strides : ${vn(e)};`}getUserCode(){let e;return this.sliceDim>1?e="uniforms.strides[j]":e="uniforms.strides",`
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let coords = getCoordsFromFlatIndex(index);
|
|
var flattenIndex = 0;
|
|
for (var j = 0; j < uniforms.sliceDim; j = j + 1) {
|
|
let indexTemp = i32(round(getIndices(coords[0], j)));
|
|
let strideNum = ${e};
|
|
flattenIndex = flattenIndex + indexTemp * strideNum;
|
|
}
|
|
|
|
setOutputFlat(index, getA(flattenIndex, coords[1]));
|
|
}
|
|
}
|
|
`}};function Jce(e){let{inputs:t,backend:n}=e,{params:s,indices:r}=t,a=r.shape,o=a[a.length-1],i=v.sizeFromShape(s.shape),[l,c,u,d]=E.prepareAndValidate(s,r),p=Ge({inputs:{x:r},backend:n,attrs:{shape:[c,o]}}),h=Ge({inputs:{x:s},backend:n,attrs:{shape:[v.sizeFromShape(s.shape)/u,u]}});if(n.shouldExecuteOnCPU([s,r])||s.dtype==="string"){let x=n.readSync(r.dataId),y=n.bufferSync(s),b=qle(x,y,s.dtype,c,o,u,d,s.shape,i);return n.makeTensorInfo(l,s.dtype,b.values)}let f=new Yce(o,[c,u]),m=[{type:"int32",data:[o]},{type:"int32",data:d}],g=n.runWebGPUProgram(f,[h,p],h.dtype,m),A=Ge({inputs:{x:g},backend:n,attrs:{shape:l}});return n.disposeData(p.dataId),n.disposeData(h.dataId),n.disposeData(g.dataId),A}var Qce={kernelName:vi,backendName:"webgpu",kernelFunc:Jce},ede=class{constructor(e,t){this.variableNames=["A","indices"],this.workGroupSize=[64,1,1],this.size=!0,this.outputShape=e.slice(),this.aShape=e,this.outputShape=t,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.shaderKey="gather"}getUserCode(){let e=tde(this.aShape,"i32");return`
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let resRC = getCoordsFromFlatIndex(index);
|
|
setOutputFlat(index, getA(${e}));
|
|
}
|
|
}
|
|
`}};function tde(e,t="int"){let n=["resRC.x","resRC.y","resRC.z","resRC.w"],s=[];for(let r=0;r<e.length;r++)r===2?s.push(`${t}(getIndices(resRC.x, resRC.z))`):s.push(`${n[r]}`);return s.join()}function IC(e){let{inputs:t,backend:n,attrs:s}=e,{x:r,indices:a}=t,{axis:o,batchDims:i}=s,l=v.parseAxisParam(o,r.shape)[0],c=E.segment_util.collectGatherOpShapeInfo(r,a,l,i),u=v.sizeFromShape(a.shape),d=[],p=Ge({inputs:{x:r},backend:n,attrs:{shape:[c.batchSize,c.outerSize,c.dimSize,c.sliceSize]}}),h=Ge({inputs:{x:a},backend:n,attrs:{shape:[c.batchSize,u/c.batchSize]}});d.push(p),d.push(h);let f=[c.batchSize,c.outerSize,u/c.batchSize,c.sliceSize];if(n.shouldExecuteOnCPU([r,a])){let y=n.tensorMap.get(h.dataId).values,b=ze(h.shape,h.dtype,y),k=n.tensorMap.get(p.dataId).values,C=ze(p.shape,p.dtype,k),N=Xle(C,b,f);return d.forEach(R=>n.disposeData(R.dataId)),n.makeTensorInfo(c.outputShape,N.dtype,N.values)}let m=new ede(p.shape,f),g=n.runWebGPUProgram(m,[p,h],p.dtype);d.push(g);let A=Ge({inputs:{x:g},backend:n,attrs:{shape:c.outputShape}});return d.forEach(x=>n.disposeData(x.dataId)),A}var nde={kernelName:bi,backendName:"webgpu",kernelFunc:IC},sde=jn({opSnippet:Wt.GREATER,cpuKernelImpl:Zle,dtype:"bool"}),rde={kernelName:wi,backendName:"webgpu",kernelFunc:sde},ade=jn({opSnippet:Wt.GREATER_EQUAL,dtype:"bool",cpuKernelImpl:Kle}),ode={kernelName:Ua,backendName:"webgpu",kernelFunc:ade},ide=jn({opSnippet:Wt.LESS,dtype:"bool",cpuKernelImpl:Jle}),lde={kernelName:Si,backendName:"webgpu",kernelFunc:ide},ude=jn({opSnippet:Wt.LESS_EQUAL,dtype:"bool",cpuKernelImpl:Yle}),cde={kernelName:Ii,backendName:"webgpu",kernelFunc:ude},dde=Tn({opType:xt.LOG,cpuKernelImpl:Qle}),pde={kernelName:Ha,backendName:"webgpu",kernelFunc:dde},hde=jn({opSnippet:Wt.LOGICAL_AND,dtype:"bool"}),fde={kernelName:Ci,backendName:"webgpu",kernelFunc:hde},mde=Tn({opType:xt.LOGICAL_NOT}),gde={kernelName:gu,backendName:"webgpu",kernelFunc:mde};function CC(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{reductionIndices:a,keepDims:o}=s;return xp(r,a,o,"max",n)}var Ade={kernelName:ja,backendName:"webgpu",kernelFunc:CC},yde=jn({opSnippet:Wt.MAX,cpuKernelImpl:tue}),xde={kernelName:qa,backendName:"webgpu",kernelFunc:yde};function bde(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{filterSize:a,strides:o,pad:i,dimRoundingMode:l}=s,c=1,u=E.computePool2DInfo(r.shape,a,o,c,i,l),d,p=[];if(u.filterHeight===1&&u.filterWidth===1){if(v.arraysEqual(u.inShape,u.outShape))return Js({inputs:{x:r},backend:n});d=new dC(u),p.push({type:"int32",data:[u.strideHeight,u.strideWidth]})}else d=new cC(u,"max"),p.push({type:"int32",data:[u.strideHeight,u.strideWidth]},{type:"int32",data:[u.padInfo.top,u.padInfo.left]},{type:"int32",data:[u.dilationHeight,u.dilationWidth]},{type:"int32",data:[u.inHeight,u.inWidth]},{type:"int32",data:[u.effectiveFilterHeight,u.effectiveFilterWidth]});return n.runWebGPUProgram(d,[r],r.dtype,p)}var vde={kernelName:Xa,backendName:"webgpu",kernelFunc:bde};function wde(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{keepDims:a,axis:o}=s;return xp(r,o,a,"mean",n)}var kde={kernelName:Ka,backendName:"webgpu",kernelFunc:wde};function Sde(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a,keepDims:o}=s;return xp(r,a,o,"min",n)}var Ide={kernelName:Za,backendName:"webgpu",kernelFunc:Sde},Cde=jn({opSnippet:Wt.MIN,cpuKernelImpl:nue}),Tde={kernelName:Ya,backendName:"webgpu",kernelFunc:Cde},Nde=class{constructor(e,t,n){this.uniforms="",this.variableNames=["x"],this.workGroupSize=[64,1,1],this.size=!0,this.outputShape=t.map((s,r)=>s[0]+e[r]+s[1]),this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.xShape=e,t.map((s,r)=>{this.uniforms+=` pad${r} : vec2<i32>;`}),this.offset=n==="reflect"?0:1,this.shaderKey=`mirrorPad_${n}`}getUserCode(){let e=this.xShape.length,t=this.xShape.map((l,c)=>`uniforms.pad${c}[0]`).join(","),n=this.xShape.map((l,c)=>`uniforms.pad${c}[0] + uniforms.xShape${e>1?`[${c}]`:""}`).join(","),s=e===1?"start":"start[i]",r=e===1?"end":"end[i]",a=e===1?"outC":"outC[i]",o=vn(e),i=e>1?["coords[0]","coords[1]","coords[2]","coords[3]"].slice(0,e):"coords";return`
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let start = ${o}(${t});
|
|
let end = ${o}(${n});
|
|
var outC = getCoordsFromFlatIndex(index);
|
|
for (var i = 0; i < ${e}; i = i + 1) {
|
|
if (${a} < ${s}) {
|
|
${a} = ${s} * 2 - ${a} - ${this.offset};
|
|
} elseif(${a} >= ${r}) {
|
|
${a} = (${r} - 1) * 2 - ${a} + ${this.offset};
|
|
}
|
|
}
|
|
let coords = outC - start;
|
|
setOutputFlat(index, getX(${i}));
|
|
}
|
|
}
|
|
`}},Ede={kernelName:Ja,backendName:"webgpu",kernelFunc:({inputs:e,attrs:t,backend:n})=>{let{x:s}=e,{paddings:r,mode:a}=t,o=n,i=r.map(u=>({type:"int32",data:[u[0],u[1]]})),l=new Nde(s.shape,r,a);return o.runWebGPUProgram(l,[s],s.dtype,i)}};function Rde(e){let{inputs:t,backend:n}=e,{x:s}=t;if(n.shouldExecuteOnCPU([s])){let a=n.tensorMap.get(s.dataId),[o,i]=rue(a.values,s.shape,s.dtype);return n.makeTensorInfo(i,s.dtype,o)}let r=new Km(s.shape,xt.NEG);return n.runWebGPUProgram(r,[s],s.dtype)}var $de={kernelName:Ti,backendName:"webgpu",kernelFunc:Rde};function _de(e){console.warn("tf.nonMaxSuppression() in webgpu locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");let{inputs:t,backend:n,attrs:s}=e,{boxes:r,scores:a}=t,{maxOutputSize:o,iouThreshold:i,scoreThreshold:l}=s,c=n.readSync(r.dataId),u=n.readSync(a.dataId),{selectedIndices:d}=Xs.nonMaxSuppressionV3Impl(c,u,o,i,l);return n.makeTensorInfo([d.length],"int32",new Int32Array(d))}var Dde={kernelName:Ei,backendName:"webgpu",kernelFunc:_de};function Pde(e){console.warn("tf.nonMaxSuppression() in webgpu locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");let{inputs:t,backend:n,attrs:s}=e,{boxes:r,scores:a}=t,{maxOutputSize:o,iouThreshold:i,scoreThreshold:l,softNmsSigma:c}=s,u=n.readSync(r.dataId),d=n.readSync(a.dataId),p=o,h=i,f=l,m=c,{selectedIndices:g,selectedScores:A}=Xs.nonMaxSuppressionV5Impl(u,d,p,h,f,m);return[n.makeTensorInfo([g.length],"int32",new Int32Array(g)),n.makeTensorInfo([A.length],"float32",new Float32Array(A))]}var Fde={kernelName:Ri,backendName:"webgpu",kernelFunc:Pde};function Ym(e){let{inputs:t,backend:n}=e,{x:s}=t;if(s.dtype==="complex64"){let r=yp({inputs:{input:s},backend:n}),a=Ym({inputs:{x:r},backend:n}),o=Zm({inputs:{input:s},backend:n}),i=Ym({inputs:{x:o},backend:n}),l=fc({inputs:{real:a,imag:i},backend:n});return n.disposeData(r.dataId),n.disposeData(a.dataId),n.disposeData(o.dataId),n.disposeData(i.dataId),l}else return gc({attrs:{shape:s.shape,dtype:s.dtype,value:s.dtype==="string"?"":0},backend:n})}var Ode={kernelName:Ki,backendName:"webgpu",kernelFunc:Ym};function TC(e){let{inputs:t,backend:n}=e,{x:s}=t;if(s.dtype==="string")throw new Error("onesLike is not supported under string dtype");if(s.dtype==="complex64"){let r=yp({inputs:{input:s},backend:n}),a=TC({inputs:{x:r},backend:n}),o=Zm({inputs:{input:s},backend:n}),i=Ym({inputs:{x:o},backend:n}),l=fc({inputs:{real:a,imag:i},backend:n});return n.disposeData(r.dataId),n.disposeData(a.dataId),n.disposeData(o.dataId),n.disposeData(i.dataId),l}else return gc({attrs:{shape:s.shape,dtype:s.dtype,value:1},backend:n})}var Mde={kernelName:$i,backendName:"webgpu",kernelFunc:TC};function zde(e){let{inputs:t,backend:n,attrs:s}=e,{axis:r}=s;if(t.length===1)return Ix({inputs:{input:t[0]},backend:n,attrs:{dim:r}});let a=t[0].shape,o=t[0].dtype;t.forEach(u=>{v.assertShapesMatch(a,u.shape,"All tensors passed to stack must have matching shapes"),v.assert(o===u.dtype,()=>"All tensors passed to stack must have matching dtypes")});let i=[],l=t.map(u=>{let d=Ix({inputs:{input:u},backend:n,attrs:{dim:r}});return i.push(d),d}),c=hC({inputs:l,backend:n,attrs:{axis:r}});return i.forEach(u=>n.disposeData(u.dataId)),c}var Lde={kernelName:Di,backendName:"webgpu",kernelFunc:zde},Bde=class{constructor(e,t){this.variableNames=["x"],this.uniforms="constantValue : f32;",this.workGroupSize=[64,1,1],this.size=!0,this.outputShape=t.map((n,s)=>n[0]+e[s]+n[1]),this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),t.map((n,s)=>{this.uniforms+=` pad${s} : vec2<i32>;`}),this.xShape=e,this.shaderKey="pad"}getUserCode(){let e=this.xShape.length,t=vn(e),n=this.xShape.map((u,d)=>`uniforms.pad${d}[0]`).join(","),s=this.xShape.map((u,d)=>`uniforms.pad${d}[0] + uniforms.xShape${e>1?`[${d}]`:""}`).join(","),r=e>1?`${t}(${n})`:`${n}`,a=e>1?`${t}(${s})`:`${s}`,o=e>1?"any(outC < start)":"outC < start",i=e>1?"any(outC >= end)":"outC >= end",l=e>1?["coords[0]","coords[1]","coords[2]","coords[3]"].slice(0,e):"coords";return`
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let start = ${r};
|
|
let end = ${a};
|
|
let outC = getCoordsFromFlatIndex(index);
|
|
|
|
if (${o} || ${i}) {
|
|
setOutputFlat(index, uniforms.constantValue);
|
|
} else {
|
|
let coords = outC - start;
|
|
setOutputFlat(index, getX(${l}));
|
|
}
|
|
}
|
|
}
|
|
`}},NC=e=>{let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{paddings:a,constantValue:o}=s;if(a.every(c=>v.arraysEqual(c,[0,0])))return Js({inputs:{x:r},backend:n});if(v.sizeFromShape(r.shape)===0){let c=a.map((u,d)=>u[0]+r.shape[d]+u[1]);return gc({backend:n,attrs:{shape:c,value:o,dtype:r.dtype}})}let i=[{type:"float32",data:[o]}];a.map(c=>i.push({type:"int32",data:[c[0],c[1]]}));let l=new Bde(r.shape,a);return n.runWebGPUProgram(l,[r],r.dtype,i)},Wde={kernelName:eo,backendName:"webgpu",kernelFunc:NC},Vde=jn({opSnippet:Wt.POW}),Ude={kernelName:to,backendName:"webgpu",kernelFunc:Vde};function Gde(e){let{inputs:t,backend:n}=e,{x:s,alpha:r}=t,a=new iC(Wt.PRELU,s.shape,r.shape);return n.runWebGPUProgram(a,[s,r],"float32")}var Hde={kernelName:no,backendName:"webgpu",kernelFunc:Gde};function jde(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a,keepDims:o}=s;return xp(r,a,o,"prod",n)}var qde={kernelName:Pi,backendName:"webgpu",kernelFunc:jde},Xde=e=>{let{backend:t,attrs:n}=e,{start:s,stop:r,step:a,dtype:o}=n,i=iue(s,r,a,o);return t.makeTensorInfo([i.length],o,i)},Kde={kernelName:xu,backendName:"webgpu",kernelFunc:Xde},EC=jn({opSnippet:Wt.DIV}),Zde={kernelName:Ma,backendName:"webgpu",kernelFunc:EC},Yde=Tn({opType:xt.RELU}),Jde={kernelName:so,backendName:"webgpu",kernelFunc:Yde},Qde=Tn({opType:xt.RELU6}),epe={kernelName:ao,backendName:"webgpu",kernelFunc:Qde},tpe=class{constructor(e,t,n,s,r){this.variableNames=["x"],this.workGroupSize=[64,1,1],this.size=!0,this.outputShape=[e[0],t,n,e[3]],this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.alignCorners=s,this.halfPixelCenters=r,this.shaderKey=`resizeBilinear_${s}_${r}_${this.outputShape[1]>1}_${this.outputShape[2]>1}`}getUserCode(){let e=this.alignCorners&&this.outputShape[1]>1,t=this.alignCorners&&this.outputShape[2]>1;return`
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let coords = getCoordsFromFlatIndex(index);
|
|
let b = coords[0];
|
|
let d = coords[3];
|
|
let rc = coords.yz;
|
|
|
|
let effectiveInSize = vec2<f32>(
|
|
${e?"f32(uniforms.xShape.y) - 1.0":"f32(uniforms.xShape.y)"},
|
|
${t?"f32(uniforms.xShape.z) - 1.0":"f32(uniforms.xShape.z)"});
|
|
|
|
let effectiveOutSize = vec2<f32>(
|
|
${e?"f32(uniforms.outShape.y) - 1.0":"f32(uniforms.outShape.y)"},
|
|
${t?"f32(uniforms.outShape.z) - 1.0":"f32(uniforms.outShape.z)"});
|
|
|
|
let effectiveInputOverOutputRatioRC =
|
|
effectiveInSize / effectiveOutSize;
|
|
|
|
// Fractional source index
|
|
let sourceFracIndexRC = ${this.halfPixelCenters?"(vec2<f32>(rc) + vec2<f32>(0.5)) * effectiveInputOverOutputRatioRC - vec2<f32>(0.5)":"vec2<f32>(rc) * effectiveInputOverOutputRatioRC"};
|
|
|
|
// Compute the four integer indices.
|
|
let sourceFloorRC = vec2<i32>(sourceFracIndexRC);
|
|
let sourceCeilRC = vec2<i32>(
|
|
min(vec2<f32>(uniforms.xShape.yz) - vec2<f32>(1.0), ceil(sourceFracIndexRC)));
|
|
|
|
let topLeft = getX(b, sourceFloorRC.x, sourceFloorRC.y, d);
|
|
let bottomLeft = getX(b, sourceCeilRC.x, sourceFloorRC.y, d);
|
|
let topRight = getX(b, sourceFloorRC.x, sourceCeilRC.y, d);
|
|
let bottomRight = getX(b, sourceCeilRC.x, sourceCeilRC.y, d);
|
|
|
|
let fracRC = sourceFracIndexRC - vec2<f32>(sourceFloorRC);
|
|
|
|
let top = topLeft + (topRight - topLeft) * fracRC.y;
|
|
let bottom = bottomLeft + (bottomRight - bottomLeft) * fracRC.y;
|
|
let newValue = top + (bottom - top) * fracRC.x;
|
|
|
|
setOutputFlat(index, newValue);
|
|
}
|
|
}
|
|
`}};function npe(e){let{inputs:t,backend:n,attrs:s}=e,{images:r}=t,{alignCorners:a,size:o,halfPixelCenters:i}=s,[l,c]=o,u=new tpe(r.shape,l,c,a,i);return n.runWebGPUProgram(u,[r],"float32")}var spe={kernelName:ro,backendName:"webgpu",kernelFunc:npe},rpe=class{constructor(e,t,n,s,r){this.variableNames=["x"],this.workGroupSize=[64,1,1],this.size=!0,this.outputShape=[e[0],t,n,e[3]],this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.alignCorners=s,this.halfPixelCenters=r,this.shaderKey=`resizeNearest_${s}_${this.outputShape[1]>1}_${this.outputShape[2]>1}_${r}`}getUserCode(){let e=this.alignCorners?"0.5":"0.0",t;this.halfPixelCenters?t="max((vec2<f32>(rc) + vec2<f32>(0.5)) * effectiveInputOverOutputRatioRC, vec2<f32>(0.0))":t="vec2<f32>(rc) * effectiveInputOverOutputRatioRC";let n=this.alignCorners&&this.outputShape[1]>1,s=this.alignCorners&&this.outputShape[2]>1;return`
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let coords = getCoordsFromFlatIndex(index);
|
|
let b = coords[0];
|
|
let d = coords[3];
|
|
let rc = coords.yz;
|
|
|
|
let effectiveInSize = vec2<f32>(
|
|
${n?"f32(uniforms.xShape.y) - 1.0":"f32(uniforms.xShape.y)"},
|
|
${s?"f32(uniforms.xShape.z) - 1.0":"f32(uniforms.xShape.z)"});
|
|
|
|
let effectiveOutSize = vec2<f32>(
|
|
${n?"f32(uniforms.outShape.y) - 1.0":"f32(uniforms.outShape.y)"},
|
|
${s?"f32(uniforms.outShape.z) - 1.0":"f32(uniforms.outShape.z)"});
|
|
|
|
let effectiveInputOverOutputRatioRC =
|
|
effectiveInSize / effectiveOutSize;
|
|
|
|
// Fractional source index
|
|
let sourceFracIndexRC = ${t};
|
|
|
|
// Compute the coordinators of nearest neighbor point.
|
|
let inputShapeRC = vec2<f32>(f32(uniforms.xShape.y), f32(uniforms.xShape.z));
|
|
let sourceNearestRC = vec2<i32>(
|
|
min(inputShapeRC - 1.0, floor(sourceFracIndexRC + ${e})));
|
|
let newValue = getX(b, sourceNearestRC.x, sourceNearestRC.y, d);
|
|
|
|
setOutputFlat(index, newValue);
|
|
}
|
|
}
|
|
`}};function ape(e){let{inputs:t,backend:n,attrs:s}=e,{images:r}=t,{alignCorners:a,halfPixelCenters:o,size:i}=s,[l,c]=i,u=new rpe(r.shape,l,c,a,o);return n.runWebGPUProgram(u,[r],r.dtype)}var ope={kernelName:vu,backendName:"webgpu",kernelFunc:ape},ipe=class{constructor(e,t){this.outputShape=[],this.variableNames=["x"],this.workGroupSize=[64,1,1],this.size=!0,this.outputShape=e,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.uniforms=`centerX : f32; centerY : f32; sinRadians : f32;
|
|
cosRadians : f32;`,this.shaderKey="rotate",this.outputShape=e,typeof t=="number"?(this.uniforms+=" fillValue : f32;",this.fillSnippet="var outputValue = uniforms.fillValue;",this.shaderKey+="_float"):(this.uniforms+=" fillValue : vec3<f32>;",this.fillSnippet="var outputValue = uniforms.fillValue[coords[3]];",this.shaderKey+="_vec3")}getUserCode(){return`
|
|
${tt()}
|
|
|
|
if (index < uniforms.size) {
|
|
let coords = getCoordsFromFlatIndex(index);
|
|
let coordXFloat = (f32(coords[2]) - uniforms.centerX) *
|
|
uniforms.cosRadians - (f32(coords[1]) - uniforms.centerY) *
|
|
uniforms.sinRadians;
|
|
let coordYFloat = (f32(coords[2]) - uniforms.centerX) *
|
|
uniforms.sinRadians + (f32(coords[1]) - uniforms.centerY) *
|
|
uniforms.cosRadians;
|
|
let coordX = i32(round(coordXFloat + uniforms.centerX));
|
|
let coordY = i32(round(coordYFloat + uniforms.centerY));
|
|
${this.fillSnippet}
|
|
if(coordX >= 0 && coordX < uniforms.xShape[2] && coordY >= 0 &&
|
|
coordY < uniforms.xShape[1]) {
|
|
outputValue = getX(coords[0], coordY, coordX, coords[3]);
|
|
}
|
|
setOutputFlat(index, outputValue);
|
|
}
|
|
}
|
|
`}},lpe={kernelName:Zi,backendName:"webgpu",kernelFunc:({inputs:e,attrs:t,backend:n})=>{let{image:s}=e,{radians:r,fillValue:a,center:o}=t,i=n,l=new ipe(s.shape,a),[c,u]=E.getImageCenter(o,s.shape[1],s.shape[2]),d=[{type:"float32",data:[c]},{type:"float32",data:[u]},{type:"float32",data:[Math.sin(r)]},{type:"float32",data:[Math.cos(r)]}];return typeof a=="number"?d.push({type:"float32",data:[Number.parseFloat(a.toFixed(2))]}):d.push({type:"float32",data:a}),i.runWebGPUProgram(l,[s],s.dtype,d)}},upe=Tn({opType:xt.RSQRT,cpuKernelImpl:lue}),cpe={kernelName:oo,backendName:"webgpu",kernelFunc:upe},dpe=class{constructor(e,t,n,s,r,a,o){this.variableNames=["updates","indices"],this.workGroupSize=[64,1,1],this.atomic=!0,this.outputShape=a,this.type=o,this.dispatchLayout=je(e),this.dispatch=Fe(this.dispatchLayout,e,this.workGroupSize),this.sliceDimGreaterThanOne=t>1,this.shaderKey=`scatter_${n}_${s}_${this.sliceDimGreaterThanOne}_${o}`;let i=vn(r.length);this.uniforms=`sliceDim : i32; strides: ${i}; size: i32;`,this.updatesRank=s,this.indicesRank=n}getUserCode(){let e="";this.indicesRank===1?e="coords[0]":this.indicesRank===2&&(e="coords[0], j");let t=`getIndices(${e})`,n=this.sliceDimGreaterThanOne?"uniforms.strides[j]":"uniforms.strides",s="",r="",a="";this.updatesRank===1?(s="coords[0]",r="flattenedIndex",a=`
|
|
fn getUpdatesCoordsFromFlatIndex(index : i32) -> i32 {
|
|
return index;
|
|
}
|
|
`):this.updatesRank===2&&(s="coords[0], coords[1]",r="vec2<i32>(flattenedIndex, coords[1])",a=`
|
|
fn getUpdatesCoordsFromFlatIndex(index : i32) -> vec2<i32> {
|
|
let d0 = index / uniforms.updatesShape[1];
|
|
let d1 = index - d0 * uniforms.updatesShape[1];
|
|
return vec2<i32>(d0, d1);
|
|
}
|
|
`);let o=`getUpdates(${s})`,i=this.type==="int32"?"ignore(atomicAdd(&(result.numbers[flatIndex]), i32(updateValue)));":`
|
|
var assumed = atomicLoad(&(result.numbers[flatIndex]));
|
|
var success = 0;
|
|
for (; success == 0;) {
|
|
let new = bitcast<f32>(assumed) + updateValue;
|
|
let newI32 = bitcast<i32>(new);
|
|
let resValue = atomicCompareExchangeWeak(&(result.numbers[flatIndex]), assumed, newI32);
|
|
assumed = resValue[0];
|
|
success = resValue[1];
|
|
}
|
|
`;return`
|
|
${a}
|
|
|
|
${tt()}
|
|
|
|
if (index < uniforms.size) {
|
|
let coords = getUpdatesCoordsFromFlatIndex(index);
|
|
var flattenedIndex = 0;
|
|
for (var j = 0; j < uniforms.sliceDim; j = j + 1) {
|
|
let indexInside = i32(round(${t}));
|
|
flattenedIndex = flattenedIndex + indexInside * ${n};
|
|
}
|
|
let updateValue = ${o};
|
|
let flatIndex = getOutputFlatIndex(${r});
|
|
|
|
${i}
|
|
}
|
|
}`}};function ppe(e){let{inputs:t,backend:n,attrs:s}=e,{indices:r,updates:a}=t,{shape:o}=s,{sliceRank:i,numUpdates:l,sliceSize:c,strides:u,outputSize:d}=E.calculateShapes(a,r,o),p=[d/c,c];if(d===0)return n.makeTensorInfo(o,r.dtype);let h=Ge({inputs:{x:r},backend:n,attrs:{shape:[l,i]}}),f=Ge({inputs:{x:a},backend:n,attrs:{shape:[l,c]}}),m=f.dtype,g=gc({backend:n,attrs:{shape:p,value:0,dtype:m}}),A=v.sizeFromShape(f.shape),x=[{type:"int32",data:[i]},{type:"int32",data:u},{type:"int32",data:[A]}],y=new dpe(f.shape,i,h.shape.length,f.shape.length,u,p,m),b=n.runWebGPUProgram(y,[f,h],m,x,g),w=Ge({inputs:{x:b},backend:n,attrs:{shape:o}});return n.disposeData(h.dataId),n.disposeData(f.dataId),n.disposeData(b.dataId),w}var hpe={kernelName:zi,backendName:"webgpu",kernelFunc:ppe},fpe=class{constructor(e,t,n){this.variableNames=["c","a","b"],this.workGroupSize=[64,1,1],this.size=!0,this.outputShape=t,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.cRank=e,this.rank=n,this.shaderKey="select"}getUserCode(){let e,t;if(this.rank>4)throw Error(`Where for rank ${this.rank} is not yet supported`);if(this.rank===1)t="resRC",e="resRC";else{let s=["resRC.x","resRC.y","resRC.z","resRC.w"],r=[],a=[];for(let o=0;o<this.outputShape.length;o++)a.push(`${s[o]}`),o<this.cRank&&r.push(`${s[o]}`);e=r.join(),t=a.join()}return`
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let resRC = getCoordsFromFlatIndex(index);
|
|
let cVal = getC(${e});
|
|
if (cVal >= 1.0) {
|
|
setOutputFlat(index, getA(${t}));
|
|
} else {
|
|
setOutputFlat(index, getB(${t}));
|
|
}
|
|
}
|
|
}
|
|
`}};function mpe(e){let{inputs:t,backend:n}=e,{condition:s,t:r,e:a}=t,o=new fpe(s.shape.length,r.shape,r.shape.length);return n.runWebGPUProgram(o,[s,r,a],zn(r.dtype,a.dtype))}var gpe={kernelName:Li,backendName:"webgpu",kernelFunc:mpe},Ape=Tn({opType:xt.SIGMOID}),ype={kernelName:lo,backendName:"webgpu",kernelFunc:Ape},xpe=Tn({opType:xt.SIN}),bpe={kernelName:io,backendName:"webgpu",kernelFunc:xpe},vpe=Tn({opType:xt.SINH}),wpe={kernelName:Wi,backendName:"webgpu",kernelFunc:vpe},RC=jn({opSnippet:Wt.SUB,cpuKernelImpl:hue,supportsComplex:!0}),kpe={kernelName:fo,backendName:"webgpu",kernelFunc:RC};function Spe(e){let{inputs:t,backend:n,attrs:s}=e,{logits:r}=t,{dim:a}=s,o=v.parseAxisParam([a],r.shape),i=CC({inputs:{x:r},backend:n,attrs:{reductionIndices:o,keepDims:!1}}),l=E.expandShapeToKeepDim(i.shape,o),c=Ge({inputs:{x:i},backend:n,attrs:{shape:l}}),u=RC({inputs:{a:r,b:c},backend:n}),d=vC({inputs:{x:u},backend:n}),p=Sx({inputs:{x:d},backend:n,attrs:{axis:o,keepDims:!1}}),h=Ge({inputs:{x:p},backend:n,attrs:{shape:l}}),f=EC({inputs:{a:d,b:h},backend:n});return n.disposeData(i.dataId),n.disposeData(c.dataId),n.disposeData(u.dataId),n.disposeData(d.dataId),n.disposeData(p.dataId),n.disposeData(h.dataId),f}var Ipe={kernelName:po,backendName:"webgpu",kernelFunc:Spe},Cpe=e=>{let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{blockShape:a,paddings:o}=s;v.assert(r.shape.length<=4,()=>"spaceToBatchND for rank > 4 with a WebGPU backend not implemented yet");let i=a.reduce((A,x)=>A*x),l=[[0,0]];l.push(...o);for(let A=1+a.length;A<r.shape.length;++A)l.push([0,0]);let c=[],u=NC({inputs:{x:r},backend:n,attrs:{paddings:l,constantValue:0}}),d=E.getReshaped(u.shape,a,i,!1),p=E.getPermuted(d.length,a.length,!1),h=E.getReshapedPermuted(u.shape,a,i,!1),f=Ge({inputs:{x:u},backend:n,attrs:{shape:d}}),m=Rl({inputs:{x:f},backend:n,attrs:{perm:p}}),g=Ge({inputs:{x:m},backend:n,attrs:{shape:h}});return c.push(u),c.push(f),c.push(m),c.forEach(A=>n.disposeData(A.dataId)),g},Tpe={kernelName:Vi,backendName:"webgpu",kernelFunc:Cpe},Npe=class{constructor(e,t,n,s,r,a,o=!0){this.variableNames=["updates","indices","defaultValue"],this.workGroupSize=[64,1,1],this.workPerThread=4,this.size=!0,this.outputShape=a,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,[this.workPerThread,1,1]);let i=t>1;this.shaderKey=`scatter_${n}_${s}_${i}`;let l=vn(r.length);this.uniforms=`updateSize : i32; sliceDim : i32; strides: ${l};`;let c="";n===1?c="i":n===2&&(c="i, j"),this.indicesSnippet=`getIndices(${c})`;let u="";s===1?u="i":s===2&&(u="i, coords[1]"),this.updatesSnippet=`getUpdates(${u})`,this.strideString=i?"uniforms.strides[j]":"uniforms.strides"}getUserCode(){return`
|
|
${tt()}
|
|
|
|
let globalIndex = index * ${this.workPerThread};
|
|
if (globalIndex < uniforms.size) {
|
|
var sum = vec4<f32>(0.0);
|
|
var found = vec4<bool>(false);
|
|
for (var i = 0; i < uniforms.updateSize; i = i + 1) {
|
|
var flattenedIndex = 0;
|
|
for (var j = 0; j < uniforms.sliceDim; j = j + 1) {
|
|
let indexInside = i32(round(${this.indicesSnippet}));
|
|
flattenedIndex = flattenedIndex + indexInside * ${this.strideString};
|
|
}
|
|
for (var innerIndex = 0; innerIndex < ${this.workPerThread}; innerIndex = innerIndex + 1) {
|
|
let curIndex = globalIndex + innerIndex;
|
|
let coords = getCoordsFromFlatIndex(curIndex);
|
|
if (flattenedIndex == coords[0]) {
|
|
sum[innerIndex] = sum[innerIndex] + ${this.updatesSnippet};
|
|
found[innerIndex] = true;
|
|
}
|
|
}
|
|
}
|
|
for (var innerIndex = 0; innerIndex < ${this.workPerThread}; innerIndex = innerIndex + 1) {
|
|
let curIndex = globalIndex + innerIndex;
|
|
if (curIndex < uniforms.size)
|
|
{
|
|
setOutputFlat(curIndex, mix(getDefaultValue(), sum[innerIndex], f32(found[innerIndex])));
|
|
}
|
|
}
|
|
}
|
|
}`}};function Epe(e){let{inputs:t,backend:n,attrs:s}=e,{sparseIndices:r,sparseValues:a,defaultValue:o}=t,{outputShape:i}=s,{sliceRank:l,numUpdates:c,strides:u,outputSize:d}=E.calculateShapes(a,r,i),p=!1,h=[{type:"int32",data:[c]},{type:"int32",data:[l]},{type:"int32",data:u}],f=new Npe(c,l,r.shape.length,a.shape.length,u,[d,1],p),m=n.runWebGPUProgram(f,[a,r,o],a.dtype,h),g=Ge({inputs:{x:m},backend:n,attrs:{shape:i}});return n.disposeData(m.dataId),g}var Rpe={kernelName:ld,backendName:"webgpu",kernelFunc:Epe};function $pe(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{numOrSizeSplits:a,axis:o}=s,i=v.parseAxisParam(o,r.shape)[0],l=E.prepareSplitSize(r,a,i),c=r.shape.length,u=new Array(c).fill(0),d=r.shape.slice();return l.map(p=>{let h=[...d];h[i]=p;let f=mc({inputs:{x:r},backend:n,attrs:{begin:u,size:h}});return u[i]+=p,f})}var _pe={kernelName:Ui,backendName:"webgpu",kernelFunc:$pe},Dpe=Tn({opType:xt.SQRT}),Ppe={kernelName:uo,backendName:"webgpu",kernelFunc:Dpe},Fpe={kernelName:Iu,backendName:"webgpu",kernelFunc:({inputs:e,backend:t})=>{let{x:n}=e,s=t,r=new Km(n.shape,xt.SQUARE);return s.runWebGPUProgram(r,[n],n.dtype)}},Ope=jn({opSnippet:Wt.SQUARED_DIFFERENCE}),Mpe={kernelName:ho,backendName:"webgpu",kernelFunc:Ope},zpe=class{constructor(e){this.variableNames=["x"],this.workPerThread=1,this.workGroupSize=[64,1,1],this.size=!0,this.outputShape=e,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,[this.workPerThread,1,1]);let t=vn(this.outputShape.length);this.uniforms=`begin : ${t}; strides : ${t}; `,this.shaderKey="stridedSlice"}getUserCode(){let e=this.outputShape.length,t="";if(e===1)t="coords * uniforms.strides + uniforms.begin";else{let s=0;t=this.outputShape.map((r,a)=>(s++,this.outputShape.length===1?`coords * uniforms.strides[${a}] + uniforms.begin[${a}]`:`coords[${s-1}] * uniforms.strides[${a}] + uniforms.begin[${a}]`)).join(",")}return`
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let coords = getCoordsFromFlatIndex(index);
|
|
setOutputFlat(index, getX(${t}));
|
|
}
|
|
}
|
|
`}};function Lpe(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{begin:a,end:o,strides:i,beginMask:l,endMask:c,ellipsisMask:u,newAxisMask:d,shrinkAxisMask:p}=s,{finalShapeSparse:h,finalShape:f,isIdentity:m,sliceDim0:g,isSimpleSlice:A,begin:x,end:y,strides:b}=Pt.sliceInfo(r.shape,a,o,i,l,c,u,d,p),w;if(m)w=Ge({inputs:{x:r},backend:n,attrs:{shape:f}});else if(g||A){v.assert(r.shape.length>=1,()=>`Input must have rank at least 1, got: ${r.shape.length}`);let k=Pt.computeOutShape(x,y,b),C=mc({inputs:{x:r},backend:n,attrs:{begin:x,size:k}});w=Ge({inputs:{x:C},backend:n,attrs:{shape:f}}),n.disposeData(C.dataId)}else if(n.shouldExecuteOnCPU([r])){let C=n.readSync(r.dataId),N=ze(r.shape,r.dtype,C),R=due(h,N,b,x);w=n.makeTensorInfo(f,r.dtype,R.values)}else{let C=new zpe(h),N=[{type:"int32",data:x},{type:"int32",data:b}],R=n.runWebGPUProgram(C,[r],r.dtype,N);w=Ge({inputs:{x:R},backend:n,attrs:{shape:f}}),n.disposeData(R.dataId)}return w}var Bpe={kernelName:Gi,backendName:"webgpu",kernelFunc:Lpe};function Wpe(e){let{inputs:t,backend:n,attrs:s}=e,{separator:r,nGramWidths:a,leftPad:o,rightPad:i,padWidth:l,preserveShortSequences:c}=s,{data:u,dataSplits:d}=t,p=n.readSync(u.dataId),h=n.readSync(d.dataId),[f,m]=pue(p,h,r,a,o,i,l,c);return[n.makeTensorInfo([f.length],"string",f),n.makeTensorInfo(d.shape,"int32",m)]}var Vpe={kernelName:ud,backendName:"webgpu",kernelFunc:Wpe},Upe=Tn({opType:xt.TANH}),Gpe={kernelName:mo,backendName:"webgpu",kernelFunc:Upe},Hpe=class{constructor(e,t){this.variableNames=["A"],this.workGroupSize=[64,1,1],this.size=!0;let n=new Array(e.length);for(let s=0;s<n.length;s++)n[s]=e[s]*t[s];this.outputShape=n,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.rank=this.outputShape.length,this.shaderKey="tile"}getUserCode(){let e=jpe(this.rank,"uniforms.");return`
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let resRC = getCoordsFromFlatIndex(index);
|
|
setOutputFlat(index, getA(${e}));
|
|
}
|
|
}
|
|
`}};function jpe(e,t=""){if(e>=5)throw Error(`Tile for rank ${e} is not yet supported`);if(e===1)return`(resRC % ${t}aShape)`;let n=["resRC.x","resRC.y","resRC.z","resRC.w"],s=[];for(let r=0;r<e;r++)s.push(`(${n[r]} % ${t}aShape[${r}])`);return s.join()}function qpe(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{reps:a}=s;if(n.shouldExecuteOnCPU([r])||r.dtype==="string"||r.shape.length>=5){let l=n.readSync(r.dataId),c=r.dtype==="string"?l.map(p=>v.decodeString(p)):l,u=ze(r.shape,r.dtype,c),d=fue(u,a);return n.makeTensorInfo(d.shape,d.dtype,d.values)}let o=new Hpe(r.shape,a);return n.runWebGPUProgram(o,[r],r.dtype)}var Xpe={kernelName:jr,backendName:"webgpu",kernelFunc:qpe},Kpe=class{constructor(e){this.variableNames=["x","indices"],this.workGroupSize=[256,1,1],this.size=!0,this.outputShape=e,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.uniforms=`inputSize : i32; firstPass : i32; negativeInf : f32;
|
|
dir : i32; inc : i32;`,this.shaderKey="swap"}getUserCode(){return`
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let outC = getCoordsFromFlatIndex(index);
|
|
let batch = outC[0];
|
|
let elemIdx = outC[1];
|
|
// We compare elements pair-wise within a group of size 2 * inc.
|
|
// The comparing rule for each group alternates between ascending
|
|
// and descending. Within each group, we compare each pair at
|
|
// positions i and i+inc. To decide whether an element at position i
|
|
// is x0 or x1, we mod it by 2 * inc, if the result is smaller than
|
|
// inc, it is in the first half of the group, we denote it as x0,
|
|
// otherwise we denote it as x1.
|
|
// For example, as shown in the Bitonic top K paper referenced
|
|
// above, Figure5(a) shows that element[1] is in the second half of
|
|
// the group when group size is 2, but it is in the first half of
|
|
// the group when group size is 4.
|
|
let isFirstInPair = elemIdx % (2 * uniforms.inc) < uniforms.inc;
|
|
var i = 0;
|
|
if (isFirstInPair) {
|
|
i = elemIdx;
|
|
} else {
|
|
i = elemIdx - uniforms.inc;
|
|
}
|
|
|
|
var i0 = 0;
|
|
if (uniforms.firstPass == 1) {
|
|
i0 = i;
|
|
} else {
|
|
i0 = i32(getIndices(batch, i));
|
|
}
|
|
|
|
var i1 = 0;
|
|
if (uniforms.firstPass == 1) {
|
|
i1 = i + uniforms.inc;
|
|
} else {
|
|
i1 = i32(getIndices(batch, i + uniforms.inc));
|
|
}
|
|
|
|
var x0 = f32(0.0);
|
|
var x1 = f32(0.0);
|
|
if (i0 < uniforms.inputSize) {
|
|
x0 = getX(batch, i0);
|
|
} else {
|
|
x0 = uniforms.negativeInf;
|
|
}
|
|
if (i1 < uniforms.inputSize) {
|
|
x1 = getX(batch, i1);
|
|
} else {
|
|
x1 = uniforms.negativeInf;
|
|
}
|
|
|
|
let reverse = elemIdx % (2 * uniforms.dir) >= uniforms.dir;
|
|
let isGreater = x0 > x1 || (x0 == x1 && i1 > i0);
|
|
if (reverse == isGreater) {
|
|
// Elements in opposite order of direction
|
|
let iTemp = i0;
|
|
i0 = i1;
|
|
i1 = iTemp;
|
|
}
|
|
if (isFirstInPair) {
|
|
setOutputFlat(index, f32(i0));
|
|
} else {
|
|
setOutputFlat(index, f32(i1));
|
|
}
|
|
}
|
|
}
|
|
`}},Zpe=class{constructor(e){this.variableNames=["x","indices"],this.workGroupSize=[256,1,1],this.size=!0,this.outputShape=e,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.uniforms="inputSize : i32; firstPass : i32; k : i32;",this.shaderKey="merge"}getUserCode(){return`
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let outC = getCoordsFromFlatIndex(index);
|
|
let batch = outC[0];
|
|
let elemIdx = outC[1];
|
|
// The output size is half of the previous size.
|
|
// If the previous sequence is | | | | _ _ _ _ | | | | _ _ _ _
|
|
// (k=4), we only need to output the indices at positions |, the
|
|
// indices at positions _ can be thrown away, see Figure5(b) After
|
|
// Phase 2 (Merge phase) in the Bitonic Top K paper referenced
|
|
// above.
|
|
// For example, the paper shows we only need to output the orange
|
|
// bars. The output sequence should look like this | | | | | | | |.
|
|
// Because the sequence is halved, to map the output index back to
|
|
// the previous sequence to find the corresponding value, we need
|
|
// to double the index. When we double the index, we basically
|
|
// interpolate a position, so 2i looks like
|
|
// | _ | _ | _ | _ | _ | _ | _. We move the | to the first k
|
|
// position of each 2k positions by - elemIdx % k. E.g. for output
|
|
// at index 4,5,6,7, we want to get the corresponding element at
|
|
// original index 8,9,10,11, for output at index 8,9,10,11,
|
|
// we want to get the corresponding element at original index
|
|
// 16,17,18,19, so on and so forth.
|
|
|
|
var i = 0;
|
|
if (elemIdx < uniforms.k) {
|
|
i = elemIdx;
|
|
} else {
|
|
i = elemIdx * 2 - elemIdx % uniforms.k;
|
|
}
|
|
var i0 = 0;
|
|
if (uniforms.firstPass == 1) {
|
|
i0 = i;
|
|
} else {
|
|
i0 = i32(getIndices(batch, i));
|
|
}
|
|
var i1 = 0;
|
|
if (uniforms.firstPass == 1) {
|
|
i1 = i + uniforms.k;
|
|
} else {
|
|
i1 = i32(getIndices(batch, i + uniforms.k));
|
|
}
|
|
|
|
let x0 = getX(batch, i0);
|
|
var x1 = f32(0.0);
|
|
if (i1 < uniforms.inputSize) {
|
|
x1 = getX(batch, i1);
|
|
} else {
|
|
x1 = x0;
|
|
}
|
|
|
|
if (x0 >= x1) {
|
|
setOutputFlat(index, f32(i0));
|
|
} else {
|
|
setOutputFlat(index, f32(i1));
|
|
}
|
|
}
|
|
}
|
|
`}};function yc(e,t){t!==null&&e.disposeData(t.dataId)}function $C(e){let t=1;for(;t<e;)t*=2;return t}function Ype(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{k:a,sorted:o}=s,i=r.shape,l=i[i.length-1];if(n.shouldExecuteOnCPU([r])){let w=n.readSync(r.dataId),[k,C]=mue(w,i,r.dtype,a,o);return[n.makeTensorInfo(k.shape,k.dtype,k.values),n.makeTensorInfo(C.shape,C.dtype,C.values)]}if(a===0)return i[i.length-1]=0,[n.makeTensorInfo(i,r.dtype,[]),n.makeTensorInfo(i,"int32",[])];if(l===1)return[r,gc({attrs:{shape:i,dtype:"int32",value:0},backend:n})];let u=v.sizeFromShape(i)/l,d=Ge({inputs:{x:r},attrs:{shape:[u,l]},backend:n}),p=$C(a),h=$C(l),f=null,m=()=>f===null?[d,d]:[d,f],g=(w,k,C)=>{let N=m(),R=new Kpe(C),_=[{type:"int32",data:[l]},{type:"int32",data:[f===null?1:0]},{type:"float32",data:[Number.NEGATIVE_INFINITY]},{type:"int32",data:[w]},{type:"int32",data:[k]}],P=f;f=n.runWebGPUProgram(R,N,"int32",_),yc(n,P)};for(let w=1;w<p;w*=2){let k=w*2;for(let C=w;C>=1;C/=2)g(k,C,[u,h])}for(let w=h;w>p;w/=2){let k=m(),C=new Zpe([u,w/2]),R=[{type:"int32",data:[l]},{type:"int32",data:[f===null?1:0]},{type:"int32",data:[p]}],F=f;f=n.runWebGPUProgram(C,k,"int32",R),yc(n,F);let _=p/2,P=_*2;for(let T=_;T>=1;T/=2)g(P,T,f.shape)}let A=f;f=mc({inputs:{x:f},backend:n,attrs:{begin:0,size:[u,a]}}),yc(n,A);let x=IC({inputs:{x:d,indices:f},backend:n,attrs:{axis:1,batchDims:1}});yc(n,d);let y=i.slice(0,-1);y.push(a),A=f,f=Ge({inputs:{x:f},attrs:{shape:y},backend:n}),yc(n,A);let b=x;return x=Ge({inputs:{x},attrs:{shape:y},backend:n}),yc(n,b),[x,f]}var Jpe={kernelName:ji,backendName:"webgpu",kernelFunc:Ype},Qpe=class{constructor(e){this.variableNames=["Image","Transforms"],this.uniforms="interpolationModeId : i32; fillModeId : i32; fillValue : f32;",this.workGroupSize=[64,1,1],this.size=!0,this.outputShape=e,this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize),this.shaderKey="transform"}getUserCode(){return`
|
|
fn mapCoord(outCoord : f32, len : f32) -> f32{
|
|
var inCoord = outCoord;
|
|
if(uniforms.fillModeId == 2) {
|
|
if (inCoord < 0.0) {
|
|
if (len <= 1.0) {
|
|
inCoord = 0.0;
|
|
} else {
|
|
let sz2 = 2.0 * len;
|
|
if (inCoord < sz2) {
|
|
inCoord = sz2 * f32(i32(f32(-inCoord / sz2))) +
|
|
inCoord;
|
|
}
|
|
if (inCoord < -len) {
|
|
inCoord = inCoord + sz2;
|
|
} else {
|
|
inCoord = -inCoord - 1.0;
|
|
}
|
|
}
|
|
} elseif (inCoord > len - 1.0) {
|
|
if (len <= 1.0) {
|
|
inCoord = 0.0;
|
|
} else {
|
|
let sz2 = 2.0 * len;
|
|
inCoord = inCoord - sz2 * f32(i32(f32(inCoord / sz2)));
|
|
if (inCoord >= len) {
|
|
inCoord = sz2 - inCoord - 1.0;
|
|
}
|
|
}
|
|
}
|
|
return clamp(inCoord, 0.0, len - 1.0);
|
|
} elseif (uniforms.fillModeId == 3) {
|
|
if (inCoord < 0.0) {
|
|
if (len <= 1.0) {
|
|
inCoord = 0.0;
|
|
} else {
|
|
let sz = len - 1.0;
|
|
inCoord = inCoord + len * (f32(i32(f32(-inCoord / sz))) + 1.0);
|
|
}
|
|
} elseif (inCoord > len - 1.0) {
|
|
if (len <= 1.0) {
|
|
inCoord = 0.0;
|
|
} else {
|
|
let sz = len - 1.0;
|
|
inCoord = inCoord - len * f32(i32(f32(inCoord / sz)));
|
|
}
|
|
}
|
|
return clamp(inCoord, 0.0, len - 1.0);
|
|
} elseif (uniforms.fillModeId == 4) {
|
|
return clamp(outCoord, 0.0, len - 1.0);
|
|
}
|
|
return outCoord;
|
|
}
|
|
fn readWithFillValue(batch : i32, coordY : i32, coordX : i32,
|
|
channel : i32) -> f32 {
|
|
var outputValue : f32;
|
|
if (0 <= coordY && coordY < uniforms.imageShape[1] && 0 <= coordX && coordX < uniforms.imageShape[2]) {
|
|
outputValue = getImage(batch, coordY, coordX, channel);
|
|
} else {
|
|
outputValue = uniforms.fillValue;
|
|
}
|
|
return outputValue;
|
|
}
|
|
|
|
${tt()}
|
|
if (index < uniforms.size) {
|
|
let coords = getCoordsFromFlatIndex(index);
|
|
var outputValue : f32;
|
|
let batch = coords[0];
|
|
let x = coords[2];
|
|
let y = coords[1];
|
|
let channel = coords[3];
|
|
let xf = f32(x);
|
|
let yf = f32(y);
|
|
let a1 = getTransforms(batch, 0);
|
|
let a2 = getTransforms(batch, 1);
|
|
let a3 = getTransforms(batch, 2);
|
|
let b1 = getTransforms(batch, 3);
|
|
let b2 = getTransforms(batch, 4);
|
|
let b3 = getTransforms(batch, 5);
|
|
let c1 = getTransforms(batch, 6);
|
|
let c2 = getTransforms(batch, 7);
|
|
let projection = c1 * xf + c2 * yf + 1.0;
|
|
if (projection == 0.0) {
|
|
outputValue = uniforms.fillValue;
|
|
} else {
|
|
let inX = (a1 * xf + a2 * yf + a3) / projection;
|
|
let inY = (b1 * xf + b2 * yf + b3) / projection;
|
|
let mapX = mapCoord(inX, f32(uniforms.imageShape[2]));
|
|
let mapY = mapCoord(inY, f32(uniforms.imageShape[1]));
|
|
|
|
if (uniforms.interpolationModeId == 1) {
|
|
let coordY = i32(round(mapY));
|
|
let coordX = i32(round(mapX));
|
|
outputValue = readWithFillValue(batch, coordY, coordX,
|
|
channel);
|
|
} else {
|
|
let yFloor = floor(mapY);
|
|
let xFloor = floor(mapX);
|
|
let yCeil = yFloor + 1.0;
|
|
let xCeil = xFloor + 1.0;
|
|
let valueYFloor = (xCeil - mapX) *
|
|
readWithFillValue(batch, i32(yFloor), i32(xFloor), channel) +
|
|
(mapX - xFloor) *
|
|
readWithFillValue(batch, i32(yFloor), i32(xCeil), channel);
|
|
let valueYCeil = (xCeil - mapX) *
|
|
readWithFillValue(batch, i32(yCeil), i32(xFloor), channel) +
|
|
(mapX - xFloor) *
|
|
readWithFillValue(batch, i32(yCeil), i32(xCeil), channel);
|
|
outputValue = (yCeil - mapY) * valueYFloor +
|
|
(mapY - yFloor) * valueYCeil;
|
|
}
|
|
}
|
|
setOutputFlat(index, outputValue);
|
|
}
|
|
}
|
|
`}};function ehe(e){let{inputs:t,backend:n,attrs:s}=e,{image:r,transforms:a}=t,{interpolation:o,fillMode:i,fillValue:l,outputShape:c}=s,[u,d,p,h]=r.shape,[f,m]=c!=null?c:[d,p],g=[u,f,m,h],A=new Qpe(g),x=o==="nearest"?1:2,y;switch(i){case"constant":y=1;break;case"reflect":y=2;break;case"wrap":y=3;break;case"nearest":y=4;break;default:y=1;break}let b=[{type:"int32",data:[x]},{type:"int32",data:[y]},{type:"float32",data:[l]}];return n.runWebGPUProgram(A,[r,a],"float32",b)}var the={kernelName:qi,backendName:"webgpu",kernelFunc:ehe};function nhe(e){let{inputs:t,backend:n,attrs:s}=e,{value:r}=t,{axis:a}=s;a<0&&(a+=r.shape.length);let o=r,i=o.shape.length,l=r.shape[a],c=new Array(i-1),u=0;for(let m=0;m<i;m++)m!==a&&(c[u++]=o.shape[m]);let d=[],p=new Array(i).fill(0),h=o.shape.slice();h[a]=1;let f=new Array(l);for(let m=0;m<f.length;m++){p[a]=m;let g=mc({inputs:{x:o},backend:n,attrs:{begin:p,size:h}}),A=Ge({inputs:{x:g},backend:n,attrs:{shape:c}});f[m]=A,d.push(g)}return d.forEach(m=>n.disposeData(m.dataId)),f}var she={kernelName:Xi,backendName:"webgpu",kernelFunc:nhe},rhe=[Fle,yue,bue,kue,Eue,$ue,Due,Fue,Bue,Gue,jue,Zue,Lle,ece,rce,lce,cce,pce,mce,yce,bce,Ice,Tce,Ece,$ce,Rce,Dce,Fce,Mce,Uce,Lce,Wce,jce,Xce,Zce,Qce,nde,rde,ode,zle,Jue,lde,cde,pde,fde,gde,Ade,xde,vde,kde,Ide,Tde,Ede,vce,$de,Dde,Fde,Wue,Mde,Lde,Wde,Hde,qde,Ude,Kde,Vue,Zde,Jde,epe,Dle,spe,ope,lpe,cpe,hpe,gpe,ype,bpe,wpe,zue,Bpe,Vpe,Ipe,Tpe,_pe,Rpe,Ppe,Fpe,Mpe,kpe,kce,Gpe,Xpe,Jpe,the,Tue,she,Ode];for(let e of rhe)rr(e);var ahe=class{constructor(e){this.device=e,this.numUsedBuffers=0,this.numFreeBuffers=0,this.freeBuffers=new Map,this.usedBuffers=new Map,this.numBytesUsed=0,this.numBytesAllocated=0}acquireBuffer(e,t){let n=_C(e,t);if(this.freeBuffers.has(n)||this.freeBuffers.set(n,[]),this.usedBuffers.has(n)||this.usedBuffers.set(n,[]),this.numBytesUsed+=e,this.numUsedBuffers++,this.freeBuffers.get(n).length>0){this.numFreeBuffers--;let r=this.freeBuffers.get(n).shift();return this.usedBuffers.get(n).push(r),r}this.numBytesAllocated+=e;let s=this.device.createBuffer({size:e,usage:t});return this.usedBuffers.get(n).push(s),s}releaseBuffer(e,t,n){if(this.freeBuffers==null)return;let s=_C(t,n);this.freeBuffers.has(s)||this.freeBuffers.set(s,[]),this.freeBuffers.get(s).push(e),this.numFreeBuffers++,this.numUsedBuffers--;let r=this.usedBuffers.get(s),a=r.indexOf(e);if(a<0)throw new Error("Cannot release a buffer that was never provided by this buffer manager");r.splice(a,1),this.numBytesUsed-=t}getNumUsedBuffers(){return this.numUsedBuffers}getNumFreeBuffers(){return this.numFreeBuffers}reset(){this.freeBuffers=new Map,this.usedBuffers=new Map,this.numUsedBuffers=0,this.numFreeBuffers=0,this.numBytesUsed=0,this.numBytesAllocated=0}dispose(){this.freeBuffers==null&&this.usedBuffers==null||(this.freeBuffers.forEach((e,t)=>{e.forEach(n=>{n.destroy()})}),this.usedBuffers.forEach((e,t)=>{e.forEach(n=>{n.destroy()})}),this.freeBuffers=null,this.usedBuffers=null,this.numUsedBuffers=0,this.numFreeBuffers=0,this.numBytesUsed=0,this.numBytesAllocated=0)}};function _C(e,t){return`${e}_${t}`}var DC=class{constructor(){this.outputShape=[0],this.variableNames=[],this.workGroupSize=[256,1,1],this.lastUniformData=[],this.inputTexture=null,this.layout=null,this.lastPixelSize={width:0,height:0},this.disposed=!1,this.shaderKey="fromPixels",this.useImport=!1}updateOutputShape(e){v.arraysEqual(this.outputShape,e)||(this.outputShape=e,this.workPerThread=e[2],this.dispatchLayout=je(this.outputShape),this.dispatch=Fe(this.dispatchLayout,this.outputShape,this.workGroupSize,[this.workPerThread,1,1]))}makeFromPixelsSource(){let e=this.useImport?"textureLoad(src, vec2<i32>(coords.yx));":"textureLoad(src, vec2<i32>(coords.yx), 0)";return`
|
|
[[binding(1), group(0)]] var src: ${this.useImport?"texture_external":"texture_2d<f32>"};
|
|
|
|
${tt()}
|
|
let flatIndexBase = index * uniforms.numChannels;
|
|
for (var i = 0; i < uniforms.numChannels; i = i + 1) {
|
|
let flatIndex = flatIndexBase + i;
|
|
if (flatIndex < uniforms.size) {
|
|
let coords = getCoordsFromFlatIndex(flatIndexBase);
|
|
let values = ${e};
|
|
result.numbers[flatIndex] = i32(floor(255.0 * values[i]));
|
|
}
|
|
}
|
|
}
|
|
`}getUserCode(){return this.makeFromPixelsSource()}setPipeline(e){this.pipeline=e}setUniform(e,t){if(!this.uniform){let n=e.createBuffer({size:t.length*4,usage:GPUBufferUsage.UNIFORM|GPUBufferUsage.COPY_DST});this.uniform=n}!t||t.length===this.lastUniformData.length&&t.every((n,s)=>n===this.lastUniformData[s])||(e.queue.writeBuffer(this.uniform,0,new Uint32Array(t)),this.lastUniformData=t)}makeInputTexture(e,t,n){return(!this.inputTexture||this.lastPixelSize.width!==t||this.lastPixelSize.height!==n)&&(this.inputTexture&&this.inputTexture.destroy(),this.inputTexture=e.createTexture({size:[t,n],format:"rgba8unorm",usage:GPUTextureUsage.COPY_DST|GPUTextureUsage.RENDER_ATTACHMENT|GPUTextureUsage.TEXTURE_BINDING}),this.lastPixelSize.width=t,this.lastPixelSize.height=n),this.inputTexture}dispose(){this.disposed||(this.uniform&&this.uniform.destroy(),this.inputTexture&&this.inputTexture.destroy(),this.disposed=!0)}getLayout(e){return this.layout===null&&(this.layout=this.createTextureLayout(e)),this.layout}createTextureLayout(e){let t=[];t.push({binding:0,visibility:GPUShaderStage.COMPUTE,buffer:{type:"storage"}}),t.push({binding:1,visibility:GPUShaderStage.COMPUTE,texture:{}}),t.push({binding:2,visibility:GPUShaderStage.COMPUTE,buffer:{}});let n=e.createBindGroupLayout({entries:t}),s=e.createPipelineLayout({bindGroupLayouts:[n]});return{bindGroupLayout:n,pipelineLayout:s}}},ohe=class extends DC{constructor(){super(...arguments);this.layout=null,this.useImport=!0}getUserCode(){return this.makeFromPixelsSource()}getLayout(e){return this.layout===null&&(this.layout=this.createTextureImportLayout(e)),this.layout}createTextureImportLayout(e){let t=[];t.push({binding:0,visibility:GPUShaderStage.COMPUTE,buffer:{type:"storage"}}),t.push({binding:1,visibility:GPUShaderStage.COMPUTE,externalTexture:{}}),t.push({binding:2,visibility:GPUShaderStage.COMPUTE,buffer:{}});let n=e.createBindGroupLayout({entries:t}),s=e.createPipelineLayout({bindGroupLayouts:[n]});return{bindGroupLayout:n,pipelineLayout:s}}},ihe=K().getNumber("WEBGPU_CPU_HANDOFF_SIZE_THRESHOLD"),PC=class extends Yl{constructor(e,t=!1){super();if(this.commandQueueOwnedIds=new WeakSet,this.tensorDisposalQueue=[],this.uniformDisposalQueue=[],this.disposed=!1,this.uploadWaitMs=0,this.downloadWaitMs=0,this.dispatchNumberInEncoder=0,!yx())throw new Error("WebGPU is not supported on this device");this.layoutCache={},this.pipelineCache={},this.device=e,this.queue=e.queue,this.currentCommandEncoder=null,this.currentComputePass=null,this.supportTimeQuery=t,this.bufferManager=new ahe(this.device),this.tensorMap=new Xc(this,ns()),this.supportTimeQuery&&(this.querySet=this.device.createQuerySet({type:"timestamp",count:2})),K().getBool("WEBGPU_USE_PROFILE_TOOL")&&(this.dummyCanvas=document.createElement("canvas"),this.dummyCanvas.width=1,this.dummyCanvas.height=1,this.dummyContext=this.dummyCanvas.getContext("webgpu"),this.dummyContext.configure({device:e,format:"bgra8unorm"}),document.body.appendChild(this.dummyCanvas))}nextDataId(){return PC.nextDataId++}floatPrecision(){return 32}defaultGpuBufferUsage(){return GPUBufferUsage.STORAGE|GPUBufferUsage.COPY_SRC|GPUBufferUsage.COPY_DST}flushDisposalQueue(){this.tensorDisposalQueue.forEach(e=>{this.maybeReleaseBuffer(e),this.tensorMap.delete(e)}),this.uniformDisposalQueue.forEach(e=>this.bufferManager.releaseBuffer(e.buffer,e.byteSize,e.usage)),this.tensorDisposalQueue=[],this.uniformDisposalQueue=[]}disposeData(e,t=!1){if(this.tensorMap.has(e)){let n=this.tensorMap.get(e);if(n.refCount--,!t&&n.refCount>0)return!1;if(this.commandQueueOwnedIds.has(e))return this.tensorDisposalQueue.push(e),!1;this.maybeReleaseBuffer(e);let{complexTensorInfos:s}=this.tensorMap.get(e);s!=null&&(this.disposeData(s.real.dataId,!0),this.disposeData(s.imag.dataId,!0)),this.tensorMap.delete(e)}return!0}memory(){return{numBytesInGPU:this.bufferManager.numBytesUsed,numBytesAllocatedInGPU:this.bufferManager.numBytesAllocated,unreliable:!1}}getBufferManager(){return this.bufferManager}acquireBuffer(e,t=this.defaultGpuBufferUsage()){return this.bufferManager.acquireBuffer(e,t)}maybeReleaseBuffer(e){let t=this.tensorMap.get(e);t!=null&&t.bufferInfo.buffer!=null&&(this.bufferManager.releaseBuffer(t.bufferInfo.buffer,t.bufferInfo.byteSize,t.bufferInfo.usage),t.bufferInfo.buffer=null)}refCount(e){return this.tensorMap.has(e)?this.tensorMap.get(e).refCount:0}incRef(e){let t=this.tensorMap.get(e);t.refCount++}decRef(e){if(this.tensorMap.has(e)){let t=this.tensorMap.get(e);t.refCount--}}write(e,t,n){if(n==="complex64"&&e!=null)throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");let s={id:this.nextDataId()},r=v.sizeFromShape(t)*Ax(n);return n==="bool"&&e instanceof Uint8Array&&(e=Int32Array.from(e)),this.tensorMap.set(s,{dtype:n,values:e,bufferInfo:{byteSize:r,usage:this.defaultGpuBufferUsage()},refCount:1}),s}move(e,t,n,s,r){if(s==="complex64")throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");let a=v.sizeFromShape(n)*Ax(s);this.tensorMap.set(e,{dtype:s,values:t,bufferInfo:{byteSize:a,usage:this.defaultGpuBufferUsage()},refCount:r})}submitQueue(){this.ensureComputePassEnded(),this.queue.submit([this.currentCommandEncoder.finish()]),this.currentCommandEncoder=null,this.dispatchNumberInEncoder=0,this.commandQueueOwnedIds=new WeakSet,this.flushDisposalQueue()}getBuffer(e){return this.uploadToGPU(e),this.tensorMap.get(e).bufferInfo.buffer}getFromPixelsProgram(e){switch(e){case"copyExternal":return this.fromPixelProgram||(this.fromPixelProgram=new DC),this.fromPixelProgram;case"import":return this.fromPixelImportProgram||(this.fromPixelImportProgram=new ohe),this.fromPixelImportProgram;default:v.assert(!1,()=>"Unsupported fromPixels shape");return}}ensureCommandEncoderReady(){this.currentCommandEncoder||(this.currentCommandEncoder=this.device.createCommandEncoder())}ensureComputePassEnded(){this.currentComputePass&&(this.currentComputePass.endPass(),this.currentComputePass=null)}getComputePass(){return this.currentComputePass||(this.currentComputePass=this.currentCommandEncoder.beginComputePass()),this.currentComputePass}async getBufferData(e){if(e.values!=null)return e.values;let t=this.acquireBuffer(e.bufferInfo.byteSize,GPUBufferUsage.COPY_DST|GPUBufferUsage.MAP_READ);this.ensureCommandEncoderReady(),this.ensureComputePassEnded(),this.currentCommandEncoder.copyBufferToBuffer(e.bufferInfo.buffer,0,t,0,e.bufferInfo.byteSize),this.submitQueue(),await t.mapAsync(GPUMapMode.READ);let n=t.getMappedRange().slice(0);return t.unmap(),t!=null&&this.bufferManager.releaseBuffer(t,e.bufferInfo.byteSize,GPUBufferUsage.COPY_DST|GPUBufferUsage.MAP_READ),K().getBool("WEBGPU_USE_PROFILE_TOOL")&&(v.assert(this.dummyContext!==void 0,()=>"Fail to get context for profiling tool"),this.dummyContext.getCurrentTexture()),n}convertAndCacheOnCPU(e,t){let n=this.tensorMap.get(e);return this.maybeReleaseBuffer(e),n.values=t,n.values}readSync(e){let t=this.tensorMap.get(e),{values:n}=t;if(n==null)throw new Error("WebGPU readSync is only available for CPU-resident tensors.");return n}async read(e){if(!this.tensorMap.has(e))throw new Error(`Tensor ${e} was not registered!`);let t=this.tensorMap.get(e),{values:n}=t;if(n!=null)return this.convertAndCacheOnCPU(e,n);let s;if(t.dtype==="complex64"){let r=await Promise.all([this.read(t.complexTensorInfos.real.dataId),this.read(t.complexTensorInfos.imag.dataId)]),a=r[0],o=r[1];s=E.mergeRealAndImagArrays(a,o)}else{let r=await this.getBufferData(t);s=tC(r,t.dtype)}return this.convertAndCacheOnCPU(e,s),s}bufferSync(e){let t=this.readSync(e.dataId),n=t;if(e.dtype==="string")try{n=t.map(s=>v.decodeString(s))}catch(s){throw new Error("Failed to decode encoded string bytes into utf-8")}return ze(e.shape,e.dtype,n)}async time(e){let t=this.activeTimers,n=[],s=!1;this.programTimersStack==null?(this.programTimersStack=n,s=!0):this.activeTimers.push(n),this.activeTimers=n,e();let r=v.flatten(this.activeTimers.map(l=>l.query)).filter(l=>l!=null),a=v.flatten(this.activeTimers.map(l=>l.name)).filter(l=>l!=null);this.activeTimers=t,s&&(this.programTimersStack=null);let o={uploadWaitMs:this.uploadWaitMs,downloadWaitMs:this.downloadWaitMs,kernelMs:null,wallMs:null},i=await Promise.all(r);return o.kernelMs=v.sum(i),o.getExtraProfileInfo=()=>i.map((l,c)=>({name:a[c],ms:l})).map(l=>`${l.name}: ${l.ms}`).join(", "),this.uploadWaitMs=0,this.downloadWaitMs=0,o}getAndSavePipeline(e,t){return e in this.pipelineCache||(this.pipelineCache[e]=t()),this.pipelineCache[e]}makeTensorInfo(e,t,n){let s;if(t==="string"&&n!=null&&n.length>0&&v.isString(n[0])){let r=n.map(a=>v.encodeString(a));s=this.write(r,e,t)}else s=this.write(n,e,t);return{dataId:s,shape:e,dtype:t}}tensorToBinding(e){if(!e)return null;let t=this.tensorMap.get(e.dataId);return{offset:0,size:t.bufferInfo.byteSize,buffer:t.bufferInfo.buffer}}async getQueryTime(e){return this.supportTimeQuery?this.getTimeFromQuerySet(e):0}uploadToGPU(e){let t=this.tensorMap.get(e);t.bufferInfo.buffer==null&&(t.bufferInfo.buffer=this.acquireBuffer(t.bufferInfo.byteSize),t.values&&this.queue.writeBuffer(t.bufferInfo.buffer,0,t.values))}makeUniformsDataView(e){let t=this.acquireBuffer(e.byteLength,GPUBufferUsage.COPY_DST|GPUBufferUsage.UNIFORM);return this.queue.writeBuffer(t,0,e),{offset:0,size:e.byteLength,buffer:t}}arrayToDataView(e,t){let n=4,s=new DataView(new ArrayBuffer(t*n)),r=0;return e.forEach(a=>{let o=a.data;if(a.type!=="int32"&&a.type!=="float32"&&a.type!=="uint32")throw new Error(`${a.type} not supported!`);a.type==="int32"?o.forEach(i=>{s.setInt32(r*n,i,!0),r++}):a.type==="uint32"?o.forEach(i=>{s.setUint32(r*n,i,!0),r++}):o.forEach(i=>{s.setFloat32(r*n,i,!0),r++})}),s}computePadding(e){let t=0,n=0,s=0,r=[];return e.forEach((a,o)=>{a.data.length===0&&(a.data=[1]);let i;switch(a.data.length){case 0:i=1;break;case 1:i=1;break;case 2:i=2;break;case 3:i=4;break;case 4:i=4;break;default:v.assert(!1,()=>`Unsupported ${a.data.length}D shape`)}n=Math.ceil(t/i)*i-t;for(let l=0;l<n;++l)r.push({type:a.type,data:[0]}),s++;r.push({type:a.type,data:a.data}),s=s+a.data.length,t+=a.data.length+n}),this.arrayToDataView(r,s)}createLayout(e){let t=[];t.push({binding:0,visibility:GPUShaderStage.COMPUTE,buffer:{type:"storage"}});for(let r=0;r<e;r++)t.push({binding:r+1,visibility:GPUShaderStage.COMPUTE,buffer:{type:"read-only-storage"}});t.push({binding:e+1,visibility:GPUShaderStage.COMPUTE,buffer:{type:"uniform"}});let n=this.device.createBindGroupLayout({entries:t}),s=this.device.createPipelineLayout({bindGroupLayouts:[n]});return{bindGroupLayout:n,pipelineLayout:s}}getCachedOrCreateLayout(e){return e in this.layoutCache||(this.layoutCache[e]=this.createLayout(e)),this.layoutCache[e]}runWebGPUProgram(e,t,n,s,r){if(!r){if(r=this.makeTensorInfo(e.outputShape,n),v.sizeFromShape(r.shape)===0){let N=this.tensorMap.get(r.dataId);return N.values=v.getTypedArrayFromDType(r.dtype,0),r}this.uploadToGPU(r.dataId)}let a=[{type:"float32",data:[NaN]}],o=t.concat(r).map(N=>N.shape),i="int32";o.map(N=>{a.push({type:i,data:N})});let l=v.computeStrides(r.shape);if(a.push({type:i,data:l}),e.size){let N=v.sizeFromShape(e.outputShape);a.push({type:i,data:[e.isVec4?N/4:N]})}s&&(a=[...a,...s]);let c=null,u=this.computePadding(a),d=u.byteLength;c=this.makeUniformsDataView(u);let p=t.map((N,R)=>{if(N.dtype==="complex64")throw new Error("GPGPUProgram does not support complex64 input. For complex64 dtypes, please separate the program into real and imaginary parts.");return this.uploadToGPU(N.dataId),{dtype:this.tensorMap.get(N.dataId).dtype,shape:N.shape,name:e.variableNames[R]}}),h=p.map(N=>N.dtype).concat(r.dtype),f=p.map(N=>E.getBroadcastDims(N.shape,r.shape)),m=p.map(N=>v.arraysEqual(N.shape,r.shape)).join("_"),g=f.map(N=>N.join("_")).join(";"),A=kC(e,o,h,g,m),{bindGroupLayout:x,pipelineLayout:y}=this.getCachedOrCreateLayout(e.variableNames.length),b=this.getAndSavePipeline(A,()=>wC(this.device,e,y,p,r)),w=this.activeTimers!=null,k=Vce(this.device,x,t.map(N=>this.tensorToBinding(N)),this.tensorToBinding(r),c);this.ensureCommandEncoderReady();let C=this.getComputePass();if(w&&this.supportTimeQuery&&C.writeTimestamp(this.querySet,0),C.setPipeline(b),C.setBindGroup(0,k),C.dispatch(e.dispatch[0],e.dispatch[1],e.dispatch[2]),w&&this.supportTimeQuery&&C.writeTimestamp(this.querySet,1),this.dispatchNumberInEncoder++,t.forEach(N=>{this.commandQueueOwnedIds.add(N.dataId)}),this.commandQueueOwnedIds.add(r.dataId),c){let N={byteSize:d,usage:GPUBufferUsage.COPY_DST|GPUBufferUsage.UNIFORM,buffer:c.buffer};this.uniformDisposalQueue.push(N)}return K().get("WEBGPU_DEFERRED_SUBMIT_BATCH_SIZE")<=this.dispatchNumberInEncoder&&this.submitQueue(),w&&this.activeTimers.push({name:e.constructor.name,query:this.getQueryTime(this.querySet)}),r}runFromPixelsProgram(e,t,n,s,r){let a=this.device.createBindGroup({layout:n.bindGroupLayout,entries:[{binding:0,resource:{buffer:t}},{binding:1,resource:s},{binding:2,resource:{buffer:e.uniform}}]});this.ensureCommandEncoderReady();let o=this.getComputePass(),i=this.activeTimers!=null;i&&this.supportTimeQuery&&o.writeTimestamp(this.querySet,0),o.setPipeline(e.pipeline),o.setBindGroup(0,a),o.dispatch(e.dispatch[0],e.dispatch[1],e.dispatch[2]),i&&this.supportTimeQuery&&o.writeTimestamp(this.querySet,1),this.commandQueueOwnedIds.add(r),this.submitQueue(),i&&this.activeTimers.push({name:e.constructor.name,query:this.getQueryTime(this.querySet)})}async getTimeFromQuerySet(e){let t=this.acquireBuffer(16,GPUBufferUsage.COPY_SRC|GPUBufferUsage.QUERY_RESOLVE),n=this.acquireBuffer(16,GPUBufferUsage.MAP_READ|GPUBufferUsage.COPY_DST);this.ensureCommandEncoderReady(),this.ensureComputePassEnded(),this.currentCommandEncoder.resolveQuerySet(e,0,2,t,0),this.currentCommandEncoder.copyBufferToBuffer(t,0,n,0,16),this.submitQueue(),await n.mapAsync(GPUMapMode.READ);let s=new BigUint64Array(n.getMappedRange()),r=Number(s[1]-s[0]);return n.unmap(),this.bufferManager.releaseBuffer(n,16,GPUBufferUsage.MAP_READ|GPUBufferUsage.COPY_DST),this.bufferManager.releaseBuffer(t,16,GPUBufferUsage.COPY_SRC|GPUBufferUsage.QUERY_RESOLVE),r/1e6}shouldExecuteOnCPU(e,t=ihe){return K().getBool("WEBGPU_CPU_FORWARD")&&e.every(n=>this.tensorMap.get(n.dataId).bufferInfo.buffer==null&&v.sizeFromShape(n.shape)<t)}numDataIds(){return this.tensorMap.numDataIds()-this.tensorDisposalQueue.length}dispose(){this.disposed||(this.bufferManager.dispose(),this.fromPixelProgram&&this.fromPixelProgram.dispose(),this.fromPixelImportProgram&&this.fromPixelImportProgram.dispose(),this.disposed=!0)}},Cx=PC;Cx.nextDataId=0;var FC={};Oe(FC,{WebGPUBackend:()=>Cx,webgpu_util:()=>eC});Nu.isBrowser()&&yx()&&al("webgpu",async()=>{K().set("CHECK_COMPUTATION_FOR_ERRORS",!1);let e={powerPreference:K().get("WEBGPU_USE_LOW_POWER_GPU")?"low-power":"high-performance"},t=await navigator.gpu.requestAdapter(e),n={},s=t.features.has("timestamp-query");s?n={requiredFeatures:["timestamp-query"]}:console.warn("This device doesn't support timestamp-query extension. Start Chrome browser with flag --disable-dawn-features=disallow_unsafe_apis then try again. Or zero will shown for the kernel time when profiling mode isenabled. Using performance.now is not workable for webgpu sinceit doesn't support synchronously to read data from GPU.");let r=await t.requestDevice(n);return new Cx(r,s)},3);var Yt;(function(e){e[e.float32=0]="float32",e[e.int32=1]="int32",e[e.bool=2]="bool",e[e.string=3]="string",e[e.complex64=4]="complex64"})(Yt||(Yt={}));var bp;(function(e){e[e.linear=0]="linear",e[e.relu=1]="relu",e[e.relu6=2]="relu6",e[e.prelu=3]="prelu",e[e.leakyrelu=4]="leakyrelu",e[e.sigmoid=5]="sigmoid",e[e.elu=6]="elu"})(bp||(bp={}));var OC;function lhe(e){OC=e.wasm.cwrap(yo,null,["number","array","number","number","array","number","number","number","number","number","number","number","number"])}function uhe(e){let{inputs:t,backend:n,attrs:s}=e,{a:r,b:a,bias:o,preluActivationWeights:i}=t;if(r.dtype!=="float32"||a.dtype!=="float32")throw new Error("_FusedMatMul for non non-float32 tensors not yet supported.");let{transposeA:l,transposeB:c,activation:u,leakyreluAlpha:d}=s,p=n.dataIdMap.get(r.dataId).id,h=n.dataIdMap.get(a.dataId).id,f=0;if(o!=null){let N=n.dataIdMap.get(o.dataId);if(N.shape.length!==1)throw new Error(`_FusedMatMul only supports rank-1 bias but got rank ${N.shape.length}.`);f=N.id}let m=i==null?0:n.dataIdMap.get(i.dataId).id,g=bp[u];if(g==null)throw new Error(`${u} activation not yet supported for FusedConv2D in the wasm backend.`);let A=l?r.shape[2]:r.shape[1],x=c?a.shape[1]:a.shape[2],y=nl.assertAndGetBroadcastShape(r.shape.slice(0,-2),a.shape.slice(0,-2)),b=n.makeOutput([...y,A,x],r.dtype),w=n.dataIdMap.get(b.dataId).id,k=new Uint8Array(new Int32Array(r.shape).buffer),C=new Uint8Array(new Int32Array(a.shape).buffer);return OC(p,k,r.shape.length,h,C,a.shape.length,l,c,g,f,m,d||0,w),b}var che={kernelName:yo,backendName:"wasm",setupFunc:lhe,kernelFunc:uhe};function Nn(e,t){let n;function s(a){n=a.wasm.cwrap(e,null,["number","number","number"])}function r(a){let{backend:o,inputs:{x:i}}=a,l=o.dataIdMap.get(i.dataId).id,c=o.makeOutput(i.shape,t||i.dtype),u=o.dataIdMap.get(c.dataId).id;return v.sizeFromShape(c.shape)===0||n(l,Yt[i.dtype],u),c}return{kernelName:e,backendName:"wasm",setupFunc:s,kernelFunc:r}}var dhe=Nn(ci);function qn(e,t,n){let s;function r(o){s=o.wasm.cwrap(e,null,["number","array","number","number","array","number","number","number"])}function a(o){let{backend:i,inputs:l}=o,{a:c,b:u}=l,d=i.dataIdMap.get(c.dataId).id,p=i.dataIdMap.get(u.dataId).id,h=n!=null?n:c.dtype,f=E.assertAndGetBroadcastShape(c.shape,u.shape),m=i.makeOutput(f,h);if(v.sizeFromShape(f)===0)return m;let g=new Uint8Array(new Int32Array(c.shape).buffer),A=new Uint8Array(new Int32Array(u.shape).buffer),x=i.dataIdMap.get(m.dataId).id,y=()=>s(d,g,c.shape.length,p,A,u.shape.length,Yt[c.dtype],x);if(t&&c.dtype==="float32")return y(),m;let b=E.getBroadcastDims(c.shape,f),w=E.getBroadcastDims(u.shape,f),k=b.every((N,R)=>N===R),C=w.every((N,R)=>N===R);if(k&&C)return y(),m;throw new Error(`Broadcasting along outer dims is not yet supported for ${c.dtype} ${e}.`)}return{kernelName:e,backendName:"wasm",setupFunc:r,kernelFunc:a}}var phe=!0,hhe=qn(Gr,phe),MC;function fhe(e){MC=e.wasm.cwrap(Ca,null,["array","number","number","number"])}function mhe(e){let{inputs:t,backend:n}=e,s=n.makeOutput(t[0].shape,t[0].dtype);if(v.sizeFromShape(s.shape)===0)return s;let r=t.map(i=>n.dataIdMap.get(i.dataId).id),a=new Uint8Array(new Int32Array(r).buffer),o=n.dataIdMap.get(s.dataId).id;return MC(a,r.length,Yt[s.dtype],o),s}var ghe={kernelName:Ca,backendName:"wasm",setupFunc:fhe,kernelFunc:mhe};function Jm(e){let{inputs:{x:t},backend:n}=e,s=n.makeOutput(t.shape,t.dtype),r=n.typedArrayFromHeap(t);return n.typedArrayFromHeap(s).set(r),s}var Ahe={kernelName:Ga,backendName:"wasm",kernelFunc:Jm},zC;function yhe(e){zC=e.wasm.cwrap(go,null,["number","array","number","number","number","array","number"])}function xc(e){let{inputs:t,backend:n,attrs:s}=e,[r,a]=bhe(t.x.shape,s.perm),o=!0;for(let f=0;f<a.length;f++)a[f]!==f&&(o=!1);let i=xhe(t.x.shape,s.perm),l={dataId:t.x.dataId,shape:r,dtype:t.x.dtype};if(o){let f=Jm({inputs:t,backend:n});return f.shape=i,f}let c=n.makeOutput(i,l.dtype),u=n.dataIdMap.get(l.dataId).id,d=n.dataIdMap.get(c.dataId).id,p=new Uint8Array(new Int32Array(a).buffer),h=new Uint8Array(new Int32Array(l.shape).buffer);return zC(u,h,l.shape.length,Yt[l.dtype],d,p,a.length),c}function xhe(e,t){let n=new Array(e.length);for(let s=0;s<n.length;s++)n[s]=e[t[s]];return n}function bhe(e,t){let n=[],s=[];for(let r=0;r<e.length;++r)e[r]!==1&&n.push(e[r]),e[t[r]]!==1&&s.push(t[r]);for(let r=0;r<s.length;++r){let a=-1;for(let o=0;o<s.length;++o)s[o]>=r&&(a===-1||s[a]>s[o])&&(a=o);s[a]=r}return[n,s]}var vhe={kernelName:go,backendName:"wasm",kernelFunc:xc,setupFunc:yhe};function Ho(e,t,n){let s=e.shape,r=e.shape.length,a=v.parseAxisParam(t,s),o=a,i=E.getAxesPermutation(o,r),l=null,c=!1;if(i!=null){let u=new Array(r);for(let h=0;h<u.length;h++)u[h]=s[i[h]];o=E.getInnerMostAxes(o.length,r),l=xc({inputs:{x:e},attrs:{perm:i},backend:n});let d=n.dataIdMap.get(e.dataId).id;n.dataIdMap.get(l.dataId).id!==d&&(c=!0)}return{transposed:l,originalAxes:a,axes:o,inputWasTransposed:c}}var LC;function whe(e){LC=e.wasm.cwrap(nu,null,["number, number, number"])}function khe(e){let{backend:t,inputs:n,attrs:s}=e,{axis:r,keepDims:a}=s,{x:o}=n,l=t.dataIdMap.get(o.dataId).id,c=o,{transposed:u,axes:d,originalAxes:p,inputWasTransposed:h}=Ho(o,r,t);if(h){let y=t.dataIdMap.get(u.dataId).id;c=u,l=y}let f=c.shape.length;E.assertAxesAreInnerMostDims("all",d,f);let[m,g]=E.computeOutAndReduceShapes(c.shape,d),A=v.sizeFromShape(g),x=t.makeOutput(m,o.dtype);if(v.sizeFromShape(c.shape)!==0){let y=t.dataIdMap.get(x.dataId).id;LC(l,A,y)}if(h&&t.disposeData(u.dataId),a){let y=E.expandShapeToKeepDim(x.shape,p);x.shape=y}return x}var She={kernelName:nu,backendName:"wasm",setupFunc:whe,kernelFunc:khe},BC;function Ihe(e){BC=e.wasm.cwrap(su,null,["number, number, number"])}function Che(e){let{backend:t,inputs:n,attrs:s}=e,{axis:r,keepDims:a}=s,{x:o}=n,l=t.dataIdMap.get(o.dataId).id,c=o,{transposed:u,axes:d,originalAxes:p,inputWasTransposed:h}=Ho(o,r,t);if(h){let y=t.dataIdMap.get(u.dataId).id;c=u,l=y}let f=c.shape.length;E.assertAxesAreInnerMostDims("any",d,f);let[m,g]=E.computeOutAndReduceShapes(c.shape,d),A=v.sizeFromShape(g),x=t.makeOutput(m,o.dtype);if(v.sizeFromShape(c.shape)!==0){let y=t.dataIdMap.get(x.dataId).id;BC(l,A,y)}if(h&&t.disposeData(u.dataId),a){let y=E.expandShapeToKeepDim(x.shape,p);x.shape=y}return x}var The={kernelName:su,backendName:"wasm",setupFunc:Ihe,kernelFunc:Che},WC;function Nhe(e){WC=e.wasm.cwrap(Ta,null,["number","number","number","number","number"])}function Ehe(e){let{backend:t,inputs:n,attrs:s}=e,{axis:r}=s,{x:a}=n,o=t.dataIdMap.get(a.dataId).id,i=o,l=a,{transposed:c,axes:u,inputWasTransposed:d}=Ho(a,r,t);if(d){let A=t.dataIdMap.get(c.dataId).id;A!==o&&(l=c,i=A)}let p=l.shape.slice(0,-1),h=t.makeOutput(p,"int32"),f=t.dataIdMap.get(h.dataId).id,m=v.sizeFromShape(h.shape),g=l.shape[u[0]];return WC(i,Yt[l.dtype],m,g,f),d&&t.disposeData(c.dataId),h}var Rhe={kernelName:Ta,backendName:"wasm",kernelFunc:Ehe,setupFunc:Nhe},VC;function $he(e){VC=e.wasm.cwrap(Na,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function _he(e){let{inputs:t,attrs:n,backend:s}=e,r=t.x,a=s.dataIdMap.get(r.dataId).id,{filterSize:o,strides:i,pad:l,dimRoundingMode:c}=n,u=E.computePool2DInfo(r.shape,o,i,1,l,c),d=u.filterHeight,p=u.filterWidth,h=u.padInfo.top,f=u.padInfo.right,m=u.padInfo.bottom,g=u.padInfo.left,A=u.strideHeight,x=u.strideWidth,y=u.inChannels;if(u.dataFormat!=="channelsLast")throw new Error(`wasm backend does not support dataFormat:'${u.dataFormat}'. Please use 'channelsLast'.`);if(u.dilationWidth!==1||u.dilationHeight!==1)throw new Error(`was backend only supports average pooling with dilation = [1, 1], got [${u.dilationHeight}, ${u.dilationWidth}].`);let b=s.makeOutput(u.outShape,"float32"),w=s.dataIdMap.get(b.dataId).id;return VC(a,r.shape[0],r.shape[1],r.shape[2],d,p,h,f,m,g,A,x,y,w),b}var Dhe={kernelName:Na,backendName:"wasm",setupFunc:$he,kernelFunc:_he};function is(e){let{inputs:t,attrs:n}=e,{x:s}=t,{shape:r}=n,a=v.sizeFromShape(s.shape),o=v.inferFromImplicitShape(r,a);return v.assert(a===v.sizeFromShape(o),()=>`new shape: ${o}, old shape: ${s.shape}. New shape and old shape must have the same number of elements.`),e.backend.incRef(s.dataId),{dataId:s.dataId,shape:o,dtype:s.dtype}}var Phe={kernelName:Fi,backendName:"wasm",kernelFunc:is},UC;function Fhe(e){UC=e.wasm.cwrap(Ea,null,["number","array","number","number","array","number","number","number","number"])}function Ohe(e){let{inputs:t,backend:n,attrs:s}=e,{a:r,b:a}=t,{transposeA:o,transposeB:i}=s;if(r.dtype!=="float32"||a.dtype!=="float32")throw new Error("BatchMatMul for non non-float32 tensors not yet supported.");let l=r.shape.length,c=a.shape.length,u=o?r.shape[l-2]:r.shape[l-1],d=i?a.shape[c-1]:a.shape[c-2],p=o?r.shape[l-1]:r.shape[l-2],h=i?a.shape[c-2]:a.shape[c-1],f=r.shape.slice(0,-2),m=a.shape.slice(0,-2),g=v.sizeFromShape(f),A=v.sizeFromShape(m),y=nl.assertAndGetBroadcastShape(r.shape.slice(0,-2),a.shape.slice(0,-2)).concat([p,h]);v.assert(u===d,()=>`Error in matMul: inner shapes (${u}) and (${d}) of Tensors with shapes ${r.shape} and ${a.shape} and transposeA=${o} and transposeB=${i} must match.`);let b=o?[g,u,p]:[g,p,u],w=i?[A,h,d]:[A,d,h],k=is({inputs:{x:r},backend:n,attrs:{shape:b}}),C=is({inputs:{x:a},backend:n,attrs:{shape:w}}),N=n.dataIdMap.get(k.dataId).id,R=n.dataIdMap.get(C.dataId).id,F=o?k.shape[2]:k.shape[1],_=i?C.shape[1]:C.shape[2],P=Math.max(g,A),T=n.makeOutput([P,F,_],k.dtype),M=n.dataIdMap.get(T.dataId).id,U=new Uint8Array(new Int32Array(k.shape).buffer),j=new Uint8Array(new Int32Array(C.shape).buffer);return UC(N,U,k.shape.length,R,j,C.shape.length,o,i,M),n.disposeData(k.dataId),n.disposeData(C.dataId),T.shape=y,T}var Mhe={kernelName:Ea,backendName:"wasm",setupFunc:Fhe,kernelFunc:Ohe};function vp(e){let{inputs:{x:t},attrs:{begin:n,size:s},backend:r}=e,[a,o]=Pt.parseSliceParams(t,n,s),i=Pt.isSliceContinous(t.shape,a,o),l=r.readSync(t.dataId),c=r.makeOutput(o,t.dtype),u=v.computeStrides(t.shape),d=r.dataIdMap.get(c.dataId);if(i){let f=Pt.computeFlatOffset(a,u);return t.dtype==="string"?d.stringBytes=l.slice(f,f+v.sizeFromShape(o)):r.typedArrayFromHeap(c).set(l.subarray(f,f+v.sizeFromShape(o))),c}if(t.dtype==="string"){let f=km(l,a,o,t.shape,t.dtype);return d.stringBytes=f,c}let p=r.typedArrayFromHeap(c),h=t.shape.length;if(h===2)zhe(l,u[0],p,a,o);else if(h===3)Lhe(l,u[0],u[1],p,a,o);else if(h===4)Bhe(l,u[0],u[1],u[2],p,a,o);else{let f=km(l,a,o,t.shape,t.dtype);p.set(f)}return c}function zhe(e,t,n,s,r){let a=0,o=s[0],i=s[1],l=o+r[0];for(let c=o;c<l;c++){let u=c*t+i;n.set(e.subarray(u,u+r[1]),a),a+=r[1]}}function Lhe(e,t,n,s,r,a){let o=0,i=r[0],l=r[1],c=r[2],u=i+a[0],d=l+a[1];for(let p=i;p<u;p++)for(let h=l;h<d;h++){let f=p*t+h*n+c;s.set(e.subarray(f,f+a[2]),o),o+=a[2]}}function Bhe(e,t,n,s,r,a,o){let i=0,l=a[0],c=a[1],u=a[2],d=l+o[0],p=c+o[1],h=u+o[2],f=a[3];for(let m=l;m<d;m++)for(let g=c;g<p;g++)for(let A=u;A<h;A++){let x=m*t+g*n+A*s+f;r.set(e.subarray(x,x+o[3]),i),i+=o[3]}}var Whe={kernelName:Bi,backendName:"wasm",kernelFunc:vp};function Vhe(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{blockShape:a,crops:o}=s,i=a.reduce((A,x)=>A*x),l=E.getReshaped(r.shape,a,i),c=E.getPermuted(l.length,a.length),u=E.getReshapedPermuted(r.shape,a,i),d=E.getSliceBeginCoords(o,a.length),p=E.getSliceSize(u,o,a.length),h=is({inputs:{x:r},backend:n,attrs:{shape:l}}),f=xc({inputs:{x:h},backend:n,attrs:{perm:c}}),m=is({inputs:{x:f},backend:n,attrs:{shape:u}}),g=vp({inputs:{x:m},backend:n,attrs:{begin:d,size:p}});return n.disposeData(h.dataId),n.disposeData(f.dataId),n.disposeData(h.dataId),g}var Uhe={kernelName:di,backendName:"wasm",kernelFunc:Vhe};function wp(e){let{inputs:{x:t},attrs:{dtype:n},backend:s}=e,r=s.makeOutput(t.shape,n),a=s.typedArrayFromHeap(t);return s.typedArrayFromHeap(r).set(a),r}var Ghe={kernelName:Ra,backendName:"wasm",kernelFunc:wp},Hhe=Nn($a),GC;function jhe(e){GC=e.wasm.cwrap(Hr,null,["number","number","number","number"])}function qhe(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{clipValueMin:a,clipValueMax:o}=s,i=n.dataIdMap.get(r.dataId).id,l=n.makeOutput(r.shape,r.dtype),c=n.dataIdMap.get(l.dataId).id;return GC(i,a,o,c),l}var Xhe={kernelName:Hr,backendName:"wasm",setupFunc:jhe,kernelFunc:qhe};function HC(e){let{inputs:t,backend:n}=e,s=v.parseAxisParam(e.attrs.axis,t[0].shape)[0],r=E.computeOutShape(t.map(h=>h.shape),s),a=t.filter(h=>v.sizeFromShape(h.shape)>0);if(a.length===1)return Jm({inputs:{x:a[0]},backend:n});let o=n.makeOutput(r,t[0].dtype);if(v.sizeFromShape(r)===0)return o;let i=a.map(h=>h.shape);if(E.assertParamsConsistent(i,s),a[0].dtype==="string"){let h=a.map(y=>{let b=v.sizeFromShape(y.shape.slice(s));return is({inputs:{x:y},backend:n,attrs:{shape:[-1,b]}})}),f=h.map(y=>({vals:n.readSync(y.dataId),shape:y.shape}));r=E.computeOutShape(h.map(y=>y.shape),1);let m=h[0].shape[0]===1,g=Fy(f,r,t[0].dtype,m),A=E.computeOutShape(a.map(y=>y.shape),s);o.shape=A;let x=n.dataIdMap.get(o.dataId);return x.stringBytes=E.fromStringArrayToUint8(g),h.forEach(y=>n.disposeData(y.dataId)),o}let l=v.sizeFromShape(a[0].shape.slice(0,s)),c=0,u=a.map(h=>{let f=v.sizeFromShape(h.shape.slice(s));return c+=f,f}),d=a.map(h=>n.typedArrayFromHeap(h)),p=n.typedArrayFromHeap(o);for(let h=0;h<l;h++){let f=h*c;for(let m=0;m<d.length;m++){let g=u[m],A=h*g,x=d[m].subarray(A,A+g);p.set(x,f),f+=g}}return o}var Khe={kernelName:pi,backendName:"wasm",kernelFunc:HC},jC;function Zhe(e){jC=e.wasm.cwrap(_a,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function Yhe(e){let{inputs:t,attrs:n,backend:s}=e,{x:r,filter:a}=t,o=s.dataIdMap.get(r.dataId).id,i=s.dataIdMap.get(a.dataId).id,{strides:l,dilations:c,pad:u,dimRoundingMode:d,dataFormat:p}=n,h=E.convertConv2DDataFormat(p),f=E.computeConv2DInfo(r.shape,a.shape,l,c,u,d,!1,h),m=f.filterHeight,g=f.filterWidth,A=f.padInfo.top,x=f.padInfo.right,y=f.padInfo.bottom,b=f.padInfo.left,w=f.dilationHeight,k=f.dilationWidth,C=f.strideHeight,N=f.strideWidth,R=f.inChannels,F=f.outChannels,_=f.padInfo.type==="SAME"?1:0;if(f.dataFormat!=="channelsLast")throw new Error(`wasm backend Conv2D does not support dataFormat:'${f.dataFormat}'. Please use 'channelsLast'.`);let P=s.makeOutput(f.outShape,"float32"),T=s.dataIdMap.get(P.dataId).id;return jC(o,r.shape[0],r.shape[1],r.shape[2],i,m,g,A,x,y,b,_,w,k,C,N,R,F,T),P}var Jhe={kernelName:_a,backendName:"wasm",setupFunc:Zhe,kernelFunc:Yhe},qC;function Qhe(e){qC=e.wasm.cwrap(Da,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function efe(e){let{backend:t,inputs:n,attrs:s}=e,{dy:r,filter:a}=n,{strides:o,pad:i,dataFormat:l,dimRoundingMode:c,inputShape:u}=s,d=1,p=E.convertConv2DDataFormat(l),h=E.computeConv2DInfo(u,a.shape,o,d,i,c,!1,p),{batchSize:f,filterHeight:m,filterWidth:g,inChannels:A,inHeight:x,inWidth:y,outChannels:b,outHeight:w,outWidth:k,strideHeight:C,strideWidth:N}=h,R=m-1-h.padInfo.top,F=g-1-h.padInfo.left,_=h.dataFormat==="channelsLast",P=v.computeStrides(h.inShape),T=v.computeStrides(r.shape),[M,U,j]=v.computeStrides(a.shape),z=P[0],X=_?P[1]:P[2],Z=_?P[2]:1,J=_?1:P[1],ee=T[0],ne=_?T[1]:T[2],Q=_?T[2]:1,te=_?1:T[1],oe=t.makeOutput(h.inShape,"float32"),fe=t.dataIdMap.get(oe.dataId).id,be=t.dataIdMap.get(r.dataId).id,we=t.dataIdMap.get(a.dataId).id;return qC(be,we,f,m,g,x,y,A,w,k,b,C,N,R,F,M,U,j,z,X,Z,J,ee,ne,Q,te,fe),oe}var tfe={kernelName:Da,backendName:"wasm",setupFunc:Qhe,kernelFunc:efe},nfe=Nn(Pa),sfe=Nn(Fa),Tx;(function(e){e[e.bilinear=0]="bilinear",e[e.nearest=1]="nearest"})(Tx||(Tx={}));var XC;function rfe(e){XC=e.wasm.cwrap(fi,null,["number","number","number","number","array","number","number","number","number","number"])}function afe(e){let{backend:t,inputs:n,attrs:s}=e,{method:r,extrapolationValue:a,cropSize:o}=s,{image:i,boxes:l,boxInd:c}=n,u=l.shape[0],[d,p]=o,h=[u,d,p,i.shape[3]],f=t.dataIdMap.get(i.dataId),m;i.dtype!=="float32"&&(m=wp({backend:t,inputs:{x:i},attrs:{dtype:"float32"}}),f=t.dataIdMap.get(m.dataId));let g=f.id,A=t.dataIdMap.get(l.dataId).id,x=t.dataIdMap.get(c.dataId).id,y=t.makeOutput(h,"float32"),b=t.dataIdMap.get(y.dataId).id,w=new Uint8Array(new Int32Array(i.shape).buffer);return XC(g,A,x,u,w,d,p,Tx[r],a,b),m!=null&&t.disposeData(m.dataId),y}var ofe={kernelName:fi,backendName:"wasm",setupFunc:rfe,kernelFunc:afe},KC;function ife(e){KC=e.wasm.cwrap(hi,null,["number","number","number","number","number","number"])}function lfe(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{axis:a,exclusive:o,reverse:i}=s,l=r.shape.length;v.assert(r.dtype==="float32"||r.dtype==="int32",()=>`cumsum does not support ${r.dtype} tensors in the WASM backend`);let c=E.getAxesPermutation([a],l),u=r;c!==null&&(u=xc({inputs:{x:r},attrs:{perm:c},backend:n}));let d=E.getInnerMostAxes(1,l)[0];E.assertAxesAreInnerMostDims("cumsum",[d],l);let p=n.makeOutput(u.shape,u.dtype),h=u.shape[d],f=n.dataIdMap.get(u.dataId).id,m=n.dataIdMap.get(p.dataId).id;KC(f,o?1:0,i?1:0,h,m,Yt[r.dtype]);let g=p;if(c!==null){let A=E.getUndoAxesPermutation(c);g=xc({inputs:{x:p},attrs:{perm:A},backend:n}),n.disposeData(u.dataId),n.disposeData(p.dataId)}return g}var ufe={kernelName:hi,backendName:"wasm",setupFunc:ife,kernelFunc:lfe},ZC;function cfe(e){ZC=e.wasm.cwrap(mi,null,["number","number","number","array","number","array","array","number","number"])}function dfe(e){let{backend:t,inputs:n,attrs:s}=e,{x:r}=n,{blockSize:a,dataFormat:o}=s,i=r.shape[0],l=o==="NHWC"?r.shape[1]:r.shape[2],c=o==="NHWC"?r.shape[2]:r.shape[3],u=o==="NHWC"?r.shape[3]:r.shape[1],d=l*a,p=c*a,h=u/(a*a),f=o==="NHWC"?[i,d,p,h]:[i,h,d,p],m=t.makeOutput(f,"float32"),A=t.dataIdMap.get(r.dataId).id,x=new Uint8Array(new Int32Array(v.computeStrides(r.shape)).buffer),y=new Uint8Array(new Int32Array(f).buffer),b=new Uint8Array(new Int32Array(v.computeStrides(f)).buffer),w=t.dataIdMap.get(m.dataId).id;return ZC(A,a,o==="NHWC"?1:0,x,r.shape.length-1,y,b,f.length,w),m}var pfe={kernelName:mi,backendName:"wasm",setupFunc:cfe,kernelFunc:dfe},YC;function hfe(e){YC=e.wasm.cwrap(Oa,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function ffe(e){let{inputs:t,attrs:n,backend:s}=e,{x:r,filter:a}=t,o=s.dataIdMap.get(r.dataId).id,i=s.dataIdMap.get(a.dataId).id,{strides:l,dilations:c,pad:u,dimRoundingMode:d}=n,p=c==null?[1,1]:c,h=E.computeConv2DInfo(r.shape,a.shape,l,p,u,d,!0),f=h.filterHeight,m=h.filterWidth,g=h.padInfo.top,A=h.padInfo.right,x=h.padInfo.bottom,y=h.padInfo.left,b=h.dilationHeight,w=h.dilationWidth,k=h.strideHeight,C=h.strideWidth,N=h.inChannels,R=h.outChannels,F=h.padInfo.type==="SAME"?1:0;if(h.dataFormat!=="channelsLast")throw new Error(`wasm backend DepthwiseConv2dNative does not support dataFormat:'${h.dataFormat}'. Please use 'channelsLast'.`);let _=s.makeOutput(h.outShape,"float32"),P=s.dataIdMap.get(_.dataId).id;return YC(o,r.shape[0],r.shape[1],r.shape[2],i,f,m,g,A,x,y,F,b,w,k,C,N,R,P),_}var mfe={kernelName:Oa,backendName:"wasm",setupFunc:hfe,kernelFunc:ffe},gfe=Nn(za),Afe=!1,yfe=qn(gi,Afe,"bool"),xfe=Nn(La,"float32");function Nx(e){let{inputs:t,attrs:n,backend:s}=e,{input:r}=t,{dim:a}=n,o=r.shape.length,i=r.shape.slice(),l=a;return a<0&&(v.assert(-(o+1)<=a,()=>`Axis must be in the interval [${-(o+1)}, ${o}]`),l=o+a+1),i.splice(l,0,1),is({inputs:{x:r},backend:s,attrs:{shape:i}})}var bfe={kernelName:Ai,backendName:"wasm",kernelFunc:Nx};function JC(e){let{attrs:{shape:t,value:n,dtype:s},backend:r}=e,a=r.makeOutput(t,s);return r.typedArrayFromHeap(a).fill(n),a}var vfe={kernelName:du,backendName:"wasm",kernelFunc:JC},QC;function wfe(e){QC=e.wasm.cwrap(xi,null,["number","number","number","number","number","number"])}function kfe(e){let{inputs:t,backend:n}=e,{image:s}=t,r=n.makeOutput(s.shape,s.dtype),a=n.dataIdMap.get(s.dataId).id,o=n.dataIdMap.get(r.dataId).id,[i,l,c,u]=s.shape;return QC(a,i,l,c,u,o),r}var Sfe={kernelName:xi,backendName:"wasm",kernelFunc:kfe,setupFunc:wfe},Ife=Nn(Ba),Cfe=!1,Tfe=qn(Wa,Cfe),e6;function Nfe(e){e6=e.wasm.cwrap(Va,null,["number","number","number","number","number","number","number"])}function Efe(e){let{backend:t,inputs:n,attrs:s}=e,{varianceEpsilon:r}=s,{x:a,mean:o,variance:i,offset:l,scale:c}=n,u=t.dataIdMap.get(a.dataId).id,d=t.dataIdMap.get(o.dataId).id,p=t.dataIdMap.get(i.dataId).id,h=l!=null?t.dataIdMap.get(l.dataId).id:0,f=c!=null?t.dataIdMap.get(c.dataId).id:0,m=t.makeOutput(a.shape,a.dtype);if(v.sizeFromShape(a.shape)===0)return m;let g=t.dataIdMap.get(m.dataId).id;return e6(u,d,p,h,f,r,g),m}var Rfe={kernelName:Va,backendName:"wasm",setupFunc:Nfe,kernelFunc:Efe},t6;function $fe(e){t6=e.wasm.cwrap(xo,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function _fe(e){let{inputs:t,attrs:n,backend:s}=e,{x:r,filter:a,bias:o,preluActivationWeights:i}=t,{strides:l,pad:c,dilations:u,dataFormat:d,dimRoundingMode:p,activation:h,leakyreluAlpha:f}=n,m=E.computeConv2DInfo(r.shape,a.shape,l,u,c,p),g=bp[h];if(g==null)throw new Error(`${h} activation not yet supported for FusedConv2D in the wasm backend.`);let A=s.dataIdMap.get(r.dataId).id,x=s.dataIdMap.get(a.dataId).id,y=m.outChannels,b=0;if(o!=null){let Q=s.dataIdMap.get(o.dataId);if(Q.shape.length!==1)throw new Error(`FusedConv2D only supports rank-1 bias but got rank ${Q.shape.length}.`);if(Q.shape[0]!==y)throw new Error(`FusedConv2D bias shape (${Q.shape}) does not match the number of output channels (${y})`);b=Q.id}let w=m.filterHeight,k=m.filterWidth,C=m.padInfo.top,N=m.padInfo.right,R=m.padInfo.bottom,F=m.padInfo.left,_=m.dilationHeight,P=m.dilationWidth,T=m.strideHeight,M=m.strideWidth,U=m.inChannels,j=m.padInfo.type==="SAME"?1:0,z=m.batchSize,X=m.inHeight,Z=m.inWidth;if(d!=="NHWC")throw new Error(`wasm backend FusedConv2D does not support dataFormat:'${d}'. Please use 'NHWC'.`);let J=s.makeOutput(m.outShape,"float32"),ee=s.dataIdMap.get(J.dataId).id,ne=i==null?0:s.dataIdMap.get(i.dataId).id;return t6(A,z,X,Z,x,w,k,b,C,N,R,F,j,_,P,T,M,U,y,g,ne,f||0,ee),J}var Dfe={kernelName:xo,backendName:"wasm",setupFunc:$fe,kernelFunc:_fe},n6;function Pfe(e){n6=e.wasm.cwrap(bo,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function Ffe(e){let{inputs:t,attrs:n,backend:s}=e,{x:r,filter:a,bias:o,preluActivationWeights:i}=t,{strides:l,pad:c,dilations:u,dataFormat:d,dimRoundingMode:p,activation:h,leakyreluAlpha:f}=n,m=E.computeConv2DInfo(r.shape,a.shape,l,u,c,p,!0),g=bp[h];if(g==null)throw new Error(`${h} activation not yet supported for FusedDepthwiseConv2D in the wasm backend.`);let A=s.dataIdMap.get(r.dataId).id,x=s.dataIdMap.get(a.dataId).id,y=m.outChannels,b=0;if(o!=null){let Q=s.dataIdMap.get(o.dataId);if(Q.shape.length!==1)throw new Error(`FusedDepthwiseConv2D only supports rank-1 bias but got rank ${Q.shape.length}.`);if(Q.shape[0]!==y)throw new Error(`FusedDepthwiseConv2D bias shape (${Q.shape}) does not match the number of output channels (${y})`);b=Q.id}let w=m.filterHeight,k=m.filterWidth,C=m.padInfo.top,N=m.padInfo.right,R=m.padInfo.bottom,F=m.padInfo.left,_=m.dilationHeight,P=m.dilationWidth,T=m.strideHeight,M=m.strideWidth,U=m.inChannels,j=m.padInfo.type==="SAME"?1:0,z=m.batchSize,X=m.inHeight,Z=m.inWidth;if(d!=="NHWC")throw new Error(`wasm backend FusedDepthwiseConv2D does not support dataFormat:'${d}'. Please use 'NHWC'.`);let J=s.makeOutput(m.outShape,"float32"),ee=s.dataIdMap.get(J.dataId).id,ne=i==null?0:s.dataIdMap.get(i.dataId).id;return n6(A,z,X,Z,x,w,k,b,C,N,R,F,j,_,P,T,M,U,y,g,ne,f||0,ee),J}var Ofe={kernelName:bo,backendName:"wasm",setupFunc:Pfe,kernelFunc:Ffe},s6;function Mfe(e){s6=e.wasm.cwrap(vi,null,["number","number","number","number","number","number","array","number"])}function zfe(e){let{backend:t,inputs:n}=e,{params:s,indices:r}=n,[a,o,i,l]=R2.prepareAndValidate(s,r),c=t.makeOutput(a,s.dtype);if(o===0)return c;let u=r.shape,d=u[u.length-1],h=t.dataIdMap.get(s.dataId).id,m=t.dataIdMap.get(r.dataId).id,g=new Uint8Array(new Int32Array(l).buffer),A=t.dataIdMap.get(c.dataId).id;return s6(h,Yt[s.dtype],m,o,d,i,g,A),c}var Lfe={kernelName:vi,backendName:"wasm",setupFunc:Mfe,kernelFunc:zfe},r6;function Bfe(e){r6=e.wasm.cwrap("Gather",null,["number","number","array","number","number","number","array","number"])}function Wfe(e){let{backend:t,inputs:n,attrs:s}=e,{x:r,indices:a}=n,{axis:o,batchDims:i}=s,l=v.parseAxisParam(o,r.shape)[0],c=t.readSync(a.dataId),u=r.shape[l];for(let R=0;R<c.length;++R){let F=c[R];v.assert(F<=u-1&&F>=0,()=>`GatherV2: the index value ${F} is not in [0, ${u-1}]`)}let d=E.segment_util.collectGatherOpShapeInfo(r,a,l,i),p=is({inputs:{x:r},attrs:{shape:[d.batchSize,d.outerSize,d.dimSize,d.sliceSize]},backend:t}),h=v.sizeFromShape(a.shape),f=is({inputs:{x:a},attrs:{shape:[d.batchSize,h/d.batchSize]},backend:t}),m=[d.batchSize,d.outerSize,h/d.batchSize,d.sliceSize],g=t.makeOutput(m,r.dtype);if(v.sizeFromShape(r.shape)===0)return g;let A=p.shape.length-1,y=t.dataIdMap.get(p.dataId).id,w=t.dataIdMap.get(f.dataId).id,k=t.dataIdMap.get(g.dataId).id,C=new Uint8Array(new Int32Array(v.computeStrides(p.shape)).buffer),N=new Uint8Array(new Int32Array(v.computeStrides(m)).buffer);return r6(y,Yt[r.dtype],C,A,w,d.batchSize,N,k),t.disposeData(p.dataId),t.disposeData(f.dataId),g.shape=d.outputShape,g}var Vfe={kernelName:bi,backendName:"wasm",setupFunc:Bfe,kernelFunc:Wfe},Ufe=!1,Gfe=qn(wi,Ufe,"bool"),Hfe=!1,jfe=qn(Ua,Hfe,"bool"),a6;function qfe(e){a6=e.wasm.cwrap(ki,null,["number","number","number","number"])}function Xfe(e){let{inputs:{x:t},attrs:{alpha:n},backend:s}=e,r=s.dataIdMap.get(t.dataId).id,a=s.makeOutput(t.shape,"float32");if(v.sizeFromShape(t.shape)!==0){let o=s.dataIdMap.get(a.dataId).id;a6(r,Yt[t.dtype],n,o)}return a}var Kfe={kernelName:ki,backendName:"wasm",setupFunc:qfe,kernelFunc:Xfe},Zfe=!1,Yfe=qn(Si,Zfe,"bool"),Jfe=!1,Qfe=qn(Ii,Jfe,"bool"),eme=Nn(Ha),tme=!1,nme=qn(Ci,tme,"bool"),o6;function sme(e){o6=e.wasm.cwrap(ja,null,["number","number","number","number"])}function rme(e){let{backend:t,inputs:n,attrs:s}=e,{reductionIndices:r,keepDims:a}=s,{x:o}=n,l=t.dataIdMap.get(o.dataId).id,c=o,{transposed:u,axes:d,originalAxes:p,inputWasTransposed:h}=Ho(o,r,t);if(h){let y=t.dataIdMap.get(u.dataId).id;c=u,l=y}let f=c.shape.length;E.assertAxesAreInnerMostDims("max",d,f);let[m,g]=E.computeOutAndReduceShapes(c.shape,d),A=v.sizeFromShape(g),x=t.makeOutput(m,o.dtype);if(v.sizeFromShape(c.shape)!==0){let y=t.dataIdMap.get(x.dataId).id;o6(l,Yt[o.dtype],A,y)}if(h&&t.disposeData(u.dataId),a){let y=E.expandShapeToKeepDim(x.shape,p);x.shape=y}return x}var ame={kernelName:ja,backendName:"wasm",setupFunc:sme,kernelFunc:rme},ome=!1,ime=qn(qa,ome),i6;function lme(e){i6=e.wasm.cwrap(Xa,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function ume(e){let{inputs:t,attrs:n,backend:s}=e,r=t.x,a=s.dataIdMap.get(r.dataId).id;v.assert(r.dtype==="float32",()=>`Error in MaxPool: only float32 input is supported. Got ${r.dtype}.`);let{filterSize:o,strides:i,pad:l,dimRoundingMode:c}=n,u=E.computePool2DInfo(r.shape,o,i,1,l,c),d=u.filterHeight,p=u.filterWidth,h=u.padInfo.top,f=u.padInfo.right,m=u.padInfo.bottom,g=u.padInfo.left,A=u.dilationHeight,x=u.dilationWidth,y=u.strideHeight,b=u.strideWidth,w=u.inChannels,k=u.outChannels;if(u.dataFormat!=="channelsLast")throw new Error(`wasm backend does not support dataFormat:'${u.dataFormat}'. Please use 'channelsLast'.`);let C=s.makeOutput(u.outShape,"float32"),N=s.dataIdMap.get(C.dataId).id;return i6(a,r.shape[0],r.shape[1],r.shape[2],d,p,h,f,m,g,A,x,y,b,w,k,N),C}var cme={kernelName:Xa,backendName:"wasm",setupFunc:lme,kernelFunc:ume},l6;function dme(e){l6=e.wasm.cwrap(Ka,null,["number, number, number"])}function pme(e){let{backend:t,inputs:n,attrs:s}=e,{axis:r,keepDims:a}=s,{x:o}=n,i=t.dataIdMap.get(o.dataId).id,l=i,c=o,{transposed:u,axes:d,originalAxes:p,inputWasTransposed:h}=Ho(o,r,t),f=d;if(h){let b=t.dataIdMap.get(u.dataId).id;b!==i&&(c=u,l=b,f=E.getInnerMostAxes(f.length,c.shape.length))}E.assertAxesAreInnerMostDims("mean",f,c.shape.length);let[m,g]=E.computeOutAndReduceShapes(c.shape,f),A=v.sizeFromShape(g),x=c;c.dtype!=="float32"&&(x=wp({backend:t,inputs:{x:c},attrs:{dtype:"float32"}}),l=t.dataIdMap.get(x.dataId).id);let y=t.makeOutput(m,"float32");if(v.sizeFromShape(c.shape)!==0){let b=t.dataIdMap.get(y.dataId).id;l6(l,A,b)}if(h&&t.disposeData(u.dataId),a){let b=E.expandShapeToKeepDim(y.shape,p);y.shape=b}return c.dtype!=="float32"&&t.disposeData(x.dataId),y}var hme={kernelName:Ka,backendName:"wasm",setupFunc:dme,kernelFunc:pme},u6;function fme(e){u6=e.wasm.cwrap(Za,null,["number","number","number","number"])}function mme(e){let{backend:t,inputs:n,attrs:s}=e,{axis:r,keepDims:a}=s,{x:o}=n,i=t.dataIdMap.get(o.dataId).id,l=i,c=o,{transposed:u,axes:d,originalAxes:p,inputWasTransposed:h}=Ho(o,r,t);if(h){let y=t.dataIdMap.get(u.dataId).id;y!==i&&(c=u,l=y)}let f=c.shape.length;E.assertAxesAreInnerMostDims("min",d,f);let[m,g]=E.computeOutAndReduceShapes(c.shape,d),A=v.sizeFromShape(g),x=t.makeOutput(m,c.dtype);if(v.sizeFromShape(c.shape)!==0){let y=t.dataIdMap.get(x.dataId).id;u6(l,Yt[o.dtype],A,y)}if(h&&t.disposeData(u.dataId),a){let y=E.expandShapeToKeepDim(x.shape,p);x.shape=y}return x}var gme={kernelName:Za,backendName:"wasm",setupFunc:fme,kernelFunc:mme},Ame=!1,yme=qn(Ya,Ame),Ex;(function(e){e[e.reflect=0]="reflect",e[e.symmetric=1]="symmetric"})(Ex||(Ex={}));var c6;function xme(e){c6=e.wasm.cwrap(Ja,null,["number","array","number","number","array","array","number","number"])}function bme(e){let{inputs:{x:t},backend:n,attrs:{paddings:s,mode:r}}=e,a=s.map((f,m)=>f[0]+t.shape[m]+f[1]),o=n.dataIdMap.get(t.dataId).id,i=n.makeOutput(a,t.dtype),l=n.dataIdMap.get(i.dataId).id,c=new Uint8Array(new Int32Array(t.shape).buffer),u=s.map(f=>f[0]),d=s.map(f=>f[1]),p=new Uint8Array(new Int32Array(u).buffer),h=new Uint8Array(new Int32Array(d).buffer);return c6(o,c,t.shape.length,Yt[t.dtype],p,h,Ex[r],l),i}var vme={kernelName:Ja,backendName:"wasm",kernelFunc:bme,setupFunc:xme},wme=!0,kme=qn(Qa,wme),Sme=Nn(Ti);function Rx(e,t){let n=new Int32Array(e.wasm.HEAPU8.buffer,t,4),s=n[0],r=n[1],a=n[2],o=n[3];return e.wasm._free(t),{pSelectedIndices:s,selectedSize:r,pSelectedScores:a,pValidOutputs:o}}var d6;function Ime(e){d6=e.wasm.cwrap(Ei,"number",["number","number","number","number","number"])}function Cme(e){let{backend:t,inputs:n,attrs:s}=e,{iouThreshold:r,maxOutputSize:a,scoreThreshold:o}=s,{boxes:i,scores:l}=n,c=t.dataIdMap.get(i.dataId).id,u=t.dataIdMap.get(l.dataId).id,d=d6(c,u,a,r,o),{pSelectedIndices:p,selectedSize:h,pSelectedScores:f,pValidOutputs:m}=Rx(t,d);return t.wasm._free(f),t.wasm._free(m),t.makeOutput([h],"int32",p)}var Tme={kernelName:Ei,backendName:"wasm",setupFunc:Ime,kernelFunc:Cme},p6;function Nme(e){p6=e.wasm.cwrap(yu,"number",["number","number","number","number","number","bool"])}function Eme(e){let{backend:t,inputs:n,attrs:s}=e,{iouThreshold:r,maxOutputSize:a,scoreThreshold:o,padToMaxOutputSize:i}=s,{boxes:l,scores:c}=n,u=t.dataIdMap.get(l.dataId).id,d=t.dataIdMap.get(c.dataId).id,p=p6(u,d,a,r,o,i),{pSelectedIndices:h,selectedSize:f,pSelectedScores:m,pValidOutputs:g}=Rx(t,p);t.wasm._free(m);let A=t.makeOutput([f],"int32",h),x=t.makeOutput([],"int32",g);return[A,x]}var Rme={kernelName:yu,backendName:"wasm",setupFunc:Nme,kernelFunc:Eme},h6;function $me(e){h6=e.wasm.cwrap(Ri,"number",["number","number","number","number","number","number"])}function _me(e){let{backend:t,inputs:n,attrs:s}=e,{iouThreshold:r,maxOutputSize:a,scoreThreshold:o,softNmsSigma:i}=s,{boxes:l,scores:c}=n,u=t.dataIdMap.get(l.dataId).id,d=t.dataIdMap.get(c.dataId).id,p=h6(u,d,a,r,o,i),{pSelectedIndices:h,selectedSize:f,pSelectedScores:m,pValidOutputs:g}=Rx(t,p);t.wasm._free(g);let A=t.makeOutput([f],"int32",h),x=t.makeOutput([f],"float32",m);return[A,x]}var Dme={kernelName:Ri,backendName:"wasm",setupFunc:$me,kernelFunc:_me},Pme=!1,Fme=qn(Ni,Pme,"bool"),f6;function Ome(e){f6=e.wasm.cwrap(_i,null,["number","number","number","number","number"])}function Mme(e){let{inputs:t,backend:n,attrs:s}=e,{indices:r}=t,{depth:a,onValue:o,offValue:i}=s,l=n.makeOutput([...r.shape,a],"int32"),c=n.dataIdMap.get(l.dataId).id,d=n.dataIdMap.get(r.dataId).id;return f6(d,a,o,i,c),l}var zme={kernelName:_i,backendName:"wasm",setupFunc:Ome,kernelFunc:Mme};function Lme(e){let{inputs:{x:t},backend:n}=e,s=n.makeOutput(t.shape,t.dtype);return n.typedArrayFromHeap(s).fill(1),s}var Bme={kernelName:$i,backendName:"wasm",kernelFunc:Lme};function Wme(e){let{inputs:t,backend:n,attrs:s}=e,{axis:r}=s;if(t.length===1)return Nx({inputs:{input:t[0]},backend:n,attrs:{dim:r}});let a=t[0].shape,o=t[0].dtype;t.forEach(u=>{v.assertShapesMatch(a,u.shape,"All tensors passed to stack must have matching shapes"),v.assert(o===u.dtype,()=>"All tensors passed to stack must have matching dtypes")});let i=[],l=t.map(u=>{let d=Nx({inputs:{input:u},backend:n,attrs:{dim:r}});return i.push(d),d}),c=HC({inputs:l,backend:n,attrs:{axis:r}});return i.forEach(u=>n.disposeData(u.dataId)),c}var Vme={kernelName:Di,backendName:"wasm",kernelFunc:Wme},m6;function Ume(e){m6=e.wasm.cwrap(eo,null,["number","array","number","number","array","array","number","number"])}function Gme(e){let{inputs:{x:t},backend:n,attrs:{paddings:s,constantValue:r}}=e,a=s.map((m,g)=>m[0]+t.shape[g]+m[1]);if(v.sizeFromShape(t.shape)===0)return JC({backend:n,attrs:{shape:a,value:r,dtype:t.dtype}});let o=n.dataIdMap.get(t.dataId).id,i=n.makeOutput(a,t.dtype),c=n.dataIdMap.get(i.dataId).id,u=new Uint8Array(new Int32Array(t.shape).buffer),d=s.map(m=>m[0]),p=s.map(m=>m[1]),h=new Uint8Array(new Int32Array(d).buffer),f=new Uint8Array(new Int32Array(p).buffer);return m6(o,u,t.shape.length,Yt[t.dtype],h,f,r,c),i}var g6={kernelName:eo,backendName:"wasm",kernelFunc:Gme,setupFunc:Ume},Hme=!1,jme=qn(to,Hme),A6;function qme(e){A6=e.wasm.cwrap(no,null,["number","number","number"])}function Xme(e){let{inputs:t,backend:n}=e,{x:s,alpha:r}=t,a=n.dataIdMap.get(s.dataId).id,o=n.dataIdMap.get(r.dataId).id,i=a,l=s,c=l;l.dtype!=="float32"&&(c=wp({backend:n,inputs:{x:s},attrs:{dtype:"float32"}}),i=n.dataIdMap.get(c.dataId).id);let u=n.makeOutput(s.shape,"float32"),d=n.dataIdMap.get(u.dataId).id;return A6(i,o,d),l.dtype!=="float32"&&n.disposeData(c.dataId),u}var Kme={kernelName:no,backendName:"wasm",setupFunc:qme,kernelFunc:Xme},y6;function Zme(e){y6=e.wasm.cwrap(Pi,null,["number","number","number","number"])}function Yme(e){let{backend:t,inputs:n,attrs:s}=e,{axis:r,keepDims:a}=s,{x:o}=n,i=t.dataIdMap.get(o.dataId).id,l=i,c=o,{transposed:u,axes:d,originalAxes:p,inputWasTransposed:h}=Ho(o,r,t),f=d;if(h){let y=t.dataIdMap.get(u.dataId).id;y!==i&&(c=u,l=y,f=E.getInnerMostAxes(f.length,c.shape.length))}E.assertAxesAreInnerMostDims("prod",f,c.shape.length);let[m,g]=E.computeOutAndReduceShapes(c.shape,f),A=v.sizeFromShape(g),x=t.makeOutput(m,c.dtype);if(v.sizeFromShape(c.shape)!==0){let y=t.dataIdMap.get(x.dataId).id;y6(l,A,Yt[x.dtype],y)}if(h&&t.disposeData(u.dataId),a){let y=E.expandShapeToKeepDim(x.shape,p);x.shape=y}return x}var Jme={kernelName:Pi,backendName:"wasm",setupFunc:Zme,kernelFunc:Yme},Qme=e=>{let{backend:t,attrs:n}=e,{start:s,stop:r,step:a,dtype:o}=n,i=zy(s,r,a,o),l=t.makeOutput([i.length],o);return t.typedArrayFromHeap(l).set(i),l},e0e={kernelName:xu,backendName:"wasm",kernelFunc:Qme},t0e=!0,n0e=qn(Ma,t0e),s0e=Nn(so),r0e=Nn(ao),x6;function a0e(e){x6=e.wasm.cwrap(ro,null,["number","number","number","number","number","number","number","number","number","number"])}function o0e(e){let{backend:t,inputs:n,attrs:s}=e,{images:r}=n,{alignCorners:a,halfPixelCenters:o,size:i}=s,[l,c]=i,[u,d,p,h]=r.shape,f=[u,l,c,h],m=t.dataIdMap.get(r.dataId),g;m.dtype!=="float32"&&(g=wp({backend:t,inputs:{x:r},attrs:{dtype:"float32"}}),m=t.dataIdMap.get(g.dataId));let A=m.id,x=t.makeOutput(f,"float32");if(v.sizeFromShape(r.shape)===0)return x;let y=t.dataIdMap.get(x.dataId).id;return x6(A,u,d,p,h,l,c,a?1:0,o?1:0,y),g!=null&&t.disposeData(g.dataId),x}var i0e={kernelName:ro,backendName:"wasm",setupFunc:a0e,kernelFunc:o0e},b6;function l0e(e){b6=e.wasm.cwrap(Oi,null,["number","array","number","array","number","number"])}function u0e(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{dims:a}=s,o=v.parseAxisParam(a,r.shape);if(r.shape.length===0)return Jm({inputs:{x:r},backend:n});let i=n.makeOutput(r.shape,r.dtype),l=n.dataIdMap.get(r.dataId).id,c=n.dataIdMap.get(i.dataId).id,u=new Uint8Array(new Int32Array(o).buffer),d=new Uint8Array(new Int32Array(r.shape).buffer);b6(l,u,o.length,d,r.shape.length,c);let p=is({inputs:{x:i},attrs:{shape:r.shape},backend:n});return n.disposeData(i.dataId),p}var c0e={kernelName:Oi,backendName:"wasm",kernelFunc:u0e,setupFunc:l0e},v6;function d0e(e){v6=e.wasm.cwrap(Zi,null,["number","number","number","number","number","number","number","number","array","number","number"])}function p0e(e){let{inputs:t,backend:n,attrs:s}=e,{image:r}=t,{radians:a,fillValue:o,center:i}=s,l=n.makeOutput(r.shape,r.dtype),c=n.dataIdMap.get(r.dataId).id,u=n.dataIdMap.get(l.dataId).id,[d,p,h,f]=r.shape,[m,g]=E.getImageCenter(i,p,h),A=o===0,x=255,y=typeof o=="number"?[o,o,o,A?0:x]:[...o,x],b=new Uint8Array(new Int32Array(y).buffer);return v6(c,d,p,h,f,a,m,g,b,y.length,u),l}var h0e={kernelName:Zi,backendName:"wasm",kernelFunc:p0e,setupFunc:d0e},f0e=Nn(Mi),m0e=Nn(oo),w6;function g0e(e){w6=e.wasm.cwrap(zi,null,["number","number","number","number","number","number","array","number","number"])}function A0e(e){let{backend:t,inputs:n,attrs:s}=e,{indices:r,updates:a}=n,{shape:o}=s,i=t.makeOutput(o,a.dtype);if(v.sizeFromShape(o)===0)return i;let{sliceRank:l,numUpdates:c,sliceSize:u,strides:d,outputSize:p}=$2.calculateShapes(a,r,o),f=t.dataIdMap.get(r.dataId).id,g=t.dataIdMap.get(a.dataId).id,A=new Uint8Array(new Int32Array(d).buffer),x=t.dataIdMap.get(i.dataId).id;return w6(f,g,Yt[a.dtype],l,c,u,A,p,x),i}var y0e={kernelName:zi,backendName:"wasm",setupFunc:g0e,kernelFunc:A0e},k6;function x0e(e){k6=e.wasm.cwrap("SelectV2",null,["number","number","number","number","number"])}function b0e(e){let{inputs:t,backend:n}=e,{condition:s,t:r,e:a}=t,o=n.dataIdMap.get(s.dataId).id,i=n.dataIdMap.get(r.dataId).id,l=n.dataIdMap.get(a.dataId).id,c=n.makeOutput(r.shape,r.dtype),u=n.dataIdMap.get(c.dataId).id,d=s.shape.length,p=r.shape.length,h=d===0||d>1||p===1?1:v.sizeFromShape(r.shape.slice(1));return k6(o,i,l,h,u),c}var v0e={kernelName:Li,backendName:"wasm",kernelFunc:b0e,setupFunc:x0e},S6;function w0e(e){S6=e.wasm.cwrap(lo,null,["number","number"])}function k0e(e){let{backend:t,inputs:{x:n}}=e,s=t.dataIdMap.get(n.dataId).id,r=t.makeOutput(n.shape,n.dtype),a=t.dataIdMap.get(r.dataId).id;return v.sizeFromShape(r.shape)===0||S6(s,a),r}var S0e={kernelName:"Sigmoid",backendName:"wasm",setupFunc:w0e,kernelFunc:k0e},I0e=Nn(io),I6;function C0e(e){I6=e.wasm.cwrap(po,null,["number","number","number","number"])}function T0e(e){let{backend:t,inputs:{logits:n},attrs:{dim:s}}=e,r=t.dataIdMap.get(n.dataId).id,a=t.makeOutput(n.shape,n.dtype),o=t.dataIdMap.get(a.dataId).id,i=n.shape[s],l=v.sizeFromShape(n.shape)/i;return v.sizeFromShape(a.shape)===0||I6(r,o,i,l),a}var N0e={kernelName:po,backendName:"wasm",setupFunc:C0e,kernelFunc:T0e};function E0e(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,{blockShape:a,paddings:o}=s,i=v.sizeFromShape(a),l=[[0,0]];l.push(...o);for(let k=1+a.length;k<r.shape.length;++k)l.push([0,0]);let c=g6.kernelFunc({inputs:{x:r},backend:n,attrs:{paddings:l,constantValue:0}}),u=E.getReshaped(c.shape,a,i,!1),d=E.getPermuted(u.length,a.length,!1),p=E.getReshapedPermuted(c.shape,a,i,!1),m=is({inputs:{x:c},backend:n,attrs:{shape:u}}),x=xc({inputs:{x:m},backend:n,attrs:{perm:d}}),w=is({inputs:{x},backend:n,attrs:{shape:p}});return n.disposeData(c.dataId),n.disposeData(m.dataId),n.disposeData(x.dataId),w}var R0e={kernelName:Vi,backendName:"wasm",kernelFunc:E0e};function $0e(e){let{inputs:t,attrs:n,backend:s}=e,{x:r}=t,{numOrSizeSplits:a,axis:o}=n,i=v.parseAxisParam(o,r.shape)[0],l=E.prepareSplitSize(r,a,i),c=new Array(r.shape.length).fill(0),u=r.shape.slice();return l.map(d=>{let p=[...u];p[i]=d;let h=vp({inputs:{x:r},attrs:{begin:c,size:p},backend:s});return c[i]+=d,h})}var _0e={kernelName:Ui,backendName:"wasm",kernelFunc:$0e},D0e=Nn(uo),P0e=Nn(Iu),F0e=!0,O0e=qn(ho,F0e),C6;function M0e(e){C6=e.wasm.cwrap(Ao,null,["number","number","number","number"])}function z0e(e){let{backend:t,inputs:n,attrs:s}=e,{alpha:r}=s,{x:a}=n,o=t.dataIdMap.get(a.dataId).id,i=t.makeOutput(a.shape,a.dtype),l=t.dataIdMap.get(i.dataId).id;return C6(o,r,Yt[a.dtype],l),i}var L0e={kernelName:Ao,backendName:"wasm",setupFunc:M0e,kernelFunc:z0e},T6;function B0e(e){T6=e.wasm.cwrap(Gi,null,["number","array","number","array","array","array","array","array","number","number"])}function W0e(e){let{backend:t,inputs:n,attrs:s}=e,{x:r}=n,{begin:a,end:o,strides:i,beginMask:l,endMask:c,ellipsisMask:u,newAxisMask:d,shrinkAxisMask:p}=s,{finalShapeSparse:h,finalShape:f,isIdentity:m,sliceDim0:g,isSimpleSlice:A,begin:x,end:y,strides:b}=Pt.sliceInfo(r.shape,a,o,i,l,c,u,d,p),w;if(m)w=is({inputs:{x:r},backend:t,attrs:{shape:f}});else if(g||A){v.assert(r.shape.length>=1,()=>`Input must have rank at least 1, got: ${r.shape.length}`);let k=Pt.computeOutShape(x,y,b),C=vp({inputs:{x:r},backend:t,attrs:{begin:x,size:k}});w=is({inputs:{x:C},backend:t,attrs:{shape:f}}),t.disposeData(C.dataId)}else{let k=t.makeOutput(h,"float32"),C=t.dataIdMap.get(r.dataId).id,N=new Uint8Array(new Int32Array(v.computeStrides(r.shape)).buffer),R=new Uint8Array(new Int32Array(x).buffer),F=new Uint8Array(new Int32Array(y).buffer),_=new Uint8Array(new Int32Array(b).buffer),P=new Uint8Array(new Int32Array(h).buffer),T=new Uint8Array(new Int32Array(v.computeStrides(h)).buffer),M=t.dataIdMap.get(k.dataId).id;T6(C,N,r.shape.length,R,F,_,P,T,h.length,M),w=is({inputs:{x:k},backend:t,attrs:{shape:f}}),t.disposeData(k.dataId)}return w}var V0e={kernelName:Gi,backendName:"wasm",setupFunc:B0e,kernelFunc:W0e},U0e=!0,G0e=qn(fo,U0e),N6;function H0e(e){N6=e.wasm.cwrap(co,null,["number","number","number","number"])}function j0e(e){let{backend:t,inputs:n,attrs:s}=e,{axis:r,keepDims:a}=s,{x:o}=n,i=t.dataIdMap.get(o.dataId).id,l=i,c=o,{transposed:u,axes:d,originalAxes:p,inputWasTransposed:h}=Ho(o,r,t),f=d;if(h){let y=t.dataIdMap.get(u.dataId).id;y!==i&&(c=u,l=y,f=E.getInnerMostAxes(f.length,c.shape.length))}E.assertAxesAreInnerMostDims("sum",f,c.shape.length);let[m,g]=E.computeOutAndReduceShapes(c.shape,f),A=v.sizeFromShape(g),x=t.makeOutput(m,c.dtype);if(v.sizeFromShape(c.shape)!==0){let y=t.dataIdMap.get(x.dataId).id;N6(l,A,Yt[x.dtype],y)}if(h&&t.disposeData(u.dataId),a){let y=E.expandShapeToKeepDim(x.shape,p);x.shape=y}return x}var q0e={kernelName:co,backendName:"wasm",setupFunc:H0e,kernelFunc:j0e},X0e=Nn(Hi),K0e=Nn(mo),E6;function Z0e(e){E6=e.wasm.cwrap(jr,null,["number","array","number","array","number","number"])}function Y0e(e){let{inputs:t,backend:n,attrs:s}=e,{x:r}=t,a=n.dataIdMap.get(r.dataId).id,{reps:o}=s,i=new Array(r.shape.length);for(let p=0;p<i.length;p++)i[p]=r.shape[p]*o[p];let l=new Uint8Array(new Int32Array(r.shape).buffer),c=new Uint8Array(new Int32Array(i).buffer),u=n.makeOutput(i,r.dtype),d=n.dataIdMap.get(u.dataId).id;return E6(a,l,r.shape.length,c,i.length,Yt[u.dtype],d),u}var J0e={kernelName:jr,backendName:"wasm",setupFunc:Z0e,kernelFunc:Y0e},R6;function Q0e(e){R6=e.wasm.cwrap(ji,null,["number","array","number","number","number","bool","number","number"])}var ege=({inputs:e,backend:t,attrs:n})=>{let{x:s}=e,{k:r,sorted:a}=n,o=t.dataIdMap.get(s.dataId).id,i=new Uint8Array(new Int32Array(s.shape).buffer),l=s.shape.slice();l[l.length-1]=r;let c=t.makeOutput(l,s.dtype),u=t.dataIdMap.get(c.dataId).id,d=t.makeOutput(l,"int32"),p=t.dataIdMap.get(d.dataId).id;return R6(o,i,s.shape.length,Yt[s.dtype],r,a,u,p),[c,d]},tge={kernelName:ji,backendName:"wasm",setupFunc:Q0e,kernelFunc:ege},$6;function nge(e){$6=e.wasm.cwrap(qi,null,["number","number","bool","number","number","number","number","number","number","array","number","number","number","number","number"])}function sge(e){let{backend:t,inputs:n,attrs:s}=e,{image:r,transforms:a}=n,{interpolation:o,fillMode:i,fillValue:l,outputShape:c}=s,[u,d,p,h]=r.shape,[f,m]=c!=null?c:[d,p],g=[u,f,m,h],A=new Uint8Array(new Int32Array(v.computeStrides(r.shape)).buffer),x=t.makeOutput(g,r.dtype),y=t.dataIdMap.get(x.dataId).id,w=t.dataIdMap.get(r.dataId).id,C=t.dataIdMap.get(a.dataId).id,N=o==="nearest"?1:2,R;switch(i){case"constant":R=1;break;case"reflect":R=2;break;case"wrap":R=3;break;case"nearest":R=4;break;default:R=1;break}return $6(w,C,a.shape[0]>1,u,f,m,h,p,d,A,r.shape.length-1,N,R,l,y),x}var rge={kernelName:qi,backendName:"wasm",setupFunc:nge,kernelFunc:sge};function age(e){let{inputs:t,backend:n,attrs:s}=e,{value:r}=t,{axis:a}=s;a<0&&(a+=r.shape.length);let o=r.shape[a],i=r.shape.length,l=new Array(i-1),c=0;for(let h=0;h<i;h++)h!==a&&(l[c++]=r.shape[h]);let u=new Array(o),d=new Array(i).fill(0),p=r.shape.slice();p[a]=1;for(let h=0;h<u.length;h++)d[a]=h,u[h]=vp({inputs:{x:r},attrs:{begin:d,size:p},backend:n});return u.map(({dataId:h,dtype:f})=>({dataId:h,dtype:f,shape:l}))}var oge={kernelName:Xi,backendName:"wasm",kernelFunc:age};function ige(e){let{inputs:{x:t},backend:n}=e,s=n.makeOutput(t.shape,t.dtype);return n.typedArrayFromHeap(s).fill(0),s}var lge={kernelName:Ki,backendName:"wasm",kernelFunc:ige},uge=[dhe,hhe,ghe,She,The,Rhe,Dhe,Mhe,Uhe,Ghe,Hhe,Xhe,Khe,Jhe,tfe,nfe,sfe,ofe,ufe,pfe,mfe,gfe,yfe,xfe,bfe,vfe,Sfe,Ife,Tfe,che,Rfe,Dfe,Ofe,Lfe,Vfe,Gfe,jfe,Ahe,Kfe,Yfe,Qfe,eme,nme,ame,ime,cme,hme,gme,yme,vme,kme,Sme,Tme,Rme,Dme,Fme,zme,Bme,Vme,g6,jme,Kme,Jme,e0e,n0e,s0e,r0e,Phe,i0e,c0e,h0e,m0e,f0e,y0e,v0e,S0e,I0e,Whe,N0e,R0e,_0e,D0e,P0e,O0e,L0e,V0e,G0e,q0e,X0e,K0e,J0e,tge,rge,vhe,oge,lge];for(let e of uge)rr(e);var $x=K();$x.registerFlag("WASM_HAS_SIMD_SUPPORT",async()=>WebAssembly.validate(new Uint8Array([0,97,115,109,1,0,0,0,1,4,1,96,0,0,3,2,1,0,10,9,1,7,0,65,0,253,15,26,11])));$x.registerFlag("WASM_HAS_MULTITHREAD_SUPPORT",async()=>{if($x.get("IS_NODE"))return!1;try{return new MessageChannel().port1.postMessage(new SharedArrayBuffer(1)),WebAssembly.validate(new Uint8Array([0,97,115,109,1,0,0,0,1,4,1,96,0,0,3,2,1,0,5,4,1,3,1,1,10,11,1,9,0,65,0,254,16,2,0,26,11]))}catch(e){return!1}});var _6=ii(rE()),cge='var Module={};function threadPrintErr(){var text=Array.prototype.slice.call(arguments).join(" ");console.error(text)}function threadAlert(){var text=Array.prototype.slice.call(arguments).join(" ");postMessage({cmd:"alert",text:text,threadId:Module["_pthread_self"]()})}var err=threadPrintErr;this.alert=threadAlert;Module["instantiateWasm"]=function(info,receiveInstance){var instance=new WebAssembly.Instance(Module["wasmModule"],info);Module["wasmModule"]=null;receiveInstance(instance);return instance.exports};function moduleLoaded(){}this.onmessage=function(e){try{if(e.data.cmd==="load"){Module["wasmModule"]=e.data.wasmModule;Module["wasmMemory"]=e.data.wasmMemory;Module["buffer"]=Module["wasmMemory"].buffer;Module["ENVIRONMENT_IS_PTHREAD"]=true;if(typeof e.data.urlOrBlob==="string"){importScripts(e.data.urlOrBlob)}else{var objectUrl=URL.createObjectURL(e.data.urlOrBlob);importScripts(objectUrl);URL.revokeObjectURL(objectUrl)}WasmBackendModuleThreadedSimd(Module).then(function(instance){Module=instance;moduleLoaded()})}else if(e.data.cmd==="objectTransfer"){Module["PThread"].receiveObjectTransfer(e.data)}else if(e.data.cmd==="run"){Module["__performance_now_clock_drift"]=performance.now()-e.data.time;Module["__emscripten_thread_init"](e.data.threadInfoStruct,0,0);var max=e.data.stackBase;var top=e.data.stackBase+e.data.stackSize;Module["establishStackSpace"](top,max);Module["_emscripten_tls_init"]();Module["PThread"].receiveObjectTransfer(e.data);Module["PThread"].setThreadStatus(Module["_pthread_self"](),1);try{var result=Module["invokeEntryPoint"](e.data.start_routine,e.data.arg);if(!Module["getNoExitRuntime"]())Module["PThread"].threadExit(result)}catch(ex){if(ex==="Canceled!"){Module["PThread"].threadCancel()}else if(ex!="unwind"){if(ex instanceof Module["ExitStatus"]){if(Module["getNoExitRuntime"]()){}else{Module["PThread"].threadExit(ex.status)}}else{Module["PThread"].threadExit(-2);throw ex}}}}else if(e.data.cmd==="cancel"){if(Module["_pthread_self"]()){Module["PThread"].threadCancel()}}else if(e.data.target==="setimmediate"){}else if(e.data.cmd==="processThreadQueue"){if(Module["_pthread_self"]()){Module["_emscripten_current_thread_process_queued_calls"]()}}else{err("worker.js received unknown command "+e.data.cmd);err(e.data)}}catch(ex){err("worker.js onmessage() captured an uncaught exception: "+ex);if(ex&&ex.stack)err(ex.stack);throw ex}};if(typeof process==="object"&&typeof process.versions==="object"&&typeof process.versions.node==="string"){self={location:{href:__filename}};var onmessage=this.onmessage;var nodeWorkerThreads=require("worker_threads");global.Worker=nodeWorkerThreads.Worker;var parentPort=nodeWorkerThreads.parentPort;parentPort.on("message",function(data){onmessage({data:data})});var nodeFS=require("fs");var nodeRead=function(filename){return nodeFS.readFileSync(filename,"utf8")};function globalEval(x){global.require=require;global.Module=Module;eval.call(null,x)}importScripts=function(f){globalEval(nodeRead(f))};postMessage=function(msg){parentPort.postMessage(msg)};if(typeof performance==="undefined"){performance={now:function(){return Date.now()}}}}',dge=ii(aE()),D6=class extends Yl{constructor(e){super();this.wasm=e,this.dataIdNextNumber=1,this.wasm.tfjs.initWithThreadsCount(O6),Dx=this.wasm.tfjs.getThreadsCount(),this.dataIdMap=new Xc(this,ns())}write(e,t,n){let s={id:this.dataIdNextNumber++};return this.move(s,e,t,n,1),s}numDataIds(){return this.dataIdMap.numDataIds()}async time(e){let t=v.now();return e(),{kernelMs:v.now()-t}}move(e,t,n,s,r){let a=this.dataIdNextNumber++;if(s==="string"){let c=t;this.dataIdMap.set(e,{id:a,stringBytes:c,shape:n,dtype:s,memoryOffset:null,refCount:r});return}let o=v.sizeFromShape(n),i=o*v.bytesPerElement(s),l=this.wasm._malloc(i);this.dataIdMap.set(e,{id:a,memoryOffset:l,shape:n,dtype:s,refCount:r}),this.wasm.tfjs.registerTensor(a,o,l),t!=null&&this.wasm.HEAPU8.set(new Uint8Array(t.buffer,t.byteOffset,i),l)}async read(e){return this.readSync(e)}readSync(e){let{memoryOffset:t,dtype:n,shape:s,stringBytes:r}=this.dataIdMap.get(e);if(n==="string")return r;let a=this.wasm.HEAPU8.slice(t,t+v.sizeFromShape(s)*v.bytesPerElement(n));return fge(a.buffer,n)}disposeData(e,t=!1){if(this.dataIdMap.has(e)){let n=this.dataIdMap.get(e);if(n.refCount--,!t&&n.refCount>0)return!1;this.wasm._free(n.memoryOffset),this.wasm.tfjs.disposeData(n.id),this.dataIdMap.delete(e)}return!0}refCount(e){return this.dataIdMap.has(e)?this.dataIdMap.get(e).refCount:0}incRef(e){let t=this.dataIdMap.get(e);t!=null&&t.refCount++}floatPrecision(){return 32}getMemoryOffset(e){return this.dataIdMap.get(e).memoryOffset}dispose(){this.wasm.tfjs.dispose(),"PThread"in this.wasm&&this.wasm.PThread.terminateAllThreads(),this.wasm=null}memory(){return{unreliable:!1}}makeOutput(e,t,n){let s;if(n==null)s=this.write(null,e,t);else{let r=this.dataIdNextNumber++;s={id:r},this.dataIdMap.set(s,{id:r,memoryOffset:n,shape:e,dtype:t,refCount:1});let a=v.sizeFromShape(e);this.wasm.tfjs.registerTensor(r,a,n)}return{dataId:s,shape:e,dtype:t}}typedArrayFromHeap({shape:e,dtype:t,dataId:n}){let s=this.wasm.HEAPU8.buffer,{memoryOffset:r}=this.dataIdMap.get(n),a=v.sizeFromShape(e);switch(t){case"float32":return new Float32Array(s,r,a);case"int32":return new Int32Array(s,r,a);case"bool":return new Uint8Array(s,r,a);default:throw new Error(`Unknown dtype ${t}`)}}};function pge(e){return(t,n)=>(v.fetch(e,{credentials:"same-origin"}).then(s=>{s.ok||t.env.a(`failed to load wasm binary file at '${e}'`),s.arrayBuffer().then(r=>{WebAssembly.instantiate(r,t).then(a=>{n(a.instance,a.module)})})}),{})}function P6(e,t,n){if(Qm!=null)return Qm;let s="tfjs-backend-wasm.wasm";return e&&t?s="tfjs-backend-wasm-threaded-simd.wasm":e&&(s="tfjs-backend-wasm-simd.wasm"),Sp!=null&&Sp[s]!=null?Sp[s]:n+s}async function hge(){let[e,t]=await Promise.all([K().getAsync("WASM_HAS_SIMD_SUPPORT"),K().getAsync("WASM_HAS_MULTITHREAD_SUPPORT")]);return new Promise((n,s)=>{let r={};r.locateFile=(i,l)=>{if(i.endsWith(".worker.js")){let c=cge,u=new Blob([c],{type:"application/javascript"});return URL.createObjectURL(u)}return i.endsWith(".wasm")?P6(e,t,kp!=null?kp:l):l+i},_x&&(r.instantiateWasm=pge(P6(e,t,kp!=null?kp:"")));let a=!1;r.onAbort=()=>{if(a||Ip)return;Ip=!0,s({message:"Make sure the server can serve the `.wasm` file relative to the bundled js file. For more details see https://github.com/tensorflow/tfjs/blob/master/tfjs-backend-wasm/README.md#using-bundlers"})};let o;t&&e&&Qm==null?(r.mainScriptUrlOrBlob=new Blob(["var WasmBackendModuleThreadedSimd = "+_6.default.toString()],{type:"text/javascript"}),o=(0,_6.default)(r)):o=(0,dge.default)(r),o.then(i=>{a=!0,Ip=!1;let l=null;i.tfjs={init:i.cwrap("init",null,[]),initWithThreadsCount:i.cwrap("init_with_threads_count",null,["number"]),getThreadsCount:i.cwrap("get_threads_count","number",[]),registerTensor:i.cwrap("register_tensor",null,["number","number","number"]),disposeData:i.cwrap("dispose_data",l,["number"]),dispose:i.cwrap("dispose",l,[])},n({wasm:i})})})}function fge(e,t){switch(t){case"float32":return new Float32Array(e);case"int32":return new Int32Array(e);case"bool":return new Uint8Array(e);default:throw new Error(`Unknown dtype ${t}`)}}var mge=["tfjs-backend-wasm.wasm","tfjs-backend-wasm-simd.wasm","tfjs-backend-wasm-threaded-simd.wasm"],Qm=null,kp=null,Sp={},Ip=!1,_x=!1;function gge(e,t=!1){if(z2("setWasmPath has been deprecated in favor of setWasmPaths and will be removed in a future release."),Ip)throw new Error("The WASM backend was already initialized. Make sure you call `setWasmPath()` before you call `tf.setBackend()` or `tf.ready()`");Qm=e,_x=t}function F6(e,t=!1){if(Ip)throw new Error("The WASM backend was already initialized. Make sure you call `setWasmPaths()` before you call `tf.setBackend()` or `tf.ready()`");if(typeof e=="string")kp=e;else{Sp=e;let n=mge.filter(s=>Sp[s]==null);if(n.length>0)throw new Error(`There were no entries found for the following binaries: ${n.join(",")}. Please either call setWasmPaths with a map providing a path for each binary, or with a string indicating the directory where all the binaries can be found.`)}_x=t}var O6=-1,Dx=-1;function Age(e){O6=e}function yge(){if(Dx===-1)throw new Error("WASM backend not initialized.");return Dx}var xge="0.0.0",bge=2;al("wasm",async()=>{let{wasm:e}=await hge();return new D6(e)},bge);var jo="3.11.0-20211108",M6={tfjs:jo,"tfjs-core":jo,"tfjs-data":jo,"tfjs-layers":jo,"tfjs-converter":jo,"tfjs-backend-cpu":jo,"tfjs-backend-webgl":jo,"tfjs-backend-wasm":jo},Cp=M6["tfjs-core"];var z6=`
|
|
precision highp float;
|
|
attribute vec2 pos;
|
|
attribute vec2 uv;
|
|
varying vec2 vUv;
|
|
uniform float flipY;
|
|
void main(void) {
|
|
vUv = uv;
|
|
gl_Position = vec4(pos.x, pos.y*flipY, 0.0, 1.);
|
|
}
|
|
`;var L6=`
|
|
precision highp float;
|
|
varying vec2 vUv;
|
|
uniform sampler2D texture;
|
|
uniform float m[20];
|
|
void main(void) {
|
|
vec4 c = texture2D(texture, vUv);
|
|
gl_FragColor.r = m[0] * c.r + m[1] * c.g + m[2] * c.b + m[3] * c.a + m[4];
|
|
gl_FragColor.g = m[5] * c.r + m[6] * c.g + m[7] * c.b + m[8] * c.a + m[9];
|
|
gl_FragColor.b = m[10] * c.r + m[11] * c.g + m[12] * c.b + m[13] * c.a + m[14];
|
|
gl_FragColor.a = m[15] * c.r + m[16] * c.g + m[17] * c.b + m[18] * c.a + m[19];
|
|
}
|
|
`,B6=`
|
|
precision highp float;
|
|
varying vec2 vUv;
|
|
uniform sampler2D texture;
|
|
uniform float m[20];
|
|
void main(void) {
|
|
vec4 c = texture2D(texture, vUv);
|
|
gl_FragColor.r = m[0] * c.r + m[1] * c.g + m[2] * c.b + m[4];
|
|
gl_FragColor.g = m[5] * c.r + m[6] * c.g + m[7] * c.b + m[9];
|
|
gl_FragColor.b = m[10] * c.r + m[11] * c.g + m[12] * c.b + m[14];
|
|
gl_FragColor.a = c.a;
|
|
}
|
|
`,W6=`
|
|
precision highp float;
|
|
varying vec2 vUv;
|
|
uniform vec2 size;
|
|
uniform sampler2D texture;
|
|
vec2 pixelate(vec2 coord, vec2 size) {
|
|
return floor( coord / size ) * size;
|
|
}
|
|
void main(void) {
|
|
gl_FragColor = vec4(0.0);
|
|
vec2 coord = pixelate(vUv, size);
|
|
gl_FragColor += texture2D(texture, coord);
|
|
}
|
|
`,V6=`
|
|
precision highp float;
|
|
varying vec2 vUv;
|
|
uniform sampler2D texture;
|
|
uniform vec2 px;
|
|
void main(void) {
|
|
gl_FragColor = vec4(0.0);
|
|
gl_FragColor += texture2D(texture, vUv + vec2(-7.0*px.x, -7.0*px.y))*0.0044299121055113265;
|
|
gl_FragColor += texture2D(texture, vUv + vec2(-6.0*px.x, -6.0*px.y))*0.00895781211794;
|
|
gl_FragColor += texture2D(texture, vUv + vec2(-5.0*px.x, -5.0*px.y))*0.0215963866053;
|
|
gl_FragColor += texture2D(texture, vUv + vec2(-4.0*px.x, -4.0*px.y))*0.0443683338718;
|
|
gl_FragColor += texture2D(texture, vUv + vec2(-3.0*px.x, -3.0*px.y))*0.0776744219933;
|
|
gl_FragColor += texture2D(texture, vUv + vec2(-2.0*px.x, -2.0*px.y))*0.115876621105;
|
|
gl_FragColor += texture2D(texture, vUv + vec2(-1.0*px.x, -1.0*px.y))*0.147308056121;
|
|
gl_FragColor += texture2D(texture, vUv )*0.159576912161;
|
|
gl_FragColor += texture2D(texture, vUv + vec2( 1.0*px.x, 1.0*px.y))*0.147308056121;
|
|
gl_FragColor += texture2D(texture, vUv + vec2( 2.0*px.x, 2.0*px.y))*0.115876621105;
|
|
gl_FragColor += texture2D(texture, vUv + vec2( 3.0*px.x, 3.0*px.y))*0.0776744219933;
|
|
gl_FragColor += texture2D(texture, vUv + vec2( 4.0*px.x, 4.0*px.y))*0.0443683338718;
|
|
gl_FragColor += texture2D(texture, vUv + vec2( 5.0*px.x, 5.0*px.y))*0.0215963866053;
|
|
gl_FragColor += texture2D(texture, vUv + vec2( 6.0*px.x, 6.0*px.y))*0.00895781211794;
|
|
gl_FragColor += texture2D(texture, vUv + vec2( 7.0*px.x, 7.0*px.y))*0.0044299121055113265;
|
|
}
|
|
`,U6=`
|
|
precision highp float;
|
|
varying vec2 vUv;
|
|
uniform sampler2D texture;
|
|
uniform vec2 px;
|
|
uniform float m[9];
|
|
void main(void) {
|
|
vec4 c11 = texture2D(texture, vUv - px); // top left
|
|
vec4 c12 = texture2D(texture, vec2(vUv.x, vUv.y - px.y)); // top center
|
|
vec4 c13 = texture2D(texture, vec2(vUv.x + px.x, vUv.y - px.y)); // top right
|
|
vec4 c21 = texture2D(texture, vec2(vUv.x - px.x, vUv.y) ); // mid left
|
|
vec4 c22 = texture2D(texture, vUv); // mid center
|
|
vec4 c23 = texture2D(texture, vec2(vUv.x + px.x, vUv.y) ); // mid right
|
|
vec4 c31 = texture2D(texture, vec2(vUv.x - px.x, vUv.y + px.y) ); // bottom left
|
|
vec4 c32 = texture2D(texture, vec2(vUv.x, vUv.y + px.y) ); // bottom center
|
|
vec4 c33 = texture2D(texture, vUv + px ); // bottom right
|
|
gl_FragColor =
|
|
c11 * m[0] + c12 * m[1] + c22 * m[2] +
|
|
c21 * m[3] + c22 * m[4] + c23 * m[5] +
|
|
c31 * m[6] + c32 * m[7] + c33 * m[8];
|
|
gl_FragColor.a = c22.a;
|
|
}
|
|
`;var Px=(e,t,n)=>{let s=new RegExp("\\b"+t+" \\w+ (\\w+)","ig");e.replace(s,(r,a)=>(n[a]=0,r))},G6=class{constructor(t,n,s){ce(this,"uniform",{});ce(this,"attribute",{});ce(this,"gl");ce(this,"id");ce(this,"compile",(t,n)=>{let s=this.gl.createShader(n);if(this.gl.shaderSource(s,t),this.gl.compileShader(s),!this.gl.getShaderParameter(s,this.gl.COMPILE_STATUS))throw new Error(`filter: gl compile failed: ${this.gl.getShaderInfoLog(s)}`);return s});this.gl=t;let r=this.compile(n,this.gl.VERTEX_SHADER),a=this.compile(s,this.gl.FRAGMENT_SHADER);if(this.id=this.gl.createProgram(),this.gl.attachShader(this.id,r),this.gl.attachShader(this.id,a),this.gl.linkProgram(this.id),!this.gl.getProgramParameter(this.id,this.gl.LINK_STATUS))throw new Error(`filter: gl link failed: ${this.gl.getProgramInfoLog(this.id)}`);this.gl.useProgram(this.id),Px(n,"attribute",this.attribute);for(let o in this.attribute)this.attribute[o]=this.gl.getAttribLocation(this.id,o);Px(n,"uniform",this.uniform),Px(s,"uniform",this.uniform);for(let o in this.uniform)this.uniform[o]=this.gl.getUniformLocation(this.id,o)}};function H6(){let e=0,t=null,n=!1,s=-1,r=[null,null],a=[],o=null,i=null,l=Xn(100,100),c={},u={INTERMEDIATE:1},d=l.getContext("webgl");if(!d)throw new Error("filter: cannot get webgl context");function p(x,y){if(!(x===l.width&&y===l.height)){if(l.width=x,l.height=y,!o){let b=new Float32Array([-1,-1,0,1,1,-1,1,1,-1,1,0,0,-1,1,0,0,1,-1,1,1,1,1,1,0]);o=d.createBuffer(),d.bindBuffer(d.ARRAY_BUFFER,o),d.bufferData(d.ARRAY_BUFFER,b,d.STATIC_DRAW),d.pixelStorei(d.UNPACK_PREMULTIPLY_ALPHA_WEBGL,!0)}d.viewport(0,0,l.width,l.height),r=[null,null]}}function h(x,y){let b=d.createFramebuffer();d.bindFramebuffer(d.FRAMEBUFFER,b);let w=d.createRenderbuffer();d.bindRenderbuffer(d.RENDERBUFFER,w);let k=d.createTexture();return d.bindTexture(d.TEXTURE_2D,k),d.texImage2D(d.TEXTURE_2D,0,d.RGBA,x,y,0,d.RGBA,d.UNSIGNED_BYTE,null),d.texParameteri(d.TEXTURE_2D,d.TEXTURE_MAG_FILTER,d.LINEAR),d.texParameteri(d.TEXTURE_2D,d.TEXTURE_MIN_FILTER,d.LINEAR),d.texParameteri(d.TEXTURE_2D,d.TEXTURE_WRAP_S,d.CLAMP_TO_EDGE),d.texParameteri(d.TEXTURE_2D,d.TEXTURE_WRAP_T,d.CLAMP_TO_EDGE),d.framebufferTexture2D(d.FRAMEBUFFER,d.COLOR_ATTACHMENT0,d.TEXTURE_2D,k,0),d.bindTexture(d.TEXTURE_2D,null),d.bindFramebuffer(d.FRAMEBUFFER,null),{fbo:b,texture:k}}function f(x){return r[x]=r[x]||h(l.width,l.height),r[x]}function m(x=0){if(!i)return;let y=null,b=null,w=!1;e===0?y=t:y=f(s).texture||null,e++,n&&!(x&u.INTERMEDIATE)?(b=null,w=e%2==0):(s=(s+1)%2,b=f(s).fbo||null),d.bindTexture(d.TEXTURE_2D,y),d.bindFramebuffer(d.FRAMEBUFFER,b),d.uniform1f(i.uniform.flipY,w?-1:1),d.drawArrays(d.TRIANGLES,0,6)}function g(x){if(c[x])return i=c[x],d.useProgram((i?i.id:null)||null),i;i=new G6(d,z6,x);let y=Float32Array.BYTES_PER_ELEMENT,b=4*y;return d.enableVertexAttribArray(i.attribute.pos),d.vertexAttribPointer(i.attribute.pos,2,d.FLOAT,!1,b,0*y),d.enableVertexAttribArray(i.attribute.uv),d.vertexAttribPointer(i.attribute.uv,2,d.FLOAT,!1,b,2*y),c[x]=i,i}let A={colorMatrix:x=>{let y=new Float32Array(x);y[4]/=255,y[9]/=255,y[14]/=255,y[19]/=255;let b=y[18]===1&&y[3]===0&&y[8]===0&&y[13]===0&&y[15]===0&&y[16]===0&&y[17]===0&&y[19]===0?B6:L6,w=g(b);d.uniform1fv(w.uniform.m,y),m()},brightness:x=>{let y=(x||0)+1;A.colorMatrix([y,0,0,0,0,0,y,0,0,0,0,0,y,0,0,0,0,0,1,0])},saturation:x=>{let y=(x||0)*2/3+1,b=(y-1)*-.5;A.colorMatrix([y,b,b,0,0,b,y,b,0,0,b,b,y,0,0,0,0,0,1,0])},desaturate:()=>{A.saturation(-1)},contrast:x=>{let y=(x||0)+1,b=-128*(y-1);A.colorMatrix([y,0,0,0,b,0,y,0,0,b,0,0,y,0,b,0,0,0,1,0])},negative:()=>{A.contrast(-2)},hue:x=>{x=(x||0)/180*Math.PI;let y=Math.cos(x),b=Math.sin(x),w=.213,k=.715,C=.072;A.colorMatrix([w+y*(1-w)+b*-w,k+y*-k+b*-k,C+y*-C+b*(1-C),0,0,w+y*-w+b*.143,k+y*(1-k)+b*.14,C+y*-C+b*-.283,0,0,w+y*-w+b*-(1-w),k+y*-k+b*k,C+y*(1-C)+b*C,0,0,0,0,0,1,0])},desaturateLuminance:()=>{A.colorMatrix([.2764723,.929708,.0938197,0,-37.1,.2764723,.929708,.0938197,0,-37.1,.2764723,.929708,.0938197,0,-37.1,0,0,0,1,0])},sepia:()=>{A.colorMatrix([.393,.7689999,.18899999,0,0,.349,.6859999,.16799999,0,0,.272,.5339999,.13099999,0,0,0,0,0,1,0])},brownie:()=>{A.colorMatrix([.5997023498159715,.34553243048391263,-.2708298674538042,0,47.43192855600873,-.037703249837783157,.8609577587992641,.15059552388459913,0,-36.96841498319127,.24113635128153335,-.07441037908422492,.44972182064877153,0,-7.562075277591283,0,0,0,1,0])},vintagePinhole:()=>{A.colorMatrix([.6279345635605994,.3202183420819367,-.03965408211312453,0,9.651285835294123,.02578397704808868,.6441188644374771,.03259127616149294,0,7.462829176470591,.0466055556782719,-.0851232987247891,.5241648018700465,0,5.159190588235296,0,0,0,1,0])},kodachrome:()=>{A.colorMatrix([1.1285582396593525,-.3967382283601348,-.03992559172921793,0,63.72958762196502,-.16404339962244616,1.0835251566291304,-.05498805115633132,0,24.732407896706203,-.16786010706155763,-.5603416277695248,1.6014850761964943,0,35.62982807460946,0,0,0,1,0])},technicolor:()=>{A.colorMatrix([1.9125277891456083,-.8545344976951645,-.09155508482755585,0,11.793603434377337,-.3087833385928097,1.7658908555458428,-.10601743074722245,0,-70.35205161461398,-.231103377548616,-.7501899197440212,1.847597816108189,0,30.950940869491138,0,0,0,1,0])},polaroid:()=>{A.colorMatrix([1.438,-.062,-.062,0,0,-.122,1.378,-.122,0,0,-.016,-.016,1.483,0,0,0,0,0,1,0])},shiftToBGR:()=>{A.colorMatrix([0,0,1,0,0,0,1,0,0,0,1,0,0,0,0,0,0,0,1,0])},convolution:x=>{let y=new Float32Array(x),b=1/l.width,w=1/l.height,k=g(U6);d.uniform1fv(k.uniform.m,y),d.uniform2f(k.uniform.px,b,w),m()},detectEdges:()=>{A.convolution.call(this,[0,1,0,1,-4,1,0,1,0])},sobelX:()=>{A.convolution.call(this,[-1,0,1,-2,0,2,-1,0,1])},sobelY:()=>{A.convolution.call(this,[-1,-2,-1,0,0,0,1,2,1])},sharpen:x=>{let y=x||1;A.convolution.call(this,[0,-1*y,0,-1*y,1+4*y,-1*y,0,-1*y,0])},emboss:x=>{let y=x||1;A.convolution.call(this,[-2*y,-1*y,0,-1*y,1,1*y,0,1*y,2*y])},blur:x=>{let y=x/7/l.width,b=x/7/l.height,w=g(V6);d.uniform2f(w.uniform.px,0,b),m(u.INTERMEDIATE),d.uniform2f(w.uniform.px,y,0),m()},pixelate:x=>{let y=x/l.width,b=x/l.height,w=g(W6);d.uniform2f(w.uniform.size,y,b),m()}};this.add=function(x){let y=Array.prototype.slice.call(arguments,1),b=A[x];a.push({func:b,args:y})},this.reset=function(){a=[]},this.get=function(){return a},this.apply=function(x){p(x.width,x.height),e=0,t||(t=d.createTexture()),d.bindTexture(d.TEXTURE_2D,t),d.texParameteri(d.TEXTURE_2D,d.TEXTURE_WRAP_S,d.CLAMP_TO_EDGE),d.texParameteri(d.TEXTURE_2D,d.TEXTURE_WRAP_T,d.CLAMP_TO_EDGE),d.texParameteri(d.TEXTURE_2D,d.TEXTURE_MIN_FILTER,d.NEAREST),d.texParameteri(d.TEXTURE_2D,d.TEXTURE_MAG_FILTER,d.NEAREST),d.texImage2D(d.TEXTURE_2D,0,d.RGBA,d.RGBA,d.UNSIGNED_BYTE,x);for(let y=0;y<a.length;y++){n=y===a.length-1;let b=a[y];b.func.apply(this,b.args||[])}return l},this.draw=function(x){return this.add("brightness",0),this.apply(x)}}async function e0(e){let t=e.shape.length===4?ot(e):e,n=nn(t,3,2),s=[Eo(n[0]),Eo(n[1]),Eo(n[2])],r=[gn(n[0]),gn(n[1]),gn(n[2])],a=await Promise.all(r.map(h=>h.data())),o=.99*Math.max(a[0][0],a[1][0],a[2][0]),i=[me(n[0],s[0]),me(n[1],s[1]),me(n[2],s[2])],l=[me(r[0],s[0]),me(r[1],s[1]),me(r[2],s[2])],c=[pe(o,l[0]),pe(o,l[1]),pe(o,l[2])],u=[L(i[0],c[0]),L(i[1],c[1]),L(i[2],c[2])],d=An([u[0],u[1],u[2]],2),p=G(d,[1,t.shape[0],t.shape[1],3]);return se([...n,...s,...r,...i,...l,...c,...u,d,t]),p}var t0=2048,it=null,on=null,bc=null,$t,oa={inputSum:0,cacheDiff:1,sumMethod:0,inputTensor:void 0};function Xn(e,t){let n;if(Ae.browser)if(Ae.worker)n=new OffscreenCanvas(e,t);else{if(typeof document=="undefined")throw new Error("attempted to run in web worker but offscreenCanvas is not supported");n=document.createElement("canvas"),n.width=e,n.height=t}else typeof Ae.Canvas!="undefined"?n=new Ae.Canvas(e,t):typeof globalThis.Canvas!="undefined"&&(n=new globalThis.Canvas(e,t));return n}function Fx(e,t){let n=t||Xn(e.width,e.height);return n.getContext("2d").drawImage(e,0,0),n}async function vc(e,t,n=!0){if(!e)return t.debug&&re("input is missing"),{tensor:null,canvas:null};if(!(e instanceof Ze)&&!(typeof Image!="undefined"&&e instanceof Image)&&!(typeof Ae.Canvas!="undefined"&&e instanceof Ae.Canvas)&&!(typeof globalThis.Canvas!="undefined"&&e instanceof globalThis.Canvas)&&!(typeof ImageData!="undefined"&&e instanceof ImageData)&&!(typeof ImageBitmap!="undefined"&&e instanceof ImageBitmap)&&!(typeof HTMLImageElement!="undefined"&&e instanceof HTMLImageElement)&&!(typeof HTMLMediaElement!="undefined"&&e instanceof HTMLMediaElement)&&!(typeof HTMLVideoElement!="undefined"&&e instanceof HTMLVideoElement)&&!(typeof HTMLCanvasElement!="undefined"&&e instanceof HTMLCanvasElement)&&!(typeof OffscreenCanvas!="undefined"&&e instanceof OffscreenCanvas))throw new Error("input type is not recognized");if(e instanceof Ze){if(e.isDisposedInternal)throw new Error("input tensor is disposed");if(!e.shape||e.shape.length!==4||e.shape[0]!==1||e.shape[3]!==3)throw new Error("input tensor shape must be [1, height, width, 3] and instead was"+(e.shape?e.shape.toString():"unknown"));return{tensor:Ln(e),canvas:t.filter.return?on:null}}else{if(typeof e.readyState!="undefined"&&e.readyState<=2)return t.debug&&re("input stream is not ready"),{tensor:null,canvas:it};let s=e.naturalWidth||e.videoWidth||e.width||e.shape&&e.shape[1]>0,r=e.naturalHeight||e.videoHeight||e.height||e.shape&&e.shape[2]>0;if(!s||!r)return t.debug&&re("cannot determine input dimensions"),{tensor:null,canvas:it};let a=s,o=r;if(a>t0&&(a=t0,o=Math.trunc(a*r/s)),o>t0&&(o=t0,a=Math.trunc(o*s/r)),(t.filter.width||0)>0?a=t.filter.width:(t.filter.height||0)>0&&(a=s*((t.filter.height||0)/r)),(t.filter.height||0)>0?o=t.filter.height:(t.filter.width||0)>0&&(o=r*((t.filter.width||0)/s)),!a||!o)throw new Error("input cannot determine dimension");(!it||(it==null?void 0:it.width)!==a||(it==null?void 0:it.height)!==o)&&(it=Xn(a,o));let i=it.getContext("2d");if(typeof ImageData!="undefined"&&e instanceof ImageData?i.putImageData(e,0,0):t.filter.flip&&typeof i.translate!="undefined"?(i.translate(s,0),i.scale(-1,1),i.drawImage(e,0,0,s,r,0,0,it==null?void 0:it.width,it==null?void 0:it.height),i.setTransform(1,0,0,1,0,0)):i.drawImage(e,0,0,s,r,0,0,it==null?void 0:it.width,it==null?void 0:it.height),(!on||it.width!==on.width||(it==null?void 0:it.height)!==(on==null?void 0:on.height))&&(on=Xn(it.width,it.height)),t.filter.enabled&&Ae.webgl.supported){if($t||($t=Ae.browser?new H6:null),Ae.filter=!!$t,!$t)return{tensor:null,canvas:it};$t.reset(),t.filter.brightness!==0&&$t.add("brightness",t.filter.brightness),t.filter.contrast!==0&&$t.add("contrast",t.filter.contrast),t.filter.sharpness!==0&&$t.add("sharpen",t.filter.sharpness),t.filter.blur!==0&&$t.add("blur",t.filter.blur),t.filter.saturation!==0&&$t.add("saturation",t.filter.saturation),t.filter.hue!==0&&$t.add("hue",t.filter.hue),t.filter.negative&&$t.add("negative"),t.filter.sepia&&$t.add("sepia"),t.filter.vintage&&$t.add("brownie"),t.filter.sepia&&$t.add("sepia"),t.filter.kodachrome&&$t.add("kodachrome"),t.filter.technicolor&&$t.add("technicolor"),t.filter.polaroid&&$t.add("polaroid"),t.filter.pixelate!==0&&$t.add("pixelate",t.filter.pixelate),$t.get()>0?on=$t.apply(it):on=$t.draw(it)}else Fx(it,on),$t&&($t=null),Ae.filter=!!$t;if(!n)return{tensor:null,canvas:on};if(!on)throw new Error("cannot create output canvas");let l,c=3;if(typeof ImageData!="undefined"&&e instanceof ImageData||e.data&&e.width&&e.height)if(Ae.browser&&Gs)l=Gs?Gs.fromPixels(e):null;else{c=e.data.length/e.height/e.width;let p=new Uint8Array(e.data.buffer);l=Ut(p,[e.height,e.width,c],"int32")}else if((!bc||on.width!==bc.width||on.height!==bc.height)&&(bc=Xn(on.width,on.height)),Gs&&Ae.browser)t.backend==="webgl"||t.backend==="humangl"||t.backend==="webgpu"?l=Gs.fromPixels(on):(bc=Fx(on),l=Gs.fromPixels(bc));else{let f=Fx(on).getContext("2d").getImageData(0,0,a,o);c=f.data.length/a/o;let m=new Uint8Array(f.data.buffer);l=Ut(m,[a,o,c])}if(c===4){let p=Wu(l,[0,0,0],[-1,-1,3]);se(l),l=p}if(!l)throw new Error("cannot create tensor from input");let u=de(l,"float32"),d=t.filter.equalization?await e0(u):mn(u,0);return se([l,u]),{tensor:d,canvas:t.filter.return?on:null}}}async function j6(e,t){let n=!1;if(e.cacheSensitivity===0)return n;if(!oa.inputTensor)oa.inputTensor=Ln(t);else if(oa.inputTensor.shape[1]!==t.shape[1]||oa.inputTensor.shape[2]!==t.shape[2])se(oa.inputTensor),oa.inputTensor=Ln(t);else{let s={};s.diff=me(t,oa.inputTensor),s.squared=L(s.diff,s.diff),s.sum=Se(s.squared);let a=(await s.sum.data())[0]/(t.shape[1]||1)/(t.shape[2]||1)/255/3;se([oa.inputTensor,s.diff,s.squared,s.sum]),oa.inputTensor=Ln(t),n=a<=e.cacheSensitivity}return n}async function q6(e,t,n){let s={};if(!t||!n||t.shape.length!==4||t.shape.length!==n.shape.length)return e.debug||re("invalid input tensor or tensor shapes do not match:",t.shape,n.shape),0;if(t.shape[0]!==1||n.shape[0]!==1||t.shape[3]!==3||n.shape[3]!==3)return e.debug||re("input tensors must be of shape [1, height, width, 3]:",t.shape,n.shape),0;s.input1=Ln(t),s.input2=t.shape[1]!==n.shape[1]||t.shape[2]!==n.shape[2]?$e.resizeBilinear(n,[t.shape[1],t.shape[2]]):Ln(n),s.diff=me(s.input1,s.input2),s.squared=L(s.diff,s.diff),s.sum=Se(s.squared);let a=(await s.sum.data())[0]/(t.shape[1]||1)/(t.shape[2]||1)/255/3;return se([s.input1,s.input2,s.diff,s.squared,s.sum]),a}var X6=class{constructor(){ce(this,"browser");ce(this,"node");ce(this,"worker");ce(this,"platform","");ce(this,"agent","");ce(this,"backends",[]);ce(this,"initial");ce(this,"filter");ce(this,"tfjs");ce(this,"offscreen");ce(this,"perfadd",!1);ce(this,"wasm",{supported:void 0,backend:void 0,simd:void 0,multithread:void 0});ce(this,"webgl",{supported:void 0,backend:void 0,version:void 0,renderer:void 0});ce(this,"webgpu",{supported:void 0,backend:void 0,adapter:void 0});ce(this,"cpu",{model:void 0,flags:[]});ce(this,"kernels",[]);ce(this,"Canvas");ce(this,"Image");ce(this,"ImageData");if(this.browser=typeof navigator!="undefined",this.node=typeof process!="undefined",this.tfjs={version:Cp},this.offscreen=typeof OffscreenCanvas!="undefined",this.initial=!0,this.worker=this.browser&&this.offscreen?typeof WorkerGlobalScope!="undefined":void 0,typeof navigator!="undefined"){let t=navigator.userAgent.match(/\(([^()]+)\)/g);if(t&&t[0]){let n=t[0].match(/\(([^()]+)\)/g);this.platform=n&&n[0]?n[0].replace(/\(|\)/g,""):"",this.agent=navigator.userAgent.replace(t[0],""),this.platform[1]&&(this.agent=this.agent.replace(t[1],"")),this.agent=this.agent.replace(/ /g," ")}}else typeof process!="undefined"&&(this.platform=`${process.platform} ${process.arch}`,this.agent=`NodeJS ${process.version}`)}async updateBackend(){this.backends=Object.keys(ns().registryFactory),this.wasm.supported=typeof WebAssembly!="undefined",this.wasm.backend=this.backends.includes("wasm"),this.wasm.supported&&this.wasm.backend&&Is()==="wasm"&&(this.wasm.simd=await K().getAsync("WASM_HAS_SIMD_SUPPORT"),this.wasm.multithread=await K().getAsync("WASM_HAS_MULTITHREAD_SUPPORT"));let t=Xn(100,100),n=t?t.getContext("webgl2"):void 0;if(this.webgl.supported=typeof n!="undefined",this.webgl.backend=this.backends.includes("webgl"),this.webgl.supported&&this.webgl.backend&&(Is()==="webgl"||Is()==="humangl")){let s=Tr().gpgpu!=="undefined"?await Tr().getGPGPUContext().gl:null;s&&(this.webgl.version=s.getParameter(s.VERSION),this.webgl.renderer=s.getParameter(s.RENDERER))}this.webgpu.supported=this.browser&&typeof navigator.gpu!="undefined",this.webgpu.backend=this.backends.includes("webgpu"),this.webgpu.supported&&(this.webgpu.adapter=(await navigator.gpu.requestAdapter()).name),this.kernels=qr(Is()).map(s=>s.kernelName.toLowerCase())}async updateCPU(){let t={model:"",flags:[]};if(this.node&&this.platform.startsWith("linux")){let n=va("fs");try{let s=n.readFileSync("/proc/cpuinfo").toString();for(let r of s.split(`
|
|
`))r.startsWith("model name")&&(t.model=r.match(/:(.*)/g)[0].replace(":","").trim()),r.startsWith("flags")&&(t.flags=r.match(/:(.*)/g)[0].replace(":","").trim().split(" ").sort())}catch(s){}}this.cpu?this.cpu=t:Object.defineProperty(this,"cpu",{value:t})}},Ae=new X6;var Ox="2.5.1";var qo;var j2e=Number.MAX_SAFE_INTEGER;async function K6(e){return Ae.initial&&(qo=null),qo?e.debug&&re("cached model:",qo.modelUrl):(qo=await et(nt(e.modelBasePath,e.face.agegenderrace.modelPath)),!qo||!qo.modelUrl?re("load model failed:",e.face.agegenderrace.modelPath):e.debug&&re("load model:",qo.modelUrl)),qo}var ln,n0=[],Mx=Number.MAX_SAFE_INTEGER,Z6=0,Y6=0;async function J6(e){var t,n;return Ae.initial&&(ln=null),ln?e.debug&&re("cached model:",ln.modelUrl):(ln=await et(nt(e.modelBasePath,((t=e.face.antispoof)==null?void 0:t.modelPath)||"")),!ln||!ln.modelUrl?re("load model failed:",(n=e.face.antispoof)==null?void 0:n.modelPath):e.debug&&re("load model:",ln.modelUrl)),ln}async function zx(e,t,n,s){var o,i;if(!ln)return null;let r=(((o=t.face.antispoof)==null?void 0:o.skipTime)||0)>he()-Y6,a=Mx<(((i=t.face.antispoof)==null?void 0:i.skipFrames)||0);return t.skipAllowed&&r&&a&&Z6===s&&n0[n]?(Mx++,n0[n]):(Mx=0,new Promise(async l=>{let c=$e.resizeBilinear(e,[(ln==null?void 0:ln.inputs[0].shape)?ln.inputs[0].shape[2]:0,(ln==null?void 0:ln.inputs[0].shape)?ln.inputs[0].shape[1]:0],!1),u=ln==null?void 0:ln.execute(c),d=(await u.data())[0];n0[n]=Math.round(100*d)/100,Z6=s,Y6=he(),se([c,u]),l(n0[n])}))}var Br={silhouette:[10,338,297,332,284,251,389,356,454,323,361,288,397,365,379,378,400,377,152,148,176,149,150,136,172,58,132,93,234,127,162,21,54,103,67,109],lipsUpperOuter:[61,185,40,39,37,0,267,269,270,409,291],lipsLowerOuter:[146,91,181,84,17,314,405,321,375,291],lipsUpperInner:[78,191,80,81,82,13,312,311,310,415,308],lipsLowerInner:[78,95,88,178,87,14,317,402,318,324,308],rightEyeUpper0:[246,161,160,159,158,157,173],rightEyeLower0:[33,7,163,144,145,153,154,155,133],rightEyeUpper1:[247,30,29,27,28,56,190],rightEyeLower1:[130,25,110,24,23,22,26,112,243],rightEyeUpper2:[113,225,224,223,222,221,189],rightEyeLower2:[226,31,228,229,230,231,232,233,244],rightEyeLower3:[143,111,117,118,119,120,121,128,245],rightEyebrowUpper:[156,70,63,105,66,107,55,193],rightEyebrowLower:[35,124,46,53,52,65],rightEyeIris:[473,474,475,476,477],leftEyeUpper0:[466,388,387,386,385,384,398],leftEyeLower0:[263,249,390,373,374,380,381,382,362],leftEyeUpper1:[467,260,259,257,258,286,414],leftEyeLower1:[359,255,339,254,253,252,256,341,463],leftEyeUpper2:[342,445,444,443,442,441,413],leftEyeLower2:[446,261,448,449,450,451,452,453,464],leftEyeLower3:[372,340,346,347,348,349,350,357,465],leftEyebrowUpper:[383,300,293,334,296,336,285,417],leftEyebrowLower:[265,353,276,283,282,295],leftEyeIris:[468,469,470,471,472],midwayBetweenEyes:[168],noseTip:[1],noseBottom:[2],noseRightCorner:[98],noseLeftCorner:[327],rightCheek:[205],leftCheek:[425]},Lx={count:468,mouth:13,symmetryLine:[13,Br.midwayBetweenEyes[0]]},Tp={leftEye:0,rightEye:1,nose:2,mouth:3,leftEar:4,rightEar:5,symmetryLine:[3,2]},Bx=[{key:"EyeUpper0",indices:[9,10,11,12,13,14,15]},{key:"EyeUpper1",indices:[25,26,27,28,29,30,31]},{key:"EyeUpper2",indices:[41,42,43,44,45,46,47]},{key:"EyeLower0",indices:[0,1,2,3,4,5,6,7,8]},{key:"EyeLower1",indices:[16,17,18,19,20,21,22,23,24]},{key:"EyeLower2",indices:[32,33,34,35,36,37,38,39,40]},{key:"EyeLower3",indices:[54,55,56,57,58,59,60,61,62]}],Np=[[.499976992607117,.652534008026123],[.500025987625122,.547487020492554],[.499974012374878,.602371990680695],[.482113003730774,.471979022026062],[.500150978565216,.527155995368958],[.499909996986389,.498252987861633],[.499523013830185,.40106201171875],[.289712011814117,.380764007568359],[.499954998493195,.312398016452789],[.499987006187439,.269918978214264],[.500023007392883,.107050001621246],[.500023007392883,.666234016418457],[.5000159740448,.679224014282227],[.500023007392883,.692348003387451],[.499976992607117,.695277988910675],[.499976992607117,.70593398809433],[.499976992607117,.719385027885437],[.499976992607117,.737019002437592],[.499967992305756,.781370997428894],[.499816000461578,.562981009483337],[.473773002624512,.573909997940063],[.104906998574734,.254140973091125],[.365929991006851,.409575998783112],[.338757991790771,.41302502155304],[.311120003461838,.409460008144379],[.274657994508743,.389131009578705],[.393361985683441,.403706014156342],[.345234006643295,.344011008739471],[.370094001293182,.346076011657715],[.319321990013123,.347265005111694],[.297903001308441,.353591024875641],[.24779200553894,.410809993743896],[.396889001131058,.842755019664764],[.280097991228104,.375599980354309],[.106310002505779,.399955987930298],[.2099249958992,.391353011131287],[.355807989835739,.534406006336212],[.471751004457474,.65040397644043],[.474155008792877,.680191993713379],[.439785003662109,.657229006290436],[.414617002010345,.66654098033905],[.450374007225037,.680860996246338],[.428770989179611,.682690978050232],[.374971002340317,.727805018424988],[.486716985702515,.547628998756409],[.485300987958908,.527395009994507],[.257764995098114,.314490020275116],[.401223003864288,.455172002315521],[.429818987846375,.548614978790283],[.421351999044418,.533740997314453],[.276895999908447,.532056987285614],[.483370006084442,.499586999416351],[.33721199631691,.282882988452911],[.296391993761063,.293242990970612],[.169294998049736,.193813979625702],[.447580009698868,.302609980106354],[.392390012741089,.353887975215912],[.354490011930466,.696784019470215],[.067304998636246,.730105042457581],[.442739009857178,.572826027870178],[.457098007202148,.584792017936707],[.381974011659622,.694710969924927],[.392388999462128,.694203019142151],[.277076005935669,.271932005882263],[.422551989555359,.563233017921448],[.385919004678726,.281364023685455],[.383103013038635,.255840003490448],[.331431001424789,.119714021682739],[.229923993349075,.232002973556519],[.364500999450684,.189113974571228],[.229622006416321,.299540996551514],[.173287004232407,.278747975826263],[.472878992557526,.666198015213013],[.446828007698059,.668527007102966],[.422762006521225,.673889994621277],[.445307999849319,.580065965652466],[.388103008270264,.693961024284363],[.403039008378983,.706539988517761],[.403629004955292,.693953037261963],[.460041999816895,.557139039039612],[.431158006191254,.692366003990173],[.452181994915009,.692366003990173],[.475387006998062,.692366003990173],[.465828001499176,.779190003871918],[.472328990697861,.736225962638855],[.473087012767792,.717857003211975],[.473122000694275,.704625964164734],[.473033010959625,.695277988910675],[.427942007780075,.695277988910675],[.426479011774063,.703539967536926],[.423162013292313,.711845993995667],[.4183090031147,.720062971115112],[.390094995498657,.639572978019714],[.013953999616206,.560034036636353],[.499913990497589,.58014702796936],[.413199990987778,.69539999961853],[.409626007080078,.701822996139526],[.468080013990402,.601534962654114],[.422728985548019,.585985004901886],[.463079988956451,.593783974647522],[.37211999297142,.47341400384903],[.334562003612518,.496073007583618],[.411671012639999,.546965003013611],[.242175996303558,.14767599105835],[.290776997804642,.201445996761322],[.327338010072708,.256527006626129],[.399509996175766,.748921036720276],[.441727995872498,.261676013469696],[.429764986038208,.187834024429321],[.412198007106781,.108901023864746],[.288955003023148,.398952007293701],[.218936994671822,.435410976409912],[.41278201341629,.398970007896423],[.257135003805161,.355440020561218],[.427684992551804,.437960982322693],[.448339998722076,.536936044692993],[.178560003638268,.45755398273468],[.247308000922203,.457193970680237],[.286267012357712,.467674970626831],[.332827985286713,.460712015628815],[.368755996227264,.447206974029541],[.398963987827301,.432654976844788],[.476410001516342,.405806005001068],[.189241006970406,.523923993110657],[.228962004184723,.348950982093811],[.490725994110107,.562400996685028],[.404670000076294,.485132992267609],[.019469000399113,.401564002037048],[.426243007183075,.420431017875671],[.396993011236191,.548797011375427],[.266469985246658,.376977026462555],[.439121007919312,.51895797252655],[.032313998788595,.644356966018677],[.419054001569748,.387154996395111],[.462783008813858,.505746960639954],[.238978996872902,.779744982719421],[.198220998048782,.831938028335571],[.107550002634525,.540755033493042],[.183610007166862,.740257024765015],[.134409993886948,.333683013916016],[.385764002799988,.883153975009918],[.490967005491257,.579378008842468],[.382384985685349,.508572995662689],[.174399003386497,.397670984268188],[.318785011768341,.39623498916626],[.343364000320435,.400596976280212],[.396100014448166,.710216999053955],[.187885001301765,.588537991046906],[.430987000465393,.944064974784851],[.318993002176285,.898285031318665],[.266247987747192,.869701027870178],[.500023007392883,.190576016902924],[.499976992607117,.954452991485596],[.366169989109039,.398822009563446],[.393207013607025,.39553701877594],[.410373002290726,.391080021858215],[.194993004202843,.342101991176605],[.388664990663528,.362284004688263],[.365961998701096,.355970978736877],[.343364000320435,.355356991291046],[.318785011768341,.35834002494812],[.301414996385574,.363156020641327],[.058132998645306,.319076001644135],[.301414996385574,.387449026107788],[.499987989664078,.618434011936188],[.415838003158569,.624195992946625],[.445681989192963,.566076993942261],[.465844005346298,.620640993118286],[.49992299079895,.351523995399475],[.288718998432159,.819945991039276],[.335278987884521,.852819979190826],[.440512001514435,.902418971061707],[.128294005990028,.791940987110138],[.408771991729736,.373893976211548],[.455606997013092,.451801002025604],[.499877005815506,.908990025520325],[.375436991453171,.924192011356354],[.11421000212431,.615022003650665],[.448662012815475,.695277988910675],[.4480200111866,.704632043838501],[.447111994028091,.715808033943176],[.444831997156143,.730794012546539],[.430011987686157,.766808986663818],[.406787008047104,.685672998428345],[.400738000869751,.681069016456604],[.392399996519089,.677703022956848],[.367855995893478,.663918972015381],[.247923001646996,.601333022117615],[.452769994735718,.420849978923798],[.43639200925827,.359887003898621],[.416164010763168,.368713974952698],[.413385987281799,.692366003990173],[.228018000721931,.683571994304657],[.468268007040024,.352671027183533],[.411361992359161,.804327011108398],[.499989002943039,.469825029373169],[.479153990745544,.442654013633728],[.499974012374878,.439637005329132],[.432112008333206,.493588984012604],[.499886006116867,.866917014122009],[.49991300702095,.821729004383087],[.456548988819122,.819200992584229],[.344549000263214,.745438992977142],[.37890899181366,.574010014533997],[.374292999505997,.780184984207153],[.319687992334366,.570737957954407],[.357154995203018,.604269981384277],[.295284003019333,.621580958366394],[.447750002145767,.862477004528046],[.410986006259918,.508723020553589],[.31395098567009,.775308012962341],[.354128003120422,.812552988529205],[.324548006057739,.703992962837219],[.189096003770828,.646299958229065],[.279776990413666,.71465802192688],[.1338230073452,.682700991630554],[.336768001317978,.644733011722565],[.429883986711502,.466521978378296],[.455527991056442,.548622965812683],[.437114000320435,.558896005153656],[.467287987470627,.529924988746643],[.414712011814117,.335219979286194],[.37704598903656,.322777986526489],[.344107985496521,.320150971412659],[.312875986099243,.32233202457428],[.283526003360748,.333190023899078],[.241245999932289,.382785975933075],[.102986000478268,.468762993812561],[.267612010240555,.424560010433197],[.297879010438919,.433175981044769],[.333433985710144,.433878004550934],[.366427004337311,.426115989685059],[.396012008190155,.416696012020111],[.420121014118195,.41022801399231],[.007561000064015,.480777025222778],[.432949006557465,.569517970085144],[.458638995885849,.479089021682739],[.473466008901596,.545744001865387],[.476087987422943,.563830018043518],[.468472003936768,.555056989192963],[.433990985155106,.582361996173859],[.483518004417419,.562983989715576],[.482482999563217,.57784903049469],[.42645001411438,.389798998832703],[.438998997211456,.39649498462677],[.450067013502121,.400434017181396],[.289712011814117,.368252992630005],[.276670008897781,.363372981548309],[.517862021923065,.471948027610779],[.710287988185883,.380764007568359],[.526226997375488,.573909997940063],[.895093023777008,.254140973091125],[.634069979190826,.409575998783112],[.661242008209229,.41302502155304],[.688880026340485,.409460008144379],[.725341975688934,.389131009578705],[.606630027294159,.40370500087738],[.654766023159027,.344011008739471],[.629905998706818,.346076011657715],[.680678009986877,.347265005111694],[.702096998691559,.353591024875641],[.75221198797226,.410804986953735],[.602918028831482,.842862963676453],[.719901978969574,.375599980354309],[.893692970275879,.399959981441498],[.790081977844238,.391354024410248],[.643998026847839,.534487962722778],[.528249025344849,.65040397644043],[.525849997997284,.680191040039062],[.560214996337891,.657229006290436],[.585384011268616,.66654098033905],[.549625992774963,.680860996246338],[.57122802734375,.682691991329193],[.624852001667023,.72809898853302],[.513050019741058,.547281980514526],[.51509702205658,.527251958847046],[.742246985435486,.314507007598877],[.598631024360657,.454979002475739],[.570338010787964,.548575043678284],[.578631997108459,.533622980117798],[.723087012767792,.532054007053375],[.516445994377136,.499638974666595],[.662801027297974,.282917976379395],[.70362401008606,.293271005153656],[.830704987049103,.193813979625702],[.552385985851288,.302568018436432],[.607609987258911,.353887975215912],[.645429015159607,.696707010269165],[.932694971561432,.730105042457581],[.557260990142822,.572826027870178],[.542901992797852,.584792017936707],[.6180260181427,.694710969924927],[.607590973377228,.694203019142151],[.722943007946014,.271963000297546],[.577413976192474,.563166975975037],[.614082992076874,.281386971473694],[.616907000541687,.255886018276215],[.668509006500244,.119913995265961],[.770092010498047,.232020974159241],[.635536015033722,.189248979091644],[.77039098739624,.299556016921997],[.826722025871277,.278755009174347],[.527121007442474,.666198015213013],[.553171992301941,.668527007102966],[.577238023281097,.673889994621277],[.554691970348358,.580065965652466],[.611896991729736,.693961024284363],[.59696102142334,.706539988517761],[.596370995044708,.693953037261963],[.539958000183105,.557139039039612],[.568841993808746,.692366003990173],[.547818005084991,.692366003990173],[.52461302280426,.692366003990173],[.534089982509613,.779141008853912],[.527670979499817,.736225962638855],[.526912987232208,.717857003211975],[.526877999305725,.704625964164734],[.526966989040375,.695277988910675],[.572058022022247,.695277988910675],[.573521018028259,.703539967536926],[.57683801651001,.711845993995667],[.581691026687622,.720062971115112],[.609944999217987,.639909982681274],[.986046016216278,.560034036636353],[.5867999792099,.69539999961853],[.590372025966644,.701822996139526],[.531915009021759,.601536989212036],[.577268004417419,.585934996604919],[.536915004253387,.593786001205444],[.627542972564697,.473352015018463],[.665585994720459,.495950996875763],[.588353991508484,.546862006187439],[.757824003696442,.14767599105835],[.709249973297119,.201507985591888],[.672684013843536,.256581008434296],[.600408971309662,.74900496006012],[.55826598405838,.261672019958496],[.570303976535797,.187870979309082],[.588165998458862,.109044015407562],[.711045026779175,.398952007293701],[.781069993972778,.435405015945435],[.587247014045715,.398931980133057],[.742869973182678,.355445981025696],[.572156012058258,.437651991844177],[.55186802148819,.536570012569427],[.821442008018494,.457556009292603],[.752701997756958,.457181990146637],[.71375697851181,.467626988887787],[.66711300611496,.460672974586487],[.631101012229919,.447153985500336],[.6008620262146,.432473003864288],[.523481011390686,.405627012252808],[.810747981071472,.523926019668579],[.771045982837677,.348959028720856],[.509127020835876,.562718033790588],[.595292985439301,.485023975372314],[.980530977249146,.401564002037048],[.573499977588654,.420000016689301],[.602994978427887,.548687994480133],[.733529984951019,.376977026462555],[.560611009597778,.519016981124878],[.967685997486115,.644356966018677],[.580985009670258,.387160003185272],[.537728011608124,.505385041236877],[.760966002941132,.779752969741821],[.801778972148895,.831938028335571],[.892440974712372,.54076099395752],[.816350996494293,.740260004997253],[.865594983100891,.333687007427216],[.614073991775513,.883246004581451],[.508952975273132,.579437971115112],[.617941975593567,.508316040039062],[.825608015060425,.397674977779388],[.681214988231659,.39623498916626],[.656635999679565,.400596976280212],[.603900015354156,.710216999053955],[.81208598613739,.588539004325867],[.56801301240921,.944564998149872],[.681007981300354,.898285031318665],[.733752012252808,.869701027870178],[.633830010890961,.398822009563446],[.606792986392975,.39553701877594],[.589659988880157,.391062021255493],[.805015981197357,.342108011245728],[.611334979534149,.362284004688263],[.634037971496582,.355970978736877],[.656635999679565,.355356991291046],[.681214988231659,.35834002494812],[.698584973812103,.363156020641327],[.941866993904114,.319076001644135],[.698584973812103,.387449026107788],[.584177017211914,.624107003211975],[.554318010807037,.566076993942261],[.534153997898102,.62064003944397],[.711217999458313,.819975018501282],[.664629995822906,.852871000766754],[.559099972248077,.902631998062134],[.871706008911133,.791940987110138],[.591234028339386,.373893976211548],[.544341027736664,.451583981513977],[.624562978744507,.924192011356354],[.88577002286911,.615028977394104],[.551338016986847,.695277988910675],[.551980018615723,.704632043838501],[.552887976169586,.715808033943176],[.555167973041534,.730794012546539],[.569944024085999,.767035007476807],[.593203008174896,.685675978660583],[.599261999130249,.681069016456604],[.607599973678589,.677703022956848],[.631937980651855,.663500010967255],[.752032995223999,.601315021514893],[.547226011753082,.420395016670227],[.563543975353241,.359827995300293],[.583841025829315,.368713974952698],[.586614012718201,.692366003990173],[.771915018558502,.683578014373779],[.531597018241882,.352482974529266],[.588370978832245,.804440975189209],[.52079701423645,.442565023899078],[.567984998226166,.493479013442993],[.543282985687256,.819254994392395],[.655317008495331,.745514988899231],[.621008992195129,.574018001556396],[.625559985637665,.78031200170517],[.680198013782501,.570719003677368],[.64276397228241,.604337990283966],[.704662978649139,.621529996395111],[.552012026309967,.862591981887817],[.589071989059448,.508637011051178],[.685944974422455,.775357007980347],[.645735025405884,.812640011310577],[.675342977046967,.703978002071381],[.810858011245728,.646304965019226],[.72012197971344,.714666962623596],[.866151988506317,.682704985141754],[.663187026977539,.644596993923187],[.570082008838654,.466325998306274],[.544561982154846,.548375964164734],[.562758982181549,.558784961700439],[.531987011432648,.530140042304993],[.585271000862122,.335177004337311],[.622952997684479,.32277899980545],[.655896008014679,.320163011550903],[.687132000923157,.322345972061157],[.716481983661652,.333200991153717],[.758756995201111,.382786989212036],[.897013008594513,.468769013881683],[.732392013072968,.424547016620636],[.70211398601532,.433162987232208],[.66652500629425,.433866024017334],[.633504986763,.426087975502014],[.603875994682312,.416586995124817],[.579657971858978,.409945011138916],[.992439985275269,.480777025222778],[.567192018032074,.569419980049133],[.54136598110199,.478899002075195],[.526564002037048,.546118021011353],[.523913025856018,.563830018043518],[.531529009342194,.555056989192963],[.566035985946655,.582329034805298],[.51631098985672,.563053965568542],[.5174720287323,.577877044677734],[.573594987392426,.389806985855103],[.560697972774506,.395331978797913],[.549755990505219,.399751007556915],[.710287988185883,.368252992630005],[.723330020904541,.363372981548309]],_l=[127,34,139,11,0,37,232,231,120,72,37,39,128,121,47,232,121,128,104,69,67,175,171,148,157,154,155,118,50,101,73,39,40,9,151,108,48,115,131,194,204,211,74,40,185,80,42,183,40,92,186,230,229,118,202,212,214,83,18,17,76,61,146,160,29,30,56,157,173,106,204,194,135,214,192,203,165,98,21,71,68,51,45,4,144,24,23,77,146,91,205,50,187,201,200,18,91,106,182,90,91,181,85,84,17,206,203,36,148,171,140,92,40,39,193,189,244,159,158,28,247,246,161,236,3,196,54,68,104,193,168,8,117,228,31,189,193,55,98,97,99,126,47,100,166,79,218,155,154,26,209,49,131,135,136,150,47,126,217,223,52,53,45,51,134,211,170,140,67,69,108,43,106,91,230,119,120,226,130,247,63,53,52,238,20,242,46,70,156,78,62,96,46,53,63,143,34,227,173,155,133,123,117,111,44,125,19,236,134,51,216,206,205,154,153,22,39,37,167,200,201,208,36,142,100,57,212,202,20,60,99,28,158,157,35,226,113,160,159,27,204,202,210,113,225,46,43,202,204,62,76,77,137,123,116,41,38,72,203,129,142,64,98,240,49,102,64,41,73,74,212,216,207,42,74,184,169,170,211,170,149,176,105,66,69,122,6,168,123,147,187,96,77,90,65,55,107,89,90,180,101,100,120,63,105,104,93,137,227,15,86,85,129,102,49,14,87,86,55,8,9,100,47,121,145,23,22,88,89,179,6,122,196,88,95,96,138,172,136,215,58,172,115,48,219,42,80,81,195,3,51,43,146,61,171,175,199,81,82,38,53,46,225,144,163,110,246,33,7,52,65,66,229,228,117,34,127,234,107,108,69,109,108,151,48,64,235,62,78,191,129,209,126,111,35,143,163,161,246,117,123,50,222,65,52,19,125,141,221,55,65,3,195,197,25,7,33,220,237,44,70,71,139,122,193,245,247,130,33,71,21,162,153,158,159,170,169,150,188,174,196,216,186,92,144,160,161,2,97,167,141,125,241,164,167,37,72,38,12,145,159,160,38,82,13,63,68,71,226,35,111,158,153,154,101,50,205,206,92,165,209,198,217,165,167,97,220,115,218,133,112,243,239,238,241,214,135,169,190,173,133,171,208,32,125,44,237,86,87,178,85,86,179,84,85,180,83,84,181,201,83,182,137,93,132,76,62,183,61,76,184,57,61,185,212,57,186,214,207,187,34,143,156,79,239,237,123,137,177,44,1,4,201,194,32,64,102,129,213,215,138,59,166,219,242,99,97,2,94,141,75,59,235,24,110,228,25,130,226,23,24,229,22,23,230,26,22,231,112,26,232,189,190,243,221,56,190,28,56,221,27,28,222,29,27,223,30,29,224,247,30,225,238,79,20,166,59,75,60,75,240,147,177,215,20,79,166,187,147,213,112,233,244,233,128,245,128,114,188,114,217,174,131,115,220,217,198,236,198,131,134,177,132,58,143,35,124,110,163,7,228,110,25,356,389,368,11,302,267,452,350,349,302,303,269,357,343,277,452,453,357,333,332,297,175,152,377,384,398,382,347,348,330,303,304,270,9,336,337,278,279,360,418,262,431,304,408,409,310,415,407,270,409,410,450,348,347,422,430,434,313,314,17,306,307,375,387,388,260,286,414,398,335,406,418,364,367,416,423,358,327,251,284,298,281,5,4,373,374,253,307,320,321,425,427,411,421,313,18,321,405,406,320,404,405,315,16,17,426,425,266,377,400,369,322,391,269,417,465,464,386,257,258,466,260,388,456,399,419,284,332,333,417,285,8,346,340,261,413,441,285,327,460,328,355,371,329,392,439,438,382,341,256,429,420,360,364,394,379,277,343,437,443,444,283,275,440,363,431,262,369,297,338,337,273,375,321,450,451,349,446,342,467,293,334,282,458,461,462,276,353,383,308,324,325,276,300,293,372,345,447,382,398,362,352,345,340,274,1,19,456,248,281,436,427,425,381,256,252,269,391,393,200,199,428,266,330,329,287,273,422,250,462,328,258,286,384,265,353,342,387,259,257,424,431,430,342,353,276,273,335,424,292,325,307,366,447,345,271,303,302,423,266,371,294,455,460,279,278,294,271,272,304,432,434,427,272,407,408,394,430,431,395,369,400,334,333,299,351,417,168,352,280,411,325,319,320,295,296,336,319,403,404,330,348,349,293,298,333,323,454,447,15,16,315,358,429,279,14,15,316,285,336,9,329,349,350,374,380,252,318,402,403,6,197,419,318,319,325,367,364,365,435,367,397,344,438,439,272,271,311,195,5,281,273,287,291,396,428,199,311,271,268,283,444,445,373,254,339,263,466,249,282,334,296,449,347,346,264,447,454,336,296,299,338,10,151,278,439,455,292,407,415,358,371,355,340,345,372,390,249,466,346,347,280,442,443,282,19,94,370,441,442,295,248,419,197,263,255,359,440,275,274,300,383,368,351,412,465,263,467,466,301,368,389,380,374,386,395,378,379,412,351,419,436,426,322,373,390,388,2,164,393,370,462,461,164,0,267,302,11,12,374,373,387,268,12,13,293,300,301,446,261,340,385,384,381,330,266,425,426,423,391,429,355,437,391,327,326,440,457,438,341,382,362,459,457,461,434,430,394,414,463,362,396,369,262,354,461,457,316,403,402,315,404,403,314,405,404,313,406,405,421,418,406,366,401,361,306,408,407,291,409,408,287,410,409,432,436,410,434,416,411,264,368,383,309,438,457,352,376,401,274,275,4,421,428,262,294,327,358,433,416,367,289,455,439,462,370,326,2,326,370,305,460,455,254,449,448,255,261,446,253,450,449,252,451,450,256,452,451,341,453,452,413,464,463,441,413,414,258,442,441,257,443,442,259,444,443,260,445,444,467,342,445,459,458,250,289,392,290,290,328,460,376,433,435,250,290,392,411,416,433,341,463,464,453,464,465,357,465,412,343,412,399,360,363,440,437,399,456,420,456,363,401,435,288,372,383,353,339,255,249,448,261,255,133,243,190,133,155,112,33,246,247,33,130,25,398,384,286,362,398,414,362,463,341,263,359,467,263,249,255,466,467,260,75,60,166,238,239,79,162,127,139,72,11,37,121,232,120,73,72,39,114,128,47,233,232,128,103,104,67,152,175,148,173,157,155,119,118,101,74,73,40,107,9,108,49,48,131,32,194,211,184,74,185,191,80,183,185,40,186,119,230,118,210,202,214,84,83,17,77,76,146,161,160,30,190,56,173,182,106,194,138,135,192,129,203,98,54,21,68,5,51,4,145,144,23,90,77,91,207,205,187,83,201,18,181,91,182,180,90,181,16,85,17,205,206,36,176,148,140,165,92,39,245,193,244,27,159,28,30,247,161,174,236,196,103,54,104,55,193,8,111,117,31,221,189,55,240,98,99,142,126,100,219,166,218,112,155,26,198,209,131,169,135,150,114,47,217,224,223,53,220,45,134,32,211,140,109,67,108,146,43,91,231,230,120,113,226,247,105,63,52,241,238,242,124,46,156,95,78,96,70,46,63,116,143,227,116,123,111,1,44,19,3,236,51,207,216,205,26,154,22,165,39,167,199,200,208,101,36,100,43,57,202,242,20,99,56,28,157,124,35,113,29,160,27,211,204,210,124,113,46,106,43,204,96,62,77,227,137,116,73,41,72,36,203,142,235,64,240,48,49,64,42,41,74,214,212,207,183,42,184,210,169,211,140,170,176,104,105,69,193,122,168,50,123,187,89,96,90,66,65,107,179,89,180,119,101,120,68,63,104,234,93,227,16,15,85,209,129,49,15,14,86,107,55,9,120,100,121,153,145,22,178,88,179,197,6,196,89,88,96,135,138,136,138,215,172,218,115,219,41,42,81,5,195,51,57,43,61,208,171,199,41,81,38,224,53,225,24,144,110,105,52,66,118,229,117,227,34,234,66,107,69,10,109,151,219,48,235,183,62,191,142,129,126,116,111,143,7,163,246,118,117,50,223,222,52,94,19,141,222,221,65,196,3,197,45,220,44,156,70,139,188,122,245,139,71,162,145,153,159,149,170,150,122,188,196,206,216,92,163,144,161,164,2,167,242,141,241,0,164,37,11,72,12,144,145,160,12,38,13,70,63,71,31,226,111,157,158,154,36,101,205,203,206,165,126,209,217,98,165,97,237,220,218,237,239,241,210,214,169,140,171,32,241,125,237,179,86,178,180,85,179,181,84,180,182,83,181,194,201,182,177,137,132,184,76,183,185,61,184,186,57,185,216,212,186,192,214,187,139,34,156,218,79,237,147,123,177,45,44,4,208,201,32,98,64,129,192,213,138,235,59,219,141,242,97,97,2,141,240,75,235,229,24,228,31,25,226,230,23,229,231,22,230,232,26,231,233,112,232,244,189,243,189,221,190,222,28,221,223,27,222,224,29,223,225,30,224,113,247,225,99,60,240,213,147,215,60,20,166,192,187,213,243,112,244,244,233,245,245,128,188,188,114,174,134,131,220,174,217,236,236,198,134,215,177,58,156,143,124,25,110,7,31,228,25,264,356,368,0,11,267,451,452,349,267,302,269,350,357,277,350,452,357,299,333,297,396,175,377,381,384,382,280,347,330,269,303,270,151,9,337,344,278,360,424,418,431,270,304,409,272,310,407,322,270,410,449,450,347,432,422,434,18,313,17,291,306,375,259,387,260,424,335,418,434,364,416,391,423,327,301,251,298,275,281,4,254,373,253,375,307,321,280,425,411,200,421,18,335,321,406,321,320,405,314,315,17,423,426,266,396,377,369,270,322,269,413,417,464,385,386,258,248,456,419,298,284,333,168,417,8,448,346,261,417,413,285,326,327,328,277,355,329,309,392,438,381,382,256,279,429,360,365,364,379,355,277,437,282,443,283,281,275,363,395,431,369,299,297,337,335,273,321,348,450,349,359,446,467,283,293,282,250,458,462,300,276,383,292,308,325,283,276,293,264,372,447,346,352,340,354,274,19,363,456,281,426,436,425,380,381,252,267,269,393,421,200,428,371,266,329,432,287,422,290,250,328,385,258,384,446,265,342,386,387,257,422,424,430,445,342,276,422,273,424,306,292,307,352,366,345,268,271,302,358,423,371,327,294,460,331,279,294,303,271,304,436,432,427,304,272,408,395,394,431,378,395,400,296,334,299,6,351,168,376,352,411,307,325,320,285,295,336,320,319,404,329,330,349,334,293,333,366,323,447,316,15,315,331,358,279,317,14,316,8,285,9,277,329,350,253,374,252,319,318,403,351,6,419,324,318,325,397,367,365,288,435,397,278,344,439,310,272,311,248,195,281,375,273,291,175,396,199,312,311,268,276,283,445,390,373,339,295,282,296,448,449,346,356,264,454,337,336,299,337,338,151,294,278,455,308,292,415,429,358,355,265,340,372,388,390,466,352,346,280,295,442,282,354,19,370,285,441,295,195,248,197,457,440,274,301,300,368,417,351,465,251,301,389,385,380,386,394,395,379,399,412,419,410,436,322,387,373,388,326,2,393,354,370,461,393,164,267,268,302,12,386,374,387,312,268,13,298,293,301,265,446,340,380,385,381,280,330,425,322,426,391,420,429,437,393,391,326,344,440,438,458,459,461,364,434,394,428,396,262,274,354,457,317,316,402,316,315,403,315,314,404,314,313,405,313,421,406,323,366,361,292,306,407,306,291,408,291,287,409,287,432,410,427,434,411,372,264,383,459,309,457,366,352,401,1,274,4,418,421,262,331,294,358,435,433,367,392,289,439,328,462,326,94,2,370,289,305,455,339,254,448,359,255,446,254,253,449,253,252,450,252,256,451,256,341,452,414,413,463,286,441,414,286,258,441,258,257,442,257,259,443,259,260,444,260,467,445,309,459,250,305,289,290,305,290,460,401,376,435,309,250,392,376,411,433,453,341,464,357,453,465,343,357,412,437,343,399,344,360,440,420,437,456,360,420,363,361,401,288,265,372,353,390,339,249,339,448,255];var wge=[127,234,132,58,172,150,149,148,152,377,378,379,397,288,361,454,356,70,63,105,66,107,336,296,334,293,300,168,6,195,4,98,97,2,326,327,33,160,158,133,153,144,362,385,387,263,373,380,57,40,37,0,267,270,287,321,314,17,84,91,78,81,13,311,308,402,14,178],kge=[33,133,362,263,1,62,308,159,145,386,374,6,102,331,2,13,14,70,105,107,336,334,300,54,10,284,50,280,234,454,58,288,152],Sge=[33,133,362,263,1,78,308],J2e=wge.map(e=>Np[e]),Q2e=kge.map(e=>Np[e]),e1e=Sge.map(e=>Np[e]);var Ep=e=>[Math.abs(e.endPoint[0]-e.startPoint[0]),Math.abs(e.endPoint[1]-e.startPoint[1])],s0=e=>[e.startPoint[0]+(e.endPoint[0]-e.startPoint[0])/2,e.startPoint[1]+(e.endPoint[1]-e.startPoint[1])/2],Wx=(e,t)=>e?[Math.trunc(Math.max(0,e.startPoint[0])),Math.trunc(Math.max(0,e.startPoint[1])),Math.trunc(Math.min(t.shape[2]||0,e.endPoint[0])-Math.max(0,e.startPoint[0])),Math.trunc(Math.min(t.shape[1]||0,e.endPoint[1])-Math.max(0,e.startPoint[1]))]:[0,0,0,0],Vx=(e,t)=>e?[e.startPoint[0]/(t.shape[2]||0),e.startPoint[1]/(t.shape[1]||0),(e.endPoint[0]-e.startPoint[0])/(t.shape[2]||0),(e.endPoint[1]-e.startPoint[1])/(t.shape[1]||0)]:[0,0,0,0],Q6=(e,t)=>{let n=[e.startPoint[0]*t[0],e.startPoint[1]*t[1]],s=[e.endPoint[0]*t[0],e.endPoint[1]*t[1]];return{startPoint:n,endPoint:s,landmarks:e.landmarks,confidence:e.confidence}},Ux=(e,t,n)=>{let s=t.shape[1],r=t.shape[2],a=$e.cropAndResize(t,[[e.startPoint[1]/s,e.startPoint[0]/r,e.endPoint[1]/s,e.endPoint[0]/r]],[0],n),o=pe(a,255);return se(a),o},Rp=(e,t)=>{let n=s0(e),s=Ep(e),r=[t*s[0]/2,t*s[1]/2];return{startPoint:[n[0]-r[0],n[1]-r[1]],endPoint:[n[0]+r[0],n[1]+r[1]],landmarks:e.landmarks,confidence:e.confidence}},$p=e=>{let t=s0(e),n=Ep(e),s=Math.max(...n)/2;return{startPoint:[Math.round(t[0]-s),Math.round(t[1]-s)],endPoint:[Math.round(t[0]+s),Math.round(t[1]+s)],landmarks:e.landmarks,confidence:e.confidence}},r0=e=>{let t=e.map(s=>s[0]),n=e.map(s=>s[1]);return{startPoint:[Math.min(...t),Math.min(...n)],endPoint:[Math.max(...t),Math.max(...n)],landmarks:e}},Gx=[[1,0,0],[0,1,0],[0,0,1]],Ige=e=>e-2*Math.PI*Math.floor((e+Math.PI)/(2*Math.PI)),Cge=(e,t)=>Ige(Math.PI/2-Math.atan2(-(t[1]-e[1]),t[0]-e[0]));var e8=(e,t)=>[[1,0,e],[0,1,t],[0,0,1]],Dl=(e,t)=>{let n=0;for(let s=0;s<e.length;s++)n+=e[s]*t[s];return n},Tge=(e,t)=>{let n=[];for(let s=0;s<e.length;s++)n.push(e[s][t]);return n},t8=(e,t)=>{let n=[],s=e.length;for(let r=0;r<s;r++){n.push([]);for(let a=0;a<s;a++)n[r].push(Dl(e[r],Tge(t,a)))}return n},n8=(e,t)=>{let n=Math.cos(e),s=Math.sin(e),r=[[n,-s,0],[s,n,0],[0,0,1]],a=e8(t[0],t[1]),o=t8(a,r),i=e8(-t[0],-t[1]);return t8(o,i)},Nge=e=>{let t=[[e[0][0],e[1][0]],[e[0][1],e[1][1]]],n=[e[0][2],e[1][2]],s=[-Dl(t[0],n),-Dl(t[1],n)];return[t[0].concat(s[0]),t[1].concat(s[1]),[0,0,1]]},Ege=(e,t)=>[Dl(e,t[0]),Dl(e,t[1])];function s8(e){let t={strides:[e/16,e/8],anchors:[2,6]},n=[];for(let s=0;s<t.strides.length;s++){let r=t.strides[s],a=Math.floor((e+r-1)/r),o=Math.floor((e+r-1)/r),i=t.anchors[s];for(let l=0;l<a;l++){let c=r*(l+.5);for(let u=0;u<o;u++){let d=r*(u+.5);for(let p=0;p<i;p++)n.push([d,c])}}}return n}function r8(e,t,n,s,r){let a=Ep(t),o=e.map(p=>[a[0]/r*(p[0]-r/2),a[1]/r*(p[1]-r/2),p[2]||0]),i=n&&n!==0&&Math.abs(n)>.2,l=i?n8(n,[0,0]):Gx,c=i?o.map(p=>[...Ege(p,l),p[2]]):o,u=i?Nge(s):Gx,d=[...s0({startPoint:t.startPoint,endPoint:t.endPoint}),1];return c.map(p=>[Math.round(p[0]+Dl(d,u[0])),Math.round(p[1]+Dl(d,u[1])),Math.round(p[2]||0)])}function Hx(e,t,n,s){let r=t.landmarks.length>=Lx.count?Lx.symmetryLine:Tp.symmetryLine,a=0,o=Gx,i;if(e&&Ae.kernels.includes("rotatewithoffset"))if(a=Cge(t.landmarks[r[0]],t.landmarks[r[1]]),a&&a!==0&&Math.abs(a)>.2){let c=s0({startPoint:t.startPoint,endPoint:t.endPoint}),u=[c[0]/n.shape[2],c[1]/n.shape[1]],d=$e.rotateWithOffset(n,a,0,u);o=n8(-a,c),i=Ux(t,d,[s,s]),se(d)}else i=Ux(t,n,[s,s]);else i=Ux(t,n,[s,s]);return[a,o,i]}var a8=6,Os,o8=[],i8=null,Ms=0,jx=()=>Ms;async function l8(e){var t,n;return Ae.initial&&(Os=null),Os?e.debug&&re("cached model:",Os.modelUrl):(Os=await et(nt(e.modelBasePath,((t=e.face.detector)==null?void 0:t.modelPath)||"")),!Os||!Os.modelUrl?re("load model failed:",(n=e.face.detector)==null?void 0:n.modelPath):e.debug&&re("load model:",Os.modelUrl)),Ms=Os.inputs[0].shape?Os.inputs[0].shape[2]:0,Ms===-1&&(Ms=64),o8=s8(Ms),i8=lr(o8),Os}function Rge(e){let t={};t.boxStarts=De(e,[0,1],[-1,2]),t.centers=le(t.boxStarts,i8),t.boxSizes=De(e,[0,3],[-1,2]),t.boxSizesNormalized=pe(t.boxSizes,Ms),t.centersNormalized=pe(t.centers,Ms),t.halfBoxSize=pe(t.boxSizesNormalized,2),t.starts=me(t.centersNormalized,t.halfBoxSize),t.ends=le(t.centersNormalized,t.halfBoxSize),t.startNormalized=L(t.starts,Ms),t.endNormalized=L(t.ends,Ms);let n=Pu([t.startNormalized,t.endNormalized],1);return Object.keys(t).forEach(s=>se(t[s])),n}async function u8(e,t){var i,l,c,u;if(!e||e.isDisposedInternal||e.shape.length!==4||e.shape[1]<1||e.shape[2]<1)return{boxes:[]};let n={};n.resized=$e.resizeBilinear(e,[Ms,Ms]),n.div=pe(n.resized,127.5),n.normalized=me(n.div,.5);let s=Os==null?void 0:Os.execute(n.normalized);if(Array.isArray(s)){let d=s.sort((p,h)=>p.size-h.size);n.concat384=vt([d[0],d[2]],2),n.concat512=vt([d[1],d[3]],2),n.concat=vt([n.concat512,n.concat384],1),n.batch=ot(n.concat,0)}else n.batch=ot(s);se(s),n.boxes=Rge(n.batch),n.logits=De(n.batch,[0,0],[-1,1]),n.sigmoid=cs(n.logits),n.scores=ot(n.sigmoid),n.nms=await $e.nonMaxSuppressionAsync(n.boxes,n.scores,((i=t.face.detector)==null?void 0:i.maxDetected)||0,((l=t.face.detector)==null?void 0:l.iouThreshold)||0,((c=t.face.detector)==null?void 0:c.minConfidence)||0);let r=await n.nms.array(),a=[],o=await n.scores.data();for(let d=0;d<r.length;d++){let p=o[r[d]];if(p>(((u=t.face.detector)==null?void 0:u.minConfidence)||0)){let h={};h.bbox=De(n.boxes,[r[d],0],[1,-1]),h.slice=De(n.batch,[r[d],a8-1],[1,-1]),h.squeeze=ot(h.slice),h.landmarks=G(h.squeeze,[a8,-1]),h.startPoint=De(h.bbox,[0,0],[-1,2]),h.endPoint=De(h.bbox,[0,2],[-1,2]),a.push({box:{startPoint:await h.startPoint.data(),endPoint:await h.endPoint.data()},landmarks:await h.landmarks.array(),confidence:p}),Object.keys(h).forEach(f=>se(h[f]))}}return Object.keys(n).forEach(d=>se(n[d])),{boxes:a,scaleFactor:[e.shape[2]/Ms,e.shape[1]/Ms]}}var Kx={};Gc(Kx,{connected:()=>Xx,kpt:()=>qx});var qx=["nose","leftEyeInside","leftEye","leftEyeOutside","rightEyeInside","rightEye","rightEyeOutside","leftEar","rightEar","leftMouth","rightMouth","leftShoulder","rightShoulder","leftElbow","rightElbow","leftWrist","rightWrist","leftPalm","rightPalm","leftIndex","rightIndex","leftPinky","rightPinky","leftHip","rightHip","leftKnee","rightKnee","leftAnkle","rightAnkle","leftHeel","rightHeel","leftFoot","rightFoot","bodyCenter","bodyTop","leftThumb","leftHand","rightThumb","rightHand"],Xx={leftLeg:["leftHip","leftKnee","leftAnkle","leftHeel","leftFoot"],rightLeg:["rightHip","rightKnee","rightAnkle","rightHeel","rightFoot"],torso:["leftShoulder","rightShoulder","rightHip","leftHip","leftShoulder"],leftArm:["leftShoulder","leftElbow","leftWrist","leftPalm"],rightArm:["rightShoulder","rightElbow","rightWrist","rightPalm"],leftHand:[],rightHand:[],head:[]};var c8={initial:!0},un=[null,null],Xo=[[0,0],[0,0]],Zx=Number.MAX_SAFE_INTEGER,Yx,a0=null,Ko=[[0,0],[0,0],[0,0],[0,0]],d8=0;async function p8(e){var t,n,s;if(c8.initial&&(un[0]=null),!un[0]&&((t=e.body.detector)==null?void 0:t.modelPath)){un[0]=await et(nt(e.modelBasePath,((n=e.body.detector)==null?void 0:n.modelPath)||""));let r=Object.values(un[0].modelSignature.inputs);Xo[0][0]=Array.isArray(r)?parseInt(r[0].tensorShape.dim[1].size):0,Xo[0][1]=Array.isArray(r)?parseInt(r[0].tensorShape.dim[2].size):0,!un[0]||!un[0].modelUrl?re("load model failed:",(s=e.body.detector)==null?void 0:s.modelPath):e.debug&&re("load model:",un[0].modelUrl)}else e.debug&&un[0]&&re("cached model:",un[0].modelUrl);return un[0]}async function h8(e){var t;if(c8.initial&&(un[1]=null),un[1])e.debug&&re("cached model:",un[1].modelUrl);else{un[1]=await et(nt(e.modelBasePath,e.body.modelPath||""));let n=Object.values(un[1].modelSignature.inputs);Xo[1][0]=Array.isArray(n)?parseInt(n[0].tensorShape.dim[1].size):0,Xo[1][1]=Array.isArray(n)?parseInt(n[0].tensorShape.dim[2].size):0,((t=e.body.modelPath)==null?void 0:t.includes("lite"))?Yx=["ld_3d","output_segmentation","output_heatmap","world_3d","output_poseflag"]:Yx=["Identity","Identity_2","Identity_3","Identity_4","Identity_1"],!un[1]||!un[1].modelUrl?re("load model failed:",e.body.modelPath):e.debug&&re("load model:",un[1].modelUrl)}return un[1]}function $ge(e,t){let n=e.map(o=>o.position[0]),s=e.map(o=>o.position[1]),r=[Math.min(...n),Math.min(...s),Math.max(...n)-Math.min(...n),Math.max(...s)-Math.min(...s)],a=[r[0]/t[0],r[1]/t[1],r[2]/t[0],r[3]/t[1]];return{keypointsBox:r,keypointsBoxRaw:a}}async function _ge(e){let t={};if(!e.shape||!e.shape[1]||!e.shape[2])return e;Ko=[[0,0],[e.shape[2]>e.shape[1]?Math.trunc((e.shape[2]-e.shape[1])/2):0,e.shape[2]>e.shape[1]?Math.trunc((e.shape[2]-e.shape[1])/2):0],[e.shape[1]>e.shape[2]?Math.trunc((e.shape[1]-e.shape[2])/2):0,e.shape[1]>e.shape[2]?Math.trunc((e.shape[1]-e.shape[2])/2):0],[0,0]],t.pad=qs(e,Ko),t.resize=$e.resizeBilinear(t.pad,[Xo[1][0],Xo[1][1]]);let n=pe(t.resize,255);return Object.keys(t).forEach(s=>se(t[s])),n}function Dge(e,t){for(let n of e)n.position=[n.position[0]*(t[0]+Ko[2][0]+Ko[2][1])/t[0]-Ko[2][0],n.position[1]*(t[1]+Ko[1][0]+Ko[1][1])/t[1]-Ko[1][0],n.position[2]],n.positionRaw=[n.position[0]/t[0],n.position[1]/t[1],n.position[2]];return e}var f8=e=>1-1/(1+Math.exp(e));async function Pge(e,t,n){var h;let s={};s.input=await _ge(e),[s.ld,s.segmentation,s.heatmap,s.world,s.poseflag]=(h=un[1])==null?void 0:h.execute(s.input,Yx);let r=(await s.poseflag.data())[0],a=Math.max(0,(r-.8)/(1-.8)),o=await s.ld.data(),i=[],l=5;for(let f=0;f<o.length/l;f++){let m=f8(o[l*f+3]),g=f8(o[l*f+4]),A=Math.trunc(100*m*g*a)/100,x=[o[l*f+0]/Xo[1][0],o[l*f+1]/Xo[1][1],o[l*f+2]+0],y=[Math.trunc(n[0]*x[0]),Math.trunc(n[1]*x[1]),x[2]];i.push({part:qx[f],positionRaw:x,position:y,score:A})}if(a<(t.body.minConfidence||0))return null;let c=Dge(i,n),u=$ge(c,[n[0],n[1]]);Object.keys(s).forEach(f=>se(s[f]));let d={};for(let[f,m]of Object.entries(Xx)){let g=[];for(let A=0;A<m.length-1;A++){let x=c.find(b=>b.part===m[A]),y=c.find(b=>b.part===m[A+1]);x&&y&&x.score>(t.body.minConfidence||0)&&y.score>(t.body.minConfidence||0)&&g.push([x.position,y.position])}d[f]=g}return{id:0,score:Math.trunc(100*a)/100,box:u.keypointsBox,boxRaw:u.keypointsBoxRaw,keypoints:c,annotations:d}}async function Jx(e,t){let n=[e.shape[2]||0,e.shape[1]||0],s=(t.body.skipTime||0)>he()-d8,r=Zx<(t.body.skipFrames||0);return t.skipAllowed&&s&&r&&a0!==null?Zx++:(a0=await Pge(e,t,n),d8=he(),Zx=0),a0?[a0]:[]}var wc=[{class:1,label:"person"},{class:2,label:"bicycle"},{class:3,label:"car"},{class:4,label:"motorcycle"},{class:5,label:"airplane"},{class:6,label:"bus"},{class:7,label:"train"},{class:8,label:"truck"},{class:9,label:"boat"},{class:10,label:"traffic light"},{class:11,label:"fire hydrant"},{class:12,label:"stop sign"},{class:13,label:"parking meter"},{class:14,label:"bench"},{class:15,label:"bird"},{class:16,label:"cat"},{class:17,label:"dog"},{class:18,label:"horse"},{class:19,label:"sheep"},{class:20,label:"cow"},{class:21,label:"elephant"},{class:22,label:"bear"},{class:23,label:"zebra"},{class:24,label:"giraffe"},{class:25,label:"backpack"},{class:26,label:"umbrella"},{class:27,label:"handbag"},{class:28,label:"tie"},{class:29,label:"suitcase"},{class:30,label:"frisbee"},{class:31,label:"skis"},{class:32,label:"snowboard"},{class:33,label:"sports ball"},{class:34,label:"kite"},{class:35,label:"baseball bat"},{class:36,label:"baseball glove"},{class:37,label:"skateboard"},{class:38,label:"surfboard"},{class:39,label:"tennis racket"},{class:40,label:"bottle"},{class:41,label:"wine glass"},{class:42,label:"cup"},{class:43,label:"fork"},{class:44,label:"knife"},{class:45,label:"spoon"},{class:46,label:"bowl"},{class:47,label:"banana"},{class:48,label:"apple"},{class:49,label:"sandwich"},{class:50,label:"orange"},{class:51,label:"broccoli"},{class:52,label:"carrot"},{class:53,label:"hot dog"},{class:54,label:"pizza"},{class:55,label:"donut"},{class:56,label:"cake"},{class:57,label:"chair"},{class:58,label:"couch"},{class:59,label:"potted plant"},{class:60,label:"bed"},{class:61,label:"dining table"},{class:62,label:"toilet"},{class:63,label:"tv"},{class:64,label:"laptop"},{class:65,label:"mouse"},{class:66,label:"remote"},{class:67,label:"keyboard"},{class:68,label:"cell phone"},{class:69,label:"microwave"},{class:70,label:"oven"},{class:71,label:"toaster"},{class:72,label:"sink"},{class:73,label:"refrigerator"},{class:74,label:"book"},{class:75,label:"clock"},{class:76,label:"vase"},{class:77,label:"scissors"},{class:78,label:"teddy bear"},{class:79,label:"hair drier"},{class:80,label:"toothbrush"}];var Qs,Pl=0,Qx=[],m8=0,eb=Number.MAX_SAFE_INTEGER;async function g8(e){if(Ae.initial&&(Qs=null),Qs)e.debug&&re("cached model:",Qs.modelUrl);else{Qs=await et(nt(e.modelBasePath,e.object.modelPath||""));let t=Object.values(Qs.modelSignature.inputs);Pl=Array.isArray(t)?parseInt(t[0].tensorShape.dim[2].size):0,!Qs||!Qs.modelUrl?re("load model failed:",e.object.modelPath):e.debug&&re("load model:",Qs.modelUrl)}return Qs}async function Fge(e,t,n){if(!e)return[];let s=[],r=await e.array(),a=ot(e);se(e);let o=nn(a,6,1);se(a);let i=An([o[1],o[0],o[3],o[2]],1),l=ot(i);se(i);let c=ot(o[4]),u=ot(o[5]);o.forEach(f=>se(f));let d=await $e.nonMaxSuppressionAsync(l,c,n.object.maxDetected,n.object.iouThreshold,n.object.minConfidence);se(l),se(c),se(u);let p=await d.data();se(d);let h=0;for(let f of p){let m=Math.trunc(100*r[0][f][4])/100,g=r[0][f][5],A=wc[g].label,[x,y]=[r[0][f][0]/Pl,r[0][f][1]/Pl],b=[x,y,r[0][f][2]/Pl-x,r[0][f][3]/Pl-y],w=[Math.trunc(b[0]*t[0]),Math.trunc(b[1]*t[1]),Math.trunc(b[2]*t[0]),Math.trunc(b[3]*t[1])];s.push({id:h++,score:m,class:g,label:A,box:w,boxRaw:b})}return s}async function tb(e,t){let n=(t.object.skipTime||0)>he()-m8,s=eb<(t.object.skipFrames||0);return t.skipAllowed&&n&&s&&Qx.length>0?(eb++,Qx):(eb=0,new Promise(async r=>{let a=[e.shape[2],e.shape[1]],o=$e.resizeBilinear(e,[Pl,Pl]),i=t.object.enabled?Qs==null?void 0:Qs.execute(o,["tower_0/detections"]):null;m8=he(),se(o);let l=await Fge(i,a,t);Qx=l,r(l)}))}var rb={};Gc(rb,{connected:()=>sb,kpt:()=>nb});var nb=["head","neck","rightShoulder","rightElbow","rightWrist","chest","leftShoulder","leftElbow","leftWrist","bodyCenter","rightHip","rightKnee","rightAnkle","leftHip","leftKnee","leftAnkle"],sb={leftLeg:["leftHip","leftKnee","leftAnkle"],rightLeg:["rightHip","rightKnee","rightAnkle"],torso:["leftShoulder","rightShoulder","rightHip","leftHip","leftShoulder"],leftArm:["leftShoulder","leftElbow","leftWrist"],rightArm:["rightShoulder","rightElbow","rightWrist"],head:[]};var cn,A8=0,Kn={id:0,keypoints:[],box:[0,0,0,0],boxRaw:[0,0,0,0],score:0,annotations:{}},ab=Number.MAX_SAFE_INTEGER;async function ob(e){return Ae.initial&&(cn=null),cn?e.debug&&re("cached model:",cn.modelUrl):(cn=await et(nt(e.modelBasePath,e.body.modelPath||"")),!cn||!cn.modelUrl?re("load model failed:",e.body.modelPath):e.debug&&re("load model:",cn.modelUrl)),cn}function Oge(e,t){let[n,s]=e.shape;return q(()=>{let r=G(e,[s*n]),a=gn(r,0).dataSync()[0];if(a>t){let o=Hs(r,0),i=Rd(o,n).dataSync()[0],l=pe(o,Re(n,"int32")).dataSync()[0];return[i,l,a]}return[0,0,a]})}async function ib(e,t){let n=(t.body.skipTime||0)>he()-A8,s=ab<(t.body.skipFrames||0);return t.skipAllowed&&n&&s&&Object.keys(Kn.keypoints).length>0?(ab++,[Kn]):(ab=0,new Promise(async r=>{var d;let a=q(()=>{if(!(cn==null?void 0:cn.inputs[0].shape))return null;let p=$e.resizeBilinear(e,[cn.inputs[0].shape[2],cn.inputs[0].shape[1]],!1);return L(p,2).sub(1)}),o;if(t.body.enabled&&(o=cn==null?void 0:cn.execute(a)),A8=he(),se(a),o){Kn.keypoints.length=0;let p=o.squeeze();se(o);let h=p.unstack(2);se(p);for(let f=0;f<h.length;f++){let[m,g,A]=Oge(h[f],t.body.minConfidence);A>(((d=t.body)==null?void 0:d.minConfidence)||0)&&Kn.keypoints.push({score:Math.round(100*A)/100,part:nb[f],positionRaw:[m/cn.inputs[0].shape[2],g/cn.inputs[0].shape[1]],position:[Math.round(e.shape[2]*m/cn.inputs[0].shape[2]),Math.round(e.shape[1]*g/cn.inputs[0].shape[1])]})}h.forEach(f=>se(f))}Kn.score=Kn.keypoints.reduce((p,h)=>h.score>p?h.score:p,0);let i=Kn.keypoints.map(p=>p.position[0]),l=Kn.keypoints.map(p=>p.position[1]);Kn.box=[Math.min(...i),Math.min(...l),Math.max(...i)-Math.min(...i),Math.max(...l)-Math.min(...l)];let c=Kn.keypoints.map(p=>p.positionRaw[0]),u=Kn.keypoints.map(p=>p.positionRaw[1]);Kn.boxRaw=[Math.min(...c),Math.min(...u),Math.max(...c)-Math.min(...c),Math.max(...u)-Math.min(...u)];for(let[p,h]of Object.entries(sb)){let f=[];for(let m=0;m<h.length-1;m++){let g=Kn.keypoints.find(x=>x.part===h[m]),A=Kn.keypoints.find(x=>x.part===h[m+1]);g&&A&&g.score>(t.body.minConfidence||0)&&A.score>(t.body.minConfidence||0)&&f.push([g.position,A.position])}Kn.annotations[p]=f}r([Kn])}))}var Mge=["angry","disgust","fear","happy","sad","surprise","neutral"],Zn,o0=[],y8=0,x8=0,lb=Number.MAX_SAFE_INTEGER,ub=[.2989,.587,.114];async function b8(e){var t,n;return Ae.initial&&(Zn=null),Zn?e.debug&&re("cached model:",Zn.modelUrl):(Zn=await et(nt(e.modelBasePath,((t=e.face.emotion)==null?void 0:t.modelPath)||"")),!Zn||!Zn.modelUrl?re("load model failed:",(n=e.face.emotion)==null?void 0:n.modelPath):e.debug&&re("load model:",Zn.modelUrl)),Zn}async function cb(e,t,n,s){var o,i;if(!Zn)return null;let r=lb<(((o=t.face.emotion)==null?void 0:o.skipFrames)||0),a=(((i=t.face.emotion)==null?void 0:i.skipTime)||0)>he()-x8;return t.skipAllowed&&a&&r&&y8===s&&o0[n]&&o0[n].length>0?(lb++,o0[n]):(lb=0,new Promise(async l=>{var u,d;let c=[];if((u=t.face.emotion)==null?void 0:u.enabled){let p={},h=(Zn==null?void 0:Zn.inputs[0].shape)?Zn.inputs[0].shape[2]:0;p.resize=$e.resizeBilinear(e,[h,h],!1),[p.red,p.green,p.blue]=nn(p.resize,3,3),p.redNorm=L(p.red,ub[0]),p.greenNorm=L(p.green,ub[1]),p.blueNorm=L(p.blue,ub[2]),p.grayscale=sf([p.redNorm,p.greenNorm,p.blueNorm]),p.grayscaleSub=me(p.grayscale,.5),p.grayscaleMul=L(p.grayscaleSub,2),p.emotion=Zn==null?void 0:Zn.execute(p.grayscaleMul),x8=he();let f=await p.emotion.data();for(let m=0;m<f.length;m++)f[m]>(((d=t.face.emotion)==null?void 0:d.minConfidence)||0)&&c.push({score:Math.min(.99,Math.trunc(100*f[m])/100),emotion:Mge[m]});c.sort((m,g)=>g.score-m.score),Object.keys(p).forEach(m=>se(p[m]))}o0[n]=c,y8=s,l(c)}))}var er,Zo=0,zge=2.3,db=Br.leftEyeLower0,pb=Br.rightEyeLower0,kc={leftBounds:[db[0],db[db.length-1]],rightBounds:[pb[0],pb[pb.length-1]]},Sc={upperCenter:3,lowerCenter:4,index:71,numCoordinates:76};async function v8(e){var t,n;return Ae.initial&&(er=null),er?e.debug&&re("cached model:",er.modelUrl):(er=await et(nt(e.modelBasePath,((t=e.face.iris)==null?void 0:t.modelPath)||"")),!er||!er.modelUrl?re("load model failed:",(n=e.face.iris)==null?void 0:n.modelPath):e.debug&&re("load model:",er.modelUrl)),Zo=er.inputs[0].shape?er.inputs[0].shape[2]:0,Zo===-1&&(Zo=64),er}function i0(e,t,n,s){for(let r=0;r<Bx.length;r++){let{key:a,indices:o}=Bx[r],i=Br[`${n}${a}`];if(!s||s.includes(a))for(let l=0;l<o.length;l++){let c=o[l];e[i[l]]=[t[c][0],t[c][1],(t[c][2]+e[i[l]][2])/2]}}}var Lge=e=>{let t=e[kc.leftBounds[0]][2],n=e[kc.rightBounds[0]][2];return t-n},w8=(e,t,n,s,r=!1,a)=>{let o=$p(Rp(r0([e[n],e[s]]),zge)),i=Ep(o),l=$e.cropAndResize(t,[[o.startPoint[1]/a,o.startPoint[0]/a,o.endPoint[1]/a,o.endPoint[0]/a]],[0],[Zo,Zo]);if(r&&Ae.kernels.includes("flipleftright")){let c=$e.flipLeftRight(l);se(l),l=c}return{box:o,boxSize:i,crop:l}},k8=(e,t,n,s=!1)=>{let r=[];for(let a=0;a<Sc.numCoordinates;a++){let o=e[a*3],i=e[a*3+1],l=e[a*3+2];r.push([(s?1-o/Zo:o/Zo)*n[0]+t.startPoint[0],i/Zo*n[1]+t.startPoint[1],l])}return{rawCoords:r,iris:r.slice(Sc.index)}},S8=(e,t,n)=>{let s=e[Br[`${n}EyeUpper0`][Sc.upperCenter]][2],r=e[Br[`${n}EyeLower0`][Sc.lowerCenter]][2],a=(s+r)/2;return t.map((o,i)=>{let l=a;return i===2?l=s:i===4&&(l=r),[o[0],o[1],l]})};async function I8(e,t,n,s){if(!er)return n.debug&&re("face mesh iris detection requested, but model is not loaded"),e;let{box:r,boxSize:a,crop:o}=w8(e,t,kc.leftBounds[0],kc.leftBounds[1],!0,s),{box:i,boxSize:l,crop:c}=w8(e,t,kc.rightBounds[0],kc.rightBounds[1],!0,s),u=vt([o,c]);se(o),se(c);let d=er.execute(u);se(u);let p=await d.data();se(d);let h=p.slice(0,Sc.numCoordinates*3),{rawCoords:f,iris:m}=k8(h,r,a,!0),g=p.slice(Sc.numCoordinates*3),{rawCoords:A,iris:x}=k8(g,i,l),y=Lge(e);Math.abs(y)<30?(i0(e,f,"left",null),i0(e,A,"right",null)):y<1?i0(e,f,"left",["EyeUpper0","EyeLower0"]):i0(e,A,"right",["EyeUpper0","EyeLower0"]);let b=S8(e,m,"left"),w=S8(e,x,"right");return e.concat(b).concat(w)}var Ic=[],tr=null,ia=0,hb=Number.MAX_SAFE_INTEGER,C8=0,T8=1.6;async function N8(e,t){var i,l,c,u,d,p,h,f;let n=(((i=t.face.detector)==null?void 0:i.skipTime)||0)>he()-C8,s=hb<(((l=t.face.detector)==null?void 0:l.skipFrames)||0);if(!t.skipAllowed||!n||!s||Ic.length===0){let m=await u8(e,t);C8=he(),Ic=[];for(let g of m.boxes){let A={startPoint:g.box.startPoint,endPoint:g.box.endPoint,landmarks:g.landmarks,confidence:g.confidence};Ic.push($p(Rp(Q6(A,m.scaleFactor),Math.sqrt(T8))))}hb=0}else hb++;let r=[],a=[],o=0;for(let m=0;m<Ic.length;m++){let g=Ic[m],A=0,x,y={id:o++,mesh:[],meshRaw:[],box:[0,0,0,0],boxRaw:[0,0,0,0],score:0,boxScore:0,faceScore:0,annotations:{}};if([A,x,y.tensor]=Hx(!1,g,e,ia),(u=t==null?void 0:t.filter)==null?void 0:u.equalization){let b=await e0(y.tensor);se(y.tensor),y.tensor=b}if(y.boxScore=Math.round(100*g.confidence)/100,(d=t.face.mesh)==null?void 0:d.enabled)if(!tr)t.debug&&re("face mesh detection requested, but model is not loaded");else{let[b,w,k]=tr.execute(y.tensor),C=await w.data();y.faceScore=Math.round(100*C[0])/100;let N=G(k,[-1,3]),R=await N.array();if(se([k,N,w,b]),y.faceScore<(((p=t.face.detector)==null?void 0:p.minConfidence)||1))g.confidence=y.faceScore;else{((h=t.face.iris)==null?void 0:h.enabled)&&(R=await I8(R,y.tensor,t,ia)),y.mesh=r8(R,g,A,x,ia),y.meshRaw=y.mesh.map(F=>[F[0]/(e.shape[2]||0),F[1]/(e.shape[1]||0),(F[2]||0)/ia]);for(let F of Object.keys(Br))y.annotations[F]=Br[F].map(_=>y.mesh[_]);g=$p({...Rp(r0(y.mesh),T8),confidence:g.confidence}),y.box=Wx(g,e),y.boxRaw=Vx(g,e),y.score=y.faceScore,a.push(g),se(y.tensor),[A,x,y.tensor]=Hx((f=t.face.detector)==null?void 0:f.rotation,g,e,ia)}}else{y.box=Wx(g,e),y.boxRaw=Vx(g,e),y.score=y.boxScore,y.mesh=g.landmarks.map(b=>[(g.startPoint[0]+g.endPoint[0])/2+(g.endPoint[0]+g.startPoint[0])*b[0]/jx(),(g.startPoint[1]+g.endPoint[1])/2+(g.endPoint[1]+g.startPoint[1])*b[1]/jx()]),y.meshRaw=y.mesh.map(b=>[b[0]/(e.shape[2]||0),b[1]/(e.shape[1]||0),(b[2]||0)/ia]);for(let b of Object.keys(Tp))y.annotations[b]=[y.mesh[Tp[b]]]}r.push(y)}return Ic=[...a],r}async function E8(e){var t,n;return Ae.initial&&(tr=null),tr?e.debug&&re("cached model:",tr.modelUrl):(tr=await et(nt(e.modelBasePath,((t=e.face.mesh)==null?void 0:t.modelPath)||"")),!tr||!tr.modelUrl?re("load model failed:",(n=e.face.mesh)==null?void 0:n.modelPath):e.debug&&re("load model:",tr.modelUrl)),ia=tr.inputs[0].shape?tr.inputs[0].shape[2]:0,ia===-1&&(ia=64),tr}var R8=_l,$8=Np;var ys,l0=[],_8=0,D8=0,fb=Number.MAX_SAFE_INTEGER;async function P8(e){var n,s;let t=nt(e.modelBasePath,((n=e.face.description)==null?void 0:n.modelPath)||"");return Ae.initial&&(ys=null),ys?e.debug&&re("cached model:",t):(ys=await et(t),ys?e.debug&&re("load model:",t):re("load model failed:",((s=e.face.description)==null?void 0:s.modelPath)||"")),ys}function mb(e){let t=e.image||e.tensor||e;if(!(ys==null?void 0:ys.inputs[0].shape))return t;let n=$e.resizeBilinear(t,[ys.inputs[0].shape[2],ys.inputs[0].shape[1]],!1),s=L(n,255);return se(n),s}async function gb(e,t,n,s){var o,i,l,c;if(!ys)return null;let r=fb<(((o=t.face.description)==null?void 0:o.skipFrames)||0),a=(((i=t.face.description)==null?void 0:i.skipTime)||0)>he()-_8;return t.skipAllowed&&r&&a&&D8===s&&((l=l0[n])==null?void 0:l.age)&&((c=l0[n])==null?void 0:c.age)>0?(fb++,l0[n]):(fb=0,new Promise(async u=>{var p,h;let d={age:0,gender:"unknown",genderScore:0,descriptor:[]};if((p=t.face.description)==null?void 0:p.enabled){let f=mb(e),m=ys==null?void 0:ys.execute(f);_8=he(),se(f);let A=await(await m.find(R=>R.shape[1]===1)).data(),x=Math.trunc(200*Math.abs(A[0]-.5))/100;x>(((h=t.face.description)==null?void 0:h.minConfidence)||0)&&(d.gender=A[0]<=.5?"female":"male",d.genderScore=Math.min(.99,x));let y=Hs(m.find(R=>R.shape[1]===100),1),b=(await y.data())[0];se(y);let k=await m.find(R=>R.shape[1]===100).data();d.age=Math.round(k[b-1]>k[b+1]?10*b-100*k[b-1]:10*b+100*k[b+1])/10;let C=m.find(R=>R.shape[1]===1024),N=C?await C.data():[];d.descriptor=Array.from(N),m.forEach(R=>se(R))}l0[n]=d,D8=s,u(d)}))}function u0(e){return[Math.abs(e.endPoint[0]-e.startPoint[0]),Math.abs(e.endPoint[1]-e.startPoint[1])]}function _p(e){return[e.startPoint[0]+(e.endPoint[0]-e.startPoint[0])/2,e.startPoint[1]+(e.endPoint[1]-e.startPoint[1])/2]}function F8(e,t,n){let s=t.shape[1],r=t.shape[2],a=[[e.startPoint[1]/s,e.startPoint[0]/r,e.endPoint[1]/s,e.endPoint[0]/r]];return $e.cropAndResize(t,a,[0],n)}function O8(e,t){let n=[e.startPoint[0]*t[0],e.startPoint[1]*t[1]],s=[e.endPoint[0]*t[0],e.endPoint[1]*t[1]],r=e.palmLandmarks.map(a=>[a[0]*t[0],a[1]*t[1]]);return{startPoint:n,endPoint:s,palmLandmarks:r,confidence:e.confidence}}function c0(e,t=1.5){let n=_p(e),s=u0(e),r=[t*s[0]/2,t*s[1]/2],a=[n[0]-r[0],n[1]-r[1]],o=[n[0]+r[0],n[1]+r[1]];return{startPoint:a,endPoint:o,palmLandmarks:e.palmLandmarks}}function d0(e){let t=_p(e),n=u0(e),r=Math.max(...n)/2,a=[t[0]-r,t[1]-r],o=[t[0]+r,t[1]+r];return{startPoint:a,endPoint:o,palmLandmarks:e.palmLandmarks}}function Bge(e){return e-2*Math.PI*Math.floor((e+Math.PI)/(2*Math.PI))}function M8(e,t){let n=Math.PI/2-Math.atan2(-(t[1]-e[1]),t[0]-e[0]);return Bge(n)}var z8=(e,t)=>[[1,0,e],[0,1,t],[0,0,1]];function Yo(e,t){let n=0;for(let s=0;s<e.length;s++)n+=e[s]*t[s];return n}function Wge(e,t){let n=[];for(let s=0;s<e.length;s++)n.push(e[s][t]);return n}function L8(e,t){let n=[],s=e.length;for(let r=0;r<s;r++){n.push([]);for(let a=0;a<s;a++)n[r].push(Yo(e[r],Wge(t,a)))}return n}function Ab(e,t){let n=Math.cos(e),s=Math.sin(e),r=[[n,-s,0],[s,n,0],[0,0,1]],a=z8(t[0],t[1]),o=L8(a,r),i=z8(-t[0],-t[1]);return L8(o,i)}function B8(e){let t=[[e[0][0],e[1][0]],[e[0][1],e[1][1]]],n=[e[0][2],e[1][2]],s=[-Yo(t[0],n),-Yo(t[1],n)];return[t[0].concat(s[0]),t[1].concat(s[1]),[0,0,1]]}function yb(e,t){return[Yo(e,t[0]),Yo(e,t[1])]}var W8=[{x:.015625,y:.015625},{x:.015625,y:.015625},{x:.046875,y:.015625},{x:.046875,y:.015625},{x:.078125,y:.015625},{x:.078125,y:.015625},{x:.109375,y:.015625},{x:.109375,y:.015625},{x:.140625,y:.015625},{x:.140625,y:.015625},{x:.171875,y:.015625},{x:.171875,y:.015625},{x:.203125,y:.015625},{x:.203125,y:.015625},{x:.234375,y:.015625},{x:.234375,y:.015625},{x:.265625,y:.015625},{x:.265625,y:.015625},{x:.296875,y:.015625},{x:.296875,y:.015625},{x:.328125,y:.015625},{x:.328125,y:.015625},{x:.359375,y:.015625},{x:.359375,y:.015625},{x:.390625,y:.015625},{x:.390625,y:.015625},{x:.421875,y:.015625},{x:.421875,y:.015625},{x:.453125,y:.015625},{x:.453125,y:.015625},{x:.484375,y:.015625},{x:.484375,y:.015625},{x:.515625,y:.015625},{x:.515625,y:.015625},{x:.546875,y:.015625},{x:.546875,y:.015625},{x:.578125,y:.015625},{x:.578125,y:.015625},{x:.609375,y:.015625},{x:.609375,y:.015625},{x:.640625,y:.015625},{x:.640625,y:.015625},{x:.671875,y:.015625},{x:.671875,y:.015625},{x:.703125,y:.015625},{x:.703125,y:.015625},{x:.734375,y:.015625},{x:.734375,y:.015625},{x:.765625,y:.015625},{x:.765625,y:.015625},{x:.796875,y:.015625},{x:.796875,y:.015625},{x:.828125,y:.015625},{x:.828125,y:.015625},{x:.859375,y:.015625},{x:.859375,y:.015625},{x:.890625,y:.015625},{x:.890625,y:.015625},{x:.921875,y:.015625},{x:.921875,y:.015625},{x:.953125,y:.015625},{x:.953125,y:.015625},{x:.984375,y:.015625},{x:.984375,y:.015625},{x:.015625,y:.046875},{x:.015625,y:.046875},{x:.046875,y:.046875},{x:.046875,y:.046875},{x:.078125,y:.046875},{x:.078125,y:.046875},{x:.109375,y:.046875},{x:.109375,y:.046875},{x:.140625,y:.046875},{x:.140625,y:.046875},{x:.171875,y:.046875},{x:.171875,y:.046875},{x:.203125,y:.046875},{x:.203125,y:.046875},{x:.234375,y:.046875},{x:.234375,y:.046875},{x:.265625,y:.046875},{x:.265625,y:.046875},{x:.296875,y:.046875},{x:.296875,y:.046875},{x:.328125,y:.046875},{x:.328125,y:.046875},{x:.359375,y:.046875},{x:.359375,y:.046875},{x:.390625,y:.046875},{x:.390625,y:.046875},{x:.421875,y:.046875},{x:.421875,y:.046875},{x:.453125,y:.046875},{x:.453125,y:.046875},{x:.484375,y:.046875},{x:.484375,y:.046875},{x:.515625,y:.046875},{x:.515625,y:.046875},{x:.546875,y:.046875},{x:.546875,y:.046875},{x:.578125,y:.046875},{x:.578125,y:.046875},{x:.609375,y:.046875},{x:.609375,y:.046875},{x:.640625,y:.046875},{x:.640625,y:.046875},{x:.671875,y:.046875},{x:.671875,y:.046875},{x:.703125,y:.046875},{x:.703125,y:.046875},{x:.734375,y:.046875},{x:.734375,y:.046875},{x:.765625,y:.046875},{x:.765625,y:.046875},{x:.796875,y:.046875},{x:.796875,y:.046875},{x:.828125,y:.046875},{x:.828125,y:.046875},{x:.859375,y:.046875},{x:.859375,y:.046875},{x:.890625,y:.046875},{x:.890625,y:.046875},{x:.921875,y:.046875},{x:.921875,y:.046875},{x:.953125,y:.046875},{x:.953125,y:.046875},{x:.984375,y:.046875},{x:.984375,y:.046875},{x:.015625,y:.078125},{x:.015625,y:.078125},{x:.046875,y:.078125},{x:.046875,y:.078125},{x:.078125,y:.078125},{x:.078125,y:.078125},{x:.109375,y:.078125},{x:.109375,y:.078125},{x:.140625,y:.078125},{x:.140625,y:.078125},{x:.171875,y:.078125},{x:.171875,y:.078125},{x:.203125,y:.078125},{x:.203125,y:.078125},{x:.234375,y:.078125},{x:.234375,y:.078125},{x:.265625,y:.078125},{x:.265625,y:.078125},{x:.296875,y:.078125},{x:.296875,y:.078125},{x:.328125,y:.078125},{x:.328125,y:.078125},{x:.359375,y:.078125},{x:.359375,y:.078125},{x:.390625,y:.078125},{x:.390625,y:.078125},{x:.421875,y:.078125},{x:.421875,y:.078125},{x:.453125,y:.078125},{x:.453125,y:.078125},{x:.484375,y:.078125},{x:.484375,y:.078125},{x:.515625,y:.078125},{x:.515625,y:.078125},{x:.546875,y:.078125},{x:.546875,y:.078125},{x:.578125,y:.078125},{x:.578125,y:.078125},{x:.609375,y:.078125},{x:.609375,y:.078125},{x:.640625,y:.078125},{x:.640625,y:.078125},{x:.671875,y:.078125},{x:.671875,y:.078125},{x:.703125,y:.078125},{x:.703125,y:.078125},{x:.734375,y:.078125},{x:.734375,y:.078125},{x:.765625,y:.078125},{x:.765625,y:.078125},{x:.796875,y:.078125},{x:.796875,y:.078125},{x:.828125,y:.078125},{x:.828125,y:.078125},{x:.859375,y:.078125},{x:.859375,y:.078125},{x:.890625,y:.078125},{x:.890625,y:.078125},{x:.921875,y:.078125},{x:.921875,y:.078125},{x:.953125,y:.078125},{x:.953125,y:.078125},{x:.984375,y:.078125},{x:.984375,y:.078125},{x:.015625,y:.109375},{x:.015625,y:.109375},{x:.046875,y:.109375},{x:.046875,y:.109375},{x:.078125,y:.109375},{x:.078125,y:.109375},{x:.109375,y:.109375},{x:.109375,y:.109375},{x:.140625,y:.109375},{x:.140625,y:.109375},{x:.171875,y:.109375},{x:.171875,y:.109375},{x:.203125,y:.109375},{x:.203125,y:.109375},{x:.234375,y:.109375},{x:.234375,y:.109375},{x:.265625,y:.109375},{x:.265625,y:.109375},{x:.296875,y:.109375},{x:.296875,y:.109375},{x:.328125,y:.109375},{x:.328125,y:.109375},{x:.359375,y:.109375},{x:.359375,y:.109375},{x:.390625,y:.109375},{x:.390625,y:.109375},{x:.421875,y:.109375},{x:.421875,y:.109375},{x:.453125,y:.109375},{x:.453125,y:.109375},{x:.484375,y:.109375},{x:.484375,y:.109375},{x:.515625,y:.109375},{x:.515625,y:.109375},{x:.546875,y:.109375},{x:.546875,y:.109375},{x:.578125,y:.109375},{x:.578125,y:.109375},{x:.609375,y:.109375},{x:.609375,y:.109375},{x:.640625,y:.109375},{x:.640625,y:.109375},{x:.671875,y:.109375},{x:.671875,y:.109375},{x:.703125,y:.109375},{x:.703125,y:.109375},{x:.734375,y:.109375},{x:.734375,y:.109375},{x:.765625,y:.109375},{x:.765625,y:.109375},{x:.796875,y:.109375},{x:.796875,y:.109375},{x:.828125,y:.109375},{x:.828125,y:.109375},{x:.859375,y:.109375},{x:.859375,y:.109375},{x:.890625,y:.109375},{x:.890625,y:.109375},{x:.921875,y:.109375},{x:.921875,y:.109375},{x:.953125,y:.109375},{x:.953125,y:.109375},{x:.984375,y:.109375},{x:.984375,y:.109375},{x:.015625,y:.140625},{x:.015625,y:.140625},{x:.046875,y:.140625},{x:.046875,y:.140625},{x:.078125,y:.140625},{x:.078125,y:.140625},{x:.109375,y:.140625},{x:.109375,y:.140625},{x:.140625,y:.140625},{x:.140625,y:.140625},{x:.171875,y:.140625},{x:.171875,y:.140625},{x:.203125,y:.140625},{x:.203125,y:.140625},{x:.234375,y:.140625},{x:.234375,y:.140625},{x:.265625,y:.140625},{x:.265625,y:.140625},{x:.296875,y:.140625},{x:.296875,y:.140625},{x:.328125,y:.140625},{x:.328125,y:.140625},{x:.359375,y:.140625},{x:.359375,y:.140625},{x:.390625,y:.140625},{x:.390625,y:.140625},{x:.421875,y:.140625},{x:.421875,y:.140625},{x:.453125,y:.140625},{x:.453125,y:.140625},{x:.484375,y:.140625},{x:.484375,y:.140625},{x:.515625,y:.140625},{x:.515625,y:.140625},{x:.546875,y:.140625},{x:.546875,y:.140625},{x:.578125,y:.140625},{x:.578125,y:.140625},{x:.609375,y:.140625},{x:.609375,y:.140625},{x:.640625,y:.140625},{x:.640625,y:.140625},{x:.671875,y:.140625},{x:.671875,y:.140625},{x:.703125,y:.140625},{x:.703125,y:.140625},{x:.734375,y:.140625},{x:.734375,y:.140625},{x:.765625,y:.140625},{x:.765625,y:.140625},{x:.796875,y:.140625},{x:.796875,y:.140625},{x:.828125,y:.140625},{x:.828125,y:.140625},{x:.859375,y:.140625},{x:.859375,y:.140625},{x:.890625,y:.140625},{x:.890625,y:.140625},{x:.921875,y:.140625},{x:.921875,y:.140625},{x:.953125,y:.140625},{x:.953125,y:.140625},{x:.984375,y:.140625},{x:.984375,y:.140625},{x:.015625,y:.171875},{x:.015625,y:.171875},{x:.046875,y:.171875},{x:.046875,y:.171875},{x:.078125,y:.171875},{x:.078125,y:.171875},{x:.109375,y:.171875},{x:.109375,y:.171875},{x:.140625,y:.171875},{x:.140625,y:.171875},{x:.171875,y:.171875},{x:.171875,y:.171875},{x:.203125,y:.171875},{x:.203125,y:.171875},{x:.234375,y:.171875},{x:.234375,y:.171875},{x:.265625,y:.171875},{x:.265625,y:.171875},{x:.296875,y:.171875},{x:.296875,y:.171875},{x:.328125,y:.171875},{x:.328125,y:.171875},{x:.359375,y:.171875},{x:.359375,y:.171875},{x:.390625,y:.171875},{x:.390625,y:.171875},{x:.421875,y:.171875},{x:.421875,y:.171875},{x:.453125,y:.171875},{x:.453125,y:.171875},{x:.484375,y:.171875},{x:.484375,y:.171875},{x:.515625,y:.171875},{x:.515625,y:.171875},{x:.546875,y:.171875},{x:.546875,y:.171875},{x:.578125,y:.171875},{x:.578125,y:.171875},{x:.609375,y:.171875},{x:.609375,y:.171875},{x:.640625,y:.171875},{x:.640625,y:.171875},{x:.671875,y:.171875},{x:.671875,y:.171875},{x:.703125,y:.171875},{x:.703125,y:.171875},{x:.734375,y:.171875},{x:.734375,y:.171875},{x:.765625,y:.171875},{x:.765625,y:.171875},{x:.796875,y:.171875},{x:.796875,y:.171875},{x:.828125,y:.171875},{x:.828125,y:.171875},{x:.859375,y:.171875},{x:.859375,y:.171875},{x:.890625,y:.171875},{x:.890625,y:.171875},{x:.921875,y:.171875},{x:.921875,y:.171875},{x:.953125,y:.171875},{x:.953125,y:.171875},{x:.984375,y:.171875},{x:.984375,y:.171875},{x:.015625,y:.203125},{x:.015625,y:.203125},{x:.046875,y:.203125},{x:.046875,y:.203125},{x:.078125,y:.203125},{x:.078125,y:.203125},{x:.109375,y:.203125},{x:.109375,y:.203125},{x:.140625,y:.203125},{x:.140625,y:.203125},{x:.171875,y:.203125},{x:.171875,y:.203125},{x:.203125,y:.203125},{x:.203125,y:.203125},{x:.234375,y:.203125},{x:.234375,y:.203125},{x:.265625,y:.203125},{x:.265625,y:.203125},{x:.296875,y:.203125},{x:.296875,y:.203125},{x:.328125,y:.203125},{x:.328125,y:.203125},{x:.359375,y:.203125},{x:.359375,y:.203125},{x:.390625,y:.203125},{x:.390625,y:.203125},{x:.421875,y:.203125},{x:.421875,y:.203125},{x:.453125,y:.203125},{x:.453125,y:.203125},{x:.484375,y:.203125},{x:.484375,y:.203125},{x:.515625,y:.203125},{x:.515625,y:.203125},{x:.546875,y:.203125},{x:.546875,y:.203125},{x:.578125,y:.203125},{x:.578125,y:.203125},{x:.609375,y:.203125},{x:.609375,y:.203125},{x:.640625,y:.203125},{x:.640625,y:.203125},{x:.671875,y:.203125},{x:.671875,y:.203125},{x:.703125,y:.203125},{x:.703125,y:.203125},{x:.734375,y:.203125},{x:.734375,y:.203125},{x:.765625,y:.203125},{x:.765625,y:.203125},{x:.796875,y:.203125},{x:.796875,y:.203125},{x:.828125,y:.203125},{x:.828125,y:.203125},{x:.859375,y:.203125},{x:.859375,y:.203125},{x:.890625,y:.203125},{x:.890625,y:.203125},{x:.921875,y:.203125},{x:.921875,y:.203125},{x:.953125,y:.203125},{x:.953125,y:.203125},{x:.984375,y:.203125},{x:.984375,y:.203125},{x:.015625,y:.234375},{x:.015625,y:.234375},{x:.046875,y:.234375},{x:.046875,y:.234375},{x:.078125,y:.234375},{x:.078125,y:.234375},{x:.109375,y:.234375},{x:.109375,y:.234375},{x:.140625,y:.234375},{x:.140625,y:.234375},{x:.171875,y:.234375},{x:.171875,y:.234375},{x:.203125,y:.234375},{x:.203125,y:.234375},{x:.234375,y:.234375},{x:.234375,y:.234375},{x:.265625,y:.234375},{x:.265625,y:.234375},{x:.296875,y:.234375},{x:.296875,y:.234375},{x:.328125,y:.234375},{x:.328125,y:.234375},{x:.359375,y:.234375},{x:.359375,y:.234375},{x:.390625,y:.234375},{x:.390625,y:.234375},{x:.421875,y:.234375},{x:.421875,y:.234375},{x:.453125,y:.234375},{x:.453125,y:.234375},{x:.484375,y:.234375},{x:.484375,y:.234375},{x:.515625,y:.234375},{x:.515625,y:.234375},{x:.546875,y:.234375},{x:.546875,y:.234375},{x:.578125,y:.234375},{x:.578125,y:.234375},{x:.609375,y:.234375},{x:.609375,y:.234375},{x:.640625,y:.234375},{x:.640625,y:.234375},{x:.671875,y:.234375},{x:.671875,y:.234375},{x:.703125,y:.234375},{x:.703125,y:.234375},{x:.734375,y:.234375},{x:.734375,y:.234375},{x:.765625,y:.234375},{x:.765625,y:.234375},{x:.796875,y:.234375},{x:.796875,y:.234375},{x:.828125,y:.234375},{x:.828125,y:.234375},{x:.859375,y:.234375},{x:.859375,y:.234375},{x:.890625,y:.234375},{x:.890625,y:.234375},{x:.921875,y:.234375},{x:.921875,y:.234375},{x:.953125,y:.234375},{x:.953125,y:.234375},{x:.984375,y:.234375},{x:.984375,y:.234375},{x:.015625,y:.265625},{x:.015625,y:.265625},{x:.046875,y:.265625},{x:.046875,y:.265625},{x:.078125,y:.265625},{x:.078125,y:.265625},{x:.109375,y:.265625},{x:.109375,y:.265625},{x:.140625,y:.265625},{x:.140625,y:.265625},{x:.171875,y:.265625},{x:.171875,y:.265625},{x:.203125,y:.265625},{x:.203125,y:.265625},{x:.234375,y:.265625},{x:.234375,y:.265625},{x:.265625,y:.265625},{x:.265625,y:.265625},{x:.296875,y:.265625},{x:.296875,y:.265625},{x:.328125,y:.265625},{x:.328125,y:.265625},{x:.359375,y:.265625},{x:.359375,y:.265625},{x:.390625,y:.265625},{x:.390625,y:.265625},{x:.421875,y:.265625},{x:.421875,y:.265625},{x:.453125,y:.265625},{x:.453125,y:.265625},{x:.484375,y:.265625},{x:.484375,y:.265625},{x:.515625,y:.265625},{x:.515625,y:.265625},{x:.546875,y:.265625},{x:.546875,y:.265625},{x:.578125,y:.265625},{x:.578125,y:.265625},{x:.609375,y:.265625},{x:.609375,y:.265625},{x:.640625,y:.265625},{x:.640625,y:.265625},{x:.671875,y:.265625},{x:.671875,y:.265625},{x:.703125,y:.265625},{x:.703125,y:.265625},{x:.734375,y:.265625},{x:.734375,y:.265625},{x:.765625,y:.265625},{x:.765625,y:.265625},{x:.796875,y:.265625},{x:.796875,y:.265625},{x:.828125,y:.265625},{x:.828125,y:.265625},{x:.859375,y:.265625},{x:.859375,y:.265625},{x:.890625,y:.265625},{x:.890625,y:.265625},{x:.921875,y:.265625},{x:.921875,y:.265625},{x:.953125,y:.265625},{x:.953125,y:.265625},{x:.984375,y:.265625},{x:.984375,y:.265625},{x:.015625,y:.296875},{x:.015625,y:.296875},{x:.046875,y:.296875},{x:.046875,y:.296875},{x:.078125,y:.296875},{x:.078125,y:.296875},{x:.109375,y:.296875},{x:.109375,y:.296875},{x:.140625,y:.296875},{x:.140625,y:.296875},{x:.171875,y:.296875},{x:.171875,y:.296875},{x:.203125,y:.296875},{x:.203125,y:.296875},{x:.234375,y:.296875},{x:.234375,y:.296875},{x:.265625,y:.296875},{x:.265625,y:.296875},{x:.296875,y:.296875},{x:.296875,y:.296875},{x:.328125,y:.296875},{x:.328125,y:.296875},{x:.359375,y:.296875},{x:.359375,y:.296875},{x:.390625,y:.296875},{x:.390625,y:.296875},{x:.421875,y:.296875},{x:.421875,y:.296875},{x:.453125,y:.296875},{x:.453125,y:.296875},{x:.484375,y:.296875},{x:.484375,y:.296875},{x:.515625,y:.296875},{x:.515625,y:.296875},{x:.546875,y:.296875},{x:.546875,y:.296875},{x:.578125,y:.296875},{x:.578125,y:.296875},{x:.609375,y:.296875},{x:.609375,y:.296875},{x:.640625,y:.296875},{x:.640625,y:.296875},{x:.671875,y:.296875},{x:.671875,y:.296875},{x:.703125,y:.296875},{x:.703125,y:.296875},{x:.734375,y:.296875},{x:.734375,y:.296875},{x:.765625,y:.296875},{x:.765625,y:.296875},{x:.796875,y:.296875},{x:.796875,y:.296875},{x:.828125,y:.296875},{x:.828125,y:.296875},{x:.859375,y:.296875},{x:.859375,y:.296875},{x:.890625,y:.296875},{x:.890625,y:.296875},{x:.921875,y:.296875},{x:.921875,y:.296875},{x:.953125,y:.296875},{x:.953125,y:.296875},{x:.984375,y:.296875},{x:.984375,y:.296875},{x:.015625,y:.328125},{x:.015625,y:.328125},{x:.046875,y:.328125},{x:.046875,y:.328125},{x:.078125,y:.328125},{x:.078125,y:.328125},{x:.109375,y:.328125},{x:.109375,y:.328125},{x:.140625,y:.328125},{x:.140625,y:.328125},{x:.171875,y:.328125},{x:.171875,y:.328125},{x:.203125,y:.328125},{x:.203125,y:.328125},{x:.234375,y:.328125},{x:.234375,y:.328125},{x:.265625,y:.328125},{x:.265625,y:.328125},{x:.296875,y:.328125},{x:.296875,y:.328125},{x:.328125,y:.328125},{x:.328125,y:.328125},{x:.359375,y:.328125},{x:.359375,y:.328125},{x:.390625,y:.328125},{x:.390625,y:.328125},{x:.421875,y:.328125},{x:.421875,y:.328125},{x:.453125,y:.328125},{x:.453125,y:.328125},{x:.484375,y:.328125},{x:.484375,y:.328125},{x:.515625,y:.328125},{x:.515625,y:.328125},{x:.546875,y:.328125},{x:.546875,y:.328125},{x:.578125,y:.328125},{x:.578125,y:.328125},{x:.609375,y:.328125},{x:.609375,y:.328125},{x:.640625,y:.328125},{x:.640625,y:.328125},{x:.671875,y:.328125},{x:.671875,y:.328125},{x:.703125,y:.328125},{x:.703125,y:.328125},{x:.734375,y:.328125},{x:.734375,y:.328125},{x:.765625,y:.328125},{x:.765625,y:.328125},{x:.796875,y:.328125},{x:.796875,y:.328125},{x:.828125,y:.328125},{x:.828125,y:.328125},{x:.859375,y:.328125},{x:.859375,y:.328125},{x:.890625,y:.328125},{x:.890625,y:.328125},{x:.921875,y:.328125},{x:.921875,y:.328125},{x:.953125,y:.328125},{x:.953125,y:.328125},{x:.984375,y:.328125},{x:.984375,y:.328125},{x:.015625,y:.359375},{x:.015625,y:.359375},{x:.046875,y:.359375},{x:.046875,y:.359375},{x:.078125,y:.359375},{x:.078125,y:.359375},{x:.109375,y:.359375},{x:.109375,y:.359375},{x:.140625,y:.359375},{x:.140625,y:.359375},{x:.171875,y:.359375},{x:.171875,y:.359375},{x:.203125,y:.359375},{x:.203125,y:.359375},{x:.234375,y:.359375},{x:.234375,y:.359375},{x:.265625,y:.359375},{x:.265625,y:.359375},{x:.296875,y:.359375},{x:.296875,y:.359375},{x:.328125,y:.359375},{x:.328125,y:.359375},{x:.359375,y:.359375},{x:.359375,y:.359375},{x:.390625,y:.359375},{x:.390625,y:.359375},{x:.421875,y:.359375},{x:.421875,y:.359375},{x:.453125,y:.359375},{x:.453125,y:.359375},{x:.484375,y:.359375},{x:.484375,y:.359375},{x:.515625,y:.359375},{x:.515625,y:.359375},{x:.546875,y:.359375},{x:.546875,y:.359375},{x:.578125,y:.359375},{x:.578125,y:.359375},{x:.609375,y:.359375},{x:.609375,y:.359375},{x:.640625,y:.359375},{x:.640625,y:.359375},{x:.671875,y:.359375},{x:.671875,y:.359375},{x:.703125,y:.359375},{x:.703125,y:.359375},{x:.734375,y:.359375},{x:.734375,y:.359375},{x:.765625,y:.359375},{x:.765625,y:.359375},{x:.796875,y:.359375},{x:.796875,y:.359375},{x:.828125,y:.359375},{x:.828125,y:.359375},{x:.859375,y:.359375},{x:.859375,y:.359375},{x:.890625,y:.359375},{x:.890625,y:.359375},{x:.921875,y:.359375},{x:.921875,y:.359375},{x:.953125,y:.359375},{x:.953125,y:.359375},{x:.984375,y:.359375},{x:.984375,y:.359375},{x:.015625,y:.390625},{x:.015625,y:.390625},{x:.046875,y:.390625},{x:.046875,y:.390625},{x:.078125,y:.390625},{x:.078125,y:.390625},{x:.109375,y:.390625},{x:.109375,y:.390625},{x:.140625,y:.390625},{x:.140625,y:.390625},{x:.171875,y:.390625},{x:.171875,y:.390625},{x:.203125,y:.390625},{x:.203125,y:.390625},{x:.234375,y:.390625},{x:.234375,y:.390625},{x:.265625,y:.390625},{x:.265625,y:.390625},{x:.296875,y:.390625},{x:.296875,y:.390625},{x:.328125,y:.390625},{x:.328125,y:.390625},{x:.359375,y:.390625},{x:.359375,y:.390625},{x:.390625,y:.390625},{x:.390625,y:.390625},{x:.421875,y:.390625},{x:.421875,y:.390625},{x:.453125,y:.390625},{x:.453125,y:.390625},{x:.484375,y:.390625},{x:.484375,y:.390625},{x:.515625,y:.390625},{x:.515625,y:.390625},{x:.546875,y:.390625},{x:.546875,y:.390625},{x:.578125,y:.390625},{x:.578125,y:.390625},{x:.609375,y:.390625},{x:.609375,y:.390625},{x:.640625,y:.390625},{x:.640625,y:.390625},{x:.671875,y:.390625},{x:.671875,y:.390625},{x:.703125,y:.390625},{x:.703125,y:.390625},{x:.734375,y:.390625},{x:.734375,y:.390625},{x:.765625,y:.390625},{x:.765625,y:.390625},{x:.796875,y:.390625},{x:.796875,y:.390625},{x:.828125,y:.390625},{x:.828125,y:.390625},{x:.859375,y:.390625},{x:.859375,y:.390625},{x:.890625,y:.390625},{x:.890625,y:.390625},{x:.921875,y:.390625},{x:.921875,y:.390625},{x:.953125,y:.390625},{x:.953125,y:.390625},{x:.984375,y:.390625},{x:.984375,y:.390625},{x:.015625,y:.421875},{x:.015625,y:.421875},{x:.046875,y:.421875},{x:.046875,y:.421875},{x:.078125,y:.421875},{x:.078125,y:.421875},{x:.109375,y:.421875},{x:.109375,y:.421875},{x:.140625,y:.421875},{x:.140625,y:.421875},{x:.171875,y:.421875},{x:.171875,y:.421875},{x:.203125,y:.421875},{x:.203125,y:.421875},{x:.234375,y:.421875},{x:.234375,y:.421875},{x:.265625,y:.421875},{x:.265625,y:.421875},{x:.296875,y:.421875},{x:.296875,y:.421875},{x:.328125,y:.421875},{x:.328125,y:.421875},{x:.359375,y:.421875},{x:.359375,y:.421875},{x:.390625,y:.421875},{x:.390625,y:.421875},{x:.421875,y:.421875},{x:.421875,y:.421875},{x:.453125,y:.421875},{x:.453125,y:.421875},{x:.484375,y:.421875},{x:.484375,y:.421875},{x:.515625,y:.421875},{x:.515625,y:.421875},{x:.546875,y:.421875},{x:.546875,y:.421875},{x:.578125,y:.421875},{x:.578125,y:.421875},{x:.609375,y:.421875},{x:.609375,y:.421875},{x:.640625,y:.421875},{x:.640625,y:.421875},{x:.671875,y:.421875},{x:.671875,y:.421875},{x:.703125,y:.421875},{x:.703125,y:.421875},{x:.734375,y:.421875},{x:.734375,y:.421875},{x:.765625,y:.421875},{x:.765625,y:.421875},{x:.796875,y:.421875},{x:.796875,y:.421875},{x:.828125,y:.421875},{x:.828125,y:.421875},{x:.859375,y:.421875},{x:.859375,y:.421875},{x:.890625,y:.421875},{x:.890625,y:.421875},{x:.921875,y:.421875},{x:.921875,y:.421875},{x:.953125,y:.421875},{x:.953125,y:.421875},{x:.984375,y:.421875},{x:.984375,y:.421875},{x:.015625,y:.453125},{x:.015625,y:.453125},{x:.046875,y:.453125},{x:.046875,y:.453125},{x:.078125,y:.453125},{x:.078125,y:.453125},{x:.109375,y:.453125},{x:.109375,y:.453125},{x:.140625,y:.453125},{x:.140625,y:.453125},{x:.171875,y:.453125},{x:.171875,y:.453125},{x:.203125,y:.453125},{x:.203125,y:.453125},{x:.234375,y:.453125},{x:.234375,y:.453125},{x:.265625,y:.453125},{x:.265625,y:.453125},{x:.296875,y:.453125},{x:.296875,y:.453125},{x:.328125,y:.453125},{x:.328125,y:.453125},{x:.359375,y:.453125},{x:.359375,y:.453125},{x:.390625,y:.453125},{x:.390625,y:.453125},{x:.421875,y:.453125},{x:.421875,y:.453125},{x:.453125,y:.453125},{x:.453125,y:.453125},{x:.484375,y:.453125},{x:.484375,y:.453125},{x:.515625,y:.453125},{x:.515625,y:.453125},{x:.546875,y:.453125},{x:.546875,y:.453125},{x:.578125,y:.453125},{x:.578125,y:.453125},{x:.609375,y:.453125},{x:.609375,y:.453125},{x:.640625,y:.453125},{x:.640625,y:.453125},{x:.671875,y:.453125},{x:.671875,y:.453125},{x:.703125,y:.453125},{x:.703125,y:.453125},{x:.734375,y:.453125},{x:.734375,y:.453125},{x:.765625,y:.453125},{x:.765625,y:.453125},{x:.796875,y:.453125},{x:.796875,y:.453125},{x:.828125,y:.453125},{x:.828125,y:.453125},{x:.859375,y:.453125},{x:.859375,y:.453125},{x:.890625,y:.453125},{x:.890625,y:.453125},{x:.921875,y:.453125},{x:.921875,y:.453125},{x:.953125,y:.453125},{x:.953125,y:.453125},{x:.984375,y:.453125},{x:.984375,y:.453125},{x:.015625,y:.484375},{x:.015625,y:.484375},{x:.046875,y:.484375},{x:.046875,y:.484375},{x:.078125,y:.484375},{x:.078125,y:.484375},{x:.109375,y:.484375},{x:.109375,y:.484375},{x:.140625,y:.484375},{x:.140625,y:.484375},{x:.171875,y:.484375},{x:.171875,y:.484375},{x:.203125,y:.484375},{x:.203125,y:.484375},{x:.234375,y:.484375},{x:.234375,y:.484375},{x:.265625,y:.484375},{x:.265625,y:.484375},{x:.296875,y:.484375},{x:.296875,y:.484375},{x:.328125,y:.484375},{x:.328125,y:.484375},{x:.359375,y:.484375},{x:.359375,y:.484375},{x:.390625,y:.484375},{x:.390625,y:.484375},{x:.421875,y:.484375},{x:.421875,y:.484375},{x:.453125,y:.484375},{x:.453125,y:.484375},{x:.484375,y:.484375},{x:.484375,y:.484375},{x:.515625,y:.484375},{x:.515625,y:.484375},{x:.546875,y:.484375},{x:.546875,y:.484375},{x:.578125,y:.484375},{x:.578125,y:.484375},{x:.609375,y:.484375},{x:.609375,y:.484375},{x:.640625,y:.484375},{x:.640625,y:.484375},{x:.671875,y:.484375},{x:.671875,y:.484375},{x:.703125,y:.484375},{x:.703125,y:.484375},{x:.734375,y:.484375},{x:.734375,y:.484375},{x:.765625,y:.484375},{x:.765625,y:.484375},{x:.796875,y:.484375},{x:.796875,y:.484375},{x:.828125,y:.484375},{x:.828125,y:.484375},{x:.859375,y:.484375},{x:.859375,y:.484375},{x:.890625,y:.484375},{x:.890625,y:.484375},{x:.921875,y:.484375},{x:.921875,y:.484375},{x:.953125,y:.484375},{x:.953125,y:.484375},{x:.984375,y:.484375},{x:.984375,y:.484375},{x:.015625,y:.515625},{x:.015625,y:.515625},{x:.046875,y:.515625},{x:.046875,y:.515625},{x:.078125,y:.515625},{x:.078125,y:.515625},{x:.109375,y:.515625},{x:.109375,y:.515625},{x:.140625,y:.515625},{x:.140625,y:.515625},{x:.171875,y:.515625},{x:.171875,y:.515625},{x:.203125,y:.515625},{x:.203125,y:.515625},{x:.234375,y:.515625},{x:.234375,y:.515625},{x:.265625,y:.515625},{x:.265625,y:.515625},{x:.296875,y:.515625},{x:.296875,y:.515625},{x:.328125,y:.515625},{x:.328125,y:.515625},{x:.359375,y:.515625},{x:.359375,y:.515625},{x:.390625,y:.515625},{x:.390625,y:.515625},{x:.421875,y:.515625},{x:.421875,y:.515625},{x:.453125,y:.515625},{x:.453125,y:.515625},{x:.484375,y:.515625},{x:.484375,y:.515625},{x:.515625,y:.515625},{x:.515625,y:.515625},{x:.546875,y:.515625},{x:.546875,y:.515625},{x:.578125,y:.515625},{x:.578125,y:.515625},{x:.609375,y:.515625},{x:.609375,y:.515625},{x:.640625,y:.515625},{x:.640625,y:.515625},{x:.671875,y:.515625},{x:.671875,y:.515625},{x:.703125,y:.515625},{x:.703125,y:.515625},{x:.734375,y:.515625},{x:.734375,y:.515625},{x:.765625,y:.515625},{x:.765625,y:.515625},{x:.796875,y:.515625},{x:.796875,y:.515625},{x:.828125,y:.515625},{x:.828125,y:.515625},{x:.859375,y:.515625},{x:.859375,y:.515625},{x:.890625,y:.515625},{x:.890625,y:.515625},{x:.921875,y:.515625},{x:.921875,y:.515625},{x:.953125,y:.515625},{x:.953125,y:.515625},{x:.984375,y:.515625},{x:.984375,y:.515625},{x:.015625,y:.546875},{x:.015625,y:.546875},{x:.046875,y:.546875},{x:.046875,y:.546875},{x:.078125,y:.546875},{x:.078125,y:.546875},{x:.109375,y:.546875},{x:.109375,y:.546875},{x:.140625,y:.546875},{x:.140625,y:.546875},{x:.171875,y:.546875},{x:.171875,y:.546875},{x:.203125,y:.546875},{x:.203125,y:.546875},{x:.234375,y:.546875},{x:.234375,y:.546875},{x:.265625,y:.546875},{x:.265625,y:.546875},{x:.296875,y:.546875},{x:.296875,y:.546875},{x:.328125,y:.546875},{x:.328125,y:.546875},{x:.359375,y:.546875},{x:.359375,y:.546875},{x:.390625,y:.546875},{x:.390625,y:.546875},{x:.421875,y:.546875},{x:.421875,y:.546875},{x:.453125,y:.546875},{x:.453125,y:.546875},{x:.484375,y:.546875},{x:.484375,y:.546875},{x:.515625,y:.546875},{x:.515625,y:.546875},{x:.546875,y:.546875},{x:.546875,y:.546875},{x:.578125,y:.546875},{x:.578125,y:.546875},{x:.609375,y:.546875},{x:.609375,y:.546875},{x:.640625,y:.546875},{x:.640625,y:.546875},{x:.671875,y:.546875},{x:.671875,y:.546875},{x:.703125,y:.546875},{x:.703125,y:.546875},{x:.734375,y:.546875},{x:.734375,y:.546875},{x:.765625,y:.546875},{x:.765625,y:.546875},{x:.796875,y:.546875},{x:.796875,y:.546875},{x:.828125,y:.546875},{x:.828125,y:.546875},{x:.859375,y:.546875},{x:.859375,y:.546875},{x:.890625,y:.546875},{x:.890625,y:.546875},{x:.921875,y:.546875},{x:.921875,y:.546875},{x:.953125,y:.546875},{x:.953125,y:.546875},{x:.984375,y:.546875},{x:.984375,y:.546875},{x:.015625,y:.578125},{x:.015625,y:.578125},{x:.046875,y:.578125},{x:.046875,y:.578125},{x:.078125,y:.578125},{x:.078125,y:.578125},{x:.109375,y:.578125},{x:.109375,y:.578125},{x:.140625,y:.578125},{x:.140625,y:.578125},{x:.171875,y:.578125},{x:.171875,y:.578125},{x:.203125,y:.578125},{x:.203125,y:.578125},{x:.234375,y:.578125},{x:.234375,y:.578125},{x:.265625,y:.578125},{x:.265625,y:.578125},{x:.296875,y:.578125},{x:.296875,y:.578125},{x:.328125,y:.578125},{x:.328125,y:.578125},{x:.359375,y:.578125},{x:.359375,y:.578125},{x:.390625,y:.578125},{x:.390625,y:.578125},{x:.421875,y:.578125},{x:.421875,y:.578125},{x:.453125,y:.578125},{x:.453125,y:.578125},{x:.484375,y:.578125},{x:.484375,y:.578125},{x:.515625,y:.578125},{x:.515625,y:.578125},{x:.546875,y:.578125},{x:.546875,y:.578125},{x:.578125,y:.578125},{x:.578125,y:.578125},{x:.609375,y:.578125},{x:.609375,y:.578125},{x:.640625,y:.578125},{x:.640625,y:.578125},{x:.671875,y:.578125},{x:.671875,y:.578125},{x:.703125,y:.578125},{x:.703125,y:.578125},{x:.734375,y:.578125},{x:.734375,y:.578125},{x:.765625,y:.578125},{x:.765625,y:.578125},{x:.796875,y:.578125},{x:.796875,y:.578125},{x:.828125,y:.578125},{x:.828125,y:.578125},{x:.859375,y:.578125},{x:.859375,y:.578125},{x:.890625,y:.578125},{x:.890625,y:.578125},{x:.921875,y:.578125},{x:.921875,y:.578125},{x:.953125,y:.578125},{x:.953125,y:.578125},{x:.984375,y:.578125},{x:.984375,y:.578125},{x:.015625,y:.609375},{x:.015625,y:.609375},{x:.046875,y:.609375},{x:.046875,y:.609375},{x:.078125,y:.609375},{x:.078125,y:.609375},{x:.109375,y:.609375},{x:.109375,y:.609375},{x:.140625,y:.609375},{x:.140625,y:.609375},{x:.171875,y:.609375},{x:.171875,y:.609375},{x:.203125,y:.609375},{x:.203125,y:.609375},{x:.234375,y:.609375},{x:.234375,y:.609375},{x:.265625,y:.609375},{x:.265625,y:.609375},{x:.296875,y:.609375},{x:.296875,y:.609375},{x:.328125,y:.609375},{x:.328125,y:.609375},{x:.359375,y:.609375},{x:.359375,y:.609375},{x:.390625,y:.609375},{x:.390625,y:.609375},{x:.421875,y:.609375},{x:.421875,y:.609375},{x:.453125,y:.609375},{x:.453125,y:.609375},{x:.484375,y:.609375},{x:.484375,y:.609375},{x:.515625,y:.609375},{x:.515625,y:.609375},{x:.546875,y:.609375},{x:.546875,y:.609375},{x:.578125,y:.609375},{x:.578125,y:.609375},{x:.609375,y:.609375},{x:.609375,y:.609375},{x:.640625,y:.609375},{x:.640625,y:.609375},{x:.671875,y:.609375},{x:.671875,y:.609375},{x:.703125,y:.609375},{x:.703125,y:.609375},{x:.734375,y:.609375},{x:.734375,y:.609375},{x:.765625,y:.609375},{x:.765625,y:.609375},{x:.796875,y:.609375},{x:.796875,y:.609375},{x:.828125,y:.609375},{x:.828125,y:.609375},{x:.859375,y:.609375},{x:.859375,y:.609375},{x:.890625,y:.609375},{x:.890625,y:.609375},{x:.921875,y:.609375},{x:.921875,y:.609375},{x:.953125,y:.609375},{x:.953125,y:.609375},{x:.984375,y:.609375},{x:.984375,y:.609375},{x:.015625,y:.640625},{x:.015625,y:.640625},{x:.046875,y:.640625},{x:.046875,y:.640625},{x:.078125,y:.640625},{x:.078125,y:.640625},{x:.109375,y:.640625},{x:.109375,y:.640625},{x:.140625,y:.640625},{x:.140625,y:.640625},{x:.171875,y:.640625},{x:.171875,y:.640625},{x:.203125,y:.640625},{x:.203125,y:.640625},{x:.234375,y:.640625},{x:.234375,y:.640625},{x:.265625,y:.640625},{x:.265625,y:.640625},{x:.296875,y:.640625},{x:.296875,y:.640625},{x:.328125,y:.640625},{x:.328125,y:.640625},{x:.359375,y:.640625},{x:.359375,y:.640625},{x:.390625,y:.640625},{x:.390625,y:.640625},{x:.421875,y:.640625},{x:.421875,y:.640625},{x:.453125,y:.640625},{x:.453125,y:.640625},{x:.484375,y:.640625},{x:.484375,y:.640625},{x:.515625,y:.640625},{x:.515625,y:.640625},{x:.546875,y:.640625},{x:.546875,y:.640625},{x:.578125,y:.640625},{x:.578125,y:.640625},{x:.609375,y:.640625},{x:.609375,y:.640625},{x:.640625,y:.640625},{x:.640625,y:.640625},{x:.671875,y:.640625},{x:.671875,y:.640625},{x:.703125,y:.640625},{x:.703125,y:.640625},{x:.734375,y:.640625},{x:.734375,y:.640625},{x:.765625,y:.640625},{x:.765625,y:.640625},{x:.796875,y:.640625},{x:.796875,y:.640625},{x:.828125,y:.640625},{x:.828125,y:.640625},{x:.859375,y:.640625},{x:.859375,y:.640625},{x:.890625,y:.640625},{x:.890625,y:.640625},{x:.921875,y:.640625},{x:.921875,y:.640625},{x:.953125,y:.640625},{x:.953125,y:.640625},{x:.984375,y:.640625},{x:.984375,y:.640625},{x:.015625,y:.671875},{x:.015625,y:.671875},{x:.046875,y:.671875},{x:.046875,y:.671875},{x:.078125,y:.671875},{x:.078125,y:.671875},{x:.109375,y:.671875},{x:.109375,y:.671875},{x:.140625,y:.671875},{x:.140625,y:.671875},{x:.171875,y:.671875},{x:.171875,y:.671875},{x:.203125,y:.671875},{x:.203125,y:.671875},{x:.234375,y:.671875},{x:.234375,y:.671875},{x:.265625,y:.671875},{x:.265625,y:.671875},{x:.296875,y:.671875},{x:.296875,y:.671875},{x:.328125,y:.671875},{x:.328125,y:.671875},{x:.359375,y:.671875},{x:.359375,y:.671875},{x:.390625,y:.671875},{x:.390625,y:.671875},{x:.421875,y:.671875},{x:.421875,y:.671875},{x:.453125,y:.671875},{x:.453125,y:.671875},{x:.484375,y:.671875},{x:.484375,y:.671875},{x:.515625,y:.671875},{x:.515625,y:.671875},{x:.546875,y:.671875},{x:.546875,y:.671875},{x:.578125,y:.671875},{x:.578125,y:.671875},{x:.609375,y:.671875},{x:.609375,y:.671875},{x:.640625,y:.671875},{x:.640625,y:.671875},{x:.671875,y:.671875},{x:.671875,y:.671875},{x:.703125,y:.671875},{x:.703125,y:.671875},{x:.734375,y:.671875},{x:.734375,y:.671875},{x:.765625,y:.671875},{x:.765625,y:.671875},{x:.796875,y:.671875},{x:.796875,y:.671875},{x:.828125,y:.671875},{x:.828125,y:.671875},{x:.859375,y:.671875},{x:.859375,y:.671875},{x:.890625,y:.671875},{x:.890625,y:.671875},{x:.921875,y:.671875},{x:.921875,y:.671875},{x:.953125,y:.671875},{x:.953125,y:.671875},{x:.984375,y:.671875},{x:.984375,y:.671875},{x:.015625,y:.703125},{x:.015625,y:.703125},{x:.046875,y:.703125},{x:.046875,y:.703125},{x:.078125,y:.703125},{x:.078125,y:.703125},{x:.109375,y:.703125},{x:.109375,y:.703125},{x:.140625,y:.703125},{x:.140625,y:.703125},{x:.171875,y:.703125},{x:.171875,y:.703125},{x:.203125,y:.703125},{x:.203125,y:.703125},{x:.234375,y:.703125},{x:.234375,y:.703125},{x:.265625,y:.703125},{x:.265625,y:.703125},{x:.296875,y:.703125},{x:.296875,y:.703125},{x:.328125,y:.703125},{x:.328125,y:.703125},{x:.359375,y:.703125},{x:.359375,y:.703125},{x:.390625,y:.703125},{x:.390625,y:.703125},{x:.421875,y:.703125},{x:.421875,y:.703125},{x:.453125,y:.703125},{x:.453125,y:.703125},{x:.484375,y:.703125},{x:.484375,y:.703125},{x:.515625,y:.703125},{x:.515625,y:.703125},{x:.546875,y:.703125},{x:.546875,y:.703125},{x:.578125,y:.703125},{x:.578125,y:.703125},{x:.609375,y:.703125},{x:.609375,y:.703125},{x:.640625,y:.703125},{x:.640625,y:.703125},{x:.671875,y:.703125},{x:.671875,y:.703125},{x:.703125,y:.703125},{x:.703125,y:.703125},{x:.734375,y:.703125},{x:.734375,y:.703125},{x:.765625,y:.703125},{x:.765625,y:.703125},{x:.796875,y:.703125},{x:.796875,y:.703125},{x:.828125,y:.703125},{x:.828125,y:.703125},{x:.859375,y:.703125},{x:.859375,y:.703125},{x:.890625,y:.703125},{x:.890625,y:.703125},{x:.921875,y:.703125},{x:.921875,y:.703125},{x:.953125,y:.703125},{x:.953125,y:.703125},{x:.984375,y:.703125},{x:.984375,y:.703125},{x:.015625,y:.734375},{x:.015625,y:.734375},{x:.046875,y:.734375},{x:.046875,y:.734375},{x:.078125,y:.734375},{x:.078125,y:.734375},{x:.109375,y:.734375},{x:.109375,y:.734375},{x:.140625,y:.734375},{x:.140625,y:.734375},{x:.171875,y:.734375},{x:.171875,y:.734375},{x:.203125,y:.734375},{x:.203125,y:.734375},{x:.234375,y:.734375},{x:.234375,y:.734375},{x:.265625,y:.734375},{x:.265625,y:.734375},{x:.296875,y:.734375},{x:.296875,y:.734375},{x:.328125,y:.734375},{x:.328125,y:.734375},{x:.359375,y:.734375},{x:.359375,y:.734375},{x:.390625,y:.734375},{x:.390625,y:.734375},{x:.421875,y:.734375},{x:.421875,y:.734375},{x:.453125,y:.734375},{x:.453125,y:.734375},{x:.484375,y:.734375},{x:.484375,y:.734375},{x:.515625,y:.734375},{x:.515625,y:.734375},{x:.546875,y:.734375},{x:.546875,y:.734375},{x:.578125,y:.734375},{x:.578125,y:.734375},{x:.609375,y:.734375},{x:.609375,y:.734375},{x:.640625,y:.734375},{x:.640625,y:.734375},{x:.671875,y:.734375},{x:.671875,y:.734375},{x:.703125,y:.734375},{x:.703125,y:.734375},{x:.734375,y:.734375},{x:.734375,y:.734375},{x:.765625,y:.734375},{x:.765625,y:.734375},{x:.796875,y:.734375},{x:.796875,y:.734375},{x:.828125,y:.734375},{x:.828125,y:.734375},{x:.859375,y:.734375},{x:.859375,y:.734375},{x:.890625,y:.734375},{x:.890625,y:.734375},{x:.921875,y:.734375},{x:.921875,y:.734375},{x:.953125,y:.734375},{x:.953125,y:.734375},{x:.984375,y:.734375},{x:.984375,y:.734375},{x:.015625,y:.765625},{x:.015625,y:.765625},{x:.046875,y:.765625},{x:.046875,y:.765625},{x:.078125,y:.765625},{x:.078125,y:.765625},{x:.109375,y:.765625},{x:.109375,y:.765625},{x:.140625,y:.765625},{x:.140625,y:.765625},{x:.171875,y:.765625},{x:.171875,y:.765625},{x:.203125,y:.765625},{x:.203125,y:.765625},{x:.234375,y:.765625},{x:.234375,y:.765625},{x:.265625,y:.765625},{x:.265625,y:.765625},{x:.296875,y:.765625},{x:.296875,y:.765625},{x:.328125,y:.765625},{x:.328125,y:.765625},{x:.359375,y:.765625},{x:.359375,y:.765625},{x:.390625,y:.765625},{x:.390625,y:.765625},{x:.421875,y:.765625},{x:.421875,y:.765625},{x:.453125,y:.765625},{x:.453125,y:.765625},{x:.484375,y:.765625},{x:.484375,y:.765625},{x:.515625,y:.765625},{x:.515625,y:.765625},{x:.546875,y:.765625},{x:.546875,y:.765625},{x:.578125,y:.765625},{x:.578125,y:.765625},{x:.609375,y:.765625},{x:.609375,y:.765625},{x:.640625,y:.765625},{x:.640625,y:.765625},{x:.671875,y:.765625},{x:.671875,y:.765625},{x:.703125,y:.765625},{x:.703125,y:.765625},{x:.734375,y:.765625},{x:.734375,y:.765625},{x:.765625,y:.765625},{x:.765625,y:.765625},{x:.796875,y:.765625},{x:.796875,y:.765625},{x:.828125,y:.765625},{x:.828125,y:.765625},{x:.859375,y:.765625},{x:.859375,y:.765625},{x:.890625,y:.765625},{x:.890625,y:.765625},{x:.921875,y:.765625},{x:.921875,y:.765625},{x:.953125,y:.765625},{x:.953125,y:.765625},{x:.984375,y:.765625},{x:.984375,y:.765625},{x:.015625,y:.796875},{x:.015625,y:.796875},{x:.046875,y:.796875},{x:.046875,y:.796875},{x:.078125,y:.796875},{x:.078125,y:.796875},{x:.109375,y:.796875},{x:.109375,y:.796875},{x:.140625,y:.796875},{x:.140625,y:.796875},{x:.171875,y:.796875},{x:.171875,y:.796875},{x:.203125,y:.796875},{x:.203125,y:.796875},{x:.234375,y:.796875},{x:.234375,y:.796875},{x:.265625,y:.796875},{x:.265625,y:.796875},{x:.296875,y:.796875},{x:.296875,y:.796875},{x:.328125,y:.796875},{x:.328125,y:.796875},{x:.359375,y:.796875},{x:.359375,y:.796875},{x:.390625,y:.796875},{x:.390625,y:.796875},{x:.421875,y:.796875},{x:.421875,y:.796875},{x:.453125,y:.796875},{x:.453125,y:.796875},{x:.484375,y:.796875},{x:.484375,y:.796875},{x:.515625,y:.796875},{x:.515625,y:.796875},{x:.546875,y:.796875},{x:.546875,y:.796875},{x:.578125,y:.796875},{x:.578125,y:.796875},{x:.609375,y:.796875},{x:.609375,y:.796875},{x:.640625,y:.796875},{x:.640625,y:.796875},{x:.671875,y:.796875},{x:.671875,y:.796875},{x:.703125,y:.796875},{x:.703125,y:.796875},{x:.734375,y:.796875},{x:.734375,y:.796875},{x:.765625,y:.796875},{x:.765625,y:.796875},{x:.796875,y:.796875},{x:.796875,y:.796875},{x:.828125,y:.796875},{x:.828125,y:.796875},{x:.859375,y:.796875},{x:.859375,y:.796875},{x:.890625,y:.796875},{x:.890625,y:.796875},{x:.921875,y:.796875},{x:.921875,y:.796875},{x:.953125,y:.796875},{x:.953125,y:.796875},{x:.984375,y:.796875},{x:.984375,y:.796875},{x:.015625,y:.828125},{x:.015625,y:.828125},{x:.046875,y:.828125},{x:.046875,y:.828125},{x:.078125,y:.828125},{x:.078125,y:.828125},{x:.109375,y:.828125},{x:.109375,y:.828125},{x:.140625,y:.828125},{x:.140625,y:.828125},{x:.171875,y:.828125},{x:.171875,y:.828125},{x:.203125,y:.828125},{x:.203125,y:.828125},{x:.234375,y:.828125},{x:.234375,y:.828125},{x:.265625,y:.828125},{x:.265625,y:.828125},{x:.296875,y:.828125},{x:.296875,y:.828125},{x:.328125,y:.828125},{x:.328125,y:.828125},{x:.359375,y:.828125},{x:.359375,y:.828125},{x:.390625,y:.828125},{x:.390625,y:.828125},{x:.421875,y:.828125},{x:.421875,y:.828125},{x:.453125,y:.828125},{x:.453125,y:.828125},{x:.484375,y:.828125},{x:.484375,y:.828125},{x:.515625,y:.828125},{x:.515625,y:.828125},{x:.546875,y:.828125},{x:.546875,y:.828125},{x:.578125,y:.828125},{x:.578125,y:.828125},{x:.609375,y:.828125},{x:.609375,y:.828125},{x:.640625,y:.828125},{x:.640625,y:.828125},{x:.671875,y:.828125},{x:.671875,y:.828125},{x:.703125,y:.828125},{x:.703125,y:.828125},{x:.734375,y:.828125},{x:.734375,y:.828125},{x:.765625,y:.828125},{x:.765625,y:.828125},{x:.796875,y:.828125},{x:.796875,y:.828125},{x:.828125,y:.828125},{x:.828125,y:.828125},{x:.859375,y:.828125},{x:.859375,y:.828125},{x:.890625,y:.828125},{x:.890625,y:.828125},{x:.921875,y:.828125},{x:.921875,y:.828125},{x:.953125,y:.828125},{x:.953125,y:.828125},{x:.984375,y:.828125},{x:.984375,y:.828125},{x:.015625,y:.859375},{x:.015625,y:.859375},{x:.046875,y:.859375},{x:.046875,y:.859375},{x:.078125,y:.859375},{x:.078125,y:.859375},{x:.109375,y:.859375},{x:.109375,y:.859375},{x:.140625,y:.859375},{x:.140625,y:.859375},{x:.171875,y:.859375},{x:.171875,y:.859375},{x:.203125,y:.859375},{x:.203125,y:.859375},{x:.234375,y:.859375},{x:.234375,y:.859375},{x:.265625,y:.859375},{x:.265625,y:.859375},{x:.296875,y:.859375},{x:.296875,y:.859375},{x:.328125,y:.859375},{x:.328125,y:.859375},{x:.359375,y:.859375},{x:.359375,y:.859375},{x:.390625,y:.859375},{x:.390625,y:.859375},{x:.421875,y:.859375},{x:.421875,y:.859375},{x:.453125,y:.859375},{x:.453125,y:.859375},{x:.484375,y:.859375},{x:.484375,y:.859375},{x:.515625,y:.859375},{x:.515625,y:.859375},{x:.546875,y:.859375},{x:.546875,y:.859375},{x:.578125,y:.859375},{x:.578125,y:.859375},{x:.609375,y:.859375},{x:.609375,y:.859375},{x:.640625,y:.859375},{x:.640625,y:.859375},{x:.671875,y:.859375},{x:.671875,y:.859375},{x:.703125,y:.859375},{x:.703125,y:.859375},{x:.734375,y:.859375},{x:.734375,y:.859375},{x:.765625,y:.859375},{x:.765625,y:.859375},{x:.796875,y:.859375},{x:.796875,y:.859375},{x:.828125,y:.859375},{x:.828125,y:.859375},{x:.859375,y:.859375},{x:.859375,y:.859375},{x:.890625,y:.859375},{x:.890625,y:.859375},{x:.921875,y:.859375},{x:.921875,y:.859375},{x:.953125,y:.859375},{x:.953125,y:.859375},{x:.984375,y:.859375},{x:.984375,y:.859375},{x:.015625,y:.890625},{x:.015625,y:.890625},{x:.046875,y:.890625},{x:.046875,y:.890625},{x:.078125,y:.890625},{x:.078125,y:.890625},{x:.109375,y:.890625},{x:.109375,y:.890625},{x:.140625,y:.890625},{x:.140625,y:.890625},{x:.171875,y:.890625},{x:.171875,y:.890625},{x:.203125,y:.890625},{x:.203125,y:.890625},{x:.234375,y:.890625},{x:.234375,y:.890625},{x:.265625,y:.890625},{x:.265625,y:.890625},{x:.296875,y:.890625},{x:.296875,y:.890625},{x:.328125,y:.890625},{x:.328125,y:.890625},{x:.359375,y:.890625},{x:.359375,y:.890625},{x:.390625,y:.890625},{x:.390625,y:.890625},{x:.421875,y:.890625},{x:.421875,y:.890625},{x:.453125,y:.890625},{x:.453125,y:.890625},{x:.484375,y:.890625},{x:.484375,y:.890625},{x:.515625,y:.890625},{x:.515625,y:.890625},{x:.546875,y:.890625},{x:.546875,y:.890625},{x:.578125,y:.890625},{x:.578125,y:.890625},{x:.609375,y:.890625},{x:.609375,y:.890625},{x:.640625,y:.890625},{x:.640625,y:.890625},{x:.671875,y:.890625},{x:.671875,y:.890625},{x:.703125,y:.890625},{x:.703125,y:.890625},{x:.734375,y:.890625},{x:.734375,y:.890625},{x:.765625,y:.890625},{x:.765625,y:.890625},{x:.796875,y:.890625},{x:.796875,y:.890625},{x:.828125,y:.890625},{x:.828125,y:.890625},{x:.859375,y:.890625},{x:.859375,y:.890625},{x:.890625,y:.890625},{x:.890625,y:.890625},{x:.921875,y:.890625},{x:.921875,y:.890625},{x:.953125,y:.890625},{x:.953125,y:.890625},{x:.984375,y:.890625},{x:.984375,y:.890625},{x:.015625,y:.921875},{x:.015625,y:.921875},{x:.046875,y:.921875},{x:.046875,y:.921875},{x:.078125,y:.921875},{x:.078125,y:.921875},{x:.109375,y:.921875},{x:.109375,y:.921875},{x:.140625,y:.921875},{x:.140625,y:.921875},{x:.171875,y:.921875},{x:.171875,y:.921875},{x:.203125,y:.921875},{x:.203125,y:.921875},{x:.234375,y:.921875},{x:.234375,y:.921875},{x:.265625,y:.921875},{x:.265625,y:.921875},{x:.296875,y:.921875},{x:.296875,y:.921875},{x:.328125,y:.921875},{x:.328125,y:.921875},{x:.359375,y:.921875},{x:.359375,y:.921875},{x:.390625,y:.921875},{x:.390625,y:.921875},{x:.421875,y:.921875},{x:.421875,y:.921875},{x:.453125,y:.921875},{x:.453125,y:.921875},{x:.484375,y:.921875},{x:.484375,y:.921875},{x:.515625,y:.921875},{x:.515625,y:.921875},{x:.546875,y:.921875},{x:.546875,y:.921875},{x:.578125,y:.921875},{x:.578125,y:.921875},{x:.609375,y:.921875},{x:.609375,y:.921875},{x:.640625,y:.921875},{x:.640625,y:.921875},{x:.671875,y:.921875},{x:.671875,y:.921875},{x:.703125,y:.921875},{x:.703125,y:.921875},{x:.734375,y:.921875},{x:.734375,y:.921875},{x:.765625,y:.921875},{x:.765625,y:.921875},{x:.796875,y:.921875},{x:.796875,y:.921875},{x:.828125,y:.921875},{x:.828125,y:.921875},{x:.859375,y:.921875},{x:.859375,y:.921875},{x:.890625,y:.921875},{x:.890625,y:.921875},{x:.921875,y:.921875},{x:.921875,y:.921875},{x:.953125,y:.921875},{x:.953125,y:.921875},{x:.984375,y:.921875},{x:.984375,y:.921875},{x:.015625,y:.953125},{x:.015625,y:.953125},{x:.046875,y:.953125},{x:.046875,y:.953125},{x:.078125,y:.953125},{x:.078125,y:.953125},{x:.109375,y:.953125},{x:.109375,y:.953125},{x:.140625,y:.953125},{x:.140625,y:.953125},{x:.171875,y:.953125},{x:.171875,y:.953125},{x:.203125,y:.953125},{x:.203125,y:.953125},{x:.234375,y:.953125},{x:.234375,y:.953125},{x:.265625,y:.953125},{x:.265625,y:.953125},{x:.296875,y:.953125},{x:.296875,y:.953125},{x:.328125,y:.953125},{x:.328125,y:.953125},{x:.359375,y:.953125},{x:.359375,y:.953125},{x:.390625,y:.953125},{x:.390625,y:.953125},{x:.421875,y:.953125},{x:.421875,y:.953125},{x:.453125,y:.953125},{x:.453125,y:.953125},{x:.484375,y:.953125},{x:.484375,y:.953125},{x:.515625,y:.953125},{x:.515625,y:.953125},{x:.546875,y:.953125},{x:.546875,y:.953125},{x:.578125,y:.953125},{x:.578125,y:.953125},{x:.609375,y:.953125},{x:.609375,y:.953125},{x:.640625,y:.953125},{x:.640625,y:.953125},{x:.671875,y:.953125},{x:.671875,y:.953125},{x:.703125,y:.953125},{x:.703125,y:.953125},{x:.734375,y:.953125},{x:.734375,y:.953125},{x:.765625,y:.953125},{x:.765625,y:.953125},{x:.796875,y:.953125},{x:.796875,y:.953125},{x:.828125,y:.953125},{x:.828125,y:.953125},{x:.859375,y:.953125},{x:.859375,y:.953125},{x:.890625,y:.953125},{x:.890625,y:.953125},{x:.921875,y:.953125},{x:.921875,y:.953125},{x:.953125,y:.953125},{x:.953125,y:.953125},{x:.984375,y:.953125},{x:.984375,y:.953125},{x:.015625,y:.984375},{x:.015625,y:.984375},{x:.046875,y:.984375},{x:.046875,y:.984375},{x:.078125,y:.984375},{x:.078125,y:.984375},{x:.109375,y:.984375},{x:.109375,y:.984375},{x:.140625,y:.984375},{x:.140625,y:.984375},{x:.171875,y:.984375},{x:.171875,y:.984375},{x:.203125,y:.984375},{x:.203125,y:.984375},{x:.234375,y:.984375},{x:.234375,y:.984375},{x:.265625,y:.984375},{x:.265625,y:.984375},{x:.296875,y:.984375},{x:.296875,y:.984375},{x:.328125,y:.984375},{x:.328125,y:.984375},{x:.359375,y:.984375},{x:.359375,y:.984375},{x:.390625,y:.984375},{x:.390625,y:.984375},{x:.421875,y:.984375},{x:.421875,y:.984375},{x:.453125,y:.984375},{x:.453125,y:.984375},{x:.484375,y:.984375},{x:.484375,y:.984375},{x:.515625,y:.984375},{x:.515625,y:.984375},{x:.546875,y:.984375},{x:.546875,y:.984375},{x:.578125,y:.984375},{x:.578125,y:.984375},{x:.609375,y:.984375},{x:.609375,y:.984375},{x:.640625,y:.984375},{x:.640625,y:.984375},{x:.671875,y:.984375},{x:.671875,y:.984375},{x:.703125,y:.984375},{x:.703125,y:.984375},{x:.734375,y:.984375},{x:.734375,y:.984375},{x:.765625,y:.984375},{x:.765625,y:.984375},{x:.796875,y:.984375},{x:.796875,y:.984375},{x:.828125,y:.984375},{x:.828125,y:.984375},{x:.859375,y:.984375},{x:.859375,y:.984375},{x:.890625,y:.984375},{x:.890625,y:.984375},{x:.921875,y:.984375},{x:.921875,y:.984375},{x:.953125,y:.984375},{x:.953125,y:.984375},{x:.984375,y:.984375},{x:.984375,y:.984375},{x:.03125,y:.03125},{x:.03125,y:.03125},{x:.09375,y:.03125},{x:.09375,y:.03125},{x:.15625,y:.03125},{x:.15625,y:.03125},{x:.21875,y:.03125},{x:.21875,y:.03125},{x:.28125,y:.03125},{x:.28125,y:.03125},{x:.34375,y:.03125},{x:.34375,y:.03125},{x:.40625,y:.03125},{x:.40625,y:.03125},{x:.46875,y:.03125},{x:.46875,y:.03125},{x:.53125,y:.03125},{x:.53125,y:.03125},{x:.59375,y:.03125},{x:.59375,y:.03125},{x:.65625,y:.03125},{x:.65625,y:.03125},{x:.71875,y:.03125},{x:.71875,y:.03125},{x:.78125,y:.03125},{x:.78125,y:.03125},{x:.84375,y:.03125},{x:.84375,y:.03125},{x:.90625,y:.03125},{x:.90625,y:.03125},{x:.96875,y:.03125},{x:.96875,y:.03125},{x:.03125,y:.09375},{x:.03125,y:.09375},{x:.09375,y:.09375},{x:.09375,y:.09375},{x:.15625,y:.09375},{x:.15625,y:.09375},{x:.21875,y:.09375},{x:.21875,y:.09375},{x:.28125,y:.09375},{x:.28125,y:.09375},{x:.34375,y:.09375},{x:.34375,y:.09375},{x:.40625,y:.09375},{x:.40625,y:.09375},{x:.46875,y:.09375},{x:.46875,y:.09375},{x:.53125,y:.09375},{x:.53125,y:.09375},{x:.59375,y:.09375},{x:.59375,y:.09375},{x:.65625,y:.09375},{x:.65625,y:.09375},{x:.71875,y:.09375},{x:.71875,y:.09375},{x:.78125,y:.09375},{x:.78125,y:.09375},{x:.84375,y:.09375},{x:.84375,y:.09375},{x:.90625,y:.09375},{x:.90625,y:.09375},{x:.96875,y:.09375},{x:.96875,y:.09375},{x:.03125,y:.15625},{x:.03125,y:.15625},{x:.09375,y:.15625},{x:.09375,y:.15625},{x:.15625,y:.15625},{x:.15625,y:.15625},{x:.21875,y:.15625},{x:.21875,y:.15625},{x:.28125,y:.15625},{x:.28125,y:.15625},{x:.34375,y:.15625},{x:.34375,y:.15625},{x:.40625,y:.15625},{x:.40625,y:.15625},{x:.46875,y:.15625},{x:.46875,y:.15625},{x:.53125,y:.15625},{x:.53125,y:.15625},{x:.59375,y:.15625},{x:.59375,y:.15625},{x:.65625,y:.15625},{x:.65625,y:.15625},{x:.71875,y:.15625},{x:.71875,y:.15625},{x:.78125,y:.15625},{x:.78125,y:.15625},{x:.84375,y:.15625},{x:.84375,y:.15625},{x:.90625,y:.15625},{x:.90625,y:.15625},{x:.96875,y:.15625},{x:.96875,y:.15625},{x:.03125,y:.21875},{x:.03125,y:.21875},{x:.09375,y:.21875},{x:.09375,y:.21875},{x:.15625,y:.21875},{x:.15625,y:.21875},{x:.21875,y:.21875},{x:.21875,y:.21875},{x:.28125,y:.21875},{x:.28125,y:.21875},{x:.34375,y:.21875},{x:.34375,y:.21875},{x:.40625,y:.21875},{x:.40625,y:.21875},{x:.46875,y:.21875},{x:.46875,y:.21875},{x:.53125,y:.21875},{x:.53125,y:.21875},{x:.59375,y:.21875},{x:.59375,y:.21875},{x:.65625,y:.21875},{x:.65625,y:.21875},{x:.71875,y:.21875},{x:.71875,y:.21875},{x:.78125,y:.21875},{x:.78125,y:.21875},{x:.84375,y:.21875},{x:.84375,y:.21875},{x:.90625,y:.21875},{x:.90625,y:.21875},{x:.96875,y:.21875},{x:.96875,y:.21875},{x:.03125,y:.28125},{x:.03125,y:.28125},{x:.09375,y:.28125},{x:.09375,y:.28125},{x:.15625,y:.28125},{x:.15625,y:.28125},{x:.21875,y:.28125},{x:.21875,y:.28125},{x:.28125,y:.28125},{x:.28125,y:.28125},{x:.34375,y:.28125},{x:.34375,y:.28125},{x:.40625,y:.28125},{x:.40625,y:.28125},{x:.46875,y:.28125},{x:.46875,y:.28125},{x:.53125,y:.28125},{x:.53125,y:.28125},{x:.59375,y:.28125},{x:.59375,y:.28125},{x:.65625,y:.28125},{x:.65625,y:.28125},{x:.71875,y:.28125},{x:.71875,y:.28125},{x:.78125,y:.28125},{x:.78125,y:.28125},{x:.84375,y:.28125},{x:.84375,y:.28125},{x:.90625,y:.28125},{x:.90625,y:.28125},{x:.96875,y:.28125},{x:.96875,y:.28125},{x:.03125,y:.34375},{x:.03125,y:.34375},{x:.09375,y:.34375},{x:.09375,y:.34375},{x:.15625,y:.34375},{x:.15625,y:.34375},{x:.21875,y:.34375},{x:.21875,y:.34375},{x:.28125,y:.34375},{x:.28125,y:.34375},{x:.34375,y:.34375},{x:.34375,y:.34375},{x:.40625,y:.34375},{x:.40625,y:.34375},{x:.46875,y:.34375},{x:.46875,y:.34375},{x:.53125,y:.34375},{x:.53125,y:.34375},{x:.59375,y:.34375},{x:.59375,y:.34375},{x:.65625,y:.34375},{x:.65625,y:.34375},{x:.71875,y:.34375},{x:.71875,y:.34375},{x:.78125,y:.34375},{x:.78125,y:.34375},{x:.84375,y:.34375},{x:.84375,y:.34375},{x:.90625,y:.34375},{x:.90625,y:.34375},{x:.96875,y:.34375},{x:.96875,y:.34375},{x:.03125,y:.40625},{x:.03125,y:.40625},{x:.09375,y:.40625},{x:.09375,y:.40625},{x:.15625,y:.40625},{x:.15625,y:.40625},{x:.21875,y:.40625},{x:.21875,y:.40625},{x:.28125,y:.40625},{x:.28125,y:.40625},{x:.34375,y:.40625},{x:.34375,y:.40625},{x:.40625,y:.40625},{x:.40625,y:.40625},{x:.46875,y:.40625},{x:.46875,y:.40625},{x:.53125,y:.40625},{x:.53125,y:.40625},{x:.59375,y:.40625},{x:.59375,y:.40625},{x:.65625,y:.40625},{x:.65625,y:.40625},{x:.71875,y:.40625},{x:.71875,y:.40625},{x:.78125,y:.40625},{x:.78125,y:.40625},{x:.84375,y:.40625},{x:.84375,y:.40625},{x:.90625,y:.40625},{x:.90625,y:.40625},{x:.96875,y:.40625},{x:.96875,y:.40625},{x:.03125,y:.46875},{x:.03125,y:.46875},{x:.09375,y:.46875},{x:.09375,y:.46875},{x:.15625,y:.46875},{x:.15625,y:.46875},{x:.21875,y:.46875},{x:.21875,y:.46875},{x:.28125,y:.46875},{x:.28125,y:.46875},{x:.34375,y:.46875},{x:.34375,y:.46875},{x:.40625,y:.46875},{x:.40625,y:.46875},{x:.46875,y:.46875},{x:.46875,y:.46875},{x:.53125,y:.46875},{x:.53125,y:.46875},{x:.59375,y:.46875},{x:.59375,y:.46875},{x:.65625,y:.46875},{x:.65625,y:.46875},{x:.71875,y:.46875},{x:.71875,y:.46875},{x:.78125,y:.46875},{x:.78125,y:.46875},{x:.84375,y:.46875},{x:.84375,y:.46875},{x:.90625,y:.46875},{x:.90625,y:.46875},{x:.96875,y:.46875},{x:.96875,y:.46875},{x:.03125,y:.53125},{x:.03125,y:.53125},{x:.09375,y:.53125},{x:.09375,y:.53125},{x:.15625,y:.53125},{x:.15625,y:.53125},{x:.21875,y:.53125},{x:.21875,y:.53125},{x:.28125,y:.53125},{x:.28125,y:.53125},{x:.34375,y:.53125},{x:.34375,y:.53125},{x:.40625,y:.53125},{x:.40625,y:.53125},{x:.46875,y:.53125},{x:.46875,y:.53125},{x:.53125,y:.53125},{x:.53125,y:.53125},{x:.59375,y:.53125},{x:.59375,y:.53125},{x:.65625,y:.53125},{x:.65625,y:.53125},{x:.71875,y:.53125},{x:.71875,y:.53125},{x:.78125,y:.53125},{x:.78125,y:.53125},{x:.84375,y:.53125},{x:.84375,y:.53125},{x:.90625,y:.53125},{x:.90625,y:.53125},{x:.96875,y:.53125},{x:.96875,y:.53125},{x:.03125,y:.59375},{x:.03125,y:.59375},{x:.09375,y:.59375},{x:.09375,y:.59375},{x:.15625,y:.59375},{x:.15625,y:.59375},{x:.21875,y:.59375},{x:.21875,y:.59375},{x:.28125,y:.59375},{x:.28125,y:.59375},{x:.34375,y:.59375},{x:.34375,y:.59375},{x:.40625,y:.59375},{x:.40625,y:.59375},{x:.46875,y:.59375},{x:.46875,y:.59375},{x:.53125,y:.59375},{x:.53125,y:.59375},{x:.59375,y:.59375},{x:.59375,y:.59375},{x:.65625,y:.59375},{x:.65625,y:.59375},{x:.71875,y:.59375},{x:.71875,y:.59375},{x:.78125,y:.59375},{x:.78125,y:.59375},{x:.84375,y:.59375},{x:.84375,y:.59375},{x:.90625,y:.59375},{x:.90625,y:.59375},{x:.96875,y:.59375},{x:.96875,y:.59375},{x:.03125,y:.65625},{x:.03125,y:.65625},{x:.09375,y:.65625},{x:.09375,y:.65625},{x:.15625,y:.65625},{x:.15625,y:.65625},{x:.21875,y:.65625},{x:.21875,y:.65625},{x:.28125,y:.65625},{x:.28125,y:.65625},{x:.34375,y:.65625},{x:.34375,y:.65625},{x:.40625,y:.65625},{x:.40625,y:.65625},{x:.46875,y:.65625},{x:.46875,y:.65625},{x:.53125,y:.65625},{x:.53125,y:.65625},{x:.59375,y:.65625},{x:.59375,y:.65625},{x:.65625,y:.65625},{x:.65625,y:.65625},{x:.71875,y:.65625},{x:.71875,y:.65625},{x:.78125,y:.65625},{x:.78125,y:.65625},{x:.84375,y:.65625},{x:.84375,y:.65625},{x:.90625,y:.65625},{x:.90625,y:.65625},{x:.96875,y:.65625},{x:.96875,y:.65625},{x:.03125,y:.71875},{x:.03125,y:.71875},{x:.09375,y:.71875},{x:.09375,y:.71875},{x:.15625,y:.71875},{x:.15625,y:.71875},{x:.21875,y:.71875},{x:.21875,y:.71875},{x:.28125,y:.71875},{x:.28125,y:.71875},{x:.34375,y:.71875},{x:.34375,y:.71875},{x:.40625,y:.71875},{x:.40625,y:.71875},{x:.46875,y:.71875},{x:.46875,y:.71875},{x:.53125,y:.71875},{x:.53125,y:.71875},{x:.59375,y:.71875},{x:.59375,y:.71875},{x:.65625,y:.71875},{x:.65625,y:.71875},{x:.71875,y:.71875},{x:.71875,y:.71875},{x:.78125,y:.71875},{x:.78125,y:.71875},{x:.84375,y:.71875},{x:.84375,y:.71875},{x:.90625,y:.71875},{x:.90625,y:.71875},{x:.96875,y:.71875},{x:.96875,y:.71875},{x:.03125,y:.78125},{x:.03125,y:.78125},{x:.09375,y:.78125},{x:.09375,y:.78125},{x:.15625,y:.78125},{x:.15625,y:.78125},{x:.21875,y:.78125},{x:.21875,y:.78125},{x:.28125,y:.78125},{x:.28125,y:.78125},{x:.34375,y:.78125},{x:.34375,y:.78125},{x:.40625,y:.78125},{x:.40625,y:.78125},{x:.46875,y:.78125},{x:.46875,y:.78125},{x:.53125,y:.78125},{x:.53125,y:.78125},{x:.59375,y:.78125},{x:.59375,y:.78125},{x:.65625,y:.78125},{x:.65625,y:.78125},{x:.71875,y:.78125},{x:.71875,y:.78125},{x:.78125,y:.78125},{x:.78125,y:.78125},{x:.84375,y:.78125},{x:.84375,y:.78125},{x:.90625,y:.78125},{x:.90625,y:.78125},{x:.96875,y:.78125},{x:.96875,y:.78125},{x:.03125,y:.84375},{x:.03125,y:.84375},{x:.09375,y:.84375},{x:.09375,y:.84375},{x:.15625,y:.84375},{x:.15625,y:.84375},{x:.21875,y:.84375},{x:.21875,y:.84375},{x:.28125,y:.84375},{x:.28125,y:.84375},{x:.34375,y:.84375},{x:.34375,y:.84375},{x:.40625,y:.84375},{x:.40625,y:.84375},{x:.46875,y:.84375},{x:.46875,y:.84375},{x:.53125,y:.84375},{x:.53125,y:.84375},{x:.59375,y:.84375},{x:.59375,y:.84375},{x:.65625,y:.84375},{x:.65625,y:.84375},{x:.71875,y:.84375},{x:.71875,y:.84375},{x:.78125,y:.84375},{x:.78125,y:.84375},{x:.84375,y:.84375},{x:.84375,y:.84375},{x:.90625,y:.84375},{x:.90625,y:.84375},{x:.96875,y:.84375},{x:.96875,y:.84375},{x:.03125,y:.90625},{x:.03125,y:.90625},{x:.09375,y:.90625},{x:.09375,y:.90625},{x:.15625,y:.90625},{x:.15625,y:.90625},{x:.21875,y:.90625},{x:.21875,y:.90625},{x:.28125,y:.90625},{x:.28125,y:.90625},{x:.34375,y:.90625},{x:.34375,y:.90625},{x:.40625,y:.90625},{x:.40625,y:.90625},{x:.46875,y:.90625},{x:.46875,y:.90625},{x:.53125,y:.90625},{x:.53125,y:.90625},{x:.59375,y:.90625},{x:.59375,y:.90625},{x:.65625,y:.90625},{x:.65625,y:.90625},{x:.71875,y:.90625},{x:.71875,y:.90625},{x:.78125,y:.90625},{x:.78125,y:.90625},{x:.84375,y:.90625},{x:.84375,y:.90625},{x:.90625,y:.90625},{x:.90625,y:.90625},{x:.96875,y:.90625},{x:.96875,y:.90625},{x:.03125,y:.96875},{x:.03125,y:.96875},{x:.09375,y:.96875},{x:.09375,y:.96875},{x:.15625,y:.96875},{x:.15625,y:.96875},{x:.21875,y:.96875},{x:.21875,y:.96875},{x:.28125,y:.96875},{x:.28125,y:.96875},{x:.34375,y:.96875},{x:.34375,y:.96875},{x:.40625,y:.96875},{x:.40625,y:.96875},{x:.46875,y:.96875},{x:.46875,y:.96875},{x:.53125,y:.96875},{x:.53125,y:.96875},{x:.59375,y:.96875},{x:.59375,y:.96875},{x:.65625,y:.96875},{x:.65625,y:.96875},{x:.71875,y:.96875},{x:.71875,y:.96875},{x:.78125,y:.96875},{x:.78125,y:.96875},{x:.84375,y:.96875},{x:.84375,y:.96875},{x:.90625,y:.96875},{x:.90625,y:.96875},{x:.96875,y:.96875},{x:.96875,y:.96875},{x:.0625,y:.0625},{x:.0625,y:.0625},{x:.0625,y:.0625},{x:.0625,y:.0625},{x:.0625,y:.0625},{x:.0625,y:.0625},{x:.1875,y:.0625},{x:.1875,y:.0625},{x:.1875,y:.0625},{x:.1875,y:.0625},{x:.1875,y:.0625},{x:.1875,y:.0625},{x:.3125,y:.0625},{x:.3125,y:.0625},{x:.3125,y:.0625},{x:.3125,y:.0625},{x:.3125,y:.0625},{x:.3125,y:.0625},{x:.4375,y:.0625},{x:.4375,y:.0625},{x:.4375,y:.0625},{x:.4375,y:.0625},{x:.4375,y:.0625},{x:.4375,y:.0625},{x:.5625,y:.0625},{x:.5625,y:.0625},{x:.5625,y:.0625},{x:.5625,y:.0625},{x:.5625,y:.0625},{x:.5625,y:.0625},{x:.6875,y:.0625},{x:.6875,y:.0625},{x:.6875,y:.0625},{x:.6875,y:.0625},{x:.6875,y:.0625},{x:.6875,y:.0625},{x:.8125,y:.0625},{x:.8125,y:.0625},{x:.8125,y:.0625},{x:.8125,y:.0625},{x:.8125,y:.0625},{x:.8125,y:.0625},{x:.9375,y:.0625},{x:.9375,y:.0625},{x:.9375,y:.0625},{x:.9375,y:.0625},{x:.9375,y:.0625},{x:.9375,y:.0625},{x:.0625,y:.1875},{x:.0625,y:.1875},{x:.0625,y:.1875},{x:.0625,y:.1875},{x:.0625,y:.1875},{x:.0625,y:.1875},{x:.1875,y:.1875},{x:.1875,y:.1875},{x:.1875,y:.1875},{x:.1875,y:.1875},{x:.1875,y:.1875},{x:.1875,y:.1875},{x:.3125,y:.1875},{x:.3125,y:.1875},{x:.3125,y:.1875},{x:.3125,y:.1875},{x:.3125,y:.1875},{x:.3125,y:.1875},{x:.4375,y:.1875},{x:.4375,y:.1875},{x:.4375,y:.1875},{x:.4375,y:.1875},{x:.4375,y:.1875},{x:.4375,y:.1875},{x:.5625,y:.1875},{x:.5625,y:.1875},{x:.5625,y:.1875},{x:.5625,y:.1875},{x:.5625,y:.1875},{x:.5625,y:.1875},{x:.6875,y:.1875},{x:.6875,y:.1875},{x:.6875,y:.1875},{x:.6875,y:.1875},{x:.6875,y:.1875},{x:.6875,y:.1875},{x:.8125,y:.1875},{x:.8125,y:.1875},{x:.8125,y:.1875},{x:.8125,y:.1875},{x:.8125,y:.1875},{x:.8125,y:.1875},{x:.9375,y:.1875},{x:.9375,y:.1875},{x:.9375,y:.1875},{x:.9375,y:.1875},{x:.9375,y:.1875},{x:.9375,y:.1875},{x:.0625,y:.3125},{x:.0625,y:.3125},{x:.0625,y:.3125},{x:.0625,y:.3125},{x:.0625,y:.3125},{x:.0625,y:.3125},{x:.1875,y:.3125},{x:.1875,y:.3125},{x:.1875,y:.3125},{x:.1875,y:.3125},{x:.1875,y:.3125},{x:.1875,y:.3125},{x:.3125,y:.3125},{x:.3125,y:.3125},{x:.3125,y:.3125},{x:.3125,y:.3125},{x:.3125,y:.3125},{x:.3125,y:.3125},{x:.4375,y:.3125},{x:.4375,y:.3125},{x:.4375,y:.3125},{x:.4375,y:.3125},{x:.4375,y:.3125},{x:.4375,y:.3125},{x:.5625,y:.3125},{x:.5625,y:.3125},{x:.5625,y:.3125},{x:.5625,y:.3125},{x:.5625,y:.3125},{x:.5625,y:.3125},{x:.6875,y:.3125},{x:.6875,y:.3125},{x:.6875,y:.3125},{x:.6875,y:.3125},{x:.6875,y:.3125},{x:.6875,y:.3125},{x:.8125,y:.3125},{x:.8125,y:.3125},{x:.8125,y:.3125},{x:.8125,y:.3125},{x:.8125,y:.3125},{x:.8125,y:.3125},{x:.9375,y:.3125},{x:.9375,y:.3125},{x:.9375,y:.3125},{x:.9375,y:.3125},{x:.9375,y:.3125},{x:.9375,y:.3125},{x:.0625,y:.4375},{x:.0625,y:.4375},{x:.0625,y:.4375},{x:.0625,y:.4375},{x:.0625,y:.4375},{x:.0625,y:.4375},{x:.1875,y:.4375},{x:.1875,y:.4375},{x:.1875,y:.4375},{x:.1875,y:.4375},{x:.1875,y:.4375},{x:.1875,y:.4375},{x:.3125,y:.4375},{x:.3125,y:.4375},{x:.3125,y:.4375},{x:.3125,y:.4375},{x:.3125,y:.4375},{x:.3125,y:.4375},{x:.4375,y:.4375},{x:.4375,y:.4375},{x:.4375,y:.4375},{x:.4375,y:.4375},{x:.4375,y:.4375},{x:.4375,y:.4375},{x:.5625,y:.4375},{x:.5625,y:.4375},{x:.5625,y:.4375},{x:.5625,y:.4375},{x:.5625,y:.4375},{x:.5625,y:.4375},{x:.6875,y:.4375},{x:.6875,y:.4375},{x:.6875,y:.4375},{x:.6875,y:.4375},{x:.6875,y:.4375},{x:.6875,y:.4375},{x:.8125,y:.4375},{x:.8125,y:.4375},{x:.8125,y:.4375},{x:.8125,y:.4375},{x:.8125,y:.4375},{x:.8125,y:.4375},{x:.9375,y:.4375},{x:.9375,y:.4375},{x:.9375,y:.4375},{x:.9375,y:.4375},{x:.9375,y:.4375},{x:.9375,y:.4375},{x:.0625,y:.5625},{x:.0625,y:.5625},{x:.0625,y:.5625},{x:.0625,y:.5625},{x:.0625,y:.5625},{x:.0625,y:.5625},{x:.1875,y:.5625},{x:.1875,y:.5625},{x:.1875,y:.5625},{x:.1875,y:.5625},{x:.1875,y:.5625},{x:.1875,y:.5625},{x:.3125,y:.5625},{x:.3125,y:.5625},{x:.3125,y:.5625},{x:.3125,y:.5625},{x:.3125,y:.5625},{x:.3125,y:.5625},{x:.4375,y:.5625},{x:.4375,y:.5625},{x:.4375,y:.5625},{x:.4375,y:.5625},{x:.4375,y:.5625},{x:.4375,y:.5625},{x:.5625,y:.5625},{x:.5625,y:.5625},{x:.5625,y:.5625},{x:.5625,y:.5625},{x:.5625,y:.5625},{x:.5625,y:.5625},{x:.6875,y:.5625},{x:.6875,y:.5625},{x:.6875,y:.5625},{x:.6875,y:.5625},{x:.6875,y:.5625},{x:.6875,y:.5625},{x:.8125,y:.5625},{x:.8125,y:.5625},{x:.8125,y:.5625},{x:.8125,y:.5625},{x:.8125,y:.5625},{x:.8125,y:.5625},{x:.9375,y:.5625},{x:.9375,y:.5625},{x:.9375,y:.5625},{x:.9375,y:.5625},{x:.9375,y:.5625},{x:.9375,y:.5625},{x:.0625,y:.6875},{x:.0625,y:.6875},{x:.0625,y:.6875},{x:.0625,y:.6875},{x:.0625,y:.6875},{x:.0625,y:.6875},{x:.1875,y:.6875},{x:.1875,y:.6875},{x:.1875,y:.6875},{x:.1875,y:.6875},{x:.1875,y:.6875},{x:.1875,y:.6875},{x:.3125,y:.6875},{x:.3125,y:.6875},{x:.3125,y:.6875},{x:.3125,y:.6875},{x:.3125,y:.6875},{x:.3125,y:.6875},{x:.4375,y:.6875},{x:.4375,y:.6875},{x:.4375,y:.6875},{x:.4375,y:.6875},{x:.4375,y:.6875},{x:.4375,y:.6875},{x:.5625,y:.6875},{x:.5625,y:.6875},{x:.5625,y:.6875},{x:.5625,y:.6875},{x:.5625,y:.6875},{x:.5625,y:.6875},{x:.6875,y:.6875},{x:.6875,y:.6875},{x:.6875,y:.6875},{x:.6875,y:.6875},{x:.6875,y:.6875},{x:.6875,y:.6875},{x:.8125,y:.6875},{x:.8125,y:.6875},{x:.8125,y:.6875},{x:.8125,y:.6875},{x:.8125,y:.6875},{x:.8125,y:.6875},{x:.9375,y:.6875},{x:.9375,y:.6875},{x:.9375,y:.6875},{x:.9375,y:.6875},{x:.9375,y:.6875},{x:.9375,y:.6875},{x:.0625,y:.8125},{x:.0625,y:.8125},{x:.0625,y:.8125},{x:.0625,y:.8125},{x:.0625,y:.8125},{x:.0625,y:.8125},{x:.1875,y:.8125},{x:.1875,y:.8125},{x:.1875,y:.8125},{x:.1875,y:.8125},{x:.1875,y:.8125},{x:.1875,y:.8125},{x:.3125,y:.8125},{x:.3125,y:.8125},{x:.3125,y:.8125},{x:.3125,y:.8125},{x:.3125,y:.8125},{x:.3125,y:.8125},{x:.4375,y:.8125},{x:.4375,y:.8125},{x:.4375,y:.8125},{x:.4375,y:.8125},{x:.4375,y:.8125},{x:.4375,y:.8125},{x:.5625,y:.8125},{x:.5625,y:.8125},{x:.5625,y:.8125},{x:.5625,y:.8125},{x:.5625,y:.8125},{x:.5625,y:.8125},{x:.6875,y:.8125},{x:.6875,y:.8125},{x:.6875,y:.8125},{x:.6875,y:.8125},{x:.6875,y:.8125},{x:.6875,y:.8125},{x:.8125,y:.8125},{x:.8125,y:.8125},{x:.8125,y:.8125},{x:.8125,y:.8125},{x:.8125,y:.8125},{x:.8125,y:.8125},{x:.9375,y:.8125},{x:.9375,y:.8125},{x:.9375,y:.8125},{x:.9375,y:.8125},{x:.9375,y:.8125},{x:.9375,y:.8125},{x:.0625,y:.9375},{x:.0625,y:.9375},{x:.0625,y:.9375},{x:.0625,y:.9375},{x:.0625,y:.9375},{x:.0625,y:.9375},{x:.1875,y:.9375},{x:.1875,y:.9375},{x:.1875,y:.9375},{x:.1875,y:.9375},{x:.1875,y:.9375},{x:.1875,y:.9375},{x:.3125,y:.9375},{x:.3125,y:.9375},{x:.3125,y:.9375},{x:.3125,y:.9375},{x:.3125,y:.9375},{x:.3125,y:.9375},{x:.4375,y:.9375},{x:.4375,y:.9375},{x:.4375,y:.9375},{x:.4375,y:.9375},{x:.4375,y:.9375},{x:.4375,y:.9375},{x:.5625,y:.9375},{x:.5625,y:.9375},{x:.5625,y:.9375},{x:.5625,y:.9375},{x:.5625,y:.9375},{x:.5625,y:.9375},{x:.6875,y:.9375},{x:.6875,y:.9375},{x:.6875,y:.9375},{x:.6875,y:.9375},{x:.6875,y:.9375},{x:.6875,y:.9375},{x:.8125,y:.9375},{x:.8125,y:.9375},{x:.8125,y:.9375},{x:.8125,y:.9375},{x:.8125,y:.9375},{x:.8125,y:.9375},{x:.9375,y:.9375},{x:.9375,y:.9375},{x:.9375,y:.9375},{x:.9375,y:.9375},{x:.9375,y:.9375},{x:.9375,y:.9375}];var xb=class{constructor(t){ce(this,"model");ce(this,"anchors");ce(this,"anchorsTensor");ce(this,"inputSize");ce(this,"inputSizeTensor");ce(this,"doubleInputSizeTensor");this.model=t,this.anchors=W8.map(n=>[n.x,n.y]),this.anchorsTensor=lr(this.anchors),this.inputSize=this.model&&this.model.inputs&&this.model.inputs[0].shape?this.model.inputs[0].shape[2]:0,this.inputSizeTensor=Xt([this.inputSize,this.inputSize]),this.doubleInputSizeTensor=Xt([this.inputSize*2,this.inputSize*2])}normalizeBoxes(t){let n={};n.boxOffsets=De(t,[0,0],[-1,2]),n.boxSizes=De(t,[0,2],[-1,2]),n.div=pe(n.boxOffsets,this.inputSizeTensor),n.boxCenterPoints=le(n.div,this.anchorsTensor),n.halfBoxSizes=pe(n.boxSizes,this.doubleInputSizeTensor),n.sub=me(n.boxCenterPoints,n.halfBoxSizes),n.startPoints=L(n.sub,this.inputSizeTensor),n.add=le(n.boxCenterPoints,n.halfBoxSizes),n.endPoints=L(n.add,this.inputSizeTensor);let s=Pu([n.startPoints,n.endPoints],1);return Object.keys(n).forEach(r=>se(n[r])),s}normalizeLandmarks(t,n){let s={};s.reshape=G(t,[-1,7,2]),s.div=pe(s.reshape,this.inputSizeTensor),s.landmarks=le(s.div,this.anchors[n]);let r=L(s.landmarks,this.inputSizeTensor);return Object.keys(s).forEach(a=>se(s[a])),r}async predict(t,n){let s={};s.resize=$e.resizeBilinear(t,[this.inputSize,this.inputSize]),s.div=pe(s.resize,127.5),s.image=me(s.div,1),s.batched=this.model.execute(s.image),s.predictions=ot(s.batched),s.slice=De(s.predictions,[0,0],[-1,1]),s.sigmoid=cs(s.slice),s.scores=ot(s.sigmoid);let r=await s.scores.data();s.boxes=De(s.predictions,[0,1],[-1,4]),s.norm=this.normalizeBoxes(s.boxes),s.nms=await $e.nonMaxSuppressionAsync(s.norm,s.scores,3*n.hand.maxDetected,n.hand.iouThreshold,n.hand.minConfidence);let a=await s.nms.array(),o=[];for(let i of a){let l={};l.box=De(s.norm,[i,0],[1,-1]),l.slice=De(s.predictions,[i,5],[1,14]),l.norm=this.normalizeLandmarks(l.slice,i),l.palmLandmarks=G(l.norm,[-1,2]);let c=await l.box.data(),u=c.slice(0,2),d=c.slice(2,4),p=await l.palmLandmarks.array(),h={startPoint:u,endPoint:d,palmLandmarks:p,confidence:r[i]},f=O8(h,[t.shape[2]/this.inputSize,t.shape[1]/this.inputSize]);o.push(f),Object.keys(l).forEach(m=>se(l[m]))}return Object.keys(s).forEach(i=>se(s[i])),o}};var Vge=5,V8=1.65,U8=[0,5,9,13,17,1,2],Uge=0,Gge=2,G8=0,bb=class{constructor(t,n){ce(this,"handDetector");ce(this,"handPoseModel");ce(this,"inputSize");ce(this,"storedBoxes");ce(this,"skipped");ce(this,"detectedHands");this.handDetector=t,this.handPoseModel=n,this.inputSize=this.handPoseModel&&this.handPoseModel.inputs[0].shape?this.handPoseModel.inputs[0].shape[2]:0,this.storedBoxes=[],this.skipped=Number.MAX_SAFE_INTEGER,this.detectedHands=0}calculateLandmarksBoundingBox(t){let n=t.map(o=>o[0]),s=t.map(o=>o[1]),r=[Math.min(...n),Math.min(...s)],a=[Math.max(...n),Math.max(...s)];return{startPoint:r,endPoint:a}}getBoxForPalmLandmarks(t,n){let s=t.map(a=>yb([...a,1],n)),r=this.calculateLandmarksBoundingBox(s);return c0(d0(r),Vge)}getBoxForHandLandmarks(t){let n=this.calculateLandmarksBoundingBox(t),s=c0(d0(n),V8);s.palmLandmarks=[];for(let r=0;r<U8.length;r++)s.palmLandmarks.push(t[U8[r]].slice(0,2));return s}transformRawCoords(t,n,s,r){let a=u0(n),o=[a[0]/this.inputSize,a[1]/this.inputSize,(a[0]+a[1])/this.inputSize/2],i=t.map(h=>[o[0]*(h[0]-this.inputSize/2),o[1]*(h[1]-this.inputSize/2),o[2]*h[2]]),l=Ab(s,[0,0]),c=i.map(h=>[...yb(h,l),h[2]]),u=B8(r),d=[..._p(n),1],p=[Yo(d,u[0]),Yo(d,u[1])];return c.map(h=>[Math.trunc(h[0]+p[0]),Math.trunc(h[1]+p[1]),Math.trunc(h[2])])}async estimateHands(t,n){let s=!1,r,a=(n.hand.skipTime||0)>he()-G8,o=this.skipped<(n.hand.skipFrames||0);n.skipAllowed&&a&&o&&(r=await this.handDetector.predict(t,n),this.skipped=0),n.skipAllowed&&this.skipped++,r&&r.length>0&&(r.length!==this.detectedHands&&this.detectedHands!==n.hand.maxDetected||!n.hand.landmarks)&&(this.detectedHands=0,this.storedBoxes=[...r],this.storedBoxes.length>0&&(s=!0));let i=[];for(let l=0;l<this.storedBoxes.length;l++){let c=this.storedBoxes[l];if(!!c)if(n.hand.landmarks){let u=n.hand.rotation?M8(c.palmLandmarks[Uge],c.palmLandmarks[Gge]):0,d=_p(c),p=[d[0]/t.shape[2],d[1]/t.shape[1]],h=n.hand.rotation&&Ae.kernels.includes("rotatewithoffset")?$e.rotateWithOffset(t,u,0,p):t.clone(),f=Ab(-u,d),m=s?this.getBoxForPalmLandmarks(c.palmLandmarks,f):c,g=F8(m,h,[this.inputSize,this.inputSize]),A=pe(g,255);se(g),se(h);let[x,y]=this.handPoseModel.execute(A);G8=he(),se(A);let b=(await x.data())[0];if(se(x),b>=n.hand.minConfidence/4){let w=G(y,[-1,3]),k=await w.array();se(y),se(w);let C=this.transformRawCoords(k,m,u,f),N=this.getBoxForHandLandmarks(C);this.storedBoxes[l]={...N,confidence:b};let R={landmarks:C,confidence:b,boxConfidence:c.confidence,fingerConfidence:b,box:{topLeft:N.startPoint,bottomRight:N.endPoint}};i.push(R)}else this.storedBoxes[l]=null;se(y)}else{let u=c0(d0(c),V8),d={confidence:c.confidence,boxConfidence:c.confidence,fingerConfidence:0,box:{topLeft:u.startPoint,bottomRight:u.endPoint},landmarks:[]};i.push(d)}}return this.storedBoxes=this.storedBoxes.filter(l=>l!==null),this.detectedHands=i.length,i.length>n.hand.maxDetected&&(i.length=n.hand.maxDetected),i}};var Yn={thumb:0,index:1,middle:2,ring:3,pinky:4,all:[0,1,2,3,4],nameMapping:{0:"thumb",1:"index",2:"middle",3:"ring",4:"pinky"},pointsMapping:{0:[[0,1],[1,2],[2,3],[3,4]],1:[[0,5],[5,6],[6,7],[7,8]],2:[[0,9],[9,10],[10,11],[11,12]],3:[[0,13],[13,14],[14,15],[15,16]],4:[[0,17],[17,18],[18,19],[19,20]]},getName:e=>Yn.nameMapping[e],getPoints:e=>Yn.pointsMapping[e]},Jo={none:0,half:1,full:2,nameMapping:{0:"none",1:"half",2:"full"},getName:e=>Jo.nameMapping[e]},Mt={verticalUp:0,verticalDown:1,horizontalLeft:2,horizontalRight:3,diagonalUpRight:4,diagonalUpLeft:5,diagonalDownRight:6,diagonalDownLeft:7,nameMapping:{0:"verticalUp",1:"verticalDown",2:"horizontalLeft",3:"horizontalRight",4:"diagonalUpRight",5:"diagonalUpLeft",6:"diagonalDownRight",7:"diagonalDownLeft"},getName:e=>Mt.nameMapping[e]},Fl=class{constructor(t){ce(this,"name");ce(this,"curls");ce(this,"directions");ce(this,"weights");ce(this,"weightsRelative");this.name=t,this.curls={},this.directions={},this.weights=[1,1,1,1,1],this.weightsRelative=[1,1,1,1,1]}curl(t,n,s){typeof this.curls[t]=="undefined"&&(this.curls[t]=[]),this.curls[t].push([n,s])}direction(t,n,s){this.directions[t]||(this.directions[t]=[]),this.directions[t].push([n,s])}weight(t,n){this.weights[t]=n;let s=this.weights.reduce((r,a)=>r+a,0);this.weightsRelative=this.weights.map(r=>r*5/s)}matchAgainst(t,n){let s=0;for(let r in t){let a=t[r],o=this.curls[r];if(typeof o=="undefined"){s+=this.weightsRelative[r];continue}for(let[i,l]of o)if(a===i){s+=l*this.weightsRelative[r];break}}for(let r in n){let a=n[r],o=this.directions[r];if(typeof o=="undefined"){s+=this.weightsRelative[r];continue}for(let[i,l]of o)if(a===i){s+=l*this.weightsRelative[r];break}}return s/10}};var{thumb:yr,index:la,middle:ua,ring:Ol,pinky:Ml}=Yn,{none:xr,half:Hge,full:br}=Jo,{verticalUp:Cc,verticalDown:uAe,horizontalLeft:vb,horizontalRight:jge,diagonalUpRight:qge,diagonalUpLeft:Tc,diagonalDownRight:cAe,diagonalDownLeft:dAe}=Mt,Qo=new Fl("thumbs up");Qo.curl(yr,xr,1);Qo.direction(yr,Cc,1);Qo.direction(yr,Tc,.25);Qo.direction(yr,qge,.25);for(let e of[Yn.index,Yn.middle,Yn.ring,Yn.pinky])Qo.curl(e,br,1),Qo.direction(e,vb,1),Qo.direction(e,jge,1);var Jt=new Fl("victory");Jt.curl(yr,Hge,.5);Jt.curl(yr,xr,.5);Jt.direction(yr,Cc,1);Jt.direction(yr,Tc,1);Jt.curl(la,xr,1);Jt.direction(la,Cc,.75);Jt.direction(la,Tc,1);Jt.curl(ua,xr,1);Jt.direction(ua,Cc,1);Jt.direction(ua,Tc,.75);Jt.curl(Ol,br,1);Jt.direction(Ol,Cc,.2);Jt.direction(Ol,Tc,1);Jt.direction(Ol,vb,.2);Jt.curl(Ml,br,1);Jt.direction(Ml,Cc,.2);Jt.direction(Ml,Tc,1);Jt.direction(Ml,vb,.2);Jt.weight(la,2);Jt.weight(ua,2);var ei=new Fl("point");ei.curl(yr,br,1);ei.curl(la,xr,.5);ei.curl(ua,br,.5);ei.curl(Ol,br,.5);ei.curl(Ml,br,.5);ei.weight(la,2);ei.weight(ua,2);var ti=new Fl("middle finger");ti.curl(yr,xr,1);ti.curl(la,br,.5);ti.curl(ua,br,.5);ti.curl(Ol,br,.5);ti.curl(Ml,br,.5);ti.weight(la,2);ti.weight(ua,2);var Nc=new Fl("open palm");Nc.curl(yr,xr,.75);Nc.curl(la,xr,.75);Nc.curl(ua,xr,.75);Nc.curl(Ol,xr,.75);Nc.curl(Ml,xr,.75);var H8=[Qo,Jt,ei,ti,Nc];var Xge=.7,zl={HALF_CURL_START_LIMIT:60,NO_CURL_START_LIMIT:130,DISTANCE_VOTE_POWER:1.1,SINGLE_ANGLE_VOTE_POWER:.9,TOTAL_ANGLE_VOTE_POWER:1.6};function j8(e,t,n,s){let r=(t-s)/(e-n),a=Math.atan(r)*180/Math.PI;return a<=0?a=-a:a>0&&(a=180-a),a}function q8(e,t){if(!e||!t)return[0,0];let n=j8(e[0],e[1],t[0],t[1]);if(e.length===2)return n;let s=j8(e[1],e[2],t[1],t[2]);return[n,s]}function X8(e,t=1){let n=0,s=0,r=0;return e>=75&&e<=105?n=1*t:e>=25&&e<=155?s=1*t:r=1*t,[n,s,r]}function Kge(e,t,n){let s=e[0]-t[0],r=e[0]-n[0],a=t[0]-n[0],o=e[1]-t[1],i=e[1]-n[1],l=t[1]-n[1],c=e[2]-t[2],u=e[2]-n[2],d=t[2]-n[2],p=Math.sqrt(s*s+o*o+c*c),h=Math.sqrt(r*r+i*i+u*u),f=Math.sqrt(a*a+l*l+d*d),m=(f*f+p*p-h*h)/(2*f*p);m>1?m=1:m<-1&&(m=-1);let g=Math.acos(m);g=57.2958*g%180;let A;return g>zl.NO_CURL_START_LIMIT?A=Jo.none:g>zl.HALF_CURL_START_LIMIT?A=Jo.half:A=Jo.full,A}function K8(e,t,n,s){let r;return s===Math.abs(e)?e>0?r=Mt.horizontalLeft:r=Mt.horizontalRight:s===Math.abs(t)?t>0?r=Mt.horizontalLeft:r=Mt.horizontalRight:n>0?r=Mt.horizontalLeft:r=Mt.horizontalRight,r}function Z8(e,t,n,s){let r;return s===Math.abs(e)?e<0?r=Mt.verticalDown:r=Mt.verticalUp:s===Math.abs(t)?t<0?r=Mt.verticalDown:r=Mt.verticalUp:n<0?r=Mt.verticalDown:r=Mt.verticalUp,r}function Zge(e,t,n,s,r,a,o,i){let l,c=Z8(e,t,n,s),u=K8(r,a,o,i);return c===Mt.verticalUp?u===Mt.horizontalLeft?l=Mt.diagonalUpLeft:l=Mt.diagonalUpRight:u===Mt.horizontalLeft?l=Mt.diagonalDownLeft:l=Mt.diagonalDownRight,l}function Yge(e,t,n,s){let r=e[0]-t[0],a=e[0]-n[0],o=t[0]-n[0],i=e[1]-t[1],l=e[1]-n[1],c=t[1]-n[1],u=Math.max(Math.abs(r),Math.abs(a),Math.abs(o)),d=Math.max(Math.abs(i),Math.abs(l),Math.abs(c)),p=0,h=0,f=0,m=d/(u+1e-5);m>1.5?p+=zl.DISTANCE_VOTE_POWER:m>.66?h+=zl.DISTANCE_VOTE_POWER:f+=zl.DISTANCE_VOTE_POWER;let g=Math.sqrt(r*r+i*i),A=Math.sqrt(a*a+l*l),x=Math.sqrt(o*o+c*c),y=Math.max(g,A,x),b=e[0],w=e[1],k=n[0],C=n[1];y===g?(k=n[0],C=n[1]):y===x&&(b=t[0],w=t[1]);let F=q8([b,w],[k,C]),_=X8(F,zl.TOTAL_ANGLE_VOTE_POWER);p+=_[0],h+=_[1],f+=_[2];for(let T of s){let M=X8(T,zl.SINGLE_ANGLE_VOTE_POWER);p+=M[0],h+=M[1],f+=M[2]}let P;return p===Math.max(p,h,f)?P=Z8(l,i,c,d):f===Math.max(h,f)?P=K8(a,r,o,u):P=Zge(l,i,c,d,a,r,o,u),P}function Y8(e){let t=[],n=[],s=[],r=[];if(!e)return{curls:s,directions:r};for(let a of Yn.all){let o=Yn.getPoints(a),i=[],l=[];for(let c of o){let u=e[c[0]],d=e[c[1]],p=q8(u,d),h=p[0],f=p[1];i.push(h),l.push(f)}t.push(i),n.push(l)}for(let a of Yn.all){let o=a===Yn.thumb?1:0,i=Yn.getPoints(a),l=e[i[o][0]],c=e[i[o+1][1]],u=e[i[3][1]],d=Kge(l,c,u),p=Yge(l,c,u,t[a].slice(o));s[a]=d,r[a]=p}return{curls:s,directions:r}}function p0(e){if(!e||e.length===0)return null;let t=Y8(e),n={};for(let s of Yn.all)n[Yn.getName(s)]={curl:Jo.getName(t.curls[s]),direction:Mt.getName(t.directions[s])};return n}function J8(e){let t=[];if(!e||e.length===0)return t;let n=Y8(e);for(let s of H8){let r=s.matchAgainst(n.curls,n.directions);r>=Xge&&t.push({name:s.name,confidence:r})}return t}var Q8={thumb:[1,2,3,4],index:[5,6,7,8],middle:[9,10,11,12],ring:[13,14,15,16],pinky:[17,18,19,20],palm:[0]},ca,da,eT;async function wb(e,t){let n=await eT.estimateHands(e,t);if(!n)return[];let s=[];for(let r=0;r<n.length;r++){let a={};if(n[r].landmarks)for(let u of Object.keys(Q8))a[u]=Q8[u].map(d=>n[r].landmarks[d]);let o=n[r].landmarks,i=[Number.MAX_SAFE_INTEGER,Number.MAX_SAFE_INTEGER,0,0],l=[0,0,0,0];if(o&&o.length>0){for(let u of o)u[0]<i[0]&&(i[0]=u[0]),u[1]<i[1]&&(i[1]=u[1]),u[0]>i[2]&&(i[2]=u[0]),u[1]>i[3]&&(i[3]=u[1]);i[2]-=i[0],i[3]-=i[1],l=[i[0]/(e.shape[2]||0),i[1]/(e.shape[1]||0),i[2]/(e.shape[2]||0),i[3]/(e.shape[1]||0)]}else i=n[r].box?[Math.trunc(Math.max(0,n[r].box.topLeft[0])),Math.trunc(Math.max(0,n[r].box.topLeft[1])),Math.trunc(Math.min(e.shape[2]||0,n[r].box.bottomRight[0])-Math.max(0,n[r].box.topLeft[0])),Math.trunc(Math.min(e.shape[1]||0,n[r].box.bottomRight[1])-Math.max(0,n[r].box.topLeft[1]))]:[0,0,0,0],l=[n[r].box.topLeft[0]/(e.shape[2]||0),n[r].box.topLeft[1]/(e.shape[1]||0),(n[r].box.bottomRight[0]-n[r].box.topLeft[0])/(e.shape[2]||0),(n[r].box.bottomRight[1]-n[r].box.topLeft[1])/(e.shape[1]||0)];let c=p0(o);s.push({id:r,score:Math.round(100*n[r].confidence)/100,boxScore:Math.round(100*n[r].boxConfidence)/100,fingerScore:Math.round(100*n[r].fingerConfidence)/100,label:"hand",box:i,boxRaw:l,keypoints:o,annotations:a,landmarks:c})}return s}async function kb(e){var n,s,r,a,o,i;Ae.initial&&(ca=null,da=null),!ca||!da?([ca,da]=await Promise.all([e.hand.enabled?et(nt(e.modelBasePath,((n=e.hand.detector)==null?void 0:n.modelPath)||""),{fromTFHub:(((s=e.hand.detector)==null?void 0:s.modelPath)||"").includes("tfhub.dev")}):null,e.hand.landmarks?et(nt(e.modelBasePath,((r=e.hand.skeleton)==null?void 0:r.modelPath)||""),{fromTFHub:(((a=e.hand.skeleton)==null?void 0:a.modelPath)||"").includes("tfhub.dev")}):null]),e.hand.enabled&&(!ca||!ca.modelUrl?re("load model failed:",((o=e.hand.detector)==null?void 0:o.modelPath)||""):e.debug&&re("load model:",ca.modelUrl),!da||!da.modelUrl?re("load model failed:",((i=e.hand.skeleton)==null?void 0:i.modelPath)||""):e.debug&&re("load model:",da.modelUrl))):(e.debug&&re("cached model:",ca.modelUrl),e.debug&&re("cached model:",da.modelUrl));let t=new xb(ca);return eT=new bb(t,da),[ca,da]}function Ll(e,t=[1,1]){let n=[e.map(i=>i[0]),e.map(i=>i[1])],s=[Math.min(...n[0]),Math.min(...n[1])],r=[Math.max(...n[0]),Math.max(...n[1])],a=[s[0],s[1],r[0]-s[0],r[1]-s[1]],o=[a[0]/t[0],a[1]/t[1],a[2]/t[0],a[3]/t[1]];return{box:a,boxRaw:o}}function tT(e,t=[1,1]){let n=[e.map(c=>c[0]),e.map(c=>c[1])],s=[Math.min(...n[0]),Math.min(...n[1])],r=[Math.max(...n[0]),Math.max(...n[1])],a=[(s[0]+r[0])/2,(s[1]+r[1])/2],o=Math.max(a[0]-s[0],a[1]-s[1],-a[0]+r[0],-a[1]+r[1]),i=[Math.trunc(a[0]-o),Math.trunc(a[1]-o),Math.trunc(2*o),Math.trunc(2*o)],l=[i[0]/t[0],i[1]/t[1],i[2]/t[0],i[3]/t[1]];return{box:i,boxRaw:l}}function h0(e,t){let n=[e[2]*t,e[3]*t];return[e[0]-(n[0]-e[2])/2,e[1]-(n[1]-e[3])/2,n[0],n[1]]}function Sb(e){return[Math.max(0,e[1]),Math.max(0,e[0]),Math.min(1,e[3]+e[1]),Math.min(1,e[2]+e[0])]}var St=[null,null],Jge=["StatefulPartitionedCall/Postprocessor/Slice","StatefulPartitionedCall/Postprocessor/ExpandDims_1"],ni=[[0,0],[0,0]],Qge=["hand","fist","pinch","point","face","tip","pinchtip"],nT=4,sT=1.6,e2e=512,t2e=1.4,f0=Number.MAX_SAFE_INTEGER,Ib=0,pa=[0,0],jt={boxes:[],hands:[]},rT={thumb:[1,2,3,4],index:[5,6,7,8],middle:[9,10,11,12],ring:[13,14,15,16],pinky:[17,18,19,20],palm:[0]};async function aT(e){var t,n;if(Ae.initial&&(St[0]=null),St[0])e.debug&&re("cached model:",St[0].modelUrl);else{m0(["tensorlistreserve","enter","tensorlistfromtensor","merge","loopcond","switch","exit","tensorliststack","nextiteration","tensorlistsetitem","tensorlistgetitem","reciprocal","shape","split","where"],e),St[0]=await et(nt(e.modelBasePath,((t=e.hand.detector)==null?void 0:t.modelPath)||""));let s=Object.values(St[0].modelSignature.inputs);ni[0][0]=Array.isArray(s)?parseInt(s[0].tensorShape.dim[1].size):0,ni[0][1]=Array.isArray(s)?parseInt(s[0].tensorShape.dim[2].size):0,!St[0]||!St[0].modelUrl?re("load model failed:",(n=e.hand.detector)==null?void 0:n.modelPath):e.debug&&re("load model:",St[0].modelUrl)}return St[0]}async function oT(e){var t,n;if(Ae.initial&&(St[1]=null),St[1])e.debug&&re("cached model:",St[1].modelUrl);else{St[1]=await et(nt(e.modelBasePath,((t=e.hand.skeleton)==null?void 0:t.modelPath)||""));let s=Object.values(St[1].modelSignature.inputs);ni[1][0]=Array.isArray(s)?parseInt(s[0].tensorShape.dim[1].size):0,ni[1][1]=Array.isArray(s)?parseInt(s[0].tensorShape.dim[2].size):0,!St[1]||!St[1].modelUrl?re("load model failed:",(n=e.hand.skeleton)==null?void 0:n.modelPath):e.debug&&re("load model:",St[1].modelUrl)}return St[1]}async function n2e(e,t){let n=[];if(!e||!St[0])return n;let s={},r=(e.shape[2]||1)/(e.shape[1]||1),a=Math.min(Math.round((e.shape[1]||0)/8)*8,e2e),o=Math.round(a*r/8)*8;s.resize=$e.resizeBilinear(e,[a,o]),s.cast=de(s.resize,"int32"),[s.rawScores,s.rawBoxes]=await St[0].executeAsync(s.cast,Jge),s.boxes=ot(s.rawBoxes,[0,2]),s.scores=ot(s.rawScores,[0]);let i=ss(s.scores,1);se(i[nT]),i.splice(nT,1),s.filtered=An(i,1),se(i),s.max=gn(s.filtered,1),s.argmax=Hs(s.filtered,1);let l=0;s.nms=await $e.nonMaxSuppressionAsync(s.boxes,s.max,t.hand.maxDetected,t.hand.iouThreshold,t.hand.minConfidence);let c=await s.nms.data(),u=await s.max.data(),d=await s.argmax.data();for(let p of Array.from(c)){let h=De(s.boxes,p,1),f=await h.data();se(h);let m=[f[1],f[0],f[3]-f[1],f[2]-f[0]],g=h0(m,t2e),A=Sb(g),x=[Math.trunc(m[0]*pa[0]),Math.trunc(m[1]*pa[1]),Math.trunc(m[2]*pa[0]),Math.trunc(m[3]*pa[1])],y=u[p],b=Qge[d[p]],w={id:l++,score:y,box:x,boxRaw:g,boxCrop:A,label:b};n.push(w)}return Object.keys(s).forEach(p=>se(s[p])),n.sort((p,h)=>h.score-p.score),n.length>(t.hand.maxDetected||1)&&(n.length=t.hand.maxDetected||1),n}async function Cb(e,t,n){let s={id:t.id,score:Math.round(100*t.score)/100,boxScore:Math.round(100*t.score)/100,fingerScore:0,box:t.box,boxRaw:t.boxRaw,label:t.label,keypoints:[],landmarks:{},annotations:{}};if(e&&St[1]&&n.hand.landmarks&&t.score>(n.hand.minConfidence||0)){let r={};r.crop=$e.cropAndResize(e,[t.boxCrop],[0],[ni[1][0],ni[1][1]],"bilinear"),r.cast=de(r.crop,"float32"),r.div=pe(r.cast,255),[r.score,r.keypoints]=St[1].execute(r.div,["Identity_1","Identity"]);let a=(await r.score.data())[0],o=(100-Math.trunc(100/(1+Math.exp(a))))/100;if(o>=(n.hand.minConfidence||0)){s.fingerScore=o,r.reshaped=G(r.keypoints,[-1,3]);let c=(await r.reshaped.array()).map(u=>[u[0]/ni[1][1],u[1]/ni[1][0],u[2]||0]).map(u=>[u[0]*t.boxRaw[2],u[1]*t.boxRaw[3],u[2]||0]);s.keypoints=c.map(u=>[pa[0]*(u[0]+t.boxRaw[0]),pa[1]*(u[1]+t.boxRaw[1]),u[2]||0]),s.landmarks=p0(s.keypoints);for(let u of Object.keys(rT))s.annotations[u]=rT[u].map(d=>s.landmarks&&s.keypoints[d]?s.keypoints[d]:null)}Object.keys(r).forEach(i=>se(r[i]))}return s}async function Tb(e,t){var r,a;if(!St[0]||!St[1]||!((r=St[0])==null?void 0:r.inputs[0].shape)||!((a=St[1])==null?void 0:a.inputs[0].shape))return[];pa=[e.shape[2]||0,e.shape[1]||0],f0++;let n=(t.hand.skipTime||0)>he()-Ib,s=f0<(t.hand.skipFrames||0);return t.skipAllowed&&n&&s?jt.hands:new Promise(async o=>{let i=3*(t.hand.skipTime||0)>he()-Ib,l=f0<3*(t.hand.skipFrames||0);t.skipAllowed&&jt.hands.length===t.hand.maxDetected?jt.hands=await Promise.all(jt.boxes.map(u=>Cb(e,u,t))):t.skipAllowed&&i&&l&&jt.hands.length>0?jt.hands=await Promise.all(jt.boxes.map(u=>Cb(e,u,t))):(jt.boxes=await n2e(e,t),Ib=he(),jt.hands=await Promise.all(jt.boxes.map(u=>Cb(e,u,t))),f0=0);let c=[...jt.boxes];if(jt.boxes.length=0,t.cacheSensitivity>0)for(let u=0;u<jt.hands.length;u++){let d=tT(jt.hands[u].keypoints,pa);if(d.box[2]/(e.shape[2]||1)>.05&&d.box[3]/(e.shape[1]||1)>.05&&jt.hands[u].fingerScore&&jt.hands[u].fingerScore>(t.hand.minConfidence||0)){let p=h0(d.box,sT),h=h0(d.boxRaw,sT),f=Sb(h);jt.boxes.push({...c[u],box:p,boxRaw:h,boxCrop:f})}}for(let u=0;u<jt.hands.length;u++){let d=Ll(jt.hands[u].keypoints,pa);jt.hands[u].box=d.box,jt.hands[u].boxRaw=d.boxRaw}o(jt.hands)})}var $b={};Gc($b,{connected:()=>A0,horizontal:()=>Nb,kpt:()=>g0,relative:()=>Rb,vertical:()=>Eb});var g0=["nose","leftEye","rightEye","leftEar","rightEar","leftShoulder","rightShoulder","leftElbow","rightElbow","leftWrist","rightWrist","leftHip","rightHip","leftKnee","rightKnee","leftAnkle","rightAnkle"],Nb=[["leftEye","rightEye"],["leftEar","rightEar"],["leftShoulder","rightShoulder"],["leftElbow","rightElbow"],["leftWrist","rightWrist"],["leftHip","rightHip"],["leftKnee","rightKnee"],["leftAnkle","rightAnkle"]],Eb=[["leftKnee","leftShoulder"],["rightKnee","rightShoulder"],["leftAnkle","leftKnee"],["rightAnkle","rightKnee"]],Rb=[[["leftHip","rightHip"],["leftShoulder","rightShoulder"]],[["leftElbow","rightElbow"],["leftShoulder","rightShoulder"]]],A0={leftLeg:["leftHip","leftKnee","leftAnkle"],rightLeg:["rightHip","rightKnee","rightAnkle"],torso:["leftShoulder","rightShoulder","rightHip","leftHip","leftShoulder"],leftArm:["leftShoulder","leftElbow","leftWrist"],rightArm:["rightShoulder","rightElbow","rightWrist"],head:[]};var iT=.005,xs={keypoints:[],padding:[[0,0],[0,0],[0,0],[0,0]]};function _b(e){for(let t of Nb){let n=e.keypoints.findIndex(r=>r.part===t[0]),s=e.keypoints.findIndex(r=>r.part===t[1]);if(e.keypoints[n]&&e.keypoints[s]&&e.keypoints[n].position[0]<e.keypoints[s].position[0]){let r=e.keypoints[n];e.keypoints[n]=e.keypoints[s],e.keypoints[s]=r}}for(let t of Eb){let n=e.keypoints.findIndex(r=>r&&r.part===t[0]),s=e.keypoints.findIndex(r=>r&&r.part===t[1]);e.keypoints[n]&&e.keypoints[s]&&e.keypoints[n].position[1]<e.keypoints[s].position[1]&&e.keypoints.splice(n,1)}for(let[t,n]of Rb){let s=e.keypoints.findIndex(c=>c&&c.part===t[0]),r=e.keypoints.findIndex(c=>c&&c.part===t[1]),a=e.keypoints.findIndex(c=>c&&c.part===n[0]),o=e.keypoints.findIndex(c=>c&&c.part===n[1]);if(!e.keypoints[a]||!e.keypoints[o])continue;let i=e.keypoints[s]?[Math.abs(e.keypoints[a].position[0]-e.keypoints[s].position[0]),Math.abs(e.keypoints[o].position[0]-e.keypoints[s].position[0])]:[0,0],l=e.keypoints[r]?[Math.abs(e.keypoints[o].position[0]-e.keypoints[r].position[0]),Math.abs(e.keypoints[a].position[0]-e.keypoints[r].position[0])]:[0,0];if(i[0]>i[1]||l[0]>l[1]){let c=e.keypoints[s];e.keypoints[s]=e.keypoints[r],e.keypoints[r]=c}}}function lT(e){for(let t=0;t<e.length;t++)if(e[t]&&xs.keypoints[t]){let n=[Math.abs(e[t].positionRaw[0]-xs.keypoints[t].positionRaw[0]),Math.abs(e[t].positionRaw[1]-xs.keypoints[t].positionRaw[1])];n[0]<iT&&n[1]<iT?e[t]=xs.keypoints[t]:xs.keypoints[t]=e[t]}else xs.keypoints[t]=e[t];return e}function uT(e,t){let n={};if(!e.shape||!e.shape[1]||!e.shape[2])return e;xs.padding=[[0,0],[e.shape[2]>e.shape[1]?Math.trunc((e.shape[2]-e.shape[1])/2):0,e.shape[2]>e.shape[1]?Math.trunc((e.shape[2]-e.shape[1])/2):0],[e.shape[1]>e.shape[2]?Math.trunc((e.shape[1]-e.shape[2])/2):0,e.shape[1]>e.shape[2]?Math.trunc((e.shape[1]-e.shape[2])/2):0],[0,0]],n.pad=qs(e,xs.padding),n.resize=$e.resizeBilinear(n.pad,[t,t]);let s=de(n.resize,"int32");return Object.keys(n).forEach(r=>se(n[r])),s}function cT(e,t){e.keypoints=e.keypoints.filter(s=>s&&s.position);for(let s of e.keypoints)s.position=[s.position[0]*(t[0]+xs.padding[2][0]+xs.padding[2][1])/t[0]-xs.padding[2][0],s.position[1]*(t[1]+xs.padding[1][0]+xs.padding[1][1])/t[1]-xs.padding[1][0]],s.positionRaw=[s.position[0]/t[0],s.position[1]/t[1]];let n=Ll(e.keypoints.map(s=>s.position),t);return e.box=n.box,e.boxRaw=n.boxRaw,e}var Dn,y0=0,Db=Number.MAX_SAFE_INTEGER,Bl={boxes:[],bodies:[],last:0};async function dT(e){return Ae.initial&&(Dn=null),Dn?e.debug&&re("cached model:",Dn.modelUrl):(m0(["size"],e),Dn=await et(nt(e.modelBasePath,e.body.modelPath||"")),!Dn||!Dn.modelUrl?re("load model failed:",e.body.modelPath):e.debug&&re("load model:",Dn.modelUrl)),y0=Dn.inputs[0].shape?Dn.inputs[0].shape[2]:0,y0===-1&&(y0=256),Dn}async function s2e(e,t,n,s){let r=e[0][0],a=[],o=0;for(let d=0;d<r.length;d++)if(o=r[d][2],o>t.body.minConfidence){let p=[(s[3]-s[1])*r[d][1]+s[1],(s[2]-s[0])*r[d][0]+s[0]];a.push({score:Math.round(100*o)/100,part:g0[d],positionRaw:p,position:[Math.round((n.shape[2]||0)*p[0]),Math.round((n.shape[1]||0)*p[1])]})}o=a.reduce((d,p)=>p.score>d?p.score:d,0);let i=[],l=Ll(a.map(d=>d.position),[n.shape[2],n.shape[1]]),c={};for(let[d,p]of Object.entries(A0)){let h=[];for(let f=0;f<p.length-1;f++){let m=a.find(A=>A.part===p[f]),g=a.find(A=>A.part===p[f+1]);m&&g&&m.score>(t.body.minConfidence||0)&&g.score>(t.body.minConfidence||0)&&h.push([m.position,g.position])}c[d]=h}let u={id:0,score:o,box:l.box,boxRaw:l.boxRaw,keypoints:a,annotations:c};return _b(u),i.push(u),i}async function r2e(e,t,n,s){let r=[];for(let a=0;a<e[0].length;a++){let o=e[0][a],i=Math.round(100*o[51+4])/100;if(i>t.body.minConfidence){let l=[];for(let p=0;p<17;p++){let h=o[3*p+2];if(h>t.body.minConfidence){let f=[(s[3]-s[1])*o[3*p+1]+s[1],(s[2]-s[0])*o[3*p+0]+s[0]];l.push({part:g0[p],score:Math.round(100*h)/100,positionRaw:f,position:[Math.round((n.shape[2]||0)*f[0]),Math.round((n.shape[1]||0)*f[1])]})}}let c=Ll(l.map(p=>p.position),[n.shape[2],n.shape[1]]),u={};for(let[p,h]of Object.entries(A0)){let f=[];for(let m=0;m<h.length-1;m++){let g=l.find(x=>x.part===h[m]),A=l.find(x=>x.part===h[m+1]);g&&A&&g.score>(t.body.minConfidence||0)&&A.score>(t.body.minConfidence||0)&&f.push([g.position,A.position])}u[p]=f}let d={id:a,score:i,box:c.box,boxRaw:c.boxRaw,keypoints:[...l],annotations:u};_b(d),r.push(d)}}return r.sort((a,o)=>o.score-a.score),r.length>t.body.maxDetected&&(r.length=t.body.maxDetected),r}async function Pb(e,t){if(!Dn||!(Dn==null?void 0:Dn.inputs[0].shape))return[];t.skipAllowed||(Bl.boxes.length=0),Db++;let n=(t.body.skipTime||0)>he()-Bl.last,s=Db<(t.body.skipFrames||0);return t.skipAllowed&&n&&s?Bl.bodies:new Promise(async r=>{let a={};Db=0,a.input=uT(e,y0),a.res=Dn==null?void 0:Dn.execute(a.input),Bl.last=he();let o=await a.res.array();Bl.bodies=a.res.shape[2]===17?await s2e(o,t,e,[0,0,1,1]):await r2e(o,t,e,[0,0,1,1]);for(let i of Bl.bodies)cT(i,[e.shape[2]||1,e.shape[1]||1]),lT(i.keypoints);Object.keys(a).forEach(i=>se(a[i])),r(Bl.bodies)})}var bs,x0=[],pT=0,Fb=Number.MAX_SAFE_INTEGER,b0=2.5;async function hT(e){if(!bs||Ae.initial){bs=await et(nt(e.modelBasePath,e.object.modelPath||""));let t=Object.values(bs.modelSignature.inputs);if(bs.inputSize=Array.isArray(t)?parseInt(t[0].tensorShape.dim[2].size):null,!bs.inputSize)throw new Error(`cannot determine model inputSize: ${e.object.modelPath}`);!bs||!bs.modelUrl?re("load model failed:",e.object.modelPath):e.debug&&re("load model:",bs.modelUrl)}else e.debug&&re("cached model:",bs.modelUrl);return bs}async function a2e(e,t,n,s){let r=0,a=[];for(let c of[1,2,4])q(async()=>{var g,A;let u=c*13,d=(g=e.find(x=>x.shape[1]===u**2&&x.shape[2]===wc.length))==null?void 0:g.squeeze(),p=(A=e.find(x=>x.shape[1]===u**2&&x.shape[2]<wc.length))==null?void 0:A.squeeze(),f=await p.reshape([-1,4,p.shape[1]/4]).argMax(2).array(),m=await d.array();for(let x=0;x<d.shape[0];x++)for(let y=0;y<d.shape[1];y++){let b=m[x][y];if(b>s.object.minConfidence&&y!==61){let w=(.5+Math.trunc(x%u))/u,k=(.5+Math.trunc(x/u))/u,C=f[x].map(U=>U*(u/c/t)),[N,R]=[w-b0/c*C[0],k-b0/c*C[1]],[F,_]=[w+b0/c*C[2]-N,k+b0/c*C[3]-R],P=[N,R,F,_];P=P.map(U=>Math.max(0,Math.min(U,1)));let T=[P[0]*n[0],P[1]*n[1],P[2]*n[0],P[3]*n[1]],M={id:r++,score:Math.round(100*b)/100,class:y+1,label:wc[y].label,box:T.map(U=>Math.trunc(U)),boxRaw:P};a.push(M)}}});e.forEach(c=>se(c));let o=a.map(c=>[c.boxRaw[1],c.boxRaw[0],c.boxRaw[3],c.boxRaw[2]]),i=a.map(c=>c.score),l=[];if(o&&o.length>0){let c=await $e.nonMaxSuppressionAsync(o,i,s.object.maxDetected,s.object.iouThreshold,s.object.minConfidence);l=await c.data(),se(c)}return a=a.filter((c,u)=>l.includes(u)).sort((c,u)=>u.score-c.score),a}async function Ob(e,t){let n=(t.object.skipTime||0)>he()-pT,s=Fb<(t.object.skipFrames||0);return t.skipAllowed&&n&&s&&x0.length>0?(Fb++,x0):(Fb=0,!Ae.kernels.includes("mod")||!Ae.kernels.includes("sparsetodense")?x0:new Promise(async r=>{let a=[e.shape[2],e.shape[1]],o=$e.resizeBilinear(e,[bs.inputSize,bs.inputSize],!1),i=pe(o,255),l=i.transpose([0,3,1,2]);se(i),se(o);let c;t.object.enabled&&(c=bs.execute(l)),pT=he(),se(l);let u=await a2e(c,bs.inputSize,a,t);x0=u,r(u)}))}var Dp=["nose","leftEye","rightEye","leftEar","rightEar","leftShoulder","rightShoulder","leftElbow","rightElbow","leftWrist","rightWrist","leftHip","rightHip","leftKnee","rightKnee","leftAnkle","rightAnkle"],o2e=Dp.length,Pp=Dp.reduce((e,t,n)=>(e[t]=n,e),{}),i2e=[["leftHip","leftShoulder"],["leftElbow","leftShoulder"],["leftElbow","leftWrist"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["rightHip","rightShoulder"],["rightElbow","rightShoulder"],["rightElbow","rightWrist"],["rightHip","rightKnee"],["rightKnee","rightAnkle"],["leftShoulder","rightShoulder"],["leftHip","rightHip"]],XAe=i2e.map(([e,t])=>[Pp[e],Pp[t]]),fT=[["nose","leftEye"],["leftEye","leftEar"],["nose","rightEye"],["rightEye","rightEar"],["nose","leftShoulder"],["leftShoulder","leftElbow"],["leftElbow","leftWrist"],["leftShoulder","leftHip"],["leftHip","leftKnee"],["leftKnee","leftAnkle"],["nose","rightShoulder"],["rightShoulder","rightElbow"],["rightElbow","rightWrist"],["rightShoulder","rightHip"],["rightHip","rightKnee"],["rightKnee","rightAnkle"]];function mT(e){let t=e.reduce(({maxX:n,maxY:s,minX:r,minY:a},{position:{x:o,y:i}})=>({maxX:Math.max(n,o),maxY:Math.max(s,i),minX:Math.min(r,o),minY:Math.min(a,i)}),{maxX:Number.NEGATIVE_INFINITY,maxY:Number.NEGATIVE_INFINITY,minX:Number.POSITIVE_INFINITY,minY:Number.POSITIVE_INFINITY});return[t.minX,t.minY,t.maxX-t.minX,t.maxY-t.minY]}function gT(e,[t,n],[s,r]){let a=t/s,o=n/r,i=(c,u)=>({id:u,score:c.score,boxRaw:[c.box[0]/r,c.box[1]/s,c.box[2]/r,c.box[3]/s],box:[Math.trunc(c.box[0]*o),Math.trunc(c.box[1]*a),Math.trunc(c.box[2]*o),Math.trunc(c.box[3]*a)],keypoints:c.keypoints.map(({score:d,part:p,position:h})=>({score:d,part:p,position:[Math.trunc(h.x*o),Math.trunc(h.y*a)],positionRaw:[h.x/s,h.y/s]}))});return e.map((c,u)=>i(c,u))}var Mb=class{constructor(t,n){ce(this,"priorityQueue");ce(this,"numberOfElements");ce(this,"getElementValue");this.priorityQueue=new Array(t),this.numberOfElements=-1,this.getElementValue=n}enqueue(t){this.priorityQueue[++this.numberOfElements]=t,this.swim(this.numberOfElements)}dequeue(){let t=this.priorityQueue[0];return this.exchange(0,this.numberOfElements--),this.sink(0),this.priorityQueue[this.numberOfElements+1]=null,t}empty(){return this.numberOfElements===-1}size(){return this.numberOfElements+1}all(){return this.priorityQueue.slice(0,this.numberOfElements+1)}max(){return this.priorityQueue[0]}swim(t){for(;t>0&&this.less(Math.floor(t/2),t);)this.exchange(t,Math.floor(t/2)),t=Math.floor(t/2)}sink(t){for(;2*t<=this.numberOfElements;){let n=2*t;if(n<this.numberOfElements&&this.less(n,n+1)&&n++,!this.less(t,n))break;this.exchange(t,n),t=n}}getValueAt(t){return this.getElementValue(this.priorityQueue[t])}less(t,n){return this.getValueAt(t)<this.getValueAt(n)}exchange(t,n){let s=this.priorityQueue[t];this.priorityQueue[t]=this.priorityQueue[n],this.priorityQueue[n]=s}};function zb(e,t,n,s){return{y:s.get(e,t,n),x:s.get(e,t,n+o2e)}}function Lb(e,t,n){let{heatmapY:s,heatmapX:r,id:a}=e,{y:o,x:i}=zb(s,r,a,n);return{x:e.heatmapX*t+i,y:e.heatmapY*t+o}}function Bb(e,t,n){return e<t?t:e>n?n:e}function AT(e,t,n,s){let r=n-e,a=s-t;return r*r+a*a}function Wb(e,t){return{x:e.x+t.x,y:e.y+t.y}}var vs,l2e=["MobilenetV1/offset_2/BiasAdd","MobilenetV1/heatmap_2/BiasAdd","MobilenetV1/displacement_fwd_2/BiasAdd","MobilenetV1/displacement_bwd_2/BiasAdd"],v0=1,Ec=16,u2e=50**2;function yT(e,t,n,s,r,a,o=2){let i=A=>({y:a.get(A.y,A.x,e),x:a.get(A.y,A.x,a.shape[2]/2+e)}),l=(A,x,y)=>({y:Bb(Math.round(A.y/Ec),0,x-1),x:Bb(Math.round(A.x/Ec),0,y-1)}),[c,u]=s.shape,d=l(t.position,c,u),p=i(d),f=Wb(t.position,p);for(let A=0;A<o;A++){let x=l(f,c,u),y=zb(x.y,x.x,n,r);f=Wb({x:x.x*Ec,y:x.y*Ec},{x:y.x,y:y.y})}let m=l(f,c,u),g=s.get(m.y,m.x,n);return{position:f,part:Dp[n],score:g}}function c2e(e,t,n,s,r){let a=fT.map(([p,h])=>[Pp[p],Pp[h]]),o=a.map(([,p])=>p),i=a.map(([p])=>p),l=t.shape[2],c=o.length,u=new Array(l),d=Lb(e.part,Ec,n);u[e.part.id]={score:e.score,part:Dp[e.part.id],position:d};for(let p=c-1;p>=0;--p){let h=o[p],f=i[p];u[h]&&!u[f]&&(u[f]=yT(p,u[h],f,t,n,r))}for(let p=0;p<c;++p){let h=i[p],f=o[p];u[h]&&!u[f]&&(u[f]=yT(p,u[h],f,t,n,s))}return u}function d2e(e,t,n,s,r){let[a,o]=r.shape,i=!0,l=Math.max(n-v0,0),c=Math.min(n+v0+1,a);for(let u=l;u<c;++u){let d=Math.max(s-v0,0),p=Math.min(s+v0+1,o);for(let h=d;h<p;++h)if(r.get(u,h,e)>t){i=!1;break}if(!i)break}return i}function p2e(e,t){let[n,s,r]=t.shape,a=new Mb(n*s*r,({score:o})=>o);for(let o=0;o<n;++o)for(let i=0;i<s;++i)for(let l=0;l<r;++l){let c=t.get(o,i,l);c<e||d2e(l,c,o,i,t)&&a.enqueue({score:c,part:{heatmapY:o,heatmapX:i,id:l}})}return a}function xT(e,{x:t,y:n},s){return e.some(({keypoints:r})=>{var o;let a=(o=r[s])==null?void 0:o.position;return a?AT(n,t,a.y,a.x)<=u2e:!1})}function h2e(e,t){return t.reduce((s,{position:r,score:a},o)=>(xT(e,r,o)||(s+=a),s),0)/t.length}function f2e(e,t,n,s,r,a){let o=[],i=p2e(a,t);for(;o.length<r&&!i.empty();){let l=i.dequeue(),c=Lb(l.part,Ec,e);if(xT(o,c,l.part.id))continue;let u=c2e(l,t,e,n,s);u=u.filter(h=>h.score>a);let d=h2e(o,u),p=mT(u);d>a&&o.push({keypoints:u,box:p,score:Math.round(100*d)/100})}return o}async function Vb(e,t){let n=q(()=>{if(!vs.inputs[0].shape)return[];let o=$e.resizeBilinear(e,[vs.inputs[0].shape[2],vs.inputs[0].shape[1]]),i=me(pe(de(o,"float32"),127.5),1),c=vs.execute(i,l2e).map(u=>ot(u,[0]));return c[1]=c[1].sigmoid(),c}),s=await Promise.all(n.map(o=>o.buffer()));for(let o of n)se(o);let r=await f2e(s[0],s[1],s[2],s[3],t.body.maxDetected,t.body.minConfidence);return vs.inputs[0].shape?gT(r,[e.shape[1],e.shape[2]],[vs.inputs[0].shape[2],vs.inputs[0].shape[1]]):[]}async function bT(e){return!vs||Ae.initial?(vs=await et(nt(e.modelBasePath,e.body.modelPath||"")),!vs||!vs.modelUrl?re("load model failed:",e.body.modelPath):e.debug&&re("load model:",vs.modelUrl)):e.debug&&re("cached model:",vs.modelUrl),vs}var zs,Ub=!1;async function Gb(e){return!zs||Ae.initial?(zs=await et(nt(e.modelBasePath,e.segmentation.modelPath||"")),!zs||!zs.modelUrl?re("load model failed:",e.segmentation.modelPath):e.debug&&re("load model:",zs.modelUrl)):e.debug&&re("cached model:",zs.modelUrl),zs}async function vT(e,t,n){var m,g;if(Ub)return{data:[],canvas:null,alpha:null};Ub=!0,zs||await Gb(n);let s=await vc(e,n),r=((m=s.canvas)==null?void 0:m.width)||0,a=((g=s.canvas)==null?void 0:g.height)||0;if(!s.tensor)return{data:[],canvas:null,alpha:null};let o={};o.resize=$e.resizeBilinear(s.tensor,[zs.inputs[0].shape?zs.inputs[0].shape[1]:0,zs.inputs[0].shape?zs.inputs[0].shape[2]:0],!1),se(s.tensor),o.norm=pe(o.resize,255),o.res=zs.execute(o.norm),o.squeeze=ot(o.res,0),o.squeeze.shape[2]===2?(o.softmax=Vu(o.squeeze),[o.bg,o.fg]=ss(o.softmax,2),o.expand=mn(o.fg,2),o.pad=mn(o.expand,0),o.crop=$e.cropAndResize(o.pad,[[0,0,.5,.5]],[0],[r,a]),o.data=ot(o.crop,0)):o.data=$e.resizeBilinear(o.squeeze,[a,r]);let i=Array.from(await o.data.data());if(Ae.node&&!Ae.Canvas&&typeof ImageData=="undefined")return n.debug&&re("canvas support missing"),Object.keys(o).forEach(A=>se(o[A])),{data:i,canvas:null,alpha:null};let l=Xn(r,a);await Gs.toPixels(o.data,l);let c=l.getContext("2d");n.segmentation.blur&&n.segmentation.blur>0&&(c.filter=`blur(${n.segmentation.blur}px)`);let u=c.getImageData(0,0,r,a),d=Xn(r,a),p=d.getContext("2d");s.canvas&&p.drawImage(s.canvas,0,0),p.globalCompositeOperation="darken",n.segmentation.blur&&n.segmentation.blur>0&&(p.filter=`blur(${n.segmentation.blur}px)`),p.drawImage(l,0,0),p.globalCompositeOperation="source-over",p.filter="none";let h=p.getImageData(0,0,r,a);for(let A=0;A<r*a;A++)h.data[4*A+3]=u.data[4*A+0];p.putImageData(h,0,0);let f=null;if(t&&d){f=Xn(r,a);let A=await vc(t,n);se(A.tensor);let x=f.getContext("2d");x.drawImage(A.canvas,0,0,f.width,f.height),x.drawImage(d,0,0)}return Object.keys(o).forEach(A=>se(o[A])),Ub=!1,{data:i,canvas:f||d,alpha:l}}var Hb=class{constructor(){ce(this,"age",null);ce(this,"agegenderrace",null);ce(this,"blazeposedetect",null);ce(this,"blazepose",null);ce(this,"centernet",null);ce(this,"efficientpose",null);ce(this,"embedding",null);ce(this,"emotion",null);ce(this,"facedetect",null);ce(this,"faceiris",null);ce(this,"facemesh",null);ce(this,"faceres",null);ce(this,"gender",null);ce(this,"handpose",null);ce(this,"handskeleton",null);ce(this,"handtrack",null);ce(this,"movenet",null);ce(this,"nanodet",null);ce(this,"posenet",null);ce(this,"segmentation",null);ce(this,"antispoof",null)}};function jb(e){for(let t of Object.keys(e.models))e.models[t]=null}async function wT(e){var t,n,s,r,a,o,i,l,c,u,d,p,h,f,m,g,A,x,y,b,w,k,C,N,R,F,_,P,T,M,U;Ae.initial&&jb(e),e.config.hand.enabled&&(!e.models.handpose&&((n=(t=e.config.hand.detector)==null?void 0:t.modelPath)==null?void 0:n.includes("handdetect"))&&([e.models.handpose,e.models.handskeleton]=await kb(e.config)),!e.models.handskeleton&&e.config.hand.landmarks&&((r=(s=e.config.hand.detector)==null?void 0:s.modelPath)==null?void 0:r.includes("handdetect"))&&([e.models.handpose,e.models.handskeleton]=await kb(e.config))),e.config.face.enabled&&!e.models.facedetect&&(e.models.facedetect=l8(e.config)),e.config.face.enabled&&((a=e.config.face.mesh)==null?void 0:a.enabled)&&!e.models.facemesh&&(e.models.facemesh=E8(e.config)),e.config.face.enabled&&((o=e.config.face.iris)==null?void 0:o.enabled)&&!e.models.faceiris&&(e.models.faceiris=v8(e.config)),e.config.face.enabled&&((i=e.config.face.antispoof)==null?void 0:i.enabled)&&!e.models.antispoof&&(e.models.antispoof=J6(e.config)),e.config.hand.enabled&&!e.models.handtrack&&((c=(l=e.config.hand.detector)==null?void 0:l.modelPath)==null?void 0:c.includes("handtrack"))&&(e.models.handtrack=aT(e.config)),e.config.hand.enabled&&e.config.hand.landmarks&&!e.models.handskeleton&&((d=(u=e.config.hand.detector)==null?void 0:u.modelPath)==null?void 0:d.includes("handtrack"))&&(e.models.handskeleton=oT(e.config)),e.config.body.enabled&&!e.models.posenet&&((h=(p=e.config.body)==null?void 0:p.modelPath)==null?void 0:h.includes("posenet"))&&(e.models.posenet=bT(e.config)),e.config.body.enabled&&!e.models.efficientpose&&((m=(f=e.config.body)==null?void 0:f.modelPath)==null?void 0:m.includes("efficientpose"))&&(e.models.efficientpose=ob(e.config)),e.config.body.enabled&&!e.models.blazepose&&((A=(g=e.config.body)==null?void 0:g.modelPath)==null?void 0:A.includes("blazepose"))&&(e.models.blazepose=h8(e.config)),e.config.body.enabled&&!e.models.blazeposedetect&&((x=e.config.body.detector)==null?void 0:x.modelPath)&&((b=(y=e.config.body)==null?void 0:y.modelPath)==null?void 0:b.includes("blazepose"))&&(e.models.blazeposedetect=p8(e.config)),e.config.body.enabled&&!e.models.efficientpose&&((k=(w=e.config.body)==null?void 0:w.modelPath)==null?void 0:k.includes("efficientpose"))&&(e.models.efficientpose=ob(e.config)),e.config.body.enabled&&!e.models.movenet&&((N=(C=e.config.body)==null?void 0:C.modelPath)==null?void 0:N.includes("movenet"))&&(e.models.movenet=dT(e.config)),e.config.object.enabled&&!e.models.nanodet&&((F=(R=e.config.object)==null?void 0:R.modelPath)==null?void 0:F.includes("nanodet"))&&(e.models.nanodet=hT(e.config)),e.config.object.enabled&&!e.models.centernet&&((P=(_=e.config.object)==null?void 0:_.modelPath)==null?void 0:P.includes("centernet"))&&(e.models.centernet=g8(e.config)),e.config.face.enabled&&((T=e.config.face.emotion)==null?void 0:T.enabled)&&!e.models.emotion&&(e.models.emotion=b8(e.config)),e.config.face.enabled&&((M=e.config.face.description)==null?void 0:M.enabled)&&!e.models.faceres&&(e.models.faceres=P8(e.config)),e.config.segmentation.enabled&&!e.models.segmentation&&(e.models.segmentation=Gb(e.config)),e.config.face.enabled&&((U=e.config.face.agegenderrace)==null?void 0:U.enabled)&&!e.models.agegenderrace&&(e.models.agegenderrace=K6(e.config));for await(let j of Object.keys(e.models))e.models[j]&&typeof e.models[j]!="undefined"&&(e.models[j]=await e.models[j])}async function kT(e){let t=["const","placeholder","noop","pad","squeeze","add","sub","mul","div"];for(let n of Object.keys(e.models))if(e.models[n]){let s=[];Array.isArray(e.models[n])?s=e.models[n].filter(r=>r!==null).map(r=>r&&r.executor?r:r.model):s=[e.models[n]];for(let r of s){if(!r){e.config.debug&&re("model marked as loaded but not defined:",n);continue}let a=[],o=r==null?void 0:r.executor;if(o&&o.graph.nodes)for(let l of Object.values(o.graph.nodes)){let c=l.op.toLowerCase();a.includes(c)||a.push(c)}else!o&&e.config.debug&&re("model signature not determined:",n);let i=[];for(let l of a)!t.includes(l)&&!e.env.kernels.includes(l)&&!e.env.kernels.includes(l.replace("_",""))&&!e.env.kernels.includes(l.replace("native",""))&&!e.env.kernels.includes(l.replace("v2",""))&&i.push(l);i.length>0&&e.config.debug&&re("model validation:",n,i)}}}var It={name:"humangl",priority:999,canvas:null,gl:null,extensions:[],webGLattr:{alpha:!1,antialias:!1,premultipliedAlpha:!1,preserveDrawingBuffer:!1,depth:!1,stencil:!1,failIfMajorPerformanceCaveat:!1,desynchronized:!0}};function m2e(){let e=It.gl;!e||(It.extensions=e.getSupportedExtensions())}async function ST(e){var t;if(e.config.backend==="humangl"&&(It.name in ns().registry&&(!It.gl||!It.gl.getParameter(It.gl.VERSION))&&(re("error: humangl backend invalid context"),jb(e)),!L2(It.name))){try{It.canvas=await Xn(100,100)}catch(s){re("error: cannot create canvas:",s);return}try{if(It.gl=(t=It.canvas)==null?void 0:t.getContext("webgl2",It.webGLattr),!It.gl.getParameter(It.gl.VERSION).includes("2.0")){re("override: using fallback webgl backend as webgl 2.0 is not detected"),e.config.backend="webgl";return}It.canvas&&(It.canvas.addEventListener("webglcontextlost",async r=>{throw re("error: humangl:",r.type),re("possible browser memory leak using webgl or conflict with multiple backend registrations"),e.emit("error"),new Error("browser webgl error")}),It.canvas.addEventListener("webglcontextrestored",r=>{re("error: humangl context restored:",r)}),It.canvas.addEventListener("webglcontextcreationerror",r=>{re("error: humangl context create:",r)}))}catch(s){re("error: cannot get WebGL context:",s);return}try{Cm(2,It.gl)}catch(s){re("error: cannot set WebGL context:",s);return}try{let s=new Fm(It.gl);al(It.name,()=>new pp(s),It.priority)}catch(s){re("error: cannot register WebGL backend:",s);return}try{qr("webgl").forEach(r=>{let a={...r,backendName:It.name};rr(a)})}catch(s){re("error: cannot update WebGL backend registration:",s);return}let n=Tr().getGPGPUContext?Tr().getGPGPUContext().gl:null;if(n)re(`humangl webgl version:${n.getParameter(n.VERSION)} renderer:${n.getParameter(n.RENDERER)}`);else{re("error: no current gl context:",n,It.gl);return}try{Sr.set("WEBGL_VERSION",2)}catch(s){re("error: cannot set WebGL backend flags:",s);return}m2e(),re("backend registered:",It.name)}}function g2e(){if(!Ae.kernels.includes("mod")){let e={kernelName:"Mod",backendName:Is(),kernelFunc:t=>q(()=>me(t.inputs.a,L(pe(t.inputs.a,t.inputs.b),t.inputs.b)))};rr(e),Ae.kernels.push("mod")}if(!Ae.kernels.includes("floormod")){let e={kernelName:"FloorMod",backendName:Is(),kernelFunc:t=>q(()=>nf(t.inputs.a/t.inputs.b)*t.inputs.b+Rd(t.inputs.a,t.inputs.b))};rr(e),Ae.kernels.push("floormod")}}async function w0(e,t=!1){if(e.state="backend",t||Ae.initial||e.config.backend&&e.config.backend.length>0&&Is()!==e.config.backend){let n=he();if(e.config.backend&&e.config.backend.length>0){if(typeof window=="undefined"&&typeof WorkerGlobalScope!="undefined"&&e.config.debug&&e.config.debug&&re("running inside web worker"),Ae.browser&&e.config.backend==="tensorflow"&&(e.config.debug&&re("override: backend set to tensorflow while running in browser"),e.config.backend="humangl"),Ae.node&&(e.config.backend==="webgl"||e.config.backend==="humangl")&&(e.config.debug&&re(`override: backend set to ${e.config.backend} while running in nodejs`),e.config.backend="tensorflow"),Ae.browser&&e.config.backend==="webgpu")if(typeof navigator=="undefined"||typeof navigator.gpu=="undefined")re("override: backend set to webgpu but browser does not support webgpu"),e.config.backend="humangl";else{let r=await navigator.gpu.requestAdapter();e.config.debug&&re("enumerated webgpu adapter:",r)}e.config.backend==="humangl"&&await ST(e);let s=Object.keys(ns().registryFactory);if(e.config.debug&&re("available backends:",s),s.includes(e.config.backend)||(re(`error: backend ${e.config.backend} not found in registry`),e.config.backend=Ae.node?"tensorflow":"webgl",e.config.debug&&re(`override: setting backend ${e.config.backend}`)),e.config.debug&&re("setting backend:",e.config.backend),e.config.backend==="wasm"){if(e.config.debug&&re("wasm path:",e.config.wasmPath),typeof($l==null?void 0:$l.setWasmPaths)!="undefined")await F6(e.config.wasmPath);else throw new Error("wasm backend is not loaded");let r=await K().getAsync("WASM_HAS_SIMD_SUPPORT"),a=await K().getAsync("WASM_HAS_MULTITHREAD_SUPPORT");e.config.debug&&re(`wasm execution: ${r?"SIMD":"no SIMD"} ${a?"multithreaded":"singlethreaded"}`),e.config.debug&&!r&&re("warning: wasm simd support is not enabled")}try{await _3(e.config.backend),await tf()}catch(r){return re("error: cannot set backend:",e.config.backend,r),!1}}if(Is()==="humangl"&&(Sr.set("CHECK_COMPUTATION_FOR_ERRORS",!1),Sr.set("WEBGL_CPU_FORWARD",!0),Sr.set("WEBGL_USE_SHAPES_UNIFORMS",!0),Sr.set("CPU_HANDOFF_SIZE_THRESHOLD",256),typeof e.config.deallocate!="undefined"&&e.config.deallocate&&(re("changing webgl: WEBGL_DELETE_TEXTURE_THRESHOLD:",!0),Sr.set("WEBGL_DELETE_TEXTURE_THRESHOLD",0)),Tr().getGPGPUContext)){let s=await Tr().getGPGPUContext().gl;e.config.debug&&re(`gl version:${s.getParameter(s.VERSION)} renderer:${s.getParameter(s.RENDERER)}`)}Is()==="webgpu",$3(),await tf(),e.performance.initBackend=Math.trunc(he()-n),e.config.backend=Is(),await Ae.updateBackend(),g2e()}return!0}function m0(e,t){for(let n of e){let s={kernelName:n,backendName:t.backend,kernelFunc:()=>{t.debug&&re("kernelFunc",n,t.backend)}};rr(s)}Ae.kernels=qr(Is()).map(n=>n.kernelName.toLowerCase())}var ha={color:"rgba(173, 216, 230, 0.6)",labelColor:"rgba(173, 216, 230, 1)",shadowColor:"black",font:'small-caps 16px "Segoe UI"',lineHeight:18,lineWidth:4,pointSize:2,roundRect:8,drawPoints:!1,drawLabels:!0,drawBoxes:!0,drawGestures:!0,drawPolygons:!0,drawGaze:!0,fillPolygons:!1,useDepth:!0,useCurves:!1},qb=0,Wl=e=>{if(e&&e.getContext)return e.getContext("2d");throw new Error("invalid canvas")},Rc=e=>Math.round(e*180/Math.PI);function Xb(e,t,n,s=0,r){e.fillStyle=r.useDepth&&s?`rgba(${127.5+2*s}, ${127.5-2*s}, 255, 0.3)`:r.color,e.beginPath(),e.arc(t,n,r.pointSize,0,2*Math.PI),e.fill()}function Fp(e,t,n,s,r,a){if(e.beginPath(),a.useCurves){let o=(t+t+s)/2,i=(n+n+r)/2;e.ellipse(o,i,s/2,r/2,0,0,2*Math.PI)}else e.lineWidth=a.lineWidth,e.moveTo(t+a.roundRect,n),e.lineTo(t+s-a.roundRect,n),e.quadraticCurveTo(t+s,n,t+s,n+a.roundRect),e.lineTo(t+s,n+r-a.roundRect),e.quadraticCurveTo(t+s,n+r,t+s-a.roundRect,n+r),e.lineTo(t+a.roundRect,n+r),e.quadraticCurveTo(t,n+r,t,n+r-a.roundRect),e.lineTo(t,n+a.roundRect),e.quadraticCurveTo(t,n,t+a.roundRect,n),e.closePath();e.stroke()}function IT(e,t,n){if(!(t.length<2)){e.beginPath(),e.moveTo(t[0][0],t[0][1]);for(let s of t){let r=s[2]||0;e.strokeStyle=n.useDepth&&r!==0?`rgba(${127.5+2*r}, ${127.5-2*r}, 255, 0.3)`:n.color,e.fillStyle=n.useDepth&&r!==0?`rgba(${127.5+2*r}, ${127.5-2*r}, 255, 0.3)`:n.color,e.lineTo(s[0],Math.round(s[1]))}e.stroke(),n.fillPolygons&&(e.closePath(),e.fill())}}function A2e(e,t,n){if(!(t.length<2)){if(!n.useCurves||t.length<=2){IT(e,t,n);return}e.moveTo(t[0][0],t[0][1]);for(let s=0;s<t.length-2;s++){let r=(t[s][0]+t[s+1][0])/2,a=(t[s][1]+t[s+1][1])/2;e.quadraticCurveTo(t[s][0],t[s][1],r,a)}e.quadraticCurveTo(t[t.length-2][0],t[t.length-2][1],t[t.length-1][0],t[t.length-1][1]),e.stroke(),n.fillPolygons&&(e.closePath(),e.fill())}}function CT(e,t,n,s=5){let r,a,o;e.beginPath(),e.moveTo(t[0],t[1]),e.lineTo(n[0],n[1]),r=Math.atan2(n[1]-t[1],n[0]-t[0]),a=s*Math.cos(r)+n[0],o=s*Math.sin(r)+n[1],e.moveTo(a,o),r+=1/3*(2*Math.PI),a=s*Math.cos(r)+n[0],o=s*Math.sin(r)+n[1],e.lineTo(a,o),r+=1/3*(2*Math.PI),a=s*Math.cos(r)+n[0],o=s*Math.sin(r)+n[1],e.lineTo(a,o),e.closePath(),e.stroke(),e.fill()}async function Kb(e,t,n){let s=Rn(ha,n);if(!(!t||!e)&&s.drawGestures){let r=Wl(e);r.font=s.font,r.fillStyle=s.color;let a=1;for(let o=0;o<t.length;o++){let i=[],l=[];if([i,l]=Object.entries(t[o]),l.length>1&&l[1].length>0){let c=i[1]>0?`#${i[1]}`:"",u=`${i[0]} ${c}: ${l[1]}`;s.shadowColor&&s.shadowColor!==""&&(r.fillStyle=s.shadowColor,r.fillText(u,8,2+a*s.lineHeight)),r.fillStyle=s.labelColor,r.fillText(u,6,0+a*s.lineHeight),a+=1}}}}async function Zb(e,t,n){var a,o,i,l,c;let s=Rn(ha,n);if(!t||!e)return;let r=Wl(e);for(let u of t){if(r.font=s.font,r.strokeStyle=s.color,r.fillStyle=s.color,s.drawBoxes&&Fp(r,u.box[0],u.box[1],u.box[2],u.box[3],s),s.drawLabels){let d=[];if(d.push(`face: ${Math.trunc(100*u.score)}%`),u.genderScore&&d.push(`${u.gender||""} ${Math.trunc(100*u.genderScore)}%`),u.age&&d.push(`age: ${u.age||""}`),u.iris&&d.push(`distance: ${u.iris}`),u.real&&d.push(`real: ${Math.trunc(100*u.real)}%`),u.emotion&&u.emotion.length>0){let p=u.emotion.map(h=>`${Math.trunc(100*h.score)}% ${h.emotion}`);p.length>3&&(p.length=3),d.push(p.join(" "))}u.rotation&&u.rotation.angle&&u.rotation.gaze&&(u.rotation.angle.roll&&d.push(`roll: ${Rc(u.rotation.angle.roll)}\xB0 yaw:${Rc(u.rotation.angle.yaw)}\xB0 pitch:${Rc(u.rotation.angle.pitch)}\xB0`),u.rotation.gaze.bearing&&d.push(`gaze: ${Rc(u.rotation.gaze.bearing)}\xB0`)),d.length===0&&d.push("face"),r.fillStyle=s.color;for(let p=d.length-1;p>=0;p--){let h=Math.max(u.box[0],0),f=p*s.lineHeight+u.box[1];s.shadowColor&&s.shadowColor!==""&&(r.fillStyle=s.shadowColor,r.fillText(d[p],h+5,f+16)),r.fillStyle=s.labelColor,r.fillText(d[p],h+4,f+15)}}if(r.lineWidth=1,u.mesh&&u.mesh.length>0){if(s.drawPoints)for(let d of u.mesh)Xb(r,d[0],d[1],d[2],s);if(s.drawPolygons){if(r.lineWidth=1,u.mesh.length>450)for(let d=0;d<_l.length/3;d++){let p=[_l[d*3+0],_l[d*3+1],_l[d*3+2]].map(h=>u.mesh[h]);IT(r,p,s)}if(u.annotations&&u.annotations.leftEyeIris&&u.annotations.leftEyeIris[0]){r.strokeStyle=s.useDepth?"rgba(255, 200, 255, 0.3)":s.color,r.beginPath();let d=Math.abs(u.annotations.leftEyeIris[3][0]-u.annotations.leftEyeIris[1][0])/2,p=Math.abs(u.annotations.leftEyeIris[4][1]-u.annotations.leftEyeIris[2][1])/2;r.ellipse(u.annotations.leftEyeIris[0][0],u.annotations.leftEyeIris[0][1],d,p,0,0,2*Math.PI),r.stroke(),s.fillPolygons&&(r.fillStyle=s.useDepth?"rgba(255, 255, 200, 0.3)":s.color,r.fill())}if(u.annotations&&u.annotations.rightEyeIris&&u.annotations.rightEyeIris[0]){r.strokeStyle=s.useDepth?"rgba(255, 200, 255, 0.3)":s.color,r.beginPath();let d=Math.abs(u.annotations.rightEyeIris[3][0]-u.annotations.rightEyeIris[1][0])/2,p=Math.abs(u.annotations.rightEyeIris[4][1]-u.annotations.rightEyeIris[2][1])/2;r.ellipse(u.annotations.rightEyeIris[0][0],u.annotations.rightEyeIris[0][1],d,p,0,0,2*Math.PI),r.stroke(),s.fillPolygons&&(r.fillStyle=s.useDepth?"rgba(255, 255, 200, 0.3)":s.color,r.fill())}if(s.drawGaze&&((a=u.rotation)==null?void 0:a.angle)){r.strokeStyle="pink";let d=u.box[0]+u.box[2]/2-u.box[3]*Rc(u.rotation.angle.yaw)/90,p=u.box[1]+u.box[3]/2+u.box[2]*Rc(u.rotation.angle.pitch)/90,h=new Path2D(`
|
|
M ${u.box[0]+u.box[2]/2} ${u.box[1]}
|
|
C
|
|
${d} ${u.box[1]},
|
|
${d} ${u.box[1]+u.box[3]},
|
|
${u.box[0]+u.box[2]/2} ${u.box[1]+u.box[3]}
|
|
`),f=new Path2D(`
|
|
M ${u.box[0]} ${u.box[1]+u.box[3]/2}
|
|
C
|
|
${u.box[0]} ${p},
|
|
${u.box[0]+u.box[2]} ${p},
|
|
${u.box[0]+u.box[2]} ${u.box[1]+u.box[3]/2}
|
|
`);r.stroke(f),r.stroke(h)}if(s.drawGaze&&((i=(o=u.rotation)==null?void 0:o.gaze)==null?void 0:i.strength)&&((c=(l=u.rotation)==null?void 0:l.gaze)==null?void 0:c.bearing)&&u.annotations.leftEyeIris&&u.annotations.rightEyeIris&&u.annotations.leftEyeIris[0]&&u.annotations.rightEyeIris[0]){r.strokeStyle="pink",r.fillStyle="pink";let d=[u.annotations.leftEyeIris[0][0]+Math.sin(u.rotation.gaze.bearing)*u.rotation.gaze.strength*u.box[3],u.annotations.leftEyeIris[0][1]+Math.cos(u.rotation.gaze.bearing)*u.rotation.gaze.strength*u.box[2]];CT(r,[u.annotations.leftEyeIris[0][0],u.annotations.leftEyeIris[0][1]],[d[0],d[1]],4);let p=[u.annotations.rightEyeIris[0][0]+Math.sin(u.rotation.gaze.bearing)*u.rotation.gaze.strength*u.box[3],u.annotations.rightEyeIris[0][1]+Math.cos(u.rotation.gaze.bearing)*u.rotation.gaze.strength*u.box[2]];CT(r,[u.annotations.rightEyeIris[0][0],u.annotations.rightEyeIris[0][1]],[p[0],p[1]],4)}}}}}async function Yb(e,t,n){var a;let s=Rn(ha,n);if(!t||!e)return;let r=Wl(e);r.lineJoin="round";for(let o=0;o<t.length;o++){if(r.strokeStyle=s.color,r.fillStyle=s.color,r.lineWidth=s.lineWidth,r.font=s.font,s.drawBoxes&&t[o].box&&((a=t[o].box)==null?void 0:a.length)===4&&(Fp(r,t[o].box[0],t[o].box[1],t[o].box[2],t[o].box[3],s),s.drawLabels&&(s.shadowColor&&s.shadowColor!==""&&(r.fillStyle=s.shadowColor,r.fillText(`body ${100*t[o].score}%`,t[o].box[0]+3,1+t[o].box[1]+s.lineHeight,t[o].box[2])),r.fillStyle=s.labelColor,r.fillText(`body ${100*t[o].score}%`,t[o].box[0]+2,0+t[o].box[1]+s.lineHeight,t[o].box[2]))),s.drawPoints&&t[o].keypoints)for(let i=0;i<t[o].keypoints.length;i++)r.fillStyle=s.useDepth&&t[o].keypoints[i].position[2]?`rgba(${127.5+2*(t[o].keypoints[i].position[2]||0)}, ${127.5-2*(t[o].keypoints[i].position[2]||0)}, 255, 0.5)`:s.color,Xb(r,t[o].keypoints[i].position[0],t[o].keypoints[i].position[1],0,s);if(s.drawLabels&&t[o].keypoints){r.font=s.font;for(let i of t[o].keypoints)r.fillStyle=s.useDepth&&i.position[2]?`rgba(${127.5+2*i.position[2]}, ${127.5-2*i.position[2]}, 255, 0.5)`:s.color,r.fillText(`${i.part} ${Math.trunc(100*i.score)}%`,i.position[0]+4,i.position[1]+4)}if(s.drawPolygons&&t[o].keypoints&&t[o].annotations)for(let i of Object.values(t[o].annotations))for(let l of i)A2e(r,l,s)}}async function Jb(e,t,n){let s=Rn(ha,n);if(!t||!e)return;let r=Wl(e);r.lineJoin="round",r.font=s.font;for(let a of t){if(s.drawBoxes&&(r.strokeStyle=s.color,r.fillStyle=s.color,Fp(r,a.box[0],a.box[1],a.box[2],a.box[3],s),s.drawLabels&&(s.shadowColor&&s.shadowColor!==""&&(r.fillStyle=s.shadowColor,r.fillText(`hand:${Math.trunc(100*a.score)}%`,a.box[0]+3,1+a.box[1]+s.lineHeight,a.box[2])),r.fillStyle=s.labelColor,r.fillText(`hand:${Math.trunc(100*a.score)}%`,a.box[0]+2,0+a.box[1]+s.lineHeight,a.box[2])),r.stroke()),s.drawPoints&&a.keypoints&&a.keypoints.length>0)for(let o of a.keypoints)r.fillStyle=s.useDepth?`rgba(${127.5+2*(o[2]||0)}, ${127.5-2*(o[2]||0)}, 255, 0.5)`:s.color,Xb(r,o[0],o[1],0,s);if(s.drawLabels&&a.annotations){let o=(i,l)=>{!i||i.length===0||!i[0]||(r.fillStyle=s.useDepth?`rgba(${127.5+2*i[i.length-1][2]}, ${127.5-2*i[i.length-1][2]}, 255, 0.5)`:s.color,r.fillText(l,i[i.length-1][0]+4,i[i.length-1][1]+4))};r.font=s.font,o(a.annotations.index,"index"),o(a.annotations.middle,"middle"),o(a.annotations.ring,"ring"),o(a.annotations.pinky,"pinky"),o(a.annotations.thumb,"thumb"),o(a.annotations.palm,"palm")}if(s.drawPolygons&&a.annotations){let o=i=>{if(!(!i||i.length===0||!i[0]))for(let l=0;l<i.length;l++)r.beginPath(),r.strokeStyle=s.useDepth?`rgba(${127.5+l*i[l][2]}, ${127.5-l*i[l][2]}, 255, 0.5)`:s.color,r.moveTo(i[l>0?l-1:0][0],i[l>0?l-1:0][1]),r.lineTo(i[l][0],i[l][1]),r.stroke()};r.lineWidth=s.lineWidth,o(a.annotations.index),o(a.annotations.middle),o(a.annotations.ring),o(a.annotations.pinky),o(a.annotations.thumb)}}}async function Qb(e,t,n){let s=Rn(ha,n);if(!t||!e)return;let r=Wl(e);r.lineJoin="round",r.font=s.font;for(let a of t)if(s.drawBoxes){if(r.strokeStyle=s.color,r.fillStyle=s.color,Fp(r,a.box[0],a.box[1],a.box[2],a.box[3],s),s.drawLabels){let o=`${a.label} ${Math.round(100*a.score)}%`;s.shadowColor&&s.shadowColor!==""&&(r.fillStyle=s.shadowColor,r.fillText(o,a.box[0]+3,1+a.box[1]+s.lineHeight,a.box[2])),r.fillStyle=s.labelColor,r.fillText(o,a.box[0]+2,0+a.box[1]+s.lineHeight,a.box[2])}r.stroke()}}async function TT(e,t,n){let s=Rn(ha,n);if(!t||!e)return;let r=Wl(e);r.lineJoin="round",r.font=s.font;for(let a=0;a<t.length;a++)if(s.drawBoxes){if(r.strokeStyle=s.color,r.fillStyle=s.color,Fp(r,t[a].box[0],t[a].box[1],t[a].box[2],t[a].box[3],s),s.drawLabels){let o=`person #${a}`;s.shadowColor&&s.shadowColor!==""&&(r.fillStyle=s.shadowColor,r.fillText(o,t[a].box[0]+3,1+t[a].box[1]+s.lineHeight,t[a].box[2])),r.fillStyle=s.labelColor,r.fillText(o,t[a].box[0]+2,0+t[a].box[1]+s.lineHeight,t[a].box[2])}r.stroke()}}async function NT(e,t){if(!e||!t)return;Wl(t).drawImage(e,0,0)}async function ET(e,t,n){if(!t||!t.performance||!t||!e)return null;let s=he(),r=Rn(ha,n),a=Promise.all([Zb(e,t.face,r),Yb(e,t.body,r),Jb(e,t.hand,r),Qb(e,t.object,r),Kb(e,t.gesture,r)]);return qb=Ae.perfadd?qb+Math.round(he()-s):Math.round(he()-s),t.performance.draw=qb,a}var y2e=e=>{let t=(d,p)=>Math.atan2(d[1]-p[1],d[0]-p[0]);if(!e.annotations.rightEyeIris||!e.annotations.leftEyeIris)return{bearing:0,strength:0};let n=[0,-.1],s=1,r=e.mesh[33][2]>e.mesh[263][2],a=r?e.mesh[473]:e.mesh[468],o=r?[(e.mesh[133][0]+e.mesh[33][0])/2,(e.mesh[133][1]+e.mesh[33][1])/2]:[(e.mesh[263][0]+e.mesh[362][0])/2,(e.mesh[263][1]+e.mesh[362][1])/2],i=r?[e.mesh[133][0]-e.mesh[33][0],e.mesh[23][1]-e.mesh[27][1]]:[e.mesh[263][0]-e.mesh[362][0],e.mesh[253][1]-e.mesh[257][1]],l=[(o[0]-a[0])/i[0]-n[0],s*(a[1]-o[1])/i[1]-n[1]],c=Math.sqrt(l[0]**2+l[1]**2);return c=Math.min(c,e.boxRaw[2]/2,e.boxRaw[3]/2),{bearing:(t([0,0],l)+Math.PI/2)%Math.PI,strength:c}},RT=(e,t)=>{let n=g=>{let A=Math.sqrt(g[0]*g[0]+g[1]*g[1]+g[2]*g[2]);return g[0]/=A,g[1]/=A,g[2]/=A,g},s=(g,A)=>{let x=g[0]-A[0],y=g[1]-A[1],b=g[2]-A[2];return[x,y,b]},r=(g,A)=>{let x=g[1]*A[2]-g[2]*A[1],y=g[2]*A[0]-g[0]*A[2],b=g[0]*A[1]-g[1]*A[0];return[x,y,b]},a=g=>{let[A,x,y,b,w,k,C,N,R]=g,F,_,P;return b<1?b>-1?(P=Math.asin(b),_=Math.atan2(-C,A),F=Math.atan2(-k,w)):(P=-Math.PI/2,_=-Math.atan2(N,R),F=0):(P=Math.PI/2,_=Math.atan2(N,R),F=0),isNaN(F)&&(F=0),isNaN(_)&&(_=0),isNaN(P)&&(P=0),{pitch:2*-F,yaw:2*-_,roll:2*-P}},o=g=>{let A=(y,b,w,k)=>Math.atan2(k-b,w-y);return{pitch:A(g[10][1],g[10][2],g[152][1],g[152][2]),yaw:A(g[33][0],g[33][2],g[263][0],g[263][2]),roll:A(g[33][0],g[33][1],g[263][0],g[263][1])}},i=e.meshRaw;if(!i||i.length<300)return{angle:{pitch:0,yaw:0,roll:0},matrix:[1,0,0,0,1,0,0,0,1],gaze:{bearing:0,strength:0}};let l=Math.max(e.boxRaw[2]*t[0],e.boxRaw[3]*t[1])/1.5,c=[i[10],i[152],i[234],i[454]].map(g=>[g[0]*t[0]/l,g[1]*t[1]/l,g[2]]),u=n(s(c[1],c[0])),d=n(s(c[3],c[2])),p=n(r(d,u));d=r(u,p);let h=[d[0],d[1],d[2],u[0],u[1],u[2],p[0],p[1],p[2]],f=a(h),m=i.length===478?y2e(e):{bearing:0,strength:0};return{angle:f,matrix:h,gaze:m}};var e5=async(e,t)=>{var p,h,f,m;let n,s,r,a,o,i,l,c,u=[];e.state="run:face",n=he();let d=await N8(t,e.config);if(e.performance.face=Ae.perfadd?(e.performance.face||0)+Math.trunc(he()-n):Math.trunc(he()-n),!t.shape||t.shape.length!==4)return[];if(!d)return[];for(let g=0;g<d.length;g++){if(e.analyze("Get Face"),!d[g].tensor||d[g].tensor.isDisposedInternal){re("Face object is disposed:",d[g].tensor);continue}let A=RT(d[g],[t.shape[2],t.shape[1]]);e.analyze("Start Emotion:"),e.config.async?o=e.config.face.emotion.enabled?cb(d[g].tensor||Ut([]),e.config,g,d.length):null:(e.state="run:emotion",n=he(),o=e.config.face.emotion.enabled?await cb(d[g].tensor||Ut([]),e.config,g,d.length):null,e.performance.emotion=Ae.perfadd?(e.performance.emotion||0)+Math.trunc(he()-n):Math.trunc(he()-n)),e.analyze("End Emotion:"),e.analyze("Start AntiSpoof:"),e.config.async?l=e.config.face.antispoof.enabled?zx(d[g].tensor||Ut([]),e.config,g,d.length):null:(e.state="run:antispoof",n=he(),l=e.config.face.antispoof.enabled?await zx(d[g].tensor||Ut([]),e.config,g,d.length):null,e.performance.antispoof=Ae.perfadd?(e.performance.antispoof||0)+Math.trunc(he()-n):Math.trunc(he()-n)),e.analyze("End AntiSpoof:"),e.analyze("Start Description:"),e.config.async?c=e.config.face.description.enabled?gb(d[g].tensor||Ut([]),e.config,g,d.length):null:(e.state="run:description",n=he(),c=e.config.face.description.enabled?await gb(d[g].tensor||Ut([]),e.config,g,d.length):null,e.performance.description=Ae.perfadd?(e.performance.description||0)+Math.trunc(he()-n):Math.trunc(he()-n)),e.analyze("End Description:"),e.config.async&&([s,a,o,i,c,r,l]=await Promise.all([s,a,o,i,c,r,l])),e.analyze("Finish Face:"),!e.config.face.iris.enabled&&((h=(p=d[g])==null?void 0:p.annotations)==null?void 0:h.leftEyeIris)&&((m=(f=d[g])==null?void 0:f.annotations)==null?void 0:m.rightEyeIris)&&(delete d[g].annotations.leftEyeIris,delete d[g].annotations.rightEyeIris);let x=d[g].annotations&&d[g].annotations.leftEyeIris&&d[g].annotations.leftEyeIris[0]&&d[g].annotations.rightEyeIris&&d[g].annotations.rightEyeIris[0]&&d[g].annotations.leftEyeIris.length>0&&d[g].annotations.rightEyeIris.length>0&&d[g].annotations.leftEyeIris[0]!==null&&d[g].annotations.rightEyeIris[0]!==null?Math.max(Math.abs(d[g].annotations.leftEyeIris[3][0]-d[g].annotations.leftEyeIris[1][0]),Math.abs(d[g].annotations.rightEyeIris[4][1]-d[g].annotations.rightEyeIris[2][1]))/t.shape[2]:0,y=e.config.face.detector.return?ot(d[g].tensor):null;se(d[g].tensor),d[g].tensor&&delete d[g].tensor,u.push({...d[g],id:g,age:c==null?void 0:c.age,gender:c==null?void 0:c.gender,genderScore:c==null?void 0:c.genderScore,embedding:c==null?void 0:c.descriptor,emotion:o,real:l,iris:x!==0?Math.trunc(500/x/11.7)/100:0,rotation:A,tensor:y}),e.analyze("End Face")}return e.analyze("End FaceMesh:"),e.config.async&&(e.performance.face&&delete e.performance.face,e.performance.age&&delete e.performance.age,e.performance.gender&&delete e.performance.gender,e.performance.emotion&&delete e.performance.emotion),u};var $T=e=>{if(!e)return[];let t=[];for(let n=0;n<e.length;n++){let s=e[n].keypoints.find(l=>l.part==="leftWrist"),r=e[n].keypoints.find(l=>l.part==="rightWrist"),a=e[n].keypoints.find(l=>l.part==="nose");a&&s&&r&&s.position[1]<a.position[1]&&r.position[1]<a.position[1]?t.push({body:n,gesture:"i give up"}):a&&s&&s.position[1]<a.position[1]?t.push({body:n,gesture:"raise left hand"}):a&&r&&r.position[1]<a.position[1]&&t.push({body:n,gesture:"raise right hand"});let o=e[n].keypoints.find(l=>l.part==="leftShoulder"),i=e[n].keypoints.find(l=>l.part==="rightShoulder");o&&i&&Math.abs(o.positionRaw[1]-i.positionRaw[1])>.1&&t.push({body:n,gesture:`leaning ${o.position[1]>i.position[1]?"left":"right"}`})}return t},_T=e=>{if(!e)return[];let t=[];for(let n=0;n<e.length;n++)if(e[n].mesh&&e[n].mesh.length>450){let s=e[n].mesh[33][2]-e[n].mesh[263][2];Math.abs(s)<10?t.push({face:n,gesture:"facing center"}):t.push({face:n,gesture:`facing ${s<0?"left":"right"}`}),Math.abs(e[n].mesh[374][1]-e[n].mesh[386][1])/Math.abs(e[n].mesh[443][1]-e[n].mesh[450][1])<.2&&t.push({face:n,gesture:"blink left eye"}),Math.abs(e[n].mesh[145][1]-e[n].mesh[159][1])/Math.abs(e[n].mesh[223][1]-e[n].mesh[230][1])<.2&&t.push({face:n,gesture:"blink right eye"});let o=Math.min(100,500*Math.abs(e[n].mesh[13][1]-e[n].mesh[14][1])/Math.abs(e[n].mesh[10][1]-e[n].mesh[152][1]));o>10&&t.push({face:n,gesture:`mouth ${Math.trunc(o)}% open`});let i=e[n].mesh[152][2];Math.abs(i)>10&&t.push({face:n,gesture:`head ${i<0?"up":"down"}`})}return t},DT=e=>{if(!e)return[];let t=[];for(let n=0;n<e.length;n++){if(!e[n].annotations||!e[n].annotations.leftEyeIris||!e[n].annotations.leftEyeIris[0]||!e[n].annotations.rightEyeIris||!e[n].annotations.rightEyeIris[0])continue;let s=e[n].annotations.leftEyeIris[3][0]-e[n].annotations.leftEyeIris[1][0],r=e[n].annotations.leftEyeIris[4][1]-e[n].annotations.leftEyeIris[2][1],a=Math.abs(s*r),o=e[n].annotations.rightEyeIris[3][0]-e[n].annotations.rightEyeIris[1][0],i=e[n].annotations.rightEyeIris[4][1]-e[n].annotations.rightEyeIris[2][1],l=Math.abs(o*i),c=!1;Math.abs(a-l)/Math.max(a,l)<.25&&(c=!0,t.push({iris:n,gesture:"facing center"}));let d=Math.abs(e[n].mesh[263][0]-e[n].annotations.leftEyeIris[0][0])/e[n].box[2],p=Math.abs(e[n].mesh[33][0]-e[n].annotations.rightEyeIris[0][0])/e[n].box[2];(d>.06||p>.06)&&(c=!1),d>p?d>.05&&t.push({iris:n,gesture:"looking right"}):p>.05&&t.push({iris:n,gesture:"looking left"});let h=Math.abs(e[n].mesh[145][1]-e[n].annotations.rightEyeIris[0][1])/e[n].box[3],f=Math.abs(e[n].mesh[374][1]-e[n].annotations.leftEyeIris[0][1])/e[n].box[3];(f<.01||h<.01||f>.022||h>.022)&&(c=!1),(f<.01||h<.01)&&t.push({iris:n,gesture:"looking down"}),(f>.022||h>.022)&&t.push({iris:n,gesture:"looking up"}),c&&t.push({iris:n,gesture:"looking center"})}return t},PT=e=>{if(!e)return[];let t=[];for(let n=0;n<e.length;n++){let s=[];if(e[n].annotations)for(let[r,a]of Object.entries(e[n].annotations))r!=="palmBase"&&Array.isArray(a)&&a[0]&&s.push({name:r.toLowerCase(),position:a[0]});if(s&&s.length>0){let r=s.reduce((o,i)=>o.position[2]<i.position[2]?o:i);t.push({hand:n,gesture:`${r.name} forward`});let a=s.reduce((o,i)=>o.position[1]<i.position[1]?o:i);t.push({hand:n,gesture:`${a.name} up`})}if(e[n].keypoints){let r=J8(e[n].keypoints);for(let a of r)t.push({hand:n,gesture:a.name})}}return t};var Pe={face:[],body:[],hand:[],gesture:[],object:[],persons:[],performance:{},timestamp:0},t5=0;function FT(e,t){var o,i,l,c,u,d,p,h,f,m,g,A,x,y,b,w,k,C,N,R,F,_,P,T,M,U,j;let n=he();if(!e)return{face:[],body:[],hand:[],gesture:[],object:[],persons:[],performance:{},timestamp:0};let s=Date.now()-e.timestamp,r=s<1e3?8-Math.log(s+1):1;if(Pe.canvas=e.canvas,!Pe.body||e.body.length!==Pe.body.length)Pe.body=JSON.parse(JSON.stringify(e.body));else for(let z=0;z<e.body.length;z++){let X=e.body[z].box.map((Q,te)=>((r-1)*Pe.body[z].box[te]+Q)/r),Z=e.body[z].boxRaw.map((Q,te)=>((r-1)*Pe.body[z].boxRaw[te]+Q)/r),J=e.body[z].keypoints.map((Q,te)=>({score:Q.score,part:Q.part,position:[Pe.body[z].keypoints[te]?((r-1)*Pe.body[z].keypoints[te].position[0]+Q.position[0])/r:Q.position[0],Pe.body[z].keypoints[te]?((r-1)*Pe.body[z].keypoints[te].position[1]+Q.position[1])/r:Q.position[1]],positionRaw:[Pe.body[z].keypoints[te]?((r-1)*Pe.body[z].keypoints[te].positionRaw[0]+Q.positionRaw[0])/r:Q.position[0],Pe.body[z].keypoints[te]?((r-1)*Pe.body[z].keypoints[te].positionRaw[1]+Q.positionRaw[1])/r:Q.position[1]]})),ee={},ne={connected:{}};((i=(o=t.body)==null?void 0:o.modelPath)==null?void 0:i.includes("efficientpose"))?ne=rb:((c=(l=t.body)==null?void 0:l.modelPath)==null?void 0:c.includes("blazepose"))?ne=Kx:((d=(u=t.body)==null?void 0:u.modelPath)==null?void 0:d.includes("movenet"))&&(ne=$b);for(let[Q,te]of Object.entries(ne.connected)){let oe=[];for(let fe=0;fe<te.length-1;fe++){let be=J.find(Ce=>Ce.part===te[fe]),we=J.find(Ce=>Ce.part===te[fe+1]);be&&we&&be.score>(t.body.minConfidence||0)&&we.score>(t.body.minConfidence||0)&&oe.push([be.position,we.position])}ee[Q]=oe}Pe.body[z]={...e.body[z],box:X,boxRaw:Z,keypoints:J,annotations:ee}}if(!Pe.hand||e.hand.length!==Pe.hand.length)Pe.hand=JSON.parse(JSON.stringify(e.hand));else for(let z=0;z<e.hand.length;z++){let X=e.hand[z].box.map((ne,Q)=>((r-1)*Pe.hand[z].box[Q]+ne)/r),Z=e.hand[z].boxRaw.map((ne,Q)=>((r-1)*Pe.hand[z].boxRaw[Q]+ne)/r);Pe.hand[z].keypoints.length!==e.hand[z].keypoints.length&&(Pe.hand[z].keypoints=e.hand[z].keypoints);let J=e.hand[z].keypoints&&e.hand[z].keypoints.length>0?e.hand[z].keypoints.map((ne,Q)=>ne.map((te,oe)=>((r-1)*(Pe.hand[z].keypoints[Q][oe]||1)+(te||0))/r)):[],ee={};if(Object.keys(Pe.hand[z].annotations).length!==Object.keys(e.hand[z].annotations).length)Pe.hand[z].annotations=e.hand[z].annotations,ee=Pe.hand[z].annotations;else if(e.hand[z].annotations)for(let ne of Object.keys(e.hand[z].annotations))ee[ne]=e.hand[z].annotations[ne]&&e.hand[z].annotations[ne][0]?e.hand[z].annotations[ne].map((Q,te)=>Q.map((oe,fe)=>((r-1)*Pe.hand[z].annotations[ne][te][fe]+oe)/r)):null;Pe.hand[z]={...e.hand[z],box:X,boxRaw:Z,keypoints:J,annotations:ee}}if(!Pe.face||e.face.length!==Pe.face.length)Pe.face=JSON.parse(JSON.stringify(e.face));else for(let z=0;z<e.face.length;z++){let X=e.face[z].box.map((ee,ne)=>((r-1)*Pe.face[z].box[ne]+ee)/r),Z=e.face[z].boxRaw.map((ee,ne)=>((r-1)*Pe.face[z].boxRaw[ne]+ee)/r),J={matrix:[0,0,0,0,0,0,0,0,0],angle:{roll:0,yaw:0,pitch:0},gaze:{bearing:0,strength:0}};J.matrix=(p=e.face[z].rotation)==null?void 0:p.matrix,J.angle={roll:((r-1)*(((f=(h=Pe.face[z].rotation)==null?void 0:h.angle)==null?void 0:f.roll)||0)+(((g=(m=e.face[z].rotation)==null?void 0:m.angle)==null?void 0:g.roll)||0))/r,yaw:((r-1)*(((x=(A=Pe.face[z].rotation)==null?void 0:A.angle)==null?void 0:x.yaw)||0)+(((b=(y=e.face[z].rotation)==null?void 0:y.angle)==null?void 0:b.yaw)||0))/r,pitch:((r-1)*(((k=(w=Pe.face[z].rotation)==null?void 0:w.angle)==null?void 0:k.pitch)||0)+(((N=(C=e.face[z].rotation)==null?void 0:C.angle)==null?void 0:N.pitch)||0))/r},J.gaze={bearing:((r-1)*(((F=(R=Pe.face[z].rotation)==null?void 0:R.gaze)==null?void 0:F.bearing)||0)+(((P=(_=e.face[z].rotation)==null?void 0:_.gaze)==null?void 0:P.bearing)||0))/r,strength:((r-1)*(((M=(T=Pe.face[z].rotation)==null?void 0:T.gaze)==null?void 0:M.strength)||0)+(((j=(U=e.face[z].rotation)==null?void 0:U.gaze)==null?void 0:j.strength)||0))/r},Pe.face[z]={...e.face[z],rotation:J,box:X,boxRaw:Z}}if(!Pe.object||e.object.length!==Pe.object.length)Pe.object=JSON.parse(JSON.stringify(e.object));else for(let z=0;z<e.object.length;z++){let X=e.object[z].box.map((J,ee)=>((r-1)*Pe.object[z].box[ee]+J)/r),Z=e.object[z].boxRaw.map((J,ee)=>((r-1)*Pe.object[z].boxRaw[ee]+J)/r);Pe.object[z]={...e.object[z],box:X,boxRaw:Z}}if(e.persons){let z=e.persons;if(!Pe.persons||z.length!==Pe.persons.length)Pe.persons=JSON.parse(JSON.stringify(z));else for(let X=0;X<z.length;X++)Pe.persons[X].box=z[X].box.map((Z,J)=>((r-1)*Pe.persons[X].box[J]+Z)/r)}e.gesture&&(Pe.gesture=e.gesture);let a=he();return t5=Ae.perfadd?t5+Math.round(a-n):Math.round(a-n),e.performance&&(Pe.performance={...e.performance,interpolate:t5}),Pe}function k0(e,t,n={order:2,multiplier:20}){let s=0;for(let r=0;r<e.length;r++){let a=!n.order||n.order===2?e[r]-t[r]:Math.abs(e[r]-t[r]);s+=!n.order||n.order===2?a*a:a**n.order}return(n.multiplier||20)*s}function OT(e,t,n={order:2,multiplier:20}){let s=k0(e,t,n),r=!n.order||n.order===2?Math.sqrt(s):s**(1/n.order);return Math.max(0,100-r)/100}function MT(e,t,n={order:2,multiplier:20,threshold:0}){if(!Array.isArray(e)||!Array.isArray(t)||e.length<64||t.length===0||e.length!==t[0].length)return{index:-1,distance:Number.POSITIVE_INFINITY,similarity:0};let s=Number.MAX_SAFE_INTEGER,r=-1;for(let a=0;a<t.length;a++){let o=k0(e,t[a],n);if(o<s&&(s=o,r=a),s<(n.threshold||0))break}return s=!n.order||n.order===2?Math.sqrt(s):s**(1/n.order),{index:r,distance:s,similarity:Math.max(0,100-s)/100}}function zT(e,t,n,s,r){var i,l,c,u,d,p,h,f,m,g,A,x,y,b,w,k;let a=0,o=[];for(let C of e){let N={id:a++,face:C,body:null,hands:{left:null,right:null},gestures:[],box:[0,0,0,0]};for(let M of t)C.box[0]>M.box[0]&&C.box[0]<M.box[0]+M.box[2]&&C.box[1]+C.box[3]>M.box[1]&&C.box[1]+C.box[3]<M.box[1]+M.box[3]&&(N.body=M);if(N.body)for(let M of n)M.box[0]+M.box[2]>N.body.box[0]&&M.box[0]+M.box[2]<N.body.box[0]+N.body.box[2]&&M.box[1]+M.box[3]>N.body.box[1]&&M.box[1]+M.box[3]<N.body.box[1]+N.body.box[3]&&N.hands&&(N.hands.left=M),M.box[0]<N.body.box[0]+N.body.box[2]&&M.box[0]>N.body.box[0]&&M.box[1]+M.box[3]>N.body.box[1]&&M.box[1]+M.box[3]<N.body.box[1]+N.body.box[3]&&N.hands&&(N.hands.right=M);for(let M of s)M.face!==void 0&&M.face===C.id?(i=N.gestures)==null||i.push(M):M.iris!==void 0&&M.iris===C.id?(l=N.gestures)==null||l.push(M):M.body!==void 0&&M.body===((c=N.body)==null?void 0:c.id)?(u=N.gestures)==null||u.push(M):M.hand!==void 0&&M.hand===((p=(d=N.hands)==null?void 0:d.left)==null?void 0:p.id)?(h=N.gestures)==null||h.push(M):M.hand!==void 0&&M.hand===((m=(f=N.hands)==null?void 0:f.right)==null?void 0:m.id)&&((g=N.gestures)==null||g.push(M));let R=[],F=[],_=M=>{M&&M.length===4&&(R.push(M[0],M[0]+M[2]),F.push(M[1],M[1]+M[3]))};_((A=N.face)==null?void 0:A.box),_((x=N.body)==null?void 0:x.box),_((b=(y=N.hands)==null?void 0:y.left)==null?void 0:b.box),_((k=(w=N.hands)==null?void 0:w.right)==null?void 0:k.box);let P=Math.min(...R),T=Math.min(...F);N.box=[P,T,Math.max(...R)-P,Math.max(...F)-T],r&&r[1]&&r[2]&&(N.boxRaw=[N.box[0]/r[2],N.box[1]/r[1],N.box[2]/r[2],N.box[3]/r[1]]),o.push(N)}return o}var S0=`
|
|
/9j/4AAQSkZJRgABAQEAYABgAAD/4QBoRXhpZgAATU0AKgAAAAgABAEaAAUAAAABAAAAPgEbAAUA
|
|
AAABAAAARgEoAAMAAAABAAIAAAExAAIAAAARAAAATgAAAAAAAABgAAAAAQAAAGAAAAABcGFpbnQu
|
|
bmV0IDQuMi4xMwAA/9sAQwAGBAUGBQQGBgUGBwcGCAoQCgoJCQoUDg8MEBcUGBgXFBYWGh0lHxob
|
|
IxwWFiAsICMmJykqKRkfLTAtKDAlKCko/9sAQwEHBwcKCAoTCgoTKBoWGigoKCgoKCgoKCgoKCgo
|
|
KCgoKCgoKCgoKCgoKCgoKCgoKCgoKCgoKCgoKCgoKCgoKCgo/8AAEQgBAAEAAwEhAAIRAQMRAf/E
|
|
AB8AAAEFAQEBAQEBAAAAAAAAAAABAgMEBQYHCAkKC//EALUQAAIBAwMCBAMFBQQEAAABfQECAwAE
|
|
EQUSITFBBhNRYQcicRQygZGhCCNCscEVUtHwJDNicoIJChYXGBkaJSYnKCkqNDU2Nzg5OkNERUZH
|
|
SElKU1RVVldYWVpjZGVmZ2hpanN0dXZ3eHl6g4SFhoeIiYqSk5SVlpeYmZqio6Slpqeoqaqys7S1
|
|
tre4ubrCw8TFxsfIycrS09TV1tfY2drh4uPk5ebn6Onq8fLz9PX29/j5+v/EAB8BAAMBAQEBAQEB
|
|
AQEAAAAAAAABAgMEBQYHCAkKC//EALURAAIBAgQEAwQHBQQEAAECdwABAgMRBAUhMQYSQVEHYXET
|
|
IjKBCBRCkaGxwQkjM1LwFWJy0QoWJDThJfEXGBkaJicoKSo1Njc4OTpDREVGR0hJSlNUVVZXWFla
|
|
Y2RlZmdoaWpzdHV2d3h5eoKDhIWGh4iJipKTlJWWl5iZmqKjpKWmp6ipqrKztLW2t7i5usLDxMXG
|
|
x8jJytLT1NXW19jZ2uLj5OXm5+jp6vLz9PX29/j5+v/aAAwDAQACEQMRAD8A+qaKACigApGOKAML
|
|
Xp8xlF5A7V4X8RtYs7PzfNImnx8sa8Kp9z3q2tEgp6angWs62ZZ5CTGoJ6DArGNz5p+UrID6EUrF
|
|
PUlW1EuN0XNW7PQ2L5j3JnoKXN0KijqNP0eYoqXBdgPuuo+ZPeupisWn2Jd4+0r924XgsQOCff3/
|
|
AJ1FzRKxDqGii6m3siiQ8F1XGfXI6YNWLfRbiRQMkcZI9fpTDluT2/h6Qy8gDPbtmtG38JeY480Z
|
|
5zSLUTZg8M28YwYxjAArXtdPt402qgHbpSaLWhma3o0Uqk7Nx9DWLaaVblgPs6qRyds2M/gRSQp9
|
|
zZOni2iWS2hlQ+kjYz9OMGrdjq89vIPPVhj+8M/lQyDq9P1WOYBlMZz1AOD+VdDaTiReOKulK0jO
|
|
tHmi0WDTlr0TyxRVhT8tJjIX+9SUxHXUV553BRQAVBcPhSBTSuxPY86+IGti0s5I7dsORy9fM3i6
|
|
8e8mfDO5P90ZrWWiJicNPpZZtxV/xrW0jQt4DOv6Vk2dEEdTY6BHuB25rpbPSo0QARjP0qTRI17W
|
|
wA/hFaMWmoQMgflQXYsDS142rU9tpqqenfNA7GgtihxkdKuRW6qMY/GkDZY8sY4Ap4hXbyB+VArk
|
|
EtuH4wPyrk/EGkOm+a3jw3suRQLc5i38SX9hJ9nnY+XnBUdPyNdFY6pa3KkkAE9l6f8AfJ/pSJT6
|
|
GhDmI+Zb4ZRycdv6ium0nUhKFydrelTsNnS2829RnrVgV6NKXNG55lWPLIM81Op+WrZkRMfmNNzT
|
|
A7GivPO4KKAEY4XNYWt3vkwPg4OK0giJdjw/xrqhm87Zs8tc7pX5A+leSajf6aHYJ50kn4AZpTep
|
|
rBWRm2Vobm4BXfyehPFdnpmnBFUY5rI2SN63tlToK0YI+KZpFF+3QdavwoKTLtoW0Toaswpk5pCb
|
|
LCxipAhoIuP2dKevHXoaYDylRyxhlwRQI4nxVoCXWZI1GfpXGtbSWjYPGP73+NIGupt6TqMsLruZ
|
|
ih4xnP5V09mQ+JLd8gn0xSYJnVaVdkook69K34zuUGunDS3Rx4qOzHVIp4rrOMY3NJQI7GivPO8K
|
|
KAILt9kZrz3xlebYiu8KCCWb0XvW0NFch6ysfO3jLVjfXLIn+pQkKorl7WxNxIPl71g2dUUdpo+l
|
|
pBGvHPet23iC8ihFosrxirkHQUFo0IF4FXI1O726CpKLacCrMJoJLYHAPpTwucHpSRJJ5e4AZI9x
|
|
UqpxzVpCuOC8cUpQUMRnXttuB4rjNdsYyeVwfXpmpGmcvcQyafMCFJjPY10eg34BUg4DcZP8jUO4
|
|
HaRq3lLNF+IHet7R7jz7c56rwa2wz9+xhiVeFy/T1PFegeaNPWigDsc0ZrzzvDNIaAM7VpNqdegr
|
|
xL4l6kywyRhseZ19lrdfAZL4jxYg3Fw20d63tJsdrDI5rm3Z3R0R0Mce1eKnQYAplIkWrMJ45oZS
|
|
NO3PHbNXIyfpSGWowSOasxLUiZdjFSqtNEMkUemKlAGKsRJjAppFAiORMjmsTVrNZEO4cfSoZSOD
|
|
1eJ7WXBUzQZ+7nkfSo7e2Ei+ZaMzxntjBX2NSU1Y6/wxqojiEFzkA8KTXYaUoWRyv3W5rSjpNHPX
|
|
+BmpSg8V6J5gUUAdhRXnneFFAGHrTfu5PpXzj8S70/aZtxzztXFbv4DKHxHI+H4GZiz9zxXXW8G3
|
|
GBXMjvLRXAx0oPGPSmMVeOnWrMTYpFI0bcg1fh54xmgovRcD3qxETSIZcRvzp+/BpEkqsBUqsM9K
|
|
q4Em4Gkxk0yRGXrVW6i8yFhkg+tJjRxGsWrxllkUMh9eK5uMz6bcebbnfG33kPcVkay2OntPKuo0
|
|
nhXI67c8qa7Lw3c+adjcEDGK1paSRhVV4s6A0or0jyRRQ1AHX0V553hRQBz+vNtt5z3xXzX8Qbdm
|
|
uic5YnOMdK3l8JnTXvlbwpYl+WySOgrp5YfLOOB9O1c62O7qQkc+9RsKChFPWp4DluOlSykaNruH
|
|
ArUgHShFNF2NT1qxGO3NBmyxGcE1N2560CFzjrUysO9JAPDDjFOVuKoQuSRTWouBkazbCa3cd8cV
|
|
wF7IISQccHBzUSWpV9C3o1x5b5GAjdQD1rs9DjC3kckbEhqKfxIzn8LOupRXqnkPccBSkUAzraK8
|
|
87wooA5rxMSI3HqK8B8bQl9Q8sffY5b/AAraXwkUviNrw9pH2W1ViMMRTdRjw4HpWNtDti9TPc4P
|
|
FQs2M5qdyyMHLcfjV63HTAoBGtap0wK0YxigpsuRDtVhVYd6GQydVwwIqdRnqKCR23I5pCMUW6gD
|
|
YNKuetAEise9KTxQBWuFyhrznxNZkXjFeN3I+tTIZg2OqmzmxNF0PO3vXp/g2+hukVl4zyPanTXv
|
|
JmVR+60dpThXpnlPceopWFAbnV0V553hSGgRynjC5FujOey14Ssp1HxNmTnc+a3kvcIpv37HoEYQ
|
|
QmMdVHSsnVbYJF5jVk0dsNzlruVIsl2wKxbjWrVHILjg1CRbZJb+ILHPzyhfStODWLQgFJFYd+el
|
|
UJM27HUIXxhga1Y5lLVLKLkMnoauxnPPrSEx7ShF+Y/n2qrc6xBbhizDAqkK1zJuvG9nbg8ZA681
|
|
ly/Ei052RO3uKAsZlx8QGd8xxvt9Aa1NH8dK7AXMcip64zigdkdrZX8F7EJLdwwNXMkrz1qRMRly
|
|
CK4TxmpidWI49felPYSOMmi80NIoOV6qRzXYeA5SskYPfirpfEjGr8LPWVHyD6U4CvQPL3ZItOYc
|
|
UDOoNFeed4Uhpks4H4iE/Z5MeleMeGULeLgjds10S+BGdL+Jc9OSBU2Huc5Nc74yvUtrcDBrJnZF
|
|
63PJdXvLy/lKWw46bvQVz82jXhkLO5Y+9ZlsYthcRnbIjY9R3q3awTRkEM3WmJI6C0ea3dGRsr1x
|
|
XY6TqW9FLHnjrUs0izpLK5DDjofSta3ckH09KRUkZuuTvFGdvPauE1Y3U6Mqbssf/rUxHPTaJPK2
|
|
ZmJPbBqzY6DCZh5xJC9s9aBJHU6dpemJjfEmfetJtI0+VPkUr/unFOxdiextHs33W07YHQHk11mk
|
|
Xb3KbZ1xIvcd6LEyWho4Nct41sTPYb16ipexCPPZN+wYGCvH1rrPAEJmvkPoc1VL4kZVvgZ6yFwK
|
|
cBXoHkkqinFaVyzo80GuE7WJRQSziPiGdthK5HQV4x4J/wBI8WPIewNdEvgRNL42emO/yj1UHNef
|
|
eNpRczbC+I17DvWT2OqJxc0sMK4TCisy41q0hfEkqj8aixdwTXNOlwvmqD9anS9tXH7uVG+hosO4
|
|
/wC0oOhrR0+6G4YNIEzsNEuCxAPNdjZruA4xxUmjINSjURksOlcbqFykbnjFA1sYGoassaknCqO5
|
|
rl7rxhGm7yBnBxuJq0rkSlYpw+NLlsfd5P8AerVsvHEqSBHwPVgcgVpyMyVXU3rXxcHYETAk+hru
|
|
/DWti6ZSTyOKzZqndHaxvvUGq2rQ+dYyqR24qWI8dvbr7LqDxyDAzXpvw6FvIxePGSM06Xxoyr/A
|
|
zviKFHNegeX1J41zUhXioGbuaSuM6wpCaBHG/EcA6HN/exxXjXw2jL67cv8A3Qa6H8CFR+NnoWpO
|
|
I4XI44rxLxrqjQzSEsQM1gdSPM9U1uR1YbmWIdXHf2rmpIb67YS28UrRlsLI3c/jW0VZGUpO5pW1
|
|
jfLNOjahawzwReYI5cjzMkDavHJ5/SrVv9uhtPtVxCPLBwzxnlT9KGghLU3tKvvPjHzbl7EGuisJ
|
|
GRxWLOg7nRXJEbDjmvSNK+aFSfSoZr0KutRkphc4NcRrdkVjL9aVio7Hk3iqS8ubhrWzUlsZY9kG
|
|
cZNc5D4aee5MclzJIFTzHAO0MfatqSOWu7bFS1srDUZEis0vIZoUxPvfcC+4/dx2xjr712XiTwXb
|
|
WmlQ6hol3cRhoFd4rlg3zY5wR0GelavQwjq7GD4etdVvSnk2wAB+9v8A8mvcfA2kXiRo0/UdcDis
|
|
ZnTTulqeoWqbUAJqWUb42X1FZlnjfjSwlGrr5S/eNdD4RkvLAAQ4yRyaUZcruVKl7TQ9I0G+mnzH
|
|
ckFwM8VuIK7ac3KF2eXiKapz5UWYxipNtMyNejNch0jSar3cjR27uoyQCRVRWom9DxTx54gu5fMi
|
|
lbKdMVjfCZPNlv5v9rFbVHpYqjGzbOn8SzFI9o715L4u0r7arYzk+lYdTqSujy7U/C0u4vHk+WwO
|
|
xuh9q3J9dgvbdVukMV1EwbDDgn04rZMwlHoZ+orZ6hfQ3RWVnQYCgZAq+8U0ln5NtBsV2yxYcfgK
|
|
JtW0CnB31LlroVwJ1nQLGDjeP7w+lb0dsFxjrWB0tHS6NuWPJ6A16ToUm63T3Gallr4S7cxiTjrX
|
|
PaxaF7dlVeSMUhxZ5jd+H7qCa4eF3DSE5x3zXN3Wk6jbyeaiFWUY6ZyPStYS5SalPmVipFbX0E4c
|
|
W0alvmPHJrag0rVvEE6LdljGpG2NRtQD+tW5XMI0uU9M8NeFo9PiQhecDIIrtrOMIoG3H4VlJm9t
|
|
C6CB06VPGM1IHLeItGS6uw+ORT7e3jsbQvj7gzUNam0JaWE+HN7NqOqX80n3FO1RXo8YzXdS+BHk
|
|
4z+KyzGPapcU2YIv7qQtiuaxvcaWqG4O6FwfSrS1JbPnrxoxkv7qIfejcitj4V2f2exumI+8+aKn
|
|
xHTT+G5d8Txlm4rjLxMsQwzWT3OiK0Mm6sEkVsAcjFc1d+FEmlGwEDPQVopaEuOpr6f4ZWNAu3tW
|
|
vHpAj5ZQcUFIWaDjGMVUMQ3cVDBmvbhY7QAV2nh+T/R1yeKhlrY31+b61FcQK6nIoJMi401WblRi
|
|
qr6PCw5UYq9y+YgOgWzNkRrx3xWjp+nx2v3FQcelAbmko9anQ4GBUNisPHWr1qMrQhS2K11HvmYV
|
|
hamcxSRZ5xRIqluS/DKAQQXZxyXrvo2FdlL4EeZjH+/ZbjNSZpswLNBrE1Gt7VE4ODVIlnh/j61F
|
|
j4lmeTGyUbq6LwdEqWbeX0YbhSqfEddP4Bddj4JIrhL5d8h7VjI6oLQqKNzelWre3yc4/ClFjaL6
|
|
wqBxxUUxwCKu5BmXRA6c+9ZjP83FSBoQuPs4BrsNBlUW659KmRrDY6G1lyQtW3Hy0lqQ1qVJnAbm
|
|
oy3b9KYJCqRj3o4zRctIlhjLHmpSuOBRbQOpLGpPFaES7UqkZzKN1KsEc87/AHUUmvPLTVGv72aQ
|
|
k7WJwKmRrQ3ud74Ltilgz4++2a6iNDXdS0gjyMU71my7GpqTbxSbMki3SViajTTHqkSeR/GeyZmg
|
|
nQHkEE1S+F+oPPavBL96I4/Cia1udVF+4dVrkW+Fq8+v4tjMDWUkdVJ6WM0cNV+F+MVmjUcZgqnP
|
|
1qpNNnkcVRLiZtxIS1UzzIF7mghlxUZpVQdq6nTVdAoAOKzkbQWhvwM6gMM1twOJYx3NOJE11Kt1
|
|
H1/pVVlwBkk+9NocXoOQ45FPj+fkUJFF2NSB700v/hTEty5ZpkjvVyUgcCq6GM9zC14/8Se6GcZQ
|
|
1574Xs5WkI2HBPHFQ1dm1KSSZ7Rotn9l0+KPHIHNacae1dy0Vjxaj5ptlhVp+2s2CJ9ppCKzuWNx
|
|
zSFc1SYrHNeNdIGpaYw25ZeRXmvheyk0jVpEdcLJ0q3ZxNKTa0O3vQHg/DNcHrsJDmsmjspnNzNt
|
|
fFIJ24GazOhC+azDmgZIOOKBsp3J2qSaZodubq58yQ4QAnmhGT3NO18pb7BORmu205LfYpyKVkWp
|
|
Oxr5gKYWoIZWgfGfloFq1qTPLubnGO1RPtxg4P0oBAkY/hBz6VNDDkZ6AU0W2WSdqkdKr9ZOaGSj
|
|
VtcLHmnOcgmmYvcz7mBLy3MbdD1q9ouiRK6bUAVeelOC1InPlidSsWMDFOCEdq3uefykqrinYqGy
|
|
rFvApMVka2DAowKAsMkRXQqwyDXn/iWyitNQ3qPl6itIvRoF8RXinW4tQ6HI6GuW8SIVBPalc6qe
|
|
5x9x97r3qruwTjrWZ0ksZ9TUmcDNAmZ9/wAoao63rR0+w22MLPtAzt6mghmfofiB76LdJBJBIp5D
|
|
d/oa7bSdWLIPnpDi9TM8TeKdas51XTbIyxd3J/pXS+E/EFxqNoFu7do5OmD60maHWrnZyDRkn/69
|
|
MlEyOR0xntVoNx+FUgYjPxg4FLCuWDZyKQr2RoRnP0qO+nEFpJITgAUzLqZnhu6+0rknOTXpOmwJ
|
|
Fbrt5yMmnHYyr6Oxb2ijaKLnPYMClwKQWK3n0hn+lachHOJ9pNNN0apQFzsY10a4v4hXQh0xpieQ
|
|
MA1XLZNjhK80cT8OdV+3Wl3A7ZZJCw+hrR1qLcjZ/CsbnfHRnFXseHJArOYYbrUs1uPhYbuatqFP
|
|
ByfSkMq3UIINYkto+87Tx6GkSxfsDbflGD7CtTw/pk4nzITtPIFMFudsukh4Rxz71paTpKwP5jcn
|
|
0qTRy0NORMDgVCqewoJTJgAoxjntTiTu7fWmFxAcnn1q3EPl+X8KZMi4gKqB1Peob/Tv7Us5bfeU
|
|
yOoq4R5nYxqT5I8xieH9J1DTbvyJELRg8ODwa9Ms5mSFV9BWiptbnNVrKdmif7Q1KLg96XIZc5Is
|
|
pNL5pqeUrmMtZs0jzV08phchaY00zH1p2ZNxjS1g+LdJOt6U9ssmxjyGp2urDjLlaZzng/wUPDqz
|
|
TSTmWeTrjpVjVk3Rvjr2rnqQ5dDvo1XUd2cTqSNk9OKxXGCeKxZ1DAxHTr2q5C/y8GokUhsz54qu
|
|
uCxzSQjQ0+FZblR2ro4bZYiMVQ0dBb7Qi5x0qzuG5QOh71LYErDufpSeWrHnimIXbjkUjLkH1Hem
|
|
gGxryc+tXI19KYmWegq9YLiLJ7mtqS945cS7QsWehqxA9dEjz4krPSxyZqbFFhGxUm6smjRM55Lk
|
|
HvSvNxXTY57kLT+9MNwKdhXGm5FIbkU7Bca1wMEVhaiuQcVhXWiZ14R6tHGanGBI2OtYkqEHjgVy
|
|
s9ErEeo6UBsHipKEZs5qpPdRxcbhx70NCSuybTNWihc5brW9Fq6vjMnFSdEIdDRi8RRKygZbHFbu
|
|
m6nb3RA3gMegNJhOm0jbXGOoxTuCc1Rz3FyoGKawz9KaAVcZqeMgCmIkB4FaUTbYwB6V00Fuzixb
|
|
0SFMuDU8Mlbs4UPeXHeiOXkUrDuXYnyKk3cVk0ap6HMxxketSMhrcwRC0dMMZFMQ3yzSeVQAeUaz
|
|
9Vj8uPd271nVV4m+GdpnHX67pCeKyLtBtNcR6xlk9RVeWTb3qRnO6trgttyIfm71z7ai8j7/AJmN
|
|
DNqUVa5Yi1AnjynHuBV+11YJhWWXcP8AZNSzqgmaEerSsf3NtIQP4mGKtRavdRgMIpVI9KjU0a7n
|
|
R6T43uYQI7qN2Tpkqciu503VVuQGAYZHQjFVc4alPlZrpKGAznpTwxOc9+lWjIlUACnM4XApiLNk
|
|
nmvnsK0NvpXZRVonmYqV52GsmanhXitTmFkSiJTSAvwrxUxXIrJ7miOfjf1pzNWxkRlqYWpgJupu
|
|
6gQbuahvIxPA6eo4pNXVioS5WmefakGhndH4INZs5DJXA10PaTurmLO21uKpSZqGMoXGnRzBiyjd
|
|
9Kx5rcQS428fSkjanLoaOliHGZFB56VswW+mtPufcBsGOAfmxz+tFkd8HpoaUx09FAtFY8DO71qb
|
|
Sms/Nb7RbecG6AEjFLS5c78t+p0djpVs9wsyQiJAdyr1rW+zqjErzSe559Sbk9S3C+MA1bjbgE1S
|
|
MSXzMVG0vNUI2tPKrAuCMnrVzNd0PhR49W/O2xrHmp4TxVMzQshpIzzQBehqesnuaI5VGzT2bitz
|
|
FEbNTC1ADS1JupgG6l3UAc14s04yR/aYRll+8BXCtLncDXFWjys9TCz5oW7GddH5qqNzWDOgQnC8
|
|
VSuo1kHzAGkPYopEY2+RWxV23Vzj5G/Kg3jWaNazhZuqNXS6TaKhB2c0jR1nJWOlhOxRxU4YkCgx
|
|
Y0OQatQyDbyaaFYe8uF4NY3iC9ltbVGj43NTIL3h7WzMihjzXVQXYYDdW9Cf2WcOJpfaRZ3g9KsQ
|
|
mupnCLIabGeaAL0LcVY3cVmzRHIxtUhetzEjZqjLUAIWpN1ArhupwagAfDKQ3Q1594v0c2bm6tx+
|
|
5Y8j+6ayrR5onThp8s7dzkZjuqAAmuBnqC7c0iwgtzSA0rWzjfGRW3ZadDu4AoNYo2rfS4v7orSh
|
|
05UA2r0pDbsTm29KRottBNyJ0wpJ9KhD7f6U0ikNWffIFBz60zVUW52ow4UcUN6EPcx44WsbgOmd
|
|
ua7TT5Bd24KHnFKnLlZFSN4koluLdueRWvp14swweG9DXoxldHlTjYtzGoo25qzEvwtUxas2jRPQ
|
|
5CNqkLVsYoYzUzdQA3dSFqBBmnqaBhuqhriCXTpVIzxUz+Fl03aSPI9QTypW2/dz0qKNw3SvOPZR
|
|
Mqin8VLKRcs3O4Cuk0w/MDjt1NBtHY6O2IIHY1pxgFaETIRwMkjtVSUEk4570MlFW5bap6dKzWm8
|
|
1tqH8aY+hp2FvGoGayNevVt7/ap4xzUvYjqTLtvLPcvJxSaVcyWsxTnFZlnT2t15xHmCtOBYwQy4
|
|
B9q7cPO+jPPxFO2qLEj5HWo42+aus4HpoX4W4FTF+KlotbHII9SFuK0MUNZqiLUDE3UbqBBupwag
|
|
Bc1DefPbyD/ZND2KjujyPWlKzuPesRZjHJXms9lMuw3StjnmphKDSLTJ7OfE3JrpbO4GQc9qlnRA
|
|
3LO82k5NbFvdADkjBoCSHyXIIIzgVQvdRigT7wzjgUzO1jHknlvG7qnp61etYFQDIpCZoqVijzXn
|
|
3iC8EmsOuaCGb/heR/s0ijkVv6fbxy3QMg5xmsnuX0Ldzut3+UYTPWk+2GJSe+M1pFtamcldalmx
|
|
1eO4XaThhWnC+TXqR2PHqL3maUJ4qRjxSEjj42qXdxVmaGs1MJoATfSbqBAG5p6mgAzTJTmNvpQU
|
|
tzzHXY83D/U1zF5FhjgV5r3Pa6FMsV5HWnLe7RhqBRdmTwagN2d2K2rPU1C5LAnPrUs6Iysbdrq6
|
|
f3gK0BrUKj/WClY05iM6xLOcQAj3NT29uznfKSzHuadzNu7NSBFjHNSm5VO9IRnajqoWMhTzXFtA
|
|
bvUfMduSeg702Qz0rS7FbTToQFwzjJqaGTFyfK5PQViyzUuFmuIdgGABya5u/vTaN5cnUHFUmLoZ
|
|
zyskwlgJweSK6zQdUEwVJeGr0aUrxPLxEfe0OrhPAqVjxWhznGRtUwatDK4jNxURbmkAm6jNABup
|
|
6tQAFqhupNtu59qUnZFwV5JHnWsHdIx96w5lz15rzT2uhRmt85xWbcxMnUGmZlB0bdxmrNvFIcfM
|
|
350mWjbs7YkDJY/jW5ZWW4jikWkdNp9mqYJFaJdEHHakUULu/VB1rLn1Ld/FgetMGYd/qWSQmSa0
|
|
/AemS32pfa7piLeLkg9z6UmQtz0W7uQ2cZx0A9BVzR7cAea6j2rPqX0L99KRat5A6Dk1wOoKZ52a
|
|
YfMORTYRLujiGWEq6/NWza2yKQVHNdOHerRy4laJo6TTnbbtb8KuM3Fdh5z3OJjbmpt3FaMxAtUZ
|
|
agBN1GaQBzTwaAAms3VbjERUGsa07RsdeFpuUuY4jUjljWTKK4j02RE4IpJYFk6imQkVl0xWarsO
|
|
mAEcUi0bNnZBR0rWtoguMCkUi21wI161mXuocEKaYXMS4u+pY/hVCSWSY4HT0pEmlouiSahdpEBl
|
|
mOceleiwWcNjClvHgJH97Hc1EmVFFi3Czy7mwIl/WtJbjP7uLgd/apQ2VNVvtsBhiPzdK5S4nAuR
|
|
nqOCaTGi9pcytPlU+XpmumtWII44rah8ZjiNIXRuWeNvvViQ/LXpJWPJbu7nCRvVkNxVsxBmqJmo
|
|
EPiXca0YLMuOlJsuKuPlsSi5IrNuG8s4HWs5VEkbwoOTKsk+FJY4rC1K53k1xTk5O7PSpwVNWRzt
|
|
4cms+WpKICtSLTETQj5q0YeBSGiys23pUguGxQMq3E59ayrm4x3yaAKiRtO2WPHcmhruKFxFajzZ
|
|
ScA44qRHoXhuMaLpxaUg6hcDLMf4F9KlhuDeXGASIl+8azZslYma68y48m1+7nFW5rtbRNhb5z1p
|
|
iMKbUg0zuW4A4rPgb7VdKXOMmpA7HRbMS7nUYiUda0lkQOBngVrS+JGdbWLRt2bAx5BqeQ/LXpnj
|
|
PQ4GJ+ashuK0MhWaoWcA0AaOmASMK7jRNPWYBmHyiuepO2x10qfcv6vYxCzYqoGK4HVYVTJrmb5l
|
|
c6oaM5TUJ8EgGsG4kLNUHT0M64OaqMMikSRsuKbnFMRLG3zVehOaGNE445NNlnVFpDMu6uie9Vo1
|
|
8z5mOAOST2pDK91cNN+5tsrH3PrW54a06KxT7fdrlh/q1Pc+tJ6IUdZGvHPLezMcnBOWbsPap5r3
|
|
ylFtbdT1xUWNWzU0/Zbwlgfmx8zGsHWtRHmMqE59aAMyNifvHPc1f0gtPdqkY5JosJHeNci2tktY
|
|
euPnNY+oXWZEVJNrZ9aun8SIq/CzodHuriIokhDIR1ronbKZr0o6o8ipoz//2Q==`,I0=`
|
|
/9j/4AAQSkZJRgABAQAAAQABAAD/2wBDAAsICAoIBwsKCQoNDAsNERwSEQ8PESIZGhQcKSQrKigk
|
|
JyctMkA3LTA9MCcnOEw5PUNFSElIKzZPVU5GVEBHSEX/2wBDAQwNDREPESESEiFFLicuRUVFRUVF
|
|
RUVFRUVFRUVFRUVFRUVFRUVFRUVFRUVFRUVFRUVFRUVFRUVFRUVFRUVFRUX/wAARCASwBLADASIA
|
|
AhEBAxEB/8QAGwABAAIDAQEAAAAAAAAAAAAAAAEDAgQFBgf/xABDEAEAAgECBAMECQIDBgUFAQAA
|
|
AQIDBBEFEiExE0FRBiJhcRQjMkJSgZGhsWLBJDNyFSVTY3OSNEPR4fAHFjWCokT/xAAYAQEAAwEA
|
|
AAAAAAAAAAAAAAAAAQIDBP/EACARAQEBAQADAQEBAQEBAAAAAAABAhEDITFBEjJRIhP/2gAMAwEA
|
|
AhEDEQA/APqYAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAKNTq8OkxzfNkisQC8eb1XtRNbzXT4q7eU2nu0MntRq/D8StMccvW29ZmdvgjsTyvZjxOLj
|
|
+s8WLxn8TFPXs6Oj9oct7c14rkxz22nrB2I49KOdTjelmszfmpMeUxv/AA28OqwZ4icWWtt/SUi4
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAmdo3nsPNe0Pt
|
|
Fh09Z0+DNWL7+9O/7A3eJcZppsV5raI27esvH6jX5ddM25p79Ilo59VbUZOe2Tm/PeGvfPfT2iKR
|
|
PLv1+DO678XmW/a97U6TtOyzTbTF538/T9WjTNecm9a7126tqk3rSYxY5ta1plRZqZNXGjyZcPXl
|
|
mZmsx+qjBrsuO16xM7eXRt04JrdTltk5OWJnfaWf0a2lty5MdZnfzSn+WOHiOutFpjHa9e8bQ2fp
|
|
+alYy462pk7zXbuxjPesbRS0f6ZZV1ET1tErzXFLHo+A+1ddZf6NrI8PJHa1vN6iJi0bxMTHwfOa
|
|
zhzd61v1846utwniM6DUdb3nBaNrVmd9vjC/ZVePYirBqMWppz4rxaPgtEAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAItaK1m09ojcHnvarjM8P0vh49+a/eY8ng9D
|
|
h1fGM1rxjtGPfvbzdbjuTJxHX48cTPNltM/KsS9Dw7S49Jp6UpHaGe2vjz1y9J7LYK13vHWe7bj2
|
|
ex1tvM80ekuxW3RnW3Vm6P5jRx8H0+OYmMcb+bapo8GKPdpC6bQwtdHU8JpWkdJ/JweL6e23iU67
|
|
d4dubSqyVi9Zi0bwIs68XGp36TtEq7ZJmZmevzdbifCKWtbJinkt6eTgZPFw32t+sRurbWVzxs1y
|
|
Rv6T8V1NZNPtfq0seTm+Kevr+SZuxXjvaPiV8N4viycto9HseG6+uu08W6Rkj7UPmFck1tE1nlmP
|
|
Ld3eA8V8HVVi1pjq6Ma/pnqce/ERMTETHaUrKgAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAADW19+TQ5p/p2bLS4v04Zmt5VjeQeJ4bjnLqsupv+Ka1+ERLv4reTmcNxcuC
|
|
vy3l0qdI2hlr66sT02ot0ZV7qqrInruzrVZLGSZ37JjqgYTG0K5lbaFVhDT1Ub456RPweY4hixWi
|
|
eSdpjvD1eWejz3FNHWYtkpvFo9EIseb3tS3SerOms22rfpPqZKzvvHSYUz70TExG6Gdbs2rljeJ/
|
|
Mx5L0vEzPaelnOi98c9J2bFNTFpit47+a+PVUvx9T9nOIfT+GV5p3yY/ds67wvsXqpxau+G09Lx+
|
|
r3TqrEAQAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAADV4ljnLw3U0jvO
|
|
O0fs2lWqyUw6XLkyfYrWZkHldBEV09eveG3Fq1mI3jd4vPrOIaid8G9MP3Y38k6fNrt/rMk9Ou8s
|
|
tfXXn49rGWInuy8SO/k5Gl1E3rG/fzbOe94wTy99mbRvTrMOOvNfJWsesywniukrG/jU6fF43WYN
|
|
TmtEeJtEQ06aSmK2+bNtEd+qfSO17unF9Hmvy1y13XWyVmN4tExLxVK8PmNq5NrT58zawam+m/yc
|
|
0Xj8NpRYSvQZ7xEOdqI3rPozxayNRXe0ct/ON03jmrKB5nV4q1yTO20Obmv4c+cx8HoeI6WZpNoj
|
|
q83niYmYscU0r8aJ6T1n49zeJ+Meqm1drb9J+Kd5p136StGVem9l9TbHxLDFp7W7+sS+q1nesT6w
|
|
+PcAzVjiGHftzQ+v4f8AJpv6On8jH9ZgIAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAABp8VrW/C9TW0ztOO3b5Nxp8VmI4bn37TWYB8f1HFtTfUfR9FWJmsdZ9I7MtJxDX5s
|
|
d8ta1y0xzteaR2277rcuhycP12SceLxMeWNpjttHwlu8I0mfQ1y+D7k5YmJmY36T36Ka43z/AF1t
|
|
cI1ds+qxVj7/AEej19PCw9HJ4NoK4OIU5Y35YmZdzVTGebVZabx5jJS+Tmns81rNLm1Wrzc9rVw4
|
|
Yibbem72mXTTS0w0M3BvEta1bWrM95ie5EanY87wXgNOL6XPfxraXLhra/W28bR/dzYzarBqJxRe
|
|
bzE7Rt5vWU9n8mPHOGmS0Ypnea1naJb+k9ncNLR7u2y/WcxXO4TOoyUrN6zD0FaW5Y3hu49FiwUi
|
|
KxCvLMR0hlW0jn6ukWw3iXjOJzbDlneOj3GaN6zDzfFOH+LE7SRGo83XNSZ2lbG2/WfdlvaT2cy6
|
|
rNFInlrv1mfJ37cK4PwTTxOoidRm2+/2/KFuyMp47XB4LivXiunrH2b2iH2qn2K/J8x4fGDNxTSZ
|
|
9Nh8OviRvTyfT6xtWI+DeXs9MNZubypASqAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAOZx6/LoOWPvWiHTcf2hiZ0e8fc2mf1E5+vP/AEeuSd7RC2uKtI6QjHfeINTfwtPf
|
|
Jvty9WPfbt/lucP03gxfJf7d/wBoReYpm97zaNeLb4Ims9Nt94auDjem1Wo5PFi1onylS+1o7l8V
|
|
bxvtupjDMdNkYtXS1+Stt+m63xImEJ4xjHER2ZxMUjeUTO3VRmydBbjLJqPi08mbeVOXJPq1sl5Q
|
|
Vbkz9+rRy35rxHqzmZlVEe/Ez5LRlW5iyfR6zffaIjq1OSNZps2a21rZInafSPJhxGMl9LStLRWM
|
|
lorM/A4dkrWbYfLZC2W/7K6eubX6b4RzT+W76K8b7G6X62cu3Sten59nsm3j+OXz3/0ANGIAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA0OIYfpOHPijvNNo+fdvtXJO18k/
|
|
/OwPFYbz2ls3jx8VqW6xMdWPEdP9D4lkx/dt79flLLHbkxTPwY6nt2512ORTRzE2x4/dpE7cvkme
|
|
E4IrW3hRMxO8THRtU1FKWtvtvK2upx22rzRCtXkqzh2jtF7ZbT122b01ndnpuWuP3Z3+Ky20qDVv
|
|
fauzVy3mejZzNK8dVjqi87KLRLYtXruqvXzkQp7Qoid88R6rcl+WGlW0/Sa22mfhCZOq2x082ix6
|
|
jkm822pO8VrPdr4dNObVeDo8XW3uzMbzK+mvxT7szE27cvnu9j7PcNjSaXx8mOIzZevbrEeic5tN
|
|
+SZnpt8J4fHD9HXHO3PPW0x/DeBtJxx29vaAJQAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAKNRim9Z5e89Nl4DzXtVh5babURHrSf7f3ec1+qnDorWrvvt5Pccb0n0zhmWk
|
|
Rvevv1+cPE2rGTFNZU26PFfxwa5dVkjelI2772nZnX6bbrEUq3o0d678u8wmuDL2ittvVjXdneeK
|
|
cGv4jpJ6U56+kS7+j118+GLXpakzHaWlp9NNY3tv+bbiYiNoQy1y30uyZJlrWmZnuym6q1iIJnop
|
|
yW2Te8bdWnnypQqzZOadokiIpSZntWN5lrxki19vNRxrUeBwnNNd+fJEY6/OejXLn3Xe/wDp9wyn
|
|
E8uo4lqqxblv7lJ26T6vpD5X7G8QycKzeBMbzMRM1/FH/wA/h9QwZ6ajDXLitvWzRgsAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAeL45w+dDrZvWv1OWd4+E+j2jX
|
|
12jx67TWw5Y6T2nzifU+rZ1y9eHwzDYxxEy18+DJodXfT5o96vafWPVbjyxDn1OOzHudbM0rt2UW
|
|
iI69mVtRXZq5tREb9VUoy2iIlRbJ0UX1VZ6btTLrI7V6yk62M2oisT1c7JmtkttVMUyZp6x0beDS
|
|
RWOvdKijDimvWd3G9pNRMfRcNfvZOb9Hpb0itJeP47k/3hgjaZnbaP1XxWW3T0movbNS0W645nbf
|
|
0nrMPpXs3xamoxdJiLbe/X1n8Uf3fKsOTw4jbaXo+EarJhtGTHMxeJ6xH7Sti9Zaj6x3HM4NxXFx
|
|
DS1mtoi8dJrv2l011QAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AGjxLhODieOIye7kr9m8d4eM4to9RwjPXFa0ZIvG9bR0fQXmPbDFvTTZPOJmEWS/V8bs9R43NxLL
|
|
G8eFbePg1bajU5/s0l1ceKLx1hbjwRE9mOpx0y2uRTSZsm3PMw2aaKtIjo6kYo9EXpET0hVLXxYK
|
|
xC6MZvyx1lFs0RHfaPiCnU12pLyHGNDbUajBekWma2npWN3p8+opa20e9LSyZLxExTlpM+vdOdcZ
|
|
a9tPS8MyUvFrzWlI6727u1pYxYrbVmb7x+TQx6au3Nqcl7/0rcmW9axGnwZJj1novmxnZXV0fFp4
|
|
ZxLBPgTGK8xzXr5fOH0bFlpmxVyY7Rato3iYfNuG2x56Wrqa8s2jz+7Lu8O12bS6jkwzN6THNNI6
|
|
tvrN68Y4rxlx1vHa0bskAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAA4XtTTm0OKfTJ/aXdcL2pyRGjwU362yb7fkJz9eTxxyZJjyltRXzUZK7TFtl9Lbwy06YzrHwa+
|
|
fJFd/wCVt8m0bQ0eS2qzcm+1K/an+zNZFL5M1pjFXeI72ky48eGnPkvNp27+TPU6nHpMfLXaIjpE
|
|
erk5dRMxOfN1mPeisfshW1ne1a1577Y6x5R3U0zze31FOWI6ze0byU098kRlzbxM9qrMlPDpyRMR
|
|
Md5Vt/Ihp5898mWZm1pjftE91uCt7fCI7dWeHDEW3t723l6rslqxWZnasR+SYhFbzhnfxJ2jyeq9
|
|
lcGXWZcmW0zWKxHLaI7794eJx5fpfEKabT8t8l5isddo3l9S4VjrwrRUwzSJt3tav3pdOL6Y6dXD
|
|
j8HFWm+/KsU4NRXPvtWazHquWVAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAa+fXYNP9u8b+kdZBsDkZOO135cWOZn4y5Wu4xqctbe9y19Kp4njt6vi+PDm8DFMWybbzPlV
|
|
5PiGtz67UxbNbeKTtWIjaIXYpnwuaftT5tXJT3vmi1pMsrU5qIrG1V1a+5DCa7b9GFbRr5J6Wnbt
|
|
Cu+Wmk0m8956z8ZWZNorbfzcbX5rZslazPux3hUt41NTntktObJ13+zX1bek01r4/HzVm0bxPXy/
|
|
+bNfDgjVa2uOY92kdfg6ufJOKvLXtttVVSqbcta2vM7zXtHpLQy5ZtMd+vWd+7Zy3mdJHXra3f0c
|
|
vUarw7zFY5rT2hH1Lavnrgx81p3U49Pk4nE5L35MO/StfNRXR5tXnrS8W67WvfyiPSPi7uLHFK1p
|
|
jrtSsbR5Lc4RzsXBaYreP4l45esRD2HD9fnw6evvWvO3Tfr0aGk0U55ra0TFInv6uzgrXFXlx0i0
|
|
77RPlC83Yj+JW7oddqr6vHzTTw9/f6dod+L1t9m0T8pcbFSmPHER3892W0zPuz+jSbVvidkcqmfP
|
|
Sel7bekrI4n4dZnPWIrHeYnZee2Wpy8dEaml4npNZblw5qzb8M9JbYgAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAABEzFYmZnaI7yCXL1XGa0jJXT0571nbee27DiXEprp8nhbxG20W8
|
|
5cbD0ikfnKO+urTPvjoZdXqctdsmTaPSvRpWmsdZ6yztfaGplvv3lWW1tyRlz1x0vkn7Vo5atTNe
|
|
Y0+1o79V2KsZsvX7Ne5mwxnyTNvsx2iGneM/rCdRSuOsTasTt5kRFtpjqmOH4t4nk7estiMNa97R
|
|
Hwhna0iuKTEdmGWa4672nZtRele1N59Zlq6vLOSsYorEc07qcW65euzRvtXvPZy52naZ7ujr6fXV
|
|
rWdukREK8+njHgmZmPc67bq6ivVWhxxgxZLztNrT1mZ/SP4VZs0zaOvfp84WUtNsXLvtv3699+rU
|
|
z7+Jtt5qURqMnPpctaR1rMSw4ZoK57eNk6xHaJRh97Ltt7lo5Z+L1HAPZvVauZ2nFTSzMTzeJEz8
|
|
to6xPfvsZntPZ9rXxabmxzefdrv0j1dXh/BcmstW1qxTHHasR3+b0GPhGl+kWmd64dNEVjf73T7X
|
|
y8vy+Ddx6O3iRakxTH5RXrMw1/lX+3Itw2MFIraN48qRHdZi0cUjmmPen9noox1iO0fNzdXEYrTt
|
|
stcmd9aX0bJ+HePmiKTitO8TMLZ1cVjrMfqpz6ys4pjfrPRWZ9rXXptUit6zO+23VyaRHEc05L1/
|
|
w9J9ys/en1ljqdVbwYw452tlnl3jyjzbmmiMeKtYjpEbLeTXPUU8ee/+qjJpsV5rbkrFqzE1tEbT
|
|
DpYNbW21Mnu29fKWna0KbqTdjXXjld0cvQ63ltGHNPSfs2n+HUbS9c2s2UASqAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAOVxPWe99HpP8ArmP4b+r1EabT3yT3iOkesvMVtN7za07zad5l
|
|
XV5GmM9vVfEstvDx0jtaVVMlq+UJ18b5cMRvPeSuK87bUt+i2Z3PtG7zXpjkzXt6R+TXyTMzvM7t
|
|
ydHqZ+zhv1+Cv/ZuqvPTHMfOYaTMil1a1K2vHSLTELq2v+KWzThGo84rH5rq8JzedqR+ZeI7WnOS
|
|
34pYTafWXR/2Pln/AMyrKOCWnvmiPyR6O1y9585lhWJvl557Q6eo4T4dYiMvW3b3UanhldHpJtGX
|
|
e09unmjsT7eb1l4trI2t0hsZfrdNO0bzy+nzU20/+NmkzO9esz+TZxWis9dttvPv+Tn21jjaW8zn
|
|
26bTG3mp1M/Wzv3t0jyWXiKZJmsTERaZhXXDbNl8WaztWenxZLstPp5pau8frDtVrNMM5cfTfpMf
|
|
3aunxxbes9d/R09Dp8ebJi09ptFr3jtt2WyrW9wy1Jx132mK+Xq9PotT0iIU19ntLtExa3T47T+q
|
|
6nBaYvsZstZ+cT/LeMnUi0TXffo1s2m8Ws2/OIMWk5Jib5L328rS2t94Sh5TV4ppklpW6PT6rh+P
|
|
NbebTHyas8E081mZy5P2W6OFhjxNTE/hr/LoRO0Kvo9dPqctKzMxEx1la5t3tdnjnMs4noievcrO
|
|
yZjeFF1OSnNV0OG62cn1GWffj7Mz5w05joovzY7xes7TE7w0xrjPeex6Ua+j1UarBFu1o6Wj0lsN
|
|
3JfQAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAACrU5o0+nvlt92P3BxuM6nxNRGCs+7Tv8
|
|
2hToxm1r3m9utrTvMsonqyt7XTmcja0u3O6FMfi5t/u0/lzdJM81p9O3zdvHTwsUR5+bfPqOfX1h
|
|
dqV+3O7bs1+T31oqmI3TEM4rvCdkDGIIhlFd2daboS0NXG2bD6bufxXU1vlmu/u4us/N0+L1tTSx
|
|
kr9qk7w89j1FNZMV3jxLzvaJ8mer+LSOZqK2xZotbvljfr/89U453rXt9lse081xZtNjx7TGKu0t
|
|
DHlrevSevaN5Y6+tJ8c7VRNMt63n3ub+6/R54rERMztDYy4a5omclYmfxKcenrjtHLvtPrCnVmdb
|
|
eFe3JXmjy6eS/DrMuLVYsta9Mdt++6qLxO+0dEc8UmInr18iUfReHcXrqccb9Z27Q61Lb13eJ9nc
|
|
1Z35rTvE9avY4bTkpG8xEfB05vYxqybc07R281naGMREdoT5JQqy9mply7Q3bV3iXG1eXw7TWSka
|
|
c258t7+tpT5/BjT7MfHqndz12Z+M4lMMKyziUJJiN1WSu9fku23RaOgKNJqbaTU1t9yelo+D0cTE
|
|
xEx1iXmM1Nt3W4PqvFweDaffx9vjDbGvxz+TP66QDRiAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAOJxzU73rp6z296zsZMkYsdr2naKxvLyObNOfNfJbvad1dXkaeOdpvsc2yuZVzfbfqybutwu
|
|
s5s8R92J3dvJb3tnO4HSMegtmt3nfZvYp8SZl0z45NfSK7onH1bNcfRFqnUKJr0Y7dVtq7prjEsK
|
|
0XVpEM6028mW20IHK41aPo3J6zs4ODhdcvPnvExFevNXpMOrxi/PlrTee7PLX6Pwa09uaNlKtHg9
|
|
dM3z5d7ReOu02nu0JzZMfblrv5R5uvrcdImZ26T1mYhxs1Os7RH93PZ7axuafNfLitvbaYU3yZYt
|
|
PXs9NwHhui1HBa5LVicsb81onrEuVqNNSuS8Y67dZ6xPZa59Il9uX41vEitImZme3q2Kxbxora0T
|
|
Md/ROSa4Ztkj7c9OafL5LuGYubmyX3iu/TfbdSfVnpvZLT/XZK233+Mbbva1xRXyiPk8pwbH4N6T
|
|
adq5a71n0tD1WDL4tPe6Xr0tDpz8YVnJHWEXYxbqlBedoef4tW0XraO09HdyztSZcbUz43C+ee9b
|
|
SVMaeOfqq7+jGckQ1Yz7+7v2RN/WXPXZPjci2+2yyJaVMuy+uSJlA2d+pNoVRbeDcSxyTE+TDDlt
|
|
pdRXLTynrHrDOyiyZeVFnY9TjvXJjres71tG8MnJ4Nqt4tp7T1jrV1nRL1x2cvABKAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAHJ49qfD09cNZ97JPX5PPw2uI6j6Vrsl/ux7tfk1mWr7dOM8iLdm
|
|
vfebREefRsWldw7SxqNbWbR7lPesrn3Vteo7dYjDpMGCvfbeXQ0uLlxRLRxROfUc34p6fCHYrXlr
|
|
EejqrjY8uzCYW7MZjdVKqK9VlaxCYrsnYExBMRMJRPZA8/xPHtmpP9W2xx76vhWOInvt/C7ike7N
|
|
vwzE9kcapGfhlevTaFbFo8RqJ5vy8/RoW09ek0msxHfp3dzNoLzp4zUmZpMbT8HJyYJi20X2n0lh
|
|
ZY1li/RaidBF4w2mK3jrHaFGp1lN+tptPp5IjBkid5mIp16TKu0abBPv33vPlM7z+iPdFNcWXU5I
|
|
tkrNce/b1W5db1nTaf3ax9q0fxDW1ebNk2phty1mOu09VOm8W19orEz23j1TwfSeERFuEYMddptW
|
|
d43dvBn21eKJ75KbW+cf/JcTgMxXTb3nbljz+TpcPmc2uyZO1KRtVtGVdi0bx07qJnllsRO6rNTe
|
|
N4XVamsy8mnvPwc3R2jPwe8TPbdlxXNOPSZfhWWpwO85OFzv57qrODkzeHntSe8Sn6Rv0a3EZ218
|
|
8nXekfr1a0ZLVnqx19dWb6demXybOO7lYMvNMdW9S/VVLo0us7tPHdtUtEwJiZU3jq2Jhham8CVG
|
|
PNODNTJXvWd3qcWSubFXJWd4tG8PK3pPd1OB6veLaa89Y61/u2xfxh5c/rsgNHOAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAANLimq+i6O0xPv392rdeZ4rq/pOqnlnelOkIt5F8Z7Wj27I2I6sb25YY
|
|
V1ImY3dbQ08LRc23vZp2j5OJG+XJWle9p2h6HHtbJXFT7OOIpX+7TxT31j5rycdTh+Dpz+XaG/sw
|
|
w18PHWseULN2trBE9UcrJKBhFU7JAQi0dEomegNDUYovM7x3jb5tO1ZvpbaTLtzRExWfWPJ08kbT
|
|
Ex5NXWYYyV5omYtHWJieyeDzuizfRs19Jn6TM7Ru1uMcJxZqTkw+5f4ebqa7SV1MR4tdrx2vEfy1
|
|
axqsNOTLjnLXytVXi3Xj8+nmsxTLM16d5npPyUzpekTtSK+U7vS6vQ/SYmK1vWPS1HOn2dvvvvE/
|
|
tDO5XlcO+LbfHSd/W3o6/BdDOXPTnj3Kz38rS6Wm4FNrRyRzTH3p6RH/AKvR8L4dXSzE3jmtHn5I
|
|
mbfqLV+m4dbLSsZInHjr3iI6zLpYaxS01rHuxHRHiT9mv6s67Vj1aqL6326MrWiYa+/Q54BxPaGe
|
|
XRZpj8MquB4+Xg8zPnB7SX30to379GxpK1xcHiKz5IS8xr8PLPixH2bftLTy05o6dHYyVjLhy0t1
|
|
izjZa3pMVv3iO/qz1G2L+NbSajbNyW7xLsY8kTDz+fJXFqKZN4iZnafi6WHL0iYlStI7OO+7axW2
|
|
crFl7dW9jvE9ULN+J3ZbdFGOy+AYWpEqN7afNXLj+1Wd23KrJVMvCzseh0+auow1yU7WhY4fCdV4
|
|
OadPefcvPuz6S7jol649Tl4AJVAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAV581NPhtkvO0R+4NPi2
|
|
r8DB4dJ9+/7Q83Po2NTqLanNbLfvPaPSFDHV66sZ5ET0hRknyW2lTtMyouz0c8usx2n7s7vScKwx
|
|
zc1vu/y85p+maJh6Th+SOWeveXR4/wDLm8v+nX5mUWa9bbrInolmu5jdTNkxYFk2Isr3TuCzeGMz
|
|
+THdEyDDJO9Ja823rt2XWnya946pGvktDXta0ztWu/ybvLE9dkcoOf4GbJPWK1j49VmLh9JtE33v
|
|
Mevb9G7WsW8l1ccREISophiJ2jpDYpijbaOjOuOJ8ujOdqxsgVcsUjaETYvbaFFrgu5lVsm0yUtu
|
|
ryg43H5m+GIj1XcJzePoL4pnrWGtxmfchr8JvfHS1622if3QljzTTLes+qrNjrkiYtCzPMxnm095
|
|
YZJ6boS5teB49Tqscza97VtvWvlv8V/FOF34RrIxTM2xXjelp/eHoeA6XnzReY3ivX/0dfivDcfE
|
|
9HbDbaLx1pb0lOs+jO7K8Lis3cN+0NKcd9PmthzV5clJ2mF9J9GHHVL108dm1SznYr/Ft0tuhLb8
|
|
mNohFbMhLWy0mJ3rPXvDvcO1karBG8/WV6Wj+7kWrvDDBlvpdRGSnbzj1hpjX4z8mOx6UYYstc2O
|
|
uSk71tG7Ns5AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAACZ2jeXneJ62dVl5KT9VTt8Z9W9xbWclPo+O
|
|
fft9qfSHEU1pv48ftYST23ZTDC/p0YtlVuvVjMbM5+LCZjYGWGdrTPxiHY4ffaf3cjTxz1v6xMS6
|
|
Olty2iXVj/Dk8n+ndrkhnGRo1v8AFdW3RCrZ5uiYsqrboncSu508yjmZRYQt50TfowYTbYGVrKrT
|
|
uTZjvukQnYhMIGVY2ZxPVWyrHVCWzXpVXkt3TE7Va+W4K7X3jv1auTNy3jdba0RZpamfroQN7Hk3
|
|
6wr1GTaN2OOJiu6Mu98NvgDi8Wy74d/yZ8PiPAiO2zU4nb6qIn1bugjfFE/ASp1ke9u15mbbRDZ1
|
|
Mb823kx0Ontn1OOkedoJCvT8I03gaKsz9q/WW+isRWsVjtHRKyrhe0XCfpWL6Vgr9fjjrEfeh5fF
|
|
feH0V5Dj3DPoOo+k4a/U5J6xH3ZZ7z3228evytOk7NvFbo0cdols47bSybt7HbddHVqUs2aW3Qnq
|
|
xVeu8LILR3SlZw3V/R8nhXn6u0/pLuPMXjeHT4Zruf6jLPvR9mZ8/g1xrvpz+TH7HUAaMAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAABRq9VXSYJyW79qx6yvmdo3l5viGs+maqYrO+OnSvx+KLeLZz2te1rZL2v
|
|
ed7WneZYWnZl5K72YV1xEyxmeqJljzIEWlVkszvbZp5soN3h2SJz3pP3odCnuWmPRxuERfJrZmtZ
|
|
mtY96fR28kbX3dXj/wAuTyf6bmK+9YX1s0cNtm3Sd4LFY2K23W1s16StiUJW7bp22RW3RluBuruz
|
|
mWEgrmCGWyNkoExKE1QlPmsqRDKeyBjaejWy2W3ttDUyz1QKslvehVqKTNosyyTvELabXptIJpaP
|
|
B39Ia2mz+JGpr51jdZefDx2hzuHZObNq58poJaGtjxJ2+LoaKP8ADRPo5+T3skx5OhpOmC0fBNQ0
|
|
5yTbn+bt8A0u9raiY6RHLVwY62mI6zMvaaHBGn0mPHt1iN5+aYVsACBXqMFNTgviyxvW0bSsAeE1
|
|
mkvw7V2w5Ote9besJx2er4rw2nEdNNekZa9aW9JeQjnxZLYskTW9Z2mJY7zz26fHrrdpbZsY7NGt
|
|
mxjvso1b9NmUwpx33XRO4K7VUTE1nmrvEx1bVo2VWiJE/XY4frY1WPlt0y17x6/FuPM0m+HJGTHO
|
|
1qu9pNVXVYt46Xj7VfRtnXXL5MfzexsALsgAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAHM4jxOMFJphmJv529Dq
|
|
ZLfjDjPEIx450+K3v2+1MeUOHSOWFc3nJkmZnf4yujpVlqunOeFpV2nctLCZUXRM7MJtsWlRkv3Q
|
|
ky5NmpWt9RnrixVm17TtEQnJabXisRMzPSIew9n+CRoccajURvqLx5/chfOest642OGcIpoOG2w7
|
|
ROW9d72+LQvXevyejcPUU5M+SvpLeOataraw2a0dLbLqTtK1G3Es4lVWWUSoldFtmcXUbpidgXzK
|
|
GEW3TuCUSncnsDFMMLSms9EC6J6FpVzbZE5ALy0809ZbFr9GtfrEoFMzuuwz0Ueey3HbaBLDXe7i
|
|
tMOfwWnP9I+NZbuttvhs1uBRtXPb4SDm3iIvf57N7Dbl0VrS5+XrltEd+Z1Jx7cNms9N4TURRw3T
|
|
+PrcO3WszEvZOD7P6aYiMlvu16S7y1QAIAABxOPcLnUY/pWCv1tI96I+9DtgmXl68Biy7/NtUu3+
|
|
O8HnFa2s0tfd75KR5fFyMWTdhrPHVnX9R0cd21S3Rzsdm1iuqs256wrmGcT0RYSx5d047X02SMmO
|
|
esd49YRE9WcdSXhZ2O1p89NRji9J+cei1xMc3wXi+KZj1j1dTTaqmor06WjvWW+ddcu8XK8BZmAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAMMmWmKu952UZ9XFZmuP3revlDTtzWnmvO8q3XGmfHb9ZanV3yxtWeWn7y4es
|
|
vPNtDqZJ6Ts5mppvdl/XXRMyfGvSNlu/RVvtOzLfoipLT1VTKbSpvfogRkvtDVyZOhkyvQcA4Dzz
|
|
XV6yvTvTHMfvK+c9U3rkW+zvA/D21urr789cdZ8vi9KDb45rejl8Rry6iJ/FV1HP4vXbBTJEfYt1
|
|
+UpiHM295bXsqrO9l8QkZ0lZEqqLeyBZHZLGvZkhIndADKJ3TMoqWQMZ6pjsxll2jsCLSrmU2lFY
|
|
36gieyu0LJk3jbsga0wdqzK20QpyztQGprL/AFMrOE05NLkt6qdVWZxNrSe5o9vWBLiUjnzXn0vL
|
|
q555dHt8HOwV928/1z/LpzXxbYccRvzTB+jucOwxh0dI22mY3ltIrHLWIjyjZKyoAAAAACJiJjaY
|
|
3iXleM8InR5J1GniZw2n3oj7s/8Ao9Wi9a3rNbRE1mNpifNFnVs65XhcWTdt47bnFuF24dm8TFEz
|
|
p7T0/pn0a+HJux1OOrOux08d1ndqY7tillVkzExLOk7yd4YxGwluViJhE45raL0na0dtlWO0+bZr
|
|
1TKi+2zptZGTamT3b/tLacvJjiY3XaTWdYxZZ6/dtPm1zrv1z78fPcbwC7EAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABhkyV
|
|
xUm152iAZWtFazNp2iGhm1Vss8uP3aevnKrNntqLdelI7VRHRnrX/HRjx/tZREVjZXeybW6KbWZt
|
|
pCZ6S08tN7Nmbb7zCrJtyoS5145bSx5mWafelr3tsKmS/o08uXyhlly7RPV2+AcBnPNdZrK+53pS
|
|
fP4ytnPVda4y4BwHxOXV6uvu96Unz+MvVxG0bQRG0bR2G0nHLb2gCUDX12LxtFmpHeazt82wT1gH
|
|
mMN4tWs+rcr2aEV8DU5sM/cvO3yb+O0csLUTSdrLphRE8tlkZI7Atr2ZMazDJVKTYSCawi7Ksq7z
|
|
1QERvLK3ZGPrKbyCrbdnMcsbeaa18/RhvvM7oGEwTG0JmYYTIML22a2e28xELM19oURPNO4lOem+
|
|
n3ZY5+prVnMc2GYU4/L4A0a15cNf6rz/AC6fC6+NxCPOuOu/5tHJTbHj+F5/l1+BYumXJMd9o3/d
|
|
MRXYASgAAAAAAABhlxUz4rY8lYtS0bTEvH8R4ffhmo6bzhtPu29Pg9mq1Gnx6rDbFmrzVsizq2df
|
|
zXkMWTeIbNL7tbXaHLwzUctvexWn3bmPL8WFnHVL326VZ91MfFVjvvVlz79kLrcf2m7j7bNHH3bl
|
|
J2SirLQoy4t1++7G0dBC/RanxI8PJPv18/WG241+alovSdrV6w6mDNGfFF4/OPSW2b1zeTPL1aAs
|
|
zAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAVZ9RXBTe3WZ7R6iZOpzZq4ac1p+UermZMl89+a/byj0Ra9815ted59PQ32hlrXXRjH
|
|
DpCLX6ML5NlNsm/ZRqstfdXzbsZt06sLZNvNB1Za8RDWyZdo7q8udq5Mu/mIMt4md2lmy7JzZuWJ
|
|
dHgfBL8RvGo1MTXTxPSPx/8AstJ1XWpIs4BwSdbeNVqq/URPu0n73/s9hEREbRG0QUpWlYrWIisR
|
|
tER5JbSccur2gCUAAAAPM8Sry8Uyz67fwuxbzVPGsE49XGbvF42V4M0TEL33ERnktsxpk3sumK2j
|
|
admFdPFZ33VS2Mdui2J3UU6LYlFSsN2O5NkCyJ6K7T1TEsbAsxdpReerKkTFGMxvYEz0rsqtbbpC
|
|
b2VT1QEzuwtbaGUxspuJU3neWdKoiu8rq12gCI92YatLcublnzbEz1aOptyZqTuDHLfxN6R0+t5X
|
|
qdJhjBp6UiPLeXl9NSMnEKxHa1+bb8nrlvxUAAAAAAAAAAABTqtNj1eC2LLXeto/R43VabJw/VTh
|
|
ydY+7b1h7ho8V4dXiGlmvbJXrS3xRZ1fGv5rzeHN02bEW3cys3xZJx5ImtqztMS3MeTeGFjqlb2O
|
|
8btql3NpbZtYsnSBLeiWfdTjtutid+ghherHS5p0+f3vsX6T8Fkw181d4lMvEWdnHaGnw/UeNh5L
|
|
T7+PpPxbjdyWcvAAQAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAo1Oprgr63ntAmTqdRqK4K9etp7Q5d7Wy2m953lNrWyWm953mVd77R0
|
|
Za1104xxlN9lV8qnJl2a9s3xUXX2ybsJyRDWtl3YWydEC+2VRkzeW6q+T4tbJm+KRdfK1cmWZnlr
|
|
vNp7RC/R6HU8SycmCk7ed57Q9ZwvgOn4fEXtHi5/O9o7fJaZ6z1uRyOEezVstq6jiEbV71xevzer
|
|
rWtKxWsRFY6REeSRrJxz22gCUAAAAAANbX6aNVpL0npMRvWfSXlKamsRMVvXm+EvZXjmpaPWHzfL
|
|
oNRjzXicfWJ8phfPxFejx72x7xMzK+sXiNoiXlq+Pi6fWV/VfTNqfLJl/WTg9Pji8R70LqvMV1Gq
|
|
j/zcv6yz+lanzzZP1lWpelTET6S81Gp1P/Gyf90s412rjtnyfqql6asREdWM9+jz9eJ6yP8Az7uh
|
|
odZqMt458tpB1JvEViI3/RhzRt13/R1MNaziiZiJn5K9ZNceKZiIiQcu/WekT+iYrWI3lzdTrs+8
|
|
8uW0fJzcur1Np/zsn6g79phVaIeetqNR/wAXJ/3SwnUaj/i5P+6UD0ldonum161h5mNRqP8Ai5P1
|
|
lNtRqJjacuT9Qd22WN5aGeZyZd/KHJy59RHbLf8AVq31Gp/4uT9ZEvS8Lr/vSs2npzRtL1z53wK+
|
|
oza/HW2XJNd99pmX0Rb8VAAAAAAAAAAAAAAcHj/C5yV+l4I9+v24jzj1cLFk8nu5jeNpeW41wmdL
|
|
knU6ev1Vp96sfdn/ANFdTrXG+eq1q5F2LLtbZoY8m8d11bbSydErsYsm+zZrO/zcnBm226uhiyRK
|
|
EtrvCrJDOJTeu8A1MWX6Lqq5N/dnpb5O5ExMbx2cPNTeJb/DM/iYPDtPvY+nzhri/jDy5/W6AuwA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAa2p1UYo5adbz+xbxMlvqJ1OqjDHLXree0ejmzNrWm953tPmTPWbWneZ7yoy5YhjrXXTjH8s75N
|
|
mtkyxt0VZM2/m175N1V03yTKubMLXVXybeYLLX2VXy7eam+b0bOg4VquJW+rry4/O9uyZOq3UjVm
|
|
9r25axMzPaIdvhns1kzbZddM0p5Y47z8/R2+HcF03Doi1a8+Xzvbv+TotJnjDXkt+K8ODHp8cY8N
|
|
IpSO0RCwF2YAAAAAAAAACvUZYw6fJkntWN3k8dfHz2vLucdz8mkjFE9bz1+UOZosX1UzPm0nqI/W
|
|
MYo9FlcPNklfFGeH/NshLGun+Cz6PtHZtVZWlRLS+jxPkRpIn7rdoupHTdA5s6SI+7H6Mfo+32Y2
|
|
+To3neSIiZ7A0IjPXpXLePlMotGW3272t85datKzHZjbTVnsDj+FG/2Y/RlGP4R+jo20u7H6N1Ql
|
|
o+H8I/REY957R+jpfReiK6eOYHLtj2tttH6KrY/6Y/R2c+kjeJiFVtLG24hxpw7/AHY/RRkw9O37
|
|
O99Hrt1YX0tfOBLjcGp4XF8c+u8fs9c4dcVcGemSI61nd3IneN1orQAAAAAAAAAAAAABFqxes1tE
|
|
TE9JiUgPKcX4RbRXnNgiZwWnrH4XPi28PdXpW9JraImsxtMS8pxXhF9DecuGJtgmf+1TWW2N/la1
|
|
L7N7T5e3Vy6W3hsYcvLbqzbO9jvvCzvDR0+XeO7crO6FmGSvRThy/RtVXJ92elvk2rRvDUzU7pl4
|
|
izsd2J3jeBpcNz+Lg5LT7+Pp+Xk3W7js5eAAgAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAADs0NTrN96Yp6edkW8Wzm6+LNTq4pvTHO9vOfRoWtt
|
|
1mes95YWvs1s2fZldddOczLPLn2ju0MmebT3YZc2/mpm3qqllN1drsbZIhr3yzvtHf4AsvlYYseb
|
|
V5Yx4KTe0+UQ6nDvZ3UazbJqd8OKeu33peq0eh0+hxcmnxxWPOfOfm0mP+steT/ji8N9mKY9suum
|
|
L37+HHaPm9DSlaVitKxWsdohI0Y22gAgAAAAAAAAAABXnyRhw3yT92Nwef4xm8bVzET0rPJH5d12
|
|
CvLhho3rN9RWs9Z23n5y6O21YhrVYbdGOCfrrLPJRpv863zVS6FS09SvZj3lVZZRdPSqmnSWdrIE
|
|
ebOkK4ldTsgW1WKqd1oMZhEVZyRAImOjGI6rJ7IiATNd46qL02bHkiaxaoNGY2n4ImPgtyV2n0Vo
|
|
Gvlx7x2beiyTk08RPevSVUxux00+Fn2n7N+n5rRFb4AAAAAAAAAAAAAAACLVres1tETWekxKQHlu
|
|
L8InR2nPp43wz3j8P/s5dLveWrFqzW0bxPeJeV4xwmdFec+CJnDM9Y/CrY1xv8qvTZ+WYdbDk5oh
|
|
5zHk283U0eo3jaZZ2N5XYjrCnLSJhOK+8d1kxvCqzSwZvousrb7k9LfJ3nB1OLeJdLhufx9LEWn3
|
|
6e7LXN9Ofy5/W4AuxAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAETaKxMzO0Qi9646Ta07RDmZ9VbPbaOlI7Qi3i+c3TPUaqcu9adKfy0722ZXvFa9
|
|
XO1OrjrESxt66ZJmcjPUanlidmhkzTZVfLN5VWvsC2b7R3U3yqrZZtO1esz2h2+F+zWTUcuXXTNM
|
|
feKR3n5+iZLVbqRzNJo9TxHLyaekz62ntD1fDOA6fQbZL7Zc/wCKY6R8odLBgxabFGPDSKUjyiFj
|
|
SZkYa3aALKAAAAAAAAAAAAAADQ4pl2pTFH3p3n5Q33E12Tn1eSfKscsLZ+orS00eJqbW+Lfnu1tF
|
|
XaJnZsz3WpCfsyp00fWSvmPdVYOmSUDd8kR3InoQosy7JmUX7MdwZ17ro7KKT1XRPRAsrO0rYndr
|
|
79V1ZBaQiJ6JgCSIJASwrO07MpV2nqBlrv1a1o2bf2qtfLXaQUTO0sb05o3jv3ZXhjS20xEphW5h
|
|
yeJjjf7UdJWNKLziyRePsz0lux1SgAQAAAAAAAAAAAAAADG9K5KTS8Rato2mJZAPIcU4ZbQZuekT
|
|
OC3afT4NXFkmlntc2GmoxWx5K71tG0vHa/RX0GpmlutJ61t6wrY2xr8dXS5uesN+tt4ef0eaa223
|
|
2dnHk3juyreM81OaFGiy/RtZET9jJ7s/2bdutd2jqKeic3iNTsd8a2h1H0jTVtP2o6W+bZbOO+gA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABje9cdJt
|
|
adohGTLXFTmvO0fy52bJfU23t0pHaqLeL5xdK9Rnvqb+cUjtCi94xxvK3JetKuHrdZvaa1ljb10y
|
|
cnIs1Wt3naJc++TmVWvMz1YWybfMGdsm3eWek0mo4jm8PT0mfW3lDf4V7P5tdMZdRviwfvZ6/TaX
|
|
DpMMYsFIpWPTzXmf+steT8jn8L4Dp+HxF77Zc/4pjpHydYGjC3oAAAAAAAAAAAAAAAAADG9opS1p
|
|
7RG7zszN6WtPe0zLua+3Joss/wBOzhzG2OsL5+IrY09dsSyYRijbHEMvOChb7KjF0yS2LQ169Mso
|
|
S24noyrPVXWejNVKbTuw3T3REdQWU6LYlVvsyiUDPfqupPRr79VuOQX1lZEqoZxIMksd0gT2VT0l
|
|
bPZVbuCaW8i8bwr32WxbcGnkjaZa9p2ndv5qbw5+aNugLItF6TEtvTX5sMb969HMpfazc0d9stqe
|
|
vVZDdAQAAAAAAAAAAAAAAAADV1+iprtPOO/2u9bektoB4TJTJpNRbHkja1Z6uto8viVht+0HDvpG
|
|
H6Tjj6zHHvbecONw7Ltfkmeqmo6Ma69DXbbZTkr1mGWO3RneOaGbZRoM30fVzSelMnT83aef1FZ7
|
|
x3h1tBqfpGnjmn369LNc3sc3kzy9bQCzIAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAa+q1dNNXr7157VhGp1Xh70x+9f9ocy283m1p5rz3mVbrjXHjt91lz
|
|
5c9+fJ1nyjyhdM8lZlOOIiqrUXikd+kMreunnI5XEdX4dZiZcG+XmtNl/F83PeeWWHDOGanieSKY
|
|
q+5H2rz2hMzWd1Iqx1yajJXHhrNrW6REeb1nCPZumn2z62Ivl7xTyr/6uhwzhGn4Zj2xxzZJ+1kn
|
|
vLoNJnjHW7TbbsAszAAAAAAAAAAAAAAAAAAAAaPFrbaSK/itEOXt0rDf4xb/ACa/GZacRvaF58Q2
|
|
IjasQnzPIhCU92tMbZGzHmotG10C6nZkwpPRmipIllEbMIZIE7solgmJBnCyk9VMM6z1BtVllEqK
|
|
z0WRILYlluriWcSDJVbusV27gwInaSWM9ECyZ3hqamnSWxFmOSOaqRx725bNnSZNs9J+OynVY+WZ
|
|
YYr7TE+nVaIr0Ais81Yn1hKAAAAAAAAAAAAAAAAAABExvG09peU4nov9n66L0j6q/WPg9Y1OJaON
|
|
ZpL0+9HWs/EWzeVz9PbmrEtnyc3h9reHy26TWdnSr2YX6657ijLXpLX0+onSamL/AHJ6W+Tbv2aW
|
|
ekTv16JzeI1Ox6KJiYiY7Slz+E6jxdN4dp3vj6fl5Og2clnKACAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAACZ2jeQRMxEbzO0Q08uqtkma4ulfO3r8lefUePMxWf
|
|
cjy9WvlzVxV6T1Z61/x0Y8f7Wc7Ur1lqVy+LqOWJ2hp6rXddon5rOF1tfmz5OkT0qzb8dWbxjp1c
|
|
biuuilJ5Z6r+IcQrixzEy8zl1E6rNt1tMztFY81sztU1eRucN4ffi2p5esRM72n0h7rS6XFo8FcO
|
|
CkVpX082nwXh3+z9FWLxHi36328vg6TZyW9ABAAAAAAAAAAAAAAAAAAAAAADj8Unm1tK/hqppHvw
|
|
y1k8/EMk+m0GOPeafiFpCZYwolnXspvHvLa9mF46gmnZmwozRUiUCBKYYsoBLOFbKAX0llEqqyzi
|
|
QXRLOJVRLOOwLIljZMEgrlhKyYYTAK5nZPN0RZjugUanHzVlz6xtLq361c+9eXItPpXX0dubTU+E
|
|
bL2lw2++O1fSW6m/VYAISAAAAAAAAAAAAAAAAAp1GbwcfTreelYEydcuMcRrM/L9nnlsV6wqpi2r
|
|
tv133mfWVkRyRtEdGFva7MzkYZNoamWN4bV4mYa9qztKIujhVppxGI8r1mJegeZpknBqKZY+7L0t
|
|
LRekWrO8TG8Ns/HJ5ZypAWZAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAADS12fp4VJ6z9qVuq1HgUiI+3bpDl589cOKZmevqprXPTbx477rDJlrhr1nq4+s182tMRP
|
|
RqaziXiZJrWekNG17ZbxWJ336M5LXRbI3dLTJrs07RMY6fan1dHLrowY+X7MVjt6N3R6Kul0EbWm
|
|
s7bz8Z+LnabQX43r7Y53php/mXj+Dnv0f1JO1x/8ZxbUzj02O15mfLtD13AvZqnDds+pmMmo26el
|
|
XX0Wh0/D8EYtNjilY7+s/NstpOOTW7QBKgAAAAAAAAAAAAAAAAAAAAAADG88tLW9I3BwJtz6nNf1
|
|
vK/DHVqYJ3pzT5y3MPZeojOWMQylEKpTVjZnDCwkqzYQyRRICATCITAJZQxhMAshnEq4ZQC2srKq
|
|
qrIBZCWNZZgwswmFloVyCu0dFcx1WyrtCBhv5NTPHXds2U5o3hIz4ffbPt+KHUcTSW5c9Jme0u2v
|
|
VYAKpAAAAAAAAAAAAAAAAYZctcVOa35R6tLrltN795/YvknNqrfhpPLH92V5isd9mWq6fHjk6rn0
|
|
ZxG8KK5Jm/wbVZiYZtqrmkqL023bkxvCiY3lJHNyRG81mHS4Rn5sNsNp64+3yaWaNrzOzHBl+i6q
|
|
mT7s9J+S+ay8mex6EIneN47SNXKAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAImYiJme0JafEs3h6fkidrZOn5eaLeJk7eOdm1Hi2vmtPTry/CHmOJcUvmvOPF1n09Pm
|
|
6HF9ZGm01qxO3R5vSY7XwzmzTy47zzTEd7en5Mfvt2/PURWdo3tvPrPlKymbktFqTtMTvHzbOLDG
|
|
f63JXbFX7FdnoODcDprZpq9TjiMMTvSn4vj8l5fxnrk91saPSa7i2hpOfbTVt5x1m0fLydzR6PDo
|
|
dPGHBXasd585n1lsRERG0dIF5OOe6tAEqgAAAAAAAAAAAAAAAAAAAAAAADX11+TRZrf0y2Gjxe22
|
|
gtH4piP3TPpXKwxtjhuYo9xq442iIblI2pC1RET2ILd9kxCqRjZmwlCSEohIJAQAAJZISDKGUd2M
|
|
MoBnVbVVCyAWVWeSuqyOwIlXZZKue4MJV2WWYT2QKbKL9YlfdRdIo35b7/Hd3KTzUrPrDh27uxpb
|
|
c2mpPwX/ABX9XAKpAAAAAAAAAAAAAACekTIp1eTwtJmv+GkyJn1oafeazbfpMzLR4jq/o8b823zX
|
|
6XNF8ERCvTcNpxLV5LauvPhx9Irv3lhztdtv8TtaWLicXrt03jzjzb2k1nid56ty3s/w+a7Uwzjn
|
|
1raejlarhmbhl/FpbxMO/fzj5p/ixSeXOvTtRfeI280ZI26tfDm3pWe63LaZx7qtGvniJ6tPLvOK
|
|
fOa9WzbJvTbza02jl3n5SSljscK1MajSxWZ96nSW88xw/VfQ9XMT9nfa3yemid43jtLeXsce88qQ
|
|
EqAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAADia3UTm1l4j7OP3Y/u
|
|
7Vp2rM+kPJW1PhYcmS0+9MzKm/jbwz31weMzbV8UppazPL9q0/BF4rk1GLDSNqxPWPhCnHmnNrtT
|
|
qPKteWPm6U6OdHaZvO+SaRNvhv12Ub/q3FhtrNVj0uKOt56z6R5y9zix1w4qY6RtWsREOJ7L6OKa
|
|
S2rvX6zNM7T6Vh3mmZyOfya7eACzIAAAAAAAAAAAAAAAAAAAAAAAAAAczjVvqMVfW/8AZ03I41bf
|
|
Lp6/OVs/UVrY47NyOzUxd4bUJpEbb3Z7IiOrKIVSjZhMLJYyhKIgmGUQSDESIEbJEgQmCITEAmGU
|
|
IiGUAyhZVhDOoM4Wx2VQtqBKuyyWEgqlhKyyuyBVaGtkbNmvk7A15l1eH2300R6TMORPSXT4ZO+O
|
|
8fFefEX63gEAAAAAAAAAAAAAAAq1WPxdLlp+Kkx+y1Fvsz8gjhaDauGK8sx07y3OE3m1tT6RaP4c
|
|
vU6yMNKUx73zT0ilY3l2eF6a+m0kRl/zbzz3+Ez5M8z26fJruW6wzYq5sV8d43raNpZjRzPPaTmx
|
|
5b6bJ9rHO3zb2WJ8GWPEscY9bgzxH2t62n19GWW0eHOzHU5XbjXZ1x8WTnz2iZ7S2M1IjH2+LX0V
|
|
KTqs8zO9ot0j8nUthi1J3UaOFMTfLFo6xMbS9BwHWTqdHOO8+/hnln5eTjYMFo1WTH5VnePzXcIm
|
|
2k4zlpPSmXy/hfF5eMfJns69OA2cgAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAADG/2LfJ874rW845mubliY7bPoto5qzHrDz0+yePNF41OotaJ7RWNtpV1OtfHqZ715fhu
|
|
j8adNpcVfeyzE2/vLuanhOu1nEctIxTTFa/+ZPbZ3eHcF0vDbTfFE2yzG03t32+DokynXl9+leDB
|
|
TTYKYccbUpWIhYCzEAAAAAAAAAAAAAAAAAAAAAAAAAAAAcXjE/4zDH9M/wAu04XF5/3jj/0f3Wz9
|
|
RUYmzDWxS2I7FSyjuzY1ZKpRKEygEwiWUIkGIk2QJNhKQhMIhkCYZQxhlAMoZwwZwgWQshVCyATL
|
|
CWc9ldpBhZXLOVdpQK7NfJPRdaWvknoDVvPvOnwuel4+TlXn3nS4VPvXj4QtEV0wAAAAAAAAAAAA
|
|
AAAAAVV02CmTxK4qRf8AFFeq0AAAanEsfPpZmO9Ji0NDLfkwdOsulrumiyzHlVzJrz4Ovoy26vB8
|
|
cTBa9NffLtMY77Rv8Yegx5ImkKdJoY1HC81Y+3OSbVn0mGGkmbY45u6tnrrTOu2xGO0RxCd+nNVj
|
|
qKxTV1vH2pjaGtnyzXXYdo96ZmGXEMk15b7/AGZiVerWPTYckZcNbx5wzc7hGbnxXxzPWk7x8pdF
|
|
0S9jh1OXgAlUAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAcPjEf4/FP9H93ccXjMf4vDP9Mx+62fqKrx+S+GvibEFSsqyYwlVK
|
|
ZYsmIMoRKYJQIPIEiQ2ATCUQygCGUIhMAyhnDCGUIFkLIV1ZxIMpVWWSrsCuyqyyyq09ECq8tfJK
|
|
66jJ2Bp5J6upwn7dv9Lk5J951uE/av8AJaIrqAAAAAAAAAAAAAAAAAAAAAAq1Mc2myxPnWf4cmtu
|
|
XT9fR0tffk0WSe28bfq5Wbamm3326MtunwfK6PCv/AxPraZ/dz9PO97/AOqf5dHhdZrw7Dv3mOb9
|
|
XOxRFM+avpe38mvkPHf/AFWlrKba7Tzt99ZxKkfR7euyNXMTrtPHfa0z+zPiM/UR8Zj+Wbdu8HpN
|
|
M2bfzrV13M4dO2pyR61dNvj44/J/oAWZgAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAADj8bj63BPzdhyeNx0wz8ZWz9RWri7Nmv
|
|
VrYu0NmqaRZHZlDGGSiwxZSgCEkCBCQSCQBMJRCYgEsoYx3Z17AlMIhlCBnDOGEM4AlhZZKq4KrK
|
|
7LLKrIFN2vdfZReAaObu6/CO9vk5OePR1uEd7fJeIrqAIAAAAAAAAAAAAAAAAAAAAGtxCk5NFliI
|
|
3mI32+XVyNTyZOHTee946PQKPoeDffw4777eW/yVs60xv+ZxOnr4Okx1t05KRv8Ao41Z5q3yed5m
|
|
XY1szXRZ5jvFJ/hxItP0aOSN9q7yrtr4f2tHFM5+KT16Yq/vK/iGSbXw4vO14UcPx5MGfNbPG18m
|
|
1oj4THRsTw7VanPXVYpi3gzMcnrvCnG11JOupwuN8+a3pEQ6jT4divjxWnJExa09pbjbM5HHu90A
|
|
JUAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAHM41H1GOf6nTc/jEf4Ws+lls/UX45uGekNujTwdm5RNIthKIZKLDFlsiQIShIC
|
|
EgCUJ7AmGTGO7IDzZQhMSDJMMYZQgZwzhhDOATuqssmVdgVWVWWyqtCBTeVF19lF+wNLNG7q8I+9
|
|
8nLyupwnt+S8RXUAQAAAAAAAAAAAAAAAAAAAAAAItWL1mto3iY2lyrcLyUxzix2ia2nvPeK+jrCL
|
|
OrTVnxpanhuPPemSs8l6RtE7dJj0ldpNP9GwRSZ3neZmV4cR/Vs4AJQAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAANHi1d9H
|
|
M+kt5ra+vPoskfDdOfqK4mn7Q3aNHBPZu0W0RdDOGFWcKLCJZeTGQQlCQSgASBsCYZQxhlAJTAmA
|
|
TsmAgGcM4YQyjsgRLC3VnaVcgwsrt3Z2V2QK7tbJ1bN5a9waeWO7p8Knt8nNyebpcK8vkvlFdQBA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAK9RXmwZI+ErEWjesx6wQeZwejeo0cccuW8
|
|
elpblJaaRGxVnCuss4ZrMvJEgCAASISCQIBlCYYpieoM0wx8k7gzIRueYM4Z79FcSy3QEsLJmWFp
|
|
BjaVVpZWlXMoGNmvkXXlr3kGtknu6XCf7OXkl1OEdl8orqgIAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAHmskcmtzV/rls0U62OXiWX4zErcc9GmkRfWVkSqqziWayxCPIANwBIhIJSxS
|
|
CRG6dwZwlhEs4BluMdzfqgZxLLdXuy3AmVdpZTKuZBjaVVpWWV2QlhZRdfZRcGpl7urwfrzfJy8r
|
|
rcH61vPyWitdMAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAHA4nHLxKZ9awnH2ZcY
|
|
jbW459aq8fZpfiI2IZwrqzhmsz3Ebm4JN0AMhCQSIASndiAziWUSriWcAyRujc80DM3RCfIETLCW
|
|
UsZEsJYSslXZAwlTddPZTkBp5e7r8Gj6rJPxhx8k9Xa4PG2C8/FaK10QAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAcfjcbZMFvnDWx9m5x2PqcNvS+zSxT7sNPxH62YZQwqzhRZO6UCB
|
|
KUAJTux3SDIRuAncQAmJZRLBMSgZ7iIAZRKd2DICUSlAljLCYWMLIFVukNfI2bNbIDTyT7zu8Ijb
|
|
Sz/qcG/2nf4T/wCE/wD2WnxWt4AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAHL9oL
|
|
+Hw2cm28VvEuPptfgyVj6yIn0no7/FtJfW8NzYMe3PaPd39d3iMug1WktNc2C9dvPbeP1aZ9xF+v
|
|
T471tHu2iflK2HkqWmvaZj5Surqc9Ps5bx+alTHqYHm68S1Vf/NmfnC2vGNTXvyT84Ql6A3cSvHM
|
|
sfaxVn5Ssrxyv3sM/lKB1xza8bwT3pePyWV4tpZ+/MfOEjfGrXiGlt2zV/PotrqcN/s5aT/+wLRj
|
|
FontMSlAlKEgndO6IAZQljDIEgeQljLCzOVdkCu/SGrkbF56NPNeKxMzMRHxENe0+89DwuNtHHzl
|
|
5PJr8NcnLW3Pbf7r1nCZm2gpae8zMrz4i/W6AgAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAETETG0xukB4HVaeMHEtRi26RedvkyjBSfX9W77QYvC4xz7dMlYlrU7M929dWJLFc6aPK0q
|
|
7YLxPS0S22FlP6q38Zac0yR92s/KVc3tHfFf8tpbcsLRvB/dR/8ALLVnU0r9uL1+dZI1mnmdvGpv
|
|
6TOy6ym+Oto2tWJ+cJ/tW+KLK5KW+zes/KU7tG+h01p64qx8Y6NXNo6Y+uPJlp8rLf0rfG7MXtHa
|
|
0x8pZxqs9e2a8f8A7Oj7HaTHn0+f6RWM23LETfr6vRW4PoL99NT8ui7F4+vEdXXtnt+fVbXjGsr/
|
|
AOZE/OsPS29nuH27YrV+VpeV9pdPXhOtw49NG9Mld55+vXcTPd42I47qo7xSfyWV9oM8d8VJ/VxM
|
|
d8l46xWF9cV7en6o/qLfxp2I9ob+eCv/AHMo9op89P8A/wBORGmyT5R+qfo2X8P7n9Q/jTsx7RR5
|
|
6ef+4/8AuHftg/8A6cWcOSO9J/WEbWr3pY7Efzp2Lcfv5YK/9zWy8d1E/ZpSv5Oba1/+Hb9lc+LP
|
|
bFt87I7E/wAabWbiurvEx4nL/pjZzc2bJkn372t85ZXx55/BX85lucC0vPxnTxlnnjm32mOiZqUu
|
|
LJ2p4TwnVavNWaYbRTfre0bQ99pcH0bT0xb78vmtiIiNojaErMwAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAHnfarF7umzRHaZrLjYrdIen9ocPi8JyTt1xzF4eUw23rCm3R4r6bMy
|
|
wt6kdTaWLdjswmNoZontsCm0K5XWjopnuDC0dGpqG5bs08/daKV672MjbSaif6oh6Z5f2LtvptRX
|
|
0tEvUN3Jfo8f7cYve0eX4zV7B5z20xc/C8eSPuZIRficfXlcPaG7ino08HWIbePpLF2NuiyOyrHK
|
|
3fZFSwuovHVfaVF4QK5YWTM9UT0EKry6Ps1Tn4zjn8NZn9nOtLseydObiWW34cf918fWfk+PYANn
|
|
KAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAq1WKM+ly4p+/WYeBxTNd6zG0xO0
|
|
vobw3FcP0bi2em20Tbmj5Srr418V9sa2Z7qKyzi07MXUylhaU7yjqhLCeiq3ddaFNxFYW7NLNG8t
|
|
zya+WO6Va9J7FW66mvwidnrXiPY3Ny8RyUn71Jj9Ht3RPjk19HK9pMHj8D1ER3rHN+jqqtTjjNps
|
|
uOe16zAifXzfTz7kNyndpYazS9qT0mszDdoxrsi6m8LazMq6zDOsq1ZEyrt1WWlXaUCqyq0rbKbi
|
|
Fdp6PReyFd8uqv8ACsfy83aXrPZHHto89/xX2/SP/dpj6y8vx6EBq5gAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAB5n2q03LfDqqx39y39npmlxbS/TOG5se29tuavzgWzeV4mtui2
|
|
O3RRSY2hdVhqO2MvI36iu9lUsrSrvDHn6spnmSiq5jooyV6tq1VV69RC32byTh43h8otMx+r6I+Z
|
|
aK/g8TwX7bXh9Mid4iW+fjl8n1ICWb57xLBOm4zqse20Tbmj8+qKdnS9q8PhcTw5tumSm0/OHMxz
|
|
0Za+uzx3sX1t0Zxurr1ZxvspWiZYWZbsbT0QK7KLrZVZJFaqt5vbezNOTg9J/FaZeJns93wCvLwb
|
|
T/GJn92uGHldIBowAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAADuAPA67F9H4l
|
|
qMW20VvO3yRWW97T4fC4rXJHSMtI/WGhVlue3b473K2KzMML4+62tujG9pnozXaOSOVFMnVbmq1t
|
|
trJRW5E7wwvUxTvCyY6CHOt7moxz6Wh9PxTzYaT61h8x1MbZK/OH0zTf+Fxf6I/htj45vL9WgLMn
|
|
mvbPFvocGWO9L7fq85p5maw9d7VYvE4JkmPu2if3eW0+PasdFNOnxfF1Y2hlykRsmY+LJ0MZjZXa
|
|
eq2eyi8oQTO0KLdZWzPRjWu6VaqtHR73g0bcI0sf0Q8Nkq93wqNuFaWP+XDTDDytwBowAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAef9q8HNpcGaI60vtPyl56k9Iew49j8ThGe
|
|
PwxFv0l4zH2U26fDfTYiyJljvsjf4sm6vJ1hrXjq2MkqLdZEVbgbMx0auGdmzNt6iHN1Ub5af6of
|
|
TdPG2nxx6Vj+HzaaTm1+nx/iyVj930ysbViPRrj45vL9SAuyc7j1efguqj+jd4/T33rD3HEcPj8O
|
|
1GP8WOY/Z4TTT7sKadHhbcsZnaCJ3TPZk6VdrKbTutmP0U2nqgrGOsr8deiuI2X09EqKM1dt3uuG
|
|
f/jdN/06/wAPE546S9rwud+Gaaf+XH8NMMPK2wGjAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAABrcRp4nDtRWPPHP8PCYusPoWSvNjtX1iYfPuWaXtX8MzCuvjfw32siu8ptXoxi
|
|
0wy5t4YulReqmazu2skbquURWFInddM7VYRGyL291KFnCcfj8e0le/Lbmn8n0N4b2Ur4nHLWmPsY
|
|
5e5a5+OXyXugBZmiY3iY9Xz7NjnTa3Ph/BeYj5PoTxftFg8Hjk2iOmWkW/Psrr418V5WrWd2faFc
|
|
V2jdnEMXWxntupmN7NiYU27iWML6dVMVnddjgVqMsdHr+CW5uE6f4Rt+7yuSsTDv+zWXn0WTHP3L
|
|
/tK+GHl+O0A1c4AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA8Dn93W56/wDM
|
|
t/L3z59qp24jn+OS38lnpr4r7ZxHQ2TEstt3PXUrt27K57rr1VT0BjKnJPRbMqMs7QlV2fYvHvrd
|
|
VknyrEfu9m8f7FZI8fVU85iJewbT45NfQBKo817W4eulzxHaZrL0rje09ItwqbfhtBVs3leai8RD
|
|
KLw1sduesL606dWFdsZT1jdhNeq6K9DlhCVUU6s4jZnt1YzAhnM71dH2bycmszY/K1d/0c6OzY4R
|
|
fwuK4p8rTstn6z8k7HrwGzkAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAHz3
|
|
Vxvr80/8y38voTwGpj/F5/8AqT/JfjTx/WVeyY6FPspc9dZPVXaOq2WEwIUTVRmjo2rNfLHRI3vZ
|
|
DJycXtX8dZh7t879nsnhcbwz23tt+r6I2nxyb+gCVBzuPY/E4PqI9K7ui19fTxNBnp60n+Aj5/pJ
|
|
3jZu1aOnnltMNussdfXbm+l3ZM9URHREdZVXTuT1Nk7boQiOkJw28PU47/htEp5eivJPLMTCZ9Vv
|
|
x7mJ3iJ9UqNHk8XR4b+tIXuhxAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAD
|
|
weqjbWZ4/wCZP8vePCaz/wDIaiP+Zb+UX408f0r9lOxWOifJhXWjfyYWllPRXYQxnrCrJHRd3YZI
|
|
6A1NJecHEsN/S0T+76bE7xE+r5dk93LW3pL6ZpMni6PDf8VIn9m2fjm8s9rgFmQxvHNS0esbMiew
|
|
PnHLyai9fS0w2aNfUTtrs3+uf5bGPqy068fF227KtSsdFlKqNGMV6myyY6sbdIQI8tlOWOi6Jhhk
|
|
j3RD0vA8nicMx9etZmHRcT2Zyb6XNT8N9/2dt0T449T2AJVAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAHhdfG3E9TH9cvdPEcXjk4zqI/q3L8aeP6xr2TsxpLOekMK6mFo6qpXSrm
|
|
OqBixvHSVmzC4OfqK7S9/wAByeLwbTW9K7fo8Fqo6Paeyl+fglI/Da0NcMPK7QC7AAB8313TiOf/
|
|
AKk/y2MHWrX4jG3E9R/1Lfyv0/aFNOrHxuU7LI7MMayGTVlHWUXhNe6Z6wIUsb9d1m20q7dkDpez
|
|
N9tRqKT5xEvRvKez9+Xis1/FSYerb5+OTyf6AFlAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAB43j9eXjN/jWJ/Z7J5L2mry8Upb8VIF8f6aGOey2eynHvOy7bowrrYSxZSwQJ2YXZ
|
|
92N4BoanrEvVexmTm4blr+HJ/aHltRHSXofYm/1Wrp5RaJaYY+X49WA0c4AD51xONuKan/qW/lbp
|
|
+0MOLRtxbU/9SU4J7KadWPjep2WQrr2WRPRk1TvsndXMpiRCb9FNu0rbTuqvKBscCjfi9PhWZeue
|
|
V9n434rafTHL1TfPxy+T/QAszAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAHmv
|
|
avHtfTZfnV6VxPajHzcNrf8ABeJFs/XnMcr4no18c+6vr2YadkY2YM57sEDLyY37Mo7MMnYGlqO0
|
|
vQ+xNfqNVb1tEfs87qZ2rL0/sVX/AHdnt65P7Q0wx8vx6UBo5wAHz/jUbcX1PT78qtO2vaCnJxjP
|
|
8Zif2amnnspp04+OjWejKJ6MKdmcMmyJn4m5ZHzEVPMwtJv0VZLbQDqezcb8RzT6Y/7vUPM+ytZt
|
|
n1OTyiIh6Ztn45N/6AFlAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABocbxeLw
|
|
nUR5xXm/Rvq8+OMuDJjntaswEeBxT0bNZ6NatZpNqz3rO0rqsdO3PxlaWEMpY+aqWXkryT0ZT2V3
|
|
7A0dVPuy9f7G124NM/iyT/Z4zWT7sw957MYfB4Fp4/FE2/WWmGHldcBowAAeM9qKcvFeb8VIly9P
|
|
0nq7ntbTbVYL+tJj93CwT76unR4/jo0nozhhTsy3Y1sWljM9Ce7HyQIm3RRlttVbaWrnt0Sh6n2U
|
|
x8vD8mSfv3/h3XN4Bi8Lg2nj8Uc36y6TeOPXugCUAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAPD8RxeBxXUU26Tbmj8+quro+02Lw+I4ssdslNvzhzazvDPbq8d7GW7Dfqz2VzG
|
|
0s2qd+iu/Zn5Ksk9BVztX1mI8930zh2LwOHabH+HHWP2fNYp4+vwYvxXiP3fUqxtWIjyjZtj45/L
|
|
faQFmQADzftfj3w6fJ6WmHmsP23rvaqnNwqLfhvEvIYZ+sV038bo0noy36MK9oZQxrdMyrlnMbMZ
|
|
QKrS1M07zEestq/RRjr4utwY/wAV4j91p9V18fQdJj8LR4ccfdpEfsuREbREJbuMAAAAAAAAAAAA
|
|
BAJAAAAEAJEAJQAJQAJEAJQAJQAJEACUJAQlAJEAJQAJQJAAAEAJEAJBAAAJAABAJEJAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAABwvanDzaPFmjvjv8A
|
|
tLztJ3h7HjGHx+FainnFeaPnHV4vFbeIU038VbHeGF+kso7Mb9mTdhKnLK3dRm7SIrHhGPxeP6Sv
|
|
9cT/AHfSnz72Zx+J7Q45/BWZ/Z9BbZ+OXyfQBZQABzeP4/E4NqI9Ii36S8Ng/wAx9C4jTxOH6ivr
|
|
jn+Hz3B/mQi/GvjdCnWNlsdI2V07LIlg6USrt2ZzZXMoFV+zPhGLxeOaavpbm/RVltEN72Yx+Jxm
|
|
b7dKUmf7L5+s9/HtRA2cqRACRACRACRACUAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAACQAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAQCQQCRACRACRCQBCQBCQB
|
|
ACRACRACRACRACL1i9LVntMbPATTwdRkxT3pea/u+gPE8Xx+DxrPHlaYt+qNfGvjvtXXsi0dOrKk
|
|
dEXjZg6VMtbP2bMtXUdpEV0/Y2nNxbNf8OP+727xvsXH+N1U/wBEfy9k3nxyb+gCVQAGOWvNivX1
|
|
rMPnGGOXNNfOJ2fSZ6w+dZKeHxDPX8N7R+6L8a+L63KdoZ7q6zvEMpnowdKJ6ywmWUyqvIKM0vQ+
|
|
x+D6rU55+9aKx+TzWa36vbezmDwODYenW+95/Nphj5L6dQBo5wAAAAAAAAAAAAAAAAAAAAAAAAAA
|
|
AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAEgAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAACAEiAAAEoA
|
|
AAAAAAAAAAAAAEAkEAkRuAkQbgkQAkQAkQAkQAl5T2nx8nEMOT8dNv0l6pwfarHvpcGWPu32/WCr
|
|
YvK4mOem6b9mGKd4Z3idmFdka0y1c892zfpMtLPaNpEV6D2Kj/Eauf6YeweQ9ieuTVz8K/3evbT4
|
|
5NfQBKoAA8FxCvJxrUx/XMvevD8Zry8fz/Haf2RfjTx/6RSOnRMyypHu9kXjowrqVSrvPRnZVl6V
|
|
kK0775MsUjvadn0nT4ow6bFijtSsVfPuFYvpPGtNTy54mfy6vorXDm8l9pEC7JIgBIgBIgBIgBIg
|
|
BIgBIhIAgBIhIAgBIgBIIBIAAhIAhIAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAJAAAAAAAAAAAAAAA
|
|
AAAAAAAAABAJQkAEAAAAAAAAAAjc3BIjdG4Mkbo5kcwMjdhzHMDPc3V8xzAs3N1fMjmBZubq+Y5g
|
|
Wbm6vmOYFm5ur5jmBZubq+Y5gWbm6vmOYFm5ur5jmBZubq+Y5gWbm6vmTzAz3N2HMnmBlu5ftFTx
|
|
OEZJ/DMW/d0t2rxKni8N1FPWkiZ9eS08e7Cy8dGGn6UhZaJljXZGnmc3UT3dPP2cnUT78xCIV6j2
|
|
H/8A9c/6f7vXPI+w8bU1U+vL/d63du5NfUiDcVSIAS8b7RV5eOb/AIqRL2TyXtNX/e2KfXH/AHlF
|
|
+NPH/pr4+2xcxx0hFpY11K7R16KM32ZWz3UaidqSgrc9kcPicWyZJjfw6T+727y3sXh2xarN+K0V
|
|
h6lvPjj3e0ASqAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAJQAAAAAkQAkQAkAAAAAAAAAAAAAAA
|
|
EgAAAAAAAAAAAAAAAAAAAAAgAAABKDcAN0bgkY8xzAyRux5kcwM9zdXNkTcFm6OZXzMeYFvMibKu
|
|
ZHMC2bo51U2RuC2bom6rc3BZzom6sBZzI52ADPnOdggFnMc6skFnMc6rc3BbznOp3RzAv50c6nml
|
|
HMC/nOf4qOY5wX85zqOc5wbHOc7X5znBsc6edr85zg2ec52vzpi4NjmY5bROG+/bllVzsNTk5dLl
|
|
n0pP8BHmMHWNmzt0aum8obm08vVjfrtnxztR0mXHzTvaZdjVRMTLkZo6yiFen9iZ2pqY/wBP93rN
|
|
3kPY+/LfPX1rE/u9XzN3HfqzdO6vmTuIZ7m7Hc3Bnu8t7TR/vHBP9E/y9Pu837SV31umn+if5Rfi
|
|
/j/01MMb1hjkrtKzBG0bMsmOZY11tOYamr6Und0LUc7XT7u3rJPqL8er9lcPhcFpbzyWm39v7O00
|
|
+FYvA4Zpsc94xxu227jv1IAgAAAAAAAAABKAAAASgASgBIgBIgBIgBIhIAAAAAAAAAAAAAAAAAAC
|
|
UACUJAAAAAAAAAAAABIAAAAAAAAAAAAAAAAAAAAg3AEbomQZbo3YzLGbAz3RNlc3YzcFs2YzdVN2
|
|
M2Bdzom6nmNwW86JurTAMuY3REJ2BB1ZRVMVBhsbSsiqeUFXLucq3lTygp5TlXcpygp5TlXcpygp
|
|
5TlXcqOUFXKjlXcrGYBXysdlswiYBVMdUTCyY6sZBWxlnMMZgGLGZZSwkDdHMiWO4MuY5mEyjcFn
|
|
N1OdVzHMC3nTzqeY5gX85zqOZPMC+Lqdbk20eb/RKOZr8QybaK/XvtH7iZ9aGlp2luzT3fg19NHS
|
|
OjbmPcYX67XH1XSZ9XIzRvMuzrK7zLkZYmYnciunb9lZ5dTk+OP+71cXeP8AZnJ/ip2nf3J/l6iL
|
|
/Fu5L9bMWZczXi6YuIbEWTzKIuyiwLt3nuO25uI4a/hx7/rLuczg8TicvFLbfdpEK6+NPH/phhjo
|
|
stLGkctUWnoxrrU3j1cnWTzZq1jzl1clo5Zcu8c+txR63iP3Tn6pv4+g4o5cVI9IiGe7CJ2iE7t3
|
|
GyN2O6dwSINwSISAlAAlACRAAlAAlACRACRCQAAAAAAAAAASgASISAAAAAAAAAAAAACQAAAAAAAA
|
|
AAAAAASAAAAAAAAAAAAAAAAIAAAQCAJljuljsCJlhMs9mOwMJYys5TkBVsjZdyHICrZPKt5E8oK4
|
|
qmKrOVOwMIqyirPY2Bjyp2ZbAI2NmSARsbMgEbI2ZAMdjZICNkbMkSCNmOzJEgx2YyzljMAwlhKy
|
|
WEwCuWErJhhMArlhLOWEgxljMpljIImWMyTKJA3N0IBO5vux3NwZbnMx3NwZczT4jf3MdPW27a3a
|
|
fJOq1XNP2KdIRfi+J2trSYfcjeF+Wm1OicVeWIiN9kai8xjY12ORqultnI1Ecsujq79XP1FovWYI
|
|
rTgeq+j8QrWZ+3Mx+r2UXeC0WG2Ti2kiN5mL807eUREvbzbaejefHJv62Iv8WUXa0WTFhVtRdlF2
|
|
rz9WUXBtc7jR9dqc2T1ttHyhvZMvJitb0jdq6XHNcNenWVN3028U99WRj6Kb02be3Tq18/SN2Lpc
|
|
3UdN9nOmZrqKX/DaJ/d0svvTLRzV3jomK6+Pd1vvWJj0ZczT0mXxNJht60hfFnQ4qu3N1cWTEgs3
|
|
Tur5k7gz3N2O5uDM3Y7m4MtxBuCQASIASIASAAAAAAACRCQAAAAAAAAEoSAAAAAAAAAAAlAAlCQA
|
|
AAAAAAAAAAASAAAAAAAAAAAAIASgAAAEJAQJQCNkbMgGOyOVnsAw5TlZ7GwMOVPKy2NgY7GzIBGx
|
|
skA2AAAAAAAAAAQkBAEghEskAxYzDPZGwK5hjMLJhjMAqmGEwumrCagomFcw2JqqtUFEsLLrV82F
|
|
o7gqljKyYYTGwMZRKUSCAQAboJnaN5Bjkneu0d5W4ccViIiOzHFWbTzNumP1Zarr8eeRMbxDW1Mx
|
|
NO67NbkhzNVnmInqzaOZrL93JyZeV0M1++7S02jvxDWxhxx033tPpC8Z6rrezWjmZyazJG2/u03h
|
|
2vFibTHoqvamiwVwY+nLGzV0+SZ1Mx8G0/45tOhzJ5lXMc3UVXRdlF1HP+iYsDPLPPy49/tz1+Te
|
|
pSIr0ho6ak5Ms5J8o2q6NImOrHV7XX488ypzTtHXo0s9t6zG7c1G1qz6ubeZiZ3UatXJG3yauSO7
|
|
cvMTEx5tPLb3prPRMVr0HB8vicNxf0+7+kt+LOJwTJyY/Bnz3tH93X36N58cWvq6LSyiyndMSlC7
|
|
mZcymLJiwLosmJVRLKLAtiU7q4lMSCzc3YxJuDMRuAlKAEgAAAlAkAAAAAABKAEgAAAAAJAAAAAA
|
|
AAAAAAAEgAAAAAAAAAAAAAkAAAAAAAAEAAAAAAAAAAAAAAAAAAAAAhIAAACAAAASgAAAAAAEAAAA
|
|
hGzJAImGMwzQDDZjNVuyNgUTVhNGxysZqDVmiu1G5NN2M4waM0+DCaN2cbGcQNGaMZq3JxMJxA1J
|
|
qx2bU4kU09slorWNwa20z02RXHbJbl26QvtFovbHWkxEdJt5y2MOHlr2U1W3jx+1hiw8vSO63lmI
|
|
XRTaEWmtY6snRHO1VpmJ+DjavpSZl2s8b7y4HFcnh0n0gha5ebJN55KRM2mdoiPN6fh+kpwXh0Wy
|
|
RHj5Otp/s5Ps1p62y31+em9aTMYt/OfVfxTiPjZ52naI7fBrI5t66xz5+a1rW7yx0eSL6iZjtEOX
|
|
qNbSletom3lENjh2fbHzbbWt3iVozruc+5ztWubf4M4ybpQ2Oboyrva0Vjza8WdDR4OkXt3n9ldX
|
|
kaePP9VtYqctYhdvt5oivTeCZ2YOxXk6ubqMfV0b9mrljfqlFcq88k7z2U5axeItDa1OPessuC8P
|
|
ya7XRWYnwqdbT/ZMilvIu4dpslNdixXja8Y5tt85djZdbDWnGOesRtXFtuw6T27No5Kx2OrKYQlC
|
|
ExKJgBnEpiyvdlEgsizKLKollFgWxLKJVRLKJBbEp3VxLKJBnuMWQJEbpBIAAAJAAAABIAAAAAAA
|
|
lAJAAAAAAAAAAAAAASAAAAAAAAAAAAAJAAAABAJABAlAAAAAAAAAAAAAAAAAAAAAAAAIAAAAAAAA
|
|
AAABAJQAAAAgAABAAI2EoBGyJhkgGPKxmqxAKpownHC+YRMdN5BrTj67R3bOn01o7p01Iv71u89o
|
|
b9a7LfBTfS1vWI2jf12VfQPSW8KX2mas+NC2iv6xMNfJpMnLtEbuuxtMRCtzF55NR5rPps1N/ctP
|
|
y6uHreE6nXZ4pak48X3rT06fB7fNeI33cbX6mI32R/MWu7XF116aDSRhxbRERs8f499bkyZeeKae
|
|
kzE2mdon81/tfxDLGOunwbzlzbx08oaHBvZHJlx48mrvaa94pu04y617576rNGLRRM0397JEd/lu
|
|
9Dw/S3x4qxffo6mm4NjwUiKY4iI9Ib1dHFY6QIaNabbrYrLfrpJtaK1rMzPZb/s+05IpP59OyLeJ
|
|
k7eNfRaOc1ue32I7fGXYpi5Y77M8OGMeOKxHSFsU3Y29deZMzirl6dlVvhLatCjJHeYQv1rXnps1
|
|
8k9/VsW6qLVmZIi1rzitlvFKRvaZ2h6TSaenC9FFY+3brM+sqeG8Prp4+kZ+lvuxPkr1mqm95nfp
|
|
DXM459676a2q1dsV7XietvNno78+CJn1cjX6mOeIm0bR33dfRU5NJjidt9t5afjG/V6JZ7I2QMNh
|
|
nyo2BhsMuVG3wAhMSbbQRAMolnE+iuGUSCyJZRKuGUSCyJZK4llEgyZMYTuCUsYSCQASISAAAlCQ
|
|
AAAAAAEoASCASAAAAAAAAAAAAlACRACQAAAAAAAAAEgCEoASCAAAAAAAAAAAAAAAAAAAAAAABAAA
|
|
AAAAAAAISAIAAAAAAQAAACASgAAAQJAQAAhIDHZhln3do7z0WS18mWsajHjmes7pg3dNi5aRMNqO
|
|
yvDHTpPRaigHZhN4hHRlaVN59JY3zRENLUavaO+yq0iNVlitJ6vNcR1MVi0zO0era1/Ea0rPvbz5
|
|
PM5MWp45qvo2GZrhmfrsnpHpHzTCseEcM/2vrr8Q1Eb4qzy44nziPN63HpYiIiI7LNHoqabBTFii
|
|
IpSNohuVxrKtWMEejPwY9G1FFmHB4mWJn7MdfnIM9JpIx15to5pbUaas/a6rqViI7MxPxqX0UT1r
|
|
O3wVzpbR2hviP5i03Y5s6a879FNtHljydhExCv8AMTPJXBnRZbz0iG5ptFjwe/l96zctMVamTJtE
|
|
yTMibu1VrdTzRMR0j0ed4lr64MVpm0RERvMz5NvX62uOJ69XhOKX1HH9bHDtFvNYnfJeOy0Z2ojX
|
|
6jjnEq6fRUmccTvN/J9H0eKcOnx45neaxEbubwHgOHg+milI3vP2resu3Wu0JQmITsmISDHZHKz2
|
|
JgFc1RMLJhGwK9iIZ7MZgEdgmAEwyiWCdwWRLKJVxKYsC2JTuriWUSDNlEsIlMAySx3SCRCQSIAS
|
|
AAACRACQAAAAAAASIASAAAAAAAAAAAAAAACRACRACQASIAAAAAAAAAAAAAAAAAAAAAAAAQCUAAAA
|
|
AAAAAAIAAAAAAAAQAAAAAACBICBICAAEJAQJQCJcLjuS2ny6fPG/LWdpd1o8T0X07SXx/e7wCdJx
|
|
Wa0jmneHQpxPDMdZmJfNtZm49weZrh0/j4o7VtSZ2+Uw0/8A7o49k92vBLc/ntFohFW9PqGXimOI
|
|
6Tu1L8T3eCx6r2t1O3JwvHjifO99v7t/Bwf2l1PXU6rS6eJ8qUm8x+so5TsekzcSjbvs4mt4rzW5
|
|
K2mbT0itesy2cHsvbvqtbmyz5xERWP2jd1tJwrTaONsOKtZ8585+cnDrzmn4Rq+IZObUROHD32n7
|
|
Vv8A0ej0uhxaXFGPFSK1j0bkY4jyZRVZVXFGUVWbGwKsk8mObekNrSW3pWf1a2aYjHbm7bNnQ1id
|
|
PW0TvuDdhJEbQABMsLW2R0ZTMQrvfbz2YWzVhpanUxEd0dWkW5c8R5uXxDX1w4pnfr5Q19XxKuOJ
|
|
2neXltVqtVxbV/RdJ715+1bypANfiOu1HENV9C0MTfNeesx2rD1PAeBYuE6aKx72W3W9/WVnBuB4
|
|
eF4dqRzZbdb5J72l160WVK02ZxCYhOwI23TsnY2BGxsnYBjsiYZsZBjMMZZSgGEolMsQDdG6NwZ7
|
|
piVe6YkFsSziVMWZRILolMSriWUSCyJTuwhMSDMRCQSI3SAlACRCQAAEoAEoASAAAAAAAAACUACR
|
|
ACQAAAAAAAAAAAAASAAAAAAAAAAAAAAAAAAACAAAAAAAAAAAAAABAAAAAAAAAAAAACBKAAAAAAAQ
|
|
JQAAAhICEbJAYTWJ7wx8KvpC0BV4ceieWGewDHlNmWwCNjZICNhIDmcZredBecdpiY69FXCOLW+i
|
|
UiZidukulmxxlx2paN4mNng+K4+I8Hy2yaTfl37TXetoCPfRxfp1qi3F48ofKMvtvxak8s6LDv61
|
|
rZji9rPaLUf5PC+bfttS0q8q3p9W/wBrRMdpUZuKdN99nzvFqPbTVz7nD8OKs+do2/mW3h4D7Xaq
|
|
ZnPrtNpqz35aRaYOHY9Zk4pNt9rR+rl6zi+OnS+WN57Rv1lXp/YrNaYtruL6zNPnGO3hxP6O5w/2
|
|
f0HDuun09Yv55Le9afznqcOvO4tBreMTHu30unnva0bWt8on+70nDuE4OHYYx4Kbesz3tPrMuhGO
|
|
IjpDOKrK9YVpsyiGUQnYGOyUgI2SlAIEmwMWMs9kTAMJYzDOYRMArmGErZhhMArlHmzmGMwDE3Ts
|
|
bAbs4swj5pgFkSziVcM4BZEsolXDKAZwyhjCYBkACQhIAAAAAAAJAAAAAAAAAAAAAAAAAAAShIAA
|
|
AAAAAAJAAAAAAAAAAAAAABAJEAAAAAAAAAAAAAAAIEoBKAAAAAAAAAAAAAAABAlAAAAAAAIAAAAA
|
|
BAkBAkBAkBAlACEgMZjdjbFW8bWrEx8YWANb6Fp+bfwab+vLDKMFK9qxH5L0bAr8OPRPKz2AY7J2
|
|
SbAjYZAI2E7AIEgIEgIEgMdkSy2NgY7MdlmyNoBXsxmFuyNgVTVjNV3KjlBRNTlXTVHKCrlIqt5T
|
|
lBhEMohlFerLlBjEMohMVTEARDKCITsAk2AEgAAAkAAAAAAAAAAAAAAAAAAAAAAAASAAAAAAAAD/
|
|
2Q==`;async function x2e(e){let t=(r,a="application/octet-stream")=>fetch(`data:${a};base64,${r}`).then(o=>o.blob()),n,s;switch(e.config.warmup){case"face":n=await t(S0);break;case"body":case"full":n=await t(I0);break;default:n=null}if(n){let r=await createImageBitmap(n);s=await e.detect(r,e.config),r.close()}return s}async function b2e(e){return new Promise(t=>{let n;switch(e.config.warmup){case"face":n="data:image/jpeg;base64,"+S0;break;case"full":case"body":n="data:image/jpeg;base64,"+I0;break;default:n=null}let s;typeof Image!="undefined"?s=new Image:Ae.Image&&(s=new Ae.Image),s.onload=async()=>{let r=Xn(s.naturalWidth,s.naturalHeight);if(!r)re("Warmup: Canvas not found"),t({});else{let a=r.getContext("2d");a&&a.drawImage(s,0,0);let o=await e.image(r),i=await e.detect(o.tensor,e.config);t(i)}},n?s.src=n:t(null)})}async function v2e(e){let t=r=>Buffer.from(r,"base64"),n;if(e.config.warmup==="face"&&(n=t(S0)),(e.config.warmup==="body"||e.config.warmup==="full")&&(n=t(I0)),!n)return null;let s;if(typeof void 0!="undefined"){let r=(void 0).decodeJpeg(n),a=r.expandDims(0);e.tf.dispose(r),s=await e.detect(a,e.config),e.tf.dispose(a)}else e.config.debug&&re("Warmup tfjs-node not loaded");return s}async function LT(e,t){let n=he();if(e.state="warmup",t&&(e.config=Rn(e.config,t)),!e.config.warmup||e.config.warmup==="none")return{error:"null"};let s;return new Promise(async r=>{typeof createImageBitmap=="function"?s=await x2e(e):typeof Image!="undefined"||Ae.Canvas!==void 0?s=await b2e(e):s=await v2e(e);let a=he();e.config.debug&&re("Warmup",e.config.warmup,Math.round(a-n),"ms"),e.emit("warmup"),r(s)})}var $c,Op,Mp,C0,WT=class{constructor(t){ce(this,"version");ce(this,"config");ce(this,"result");ce(this,"state");ce(this,"process");ce(this,"tf");ce(this,"env");ce(this,"draw");ce(this,"models");ce(this,"events");ce(this,"faceTriangulation");ce(this,"faceUVMap");ce(this,"performance");jc(this,$c,void 0);jc(this,Op,void 0);jc(this,Mp,void 0);ce(this,"gl");ce(this,"analyze",(...t)=>{if(!Hc(this,Op))return;let n=this.tf.engine().state.numTensors,s=Hc(this,$c);qc(this,$c,n);let r=n-s;r!==0&&re(...t,r)});jc(this,C0,t=>{if(!Hc(this,Mp))return null;if(!t)return"input is not defined";if(this.env.node&&!(t instanceof Ze))return"input must be a tensor";try{this.tf.getBackend()}catch(n){return"backend not loaded"}return null});ce(this,"similarity",OT);ce(this,"distance",k0);ce(this,"match",MT);ce(this,"emit",t=>{var n;this.events&&this.events.dispatchEvent&&((n=this.events)==null||n.dispatchEvent(new Event(t)))});this.env=Ae,wa.wasmPath=Cp.includes("-")?"https://vladmandic.github.io/tfjs/dist/":`https://cdn.jsdelivr.net/npm/@tensorflow/tfjs-backend-wasm@${Cp}/dist/`,wa.modelBasePath=Ae.browser?"../models/":"file://models/",wa.backend=Ae.browser?"humangl":"tensorflow",this.version=Ox,Object.defineProperty(this,"version",{value:Ox}),this.config=JSON.parse(JSON.stringify(wa)),Object.seal(this.config),t&&(this.config=Rn(this.config,t)),this.tf=$l,this.state="idle",qc(this,$c,0),qc(this,Op,!1),qc(this,Mp,!1),this.performance={},this.events=typeof EventTarget!="undefined"?new EventTarget:void 0,this.models=new Hb,this.draw={options:ha,canvas:(n,s)=>NT(n,s),face:(n,s,r)=>Zb(n,s,r),body:(n,s,r)=>Yb(n,s,r),hand:(n,s,r)=>Jb(n,s,r),gesture:(n,s,r)=>Kb(n,s,r),object:(n,s,r)=>Qb(n,s,r),person:(n,s,r)=>TT(n,s,r),all:(n,s,r)=>ET(n,s,r)},this.result={face:[],body:[],hand:[],gesture:[],object:[],performance:{},timestamp:0,persons:[]},this.process={tensor:null,canvas:null},this.faceTriangulation=R8,this.faceUVMap=$8,this.gl=It,this.emit("create")}reset(){let t=this.config.backend;this.config=JSON.parse(JSON.stringify(wa)),this.config.backend=t}validate(t){return Zg(wa,t||this.config)}now(){return he()}image(t,n=!0){return vc(t,this.config,n)}async segmentation(t,n){return vT(t,n,this.config)}enhance(t){return mb(t)}compare(t,n){return q6(this.config,t,n)}async init(){await w0(this,!0),await this.tf.ready()}async load(t){this.state="load";let n=he(),s=Object.values(this.models).filter(o=>o).length;t&&(this.config=Rn(this.config,t)),this.env.initial&&(this.config.debug&&re(`version: ${this.version}`),this.config.debug&&re(`tfjs version: ${this.tf.version_core}`),await w0(this)||re("error: backend check failed"),await tf(),this.env.browser&&(this.config.debug&&re("configuration:",this.config),this.config.debug&&re("environment:",this.env),this.config.debug&&re("tf flags:",this.tf.ENV.flags))),await wT(this),this.env.initial&&this.config.debug&&re("tf engine state:",this.tf.engine().state.numBytes,"bytes",this.tf.engine().state.numTensors,"tensors"),this.env.initial=!1,Object.values(this.models).filter(o=>o).length!==s&&(await kT(this),this.emit("load"));let a=Math.trunc(he()-n);a>(this.performance.loadModels||0)&&(this.performance.loadModels=this.env.perfadd?(this.performance.loadModels||0)+a:a)}next(t=this.result){return FT(t,this.config)}async warmup(t){let n=he(),s=await LT(this,t),r=he();return this.performance.warmup=Math.trunc(r-n),s}async profile(t,n){let s=await this.tf.profile(()=>this.detect(t,n)),r={};for(let i of s.kernels)r[i.name]?r[i.name]+=i.kernelTimeMs:r[i.name]=i.kernelTimeMs;let a=[];Object.entries(r).forEach(i=>a.push({name:i[0],ms:i[1]})),a.sort((i,l)=>l.ms-i.ms),a.length=20;let o={};for(let i of a)o[i.name]=i.ms;return o}async detect(t,n){return this.state="detect",new Promise(async s=>{var g,A,x,y,b,w,k,C,N,R,F,_,P,T,M,U,j,z,X,Z,J,ee;this.state="config";let r;this.config=Rn(this.config,n),this.state="check";let a=Hc(this,C0).call(this,t);a&&(re(a,t),s({error:a}));let o=he();await w0(this),await this.load(),r=he(),this.state="image";let i=await vc(t,this.config);if(this.process=i,this.performance.inputProcess=this.env.perfadd?(this.performance.inputProcess||0)+Math.trunc(he()-r):Math.trunc(he()-r),this.analyze("Get Image:"),!i.tensor){this.config.debug&&re("could not convert input to tensor"),s({error:"could not convert input to tensor"});return}this.emit("image"),r=he(),this.config.skipAllowed=await j6(this.config,i.tensor),this.performance.totalFrames||(this.performance.totalFrames=0),this.performance.cachedFrames||(this.performance.cachedFrames=0),this.performance.totalFrames++,this.config.skipAllowed&&this.performance.cachedFrames++,this.performance.cacheCheck=this.env.perfadd?(this.performance.cacheCheck||0)+Math.trunc(he()-r):Math.trunc(he()-r),this.analyze("Check Changed:");let l=[],c=[],u=[],d=[];this.state="detect:face",this.config.async?(l=this.config.face.enabled?e5(this,i.tensor):[],this.performance.face&&delete this.performance.face):(r=he(),l=this.config.face.enabled?await e5(this,i.tensor):[],this.performance.face=this.env.perfadd?(this.performance.face||0)+Math.trunc(he()-r):Math.trunc(he()-r)),this.config.async&&(this.config.body.maxDetected===-1||this.config.hand.maxDetected===-1)&&(l=await l),this.analyze("Start Body:"),this.state="detect:body";let p=this.config.body.maxDetected===-1?Rn(this.config,{body:{maxDetected:this.config.face.enabled?1*l.length:1}}):this.config;this.config.async?(((g=this.config.body.modelPath)==null?void 0:g.includes("posenet"))?c=this.config.body.enabled?Vb(i.tensor,p):[]:((A=this.config.body.modelPath)==null?void 0:A.includes("blazepose"))?c=this.config.body.enabled?Jx(i.tensor,p):[]:((x=this.config.body.modelPath)==null?void 0:x.includes("efficientpose"))?c=this.config.body.enabled?ib(i.tensor,p):[]:((y=this.config.body.modelPath)==null?void 0:y.includes("movenet"))&&(c=this.config.body.enabled?Pb(i.tensor,p):[]),this.performance.body&&delete this.performance.body):(r=he(),((b=this.config.body.modelPath)==null?void 0:b.includes("posenet"))?c=this.config.body.enabled?await Vb(i.tensor,p):[]:((w=this.config.body.modelPath)==null?void 0:w.includes("blazepose"))?c=this.config.body.enabled?await Jx(i.tensor,p):[]:((k=this.config.body.modelPath)==null?void 0:k.includes("efficientpose"))?c=this.config.body.enabled?await ib(i.tensor,p):[]:((C=this.config.body.modelPath)==null?void 0:C.includes("movenet"))&&(c=this.config.body.enabled?await Pb(i.tensor,p):[]),this.performance.body=this.env.perfadd?(this.performance.body||0)+Math.trunc(he()-r):Math.trunc(he()-r)),this.analyze("End Body:"),this.analyze("Start Hand:"),this.state="detect:hand";let h=this.config.hand.maxDetected===-1?Rn(this.config,{hand:{maxDetected:this.config.face.enabled?2*l.length:1}}):this.config;this.config.async?(((R=(N=this.config.hand.detector)==null?void 0:N.modelPath)==null?void 0:R.includes("handdetect"))?u=this.config.hand.enabled?wb(i.tensor,h):[]:((_=(F=this.config.hand.detector)==null?void 0:F.modelPath)==null?void 0:_.includes("handtrack"))&&(u=this.config.hand.enabled?Tb(i.tensor,h):[]),this.performance.hand&&delete this.performance.hand):(r=he(),((T=(P=this.config.hand.detector)==null?void 0:P.modelPath)==null?void 0:T.includes("handdetect"))?u=this.config.hand.enabled?await wb(i.tensor,h):[]:((U=(M=this.config.hand.detector)==null?void 0:M.modelPath)==null?void 0:U.includes("handtrack"))&&(u=this.config.hand.enabled?await Tb(i.tensor,h):[]),this.performance.hand=this.env.perfadd?(this.performance.hand||0)+Math.trunc(he()-r):Math.trunc(he()-r)),this.analyze("End Hand:"),this.analyze("Start Object:"),this.state="detect:object",this.config.async?(((j=this.config.object.modelPath)==null?void 0:j.includes("nanodet"))?d=this.config.object.enabled?Ob(i.tensor,this.config):[]:((z=this.config.object.modelPath)==null?void 0:z.includes("centernet"))&&(d=this.config.object.enabled?tb(i.tensor,this.config):[]),this.performance.object&&delete this.performance.object):(r=he(),((X=this.config.object.modelPath)==null?void 0:X.includes("nanodet"))?d=this.config.object.enabled?await Ob(i.tensor,this.config):[]:((Z=this.config.object.modelPath)==null?void 0:Z.includes("centernet"))&&(d=this.config.object.enabled?await tb(i.tensor,this.config):[]),this.performance.object=this.env.perfadd?(this.performance.object||0)+Math.trunc(he()-r):Math.trunc(he()-r)),this.analyze("End Object:"),this.state="detect:await",this.config.async&&([l,c,u,d]=await Promise.all([l,c,u,d])),this.state="detect:gesture";let f=[];this.config.gesture.enabled&&(r=he(),f=[..._T(l),...$T(c),...PT(u),...DT(l)],this.config.async?this.performance.gesture&&delete this.performance.gesture:this.performance.gesture=this.env.perfadd?(this.performance.gesture||0)+Math.trunc(he()-r):Math.trunc(he()-r)),this.performance.total=this.env.perfadd?(this.performance.total||0)+Math.trunc(he()-o):Math.trunc(he()-o);let m=((ee=(J=this.process)==null?void 0:J.tensor)==null?void 0:ee.shape)||[];this.result={face:l,body:c,hand:u,gesture:f,object:d,performance:this.performance,canvas:this.process.canvas,timestamp:Date.now(),get persons(){return zT(l,c,u,f,m)}},se(i.tensor),this.emit("detect"),this.state="idle",s(this.result)})}};$c=new WeakMap,Op=new WeakMap,Mp=new WeakMap,C0=new WeakMap;return w2e;})();
|
|
/**
|
|
* @license
|
|
* Copyright 2017 Google LLC. All Rights Reserved.
|
|
* Licensed under the Apache License, Version 2.0 (the "License");
|
|
* you may not use this file except in compliance with the License.
|
|
* You may obtain a copy of the License at
|
|
*
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
* =============================================================================
|
|
*/
|
|
/**
|
|
* @license
|
|
* Copyright 2018 Google LLC
|
|
*
|
|
* Use of this source code is governed by an MIT-style
|
|
* license that can be found in the LICENSE file or at
|
|
* https://opensource.org/licenses/MIT.
|
|
* =============================================================================
|
|
*/
|
|
/**
|
|
* @license
|
|
* Copyright 2018 Google LLC. All Rights Reserved.
|
|
* Licensed under the Apache License, Version 2.0 (the "License");
|
|
* you may not use this file except in compliance with the License.
|
|
* You may obtain a copy of the License at
|
|
*
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
*
|
|
* =============================================================================
|
|
*/
|
|
/**
|
|
* @license
|
|
* Copyright 2018 Google LLC. All Rights Reserved.
|
|
* Licensed under the Apache License, Version 2.0 (the "License");
|
|
* you may not use this file except in compliance with the License.
|
|
* You may obtain a copy of the License at
|
|
*
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
* =============================================================================
|
|
*/
|
|
/**
|
|
* @license
|
|
* Copyright 2019 Google LLC
|
|
*
|
|
* Use of this source code is governed by an MIT-style
|
|
* license that can be found in the LICENSE file or at
|
|
* https://opensource.org/licenses/MIT.
|
|
* =============================================================================
|
|
*/
|
|
/**
|
|
* @license
|
|
* Copyright 2019 Google LLC. All Rights Reserved.
|
|
* Licensed under the Apache License, Version 2.0 (the "License");
|
|
* you may not use this file except in compliance with the License.
|
|
* You may obtain a copy of the License at
|
|
*
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
*
|
|
* =============================================================================
|
|
*/
|
|
/**
|
|
* @license
|
|
* Copyright 2019 Google LLC. All Rights Reserved.
|
|
* Licensed under the Apache License, Version 2.0 (the "License");
|
|
* you may not use this file except in compliance with the License.
|
|
* You may obtain a copy of the License at
|
|
*
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
* =============================================================================
|
|
*/
|
|
/**
|
|
* @license
|
|
* Copyright 2020 Google Inc. All Rights Reserved.
|
|
* Licensed under the Apache License, Version 2.0 (the "License");
|
|
* you may not use this file except in compliance with the License.
|
|
* You may obtain a copy of the License at
|
|
*
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
* =============================================================================
|
|
*/
|
|
/**
|
|
* @license
|
|
* Copyright 2020 Google LLC
|
|
*
|
|
* Use of this source code is governed by an MIT-style
|
|
* license that can be found in the LICENSE file or at
|
|
* https://opensource.org/licenses/MIT.
|
|
* =============================================================================
|
|
*/
|
|
/**
|
|
* @license
|
|
* Copyright 2020 Google LLC. All Rights Reserved.
|
|
* Licensed under the Apache License, Version 2.0 (the "License");
|
|
* you may not use backend file except in compliance with the License.
|
|
* You may obtain a copy of the License at
|
|
*
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
* =============================================================================
|
|
*/
|
|
/**
|
|
* @license
|
|
* Copyright 2020 Google LLC. All Rights Reserved.
|
|
* Licensed under the Apache License, Version 2.0 (the "License");
|
|
* you may not use this file except in compliance with the License.
|
|
* You may obtain a copy of the License at
|
|
*
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
* =============================================================================
|
|
*/
|
|
/**
|
|
* @license
|
|
* Copyright 2020 Google LLC. All Rights Reserved.
|
|
* Licensed under the Apache License, Version 2.0 (the License);
|
|
* you may not use this file except in compliance with the License.
|
|
* You may obtain a copy of the License at
|
|
*
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an AS IS BASIS,
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
* =============================================================================
|
|
*/
|
|
/**
|
|
* @license
|
|
* Copyright 2021 Google LLC. All Rights Reserved.
|
|
* Licensed under the Apache License, Version 2.0 (the "License");
|
|
* you may not use this file except in compliance with the License.
|
|
* You may obtain a copy of the License at
|
|
*
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
* =============================================================================
|
|
*/
|
|
/**
|
|
* @license
|
|
* Copyright 2021 Google LLC. All Rights Reserved.
|
|
* Licensed under the Apache License, Version 2.0 (the "License");
|
|
* you may not use this file except in compliance with the License.
|
|
* You may obtain a copy of the License at
|
|
*
|
|
* https://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
* =============================================================================
|
|
*/
|
|
/**
|
|
* @license
|
|
* Copyright 2021 Google LLC. All Rights Reserved.
|
|
* Licensed under the Apache License, Version 2.0 (the License);
|
|
* you may not use this file except in compliance with the License.
|
|
* You may obtain a copy of the License at
|
|
*
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an AS IS BASIS,
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
* =============================================================================
|
|
*/
|
|
/**
|
|
* Human main module
|
|
* @default Human Library
|
|
* @summary <https://github.com/vladmandic/human>
|
|
* @author <https://github.com/vladmandic>
|
|
* @copyright <https://github.com/vladmandic>
|
|
* @license MIT
|
|
*/
|
|
/**
|
|
* @license
|
|
* Copyright 2018 Google LLC. All Rights Reserved.
|
|
* Licensed under the Apache License, Version 2.0 (the "License");
|
|
* you may not use this file except in compliance with the License.
|
|
* You may obtain a copy of the License at
|
|
*
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
* =============================================================================
|
|
*/
|
|
/** @license See the LICENSE file. */
|