4224 lines
1.1 MiB
4224 lines
1.1 MiB
|
|
/*
|
|
Face-API
|
|
homepage: <https://github.com/vladmandic/face-api>
|
|
author: <https://github.com/vladmandic>'
|
|
*/
|
|
|
|
var H_=Object.create,ud=Object.defineProperty,j_=Object.getPrototypeOf,q_=Object.prototype.hasOwnProperty,K_=Object.getOwnPropertyNames,X_=Object.getOwnPropertyDescriptor;var Vw=e=>ud(e,"__esModule",{value:!0});var Y_=(e,t)=>()=>(t||(t={exports:{}},e(t.exports,t)),t.exports),cd=(e,t)=>{Vw(e);for(var n in t)ud(e,n,{get:t[n],enumerable:!0})},Z_=(e,t,n)=>{if(Vw(e),t&&typeof t=="object"||typeof t=="function")for(let a of K_(t))!q_.call(e,a)&&a!=="default"&&ud(e,a,{get:()=>t[a],enumerable:!(n=X_(t,a))||n.enumerable});return e},J_=e=>e&&e.__esModule?e:Z_(ud(e!=null?H_(j_(e)):{},"default",{value:e,enumerable:!0}),e);var U2=Y_((ure,V2)=>{cd(ure,{isNodejs:()=>cre});function cre(){return typeof global=="object"&&!0&&typeof V2!="undefined"&&typeof process!="undefined"&&!!process.version}});var Bg={};cd(Bg,{Abs:()=>So,Acos:()=>Co,Acosh:()=>_o,AdadeltaOptimizer:()=>Lh,AdagradOptimizer:()=>zh,AdamOptimizer:()=>Bh,AdamaxOptimizer:()=>Wh,Add:()=>Lr,AddN:()=>Ss,All:()=>yd,Any:()=>bd,ArgMax:()=>Cs,ArgMin:()=>lc,Asin:()=>Eo,Asinh:()=>Ao,Atan:()=>Fo,Atan2:()=>Do,Atanh:()=>$o,AvgPool:()=>_s,AvgPool3D:()=>uc,AvgPool3DGrad:()=>vd,AvgPoolGrad:()=>xd,BackendWasm:()=>L2,BatchMatMul:()=>Es,BatchToSpaceND:()=>cc,Bincount:()=>wd,BroadcastTo:()=>r0,Callback:()=>LI,CallbackList:()=>Pk,Cast:()=>As,Ceil:()=>Ro,ClipByValue:()=>zr,Complex:()=>kd,ComplexAbs:()=>pc,Concat:()=>Mo,Conv2D:()=>Fs,Conv2DBackpropFilter:()=>Id,Conv2DBackpropInput:()=>$s,Conv3D:()=>dc,Conv3DBackpropFilterV2:()=>Nd,Conv3DBackpropInputV2:()=>Td,Cos:()=>Ds,Cosh:()=>Po,CropAndResize:()=>Oo,Cumsum:()=>Rs,CustomCallback:()=>Lk,DataStorage:()=>hd,DenseBincount:()=>Sd,DepthToSpace:()=>Lo,DepthwiseConv2dNative:()=>Ms,DepthwiseConv2dNativeBackpropFilter:()=>Cd,DepthwiseConv2dNativeBackpropInput:()=>_d,Diag:()=>Ed,Dilation2D:()=>hc,Dilation2DBackpropFilter:()=>Fd,Dilation2DBackpropInput:()=>Ad,ENV:()=>Gg,EarlyStopping:()=>BI,Elu:()=>zo,EluGrad:()=>$d,Environment:()=>n0,Equal:()=>Wo,Erf:()=>Bo,Exp:()=>Os,ExpandDims:()=>Vo,Expm1:()=>Uo,FFT:()=>Dd,Fill:()=>mc,FlipLeftRight:()=>Go,Floor:()=>Ls,FloorDiv:()=>zs,FromPixels:()=>qd,FusedBatchNorm:()=>Bs,FusedConv2D:()=>yi,FusedDepthwiseConv2D:()=>bi,GatherNd:()=>jo,GatherV2:()=>Ho,GraphModel:()=>fN,Greater:()=>qo,GreaterEqual:()=>Ws,History:()=>Ok,IFFT:()=>Rd,Identity:()=>Ko,Imag:()=>Md,InputSpec:()=>Xt,IsFinite:()=>Xo,IsInf:()=>Yo,IsNan:()=>Zo,KernelBackend:()=>sc,LRN:()=>yc,LRNGrad:()=>Od,LayerVariable:()=>Fk,LayersModel:()=>wr,LeakyRelu:()=>Vs,Less:()=>Jo,LessEqual:()=>Qo,LinSpace:()=>Pd,Log:()=>Us,Log1p:()=>el,LogSoftmax:()=>s0,LogicalAnd:()=>tl,LogicalNot:()=>fc,LogicalOr:()=>gc,Max:()=>Gs,MaxPool:()=>js,MaxPool3D:()=>bc,MaxPool3DGrad:()=>zd,MaxPoolGrad:()=>Ld,MaxPoolWithArgmax:()=>Bd,Maximum:()=>Hs,Mean:()=>qs,Min:()=>Ks,Minimum:()=>Xs,MirrorPad:()=>xc,Mod:()=>nl,MomentumOptimizer:()=>Vh,Multinomial:()=>Wd,Multiply:()=>Ys,Neg:()=>al,NonMaxSuppressionV3:()=>sl,NonMaxSuppressionV4:()=>il,NonMaxSuppressionV5:()=>ol,NotEqual:()=>rl,OP_SCOPE_SUFFIX:()=>g0,OneHot:()=>Zs,OnesLike:()=>ll,Optimizer:()=>br,Pack:()=>ul,PadV2:()=>Js,Pool:()=>WE,Pow:()=>Qs,Prelu:()=>ei,Prod:()=>cl,RMSPropOptimizer:()=>Uh,RNN:()=>tr,Range:()=>vc,Rank:()=>Jg,Real:()=>Vd,RealDiv:()=>Ps,Reciprocal:()=>pl,Reduction:()=>dn,Relu:()=>ti,Relu6:()=>ai,Reshape:()=>dl,ResizeBilinear:()=>ni,ResizeBilinearGrad:()=>Gd,ResizeNearestNeighbor:()=>wc,ResizeNearestNeighborGrad:()=>Ud,Reverse:()=>ri,RotateWithOffset:()=>Sl,Round:()=>si,Rsqrt:()=>ii,SGDOptimizer:()=>Kc,ScatterNd:()=>hl,Select:()=>ml,Selu:()=>fl,Sequential:()=>Zl,Sigmoid:()=>li,Sign:()=>bl,Sin:()=>oi,Sinh:()=>yl,Slice:()=>gl,Softmax:()=>pi,Softplus:()=>xl,SpaceToBatchND:()=>kc,SparseToDense:()=>Hd,SplitV:()=>vl,Sqrt:()=>ui,Square:()=>Ic,SquaredDifference:()=>di,Step:()=>Wr,StridedSlice:()=>wl,Sub:()=>hi,Sum:()=>ci,SymbolicTensor:()=>Aa,Tan:()=>kl,Tanh:()=>mi,Tensor:()=>Ae,TensorBuffer:()=>Ot,Tile:()=>Br,TopK:()=>Il,Transpose:()=>fi,Unique:()=>jd,Unpack:()=>Nl,UnsortedSegmentSum:()=>Nc,Variable:()=>Vr,ZerosLike:()=>Tl,_FusedMatMul:()=>gi,abs:()=>Lt,acos:()=>Iy,acosh:()=>Ny,add:()=>Z,addN:()=>J0,all:()=>lh,any:()=>Rc,argMax:()=>Mc,argMin:()=>Ty,asin:()=>Sy,asinh:()=>Cy,atan:()=>_y,atan2:()=>Ey,atanh:()=>Ay,avgPool:()=>Xn,avgPool3d:()=>Dy,backend:()=>Z0,backend_util:()=>_,basicLSTMCell:()=>g$,batchNorm:()=>fr,batchNorm2d:()=>n1,batchNorm3d:()=>a1,batchNorm4d:()=>r1,batchToSpaceND:()=>Oc,bincount:()=>s1,booleanMaskAsync:()=>vM,broadcastTo:()=>Lc,browser:()=>ki,buffer:()=>Le,callbacks:()=>l4,cast:()=>ce,ceil:()=>Ry,clipByValue:()=>qt,clone:()=>jr,complex:()=>Ur,concat:()=>Qe,concat1d:()=>i1,concat2d:()=>o1,concat3d:()=>l1,concat4d:()=>u1,constraints:()=>rk,conv1d:()=>ch,conv2d:()=>At,conv2dTranspose:()=>ph,conv3d:()=>Py,conv3dTranspose:()=>L$,copyRegisteredKernels:()=>GE,cos:()=>zc,cosh:()=>dh,cosineWindow:()=>lb,cumsum:()=>hh,customGrad:()=>qa,data:()=>yN,denseBincount:()=>p1,deprecationWarn:()=>ky,depthToSpace:()=>Oy,depthwiseConv2d:()=>Yr,deregisterOp:()=>c4,device_util:()=>eh,diag:()=>j$,dilation2d:()=>Ly,disableDeprecationWarnings:()=>_F,dispose:()=>Fe,disposeVariables:()=>EF,div:()=>ve,divNoNan:()=>zy,dot:()=>d1,dropout:()=>D1,elu:()=>Pl,enableDebugMode:()=>CF,enableProdMode:()=>SF,enclosingPowerOfTwo:()=>R1,engine:()=>Kr,env:()=>ee,equal:()=>Zr,erf:()=>By,exp:()=>pn,expandDims:()=>Fn,expm1:()=>Wy,eye:()=>Vy,fft:()=>jc,fill:()=>wn,findBackend:()=>PF,findBackendFactory:()=>OF,floor:()=>Ol,floorDiv:()=>oh,fused:()=>es,gather:()=>Ti,gatherND:()=>$1,gather_util:()=>fy,getBackend:()=>RF,getGradient:()=>Kg,getKernel:()=>Kd,getKernelsForBackend:()=>Xd,grad:()=>wD,grads:()=>kD,greater:()=>ca,greaterEqual:()=>Jr,ifft:()=>Ul,imag:()=>mh,image:()=>Ya,inTopKAsync:()=>FM,initializers:()=>pk,input:()=>Nk,io:()=>Ht,irfft:()=>Eh,isFinite:()=>h1,isInf:()=>m1,isNaN:()=>f1,keep:()=>jt,kernel_impls:()=>Za,layers:()=>Ik,leakyRelu:()=>Bc,less:()=>fh,lessEqual:()=>Si,linalg:()=>j1,linspace:()=>g1,loadGraphModel:()=>cV,loadLayersModel:()=>_W,localResponseNormalization:()=>Uy,log:()=>$n,log1p:()=>gh,logSigmoid:()=>b1,logSoftmax:()=>bh,logSumExp:()=>jy,logicalAnd:()=>pa,logicalNot:()=>Wc,logicalOr:()=>xh,logicalXor:()=>k1,losses:()=>jP,matMul:()=>ze,math:()=>$0,max:()=>Yn,maxPool:()=>Ft,maxPool3d:()=>qy,maxPoolWithArgmax:()=>I1,maximum:()=>Ka,mean:()=>St,memory:()=>sh,metrics:()=>MI,min:()=>zl,minimum:()=>Bl,mirrorPad:()=>Ky,mod:()=>Xy,model:()=>SW,models:()=>PI,moments:()=>vh,movingAverage:()=>IM,mul:()=>L,multiRNNCell:()=>ZD,multinomial:()=>N1,neg:()=>Tt,nextFrame:()=>Gh,norm:()=>Dh,notEqual:()=>_i,oneHot:()=>$l,ones:()=>Xa,onesLike:()=>Dn,op:()=>P,outerProduct:()=>nR,pad:()=>Zn,pad1d:()=>sR,pad2d:()=>oR,pad3d:()=>uR,pad4d:()=>pR,pool:()=>T1,pow:()=>gr,prelu:()=>Uc,print:()=>S0,prod:()=>wh,profile:()=>AF,rand:()=>vR,randomGamma:()=>NR,randomNormal:()=>S1,randomUniform:()=>Wl,range:()=>kh,ready:()=>DF,real:()=>Gc,reciprocal:()=>Jy,registerBackend:()=>ih,registerCallbackConstructor:()=>EW,registerGradient:()=>i0,registerKernel:()=>Sc,registerOp:()=>u4,regularizers:()=>OI,relu:()=>Ye,relu6:()=>Ih,removeBackend:()=>MF,reshape:()=>j,reverse:()=>Rn,reverse1d:()=>DR,reverse2d:()=>MR,reverse3d:()=>OR,reverse4d:()=>zR,rfft:()=>qc,round:()=>Qy,rsqrt:()=>Nh,scalar:()=>de,scatterND:()=>F1,scatter_util:()=>gy,selu:()=>Th,separableConv2d:()=>Ei,sequential:()=>CW,serialization:()=>re,setBackend:()=>$F,setPlatform:()=>LF,setWasmPath:()=>Qae,setWasmPaths:()=>ere,setdiff1dAsync:()=>C1,sigmoid:()=>ua,sign:()=>eb,signal:()=>HP,sin:()=>Sh,sinh:()=>Ch,slice:()=>We,slice1d:()=>_h,slice2d:()=>tb,slice3d:()=>Vl,slice4d:()=>Hc,slice_util:()=>cn,softmax:()=>Na,softplus:()=>Ll,spaceToBatchND:()=>Vc,sparseToDense:()=>ob,spectral:()=>GP,split:()=>Mn,sqrt:()=>en,square:()=>lt,squaredDifference:()=>Ah,squeeze:()=>Qr,stack:()=>$t,step:()=>Gl,stridedSlice:()=>nb,sub:()=>fe,sum:()=>Te,sumOutType:()=>Qd,tan:()=>ab,tanh:()=>Ml,tensor:()=>qn,tensor1d:()=>tt,tensor2d:()=>Ta,tensor3d:()=>ah,tensor4d:()=>Sa,tensor5d:()=>pM,tensor6d:()=>dM,tensor_util:()=>Ia,test_util:()=>q0,tidy:()=>D,tile:()=>ja,time:()=>FF,topk:()=>rb,train:()=>Fi,transpose:()=>Ue,truncatedNormal:()=>Fh,unique:()=>$h,unregisterGradient:()=>UE,unregisterKernel:()=>VE,unsortedSegmentSum:()=>sb,unstack:()=>dt,upcastType:()=>la,util:()=>k,valueAndGrad:()=>ID,valueAndGrads:()=>ND,variable:()=>_1,variableGrads:()=>y1,version:()=>lee,version_converter:()=>gN,version_core:()=>Y0,version_layers:()=>mm,version_wasm:()=>tre,where:()=>vn,whereAsync:()=>ib,zeros:()=>vt,zerosLike:()=>Ge});var Q_=Object.create,pd=Object.defineProperty,eE=Object.getPrototypeOf,tE=Object.prototype.hasOwnProperty,nE=Object.getOwnPropertyNames,aE=Object.getOwnPropertyDescriptor,Uw=e=>pd(e,"__esModule",{value:!0}),un=(e,t)=>()=>(t||(t={exports:{}},e(t.exports,t)),t.exports),Oe=(e,t)=>{Uw(e);for(var n in t)pd(e,n,{get:t[n],enumerable:!0})},rE=(e,t,n)=>{if(Uw(e),t&&typeof t=="object"||typeof t=="function")for(let a of nE(t))!tE.call(e,a)&&a!=="default"&&pd(e,a,{get:()=>t[a],enumerable:!(n=aE(t,a))||n.enumerable});return e},Io=e=>e&&e.__esModule?e:rE(pd(e!=null?Q_(eE(e)):{},"default",{value:e,enumerable:!0}),e),sE=un(()=>{}),iE=un((e,t)=>{(function(n,a,r){function s(c){var u=this,p=l();u.next=function(){var d=2091639*u.s0+u.c*23283064365386963e-26;return u.s0=u.s1,u.s1=u.s2,u.s2=d-(u.c=d|0)},u.c=1,u.s0=p(" "),u.s1=p(" "),u.s2=p(" "),u.s0-=p(c),u.s0<0&&(u.s0+=1),u.s1-=p(c),u.s1<0&&(u.s1+=1),u.s2-=p(c),u.s2<0&&(u.s2+=1),p=null}function i(c,u){return u.c=c.c,u.s0=c.s0,u.s1=c.s1,u.s2=c.s2,u}function o(c,u){var p=new s(c),d=u&&u.state,h=p.next;return h.int32=function(){return p.next()*4294967296|0},h.double=function(){return h()+(h()*2097152|0)*11102230246251565e-32},h.quick=h,d&&(typeof d=="object"&&i(d,p),h.state=function(){return i(p,{})}),h}function l(){var c=4022871197,u=function(p){p=p.toString();for(var d=0;d<p.length;d++){c+=p.charCodeAt(d);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}a&&a.exports?a.exports=o:r&&r.amd?r(function(){return o}):this.alea=o})(e,typeof t=="object"&&t,typeof define=="function"&&define)}),oE=un((e,t)=>{(function(n,a,r){function s(l){var c=this,u="";c.x=0,c.y=0,c.z=0,c.w=0,c.next=function(){var d=c.x^c.x<<11;return c.x=c.y,c.y=c.z,c.z=c.w,c.w^=c.w>>>19^d^d>>>8},l===(l|0)?c.x=l:u+=l;for(var p=0;p<u.length+64;p++)c.x^=u.charCodeAt(p)|0,c.next()}function i(l,c){return c.x=l.x,c.y=l.y,c.z=l.z,c.w=l.w,c}function o(l,c){var u=new s(l),p=c&&c.state,d=function(){return(u.next()>>>0)/4294967296};return d.double=function(){do var h=u.next()>>>11,m=(u.next()>>>0)/4294967296,f=(h+m)/(1<<21);while(f===0);return f},d.int32=u.next,d.quick=d,p&&(typeof p=="object"&&i(p,u),d.state=function(){return i(u,{})}),d}a&&a.exports?a.exports=o:r&&r.amd?r(function(){return o}):this.xor128=o})(e,typeof t=="object"&&t,typeof define=="function"&&define)}),lE=un((e,t)=>{(function(n,a,r){function s(l){var c=this,u="";c.next=function(){var d=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^(d^d<<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 p=0;p<u.length+64;p++)c.x^=u.charCodeAt(p)|0,p==u.length&&(c.d=c.x<<10^c.x>>>4),c.next()}function i(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 o(l,c){var u=new s(l),p=c&&c.state,d=function(){return(u.next()>>>0)/4294967296};return d.double=function(){do var h=u.next()>>>11,m=(u.next()>>>0)/4294967296,f=(h+m)/(1<<21);while(f===0);return f},d.int32=u.next,d.quick=d,p&&(typeof p=="object"&&i(p,u),d.state=function(){return i(u,{})}),d}a&&a.exports?a.exports=o:r&&r.amd?r(function(){return o}):this.xorwow=o})(e,typeof t=="object"&&t,typeof define=="function"&&define)}),uE=un((e,t)=>{(function(n,a,r){function s(l){var c=this;c.next=function(){var p=c.x,d=c.i,h,m,f;return h=p[d],h^=h>>>7,m=h^h<<24,h=p[d+1&7],m^=h^h>>>10,h=p[d+3&7],m^=h^h>>>3,h=p[d+4&7],m^=h^h<<7,h=p[d+7&7],h=h^h<<13,m^=h^h<<9,p[d]=m,c.i=d+1&7,m};function u(p,d){var h,m,f=[];if(d===(d|0))m=f[0]=d;else for(d=""+d,h=0;h<d.length;++h)f[h&7]=f[h&7]<<15^d.charCodeAt(h)+f[h+1&7]<<13;for(;f.length<8;)f.push(0);for(h=0;h<8&&f[h]===0;++h);for(h==8?m=f[7]=-1:m=f[h],p.x=f,p.i=0,h=256;h>0;--h)p.next()}u(c,l)}function i(l,c){return c.x=l.x.slice(),c.i=l.i,c}function o(l,c){l==null&&(l=+new Date);var u=new s(l),p=c&&c.state,d=function(){return(u.next()>>>0)/4294967296};return d.double=function(){do var h=u.next()>>>11,m=(u.next()>>>0)/4294967296,f=(h+m)/(1<<21);while(f===0);return f},d.int32=u.next,d.quick=d,p&&(p.x&&i(p,u),d.state=function(){return i(u,{})}),d}a&&a.exports?a.exports=o:r&&r.amd?r(function(){return o}):this.xorshift7=o})(e,typeof t=="object"&&t,typeof define=="function"&&define)}),cE=un((e,t)=>{(function(n,a,r){function s(l){var c=this;c.next=function(){var p=c.w,d=c.X,h=c.i,m,f;return c.w=p=p+1640531527|0,f=d[h+34&127],m=d[h=h+1&127],f^=f<<13,m^=m<<17,f^=f>>>15,m^=m>>>12,f=d[h]=f^m,c.i=h,f+(p^p>>>16)|0};function u(p,d){var h,m,f,g,y,b=[],v=128;for(d===(d|0)?(m=d,d=null):(d=d+"\0",m=0,v=Math.max(v,d.length)),f=0,g=-32;g<v;++g)d&&(m^=d.charCodeAt((g+32)%d.length)),g===0&&(y=m),m^=m<<10,m^=m>>>15,m^=m<<4,m^=m>>>13,g>=0&&(y=y+1640531527|0,h=b[g&127]^=m+y,f=h==0?f+1:0);for(f>=128&&(b[(d&&d.length||0)&127]=-1),f=127,g=4*128;g>0;--g)m=b[f+34&127],h=b[f=f+1&127],m^=m<<13,h^=h<<17,m^=m>>>15,h^=h>>>12,b[f]=m^h;p.w=y,p.X=b,p.i=f}u(c,l)}function i(l,c){return c.i=l.i,c.w=l.w,c.X=l.X.slice(),c}function o(l,c){l==null&&(l=+new Date);var u=new s(l),p=c&&c.state,d=function(){return(u.next()>>>0)/4294967296};return d.double=function(){do var h=u.next()>>>11,m=(u.next()>>>0)/4294967296,f=(h+m)/(1<<21);while(f===0);return f},d.int32=u.next,d.quick=d,p&&(p.X&&i(p,u),d.state=function(){return i(u,{})}),d}a&&a.exports?a.exports=o:r&&r.amd?r(function(){return o}):this.xor4096=o})(e,typeof t=="object"&&t,typeof define=="function"&&define)}),pE=un((e,t)=>{(function(n,a,r){function s(l){var c=this,u="";c.next=function(){var d=c.b,h=c.c,m=c.d,f=c.a;return d=d<<25^d>>>7^h,h=h-m|0,m=m<<24^m>>>8^f,f=f-d|0,c.b=d=d<<20^d>>>12^h,c.c=h=h-m|0,c.d=m<<16^h>>>16^f,c.a=f-d|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 p=0;p<u.length+20;p++)c.b^=u.charCodeAt(p)|0,c.next()}function i(l,c){return c.a=l.a,c.b=l.b,c.c=l.c,c.d=l.d,c}function o(l,c){var u=new s(l),p=c&&c.state,d=function(){return(u.next()>>>0)/4294967296};return d.double=function(){do var h=u.next()>>>11,m=(u.next()>>>0)/4294967296,f=(h+m)/(1<<21);while(f===0);return f},d.int32=u.next,d.quick=d,p&&(typeof p=="object"&&i(p,u),d.state=function(){return i(u,{})}),d}a&&a.exports?a.exports=o:r&&r.amd?r(function(){return o}):this.tychei=o})(e,typeof t=="object"&&t,typeof define=="function"&&define)}),dE=un(()=>{}),hE=un((e,t)=>{(function(n,a){var r=this,s=256,i=6,o=52,l="random",c=a.pow(s,i),u=a.pow(2,o),p=u*2,d=s-1,h;function m(N,T,C){var $=[];T=T==!0?{entropy:!0}:T||{};var F=b(y(T.entropy?[N,x(n)]:N==null?v():N,3),$),O=new f($),B=function(){for(var W=O.g(i),G=c,q=0;W<u;)W=(W+q)*s,G*=s,q=O.g(1);for(;W>=p;)W/=2,G/=2,q>>>=1;return(W+q)/G};return B.int32=function(){return O.g(4)|0},B.quick=function(){return O.g(4)/4294967296},B.double=B,b(x(O.S),n),(T.pass||C||function(W,G,q,H){return H&&(H.S&&g(H,O),W.state=function(){return g(O,{})}),q?(a[l]=W,G):W})(B,F,"global"in T?T.global:this==a,T.state)}a["seed"+l]=m;function f(N){var T,C=N.length,$=this,F=0,O=$.i=$.j=0,B=$.S=[];for(C||(N=[C++]);F<s;)B[F]=F++;for(F=0;F<s;F++)B[F]=B[O=d&O+N[F%C]+(T=B[F])],B[O]=T;($.g=function(W){for(var G,q=0,H=$.i,K=$.j,X=$.S;W--;)G=X[H=d&H+1],q=q*s+X[d&(X[H]=X[K=d&K+G])+(X[K]=G)];return $.i=H,$.j=K,q})(s)}function g(N,T){return T.i=N.i,T.j=N.j,T.S=N.S.slice(),T}function y(N,T){var C=[],$=typeof N,F;if(T&&$=="object")for(F in N)try{C.push(y(N[F],T-1))}catch(O){}return C.length?C:$=="string"?N:N+"\0"}function b(N,T){for(var C=N+"",$,F=0;F<C.length;)T[d&F]=d&($^=T[d&F]*19)+C.charCodeAt(F++);return x(T)}function v(){try{var N;return h&&(N=h.randomBytes)?N=N(s):(N=new Uint8Array(s),(r.crypto||r.msCrypto).getRandomValues(N)),x(N)}catch($){var T=r.navigator,C=T&&T.plugins;return[+new Date,r,C,r.screen,x(n)]}}function x(N){return String.fromCharCode.apply(0,N)}if(b(a.random(),n),typeof t=="object"&&t.exports){t.exports=m;try{h=dE()}catch(N){}}else typeof define=="function"&&define.amd&&define(function(){return m})})([],Math)}),dd=un((e,t)=>{var n=iE(),a=oE(),r=lE(),s=uE(),i=cE(),o=pE(),l=hE();l.alea=n,l.xor128=a,l.xorwow=r,l.xorshift7=s,l.xor4096=i,l.tychei=o,t.exports=l}),mE=un(()=>{}),rc=un(()=>{}),fE=un(()=>{}),gE=un(()=>{}),yE=un((e,t)=>{var n=function(){var a=typeof document!="undefined"&&document.currentScript?document.currentScript.src:void 0;return typeof __filename!="undefined"&&(a=a||__filename),function(r){r=r||{};function s(){return Q.buffer!=Je&&An(Q.buffer),_n}function i(){return Q.buffer!=Je&&An(Q.buffer),Qt}function o(){return Q.buffer!=Je&&An(Q.buffer),xn}function l(){return Q.buffer!=Je&&An(Q.buffer),ln}function c(){return Q.buffer!=Je&&An(Q.buffer),za}var u=typeof r!="undefined"?r:{},p={},d;for(d in u)u.hasOwnProperty(d)&&(p[d]=u[d]);var h=[],m="./this.program",f=function(w,S){throw S},g=!1,y=!1,b=!1,v=!1;g=typeof window=="object",y=typeof importScripts=="function",b=typeof process=="object"&&typeof process.versions=="object"&&typeof process.versions.node=="string",v=!g&&!b&&!y;var x=u.ENVIRONMENT_IS_PTHREAD||!1;x&&(Je=u.buffer,ra=u.DYNAMIC_BASE,ka=u.DYNAMICTOP_PTR);var N="";function T(w){return u.locateFile?u.locateFile(w,N):N+w}var C,$,F,O,B,W;if(b){y?N=rc().dirname(N)+"/":N=__dirname+"/",C=function(w,S){return B||(B=require("fs")),W||(W=rc()),w=W.normalize(w),B.readFileSync(w,S?null:"utf8")},F=function(w){var S=C(w,!0);return S.buffer||(S=new Uint8Array(S)),ke(S.buffer),S},process.argv.length>1&&(m=process.argv[1].replace(/\\/g,"/")),h=process.argv.slice(2),process.on("uncaughtException",function(w){if(!(w instanceof Bw))throw w}),process.on("unhandledRejection",cr),f=function(w){process.exit(w)},u.inspect=function(){return"[Emscripten Module object]"};var G;try{G=fE()}catch(w){throw console.error('The "worker_threads" module is not supported in this node.js build - perhaps a newer version is needed?'),w}Worker=G.Worker}else v?(typeof read!="undefined"&&(C=function(w){return read(w)}),F=function(w){var S;return typeof readbuffer=="function"?new Uint8Array(readbuffer(w)):(S=read(w,"binary"),ke(typeof S=="object"),S)},typeof scriptArgs!="undefined"?h=scriptArgs:typeof arguments!="undefined"&&(h=arguments),typeof quit=="function"&&(f=function(w){quit(w)}),typeof print!="undefined"&&(typeof console=="undefined"&&(console={}),console.log=print,console.warn=console.error=typeof printErr!="undefined"?printErr:print)):(g||y)&&(y?N=self.location.href:document.currentScript&&(N=document.currentScript.src),typeof a!="undefined"&&a&&(N=a),N.indexOf("blob:")!==0?N=N.substr(0,N.lastIndexOf("/")+1):N="",b?(C=function(w,S){return B||(B=require("fs")),W||(W=rc()),w=W.normalize(w),B.readFileSync(w,S?null:"utf8")},F=function(w){var S=C(w,!0);return S.buffer||(S=new Uint8Array(S)),ke(S.buffer),S}):(C=function(w){var S=new XMLHttpRequest;return S.open("GET",w,!1),S.send(null),S.responseText},y&&(F=function(w){var S=new XMLHttpRequest;return S.open("GET",w,!1),S.responseType="arraybuffer",S.send(null),new Uint8Array(S.response)}),$=function(w,S,R){var U=new XMLHttpRequest;U.open("GET",w,!0),U.responseType="arraybuffer",U.onload=function(){if(U.status==200||U.status==0&&U.response){S(U.response);return}R()},U.onerror=R,U.send(null)}),O=function(w){document.title=w});b&&typeof performance=="undefined"&&(performance=gE().performance);var q=u.print||console.log.bind(console),H=u.printErr||console.warn.bind(console);for(d in p)p.hasOwnProperty(d)&&(u[d]=p[d]);p=null,u.arguments&&(h=u.arguments),u.thisProgram&&(m=u.thisProgram),u.quit&&(f=u.quit);var K=Atomics.load,X=Atomics.store,ne=Atomics.compareExchange,J;u.wasmBinary&&(J=u.wasmBinary);var se;u.noExitRuntime&&(se=u.noExitRuntime),typeof WebAssembly!="object"&&H("no native wasm support detected");var Q,oe=new WebAssembly.Table({initial:169,maximum:169+0,element:"anyfunc"}),ae,pe=0,ue=0,ge=!1,ye=0;function ke(w,S){w||cr("Assertion failed: "+S)}function Se(w){var S=u["_"+w];return ke(S,"Cannot call unknown function "+w+", make sure it is exported"),S}function Ce(w,S,R,U,he){var le={string:function(jn){var Mr=0;if(jn!=null&&jn!==0){var ac=(jn.length<<2)+1;Mr=xo(ac),ot(jn,Mr,ac)}return Mr},array:function(jn){var Mr=xo(jn.length);return ct(jn,Mr),Mr}};function ie(jn){return S==="string"?Ve(jn):S==="boolean"?Boolean(jn):jn}var xe=Se(w),et=[],Mt=0;if(U)for(var rn=0;rn<U.length;rn++){var wo=le[R[rn]];wo?(Mt===0&&(Mt=ec()),et[rn]=wo(U[rn])):et[rn]=U[rn]}var nc=xe.apply(null,et);return nc=ie(nc),Mt!==0&&vo(Mt),nc}function Re(w,S,R,U){R=R||[];var he=R.every(function(ie){return ie==="number"}),le=S!=="string";return le&&he&&!U?Se(w):function(){return Ce(w,S,R,arguments,U)}}function Ke(w,S,R){for(var U=S+R,he="";!(S>=U);){var le=w[S++];if(!le)return he;if(!(le&128)){he+=String.fromCharCode(le);continue}var ie=w[S++]&63;if((le&224)==192){he+=String.fromCharCode((le&31)<<6|ie);continue}var xe=w[S++]&63;if((le&240)==224?le=(le&15)<<12|ie<<6|xe:le=(le&7)<<18|ie<<12|xe<<6|w[S++]&63,le<65536)he+=String.fromCharCode(le);else{var et=le-65536;he+=String.fromCharCode(55296|et>>10,56320|et&1023)}}return he}function Ve(w,S){return w?Ke(i(),w,S):""}function rt(w,S,R,U){if(!(U>0))return 0;for(var he=R,le=R+U-1,ie=0;ie<w.length;++ie){var xe=w.charCodeAt(ie);if(xe>=55296&&xe<=57343){var et=w.charCodeAt(++ie);xe=65536+((xe&1023)<<10)|et&1023}if(xe<=127){if(R>=le)break;S[R++]=xe}else if(xe<=2047){if(R+1>=le)break;S[R++]=192|xe>>6,S[R++]=128|xe&63}else if(xe<=65535){if(R+2>=le)break;S[R++]=224|xe>>12,S[R++]=128|xe>>6&63,S[R++]=128|xe&63}else{if(R+3>=le)break;S[R++]=240|xe>>18,S[R++]=128|xe>>12&63,S[R++]=128|xe>>6&63,S[R++]=128|xe&63}}return S[R]=0,R-he}function ot(w,S,R){return rt(w,i(),S,R)}function He(w){for(var S=0,R=0;R<w.length;++R){var U=w.charCodeAt(R);U>=55296&&U<=57343&&(U=65536+((U&1023)<<10)|w.charCodeAt(++R)&1023),U<=127?++S:U<=2047?S+=2:U<=65535?S+=3:S+=4}return S}function ct(w,S){s().set(w,S)}var pt=65536;function Gn(w,S){return w%S>0&&(w+=S-w%S),w}var Je,_n,Qt,En,na,xn,ln,aa,za;function An(w){Je=w,u.HEAP8=_n=new Int8Array(w),u.HEAP16=En=new Int16Array(w),u.HEAP32=xn=new Int32Array(w),u.HEAPU8=Qt=new Uint8Array(w),u.HEAPU16=na=new Uint16Array(w),u.HEAPU32=ln=new Uint32Array(w),u.HEAPF32=aa=new Float32Array(w),u.HEAPF64=za=new Float64Array(w)}var uo=5256480,zu=uo,wa=13600,ra=5256480,ka=12672,co=u.INITIAL_MEMORY||16777216;if(x)Q=u.wasmMemory,Je=u.buffer;else if(u.wasmMemory)Q=u.wasmMemory;else if(Q=new WebAssembly.Memory({initial:co/pt,maximum:2147483648/pt,shared:!0}),!(Q.buffer instanceof SharedArrayBuffer))throw H("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");Q&&(Je=Q.buffer),co=Je.byteLength,An(Je),x||(o()[ka>>2]=ra);function po(w){for(;w.length>0;){var S=w.shift();if(typeof S=="function"){S(u);continue}var R=S.func;typeof R=="number"?S.arg===void 0?u.dynCall_v(R):u.dynCall_vi(R,S.arg):R(S.arg===void 0?null:S.arg)}}var ys=[],Bu=[],Ff=[],Wu=[],Vp=[],Vu=!1;x&&(Vu=!0);function sa(){if(!x){if(u.preRun)for(typeof u.preRun=="function"&&(u.preRun=[u.preRun]);u.preRun.length;)Rf(u.preRun.shift());po(ys)}}function Up(){Vu=!0,po(Bu)}function $f(){x||po(Ff)}function Df(){if(!x){if(u.postRun)for(typeof u.postRun=="function"&&(u.postRun=[u.postRun]);u.postRun.length;)bs(u.postRun.shift());po(Vp)}}function Rf(w){ys.unshift(w)}function bs(w){Vp.unshift(w)}var ho=Math.ceil,Mf=Math.floor,ur=0,Uu=null,xs=null;function Pf(w){ke(!x,"addRunDependency cannot be used in a pthread worker"),ur++,u.monitorRunDependencies&&u.monitorRunDependencies(ur)}function Of(w){if(ur--,u.monitorRunDependencies&&u.monitorRunDependencies(ur),ur==0&&(Uu!==null&&(clearInterval(Uu),Uu=null),xs)){var S=xs;xs=null,S()}}u.preloadedImages={},u.preloadedAudios={};function cr(w){throw u.onAbort&&u.onAbort(w),x&&console.error("Pthread aborting at "+new Error().stack),w+="",q(w),H(w),ge=!0,ye=1,w="abort("+w+"). Build with -s ASSERTIONS=1 for more info.",new WebAssembly.RuntimeError(w)}function Gu(w,S){return String.prototype.startsWith?w.startsWith(S):w.indexOf(S)===0}var Lf="data:application/octet-stream;base64,";function Gp(w){return Gu(w,Lf)}var zf="file://";function Hp(w){return Gu(w,zf)}var ia="tfjs-backend-wasm-threaded-simd.wasm";Gp(ia)||(ia=T(ia));function jp(){try{if(J)return new Uint8Array(J);if(F)return F(ia);throw"both async and sync fetching of the wasm failed"}catch(w){cr(w)}}function Bf(){return!J&&(g||y)&&typeof fetch=="function"&&!Hp(ia)?fetch(ia,{credentials:"same-origin"}).then(function(w){if(!w.ok)throw"failed to load wasm binary file at '"+ia+"'";return w.arrayBuffer()}).catch(function(){return jp()}):new Promise(function(w,S){w(jp())})}function Wf(){var w={a:$g};function S(ie,xe){var et=ie.exports;if(u.asm=et,ae=xe,!x){var Mt=me.unusedWorkers.length;me.unusedWorkers.forEach(function(rn){me.loadWasmModuleToWorker(rn,function(){--Mt||Of("wasm-instantiate")})})}}x||Pf("wasm-instantiate");function R(ie){S(ie.instance,ie.module)}function U(ie){return Bf().then(function(xe){return WebAssembly.instantiate(xe,w)}).then(ie,function(xe){H("failed to asynchronously prepare wasm: "+xe),cr(xe)})}function he(){if(!J&&typeof WebAssembly.instantiateStreaming=="function"&&!Gp(ia)&&!Hp(ia)&&typeof fetch=="function")fetch(ia,{credentials:"same-origin"}).then(function(ie){var xe=WebAssembly.instantiateStreaming(ie,w);return xe.then(R,function(et){H("wasm streaming compile failed: "+et),H("falling back to ArrayBuffer instantiation"),U(R)})});else return U(R)}if(u.instantiateWasm)try{var le=u.instantiateWasm(w,S);return le}catch(ie){return H("Module.instantiateWasm callback failed with error: "+ie),!1}return he(),{}}var Vf={};function Uf(){me.initRuntime()}x||Bu.push({func:function(){qu()}});var qp=0,Kp=0,Xp=0;function mo(w,S,R){w=w|0,S=S|0,R=R|0,qp=w,Xp=S,Kp=R}u.__register_pthread_ptr=mo;var Hu={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},fo=13584;function go(w,S){if(w<=0||w>s().length||w&!0||S<0)return-28;if(S==0)return 0;S>=2147483647&&(S=Infinity);var R=Atomics.load(o(),fo>>2),U=0;if(R==w){var he=Atomics.compareExchange(o(),fo>>2,R,0);if(he==R&&(--S,U=1,S<=0))return 1}var le=Atomics.notify(o(),w>>2,S);if(le>=0)return le+U;throw"Atomics.notify returned an unexpected value "+le}u._emscripten_futex_wake=go;function Gf(w){if(x)throw"Internal Error! _kill_thread() can only ever be called from main application thread!";if(!w)throw"Internal Error! Null pthread_ptr in _kill_thread!";o()[w+12>>2]=0;var S=me.pthreads[w];S.worker.terminate(),me.freeThreadData(S),me.runningWorkers.splice(me.runningWorkers.indexOf(S.worker),1),S.worker.pthread=void 0}function Hf(w){if(x)throw"Internal Error! _cancel_thread() can only ever be called from main application thread!";if(!w)throw"Internal Error! Null pthread_ptr in _cancel_thread!";var S=me.pthreads[w];S.worker.postMessage({cmd:"cancel"})}function jf(w){if(x)throw"Internal Error! _cleanup_thread() can only ever be called from main application thread!";if(!w)throw"Internal Error! Null pthread_ptr in _cleanup_thread!";o()[w+12>>2]=0;var S=me.pthreads[w];if(S){var R=S.worker;me.returnWorkerToPool(R)}}var me={MAIN_THREAD_ID:1,mainThreadInfo:{schedPolicy:0,schedPrio:0},unusedWorkers:[],runningWorkers:[],initRuntime:function(){mo(me.mainThreadBlock,!y,1),Pw(me.mainThreadBlock)},initMainThreadBlock:function(){for(var w=8,S=0;S<w;++S)me.allocateUnusedWorker();me.mainThreadBlock=12832;for(var S=0;S<232/4;++S)l()[me.mainThreadBlock/4+S]=0;o()[me.mainThreadBlock+12>>2]=me.mainThreadBlock;var R=me.mainThreadBlock+156;o()[R>>2]=R;for(var U=13072,S=0;S<128;++S)l()[U/4+S]=0;Atomics.store(l(),me.mainThreadBlock+104>>2,U),Atomics.store(l(),me.mainThreadBlock+40>>2,me.mainThreadBlock),Atomics.store(l(),me.mainThreadBlock+44>>2,42)},initWorker:function(){},pthreads:{},exitHandlers:null,setThreadStatus:function(){},runExitHandlers:function(){if(me.exitHandlers!==null){for(;me.exitHandlers.length>0;)me.exitHandlers.pop()();me.exitHandlers=null}x&&pe&&Mw()},threadExit:function(w){var S=Ba();S&&(Atomics.store(l(),S+4>>2,w),Atomics.store(l(),S+0>>2,1),Atomics.store(l(),S+60>>2,1),Atomics.store(l(),S+64>>2,0),me.runExitHandlers(),go(S+0,2147483647),mo(0,0,0),pe=0,x&&postMessage({cmd:"exit"}))},threadCancel:function(){me.runExitHandlers(),Atomics.store(l(),pe+4>>2,-1),Atomics.store(l(),pe+0>>2,1),go(pe+0,2147483647),pe=ue=0,mo(0,0,0),postMessage({cmd:"cancelDone"})},terminateAllThreads:function(){for(var w in me.pthreads){var S=me.pthreads[w];S&&S.worker&&me.returnWorkerToPool(S.worker)}me.pthreads={};for(var R=0;R<me.unusedWorkers.length;++R){var U=me.unusedWorkers[R];U.terminate()}me.unusedWorkers=[];for(var R=0;R<me.runningWorkers.length;++R){var U=me.runningWorkers[R],S=U.pthread;me.freeThreadData(S),U.terminate()}me.runningWorkers=[]},freeThreadData:function(w){if(w){if(w.threadInfoStruct){var S=o()[w.threadInfoStruct+104>>2];o()[w.threadInfoStruct+104>>2]=0,Qu(S),Qu(w.threadInfoStruct)}w.threadInfoStruct=0,w.allocatedOwnStack&&w.stackBase&&Qu(w.stackBase),w.stackBase=0,w.worker&&(w.worker.pthread=null)}},returnWorkerToPool:function(w){delete me.pthreads[w.pthread.thread],me.unusedWorkers.push(w),me.runningWorkers.splice(me.runningWorkers.indexOf(w),1),me.freeThreadData(w.pthread),w.pthread=void 0},receiveObjectTransfer:function(w){},loadWasmModuleToWorker:function(w,S){w.onmessage=function(R){var U=R.data,he=U.cmd;if(w.pthread&&(me.currentProxiedOperationCallerThread=w.pthread.threadInfoStruct),U.targetThread&&U.targetThread!=Ba()){var le=me.pthreads[U.targetThread];le?le.worker.postMessage(R.data,U.transferList):console.error('Internal error! Worker sent a message "'+he+'" to target pthread '+U.targetThread+", but that thread no longer exists!"),me.currentProxiedOperationCallerThread=void 0;return}if(he==="processQueuedMainThreadWork")Mg();else if(he==="spawnThread")td(R.data);else if(he==="cleanupThread")jf(U.thread);else if(he==="killThread")Gf(U.thread);else if(he==="cancelThread")Hf(U.thread);else if(he==="loaded")w.loaded=!0,S&&S(w),w.runPthread&&(w.runPthread(),delete w.runPthread);else if(he==="print")q("Thread "+U.threadId+": "+U.text);else if(he==="printErr")H("Thread "+U.threadId+": "+U.text);else if(he==="alert")alert("Thread "+U.threadId+": "+U.text);else if(he==="exit"){var ie=w.pthread&&Atomics.load(l(),w.pthread.thread+68>>2);ie&&me.returnWorkerToPool(w)}else he==="cancelDone"?me.returnWorkerToPool(w):he==="objectTransfer"?me.receiveObjectTransfer(R.data):R.data.target==="setimmediate"?w.postMessage(R.data):H("worker sent an unknown command "+he);me.currentProxiedOperationCallerThread=void 0},w.onerror=function(R){H("pthread sent an error! "+R.filename+":"+R.lineno+": "+R.message)},b&&(w.on("message",function(R){w.onmessage({data:R})}),w.on("error",function(R){w.onerror(R)}),w.on("exit",function(R){console.log("worker exited - TODO: update the worker queue?")})),w.postMessage({cmd:"load",urlOrBlob:u.mainScriptUrlOrBlob||a,wasmMemory:Q,wasmModule:ae,DYNAMIC_BASE:ra,DYNAMICTOP_PTR:ka})},allocateUnusedWorker:function(){var w=T("tfjs-backend-wasm-threaded-simd.worker.js");me.unusedWorkers.push(new Worker(w))},getNewWorker:function(){return me.unusedWorkers.length==0&&(me.allocateUnusedWorker(),me.loadWasmModuleToWorker(me.unusedWorkers[0])),me.unusedWorkers.length>0?me.unusedWorkers.pop():null},busySpinWait:function(w){for(var S=performance.now()+w;performance.now()<S;);}};function qf(w,S){uo=zu=w,wa=S,vo(w)}u.establishStackSpace=qf;function Kf(){return se}u.getNoExitRuntime=Kf;function Xf(w,S,R,U){cr("Assertion failed: "+Ve(w)+", at: "+[S?Ve(S):"unknown filename",R,U?Ve(U):"unknown function"])}function Yf(w,S){var R=_main(w,S)}var vs;b?vs=function(){var w=process.hrtime();return w[0]*1e3+w[1]/1e6}:x?vs=function(){return performance.now()-u.__performance_now_clock_drift}:typeof dateNow!="undefined"?vs=dateNow:vs=function(){return performance.now()};function Zf(w){return o()[$w()>>2]=w,w}function Jf(w,S){if(x)return $r(1,1,w,S);Wu.unshift({func:w,arg:S})}function Qf(w,S){if(w==S)postMessage({cmd:"processQueuedMainThreadWork"});else if(x)postMessage({targetThread:w,cmd:"processThreadQueue"});else{var R=me.pthreads[w],U=R&&R.worker;if(!U)return;U.postMessage({cmd:"processThreadQueue"})}return 1}function eg(){cr()}function tg(w,S){w=w|0,S=S|0}function ng(w,S,R){if(w<=0||w>s().length||w&!0)return-28;if(y){var U=Atomics.wait(o(),w>>2,S,R);if(U==="timed-out")return-73;if(U==="not-equal")return-6;if(U==="ok")return 0;throw"Atomics.wait returned an unexpected value "+U}else{var he=Atomics.load(o(),w>>2);if(S!=he)return-6;var le=performance.now(),ie=le+R;Atomics.store(o(),fo>>2,w);for(var xe=w;w==xe;){if(le=performance.now(),le>ie)return-73;Mg(),w=Atomics.load(o(),fo>>2)}return 0}}function ag(){return Xp|0}function rg(){return Kp|0}function sg(w,S,R){i().copyWithin(w,S,S+R)}function ig(){return navigator.hardwareConcurrency}function $r(w,S){for(var R=arguments.length-2,U=ec(),he=xo(R*8),le=he>>3,ie=0;ie<R;ie++)c()[le+ie]=arguments[2+ie];var xe=Lw(w,R,he,S);return vo(U),xe}var ws=[];function yo(w,S){yo.array||(yo.array=[]);var R=yo.array;R.length=0;for(var U;U=i()[w++];)U===100||U===102?(S=S+7&~7,R.push(c()[S>>3]),S+=8):(S=S+3&~3,R.push(o()[S>>2]),S+=4);return R}function og(w,S,R){ws.length=S;for(var U=R>>3,he=0;he<S;he++)ws[he]=c()[U+he];var le=w<0,ie=le?Vf[-w-1]:Fg[w];if(le){var xe=ws[1],et=ws[2],Mt=yo(xe,et);return ie.apply(null,Mt)}return ie.apply(null,ws)}function lg(){return i().length}function ug(w){try{return Q.grow(w-Je.byteLength+65535>>>16),An(Q.buffer),1}catch(S){}}function cg(w){w=w>>>0;var S=lg();if(w<=S)return!1;var R=65536,U=2147483648;if(w>U)return!1;for(var he=16777216,le=1;le<=4;le*=2){var ie=S*(1+.2/le);ie=Math.min(ie,w+100663296);var xe=Math.min(U,Gn(Math.max(he,w,ie),R)),et=ug(xe);if(et)return!0}return!1}var Pe={keyEvent:0,mouseEvent:0,wheelEvent:0,uiEvent:0,focusEvent:0,deviceOrientationEvent:0,deviceMotionEvent:0,fullscreenChangeEvent:0,pointerlockChangeEvent:0,visibilityChangeEvent:0,touchEvent:0,previousFullscreenElement:null,previousScreenX:null,previousScreenY:null,removeEventListenersRegistered:!1,removeAllEventListeners:function(){for(var w=Pe.eventHandlers.length-1;w>=0;--w)Pe._removeHandler(w);Pe.eventHandlers=[],Pe.deferredCalls=[]},registerRemoveEventListeners:function(){Pe.removeEventListenersRegistered||(Wu.push(Pe.removeAllEventListeners),Pe.removeEventListenersRegistered=!0)},deferredCalls:[],deferCall:function(w,S,R){function U(ie,xe){if(ie.length!=xe.length)return!1;for(var et in ie)if(ie[et]!=xe[et])return!1;return!0}for(var he in Pe.deferredCalls){var le=Pe.deferredCalls[he];if(le.targetFunction==w&&U(le.argsList,R))return}Pe.deferredCalls.push({targetFunction:w,precedence:S,argsList:R}),Pe.deferredCalls.sort(function(ie,xe){return ie.precedence<xe.precedence})},removeDeferredCalls:function(w){for(var S=0;S<Pe.deferredCalls.length;++S)Pe.deferredCalls[S].targetFunction==w&&(Pe.deferredCalls.splice(S,1),--S)},canPerformEventHandlerRequests:function(){return Pe.inEventHandler&&Pe.currentEventHandler.allowsDeferredCalls},runDeferredCalls:function(){if(Pe.canPerformEventHandlerRequests())for(var w=0;w<Pe.deferredCalls.length;++w){var S=Pe.deferredCalls[w];Pe.deferredCalls.splice(w,1),--w,S.targetFunction.apply(null,S.argsList)}},inEventHandler:0,currentEventHandler:null,eventHandlers:[],removeAllHandlersOnTarget:function(w,S){for(var R=0;R<Pe.eventHandlers.length;++R)Pe.eventHandlers[R].target==w&&(!S||S==Pe.eventHandlers[R].eventTypeString)&&Pe._removeHandler(R--)},_removeHandler:function(w){var S=Pe.eventHandlers[w];S.target.removeEventListener(S.eventTypeString,S.eventListenerFunc,S.useCapture),Pe.eventHandlers.splice(w,1)},registerOrRemoveHandler:function(w){var S=function(U){++Pe.inEventHandler,Pe.currentEventHandler=w,Pe.runDeferredCalls(),w.handlerFunc(U),Pe.runDeferredCalls(),--Pe.inEventHandler};if(w.callbackfunc)w.eventListenerFunc=S,w.target.addEventListener(w.eventTypeString,S,w.useCapture),Pe.eventHandlers.push(w),Pe.registerRemoveEventListeners();else for(var R=0;R<Pe.eventHandlers.length;++R)Pe.eventHandlers[R].target==w.target&&Pe.eventHandlers[R].eventTypeString==w.eventTypeString&&Pe._removeHandler(R--)},queueEventHandlerOnThread_iiii:function(w,S,R,U,he){var le=ec(),ie=xo(12);o()[ie>>2]=R,o()[ie+4>>2]=U,o()[ie+8>>2]=he,Pg(w,637534208,S,U,ie),vo(le)},getTargetThreadForEventCallback:function(w){switch(w){case 1:return 0;case 2:return me.currentProxiedOperationCallerThread;default:return w}},getNodeNameForTarget:function(w){return w?w==window?"#window":w==screen?"#screen":w&&w.nodeName?w.nodeName:"":""},fullscreenEnabled:function(){return document.fullscreenEnabled||document.webkitFullscreenEnabled}};function pg(w){var S=He(w)+1,R=Ju(S);return ot(w,R,S),R}function dg(w,S,R,U){var he=ec(),le=xo(12),ie=0;S&&(ie=pg(S)),o()[le>>2]=ie,o()[le+4>>2]=R,o()[le+8>>2]=U,Pg(w,657457152,0,ie,le),vo(he)}function hg(w,S,R,U){S=S?Ve(S):"",dg(w,S,R,U)}function mg(w){return w>2?Ve(w):w}var fg=[0,typeof document!="undefined"?document:0,typeof window!="undefined"?window:0];function gg(w){w=mg(w);var S=fg[w]||(typeof document!="undefined"?document.querySelector(w):void 0);return S}function ju(w){return gg(w)}function Yp(w,S,R){var U=ju(w);if(!U)return-4;if(U.canvasSharedPtr&&(o()[U.canvasSharedPtr>>2]=S,o()[U.canvasSharedPtr+4>>2]=R),U.offscreenCanvas||!U.controlTransferredOffscreen){U.offscreenCanvas&&(U=U.offscreenCanvas);var he=!1;if(U.GLctxObject&&U.GLctxObject.GLctx){var le=U.GLctxObject.GLctx.getParameter(2978);he=le[0]===0&&le[1]===0&&le[2]===U.width&&le[3]===U.height}U.width=S,U.height=R,he&&U.GLctxObject.GLctx.viewport(0,0,S,R)}else if(U.canvasSharedPtr){var ie=o()[U.canvasSharedPtr+8>>2];return hg(ie,w,S,R),1}else return-4;return 0}function Zp(w,S,R){return x?$r(2,1,w,S,R):Yp(w,S,R)}function yg(w,S,R){var U=ju(w);return U?Yp(w,S,R):Zp(w,S,R)}function bg(w){w=w|0}function xg(w,S){w=w|0,S=S|0}function vg(w){var S=w.getExtension("ANGLE_instanced_arrays");if(S)return w.vertexAttribDivisor=function(R,U){S.vertexAttribDivisorANGLE(R,U)},w.drawArraysInstanced=function(R,U,he,le){S.drawArraysInstancedANGLE(R,U,he,le)},w.drawElementsInstanced=function(R,U,he,le,ie){S.drawElementsInstancedANGLE(R,U,he,le,ie)},1}function wg(w){var S=w.getExtension("OES_vertex_array_object");if(S)return w.createVertexArray=function(){return S.createVertexArrayOES()},w.deleteVertexArray=function(R){S.deleteVertexArrayOES(R)},w.bindVertexArray=function(R){S.bindVertexArrayOES(R)},w.isVertexArray=function(R){return S.isVertexArrayOES(R)},1}function kg(w){var S=w.getExtension("WEBGL_draw_buffers");if(S)return w.drawBuffers=function(R,U){S.drawBuffersWEBGL(R,U)},1}var Be={counter:1,lastError:0,buffers:[],mappedBuffers:{},programs:[],framebuffers:[],renderbuffers:[],textures:[],uniforms:[],shaders:[],vaos:[],contexts:{},currentContext:null,offscreenCanvases:{},timerQueriesEXT:[],programInfos:{},stringCache:{},unpackAlignment:4,init:function(){for(var w=new Float32Array(Be.MINI_TEMP_BUFFER_SIZE),S=0;S<Be.MINI_TEMP_BUFFER_SIZE;S++)Be.miniTempBufferFloatViews[S]=w.subarray(0,S+1);for(var R=new Int32Array(Be.MINI_TEMP_BUFFER_SIZE),S=0;S<Be.MINI_TEMP_BUFFER_SIZE;S++)Be.miniTempBufferIntViews[S]=R.subarray(0,S+1)},recordError:function(w){Be.lastError||(Be.lastError=w)},getNewId:function(w){for(var S=Be.counter++,R=w.length;R<S;R++)w[R]=null;return S},MINI_TEMP_BUFFER_SIZE:256,miniTempBufferFloatViews:[0],miniTempBufferIntViews:[0],getSource:function(w,S,R,U){for(var he="",le=0;le<S;++le){var ie=U?o()[U+le*4>>2]:-1;he+=Ve(o()[R+le*4>>2],ie<0?void 0:ie)}return he},createContext:function(w,S){var R=w.getContext("webgl",S);if(!R)return 0;var U=Be.registerContext(R,S);return U},registerContext:function(w,S){var R=Ju(8);o()[R+4>>2]=Ba();var U={handle:R,attributes:S,version:S.majorVersion,GLctx:w};return w.canvas&&(w.canvas.GLctxObject=U),Be.contexts[R]=U,(typeof S.enableExtensionsByDefault=="undefined"||S.enableExtensionsByDefault)&&Be.initExtensions(U),R},makeContextCurrent:function(w){return Be.currentContext=Be.contexts[w],u.ctx=Dr=Be.currentContext&&Be.currentContext.GLctx,!(w&&!Dr)},getContext:function(w){return Be.contexts[w]},deleteContext:function(w){Be.currentContext===Be.contexts[w]&&(Be.currentContext=null),typeof Pe=="object"&&Pe.removeAllHandlersOnTarget(Be.contexts[w].GLctx.canvas),Be.contexts[w]&&Be.contexts[w].GLctx.canvas&&(Be.contexts[w].GLctx.canvas.GLctxObject=void 0),Qu(Be.contexts[w].handle),Be.contexts[w]=null},initExtensions:function(w){if(w||(w=Be.currentContext),!w.initExtensionsDone){w.initExtensionsDone=!0;var S=w.GLctx;vg(S),wg(S),kg(S),S.disjointTimerQueryExt=S.getExtension("EXT_disjoint_timer_query");var R=["OES_texture_float","OES_texture_half_float","OES_standard_derivatives","OES_vertex_array_object","WEBGL_compressed_texture_s3tc","WEBGL_depth_texture","OES_element_index_uint","EXT_texture_filter_anisotropic","EXT_frag_depth","WEBGL_draw_buffers","ANGLE_instanced_arrays","OES_texture_float_linear","OES_texture_half_float_linear","EXT_blend_minmax","EXT_shader_texture_lod","EXT_texture_norm16","WEBGL_compressed_texture_pvrtc","EXT_color_buffer_half_float","WEBGL_color_buffer_float","EXT_sRGB","WEBGL_compressed_texture_etc1","EXT_disjoint_timer_query","WEBGL_compressed_texture_etc","WEBGL_compressed_texture_astc","EXT_color_buffer_float","WEBGL_compressed_texture_s3tc_srgb","EXT_disjoint_timer_query_webgl2","WEBKIT_WEBGL_compressed_texture_pvrtc"],U=S.getSupportedExtensions()||[];U.forEach(function(he){R.indexOf(he)!=-1&&S.getExtension(he)})}},populateUniformTable:function(w){for(var S=Be.programs[w],R=Be.programInfos[w]={uniforms:{},maxUniformLength:0,maxAttributeLength:-1,maxUniformBlockNameLength:-1},U=R.uniforms,he=Dr.getProgramParameter(S,35718),le=0;le<he;++le){var ie=Dr.getActiveUniform(S,le),xe=ie.name;R.maxUniformLength=Math.max(R.maxUniformLength,xe.length+1),xe.slice(-1)=="]"&&(xe=xe.slice(0,xe.lastIndexOf("[")));var et=Dr.getUniformLocation(S,xe);if(et){var Mt=Be.getNewId(Be.uniforms);U[xe]=[ie.size,Mt],Be.uniforms[Mt]=et;for(var rn=1;rn<ie.size;++rn){var wo=xe+"["+rn+"]";et=Dr.getUniformLocation(S,wo),Mt=Be.getNewId(Be.uniforms),Be.uniforms[Mt]=et}}}}},Ig=["default","low-power","high-performance"];function Ng(w,S){var R={},U=S>>2;R.alpha=!!o()[U+(0>>2)],R.depth=!!o()[U+(4>>2)],R.stencil=!!o()[U+(8>>2)],R.antialias=!!o()[U+(12>>2)],R.premultipliedAlpha=!!o()[U+(16>>2)],R.preserveDrawingBuffer=!!o()[U+(20>>2)];var he=o()[U+(24>>2)];R.powerPreference=Ig[he],R.failIfMajorPerformanceCaveat=!!o()[U+(28>>2)],R.majorVersion=o()[U+(32>>2)],R.minorVersion=o()[U+(36>>2)],R.enableExtensionsByDefault=o()[U+(40>>2)],R.explicitSwapControl=o()[U+(44>>2)],R.proxyContextToMainThread=o()[U+(48>>2)],R.renderViaOffscreenBackBuffer=o()[U+(52>>2)];var le=ju(w);if(!le)return-4;if(R.explicitSwapControl)return-1;var ie=Be.createContext(le,R);return ie}function Tg(w,S){return Ng(w,S)}var ks={splitPath:function(w){var S=/^(\/?|)([\s\S]*?)((?:\.{1,2}|[^\/]+?|)(\.[^.\/]*|))(?:[\/]*)$/;return S.exec(w).slice(1)},normalizeArray:function(w,S){for(var R=0,U=w.length-1;U>=0;U--){var he=w[U];he==="."?w.splice(U,1):he===".."?(w.splice(U,1),R++):R&&(w.splice(U,1),R--)}if(S)for(;R;R--)w.unshift("..");return w},normalize:function(w){var S=w.charAt(0)==="/",R=w.substr(-1)==="/";return w=ks.normalizeArray(w.split("/").filter(function(U){return!!U}),!S).join("/"),!w&&!S&&(w="."),w&&R&&(w+="/"),(S?"/":"")+w},dirname:function(w){var S=ks.splitPath(w),R=S[0],U=S[1];return!R&&!U?".":(U&&(U=U.substr(0,U.length-1)),R+U)},basename:function(w){if(w==="/")return"/";var S=w.lastIndexOf("/");return S===-1?w:w.substr(S+1)},extname:function(w){return ks.splitPath(w)[3]},join:function(){var w=Array.prototype.slice.call(arguments,0);return ks.normalize(w.join("/"))},join2:function(w,S){return ks.normalize(w+"/"+S)}},bo={mappings:{},buffers:[null,[],[]],printChar:function(w,S){var R=bo.buffers[w];S===0||S===10?((w===1?q:H)(Ke(R,0)),R.length=0):R.push(S)},varargs:void 0,get:function(){bo.varargs+=4;var w=o()[bo.varargs-4>>2];return w},getStr:function(w){var S=Ve(w);return S},get64:function(w,S){return w}};function Jp(w){return x?$r(3,1,w):0}function Qp(w,S,R,U,he){if(x)return $r(4,1,w,S,R,U,he)}function ed(w,S,R,U){if(x)return $r(5,1,w,S,R,U);for(var he=0,le=0;le<R;le++){for(var ie=o()[S+le*8>>2],xe=o()[S+(le*8+4)>>2],et=0;et<xe;et++)bo.printChar(w,i()[ie+et]);he+=xe}return o()[U>>2]=he,0}function Sg(w){var S=me.exitHandlers.pop();w&&S()}function Cg(w,S){me.exitHandlers===null&&(me.exitHandlers=[]),me.exitHandlers.push(function(){zw(w,S)})}function td(w){if(x)throw"Internal Error! _spawn_thread() can only ever be called from main application thread!";var S=me.getNewWorker();if(S.pthread!==void 0)throw"Internal error!";if(!w.pthread_ptr)throw"Internal error, no pthread ptr!";me.runningWorkers.push(S);for(var R=Ju(128*4),U=0;U<128;++U)o()[R+U*4>>2]=0;var he=w.stackBase+w.stackSize,le=me.pthreads[w.pthread_ptr]={worker:S,stackBase:w.stackBase,stackSize:w.stackSize,allocatedOwnStack:w.allocatedOwnStack,thread:w.pthread_ptr,threadInfoStruct:w.pthread_ptr},ie=le.threadInfoStruct>>2;Atomics.store(l(),ie+(0>>2),0),Atomics.store(l(),ie+(4>>2),0),Atomics.store(l(),ie+(8>>2),0),Atomics.store(l(),ie+(68>>2),w.detached),Atomics.store(l(),ie+(104>>2),R),Atomics.store(l(),ie+(48>>2),0),Atomics.store(l(),ie+(40>>2),le.threadInfoStruct),Atomics.store(l(),ie+(44>>2),42),Atomics.store(l(),ie+(108>>2),w.stackSize),Atomics.store(l(),ie+(84>>2),w.stackSize),Atomics.store(l(),ie+(80>>2),he),Atomics.store(l(),ie+(108+8>>2),he),Atomics.store(l(),ie+(108+12>>2),w.detached),Atomics.store(l(),ie+(108+20>>2),w.schedPolicy),Atomics.store(l(),ie+(108+24>>2),w.schedPrio);var xe=Dw(),et=xe+40;Atomics.store(l(),ie+(176>>2),et),S.pthread=le;var Mt={cmd:"run",start_routine:w.startRoutine,arg:w.arg,threadInfoStruct:w.pthread_ptr,selfThreadId:w.pthread_ptr,parentThreadId:w.parent_pthread_ptr,stackBase:w.stackBase,stackSize:w.stackSize};S.runPthread=function(){Mt.time=performance.now(),S.postMessage(Mt,w.transferList)},S.loaded&&(S.runPthread(),delete S.runPthread)}function _g(w,S,R){if(!S&&!R)return Hu.EINVAL;if(!w)return H("pthread_getschedparam called with a null thread pointer!"),Hu.ESRCH;var U=o()[w+12>>2];if(U!==w)return H("pthread_getschedparam attempted on thread "+w+", which does not point to a valid thread, or does not exist anymore!"),Hu.ESRCH;var he=Atomics.load(l(),w+108+20>>2),le=Atomics.load(l(),w+108+24>>2);return S&&(o()[S>>2]=he),R&&(o()[R>>2]=le),0}function Ba(){return qp|0}u._pthread_self=Ba;function Eg(w,S,R,U){if(typeof SharedArrayBuffer=="undefined")return H("Current environment does not support SharedArrayBuffer, pthreads are not available!"),6;if(!w)return H("pthread_create called with a null thread pointer!"),28;var he=[],le=0;if(x&&(he.length===0||le))return Ow(687865856,w,S,R,U);if(le)return le;var ie=0,xe=0,et=0,Mt=0,rn=0;if(S){ie=o()[S>>2],ie+=81920,xe=o()[S+8>>2],et=o()[S+12>>2]!==0;var wo=o()[S+16>>2]===0;if(wo){var nc=o()[S+20>>2],jn=o()[S+24>>2],Mr=me.currentProxiedOperationCallerThread?me.currentProxiedOperationCallerThread:Ba();_g(Mr,S+20,S+24),Mt=o()[S+20>>2],rn=o()[S+24>>2],o()[S+20>>2]=nc,o()[S+24>>2]=jn}else Mt=o()[S+20>>2],rn=o()[S+24>>2]}else ie=2097152;var ac=xe==0;ac?xe=Rw(16,ie):(xe-=ie,ke(xe>0));for(var ko=Ju(232),Lg=0;Lg<232>>2;++Lg)l()[(ko>>2)+Lg]=0;o()[w>>2]=ko,o()[ko+12>>2]=ko;var Ww=ko+156;o()[Ww>>2]=Ww;var zg={stackBase:xe,stackSize:ie,allocatedOwnStack:ac,schedPolicy:Mt,schedPrio:rn,detached:et,startRoutine:R,pthread_ptr:ko,parent_pthread_ptr:Ba(),arg:U,transferList:he};return x?(zg.cmd="spawnThread",postMessage(zg,he)):td(zg),0}function Ag(w){return w=+w,w>=0?+Mf(w+.5):+ho(w-.5)}function nd(w){if(x)return $r(6,1,w);switch(w){case 30:return 16384;case 85:var S=2147483648;return S/16384;case 132:case 133:case 12:case 137:case 138:case 15:case 235:case 16:case 17:case 18:case 19:case 20:case 149:case 13:case 10:case 236:case 153:case 9:case 21:case 22:case 159:case 154:case 14:case 77:case 78:case 139:case 80:case 81:case 82:case 68:case 67:case 164:case 11:case 29:case 47:case 48:case 95:case 52:case 51:case 46:case 79:return 200809;case 27:case 246:case 127:case 128:case 23:case 24:case 160:case 161:case 181:case 182:case 242:case 183:case 184:case 243:case 244:case 245:case 165:case 178:case 179:case 49:case 50:case 168:case 169:case 175:case 170:case 171:case 172:case 97:case 76:case 32:case 173:case 35:return-1;case 176:case 177:case 7:case 155:case 8:case 157:case 125:case 126:case 92:case 93:case 129:case 130:case 131:case 94:case 91:return 1;case 74:case 60:case 69:case 70:case 4:return 1024;case 31:case 42:case 72:return 32;case 87:case 26:case 33:return 2147483647;case 34:case 1:return 47839;case 38:case 36:return 99;case 43:case 37:return 2048;case 0:return 2097152;case 3:return 65536;case 28:return 32768;case 44:return 32767;case 75:return 16384;case 39:return 1e3;case 89:return 700;case 71:return 256;case 40:return 255;case 2:return 100;case 180:return 64;case 25:return 20;case 5:return 16;case 6:return 6;case 73:return 4;case 84:return typeof navigator=="object"&&navigator.hardwareConcurrency||1}return Zf(28),-1}x?me.initWorker():me.initMainThreadBlock();var Dr;Be.init();var Fg=[null,Jf,Zp,Jp,Qp,ed,nd],$g={e:Xf,r:Yf,w:Qf,a:eg,l:tg,d:ng,c:go,h:vs,g:ag,x:rg,q:sg,B:ig,t:og,A:cg,u:yg,k:bg,s:xg,v:Tg,m:Jp,o:Qp,i:ed,p:Uf,memory:Q||u.wasmMemory,y:Sg,z:Cg,j:Eg,b:Ba,f:Ag,n:nd,table:oe},ad=Wf();u.asm=ad;var qu=u.___wasm_call_ctors=function(){return(qu=u.___wasm_call_ctors=u.asm.C).apply(null,arguments)},Ku=u._init=function(){return(Ku=u._init=u.asm.D).apply(null,arguments)},rd=u._register_tensor=function(){return(rd=u._register_tensor=u.asm.E).apply(null,arguments)},Is=u._dispose_data=function(){return(Is=u._dispose_data=u.asm.F).apply(null,arguments)},Xu=u._dispose=function(){return(Xu=u._dispose=u.asm.G).apply(null,arguments)},Dg=u._Abs=function(){return(Dg=u._Abs=u.asm.H).apply(null,arguments)},Rg=u._Add=function(){return(Rg=u._Add=u.asm.I).apply(null,arguments)},Yu=u._AddN=function(){return(Yu=u._AddN=u.asm.J).apply(null,arguments)},sd=u._ArgMax=function(){return(sd=u._ArgMax=u.asm.K).apply(null,arguments)},id=u._AvgPool=function(){return(id=u._AvgPool=u.asm.L).apply(null,arguments)},V=u._BatchMatMul=function(){return(V=u._BatchMatMul=u.asm.M).apply(null,arguments)},te=u._ClipByValue=function(){return(te=u._ClipByValue=u.asm.N).apply(null,arguments)},Ie=u._Conv2D=function(){return(Ie=u._Conv2D=u.asm.O).apply(null,arguments)},_e=u._Conv2DBackpropInput=function(){return(_e=u._Conv2DBackpropInput=u.asm.P).apply(null,arguments)},at=u._Cos=function(){return(at=u._Cos=u.asm.Q).apply(null,arguments)},_t=u._CropAndResize=function(){return(_t=u._CropAndResize=u.asm.R).apply(null,arguments)},Xe=u._Cumsum=function(){return(Xe=u._Cumsum=u.asm.S).apply(null,arguments)},je=u._DepthToSpace=function(){return(je=u._DepthToSpace=u.asm.T).apply(null,arguments)},Ut=u._DepthwiseConv2dNative=function(){return(Ut=u._DepthwiseConv2dNative=u.asm.U).apply(null,arguments)},pr=u._Equal=function(){return(pr=u._Equal=u.asm.V).apply(null,arguments)},dr=u._Exp=function(){return(dr=u._Exp=u.asm.W).apply(null,arguments)},od=u._FlipLeftRight=function(){return(od=u._FlipLeftRight=u.asm.X).apply(null,arguments)},Zu=u._Floor=function(){return(Zu=u._Floor=u.asm.Y).apply(null,arguments)},Hn=u._FloorDiv=function(){return(Hn=u._FloorDiv=u.asm.Z).apply(null,arguments)},Rr=u._FusedBatchNorm=function(){return(Rr=u._FusedBatchNorm=u.asm._).apply(null,arguments)},ld=u._FusedConv2D=function(){return(ld=u._FusedConv2D=u.asm.$).apply(null,arguments)},FC=u._FusedDepthwiseConv2D=function(){return(FC=u._FusedDepthwiseConv2D=u.asm.aa).apply(null,arguments)},$C=u._Gather=function(){return($C=u._Gather=u.asm.ba).apply(null,arguments)},DC=u._GatherNd=function(){return(DC=u._GatherNd=u.asm.ca).apply(null,arguments)},RC=u._Greater=function(){return(RC=u._Greater=u.asm.da).apply(null,arguments)},MC=u._GreaterEqual=function(){return(MC=u._GreaterEqual=u.asm.ea).apply(null,arguments)},PC=u._LeakyRelu=function(){return(PC=u._LeakyRelu=u.asm.fa).apply(null,arguments)},OC=u._Less=function(){return(OC=u._Less=u.asm.ga).apply(null,arguments)},LC=u._LessEqual=function(){return(LC=u._LessEqual=u.asm.ha).apply(null,arguments)},zC=u._Log=function(){return(zC=u._Log=u.asm.ia).apply(null,arguments)},BC=u._LogicalAnd=function(){return(BC=u._LogicalAnd=u.asm.ja).apply(null,arguments)},WC=u._Max=function(){return(WC=u._Max=u.asm.ka).apply(null,arguments)},VC=u._MaxPool=function(){return(VC=u._MaxPool=u.asm.la).apply(null,arguments)},UC=u._Maximum=function(){return(UC=u._Maximum=u.asm.ma).apply(null,arguments)},GC=u._Mean=function(){return(GC=u._Mean=u.asm.na).apply(null,arguments)},HC=u._Min=function(){return(HC=u._Min=u.asm.oa).apply(null,arguments)},jC=u._Minimum=function(){return(jC=u._Minimum=u.asm.pa).apply(null,arguments)},qC=u._Multiply=function(){return(qC=u._Multiply=u.asm.qa).apply(null,arguments)},KC=u._Neg=function(){return(KC=u._Neg=u.asm.ra).apply(null,arguments)},XC=u._NonMaxSuppressionV3=function(){return(XC=u._NonMaxSuppressionV3=u.asm.sa).apply(null,arguments)},YC=u._NonMaxSuppressionV4=function(){return(YC=u._NonMaxSuppressionV4=u.asm.ta).apply(null,arguments)},ZC=u._NonMaxSuppressionV5=function(){return(ZC=u._NonMaxSuppressionV5=u.asm.ua).apply(null,arguments)},JC=u._NotEqual=function(){return(JC=u._NotEqual=u.asm.va).apply(null,arguments)},QC=u._OneHot=function(){return(QC=u._OneHot=u.asm.wa).apply(null,arguments)},e_=u._PadV2=function(){return(e_=u._PadV2=u.asm.xa).apply(null,arguments)},t_=u._Pow=function(){return(t_=u._Pow=u.asm.ya).apply(null,arguments)},n_=u._Prelu=function(){return(n_=u._Prelu=u.asm.za).apply(null,arguments)},a_=u._Prod=function(){return(a_=u._Prod=u.asm.Aa).apply(null,arguments)},r_=u._RealDiv=function(){return(r_=u._RealDiv=u.asm.Ba).apply(null,arguments)},s_=u._Relu=function(){return(s_=u._Relu=u.asm.Ca).apply(null,arguments)},i_=u._Relu6=function(){return(i_=u._Relu6=u.asm.Da).apply(null,arguments)},o_=u._ResizeBilinear=function(){return(o_=u._ResizeBilinear=u.asm.Ea).apply(null,arguments)},l_=u._Reverse=function(){return(l_=u._Reverse=u.asm.Fa).apply(null,arguments)},u_=u._RotateWithOffset=function(){return(u_=u._RotateWithOffset=u.asm.Ga).apply(null,arguments)},c_=u._Round=function(){return(c_=u._Round=u.asm.Ha).apply(null,arguments)},p_=u._Rsqrt=function(){return(p_=u._Rsqrt=u.asm.Ia).apply(null,arguments)},d_=u._ScatterNd=function(){return(d_=u._ScatterNd=u.asm.Ja).apply(null,arguments)},h_=u._SelectV2=function(){return(h_=u._SelectV2=u.asm.Ka).apply(null,arguments)},m_=u._Sigmoid=function(){return(m_=u._Sigmoid=u.asm.La).apply(null,arguments)},f_=u._Sin=function(){return(f_=u._Sin=u.asm.Ma).apply(null,arguments)},g_=u._Softmax=function(){return(g_=u._Softmax=u.asm.Na).apply(null,arguments)},y_=u._Sqrt=function(){return(y_=u._Sqrt=u.asm.Oa).apply(null,arguments)},b_=u._Square=function(){return(b_=u._Square=u.asm.Pa).apply(null,arguments)},x_=u._SquaredDifference=function(){return(x_=u._SquaredDifference=u.asm.Qa).apply(null,arguments)},v_=u._Step=function(){return(v_=u._Step=u.asm.Ra).apply(null,arguments)},w_=u._StridedSlice=function(){return(w_=u._StridedSlice=u.asm.Sa).apply(null,arguments)},k_=u._Sub=function(){return(k_=u._Sub=u.asm.Ta).apply(null,arguments)},I_=u._Sum=function(){return(I_=u._Sum=u.asm.Ua).apply(null,arguments)},N_=u._Tanh=function(){return(N_=u._Tanh=u.asm.Va).apply(null,arguments)},T_=u._Tile=function(){return(T_=u._Tile=u.asm.Wa).apply(null,arguments)},S_=u._TopK=function(){return(S_=u._TopK=u.asm.Xa).apply(null,arguments)},C_=u._Transpose=function(){return(C_=u._Transpose=u.asm.Ya).apply(null,arguments)},__=u.__FusedMatMul=function(){return(__=u.__FusedMatMul=u.asm.Za).apply(null,arguments)},Ju=u._malloc=function(){return(Ju=u._malloc=u.asm._a).apply(null,arguments)},Qu=u._free=function(){return(Qu=u._free=u.asm.$a).apply(null,arguments)},E_=u.___em_js__initPthreadsJS=function(){return(E_=u.___em_js__initPthreadsJS=u.asm.ab).apply(null,arguments)},$w=u.___errno_location=function(){return($w=u.___errno_location=u.asm.bb).apply(null,arguments)},Dw=u._emscripten_get_global_libc=function(){return(Dw=u._emscripten_get_global_libc=u.asm.cb).apply(null,arguments)},Rw=u._memalign=function(){return(Rw=u._memalign=u.asm.db).apply(null,arguments)},Mw=u.___pthread_tsd_run_dtors=function(){return(Mw=u.___pthread_tsd_run_dtors=u.asm.eb).apply(null,arguments)},Mg=u._emscripten_main_thread_process_queued_calls=function(){return(Mg=u._emscripten_main_thread_process_queued_calls=u.asm.fb).apply(null,arguments)},A_=u._emscripten_current_thread_process_queued_calls=function(){return(A_=u._emscripten_current_thread_process_queued_calls=u.asm.gb).apply(null,arguments)},Pw=u._emscripten_register_main_browser_thread_id=function(){return(Pw=u._emscripten_register_main_browser_thread_id=u.asm.hb).apply(null,arguments)},F_=u._emscripten_main_browser_thread_id=function(){return(F_=u._emscripten_main_browser_thread_id=u.asm.ib).apply(null,arguments)},$_=u._emscripten_async_run_in_main_thread=function(){return($_=u._emscripten_async_run_in_main_thread=u.asm.jb).apply(null,arguments)},D_=u._emscripten_sync_run_in_main_thread=function(){return(D_=u._emscripten_sync_run_in_main_thread=u.asm.kb).apply(null,arguments)},R_=u._emscripten_sync_run_in_main_thread_0=function(){return(R_=u._emscripten_sync_run_in_main_thread_0=u.asm.lb).apply(null,arguments)},M_=u._emscripten_sync_run_in_main_thread_1=function(){return(M_=u._emscripten_sync_run_in_main_thread_1=u.asm.mb).apply(null,arguments)},P_=u._emscripten_sync_run_in_main_thread_2=function(){return(P_=u._emscripten_sync_run_in_main_thread_2=u.asm.nb).apply(null,arguments)},O_=u._emscripten_sync_run_in_main_thread_xprintf_varargs=function(){return(O_=u._emscripten_sync_run_in_main_thread_xprintf_varargs=u.asm.ob).apply(null,arguments)},L_=u._emscripten_sync_run_in_main_thread_3=function(){return(L_=u._emscripten_sync_run_in_main_thread_3=u.asm.pb).apply(null,arguments)},Ow=u._emscripten_sync_run_in_main_thread_4=function(){return(Ow=u._emscripten_sync_run_in_main_thread_4=u.asm.qb).apply(null,arguments)},z_=u._emscripten_sync_run_in_main_thread_5=function(){return(z_=u._emscripten_sync_run_in_main_thread_5=u.asm.rb).apply(null,arguments)},B_=u._emscripten_sync_run_in_main_thread_6=function(){return(B_=u._emscripten_sync_run_in_main_thread_6=u.asm.sb).apply(null,arguments)},W_=u._emscripten_sync_run_in_main_thread_7=function(){return(W_=u._emscripten_sync_run_in_main_thread_7=u.asm.tb).apply(null,arguments)},Lw=u._emscripten_run_in_main_runtime_thread_js=function(){return(Lw=u._emscripten_run_in_main_runtime_thread_js=u.asm.ub).apply(null,arguments)},Pg=u._emscripten_async_queue_on_thread_=function(){return(Pg=u._emscripten_async_queue_on_thread_=u.asm.vb).apply(null,arguments)},V_=u._emscripten_tls_init=function(){return(V_=u._emscripten_tls_init=u.asm.wb).apply(null,arguments)},ec=u.stackSave=function(){return(ec=u.stackSave=u.asm.xb).apply(null,arguments)},xo=u.stackAlloc=function(){return(xo=u.stackAlloc=u.asm.yb).apply(null,arguments)},vo=u.stackRestore=function(){return(vo=u.stackRestore=u.asm.zb).apply(null,arguments)},zw=u.dynCall_vi=function(){return(zw=u.dynCall_vi=u.asm.Ab).apply(null,arguments)},U_=u.dynCall_v=function(){return(U_=u.dynCall_v=u.asm.Bb).apply(null,arguments)},G_=u.dynCall_ii=function(){return(G_=u.dynCall_ii=u.asm.Cb).apply(null,arguments)};u.asm=ad,u.cwrap=Re,u.PThread=me,u.PThread=me,u._pthread_self=Ba,u.wasmMemory=Q,u.ExitStatus=Bw;var tc;u.then=function(w){if(tc)w(u);else{var S=u.onRuntimeInitialized;u.onRuntimeInitialized=function(){S&&S(),w(u)}}return u};function Bw(w){this.name="ExitStatus",this.message="Program terminated with exit("+w+")",this.status=w}xs=function w(){tc||Og(),tc||(xs=w)};function Og(w){if(w=w||h,ur>0||(sa(),ur>0))return;function S(){tc||(tc=!0,u.calledRun=!0,!ge&&(Up(),$f(),u.onRuntimeInitialized&&u.onRuntimeInitialized(),Df()))}u.setStatus?(u.setStatus("Running..."),setTimeout(function(){setTimeout(function(){u.setStatus("")},1),S()},1)):S()}if(u.run=Og,u.preInit)for(typeof u.preInit=="function"&&(u.preInit=[u.preInit]);u.preInit.length>0;)u.preInit.pop()();return x||(se=!0),x||Og(),r}}();typeof e=="object"&&typeof t=="object"?t.exports=n:typeof define=="function"&&define.amd?define([],function(){return n}):typeof e=="object"&&(e.WasmBackendModuleThreadedSimd=n)}),bE=un((e,t)=>{var n=function(){var a=typeof document!="undefined"&&document.currentScript?document.currentScript.src:void 0;return typeof __filename!="undefined"&&(a=a||__filename),function(r){r=r||{};var s=typeof r!="undefined"?r:{},i={},o;for(o in s)s.hasOwnProperty(o)&&(i[o]=s[o]);var l=[],c="./this.program",u=function(V,te){throw te},p=!1,d=!1,h=!1,m=!1;p=typeof window=="object",d=typeof importScripts=="function",h=typeof process=="object"&&typeof process.versions=="object"&&typeof process.versions.node=="string",m=!p&&!h&&!d;var f="";function g(V){return s.locateFile?s.locateFile(V,f):f+V}var y,b,v,x,N,T;h?(d?f=rc().dirname(f)+"/":f=__dirname+"/",y=function(V,te){return N||(N=require("fs")),T||(T=rc()),V=T.normalize(V),N.readFileSync(V,te?null:"utf8")},v=function(V){var te=y(V,!0);return te.buffer||(te=new Uint8Array(te)),H(te.buffer),te},process.argv.length>1&&(c=process.argv[1].replace(/\\/g,"/")),l=process.argv.slice(2),process.on("uncaughtException",function(V){if(!(V instanceof Xu))throw V}),process.on("unhandledRejection",ys),u=function(V){process.exit(V)},s.inspect=function(){return"[Emscripten Module object]"}):m?(typeof read!="undefined"&&(y=function(V){return read(V)}),v=function(V){var te;return typeof readbuffer=="function"?new Uint8Array(readbuffer(V)):(te=read(V,"binary"),H(typeof te=="object"),te)},typeof scriptArgs!="undefined"?l=scriptArgs:typeof arguments!="undefined"&&(l=arguments),typeof quit=="function"&&(u=function(V){quit(V)}),typeof print!="undefined"&&(typeof console=="undefined"&&(console={}),console.log=print,console.warn=console.error=typeof printErr!="undefined"?printErr:print)):(p||d)&&(d?f=self.location.href:document.currentScript&&(f=document.currentScript.src),a&&(f=a),f.indexOf("blob:")!==0?f=f.substr(0,f.lastIndexOf("/")+1):f="",y=function(V){var te=new XMLHttpRequest;return te.open("GET",V,!1),te.send(null),te.responseText},d&&(v=function(V){var te=new XMLHttpRequest;return te.open("GET",V,!1),te.responseType="arraybuffer",te.send(null),new Uint8Array(te.response)}),b=function(V,te,Ie){var _e=new XMLHttpRequest;_e.open("GET",V,!0),_e.responseType="arraybuffer",_e.onload=function(){if(_e.status==200||_e.status==0&&_e.response){te(_e.response);return}Ie()},_e.onerror=Ie,_e.send(null)},x=function(V){document.title=V});var C=s.print||console.log.bind(console),$=s.printErr||console.warn.bind(console);for(o in i)i.hasOwnProperty(o)&&(s[o]=i[o]);i=null,s.arguments&&(l=s.arguments),s.thisProgram&&(c=s.thisProgram),s.quit&&(u=s.quit);var F;s.wasmBinary&&(F=s.wasmBinary);var O;s.noExitRuntime&&(O=s.noExitRuntime),typeof WebAssembly!="object"&&$("no native wasm support detected");var B,W=new WebAssembly.Table({initial:151,maximum:151+0,element:"anyfunc"}),G=!1,q=0;function H(V,te){V||ys("Assertion failed: "+te)}function K(V){var te=s["_"+V];return H(te,"Cannot call unknown function "+V+", make sure it is exported"),te}function X(V,te,Ie,_e,at){var _t={string:function(Hn){var Rr=0;if(Hn!=null&&Hn!==0){var ld=(Hn.length<<2)+1;Rr=Ku(ld),ae(Hn,Rr,ld)}return Rr},array:function(Hn){var Rr=Ku(Hn.length);return pe(Hn,Rr),Rr}};function Xe(Hn){return te==="string"?Q(Hn):te==="boolean"?Boolean(Hn):Hn}var je=K(V),Ut=[],pr=0;if(_e)for(var dr=0;dr<_e.length;dr++){var od=_t[Ie[dr]];od?(pr===0&&(pr=qu()),Ut[dr]=od(_e[dr])):Ut[dr]=_e[dr]}var Zu=je.apply(null,Ut);return Zu=Xe(Zu),pr!==0&&rd(pr),Zu}function ne(V,te,Ie,_e){Ie=Ie||[];var at=Ie.every(function(Xe){return Xe==="number"}),_t=te!=="string";return _t&&at&&!_e?K(V):function(){return X(V,te,Ie,arguments,_e)}}var J=typeof TextDecoder!="undefined"?new TextDecoder("utf8"):void 0;function se(V,te,Ie){for(var _e=te+Ie,at=te;V[at]&&!(at>=_e);)++at;if(at-te>16&&V.subarray&&J)return J.decode(V.subarray(te,at));for(var _t="";te<at;){var Xe=V[te++];if(!(Xe&128)){_t+=String.fromCharCode(Xe);continue}var je=V[te++]&63;if((Xe&224)==192){_t+=String.fromCharCode((Xe&31)<<6|je);continue}var Ut=V[te++]&63;if((Xe&240)==224?Xe=(Xe&15)<<12|je<<6|Ut:Xe=(Xe&7)<<18|je<<12|Ut<<6|V[te++]&63,Xe<65536)_t+=String.fromCharCode(Xe);else{var pr=Xe-65536;_t+=String.fromCharCode(55296|pr>>10,56320|pr&1023)}}return _t}function Q(V,te){return V?se(ye,V,te):""}function oe(V,te,Ie,_e){if(!(_e>0))return 0;for(var at=Ie,_t=Ie+_e-1,Xe=0;Xe<V.length;++Xe){var je=V.charCodeAt(Xe);if(je>=55296&&je<=57343){var Ut=V.charCodeAt(++Xe);je=65536+((je&1023)<<10)|Ut&1023}if(je<=127){if(Ie>=_t)break;te[Ie++]=je}else if(je<=2047){if(Ie+1>=_t)break;te[Ie++]=192|je>>6,te[Ie++]=128|je&63}else if(je<=65535){if(Ie+2>=_t)break;te[Ie++]=224|je>>12,te[Ie++]=128|je>>6&63,te[Ie++]=128|je&63}else{if(Ie+3>=_t)break;te[Ie++]=240|je>>18,te[Ie++]=128|je>>12&63,te[Ie++]=128|je>>6&63,te[Ie++]=128|je&63}}return te[Ie]=0,Ie-at}function ae(V,te,Ie){return oe(V,ye,te,Ie)}function pe(V,te){ge.set(V,te)}var ue,ge,ye,ke,Se,Ce,Re,Ke,Ve;function rt(V){ue=V,s.HEAP8=ge=new Int8Array(V),s.HEAP16=ke=new Int16Array(V),s.HEAP32=Ce=new Int32Array(V),s.HEAPU8=ye=new Uint8Array(V),s.HEAPU16=Se=new Uint16Array(V),s.HEAPU32=Re=new Uint32Array(V),s.HEAPF32=Ke=new Float32Array(V),s.HEAPF64=Ve=new Float64Array(V)}var ot=s.INITIAL_MEMORY||16777216;function He(V){for(;V.length>0;){var te=V.shift();if(typeof te=="function"){te(s);continue}var Ie=te.func;typeof Ie=="number"?te.arg===void 0?s.dynCall_v(Ie):s.dynCall_vi(Ie,te.arg):Ie(te.arg===void 0?null:te.arg)}}var ct=[],pt=[],Gn=[],Je=[],_n=!1,Qt=!1;function En(){if(s.preRun)for(typeof s.preRun=="function"&&(s.preRun=[s.preRun]);s.preRun.length;)za(s.preRun.shift());He(ct)}function na(){_n=!0,He(pt)}function xn(){He(Gn)}function ln(){Qt=!0}function aa(){if(s.postRun)for(typeof s.postRun=="function"&&(s.postRun=[s.postRun]);s.postRun.length;)An(s.postRun.shift());He(Je)}function za(V){ct.unshift(V)}function An(V){Je.unshift(V)}var uo=Math.ceil,zu=Math.floor,wa=0,ra=null,ka=null;function co(V){wa++,s.monitorRunDependencies&&s.monitorRunDependencies(wa)}function po(V){if(wa--,s.monitorRunDependencies&&s.monitorRunDependencies(wa),wa==0&&(ra!==null&&(clearInterval(ra),ra=null),ka)){var te=ka;ka=null,te()}}s.preloadedImages={},s.preloadedAudios={};function ys(V){throw s.onAbort&&s.onAbort(V),V+="",C(V),$(V),G=!0,q=1,V="abort("+V+"). Build with -s ASSERTIONS=1 for more info.",new WebAssembly.RuntimeError(V)}function Bu(V,te){return String.prototype.startsWith?V.startsWith(te):V.indexOf(te)===0}var Ff="data:application/octet-stream;base64,";function Wu(V){return Bu(V,Ff)}var Vp="file://";function Vu(V){return Bu(V,Vp)}var sa="tfjs-backend-wasm.wasm";Wu(sa)||(sa=g(sa));function Up(){try{if(F)return new Uint8Array(F);if(v)return v(sa);throw"both async and sync fetching of the wasm failed"}catch(V){ys(V)}}function $f(){return!F&&(p||d)&&typeof fetch=="function"&&!Vu(sa)?fetch(sa,{credentials:"same-origin"}).then(function(V){if(!V.ok)throw"failed to load wasm binary file at '"+sa+"'";return V.arrayBuffer()}).catch(function(){return Up()}):new Promise(function(V,te){V(Up())})}function Df(){var V={env:cr,wasi_snapshot_preview1:cr};function te(Xe,je){var Ut=Xe.exports;s.asm=Ut,B=Ut.memory,rt(B.buffer),po("wasm-instantiate")}co("wasm-instantiate");function Ie(Xe){te(Xe.instance)}function _e(Xe){return $f().then(function(je){return WebAssembly.instantiate(je,V)}).then(Xe,function(je){$("failed to asynchronously prepare wasm: "+je),ys(je)})}function at(){if(!F&&typeof WebAssembly.instantiateStreaming=="function"&&!Wu(sa)&&!Vu(sa)&&typeof fetch=="function")fetch(sa,{credentials:"same-origin"}).then(function(Xe){var je=WebAssembly.instantiateStreaming(Xe,V);return je.then(Ie,function(Ut){$("wasm streaming compile failed: "+Ut),$("falling back to ArrayBuffer instantiation"),_e(Ie)})});else return _e(Ie)}if(s.instantiateWasm)try{var _t=s.instantiateWasm(V,te);return _t}catch(Xe){return $("Module.instantiateWasm callback failed with error: "+Xe),!1}return at(),{}}pt.push();function Rf(V){rt(B.buffer)}var bs={splitPath:function(V){var te=/^(\/?|)([\s\S]*?)((?:\.{1,2}|[^\/]+?|)(\.[^.\/]*|))(?:[\/]*)$/;return te.exec(V).slice(1)},normalizeArray:function(V,te){for(var Ie=0,_e=V.length-1;_e>=0;_e--){var at=V[_e];at==="."?V.splice(_e,1):at===".."?(V.splice(_e,1),Ie++):Ie&&(V.splice(_e,1),Ie--)}if(te)for(;Ie;Ie--)V.unshift("..");return V},normalize:function(V){var te=V.charAt(0)==="/",Ie=V.substr(-1)==="/";return V=bs.normalizeArray(V.split("/").filter(function(_e){return!!_e}),!te).join("/"),!V&&!te&&(V="."),V&&Ie&&(V+="/"),(te?"/":"")+V},dirname:function(V){var te=bs.splitPath(V),Ie=te[0],_e=te[1];return!Ie&&!_e?".":(_e&&(_e=_e.substr(0,_e.length-1)),Ie+_e)},basename:function(V){if(V==="/")return"/";var te=V.lastIndexOf("/");return te===-1?V:V.substr(te+1)},extname:function(V){return bs.splitPath(V)[3]},join:function(){var V=Array.prototype.slice.call(arguments,0);return bs.normalize(V.join("/"))},join2:function(V,te){return bs.normalize(V+"/"+te)}},ho={mappings:{},buffers:[null,[],[]],printChar:function(V,te){var Ie=ho.buffers[V];te===0||te===10?((V===1?C:$)(se(Ie,0)),Ie.length=0):Ie.push(te)},varargs:void 0,get:function(){ho.varargs+=4;var V=Ce[ho.varargs-4>>2];return V},getStr:function(V){var te=Q(V);return te},get64:function(V,te){return V}};function Mf(V){return 0}function ur(V,te,Ie,_e,at){}function Uu(V,te,Ie,_e){for(var at=0,_t=0;_t<Ie;_t++){for(var Xe=Ce[te+_t*8>>2],je=Ce[te+(_t*8+4)>>2],Ut=0;Ut<je;Ut++)ho.printChar(V,ye[Xe+Ut]);at+=je}return Ce[_e>>2]=at,0}function xs(V){sd(V)}function Pf(V){xs(V)}function Of(V){return V=+V,V>=0?+zu(V+.5):+uo(V-.5)}var cr={emscripten_notify_memory_growth:Rf,fd_close:Mf,fd_seek:ur,fd_write:Uu,proc_exit:Pf,roundf:Of},Gu=Df();s.asm=Gu;var Lf=s._init=function(){return(Lf=s._init=s.asm.init).apply(null,arguments)},Gp=s._register_tensor=function(){return(Gp=s._register_tensor=s.asm.register_tensor).apply(null,arguments)},zf=s._dispose_data=function(){return(zf=s._dispose_data=s.asm.dispose_data).apply(null,arguments)},Hp=s._dispose=function(){return(Hp=s._dispose=s.asm.dispose).apply(null,arguments)},ia=s._Abs=function(){return(ia=s._Abs=s.asm.Abs).apply(null,arguments)},jp=s._Add=function(){return(jp=s._Add=s.asm.Add).apply(null,arguments)},Bf=s._AddN=function(){return(Bf=s._AddN=s.asm.AddN).apply(null,arguments)},Wf=s._ArgMax=function(){return(Wf=s._ArgMax=s.asm.ArgMax).apply(null,arguments)},Vf=s._AvgPool=function(){return(Vf=s._AvgPool=s.asm.AvgPool).apply(null,arguments)},Uf=s._BatchMatMul=function(){return(Uf=s._BatchMatMul=s.asm.BatchMatMul).apply(null,arguments)},qp=s._ClipByValue=function(){return(qp=s._ClipByValue=s.asm.ClipByValue).apply(null,arguments)},Kp=s._Conv2D=function(){return(Kp=s._Conv2D=s.asm.Conv2D).apply(null,arguments)},Xp=s._Conv2DBackpropInput=function(){return(Xp=s._Conv2DBackpropInput=s.asm.Conv2DBackpropInput).apply(null,arguments)},mo=s._Cos=function(){return(mo=s._Cos=s.asm.Cos).apply(null,arguments)},Hu=s._CropAndResize=function(){return(Hu=s._CropAndResize=s.asm.CropAndResize).apply(null,arguments)},fo=s._Cumsum=function(){return(fo=s._Cumsum=s.asm.Cumsum).apply(null,arguments)},go=s._DepthToSpace=function(){return(go=s._DepthToSpace=s.asm.DepthToSpace).apply(null,arguments)},Gf=s._DepthwiseConv2dNative=function(){return(Gf=s._DepthwiseConv2dNative=s.asm.DepthwiseConv2dNative).apply(null,arguments)},Hf=s._Equal=function(){return(Hf=s._Equal=s.asm.Equal).apply(null,arguments)},jf=s._Exp=function(){return(jf=s._Exp=s.asm.Exp).apply(null,arguments)},me=s._FlipLeftRight=function(){return(me=s._FlipLeftRight=s.asm.FlipLeftRight).apply(null,arguments)},qf=s._Floor=function(){return(qf=s._Floor=s.asm.Floor).apply(null,arguments)},Kf=s._FloorDiv=function(){return(Kf=s._FloorDiv=s.asm.FloorDiv).apply(null,arguments)},Xf=s._FusedBatchNorm=function(){return(Xf=s._FusedBatchNorm=s.asm.FusedBatchNorm).apply(null,arguments)},Yf=s._FusedConv2D=function(){return(Yf=s._FusedConv2D=s.asm.FusedConv2D).apply(null,arguments)},vs=s._FusedDepthwiseConv2D=function(){return(vs=s._FusedDepthwiseConv2D=s.asm.FusedDepthwiseConv2D).apply(null,arguments)},Zf=s._Gather=function(){return(Zf=s._Gather=s.asm.Gather).apply(null,arguments)},Jf=s._GatherNd=function(){return(Jf=s._GatherNd=s.asm.GatherNd).apply(null,arguments)},Qf=s._Greater=function(){return(Qf=s._Greater=s.asm.Greater).apply(null,arguments)},eg=s._GreaterEqual=function(){return(eg=s._GreaterEqual=s.asm.GreaterEqual).apply(null,arguments)},tg=s._LeakyRelu=function(){return(tg=s._LeakyRelu=s.asm.LeakyRelu).apply(null,arguments)},ng=s._Less=function(){return(ng=s._Less=s.asm.Less).apply(null,arguments)},ag=s._LessEqual=function(){return(ag=s._LessEqual=s.asm.LessEqual).apply(null,arguments)},rg=s._Log=function(){return(rg=s._Log=s.asm.Log).apply(null,arguments)},sg=s._LogicalAnd=function(){return(sg=s._LogicalAnd=s.asm.LogicalAnd).apply(null,arguments)},ig=s._Max=function(){return(ig=s._Max=s.asm.Max).apply(null,arguments)},$r=s._MaxPool=function(){return($r=s._MaxPool=s.asm.MaxPool).apply(null,arguments)},ws=s._Maximum=function(){return(ws=s._Maximum=s.asm.Maximum).apply(null,arguments)},yo=s._Mean=function(){return(yo=s._Mean=s.asm.Mean).apply(null,arguments)},og=s._Min=function(){return(og=s._Min=s.asm.Min).apply(null,arguments)},lg=s._Minimum=function(){return(lg=s._Minimum=s.asm.Minimum).apply(null,arguments)},ug=s._Multiply=function(){return(ug=s._Multiply=s.asm.Multiply).apply(null,arguments)},cg=s._Neg=function(){return(cg=s._Neg=s.asm.Neg).apply(null,arguments)},Pe=s._NonMaxSuppressionV3=function(){return(Pe=s._NonMaxSuppressionV3=s.asm.NonMaxSuppressionV3).apply(null,arguments)},pg=s._NonMaxSuppressionV4=function(){return(pg=s._NonMaxSuppressionV4=s.asm.NonMaxSuppressionV4).apply(null,arguments)},dg=s._NonMaxSuppressionV5=function(){return(dg=s._NonMaxSuppressionV5=s.asm.NonMaxSuppressionV5).apply(null,arguments)},hg=s._NotEqual=function(){return(hg=s._NotEqual=s.asm.NotEqual).apply(null,arguments)},mg=s._OneHot=function(){return(mg=s._OneHot=s.asm.OneHot).apply(null,arguments)},fg=s._PadV2=function(){return(fg=s._PadV2=s.asm.PadV2).apply(null,arguments)},gg=s._Pow=function(){return(gg=s._Pow=s.asm.Pow).apply(null,arguments)},ju=s._Prelu=function(){return(ju=s._Prelu=s.asm.Prelu).apply(null,arguments)},Yp=s._Prod=function(){return(Yp=s._Prod=s.asm.Prod).apply(null,arguments)},Zp=s._RealDiv=function(){return(Zp=s._RealDiv=s.asm.RealDiv).apply(null,arguments)},yg=s._Relu=function(){return(yg=s._Relu=s.asm.Relu).apply(null,arguments)},bg=s._Relu6=function(){return(bg=s._Relu6=s.asm.Relu6).apply(null,arguments)},xg=s._ResizeBilinear=function(){return(xg=s._ResizeBilinear=s.asm.ResizeBilinear).apply(null,arguments)},vg=s._Reverse=function(){return(vg=s._Reverse=s.asm.Reverse).apply(null,arguments)},wg=s._RotateWithOffset=function(){return(wg=s._RotateWithOffset=s.asm.RotateWithOffset).apply(null,arguments)},kg=s._Round=function(){return(kg=s._Round=s.asm.Round).apply(null,arguments)},Be=s._Rsqrt=function(){return(Be=s._Rsqrt=s.asm.Rsqrt).apply(null,arguments)},Ig=s._ScatterNd=function(){return(Ig=s._ScatterNd=s.asm.ScatterNd).apply(null,arguments)},Ng=s._SelectV2=function(){return(Ng=s._SelectV2=s.asm.SelectV2).apply(null,arguments)},Tg=s._Sigmoid=function(){return(Tg=s._Sigmoid=s.asm.Sigmoid).apply(null,arguments)},ks=s._Sin=function(){return(ks=s._Sin=s.asm.Sin).apply(null,arguments)},bo=s._Softmax=function(){return(bo=s._Softmax=s.asm.Softmax).apply(null,arguments)},Jp=s._Sqrt=function(){return(Jp=s._Sqrt=s.asm.Sqrt).apply(null,arguments)},Qp=s._Square=function(){return(Qp=s._Square=s.asm.Square).apply(null,arguments)},ed=s._SquaredDifference=function(){return(ed=s._SquaredDifference=s.asm.SquaredDifference).apply(null,arguments)},Sg=s._Step=function(){return(Sg=s._Step=s.asm.Step).apply(null,arguments)},Cg=s._StridedSlice=function(){return(Cg=s._StridedSlice=s.asm.StridedSlice).apply(null,arguments)},td=s._Sub=function(){return(td=s._Sub=s.asm.Sub).apply(null,arguments)},_g=s._Sum=function(){return(_g=s._Sum=s.asm.Sum).apply(null,arguments)},Ba=s._Tanh=function(){return(Ba=s._Tanh=s.asm.Tanh).apply(null,arguments)},Eg=s._Tile=function(){return(Eg=s._Tile=s.asm.Tile).apply(null,arguments)},Ag=s._TopK=function(){return(Ag=s._TopK=s.asm.TopK).apply(null,arguments)},nd=s._Transpose=function(){return(nd=s._Transpose=s.asm.Transpose).apply(null,arguments)},Dr=s.__FusedMatMul=function(){return(Dr=s.__FusedMatMul=s.asm._FusedMatMul).apply(null,arguments)},Fg=s._malloc=function(){return(Fg=s._malloc=s.asm.malloc).apply(null,arguments)},$g=s._free=function(){return($g=s._free=s.asm.free).apply(null,arguments)},ad=s.__start=function(){return(ad=s.__start=s.asm._start).apply(null,arguments)},qu=s.stackSave=function(){return(qu=s.stackSave=s.asm.stackSave).apply(null,arguments)},Ku=s.stackAlloc=function(){return(Ku=s.stackAlloc=s.asm.stackAlloc).apply(null,arguments)},rd=s.stackRestore=function(){return(rd=s.stackRestore=s.asm.stackRestore).apply(null,arguments)};s.asm=Gu,s.cwrap=ne;var Is;s.then=function(V){if(Is)V(s);else{var te=s.onRuntimeInitialized;s.onRuntimeInitialized=function(){te&&te(),V(s)}}return s};function Xu(V){this.name="ExitStatus",this.message="Program terminated with exit("+V+")",this.status=V}var Dg=!1;ka=function V(){Is||Yu(),Is||(ka=V)};function Rg(V){var te=s.__start;try{te();var Ie=0;sd(Ie,!0)}catch(at){if(at instanceof Xu)return;if(at=="unwind"){O=!0;return}else{var _e=at;at&&typeof at=="object"&&at.stack&&(_e=[at,at.stack]),$("exception thrown: "+_e),u(1,at)}}finally{Dg=!0}}function Yu(V){if(V=V||l,wa>0||(En(),wa>0))return;function te(){Is||(Is=!0,s.calledRun=!0,!G&&(na(),xn(),s.onRuntimeInitialized&&s.onRuntimeInitialized(),id&&Rg(V),aa()))}s.setStatus?(s.setStatus("Running..."),setTimeout(function(){setTimeout(function(){s.setStatus("")},1),te()},1)):te()}s.run=Yu;function sd(V,te){te&&O&&V===0||(O||(G=!0,q=V,ln(),s.onExit&&s.onExit(V)),u(V,new Xu(V)))}if(s.preInit)for(typeof s.preInit=="function"&&(s.preInit=[s.preInit]);s.preInit.length>0;)s.preInit.pop()();var id=!0;return s.noInitialRun&&(id=!1),O=!0,Yu(),r}}();typeof e=="object"&&typeof t=="object"?t.exports=n:typeof define=="function"&&define.amd?define([],function(){return n}):typeof e=="object"&&(e.WasmBackendModule=n)}),xE=1e-7,vE=1e-4,hd=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}},sc=class{decComplexRef(e){}time(e){return Wa("time")}read(e){return Wa("read")}readSync(e){return Wa("readSync")}numDataIds(){return Wa("numDataIds")}disposeData(e){return Wa("disposeData")}write(e,t,n){return Wa("write")}move(e,t,n,a){return Wa("move")}memory(){return Wa("memory")}floatPrecision(){return Wa("floatPrecision")}epsilon(){return this.floatPrecision()===32?xE:vE}dispose(){return Wa("dispose")}};function Wa(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 Gw(e){let t=e.length,n=0,a=0;for(;t>0;)a=Math.random()*t|0,t--,n=e[t],e[t]=e[a],e[a]=n}function wE(e,t){if(e.length!==t.length)throw 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,a,r,s=0;for(;n>0;)s=Math.random()*n|0,n--,a=e[n],r=t[n],e[n]=e[s],t[n]=t[s],e[s]=a,t[s]=r}function ic(e,t,n){return Math.max(e,Math.min(t,n))}function kE(e){return e%2==0?e:e+1}function IE(e){let t=0;for(let n=0;n<e.length;n++)t+=e[n];return t}function NE(e,t){let n=Math.random();return t*n+(1-n)*e}function TE(e,t){let n=0;for(let a=0;a<e.length;a++){let r=Number(e[a])-Number(t[a]);n+=r*r}return n}function A(e,t){if(!e)throw new Error(typeof t=="string"?t:t())}function sn(e,t,n=""){A(hr(e,t),()=>n+` Shapes ${e} and ${t} must match`)}function Ns(e){A(e!=null,()=>"The input to the tensor constructor must be a non-null value.")}function Ts(e,t=[],n=!1){if(t==null&&(t=[]),Array.isArray(e)||on(e)&&!n)for(let a=0;a<e.length;++a)Ts(e[a],t,n);else t.push(e);return t}function Pt(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 SE(e){return e.length===0}function hr(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 Gt(e){return e%1==0}function CE(e){if(Math.tanh!=null)return Math.tanh(e);if(e===Infinity)return 1;if(e===-Infinity)return-1;{let t=Math.exp(2*e);return(t-1)/(t+1)}}function _E(e){let t=Math.ceil(Math.sqrt(e));return[t,Math.ceil(e/t)]}function EE(e){let t=new Uint32Array(e);for(let n=0;n<e;++n)t[n]=n;return Gw(t),t}function oc(e,t){return t<=e.length?e:e+" ".repeat(t-e.length)}function AE(e,t=a=>0,n){return new Promise((a,r)=>{let s=0,i=()=>{if(e()){a();return}s++;let o=t(s);if(n!=null&&s>=n){r();return}setTimeout(i,o)};i()})}function FE(e,t){let n=1,a=-1;for(let s=0;s<e.length;++s)if(e[s]>=0)n*=e[s];else if(e[s]===-1){if(a!==-1)throw Error(`Shapes can only have 1 implicit size. Found -1 at dim ${a} and dim ${s}`);a=s}else if(e[s]<0)throw Error(`Shapes can not be < 0. Found ${e[s]} at dim ${s}`);if(a===-1){if(t>0&&t!==n)throw Error(`Size(${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} / ${n}`);let r=e.slice();return r[a]=t/n,r}function oa(e,t){let n=t.length;return e=e==null?t.map((a,r)=>r):[].concat(e),A(e.every(a=>a>=-n&&a<n),()=>`All values in axis param must be in range [-${n}, ${n}) but got axis ${e}`),A(e.every(a=>Gt(a)),()=>`All values in axis param must be integers but got axis ${e}`),e.map(a=>a<0?n+a:a)}function Hw(e,t){let n=[],a=[],r=t!=null&&Array.isArray(t)&&t.length===0,s=t==null||r?null:oa(t,e).sort(),i=0;for(let o=0;o<e.length;++o){if(s!=null){if(s[i]===o&&e[o]!==1)throw new Error(`Can't squeeze axis ${o} since its dim '${e[o]}' is not 1`);(s[i]==null||s[i]>o)&&e[o]===1&&(n.push(e[o]),a.push(o)),s[i]<=o&&i++}e[o]!==1&&(n.push(e[o]),a.push(o))}return{newShape:n,keptDims:a}}function jw(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 qw(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 Kw(e,t){for(let n=0;n<e.length;n++){let a=e[n];if(isNaN(a)||!isFinite(a))throw Error(`A tensor of type ${t} being uploaded contains ${a}.`)}}function Xw(e){return e==="bool"||e==="complex64"||e==="float32"||e==="int32"||e==="string"}function $E(e,t){return!(t==="complex64"||t==="float32"&&e!=="complex64"||t==="int32"&&e!=="float32"&&e!=="complex64"||t==="bool"&&e==="bool")}function on(e){return e instanceof Float32Array||e instanceof Int32Array||e instanceof Uint8Array}function Yw(e){if(e==="float32"||e==="int32")return 4;if(e==="complex64")return 8;if(e==="bool")return 1;throw new Error(`Unknown dtype ${e}`)}function Zw(e){if(e==null)return 0;let t=0;return e.forEach(n=>t+=n.length),t}function Pr(e){return typeof e=="string"||e instanceof String}function Jw(e){return typeof e=="boolean"}function Qw(e){return typeof e=="number"}function md(e){return Array.isArray(e)?md(e[0]):e instanceof Float32Array?"float32":e instanceof Int32Array||e instanceof Uint8Array?"int32":Qw(e)?"float32":Pr(e)?"string":Jw(e)?"bool":"float32"}function Or(e){return!!(e&&e.constructor&&e.call&&e.apply)}function fd(e,t){for(let n=t;n<e;++n)if(e%n==0)return n;return e}function No(e){let t=e.length;if(t<2)return[];let n=new Array(t-1);n[t-2]=e[t-1];for(let a=t-3;a>=0;--a)n[a]=n[a+1]*e[a+1];return n}function e0(e,t,n){let a=new Array;if(t.length===1){let r=t[0];for(let s=0;s<r;s++)a[s]=n[e+s]}else{let r=t[0],s=t.slice(1),i=s.reduce((o,l)=>o*l);for(let o=0;o<r;o++)a[o]=e0(e+o*i,s,n)}return a}function To(e,t){if(e.length===0)return t[0];let n=e.reduce((a,r)=>a*r);if(n===0)return[];if(n!==t.length)throw new Error(`[${e}] does not match the input size ${t.length}.`);return e0(0,e,t)}function Wg(e,t){let n=gd(e,t);for(let a=0;a<n.length;a++)n[a]=1;return n}function gd(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 DE(e,t){let n=e.reduce((a,r)=>a*r,1);if(t==null||t==="float32")return To(e,new Float32Array(n));if(t==="int32")return To(e,new Int32Array(n));if(t==="bool")return To(e,new Uint8Array(n));throw new Error(`Unknown data type ${t}`)}function Vg(e){e.forEach(t=>{A(Number.isInteger(t)&&t>=0,()=>`Tensor must have a shape comprised of positive integers but got shape [${e}].`)})}function RE(e,t,n){if(t===0)return 0;if(t===1)return e[0];let a=e[e.length-1];for(let r=0;r<e.length-1;++r)a+=n[r]*e[r];return a}function ME(e,t,n){if(t===0)return[];if(t===1)return[e];let a=new Array(t);for(let r=0;r<a.length-1;++r)a[r]=Math.floor(e/n[r]),e-=a[r]*n[r];return a[a.length-1]=e,a}function Ug(e){return e&&e.then&&typeof e.then=="function"}var t0="tfjsflags",n0=class{constructor(e){this.global=e,this.flags={},this.flagRegistry={},this.urlFlags={},this.populateURLFlags()}setPlatform(e,t){this.platform!=null&&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 a=this.urlFlags[e];console.warn(`Setting feature override from URL ${e}: ${a}.`),this.set(e,a)}}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(Ug(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=PE(this.global.location.search);t0 in e&&e[t0].split(",").forEach(t=>{let[n,a]=t.split(":");this.urlFlags[n]=OE(n,a)})}};function PE(e){let t={};return e.replace(/[?&]([^=?&]+)(?:=([^&]*))?/g,(n,...a)=>(LE(t,a[0],a[1]),a.join("="))),t}function LE(e,t,n){e[decodeURIComponent(t)]=decodeURIComponent(n||"")}function OE(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 ${e}.`)}function ee(){return Gg}var Gg=null;function zE(e){Gg=e}var Hg;function a0(){if(Hg==null){let e;if(typeof window!="undefined")e=window;else if(typeof global!="undefined")e=global;else if(typeof process!="undefined")e=process;else if(typeof self!="undefined")e=self;else throw new Error("Could not find a global object");Hg=e}return Hg}function BE(){let e=a0();return e._tfGlobals==null&&(e._tfGlobals=new Map),e._tfGlobals}function jg(e,t){let n=BE();if(n.has(e))return n.get(e);{let a=t();return n.set(e,a),n.get(e)}}var So="Abs",Co="Acos",_o="Acosh",Lr="Add",Ss="AddN",yd="All",bd="Any",Cs="ArgMax",lc="ArgMin",Eo="Asin",Ao="Asinh",Fo="Atan",$o="Atanh",Do="Atan2",_s="AvgPool",xd="AvgPoolGrad",uc="AvgPool3D",vd="AvgPool3DGrad",Es="BatchMatMul",cc="BatchToSpaceND",wd="Bincount",r0="BroadcastTo",As="Cast",Ro="Ceil",zr="ClipByValue",kd="Complex",pc="ComplexAbs",Mo="Concat",Fs="Conv2D",Id="Conv2DBackpropFilter",$s="Conv2DBackpropInput",dc="Conv3D",Nd="Conv3DBackpropFilterV2",Td="Conv3DBackpropInputV2",Ds="Cos",Po="Cosh",Rs="Cumsum",Oo="CropAndResize",Sd="DenseBincount",Lo="DepthToSpace",Ms="DepthwiseConv2dNative",Cd="DepthwiseConv2dNativeBackpropFilter",_d="DepthwiseConv2dNativeBackpropInput",Ed="Diag",hc="Dilation2D",Ad="Dilation2DBackpropInput",Fd="Dilation2DBackpropFilter",Ps="RealDiv",zo="Elu",$d="EluGrad",Bo="Erf",Wo="Equal",Os="Exp",Vo="ExpandDims",Uo="Expm1",Dd="FFT",mc="Fill",Go="FlipLeftRight",Ls="Floor",zs="FloorDiv",Bs="FusedBatchNorm",Ho="GatherV2",jo="GatherNd",qo="Greater",Ws="GreaterEqual",Ko="Identity",Rd="IFFT",Md="Imag",Xo="IsFinite",Yo="IsInf",Zo="IsNan",Vs="LeakyRelu",Jo="Less",Qo="LessEqual",Pd="LinSpace",Us="Log",el="Log1p",tl="LogicalAnd",fc="LogicalNot",gc="LogicalOr",s0="LogSoftmax",yc="LRN",Od="LRNGrad",Gs="Max",Hs="Maximum",js="MaxPool",Ld="MaxPoolGrad",bc="MaxPool3D",zd="MaxPool3DGrad",Bd="MaxPoolWithArgmax",qs="Mean",Ks="Min",Xs="Minimum",xc="MirrorPad",nl="Mod",Wd="Multinomial",Ys="Multiply",al="Neg",rl="NotEqual",sl="NonMaxSuppressionV3",il="NonMaxSuppressionV4",ol="NonMaxSuppressionV5",ll="OnesLike",Zs="OneHot",ul="Pack",Js="PadV2",WE="Pool",Qs="Pow",ei="Prelu",cl="Prod",vc="Range",Vd="Real",pl="Reciprocal",ti="Relu",dl="Reshape",wc="ResizeNearestNeighbor",Ud="ResizeNearestNeighborGrad",ni="ResizeBilinear",Gd="ResizeBilinearGrad",ai="Relu6",ri="Reverse",si="Round",ii="Rsqrt",hl="ScatterNd",ml="Select",fl="Selu",gl="Slice",oi="Sin",yl="Sinh",bl="Sign",li="Sigmoid",xl="Softplus",ui="Sqrt",ci="Sum",kc="SpaceToBatchND",vl="SplitV",pi="Softmax",di="SquaredDifference",Ic="Square",hi="Sub",Hd="SparseToDense",wl="StridedSlice",kl="Tan",mi="Tanh",Br="Tile",Il="TopK",fi="Transpose",jd="Unique",Nl="Unpack",Nc="UnsortedSegmentSum",Tl="ZerosLike",Wr="Step",qd="FromPixels",Sl="RotateWithOffset",gi="_FusedMatMul",yi="FusedConv2D",bi="FusedDepthwiseConv2D",Cl=jg("kernelRegistry",()=>new Map),Tc=jg("gradRegistry",()=>new Map);function Kd(e,t){let n=qg(e,t);return Cl.get(n)}function Kg(e){return Tc.get(e)}function Xd(e){let t=Cl.entries(),n=[];for(;;){let{done:a,value:r}=t.next();if(a)break;let[s,i]=r,[o]=s.split("_");o===e&&n.push(i)}return n}function Sc(e){let{kernelName:t,backendName:n}=e,a=qg(t,n);Cl.has(a)&&console.warn(`The kernel '${t}' for backend '${n}' is already registered`),Cl.set(a,e)}function i0(e){let{kernelName:t}=e;Tc.has(t)&&ee().getBool("DEBUG")&&console.warn(`Overriding the gradient for '${t}'`),Tc.set(t,e)}function VE(e,t){let n=qg(e,t);if(!Cl.has(n))throw new Error(`The kernel '${e}' for backend '${t}' is not registered`);Cl.delete(n)}function UE(e){if(!Tc.has(e))throw new Error(`The gradient '${e}' for backend is not registered`);Tc.delete(e)}function GE(e,t){Xd(e).forEach(n=>{let a=Object.assign({},n,{backendName:t});Sc(a)})}function qg(e,t){return`${t}_${e}`}var k={};Oe(k,{arraysEqual:()=>hr,assert:()=>A,assertNonNegativeIntegerDimensions:()=>Vg,assertNonNull:()=>Ns,assertShapesMatch:()=>sn,bytesFromStringArray:()=>Zw,bytesPerElement:()=>Yw,checkConversionForErrors:()=>Kw,clamp:()=>ic,computeStrides:()=>No,createScalarValue:()=>HE,createShuffledIndices:()=>EE,decodeString:()=>Zd,distSquared:()=>TE,encodeString:()=>Cc,fetch:()=>jE,flatten:()=>Ts,getArrayFromDType:()=>qw,getTypedArrayFromDType:()=>jw,hasEncodingLoss:()=>$E,indexToLoc:()=>ME,inferDtype:()=>md,inferFromImplicitShape:()=>FE,isBoolean:()=>Jw,isFunction:()=>Or,isInt:()=>Gt,isNumber:()=>Qw,isPromise:()=>Ug,isScalarShape:()=>SE,isString:()=>Pr,isTypedArray:()=>on,isValidDtype:()=>Xw,locToIndex:()=>RE,makeOnesTypedArray:()=>Wg,makeZerosNestedTypedArray:()=>DE,makeZerosTypedArray:()=>gd,nearestDivisor:()=>fd,nearestLargerEven:()=>kE,now:()=>Xg,parseAxisParam:()=>oa,randUniform:()=>NE,repeatedTry:()=>AE,rightPad:()=>oc,shuffle:()=>Gw,shuffleCombo:()=>wE,sizeFromShape:()=>Pt,sizeToSquarishShape:()=>_E,squeezeShape:()=>Hw,sum:()=>IE,tanh:()=>CE,toNestedArray:()=>To,toTypedArray:()=>Yd});function HE(e,t){return t==="string"?Cc(e):Yd([e],t)}function qE(e,t){return e instanceof Float32Array&&t==="float32"||e instanceof Int32Array&&t==="int32"||e instanceof Uint8Array&&t==="bool"}function Yd(e,t){if(t==="string")throw new Error("Cannot convert a string[] to a TypedArray");if(Array.isArray(e)&&(e=Ts(e)),ee().getBool("DEBUG")&&Kw(e,t),qE(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 a=0;a<n.length;++a)Math.round(e[a])!==0&&(n[a]=1);return n}else throw new Error(`Unknown data type ${t}`)}function Xg(){return ee().platform.now()}function jE(e,t){return ee().platform.fetch(e,t)}function Cc(e,t="utf-8"){return t=t||"utf-8",ee().platform.encode(e,t)}function Zd(e,t="utf-8"){return t=t||"utf-8",ee().platform.decode(e,t)}var YE=class{constructor(e,t){this.backendTimer=e,this.logger=t,t==null&&(this.logger=new XE)}profileKernel(e,t,n){let a,r=()=>{a=n()},s=this.backendTimer.time(r);if(ee().getBool("CHECK_COMPUTATION_FOR_ERRORS"))for(let i=0;i<a.length;i++){let o=a[i];o.data().then(l=>{KE(l,o.dtype,e)})}return{kernelName:e,outputs:a,inputs:t,timeMs:s.then(i=>i.kernelMs),extraInfo:s.then(i=>i.getExtraProfileInfo!=null?i.getExtraProfileInfo():"")}}logKernelProfile(e){let{kernelName:t,outputs:n,timeMs:a,inputs:r,extraInfo:s}=e;n.forEach(i=>{Promise.all([i.data(),a,s]).then(o=>{this.logger.logKernelProfile(t,i,o[0],o[1],r,o[2])})})}};function KE(e,t,n){if(t!=="float32")return!1;for(let a=0;a<e.length;a++){let r=e[a];if(isNaN(r)||!isFinite(r))return console.warn(`Found ${r} in the result of '${n}'`),!0}return!1}var XE=class{logKernelProfile(e,t,n,a,r,s){let i=typeof a=="number"?oc(`${a}ms`,9):a.error,o=oc(e,25),l=t.rank,c=t.size,u=oc(t.shape.toString(),14),p="";for(let d in r){let h=r[d];if(h!=null){let m=h.shape||t.shape,f=m.length;p+=`${d}: ${f}D ${f>0?m:""} `}}console.log(`%c${o} %c${i} %c${l}D ${u} %c${c} %c${p} %c${s}`,"font-weight:bold","color:red","color:blue","color: orange","color: green","color: steelblue")}};function ZE(e,t,n){let a={},r={};for(let l=0;l<t.length;l++)a[t[l].id]=!0;for(let l=0;l<e.length;l++){let c=e[l],u=c.inputs;for(let p in u){let d=u[p],h=!1;for(let m=0;m<t.length;m++)if(a[d.id]){c.outputs.forEach(f=>a[f.id]=!0),h=!0,r[c.id]=!0;break}if(h)break}}let s={};s[n.id]=!0;let i={};for(let l=e.length-1;l>=0;l--){let c=e[l],u=c.inputs;for(let p=0;p<c.outputs.length;p++)if(s[c.outputs[p].id]){for(let d in u)s[u[d].id]=!0,i[c.id]=!0;break}}let o=[];for(let l=0;l<e.length;l++){let c=e[l];if(r[c.id]&&i[c.id]){let u={};for(let d in c.inputs){let h=c.inputs[d];a[h.id]&&(u[d]=h)}let p=Object.assign({},c);p.inputs=u,p.outputs=c.outputs,o.push(p)}}return o}function JE(e,t,n,a){for(let r=t.length-1;r>=0;r--){let s=t[r],i=[];if(s.outputs.forEach(l=>{let c=e[l.id];c!=null?i.push(c):i.push(null)}),s.gradient==null)throw new Error(`Cannot compute gradient: gradient function not found for ${s.kernelName}.`);let o=s.gradient(i);for(let l in s.inputs){if(!(l in o))throw new Error(`Cannot backprop through input ${l}. Available gradients found: ${Object.keys(o)}.`);let c=n(()=>o[l]());if(c.dtype!=="float32")throw new Error(`Error in gradient for op ${s.kernelName}. The gradient of input ${l} must have 'float32' dtype, but has '${c.dtype}'`);let u=s.inputs[l];if(!hr(c.shape,u.shape))throw new Error(`Error in gradient for op ${s.kernelName}. The gradient of input '${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 p=e[u.id];e[u.id]=a(p,c),p.dispose()}}}}var o0=20,_c=3,Yg=7;function eA(e,t,n,a){let r=No(t),s=QE(e,t,n,r),i=t.length,o=Jd(e,t,n,r,s),l=["Tensor"];return a&&(l.push(` dtype: ${n}`),l.push(` rank: ${i}`),l.push(` shape: [${t}]`),l.push(" values:")),l.push(o.map(c=>" "+c).join(`
|
|
`)),l.join(`
|
|
`)}function QE(e,t,n,a){let r=Pt(t),s=a[a.length-1],i=new Array(s).fill(0),o=t.length,l=n==="complex64"?Ac(e):e;if(o>1)for(let c=0;c<r/s;c++){let u=c*s;for(let p=0;p<s;p++)i[p]=Math.max(i[p],Ec(l[u+p],0,n).length)}return i}function Ec(e,t,n){let a;return Array.isArray(e)?a=`${parseFloat(e[0].toFixed(Yg))} + ${parseFloat(e[1].toFixed(Yg))}j`:Pr(e)?a=`'${e}'`:n==="bool"?a=l0(e):a=parseFloat(e.toFixed(Yg)).toString(),oc(a,t)}function l0(e){return e===0?"false":"true"}function Jd(e,t,n,a,r,s=!0){let i=n==="complex64"?2:1,o=t[0],l=t.length;if(l===0){if(n==="complex64"){let f=Ac(e);return[Ec(f[0],0,n)]}return n==="bool"?[l0(e[0])]:[e[0].toString()]}if(l===1){if(o>o0){let g=_c*i,y=Array.from(e.slice(0,g)),b=Array.from(e.slice((o-_c)*i,o*i));return n==="complex64"&&(y=Ac(y),b=Ac(b)),["["+y.map((v,x)=>Ec(v,r[x],n)).join(", ")+", ..., "+b.map((v,x)=>Ec(v,r[o-_c+x],n)).join(", ")+"]"]}let f=n==="complex64"?Ac(e):Array.from(e);return["["+f.map((g,y)=>Ec(g,r[y],n)).join(", ")+"]"]}let c=t.slice(1),u=a.slice(1),p=a[0]*i,d=[];if(o>o0){for(let f=0;f<_c;f++){let g=f*p,y=g+p;d.push(...Jd(e.slice(g,y),c,n,u,r,!1))}d.push("...");for(let f=o-_c;f<o;f++){let g=f*p,y=g+p;d.push(...Jd(e.slice(g,y),c,n,u,r,f===o-1))}}else for(let f=0;f<o;f++){let g=f*p,y=g+p;d.push(...Jd(e.slice(g,y),c,n,u,r,f===o-1))}let h=l===2?",":"";d[0]="["+d[0]+h;for(let f=1;f<d.length-1;f++)d[f]=" "+d[f]+h;let m=`,
|
|
`;for(let f=2;f<l;f++)m+=`
|
|
`;return d[d.length-1]=" "+d[d.length-1]+"]"+(s?"":m),d}function Ac(e){let t=[];for(let n=0;n<e.length;n+=2)t.push([e[n],e[n+1]]);return t}var Ot=class{constructor(e,t,n){if(this.dtype=t,this.shape=e.slice(),this.size=Pt(e),n!=null){let a=n.length;A(a===this.size,()=>`Length of values '${a}' does not match the size inferred by the shape '${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||qw(t,this.size),this.strides=No(e)}set(e,...t){t.length===0&&(t=[0]),A(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 a of e){if(a<0||a>=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 a=0;a<e.length-1;++a)n+=this.strides[a]*e[a];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 Va().makeTensor(this.values,this.shape,this.dtype)}},Va=null,_l=null,tA=null;function nA(e){Va=e}function aA(e){_l=e}function rA(e){tA=e}var Ae=class{constructor(e,t,n,a){this.kept=!1,this.isDisposedInternal=!1,this.shape=e.slice(),this.dtype=t||"float32",this.size=Pt(e),this.strides=No(e),this.dataId=n,this.id=a,this.rankType=this.rank<5?this.rank.toString():"higher"}get rank(){return this.shape.length}async buffer(){let e=await this.data();return _l.buffer(this.shape,this.dtype,e)}bufferSync(){return _l.buffer(this.shape,this.dtype,this.dataSync())}async array(){let e=await this.data();return To(this.shape,e)}arraySync(){return To(this.shape,this.dataSync())}async data(){this.throwIfDisposed();let e=Va().read(this.dataId);if(this.dtype==="string"){let t=await e;try{return t.map(n=>Zd(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=Va().readSync(this.dataId);if(this.dtype==="string")try{return e.map(t=>Zd(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 Va().read(this.dataId);return this.dtype==="string"?e:new Uint8Array(e.buffer)}dispose(){this.isDisposed||(Va().disposeTensor(this),this.isDisposedInternal=!0)}get isDisposed(){return this.isDisposedInternal}throwIfDisposed(){if(this.isDisposed)throw new Error("Tensor is disposed.")}print(e=!1){return _l.print(this,e)}clone(){return this.throwIfDisposed(),_l.clone(this)}toString(e=!1){let t=this.dataSync();return eA(t,this.shape,this.dtype,e)}cast(e){return this.throwIfDisposed(),_l.cast(this,e)}variable(e=!0,t,n){return this.throwIfDisposed(),Va().makeVariable(this,e,t,n)}};Object.defineProperty(Ae,Symbol.hasInstance,{value:e=>!!e&&e.data!=null&&e.dataSync!=null&&e.throwIfDisposed!=null});function Y(){return jg("Tensor",()=>Ae)}Y();var Vr=class extends Ae{constructor(e,t,n,a){super(e.shape,e.dtype,e.dataId,a);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(!hr(e.shape,this.shape))throw new Error(`shape of the new value (${e.shape}) and previous value (${this.shape}) must match`);Va().disposeTensor(this),this.dataId=e.dataId,Va().incRef(this,null)}dispose(){Va().disposeVariable(this),this.isDisposedInternal=!0}};Object.defineProperty(Vr,Symbol.hasInstance,{value:e=>e instanceof Ae&&e.assign!=null&&e.assign instanceof Function});var Ia={};Oe(Ia,{assertTypesMatch:()=>u0,getTensorsInContainer:()=>Zg,isTensorInList:()=>sA,makeTypesMatch:()=>Nt});var Jg;(function(e){e.R0="R0",e.R1="R1",e.R2="R2",e.R3="R3",e.R4="R4",e.R5="R5",e.R6="R6"})(Jg||(Jg={}));var Qg;(function(e){e.float32="float32",e.int32="int32",e.bool="int32",e.complex64="complex64"})(Qg||(Qg={}));var ey;(function(e){e.float32="float32",e.int32="int32",e.bool="bool",e.complex64="complex64"})(ey||(ey={}));var ty;(function(e){e.float32="float32",e.int32="float32",e.bool="float32",e.complex64="complex64"})(ty||(ty={}));var ny;(function(e){e.float32="complex64",e.int32="complex64",e.bool="complex64",e.complex64="complex64"})(ny||(ny={}));var iA={float32:ty,int32:Qg,bool:ey,complex64:ny};function la(e,t){if(e==="string"||t==="string"){if(e==="string"&&t==="string")return"string";throw new Error(`Can not upcast ${e} with ${t}`)}return iA[e][t]}function Qd(e){return la(e,"int32")}function Nt(e,t){if(e.dtype===t.dtype)return[e,t];let n=la(e.dtype,t.dtype);return[e.cast(n),t.cast(n)]}function u0(e,t){A(e.dtype===t.dtype,()=>`The dtypes of the first(${e.dtype}) and second(${t.dtype}) input must match`)}function sA(e,t){return t.some(n=>n.id===e.id)}function Zg(e){let t=[],n=new Set;return c0(e,t,n),t}function c0(e,t,n){if(e==null)return;if(e instanceof Ae){t.push(e);return}if(!oA(e))return;let a=e;for(let r in a){let s=a[r];n.has(s)||(n.add(s),c0(s,t,n))}}function oA(e){return Array.isArray(e)||typeof e=="object"}function ay(e){return e.kernelName!=null}var p0=class{constructor(){this.registeredVariables={},this.nextTapeNodeId=0,this.numBytes=0,this.numTensors=0,this.numStringTensors=0,this.numDataBuffers=0,this.gradientDepth=0,this.kernelDepth=0,this.scopeStack=[],this.numDataMovesStack=[],this.nextScopeId=0,this.tensorInfo=new WeakMap,this.profiling=!1,this.activeProfile={newBytes:0,newTensors:0,peakBytes:0,kernels:[],result:null,get kernelNames(){return Array.from(new Set(this.kernels.map(e=>e.name)))}}}dispose(){for(let e in this.registeredVariables)this.registeredVariables[e].dispose()}},Fc=class{constructor(e){this.ENV=e,this.registry={},this.registryFactory={},this.pendingBackendInitId=0,this.state=new p0}async ready(){if(this.pendingBackendInit!=null)return this.pendingBackendInit.then(()=>{});if(this.backendInstance!=null)return;let e=this.getSortedBackends();for(let t=0;t<e.length;t++){let n=e[t];if(await this.initializeBackend(n).success){await this.setBackend(n);return}}throw new Error("Could not initialize any backends, all backend initializations failed.")}get backend(){if(this.pendingBackendInit!=null)throw new Error(`Backend '${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 '${e}' has not yet been initialized. Make sure to await tf.ready() or await tf.setBackend() before calling other methods`);this.setBackend(e)}return this.backendInstance}backendNames(){return Object.keys(this.registryFactory)}findBackend(e){if(!(e in this.registry))if(e in this.registryFactory){let{asyncInit:t}=this.initializeBackend(e);if(t)return null}else return null;return this.registry[e]}findBackendFactory(e){return e in this.registryFactory?this.registryFactory[e].factory:null}registerBackend(e,t,n=1){return e in this.registryFactory?(console.warn(`${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 YE(this.backendInstance),!0}setupRegisteredKernels(){Xd(this.backendName).forEach(e=>{e.setupFunc!=null&&e.setupFunc(this.backendInstance)})}disposeRegisteredKernels(e){Xd(e).forEach(t=>{t.disposeFunc!=null&&t.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 sc)&&typeof n.then=="function"){let a=++this.pendingBackendInitId,r=n.then(s=>a<this.pendingBackendInitId?!1:(this.registry[e]=s,this.pendingBackendInit=null,!0)).catch(s=>(a<this.pendingBackendInitId||(this.pendingBackendInit=null,console.warn(`Initialization of backend ${e} failed`),console.warn(s.stack||s.message)),!1));return this.pendingBackendInit=r,{success:r,asyncInit:!0}}else return this.registry[e]=n,{success:!0,asyncInit:!1}}catch(n){return console.warn(`Initialization of backend ${e} failed`),console.warn(n.stack||n.message),{success:!1,asyncInit:!1}}}removeBackend(e){if(!(e in this.registryFactory))throw new Error(`${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:a,asyncInit:r}=this.initializeBackend(n);if(r||a)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),a=n.backend,r=this.readSync(t);a.disposeData(t),n.backend=e,e.move(t,r,n.shape,n.dtype),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 a;return this.scopedRun(()=>this.startScope(n),()=>this.endScope(a),()=>(a=t(),a instanceof Promise&&console.error("Cannot return a Promise inside of tidy."),a))}scopedRun(e,t,n){e();try{let a=n();return t(),a}catch(a){throw t(),a}}nextTensorId(){return Fc.nextTensorId++}nextVariableId(){return Fc.nextVariableId++}clone(e){let t=this.makeTensorFromDataId(e.dataId,e.shape,e.dtype),n={x:e},a=s=>({x:()=>{let i="float32",o={x:s},l={dtype:i};return M.runKernel(As,o,l)}}),r=[];return this.addTapeNode(this.state.activeScope.name,n,[t],a,r,{}),t}runKernel(e,t,n){if(Kd(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 a=this.backend.numDataIds(),r=0;n.forEach(o=>{r+=o.dtype==="complex64"?3:1});let s=this.state.numDataMovesStack[this.state.numDataMovesStack.length-1],i=a-t-r-s;if(i>0)throw new Error(`Backend '${this.backendName}' has an internal memory leak (${i} data ids) after running '${e}'`)}runKernelFunc(e){let t,n=[],a=this.isTapeOn(),r=this.state.numBytes,s=this.state.numTensors;this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack.push(0);let i;this.backendName==null&&this.backend;let o,l=ay(e)?e.kernelName:this.state.activeScope!=null?this.state.activeScope.name:"";if(ay(e)){let{kernelName:h,inputs:m,attrs:f}=e;this.backendName==null&&this.backend;let g=Kd(h,this.backendName);A(g!=null,()=>`Cannot find registered kernel '${h}' for backend '${this.backendName}'`),i=()=>{let y=this.backend.numDataIds();o=g.kernelFunc({inputs:m,attrs:f,backend:this.backend});let b=Array.isArray(o)?o:[o];this.shouldCheckForMemLeaks()&&this.checkKernelForMemLeak(h,y,b);let v=b.map(x=>{if(x.rank!=null)return x;let{dataId:N,shape:T,dtype:C}=x;return this.makeTensorFromDataId(N,T,C)});if(a){let x=this.getTensorsForGradient(h,m,v);n=this.saveTensorsForBackwardMode(x)}return v}}else{let{forwardFunc:h}=e,m=f=>{!a||(n=f.map(g=>this.keep(this.clone(g))))};i=()=>{let f=this.backend.numDataIds();o=this.tidy(()=>h(this.backend,m));let g=Array.isArray(o)?o:[o];return this.shouldCheckForMemLeaks()&&this.checkKernelForMemLeak(l,f,g),g}}let{inputs:c,attrs:u}=e,p=ay(e)?null:e.backwardsFunc,d;return this.scopedRun(()=>this.state.kernelDepth++,()=>this.state.kernelDepth--,()=>{!this.ENV.getBool("DEBUG")&&!this.state.profiling?t=i():(d=this.profiler.profileKernel(l,c,()=>i()),this.ENV.getBool("DEBUG")&&this.profiler.logKernelProfile(d),t=d.outputs)}),a&&this.addTapeNode(l,c,t,p,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-s,totalTensorsSnapshot:this.state.numTensors,inputShapes:Object.keys(c).map(h=>c[h]!=null?c[h].shape:null),outputShapes:t.map(h=>h.shape),kernelTimeMs:d.timeMs,extraInfo:d.extraInfo}),Array.isArray(o)?t:t[0]}saveTensorsForBackwardMode(e){return e.map(t=>this.keep(this.clone(t)))}getTensorsForGradient(e,t,n){let a=Kg(e);if(a!=null){let r=a.inputsToSave||[],s=a.outputsToSave||[],i;a.saveAllInputs?(A(Array.isArray(t),()=>"saveAllInputs is true, expected inputs to be an array."),i=Object.keys(t).map(l=>t[l])):i=r.map(l=>t[l]);let o=n.filter((l,c)=>s[c]);return i.concat(o)}return[]}makeTensor(e,t,n,a){if(e==null)throw new Error("Values passed to engine.makeTensor() are null");n=n||"float32",a=a||this.backend;let r=e;n==="string"&&Pr(e[0])&&(r=e.map(o=>Cc(o)));let s=a.write(r,t,n),i=new Ae(t,n,s,this.nextTensorId());if(this.incRef(i,a),n==="string"){let o=this.state.tensorInfo.get(s),l=Zw(r);this.state.numBytes+=l-o.bytes,o.bytes=l}return i}makeTensorFromDataId(e,t,n,a){n=n||"float32";let r=new Ae(t,n,e,this.nextTensorId());return this.incRef(r,a),r}makeVariable(e,t=!0,n,a){n=n||this.nextVariableId().toString(),a!=null&&a!==e.dtype&&(e=e.cast(a));let r=new Vr(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}incRef(e,t){let n=this.state.tensorInfo.has(e.dataId)?this.state.tensorInfo.get(e.dataId).refCount:0;if(this.state.numTensors++,e.dtype==="string"&&this.state.numStringTensors++,n===0){this.state.numDataBuffers++;let a=0;e.dtype!=="complex64"&&e.dtype!=="string"&&(a=e.size*Yw(e.dtype)),this.state.tensorInfo.set(e.dataId,{backend:t||this.backend,dtype:e.dtype,shape:e.shape,bytes:a,refCount:0}),this.state.numBytes+=a}this.state.tensorInfo.get(e.dataId).refCount++,e instanceof Vr||this.track(e)}disposeTensor(e){if(!this.state.tensorInfo.has(e.dataId))return;this.state.numTensors--,e.dtype==="string"&&this.state.numStringTensors--;let t=this.state.tensorInfo.get(e.dataId);t.refCount<=1?(e.dtype!=="complex64"&&(this.state.numBytes-=t.bytes),this.state.numDataBuffers--,t.backend.disposeData(e.dataId),this.state.tensorInfo.delete(e.dataId)):(t.backend.decComplexRef(e.dataId),this.state.tensorInfo.get(e.dataId).refCount--)}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(a=>a.totalBytesSnapshot)),this.state.activeProfile.newBytes=this.state.numBytes-t,this.state.activeProfile.newTensors=this.state.numTensors-n;for(let a of this.state.activeProfile.kernels)a.kernelTimeMs=await a.kernelTimeMs,a.extraInfo=await a.extraInfo;return this.state.activeProfile}isTapeOn(){return this.state.gradientDepth>0&&this.state.kernelDepth===0}addTapeNode(e,t,n,a,r,s){let i={id:this.state.nextTapeNodeId++,kernelName:e,inputs:t,outputs:n,saved:r},o=Kg(e);o!=null&&(a=o.gradFunc),a!=null&&(i.gradient=l=>(l=l.map((c,u)=>{if(c==null){let p=n[u],d=gd(p.size,p.dtype);return this.makeTensor(d,p.shape,p.dtype)}return c}),a(l.length>1?l:l[0],r,s))),this.state.activeTape.push(i)}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=Zg(e),n=new Set(t.map(r=>r.id));for(let r=0;r<this.state.activeScope.track.length;r++){let s=this.state.activeScope.track[r];!s.kept&&!n.has(s.id)&&s.dispose()}let a=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===a.id&&this.track(r)})}gradients(e,t,n,a=!1){if(A(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));A(r instanceof Ae,()=>"The result y returned by f() must be a tensor.");let s=ZE(this.state.activeTape,t,r);if(!a&&s.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 i={};i[r.id]=n==null?lA(r.shape):n,JE(i,s,l=>this.tidy(l),uA);let o=t.map(l=>i[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:o}})}customGrad(e){return A(Or(e),()=>"The f passed in customGrad(f) must be a function."),(...t)=>{A(t.every(i=>i instanceof Ae),()=>"The args passed in customGrad(f)(x1, x2,...) must all be tensors");let n,a={};t.forEach((i,o)=>{a[o]=i});let r=(i,o)=>(n=e(...t,o),A(n.value instanceof Ae,()=>"The function f passed in customGrad(f) must return an object where `obj.value` is a tensor"),A(Or(n.gradFunc),()=>"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function."),n.value),s=(i,o)=>{let l=n.gradFunc(i,o),c=Array.isArray(l)?l:[l];A(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(...)."),A(c.every(p=>p instanceof Ae),()=>"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((p,d)=>{u[d]=()=>p}),u};return this.runKernelFunc({forwardFunc:r,backwardsFunc:s,inputs:a})}}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=Xg(),n=await this.backend.time(e);return n.wallMs=Xg()-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 p0;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}};Fc.nextTensorId=0;Fc.nextVariableId=0;function lA(e){let t=Wg(Pt(e),"float32");return M.makeTensor(t,e,"float32")}function d0(){let e=a0();if(e._tfengine==null){let t=new n0(e);e._tfengine=new Fc(t)}return zE(e._tfengine.ENV),nA(()=>e._tfengine),e._tfengine}var M=d0();function uA(e,t){let n={a:e,b:t};return M.runKernel(Lr,n)}var eh={};Oe(eh,{isBrowser:()=>h0,isMobile:()=>cA});function pA(){return typeof navigator!="undefined"&&navigator!=null}function cA(){if(pA()){let e=navigator.userAgent||navigator.vendor||window.opera;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(e)||/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(e.substr(0,4))}return!1}function h0(){return typeof window!="undefined"&&window.document!=null||typeof WorkerGlobalScope!="undefined"}var Ua=ee();Ua.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.")});Ua.registerFlag("IS_BROWSER",()=>h0());Ua.registerFlag("IS_NODE",()=>typeof process!="undefined"&&typeof process.versions!="undefined"&&typeof process.versions.node!="undefined");Ua.registerFlag("IS_CHROME",()=>typeof navigator!="undefined"&&navigator!=null&&navigator.userAgent!=null&&/Chrome/.test(navigator.userAgent)&&/Google Inc/.test(navigator.vendor));Ua.registerFlag("PROD",()=>!1);Ua.registerFlag("TENSORLIKE_CHECK_SHAPE_CONSISTENCY",()=>Ua.getBool("DEBUG"));Ua.registerFlag("DEPRECATION_WARNINGS_ENABLED",()=>!0);Ua.registerFlag("IS_TEST",()=>!1);Ua.registerFlag("CHECK_COMPUTATION_FOR_ERRORS",()=>!0);function Ga(e,t){let n=e;if(on(e))return t==="string"?[]:[e.length];if(!Array.isArray(e))return[];let a=[];for(;Array.isArray(n)||on(n)&&t!=="string";)a.push(n.length),n=n[0];return Array.isArray(e)&&ee().getBool("TENSORLIKE_CHECK_SHAPE_CONSISTENCY")&&m0(e,a,[]),a}function m0(e,t,n){if(n=n||[],!Array.isArray(e)&&!on(e)){A(t.length===0,()=>`Element arr[${n.join("][")}] is a primitive, but should be an array/TypedArray of ${t[0]} elements`);return}A(t.length>0,()=>`Element arr[${n.join("][")}] should be a primitive, but is an array of ${e.length} elements`),A(e.length===t[0],()=>`Element arr[${n.join("][")}] should have ${t[0]} elements, but has ${e.length} elements`);let a=t.slice(1);for(let r=0;r<e.length;++r)m0(e[r],a,n.concat(r))}function f0(e,t,n,a){if(e!=="string_or_numeric"){if(e==null)throw new Error("Expected dtype cannot be null.");if(e!=="numeric"&&e!==t||e==="numeric"&&t==="string")throw new Error(`Argument '${n}' passed to '${a}' must be ${e} tensor, but got ${t} tensor`)}}function E(e,t,n,a="numeric"){if(e instanceof Ae)return f0(a,e.dtype,t,n),e;let r=md(e);if(r!=="string"&&["bool","int32","float32"].indexOf(a)>=0&&(r=a),f0(a,r,t,n),e==null||!on(e)&&!Array.isArray(e)&&typeof e!="number"&&typeof e!="boolean"&&typeof e!="string"){let o=e==null?"null":e.constructor.name;throw new Error(`Argument '${t}' passed to '${n}' must be a Tensor or TensorLike, but got '${o}'`)}let s=Ga(e,r);!on(e)&&!Array.isArray(e)&&(e=[e]);let i=r!=="string"?Yd(e,r):Ts(e,[],!0);return M.makeTensor(i,s,r)}function $c(e,t,n,a="numeric"){if(!Array.isArray(e))throw new Error(`Argument ${t} passed to ${n} must be a \`Tensor[]\` or \`TensorLike[]\``);return e.map((r,s)=>E(r,`${t}[${s}]`,n,a))}var g0="__op";function P(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],a=e[n];n.endsWith("_")&&(n=n.substring(0,n.length-1)),n=n+g0;let r=(...s)=>{M.startScope(n);try{let i=a(...s);return Ug(i)&&console.error("Cannot return a Promise inside of tidy."),M.endScope(i),i}catch(i){throw M.endScope(null),i}};return Object.defineProperty(r,"name",{value:n,configurable:!0}),r}function dA(e,t){let n=E(e,"real","complex"),a=E(t,"imag","complex");sn(n.shape,a.shape,`real and imag shapes, ${n.shape} and ${a.shape}, must match in call to tf.complex().`);let r={real:n,imag:a};return M.runKernel(kd,r)}var Ur=P({complex_:dA});function Gr(e,t,n,a){if(a==null&&(a=md(e)),a==="complex64")throw new Error("Cannot construct a complex64 tensor directly. Please use tf.complex(real, imag).");if(!on(e)&&!Array.isArray(e)&&typeof e!="number"&&typeof e!="boolean"&&typeof e!="string")throw new Error("values passed to tensor(values) must be a number/boolean/string or an array of numbers/booleans/strings, or a TypedArray");if(t!=null){Vg(t);let r=Pt(t),s=Pt(n);A(r===s,()=>`Based on the provided shape, [${t}], the tensor should have ${r} values but has ${s}`);for(let i=0;i<n.length;++i){let o=n[i],l=i===n.length-1?o!==Pt(t.slice(i)):!0;A(n[i]===t[i]||!l,()=>`Error creating a new Tensor. Inferred shape (${n}) does not match the provided shape (${t}). `)}}return!on(e)&&!Array.isArray(e)&&(e=[e]),t=t||n,e=a!=="string"?Yd(e,a):Ts(e,[],!0),M.makeTensor(e,t,a)}function qn(e,t,n){let a=Ga(e,n);return Gr(e,t,a,n)}var ry={float32:4,float16:2,int32:4,uint16:2,uint8:1,bool:1,complex64:8},th=4;async function mA(e,t){let n=[],a=[],r=Array.isArray(e)?e.map(i=>i.name):Object.keys(e);for(let i=0;i<r.length;++i){let o=r[i],l=Array.isArray(e)?e[i].tensor:e[o];if(l.dtype!=="float32"&&l.dtype!=="int32"&&l.dtype!=="bool"&&l.dtype!=="string"&&l.dtype!=="complex64")throw new Error(`Unsupported dtype in weight '${o}': ${l.dtype}`);let c={name:o,shape:l.shape,dtype:l.dtype};if(l.dtype==="string"){let u=new Promise(async p=>{let d=await l.bytes(),h=d.reduce((g,y)=>g+y.length,0)+th*d.length,m=new Uint8Array(h),f=0;for(let g=0;g<d.length;g++){let y=d[g],b=new Uint8Array(new Uint32Array([y.length]).buffer);m.set(b,f),f+=th,m.set(y,f),f+=y.length}p(m)});a.push(u)}else a.push(l.data());t!=null&&(c.group=t),n.push(c)}let s=await Promise.all(a);return{data:hA(s),specs:n}}function y0(e,t){let n={},a,r=0;for(let s of t){let i=s.name,o=s.dtype,l=s.shape,c=Pt(l),u;if("quantization"in s){let p=s.quantization;if(p.dtype==="uint8"||p.dtype==="uint16"){if(!("min"in p&&"scale"in p))throw new Error(`Weight ${s.name} with quantization ${p.dtype} doesn't have corresponding metadata min and scale.`)}else if(p.dtype==="float16"){if(o!=="float32")throw new Error(`Weight ${s.name} is quantized with ${p.dtype} which only supports weights of type float32 not ${o}.`)}else throw new Error(`Weight ${s.name} has unknown quantization dtype ${p.dtype}. Supported quantization dtypes are: 'uint8', 'uint16', and 'float16'.`);let d=ry[p.dtype],h=e.slice(r,r+c*d),m=p.dtype==="uint8"?new Uint8Array(h):new Uint16Array(h);if(o==="float32")if(p.dtype==="uint8"||p.dtype==="uint16"){u=new Float32Array(m.length);for(let f=0;f<m.length;f++){let g=m[f];u[f]=g*p.scale+p.min}}else if(p.dtype==="float16")a===void 0&&(a=fA()),u=a(m);else throw new Error(`Unsupported quantization type ${p.dtype} for weight type float32.`);else if(o==="int32"){if(p.dtype!=="uint8"&&p.dtype!=="uint16")throw new Error(`Unsupported quantization type ${p.dtype} for weight type int32.`);u=new Int32Array(m.length);for(let f=0;f<m.length;f++){let g=m[f];u[f]=Math.round(g*p.scale+p.min)}}else throw new Error(`Unsupported dtype in weight '${i}': ${o}`);r+=c*d}else if(o==="string"){let p=Pt(s.shape);u=[];for(let d=0;d<p;d++){let h=new Uint32Array(e.slice(r,r+th))[0];r+=th;let m=new Uint8Array(e.slice(r,r+h));u.push(m),r+=h}}else{let p=ry[o],d=e.slice(r,r+c*p);if(o==="float32")u=new Float32Array(d);else if(o==="int32")u=new Int32Array(d);else if(o==="bool")u=new Uint8Array(d);else if(o==="complex64"){u=new Float32Array(d);let h=new Float32Array(u.length/2),m=new Float32Array(u.length/2);for(let y=0;y<h.length;y++)h[y]=u[y*2],m[y]=u[y*2+1];let f=qn(h,l,"float32"),g=qn(m,l,"float32");n[i]=Ur(f,g),f.dispose(),g.dispose()}else throw new Error(`Unsupported dtype in weight '${i}': ${o}`);r+=c*p}o!=="complex64"&&(n[i]=qn(u,l,o))}return n}function hA(e){if(e===null)throw new Error(`Invalid input value: ${JSON.stringify(e)}`);let t=0,n=[];e.forEach(s=>{if(t+=s.byteLength,n.push(s.byteLength===s.buffer.byteLength?s:new s.constructor(s)),!(s instanceof Float32Array||s instanceof Int32Array||s instanceof Uint8Array))throw new Error(`Unsupported TypedArray subtype: ${s.constructor.name}`)});let a=new Uint8Array(t),r=0;return n.forEach(s=>{a.set(new Uint8Array(s.buffer),r),r+=s.byteLength}),a.buffer}var sy=typeof Buffer!="undefined"&&(typeof Blob=="undefined"||typeof atob=="undefined"||typeof btoa=="undefined");function b0(e){return sy?Buffer.byteLength(e):new Blob([e]).size}function gA(e){if(sy)return Buffer.from(e).toString("base64");let t=new Uint8Array(e),n="";for(let a=0,r=t.length;a<r;a++)n+=String.fromCharCode(t[a]);return btoa(n)}function yA(e){if(sy){let a=Buffer.from(e,"base64");return a.buffer.slice(a.byteOffset,a.byteOffset+a.byteLength)}let t=atob(e),n=new Uint8Array(t.length);for(let a=0;a<t.length;++a)n.set([t.charCodeAt(a)],a);return n.buffer}function iy(e){if(e.length===1)return e[0];let t=0;e.forEach(r=>{t+=r.byteLength});let n=new Uint8Array(t),a=0;return e.forEach(r=>{n.set(new Uint8Array(r),a),a+=r.byteLength}),n.buffer}function x0(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 Dc(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:b0(JSON.stringify(e.modelTopology)),weightSpecsBytes:e.weightSpecs==null?0:b0(JSON.stringify(e.weightSpecs)),weightDataBytes:e.weightData==null?0:e.weightData.byteLength}}function bA(){let e=n=>{let a=n<<13,r=0;for(;(a&8388608)==0;)r-=8388608,a<<=1;return a&=~8388608,r+=947912704,a|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 xA(){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 vA(){let e=new Uint32Array(64);for(let t=0;t<64;t++)e[t]=1024;return e[0]=e[32]=0,e}function fA(){let e=bA(),t=xA(),n=vA();return a=>{let r=new ArrayBuffer(4*a.length),s=new Uint32Array(r);for(let i=0;i<a.length;i++){let o=a[i],l=e[n[o>>10]+(o&1023)]+t[o>>10];s[i]=l}return new Float32Array(r)}}var Et=class{constructor(){this.saveRouters=[],this.loadRouters=[]}static getInstance(){return Et.instance==null&&(Et.instance=new Et),Et.instance}static registerSaveRouter(e){Et.getInstance().saveRouters.push(e)}static registerLoadRouter(e){Et.getInstance().loadRouters.push(e)}static getSaveHandlers(e){return Et.getHandlers(e,"save")}static getLoadHandlers(e,t){return Et.getHandlers(e,"load",t)}static getHandlers(e,t,n){let a=[];return(t==="load"?Et.getInstance().loadRouters:Et.getInstance().saveRouters).forEach(r=>{let s=r(e,n);s!==null&&a.push(s)}),a}},wA=e=>Et.registerSaveRouter(e),kA=e=>Et.registerLoadRouter(e),IA=e=>Et.getSaveHandlers(e),NA=(e,t)=>Et.getLoadHandlers(e,t),oy="tensorflowjs",ly=1,xi="models_store",Hr="model_info_store";function v0(){if(!ee().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 uy(e){let t=e.result;t.createObjectStore(xi,{keyPath:"modelPath"}),t.createObjectStore(Hr,{keyPath:"modelPath"})}var vi=class{constructor(e){if(this.indexedDB=v0(),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,a)=>{let r=this.indexedDB.open(oy,ly);r.onupgradeneeded=()=>uy(r),r.onsuccess=()=>{let s=r.result;if(t==null){let i=s.transaction(xi,"readonly"),o=i.objectStore(xi).get(this.modelPath);o.onsuccess=()=>{if(o.result==null)return s.close(),a(new Error(`Cannot find model with path '${this.modelPath}' in IndexedDB.`));n(o.result.modelArtifacts)},o.onerror=l=>(s.close(),a(o.error)),i.oncomplete=()=>s.close()}else{let i=Dc(t),o=s.transaction(Hr,"readwrite"),l=o.objectStore(Hr),c=l.put({modelPath:this.modelPath,modelArtifactsInfo:i}),u;c.onsuccess=()=>{u=s.transaction(xi,"readwrite");let p=u.objectStore(xi).put({modelPath:this.modelPath,modelArtifacts:t,modelArtifactsInfo:i});p.onsuccess=()=>n({modelArtifactsInfo:i}),p.onerror=d=>{l=o.objectStore(Hr);let h=l.delete(this.modelPath);h.onsuccess=()=>(s.close(),a(p.error)),h.onerror=m=>(s.close(),a(p.error))}},c.onerror=p=>(s.close(),a(c.error)),o.oncomplete=()=>{u==null?s.close():u.oncomplete=()=>s.close()}}},r.onerror=s=>a(r.error)})}};vi.URL_SCHEME="indexeddb://";var w0=e=>ee().getBool("IS_BROWSER")&&!Array.isArray(e)&&e.startsWith(vi.URL_SCHEME)?TA(e.slice(vi.URL_SCHEME.length)):null;Et.registerSaveRouter(w0);Et.registerLoadRouter(w0);function TA(e){return new vi(e)}function SA(e){return e.startsWith(vi.URL_SCHEME)?e.slice(vi.URL_SCHEME.length):e}var CA=class{constructor(){this.indexedDB=v0()}async listModels(){return new Promise((e,t)=>{let n=this.indexedDB.open(oy,ly);n.onupgradeneeded=()=>uy(n),n.onsuccess=()=>{let a=n.result,r=a.transaction(Hr,"readonly"),s=r.objectStore(Hr).getAll();s.onsuccess=()=>{let i={};for(let o of s.result)i[o.modelPath]=o.modelArtifactsInfo;e(i)},s.onerror=i=>(a.close(),t(s.error)),r.oncomplete=()=>a.close()},n.onerror=a=>t(n.error)})}async removeModel(e){return e=SA(e),new Promise((t,n)=>{let a=this.indexedDB.open(oy,ly);a.onupgradeneeded=()=>uy(a),a.onsuccess=()=>{let r=a.result,s=r.transaction(Hr,"readwrite"),i=s.objectStore(Hr),o=i.get(e),l;o.onsuccess=()=>{if(o.result==null)return r.close(),n(new Error(`Cannot find model with path '${e}' in IndexedDB.`));{let c=i.delete(e),u=()=>{l=r.transaction(xi,"readwrite");let p=l.objectStore(xi).delete(e);p.onsuccess=()=>t(o.result.modelArtifactsInfo),p.onerror=d=>n(o.error)};c.onsuccess=u,c.onerror=p=>(u(),r.close(),n(o.error))}},o.onerror=c=>(r.close(),n(o.error)),s.oncomplete=()=>{l==null?r.close():l.oncomplete=()=>r.close()}},a.onerror=r=>n(a.error)})}},mr="/",El="tensorflowjs_models",k0="info",_A="model_topology",EA="weight_specs",AA="weight_data",FA="model_metadata";function I0(e){return{info:[El,e,k0].join(mr),topology:[El,e,_A].join(mr),weightSpecs:[El,e,EA].join(mr),weightData:[El,e,AA].join(mr),modelMetadata:[El,e,FA].join(mr)}}function $A(e){let t=e.split(mr);if(t.length<3)throw new Error(`Invalid key format: ${e}`);return t.slice(1,t.length-1).join(mr)}function DA(e){return e.startsWith(wi.URL_SCHEME)?e.slice(wi.URL_SCHEME.length):e}var wi=class{constructor(e){if(!ee().getBool("IS_BROWSER")||typeof window=="undefined"||typeof window.localStorage=="undefined")throw new Error("The current environment does not support local storage.");if(this.LS=window.localStorage,e==null||!e)throw new Error("For local storage, modelPath must not be null, undefined or empty.");this.modelPath=e,this.keys=I0(this.modelPath)}async save(e){if(e.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");{let t=JSON.stringify(e.modelTopology),n=JSON.stringify(e.weightSpecs),a=Dc(e);try{this.LS.setItem(this.keys.info,JSON.stringify(a)),this.LS.setItem(this.keys.topology,t),this.LS.setItem(this.keys.weightSpecs,n),this.LS.setItem(this.keys.weightData,gA(e.weightData));let r={format:e.format,generatedBy:e.generatedBy,convertedBy:e.convertedBy};return e.signature!=null&&(r.signature=e.signature),e.userDefinedMetadata!=null&&(r.userDefinedMetadata=e.userDefinedMetadata),e.modelInitializer!=null&&(r.modelInitializer=e.modelInitializer),this.LS.setItem(this.keys.modelMetadata,JSON.stringify(r)),{modelArtifactsInfo:a}}catch(r){throw this.LS.removeItem(this.keys.info),this.LS.removeItem(this.keys.topology),this.LS.removeItem(this.keys.weightSpecs),this.LS.removeItem(this.keys.weightData),this.LS.removeItem(this.keys.modelMetadata),new Error(`Failed to save model '${this.modelPath}' to local storage: size quota being exceeded is a possible cause of this failure: modelTopologyBytes=${a.modelTopologyBytes}, weightSpecsBytes=${a.weightSpecsBytes}, weightDataBytes=${a.weightDataBytes}.`)}}}async load(){let e=JSON.parse(this.LS.getItem(this.keys.info));if(e==null)throw new Error(`In local storage, there is no model with name '${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 '${this.modelPath}' is missing.`);t.modelTopology=n;let a=JSON.parse(this.LS.getItem(this.keys.weightSpecs));if(a==null)throw new Error(`In local storage, the weight specs of model '${this.modelPath}' are missing.`);t.weightSpecs=a;let r=this.LS.getItem(this.keys.modelMetadata);if(r!=null){let i=JSON.parse(r);t.format=i.format,t.generatedBy=i.generatedBy,t.convertedBy=i.convertedBy,i.signature!=null&&(t.signature=i.signature),i.userDefinedMetadata!=null&&(t.userDefinedMetadata=i.userDefinedMetadata),i.modelInitializer!=null&&(t.modelInitializer=i.modelInitializer)}let s=this.LS.getItem(this.keys.weightData);if(s==null)throw new Error(`In local storage, the binary weight values of model '${this.modelPath}' are missing.`);return t.weightData=yA(s),t}};wi.URL_SCHEME="localstorage://";var N0=e=>ee().getBool("IS_BROWSER")&&!Array.isArray(e)&&e.startsWith(wi.URL_SCHEME)?RA(e.slice(wi.URL_SCHEME.length)):null;Et.registerSaveRouter(N0);Et.registerLoadRouter(N0);function RA(e){return new wi(e)}var MA=class{constructor(){A(ee().getBool("IS_BROWSER"),()=>"Current environment is not a web browser"),A(typeof window=="undefined"||typeof window.localStorage!="undefined",()=>"Current browser does not appear to support localStorage"),this.LS=window.localStorage}async listModels(){let e={},t=El+mr,n=mr+k0;for(let a=0;a<this.LS.length;++a){let r=this.LS.key(a);if(r.startsWith(t)&&r.endsWith(n)){let s=$A(r);e[s]=JSON.parse(this.LS.getItem(r))}}return e}async removeModel(e){e=DA(e);let t=I0(e);if(this.LS.getItem(t.info)==null)throw new Error(`Cannot find model at path '${e}'`);let n=JSON.parse(this.LS.getItem(t.info));return this.LS.removeItem(t.info),this.LS.removeItem(t.topology),this.LS.removeItem(t.weightSpecs),this.LS.removeItem(t.weightData),n}},Al="://",Kn=class{constructor(){this.managers={}}static getInstance(){return Kn.instance==null&&(Kn.instance=new Kn),Kn.instance}static registerManager(e,t){A(e!=null,()=>"scheme must not be undefined or null."),e.endsWith(Al)&&(e=e.slice(0,e.indexOf(Al))),A(e.length>0,()=>"scheme must not be an empty string.");let n=Kn.getInstance();A(n.managers[e]==null,()=>`A model store manager is already registered for scheme '${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 nh(e){if(e.indexOf(Al)===-1)throw new Error(`The url string provided does not contain a scheme. Supported schemes are: ${Kn.getSchemes().join(",")}`);return{scheme:e.split(Al)[0],path:e.split(Al)[1]}}async function T0(e,t,n=!1){A(e!==t,()=>`Old path and new path are the same: '${e}'`);let a=Et.getLoadHandlers(e);A(a.length>0,()=>`Copying failed because no load handler is found for source URL ${e}.`),A(a.length<2,()=>`Copying failed because more than one (${a.length}) load handlers for source URL ${e}.`);let r=a[0],s=Et.getSaveHandlers(t);A(s.length>0,()=>`Copying failed because no save handler is found for destination URL ${t}.`),A(s.length<2,()=>`Copying failed because more than one (${a.length}) save handlers for destination URL ${t}.`);let i=s[0],o=nh(e).scheme,l=nh(e).path,c=o===nh(e).scheme,u=await r.load();n&&c&&await Kn.getManager(o).removeModel(l);let p=await i.save(u);return n&&!c&&await Kn.getManager(o).removeModel(l),p.modelArtifactsInfo}async function PA(){let e=Kn.getSchemes(),t={};for(let n of e){let a=await Kn.getManager(n).listModels();for(let r in a){let s=n+Al+r;t[s]=a[r]}}return t}async function OA(e){let t=nh(e);return Kn.getManager(t.scheme).removeModel(t.path)}async function LA(e,t){return T0(e,t,!1)}async function zA(e,t){return T0(e,t,!0)}var BA=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(ee().get("IS_BROWSER")){ee().setPlatform("browser",new BA);try{Kn.registerManager(wi.URL_SCHEME,new MA)}catch(e){}try{Kn.registerManager(vi.URL_SCHEME,new CA)}catch(e){}}var WA={importFetch:()=>sE()},cy,VA=class{constructor(){this.util=require("util"),this.textEncoder=new this.util.TextEncoder}fetch(e,t){return ee().global.fetch!=null?ee().global.fetch(e,t):(cy==null&&(cy=WA.importFetch()),cy(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)}};ee().get("IS_NODE")&&ee().setPlatform("node",new VA);function Le(e,t="float32",n){return t=t||"float32",Vg(e),new Ot(e,t,n)}function UA(e,t){let n=E(e,"x","cast");if(!Xw(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 a={x:n},r={dtype:t};return M.runKernel(As,a,r)}var ce=P({cast_:UA});function GA(e){let t={x:E(e,"x","clone","string_or_numeric")};return M.runKernel(Ko,t)}var jr=P({clone_:GA});function S0(e,t=!1){console.log(e.toString(t))}d0();var HA={buffer:Le,cast:ce,clone:jr,print:S0};aA(HA);var Ht={};Oe(Ht,{browserFiles:()=>jA,browserHTTPRequest:()=>KA,concatenateArrayBuffers:()=>iy,copyModel:()=>LA,decodeWeights:()=>y0,encodeWeights:()=>mA,fromMemory:()=>XA,getLoadHandlers:()=>NA,getModelArtifactsInfoForJSON:()=>Dc,getSaveHandlers:()=>IA,http:()=>dy,isHTTPScheme:()=>py,listModels:()=>PA,loadWeights:()=>qA,moveModel:()=>zA,registerLoadRouter:()=>kA,registerSaveRouter:()=>wA,removeModel:()=>OA,weightsLoaderFactory:()=>C0,withSaveHandler:()=>YA});var ZA="model",JA=".json",QA=".weights.bin";function _0(e){return new Promise(t=>setTimeout(t)).then(e)}var Fl=class{constructor(e){if(!ee().getBool("IS_BROWSER"))throw new Error("browserDownloads() cannot proceed because the current environment is not a browser.");e.startsWith(Fl.URL_SCHEME)&&(e=e.slice(Fl.URL_SCHEME.length)),(e==null||e.length===0)&&(e=ZA),this.modelTopologyFileName=e+JA,this.weightDataFileName=e+QA}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}],a={modelTopology:e.modelTopology,format:e.format,generatedBy:e.generatedBy,convertedBy:e.convertedBy,weightsManifest:n};e.signature!=null&&(a.signature=e.signature),e.userDefinedMetadata!=null&&(a.userDefinedMetadata=e.userDefinedMetadata),e.modelInitializer!=null&&(a.modelInitializer=e.modelInitializer);let r=window.URL.createObjectURL(new Blob([JSON.stringify(a)],{type:"application/json"})),s=this.jsonAnchor==null?document.createElement("a"):this.jsonAnchor;if(s.download=this.modelTopologyFileName,s.href=r,await _0(()=>s.dispatchEvent(new MouseEvent("click"))),e.weightData!=null){let i=this.weightDataAnchor==null?document.createElement("a"):this.weightDataAnchor;i.download=this.weightDataFileName,i.href=t,await _0(()=>i.dispatchEvent(new MouseEvent("click")))}return{modelArtifactsInfo:Dc(e)}}}};Fl.URL_SCHEME="downloads://";var eF=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.files=e}async load(){let e=this.files[0],t=this.files.slice(1);return new Promise((n,a)=>{let r=new FileReader;r.onload=s=>{let i=JSON.parse(s.target.result),o=i.modelTopology;if(o==null){a(new Error(`modelTopology field is missing from file ${e.name}`));return}t.length===0&&n({modelTopology:o});let l=i.weightsManifest;if(l==null){a(new Error(`weightManifest field is missing from file ${e.name}`));return}let c;try{c=this.checkManifestAndWeightFiles(l,t)}catch(h){a(h);return}let u=[],p=[],d=[];l.forEach(h=>{h.paths.forEach(m=>{p.push(m),d.push(null)}),u.push(...h.weights)}),l.forEach(h=>{h.paths.forEach(m=>{let f=new FileReader;f.onload=g=>{let y=g.target.result,b=p.indexOf(m);if(d[b]=y,d.indexOf(null)===-1){let v={modelTopology:o,weightSpecs:u,weightData:iy(d),format:i.format,generatedBy:i.generatedBy,convertedBy:i.convertedBy};i.signature!=null&&(v.signature=i.signature),i.userDefinedMetadata!=null&&(v.userDefinedMetadata=i.userDefinedMetadata),i.modelInitializer!=null&&(v.modelInitializer=i.modelInitializer),n(v)}},f.onerror=g=>a(`Failed to weights data from file of path '${m}'.`),f.readAsArrayBuffer(c[m])})})},r.onerror=s=>a(`Failed to read model topology and weights manifest JSON from file '${e.name}'. BrowserFiles supports loading Keras-style tf.Model artifacts only.`),r.readAsText(e)})}checkManifestAndWeightFiles(e,t){let n=[],a=t.map(s=>x0(s.name)),r={};for(let s of e)s.paths.forEach(i=>{let o=x0(i);if(n.indexOf(o)!==-1)throw new Error(`Duplicate file basename found in weights manifest: '${o}'`);if(n.push(o),a.indexOf(o)===-1)throw new Error(`Weight file with basename '${o}' is not provided.`);r[i]=t[a.indexOf(o)]});if(n.length!==t.length)throw new Error(`Mismatch in the number of files in weights manifest (${n.length}) and the number of weight files provided (${t.length}).`);return r}},nF=e=>ee().getBool("IS_BROWSER")&&!Array.isArray(e)&&e.startsWith(Fl.URL_SCHEME)?tF(e.slice(Fl.URL_SCHEME.length)):null;Et.registerSaveRouter(nF);function tF(e="model"){return new Fl(e)}function jA(e){return new eF(e)}function E0(e,t,n,a){i(e),n=n==null?0:n,a=a==null?1:a,o(n,a);let r=0,s=l=>(l.then(c=>{let u=n+ ++r/e.length*(a-n);return t(u),c}),l);function i(l){A(l!=null&&Array.isArray(l)&&l.length>0,()=>"promises must be a none empty array")}function o(l,c){A(l>=0&&l<=1,()=>`Progress fraction must be in range [0, 1], but got startFraction ${l}`),A(c>=0&&c<=1,()=>`Progress fraction must be in range [0, 1], but got endFraction ${c}`),A(c>=l,()=>`startFraction must be no more than endFraction, but got startFraction ${l} and endFraction ${c}`)}return Promise.all(e.map(s))}async function A0(e,t){t==null&&(t={});let n=t.fetchFunc==null?ee().platform.fetch:t.fetchFunc,a=e.map(c=>n(c,t.requestInit,{isBinary:!0})),r=0,s=.5,i=(t.onProgress==null?await Promise.all(a):await E0(a,t.onProgress,r,s)).map(c=>c.arrayBuffer()),o=.5,l=1;return t.onProgress==null?await Promise.all(i):await E0(i,t.onProgress,o,l)}async function qA(e,t="",n,a){return C0(r=>A0(r,{requestInit:a}))(e,t,n)}function C0(e){return async(t,n="",a)=>{let r=t.map(()=>!1),s={},i=a!=null?a.map(()=>!1):[],o=[];if(t.forEach((h,m)=>{let f=0;h.weights.forEach(g=>{let y="quantization"in g?g.quantization.dtype:g.dtype,b=ry[y]*Pt(g.shape),v=()=>{r[m]=!0,s[m]==null&&(s[m]=[]),s[m].push({manifestEntry:g,groupOffset:f,sizeBytes:b})};a!=null?a.forEach((x,N)=>{x===g.name&&(v(),i[N]=!0)}):v(),o.push(g.name),f+=b})}),!i.every(h=>h)){let h=a.filter((m,f)=>!i[f]);throw new Error(`Could not find weights in manifest with names: ${h.join(", ")}.
|
|
Manifest JSON has weights with names: ${o.join(", ")}.`)}let l=r.reduce((h,m,f)=>(m&&h.push(f),h),[]),c=[];l.forEach(h=>{t[h].paths.forEach(m=>{let f=n+(n.endsWith("/")?"":"/")+m;c.push(f)})});let u=await e(c),p={},d=0;return l.forEach(h=>{let m=t[h].paths.length,f=0;for(let v=0;v<m;v++)f+=u[d+v].byteLength;let g=new ArrayBuffer(f),y=new Uint8Array(g),b=0;for(let v=0;v<m;v++){let x=new Uint8Array(u[d+v]);y.set(x,b),b+=x.byteLength}s[h].forEach(v=>{let x=g.slice(v.groupOffset,v.groupOffset+v.sizeBytes),N=y0(x,[v.manifestEntry]);for(let T in N)p[T]=N[T]}),d+=m}),p}}var aF="application/octet-stream",rF="application/json",hy=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?(A(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=ee().platform.fetch,A(e!=null&&e.length>0,()=>"URL path for http must not be null, undefined or empty."),Array.isArray(e)&&A(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}],a={modelTopology:e.modelTopology,format:e.format,generatedBy:e.generatedBy,convertedBy:e.convertedBy,weightsManifest:n};e.signature!=null&&(a.signature=e.signature),e.userDefinedMetadata!=null&&(a.userDefinedMetadata=e.userDefinedMetadata),e.modelInitializer!=null&&(a.modelInitializer=e.modelInitializer),t.body.append("model.json",new Blob([JSON.stringify(a)],{type:rF}),"model.json"),e.weightData!=null&&t.body.append("model.weights.bin",new Blob([e.weightData],{type:aF}),"model.weights.bin");let r=await this.fetch(this.path,t);if(r.ok)return{modelArtifactsInfo:Dc(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(h){let m=`Failed to parse model JSON of response from ${this.path}.`;throw this.path.endsWith(".pb")?m+=" 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.":m+=" Please make sure the server is serving valid JSON for this request.",new Error(m)}let n=t.modelTopology,a=t.weightsManifest,r=t.generatedBy,s=t.convertedBy,i=t.format,o=t.signature,l=t.userDefinedMetadata;if(n==null&&a==null)throw new Error(`The JSON from HTTP path ${this.path} contains neither model topology or manifest for weights.`);let c,u;a!=null&&([c,u]=await this.loadWeights(a));let p={modelTopology:n,weightSpecs:c,weightData:u,generatedBy:r,convertedBy:s,format:i};o!=null&&(p.signature=o),l!=null&&(p.userDefinedMetadata=l);let d=t.modelInitializer;return d&&(p.modelInitializer=d),p}async loadWeights(e){let t=Array.isArray(this.path)?this.path[1]:this.path,[n,a]=sF(t),r=this.weightPathPrefix||n,s=[];for(let c of e)s.push(...c.weights);let i=[],o=[];for(let c of e)for(let u of c.paths)this.weightUrlConverter!=null?o.push(this.weightUrlConverter(u)):i.push(r+u+a);this.weightUrlConverter&&i.push(...await Promise.all(o));let l=await A0(i,{requestInit:this.requestInit,fetchFunc:this.fetch,onProgress:this.onProgress});return[s,iy(l)]}};hy.URL_SCHEME_REGEX=/^https?:\/\//;function sF(e){let t=e.lastIndexOf("/"),n=e.lastIndexOf("?"),a=e.substring(0,t),r=n>t?e.substring(n):"";return[a+"/",r]}function py(e){return e.match(hy.URL_SCHEME_REGEX)!=null}var F0=(e,t)=>{if(typeof fetch=="undefined"&&(t==null||t.fetchFunc==null))return null;{let n=!0;if(Array.isArray(e)?n=e.every(a=>py(a)):n=py(e),n)return dy(e,t)}return null};Et.registerSaveRouter(F0);Et.registerLoadRouter(F0);function dy(e,t){return new hy(e,t)}function KA(e,t){return dy(e,t)}var my=class{constructor(e){this.modelArtifacts=e}async load(){return this.modelArtifacts}},iF=class{constructor(e){this.saveHandler=e}async save(e){return this.saveHandler(e)}};function XA(e,t,n,a){return arguments.length===1?e.modelTopology!=null||e.weightSpecs!=null?new my(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 my({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 my({modelTopology:e,weightSpecs:t,weightData:n,trainingConfig:a}))}function YA(e){return new iF(e)}var $0={};Oe($0,{confusionMatrix:()=>oF});function lF(e,t,n=!1,a=!1){let r=E(e,"a","matMul"),s=E(t,"b","matMul");[r,s]=Nt(r,s);let i={a:r,b:s},o={transposeA:n,transposeB:a};return M.runKernel(Es,i,o)}var ze=P({matMul_:lF});function uF(e,t,n=1,a=0){if(t<2)throw new Error(`Error in oneHot: depth must be >=2, but it is ${t}`);let r={indices:E(e,"indices","oneHot","int32")},s={depth:t,onValue:n,offValue:a};return M.runKernel(Zs,r,s)}var $l=P({oneHot_:uF});function cF(e,t){let n=E(e,"x","transpose");if(t==null&&(t=n.shape.map((s,i)=>i).reverse()),A(n.rank===t.length,()=>`Error in transpose: rank of input ${n.rank} must match length of perm ${t}.`),t.forEach(s=>{A(s>=0&&s<n.rank,()=>`All entries in 'perm' must be between 0 and ${n.rank-1} but got ${t}`)}),n.rank<=1)return n.clone();let a={x:n},r={perm:t};return M.runKernel(fi,a,r)}var Ue=P({transpose_:cF});function pF(e,t,n){let a=E(e,"labels","confusionMatrix"),r=E(t,"predictions","confusionMatrix");A(n==null||n>0&&Number.isInteger(n),()=>`If provided, numClasses must be a positive integer, but got ${n}`),A(a.rank===1,()=>`Expected the rank of labels to be 1, but got ${a.rank}`),A(r.rank===1,()=>`Expected the rank of predictions to be 1, but got ${r.rank}`),A(a.shape[0]===r.shape[0],()=>`Mismatch in the number of examples: ${a.shape[0]} vs. ${r.shape[0]}. Labels and predictions should have the same number of elements.`),A(n>0&&Number.isInteger(n),()=>`numClasses is required to be a positive integer, but got ${n}`);let s=$l(ce(a,"int32"),n),i=$l(ce(r,"int32"),n),o=Ue(s),l=ze(o,i);return ce(l,"int32")}var oF=P({confusionMatrix_:pF}),ki={};Oe(ki,{fromPixels:()=>hF,toPixels:()=>dF});function ah(e,t,n){if(Ns(e),t!=null&&t.length!==3)throw new Error("tensor3d() requires shape to have three numbers");let a=Ga(e,n);if(a.length!==3&&a.length!==1)throw new Error("tensor3d() requires values to be number[][][] or flat/TypedArray");if(a.length===1&&t==null)throw new Error("tensor3d() requires shape to be provided when `values` are a flat array");return Gr(e,t,a,n)}var Dl;function mF(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,a=!1,r=!1,s=!1,i=!1,o=!1;if(e.data instanceof Uint8Array)n=!0;else if(typeof ImageData!="undefined"&&e instanceof ImageData)a=!0;else if(typeof HTMLVideoElement!="undefined"&&e instanceof HTMLVideoElement)r=!0;else if(typeof HTMLImageElement!="undefined"&&e instanceof HTMLImageElement)s=!0;else if(e.getContext!=null)i=!0;else if(typeof ImageBitmap!="undefined"&&e instanceof ImageBitmap)o=!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 d=2;if(r&&e.readyState<d)throw new Error("The video element has not loaded data yet. Please wait for `loadeddata` event on the <video> element.")}if(Kd(qd,M.backendName)!=null){let d={pixels:e},h={numChannels:t};return M.runKernel(qd,d,h)}let[l,c]=r?[e.videoWidth,e.videoHeight]:[e.width,e.height],u;i?u=e.getContext("2d").getImageData(0,0,l,c).data:a||n?u=e.data:(s||r||o)&&(Dl==null&&(Dl=document.createElement("canvas").getContext("2d")),Dl.canvas.width=l,Dl.canvas.height=c,Dl.drawImage(e,0,0,l,c),u=Dl.getImageData(0,0,l,c).data);let p;if(t===4)p=new Int32Array(u);else{let d=l*c;p=new Int32Array(d*t);for(let h=0;h<d;h++)for(let m=0;m<t;++m)p[h*t+m]=u[h*4+m]}return ah(p,[c,l,t],"int32")}async function dF(e,t){let n=E(e,"img","toPixels");if(!(e instanceof Ae)){let c=n;n=ce(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[a,r]=n.shape.slice(0,2),s=n.rank===2?1:n.shape[2];if(s>4||s===2)throw new Error(`toPixels only supports depth of size 1, 3 or 4 but got ${s}`);if(n.dtype!=="float32"&&n.dtype!=="int32")throw new Error(`Unsupported type for toPixels: ${n.dtype}. Please use float32 or int32 tensors.`);let i=await n.data(),o=n.dtype==="float32"?255:1,l=new Uint8ClampedArray(r*a*4);for(let c=0;c<a*r;++c){let u=[0,0,0,255];for(let d=0;d<s;d++){let h=i[c*s+d];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}.`);s===1?(u[0]=h*o,u[1]=h*o,u[2]=h*o):u[d]=h*o}let p=c*4;l[p+0]=Math.round(u[0]),l[p+1]=Math.round(u[1]),l[p+2]=Math.round(u[2]),l[p+3]=Math.round(u[3])}if(t!=null){t.width=r,t.height=a;let c=t.getContext("2d"),u=new ImageData(l,r,a);c.putImageData(u,0,0)}return n!==e&&n.dispose(),l}var hF=P({fromPixels_:mF}),fy={};Oe(fy,{prepareAndValidate:()=>D0});function D0(e,t){let n=e.shape.length,a=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(a<1)throw new Error(`tf.gatherND() expects the indices to be rank 1 or higher, but the rank was ${a}.`);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[a-1]>n)throw new Error(`index innermost dimension length must be <= tensor rank; saw: ${t.shape[a-1]} vs. ${n}`);if(Pt(e.shape)===0)throw new Error(`Requested more than 0 entries, but input is empty. Input shape: ${e.shape}.`);let r=t.shape,s=r[r.length-1],i=1;for(let p=0;p<r.length-1;++p)i*=r[p];let o=e.shape,l=r.slice();l.pop();let c=1;for(let p=s;p<n;++p)c*=o[p],l.push(o[p]);let u=[...No(e.shape).map(p=>p/c),1].slice(0,s);return[l,i,c,u]}var gy={};Oe(gy,{calculateShapes:()=>R0,validateInput:()=>by,validateUpdateShape:()=>yy});function yy(e,t,n){let a=t.rank>1?t.shape[t.rank-1]:1,r=t.rank>1?t.rank-1:1,s=`Must have updates.shape = indices.shape[:batchDim] + shape[sliceDim:], got updates.shape: ${n.shape}, indices.shape: ${t.shape}, shape: ${e}, sliceDim: ${a}, and batchDim: ${r}.`;if(n.rank<r)throw new Error(s+` update.rank < ${r}. `);if(e.length<a+(n.rank-r))throw new Error(s+` Output shape length < ${a+(n.rank-r)}`);if(n.rank!==r+e.length-a)throw new Error(s+` update.rank != ${r+e.length-a}`);for(let i=0;i<r;++i)if(n.shape[i]!==t.shape[i])throw new Error(s+` updates.shape[${i}] (${n.shape[i]}) != indices.shape[${i}] (${t.shape[i]}).`);for(let i=0;i<n.rank-r;++i)if(n.shape[i+r]!==e[i+a])throw new Error(s+` updates.shape[${i+r}] (${n.shape[i+r]}) != shape[${i+r}] (${e[i+r]})`)}function by(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: ${e.shape}`)}yy(n,t,e)}function R0(e,t,n){let a=t.shape.length,r=a>1?t.shape[a-1]:1,s=n.length,i=1;for(let p=r;p<s;++p)i*=n[p];let o=r<1?1:r,l=Pt(t.shape)/o,c=[...No(n.slice(0,r)),1],u=Pt(n);return{sliceRank:r,numUpdates:l,sliceSize:i,strides:c,outputSize:u}}var cn={};Oe(cn,{assertParamsValid:()=>fF,computeFlatOffset:()=>yF,computeOutShape:()=>M0,getNormalizedAxes:()=>O0,isSliceContinous:()=>gF,maskToAxes:()=>rh,parseSliceParams:()=>U0,sliceInfo:()=>bF,startForAxis:()=>W0,startIndicesWithElidedDims:()=>L0,stopForAxis:()=>V0,stopIndicesWithElidedDims:()=>z0,stridesForAxis:()=>B0,stridesWithElidedDims:()=>P0});function fF(e,t,n){let a=e.shape.length;A(a===t.length,()=>`Error in slice${a}D: Length of begin ${t} must match the rank of the array (${a}).`),A(a===n.length,()=>`Error in slice${a}D: Length of size ${n} must match the rank of the array (${a}).`);for(let r=0;r<a;++r)A(t[r]+n[r]<=e.shape[r],()=>`Error in slice${a}D: begin[${r}] + size[${r}] (${t[r]+n[r]}) would overflow input.shape[${r}] (${e.shape[r]})`)}function rh(e){let t=[],n=0;for(;e>0;)e&1&&t.push(n),e/=2,n++;return t}function M0(e,t,n){let a=[];for(let r=0;r<e.length;r++)a[r]=Math.ceil((t[r]-e[r])/n[r]);return a}function P0(e,t,n,a){let r=[...e];for(let s=r.length;s<a.length;s++)r.push(1);for(let s=0;s<n;s++)s===0?r[t]=1:(r.splice(t,0,1),r.pop());return r}function G0(e,t,n){return n<=e?n:n-(t-1)}function H0(e,t){let n=[];for(let a=0;a<e;a++)n.push(t+a);return n}function O0(e,t,n,a,r,s,i,o,l){let c=e.length,u=new Array(c),p=new Array(c),d=new Array(c);if(t.length&&n>0){let h=t[0],m=n+1;u=L0(i,h,m,a,e),p=z0(o,h,m,r,e),d=P0(s,h,m,e)}else for(let h=0;h<c;h++)u[h]=W0(i,a,s,e,h,l),p[h]=V0(o,r,s,e,h,l),d[h]=B0(s,h,l);return{begin:u,end:p,strides:d}}function L0(e,t,n,a,r){let s=[...r],i=H0(n,t);for(let o=0;o<s.length;o++)if(i.indexOf(o)>-1)s[o]=0;else{let l=G0(t,n,o),c=a[l];e&1<<l&&(c=0),s[o]=c}return s}function z0(e,t,n,a,r){let s=[...r],i=H0(n,t);for(let o=0;o<s.length;o++)if(i.indexOf(o)>-1)s[o]=Number.MAX_SAFE_INTEGER;else{let l=G0(t,n,o),c=a[l];e&1<<l&&(c=Number.MAX_SAFE_INTEGER),s[o]=c}for(let o=0;o<s.length;o++){let l=r[o];s[o]<0&&(s[o]+=l),s[o]=ic(0,s[o],r[o])}return s}function B0(e,t,n){let a=e[t];return(n&1<<t||a==null)&&(a=1),a}function W0(e,t,n,a,r,s){let i=t[r],o=n[r]||1;(e&1<<r||s&1<<r||i==null)&&(o>0?i=Number.MIN_SAFE_INTEGER:i=Number.MAX_SAFE_INTEGER);let l=a[r];return i<0&&(i+=l),i=ic(0,i,l-1),i}function V0(e,t,n,a,r,s){let i=t[r],o=n[r]||1;(e&1<<r||s&1<<r||i==null)&&(o>0?i=Number.MAX_SAFE_INTEGER:i=Number.MIN_SAFE_INTEGER);let l=a[r];return i<0&&(i+=l),o>0?i=ic(0,i,l):i=ic(-1,i,l-1),i}function gF(e,t,n){let a=n.length;for(let r=0;r<n.length;r++)if(n[r]>1){a=r;break}for(let r=a+1;r<n.length;r++)if(t[r]>0||n[r]!==e[r])return!1;return!0}function yF(e,t){let n=e.length>0?e[e.length-1]:1;for(let a=0;a<e.length-1;a++)n+=e[a]*t[a];return n}function U0(e,t,n){let a,r=e.shape.length;typeof t=="number"?a=[t,...new Array(r-1).fill(0)]:t.length<r?a=t.concat(new Array(r-t.length).fill(0)):a=t.slice(),a.forEach(i=>{A(i!==-1,()=>"slice() does not support negative begin indexing.")});let s;return n==null?s=new Array(r).fill(-1):typeof n=="number"?s=[n,...new Array(r-1).fill(-1)]:n.length<r?s=n.concat(new Array(r-n.length).fill(-1)):s=n,s=s.map((i,o)=>i>=0?i:(A(i===-1,()=>`Negative size values should be exactly -1 but got ${i} for the slice() size at index ${o}.`),e.shape[o]-a[o])),[a,s]}function bF(e,t,n,a,r,s,i,o,l){let c=t.slice(),u=n.slice(),p=a;a==null&&(p=new Array(c.length));let d=rh(i);if(d.length>1)throw new Error("Multiple ellipses in slice is not allowed.");if(i!==0&&o!==0)throw new Error("Using both ellipsisMask and newAxisMask is not yet supported.");if(i!==0&&l!==0)throw new Error("Using both ellipsisMask and shrinkAxisMask is not yet supported.");let h=e.length-c.length,m=rh(o),f=e.slice();m.forEach(T=>{c[T]=0,u[T]=1,f.splice(T,0,1)});let{begin:g,end:y,strides:b}=O0(f,d,h,c,u,p,r,s,i);c=g,u=y,p=b;let v=rh(l);v.forEach(T=>{u[T]=c[T]+1,p[T]=1});let x=M0(c,u,p),N=x.filter((T,C)=>v.indexOf(C)===-1);return{nonStrided:p.every(T=>T===1),$begin:c,$end:u,$strides:p,size:x,newShape:f,outShape:N}}var re={};Oe(re,{Serializable:()=>j0,SerializationMap:()=>Ii,registerClass:()=>qr});var j0=class{getClassName(){return this.constructor.className}static fromConfig(e,t){return new e(t)}},Ii=class{constructor(){this.classNameMap={}}static getMap(){return Ii.instance==null&&(Ii.instance=new Ii),Ii.instance}static register(e){Ii.getMap().classNameMap[e.className]=[e,e.fromConfig]}};function qr(e){A(e.className!=null,()=>"Class being registered does not have the static className property defined."),A(typeof e.className=="string",()=>"className is required to be a string, but got type "+typeof e.className),A(e.className.length>0,()=>"Class being registered has an empty-string as its className, which is disallowed."),Ii.register(e)}var q0={};Oe(q0,{TEST_EPSILON_FLOAT16:()=>K0,encodeStrings:()=>X0,expectArrayBuffersEqual:()=>NF,expectArraysClose:()=>xF,expectArraysEqual:()=>wF,expectNumbersClose:()=>kF,expectPromiseToFail:()=>vF,expectValuesInRange:()=>IF,testEpsilon:()=>xy});var TF=.001,K0=.1;function xF(e,t,n){return n==null&&(n=xy()),vy(e,t,(a,r)=>wy(a,r,n))}function xy(){return M.backend.floatPrecision()===32?TF:K0}function vy(e,t,n){let a=!0;if((on(e)||on(t))&&(a=!1),on(e)&&on(t)&&(a=!0),a){let i=e.constructor.name,o=t.constructor.name;if(i!==o)throw new Error(`Arrays are of different type. Actual: ${i}. Expected: ${o}`)}if(Array.isArray(e)&&Array.isArray(t)){let i=Ga(e),o=Ga(t);if(!hr(i,o))throw new Error(`Arrays have different shapes. Actual: [${i}]. Expected: [${o}]`)}let r=on(e)?e:Ts(e),s=on(t)?t:Ts(t);if(r.length!==s.length)throw new Error(`Arrays have different lengths actual: ${r.length} vs expected: ${s.length}.
|
|
Actual: ${r}.
|
|
Expected: ${s}.`);for(let i=0;i<s.length;++i){let o=r[i],l=s[i];if(!n(o,l))throw new Error(`Arrays differ: actual[${i}] = ${o}, expected[${i}] = ${l}.
|
|
Actual: ${r}.
|
|
Expected: ${s}.`)}}function vF(e,t){e().then(()=>t.fail(),()=>t())}function wF(e,t){let n=typeof t=="string"||typeof t=="number"||typeof t=="boolean"?[t]:t;return Pr(e)||Pr(e[0])||Pr(t)||Pr(t[0])?vy(e,n,(a,r)=>a==r):vy(e,t,(a,r)=>wy(a,r,0))}function kF(e,t,n){if(n==null&&(n=xy()),!wy(e,t,n))throw new Error(`Numbers differ: actual === ${e}, expected === ${t}`)}function wy(e,t,n){return!isFinite(e)&&!isFinite(t)?!0:!(isNaN(e)||isNaN(t)||Math.abs(e-t)>n)}function IF(e,t,n){for(let a=0;a<e.length;a++)if(e[a]<t||e[a]>n)throw new Error(`Value out of range:${e[a]} low: ${t}, high: ${n}`)}function NF(e,t){expect(new Float32Array(e)).toEqual(new Float32Array(t))}function X0(e){for(let t=0;t<e.length;t++){let n=e[t];Array.isArray(n)?X0(n):e[t]=Cc(n)}return e}var Y0="3.0.0";function SF(){ee().set("PROD",!0)}function CF(){ee().set("DEBUG",!0)}function _F(){ee().set("DEPRECATION_WARNINGS_ENABLED",!1),console.warn("TensorFlow.js deprecation warnings have been disabled.")}function ky(e){ee().getBool("DEPRECATION_WARNINGS_ENABLED")&&console.warn(e+" You can disable deprecation warnings with tf.disableDeprecationWarnings().")}rA(ky);function EF(){M.disposeVariables()}function Kr(){return M}function sh(){return M.memory()}function AF(e){return M.profile(e)}function D(e,t){return M.tidy(e,t)}function Fe(e){Zg(e).forEach(t=>t.dispose())}function jt(e){return M.keep(e)}function FF(e){return M.time(e)}function $F(e){return M.setBackend(e)}function DF(){return M.ready()}function RF(){return M.backendName}function MF(e){M.removeBackend(e)}function PF(e){return M.findBackend(e)}function OF(e){return M.findBackendFactory(e)}function ih(e,t,n=1){return M.registerBackend(e,t,n)}function Z0(){return M.backend}function LF(e,t){ee().setPlatform(e,t)}function zF(e,t){let n=E(e,"a","add"),a=E(t,"b","add");[n,a]=Nt(n,a);let r={a:n,b:a};return M.runKernel(Lr,r)}var Z=P({add_:zF});function BF(e,t){let n=E(e,"a","floorDiv"),a=E(t,"b","floorDiv");[n,a]=Nt(n,a);let r={a:n,b:a};return M.runKernel(zs,r)}var oh=P({floorDiv_:BF});function WF(e,t){let n=E(e,"a","div"),a=E(t,"b","div");if([n,a]=Nt(n,a),n.dtype==="int32"&&a.dtype==="int32")return oh(n,a);let r={a:n,b:a},s={};return M.runKernel(Ps,r,s)}var ve=P({div_:WF});function VF(e,t){let n=E(e,"a","mul"),a=E(t,"b","mul");[n,a]=Nt(n,a);let r={a:n,b:a};return M.runKernel(Ys,r)}var L=P({mul_:VF});function UF(e){let t=E(e,"x","abs");if(t.dtype==="complex64"){let n={x:t};return M.runKernel(pc,n)}else{let n={x:t};return M.runKernel(So,n)}}var Lt=P({abs_:UF});function GF(e){let t={x:E(e,"x","acos")};return M.runKernel(Co,t)}var Iy=P({acos_:GF});function HF(e){let t={x:E(e,"x","acosh")};return M.runKernel(_o,t)}var Ny=P({acosh_:HF});function jF(e){A(Array.isArray(e),()=>"The argument passed to tf.addN() must be a list of tensors"),A(e.length>=1,()=>`Must pass at least one tensor to tf.addN(), but got ${e.length}`);let t=e.map((r,s)=>E(r,`tensors${s}`,"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(!hr(r.shape,n.shape))throw new Error("All tensors passed to tf.addN() must have the same shape")});let a=t;return M.runKernel(Ss,a)}var J0=P({addN_:jF});function qF(e,t=null,n=!1){let a={x:E(e,"x","all","bool")},r={axis:t,keepDims:n};return M.runKernel(yd,a,r)}var lh=P({all_:qF});function KF(e,t=null,n=!1){let a={x:E(e,"x","any","bool")},r={axis:t,keepDims:n};return M.runKernel(bd,a,r)}var Rc=P({any_:KF});function XF(e,t=0){let n={x:E(e,"x","argMax")},a={axis:t};return M.runKernel(Cs,n,a)}var Mc=P({argMax_:XF});function YF(e,t=0){let n={x:E(e,"x","argMin")},a={axis:t};return M.runKernel(lc,n,a)}var Ty=P({argMin_:YF});function ZF(e){let t={x:E(e,"x","asin")};return M.runKernel(Eo,t)}var Sy=P({asin_:ZF});function JF(e){let t={x:E(e,"x","asinh")};return M.runKernel(Ao,t)}var Cy=P({asinh_:JF});function QF(e){let t={x:E(e,"x","atan")};return M.runKernel(Fo,t)}var _y=P({atan_:QF});function e$(e,t){let n=E(e,"a","atan2"),a=E(t,"b","atan2");[n,a]=Nt(n,a);let r={a:n,b:a};return M.runKernel(Do,r)}var Ey=P({atan2_:e$});function t$(e){let t={x:E(e,"x","atanh")};return M.runKernel($o,t)}var Ay=P({atanh_:t$});function n$(e,t,n,a,r="NHWC",s){let i=e[3],o=[...t,i],l=Q0(r);return Pc(e,o,n,s,a,null,null,l)}function e1(e,t,n,a,r,s,i="channelsLast"){let[o,l]=uh(t),c;if(i==="channelsLast")c=[o,l,e[3],e[3]];else if(i==="channelsFirst")c=[o,l,e[1],e[1]];else throw new Error(`Unknown dataFormat ${i}`);return Pc(e,c,n,a,r,s,!1,i)}function a$(e,t,n,a,r,s,i="NDHWC"){let[o,l,c]=Fy(t),u,p;if(i==="NDHWC")p="channelsLast",u=[o,l,c,e[4],e[4]];else if(i==="NCDHW")p="channelsFirst",u=[o,l,c,e[1],e[1]];else throw new Error(`Unknown dataFormat ${i}`);return t1(e,u,n,a,r,!1,p,s)}function Pc(e,t,n,a,r,s,i=!1,o="channelsLast"){let[l,c,u,p]=[-1,-1,-1,-1];if(o==="channelsLast")[l,c,u,p]=e;else if(o==="channelsFirst")[l,p,c,u]=e;else throw new Error(`Unknown dataFormat ${o}`);let[d,h,,m]=t,[f,g]=uh(n),[y,b]=uh(a),v=Rl(d,y),x=Rl(h,b),{padInfo:N,outHeight:T,outWidth:C}=r$(r,c,u,f,g,v,x,s,o),$=i?m*p:m,F;return o==="channelsFirst"?F=[l,$,T,C]:o==="channelsLast"&&(F=[l,T,C,$]),{batchSize:l,dataFormat:o,inHeight:c,inWidth:u,inChannels:p,outHeight:T,outWidth:C,outChannels:$,padInfo:N,strideHeight:f,strideWidth:g,filterHeight:d,filterWidth:h,effectiveFilterHeight:v,effectiveFilterWidth:x,dilationHeight:y,dilationWidth:b,inShape:e,outShape:F,filterShape:t}}function t1(e,t,n,a,r,s=!1,i="channelsLast",o){let[l,c,u,p,d]=[-1,-1,-1,-1,-1];if(i==="channelsLast")[l,c,u,p,d]=e;else if(i==="channelsFirst")[l,d,c,u,p]=e;else throw new Error(`Unknown dataFormat ${i}`);let[h,m,f,,g]=t,[y,b,v]=Fy(n),[x,N,T]=Fy(a),C=Rl(h,x),$=Rl(m,N),F=Rl(f,T),{padInfo:O,outDepth:B,outHeight:W,outWidth:G}=s$(r,c,u,p,y,b,v,C,$,F,o),q=s?g*d:g,H;return i==="channelsFirst"?H=[l,q,B,W,G]:i==="channelsLast"&&(H=[l,B,W,G,q]),{batchSize:l,dataFormat:i,inDepth:c,inHeight:u,inWidth:p,inChannels:d,outDepth:B,outHeight:W,outWidth:G,outChannels:q,padInfo:O,strideDepth:y,strideHeight:b,strideWidth:v,filterDepth:h,filterHeight:m,filterWidth:f,effectiveFilterDepth:C,effectiveFilterHeight:$,effectiveFilterWidth:F,dilationDepth:x,dilationHeight:N,dilationWidth:T,inShape:e,outShape:H,filterShape:t}}function i$(e,t,n,a,r){a==null&&(a=$y(e,t,n));let s=e[0],i=e[1],o=Ni((s-t+2*a)/n+1,r),l=Ni((i-t+2*a)/n+1,r);return[o,l]}function o$(e,t,n,a,r,s){r==null&&(r=$y(e,t,a));let i=e[0],o=e[1],l=e[2],c=Ni((i-t+2*r)/a+1,s),u=Ni((o-t+2*r)/a+1,s),p=Ni((l-t+2*r)/a+1,s);return[c,u,p,n]}function $y(e,t,n,a=1){let r=Rl(t,a);return Math.floor((e[0]*(n-1)-n+r)/2)}function uh(e){return typeof e=="number"?[e,e,e]:e.length===2?[e[0],e[1],1]:e}function Fy(e){return typeof e=="number"?[e,e,e]:e}function Rl(e,t){return t<=1?e:e+(e-1)*(t-1)}function r$(e,t,n,a,r,s,i,o,l){let c,u,p;if(typeof e=="number"){c={top:e,bottom:e,left:e,right:e,type:e===0?"VALID":"NUMBER"};let d=i$([t,n],s,a,e,o);u=d[0],p=d[1]}else if(e==="same"){u=Math.ceil(t/a),p=Math.ceil(n/r);let d=Math.max(0,(u-1)*a+s-t),h=Math.max(0,(p-1)*r+i-n),m=Math.floor(d/2),f=d-m,g=Math.floor(h/2),y=h-g;c={top:m,bottom:f,left:g,right:y,type:"SAME"}}else if(e==="valid")c={top:0,bottom:0,left:0,right:0,type:"VALID"},u=Math.ceil((t-s+1)/a),p=Math.ceil((n-i+1)/r);else if(typeof e=="object"){let d=l==="channelsLast"?e[1][0]:e[2][0],h=l==="channelsLast"?e[1][1]:e[2][1],m=l==="channelsLast"?e[2][0]:e[3][0],f=l==="channelsLast"?e[2][1]:e[3][1];c={top:d,bottom:h,left:m,right:f,type:d===0&&h===0&&m===0&&f===0?"VALID":"EXPLICIT"},u=Ni((t-s+d+h)/a+1,o),p=Ni((n-i+m+f)/r+1,o)}else throw Error(`Unknown padding parameter: ${e}`);return{padInfo:c,outHeight:u,outWidth:p}}function s$(e,t,n,a,r,s,i,o,l,c,u){let p,d,h,m;if(typeof e=="number"){p={top:e,bottom:e,left:e,right:e,front:e,back:e,type:e===0?"VALID":"NUMBER"};let f=o$([t,n,a,1],o,1,r,e,u);d=f[0],h=f[1],m=f[2]}else if(e==="same"){d=Math.ceil(t/r),h=Math.ceil(n/s),m=Math.ceil(a/i);let f=(d-1)*r+o-t,g=(h-1)*s+l-n,y=(m-1)*i+c-a,b=Math.floor(f/2),v=f-b,x=Math.floor(g/2),N=g-x,T=Math.floor(y/2),C=y-T;p={top:x,bottom:N,left:T,right:C,front:b,back:v,type:"SAME"}}else if(e==="valid")p={top:0,bottom:0,left:0,right:0,front:0,back:0,type:"VALID"},d=Math.ceil((t-o+1)/r),h=Math.ceil((n-l+1)/s),m=Math.ceil((a-c+1)/i);else throw Error(`Unknown padding parameter: ${e}`);return{padInfo:p,outDepth:d,outHeight:h,outWidth:m}}function Ni(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 Xr(e){let[t,n,a]=uh(e);return t===1&&n===1&&a===1}function Ha(e,t){return Xr(e)||Xr(t)}function Q0(e){if(e==="NHWC")return"channelsLast";if(e==="NCHW")return"channelsFirst";throw new Error(`Unknown dataFormat ${e}`)}function l$(e,t){let n={x:E(e,"x","reshape","string_or_numeric")},a={shape:t};return M.runKernel(dl,n,a)}var j=P({reshape_:l$});function u$(e,t,n,a,r){let s=E(e,"x","avgPool","float32"),i=1;A(Ha(n,i),()=>`Error in avgPool: Either strides or dilations must be 1. Got strides ${n} and dilations '${i}'`);let o=s,l=!1;s.rank===3&&(l=!0,o=j(s,[1,s.shape[0],s.shape[1],s.shape[2]])),A(o.rank===4,()=>`Error in avgPool: x must be rank 4 but got rank ${o.rank}.`),r!=null&&A(Gt(a),()=>`Error in avgPool: pad must be an integer when using, dimRoundingMode ${r} but got pad ${a}.`);let c={x:o},u={filterSize:t,strides:n,pad:a,dimRoundingMode:r},p=M.runKernel(_s,c,u);return p=ce(p,s.dtype),l?j(p,[p.shape[1],p.shape[2],p.shape[3]]):p}var Xn=P({avgPool_:u$});function c$(e,t,n,a,r,s="NDHWC"){let i=E(e,"x","avgPool3d","float32"),o=i,l=!1;i.rank===4&&(l=!0,o=j(i,[1,i.shape[0],i.shape[1],i.shape[2],i.shape[3]])),A(o.rank===5,()=>`Error in avgPool3d: x must be rank 5 but got rank ${o.rank}.`),A(s==="NDHWC",()=>`Error in avgPool3d: Only NDHWC is currently supported, but got dataFormat of ${s}`),r!=null&&A(Gt(a),()=>`Error in avgPool3d: pad must be an integer when using, dimRoundingMode ${r} but got pad ${a}.`);let c={x:o},u={filterSize:t,strides:n,pad:a,dimRoundingMode:r,dataFormat:s},p=M.runKernel(uc,c,u);return p=ce(p,o.dtype),l?j(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var Dy=P({avgPool3d_:c$});function p$(e,t=0){A(e.length>=1,()=>"Pass at least one tensor to concat");let n=$c(e,"tensors","concat","string_or_numeric");if(n[0].dtype==="complex64"&&n.forEach(s=>{if(s.dtype!=="complex64")throw new Error(`Cannot concatenate complex64 tensors with a tensor
|
|
with dtype ${s.dtype}. `)}),n.length===1)return jr(n[0]);let a=n,r={axis:t};return M.runKernel(Mo,a,r)}var Qe=P({concat_:p$});function d$(e){let t={x:E(e,"x","sigmoid")};return M.runKernel(li,t)}var ua=P({sigmoid_:d$});function h$(e,t,n){let a=E(e,"x","slice","string_or_numeric");if(a.rank===0)throw new Error("Slicing scalar is not possible");let r={x:a},s={begin:t,size:n};return M.runKernel(gl,r,s)}var We=P({slice_:h$});function m$(e){let t={x:E(e,"x","tanh")};return M.runKernel(mi,t)}var Ml=P({tanh_:m$});function f$(e,t,n,a,r,s){let i=E(e,"forgetBias","basicLSTMCell"),o=E(t,"lstmKernel","basicLSTMCell"),l=E(n,"lstmBias","basicLSTMCell"),c=E(a,"data","basicLSTMCell"),u=E(r,"c","basicLSTMCell"),p=E(s,"h","basicLSTMCell"),d=Qe([c,p],1),h=ze(d,o),m=Z(h,l),f=m.shape[0],g=m.shape[1]/4,y=[f,g],b=We(m,[0,0],y),v=We(m,[0,g],y),x=We(m,[0,g*2],y),N=We(m,[0,g*3],y),T=Z(L(ua(b),Ml(v)),L(u,ua(Z(i,x)))),C=L(Ml(T),ua(N));return[T,C]}var g$=P({basicLSTMCell_:f$});function y$(e,t,n){let a=E(e,"x","batchToSpaceND"),r=t.reduce((o,l)=>o*l);A(a.rank>=1+t.length,()=>`input rank is ${a.rank} but should be > than blockShape.length ${t.length}`),A(n.length===t.length,()=>`crops.length is ${n.length} but should be equal to blockShape.length ${t.length}`),A(a.shape[0]%r==0,()=>`input tensor batch is ${a.shape[0]} but is not divisible by the product of the elements of blockShape ${t.join(" * ")} === ${r}`);let s={x:a},i={blockShape:t,crops:n};return M.runKernel(cc,s,i)}var Oc=P({batchToSpaceND_:y$});function b$(e){let t;return e.rank===0||e.rank===1?t=j(e,[1,1,1,e.size]):e.rank===2?t=j(e,[1,1,e.shape[0],e.shape[1]]):e.rank===3?t=j(e,[1,e.shape[0],e.shape[1],e.shape[2]]):t=e,t}function x$(e,t,n,a,r,s){s==null&&(s=.001);let i=E(e,"x","batchNorm"),o=E(t,"mean","batchNorm"),l=E(n,"variance","batchNorm"),c;r!=null&&(c=E(r,"scale","batchNorm"));let u;a!=null&&(u=E(a,"offset","batchNorm")),A(o.rank===l.rank,()=>"Batch normalization gradient requires mean and variance to have equal ranks."),A(u==null||o.rank===u.rank,()=>"Batch normalization gradient requires mean and offset to have equal ranks."),A(c==null||o.rank===c.rank,()=>"Batch normalization gradient requires mean and scale to have equal ranks.");let p={x:b$(i),scale:c,offset:u,mean:o,variance:l},d={varianceEpsilon:s},h=M.runKernel(Bs,p,d);return j(h,i.shape)}var fr=P({batchNorm_:x$});function v$(e,t,n,a,r,s){let i=E(e,"x","batchNorm"),o=E(t,"mean","batchNorm"),l=E(n,"variance","batchNorm"),c;r!=null&&(c=E(r,"scale","batchNorm"));let u;return a!=null&&(u=E(a,"offset","batchNorm")),A(i.rank===2,()=>`Error in batchNorm2D: x must be rank 2 but got rank ${i.rank}.`),A(o.rank===2||o.rank===1,()=>`Error in batchNorm2D: mean must be rank 2 or rank 1 but got rank ${o.rank}.`),A(l.rank===2||l.rank===1,()=>`Error in batchNorm2D: variance must be rank 2 or rank 1 but got rank ${l.rank}.`),c!=null&&A(c.rank===2||c.rank===1,()=>`Error in batchNorm2D: scale must be rank 2 or rank 1 but got rank ${c.rank}.`),u!=null&&A(u.rank===2||u.rank===1,()=>`Error in batchNorm2D: offset must be rank 2 or rank 1 but got rank ${u.rank}.`),fr(i,o,l,u,c,s)}var n1=P({batchNorm2d_:v$});function w$(e,t,n,a,r,s){let i=E(e,"x","batchNorm"),o=E(t,"mean","batchNorm"),l=E(n,"variance","batchNorm"),c;r!=null&&(c=E(r,"scale","batchNorm"));let u;return a!=null&&(u=E(a,"offset","batchNorm")),A(i.rank===3,()=>`Error in batchNorm3D: x must be rank 3 but got rank ${i.rank}.`),A(o.rank===3||o.rank===1,()=>`Error in batchNorm3D: mean must be rank 3 or rank 1 but got rank ${o.rank}.`),A(l.rank===3||l.rank===1,()=>`Error in batchNorm3D: variance must be rank 3 or rank 1 but got rank ${l.rank}.`),c!=null&&A(c.rank===3||c.rank===1,()=>`Error in batchNorm3D: scale must be rank 3 or rank 1 but got rank ${c.rank}.`),u!=null&&A(u.rank===3||u.rank===1,()=>`Error in batchNorm3D: offset must be rank 3 or rank 1 but got rank ${u.rank}.`),fr(i,o,l,u,c,s)}var a1=P({batchNorm3d_:w$});function k$(e,t,n,a,r,s){let i=E(e,"x","batchNorm"),o=E(t,"mean","batchNorm"),l=E(n,"variance","batchNorm"),c;r!=null&&(c=E(r,"scale","batchNorm"));let u;return a!=null&&(u=E(a,"offset","batchNorm")),A(i.rank===4,()=>`Error in batchNorm4D: x must be rank 4 but got rank ${i.rank}.`),A(o.rank===4||o.rank===1,()=>`Error in batchNorm4D: mean must be rank 4 or rank 1 but got rank ${o.rank}.`),A(l.rank===4||l.rank===1,()=>`Error in batchNorm4D: variance must be rank 4 or rank 1 but got rank ${l.rank}.`),c!=null&&A(c.rank===4||c.rank===1,()=>`Error in batchNorm4D: scale must be rank 4 or rank 1 but got rank ${c.rank}.`),u!=null&&A(u.rank===4||u.rank===1,()=>`Error in batchNorm4D: offset must be rank 4 or rank 1 but got rank ${u.rank}.`),fr(i,o,l,u,c,s)}var r1=P({batchNorm4d_:k$});function I$(e,t,n){let a=E(e,"x","bincount"),r=E(t,"weights","bincount");A(a.dtype==="int32",()=>`Error in bincount: input dtype must be int32, but got ${a.dtype}`),A(n>=0,()=>`size must be non-negative, but got ${n}.`),A(r.size===a.size||r.size===0,()=>`Error in bincount: weights must have the same size as input or0-length, but got input shape: ${a.shape}, weights shape: ${r.shape}.`);let s={x:a,weights:r},i={size:n};return M.runKernel(wd,s,i)}var s1=P({bincount_:I$});function N$(e,t){let n=E(e,"broadcastTo","x"),a=n.shape;if(t.some(l=>!(l>0)||l%1!=0))throw new Error(`broadcastTo(): Invalid broadcast shape [${t}].`);if(t.length<n.rank)throw new Error(`broadcastTo(): shape.length=${t.length} < input.rank=${n.rank}.`);if(t.length>n.rank){let l=n.shape.slice();for(;l.length<t.length;)l.unshift(1);n=j(n,l)}let r=n.shape,s=Array.from(t);for(let l=t.length-1;l>=0;l--)if(r[l]===t[l])s[l]=1;else if(n.shape[l]!==1)throw new Error(`broadcastTo(): [${a}] cannot be broadcast to [${t}].`);if(s.map((l,c)=>l>1?c:-1).filter(l=>l>=0).length===0)return jr(n);let i={x:n},o={reps:s};return M.runKernel(Br,i,o)}var Lc=P({broadcastTo_:N$});function T$(e){let t={x:E(e,"x","ceil")};return M.runKernel(Ro,t)}var Ry=P({ceil_:T$});function S$(e,t,n){let a=E(e,"x","clipByValue");A(t<=n,()=>`Error in clip: min (${t}) must be less than or equal to max (${n}).`);let r={x:a},s={clipValueMin:t,clipValueMax:n};return M.runKernel(zr,r,s)}var qt=P({clipByValue_:S$});function C$(e){return Qe(e,0)}var i1=P({concat1d_:C$});function _$(e,t){return Qe(e,t)}var o1=P({concat2d_:_$});function E$(e,t){return Qe(e,t)}var l1=P({concat3d_:E$});function A$(e,t){return Qe(e,t)}var u1=P({concat4d_:A$});function F$(e,t,n,a,r="NHWC",s=[1,1],i){let o=E(e,"x","conv2d"),l=E(t,"filter","conv2d"),c=o,u=!1;o.rank===3&&(u=!0,c=j(o,[1,o.shape[0],o.shape[1],o.shape[2]])),A(c.rank===4,()=>`Error in conv2d: input must be rank 4, but got rank ${c.rank}.`),A(l.rank===4,()=>`Error in conv2d: filter must be rank 4, but got rank ${l.rank}.`),i!=null&&A(Gt(a),()=>`Error in conv2d: pad must be an integer when using, dimRoundingMode ${i} but got pad ${a}.`);let p=r==="NHWC"?c.shape[3]:c.shape[1];A(p===l.shape[2],()=>`Error in conv2d: depth of input (${p}) must match input depth for filter ${l.shape[2]}.`),A(Ha(n,s),()=>`Error in conv2D: Either strides or dilations must be 1. Got strides ${n} and dilations '${s}'`);let d={x:c,filter:l},h={strides:n,pad:a,dataFormat:r,dilations:s,dimRoundingMode:i},m=M.runKernel(Fs,d,h);return u?j(m,[m.shape[1],m.shape[2],m.shape[3]]):m}var At=P({conv2d_:F$});function $$(e,t,n,a,r="NWC",s=1,i){let o=E(e,"x","conv1d"),l=E(t,"filter","conv1d"),c=o,u=!1;o.rank===2&&(u=!0,c=j(o,[1,o.shape[0],o.shape[1]])),A(c.rank===3,()=>`Error in conv1d: input must be rank 3, but got rank ${c.rank}.`),A(l.rank===3,()=>`Error in conv1d: filter must be rank 3, but got rank ${l.rank}.`),i!=null&&A(Gt(a),()=>`Error in conv1d: pad must be an integer when using, dimRoundingMode ${i} but got pad ${a}.`),A(c.shape[2]===l.shape[1],()=>`Error in conv1d: depth of input (${c.shape[2]}) must match input depth for filter ${l.shape[1]}.`),A(Ha(n,s),()=>`Error in conv1D: Either stride or dilation must be 1. Got stride ${n} and dilation '${s}'`),A(r==="NWC",()=>`Error in conv1d: got dataFormat of ${r} but only NWC is currently supported.`);let p=j(l,[1,l.shape[0],l.shape[1],l.shape[2]]),d=j(c,[c.shape[0],1,c.shape[1],c.shape[2]]),h=At(d,p,[1,n],a,"NHWC",[1,s],i);return u?j(h,[h.shape[2],h.shape[3]]):j(h,[h.shape[0],h.shape[2],h.shape[3]])}var ch=P({conv1d_:$$});function D$(e,t,n,a,r,s="NHWC",i){A(e.length===t.rank,()=>`Length of inShape (${e.length}) and rank of dy (${t.rank}) must match`);let o=e,l=t,c=!1;t.rank===3&&(c=!0,l=j(t,[1,t.shape[0],t.shape[1],t.shape[2]]),o=[1,e[0],e[1],e[2]]),A(o.length===4,()=>`Error in conv2dDerInput: inShape must be length 4, but got length ${o.length}.`),A(l.rank===4,()=>`Error in conv2dDerInput: dy must be rank 4, but got rank ${l.rank}`),A(n.rank===4,()=>`Error in conv2dDerInput: filter must be rank 4, but got rank ${n.rank}`);let u=s==="NHWC"?o[3]:o[1],p=s==="NHWC"?l.shape[3]:l.shape[1];A(u===n.shape[2],()=>`Error in conv2dDerInput: depth of input (${u}) must match input depth for filter ${n.shape[2]}.`),A(p===n.shape[3],()=>`Error in conv2dDerInput: depth of output (${p}) must match output depth for filter ${n.shape[3]}.`),i!=null&&A(Gt(r),()=>`Error in conv2dDerInput: pad must be an integer when using, dimRoundingMode ${i} but got pad ${r}.`);let d={dy:l,filter:n},h={strides:a,pad:r,dataFormat:s,dimRoundingMode:i,inputShape:o},m=M.runKernel($s,d,h);return c?j(m,[m.shape[1],m.shape[2],m.shape[3]]):m}var My=P({conv2DBackpropInput_:D$});function R$(e,t,n,a,r,s){let i=E(e,"x","conv2dTranspose"),o=E(t,"filter","conv2dTranspose");return My(n,i,o,a,r,"NHWC",s)}var ph=P({conv2dTranspose_:R$});function M$(e,t,n,a,r="NDHWC",s=[1,1,1]){let i=E(e,"x","conv3d"),o=E(t,"filter","conv3d"),l=i,c=!1;i.rank===4&&(c=!0,l=j(i,[1,i.shape[0],i.shape[1],i.shape[2],i.shape[3]])),A(l.rank===5,()=>`Error in conv3d: input must be rank 5, but got rank ${l.rank}.`),A(o.rank===5,()=>`Error in conv3d: filter must be rank 5, but got rank ${o.rank}.`),A(l.shape[4]===o.shape[3],()=>`Error in conv3d: depth of input (${l.shape[4]}) must match input depth for filter ${o.shape[3]}.`),A(Ha(n,s),()=>`Error in conv3D: Either strides or dilations must be 1. Got strides ${n} and dilations '${s}'`),A(r==="NDHWC",()=>`Error in conv3d: got dataFormat of ${r} but only NDHWC is currently supported.`);let u={x:l,filter:o},p={strides:n,pad:a,dataFormat:r,dilations:s},d=M.runKernel(dc,u,p);return c?j(d,[d.shape[1],d.shape[2],d.shape[3],d.shape[4]]):d}var Py=P({conv3d_:M$});function P$(e,t,n,a,r){A(e.length===t.rank,()=>`Length of inShape (${e.length}) and rank of dy (${t.rank}) must match`);let s=e,i=t,o=!1;t.rank===4&&(o=!0,i=j(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]]),s=[1,e[0],e[1],e[2],e[3]]);let l=s[4],c=i.shape[4];A(s.length===5,()=>`Error in conv3dDerInput: inShape must be length 5, but got length ${s.length}.`),A(i.rank===5,()=>`Error in conv3dDerInput: dy must be rank 5, but got rank ${i.rank}`),A(n.rank===5,()=>`Error in conv3dDerInput: filter must be rank 5, but got rank ${n.rank}`),A(l===n.shape[3],()=>`Error in conv3dDerInput: depth of input (${l}) must match input depth for filter ${n.shape[3]}.`),A(c===n.shape[4],()=>`Error in conv3dDerInput: depth of output (${c}) must match output depth for filter ${n.shape[4]}.`);let u={dy:i,filter:n},p={pad:r,strides:a,inputShape:s},d=M.runKernel(Td,u,p);return o?j(d,[d.shape[1],d.shape[2],d.shape[3],d.shape[4]]):d}var c1=P({conv3DBackpropInput_:P$});function O$(e,t,n,a,r){let s=E(e,"x","conv3dTranspose"),i=E(t,"filter","conv3dTranspose");return c1(n,s,i,a,r)}var L$=P({conv3dTranspose_:O$});function z$(e){let t={x:E(e,"x","cos")};return M.runKernel(Ds,t)}var zc=P({cos_:z$});function B$(e){let t={x:E(e,"x","cosh")};return M.runKernel(Po,t)}var dh=P({cosh_:B$});function W$(e,t=0,n=!1,a=!1){let r={x:E(e,"x","cumsum")},s={axis:t,exclusive:n,reverse:a};return M.runKernel(Rs,r,s)}var hh=P({cumsum_:W$});function V$(e,t,n,a=!1){let r=E(e,"x","denseBincount"),s=E(t,"weights","denseBincount");A(r.dtype==="int32",()=>`Error in denseBincount: input dtype must be int32, but got ${r.dtype}`),A(r.rank<=2,()=>`Error in denseBincount: input must be at most rank 2, but got rank ${r.rank}.`),A(n>=0,()=>`size must be non-negative, but got ${n}.`),A(s.size===r.size||s.size===0,()=>`Error in denseBincount: weights must have the same shape as x or 0-length, but got x shape: ${r.shape}, weights shape: ${s.shape}.`);let i={x:r,weights:s},o={size:n,binaryOutput:a};return M.runKernel(Sd,i,o)}var p1=P({denseBincount_:V$});function U$(e,t,n="NHWC"){let a=E(e,"x","depthToSpace"),r=n==="NHWC"?a.shape[1]:a.shape[2],s=n==="NHWC"?a.shape[2]:a.shape[3],i=n==="NHWC"?a.shape[3]:a.shape[1];A(r*t>=0,()=>`Negative dimension size caused by overflow when multiplying
|
|
${r} and ${t} for depthToSpace with input shape
|
|
${a.shape}`),A(s*t>=0,()=>`Negative dimension size caused by overflow when multiplying
|
|
${s} and ${t} for depthToSpace with input shape
|
|
${a.shape}`),A(i%(t*t)==0,()=>`Dimension size must be evenly divisible by ${t*t} but is ${i} for depthToSpace with input shape ${a.shape}`);let o={x:a},l={blockSize:t,dataFormat:n};return M.runKernel(Lo,o,l)}var Oy=P({depthToSpace_:U$});function G$(e,t,n,a,r="NHWC",s=[1,1],i){let o=E(e,"x","depthwiseConv2d"),l=E(t,"filter","depthwiseConv2d"),c=o,u=!1;o.rank===3&&(u=!0,c=j(o,[1,o.shape[0],o.shape[1],o.shape[2]])),A(c.rank===4,()=>`Error in depthwiseConv2d: input must be rank 4, but got rank ${c.rank}.`),A(l.rank===4,()=>`Error in depthwiseConv2d: filter must be rank 4, but got rank ${l.rank}.`),A(c.shape[3]===l.shape[2],()=>`Error in depthwiseConv2d: number of input channels (${c.shape[3]}) must match the inChannels dimension in filter ${l.shape[2]}.`),i!=null&&A(Gt(a),()=>`Error in depthwiseConv2d: pad must be an integer when using, dimRoundingMode ${i} but got pad ${a}.`);let p={x:c,filter:l},d={strides:n,pad:a,dataFormat:r,dilations:s,dimRoundingMode:i},h=M.runKernel(Ms,p,d);return u?j(h,[h.shape[1],h.shape[2],h.shape[3]]):h}var Yr=P({depthwiseConv2d_:G$});function H$(e){let t={x:E(e,"x","diag")};return M.runKernel(Ed,t)}var j$=P({diag_:H$});function q$(e,t,n,a,r=[1,1],s="NHWC"){let i=E(e,"x","dilation2d"),o=E(t,"filter","dilation2d");A(i.rank===3||i.rank===4,()=>`Error in dilation2d: input must be rank 3 or 4, but got rank ${i.rank}.`),A(o.rank===3,()=>`Error in dilation2d: filter must be rank 3, but got rank ${o.rank}.`),A(s==="NHWC",()=>`Error in dilation2d: Only NHWC is currently supported, but got dataFormat of ${s}`);let l=i,c=!1;i.rank===3&&(l=j(i,[1,i.shape[0],i.shape[1],i.shape[2]]),c=!0);let u={x:l,filter:o},p={strides:n,pad:a,dilations:r},d=M.runKernel(hc,u,p);return c?j(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var Ly=P({dilation2d_:q$});function K$(e,t){let n=e.length,a=[];for(let r=0;r<n;r++){let s=n-1-r,i=e[s]||1;(t[t.length-1-r]||1)>1&&i===1&&a.unshift(s)}return a}function zt(e,t){let n=[];for(let a=0;a<t.length;a++){let r=e[e.length-a-1],s=t.length-a-1,i=t[s];(r==null||r===1&&i>1)&&n.unshift(s)}return n}function xt(e,t){let n=[],a=Math.max(e.length,t.length);for(let r=0;r<a;r++){let s=e[e.length-r-1];s==null&&(s=1);let i=t[t.length-r-1];if(i==null&&(i=1),s===1)n.unshift(i);else if(i===1)n.unshift(s);else if(s!==i){let o=`Operands could not be broadcast together with shapes ${e} and ${t}.`;throw Error(o)}else n.unshift(s)}return n}function X$(e,t){let n=E(e,"a","equal"),a=E(t,"b","equal");[n,a]=Nt(n,a),xt(n.shape,a.shape);let r={a:n,b:a};return M.runKernel(Wo,r)}var Zr=P({equal_:X$});function Y$(e,t,n){let a=E(t,"a","where"),r=E(n,"b","where"),s=E(e,"condition","where","bool"),i=xt(a.shape,r.shape),o=Lc(a,i),l=Lc(r,i);s.rank===1&&A(s.shape[0]===a.shape[0],()=>"The first dimension of `a` must match the size of `condition`."),s.rank!==1&&sn(s.shape,l.shape,"Error in where: ");let c={condition:s,t:o,e:l};return M.runKernel(ml,c)}var vn=P({where_:Y$});function Z$(e){let t={x:E(e,"x","zerosLike")};return M.runKernel(Tl,t)}var Ge=P({zerosLike_:Z$});function J$(e,t){let n=E(e,"a","div"),a=E(t,"b","div");[n,a]=Nt(n,a);let r=ve(n,a),s=Ge(r),i=Zr(a,s);return vn(i,s,r)}var zy=P({divNoNan_:J$});function Q$(e,t){let n=E(e,"t1","dot"),a=E(t,"t2","dot");A((n.rank===1||n.rank===2)&&(a.rank===1||a.rank===2),()=>`Error in dot: inputs must all be rank 1 or 2, but got ranks ${n.rank} and ${a.rank}.`);let r=n.rank===1?n.size:n.shape[1],s=a.rank===1?a.size:a.shape[0];if(A(r===s,()=>`Error in dot: inner dimensions of inputs must match, but got ${r} and ${s}.`),n.rank===1&&a.rank===1){let i=j(n,[1,-1]),o=j(a,[-1,1]),l=ze(i,o);return j(l,[])}else if(n.rank===1&&a.rank===2){let i=j(n,[1,-1]),o=j(a,[a.shape[0],a.shape[1]]),l=ze(i,o);return j(l,[l.size])}else if(n.rank===2&&a.rank===1){let i=j(a,[-1,1]),o=ze(n,i);return j(o,[o.size])}else{let i=j(a,[a.shape[0],a.shape[1]]);return ze(n,i)}}var d1=P({dot_:Q$});function eD(e){let t={x:E(e,"x","elu")};return M.runKernel(zo,t)}var Pl=P({elu_:eD});function tD(e){let t=E(e,"x","erf");A(t.dtype==="int32"||t.dtype==="float32",()=>"Input dtype must be `int32` or `float32`."),t.dtype==="int32"&&(t=ce(t,"float32"));let n={x:t};return M.runKernel(Bo,n)}var By=P({erf_:tD});function nD(e){let t={x:E(e,"x","exp")};return M.runKernel(Os,t)}var pn=P({exp_:nD});function aD(e,t=0){let n=E(e,"x","expandDims","string_or_numeric");A(t<=n.rank,()=>"Axis must be <= rank of the tensor");let a={input:n},r={dim:t};return M.runKernel(Vo,a,r)}var Fn=P({expandDims_:aD});function rD(e){let t={x:E(e,"x","expm1")};return M.runKernel(Uo,t)}var Wy=P({expm1_:rD});function sD(e,t){let n=E(e,"x","tile","string_or_numeric");A(n.rank===t.length,()=>`Error in transpose: rank of input ${n.rank} must match length of reps ${t}.`);let a={x:n},r={reps:t};return M.runKernel(Br,a,r)}var ja=P({tile_:sD});function iD(e,t,n,a="float32"){t==null&&(t=e);let r=Le([e,t],a),s=e<=t?e:t;for(let o=0;o<s;++o)r.set(1,o,o);let i=j(r.toTensor(),[e,t]);if(n==null)return i;if(n.length===1)return ja(Fn(i,0),[n[0],1,1]);if(n.length===2)return ja(Fn(Fn(i,0),0),[n[0],n[1],1,1]);if(n.length===3)return ja(Fn(Fn(Fn(i,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 Vy=P({eye_:iD});function wn(e,t,n){let a={shape:e,value:t,dtype:n};return M.runKernel(mc,{},a)}function oD(e){let t={x:E(e,"x","floor")};return M.runKernel(Ls,t)}var Ol=P({floor_:oD});function lD(e,t,n=0,a=0){let r=E(e,"x","gather"),s=E(t,"indices","gather","int32"),i={x:r,indices:s},o={axis:n,batchDims:a};return M.runKernel(Ho,i,o)}var Ti=P({gather_:lD});function uD(e,t){let n=E(e,"a","greater"),a=E(t,"b","greater");[n,a]=Nt(n,a),xt(n.shape,a.shape);let r={a:n,b:a};return M.runKernel(qo,r)}var ca=P({greater_:uD});function cD(e,t){let n=E(e,"a","greaterEqual"),a=E(t,"b","greaterEqual");[n,a]=Nt(n,a),xt(n.shape,a.shape);let r={a:n,b:a};return M.runKernel(Ws,r)}var Jr=P({greaterEqual_:cD});function pD(e){let t={input:E(e,"input","imag")};return M.runKernel(Md,t)}var mh=P({imag_:pD});function dD(e){let t={x:E(e,"x","isFinite")};return M.runKernel(Xo,t)}var h1=P({isFinite_:dD});function hD(e){let t={x:E(e,"x","isInf")};return M.runKernel(Yo,t)}var m1=P({isInf_:hD});function mD(e){let t={x:E(e,"x","isNaN")};return M.runKernel(Zo,t)}var f1=P({isNaN_:mD});function fD(e,t=.2){let n={x:E(e,"x","leakyRelu")},a={alpha:t};return M.runKernel(Vs,n,a)}var Bc=P({leakyRelu_:fD});function gD(e,t){let n=E(e,"a","less"),a=E(t,"b","less");[n,a]=Nt(n,a),xt(n.shape,a.shape);let r={a:n,b:a};return M.runKernel(Jo,r)}var fh=P({less_:gD});function yD(e,t){let n=E(e,"a","lessEqual"),a=E(t,"b","lessEqual");[n,a]=Nt(n,a),xt(n.shape,a.shape);let r={a:n,b:a};return M.runKernel(Qo,r)}var Si=P({lessEqual_:yD});function g1(e,t,n){if(n<=0)throw new Error("The number of values should be positive.");let a={start:e,stop:t,num:n};return M.runKernel(Pd,{},a)}function bD(e,t=5,n=1,a=1,r=.5){let s=E(e,"x","localResponseNormalization");A(s.rank===4||s.rank===3,()=>`Error in localResponseNormalization: x must be rank 3 or 4 but got
|
|
rank ${s.rank}.`),A(Gt(t),()=>`Error in localResponseNormalization: depthRadius must be an integer but got depthRadius ${t}.`);let i=s,o=!1;s.rank===3&&(o=!0,i=j(s,[1,s.shape[0],s.shape[1],s.shape[2]]));let l={x:i},c={depthRadius:t,bias:n,alpha:a,beta:r},u=M.runKernel(yc,l,c);return o?j(u,[u.shape[1],u.shape[2],u.shape[3]]):u}var Uy=P({localResponseNormalization_:bD});function xD(e){let t={x:E(e,"x","log")};return M.runKernel(Us,t)}var $n=P({log_:xD});function vD(e){let t={x:E(e,"x","log1p")};return M.runKernel(el,t)}var gh=P({log1p_:vD});function wD(e){return A(Or(e),()=>"The f passed in grad(f) must be a function"),(t,n)=>{let a=E(t,"x","tf.grad","string_or_numeric"),r=n!=null?E(n,"dy","tf.grad"):null;return M.tidy(()=>{let{value:s,grads:i}=M.gradients(()=>e(a),[a],r);return r!=null&&sn(s.shape,r.shape,"The shape of dy passed in grad(f)(x, dy) must match the shape returned by f(x)"),yh(i),i[0]})}}function kD(e){return A(Or(e),()=>"The f passed in grads(f) must be a function"),(t,n)=>{A(Array.isArray(t),()=>"The args passed in grads(f)(args) must be an array of `Tensor`s or `TensorLike`s");let a=$c(t,"args","tf.grads","string_or_numeric"),r=n!=null?E(n,"dy","tf.grads"):null;return M.tidy(()=>{let{value:s,grads:i}=M.gradients(()=>e(...a),a,r);return r!=null&&sn(s.shape,r.shape,"The shape of dy passed in grads(f)([x1,...], dy) must match the shape returned by f([x1,...])"),yh(i),i})}}function ID(e){return A(Or(e),()=>"The f passed in valueAndGrad(f) must be a function"),(t,n)=>{A(t instanceof Ae,()=>"The x passed in valueAndGrad(f)(x) must be a tensor"),A(n==null||n instanceof Ae,()=>"The dy passed in valueAndGrad(f)(x, dy) must be a tensor");let{grads:a,value:r}=M.gradients(()=>e(t),[t],n);return yh(a),{grad:a[0],value:r}}}function ND(e){return A(Or(e),()=>"The f passed in valueAndGrads(f) must be a function"),(t,n)=>{A(Array.isArray(t)&&t.every(r=>r instanceof Ae),()=>"The args passed in valueAndGrads(f)(args) must be array of tensors"),A(n==null||n instanceof Ae,()=>"The dy passed in valueAndGrads(f)(args, dy) must be a tensor");let a=M.gradients(()=>e(...t),t,n);return n!=null&&sn(a.value.shape,n.shape,"The shape of dy passed in valueAndGrads(f)([x1,...], dy) must match the shape returned by f([x1,...])"),yh(a.grads),a}}function y1(e,t){A(Or(e),()=>"The f passed in variableGrads(f) must be a function"),A(t==null||Array.isArray(t)&&t.every(c=>c instanceof Vr),()=>"The varList passed in variableGrads(f, varList) must be an array of variables");let n=t!=null;if(!n){t=[];for(let c in M.registeredVariables)t.push(M.registeredVariables[c])}let a=n?t.filter(c=>!c.trainable):null,r=t.length;t=t.filter(c=>c.trainable),A(t.length>0,()=>`variableGrads() expects at least one of the input variables to be trainable, but none of the ${r} variables is trainable.`);let s=!0,{value:i,grads:o}=M.gradients(e,t,null,s);A(o.some(c=>c!=null),()=>"Cannot find a connection between any variable and the result of the loss function y=f(x). Please make sure the operations that use variables are inside the function f passed to minimize()."),A(i.rank===0,()=>`The f passed in variableGrads(f) must return a scalar, but it returned a rank-${i.rank} tensor`);let l={};return t.forEach((c,u)=>{o[u]!=null&&(l[c.name]=o[u])}),a!=null&&a.forEach(c=>l[c.name]=null),{value:i,grads:l}}function qa(e){return M.customGrad(e)}function yh(e){if(e.filter(t=>t==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 TD(e){let t={x:E(e,"x","neg")};return M.runKernel(al,t)}var Tt=P({neg_:TD});function SD(e){let t={x:E(e,"x","softplus")};return M.runKernel(xl,t)}var Ll=P({softplus_:SD});function CD(e){let t=E(e,"x","logSigmoid");return qa(n=>({value:Tt(Ll(Tt(n))),gradFunc:a=>L(a,ua(Tt(n)))}))(t)}var b1=P({logSigmoid_:CD});function _D(e,t=null,n=!1){let a={x:E(e,"x","max")},r={reductionIndices:t,keepDims:n};return M.runKernel(Gs,a,r)}var Yn=P({max_:_D});function ED(e,t){let n=E(e,"a","sub"),a=E(t,"b","sub");[n,a]=Nt(n,a);let r={a:n,b:a};return M.runKernel(hi,r)}var fe=P({sub_:ED});function AD(e,t=null,n=!1){let a=E(e,"x","sum");a.dtype==="bool"&&(a=ce(a,"int32"));let r={x:a},s={axis:t,keepDims:n};return M.runKernel(ci,r,s)}var Te=P({sum_:AD});function FD(e,t=-1){let n=E(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 ${t}`);return qa((a,r)=>{let s=!0,i=Yn(a,t,!0),o=fe(a,i),l=fe(ce(o,"float32"),$n(Te(pn(o),t,s)));return r([l]),{value:l,gradFunc:(c,u)=>{let[p]=u,d=!0,h=pn(p);return fe(c,L(Te(c,t,d),h))}}})(n)}var bh=P({logSoftmax_:FD});function Gy(e,t){for(let n=0;n<e.length;++n)if(e[e.length-n-1]!==t-1-n)return!1;return!0}function x1(e,t,n){let a=e.length+t.length,r=[],s=0,i=0;for(let o=0;o<a;o++)n.indexOf(o)===-1?r.push(e[s++]):r.push(t[i++]);return r}function v1(e,t){let n=[],a=e.length;for(let s=0;s<a;s++)t.indexOf(s)===-1&&n.push(e[s]);let r=t.map(s=>e[s]);return[n,r]}function Ci(e,t){let n=t.map(a=>1);return x1(e,n,t)}function $D(e,t,n){A(Gy(t,n),()=>`${e} supports only inner-most axes for now. Got axes ${t} and rank-${n} input.`)}function w1(e,t){if(Gy(e,t))return null;let n=[];for(let a=0;a<t;++a)e.indexOf(a)===-1&&n.push(a);return e.forEach(a=>n.push(a)),n}function Hy(e){return e.map((t,n)=>[n,t]).sort((t,n)=>t[1]-n[1]).map(t=>t[0])}function DD(e,t){let n=[];for(let a=t-e;a<t;++a)n.push(a);return n}function RD(e,t=null,n=!1){let a=E(e,"x","logSumExp"),r=oa(t,a.shape),s=Yn(a,r,!0),i=fe(a,s),o=pn(i),l=Te(o,r),c=$n(l),u=Z(j(s,c.shape),c);if(n){let p=Ci(u.shape,r);return j(u,p)}return u}var jy=P({logSumExp_:RD});function MD(e,t){let n=E(e,"a","logicalAnd","bool"),a=E(t,"b","logicalAnd","bool");xt(n.shape,a.shape);let r={a:n,b:a};return M.runKernel(tl,r)}var pa=P({logicalAnd_:MD});function PD(e){let t={x:E(e,"x","logicalNot","bool")};return M.runKernel(fc,t)}var Wc=P({logicalNot_:PD});function OD(e,t){let n=E(e,"a","logicalOr","bool"),a=E(t,"b","logicalOr","bool");xt(n.shape,a.shape);let r={a:n,b:a};return M.runKernel(gc,r)}var xh=P({logicalOr_:OD});function LD(e,t){let n=E(e,"a","logicalXor","bool"),a=E(t,"b","logicalXor","bool");return xt(n.shape,a.shape),pa(xh(e,t),Wc(pa(e,t)))}var k1=P({logicalXor_:LD});function zD(e,t,n,a,r){let s=E(e,"x","maxPool"),i=1,o=s,l=!1;s.rank===3&&(l=!0,o=j(s,[1,s.shape[0],s.shape[1],s.shape[2]])),A(o.rank===4,()=>`Error in maxPool: input must be rank 4 but got rank ${o.rank}.`),A(Ha(n,i),()=>`Error in maxPool: Either strides or dilations must be 1. Got strides ${n} and dilations '${i}'`),r!=null&&A(Gt(a),()=>`Error in maxPool: pad must be an integer when using, dimRoundingMode ${r} but got pad ${a}.`);let c={x:o},u={filterSize:t,strides:n,pad:a,dimRoundingMode:r},p=M.runKernel(js,c,u);return l?j(p,[p.shape[1],p.shape[2],p.shape[3]]):p}var Ft=P({maxPool_:zD});function BD(e,t=[1,1,1],n,a,r,s="NDHWC"){let i=E(e,"x","maxPool3d"),o=i,l=!1;i.rank===4&&(l=!0,o=j(i,[1,i.shape[0],i.shape[1],i.shape[2],i.shape[3]])),A(o.rank===5,()=>`Error in maxPool3d: x must be rank 5 but got rank ${o.rank}.`),A(s==="NDHWC",()=>`Error in maxPool3d: Only NDHWC is currently supported, but got dataFormat of ${s}`),r!=null&&A(Gt(a),()=>`Error in maxPool3d: pad must be an integer when using, dimRoundingMode ${r} but got pad ${a}.`);let c={x:o},u={filterSize:t,strides:n,pad:a,dimRoundingMode:r,dataFormat:s},p=M.runKernel(bc,c,u);return l?j(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var qy=P({maxPool3d_:BD});function WD(e,t,n,a,r=!1){let s={x:E(e,"x","maxPoolWithArgmax")},i={filterSize:t,strides:n,pad:a,includeBatchInIndex:r},o=M.runKernel(Bd,s,i);return{result:o[0],indexes:o[1]}}var I1=P({maxPoolWithArgmax_:WD});function VD(e,t){let n=E(e,"a","maximum"),a=E(t,"b","maximum");[n,a]=Nt(n,a),n.dtype==="bool"&&(n=ce(n,"int32"),a=ce(a,"int32")),xt(n.shape,a.shape);let r={a:n,b:a};return M.runKernel(Hs,r)}var Ka=P({maximum_:VD});function UD(e,t=null,n=!1){let a={x:E(e,"x","mean")},r={axis:t,keepDims:n};return M.runKernel(qs,a,r)}var St=P({mean_:UD});function GD(e,t=null,n=!1){let a={x:E(e,"x","min")},r={axis:t,keepDims:n};return M.runKernel(Ks,a,r)}var zl=P({min_:GD});function HD(e,t){let n=E(e,"a","minimum"),a=E(t,"b","minimum");[n,a]=Nt(n,a),n.dtype==="bool"&&(n=ce(n,"int32"),a=ce(a,"int32")),xt(n.shape,a.shape);let r={a:n,b:a};return M.runKernel(Xs,r)}var Bl=P({minimum_:HD});function jD(e,t,n){A(n==="reflect"||n==="symmetric",()=>`Invalid mode. Mode must be either reflect or symmetric. Got ${n}.`);let a=E(e,"x","mirrorPad");if(a.rank===0)throw new Error("mirrorPad(scalar) is not defined. Pass non-scalar to mirrorPad");A(t.length===a.rank,()=>`Padding doesn't match input. Must be ${a.rank}. Got ${t.length}.`);let r=n==="reflect"?1:0;for(let o=0;o<a.rank;o++)A(t[o].length===2,()=>"Invalid number of paddings. Must be length of 2 each."),A(t[o][0]>=0&&t[o][0]<=a.shape[o]-r&&t[o][1]>=0&&t[o][1]<=a.shape[o]-r,()=>`Padding in dimension ${o} cannot be greater than or equal to ${a.shape[o]-r} or less than 0 for input of shape ${a.shape}`);let s={paddings:t,mode:n},i={x:a};return M.runKernel(xc,i,s)}var Ky=P({mirrorPad_:jD});function qD(e,t){let n=E(e,"a","mod"),a=E(t,"b","mod");[n,a]=Nt(n,a);let r={a:n,b:a};return M.runKernel(nl,r)}var Xy=P({mod_:qD});function KD(e){let t=E(e,"x","square"),n={};return M.runKernel("Square",{x:t},n)}var lt=P({square_:KD});function XD(e,t=null,n=!1){e=E(e,"x","moments");let a=oa(t,e.shape),r=St(e,a,n),s=r.shape;n||(s=Ci(r.shape,a));let i=lt(fe(ce(e,"float32"),j(r,s))),o=St(i,a,n);return{mean:r,variance:o}}var vh=P({moments_:XD});function YD(e,t,n,a){let r=E(t,"data","multiRNNCell"),s=$c(n,"c","multiRNNCell"),i=$c(a,"h","multiRNNCell"),o=r,l=[];for(let p=0;p<e.length;p++){let d=e[p](o,s[p],i[p]);l.push(d[0]),l.push(d[1]),o=d[1]}let c=[],u=[];for(let p=0;p<l.length;p+=2)c.push(l[p]),u.push(l[p+1]);return[c,u]}var ZD=P({multiRNNCell_:YD});function JD(e,t,n,a=!1){let r=E(e,"logits","multinomial"),s=r.size,i=r.rank;if(s<2)throw new Error(`Error in multinomial: you need at least 2 outcomes, but got ${s}.`);if(i>2)throw new Error(`Rank of probabilities must be 1 or 2, but is ${i}`);n=n||Math.random();let o={logits:i===1?j(r,[1,-1]):r},l={numSamples:t,seed:n,normalized:a},c=M.runKernel(Wd,o,l);return i===1?j(c,[c.size]):c}var N1=P({multinomial_:JD});function QD(e,t){let n=E(e,"a","notEqual"),a=E(t,"b","notEqual");[n,a]=Nt(n,a),xt(n.shape,a.shape);let r={a:n,b:a};return M.runKernel(rl,r)}var _i=P({notEqual_:QD});function vt(e,t="float32"){if(t==="complex64"){let a=vt(e,"float32"),r=vt(e,"float32");return Ur(a,r)}let n=gd(Pt(e),t);return M.makeTensor(n,e,t)}function Xa(e,t="float32"){if(t==="complex64"){let a=Xa(e,"float32"),r=vt(e,"float32");return Ur(a,r)}let n=Wg(Pt(e),t);return M.makeTensor(n,e,t)}function eR(e){let t={x:E(e,"x","onesLike")};return M.runKernel(ll,t)}var Dn=P({onesLike_:eR});function tR(e,t){let n=E(e,"v1","outerProduct"),a=E(t,"v2","outerProduct");A(n.rank===1&&a.rank===1,()=>`Error in outerProduct: inputs must be rank 1, but got ranks ${n.rank} and ${a.rank}.`);let r=j(n,[-1,1]),s=j(a,[1,-1]);return ze(r,s)}var nR=P({outerProduct_:tR});function aR(e,t,n=0){let a=E(e,"x","pad");if(a.rank===0)throw new Error("pad(scalar) is not defined. Pass non-scalar to pad");let r={paddings:t,constantValue:n},s={x:a};return M.runKernel(Js,s,r)}var Zn=P({pad_:aR});function rR(e,t,n=0){return A(t.length===2,()=>"Invalid number of paddings. Must be length of 2."),Zn(e,[t],n)}var sR=P({pad1d_:rR});function iR(e,t,n=0){return A(t.length===2&&t[0].length===2&&t[1].length===2,()=>"Invalid number of paddings. Must be length of 2 each."),Zn(e,t,n)}var oR=P({pad2d_:iR});function lR(e,t,n=0){return A(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."),Zn(e,t,n)}var uR=P({pad3d_:lR});function cR(e,t,n=0){return A(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."),Zn(e,t,n)}var pR=P({pad4d_:cR});function dR(e,t,n){let a=E(e,"x","spaceToBatchND");A(a.rank>=1+t.length,()=>`input rank ${a.rank} should be > than [blockShape] ${t.length}`),A(n.length===t.length,()=>`paddings.shape[0] ${n.length} must be equal to [blockShape] ${t.length}`),A(a.shape.reduce((i,o,l)=>l>0&&l<=t.length?i&&(o+n[l-1][0]+n[l-1][1])%t[l-1]==0:i,!0),()=>`input spatial dimensions ${a.shape.slice(1)} with paddings ${n.toString()} must be divisible by blockShapes ${t.toString()}`);let r={x:a},s={blockShape:t,paddings:n};return M.runKernel(kc,r,s)}var Vc=P({spaceToBatchND_:dR});function fR(e,t,n,a,r,s){r==null&&(r=[1,1]),s==null&&(s=1),a===0&&(a="valid");let i=E(e,"x","maxPool"),o=i,l=!1;i.rank===3&&(l=!0,o=j(i,[1,i.shape[0],i.shape[1],i.shape[2]])),A(Ha(s,r),()=>`Error in pool: Either strides or dilations must be 1. Got strides ${s} and dilations '${r}'`);let c=e1(o.shape,t,s,r,a),u=[c.dilationHeight,c.dilationWidth],p;a==="same"?p=mR([c.filterHeight,c.filterWidth],u):p=[[0,0],[0,0]];let d=u[0]===1&&u[1]===1,[h,m]=hR([c.inHeight,c.inWidth],u,p),f=d?a:"valid",g=d?o:Vc(o,u,h),y=(n==="avg"?()=>Xn(g,t,s,f):()=>Ft(g,t,s,f))(),b=d?y:Oc(y,u,m);return l?j(b,[b.shape[1],b.shape[2],b.shape[3]]):b}function hR(e,t,n){let a=n.map(u=>u[0]),r=n.map(u=>u[1]),s=e.concat(a,r),i=t.map((u,p)=>(u-s[p]%u)%u),o=r.map((u,p)=>u+i[p]),l=t.map((u,p)=>[a[p],o[p]]),c=t.map((u,p)=>[0,i[p]]);return[l,c]}function mR(e,t){let n=e.map((s,i)=>s+(s-1)*(t[i]-1)).map(s=>s-1),a=n.map(s=>Math.floor(s/2)),r=n.map((s,i)=>s-a[i]);return n.map((s,i)=>[a[i],r[i]])}var T1=P({pool_:fR});function gR(e,t){let n=E(e,"base","pow"),a=E(t,"exp","pow");[n,a]=Nt(n,a);let r={a:n,b:a};return M.runKernel(Qs,r)}var gr=P({pow_:gR});function yR(e,t){let n=E(e,"x","prelu"),a=E(t,"alpha","prelu"),r={x:n,alpha:a};return M.runKernel(ei,r)}var Uc=P({prelu_:yR});function bR(e,t=null,n=!1){let a=E(e,"x","prod");a.dtype==="bool"&&(a=ce(a,"int32"));let r={x:a},s={axis:t,keepDims:n};return M.runKernel(cl,r,s)}var wh=P({prod_:bR});function xR(e,t,n){let a=Pt(e),r=null;if(n==null||n==="float32")r=new Float32Array(a);else if(n==="int32")r=new Int32Array(a);else if(n==="bool")r=new Uint8Array(a);else throw new Error(`Unknown data type ${n}`);for(let s=0;s<a;s++)r[s]=t();return M.makeTensor(r,e,n)}var vR=P({rand_:xR}),Yy=Io(dd()),Zy=class{constructor(e,t,n,a,r){this.mean=e,this.stdDev=t,this.dtype=n,this.nextVal=NaN,this.truncated=a,this.truncated&&(this.upper=this.mean+this.stdDev*2,this.lower=this.mean-this.stdDev*2);let s=r||Math.random();this.random=Yy.alea(s.toString())}nextValue(){if(!isNaN(this.nextVal)){let a=this.nextVal;return this.nextVal=NaN,a}let e,t,n=!1;for(;!n;){let a,r,s;do a=2*this.random()-1,r=2*this.random()-1,s=a*a+r*r;while(s>=1||s===0);let i=Math.sqrt(-2*Math.log(s)/s);e=this.mean+this.stdDev*a*i,t=this.mean+this.stdDev*r*i,(!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}},wR=class{constructor(e,t,n,a){this.alpha=e,this.beta=1/t,this.dtype=n;let r=a||Math.random();this.randu=Yy.alea(r.toString()),this.randn=new Zy(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,a,r,s;for(;;){do a=this.randn.nextValue(),s=1+this.c*a;while(s<=0);if(s*=s*s,e=a*a,t=1-.331*e*e,n=.5*e+this.d*(1-s+Math.log(s)),r=this.randu(),r<t||Math.log(r)<n)break}return s=1/this.beta*this.d*s,this.alpha<1&&(s*=Math.pow(this.randu(),1/this.alpha)),this.convertValue(s)}convertValue(e){return this.dtype==="float32"?e:Math.round(e)}},kR=class{constructor(e=0,t=1,n,a){if(this.canReturnFloat=()=>this.dtype==null||this.dtype==="float32",this.min=e,this.range=t-e,this.dtype=n,a==null&&(a=Math.random()),typeof a=="number"&&(a=a.toString()),!this.canReturnFloat()&&this.range<=1)throw new Error(`The difference between ${e} - ${t} <= 1 and dtype is not float`);this.random=Yy.alea(a)}convertValue(e){return this.canReturnFloat()?e:Math.round(e)}nextValue(){return this.convertValue(this.min+this.range*this.random())}};function IR(e,t,n=1,a="float32",r){if(n==null&&(n=1),a==null&&(a="float32"),a!=="float32"&&a!=="int32")throw new Error(`Unsupported data type ${a}`);let s=new wR(t,n,a,r),i=Le(e,a);for(let o=0;o<i.values.length;o++)i.values[o]=s.nextValue();return i.toTensor()}var NR=P({randomGamma_:IR});function TR(e,t=0,n=1,a,r){if(a!=null&&a==="bool")throw new Error(`Unsupported data type ${a}`);let s=new Zy(t,n,a,!1,r),i=Le(e,a);for(let o=0;o<i.values.length;o++)i.values[o]=s.nextValue();return i.toTensor()}var S1=P({randomNormal_:TR});function SR(e,t=0,n=1,a="float32",r){let s=Le(e,a),i=new kR(t,n,null,r);for(let o=0;o<s.values.length;o++)s.values[o]=i.nextValue();return s.toTensor()}var Wl=P({randomUniform_:SR});function kh(e,t,n=1,a="float32"){if(n===0)throw new Error("Cannot have a step of zero");let r={start:e,stop:t,step:n,dtype:a};return M.runKernel(vc,{},r)}function CR(e){let t={input:E(e,"input","real")};return M.runKernel(Vd,t)}var Gc=P({real_:CR});function _R(e){let t={x:E(e,"x","reciprocal")};return M.runKernel(pl,t)}var Jy=P({reciprocal_:_R});function ER(e){let t={x:E(e,"x","relu")};return M.runKernel(ti,t)}var Ye=P({relu_:ER});function AR(e){let t={x:E(e,"x","relu6")};return M.runKernel(ai,t)}var Ih=P({relu6_:AR});function FR(e,t){let n={x:E(e,"x","reverse")},a={dims:t};return M.runKernel(ri,n,a)}var Rn=P({reverse_:FR});function $R(e){let t=E(e,"x","reverse");return A(t.rank===1,()=>`Error in reverse1D: x must be rank 1 but got rank ${t.rank}.`),Rn(t,0)}var DR=P({reverse1d_:$R});function RR(e,t){let n=E(e,"x","reverse");return A(n.rank===2,()=>`Error in reverse2D: x must be rank 2 but got rank ${n.rank}.`),Rn(n,t)}var MR=P({reverse2d_:RR});function PR(e,t){let n=E(e,"x","reverse");return A(n.rank===3,()=>`Error in reverse3D: x must be rank 3 but got rank ${n.rank}.`),Rn(n,t)}var OR=P({reverse3d_:PR});function LR(e,t){let n=E(e,"x","reverse");return A(n.rank===4,()=>`Error in reverse4D: x must be rank 4 but got rank ${n.rank}.`),Rn(n,t)}var zR=P({reverse4d_:LR});function BR(e){let t={x:E(e,"x","round")};return M.runKernel(si,t)}var Qy=P({round_:BR});function WR(e){let t={x:E(e,"x","rsqrt")};return M.runKernel(ii,t)}var Nh=P({rsqrt_:WR});function de(e,t){if((on(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"&&on(e)&&!(e instanceof Uint8Array))throw new Error("When making a scalar from encoded string, the value must be `Uint8Array`.");return Gr(e,[],[],t)}function VR(e){let t={x:E(e,"x","selu")};return M.runKernel(fl,t)}var Th=P({selu_:VR});function UR(e,t,n,a,r,s=[1,1],i="NHWC"){let o=E(e,"x","separableConv2d"),l=E(t,"depthwiseFilter","separableConv2d"),c=E(n,"pointwiseFilter","separableConv2d"),u=o,p=!1;if(o.rank===3&&(p=!0,u=j(o,[1,o.shape[0],o.shape[1],o.shape[2]])),i==="NCHW")throw new Error("separableConv2d currently does not support dataFormat NCHW; only NHWC is supported");A(u.rank===4,()=>`Error in separableConv2d: input must be rank 4, but got rank ${u.rank}.`),A(l.rank===4,()=>`Error in separableConv2d: depthwise filter must be rank 4, but got rank ${l.rank}.`),A(c.rank===4,()=>`Error in separableConv2d: pointwise filter must be rank 4, but got rank ${l.rank}.`),A(c.shape[0]===1,()=>`Error in separableConv2d: the first dimension of pointwise filter must be 1, but got ${c.shape[0]}.`),A(c.shape[1]===1,()=>`Error in separableConv2d: the second dimension of pointwise filter must be 1, but got ${c.shape[1]}.`);let d=l.shape[2],h=l.shape[3];A(c.shape[2]===d*h,()=>`Error in separableConv2d: the third dimension of pointwise filter must be ${d*h}, but got ${c.shape[2]}.`);let m=Yr(u,l,a,r,i,s),f=At(m,c,1,"valid",i);return p?j(f,[f.shape[1],f.shape[2],f.shape[3]]):f}var Ei=P({separableConv2d_:UR});async function GR(e,t){let n=E(e,"x","setdiff1d"),a=E(t,"y","setdiff1d");A(n.dtype===a.dtype,()=>`x and y should have the same dtype, but got x (${n.dtype}) and y (${a.dtype}).`),A(n.rank===1,()=>`x should be 1D tensor, but got x (${n.shape}).`),A(a.rank===1,()=>`y should be 1D tensor, but got y (${a.shape}).`);let r=await n.data(),s=await a.data(),i=new Set(s),o=0;for(let u=0;u<r.length;u++)i.has(r[u])||o++;let l=new Ot([o],n.dtype),c=new Ot([o],"int32");for(let u=0,p=0;u<r.length;u++)i.has(r[u])||(l.values[p]=r[u],c.values[p]=u,p++);return[l.toTensor(),c.toTensor()]}var C1=GR;function HR(e){let t={x:E(e,"x","sign")};return M.runKernel(bl,t)}var eb=P({sign_:HR});function jR(e){let t={x:E(e,"x","sin")};return M.runKernel(oi,t)}var Sh=P({sin_:jR});function qR(e){let t={x:E(e,"x","sinh")};return M.runKernel(yl,t)}var Ch=P({sinh_:qR});function KR(e,t,n){let a=E(e,"x","slice1d");return A(a.rank===1,()=>`slice1d expects a rank-1 tensor, but got a rank-${a.rank} tensor`),We(a,[t],[n])}var _h=P({slice1d_:KR});function XR(e,t,n){let a=E(e,"x","slice2d");return A(a.rank===2,()=>`slice2d expects a rank-2 tensor, but got a rank-${a.rank} tensor`),We(a,t,n)}var tb=P({slice2d_:XR});function YR(e,t,n){let a=E(e,"x","slice3d");return A(a.rank===3,()=>`slice3d expects a rank-3 tensor, but got a rank-${a.rank} tensor`),We(a,t,n)}var Vl=P({slice3d_:YR});function ZR(e,t,n){let a=E(e,"x","slice4d");return A(a.rank===4,()=>`slice4d expects a rank-4 tensor, but got a rank-${a.rank} tensor`),We(a,t,n)}var Hc=P({slice4d_:ZR});function JR(e,t=-1){let n=E(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 a={logits:n},r={dim:t};return M.runKernel(pi,a,r)}var Na=P({softmax_:JR});function QR(e){A(e.dtype==="complex64",()=>`The dtype for tf.spectral.fft() must be complex64 but got ${e.dtype}.`);let t={input:e};return M.runKernel(Dd,t)}var jc=P({fft_:QR});function eM(e){A(e.dtype==="complex64",()=>`The dtype for tf.spectral.ifft() must be complex64 but got ${e.dtype}.`);let t={input:e};return M.runKernel(Rd,t)}var Ul=P({ifft_:eM});function tM(e){let t=e.shape[e.shape.length-1],n=e.size/t,a;if(t<=2){let r=j(e,[n,t]);a=Ul(r)}else{let r=[n,2*(t-1)],s=j(Gc(e),[n,t]),i=j(mh(e),[n,t]),o=Rn(We(s,[0,1],[n,t-2]),1),l=L(Rn(We(i,[0,1],[n,t-2]),1),de(-1)),c=Qe([s,o],1),u=Qe([i,l],1),p=j(Ur(c,u),[r[0],r[1]]);a=Ul(p)}if(a=Gc(a),e.rank===3&&e.shape[0]!==0){let r=a,s=e.shape[0];a=j(a,[s,a.shape[0]/s,a.shape[1]]),r.dispose()}return a}var Eh=P({irfft_:tM});function nM(e,t,n=0){let a={x:E(e,"x","split")},r={numOrSizeSplits:t,axis:n};return M.runKernel(vl,a,r)}var Mn=P({split_:nM});function aM(e,t){A(e.dtype==="float32",()=>`The dtype for rfft() must be real value but got ${e.dtype}`);let n=e.shape[e.shape.length-1],a=e.size/n,r;if(t!=null&&t<n){let m=e.shape.map(g=>0),f=e.shape.map(g=>g);f[e.shape.length-1]=t,r=We(e,m,f),n=t}else if(t!=null&&t>n){let m=e.shape.map(f=>f);m[e.shape.length-1]=t-n,r=Qe([e,vt(m)],e.shape.length-1),n=t}else r=e;let s=Ge(r),i=j(Ur(r,s),[a,n]),o=jc(i),l=Math.floor(n/2)+1,c=Gc(o),u=mh(o),p=Mn(c,[l,n-l],c.shape.length-1),d=Mn(u,[l,n-l],u.shape.length-1),h=r.shape.slice();return h[r.shape.length-1]=l,j(Ur(p[0],d[0]),h)}var qc=P({rfft_:aM});function rM(e){let t={x:E(e,"x","sqrt")};return M.runKernel(ui,t)}var en=P({sqrt_:rM});function sM(e,t){let n=E(e,"a","squaredDifference"),a=E(t,"b","squaredDifference");[n,a]=Nt(n,a),xt(n.shape,a.shape);let r={a:n,b:a},s={};return M.runKernel(di,r,s)}var Ah=P({squaredDifference_:sM});function iM(e,t){let n=E(e,"x","squeeze");return j(n,Hw(n.shape,t).newShape)}var Qr=P({squeeze_:iM});function oM(e,t=0){let n=$c(e,"tensors","stack","string_or_numeric");A(n.length>=1,()=>"Pass at least one tensor to tf.stack"),n.length>0&&A(t<=n[0].rank,()=>"Axis must be <= rank of the tensor");let a=n,r={axis:t};return M.runKernel(ul,a,r)}var $t=P({stack_:oM});function lM(e,t=0){let n={x:E(e,"x","step")},a={alpha:t};return M.runKernel(Wr,n,a)}var Gl=P({step_:lM});function uM(e,t,n,a,r=0,s=0,i=0,o=0,l=0){let c={x:E(e,"x","stridedSlice")},u={begin:t,end:n,strides:a,beginMask:r,endMask:s,ellipsisMask:i,newAxisMask:o,shrinkAxisMask:l};return M.runKernel(wl,c,u)}var nb=P({stridedSlice_:uM});function cM(e){let t={x:E(e,"x","tan")};return M.runKernel(kl,t)}var ab=P({tan_:cM});function tt(e,t){Ns(e);let n=Ga(e,t);if(n.length!==1)throw new Error("tensor1d() requires values to be a flat/TypedArray");return Gr(e,null,n,t)}function Ta(e,t,n){if(Ns(e),t!=null&&t.length!==2)throw new Error("tensor2d() requires shape to have two numbers");let a=Ga(e,n);if(a.length!==2&&a.length!==1)throw new Error("tensor2d() requires values to be number[][] or flat/TypedArray");if(a.length===1&&t==null)throw new Error("tensor2d() requires shape to be provided when `values` are a flat/TypedArray");return Gr(e,t,a,n)}function Sa(e,t,n){if(Ns(e),t!=null&&t.length!==4)throw new Error("tensor4d() requires shape to have four numbers");let a=Ga(e,n);if(a.length!==4&&a.length!==1)throw new Error("tensor4d() requires values to be number[][][][] or flat/TypedArray");if(a.length===1&&t==null)throw new Error("tensor4d() requires shape to be provided when `values` are a flat array");return Gr(e,t,a,n)}function pM(e,t,n){if(Ns(e),t!=null&&t.length!==5)throw new Error("tensor5d() requires shape to have five numbers");let a=Ga(e,n);if(a.length!==5&&a.length!==1)throw new Error("tensor5d() requires values to be number[][][][][] or flat/TypedArray");if(a.length===1&&t==null)throw new Error("tensor5d() requires shape to be provided when `values` are a flat array");return Gr(e,t,a,n)}function dM(e,t,n){if(Ns(e),t!=null&&t.length!==6)throw new Error("tensor6d() requires shape to have six numbers");let a=Ga(e,n);if(a.length!==6&&a.length!==1)throw new Error("tensor6d() requires values to be number[][][][][][] or flat/TypedArray");if(a.length===1&&t==null)throw new Error("tensor6d() requires shape to be provided when `values` are a flat array");return t=t||a,Gr(e,t,a,n)}function hM(e,t=1,n=!0){let a=E(e,"x","topk");if(a.rank===0)throw new Error("topk() expects the input to be of rank 1 or higher");let r=a.shape[a.shape.length-1];if(t>r)throw new Error(`'k' passed to topk() must be <= the last dimension (${r}) but got ${t}`);let s={x:a},i={k:t,sorted:n},[o,l]=M.runKernel(Il,s,i);return{values:o,indices:l}}var rb=P({topk_:hM});function mM(e,t=0,n=1,a,r){if(a!=null&&a==="bool")throw new Error("Unsupported data type $ { dtype }");let s=new Zy(t,n,a,!0,r),i=Le(e,a);for(let o=0;o<i.values.length;o++)i.values[o]=s.nextValue();return i.toTensor()}var Fh=P({truncatedNormal_:mM});function fM(e,t=0){let n=E(e,"x","unique","string_or_numeric");A(n.rank>0,()=>"The input tensor must be at least 1D");let a={x:n},r={axis:t},[s,i]=M.runKernel(jd,a,r);return{values:s,indices:i}}var $h=P({unique_:fM});function gM(e,t,n){let a=E(e,"x","unsortedSegmentSum"),r=E(t,"segmentIds","unsortedSegmentSum","int32");A(Gt(n),()=>"numSegments must be of dtype int");let s={x:a,segmentIds:r},i={numSegments:n};return M.runKernel(Nc,s,i)}var sb=P({unsortedSegmentSum_:gM});function yM(e,t=0){let n=E(e,"x","unstack","string_or_numeric");A(t>=-n.shape.length&&t<n.shape.length,()=>`Axis = ${t} is not in [-${n.shape.length}, ${n.shape.length})`);let a={value:n},r={axis:t};return M.runKernel(Nl,a,r)}var dt=P({unstack_:yM});function _1(e,t=!0,n,a){return M.makeVariable(e,t,n,a)}function E1(e,t){let n=[];for(let s=0;s<t.length;s++)t[s]&&n.push(s);let a=Le(e,"int32"),r=Le([n.length,e.length],"int32");for(let s=0;s<n.length;s++){let i=a.indexToLoc(n[s]),o=s*e.length;r.values.set(i,o)}return r.toTensor()}async function bM(e){let t=E(e,"condition","whereAsync","bool"),n=await t.data(),a=E1(t.shape,n);return e!==t&&t.dispose(),a}var ib=bM;async function xM(e,t,n){let a=E(e,"tensor","boolMask"),r=E(t,"mask","boolMask","bool"),s=n==null?0:n,i=r.rank,o=a.shape;A(i>0,()=>"mask cannot be scalar"),sn(o.slice(s,s+i),r.shape,"mask's shape must match the first K dimensions of tensor's shape,");let l=1;for(let f=s;f<s+i;f++)l*=o[f];let c=o.slice(0,s).concat([l],o.slice(s+i)),u=j(a,c),p=j(r,[-1]),d=await ib(p),h=Qr(d,[1]),m=Ti(u,h,s);return e!==a&&a.dispose(),t!==r&&r.dispose(),h.dispose(),u.dispose(),p.dispose(),d.dispose(),m}var vM=xM;function wM(e,t="euclidean",n=null,a=!1){e=E(e,"x","norm");let r=A1(e,t,n),s=r.shape;if(a){let i=oa(n,e.shape);s=Ci(r.shape,i)}return j(r,s)}function A1(e,t,n=null){if(e.rank===0)return Lt(e);if(e.rank!==1&&n===null)return A1(j(e,[-1]),t,n);if(e.rank===1||typeof n=="number"||Array.isArray(n)&&n.length===1){if(t===1)return Te(Lt(e),n);if(t===Infinity)return Yn(Lt(e),n);if(t===-Infinity)return zl(Lt(e),n);if(t==="euclidean"||t===2)return en(Te(gr(Lt(e),de(2,"int32")),n));throw new Error(`Error in norm: invalid ord value: ${t}`)}if(Array.isArray(n)&&n.length===2){if(t===1)return Yn(Te(Lt(e),n[0]),n[1]-1);if(t===Infinity)return Yn(Te(Lt(e),n[1]),n[0]);if(t===-Infinity)return zl(Te(Lt(e),n[1]),n[0]);if(t==="fro"||t==="euclidean")return en(Te(lt(e),n));throw new Error(`Error in norm: invalid ord value: ${t}`)}throw new Error(`Error in norm: invalid axis: ${n}`)}var Dh=P({norm_:wM});function kM(e,t,n,a,r=!0){let s=E(e,"v","movingAverage"),i=E(t,"x","movingAverage"),o=E(n,"decay","movingAverage");u0(s,i),A(hr(s.shape,i.shape),()=>"Shape mismatch in v and x");let l=de(1),c=fe(l,o),u=L(fe(i,s),c);if(r){A(a!=null,()=>"When using zeroDebias: true, step is required.");let p=E(a,"step","movingAverage");u=ve(u,fe(l,gr(o,p)))}return Z(s,u)}var IM=P({movingAverage_:kM});function NM(e,t,n){let a=E(e,"indices","scatterND","int32"),r=E(t,"updates","scatterND");by(r,a,n);let s={indices:a,updates:r},i={shape:n};return M.runKernel(hl,s,i)}var F1=P({scatterND_:NM});function TM(e,t,n,a){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,s=e.rank>1?e.shape[1]:1;if(n.length!==s)throw new Error(`outputShape has incorrect number of elements:, ${n.length}, should be: ${s}.`);let i=t.size;if(!(t.rank===0||t.rank===1&&i===r))throw new Error(`sparseValues has incorrect shape ${t.shape}, should be [] or [${r}]`);if(t.dtype!==a.dtype)throw new Error("sparseValues.dtype must match defaultValues.dtype")}function SM(e,t,n,a=0){let r=E(e,"sparseIndices","sparseToDense","int32"),s=E(t,"sparseValues","sparseToDense"),i=E(a,"defaultValue","sparseToDense",s.dtype);TM(r,s,n,i);let o={sparseIndices:r,sparseValues:s,defaultValue:i},l={outputShape:n};return M.runKernel(Hd,o,l)}var ob=P({sparseToDense_:SM});function CM(e,t){let n=E(t,"indices","gatherND","int32"),a={params:E(e,"x","gatherND"),indices:n};return M.runKernel(jo,a)}var $1=P({gatherND_:CM});function _M(e,t){if(t==null)return e.shape.slice();if(hr(e.shape,t))return t;if(e.shape.length===t.length){let n=[];for(let a=0;a<e.shape.length;a++)t[a]==null&&e.shape[a]!=null?n.push(e.shape[a]):n.push(t[a]);return n}return t}function EM(e,t,n,a){let r=E(e,"x","dropout");if(A(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.`),A(t>=0&&t<1,()=>`rate must be a float in the range [0, 1), but got ${t}.`),t===0)return e instanceof Ae?r.clone():r;let s=_M(r,n),i=1-t,o=ve(Ol(Z(Wl(s,0,1,"float32",a),i)),i);return L(r,o)}var D1=P({dropout_:EM});function R1(e){return Math.floor(Math.pow(2,Math.ceil(Math.log(e)/Math.log(2))))}function lb(e,t,n){let a=1-e%2,r=new Float32Array(e);for(let s=0;s<e;++s){let i=2*Math.PI*s/(e+a-1);r[s]=t-n*Math.cos(i)}return tt(r,"float32")}async function AM(e,t,n=1){let a=E(e,"predictions","inTopK"),r=E(t,"targets","inTopK");A(a.rank>1,()=>`inTopK() expects the predictions to be of rank 2 or higher, but got ${a.rank}`),A(a.rank-1===r.rank,()=>`predictions rank should be 1 larger than targets rank, but got predictions rank ${a.rank} and targets rank ${r.rank}`),sn(a.shape.slice(0,a.shape.length-1),r.shape,"predictions's shape should be align with the targets' shape, except the last dimension.");let s=a.shape[a.shape.length-1];A(n>0&&n<=s,()=>`'k' passed to inTopK() must be > 0 && <= the predictions last dimension (${s}), but got ${n}`);let i=await a.data(),o=await r.data(),[l,c]=[i.length/s,s],u=jw("bool",l);for(let p=0;p<l;p++){let d=p*c,h=i.subarray(d,d+c),m=[];for(let f=0;f<h.length;f++)m.push({value:h[f],index:f});m.sort((f,g)=>g.value-f.value),u[p]=0;for(let f=0;f<n;f++)if(m[f].index===o[p]){u[p]=1;break}}return e!==a&&a.dispose(),t!==r&&r.dispose(),qn(u,r.shape,"bool")}var FM=AM,es={};Oe(es,{conv2d:()=>$M,depthwiseConv2d:()=>DM,matMul:()=>RM});function MM(e,t,n,a,r,s="NHWC",i){let o=e;e.rank===3&&(o=j(e,[1,e.shape[0],e.shape[1],e.shape[2]]));let l=t;l.rank===3&&(l=j(t,[1,t.shape[0],t.shape[1],t.shape[2]])),A(o.rank===4,()=>`Error in conv2dDerFilter: input must be rank 4, but got shape ${o.shape}.`),A(l.rank===4,()=>`Error in conv2dDerFilter: dy must be rank 4, but got shape ${l.shape}.`),A(n.length===4,()=>`Error in conv2dDerFilter: filterShape must be length 4, but got ${n}.`);let c=s==="NHWC"?o.shape[3]:o.shape[1],u=s==="NHWC"?l.shape[3]:l.shape[1];A(c===n[2],()=>`Error in conv2dDerFilter: depth of input ${c}) must match input depth in filter (${n[2]}.`),A(u===n[3],()=>`Error in conv2dDerFilter: depth of dy (${u}) must match output depth for filter (${n[3]}).`),i!=null&&A(Gt(r),()=>`Error in conv2dDerFilter: pad must be an integer when using, dimRoundingMode ${i} but got pad ${r}.`);let p={x:o,dy:l},d={strides:a,pad:r,dataFormat:s,dimRoundingMode:i,filterShape:n};return M.runKernel(Id,p,d)}var ub=P({conv2DBackpropFilter_:MM});function Rh(e,t,n){if(n==null||n==="linear")return e;if(n==="relu")return L(e,Gl(t));throw new Error(`Cannot compute gradient for fused activation ${n}.`)}function Mh(e,t){let n=t,a=zt(e.shape,t.shape);return a.length>0&&(n=Te(n,a)),j(n,e.shape)}function Ph(e,t,n,a){if(t==="linear")return e;if(t==="relu")return Ye(e);if(t==="elu")return Pl(e);if(t==="relu6")return Ih(e);if(t==="prelu")return Uc(e,n);if(t==="leakyrelu")return Bc(e,a);throw new Error(`Unknown fused activation ${t}.`)}var Oh=(e,t)=>!(e>0)||t==="linear";function PM({x:e,filter:t,strides:n,pad:a,dataFormat:r="NHWC",dilations:s=[1,1],dimRoundingMode:i,bias:o,activation:l="linear",preluActivationWeights:c,leakyreluAlpha:u}){if(l=l||"linear",Oh(M.state.gradientDepth,l)===!1){let N=At(e,t,n,a,r,s,i);return o!=null&&(N=Z(N,o)),Ph(N,l,c,u)}let p=E(e,"x","conv2d"),d=E(t,"filter","conv2d"),h=p,m=!1;p.rank===3&&(m=!0,h=j(p,[1,p.shape[0],p.shape[1],p.shape[2]])),A(h.rank===4,()=>`Error in fused conv2d: input must be rank 4, but got rank ${h.rank}.`),A(d.rank===4,()=>`Error in fused conv2d: filter must be rank 4, but got rank ${d.rank}.`),i!=null&&A(Gt(a),()=>`Error in fused conv2d: pad must be an integer when using, dimRoundingMode ${i} but got pad ${a}.`),A(h.shape[3]===d.shape[2],()=>`Error in conv2d: depth of input (${h.shape[3]}) must match input depth for filter ${d.shape[2]}.`),A(Ha(n,s),()=>`Error in conv2D: Either strides or dilations must be 1. Got strides ${n} and dilations '${s}'`),A(r==="NHWC",()=>`Error in conv2d: got dataFormat of ${r} but only NHWC is currently supported.`);let f=Pc(h.shape,d.shape,n,s,a,i),g;o!=null&&(g=E(o,"bias","fused conv2d"),[g]=Nt(g,p),xt(f.outShape,g.shape));let y;c!=null&&(y=E(c,"prelu weights","fused conv2d"));let b=(N,T)=>{let[C,$,F,O]=T,B=Rh(N,F,l);A(Xr(s),()=>`Error in gradient of fused conv2D: dilation rates greater than 1 are not yet supported in gradients. Got dilations '${s}'`);let W=My($.shape,B,C,n,a),G=ub($,B,C.shape,n,a),q=[W,G];if(O!=null){let H=Mh(O,B);q.push(H)}return q},v={x:h,filter:d,bias:g,preluActivationWeights:y},x={strides:n,pad:a,dataFormat:r,dilations:s,dimRoundingMode:i,activation:l,leakyreluAlpha:u};return o==null?qa((N,T,C)=>{let $=M.runKernel(yi,v,x);return C([T,N,$]),m&&($=j($,[$.shape[1],$.shape[2],$.shape[3]])),{value:$,gradFunc:b}})(h,d):qa((N,T,C,$)=>{let F=M.runKernel(yi,v,x);return $([T,N,F,C]),m&&(F=j(F,[F.shape[1],F.shape[2],F.shape[3]])),{value:F,gradFunc:b}})(h,d,g)}var $M=P({fusedConv2d_:PM});function OM(e,t,n,a,r,s=[1,1],i){let o=e;e.rank===3&&(o=j(e,[1,e.shape[0],e.shape[1],e.shape[2]]));let l=t;l.rank===3&&(l=j(t,[1,t.shape[0],t.shape[1],t.shape[2]]));let c={x:o,dy:l},u={strides:a,pad:r,dimRoundingMode:i,dilations:s,filterShape:n};return M.runKernel(Cd,c,u)}var M1=P({depthwiseConv2dNativeBackpropFilter_:OM});function LM(e,t,n,a,r,s=[1,1],i){let o=t,l=!1;t.rank===3&&(l=!0,o=j(t,[1,t.shape[0],t.shape[1],t.shape[2]]));let c={dy:o,filter:n},u={strides:a,pad:r,dimRoundingMode:i,dilations:s,inputShape:e},p=M.runKernel(_d,c,u);return l?j(p,[p.shape[1],p.shape[2],p.shape[3]]):p}var P1=P({depthwiseConv2dNativeBackpropInput_:LM});function zM({x:e,filter:t,strides:n,pad:a,dataFormat:r="NHWC",dilations:s=[1,1],dimRoundingMode:i,bias:o,activation:l="linear",preluActivationWeights:c,leakyreluAlpha:u}){if(Oh(M.state.gradientDepth,l)===!1){let N=Yr(e,t,n,a,r,s,i);return o!=null&&(N=Z(N,o)),Ph(N,l,c,u)}let p=E(e,"x","depthwiseConv2d"),d=E(t,"filter","depthwiseConv2d"),h=p,m=!1;p.rank===3&&(m=!0,h=j(p,[1,p.shape[0],p.shape[1],p.shape[2]])),A(h.rank===4,()=>`Error in fused depthwiseConv2d: input must be rank 4, but got rank ${h.rank}.`),A(d.rank===4,()=>`Error in fused depthwiseConv2d: filter must be rank 4, but got rank ${d.rank}.`),A(h.shape[3]===d.shape[2],()=>`Error in fused depthwiseConv2d: number of input channels (${h.shape[3]}) must match the inChannels dimension in filter ${d.shape[2]}.`),s==null&&(s=[1,1]),A(Ha(n,s),()=>`Error in fused depthwiseConv2d: Either strides or dilations must be 1. Got strides ${n} and dilations '${s}'`),i!=null&&A(Gt(a),()=>`Error in fused depthwiseConv2d: pad must be an integer when using dimRoundingMode ${i} but got pad ${a}.`);let f=Pc(h.shape,d.shape,n,s,a,i,!0),g;o!=null&&(g=E(o,"bias","fused conv2d"),[g]=Nt(g,p),xt(f.outShape,g.shape));let y;c!=null&&(y=E(c,"prelu weights","fused depthwiseConv2d"));let b=(N,T)=>{A(Xr(s),()=>`Error in gradient of fused depthwiseConv2d: dilation rates greater than 1 are not yet supported. Got dilations '${s}'`);let[C,$,F,O]=T,B=Rh(N,F,l),W=P1($.shape,B,C,n,a,s,i),G=M1($,B,C.shape,n,a,s,i);if(O!=null){let q=Mh(g,B);return[W,G,q]}return[W,G]},v={x:h,filter:d,bias:g,preluActivationWeights:y},x={strides:n,pad:a,dataFormat:r,dilations:s,dimRoundingMode:i,activation:l,leakyreluAlpha:u};return o==null?qa((N,T,C)=>{let $=M.runKernel(bi,v,x);return C([T,N,$]),m&&($=j($,[$.shape[1],$.shape[2],$.shape[3]])),{value:$,gradFunc:b}})(h,d):qa((N,T,C,$)=>{let F=M.runKernel(bi,v,x);return $([T,N,F,C]),m&&(F=j(F,[F.shape[1],F.shape[2],F.shape[3]])),{value:F,gradFunc:b}})(h,d,g)}var DM=P({fusedDepthwiseConv2d_:zM});function BM({a:e,b:t,transposeA:n=!1,transposeB:a=!1,bias:r,activation:s="linear",preluActivationWeights:i,leakyreluAlpha:o}){if(Oh(M.state.gradientDepth,s)===!1){let O=ze(e,t,n,a);return r!=null&&(O=Z(O,r)),Ph(O,s,i,o)}let l=E(e,"a","fused matMul"),c=E(t,"b","fused matMul");[l,c]=Nt(l,c);let u=n?l.shape[l.rank-2]:l.shape[l.rank-1],p=a?c.shape[c.rank-1]:c.shape[c.rank-2],d=n?l.shape[l.rank-1]:l.shape[l.rank-2],h=a?c.shape[c.rank-2]:c.shape[c.rank-1],m=l.shape.slice(0,-2),f=c.shape.slice(0,-2),g=Pt(m),y=Pt(f);A(l.rank>=2&&c.rank>=2&&l.rank===c.rank,()=>`Error in fused matMul: inputs must have the same rank of at least 2, got ranks ${l.rank} and ${c.rank}.`),A(hr(m,f),()=>`Error in fused matMul: outer dimensions (${m}) and (${f}) of Tensors with shapes ${l.shape} and ${c.shape} must match.`),A(u===p,()=>`Error in fused matMul: inner shapes (${u}) and (${p}) of Tensors with shapes ${l.shape} and ${c.shape} and transposeA=${n} and transposeB=${a} must match.`);let b=l.shape.slice(0,-2).concat([d,h]),v=n?j(l,[g,u,d]):j(l,[g,d,u]),x=a?j(c,[y,h,p]):j(c,[y,p,h]),N;r!=null&&(N=E(r,"bias","fused matMul"),[N]=Nt(N,l),xt(b,N.shape));let T;i!=null&&(T=E(i,"prelu weights","fused matMul"));let C=(O,B)=>{let[W,G,q,H]=B,K=Rh(j(O,q.shape),q,s),X,ne;if(!n&&!a?(X=ze(K,G,!1,!0),ne=ze(W,K,!0,!1)):!n&&a?(X=ze(K,G,!1,!1),ne=ze(K,W,!0,!1)):n&&!a?(X=ze(G,K,!1,!0),ne=ze(W,K,!1,!1)):(X=ze(G,K,!0,!0),ne=ze(K,W,!0,!0)),r!=null){let J=Mh(H,K);return[X,ne,J]}else return[X,ne]},$={a:v,b:x,bias:N,preluActivationWeights:T},F={transposeA:n,transposeB:a,activation:s,leakyreluAlpha:o};return r==null?qa((O,B,W)=>{let G=M.runKernel(gi,$,F);return W([O,B,G]),{value:j(G,b),gradFunc:C}})(v,x):qa((O,B,W,G)=>{let q=M.runKernel(gi,$,F);return G([O,B,q,W]),{value:j(q,b),gradFunc:C}})(v,x,N)}var RM=P({fusedMatMul_:BM});function WM(e){return lb(e,.54,.46)}var VM=P({hammingWindow_:WM});function UM(e){return lb(e,.5,.5)}var O1=P({hannWindow_:UM});function GM(e,t,n,a=!1,r=0){let s=0,i=[];for(;s+t<=e.size;)i.push(We(e,s,t)),s+=n;if(a)for(;s<e.size;){let o=s+t-e.size,l=Qe([We(e,s,t-o),wn([o],r)]);i.push(l),s+=n}return i.length===0?Ta([],[0,t]):j(Qe(i),[i.length,t])}var L1=P({frame_:GM});function HM(e,t,n,a,r=O1){a==null&&(a=R1(t));let s=L1(e,t,n),i=L(s,r(t)),o=[];for(let l=0;l<s.shape[0];l++)o.push(qc(We(i,[l,0],[1,t]),a));return Qe(o)}var jM=P({stft_:HM});function qM(e,t,n,a,r="bilinear",s=0){let i=E(e,"image","cropAndResize"),o=E(t,"boxes","cropAndResize","float32"),l=E(n,"boxInd","cropAndResize","int32"),c=o.shape[0];A(i.rank===4,()=>`Error in cropAndResize: image must be rank 4,but got rank ${i.rank}.`),A(o.rank===2&&o.shape[1]===4,()=>`Error in cropAndResize: boxes must be have size [${c},4] but had shape ${o.shape}.`),A(l.rank===1&&l.shape[0]===c,()=>`Error in cropAndResize: boxInd must be have size [${c}] but had shape ${o.shape}.`),A(a.length===2,()=>`Error in cropAndResize: cropSize must be of length 2, but got length ${a.length}.`),A(a[0]>=1&&a[1]>=1,()=>`cropSize must be atleast [1,1], but was ${a}`),A(r==="bilinear"||r==="nearest",()=>`method must be bilinear or nearest, but was ${r}`);let u={image:i,boxes:o,boxInd:l},p={method:r,extrapolationValue:s,cropSize:a};return M.runKernel(Oo,u,p)}var KM=P({cropAndResize_:qM});function XM(e){let t=E(e,"image","flipLeftRight","float32");A(t.rank===4,()=>`Error in flipLeftRight: image must be rank 4,but got rank ${t.rank}.`);let n={image:t};return M.runKernel(Go,n,{})}var YM=P({flipLeftRight_:XM});function ZM(e,t,n=0,a=.5){let r=E(e,"image","rotateWithOffset","float32");A(r.rank===4,()=>`Error in rotateWithOffset: image must be rank 4,but got rank ${r.rank}.`);let s={image:r},i={radians:t,fillValue:n,center:a};return M.runKernel(Sl,s,i)}var JM=P({rotateWithOffset_:ZM});function Hl(e,t,n,a,r,s){a==null&&(a=.5),r==null&&(r=Number.NEGATIVE_INFINITY),s==null&&(s=0);let i=e.shape[0];return n=Math.min(n,i),A(0<=a&&a<=1,()=>`iouThreshold must be in [0, 1], but was '${a}'`),A(e.rank===2,()=>`boxes must be a 2D tensor, but was of rank '${e.rank}'`),A(e.shape[1]===4,()=>`boxes must have 4 columns, but 2nd dimension was ${e.shape[1]}`),A(t.rank===1,()=>"scores must be a 1D tensor"),A(t.shape[0]===i,()=>`scores has incompatible shape with boxes. Expected ${i}, but was ${t.shape[0]}`),A(0<=s&&s<=1,()=>`softNmsSigma must be in [0, 1], but was '${s}'`),{maxOutputSize:n,iouThreshold:a,scoreThreshold:r,softNmsSigma:s}}function QM(e,t,n,a=.5,r=Number.NEGATIVE_INFINITY){let s=E(e,"boxes","nonMaxSuppression"),i=E(t,"scores","nonMaxSuppression"),o=Hl(s,i,n,a,r);n=o.maxOutputSize,a=o.iouThreshold,r=o.scoreThreshold;let l={maxOutputSize:n,iouThreshold:a,scoreThreshold:r};return M.runKernel(sl,{boxes:s,scores:i},l)}var eP=P({nonMaxSuppression_:QM});function nP(e,t,n){let a=tP(e,t,n),r=a<0?-(a+1):a;e.splice(r,0,t)}function tP(e,t,n){return rP(e,t,n||aP)}function aP(e,t){return e>t?1:e<t?-1:0}function rP(e,t,n){let a=0,r=e.length,s=0,i=!1;for(;a<r;){s=a+(r-a>>>1);let o=n(t,e[s]);o>0?a=s+1:(r=s,i=!o)}return i?a:-a-1}function z1(e,t,n,a,r){return cb(e,t,n,a,r,0)}function B1(e,t,n,a,r,s){return cb(e,t,n,a,r,0,!1,s,!0)}function W1(e,t,n,a,r,s){return cb(e,t,n,a,r,s,!0)}function cb(e,t,n,a,r,s,i=!1,o=!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(V1);let u=s>0?-.5/s:0,p=[],d=[];for(;p.length<n&&c.length>0;){let g=c.pop(),{score:y,boxIndex:b,suppressBeginIndex:v}=g;if(y<r)break;let x=!1;for(let N=p.length-1;N>=v;--N){let T=sP(e,b,p[N]);if(T>=a){x=!0;break}if(g.score=g.score*iP(a,u,T),g.score<=r)break}g.suppressBeginIndex=p.length,x||(g.score===y?(p.push(b),d.push(g.score)):g.score>r&&nP(c,g,V1))}let h=p.length,m=n-h;o&&m>0&&(p.push(...new Array(m).fill(0)),d.push(...new Array(m).fill(0)));let f={selectedIndices:p};return i&&(f.selectedScores=d),l&&(f.validOutputs=h),f}function sP(e,t,n){let a=e.subarray(t*4,t*4+4),r=e.subarray(n*4,n*4+4),s=Math.min(a[0],a[2]),i=Math.min(a[1],a[3]),o=Math.max(a[0],a[2]),l=Math.max(a[1],a[3]),c=Math.min(r[0],r[2]),u=Math.min(r[1],r[3]),p=Math.max(r[0],r[2]),d=Math.max(r[1],r[3]),h=(o-s)*(l-i),m=(p-c)*(d-u);if(h<=0||m<=0)return 0;let f=Math.max(s,c),g=Math.max(i,u),y=Math.min(o,p),b=Math.min(l,d),v=Math.max(y-f,0)*Math.max(b-g,0);return v/(h+m-v)}function iP(e,t,n){let a=Math.exp(t*n*n);return n<=e?a:0}function V1(e,t){return e.score-t.score||e.score===t.score&&t.boxIndex-e.boxIndex}async function oP(e,t,n,a=.5,r=Number.NEGATIVE_INFINITY){let s=E(e,"boxes","nonMaxSuppressionAsync"),i=E(t,"scores","nonMaxSuppressionAsync"),o=Hl(s,i,n,a,r);n=o.maxOutputSize,a=o.iouThreshold,r=o.scoreThreshold;let l=await Promise.all([s.data(),i.data()]),c=l[0],u=l[1],{selectedIndices:p}=z1(c,u,n,a,r);return s!==e&&s.dispose(),i!==t&&i.dispose(),tt(p,"int32")}var lP=oP;function uP(e,t,n,a=.5,r=Number.NEGATIVE_INFINITY,s=0){let i=E(e,"boxes","nonMaxSuppression"),o=E(t,"scores","nonMaxSuppression"),l=Hl(i,o,n,a,r,s);n=l.maxOutputSize,a=l.iouThreshold,r=l.scoreThreshold,s=l.softNmsSigma;let c={boxes:i,scores:o},u={maxOutputSize:n,iouThreshold:a,scoreThreshold:r,softNmsSigma:s},p=M.runKernel(ol,c,u);return{selectedIndices:p[0],selectedScores:p[1]}}var cP=P({nonMaxSuppressionWithScore_:uP});async function pP(e,t,n,a=.5,r=Number.NEGATIVE_INFINITY,s=0){let i=E(e,"boxes","nonMaxSuppressionAsync"),o=E(t,"scores","nonMaxSuppressionAsync"),l=Hl(i,o,n,a,r,s);n=l.maxOutputSize,a=l.iouThreshold,r=l.scoreThreshold,s=l.softNmsSigma;let c=await Promise.all([i.data(),o.data()]),u=c[0],p=c[1],{selectedIndices:d,selectedScores:h}=W1(u,p,n,a,r,s);return i!==e&&i.dispose(),o!==t&&o.dispose(),{selectedIndices:tt(d,"int32"),selectedScores:tt(h)}}var dP=pP;function hP(e,t,n,a=.5,r=Number.NEGATIVE_INFINITY,s=!1){let i=E(e,"boxes","nonMaxSuppression"),o=E(t,"scores","nonMaxSuppression"),l=Hl(i,o,n,a,r,null),c=l.maxOutputSize,u=l.iouThreshold,p=l.scoreThreshold,d={boxes:i,scores:o},h={maxOutputSize:c,iouThreshold:u,scoreThreshold:p,padToMaxOutputSize:s},m=M.runKernel(il,d,h);return{selectedIndices:m[0],validOutputs:m[1]}}var mP=P({nonMaxSuppressionPadded_:hP});async function fP(e,t,n,a=.5,r=Number.NEGATIVE_INFINITY,s=!1){let i=E(e,"boxes","nonMaxSuppressionAsync"),o=E(t,"scores","nonMaxSuppressionAsync"),l=Hl(i,o,n,a,r,null),c=l.maxOutputSize,u=l.iouThreshold,p=l.scoreThreshold,[d,h]=await Promise.all([i.data(),o.data()]),{selectedIndices:m,validOutputs:f}=B1(d,h,c,u,p,s);return i!==e&&i.dispose(),o!==t&&o.dispose(),{selectedIndices:tt(m,"int32"),validOutputs:de(f,"int32")}}var gP=fP;function yP(e,t,n=!1,a=!1){let r=E(e,"images","resizeBilinear");A(r.rank===3||r.rank===4,()=>`Error in resizeBilinear: x must be rank 3 or 4, but got rank ${r.rank}.`),A(t.length===2,()=>`Error in resizeBilinear: new shape must 2D, but got shape ${t}.`),A(a===!1||n===!1,()=>"Error in resizeBilinear: If halfPixelCenters is true, alignCorners must be false.");let s=r,i=!1;r.rank===3&&(i=!0,s=j(r,[1,r.shape[0],r.shape[1],r.shape[2]]));let[]=t,o={images:s},l={alignCorners:n,halfPixelCenters:a,size:t},c=M.runKernel(ni,o,l);return i?j(c,[c.shape[1],c.shape[2],c.shape[3]]):c}var U1=P({resizeBilinear_:yP});function bP(e,t,n=!1,a=!1){let r=E(e,"images","resizeNearestNeighbor");A(r.rank===3||r.rank===4,()=>`Error in resizeNearestNeighbor: x must be rank 3 or 4, but got rank ${r.rank}.`),A(t.length===2,()=>`Error in resizeNearestNeighbor: new shape must 2D, but got shape ${t}.`),A(r.dtype==="float32"||r.dtype==="int32",()=>"`images` must have `int32` or `float32` as dtype"),A(a===!1||n===!1,()=>"Error in resizeNearestNeighbor: If halfPixelCenters is true, alignCorners must be false.");let s=r,i=!1;r.rank===3&&(i=!0,s=j(r,[1,r.shape[0],r.shape[1],r.shape[2]]));let[]=t,o={images:s},l={alignCorners:n,halfPixelCenters:a,size:t},c=M.runKernel(wc,o,l);return i?j(c,[c.shape[1],c.shape[2],c.shape[3]]):c}var G1=P({resizeNearestNeighbor_:bP});function xP(e,t,n){A(t%1==0,()=>`bandPart(): numLower must be an integer, got ${t}.`),A(n%1==0,()=>`bandPart(): numUpper must be an integer, got ${n}.`);let a=E(e,"a","bandPart");A(a.rank>=2,()=>`bandPart(): Rank must be at least 2, got ${a.rank}.`);let r=a.shape,[s,i]=a.shape.slice(-2);if(!(t<=s))throw new Error(`bandPart(): numLower (${t}) must not be greater than the number of rows (${s}).`);if(!(n<=i))throw new Error(`bandPart(): numUpper (${n}) must not be greater than the number of columns (${i}).`);t<0&&(t=s),n<0&&(n=i);let o=j(kh(0,s,1,"int32"),[-1,1]),l=kh(0,i,1,"int32"),c=fe(o,l),u=pa(Si(c,de(+t,"int32")),Jr(c,de(-n,"int32"))),p=vt([s,i],a.dtype);return j($t(dt(j(a,[-1,s,i])).map(d=>vn(u,d,p))),r)}var vP=P({bandPart_:xP});function wP(e){let t;if(Array.isArray(e)){t=!1,A(e!=null&&e.length>0,()=>"Gram-Schmidt process: input must not be null, undefined, or empty");let r=e[0].shape[0];for(let s=1;s<e.length;++s)A(e[s].shape[0]===r,()=>`Gram-Schmidt: Non-unique lengths found in the input vectors: (${e[s].shape[0]} vs. ${r})`)}else t=!0,e=Mn(e,e.shape[0],0).map(r=>Qr(r,[0]));A(e.length<=e[0].shape[0],()=>`Gram-Schmidt: Number of vectors (${e.length}) exceeds number of dimensions (${e[0].shape[0]}).`);let n=[],a=e;for(let r=0;r<e.length;++r)n.push(M.tidy(()=>{let s=a[r];if(r>0)for(let i=0;i<r;++i){let o=L(Te(L(n[i],s)),n[i]);s=fe(s,o)}return ve(s,Dh(s,"euclidean"))}));return t?$t(n,0):n}var kP=P({gramSchmidt_:wP});function IP(e,t=!1){if(A(e.rank>=2,()=>`qr() requires input tensor to have a rank >= 2, but got rank ${e.rank}`),e.rank===2)return H1(e,t);{let n=e.shape.slice(0,e.shape.length-2).reduce((l,c)=>l*c),a=dt(j(e,[n,e.shape[e.shape.length-2],e.shape[e.shape.length-1]]),0),r=[],s=[];a.forEach(l=>{let[c,u]=H1(l,t);r.push(c),s.push(u)});let i=j($t(r,0),e.shape),o=j($t(s,0),e.shape);return[i,o]}}function H1(e,t=!1){return M.tidy(()=>{A(e.shape.length===2,()=>`qr2d() requires a 2D Tensor, but got a ${e.shape.length}D Tensor.`);let n=e.shape[0],a=e.shape[1],r=Vy(n),s=jr(e),i=Ta([[1]],[1,1]),o=jr(i),l=n>=a?a:n;for(let c=0;c<l;++c){let u=s,p=o,d=r;[o,s,r]=M.tidy(()=>{let h=We(s,[c,c],[n-c,1]),m=Dh(h),f=We(s,[c,c],[1,1]),g=vn(ca(f,0),Ta([[-1]]),Ta([[1]])),y=fe(f,L(g,m)),b=ve(h,y);b.shape[0]===1?o=jr(i):o=Qe([i,We(b,[1,0],[b.shape[0]-1,b.shape[1]])],0);let v=Tt(ve(ze(g,y),m)),x=We(s,[c,0],[n-c,a]),N=L(v,o),T=Ue(o);if(c===0)s=fe(x,ze(N,ze(T,x)));else{let F=fe(x,ze(N,ze(T,x)));s=Qe([We(s,[0,0],[c,a]),F],0)}let C=Ue(N),$=We(r,[0,c],[n,r.shape[1]-c]);if(c===0)r=fe($,ze(ze($,o),C));else{let F=fe($,ze(ze($,o),C));r=Qe([We(r,[0,0],[n,c]),F],1)}return[o,s,r]}),Fe([u,p,d])}return!t&&n>a&&(r=We(r,[0,0],[n,a]),s=We(s,[0,0],[a,a])),[r,s]})}var NP=P({qr_:IP}),dn;(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"})(dn||(dn={}));function TP(e,t,n=dn.SUM_BY_NONZERO_WEIGHTS){let a=E(e,"losses","computeWeightedLoss"),r=null;t!=null&&(r=E(t,"weights","computeWeightedLoss"));let s=r==null?a:L(a,r);if(n===dn.NONE)return s;if(n===dn.SUM)return Te(s);if(n===dn.MEAN){if(r==null)return St(s);{let i=a.size/r.size,o=ve(Te(s),Te(r));return i>1?ve(o,de(i)):o}}if(n===dn.SUM_BY_NONZERO_WEIGHTS){if(r==null)return ve(Te(s),de(a.size));{let i=L(r,Xa(a.shape)),o=ce(Te(_i(i,de(0))),"float32");return ve(Te(s),o)}}throw Error(`Unknown reduction: ${n}`)}var yr=P({computeWeightedLoss_:TP});function SP(e,t,n,a=dn.SUM_BY_NONZERO_WEIGHTS){let r=E(e,"labels","absoluteDifference"),s=E(t,"predictions","absoluteDifference"),i=null;n!=null&&(i=E(n,"weights","absoluteDifference")),sn(r.shape,s.shape,"Error in absoluteDifference: ");let o=Lt(fe(r,s));return yr(o,i,a)}var CP=P({absoluteDifference_:SP});function _P(e,t,n,a,r=dn.SUM_BY_NONZERO_WEIGHTS){let s=E(e,"labels","cosineDistance"),i=E(t,"predictions","cosineDistance"),o=null;a!=null&&(o=E(a,"weights","cosineDistance")),sn(s.shape,i.shape,"Error in cosineDistance: ");let l=de(1),c=fe(l,Te(L(s,i),n,!0));return yr(c,o,r)}var EP=P({cosineDistance_:_P});function AP(e,t,n,a=dn.SUM_BY_NONZERO_WEIGHTS){let r=E(e,"labels","hingeLoss"),s=E(t,"predictions","hingeLoss"),i=null;n!=null&&(i=E(n,"weights","hingeLoss")),sn(r.shape,s.shape,"Error in hingeLoss: ");let o=de(1);r=fe(L(de(2),r),o);let l=Ye(fe(o,L(r,s)));return yr(l,i,a)}var FP=P({hingeLoss_:AP});function $P(e,t,n,a=1,r=dn.SUM_BY_NONZERO_WEIGHTS){let s=E(e,"labels","huberLoss"),i=E(t,"predictions","huberLoss"),o=null;n!=null&&(o=E(n,"weights","huberLoss")),sn(s.shape,i.shape,"Error in huberLoss: ");let l=de(a),c=Lt(fe(i,s)),u=Bl(c,l),p=fe(c,u),d=Z(L(de(.5),lt(u)),L(l,p));return yr(d,o,r)}var DP=P({huberLoss_:$P});function RP(e,t,n,a=1e-7,r=dn.SUM_BY_NONZERO_WEIGHTS){let s=E(e,"labels","logLoss"),i=E(t,"predictions","logLoss"),o=null;n!=null&&(o=E(n,"weights","logLoss")),sn(s.shape,i.shape,"Error in logLoss: ");let l=de(1),c=de(a),u=Tt(L(s,$n(Z(i,c)))),p=L(fe(l,s),$n(Z(fe(l,i),c))),d=fe(u,p);return yr(d,o,r)}var MP=P({logLoss_:RP});function PP(e,t,n,a=dn.SUM_BY_NONZERO_WEIGHTS){let r=E(e,"labels","meanSquaredError"),s=E(t,"predictions","meanSquaredError"),i=null;n!=null&&(i=E(n,"weights","meanSquaredError")),sn(r.shape,s.shape,"Error in meanSquaredError: ");let o=Ah(r,s);return yr(o,i,a)}var OP=P({meanSquaredError_:PP});function LP(e,t){let n=E(e,"labels","sigmoidCrossEntropyWithLogits"),a=E(t,"logits","sigmoidCrossEntropyWithLogits");sn(n.shape,a.shape,"Error in sigmoidCrossEntropyWithLogits: ");let r=Ye(a),s=L(a,n),i=gh(pn(Tt(Lt(a))));return Z(fe(r,s),i)}function zP(e,t,n,a=0,r=dn.SUM_BY_NONZERO_WEIGHTS){let s=E(e,"multiClassLabels","sigmoidCrossEntropy"),i=E(t,"logits","sigmoidCrossEntropy"),o=null;if(n!=null&&(o=E(n,"weights","sigmoidCrossEntropy")),sn(s.shape,i.shape,"Error in sigmoidCrossEntropy: "),a>0){let c=de(a),u=de(1),p=de(.5);s=Z(L(s,fe(u,c)),L(p,c))}let l=LP(s,i);return yr(l,o,r)}var BP=P({sigmoidCrossEntropy_:zP});function WP(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 qa((a,r,s)=>{let i=jy(r,[n],!0),o=fe(ce(r,"float32"),i);s([a,o]);let l=Tt(L(o,a));return{value:Te(l,[n]),gradFunc:(c,u)=>{let[p,d]=u,h=Ci(c.shape,[n]);return[L(j(c,h),fe(ce(p,"float32"),pn(d))),L(j(c,h),fe(pn(d),ce(p,"float32")))]}}})(e,t)}function VP(e,t,n,a=0,r=dn.SUM_BY_NONZERO_WEIGHTS){let s=E(e,"onehotLabels","softmaxCrossEntropy"),i=E(t,"logits","softmaxCrossEntropy"),o=null;if(n!=null&&(o=E(n,"weights","softmaxCrossEntropy")),sn(s.shape,i.shape,"Error in softmaxCrossEntropy: "),a>0){let c=de(a),u=de(1),p=de(s.shape[1]);s=Z(L(s,fe(u,c)),ve(c,p))}let l=WP(s,i);return yr(l,o,r)}var UP=P({softmaxCrossEntropy_:VP}),GP={fft:jc,ifft:Ul,rfft:qc,irfft:Eh},HP={hammingWindow:VM,hannWindow:O1,frame:L1,stft:jM},Ya={flipLeftRight:YM,resizeNearestNeighbor:G1,resizeBilinear:U1,rotateWithOffset:JM,cropAndResize:KM,nonMaxSuppression:eP,nonMaxSuppressionAsync:lP,nonMaxSuppressionWithScore:cP,nonMaxSuppressionWithScoreAsync:dP,nonMaxSuppressionPadded:mP,nonMaxSuppressionPaddedAsync:gP},j1={bandPart:vP,gramSchmidt:kP,qr:NP},jP={absoluteDifference:CP,computeWeightedLoss:yr,cosineDistance:EP,hingeLoss:FP,huberLoss:DP,logLoss:MP,meanSquaredError:OP,sigmoidCrossEntropy:BP,softmaxCrossEntropy:UP},br=class extends j0{minimize(e,t=!1,n){let{value:a,grads:r}=this.computeGradients(e,n);if(n!=null){let s=n.map(i=>({name:i.name,tensor:r[i.name]}));this.applyGradients(s)}else this.applyGradients(r);return Fe(r),t?a:(a.dispose(),null)}get iterations(){return this.iterations_==null&&(this.iterations_=0),this.iterations_}incrementIterations(){this.iterations_=this.iterations+1}computeGradients(e,t){return y1(e,t)}dispose(){this.iterations_!=null&&Fe(this.iterations_)}async saveIterations(){return this.iterations_==null&&(this.iterations_=0),{name:"iter",tensor:de(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 ${this.getClassName()}`)}async extractIterations(e){return this.iterations_=(await e[0].tensor.data())[0],e.slice(1)}};Object.defineProperty(br,Symbol.hasInstance,{value:e=>e.minimize!=null&&e.computeGradients!=null&&e.applyGradients!=null});var Lh=class extends br{constructor(e,t,n=null){super();this.learningRate=e,this.rho=t,this.epsilon=n,this.accumulatedGrads=[],this.accumulatedUpdates=[],n==null&&(this.epsilon=M.backend.epsilon())}applyGradients(e){(Array.isArray(e)?e.map(t=>t.name):Object.keys(e)).forEach((t,n)=>{let a=M.registeredVariables[t],r=!1;this.accumulatedGrads[n]==null&&(this.accumulatedGrads[n]={originalName:`${t}/accum_grad`,variable:D(()=>Ge(a).variable(r))}),this.accumulatedUpdates[n]==null&&(this.accumulatedUpdates[n]={originalName:`${t}/accum_var`,variable:D(()=>Ge(a).variable(r))});let s=Array.isArray(e)?e[n].tensor:e[t];if(s==null)return;let i=this.accumulatedGrads[n].variable,o=this.accumulatedUpdates[n].variable;D(()=>{let l=Z(L(i,this.rho),L(lt(s),1-this.rho)),c=L(ve(en(Z(o,this.epsilon)),en(Z(i,this.epsilon))),s),u=Z(L(o,this.rho),L(lt(c),1-this.rho));i.assign(l),o.assign(u);let p=Z(L(c,-this.learningRate),a);a.assign(p)})}),this.incrementIterations()}dispose(){this.accumulatedUpdates!=null&&(Fe(this.accumulatedGrads.map(e=>e.variable)),Fe(this.accumulatedUpdates.map(e=>e.variable)))}async getWeights(){let e=[...this.accumulatedGrads,...this.accumulatedUpdates];return[await this.saveIterations()].concat(e.map(t=>({name:t.originalName,tensor:t.variable})))}async setWeights(e){e=await this.extractIterations(e);let t=e.length/2,n=!1;this.accumulatedGrads=e.slice(0,t).map(a=>({originalName:a.name,variable:a.tensor.variable(n)})),this.accumulatedUpdates=e.slice(t,t*2).map(a=>({originalName:a.name,variable:a.tensor.variable(n)}))}getConfig(){return{learningRate:this.learningRate,rho:this.rho,epsilon:this.epsilon}}static fromConfig(e,t){return new e(t.learningRate,t.rho,t.epsilon)}};Lh.className="Adadelta";qr(Lh);var zh=class extends br{constructor(e,t=.1){super();this.learningRate=e,this.initialAccumulatorValue=t,this.accumulatedGrads=[]}applyGradients(e){(Array.isArray(e)?e.map(t=>t.name):Object.keys(e)).forEach((t,n)=>{let a=M.registeredVariables[t];if(this.accumulatedGrads[n]==null){let i=!1;this.accumulatedGrads[n]={originalName:`${t}/accumulator`,variable:D(()=>wn(a.shape,this.initialAccumulatorValue).variable(i))}}let r=Array.isArray(e)?e[n].tensor:e[t];if(r==null)return;let s=this.accumulatedGrads[n].variable;D(()=>{let i=Z(s,lt(r));s.assign(i);let o=Z(L(ve(r,en(Z(i,M.backend.epsilon()))),-this.learningRate),a);a.assign(o)})}),this.incrementIterations()}dispose(){this.accumulatedGrads!=null&&Fe(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)}};zh.className="Adagrad";qr(zh);var Bh=class extends br{constructor(e,t,n,a=null){super();this.learningRate=e,this.beta1=t,this.beta2=n,this.epsilon=a,this.accumulatedFirstMoment=[],this.accumulatedSecondMoment=[],D(()=>{this.accBeta1=de(t).variable(),this.accBeta2=de(n).variable()}),a==null&&(this.epsilon=M.backend.epsilon())}applyGradients(e){let t=Array.isArray(e)?e.map(n=>n.name):Object.keys(e);D(()=>{let n=fe(1,this.accBeta1),a=fe(1,this.accBeta2);t.forEach((r,s)=>{let i=M.registeredVariables[r],o=!1;this.accumulatedFirstMoment[s]==null&&(this.accumulatedFirstMoment[s]={originalName:`${r}/m`,variable:D(()=>Ge(i).variable(o))}),this.accumulatedSecondMoment[s]==null&&(this.accumulatedSecondMoment[s]={originalName:`${r}/v`,variable:D(()=>Ge(i).variable(o))});let l=Array.isArray(e)?e[s].tensor:e[r];if(l==null)return;let c=this.accumulatedFirstMoment[s].variable,u=this.accumulatedSecondMoment[s].variable,p=Z(L(c,this.beta1),L(l,1-this.beta1)),d=Z(L(u,this.beta2),L(lt(l),1-this.beta2)),h=ve(p,n),m=ve(d,a);c.assign(p),u.assign(d);let f=Z(L(ve(h,Z(en(m),this.epsilon)),-this.learningRate),i);i.assign(f)}),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&&Fe(this.accumulatedFirstMoment.map(e=>e.variable)),this.accumulatedSecondMoment!=null&&Fe(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),D(()=>{this.accBeta1.assign(gr(this.beta1,this.iterations_+1)),this.accBeta2.assign(gr(this.beta2,this.iterations_+1))});let t=e.length/2,n=!1;this.accumulatedFirstMoment=e.slice(0,t).map(a=>({originalName:a.name,variable:a.tensor.variable(n)})),this.accumulatedSecondMoment=e.slice(t,t*2).map(a=>({originalName:a.name,variable:a.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)}};Bh.className="Adam";qr(Bh);var Wh=class extends br{constructor(e,t,n,a=null,r=0){super();this.learningRate=e,this.beta1=t,this.beta2=n,this.epsilon=a,this.decay=r,this.accumulatedFirstMoment=[],this.accumulatedWeightedInfNorm=[],D(()=>{this.iteration=de(0).variable(),this.accBeta1=de(t).variable()}),a==null&&(this.epsilon=M.backend.epsilon())}applyGradients(e){let t=Array.isArray(e)?e.map(n=>n.name):Object.keys(e);D(()=>{let n=fe(1,this.accBeta1),a=ve(-this.learningRate,Z(L(this.iteration,this.decay),1));t.forEach((r,s)=>{let i=M.registeredVariables[r],o=!1;this.accumulatedFirstMoment[s]==null&&(this.accumulatedFirstMoment[s]={originalName:`${r}/m`,variable:Ge(i).variable(o)}),this.accumulatedWeightedInfNorm[s]==null&&(this.accumulatedWeightedInfNorm[s]={originalName:`${r}/v`,variable:Ge(i).variable(o)});let l=Array.isArray(e)?e[s].tensor:e[r];if(l==null)return;let c=this.accumulatedFirstMoment[s].variable,u=this.accumulatedWeightedInfNorm[s].variable,p=Z(L(c,this.beta1),L(l,1-this.beta1)),d=L(u,this.beta2),h=Lt(l),m=Ka(d,h);c.assign(p),u.assign(m);let f=Z(L(ve(a,n),ve(p,Z(m,this.epsilon))),i);i.assign(f)}),this.iteration.assign(Z(this.iteration,1)),this.accBeta1.assign(L(this.accBeta1,this.beta1))}),this.incrementIterations()}dispose(){this.accBeta1.dispose(),this.iteration.dispose(),this.accumulatedFirstMoment!=null&&Fe(this.accumulatedFirstMoment.map(e=>e.variable)),this.accumulatedWeightedInfNorm!=null&&Fe(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)}};Wh.className="Adamax";qr(Wh);var Kc=class extends br{constructor(e){super();this.learningRate=e,this.setLearningRate(e)}applyGradients(e){(Array.isArray(e)?e.map(t=>t.name):Object.keys(e)).forEach((t,n)=>{let a=Array.isArray(e)?e[n].tensor:e[t];if(a==null)return;let r=M.registeredVariables[t];D(()=>{let s=Z(L(this.c,a),r);r.assign(s)})}),this.incrementIterations()}setLearningRate(e){this.learningRate=e,this.c!=null&&this.c.dispose(),this.c=jt(de(-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)}};Kc.className="SGD";qr(Kc);var Vh=class extends Kc{constructor(e,t,n=!1){super(e);this.learningRate=e,this.momentum=t,this.useNesterov=n,this.accumulations=[],this.m=de(this.momentum)}applyGradients(e){(Array.isArray(e)?e.map(t=>t.name):Object.keys(e)).forEach((t,n)=>{let a=M.registeredVariables[t];if(this.accumulations[n]==null){let i=!1;this.accumulations[n]={originalName:`${t}/momentum`,variable:D(()=>Ge(a).variable(i))}}let r=this.accumulations[n].variable,s=Array.isArray(e)?e[n].tensor:e[t];s!=null&&D(()=>{let i,o=Z(L(this.m,r),s);this.useNesterov?i=Z(L(this.c,Z(s,L(o,this.m))),a):i=Z(L(this.c,o),a),r.assign(o),a.assign(i)})}),this.incrementIterations()}dispose(){this.m.dispose(),this.accumulations!=null&&Fe(this.accumulations.map(e=>e.variable))}setMomentum(e){this.momentum=e}async getWeights(){return[await this.saveIterations()].concat(this.accumulations.map(e=>({name:e.originalName,tensor:e.variable})))}async setWeights(e){e=await this.extractIterations(e);let t=!1;this.accumulations=e.map(n=>({originalName:n.name,variable:n.tensor.variable(t)}))}getConfig(){return{learningRate:this.learningRate,momentum:this.momentum,useNesterov:this.useNesterov}}static fromConfig(e,t){return new e(t.learningRate,t.momentum,t.useNesterov)}};Vh.className="Momentum";qr(Vh);var Uh=class extends br{constructor(e,t=.9,n=0,a=null,r=!1){super();if(this.learningRate=e,this.decay=t,this.momentum=n,this.epsilon=a,this.accumulatedMeanSquares=[],this.accumulatedMoments=[],this.accumulatedMeanGrads=[],this.centered=r,a==null&&(this.epsilon=M.backend.epsilon()),e==null)throw new Error("learningRate for RMSPropOptimizer must be defined.")}applyGradients(e){(Array.isArray(e)?e.map(t=>t.name):Object.keys(e)).forEach((t,n)=>{let a=M.registeredVariables[t],r=!1;this.accumulatedMeanSquares[n]==null&&(this.accumulatedMeanSquares[n]={originalName:`${t}/rms`,variable:D(()=>Ge(a).variable(r))}),this.accumulatedMoments[n]==null&&(this.accumulatedMoments[n]={originalName:`${t}/momentum`,variable:D(()=>Ge(a).variable(r))}),this.accumulatedMeanGrads[n]==null&&this.centered&&(this.accumulatedMeanGrads[n]={originalName:`${t}/mg`,variable:D(()=>Ge(a).variable(r))});let s=Array.isArray(e)?e[n].tensor:e[t];if(s==null)return;let i=this.accumulatedMeanSquares[n].variable,o=this.accumulatedMoments[n].variable;D(()=>{let l=Z(L(i,this.decay),L(lt(s),1-this.decay));if(this.centered){let c=this.accumulatedMeanGrads[n].variable,u=Z(L(c,this.decay),L(s,1-this.decay)),p=ve(L(s,this.learningRate),en(fe(l,Z(lt(u),this.epsilon)))),d=Z(L(o,this.momentum),p);i.assign(l),c.assign(u),o.assign(d);let h=fe(a,d);a.assign(h)}else{let c=Z(L(i,this.decay),L(lt(s),1-this.decay)),u=Z(L(o,this.momentum),ve(L(s,this.learningRate),en(Z(c,this.epsilon))));i.assign(c),o.assign(u);let p=fe(a,u);a.assign(p)}})}),this.incrementIterations()}dispose(){this.accumulatedMeanSquares!=null&&Fe(this.accumulatedMeanSquares.map(e=>e.variable)),this.accumulatedMeanGrads!=null&&this.centered&&Fe(this.accumulatedMeanGrads.map(e=>e.variable)),this.accumulatedMoments!=null&&Fe(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(a=>({originalName:a.name,variable:a.tensor.variable(n)})),this.accumulatedMoments=e.slice(t,t*2).map(a=>({originalName:a.name,variable:a.tensor.variable(n)})),this.centered&&(this.accumulatedMeanGrads=e.slice(t*2,t*3).map(a=>({originalName:a.name,variable:a.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)}};Uh.className="RMSProp";qr(Uh);var Ai=class{static sgd(e){return new Kc(e)}static momentum(e,t,n=!1){return new Vh(e,t,n)}static rmsprop(e,t=.9,n=0,a=null,r=!1){return new Uh(e,t,n,a,r)}static adam(e=.001,t=.9,n=.999,a=null){return new Bh(e,t,n,a)}static adadelta(e=.001,t=.95,n=null){return new Lh(e,t,n)}static adamax(e=.002,t=.9,n=.999,a=null,r=0){return new Wh(e,t,n,a,r)}static adagrad(e,t=.1){return new zh(e,t)}},Fi={sgd:Ai.sgd,momentum:Ai.momentum,adadelta:Ai.adadelta,adagrad:Ai.adagrad,rmsprop:Ai.rmsprop,adamax:Ai.adamax,adam:Ai.adam},qP=(()=>typeof requestAnimationFrame!="undefined"?requestAnimationFrame:typeof setImmediate!="undefined"?setImmediate:e=>e())();function Gh(){return new Promise(e=>qP(()=>e()))}var _={};Oe(_,{ERF_A1:()=>rO,ERF_A2:()=>sO,ERF_A3:()=>iO,ERF_A4:()=>oO,ERF_A5:()=>lO,ERF_P:()=>aO,PARALLELIZE_THRESHOLD:()=>pb,SELU_SCALE:()=>K1,SELU_SCALEALPHA:()=>q1,applyActivation:()=>Ph,assertAndGetBroadcastShape:()=>xt,assertAxesAreInnerMostDims:()=>$D,assertParamsConsistent:()=>KP,assignToTypedArray:()=>gO,axesAreInnerMostDims:()=>Gy,calculateShapes:()=>R0,combineLocations:()=>x1,complexWithEvenIndex:()=>hO,complexWithOddIndex:()=>mO,computeConv2DInfo:()=>Pc,computeConv3DInfo:()=>t1,computeDefaultPad:()=>$y,computeDilation2DInfo:()=>n$,computeOptimalWindowSize:()=>YP,computeOutAndReduceShapes:()=>v1,computeOutShape:()=>XP,computePool2DInfo:()=>e1,computePool3DInfo:()=>a$,convertConv2DDataFormat:()=>Q0,eitherStridesOrDilationsAreOne:()=>Ha,expandShapeToKeepDim:()=>Ci,exponent:()=>bO,exponents:()=>yO,fromStringArrayToUint8:()=>wO,fromUint8ToStringArray:()=>vO,getAxesPermutation:()=>w1,getBroadcastDims:()=>K$,getComplexWithIndex:()=>fO,getFusedBiasGradient:()=>Mh,getFusedDyActivation:()=>Rh,getImageCenter:()=>ZP,getInnerMostAxes:()=>DD,getPermuted:()=>QP,getReductionAxes:()=>zt,getReshaped:()=>JP,getReshapedPermuted:()=>eO,getSliceBeginCoords:()=>tO,getSliceSize:()=>nO,getUndoAxesPermutation:()=>Hy,log:()=>cO,mergeRealAndImagArrays:()=>pO,prepareAndValidate:()=>D0,prepareSplitSize:()=>xO,segment_util:()=>X1,shouldFuse:()=>Oh,slice_util:()=>cn,splitRealAndImagArrays:()=>dO,tupleValuesAreOne:()=>Xr,upcastType:()=>la,validateInput:()=>by,validateUpdateShape:()=>yy,warn:()=>uO});function KP(e,t){let n=e[0].length;e.forEach((r,s)=>{A(r.length===n,()=>`Error in concat${n}D: rank of tensors[${s}] must be the same as the rank of the rest (${n})`)}),A(t>=0&&t<n,()=>`Error in concat${n}D: axis must be between 0 and ${n-1}.`);let a=e[0];e.forEach((r,s)=>{for(let i=0;i<n;i++)A(i===t||r[i]===a[i],()=>`Error in concat${n}D: Shape of tensors[${s}] (${r}) does not match the shape of the rest (${a}) along the non-concatenated axis ${s}.`)})}function XP(e,t){let n=e[0].slice();for(let a=1;a<e.length;a++)n[t]+=e[a][t];return n}var pb=30;function YP(e){return e<=pb?e:fd(e,Math.floor(Math.sqrt(e)))}function ZP(e,t,n){let a=n*(typeof e=="number"?e:e[0]),r=t*(typeof e=="number"?e:e[1]);return[a,r]}function JP(e,t,n,a=!0){let r=[];if(a)r=r.concat(t.slice(0)),r.push(e[0]/n),r=r.concat(e.slice(1));else{r=r.concat(e[0]);let s=t.length;for(let i=0;i<s;++i)r=r.concat([e[i+1]/t[i],t[i]]);r=r.concat(e.slice(s+1))}return r}function QP(e,t,n=!0){let a=[];if(n){a.push(t);for(let r=t+1;r<e;++r)r<=2*t?(a.push(r),a.push(r-(t+1))):a.push(r)}else{let r=[],s=[];for(let i=1;i<e;++i)i>=t*2+1||i%2==1?s.push(i):r.push(i);a.push(...r),a.push(0),a.push(...s)}return a}function eO(e,t,n,a=!0){let r=[];a?r.push(e[0]/n):r.push(e[0]*n);for(let s=1;s<e.length;++s)s<=t.length?a?r.push(t[s-1]*e[s]):r.push(e[s]/t[s-1]):r.push(e[s]);return r}function tO(e,t){let n=[0];for(let a=0;a<t;++a)n.push(e[a][0]);return n}function nO(e,t,n){let a=e.slice(0,1);for(let r=0;r<n;++r)a.push(e[r+1]-t[r][0]-t[r][1]);return a}var q1=1.7580993408473768,K1=1.0507009873554805,aO=.3275911,rO=.254829592,sO=-.284496736,iO=1.421413741,oO=-1.453152027,lO=1.061405429;function uO(...e){ee().getBool("IS_TEST")||console.warn(...e)}function cO(...e){ee().getBool("IS_TEST")||console.log(...e)}function pO(e,t){if(e.length!==t.length)throw new Error(`Cannot merge real and imag arrays of different lengths. real:${e.length}, imag: ${t.length}.`);let n=new Float32Array(e.length*2);for(let a=0;a<n.length;a+=2)n[a]=e[a/2],n[a+1]=t[a/2];return n}function dO(e){let t=new Float32Array(e.length/2),n=new Float32Array(e.length/2);for(let a=0;a<e.length;a+=2)t[a/2]=e[a],n[a/2]=e[a+1];return{real:t,imag:n}}function hO(e){let t=Math.ceil(e.length/4),n=new Float32Array(t),a=new Float32Array(t);for(let r=0;r<e.length;r+=4)n[Math.floor(r/4)]=e[r],a[Math.floor(r/4)]=e[r+1];return{real:n,imag:a}}function mO(e){let t=Math.floor(e.length/4),n=new Float32Array(t),a=new Float32Array(t);for(let r=2;r<e.length;r+=4)n[Math.floor(r/4)]=e[r],a[Math.floor(r/4)]=e[r+1];return{real:n,imag:a}}function fO(e,t){let n=e[t*2],a=e[t*2+1];return{real:n,imag:a}}function gO(e,t,n,a){e[a*2]=t,e[a*2+1]=n}function yO(e,t){let n=new Float32Array(e/2),a=new Float32Array(e/2);for(let r=0;r<Math.ceil(e/2);r++){let s=(t?2:-2)*Math.PI*(r/e);n[r]=Math.cos(s),a[r]=Math.sin(s)}return{real:n,imag:a}}function bO(e,t,n){let a=(n?2:-2)*Math.PI*(e/t),r=Math.cos(a),s=Math.sin(a);return{real:r,imag:s}}function xO(e,t,n=0){let a=[];if(typeof t=="number")A(e.shape[n]%t==0,()=>"Number of splits must evenly divide the axis."),a=new Array(t).fill(e.shape[n]/t);else{let r=t.reduce((i,o)=>(o===-1&&(i+=1),i),0);A(r<=1,()=>"There should be only one negative value in split array.");let s=t.indexOf(-1);if(s!==-1){let i=t.reduce((o,l)=>l>0?o+l:o);t[s]=e.shape[n]-i}A(e.shape[n]===t.reduce((i,o)=>i+o),()=>"The sum of sizes must match the size of the axis dimension."),a=t}return a}var X1={};Oe(X1,{collectGatherOpShapeInfo:()=>NO,computeOutShape:()=>IO,segOpComputeOptimalWindowSize:()=>kO});function kO(e,t){let n=!1,a;for(e<=pb?(a=e,n=!0):a=fd(e,Math.floor(Math.sqrt(e)));!n;)a>t||a===e?n=!0:a=fd(e,a+1);return a}function IO(e,t,n){let a=[],r=e.length;for(let s=0;s<r;s++)s!==t?a.push(e[s]):a.push(n);return a}function NO(e,t,n,a){let r=t.shape.length,s=e.shape.length;if(a!==0&&(a<-r||a>r))throw new Error(`Expect batchDims in the range of [-${r}, ${r}], but got ${a}`);if(a<0&&(a+=r),a>s)throw new Error(`batchDims (${a}) must be less than rank(x) (
|
|
${s}).`);if(n<a)throw new Error(`batchDims (${a}) must be less than or equal to axis (${n}).`);for(let p=0;p<a;++p)if(e.shape[p]!==t.shape[p])throw new Error(`x.shape[${p}]: ${e.shape[p]} should be equal to indices.shape[${p}]: ${t.shape[p]}.`);let i=e.shape[n],o=[],l=1,c=1,u=1;for(let p=0;p<a;++p)o.push(e.shape[p]),l*=e.shape[p];for(let p=a;p<n;p++)o.push(e.shape[p]),c*=e.shape[p];for(let p=a;p<r;p++)o.push(t.shape[p]);for(let p=n+1;p<s;p++)o.push(e.shape[p]),u*=e.shape[p];return{batchSize:l,sliceSize:u,outerSize:c,dimSize:i,outputShape:o}}function vO(e){try{return e.map(t=>Zd(t))}catch(t){throw new Error(`Failed to decode encoded string bytes into utf-8, error: ${t}`)}}function wO(e){return e.map(t=>Cc(t))}var Za={};Oe(Za,{nonMaxSuppressionV3Impl:()=>z1,nonMaxSuppressionV4Impl:()=>B1,nonMaxSuppressionV5Impl:()=>W1,whereImpl:()=>E1});var Y1={kernelName:So,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(e,Gl(ce(n,"float32"),-1))}}},TO={kernelName:Co,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>{let a=lt(ce(n,"float32")),r=en(fe(de(1),a));return Tt(ve(e,r))}}}},SO={kernelName:_o,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>{let a=en(fe(lt(ce(n,"float32")),1));return ve(e,a)}}}},CO={kernelName:Lr,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,a]=t,r=xt(n.shape,a.shape);return{a:()=>{let s=e,i=zt(n.shape,r);return i.length>0&&(s=Te(s,i)),j(s,n.shape)},b:()=>{let s=e,i=zt(a.shape,r);return i.length>0&&(s=Te(s,i)),j(s,a.shape)}}}},_O={kernelName:Ss,saveAllInputs:!0,gradFunc:(e,t)=>{let n={};return t.forEach((a,r)=>{n[r]=()=>e.clone()}),n}},EO={kernelName:Cs,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>Ge(n)}}},AO={kernelName:lc,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>Ge(n)}}},FO={kernelName:Eo,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>ve(e,en(fe(de(1),lt(ce(n,"float32")))))}}},$O={kernelName:Ao,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>{let a=en(Z(de(1),lt(ce(n,"float32"))));return ve(e,a)}}}},DO={kernelName:Do,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,a]=t,r=xt(n.shape,a.shape);return{a:()=>{let s=Z(lt(n),lt(a)),i=L(e,ve(a,s)),o=zt(n.shape,r);return o.length>0&&(i=Te(i,o)),j(i,n.shape)},b:()=>{let s=Z(lt(n),lt(a)),i=Tt(L(e,ve(n,s))),o=zt(a.shape,r);return o.length>0&&(i=Te(i,o)),j(i,a.shape)}}}},RO={kernelName:Fo,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>ve(e,Z(lt(ce(n,"float32")),1))}}},MO={kernelName:$o,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>ve(e,fe(de(1),lt(ce(n,"float32"))))}}};function PO(e,t,n,a,r,s){let i=E(e,"dy","avgPool3dGrad"),o=E(t,"input","avgPool3dGrad"),l=i,c=o,u=!1;o.rank===4&&(u=!0,l=j(i,[1,i.shape[0],i.shape[1],i.shape[2],i.shape[3]]),c=j(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]])),A(l.rank===5,()=>`Error in avgPool3dGrad: dy must be rank 5 but got rank ${l.rank}.`),A(c.rank===5,()=>`Error in avgPool3dGrad: input must be rank 5 but got rank ${c.rank}.`),s!=null&&A(Gt(r),()=>`Error in avgPool3dGrad: pad must be an integer when using, dimRoundingMode ${s} but got pad ${r}.`);let p={dy:l,input:c},d={filterSize:n,strides:a,pad:r,dimRoundingMode:s},h=M.runKernel(vd,p,d);return u?j(h,[h.shape[1],h.shape[2],h.shape[3],h.shape[4]]):h}var OO=P({avgPool3dGrad_:PO}),LO={kernelName:uc,inputsToSave:["x"],gradFunc:(e,t,n)=>{let[a]=t,{filterSize:r,strides:s,pad:i,dimRoundingMode:o}=n;return{x:()=>OO(e,a,r,s,i,o)}}};function zO(e,t,n,a,r){let s=E(e,"dy","avgPoolGrad"),i=E(t,"input","avgPoolGrad");A(i.rank===s.rank,()=>`Rank of input (${i.rank}) does not match rank of dy (${s.rank})`);let o=i,l=s,c=!1;i.rank===3&&(c=!0,o=j(i,[1,i.shape[0],i.shape[1],i.shape[2]]),l=j(s,[1,s.shape[0],s.shape[1],s.shape[2]])),A(l.rank===4,()=>`Error in avgPoolGrad: dy must be rank 4 but got rank ${l.rank}.`),A(o.rank===4,()=>`Error in avgPoolGrad: input must be rank 4 but got rank ${o.rank}.`);let u={dy:l,input:o},p={filterSize:n,strides:a,pad:r},d=M.runKernel(xd,u,p);return c?j(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var BO=P({avgPoolGrad_:zO}),WO={kernelName:_s,inputsToSave:["x"],gradFunc:(e,t,n)=>{let[a]=t,{filterSize:r,strides:s,pad:i}=n;return{x:()=>BO(e,a,r,s,i)}}},VO={kernelName:Es,inputsToSave:["a","b"],gradFunc:(e,t,n)=>{let[a,r]=t,{transposeA:s,transposeB:i}=n;return!s&&!i?{a:()=>ze(e,r,!1,!0),b:()=>ze(a,e,!0,!1)}:!s&&i?{a:()=>ze(e,r,!1,!1),b:()=>ze(e,a,!0,!1)}:s&&!i?{a:()=>ze(r,e,!1,!0),b:()=>ze(a,e,!1,!1)}:{a:()=>ze(r,e,!0,!0),b:()=>ze(e,a,!0,!0)}}},UO={kernelName:cc,gradFunc:(e,t,n)=>{let{blockShape:a,crops:r}=n;return{x:()=>Vc(e,a,r)}}},GO={kernelName:r0,gradFunc:(e,t,n)=>{let a=n,r=a.inputShape,s=a.shape,i=Array.from(s);for(let l=r.length-1;l>=0;l--)if(r[l]===s[l])i[l]=1;else if(r[l]!==1)throw new Error(`broadcastTo(): [${r}] cannot be broadcast to [${s}].`);let o=[];for(let l=0;l<i.length;l++)i[l]>1&&o.push(l);return{x:()=>Te(e,o,!0)}}},HO={kernelName:As,gradFunc:e=>({x:()=>e.clone()})},jO={kernelName:Ro,gradFunc:e=>({x:()=>Ge(e)})},qO={kernelName:zr,inputsToSave:["x"],gradFunc:(e,t,n)=>{let[a]=t,{clipValueMin:r,clipValueMax:s}=n;return{x:()=>vn(pa(Jr(a,r),Si(a,s)),e,Ge(e))}}},KO={kernelName:pc,inputsToSave:["x"],gradFunc:Y1.gradFunc},XO={kernelName:Mo,saveAllInputs:!0,gradFunc:(e,t,n)=>{let a=t.map(o=>o.shape),{axis:r}=n,s=oa(r,t[0].shape)[0],i=a.map(o=>o[s]);return Mn(e,i,s).map(o=>()=>o)}},YO={kernelName:Fs,inputsToSave:["x","filter"],gradFunc:(e,t,n)=>{let[a,r]=t,{dilations:s,strides:i,pad:o,dataFormat:l}=n;return A(Xr(s),()=>`Error in gradient of conv2D: dilation rates greater than 1 are not yet supported in gradients. Got dilations '${s}'`),{x:()=>My(a.shape,e,r,i,o,l),filter:()=>ub(a,e,r.shape,i,o,l)}}},ZO={kernelName:$s,inputsToSave:["dy","filter"],gradFunc:(e,t,n)=>{let[a,r]=t,{strides:s,pad:i,dataFormat:o,dimRoundingMode:l}=n;return{dy:()=>At(e,r,s,i,o,1,l),filter:()=>ub(e,a,r.shape,s,i,o,l)}}};function JO(e,t,n,a,r){let s=e;e.rank===4&&(s=j(e,[1,e.shape[0],e.shape[1],e.shape[2],e.shape[3]]));let i=t;i.rank===4&&(i=j(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]])),A(s.rank===5,()=>`Error in conv3dDerFilter: input must be rank 5, but got shape ${s.shape}.`),A(i.rank===5,()=>`Error in conv3dDerFilter: dy must be rank 5, but got shape ${i.shape}.`),A(n.length===5,()=>`Error in conv3dDerFilter: filterShape must be length 5, but got ${n}.`),A(s.shape[4]===n[3],()=>`Error in conv3dDerFilter: depth of input ${s.shape[4]}) must match input depth in filter (${n[3]}.`),A(i.shape[4]===n[4],()=>`Error in conv3dDerFilter: depth of dy (${i.shape[4]}) must match output depth for filter (${n[4]}).`);let o={x:s,dy:i},l={strides:a,pad:r,filterShape:n};return M.runKernel(Nd,o,l)}var QO=P({conv3DBackpropFilter_:JO}),eL={kernelName:dc,inputsToSave:["x","filter"],gradFunc:(e,t,n)=>{let{dilations:a,strides:r,pad:s}=n;A(Xr(a),()=>`Error in gradient of conv3D: dilation rates greater than 1 are not yet supported in gradients. Got dilations '${a}'`);let[i,o]=t;return{x:()=>c1(i.shape,e,o,r,s),filter:()=>QO(i,e,o.shape,r,s)}}},tL={kernelName:Ds,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(Tt(Sh(ce(n,"float32"))),e)}}},nL={kernelName:Po,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(Ch(ce(n,"float32")),e)}}},aL={kernelName:Rs,inputsToSave:["x"],gradFunc:(e,t,n)=>{let[a]=t,{axis:r,exclusive:s,reverse:i}=n;return{x:()=>{let o=w1([r],a.rank),l=hh(e,r,s,!i);return o!=null&&(l=Ue(l,o)),l}}}},rL={kernelName:Ms,inputsToSave:["x","filter"],gradFunc:(e,t,n)=>{let{dilations:a,strides:r,pad:s,dimRoundingMode:i}=n,o=a==null?[1,1]:a;A(Xr(o),()=>`Error in gradient of depthwiseConv2dNative: dilation rates greater than 1 are not yet supported. Got dilations '${o}'`);let[l,c]=t;return A(l.rank===4,()=>`Error in gradient of depthwiseConv2dNative: input must be rank 4, but got rank ${l.rank}.`),A(c.rank===4,()=>`Error in gradient of depthwiseConv2dNative: filter must be rank 4, but got rank ${c.rank}.`),A(l.shape[3]===c.shape[2],()=>`Error in gradient of depthwiseConv2d: number of input channels (${l.shape[3]}) must match the inChannels dimension in filter ${c.shape[2]}.`),A(Ha(r,o),()=>`Error in gradient of depthwiseConv2d: Either strides or dilations must be 1. Got strides ${r} and dilations '${o}'.`),i!=null&&A(Gt(s),()=>`Error in depthwiseConv2d: pad must be an integer when using, dimRoundingMode ${i} but got pad ${s}.`),{x:()=>P1(l.shape,e,c,r,s,a,i),filter:()=>M1(l,e,c.shape,r,s,a,i)}}},sL={kernelName:hc,inputsToSave:["x","filter"],gradFunc:(e,t,n)=>{let[a,r]=t,s={x:a,filter:r,dy:e},i={x:a,filter:r,dy:e};return{x:()=>M.runKernel(Ad,s,n),filter:()=>M.runKernel(Fd,i,n)}}},iL={kernelName:zo,outputsToSave:[!0],gradFunc:(e,t)=>{let[n]=t,a={dy:e,y:n};return{x:()=>M.runKernel($d,a)}}},oL={kernelName:Bo,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t,a=L(pn(Tt(lt(n))),2/Math.sqrt(Math.PI));return{x:()=>L(e,a)}}},lL={kernelName:Os,outputsToSave:[!0],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(e,n)}}},uL={kernelName:Vo,inputsToSave:["input"],gradFunc:(e,t)=>{let[n]=t;return{input:()=>j(e,n.shape)}}},cL={kernelName:Uo,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(e,pn(n))}}},pL={kernelName:Ls,gradFunc:e=>({x:()=>Ge(e)})},dL={kernelName:zs,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,a]=t,r=xt(n.shape,a.shape);return{a:()=>{let s=ve(e,ce(a,"float32")),i=zt(n.shape,r);return i.length>0?j(Te(s,i),n.shape):s},b:()=>{let s=L(e,ce(n,"float32")),i=zt(a.shape,r);i.length>0&&(s=j(Te(s,i),a.shape));let o=lt(a);return Tt(ve(s,ce(o,"float32")))}}}},hL={kernelName:Bs,inputsToSave:["x","mean","variance","scale"],gradFunc:(e,t,n)=>{let{varianceEpsilon:a}=n,[r,s,i,o]=t,l=o==null?de(1):o,c=zt(s.shape,r.shape),u=[];if(s.rank===1){for(let f=0;f<r.shape.length-1;++f)u.push(r.shape[f]);u.push(1)}let p=fe(r,s),d=L(e,l),h=Nh(Z(i,de(a))),m=L(L(L(h,h),h),de(-.5));return{x:()=>s.rank===1?j(L(L(e,ja(j(h,[1,1,1,s.shape[0]]),u)),l),r.shape):j(L(L(e,h),l),r.shape),mean:()=>{let f=L(L(h,de(-1)),d);return s.rank===1&&(f=Te(f,c)),j(f,s.shape)},variance:()=>{let f=L(L(m,p),d);return s.rank===1&&(f=Te(f,c)),j(f,s.shape)},scale:()=>{let f=L(p,h),g=L(e,f);return s.rank===1&&(g=Te(g,c)),j(g,s.shape)},offset:()=>{let f=e;return s.rank===1&&(f=Te(f,c)),j(f,s.shape)}}}},mL={kernelName:Ho,inputsToSave:["x","indices"],gradFunc:(e,t,n)=>{let[a,r]=t,{axis:s}=n,i=oa(s,a.shape)[0];return{x:()=>{let o=a.shape,l=r.size,c=o.slice(0,i),u=c.length,p=o.slice(s,o.length).slice(1),d=p.length,h=Z1(0,u),m=Z1(u+1,u+1+d),f=J1([c,[l],p]),g=j(e,f),y=j(r,[l]),b=J1([[u],h,m]),v=Ue(g,b),x=sb(v,y,a.shape[i]),N=Hy(b);return x=Ue(x,N),x},indices:()=>r}}};function Z1(e,t){let n=[];for(let a=e;a<t;++a)n.push(a);return n}function J1(e){let t=[];for(let n=0;n<e.length;++n)for(let a=0;a<e[n].length;++a)t.push(e[n][a]);return t}var fL={kernelName:Ws,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,a]=t;return{a:()=>Ge(n),b:()=>Ge(a)}}},gL={kernelName:Ko,gradFunc:e=>({x:()=>ce(e,"float32")})},yL={kernelName:Xo,gradFunc:e=>({x:()=>Ge(e)})},bL={kernelName:Yo,gradFunc:e=>({x:()=>Ge(e)})},xL={kernelName:Zo,gradFunc:e=>({x:()=>Ge(e)})},vL={kernelName:Vs,inputsToSave:["x"],gradFunc:(e,t,n)=>{let[a]=t,{alpha:r}=n,s=ca(a,0);return{x:()=>vn(s,e,L(e,r))}}},wL={kernelName:el,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>ve(e,Z(n,1))}}},kL={kernelName:Us,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>ve(e,ce(n,"float32"))}}},IL={kernelName:s0,inputsToSave:[],outputsToSave:[!0],gradFunc:(e,t,n)=>{let[a]=t,{axis:r}=n;return{logits:()=>{let s=!0,i=pn(a);return fe(e,L(Te(e,r,s),i))}}}};function NL(e,t,n,a=5,r=1,s=1,i=.5){let o={x:e,y:t,dy:n},l={depthRadius:a,bias:r,alpha:s,beta:i};return M.runKernel(Od,o,l)}var TL=P({localResponseNormalizationBackprop_:NL}),SL={kernelName:yc,inputsToSave:["x"],outputsToSave:[!0],gradFunc:(e,t,n)=>{let[a,r]=t,{depthRadius:s,bias:i,alpha:o,beta:l}=n;return{x:()=>TL(a,r,e,s,i,o,l)}}};function Q1(e,t,n,a){return t.rank<n.rank&&(t=j(t,Ci(t.shape,a))),e.rank<n.rank&&(e=j(e,Ci(e.shape,a))),{x:()=>L(e,ce(Zr(n,t),e.dtype))}}var ek={kernelName:Gs,inputsToSave:["x"],outputsToSave:[!0],gradFunc:(e,t,n)=>{let a=n,{reductionIndices:r}=a,s=t[0],i=t[1],o=oa(r,s.shape),l=Q1(e,i,s,o);return{x:()=>l.x()}}},CL={kernelName:Hs,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,a]=t;return{a:()=>L(e,ce(Jr(n,a),"float32")),b:()=>L(e,ce(fh(n,a),"float32"))}}};function _L(e,t,n,a,r,s,i){let o=E(e,"dy","maxPool3dGrad"),l=E(t,"input","maxPool3dGrad"),c=E(n,"output","maxPool3dGrad"),u=o,p=l,d=c,h=!1;l.rank===4&&(h=!0,u=j(o,[1,o.shape[0],o.shape[1],o.shape[2],o.shape[3]]),p=j(l,[1,l.shape[0],l.shape[1],l.shape[2],l.shape[3]]),d=j(c,[1,c.shape[0],c.shape[1],c.shape[2],c.shape[3]])),A(u.rank===5,()=>`Error in maxPool3dGrad: dy must be rank 5 but got rank ${u.rank}.`),A(p.rank===5,()=>`Error in maxPool3dGrad: input must be rank 5 but got rank ${p.rank}.`),A(d.rank===5,()=>`Error in maxPool3dGrad: output must be rank 5 but got rank ${d.rank}.`),i!=null&&A(Gt(s),()=>`Error in maxPool3dGrad: pad must be an integer when using, dimRoundingMode ${i} but got pad ${s}.`);let m={dy:u,input:p,output:d},f={filterSize:a,strides:r,pad:s,dimRoundingMode:i},g=M.runKernel(zd,m,f);return h?j(g,[g.shape[1],g.shape[2],g.shape[3],g.shape[4]]):g}var EL=P({maxPool3dGrad_:_L}),AL={kernelName:bc,inputsToSave:["x"],outputsToSave:[!0],gradFunc:(e,t,n)=>{let[a,r]=t,{filterSize:s,strides:i,pad:o,dimRoundingMode:l}=n;return{x:()=>EL(e,a,r,s,i,o,l)}}};function FL(e,t,n,a,r,s,i){let o=E(e,"dy","maxPoolGrad"),l=E(t,"input","maxPoolGrad"),c=E(n,"output","maxPoolGrad");A(l.rank===o.rank,()=>`Rank of input (${l.rank}) does not match rank of dy (${o.rank})`),A(o.rank===4,()=>`Error in maxPoolGrad: dy must be rank 4 but got rank ${o.rank}.`),A(l.rank===4,()=>`Error in maxPoolGrad: input must be rank 4 but got rank ${l.rank}.`),i!=null&&A(Gt(s),()=>`Error in maxPoolGrad: pad must be an integer when using, dimRoundingMode ${i} but got pad ${s}.`);let u={dy:o,input:l,output:c},p={filterSize:a,strides:r,pad:s,dimRoundingMode:i};return M.runKernel(Ld,u,p)}var $L=P({maxPoolGrad_:FL}),DL={kernelName:js,inputsToSave:["x"],outputsToSave:[!0],gradFunc:(e,t,n)=>{let[a,r]=t,{filterSize:s,strides:i,pad:o}=n;return{x:()=>$L(e,a,r,s,i,o)}}},RL={kernelName:qs,inputsToSave:["x"],gradFunc:(e,t,n)=>{let[a]=t,{axis:r}=n,s=oa(r,a.shape),i=v1(a.shape,s)[1],o=Pt(i);return{x:()=>{let l=a.shape.slice();s.forEach(u=>{l[u]=1});let c=j(e,l);return ve(L(c,Xa(a.shape,"float32")),o)}}}},ML={kernelName:Ks,inputsToSave:["x"],outputsToSave:[!0],gradFunc:(e,t,n)=>{let a=n,{axis:r}=a,[s,i]=t,o=oa(r,s.shape),l=Q1(e,i,s,o);return{x:()=>l.x()}}},PL={kernelName:Xs,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,a]=t;return{a:()=>L(e,ce(Si(n,a),"float32")),b:()=>L(e,ce(ca(n,a),"float32"))}}},OL={kernelName:xc,inputsToSave:["x"],gradFunc:(e,t,n)=>{let a=t[0],{paddings:r}=n,s=r.map(i=>i[0]);return{x:()=>We(e,s,a.shape)}}},LL={kernelName:nl,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,a]=t,r=xt(n.shape,a.shape);return{a:()=>{let s=zt(n.shape,r);return s.length>0?j(Te(e,s),n.shape):e},b:()=>{let s=L(e,Tt(Ol(ve(n,a)))),i=zt(a.shape,r);return i.length>0?j(Te(s,i),a.shape):s}}}},zL={kernelName:Ys,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,a]=t,r=xt(n.shape,a.shape);return{a:()=>{let s=L(e,ce(a,"float32")),i=zt(n.shape,r);return i.length>0?j(Te(s,i),n.shape):s},b:()=>{let s=L(e,ce(n,"float32")),i=zt(a.shape,r);return i.length>0?j(Te(s,i),a.shape):s}}}},BL={kernelName:al,gradFunc:e=>({x:()=>Tt(e)})},WL={kernelName:Zs,inputsToSave:["indices"],gradFunc:(e,t)=>{let n=t[0];return{indices:()=>vt(n.shape,"float32")}}},VL={kernelName:ll,gradFunc:e=>({x:()=>Ge(e)})},UL={kernelName:ul,saveAllInputs:!0,gradFunc:(e,t,n)=>{let{axis:a}=n;return dt(e,a).map(r=>()=>r)}},tk={kernelName:Js,inputsToSave:["x"],gradFunc:(e,t,n)=>{let a=t[0],{paddings:r}=n,s=r.map(i=>i[0]);return{x:()=>We(e,s,a.shape)}}},GL={kernelName:Qs,inputsToSave:["a","b"],outputsToSave:[!0],gradFunc:(e,t)=>{let[n,a,r]=t,s=n,i=a,o=xt(s.shape,i.shape);return{a:()=>{let l=ce(i,"float32"),c=L(e,L(l,gr(s,fe(l,de(1))))),u=zt(s.shape,o);return u.length>0&&(c=Te(c,u)),j(c,s.shape)},b:()=>{let l=ca(s,0),c=vn(l,$n(s),Ge(s)),u=L(e,L(r,c)),p=zt(i.shape,o);return p.length>0&&(u=Te(u,p)),j(u,i.shape)}}}},HL={kernelName:ei,inputsToSave:["x","alpha"],gradFunc:(e,t)=>{let[n,a]=t,r=ca(n,0);return{x:()=>vn(r,e,L(e,a)),alpha:()=>{let s=vn(r,Ge(e),L(e,n)),i=zt(a.shape,e.shape);return i.length>0&&(s=Te(s,i)),j(s,a.shape)}}}},jL={kernelName:Ps,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,a]=t,r=xt(n.shape,a.shape);return{a:()=>{let s=ve(e,ce(a,"float32")),i=zt(n.shape,r);return i.length>0?j(Te(s,i),n.shape):s},b:()=>{let s=L(e,ce(n,"float32")),i=zt(a.shape,r);i.length>0&&(s=j(Te(s,i),a.shape));let o=lt(a);return Tt(ve(s,ce(o,"float32")))}}}},qL={kernelName:pl,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>ve(e,Tt(lt(n)))}}},KL={kernelName:ai,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t,a=L(Si(n,6),Gl(n));return{x:()=>L(e,ce(a,"float32"))}}},XL={kernelName:ti,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(e,ce(Gl(n),"float32"))}}},YL={kernelName:dl,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>j(e,n.shape)}}},ZL={kernelName:ni,inputsToSave:["images"],gradFunc:(e,t,n)=>{let[a]=t,r={dy:e,images:a};return{images:()=>M.runKernel(Gd,r,n)}}},JL={kernelName:wc,inputsToSave:["images"],gradFunc:(e,t,n)=>{let[a]=t,r={dy:e,images:a};return{images:()=>M.runKernel(Ud,r,n)}}},QL={kernelName:ri,gradFunc:(e,t,n)=>{let{dims:a}=n,r=oa(a,e.shape);return{x:()=>Rn(e,r)}}},e3={kernelName:si,gradFunc:e=>({x:()=>Ge(e)})},t3={kernelName:ii,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>Tt(ve(e,L(gr(n,1.5),2)))}}},n3={kernelName:ml,inputsToSave:["condition"],gradFunc:(e,t)=>{let[n]=t;return{condition:()=>ce(Ge(n),"float32"),t:()=>L(e,ce(n,e.dtype)),e:()=>L(e,ce(Wc(n),e.dtype))}}},a3={kernelName:fl,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>{let a=ca(n,de(0)),r=de(q1),s=de(K1),i=L(e,s),o=L(L(e,r),pn(ce(n,"float32")));return vn(a,i,o)}}}},r3={kernelName:li,outputsToSave:[!0],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(e,L(n,fe(de(1),n)))}}},s3={kernelName:bl,gradFunc:e=>({x:()=>Ge(e)})},i3={kernelName:oi,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(zc(ce(n,"float32")),e)}}},o3={kernelName:yl,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(dh(ce(n,"float32")),e)}}},l3={kernelName:gl,inputsToSave:["x"],gradFunc:(e,t,n)=>{let[a]=t,{begin:r,size:s}=n,i=a.shape,[o,l]=U0(a,r,s),c=[];for(let u=0;u<e.rank;u++)c.push([o[u],i[u]-o[u]-l[u]]);return{x:()=>Zn(e,c)}}},u3={kernelName:pi,outputsToSave:[!0],gradFunc:(e,t,n)=>{let[a]=t,{dim:r}=n,s=!0,i=L(e,a);return{logits:()=>fe(i,L(Te(i,[r],s),a))}}},c3={kernelName:xl,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(e,ua(n))}}},nk={kernelName:kc,gradFunc:(e,t,n)=>{let{blockShape:a,paddings:r}=n;return{x:()=>Oc(e,a,r)}}},ak={kernelName:vl,gradFunc:(e,t,n)=>{let{axis:a}=n;return{x:()=>Qe(e,a)}}},p3={kernelName:ui,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>ve(e,L(en(ce(n,"float32")),2))}}},d3={kernelName:Ic,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(e,L(ce(n,"float32"),2))}}},h3={kernelName:di,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,a]=t,r=de(2);return{a:()=>L(e,L(r,fe(n,a))),b:()=>L(e,L(r,fe(a,n)))}}},m3={kernelName:Wr,gradFunc:e=>({x:()=>Ge(e)})},f3={kernelName:hi,inputsToSave:["a","b"],gradFunc:(e,t)=>{let[n,a]=t,r=xt(n.shape,a.shape);return{a:()=>{let s=e,i=zt(n.shape,r);return i.length>0&&(s=Te(s,i)),j(s,n.shape)},b:()=>{let s=e,i=zt(a.shape,r);return i.length>0&&(s=Te(s,i)),j(Tt(s),a.shape)}}}},g3={kernelName:ci,inputsToSave:["x"],gradFunc:(e,t,n)=>{let[a]=t,r=a.shape.slice(),{axis:s}=n;oa(s,a.shape).forEach(l=>{r[l]=1});let i=j(e,r),o=L(i,Xa(a.shape,"float32"));return{x:()=>o}}},y3={kernelName:kl,inputsToSave:["x"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>ve(e,lt(zc(n)))}}},b3={kernelName:mi,outputsToSave:[!0],gradFunc:(e,t)=>{let[n]=t;return{x:()=>L(fe(de(1),lt(n)),e)}}},x3={kernelName:Br,inputsToSave:["x"],gradFunc:(e,t,n)=>{let[a]=t,{reps:r}=n;return{x:()=>{let s=Ge(a);if(a.rank===1)for(let i=0;i<r[0];++i)s=Z(s,We(e,[i*a.shape[0]],[a.shape[0]]));else if(a.rank===2)for(let i=0;i<r[0];++i)for(let o=0;o<r[1];++o)s=Z(s,We(e,[i*a.shape[0],o*a.shape[1]],[a.shape[0],a.shape[1]]));else if(a.rank===3)for(let i=0;i<r[0];++i)for(let o=0;o<r[1];++o)for(let l=0;l<r[2];++l)s=Z(s,We(e,[i*a.shape[0],o*a.shape[1],l*a.shape[2]],[a.shape[0],a.shape[1],a.shape[2]]));else if(a.rank===4)for(let i=0;i<r[0];++i)for(let o=0;o<r[1];++o)for(let l=0;l<r[2];++l)for(let c=0;c<r[3];++c)s=Z(s,We(e,[i*a.shape[0],o*a.shape[1],l*a.shape[2],c*a.shape[3]],[a.shape[0],a.shape[1],a.shape[2],a.shape[3]]));else throw new Error(`Gradient for tile operation is not implemented for rank-${a.rank} tensors yet.`);return s}}}},v3={kernelName:fi,gradFunc:(e,t,n)=>{let a=n,{perm:r}=a,s=Hy(r);return{x:()=>Ue(e,s)}}},w3={kernelName:Nl,gradFunc:(e,t,n)=>{let a=n,{axis:r}=a;return{value:()=>$t(e,r)}}},I3={kernelName:Nc,inputsToSave:["segmentIds"],gradFunc:(e,t)=>{let[n]=t;return{x:()=>k3(e,n)}}};function k3(e,t){let n=Ka(t,Ge(t)),a=Ti(e,n),r=Jr(t,de(0,"int32")),s=a.rank-r.rank;for(let o=0;o<s;++o)r=Fn(r,o+1);r=pa(r,Xa(a.shape,"bool"));let i=Ge(a);return vn(r,a,i)}var N3={kernelName:Tl,gradFunc:e=>({x:()=>Ge(e)})},T3=[Y1,TO,SO,CO,_O,EO,AO,FO,$O,DO,RO,MO,LO,WO,VO,UO,GO,HO,jO,qO,KO,XO,ZO,YO,eL,tL,nL,aL,rL,sL,jL,iL,oL,lL,uL,cL,dL,pL,hL,mL,fL,gL,yL,bL,xL,vL,wL,kL,IL,SL,ek,ek,CL,AL,DL,RL,ML,PL,OL,LL,zL,BL,WL,VL,UL,tk,tk,GL,HL,qL,KL,XL,YL,ZL,JL,QL,e3,t3,n3,a3,r3,s3,i3,o3,l3,u3,c3,nk,nk,ak,ak,p3,h3,d3,m3,f3,g3,y3,b3,x3,v3,w3,I3,N3];for(let e of T3)i0(e);Y().prototype.abs=function(){return this.throwIfDisposed(),Lt(this)};Y().prototype.acos=function(){return this.throwIfDisposed(),Iy(this)};Y().prototype.acosh=function(){return this.throwIfDisposed(),Ny(this)};Y().prototype.add=function(e){return this.throwIfDisposed(),Z(this,e)};Y().prototype.all=function(e,t){return this.throwIfDisposed(),lh(this,e,t)};Y().prototype.any=function(e,t){return this.throwIfDisposed(),Rc(this,e,t)};Y().prototype.argMax=function(e){return this.throwIfDisposed(),Mc(this,e)};Y().prototype.argMin=function(e){return this.throwIfDisposed(),Ty(this,e)};Y().prototype.asScalar=function(){return this.throwIfDisposed(),A(this.size===1,()=>"The array must have only 1 element."),j(this,[])};Y().prototype.asType=function(e){return this.throwIfDisposed(),ce(this,e)};Y().prototype.as1D=function(){return this.throwIfDisposed(),j(this,[this.size])};Y().prototype.as2D=function(e,t){return this.throwIfDisposed(),j(this,[e,t])};Y().prototype.as3D=function(e,t,n){return this.throwIfDisposed(),j(this,[e,t,n])};Y().prototype.as4D=function(e,t,n,a){return this.throwIfDisposed(),j(this,[e,t,n,a])};Y().prototype.as5D=function(e,t,n,a,r){return this.throwIfDisposed(),j(this,[e,t,n,a,r])};Y().prototype.asin=function(){return this.throwIfDisposed(),Sy(this)};Y().prototype.asinh=function(){return this.throwIfDisposed(),Cy(this)};Y().prototype.atan=function(){return this.throwIfDisposed(),_y(this)};Y().prototype.atan2=function(e){return this.throwIfDisposed(),Ey(this,e)};Y().prototype.atanh=function(){return this.throwIfDisposed(),Ay(this)};Y().prototype.avgPool=function(e,t,n,a){return this.throwIfDisposed(),Xn(this,e,t,n,a)};Y().prototype.batchToSpaceND=function(e,t){return this.throwIfDisposed(),Oc(this,e,t)};Y().prototype.batchNorm=function(e,t,n,a,r){return this.throwIfDisposed(),fr(this,e,t,n,a,r)};Y().prototype.broadcastTo=function(e){return this.throwIfDisposed(),Lc(this,e)};Y().prototype.cast=function(e){return this.throwIfDisposed(),ce(this,e)};Y().prototype.ceil=function(){return this.throwIfDisposed(),Ry(this)};Y().prototype.clipByValue=function(e,t){return this.throwIfDisposed(),qt(this,e,t)};Y().prototype.concat=function(e,t){return this.throwIfDisposed(),e instanceof Ae&&(e=[e]),Qe([this,...e],t)};Y().prototype.conv1d=function(e,t,n,a,r,s){return this.throwIfDisposed(),ch(this,e,t,n,a,r,s)};Y().prototype.conv2dTranspose=function(e,t,n,a,r){return this.throwIfDisposed(),ph(this,e,t,n,a,r)};Y().prototype.conv2d=function(e,t,n,a,r,s){return this.throwIfDisposed(),At(this,e,t,n,a,r,s)};Y().prototype.cos=function(){return this.throwIfDisposed(),zc(this)};Y().prototype.cosh=function(){return this.throwIfDisposed(),dh(this)};Y().prototype.cumsum=function(e,t,n){return this.throwIfDisposed(),hh(this,e,t,n)};Y().prototype.depthToSpace=function(e,t){return this.throwIfDisposed(),Oy(this,e,t)};Y().prototype.depthwiseConv2d=function(e,t,n,a,r,s){return this.throwIfDisposed(),Yr(this,e,t,n,a,r,s)};Y().prototype.dilation2d=function(e,t,n,a,r){return this.throwIfDisposed(),Ly(this,e,t,n,a,r)};Y().prototype.divNoNan=function(e){return this.throwIfDisposed(),zy(this,e)};Y().prototype.div=function(e){return this.throwIfDisposed(),ve(this,e)};Y().prototype.dot=function(e){return this.throwIfDisposed(),d1(this,e)};Y().prototype.elu=function(){return this.throwIfDisposed(),Pl(this)};Y().prototype.equal=function(e){return this.throwIfDisposed(),Zr(this,e)};Y().prototype.erf=function(){return this.throwIfDisposed(),By(this)};Y().prototype.exp=function(){return this.throwIfDisposed(),pn(this)};Y().prototype.expandDims=function(e){return this.throwIfDisposed(),Fn(this,e)};Y().prototype.expm1=function(){return this.throwIfDisposed(),Wy(this)};Y().prototype.fft=function(){return this.throwIfDisposed(),jc(this)};Y().prototype.flatten=function(){return this.throwIfDisposed(),j(this,[this.size])};Y().prototype.floor=function(){return this.throwIfDisposed(),Ol(this)};Y().prototype.floorDiv=function(e){return this.throwIfDisposed(),oh(this,e)};Y().prototype.gather=function(e,t){return this.throwIfDisposed(),Ti(this,e,t)};Y().prototype.greaterEqual=function(e){return this.throwIfDisposed(),Jr(this,e)};Y().prototype.greater=function(e){return this.throwIfDisposed(),ca(this,e)};Y().prototype.ifft=function(){return this.throwIfDisposed(),Ul(this)};Y().prototype.irfft=function(){return this.throwIfDisposed(),Eh(this)};Y().prototype.isFinite=function(){return this.throwIfDisposed(),h1(this)};Y().prototype.isInf=function(){return this.throwIfDisposed(),m1(this)};Y().prototype.isNaN=function(){return this.throwIfDisposed(),f1(this)};Y().prototype.leakyRelu=function(e){return this.throwIfDisposed(),Bc(this,e)};Y().prototype.lessEqual=function(e){return this.throwIfDisposed(),Si(this,e)};Y().prototype.less=function(e){return this.throwIfDisposed(),fh(this,e)};Y().prototype.localResponseNormalization=function(e,t,n,a){return this.throwIfDisposed(),Uy(this,e,t,n,a)};Y().prototype.logSigmoid=function(){return this.throwIfDisposed(),b1(this)};Y().prototype.logSoftmax=function(e){return this.throwIfDisposed(),bh(this,e)};Y().prototype.logSumExp=function(e,t){return this.throwIfDisposed(),jy(this,e,t)};Y().prototype.log=function(){return this.throwIfDisposed(),$n(this)};Y().prototype.log1p=function(){return this.throwIfDisposed(),gh(this)};Y().prototype.logicalAnd=function(e){return this.throwIfDisposed(),pa(this,e)};Y().prototype.logicalNot=function(){return this.throwIfDisposed(),Wc(this)};Y().prototype.logicalOr=function(e){return this.throwIfDisposed(),xh(this,e)};Y().prototype.logicalXor=function(e){return this.throwIfDisposed(),k1(this,e)};Y().prototype.matMul=function(e,t,n){return this.throwIfDisposed(),ze(this,e,t,n)};Y().prototype.maxPool=function(e,t,n,a){return this.throwIfDisposed(),Ft(this,e,t,n,a)};Y().prototype.max=function(e,t){return this.throwIfDisposed(),Yn(this,e,t)};Y().prototype.maximum=function(e){return this.throwIfDisposed(),Ka(this,e)};Y().prototype.mean=function(e,t){return this.throwIfDisposed(),St(this,e,t)};Y().prototype.min=function(e,t){return this.throwIfDisposed(),zl(this,e,t)};Y().prototype.minimum=function(e){return this.throwIfDisposed(),Bl(this,e)};Y().prototype.mirrorPad=function(e,t){return this.throwIfDisposed(),Ky(this,e,t)};Y().prototype.mod=function(e){return this.throwIfDisposed(),Xy(this,e)};Y().prototype.mul=function(e){return this.throwIfDisposed(),L(this,e)};Y().prototype.neg=function(){return this.throwIfDisposed(),Tt(this)};Y().prototype.norm=function(e,t,n){return this.throwIfDisposed(),Dh(this,e,t,n)};Y().prototype.notEqual=function(e){return this.throwIfDisposed(),_i(this,e)};Y().prototype.oneHot=function(e,t=1,n=0){return this.throwIfDisposed(),$l(this,e,t,n)};Y().prototype.onesLike=function(){return this.throwIfDisposed(),Dn(this)};Y().prototype.pad=function(e,t){return this.throwIfDisposed(),Zn(this,e,t)};Y().prototype.pool=function(e,t,n,a,r){return this.throwIfDisposed(),T1(this,e,t,n,a,r)};Y().prototype.pow=function(e){return this.throwIfDisposed(),gr(this,e)};Y().prototype.prelu=function(e){return this.throwIfDisposed(),Uc(this,e)};Y().prototype.prod=function(e,t){return this.throwIfDisposed(),wh(this,e,t)};Y().prototype.reciprocal=function(){return this.throwIfDisposed(),Jy(this)};Y().prototype.relu=function(){return this.throwIfDisposed(),Ye(this)};Y().prototype.relu6=function(){return this.throwIfDisposed(),Ih(this)};Y().prototype.reshapeAs=function(e){return this.throwIfDisposed(),j(this,e.shape)};Y().prototype.reshape=function(e){return this.throwIfDisposed(),j(this,e)};Y().prototype.resizeBilinear=function(e,t,n){return this.throwIfDisposed(),U1(this,e,t,n)};Y().prototype.resizeNearestNeighbor=function(e,t,n){return this.throwIfDisposed(),G1(this,e,t,n)};Y().prototype.reverse=function(e){return this.throwIfDisposed(),Rn(this,e)};Y().prototype.rfft=function(){return this.throwIfDisposed(),qc(this)};Y().prototype.round=function(){return this.throwIfDisposed(),Qy(this)};Y().prototype.rsqrt=function(){return this.throwIfDisposed(),Nh(this)};Y().prototype.selu=function(){return this.throwIfDisposed(),Th(this)};Y().prototype.separableConv2d=function(e,t,n,a,r,s){return this.throwIfDisposed(),Ei(this,e,t,n,a,r,s)};Y().prototype.sigmoid=function(){return this.throwIfDisposed(),ua(this)};Y().prototype.sign=function(){return this.throwIfDisposed(),eb(this)};Y().prototype.sin=function(){return this.throwIfDisposed(),Sh(this)};Y().prototype.sinh=function(){return this.throwIfDisposed(),Ch(this)};Y().prototype.slice=function(e,t){return this.throwIfDisposed(),We(this,e,t)};Y().prototype.softmax=function(e){return this.throwIfDisposed(),Na(this,e)};Y().prototype.softplus=function(){return this.throwIfDisposed(),Ll(this)};Y().prototype.spaceToBatchND=function(e,t){return this.throwIfDisposed(),Vc(this,e,t)};Y().prototype.split=function(e,t){return this.throwIfDisposed(),Mn(this,e,t)};Y().prototype.sqrt=function(){return this.throwIfDisposed(),en(this)};Y().prototype.square=function(){return this.throwIfDisposed(),lt(this)};Y().prototype.squaredDifference=function(e){return this.throwIfDisposed(),Ah(this,e)};Y().prototype.squeeze=function(e){return this.throwIfDisposed(),Qr(this,e)};Y().prototype.stack=function(e,t){this.throwIfDisposed();let n=e instanceof Ae?[this,e]:[this,...e];return $t(n,t)};Y().prototype.step=function(e){return this.throwIfDisposed(),Gl(this,e)};Y().prototype.stridedSlice=function(e,t,n,a,r,s,i,o){return this.throwIfDisposed(),nb(this,e,t,n,a,r,s,i,o)};Y().prototype.sub=function(e){return this.throwIfDisposed(),fe(this,e)};Y().prototype.sum=function(e,t){return this.throwIfDisposed(),Te(this,e,t)};Y().prototype.tan=function(){return this.throwIfDisposed(),ab(this)};Y().prototype.tanh=function(){return this.throwIfDisposed(),Ml(this)};Y().prototype.tile=function(e){return this.throwIfDisposed(),ja(this,e)};Y().prototype.toBool=function(){return this.throwIfDisposed(),ce(this,"bool")};Y().prototype.toFloat=function(){return this.throwIfDisposed(),ce(this,"float32")};Y().prototype.toInt=function(){return this.throwIfDisposed(),ce(this,"int32")};Y().prototype.topk=function(e,t){return this.throwIfDisposed(),rb(this,e,t)};Y().prototype.transpose=function(e){return this.throwIfDisposed(),Ue(this,e)};Y().prototype.unique=function(e){return this.throwIfDisposed(),$h(this,e)};Y().prototype.unsortedSegmentSum=function(e,t){return this.throwIfDisposed(),sb(this,e,t)};Y().prototype.unstack=function(e){return this.throwIfDisposed(),dt(this,e)};Y().prototype.where=function(e,t){return this.throwIfDisposed(),vn(e,this,t)};Y().prototype.zerosLike=function(){return this.throwIfDisposed(),Ge(this)};var rk={};Oe(rk,{maxNorm:()=>S3,minMaxNorm:()=>E3,nonNeg:()=>_3,unitNorm:()=>C3});var db;function Bt(){return db==null&&(db=Z0().epsilon()),db}function Ca(){return"channelsLast"}var xr=class extends Error{constructor(e){super(e);Object.setPrototypeOf(this,xr.prototype)}},_a=class extends Error{constructor(e){super(e);Object.setPrototypeOf(this,_a.prototype)}},z=class extends Error{constructor(e){super(e);Object.setPrototypeOf(this,z.prototype)}},$e=class extends Error{constructor(e){super(e);Object.setPrototypeOf(this,$e.prototype)}},sk=class extends Error{constructor(e){super(e);Object.setPrototypeOf(this,sk.prototype)}},A3=class extends Error{constructor(e){super(e);Object.setPrototypeOf(this,A3.prototype)}};function $i(e,t){if(Array.isArray(e)){let n=[];for(let a=0;a<t;a++)n=n.concat(e);return n}else{let n=new Array(t);return n.fill(e),n}}function Ja(e,t){if(!e)throw new sk(t)}function ik(e,t){let n=0;for(let a of e)a===t&&n++;return n}function kn(e){return e.length===1?e[0]:e}function yt(e){return Array.isArray(e)?e:[e]}function vr(e){let t=e.replace(/(.)([A-Z][a-z0-9]+)/g,"$1_$2").replace(/([a-z])([A-Z])/g,"$1_$2").toLowerCase();return t[0]!=="_"?t:"private"+t}function Di(e){return e.length<=1||e.indexOf("_")===-1?e:e.replace(/[_]+(\w|$)/g,(t,n)=>n.toUpperCase())}var da={};function hb(e){if(e==null)return null;let t={};return t.className=e.getClassName(),t.config=e.getConfig(),t}function mb(e){if(!(e==null||typeof e!="object"))if(Array.isArray(e))e.forEach(t=>mb(t));else{let t=Object.keys(e);for(let n of t){let a=e[n];a!=null&&typeof a=="object"&&(!Array.isArray(a)&&a.type==="ndarray"&&typeof a.value=="number"?e[n]=a.value:mb(a))}}}function Xc(e,t={},n={},a="object",r=!1){if(typeof e=="string"){let s=e,i;if(s in n)i=n[s];else if(s in da)i=da[s];else if(i=t[s],i==null)throw new z(`Unknown ${a}: ${e}. This may be due to one of the following reasons:
|
|
1. The ${a} is defined in Python, in which case it needs to be ported to TensorFlow.js or your JavaScript code.
|
|
2. The custom ${a} is defined in JavaScript, but is not registered properly with tf.serialization.registerClass().`);return i}else{let s=e;if(s.className==null||s.config==null)throw new z(`${a}: Improper config format: ${JSON.stringify(s)}.
|
|
'className' and 'config' must set.`);let i=s.className,o,l;if(i in n?[o,l]=n[i]:i in da?[o,l]=da.className:i in t&&([o,l]=t[i]),o==null)throw new z(`Unknown ${a}: ${i}. This may be due to one of the following reasons:
|
|
1. The ${a} is defined in Python, in which case it needs to be ported to TensorFlow.js or your JavaScript code.
|
|
2. The custom ${a} is defined in JavaScript, but is not registered properly with tf.serialization.registerClass().`);if(l!=null){let c={};for(let h of Object.keys(da))c[h]=da[h];for(let h of Object.keys(n))c[h]=n[h];let u=s.config;u.customObjects=c;let p=Object.assign({},da);for(let h of Object.keys(n))da[h]=n[h];mb(s.config);let d=l(o,s.config,n,r);return da=Object.assign({},p),d}else{let c=Object.assign({},da);for(let p of Object.keys(n))da[p]=n[p];let u=new o(s.config);return da=Object.assign({},c),u}}}function F3(e,t){return e<t?-1:e>t?1:0}function Hh(e,t){return-1*F3(e,t)}function ts(e){if(e==null)return e;let t=[];for(let n of e)t.indexOf(n)===-1&&t.push(n);return t}function $3(e){if(e==null)throw new z(`Invalid value in obj: ${JSON.stringify(e)}`);for(let t in e)if(e.hasOwnProperty(t))return!1;return!0}function Ri(e,t,n){if(n!=null&&e.indexOf(n)<0)throw new z(`${n} is not a valid ${t}. Valid values are ${e} or null/undefined.`)}function fb(e,t,n=0,a=Infinity){return Ja(n>=0),Ja(a>=n),Array.isArray(e)&&e.length>=n&&e.length<=a&&e.every(r=>typeof r===t)}function Kt(e,t){Array.isArray(e)?(k.assert(e.length>0,()=>`${t} is unexpectedly an empty array.`),e.forEach((n,a)=>Kt(n,`element ${a+1} of ${t}`))):k.assert(Number.isInteger(e)&&e>0,()=>`Expected ${t} to be a positive integer, but got ${ok(e)}.`)}function ok(e){return e===null?"null":Array.isArray(e)?"["+e.map(t=>ok(t)).join(",")+"]":typeof e=="string"?`"${e}"`:`${e}`}function D3(e,t){let n=k.now(),a;return(...r)=>{let s=k.now();return s-n<t||(n=s,a=e(...r)),a}}function lk(e){return e==="relu"?"relu":e==="linear"?"linear":e==="elu"?"elu":null}function gb(e,t){return D(()=>en(Te(L(e,e),t,!0)))}var Yc=class extends re.Serializable{getConfig(){return{}}},yb=class extends Yc{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 D(()=>{let t=gb(e,this.axis),n=qt(t,0,this.maxValue);return L(e,ve(n,Z(Bt(),t)))})}getConfig(){return{maxValue:this.maxValue,axis:this.axis}}};yb.className="MaxNorm";re.registerClass(yb);var bb=class extends Yc{constructor(e){super();this.defaultAxis=0,this.axis=e.axis!=null?e.axis:this.defaultAxis}apply(e){return D(()=>ve(e,Z(Bt(),gb(e,this.axis))))}getConfig(){return{axis:this.axis}}};bb.className="UnitNorm";re.registerClass(bb);var xb=class extends Yc{apply(e){return Ye(e)}};xb.className="NonNeg";re.registerClass(xb);var vb=class extends Yc{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 D(()=>{let t=gb(e,this.axis),n=Z(L(this.rate,qt(t,this.minValue,this.maxValue)),L(1-this.rate,t));return L(e,ve(n,Z(Bt(),t)))})}getConfig(){return{minValue:this.minValue,maxValue:this.maxValue,rate:this.rate,axis:this.axis}}};vb.className="MinMaxNorm";re.registerClass(vb);var uk={maxNorm:"MaxNorm",minMaxNorm:"MinMaxNorm",nonNeg:"NonNeg",unitNorm:"UnitNorm"};function Wt(e){return hb(e)}function ck(e,t={}){return Xc(e,re.SerializationMap.getMap().classNameMap,t,"constraint")}function Vt(e){if(e==null)return null;if(typeof e=="string"){let t={className:e in uk?uk[e]:e,config:{}};return ck(t)}else return e instanceof Yc?e:ck(e)}function S3(e){return new yb(e)}function C3(e){return new bb(e)}function _3(){return new xb}function E3(e){return new vb(e)}var pk={};Oe(pk,{constant:()=>P3,glorotNormal:()=>U3,glorotUniform:()=>V3,heNormal:()=>G3,heUniform:()=>H3,identity:()=>B3,leCunNormal:()=>j3,leCunUniform:()=>q3,ones:()=>M3,orthogonal:()=>K3,randomNormal:()=>L3,randomUniform:()=>O3,truncatedNormal:()=>z3,varianceScaling:()=>W3,zeros:()=>R3});var X3=["channelsFirst","channelsLast"],Y3=["nearest","bilinear"],Z3=["valid","same","causal"],J3=["max","avg"],Q3=["sum","mul","concat","ave"],jl=new Map;function Dt(e){Ri(X3,"DataFormat",e)}function ez(e){Ri(Y3,"InterpolationFormat",e)}function Jn(e){Ri(Z3,"PaddingMode",e)}function dk(e){Ri(J3,"PoolMode",e)}var Zc=[],hk="/";function Mi(e,t){Zc.push(e);try{let n=t();return Zc.pop(),n}catch(n){throw Zc.pop(),n}}function tz(){return Zc.length===0?"":Zc.join(hk)+hk}function fk(e){if(!mk(e))throw new Error("Not a valid tensor name: '"+e+"'");return tz()+e}function gk(e){if(!mk(e))throw new Error("Not a valid tensor name: '"+e+"'");jl.has(e)||jl.set(e,0);let t=jl.get(e);if(jl.set(e,jl.get(e)+1),t>0){let n=`${e}_${t}`;return jl.set(n,1),n}else return e}var nz=new RegExp(/^[A-Za-z0-9][-A-Za-z0-9\._\/]*$/);function mk(e){return!!e.match(nz)}function az(e){return e===parseInt(e.toString(),10)}function ns(e,t,n){t==null&&(t=0),n==null&&(n=e.length);let a=1;for(let r=t;r<n;++r)a*=e[r];return a}function yk(e){return e=Array.isArray(e)?new Float32Array(e):e,tt(e)}function ql(e){return zl(yk(e)).dataSync()[0]}function as(e){return Yn(yk(e)).dataSync()[0]}function Ea(e,t){if(t<e)throw new z(`end (${t}) < begin (${e}) is forbidden.`);let n=[];for(let a=e;a<t;++a)n.push(a);return n}function Jc(e,t){return e.asType(t)}function Qc(e,t=-1){let n=e.shape.slice();return t<0&&(t=n.length+t+1),n.splice(t,0,1),e.reshape(n)}function rz(e,t){return D(()=>{if(e.shape.length!==2)throw new z(`repeat() expects a rank-2 tensor, but received a rank-${e.shape.length} tensor.`);let n=Qc(e,1);return wb(n,[1,t,1])})}function sz(e){let t=[ns(e.shape)];return e.reshape(t)}function iz(e){if(e.rank<=1)throw new z(`batchFlatten requires a minimum rank of 2. Got rank: ${e.rank}.`);let t=[e.shape[0],ns(e.shape,1)];return e.reshape(t)}function Pi(e,t,n){return D(()=>{switch(e.rank){case 1:return _h(e,t,n);case 2:return tb(e,[t,0],[n,e.shape[1]]);case 3:return Vl(e,[t,0,0],[n,e.shape[1],e.shape[2]]);case 4:return Hc(e,[t,0,0,0],[n,e.shape[1],e.shape[2],e.shape[3]]);case 5:return We(e,[t,0,0,0,0],[n,e.shape[1],e.shape[2],e.shape[3],e.shape[4]]);case 6:return We(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 z(`sliceAlongFirstAxis() received an unsupported tensor rank: ${e.rank}`)}})}function kb(e,t,n){return D(()=>{switch(e.rank){case 1:return _h(e,t,n);case 2:return tb(e,[0,t],[e.shape[0],n]);case 3:return Vl(e,[0,0,t],[e.shape[0],e.shape[1],n]);case 4:return Hc(e,[0,0,0,t],[e.shape[0],e.shape[1],e.shape[2],n]);default:throw new z(`sliceAlongLastAxis() received an unsupported tensor rank: ${e.rank}`)}})}function jh(e,t,n,a){return D(()=>{switch(e.rank){case 1:return _h(e,t,n);case 2:switch(a){case 1:return Pi(e,t,n);case 2:return kb(e,t,n);default:throw new z(`The axis is not within the rank of the tensor ${a}`)}case 3:switch(a){case 1:return Pi(e,t,n);case 2:return Vl(e,[0,t,0],[e.shape[0],n,e.shape[2]]);case 3:return kb(e,t,n);default:throw new z(`The axis is not within the rank of the tensor ${a}`)}case 4:switch(a){case 1:return Pi(e,t,n);case 2:return Hc(e,[0,t,0,0],[e.shape[0],n,e.shape[2],e.shape[3]]);case 3:return Hc(e,[0,0,t,0],[e.shape[0],e.shape[1],n,e.shape[3]]);case 4:return kb(e,t,n);default:throw new z(`The axis is not within the rank of the tensor ${a}`)}default:throw new z(`sliceAlongLastAxis() received an unsupported tensor rank: ${e.rank}`)}})}function Ib(e,t=-1){let n;return t<0&&(n=e[0].rank,n!==0?t=n:t=0),t===e[0].rank&&(t=-1),Qe(e,t)}function bk(e,t){switch(e.rank){case 1:return i1([e,t]);case 2:return o1([e,t],0);case 3:return l1([e,t],0);case 4:return u1([e,t],0);default:throw new z(`concatAlongFirstAxis() received an unsupported tensor rank: ${e.rank}`)}}function wb(e,t){if(Array.isArray(t)||(t=[t]),e.rank!==t.length)throw new z(`The length of input n (${t.length}) does not match the number of dimensions in input x (${e.rank})`);return ja(e,t)}function qh(e,t=0,n=1,a,r){return S1(e,t,n,a,r)}function Qa(e,t,n,a){if(e.rank<2||t.rank<2)throw new $e(`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],s=t.shape.slice(-2)[0];if(r!==s)throw new $e(`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,s=!1;return es.matMul({a:e,b:t,transposeA:r,transposeB:s,bias:a?Nb(e.rank,a,Ca()):null,activation:n})}else{let r=e.shape.slice(),s=r.pop();e=e.reshape([-1,s]);let i=t.shape.slice(),o=i.pop(),l=i.pop(),c=[...i,o],u=Array.from({length:t.rank},(m,f)=>f===0?t.rank-2:f<=t.rank-2?f-1:f);t=t.transpose(u).reshape([l,-1]);let p=[...r,...c],d=!1,h=!1;return es.matMul({a:e,b:t,transposeA:d,transposeB:h,bias:a?Nb(e.rank,a,Ca()):null,activation:n}).reshape(p)}}function xk(e,t,n){return D(()=>(Array.isArray(t)?t=tt(t,"int32"):t=t.toInt(),Ti(e,t,n)))}function ep(e){return L(e,e)}function Nb(e,t,n){let a=t.shape;if(t.rank!==1&&t.rank!==e)throw new z(`Unexpected bias dimensions: ${t.rank}; expected it to be 1 or ${e}`);if(e===5){if(n==="channelsFirst")return a.length===1?t.reshape([1,a[0],1,1,1]):t.reshape([1,a[3],a[0],a[1],a[2]]);if(n==="channelsLast")return a.length===1?t.reshape([1,1,1,1,a[0]]):t.reshape([1].concat(a))}else if(e===4){if(n==="channelsFirst")return a.length===1?t.reshape([1,a[0],1,1]):t.reshape([1,a[2],a[0],a[1]]);if(n==="channelsLast")return a.length===1?t.reshape([1,1,1,a[0]]):t.reshape([1].concat(a))}else if(e===3){if(n==="channelsFirst")return a.length===1?t.reshape([1,a[0],1]):t.reshape([1,a[1],a[0]]);if(n==="channelsLast")return a.length===1?t.reshape([1,1,a[0]]):t.reshape([1].concat(a))}else if(e<3)return t;throw new z(`Unsupported input rank by biasAdd: ${t.rank}`)}function er(e,t,n){return D(()=>(n==null&&(n=Ca()),Dt(n),e.add(Nb(e.rank,t,n))))}function oz(e,t=1){if(t!==1)throw new $e(`Support for alpha values other than 1 (${t}) is not implemented yet.`);return Pl(e)}function lz(e){return D(()=>ve(e,Lt(e).add(1)))}function vk(e,t,n,a){return D(()=>D1(e,t,n,a))}function uz(e){return D(()=>{let t=Z(.5,L(.2,e));return qt(t,0,1)})}function tp(e,t,n=!1){return n?e():t()}var cz=["fanIn","fanOut","fanAvg"],pz=["normal","uniform","truncatedNormal"];function dz(e){Ri(cz,"FanMode",e)}function hz(e){Ri(pz,"Distribution",e)}var ha=class extends re.Serializable{fromConfigUsesCustomObjects(){return!1}getConfig(){return{}}},Tb=class extends ha{apply(e,t){return vt(e,t)}};Tb.className="Zeros";re.registerClass(Tb);var Kh=class extends ha{apply(e,t){return Xa(e,t)}};Kh.className="Ones";re.registerClass(Kh);var Sb=class extends ha{constructor(e){super();if(typeof e!="object")throw new z(`Expected argument of type ConstantConfig but got ${e}`);if(e.value===void 0)throw new z(`config must have value set but got ${e}`);this.value=e.value}apply(e,t){return D(()=>L(de(this.value),Xa(e,t)))}getConfig(){return{value:this.value}}};Sb.className="Constant";re.registerClass(Sb);var Cb=class extends ha{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 Wl(e,this.minval,this.maxval,t)}getConfig(){return{minval:this.minval,maxval:this.maxval,seed:this.seed}}};Cb.className="RandomUniform";re.registerClass(Cb);var _b=class extends ha{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 $e(`randomNormal does not support dType ${t}.`);return qh(e,this.mean,this.stddev,t,this.seed)}getConfig(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}}};_b.className="RandomNormal";re.registerClass(_b);var Eb=class extends ha{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 $e(`truncatedNormal does not support dType ${t}.`);return Fh(e,this.mean,this.stddev,t,this.seed)}getConfig(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}}};Eb.className="TruncatedNormal";re.registerClass(Eb);var Ab=class extends ha{constructor(e){super();this.gain=e.gain!=null?e.gain:1}apply(e,t){return D(()=>{if(e.length!==2||e[0]!==e[1])throw new z("Identity matrix initializer can only be used for 2D square matrices.");return L(this.gain,Vy(e[0]))})}getConfig(){return{gain:this.gain}}};Ab.className="Identity";re.registerClass(Ab);function mz(e,t="channelsLast"){let n,a;if(Dt(t),e.length===2)n=e[0],a=e[1];else if([3,4,5].indexOf(e.length)!==-1){if(t==="channelsFirst"){let r=ns(e,2);n=e[1]*r,a=e[0]*r}else if(t==="channelsLast"){let r=ns(e,0,e.length-2);n=e[e.length-2]*r,a=e[e.length-1]*r}}else{let r=ns(e);n=Math.sqrt(r),a=Math.sqrt(r)}return[n,a]}var In=class extends ha{constructor(e){super();if(e.scale<0)throw new z(`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,dz(this.mode),this.distribution=e.distribution==null?"normal":e.distribution,hz(this.distribution),this.seed=e.seed}apply(e,t){let n=mz(e),a=n[0],r=n[1],s=this.scale;if(this.mode==="fanIn"?s/=Math.max(1,a):this.mode==="fanOut"?s/=Math.max(1,r):s/=Math.max(1,(a+r)/2),this.distribution==="normal"){let i=Math.sqrt(s);if(t=t||"float32",t!=="float32"&&t!=="int32")throw new $e(`${this.getClassName()} does not support dType ${t}.`);return Fh(e,0,i,t,this.seed)}else{let i=Math.sqrt(3*s);return Wl(e,-i,i,t)}}getConfig(){return{scale:this.scale,mode:this.mode,distribution:this.distribution,seed:this.seed}}};In.className="VarianceScaling";re.registerClass(In);var Xh=class extends In{constructor(e){super({scale:1,mode:"fanAvg",distribution:"uniform",seed:e==null?null:e.seed})}getClassName(){return In.className}};Xh.className="GlorotUniform";re.registerClass(Xh);var Yh=class extends In{constructor(e){super({scale:1,mode:"fanAvg",distribution:"normal",seed:e==null?null:e.seed})}getClassName(){return In.className}};Yh.className="GlorotNormal";re.registerClass(Yh);var Zh=class extends In{constructor(e){super({scale:2,mode:"fanIn",distribution:"normal",seed:e==null?null:e.seed})}getClassName(){return In.className}};Zh.className="HeNormal";re.registerClass(Zh);var Jh=class extends In{constructor(e){super({scale:2,mode:"fanIn",distribution:"uniform",seed:e==null?null:e.seed})}getClassName(){return In.className}};Jh.className="HeUniform";re.registerClass(Jh);var Qh=class extends In{constructor(e){super({scale:1,mode:"fanIn",distribution:"normal",seed:e==null?null:e.seed})}getClassName(){return In.className}};Qh.className="LeCunNormal";re.registerClass(Qh);var em=class extends In{constructor(e){super({scale:1,mode:"fanIn",distribution:"uniform",seed:e==null?null:e.seed})}getClassName(){return In.className}};em.className="LeCunNormal";re.registerClass(em);var Fb=class extends ha{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 $e("Random seed is not implemented for Orthogonal Initializer yet.")}apply(e,t){return D(()=>{if(e.length<2)throw new $e("Shape must be at least 2D.");e[0]*e[1]>2e3&&console.warn(`Orthogonal initializer is being called on a matrix with more than 2000 (${e[0]*e[1]}) elements: Slowness may result.`);let n=e[0]>e[1]?[e[1],e[0]]:e,a=qh(n,0,1,"float32"),r=j1.gramSchmidt(a);return e[0]>e[1]&&(r=r.transpose()),L(this.gain,r)})}getConfig(){return{gain:this.gain,seed:this.seed}}};Fb.className="Orthogonal";re.registerClass(Fb);var wk={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 kk(e,t={}){return Xc(e,re.SerializationMap.getMap().classNameMap,t,"initializer")}function Ct(e){return hb(e)}function wt(e){if(typeof e=="string"){let t=e in wk?wk[e]:e;if(t==="GlorotNormal")return new Yh;if(t==="GlorotUniform")return new Xh;if(t==="HeNormal")return new Zh;if(t==="HeUniform")return new Jh;if(t==="LeCunNormal")return new Qh;if(t==="LeCunUniform")return new em;{let n={};return n.className=t,n.config={},kk(n)}}else return e instanceof ha?e:kk(e)}function R3(){return new Tb}function M3(){return new Kh}function P3(e){return new Sb(e)}function O3(e){return new Cb(e)}function L3(e){return new _b(e)}function z3(e){return new Eb(e)}function B3(e){return new Ab(e)}function W3(e){return new In(e)}function V3(e){return new Xh(e)}function U3(e){return new Yh(e)}function G3(e){return new Zh(e)}function H3(e){return new Jh(e)}function j3(e){return new Qh(e)}function q3(e){return new em(e)}function K3(e){return new Fb(e)}var Ik={};Oe(Ik,{Layer:()=>qe,RNN:()=>tr,RNNCell:()=>np,activation:()=>Az,add:()=>zz,alphaDropout:()=>kB,average:()=>Bz,averagePooling1d:()=>$b,averagePooling2d:()=>Db,averagePooling3d:()=>Rb,avgPool1d:()=>Xz,avgPool2d:()=>Zz,avgPool3d:()=>Qz,avgPooling1d:()=>Yz,avgPooling2d:()=>Jz,avgPooling3d:()=>eB,batchNormalization:()=>jz,bidirectional:()=>mB,concatenate:()=>Wz,conv1d:()=>kz,conv2d:()=>Iz,conv2dTranspose:()=>Nz,conv3d:()=>Tz,convLstm2d:()=>cB,convLstm2dCell:()=>pB,cropping2D:()=>Cz,dense:()=>Fz,depthwiseConv2d:()=>Ez,dot:()=>Hz,dropout:()=>$z,elu:()=>gz,embedding:()=>Lz,flatten:()=>Rz,gaussianDropout:()=>wB,gaussianNoise:()=>vB,globalAveragePooling1d:()=>tB,globalAveragePooling2d:()=>nB,globalMaxPool1d:()=>gB,globalMaxPool2d:()=>yB,globalMaxPooling1d:()=>Tk,globalMaxPooling2d:()=>Sk,gru:()=>rB,gruCell:()=>sB,input:()=>Nk,inputLayer:()=>fz,layerNormalization:()=>qz,leakyReLU:()=>bz,lstm:()=>iB,lstmCell:()=>oB,masking:()=>IB,maxPool1d:()=>bB,maxPool2d:()=>xB,maxPooling1d:()=>Ck,maxPooling2d:()=>_k,maxPooling3d:()=>aB,maximum:()=>Vz,minimum:()=>Uz,multiply:()=>Gz,permute:()=>Oz,prelu:()=>xz,reLU:()=>yz,repeatVector:()=>Mz,reshape:()=>Pz,rnn:()=>dB,separableConv2d:()=>Sz,simpleRNN:()=>lB,simpleRNNCell:()=>uB,softmax:()=>vz,spatialDropout1d:()=>Dz,stackedRNNCells:()=>hB,thresholdedReLU:()=>wz,timeDistributed:()=>fB,upSampling2d:()=>_z,zeroPadding2d:()=>Kz});var NB=0;function Ek(){return NB++}var tm={};function nm(e=""){return e in tm||(tm[e]=0),tm[e]+=1,e+tm[e].toString()}function Mb(e){return Array.isArray(e)&&Array.isArray(e[0])}function am(e){return e.length===0?[]:Array.isArray(e[0])?e:[e]}function Me(e){let t;if(Array.isArray(e)){if(e.length!==1)throw new z(`Expected Tensor length to be 1; got ${e.length}`);t=e[0]}else t=e;return t}function ht(e){if(Array.isArray(e)&&Array.isArray(e[0])){if(e.length===1)return e=e,e[0];throw new z(`Expected exactly 1 Shape; got ${e.length}`)}else return e}function rm(e){let t=0;for(let n of e)n.shape.length===0?t+=1:t+=n.shape.reduce((a,r)=>a*r);return t}var Ak="Variable",Fk=class{constructor(e,t="float32",n=Ak,a=!0,r=null){this.dtype=t==null?"float32":t,this.shape=e.shape,this.id=Ek(),n=n==null?Ak:n,this.originalName=fk(n),this.name=gk(this.originalName),this.trainable_=a,this.constraint=r,this.val=_1(e,this.trainable_,this.name,this.dtype)}read(){return this.assertNotDisposed(),this.val}write(e){return this.assertNotDisposed(),TB(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 TB(e,t){if(e.shape.toString()!==t.shape.toString())throw new Error("Shape mismatch: "+JSON.stringify(e.shape)+" vs. "+JSON.stringify(t.shape))}function Pb(e){return e.map(t=>t.read())}function Ob(e){e.forEach(t=>{t[0].write(t[1])})}var Xt=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||{}}},Aa=class{constructor(e,t,n,a,r,s,i){this.dtype=e,this.shape=t,this.sourceLayer=n,this.inputs=a,this.callArgs=r,this.outputTensorIndex=i,this.id=Ek(),s!=null&&(this.originalName=fk(s),this.name=gk(this.originalName)),this.rank=t.length}},SB=0,sm=class{constructor(e,t){this.callArgs=t,this.id=SB++,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}}},CB=0,qe=class extends re.Serializable{constructor(e={}){super();this._callHook=null,this._addedWeightNames=[],this._stateful=!1,this.id=CB++,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=vr(n)+"_"+nm(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 a=e.dtype;a==null&&(a=e.inputDType),a==null&&(a="float32"),this.dtype=a}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 _a(`The layer has never been called and thus has no defined ${t}.`);if(this.inboundNodes.length<=e)throw new z(`Asked to get ${t} at node ${e}, but the layer has only ${this.inboundNodes.length} inbound nodes.`);return this.inboundNodes[e]}getInputAt(e){return kn(this.getNodeAtIndex(e,"input").inputTensors)}getOutputAt(e){return kn(this.getNodeAtIndex(e,"output").outputTensors)}get input(){if(this.inboundNodes.length>1)throw new xr(`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 xr(`Layer ${this.name} is not connected, no input to return.`);return kn(this.getNodeAtIndex(0,"input").inputTensors)}get output(){if(this.inboundNodes.length===0)throw new xr(`Layer ${this.name} has no inbound nodes.`);if(this.inboundNodes.length>1)throw new xr(`Layer ${this.name} has multiple inbound nodes, hence the notion of "layer output" is ill-defined. Use \`getOutputAt(nodeIndex)\` instead.`);return kn(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=yt(e),this.inputSpec==null||this.inputSpec.length===0)return;let t=yt(this.inputSpec);if(e.length!==t.length)throw new z(`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 a=e[n],r=t[n];if(r==null)continue;let s=a.rank;if(r.ndim!=null&&s!==r.ndim)throw new z(`Input ${n} is incompatible with layer ${this.name}: expected ndim=${r.ndim}, found ndim=${s}`);if(r.maxNDim!=null&&s>r.maxNDim)throw new z(`Input ${n} is incompatible with layer ${this.name}: expected max_ndim=${r.maxNDim}, found ndim=${s}`);if(r.minNDim!=null&&s<r.minNDim)throw new z(`Input ${n} is incompatible with layer ${this.name}: expected min_ndim=${r.minNDim}, found ndim=${s}.`);if(r.dtype!=null&&a.dtype!==r.dtype)throw new z(`Input ${n} is incompatible with layer ${this.name} : expected dtype=${r.dtype}, found dtype=${a.dtype}.`);if(r.axes){let i=a.shape;for(let o in r.axes){let l=Number(o),c=r.axes[o],u=l>=0?i[l]:i[i.length+l];if(c!=null&&[c,null].indexOf(u)===-1)throw new z(`Input ${n} is incompatible with layer ${this.name}: expected axis ${l} of input shape to have value ${c} but got shape ${i}.`)}}if(r.shape!=null)for(let i=0;i<r.shape.length;++i){let o=r.shape[i],l=a.shape[i];if(o!=null&&l!=null&&o!==l)throw new z(`Input ${n} is incompatible with layer ${this.name}: expected shape=${r.shape}, found shape=${a.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=yt(e),a=!0;for(let s of n)if(!(s instanceof Aa)){a=!1;break}let r=!0;for(let s of n)if(s instanceof Aa){r=!1;break}if(a===r)throw new z("Arguments to apply() must be all SymbolicTensors or all Tensors");return Mi(this.name,()=>{if(!this.built){this.assertInputCompatibility(e);let s=[];for(let i of yt(e))s.push(i.shape);this.build(kn(s)),this.built=!0,this.initialWeights&&this.setWeights(this.initialWeights),this._refCount===null&&r&&(this._refCount=1)}if(this.assertInputCompatibility(e),r){let s=this.call(e,t),i=yt(s),o=[];for(let l of i)n.indexOf(l)!==-1&&(l=l.clone()),o.push(l);if(s=kn(o),this.activityRegularizer!=null)throw new $e("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return s}else{let s=_B(e),i=this.computeOutputShape(s),o,l=EB(e);if(this.warnOnIncompatibleInputShape(Array.isArray(e)?s[0]:s),i!=null&&i.length>0&&Array.isArray(i[0])?o=i.map((c,u)=>new Aa(l,c,this,yt(e),t,this.name,u)):o=new Aa(l,i,this,yt(e),t,this.name),this.addInboundNode(e,o,null,null,s,i,t),this._refCount++,this.activityRegularizer!=null)throw new $e("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return o}})}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 ${this.name}`);else{let t=!1;this.batchInputShape.forEach((n,a)=>{n!=null&&e[a]!=null&&e[a]!==n&&(t=!0)}),t&&console.warn(`The shape of the input tensor (${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 xr(`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 xr(`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 _a(`You tried to call countParams() on ${this.name}, but the layer is not built yet. Build it first by calling build(batchInputShape).`);return rm(this.weights)}build(e){this.built=!0}getWeights(e=!1){return Pb(e?this.trainableWeights:this.weights)}setWeights(e){D(()=>{let t=this.weights;if(t.length!==e.length)throw new z(`You called setWeights(weights) on layer "${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=[],a=Pb(t);for(let r=0;r<a.length;++r){let s=a[r],i=t[r],o=e[r];if(!k.arraysEqual(s.shape,o.shape))throw new z(`Layer weight shape ${s.shape} not compatible with provided weight shape ${o.shape}`);n.push([i,o])}Ob(n)})}addWeight(e,t,n,a,r,s,i){if(this._addedWeightNames.indexOf(e)!==-1)throw new z(`Duplicate weight name ${e} for layer ${this.name}`);this._addedWeightNames.push(e),n==null&&(n="float32"),this.fastWeightInitDuringBuild&&(a=wt("zeros"));let o=a.apply(t,n),l=new Fk(o,n,e,s,i);return o.dispose(),r!=null&&this.addLoss(()=>r.apply(l.read())),s==null&&(s=!0),s?this._trainableWeights.push(l):this._nonTrainableWeights.push(l),l}setFastWeightInitDuringBuild(e){this.fastWeightInitDuringBuild=e}addLoss(e){e==null||Array.isArray(e)&&e.length===0||(e=yt(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 ${this.name} does not support masking, but was passed an inputMask.`);return null}return t}addInboundNode(e,t,n,a,r,s,i=null){let o=yt(e);t=yt(t),n=yt(n),a=yt(a),r=am(r),s=am(s);let l=[],c=[],u=[];for(let p of o)l.push(p.sourceLayer),c.push(p.nodeIndex),u.push(p.tensorIndex);new sm({outboundLayer:this,inboundLayers:l,nodeIndices:c,tensorIndices:u,inputTensors:o,outputTensors:t,inputMasks:n,outputMasks:a,inputShapes:r,outputShapes:s},i);for(let p=0;p<t.length;p++)t[p].sourceLayer=this,t[p].nodeIndex=this.inboundNodes.length-1,t[p].tensorIndex=p}getConfig(){let e={name:this.name,trainable:this.trainable};return this.batchInputShape!=null&&(e.batchInputShape=this.batchInputShape),this.dtype!=null&&(e.dtype=this.dtype),e}disposeWeights(){return this.weights.forEach(e=>e.dispose()),this.weights.length}assertNotDisposed(){if(this._refCount===0)throw new Error(`Layer '${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 _B(e){e=yt(e);let t=[];for(let n of e)t.push(n.shape);return kn(t)}function EB(e){return"float32"}function $k(e,t,n){if((t==null||n!=null&&n>0)&&(t=e.sourceLayer,n=e.nodeIndex),t.inboundNodes.length===0)return[e];{let a=t.inboundNodes[n];if(a.inboundLayers.length===0)return a.inputTensors;{let r=[];for(let s=0;s<a.inboundLayers.length;s++){let i=a.inputTensors[s],o=a.inboundLayers[s],l=a.nodeIndices[s],c=$k(i,o,l);for(let u of c)r.indexOf(u)===-1&&r.push(u)}return r}}}var Kl=class extends qe{constructor(e){super({dtype:e.dtype,name:e.name!=null?e.name:nm("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 z("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 z("An InputLayer should be passed either a `batchInputShape` or an `inputShape`.");t=[e.batchSize].concat(e.inputShape)}else if(e.batchSize!=null)throw new z("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 a=new Aa(this.dtype,this.batchInputShape,this,[],{},this.name);a.nodeIndex=0,a.tensorIndex=0,new sm({outboundLayer:this,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:[a],outputTensors:[a],inputMasks:[null],outputMasks:[null],inputShapes:[t],outputShapes:[t]})}apply(e,t){throw new z(`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}}};Kl.className="InputLayer";re.registerClass(Kl);function Dk(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 z("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 Kl({batchInputShape:t,name:e.name,dtype:n,sparse:e.sparse}).inboundNodes[0].outputTensors[0]}async function rs(e){if(e==null)return;let t=[],n=[],a=[];for(let r in e){let s=e[r];if(typeof s!="number"){let i=s;t.push(i.data()),n.push(r),a.push(i)}}if(t.length>0){let r=await Promise.all(t);for(let s=0;s<r.length;++s)e[n[s]]=r[s][0];Fe(a)}}function Rk(e){if(e!=null)for(let t in e){let n=e[t];typeof n!="number"&&n.dispose()}}var Mk;(function(e){e[e.SILENT=0]="SILENT",e[e.VERBOSE=1]="VERBOSE"})(Mk||(Mk={}));var AB=125,Xl=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){}},Pk=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)}},FB=class extends Xl{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 a in t){let r=t[a];if(typeof r=="number")this.totals.hasOwnProperty(a)||(this.totals[a]=0),this.totals[a]=this.totals[a]+r*n;else{let s;a in this.totals?s=this.totals[a]:this.totals[a]=0;let i=D(()=>Z(this.totals[a],L(r,n)));this.totals[a]=i,s!=null&&s.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:D(()=>{let a=L(ve(1,this.seen),this.totals[n]);t[n]=a,this.totals[n].dispose(),jt(t[n])}))}},Ok=class extends Xl{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 s=this.history[r];for(let i=0;i<s.length;++i)if(typeof s[i]!="number"){let o=s[i];e.push(o.data()),t.push(r),n.push(i)}}let a=await Promise.all(e);for(let r=0;r<a.length;++r)this.history[t[r]][n[r]].dispose(),this.history[t[r]][n[r]]=a[r][0]}},Lk=class extends Xl{constructor(e,t){super();if(this.currentEpoch=0,this.yieldEvery=t||"auto",this.yieldEvery==="auto"&&(this.yieldEvery=AB),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");k.isNumber(this.yieldEvery)&&(this.maybeWait=D3(this.maybeWait.bind(this),this.yieldEvery)),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 a=[];this.yield!=null&&(await rs(n),a.push(this.yield(e,t,n))),a.push(Gh()),await Promise.all(a)}async onEpochBegin(e,t){this.currentEpoch=e,this.epochBegin!=null&&(await rs(t),await this.epochBegin(e,t))}async onEpochEnd(e,t){let n=[];this.epochEnd!=null&&(await rs(t),n.push(this.epochEnd(e,t))),this.yieldEvery==="epoch"&&n.push(Gh()),await Promise.all(n)}async onBatchBegin(e,t){this.batchBegin!=null&&(await rs(t),await this.batchBegin(e,t))}async onBatchEnd(e,t){let n=[];this.batchEnd!=null&&(await rs(t),n.push(this.batchEnd(e,t))),this.yieldEvery==="batch"?n.push(Gh()):k.isNumber(this.yieldEvery)&&n.push(this.maybeWait(this.currentEpoch,e,t)),await Promise.all(n)}async onTrainBegin(e){this.trainBegin!=null&&(await rs(e),await this.trainBegin(e))}async onTrainEnd(e){this.trainEnd!=null&&(await rs(e),await this.trainEnd(e))}};function zk(e,t){return e==null&&(e={}),e instanceof Xl?[e]:Array.isArray(e)&&e[0]instanceof Xl?e:yt(e).map(n=>new Lk(n,t))}var ma=class{constructor(){}static registerCallbackConstructor(e,t){k.assert(e>=0&&Number.isInteger(e),()=>`Verbosity level is expected to be an integer >= 0, but got ${e}`),ma.checkForDuplicate(t),ma.constructors[e]==null&&(ma.constructors[e]=[]),ma.constructors[e].push(t)}static checkForDuplicate(e){for(let t in ma.constructors)ma.constructors[+t].forEach(n=>{if(n===e)throw new z("Duplicate callback constructor.")})}static clear(){ma.constructors={}}static createCallbacks(e){let t=[];for(let n in ma.constructors){let a=+n;e>=a&&t.push(...ma.constructors[a])}return t.map(n=>new n)}};ma.constructors={};function Bk(e,t,n,a,r,s,i,o,l){let c=new Ok,u=[new FB,...ma.createCallbacks(t)];e!=null&&u.push(...e),u.push(c);let p=new Pk(u);return p.setParams({epochs:n,initialEpoch:a,samples:r,steps:s,batchSize:i,verbose:t,doValidation:o,metrics:l}),{callbackList:p,history:c}}function Fa(e,t={},n=!1){return Xc(e,re.SerializationMap.getMap().classNameMap,t,"layer",n)}function im(e,t){return D(()=>{e.dtype!=="float32"&&(e=e.asType("float32"));let n=Te(ep(e),t,!0),a=wn(n.shape,Bt()),r=en(Ka(n,a));return ve(e,r)})}function Oi(e,t){return D(()=>St(ep(fe(t,e)),-1))}function om(e,t){return D(()=>St(Lt(fe(t,e)),-1))}function Yl(e,t){return D(()=>{let n=fe(e,t),a=qt(Lt(e),Bt(),Number.MAX_VALUE),r=Lt(ve(n,a));return L(100,St(r,-1))})}function $B(e,t){return D(()=>{let n=qt(t,Bt(),Number.MAX_VALUE),a=$n(Z(1,n)),r=qt(e,Bt(),Number.MAX_VALUE),s=$n(Z(1,r));return St(ep(fe(a,s)),-1)})}function DB(e,t){return D(()=>{let n=Ka(0,fe(1,L(e,t)));return St(ep(n),-1)})}function RB(e,t){return D(()=>{let n=Ka(0,fe(1,L(e,t)));return St(n,-1)})}function MB(e,t){return D(()=>{let n=Te(L(e,t),-1),a=Yn(L(fe(1,e),t),-1);return Ka(0,Z(1,fe(a,n)))})}function PB(e,t){return D(()=>{let n=Math.log(2),a=fe(t,e),r=fe(Z(a,Ll(L(-2,a))),n);return St(r,-1)})}function ap(e,t,n=!1){return D(()=>{if(n)t=Na(t);else{let a=Te(t,t.shape.length-1,!0);t=ve(t,a)}return t=qt(t,Bt(),1-Bt()),Tt(Te(L(e.toFloat(),$n(t)),t.shape.length-1))})}function lm(e,t,n=!1){return D(()=>{let a=Ol(sz(e)).toInt();t=qt(t,Bt(),1-Bt());let r=t.shape,s=$l(a,r[r.length-1]).reshape(r);return ap(s,t,n)})}function OB(e,t){if(!k.arraysEqual(e.shape,t.shape))throw new z(`logits and labels must have the same shape, but got shapes ${JSON.stringify(e.shape)} and ${JSON.stringify(t.shape)}`);return D(()=>{let n=t.relu(),a=t.abs().neg();return n.sub(t.mul(e)).add(a.exp().log1p())})}function um(e,t){return D(()=>{let n;return n=qt(t,Bt(),1-Bt()),n=$n(ve(n,fe(1,n))),St(OB(e,n),-1)})}function LB(e,t){return D(()=>{let n=qt(e,Bt(),1),a=qt(t,Bt(),1);return Te(L(e,$n(ve(n,a))),-1)})}function zB(e,t){return D(()=>{let n=$n(Z(Bt(),t));return St(fe(t,L(e,n)),-1)})}function Lb(e,t){return D(()=>{let n=im(e,-1),a=im(t,-1),r=L(n,a);return Tt(Te(r,-1))})}var cm={meanSquaredError:Oi,meanAbsoluteError:om,meanAbsolutePercentageError:Yl,meanSquaredLogarithmicError:$B,squaredHinge:DB,hinge:RB,categoricalHinge:MB,logcosh:PB,categoricalCrossentropy:ap,sparseCategoricalCrossentropy:lm,binaryCrossentropy:um,kullbackLeiblerDivergence:LB,poisson:zB,cosineProximity:Lb};function zb(e){if(typeof e=="string"){if(e in cm)return cm[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 z(t)}else return e}function Bb(e,t){return D(()=>{let n=L(.5,Dn(t)),a=Jc(ca(t,n),e.dtype);return St(Zr(e,a),-1)})}function Wb(e,t){return D(()=>Jc(Zr(Mc(e,-1),Mc(t,-1)),"float32"))}function Wk(e,t){return D(()=>pa(e.equal(1),t.equal(1)).sum().cast("float32"))}function BB(e,t){return D(()=>pa(e.equal(1),t.equal(0)).sum().cast("float32"))}function WB(e,t){return D(()=>pa(e.equal(0),t.equal(1)).sum().cast("float32"))}function Vk(e,t){return D(()=>{let n=Wk(e,t),a=WB(e,t),r=n.add(a);return vn(ca(r,0),n.div(r),0).cast("float32")})}function VB(e,t){return D(()=>{let n=Wk(e,t),a=BB(e,t),r=n.add(a);return vn(ca(r,0),n.div(r),0).cast("float32")})}function Uk(e,t){return um(e,t)}function Gk(e,t){return e.rank===t.rank&&(e=e.squeeze([e.rank-1])),t=t.argMax(-1),t.dtype!==e.dtype&&(t=t.asType(e.dtype)),Zr(e,t).asType("float32")}var UB=Oi,GB=Oi,HB=om,jB=om,qB=Yl,KB=Yl,Vb=ap,XB=Lb,Hk=lm,pm={binaryAccuracy:Bb,categoricalAccuracy:Wb,precision:Vk,categoricalCrossentropy:Vb,sparseCategoricalCrossentropy:Hk,mse:UB,MSE:GB,mae:HB,MAE:jB,mape:qB,MAPE:KB,cosine:XB};function YB(e){if(typeof e=="string"&&e in pm)return pm[e];if(typeof e!="string"&&e!=null)return e;throw new z(`Unknown metric ${e}`)}function dm(e){if(Ja(e!==null,`Unknown LossOrMetricFn ${e}`),typeof e=="string")return e;{let t;for(let n of Object.keys(cm))if(cm[n]===e){t=n;break}if(t!==void 0)return t;for(let n of Object.keys(pm))if(pm[n]===e){t=n;break}return t!==void 0?t:e.name}}function ZB(e){let t={Adagrad:()=>Fi.adagrad(.01),Adadelta:()=>Fi.adadelta(1,.95,Bt()),Adam:()=>Fi.adam(.001,.9,.999,Bt()),Adamax:()=>Fi.adamax(.002,.9,.999,Bt(),0),RMSProp:()=>Fi.rmsprop(.001,.9,0,Bt()),SGD:()=>Fi.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 z(`Unknown Optimizer ${e}`)}var jk=1*1024*1024;function qk(e,t,n=!1){if(e==null||typeof e!="object"||Object.getPrototypeOf(e)!==Object.prototype||!Ub(e))throw new Error("User-defined metadata is expected to be a JSON object, but is not.");if(n){let a=JSON.stringify(e);a.length>jk&&console.warn(`User-defined metadata of model "${t}" is too large in size (length=${a.length} when serialized). It is not recommended to store such large objects in user-defined metadata. Please make sure its serialized length is <= ${jk}.`)}}function Ub(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"||!Ub(e[n]))return!1;return!0}else if(Array.isArray(e)){for(let t of e)if(!Ub(t))return!1;return!0}else return!1;else{let t=typeof e;return t==="string"||t==="number"||t==="boolean"}}function nW(e,t,n,a=console.log){let r=QB(e),s=["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 i;if(!r){s.push("Receives inputs"),i=[];for(let u in e.nodesByDepth)i.push(...e.nodesByDepth[u])}a("_".repeat(t)),hm(s,n,a),a("=".repeat(t));let o=e.layers;for(let u=0;u<o.length;++u)r?eW(o[u],n,a):tW(o[u],n,i,a),a((u===o.length-1?"=":"_").repeat(t));e.checkTrainableWeightsConsistency();let l=JB(e),c=rm(e.nonTrainableWeights);a(`Total params: ${l+c}`),a(`Trainable params: ${l}`),a(`Non-trainable params: ${c}`),a("_".repeat(t))}function JB(e){let t;return e.collectedTrainableWeights!=null?t=rm(e.collectedTrainableWeights):t=rm(e.trainableWeights),t}function QB(e){let t=!0,n=[],a=[];for(let r in e.nodesByDepth)n.push(e.nodesByDepth[r]);for(let r of n){if(r.length>1||r.length===1&&r[0].inboundLayers.length>1){t=!1;break}a.push(...r)}if(t)for(let r of e.layers){let s=!1;for(let i of r.inboundNodes)if(a.indexOf(i)!==-1)if(s){t=!1;break}else s=!0;if(!t)break}return t}function hm(e,t,n=console.log){let a="";for(let r=0;r<e.length;++r)r>0&&(a=a.slice(0,a.length-1)+" "),a+=e[r],a=a.slice(0,t[r]),a+=" ".repeat(t[r]-a.length);n(a)}function eW(e,t,n){let a;try{a=JSON.stringify(e.outputShape)}catch(o){a="multiple"}let r=e.name,s=e.getClassName(),i=[`${r} (${s})`,a,e.countParams().toString()];hm(i,t,n)}function tW(e,t,n,a){let r;try{r=JSON.stringify(e.outputShape)}catch(u){r="multiple"}let s=[];for(let u of e.inboundNodes)if(!(n!=null&&n.length>0&&n.indexOf(u)===-1))for(let p=0;p<u.inboundLayers.length;++p){let d=u.inboundLayers[p].name,h=u.nodeIndices[p],m=u.tensorIndices[p];s.push(`${d}[${h}][${m}]`)}let i=e.name,o=e.getClassName(),l=s.length===0?"":s[0],c=[`${i} (${o})`,r,e.countParams().toString(),l];hm(c,t,a);for(let u=1;u<s.length;++u)hm(["","","",s[u]],t,a)}function Kk(e,t,n){return(e==="inboundNodes"||e==="outputLayers"||e==="inputLayers")&&t===0&&typeof n=="string"}function rp(e,t){if(e===null)return null;if(typeof e=="string")return Di(e);if(typeof e=="number"||typeof e=="boolean")return e;if(e instanceof Array){let n=[],a=e.length;for(let r=0;r<a;++r){let s=e[r];Kk(t,r,s)?n.push(s):n.push(rp(s,t))}return n}else{let n={};for(let a of Object.keys(e)){let r=e[a];if(a==="name"&&typeof r=="string")n[a]=r;else{let s=Di(a);n[s]=rp(r,s)}}return n}}function Gb(e,t){if(e==null)return null;if(typeof e=="string")return vr(e);if(typeof e=="number"||typeof e=="boolean")return e;if(e instanceof Array){let n=[],a=e.length;for(let r=0;r<a;++r){let s=e[r];Kk(t,r,s)?n.push(s):n.push(Gb(s,t))}return n}else{let n={};for(let a of Object.keys(e)){let r=e[a],s=vr(a);(a==="name"||a==="className")&&typeof r=="string"?n[s]=r:n[s]=Gb(r,a)}return n}}var mm="3.0.0";function aW(e,t){if(e.dtype==null||e.dtype===t.dtype)return t;try{return ce(t,e.dtype)}catch(n){throw new z(`The dtype of the feed (${t.dtype}) can not be cast to the dtype of the key '${e.name}' (${e.dtype}).`)}}var Li=class{constructor(e){if(this.id2Value={},this.id2Mask={},this.name2Id={},e instanceof Li)for(let t in e.id2Value)this.id2Value[t]=e.id2Value[t],t in e.id2Mask&&(this.id2Mask[t]=e.id2Mask[t]);else{if(e==null)return;for(let t of e)this.add(t.key,t.value)}}add(e,t,n){if(this.id2Value[e.id]==null)this.id2Value[e.id]=aW(e,t),this.name2Id[e.name]=e.id,n!=null&&(this.id2Mask[e.id]=n);else throw new z(`Duplicate key: name=${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 Aa){if(this.id2Value[e.id]==null)throw new z(`Nonexistent key: ${e.name}`);return this.id2Value[e.id]}else{let t=this.name2Id[e];if(t==null)throw new z(`Feed dict has no SymbolicTensor name: ${e}`);return this.id2Value[t]}}getMask(e){if(e instanceof Aa){if(this.id2Value[e.id]==null)throw new z(`Nonexistent key: ${e.name}`);return this.id2Mask[e.id]}else{let t=this.name2Id[e];if(t==null)throw new z(`Feed dict has no SymbolicTensor name: ${e}`);return this.id2Mask[t]}}disposeMasks(){this.id2Mask!=null&&Fe(this.id2Mask)}},Hb={},Xk={};function sp(e,t,n,a){let r=n==null?!1:n.training,s=Array.isArray(e),i=s?e:[e],o=i.map(m=>m.name),l=[],c=t.names();for(let m of o)c.indexOf(m)!==-1?l.push(t.getValue(m)):l.push(null);a!=null&&(a.maxNumTensors=-Infinity,a.minNumTensors=Infinity);let u=o.join(",")+"|"+t.names().join(","),p,d;if(Hb[u]==null){let m=rW(i,t);p=m.sorted,d=m.recipientCounts,Hb[u]=p,Xk[u]=d}p=Hb[u],d={},r||Object.assign(d,Xk[u]);let h=new Li(t);for(let m=0;m<p.length;++m){if(a!=null){let F=sh().numTensors;F>a.maxNumTensors&&(a.maxNumTensors=F),F<a.minNumTensors&&(a.minNumTensors=F)}let f=p[m],g=f.sourceLayer;if(g instanceof Kl)continue;let y=[],b=[],v=[],x=!1;for(let F of f.inputs){let O=h.getValue(F),B=h.getMask(F);y.push(O),b.push(B),B!=null&&(x=!0),r||(d[F.name]--,d[F.name]===0&&!t.hasKey(F)&&o.indexOf(F.name)===-1&&!O.isDisposed&&F.sourceLayer.stateful!==!0&&v.push(O))}x&&(n=n||{},n.mask=b[0]);let N=yt(g.apply(y,n)),T=null;g.supportsMasking&&(T=g.computeMask(y,b));let C=sW(f),$=Array.isArray(C)?C:[C];for(let F=0;F<$.length;++F){h.hasKey($[F])||h.add($[F],N[F],Array.isArray(T)?T[0]:T);let O=o.indexOf($[F].name);O!==-1&&(l[O]=N[F])}r||Fe(v)}return h.disposeMasks(),s?l:l[0]}function rW(e,t){k.assert(e!=null&&e.length>0,()=>"Expected at least one fetch, got none");let n=[],a={};if(e.length===1){let r=Yk(e[0],t);n=r.sorted,a=r.recipientMap}else{let r=new Set;for(let s of e){let{sorted:i,recipientMap:o}=Yk(s,t);for(let l of i)r.has(l.name)||(n.push(l),r.add(l.name));for(let l in o)a[l]==null&&(a[l]=new Set),o[l].forEach(c=>a[l].add(c))}}return{sorted:n,recipientCounts:iW(a)}}function iW(e){let t={};for(let n in e)t[n]=e[n].size;return t}function Yk(e,t){let n=new Set,a=[],r={};for(let o of t.names())n.add(o);let s=[],i=[];for(s.push(e);s.length>0;){let o=s[s.length-1];if(n.has(o.name)){s.pop();continue}let l=i[i.length-1]===s.length-1;if(o.inputs.length===0||l)s.pop(),a.push(o),n.add(o.name),l&&i.pop();else{i.push(s.length-1);for(let c of o.inputs)r[c.name]==null&&(r[c.name]=new Set),r[c.name].add(o.name),!n.has(c.name)&&s.push(c)}}return{sorted:a,recipientMap:r}}function sW(e){let t;if(e.sourceLayer.inboundNodes.length===1)t=e.sourceLayer.output;else{let n=null;for(let a=0;a<e.sourceLayer.inboundNodes.length;++a)for(let r of e.sourceLayer.inboundNodes[a].outputTensors)if(r.id===e.id){n=a;break}t=e.sourceLayer.getOutputAt(n)}return t}var nr=class extends qe{constructor(e){super({});if(this.containerNodes=new Set,this.name=e.name,this.name==null){let y=this.getClassName().toLowerCase();this.name=nm(y)}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],ts(this.inputs).length!==this.inputs.length)throw new z(`The list of inputs passed to the model is redundant. All inputs should only appear once. Found: ${this.inputs.map(y=>y.name)}`);ts(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(y=>y.name)}`),this.inputLayers=[],this.inputLayersNodeIndices=[],this.inputLayersTensorIndices=[],this.outputLayers=[],this.outputLayersNodeIndices=[],this.outputLayersTensorIndices=[],this.layers=[],this.internalContainerRefs=[];for(let y of this.outputs){let b=y.sourceLayer,v=y.nodeIndex,x=y.tensorIndex;this.outputLayers.push(b),this.outputLayersNodeIndices.push(v),this.outputLayersTensorIndices.push(x)}for(let y of this.inputs){let b=y.sourceLayer,v=y.nodeIndex,x=y.tensorIndex;Ja(v===0,"input layer has >1 nodes"),Ja(x===0,"input layer has >1 tensors"),this.inputLayers.push(b),this.inputLayersNodeIndices.push(v),this.inputLayersTensorIndices.push(x)}this.inputNames=[],this.outputNames=[],this.feedInputShapes=[],this.feedInputNames=[],this.feedOutputNames=[];for(let y=0;y<this.inputLayers.length;y++){let b=this.inputLayers[y];if(!(b instanceof Kl))throw new TypeError(`Input layers to a LayersModel must be InputLayer objects. Received inputs: ${e.inputs}. Input ${y} (0-based) originates from layer type ${b.getClassName()}.`);this.inputNames.push(b.name),this.feedInputShapes.push(b.batchInputShape),this.feedInputNames.push(b.name)}for(let y of this.outputLayers)this.outputNames.push(y.name);this.internalInputShapes=this.inputs.map(y=>y.shape),this.internalOutputShapes=this.outputs.map(y=>y.shape);let t={},n={},a={},r={},s={},i=[],o=(y,b,v,x,N,T)=>{(x==null||N==null||T==null)&&(x=y.sourceLayer,N=y.nodeIndex,T=y.tensorIndex);let C=x.inboundNodes[N];if(v.indexOf(C)!==-1)throw new _a(`The tensor ${y.name} at layer "${x.name}" is part of a cycle.`);if(b.indexOf(C)!==-1)return;this.containerNodes.add(nr.nodeKey(x,N)),x.id in s||(s[x.id]=Object.keys(s).length),v.indexOf(C)===-1&&v.push(C);let $=C.inboundLayers.length;for(let F=0;F<$;F++){let O=C.inputTensors[F],B=C.inboundLayers[F],W=C.nodeIndices[F],G=C.tensorIndices[F];o(O,b,v,B,W,G)}for(b.push(C);v.indexOf(C)>=0;)v.splice(v.indexOf(C),1);i.push(C)},l=[],c=[];for(let y of this.outputs)o(y,l,c);let u=i.slice().reverse();for(let y of u){n[y.id]=y,y.id in t||(t[y.id]=0);let b=t[y.id],v=a[y.outboundLayer.id]==null?0:a[y.outboundLayer.id];b=Math.max(b,v),a[y.outboundLayer.id]=b,r[y.outboundLayer.id]=y.outboundLayer,t[y.id]=b;for(let x=0;x<y.inboundLayers.length;x++){let N=y.inboundLayers[x],T=y.nodeIndices[x],C=N.inboundNodes[T],$=t[C.id]==null?0:t[C.id];t[C.id]=Math.max(b+1,$),n[C.id]=C}}let p={};for(let y in t){let b=t[y];b in p||(p[b]=[]),p[b].push(n[y])}let d={};for(let y in a){let b=a[y];b in d||(d[b]=[]),d[b].push(r[y])}let h=Object.keys(d).map(y=>parseInt(y,10)).sort(Hh);this.layers=[];for(let y of h){let b=d[y];b.sort((v,x)=>{let N=s[v.id],T=s[x.id];return N<T?-1:N>T?1:0});for(let v of b)v instanceof nr&&this.internalContainerRefs.push(v),this.layers.push(v)}this.layersByDepth=d,h=Object.keys(p).map(y=>parseInt(y,10)).sort(Hh);let m=this.inputs.slice(),f=[];for(let y of h)for(let b of p[y]){let v=b.outboundLayer;if(v!=null){for(let x of b.inputTensors)if(m.indexOf(x)===-1)throw new _a(`Graph disconnected: cannot obtain value for tensor ${x} at layer "${v.name}". The following previous layers were accessed without issue: ${f}`);for(let x of b.outputTensors)m.push(x);f.push(v.name)}}this.nodesByDepth=p;let g=this.layers.map(y=>y.name);for(let y of g){let b=g.filter(v=>v===y).length;if(b!==1)throw new _a(`The name "${y}" is used ${b} times in the model. All layer names should be unique. Layer names: `+JSON.stringify(g))}this.outboundNodes=[],this.inboundNodes=[],new sm({outboundLayer:this,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:this.inputs,outputTensors:this.outputs,inputMasks:this.inputs.map(y=>null),outputMasks:this.outputs.map(y=>null),inputShapes:this.inputs.map(y=>y.shape),outputShapes:this.outputs.map(y=>y.shape)}),this.built=!0,this._refCount=1}assertNotDisposed(){if(this._refCount===0)throw new Error(`Container '${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 z("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={},a=0;for(let s of this.layers)for(let i of s.weights){if(n[i.originalName]!=null)throw new z(`Duplicate weight name: ${i.originalName}`);n[i.originalName]=i,a++}let r=[];for(let s in e){let i=s;if(n[s]==null){let o=s.split("/");i=o.slice(0,-2).concat([o[o.length-1]]).join("/")}if(n[i]!=null)r.push([n[i],e[s]]);else if(t)throw new z(`Provided weight data has no target variable: ${s}`);delete n[i]}if(t){let s=[];for(let i in n)s.push(i);if(s.length>0)throw new z(`${s.length} of ${a} weights are not set: ${s}`)}Ob(r)}updatedConfig(){let e=this.getConfig(),t={};return t.className=this.getClassName(),t.config=e,t.kerasVersion=`tfjs-layers ${mm}`,t.backend="TensorFlow.js",t}toJSON(e,t=!0){let n=Gb(this.updatedConfig());return t?JSON.stringify(n):n}call(e,t){return D(()=>{e=yt(e);let n=new Li;for(let a=0;a<this.inputs.length;++a)n.add(this.inputs[a],e[a]);return sp(this.outputs,n,t)})}computeMask(e,t){return D(()=>{e=yt(e);let n;return t==null?n=$i(null,e.length):n=yt(t),this.runInternalGraph(e,n)[1]})}computeOutputShape(e){let t=am(e);if(t.length!==this.inputLayers.length)throw new z(`Invalid inputShape argument ${e}: model has ${this.inputLayers.length} tensor inputs.`);let n={};for(let i=0;i<t.length;i++){let o=this.inputLayers[i],l=t[i],c=o.name+"_0_0";n[c]=l}let a=Object.keys(this.nodesByDepth).map(i=>parseInt(i,10)).sort(Hh);if(a.length>1)for(let i of a){let o=this.nodesByDepth[i];for(let l of o){let c=l.outboundLayer;if(this.inputLayers.map(m=>m.id).indexOf(c.id)!==-1)continue;let u=[];for(let m=0;m<l.inboundLayers.length;m++){let f=l.inboundLayers[m],g=l.nodeIndices[m],y=l.tensorIndices[m],b=`${f.name}_${g}_${y}`,v=n[b];u.push(v)}let p=c.computeOutputShape(kn(u)),d=am(p),h=c.inboundNodes.indexOf(l);for(let m=0;m<d.length;m++){let f=`${c.name}_${h}_${m}`;n[f]=d[m]}}}let r=[],s=[];for(let i=0;i<this.outputLayers.length;i++){let o=this.outputLayers[i],l=this.outputLayersNodeIndices[i],c=this.outputLayersTensorIndices[i],u=`${o.name}_${l}_${c}`;s.push(u)}for(let i=0;i<s.length;i++){let o=s[i];Ja(o in n),r.push(n[o])}return kn(r)}runInternalGraph(e,t){t==null&&(t=$i(null,e.length));let n={};for(let o=0;o<this.inputs.length;++o){let l=this.inputs[o],c=e[o],u=t[o];n[l.id]=[c,u]}let a=Object.keys(this.nodesByDepth).map(o=>parseInt(o,10)).sort(Hh);for(let o of a){let l=this.nodesByDepth[o];for(let c of l){let u=c.outboundLayer,p=c.inputTensors,d=c.outputTensors,h=new Array;for(let m of p)m.id in n&&h.push(n[m.id]);if(h.length===p.length){let m={},f,g,y,b;if(c.callArgs!=null&&(m=c.callArgs),h.length===1){let[v,x]=h[0];m.mask==null&&(m.mask=x),y=yt(u.call(v,m)),b=yt(u.computeMask(v,x)),f=[v],g=[x]}else f=h.map(v=>v[0]),g=h.map(v=>v[1]),m.mask==null&&(m.mask=g),y=yt(u.call(f,m)),b=yt(u.computeMask(f,g));if(u.activityRegularizer)throw new $e("LayersModel invocation with concrete Tensor value(s) in the presence of activity regularizer(s) is not supported yet.");for(let v=0;v<d.length;++v){let x=d[v],N=y[v],T=b[v];n[x.id]=[N,T]}}}}let r=[],s=[],i=[];for(let o of this.outputs){Ja(o.id in n,`Could not compute output ${o.name} : ${o.id}`);let[l,c]=n[o.id];i.push(l.shape),r.push(l),s.push(c)}return[r,s,i]}buildNodeConversionMap(e){let t={},n;for(let a of this.layers){n=a instanceof nr?1:0;for(let r=0;r<a.inboundNodes.length;r++){let s=nr.nodeKey(a,r);this.containerNodes.has(s)&&(t[s]=n,n+=1)}}return t}getLayer(e,t){if(t!=null){if(this.layers.length<=t)throw new z(`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 z("Provide either a layer name or layer index");for(let n of this.layers)if(n.name===e)return n;throw new z(`No such layer: ${e}`)}calculateLosses(){return D(()=>{let e=[];for(let t of this.layers)for(let n=0;n<t.inboundNodes.length;++n){let a=nr.nodeKey(t,n);this.containerNodes.has(a)&&e.push(...t.calculateLosses())}return e})}getConfig(){let e={name:this.name},t=this.buildNodeConversionMap(this.layers),n=[];for(let s of this.layers){let i=s.getClassName(),o=s.getConfig(),l=[];for(let u=0;u<s.inboundNodes.length;u++){let p=s.inboundNodes[u],d=nr.nodeKey(s,u),h={};if(this.containerNodes.has(d)){if(p.callArgs)try{JSON.stringify(p.callArgs),h=p.callArgs}catch(m){console.warn(`Layer ${s.name} was passed non-serializable keyword arguments: ${p.callArgs}. They will not be included in the serialized model (and thus will be missing at deserialization time).`),h={}}if(p.inboundLayers.length>0){let m=[];for(let f=0;f<p.inboundLayers.length;f++){let g=p.inboundLayers[f],y=p.nodeIndices[f],b=p.tensorIndices[f],v=nr.nodeKey(g,y),x=t[v];x==null&&(x=0),m.push([g.name,x,b,h])}l.push(m)}}}let c={};c.name=s.name,c.className=i,c.config=o,c.inboundNodes=l,n.push(c)}e.layers=n;let a=[];for(let s=0;s<this.inputLayers.length;s++){let i=this.inputLayers[s],o=this.inputLayersNodeIndices[s],l=nr.nodeKey(i,o);if(!this.containerNodes.has(l))continue;let c=t[l];c==null&&(c=0);let u=this.inputLayersTensorIndices[s];a.push([i.name,c,u])}e.inputLayers=a;let r=[];for(let s=0;s<this.outputLayers.length;s++){let i=this.outputLayers[s],o=this.outputLayersNodeIndices[s],l=nr.nodeKey(i,o);if(!this.containerNodes.has(l))continue;let c=t[l];c==null&&(c=0);let u=this.outputLayersTensorIndices[s];r.push([i.name,c,u])}return e.outputLayers=r,e}static fromConfig(e,t,n={},a=!1){let r={},s={};function i(f,g){f.name in s?s[f.name].push(g):s[f.name]=[g]}function o(f,g){let y=[],b;for(let v of g){let x=v[0],N=v[1],T=v[2];if(b=v[3]==null?{}:v[3],!(x in r)){i(f,g);return}let C=r[x];if(C.inboundNodes.length<=N){i(f,g);return}let $=C.inboundNodes[N];y.push($.outputTensors[T])}y.length>0&&f.apply(kn(y),b)}function l(f){let g=f.name,y=Fa(f,t.customObjects!=null?t.customObjects:{});y.setFastWeightInitDuringBuild(a),r[g]=y,f.inboundNodes.forEach(b=>{if(!(b instanceof Array))throw new z(`Corrupted configuration, expected array for nodeData: ${b}`);i(y,b)})}let c=t.name,u=t.layers;for(let f of u)l(f);for(;!$3(s);)for(let f of u){let g=r[f.name];if(g.name in s){let y=s[g.name];delete s[g.name];for(let b of y)o(g,b)}}let p=[],d=[],h=t.inputLayers;for(let f of h){let g=f[0],y=f[1],b=f[2];Ja(g in r);let v=r[g].inboundNodes[y].outputTensors;p.push(v[b])}let m=t.outputLayers;for(let f of m){let g=f[0],y=f[1],b=f[2];Ja(g in r);let v=r[g].inboundNodes[y].outputTensors;d.push(v[b])}return new e({inputs:p,outputs:d,name:c})}get stateful(){if(this._stateful)throw new z("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(){D(()=>{this.layers.forEach(e=>{e.stateful&&e.resetStates()})})}};function oW(e,t,n){let a=t.length;if(e==null||Array.isArray(e)&&e.length===0)return t.map(r=>null);if(a===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!==a)throw new Error(`Provided ${n} is an array of ${e.length} element(s), but the model has ${a} outputs. Make sure a set of weights is provided for each model output.`);return e}else if(typeof e=="object"&&Object.keys(e).length>0&&typeof e[Object.keys(e)[0]]=="object"){let r=[];return t.forEach(s=>{s in e?r.push(e[s]):r.push(null)}),r}else throw new Error(`The model has multiple (${a}) outputs, so ${n} must be either an array with ${a} elements or an object with ${t} keys. Provided ${n} not understood: ${JSON.stringify(e)}`)}function Zk(e,t){return oW(e,t,"classWeight")}async function Jk(e,t,n,a){if(t!=null||a!=null)throw new Error("Support sampleWeight is not implemented yet");if(n!=null){let r=D(()=>{if(e.shape.length===1)return e.clone();if(e.shape.length===2)if(e.shape[1]>1){let o=1;return e.argMax(o)}else{if(e.shape[1]===1)return e.reshape([e.shape[0]]);throw new Error(`Encountered unexpected last-dimension size (${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.`)}),s=Array.from(await r.data());Fe(r);let i=[];return s.forEach(o=>{if(n[o]==null)throw new Error(`classWeight must contain all classes in the training data. The class ${o} exists in the data but not in classWeight`);i.push(n[o])}),tt(i,"float32")}else return null}function lW(e,t){return L(e,t)}var uW=32;function eI(e,t){let n,a,r=t;n=r.xs,a=r.ys,k.assert(n!=null&&a!=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 s=Qk("input",e.inputNames,n),i=Qk("output",e.outputNames,a),o=s[0].shape[0];k.assert(s.length===e.inputs.length,()=>`LayersModel has ${e.inputs.length} inputs, but the dataset provides ${s.length} inputs. (Expected input keys: ${JSON.stringify(e.inputNames)})`),k.assert(i.length===e.outputs.length,()=>`LayersModel has ${e.outputs.length} outputs, but the dataset provides ${i.length} outputs. (Expected output keys: ${JSON.stringify(e.outputNames)})`);for(let l=0;l<s.length;l++)k.assert(s[l].shape[0]===o,()=>`Batch size mismatch: input ${e.inputNames[l]} has ${s[l].shape[0]}; expected ${o} based on input ${e.inputNames[0]}.`);for(let l=0;l<i.length;l++)k.assert(i[l].shape[0]===o,()=>`Batch size mismatch: output ${e.outputNames[l]} has ${i[l].shape[0]}; expected ${o} based on input ${e.inputNames[0]}.`);return{xs:s,ys:i}}function Qk(e,t,n){if(n instanceof Ae)return[n];if(Array.isArray(n))return k.assert(n.length===t.length,()=>`Received an array of ${n.length} Tensors, but expected ${t.length} to match the ${e} keys ${t}.`),n;{let a=[];for(let r of t){if(n[r]==null)throw new z(`The feature data generated by the dataset lacks the required ${e} key '${r}'.`);a.push(n[r])}return a}}function cW(e){if(e.length===3)throw new $e("Validation with sample weights is not implemented yet.");return{xs:e[0],ys:e[1]}}async function dW(e,t,n){let a=n.batchesPerEpoch!=null;if(k.assert(e.optimizer!=null,()=>"You must compile a model before training/testing. Use LayersModel.compile(modelCompileConfig)."),k.assert(n!=null,()=>"For fitDataset(), the 2nd argument (config) is required, but it is not provided in this call."),k.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}`),k.assert(!a||n.batchesPerEpoch>0&&Number.isInteger(n.batchesPerEpoch),()=>`For fitDataset(), config.batchesPerEpoch is expected to be a positive integer if specified, but got ${n.batchesPerEpoch}`),k.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,s,i;if(r)if(tI(n.validationData))k.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=cW(n.validationData);s=g.xs,i=g.ys}let o=e.makeTrainFunction(),l=e.getDedupedMetricsNames(),c;r?c=l.slice().concat(l.map(g=>"val_"+g)):c=l.slice();let u=zk(n.callbacks,n.yieldEvery),p=n.verbose==null?1:n.verbose,{callbackList:d,history:h}=Bk(u,p,n.epochs,null,null,pW(t,n),null,r,c);d.setModel(e),e.history=h,await d.onTrainBegin(),e.stopTraining_=!1;let m=n.initialEpoch==null?0:n.initialEpoch,f=await t.iterator();for(;m<n.epochs;){let g={};await d.onEpochBegin(m);let y=0,b=0;for(a||(f=await t.iterator());a?y<n.batchesPerEpoch:!0;){let v=await f.next();if(a&&v.done){console.warn(`You provided \`batchesPerEpoch\` as ${n.batchesPerEpoch}, but your dataset iterator ran out of data after ${y} 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(v.value!=null){let{xs:x,ys:N}=eI(e,v.value),T={};T.batch=b,T.size=x[0].shape[0],await d.onBatchBegin(b,T);let C=[];if(n.classWeight!=null){let O=Zk(n.classWeight,e.outputNames);for(let B=0;B<O.length;++B)C.push(await Jk(N[B],null,O[B]))}let $=x.concat(N).concat(C),F=o($);Fe($);for(let O=0;O<l.length;++O){let B=l[O],W=F[O];T[B]=W,jt(W)}await d.onBatchEnd(b,T),Rk(T),b++,y++}if(a?y>=n.batchesPerEpoch:v.done){if(r){let x;tI(n.validationData)?x=yt(await e.evaluateDataset(n.validationData,{batches:n.validationBatches})):x=yt(e.evaluate(s,i,{batchSize:n.validationBatchSize==null?uW:n.validationBatchSize,verbose:0}));for(let N=0;N<e.metricsNames.length;++N)g[`val_${e.metricsNames[N]}`]=x[N]}break}if(e.stopTraining_)break}if(await d.onEpochEnd(m,g),m++,e.stopTraining_)break}return await d.onTrainEnd(),await e.history.syncData(),e.history}finally{e.isTraining=!1}}function pW(e,t){let n=null;return t.batchesPerEpoch!=null?n=t.batchesPerEpoch:Number.isFinite(e.size)&&(n=e.size),n}function tI(e){return typeof e.iterator=="function"}function hW(e){return typeof e.next=="function"}async function mW(e,t,n){n=n||{};let a=n.batches!=null,r=e.testFunction,s=[];if(n.verbose>0)throw new $e("Verbose mode is not implemented yet.");k.assert(!a||n.batches>0&&Number.isInteger(n.batches),()=>`Test loop expects \`batches\` to be a positive integer, but received ${JSON.stringify(n.batches)}`);let i=hW(t)?t:await t.iterator(),o=0,l=0;for(;a?l<n.batches:!0;){let c=await i.next();if(s=D(()=>{if(c.value){let{xs:u,ys:p}=eI(e,c.value),d=u.concat(p),h=D(()=>r(d));if(Fe(d),l===0)for(let f=0;f<h.length;++f)s.push(de(0));let m=d[0].shape[0];for(let f=0;f<h.length;++f){let g=h[f],y=s[f];s[f]=D(()=>Z(s[f],L(m,g))),l>0&&Fe(y)}Fe(h),o+=m,++l}return s}),c.done){a&&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<s.length;++c){let u=s[c];s[c]=ve(s[c],o),Fe(u)}return kn(s)}function jb(e){k.assert(e>0&&Number.isInteger(e),()=>`batchSize is required to be a positive integer, but got ${e}`)}function ip(e,t,n){return e==null?[null]:Array.isArray(e)?e.map(a=>Pi(a,t,n-t)):Pi(e,t,n-t)}function qb(e,t){return D(()=>e==null?null:Array.isArray(e)?e.map(n=>qb(n,t)):xk(e,t.dtype==="int32"?t:t.toInt()))}function Kb(e,t){let n=[],a=0,r=null;for(;a<e;)r=a+t,r>=e&&(r=e),n.push([a,r]),a=r;return n}async function fW(e,t,n,a,r,s,i,o,l,c,u,p,d,h,m){r==null&&(r=32),s==null&&(s=1),u==null&&(u=!0),d==null&&(d=0);let f=!1;if(l!=null&&c!=null&&(f=!0),m!=null&&(f=!0,h==null))throw new z("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"),y;g!=null&&(y=Ea(0,g)),i==null&&(i=1);let{callbackList:b,history:v}=Bk(o,i,s,d,g,h,r,f,p);b.setModel(e),e.history=v,await b.onTrainBegin(),e.stopTraining_=!1;for(let x=d;x<s;++x){await b.onEpochBegin(x);let N={};if(h!=null)throw new $e("stepsPerEpoch mode is not implemented yet.");{if(u==="batch")throw new $e("batch shuffling is not implemneted yet");u&&k.shuffle(y);let T=tt(y),C=Kb(g,r);for(let $=0;$<C.length;++$){let F={};if(await b.onBatchBegin($,F),D(()=>{let O=C[$][0],B=C[$][1],W=Pi(T,O,B-O);F.batch=$,F.size=B-O;let G=qb(n,W),q=t(G);for(let H=0;H<a.length;++H){let K=a[H],X=q[H];F[K]=X,jt(X)}if($===C.length-1&&f){let H=e.testLoop(l,c,r);for(let K=0;K<a.length;++K){let X=a[K],ne=H[K];jt(ne),N["val_"+X]=ne}}}),await b.onBatchEnd($,F),Rk(F),e.stopTraining_)break}T.dispose()}if(await b.onEpochEnd(x,N),e.stopTraining_)break}return await b.onTrainEnd(),await e.history.syncData(),e.history}async function gW(e,t,n,a={}){if(e.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");e.isTraining=!0;let r,s,i,o,l,c,u;try{let p=a.batchSize==null?32:a.batchSize;jb(p);let d=!1,h=await e.standardizeUserData(t,n,a.sampleWeight,a.classWeight,d,p);r=h[0],s=h[1],u=h[2];let m=!1,f;if(a.validationData!=null&&a.validationData.length>0){if(m=!0,a.validationData.length===2)i=a.validationData[0],o=a.validationData[1];else throw a.validationData.length===3?new $e("validationData including sample weights is not supported yet."):new z(`When passing validation data, it must contain 2 (valX, valY) or 3 (valX, valY, valSampleWeight) items; ${a.validationData} is invalid.`);let T=!0,C=await e.standardizeUserData(i,o,null,null,T,p);l=C[0],c=C[1],f=l.concat(c)}else if(a.validationSplit!=null&&a.validationSplit>0&&a.validationSplit<1){m=!0;let T=Math.floor(r[0].shape[0]*(1-a.validationSplit)),C=r[0].shape[0];l=ip(r,T,C),r=ip(r,0,T),c=ip(s,T,C),s=ip(s,0,T),f=l.concat(c)}else a.validationSteps!=null&&(m=!0);let g=r.concat(s).concat(u);e.checkTrainableWeightsConsistency();let y=e.makeTrainFunction(),b=e.getDedupedMetricsNames(),v,x;m?(e.makeTestFunction(),v=e.testFunction,x=b.slice().concat(b.map(T=>"val_"+T))):(v=null,f=[],x=b.slice());let N=zk(a.callbacks,a.yieldEvery);return await fW(e,y,g,b,p,a.epochs,a.verbose,N,v,f,a.shuffle,x,a.initialEpoch,null,null)}finally{e.isTraining=!1,zi(r,t),zi(s,n),zi(l,i),zi(c,o),u!=null&&Fe(u)}}function nI(e){let t=[];e instanceof Ae&&(e=[e]);for(let n=0;n<e.length;++n){let a=e[n];if(a.rank===1)t.push(Qc(a,1));else{if(a.rank===0)throw new Error("Expected tensor to be at least 1D, but received a 0D tensor (scalar).");t.push(a)}}return t}function zi(e,t){if(e==null)return;let n=[];if(t instanceof Ae)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 s=t[r];n.push(s.id)}let a=[];if(e instanceof Ae)n.indexOf(e.id)===-1&&a.push(e);else if(Array.isArray(e))e.forEach(r=>{n.indexOf(r.id)===-1&&a.push(r)});else if(e!=null)for(let r in e){let s=e[r];n.indexOf(s.id)===-1&&a.push(s)}a.forEach(r=>{r.isDisposed||r.dispose()})}function yW(e){return e instanceof Ae}function Xb(e){return Array.isArray(e)}function aI(e){return!yW(e)&&!Xb(e)}function rI(e,t,n,a=!0,r=""){if(t==null||t.length===0){if(e!=null){let i=!1;if(Xb(e)&&e.length>0)i=!0;else if(aI(e)){for(let o in e)if(e.hasOwnProperty(o)){i=!0;break}}else i=!0;if(i)throw new z(`Error when checking model ${r} expected no data, but got ${e}`)}return[]}if(e==null)return t.map(i=>null);let s;if(aI(e)){e=e,s=[];for(let i of t){if(e[i]==null)throw new z(`No data provided for "${i}". Need data for each key in: ${t}`);s.push(e[i])}}else if(Xb(e)){if(e=e,e.length!==t.length)throw new z(`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}`);s=e}else{if(e=e,t.length>1)throw new z(`The model ${r} expects ${t.length} Tensor(s), but only received one Tensor. Found: Tensor with shape ${e.shape}`);s=[e]}if(s=nI(s),n!=null)for(let i=0;i<t.length;++i){if(n[i]==null)continue;let o=s[i];if(o.shape.length!==n[i].length)throw new z(`Error when checking ${r}: expected ${t[i]} to have ${n[i].length} dimension(s). but got array with shape ${o.shape}`);for(let l=0;l<n[i].length;++l){if(l===0&&!a)continue;let c=o.shape[l],u=n[i][l];if(u!=null&&u>=0&&c!==u)throw new z(`Error when checking ${r}: expected ${t[i]} to have shape [${n[i]}], but got array with shape [${o.shape}].`)}}return s}function bW(e,t,n){let a=ts(e.map(s=>s.shape[0]));a.sort();let r=ts(t.map(s=>s.shape[0]));if(r.sort(),a.length>1)throw new z(`All input Tensors (x) should have the same number of samples. Got array shapes: ${JSON.stringify(e.map(s=>s.shape))}`);if(r.length>1)throw new z(`All target Tensors (y) should have the same number of samples. Got array shapes: ${JSON.stringify(t.map(s=>s.shape))}`);if(a.length>0&&r.length>0&&!k.arraysEqual(a,r))throw new z(`Input Tensors should have the same number of samples as target Tensors. Found ${a[0]} input sample(s) and ${r[0]} target sample(s).`)}function xW(e,t,n){let a=[Oi,um,ap];for(let r=0;r<e.length;++r){let s=e[r],i=t[r],o=n[r];if(i!=null){if(i===ap&&s.shape[s.shape.length-1]===1)throw new z(`You are passing a target array of shape ${s.shape} while using a loss 'categorical_crossentropy'. 'categorical_crossentropy'expects targets to be binary matrices (1s and 0s) of shape [samples, classes].`);if(a.indexOf(i)!==-1){let l=s.shape.slice(1),c=o.slice(1);for(let u=0;u<l.length;++u){let p=l[u],d=c[u];if(d!=null&&p!==d)throw new z(`A target Tensor with shape ${s.shape} was passed for an output of shape ${o}, while using a loss function that expects targets to have the same shape as the output.`)}}}}}function sI(e,t,n,a=!0,r=""){let s;if(Array.isArray(e)){if(e.length!==t.length)throw new z(`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).`);s=e}else{if(t.length>1)throw new z(`The model expects ${t.length} ${r} Tensors, but only received one Tensor. Found: array with shape ${JSON.stringify(e.shape)}.`);s=[e]}if(n!=null)for(let i=0;i<t.length;++i){if(n[i]==null)continue;let o=s[i];if(o.shape.length!==n[i].length)throw new z(`Error when checking ${r}: expected ${t[i]} to have ${n[i].length} dimension(s), but got array with shape ${JSON.stringify(o.shape)}`);for(let l=0;l<n[i].length;++l){if(l===0&&!a)continue;let c=o.shape[l],u=n[i][l];if(u!=null&&u!==c)throw new z(`Error when checking ${r}: expected ${t[i]} to have shape ${JSON.stringify(n[i])} but got array with shape ${JSON.stringify(o.shape)}.`)}}}function vW(e,t){if(e==null||Array.isArray(e)&&e.length===0)return t.map(a=>[]);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: ${e}`);if(Array.isArray(n))return t.map(a=>n);{let a=[];for(let r of t){let s=n.hasOwnProperty(r)?n[r]:[];Array.isArray(s)||(s=[s]),a.push(s)}return a}}var wW="layers-model",wr=class extends nr{constructor(e){super(e);this.isTraining=!1}summary(e,t,n=console.log){if(!this.built)throw new z("This model has never been called, thus its weights have not been created yet. So no summary can be displayed. Build the model first (e.g., by calling it on some test data).");nW(this,e,t,n)}compile(e){if(e.loss==null&&(e.loss=[]),this.loss=e.loss,typeof e.optimizer=="string")this.optimizer_=ZB(e.optimizer),this.isOptimizerOwned=!0;else{if(!(e.optimizer instanceof br))throw new z("User-defined optimizer must be an instance of tf.Optimizer.");this.optimizer_=e.optimizer,this.isOptimizerOwned=!1}let t=[];if(!Array.isArray(e.loss)&&typeof e.loss!="string"&&typeof e.loss!="function"){e.loss=e.loss;for(let s in e.loss)if(this.outputNames.indexOf(s)===-1)throw new z(`Unknown entry in loss dictionary: "${s}". Only expected the following keys: ${this.outputNames}`);for(let s of this.outputNames)e.loss[s]==null&&console.warn(`Output "${s}" is missing from loss dictionary. We assume this was done on purpose, and we will not be expecting data to be passed to ${s} during training`),t.push(zb(e.loss[s]))}else if(Array.isArray(e.loss)){if(e.loss.length!==this.outputs.length)throw new z(`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(s=>zb(s))}else{let s=zb(e.loss);this.outputs.forEach(i=>{t.push(s)})}this.lossFunctions=t,this.feedOutputNames=[],this.feedOutputShapes=[],this.feedLossFns=[];for(let s=0;s<this.outputs.length;++s){let i=this.internalOutputShapes[s],o=this.outputNames[s];this.feedOutputNames.push(o),this.feedOutputShapes.push(i),this.feedLossFns.push(this.lossFunctions[s])}let n=[];this.metrics=e.metrics,this.metricsNames=["loss"],this.metricsTensors=[],Mi("loss",()=>{for(let s=0;s<this.outputs.length;++s){if(n.indexOf(s)!==-1)continue;let i=this.lossFunctions[s];this.outputs.length>1&&(this.metricsTensors.push([i,s]),this.metricsNames.push(this.outputNames[s]+"_loss"))}});let a=vW(e.metrics,this.outputNames),r=(s,i,o)=>{this.outputNames.length>1&&(i=this.outputNames[s]+"_"+i),this.metricsNames.push(i),this.metricsTensors.push([o,s])};Mi("metric",()=>{for(let s=0;s<this.outputs.length;++s){if(n.indexOf(s)!==-1)continue;let i=a[s];(o=>{let l="",c,u,p;for(let d of o){if(typeof d=="string"&&["accuracy","acc","crossentropy","ce"].indexOf(d)!==-1){let m=this.internalOutputShapes[s];m[m.length-1]===1||this.lossFunctions[s]===um?["accuracy","acc"].indexOf(d)!==-1?u=Bb:["crossentropy","ce"].indexOf(d)!==-1&&(u=Uk):this.lossFunctions[s]===lm?["accuracy","acc"].indexOf(d)!==-1?u=Gk:["crossentropy","ce"].indexOf(d)!==-1&&(u=Hk):["accuracy","acc"].indexOf(d)!==-1?u=Wb:["crossentropy","ce"].indexOf(d)!==-1&&(u=Vb);let f;["accuracy","acc"].indexOf(d)!==-1?f="acc":["crossentropy","ce"].indexOf(d)!==-1&&(f="ce"),p=u,c=l+f}else p=YB(d),c=l+dm(d);let h;Mi(c,()=>{h=p}),r(s,c,h)}})(i)}}),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 a=n.batchSize==null?32:n.batchSize;jb(a);let r=!0,s=this.standardizeUserDataXY(e,t,r,a);try{let i=s[0].concat(s[1]);this.makeTestFunction();let o=this.testFunction,l=this.testLoop(o,i,a,n.verbose,n.steps);return kn(l)}finally{zi(s[0],e),zi(s[1],t)}}async evaluateDataset(e,t){return this.makeTestFunction(),mW(this,e,t)}checkNumSamples(e,t,n,a="steps"){let r;if(n!=null){if(r=null,t!=null)throw new z(`If ${a} 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 z(`Either the input data should have a defined shape, or ${a} shoud be specified.`);return r}execute(e,t){if(Array.isArray(t)&&t.length===0)throw new z("`outputs` is an empty Array, which is not allowed.");let n=Array.isArray(t),a=n?t:[t],r=this.retrieveSymbolicTensors(a),s=new Li;if(e instanceof Ae&&(e=[e]),Array.isArray(e)){if(e.length!==this.inputs.length)throw new z(`The number of inputs provided (${e.length}) does not match the number of inputs of this model (${this.inputs.length}).`);for(let o=0;o<this.inputs.length;++o)s.add(this.inputs[o],e[o])}else for(let o of this.inputs){let l=e[o.name];if(l==null)throw new z(`No value is provided for the model's input ${o.name}`);s.add(o,l)}let i=sp(r,s);return n?i:i[0]}retrieveSymbolicTensors(e){let t=$i(null,e.length),n=e.length;for(let a of this.layers){let r=Array.isArray(a.output)?a.output:[a.output],s=r.map(i=>i.name);for(let i=0;i<e.length;++i){let o=s.indexOf(e[i]);if(o!==-1&&(t[i]=r[o],n--),n===0)break}if(n===0)break}if(n>0){let a=[];throw t.forEach((r,s)=>{r==null&&a.push(e[s])}),new z(`Cannot find SymbolicTensors for output name(s): ${JSON.stringify(a)}`)}return t}predictLoop(e,t=32,n=!1){return D(()=>{let a=this.checkNumSamples(e);if(n)throw new $e("Verbose predictLoop() is not implemented yet.");let r=Kb(a,t),s=this.outputs.map(i=>[]);for(let i=0;i<r.length;++i)D(()=>{let o=r[i][0],l=r[i][1],c=ip(e,o,l),u=[];if(Array.isArray(c))for(let d=0;d<c.length;++d)u.push({key:this.inputs[d],value:c[d]});else u.push({key:this.inputs[0],value:c});let p=new Li(u);return sp(this.outputs,p)}).forEach((o,l)=>s[l].push(o));return kn(s.map(i=>Qe(i,0)))})}predict(e,t={}){let n=nI(e);sI(n,this.inputNames,this.feedInputShapes,!1);try{let a=t.batchSize==null?32:t.batchSize;return jb(a),this.predictLoop(n,a)}finally{zi(n,e)}}predictOnBatch(e){sI(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,a){if(this.optimizer_==null)throw new _a("You must compile a model before training/testing. Use LayersModel.compile(modelCompileArgs).");let r=[];for(let s=0;s<this.feedOutputShapes.length;++s){let i=this.feedOutputShapes[s];this.feedLossFns[s]===lm?r.push(i.slice(0,i.length-1).concat([1])):r.push(i)}if(e=rI(e,this.feedInputNames,this.feedInputShapes,!1,"input"),t=rI(t,this.feedOutputNames,r,!1,"target"),bW(e,t,null),xW(t,this.feedLossFns,this.feedOutputShapes),this.stateful&&a!=null&&a>0&&e[0].shape[0]%a!=0)throw new z(`In a stateful network, you should only pass inputs with a number of samples that is divisible by the batch size ${a}. Found: ${e[0].shape[0]} sample(s).`);return[e,t]}async standardizeUserData(e,t,n,a,r=!0,s){let[i,o]=this.standardizeUserDataXY(e,t,r,s);if(n!=null)throw new Error("sample weight is not supported yet.");let l=null;if(a!=null){let c=Zk(a,this.outputNames);l=[];for(let u=0;u<c.length;++u)l.push(await Jk(o[u],null,c[u]))}return[i,o,l]}testLoop(e,t,n,a=0,r){return D(()=>{let s=this.checkNumSamples(t,n,r,"steps"),i=[];if(a>0)throw new $e("Verbose mode is not implemented yet.");if(r!=null)throw new $e("steps mode in testLoop() is not implemented yet");{let o=Kb(s,n),l=tt(Ea(0,s));for(let c=0;c<o.length;++c){let u=o[c][0],p=o[c][1],d=Pi(l,u,p-u),h=qb(t,d),m=e(h);if(c===0)for(let f=0;f<m.length;++f)i.push(de(0));for(let f=0;f<m.length;++f){let g=m[f];i[f]=Z(i[f],L(p-u,g))}}for(let c=0;c<i.length;++c)i[c]=ve(i[c],s)}return i})}getDedupedMetricsNames(){let e=this.metricsNames,t=[];for(let n=0;n<e.length;++n){let a=e[n],r=a;ik(e,a)>1&&(r+=`_${ik(e.slice(0,n),a)}`),t.push(r)}return t}makeTrainFunction(){return e=>{let t=[],n=e.slice(0,this.inputs.length),a=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),s=[],i=()=>{let c=[];for(let h=0;h<this.inputs.length;++h)c.push({key:this.inputs[h],value:n[h]});let u=new Li(c),p=sp(this.outputs,u,{training:!0}),d;for(let h=0;h<this.lossFunctions.length;++h){let m=this.lossFunctions[h](a[h],p[h]);r[h]!=null&&(m=lW(m,r[h]));let f=St(m);t.push(f),h===0?d=m:d=Z(d,m)}for(let h=0;h<this.metricsTensors.length;++h){let m;if(this.outputs.length>1&&h<this.outputs.length)m=t[h];else{let f=this.metricsTensors[h][0],g=this.metricsTensors[h][1];m=St(f(a[g],p[g]))}jt(m),s.push(m)}return d=St(d),this.calculateLosses().forEach(h=>{d=Z(d,h)}),d},o=this.collectedTrainableWeights.map(c=>c.read()),l=!0;return[this.optimizer_.minimize(i,l,o)].concat(s)}}makeTestFunction(){this.testFunction=e=>D(()=>{let t=[],n,a=e.slice(0,this.inputs.length),r=e.slice(this.inputs.length,this.inputs.length+this.outputs.length),s=[];for(let l=0;l<this.inputs.length;++l)s.push({key:this.inputs[l],value:a[l]});let i=new Li(s),o=sp(this.outputs,i);for(let l=0;l<this.lossFunctions.length;++l){let c=this.lossFunctions[l],u=St(c(r[l],o[l]));l===0?n=u:n=Z(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],p=St(c(r[u],o[u]));t.push(p)}return t})}async fit(e,t,n={}){return gW(this,e,t,n)}async fitDataset(e,t){return dW(this,e,t)}async trainOnBatch(e,t){let n=await this.standardizeUserData(e,t),a=n[0],r=n[1],s=this.makeTrainFunction()(a.concat(r)),i=[];for(let o of s){let l=await o.data();i.push(l[0])}return Fe(s),kn(i)}getNamedWeights(e){let t=[],n=e!=null&&e.trainableOnly,a=n?this.trainableWeights:this.weights,r=this.getWeights(n);for(let s=0;s<a.length;++s)n&&!a[s].trainable||t.push({name:a[s].originalName,tensor:r[s]});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=sh().numTensors;this.optimizer_.dispose(),e.numDisposedVariables+=t-sh().numTensors}return e}getLossIdentifiers(){let e;if(typeof this.loss=="string")e=vr(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=>vr(t))}else{let t=Object.keys(this.loss);e={};let n=this.loss;for(let a of t)if(typeof n[a]=="string")e[a]=vr(n[a]);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[vr(dm(this.metrics))];if(Array.isArray(this.metrics))return this.metrics.map(e=>vr(dm(e)));{let e={};for(let t in this.metrics)e[t]=vr(dm(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=rp(e.optimizer_config),n=Fa(t),a;if(typeof e.loss=="string")a=Di(e.loss);else if(Array.isArray(e.loss))a=e.loss.map(s=>Di(s));else if(e.loss!=null){a={};for(let s in e.loss)a[s]=Di(e.loss[s])}let r;if(Array.isArray(e.metrics))r=e.metrics.map(s=>Di(s));else if(e.metrics!=null){r={};for(let s in e.metrics)r[s]=Di(e.metrics[s])}this.compile({loss:a,metrics:r,optimizer:n})}async save(e,t){if(typeof e=="string"){let i=Ht.getSaveHandlers(e);if(i.length===0)throw new z(`Cannot find any save handlers for URL '${e}'`);if(i.length>1)throw new z(`Found more than one (${i.length}) save handlers for URL '${e}'`);e=i[0]}if(e.save==null)throw new z("LayersModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");let n=await Ht.encodeWeights(this.getNamedWeights(t)),a=!1,r=null,s={modelTopology:this.toJSON(r,a),format:wW,generatedBy:`TensorFlow.js tfjs-layers v${mm}`,convertedBy:null};if((t==null?!1:t.includeOptimizer)&&this.optimizer!=null){s.trainingConfig=this.getTrainingConfig();let i="optimizer",{data:o,specs:l}=await Ht.encodeWeights(await this.optimizer.getWeights(),i);n.specs.push(...l),n.data=Ht.concatenateArrayBuffers([n.data,o])}if(this.userDefinedMetadata!=null){let i=!0;qk(this.userDefinedMetadata,this.name,i),s.userDefinedMetadata=this.userDefinedMetadata}return s.weightData=n.data,s.weightSpecs=n.specs,e.save(s)}setUserDefinedMetadata(e){qk(e,this.name),this.userDefinedMetadata=e}getUserDefinedMetadata(){return this.userDefinedMetadata}};wr.className="Model";re.registerClass(wr);var iI=class extends wr{};iI.className="Functional";re.registerClass(iI);async function kW(e,t){"modelTopology"in e||(e={modelTopology:e}),e=e;let n=e.modelTopology;n.model_config!=null&&(n=n.model_config);let a=rp(n),r=Fa(a,t);if(e.weightsManifest!=null){let s=await Ht.loadWeights(e.weightsManifest,e.pathPrefix,r.weights.map(o=>o.originalName)),i={};for(let o of r.weights)i[o.originalName]=s[o.originalName];r.loadWeights(i),Fe(s)}return r}async function NW(e,t){if(t==null&&(t={}),typeof e=="string"){let n=Ht.getLoadHandlers(e,t);if(n.length===0)n.push(Ht.browserHTTPRequest(e,t));else if(n.length>1)throw new z(`Found more than one (${n.length}) load handlers for URL '${e}'`);e=n[0]}return IW(e,void 0,t)}async function IW(e,t,n){if(n==null&&(n={}),e.load==null)throw new z("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");let a=await e.load(),r=a.modelTopology;r.model_config!=null&&(r=r.model_config);let s=n.strict==null?!0:n.strict,i=a.weightData!=null&&a.weightSpecs!=null&&s,o=Fa(rp(r),t,i),l=a.trainingConfig;if(l!=null&&o.loadTrainingConfig(l),a.userDefinedMetadata!=null&&o.setUserDefinedMetadata(a.userDefinedMetadata),a.weightData!=null){if(a.weightSpecs==null)throw new z("LayersModel artifacts contains weight data, but not weight specs. Therefore loading of weights cannot proceed.");let{modelWeights:c,optimizerWeights:u}=TW(a.weightData,a.weightSpecs);o.loadWeights(c,s),o.optimizer!=null&&u.length>0&&await o.optimizer.setWeights(u),Fe(c),Fe(u.map(p=>p.tensor))}return o}function TW(e,t){let n=Ht.decodeWeights(e,t),a={},r=[];return t.forEach(s=>{s.group==="optimizer"?r.push({name:s.name,tensor:n[s.name]}):a[s.name]=n[s.name]}),{modelWeights:a,optimizerWeights:r}}var Zl=class extends wr{constructor(e){super({inputs:[],outputs:[]});if(e=e||{},this.trainable=!0,this.built=!1,this.name=e.name!=null?e.name:nm("sequential_"),e.layers!=null)for(let t of e.layers)this.add(t)}checkShape(e){if(e.inboundNodes[0].outputTensors[0].shape.some(t=>t<0))throw new z(`Negative dimension size caused by adding layer ${e.name} with input shape [${e.inboundNodes[0].inputTensors[0].shape}]`)}add(e){let t=e instanceof Zl||e instanceof wr,n;if(t){if(n=e,n.outputs.length!==1)throw new z("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 z("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 z("The first layer in a Sequential model must get an `inputShape` or `batchInputShape` argument.");let a=Dk({batchShape:e.batchInputShape,dtype:e.dtype,name:e.name+"_input"});e.apply(a)}if(t)this.outputs=n.outputs,this.inputs=n.inputs;else{if(e.inboundNodes.length!==1)throw new z(`A layer added to a Sequential model must not already be connected somewhere else. LayersModel received layer ${e.name} which has ${e.inboundNodes.length} pre-existing inbound connections.`);if(e.inboundNodes[0].outputTensors.length!==1)throw new z("All layers in a Sequential model should have a single output tensor. For multi-output layers, use the functional API.");this.checkShape(e),this.outputs=[e.inboundNodes[0].outputTensors[0]],this.inputs=$k(this.outputs[0])}this.inboundNodes=[],new sm({outboundLayer:this,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:this.inputs,outputTensors:this.outputs,inputMasks:$i(null,this.inputs.length),outputMasks:[null],inputShapes:this.inputs.map(a=>a.shape),outputShapes:this.outputs[0].shape})}else{let a=e.apply(this.outputs[0]);if(Array.isArray(a))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=[a],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(ht(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 wr({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 _a("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 _a("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 _a("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 _a("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={},a=!1){let r,s={};if(t instanceof Array){if(t[0].className==null||t[0].className==="Merge")throw new z("Legacy serialization format not supported yet.");r=t}else k.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,s=t;let i=new e(s);if(!(i instanceof Zl))throw new $e(`Sequential.fromConfig called on non-Sequential input: ${i}`);for(let o of r){let l=Fa(o,void 0,a);a&&l.setFastWeightInitDuringBuild(!0),i.add(l)}return i}set stopTraining(e){if(this.model==null)throw new z("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 z("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}}};Zl.className="Sequential";re.registerClass(Zl);function SW(e){return new wr(e)}function CW(e){return new Zl(e)}function _W(e,t){return t==null&&(t={}),NW(e,t)}function Nk(e){return Dk(e)}function EW(e,t){ma.registerCallbackConstructor(e,t)}var Pn=class extends re.Serializable{getConfig(){return{}}},oI=class extends Pn{apply(e,t=1){return oz(e,t)}};oI.className="elu";re.registerClass(oI);var lI=class extends Pn{apply(e){return Th(e)}};lI.className="selu";re.registerClass(lI);var uI=class extends Pn{apply(e){return Ye(e)}};uI.className="relu";re.registerClass(uI);var cI=class extends Pn{apply(e){return D(()=>Bl(6,Ye(e)))}};cI.className="relu6";re.registerClass(cI);var pI=class extends Pn{apply(e){return e}};pI.className="linear";re.registerClass(pI);var dI=class extends Pn{apply(e){return ua(e)}};dI.className="sigmoid";re.registerClass(dI);var hI=class extends Pn{apply(e){return uz(e)}};hI.className="hardSigmoid";re.registerClass(hI);var mI=class extends Pn{apply(e){return Ll(e)}};mI.className="softplus";re.registerClass(mI);var fI=class extends Pn{apply(e){return lz(e)}};fI.className="softsign";re.registerClass(fI);var gI=class extends Pn{apply(e){return Ml(e)}};gI.className="tanh";re.registerClass(gI);var Yb=class extends Pn{apply(e,t=-1){return Na(e,t)}};Yb.className="softmax";re.registerClass(Yb);var yI=class extends Pn{apply(e,t=-1){return bh(e,t)}};yI.className="logSoftmax";re.registerClass(yI);var bI=class extends Pn{apply(e,t=1){return D(()=>ua(e.mul(t)).mul(e))}};bI.className="swish";re.registerClass(bI);function ss(e){return e.getClassName()}function Zb(e,t={}){return Xc(e,re.SerializationMap.getMap().classNameMap,t,"activation")}function is(e){if(e==null){let t={};return t.className="linear",t.config={},Zb(t)}if(typeof e=="string"){let t={};return t.className=e,t.config={},Zb(t)}else return e instanceof Pn?e:Zb(e)}function Jb(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 xI=class extends re.Serializable{},op=class extends xI{constructor(e){super();Jb(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 D(()=>{let t=vt([1]);return this.hasL1&&(t=Z(t,Te(L(this.l1,Lt(e))))),this.hasL2&&(t=Z(t,Te(L(this.l2,ep(e))))),t.asScalar()})}getConfig(){return{l1:this.l1,l2:this.l2}}static fromConfig(e,t){return new e({l1:t.l1,l2:t.l2})}};op.className="L1L2";re.registerClass(op);function AW(e){return Jb(e),new op({l1:e!=null?e.l1:null,l2:0})}function FW(e){return Jb(e),new op({l2:e!=null?e.l2:null,l1:0})}var vI={l1l2:"L1L2"};function mt(e){return hb(e)}function wI(e,t={}){return Xc(e,re.SerializationMap.getMap().classNameMap,t,"regularizer")}function kt(e){if(e==null)return null;if(typeof e=="string"){let t={className:e in vI?vI[e]:e,config:{}};return wI(t)}else return e instanceof xI?e:wI(e)}var Qb=class extends qe{constructor(e){super(e==null?{}:e);this.supportsMasking=!0,e!=null&&(this.maxValue=e.maxValue)}call(e,t){e=Me(e);let n=Ye(e);return this.maxValue!=null&&(n=qt(n,0,this.maxValue)),n}computeOutputShape(e){return e}getConfig(){let e={maxValue:this.maxValue},t=super.getConfig();return Object.assign(e,t),e}};Qb.className="ReLU";re.registerClass(Qb);var ex=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=Me(e);return Bc(n,this.alpha)}computeOutputShape(e){return e}getConfig(){let e={alpha:this.alpha},t=super.getConfig();return Object.assign(e,t),e}};ex.className="LeakyReLU";re.registerClass(ex);var tx=class extends qe{constructor(e){super(e==null?{}:e);if(this.DEFAULT_ALPHA_INITIALIZER="zeros",e==null&&(e={}),this.supportsMasking=!0,this.alphaInitializer=wt(e.alphaInitializer||this.DEFAULT_ALPHA_INITIALIZER),this.alphaRegularizer=kt(e.alphaRegularizer),this.alphaConstraint=Vt(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 z(`Expected sharedAxes to be a number or an array of numbers, but got ${e.sharedAxes}`)}build(e){e=ht(e);let t=e.slice(1);if(this.sharedAxes!=null)for(let a of this.sharedAxes)t[a-1]=1;this.alpha=this.addWeight("alpha",t,"float32",this.alphaInitializer,this.alphaRegularizer,!0,this.alphaConstraint);let n={};if(this.sharedAxes!=null)for(let a=1;a<e.length;++a)n[a]=e[a];this.inputSpec=[new Xt({ndim:e.length,axes:n})],this.built=!0}call(e,t){return e=Me(e),Uc(e,this.alpha.read())}getConfig(){let e={alphaInitializer:Ct(this.alphaInitializer),alphaRegularizer:mt(this.alphaRegularizer),alphaConstraint:Wt(this.alphaConstraint),sharedAxes:this.sharedAxes},t=super.getConfig();return Object.assign(e,t),e}};tx.className="PReLU";re.registerClass(tx);var nx=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 $e(`Non-default alpha value (${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=Me(e);return Pl(n)}computeOutputShape(e){return e}getConfig(){let e={alpha:this.alpha},t=super.getConfig();return Object.assign(e,t),e}};nx.className="ELU";re.registerClass(nx);var ax=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=Me(e);return n.mul(Jc(n.greater(this.theta),"float32"))}computeOutputShape(e){return e}getConfig(){let e={theta:this.theta},t=super.getConfig();return Object.assign(e,t),e}};ax.className="ThresholdedReLU";re.registerClass(ax);var rx=class extends qe{constructor(e){super(e==null?{}:e);this.DEFAULT_AXIS=1,e==null&&(e={}),this.softmax=new Yb().apply,this.axis=e.axis==null?this.DEFAULT_AXIS:e.axis}call(e,t){let n=Me(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}};rx.className="Softmax";re.registerClass(rx);function Jl(e,t,n){if(typeof e=="number")return $i(e,t);if(e.length!==t)throw new z(`The ${n} argument must be an integer or tuple of ${t} integers. Received: ${e.length} elements.`);for(let a=0;a<t;++a){let r=e[a];if(!az(r))throw new z(`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 $a(e,t,n,a,r=1){if(e==null)return e;let s=t+(t-1)*(r-1),i;return n==="same"?i=e:i=e-s+1,Math.floor((i+a-1)/a)}function fm(e,t,n,a){if(e==null)return null;if(a==="valid")e=e*t+as([n-t,0]);else if(a==="same")e=e*t;else throw new z(`Unsupport padding mode: ${a}.`);return e}function sx(e,t){return D(()=>(Dt(t),t==="channelsFirst"?Ue(e,[0,2,3,1]):e))}function kI(e,t){return D(()=>(Dt(t),t==="channelsFirst"?Ue(e,[0,2,3,4,1]):e))}function $W(e,t,n,a=1,r="valid",s,i=1){return D(()=>{if(s==null&&(s=Ca()),Dt(s),e.shape.length!==3)throw new z(`The input of a conv1dWithBias operation should be 3, but is ${e.shape.length} instead.`);if(t.shape.length!==3)throw new z(`The kernel for a conv1dWithBias operation should be 3, but is ${t.shape.length} instead`);if(n!=null&&n.shape.length!==1)throw new z(`The bias for a conv1dWithBias operation should be 1, but is ${t.shape.length} instead`);if(s==="channelsFirst"&&(e=Ue(e,[0,2,1])),r==="causal")throw new $e("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");let o=ch(e,t,a,r==="same"?"same":"valid","NWC",i);return n!=null&&(o=er(o,n)),o})}function II(e,t,n,a=[1,1],r="valid",s,i,o=null){return D(()=>{if(s==null&&(s=Ca()),Dt(s),e.rank!==3&&e.rank!==4)throw new z(`conv2dWithBiasActivation expects input to be of rank 3 or 4, but received ${e.rank}.`);if(t.rank!==3&&t.rank!==4)throw new z(`conv2dWithBiasActivation expects kernel to be of rank 3 or 4, but received ${e.rank}.`);let l=sx(e,s);if(r==="causal")throw new $e("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");return l=es.conv2d({x:l,filter:t,strides:a,pad:r==="same"?"same":"valid",dilations:i,dataFormat:"NHWC",bias:n,activation:o}),s==="channelsFirst"&&(l=Ue(l,[0,3,1,2])),l})}function DW(e,t,n,a=[1,1,1],r="valid",s,i){return D(()=>{if(s==null&&(s=Ca()),Dt(s),e.rank!==4&&e.rank!==5)throw new z(`conv3dWithBias expects input to be of rank 4 or 5, but received ${e.rank}.`);if(t.rank!==4&&t.rank!==5)throw new z(`conv3dWithBias expects kernel to be of rank 4 or 5, but received ${e.rank}.`);let o=kI(e,s);if(r==="causal")throw new $e("The support for CAUSAL padding mode in conv3dWithBias is not implemented yet.");return o=Py(o,t,a,r==="same"?"same":"valid","NDHWC",i),n!=null&&(o=er(o,n)),s==="channelsFirst"&&(o=Ue(o,[0,4,1,2,3])),o})}var ix=class extends qe{constructor(e,t){super(t);if(this.bias=null,this.DEFAULT_KERNEL_INITIALIZER="glorotNormal",this.DEFAULT_BIAS_INITIALIZER="zeros",ix.verifyArgs(t),this.rank=e,Kt(this.rank,"rank"),this.rank!==1&&this.rank!==2&&this.rank!==3)throw new $e(`Convolution layer for rank other than 1, 2, or 3 (${this.rank}) is not implemented yet.`);if(this.kernelSize=Jl(t.kernelSize,e,"kernelSize"),this.strides=Jl(t.strides==null?1:t.strides,e,"strides"),this.padding=t.padding==null?"valid":t.padding,Jn(this.padding),this.dataFormat=t.dataFormat==null?"channelsLast":t.dataFormat,Dt(this.dataFormat),this.activation=is(t.activation),this.useBias=t.useBias==null?!0:t.useBias,this.biasInitializer=wt(t.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.biasConstraint=Vt(t.biasConstraint),this.biasRegularizer=kt(t.biasRegularizer),this.activityRegularizer=kt(t.activityRegularizer),this.dilationRate=Jl(t.dilationRate==null?1:t.dilationRate,e,"dilationRate"),this.rank===1&&Array.isArray(this.dilationRate)&&this.dilationRate.length!==1)throw new z(`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 z(`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 z(`dilationRate must be a number or array of three numbers for 3D convolution, but received ${JSON.stringify(this.dilationRate)}`)}}static verifyArgs(e){if(Ja("kernelSize"in e,"required key 'kernelSize' not in config"),typeof e.kernelSize!="number"&&!fb(e.kernelSize,"number",1,3))throw new z(`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:ss(this.activation),useBias:this.useBias,biasInitializer:Ct(this.biasInitializer),biasRegularizer:mt(this.biasRegularizer),activityRegularizer:mt(this.activityRegularizer),biasConstraint:Wt(this.biasConstraint)},t=super.getConfig();return Object.assign(e,t),e}},lp=class extends ix{constructor(e,t){super(e,t);this.kernel=null,lp.verifyArgs(t),this.filters=t.filters,Kt(this.filters,"filters"),this.kernelInitializer=wt(t.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.kernelConstraint=Vt(t.kernelConstraint),this.kernelRegularizer=kt(t.kernelRegularizer)}build(e){e=ht(e);let t=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[t]==null)throw new z(`The channel dimension of the input should be defined. Found ${e[t]}`);let n=e[t],a=this.kernelSize.concat([n,this.filters]);this.kernel=this.addWeight("kernel",a,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],null,this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[{ndim:this.rank+2,axes:{[t]:n}}],this.built=!0}call(e,t){return D(()=>{e=Me(e);let n,a=this.bias==null?null:this.bias.read(),r=lk(this.activation.getClassName());if(r!=null&&this.rank===2)n=II(e,this.kernel.read(),a,this.strides,this.padding,this.dataFormat,this.dilationRate,r);else{if(this.rank===1)n=$W(e,this.kernel.read(),a,this.strides[0],this.padding,this.dataFormat,this.dilationRate[0]);else if(this.rank===2)n=II(e,this.kernel.read(),a,this.strides,this.padding,this.dataFormat,this.dilationRate);else if(this.rank===3)n=DW(e,this.kernel.read(),a,this.strides,this.padding,this.dataFormat,this.dilationRate);else throw new $e("convolutions greater than 3D are not implemented yet.");this.activation!=null&&(n=this.activation.apply(n))}return n})}computeOutputShape(e){e=ht(e);let t=[],n=this.dataFormat==="channelsLast"?e.slice(1,e.length-1):e.slice(2);for(let r=0;r<n.length;++r){let s=$a(n[r],this.kernelSize[r],this.padding,this.strides[r],typeof this.dilationRate=="number"?this.dilationRate:this.dilationRate[r]);t.push(s)}let a=[e[0]];return this.dataFormat==="channelsLast"?(a=a.concat(t),a.push(this.filters)):(a.push(this.filters),a=a.concat(t)),a}getConfig(){let e={filters:this.filters,kernelInitializer:Ct(this.kernelInitializer),kernelRegularizer:mt(this.kernelRegularizer),kernelConstraint:Wt(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 z(`Convolution layer expected config.filters to be a 'number' > 0 but got ${JSON.stringify(e.filters)}`)}},up=class extends lp{constructor(e){super(2,e);up.verifyArgs(e)}getConfig(){let e=super.getConfig();return delete e.rank,e}static verifyArgs(e){if(typeof e.kernelSize!="number"&&!fb(e.kernelSize,"number",1,2))throw new z(`Conv2D expects config.kernelSize to be number or number[] with length 1 or 2, but received ${JSON.stringify(e.kernelSize)}.`)}};up.className="Conv2D";re.registerClass(up);var gm=class extends lp{constructor(e){super(3,e);gm.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 z(`Conv3D expects config.kernelSize to be number or [number, number, number], but received ${JSON.stringify(e.kernelSize)}.`)}};gm.className="Conv3D";re.registerClass(gm);var ox=class extends up{constructor(e){super(e);if(this.inputSpec=[new Xt({ndim:4})],this.padding!=="same"&&this.padding!=="valid")throw new z(`Conv2DTranspose currently supports only padding modes 'same' and 'valid', but received padding mode ${this.padding}`)}build(e){if(e=ht(e),e.length!==4)throw new z("Input should have rank 4; Received input shape: "+JSON.stringify(e));let t=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[t]==null)throw new z("The channel dimension of the inputs should be defined. Found `None`.");let n=e[t],a=this.kernelSize.concat([this.filters,n]);this.kernel=this.addWeight("kernel",a,"float32",this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[new Xt({ndim:4,axes:{[t]:n}})],this.built=!0}call(e,t){return D(()=>{let n=Me(e);if(n.shape.length!==4)throw new z(`Conv2DTranspose.call() expects input tensor to be rank-4, but received a tensor of rank-${n.shape.length}`);let a=n.shape,r=a[0],s,i;this.dataFormat==="channelsFirst"?(s=2,i=3):(s=1,i=2);let o=a[s],l=a[i],c=this.kernelSize[0],u=this.kernelSize[1],p=this.strides[0],d=this.strides[1],h=fm(o,p,c,this.padding),m=fm(l,d,u,this.padding),f=[r,h,m,this.filters];this.dataFormat!=="channelsLast"&&(n=Ue(n,[0,2,3,1]));let g=ph(n,this.kernel.read(),f,this.strides,this.padding);return this.dataFormat!=="channelsLast"&&(g=Ue(g,[0,3,1,2])),this.bias!=null&&(g=er(g,this.bias.read(),this.dataFormat)),this.activation!=null&&(g=this.activation.apply(g)),g})}computeOutputShape(e){e=ht(e);let t=e.slice(),n,a,r;this.dataFormat==="channelsFirst"?(n=1,a=2,r=3):(n=3,a=1,r=2);let s=this.kernelSize[0],i=this.kernelSize[1],o=this.strides[0],l=this.strides[1];return t[n]=this.filters,t[a]=fm(t[a],o,s,this.padding),t[r]=fm(t[r],l,i,this.padding),t}getConfig(){let e=super.getConfig();return delete e.dilationRate,e}};ox.className="Conv2DTranspose";re.registerClass(ox);var NI=class extends lp{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 z("The `filters` configuration field is required by SeparableConv, but is unspecified.");if(t.kernelInitializer!=null||t.kernelRegularizer!=null||t.kernelConstraint!=null)throw new z("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 z(`SeparableConv${this.rank}D supports only padding modes: 'same' and 'valid', but received ${JSON.stringify(t.padding)}`);this.depthMultiplier=t.depthMultiplier==null?1:t.depthMultiplier,this.depthwiseInitializer=wt(t.depthwiseInitializer||this.DEFAULT_DEPTHWISE_INITIALIZER),this.depthwiseRegularizer=kt(t.depthwiseRegularizer),this.depthwiseConstraint=Vt(t.depthwiseConstraint),this.pointwiseInitializer=wt(t.depthwiseInitializer||this.DEFAULT_POINTWISE_INITIALIZER),this.pointwiseRegularizer=kt(t.pointwiseRegularizer),this.pointwiseConstraint=Vt(t.pointwiseConstraint)}build(e){if(e=ht(e),e.length<this.rank+2)throw new z(`Inputs to SeparableConv${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 z(`The channel dimension of the inputs should be defined, but found ${JSON.stringify(e[t])}`);let n=e[t],a=this.kernelSize.concat([n,this.depthMultiplier]),r=[];for(let i=0;i<this.rank;++i)r.push(1);r.push(n*this.depthMultiplier,this.filters);let s=!0;this.depthwiseKernel=this.addWeight("depthwise_kernel",a,"float32",this.depthwiseInitializer,this.depthwiseRegularizer,s,this.depthwiseConstraint),this.pointwiseKernel=this.addWeight("pointwise_kernel",r,"float32",this.pointwiseInitializer,this.pointwiseRegularizer,s,this.pointwiseConstraint),this.useBias?this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,s,this.biasConstraint):this.bias=null,this.inputSpec=[new Xt({ndim:this.rank+2,axes:{[t]:n}})],this.built=!0}call(e,t){return D(()=>{e=Me(e);let n;if(this.rank===1)throw new $e("1D separable convolution is not implemented yet.");return this.rank===2&&(this.dataFormat==="channelsFirst"&&(e=Ue(e,[0,2,3,1])),n=Ei(e,this.depthwiseKernel.read(),this.pointwiseKernel.read(),this.strides,this.padding,this.dilationRate,"NHWC")),this.useBias&&(n=er(n,this.bias.read(),this.dataFormat)),this.activation!=null&&(n=this.activation.apply(n)),this.dataFormat==="channelsFirst"&&(n=Ue(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=Ct(this.depthwiseInitializer),e.pointwiseInitializer=Ct(this.pointwiseInitializer),e.depthwiseRegularizer=mt(this.depthwiseRegularizer),e.pointwiseRegularizer=mt(this.pointwiseRegularizer),e.depthwiseConstraint=Wt(this.depthwiseConstraint),e.pointwiseConstraint=Wt(this.pointwiseConstraint),e}};NI.className="SeparableConv";var lx=class extends NI{constructor(e){super(2,e)}};lx.className="SeparableConv2D";re.registerClass(lx);var ym=class extends lp{constructor(e){super(1,e);ym.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"&&!fb(e.kernelSize,"number",1,1))throw new z(`Conv1D expects config.kernelSize to be number or number[] with length 1, but received ${JSON.stringify(e.kernelSize)}.`)}};ym.className="Conv1D";re.registerClass(ym);var ux=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 D(()=>{if(e=Me(e),this.dataFormat==="channelsLast"){let n=jh(e,this.cropping[0][0],e.shape[1]-this.cropping[0][0]-this.cropping[0][1],2);return jh(n,this.cropping[1][0],e.shape[2]-this.cropping[1][1]-this.cropping[1][0],3)}else{let n=jh(e,this.cropping[0][0],e.shape[2]-this.cropping[0][0]-this.cropping[0][1],3);return jh(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}};ux.className="Cropping2D";re.registerClass(ux);var cx=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,Dt(this.dataFormat),this.interpolation=e.interpolation==null?"nearest":e.interpolation,ez(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 D(()=>{let n=Me(e),a=n.shape;if(this.dataFormat==="channelsFirst"){n=Ue(n,[0,2,3,1]);let r=this.size[0]*a[2],s=this.size[1]*a[3],i=this.interpolation==="nearest"?n.resizeNearestNeighbor([r,s]):n.resizeBilinear([r,s]);return Ue(i,[0,3,1,2])}else{let r=this.size[0]*a[1],s=this.size[1]*a[2];return this.interpolation==="nearest"?n.resizeNearestNeighbor([r,s]):n.resizeBilinear([r,s])}})}getConfig(){let e={size:this.size,dataFormat:this.dataFormat},t=super.getConfig();return Object.assign(e,t),e}};cx.className="UpSampling2D";re.registerClass(cx);function RW(e,t,n=[1,1],a="valid",r,s){return D(()=>{r==null&&(r=Ca()),Dt(r);let i=sx(e,r);if(e.rank!==4)throw new z(`Input for depthwiseConv2d is required to be 4-D, but is instead ${e.rank}-D`);if(t.rank!==4)throw new z(`depthwiseKernel is required to be 4-D, but is instead ${t.rank}-D`);return i=Yr(i,t,n,a==="same"?"same":"valid","NHWC",s),r==="channelsFirst"&&(i=Ue(i,[0,3,1,2])),i})}var px=class extends ix{constructor(e){super(2,e);this.depthwiseKernel=null,this.depthMultiplier=e.depthMultiplier==null?1:e.depthMultiplier,this.depthwiseInitializer=wt(e.depthwiseInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.depthwiseConstraint=Vt(e.depthwiseConstraint),this.depthwiseRegularizer=kt(e.depthwiseRegularizer)}build(e){if(e=ht(e),e.length<4)throw new z(`Inputs to DepthwiseConv2D should have rank 4. Received input shape: ${JSON.stringify(e)}.`);let t=this.dataFormat==="channelsFirst"?1:3;if(e[t]==null||e[t]<0)throw new z(`The channel dimension of the inputs to DepthwiseConv2D should be defined, but is not (${e[t]}).`);let n=e[t],a=[this.kernelSize[0],this.kernelSize[1],n,this.depthMultiplier];this.depthwiseKernel=this.addWeight("depthwise_kernel",a,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 D(()=>{e=Me(e);let n=RW(e,this.depthwiseKernel.read(),this.strides,this.padding,this.dataFormat,null);return this.useBias&&(n=er(n,this.bias.read(),this.dataFormat)),this.activation!=null&&(n=this.activation.apply(n)),n})}computeOutputShape(e){e=ht(e);let t=this.dataFormat==="channelsFirst"?e[2]:e[1],n=this.dataFormat==="channelsFirst"?e[3]:e[2],a=this.dataFormat==="channelsFirst"?e[1]*this.depthMultiplier:e[3]*this.depthMultiplier,r=$a(t,this.kernelSize[0],this.padding,this.strides[0]),s=$a(n,this.kernelSize[1],this.padding,this.strides[1]);return this.dataFormat==="channelsFirst"?[e[0],a,r,s]:[e[0],r,s,a]}getConfig(){let e=super.getConfig();return e.depthMultiplier=this.depthMultiplier,e.depthwiseInitializer=Ct(this.depthwiseInitializer),e.depthwiseRegularizer=mt(this.depthwiseRegularizer),e.depthwiseConstraint=Wt(this.depthwiseRegularizer),e}};px.className="DepthwiseConv2D";re.registerClass(px);function TI(e,t,n,a){if(Array.isArray(e)){if(t!=null||n!=null)throw new z("When inputs is an array, neither initialState or constants should be provided");a!=null&&(n=e.slice(e.length-a,e.length),e=e.slice(0,e.length-a)),e.length>1&&(t=e.slice(1,e.length)),e=e[0]}function r(s){return s==null||Array.isArray(s)?s:[s]}return t=r(t),n=r(n),{inputs:e,initialState:t,constants:n}}function SI(e,t,n,a=!1,r,s,i=!1,o=!1){return D(()=>{let l=t.shape.length;if(l<3)throw new z(`Input should be at least 3D, but is ${l}D.`);let c=[1,0].concat(Ea(2,l));if(t=Ue(t,c),s!=null)throw new $e("The rnn() functoin of the deeplearn.js backend does not support constants yet.");i&&console.warn("Backend rnn(): the unroll = true option is not applicable to the imperative deeplearn.js backend."),r!=null&&(r=r.asType("bool").asType("float32"),r.rank===l-1&&(r=Fn(r,-1)),r=Ue(r,c)),a&&(t=Rn(t,0),r!=null&&(r=Rn(r,0)));let u=[],p,d=n,h=t.shape[0],m=dt(t),f;r!=null&&(f=dt(r));for(let y=0;y<h;++y){let b=m[y],v=D(()=>e(b,d));if(r==null)p=v[0],d=v[1];else{let x=D(()=>{let N=f[y],T=Dn(N).sub(N),C=v[0].mul(N).add(d[0].mul(T)),$=d.map((F,O)=>v[1][O].mul(N).add(F.mul(T)));return{output:C,newStates:$}});p=x.output,d=x.newStates}o&&u.push(p)}let g;return o&&(g=$t(u,1)),[p,g,d]})}var tr=class extends qe{constructor(e){super(e);let t;if(e.cell==null)throw new z("cell property is missing for the constructor of RNN.");if(Array.isArray(e.cell)?t=new bm({cells:e.cell}):t=e.cell,t.stateSize==null)throw new z("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 Xt({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 Ea(0,e).map(t=>null)}else return this.states_}setStates(e){this.states_=e}computeOutputShape(e){Mb(e)&&(e=e[0]),e=e;let t=this.cell.stateSize;Array.isArray(t)||(t=[t]);let n=t[0],a;if(this.returnSequences?a=[e[0],e[1],n]:a=[e[0],n],this.returnState){let r=[];for(let s of t)r.push([e[0],s]);return[a].concat(r)}else return a}computeMask(e,t){return D(()=>{Array.isArray(t)&&(t=t[0]);let n=this.returnSequences?t:null;if(this.returnState){let a=this.states.map(r=>null);return[n].concat(a)}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 $e("Constants support is not implemented in RNN yet.");Mb(e)&&(e=e[0]),e=e;let n=this.stateful?e[0]:null,a=e.slice(2);this.inputSpec[0]=new Xt({shape:[n,null,...a]});let r=[e[0]].concat(e.slice(2));if(t!=null)throw new $e("Constants support is not implemented in RNN yet.");this.cell.build(r);let s;if(Array.isArray(this.cell.stateSize)?s=this.cell.stateSize:s=[this.cell.stateSize],this.stateSpec!=null){if(!k.arraysEqual(this.stateSpec.map(i=>i.shape[i.shape.length-1]),s))throw new z(`An initialState was passed that is not compatible with cell.stateSize. Received stateSpec=${this.stateSpec}; However cell.stateSize is ${this.cell.stateSize}`)}else this.stateSpec=s.map(i=>new Xt({shape:[null,i]}));this.stateful&&this.resetStates()}resetStates(e,t=!1){D(()=>{if(!this.stateful)throw new xr("Cannot call resetStates() on an RNN Layer that is not stateful.");let n=this.inputSpec[0].shape[0];if(n==null)throw new z("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(a=>vt([n,a])):this.states_=[vt([n,this.cell.stateSize])];else if(e==null)Fe(this.states_),this.keptStates!=null&&(Fe(this.keptStates),this.keptStates=[]),Array.isArray(this.cell.stateSize)?this.states_=this.cell.stateSize.map(a=>vt([n,a])):this.states_[0]=vt([n,this.cell.stateSize]);else{if(Array.isArray(e)||(e=[e]),e.length!==this.states_.length)throw new z(`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()):Fe(this.states_);for(let a=0;a<this.states_.length;++a){let r=e[a],s=Array.isArray(this.cell.stateSize)?this.cell.stateSize[a]:this.cell.stateSize,i=[n,s];if(!k.arraysEqual(r.shape,i))throw new z(`State ${a} is incompatible with layer ${this.name}: expected shape=${i}, received shape=${r.shape}`);this.states_[a]=r}}this.states_=this.states_.map(a=>jt(a.clone()))})}apply(e,t){let n=t==null?null:t.initialState,a=t==null?null:t.constants;t==null&&(t={});let r=TI(e,n,a,this.numConstants);e=r.inputs,n=r.initialState,a=r.constants;let s=[],i=[];if(n!=null){t.initialState=n,s=s.concat(n),this.stateSpec=[];for(let o of n)this.stateSpec.push(new Xt({shape:o.shape}));i=i.concat(this.stateSpec)}if(a!=null&&(t.constants=a,s=s.concat(a),this.numConstants=a.length),s[0]instanceof Aa){let o=[e].concat(s),l=this.inputSpec.concat(i),c=this.inputSpec;this.inputSpec=l;let u=super.apply(o,t);return this.inputSpec=c,u}else return super.apply(e,t)}call(e,t){return D(()=>{let n=t==null?null:t.mask,a=t==null?null:t.training,r=t==null?null:t.initialState;e=Me(e),r==null&&(this.stateful?r=this.states_:r=this.getInitialState(e));let s=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1;if(r.length!==s)throw new z(`RNN Layer has ${s} 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 i={training:a},o=SI((d,h)=>{let m=this.cell.call([d].concat(h),i);return[m[0],m.slice(1)]},e,r,this.goBackwards,n,null,this.unroll,this.returnSequences),l=o[0],c=o[1],u=o[2];this.stateful&&this.resetStates(u,a);let p=this.returnSequences?c:l;return this.returnState?[p].concat(u):p})}getInitialState(e){return D(()=>{let t=vt(e.shape);return t=Te(t,[1,2]),t=Qc(t),Array.isArray(this.cell.stateSize)?this.cell.stateSize.map(n=>n>1?wb(t,[1,n]):t):this.cell.stateSize>1?[wb(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()===tr.className&&(t.cell={className:this.cell.getClassName(),config:n}),Object.assign({},n,e,t)}static fromConfig(e,t,n={}){let a=t.cell,r=Fa(a,n);return new e(Object.assign(t,{cell:r}))}};tr.className="RNN";re.registerClass(tr);var np=class extends qe{},xm=class extends np{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,Kt(this.units,"units"),this.activation=is(e.activation==null?this.DEFAULT_ACTIVATION:e.activation),this.useBias=e.useBias==null?!0:e.useBias,this.kernelInitializer=wt(e.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.recurrentInitializer=wt(e.recurrentInitializer||this.DEFAULT_RECURRENT_INITIALIZER),this.biasInitializer=wt(e.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.kernelRegularizer=kt(e.kernelRegularizer),this.recurrentRegularizer=kt(e.recurrentRegularizer),this.biasRegularizer=kt(e.biasRegularizer),this.kernelConstraint=Vt(e.kernelConstraint),this.recurrentConstraint=Vt(e.recurrentConstraint),this.biasConstraint=Vt(e.biasConstraint),this.dropout=ql([1,as([0,e.dropout==null?0:e.dropout])]),this.recurrentDropout=ql([1,as([0,e.recurrentDropout==null?0:e.recurrentDropout])]),this.stateSize=this.units,this.dropoutMask=null,this.recurrentDropoutMask=null}build(e){e=ht(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 D(()=>{if(e=e,e.length!==2)throw new z(`SimpleRNNCell expects 2 input Tensors, got ${e.length}.`);let n=e[1];e=e[0];let a=t.training==null?!1:t.training;0<this.dropout&&this.dropout<1&&this.dropoutMask==null&&(this.dropoutMask=os({ones:()=>Dn(e),rate:this.dropout,training:a})),0<this.recurrentDropout&&this.recurrentDropout<1&&this.recurrentDropoutMask==null&&(this.recurrentDropoutMask=os({ones:()=>Dn(n),rate:this.recurrentDropout,training:a}));let r,s=this.dropoutMask,i=this.recurrentDropoutMask;s!=null?r=Qa(L(e,s),this.kernel.read()):r=Qa(e,this.kernel.read()),this.bias!=null&&(r=er(r,this.bias.read())),i!=null&&(n=L(n,i));let o=Z(r,Qa(n,this.recurrentKernel.read()));return this.activation!=null&&(o=this.activation.apply(o)),[o,o]})}getConfig(){let e=super.getConfig(),t={units:this.units,activation:ss(this.activation),useBias:this.useBias,kernelInitializer:Ct(this.kernelInitializer),recurrentInitializer:Ct(this.recurrentInitializer),biasInitializer:Ct(this.biasInitializer),kernelRegularizer:mt(this.kernelRegularizer),recurrentRegularizer:mt(this.recurrentRegularizer),biasRegularizer:mt(this.biasRegularizer),activityRegularizer:mt(this.activityRegularizer),kernelConstraint:Wt(this.kernelConstraint),recurrentConstraint:Wt(this.recurrentConstraint),biasConstraint:Wt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout};return Object.assign({},e,t)}};xm.className="SimpleRNNCell";re.registerClass(xm);var dx=class extends tr{constructor(e){e.cell=new xm(e),super(e)}call(e,t){return D(()=>{this.cell.dropoutMask!=null&&(Fe(this.cell.dropoutMask),this.cell.dropoutMask=null),this.cell.recurrentDropoutMask!=null&&(Fe(this.cell.recurrentDropoutMask),this.cell.recurrentDropoutMask=null);let n=t==null?null:t.mask,a=t==null?null:t.training,r=t==null?null:t.initialState;return super.call(e,{mask:n,training:a,initialState:r})})}static fromConfig(e,t){return new e(t)}};dx.className="SimpleRNN";re.registerClass(dx);var vm=class extends np{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 z("GRUCell does not support reset_after parameter set to true.");this.units=e.units,Kt(this.units,"units"),this.activation=is(e.activation===void 0?this.DEFAULT_ACTIVATION:e.activation),this.recurrentActivation=is(e.recurrentActivation===void 0?this.DEFAULT_RECURRENT_ACTIVATION:e.recurrentActivation),this.useBias=e.useBias==null?!0:e.useBias,this.kernelInitializer=wt(e.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.recurrentInitializer=wt(e.recurrentInitializer||this.DEFAULT_RECURRENT_INITIALIZER),this.biasInitializer=wt(e.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.kernelRegularizer=kt(e.kernelRegularizer),this.recurrentRegularizer=kt(e.recurrentRegularizer),this.biasRegularizer=kt(e.biasRegularizer),this.kernelConstraint=Vt(e.kernelConstraint),this.recurrentConstraint=Vt(e.recurrentConstraint),this.biasConstraint=Vt(e.biasConstraint),this.dropout=ql([1,as([0,e.dropout==null?0:e.dropout])]),this.recurrentDropout=ql([1,as([0,e.recurrentDropout==null?0:e.recurrentDropout])]),this.implementation=e.implementation,this.stateSize=this.units,this.dropoutMask=null,this.recurrentDropoutMask=null}build(e){e=ht(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 D(()=>{if(e=e,e.length!==2)throw new z(`GRUCell expects 2 input Tensors (inputs, h, c), got ${e.length}.`);let n=t.training==null?!1:t.training,a=e[1];e=e[0],0<this.dropout&&this.dropout<1&&this.dropoutMask==null&&(this.dropoutMask=os({ones:()=>Dn(e),rate:this.dropout,training:n,count:3})),0<this.recurrentDropout&&this.recurrentDropout<1&&this.recurrentDropoutMask==null&&(this.recurrentDropoutMask=os({ones:()=>Dn(a),rate:this.recurrentDropout,training:n,count:3}));let r=this.dropoutMask,s=this.recurrentDropoutMask,i,o,l;0<this.dropout&&this.dropout<1&&(e=L(e,r[0]));let c=Qa(e,this.kernel.read());this.useBias&&(c=er(c,this.bias.read())),0<this.recurrentDropout&&this.recurrentDropout<1&&(a=L(a,s[0]));let u=this.recurrentKernel.read(),[p,d]=Mn(u,[2*this.units,this.units],u.rank-1),h=Qa(a,p),[m,f,g]=Mn(c,3,c.rank-1),[y,b]=Mn(h,2,h.rank-1);i=this.recurrentActivation.apply(Z(m,y)),o=this.recurrentActivation.apply(Z(f,b));let v=Qa(L(o,a),d);l=this.activation.apply(Z(g,v));let x=Z(L(i,a),L(Z(1,Tt(i)),l));return[x,x]})}getConfig(){let e=super.getConfig(),t={units:this.units,activation:ss(this.activation),recurrentActivation:ss(this.recurrentActivation),useBias:this.useBias,kernelInitializer:Ct(this.kernelInitializer),recurrentInitializer:Ct(this.recurrentInitializer),biasInitializer:Ct(this.biasInitializer),kernelRegularizer:mt(this.kernelRegularizer),recurrentRegularizer:mt(this.recurrentRegularizer),biasRegularizer:mt(this.biasRegularizer),activityRegularizer:mt(this.activityRegularizer),kernelConstraint:Wt(this.kernelConstraint),recurrentConstraint:Wt(this.recurrentConstraint),biasConstraint:Wt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation,resetAfter:!1};return Object.assign({},e,t)}};vm.className="GRUCell";re.registerClass(vm);var hx=class extends tr{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 vm(e),super(e)}call(e,t){return D(()=>{this.cell.dropoutMask!=null&&(Fe(this.cell.dropoutMask),this.cell.dropoutMask=null),this.cell.recurrentDropoutMask!=null&&(Fe(this.cell.recurrentDropoutMask),this.cell.recurrentDropoutMask=null);let n=t==null?null:t.mask,a=t==null?null:t.training,r=t==null?null:t.initialState;return super.call(e,{mask:n,training:a,initialState:r})})}static fromConfig(e,t){return t.implmentation===0&&(t.implementation=1),new e(t)}};hx.className="GRU";re.registerClass(hx);var cp=class extends np{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,Kt(this.units,"units"),this.activation=is(e.activation===void 0?this.DEFAULT_ACTIVATION:e.activation),this.recurrentActivation=is(e.recurrentActivation===void 0?this.DEFAULT_RECURRENT_ACTIVATION:e.recurrentActivation),this.useBias=e.useBias==null?!0:e.useBias,this.kernelInitializer=wt(e.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.recurrentInitializer=wt(e.recurrentInitializer||this.DEFAULT_RECURRENT_INITIALIZER),this.biasInitializer=wt(e.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.unitForgetBias=e.unitForgetBias,this.kernelRegularizer=kt(e.kernelRegularizer),this.recurrentRegularizer=kt(e.recurrentRegularizer),this.biasRegularizer=kt(e.biasRegularizer),this.kernelConstraint=Vt(e.kernelConstraint),this.recurrentConstraint=Vt(e.recurrentConstraint),this.biasConstraint=Vt(e.biasConstraint),this.dropout=ql([1,as([0,e.dropout==null?0:e.dropout])]),this.recurrentDropout=ql([1,as([0,e.recurrentDropout==null?0:e.recurrentDropout])]),this.implementation=e.implementation,this.stateSize=[this.units,this.units],this.dropoutMask=null,this.recurrentDropoutMask=null}build(e){var t;e=ht(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 a;if(this.useBias){if(this.unitForgetBias){let r=this.biasInitializer,s=this.units;a=new(t=class extends ha{apply(i,o){let l=r.apply([s]),c=new Kh().apply([s]),u=r.apply([s*2]);return bk(bk(l,c),u)}},t.className="CustomInit",t)}else a=this.biasInitializer;this.bias=this.addWeight("bias",[this.units*4],null,a,this.biasRegularizer,!0,this.biasConstraint)}else this.bias=null;this.built=!0}call(e,t){return D(()=>{let n=t.training==null?!1:t.training;if(e=e,e.length!==3)throw new z(`LSTMCell expects 3 input Tensors (inputs, h, c), got ${e.length}.`);let a=e[1],r=e[2];e=e[0],0<this.dropout&&this.dropout<1&&this.dropoutMask==null&&(this.dropoutMask=os({ones:()=>Dn(e),rate:this.dropout,training:n,count:4})),0<this.recurrentDropout&&this.recurrentDropout<1&&this.recurrentDropoutMask==null&&(this.recurrentDropoutMask=os({ones:()=>Dn(a),rate:this.recurrentDropout,training:n,count:4}));let s=this.dropoutMask,i=this.recurrentDropoutMask,o,l,c,u;0<this.dropout&&this.dropout<1&&(e=L(e,s[0]));let p=Qa(e,this.kernel.read());0<this.recurrentDropout&&this.recurrentDropout<1&&(a=L(a,i[0])),p=Z(p,Qa(a,this.recurrentKernel.read())),this.useBias&&(p=er(p,this.bias.read()));let[d,h,m,f]=Mn(p,4,p.rank-1);o=this.recurrentActivation.apply(d),l=this.recurrentActivation.apply(h),c=Z(L(l,r),L(o,this.activation.apply(m))),u=this.recurrentActivation.apply(f);let g=L(u,this.activation.apply(c));return[g,g,c]})}getConfig(){let e=super.getConfig(),t={units:this.units,activation:ss(this.activation),recurrentActivation:ss(this.recurrentActivation),useBias:this.useBias,kernelInitializer:Ct(this.kernelInitializer),recurrentInitializer:Ct(this.recurrentInitializer),biasInitializer:Ct(this.biasInitializer),unitForgetBias:this.unitForgetBias,kernelRegularizer:mt(this.kernelRegularizer),recurrentRegularizer:mt(this.recurrentRegularizer),biasRegularizer:mt(this.biasRegularizer),activityRegularizer:mt(this.activityRegularizer),kernelConstraint:Wt(this.kernelConstraint),recurrentConstraint:Wt(this.recurrentConstraint),biasConstraint:Wt(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation};return Object.assign({},e,t)}};cp.className="LSTMCell";re.registerClass(cp);var mx=class extends tr{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 cp(e),super(e)}call(e,t){return D(()=>{this.cell.dropoutMask!=null&&(Fe(this.cell.dropoutMask),this.cell.dropoutMask=null),this.cell.recurrentDropoutMask!=null&&(Fe(this.cell.recurrentDropoutMask),this.cell.recurrentDropoutMask=null);let n=t==null?null:t.mask,a=t==null?null:t.training,r=t==null?null:t.initialState;return super.call(e,{mask:n,training:a,initialState:r})})}static fromConfig(e,t){return t.implmentation===0&&(t.implementation=1),new e(t)}};mx.className="LSTM";re.registerClass(mx);var bm=class extends np{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 D(()=>{e=e;let n=e.slice(1),a=[];for(let i of this.cells.slice().reverse())Array.isArray(i.stateSize)?a.push(n.splice(0,i.stateSize.length)):a.push(n.splice(0,1));a.reverse();let r=[],s;for(let i=0;i<this.cells.length;++i){let o=this.cells[i];n=a[i],i===0?s=[e[0]].concat(n):s=[s[0]].concat(n),s=o.call(s,t),r.push(s.slice(1))}n=[];for(let i of r.slice().reverse())n.push(...i);return[s[0]].concat(n)})}build(e){Mb(e)&&(e=e[0]),e=e;let t;this.cells.forEach((n,a)=>{Mi(`RNNCell_${a}`,()=>{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=a=>({className:a.getClassName(),config:a.getConfig()}),n={cells:this.cells.map(t)};return Object.assign({},e,n)}static fromConfig(e,t,n={}){let a=[];for(let r of t.cells)a.push(Fa(r,n));return new e({cells:a})}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 Pb(e)}setWeights(e){let t=[];for(let n of this.cells){let a=n.weights.length,r=e.splice(a);for(let s=0;s<n.weights.length;++s)t.push([n.weights[s],r[s]])}Ob(t)}};bm.className="StackedRNNCells";re.registerClass(bm);function os(e){let{ones:t,rate:n,training:a=!1,count:r=1}=e,s=()=>vk(t(),n),i=()=>tp(s,t,a);return!r||r<=1?jt(i().clone()):Array(r).fill(void 0).map(i).map(o=>jt(o.clone()))}var MW=function(e,t){var n={};for(var a in e)Object.prototype.hasOwnProperty.call(e,a)&&t.indexOf(a)<0&&(n[a]=e[a]);if(e!=null&&typeof Object.getOwnPropertySymbols=="function")for(var r=0,a=Object.getOwnPropertySymbols(e);r<a.length;r++)t.indexOf(a[r])<0&&Object.prototype.propertyIsEnumerable.call(e,a[r])&&(n[a[r]]=e[a[r]]);return n},CI=class extends tr{constructor(e){if(e.unroll)throw new $e("Unrolling is not possible with convolutional RNNs.");if(Array.isArray(e.cell))throw new $e("It is not possible at the moment to stack convolutional cells.");super(e);this.inputSpec=[new Xt({ndim:5})]}call(e,t){return D(()=>{if(this.cell.dropoutMask!=null&&(Fe(this.cell.dropoutMask),this.cell.dropoutMask=null),this.cell.recurrentDropoutMask!=null&&(Fe(this.cell.recurrentDropoutMask),this.cell.recurrentDropoutMask=null),t&&t.constants)throw new z("ConvRNN2D cell does not support constants");let n=t==null?null:t.mask,a=t==null?null:t.training,r=t==null?null:t.initialState;return super.call(e,{mask:n,training:a,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 D(()=>{let{stateSize:t}=this.cell,n=e.shape,a=this.computeSingleOutputShape(n),r=[a[0],...a.slice(2)],s=vt(r);return Array.isArray(t)?Array(t.length).fill(s):[s]})}resetStates(e,t=!1){D(()=>{if(!this.stateful)throw new xr("Cannot call resetStates() on an RNN Layer that is not stateful.");let n=this.inputSpec[0].shape,a=this.computeSingleOutputShape(n),r=[a[0],...a.slice(2)];if(n[0]==null)throw new z("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(()=>vt(r)):this.states_=[vt(r)];else if(e==null)Fe(this.states_),this.keptStates!=null&&(Fe(this.keptStates),this.keptStates=[]),Array.isArray(this.cell.stateSize)?this.states_=this.cell.stateSize.map(()=>vt(r)):this.states_[0]=vt(r);else{if(Array.isArray(e)||(e=[e]),e.length!==this.states_.length)throw new z(`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()):Fe(this.states_);for(let s=0;s<this.states_.length;++s){let i=e[s],o=r;if(!k.arraysEqual(i.shape,o))throw new z(`State ${s} is incompatible with layer ${this.name}: expected shape=${o}, received shape=${i.shape}`);this.states_[s]=i}}this.states_=this.states_.map(s=>jt(s.clone()))})}computeSingleOutputShape(e){let{dataFormat:t,filters:n,kernelSize:a,padding:r,strides:s,dilationRate:i}=this.cell,o=t==="channelsFirst",l=e[o?3:2],c=e[o?4:3],u=$a(l,a[0],r,s[0],i[0]),p=$a(c,a[1],r,s[1],i[1]);return[...e.slice(0,2),...o?[n,u,p]:[u,p,n]]}};CI.className="ConvRNN2D";var wm=class extends cp{constructor(e){let{filters:t,kernelSize:n,strides:a,padding:r,dataFormat:s,dilationRate:i}=e;super(Object.assign({},e,{units:t}));this.filters=t,Kt(this.filters,"filters"),this.kernelSize=Jl(n,2,"kernelSize"),this.kernelSize.forEach(o=>Kt(o,"kernelSize")),this.strides=Jl(a||1,2,"strides"),this.strides.forEach(o=>Kt(o,"strides")),this.padding=r||"valid",Jn(this.padding),this.dataFormat=s||"channelsLast",Dt(this.dataFormat),this.dilationRate=Jl(i||1,2,"dilationRate"),this.dilationRate.forEach(o=>Kt(o,"dilationRate"))}build(e){var t;e=ht(e);let n=this.dataFormat==="channelsFirst"?1:e.length-1;if(e[n]==null)throw new z(`The channel dimension of the input should be defined. Found ${e[n]}`);let a=e[n],r=4,s=this.kernelSize.concat([a,this.filters*r]);this.kernel=this.addWeight("kernel",s,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint);let i=this.kernelSize.concat([this.filters,this.filters*r]);if(this.recurrentKernel=this.addWeight("recurrent_kernel",i,null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias){let o;if(this.unitForgetBias){let l=this.biasInitializer,c=this.filters;o=new(t=class extends ha{apply(u,p){let d=l.apply([c]),h=Xa([c]),m=l.apply([c*2]);return Ib([d,h,m])}},t.className="CustomInit",t)}else o=this.biasInitializer;this.bias=this.addWeight("bias",[this.filters*r],null,o,this.biasRegularizer,!0,this.biasConstraint)}this.built=!0}call(e,t){return D(()=>{if(e.length!==3)throw new z(`ConvLSTM2DCell expects 3 input Tensors (inputs, h, c), got ${e.length}.`);let n=t.training||!1,a=e[0],r=e[1],s=e[2],i=4;0<this.dropout&&this.dropout<1&&this.dropoutMask==null&&(this.dropoutMask=os({ones:()=>Dn(a),rate:this.dropout,training:n,count:i}));let o=this.dropoutMask,l=(J,se,Q)=>!se||!se[Q]?J:L(se[Q],J),c=l(a,o,0),u=l(a,o,1),p=l(a,o,2),d=l(a,o,3);0<this.recurrentDropout&&this.recurrentDropout<1&&this.recurrentDropoutMask==null&&(this.recurrentDropoutMask=os({ones:()=>Dn(r),rate:this.recurrentDropout,training:n,count:i}));let h=this.recurrentDropoutMask,m=l(r,h,0),f=l(r,h,1),g=l(r,h,2),y=l(r,h,3),b=3,[v,x,N,T]=Mn(this.kernel.read(),i,b),[C,$,F,O]=this.useBias?Mn(this.bias.read(),i):[null,null,null,null];c=this.inputConv(c,v,C,this.padding),u=this.inputConv(u,x,$,this.padding),p=this.inputConv(p,N,F,this.padding),d=this.inputConv(d,T,O,this.padding);let[B,W,G,q]=Mn(this.recurrentKernel.read(),i,b);m=this.recurrentConv(m,B),f=this.recurrentConv(f,W),g=this.recurrentConv(g,G),y=this.recurrentConv(y,q);let H=this.recurrentActivation.apply(Z(c,m)),K=this.recurrentActivation.apply(Z(u,f)),X=Z(L(K,s),L(H,this.activation.apply(Z(p,g)))),ne=L(this.recurrentActivation.apply(Z(d,y)),this.activation.apply(X));return[ne,ne,X]})}getConfig(){let e=super.getConfig(),{units:t}=e,n=MW(e,["units"]),a={filters:this.filters,kernelSize:this.kernelSize,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,strides:this.strides};return Object.assign({},n,a)}inputConv(e,t,n,a){let r=At(e,t,this.strides,a||"valid",this.dataFormat==="channelsFirst"?"NCHW":"NHWC",this.dilationRate);return n?er(r,n,this.dataFormat):r}recurrentConv(e,t){return At(e,t,1,"same",this.dataFormat==="channelsFirst"?"NCHW":"NHWC")}};wm.className="ConvLSTM2DCell";re.registerClass(wm);var fx=class extends CI{constructor(e){let t=new wm(e);super(Object.assign({},e,{cell:t}))}static fromConfig(e,t){return new e(t)}};fx.className="ConvLSTM2D";re.registerClass(fx);var km=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 a=0;a<this.noiseShape.length;++a)n.push(this.noiseShape[a]==null?t[a]:this.noiseShape[a]);return n}call(e,t){return D(()=>{this.invokeCallHook(e,t);let n=Me(e);if(0<this.rate&&this.rate<1){let a=t.training==null?!1:t.training,r=this.getNoiseShape(n);return tp(()=>vk(n,this.rate,r,this.seed),()=>n,a)}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()}};km.className="Dropout";re.registerClass(km);var gx=class extends km{constructor(e){super(e);this.inputSpec=[{ndim:3}]}getNoiseShape(e){let t=e.shape;return[t[0],1,t[2]]}};gx.className="SpatialDropout1D";re.registerClass(gx);var yx=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,Kt(this.units,"units"),this.activation=is(e.activation),e.useBias!=null&&(this.useBias=e.useBias),this.kernelInitializer=wt(e.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.biasInitializer=wt(e.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.kernelConstraint=Vt(e.kernelConstraint),this.biasConstraint=Vt(e.biasConstraint),this.kernelRegularizer=kt(e.kernelRegularizer),this.biasRegularizer=kt(e.biasRegularizer),this.activityRegularizer=kt(e.activityRegularizer),this.supportsMasking=!0,this.inputSpec=[{minNDim:2}]}build(e){e=ht(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=ht(e);let t=e.slice();return t[t.length-1]=this.units,t}call(e,t){return D(()=>{this.invokeCallHook(e,t);let n=Me(e),a=lk(this.activation.getClassName()),r;return a!=null?r=Qa(n,this.kernel.read(),a,this.bias?this.bias.read():null):(r=Qa(n,this.kernel.read()),this.bias!=null&&(r=er(r,this.bias.read())),this.activation!=null&&(r=this.activation.apply(r))),r})}getConfig(){let e={units:this.units,activation:ss(this.activation),useBias:this.useBias,kernelInitializer:Ct(this.kernelInitializer),biasInitializer:Ct(this.biasInitializer),kernelRegularizer:mt(this.kernelRegularizer),biasRegularizer:mt(this.biasRegularizer),activityRegularizer:mt(this.activityRegularizer),kernelConstraint:Wt(this.kernelConstraint),biasConstraint:Wt(this.biasConstraint)},t=super.getConfig();return Object.assign(e,t),e}};yx.className="Dense";re.registerClass(yx);var bx=class extends qe{constructor(e){e=e||{},super(e),this.inputSpec=[{minNDim:3}],this.dataFormat=e.dataFormat}computeOutputShape(e){e=ht(e);for(let t of e.slice(1))if(t==null)throw new z(`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],ns(e,1)]}call(e,t){return D(()=>{this.invokeCallHook(e,t);let n=Me(e);if(this.dataFormat==="channelsFirst"&&n.rank>1){let a=[0];for(let r=2;r<n.rank;++r)a.push(r);a.push(1),n=n.transpose(a)}return iz(n)})}getConfig(){let e={};this.dataFormat!=null&&(e.dataFormat=this.dataFormat);let t=super.getConfig();return Object.assign(e,t),e}};bx.className="Flatten";re.registerClass(bx);var xx=class extends qe{constructor(e){super(e);this.supportsMasking=!0,this.activation=is(e.activation)}call(e,t){return D(()=>{this.invokeCallHook(e,t);let n=Me(e);return this.activation.apply(n)})}getConfig(){let e={activation:ss(this.activation)},t=super.getConfig();return Object.assign(e,t),e}};xx.className="Activation";re.registerClass(xx);var vx=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 D(()=>(e=Me(e),rz(e,this.n)))}getConfig(){let e={n:this.n},t=super.getConfig();return Object.assign(e,t),e}};vx.className="RepeatVector";re.registerClass(vx);var wx=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.",a=t.slice(),r=1,s=null;for(let o=0;o<a.length;++o){let l=a[o];if(this.isUnknown(l))if(s===null)s=o;else throw new z("Can only specifiy one unknown dimension.");else r*=l}let i=ns(e);if(s!==null){if(r===0||i%r!=0)throw new z(n);a[s]=i/r}else if(i!==r)throw new z(n);return a}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 D(()=>{this.invokeCallHook(e,t);let n=Me(e),a=n.shape,r=a.slice(0,1).concat(this.fixUnknownDimension(a.slice(1),this.targetShape));return n.reshape(r)})}getConfig(){let e={targetShape:this.targetShape},t=super.getConfig();return Object.assign(e,t),e}};wx.className="Reshape";re.registerClass(wx);var kx=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=Ea(1,e.dims.length+1);if(!k.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 Xt({ndim:this.dims.length+1})]}computeOutputShape(e){e=ht(e);let t=e.slice();return this.dims.forEach((n,a)=>{t[a+1]=e[n]}),t}call(e,t){return Ue(Me(e),this.dimsIncludingBatch)}getConfig(){let e={dims:this.dims},t=super.getConfig();return Object.assign(e,t),e}};kx.className="Permute";re.registerClass(kx);var Ix=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=Me(e),a=-1;return Rc(_i(n,this.maskValue),a)}call(e,t){return D(()=>{this.invokeCallHook(e,t);let n=Me(e),a=-1,r=!0,s=Rc(_i(n,this.maskValue),a,r);return n.mul(s.asType(n.dtype))})}};Ix.className="Masking";re.registerClass(Ix);var Nx=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(yt(e.inputLength))}this.inputDim=e.inputDim,Kt(this.inputDim,"inputDim"),this.outputDim=e.outputDim,Kt(this.outputDim,"outputDim"),this.embeddingsInitializer=wt(e.embeddingsInitializer||this.DEFAULT_EMBEDDINGS_INITIALIZER),this.embeddingsRegularizer=kt(e.embeddingsRegularizer),this.activityRegularizer=kt(e.activityRegularizer),this.embeddingsConstraint=Vt(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 D(()=>this.maskZero?(e=Me(e),_i(e,Ge(e))):null)}computeOutputShape(e){if(e=ht(e),this.inputLength==null)return[...e,this.outputDim];let t=yt(this.inputLength);if(t.length!==e.length-1)throw new z(`"inputLength" is ${this.inputLength}, but received input shape has shape ${e}`);{let n=0;for(let a=0;a<t.length;++a){let r=t[a],s=e[a+1];if(r!=null&&s!=null&&r!==s)throw new z(`"inputLength" is ${this.inputLength}, but received input shape has shape ${e}`);r==null&&(t[n]=s),n++}}return[e[0],...t,this.outputDim]}call(e,t){return D(()=>{this.invokeCallHook(e,t);let n=Me(e);return n.dtype!=="int32"&&(n=Jc(n,"int32")),xk(this.embeddings.read(),n.as1D()).reshape(ht(this.computeOutputShape(n.shape)))})}getConfig(){let e={inputDim:this.inputDim,outputDim:this.outputDim,embeddingsInitializer:Ct(this.embeddingsInitializer),embeddingsRegularizer:mt(this.embeddingsRegularizer),activityRegularizer:mt(this.activityRegularizer),embeddingsConstraint:Wt(this.embeddingsConstraint),maskZero:this.maskZero,inputLength:this.inputLength},t=super.getConfig();return Object.assign(e,t),e}};Nx.className="Embedding";re.registerClass(Nx);var Bi=class extends qe{constructor(e){super(e||{});this.supportsMasking=!0}mergeFunction(e){throw new $e}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 a=0;a<t.length;++a){let r=e[e.length-t.length+a],s=t[a];if(r==null||s==null||r<0||s<0)n.push(null);else if(r===1)n.push(s);else if(s===1)n.push(r);else{if(r!==s)throw new z("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=[ht(e)]),e=e,e.length<2)throw new z(`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=ts(t),t.length>1)throw new z(`Can not merge tensors with different batch sizes. Got tensors with shapes: ${JSON.stringify(e)}.`);let n=e[0]==null?null:e[0].slice(1);for(let r=1;r<e.length;++r){let s=e[r]==null?null:e[r].slice(1);n=this.computeElementwiseOpOutputShape(n,s)}let a=e.map(r=>r.length);e.indexOf(null)===-1&&ts(a).length===1?this.reshapeRequired=!1:this.reshapeRequired=!0}call(e,t){return D(()=>{if(e=e,this.reshapeRequired){let n=[],a=e.map(r=>r.rank);if(a.indexOf(null)===-1){let r=as(a);for(let s of e){let i=s.rank;for(let o=0;o<r-i;++o)s=Qc(s,1);n.push(s)}return this.mergeFunction(n)}else{let r=!1;for(let o of e){let l=o.rank;if(l==null){let c=o.shape,u=c[0],p=c.slice(1).concat([u]),d=o.reshape([u].concat(ns(c.slice(1))));d=Ue(d,[1,0]),d=d.reshape(p),n.push(d),r=!0}else if(l>1){let c=Ea(1,l).concat([0]);n.push(Ue(o,c)),r=!0}else n.push(o)}let s=this.mergeFunction(n),i=s.rank;if(r){if(i==null){let o=s.shape,l=o.length,c=o[l-1],u=[c].concat(o.slice(0,o.length-1));s=Ue(s.reshape([-1,c]),[1,0]).reshape(u)}else if(i>1){let o=[i-1].concat(Ea(0,i-1));s=Ue(s,o)}}return s}}else return this.mergeFunction(e)})}computeOutputShape(e){e=e;let t;e[0]==null?t=null:t=e[0].slice(1);for(let a=1;a<e.length;++a){let r=e[a]==null?null:e[a].slice(1);t=this.computeElementwiseOpOutputShape(t,r)}let n=[];for(let a of e)a!=null&&a[0]!==null&&n.push(a[0]);return n=ts(n),n.length===1?t=n.concat(t):t=[null].concat(t),t}computeMask(e,t){return D(()=>{if(t==null)return null;if(!Array.isArray(t))throw new z("`mask` should be an Array");if(!Array.isArray(e))throw new z("`inputs` should be an Array");if(t.length!==e.length)throw new z(`The Array 'inputs' and 'mask' are expected to have the same length, but have different lengths (${e.length} vs ${t.length})`);if(t.every(a=>a==null))return null;t=t.map(a=>a==null?a:Fn(a,0));let n=t[0];for(let a=1;a<t.length-1;++a)n=pa(n,t[a]);return n})}},Tx=class extends Bi{constructor(e){super(e)}mergeFunction(e){return D(()=>{let t=e[0].clone();for(let n=1;n<e.length;++n)t=Z(t,e[n]);return t})}};Tx.className="Add";re.registerClass(Tx);var Sx=class extends Bi{constructor(e){super(e)}mergeFunction(e){return D(()=>{let t=e[0].clone();for(let n=1;n<e.length;++n)t=L(t,e[n]);return t})}};Sx.className="Multiply";re.registerClass(Sx);var Cx=class extends Bi{constructor(e){super(e)}mergeFunction(e){return D(()=>{let t=e[0].clone();for(let n=1;n<e.length;++n)t=Z(t,e[n]);return L(1/e.length,t)})}};Cx.className="Average";re.registerClass(Cx);var _x=class extends Bi{constructor(e){super(e)}mergeFunction(e){return D(()=>{let t=e[0];for(let n=1;n<e.length;++n)t=Ka(t,e[n]);return t})}};_x.className="Maximum";re.registerClass(_x);var Ex=class extends Bi{constructor(e){super(e)}mergeFunction(e){return D(()=>{let t=e[0];for(let n=1;n<e.length;++n)t=Bl(t,e[n]);return t})}};Ex.className="Minimum";re.registerClass(Ex);var Ax=class extends Bi{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 z("A `Concatenate` layer should be called on a list of at least 2 inputs");e=e;let t=!0;for(let a of e)if(a!=null){t=!1;break}if(t)return;let n=[];for(let a=0;a<e.length;++a){let r=e[a].slice();r.splice(this.axis,1);let s=!1;for(let i of n)if(k.arraysEqual(i,r)){s=!0;break}s||n.push(r)}if(n.length>1)throw new z("A `Concatenate` layer requires inputs with matching shapes except for the concat axis. Got input shapes: "+JSON.stringify(e))}mergeFunction(e){return D(()=>Ib(e,this.axis))}computeOutputShape(e){if(!(Array.isArray(e)&&Array.isArray(e[0])))throw new z("A `Concatenate` layer should be called on a list of inputs.");let t=e,n=t[0].slice(),a=this.axis<0?n.length+this.axis:this.axis;for(let r of t.slice(1)){if(n[a]==null||r[a]==null){n[a]=null;break}n[a]+=r[a]}return n}computeMask(e,t){if(t==null)return null;if(!Array.isArray(t))throw new z("`mask` should be an array for Concatenate");if(!Array.isArray(e))throw new z("`inputs` should be an array for Concatenate");if(t.length!==e.length)throw new z(`Mismatch in the length of mask (${t.length}) and the legnth of inputs (${e.length})`);return D(()=>{let n=!0;if(t.forEach(s=>{if(s!=null){n=!1;return}}),n)return null;let a=[];for(let s=0;s<e.length;++s)t[s]==null?a.push(Dn(e[s]).asType("bool")):t[s].rank<e[s].rank?a.push(Fn(t[s],-1)):a.push(t[s]);let r=Qe(a,this.axis);return lh(r,-1,!1)})}getConfig(){let e={axis:this.axis},t=super.getConfig();return Object.assign(e,t),e}};Ax.className="Concatenate";re.registerClass(Ax);function pp(e,t){for(;e<0;)e+=t;return e}function PW(e,t,n){if(e.shape.length>3||t.shape.length>3)throw new $e("batchDot is not implemented for tensors of 4D or higher rank yet");if(k.assert(e.shape.length>=2,()=>`batchDot requires the rank of x to be >= 2, but got ${e.shape.length}`),k.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 $e("batchDot is not implemented for complex64-type Tensors yet.");let a=e.shape.length,r=t.shape.length;n==null&&(n=[a-1,r-2]);let s=n;return D(()=>{let i;if(a>r){i=a-r;let l=[];for(let c=0;c<i;++c)l.push(1);t=t.reshape(t.shape.concat(l))}else if(r>a){i=r-a;let l=[];for(let c=0;c<i;++c)l.push(1);e=e.reshape(e.shape.concat(l))}else i=0;let o;if(e.shape.length===2&&t.shape.length===2)s[0]===s[1]?o=e.mul(t).sum(s[0]):o=e.transpose([1,0]).mul(t).sum(s[1]);else{let l=s[0]!==e.shape.length-1,c=s[1]===t.shape.length-1;o=e.matMul(t,l,c)}if(i>0){let l;a>r?l=a+r-3:l=a-1;let c=[];for(let u=l;u<l+i;++u)c.push(u);o=o.squeeze(c)}return o.shape.length===1&&(o=o.expandDims(1)),o})}var Fx=class extends Bi{constructor(e){super(e);this.axes=e.axes,this.normalize=e.normalize==null?!1:e.normalize,this.supportsMasking=!0,this.reshapeRequired=!1}build(e){k.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 $e("Dot layer does not support tensors of 4D or higher rank yet.");let a=this.interpretAxes(t,n);if(t[a[0]]!==n[a[1]])throw new z(`Dimension incompatibility: ${t[a[0]]} !== ${n[a[1]]}`)}mergeFunction(e){if(e.length!==2)throw new z(`A \`Dot\` layer must be called on exactly 2 inputs, but received ${e.length} input(s).`);let t=e[0],n=e[1],a;return Array.isArray(this.axes)?a=this.axes.map((r,s)=>pp(r,e[s].shape.length)):a=[pp(this.axes,t.shape.length),pp(this.axes,n.shape.length)],this.normalize&&(t=im(t,a[0]),n=im(n,a[1])),PW(t,n,a)}interpretAxes(e,t){let n;return Array.isArray(this.axes)?n=this.axes:n=[pp(this.axes,e.length),pp(this.axes,t.length)],n}computeOutputShape(e){k.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 $e("Dot layer does not support tensors of 4D or higher rank yet.");let a=this.interpretAxes(t,n);t.splice(a[0],1),n.splice(a[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}};Fx.className="Dot";re.registerClass(Fx);var $x=class extends qe{constructor(e){super(e);this.supportsMasking=!0,this.stddev=e.stddev}computeOutputShape(e){return e}getConfig(){let e=super.getConfig(),t={stddev:this.stddev};return Object.assign(t,e),t}call(e,t){return D(()=>{this.invokeCallHook(e,t);let n=Me(e);return tp(()=>qh(n.shape,0,this.stddev).add(n),()=>n,t.training||!1)})}};$x.className="GaussianNoise";re.registerClass($x);var Dx=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 D(()=>{this.invokeCallHook(e,t);let n=Me(e);return this.rate>0&&this.rate<1?tp(()=>{let a=Math.sqrt(this.rate/(1-this.rate));return n.mul(qh(n.shape,1,a))},()=>n,t.training||!1):n})}};Dx.className="GaussianDropout";re.registerClass(Dx);var Rx=class extends qe{constructor(e){super(e);this.supportsMasking=!0,this.rate=e.rate,this.noiseShape=e.noiseShape}_getNoiseShape(e){return this.noiseShape||Me(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 D(()=>{if(this.rate<1&&this.rate>0){let n=this._getNoiseShape(e);return tp(()=>{let a=Me(e),r=1.6732632423543772,s=1.0507009873554805,i=-r*s,o=Jr(Wl(n),this.rate);o=Jc(o,"float32");let l=((1-this.rate)*(1+this.rate*i**2))**-.5,c=-l*i*this.rate;return a.mul(o).add(o.add(-1).mul(i)).mul(l).add(c)},()=>Me(e),t.training||!1)}return e})}};Rx.className="AlphaDropout";re.registerClass(Rx);function dp(e,t,n,a,r,s=.001){let i;if(e.rank===2)i=n1(e,t,n,a,r,s);else if(e.rank===3)i=a1(e,t,n,a,r,s);else if(e.rank===4)i=r1(e,t,n,a,r,s);else throw new $e(`batchNormalization is not implemented for array of rank ${e.rank} yet`);return i}function OW(e,t,n,a,r=.001){return D(()=>{let s=vh(e,a),i=s.mean,o=s.variance;return[dp(e,i,o,n,t,r),i,o]})}function LW(e,t,n,a,r=.001){return D(()=>{let s=vh(e,a),i=s.mean,o=s.variance,l=[];for(let h of Ea(0,e.rank))a.indexOf(h)!==-1?l.push(1):l.push(e.shape[h]);let c=i.reshape(l),u=o.reshape(l),p=t==null?null:t.reshape(l),d=n==null?null:n.reshape(l);return[dp(e,c,u,d,p,r),i,o]})}function zW(e,t,n,a,r=.001){return k.arraysEqual(a.slice().sort(),Ea(0,e.rank-1))?OW(e,t,n,a,r):LW(e,t,n,a,r)}var Mx=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=wt(e.betaInitializer||"zeros"),this.gammaInitializer=wt(e.gammaInitializer||"ones"),this.movingMeanInitializer=wt(e.movingMeanInitializer||"zeros"),this.movingVarianceInitializer=wt(e.movingVarianceInitializer||"ones"),this.betaConstraint=Vt(e.betaConstraint),this.gammaConstraint=Vt(e.gammaConstraint),this.betaRegularizer=kt(e.betaRegularizer),this.gammaRegularizer=kt(e.gammaRegularizer)}build(e){e=ht(e);let t=this.axis>=0?this.axis:this.axis+e.length,n=e[t];if(n==null)throw new z(`Axis ${t} of input tensor should have a defined dimension but the layer received an input with shape ${JSON.stringify(e)}.`);this.inputSpec=[new Xt({ndim:e.length,axes:{[t]:n}})];let a=[n];this.scale&&(this.gamma=this.addWeight("gamma",a,null,this.gammaInitializer,this.gammaRegularizer,!0,this.gammaConstraint)),this.center&&(this.beta=this.addWeight("beta",a,null,this.betaInitializer,this.betaRegularizer,!0,this.betaConstraint)),this.movingMean=this.addWeight("moving_mean",a,null,this.movingMeanInitializer,null,!1),this.movingVariance=this.addWeight("moving_variance",a,null,this.movingVarianceInitializer,null,!1),this.built=!0}call(e,t){return D(()=>{let n=t.training==null?!1:t.training,a=Me(e),r=a.shape,s=r.length,i=Ea(0,s),o=this.axis>=0?this.axis:this.axis+s;i.splice(o,1);let l=$i(1,s);l[o]=r[o];let c=i.slice();c.sort();let u=!k.arraysEqual(c,Ea(0,s).slice(0,s-1)),p=()=>{if(u){let g=this.movingMean.read().reshape(l),y=this.movingVariance.read().reshape(l),b=this.center?this.beta.read().reshape(l):null,v=this.scale?this.gamma.read().reshape(l):null;return dp(a,g,y,b,v,this.epsilon)}else return dp(a,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 p();let[d,h,m]=zW(a,this.gamma.read(),this.beta.read(),i,this.epsilon),f=(g,y,b)=>{D(()=>{let v=1-b,x=g.read(),N=x.sub(y).mul(v);g.write(x.sub(N))})};return(()=>{f(this.movingMean,h,this.momentum),f(this.movingVariance,m,this.momentum)})(),d})}getConfig(){let e={axis:this.axis,momentum:this.momentum,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:Ct(this.betaInitializer),gammaInitializer:Ct(this.gammaInitializer),movingMeanInitializer:Ct(this.movingMeanInitializer),movingVarianceInitializer:Ct(this.movingVarianceInitializer),betaRegularizer:mt(this.betaRegularizer),gammaRegularizer:mt(this.gammaRegularizer),betaConstraint:Wt(this.betaConstraint),gammaConstraint:Wt(this.gammaConstraint)},t=super.getConfig();return Object.assign(e,t),e}};Mx.className="BatchNormalization";re.registerClass(Mx);var Px=class extends qe{constructor(e){if(e==null&&(e={}),super(e),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=wt(e.betaInitializer||"zeros"),this.gammaInitializer=wt(e.gammaInitializer||"ones"),this.betaRegularizer=kt(e.betaRegularizer),this.gammaRegularizer=kt(e.gammaRegularizer),this.supportsMasking=!0}build(e){e=ht(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!==ts(this.axis).length)throw new Error(`Found duplicate axes in: ${this.axis}`);let n=this.axis.map(r=>e[r]),a=!0;this.scale?this.gamma=this.addWeight("gamma",n,"float32",this.gammaInitializer,this.gammaRegularizer,a):this.gamma=null,this.center?this.beta=this.addWeight("beta",n,"float32",this.betaInitializer,this.betaRegularizer,a):this.beta=null,this.built=!0}call(e,t){let n=Me(e),a=n.shape,r=a.length;return D(()=>{let s=!0,{mean:i,variance:o}=vh(n,this.axis,s),l=$i(1,r);for(let m of this.axis)l[m]=a[m];let c=m=>m!=null&&m.shape.length!==r&&this.axis!==[r-1]?m.reshape(l):m,u=c(this.gamma.read()),p=c(this.beta.read()),d=[],h=[];for(let m=0;m<r;++m)this.axis.indexOf(m)!==-1?(d.push(a[m]),h.push(1)):(d.push(1),h.push(a[m]));return i=i.tile(d),o=o.tile(d),u=u.tile(h),p=p.tile(h),dp(n,i,o,p,u,this.epsilon)})}getConfig(){let e={axis:this.axis,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:Ct(this.betaInitializer),gammaInitializer:Ct(this.gammaInitializer),betaRegularizer:mt(this.betaRegularizer),gammaRegularizer:mt(this.gammaRegularizer)},t=super.getConfig();return Object.assign(e,t),e}};Px.className="LayerNormalization";re.registerClass(Px);function BW(e,t,n){return D(()=>{if(e.rank!==4)throw new z(`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 z("spatial2dPadding expects `padding` to be an Array of two Arrays, each of which is an Array of two integers.");if(n==null&&(n=Ca()),n!=="channelsLast"&&n!=="channelsFirst")throw new z(`Unknown data format: ${n}. Supported data formats are 'channelsLast' and 'channelsFirst.`);let a;return n==="channelsFirst"?a=[[0,0],[0,0],t[0],t[1]]:a=[[0,0],t[0],t[1],[0,0]],Zn(e,a)})}var Ox=class extends qe{constructor(e){if(e==null&&(e={}),super(e),this.dataFormat=e.dataFormat==null?Ca():e.dataFormat,e.padding==null)this.padding=[[1,1],[1,1]];else if(typeof e.padding=="number")this.padding=[[e.padding,e.padding],[e.padding,e.padding]];else{if(e.padding=e.padding,e.padding.length!==2)throw new z(`ZeroPadding2D expects padding to be a length-2 array, but received a length-${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 z(`ZeroPadding2D expects height padding to be a length-2 array, but received a length-${e.padding[0].length} array.`);if(t=e.padding[0],e.padding[1].length!==2)throw new z(`ZeroPadding2D expects width padding to be a length-2 array, but received a length-${e.padding[1].length} array.`);n=e.padding[1]}this.padding=[t,n]}this.inputSpec=[new Xt({ndim:4})]}computeOutputShape(e){e=ht(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 D(()=>BW(Me(e),this.padding,this.dataFormat))}getConfig(){let e={padding:this.padding,dataFormat:this.dataFormat},t=super.getConfig();return Object.assign(e,t),e}};Ox.className="ZeroPadding2D";re.registerClass(Ox);function Im(e,t,n,a,r,s){return D(()=>{Dt(r),dk(s),Jn(a),n==null&&(n=[1,1]),a==null&&(a="valid"),r==null&&(r=Ca()),s==null&&(s="max"),e=sx(e,r);let i,o=a==="same"?"same":"valid";return s==="max"?i=Ft(e,t,n,o):i=Xn(e,t,n,o),r==="channelsFirst"&&(i=Ue(i,[0,3,1,2])),i})}function _I(e,t,n,a,r,s){return D(()=>{Dt(r),dk(s),Jn(a),n==null&&(n=[1,1,1]),a==null&&(a="valid"),r==null&&(r=Ca()),s==null&&(s="max"),e=kI(e,r);let i,o=a==="same"?"same":"valid";return s==="max"?i=qy(e,t,n,o):i=Dy(e,t,n,o),r==="channelsFirst"&&(i=Ue(i,[0,4,1,2,3])),i})}var EI=class extends qe{constructor(e){if(e.poolSize==null&&(e.poolSize=2),super(e),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 z(`poolSize for 1D convolutional layer must be a number or an Array of a single number, but received ${JSON.stringify(e.poolSize)}`);if(Kt(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 z(`strides for 1D convolutional layer must be a number or an Array of a single number, but received ${JSON.stringify(e.strides)}`);Kt(this.strides,"strides"),this.padding=e.padding==null?"valid":e.padding,Jn(this.padding),this.inputSpec=[new Xt({ndim:3})]}computeOutputShape(e){e=ht(e);let t=$a(e[1],this.poolSize[0],this.padding,this.strides[0]);return[e[0],t,e[2]]}call(e,t){return D(()=>{this.invokeCallHook(e,t),e=Qc(Me(e),2);let n=this.poolingFunction(Me(e),[this.poolSize[0],1],[this.strides[0],1],this.padding,"channelsLast");return Qr(n,[2])})}getConfig(){let e={poolSize:this.poolSize,padding:this.padding,strides:this.strides},t=super.getConfig();return Object.assign(e,t),e}},Lx=class extends EI{constructor(e){super(e)}poolingFunction(e,t,n,a,r){return Dt(r),Jn(a),Im(e,t,n,a,r,"max")}};Lx.className="MaxPooling1D";re.registerClass(Lx);var zx=class extends EI{constructor(e){super(e)}poolingFunction(e,t,n,a,r){return Dt(r),Jn(a),Im(e,t,n,a,r,"avg")}};zx.className="AveragePooling1D";re.registerClass(zx);var AI=class extends qe{constructor(e){if(e.poolSize==null&&(e.poolSize=[2,2]),super(e),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 z(`If the strides property of a 2D pooling layer is an Array, it is expected to have a length of 2, but received length ${e.strides.length}.`);this.strides=e.strides}else this.strides=[e.strides,e.strides];Kt(this.poolSize,"poolSize"),Kt(this.strides,"strides"),this.padding=e.padding==null?"valid":e.padding,this.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,Dt(this.dataFormat),Jn(this.padding),this.inputSpec=[new Xt({ndim:4})]}computeOutputShape(e){e=ht(e);let t=this.dataFormat==="channelsFirst"?e[2]:e[1],n=this.dataFormat==="channelsFirst"?e[3]:e[2];return t=$a(t,this.poolSize[0],this.padding,this.strides[0]),n=$a(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 D(()=>(this.invokeCallHook(e,t),this.poolingFunction(Me(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}},Bx=class extends AI{constructor(e){super(e)}poolingFunction(e,t,n,a,r){return Dt(r),Jn(a),Im(e,t,n,a,r,"max")}};Bx.className="MaxPooling2D";re.registerClass(Bx);var Wx=class extends AI{constructor(e){super(e)}poolingFunction(e,t,n,a,r){return Dt(r),Jn(a),Im(e,t,n,a,r,"avg")}};Wx.className="AveragePooling2D";re.registerClass(Wx);var FI=class extends qe{constructor(e){if(e.poolSize==null&&(e.poolSize=[2,2,2]),super(e),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 z(`If the strides property of a 3D pooling layer is an Array, it is expected to have a length of 3, but received length ${e.strides.length}.`);this.strides=e.strides}else this.strides=[e.strides,e.strides,e.strides];Kt(this.poolSize,"poolSize"),Kt(this.strides,"strides"),this.padding=e.padding==null?"valid":e.padding,this.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,Dt(this.dataFormat),Jn(this.padding),this.inputSpec=[new Xt({ndim:5})]}computeOutputShape(e){e=ht(e);let t=this.dataFormat==="channelsFirst"?e[2]:e[1],n=this.dataFormat==="channelsFirst"?e[3]:e[2],a=this.dataFormat==="channelsFirst"?e[4]:e[3];return t=$a(t,this.poolSize[0],this.padding,this.strides[0]),n=$a(n,this.poolSize[1],this.padding,this.strides[1]),a=$a(a,this.poolSize[2],this.padding,this.strides[2]),this.dataFormat==="channelsFirst"?[e[0],e[1],t,n,a]:[e[0],t,n,a,e[4]]}call(e,t){return D(()=>(this.invokeCallHook(e,t),this.poolingFunction(Me(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}},Vx=class extends FI{constructor(e){super(e)}poolingFunction(e,t,n,a,r){return Dt(r),Jn(a),_I(e,t,n,a,r,"max")}};Vx.className="MaxPooling3D";re.registerClass(Vx);var Ux=class extends FI{constructor(e){super(e)}poolingFunction(e,t,n,a,r){return Dt(r),Jn(a),_I(e,t,n,a,r,"avg")}};Ux.className="AveragePooling3D";re.registerClass(Ux);var $I=class extends qe{constructor(e){super(e);this.inputSpec=[new Xt({ndim:3})]}computeOutputShape(e){return[e[0],e[2]]}call(e,t){throw new $e}},Gx=class extends $I{constructor(e){super(e||{})}call(e,t){return D(()=>{let n=Me(e);return St(n,1)})}};Gx.className="GlobalAveragePooling1D";re.registerClass(Gx);var Hx=class extends $I{constructor(e){super(e||{})}call(e,t){return D(()=>{let n=Me(e);return Yn(n,1)})}};Hx.className="GlobalMaxPooling1D";re.registerClass(Hx);var DI=class extends qe{constructor(e){super(e);this.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,Dt(this.dataFormat),this.inputSpec=[new Xt({ndim:4})]}computeOutputShape(e){return e=e,this.dataFormat==="channelsLast"?[e[0],e[3]]:[e[0],e[1]]}call(e,t){throw new $e}getConfig(){let e={dataFormat:this.dataFormat},t=super.getConfig();return Object.assign(e,t),e}},jx=class extends DI{call(e,t){return D(()=>{let n=Me(e);return this.dataFormat==="channelsLast"?St(n,[1,2]):St(n,[2,3])})}};jx.className="GlobalAveragePooling2D";re.registerClass(jx);var qx=class extends DI{call(e,t){return D(()=>{let n=Me(e);return this.dataFormat==="channelsLast"?Yn(n,[1,2]):Yn(n,[2,3])})}};qx.className="GlobalMaxPooling2D";re.registerClass(qx);var RI=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 a=t.layer,r=Fa(a,n);delete t.layer;let s={layer:r};return Object.assign(s,t),new e(s)}},Kx=class extends RI{constructor(e){super(e);this.supportsMasking=!0}build(e){if(e=ht(e),e.length<3)throw new z(`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=ht(e);let t=[e[0]].concat(e.slice(2)),n=this.layer.computeOutputShape(t),a=e[1];return[n[0],a].concat(n.slice(1))}call(e,t){return D(()=>(e=Me(e),SI((n,a)=>[Me(this.layer.call(n,t)),[]],e,[],!1,null,null,!1,!0)[1]))}};Kx.className="TimeDistributed";re.registerClass(Kx);function WW(e){Ri(Q3,"BidirectionalMergeMode",e)}var VW="concat",Xx=class extends RI{constructor(e){super(e);let t=e.layer.getConfig(),n={};n.className=e.layer.getClassName(),n.config=t,this.forwardLayer=Fa(n),t.goBackwards=t.goBackwards!==!0;let a={};if(a.className=e.layer.getClassName(),a.config=t,this.backwardLayer=Fa(a),this.forwardLayer.name="forward_"+this.forwardLayer.name,this.backwardLayer.name="backward_"+this.backwardLayer.name,this.mergeMode=e.mergeMode===void 0?VW:e.mergeMode,WW(this.mergeMode),e.weights)throw new $e("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,a,r;return this.returnState&&(r=t.slice(1)),n=t[0],n=n,this.mergeMode==="concat"?(n[n.length-1]*=2,a=[n]):this.mergeMode==null?a=[n,n.slice()]:a=[n],this.returnState?this.mergeMode==null?a.concat(r).concat(r.slice()):[n].concat(r).concat(r.slice()):kn(a)}apply(e,t){let n=t==null?null:t.initialState,a=t==null?null:t.constants;t==null&&(t={});let r=TI(e,n,a,this.numConstants);if(e=r.inputs,n=r.initialState,a=r.constants,Array.isArray(e)&&(n=e.slice(1),e=e[0]),(n==null||n.length===0)&&a==null)return super.apply(e,t);let s=[],i=[];if(n!=null){let l=n.length;if(l%2>0)throw new z("When passing `initialState` to a Bidrectional RNN, the state should be an Array containing the states of the underlying RNNs.");t.initialState=n,s.push(...n);let c=n.map(u=>new Xt({shape:u.shape}));this.forwardLayer.stateSpec=c.slice(0,l/2),this.backwardLayer.stateSpec=c.slice(l/2),i.push(...c)}if(a!=null)throw new $e("Support for constants in Bidirectional layers is not implemented yet.");let o=s[0]instanceof Aa;for(let l of s)if(l instanceof Aa!==o)throw new z("The initial state of a Bidirectional layer cannot be specified as a mix of symbolic and non-symbolic tensors");if(o){let l=[e].concat(s),c=this.inputSpec.concat(i),u=this.inputSpec;this.inputSpec=c;let p=super.apply(l,t);return this.inputSpec=u,p}else return super.apply(e,t)}call(e,t){return D(()=>{let n=t.initialState,a,r;if(n==null)a=this.forwardLayer.call(e,t),r=this.backwardLayer.call(e,t);else{let o=n.slice(0,n.length/2),l=n.slice(n.length/2);a=this.forwardLayer.call(e,Object.assign(t,{initialState:o})),r=this.backwardLayer.call(e,Object.assign(t,{initialState:l}))}let s;this.returnState&&(Array.isArray(a)&&(s=a.slice(1).concat(r.slice(1))),a=a[0],r=r[0]),this.returnSequences&&(r=Rn(r,1));let i;return this.mergeMode==="concat"?i=Ib([a,r]):this.mergeMode==="sum"?i=Z(a,r):this.mergeMode==="ave"?i=L(.5,Z(a,r)):this.mergeMode==="mul"?i=L(a,r):this.mergeMode==null&&(i=[a,r]),this.returnState?this.mergeMode==null?i.concat(s):[i].concat(s):i})}resetStates(e){this.forwardLayer.resetStates(),this.backwardLayer.resetStates()}build(e){Mi(this.forwardLayer.name,()=>{this.forwardLayer.build(e)}),Mi(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 a=this.forwardLayer.states.map(r=>null);return Array.isArray(n)?n.concat(a).concat(a):[n].concat(a).concat(a)}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=Fa(t.layer);if(delete t.layer,t.numConstants!=null)throw new $e("Deserialization of a Bidirectional layer with numConstants present is not supported yet.");let a=t;return a.layer=n,new e(a)}};Xx.className="Bidirectional";re.registerClass(Xx);function fz(e){return new Kl(e)}function gz(e){return new nx(e)}function yz(e){return new Qb(e)}function bz(e){return new ex(e)}function xz(e){return new tx(e)}function vz(e){return new rx(e)}function wz(e){return new ax(e)}function kz(e){return new ym(e)}function Iz(e){return new up(e)}function Nz(e){return new ox(e)}function Tz(e){return new gm(e)}function Sz(e){return new lx(e)}function Cz(e){return new ux(e)}function _z(e){return new cx(e)}function Ez(e){return new px(e)}function Az(e){return new xx(e)}function Fz(e){return new yx(e)}function $z(e){return new km(e)}function Dz(e){return new gx(e)}function Rz(e){return new bx(e)}function Mz(e){return new vx(e)}function Pz(e){return new wx(e)}function Oz(e){return new kx(e)}function Lz(e){return new Nx(e)}function zz(e){return new Tx(e)}function Bz(e){return new Cx(e)}function Wz(e){return new Ax(e)}function Vz(e){return new _x(e)}function Uz(e){return new Ex(e)}function Gz(e){return new Sx(e)}function Hz(e){return new Fx(e)}function jz(e){return new Mx(e)}function qz(e){return new Px(e)}function Kz(e){return new Ox(e)}function $b(e){return new zx(e)}function Xz(e){return $b(e)}function Yz(e){return $b(e)}function Db(e){return new Wx(e)}function Zz(e){return Db(e)}function Jz(e){return Db(e)}function Rb(e){return new Ux(e)}function Qz(e){return Rb(e)}function eB(e){return Rb(e)}function tB(e){return new Gx(e)}function nB(e){return new jx(e)}function Tk(e){return new Hx(e)}function Sk(e){return new qx(e)}function Ck(e){return new Lx(e)}function _k(e){return new Bx(e)}function aB(e){return new Vx(e)}function rB(e){return new hx(e)}function sB(e){return new vm(e)}function iB(e){return new mx(e)}function oB(e){return new cp(e)}function lB(e){return new dx(e)}function uB(e){return new xm(e)}function cB(e){return new fx(e)}function pB(e){return new wm(e)}function dB(e){return new tr(e)}function hB(e){return new bm(e)}function mB(e){return new Xx(e)}function fB(e){return new Kx(e)}var gB=Tk,yB=Sk,bB=Ck,xB=_k;function vB(e){return new $x(e)}function wB(e){return new Dx(e)}function kB(e){return new Rx(e)}function IB(e){return new Ix(e)}var MI={};Oe(MI,{MAPE:()=>QW,MSE:()=>n4,binaryAccuracy:()=>UW,binaryCrossentropy:()=>GW,categoricalAccuracy:()=>jW,categoricalCrossentropy:()=>qW,cosineProximity:()=>YW,mape:()=>e4,meanAbsoluteError:()=>ZW,meanAbsolutePercentageError:()=>JW,meanSquaredError:()=>t4,mse:()=>a4,precision:()=>KW,recall:()=>XW,sparseCategoricalAccuracy:()=>HW});function UW(e,t){return Bb(e,t)}function GW(e,t){return Uk(e,t)}function HW(e,t){return Gk(e,t)}function jW(e,t){return Wb(e,t)}function qW(e,t){return Vb(e,t)}function KW(e,t){return Vk(e,t)}function XW(e,t){return VB(e,t)}function YW(e,t){return Lb(e,t)}function ZW(e,t){return om(e,t)}function JW(e,t){return Yl(e,t)}function QW(e,t){return Yl(e,t)}function e4(e,t){return Yl(e,t)}function t4(e,t){return Oi(e,t)}function n4(e,t){return Oi(e,t)}function a4(e,t){return Oi(e,t)}var PI={};Oe(PI,{modelFromJSON:()=>kW});var OI={};Oe(OI,{l1:()=>s4,l1l2:()=>r4,l2:()=>i4});function r4(e){return new op(e)}function s4(e){return AW(e)}function i4(e){return FW(e)}var LI=class extends Xl{constructor(){super(...arguments);this.model=null}setModel(e){if(!(e instanceof wr))throw new Error("model must be a LayersModel, not some other Container");this.model=e}};function Nm(e,t){return e<t}function zI(e,t){return e>t}var BI=class extends LI{constructor(e){super();if(e==null&&(e={}),e.restoreBestWeights)throw new $e("restoreBestWeights = True is not implemented in EarlyStopping yet.");this.monitor=e.monitor||"val_loss",this.minDelta=Math.abs(e.minDelta||0),this.patience=e.patience||0,this.verbose=e.verbose||0,this.mode=e.mode||"auto",this.baseline=e.baseline,["auto","min","max"].indexOf(this.mode)===-1&&(console.warn(`EarlyStopping mode '${this.mode}' is invalid. Falling back to mode 'auto'.`),this.mode="auto"),this.mode==="min"?this.monitorFunc=Nm:this.mode==="max"?this.monitorFunc=zI:this.monitor.indexOf("acc")!==-1?this.monitorFunc=zI:this.monitorFunc=Nm,this.monitorFunc===Nm&&(this.minDelta*=-1)}async onTrainBegin(e){this.wait=0,this.stoppedEpoch=0,this.baseline!=null?this.best=this.baseline:this.best=this.monitorFunc===Nm?Infinity:-Infinity}async onEpochEnd(e,t){await rs(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 o4(e){return new BI(e)}var l4={earlyStopping:o4},Da;(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"})(Da||(Da={}));var WI;(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={}))})(WI||(WI={}));var Yx={};function u4(e,t){let n={tfOpName:e,category:"custom",inputs:[],attrs:[],customExecutor:t};Yx[e]=n}function VI(e){return Yx[e]}function c4(e){delete Yx[e]}function I(e,t,n,a,r){let s=t.inputParams[e];if(s&&s.inputIndexStart!==void 0){let o=s.inputIndexStart,l=s.inputIndexEnd===0?void 0:s.inputIndexEnd===void 0?o+1:s.inputIndexEnd;if(s.type==="tensor")return Nn(t.inputNames[s.inputIndexStart],n,a,r);if(s.type==="tensors")return t.inputNames.slice(o,l).map(p=>Nn(p,n,a,r));let c=Nn(t.inputNames.slice(o)[0],n,a,r),u=c.dataSync();return s.type==="number"?u[0]:k.toNestedArray(c.shape,u)}let i=t.attrParams[e];return i&&i.value}function Nn(e,t,n,a){let[r,s]=On(e);if(a!=null){let o=a.getHashTableHandleByName(r);if(o!=null)return o}let i=n.currentContextIds.find(o=>!!t[Tm(r,o)]);return i!==void 0?t[Tm(r,i)][s]:void 0}function p4(e,t,n){return t[Tm(e,n.currentContextId)]}function kr(e,t){let[n,a]=On(e);return[Tm(n,t&&t.currentContextId),a]}function Tm(e,t){return t?`${e}-${t}`:e}function On(e){let t=e.split(":");return t.length===1?[e,0]:[t[0],Number(t[t.length-1])]}function Sm(e,t,n){let a=I("pad",e,t,n);if(a==="explicit"){a=I("explicitPaddings",e,t,n);let r=[[0,0],[0,0],[0,0],[0,0]];for(let s=0;s<4;s++)r[s][0]=a[s*2],r[s][1]=a[s*2+1];return r}return a}function Ir(e){return e.kept?e:jr(e)}var UI={};Oe(UI,{json:()=>d4});var d4=[{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}]}],GI={};Oe(GI,{json:()=>h4});var h4=[{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}]}],HI={};Oe(HI,{json:()=>m4});var m4=[{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"}]}],jI={};Oe(jI,{json:()=>f4});var f4=[{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"}]}],qI={};Oe(qI,{json:()=>g4});var g4=[{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"}]}],KI={};Oe(KI,{json:()=>y4});var y4=[{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}]}],XI={};Oe(XI,{json:()=>b4});var b4=[{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"}]}],YI={};Oe(YI,{json:()=>x4});var x4=[{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"}]}],ZI={};Oe(ZI,{json:()=>v4});var v4=[{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}]}],JI={};Oe(JI,{json:()=>w4});var w4=[{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"}]}],QI={};Oe(QI,{json:()=>k4});var k4=[{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}]}],eN={};Oe(eN,{json:()=>I4});var I4=[{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}]}],tN={};Oe(tN,{json:()=>N4});var N4=[{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}]}],nN={};Oe(nN,{json:()=>T4});var T4=[{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"}]}],aN={};Oe(aN,{json:()=>S4});var S4=[{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}]}],rN={};Oe(rN,{json:()=>C4});var C4=[{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}]}],sN={};Oe(sN,{json:()=>_4});var _4=[{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:[]}],oN=class{static get Instance(){return this._instance||(this._instance=new this)}constructor(){let e=[UI,GI,HI,jI,qI,KI,XI,QI,JI,YI,eN,tN,nN,aN,rN,sN,ZI],t=[].concat(...e.map(n=>n.json));this.opMappers=t.reduce((n,a)=>(n[a.tfOpName]=a,n),{})}transformGraph(e,t={}){let n=e.node,a=[],r=[],s=[],i=n.reduce((m,f)=>(m[f.name]=this.mapNode(f),f.op.startsWith("Placeholder")?a.push(m[f.name]):f.op==="Const"?r.push(m[f.name]):(f.input==null||f.input.length===0)&&s.push(m[f.name]),m),{}),o=[],l=[],c={},u={};t!=null&&(c=this.mapSignatureEntries(t.inputs),u=this.mapSignatureEntries(t.outputs));let p=Object.keys(i);p.forEach(m=>{let f=i[m];f.inputNames.forEach(g=>{let[y]=kr(g);f.inputs.push(i[y]),i[y].children.push(f)})}),Object.keys(u).length===0?p.forEach(m=>{let f=i[m];f.children.length===0&&l.push(f)}):Object.keys(u).forEach(m=>{let[f]=kr(m),g=i[f];g!=null&&(g.signatureKey=u[m],l.push(g))}),Object.keys(c).length>0?Object.keys(c).forEach(m=>{let[f]=kr(m),g=i[f];g&&(g.signatureKey=c[m],o.push(g))}):o=a;let d={};e.library!=null&&e.library.function!=null&&(d=e.library.function.reduce((m,f)=>(m[f.signature.name]=this.mapFunction(f),m),{}));let h={nodes:i,inputs:o,outputs:l,weights:r,placeholders:a,signature:t,functions:d};return s.length>0&&(h.initNodes=s),h}mapSignatureEntries(e){return Object.keys(e||{}).reduce((t,n)=>(t[e[n].name]=n,t),{})}mapNode(e){let t=VI(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(a=>a.startsWith("^")?a.substr(1):a),inputs:[],children:[],inputParams:{},attrParams:{},rawAttrs:e.attr};return t.inputs!=null&&(n.inputParams=t.inputs.reduce((a,r)=>(a[r.name]={type:r.type,inputIndexStart:r.start,inputIndexEnd:r.end},a),{})),t.attrs!=null&&(n.attrParams=t.attrs.reduce((a,r)=>{let s=r.type,i;switch(r.type){case"string":i=Zx(e.attr,r.tfName,r.defaultValue),i===void 0&&!!r.tfDeprecatedName&&(i=Zx(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"string[]":i=sv(e.attr,r.tfName,r.defaultValue),i===void 0&&!!r.tfDeprecatedName&&(i=sv(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"number":i=Qx(e.attr,r.tfName,r.defaultValue||0),i===void 0&&!!r.tfDeprecatedName&&(i=Qx(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"number[]":i=rv(e.attr,r.tfName,r.defaultValue),i===void 0&&!!r.tfDeprecatedName&&(i=rv(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"bool":i=Jx(e.attr,r.tfName,r.defaultValue),i===void 0&&!!r.tfDeprecatedName&&(i=Jx(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"bool[]":i=ov(e.attr,r.tfName,r.defaultValue),i===void 0&&!!r.tfDeprecatedName&&(i=ov(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"shape":i=av(e.attr,r.tfName,r.defaultValue),i===void 0&&!!r.tfDeprecatedName&&(i=av(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"shape[]":i=iv(e.attr,r.tfName,r.defaultValue),i===void 0&&!!r.tfDeprecatedName&&(i=iv(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"dtype":i=tv(e.attr,r.tfName,r.defaultValue),i===void 0&&!!r.tfDeprecatedName&&(i=tv(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"dtype[]":i=nv(e.attr,r.tfName,r.defaultValue),i===void 0&&!!r.tfDeprecatedName&&(i=nv(e.attr,r.tfDeprecatedName,r.defaultValue));break;case"func":i=iN(e.attr,r.tfName,r.defaultValue),i===void 0&&!!r.tfDeprecatedName&&(i=iN(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 a[r.name]={value:i,type:s},a},{})),n}mapFunction(e){let t=e.nodeDef,n=[],a=[],r={};t!=null&&(r=t.reduce((c,u)=>(c[u.name]=this.mapNode(u),u.op==="Const"&&a.push(c[u.name]),c),{}));let s=[],i=[];e.signature.inputArg.forEach(c=>{let[u]=kr(c.name),p={name:u,op:"Placeholder",inputs:[],inputNames:[],category:"graph",inputParams:{},attrParams:{dtype:{value:ev(c.type),type:"dtype"}},children:[]};p.signatureKey=c.name,s.push(p),r[u]=p}),Object.keys(r).forEach(c=>{let u=r[c];u.inputNames.forEach(p=>{let[d]=kr(p);u.inputs.push(r[d]),r[d].children.push(u)})});let o=e.ret;e.signature.outputArg.forEach(c=>{let[u,p]=kr(o[c.name]),d=r[u];d!=null&&(d.defaultOutput=p,i.push(d))});let l=this.mapArgsToSignature(e);return{nodes:r,inputs:s,outputs:i,weights:a,placeholders:n,signature:l}}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 E4(e){let t=ee().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 lN(e,t){let n=Array.isArray(e)?String.fromCharCode.apply(null,e):E4(e);return t?n:n.toLowerCase()}function Zx(e,t,n,a=!1){let r=e[t];return r!=null?lN(r.s,a):n}function Jx(e,t,n){let a=e[t];return a?a.b:n}function Qx(e,t,n){let a=e[t]||{},r=a.i!=null?a.i:a.f!=null?a.f:n;return typeof r=="number"?r:parseInt(r,10)}function ev(e){switch(typeof e=="string"&&(e=Da[e]),e){case Da.DT_FLOAT:return"float32";case Da.DT_INT32:case Da.DT_INT64:case Da.DT_INT8:case Da.DT_UINT8:return"int32";case Da.DT_BOOL:return"bool";case Da.DT_DOUBLE:return"float32";case Da.DT_STRING:return"string";default:return null}}function iN(e,t,n){let a=e[t];return a&&a.func?a.func.name:n}function tv(e,t,n){let a=e[t];return a&&a.type?ev(a.type):n}function nv(e,t,n){let a=e[t];return a&&a.list&&a.list.type?a.list.type.map(r=>ev(r)):n}function uN(e){if(!e.unknownRank)return e.dim!=null?e.dim.map(t=>typeof t.size=="number"?t.size:parseInt(t.size,10)):[]}function av(e,t,n){let a=e[t];return a&&a.shape?uN(a.shape):n}function rv(e,t,n){let a=e[t];return a?((a.list.f&&a.list.f.length?a.list.f:a.list.i)||[]).map(r=>typeof r=="number"?r:parseInt(r,10)):n}function sv(e,t,n,a=!1){let r=e[t];return r&&r.list&&r.list.s?r.list.s.map(s=>lN(s,a)):n}function iv(e,t,n){let a=e[t];return a&&a.list&&a.list.shape?a.list.shape.map(r=>uN(r)):n}function ov(e,t,n){let a=e[t];return a&&a.list&&a.list.b?a.list.b:n}var A4=class{constructor(e,t,n){this.node=e,this.tensorMap=t,this.context=n,this.inputs=[],this.attrs={},this.inputs=e.inputNames.map(a=>this.getInput(a)),e.rawAttrs!=null&&(this.attrs=Object.keys(e.rawAttrs).reduce((a,r)=>(a[r]=this.getAttr(r),a),{}))}getInput(e){return Nn(e,this.tensorMap,this.context)}getAttr(e,t){let n=this.node.rawAttrs[e];if(n.tensor!=null)return Nn(e,this.tensorMap,this.context);if(n.i!=null||n.f!=null)return Qx(this.node.rawAttrs,e,t);if(n.s!=null)return Zx(this.node.rawAttrs,e,t);if(n.b!=null)return Jx(this.node.rawAttrs,e,t);if(n.shape!=null)return av(this.node.rawAttrs,e,t);if(n.type!=null)return tv(this.node.rawAttrs,e,t);if(n.list!=null){if(n.list.i!=null||n.list.f!=null)return rv(this.node.rawAttrs,e,t);if(n.list.s!=null)return sv(this.node.rawAttrs,e,t);if(n.list.shape!=null)return iv(this.node.rawAttrs,e,t);if(n.list.b!=null)return ov(this.node.rawAttrs,e,t);if(n.list.type!=null)return nv(this.node.rawAttrs,e,t)}return t}},F4=(e,t,n)=>{switch(e.op){case"BiasAdd":case"AddV2":case"Add":return[Z(I("a",e,t,n),I("b",e,t,n))];case"AddN":return[J0(I("tensors",e,t,n))];case"FloorMod":case"Mod":return[Xy(I("a",e,t,n),I("b",e,t,n))];case"Mul":return[L(I("a",e,t,n),I("b",e,t,n))];case"RealDiv":case"Div":return[ve(I("a",e,t,n),I("b",e,t,n))];case"DivNoNan":return[zy(I("a",e,t,n),I("b",e,t,n))];case"FloorDiv":return[oh(I("a",e,t,n),I("b",e,t,n))];case"Sub":return[fe(I("a",e,t,n),I("b",e,t,n))];case"Minimum":return[Bl(I("a",e,t,n),I("b",e,t,n))];case"Maximum":return[Ka(I("a",e,t,n),I("b",e,t,n))];case"Pow":return[gr(I("a",e,t,n),I("b",e,t,n))];case"SquaredDifference":return[Ah(I("a",e,t,n),I("b",e,t,n))];default:throw TypeError(`Node type ${e.op} is not implemented`)}},$4=(e,t,n)=>{switch(e.op){case"Abs":case"ComplexAbs":return[Lt(I("x",e,t,n))];case"Acos":return[Iy(I("x",e,t,n))];case"Acosh":return[Ny(I("x",e,t,n))];case"Asin":return[Sy(I("x",e,t,n))];case"Asinh":return[Cy(I("x",e,t,n))];case"Atan":return[_y(I("x",e,t,n))];case"Atan2":return[Ey(I("x",e,t,n),I("y",e,t,n))];case"Atanh":return[Ay(I("x",e,t,n))];case"Ceil":return[Ry(I("x",e,t,n))];case"Complex":return[Ur(I("real",e,t,n),I("imag",e,t,n))];case"Cos":return[zc(I("x",e,t,n))];case"Cosh":return[dh(I("x",e,t,n))];case"Elu":return[Pl(I("x",e,t,n))];case"Erf":return[By(I("x",e,t,n))];case"Exp":return[pn(I("x",e,t,n))];case"Expm1":return[Wy(I("x",e,t,n))];case"Floor":return[Ol(I("x",e,t,n))];case"Log":return[$n(I("x",e,t,n))];case"Log1p":return[gh(I("x",e,t,n))];case"Imag":return[mh(I("x",e,t,n))];case"Neg":return[Tt(I("x",e,t,n))];case"Reciprocal":return[Jy(I("x",e,t,n))];case"Real":return[Gc(I("x",e,t,n))];case"Relu":return[Ye(I("x",e,t,n))];case"Round":return[Qy(I("x",e,t,n))];case"Selu":return[Th(I("x",e,t,n))];case"Sigmoid":return[ua(I("x",e,t,n))];case"Sin":return[Sh(I("x",e,t,n))];case"Sign":return[eb(I("x",e,t,n))];case"Sinh":return[Ch(I("x",e,t,n))];case"Softplus":return[Ll(I("x",e,t,n))];case"Sqrt":return[en(I("x",e,t,n))];case"Square":return[lt(I("x",e,t,n))];case"Tanh":return[Ml(I("x",e,t,n))];case"Tan":return[ab(I("x",e,t,n))];case"ClipByValue":return[qt(I("x",e,t,n),I("clipValueMin",e,t,n),I("clipValueMax",e,t,n))];case"Relu6":return[Ih(I("x",e,t,n))];case"Rsqrt":return[Nh(Nn(e.inputNames[0],t,n))];case"Prod":return[wh(I("x",e,t,n),I("axes",e,t,n))];case"LeakyRelu":return[Bc(I("x",e,t,n),I("alpha",e,t,n))];case"Prelu":return[Uc(I("x",e,t,n),I("alpha",e,t,n))];default:throw TypeError(`Node type ${e.op} is not implemented`)}};function fa(e,t,n=""){k.assert(D4(e,t),()=>n+` Shapes ${e} and ${t} must match`)}function D4(e,t){if(e.length!==t.length)return!1;for(let n=0;n<e.length;n++)if(e[n]!==-1&&t[n]!==-1&&e[n]!==t[n])return!1;return!0}var R4=class{constructor(e,t,n,a,r,s,i){this.name=e,this.dtype=t,this.maxSize=n,this.elementShape=a,this.identicalElementShapes=r,this.dynamicSize=s,this.clearAfterRead=i,this.tensors=[],this.closed_=!1,this.idTensor=de(0),jt(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),fa(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,jt(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,a)=>this.write(n,t[a]))}gather(e,t){if(!!t&&t!==this.dtype)throw new Error(`TensorArray dtype is ${this.dtype} but gather requested dtype ${t}`);if(e)e=e.slice(0,this.size());else{e=[];for(let a=0;a<this.size();a++)e.push(a)}if(e.length===0)return qn([],[0].concat(this.elementShape));let n=this.readMany(e);return fa(this.elementShape,n[0].shape,"TensorArray shape mismatch: "),$t(n,0)}concat(e){if(!!e&&e!==this.dtype)throw new Error(`TensorArray dtype is ${this.dtype} but concat requested dtype ${e}`);if(this.size()===0)return qn([],[0].concat(this.elementShape));let t=[];for(let a=0;a<this.size();a++)t.push(a);let n=this.readMany(t);return fa(this.elementShape,n[0].shape,`TensorArray shape mismatch: tensor array shape (${this.elementShape}) vs first tensor shape (${n[0].shape})`),Qe(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,dt(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,a=e.map(o=>(n+=o,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,s=[];D(()=>{t=j(t,[1,n,r]);for(let o=0;o<e.length;++o){let l=o===0?0:a[o-1],c=[0,l,0],u=[1,e[o],r];s[o]=j(We(t,c,u),this.elementShape)}return s});let i=[];for(let o=0;o<e.length;o++)i[o]=o;this.writeMany(i,s)}},hp=class{constructor(e,t,n,a=-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}`);fa(t,r.shape,"TensorList shape mismatch: "),jt(r)}),this.idTensor=de(0),this.maxNumElements=a,jt(this.idTensor)}get id(){return this.idTensor.id}copy(){return new hp([...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.`);return fa(e,this.elementShape,"TensorList shape mismatch: "),D(()=>{let a=this.tensors.map(r=>j(r,e));return $t(a,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=this.tensors.pop();return fa(n.shape,e,"TensorList shape mismatch: "),j(n,e)}pushBack(e){if(e.dtype!==this.elementDtype)throw new Error(`Invalid data types; op elements ${e.dtype}, but list elements ${this.elementDtype}`);if(fa(e.shape,this.elementShape,"TensorList shape mismatch: "),this.maxNumElements===this.size())throw new Error("Trying to push element into a full list.");jt(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.`);return fa(this.tensors[e].shape,t,"TensorList shape mismatch: "),this.tensors[e]}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.`);fa(this.elementShape,t.shape,"TensorList shape mismatch: "),jt(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}`);return fa(this.elementShape,n,"TensorList shape mismatch: "),e=e.slice(0,this.size()),e.length===0?qn([],[0].concat(this.elementShape)):D(()=>{let a=e.map(r=>j(this.tensors[r],n));return $t(a,0)})}concat(e,t){if(!!e&&e!==this.elementDtype)throw new Error(`TensorList dtype is ${this.elementDtype} but concat requested dtype ${e}`);return fa(this.elementShape,t,"TensorList shape mismatch: "),this.size()===0?qn([],[0].concat(this.elementShape)):D(()=>{let n=this.tensors.map(a=>j(a,t));return Qe(n,0)})}};function M4(e,t,n){let a=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);fa(r,t,"TensorList shape mismatch: ");let s=dt(e);return new hp(s,t,a)}function P4(e,t,n){return new hp([],e,t,n)}function O4(e,t,n,a){if(t.length!==e.shape[0])throw new Error(`Expected len(indices) == tensor.shape[0], but saw: ${t.length} vs. ${e.shape[0]}`);let r=Math.max(...t);if(a!=null&&a!==-1&&r>=a)throw new Error(`Max index must be < array size (${r} vs. ${a})`);let s=new hp([],n,e.dtype,a),i=dt(e,0);return t.forEach((o,l)=>{s.setItem(o,i[l])}),s}function L4(e,t,n){let a=0,r=t.map(l=>(a+=l,a));if(a!==e.shape[0])throw new Error(`Expected sum of lengths to be equal to
|
|
tensor.shape[0], but sum of lengths is
|
|
${a}, and tensor's shape is: ${e.shape}`);let s=a===0?0:e.size/a,i=D(()=>{let l=[];e=j(e,[1,a,s]);for(let c=0;c<t.length;++c){let u=c===0?0:r[c-1],p=[0,u,0],d=[1,t[c],s];l[c]=j(We(e,p,d),n)}return e.dispose(),l}),o=new hp([],n,e.dtype,t.length);for(let l=0;l<i.length;l++)o.setItem(l,i[l]);return o}var z4=async(e,t,n)=>{switch(e.op){case"If":case"StatelessIf":{let a=I("thenBranch",e,t,n),r=I("elseBranch",e,t,n),s=I("cond",e,t,n),i=I("args",e,t,n);return(await s.data())[0]?n.functionMap[a].executeFunctionAsync(i,n.tensorArrayMap,n.tensorListMap):n.functionMap[r].executeFunctionAsync(i,n.tensorArrayMap,n.tensorListMap)}case"While":case"StatelessWhile":{let a=I("body",e,t,n),r=I("cond",e,t,n),s=I("args",e,t,n),i=await n.functionMap[r].executeFunctionAsync(s,n.tensorArrayMap,n.tensorListMap),o=s.map(u=>u.id),l=await i[0].data();i.forEach(u=>{!u.kept&&o.indexOf(u.id)===-1&&u.dispose()});let c=s;for(;l[0];){let u=c;c=await n.functionMap[a].executeFunctionAsync(c,n.tensorArrayMap,n.tensorListMap);let p=c.map(h=>h.id);u.forEach(h=>{!h.kept&&o.indexOf(h.id)===-1&&p.indexOf(h.id)===-1&&h.dispose()});let d=await n.functionMap[r].executeFunctionAsync(c,n.tensorArrayMap,n.tensorListMap);l=await d[0].data(),d.forEach(h=>{!h.kept&&o.indexOf(h.id)===-1&&p.indexOf(h.id)===-1&&h.dispose()})}return c}case"LoopCond":{let a=I("pred",e,t,n);return[Ir(a)]}case"Switch":{let a=I("pred",e,t,n),r=I("data",e,t,n);return r.kept||(r=Ir(r)),(await a.data())[0]?[void 0,r]:[r,void 0]}case"Merge":{let a=e.inputNames.find(r=>Nn(r,t,n)!==void 0);if(a){let r=Nn(a,t,n);return[Ir(r)]}return}case"Enter":{let a=I("frameName",e,t,n),r=I("tensor",e,t,n);return n.enterFrame(a),[Ir(r)]}case"Exit":{let a=I("tensor",e,t,n);return n.exitFrame(),[Ir(a)]}case"NextIteration":{let a=I("tensor",e,t,n);return n.nextIteration(),[Ir(a)]}case"TensorArrayV3":{let a=I("size",e,t,n),r=I("dtype",e,t,n),s=I("elementShape",e,t,n),i=I("dynamicSize",e,t,n),o=I("clearAfterRead",e,t,n),l=I("identicalElementShapes",e,t,n),c=I("name",e,t,n),u=new R4(c,r,a,s,l,i,o);return n.addTensorArray(u),[u.idTensor,de(1)]}case"TensorArrayWriteV3":{let a=I("tensorArrayId",e,t,n),r=I("index",e,t,n),s=I("tensor",e,t,n),i=n.getTensorArray(a.id);return i.write(r,s),[i.idTensor]}case"TensorArrayReadV3":{let a=I("tensorArrayId",e,t,n),r=I("index",e,t,n);return[n.getTensorArray(a.id).read(r)]}case"TensorArrayGatherV3":{let a=I("tensorArrayId",e,t,n),r=I("indices",e,t,n),s=I("dtype",e,t,n);return[n.getTensorArray(a.id).gather(r,s)]}case"TensorArrayScatterV3":{let a=I("tensorArrayId",e,t,n),r=I("indices",e,t,n),s=I("tensor",e,t,n),i=n.getTensorArray(a.id);return i.scatter(r,s),[i.idTensor]}case"TensorArrayConcatV3":{let a=I("tensorArrayId",e,t,n),r=n.getTensorArray(a.id),s=I("dtype",e,t,n);return[r.concat(s)]}case"TensorArraySplitV3":{let a=I("tensorArrayId",e,t,n),r=I("tensor",e,t,n),s=I("lengths",e,t,n),i=n.getTensorArray(a.id);return i.split(s,r),[i.idTensor]}case"TensorArraySizeV3":{let a=I("tensorArrayId",e,t,n),r=n.getTensorArray(a.id);return[de(r.size(),"int32")]}case"TensorArrayCloseV3":{let a=I("tensorArrayId",e,t,n),r=n.getTensorArray(a.id);return r.clearAndClose(),[r.idTensor]}case"TensorListSetItem":{let a=I("tensorListId",e,t,n),r=I("index",e,t,n),s=I("tensor",e,t,n),i=n.getTensorList(a.id);return i.setItem(r,s),[i.idTensor]}case"TensorListGetItem":{let a=I("tensorListId",e,t,n),r=I("index",e,t,n),s=I("elementShape",e,t,n),i=I("elementDType",e,t,n);return[n.getTensorList(a.id).getItem(r,s,i)]}case"TensorListScatterV2":case"TensorListScatter":{let a=I("indices",e,t,n),r=I("tensor",e,t,n),s=I("elementShape",e,t,n),i=I("numElements",e,t,n),o=O4(r,a,s,i);return n.addTensorList(o),[o.idTensor]}case"TensorListReserve":case"EmptyTensorList":{let a=I("elementShape",e,t,n),r=I("elementDType",e,t,n),s;e.op==="TensorListReserve"?s="numElements":s="maxNumElements";let i=I(s,e,t,n),o=P4(a,r,i);return n.addTensorList(o),[o.idTensor]}case"TensorListGather":{let a=I("tensorListId",e,t,n),r=I("indices",e,t,n),s=I("elementShape",e,t,n),i=I("elementDType",e,t,n);return[n.getTensorList(a.id).gather(r,i,s)]}case"TensorListStack":{let a=I("tensorListId",e,t,n),r=I("elementShape",e,t,n),s=I("elementDType",e,t,n),i=I("numElements",e,t,n);return[n.getTensorList(a.id).stack(r,s,i)]}case"TensorListFromTensor":{let a=I("tensor",e,t,n),r=I("elementShape",e,t,n),s=I("elementDType",e,t,n),i=M4(a,r,s);return n.addTensorList(i),[i.idTensor]}case"TensorListConcat":{let a=I("tensorListId",e,t,n),r=n.getTensorList(a.id),s=I("dtype",e,t,n),i=I("elementShape",e,t,n);return[r.concat(s,i)]}case"TensorListPushBack":{let a=I("tensorListId",e,t,n),r=I("tensor",e,t,n),s=n.getTensorList(a.id);return s.pushBack(r),[s.idTensor]}case"TensorListPopBack":{let a=I("tensorListId",e,t,n),r=I("elementShape",e,t,n),s=I("elementDType",e,t,n);return[n.getTensorList(a.id).popBack(r,s)]}case"TensorListSplit":{let a=I("tensor",e,t,n),r=I("elementShape",e,t,n),s=I("lengths",e,t,n),i=L4(a,s,r);return n.addTensorList(i),[i.idTensor]}default:throw TypeError(`Node type ${e.op} is not implemented`)}};function cN(e,t,n){let[a,r]=I("fusedOps",e,t,n),s=a==="biasadd",i=r==="prelu",o=a==="fusedbatchnorm",l=I("numArgs",e,t,n);if(s){if(i&&l!==2)throw new Error("FusedConv2d and DepthwiseConv2d with BiasAdd and Prelu must have two extra arguments: bias and alpha.");if(!i&&l!==1)throw new Error("FusedConv2d and DepthwiseConv2d with BiasAdd must have one extra argument: bias.")}if(o)throw new Error("FusedConv2d and DepthwiseConv2d with FusedBatchNorm is not supported.");let c=I("strides",e,t,n),u=Sm(e,t,n),p=I("dataFormat",e,t,n).toUpperCase(),d=I("dilations",e,t,n),[h,m]=I("args",e,t,n),f=I("leakyreluAlpha",e,t,n);return{stride:c,pad:u,dataFormat:p,dilations:d,biasArg:h,preluArg:m,activationFunc:r,leakyreluAlpha:f}}var B4=(e,t,n)=>{switch(e.op){case"Conv1D":{let a=I("stride",e,t,n),r=I("pad",e,t,n),s=I("dataFormat",e,t,n).toUpperCase(),i=I("dilation",e,t,n);return[ch(I("x",e,t,n),I("filter",e,t,n),a,r,s,i)]}case"Conv2D":{let a=I("strides",e,t,n),r=Sm(e,t,n),s=I("dataFormat",e,t,n).toUpperCase(),i=I("dilations",e,t,n);return[At(I("x",e,t,n),I("filter",e,t,n),[a[1],a[2]],r,s,[i[1],i[2]])]}case"_FusedConv2D":{let{stride:a,pad:r,dataFormat:s,dilations:i,biasArg:o,preluArg:l,activationFunc:c,leakyreluAlpha:u}=cN(e,t,n);return[es.conv2d({x:I("x",e,t,n),filter:I("filter",e,t,n),strides:[a[1],a[2]],pad:r,dataFormat:s,dilations:[i[1],i[2]],bias:o,activation:c,preluActivationWeights:l,leakyreluAlpha:u})]}case"FusedDepthwiseConv2dNative":{let{stride:a,pad:r,dataFormat:s,dilations:i,biasArg:o,preluArg:l,activationFunc:c,leakyreluAlpha:u}=cN(e,t,n);return[es.depthwiseConv2d({x:I("x",e,t,n),filter:I("filter",e,t,n),strides:[a[1],a[2]],pad:r,dataFormat:s,dilations:[i[1],i[2]],bias:o,activation:c,preluActivationWeights:l,leakyreluAlpha:u})]}case"Conv2DBackpropInput":case"Conv2dTranspose":{let a=I("outputShape",e,t,n),r=I("strides",e,t,n),s=Sm(e,t,n);return[ph(I("x",e,t,n),I("filter",e,t,n),a,[r[1],r[2]],s)]}case"DepthwiseConv2dNative":case"DepthwiseConv2d":{let a=I("strides",e,t,n),r=Sm(e,t,n),s=I("dilations",e,t,n),i=I("dataFormat",e,t,n).toUpperCase();return[Yr(I("input",e,t,n),I("filter",e,t,n),[a[1],a[2]],r,i,[s[1],s[2]])]}case"Conv3D":{let a=I("strides",e,t,n),r=I("pad",e,t,n),s=I("dataFormat",e,t,n).toUpperCase(),i=I("dilations",e,t,n);return[Py(I("x",e,t,n),I("filter",e,t,n),[a[1],a[2],a[3]],r,s,[i[1],i[2],i[3]])]}case"AvgPool":{let a=I("strides",e,t,n),r=I("pad",e,t,n),s=I("kernelSize",e,t,n);return[Xn(I("x",e,t,n),[s[1],s[2]],[a[1],a[2]],r)]}case"MaxPool":{let a=I("strides",e,t,n),r=I("pad",e,t,n),s=I("kernelSize",e,t,n);return[Ft(I("x",e,t,n),[s[1],s[2]],[a[1],a[2]],r)]}case"MaxPoolWithArgmax":{let a=I("strides",e,t,n),r=I("pad",e,t,n),s=I("kernelSize",e,t,n),i=I("includeBatchInIndex",e,t,n),{result:o,indexes:l}=I1(I("x",e,t,n),[s[1],s[2]],[a[1],a[2]],r,i);return[o,l]}case"AvgPool3D":{let a=I("strides",e,t,n),r=I("pad",e,t,n),s=I("kernelSize",e,t,n);return[Dy(I("x",e,t,n),[s[1],s[2],s[3]],[a[1],a[2],a[3]],r)]}case"MaxPool3D":{let a=I("strides",e,t,n),r=I("pad",e,t,n),s=I("kernelSize",e,t,n);return[qy(I("x",e,t,n),[s[1],s[2],s[3]],[a[1],a[2],a[3]],r)]}case"Dilation2D":{let a=I("strides",e,t,n),r=I("pad",e,t,n),s=I("dilations",e,t,n),i=a[1],o=a[2],l=s[1],c=s[2];return[Ly(I("x",e,t,n),I("filter",e,t,n),[i,o],r,[l,c],"NHWC")]}default:throw TypeError(`Node type ${e.op} is not implemented`)}},W4=(e,t,n)=>{switch(e.op){case"Fill":{let a=I("shape",e,t,n),r=I("dtype",e,t,n),s=I("value",e,t,n);return[wn(a,s,r)]}case"LinSpace":{let a=I("start",e,t,n),r=I("stop",e,t,n),s=I("num",e,t,n);return[g1(a,r,s)]}case"Multinomial":{let a=I("logits",e,t,n),r=I("numSamples",e,t,n),s=I("seed",e,t,n);return[N1(a,r,s)]}case"OneHot":{let a=I("indices",e,t,n),r=I("depth",e,t,n),s=I("onValue",e,t,n),i=I("offValue",e,t,n);return[$l(a,r,s,i)]}case"Ones":return[Xa(I("shape",e,t,n),I("dtype",e,t,n))];case"OnesLike":return[Dn(I("x",e,t,n))];case"RandomUniform":return[Wl(I("shape",e,t,n),I("minval",e,t,n),I("maxval",e,t,n),I("dtype",e,t,n))];case"Range":{let a=I("start",e,t,n),r=I("stop",e,t,n),s=I("step",e,t,n);return[kh(a,r,s,I("dtype",e,t,n))]}case"TruncatedNormal":{let a=I("shape",e,t,n),r=I("mean",e,t,n),s=I("stdDev",e,t,n),i=I("seed",e,t,n);return[Fh(a,r,s,I("dtype",e,t,n),i)]}case"Zeros":return[vt(I("shape",e,t,n),I("dtype",e,t,n))];case"ZerosLike":return[Ge(I("x",e,t,n))];default:throw TypeError(`Node type ${e.op} is not implemented`)}};function lv(e,t,n){let a=I("boxes",e,t,n),r=I("scores",e,t,n),s=I("maxOutputSize",e,t,n),i=I("iouThreshold",e,t,n),o=I("scoreThreshold",e,t,n),l=I("softNmsSigma",e,t,n);return{boxes:a,scores:r,maxOutputSize:s,iouThreshold:i,scoreThreshold:o,softNmsSigma:l}}var V4=async(e,t,n)=>{switch(e.op){case"NonMaxSuppressionV5":{let{boxes:a,scores:r,maxOutputSize:s,iouThreshold:i,scoreThreshold:o,softNmsSigma:l}=lv(e,t,n),c=await Ya.nonMaxSuppressionWithScoreAsync(a,r,s,i,o,l);return[c.selectedIndices,c.selectedScores]}case"NonMaxSuppressionV4":{let{boxes:a,scores:r,maxOutputSize:s,iouThreshold:i,scoreThreshold:o}=lv(e,t,n),l=I("padToMaxOutputSize",e,t,n),c=await Ya.nonMaxSuppressionPaddedAsync(a,r,s,i,o,l);return[c.selectedIndices,c.validOutputs]}case"NonMaxSuppressionV3":case"NonMaxSuppressionV2":{let{boxes:a,scores:r,maxOutputSize:s,iouThreshold:i,scoreThreshold:o}=lv(e,t,n);return[await Ya.nonMaxSuppressionAsync(a,r,s,i,o)]}case"Where":{let a=ce(I("condition",e,t,n),"bool"),r=[await ib(a)];return a.dispose(),r}case"ListDiff":return C1(I("x",e,t,n),I("y",e,t,n));default:throw TypeError(`Node type ${e.op} is not implemented`)}},U4=(e,t,n)=>{switch(e.op){case"TopKV2":{let a=I("x",e,t,n),r=I("k",e,t,n),s=I("sorted",e,t,n),i=rb(a,r,s);return[i.values,i.indices]}case"Unique":{let a=I("x",e,t,n),r=$h(a);return[r.values,r.indices]}case"UniqueV2":{let a=I("x",e,t,n),r=I("axis",e,t,n),s=$h(a,r);return[s.values,s.indices]}default:throw TypeError(`Node type ${e.op} is not implemented`)}},G4=(e,t,n)=>{switch(e.op){case"Const":return t[e.name];case"PlaceholderWithDefault":let a=I("default",e,t,n);return[Nn(e.name,t,n)||a];case"Placeholder":return[Nn(e.name,t,n)];case"Identity":case"StopGradient":case"FakeQuantWithMinMaxVars":{let c=I("x",e,t,n);return[Ir(c)]}case"IdentityN":return I("x",e,t,n).map(c=>Ir(c));case"Snapshot":let r=I("x",e,t,n);return[Ir(r)];case"Shape":return[tt(I("x",e,t,n).shape,"int32")];case"ShapeN":return I("x",e,t,n).map(c=>tt(c.shape));case"Size":return[de(I("x",e,t,n).size,"int32")];case"Rank":return[de(I("x",e,t,n).rank,"int32")];case"NoOp":return[de(1)];case"Print":let s=I("x",e,t,n),i=I("data",e,t,n),o=I("message",e,t,n),l=I("summarize",e,t,n);console.warn("The graph has a tf.print() operation,usually used for debugging, which slows down performance."),console.log(o);for(let c=0;c<i.length;c++)console.log(Array.prototype.slice.call(i[c].dataSync()).slice(0,l));return[s];default:throw TypeError(`Node type ${e.op} is not implemented`)}},H4=class{constructor(e,t){this.keyDType=e,this.valueDType=t,this.handle=de(0),this.tensorMap=new Map,jt(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}async import(e,t){this.checkKeyAndValueTensor(e,t);let n=await e.data();return this.tensorMap.forEach(a=>a.dispose()),this.tensorMap.clear(),D(()=>{let a=dt(t),r=n.length,s=a.length;k.assert(r===s,()=>`The number of elements doesn't match, keys has ${r} elements, the values has ${s} elements.`);for(let i=0;i<r;i++){let o=n[i],l=a[i];jt(l),this.tensorMap.set(o,l)}return this.handle})}async find(e,t){this.checkKeyAndValueTensor(e,t);let n=await e.data();return D(()=>{let a=[];for(let r=0;r<n.length;r++){let s=n[r],i=this.findWithDefault(s,t);a.push(i)}return $t(a)})}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}`)}},j4=async(e,t,n,a)=>{switch(e.op){case"HashTable":case"HashTableV2":{let r=I("keyDType",e,t,n),s=I("valueDType",e,t,n),i=new H4(r,s);return a.addHashTable(e.name,i),[i.handle]}case"LookupTableImport":case"LookupTableImportV2":{let r=I("tableHandle",e,t,n,a),s=I("keys",e,t,n),i=I("values",e,t,n);return[await a.getHashTableById(r.id).import(s,i)]}case"LookupTableFind":case"LookupTableFindV2":{let r=I("tableHandle",e,t,n,a),s=I("keys",e,t,n),i=I("defaultValue",e,t,n);return[await a.getHashTableById(r.id).find(s,i)]}default:throw TypeError(`Node type ${e.op} is not implemented`)}},q4=(e,t,n)=>{switch(e.op){case"ResizeBilinear":{let a=I("images",e,t,n),r=I("size",e,t,n),s=I("alignCorners",e,t,n),i=I("halfPixelCenters",e,t,n);return[Ya.resizeBilinear(a,[r[0],r[1]],s,i)]}case"ResizeNearestNeighbor":{let a=I("images",e,t,n),r=I("size",e,t,n),s=I("alignCorners",e,t,n),i=I("halfPixelCenters",e,t,n);return[Ya.resizeNearestNeighbor(a,[r[0],r[1]],s,i)]}case"CropAndResize":{let a=I("image",e,t,n),r=I("boxes",e,t,n),s=I("boxInd",e,t,n),i=I("cropSize",e,t,n),o=I("method",e,t,n),l=I("extrapolationValue",e,t,n);return[Ya.cropAndResize(a,r,s,i,o,l)]}default:throw TypeError(`Node type ${e.op} is not implemented`)}},K4=(e,t,n)=>{switch(e.op){case"Equal":return[Zr(I("a",e,t,n),I("b",e,t,n))];case"NotEqual":return[_i(I("a",e,t,n),I("b",e,t,n))];case"Greater":return[ca(I("a",e,t,n),I("b",e,t,n))];case"GreaterEqual":return[Jr(I("a",e,t,n),I("b",e,t,n))];case"Less":return[fh(I("a",e,t,n),I("b",e,t,n))];case"LessEqual":return[Si(I("a",e,t,n),I("b",e,t,n))];case"LogicalAnd":return[pa(I("a",e,t,n),I("b",e,t,n))];case"LogicalNot":return[Wc(I("a",e,t,n))];case"LogicalOr":return[xh(I("a",e,t,n),I("b",e,t,n))];case"Select":case"SelectV2":return[vn(I("condition",e,t,n),I("a",e,t,n),I("b",e,t,n))];default:throw TypeError(`Node type ${e.op} is not implemented`)}},X4=(e,t,n)=>{switch(e.op){case"BatchMatMul":case"BatchMatMulV2":case"MatMul":return[ze(I("a",e,t,n),I("b",e,t,n),I("transposeA",e,t,n),I("transposeB",e,t,n))];case"Transpose":return[Ue(I("x",e,t,n),I("perm",e,t,n))];case"_FusedMatMul":let[a,r]=I("fusedOps",e,t,n),s=a==="biasadd",i=r==="prelu",o=I("numArgs",e,t,n),l=I("leakyreluAlpha",e,t,n);if(s){if(i&&o!==2)throw new Error("Fused MatMul with BiasAdd and Prelu must have two extra arguments: bias and alpha.");if(!i&&o!==1)throw new Error("Fused MatMul with BiasAdd must have one extra argument: bias.")}let[c,u]=I("args",e,t,n);return[es.matMul({a:I("a",e,t,n),b:I("b",e,t,n),transposeA:I("transposeA",e,t,n),transposeB:I("transposeB",e,t,n),bias:c,activation:r,preluActivationWeights:u,leakyreluAlpha:l})];default:throw TypeError(`Node type ${e.op} is not implemented`)}},Y4=(e,t,n)=>{switch(e.op){case"FusedBatchNorm":case"FusedBatchNormV2":return[fr(I("x",e,t,n),I("mean",e,t,n),I("variance",e,t,n),I("offset",e,t,n),I("scale",e,t,n),I("epsilon",e,t,n))];case"FusedBatchNormV3":return[fr(I("x",e,t,n),I("mean",e,t,n),I("variance",e,t,n),I("offset",e,t,n),I("scale",e,t,n),I("epsilon",e,t,n))];case"LRN":return[Uy(I("x",e,t,n),I("radius",e,t,n),I("bias",e,t,n),I("alpha",e,t,n),I("beta",e,t,n))];case"Softmax":return[Na(I("x",e,t,n))];case"LogSoftmax":return[bh(I("x",e,t,n))];case"SparseToDense":return[ob(I("sparseIndices",e,t,n),I("outputShape",e,t,n),I("sparseValues",e,t,n),I("defaultValue",e,t,n))];default:throw TypeError(`Node type ${e.op} is not implemented`)}},Z4=(e,t,n)=>{switch(e.op){case"Max":{let i=I("axis",e,t,n),o=I("keepDims",e,t,n);return[Yn(I("x",e,t,n),i,o)]}case"Mean":{let i=I("axis",e,t,n),o=I("keepDims",e,t,n);return[St(I("x",e,t,n),i,o)]}case"Min":{let i=I("axis",e,t,n),o=I("keepDims",e,t,n);return[zl(I("x",e,t,n),i,o)]}case"Sum":{let i=I("axis",e,t,n),o=I("keepDims",e,t,n);return[Te(I("x",e,t,n),i,o)]}case"All":{let i=I("axis",e,t,n),o=I("keepDims",e,t,n);return[lh(I("x",e,t,n),i,o)]}case"Any":{let i=I("axis",e,t,n),o=I("keepDims",e,t,n);return[Rc(I("x",e,t,n),i,o)]}case"ArgMax":{let i=I("axis",e,t,n);return[Mc(I("x",e,t,n),i)]}case"ArgMin":{let i=I("axis",e,t,n);return[Ty(I("x",e,t,n),i)]}case"Prod":{let i=I("axis",e,t,n),o=I("keepDims",e,t,n);return[wh(I("x",e,t,n),i,o)]}case"Cumsum":{let i=I("axis",e,t,n),o=I("exclusive",e,t,n),l=I("reverse",e,t,n);return[hh(I("x",e,t,n),i,o,l)]}case"Bincount":let a=I("x",e,t,n),r=I("weights",e,t,n),s=I("size",e,t,n);return[s1(a,r,s)];case"DenseBincount":{let i=I("x",e,t,n),o=I("weights",e,t,n),l=I("size",e,t,n),c=I("binaryOutput",e,t,n);return[p1(i,o,l,c)]}default:throw TypeError(`Node type ${e.op} is not implemented`)}},J4=(e,t,n)=>{switch(e.op){case"ConcatV2":case"Concat":{let a=I("n",e,t,n),r=I("axis",e,t,n),s=I("tensors",e,t,n);return s=s.slice(0,a),[Qe(s,r)]}case"Gather":{let a=I("x",e,t,n),r=I("indices",e,t,n);return[Ti(a,ce(r,"int32"),0)]}case"GatherV2":{let a=I("axis",e,t,n),r=I("batchDims",e,t,n),s=I("x",e,t,n),i=I("indices",e,t,n);return[Ti(s,ce(i,"int32"),a,r)]}case"Reverse":{let a=I("dims",e,t,n),r=[];for(let i=0;i<a.length;i++)a[i]&&r.push(i);let s=I("x",e,t,n);return[Rn(s,r)]}case"ReverseV2":{let a=I("axis",e,t,n),r=I("x",e,t,n);return[Rn(r,a)]}case"Slice":{let a=I("begin",e,t,n),r=I("size",e,t,n);return[We(I("x",e,t,n),a,r)]}case"StridedSlice":{let a=I("begin",e,t,n),r=I("end",e,t,n),s=I("strides",e,t,n),i=I("beginMask",e,t,n),o=I("endMask",e,t,n),l=I("ellipsisMask",e,t,n),c=I("newAxisMask",e,t,n),u=I("shrinkAxisMask",e,t,n),p=I("x",e,t,n);return[nb(p,a,r,s,i,o,l,c,u)]}case"Pack":return D(()=>{let a=I("axis",e,t,n),r=I("tensors",e,t,n),s=r[0].shape,i=Qr(r[0]).shape,o=r.map(l=>{let c=k.arraysEqual(l.shape,s);if(!c&&!k.arraysEqual(Qr(l).shape,i))throw new Error("the input tensors shape does not match");return c?l:j(l,s)});return[$t(o,a)]});case"Unpack":{let a=I("axis",e,t,n),r=I("tensor",e,t,n);return dt(r,a)}case"Tile":{let a=I("reps",e,t,n);return[ja(I("x",e,t,n),a)]}case"Split":case"SplitV":{let a=I("axis",e,t,n),r=I("numOrSizeSplits",e,t,n),s=I("x",e,t,n);return Mn(s,r,a)}case"ScatterNd":{let a=I("indices",e,t,n),r=I("values",e,t,n),s=I("shape",e,t,n);return[F1(a,r,s)]}case"GatherNd":{let a=I("x",e,t,n),r=I("indices",e,t,n);return[$1(a,r)]}case"SparseToDense":{let a=I("sparseIndices",e,t,n),r=I("outputShape",e,t,n),s=I("sparseValues",e,t,n),i=I("defaultValue",e,t,n);return[ob(a,s,r,s.dtype===i.dtype?i:ce(i,s.dtype))]}default:throw TypeError(`Node type ${e.op} is not implemented`)}},Q4=(e,t,n)=>{switch(e.op){case"FFT":return[jc(I("x",e,t,n))];case"IFFT":return[Ul(I("x",e,t,n))];case"RFFT":return[qc(I("x",e,t,n))];case"IRFFT":return[Eh(I("x",e,t,n))];default:throw TypeError(`Node type ${e.op} is not implemented`)}},eV=(e,t,n)=>{switch(e.op){case"Cast":return[ce(I("x",e,t,n),I("dtype",e,t,n))];case"ExpandDims":{let a=I("axis",e,t,n);return[Fn(I("x",e,t,n),a)]}case"Squeeze":{let a=I("axis",e,t,n);return[Qr(I("x",e,t,n),a)]}case"Reshape":return[j(I("x",e,t,n),I("shape",e,t,n))];case"MirrorPad":return[Ky(I("x",e,t,n),I("padding",e,t,n),I("mode",e,t,n))];case"PadV2":case"Pad":return[Zn(I("x",e,t,n),I("padding",e,t,n),I("constantValue",e,t,n))];case"SpaceToBatchND":{let a=I("blockShape",e,t,n),r=I("paddings",e,t,n);return[Vc(I("x",e,t,n),a,r)]}case"BatchToSpaceND":{let a=I("blockShape",e,t,n),r=I("crops",e,t,n);return[Oc(I("x",e,t,n),a,r)]}case"DepthToSpace":{let a=I("blockSize",e,t,n),r=I("dataFormat",e,t,n).toUpperCase();return[Oy(I("x",e,t,n),a,r)]}case"BroadcastTo":return[Lc(I("x",e,t,n),I("shape",e,t,n))];default:throw TypeError(`Node type ${e.op} is not implemented`)}};function pN(e,t,n,a){let r=((s,i,o)=>{switch(s.category){case"arithmetic":return D(()=>F4(s,i,o));case"basic_math":return D(()=>$4(s,i,o));case"control":return z4(s,i,o);case"convolution":return D(()=>B4(s,i,o));case"creation":return D(()=>W4(s,i,o));case"dynamic":return V4(s,i,o);case"evaluation":return D(()=>U4(s,i,o));case"image":return D(()=>q4(s,i,o));case"graph":return D(()=>G4(s,i,o));case"logical":return D(()=>K4(s,i,o));case"matrices":return D(()=>X4(s,i,o));case"normalization":return D(()=>Y4(s,i,o));case"reduction":return D(()=>Z4(s,i,o));case"slice_join":return D(()=>J4(s,i,o));case"spectral":return D(()=>Q4(s,i,o));case"transformation":return D(()=>eV(s,i,o));case"hash_table":return j4(s,i,o,a);case"custom":let l=VI(s.op);if(l&&l.customExecutor)return l.customExecutor(new A4(s,i,o));throw TypeError(`Custom op ${s.op} is not registered.`);default:throw TypeError(`Unknown op '${s.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 k.isPromise(r)?r.then(s=>[].concat(s)):[].concat(r)}var dN=class{constructor(e={},t={},n={},a={}){this.weightMap=e,this.tensorArrayMap=t,this.tensorListMap=n,this.functionMap=a,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 mN(e,t,n,a){let r=new Set,s=[],i=null,o=null,l=new Set,c=Object.keys(e).map(d=>On(d)[0]),u=[];a!=null&&(u=a.map(d=>On(d.name)[0]));let p=[...t];for(;p.length>0;){let d=p.pop();if((hN(d)||tV(d)||nV(d))&&i==null&&(i=d,o=i.children.map(h=>h.name).filter(h=>r.has(h))),r.add(d.name),n[d.name]==null&&c.indexOf(d.name)===-1&&u.indexOf(d.name)===-1){if(d.inputs.length===0){s.push(d.name);continue}d.inputs.forEach(h=>{l.has(h.name)||(l.add(h.name),p.push(h))})}}return{inputs:e,outputs:t,usedNodes:r,missingInputs:s,dynamicNode:i,syncInputs:o}}function aV(e,t,n){let{usedNodes:a,inputs:r}=n,s=[],i=Object.keys(r).map(u=>On(u)[0]).map(u=>e.nodes[u]),o=e.initNodes;i.forEach(u=>{a.has(u.name)&&s.push(u)}),e.weights.forEach(u=>{a.has(u.name)&&s.push(u)}),o!=null&&o.forEach(u=>{a.has(u.name)&&s.push(u)});let l=new Set,c=[];for(;s.length>0;){let u=s.pop();l.add(u.name),t[u.name]||c.push(u),u.children.forEach(p=>{!l.has(p.name)&&a.has(p.name)&&p.inputs.every(d=>l.has(d.name))&&s.push(p)})}return c}var rV=["Switch","Merge","Enter","Exit","NextIteration","StatelessIf","StatelessWhile","if","While"],sV=["NonMaxSuppressionV2","NonMaxSuppressionV3","NonMaxSuppressionV5","Where"],iV=["HashTable","HashTableV2","LookupTableImport","LookupTableImportV2","LookupTableFind","LookupTableFindV2"];function hN(e){return rV.indexOf(e.op)>=0}function tV(e){return sV.indexOf(e.op)>=0}function nV(e){return iV.indexOf(e.op)>=0}var uv=class{constructor(e,t){this.graph=e,this.parent=t,this.compiledMap=new Map,this._weightMap={},this.SEPERATOR=",",this._functions={},this._functionExecutorMap={},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 uv(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(a=>a.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(),a=t.map(r=>r.name).sort();return n.join(this.SEPERATOR)+"--"+a.join(this.SEPERATOR)}compile(e,t){let n=mN(e,t,this.weightMap,this._initNodes),{missingInputs:a,dynamicNode:r,syncInputs:s}=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 [${s}]`);if(a.length>0){let i=t.map(l=>l.name),o=Object.keys(e);throw new Error(`Cannot compute the outputs [${i}] from the provided inputs [${o}]. Missing the following inputs: [${a}]`)}return aV(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 a=n.map(u=>this.graph.nodes[On(u)[0]]),r=t.map(u=>On(u)[0]),s=r.map(u=>this.graph.nodes[u]);s.length===0&&(s=this._outputs);let i=this.getCompilationKey(a,s),o=this.compiledMap.get(i);o==null&&(o=this.compile(e,s),this.compiledMap.set(i,o));let l={},c={};return D(()=>{let u=new dN(this.weightMap,l,c,this.functionExecutorMap),p=Object.assign({},this.weightMap);Object.keys(e).forEach(m=>{let[f,g]=On(m),y=[];y[g]=e[m],p[f]=y});let d=this.getFrozenTensorIds(p),h={};for(let m=0;m<o.length;m++){let f=o[m];if(!p[f.name]){let g=pN(f,p,u,this._resourceManager);if(k.isPromise(g))throw new Error(`The execution of the op '${f.op}' returned a promise. Please use model.executeAsync() instead.`);p[f.name]=g,this.checkTensorForDisposal(f.name,f,p,u,d,r,h)}}return this.parent==null&&u.dispose(d),t.map(m=>Nn(m,p,u))})}getFrozenTensorIds(e){let t=[].concat.apply([],Object.keys(e).map(n=>e[n]).map(n=>n.map(a=>a.id)));return new Set(t)}checkTensorForDisposal(e,t,n,a,r,s,i){t.category==="control"||s.indexOf(e)!==-1||(n[e].forEach(o=>{o!=null&&(i[o.id]=(i[o.id]||0)+t.children.length)}),t.inputs.forEach(o=>{if(o.category!=="control"){let l=p4(o.name,n,a);l!=null&&l.forEach(c=>{if(c&&!r.has(c.id)){let u=i[c.id];u===1?(c.dispose(),delete i[c.id]):u!=null&&i[c.id]--}})}}))}async executeAsync(e,t){return this._executeAsync(e,t)}async _executeAsync(e,t,n=!1,a={},r={}){n||(e=this.mapInputs(e),this.checkInputs(e),this.checkInputShapeAndType(e),t=this.mapOutputs(t),this.checkOutputs(t));let s=new dN(this.weightMap,a,r,this.functionExecutorMap),i=await this.executeWithControlFlow(e,s,t,n),o=t.map(p=>Nn(p,i,s)),l=o.map(p=>p.id),c=Object.keys(e).map(p=>e[p].id),u=new Set([...l,...c,...this.weightIds]);return Object.keys(i).forEach(p=>{i[p].forEach(d=>{d&&!d.isDisposed&&!u.has(d.id)&&d.dispose()})}),this.parent==null&&s.dispose(u),o}async executeFunctionAsync(e,t,n){let a=e.reduce((r,s,i)=>(r[this.inputs[i].name]=s,r),{});return this._executeAsync(a,this.outputNodes,!0,t,n)}async executeWithControlFlow(e,t,n,a){let r=Object.keys(e),s=r.map(b=>this.graph.nodes[On(b)[0]]),i=n.map(b=>On(b)[0]),o=i.map(b=>this.graph.nodes[b]);o.length===0&&(o=this._outputs);let{usedNodes:l,missingInputs:c,dynamicNode:u,syncInputs:p}=mN(e,o,this.weightMap,this._initNodes),d=[...s,...this.graph.weights,...this._initNodes||[]].map(b=>({node:b,contexts:t.currentContext})),h=Object.assign({},this.weightMap);Object.keys(e).forEach(b=>{let[v,x]=On(b),N=[];N[x]=e[b],h[v]=N});let m={},f=this.getFrozenTensorIds(h),g={};for(;d.length>0;){let b=this.processStack(s,d,t,h,g,f,i,m,l);await Promise.all(b)}u==null&&!a&&console.warn("This model execution did not contain any nodes with control flow or dynamic output shapes. You can use model.execute() instead.");let y=o.filter(b=>!hN(b)&&!Nn(b.name,h,t)).map(b=>b.name);if(y.length>0){let b="";throw u!=null&&(b=`Alternatively, to avoid the dynamic ops, use model.execute() and specify the inputs [${p}]`),new Error(`Cannot compute the outputs [${y}] from the provided inputs [${r}]. Consider providing the following inputs: [${c}]. ${b}`)}return h}processStack(e,t,n,a,r,s,i,o,l){let c=[];for(;t.length>0;){let u=t.pop();n.currentContext=u.contexts;let p="";if(u.node.op==="Enter"&&I("isConstant",u.node,a,n)&&([p]=kr(u.node.name,n)),a[u.node.name]==null){let d=pN(u.node,a,n,this._resourceManager);p||([p]=kr(u.node.name,n));let h=n.currentContext;k.isPromise(d)?c.push(d.then(m=>(a[p]=m,n.currentContext=h,this.checkTensorForDisposal(p,u.node,a,n,s,i,o),this.processChildNodes(u.node,t,n,a,r,l),m))):(a[p]=d,this.checkTensorForDisposal(p,u.node,a,n,s,i,o),this.processChildNodes(u.node,t,n,a,r,l))}else this.processChildNodes(u.node,t,n,a,r,l)}return c}processChildNodes(e,t,n,a,r,s){e.children.forEach(i=>{let[o]=kr(i.name,n);r[o]||!s.has(i.name)||(i.op==="Merge"?i.inputNames.some(l=>!!Nn(l,a,n))&&(r[o]=!0,t.push({contexts:n.currentContext,node:i})):i.inputNames.every(l=>!!Nn(l,a,n))&&(r[o]=!0,t.push({contexts:n.currentContext,node:i})))})}dispose(){Object.keys(this.weightMap).forEach(e=>this.weightMap[e].forEach(t=>t.dispose()))}checkInputShapeAndType(e){Object.keys(e).forEach(t=>{let n=e[t],[a]=On(t),r=this.graph.nodes[a];if(r.attrParams.shape&&r.attrParams.shape.value){let s=r.attrParams.shape.value,i=s.length===n.shape.length&&n.shape.every((o,l)=>s[l]===-1||s[l]===o);k.assert(i,()=>`The shape of dict['${r.name}'] provided in model.execute(dict) must be [${s}], but was [${n.shape}]`)}r.attrParams.dtype&&r.attrParams.dtype.value&&k.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 ${n.dtype}`)})}mapInputs(e){let t={};for(let n in e)if(this._signature!=null&&this._signature.inputs!=null&&this._signature.inputs[n]!=null){let a=this._signature.inputs[n];t[a.name]=e[n]}else t[n]=e[n];return t}checkInputs(e){let t=Object.keys(e).filter(n=>{let[a]=On(n);return this.graph.nodes[a]==null});if(t.length>0)throw new Error(`The dict provided in model.execute(dict) has keys: [${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]=On(t);if(!this.graph.nodes[n])throw new Error(`The output '${t}' is not found in the graph`)})}},oV=class{constructor(e={},t={}){this.hashTableNameToHandle=e,this.hashTableMap=t}addHashTable(e,t){this.hashTableNameToHandle[e]=t.handle,this.hashTableMap[t.id]=t}getHashTableHandleByName(e){return this.hashTableNameToHandle[e]}getHashTableById(e){return this.hashTableMap[e]}dispose(){for(let e in this.hashTableMap)this.hashTableMap[e].clearAndClose(),delete this.hashTableMap[e];for(let e in this.hashTableNameToHandle)this.hashTableNameToHandle[e].dispose(),delete this.hashTableNameToHandle[e]}},lV="?tfjs-format=file",uV="model.json",fN=class{constructor(e,t={}){this.modelUrl=e,this.loadOptions=t,this.version="n/a",t==null&&(this.loadOptions={}),this.resourceManager=new oV}get modelVersion(){return this.version}get inputNodes(){return this.executor.inputNodes}get outputNodes(){return this.executor.outputNodes}get inputs(){return this.executor.inputs}get outputs(){return this.executor.outputs}get weights(){return this.executor.weightMap}get metadata(){return this.artifacts.userDefinedMetadata}get modelSignature(){return this.signature}findIOHandler(){let e=this.modelUrl;if(e.load!=null)this.handler=e;else if(this.loadOptions.requestInit!=null)this.handler=Ht.browserHTTPRequest(e,this.loadOptions);else{let t=Ht.getLoadHandlers(e,this.loadOptions);if(t.length===0)t.push(Ht.browserHTTPRequest(e,this.loadOptions));else if(t.length>1)throw new Error(`Found more than one (${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}.${t.versions.minConsumer}`;let a=Ht.decodeWeights(this.artifacts.weightData,this.artifacts.weightSpecs);if(this.executor=new uv(oN.Instance.transformGraph(t,this.signature)),this.executor.weightMap=this.convertTensorMapToTensorsMap(a),this.executor.resourceManager=this.resourceManager,e.modelInitializer!=null&&e.modelInitializer.node!=null){let r=oN.Instance.transformGraph(e.modelInitializer);this.initializer=new uv(r),this.initializer.weightMap=this.executor.weightMap,this.initializer.resourceManager=this.resourceManager,this.initializer.executeAsync({},[])}return!0}async save(e,t){if(typeof e=="string"){let n=Ht.getSaveHandlers(e);if(n.length===0)throw new Error(`Cannot find any save handlers for URL '${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 Ae)&&!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 ${e.length} input tensors.`);return this.inputNodes.reduce((t,n,a)=>(t[n]=e[a],t),{})}normalizeOutputs(e){return e=e||this.outputNodes,Array.isArray(e)?e:[e]}execute(e,t){e=this.normalizeInputs(e),t=this.normalizeOutputs(t);let n=this.executor.execute(e,t);return n.length>1?n:n[0]}async executeAsync(e,t){e=this.normalizeInputs(e),t=this.normalizeOutputs(t);let n=await this.executor.executeAsync(e,t);return n.length>1?n:n[0]}convertTensorMapToTensorsMap(e){return Object.keys(e).reduce((t,n)=>(t[n]=[e[n]],t),{})}dispose(){this.executor.dispose(),this.initializer&&this.initializer.dispose(),this.resourceManager.dispose()}};async function cV(e,t={}){if(e==null)throw new Error("modelUrl in loadGraphModel() cannot be null. Please provide a url or an IOHandler that loads the model");t==null&&(t={}),t.fromTFHub&&e.load==null&&(e.endsWith("/")||(e=e+"/"),e=`${e}${uV}${lV}`);let n=new fN(e,t);return await n.load(),n}var gN="3.0.0",yN={};Oe(yN,{CSVDataset:()=>xN,Dataset:()=>Ql,FileDataSource:()=>vN,TextLineDataset:()=>bN,URLDataSource:()=>wN,array:()=>pV,csv:()=>hV,func:()=>mV,generator:()=>fV,microphone:()=>yV,version_data:()=>kN,webcam:()=>gV,zip:()=>dV});var bV=Io(dd()),xV=Io(dd());function vV(e,t){return Cm(e,t)}function Cm(e,t,n=new Map,a=new Set){if(e==null)return null;if(a.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(eu(e)){let s=Array.isArray(e)?[]:{};a.add(e);for(let i in e){let o=e[i],l=Cm(o,t,n,a);s[i]=l}return a.delete(e),s}else throw new Error(`Can't recurse into non-iterable type: ${e}`);else return n.set(e,r.value),r.value}function wV(e,t=NN){return IN(e,t)}function IN(e,t,n=new Set){let a=e[0];if(n.has(a))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(eu(a)){let s=Array.isArray(a)?[]:{};n.add(a);for(let i in a){let o=e.map(c=>c[i]),l=IN(o,t,n);s[i]=l}return n.delete(a),s}else throw new Error(`Can't recurse into non-iterable type: ${a}`);else return r.value}function NN(e){return e===null?null:eu(e[0])?{value:null,recurse:!0}:{value:e,recurse:!1}}async function TN(e,t){let n=new Map;Cm(e,t,n);for(let a of Array.from(n.keys())){let r=n.get(a);if(k.isPromise(r)){let s=await r;n.set(a,s)}}return Cm(e,t,n)}function eu(e){return e!=null&&!ArrayBuffer.isView(e)&&(Array.isArray(e)||typeof e=="object"&&!(e instanceof Ae))}function IV(e){return e==null||kV(e)||Array.isArray(e)||typeof e=="object"&&e instanceof Ae||k.isTypedArray(e)}function kV(e){return e===null||typeof e!="object"&&typeof e!="function"}function TV(e){return vV(e,NV)}function NV(e){return e instanceof Ae?{value:e.clone(),recurse:!1}:eu(e)?{value:null,recurse:!0}:{value:e,recurse:!1}}var SN=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}},cv=class extends SN{constructor(){super(cv.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 a=0;a<n;a++)t[a]=this.get(this.wrap(this.begin+a));this.data=t,this.capacity=e,this.doubledCapacity=2*this.capacity,this.begin=0,this.end=n}};cv.INITIAL_CAPACITY=32;function CN(e){return new SV(e)}function pv(e){return new CV(e)}function _V(e,t){return new _N(e,t)}function AV(e,t=ls.FAIL){return new EV(e,t)}var Yt=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 OV(this,e)}filter(e){return new MV(this,e)}map(e){return new PV(this,e)}mapAsync(e){return new EN(this,e)}serialMapAsync(e){return new EN(this,e).serial()}flatmap(e){return new LV(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 RV(this,e,t)}columnMajorBatch(e,t=!0,n=NN){return this.rowMajorBatch(e,t).map(a=>wV(a,n))}concatenate(e,t){return new _N(CN([this,e]),t)}take(e){return e<0||e==null?this:new DV(this,e)}skip(e){return e<0||e==null?this:new $V(this,e)}prefetch(e){return new AN(this,e)}shuffle(e,t){return new zV(this,e,t)}serial(){return new FV(this)}},SV=class extends Yt{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:TV(e),done:!1}}},CV=class extends Yt{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: ${e.message}`,e}}},FV=class extends Yt{constructor(e){super();this.upstream=e,this.lastRead=Promise.resolve({value:null,done:!1})}summary(){return`${this.upstream.summary()} -> Serial`}async next(){return this.lastRead=this.lastRead.then(()=>this.serialNext()),this.lastRead}async serialNext(){return this.upstream.next()}},$V=class extends Yt{constructor(e,t){super();this.upstream=e,this.maxCount=t,this.count=0,this.lastRead=Promise.resolve({value:null,done:!1})}summary(){return`${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;Fe(e.value)}return this.upstream.next()}},DV=class extends Yt{constructor(e,t){super();this.upstream=e,this.maxCount=t,this.count=0}summary(){return`${this.upstream.summary()} -> Take`}async next(){return this.count++>=this.maxCount?{value:null,done:!0}:this.upstream.next()}},RV=class extends Yt{constructor(e,t,n=!0){super();this.upstream=e,this.batchSize=t,this.enableSmallLastBatch=n,this.lastRead=Promise.resolve({value:null,done:!1})}summary(){return`${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}}},MV=class extends Yt{constructor(e,t){super();this.upstream=e,this.predicate=t,this.lastRead=Promise.resolve({value:null,done:!1})}summary(){return`${this.upstream.summary()} -> Filter`}async next(){return this.lastRead=this.lastRead.then(()=>this.serialNext()),this.lastRead}async serialNext(){for(;;){let e=await this.upstream.next();if(e.done||this.predicate(e.value))return e;Fe(e.value)}}},PV=class extends Yt{constructor(e,t){super();this.upstream=e,this.transform=t}summary(){return`${this.upstream.summary()} -> Map`}async next(){let e=await this.upstream.next();if(e.done)return{value:null,done:!0};let t=Ia.getTensorsInContainer(e.value),n=this.transform(e.value),a=Ia.getTensorsInContainer(n);for(let r of t)Ia.isTensorInList(r,a)||r.dispose();return{value:n,done:!1}}},OV=class extends Yt{constructor(e,t){super();this.upstream=e,this.handler=t,this.count=0,this.lastRead=Promise.resolve({value:null,done:!1})}summary(){return`${this.upstream.summary()} -> handleErrors`}async next(){return this.lastRead=this.lastRead.then(()=>this.serialNext()),this.lastRead}async serialNext(){for(;;)try{return await this.upstream.next()}catch(e){if(!this.handler(e))return{value:null,done:!0}}}},EN=class extends Yt{constructor(e,t){super();this.upstream=e,this.transform=t}summary(){return`${this.upstream.summary()} -> AsyncMap`}async next(){let e=await this.upstream.next();if(e.done)return{value:null,done:!0};let t=Ia.getTensorsInContainer(e.value),n=await this.transform(e.value),a=Ia.getTensorsInContainer(n);for(let r of t)Ia.isTensorInList(r,a)||r.dispose();return{value:n,done:!1}}},dv=class extends Yt{constructor(){super();this.outputQueue=new cv,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}}},LV=class extends dv{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=Ia.getTensorsInContainer(e.value),n=this.transform(e.value),a=Ia.getTensorsInContainer(n);this.outputQueue.pushAll(n);for(let r of t)Ia.isTensorInList(r,a)||r.dispose();return!0}},_N=class extends Yt{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}},ls;(function(e){e[e.FAIL=0]="FAIL",e[e.SHORTEST=1]="SHORTEST",e[e.LONGEST=2]="LONGEST"})(ls||(ls={}));var EV=class extends Yt{constructor(e,t=ls.FAIL){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 a(s){return s instanceof Yt?{value:s.next().then(i=>(t++,i.done&&n++,i.value)),recurse:!1}:{value:null,recurse:!0}}let r=await TN(this.iterators,a);if(t===n)return{value:null,done:!0};if(n>0)switch(this.mismatchMode){case ls.FAIL:throw new Error(`Zipped streams should have the same length. Mismatched at element ${this.count}.`);case ls.SHORTEST:return{value:null,done:!0};case ls.LONGEST:default:}return this.count++,{value:r,done:!1}}async next(){return this.currentPromise=this.nextState(this.currentPromise),this.currentPromise}},AN=class extends Yt{constructor(e,t){super();this.upstream=e,this.bufferSize=t,this.buffer=new SN(t)}summary(){return`${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()}},zV=class extends AN{constructor(e,t,n){super(e,t);this.upstream=e,this.windowSize=t,this.upstreamExhausted=!1,this.random=xV.alea(n||k.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}}},Ql=class{constructor(){this.size=null}batch(e,t=!0){let n=this;k.assert(e>0,()=>`batchSize needs to be positive, but it is
|
|
${e}`);let a;return this.size===Infinity||this.size==null?a=this.size:t?a=Math.ceil(this.size/e):a=Math.floor(this.size/e),Ln(async()=>(await n.iterator()).columnMajorBatch(e,t,BV),a)}concatenate(e){let t=this,n;return this.size===Infinity||e.size===Infinity?n=Infinity:this.size!=null&&e.size!=null?n=this.size+e.size:n=null,Ln(async()=>(await t.iterator()).concatenate(await e.iterator()),n)}filter(e){let t=this,n;return this.size===Infinity?n=Infinity:n=null,Ln(async()=>(await t.iterator()).filter(a=>D(()=>e(a))),n)}async forEachAsync(e){return(await this.iterator()).forEachAsync(e)}map(e){let t=this;return Ln(async()=>(await t.iterator()).map(n=>D(()=>e(n))),this.size)}mapAsync(e){let t=this;return Ln(async()=>(await t.iterator()).mapAsync(e),this.size)}prefetch(e){if(e==null)throw new RangeError("`Dataset.prefetch()` requires bufferSize to be specified.");let t=this;return Ln(async()=>(await t.iterator()).prefetch(e),this.size)}repeat(e){let t=this,n;return this.size!=null&&e>0?n=this.size*e:e===0?n=0:this.size!=null&&(e===void 0||e<0)?n=Infinity:n=null,Ln(async()=>{let a=pv(async()=>({value:await t.iterator(),done:!1}));return _V(a.take(e))},n)}skip(e){let t=this,n;return this.size!=null&&e>=0&&this.size>=e?n=this.size-e:this.size!=null&&(this.size<e||e===void 0||e<0)?n=0:n=null,Ln(async()=>(await t.iterator()).skip(e),n)}shuffle(e,t,n=!0){if(e==null||e<0)throw this.size==null?new RangeError("`Dataset.shuffle()` requires bufferSize to be specified."):new RangeError(`\`Dataset.shuffle()\` requires bufferSize to be specified. If your data fits in main memory (for regular JS objects), and/or GPU memory (for \`tf.Tensor\`s), consider setting bufferSize to the dataset size (${this.size} elements)`);let a=this,r=bV.alea(t||k.now().toString());return Ln(async()=>{let s=r.int32();return n&&(s+=r.int32()),(await a.iterator()).shuffle(e,s.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,Ln(async()=>(await t.iterator()).take(e),n)}async toArray(){if(this.size===Infinity)throw new Error("Can not convert infinite data stream to array.");return(await this.iterator()).toArray()}async toArrayForTest(){if(this.size===Infinity)throw new Error("Can not convert infinite data stream to array.");return(await this.iterator()).toArrayForTest()}};Ql.MAX_BUFFER_SIZE=1e4;function Ln(e,t=null){return new class extends Ql{constructor(){super(...arguments);this.size=t}async iterator(){return e()}}}function pV(e){return Ln(async()=>CN(e),e.length)}function dV(e){if(!eu(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 Ln(async()=>{let n=await TN(e,a=>{if(a instanceof Ql)return{value:a.iterator(),recurse:!1};if(eu(a))return{value:null,recurse:!0};throw new Error("Leaves of the structure passed to zip() must be Datasets, not primitives.")});return AV(n,ls.SHORTEST)},t)}function BV(e){if(e===null)return null;let t=e[0];return IV(t)?{value:WV(e),recurse:!1}:{value:null,recurse:!0}}function WV(e){if(e.length===0)throw new Error("Can't make a batch of zero elements.");return e[0]instanceof Ae?$t(e):qn(e)}var bN=class extends Ql{constructor(e){super();this.input=e}async iterator(){return(await this.input.iterator()).decodeUTF8().split(`
|
|
`).map(e=>(e.endsWith("\r")&&(e=e.slice(0,-1)),e))}},_m='"',mp=Symbol("out"),FN=Symbol("field"),Em=Symbol("quote"),hv=Symbol("quoteafterquote"),$N=Symbol("quoteinquote"),xN=class extends Ql{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 bN(e),t||(t={}),this.hasHeader=t.hasHeader!==!1,this.fullColumnNames=t.columnNames,this.columnConfigs=t.columnConfigs,this.configuredColumnsOnly=t.configuredColumnsOnly,t.delimWhitespace?(k.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&&k.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((a,r)=>(a[r]=a[r]+1||1,a),{}),n=Object.keys(t).filter(a=>t[a]>1);if(k.assert(n.length===0,()=>"Duplicate column names found: "+n.toString()),this.columnConfigs){for(let a of Object.keys(this.columnConfigs))if(this.fullColumnNames.indexOf(a)===-1)throw new Error('The key "'+a+'" provided in columnConfigs does not match any of the column names ('+this.fullColumnNames.toString()+").")}this.columnNamesValidated=!0}async maybeReadHeaderLine(){if(this.hasHeader){let e=await(await this.base.iterator()).next();if(e.done)throw new Error("No data was found for CSV parsing.");let t=e.value;return this.parseRow(t,!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={},a={};for(let r=0;r<this.fullColumnNames.length;r++){let s=this.fullColumnNames[r],i=this.columnConfigs?this.columnConfigs[s]:null;if(!(this.configuredColumnsOnly&&!i)){let o=t[r],l=null;if(o==="")if(i&&i.default!==void 0)l=i.default;else{if(i&&(i.required||i.isLabel))throw new Error(`Required column ${s} is empty in this line: ${e}`);l=void 0}else{let c=Number(o);if(isNaN(c))i&&i.dtype==="bool"?l=this.getBoolean(o):l=o;else if(!i||!i.dtype)l=c;else switch(i.dtype){case"float32":l=c;break;case"int32":l=Math.floor(c);break;case"bool":l=this.getBoolean(o);break;default:l=c}}i&&i.isLabel?a[s]=l:n[s]=l}}return Object.keys(a).length===0?n:{xs:n,ys:a}}getBoolean(e){return e==="1"||e.toLowerCase()==="true"?1:0}parseRow(e,t=!0){let n=[],a=0,r=e.length,s=mp;for(let i=0;i<r;i++)switch(s){case mp:switch(e.charAt(i)){case _m:a=i+1,s=Em;break;case this.delimiter:if(a=i+1,this.delimiter===" "&&this.delimWhitespace)break;n.push(""),s=mp;break;default:s=FN,a=i;break}break;case FN:switch(e.charAt(i)){case this.delimiter:n.push(e.substring(a,i)),s=mp,a=i+1;break;default:}break;case Em:switch(e.charAt(i)){case _m:s=hv;break;default:}break;case hv:switch(e.charAt(i)){case this.delimiter:n.push(e.substring(a,i-1)),s=mp,a=i+1;break;case _m:s=Em;break;default:s=$N;break}break;case $N:switch(e.charAt(i)){case _m:s=Em;break;default:}break;default:}if(s===hv?n.push(e.substring(a,r-1)):n.push(e.substring(a)),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}},DN=class extends Yt{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(ee().get("IS_NODE"))throw new Error("microphone API is only supported in browser environment.");let t=new DN(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 a=this.flattenQueue(n.freqDataQueue);e=this.getTensorFromAudioDataArray(a,[this.numFrames,this.columnTruncateLength,1])}if(this.includeWaveform){let a=this.flattenQueue(n.timeDataQueue);t=this.getTensorFromAudioDataArray(a,[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(a=>{let r=setInterval(()=>{this.includeSpectrogram&&(this.analyser.getFloatFrequencyData(this.freqData),this.freqData[0]===-Infinity&&a({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),a({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((a,r)=>n.set(a,r*t)),n}getTensorFromAudioDataArray(e,t){let n=new Float32Array(k.sizeFromShape(t));return n.set(e,n.length-e.length),qn(n,t)}},RN=class extends Yt{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=tt([0],"int32"),this.webcamConfig.centerCrop){let n=this.webcamConfig.resizeWidth*1/this.webcamVideoElement.width,a=this.webcamConfig.resizeHeight*1/this.webcamVideoElement.height,r=(1-n)/2,s=(1-a)/2,i=r+n,o=a+s;this.cropBox=Ta([s,r,o,i],[1,4])}else this.cropBox=Ta([0,0,1,1],[1,4])}summary(){return"webcam"}static async create(e,t={}){if(ee().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 RN(e,t);return await n.start(),n}async start(){this.webcamConfig.facingMode&&k.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=ki.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: ${t.message}`)}finally{e.dispose()}else return{value:e,done:!1}}needToResize(){return!!(this.webcamConfig.resizeWidth&&this.webcamConfig.resizeHeight&&(this.webcamVideoElement.width!==this.webcamConfig.resizeWidth||this.webcamVideoElement.height!==this.webcamConfig.resizeHeight))}cropAndResizeFrame(e){return D(()=>{let t=Fn(ce(e,"float32"),0),n;n=Ya.cropAndResize(t,this.cropBox,this.cropBoxInd,this.cropSize,"bilinear");let a=n.shape;return j(n,a.slice(1))})}async capture(){return(await this.next()).value}stop(){this.stream.getTracks().forEach(e=>e.stop());try{this.webcamVideoElement.srcObject=null}catch(e){console.log(e),this.webcamVideoElement.src=null}this.isClosed=!0}toArray(){throw new Error("Can not convert infinite video stream to array.")}},MN=class{},PN=class extends Yt{split(e){return new VV(this,e)}},VV=class extends PN{constructor(e,t){super();this.upstream=e,this.impl=new UV(e,t)}summary(){return this.impl.summary()}async next(){return this.impl.next()}},UV=class extends dv{constructor(e,t){super();this.upstream=e,this.separator=t,this.carryover=""}summary(){return`${this.upstream.summary()} -> Split('${this.separator}')`}async pump(){let e=await this.upstream.next();if(e.done)return this.carryover===""?!1:(this.outputQueue.push(this.carryover),this.carryover="",!0);let t=e.value.split(this.separator);t[0]=this.carryover+t[0];for(let n of t.slice(0,-1))this.outputQueue.push(n);return this.carryover=t[t.length-1],!0}},HV=class extends Yt{decodeUTF8(){return new GV(this)}},GV=class extends PN{constructor(e){super();this.upstream=e,this.impl=new jV(e)}summary(){return this.impl.summary()}async next(){return this.impl.next()}},jV=class extends dv{constructor(e){super();if(this.upstream=e,ee().get("IS_BROWSER"))this.decoder=new TextDecoder("utf-8");else{let{StringDecoder:t}=mE();this.decoder=new t("utf8")}}summary(){return`${this.upstream.summary()} -> Utf8`}async pump(){let e=await this.upstream.next(),t;if(e.done)return!1;t=e.value;let n;return ee().get("IS_BROWSER")?n=this.decoder.decode(t,{stream:!0}):n=this.decoder.write(Buffer.from(t.buffer)),this.outputQueue.push(n),!0}},ON=class extends HV{constructor(e,t={}){super();this.file=e,this.options=t,k.assert(e instanceof Uint8Array||(ee().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 ${this.file}`}async next(){return this.offset>=(this.file instanceof Uint8Array?this.file.byteLength:this.file.size)?{value:null,done:!0}:{value:await new Promise((e,t)=>{let n=this.offset+this.chunkSize;if(this.file instanceof Uint8Array)e(new Uint8Array(this.file.slice(this.offset,n)));else{let a=new FileReader;a.onload=s=>{let i=a.result;if(i instanceof ArrayBuffer&&(i=new Uint8Array(i)),!(i instanceof Uint8Array))return t(new TypeError("FileReader returned unknown type."));e(i)},a.onabort=s=>t(new Error("Aborted")),a.onerror=s=>t(new Error(s.type));let r=this.file.slice(this.offset,n);a.readAsArrayBuffer(r)}this.offset=n}),done:!1}}};async function KV(e,t={}){let n,a;typeof e=="string"?n=e:(n=e.url,a=qV(e));let r=await k.fetch(n,a);if(r.ok){let s=new Uint8Array(await r.arrayBuffer());return new ON(s,t)}else throw new Error(r.statusText)}var qV=e=>({method:e.method,headers:e.headers,body:e.body,mode:e.mode,credentials:e.credentials,cache:e.cache,redirect:e.redirect,referrer:e.referrer,integrity:e.integrity});function LN(e){return typeof e=="string"&&e.substr(0,7)==="file://"}var vN=class extends MN{constructor(e,t={}){super();this.input=e,this.options=t}async iterator(){if(LN(this.input)&&ee().get("IS_NODE")){let e=require("fs");this.input=e.readFileSync(this.input.substr(7))}return new ON(this.input,this.options)}},wN=class extends MN{constructor(e,t={}){super();this.url=e,this.fileOptions=t}async iterator(){return LN(this.url)?new vN(this.url,this.fileOptions).iterator():KV(this.url,this.fileOptions)}};function hV(e,t={}){return new xN(new wN(e),t)}function mV(e){let t=pv(e);return Ln(async()=>t)}function fV(e){return Ln(async()=>{let t=await e();return pv(()=>t.next())})}async function gV(e,t){return RN.create(e,t)}async function yV(e){return DN.create(e)}var kN="3.0.0";function we(e,t){Array.isArray(e)||(e=[e]),e.forEach(n=>{n!=null&&k.assert(n.dtype!=="complex64",()=>`${t} does not support complex64 tensors in the CPU backend.`)})}var XV=Za.whereImpl,YV=class extends sc{constructor(){super();this.blockSize=48,this.firstUse=!0,this.data=new hd(this,Kr())}write(e,t,n){this.firstUse&&(this.firstUse=!1,ee().get("IS_NODE")&&_.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 a={};return this.data.set(a,{values:e,dtype:n,refCount:1}),a}makeTensorInfo(e,t,n){let a;if(t==="string"&&n!=null&&n.length>0&&k.isString(n[0])){let r=n.map(s=>k.encodeString(s));a=this.write(r,e,t)}else a=this.write(n,e,t);return{dataId:a,shape:e,dtype:t}}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,a){this.data.set(e,{values:t,dtype:a,refCount:1})}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 a=this.readSync(n.real.dataId),r=this.readSync(n.imag.dataId);return _.mergeRealAndImagArrays(a,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(a=>k.decodeString(a))}catch(a){throw new Error("Failed to decode encoded string bytes into utf-8")}return Le(e.shape,e.dtype,n)}makeOutput(e,t,n){let a=this.write(e,t,n);return Kr().makeTensorFromDataId(a,t,n,this)}disposeData(e){if(this.data.has(e)){let{complexTensorInfos:t}=this.data.get(e);t!=null&&(this.disposeData(t.real.dataId),this.disposeData(t.imag.dataId)),this.data.delete(e)}}disposeIntermediateTensorInfo(e){let t=e.dataId;if(this.data.has(t)){let n=this.data.get(t);n.refCount--,n.refCount<1&&this.disposeData(t)}}async time(e){let t=k.now();return e(),{kernelMs:k.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){we([e],"where");let t=this.readSync(e.dataId);return XV(e.shape,t)}dispose(){}floatPrecision(){return 32}epsilon(){return super.epsilon()}},zN={};Oe(zN,{addImpl:()=>WN,bincountImpl:()=>mv,bincountReduceImpl:()=>VN,ceilImpl:()=>UN,concatImpl:()=>fv,expImpl:()=>GN,expm1Impl:()=>HN,floorImpl:()=>jN,gatherV2Impl:()=>qN,greaterImpl:()=>KN,lessImpl:()=>XN,linSpaceImpl:()=>YN,logImpl:()=>ZN,maxImpl:()=>JN,maximumImpl:()=>QN,minimumImpl:()=>eT,multiplyImpl:()=>gv,negImpl:()=>tT,notEqualImpl:()=>nT,prodImpl:()=>aT,rangeImpl:()=>bv,rsqrtImpl:()=>rT,simpleAbsImpl:()=>BN,sliceImpl:()=>Am,squaredDifferenceImpl:()=>sT,stridedSliceImpl:()=>iT,subImpl:()=>oT,tileImpl:()=>lT,topKImpl:()=>uT,transposeImpl:()=>yv,uniqueImpl:()=>cT});function BN(e){let t=new Float32Array(e.length);for(let n=0;n<e.length;++n)t[n]=Math.abs(e[n]);return t}var ZV=e=>{let{x:t}=e.inputs,n=e.backend;we(t,"abs");let a=new Float32Array(k.sizeFromShape(t.shape)),r=n.data.get(t.dataId).values;return a=BN(r),n.makeOutput(a,t.shape,"float32")},JV={kernelName:So,backendName:"cpu",kernelFunc:ZV};function Rt(e){return(t,n,a,r,s)=>{let i=_.assertAndGetBroadcastShape(t,n),o=i.length,l=k.computeStrides(i),c=k.sizeFromShape(i),u=k.getTypedArrayFromDType(s,c),p=t.length,d=n.length,h=k.computeStrides(t),m=k.computeStrides(n),f=_.getBroadcastDims(t,i),g=_.getBroadcastDims(n,i);if(f.length+g.length===0)for(let y=0;y<u.length;++y)u[y]=e(a[y%a.length],r[y%r.length]);else for(let y=0;y<u.length;++y){let b=k.indexToLoc(y,o,l),v=b.slice(-p);f.forEach(C=>v[C]=0);let x=k.locToIndex(v,p,h),N=b.slice(-d);g.forEach(C=>N[C]=0);let T=k.locToIndex(N,d,m);u[y]=e(a[x],r[T])}return[u,i]}}function zn(e){let{inputs:t,backend:n}=e,{real:a,imag:r}=t,s=n.data.get(a.dataId).values,i=n.data.get(r.dataId).values,o=n.makeTensorInfo(a.shape,"complex64"),l=n.data.get(o.dataId);return l.complexTensorInfos={real:n.makeTensorInfo(a.shape,"float32",s),imag:n.makeTensorInfo(r.shape,"float32",i)},o}var QV={kernelName:kd,backendName:"cpu",kernelFunc:zn};function Fm(e,t,n="float32"){if(n==="complex64"){let r=Fm(e,t,"float32"),s=Fm(e,t,"float32");return zn({inputs:{real:r,imag:s},backend:e})}let a=k.makeZerosTypedArray(k.sizeFromShape(t),n);return e.makeTensorInfo(t,n,a)}function ar(e){let{inputs:t,backend:n}=e,{x:a}=t;return n.incRef(a.dataId),{dataId:a.dataId,shape:a.shape,dtype:a.dtype}}var eU={kernelName:Ko,backendName:"cpu",kernelFunc:ar};function Wi(e){let{inputs:t,backend:n}=e,{input:a}=t,r=n.data.get(a.dataId).complexTensorInfos.real,s=n.data.get(r.dataId).values;return n.makeTensorInfo(r.shape,r.dtype,s)}var tU={kernelName:Vd,backendName:"cpu",kernelFunc:Wi};function us(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{dtype:s}=a;if(s==="complex64"){if(r.dtype==="complex64")return ar({inputs:{x:r},backend:n});let i=Fm(n,r.shape,r.dtype),o=us({inputs:{x:r},backend:n,attrs:{dtype:"float32"}}),l=zn({inputs:{real:o,imag:i},backend:n});return n.disposeIntermediateTensorInfo(i),n.disposeIntermediateTensorInfo(o),l}if(r.dtype==="complex64"){let i=Wi({inputs:{input:r},backend:n}),o=us({inputs:{x:i},backend:n,attrs:{dtype:s}});return n.disposeIntermediateTensorInfo(i),o}if(!k.hasEncodingLoss(r.dtype,s)){let i=ar({inputs:{x:r},backend:n});return{dataId:i.dataId,shape:i.shape,dtype:s}}if(s==="int32"){let i=n.data.get(r.dataId).values,o=Int32Array.from(i);return n.makeTensorInfo(r.shape,"int32",o)}if(s==="bool"){let i=n.data.get(r.dataId).values,o=k.toTypedArray([0],r.dtype),[l,c]=Rt((u,p)=>u!==p?1:0)(r.shape,[],i,o,"bool");return n.makeTensorInfo(c,"bool",l)}throw new Error(`Error in Cast: failed to cast ${r.dtype} to ${s}`)}var nU={kernelName:As,backendName:"cpu",kernelFunc:us};function Zt(e,t,n,a){return n==null?({inputs:r,backend:s})=>{let{a:i,b:o}=r,l=s;we([i,o],e);let c=l.data.get(i.dataId).values,u=l.data.get(o.dataId).values,p=a||i.dtype,[d,h]=t(i.shape,o.shape,c,u,p);return l.makeTensorInfo(h,p,d)}:({inputs:r,backend:s})=>{let{a:i,b:o}=r,l=s;if(i.dtype==="complex64"||o.dtype==="complex64"){let c=us({inputs:{x:i},backend:l,attrs:{dtype:"complex64"}}),u=l.data.get(c.dataId),p=u.complexTensorInfos.real,d=u.complexTensorInfos.imag,h=l.data.get(p.dataId).values,m=l.data.get(d.dataId).values,f=us({inputs:{x:o},backend:l,attrs:{dtype:"complex64"}}),g=l.data.get(f.dataId),y=g.complexTensorInfos.real,b=g.complexTensorInfos.imag,v=l.data.get(y.dataId).values,x=l.data.get(b.dataId).values,[N,T,C]=n(i.shape,o.shape,h,m,v,x),$=l.makeTensorInfo(C,"float32",N),F=l.makeTensorInfo(C,"float32",T),O=zn({inputs:{real:$,imag:F},backend:l});return l.disposeIntermediateTensorInfo(c),l.disposeIntermediateTensorInfo(f),l.disposeIntermediateTensorInfo($),l.disposeIntermediateTensorInfo(F),O}else{let c=l.data.get(i.dataId).values,u=l.data.get(o.dataId).values,p=a||i.dtype,[d,h]=t(i.shape,o.shape,c,u,p);return l.makeTensorInfo(h,p,d)}}}function xv(e){return(t,n,a,r,s,i)=>{let o=_.assertAndGetBroadcastShape(t,n),l=k.sizeFromShape(o),c=o.length,u=k.computeStrides(o),p=k.getTypedArrayFromDType("float32",l),d=k.getTypedArrayFromDType("float32",l),h=_.getBroadcastDims(t,o),m=_.getBroadcastDims(n,o),f=_.mergeRealAndImagArrays(a,r),g=_.mergeRealAndImagArrays(s,i),y=t.length,b=k.computeStrides(t),v=n.length,x=k.computeStrides(n);if(h.length+m.length===0)for(let N=0;N<p.length;N++){let T=N%f.length,C=N%g.length,$=e(f[T*2],f[T*2+1],g[C*2],g[C*2+1]);p[N]=$.real,d[N]=$.imag}else for(let N=0;N<p.length;N++){let T=k.indexToLoc(N,c,u),C=T.slice(-y);h.forEach(W=>C[W]=0);let $=k.locToIndex(C,y,b),F=T.slice(-v);m.forEach(W=>F[W]=0);let O=k.locToIndex(F,v,x),B=e(f[$*2],f[$*2+1],g[O*2],g[O*2+1]);p[N]=B.real,d[N]=B.imag}return[p,d,o]}}var WN=Rt((e,t)=>e+t),aU=xv((e,t,n,a)=>({real:e+n,imag:t+a})),fp=Zt(Lr,WN,aU),rU={kernelName:Lr,backendName:"cpu",kernelFunc:fp};function mv(e,t,n,a,r){let s=k.sizeFromShape(a),i=k.makeZerosTypedArray(r,n);for(let o=0;o<e.length;o++){let l=e[o];if(l<0)throw new Error("Input x must be non-negative!");l>=r||(s>0?i[l]+=t[o]:i[l]+=1)}return i}function VN(e,t,n,a=!1){let r=e.shape[0],s=e.shape[1],i=Le([r,n],t.dtype);for(let o=0;o<r;o++)for(let l=0;l<s;l++){let c=e.get(o,l);if(c<0)throw new Error("Input x must be non-negative!");c>=n||(a?i.set(1,o,c):t.size>0?i.set(i.get(o,c)+t.get(o,l),o,c):i.set(i.get(o,c)+1,o,c))}return i}function tu(e){return(t,n,a)=>{let r=k.getTypedArrayFromDType(n,t.length);for(let s=0;s<t.length;++s)r[s]=e(t[s],a);return r}}function it(e,t,n){return({inputs:a,attrs:r,backend:s})=>{let{x:i}=a;if(we(i,e),i.dtype==="string"||n==="string")throw new Error("unaryKernelFunc does not support string input/output");let o=s,l=o.data.get(i.dataId).values,c=k.sizeFromShape(i.shape),u=n||i.dtype,p=k.getArrayFromDType(u,c);for(let d=0;d<c;++d)p[d]=t(l[d],r);return o.makeTensorInfo(i.shape,u,p)}}function nu(e,t,n){return({inputs:a,attrs:r,backend:s})=>{let{x:i}=a;if(we(i,e),i.dtype==="string"||n==="string")throw new Error("unaryKernelFunc does not support string input/output");let o=s,l=o.data.get(i.dataId).values,c=n||i.dtype,u=t(l,c,r);return o.makeTensorInfo(i.shape,c,u)}}var UN=tu(e=>Math.ceil(e)),sU=nu(Ro,UN),iU={kernelName:Ro,backendName:"cpu",kernelFunc:sU};function fv(e,t,n,a){let r=k.getArrayFromDType(n,k.sizeFromShape(t));if(a&&n!=="string"){let s=0;e.forEach(i=>{let o=k.sizeFromShape(i.shape);r.set(i.vals,s),s+=o})}else{let s=0;e.forEach(i=>{let o=n==="string"?_.fromUint8ToStringArray(i.vals):i.vals,l=0;for(let c=0;c<i.shape[0];++c){let u=c*t[1]+s;for(let p=0;p<i.shape[1];++p)r[u+p]=o[l++]}s+=i.shape[1]})}return r}var GN=tu(e=>Math.exp(e)),pT=nu(Os,GN),oU={kernelName:Os,backendName:"cpu",kernelFunc:pT},HN=tu(e=>Math.expm1(e)),lU=nu(Uo,HN),uU={kernelName:Uo,backendName:"cpu",kernelFunc:lU},jN=tu(e=>Math.floor(e)),cU=nu(Ls,jN),pU={kernelName:Ls,backendName:"cpu",kernelFunc:cU};function qN(e,t,n){let a=Le(n,e.dtype);for(let r=0;r<a.size;++r){let s=a.indexToLoc(r).slice(),i=s[0],o=s[2],l=t.locToIndex([i,o]);s[2]=t.values[l];let c=e.locToIndex(s);a.values[r]=e.values[c]}return a}var KN=Rt((e,t)=>e>t?1:0),dU=Zt(qo,KN,null,"bool"),hU={kernelName:qo,backendName:"cpu",kernelFunc:dU},XN=Rt((e,t)=>e<t?1:0),mU=Zt(Jo,XN,null,"bool"),fU={kernelName:Jo,backendName:"cpu",kernelFunc:mU};function YN(e,t,n){let a=(t-e)/(n-1),r=k.makeZerosTypedArray(n,"float32");r[0]=e;for(let s=1;s<r.length;s++)r[s]=r[s-1]+a;return r}var ZN=tu(e=>Math.log(e)),gU=nu(Us,ZN),yU={kernelName:Us,backendName:"cpu",kernelFunc:gU};function JN(e,t,n,a){let r=k.getTypedArrayFromDType(a,k.sizeFromShape(n));for(let s=0;s<r.length;++s){let i=s*t,o=e[i];for(let l=0;l<t;++l){let c=e[i+l];c>o&&(o=c)}r[s]=o}return r}var QN=Rt((e,t)=>Math.max(e,t)),bU=Zt(Hs,QN),xU={kernelName:Hs,backendName:"cpu",kernelFunc:bU},eT=Rt((e,t)=>Math.min(e,t)),vU=Zt(Xs,eT),wU={kernelName:Xs,backendName:"cpu",kernelFunc:vU},gv=Rt((e,t)=>e*t),kU=xv((e,t,n,a)=>({real:e*n-t*a,imag:e*a+t*n})),vv=Zt(Ys,gv,kU),IU={kernelName:Ys,backendName:"cpu",kernelFunc:vv};function tT(e,t,n){let a=k.createScalarValue(-1,n);return gv([],t,a,e,n)}function NU(e){let{inputs:t,backend:n}=e,{x:a}=t;we(a,"neg");let r=n.data.get(a.dataId).values,[s,i]=tT(r,a.shape,a.dtype);return n.makeTensorInfo(i,a.dtype,s)}var TU={kernelName:al,backendName:"cpu",kernelFunc:NU},nT=Rt((e,t)=>e!==t?1:0),SU=Zt(rl,nT,null,"bool"),CU={kernelName:rl,backendName:"cpu",kernelFunc:SU};function yv(e,t,n,a,r){let s=t.length,i=k.sizeFromShape(t),o=k.computeStrides(t),l=k.computeStrides(r),c=k.getTypedArrayFromDType(n,k.sizeFromShape(r));for(let u=0;u<i;++u){let p=k.indexToLoc(u,s,o),d=new Array(p.length);for(let m=0;m<d.length;m++)d[m]=p[a[m]];let h=k.locToIndex(d,s,l);c[h]=e[u]}return c}function ga(e){let{inputs:t,attrs:n,backend:a}=e,{x:r}=t,{perm:s}=n;we(r,"transpose");let i=r.shape.length,o=new Array(i);for(let u=0;u<o.length;u++)o[u]=r.shape[s[u]];let l=a.data.get(r.dataId).values,c=yv(l,r.shape,r.dtype,s,o);return{dataId:a.write(c,o,r.dtype),shape:o,dtype:r.dtype}}var _U={kernelName:fi,backendName:"cpu",kernelFunc:ga};function aT(e,t,n,a){let[r,s]=_.computeOutAndReduceShapes(e,a),i=la(t,"int32"),o=k.makeZerosTypedArray(k.sizeFromShape(r),i),l=k.sizeFromShape(s);for(let c=0;c<o.length;++c){let u=c*l,p=1;for(let d=0;d<l;++d)p*=n[u+d];o[c]=p}return{outVals:o,outShape:r,outDtype:i}}function EU(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{axis:s,keepDims:i}=a;we(r,"prod");let o=r.shape.length,l=k.parseAxisParam(s,r.shape),c=_.getAxesPermutation(l,o),u=l,p=r,d=[];c!=null&&(p=ga({inputs:{x:r},backend:n,attrs:{perm:c}}),d.push(p),u=_.getInnerMostAxes(u.length,o));let h=n.data.get(p.dataId).values,{outVals:m,outShape:f,outDtype:g}=aT(p.shape,p.dtype,h,u),y=f;return i&&(y=_.expandShapeToKeepDim(f,l)),d.forEach(b=>n.disposeIntermediateTensorInfo(b)),n.makeTensorInfo(y,g,m)}var AU={kernelName:cl,backendName:"cpu",kernelFunc:EU};function bv(e,t,n,a){let r=e===t,s=e<t&&n<0,i=t<e&&n>1;if(r||s||i)return k.makeZerosTypedArray(0,a);let o=Math.abs(Math.ceil((t-e)/n)),l=k.makeZerosTypedArray(o,a);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 rT=tu(e=>1/Math.sqrt(e)),FU=nu(ii,rT),$U={kernelName:ii,backendName:"cpu",kernelFunc:FU};function Am(e,t,n,a,r){let s=cn.isSliceContinous(a,t,n),i=k.sizeFromShape(n),o=k.computeStrides(a);if(s){let p=cn.computeFlatOffset(t,o);return r==="string"?e.slice(p,p+i):e.subarray(p,p+i)}let l=r==="string"?_.fromUint8ToStringArray(e):e,c=Le(a,r,l),u=Le(n,r);for(let p=0;p<u.size;++p){let d=u.indexToLoc(p),h=d.map((m,f)=>m+t[f]);u.set(c.get(...h),...d)}return r==="string"?_.fromStringArrayToUint8(u.values):u.values}function Vi(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{begin:s,size:i}=a;we(r,"slice");let[o,l]=cn.parseSliceParams(r,s,i);cn.assertParamsValid(r,o,l);let c=n.data.get(r.dataId).values,u=Am(c,o,l,r.shape,r.dtype);return n.makeTensorInfo(l,r.dtype,u)}var DU={kernelName:gl,backendName:"cpu",kernelFunc:Vi},sT=Rt((e,t)=>{let n=e-t;return n*n}),RU=Zt(di,sT),MU={kernelName:di,backendName:"cpu",kernelFunc:RU};function iT(e,t,n,a){let r=Le(e,t.dtype);for(let s=0;s<r.size;s++){let i=r.indexToLoc(s),o=new Array(i.length);for(let l=0;l<o.length;l++)o[l]=i[l]*n[l]+a[l];r.set(t.get(...o),...i)}return r}var oT=Rt((e,t)=>e-t),PU=xv((e,t,n,a)=>({real:e-n,imag:t-a})),wv=Zt(hi,oT,PU),OU={kernelName:hi,backendName:"cpu",kernelFunc:wv};function lT(e,t){let n=new Array(e.rank);for(let r=0;r<n.length;r++)n[r]=e.shape[r]*t[r];let a=Le(n,e.dtype);for(let r=0;r<a.values.length;++r){let s=a.indexToLoc(r),i=new Array(e.rank);for(let l=0;l<i.length;l++)i[l]=s[l]%e.shape[l];let o=e.locToIndex(i);a.values[r]=e.values[o]}return a}function uT(e,t,n,a,r){let s=t[t.length-1],[i,o]=[e.length/s,s],l=k.getTypedArrayFromDType(n,i*a),c=k.getTypedArrayFromDType("int32",i*a);for(let p=0;p<i;p++){let d=p*o,h=e.subarray(d,d+o),m=[];for(let b=0;b<h.length;b++)m.push({value:h[b],index:b});m.sort((b,v)=>v.value-b.value);let f=p*a,g=l.subarray(f,f+a),y=c.subarray(f,f+a);for(let b=0;b<a;b++)g[b]=m[b].value,y[b]=m[b].index}let u=t.slice();return u[u.length-1]=a,[Le(u,n,l),Le(u,"int32",c)]}function cT(e,t,n,a){let r=k.parseAxisParam(t,n)[0],s=[1,n[0],1];for(let m=0;m<r;m++)s[0]*=n[m];s[1]=n[r];for(let m=r+1;m<n.length;m++)s[2]*=n[m];let i={},o=new Int32Array(n[r]),l=new Ot(s,a,e),c=[],u=s[0]===1&&s[2]===1;for(let m=0;m<n[r];m++){let f;if(u)f=e[m].toString();else{let g=[];for(let y=0;y<s[0];y++)for(let b=0;b<s[2];b++)g.push(l.get(y,m,b));f=g.join(",")}if(i[f]!==void 0)o[m]=i[f];else{let g=Object.keys(i).length;i[f]=g,o[m]=g,c.push(m)}}let p=s.slice();p[1]=Object.keys(i).length;let d=new Ot(p,a);c.forEach((m,f)=>{for(let g=0;g<s[0];g++)for(let y=0;y<s[2];y++)d.set(l.get(g,m,y),g,f,y)});let h=n.slice();return h[r]=p[1],{outputValues:d.values,outputShape:h,indices:o}}var LU="3.0.0";ih("cpu",()=>new YV,1);var dT=it(zo,e=>e>=0?e:Math.exp(e)-1),zU={kernelName:zo,backendName:"cpu",kernelFunc:dT};function hT(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{alpha:s}=a;we([r],"leakyRelu");let i=k.sizeFromShape(r.shape),o=n.data.get(r.dataId).values,l=k.getTypedArrayFromDType("float32",i);for(let c=0;c<o.length;c++)l[c]=o[c]<0?s*o[c]:o[c];return n.makeTensorInfo(r.shape,"float32",l)}var BU={kernelName:Vs,backendName:"cpu",kernelFunc:hT},WU=Rt((e,t)=>e<0?t*e:e);function mT(e){let{inputs:t,backend:n}=e,{x:a,alpha:r}=t;we([a,r],"prelu");let s=n.data.get(a.dataId).values,i=n.data.get(r.dataId).values,[o,l]=WU(a.shape,r.shape,s,i,a.dtype);return n.makeTensorInfo(l,a.dtype,o)}var VU={kernelName:ei,backendName:"cpu",kernelFunc:mT},fT=it(ti,e=>Math.max(0,e)),UU={kernelName:ti,backendName:"cpu",kernelFunc:fT},gT=it(ai,e=>Math.min(Math.max(0,e),6)),GU={kernelName:ai,backendName:"cpu",kernelFunc:gT};function kv(e,t,n,a,r){if(n==="linear")return ar({inputs:{x:t},backend:e});if(n==="relu")return fT({inputs:{x:t},backend:e});if(n==="elu")return dT({inputs:{x:t},backend:e});if(n==="relu6")return gT({inputs:{x:t},backend:e});if(n==="prelu")return mT({inputs:{x:t,alpha:a},backend:e});if(n==="leakyrelu")return hT({inputs:{x:t},backend:e,attrs:{alpha:r}});throw new Error(`Activation ${n} has not been implemented for the CPU backend.`)}function It(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{shape:s}=a,i=k.sizeFromShape(r.shape),o=k.inferFromImplicitShape(s,i),l=k.sizeFromShape(o);k.assert(i===l,()=>`The new shape (${o}) has ${l} elements and the old shape (${r.shape}) has ${i} elements. The new shape and old shape must have the same number of elements.`),n.incRef(r.dataId);let c=n.data.get(r.dataId);if(c.complexTensorInfos!=null){let u=c.complexTensorInfos.real,p=c.complexTensorInfos.imag;u.shape=o,p.shape=o}return{dataId:r.dataId,shape:o,dtype:r.dtype}}var HU={kernelName:dl,backendName:"cpu",kernelFunc:It};function yT(e){let{inputs:t,backend:n,attrs:a}=e,{a:r,b:s}=t,{transposeA:i,transposeB:o}=a;we([r,s],"matMul");let l=r.shape.length,c=s.shape.length,u=i?r.shape[l-2]:r.shape[l-1],p=o?s.shape[c-1]:s.shape[c-2],d=i?r.shape[l-1]:r.shape[l-2],h=o?s.shape[c-2]:s.shape[c-1],m=r.shape.slice(0,-2),f=s.shape.slice(0,-2),g=k.sizeFromShape(m),y=k.sizeFromShape(f),b=g===y||g===1||y===1;k.assert(l>=2&&c>=2&&b,()=>`Error in matMul: the input batch dimensions must either be the same or at least one input batch dimension must be 1. Got input batch dimensions of (${m}) and (${f}).`);let v=(g>y?r.shape.slice(0,-2):s.shape.slice(0,-2)).concat([d,h]);k.assert(u===p,()=>`Error in matMul: inner shapes (${u}) and (${p}) of Tensors with shapes ${r.shape} and ${s.shape} and transposeA=${i} and transposeB=${o} must match.`);let x=i?[g,u,d]:[g,d,u],N=o?[y,h,p]:[y,p,h],T=It({inputs:{x:r},backend:n,attrs:{shape:x}}),C=It({inputs:{x:s},backend:n,attrs:{shape:N}}),$=i?T.shape[1]:T.shape[2],F=i?T.shape[2]:T.shape[1],O=o?C.shape[1]:C.shape[2],B=Math.max(g,y),W=n.data.get(T.dataId).values,G=n.data.get(C.dataId).values,q=k.computeStrides(T.shape),H=k.computeStrides(C.shape),[K,X,ne]=i?[q[0],1,q[1]]:[q[0],q[1],1],[J,se,Q]=o?[1,H[1],H[0]]:[H[1],1,H[0]],oe=F*O,ae=Le([B,F,O],T.dtype),pe=ae.values,ue=n.blockSize;for(let ge=0;ge<B;ge++)for(let ye=0;ye<F;ye+=ue)for(let ke=0;ke<O;ke+=ue)for(let Se=0;Se<$;Se+=ue){let Ce=Math.min(ye+ue,F),Re=Math.min(ke+ue,O),Ke=Math.min(Se+ue,$);for(let Ve=ye;Ve<Ce;Ve++)for(let rt=ke;rt<Re;rt++){let ot=0;for(let He=Se;He<Ke;He++){let ct=Math.min(ge,g-1)*K,pt=Math.min(ge,y-1)*Q,Gn=W[ct+Ve*X+He*ne],Je=G[He*J+rt*se+pt];ot+=Gn*Je}pe[ge*oe+(Ve*O+rt)]+=ot}}return n.disposeIntermediateTensorInfo(T),n.disposeIntermediateTensorInfo(C),n.makeTensorInfo(v,ae.dtype,ae.values)}var jU={kernelName:Es,backendName:"cpu",kernelFunc:yT};function qU(e){let{inputs:t,backend:n,attrs:a}=e,{a:r,b:s,bias:i,preluActivationWeights:o}=t,{transposeA:l,transposeB:c,activation:u,leakyreluAlpha:p}=a,d,h,m,f=[];d=yT({inputs:{a:r,b:s},attrs:{transposeA:l,transposeB:c},backend:n}),i&&(h=fp({inputs:{a:d,b:i},backend:n}),f.push(d),d=h),u&&(m=kv(n,d,u,o,p),f.push(d),d=m);for(let g of f)n.disposeIntermediateTensorInfo(g);return d}var KU={kernelName:gi,backendName:"cpu",kernelFunc:qU},XU=it(Co,e=>Math.acos(e)),YU={kernelName:Co,backendName:"cpu",kernelFunc:XU},ZU=it(_o,e=>Math.acosh(e)),JU={kernelName:_o,backendName:"cpu",kernelFunc:ZU};function QU(e){let{inputs:t,backend:n}=e,a=t;we(t,"addN");let r=a.map(o=>n.data.get(o.dataId).values),s=Le(a[0].shape,a[0].dtype),i=s.values;for(let o=0;o<a.length;o++){let l=r[o];for(let c=0;c<i.length;c++)i[c]+=l[c]}return n.makeTensorInfo(s.shape,s.dtype,s.values)}var eG={kernelName:Ss,backendName:"cpu",kernelFunc:QU};function tG(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{axis:s,keepDims:i}=a;we(r,"all");let o=k.parseAxisParam(s,r.shape),l=o,c=_.getAxesPermutation(l,r.shape.length),u=r;c!=null&&(u=ga({inputs:{x:r},backend:n,attrs:{perm:c}}),l=_.getInnerMostAxes(l.length,r.shape.length)),_.assertAxesAreInnerMostDims("all",l,u.shape.length);let[p,d]=_.computeOutAndReduceShapes(u.shape,l),h=k.sizeFromShape(d),m=k.makeZerosTypedArray(k.sizeFromShape(p),u.dtype),f=n.data.get(u.dataId).values;for(let y=0;y<m.length;++y){let b=y*h,v=f[b];for(let x=0;x<h;++x){let N=f[b+x];v=v&&N}m[y]=v}c!=null&&n.disposeIntermediateTensorInfo(u);let g=n.makeTensorInfo(p,u.dtype,m);if(i){let y=_.expandShapeToKeepDim(p,o),b=It({inputs:{x:g},backend:n,attrs:{shape:y}});return n.disposeIntermediateTensorInfo(g),b}return g}var nG={kernelName:yd,backendName:"cpu",kernelFunc:tG};function aG(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{axis:s,keepDims:i}=a;we(r,"any");let o=k.parseAxisParam(s,r.shape),l=o,c=_.getAxesPermutation(l,r.shape.length),u=r;c!=null&&(u=ga({inputs:{x:r},backend:n,attrs:{perm:c}}),l=_.getInnerMostAxes(l.length,r.shape.length)),_.assertAxesAreInnerMostDims("any",l,u.shape.length);let[p,d]=_.computeOutAndReduceShapes(u.shape,l),h=k.sizeFromShape(d),m=k.makeZerosTypedArray(k.sizeFromShape(p),u.dtype),f=n.data.get(u.dataId).values;for(let y=0;y<m.length;++y){let b=y*h,v=f[b];for(let x=0;x<h;++x){let N=f[b+x];v=v||N}m[y]=v}c!=null&&n.disposeIntermediateTensorInfo(u);let g=n.makeTensorInfo(p,u.dtype,m);if(i){let y=_.expandShapeToKeepDim(p,o),b=It({inputs:{x:g},backend:n,attrs:{shape:y}});return n.disposeIntermediateTensorInfo(g),b}return g}var rG={kernelName:bd,backendName:"cpu",kernelFunc:aG};function sG(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{axis:s}=a;we(r,"argMax");let i=k.parseAxisParam(s,r.shape),o=_.getAxesPermutation(i,r.shape.length),l=r,c=[];o!=null&&(l=ga({inputs:{x:r},backend:n,attrs:{perm:o}}),c.push(l),i=_.getInnerMostAxes(i.length,l.shape.length)),i=[i[0]],_.assertAxesAreInnerMostDims("argMax",i,l.shape.length);let[u,p]=_.computeOutAndReduceShapes(l.shape,i),d=k.sizeFromShape(u),h=k.makeZerosTypedArray(d,"int32"),m=k.sizeFromShape(p),f=n.data.get(l.dataId).values;for(let g=0;g<h.length;++g){let y=g*m,b=f[y],v=0;for(let x=0;x<m;++x){let N=f[y+x];N>b&&(b=N,v=x)}h[g]=v}return c.forEach(g=>n.disposeIntermediateTensorInfo(g)),n.makeTensorInfo(u,"int32",h)}var iG={kernelName:Cs,backendName:"cpu",kernelFunc:sG};function oG(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{axis:s}=a;we(r,"argMin");let i=k.parseAxisParam(s,r.shape),o=_.getAxesPermutation(i,r.shape.length),l=r,c=[];o!=null&&(l=ga({inputs:{x:r},backend:n,attrs:{perm:o}}),c.push(l),i=_.getInnerMostAxes(i.length,l.shape.length)),i=[i[0]],_.assertAxesAreInnerMostDims("argMin",i,l.shape.length);let[u,p]=_.computeOutAndReduceShapes(l.shape,i),d=k.sizeFromShape(u),h=k.makeZerosTypedArray(d,"int32"),m=k.sizeFromShape(p),f=n.data.get(l.dataId).values;for(let g=0;g<h.length;++g){let y=g*m,b=f[y],v=0;for(let x=0;x<m;++x){let N=f[y+x];N<b&&(b=N,v=x)}h[g]=v}return c.forEach(g=>n.disposeIntermediateTensorInfo(g)),n.makeTensorInfo(u,"int32",h)}var lG={kernelName:lc,backendName:"cpu",kernelFunc:oG},uG=it(Eo,e=>Math.asin(e)),cG={kernelName:Eo,backendName:"cpu",kernelFunc:uG},pG=it(Ao,e=>Math.asinh(e)),dG={kernelName:Ao,backendName:"cpu",kernelFunc:pG},hG=it(Fo,e=>Math.atan(e)),mG={kernelName:Fo,backendName:"cpu",kernelFunc:hG},fG=Rt((e,t)=>Math.atan2(e,t)),gG=Zt(Do,fG),yG={kernelName:Do,backendName:"cpu",kernelFunc:gG},bG=it($o,e=>Math.atanh(e)),xG={kernelName:$o,backendName:"cpu",kernelFunc:bG};function Iv(e,t,n,a,r,s){let i=r.strideHeight,o=r.strideWidth,l=r.dilationHeight,c=r.dilationWidth,u=r.effectiveFilterHeight,p=r.effectiveFilterWidth,d=r.padInfo.top,h=r.padInfo.left,m=s==="max"?Number.NEGATIVE_INFINITY:Number.POSITIVE_INFINITY,f=Le(r.outShape,n),g=f.values,y=r.outShape[1]*r.outShape[2]*r.outShape[3],b=r.outShape[2]*r.outShape[3],v=r.outShape[3];for(let x=0;x<r.batchSize;++x){let N=x*y,T=x*a[0];for(let C=0;C<r.inChannels;++C)for(let $=0;$<r.outHeight;++$){let F=$*i-d,O=Math.max(0,F),B=Math.min(r.inHeight,u+F),W=N+$*b;for(let G=0;G<r.outWidth;++G){let q=G*o-h,H=Math.max(0,q),K=Math.min(r.inWidth,p+q),X=m,ne=0,J=0;for(let Q=O;Q<B;Q+=l){let oe=T+Q*a[1];for(let ae=H;ae<K;ae+=c){let pe=oe+ae*a[2],ue=e[pe+C];s==="max"&&ue>X?X=ue:s==="avg"&&(ne+=ue,J++)}if(isNaN(X))break}let se=W+G*v+C;g[se]=s==="avg"?ne/J:X}}}return f}function bT(e,t,n,a,r=!1,s=!1){let i=Le(a.outShape,"int32"),o=a.strideHeight,l=a.strideWidth,c=a.dilationHeight,u=a.dilationWidth,p=a.effectiveFilterHeight,d=a.effectiveFilterWidth,h=a.padInfo.top,m=a.padInfo.left,f=Le(t,n,e);for(let g=0;g<a.batchSize;++g)for(let y=0;y<a.inChannels;++y)for(let b=0;b<a.outHeight;++b){let v=b*o-h,x=v;for(;x<0;)x+=c;let N=Math.min(a.inHeight,p+v);for(let T=0;T<a.outWidth;++T){let C=T*l-m,$=C;for(;$<0;)$+=u;let F=Math.min(a.inWidth,d+C),O=Number.NEGATIVE_INFINITY,B=-1;for(let W=x;W<N;W+=c){let G=W-v;for(let q=$;q<F;q+=u){let H=q-C,K=f.get(g,W,q,y);K>O&&(O=K,r?B=s?((g*a.inHeight+W)*a.inWidth+q)*a.inChannels+y:(W*a.inWidth+q)*a.inChannels+y:B=G*d+H)}}i.set(B,g,b,T,y)}}return i}function xT(e,t,n,a,r,s){let i=r.strideDepth,o=r.strideHeight,l=r.strideWidth,c=r.dilationDepth,u=r.dilationHeight,p=r.dilationWidth,d=r.effectiveFilterDepth,h=r.effectiveFilterHeight,m=r.effectiveFilterWidth,f=r.padInfo.front,g=r.padInfo.top,y=r.padInfo.left,b=s==="max"?Number.NEGATIVE_INFINITY:Number.POSITIVE_INFINITY,v=Le(r.outShape,n),x=v.values,N=r.outShape[1]*r.outShape[2]*r.outShape[3]*r.outShape[4],T=r.outShape[2]*r.outShape[3]*r.outShape[4],C=r.outShape[3]*r.outShape[4],$=r.outShape[4];for(let F=0;F<r.batchSize;++F){let O=F*N,B=F*a[0];for(let W=0;W<r.inChannels;++W)for(let G=0;G<r.outDepth;++G){let q=G*i-f,H=q;for(;H<0;)H+=c;let K=Math.min(r.inDepth,d+q),X=O+G*T;for(let ne=0;ne<r.outHeight;++ne){let J=ne*o-g,se=J;for(;se<0;)se+=u;let Q=Math.min(r.inHeight,h+J),oe=X+ne*C;for(let ae=0;ae<r.outWidth;++ae){let pe=ae*l-y,ue=pe;for(;ue<0;)ue+=p;let ge=Math.min(r.inWidth,m+pe),ye=oe+ae*$,ke=b,Se=0,Ce=0;for(let Ke=H;Ke<K;Ke+=c){let Ve=B+Ke*a[1];for(let rt=se;rt<Q;rt+=u){let ot=Ve+rt*a[2];for(let He=ue;He<ge;He+=p){let ct=ot+He*a[3],pt=e[ct+W];if(s==="max"&&pt>ke?ke=pt:s==="avg"&&(Se+=pt,Ce++),isNaN(ke))break}if(isNaN(ke))break}if(isNaN(ke))break}let Re=ye+W;x[Re]=s==="avg"?Se/Ce:ke}}}}return v}function vG(e,t){let n=Le(t.outShape,"int32"),a=t.strideDepth,r=t.strideHeight,s=t.strideWidth,i=t.dilationDepth,o=t.dilationHeight,l=t.dilationWidth,c=t.effectiveFilterDepth,u=t.effectiveFilterHeight,p=t.effectiveFilterWidth,d=t.padInfo.front,h=t.padInfo.top,m=t.padInfo.left;for(let f=0;f<t.batchSize;++f)for(let g=0;g<t.inChannels;++g)for(let y=0;y<t.outDepth;++y){let b=y*a-d,v=b;for(;v<0;)v+=i;let x=Math.min(t.inDepth,c+b);for(let N=0;N<t.outHeight;++N){let T=N*r-h,C=T;for(;C<0;)C+=o;let $=Math.min(t.inHeight,u+T);for(let F=0;F<t.outWidth;++F){let O=F*s-m,B=O;for(;B<0;)B+=l;let W=Math.min(t.inWidth,p+O),G=Number.NEGATIVE_INFINITY,q=-1;for(let H=v;H<x;H+=i){let K=H-b;for(let X=C;X<$;X+=o){let ne=X-T;for(let J=B;J<W;J+=l){let se=J-O,Q=e.get(f,H,X,J,g);Q>=G&&(G=Q,q=K*u*p+ne*u+se)}}}n.set(q,f,y,N,F,g)}}}return n}function wG(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t;we(r,"avgPool");let{filterSize:s,strides:i,pad:o,dimRoundingMode:l}=a,c=1;k.assert(_.eitherStridesOrDilationsAreOne(i,c),()=>`Error in avgPool: Either strides or dilations must be 1. Got strides ${i} and dilations '${c}'`);let u=_.computePool2DInfo(r.shape,s,i,c,o,l),p;if(u.filterWidth===1&&u.filterHeight===1&&k.arraysEqual(u.inShape,u.outShape))p=ar({inputs:{x:r},backend:n});else{let d=n.data.get(r.dataId).values,h=k.computeStrides(r.shape),m=Iv(d,r.shape,r.dtype,h,u,"avg");p=n.makeTensorInfo(u.outShape,r.dtype,m.values)}return p}var kG={kernelName:_s,backendName:"cpu",kernelFunc:wG};function IG(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{filterSize:s,strides:i,pad:o,dimRoundingMode:l,dataFormat:c}=a;we(r,"avgPool3d");let u=_.computePool3DInfo(r.shape,s,i,1,o,l,c),p=n.data.get(r.dataId).values,d=xT(p,r.shape,r.dtype,k.computeStrides(r.shape),u,"avg");return n.makeTensorInfo(d.shape,"float32",d.values)}var NG={kernelName:uc,backendName:"cpu",kernelFunc:IG};function TG(e){let{inputs:t,backend:n,attrs:a}=e,{dy:r,input:s}=t,{filterSize:i,strides:o,pad:l,dimRoundingMode:c}=a;we([r,s],"avgPool3DGrad");let u=_.computePool3DInfo(s.shape,i,o,1,l,c),p=u.strideDepth,d=u.strideHeight,h=u.strideWidth,m=u.filterDepth,f=u.filterHeight,g=u.filterWidth,y=u.dilationDepth,b=u.dilationHeight,v=u.dilationWidth,x=u.effectiveFilterDepth,N=u.effectiveFilterHeight,T=u.effectiveFilterWidth,C=x-1-u.padInfo.front,$=T-1-u.padInfo.left,F=N-1-u.padInfo.top,O=Le(s.shape,"float32"),B=1/(m*f*g),W=n.bufferSync(r);for(let G=0;G<u.batchSize;++G)for(let q=0;q<u.inChannels;++q)for(let H=0;H<u.inDepth;++H)for(let K=0;K<u.inHeight;++K)for(let X=0;X<u.inWidth;++X){let ne=H-C,J=K-F,se=X-$,Q=0;for(let oe=0;oe<x;oe+=y){let ae=(ne+oe)/p;if(!(ae<0||ae>=u.outDepth||Math.floor(ae)!==ae))for(let pe=0;pe<N;pe+=b){let ue=(J+pe)/d;if(!(ue<0||ue>=u.outHeight||Math.floor(ue)!==ue))for(let ge=0;ge<T;ge+=v){let ye=(se+ge)/h;ye<0||ye>=u.outWidth||Math.floor(ye)!==ye||(Q+=W.get(G,ae,ue,ye,q))}}}O.set(Q*B,G,H,K,X,q)}return n.makeTensorInfo(O.shape,O.dtype,O.values)}var SG={kernelName:vd,backendName:"cpu",kernelFunc:TG};function CG(e){let{inputs:t,backend:n,attrs:a}=e,{dy:r,input:s}=t,i=s;we([r,s],"avgPoolGrad");let{filterSize:o,strides:l,pad:c}=a,u=_.computePool2DInfo(i.shape,o,l,1,c),p=u.strideHeight,d=u.strideWidth,h=u.filterHeight,m=u.filterWidth,f=u.dilationHeight,g=u.dilationWidth,y=u.effectiveFilterHeight,b=u.effectiveFilterWidth,v=b-1-u.padInfo.left,x=y-1-u.padInfo.top,N=Le(i.shape,"float32"),T=1/(h*m),C=n.data.get(r.dataId).values,$=Le(r.shape,"float32",C);for(let F=0;F<u.batchSize;++F)for(let O=0;O<u.inChannels;++O)for(let B=0;B<u.inHeight;++B)for(let W=0;W<u.inWidth;++W){let G=B-x,q=W-v,H=0;for(let K=0;K<y;K+=f){let X=(G+K)/p;if(!(X<0||X>=u.outHeight||Math.floor(X)!==X))for(let ne=0;ne<b;ne+=g){let J=(q+ne)/d;J<0||J>=u.outWidth||Math.floor(J)!==J||(H+=$.get(F,X,J,O))}}N.set(H*T,F,B,W,O)}return n.makeTensorInfo(N.shape,N.dtype,N.values)}var _G={kernelName:xd,backendName:"cpu",kernelFunc:CG};function EG(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,scale:s,offset:i,mean:o,variance:l}=t;k.assert(o.shape.length===l.shape.length,()=>"Batch normalization gradient requires mean and variance to have equal ranks."),k.assert(i==null||o.shape.length===i.shape.length,()=>"Batch normalization gradient requires mean and offset to have equal ranks."),k.assert(s==null||o.shape.length===s.shape.length,()=>"Batch normalization gradient requires mean and scale to have equal ranks."),we([r,o,l,s,i],"batchNorm");let{varianceEpsilon:c}=a;c==null&&(c=.001);let u=n.data.get(r.dataId).values,p=n.data.get(o.dataId).values,d=n.data.get(l.dataId).values,h=s?n.data.get(s.dataId).values:new Float32Array([1]),m=i?n.data.get(i.dataId).values:new Float32Array([0]),f=new Float32Array(u.length),g=m.length,y=h.length,b=d.length,v=p.length,x=0,N=0,T=0,C=0;for(let $=0;$<u.length;++$)f[$]=m[x++]+(u[$]-p[N++])*h[T++]/Math.sqrt(d[C++]+c),x>=g&&(x=0),N>=v&&(N=0),T>=y&&(T=0),C>=b&&(C=0);return n.makeTensorInfo(r.shape,r.dtype,f)}var AG={kernelName:Bs,backendName:"cpu",kernelFunc:EG};function FG(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{blockShape:s,crops:i}=a;we([r],"batchToSpaceND");let o=s.reduce((y,b)=>y*b),l=_.getReshaped(r.shape,s,o),c=_.getPermuted(l.length,s.length),u=_.getReshapedPermuted(r.shape,s,o),p=_.getSliceBeginCoords(i,s.length),d=_.getSliceSize(u,i,s.length),h=It({inputs:{x:r},backend:n,attrs:{shape:l}}),m=ga({inputs:{x:h},backend:n,attrs:{perm:c}}),f=It({inputs:{x:m},backend:n,attrs:{shape:u}}),g=Vi({inputs:{x:f},backend:n,attrs:{begin:p,size:d}});return n.disposeIntermediateTensorInfo(h),n.disposeIntermediateTensorInfo(m),n.disposeIntermediateTensorInfo(f),g}var $G={kernelName:cc,backendName:"cpu",kernelFunc:FG};function DG(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,weights:s}=t,{size:i}=a,o=n.data.get(r.dataId).values,l=n.data.get(s.dataId).values,c=mv(o,l,s.dtype,s.shape,i);return n.makeTensorInfo([i],s.dtype,c)}var RG={kernelName:wd,backendName:"cpu",kernelFunc:DG},MG=it(zr,(e,t)=>{let n=t;return e>n.clipValueMax?n.clipValueMax:e<n.clipValueMin?n.clipValueMin:e}),PG={kernelName:zr,backendName:"cpu",kernelFunc:MG},OG=e=>{let{x:t}=e.inputs,n=e.backend,a=new Float32Array(k.sizeFromShape(t.shape)),r=n.data.get(t.dataId),s=r.complexTensorInfos.real,i=r.complexTensorInfos.imag,o=n.data.get(s.dataId).values,l=n.data.get(i.dataId).values;for(let c=0;c<o.length;c++){let u=o[c],p=l[c];a[c]=Math.hypot(u,p)}return n.makeOutput(a,t.shape,"float32")},LG={kernelName:pc,backendName:"cpu",kernelFunc:OG};function au(e){let{inputs:t,backend:n}=e,{input:a}=t,r=n.data.get(a.dataId).complexTensorInfos.imag,s=n.data.get(r.dataId).values;return n.makeTensorInfo(r.shape,r.dtype,s)}var zG={kernelName:Md,backendName:"cpu",kernelFunc:au};function ru(e){let{inputs:t,backend:n,attrs:a}=e,{axis:r}=a,s=k.parseAxisParam(r,t[0].shape)[0],i=_.computeOutShape(t.map(f=>f.shape),s);if(k.sizeFromShape(i)===0)return n.makeTensorInfo(i,t[0].dtype,[]);let o=t.filter(f=>k.sizeFromShape(f.shape)>0);if(o.length===1)return ar({inputs:{x:o[0]},backend:n});let l=o.map(f=>f.shape);if(_.assertParamsConsistent(l,s),o[0].dtype==="complex64"){let f=o.map(x=>Wi({inputs:{input:x},backend:n})),g=o.map(x=>au({inputs:{input:x},backend:n})),y=ru({inputs:f,backend:n,attrs:{axis:s}}),b=ru({inputs:g,backend:n,attrs:{axis:s}}),v=zn({inputs:{real:y,imag:b},backend:n});return f.forEach(x=>n.disposeIntermediateTensorInfo(x)),g.forEach(x=>n.disposeIntermediateTensorInfo(x)),n.disposeIntermediateTensorInfo(y),n.disposeIntermediateTensorInfo(b),v}let c=o.map(f=>{let g=k.sizeFromShape(f.shape.slice(s));return It({inputs:{x:f},backend:n,attrs:{shape:[-1,g]}})}),u=c.map(f=>({vals:n.data.get(f.dataId).values,shape:f.shape}));i=_.computeOutShape(c.map(f=>f.shape),1);let p=c[0].shape[0]===1,d=fv(u,i,t[0].dtype,p),h=_.computeOutShape(o.map(f=>f.shape),s),m=n.makeTensorInfo(h,t[0].dtype,d);return c.forEach(f=>n.disposeIntermediateTensorInfo(f)),m}var BG={kernelName:Mo,backendName:"cpu",kernelFunc:ru};function vT(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,filter:s}=t,{strides:i,pad:o,dataFormat:l,dilations:c,dimRoundingMode:u}=a;we([r,s],"conv2d");let p=_.convertConv2DDataFormat(l),d=_.computeConv2DInfo(r.shape,s.shape,i,c,o,u,!1,p),h=d.filterHeight,m=d.filterWidth,f=d.dilationHeight,g=d.dilationWidth,y=d.padInfo.left,b=d.padInfo.top,v=d.dataFormat==="channelsLast",x=new Ot(d.outShape,r.dtype),N=k.computeStrides(r.shape),T=k.computeStrides(s.shape),C=N[0],$=v?N[1]:N[2],F=v?N[2]:1,O=v?1:N[1],B=x.strides[0],W=v?x.strides[1]:x.strides[2],G=v?x.strides[2]:1,q=v?1:x.strides[1],H=n.data.get(r.dataId).values,K=n.data.get(s.dataId).values,X=x.values;for(let ne=0;ne<d.batchSize;++ne){let J=ne*C,se=ne*B;for(let Q=0;Q<d.outHeight;++Q){let oe=se+Q*W,ae=Q*d.strideHeight-b;for(let pe=0;pe<h;++pe){let ue=ae+pe*f;if(ue<0||ue>=d.inHeight)continue;let ge=pe*T[0],ye=J+ue*$;for(let ke=0;ke<d.outWidth;++ke){let Se=oe+ke*G,Ce=ke*d.strideWidth-y;for(let Re=0;Re<m;++Re){let Ke=Ce+Re*g;if(Ke<0||Ke>=d.inWidth)continue;let Ve=ge+Re*T[1],rt=ye+Ke*F,ot=Ve;for(let He=0;He<d.inChannels;++He){let ct=H[rt+He*O];for(let pt=0;pt<d.outChannels;++pt)X[Se+pt*q]+=ct*K[ot+pt];ot+=d.outChannels}}}}}}return n.makeTensorInfo(x.shape,x.dtype,X)}var WG={kernelName:Fs,backendName:"cpu",kernelFunc:vT};function VG(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,dy:s}=t,{strides:i,pad:o,dataFormat:l,dimRoundingMode:c,filterShape:u}=a;we([r,s],"conv2dBackpropFilter");let p=_.convertConv2DDataFormat(l),d=_.computeConv2DInfo(r.shape,u,i,1,o,c,!1,p),{strideHeight:h,strideWidth:m,filterHeight:f,filterWidth:g}=d,y=d.dataFormat==="channelsLast",b=new Ot(d.filterShape,"float32"),v=d.padInfo.left,x=d.padInfo.top,N=n.data.get(r.dataId).values,T=n.data.get(s.dataId).values,C=new Ot(r.shape,r.dtype,N),$=new Ot(s.shape,s.dtype,T);for(let F=0;F<f;++F){let O=Math.max(0,Math.ceil((x-F)/h)),B=Math.min(d.outHeight,(d.inHeight+x-F)/h);for(let W=0;W<g;++W){let G=Math.max(0,Math.ceil((v-W)/m)),q=Math.min(d.outWidth,(d.inWidth+v-W)/m);for(let H=0;H<d.inChannels;++H)for(let K=0;K<d.outChannels;++K){let X=0;for(let ne=0;ne<d.batchSize;++ne)for(let J=O;J<B;++J){let se=F+J*h-x;for(let Q=G;Q<q;++Q){let oe=W+Q*m-v;y?X+=C.get(ne,se,oe,H)*$.get(ne,J,Q,K):X+=C.get(ne,H,se,oe)*$.get(ne,K,J,Q)}}b.set(X,F,W,H,K)}}}return n.makeTensorInfo(b.shape,b.dtype,b.values)}var UG={kernelName:Id,backendName:"cpu",kernelFunc:VG};function GG(e){let{inputs:t,backend:n,attrs:a}=e,{dy:r,filter:s}=t,{inputShape:i,strides:o,pad:l,dataFormat:c,dimRoundingMode:u}=a;we([r,s],"conv2dBackpropInput");let p=k.computeStrides(s.shape),d=k.computeStrides(r.shape),h=_.convertConv2DDataFormat(c),m=_.computeConv2DInfo(i,s.shape,o,1,l,u,!1,h),f=new Ot(m.inShape,"float32"),g=f.values,y=n.data.get(r.dataId).values,b=n.data.get(s.dataId).values,[v,x,N]=p,{batchSize:T,filterHeight:C,filterWidth:$,inChannels:F,inHeight:O,inWidth:B,outChannels:W,outHeight:G,outWidth:q,strideHeight:H,strideWidth:K}=m;h=m.dataFormat;let X=C-1-m.padInfo.top,ne=$-1-m.padInfo.left,J=h==="channelsLast",se=f.strides[0],Q=J?f.strides[1]:f.strides[2],oe=J?f.strides[2]:1,ae=J?1:f.strides[1],pe=d[0],ue=J?d[1]:d[2],ge=J?d[2]:1,ye=J?1:d[1];for(let ke=0;ke<T;++ke)for(let Se=0;Se<F;++Se)for(let Ce=0;Ce<O;++Ce){let Re=Ce-X,Ke=Math.max(0,Math.ceil(Re/H)),Ve=Math.min(G,(C+Re)/H);for(let rt=0;rt<B;++rt){let ot=rt-ne,He=Math.max(0,Math.ceil(ot/K)),ct=Math.min(q,($+ot)/K),pt=0;for(let Je=Ke;Je<Ve;++Je){let _n=Je*H-Re;for(let Qt=He;Qt<ct;++Qt){let En=Qt*K-ot,na=pe*ke+ue*Je+ge*Qt,xn=v*(C-1-_n)+x*($-1-En)+N*Se;for(let ln=0;ln<W;++ln){let aa=y[na+ye*ln],za=b[xn+ln];pt+=aa*za}}}let Gn=se*ke+Q*Ce+oe*rt+ae*Se;g[Gn]=pt}}return n.makeTensorInfo(f.shape,f.dtype,f.values)}var HG={kernelName:$s,backendName:"cpu",kernelFunc:GG};function jG(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,filter:s}=t,{strides:i,pad:o,dilations:l}=a;we([r,s],"conv3d");let c=_.computeConv3DInfo(r.shape,s.shape,i,l,o),{filterDepth:u,filterHeight:p,filterWidth:d,dilationDepth:h,dilationHeight:m,dilationWidth:f,padInfo:g}=c,y=g.front,b=g.left,v=g.top,x=new Ot(c.outShape,r.dtype),N=n.data.get(r.dataId).values,T=n.data.get(s.dataId).values,C=x.values,$=k.computeStrides(r.shape),F=k.computeStrides(s.shape);for(let O=0;O<c.batchSize;++O){let B=O*$[0],W=O*x.strides[0];for(let G=0;G<c.outDepth;++G){let q=W+G*x.strides[1],H=G*c.strideDepth-y;for(let K=0;K<u;++K){let X=H+K*h;if(X<0||X>=c.inDepth)continue;let ne=K*F[0],J=B+X*$[1];for(let se=0;se<c.outHeight;++se){let Q=q+se*x.strides[2],oe=se*c.strideHeight-v;for(let ae=0;ae<p;++ae){let pe=oe+ae*m;if(pe<0||pe>=c.inHeight)continue;let ue=ne+ae*F[1],ge=J+pe*$[2];for(let ye=0;ye<c.outWidth;++ye){let ke=Q+ye*c.outChannels,Se=ye*c.strideWidth-b;for(let Ce=0;Ce<d;++Ce){let Re=Se+Ce*f;if(Re<0||Re>=c.inWidth)continue;let Ke=ue+Ce*F[2],Ve=ge+Re*c.inChannels,rt=Ke;for(let ot=0;ot<c.inChannels;++ot){let He=N[Ve+ot];for(let ct=0;ct<c.outChannels;++ct)C[ke+ct]+=He*T[rt+ct];rt+=c.outChannels}}}}}}}}return n.makeTensorInfo(x.shape,x.dtype,x.values)}var qG={kernelName:dc,backendName:"cpu",kernelFunc:jG};function KG(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,dy:s}=t,{strides:i,pad:o,filterShape:l}=a;we([r,s],"conv3dBackpropFilterV2");let c=k.computeStrides(r.shape),u=k.computeStrides(s.shape),p=_.computeConv3DInfo(r.shape,l,i,1,o),d=p.strideDepth,h=p.strideHeight,m=p.strideWidth,f=p.filterDepth,g=p.filterHeight,y=p.filterWidth,b=new Ot(p.filterShape,"float32"),v=b.values,[x,N,T,C]=b.strides,$=n.data.get(s.dataId).values,[F,O,B,W]=u,G=n.data.get(r.dataId).values,[q,H,K,X]=c,ne=p.padInfo.front,J=p.padInfo.left,se=p.padInfo.top;for(let Q=0;Q<f;++Q){let oe=Math.max(0,Math.ceil((ne-Q)/d)),ae=Math.min(p.outDepth,(p.inDepth+ne-Q)/d),pe=Q*x;for(let ue=0;ue<g;++ue){let ge=Math.max(0,Math.ceil((se-ue)/h)),ye=Math.min(p.outHeight,(p.inHeight+se-ue)/h),ke=ue*N+pe;for(let Se=0;Se<y;++Se){let Ce=Math.max(0,Math.ceil((J-Se)/m)),Re=Math.min(p.outWidth,(p.inWidth+J-Se)/m),Ke=Se*T+ke;for(let Ve=0;Ve<p.inChannels;++Ve){let rt=Ve*C+Ke;for(let ot=0;ot<p.outChannels;++ot){let He=0;for(let ct=0;ct<p.batchSize;++ct){let pt=ct*q,Gn=ct*F;for(let Je=oe;Je<ae;++Je){let _n=(Q+Je*d-ne)*H+pt,Qt=Je*O+Gn;for(let En=ge;En<ye;++En){let na=(ue+En*h-se)*K+_n,xn=En*B+Qt;for(let ln=Ce;ln<Re;++ln){let aa=(Se+ln*m-J)*X+na,za=ln*W+xn;He+=G[aa+Ve]*$[za+ot]}}}}v[rt+ot]=He}}}}}return n.makeTensorInfo(b.shape,b.dtype,b.values)}var XG={kernelName:Nd,backendName:"cpu",kernelFunc:KG};function YG(e){let{inputs:t,backend:n,attrs:a}=e,{dy:r,filter:s}=t,{pad:i,strides:o,inputShape:l}=a;we([r],"conv3dBackpropInputV2");let c=k.computeStrides(r.shape),u=k.computeStrides(s.shape),p=_.computeConv3DInfo(l,s.shape,o,1,i),d=new Ot(p.inShape,"float32"),h=d.values,[m,f,g,y]=d.strides,b=n.data.get(r.dataId).values,[v,x,N,T]=c,C=n.data.get(s.dataId).values,[$,F,O,B]=u,{batchSize:W,filterDepth:G,filterHeight:q,filterWidth:H,inChannels:K,inDepth:X,inHeight:ne,inWidth:J,outChannels:se,outDepth:Q,outHeight:oe,outWidth:ae,strideDepth:pe,strideHeight:ue,strideWidth:ge}=p,ye=G-1-p.padInfo.front,ke=q-1-p.padInfo.top,Se=H-1-p.padInfo.left;for(let Ce=0;Ce<W;++Ce)for(let Re=0;Re<K;++Re)for(let Ke=0;Ke<X;++Ke){let Ve=Ke-ye,rt=Math.max(0,Math.ceil(Ve/pe)),ot=Math.min(Q,(G+Ve)/pe);for(let He=0;He<ne;++He){let ct=He-ke,pt=Math.max(0,Math.ceil(ct/ue)),Gn=Math.min(oe,(q+ct)/ue);for(let Je=0;Je<J;++Je){let _n=Je-Se,Qt=Math.max(0,Math.ceil(_n/ge)),En=Math.min(ae,(H+_n)/ge),na=0;for(let xn=rt;xn<ot;++xn){let ln=xn*pe-Ve;for(let aa=pt;aa<Gn;++aa){let za=aa*ue-ct;for(let An=Qt;An<En;++An){let uo=An*ge-_n,zu=v*Ce+x*xn+N*aa+T*An,wa=$*(G-1-ln)+F*(q-1-za)+O*(H-1-uo)+B*Re;for(let ra=0;ra<se;++ra){let ka=b[zu+ra],co=C[wa+ra];na+=ka*co}}}}h[m*Ce+f*Ke+g*He+y*Je+Re]=na}}}return n.makeTensorInfo(d.shape,d.dtype,d.values)}var ZG={kernelName:Td,backendName:"cpu",kernelFunc:YG},JG=it(Ds,e=>Math.cos(e)),QG={kernelName:Ds,backendName:"cpu",kernelFunc:JG},eH=it(Po,e=>Math.cosh(e)),tH={kernelName:Po,backendName:"cpu",kernelFunc:eH};function nH(e){let{inputs:t,backend:n,attrs:a}=e,{image:r,boxes:s,boxInd:i}=t,{cropSize:o,method:l,extrapolationValue:c}=a,[u,p,d,h]=r.shape,m=s.shape[0],[f,g]=o,y=Le([m,f,g,h],"float32"),b=n.data.get(s.dataId).values,v=n.data.get(i.dataId).values,x=n.data.get(r.dataId).values,N=k.computeStrides(r.shape),T=k.computeStrides(y.shape);for(let C=0;C<m;C++){let $=C*4,F=b[$],O=b[$+1],B=b[$+2],W=b[$+3],G=v[C];if(G>=u)continue;let q=f>1?(B-F)*(p-1)/(f-1):0,H=g>1?(W-O)*(d-1)/(g-1):0;for(let K=0;K<f;K++){let X=f>1?F*(p-1)+K*q:.5*(F+B)*(p-1);if(X<0||X>p-1){for(let ne=0;ne<g;ne++)for(let J=0;J<h;J++){let se=J+ne*T[2]+K*T[1]+C*T[0];y.values[se]=c}continue}if(l==="bilinear"){let ne=Math.floor(X),J=Math.ceil(X),se=X-ne;for(let Q=0;Q<g;Q++){let oe=g>1?O*(d-1)+Q*H:.5*(O+W)*(d-1);if(oe<0||oe>d-1){for(let ge=0;ge<h;ge++){let ye=ge+Q*T[2]+K*T[1]+C*T[0];y.values[ye]=c}continue}let ae=Math.floor(oe),pe=Math.ceil(oe),ue=oe-ae;for(let ge=0;ge<h;ge++){let ye=ge+ae*N[2]+ne*N[1]+G*N[0],ke=x[ye];ye=ge+pe*N[2]+ne*N[1]+G*N[0];let Se=x[ye];ye=ge+ae*N[2]+J*N[1]+G*N[0];let Ce=x[ye];ye=ge+pe*N[2]+J*N[1]+G*N[0];let Re=x[ye],Ke=ke+(Se-ke)*ue,Ve=Ce+(Re-Ce)*ue;ye=ge+Q*T[2]+K*T[1]+C*T[0],y.values[ye]=Ke+(Ve-Ke)*se}}}else for(let ne=0;ne<g;++ne){let J=g>1?O*(d-1)+ne*H:.5*(O+W)*(d-1);if(J<0||J>d-1){for(let oe=0;oe<h;oe++){let ae=oe+ne*T[2]+K*T[1]+C*T[0];y.values[ae]=c}continue}let se=Math.round(J),Q=Math.round(X);for(let oe=0;oe<h;oe++){let ae=oe+se*N[2]+Q*N[1]+G*N[0],pe=oe+ne*T[2]+K*T[1]+C*T[0];y.values[pe]=x[ae]}}}}return n.makeTensorInfo(y.shape,y.dtype,y.values)}var aH={kernelName:Oo,backendName:"cpu",kernelFunc:nH};function rH(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{axis:s,exclusive:i,reverse:o}=a;we(r,"cumsum");let l=_.getAxesPermutation([s],r.shape.length),c=r;l!=null&&(c=ga({inputs:{x:r},backend:n,attrs:{perm:l}}));let u=_.getInnerMostAxes(1,r.shape.length)[0];if(u!==c.shape.length-1)throw new Error(`backend.cumsum in CPU expects an inner-most axis=${c.shape.length-1} but got axis=${u}`);let p=la(c.dtype,"int32"),d=k.makeZerosTypedArray(k.sizeFromShape(c.shape),p),h=n.data.get(c.dataId).values,m=c.shape[c.shape.length-1],f=o?(y,b)=>y+m-b-1:(y,b)=>y+b;for(let y=0;y<h.length;y+=m)for(let b=0;b<m;b++){let v=f(y,b);if(b===0)d[v]=i?0:h[v];else{let x=f(y,b-1);d[v]=i?h[x]+d[x]:h[v]+d[x]}}let g=n.makeTensorInfo(c.shape,p,d);if(l!=null){let y=_.getUndoAxesPermutation(l),b=ga({inputs:{x:g},backend:n,attrs:{perm:y}});return n.disposeIntermediateTensorInfo(g),n.disposeIntermediateTensorInfo(c),b}return g}var sH={kernelName:Rs,backendName:"cpu",kernelFunc:rH};function iH(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,weights:s}=t,{size:i,binaryOutput:o}=a;if(r.shape.length===1){let l=n.data.get(r.dataId).values,c=n.data.get(s.dataId).values,u=mv(l,c,s.dtype,s.shape,i);return n.makeTensorInfo([i],s.dtype,u)}else if(r.shape.length===2){let l=n.bufferSync(r),c=n.bufferSync(s),u=VN(l,c,i,o);return n.makeTensorInfo(u.shape,s.dtype,u.values)}throw new Error(`Error in denseBincount: input must be at most rank 2, but got rank${r.shape.length}.`)}var oH={kernelName:Sd,backendName:"cpu",kernelFunc:iH};function lH(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{blockSize:s,dataFormat:i}=a;k.assert(i==="NHWC",()=>`Only NHWC dataFormat supported on CPU for depthToSpace. Got ${i}`),k.assert(s>1,()=>`blockSize should be > 1 for depthToSpace, but was: ${s}`);let o=r.shape[0],l=r.shape[1],c=r.shape[2],u=r.shape[3],p=l*s,d=c*s,h=u/(s*s),m=n.data.get(r.dataId).values,f=new Float32Array(o*p*d*h),g=0;for(let y=0;y<o;++y)for(let b=0;b<p;++b){let v=Math.floor(b/s),x=b%s;for(let N=0;N<d;++N){let T=Math.floor(N/s),C=N%s,$=(x*s+C)*h;for(let F=0;F<h;++F){let O=F+$+u*(T+c*(v+l*y));f[g++]=m[O]}}}return n.makeTensorInfo([o,p,d,h],r.dtype,f)}var uH={kernelName:Lo,backendName:"cpu",kernelFunc:lH};function wT(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,filter:s}=t,{strides:i,pad:o,dilations:l,dimRoundingMode:c}=a;we([r,s],"depthwiseConv2DNative");let u=k.computeStrides(r.shape),p=k.computeStrides(s.shape),d=l;d==null&&(d=[1,1]),k.assert(_.eitherStridesOrDilationsAreOne(i,d),()=>`Error in depthwiseConv2d: Either strides or dilations must be 1. Got strides ${i} and dilations '${d}'`);let h=_.computeConv2DInfo(r.shape,s.shape,i,d,o,c,!0),{filterHeight:m,filterWidth:f,dilationHeight:g,dilationWidth:y,padInfo:b}=h,v=b.left,x=b.top,N=h.outChannels/h.inChannels,T=new Ot(h.outShape,r.dtype),C=n.data.get(r.dataId).values,$=n.data.get(s.dataId).values,F=T.values;for(let O=0;O<h.batchSize;++O){let B=O*u[0],W=O*T.strides[0];for(let G=0;G<h.outHeight;++G){let q=W+G*T.strides[1],H=G*h.strideHeight-v;for(let K=0;K<m;++K){let X=H+K*g;if(X<0||X>=h.inHeight)continue;let ne=K*p[0],J=B+X*u[1];for(let se=0;se<h.outWidth;++se){let Q=q+se*T.strides[2],oe=se*h.strideWidth-x;for(let ae=0;ae<f;++ae){let pe=oe+ae*y;if(pe<0||pe>=h.inWidth)continue;let ue=ne+ae*p[1],ge=J+pe*h.inChannels,ye=Q,ke=ue;for(let Se=0;Se<h.inChannels;++Se){let Ce=C[ge+Se];for(let Re=0;Re<N;++Re)F[ye+Re]+=Ce*$[ke+Re];ye+=N,ke+=N}}}}}}return n.makeTensorInfo(T.shape,T.dtype,T.values)}var cH={kernelName:Ms,backendName:"cpu",kernelFunc:wT};function pH(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,dy:s}=t,{strides:i,dilations:o,pad:l,dimRoundingMode:c,filterShape:u}=a;we([r,s],"depthwiseConv2dNativeBackpropFilter");let p=_.computeConv2DInfo(r.shape,u,i,o,l,c,!0),{strideHeight:d,strideWidth:h,filterHeight:m,filterWidth:f}=p,g=new Ot(p.filterShape,"float32"),y=p.padInfo.left,b=p.padInfo.top,v=p.outChannels/p.inChannels,x=n.data.get(r.dataId).values,N=new Ot(r.shape,r.dtype,x),T=n.data.get(s.dataId).values,C=new Ot(s.shape,s.dtype,T);for(let $=0;$<m;++$){let F=Math.max(0,Math.ceil((b-$)/d)),O=Math.min(p.outHeight,(p.inHeight+b-$)/d);for(let B=0;B<f;++B){let W=Math.max(0,Math.ceil((y-B)/h)),G=Math.min(p.outWidth,(p.inWidth+y-B)/h);for(let q=0;q<p.outChannels;++q){let H=Math.trunc(q/v),K=q%v,X=0;for(let ne=0;ne<p.batchSize;++ne)for(let J=F;J<O;++J){let se=$+J*d-b;for(let Q=W;Q<G;++Q){let oe=B+Q*h-y;X+=N.get(ne,se,oe,H)*C.get(ne,J,Q,q)}}g.set(X,$,B,H,K)}}}return n.makeTensorInfo(g.shape,g.dtype,g.values)}var dH={kernelName:Cd,backendName:"cpu",kernelFunc:pH};function hH(e){let{inputs:t,backend:n,attrs:a}=e,{dy:r,filter:s}=t,{strides:i,dilations:o,pad:l,dimRoundingMode:c,inputShape:u}=a;we([r,s],"depthwiseConv2DNativeBackpropInput");let p=k.computeStrides(r.shape),d=k.computeStrides(s.shape),h=_.computeConv2DInfo(u,s.shape,i,o,l,c,!0),m=new Ot(h.inShape,"float32"),f=m.values,[g,y,b]=m.strides,v=n.data.get(r.dataId).values,[x,N,T]=p,C=n.data.get(s.dataId).values,[$,F,O]=d,{batchSize:B,filterHeight:W,filterWidth:G,inChannels:q,inHeight:H,inWidth:K,outChannels:X,outHeight:ne,outWidth:J,strideHeight:se,strideWidth:Q}=h,oe=W-1-h.padInfo.top,ae=G-1-h.padInfo.left,pe=X/q;for(let ue=0;ue<B;++ue)for(let ge=0;ge<q;++ge)for(let ye=0;ye<H;++ye){let ke=ye-oe,Se=Math.max(0,Math.ceil(ke/se)),Ce=Math.min(ne,(W+ke)/se);for(let Re=0;Re<K;++Re){let Ke=Re-ae,Ve=Math.max(0,Math.ceil(Ke/Q)),rt=Math.min(J,(G+Ke)/Q),ot=0;for(let He=Se;He<Ce;++He){let ct=He*se-ke;for(let pt=Ve;pt<rt;++pt){let Gn=pt*Q-Ke,Je=x*ue+N*He+T*pt,_n=$*(W-1-ct)+F*(G-1-Gn)+O*ge;for(let Qt=0;Qt<pe;++Qt){let En=ge*pe+Qt,na=v[Je+En],xn=C[_n+Qt];ot+=na*xn}}}f[g*ue+y*ye+b*Re+ge]=ot}}return n.makeTensorInfo(m.shape,m.dtype,m.values)}var mH={kernelName:_d,backendName:"cpu",kernelFunc:hH};function fH(e){let{inputs:t,backend:n}=e,{x:a}=t,r=k.sizeFromShape(a.shape),s=n.data.get(a.dataId).values,i=Le([r,r],a.dtype),o=i.values;for(let c=0;c<s.length;c++)o[c*r+c]=s[c];let l=[...a.shape,...a.shape];return n.makeTensorInfo(l,i.dtype,i.values)}var gH={kernelName:Ed,backendName:"cpu",kernelFunc:fH},yH={kernelName:hc,backendName:"cpu",kernelFunc:({inputs:e,backend:t,attrs:n})=>{let{x:a,filter:r}=e,{strides:s,pad:i,dilations:o}=n,l=t,c=l.data.get(a.dataId).values,u=a.shape.length,p=l.data.get(r.dataId).values,d=r.shape.length,{batchSize:h,inHeight:m,inWidth:f,inChannels:g,outHeight:y,outWidth:b,padInfo:v,strideHeight:x,strideWidth:N,filterHeight:T,filterWidth:C,dilationHeight:$,dilationWidth:F,outShape:O}=_.computeDilation2DInfo(a.shape,r.shape,s,i,"NHWC",o),B=k.sizeFromShape(O),W=O.length,G=k.getArrayFromDType(a.dtype,B);for(let q=0;q<h;++q)for(let H=0;H<y;++H){let K=H*x-v.top;for(let X=0;X<b;++X){let ne=X*N-v.left;for(let J=0;J<g;++J){let se=Number.MIN_SAFE_INTEGER;for(let oe=0;oe<T;++oe){let ae=K+oe*$;if(ae>=0&&ae<m)for(let pe=0;pe<C;++pe){let ue=ne+pe*F;if(ue>=0&&ue<f){let ge=k.locToIndex([q,ae,ue,J],u,k.computeStrides(a.shape)),ye=k.locToIndex([oe,pe,J],d,k.computeStrides(r.shape)),ke=c[ge]+p[ye];ke>se&&(se=ke)}}}let Q=k.locToIndex([q,H,X,J],W,k.computeStrides(O));G[Q]=se}}}return{dataId:l.write(k.toTypedArray(G,a.dtype),O,a.dtype),shape:O,dtype:a.dtype}}},bH={kernelName:Fd,backendName:"cpu",kernelFunc:({inputs:e,backend:t,attrs:n})=>{let{x:a,filter:r,dy:s}=e,{strides:i,pad:o,dilations:l}=n,c=t,u=k.toNestedArray(a.shape,c.data.get(a.dataId).values),p=k.toNestedArray(r.shape,c.data.get(r.dataId).values),{batchSize:d,inHeight:h,inWidth:m,inChannels:f,outHeight:g,outWidth:y,padInfo:b,strideHeight:v,strideWidth:x,filterHeight:N,filterWidth:T,dilationHeight:C,dilationWidth:$,outShape:F}=_.computeDilation2DInfo(a.shape,r.shape,i,o,"NHWC",l);k.assert(s.rank===F.length,()=>`Error in ${Fd}, dy must have the same rank as output ${F.length}, but got ${s.rank}`);let O=k.toNestedArray(F,c.data.get(s.dataId).values),B=k.makeZerosNestedTypedArray(r.shape,r.dtype);for(let W=0;W<d;++W)for(let G=0;G<g;++G){let q=G*v-b.top;for(let H=0;H<y;++H){let K=H*x-b.left;for(let X=0;X<f;++X){let ne=Number.MIN_SAFE_INTEGER,J=0,se=0;for(let Q=0;Q<N;++Q){let oe=q+Q*C;if(oe>=0&&oe<h)for(let ae=0;ae<T;++ae){let pe=K+ae*$;if(pe>=0&&pe<m){let ue=u[W][oe][pe][X]+p[Q][ae][X];ue>ne&&(ne=ue,J=Q,se=ae)}}}B[J][se][X]+=O[W][G][H][X]}}}return{dataId:c.write(k.toTypedArray(B,a.dtype),r.shape,r.dtype),shape:r.shape,dtype:r.dtype}}},xH={kernelName:Ad,backendName:"cpu",kernelFunc:({inputs:e,backend:t,attrs:n})=>{let{x:a,filter:r,dy:s}=e,{strides:i,pad:o,dilations:l}=n,c=t,u=k.toNestedArray(a.shape,c.data.get(a.dataId).values),p=k.toNestedArray(r.shape,c.data.get(r.dataId).values),{batchSize:d,inHeight:h,inWidth:m,inChannels:f,outHeight:g,outWidth:y,padInfo:b,strideHeight:v,strideWidth:x,filterHeight:N,filterWidth:T,dilationHeight:C,dilationWidth:$,outShape:F}=_.computeDilation2DInfo(a.shape,r.shape,i,o,"NHWC",l);k.assert(s.rank===F.length,()=>`Error in ${Ad}, dy must have the same rank as output ${F.length}, but got ${s.rank}`);let O=k.toNestedArray(F,c.data.get(s.dataId).values),B=k.makeZerosNestedTypedArray(a.shape,a.dtype);for(let W=0;W<d;++W)for(let G=0;G<g;++G){let q=G*v-b.top;for(let H=0;H<y;++H){let K=H*x-b.left;for(let X=0;X<f;++X){let ne=Number.MIN_SAFE_INTEGER,J=q<0?0:q,se=K<0?0:K;for(let Q=0;Q<N;++Q){let oe=q+Q*C;if(oe>=0&&oe<h)for(let ae=0;ae<T;++ae){let pe=K+ae*$;if(pe>=0&&pe<m){let ue=u[W][oe][pe][X]+p[Q][ae][X];ue>ne&&(ne=ue,J=oe,se=pe)}}}B[W][J][se][X]+=O[W][G][H][X]}}}return{dataId:c.write(k.toTypedArray(B,a.dtype),a.shape,a.dtype),shape:a.shape,dtype:a.dtype}}};function vH(e){let{inputs:t,backend:n}=e,{dy:a,y:r}=t;we([a,r],"eluGrad");let s=new Float32Array(k.sizeFromShape(r.shape)),i=n.data.get(r.dataId).values,o=n.data.get(a.dataId).values;for(let l=0;l<i.length;++l){let c=i[l];c>=1?s[l]=o[l]:s[l]=o[l]*(c+1)}return n.makeTensorInfo(r.shape,"float32",s)}var wH={kernelName:$d,backendName:"cpu",kernelFunc:vH},kH=Rt((e,t)=>e===t?1:0),kT=Zt(Wo,kH,null,"bool"),IH={kernelName:Wo,backendName:"cpu",kernelFunc:kT},NH=_.ERF_P,TH=_.ERF_A1,SH=_.ERF_A2,CH=_.ERF_A3,_H=_.ERF_A4,EH=_.ERF_A5,AH=it(Bo,e=>{let t=Math.sign(e),n=Math.abs(e),a=1/(1+NH*n);return t*(1-((((EH*a+_H)*a+CH)*a+SH)*a+TH)*a*Math.exp(-n*n))}),FH={kernelName:Bo,backendName:"cpu",kernelFunc:AH};function $m(e){let{inputs:t,backend:n,attrs:a}=e,{input:r}=t,{dim:s}=a,i=r.shape.length,o=r.shape.slice(),l=s;return s<0&&(k.assert(-(i+1)<=s,()=>`Axis must be in the interval [${-(i+1)}, ${i}]`),l=i+s+1),o.splice(l,0,1),It({inputs:{x:r},backend:n,attrs:{shape:o}})}var $H={kernelName:Vo,backendName:"cpu",kernelFunc:$m},DH=Rt((e,t)=>e/t),Nv=Zt(Ps,DH),Tv={kernelName:Ps,backendName:"cpu",kernelFunc:Nv};function IT(e,t,n){let a=e.shape,r=a[0],s=a[1],i=n.data.get(e.dataId),o=i.complexTensorInfos.real,l=i.complexTensorInfos.imag,c=[r,s],u=k.sizeFromShape(c),p=k.getTypedArrayFromDType("float32",u),d=k.getTypedArrayFromDType("float32",u);for(let g=0;g<r;g++){let y=Vi({inputs:{x:o},backend:n,attrs:{begin:[g,0],size:[1,s]}}),b=Vi({inputs:{x:l},backend:n,attrs:{begin:[g,0],size:[1,s]}}),v=zn({inputs:{real:y,imag:b},backend:n}),{real:x,imag:N}=RH(v,t,n),T=_.mergeRealAndImagArrays(x,N);for(let C=0;C<s;C++){let $=_.getComplexWithIndex(T,C);p[g*s+C]=$.real,d[g*s+C]=$.imag}n.disposeIntermediateTensorInfo(y),n.disposeIntermediateTensorInfo(b),n.disposeIntermediateTensorInfo(v)}let h=n.makeTensorInfo(c,"float32",p),m=n.makeTensorInfo(c,"float32",d),f=zn({inputs:{real:h,imag:m},backend:n});return n.disposeIntermediateTensorInfo(h),n.disposeIntermediateTensorInfo(m),f}function RH(e,t,n){let a=k.sizeFromShape(e.shape),r=n.data.get(e.dataId),s=n.data.get(r.complexTensorInfos.real.dataId).values,i=n.data.get(r.complexTensorInfos.imag.dataId).values;if(MH(a)){let o=Sv(s,i,a,t,n),l=[e.shape[0],e.shape[1]];if(t){let c=n.makeTensorInfo(l,"float32",o.real),u=n.makeTensorInfo(l,"float32",o.imag),p=n.makeTensorInfo([],"float32",k.createScalarValue(a,"float32")),d=ar({inputs:{x:p},backend:n}),h=Tv.kernelFunc({inputs:{a:c,b:p},backend:n}),m=Tv.kernelFunc({inputs:{a:u,b:d},backend:n}),f=n.data.get(h.dataId).values,g=n.data.get(m.dataId).values;return n.disposeIntermediateTensorInfo(c),n.disposeIntermediateTensorInfo(u),n.disposeIntermediateTensorInfo(p),n.disposeIntermediateTensorInfo(d),n.disposeIntermediateTensorInfo(h),n.disposeIntermediateTensorInfo(m),{real:f,imag:g}}return o}else{let o=_.mergeRealAndImagArrays(s,i),l=PH(o,a,t);return _.splitRealAndImagArrays(l)}}function MH(e){return(e&e-1)==0}function Sv(e,t,n,a,r){if(n===1)return{real:e,imag:t};let s=_.mergeRealAndImagArrays(e,t),i=n/2,o=_.complexWithEvenIndex(s),l=o.real,c=o.imag,u=[l.length],p=r.makeTensorInfo(u,"float32",l),d=r.makeTensorInfo(u,"float32",c),h=zn({inputs:{real:p,imag:d},backend:r}),m=_.complexWithOddIndex(s),f=m.real,g=m.imag,y=[f.length],b=r.makeTensorInfo(y,"float32",f),v=r.makeTensorInfo(y,"float32",g),x=zn({inputs:{real:b,imag:v},backend:r}),N=Sv(l,c,i,a,r),T=N.real,C=N.imag,$=[T.length],F=r.makeTensorInfo($,"float32",T),O=r.makeTensorInfo($,"float32",C),B=zn({inputs:{real:F,imag:O},backend:r}),W=Sv(f,g,i,a,r),G=W.real,q=W.imag,H=[G.length],K=r.makeTensorInfo(H,"float32",G),X=r.makeTensorInfo(H,"float32",q),ne=zn({inputs:{real:K,imag:X},backend:r}),J=_.exponents(n,a),se=[J.real.length],Q=r.makeTensorInfo(se,"float32",J.real),oe=r.makeTensorInfo(se,"float32",J.imag),ae=zn({inputs:{real:Q,imag:oe},backend:r}),pe=vv({inputs:{a:ae,b:ne},backend:r}),ue=fp({inputs:{a:B,b:pe},backend:r}),ge=wv({inputs:{a:B,b:pe},backend:r}),ye=Wi({inputs:{input:ue},backend:r}),ke=Wi({inputs:{input:ge},backend:r}),Se=au({inputs:{input:ue},backend:r}),Ce=au({inputs:{input:ge},backend:r}),Re=ru({inputs:[ye,ke],backend:r,attrs:{axis:0}}),Ke=ru({inputs:[Se,Ce],backend:r,attrs:{axis:0}}),Ve=r.data.get(Re.dataId).values,rt=r.data.get(Ke.dataId).values;return r.disposeIntermediateTensorInfo(p),r.disposeIntermediateTensorInfo(d),r.disposeIntermediateTensorInfo(h),r.disposeIntermediateTensorInfo(b),r.disposeIntermediateTensorInfo(v),r.disposeIntermediateTensorInfo(x),r.disposeIntermediateTensorInfo(F),r.disposeIntermediateTensorInfo(O),r.disposeIntermediateTensorInfo(B),r.disposeIntermediateTensorInfo(K),r.disposeIntermediateTensorInfo(X),r.disposeIntermediateTensorInfo(ne),r.disposeIntermediateTensorInfo(Q),r.disposeIntermediateTensorInfo(oe),r.disposeIntermediateTensorInfo(ae),r.disposeIntermediateTensorInfo(pe),r.disposeIntermediateTensorInfo(ue),r.disposeIntermediateTensorInfo(ge),r.disposeIntermediateTensorInfo(ye),r.disposeIntermediateTensorInfo(Se),r.disposeIntermediateTensorInfo(ke),r.disposeIntermediateTensorInfo(Ce),r.disposeIntermediateTensorInfo(Re),r.disposeIntermediateTensorInfo(Ke),{real:Ve,imag:rt}}function PH(e,t,n){let a=new Float32Array(t*2);for(let r=0;r<t;r++){let s=0,i=0;for(let o=0;o<t;o++){let l=_.exponent(r*o,t,n),c=_.getComplexWithIndex(e,o);s+=c.real*l.real-c.imag*l.imag,i+=c.real*l.imag+c.imag*l.real}n&&(s/=t,i/=t),_.assignToTypedArray(a,s,i,r)}return a}function OH(e){let{inputs:t,backend:n}=e,{input:a}=t,r=k.sizeFromShape(a.shape),s=a.shape[a.shape.length-1],i=r/s,o=It({inputs:{x:a},backend:n,attrs:{shape:[i,s]}}),l=IT(o,!1,n),c=It({inputs:{x:l},backend:n,attrs:{shape:a.shape}});return n.disposeIntermediateTensorInfo(o),n.disposeIntermediateTensorInfo(l),c}var LH={kernelName:Dd,backendName:"cpu",kernelFunc:OH};function Cv(e){let{backend:t,attrs:n}=e,{shape:a,value:r,dtype:s}=n,i=s||k.inferDtype(r),o=k.getArrayFromDType(i,k.sizeFromShape(a));return zH(o,r,i),t.makeTensorInfo(a,i,o)}var BH={kernelName:mc,backendName:"cpu",kernelFunc:Cv};function zH(e,t,n){e.fill(t)}var WH={kernelName:Go,backendName:"cpu",kernelFunc:({inputs:e,attrs:t,backend:n})=>{let{image:a}=e,r=n,s=k.getTypedArrayFromDType(a.dtype,k.sizeFromShape(a.shape)),[i,o,l,c]=a.shape,u=r.data.get(a.dataId).values;for(let p=0;p<i;p++){let d=p*l*o*c;for(let h=0;h<o;h++){let m=h*(l*c);for(let f=0;f<l;f++){let g=f*c;for(let y=0;y<c;y++){let b=[i,h,f,y][2],v=Math.round(l-b),x=d+m+g+y,N=u[x];if(v>=0&&v<l){let T=v*c,C=d+m+T+y;N=u[C]}s[x]=N}}}}return{dataId:r.write(s,a.shape,a.dtype),shape:a.shape,dtype:a.dtype}}},VH=Rt((e,t)=>Math.floor(e/t)),UH=Zt(zs,VH,null,"int32"),GH={kernelName:zs,backendName:"cpu",kernelFunc:UH};function HH(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,filter:s,bias:i,preluActivationWeights:o}=t,{strides:l,pad:c,dataFormat:u,dilations:p,dimRoundingMode:d,activation:h,leakyreluAlpha:m}=a,f=vT({inputs:{x:r,filter:s},backend:n,attrs:{strides:l,pad:c,dataFormat:u,dilations:p,dimRoundingMode:d}});if(i){let g=f;f=fp({inputs:{a:f,b:i},backend:n}),n.disposeIntermediateTensorInfo(g)}if(h){let g=f;f=kv(n,f,h,o,m),n.disposeIntermediateTensorInfo(g)}return f}var jH={kernelName:yi,backendName:"cpu",kernelFunc:HH};function qH(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,filter:s,bias:i,preluActivationWeights:o}=t,{strides:l,pad:c,dataFormat:u,dilations:p,dimRoundingMode:d,activation:h,leakyreluAlpha:m}=a,f=wT({inputs:{x:r,filter:s},backend:n,attrs:{strides:l,pad:c,dataFormat:u,dilations:p,dimRoundingMode:d}});if(i){let g=f;f=fp({inputs:{a:f,b:i},backend:n}),n.disposeIntermediateTensorInfo(g)}if(h){let g=f;f=kv(n,f,h,o,m),n.disposeIntermediateTensorInfo(g)}return f}var KH={kernelName:bi,backendName:"cpu",kernelFunc:qH};function XH(e){let{inputs:t,backend:n}=e,{params:a,indices:r}=t,s=k.sizeFromShape(a.shape),i=r.shape,o=i[i.length-1],[l,c,u,p]=_.prepareAndValidate(a,r);if(c===0)return n.makeTensorInfo(l,a.dtype,[]);let d=Le([c,u],a.dtype),h=n.data.get(r.dataId).values,m=n.data.get(a.dataId).values;for(let f=0;f<c;f++){let g=[],y=0;for(let b=0;b<o;b++){let v=h[f*o+b];y+=v*p[b],g.push(v)}if(y<0||y>=s/u)throw new Error(`Invalid indices: ${g} does not index into ${a.shape}`);for(let b=0;b<u;b++)d.values[f*u+b]=m[y*u+b]}return n.makeTensorInfo(l,d.dtype,d.values)}var YH={kernelName:jo,backendName:"cpu",kernelFunc:XH};function ZH(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,indices:s}=t,{axis:i,batchDims:o}=a;we([r,s],"gatherV2");let l=o;o==null&&(l=0);let c=k.sizeFromShape(s.shape),u=k.parseAxisParam(i,r.shape)[0],p=_.segment_util.collectGatherOpShapeInfo(r,s,u,l),d=It({inputs:{x:r},backend:n,attrs:{shape:[p.batchSize,p.outerSize,p.dimSize,p.sliceSize]}}),h=It({inputs:{x:s},backend:n,attrs:{shape:[p.batchSize,c/p.batchSize]}}),m=[p.batchSize,p.outerSize,c/p.batchSize,p.sliceSize],f=n.bufferSync(h),g=n.bufferSync(d),y=qN(g,f,m);return n.disposeIntermediateTensorInfo(d),n.disposeIntermediateTensorInfo(h),n.makeTensorInfo(p.outputShape,y.dtype,y.values)}var JH={kernelName:Ho,backendName:"cpu",kernelFunc:ZH},QH=Rt((e,t)=>e>=t?1:0),e6=Zt(Ws,QH,null,"bool"),t6={kernelName:Ws,backendName:"cpu",kernelFunc:e6};function n6(e){let{inputs:t,backend:n}=e,{input:a}=t,r=k.sizeFromShape(a.shape),s=a.shape[a.shape.length-1],i=r/s,o=It({inputs:{x:a},backend:n,attrs:{shape:[i,s]}}),l=IT(o,!0,n),c=It({inputs:{x:l},backend:n,attrs:{shape:a.shape}});return n.disposeIntermediateTensorInfo(o),n.disposeIntermediateTensorInfo(l),c}var a6={kernelName:Rd,backendName:"cpu",kernelFunc:n6},r6=it(Xo,e=>Number.isFinite(e)?1:0,"bool"),s6={kernelName:Xo,backendName:"cpu",kernelFunc:r6},i6=it(Yo,e=>Math.abs(e)===Infinity?1:0,"bool"),o6={kernelName:Yo,backendName:"cpu",kernelFunc:i6},l6=it(Zo,e=>Number.isNaN(e)?1:0,"bool"),u6={kernelName:Zo,backendName:"cpu",kernelFunc:l6},c6=Rt((e,t)=>e<=t?1:0),p6=Zt(Qo,c6,null,"bool"),d6={kernelName:Qo,backendName:"cpu",kernelFunc:p6};function h6(e){let{backend:t,attrs:n}=e,{start:a,stop:r,num:s}=n,i=YN(a,r,s);return t.makeTensorInfo([i.length],"float32",i)}var m6={kernelName:Pd,backendName:"cpu",kernelFunc:h6},f6=it(el,e=>Math.log1p(e)),g6={kernelName:el,backendName:"cpu",kernelFunc:f6},y6=Rt((e,t)=>e&&t),b6=Zt(tl,y6,null,"bool"),x6={kernelName:tl,backendName:"cpu",kernelFunc:b6},v6=it(fc,e=>e?0:1,"bool"),w6={kernelName:fc,backendName:"cpu",kernelFunc:v6},k6=Rt((e,t)=>e||t),I6=Zt(gc,k6,null,"bool"),N6={kernelName:gc,backendName:"cpu",kernelFunc:I6};function T6(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{depthRadius:s,bias:i,alpha:o,beta:l}=a;we(r,"LRN");let c=r.shape[3],u=c-1,p=n.data.get(r.dataId).values,d=k.sizeFromShape(r.shape),h=new Float32Array(d);function m(f){let g=f%c,y=f-g+Math.max(0,g-s),b=f-g+Math.min(g+s,u),v=0;for(;y<=b;y++){let x=p[y];v+=x*x}return v}for(let f=0;f<d;f++){let g=m(f),y=p[f]*Math.pow(i+o*g,-l);h[f]=y}return n.makeTensorInfo(r.shape,r.dtype,h)}var S6={kernelName:yc,backendName:"cpu",kernelFunc:T6};function C6(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,y:s,dy:i}=t,{depthRadius:o,bias:l,alpha:c,beta:u}=a;we(i,"LRNGrad");let p=k.sizeFromShape(i.shape),d=i.shape[3],h=n.data.get(i.dataId).values,m=n.data.get(r.dataId).values,f=n.data.get(s.dataId).values,g=new Float32Array(p),y=p;for(let b=0;b<y;b++){let v=b%d,x=b-v+Math.max(0,v-o),N=b-v+Math.min(d,v+o+1),T=0;for(let C=x;C<N;C++)T+=Math.pow(m[C],2);T=c*T+l;for(let C=x;C<N;C++){let $=-2*c*u*m[C]*f[b]/T;b===C&&($+=Math.pow(T,-u)),$*=h[b],g[C]+=$}}return n.makeTensorInfo(i.shape,r.dtype,g)}var _6={kernelName:Od,backendName:"cpu",kernelFunc:C6};function NT(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{reductionIndices:s,keepDims:i}=a,o=n,l=r.shape,c=l.length,u=k.parseAxisParam(s,l),p=u,d=_.getAxesPermutation(p,c),h=o.data.get(r.dataId).values;if(d!=null){let x=new Array(c);for(let N=0;N<x.length;N++)x[N]=l[d[N]];h=yv(h,l,r.dtype,d,x),p=_.getInnerMostAxes(p.length,c),l=x}we(r,"max"),_.assertAxesAreInnerMostDims("max",p,c);let[m,f]=_.computeOutAndReduceShapes(l,p),g=k.sizeFromShape(f),y=JN(h,g,m,r.dtype),b=o.write(y,m,r.dtype),v=m;return i&&(v=_.expandShapeToKeepDim(m,u)),{dataId:b,shape:v,dtype:r.dtype}}var E6={kernelName:Gs,backendName:"cpu",kernelFunc:NT};function A6(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t;we(r,"maxPool");let{filterSize:s,strides:i,pad:o,dimRoundingMode:l}=a,c=1;k.assert(_.eitherStridesOrDilationsAreOne(i,c),()=>`Error in maxPool: Either strides or dilations must be 1. Got strides ${i} and dilations '${c}'`);let u=_.computePool2DInfo(r.shape,s,i,c,o,l),p;if(u.filterWidth===1&&u.filterHeight===1&&k.arraysEqual(u.inShape,u.outShape))p=ar({inputs:{x:r},backend:n});else{let d=n.data.get(r.dataId).values,h=k.computeStrides(r.shape),m=Iv(d,r.shape,r.dtype,h,u,"max");p=n.makeTensorInfo(u.outShape,r.dtype,m.values)}return p}var F6={kernelName:js,backendName:"cpu",kernelFunc:A6};function $6(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{filterSize:s,strides:i,pad:o,dimRoundingMode:l,dataFormat:c}=a;we(r,"maxPool3d");let u=_.computePool3DInfo(r.shape,s,i,1,o,l,c),p=n.data.get(r.dataId).values,d=xT(p,r.shape,r.dtype,k.computeStrides(r.shape),u,"max");return n.makeTensorInfo(d.shape,"float32",d.values)}var D6={kernelName:bc,backendName:"cpu",kernelFunc:$6};function R6(e){let{inputs:t,backend:n,attrs:a}=e,{dy:r,input:s}=t,{filterSize:i,strides:o,pad:l,dimRoundingMode:c}=a;we([r,s],"maxPool3DGrad");let u=_.computePool3DInfo(s.shape,i,o,1,l,c),p=n.bufferSync(s),d=vG(p,u),h=u.strideDepth,m=u.strideHeight,f=u.strideWidth,g=u.dilationDepth,y=u.dilationHeight,b=u.dilationWidth,v=u.effectiveFilterDepth,x=u.effectiveFilterHeight,N=u.effectiveFilterWidth,T=v-1-u.padInfo.front,C=N-1-u.padInfo.left,$=x-1-u.padInfo.top,F=Le(s.shape,"float32"),O=n.bufferSync(r);for(let B=0;B<u.batchSize;++B)for(let W=0;W<u.inChannels;++W)for(let G=0;G<u.inDepth;++G)for(let q=0;q<u.inHeight;++q)for(let H=0;H<u.inWidth;++H){let K=G-T,X=q-$,ne=H-C,J=0;for(let se=0;se<v;se+=g){let Q=(K+se)/h;if(!(Q<0||Q>=u.outDepth||Math.floor(Q)!==Q))for(let oe=0;oe<x;oe+=y){let ae=(X+oe)/m;if(!(ae<0||ae>=u.outHeight||Math.floor(ae)!==ae))for(let pe=0;pe<N;pe+=b){let ue=(ne+pe)/f;if(ue<0||ue>=u.outWidth||Math.floor(ue)!==ue)continue;let ge=v*x*N-1-d.get(B,Q,ae,ue,W),ye=se*x*N+oe*N+pe,ke=ge===ye?1:0;ke!==0&&(J+=O.get(B,Q,ae,ue,W)*ke)}}}F.set(J,B,G,q,H,W)}return n.makeTensorInfo(F.shape,F.dtype,F.values)}var M6={kernelName:zd,backendName:"cpu",kernelFunc:R6};function P6(e){let{inputs:t,backend:n,attrs:a}=e,{dy:r,input:s,output:i}=t,o=s;we([s,i],"maxPoolGrad");let{filterSize:l,strides:c,pad:u,dimRoundingMode:p}=a,d=_.computePool2DInfo(o.shape,l,c,1,u,p),h=n.data.get(o.dataId).values,m=Le(d.outShape,o.dtype,bT(h,o.shape,o.dtype,d).values),f=d.strideHeight,g=d.strideWidth,y=d.dilationHeight,b=d.dilationWidth,v=d.effectiveFilterHeight,x=d.effectiveFilterWidth,N=x-1-d.padInfo.left,T=v-1-d.padInfo.top,C=Le(o.shape,"float32"),$=n.data.get(r.dataId).values,F=Le(r.shape,"float32",$);for(let O=0;O<d.batchSize;++O)for(let B=0;B<d.inChannels;++B)for(let W=0;W<d.inHeight;++W)for(let G=0;G<d.inWidth;++G){let q=W-T,H=G-N,K=0;for(let X=0;X<v;X+=y){let ne=(q+X)/f;if(!(ne<0||ne>=d.outHeight||Math.floor(ne)!==ne))for(let J=0;J<x;J+=b){let se=(H+J)/g;if(se<0||se>=d.outWidth||Math.floor(se)!==se)continue;let Q=v*x-1-m.get(O,ne,se,B),oe=X*x+J,ae=Q===oe?1:0;ae!==0&&(K+=F.get(O,ne,se,B)*ae)}}C.set(K,O,W,G,B)}return n.makeTensorInfo(C.shape,C.dtype,C.values)}var O6={kernelName:Ld,backendName:"cpu",kernelFunc:P6};function L6(e,t,n,a,r){let s=k.computeStrides(t),i=Iv(e,t,n,s,r,"max"),o=bT(e,t,n,r,!0,a);return[i.values,o.values]}var z6={kernelName:Bd,backendName:"cpu",kernelFunc:({inputs:e,attrs:t,backend:n})=>{let{x:a}=e,{filterSize:r,strides:s,pad:i,includeBatchInIndex:o}=t,l=n;we(a,"MaxPoolWithArgmax");let c=l.data.get(a.dataId).values,u=_.computePool2DInfo(a.shape,r,s,[1,1],i),[p,d]=L6(c,a.shape,a.dtype,o,u),h=l.write(p,u.outShape,a.dtype),m=l.write(d,u.outShape,a.dtype);return[{dataId:h,shape:u.outShape,dtype:a.dtype},{dataId:m,shape:u.outShape,dtype:"int32"}]}};function Dm(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{axis:s,keepDims:i}=a;we(r,"sum");let o;r.dtype==="bool"?o=us({inputs:{x:r},backend:n,attrs:{dtype:"int32"}}):o=ar({inputs:{x:r},backend:n});let l=o.shape.length,c=k.parseAxisParam(s,o.shape),u=_.getAxesPermutation(c,l),p=c,d=o;u!=null&&(d=ga({inputs:{x:o},backend:n,attrs:{perm:u}}),p=_.getInnerMostAxes(p.length,l)),_.assertAxesAreInnerMostDims("sum",p,d.shape.length);let[h,m]=_.computeOutAndReduceShapes(d.shape,p),f=_.upcastType(d.dtype,"int32"),g=Fm(n,h,f),y=k.sizeFromShape(m),b=n.data.get(g.dataId).values,v=n.data.get(d.dataId).values;for(let x=0;x<b.length;++x){let N=x*y,T=0;for(let C=0;C<y;++C)T+=v[N+C];b[x]=T}if(i){let x=_.expandShapeToKeepDim(g.shape,c),N=g;g=It({inputs:{x:g},backend:n,attrs:{shape:x}}),n.disposeIntermediateTensorInfo(N)}return n.disposeIntermediateTensorInfo(o),u!=null&&n.disposeIntermediateTensorInfo(d),g}var B6={kernelName:ci,backendName:"cpu",kernelFunc:Dm};function W6(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{axis:s,keepDims:i}=a,o=k.parseAxisParam(s,r.shape),l=_.computeOutAndReduceShapes(r.shape,o)[1],c=k.sizeFromShape(l),u=[],p=n.makeTensorInfo([],"float32",new Float32Array([c]));u.push(p);let d=us({inputs:{x:r},backend:n,attrs:{dtype:"float32"}});u.push(d);let h=Nv({inputs:{a:d,b:p},backend:n});u.push(h);let m=Dm({inputs:{x:h},backend:n,attrs:{axis:s,keepDims:i}});return u.forEach(f=>n.disposeIntermediateTensorInfo(f)),m}var V6={kernelName:qs,backendName:"cpu",kernelFunc:W6};function U6(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{axis:s,keepDims:i}=a;we(r,"min");let o=k.parseAxisParam(s,r.shape),l=o,c=_.getAxesPermutation(l,r.shape.length),u=r;c!=null&&(u=ga({inputs:{x:r},backend:n,attrs:{perm:c}}),l=_.getInnerMostAxes(l.length,r.shape.length)),_.assertAxesAreInnerMostDims("min",l,u.shape.length);let[p,d]=_.computeOutAndReduceShapes(u.shape,l),h=k.sizeFromShape(d),m=k.makeZerosTypedArray(k.sizeFromShape(p),u.dtype),f=n.data.get(u.dataId).values;for(let y=0;y<m.length;++y){let b=y*h,v=f[b];for(let x=0;x<h;++x){let N=f[b+x];N<v&&(v=N)}m[y]=v}c!=null&&n.disposeIntermediateTensorInfo(u);let g=n.makeTensorInfo(p,u.dtype,m);if(i){let y=_.expandShapeToKeepDim(p,o),b=It({inputs:{x:g},backend:n,attrs:{shape:y}});return n.disposeIntermediateTensorInfo(g),b}return g}var G6={kernelName:Ks,backendName:"cpu",kernelFunc:U6};function H6(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{paddings:s,mode:i}=a;we(r,"mirrorPad");let o=s.map((b,v)=>b[0]+r.shape[v]+b[1]),l=s.map(b=>b[0]),c=s.map((b,v)=>b[0]+r.shape[v]),u=i==="reflect"?0:1,p=n.data.get(r.dataId).values,d=r.shape.length,h=k.computeStrides(r.shape),m=k.sizeFromShape(o),f=o.length,g=k.computeStrides(o),y=k.getTypedArrayFromDType(r.dtype,m);for(let b=0;b<m;b++){let v=k.indexToLoc(b,f,g);for(let N=0;N<f;N++)v[N]<l[N]?v[N]=l[N]*2-v[N]-u:v[N]>=c[N]&&(v[N]=(c[N]-1)*2-v[N]+u);v=v.map((N,T)=>N-l[T]);let x=k.locToIndex(v,d,h);y[b]=p[x]}return{dataId:n.write(y,o,r.dtype),shape:o,dtype:r.dtype}}var j6={kernelName:xc,backendName:"cpu",kernelFunc:H6},q6=Rt((e,t)=>{let n=e%t;return e<0&&t<0||e>=0&&t>=0?n:(n+t)%t}),K6=Zt(nl,q6),X6={kernelName:nl,backendName:"cpu",kernelFunc:K6},Y6=Io(dd());function TT(e){let{inputs:t,backend:n,attrs:a}=e,{logits:r}=t,{dim:s}=a,i=r.shape.length,o=s;if(o===-1&&(o=i-1),o!==i-1)throw Error(`Softmax along a non-last dimension is not yet supported. Logits was rank ${i} and dim was ${o}`);let l=k.parseAxisParam([o],r.shape),c=NT({inputs:{x:r},backend:n,attrs:{reductionIndices:l,keepDims:!1}}),u=_.expandShapeToKeepDim(c.shape,l),p=It({inputs:{x:c},backend:n,attrs:{shape:u}}),d=wv({inputs:{a:r,b:p},backend:n}),h=pT({inputs:{x:d},backend:n}),m=Dm({inputs:{x:h},backend:n,attrs:{axis:l,keepDims:!1}}),f=It({inputs:{x:m},backend:n,attrs:{shape:u}}),g=Nv({inputs:{a:h,b:f},backend:n});return n.disposeIntermediateTensorInfo(c),n.disposeIntermediateTensorInfo(p),n.disposeIntermediateTensorInfo(d),n.disposeIntermediateTensorInfo(h),n.disposeIntermediateTensorInfo(m),n.disposeIntermediateTensorInfo(f),g}var Z6={kernelName:pi,backendName:"cpu",kernelFunc:TT};function J6(e){let{inputs:t,backend:n,attrs:a}=e,{logits:r}=t,{numSamples:s,seed:i,normalized:o}=a;we(r,"multinomial");let l=o?r:TT({inputs:{logits:r},backend:n,attrs:{dim:-1}}),c=l.shape[0],u=l.shape[1],p=n.data.get(l.dataId).values,d=[c,s],h=k.makeZerosTypedArray(k.sizeFromShape(d),"int32");for(let m=0;m<c;++m){let f=m*u,g=new Float32Array(u-1);g[0]=p[f];for(let v=1;v<g.length;++v)g[v]=g[v-1]+p[f+v];let y=Y6.alea(i.toString()),b=m*s;for(let v=0;v<s;++v){let x=y();h[b+v]=g.length;for(let N=0;N<g.length;N++)if(x<g[N]){h[b+v]=N;break}}}return o||n.disposeIntermediateTensorInfo(l),n.makeTensorInfo(d,"int32",h)}var Q6={kernelName:Wd,backendName:"cpu",kernelFunc:J6},ej=Za.nonMaxSuppressionV3Impl;function tj(e){let{inputs:t,backend:n,attrs:a}=e,{boxes:r,scores:s}=t,{maxOutputSize:i,iouThreshold:o,scoreThreshold:l}=a;we(r,"NonMaxSuppression");let c=n.data.get(r.dataId).values,u=n.data.get(s.dataId).values,{selectedIndices:p}=ej(c,u,i,o,l);return n.makeTensorInfo([p.length],"int32",new Int32Array(p))}var nj={kernelName:sl,backendName:"cpu",kernelFunc:tj},aj=Za.nonMaxSuppressionV4Impl;function rj(e){let{inputs:t,backend:n,attrs:a}=e,{boxes:r,scores:s}=t,{maxOutputSize:i,iouThreshold:o,scoreThreshold:l,padToMaxOutputSize:c}=a;we(r,"NonMaxSuppressionPadded");let u=n.data.get(r.dataId).values,p=n.data.get(s.dataId).values,{selectedIndices:d,validOutputs:h}=aj(u,p,i,o,l,c);return[n.makeTensorInfo([d.length],"int32",new Int32Array(d)),n.makeTensorInfo([],"int32",new Int32Array([h]))]}var sj={kernelName:il,backendName:"cpu",kernelFunc:rj},ij=Za.nonMaxSuppressionV5Impl;function oj(e){let{inputs:t,backend:n,attrs:a}=e,{boxes:r,scores:s}=t,{maxOutputSize:i,iouThreshold:o,scoreThreshold:l,softNmsSigma:c}=a;we(r,"NonMaxSuppressionWithScore");let u=n.data.get(r.dataId).values,p=n.data.get(s.dataId).values,d=i,h=o,m=l,f=c,{selectedIndices:g,selectedScores:y}=ij(u,p,d,h,m,f);return[n.makeTensorInfo([g.length],"int32",new Int32Array(g)),n.makeTensorInfo([y.length],"float32",new Float32Array(y))]}var lj={kernelName:ol,backendName:"cpu",kernelFunc:oj};function uj(e){let{inputs:t,backend:n,attrs:a}=e,{indices:r}=t,{depth:s,onValue:i,offValue:o}=a;we(r,"oneHot");let l=k.sizeFromShape(r.shape),c=new Float32Array(l*s);c.fill(o);let u=n.data.get(r.dataId).values;for(let p=0;p<l;++p)u[p]>=0&&u[p]<s&&(c[p*s+u[p]]=i);return n.makeTensorInfo([...r.shape,s],"int32",c)}var cj={kernelName:Zs,backendName:"cpu",kernelFunc:uj};function Rm(e){let{inputs:t,backend:n}=e,{x:a}=t;if(a.dtype==="string")throw new Error("zerosLike is not supported for string tensors");if(a.dtype==="complex64"){let r=Wi({inputs:{input:a},backend:n}),s=Rm({inputs:{x:r},backend:n}),i=au({inputs:{input:a},backend:n}),o=Rm({inputs:{x:i},backend:n}),l=zn({inputs:{real:s,imag:o},backend:n});return n.disposeIntermediateTensorInfo(r),n.disposeIntermediateTensorInfo(s),n.disposeIntermediateTensorInfo(i),n.disposeIntermediateTensorInfo(o),l}else return Cv({backend:n,attrs:{shape:a.shape,value:0,dtype:a.dtype}})}var pj={kernelName:Tl,backendName:"cpu",kernelFunc:Rm};function ST(e){let{inputs:t,backend:n}=e,{x:a}=t;if(a.dtype==="string")throw new Error("onesLike is not supported for string tensors");if(a.dtype==="complex64"){let r=Wi({inputs:{input:a},backend:n}),s=ST({inputs:{x:r},backend:n}),i=au({inputs:{input:a},backend:n}),o=Rm({inputs:{x:i},backend:n}),l=zn({inputs:{real:s,imag:o},backend:n});return n.disposeIntermediateTensorInfo(r),n.disposeIntermediateTensorInfo(s),n.disposeIntermediateTensorInfo(i),n.disposeIntermediateTensorInfo(o),l}else return Cv({backend:n,attrs:{shape:a.shape,value:1,dtype:a.dtype}})}var dj={kernelName:ll,backendName:"cpu",kernelFunc:ST};function CT(e){let{inputs:t,backend:n,attrs:a}=e,{axis:r}=a;if(t.length===1)return $m({inputs:{input:t[0]},backend:n,attrs:{dim:r}});let s=t[0].shape,i=t[0].dtype;t.forEach(u=>{k.assertShapesMatch(s,u.shape,"All tensors passed to stack must have matching shapes"),k.assert(i===u.dtype,()=>"All tensors passed to stack must have matching dtypes")});let o=[],l=t.map(u=>{let p=$m({inputs:{input:u},backend:n,attrs:{dim:r}});return o.push(p),p}),c=ru({inputs:l,backend:n,attrs:{axis:r}});return o.forEach(u=>n.disposeIntermediateTensorInfo(u)),c}var hj={kernelName:ul,backendName:"cpu",kernelFunc:CT};function mj(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{paddings:s,constantValue:i}=a;we(r,"pad");let o=s.map((y,b)=>y[0]+r.shape[b]+y[1]),l=s.map(y=>y[0]),c=n.data.get(r.dataId).values,u=k.sizeFromShape(r.shape),p=r.shape.length,d=k.computeStrides(r.shape),h=k.sizeFromShape(o),m=o.length,f=k.computeStrides(o),g=k.getTypedArrayFromDType(r.dtype,h);i!==0&&g.fill(i);for(let y=0;y<u;y++){let b=k.indexToLoc(y,p,d).map((x,N)=>x+l[N]),v=k.locToIndex(b,m,f);g[v]=c[y]}return{dataId:n.write(g,o,r.dtype),shape:o,dtype:r.dtype}}var _T={kernelName:Js,backendName:"cpu",kernelFunc:mj},fj=Rt((e,t)=>Math.pow(e,t)),gj=Zt(Qs,fj),yj={kernelName:Qs,backendName:"cpu",kernelFunc:gj};function bj(e){let{backend:t,attrs:n}=e,{start:a,stop:r,dtype:s,step:i}=n,o=bv(a,r,i,s);return t.makeTensorInfo([o.length],s,o)}var xj={kernelName:vc,backendName:"cpu",kernelFunc:bj},vj=it(pl,e=>1/e),wj={kernelName:pl,backendName:"cpu",kernelFunc:vj};function kj(e){let{inputs:t,backend:n,attrs:a}=e,{images:r}=t,{alignCorners:s,halfPixelCenters:i,size:o}=a;we(r,"resizeBilinear");let l=k.computeStrides(r.shape),[c,u]=o,[p,d,h,m]=r.shape,f=n.data.get(r.dataId).values,g=new Float32Array(k.sizeFromShape([p,c,u,m])),y=[s&&c>1?d-1:d,s&&u>1?h-1:h],b=[s&&c>1?c-1:c,s&&u>1?u-1:u],v=0,x=y[0]/b[0],N=y[1]/b[1];for(let T=0;T<p;T++)for(let C=0;C<c;C++){let $;i?$=x*(C+.5)-.5:$=x*C;let F=Math.max(0,Math.floor($)),O=$-F,B=Math.min(d-1,Math.ceil($)),W=T*l[0]+F*l[1],G=T*l[0]+B*l[1];for(let q=0;q<u;q++){let H;i?H=N*(q+.5)-.5:H=N*q;let K=Math.max(0,Math.floor(H)),X=H-K,ne=Math.min(h-1,Math.ceil(H)),J=W+K*l[2],se=G+K*l[2],Q=W+ne*l[2],oe=G+ne*l[2];for(let ae=0;ae<m;ae++){let pe=f[J+ae],ue=f[se+ae],ge=f[Q+ae],ye=f[oe+ae],ke=pe+(ge-pe)*X,Se=ue+(ye-ue)*X,Ce=ke+(Se-ke)*O;g[v++]=Ce}}}return n.makeTensorInfo([p,c,u,m],"float32",g)}var Ij={kernelName:ni,backendName:"cpu",kernelFunc:kj};function Nj(e){let{inputs:t,backend:n,attrs:a}=e,{images:r,dy:s}=t,{alignCorners:i}=a;we([s,r],"resizeBilinearGrad");let o=k.computeStrides(r.shape),[l,c,u,p]=r.shape,[,d,h]=s.shape,m=new Float32Array(l*c*u*p),f=[i&&d>1?c-1:c,i&&h>1?u-1:u],g=[i&&d>1?d-1:d,i&&h>1?h-1:h],y=f[0]/g[0],b=f[1]/g[1],v=n.data.get(s.dataId).values,x=0;for(let N=0;N<l;N++){let T=N*o[0];for(let C=0;C<d;C++){let $=C*y,F=Math.floor($),O=Math.min(Math.ceil($),c-1),B=T+F*o[1],W=T+O*o[1],G=$-F,q=1-G;for(let H=0;H<h;H++){let K=H*b,X=Math.floor(K),ne=Math.min(Math.ceil(K),u-1),J=K-X,se=1-J,Q=B+X*o[2],oe=B+ne*o[2],ae=W+X*o[2],pe=W+ne*o[2],ue=q*se,ge=q*J,ye=G*se,ke=G*J;for(let Se=0;Se<p;Se++){let Ce=v[x++];m[Q+Se]+=Ce*ue,m[oe+Se]+=Ce*ge,m[ae+Se]+=Ce*ye,m[pe+Se]+=Ce*ke}}}}return n.makeTensorInfo([l,u,c,p],"float32",m)}var Tj={kernelName:Gd,backendName:"cpu",kernelFunc:Nj};function Sj(e){let{inputs:t,backend:n,attrs:a}=e,{images:r}=t,{alignCorners:s,halfPixelCenters:i,size:o}=a;we(r,"resizeNearestNeighbor");let l=k.computeStrides(r.shape),[c,u]=o,[p,d,h,m]=r.shape,f=n.data.get(r.dataId).values,g=new Float32Array(p*c*u*m),y=[s&&c>1?d-1:d,s&&u>1?h-1:h],b=[s&&c>1?c-1:c,s&&u>1?u-1:u],v=y[0]/b[0],x=y[1]/b[1],N=0;for(let T=0;T<p;T++){let C=T*l[0];for(let $=0;$<c;$++){let F=i?v*($+.5):v*$,O=Math.min(d-1,s?Math.round(F):Math.floor(F));i&&(O=Math.max(0,O));let B=C+O*l[1];for(let W=0;W<u;W++){let G=i?x*(W+.5):x*W,q=Math.min(h-1,s?Math.round(G):Math.floor(G));i&&(q=Math.max(0,q));let H=B+q*l[2];for(let K=0;K<m;K++){let X=f[H+K];g[N++]=X}}}}return n.makeTensorInfo([p,c,u,m],r.dtype,g)}var Cj={kernelName:wc,backendName:"cpu",kernelFunc:Sj};function _j(e){let{inputs:t,backend:n,attrs:a}=e,{images:r,dy:s}=t,{alignCorners:i}=a;we([s,r],"resizeNearestNeighborGrad");let o=k.computeStrides(r.shape),l=k.computeStrides(s.shape),[c,u,p,d]=r.shape,[,h,m]=s.shape,f=new Float32Array(c*u*p*d),g=n.data.get(s.dataId).values,y=[i&&h>1?u-1:u,i&&m>1?p-1:p],b=[i&&h>1?h-1:h,i&&m>1?m-1:m],v=y[0]/b[0],x=y[1]/b[1],N=1/v,T=1/x,C=Math.ceil(N)*2+2,$=Math.ceil(T)*2+2;for(let F=0;F<c;F++){let O=F*o[0];for(let B=0;B<u;B++){let W=O+B*o[1],G=Math.floor(B*N),q=Math.floor(G-C/2);for(let H=0;H<p;H++){let K=W+H*o[2],X=Math.floor(H*T),ne=Math.floor(X-$/2);for(let J=0;J<d;J++){let se=0;for(let Q=0;Q<C;Q++){let oe=Q+q;if(oe<0||oe>=h)continue;let ae=O+oe*l[1],pe=oe*v,ue=Math.min(u-1,i?Math.round(pe):Math.floor(pe));if(B===ue)for(let ge=0;ge<$;ge++){let ye=ge+ne;if(ye<0||ye>=m)continue;let ke=ae+ye*l[2],Se=ye*x,Ce=Math.min(p-1,i?Math.round(Se):Math.floor(Se));H===Ce&&(se+=g[ke+J])}}f[K+J]=se}}}}return n.makeTensorInfo(r.shape,r.dtype,f)}var Ej={kernelName:Ud,backendName:"cpu",kernelFunc:_j};function Aj(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{dims:s}=a;we(r,"reverse");let i=r.shape.length,o=k.parseAxisParam(s,r.shape);if(i===0)return ar({inputs:{x:r},backend:n});let l=new Ot(r.shape,r.dtype),c=n.bufferSync(r);for(let u=0;u<l.size;u++){let p=l.indexToLoc(u),d=p.slice();o.forEach(h=>d[h]=r.shape[h]-1-d[h]),l.set(c.get(...d),...p)}return n.makeTensorInfo(l.shape,l.dtype,l.values)}var Fj={kernelName:ri,backendName:"cpu",kernelFunc:Aj},$j={kernelName:Sl,backendName:"cpu",kernelFunc:({inputs:e,attrs:t,backend:n})=>{let{image:a}=e,{radians:r,fillValue:s,center:i}=t,o=n,l=k.getTypedArrayFromDType(a.dtype,k.sizeFromShape(a.shape)),[c,u,p,d]=a.shape,[h,m]=_.getImageCenter(i,u,p),f=255,g=Math.sin(r),y=Math.cos(r),b=o.data.get(a.dataId).values;for(let v=0;v<c;v++){let x=v*p*u*d;for(let N=0;N<u;N++){let T=N*(p*d);for(let C=0;C<p;C++){let $=C*d;for(let F=0;F<d;F++){let O=[c,N,C,F],B=O[2],W=O[1],G=(B-h)*y-(W-m)*g,q=(B-h)*g+(W-m)*y;G=Math.round(G+h),q=Math.round(q+m);let H=s;if(typeof s!="number"&&(F===3?H=f:H=s[F]),G>=0&&G<p&&q>=0&&q<u){let X=q*(p*d),ne=G*d,J=x+X+ne+F;H=b[J]}let K=x+T+$+F;l[K]=H}}}}return{dataId:o.write(l,a.shape,a.dtype),shape:a.shape,dtype:a.dtype}}},Dj=it(si,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}),Rj={kernelName:si,backendName:"cpu",kernelFunc:Dj};function ET(e,t,n,a,r,s,i,o,l,c){let u=[a/r,r],p=e.values,d=t.values;if(a===0)return Le(n,t.dtype);let h=Le(u,t.dtype);h.values.fill(l);for(let m=0;m<s;m++){let f=[],g=0;for(let y=0;y<i;y++){let b=p[m*i+y];f.push(b),g+=b*o[y]}if(g<0||g>=a/r)throw new Error(`Invalid indices: ${f} does not index into ${n}`);for(let y=0;y<r;y++)c?h.values[g*r+y]+=d[m*r+y]:h.values[g*r+y]=t.rank===0?d[0]:d[m*r+y]}return h}function Mj(e){let{inputs:t,backend:n,attrs:a}=e,{indices:r,updates:s}=t,{shape:i}=a,{sliceRank:o,numUpdates:l,sliceSize:c,strides:u,outputSize:p}=_.calculateShapes(s,r,i),d=!0,h=n.bufferSync(r),m=n.bufferSync(s),f=ET(h,m,i,p,c,l,o,u,0,d);return n.makeTensorInfo(i,f.dtype,f.values)}var Pj={kernelName:hl,backendName:"cpu",kernelFunc:Mj};function Oj(e){let{inputs:t,backend:n}=e,{condition:a,t:r,e:s}=t;we([a,r,s],"select");let i=a.shape.length,o=n.data.get(a.dataId).values,l=n.data.get(r.dataId).values,c=n.data.get(s.dataId).values,u=la(r.dtype,s.dtype),p=k.makeZerosTypedArray(k.sizeFromShape(r.shape),u),d=0,h=i===0||i>1||r.shape.length===1?1:k.sizeFromShape(r.shape.slice(1));for(let m=0;m<o.length;m++)for(let f=0;f<h;f++)o[m]===1?p[d++]=l[m]:p[d++]=c[m];return n.makeTensorInfo(r.shape,u,p)}var Lj={kernelName:ml,backendName:"cpu",kernelFunc:Oj},zj=_.SELU_SCALEALPHA,Bj=_.SELU_SCALE,Wj=it(fl,e=>e>=0?Bj*e:zj*(Math.exp(e)-1)),Vj={kernelName:fl,backendName:"cpu",kernelFunc:Wj},Uj=it(li,e=>1/(1+Math.exp(-e))),Gj={kernelName:li,backendName:"cpu",kernelFunc:Uj},Hj=it(bl,e=>e<0?-1:e>0?1:0),jj={kernelName:bl,backendName:"cpu",kernelFunc:Hj},qj=it(oi,e=>Math.sin(e)),Kj={kernelName:oi,backendName:"cpu",kernelFunc:qj},Xj=it(yl,e=>Math.sinh(e)),Yj={kernelName:yl,backendName:"cpu",kernelFunc:Xj},Zj=11920928955078125e-23,AT=Math.log(Zj)+2,Jj=it(xl,e=>{let t=e>-AT,n=e<AT,a=Math.exp(e),r;return n?r=a:t?r=e:r=Math.log(1+a),r}),Qj={kernelName:xl,backendName:"cpu",kernelFunc:Jj};function eq(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{blockShape:s,paddings:i}=a;we([r],"spaceToBatchND");let o=k.sizeFromShape(s),l=[[0,0]];l.push(...i);for(let g=1+s.length;g<r.shape.length;++g)l.push([0,0]);let c=_T.kernelFunc({inputs:{x:r},backend:n,attrs:{paddings:l,constantValue:0}}),u=_.getReshaped(c.shape,s,o,!1),p=_.getPermuted(u.length,s.length,!1),d=_.getReshapedPermuted(c.shape,s,o,!1),h=It({inputs:{x:c},backend:n,attrs:{shape:u}}),m=ga({inputs:{x:h},backend:n,attrs:{perm:p}}),f=It({inputs:{x:m},backend:n,attrs:{shape:d}});return n.disposeIntermediateTensorInfo(c),n.disposeIntermediateTensorInfo(h),n.disposeIntermediateTensorInfo(m),f}var tq={kernelName:kc,backendName:"cpu",kernelFunc:eq};function nq(e){let{inputs:t,backend:n,attrs:a}=e,{sparseIndices:r,sparseValues:s,defaultValue:i}=t,{outputShape:o}=a,{sliceRank:l,numUpdates:c,sliceSize:u,strides:p,outputSize:d}=_.calculateShapes(s,r,o),h=!1,m=n.bufferSync(r),f=n.bufferSync(s),g=n.data.get(i.dataId).values[0],y=ET(m,f,o,d,u,c,l,p,g,h);return n.makeTensorInfo(o,y.dtype,y.values)}var aq={kernelName:Hd,backendName:"cpu",kernelFunc:nq};function rq(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{numOrSizeSplits:s,axis:i}=a,o=k.parseAxisParam(i,r.shape)[0],l=_.prepareSplitSize(r,s,o),c=new Array(r.shape.length).fill(0),u=r.shape.slice();return l.map(p=>{let d=[...u];d[o]=p;let h=Vi({inputs:{x:r},backend:n,attrs:{begin:c,size:d}});return c[o]+=p,h})}var sq={kernelName:vl,backendName:"cpu",kernelFunc:rq},iq=it(ui,e=>Math.sqrt(e)),oq={kernelName:ui,backendName:"cpu",kernelFunc:iq},lq={kernelName:Ic,backendName:"cpu",kernelFunc:({inputs:e,backend:t})=>{let{x:n}=e,a=t;we(n,"square");let r=a.data.get(n.dataId).values,s=new Float32Array(r.length);for(let i=0;i<r.length;++i){let o=r[i];s[i]=o*o}return{dataId:a.write(s,n.shape,n.dtype),shape:n.shape,dtype:n.dtype}}},uq=it(Wr,(e,t)=>{let n=t;return isNaN(e)?NaN:e>0?1:n.alpha}),cq={kernelName:Wr,backendName:"cpu",kernelFunc:uq};function pq(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{begin:s,end:i,strides:o,beginMask:l,endMask:c,ellipsisMask:u,newAxisMask:p,shrinkAxisMask:d}=a;we(r,"stridedSlice");let{nonStrided:h,$begin:m,$strides:f,size:g,newShape:y,outShape:b}=cn.sliceInfo(r.shape,s,i,o,l,c,u,p,d),v=It({inputs:{x:r},backend:n,attrs:{shape:y}}),x;if(h){let T=Vi({inputs:{x:v},backend:n,attrs:{begin:m,size:g}});x=It({inputs:{x:T},backend:n,attrs:{shape:b}}),n.disposeIntermediateTensorInfo(T)}else if(b.some(T=>T===0))x=n.makeTensorInfo(b,r.dtype,[]);else{let T=n.bufferSync(v),C=iT(b,T,f,m);x=n.makeTensorInfo(C.shape,C.dtype,C.values)}let N=It({inputs:{x},backend:n,attrs:{shape:b}});return n.disposeIntermediateTensorInfo(v),n.disposeIntermediateTensorInfo(x),N}var dq={kernelName:wl,backendName:"cpu",kernelFunc:pq},hq=it(kl,e=>Math.tan(e)),mq={kernelName:kl,backendName:"cpu",kernelFunc:hq},fq=it(mi,e=>Math.tanh(e)),gq={kernelName:mi,backendName:"cpu",kernelFunc:fq};function yq(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{reps:s}=a;we(r,"tile");let i=lT(n.bufferSync(r),s);return n.makeTensorInfo(i.shape,i.dtype,i.values)}var bq={kernelName:Br,backendName:"cpu",kernelFunc:yq};function xq(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{k:s,sorted:i}=a;we(r,"topk");let o=n.data.get(r.dataId).values,[l,c]=uT(o,r.shape,r.dtype,s,i);return[n.makeTensorInfo(l.shape,l.dtype,l.values),n.makeTensorInfo(c.shape,c.dtype,c.values)]}var vq={kernelName:Il,backendName:"cpu",kernelFunc:xq};function wq(e){let{inputs:t,attrs:n,backend:a}=e,{axis:r}=n,{x:s}=t;we(s,"unique");let i=a.data.get(s.dataId).values,{outputValues:o,outputShape:l,indices:c}=cT(i,r,s.shape,s.dtype);return[a.makeTensorInfo(l,s.dtype,o),a.makeTensorInfo([c.length],"int32",c)]}var kq={kernelName:jd,backendName:"cpu",kernelFunc:wq};function Iq(e){let{inputs:t,backend:n,attrs:a}=e,{value:r}=t,{axis:s}=a;s<0&&(s+=r.shape.length);let i=r.shape.length,o=r.shape[s],l=new Array(i-1),c=0;for(let h=0;h<i;h++)h!==s&&(l[c++]=r.shape[h]);let u=new Array(i).fill(0),p=r.shape.slice();p[s]=1;let d=new Array(o);for(let h=0;h<d.length;h++){u[s]=h;let m=Vi({inputs:{x:r},backend:n,attrs:{begin:u,size:p}});d[h]=It({inputs:{x:m},backend:n,attrs:{shape:l}}),n.disposeIntermediateTensorInfo(m)}return d}var Nq={kernelName:Nl,backendName:"cpu",kernelFunc:Iq};function Tq(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,segmentIds:s}=t,{numSegments:i}=a;we(r,"unsortedSegmentSum");let o=r.shape.length,l=s.shape.length,c=[],u=[],p=o-l,d=s;for(let m=0;m<p;++m){let f=$m({inputs:{input:d},backend:n,attrs:{dim:m+1}});d=f,u.push(f)}for(let m=0;m<i;++m){let f=k.createScalarValue(m,"int32"),g=n.makeTensorInfo([],"int32",f),y=kT({inputs:{a:g,b:d},backend:n}),b=us({inputs:{x:y},backend:n,attrs:{dtype:"float32"}}),v=vv({inputs:{a:b,b:r},backend:n}),x=Dm({inputs:{x:v},backend:n,attrs:{axis:0,keepDims:!1}});c.push(x),u.push(g),u.push(y),u.push(b),u.push(v),u.push(x)}let h=CT({inputs:c,backend:n,attrs:{axis:0}});return u.forEach(m=>n.disposeIntermediateTensorInfo(m)),h}var Sq={kernelName:Nc,backendName:"cpu",kernelFunc:Tq},Cq=[KU,JV,YU,JU,rU,eG,nG,rG,iG,lG,cG,dG,mG,yG,xG,kG,NG,SG,_G,jU,AG,$G,RG,nU,iU,PG,QV,LG,BG,UG,HG,WG,XG,ZG,qG,QG,tH,aH,sH,oH,uH,cH,dH,mH,gH,yH,xH,bH,Tv,zU,wH,IH,FH,oU,$H,uU,LH,BH,WH,pU,GH,jH,KH,YH,JH,hU,t6,eU,a6,zG,s6,o6,u6,BU,fU,d6,m6,yU,g6,x6,w6,N6,S6,_6,xU,F6,D6,M6,O6,z6,E6,V6,G6,wU,j6,X6,Q6,IU,TU,nj,sj,lj,CU,cj,dj,hj,_T,yj,VU,AU,xj,tU,wj,UU,GU,HU,Ij,Tj,Cj,Ej,Fj,$j,Rj,$U,Pj,Lj,Vj,Gj,jj,Kj,Yj,DU,Z6,Qj,tq,aq,sq,oq,lq,MU,cq,dq,OU,B6,mq,gq,bq,vq,_U,kq,Nq,Sq,pj];for(let e of Cq)Sc(e);var Ui={},_v={alpha:!1,antialias:!1,premultipliedAlpha:!1,preserveDrawingBuffer:!1,depth:!1,stencil:!1,failIfMajorPerformanceCaveat:!0};function _q(e,t){Ui[e]=t}function rr(e){if(!(e in Ui)){let n=Eq(e);if(n!==null)Ui[e]=n;else return console.log("Could not get context for WebGL version",e),null}let t=Ui[e];return t.isContextLost()?(delete Ui[e],rr(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),Ui[e])}function Aq(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 Eq(e){if(e!==1&&e!==2)throw new Error("Cannot get WebGL rendering context, WebGL is disabled.");let t=Aq(e);return t.addEventListener("webglcontextlost",n=>{n.preventDefault(),delete Ui[e]},!1),e===1?t.getContext("webgl",_v)||t.getContext("experimental-webgl",_v):t.getContext("webgl2",_v)}var gp;(function(e){e[e.DENSE=0]="DENSE",e[e.SHARED_BATCH=1]="SHARED_BATCH"})(gp||(gp={}));var Qn;(function(e){e[e.RENDER=0]="RENDER",e[e.UPLOAD=1]="UPLOAD",e[e.PIXELS=2]="PIXELS",e[e.DOWNLOAD=3]="DOWNLOAD"})(Qn||(Qn={}));var tn;(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"})(tn||(tn={}));function yp(e,t){return[t,e]}function Fq(e,t){return e*t}function bp(e){let t=k.sizeFromShape(e),n=Math.ceil(t/4);return k.sizeToSquarishShape(n)}function su(e,t){return[Math.max(1,Math.ceil(t/2)),Math.max(1,Math.ceil(e/2))]}function $q(e,t){let[n,a]=su(e,t);return n*a*4}function Ev(e,t){let n=e,a,r,s,i,o,l,c,u,p,d;return ee().getNumber("WEBGL_VERSION")===2?(a=n.R32F,r=n.R16F,s=n.RGBA16F,i=n.RGBA32F,o=n.RED,c=4,u=1,p=n.HALF_FLOAT,d=n.FLOAT):(a=e.RGBA,r=e.RGBA,s=e.RGBA,i=n.RGBA,o=e.RGBA,c=4,u=4,p=t!=null?t.HALF_FLOAT_OES:null,d=e.FLOAT),l=e.RGBA,{internalFormatFloat:a,internalFormatHalfFloat:r,internalFormatPackedHalfFloat:s,internalFormatPackedFloat:i,textureFormatFloat:o,downloadTextureFormat:l,downloadUnpackNumChannels:c,defaultNumChannels:u,textureTypeHalfFloat:p,textureTypeFloat:d}}function Ne(e,t){let n=t();return ee().getBool("DEBUG")&&Dq(e),n}function Dq(e){let t=e.getError();if(t!==e.NO_ERROR)throw new Error("WebGL Error: "+Rq(e,t))}var Mq=596e-10,Pq=65504;function Oq(e){return!!(ee().getBool("WEBGL_RENDER_FLOAT32_ENABLED")||e===0||Mq<Math.abs(e)&&Math.abs(e)<Pq)}function Rq(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 Mm(e,t){return Nr(e,()=>e.getExtension(t),'Extension "'+t+'" not supported on this browser.')}function Lq(e,t){let n=Nr(e,()=>e.createShader(e.VERTEX_SHADER),"Unable to create vertex WebGLShader.");if(Ne(e,()=>e.shaderSource(n,t)),Ne(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 Bq(e,t){let n=Nr(e,()=>e.createShader(e.FRAGMENT_SHADER),"Unable to create fragment WebGLShader.");if(Ne(e,()=>e.shaderSource(n,t)),Ne(e,()=>e.compileShader(n)),e.getShaderParameter(n,e.COMPILE_STATUS)===!1)throw zq(t,e.getShaderInfoLog(n)),new Error("Failed to compile fragment shader.");return n}var Wq=/ERROR: [0-9]+:([0-9]+):/g;function zq(e,t){let n=Wq.exec(t);if(n==null){console.log(`Couldn't parse line number in error: ${t}`),console.log(e);return}let a=+n[1],r=e.split(`
|
|
`),s=r.length.toString().length+2,i=r.map((p,d)=>k.rightPad((d+1).toString(),s)+p),o=0;for(let p=0;p<i.length;p++)o=Math.max(i[p].length,o);let l=i.slice(0,a-1),c=i.slice(a-1,a),u=i.slice(a);console.log(l.join(`
|
|
`)),console.log(t.split(`
|
|
`)[0]),console.log(`%c ${k.rightPad(c[0],o)}`,"border:1px solid red; background-color:#e3d2d2; color:#a61717"),console.log(u.join(`
|
|
`))}function Vq(e){return Nr(e,()=>e.createProgram(),"Unable to create WebGLProgram.")}function Uq(e,t){if(Ne(e,()=>e.linkProgram(t)),e.getProgramParameter(t,e.LINK_STATUS)===!1)throw console.log(e.getProgramInfoLog(t)),new Error("Failed to link vertex and fragment shaders.")}function Av(e,t){if(Ne(e,()=>e.validateProgram(t)),e.getProgramParameter(t,e.VALIDATE_STATUS)===!1)throw console.log(e.getProgramInfoLog(t)),new Error("Shader program validation failed.")}function Gq(e,t){let n=Nr(e,()=>e.createBuffer(),"Unable to create WebGLBuffer");return Ne(e,()=>e.bindBuffer(e.ARRAY_BUFFER,n)),Ne(e,()=>e.bufferData(e.ARRAY_BUFFER,t,e.STATIC_DRAW)),n}function Hq(e,t){let n=Nr(e,()=>e.createBuffer(),"Unable to create WebGLBuffer");return Ne(e,()=>e.bindBuffer(e.ELEMENT_ARRAY_BUFFER,n)),Ne(e,()=>e.bufferData(e.ELEMENT_ARRAY_BUFFER,t,e.STATIC_DRAW)),n}function jq(e){return Nr(e,()=>e.createTexture(),"Unable to create WebGLTexture.")}function qq(e,t){let n=ee().getNumber("WEBGL_MAX_TEXTURE_SIZE");if(e<=0||t<=0){let a=`[${e}x${t}]`;throw new Error("Requested texture size "+a+" is invalid.")}if(e>n||t>n){let a=`[${e}x${t}]`,r=`[${n}x${n}]`;throw new Error("Requested texture size "+a+" greater than WebGL maximum on this browser / GPU "+r+".")}}function Kq(e){return Nr(e,()=>e.createFramebuffer(),"Unable to create WebGLFramebuffer.")}function FT(e,t,n,a,r,s,i){let o=e.getAttribLocation(t,n);return o===-1?!1:(Ne(e,()=>e.bindBuffer(e.ARRAY_BUFFER,a)),Ne(e,()=>e.vertexAttribPointer(o,r,e.FLOAT,!1,s,i)),Ne(e,()=>e.enableVertexAttribArray(o)),!0)}function Yq(e,t,n){Xq(e,n),Ne(e,()=>e.activeTexture(e.TEXTURE0+n)),Ne(e,()=>e.bindTexture(e.TEXTURE_2D,t))}function Zq(e,t,n){return Nr(e,()=>e.getUniformLocation(t,n),'uniform "'+n+'" not present in program.')}function Jq(e,t,n){return e.getUniformLocation(t,n)}function Qq(e,t,n,a){Ne(e,()=>Yq(e,t,a)),Ne(e,()=>e.uniform1i(n,a))}function Fv(e,t,n){Ne(e,()=>e.bindFramebuffer(e.FRAMEBUFFER,n)),Ne(e,()=>e.framebufferTexture2D(e.FRAMEBUFFER,e.COLOR_ATTACHMENT0,e.TEXTURE_2D,t,0))}function $T(e,t){Ne(e,()=>e.bindFramebuffer(e.FRAMEBUFFER,t)),Ne(e,()=>e.framebufferTexture2D(e.FRAMEBUFFER,e.COLOR_ATTACHMENT0,e.TEXTURE_2D,null,0))}function Pm(e){let t=e.checkFramebufferStatus(e.FRAMEBUFFER);if(t!==e.FRAMEBUFFER_COMPLETE)throw new Error("Error binding framebuffer: "+e5(e,t))}function e5(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 ${t}`}}function Nr(e,t,n){let a=Ne(e,()=>t());if(a==null)throw new Error(n);return a}function Xq(e,t){let n=e.MAX_COMBINED_TEXTURE_IMAGE_UNITS-1,a=t+e.TEXTURE0;if(a<e.TEXTURE0||a>n){let r=`[gl.TEXTURE0, gl.TEXTURE${n}]`;throw new Error(`textureUnit must be in ${r}.`)}}function iu(e,t=2){return k.sizeFromShape(e.slice(0,e.length-t))}function ou(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 $v(e){let t=[1,1,1];return e.length===0||e.length===1&&e[0]===1||(t=[iu(e),...ou(e)]),t}function t5(e,t=!1){let n=ee().getNumber("WEBGL_MAX_TEXTURE_SIZE");t&&(n=n*2,e=e.map((r,s)=>s>=e.length-2?k.nearestLargerEven(e[s]):e[s]),e.length===1&&(e=[2,e[0]])),e.length!==2&&(e=k.squeezeShape(e).newShape);let a=k.sizeFromShape(e);if(e.length<=1&&a<=n)return[1,a];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=iu(e),s=2,i=2;return e.length&&([s,i]=ou(e)),a=r*(s/2)*(i/2),k.sizeToSquarishShape(a).map(o=>o*2)}return k.sizeToSquarishShape(a)}function Om(e){return e%2==0}function Lm(e,t){if(e=e.slice(-2),t=t.slice(-2),k.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],a=t.slice(-1)[0];if(n===a||Om(n)&&Om(a)&&(e[0]===1||t[0]===1))return!0}return e[1]===t[1]&&Om(e[0])&&Om(t[0])}var Dv,Rv;function n5(e){if(Dv==null){let t=rr(e);Dv=t.getParameter(t.MAX_TEXTURE_SIZE)}return Dv}function a5(e){if(Rv==null){let t=rr(e);Rv=t.getParameter(t.MAX_TEXTURE_IMAGE_UNITS)}return Math.min(16,Rv)}function r5(e){if(e===0)return 0;let t,n=rr(e);return ya(n,"EXT_disjoint_timer_query_webgl2")&&e===2?t=2:ya(n,"EXT_disjoint_timer_query")?t=1:t=0,t}function ya(e,t){return e.getExtension(t)!=null}function DT(e){try{if(rr(e)!=null)return!0}catch(t){return console.log("Error when getting WebGL context: ",t),!1}return!1}function s5(e){if(e===0)return!1;let t=rr(e);if(e===1){if(!ya(t,"OES_texture_float"))return!1}else if(!ya(t,"EXT_color_buffer_float"))return!1;return Mv(t)}function o5(e){if(e===0)return!1;let t=rr(e);if(e===1){if(!ya(t,"OES_texture_float")||!ya(t,"WEBGL_color_buffer_float"))return!1}else{if(ya(t,"EXT_color_buffer_float"))return Mv(t);let n="EXT_color_buffer_half_float";if(ya(t,n)){let a=t.getExtension(n);return i5(t,a)}return!1}return Mv(t)}function Mv(e){let t=Ev(e),n=e.createTexture();e.bindTexture(e.TEXTURE_2D,n);let a=1,r=1;e.texImage2D(e.TEXTURE_2D,0,t.internalFormatFloat,a,r,0,t.textureFormatFloat,t.textureTypeFloat,null);let s=e.createFramebuffer();e.bindFramebuffer(e.FRAMEBUFFER,s),e.framebufferTexture2D(e.FRAMEBUFFER,e.COLOR_ATTACHMENT0,e.TEXTURE_2D,n,0);let i=e.checkFramebufferStatus(e.FRAMEBUFFER)===e.FRAMEBUFFER_COMPLETE;return e.bindTexture(e.TEXTURE_2D,null),e.bindFramebuffer(e.FRAMEBUFFER,null),e.deleteTexture(n),e.deleteFramebuffer(s),i}function i5(e,t){let n=Ev(e,t),a=e.createTexture();e.bindTexture(e.TEXTURE_2D,a);let r=1,s=1;e.texImage2D(e.TEXTURE_2D,0,n.internalFormatHalfFloat,r,s,0,n.textureFormatFloat,n.textureTypeHalfFloat,null);let i=e.createFramebuffer();e.bindFramebuffer(e.FRAMEBUFFER,i),e.framebufferTexture2D(e.FRAMEBUFFER,e.COLOR_ATTACHMENT0,e.TEXTURE_2D,a,0);let o=e.checkFramebufferStatus(e.FRAMEBUFFER)===e.FRAMEBUFFER_COMPLETE;return e.bindTexture(e.TEXTURE_2D,null),e.bindFramebuffer(e.FRAMEBUFFER,null),e.deleteTexture(a),e.deleteFramebuffer(i),o}function l5(e){return e!==2?!1:rr(e).fenceSync!=null}function xp(e,t){Array.isArray(e)||(e=[e]),e.forEach(n=>{n!=null&&k.assert(n.dtype!=="complex64",()=>`${t} does not support complex64 tensors in the WebGL backend.`)})}var Ee=ee();Ee.registerFlag("HAS_WEBGL",()=>Ee.getNumber("WEBGL_VERSION")>0);Ee.registerFlag("WEBGL_VERSION",()=>DT(2)?2:DT(1)?1:0);Ee.registerFlag("WEBGL_CHECK_NUMERICAL_PROBLEMS",()=>!1);Ee.registerFlag("WEBGL_BUFFER_SUPPORTED",()=>Ee.get("WEBGL_VERSION")===2);Ee.registerFlag("WEBGL_CPU_FORWARD",()=>!0);Ee.registerFlag("WEBGL_FORCE_F16_TEXTURES",()=>!1);Ee.registerFlag("WEBGL_PACK",()=>Ee.getBool("HAS_WEBGL"));Ee.registerFlag("WEBGL_PACK_NORMALIZATION",()=>Ee.getBool("WEBGL_PACK"));Ee.registerFlag("WEBGL_PACK_CLIP",()=>Ee.getBool("WEBGL_PACK"));Ee.registerFlag("WEBGL_PACK_DEPTHWISECONV",()=>!1);Ee.registerFlag("WEBGL_PACK_BINARY_OPERATIONS",()=>Ee.getBool("WEBGL_PACK"));Ee.registerFlag("WEBGL_PACK_UNARY_OPERATIONS",()=>Ee.getBool("WEBGL_PACK"));Ee.registerFlag("WEBGL_PACK_ARRAY_OPERATIONS",()=>Ee.getBool("WEBGL_PACK"));Ee.registerFlag("WEBGL_PACK_IMAGE_OPERATIONS",()=>Ee.getBool("WEBGL_PACK"));Ee.registerFlag("WEBGL_PACK_REDUCE",()=>Ee.getBool("WEBGL_PACK"));Ee.registerFlag("WEBGL_LAZILY_UNPACK",()=>Ee.getBool("WEBGL_PACK"));Ee.registerFlag("WEBGL_CONV_IM2COL",()=>Ee.getBool("WEBGL_PACK"));Ee.registerFlag("WEBGL_MAX_TEXTURE_SIZE",()=>n5(Ee.getNumber("WEBGL_VERSION")));Ee.registerFlag("WEBGL_MAX_TEXTURES_IN_SHADER",()=>a5(Ee.getNumber("WEBGL_VERSION")));Ee.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION",()=>{let e=Ee.getNumber("WEBGL_VERSION");return e===0?0:r5(e)});Ee.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE",()=>Ee.getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0&&!eh.isMobile());Ee.registerFlag("WEBGL_RENDER_FLOAT32_CAPABLE",()=>s5(Ee.getNumber("WEBGL_VERSION")));Ee.registerFlag("WEBGL_RENDER_FLOAT32_ENABLED",()=>Ee.getBool("WEBGL_FORCE_F16_TEXTURES")?!1:Ee.getBool("WEBGL_RENDER_FLOAT32_CAPABLE"));Ee.registerFlag("WEBGL_DOWNLOAD_FLOAT_ENABLED",()=>o5(Ee.getNumber("WEBGL_VERSION")));Ee.registerFlag("WEBGL_FENCE_API_ENABLED",()=>l5(Ee.getNumber("WEBGL_VERSION")));Ee.registerFlag("WEBGL_SIZE_UPLOAD_UNIFORM",()=>Ee.getBool("WEBGL_RENDER_FLOAT32_ENABLED")?4:0);Ee.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}.`)});function hn(){let e,t,n,a,r,s,i,o,l,c;return ee().getNumber("WEBGL_VERSION")===2?(e="#version 300 es",t="in",n="out",a="in",r="texture",s="outputColor",i="out vec4 outputColor;",o=`
|
|
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",a="varying",r="texture2D",s="gl_FragColor",i="",o=`
|
|
#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:a,texture2D:r,output:s,defineOutput:i,defineSpecialNaN:o,defineSpecialInf:l,defineRound:c}}function Gi(e,t,n="index"){let a=k.computeStrides(t);return a.map((r,s)=>{let i=`int ${e[s]} = ${n} / ${r}`,o=s===a.length-1?`int ${e[s+1]} = ${n} - ${e[s]} * ${r}`:`index -= ${e[s]} * ${r}`;return`${i}; ${o};`}).join("")}function Pv(e){let t=k.computeStrides(e).map(n=>n.toString());return`
|
|
int getFlatIndex(ivec3 coords) {
|
|
return coords.x * ${t[0]} + coords.y * ${t[1]} + coords.z;
|
|
}
|
|
`}var RT=`
|
|
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;
|
|
}
|
|
`,u5=class{constructor(e){this.variableNames=["A"],this.packedInputs=!1,this.packedOutput=!0,this.outPackingScheme=gp.DENSE;let t=bp(e),n=hn();this.outputShape=e,this.userCode=`
|
|
ivec3 outCoordsFromFlatIndex(int index) {
|
|
${Gi(["r","c","d"],e)}
|
|
return ivec3(r, c, d);
|
|
}
|
|
|
|
void main() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${t[0]}, ${t[1]}));
|
|
int index = 4 * (resTexRC.x * ${t[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);
|
|
}
|
|
|
|
${n.output} = result;
|
|
}
|
|
`}},c5=class{constructor(e){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outPackingScheme=gp.DENSE;let t=bp(e),n=hn();this.outputShape=e,this.userCode=`
|
|
ivec3 outCoordsFromFlatIndex(int index) {
|
|
${Gi(["r","c","d"],e)}
|
|
return ivec3(r, c, d);
|
|
}
|
|
|
|
void main() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${t[0]}, ${t[1]}));
|
|
int index = 4 * (resTexRC.x * ${t[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));
|
|
}
|
|
|
|
${n.output} = result;
|
|
}
|
|
`}},p5=class{constructor(e){this.variableNames=["A"],this.outTexUsage=Qn.DOWNLOAD;let t=hn();this.outputShape=e,this.userCode=`
|
|
${RT}
|
|
|
|
void main() {
|
|
float x = getAAtOutCoords();
|
|
${t.output} = encode_float(x);
|
|
}
|
|
`}},d5=class{constructor(e){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outTexUsage=Qn.DOWNLOAD;let t=hn();this.outputShape=e,this.userCode=`
|
|
${RT}
|
|
|
|
void main() {
|
|
ivec3 coords = getOutputCoords();
|
|
float x = getChannel(getAAtOutCoords(), vec2(coords.y, coords.z));
|
|
${t.output} = encode_float(x);
|
|
}
|
|
`}},h5=class{constructor(e,t,n=!1){this.variableNames=["A"];let a=hn(),[r,s]=t;this.outputShape=e;let i="result";n&&(i="floor(result * 255. + 0.5)"),this.userCode=`
|
|
${Pv(e)}
|
|
|
|
void main() {
|
|
ivec3 coords = getOutputCoords();
|
|
|
|
int flatIndex = getFlatIndex(coords);
|
|
int offset = imod(flatIndex, 4);
|
|
|
|
flatIndex = idiv(flatIndex, 4, 1.);
|
|
|
|
int r = flatIndex / ${s};
|
|
int c = imod(flatIndex, ${s});
|
|
vec2 uv = (vec2(c, r) + halfCR) / vec2(${s}.0, ${r}.0);
|
|
vec4 values = ${a.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];
|
|
}
|
|
|
|
${a.output} = vec4(${i}, 0., 0., 0.);
|
|
}
|
|
`}},m5=class{constructor(e,t,n=!1){this.variableNames=["A"],this.packedInputs=!1,this.packedOutput=!0;let a=hn(),[r,s]=t;this.outputShape=e;let i="",o="result";n&&(o="floor(result * 255. + 0.5)");for(let l=0;l<=1;l++)for(let c=0;c<=1;c++){let u=l*2+c;i+=`
|
|
localCoords = coords;
|
|
if(localCoords[2] + ${c} < ${e[2]}) {
|
|
localCoords[2] += ${c};
|
|
if(localCoords[1] + ${l} < ${e[1]}) {
|
|
localCoords[1] += ${l};
|
|
|
|
flatIndex = getFlatIndex(localCoords);
|
|
offset = imod(flatIndex, 4);
|
|
|
|
flatIndex = idiv(flatIndex, 4, 1.);
|
|
|
|
r = flatIndex / ${s};
|
|
c = imod(flatIndex, ${s});
|
|
uv = (vec2(c, r) + halfCR) / vec2(${s}.0, ${r}.0);
|
|
values = ${a.texture2D}(A, uv);
|
|
|
|
if(offset == 0) {
|
|
result[${u}] = values[0];
|
|
} else if(offset == 1) {
|
|
result[${u}] = values[1];
|
|
} else if(offset == 2) {
|
|
result[${u}] = values[2];
|
|
} else {
|
|
result[${u}] = values[3];
|
|
}
|
|
}
|
|
}
|
|
`}this.userCode=`
|
|
${Pv(e)}
|
|
|
|
void main() {
|
|
ivec3 coords = getOutputCoords();
|
|
|
|
vec4 result = vec4(0.);
|
|
int flatIndex, r, c, offset;
|
|
ivec3 localCoords;
|
|
vec2 uv;
|
|
vec4 values;
|
|
|
|
${i}
|
|
|
|
${a.output} = ${o};
|
|
}
|
|
`}};function f5(e){let t=hn(),n=`${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 Lq(e,n)}function g5(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 Gq(e,t)}function y5(e){let t=new Uint16Array([0,1,2,2,1,3]);return Hq(e,t)}function vp(e,t,n,a,r,s){qq(t,n);let i=jq(e),o=e.TEXTURE_2D;return Ne(e,()=>e.bindTexture(o,i)),Ne(e,()=>e.texParameteri(o,e.TEXTURE_WRAP_S,e.CLAMP_TO_EDGE)),Ne(e,()=>e.texParameteri(o,e.TEXTURE_WRAP_T,e.CLAMP_TO_EDGE)),Ne(e,()=>e.texParameteri(o,e.TEXTURE_MIN_FILTER,e.NEAREST)),Ne(e,()=>e.texParameteri(o,e.TEXTURE_MAG_FILTER,e.NEAREST)),Ne(e,()=>e.texImage2D(o,0,a,t,n,0,r,s,null)),Ne(e,()=>e.bindTexture(e.TEXTURE_2D,null)),i}function MT(e){return e.internalFormatFloat}function b5(e,t,n,a){let[r,s]=yp(t,n);return vp(e,r,s,MT(a),a.textureFormatFloat,e.FLOAT)}function PT(e){return e.internalFormatHalfFloat}function x5(e,t,n,a){let[r,s]=yp(t,n);return vp(e,r,s,PT(a),a.textureFormatFloat,a.textureTypeHalfFloat)}function OT(e){return e.downloadTextureFormat}function v5(e,t,n,a){let[r,s]=yp(t,n);return vp(e,r,s,OT(a),e.RGBA,e.UNSIGNED_BYTE)}function LT(e){return e.internalFormatPackedFloat}function w5(e,t,n,a){let[r,s]=su(t,n);return vp(e,r,s,LT(a),e.RGBA,e.FLOAT)}function zT(e){return e.internalFormatPackedHalfFloat}function k5(e,t,n,a){let[r,s]=su(t,n);return vp(e,r,s,zT(a),e.RGBA,a.textureTypeHalfFloat)}function I5(e,t,n){let a=0,r=3*4,s=3*4+2*4;return Ne(e,()=>e.bindBuffer(e.ARRAY_BUFFER,n)),FT(e,t,"clipSpacePos",n,3,s,a)&&FT(e,t,"uv",n,2,s,r)}function N5(e,t,n,a,r,s){Ne(e,()=>e.bindTexture(e.TEXTURE_2D,t));let i,o,l;r instanceof Uint8Array?(i=new Uint8Array(n*a*4),o=e.UNSIGNED_BYTE,l=e.RGBA):(i=new Float32Array(n*a*4),o=e.FLOAT,l=s.internalFormatPackedFloat),i.set(r),Ne(e,()=>e.texImage2D(e.TEXTURE_2D,0,l,n,a,0,e.RGBA,o,i)),Ne(e,()=>e.bindTexture(e.TEXTURE_2D,null))}function T5(e,t,n){Ne(e,()=>e.bindTexture(e.TEXTURE_2D,t)),n.data instanceof Uint8Array?Ne(e,()=>e.texImage2D(e.TEXTURE_2D,0,e.RGBA,n.width,n.height,0,e.RGBA,e.UNSIGNED_BYTE,n.data)):Ne(e,()=>e.texImage2D(e.TEXTURE_2D,0,e.RGBA,e.RGBA,e.UNSIGNED_BYTE,n)),Ne(e,()=>e.bindTexture(e.TEXTURE_2D,null))}function S5(e,t,n,a){let r=e.createBuffer();Ne(e,()=>e.bindBuffer(e.PIXEL_PACK_BUFFER,r));let s=4*4*t*n;return Ne(e,()=>e.bufferData(e.PIXEL_PACK_BUFFER,s,e.STREAM_READ)),Ne(e,()=>e.readPixels(0,0,n,t,e.RGBA,e.FLOAT,0)),Ne(e,()=>e.bindBuffer(e.PIXEL_PACK_BUFFER,null)),r}function C5(e,t,n){let a=e,r=new Float32Array(n);return a.bindBuffer(a.PIXEL_PACK_BUFFER,t),a.getBufferSubData(a.PIXEL_PACK_BUFFER,0,r),a.bindBuffer(a.PIXEL_PACK_BUFFER,null),r}function _5(e,t,n,a){let[r,s]=yp(t,n),i=4,o=new Uint8Array(Fq(t*n,i));return Ne(e,()=>e.readPixels(0,0,r,s,a.downloadTextureFormat,e.UNSIGNED_BYTE,o)),new Float32Array(o.buffer)}function E5(e,t,n,a,r,s,i,o){let l=e,c=new Float32Array($q(s,i));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 A5(e,t,n){let a=new Float32Array(t*n*4);return Ne(e,()=>e.readPixels(0,0,n,t,e.RGBA,e.FLOAT,a)),a}var $5=class{constructor(e){this.outputTexture=null,this.program=null,this.disposed=!1,this.vertexAttrsAreBound=!1,this.itemsToPoll=[];let t=ee().getNumber("WEBGL_VERSION");e!=null?(this.gl=e,_q(t,e)):this.gl=rr(t);let n="WEBGL_color_buffer_float",a="EXT_color_buffer_half_float";if(ee().getNumber("WEBGL_VERSION")===1){let r="OES_texture_float",s="OES_texture_half_float";if(this.textureFloatExtension=Mm(this.gl,r),ya(this.gl,s))this.textureHalfFloatExtension=Mm(this.gl,s);else if(ee().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),ya(this.gl,a))this.colorBufferHalfFloatExtension=Mm(this.gl,a);else if(ee().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",ya(this.gl,n))this.colorBufferFloatExtension=this.gl.getExtension(n);else if(ya(this.gl,a))this.colorBufferHalfFloatExtension=this.gl.getExtension(a);else throw new Error("GL context does not support color renderable floats");this.vertexBuffer=g5(this.gl),this.indexBuffer=y5(this.gl),this.framebuffer=Kq(this.gl),this.textureConfig=Ev(this.gl,this.textureHalfFloatExtension)}get debug(){return ee().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;Ne(e,()=>e.finish()),Ne(e,()=>e.bindFramebuffer(e.FRAMEBUFFER,null)),Ne(e,()=>e.deleteFramebuffer(this.framebuffer)),Ne(e,()=>e.bindBuffer(e.ARRAY_BUFFER,null)),Ne(e,()=>e.bindBuffer(e.ELEMENT_ARRAY_BUFFER,null)),Ne(e,()=>e.deleteBuffer(this.indexBuffer)),this.disposed=!0}createFloat32MatrixTexture(e,t){return this.throwIfDisposed(),b5(this.gl,e,t,this.textureConfig)}createFloat16MatrixTexture(e,t){return this.throwIfDisposed(),x5(this.gl,e,t,this.textureConfig)}createUnsignedBytesMatrixTexture(e,t){return this.throwIfDisposed(),v5(this.gl,e,t,this.textureConfig)}uploadPixelDataToTexture(e,t){this.throwIfDisposed(),T5(this.gl,e,t)}uploadDenseMatrixToTexture(e,t,n,a){this.throwIfDisposed(),N5(this.gl,e,t,n,a,this.textureConfig)}createFloat16PackedMatrixTexture(e,t){return this.throwIfDisposed(),k5(this.gl,e,t,this.textureConfig)}createPackedMatrixTexture(e,t){return this.throwIfDisposed(),w5(this.gl,e,t,this.textureConfig)}deleteMatrixTexture(e){this.throwIfDisposed(),this.outputTexture===e&&($T(this.gl,this.framebuffer),this.outputTexture=null),Ne(this.gl,()=>this.gl.deleteTexture(e))}downloadByteEncodedFloatMatrixFromOutputTexture(e,t,n){return this.downloadMatrixDriver(e,()=>_5(this.gl,t,n,this.textureConfig))}downloadPackedMatrixFromBuffer(e,t,n,a,r,s){return E5(this.gl,e,t,n,a,r,s,this.textureConfig)}downloadFloat32MatrixFromBuffer(e,t){return C5(this.gl,e,t)}createBufferFromTexture(e,t,n){this.bindTextureToFrameBuffer(e);let a=S5(this.gl,t,n,this.textureConfig);return this.unbindTextureToFrameBuffer(),a}createAndWaitForFence(){let e=this.createFence(this.gl);return this.pollFence(e)}createFence(e){let t,n;if(ee().getBool("WEBGL_FENCE_API_ENABLED")){let a=e,r=a.fenceSync(a.SYNC_GPU_COMMANDS_COMPLETE,0);e.flush(),n=()=>{let s=a.clientWaitSync(r,0,0);return s===a.ALREADY_SIGNALED||s===a.CONDITION_SATISFIED},t=r}else ee().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0?(t=this.beginQuery(),this.endQuery(),n=()=>this.isQueryAvailable(t,ee().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))):n=()=>!0;return{query:t,isFencePassed:n}}downloadMatrixFromPackedTexture(e,t,n){return this.downloadMatrixDriver(e,()=>A5(this.gl,t,n))}createProgram(e){this.throwIfDisposed();let t=this.gl,n=Bq(t,e),a=f5(t),r=Vq(t);return Ne(t,()=>t.attachShader(r,a)),Ne(t,()=>t.attachShader(r,n)),Uq(t,r),this.debug&&Av(t,r),this.vertexAttrsAreBound||(this.setProgram(r),this.vertexAttrsAreBound=I5(t,this.program,this.vertexBuffer)),r}deleteProgram(e){this.throwIfDisposed(),e===this.program&&(this.program=null),e!=null&&Ne(this.gl,()=>this.gl.deleteProgram(e))}setProgram(e){this.throwIfDisposed(),this.program=e,this.program!=null&&this.debug&&Av(this.gl,this.program),Ne(this.gl,()=>this.gl.useProgram(e))}getUniformLocation(e,t,n=!0){return this.throwIfDisposed(),n?Zq(this.gl,e,t):Jq(this.gl,e,t)}getAttributeLocation(e,t){return this.throwIfDisposed(),Ne(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(),Qq(this.gl,e,t,n)}setOutputMatrixTexture(e,t,n){this.setOutputMatrixTextureDriver(e,n,t)}setOutputPackedMatrixTexture(e,t,n){this.throwIfDisposed();let[a,r]=su(t,n);this.setOutputMatrixTextureDriver(e,a,r)}setOutputMatrixWriteRegion(e,t,n,a){this.setOutputMatrixWriteRegionDriver(n,e,a,t)}setOutputPackedMatrixWriteRegion(e,t,n,a){throw new Error("setOutputPackedMatrixWriteRegion not implemented.")}debugValidate(){this.program!=null&&Av(this.gl,this.program),Pm(this.gl)}executeProgram(){this.throwIfDisposed(),this.throwIfNoProgram();let e=this.gl;this.debug&&this.debugValidate(),Ne(e,()=>e.drawElements(e.TRIANGLES,6,e.UNSIGNED_SHORT,0))}blockUntilAllProgramsCompleted(){this.throwIfDisposed(),Ne(this.gl,()=>this.gl.finish())}getQueryTimerExtension(){return this.disjointQueryTimerExtension==null&&(this.disjointQueryTimerExtension=Mm(this.gl,ee().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(ee().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){let n=this.gl,a=this.getQueryTimerExtensionWebGL2(),r=n.createQuery();return n.beginQuery(a.TIME_ELAPSED_EXT,r),r}let e=this.getQueryTimerExtensionWebGL1(),t=e.createQueryEXT();return e.beginQueryEXT(e.TIME_ELAPSED_EXT,t),t}endQuery(){if(ee().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 k.repeatedTry(()=>this.disposed||this.isQueryAvailable(e,ee().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))),this.getQueryTime(e,ee().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,a=this.getQueryTimerExtensionWebGL2(),r=n.getQueryParameter(e,n.QUERY_RESULT_AVAILABLE);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(a.GPU_DISJOINT_EXT)),r&&!this.disjoint}else{let n=this.getQueryTimerExtensionWebGL1(),a=n.getQueryObjectEXT(e,n.QUERY_RESULT_AVAILABLE_EXT);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(n.GPU_DISJOINT_EXT)),a&&!this.disjoint}}pollFence(e){return new Promise(t=>{this.addItemToPoll(()=>e.isFencePassed(),()=>t())})}pollItems(){let e=F5(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)&&k.repeatedTry(()=>(this.pollItems(),this.itemsToPoll.length===0))}bindTextureToFrameBuffer(e){this.throwIfDisposed(),Fv(this.gl,e,this.framebuffer),this.debug&&Pm(this.gl)}unbindTextureToFrameBuffer(){this.outputTexture!=null?(Fv(this.gl,this.outputTexture,this.framebuffer),this.debug&&Pm(this.gl)):$T(this.gl,this.framebuffer)}downloadMatrixDriver(e,t){this.bindTextureToFrameBuffer(e);let n=t();return this.unbindTextureToFrameBuffer(),n}setOutputMatrixTextureDriver(e,t,n){this.throwIfDisposed();let a=this.gl;Fv(a,e,this.framebuffer),this.debug&&Pm(a),this.outputTexture=e,Ne(a,()=>a.viewport(0,0,t,n)),Ne(a,()=>a.scissor(0,0,t,n))}setOutputMatrixWriteRegionDriver(e,t,n,a){this.throwIfDisposed(),Ne(this.gl,()=>this.gl.scissor(e,t,n,a))}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 F5(e){let t=0;for(;t<e.length&&e[t]();++t);return t-1}var{getBroadcastDims:BT}=_;function W5(e,t,n,a){let r=[];e.forEach(h=>{let m=k.sizeFromShape(h.shapeInfo.logicalShape);h.shapeInfo.isUniform?r.push(`uniform float ${h.name}${m>1?`[${m}]`:""};`):(r.push(`uniform sampler2D ${h.name};`),r.push(`uniform int offset${h.name};`))});let s=r.join(`
|
|
`),i=e.map(h=>D5(h,t,a)).join(`
|
|
`),o=t.texShape,l=hn(),c=P5(l),u,p,d=z5(l);return t.isPacked?(u=R5(t.logicalShape,o),p=L5(l)):(u=M5(t.logicalShape,o),p=O5(l)),a&&(d+=B5),[d,c,p,s,u,i,n].join(`
|
|
`)}function lu(e){let t=e.shapeInfo.logicalShape;switch(t.length){case 0:return V5(e);case 1:return U5(e);case 2:return G5(e);case 3:return H5(e);case 4:return j5(e);case 5:return q5(e);case 6:return K5(e);default:throw new Error(`${t.length}-D input sampling is not yet supported`)}}function WT(e){switch(e.shapeInfo.logicalShape.length){case 0:return X5(e);case 1:return Y5(e);case 2:return Z5(e);case 3:return J5(e);default:return Q5(e)}}function D5(e,t,n=!1){let a="";n?a+=WT(e):a+=lu(e);let r=e.shapeInfo.logicalShape,s=t.logicalShape;return r.length<=s.length&&(n?a+=e8(e,t):a+=t8(e,t)),a}function R5(e,t){switch(e.length){case 0:return VT();case 1:return n8(e,t);case 2:return s8(e,t);case 3:return a8(e,t);default:return r8(e,t)}}function M5(e,t){switch(e.length){case 0:return VT();case 1:return i8(e,t);case 2:return p8(e,t);case 3:return o8(e,t);case 4:return l8(e,t);case 5:return u8(e,t);case 6:return c8(e,t);default:throw new Error(`${e.length}-D output sampling is not yet supported`)}}function P5(e){return`
|
|
float sampleTexture(sampler2D textureSampler, vec2 uv) {
|
|
return ${e.texture2D}(textureSampler, uv).r;
|
|
}
|
|
`}function O5(e){return`
|
|
void setOutput(float val) {
|
|
${e.output} = vec4(val, 0, 0, 0);
|
|
}
|
|
`}function L5(e){return`
|
|
void setOutput(vec4 val) {
|
|
${e.output} = val;
|
|
}
|
|
`}function z5(e){return`${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);
|
|
}
|
|
|
|
${d8}
|
|
${h8}
|
|
${m8}
|
|
`}var d8=`
|
|
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);
|
|
}
|
|
`,h8=`
|
|
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);
|
|
}
|
|
`,m8=`
|
|
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);
|
|
}
|
|
`,B5=`
|
|
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 VT(){return`
|
|
int getOutputCoords() {
|
|
return 0;
|
|
}
|
|
`}function n8(e,t){let n=[Math.ceil(t[0]/2),Math.ceil(t[1]/2)];return n[0]===1?`
|
|
int getOutputCoords() {
|
|
return 2 * int(resultUV.x * ${n[1]}.0);
|
|
}
|
|
`:n[1]===1?`
|
|
int getOutputCoords() {
|
|
return 2 * int(resultUV.y * ${n[0]}.0);
|
|
}
|
|
`:`
|
|
int getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${n[0]}, ${n[1]}));
|
|
return 2 * (resTexRC.x * ${n[1]} + resTexRC.y);
|
|
}
|
|
`}function i8(e,t){return t[0]===1?`
|
|
int getOutputCoords() {
|
|
return int(resultUV.x * ${t[1]}.0);
|
|
}
|
|
`:t[1]===1?`
|
|
int getOutputCoords() {
|
|
return int(resultUV.y * ${t[0]}.0);
|
|
}
|
|
`:`
|
|
int getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${t[0]}, ${t[1]}));
|
|
return resTexRC.x * ${t[1]} + resTexRC.y;
|
|
}
|
|
`}function a8(e,t){let n=[Math.ceil(t[0]/2),Math.ceil(t[1]/2)],a=Math.ceil(e[2]/2),r=a*Math.ceil(e[1]/2);return`
|
|
ivec3 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${n[0]}, ${n[1]}));
|
|
int index = resTexRC.x * ${n[1]} + resTexRC.y;
|
|
|
|
int b = index / ${r};
|
|
index -= b * ${r};
|
|
|
|
int r = 2 * (index / ${a});
|
|
int c = imod(index, ${a}) * 2;
|
|
|
|
return ivec3(b, r, c);
|
|
}
|
|
`}function o8(e,t){let n=Gi(["r","c","d"],e);return`
|
|
ivec3 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${t[0]}, ${t[1]}));
|
|
int index = resTexRC.x * ${t[1]} + resTexRC.y;
|
|
${n}
|
|
return ivec3(r, c, d);
|
|
}
|
|
`}function r8(e,t){let n=[Math.ceil(t[0]/2),Math.ceil(t[1]/2)],a=Math.ceil(e[e.length-1]/2),r=a*Math.ceil(e[e.length-2]/2),s=r,i="",o="b, r, c";for(let l=2;l<e.length-1;l++)s*=e[e.length-l-1],i=`
|
|
int b${l} = index / ${s};
|
|
index -= b${l} * ${s};
|
|
`+i,o=`b${l}, `+o;return`
|
|
ivec${e.length} getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${n[0]}, ${n[1]}));
|
|
int index = resTexRC.x * ${n[1]} + resTexRC.y;
|
|
|
|
${i}
|
|
|
|
int b = index / ${r};
|
|
index -= b * ${r};
|
|
|
|
int r = 2 * (index / ${a});
|
|
int c = imod(index, ${a}) * 2;
|
|
|
|
return ivec${e.length}(${o});
|
|
}
|
|
`}function l8(e,t){let n=Gi(["r","c","d","d2"],e);return`
|
|
ivec4 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${t[0]}, ${t[1]}));
|
|
int index = resTexRC.x * ${t[1]} + resTexRC.y;
|
|
${n}
|
|
return ivec4(r, c, d, d2);
|
|
}
|
|
`}function u8(e,t){let n=Gi(["r","c","d","d2","d3"],e);return`
|
|
ivec5 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx * vec2(${t[0]},
|
|
${t[1]}));
|
|
|
|
int index = resTexRC.x * ${t[1]} + resTexRC.y;
|
|
|
|
${n}
|
|
|
|
ivec5 outShape = ivec5(r, c, d, d2, d3);
|
|
return outShape;
|
|
}
|
|
`}function c8(e,t){let n=Gi(["r","c","d","d2","d3","d4"],e);return`
|
|
ivec6 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${t[0]}, ${t[1]}));
|
|
int index = resTexRC.x * ${t[1]} + resTexRC.y;
|
|
|
|
${n}
|
|
|
|
ivec6 result = ivec6(r, c, d, d2, d3, d4);
|
|
return result;
|
|
}
|
|
`}function s8(e,t){let n=[Math.ceil(t[0]/2),Math.ceil(t[1]/2)];if(k.arraysEqual(e,t))return`
|
|
ivec2 getOutputCoords() {
|
|
return 2 * ivec2(resultUV.yx * vec2(${n[0]}, ${n[1]}));
|
|
}
|
|
`;let a=Math.ceil(e[1]/2);return`
|
|
ivec2 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${n[0]}, ${n[1]}));
|
|
|
|
int index = resTexRC.x * ${n[1]} + resTexRC.y;
|
|
int r = 2 * (index / ${a});
|
|
int c = imod(index, ${a}) * 2;
|
|
|
|
return ivec2(r, c);
|
|
}
|
|
`}function p8(e,t){return k.arraysEqual(e,t)?`
|
|
ivec2 getOutputCoords() {
|
|
return ivec2(resultUV.yx * vec2(${t[0]}, ${t[1]}));
|
|
}
|
|
`:e[1]===1?`
|
|
ivec2 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${t[0]}, ${t[1]}));
|
|
int index = resTexRC.x * ${t[1]} + resTexRC.y;
|
|
return ivec2(index, 0);
|
|
}
|
|
`:e[0]===1?`
|
|
ivec2 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${t[0]}, ${t[1]}));
|
|
int index = resTexRC.x * ${t[1]} + resTexRC.y;
|
|
return ivec2(0, index);
|
|
}
|
|
`:`
|
|
ivec2 getOutputCoords() {
|
|
ivec2 resTexRC = ivec2(resultUV.yx *
|
|
vec2(${t[0]}, ${t[1]}));
|
|
int index = resTexRC.x * ${t[1]} + resTexRC.y;
|
|
int r = index / ${e[1]};
|
|
int c = index - r * ${e[1]};
|
|
return ivec2(r, c);
|
|
}
|
|
`}function Hi(e){return`offset${e}`}function X5(e){let t=e.name,n="get"+t.charAt(0).toUpperCase()+t.slice(1),a=hn();return`
|
|
vec4 ${n}() {
|
|
return ${a.texture2D}(${t}, halfCR);
|
|
}
|
|
`}function V5(e){let t=e.name,n="get"+t.charAt(0).toUpperCase()+t.slice(1);if(e.shapeInfo.isUniform)return`float ${n}() {return ${t};}`;let[a,r]=e.shapeInfo.texShape;if(a===1&&r===1)return`
|
|
float ${n}() {
|
|
return sampleTexture(${t}, halfCR);
|
|
}
|
|
`;let[s,i]=e.shapeInfo.texShape,o=Hi(t);return`
|
|
float ${n}() {
|
|
vec2 uv = uvFromFlat(${s}, ${i}, ${o});
|
|
return sampleTexture(${t}, uv);
|
|
}
|
|
`}function Y5(e){let t=e.name,n="get"+t.charAt(0).toUpperCase()+t.slice(1),a=e.shapeInfo.texShape,r=[Math.ceil(a[0]/2),Math.ceil(a[1]/2)],s=hn();return`
|
|
vec4 ${n}(int index) {
|
|
vec2 uv = packedUVfrom1D(
|
|
${r[0]}, ${r[1]}, index);
|
|
return ${s.texture2D}(${t}, uv);
|
|
}
|
|
`}function U5(e){let t=e.name,n="get"+t.charAt(0).toUpperCase()+t.slice(1);if(e.shapeInfo.isUniform)return`
|
|
float ${n}(int index) {
|
|
${uu(e)}
|
|
}
|
|
`;let a=e.shapeInfo.texShape,r=a[0],s=a[1];if(s===1&&r===1)return`
|
|
float ${n}(int index) {
|
|
return sampleTexture(${t}, halfCR);
|
|
}
|
|
`;let i=Hi(t);return s===1?`
|
|
float ${n}(int index) {
|
|
vec2 uv = vec2(0.5, (float(index + ${i}) + 0.5) / ${r}.0);
|
|
return sampleTexture(${t}, uv);
|
|
}
|
|
`:r===1?`
|
|
float ${n}(int index) {
|
|
vec2 uv = vec2((float(index + ${i}) + 0.5) / ${s}.0, 0.5);
|
|
return sampleTexture(${t}, uv);
|
|
}
|
|
`:`
|
|
float ${n}(int index) {
|
|
vec2 uv = uvFromFlat(${r}, ${s}, index + ${i});
|
|
return sampleTexture(${t}, uv);
|
|
}
|
|
`}function Z5(e){let t=e.shapeInfo.logicalShape,n=e.name,a="get"+n.charAt(0).toUpperCase()+n.slice(1),r=e.shapeInfo.texShape,s=r[0],i=r[1],o=hn();if(r!=null&&k.arraysEqual(t,r))return`
|
|
vec4 ${a}(int row, int col) {
|
|
vec2 uv = (vec2(col, row) + halfCR) / vec2(${i}.0, ${s}.0);
|
|
|
|
return ${o.texture2D}(${n}, uv);
|
|
}
|
|
`;let l=[Math.ceil(r[0]/2),Math.ceil(r[1]/2)],c=Math.ceil(t[1]/2);return`
|
|
vec4 ${a}(int row, int col) {
|
|
vec2 uv = packedUVfrom2D(${c}, ${l[0]}, ${l[1]}, row, col);
|
|
return ${o.texture2D}(${n}, uv);
|
|
}
|
|
`}function G5(e){let t=e.shapeInfo.logicalShape,n=e.name,a="get"+n.charAt(0).toUpperCase()+n.slice(1),r=e.shapeInfo.texShape;if(r!=null&&k.arraysEqual(t,r)){let p=r[0],d=r[1];return`
|
|
float ${a}(int row, int col) {
|
|
vec2 uv = (vec2(col, row) + halfCR) / vec2(${d}.0, ${p}.0);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`}let{newShape:s,keptDims:i}=k.squeezeShape(t),o=s;if(o.length<t.length){let p=cu(e,o),d=["row","col"];return`
|
|
${lu(p)}
|
|
float ${a}(int row, int col) {
|
|
return ${a}(${pu(d,i)});
|
|
}
|
|
`}if(e.shapeInfo.isUniform)return`
|
|
float ${a}(int row, int col) {
|
|
int index = round(dot(vec2(row, col), vec2(${t[1]}, 1)));
|
|
${uu(e)}
|
|
}
|
|
`;let l=r[0],c=r[1],u=Hi(n);return c===1?`
|
|
float ${a}(int row, int col) {
|
|
float index = dot(vec3(row, col, ${u}), vec3(${t[1]}, 1, 1));
|
|
vec2 uv = vec2(0.5, (index + 0.5) / ${l}.0);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`:l===1?`
|
|
float ${a}(int row, int col) {
|
|
float index = dot(vec3(row, col, ${u}), vec3(${t[1]}, 1, 1));
|
|
vec2 uv = vec2((index + 0.5) / ${c}.0, 0.5);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`:`
|
|
float ${a}(int row, int col) {
|
|
// Explicitly use integer operations as dot() only works on floats.
|
|
int index = row * ${t[1]} + col + ${u};
|
|
vec2 uv = uvFromFlat(${l}, ${c}, index);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`}function J5(e){let t=e.shapeInfo.logicalShape,n=e.name,a="get"+n.charAt(0).toUpperCase()+n.slice(1),r=e.shapeInfo.texShape,s=[Math.ceil(r[0]/2),Math.ceil(r[1]/2)];if(t[0]===1){let p=t.slice(1),d=[1,2],h=cu(e,p),m=["b","row","col"];return`
|
|
${WT(h)}
|
|
vec4 ${a}(int b, int row, int col) {
|
|
return ${a}(${pu(m,d)});
|
|
}
|
|
`}let i=s[0],o=s[1],l=Math.ceil(t[2]/2),c=l*Math.ceil(t[1]/2),u=hn();return`
|
|
vec4 ${a}(int b, int row, int col) {
|
|
vec2 uv = packedUVfrom3D(
|
|
${i}, ${o}, ${c}, ${l}, b, row, col);
|
|
return ${u.texture2D}(${n}, uv);
|
|
}
|
|
`}function H5(e){let t=e.shapeInfo.logicalShape,n=e.name,a="get"+n.charAt(0).toUpperCase()+n.slice(1),r=t[1]*t[2],s=t[2],{newShape:i,keptDims:o}=k.squeezeShape(t),l=i;if(l.length<t.length){let m=cu(e,l),f=["row","col","depth"];return`
|
|
${lu(m)}
|
|
float ${a}(int row, int col, int depth) {
|
|
return ${a}(${pu(f,o)});
|
|
}
|
|
`}if(e.shapeInfo.isUniform)return`
|
|
float ${a}(int row, int col, int depth) {
|
|
int index = round(dot(vec3(row, col, depth),
|
|
vec3(${r}, ${s}, 1)));
|
|
${uu(e)}
|
|
}
|
|
`;let c=e.shapeInfo.texShape,u=c[0],p=c[1],d=e.shapeInfo.flatOffset;if(p===r&&d==null)return`
|
|
float ${a}(int row, int col, int depth) {
|
|
float texR = float(row);
|
|
float texC = dot(vec2(col, depth), vec2(${s}, 1));
|
|
vec2 uv = (vec2(texC, texR) + halfCR) /
|
|
vec2(${p}.0, ${u}.0);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`;if(p===s&&d==null)return`
|
|
float ${a}(int row, int col, int depth) {
|
|
float texR = dot(vec2(row, col), vec2(${t[1]}, 1));
|
|
float texC = float(depth);
|
|
vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${p}.0, ${u}.0);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`;let h=Hi(n);return`
|
|
float ${a}(int row, int col, int depth) {
|
|
// Explicitly use integer operations as dot() only works on floats.
|
|
int index = row * ${r} + col * ${s} + depth + ${h};
|
|
vec2 uv = uvFromFlat(${u}, ${p}, index);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`}function Q5(e){let t=e.shapeInfo.logicalShape,n=t.length,a=e.name,r="get"+a.charAt(0).toUpperCase()+a.slice(1),s=e.shapeInfo.texShape,i=[Math.ceil(s[0]/2),Math.ceil(s[1]/2)],o=i[0],l=i[1],c=Math.ceil(t[n-1]/2),u=c*Math.ceil(t[n-2]/2),p="int b, int row, int col",d=`b * ${u} + (row / 2) * ${c} + (col / 2)`;for(let m=2;m<n-1;m++)p=`int b${m}, `+p,u*=t[n-m-1],d=`b${m} * ${u} + `+d;let h=hn();return`
|
|
vec4 ${r}(${p}) {
|
|
int index = ${d};
|
|
int texR = index / ${l};
|
|
int texC = index - texR * ${l};
|
|
vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${l}, ${o});
|
|
return ${h.texture2D}(${a}, uv);
|
|
}
|
|
`}function j5(e){let t=e.shapeInfo.logicalShape,n=e.name,a="get"+n.charAt(0).toUpperCase()+n.slice(1),r=t[3],s=t[2]*r,i=t[1]*s,{newShape:o,keptDims:l}=k.squeezeShape(t);if(o.length<t.length){let m=cu(e,o),f=["row","col","depth","depth2"];return`
|
|
${lu(m)}
|
|
float ${a}(int row, int col, int depth, int depth2) {
|
|
return ${a}(${pu(f,l)});
|
|
}
|
|
`}if(e.shapeInfo.isUniform)return`
|
|
float ${a}(int row, int col, int depth, int depth2) {
|
|
int index = round(dot(vec4(row, col, depth, depth2),
|
|
vec4(${i}, ${s}, ${r}, 1)));
|
|
${uu(e)}
|
|
}
|
|
`;let c=e.shapeInfo.flatOffset,u=e.shapeInfo.texShape,p=u[0],d=u[1];if(d===i&&c==null)return`
|
|
float ${a}(int row, int col, int depth, int depth2) {
|
|
float texR = float(row);
|
|
float texC =
|
|
dot(vec3(col, depth, depth2),
|
|
vec3(${s}, ${r}, 1));
|
|
vec2 uv = (vec2(texC, texR) + halfCR) /
|
|
vec2(${d}.0, ${p}.0);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`;if(d===r&&c==null)return`
|
|
float ${a}(int row, int col, int depth, int depth2) {
|
|
float texR = dot(vec3(row, col, depth),
|
|
vec3(${t[1]*t[2]}, ${t[2]}, 1));
|
|
float texC = float(depth2);
|
|
vec2 uv = (vec2(texC, texR) + halfCR) /
|
|
vec2(${d}.0, ${p}.0);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`;let h=Hi(n);return`
|
|
float ${a}(int row, int col, int depth, int depth2) {
|
|
// Explicitly use integer operations as dot() only works on floats.
|
|
int index = row * ${i} + col * ${s} +
|
|
depth * ${r} + depth2;
|
|
vec2 uv = uvFromFlat(${p}, ${d}, index + ${h});
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`}function q5(e){let t=e.shapeInfo.logicalShape,n=e.name,a="get"+n.charAt(0).toUpperCase()+n.slice(1),r=t[4],s=t[3]*r,i=t[2]*s,o=t[1]*i,{newShape:l,keptDims:c}=k.squeezeShape(t);if(l.length<t.length){let f=cu(e,l),g=["row","col","depth","depth2","depth3"];return`
|
|
${lu(f)}
|
|
float ${a}(int row, int col, int depth, int depth2, int depth3) {
|
|
return ${a}(${pu(g,c)});
|
|
}
|
|
`}if(e.shapeInfo.isUniform)return`
|
|
float ${a}(int row, int col, int depth, int depth2, int depth3) {
|
|
float index = dot(
|
|
vec4(row, col, depth, depth2),
|
|
vec4(${o}, ${i}, ${s}, ${r})) +
|
|
depth3;
|
|
${uu(e)}
|
|
}
|
|
`;let u=e.shapeInfo.flatOffset,p=e.shapeInfo.texShape,d=p[0],h=p[1];if(h===o&&u==null)return`
|
|
float ${a}(int row, int col, int depth, int depth2, int depth3) {
|
|
int texR = row;
|
|
float texC = dot(vec4(col, depth, depth2, depth3),
|
|
vec4(${i}, ${s}, ${r}, 1));
|
|
vec2 uv = (vec2(texC, texR) + halfCR) /
|
|
vec2(${h}.0, ${d}.0);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`;if(h===r&&u==null)return`
|
|
float ${a}(int row, int col, int depth, int depth2, int depth3) {
|
|
float texR = dot(
|
|
vec4(row, col, depth, depth2),
|
|
vec4(${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, ${d}.0);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`;let m=Hi(n);return`
|
|
float ${a}(int row, int col, int depth, int depth2, int depth3) {
|
|
// Explicitly use integer operations as dot() only works on floats.
|
|
int index = row * ${o} + col * ${i} + depth * ${s} +
|
|
depth2 * ${r} + depth3 + ${m};
|
|
vec2 uv = uvFromFlat(${d}, ${h}, index);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`}function K5(e){let t=e.shapeInfo.logicalShape,n=e.name,a="get"+n.charAt(0).toUpperCase()+n.slice(1),{newShape:r,keptDims:s}=k.squeezeShape(t);if(r.length<t.length){let g=cu(e,r),y=["row","col","depth","depth2","depth3","depth4"];return`
|
|
${lu(g)}
|
|
float ${a}(int row, int col, int depth,
|
|
int depth2, int depth3, int depth4) {
|
|
return ${a}(${pu(y,s)});
|
|
}
|
|
`}let i=t[5],o=t[4]*i,l=t[3]*o,c=t[2]*l,u=t[1]*c;if(e.shapeInfo.isUniform)return`
|
|
float ${a}(int row, int col, int depth,
|
|
int depth2, int depth3, int depth4) {
|
|
int index = round(dot(
|
|
vec4(row, col, depth, depth2),
|
|
vec4(${u}, ${c}, ${l}, ${o})) +
|
|
dot(
|
|
vec2(depth3, depth4),
|
|
vec2(${i}, 1)));
|
|
${uu(e)}
|
|
}
|
|
`;let p=e.shapeInfo.flatOffset,d=e.shapeInfo.texShape,h=d[0],m=d[1];if(m===u&&p==null)return`
|
|
float ${a}(int row, int col, int depth,
|
|
int depth2, int depth3, int depth4) {
|
|
int texR = row;
|
|
float texC = dot(vec4(col, depth, depth2, depth3),
|
|
vec4(${c}, ${l}, ${o}, ${i})) +
|
|
float(depth4);
|
|
vec2 uv = (vec2(texC, texR) + halfCR) /
|
|
vec2(${m}.0, ${h}.0);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`;if(m===i&&p==null)return`
|
|
float ${a}(int row, int col, int depth,
|
|
int depth2, int depth3, int depth4) {
|
|
float texR = dot(vec4(row, col, depth, depth2),
|
|
vec4(${t[1]*t[2]*t[3]*t[4]},
|
|
${t[2]*t[3]*t[4]},
|
|
${t[3]*t[4]},
|
|
${t[4]})) + float(depth3);
|
|
int texC = depth4;
|
|
vec2 uv = (vec2(texC, texR) + halfCR) /
|
|
vec2(${m}.0, ${h}.0);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`;let f=Hi(n);return`
|
|
float ${a}(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 * ${o} + depth3 * ${i} + depth4 + ${f};
|
|
vec2 uv = uvFromFlat(${h}, ${m}, index);
|
|
return sampleTexture(${n}, uv);
|
|
}
|
|
`}function uu(e){let t=e.name,n=k.sizeFromShape(e.shapeInfo.logicalShape);return n<2?`return ${t};`:`
|
|
for (int i = 0; i < ${n}; i++) {
|
|
if (i == index) {
|
|
return ${t}[i];
|
|
}
|
|
}
|
|
`}function e8(e,t){let n=e.name,a=n.charAt(0).toUpperCase()+n.slice(1),r="get"+a+"AtOutCoords",s=e.shapeInfo.logicalShape.length,i=t.logicalShape.length,o=BT(e.shapeInfo.logicalShape,t.logicalShape),l=ft(i),c=i-s,u,p=["x","y","z","w","u","v"];s===0?u="":i<2&&o.length>=1?u="coords = 0;":u=o.map(g=>`coords.${p[g+c]} = 0;`).join(`
|
|
`);let d="";i<2&&s>0?d="coords":d=e.shapeInfo.logicalShape.map((g,y)=>`coords.${p[y+c]}`).join(", ");let h="return outputValue;",m=k.sizeFromShape(e.shapeInfo.logicalShape)===1,f=k.sizeFromShape(t.logicalShape)===1;if(s===1&&!m&&!f)h=`
|
|
return vec4(outputValue.xy, outputValue.xy);
|
|
`;else if(m&&!f)i===1?h=`
|
|
return vec4(outputValue.x, outputValue.x, 0., 0.);
|
|
`:h=`
|
|
return vec4(outputValue.x);
|
|
`;else if(o.length){let g=s-2,y=s-1;o.indexOf(g)>-1&&o.indexOf(y)>-1?h="return vec4(outputValue.x);":o.indexOf(g)>-1?h="return vec4(outputValue.x, outputValue.y, outputValue.x, outputValue.y);":o.indexOf(y)>-1&&(h="return vec4(outputValue.xx, outputValue.zz);")}return`
|
|
vec4 ${r}() {
|
|
${l} coords = getOutputCoords();
|
|
${u}
|
|
vec4 outputValue = get${a}(${d});
|
|
${h}
|
|
}
|
|
`}function t8(e,t){let n=e.name,a=n.charAt(0).toUpperCase()+n.slice(1),r="get"+a+"AtOutCoords",s=t.texShape,i=e.shapeInfo.texShape,o=e.shapeInfo.logicalShape.length,l=t.logicalShape.length;if(!e.shapeInfo.isUniform&&o===l&&e.shapeInfo.flatOffset==null&&k.arraysEqual(i,s))return`
|
|
float ${r}() {
|
|
return sampleTexture(${n}, resultUV);
|
|
}
|
|
`;let c=ft(l),u=BT(e.shapeInfo.logicalShape,t.logicalShape),p=l-o,d,h=["x","y","z","w","u","v"];o===0?d="":l<2&&u.length>=1?d="coords = 0;":d=u.map(f=>`coords.${h[f+p]} = 0;`).join(`
|
|
`);let m="";return l<2&&o>0?m="coords":m=e.shapeInfo.logicalShape.map((f,g)=>`coords.${h[g+p]}`).join(", "),`
|
|
float ${r}() {
|
|
${c} coords = getOutputCoords();
|
|
${d}
|
|
return get${a}(${m});
|
|
}
|
|
`}function ft(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 ${e} is not yet supported`)}function cu(e,t){let n=JSON.parse(JSON.stringify(e));return n.shapeInfo.logicalShape=t,n}function pu(e,t){return t.map(n=>e[n]).join(", ")}function f8(e,t,n,a){let r=t.userCode,s=n.map((h,m)=>{let f={logicalShape:h.shape,texShape:h.isUniform?null:h.texData.texShape,isUniform:h.isUniform,isPacked:h.isUniform?!1:h.texData.isPacked,flatOffset:null};return h.texData!=null&&h.texData.slice!=null&&h.texData.slice.flatOffset>0&&(f.flatOffset=h.texData.slice.flatOffset),{name:t.variableNames[m],shapeInfo:f}}),i=s.map(h=>h.shapeInfo),o={logicalShape:a.shape,texShape:a.texData.texShape,isUniform:!1,isPacked:a.texData.isPacked,flatOffset:null},l=W5(s,o,r,t.packedInputs),c=e.createProgram(l),u=null,p=e.getUniformLocation(c,"NAN",!1);ee().getNumber("WEBGL_VERSION")===1&&(u=e.getUniformLocation(c,"INFINITY",!1));let d={};for(let h=0;h<t.variableNames.length;h++){let m=t.variableNames[h],f=!1;d[m]=e.getUniformLocation(c,m,f),d[`offset${m}`]=e.getUniformLocation(c,`offset${m}`,f)}return{program:t,source:l,webGLProgram:c,uniformLocations:d,inShapeInfos:i,outShapeInfo:o,infLoc:u,nanLoc:p}}function UT(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,a)=>{let r=n.logicalShape,s=t[a],i=s.shape;if(!k.arraysEqual(r,i))throw Error(`Binary was compiled with different shapes than the current args. Shapes ${r} and ${i} must match`);if(n.isUniform&&s.isUniform)return;let o=n.texShape,l=s.isUniform?null:s.texData.texShape;if(!k.arraysEqual(o,l))throw Error(`Binary was compiled with different texture shapes than the current args. Shape ${o} and ${l} must match`)})}function g8(e,t,n,a,r){UT(t.inShapeInfos,n),UT([t.outShapeInfo],[a]);let s=a.texData.texture,i=a.texData.texShape;a.texData.isPacked?e.setOutputPackedMatrixTexture(s,i[0],i[1]):e.setOutputMatrixTexture(s,i[0],i[1]),e.setProgram(t.webGLProgram),ee().getNumber("WEBGL_VERSION")===1&&t.infLoc!==null&&e.gl.uniform1f(t.infLoc,Infinity),t.nanLoc!==null&&e.gl.uniform1f(t.nanLoc,NaN),n.forEach((o,l)=>{let c=t.program.variableNames[l],u=t.uniformLocations[c],p=t.uniformLocations[`offset${c}`];if(u!=null){if(o.isUniform){if(k.sizeFromShape(o.shape)<2)e.gl.uniform1f(u,o.uniformValues[0]);else{let d=o.uniformValues;d instanceof Float32Array||(d=new Float32Array(d)),e.gl.uniform1fv(u,d)}return}o.texData.slice!=null&&p!=null&&e.gl.uniform1i(p,o.texData.slice.flatOffset),e.setInputMatrixTexture(o.texData.texture,u,l)}}),r!=null&&r(e,t.webGLProgram),e.executeProgram()}function y8(e,t,n){let a="";t.concat(n).forEach(i=>{let o=i.texData!=null&&i.texData.slice!=null&&i.texData.slice.flatOffset>0,l=i.isUniform?"uniform":i.texData.texShape;a+=`${i.shape}_${l}_${o}`});let r=e.userCode,s=e.constructor.name;return s+="_"+a+"_"+r,s}var{addImpl:b8,bincountImpl:GT,bincountReduceImpl:x8,ceilImpl:v8,concatImpl:w8,expImpl:k8,expm1Impl:I8,floorImpl:N8,gatherV2Impl:T8,greaterImpl:S8,lessImpl:C8,linSpaceImpl:_8,logImpl:E8,maxImpl:A8,maximumImpl:F8,minimumImpl:$8,multiplyImpl:D8,negImpl:R8,prodImpl:M8,rangeImpl:P8,rsqrtImpl:O8,simpleAbsImpl:HT,sliceImpl:L8,stridedSliceImpl:z8,subImpl:B8,tileImpl:W8,topKImpl:V8,transposeImpl:Ov,uniqueImpl:U8}=zN;function jT(e,t){return["x","y","z","w","u","v"].slice(0,t).map(n=>`${e}.${n}`)}function mn(e,t){return t===1?[e]:jT(e,t)}function G8(e,t){if(e===1)return"rc";let n="";for(let a=0;a<e;a++)n+=t[a],a<e-1&&(n+=",");return n}var K8=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=mn("rc",t),a=ft(t),r=H8(t,e,n),s=j8(t,e[e.length-1],e[e.length-2],n),i=q8(e,n);this.userCode=`
|
|
void main() {
|
|
${a} rc = getOutputCoords();
|
|
|
|
if(${r}) {
|
|
setOutput(vec4(0));
|
|
} else {
|
|
${s}
|
|
|
|
setOutput(vec4(${i}));
|
|
}
|
|
}
|
|
`}}};function X8(e,t){let n=[];for(let a=0;a<=1;a++)for(let r=0;r<=1;r++){let s=`${a===0?"r":"rp1"}, ${r===0?"c":"cp1"}`;for(let i=2;i<e;i++)s=`${t[t.length-1-i]},`+s;n.push(s)}return n}function H8(e,t,n){if(e===1)return`rc > ${t[0]}`;let a="";for(let r=e-2;r<e;r++)a+=`${n[r]} >= ${t[r]}`,r<e-1&&(a+="||");return a}function j8(e,t,n,a){if(e===1)return"";let r=a.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 q8(e,t){let n=e.length,a=X8(n,t);return n===1?`getA(rc),
|
|
rc + 1 >= ${e[0]} ? 0. : getA(rc + 1),
|
|
0, 0`:`getA(${a[0]}),
|
|
cEdge ? 0. : getA(${a[1]}),
|
|
rEdge ? 0. : getA(${a[2]}),
|
|
rEdge || cEdge ? 0. : getA(${a[3]})`}var qT=class{constructor(e,t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e;let n="";for(let a=0;a<4;a++){let r="thisRC = rc;";a%2==1&&(r+="thisRC.z += 1;"),a>1&&(r+="thisRC.y += 1;"),n+=`
|
|
${r}
|
|
${a>0?"if(thisRC.y < rows && thisRC.z < cols){":""}
|
|
int flatIndex = getFlatIndex(thisRC);
|
|
|
|
ivec3 inputRC = inputCoordsFromReshapedOutCoords(flatIndex);
|
|
vec2 inputRCInnerDims = vec2(float(inputRC.y),float(inputRC.z));
|
|
|
|
result[${a}] =
|
|
getChannel(getA(inputRC.x, inputRC.y, inputRC.z), inputRCInnerDims);
|
|
${a>0?"}":""}
|
|
`}this.userCode=`
|
|
${Y8(t)}
|
|
${Pv(e)}
|
|
|
|
void main() {
|
|
ivec3 rc = getOutputCoords();
|
|
|
|
vec4 result = vec4(0.);
|
|
|
|
ivec3 thisRC;
|
|
int rows = ${e[1]};
|
|
int cols = ${e[2]};
|
|
|
|
${n}
|
|
|
|
setOutput(result);
|
|
}
|
|
`}};function Y8(e){return`
|
|
ivec3 inputCoordsFromReshapedOutCoords(int index) {
|
|
${Gi(["r","c","d"],e)}
|
|
return ivec3(r, c, d);
|
|
}
|
|
`}var Z8=class{constructor(e){this.gpgpu=e,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0,this.freeTextures={},this.logEnabled=!1,this.usedTextures={}}acquireTexture(e,t,n){let a=XT(t,n),r=YT(e,a,n);r in this.freeTextures||(this.freeTextures[r]=[]),r in this.usedTextures||(this.usedTextures[r]=[]);let s=KT(e,a,this.gpgpu.gl,this.gpgpu.textureConfig,n);if(this.freeTextures[r].length>0){this.numFreeTextures--,this.numUsedTextures++,this._numBytesFree-=s,this.log();let o=this.freeTextures[r].shift();return this.usedTextures[r].push(o),o}let i;return a===tn.PACKED_2X2_FLOAT32?i=this.gpgpu.createPackedMatrixTexture(e[0],e[1]):a===tn.PACKED_2X2_FLOAT16?i=this.gpgpu.createFloat16PackedMatrixTexture(e[0],e[1]):a===tn.UNPACKED_FLOAT32?i=this.gpgpu.createFloat32MatrixTexture(e[0],e[1]):a===tn.UNPACKED_FLOAT16?i=this.gpgpu.createFloat16MatrixTexture(e[0],e[1]):a===tn.PACKED_4X1_UNSIGNED_BYTE&&(i=this.gpgpu.createUnsignedBytesMatrixTexture(e[0],e[1])),this.usedTextures[r].push(i),this.numUsedTextures++,this._numBytesAllocated+=s,this.log(),i}releaseTexture(e,t,n,a){if(this.freeTextures==null)return;let r=XT(n,a),s=YT(t,r,a);s in this.freeTextures||(this.freeTextures[s]=[]);let i=KT(t,r,this.gpgpu.gl,this.gpgpu.textureConfig,a),o=ee().get("WEBGL_DELETE_TEXTURE_THRESHOLD");o!==-1&&this._numBytesAllocated>o?(this.gpgpu.deleteMatrixTexture(e),this._numBytesAllocated-=i):(this.freeTextures[s].push(e),this.numFreeTextures++,this._numBytesFree+=i),this.numUsedTextures--;let l=this.usedTextures[s],c=l.indexOf(e);if(c<0)throw new Error("Cannot release a texture that was never provided by this texture manager");l.splice(c,1),this.log()}log(){if(!this.logEnabled)return;let e=this.numFreeTextures+this.numUsedTextures;console.log("Free/Used",`${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 J8(e,t){let n=e;if(t===n.R32F)return 4;if(t===n.R16F)return 2;if(t===n.RGBA32F||t===e.RGBA)return 16;if(t===n.RGBA16F)return 8;throw new Error(`Unknown internal format ${t}`)}function KT(e,t,n,a,r){let s=Q8(t,a),i;if(r){let[l,c]=su(e[0],e[1]);i=l*c}else{let[l,c]=yp(e[0],e[1]);i=l*c}let o=J8(n,s);return i*o}function Q8(e,t){switch(e){case tn.PACKED_2X2_FLOAT32:return LT(t);case tn.PACKED_2X2_FLOAT16:return zT(t);case tn.UNPACKED_FLOAT32:return MT(t);case tn.UNPACKED_FLOAT16:return PT(t);case tn.PACKED_4X1_UNSIGNED_BYTE:return OT(t);default:throw new Error(`Unknown physical texture type ${e}`)}}function eK(e){return ee().getBool("WEBGL_RENDER_FLOAT32_ENABLED")?e?tn.PACKED_2X2_FLOAT32:tn.UNPACKED_FLOAT32:e?tn.PACKED_2X2_FLOAT16:tn.UNPACKED_FLOAT16}function XT(e,t){if(e===Qn.UPLOAD)return tn.PACKED_2X2_FLOAT32;if(e===Qn.RENDER||e==null)return eK(t);if(e===Qn.DOWNLOAD||e===Qn.PIXELS)return tn.PACKED_4X1_UNSIGNED_BYTE;throw new Error(`Unknown logical texture type ${e}`)}function YT(e,t,n){return`${e[0]}_${e[1]}_${t}_${n}`}var cs=class{constructor(e,t){this.variableNames=["A"],this.outputShape=e,this.userCode=`
|
|
float unaryOperation(float x) {
|
|
${t}
|
|
}
|
|
|
|
void main() {
|
|
float x = getAAtOutCoords();
|
|
float y = unaryOperation(x);
|
|
|
|
setOutput(y);
|
|
}
|
|
`}},Ra="if (isnan(x)) return x;",tK="return x;",ZT="return abs(x);",nK="return (x >= 0.0) ? x : (exp(x) - 1.0);",aK=Ra+`
|
|
return (x < 0.0) ? 0.0 : x;
|
|
`,rK=Ra+`
|
|
return (x < 0.0) ? 0.0 : min(6.0, x);
|
|
`,zm="return x;",sK="return x;",iK=`
|
|
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;
|
|
`,oK=`
|
|
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;
|
|
`,lK=`
|
|
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;
|
|
`,du=class{constructor(e,t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e,this.userCode=`
|
|
vec4 unaryOperation(vec4 x) {
|
|
${t}
|
|
}
|
|
|
|
void main() {
|
|
vec4 x = getAAtOutCoords();
|
|
vec4 y = unaryOperation(x);
|
|
|
|
setOutput(y);
|
|
}
|
|
`}},uK=class{constructor(e){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outputShape=e;let t=e.length,n=mn("rc",t),a=ft(t),r=G8(t,n),s=n.slice(-2),i=t<=1?"rc":`vec2(${s.join(",")})`;this.userCode=`
|
|
void main() {
|
|
${a} rc = getOutputCoords();
|
|
vec4 packedInput = getA(${r});
|
|
|
|
setOutput(getChannel(packedInput, ${i}));
|
|
}
|
|
`}},cK=Za.whereImpl,pK=1e-7,dK=1e-4,Lv={};function hK(e){return e in Lv||(Lv[e]={}),Lv[e]}var mK=128,fK=600;function gK(){return ee().global.screen==null?1024:ee().global.screen.height*ee().global.screen.width*window.devicePixelRatio*fK/1024/1024}var bK=class extends sc{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.warnedAboutMemory=!1,this.warnedAboutCPUBackend=!1,this.pendingDeletes=0,this.disposed=!1,!ee().getBool("HAS_WEBGL"))throw new Error("WebGL is not supported on this device");if(e==null){let t=rr(ee().getNumber("WEBGL_VERSION"));this.binaryCache=hK(ee().getNumber("WEBGL_VERSION")),this.gpgpu=new $5(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 Z8(this.gpgpu),this.numMBBeforeWarning=gK(),this.texData=new hd(this,Kr())}numDataIds(){return this.texData.numDataIds()+(this.cpuBackend?this.cpuBackend.numDataIds():0)-this.pendingDeletes}write(e,t,n){if((ee().getBool("WEBGL_CHECK_NUMERICAL_PROBLEMS")||ee().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 a={};return this.texData.set(a,{shape:t,dtype:n,values:e,usage:Qn.UPLOAD,refCount:1,complexParentRefCount:0}),a}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--}}decComplexRef(e){if(this.texData.has(e)){let t=this.texData.get(e);t.complexParentRefCount>0&&t.refCount--}}move(e,t,n,a){if(ee().getBool("DEBUG")&&this.checkNumericalProblems(t),a==="complex64")throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");this.texData.set(e,{shape:n,dtype:a,values:t,usage:Qn.UPLOAD,refCount:1,complexParentRefCount:0})}disposeIntermediateTensorInfo(e){let t=e.dataId;if(this.texData.has(t)){let n=this.texData.get(t);n.refCount--,n.refCount<1&&this.disposeData(t)}}readSync(e){let t=this.texData.get(e),{values:n,dtype:a,complexTensorInfos:r,slice:s,shape:i,isPacked:o}=t;if(s!=null){let p;o?p=new du(i,zm):p=new cs(i,zm);let d=this.runWebGLProgram(p,[{dataId:e,shape:i,dtype:a}],a),h=this.readSync(d.dataId);return this.disposeIntermediateTensorInfo(d),h}if(n!=null)return this.convertAndCacheOnCPU(e);if(a==="string")return n;let l=this.activeTimers!=null,c;l&&(c=k.now());let u;if(a==="complex64"){let p=this.readSync(r.real.dataId),d=this.readSync(r.imag.dataId);u=_.mergeRealAndImagArrays(p,d)}else u=this.getValuesFromTexture(e);return l&&(this.downloadWaitMs+=k.now()-c),this.convertAndCacheOnCPU(e,u)}async read(e){if(this.pendingRead.has(e)){let h=this.pendingRead.get(e);return new Promise(m=>h.push(m))}let t=this.texData.get(e),{values:n,shape:a,slice:r,dtype:s,complexTensorInfos:i,isPacked:o}=t;if(r!=null){let h;o?h=new du(a,zm):h=new cs(a,zm);let m=this.runWebGLProgram(h,[{dataId:e,shape:a,dtype:s}],s),f=this.read(m.dataId);return this.disposeIntermediateTensorInfo(m),f}if(n!=null)return this.convertAndCacheOnCPU(e);if(!ee().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")&&ee().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(s!=="complex64"&&ee().get("WEBGL_BUFFER_SUPPORTED")){c=this.decode(e);let h=this.texData.get(c.dataId);l=this.gpgpu.createBufferFromTexture(h.texture,...bp(a))}this.pendingRead.set(e,[]),s!=="complex64"&&await this.gpgpu.createAndWaitForFence();let u;if(s==="complex64"){let h=await Promise.all([this.read(i.real.dataId),this.read(i.imag.dataId)]),m=h[0],f=h[1];u=_.mergeRealAndImagArrays(m,f)}else if(l==null)u=this.getValuesFromTexture(e);else{let h=k.sizeFromShape(a);u=this.gpgpu.downloadFloat32MatrixFromBuffer(l,h)}c!=null&&this.disposeIntermediateTensorInfo(c);let p=this.convertAndCacheOnCPU(e,u),d=this.pendingRead.get(e);return this.pendingRead.delete(e),d.forEach(h=>h(p)),this.pendingDisposal.has(e)&&(this.pendingDisposal.delete(e),this.disposeData(e),this.pendingDeletes--),p}bufferSync(e){let t=this.readSync(e.dataId),n=t;if(e.dtype==="string")try{n=t.map(a=>k.decodeString(a))}catch(a){throw new Error("Failed to decode encoded string bytes into utf-8")}return Le(e.shape,e.dtype,n)}checkNumericalProblems(e){if(e!=null)for(let t=0;t<e.length;t++){let n=e[t];if(!Oq(n))throw ee().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 ${n} cannot be represented on this device.`)}}getValuesFromTexture(e){let{shape:t,dtype:n,isPacked:a}=this.texData.get(e),r=k.sizeFromShape(t);if(ee().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")){let p=this.decode(e),d=this.texData.get(p.dataId),h=this.gpgpu.downloadMatrixFromPackedTexture(d.texture,...bp(t)).subarray(0,r);return this.disposeIntermediateTensorInfo(p),h}let s=ee().getBool("WEBGL_PACK")&&a===!0,i=s?$v(t):t,o=s?new d5(i):new p5(i),l=this.runWebGLProgram(o,[{shape:i,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}async time(e){let t=this.activeTimers,n=[],a=!1;this.programTimersStack==null?(this.programTimersStack=n,a=!0):this.activeTimers.push(n),this.activeTimers=n,e();let r=k.flatten(this.activeTimers.map(o=>o.query)).filter(o=>o!=null),s=k.flatten(this.activeTimers.map(o=>o.name)).filter(o=>o!=null);this.activeTimers=t,a&&(this.programTimersStack=null);let i={uploadWaitMs:this.uploadWaitMs,downloadWaitMs:this.downloadWaitMs,kernelMs:null,wallMs:null};if(ee().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0){let o=await Promise.all(r);i.kernelMs=k.sum(o),i.getExtraProfileInfo=()=>o.map((l,c)=>({name:s[c],ms:l})).map(l=>`${l.name}: ${l.ms}`).join(", ")}else i.kernelMs={error:"WebGL query timers are not supported in this environment."};return this.uploadWaitMs=0,this.downloadWaitMs=0,i}memory(){return{unreliable:!1,numBytesInGPU:this.numBytesInGPU,numBytesInGPUAllocated:this.textureManager.numBytesAllocated,numBytesInGPUFree:this.textureManager.numBytesFree}}startTimer(){return ee().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?this.gpgpu.beginQuery():{startMs:k.now(),endMs:null}}endTimer(e){return ee().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?(this.gpgpu.endQuery(),e):(e.endMs=k.now(),e)}async getQueryTime(e){if(ee().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0)return this.gpgpu.waitForQueryAndGetTime(e);let t=e;return t.endMs-t.startMs}disposeData(e){if(this.pendingDisposal.has(e))return;if(this.pendingRead.has(e)){this.pendingDisposal.add(e),this.pendingDeletes++;return}if(!this.texData.has(e))return;if(this.texData.get(e).complexParentRefCount>0){this.texData.get(e).refCount--;return}this.releaseGPUData(e);let{complexTensorInfos:t}=this.texData.get(e);t!=null&&(this.texData.get(t.real.dataId).complexParentRefCount--,this.disposeIntermediateTensorInfo(t.real),this.texData.get(t.imag.dataId).complexParentRefCount--,this.disposeIntermediateTensorInfo(t.imag)),this.texData.delete(e)}releaseGPUData(e){let{texture:t,dtype:n,texShape:a,usage:r,isPacked:s,slice:i}=this.texData.get(e),o=i&&i.origDataId||e,l=this.dataRefCount.get(o);l>1?this.dataRefCount.set(o,l-1):(this.dataRefCount.delete(o),t!=null&&(this.numBytesInGPU-=this.computeBytes(a,n),this.textureManager.releaseTexture(t,a,r,s)));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)}getCPUBackend(){return ee().getBool("WEBGL_CPU_FORWARD")?(this.cpuBackend==null&&(this.cpuBackend=Kr().findBackend("cpu")),this.cpuBackend):null}shouldExecuteOnCPU(e,t=mK){let n=this.getCPUBackend();return!ee().getBool("IS_TEST")&&!this.warnedAboutCPUBackend&&n==null&&(console.warn("Your application contains ops that are small enough to be executed on the CPU backend, however the CPU backend cannot be found. Consider importing the CPU backend (@tensorflow/tfjs-backend-cpu) for better performance."),this.warnedAboutCPUBackend=!0),n!=null&&e.every(a=>this.texData.get(a.dataId).texture==null&&k.sizeFromShape(a.shape)<t)}getGPGPUContext(){return this.gpgpu}where(e){_.warn("tf.where() in webgl locks the UI thread. Call tf.whereAsync() instead");let t=e.dataSync();return cK(e.shape,t)}packedUnaryOp(e,t,n){let a=new du(e.shape,t);return this.compileAndRun(a,[e],n)}abs(e){if(this.shouldExecuteOnCPU([e])&&e.dtype!=="complex64"){let n=HT(this.texData.get(e.dataId).values);return this.makeOutput(e.shape,e.dtype,n)}if(ee().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(e,ZT,e.dtype);let t=new cs(e.shape,ZT);return this.compileAndRun(t,[e])}makeTensorInfo(e,t,n){let a;if(t==="string"&&n!=null&&n.length>0&&k.isString(n[0])){let r=n.map(s=>k.encodeString(s));a=this.write(r,e,t)}else a=this.write(n,e,t);return this.texData.get(a).usage=null,{dataId:a,shape:e,dtype:t}}makeOutput(e,t,n){let{dataId:a}=this.makeTensorInfo(e,t,n);return Kr().makeTensorFromDataId(a,e,t,this)}unpackTensor(e){let t=new uK(e.shape);return this.runWebGLProgram(t,[e],e.dtype)}packTensor(e){let t=new K8(e.shape),n=!0;return this.runWebGLProgram(t,[e],e.dtype,null,n)}packedReshape(e,t){let n=[iu(e.shape),...ou(e.shape)],a={dtype:e.dtype,shape:n,dataId:e.dataId},r=[iu(t),...ou(t)],s=new qT(r,n),i=!0,o=this.runWebGLProgram(s,[a],e.dtype,null,i);return{dataId:o.dataId,shape:t,dtype:o.dtype}}decode(e){let t=this.texData.get(e),{isPacked:n,shape:a,dtype:r}=t,s=$v(a),i;n?i=new c5(s):i=new u5(s);let o=!0,l=this.runWebGLProgram(i,[{shape:s,dtype:r,dataId:e}],r,null,o);return{dtype:r,shape:a,dataId:l.dataId}}runWebGLProgram(e,t,n,a,r=!1){let s=this.makeTensorInfo(e.outputShape,n),i=this.texData.get(s.dataId);if(e.packedOutput&&(i.isPacked=!0),e.outPackingScheme===gp.DENSE){let m=bp(e.outputShape);i.texShape=m.map(f=>f*2)}if(e.outTexUsage!=null&&(i.usage=e.outTexUsage),k.sizeFromShape(s.shape)===0)return i.values=k.getTypedArrayFromDType(s.dtype,0),s;let o=[],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 f=this.texData.get(m.dataId);if(f.texture==null){if(!e.packedInputs&&k.sizeFromShape(m.shape)<=ee().getNumber("WEBGL_SIZE_UPLOAD_UNIFORM"))return{shape:m.shape,texData:null,isUniform:!0,uniformValues:f.values};e.packedInputs&&(f.isPacked=!0,f.shape=m.shape)}else if(!!f.isPacked!=!!e.packedInputs)m=f.isPacked?this.unpackTensor(m):this.packTensor(m),o.push(m),f=this.texData.get(m.dataId);else if(f.isPacked&&!Lm(f.shape,m.shape)){let g=m,y=m.shape;m.shape=f.shape,m=this.packedReshape(m,y),o.push(m),f=this.texData.get(m.dataId),g.shape=y}return this.uploadToGPU(m.dataId),{shape:m.shape,texData:f,isUniform:!1}});this.uploadToGPU(s.dataId);let c={shape:s.shape,texData:i,isUniform:!1},u=y8(e,l,c),p=this.getAndSaveBinary(u,()=>f8(this.gpgpu,e,l,c)),d=this.activeTimers!=null,h;if(d&&(h=this.startTimer()),g8(this.gpgpu,p,l,c,a),o.forEach(m=>this.disposeIntermediateTensorInfo(m)),d&&(h=this.endTimer(h),this.activeTimers.push({name:e.constructor.name,query:this.getQueryTime(h)})),!ee().getBool("WEBGL_LAZILY_UNPACK")&&i.isPacked&&r===!1){let m=this.unpackTensor(s);return this.disposeIntermediateTensorInfo(s),m}return s}compileAndRun(e,t,n,a,r=!1){n=n||t[0].dtype;let s=this.runWebGLProgram(e,t,n,a,r);return Kr().makeTensorFromDataId(s.dataId,s.shape,s.dtype)}getAndSaveBinary(e,t){return e in this.binaryCache||(this.binaryCache[e]=t()),this.binaryCache[e]}getTextureManager(){return this.textureManager}dispose(){this.disposed||(ee().getBool("IS_TEST")||Object.keys(this.binaryCache).forEach(e=>{this.gpgpu.deleteProgram(this.binaryCache[e].webGLProgram),delete this.binaryCache[e]}),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=D(()=>{if(!ee().get("WEBGL_RENDER_FLOAT32_ENABLED")){let e=ee().getBool("DEBUG");ee().set("DEBUG",!1);let t=this.abs(de(1e-8)).dataSync()[0];if(ee().set("DEBUG",e),t>0)return 32}return 16})),this.floatPrecisionValue}epsilon(){return this.floatPrecision()===32?pK:dK}uploadToGPU(e){let t=this.texData.get(e),{shape:n,dtype:a,values:r,texture:s,usage:i,isPacked:o}=t;if(s!=null)return;let l=this.activeTimers!=null,c;l&&(c=k.now());let u=t.texShape;if(u==null&&(u=t5(n,o),t.texShape=u),r!=null){let p=$v(n),d,h=u[1],m=u[0],f=r instanceof Uint8Array;o?([h,m]=su(u[0],u[1]),d=new m5(p,[m,h],f)):d=new h5(p,[m,h],f);let g=this.makeTensorInfo([m,h],a);f?this.texData.get(g.dataId).usage=Qn.PIXELS:this.texData.get(g.dataId).usage=Qn.UPLOAD,this.gpgpu.uploadDenseMatrixToTexture(this.getTexture(g.dataId),h,m,r);let y=!0,b=this.runWebGLProgram(d,[g],a,null,y),v=this.texData.get(b.dataId);t.texture=v.texture,t.texShape=v.texShape,t.isPacked=v.isPacked,t.usage=v.usage,this.disposeIntermediateTensorInfo(g),this.texData.delete(b.dataId),t.values=null,l&&(this.uploadWaitMs+=k.now()-c)}else{let p=this.acquireTexture(u,i,a,o);t.texture=p}}convertAndCacheOnCPU(e,t){let n=this.texData.get(e),{dtype:a}=n;return this.releaseGPUData(e),t!=null&&(n.values=yK(t,a)),n.values}acquireTexture(e,t,n,a){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,a)}computeBytes(e,t){return e[0]*e[1]*k.bytesPerElement(t)}};function yK(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 a=0;a<n.length;++a)n[a]=Math.round(e[a]);return n}else throw new Error(`Unknown dtype ${t}`)}var xK="3.0.0";eh.isBrowser()&&ih("webgl",()=>new bK,2);var JT=`
|
|
if (isnan(a)) return a;
|
|
if (isnan(b)) return b;
|
|
`,hu=class{constructor(e,t,n){this.variableNames=["A","B"],this.outputShape=_.assertAndGetBroadcastShape(t,n),this.userCode=`
|
|
float binaryOperation(float a, float b) {
|
|
${e}
|
|
}
|
|
|
|
void main() {
|
|
float a = getAAtOutCoords();
|
|
float b = getBAtOutCoords();
|
|
setOutput(binaryOperation(a, b));
|
|
}
|
|
`}},Bm=`
|
|
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;
|
|
`,wp=class{constructor(e,t,n,a=!1){this.variableNames=["A","B"],this.supportsBroadcasting=!0,this.packedInputs=!0,this.packedOutput=!0,this.outputShape=_.assertAndGetBroadcastShape(t,n);let r=this.outputShape.length,s="";if(a)if(r===0||k.sizeFromShape(this.outputShape)===1)s=`
|
|
result.y = 0.;
|
|
result.z = 0.;
|
|
result.w = 0.;
|
|
`;else if(s=`
|
|
${ft(r)} coords = getOutputCoords();
|
|
`,r===1)s+=`
|
|
result.y = (coords + 1) >= ${this.outputShape[0]} ? 0. : result.y;
|
|
result.z = 0.;
|
|
result.w = 0.;
|
|
`;else{let i=mn("coords",r);s+=`
|
|
bool nextRowOutOfBounds =
|
|
(${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);
|
|
${s}
|
|
|
|
setOutput(result);
|
|
}
|
|
`}};function Bn(e){let{inputs:t,backend:n}=e,{x:a}=t;return n.incRef(a.dataId),{dataId:a.dataId,shape:a.shape,dtype:a.dtype}}var vK={kernelName:Ko,backendName:"webgl",kernelFunc:Bn};function ps(e){let{inputs:t,backend:n}=e,{real:a,imag:r}=t,s=n.makeTensorInfo(a.shape,"complex64"),i=n.texData.get(s.dataId),o=Bn({inputs:{x:a},backend:n}),l=n.texData.get(o.dataId);l.complexParentRefCount++;let c=Bn({inputs:{x:r},backend:n}),u=n.texData.get(c.dataId);return u.complexParentRefCount++,i.complexTensorInfos={real:o,imag:c},s}var wK={kernelName:kd,backendName:"webgl",kernelFunc:ps},QT="return (a < 0.) ? b * a : a;",eS=`
|
|
vec4 aLessThanZero = vec4(lessThan(a, vec4(0.)));
|
|
return (aLessThanZero * (b * a)) + ((vec4(1.0) - aLessThanZero) * a);
|
|
`;function kK(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{alpha:s}=a,i=n.makeTensorInfo([],"float32",k.createScalarValue(s,"float32")),o=ee().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new wp(eS,r.shape,i.shape):new hu(QT,r.shape,i.shape),l=n.runWebGLProgram(o,[r,i],r.dtype);return n.disposeIntermediateTensorInfo(i),l}var IK={kernelName:Vs,backendName:"webgl",kernelFunc:kK},tS="return (a < 0.) ? b * a : a;",nS=`
|
|
vec4 aLessThanZero = vec4(lessThan(a, vec4(0.)));
|
|
return (aLessThanZero * (b * a)) + ((vec4(1.0) - aLessThanZero) * a);
|
|
`;function NK(e){let{inputs:t,backend:n}=e,{x:a,alpha:r}=t,s=ee().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new wp(nS,a.shape,r.shape):new hu(tS,a.shape,r.shape);return n.runWebGLProgram(s,[a,r],a.dtype)}var TK={kernelName:ei,backendName:"webgl",kernelFunc:NK},aS="if (isnan(x)) return x;",SK=`
|
|
if (isnan(a)) return a;
|
|
if (isnan(b)) return b;
|
|
`,CK=`
|
|
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 Ze({opSnippet:e,packedOpSnippet:t,cpuKernelImpl:n,dtype:a}){return({inputs:r,backend:s})=>{let{x:i}=r,o=s,l=a||i.dtype;if(o.shouldExecuteOnCPU([i])&&n!=null){let p=o.texData.get(i.dataId),d=n(p.values,l);return o.makeTensorInfo(i.shape,l,d)}let c=ee().getBool("WEBGL_PACK_UNARY_OPERATIONS")&&t!=null,u;return c?u=new du(i.shape,t):u=new cs(i.shape,e),o.runWebGLProgram(u,[i],l)}}function nn({opSnippet:e,packedOpSnippet:t,checkOutOfBounds:n=!1,supportsComplex:a=!1,cpuKernelImpl:r,dtype:s}){return({inputs:i,backend:o})=>{let{a:l,b:c}=i,u=o;if(a&&l.dtype==="complex64"){let m=u.texData.get(l.dataId),f=u.texData.get(c.dataId),[g,y]=[[m.complexTensorInfos.real,f.complexTensorInfos.real],[m.complexTensorInfos.imag,f.complexTensorInfos.imag]].map(v=>{let[x,N]=v,T={dataId:x.dataId,dtype:x.dtype,shape:l.shape},C={dataId:N.dataId,dtype:N.dtype,shape:c.shape},$=new hu(e,l.shape,c.shape);return u.runWebGLProgram($,[T,C],la(x.dtype,N.dtype))}),b=ps({inputs:{real:g,imag:y},backend:u});return u.disposeIntermediateTensorInfo(g),u.disposeIntermediateTensorInfo(y),b}let p=s||la(l.dtype,c.dtype);if(u.shouldExecuteOnCPU([l,c])&&r!=null){let m=u.texData.get(l.dataId),f=u.texData.get(c.dataId),[g,y]=r(l.shape,c.shape,m.values,f.values,p),b=u.makeTensorInfo(y,p),v=u.texData.get(b.dataId);return v.values=g,b}let d=ee().getBool("WEBGL_PACK_BINARY_OPERATIONS")&&t!=null,h;return d?h=new wp(t,l.shape,c.shape,n):h=new hu(e,l.shape,c.shape),u.runWebGLProgram(h,[l,c],p)}}function Wm(e,t=!1){if(e==="linear")return t?sK:tK;if(e==="relu")return t?oK:aK;if(e==="elu")return t?iK:nK;if(e==="relu6")return t?lK:rK;if(e==="prelu")return t?nS:tS;if(e==="leakyrelu")return t?eS:QT;throw new Error(`Activation ${e} has not been implemented for the WebGL backend.`)}var rS=class{constructor(e,t,n,a=!1,r=!1,s=!1,i=null,o=!1,l=!1){this.variableNames=["matrixA","matrixB"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=n;let c=a?e[1]:e[2],u=Math.ceil(c/2),p=a?"i * 2, rc.y":"rc.y, i * 2",d=r?"rc.z, i * 2":"i * 2, rc.z",h=a?["a.xxyy","a.zzww"]:["a.xxzz","a.yyww"],m=r?["b.xzxz","b.ywyw"]:["b.xyxy","b.zwzw"],f="",g="";i&&(o?f=`vec4 activation(vec4 a) {
|
|
vec4 b = getPreluActivationWeightsAtOutCoords();
|
|
${i}
|
|
}`:l?f=`vec4 activation(vec4 a) {
|
|
vec4 b = getLeakyreluAlphaAtOutCoords();
|
|
${i}
|
|
}`:f=`vec4 activation(vec4 x) {
|
|
${i}
|
|
}`,g="result = activation(result);");let y=s?"result += getBiasAtOutCoords();":"";s&&this.variableNames.push("bias"),o&&this.variableNames.push("preluActivationWeights"),l&&this.variableNames.push("leakyreluAlpha");let b="rc.x",v="rc.x";e[0]<t[0]?b=`int(min(float(rc.x), ${e[0]-1}.))`:t[0]<e[0]&&(v=`int(min(float(rc.x), ${t[0]-1}.))`),this.userCode=`
|
|
${f}
|
|
|
|
const float sharedDimension = ${u}.0;
|
|
|
|
vec4 dot2x2ARowBCol(ivec3 rc) {
|
|
vec4 result = vec4(0);
|
|
for (int i = 0; i < ${u}; i++) {
|
|
int batchA = ${b};
|
|
int batchB = ${v};
|
|
vec4 a = getMatrixA(batchA, ${p});
|
|
vec4 b = getMatrixB(batchB, ${d});
|
|
|
|
// These swizzled products need to be separately added.
|
|
// See: https://github.com/tensorflow/tfjs/issues/1735
|
|
result += (${h[0]} * ${m[0]});
|
|
result += (${h[1]} * ${m[1]});
|
|
}
|
|
return result;
|
|
}
|
|
|
|
void main() {
|
|
ivec3 rc = getOutputCoords();
|
|
vec4 result = dot2x2ARowBCol(rc);
|
|
|
|
${y}
|
|
|
|
${g}
|
|
|
|
setOutput(result);
|
|
}
|
|
`}},sS={REAL:"return areal * breal - aimag * bimag;",IMAG:"return areal * bimag + aimag * breal;"},iS=class{constructor(e,t,n){this.variableNames=["AReal","AImag","BReal","BImag"],this.outputShape=_.assertAndGetBroadcastShape(t,n),this.userCode=`
|
|
float binaryOpComplex(
|
|
float areal, float aimag, float breal, float bimag) {
|
|
${e}
|
|
}
|
|
|
|
void main() {
|
|
float areal = getARealAtOutCoords();
|
|
float aimag = getAImagAtOutCoords();
|
|
float breal = getBRealAtOutCoords();
|
|
float bimag = getBImagAtOutCoords();
|
|
setOutput(binaryOpComplex(areal, aimag, breal, bimag));
|
|
}
|
|
`}},oS="return a * b;";function lS(e){let{inputs:t,backend:n}=e,{a,b:r}=t,s=_.upcastType(a.dtype,r.dtype);if(a.dtype==="complex64"){let o=n.texData.get(a.dataId),l=n.texData.get(r.dataId),c=new iS(sS.REAL,a.shape,r.shape),u=new iS(sS.IMAG,a.shape,r.shape),p=[{dataId:o.complexTensorInfos.real.dataId,dtype:o.complexTensorInfos.real.dtype,shape:a.shape},{dataId:o.complexTensorInfos.imag.dataId,dtype:o.complexTensorInfos.imag.dtype,shape:a.shape},{dataId:l.complexTensorInfos.real.dataId,dtype:l.complexTensorInfos.real.dtype,shape:r.shape},{dataId:l.complexTensorInfos.imag.dataId,dtype:l.complexTensorInfos.imag.dtype,shape:r.shape}],d=n.runWebGLProgram(c,p,"float32"),h=n.runWebGLProgram(u,p,"float32"),m=ps({inputs:{real:d,imag:h},backend:n});return n.disposeIntermediateTensorInfo(d),n.disposeIntermediateTensorInfo(h),m}if(n.shouldExecuteOnCPU([a,r])){let o=n.texData.get(a.dataId),l=n.texData.get(r.dataId),[c,u]=D8(a.shape,r.shape,o.values,l.values,s),p=n.makeTensorInfo(u,s),d=n.texData.get(p.dataId);return d.values=c,p}let i;return ee().getBool("WEBGL_PACK_BINARY_OPERATIONS")?i=new wp(oS,a.shape,r.shape):i=new hu(oS,a.shape,r.shape),n.runWebGLProgram(i,[a,r],s)}var _K={kernelName:Ys,backendName:"webgl",kernelFunc:lS};function EK(e,t,n){let a=[iu(e.shape),...ou(e.shape)],r={dtype:e.dtype,shape:a,dataId:e.dataId},s=[iu(t),...ou(t)],i=new qT(s,a),o=!0,l=n.runWebGLProgram(i,[r],e.dtype,null,o);return{dataId:l.dataId,shape:t,dtype:l.dtype}}function be(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{shape:s}=a,i=n,o=k.sizeFromShape(r.shape),l=k.inferFromImplicitShape(s,o),c=k.sizeFromShape(l);k.assert(o===c,()=>`The new shape (${l}) has ${c} elements and the old shape (${r.shape}) has ${o} elements. The new shape and old shape must have the same number of elements.`);let u=i.texData.get(r.dataId);return u.isPacked&&!Lm(r.shape,l)&&!(u.texture!==null&&Lm(u.shape,l))?EK(r,l,i):(i.incRef(r.dataId),{dataId:r.dataId,shape:l,dtype:r.dtype})}var AK={kernelName:dl,backendName:"webgl",kernelFunc:be},uS=class{constructor(e,t){this.variableNames=["x"];let{windowSize:n,batchSize:a,inSize:r,outSize:s}=e;this.outputShape=[a,s];let i=Math.floor(n/4)*4,o=n%4,l="sumValue += dot(values, ones);";if(t!=null){let u=1/t;l=`sumValue += dot(values * ${k.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 < ${i}; 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 + ${i};
|
|
if (${o===1}) {
|
|
vec4 values = vec4(getValue(batch, inIdx), 0.0, 0.0, 0.0);
|
|
|
|
${l}
|
|
} else if (${o===2}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, inIdx),
|
|
getValue(batch, inIdx + 1), 0.0, 0.0);
|
|
|
|
${l}
|
|
} else if (${o===3}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, inIdx),
|
|
getValue(batch, inIdx + 1),
|
|
getValue(batch, inIdx + 2), 0.0);
|
|
|
|
${l}
|
|
}
|
|
setOutput(sumValue);
|
|
}
|
|
`}},FK=class{constructor(e,t){this.variableNames=["x"];let{windowSize:n,batchSize:a,inSize:r,outSize:s}=e;this.outputShape=[a,s];let i="0.0",o="";t==="prod"?i="1.0":t==="min"?(i="1.0 / 1e-20",o="min"):t==="max"&&(i="-1.0 / 1e-20",o="max");let l=`${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,p=`
|
|
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 = ${o}(values, minMaxValue);
|
|
}
|
|
`,d="vec4";t==="all"?(i="1.0",p=`
|
|
bool reducedAllValue = all(values);
|
|
float floatedReducedAllValue = float(reducedAllValue);
|
|
allValue = float(allValue >= 1.0 && floatedReducedAllValue >= 1.0);
|
|
`,d="bvec4"):t==="any"&&(i="0.0",p=`
|
|
bool reducedAnyValue = any(values);
|
|
float floatedReducedAnyValue = float(reducedAnyValue);
|
|
anyValue = float(anyValue >= 1.0 || floatedReducedAnyValue >= 1.0);
|
|
`,d="bvec4");let h="";r%n>0&&(h=`
|
|
if (inIdx < 0 || inIdx >= ${r}) {
|
|
return initializationValue;
|
|
}
|
|
`),this.userCode=`
|
|
const float initializationValue = ${i};
|
|
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 * ${n};
|
|
|
|
vec4 minMaxValue = vec4(${i});
|
|
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;
|
|
${d} values = ${d}(
|
|
getValue(batch, inIdx),
|
|
getValue(batch, inIdx + 1),
|
|
getValue(batch, inIdx + 2),
|
|
getValue(batch, inIdx + 3)
|
|
);
|
|
|
|
${p}
|
|
}
|
|
|
|
int inIdx = inOffset + ${c};
|
|
if (${u===1}) {
|
|
${d} values = ${d}(
|
|
getValue(batch, inIdx),
|
|
initializationValue,
|
|
initializationValue,
|
|
initializationValue
|
|
);
|
|
|
|
${p}
|
|
} else if (${u===2}) {
|
|
${d} values = ${d}(
|
|
getValue(batch, inIdx),
|
|
getValue(batch, inIdx + 1),
|
|
initializationValue,
|
|
initializationValue
|
|
);
|
|
|
|
${p}
|
|
} else if (${u===3}) {
|
|
${d} values = ${d}(
|
|
getValue(batch, inIdx),
|
|
getValue(batch, inIdx + 1),
|
|
getValue(batch, inIdx + 2),
|
|
initializationValue
|
|
);
|
|
|
|
${p}
|
|
}
|
|
setOutput(${l});
|
|
}
|
|
`}};function $K(e){let t=[];for(;t.length===0||t[t.length-1].outSize!==1;){let n=t.length?t[t.length-1].outSize:e[1],a=_.computeOptimalWindowSize(n);t.push({inSize:n,windowSize:a,outSize:Math.ceil(n/a)})}return t}function ji(e,t,n,a){let r=$K(e.shape),s=e;for(let i=0;i<r.length;i++){let{inSize:o,windowSize:l,outSize:c}=r[i],u,p;n==="mean"?u=i===0?new uS({windowSize:l,inSize:o,batchSize:e.shape[0],outSize:c},o):new uS({windowSize:l,inSize:o,batchSize:e.shape[0],outSize:c}):u=new FK({windowSize:l,inSize:o,batchSize:e.shape[0],outSize:c},n),p=s,s=a.runWebGLProgram(u,[s],t),p.dataId!==e.dataId&&a.disposeIntermediateTensorInfo(p)}return s}var RK=class{constructor(e,t){this.variableNames=["A"];let n=new Array(e.length);for(let s=0;s<n.length;s++)n[s]=e[t[s]];this.outputShape=n,this.rank=n.length;let a=ft(this.rank),r=DK(t);this.userCode=`
|
|
void main() {
|
|
${a} resRC = getOutputCoords();
|
|
setOutput(getA(${r}));
|
|
}
|
|
`}};function DK(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"],a=new Array(t);for(let r=0;r<e.length;r++)a[e[r]]=n[r];return a.join()}var MK=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 ${this.rank} is not yet supported.`);let a=ft(this.rank),r=jT("rc",this.rank),s=new Array(this.rank);for(let c=0;c<t.length;c++)s[t[c]]=r[c];let i=`vec2(${s.slice(-2).join()})`,o=`++${r[this.rank-1]} < ${n[this.rank-1]}`,l=`getChannel(getA(${s.join()}), ${i})`;this.userCode=`
|
|
void main() {
|
|
${a} rc = getOutputCoords();
|
|
vec4 result = vec4(0.);
|
|
result[0] = ${l};
|
|
if(${o}) {
|
|
result[1] = ${l};
|
|
}
|
|
--${r[this.rank-1]};
|
|
if(++${r[this.rank-2]} < ${n[this.rank-2]}) {
|
|
result[2] = ${l};
|
|
if(${o}) {
|
|
result[3] = ${l};
|
|
}
|
|
}
|
|
setOutput(result);
|
|
}
|
|
`}};function Vm(e,t,n){let a=ee().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new MK(e.shape,t):new RK(e.shape,t);return n.runWebGLProgram(a,[e],e.dtype)}function PK(e,t,n,a){let r=t,s=e.shape.length,i=k.parseAxisParam(r,e.shape),o=i,l=_.getAxesPermutation(o,s),c=l!=null,u=e;c&&(u=Vm(e,l,a),o=_.getInnerMostAxes(o.length,s)),_.assertAxesAreInnerMostDims("sum",o,s);let[p,d]=_.computeOutAndReduceShapes(u.shape,o),h=p;n&&(h=_.expandShapeToKeepDim(p,i));let m=k.sizeFromShape(d),f=k.sizeFromShape(e.shape)/m,g=be({inputs:{x:u},attrs:{shape:[f,m]},backend:a}),y=Qd(e.dtype),b=ji(g,y,"sum",a),v=be({inputs:{x:b},attrs:{shape:h},backend:a});return a.disposeIntermediateTensorInfo(g),a.disposeIntermediateTensorInfo(b),c&&a.disposeIntermediateTensorInfo(u),v}function zv(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{axis:s,keepDims:i}=a;return PK(r,s,i,n)}var OK={kernelName:ci,backendName:"webgl",kernelFunc:zv};function Tn(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{perm:s}=a,i=n,o=r.shape.length,l=new Array(o);for(let u=0;u<l.length;u++)l[u]=r.shape[s[u]];let c;if(i.shouldExecuteOnCPU([r])){let u=i.texData.get(r.dataId).values,p=Ov(u,r.shape,r.dtype,s,l);c=i.makeTensorInfo(l,r.dtype);let d=i.texData.get(c.dataId);d.values=p}else c=Vm(r,s,i);return c}var LK={kernelName:fi,backendName:"webgl",kernelFunc:Tn},cS=1e3;function Um({a:e,b:t,transposeA:n,transposeB:a,backend:r,bias:s=null,preluActivationWeights:i=null,leakyreluAlpha:o=0,activation:l=null}){let c=e.shape.length,u=t.shape.length,p=n?e.shape[c-2]:e.shape[c-1],d=a?t.shape[u-1]:t.shape[u-2],h=n?e.shape[c-1]:e.shape[c-2],m=a?t.shape[u-2]:t.shape[u-1],f=e.shape.slice(0,-2),g=t.shape.slice(0,-2),y=k.sizeFromShape(f),b=k.sizeFromShape(g),v=y===b||y===1||b===1;k.assert(c>=2&&u>=2&&v,()=>`Error in matMul: the input batch dimensions must either be the same or at least one input batch dimension must be 1. Got input batch dimensions of (${f}) and (${g}).`);let x=(y>b?e.shape.slice(0,-2):t.shape.slice(0,-2)).concat([h,m]);k.assert(p===d,()=>`Error in matMul: inner shapes (${p}) and (${d}) of Tensors with shapes ${e.shape} and ${t.shape} and transposeA=${n} and transposeB=${a} must match.`);let N=n?[y,p,h]:[y,h,p],T=a?[b,m,d]:[b,d,m],C=be({inputs:{x:e},backend:r,attrs:{shape:N}}),$=be({inputs:{x:t},backend:r,attrs:{shape:T}}),F=[C,$],O=Math.max(y,b),B=n?C.shape[1]:C.shape[2],W=s!=null,G=i!=null,q=l==="leakyrelu",H=l!=null?Wm(l,!0):null,K=W||G||q||H!=null,X;if((h===1||m===1)&&B>cS&&K===!1){let J=C,se=$;n&&(J=Tn({inputs:{x:C},backend:r,attrs:{perm:[0,2,1]}}),F.push(J)),a&&(se=Tn({inputs:{x:$},backend:r,attrs:{perm:[0,2,1]}}),F.push(se));let Q=m!==1,oe=m===1,ae=J;Q&&(ae=be({inputs:{x:J},backend:r,attrs:{shape:[O,B,1]}}),F.push(ae));let pe=m===1?2:1,ue=se;oe&&(ue=be({inputs:{x:se},backend:r,attrs:{shape:[O,1,B]}}),F.push(ue));let ge=lS({inputs:{a:ae,b:ue},backend:r});X=zv({inputs:{x:ge},backend:r,attrs:{axis:pe,keepDims:!0}}),F.push(ge)}else{let J=la(e.dtype,t.dtype),se=new rS(N,T,[O,h,m],n,a,W,H,G,q),Q=[C,$];if(s!=null&&Q.push(s),G&&Q.push(i),q){let oe=r.makeTensorInfo([],"float32",k.createScalarValue(o,"float32"));Q.push(oe),F.push(oe)}X=r.runWebGLProgram(se,Q,J)}let ne=be({inputs:{x:X},backend:r,attrs:{shape:x}});F.push(X);for(let J of F)r.disposeIntermediateTensorInfo(J);return ne}function zK(e){let{inputs:t,backend:n,attrs:a}=e,{a:r,b:s,bias:i,preluActivationWeights:o}=t,{transposeA:l,transposeB:c,activation:u,leakyreluAlpha:p}=a;return Um({a:r,b:s,transposeA:l,transposeB:c,backend:n,bias:i,preluActivationWeights:o,leakyreluAlpha:p,activation:u})}var BK={kernelName:gi,backendName:"webgl",kernelFunc:zK},pS="return abs(x);";function WK(e){let{inputs:t,backend:n}=e,{x:a}=t;if(n.shouldExecuteOnCPU([a])&&a.dtype!=="complex64"){let s=n.texData.get(a.dataId),i=HT(s.values);return n.makeTensorInfo(a.shape,a.dtype,i)}let r;return ee().getBool("WEBGL_PACK_UNARY_OPERATIONS")?r=new du(a.shape,pS):r=new cs(a.shape,pS),n.runWebGLProgram(r,[a],a.dtype)}var VK={kernelName:So,backendName:"webgl",kernelFunc:WK},UK=Ra+`
|
|
if (abs(x) > 1.) {
|
|
return NAN;
|
|
}
|
|
return acos(x);
|
|
`,GK=Ze({opSnippet:UK}),HK={kernelName:Co,backendName:"webgl",kernelFunc:GK},jK=Ra+`
|
|
if (x < 1.0) return NAN;
|
|
return log(x + sqrt(x * x - 1.0));`,qK=Ze({opSnippet:jK}),KK={kernelName:_o,backendName:"webgl",kernelFunc:qK},dS="return a + b;",XK=nn({opSnippet:dS,packedOpSnippet:dS,supportsComplex:!0,cpuKernelImpl:b8}),YK={kernelName:Lr,backendName:"webgl",kernelFunc:XK},ZK=class{constructor(e,t){this.outputShape=[],this.outputShape=e,this.variableNames=t.map((r,s)=>`T${s}`);let n=[];this.variableNames.forEach(r=>{n.push(`float v${r} = get${r}AtOutCoords();`)});let a=this.variableNames.map(r=>`v${r}`).join(" + ");this.userCode=`
|
|
void main() {
|
|
${n.join(`
|
|
`)}
|
|
|
|
float result = ${a};
|
|
setOutput(result);
|
|
}
|
|
`}},JK=class{constructor(e,t){this.outputShape=[],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e,this.variableNames=t.map((r,s)=>`T${s}`);let n=[];this.variableNames.forEach(r=>{n.push(`vec4 v${r} = get${r}AtOutCoords();`)});let a=this.variableNames.map(r=>`v${r}`).join(" + ");this.userCode=`
|
|
void main() {
|
|
${n.join(`
|
|
`)}
|
|
|
|
vec4 result = ${a};
|
|
setOutput(result);
|
|
}
|
|
`}};function Gm(e){let{inputs:t,backend:n}=e,a=t;if(a.length===1)return Bn({inputs:{x:a[0]},backend:n});if(a.length>ee().get("WEBGL_MAX_TEXTURES_IN_SHADER")){let o=Math.floor(a.length/2),l=Gm({inputs:a.slice(0,o),backend:n}),c=Gm({inputs:a.slice(o),backend:n});return Gm({inputs:[l,c],backend:n})}let r=a.map(o=>o.dtype).reduce((o,l)=>la(o,l)),s=a.map(o=>o.shape),i=ee().getBool("WEBGL_PACK")?new JK(a[0].shape,s):new ZK(a[0].shape,s);return n.runWebGLProgram(i,a,r)}var QK={kernelName:Ss,backendName:"webgl",kernelFunc:Gm};function eX(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{axis:s,keepDims:i}=a,o=r.shape.length,l=k.parseAxisParam(s,r.shape),c=l,u=_.getAxesPermutation(c,o),p=r;u!=null&&(p=Tn({inputs:{x:r},backend:n,attrs:{perm:u}}),c=_.getInnerMostAxes(c.length,o)),_.assertAxesAreInnerMostDims("all",c,o);let[d,h]=_.computeOutAndReduceShapes(p.shape,c),m=k.sizeFromShape(h),f=be({inputs:{x:p},backend:n,attrs:{shape:[-1,m]}}),g=ji(f,f.dtype,"all",n),y;if(i){let b=_.expandShapeToKeepDim(d,l);y=be({inputs:{x:g},backend:n,attrs:{shape:b}})}else y=be({inputs:{x:g},backend:n,attrs:{shape:d}});return n.disposeIntermediateTensorInfo(f),n.disposeIntermediateTensorInfo(g),u!=null&&n.disposeIntermediateTensorInfo(p),y}var tX={kernelName:yd,backendName:"webgl",kernelFunc:eX};function nX(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{axis:s,keepDims:i}=a,o=r.shape.length,l=k.parseAxisParam(s,r.shape),c=l,u=_.getAxesPermutation(c,o),p=r;u!=null&&(p=Tn({inputs:{x:r},backend:n,attrs:{perm:u}}),c=_.getInnerMostAxes(c.length,o)),_.assertAxesAreInnerMostDims("any",c,o);let[d,h]=_.computeOutAndReduceShapes(p.shape,c),m=k.sizeFromShape(h),f=be({inputs:{x:p},backend:n,attrs:{shape:[-1,m]}}),g=ji(f,f.dtype,"any",n),y;if(i){let b=_.expandShapeToKeepDim(d,l);y=be({inputs:{x:g},backend:n,attrs:{shape:b}})}else y=be({inputs:{x:g},backend:n,attrs:{shape:d}});return n.disposeIntermediateTensorInfo(f),n.disposeIntermediateTensorInfo(g),u!=null&&n.disposeIntermediateTensorInfo(p),y}var aX={kernelName:bd,backendName:"webgl",kernelFunc:nX},rX=class{constructor(e,t,n){this.variableNames=["A"];let{windowSize:a,batchSize:r,outSize:s}=e;n||this.variableNames.push("bestIndicesA"),this.outputShape=[r,s];let i=t==="max"?">":"<",o=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 * ${a};
|
|
|
|
int bestIndex = inOffset;
|
|
float bestValue = getA(batch, bestIndex);
|
|
|
|
for (int i = 0; i < ${a}; i++) {
|
|
int inIdx = ${o};
|
|
float candidate = getA(batch, inIdx);
|
|
if (candidate ${i} bestValue) {
|
|
bestValue = candidate;
|
|
bestIndex = inIdx;
|
|
}
|
|
}
|
|
setOutput(float(bestIndex));
|
|
}
|
|
`}},sX=class{constructor(e,t,n,a){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,k.assert(e.length>2,()=>`Packed arg${n.charAt(0).toUpperCase()+n.slice(1)} supports only inputs with rank above 2.`);let r=e[e.length-1],s=Math.ceil(r/t);this.outputShape=e.slice(0,-1),s>1&&this.outputShape.push(s),a||this.variableNames.push("bestIndicesA");let i=this.outputShape,o=i.length,l=ft(o),c=mn("coords",o),u,p;if(s===1){p=o+1;let C=ft(p);u=`
|
|
${C} sourceLocR = ${C}(${c.join()}, 0);
|
|
++${c[o-1]};
|
|
${C} sourceLocG = ${C}(${c.join()}, 0);
|
|
++${c[o-2]};
|
|
${C} sourceLocA = ${C}(${c.join()}, 0);
|
|
--${c[o-1]};
|
|
${C} sourceLocB = ${C}(${c.join()}, 0);
|
|
--${c[o-2]};`}else p=o,u=`
|
|
${l} sourceLocR = coords;
|
|
++${c[o-1]};
|
|
${l} sourceLocG = coords;
|
|
++${c[o-2]};
|
|
${l} sourceLocA = coords;
|
|
--${c[o-1]};
|
|
${l} sourceLocB = coords;
|
|
--${c[o-2]};`;let d=["x","y","z","w","u","v"].slice(0,p),h="."+d[p-1],m=d.map(C=>"int "+C),f=mn("sourceLocR",p-1).concat("inIdx.r"),g=mn("sourceLocG",p-1).concat("inIdx.g"),y=mn("sourceLocB",p-1).concat("inIdx.b"),b=mn("sourceLocA",p-1).concat("inIdx.a"),v=n==="max"?"greaterThan":"lessThan",x=a?"":`
|
|
inIdx = round(vec4(getBestIndicesAChannel(${f.join()}),
|
|
getBestIndicesAChannel(${g.join()}),
|
|
getBestIndicesAChannel(${y.join()}),
|
|
getBestIndicesAChannel(${b.join()})));`,N=`vec4(
|
|
getAChannel(${f.join()}),
|
|
hasNextCol ? getAChannel(${g.join()}) : 0.,
|
|
hasNextRow ? getAChannel(${y.join()}) : 0.,
|
|
hasNextRow && hasNextCol ? getAChannel(${b.join()}) : 0.)`,T=a?"":`
|
|
float getBestIndicesAChannel(${m.join()}) {
|
|
return getChannel(getBestIndicesA(${d.join()}),
|
|
vec2(${d.slice(-2).join()}));
|
|
}`;this.userCode=`
|
|
float getAChannel(${m.join()}) {
|
|
return getChannel(getA(${d.join()}),
|
|
vec2(${d.slice(-2).join()}));
|
|
}
|
|
${T}
|
|
void main() {
|
|
${l} coords = getOutputCoords();
|
|
bool hasNextCol = ${c[o-1]} < ${i[o-1]-1};
|
|
bool hasNextRow = ${c[o-2]} < ${i[o-2]-1};
|
|
${u}
|
|
ivec4 srcIdx = ivec4(sourceLocR${h}, sourceLocG${h},
|
|
sourceLocB${h}, sourceLocA${h}) * ${t};
|
|
ivec4 inIdx = srcIdx;
|
|
vec4 bestIndex = vec4(inIdx);
|
|
vec4 bestValue = ${N};
|
|
|
|
for (int i = 0; i < ${t}; i++) {
|
|
inIdx = srcIdx;
|
|
${x}
|
|
vec4 candidate = ${N};
|
|
bvec4 nan = isnan(candidate);
|
|
bvec4 replace = bvec4(
|
|
vec4(${v}(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 hS(e,t,n,a=null){let r=t.shape[0],s=t.shape[1];a!=null&&(r=a.shape[0],s=a.shape[1]);let i=_.computeOptimalWindowSize(s),o={windowSize:i,inSize:s,batchSize:r,outSize:Math.ceil(s/i)},l=new rX(o,n,a==null),c=[t];a!=null&&c.push(a);let u=e.runWebGLProgram(l,c,"int32");if(u.shape[1]===1)return u;let p=hS(e,t,n,u);return e.disposeIntermediateTensorInfo(u),p}function mS(e,t,n,a=null){let r=a!=null?a.shape:t.shape,s=r[r.length-1],i=_.computeOptimalWindowSize(s),o=new sX(r,i,n,a==null),l=a==null?[t]:[t,a],c=e.runWebGLProgram(o,l,"int32");if(c.shape.length===t.shape.length){let u=mS(e,t,n,c);return e.disposeIntermediateTensorInfo(c),u}return c}function fS(e,t,n,a){let r=[n];if(_.assertAxesAreInnerMostDims("arg"+a.charAt(0).toUpperCase()+a.slice(1),r,t.shape.length),!ee().getBool("WEBGL_PACK_REDUCE")||t.shape.length<=2){let s=[],[i,o]=_.computeOutAndReduceShapes(t.shape,r),l=k.sizeFromShape(o),c=be({inputs:{x:t},backend:e,attrs:{shape:[-1,l]}});s.push(c);let u=hS(e,c,a);s.push(u);let p=be({inputs:{x:u},backend:e,attrs:{shape:i}});return s.forEach(d=>e.disposeIntermediateTensorInfo(d)),p}return mS(e,t,a)}function iX(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{axis:s}=a,i=k.parseAxisParam(s,r.shape),o=_.getAxesPermutation(i,r.shape.length),l=r,c=[];o!=null&&(l=Tn({inputs:{x:r},backend:n,attrs:{perm:o}}),c.push(l),i=_.getInnerMostAxes(i.length,l.shape.length)),_.assertAxesAreInnerMostDims("argMax",[i[0]],l.shape.length);let u=fS(n,l,i[0],"max");return c.forEach(p=>n.disposeIntermediateTensorInfo(p)),u}var oX={kernelName:Cs,backendName:"webgl",kernelFunc:iX};function lX(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{axis:s}=a,i=k.parseAxisParam(s,r.shape),o=_.getAxesPermutation(i,r.shape.length),l=r,c=[];o!=null&&(l=Tn({inputs:{x:r},backend:n,attrs:{perm:o}}),c.push(l),i=_.getInnerMostAxes(i.length,l.shape.length)),_.assertAxesAreInnerMostDims("argMin",[i[0]],l.shape.length);let u=fS(n,l,i[0],"min");return c.forEach(p=>n.disposeIntermediateTensorInfo(p)),u}var uX={kernelName:lc,backendName:"webgl",kernelFunc:lX},cX=Ra+`
|
|
if (abs(x) > 1.) {
|
|
return NAN;
|
|
}
|
|
return asin(x);
|
|
`,pX=Ze({opSnippet:cX}),dX={kernelName:Eo,backendName:"webgl",kernelFunc:pX},hX=Ra+"return log(x + sqrt(x * x + 1.0));",mX=Ze({opSnippet:hX}),fX={kernelName:Ao,backendName:"webgl",kernelFunc:mX},gX=Ra+`
|
|
return atan(x);
|
|
`,yX=Ze({opSnippet:gX}),bX={kernelName:Fo,backendName:"webgl",kernelFunc:yX},xX=SK+`
|
|
return atan(a, b);
|
|
`,vX=`
|
|
vec4 result = atan(a, b);
|
|
vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));
|
|
`+CK+`
|
|
return result;
|
|
`,wX=nn({opSnippet:xX,packedOpSnippet:vX}),kX={kernelName:Do,backendName:"webgl",kernelFunc:wX},IX=Ra+`
|
|
if ((x < -1.0) || (x > 1.0)) return NAN;
|
|
return (log(1.0 + x) - log(1.0 - x)) / 2.0;`,NX=Ze({opSnippet:IX}),TX={kernelName:$o,backendName:"webgl",kernelFunc:NX},kp=class{constructor(e,t,n,a=!1,r=!1){if(this.variableNames=["x"],t==="avg"&&n)throw new Error("Cannot compute positions for average pool.");let s=e.filterWidth,i=e.strideHeight,o=e.strideWidth,l=e.dilationHeight,c=e.dilationWidth,u=e.effectiveFilterHeight,p=e.effectiveFilterWidth,d=e.padInfo.top,h=e.padInfo.left;this.outputShape=e.outShape;let m=t==="avg",f=`((batch * ${e.inHeight} + xR) * ${e.inWidth} + xC) * ${e.inChannels} + d`,g=`(xR * ${e.inWidth} + xC) * ${e.inChannels} + d`,y="0.0";if(m||(y="-1.0 / 1e-20"),n){let C=">=";this.userCode=`
|
|
const ivec2 strides = ivec2(${i}, ${o});
|
|
const ivec2 pads = ivec2(${d}, ${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 < ${p};
|
|
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 = ${a?r?f:g:`wR * ${p} + wC`};
|
|
}
|
|
}
|
|
}
|
|
setOutput(float(minMaxPosition));
|
|
}
|
|
`;return}let b="max",v=`${t}(${t}(${t}(minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])`;t==="avg"&&(v="avgValue / count");let x=Math.floor(s/4)*4,N=s%4,T=`
|
|
if (${m}) {
|
|
avgValue += dot(values, ones);
|
|
} else {
|
|
minMaxValue = ${b}(values, minMaxValue);
|
|
}
|
|
`;this.userCode=`
|
|
const ivec2 strides = ivec2(${i}, ${o});
|
|
const ivec2 pads = ivec2(${d}, ${h});
|
|
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 xR, int xC, int d) {
|
|
if (xC < 0 || xC >= ${e.inWidth}) {
|
|
return initializationValue;
|
|
}
|
|
count += 1.0;
|
|
return getX(batch, xR, xC, d);
|
|
}
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int batch = coords[0];
|
|
int d = coords[3];
|
|
|
|
ivec2 xRCCorner = coords.yz * strides - pads;
|
|
int xRCorner = xRCCorner.x;
|
|
int xCCorner = xRCCorner.y;
|
|
|
|
// max/min x(?, ?, d) to get y(yR, yC, d).
|
|
// ? = to be determined
|
|
vec4 minMaxValue = vec4(${y});
|
|
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 < ${x}; 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)
|
|
);
|
|
|
|
${T}
|
|
}
|
|
|
|
int xC = xCCorner + ${x};
|
|
if (${N===1}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, xR, xC, d),
|
|
initializationValue,
|
|
initializationValue,
|
|
initializationValue
|
|
);
|
|
|
|
${T}
|
|
} else if (${N===2}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, xR, xC, d),
|
|
getValue(batch, xR, xC + ${c}, d),
|
|
initializationValue,
|
|
initializationValue
|
|
);
|
|
|
|
${T}
|
|
} else if (${N===3}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, xR, xC, d),
|
|
getValue(batch, xR, xC + ${c}, d),
|
|
getValue(batch, xR, xC + 2 * ${c}, d),
|
|
initializationValue
|
|
);
|
|
|
|
${T}
|
|
}
|
|
}
|
|
setOutput(${v});
|
|
}
|
|
`}},Bv=class{constructor(e,t,n,a=!1,r=!1){if(this.variableNames=["x"],t==="avg"&&n)throw new Error("Cannot compute positions for average pool.");let s=e.filterWidth,i=e.strideDepth,o=e.strideHeight,l=e.strideWidth,c=e.dilationDepth,u=e.dilationHeight,p=e.dilationWidth,d=e.effectiveFilterDepth,h=e.effectiveFilterHeight,m=e.effectiveFilterWidth,f=e.padInfo.front,g=e.padInfo.top,y=e.padInfo.left;this.outputShape=e.outShape;let b=t==="avg",v="0.0";if(b||(v="-1.0 / 1e-20"),n){let F=">=";this.userCode=`
|
|
const ivec3 strides =
|
|
ivec3(${i}, ${o}, ${l});
|
|
const ivec3 pads = ivec3(${f}, ${g}, ${y});
|
|
|
|
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 < ${d};
|
|
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 < ${m};
|
|
wC += ${p}) {
|
|
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 ${F} currMinMaxValue) {
|
|
minMaxValue = value;
|
|
minMaxValueFound = 1.0;
|
|
minMaxPosition = ${a?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} * ${m} +
|
|
wR * ${m} + wC`};
|
|
}
|
|
}
|
|
}
|
|
}
|
|
setOutput(float(minMaxPosition));
|
|
}
|
|
`;return}let x="max",N=`${t}(${t}(${t}(minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])`;t==="avg"&&(N="avgValue / count");let T=Math.floor(s/4)*4,C=s%4,$=`
|
|
if (${b}) {
|
|
avgValue += dot(values, ones);
|
|
} else {
|
|
minMaxValue = ${x}(values, minMaxValue);
|
|
}
|
|
`;this.userCode=`
|
|
const ivec3 strides =
|
|
ivec3(${i}, ${o}, ${l});
|
|
const ivec3 pads = ivec3(${f}, ${g}, ${y});
|
|
const float initializationValue = ${v};
|
|
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 >= ${e.inWidth}) {
|
|
return initializationValue;
|
|
}
|
|
count += 1.0;
|
|
return getX(batch, xD, xR, xC, ch);
|
|
}
|
|
|
|
void main() {
|
|
ivec5 coords = getOutputCoords();
|
|
int batch = coords.x;
|
|
int ch = coords.u;
|
|
|
|
ivec3 xCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;
|
|
int xDCorner = xCorner.x;
|
|
int xRCorner = xCorner.y;
|
|
int xCCorner = xCorner.z;
|
|
|
|
// max/min x(?, ?, ?, d) to get y(yD, yR, yC, ch).
|
|
// ? = to be determined
|
|
vec4 minMaxValue = vec4(${v});
|
|
float avgValue = 0.0;
|
|
count = 0.0;
|
|
|
|
for (int wD = 0; wD < ${d};
|
|
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 < ${T}; wC += 4) {
|
|
int xC = xCCorner + wC * ${p};
|
|
|
|
vec4 values = vec4(
|
|
getValue(batch, xD, xR, xC, ch),
|
|
getValue(batch, xD, xR, xC + ${p}, ch),
|
|
getValue(batch, xD, xR, xC + 2 * ${p}, ch),
|
|
getValue(batch, xD, xR, xC + 3 * ${p}, ch)
|
|
);
|
|
|
|
${$}
|
|
}
|
|
|
|
int xC = xCCorner + ${T};
|
|
if (${C===1}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, xD, xR, xC, ch),
|
|
initializationValue,
|
|
initializationValue,
|
|
initializationValue
|
|
);
|
|
|
|
${$}
|
|
} else if (${C===2}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, xD, xR, xC, ch),
|
|
getValue(batch, xD, xR, xC + ${p}, ch),
|
|
initializationValue,
|
|
initializationValue
|
|
);
|
|
|
|
${$}
|
|
} else if (${C===3}) {
|
|
vec4 values = vec4(
|
|
getValue(batch, xD, xR, xC, ch),
|
|
getValue(batch, xD, xR, xC + ${p}, ch),
|
|
getValue(batch, xD, xR, xC + 2 * ${p}, ch),
|
|
initializationValue
|
|
);
|
|
|
|
${$}
|
|
}
|
|
}
|
|
setOutput(${N});
|
|
}
|
|
}
|
|
`}};function SX(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t;xp(r,"avgPool");let{filterSize:s,strides:i,pad:o,dimRoundingMode:l}=a,c=1;k.assert(_.eitherStridesOrDilationsAreOne(i,c),()=>`Error in avgPool: Either strides or dilations must be 1. Got strides ${i} and dilations '${c}'`);let u=_.computePool2DInfo(r.shape,s,i,c,o,l);if(u.filterWidth===1&&u.filterHeight===1&&k.arraysEqual(u.inShape,u.outShape))return Bn({inputs:{x:r},backend:n});let p=new kp(u,"avg",!1);return n.runWebGLProgram(p,[r],"float32")}var CX={kernelName:_s,backendName:"webgl",kernelFunc:SX};function _X(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{filterSize:s,strides:i,pad:o,dimRoundingMode:l,dataFormat:c}=a,u=[1,1,1],p=_.computePool3DInfo(r.shape,s,i,u,o,l,c),d=new Bv(p,"avg",!1);return n.runWebGLProgram(d,[r],"float32")}var EX={kernelName:uc,backendName:"webgl",kernelFunc:_X},AX=class{constructor(e){this.variableNames=["dy"],this.outputShape=e.inShape;let t=e.filterHeight,n=e.filterWidth,a=e.strideHeight,r=e.strideWidth,s=e.dilationHeight,i=e.dilationWidth,o=e.effectiveFilterHeight,l=e.effectiveFilterWidth,c=o-1-e.padInfo.top,u=l-1-e.padInfo.left,p=1/(t*n);this.userCode=`
|
|
const ivec2 pads = ivec2(${c}, ${u});
|
|
const float avgMultiplier = float(${p});
|
|
|
|
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 < ${o};
|
|
wR += ${s}) {
|
|
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 < ${l};
|
|
wC+= ${i}) {
|
|
float dyC = float(dyCCorner + wC) / ${r}.0;
|
|
|
|
if (dyC < 0.0 || dyC >= ${e.outWidth}.0 ||
|
|
fract(dyC) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyC = int(dyC);
|
|
|
|
float dyValue = getDy(b, idyR, idyC, d);
|
|
|
|
dotProd += dyValue * avgMultiplier;
|
|
}
|
|
}
|
|
setOutput(dotProd);
|
|
}
|
|
`}},FX=class{constructor(e){this.variableNames=["dy"],this.outputShape=e.inShape;let t=e.filterDepth,n=e.filterHeight,a=e.filterWidth,r=e.strideDepth,s=e.strideHeight,i=e.strideWidth,o=e.dilationDepth,l=e.dilationHeight,c=e.dilationWidth,u=e.effectiveFilterDepth,p=e.effectiveFilterHeight,d=e.effectiveFilterWidth,h=u-1-e.padInfo.front,m=p-1-e.padInfo.top,f=d-1-e.padInfo.left,g=1/(t*n*a);this.userCode=`
|
|
const ivec3 pads = ivec3(${h}, ${m}, ${f});
|
|
const float avgMultiplier = float(${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 += ${o}) {
|
|
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 < ${p};
|
|
wR += ${l}) {
|
|
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 < ${d};
|
|
wC += ${c}) {
|
|
float dyC = float(dyCCorner + wC) / ${i}.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 $X(e){let{inputs:t,backend:n,attrs:a}=e,{dy:r,input:s}=t,i=s,{filterSize:o,strides:l,pad:c,dimRoundingMode:u}=a,p=[1,1,1],d=_.computePool3DInfo(i.shape,o,l,p,c,u),h=new FX(d);return n.runWebGLProgram(h,[r],i.dtype)}var DX={kernelName:vd,backendName:"webgl",kernelFunc:$X};function RX(e){let{inputs:t,backend:n,attrs:a}=e,{dy:r,input:s}=t,i=s;xp([r,s],"avgPoolGrad");let{filterSize:o,strides:l,pad:c}=a,u=_.computePool2DInfo(i.shape,o,l,1,c),p=new AX(u);return n.runWebGLProgram(p,[r],i.dtype)}var MX={kernelName:xd,backendName:"webgl",kernelFunc:RX};function PX(e){let{inputs:t,backend:n,attrs:a}=e,{a:r,b:s}=t,{transposeA:i,transposeB:o}=a;return Um({a:r,b:s,transposeA:i,transposeB:o,backend:n})}var OX={kernelName:Es,backendName:"webgl",kernelFunc:PX},LX=class{constructor(e,t,n,a,r,s){this.outputShape=[],this.variableNames=["x","mean","variance"],_.assertAndGetBroadcastShape(e,t),_.assertAndGetBroadcastShape(e,n);let i="0.0";a!=null&&(_.assertAndGetBroadcastShape(e,a),this.variableNames.push("offset"),i="getOffsetAtOutCoords()");let o="1.0";r!=null&&(_.assertAndGetBroadcastShape(e,r),this.variableNames.push("scale"),o="getScaleAtOutCoords()"),this.outputShape=e,this.userCode=`
|
|
void main() {
|
|
float x = getXAtOutCoords();
|
|
float mean = getMeanAtOutCoords();
|
|
float variance = getVarianceAtOutCoords();
|
|
float offset = ${i};
|
|
float scale = ${o};
|
|
float inv = scale * inversesqrt(variance + float(${s}));
|
|
setOutput(dot(vec3(x, -mean, offset), vec3(inv, inv, 1)));
|
|
}
|
|
`}},zX=class{constructor(e,t,n,a,r,s){this.packedInputs=!0,this.packedOutput=!0,this.variableNames=["x","mean","variance"],_.assertAndGetBroadcastShape(e,t),_.assertAndGetBroadcastShape(e,n);let i="vec4(0.0)";a!=null&&(_.assertAndGetBroadcastShape(e,a),this.variableNames.push("offset"),i="getOffsetAtOutCoords()");let o="vec4(1.0)";r!=null&&(_.assertAndGetBroadcastShape(e,r),this.variableNames.push("scale"),o="getScaleAtOutCoords()"),this.outputShape=e,this.userCode=`
|
|
void main() {
|
|
vec4 offset = ${i};
|
|
vec4 scale = ${o};
|
|
|
|
vec4 x = getXAtOutCoords();
|
|
vec4 mean = getMeanAtOutCoords();
|
|
vec4 variance = getVarianceAtOutCoords();
|
|
|
|
vec4 inv = scale * inversesqrt(variance + vec4(${s}));
|
|
|
|
setOutput((x - mean) * inv + offset);
|
|
}
|
|
`}},BX=({inputs:e,backend:t,attrs:n})=>{let{x:a,mean:r,variance:s,offset:i,scale:o}=e;k.assert(r.shape.length===s.shape.length,()=>"Batch normalization gradient requires mean and variance to have equal ranks."),k.assert(i==null||r.shape.length===i.shape.length,()=>"Batch normalization gradient requires mean and offset to have equal ranks."),k.assert(o==null||r.shape.length===o.shape.length,()=>"Batch normalization gradient requires mean and scale to have equal ranks.");let{varianceEpsilon:l}=n;l==null&&(l=.001);let c=[a,r,s],u=null;i!=null&&(u=i.shape,c.push(i));let p=null;o!=null&&(p=o.shape,c.push(o));let d=ee().getBool("WEBGL_PACK_NORMALIZATION")?new zX(a.shape,r.shape,s.shape,u,p,l):new LX(a.shape,r.shape,s.shape,u,p,l);return t.runWebGLProgram(d,c,c[0].dtype)},WX={kernelName:Bs,backendName:"webgl",kernelFunc:BX},UX=class{constructor(e){this.variableNames=["source"],this.outputShape=e,this.rank=e.length;let t=ft(this.rank),n=`uniform int start[${this.rank}];`,a=VX(this.rank),r,s=e.map((i,o)=>`sourceLoc.${Wv[o]} = start[${o}] + coords.${Wv[o]};`);r=`
|
|
${t} sourceLoc;
|
|
${t} coords = getOutputCoords();
|
|
${s.join(`
|
|
`)}
|
|
`,this.userCode=`
|
|
${n}
|
|
void main() {
|
|
${r}
|
|
setOutput(getSource(${a}));
|
|
}
|
|
`}getCustomSetupFunc(e){if(e.length!==this.rank)throw Error(`The rank (${this.rank}) of the program must match the length of start (${e.length})`);return(t,n)=>{this.startLoc==null&&(this.startLoc=t.getUniformLocationNoThrow(n,"start"),this.startLoc==null)||t.gl.uniform1iv(this.startLoc,e)}}},Wv=["x","y","z","w","u","v"];function VX(e){if(e===1)return"sourceLoc";if(e<=6)return Wv.slice(0,e).map(t=>"sourceLoc."+t).join(",");throw Error(`Slicing for rank ${e} is not yet supported`)}var GX=class{constructor(e){this.variableNames=["source"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e,this.rank=e.length;let t=ft(this.rank),n=mn("coords",this.rank),a=mn("sourceLoc",this.rank),r=this.rank===1?"sourceLoc":`vec2(${a.slice(-2).join()})`,s=`getChannel(getSource(${a.join()}), ${r})`,i=`
|
|
result.x = ${s};
|
|
if (++${n[this.rank-1]} < ${e[this.rank-1]}) {
|
|
++${a[this.rank-1]};
|
|
result.y = ${s};
|
|
--${a[this.rank-1]};
|
|
}
|
|
`,o=this.rank===1?"":`
|
|
--${n[this.rank-1]};
|
|
if (++${n[this.rank-2]} < ${e[this.rank-2]}) {
|
|
++${a[this.rank-2]};
|
|
result.z = ${s};
|
|
if (++${n[this.rank-1]} < ${e[this.rank-1]}) {
|
|
++${a[this.rank-1]};
|
|
result.w = ${s};
|
|
}
|
|
}
|
|
`,l=this.rank<=4?`sourceLoc = coords +
|
|
${t}(${e.map((c,u)=>`start[${u}]`).join()});`:e.map((c,u)=>`${a[u]} = ${n[u]} + start[${u}];`).join(`
|
|
`);this.userCode=`
|
|
uniform int start[${this.rank}];
|
|
void main() {
|
|
${t} coords = getOutputCoords();
|
|
${t} sourceLoc;
|
|
${l}
|
|
vec4 result = vec4(0.);
|
|
${i}
|
|
${o}
|
|
setOutput(result);
|
|
}
|
|
`}getCustomSetupFunc(e){if(e.length!==this.rank)throw Error(`The rank (${this.rank}) of the program must match the length of start (${e.length})`);return(t,n)=>{this.startLoc==null&&(this.startLoc=t.getUniformLocationNoThrow(n,"start"),this.startLoc==null)||t.gl.uniform1iv(this.startLoc,e)}}};function HX(e,t,n,a){let r=a.texData.get(e.dataId),s=a.makeTensorInfo(n,e.dtype),i=a.texData.get(s.dataId);Object.assign(i,r),i.complexParentRefCount=0,i.refCount=1,i.shape=n,i.dtype=e.dtype;let o=cn.computeFlatOffset(t,k.computeStrides(e.shape));r.slice&&(o+=r.slice.flatOffset),i.slice={flatOffset:o,origDataId:r.slice&&r.slice.origDataId||e.dataId};let l=a.dataRefCount.get(i.slice.origDataId)||1;return a.dataRefCount.set(i.slice.origDataId,l+1),s}function Ip(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{begin:s,size:i}=a,[o,l]=cn.parseSliceParams(r,s,i);if(cn.assertParamsValid(r,o,l),k.sizeFromShape(l)===0)return n.makeTensorInfo(l,r.dtype,[]);if(n.shouldExecuteOnCPU([r])||r.dtype==="string"){let p=n.texData.get(r.dataId),d=L8(p.values,o,l,r.shape,r.dtype);return n.makeTensorInfo(l,r.dtype,d)}let{isPacked:c}=n.texData.get(r.dataId),u=cn.isSliceContinous(r.shape,o,l);if(c||!u){let p=ee().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new GX(l):new UX(l),d=p.getCustomSetupFunc(o);return n.runWebGLProgram(p,[r],r.dtype,d)}return n.uploadToGPU(r.dataId),HX(r,o,l,n)}var jX={kernelName:gl,backendName:"webgl",kernelFunc:Ip},qX=e=>{let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{blockShape:s,crops:i}=a;k.assert(r.shape.length<=4,()=>"batchToSpaceND for rank > 4 with a WebGL backend not implemented yet");let o=s.reduce((b,v)=>b*v),l=_.getReshaped(r.shape,s,o),c=_.getPermuted(l.length,s.length),u=_.getReshapedPermuted(r.shape,s,o),p=_.getSliceBeginCoords(i,s.length),d=_.getSliceSize(u,i,s.length),h=[],m=be({inputs:{x:r},backend:n,attrs:{shape:l}}),f=Tn({inputs:{x:m},backend:n,attrs:{perm:c}}),g=be({inputs:{x:f},backend:n,attrs:{shape:u}}),y=Ip({inputs:{x:g},backend:n,attrs:{begin:p,size:d}});return h.push(m),h.push(f),h.push(g),h.forEach(b=>n.disposeIntermediateTensorInfo(b)),y},KX={kernelName:cc,backendName:"webgl",kernelFunc:qX};function XX(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,weights:s}=t,{size:i}=a,o=n.readSync(r.dataId),l=n.readSync(s.dataId),c=GT(o,l,s.dtype,s.shape,i);return n.makeTensorInfo([i],s.dtype,c)}var YX={kernelName:wd,backendName:"webgl",kernelFunc:XX},ZX="return float(a != b);",gS=nn({opSnippet:ZX,dtype:"bool"}),JX={kernelName:rl,backendName:"webgl",kernelFunc:gS};function Np(e){let{inputs:t,backend:n}=e,{input:a}=t,r=n.texData.get(a.dataId);return Bn({inputs:{x:r.complexTensorInfos.real},backend:n})}var QX={kernelName:Vd,backendName:"webgl",kernelFunc:Np},eY="return float(int(x));";function tY(e,t){let n=new cs(e.shape,eY),a=t.runWebGLProgram(n,[e],"int32");return{dataId:a.dataId,shape:a.shape,dtype:a.dtype}}function Vv(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{dtype:s}=a;if(s==="complex64"){if(r.dtype==="complex64")return Bn({inputs:{x:r},backend:n});let i=vt(r.shape),o=Vv({inputs:{x:r},backend:n,attrs:{dtype:"float32"}}),l=ps({inputs:{real:o,imag:i},backend:n});return i.dispose(),n.disposeIntermediateTensorInfo(o),l}if(r.dtype==="complex64"){let i=Np({inputs:{input:r},backend:n}),o=Vv({inputs:{x:i},backend:n,attrs:{dtype:s}});return n.disposeIntermediateTensorInfo(i),o}if(!k.hasEncodingLoss(r.dtype,s)){let i=Bn({inputs:{x:r},backend:n});return{dataId:i.dataId,shape:i.shape,dtype:s}}if(s==="int32")return tY(r,n);if(s==="bool"){let i=n.makeTensorInfo([],"bool",k.getTypedArrayFromDType("bool",1)),o=gS({inputs:{a:r,b:i},backend:n});return n.disposeIntermediateTensorInfo(i),o}throw new Error(`Error in Cast: failed to cast ${r.dtype} to ${s}`)}var nY={kernelName:As,backendName:"webgl",kernelFunc:Vv},yS="return ceil(x);",aY=Ze({opSnippet:yS,packedOpSnippet:yS,cpuKernelImpl:v8}),rY={kernelName:Ro,backendName:"webgl",kernelFunc:aY},sY=class{constructor(e){this.variableNames=["A"],this.outputShape=e,this.userCode=`
|
|
uniform float minVal;
|
|
uniform float maxVal;
|
|
|
|
void main() {
|
|
float value = getAAtOutCoords();
|
|
if (isnan(value)) {
|
|
setOutput(value);
|
|
return;
|
|
}
|
|
|
|
setOutput(clamp(value, minVal, maxVal));
|
|
}
|
|
`}getCustomSetupFunc(e,t){return(n,a)=>{this.minLoc==null&&(this.minLoc=n.getUniformLocationNoThrow(a,"minVal"),this.maxLoc=n.getUniformLocationNoThrow(a,"maxVal")),n.gl.uniform1f(this.minLoc,e),n.gl.uniform1f(this.maxLoc,t)}}},iY=class{constructor(e){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e,this.userCode=`
|
|
uniform float minVal;
|
|
uniform float maxVal;
|
|
|
|
void main() {
|
|
vec4 value = getAAtOutCoords();
|
|
|
|
if (any(isnan(value))) {
|
|
setOutput(value);
|
|
return;
|
|
}
|
|
|
|
setOutput(clamp(value, vec4(minVal), vec4(maxVal)));
|
|
}
|
|
`}getCustomSetupFunc(e,t){return(n,a)=>{this.minLoc==null&&(this.minLoc=n.getUniformLocationNoThrow(a,"minVal"),this.maxLoc=n.getUniformLocationNoThrow(a,"maxVal")),n.gl.uniform1f(this.minLoc,e),n.gl.uniform1f(this.maxLoc,t)}}};function oY(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{clipValueMin:s,clipValueMax:i}=a,o;ee().getBool("WEBGL_PACK_CLIP")?o=new iY(r.shape):o=new sY(r.shape);let l=o.getCustomSetupFunc(s,i);return n.runWebGLProgram(o,[r],r.dtype,l)}var lY={kernelName:zr,backendName:"webgl",kernelFunc:oY},uY=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 bS(e,t){return{dataId:t.dataId,dtype:t.dtype,shape:e.shape}}function cY(e){let{inputs:t,backend:n}=e,{x:a}=t,r=n.texData.get(a.dataId),s=new uY(a.shape),i=[bS(a,r.complexTensorInfos.real),bS(a,r.complexTensorInfos.imag)];return n.runWebGLProgram(s,i,i[0].dtype)}var pY={kernelName:pc,backendName:"webgl",kernelFunc:cY},dY=class{constructor(e){this.outputShape=[],this.outputShape=_.computeOutShape(e,1),this.variableNames=e.map((s,i)=>`T${i}`);let t=new Array(e.length-1);t[0]=e[0][1];for(let s=1;s<t.length;s++)t[s]=t[s-1]+e[s][1];let n=[`if (yC < ${t[0]}) setOutput(getT0(yR, yC));`];for(let s=1;s<t.length;s++){let i=t[s-1];n.push(`else if (yC < ${t[s]}) setOutput(getT${s}(yR, yC-${i}));`)}let a=t.length,r=t[t.length-1];n.push(`else setOutput(getT${a}(yR, yC-${r}));`),this.userCode=`
|
|
void main() {
|
|
ivec2 coords = getOutputCoords();
|
|
int yR = coords.x;
|
|
int yC = coords.y;
|
|
|
|
${n.join(`
|
|
`)}
|
|
}
|
|
`}},hY=class{constructor(e,t){this.packedInputs=!0,this.packedOutput=!0,this.outputShape=[],this.outputShape=_.computeOutShape(e,t);let n=this.outputShape,a=n.length,r=ft(a),s=mn("coords",a),i=["x","y","z","w","u","v"].slice(0,a);this.variableNames=e.map((m,f)=>`T${f}`);let o=new Array(e.length-1);o[0]=e[0][t];for(let m=1;m<o.length;m++)o[m]=o[m-1]+e[m][t];let l=i[t],c=i.slice(-2),u=i.join(),p=`if (${l} < ${o[0]}) {
|
|
return getChannel(
|
|
getT0(${u}), vec2(${c.join()}));
|
|
}`;for(let m=1;m<o.length;m++){let f=o[m-1];p+=`
|
|
if (${l} < ${o[m]} && ${l} >= ${o[m-1]}) {
|
|
return getChannel(
|
|
getT${m}(${Hm(i,l,f)}),
|
|
vec2(${Hm(c,l,f)}));
|
|
}`}let d=o.length,h=o[o.length-1];p+=`
|
|
return getChannel(
|
|
getT${d}(${Hm(i,l,h)}),
|
|
vec2(${Hm(c,l,h)}));`,this.userCode=`
|
|
float getValue(${i.map(m=>"int "+m)}) {
|
|
${p}
|
|
}
|
|
|
|
void main() {
|
|
${r} coords = getOutputCoords();
|
|
vec4 result = vec4(getValue(${s}), 0., 0., 0.);
|
|
|
|
${s[a-1]} = ${s[a-1]} + 1;
|
|
if (${s[a-1]} < ${n[a-1]}) {
|
|
result.g = getValue(${s});
|
|
}
|
|
|
|
${s[a-2]} = ${s[a-2]} + 1;
|
|
if (${s[a-2]} < ${n[a-2]}) {
|
|
result.a = getValue(${s});
|
|
}
|
|
|
|
${s[a-1]} = ${s[a-1]} - 1;
|
|
if (${s[a-2]} < ${n[a-2]} &&
|
|
${s[a-1]} < ${n[a-1]}) {
|
|
result.b = getValue(${s});
|
|
}
|
|
setOutput(result);
|
|
}
|
|
`}};function Hm(e,t,n){let a=e.indexOf(t);return e.map((r,s)=>s===a?`${r} - ${n}`:r).join()}function jm(e){let{inputs:t,backend:n}=e,{input:a}=t,r=n.texData.get(a.dataId);return Bn({inputs:{x:r.complexTensorInfos.imag},backend:n})}var mY={kernelName:Md,backendName:"webgl",kernelFunc:jm};function mu(e,t,n){let a=e[0].dtype;if(a==="complex64"){let c=e.map(m=>Np({inputs:{input:m},backend:n})),u=e.map(m=>jm({inputs:{input:m},backend:n})),p=mu(c,t,n),d=mu(u,t,n),h=ps({inputs:{real:p,imag:d},backend:n});return c.forEach(m=>n.disposeIntermediateTensorInfo(m)),u.forEach(m=>n.disposeIntermediateTensorInfo(m)),n.disposeIntermediateTensorInfo(p),n.disposeIntermediateTensorInfo(d),h}if(a==="string"){let{tensors2D:c,outShape:u}=xS(e,t,n),p=c.map(g=>({vals:n.readSync(g.dataId),shape:g.shape})),d=c[0].shape[0]===1,h=w8(p,u,a,d),m=_.computeOutShape(e.map(g=>g.shape),t),f=n.makeTensorInfo(m,a,h);return c.forEach(g=>n.disposeIntermediateTensorInfo(g)),f}if(e.length>ee().getNumber("WEBGL_MAX_TEXTURES_IN_SHADER")){let c=Math.floor(e.length/2),u=mu(e.slice(0,c),t,n),p=mu(e.slice(c),t,n),d=mu([u,p],t,n);return n.disposeIntermediateTensorInfo(u),n.disposeIntermediateTensorInfo(p),d}if(ee().getBool("WEBGL_PACK_ARRAY_OPERATIONS")&&e[0].shape.length>1){let c=new hY(e.map(u=>u.shape),t);return n.runWebGLProgram(c,e,a)}let{tensors2D:r,outShape:s}=xS(e,t,n),i=new dY(r.map(c=>c.shape)),o=n.runWebGLProgram(i,r,a);r.forEach(c=>n.disposeIntermediateTensorInfo(c));let l=be({inputs:{x:o},attrs:{shape:s},backend:n});return n.disposeIntermediateTensorInfo(o),l}function xS(e,t,n){let a=_.computeOutShape(e.map(r=>r.shape),t);return{tensors2D:e.map(r=>be({inputs:{x:r},attrs:{shape:[-1,k.sizeFromShape(r.shape.slice(t))]},backend:n})),outShape:a}}function vS(e){let{inputs:t,backend:n,attrs:a}=e,{axis:r}=a,s=k.parseAxisParam(r,t[0].shape)[0],i=_.computeOutShape(t.map(c=>c.shape),s);if(k.sizeFromShape(i)===0)return n.makeTensorInfo(i,t[0].dtype,[]);let o=t.filter(c=>k.sizeFromShape(c.shape)>0);if(o.length===1)return Bn({inputs:{x:o[0]},backend:n});let l=o.map(c=>c.shape);return _.assertParamsConsistent(l,s),mu(o,s,n)}var fY={kernelName:Mo,backendName:"webgl",kernelFunc:vS},wS=class{constructor(e,t=!1,n=null,a=!1,r=!1){this.variableNames=["x","W"],this.outputShape=e.outShape;let s=e.padInfo.top,i=e.padInfo.left,o=e.strideHeight,l=e.strideWidth,c=e.dilationHeight,u=e.dilationWidth,p=e.filterHeight,d=e.filterWidth,h=Math.floor(e.inChannels/4)*4,m=e.inChannels%4,f=e.dataFormat==="channelsLast",g=f?1:2,y=f?2:3,b=f?3:1,v="",x="";n&&(a?v=`float activation(float a) {
|
|
float b = getPreluActivationWeightsAtOutCoords();
|
|
${n}
|
|
}`:r?v=`float activation(float a) {
|
|
float b = getLeakyreluAlphaAtOutCoords();
|
|
${n}
|
|
}`:v=`
|
|
float activation(float x) {
|
|
${n}
|
|
}
|
|
`,x="result = activation(result);");let N=t?"result += getBiasAtOutCoords();":"";t&&this.variableNames.push("bias"),a&&this.variableNames.push("preluActivationWeights"),r&&this.variableNames.push("leakyreluAlpha"),this.userCode=`
|
|
${v}
|
|
|
|
const ivec2 strides = ivec2(${o}, ${l});
|
|
const ivec2 pads = ivec2(${s}, ${i});
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int batch = coords[0];
|
|
int d2 = coords[${b}];
|
|
|
|
ivec2 xRCCorner =
|
|
ivec2(coords[${g}], coords[${y}]) * 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 < ${p}; wR++) {
|
|
int xR = xRCorner + wR * ${c};
|
|
|
|
if (xR < 0 || xR >= ${e.inHeight}) {
|
|
continue;
|
|
}
|
|
|
|
for (int wC = 0; wC < ${d}; wC++) {
|
|
int xC = xCCorner + wC * ${u};
|
|
|
|
if (xC < 0 || xC >= ${e.inWidth}) {
|
|
continue;
|
|
}
|
|
|
|
for (int d1 = 0; d1 < ${h}; d1 += 4) {
|
|
vec4 wValues = vec4(
|
|
getW(wR, wC, d1, d2),
|
|
getW(wR, wC, d1 + 1, d2),
|
|
getW(wR, wC, d1 + 2, d2),
|
|
getW(wR, wC, d1 + 3, d2)
|
|
);
|
|
|
|
if (${f}) {
|
|
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 (${m===1}) {
|
|
|
|
if (${f}) {
|
|
dotProd +=
|
|
getX(batch, xR, xC, ${h}) *
|
|
getW(wR, wC, ${h}, d2);
|
|
} else {
|
|
dotProd +=
|
|
getX(batch, ${h}, xR, xC) *
|
|
getW(wR, wC, ${h}, d2);
|
|
}
|
|
|
|
} else if (${m===2}) {
|
|
vec2 wValues = vec2(
|
|
getW(wR, wC, ${h}, d2),
|
|
getW(wR, wC, ${h} + 1, d2)
|
|
);
|
|
|
|
if (${f}) {
|
|
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, ${h} + 1, xR, xC)
|
|
);
|
|
dotProd += dot(xValues, wValues);
|
|
}
|
|
|
|
} else if (${m===3}) {
|
|
vec3 wValues = vec3(
|
|
getW(wR, wC, ${h}, d2),
|
|
getW(wR, wC, ${h} + 1, d2),
|
|
getW(wR, wC, ${h} + 2, d2)
|
|
);
|
|
|
|
if (${f}) {
|
|
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;
|
|
${N}
|
|
${x}
|
|
setOutput(result);
|
|
}
|
|
`}},gY=class{constructor(e){this.variableNames=["x","W"],this.outputShape=e.outShape;let t=e.padInfo.front,n=e.padInfo.top,a=e.padInfo.left,r=e.strideDepth,s=e.strideHeight,i=e.strideWidth,o=e.dilationDepth,l=e.dilationHeight,c=e.dilationWidth,u=e.filterDepth,p=e.filterHeight,d=e.filterWidth,h=Math.floor(e.inChannels/4)*4,m=e.inChannels%4;this.userCode=`
|
|
const ivec3 strides = ivec3(${r}, ${s}, ${i});
|
|
const ivec3 pads = ivec3(${t}, ${n}, ${a});
|
|
|
|
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 * ${o};
|
|
|
|
if (xF < 0 || xF >= ${e.inDepth}) {
|
|
continue;
|
|
}
|
|
|
|
for (int wR = 0; wR < ${p}; wR++) {
|
|
int xR = xRCorner + wR * ${l};
|
|
|
|
if (xR < 0 || xR >= ${e.inHeight}) {
|
|
continue;
|
|
}
|
|
|
|
for (int wC = 0; wC < ${d}; wC++) {
|
|
int xC = xCCorner + wC * ${c};
|
|
|
|
if (xC < 0 || xC >= ${e.inWidth}) {
|
|
continue;
|
|
}
|
|
|
|
for (int d1 = 0; d1 < ${h}; d1 += 4) {
|
|
vec4 xValues = vec4(
|
|
getX(batch, xF, xR, xC, d1),
|
|
getX(batch, xF, xR, xC, d1 + 1),
|
|
getX(batch, xF, xR, xC, d1 + 2),
|
|
getX(batch, xF, xR, xC, d1 + 3)
|
|
);
|
|
vec4 wValues = vec4(
|
|
getW(wF, wR, wC, d1, d2),
|
|
getW(wF, wR, wC, d1 + 1, d2),
|
|
getW(wF, wR, wC, d1 + 2, d2),
|
|
getW(wF, wR, wC, d1 + 3, d2)
|
|
);
|
|
|
|
dotProd += dot(xValues, wValues);
|
|
}
|
|
|
|
if (${m===1}) {
|
|
dotProd +=
|
|
getX(batch, xF, xR, xC, ${h}) *
|
|
getW(wF, wR, wC, ${h}, d2);
|
|
} else if (${m===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, ${h} + 1, d2)
|
|
);
|
|
dotProd += dot(xValues, wValues);
|
|
} else if (${m===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);
|
|
}
|
|
`}},yY=class{constructor(e,t,n){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e;let{filterWidth:a,inChannels:r,strideWidth:s,strideHeight:i,padInfo:o,outWidth:l,dilationWidth:c,dilationHeight:u,dataFormat:p}=n,{left:d,top:h}=o,m=r*a,f=hn(),g=p==="channelsLast",y=g?0:1,b=g?1:2,v="";for(let x=0;x<=1;x++)for(let N=0;N<=1;N++)v+=`
|
|
blockIndex = rc.y + ${N};
|
|
pos = rc.x + ${x};
|
|
|
|
if(blockIndex < ${e[1]} && pos < ${e[0]}) {
|
|
offsetY = int(blockIndex / (${l})) * ${i} - ${h};
|
|
d0 = offsetY + ${u} * (pos / ${m});
|
|
|
|
if(d0 < ${t[y]} && d0 >= 0) {
|
|
|
|
offsetX = int(mod(float(blockIndex), ${l}.) * ${s}. - ${d}.);
|
|
d1 = offsetX + ${c} * (int(mod(float(pos), ${m}.) / ${r}.));
|
|
|
|
if(d1 < ${t[b]} && d1 >= 0) {
|
|
|
|
ch = int(mod(float(pos), ${r}.));
|
|
|
|
if (${g}) {
|
|
innerDims = vec2(d1, ch);
|
|
result[${x*2+N}] = getChannel(
|
|
getA(d0, int(innerDims.x),
|
|
int(innerDims.y)), innerDims);
|
|
} else {
|
|
innerDims = vec2(d0, d1);
|
|
result[${x*2+N}] = 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;
|
|
|
|
${v}
|
|
|
|
${f.output} = result;
|
|
}
|
|
`}};function kS({x:e,filter:t,convInfo:n,backend:a,bias:r=null,preluActivationWeights:s=null,leakyreluAlpha:i=0,activation:o=null}){let l=e.shape,c=a.texData.get(e.dataId),u=n.inChannels,p=l[0]*l[1]*l[2],d=n.outChannels,h=n.dataFormat==="channelsLast",m=!1,f=!1,g,y=[],b=(p===1||d===1)&&u>cS,v=l[2]%2!=0&&!!c.isPacked;if(b||!ee().getBool("WEBGL_LAZILY_UNPACK")||!ee().getBool("WEBGL_PACK_BINARY_OPERATIONS")||!v){let x=h?l[0]*l[1]*l[2]:l[0]*l[2]*l[3],N=be({inputs:{x:e},backend:a,attrs:{shape:[1,x,n.inChannels]}}),T=be({inputs:{x:t},backend:a,attrs:{shape:[1,n.inChannels,n.outChannels]}}),C=Um({a:N,b:T,transposeA:m,transposeB:f,backend:a,bias:r,activation:o,preluActivationWeights:s,leakyreluAlpha:i});g=be({inputs:{x:C},backend:a,attrs:{shape:n.outShape}}),y.push(N),y.push(T),y.push(C)}else{let x=h?l[0]*l[1]*(l[2]+1):l[0]*l[2]*(l[3]+1),N={dataId:e.dataId,shape:[1,x,n.inChannels],dtype:e.dtype},T=c.shape;c.shape=c.shape.slice(),c.shape[c.shape.length-2]++,k.assert(Lm(c.shape,N.shape),()=>`packed reshape ${c.shape} to ${N.shape} isn't free`);let C=be({inputs:{x:t},backend:a,attrs:{shape:[1,n.inChannels,n.outChannels]}});y.push(C);let $=Um({a:N,b:C,backend:a,transposeA:m,transposeB:f,bias:r,activation:o,preluActivationWeights:s,leakyreluAlpha:i}),F=a.texData.get($.dataId);k.assert(F.isPacked,()=>"batchMatMul result is expected to be packed"),c.shape=T,F.shape=n.outShape,g=Bn({inputs:{x:$},backend:a}),g.shape=n.outShape,y.push($)}for(let x of y)a.disposeIntermediateTensorInfo(x);return g}function IS({x:e,filter:t,convInfo:n,backend:a,bias:r=null,preluActivationWeights:s=null,leakyreluAlpha:i=0,activation:o=null}){let{filterWidth:l,filterHeight:c,inChannels:u,outWidth:p,outHeight:d,dataFormat:h}=n,m=h==="channelsLast",f=l*c*u,g=d*p,y=[f,g],b=!0,v=!1,x=[],N=be({inputs:{x:e},backend:a,attrs:{shape:e.shape.slice(1)}}),T=be({inputs:{x:t},backend:a,attrs:{shape:[1,f,k.sizeFromShape(t.shape)/f]}});x.push(N),x.push(T);let C=new yY(y,N.shape,n),$=a.runWebGLProgram(C,[N],"float32"),F=be({inputs:{x:$},backend:a,attrs:{shape:[1,y[0],y[1]]}});x.push($),x.push(F);let O=r!=null,B=s!=null,W=o==="leakyrelu",G=o?Wm(o,!0):null,q=new rS(F.shape,T.shape,[1,g,n.outChannels],b,v,O,G,B,W),H=[F,T];if(r&&H.push(r),B&&H.push(s),W){let J=a.makeTensorInfo([],"float32",k.createScalarValue(i,"float32"));H.push(J),x.push(J)}let K=a.runWebGLProgram(q,H,"float32"),X=m?[1,d,p,n.outChannels]:[1,n.outChannels,d,p],ne=be({inputs:{x:K},backend:a,attrs:{shape:X}});x.push(K);for(let J of x)a.disposeIntermediateTensorInfo(J);return ne}function bY(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,filter:s}=t,{strides:i,pad:o,dataFormat:l,dilations:c,dimRoundingMode:u}=a,p=_.convertConv2DDataFormat(l),d=_.computeConv2DInfo(r.shape,s.shape,i,c,o,u,!1,p),h;if(d.filterHeight===1&&d.filterWidth===1&&d.dilationHeight===1&&d.dilationWidth===1&&d.strideHeight===1&&d.strideWidth===1&&(d.padInfo.type==="SAME"||d.padInfo.type==="VALID"))h=kS({x:r,filter:s,convInfo:d,backend:n});else if(ee().getBool("WEBGL_CONV_IM2COL")&&r.shape[0]===1)h=IS({x:r,filter:s,convInfo:d,backend:n});else{let f=new wS(d);h=n.runWebGLProgram(f,[r,s],"float32")}let m=be({inputs:{x:h},backend:n,attrs:{shape:d.outShape}});return n.disposeIntermediateTensorInfo(h),m}var xY={kernelName:Fs,backendName:"webgl",kernelFunc:bY},vY=class{constructor(e){this.variableNames=["x","dy"],this.outputShape=e.filterShape;let t=e.strideHeight,n=e.strideWidth,a=e.padInfo.top,r=e.padInfo.left,s=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} - ${a};
|
|
|
|
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 >= ${e.inWidth}) {
|
|
continue;
|
|
}
|
|
|
|
if (${s}) {
|
|
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);
|
|
}
|
|
`}},wY=class{constructor(e){this.variableNames=["dy","W"],this.outputShape=e.inShape;let t=e.filterHeight,n=e.filterWidth,a=e.strideHeight,r=e.strideWidth,s=e.dataFormat==="channelsLast",i=t-1-e.padInfo.top,o=n-1-e.padInfo.left,l=s?1:2,c=s?2:3,u=s?3:1;this.userCode=`
|
|
const ivec2 pads = ivec2(${i}, ${o});
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int batch = coords[0];
|
|
int d1 = coords[${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) / ${a}.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 < ${e.outChannels}; d2++) {
|
|
|
|
if (${s}) {
|
|
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);
|
|
}
|
|
`}},kY=class{constructor(e){this.variableNames=["x","dy"],this.outputShape=e.filterShape;let t=e.strideDepth,n=e.strideHeight,a=e.strideWidth,r=e.padInfo.front,s=e.padInfo.top,i=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} - ${s};
|
|
|
|
if (xR < 0 || xR >= ${e.inHeight}) {
|
|
continue;
|
|
}
|
|
|
|
for (int yC = 0; yC < ${e.outWidth}; yC++) {
|
|
int xC = wC + yC * ${a} - ${i};
|
|
|
|
if (xC < 0 || xC >= ${e.inWidth}) {
|
|
continue;
|
|
}
|
|
|
|
float dyValue = getDy(b, yF, yR, yC, d2);
|
|
float xValue = getX(b, xF, xR, xC, d1);
|
|
dotProd += (xValue * dyValue);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
setOutput(dotProd);
|
|
}
|
|
`}},IY=class{constructor(e){this.variableNames=["dy","W"],this.outputShape=e.inShape;let t=e.filterDepth,n=e.filterHeight,a=e.filterWidth,r=e.strideDepth,s=e.strideHeight,i=e.strideWidth,o=t-1-e.padInfo.front,l=n-1-e.padInfo.top,c=a-1-e.padInfo.left;this.userCode=`
|
|
const ivec3 pads = ivec3(${o}, ${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) / ${s}.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 < ${a}; wC++) {
|
|
float dyC = float(dyCCorner + wC) / ${i}.0;
|
|
|
|
if (dyC < 0.0 || dyC >= ${e.outWidth}.0 ||
|
|
fract(dyC) > 0.0) {
|
|
continue;
|
|
}
|
|
int idyC = int(dyC);
|
|
|
|
int wCPerm = ${a} - 1 - wC;
|
|
|
|
for (int d2 = 0; d2 < ${e.outChannels}; d2++) {
|
|
float xValue = getDy(batch, idyF, idyR, idyC, d2);
|
|
float wValue = getW(wFPerm, wRPerm, wCPerm, d1, d2);
|
|
dotProd += xValue * wValue;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
setOutput(dotProd);
|
|
}
|
|
`}};function NY(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,dy:s}=t,{strides:i,pad:o,dataFormat:l,dimRoundingMode:c,filterShape:u}=a,p=_.convertConv2DDataFormat(l),d=_.computeConv2DInfo(r.shape,u,i,1,o,c,!1,p),h=new vY(d);return n.runWebGLProgram(h,[r,s],"float32")}var TY={kernelName:Id,backendName:"webgl",kernelFunc:NY};function SY(e){let{inputs:t,backend:n,attrs:a}=e,{dy:r,filter:s}=t,{inputShape:i,strides:o,pad:l,dataFormat:c,dimRoundingMode:u}=a,p=_.convertConv2DDataFormat(c),d=_.computeConv2DInfo(i,s.shape,o,1,l,u,!1,p),h=new wY(d);return n.runWebGLProgram(h,[r,s],"float32")}var CY={kernelName:$s,backendName:"webgl",kernelFunc:SY};function _Y(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,filter:s}=t,{strides:i,pad:o,dilations:l}=a,c=_.computeConv3DInfo(r.shape,s.shape,i,l,o),u=new gY(c);return n.runWebGLProgram(u,[r,s],"float32")}var EY={kernelName:dc,backendName:"webgl",kernelFunc:_Y};function AY(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,dy:s}=t,{strides:i,pad:o,filterShape:l}=a,c=_.computeConv3DInfo(r.shape,l,i,1,o),u=new kY(c);return n.runWebGLProgram(u,[r,s],"float32")}var FY={kernelName:Nd,backendName:"webgl",kernelFunc:AY};function $Y(e){let{inputs:t,backend:n,attrs:a}=e,{dy:r,filter:s}=t,{pad:i,strides:o,inputShape:l}=a,c=_.computeConv3DInfo(l,s.shape,o,1,i),u=new IY(c);return n.runWebGLProgram(u,[r,s],"float32")}var DY={kernelName:Td,backendName:"webgl",kernelFunc:$Y},RY=aS+`
|
|
return cos(x);
|
|
`,MY=Ze({opSnippet:RY}),PY={kernelName:Ds,backendName:"webgl",kernelFunc:MY},OY=`
|
|
float e2x = exp(-x);
|
|
return (e2x + 1.0 / e2x) / 2.0;
|
|
`,LY=Ze({opSnippet:OY}),zY={kernelName:Po,backendName:"webgl",kernelFunc:LY},BY=class{constructor(e,t,n,a,r){this.variableNames=["Image","Boxes","BoxInd"],this.outputShape=[];let[s,i,o,l]=e,[c]=t,[u,p]=n;this.outputShape=[c,u,p,l];let d=a==="bilinear"?1:0,[h,m]=[`${i-1}.0`,`${o-1}.0`],[f,g,y]=u>1?[`${(i-1)/(u-1)}`,"(y2-y1) * height_ratio",`y1*${h} + float(y)*(height_scale)`]:["0.0","0.0",`0.5 * (y1+y2) * ${h}`],[b,v,x]=p>1?[`${(o-1)/(p-1)}`,"(x2-x1) * width_ratio",`x1*${m} + float(x)*(width_scale)`]:["0.0","0.0",`0.5 * (x1+x2) * ${m}`];this.userCode=`
|
|
const float height_ratio = float(${f});
|
|
const float width_ratio = float(${b});
|
|
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 >= ${s}) {
|
|
return;
|
|
}
|
|
|
|
float height_scale = ${g};
|
|
float width_scale = ${v};
|
|
|
|
float in_y = ${y};
|
|
if( in_y < 0.0 || in_y > ${h} ) {
|
|
setOutput(float(${r}));
|
|
return;
|
|
}
|
|
float in_x = ${x};
|
|
if( in_x < 0.0 || in_x > ${m} ) {
|
|
setOutput(float(${r}));
|
|
return;
|
|
}
|
|
|
|
vec2 sourceFracIndexCR = vec2(in_x,in_y);
|
|
if(${d} == 1) {
|
|
// Compute the four integer indices.
|
|
ivec2 sourceFloorCR = ivec2(sourceFracIndexCR);
|
|
ivec2 sourceCeilCR = ivec2(ceil(sourceFracIndexCR));
|
|
|
|
float topLeft = getImage(b, sourceFloorCR.y, sourceFloorCR.x, d);
|
|
float bottomLeft = getImage(b, sourceCeilCR.y, sourceFloorCR.x, d);
|
|
float topRight = getImage(b, sourceFloorCR.y, sourceCeilCR.x, d);
|
|
float bottomRight = getImage(b, sourceCeilCR.y, sourceCeilCR.x, d);
|
|
|
|
vec2 fracCR = sourceFracIndexCR - vec2(sourceFloorCR);
|
|
|
|
float top = topLeft + (topRight - topLeft) * fracCR.x;
|
|
float bottom = bottomLeft + (bottomRight - bottomLeft) * fracCR.x;
|
|
float newValue = top + (bottom - top) * fracCR.y;
|
|
setOutput(newValue);
|
|
} else {
|
|
// Compute the coordinators of nearest neighbor point.
|
|
ivec2 sourceNearestCR = ivec2(floor(
|
|
sourceFracIndexCR + vec2(0.5,0.5)));
|
|
float newValue = getImage(b, sourceNearestCR.y, sourceNearestCR.x, d);
|
|
setOutput(newValue);
|
|
}
|
|
}
|
|
`}},WY=e=>{let{inputs:t,backend:n,attrs:a}=e,{image:r,boxes:s,boxInd:i}=t,{cropSize:o,method:l,extrapolationValue:c}=a,u=new BY(r.shape,s.shape,o,l,c);return n.runWebGLProgram(u,[r,s,i],"float32")},VY={kernelName:Oo,backendName:"webgl",kernelFunc:WY},SS=class{constructor(e,t,n){this.variableNames=["x"],this.outputShape=e;let a=e.length,r=t?"0.0":`getX(${NS(a,"coords")})`,s=e[e.length-1],i="",o="";t?(i=n?`end != ${s-1}`:"end != 0",o=n?"end + 1":"end - 1"):(i=n?`end + pow2 < ${s}`:"end >= pow2",o=n?"end + pow2":"end - pow2"),this.userCode=`
|
|
uniform float index;
|
|
void main() {
|
|
${ft(a)} coords = getOutputCoords();
|
|
int end = ${TS(a,"coords")};
|
|
float val = ${r};
|
|
int pow2 = int(pow(2.0, index));
|
|
if (${i}) {
|
|
int idx = ${o};
|
|
${TS(a,"coords")} = idx;
|
|
val += getX(${NS(a,"coords")});
|
|
}
|
|
setOutput(val);
|
|
}
|
|
`}getCustomSetupFunc(e){return(t,n)=>{this.index==null&&(this.index=t.getUniformLocation(n,"index")),t.gl.uniform1f(this.index,e)}}};function NS(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 ${e} is not yet supported`)}function TS(e,t){if(e===1)return`${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 ${e} is not yet supported`)}function UY(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{axis:s,exclusive:i,reverse:o}=a,l=r.shape.length,c=_.getAxesPermutation([s],l),u=r;c!=null&&(u=Tn({inputs:{x:r},backend:n,attrs:{perm:c}}));let p=_.getInnerMostAxes(1,l)[0];if(p!==l-1)throw new Error(`WebGL cumsum shader expects an inner-most axis=${r.shape.length-1} but got axis=${s}`);let d=r.shape[p],h=Bn({inputs:{x:u},backend:n});for(let m=0;m<=Math.ceil(Math.log2(d))-1;m++){let f=new SS(u.shape,!1,o),g=f.getCustomSetupFunc(m),y=h;h=n.runWebGLProgram(f,[h],h.dtype,g),n.disposeIntermediateTensorInfo(y)}if(i){let m=new SS(u.shape,i,o),f=h;h=n.runWebGLProgram(m,[h],h.dtype),n.disposeIntermediateTensorInfo(f)}if(c!=null){let m=_.getUndoAxesPermutation(c),f=Tn({inputs:{x:h},backend:n,attrs:{perm:m}});return n.disposeIntermediateTensorInfo(h),n.disposeIntermediateTensorInfo(u),f}return h}var GY={kernelName:Rs,backendName:"webgl",kernelFunc:UY};function HY(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,weights:s}=t,{size:i,binaryOutput:o}=a;if(r.shape.length===1){let l=n.readSync(r.dataId),c=n.readSync(s.dataId),u=GT(l,c,s.dtype,s.shape,i);return n.makeTensorInfo([i],s.dtype,u)}else if(r.shape.length===2){let l=n.bufferSync(r),c=n.bufferSync(s),u=x8(l,c,i,o);return n.makeTensorInfo(u.shape,s.dtype,u.values)}throw new Error(`Error in denseBincount: input must be at most rank 2, but got rank${r.shape.length}.`)}var jY={kernelName:Sd,backendName:"webgl",kernelFunc:HY},qY=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 KY(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{blockSize:s,dataFormat:i}=a;k.assert(s>1,()=>`blockSize should be > 1 for depthToSpace, but was: ${s}`);let o=r.shape[0],l=i==="NHWC"?r.shape[1]:r.shape[2],c=i==="NHWC"?r.shape[2]:r.shape[3],u=i==="NHWC"?r.shape[3]:r.shape[1],p=l*s,d=c*s,h=u/(s*s),m=i==="NHWC"?[o,p,d,h]:[o,h,p,d],f=new qY(m,s,i);return n.runWebGLProgram(f,[r],r.dtype)}var XY={kernelName:Lo,backendName:"webgl",kernelFunc:KY},CS=class{constructor(e,t=!1,n=null,a=!1,r=!1){this.variableNames=["x","W"],this.outputShape=e.outShape;let s=e.inHeight,i=e.inWidth,o=e.padInfo.top,l=e.padInfo.left,c=e.strideHeight,u=e.strideWidth,p=e.dilationHeight,d=e.dilationWidth,h=e.filterHeight,m=e.filterWidth,f=e.outChannels/e.inChannels,g="",y="";n&&(a?g=`float activation(float a) {
|
|
float b = getPreluActivationWeightsAtOutCoords();
|
|
${n}
|
|
}`:r?g=`float activation(float a) {
|
|
float b = getLeakyreluAlphaAtOutCoords();
|
|
${n}
|
|
}`:g=`
|
|
float activation(float x) {
|
|
${n}
|
|
}
|
|
`,y="result = activation(result);");let b=t?"result += getBiasAtOutCoords();":"";t&&this.variableNames.push("bias"),a&&this.variableNames.push("preluActivationWeights"),r&&this.variableNames.push("leakyreluAlpha"),this.userCode=`
|
|
${g}
|
|
|
|
const ivec2 strides = ivec2(${c}, ${u});
|
|
const ivec2 pads = ivec2(${o}, ${l});
|
|
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int batch = coords.x;
|
|
ivec2 xRCCorner = coords.yz * strides - pads;
|
|
int d2 = coords.w;
|
|
int d1 = d2 / ${f};
|
|
int q = d2 - d1 * ${f};
|
|
|
|
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 < ${h}; wR++) {
|
|
int xR = xRCorner + wR * ${p};
|
|
|
|
if (xR < 0 || xR >= ${s}) {
|
|
continue;
|
|
}
|
|
|
|
for (int wC = 0; wC < ${m}; wC++) {
|
|
int xC = xCCorner + wC * ${d};
|
|
|
|
if (xC < 0 || xC >= ${i}) {
|
|
continue;
|
|
}
|
|
|
|
float xVal = getX(batch, xR, xC, d1);
|
|
float wVal = getW(wR, wC, d1, q);
|
|
dotProd += xVal * wVal;
|
|
}
|
|
}
|
|
|
|
float result = dotProd;
|
|
${b}
|
|
${y}
|
|
setOutput(result);
|
|
}
|
|
`}},_S=class{constructor(e,t=!1,n=null,a=!1,r=!1){this.variableNames=["x","W"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e.outShape;let s=e.inHeight,i=e.inWidth,o=e.padInfo.top,l=e.padInfo.left,c=e.strideHeight,u=e.strideWidth,p=e.dilationHeight,d=e.dilationWidth,h=e.filterHeight,m=e.filterWidth,f=m,g="int xR; int xC; int xCOffset;";for(let x=0;x<h;x++)for(let N=0;N<m;N++)g+=`
|
|
vec4 xTexelR${x}C${N*2} = vec4(0.);
|
|
vec4 wR${x}C${N} = vec4(0.);
|
|
vec4 xR${x}C${N} = vec4(0.);`;for(let x=0;x<h;x++)for(let N=0;N<f;N++){let T=N*2;if(g+=`
|
|
xR = xRCorner + ${x*p};
|
|
xC = xCCorner + ${T*d};
|
|
`,u===1){if(T<m&&(l%2==1?g+=`
|
|
xCOffset = xC + 1;
|
|
if(xR >= 0 && xR < ${s} && xCOffset >= 0 && xCOffset < ${i}) {
|
|
xTexelR${x}C${T} = getX(batch, xR, xCOffset, d1);
|
|
|
|
// Need to manually clear unused channels in case
|
|
// we're reading from recycled texture.
|
|
if(xCOffset + 1 >= ${i}) {
|
|
xTexelR${x}C${T}.zw = vec2(0.);
|
|
}
|
|
} else {
|
|
xTexelR${x}C${T} = vec4(0.);
|
|
}
|
|
|
|
xCOffset = xC + 1 - 2;
|
|
if(xR >= 0 && xR < ${s} && xCOffset >= 0 && xCOffset < ${i}) {
|
|
vec4 previous = getX(batch, xR, xCOffset, d1);
|
|
|
|
// Need to manually clear unused channels in case
|
|
// we're reading from recycled texture.
|
|
if(xCOffset + 1 >= ${i}) {
|
|
previous.zw = vec2(0.);
|
|
}
|
|
|
|
xR${x}C${T} = vec4(previous.zw, xTexelR${x}C${T}.xy);
|
|
} else {
|
|
xR${x}C${T} = vec4(0, 0, xTexelR${x}C${T}.xy);
|
|
}
|
|
`:g+=`
|
|
if(xR >= 0 && xR < ${s} && xC >= 0 && xC < ${i}) {
|
|
xTexelR${x}C${T} = getX(batch, xR, xC, d1);
|
|
} else {
|
|
xTexelR${x}C${T} = vec4(0.);
|
|
}
|
|
|
|
xR${x}C${T} = xTexelR${x}C${T};
|
|
`,T+1<m)){let C=l%2==0?k.nearestLargerEven(d):d;d%2==0&&l%2==1||d%2!=0&&l%2!=1?(g+=`
|
|
xCOffset = xC + ${l%2} + ${C};
|
|
|
|
if(xR >= 0 && xR < ${s} &&
|
|
xCOffset >= 0 && xCOffset < ${i}) {
|
|
xTexelR${x}C${T+2} = getX(batch, xR, xCOffset, d1);
|
|
}
|
|
`,d>1&&(g+=`
|
|
xCOffset -= 2;
|
|
if(xR >= 0 && xR < ${s} &&
|
|
xCOffset >= 0 && xCOffset < ${i}) {
|
|
xTexelR${x}C${T} = getX(batch, xR, xCOffset, d1);
|
|
} else {
|
|
xTexelR${x}C${T} = vec4(0.);
|
|
}
|
|
`),g+=`
|
|
xR${x}C${T+1} = vec4(
|
|
xTexelR${x}C${T}.zw, xTexelR${x}C${T+2}.xy);
|
|
`):g+=`
|
|
xCOffset = xC + ${C};
|
|
|
|
if(xR >= 0 && xR < ${s} &&
|
|
xCOffset >= 0 && xCOffset < ${i}) {
|
|
xTexelR${x}C${T+2} = getX(batch, xR, xCOffset, d1);
|
|
}
|
|
|
|
xR${x}C${T+1} = xTexelR${x}C${T+2};
|
|
`}}else T<m&&(g+=`
|
|
if(xR >= 0 && xR < ${s}) {
|
|
`,l%2==1?(g+=`
|
|
xCOffset = xC + 1 - ${u};
|
|
if(xCOffset >= 0 && xCOffset < ${i}) {
|
|
xTexelR${x}C${T} = getX(batch, xR, xCOffset, d1);
|
|
} else {
|
|
xTexelR${x}C${T} = vec4(0.);
|
|
}
|
|
|
|
if(xC + 1 >= 0 && xC + 1 < ${i}) {
|
|
xTexelR${x}C${T+2} = getX(batch, xR, xC + 1, d1);
|
|
} else {
|
|
xTexelR${x}C${T+2} = vec4(0.);
|
|
}
|
|
|
|
xR${x}C${T} = vec4(
|
|
xTexelR${x}C${T}.zw, xTexelR${x}C${T+2}.zw);
|
|
`,T+1<m&&(g+=`
|
|
vec4 final = vec4(0.);
|
|
xCOffset = xC + 1 + ${u};
|
|
if(xCOffset >= 0 && xCOffset < ${i}) {
|
|
final = getX(batch, xR, xCOffset, d1);
|
|
}
|
|
xR${x}C${T+1} = vec4(xTexelR${x}C${T+2}.xy, final.xy);
|
|
`)):(g+=`
|
|
if(xC >= 0 && xC < ${i}) {
|
|
xTexelR${x}C${T} = getX(batch, xR, xC, d1);
|
|
} else {
|
|
xTexelR${x}C${T} = vec4(0.);
|
|
}
|
|
|
|
xCOffset = xC + ${u};
|
|
if(xCOffset >= 0 && xCOffset < ${i}) {
|
|
xTexelR${x}C${T+2} = getX(batch, xR, xCOffset, d1);
|
|
} else {
|
|
xTexelR${x}C${T+2} = vec4(0.);
|
|
}
|
|
|
|
xR${x}C${T} = vec4(
|
|
xTexelR${x}C${T}.xy, xTexelR${x}C${T+2}.xy);
|
|
`,T+1<m&&(g+=`
|
|
xR${x}C${T+1} = vec4(
|
|
xTexelR${x}C${T}.zw, xTexelR${x}C${T+2}.zw);
|
|
`)),g+="}");T<m&&(g+=`
|
|
vec4 wTexelR${x}C${T} = getW(${x}, ${T}, d1, q);
|
|
wR${x}C${T} = vec4(wTexelR${x}C${T}.xz, wTexelR${x}C${T}.xz);
|
|
`,T+1<m&&(g+=`
|
|
vec4 wTexelR${x}C${T+1} = getW(${x}, ${T+1}, d1, q);
|
|
wR${x}C${T+1} =
|
|
vec4(wTexelR${x}C${T+1}.xz, wTexelR${x}C${T+1}.xz);`))}for(let x=0;x<h;x++)for(let N=0;N<m;N++)g+=`dotProd += xR${x}C${N} * wR${x}C${N};`;let y="",b="";n&&(a?y=`vec4 activation(vec4 a) {
|
|
vec4 b = getPreluActivationWeightsAtOutCoords();
|
|
${n}
|
|
}`:r?y=`vec4 activation(vec4 a) {
|
|
vec4 b = getLeakyreluAlphaAtOutCoords();
|
|
${n}
|
|
}`:y=`vec4 activation(vec4 x) {
|
|
${n}
|
|
}`,b="result = activation(result);");let v=t?"result += getBiasAtOutCoords();":"";t&&this.variableNames.push("bias"),a&&this.variableNames.push("preluActivationWeights"),r&&this.variableNames.push("leakyreluAlpha"),this.userCode=`
|
|
${y}
|
|
|
|
const ivec2 strides = ivec2(${c}, ${u});
|
|
const ivec2 pads = ivec2(${o}, ${l});
|
|
|
|
void main() {
|
|
|
|
ivec4 coords = getOutputCoords();
|
|
int batch = coords.x;
|
|
ivec2 xRCCorner = coords.yz * strides - pads;
|
|
int d2 = coords.w;
|
|
int d1 = d2;
|
|
int q = 0;
|
|
int xRCorner = xRCCorner.x;
|
|
int xCCorner = xRCCorner.y;
|
|
|
|
vec4 dotProd = vec4(0.);
|
|
|
|
${g}
|
|
|
|
vec4 result = dotProd;
|
|
${v}
|
|
${b}
|
|
setOutput(result);
|
|
}
|
|
`}};function YY(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,filter:s}=t,{strides:i,pad:o,dilations:l,dimRoundingMode:c}=a,u=l;u==null&&(u=[1,1]),k.assert(_.eitherStridesOrDilationsAreOne(i,u),()=>`Error in depthwiseConv2d: Either strides or dilations must be 1. Got strides ${i} and dilations '${u}'`);let p=_.computeConv2DInfo(r.shape,s.shape,i,u,o,c,!0),d;return ee().getBool("WEBGL_PACK_DEPTHWISECONV")&&p.strideWidth<=2&&p.outChannels/p.inChannels==1?d=new _S(p):d=new CS(p),n.runWebGLProgram(d,[r,s],"float32")}var ZY={kernelName:Ms,backendName:"webgl",kernelFunc:YY},JY=class{constructor(e){this.variableNames=["x","dy"],this.outputShape=e.filterShape;let t=e.strideHeight,n=e.strideWidth,a=e.padInfo.top,r=e.padInfo.left,s=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 * ${s} + 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} - ${a};
|
|
|
|
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 >= ${e.inWidth}) {
|
|
continue;
|
|
}
|
|
|
|
float dyValue = getDy(b, yR, yC, d2);
|
|
float xValue = getX(b, xR, xC, d1);
|
|
dotProd += (xValue * dyValue);
|
|
}
|
|
}
|
|
}
|
|
setOutput(dotProd);
|
|
}
|
|
`}},QY=class{constructor(e){this.variableNames=["dy","W"],this.outputShape=e.inShape;let t=e.filterHeight,n=e.filterWidth,a=e.strideHeight,r=e.strideWidth,s=t-1-e.padInfo.top,i=n-1-e.padInfo.left,o=e.outChannels/e.inChannels;this.userCode=`
|
|
const ivec2 pads = ivec2(${s}, ${i});
|
|
|
|
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) / ${a}.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 < ${o}; dm++) {
|
|
int d2 = d1 * ${o} + dm;
|
|
float xValue = getDy(batch, idyR, idyC, d2);
|
|
float wValue = getW(wRPerm, wCPerm, d1, dm);
|
|
dotProd += xValue * wValue;
|
|
}
|
|
}
|
|
}
|
|
setOutput(dotProd);
|
|
}
|
|
`}};function e7(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,dy:s}=t,{strides:i,dilations:o,pad:l,dimRoundingMode:c,filterShape:u}=a,p=_.computeConv2DInfo(r.shape,u,i,o,l,c,!0),d=new JY(p);return n.runWebGLProgram(d,[r,s],"float32")}var t7={kernelName:Cd,backendName:"webgl",kernelFunc:e7};function n7(e){let{inputs:t,backend:n,attrs:a}=e,{dy:r,filter:s}=t,{strides:i,dilations:o,pad:l,dimRoundingMode:c,inputShape:u}=a,p=_.computeConv2DInfo(u,s.shape,i,o,l,c,!0),d=new QY(p);return n.runWebGLProgram(d,[r,s],"float32")}var a7={kernelName:_d,backendName:"webgl",kernelFunc:n7},r7=class{constructor(e){this.variableNames=["X"],this.outputShape=[e,e],this.userCode=`
|
|
void main() {
|
|
ivec2 coords = getOutputCoords();
|
|
float val = coords[0] == coords[1] ? getX(coords[0]) : 0.0;
|
|
setOutput(val);
|
|
}
|
|
`}};function s7(e){let{inputs:t,backend:n}=e,{x:a}=t,r=[...a.shape,...a.shape],s=k.sizeFromShape(a.shape),i=be({inputs:{x:a},backend:n,attrs:{shape:[s]}}),o=new r7(s),l=n.runWebGLProgram(o,[i],i.dtype),c=be({inputs:{x:l},backend:n,attrs:{shape:r}});return n.disposeIntermediateTensorInfo(i),n.disposeIntermediateTensorInfo(l),c}var i7={kernelName:Ed,backendName:"webgl",kernelFunc:s7},o7=class{constructor(e){this.variableNames=["x","W"],this.outputShape=e.outShape;let{inHeight:t,inWidth:n,padInfo:a,strideHeight:r,strideWidth:s,filterHeight:i,filterWidth:o,dilationHeight:l,dilationWidth:c}=e,{top:u,left:p}=a;this.userCode=`
|
|
const ivec2 strides = ivec2(${r}, ${s});
|
|
const ivec2 pads = ivec2(${u}, ${p});
|
|
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 < ${i}; h++) {
|
|
int hIn = hBeg + h * ${l};
|
|
|
|
if (hIn >= 0 && hIn < ${t}) {
|
|
for (int w = 0; w < ${o}; w++) {
|
|
int wIn = wBeg + w * ${c};
|
|
|
|
if (wIn >= 0 && wIn < ${n}) {
|
|
float xVal = getX(batch, hIn, wIn, d1);
|
|
float wVal = getW(h, w, d1);
|
|
|
|
float val = xVal + wVal;
|
|
if (val > curVal) {
|
|
curVal = val;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
float result = curVal;
|
|
setOutput(result);
|
|
}
|
|
`}};function l7(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,filter:s}=t,{strides:i,pad:o,dilations:l}=a,c=_.computeDilation2DInfo(r.shape,s.shape,i,o,"NHWC",l),u,p=new o7(c);u=n.runWebGLProgram(p,[r,s],"float32");let d=be({inputs:{x:u},backend:n,attrs:{shape:c.outShape}});return n.disposeIntermediateTensorInfo(u),d}var u7={kernelName:hc,backendName:"webgl",kernelFunc:l7},c7="return (x >= 0.0) ? x : (exp(x) - 1.0);",p7=`
|
|
vec4 result;
|
|
|
|
result.r = (x.r >= 0.0) ? x.r : (exp(x.r) - 1.0);
|
|
result.g = (x.g >= 0.0) ? x.g : (exp(x.g) - 1.0);
|
|
result.b = (x.b >= 0.0) ? x.b : (exp(x.b) - 1.0);
|
|
result.a = (x.a >= 0.0) ? x.a : (exp(x.a) - 1.0);
|
|
|
|
return result;
|
|
`,d7=Ze({opSnippet:c7,packedOpSnippet:p7}),h7={kernelName:zo,backendName:"webgl",kernelFunc:d7},m7="return (b >= 1.0) ? a : a * (b + 1.0);",f7=`
|
|
vec4 bGTEZero = vec4(greaterThanEqual(b, vec4(0.)));
|
|
return (bGTEZero * a) + ((vec4(1.0) - bGTEZero) * (a * (b + vec4(1.0))));
|
|
`,g7=e=>{let{inputs:t,backend:n}=e,{dy:a,y:r}=t,s=ee().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new wp(f7,a.shape,r.shape):new hu(m7,a.shape,r.shape);return n.runWebGLProgram(s,[a,r],a.dtype)},y7={kernelName:$d,backendName:"webgl",kernelFunc:g7},b7=`
|
|
return vec4(equal(a, b));
|
|
`,x7="return float(a == b);",v7=nn({opSnippet:x7,packedOpSnippet:b7,dtype:"bool"}),w7={kernelName:Wo,backendName:"webgl",kernelFunc:v7},k7=`
|
|
// Error function is calculated approximately with elementary function.
|
|
// See "Handbook of Mathematical Functions with Formulas,
|
|
// Graphs, and Mathematical Tables", Abramowitz and Stegun.
|
|
float p = ${_.ERF_P};
|
|
float a1 = ${_.ERF_A1};
|
|
float a2 = ${_.ERF_A2};
|
|
float a3 = ${_.ERF_A3};
|
|
float a4 = ${_.ERF_A4};
|
|
float a5 = ${_.ERF_A5};
|
|
|
|
float sign = sign(x);
|
|
x = abs(x);
|
|
float t = 1.0 / (1.0 + p * x);
|
|
return sign * (1.0 - (((((a5*t + a4)*t) + a3)*t + a2)*t + a1)*t*exp(-x*x));
|
|
`,I7=Ze({opSnippet:k7}),N7={kernelName:Bo,backendName:"webgl",kernelFunc:I7},ES="return exp(x);",AS=Ze({opSnippet:ES,packedOpSnippet:ES,cpuKernelImpl:k8}),T7={kernelName:Os,backendName:"webgl",kernelFunc:AS};function Uv(e){let{inputs:t,attrs:n,backend:a}=e,{dim:r}=n,{input:s}=t,i=s.shape.length,o=s.shape.slice(),l=r;return r<0&&(k.assert(-(i+1)<=r,()=>`Axis must be in the interval [${-(i+1)}, ${i}]`),l=i+r+1),o.splice(l,0,1),be({inputs:{x:s},backend:a,attrs:{shape:o}})}var S7={kernelName:Vo,backendName:"webgl",kernelFunc:Uv},FS="return exp(x) - 1.0;",C7=Ze({opSnippet:FS,packedOpSnippet:FS,cpuKernelImpl:I8}),_7={kernelName:Uo,backendName:"webgl",kernelFunc:C7},$S=class{constructor(e,t,n){this.variableNames=["real","imag"];let a=t[1];this.outputShape=t;let r=n?`2.0 * ${Math.PI}`:`-2.0 * ${Math.PI}`,s=n?`${a}.0`:"1.0",i;if(e==="real")i="return real * expR - imag * expI;";else if(e==="imag")i="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) {
|
|
${i}
|
|
}
|
|
|
|
float mulMatDFT(int batch, int index) {
|
|
float indexRatio = float(index) / float(${a});
|
|
float exponentMultiplierTimesIndexRatio =
|
|
exponentMultiplier * indexRatio;
|
|
|
|
float result = 0.0;
|
|
|
|
for (int i = 0; i < ${a}; 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) / ${s};
|
|
}
|
|
|
|
return result;
|
|
}
|
|
|
|
void main() {
|
|
ivec2 coords = getOutputCoords();
|
|
setOutput(mulMatDFT(coords[0], coords[1]));
|
|
}
|
|
`}};function DS(e,t,n){let a=n.texData.get(e.dataId),r=k.sizeFromShape(e.shape),s=e.shape[e.shape.length-1],i=r/s,o=be({inputs:{x:e},backend:n,attrs:{shape:[i,s]}}),l=o.shape,c=new $S("real",l,t),u=new $S("imag",l,t),p=[{dataId:a.complexTensorInfos.real.dataId,dtype:a.complexTensorInfos.real.dtype,shape:l},{dataId:a.complexTensorInfos.imag.dataId,dtype:a.complexTensorInfos.imag.dtype,shape:l}],d=n.runWebGLProgram(c,p,"float32"),h=n.runWebGLProgram(u,p,"float32"),m=ps({inputs:{real:d,imag:h},backend:n});n.disposeIntermediateTensorInfo(d),n.disposeIntermediateTensorInfo(h);let f=be({inputs:{x:m},backend:n,attrs:{shape:e.shape}});return n.disposeIntermediateTensorInfo(o),n.disposeIntermediateTensorInfo(m),f}function E7(e){let{inputs:t,backend:n}=e,{input:a}=t;return DS(a,!1,n)}var A7={kernelName:Dd,backendName:"webgl",kernelFunc:E7},F7=class{constructor(e,t){this.outputShape=[],this.variableNames=["x"],this.outputShape=e,this.userCode=`
|
|
uniform float value;
|
|
void main() {
|
|
// Input can be obtained from uniform value.
|
|
setOutput(value);
|
|
}
|
|
`}getCustomSetupFunc(e){return(t,n)=>{this.valueLoc==null&&(this.valueLoc=t.getUniformLocationNoThrow(n,"value")),t.gl.uniform1f(this.valueLoc,e)}}};function Gv(e){let{backend:t,attrs:n}=e,{shape:a,value:r}=n,{dtype:s}=n;if(s=s||k.inferDtype(r),s==="string"){let i=k.getArrayFromDType(s,k.sizeFromShape(a));return i.fill(r),t.makeTensorInfo(a,s,i)}else{let i=new F7(a,r),o=i.getCustomSetupFunc(r);return t.runWebGLProgram(i,[],s,o)}}var $7={kernelName:mc,backendName:"webgl",kernelFunc:Gv},D7=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;
|
|
float outputValue;
|
|
if(coordX >= 0 && coordX < ${t}) {
|
|
outputValue = getImage(coords[0], coords[1], coordX, coords[3]);
|
|
} else {
|
|
outputValue = getImage(coords[0], coords[1], coords[2], coords[3]);
|
|
}
|
|
setOutput(outputValue);
|
|
}
|
|
`}},R7={kernelName:Go,backendName:"webgl",kernelFunc:({inputs:e,backend:t})=>{let{image:n}=e,a=t,r=new D7(n.shape);return a.runWebGLProgram(r,[n],n.dtype)}},RS="return floor(x);",M7=Ze({opSnippet:RS,packedOpSnippet:RS,cpuKernelImpl:N8}),P7={kernelName:Ls,backendName:"webgl",kernelFunc:M7},O7=`
|
|
float s = sign(a) * sign(b);
|
|
int ia = round(a);
|
|
int ib = round(b);
|
|
if (ib != 0) {
|
|
// Windows (D3D) wants guaranteed non-zero int division at compile-time.
|
|
return float(idiv(ia, ib, s));
|
|
} else {
|
|
return NAN;
|
|
}
|
|
`,L7=`
|
|
ivec4 ia = round(a);
|
|
ivec4 ib = round(b);
|
|
bvec4 cond = notEqual(ib, ivec4(0));
|
|
ivec4 result = ivec4(0);
|
|
vec4 s = sign(a) * sign(b);
|
|
|
|
// Windows (D3D) wants guaranteed non-zero int division at compile-time.
|
|
if (cond[0]) {
|
|
result[0] = idiv(ia[0], ib[0], s[0]);
|
|
}
|
|
if (cond[1]) {
|
|
result[1] = idiv(ia[1], ib[1], s[1]);
|
|
}
|
|
if (cond[2]) {
|
|
result[2] = idiv(ia[2], ib[2], s[2]);
|
|
}
|
|
if (cond[3]) {
|
|
result[3] = idiv(ia[3], ib[3], s[3]);
|
|
}
|
|
return vec4(result);
|
|
`,z7=nn({opSnippet:O7,packedOpSnippet:L7,dtype:"int32"}),B7={kernelName:zs,backendName:"webgl",kernelFunc:z7},W7=class{constructor(e){this.variableNames=["A"];let t=hn(),[n,a]=e;this.outputShape=e,this.userCode=`
|
|
void main() {
|
|
ivec3 coords = getOutputCoords();
|
|
int texR = coords[0];
|
|
int texC = coords[1];
|
|
int depth = coords[2];
|
|
vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${a}.0, ${n}.0);
|
|
|
|
vec4 values = ${t.texture2D}(A, uv);
|
|
float value;
|
|
if (depth == 0) {
|
|
value = values.r;
|
|
} else if (depth == 1) {
|
|
value = values.g;
|
|
} else if (depth == 2) {
|
|
value = values.b;
|
|
} else if (depth == 3) {
|
|
value = values.a;
|
|
}
|
|
|
|
setOutput(floor(value * 255.0 + 0.5));
|
|
}
|
|
`}},V7=class{constructor(e){this.variableNames=["A"],this.packedInputs=!1,this.packedOutput=!0;let t=hn(),[n,a]=e;this.outputShape=e,this.userCode=`
|
|
void main() {
|
|
ivec3 coords = getOutputCoords();
|
|
int texR = coords[0];
|
|
int texC = coords[1];
|
|
int depth = coords[2];
|
|
|
|
vec4 result = vec4(0.);
|
|
|
|
for(int row=0; row<=1; row++) {
|
|
for(int col=0; col<=1; col++) {
|
|
texC = coords[1] + row;
|
|
depth = coords[2] + col;
|
|
|
|
vec2 uv = (vec2(texC, texR) + halfCR) /
|
|
vec2(${a}.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);
|
|
}
|
|
}
|
|
|
|
${t.output} = result;
|
|
}
|
|
`}},G7={kernelName:qd,backendName:"webgl",kernelFunc:U7},fu;function U7(e){let{inputs:t,backend:n,attrs:a}=e,{pixels:r}=t,{numChannels:s}=a,i=typeof HTMLVideoElement!="undefined"&&r instanceof HTMLVideoElement,o=typeof HTMLImageElement!="undefined"&&r instanceof HTMLImageElement,l=typeof ImageBitmap!="undefined"&&r instanceof ImageBitmap,[c,u]=i?[r.videoWidth,r.videoHeight]:[r.width,r.height],p=[u,c],d=[u,c,s];(o||i||l)&&(fu==null&&(fu=document.createElement("canvas").getContext("2d")),fu.canvas.width=c,fu.canvas.height=u,fu.drawImage(r,0,0,c,u),r=fu.canvas);let h=n.makeTensorInfo(p,"int32");n.texData.get(h.dataId).usage=Qn.PIXELS,n.gpgpu.uploadPixelDataToTexture(n.getTexture(h.dataId),r);let m=ee().getBool("WEBGL_PACK")?new V7(d):new W7(d),f=n.runWebGLProgram(m,[h],"int32");return n.disposeData(h.dataId),f}function H7(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,filter:s,bias:i,preluActivationWeights:o}=t,{strides:l,pad:c,dataFormat:u,dilations:p,dimRoundingMode:d,activation:h,leakyreluAlpha:m}=a,f=_.convertConv2DDataFormat(u),g=_.computeConv2DInfo(r.shape,s.shape,l,p,c,d,!1,f),y,b=[];if(g.filterHeight===1&&g.filterWidth===1&&g.dilationHeight===1&&g.dilationWidth===1&&g.strideHeight===1&&g.strideWidth===1&&(g.padInfo.type==="SAME"||g.padInfo.type==="VALID"))y=kS({x:r,filter:s,convInfo:g,backend:n,bias:i,activation:h,preluActivationWeights:o,leakyreluAlpha:m});else if(ee().getBool("WEBGL_CONV_IM2COL")&&r.shape[0]===1)y=IS({x:r,filter:s,convInfo:g,backend:n,bias:i,activation:h,preluActivationWeights:o,leakyreluAlpha:m});else{let x=i!=null,N=o!=null,T=h==="leakyrelu",C=h?Wm(h,!1):null,$=new wS(g,x,C,N,T),F=[r,s];if(i&&F.push(i),o&&F.push(o),T){let O=n.makeTensorInfo([],"float32",k.createScalarValue(m,"float32"));F.push(O),b.push(O)}y=n.runWebGLProgram($,F,"float32")}let v=be({inputs:{x:y},backend:n,attrs:{shape:g.outShape}});return b.push(y),b.forEach(x=>n.disposeIntermediateTensorInfo(x)),v}var j7={kernelName:yi,backendName:"webgl",kernelFunc:H7};function q7(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,filter:s,bias:i,preluActivationWeights:o}=t,{strides:l,pad:c,dilations:u,dimRoundingMode:p,activation:d,leakyreluAlpha:h}=a,m=[],f=u;f==null&&(f=[1,1]),k.assert(_.eitherStridesOrDilationsAreOne(l,f),()=>`Error in depthwiseConv2d: Either strides or dilations must be 1. Got strides ${l} and dilations '${f}'`);let g=_.computeConv2DInfo(r.shape,s.shape,l,f,c,p,!0),y=ee().getBool("WEBGL_PACK_DEPTHWISECONV")&&g.strideWidth<=2&&g.outChannels/g.inChannels==1,b=d?Wm(d,y):null,v=[r,s],x=i!=null,N=o!=null,T=d==="leakyrelu";if(x&&v.push(i),N&&v.push(o),T){let F=n.makeTensorInfo([],"float32",k.createScalarValue(h,"float32"));v.push(F),m.push(F)}let C;y?C=new _S(g,x,b,N,T):C=new CS(g,x,b,N,T);let $=n.runWebGLProgram(C,v,"float32");return m.forEach(F=>n.disposeIntermediateTensorInfo(F)),$}var K7={kernelName:bi,backendName:"webgl",kernelFunc:q7},X7=class{constructor(e,t,n){this.sliceDim=e,this.strides=t,this.variableNames=["x","indices"],this.outputShape=n;let a=ft(t.length),r=ft(n.length),s=this.sliceDim>1?"strides[j]":"strides";this.userCode=`
|
|
${a} strides = ${a}(${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 * ${s};
|
|
}
|
|
setOutput(getX(flattenIndex, coords[1]));
|
|
}
|
|
`}};function Y7(e){let{inputs:t,backend:n}=e,{params:a,indices:r}=t,s=r.shape,i=s[s.length-1],[o,l,c,u]=_.prepareAndValidate(a,r),p=be({inputs:{x:r},backend:n,attrs:{shape:[l,i]}}),d=be({inputs:{x:a},backend:n,attrs:{shape:[k.sizeFromShape(a.shape)/c,c]}}),h=new X7(i,u,[l,c]),m=n.runWebGLProgram(h,[d,p],d.dtype),f=be({inputs:{x:m},backend:n,attrs:{shape:o}});return n.disposeIntermediateTensorInfo(p),n.disposeIntermediateTensorInfo(d),n.disposeIntermediateTensorInfo(m),f}var Z7={kernelName:jo,backendName:"webgl",kernelFunc:Y7},Q7=class{constructor(e,t){this.variableNames=["A","indices"],this.outputShape=t,this.rank=t.length;let n=ft(this.rank),a=J7(e,2);this.userCode=`
|
|
void main() {
|
|
${n} resRC = getOutputCoords();
|
|
setOutput(getA(${a}));
|
|
}
|
|
`}};function J7(e,t){let n=["resRC.x","resRC.y","resRC.z","resRC.w"],a=[];for(let r=0;r<e.length;r++)r===2?a.push("int(getIndices(resRC.x, resRC.z))"):a.push(`${n[r]}`);return a.join()}function eZ(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,indices:s}=t,{axis:i,batchDims:o}=a,l=k.parseAxisParam(i,r.shape)[0],c=_.segment_util.collectGatherOpShapeInfo(r,s,l,o),u=k.sizeFromShape(s.shape),p=[],d=be({inputs:{x:r},backend:n,attrs:{shape:[c.batchSize,c.outerSize,c.dimSize,c.sliceSize]}}),h=be({inputs:{x:s},backend:n,attrs:{shape:[c.batchSize,u/c.batchSize]}});p.push(d),p.push(h);let m=[c.batchSize,c.outerSize,u/c.batchSize,c.sliceSize];if(n.shouldExecuteOnCPU([r,s])||r.dtype==="string"){let b=n.bufferSync(h),v=n.bufferSync(d),x=T8(v,b,m);return p.forEach(N=>n.disposeIntermediateTensorInfo(N)),n.makeTensorInfo(c.outputShape,x.dtype,x.values)}let f=new Q7(d.shape,m),g=n.runWebGLProgram(f,[d,h],d.dtype);p.push(g);let y=be({inputs:{x:g},backend:n,attrs:{shape:c.outputShape}});return p.forEach(b=>n.disposeIntermediateTensorInfo(b)),y}var tZ={kernelName:Ho,backendName:"webgl",kernelFunc:eZ},nZ="return float(a > b);",aZ=`
|
|
return vec4(greaterThan(a, b));
|
|
`,rZ=nn({opSnippet:nZ,packedOpSnippet:aZ,cpuKernelImpl:S8,dtype:"bool"}),sZ={kernelName:qo,backendName:"webgl",kernelFunc:rZ},iZ="return float(a >= b);",oZ=`
|
|
return vec4(greaterThanEqual(a, b));
|
|
`,lZ=nn({opSnippet:iZ,packedOpSnippet:oZ,dtype:"bool"}),uZ={kernelName:Ws,backendName:"webgl",kernelFunc:lZ};function cZ(e){let{inputs:t,backend:n}=e,{input:a}=t;return DS(a,!0,n)}var pZ={kernelName:Rd,backendName:"webgl",kernelFunc:cZ},dZ="return float(!isnan(x) && !isinf(x));",hZ=Ze({opSnippet:dZ,dtype:"bool"}),mZ={kernelName:Xo,backendName:"webgl",kernelFunc:hZ},fZ="return float(isinf(x));",gZ=Ze({opSnippet:fZ,dtype:"bool"}),yZ={kernelName:Yo,backendName:"webgl",kernelFunc:gZ},bZ="return float(isnan(x));",xZ=Ze({opSnippet:bZ,dtype:"bool"}),vZ={kernelName:Zo,backendName:"webgl",kernelFunc:xZ},wZ="return float(a < b);",kZ=`
|
|
return vec4(lessThan(a, b));
|
|
`,IZ=nn({opSnippet:wZ,packedOpSnippet:kZ,cpuKernelImpl:C8,dtype:"bool"}),NZ={kernelName:Jo,backendName:"webgl",kernelFunc:IZ},TZ="return float(a <= b);",SZ=`
|
|
return vec4(lessThanEqual(a, b));
|
|
`,CZ=nn({opSnippet:TZ,packedOpSnippet:SZ,dtype:"bool"}),_Z={kernelName:Qo,backendName:"webgl",kernelFunc:CZ};function EZ(e){let{backend:t,attrs:n}=e,{start:a,stop:r,num:s}=n,i=_8(a,r,s);return t.makeTensorInfo([i.length],"float32",i)}var AZ={kernelName:Pd,backendName:"webgl",kernelFunc:EZ},FZ=`if (x < 0.0) return NAN;
|
|
return log(x);`,$Z=`
|
|
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;
|
|
`,DZ=Ze({opSnippet:FZ,packedOpSnippet:$Z,cpuKernelImpl:E8}),RZ={kernelName:Us,backendName:"webgl",kernelFunc:DZ},MZ="return log(1.0 + x);",PZ=Ze({opSnippet:MZ}),OZ={kernelName:el,backendName:"webgl",kernelFunc:PZ},LZ="return float(a >= 1.0 && b >= 1.0);",zZ=`
|
|
return vec4(
|
|
vec4(greaterThanEqual(a, vec4(1.0))) *
|
|
vec4(greaterThanEqual(b, vec4(1.0))));
|
|
`,BZ=nn({opSnippet:LZ,packedOpSnippet:zZ,dtype:"bool"}),WZ={kernelName:tl,backendName:"webgl",kernelFunc:BZ},VZ="return float(!(x >= 1.0));",UZ=Ze({opSnippet:VZ}),GZ={kernelName:fc,backendName:"webgl",kernelFunc:UZ},HZ="return float(a >= 1.0 || b >= 1.0);",jZ=`
|
|
return min(
|
|
vec4(greaterThanEqual(a, vec4(1.0))) +
|
|
vec4(greaterThanEqual(b, vec4(1.0))),
|
|
vec4(1.0));
|
|
`,qZ=nn({opSnippet:HZ,packedOpSnippet:jZ,dtype:"bool"}),KZ={kernelName:gc,backendName:"webgl",kernelFunc:qZ},XZ=class{constructor(e,t,n,a,r){this.variableNames=["x"],this.outputShape=[];let s=t,i=e[3]-1;this.outputShape=e;let o,l=`float(${n}) + float(${a}) * sum`;r===.5?o=`inversesqrt(${l})`:r===1?o=`1.0/(${l})`:o=`exp(log(${l}) * float(-${r}));`,this.userCode=`
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int b = coords[0];
|
|
int r = coords[1];
|
|
int c = coords[2];
|
|
int d = coords[3];
|
|
float x = getX(b, r, c, d);
|
|
float sum = 0.0;
|
|
for (int j = -${s}; j <= ${s}; j++) {
|
|
int idx = d + j;
|
|
if (idx >= 0 && idx <= ${i}) {
|
|
float z = getX(b, r, c, idx);
|
|
sum += z * z;
|
|
}
|
|
}
|
|
float val = x * ${o};
|
|
setOutput(val);
|
|
}
|
|
`}},YZ=class{constructor(e,t,n,a,r){this.variableNames=["x"],this.outputShape=[],this.packedInputs=!0,this.packedOutput=!0;let s=t,i=e[3]-1;this.outputShape=e;let o,l=`float(${n}) + float(${a}) * sum`;r===.5?o=`inversesqrt(${l})`:r===1?o=`1.0/(${l})`:o=`exp(log(${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 - ${s};
|
|
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 = - ${s}; j <= ${s}; j++) {
|
|
ivec2 idx = depth + j;
|
|
bvec2 aboveLowerBound = greaterThanEqual(idx, ivec2(0));
|
|
bvec2 belowUpperBound = lessThanEqual(idx, ivec2(${i}));
|
|
|
|
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 * ${o};
|
|
setOutput(result);
|
|
}
|
|
`}},ZZ=e=>{let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{depthRadius:s,bias:i,alpha:o,beta:l}=a,c=ee().getBool("WEBGL_PACK_NORMALIZATION")?new YZ(r.shape,s,i,o,l):new XZ(r.shape,s,i,o,l);return n.runWebGLProgram(c,[r],r.dtype)},JZ={kernelName:yc,backendName:"webgl",kernelFunc:ZZ},QZ=class{constructor(e,t,n,a,r){this.variableNames=["inputImage","outputImage","dy"],this.outputShape=[],this.outputShape=e,this.depth=e[3],this.depthRadius=t,this.bias=n,this.alpha=a,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 = ${this.depth};
|
|
|
|
float norm = 0.0;
|
|
for (int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k) {
|
|
if (k < depthBegin){
|
|
continue;
|
|
}
|
|
else if (k >= depthBegin && k < depthEnd) {
|
|
norm += getInputImage(b, r, c, k) * getInputImage(b, r, c, k);
|
|
}
|
|
else {
|
|
break;
|
|
}
|
|
}
|
|
|
|
norm = float(${a}) * norm + float(${n});
|
|
|
|
for(int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k){
|
|
if (k < depthBegin){
|
|
continue;
|
|
}
|
|
else if (k >= depthBegin && k < depthEnd){
|
|
float dyi = -2.0 * float(${a})
|
|
* 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);
|
|
}
|
|
`}},e9=e=>{let{inputs:t,backend:n,attrs:a}=e,{x:r,y:s,dy:i}=t,{depthRadius:o,bias:l,alpha:c,beta:u}=a,p=new QZ(r.shape,o,l,c,u);return n.runWebGLProgram(p,[r,s,i],r.dtype)},t9={kernelName:Od,backendName:"webgl",kernelFunc:e9};function n9(e,t,n,a){let r=k.sizeFromShape(t),s=k.sizeFromShape(e.shape)/r,i=be({inputs:{x:e},attrs:{shape:[s,r]},backend:a}),o=ji(i,e.dtype,"max",a),l=be({inputs:{x:o},attrs:{shape:n},backend:a});return a.disposeIntermediateTensorInfo(i),a.disposeIntermediateTensorInfo(o),l}function MS(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{reductionIndices:s,keepDims:i}=a,o=r.shape.length,l=k.parseAxisParam(s,r.shape),c=l,u=_.getAxesPermutation(c,o),p=u!=null,d=n.shouldExecuteOnCPU([r]),h=r;if(p){if(d){let b=n.texData.get(h.dataId).values,v=new Array(o);for(let T=0;T<v.length;T++)v[T]=r.shape[u[T]];let x=Ov(b,r.shape,r.dtype,u,v);h=n.makeTensorInfo(v,r.dtype);let N=n.texData.get(h.dataId);N.values=x}else h=Vm(r,u,n);c=_.getInnerMostAxes(c.length,o)}_.assertAxesAreInnerMostDims("max",c,o);let[m,f]=_.computeOutAndReduceShapes(h.shape,c),g=m;i&&(g=_.expandShapeToKeepDim(m,l));let y;if(d){let b=n.texData.get(h.dataId).values,v=A8(b,k.sizeFromShape(f),g,r.dtype);y=n.makeTensorInfo(g,r.dtype);let x=n.texData.get(y.dataId);x.values=v}else y=n9(h,f,g,n);return p&&n.disposeIntermediateTensorInfo(h),y}var a9={kernelName:Gs,backendName:"webgl",kernelFunc:MS},r9=JT+`
|
|
return max(a, b);
|
|
`,s9=`
|
|
vec4 result = vec4(max(a, b));
|
|
vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));
|
|
`+Bm+`
|
|
return result;
|
|
`,i9=nn({opSnippet:r9,packedOpSnippet:s9,cpuKernelImpl:F8}),o9={kernelName:Hs,backendName:"webgl",kernelFunc:i9};function l9(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t;xp(r,"maxPool");let{filterSize:s,strides:i,pad:o,dimRoundingMode:l}=a,c=1;k.assert(_.eitherStridesOrDilationsAreOne(i,c),()=>`Error in maxPool: Either strides or dilations must be 1. Got strides ${i} and dilations '${c}'`);let u=_.computePool2DInfo(r.shape,s,i,c,o,l);if(u.filterWidth===1&&u.filterHeight===1&&k.arraysEqual(u.inShape,u.outShape))return Bn({inputs:{x:r},backend:n});let p=new kp(u,"max",!1);return n.runWebGLProgram(p,[r],r.dtype)}var u9={kernelName:js,backendName:"webgl",kernelFunc:l9};function c9(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{filterSize:s,strides:i,pad:o,dataFormat:l,dimRoundingMode:c}=a,u=[1,1,1],p=_.computePool3DInfo(r.shape,s,i,u,o,c,l),d=new Bv(p,"max",!1);return n.runWebGLProgram(d,[r],r.dtype)}var p9={kernelName:bc,backendName:"webgl",kernelFunc:c9},d9=class{constructor(e){this.variableNames=["dy","maxPos"],this.outputShape=e.inShape;let t=e.strideHeight,n=e.strideWidth,a=e.dilationHeight,r=e.effectiveFilterHeight,s=e.effectiveFilterWidth,i=r-1-e.padInfo.top,o=s-1-e.padInfo.left,l=r*s-1;this.userCode=`
|
|
const ivec2 pads = ivec2(${i}, ${o});
|
|
|
|
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 += ${a}) {
|
|
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 < ${s}; 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 * ${s} + wC;
|
|
float mask = float(maxPosValue == curPosValue ? 1.0 : 0.0);
|
|
|
|
dotProd += dyValue * mask;
|
|
}
|
|
}
|
|
setOutput(dotProd);
|
|
}
|
|
`}},h9=class{constructor(e){this.variableNames=["dy","maxPos"],this.outputShape=e.inShape;let t=e.strideDepth,n=e.strideHeight,a=e.strideWidth,r=e.dilationDepth,s=e.dilationHeight,i=e.dilationWidth,o=e.effectiveFilterDepth,l=e.effectiveFilterHeight,c=e.effectiveFilterWidth,u=o-1-e.padInfo.front,p=l-1-e.padInfo.top,d=c-1-e.padInfo.left,h=o*l*c-1;this.userCode=`
|
|
const ivec3 pads = ivec3(${u}, ${p}, ${d});
|
|
|
|
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 < ${o};
|
|
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 += ${s}) {
|
|
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 += ${i}) {
|
|
float dyC = float(dyCCorner + wC) / ${a}.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 m9(e){let{inputs:t,backend:n,attrs:a}=e,{dy:r,input:s}=t,i=s,{filterSize:o,strides:l,pad:c,dimRoundingMode:u}=a,p=[1,1,1],d=_.computePool3DInfo(i.shape,o,l,p,c,u),h=new Bv(d,"max",!0),m=n.runWebGLProgram(h,[i],i.dtype),f=new h9(d),g=n.runWebGLProgram(f,[r,m],i.dtype);return n.disposeIntermediateTensorInfo(m),g}var f9={kernelName:zd,backendName:"webgl",kernelFunc:m9};function g9(e){let{inputs:t,backend:n,attrs:a}=e,{dy:r,input:s,output:i}=t,o=s;xp([s,i],"maxPoolGrad");let{filterSize:l,strides:c,pad:u,dimRoundingMode:p}=a,d=_.computePool2DInfo(o.shape,l,c,1,u,p),h=!0,m=new kp(d,"max",h),f=n.runWebGLProgram(m,[o],o.dtype),g=new d9(d),y=n.runWebGLProgram(g,[r,f],o.dtype);return n.disposeIntermediateTensorInfo(f),y}var y9={kernelName:Ld,backendName:"webgl",kernelFunc:g9};function b9(e,t,n,a){let r=new kp(n,"max",!1),s=a.runWebGLProgram(r,[e],"float32");r=new kp(n,"max",!0,!0,t);let i=a.runWebGLProgram(r,[e],"float32");return[s,i]}var x9={kernelName:Bd,backendName:"webgl",kernelFunc:({inputs:e,attrs:t,backend:n})=>{let{x:a}=e,{filterSize:r,strides:s,pad:i,includeBatchInIndex:o}=t,l=n;k.assert(a.shape.length===4,()=>`Error in maxPool: input must be rank 4 but got rank ${a.shape.length}.`);let c=[1,1];k.assert(_.eitherStridesOrDilationsAreOne(s,c),()=>`Error in maxPool: Either strides or dilations must be 1. Got strides ${s} and dilations '${c}'`);let u=_.computePool2DInfo(a.shape,r,s,c,i),[p,d]=b9(a,o,u,l);return[p,d]}};function v9(e,t,n,a){let r=k.sizeFromShape(t),s=k.sizeFromShape(e.shape)/r,i=be({inputs:{x:e},attrs:{shape:[s,r]},backend:a}),o=ji(i,"float32","mean",a),l=be({inputs:{x:o},attrs:{shape:n},backend:a});return a.disposeIntermediateTensorInfo(i),a.disposeIntermediateTensorInfo(o),l}var w9={kernelName:qs,backendName:"webgl",kernelFunc:({inputs:e,attrs:t,backend:n})=>{let{x:a}=e,{keepDims:r,axis:s}=t,i=n,o=a.shape.length,l=k.parseAxisParam(s,a.shape),c=l,u=_.getAxesPermutation(c,o),p=u!=null,d=i.shouldExecuteOnCPU([a]),h=[],m=a;if(p){if(d){let v=i.texData.get(m.dataId).values,x=new Array(o);for(let C=0;C<x.length;C++)x[C]=a.shape[u[C]];let N=Ov(v,a.shape,a.dtype,u,x);m=i.makeTensorInfo(x,a.dtype);let T=i.texData.get(m.dataId);T.values=N}else m=Vm(a,u,i);h.push(m),c=_.getInnerMostAxes(c.length,o)}_.assertAxesAreInnerMostDims("sum",c,o);let[f,g]=_.computeOutAndReduceShapes(m.shape,c),y=f;r&&(y=_.expandShapeToKeepDim(f,l));let b=v9(m,g,y,i);for(let v of h)i.disposeIntermediateTensorInfo(v);return b}};function k9(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{axis:s,keepDims:i}=a,o=r.shape.length,l=k.parseAxisParam(s,r.shape),c=l,u=_.getAxesPermutation(c,o),p=r;u!=null&&(p=Tn({inputs:{x:r},backend:n,attrs:{perm:u}}),c=_.getInnerMostAxes(c.length,r.shape.length)),_.assertAxesAreInnerMostDims("min",c,o);let[d,h]=_.computeOutAndReduceShapes(p.shape,c),m=k.sizeFromShape(h),f=be({inputs:{x:p},backend:n,attrs:{shape:[-1,m]}}),g=ji(f,f.dtype,"min",n),y;if(i){let b=_.expandShapeToKeepDim(d,l);y=be({inputs:{x:g},backend:n,attrs:{shape:b}})}else y=be({inputs:{x:g},backend:n,attrs:{shape:d}});return n.disposeIntermediateTensorInfo(f),n.disposeIntermediateTensorInfo(g),u!=null&&n.disposeIntermediateTensorInfo(p),y}var I9={kernelName:Ks,backendName:"webgl",kernelFunc:k9},N9=JT+`
|
|
return min(a, b);
|
|
`,T9=`
|
|
vec4 result = vec4(min(a, b));
|
|
vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));
|
|
`+Bm+`
|
|
return result;
|
|
`,S9=nn({opSnippet:N9,packedOpSnippet:T9,cpuKernelImpl:$8}),C9={kernelName:Xs,backendName:"webgl",kernelFunc:S9},_9=class{constructor(e,t,n){this.variableNames=["x"],this.outputShape=t.map((c,u)=>c[0]+e[u]+c[1]);let a=e.length,r=ft(a),s=t.map(c=>c[0]).join(","),i=t.map((c,u)=>c[0]+e[u]).join(","),o=["coords[0]","coords[1]","coords[2]","coords[3]"].slice(0,a),l=n==="reflect"?0:1;if(a===1){this.userCode=`
|
|
int start = ${s};
|
|
int end = ${i};
|
|
|
|
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 = ${r}(${s});
|
|
${r} end = ${r}(${i});
|
|
|
|
void main() {
|
|
${r} outC = getOutputCoords();
|
|
for (int i = 0; i < ${a}; 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};
|
|
}
|
|
}
|
|
${r} coords = outC - start;
|
|
setOutput(getX(${o}));
|
|
}
|
|
`}},E9=class{constructor(e,t,n){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t.map((h,m)=>h[0]+e[m]+h[1]);let a=e.length,r=ft(a),s=t.map(h=>h[0]).join(","),i=t.map((h,m)=>h[0]+e[m]).join(","),o=mn("rc",a),l=mn("source",a),c=`${o[a-1]} < ${this.outputShape[a-1]}`,u=a===1?"source":`vec2(${l.slice(-2).join()})`,p=n==="reflect"?0:1,d="";if(a===1){let h=`
|
|
${r} source = rc;
|
|
if (source < start) {
|
|
source = start * 2 - source - ${p};
|
|
} else if (source >= end) {
|
|
source = (end - 1) * 2 - source + ${p};
|
|
}
|
|
source -= start;
|
|
`;d=`
|
|
${r} rc = outputLoc;
|
|
${h}
|
|
result[0] = getChannel(getX(${l.join()}), ${u});
|
|
${o[a-1]} += 1;
|
|
if(${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 - ${p}) +
|
|
gte * ((end - 1) * 2 - source + ${p});
|
|
source -= start;
|
|
`;d=`
|
|
${r} rc = outputLoc;
|
|
${h}
|
|
result[0] = getChannel(getX(${l.join()}), ${u});
|
|
${o[a-1]} += 1;
|
|
if(${c}) {
|
|
${h}
|
|
result[1] = getChannel(getX(${l.join()}), ${u});
|
|
}
|
|
rc = outputLoc;
|
|
${o[a-2]} += 1;
|
|
if(${o[a-2]} < ${this.outputShape[a-2]}) {
|
|
${h}
|
|
result[2] = getChannel(getX(${l.join()}), ${u});
|
|
${o[a-1]} += 1;
|
|
if(${c}) {
|
|
${h}
|
|
result[3] = getChannel(getX(${l.join()}), ${u});
|
|
}
|
|
}
|
|
`}this.userCode=`
|
|
const ${r} start = ${r}(${s});
|
|
const ${r} end = ${r}(${i});
|
|
|
|
void main() {
|
|
${r} outputLoc = getOutputCoords();
|
|
vec4 result = vec4(0.);
|
|
${d}
|
|
setOutput(result);
|
|
}
|
|
`}},A9=({inputs:e,backend:t,attrs:n})=>{let{x:a}=e,{paddings:r,mode:s}=n,i=ee().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new E9(a.shape,r,s):new _9(a.shape,r,s);return t.runWebGLProgram(i,[a],a.dtype)},F9={kernelName:xc,backendName:"webgl",kernelFunc:A9},$9=`if (b == 0.0) return NAN;
|
|
return mod(a, b);`,D9=`
|
|
vec4 result = mod(a, b);
|
|
vec4 isNaN = vec4(equal(b, vec4(0.0)));
|
|
`+Bm+`
|
|
return result;
|
|
`,R9=nn({opSnippet:$9,packedOpSnippet:D9}),M9={kernelName:nl,backendName:"webgl",kernelFunc:R9},P9=class{constructor(e,t,n){this.variableNames=["probs"],this.outputShape=[e,n],this.userCode=`
|
|
uniform float seed;
|
|
|
|
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}));
|
|
}
|
|
`}getCustomSetupFunc(e){return(t,n)=>{this.seedLoc==null&&(this.seedLoc=t.getUniformLocation(n,"seed")),t.gl.uniform1f(this.seedLoc,e)}}},O9=`
|
|
if (a == b) {
|
|
return 1.0;
|
|
};
|
|
return a / b;`,L9=`
|
|
// 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;
|
|
`,PS=nn({opSnippet:O9,packedOpSnippet:L9,checkOutOfBounds:!0}),z9={kernelName:Ps,backendName:"webgl",kernelFunc:PS},OS="return a - b;",LS=nn({opSnippet:OS,packedOpSnippet:OS,supportsComplex:!0,cpuKernelImpl:B8}),B9={kernelName:hi,backendName:"webgl",kernelFunc:LS};function zS(e){let{inputs:t,backend:n,attrs:a}=e,{logits:r}=t,{dim:s}=a,i=k.parseAxisParam([s],r.shape),o=MS({inputs:{x:r},backend:n,attrs:{reductionIndices:i,keepDims:!1}}),l=_.expandShapeToKeepDim(o.shape,i),c=be({inputs:{x:o},backend:n,attrs:{shape:l}}),u=LS({inputs:{a:r,b:c},backend:n}),p=AS({inputs:{x:u},backend:n}),d=zv({inputs:{x:p},backend:n,attrs:{axis:i,keepDims:!1}}),h=be({inputs:{x:d},backend:n,attrs:{shape:l}}),m=PS({inputs:{a:p,b:h},backend:n});return n.disposeIntermediateTensorInfo(o),n.disposeIntermediateTensorInfo(c),n.disposeIntermediateTensorInfo(u),n.disposeIntermediateTensorInfo(p),n.disposeIntermediateTensorInfo(d),n.disposeIntermediateTensorInfo(h),m}var W9={kernelName:pi,backendName:"webgl",kernelFunc:zS};function V9(e){let{inputs:t,backend:n,attrs:a}=e,{logits:r}=t,{numSamples:s,seed:i,normalized:o}=a,l=o?r:zS({inputs:{logits:r},backend:n,attrs:{dim:r.shape.length-1}}),c=l.shape[0],u=l.shape[1],p=new P9(c,u,s),d=p.getCustomSetupFunc(i),h=n.runWebGLProgram(p,[l],"int32",d);return o||n.disposeIntermediateTensorInfo(l),h}var U9={kernelName:Wd,backendName:"webgl",kernelFunc:V9},BS="return -x;";function G9(e){let{inputs:t,backend:n}=e,{x:a}=t;if(n.shouldExecuteOnCPU([a])){let s=n.texData.get(a.dataId),[i,o]=R8(s.values,a.shape,a.dtype);return n.makeTensorInfo(o,a.dtype,i)}let r;return ee().getBool("WEBGL_PACK_UNARY_OPERATIONS")?r=new du(a.shape,BS):r=new cs(a.shape,BS),n.runWebGLProgram(r,[a],a.dtype)}var H9={kernelName:al,backendName:"webgl",kernelFunc:G9},j9=Za.nonMaxSuppressionV3Impl;function q9(e){_.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");let{inputs:t,backend:n,attrs:a}=e,{boxes:r,scores:s}=t,{maxOutputSize:i,iouThreshold:o,scoreThreshold:l}=a,c=n.readSync(r.dataId),u=n.readSync(s.dataId),{selectedIndices:p}=j9(c,u,i,o,l);return n.makeTensorInfo([p.length],"int32",new Int32Array(p))}var K9={kernelName:sl,backendName:"webgl",kernelFunc:q9},X9=Za.nonMaxSuppressionV4Impl;function Y9(e){_.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");let{inputs:t,backend:n,attrs:a}=e,{boxes:r,scores:s}=t,{maxOutputSize:i,iouThreshold:o,scoreThreshold:l,padToMaxOutputSize:c}=a,u=n.readSync(r.dataId),p=n.readSync(s.dataId),{selectedIndices:d,validOutputs:h}=X9(u,p,i,o,l,c);return[n.makeTensorInfo([d.length],"int32",new Int32Array(d)),n.makeTensorInfo([],"int32",new Int32Array([h]))]}var Z9={kernelName:il,backendName:"webgl",kernelFunc:Y9},J9=Za.nonMaxSuppressionV5Impl;function Q9(e){_.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");let{inputs:t,backend:n,attrs:a}=e,{boxes:r,scores:s}=t,{maxOutputSize:i,iouThreshold:o,scoreThreshold:l,softNmsSigma:c}=a,u=n.readSync(r.dataId),p=n.readSync(s.dataId),d=i,h=o,m=l,f=c,{selectedIndices:g,selectedScores:y}=J9(u,p,d,h,m,f);return[n.makeTensorInfo([g.length],"int32",new Int32Array(g)),n.makeTensorInfo([y.length],"float32",new Float32Array(y))]}var eJ={kernelName:ol,backendName:"webgl",kernelFunc:Q9},tJ=class{constructor(e,t,n,a){this.variableNames=["indices"],this.outputShape=[e,t],this.userCode=`
|
|
void main() {
|
|
ivec2 coords = getOutputCoords();
|
|
int index = round(getIndices(coords.x));
|
|
setOutput(mix(float(${a}), float(${n}),
|
|
float(index == coords.y)));
|
|
}
|
|
`}},nJ=e=>{let{inputs:t,backend:n,attrs:a}=e,{indices:r}=t,{depth:s,onValue:i,offValue:o}=a,l=k.sizeFromShape(r.shape),c=new tJ(l,s,i,o),u=be({inputs:{x:r},backend:n,attrs:{shape:[l]}}),p=n.runWebGLProgram(c,[u],r.dtype);n.disposeIntermediateTensorInfo(u);let d=[...r.shape,s],h=be({inputs:{x:p},backend:n,attrs:{shape:d}});return n.disposeIntermediateTensorInfo(p),h},aJ={kernelName:Zs,backendName:"webgl",kernelFunc:nJ};function qm(e){let{inputs:t,backend:n}=e,{x:a}=t;if(a.dtype==="complex64"){let r=Np({inputs:{input:a},backend:n}),s=qm({inputs:{x:r},backend:n}),i=jm({inputs:{input:a},backend:n}),o=qm({inputs:{x:i},backend:n}),l=ps({inputs:{real:s,imag:o},backend:n});return n.disposeIntermediateTensorInfo(r),n.disposeIntermediateTensorInfo(s),n.disposeIntermediateTensorInfo(i),n.disposeIntermediateTensorInfo(o),l}else return Gv({attrs:{shape:a.shape,dtype:a.dtype,value:a.dtype==="string"?"":0},backend:n})}var rJ={kernelName:Tl,backendName:"webgl",kernelFunc:qm};function WS(e){let{inputs:t,backend:n}=e,{x:a}=t;if(a.dtype==="string")throw new Error("onesLike is not supported under string dtype");if(a.dtype==="complex64"){let r=Np({inputs:{input:a},backend:n}),s=WS({inputs:{x:r},backend:n}),i=jm({inputs:{input:a},backend:n}),o=qm({inputs:{x:i},backend:n}),l=ps({inputs:{real:s,imag:o},backend:n});return n.disposeIntermediateTensorInfo(r),n.disposeIntermediateTensorInfo(s),n.disposeIntermediateTensorInfo(i),n.disposeIntermediateTensorInfo(o),l}else return Gv({attrs:{shape:a.shape,dtype:a.dtype,value:1},backend:n})}var sJ={kernelName:ll,backendName:"webgl",kernelFunc:WS};function iJ(e){let{inputs:t,backend:n,attrs:a}=e,{axis:r}=a;if(t.length===1)return Uv({inputs:{input:t[0]},backend:n,attrs:{dim:r}});let s=t[0].shape,i=t[0].dtype;t.forEach(u=>{k.assertShapesMatch(s,u.shape,"All tensors passed to stack must have matching shapes"),k.assert(i===u.dtype,()=>"All tensors passed to stack must have matching dtypes")});let o=[],l=t.map(u=>{let p=Uv({inputs:{input:u},backend:n,attrs:{dim:r}});return o.push(p),p}),c=vS({inputs:l,backend:n,attrs:{axis:r}});return o.forEach(u=>n.disposeIntermediateTensorInfo(u)),c}var oJ={kernelName:ul,backendName:"webgl",kernelFunc:iJ},lJ=class{constructor(e,t,n){this.variableNames=["x"],this.outputShape=t.map((l,c)=>l[0]+e[c]+l[1]);let a=e.length,r=ft(a),s=t.map(l=>l[0]).join(","),i=t.map((l,c)=>l[0]+e[c]).join(","),o=["coords[0]","coords[1]","coords[2]","coords[3]"].slice(0,a);if(a===1){this.userCode=`
|
|
int start = ${s};
|
|
int end = ${i};
|
|
|
|
void main() {
|
|
int outC = getOutputCoords();
|
|
if (outC < start || outC >= end) {
|
|
setOutput(float(${n}));
|
|
} else {
|
|
setOutput(getX(outC - start));
|
|
}
|
|
}
|
|
`;return}this.userCode=`
|
|
${r} start = ${r}(${s});
|
|
${r} end = ${r}(${i});
|
|
|
|
void main() {
|
|
${r} outC = getOutputCoords();
|
|
if (any(lessThan(outC, start)) || any(greaterThanEqual(outC, end))) {
|
|
setOutput(float(${n}));
|
|
} else {
|
|
${r} coords = outC - start;
|
|
setOutput(getX(${o}));
|
|
}
|
|
}
|
|
`}},uJ=class{constructor(e,t,n){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t.map((m,f)=>m[0]+e[f]+m[1]);let a=e.length,r=ft(a),s=t.map(m=>m[0]).join(","),i=t.map((m,f)=>m[0]+e[f]).join(","),o=mn("rc",a),l=mn("source",a),c=`${o[a-1]} < ${this.outputShape[a-1]}`,u=a===1?"source":`vec2(${l.slice(-2).join()})`,p=[`${r} rc = outputLoc;`,`${o[a-1]} += 1;
|
|
if(${c}) {
|
|
`,a===1?"":`}
|
|
rc = outputLoc;
|
|
${o[a-2]} += 1;
|
|
if(${o[a-2]} < ${this.outputShape[a-2]}) {`,a===1?"":` ${o[a-1]} += 1;
|
|
if(${c}) {`],d=a===1?"rc < start || rc >= end":"any(lessThan(rc, start)) || any(greaterThanEqual(rc, end))",h="";for(let m=0,f=a===1?2:4;m<f;m++)h+=`
|
|
${p[m]}
|
|
if (${d}) {
|
|
result[${m}] = float(${n});
|
|
} else {
|
|
${r} source = rc - start;
|
|
result[${m}] = getChannel(getX(${l.join()}), ${u});
|
|
}
|
|
`;h+=a===1?"} ":"}}",this.userCode=`
|
|
const ${r} start = ${r}(${s});
|
|
const ${r} end = ${r}(${i});
|
|
|
|
void main() {
|
|
${r} outputLoc = getOutputCoords();
|
|
vec4 result = vec4(0.);
|
|
${h}
|
|
setOutput(result);
|
|
}
|
|
`}},VS=e=>{let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{paddings:s,constantValue:i}=a,o=ee().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new uJ(r.shape,s,i):new lJ(r.shape,s,i);return n.runWebGLProgram(o,[r],r.dtype)},cJ={kernelName:Js,backendName:"webgl",kernelFunc:VS},pJ=`
|
|
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);
|
|
`,dJ=`
|
|
// 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));
|
|
`+Bm+`
|
|
return result;
|
|
`,hJ=nn({opSnippet:pJ,packedOpSnippet:dJ}),mJ={kernelName:Qs,backendName:"webgl",kernelFunc:hJ};function fJ(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{axis:s,keepDims:i}=a,o=r.shape.length,l=[],c=k.parseAxisParam(s,r.shape),u=c,p=_.getAxesPermutation(u,o),d=r;p!=null&&(d=Tn({inputs:{x:r},backend:n,attrs:{perm:p}}),u=_.getInnerMostAxes(u.length,o),l.push(d)),_.assertAxesAreInnerMostDims("prod",u,o);let h;if(n.shouldExecuteOnCPU([d])){let m=n.texData.get(d.dataId).values,{outVals:f,outShape:g,outDtype:y}=M8(d.shape,d.dtype,m,u);h=n.makeTensorInfo(g,y,f)}else{let[m,f]=_.computeOutAndReduceShapes(d.shape,u),g=k.sizeFromShape(f),y=be({inputs:{x:d},backend:n,attrs:{shape:[-1,g]}}),b=Qd(r.dtype),v=ji(y,b,"prod",n);h=be({inputs:{x:v},backend:n,attrs:{shape:m}}),l.push(y),l.push(v)}if(i){l.push(h);let m=_.expandShapeToKeepDim(h.shape,c);h=be({inputs:{x:h},backend:n,attrs:{shape:m}})}return l.forEach(m=>n.disposeIntermediateTensorInfo(m)),h}var gJ={kernelName:cl,backendName:"webgl",kernelFunc:fJ},US=e=>{let{backend:t,attrs:n}=e,{start:a,stop:r,step:s,dtype:i}=n,o=P8(a,r,s,i);return t.makeTensorInfo([o.length],i,o)},yJ={kernelName:vc,backendName:"webgl",kernelFunc:US},bJ="return 1.0 / x;",xJ=Ze({opSnippet:bJ}),vJ={kernelName:pl,backendName:"webgl",kernelFunc:xJ},wJ=Ra+`
|
|
return (x < 0.0) ? 0.0 : x;
|
|
`,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;
|
|
`,IJ=Ze({opSnippet:wJ,packedOpSnippet:kJ}),NJ={kernelName:ti,backendName:"webgl",kernelFunc:IJ},TJ=Ra+`
|
|
return (x < 0.0) ? 0.0 : min(6.0, x);
|
|
`,SJ=`
|
|
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;
|
|
`,CJ=Ze({opSnippet:TJ,packedOpSnippet:SJ}),_J={kernelName:ai,backendName:"webgl",kernelFunc:CJ},EJ=class{constructor(e,t,n,a,r){this.variableNames=["A"],this.outputShape=[];let[s,i,o,l]=e;this.outputShape=[s,t,n,l];let c=[a&&t>1?i-1:i,a&&n>1?o-1:o],u=[a&&t>1?t-1:t,a&&n>1?n-1:n],p;r?p="(vec2(yRC) + vec2(0.5)) * effectiveInputOverOutputRatioRC - vec2(0.5)":p="vec2(yRC) * effectiveInputOverOutputRatioRC",this.userCode=`
|
|
const vec2 effectiveInputOverOutputRatioRC = vec2(
|
|
${c[0]/u[0]},
|
|
${c[1]/u[1]});
|
|
const vec2 inputShapeRC = vec2(${i}.0, ${o}.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 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);
|
|
}
|
|
`}},AJ=class{constructor(e,t,n,a,r){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=[];let[s,i,o,l]=e;this.outputShape=[s,t,n,l];let c=[a&&t>1?i-1:i,a&&n>1?o-1:o],u=[a&&t>1?t-1:t,a&&n>1?n-1:n],p;r?p="(vec3(yRC) + vec3(0.5)) * effectiveInputOverOutputRatioRC - vec3(0.5)":p="vec3(yRC) * effectiveInputOverOutputRatioRC",this.userCode=`
|
|
const vec3 effectiveInputOverOutputRatioRC = vec3(
|
|
${c[0]/u[0]},
|
|
${c[1]/u[1]},
|
|
${c[1]/u[1]});
|
|
const vec3 inputShapeRC = vec3(${i}.0, ${o}.0,
|
|
${o}.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 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 FJ(e){let{inputs:t,backend:n,attrs:a}=e,{images:r}=t,{alignCorners:s,halfPixelCenters:i,size:o}=a,[l,c]=o,u=ee().getBool("WEBGL_PACK_IMAGE_OPERATIONS")?new AJ(r.shape,l,c,s,i):new EJ(r.shape,l,c,s,i);return n.runWebGLProgram(u,[r],"float32")}var $J={kernelName:ni,backendName:"webgl",kernelFunc:FJ},DJ=class{constructor(e,t,n){this.variableNames=["dy"],this.outputShape=[],this.outputShape=t;let[,a,r]=t,[,s,i]=e,o=[n&&s>1?a-1:a,n&&i>1?r-1:r],l=[n&&s>1?s-1:s,n&&i>1?i-1:i],c=o[0]/l[0],u=o[1]/l[1],p=1/c,d=1/u,h=Math.ceil(p)*2+2,m=Math.ceil(d)*2+2;this.userCode=`
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int b = coords[0];
|
|
int d = coords[3];
|
|
int r = coords[1];
|
|
int c = coords[2];
|
|
|
|
float accumulator = 0.0;
|
|
|
|
const float heightScale = float(${c});
|
|
const float widthScale = float(${u});
|
|
|
|
const float invHeightScale = float(${p});
|
|
const float invWidthScale = float(${d});
|
|
|
|
const int winHeight = int(${h});
|
|
const int winWidth = int(${m});
|
|
|
|
// 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 >= ${s}) {
|
|
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 >= ${i}) {
|
|
continue;
|
|
}
|
|
|
|
float dxR = float(dyR) * heightScale;
|
|
int topDxRIndex = int(floor(dxR));
|
|
int bottomDxRIndex = int(min(ceil(dxR), ${a-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 RJ(e){let{inputs:t,backend:n,attrs:a}=e,{images:r,dy:s}=t,{alignCorners:i}=a,o=new DJ(s.shape,r.shape,i);return n.runWebGLProgram(o,[s],s.dtype)}var MJ={kernelName:Gd,backendName:"webgl",kernelFunc:RJ},PJ=class{constructor(e,t,n,a,r){this.variableNames=["A"],this.outputShape=[];let[s,i,o,l]=e;this.outputShape=[s,t,n,l];let c=[a&&t>1?i-1:i,a&&n>1?o-1:o],u=[a&&t>1?t-1:t,a&&n>1?n-1:n],p=a?"0.5":"0.0",d;r?d="max((vec2(yRC) + vec2(0.5)) * effectiveInputOverOutputRatioRC, vec2(0.0))":d="vec2(yRC) * effectiveInputOverOutputRatioRC",this.userCode=`
|
|
const vec2 effectiveInputOverOutputRatioRC = vec2(
|
|
${c[0]/u[0]},
|
|
${c[1]/u[1]});
|
|
const vec2 inputShapeRC = vec2(${i}.0, ${o}.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 coordinators of nearest neighbor point.
|
|
ivec2 sourceNearestRC = ivec2(
|
|
min(inputShapeRC - 1.0, floor(sourceFracIndexRC + ${p})));
|
|
float newValue = getA(b, sourceNearestRC.x, sourceNearestRC.y, d);
|
|
|
|
setOutput(newValue);
|
|
}
|
|
`}};function OJ(e){let{inputs:t,backend:n,attrs:a}=e,{images:r}=t,{alignCorners:s,halfPixelCenters:i,size:o}=a,[l,c]=o,u=new PJ(r.shape,l,c,s,i);return n.runWebGLProgram(u,[r],r.dtype)}var LJ={kernelName:wc,backendName:"webgl",kernelFunc:OJ},zJ=class{constructor(e,t,n){this.variableNames=["dy"],this.outputShape=[],this.outputShape=t;let[,a,r]=t,[,s,i]=e,o=[n&&s>1?a-1:a,n&&i>1?r-1:r],l=[n&&s>1?s-1:s,n&&i>1?i-1:i],c=o[0]/l[0],u=o[1]/l[1],p=1/c,d=1/u,h=Math.ceil(p)*2+2,m=Math.ceil(d)*2+2;this.userCode=`
|
|
void main() {
|
|
ivec4 coords = getOutputCoords();
|
|
int b = coords[0];
|
|
int d = coords[3];
|
|
int r = coords[1];
|
|
int c = coords[2];
|
|
|
|
float accumulator = 0.0;
|
|
|
|
const float heightScale = float(${c});
|
|
const float widthScale = float(${u});
|
|
|
|
const float invHeightScale = float(${p});
|
|
const float invWidthScale = float(${d});
|
|
|
|
const int winHeight = int(${h});
|
|
const int winWidth = int(${m});
|
|
|
|
// 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 >= ${s}) {
|
|
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 >= ${i}) {
|
|
continue;
|
|
}
|
|
|
|
float sourceFracRow =
|
|
float(${o[0]}) *
|
|
(float(dyR) / float(${l[0]}));
|
|
|
|
float sourceFracCol =
|
|
float(${o[1]}) *
|
|
(float(dyC) / float(${l[1]}));
|
|
|
|
int sourceNearestRow = int(min(
|
|
float(int(${a}) - 1),
|
|
${n} ? float(round(sourceFracRow)) :
|
|
float(floor(sourceFracRow))));
|
|
|
|
int sourceNearestCol = int(min(
|
|
float(int(${r}) - 1),
|
|
${n} ? float(round(sourceFracCol)) :
|
|
float(floor(sourceFracCol))));
|
|
|
|
if (r == sourceNearestRow && c == sourceNearestCol) {
|
|
accumulator += getDy(b, dyR, dyC, d);
|
|
}
|
|
}
|
|
}
|
|
// End loop over dy
|
|
|
|
setOutput(accumulator);
|
|
}
|
|
`}};function BJ(e){let{inputs:t,backend:n,attrs:a}=e,{images:r,dy:s}=t,{alignCorners:i}=a,o=new zJ(s.shape,r.shape,i);return n.runWebGLProgram(o,[s],s.dtype)}var WJ={kernelName:Ud,backendName:"webgl",kernelFunc:BJ},VJ=class{constructor(e,t){this.variableNames=["x"];let n=e.length;if(n>4)throw new Error(`WebGL backend: Reverse of rank-${n} tensor is not yet supported`);if(this.outputShape=e,n===1){this.userCode=`
|
|
void main() {
|
|
int coord = getOutputCoords();
|
|
setOutput(getX(${e[0]} - coord - 1));
|
|
}
|
|
`;return}let a=i=>t.indexOf(i)!==-1&&e[i]!==1?`${e[i]} - coords[${i}] - 1`:`coords[${i}]`,r=e.map((i,o)=>a(o)).join(","),s=ft(n);this.userCode=`
|
|
void main() {
|
|
${s} coords = getOutputCoords();
|
|
setOutput(getX(${r}));
|
|
}
|
|
`}},UJ=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-${n} tensor is not yet supported`);this.outputShape=e;let a=mn("rc",n),r=`${a[n-1]} + 1 < ${this.outputShape[n-1]}`,s=`${a[n-2]} + 1 < ${this.outputShape[n-2]}`,i=ft(n);n===1?this.userCode=`
|
|
void main(){
|
|
int rc = getOutputCoords();
|
|
vec4 result = vec4(0.);
|
|
result.r = getChannel(getX(${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() {
|
|
${i} rc = getOutputCoords();
|
|
vec4 result = vec4(0.);
|
|
result.r = ${o(a.slice())};
|
|
if(${r}){
|
|
result.g = ${l(a.slice())};
|
|
}
|
|
if(${s}) {
|
|
result.b = ${c(a.slice())};
|
|
if(${r}) {
|
|
result.a = ${u(a.slice())};
|
|
}
|
|
}
|
|
setOutput(result);
|
|
}
|
|
`;function o(h){return p(h)}function l(h){return h[n-1]="("+h[n-1]+" + 1)",p(h)}function c(h){return h[n-2]="("+h[n-2]+" + 1)",p(h)}function u(h){return h[n-1]="("+h[n-1]+" + 1)",h[n-2]="("+h[n-2]+" + 1)",p(h)}function p(h){let m=e.map((y,b)=>d(b,h)),f=m.join(","),g=m.slice(-2).join(",");return`getChannel(getX(${f}), vec2(${g}))`}function d(h,m){return t.indexOf(h)!==-1&&e[h]!==1?`${e[h]} - ${m[h]} - 1`:`${m[h]}`}}};function GJ(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{dims:s}=a,i=r.shape.length,o=k.parseAxisParam(s,r.shape);if(i===0)return Bn({inputs:{x:r},backend:n});let l=ee().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new UJ(r.shape,o):new VJ(r.shape,o);return n.runWebGLProgram(l,[r],r.dtype)}var HJ={kernelName:ri,backendName:"webgl",kernelFunc:GJ},jJ=class{constructor(e,t,n,a){this.variableNames=["Image"],this.outputShape=[];let r=e[1],s=e[2],i=Math.sin(t).toFixed(3),o=Math.cos(t).toFixed(3);this.outputShape=e;let[l,c]=_.getImageCenter(a,r,s),u=l.toFixed(3),p=c.toFixed(3),d="";typeof n=="number"?d=`float outputValue = ${n.toFixed(2)};`:d=`
|
|
vec3 fill = vec3(${n.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) - ${u}) * ${o} - (float(y) - ${p}) * ${i};
|
|
float coordYFloat = (float(x) - ${u}) * ${i} + (float(y) - ${p}) * ${o};
|
|
int coordX = int(round(coordXFloat + ${u}));
|
|
int coordY = int(round(coordYFloat + ${p}));
|
|
${d}
|
|
if(coordX >= 0 && coordX < ${s} && coordY >= 0 && coordY < ${r}) {
|
|
outputValue = getImage(coords[0], coordY, coordX, coords[3]);
|
|
}
|
|
setOutput(outputValue);
|
|
}
|
|
`}},qJ={kernelName:Sl,backendName:"webgl",kernelFunc:({inputs:e,attrs:t,backend:n})=>{let{image:a}=e,{radians:r,fillValue:s,center:i}=t,o=n,l=new jJ(a.shape,r,s,i);return o.runWebGLProgram(l,[a],a.dtype)}},KJ=`
|
|
// OpenGL ES does not support round function.
|
|
// The algorithm is based on banker's rounding.
|
|
float base = floor(x);
|
|
if ((x - base) < 0.5) {
|
|
return floor(x);
|
|
} else if ((x - base) > 0.5) {
|
|
return ceil(x);
|
|
} else {
|
|
if (mod(base, 2.0) == 0.0) {
|
|
return base;
|
|
} else {
|
|
return base + 1.0;
|
|
}
|
|
}
|
|
`,XJ=Ze({opSnippet:KJ}),YJ={kernelName:si,backendName:"webgl",kernelFunc:XJ},ZJ="return inversesqrt(x);",JJ=Ze({opSnippet:ZJ,cpuKernelImpl:O8}),QJ={kernelName:ii,backendName:"webgl",kernelFunc:JJ},GS=class{constructor(e,t,n,a,r,s,i=!0){this.variableNames=["updates","indices","defaultValue"],this.outputShape=s;let o=ft(r.length),l=ft(s.length),c="";n===1?c="i":n===2&&(c="i, j");let u=`getIndices(${c})`,p="";a===1?p="i":a===2&&(p="i, coords[1]");let d=`getUpdates(${p})`,h=t>1?"strides[j]":"strides";this.userCode=`
|
|
${o} strides = ${o}(${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 += ${d};
|
|
found = true;
|
|
}
|
|
}
|
|
setOutput(mix(getDefaultValue(), sum, float(found)));
|
|
}
|
|
`}};function eQ(e){let{inputs:t,backend:n,attrs:a}=e,{indices:r,updates:s}=t,{shape:i}=a,{sliceRank:o,numUpdates:l,sliceSize:c,strides:u,outputSize:p}=_.calculateShapes(s,r,i),d=[p/c,c];if(p===0)return n.makeTensorInfo(i,r.dtype);let h=be({inputs:{x:r},backend:n,attrs:{shape:[l,o]}}),m=be({inputs:{x:s},backend:n,attrs:{shape:[l,c]}}),f=n.makeTensorInfo([],"float32",new Float32Array([0])),g=new GS(l,o,h.shape.length,m.shape.length,u,d),y=n.runWebGLProgram(g,[m,h,f],m.dtype),b=be({inputs:{x:y},backend:n,attrs:{shape:i}});return n.disposeIntermediateTensorInfo(h),n.disposeIntermediateTensorInfo(m),n.disposeIntermediateTensorInfo(y),n.disposeIntermediateTensorInfo(f),b}var tQ={kernelName:hl,backendName:"webgl",kernelFunc:eQ},nQ=class{constructor(e,t,n){this.variableNames=["c","a","b"],this.outputShape=t;let a,r;if(n>4)throw Error(`Where for rank ${n} is not yet supported`);if(n===1)r="resRC",a="resRC";else{let i=["resRC.x","resRC.y","resRC.z","resRC.w"],o=[],l=[];for(let c=0;c<t.length;c++)l.push(`${i[c]}`),c<e&&o.push(`${i[c]}`);a=o.join(),r=l.join()}let s=ft(n);this.userCode=`
|
|
void main() {
|
|
${s} resRC = getOutputCoords();
|
|
float cVal = getC(${a});
|
|
if (cVal >= 1.0) {
|
|
setOutput(getA(${r}));
|
|
} else {
|
|
setOutput(getB(${r}));
|
|
}
|
|
}
|
|
`}};function aQ(e){let{inputs:t,backend:n}=e,{condition:a,t:r,e:s}=t,i=new nQ(a.shape.length,r.shape,r.shape.length);return n.runWebGLProgram(i,[a,r,s],la(r.dtype,s.dtype))}var rQ={kernelName:ml,backendName:"webgl",kernelFunc:aQ},sQ=`
|
|
// Stable and Attracting Fixed Point (0, 1) for Normalized Weights.
|
|
// see: https://arxiv.org/abs/1706.02515
|
|
float scaleAlpha = ${_.SELU_SCALEALPHA};
|
|
float scale = ${_.SELU_SCALE};
|
|
return (x >= 0.0) ? scale * x : scaleAlpha * (exp(x) - 1.0);
|
|
`,iQ=Ze({opSnippet:sQ}),oQ={kernelName:fl,backendName:"webgl",kernelFunc:iQ},lQ="return 1.0 / (1.0 + exp(-1.0 * x));",uQ=Ze({opSnippet:lQ}),cQ={kernelName:li,backendName:"webgl",kernelFunc:uQ},pQ=`
|
|
if (isnan(x)) { return 0.0; }
|
|
return sign(x);
|
|
`,dQ=Ze({opSnippet:pQ}),hQ={kernelName:bl,backendName:"webgl",kernelFunc:dQ},mQ=aS+`
|
|
return sin(x);
|
|
`,fQ=Ze({opSnippet:mQ}),gQ={kernelName:oi,backendName:"webgl",kernelFunc:fQ},yQ=`
|
|
float e2x = exp(x);
|
|
return (e2x - 1.0 / e2x) / 2.0;
|
|
`,bQ=Ze({opSnippet:yQ}),xQ={kernelName:yl,backendName:"webgl",kernelFunc:bQ},vQ=`
|
|
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;
|
|
`,wQ=Ze({opSnippet:vQ}),kQ={kernelName:xl,backendName:"webgl",kernelFunc:wQ},IQ=e=>{let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{blockShape:s,paddings:i}=a;k.assert(r.shape.length<=4,()=>"spaceToBatchND for rank > 4 with a WebGL backend not implemented yet");let o=s.reduce((y,b)=>y*b),l=[[0,0]];l.push(...i);for(let y=1+s.length;y<r.shape.length;++y)l.push([0,0]);let c=[],u=VS({inputs:{x:r},backend:n,attrs:{paddings:l,constantValue:0}}),p=_.getReshaped(u.shape,s,o,!1),d=_.getPermuted(p.length,s.length,!1),h=_.getReshapedPermuted(u.shape,s,o,!1),m=be({inputs:{x:u},backend:n,attrs:{shape:p}}),f=Tn({inputs:{x:m},backend:n,attrs:{perm:d}}),g=be({inputs:{x:f},backend:n,attrs:{shape:h}});return c.push(u),c.push(m),c.push(f),c.forEach(y=>n.disposeIntermediateTensorInfo(y)),g},NQ={kernelName:kc,backendName:"webgl",kernelFunc:IQ};function TQ(e){let{inputs:t,backend:n,attrs:a}=e,{sparseIndices:r,sparseValues:s,defaultValue:i}=t,{outputShape:o}=a,{sliceRank:l,numUpdates:c,strides:u,outputSize:p}=_.calculateShapes(s,r,o),d=!1,h=new GS(c,l,r.shape.length,s.shape.length,u,[p,1],d),m=n.runWebGLProgram(h,[s,r,i],s.dtype),f=be({inputs:{x:m},backend:n,attrs:{shape:o}});return n.disposeIntermediateTensorInfo(m),f}var SQ={kernelName:Hd,backendName:"webgl",kernelFunc:TQ};function CQ(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{numOrSizeSplits:s,axis:i}=a,o=k.parseAxisParam(i,r.shape)[0],l=_.prepareSplitSize(r,s,o),c=r.shape.length,u=new Array(c).fill(0),p=r.shape.slice();return l.map(d=>{let h=[...p];h[o]=d;let m=Ip({inputs:{x:r},backend:n,attrs:{begin:u,size:h}});return u[o]+=d,m})}var _Q={kernelName:vl,backendName:"webgl",kernelFunc:CQ},EQ="return sqrt(x);",AQ=Ze({opSnippet:EQ}),FQ={kernelName:ui,backendName:"webgl",kernelFunc:AQ},$Q="return x * x;",DQ=Ze({opSnippet:$Q}),RQ={kernelName:Ic,backendName:"webgl",kernelFunc:DQ},HS="return (a - b) * (a - b);",MQ=nn({opSnippet:HS,packedOpSnippet:HS}),PQ={kernelName:di,backendName:"webgl",kernelFunc:MQ};function OQ({inputs:e,attrs:t,backend:n}){let{x:a}=e,r=Ra+`
|
|
return x > 0.0 ? 1.0 : float(${t.alpha});
|
|
`,s=new cs(a.shape,r);return n.runWebGLProgram(s,[a],a.dtype)}var LQ={kernelName:Wr,backendName:"webgl",kernelFunc:OQ},zQ=class{constructor(e,t,n){this.variableNames=["x"],this.outputShape=n;let a=n.length,r=ft(n.length),s=ft(n.length),i="";if(a===1)i="coords * strides + begin";else{let o=0;i=n.map((l,c)=>(o++,n.length===1?`coords * strides[${c}] + begin[${c}]`:`coords[${o-1}] * strides[${c}] + begin[${c}]`)).join(",")}this.userCode=`
|
|
${r} begin = ${r}(${e});
|
|
${r} strides = ${r}(${t});
|
|
|
|
void main() {
|
|
${s} coords = getOutputCoords();
|
|
setOutput(getX(${i}));
|
|
}
|
|
`}};function BQ(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{begin:s,end:i,strides:o,beginMask:l,endMask:c,ellipsisMask:u,newAxisMask:p,shrinkAxisMask:d}=a,{nonStrided:h,$begin:m,$strides:f,size:g,newShape:y,outShape:b}=cn.sliceInfo(r.shape,s,i,o,l,c,u,p,d),v=be({inputs:{x:r},backend:n,attrs:{shape:y}}),x;if(h){let T=Ip({inputs:{x:v},backend:n,attrs:{begin:m,size:g}});x=be({inputs:{x:T},backend:n,attrs:{shape:b}}),n.disposeIntermediateTensorInfo(T)}else if(b.some(T=>T===0))x=n.makeTensorInfo(b,r.dtype,[]);else if(n.shouldExecuteOnCPU([v])){let T=n.texData.get(v.dataId).values,C=Le(v.shape,v.dtype,T),$=z8(b,C,f,m);x=n.makeTensorInfo(b,v.dtype,$.values)}else{let T=new zQ(m,f,b);x=n.runWebGLProgram(T,[v],v.dtype)}let N=be({inputs:{x},backend:n,attrs:{shape:b}});return n.disposeIntermediateTensorInfo(v),n.disposeIntermediateTensorInfo(x),N}var WQ={kernelName:wl,backendName:"webgl",kernelFunc:BQ},VQ="return tan(x);",UQ=Ze({opSnippet:VQ}),GQ={kernelName:kl,backendName:"webgl",kernelFunc:UQ},HQ=`
|
|
float e2x = exp(-2.0 * abs(x));
|
|
return sign(x) * (1.0 - e2x) / (1.0 + e2x);
|
|
`,jQ=Ze({opSnippet:HQ}),qQ={kernelName:mi,backendName:"webgl",kernelFunc:jQ},XQ=class{constructor(e,t){this.variableNames=["A"];let n=new Array(e.length);for(let s=0;s<n.length;s++)n[s]=e[s]*t[s];this.outputShape=n,this.rank=n.length;let a=ft(this.rank),r=KQ(e);this.userCode=`
|
|
void main() {
|
|
${a} resRC = getOutputCoords();
|
|
setOutput(getA(${r}));
|
|
}
|
|
`}};function KQ(e){let t=e.length;if(t>5)throw Error(`Tile for rank ${t} is not yet supported`);if(t===1)return`imod(resRC, ${e[0]})`;let n=["resRC.x","resRC.y","resRC.z","resRC.w","resRC.u"],a=[];for(let r=0;r<e.length;r++)a.push(`imod(${n[r]}, ${e[r]})`);return a.join()}function jS(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{reps:s}=a;if(r.dtype==="string"){let o=n.readSync(r.dataId).map(u=>k.decodeString(u)),l=Le(r.shape,r.dtype,o),c=W8(l,s);return n.makeTensorInfo(c.shape,c.dtype,c.values)}let i=new XQ(r.shape,s);return n.runWebGLProgram(i,[r],r.dtype)}var YQ={kernelName:Br,backendName:"webgl",kernelFunc:jS};function ZQ(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{k:s,sorted:i}=a,o=n.readSync(r.dataId),[l,c]=V8(o,r.shape,r.dtype,s,i);return[n.makeTensorInfo(l.shape,l.dtype,l.values),n.makeTensorInfo(c.shape,c.dtype,c.values)]}var JQ={kernelName:Il,backendName:"webgl",kernelFunc:ZQ};function QQ(e){let{inputs:t,attrs:n,backend:a}=e,{axis:r}=n,{x:s}=t;xp(s,"unique"),console.warn("WARNING: ","UI might be locked temporarily as data is being downloaded");let i=a.readSync(s.dataId),{outputValues:o,outputShape:l,indices:c}=U8(i,r,s.shape,s.dtype);return[a.makeTensorInfo(l,s.dtype,o),a.makeTensorInfo([c.length],"int32",c)]}var eee={kernelName:jd,backendName:"webgl",kernelFunc:QQ};function tee(e){let{inputs:t,backend:n,attrs:a}=e,{value:r}=t,{axis:s}=a;s<0&&(s+=r.shape.length);let i=r,o=i.shape.length,l=r.shape[s],c=new Array(o-1),u=0;for(let f=0;f<o;f++)f!==s&&(c[u++]=i.shape[f]);let p=[],d=new Array(o).fill(0),h=i.shape.slice();h[s]=1;let m=new Array(l);for(let f=0;f<m.length;f++){d[s]=f;let g=Ip({inputs:{x:i},backend:n,attrs:{begin:d,size:h}}),y=be({inputs:{x:g},backend:n,attrs:{shape:c}});m[f]=y,p.push(g)}return p.forEach(f=>n.disposeIntermediateTensorInfo(f)),m}var nee={kernelName:Nl,backendName:"webgl",kernelFunc:tee},aee=class{constructor(e,t){this.variableNames=["x","segmentIds"];let n=e.windowSize,a=e.batchSize,r=e.inSize,s=e.numSegments,i=s*Math.ceil(r/n);this.outputShape=[a,i];let o="0.0",l="sumValue",c=Math.floor(n/4)*4,u=n%4,p=`
|
|
sumValue += dot(values, segFilter);
|
|
`,d="";r%n>0&&(d=`
|
|
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 = ${o};
|
|
|
|
float getValue(int batch, int inIdx) {
|
|
${d}
|
|
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(
|
|
${s})) * float(${n}));
|
|
int currentSeg = int(mod(float(outIdx), float(${s})));
|
|
|
|
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
|
|
);
|
|
|
|
${p}
|
|
}
|
|
|
|
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
|
|
);
|
|
|
|
${p}
|
|
} else if (${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
|
|
);
|
|
|
|
${p}
|
|
} else if (${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
|
|
);
|
|
|
|
${p}
|
|
}
|
|
setOutput(${l});
|
|
}
|
|
`}};function ree(e){let{inputs:t,backend:n,attrs:a}=e,{x:r,segmentIds:s}=t,{numSegments:i}=a,o=r.shape.length,l=[],c=0,u=_.getAxesPermutation([c],o),p=r;u!=null&&(p=Tn({inputs:{x:r},backend:n,attrs:{perm:u}}),l.push(p),c=_.getInnerMostAxes(1,o)[0]);let d=_.segment_util.computeOutShape(p.shape,c,i),h=k.sizeFromShape([p.shape[c]]),m=be({inputs:{x:p},backend:n,attrs:{shape:[-1,h]}});l.push(m);let f=Qd(r.dtype),g=(x,N,T,C,$)=>{let F=x.shape[0],O=x.shape[1],B=_.segment_util.segOpComputeOptimalWindowSize(O,$),W={windowSize:B,inSize:O,batchSize:F,numSegments:$},G=new aee(W,N),q=n.compileAndRun(G,[x,T],C);if(l.push(q),q.shape[1]===$)return q;let H=US({backend:n,attrs:{start:0,stop:$,step:1,dtype:"float32"}}),K=jS({inputs:{x:H},backend:n,attrs:{reps:[O/B]}});return l.push(H),l.push(K),g(q,N,K,C,$)},y=g(m,"unsortedSegmentSum",s,f,i),b=be({inputs:{x:y},backend:n,attrs:{shape:d}}),v=b;if(u!=null){l.push(b);let x=_.getUndoAxesPermutation(u);v=Tn({inputs:{x:v},backend:n,attrs:{perm:x}})}return l.forEach(x=>n.disposeIntermediateTensorInfo(x)),v}var see={kernelName:Nc,backendName:"webgl",kernelFunc:ree},iee=[JZ,t9,BK,VK,HK,KK,YK,QK,tX,aX,oX,uX,dX,fX,kX,bX,TX,EX,CX,DX,MX,OX,WX,KX,YX,nY,rY,lY,pY,wK,fY,TY,CY,xY,FY,DY,EY,PY,zY,VY,GY,jY,XY,t7,a7,ZY,i7,u7,h7,y7,w7,N7,T7,S7,_7,A7,$7,R7,P7,B7,G7,j7,K7,Z7,tZ,sZ,uZ,vK,pZ,mY,mZ,yZ,vZ,IK,NZ,_Z,AZ,OZ,RZ,WZ,GZ,KZ,a9,p9,u9,f9,y9,x9,o9,w9,I9,C9,F9,M9,U9,_K,H9,K9,Z9,eJ,JX,aJ,sJ,oJ,cJ,mJ,TK,gJ,yJ,QX,z9,vJ,_J,NJ,AK,$J,MJ,LJ,WJ,HJ,qJ,YJ,QJ,tQ,rQ,oQ,cQ,hQ,gQ,xQ,jX,W9,kQ,NQ,SQ,_Q,FQ,RQ,PQ,LQ,WQ,B9,OK,GQ,qQ,YQ,JQ,LK,eee,nee,see,rJ];for(let e of iee)Sc(e);var oee="3.0.0",lee={"tfjs-core":Y0,"tfjs-backend-cpu":LU,"tfjs-backend-webgl":xK,"tfjs-data":kN,"tfjs-layers":mm,"tfjs-converter":gN,tfjs:oee},Wn;(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"})(Wn||(Wn={}));var Tp;(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"})(Tp||(Tp={}));var qS;function uee(e){qS=e.wasm.cwrap(gi,null,["number","array","number","number","array","number","number","number","number","number","number","number","number"])}function cee(e){let{inputs:t,backend:n,attrs:a}=e,{a:r,b:s,bias:i,preluActivationWeights:o}=t;if(r.dtype!=="float32"||s.dtype!=="float32")throw new Error("_FusedMatMul for non non-float32 tensors not yet supported.");let{transposeA:l,transposeB:c,activation:u,leakyreluAlpha:p}=a,d=n.dataIdMap.get(r.dataId).id,h=n.dataIdMap.get(s.dataId).id,m=0;if(i!=null){let $=n.dataIdMap.get(i.dataId);if($.shape.length!==1)throw new Error(`_FusedMatMul only supports rank-1 bias but got rank ${$.shape.length}.`);m=$.id}let f=o==null?0:n.dataIdMap.get(o.dataId).id,g=Tp[u];if(g==null)throw new Error(`${u} activation not yet supported for FusedConv2D in the wasm backend.`);let y=l?r.shape[2]:r.shape[1],b=c?s.shape[1]:s.shape[2],v=r.shape[0],x=n.makeOutput([v,y,b],r.dtype),N=n.dataIdMap.get(x.dataId).id,T=new Uint8Array(new Int32Array(r.shape).buffer),C=new Uint8Array(new Int32Array(s.shape).buffer);return qS(d,T,r.shape.length,h,C,s.shape.length,l,c,g,m,f,p||0,N),x}var pee={kernelName:gi,backendName:"wasm",setupFunc:uee,kernelFunc:cee};function Vn(e){let t;function n(r){t=r.wasm.cwrap(e,null,["number","number"])}function a(r){let{backend:s,inputs:{x:i}}=r,o=s.dataIdMap.get(i.dataId).id,l=s.makeOutput(i.shape,i.dtype),c=s.dataIdMap.get(l.dataId).id;return k.sizeFromShape(l.shape)===0||t(o,c),l}return{kernelName:e,backendName:"wasm",setupFunc:n,kernelFunc:a}}var dee=Vn(So);function fn(e,t,n){let a;function r(i){a=i.wasm.cwrap(e,null,["number","array","number","number","array","number","number","number"])}function s(i){let{backend:o,inputs:l}=i,{a:c,b:u}=l,p=o.dataIdMap.get(c.dataId).id,d=o.dataIdMap.get(u.dataId).id,h=n!=null?n:c.dtype,m=_.assertAndGetBroadcastShape(c.shape,u.shape),f=o.makeOutput(m,h);if(k.sizeFromShape(m)===0)return f;let g=new Uint8Array(new Int32Array(c.shape).buffer),y=new Uint8Array(new Int32Array(u.shape).buffer),b=o.dataIdMap.get(f.dataId).id,v=()=>a(p,g,c.shape.length,d,y,u.shape.length,Wn[c.dtype],b);if(t&&c.dtype==="float32")return v(),f;let x=_.getBroadcastDims(c.shape,m),N=_.getBroadcastDims(u.shape,m),T=x.every(($,F)=>$===F),C=N.every(($,F)=>$===F);if(T&&C)return v(),f;throw new Error(`Broadcasting along outer dims is not yet supported for ${c.dtype} ${e}.`)}return{kernelName:e,backendName:"wasm",setupFunc:r,kernelFunc:s}}var hee=!0,mee=fn(Lr,hee),KS;function fee(e){KS=e.wasm.cwrap(Ss,null,["array","number","number","number"])}function gee(e){let{inputs:t,backend:n}=e,a=n.makeOutput(t[0].shape,t[0].dtype);if(k.sizeFromShape(a.shape)===0)return a;let r=t.map(o=>n.dataIdMap.get(o.dataId).id),s=new Uint8Array(new Int32Array(r).buffer),i=n.dataIdMap.get(a.dataId).id;return KS(s,r.length,Wn[a.dtype],i),a}var yee={kernelName:Ss,backendName:"wasm",setupFunc:fee,kernelFunc:gee};function Km(e){let{inputs:{x:t},backend:n}=e,a=n.makeOutput(t.shape,t.dtype),r=n.typedArrayFromHeap(t);return n.typedArrayFromHeap(a).set(r),a}var bee={kernelName:Ko,backendName:"wasm",kernelFunc:Km},XS;function xee(e){XS=e.wasm.cwrap(fi,null,["number","array","number","number","number","array","number"])}function Xm(e){let{inputs:t,backend:n,attrs:a}=e,[r,s]=wee(t.x.shape,a.perm),i=!0;for(let m=0;m<s.length;m++)s[m]!==m&&(i=!1);let o=vee(t.x.shape,a.perm),l={dataId:t.x.dataId,shape:r,dtype:t.x.dtype};if(i){let m=Km({inputs:t,backend:n});return m.shape=o,m}let c=n.makeOutput(o,l.dtype),u=n.dataIdMap.get(l.dataId).id,p=n.dataIdMap.get(c.dataId).id,d=new Uint8Array(new Int32Array(s).buffer),h=new Uint8Array(new Int32Array(l.shape).buffer);return XS(u,h,l.shape.length,Wn[l.dtype],p,d,s.length),c}function vee(e,t){let n=new Array(e.length);for(let a=0;a<n.length;a++)n[a]=e[t[a]];return n}function wee(e,t){let n=[],a=[];for(let r=0;r<e.length;++r)e[r]!==1&&n.push(e[r]),e[t[r]]!==1&&a.push(t[r]);for(let r=0;r<a.length;++r){let s=-1;for(let i=0;i<a.length;++i)a[i]>=r&&(s===-1||a[s]>a[i])&&(s=i);a[s]=r}return[n,a]}var kee={kernelName:fi,backendName:"wasm",kernelFunc:Xm,setupFunc:xee};function gu(e,t,n){let a=e.shape,r=e.shape.length,s=k.parseAxisParam(t,a),i=s,o=_.getAxesPermutation(i,r),l=null,c=!1;if(o!=null){let u=new Array(r);for(let d=0;d<u.length;d++)u[d]=a[o[d]];i=_.getInnerMostAxes(i.length,r),l=Xm({inputs:{x:e},attrs:{perm:o},backend:n});let p=n.dataIdMap.get(e.dataId).id;n.dataIdMap.get(l.dataId).id!==p&&(c=!0)}return{transposed:l,originalAxes:s,axes:i,inputWasTransposed:c}}var YS;function Iee(e){YS=e.wasm.cwrap(Cs,null,["number","number","number","number","number"])}function Nee(e){let{backend:t,inputs:n,attrs:a}=e,{axis:r}=a,{x:s}=n,i=t.dataIdMap.get(s.dataId).id,o=i,l=s,{transposed:c,axes:u,inputWasTransposed:p}=gu(s,r,t);if(p){let y=t.dataIdMap.get(c.dataId).id;y!==i&&(l=c,o=y)}let d=l.shape.slice(0,-1),h=t.makeOutput(d,"int32"),m=t.dataIdMap.get(h.dataId).id,f=k.sizeFromShape(h.shape),g=l.shape[u[0]];return YS(o,Wn[l.dtype],f,g,m),p&&t.disposeData(c.dataId),h}var Tee={kernelName:Cs,backendName:"wasm",kernelFunc:Nee,setupFunc:Iee},ZS;function See(e){ZS=e.wasm.cwrap(_s,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function Cee(e){let{inputs:t,attrs:n,backend:a}=e,r=t.x,s=a.dataIdMap.get(r.dataId).id,{filterSize:i,strides:o,pad:l,dimRoundingMode:c}=n,u=_.computePool2DInfo(r.shape,i,o,1,l,c),p=u.filterHeight,d=u.filterWidth,h=u.padInfo.top,m=u.padInfo.right,f=u.padInfo.bottom,g=u.padInfo.left,y=u.strideHeight,b=u.strideWidth,v=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 x=a.makeOutput(u.outShape,"float32"),N=a.dataIdMap.get(x.dataId).id;return ZS(s,r.shape[0],r.shape[1],r.shape[2],p,d,h,m,f,g,y,b,v,N),x}var _ee={kernelName:_s,backendName:"wasm",setupFunc:See,kernelFunc:Cee};function Ma(e){let{inputs:t,attrs:n}=e,{x:a}=t,{shape:r}=n,s=k.sizeFromShape(a.shape),i=k.inferFromImplicitShape(r,s);return k.assert(s===k.sizeFromShape(i),()=>`new shape: ${i}, old shape: ${a.shape}. New shape and old shape must have the same number of elements.`),{dataId:a.dataId,shape:i,dtype:a.dtype}}var Eee={kernelName:dl,backendName:"wasm",kernelFunc:Ma},JS;function Aee(e){JS=e.wasm.cwrap(Es,null,["number","array","number","number","array","number","number","number","number"])}function Fee(e){let{inputs:t,backend:n,attrs:a}=e,{a:r,b:s}=t,{transposeA:i,transposeB:o}=a;if(r.dtype!=="float32"||s.dtype!=="float32")throw new Error("BatchMatMul for non non-float32 tensors not yet supported.");let l=r.shape.length,c=s.shape.length,u=i?r.shape[l-2]:r.shape[l-1],p=o?s.shape[c-1]:s.shape[c-2],d=i?r.shape[l-1]:r.shape[l-2],h=o?s.shape[c-2]:s.shape[c-1],m=r.shape.slice(0,-2),f=s.shape.slice(0,-2),g=k.sizeFromShape(m),y=k.sizeFromShape(f),b=g===y||g===1||y===1;k.assert(l>=2&&c>=2&&b,()=>`Error in matMul: the input batch dimensions must either be the same or at least one input batch dimension must be 1. Got input batch dimensions of (${m}) and (${f}).`);let v=(g>y?r.shape.slice(0,-2):s.shape.slice(0,-2)).concat([d,h]);k.assert(u===p,()=>`Error in matMul: inner shapes (${u}) and (${p}) of Tensors with shapes ${r.shape} and ${s.shape} and transposeA=${i} and transposeB=${o} must match.`);let x=i?[g,u,d]:[g,d,u],N=o?[y,h,p]:[y,p,h],T=Ma({inputs:{x:r},backend:n,attrs:{shape:x}}),C=Ma({inputs:{x:s},backend:n,attrs:{shape:N}}),$=n.dataIdMap.get(T.dataId).id,F=n.dataIdMap.get(C.dataId).id,O=i?T.shape[2]:T.shape[1],B=o?C.shape[1]:C.shape[2],W=Math.max(g,y),G=n.makeOutput([W,O,B],T.dtype),q=n.dataIdMap.get(G.dataId).id,H=new Uint8Array(new Int32Array(T.shape).buffer),K=new Uint8Array(new Int32Array(C.shape).buffer);return JS($,H,T.shape.length,F,K,C.shape.length,i,o,q),G.shape=v,G}var $ee={kernelName:Es,backendName:"wasm",setupFunc:Aee,kernelFunc:Fee};function Ym(e){let{inputs:{x:t},attrs:{dtype:n},backend:a}=e,r=a.makeOutput(t.shape,n),s=a.typedArrayFromHeap(t);return a.typedArrayFromHeap(r).set(s),r}var Dee={kernelName:As,backendName:"wasm",kernelFunc:Ym},QS;function Ree(e){QS=e.wasm.cwrap(zr,null,["number","number","number","number"])}function Mee(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{clipValueMin:s,clipValueMax:i}=a,o=n.dataIdMap.get(r.dataId).id,l=n.makeOutput(r.shape,r.dtype),c=n.dataIdMap.get(l.dataId).id;return QS(o,s,i,c),l}var Pee={kernelName:zr,backendName:"wasm",setupFunc:Ree,kernelFunc:Mee};function e2(e){let{inputs:t,backend:n}=e,a=k.parseAxisParam(e.attrs.axis,t[0].shape)[0],r=_.computeOutShape(t.map(h=>h.shape),a),s=t.filter(h=>k.sizeFromShape(h.shape)>0);if(s.length===1)return Km({inputs:{x:s[0]},backend:n});let i=n.makeOutput(r,t[0].dtype);if(k.sizeFromShape(r)===0)return i;let o=s.map(h=>h.shape);if(_.assertParamsConsistent(o,a),s[0].dtype==="string"){let h=s.map(v=>{let x=k.sizeFromShape(v.shape.slice(a));return Ma({inputs:{x:v},backend:n,attrs:{shape:[-1,x]}})}),m=h.map(v=>({vals:n.readSync(v.dataId),shape:v.shape}));r=_.computeOutShape(h.map(v=>v.shape),1);let f=h[0].shape[0]===1,g=fv(m,r,t[0].dtype,f),y=_.computeOutShape(s.map(v=>v.shape),a);i.shape=y;let b=n.dataIdMap.get(i.dataId);return b.stringBytes=_.fromStringArrayToUint8(g),i}let l=k.sizeFromShape(s[0].shape.slice(0,a)),c=0,u=s.map(h=>{let m=k.sizeFromShape(h.shape.slice(a));return c+=m,m}),p=s.map(h=>n.typedArrayFromHeap(h)),d=n.typedArrayFromHeap(i);for(let h=0;h<l;h++){let m=h*c;for(let f=0;f<p.length;f++){let g=u[f],y=h*g,b=p[f].subarray(y,y+g);d.set(b,m),m+=g}}return i}var Oee={kernelName:Mo,backendName:"wasm",kernelFunc:e2},t2;function Lee(e){t2=e.wasm.cwrap(Fs,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function zee(e){let{inputs:t,attrs:n,backend:a}=e,{x:r,filter:s}=t,i=a.dataIdMap.get(r.dataId).id,o=a.dataIdMap.get(s.dataId).id,{strides:l,dilations:c,pad:u,dimRoundingMode:p,dataFormat:d}=n,h=_.convertConv2DDataFormat(d),m=_.computeConv2DInfo(r.shape,s.shape,l,c,u,p,!1,h),f=m.filterHeight,g=m.filterWidth,y=m.padInfo.top,b=m.padInfo.right,v=m.padInfo.bottom,x=m.padInfo.left,N=m.dilationHeight,T=m.dilationWidth,C=m.strideHeight,$=m.strideWidth,F=m.inChannels,O=m.outChannels,B=m.padInfo.type==="SAME"?1:0;if(m.dataFormat!=="channelsLast")throw new Error(`wasm backend Conv2D does not support dataFormat:'${m.dataFormat}'. Please use 'channelsLast'.`);let W=a.makeOutput(m.outShape,"float32"),G=a.dataIdMap.get(W.dataId).id;return t2(i,r.shape[0],r.shape[1],r.shape[2],o,f,g,y,b,v,x,B,N,T,C,$,F,O,G),W}var Bee={kernelName:Fs,backendName:"wasm",setupFunc:Lee,kernelFunc:zee},n2;function Wee(e){n2=e.wasm.cwrap($s,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function Vee(e){let{backend:t,inputs:n,attrs:a}=e,{dy:r,filter:s}=n,{strides:i,pad:o,dataFormat:l,dimRoundingMode:c,inputShape:u}=a,p=1,d=_.convertConv2DDataFormat(l),h=_.computeConv2DInfo(u,s.shape,i,p,o,c,!1,d),{batchSize:m,filterHeight:f,filterWidth:g,inChannels:y,inHeight:b,inWidth:v,outChannels:x,outHeight:N,outWidth:T,strideHeight:C,strideWidth:$}=h,F=f-1-h.padInfo.top,O=g-1-h.padInfo.left,B=h.dataFormat==="channelsLast",W=k.computeStrides(h.inShape),G=k.computeStrides(r.shape),[q,H,K]=k.computeStrides(s.shape),X=W[0],ne=B?W[1]:W[2],J=B?W[2]:1,se=B?1:W[1],Q=G[0],oe=B?G[1]:G[2],ae=B?G[2]:1,pe=B?1:G[1],ue=t.makeOutput(h.inShape,"float32"),ge=t.dataIdMap.get(ue.dataId).id,ye=t.dataIdMap.get(r.dataId).id,ke=t.dataIdMap.get(s.dataId).id;return n2(ye,ke,m,f,g,b,v,y,N,T,x,C,$,F,O,q,H,K,X,ne,J,se,Q,oe,ae,pe,ge),ue}var Uee={kernelName:$s,backendName:"wasm",setupFunc:Wee,kernelFunc:Vee},Gee=Vn(Ds),Hv;(function(e){e[e.bilinear=0]="bilinear",e[e.nearest=1]="nearest"})(Hv||(Hv={}));var a2;function Hee(e){a2=e.wasm.cwrap(Oo,null,["number","number","number","number","array","number","number","number","number","number"])}function jee(e){let{backend:t,inputs:n,attrs:a}=e,{method:r,extrapolationValue:s,cropSize:i}=a,{image:o,boxes:l,boxInd:c}=n,u=l.shape[0],[p,d]=i,h=[u,p,d,o.shape[3]],m=t.dataIdMap.get(o.dataId),f;o.dtype!=="float32"&&(f=Ym({backend:t,inputs:{x:o},attrs:{dtype:"float32"}}),m=t.dataIdMap.get(f.dataId));let g=m.id,y=t.dataIdMap.get(l.dataId).id,b=t.dataIdMap.get(c.dataId).id,v=t.makeOutput(h,"float32"),x=t.dataIdMap.get(v.dataId).id,N=new Uint8Array(new Int32Array(o.shape).buffer);return a2(g,y,b,u,N,p,d,Hv[r],s,x),f!=null&&t.disposeData(f.dataId),v}var qee={kernelName:Oo,backendName:"wasm",setupFunc:Hee,kernelFunc:jee},r2;function Kee(e){r2=e.wasm.cwrap(Rs,null,["number","number","number","number","number","number"])}function Xee(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{axis:s,exclusive:i,reverse:o}=a,l=r.shape.length;k.assert(r.dtype==="float32"||r.dtype==="int32",()=>`cumsum does not support ${r.dtype} tensors in the WASM backend`);let c=_.getAxesPermutation([s],l),u=r;c!==null&&(u=Xm({inputs:{x:r},attrs:{perm:c},backend:n}));let p=_.getInnerMostAxes(1,l)[0];_.assertAxesAreInnerMostDims("cumsum",[p],l);let d=n.makeOutput(u.shape,u.dtype),h=u.shape[p],m=n.dataIdMap.get(u.dataId).id,f=n.dataIdMap.get(d.dataId).id;r2(m,i?1:0,o?1:0,h,f,Wn[r.dtype]);let g=d;if(c!==null){let y=_.getUndoAxesPermutation(c);g=Xm({inputs:{x:d},attrs:{perm:y},backend:n}),n.disposeData(u.dataId),n.disposeData(d.dataId)}return g}var Yee={kernelName:Rs,backendName:"wasm",setupFunc:Kee,kernelFunc:Xee},s2;function Zee(e){s2=e.wasm.cwrap(Lo,null,["number","number","number","array","number","array","array","number","number"])}function Jee(e){let{backend:t,inputs:n,attrs:a}=e,{x:r}=n,{blockSize:s,dataFormat:i}=a;k.assert(s>1,()=>`blockSize should be > 1 for depthToSpace, but was: ${s}`);let o=r.shape[0],l=i==="NHWC"?r.shape[1]:r.shape[2],c=i==="NHWC"?r.shape[2]:r.shape[3],u=i==="NHWC"?r.shape[3]:r.shape[1],p=l*s,d=c*s,h=u/(s*s),m=i==="NHWC"?[o,p,d,h]:[o,h,p,d],f=t.makeOutput(m,"float32"),g=t.dataIdMap.get(r.dataId).id,y=new Uint8Array(new Int32Array(k.computeStrides(r.shape)).buffer),b=new Uint8Array(new Int32Array(m).buffer),v=new Uint8Array(new Int32Array(k.computeStrides(m)).buffer),x=t.dataIdMap.get(f.dataId).id;return s2(g,s,i==="NHWC"?1:0,y,r.shape.length-1,b,v,m.length,x),f}var Qee={kernelName:Lo,backendName:"wasm",setupFunc:Zee,kernelFunc:Jee},i2;function ete(e){i2=e.wasm.cwrap(Ms,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function tte(e){let{inputs:t,attrs:n,backend:a}=e,{x:r,filter:s}=t,i=a.dataIdMap.get(r.dataId).id,o=a.dataIdMap.get(s.dataId).id,{strides:l,dilations:c,pad:u,dimRoundingMode:p}=n,d=c==null?[1,1]:c,h=_.computeConv2DInfo(r.shape,s.shape,l,d,u,p,!0),m=h.filterHeight,f=h.filterWidth,g=h.padInfo.top,y=h.padInfo.right,b=h.padInfo.bottom,v=h.padInfo.left,x=h.dilationHeight,N=h.dilationWidth,T=h.strideHeight,C=h.strideWidth,$=h.inChannels,F=h.outChannels,O=h.padInfo.type==="SAME"?1:0;if(h.dataFormat!=="channelsLast")throw new Error(`wasm backend DepthwiseConv2dNative does not support dataFormat:'${h.dataFormat}'. Please use 'channelsLast'.`);let B=a.makeOutput(h.outShape,"float32"),W=a.dataIdMap.get(B.dataId).id;return i2(i,r.shape[0],r.shape[1],r.shape[2],o,m,f,g,y,b,v,O,x,N,T,C,$,F,W),B}var nte={kernelName:Ms,backendName:"wasm",setupFunc:ete,kernelFunc:tte},ate=!1,rte=fn(Wo,ate,"bool"),ste=Vn(Os);function jv(e){let{inputs:t,attrs:n,backend:a}=e,{input:r}=t,{dim:s}=n,i=r.shape.length,o=r.shape.slice(),l=s;return s<0&&(k.assert(-(i+1)<=s,()=>`Axis must be in the interval [${-(i+1)}, ${i}]`),l=i+s+1),o.splice(l,0,1),Ma({inputs:{x:r},backend:a,attrs:{shape:o}})}var ite={kernelName:Vo,backendName:"wasm",kernelFunc:jv};function ote(e){let{attrs:{shape:t,value:n,dtype:a},backend:r}=e,s=r.makeOutput(t,a);return r.typedArrayFromHeap(s).fill(n),s}var lte={kernelName:mc,backendName:"wasm",kernelFunc:ote},o2;function ute(e){o2=e.wasm.cwrap(Go,null,["number","number","number","number","number","number"])}function cte(e){let{inputs:t,backend:n}=e,{image:a}=t,r=n.makeOutput(a.shape,a.dtype),s=n.dataIdMap.get(a.dataId).id,i=n.dataIdMap.get(r.dataId).id,[o,l,c,u]=a.shape;return o2(s,o,l,c,u,i),r}var pte={kernelName:Go,backendName:"wasm",kernelFunc:cte,setupFunc:ute},dte=Vn(Ls),hte=!1,mte=fn(zs,hte),l2;function fte(e){l2=e.wasm.cwrap(Bs,null,["number","number","number","number","number","number","number"])}function gte(e){let{backend:t,inputs:n,attrs:a}=e,{varianceEpsilon:r}=a,{x:s,mean:i,variance:o,offset:l,scale:c}=n,u=t.dataIdMap.get(s.dataId).id,p=t.dataIdMap.get(i.dataId).id,d=t.dataIdMap.get(o.dataId).id,h=l!=null?t.dataIdMap.get(l.dataId).id:0,m=c!=null?t.dataIdMap.get(c.dataId).id:0,f=t.makeOutput(s.shape,s.dtype);if(k.sizeFromShape(s.shape)===0)return f;let g=t.dataIdMap.get(f.dataId).id;return l2(u,p,d,h,m,r,g),f}var yte={kernelName:Bs,backendName:"wasm",setupFunc:fte,kernelFunc:gte},u2;function bte(e){u2=e.wasm.cwrap(yi,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function xte(e){let{inputs:t,attrs:n,backend:a}=e,{x:r,filter:s,bias:i,preluActivationWeights:o}=t,{strides:l,pad:c,dilations:u,dataFormat:p,dimRoundingMode:d,activation:h,leakyreluAlpha:m}=n,f=_.computeConv2DInfo(r.shape,s.shape,l,u,c,d),g=Tp[h];if(g==null)throw new Error(`${h} activation not yet supported for FusedConv2D in the wasm backend.`);let y=a.dataIdMap.get(r.dataId).id,b=a.dataIdMap.get(s.dataId).id,v=f.outChannels,x=0;if(i!=null){let ae=a.dataIdMap.get(i.dataId);if(ae.shape.length!==1)throw new Error(`FusedConv2D only supports rank-1 bias but got rank ${ae.shape.length}.`);if(ae.shape[0]!==v)throw new Error(`FusedConv2D bias shape (${ae.shape}) does not match the number of output channels (${v})`);x=ae.id}let N=f.filterHeight,T=f.filterWidth,C=f.padInfo.top,$=f.padInfo.right,F=f.padInfo.bottom,O=f.padInfo.left,B=f.dilationHeight,W=f.dilationWidth,G=f.strideHeight,q=f.strideWidth,H=f.inChannels,K=f.padInfo.type==="SAME"?1:0,X=f.batchSize,ne=f.inHeight,J=f.inWidth;if(p!=="NHWC")throw new Error(`wasm backend FusedConv2D does not support dataFormat:'${p}'. Please use 'NHWC'.`);let se=a.makeOutput(f.outShape,"float32"),Q=a.dataIdMap.get(se.dataId).id,oe=o==null?0:a.dataIdMap.get(o.dataId).id;return u2(y,X,ne,J,b,N,T,x,C,$,F,O,K,B,W,G,q,H,v,g,oe,m||0,Q),se}var vte={kernelName:yi,backendName:"wasm",setupFunc:bte,kernelFunc:xte},c2;function wte(e){c2=e.wasm.cwrap(bi,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 kte(e){let{inputs:t,attrs:n,backend:a}=e,{x:r,filter:s,bias:i,preluActivationWeights:o}=t,{strides:l,pad:c,dilations:u,dataFormat:p,dimRoundingMode:d,activation:h,leakyreluAlpha:m}=n,f=_.computeConv2DInfo(r.shape,s.shape,l,u,c,d,!0),g=Tp[h];if(g==null)throw new Error(`${h} activation not yet supported for FusedDepthwiseConv2D in the wasm backend.`);let y=a.dataIdMap.get(r.dataId).id,b=a.dataIdMap.get(s.dataId).id,v=f.outChannels,x=0;if(i!=null){let ae=a.dataIdMap.get(i.dataId);if(ae.shape.length!==1)throw new Error(`FusedDepthwiseConv2D only supports rank-1 bias but got rank ${ae.shape.length}.`);if(ae.shape[0]!==v)throw new Error(`FusedDepthwiseConv2D bias shape (${ae.shape}) does not match the number of output channels (${v})`);x=ae.id}let N=f.filterHeight,T=f.filterWidth,C=f.padInfo.top,$=f.padInfo.right,F=f.padInfo.bottom,O=f.padInfo.left,B=f.dilationHeight,W=f.dilationWidth,G=f.strideHeight,q=f.strideWidth,H=f.inChannels,K=f.padInfo.type==="SAME"?1:0,X=f.batchSize,ne=f.inHeight,J=f.inWidth;if(p!=="NHWC")throw new Error(`wasm backend FusedDepthwiseConv2D does not support dataFormat:'${p}'. Please use 'NHWC'.`);let se=a.makeOutput(f.outShape,"float32"),Q=a.dataIdMap.get(se.dataId).id,oe=o==null?0:a.dataIdMap.get(o.dataId).id;return c2(y,X,ne,J,b,N,T,x,C,$,F,O,K,B,W,G,q,H,v,g,oe,m||0,Q),se}var Ite={kernelName:bi,backendName:"wasm",setupFunc:wte,kernelFunc:kte},p2;function Nte(e){p2=e.wasm.cwrap(jo,null,["number","number","number","number","number","number","array","number"])}function Tte(e){let{backend:t,inputs:n}=e,{params:a,indices:r}=n,[s,i,o,l]=fy.prepareAndValidate(a,r),c=t.makeOutput(s,a.dtype);if(i===0)return c;let u=r.shape,p=u[u.length-1],d=t.dataIdMap.get(a.dataId).id,h=t.dataIdMap.get(r.dataId).id,m=new Uint8Array(new Int32Array(l).buffer),f=t.dataIdMap.get(c.dataId).id;return p2(d,Wn[a.dtype],h,i,p,o,m,f),c}var Ste={kernelName:jo,backendName:"wasm",setupFunc:Nte,kernelFunc:Tte},d2;function Cte(e){d2=e.wasm.cwrap("Gather",null,["number","number","array","number","number","number","array","number"])}function _te(e){let{backend:t,inputs:n,attrs:a}=e,{x:r,indices:s}=n,{axis:i,batchDims:o}=a,l=k.parseAxisParam(i,r.shape)[0],c=_.segment_util.collectGatherOpShapeInfo(r,s,l,o),u=Ma({inputs:{x:r},attrs:{shape:[c.batchSize,c.outerSize,c.dimSize,c.sliceSize]},backend:t}),p=k.sizeFromShape(s.shape),d=Ma({inputs:{x:s},attrs:{shape:[c.batchSize,p/c.batchSize]},backend:t}),h=[c.batchSize,c.outerSize,p/c.batchSize,c.sliceSize],m=t.makeOutput(h,r.dtype);if(k.sizeFromShape(r.shape)===0)return m;let f=u.shape.length-1,g=t.dataIdMap.get(u.dataId).id,y=t.dataIdMap.get(d.dataId).id,b=t.dataIdMap.get(m.dataId).id,v=new Uint8Array(new Int32Array(k.computeStrides(u.shape)).buffer),x=new Uint8Array(new Int32Array(k.computeStrides(h)).buffer);return d2(g,Wn[r.dtype],v,f,y,c.batchSize,x,b),m.shape=c.outputShape,m}var Ete={kernelName:Ho,backendName:"wasm",setupFunc:Cte,kernelFunc:_te},Ate=!1,Fte=fn(qo,Ate,"bool"),$te=!1,Dte=fn(Ws,$te,"bool"),h2;function Rte(e){h2=e.wasm.cwrap(Vs,null,["number","number","number"])}function Mte(e){let{inputs:{x:t},attrs:{alpha:n},backend:a}=e,r=a.dataIdMap.get(t.dataId).id,s=a.makeOutput(t.shape,t.dtype);if(k.sizeFromShape(t.shape)!==0){let i=a.dataIdMap.get(s.dataId).id;h2(r,n,i)}return s}var Pte={kernelName:Vs,backendName:"wasm",setupFunc:Rte,kernelFunc:Mte},Ote=!1,Lte=fn(Jo,Ote,"bool"),zte=!1,Bte=fn(Qo,zte,"bool"),Wte=Vn(Us),Vte=!1,Ute=fn(tl,Vte,"bool"),m2;function Gte(e){m2=e.wasm.cwrap(Gs,null,["number, number, number"])}function Hte(e){let{backend:t,inputs:n,attrs:a}=e,{reductionIndices:r,keepDims:s}=a,{x:i}=n,o=t.dataIdMap.get(i.dataId).id,l=i,{transposed:c,axes:u,originalAxes:p,inputWasTransposed:d}=gu(i,r,t);if(d){let b=t.dataIdMap.get(c.dataId).id;l=c,o=b}let h=l.shape.length;_.assertAxesAreInnerMostDims("max",u,h);let[m,f]=_.computeOutAndReduceShapes(l.shape,u),g=k.sizeFromShape(f),y=t.makeOutput(m,i.dtype);if(k.sizeFromShape(l.shape)!==0){let b=t.dataIdMap.get(y.dataId).id;m2(o,g,b)}if(d&&t.disposeData(c.dataId),s){let b=_.expandShapeToKeepDim(y.shape,p);y.shape=b}return y}var jte={kernelName:Gs,backendName:"wasm",setupFunc:Gte,kernelFunc:Hte},qte=!1,Kte=fn(Hs,qte),f2;function Xte(e){f2=e.wasm.cwrap(js,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function Yte(e){let{inputs:t,attrs:n,backend:a}=e,r=t.x,s=a.dataIdMap.get(r.dataId).id,{filterSize:i,strides:o,pad:l,dimRoundingMode:c}=n,u=_.computePool2DInfo(r.shape,i,o,1,l,c),p=u.filterHeight,d=u.filterWidth,h=u.padInfo.top,m=u.padInfo.right,f=u.padInfo.bottom,g=u.padInfo.left,y=u.dilationHeight,b=u.dilationWidth,v=u.strideHeight,x=u.strideWidth,N=u.inChannels,T=u.outChannels;if(u.dataFormat!=="channelsLast")throw new Error(`wasm backend does not support dataFormat:'${u.dataFormat}'. Please use 'channelsLast'.`);let C=a.makeOutput(u.outShape,"float32"),$=a.dataIdMap.get(C.dataId).id;return f2(s,r.shape[0],r.shape[1],r.shape[2],p,d,h,m,f,g,y,b,v,x,N,T,$),C}var Zte={kernelName:js,backendName:"wasm",setupFunc:Xte,kernelFunc:Yte},g2;function Jte(e){g2=e.wasm.cwrap(qs,null,["number, number, number"])}function Qte(e){let{backend:t,inputs:n,attrs:a}=e,{axis:r,keepDims:s}=a,{x:i}=n,o=t.dataIdMap.get(i.dataId).id,l=o,c=i,{transposed:u,axes:p,originalAxes:d,inputWasTransposed:h}=gu(i,r,t),m=p;if(h){let x=t.dataIdMap.get(u.dataId).id;x!==o&&(c=u,l=x,m=_.getInnerMostAxes(m.length,c.shape.length))}_.assertAxesAreInnerMostDims("mean",m,c.shape.length);let[f,g]=_.computeOutAndReduceShapes(c.shape,m),y=k.sizeFromShape(g),b=c;c.dtype!=="float32"&&(b=Ym({backend:t,inputs:{x:c},attrs:{dtype:"float32"}}),l=t.dataIdMap.get(b.dataId).id);let v=t.makeOutput(f,"float32");if(k.sizeFromShape(c.shape)!==0){let x=t.dataIdMap.get(v.dataId).id;g2(l,y,x)}if(h&&t.disposeData(u.dataId),s){let x=_.expandShapeToKeepDim(v.shape,d);v.shape=x}return c.dtype!=="float32"&&t.disposeData(b.dataId),v}var ene={kernelName:qs,backendName:"wasm",setupFunc:Jte,kernelFunc:Qte},y2;function tne(e){y2=e.wasm.cwrap(Ks,null,["number, number, number"])}function nne(e){let{backend:t,inputs:n,attrs:a}=e,{axis:r,keepDims:s}=a,{x:i}=n,o=t.dataIdMap.get(i.dataId).id,l=o,c=i,{transposed:u,axes:p,originalAxes:d,inputWasTransposed:h}=gu(i,r,t);if(h){let v=t.dataIdMap.get(u.dataId).id;v!==o&&(c=u,l=v)}let m=c.shape.length;_.assertAxesAreInnerMostDims("min",p,m);let[f,g]=_.computeOutAndReduceShapes(c.shape,p),y=k.sizeFromShape(g),b=t.makeOutput(f,c.dtype);if(k.sizeFromShape(c.shape)!==0){let v=t.dataIdMap.get(b.dataId).id;y2(l,y,v)}if(h&&t.disposeData(u.dataId),s){let v=_.expandShapeToKeepDim(b.shape,d);b.shape=v}return b}var ane={kernelName:Ks,backendName:"wasm",setupFunc:tne,kernelFunc:nne},rne=!1,sne=fn(Xs,rne),ine=!0,one=fn(Ys,ine),lne=Vn(al);function qv(e,t){let n=new Int32Array(e.wasm.HEAPU8.buffer,t,4),a=n[0],r=n[1],s=n[2],i=n[3];return e.wasm._free(t),{pSelectedIndices:a,selectedSize:r,pSelectedScores:s,pValidOutputs:i}}var b2;function une(e){b2=e.wasm.cwrap(sl,"number",["number","number","number","number","number"])}function cne(e){let{backend:t,inputs:n,attrs:a}=e,{iouThreshold:r,maxOutputSize:s,scoreThreshold:i}=a,{boxes:o,scores:l}=n,c=t.dataIdMap.get(o.dataId).id,u=t.dataIdMap.get(l.dataId).id,p=b2(c,u,s,r,i),{pSelectedIndices:d,selectedSize:h,pSelectedScores:m,pValidOutputs:f}=qv(t,p);return t.wasm._free(m),t.wasm._free(f),t.makeOutput([h],"int32",d)}var pne={kernelName:sl,backendName:"wasm",setupFunc:une,kernelFunc:cne},x2;function dne(e){x2=e.wasm.cwrap(il,"number",["number","number","number","number","number","bool"])}function hne(e){let{backend:t,inputs:n,attrs:a}=e,{iouThreshold:r,maxOutputSize:s,scoreThreshold:i,padToMaxOutputSize:o}=a,{boxes:l,scores:c}=n,u=t.dataIdMap.get(l.dataId).id,p=t.dataIdMap.get(c.dataId).id,d=x2(u,p,s,r,i,o),{pSelectedIndices:h,selectedSize:m,pSelectedScores:f,pValidOutputs:g}=qv(t,d);t.wasm._free(f);let y=t.makeOutput([m],"int32",h),b=t.makeOutput([],"int32",g);return[y,b]}var mne={kernelName:il,backendName:"wasm",setupFunc:dne,kernelFunc:hne},v2;function fne(e){v2=e.wasm.cwrap(ol,"number",["number","number","number","number","number","number"])}function gne(e){let{backend:t,inputs:n,attrs:a}=e,{iouThreshold:r,maxOutputSize:s,scoreThreshold:i,softNmsSigma:o}=a,{boxes:l,scores:c}=n,u=t.dataIdMap.get(l.dataId).id,p=t.dataIdMap.get(c.dataId).id,d=v2(u,p,s,r,i,o),{pSelectedIndices:h,selectedSize:m,pSelectedScores:f,pValidOutputs:g}=qv(t,d);t.wasm._free(g);let y=t.makeOutput([m],"int32",h),b=t.makeOutput([m],"float32",f);return[y,b]}var yne={kernelName:ol,backendName:"wasm",setupFunc:fne,kernelFunc:gne},bne=!1,xne=fn(rl,bne,"bool"),w2;function vne(e){w2=e.wasm.cwrap(Zs,null,["number","number","number","number","number"])}function wne(e){let{inputs:t,backend:n,attrs:a}=e,{indices:r}=t,{depth:s,onValue:i,offValue:o}=a,l=n.makeOutput([...r.shape,s],"int32"),c=n.dataIdMap.get(l.dataId).id,u=n.dataIdMap.get(r.dataId).id;return w2(u,s,i,o,c),l}var kne={kernelName:Zs,backendName:"wasm",setupFunc:vne,kernelFunc:wne};function Ine(e){let{inputs:{x:t},backend:n}=e,a=n.makeOutput(t.shape,t.dtype);return n.typedArrayFromHeap(a).fill(1),a}var Nne={kernelName:ll,backendName:"wasm",kernelFunc:Ine};function Tne(e){let{inputs:t,backend:n,attrs:a}=e,{axis:r}=a;if(t.length===1)return jv({inputs:{input:t[0]},backend:n,attrs:{dim:r}});let s=t[0].shape,i=t[0].dtype;t.forEach(l=>{k.assertShapesMatch(s,l.shape,"All tensors passed to stack must have matching shapes"),k.assert(i===l.dtype,()=>"All tensors passed to stack must have matching dtypes")});let o=t.map(l=>jv({inputs:{input:l},backend:n,attrs:{dim:r}}));return e2({inputs:o,backend:n,attrs:{axis:r}})}var Sne={kernelName:ul,backendName:"wasm",kernelFunc:Tne},k2;function Cne(e){k2=e.wasm.cwrap(Js,null,["number","array","number","number","array","array","number","number"])}function _ne(e){let{inputs:{x:t},backend:n,attrs:{paddings:a,constantValue:r}}=e,s=a.map((m,f)=>m[0]+t.shape[f]+m[1]),i=n.dataIdMap.get(t.dataId).id,o=n.makeOutput(s,t.dtype),l=n.dataIdMap.get(o.dataId).id,c=new Uint8Array(new Int32Array(t.shape).buffer),u=a.map(m=>m[0]),p=a.map(m=>m[1]),d=new Uint8Array(new Int32Array(u).buffer),h=new Uint8Array(new Int32Array(p).buffer);return k2(i,c,t.shape.length,Wn[t.dtype],d,h,r,l),o}var Ene={kernelName:Js,backendName:"wasm",kernelFunc:_ne,setupFunc:Cne},Ane=!1,Fne=fn(Qs,Ane),I2;function $ne(e){I2=e.wasm.cwrap(ei,null,["number","number","number"])}function Dne(e){let{inputs:t,backend:n}=e,{x:a,alpha:r}=t,s=n.dataIdMap.get(a.dataId).id,i=n.dataIdMap.get(r.dataId).id,o=n.makeOutput(a.shape,"float32"),l=n.dataIdMap.get(o.dataId).id;return I2(s,i,l),o}var Rne={kernelName:ei,backendName:"wasm",setupFunc:$ne,kernelFunc:Dne},N2;function Mne(e){N2=e.wasm.cwrap(cl,null,["number","number","number","number"])}function Pne(e){let{backend:t,inputs:n,attrs:a}=e,{axis:r,keepDims:s}=a,{x:i}=n,o=t.dataIdMap.get(i.dataId).id,l=o,c=i,{transposed:u,axes:p,originalAxes:d,inputWasTransposed:h}=gu(i,r,t),m=p;if(h){let v=t.dataIdMap.get(u.dataId).id;v!==o&&(c=u,l=v,m=_.getInnerMostAxes(m.length,c.shape.length))}_.assertAxesAreInnerMostDims("prod",m,c.shape.length);let[f,g]=_.computeOutAndReduceShapes(c.shape,m),y=k.sizeFromShape(g),b=t.makeOutput(f,c.dtype);if(k.sizeFromShape(c.shape)!==0){let v=t.dataIdMap.get(b.dataId).id;N2(l,y,Wn[b.dtype],v)}if(h&&t.disposeData(u.dataId),s){let v=_.expandShapeToKeepDim(b.shape,d);b.shape=v}return b}var One={kernelName:cl,backendName:"wasm",setupFunc:Mne,kernelFunc:Pne},Lne=e=>{let{backend:t,attrs:n}=e,{start:a,stop:r,step:s,dtype:i}=n,o=bv(a,r,s,i),l=t.makeOutput([o.length],i);return t.typedArrayFromHeap(l).set(o),l},zne={kernelName:vc,backendName:"wasm",kernelFunc:Lne},Bne=!0,Wne=fn(Ps,Bne),Vne=Vn(ti),Une=Vn(ai),T2;function Gne(e){T2=e.wasm.cwrap(ni,null,["number","number","number","number","number","number","number","number","number","number"])}function Hne(e){let{backend:t,inputs:n,attrs:a}=e,{images:r}=n,{alignCorners:s,halfPixelCenters:i,size:o}=a,[l,c]=o,[u,p,d,h]=r.shape,m=[u,l,c,h],f=t.dataIdMap.get(r.dataId),g;f.dtype!=="float32"&&(g=Ym({backend:t,inputs:{x:r},attrs:{dtype:"float32"}}),f=t.dataIdMap.get(g.dataId));let y=f.id,b=t.makeOutput(m,"float32");if(k.sizeFromShape(r.shape)===0)return b;let v=t.dataIdMap.get(b.dataId).id;return T2(y,u,p,d,h,l,c,s?1:0,i?1:0,v),g!=null&&t.disposeData(g.dataId),b}var jne={kernelName:ni,backendName:"wasm",setupFunc:Gne,kernelFunc:Hne},S2;function qne(e){S2=e.wasm.cwrap(ri,null,["number","array","number","array","number","number"])}function Kne(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,{dims:s}=a,i=k.parseAxisParam(s,r.shape);if(r.shape.length===0)return Km({inputs:{x:r},backend:n});let o=n.makeOutput(r.shape,r.dtype),l=n.dataIdMap.get(r.dataId).id,c=n.dataIdMap.get(o.dataId).id,u=new Uint8Array(new Int32Array(i).buffer),p=new Uint8Array(new Int32Array(r.shape).buffer);return S2(l,u,i.length,p,r.shape.length,c),Ma({inputs:{x:o},attrs:{shape:r.shape},backend:n})}var Xne={kernelName:ri,backendName:"wasm",kernelFunc:Kne,setupFunc:qne},C2;function Yne(e){C2=e.wasm.cwrap(Sl,null,["number","number","number","number","number","number","number","number","array","number","number"])}function Zne(e){let{inputs:t,backend:n,attrs:a}=e,{image:r}=t,{radians:s,fillValue:i,center:o}=a,l=n.makeOutput(r.shape,r.dtype),c=n.dataIdMap.get(r.dataId).id,u=n.dataIdMap.get(l.dataId).id,[p,d,h,m]=r.shape,[f,g]=_.getImageCenter(o,d,h),y=i===0,b=255,v=typeof i=="number"?[i,i,i,y?0:b]:[...i,b],x=new Uint8Array(new Int32Array(v).buffer);return C2(c,p,d,h,m,s,f,g,x,v.length,u),l}var Jne={kernelName:Sl,backendName:"wasm",kernelFunc:Zne,setupFunc:Yne},Qne=Vn(si),eae=Vn(ii),_2;function tae(e){_2=e.wasm.cwrap(hl,null,["number","number","number","number","number","number","array","number","number"])}function nae(e){let{backend:t,inputs:n,attrs:a}=e,{indices:r,updates:s}=n,{shape:i}=a,o=t.makeOutput(i,s.dtype);if(k.sizeFromShape(i)===0)return o;let{sliceRank:l,numUpdates:c,sliceSize:u,strides:p,outputSize:d}=gy.calculateShapes(s,r,i),h=t.dataIdMap.get(r.dataId).id,m=t.dataIdMap.get(s.dataId).id,f=new Uint8Array(new Int32Array(p).buffer),g=t.dataIdMap.get(o.dataId).id;return _2(h,m,Wn[s.dtype],l,c,u,f,d,g),o}var aae={kernelName:hl,backendName:"wasm",setupFunc:tae,kernelFunc:nae},E2;function rae(e){E2=e.wasm.cwrap("SelectV2",null,["number","number","number","number","number"])}function sae(e){let{inputs:t,backend:n}=e,{condition:a,t:r,e:s}=t,i=n.dataIdMap.get(a.dataId).id,o=n.dataIdMap.get(r.dataId).id,l=n.dataIdMap.get(s.dataId).id,c=n.makeOutput(r.shape,r.dtype),u=n.dataIdMap.get(c.dataId).id,p=a.shape.length,d=r.shape.length,h=p===0||p>1||d===1?1:k.sizeFromShape(r.shape.slice(1));return E2(i,o,l,h,u),c}var iae={kernelName:ml,backendName:"wasm",kernelFunc:sae,setupFunc:rae},A2;function oae(e){A2=e.wasm.cwrap(li,null,["number","number"])}function lae(e){let{backend:t,inputs:{x:n}}=e,a=t.dataIdMap.get(n.dataId).id,r=t.makeOutput(n.shape,n.dtype),s=t.dataIdMap.get(r.dataId).id;return k.sizeFromShape(r.shape)===0||A2(a,s),r}var uae={kernelName:"Sigmoid",backendName:"wasm",setupFunc:oae,kernelFunc:lae},cae=Vn(oi);function Zm(e){let{inputs:{x:t},attrs:{begin:n,size:a},backend:r}=e,[s,i]=cn.parseSliceParams(t,n,a),o=cn.isSliceContinous(t.shape,s,i),l=r.readSync(t.dataId),c=r.makeOutput(i,t.dtype),u=k.computeStrides(t.shape),p=r.dataIdMap.get(c.dataId);if(o){let m=cn.computeFlatOffset(s,u);return t.dtype==="string"?p.stringBytes=l.slice(m,m+k.sizeFromShape(i)):r.typedArrayFromHeap(c).set(l.subarray(m,m+k.sizeFromShape(i))),c}if(t.dtype==="string"){let m=Am(l,s,i,t.shape,t.dtype);return p.stringBytes=m,c}let d=r.typedArrayFromHeap(c),h=t.shape.length;if(h===2)pae(l,u[0],d,s,i);else if(h===3)dae(l,u[0],u[1],d,s,i);else if(h===4)hae(l,u[0],u[1],u[2],d,s,i);else{let m=Am(l,s,i,t.shape,t.dtype);d.set(m)}return c}function pae(e,t,n,a,r){let s=0,i=a[0],o=a[1],l=i+r[0];for(let c=i;c<l;c++){let u=c*t+o;n.set(e.subarray(u,u+r[1]),s),s+=r[1]}}function dae(e,t,n,a,r,s){let i=0,o=r[0],l=r[1],c=r[2],u=o+s[0],p=l+s[1];for(let d=o;d<u;d++)for(let h=l;h<p;h++){let m=d*t+h*n+c;a.set(e.subarray(m,m+s[2]),i),i+=s[2]}}function hae(e,t,n,a,r,s,i){let o=0,l=s[0],c=s[1],u=s[2],p=l+i[0],d=c+i[1],h=u+i[2],m=s[3];for(let f=l;f<p;f++)for(let g=c;g<d;g++)for(let y=u;y<h;y++){let b=f*t+g*n+y*a+m;r.set(e.subarray(b,b+i[3]),o),o+=i[3]}}var mae={kernelName:gl,backendName:"wasm",kernelFunc:Zm},F2;function fae(e){F2=e.wasm.cwrap(pi,null,["number","number","number","number"])}function gae(e){let{backend:t,inputs:{logits:n},attrs:{dim:a}}=e,r=t.dataIdMap.get(n.dataId).id,s=t.makeOutput(n.shape,n.dtype),i=t.dataIdMap.get(s.dataId).id,o=n.shape[a],l=k.sizeFromShape(n.shape)/o;return k.sizeFromShape(s.shape)===0||F2(r,i,o,l),s}var yae={kernelName:pi,backendName:"wasm",setupFunc:fae,kernelFunc:gae};function bae(e){let{inputs:t,attrs:n,backend:a}=e,{x:r}=t,{numOrSizeSplits:s,axis:i}=n,o=k.parseAxisParam(i,r.shape)[0],l=_.prepareSplitSize(r,s,o),c=new Array(r.shape.length).fill(0),u=r.shape.slice();return l.map(p=>{let d=[...u];d[o]=p;let h=Zm({inputs:{x:r},attrs:{begin:c,size:d},backend:a});return c[o]+=p,h})}var xae={kernelName:vl,backendName:"wasm",kernelFunc:bae},vae=Vn(ui),wae=Vn(Ic),kae=!0,Iae=fn(di,kae),$2;function Nae(e){$2=e.wasm.cwrap(Wr,null,["number","number","number"])}function Tae(e){let{backend:t,inputs:n,attrs:a}=e,{alpha:r}=a,{x:s}=n,i=t.dataIdMap.get(s.dataId).id,o=t.makeOutput(s.shape,s.dtype),l=t.dataIdMap.get(o.dataId).id;return $2(i,r,l),o}var Sae={kernelName:Wr,backendName:"wasm",setupFunc:Nae,kernelFunc:Tae},D2;function Cae(e){D2=e.wasm.cwrap(wl,null,["number","array","number","array","array","array","array","array","number","number"])}function _ae(e){let{backend:t,inputs:n,attrs:a}=e,{x:r}=n,{begin:s,end:i,strides:o}=a;o==null&&(o=new Array(s.length));let{beginMask:l,endMask:c,ellipsisMask:u,newAxisMask:p,shrinkAxisMask:d}=a,h=_.slice_util.maskToAxes(u);if(h.length>1)throw new Error("Multiple ellipses in slice is not allowed.");if(u!==0&&p!==0)throw new Error("Using both ellipsisMask and newAxisMask is not yet supported.");if(u!==0&&d!==0)throw new Error("Using both ellipsisMask and shrinkAxisMask is not yet supported.");let m=r.shape.length-s.length,f=_.slice_util.maskToAxes(p),g=r.shape.slice();f.forEach(F=>{s[F]=0,i[F]=1,g.splice(F,0,1)});let y=Ma({inputs:{x:r},attrs:{shape:g},backend:t}),{begin:b,end:v,strides:x}=_.slice_util.getNormalizedAxes(y.shape,h,m,s,i,o,l,c,u);s=b,i=v,o=x;let N=_.slice_util.maskToAxes(d);N.forEach(F=>{i[F]=s[F]+1,o[F]=1});let T=_.slice_util.computeOutShape(s,i,o),C=T.filter((F,O)=>N.indexOf(O)===-1);if(o.every(F=>F===1)){let F=Zm({inputs:{x:r},attrs:{begin:s,size:T},backend:t});return Ma({inputs:{x:F},attrs:{shape:C},backend:t})}let $=t.makeOutput(C,"float32");if(!C.some(F=>F===0)){let F=t.dataIdMap.get(y.dataId).id,O=new Uint8Array(new Int32Array(k.computeStrides(y.shape)).buffer),B=new Uint8Array(new Int32Array(s).buffer),W=new Uint8Array(new Int32Array(i).buffer),G=new Uint8Array(new Int32Array(o).buffer),q=new Uint8Array(new Int32Array(C).buffer),H=new Uint8Array(new Int32Array(k.computeStrides(C)).buffer),K=t.dataIdMap.get($.dataId).id;D2(F,O,y.shape.length,B,W,G,q,H,C.length,K)}return Ma({inputs:{x:$},attrs:{shape:C},backend:t})}var Eae={kernelName:wl,backendName:"wasm",setupFunc:Cae,kernelFunc:_ae},Aae=!0,Fae=fn(hi,Aae),R2;function $ae(e){R2=e.wasm.cwrap(ci,null,["number, number, number"])}function Dae(e){let{backend:t,inputs:n,attrs:a}=e,{axis:r,keepDims:s}=a,{x:i}=n,o=t.dataIdMap.get(i.dataId).id,l=o,c=i,{transposed:u,axes:p,originalAxes:d,inputWasTransposed:h}=gu(i,r,t),m=p;if(h){let v=t.dataIdMap.get(u.dataId).id;v!==o&&(c=u,l=v,m=_.getInnerMostAxes(m.length,c.shape.length))}_.assertAxesAreInnerMostDims("sum",m,c.shape.length);let[f,g]=_.computeOutAndReduceShapes(c.shape,m),y=k.sizeFromShape(g),b=t.makeOutput(f,c.dtype);if(k.sizeFromShape(c.shape)!==0){let v=t.dataIdMap.get(b.dataId).id;R2(l,y,v)}if(h&&t.disposeData(u.dataId),s){let v=_.expandShapeToKeepDim(b.shape,d);b.shape=v}return b}var Rae={kernelName:ci,backendName:"wasm",setupFunc:$ae,kernelFunc:Dae},Mae=Vn(mi),M2;function Pae(e){M2=e.wasm.cwrap(Br,null,["number","array","number","array","number","number"])}function Oae(e){let{inputs:t,backend:n,attrs:a}=e,{x:r}=t,s=n.dataIdMap.get(r.dataId).id,{reps:i}=a,o=new Array(r.shape.length);for(let d=0;d<o.length;d++)o[d]=r.shape[d]*i[d];let l=new Uint8Array(new Int32Array(r.shape).buffer),c=new Uint8Array(new Int32Array(o).buffer),u=n.makeOutput(o,r.dtype),p=n.dataIdMap.get(u.dataId).id;return M2(s,l,r.shape.length,c,o.length,Wn[u.dtype],p),u}var Lae={kernelName:Br,backendName:"wasm",setupFunc:Pae,kernelFunc:Oae},P2;function zae(e){P2=e.wasm.cwrap(Il,null,["number","array","number","number","number","bool","number","number"])}var Bae=({inputs:e,backend:t,attrs:n})=>{let{x:a}=e,{k:r,sorted:s}=n,i=t.dataIdMap.get(a.dataId).id,o=new Uint8Array(new Int32Array(a.shape).buffer),l=a.shape.slice();l[l.length-1]=r;let c=t.makeOutput(l,a.dtype),u=t.dataIdMap.get(c.dataId).id,p=t.makeOutput(l,"int32"),d=t.dataIdMap.get(p.dataId).id;return P2(i,o,a.shape.length,Wn[a.dtype],r,s,u,d),[c,p]},Wae={kernelName:Il,backendName:"wasm",setupFunc:zae,kernelFunc:Bae};function Vae(e){let{inputs:t,backend:n,attrs:a}=e,{value:r}=t,{axis:s}=a;s<0&&(s+=r.shape.length);let i=r.shape[s],o=r.shape.length,l=new Array(o-1),c=0;for(let h=0;h<o;h++)h!==s&&(l[c++]=r.shape[h]);let u=new Array(i),p=new Array(o).fill(0),d=r.shape.slice();d[s]=1;for(let h=0;h<u.length;h++)p[s]=h,u[h]=Zm({inputs:{x:r},attrs:{begin:p,size:d},backend:n});return u.map(({dataId:h,dtype:m})=>({dataId:h,dtype:m,shape:l}))}var Uae={kernelName:Nl,backendName:"wasm",kernelFunc:Vae};function Gae(e){let{inputs:{x:t},backend:n}=e,a=n.makeOutput(t.shape,t.dtype);return n.typedArrayFromHeap(a).fill(0),a}var Hae={kernelName:Tl,backendName:"wasm",kernelFunc:Gae},jae=[dee,mee,yee,Tee,_ee,$ee,Dee,Pee,Oee,Bee,Uee,Gee,qee,Yee,Qee,nte,rte,ste,ite,lte,pte,dte,mte,pee,yte,vte,Ite,Ste,Ete,Fte,Dte,bee,Pte,Lte,Bte,Wte,Ute,jte,Kte,Zte,ene,ane,sne,one,lne,pne,mne,yne,xne,kne,Nne,Sne,Ene,Fne,Rne,One,zne,Wne,Vne,Une,Eee,jne,Xne,Jne,eae,Qne,aae,iae,uae,cae,mae,yae,xae,vae,wae,Iae,Sae,Eae,Fae,Rae,Mae,Lae,Wae,kee,Uae,Hae];for(let e of jae)Sc(e);var Kv=ee();Kv.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])));Kv.registerFlag("WASM_HAS_MULTITHREAD_SUPPORT",async()=>{if(Kv.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 O2=Io(yE()),qae='var threadInfoStruct=0;var selfThreadId=0;var parentThreadId=0;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:selfThreadId})}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};this.onmessage=function(e){try{if(e.data.cmd==="load"){Module["DYNAMIC_BASE"]=e.data.DYNAMIC_BASE;Module["DYNAMICTOP_PTR"]=e.data.DYNAMICTOP_PTR;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)}Module=WasmBackendModuleThreadedSimd(Module);postMessage({"cmd":"loaded"})}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;threadInfoStruct=e.data.threadInfoStruct;Module["__register_pthread_ptr"](threadInfoStruct,0,0);selfThreadId=e.data.selfThreadId;parentThreadId=e.data.parentThreadId;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["dynCall_ii"](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"){Atomics.store(Module["HEAPU32"],threadInfoStruct+4>>2,ex instanceof Module["ExitStatus"]?ex.status:-2);Atomics.store(Module["HEAPU32"],threadInfoStruct+0>>2,1);Module["_emscripten_futex_wake"](threadInfoStruct+0,2147483647);if(!(ex instanceof Module["ExitStatus"]))throw ex}}}else if(e.data.cmd==="cancel"){if(threadInfoStruct){Module["PThread"].threadCancel()}}else if(e.data.target==="setimmediate"){}else if(e.data.cmd==="processThreadQueue"){if(threadInfoStruct){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.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");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()}}}}',Kae=Io(bE()),L2=class extends sc{constructor(e){super();this.wasm=e,this.dataIdNextNumber=1,this.wasm.tfjs.init(),this.dataIdMap=new hd(this,Kr())}write(e,t,n){let a={};return this.move(a,e,t,n),a}numDataIds(){return this.dataIdMap.numDataIds()}async time(e){let t=k.now();return e(),{kernelMs:k.now()-t}}move(e,t,n,a){let r=this.dataIdNextNumber++;if(a==="string"){let l=t;this.dataIdMap.set(e,{id:r,stringBytes:l,shape:n,dtype:a,memoryOffset:null});return}let s=k.sizeFromShape(n),i=s*k.bytesPerElement(a),o=this.wasm._malloc(i);this.dataIdMap.set(e,{id:r,memoryOffset:o,shape:n,dtype:a}),this.wasm.tfjs.registerTensor(r,s,o),t!=null&&this.wasm.HEAPU8.set(new Uint8Array(t.buffer,t.byteOffset,i),o)}async read(e){return this.readSync(e)}readSync(e){let{memoryOffset:t,dtype:n,shape:a,stringBytes:r}=this.dataIdMap.get(e);if(n==="string")return r;let s=this.wasm.HEAPU8.slice(t,t+k.sizeFromShape(a)*k.bytesPerElement(n));return Xae(s.buffer,n)}disposeData(e){let t=this.dataIdMap.get(e);this.wasm._free(t.memoryOffset),this.wasm.tfjs.disposeData(t.id),this.dataIdMap.delete(e)}floatPrecision(){return 32}getMemoryOffset(e){return this.dataIdMap.get(e).memoryOffset}dispose(){this.wasm.tfjs.dispose(),this.wasm=null}memory(){return{unreliable:!1}}makeOutput(e,t,n){let a;if(n==null)a=this.write(null,e,t);else{a={};let r=this.dataIdNextNumber++;this.dataIdMap.set(a,{id:r,memoryOffset:n,shape:e,dtype:t});let s=k.sizeFromShape(e);this.wasm.tfjs.registerTensor(r,s,n)}return{dataId:a,shape:e,dtype:t}}typedArrayFromHeap({shape:e,dtype:t,dataId:n}){let a=this.wasm.HEAPU8.buffer,{memoryOffset:r}=this.dataIdMap.get(n),s=k.sizeFromShape(e);switch(t){case"float32":return new Float32Array(a,r,s);case"int32":return new Int32Array(a,r,s);case"bool":return new Uint8Array(a,r,s);default:throw new Error(`Unknown dtype ${t}`)}}};function Yae(e){return(t,n)=>(k.fetch(e,{credentials:"same-origin"}).then(a=>{a.ok||t.env.a(`failed to load wasm binary file at '${e}'`),a.arrayBuffer().then(r=>{WebAssembly.instantiate(r,t).then(s=>{n(s.instance)})})}),{})}function z2(e,t,n){if(Jm!=null)return Jm;let a="tfjs-backend-wasm.wasm";return e&&t?a="tfjs-backend-wasm-threaded-simd.wasm":e&&(a="tfjs-backend-wasm-simd.wasm"),Sp!=null&&Sp[a]!=null?Sp[a]:n+a}async function Zae(){let[e,t]=await Promise.all([ee().getAsync("WASM_HAS_SIMD_SUPPORT"),ee().getAsync("WASM_HAS_MULTITHREAD_SUPPORT")]);return new Promise((n,a)=>{let r={};r.locateFile=(l,c)=>{if(l.endsWith(".worker.js")){let u=qae,p=new Blob([u],{type:"application/javascript"});return URL.createObjectURL(p)}return l.endsWith(".wasm")?z2(e,t,Cp!=null?Cp:c):c+l},Xv&&(r.instantiateWasm=Yae(z2(e,t,Cp!=null?Cp:"")));let s;t&&e&&Jm==null?(s=O2.default(r),s.mainScriptUrlOrBlob=new Blob(["var WasmBackendModuleThreadedSimd = "+O2.default.toString()],{type:"text/javascript"})):s=Kae.default(r);let i=null;s.tfjs={init:s.cwrap("init",null,[]),registerTensor:s.cwrap("register_tensor",null,["number","number","number"]),disposeData:s.cwrap("dispose_data",i,["number"]),dispose:s.cwrap("dispose",i,[])};let o=!1;s.onRuntimeInitialized=()=>{o=!0,_p=!1,n({wasm:s})},s.onAbort=()=>{o||_p||(_p=!0,a({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"}))}})}function Xae(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 Jae=["tfjs-backend-wasm.wasm","tfjs-backend-wasm-simd.wasm","tfjs-backend-wasm-threaded-simd.wasm"],Jm=null,Cp=null,Sp={},_p=!1,Xv=!1;function Qae(e,t=!1){if(ky("setWasmPath has been deprecated in favor of setWasmPaths and will be removed in a future release."),_p)throw new Error("The WASM backend was already initialized. Make sure you call `setWasmPath()` before you call `tf.setBackend()` or `tf.ready()`");Jm=e,Xv=t}function ere(e,t=!1){if(_p)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")Cp=e;else{Sp=e;let n=Jae.filter(a=>Sp[a]==null);if(n.length>0)throw new Error(`There were no entries found for the following binaries: ${n.join(",")}. Please either call setWasmPaths with a map providing a path for each binary, or with a string indicating the directory where all the binaries can be found.`)}Xv=t}var tre="3.0.0",nre=2;ih("wasm",async()=>{let{wasm:e}=await Zae();return new L2(e)},nre);var ww={};cd(ww,{AnchorPosition:()=>or,DrawBox:()=>nf,DrawBoxOptions:()=>cw,DrawFaceLandmarks:()=>vw,DrawFaceLandmarksOptions:()=>xw,DrawTextField:()=>hs,DrawTextFieldOptions:()=>$p,drawContour:()=>Tr,drawDetections:()=>hre,drawFaceExpressions:()=>yre,drawFaceLandmarks:()=>bre});function Tr(e,t,n=!1){if(e.beginPath(),t.slice(1).forEach(({x:a,y:r},s)=>{let i=t[s];e.moveTo(i.x,i.y),e.lineTo(a,r)}),n){let a=t[t.length-1],r=t[0];if(!a||!r)return;e.moveTo(a.x,a.y),e.lineTo(r.x,r.y)}e.stroke()}var Yv={};cd(Yv,{computeReshapedDimensions:()=>Qv,getCenterPoint:()=>Xi,isDimensions:()=>ef,isEven:()=>Qm,isFloat:()=>Jv,isTensor:()=>qi,isTensor1D:()=>are,isTensor2D:()=>Zv,isTensor3D:()=>Sr,isTensor4D:()=>ea,isValidNumber:()=>Pa,isValidProbablitiy:()=>yu,range:()=>sr,round:()=>Ki});var gn=class{constructor(t,n){if(!Pa(t)||!Pa(n))throw new Error(`Dimensions.constructor - expected width and height to be valid numbers, instead have ${JSON.stringify({width:t,height:n})}`);this._width=t,this._height=n}get width(){return this._width}get height(){return this._height}reverse(){return new gn(1/this.width,1/this.height)}};function qi(e,t){return e instanceof Ae&&e.shape.length===t}function are(e){return qi(e,1)}function Zv(e){return qi(e,2)}function Sr(e){return qi(e,3)}function ea(e){return qi(e,4)}function Jv(e){return e%1!=0}function Qm(e){return e%2==0}function Ki(e,t=2){let n=10**t;return Math.floor(e*n)/n}function ef(e){return e&&e.width&&e.height}function Qv({width:e,height:t},n){let a=n/Math.max(t,e);return new gn(Math.round(e*a),Math.round(t*a))}function Xi(e){return e.reduce((t,n)=>t.add(n),new De(0,0)).div(new De(e.length,e.length))}function sr(e,t,n){return Array(e).fill(0).map((a,r)=>t+r*n)}function Pa(e){return!!e&&e!==Infinity&&e!==-Infinity&&!Number.isNaN(e)||e===0}function yu(e){return Pa(e)&&e>=0&&e<=1}var De=class{constructor(t,n){this._x=t,this._y=n}get x(){return this._x}get y(){return this._y}add(t){return new De(this.x+t.x,this.y+t.y)}sub(t){return new De(this.x-t.x,this.y-t.y)}mul(t){return new De(this.x*t.x,this.y*t.y)}div(t){return new De(this.x/t.x,this.y/t.y)}abs(){return new De(Math.abs(this.x),Math.abs(this.y))}magnitude(){return Math.sqrt(this.x**2+this.y**2)}floor(){return new De(Math.floor(this.x),Math.floor(this.y))}};var ut=class{static isRect(t){return!!t&&[t.x,t.y,t.width,t.height].every(Pa)}static assertIsValidBox(t,n,a=!1){if(!ut.isRect(t))throw new Error(`${n} - invalid box: ${JSON.stringify(t)}, expected object with properties x, y, width, height`);if(!a&&(t.width<0||t.height<0))throw new Error(`${n} - width (${t.width}) and height (${t.height}) must be positive numbers`)}constructor(t,n=!0){let a=t||{},r=[a.left,a.top,a.right,a.bottom].every(Pa),s=[a.x,a.y,a.width,a.height].every(Pa);if(!s&&!r)throw new Error(`Box.constructor - expected box to be IBoundingBox | IRect, instead have ${JSON.stringify(a)}`);let[i,o,l,c]=s?[a.x,a.y,a.width,a.height]:[a.left,a.top,a.right-a.left,a.bottom-a.top];ut.assertIsValidBox({x:i,y:o,width:l,height:c},"Box.constructor",n),this._x=i,this._y=o,this._width=l,this._height=c}get x(){return this._x}get y(){return this._y}get width(){return this._width}get height(){return this._height}get left(){return this.x}get top(){return this.y}get right(){return this.x+this.width}get bottom(){return this.y+this.height}get area(){return this.width*this.height}get topLeft(){return new De(this.left,this.top)}get topRight(){return new De(this.right,this.top)}get bottomLeft(){return new De(this.left,this.bottom)}get bottomRight(){return new De(this.right,this.bottom)}round(){let[t,n,a,r]=[this.x,this.y,this.width,this.height].map(s=>Math.round(s));return new ut({x:t,y:n,width:a,height:r})}floor(){let[t,n,a,r]=[this.x,this.y,this.width,this.height].map(s=>Math.floor(s));return new ut({x:t,y:n,width:a,height:r})}toSquare(){let{x:t,y:n,width:a,height:r}=this,s=Math.abs(a-r);return a<r&&(t-=s/2,a+=s),r<a&&(n-=s/2,r+=s),new ut({x:t,y:n,width:a,height:r})}rescale(t){let n=ef(t)?t.width:t,a=ef(t)?t.height:t;return new ut({x:this.x*n,y:this.y*a,width:this.width*n,height:this.height*a})}pad(t,n){let[a,r,s,i]=[this.x-t/2,this.y-n/2,this.width+t,this.height+n];return new ut({x:a,y:r,width:s,height:i})}clipAtImageBorders(t,n){let{x:a,y:r,right:s,bottom:i}=this,o=Math.max(a,0),l=Math.max(r,0),c=s-o,u=i-l,p=Math.min(c,t-o),d=Math.min(u,n-l);return new ut({x:o,y:l,width:p,height:d}).floor()}shift(t,n){let{width:a,height:r}=this,s=this.x+t,i=this.y+n;return new ut({x:s,y:i,width:a,height:r})}padAtBorders(t,n){let a=this.width+1,r=this.height+1,s=1,i=1,o=a,l=r,c=this.left,u=this.top,p=this.right,d=this.bottom;return p>n&&(o=-p+n+a,p=n),d>t&&(l=-d+t+r,d=t),c<1&&(l=2-c,c=1),u<1&&(l=2-u,u=1),{dy:i,edy:l,dx:s,edx:o,y:u,ey:d,x:c,ex:p,w:a,h:r}}calibrate(t){return new ut({left:this.left+t.left*this.width,top:this.top+t.top*this.height,right:this.right+t.right*this.width,bottom:this.bottom+t.bottom*this.height}).toSquare().round()}};var bu=class extends ut{constructor(t,n,a,r,s=!1){super({left:t,top:n,right:a,bottom:r},s)}};var ds=class{constructor(t,n,a,r,s){this._imageDims=new gn(s.width,s.height),this._score=t,this._classScore=n,this._className=a,this._box=new ut(r).rescale(this._imageDims)}get score(){return this._score}get classScore(){return this._classScore}get className(){return this._className}get box(){return this._box}get imageDims(){return this._imageDims}get imageWidth(){return this.imageDims.width}get imageHeight(){return this.imageDims.height}get relativeBox(){return new ut(this._box).rescale(this.imageDims.reverse())}forSize(t,n){return new ds(this.score,this.classScore,this.className,this.relativeBox,{width:t,height:n})}};var bt=class extends ds{constructor(t,n,a){super(t,t,"",n,a)}forSize(t,n){let{score:a,relativeBox:r,imageDims:s}=super.forSize(t,n);return new bt(a,r,s)}};function ew(e,t,n=!0){let a=Math.max(0,Math.min(e.right,t.right)-Math.max(e.left,t.left)),r=Math.max(0,Math.min(e.bottom,t.bottom)-Math.max(e.top,t.top)),s=a*r;return n?s/(e.area+t.area-s):s/Math.min(e.area,t.area)}function tw(e){let t=e.map(o=>o.x),n=e.map(o=>o.y),a=t.reduce((o,l)=>l<o?l:o,Infinity),r=n.reduce((o,l)=>l<o?l:o,Infinity),s=t.reduce((o,l)=>o<l?l:o,0),i=n.reduce((o,l)=>o<l?l:o,0);return new bu(a,r,s,i)}function nw(e,t,n,a=!0){let r=t.map((i,o)=>({score:i,boxIndex:o})).sort((i,o)=>i.score-o.score).map(i=>i.boxIndex),s=[];for(;r.length>0;){let i=r.pop();s.push(i);let o=r,l=[];for(let c=0;c<o.length;c++){let u=o[c],p=e[i],d=e[u];l.push(ew(p,d,a))}r=r.filter((c,u)=>l[u]<=n)}return s}function Oa(e,t){return D(()=>{let[n,a,r]=t,s=wn([...e.shape.slice(0,3),1],n,"float32"),i=wn([...e.shape.slice(0,3),1],a,"float32"),o=wn([...e.shape.slice(0,3),1],r,"float32"),l=Qe([s,i,o],3);return fe(e,l)})}function aw(e,t=!1){return D(()=>{let[n,a]=e.shape.slice(1);if(n===a)return e;let r=Math.abs(n-a),s=Math.round(r*(t?.5:1)),i=n>a?2:1,o=d=>{let h=e.shape.slice();return h[i]=d,wn(h,0,"float32")},l=o(s),c=r-l.shape[i],p=[t&&c?o(c):null,e,l].filter(d=>!!d).map(d=>ce(d,"float32"));return Qe(p,i)})}function rre(e){let t=e.slice();for(let n=t.length-1;n>0;n--){let a=Math.floor(Math.random()*(n+1)),r=t[n];t[n]=t[a],t[a]=r}return t}function Ep(e){return 1/(1+Math.exp(-e))}function sre(e){return Math.log(e/(1-e))}var xu=class extends ut{constructor(t,n,a,r,s=!1){super({x:t,y:n,width:a,height:r},s)}};var ire=.5,ore=.43,lre=.45,ta=class{constructor(t,n,a=new De(0,0)){let{width:r,height:s}=n;this._imgDims=new gn(r,s),this._shift=a,this._positions=t.map(i=>i.mul(new De(r,s)).add(a))}get shift(){return new De(this._shift.x,this._shift.y)}get imageWidth(){return this._imgDims.width}get imageHeight(){return this._imgDims.height}get positions(){return this._positions}get relativePositions(){return this._positions.map(t=>t.sub(this._shift).div(new De(this.imageWidth,this.imageHeight)))}forSize(t,n){return new this.constructor(this.relativePositions,{width:t,height:n})}shiftBy(t,n){return new this.constructor(this.relativePositions,this._imgDims,new De(t,n))}shiftByPoint(t){return this.shiftBy(t.x,t.y)}align(t,n={}){if(t){let s=t instanceof bt?t.box.floor():new ut(t);return this.shiftBy(s.x,s.y).align(null,n)}let{useDlibAlignment:a,minBoxPadding:r}={useDlibAlignment:!1,minBoxPadding:.2,...n};return a?this.alignDlib():this.alignMinBbox(r)}alignDlib(){let t=this.getRefPointsForAlignment(),[n,a,r]=t,s=p=>r.sub(p).magnitude(),i=(s(n)+s(a))/2,o=Math.floor(i/lre),l=Xi(t),c=Math.floor(Math.max(0,l.x-ire*o)),u=Math.floor(Math.max(0,l.y-ore*o));return new xu(c,u,Math.min(o,this.imageWidth+c),Math.min(o,this.imageHeight+u))}alignMinBbox(t){let n=tw(this.positions);return n.pad(n.width*t,n.height*t)}getRefPointsForAlignment(){throw new Error("getRefPointsForAlignment not implemented by base class")}};var B2=class extends ta{getRefPointsForAlignment(){let t=this.positions;return[t[0],t[1],Xi([t[3],t[4]])]}};var vu=class extends ta{getJawOutline(){return this.positions.slice(0,17)}getLeftEyeBrow(){return this.positions.slice(17,22)}getRightEyeBrow(){return this.positions.slice(22,27)}getNose(){return this.positions.slice(27,36)}getLeftEye(){return this.positions.slice(36,42)}getRightEye(){return this.positions.slice(42,48)}getMouth(){return this.positions.slice(48,68)}getRefPointsForAlignment(){return[this.getLeftEye(),this.getRightEye(),this.getMouth()].map(Xi)}};var Ap=class{constructor(t,n){this._label=t,this._distance=n}get label(){return this._label}get distance(){return this._distance}toString(t=!0){return`${this.label}${t?` (${Ki(this.distance)})`:""}`}};var Fp=class extends ut{static assertIsValidLabeledBox(t,n){if(ut.assertIsValidBox(t,n),!Pa(t.label))throw new Error(`${n} - expected property label (${t.label}) to be a number`)}constructor(t,n){super(t);this._label=n}get label(){return this._label}};var Cr=class{constructor(t,n){if(typeof t!="string")throw new Error("LabeledFaceDescriptors - constructor expected label to be a string");if(!Array.isArray(n)||n.some(a=>!(a instanceof Float32Array)))throw new Error("LabeledFaceDescriptors - constructor expected descriptors to be an array of Float32Array");this._label=t,this._descriptors=n}get label(){return this._label}get descriptors(){return this._descriptors}toJSON(){return{label:this.label,descriptors:this.descriptors.map(t=>Array.from(t))}}static fromJSON(t){let n=t.descriptors.map(a=>new Float32Array(a));return new Cr(t.label,n)}};var W2=class extends Fp{static assertIsValidPredictedBox(t,n){if(Fp.assertIsValidLabeledBox(t,n),!yu(t.score)||!yu(t.classScore))throw new Error(`${n} - expected properties score (${t.score}) and (${t.classScore}) to be a number between [0, 1]`)}constructor(t,n,a,r){super(t,n);this._score=a,this._classScore=r}get score(){return this._score}get classScore(){return this._classScore}};function ir(e){return e.detection instanceof bt}function Yi(e,t){return{...e,...{detection:t}}}function rw(){let e=window.fetch;if(!e)throw new Error("fetch - missing fetch implementation for browser environment");let t=()=>{throw new Error("readFile - filesystem not available for browser environment")};return{Canvas:HTMLCanvasElement,CanvasRenderingContext2D,Image:HTMLImageElement,ImageData,Video:HTMLVideoElement,createCanvasElement:()=>document.createElement("canvas"),createImageElement:()=>document.createElement("img"),fetch:e,readFile:t}}function tf(e){let t="";if(!e)try{e=require("fs")}catch(a){t=a.toString()}return{readFile:e?a=>new Promise((r,s)=>{e.readFile(a,(i,o)=>i?s(i):r(o))}):()=>{throw new Error(`readFile - failed to require fs in nodejs environment with error: ${t}`)}}}function sw(){let e=global.Canvas||global.HTMLCanvasElement,t=global.Image||global.HTMLImageElement,n=()=>{if(e)return new e;throw new Error("createCanvasElement - missing Canvas implementation for nodejs environment")},a=()=>{if(t)return new t;throw new Error("createImageElement - missing Image implementation for nodejs environment")},r=global.fetch,s=tf();return{Canvas:e||class{},CanvasRenderingContext2D:global.CanvasRenderingContext2D||class{},Image:t||class{},ImageData:global.ImageData||class{},Video:global.HTMLVideoElement||class{},createCanvasElement:n,createImageElement:a,fetch:r,...s}}function iw(){return typeof window=="object"&&typeof document!="undefined"&&typeof HTMLImageElement!="undefined"&&typeof HTMLCanvasElement!="undefined"&&typeof HTMLVideoElement!="undefined"&&typeof ImageData!="undefined"&&typeof CanvasRenderingContext2D!="undefined"}var ow=J_(U2()),Jt;function pre(){if(!Jt)throw new Error("getEnv - environment is not defined, check isNodejs() and isBrowser()");return Jt}function lw(e){Jt=e}function uw(){return iw()?lw(rw()):ow.isNodejs()?lw(sw()):null}function dre(e){if(Jt||uw(),!Jt)throw new Error("monkeyPatch - environment is not defined, check isNodejs() and isBrowser()");let{Canvas:t=Jt.Canvas,Image:n=Jt.Image}=e;Jt.Canvas=t,Jt.Image=n,Jt.createCanvasElement=e.createCanvasElement||(()=>new t),Jt.createImageElement=e.createImageElement||(()=>new n),Jt.ImageData=e.ImageData||Jt.ImageData,Jt.Video=e.Video||Jt.Video,Jt.fetch=e.fetch||Jt.fetch,Jt.readFile=e.readFile||Jt.readFile}var st={getEnv:pre,setEnv:lw,initialize:uw,createBrowserEnv:rw,createFileSystem:tf,createNodejsEnv:sw,monkeyPatch:dre,isBrowser:iw,isNodejs:ow.isNodejs};uw();function Zi(e){return!st.isNodejs()&&typeof e=="string"?document.getElementById(e):e}function Sn(e){let{Canvas:t,CanvasRenderingContext2D:n}=st.getEnv();if(e instanceof n)return e;let a=Zi(e);if(!(a instanceof t))throw new Error("resolveContext2d - expected canvas to be of instance of Canvas");let r=a.getContext("2d");if(!r)throw new Error("resolveContext2d - canvas 2d context is null");return r}var or;(function(e){e.TOP_LEFT="TOP_LEFT",e.TOP_RIGHT="TOP_RIGHT",e.BOTTOM_LEFT="BOTTOM_LEFT",e.BOTTOM_RIGHT="BOTTOM_RIGHT"})(or||(or={}));var $p=class{constructor(t={}){let{anchorPosition:n,backgroundColor:a,fontColor:r,fontSize:s,fontStyle:i,padding:o}=t;this.anchorPosition=n||or.TOP_LEFT,this.backgroundColor=a||"rgba(0, 0, 0, 0.5)",this.fontColor=r||"rgba(255, 255, 255, 1)",this.fontSize=s||14,this.fontStyle=i||"Georgia",this.padding=o||4}},hs=class{constructor(t,n,a={}){this.text=typeof t=="string"?[t]:t instanceof hs?t.text:t,this.anchor=n,this.options=new $p(a)}measureWidth(t){let{padding:n}=this.options;return this.text.map(a=>t.measureText(a).width).reduce((a,r)=>a<r?r:a,0)+2*n}measureHeight(){let{fontSize:t,padding:n}=this.options;return this.text.length*t+2*n}getUpperLeft(t,n){let{anchorPosition:a}=this.options,r=a===or.BOTTOM_RIGHT||a===or.TOP_RIGHT,s=a===or.BOTTOM_LEFT||a===or.BOTTOM_RIGHT,i=this.measureWidth(t),o=this.measureHeight(),l=r?this.anchor.x-i:this.anchor.x,c=s?this.anchor.y-o:this.anchor.y;if(n){let{width:u,height:p}=n,d=Math.max(Math.min(l,u-i),0),h=Math.max(Math.min(c,p-o),0);return{x:d,y:h}}return{x:l,y:c}}draw(t){let n=Zi(t),a=Sn(n),{backgroundColor:r,fontColor:s,fontSize:i,fontStyle:o,padding:l}=this.options;a.font=`${i}px ${o}`;let c=this.measureWidth(a),u=this.measureHeight();a.fillStyle=r;let p=this.getUpperLeft(a,n);a.fillRect(p.x,p.y,c,u),a.fillStyle=s,this.text.forEach((d,h)=>{let m=l+p.x,f=l+p.y+(h+1)*i;a.fillText(d,m,f)})}};var cw=class{constructor(t={}){let{boxColor:n,lineWidth:a,label:r,drawLabelOptions:s}=t;this.boxColor=n||"rgba(0, 0, 255, 1)",this.lineWidth=a||2,this.label=r;let i={anchorPosition:or.BOTTOM_LEFT,backgroundColor:this.boxColor};this.drawLabelOptions=new $p({...i,...s})}},nf=class{constructor(t,n={}){this.box=new ut(t),this.options=new cw(n)}draw(t){let n=Sn(t),{boxColor:a,lineWidth:r}=this.options,{x:s,y:i,width:o,height:l}=this.box;n.strokeStyle=a,n.lineWidth=r,n.strokeRect(s,i,o,l);let{label:c}=this.options;c&&new hs([c],{x:s-r/2,y:i},this.options.drawLabelOptions).draw(t)}};function hre(e,t){(Array.isArray(t)?t:[t]).forEach(a=>{let r=a instanceof bt?a.score:ir(a)?a.detection.score:void 0,s=a instanceof bt?a.box:ir(a)?a.detection.box:new ut(a),i=r?`${Ki(r)}`:void 0;new nf(s,{label:i}).draw(e)})}function Dp(e){let{Image:t,Video:n}=st.getEnv();return e instanceof t&&e.complete||e instanceof n&&e.readyState>=3}function pw(e){return new Promise((t,n)=>{if(e instanceof st.getEnv().Canvas||Dp(e))return t(null);function a(s){!s.currentTarget||(s.currentTarget.removeEventListener("load",r),s.currentTarget.removeEventListener("error",a),n(s))}function r(s){!s.currentTarget||(s.currentTarget.removeEventListener("load",r),s.currentTarget.removeEventListener("error",a),t(s))}e.addEventListener("load",r),e.addEventListener("error",a)})}function dw(e){return new Promise((t,n)=>{e instanceof Blob||n(new Error("bufferToImage - expected buf to be of type: Blob"));let a=new FileReader;a.onload=()=>{typeof a.result!="string"&&n(new Error("bufferToImage - expected reader.result to be a string, in onload"));let r=st.getEnv().createImageElement();r.onload=()=>t(r),r.onerror=n,r.src=a.result},a.onerror=n,a.readAsDataURL(e)})}function Ji(e){let{Image:t,Video:n}=st.getEnv();return e instanceof t?new gn(e.naturalWidth,e.naturalHeight):e instanceof n?new gn(e.videoWidth,e.videoHeight):new gn(e.width,e.height)}function wu({width:e,height:t}){let{createCanvasElement:n}=st.getEnv(),a=n();return a.width=e,a.height=t,a}function Rp(e,t){let{ImageData:n}=st.getEnv();if(!(e instanceof n)&&!Dp(e))throw new Error("createCanvasFromMedia - media has not finished loading yet");let{width:a,height:r}=t||Ji(e),s=wu({width:a,height:r});return e instanceof n?Sn(s).putImageData(e,0,0):Sn(s).drawImage(e,0,0,a,r),s}async function hw(e,t){let n=t||st.getEnv().createCanvasElement(),[a,r,s]=e.shape.slice(ea(e)?1:0),i=D(()=>e.as3D(a,r,s).toInt());return await ki.toPixels(i,n),i.dispose(),n}function af(e){let{Image:t,Canvas:n,Video:a}=st.getEnv();return e instanceof t||e instanceof n||e instanceof a}function mw(e,t,n=!1){let{Image:a,Canvas:r}=st.getEnv();if(!(e instanceof a||e instanceof r))throw new Error("imageToSquare - expected arg0 to be HTMLImageElement | HTMLCanvasElement");let s=Ji(e),i=t/Math.max(s.height,s.width),o=i*s.width,l=i*s.height,c=wu({width:t,height:t}),u=e instanceof r?e:Rp(e),p=Math.abs(o-l)/2,d=n&&o<l?p:0,h=n&&l<o?p:0;return Sn(c).drawImage(u,d,h,o,l),c}var _r=class{constructor(t,n=!1){this._imageTensors=[];this._canvases=[];this._treatAsBatchInput=!1;this._inputDimensions=[];if(!Array.isArray(t))throw new Error(`NetInput.constructor - expected inputs to be an Array of TResolvedNetInput or to be instanceof tf.Tensor4D, instead have ${t}`);this._treatAsBatchInput=n,this._batchSize=t.length,t.forEach((a,r)=>{if(Sr(a)){this._imageTensors[r]=a,this._inputDimensions[r]=a.shape;return}if(ea(a)){let i=a.shape[0];if(i!==1)throw new Error(`NetInput - tf.Tensor4D with batchSize ${i} passed, but not supported in input array`);this._imageTensors[r]=a,this._inputDimensions[r]=a.shape.slice(1);return}let s=a instanceof st.getEnv().Canvas?a:Rp(a);this._canvases[r]=s,this._inputDimensions[r]=[s.height,s.width,3]})}get imageTensors(){return this._imageTensors}get canvases(){return this._canvases}get isBatchInput(){return this.batchSize>1||this._treatAsBatchInput}get batchSize(){return this._batchSize}get inputDimensions(){return this._inputDimensions}get inputSize(){return this._inputSize}get reshapedInputDimensions(){return sr(this.batchSize,0,1).map((t,n)=>this.getReshapedInputDimensions(n))}getInput(t){return this.canvases[t]||this.imageTensors[t]}getInputDimensions(t){return this._inputDimensions[t]}getInputHeight(t){return this._inputDimensions[t][0]}getInputWidth(t){return this._inputDimensions[t][1]}getReshapedInputDimensions(t){if(typeof this.inputSize!="number")throw new Error("getReshapedInputDimensions - inputSize not set, toBatchTensor has not been called yet");let n=this.getInputWidth(t),a=this.getInputHeight(t);return Qv({width:n,height:a},this.inputSize)}toBatchTensor(t,n=!0){return this._inputSize=t,D(()=>{let a=sr(this.batchSize,0,1).map(s=>{let i=this.getInput(s);if(i instanceof Ae){let o=ea(i)?i:i.expandDims();return o=aw(o,n),(o.shape[1]!==t||o.shape[2]!==t)&&(o=Ya.resizeBilinear(o,[t,t])),o.as3D(t,t,3)}if(i instanceof st.getEnv().Canvas)return ki.fromPixels(mw(i,t,n));throw new Error(`toBatchTensor - at batchIdx ${s}, expected input to be instanceof tf.Tensor or instanceof HTMLCanvasElement, instead have ${i}`)});return $t(a.map(s=>ce(s,"float32"))).as4D(this.batchSize,t,t,3)})}};async function gt(e){if(e instanceof _r)return e;let t=Array.isArray(e)?e:[e];if(!t.length)throw new Error("toNetInput - empty array passed as input");let n=r=>Array.isArray(e)?` at input index ${r}:`:"",a=t.map(Zi);return a.forEach((r,s)=>{if(!af(r)&&!Sr(r)&&!ea(r))throw typeof t[s]=="string"?new Error(`toNetInput -${n(s)} string passed, but could not resolve HTMLElement for element id ${t[s]}`):new Error(`toNetInput -${n(s)} expected media to be of type HTMLImageElement | HTMLVideoElement | HTMLCanvasElement | tf.Tensor3D, or to be an element id`);if(ea(r)){let i=r.shape[0];if(i!==1)throw new Error(`toNetInput -${n(s)} tf.Tensor4D with batchSize ${i} passed, but not supported in input array`)}}),await Promise.all(a.map(r=>af(r)&&pw(r))),new _r(a,Array.isArray(e))}async function ku(e,t){let{Canvas:n}=st.getEnv(),a=e;if(!(e instanceof n)){let i=await gt(e);if(i.batchSize>1)throw new Error("extractFaces - batchSize > 1 not supported");let o=i.getInput(0);a=o instanceof n?o:await hw(o)}let r=Sn(a);return t.map(i=>i instanceof bt?i.forSize(a.width,a.height).box.floor():i).map(i=>i.clipAtImageBorders(a.width,a.height)).map(({x:i,y:o,width:l,height:c})=>{let u=wu({width:l,height:c});return Sn(u).putImageData(r.getImageData(i,o,l,c),0,0),u})}async function Iu(e,t){if(!Sr(e)&&!ea(e))throw new Error("extractFaceTensors - expected image tensor to be 3D or 4D");if(ea(e)&&e.shape[0]>1)throw new Error("extractFaceTensors - batchSize > 1 not supported");return D(()=>{let[n,a,r]=e.shape.slice(ea(e)?1:0);return t.map(o=>o instanceof bt?o.forSize(a,n).box:o).map(o=>o.clipAtImageBorders(a,n)).map(({x:o,y:l,width:c,height:u})=>Vl(e.as3D(n,a,r),[l,o,0],[u,c,r]))})}async function Qi(e,t){let{fetch:n}=st.getEnv(),a=await n(e,t);if(!(a.status<400))throw new Error(`failed to fetch: (${a.status}) ${a.statusText}, from url: ${a.url}`);return a}async function mre(e){let t=await Qi(e),n=await t.blob();if(!n.type.startsWith("image/"))throw new Error(`fetchImage - expected blob type to be of type image/*, instead have: ${n.type}, for url: ${t.url}`);return dw(n)}async function fw(e){return(await Qi(e)).json()}async function fre(e){return new Float32Array(await(await Qi(e)).arrayBuffer())}function rf(e,t){let n=`${t}-weights_manifest.json`;if(!e)return{modelBaseUri:"",manifestUri:n};if(e==="/")return{modelBaseUri:"/",manifestUri:`/${n}`};let a=e.startsWith("http://")?"http://":e.startsWith("https://")?"https://":"";e=e.replace(a,"");let r=e.split("/").filter(o=>o),s=e.endsWith(".json")?r[r.length-1]:n,i=a+(e.endsWith(".json")?r.slice(0,r.length-1):r).join("/");return i=e.startsWith("/")?`/${i}`:i,{modelBaseUri:i,manifestUri:i==="/"?`/${s}`:`${i}/${s}`}}async function gw(e,t){let{manifestUri:n,modelBaseUri:a}=rf(e,t),r=await fw(n);return Ht.loadWeights(r,a)}function gre(e,t,n=!1){let{width:a,height:r}=n?Ji(t):t;return e.width=a,e.height=r,{width:a,height:r}}var an=class{constructor(t){this._params=void 0;this._paramMappings=[];this._name=t}get params(){return this._params}get paramMappings(){return this._paramMappings}get isLoaded(){return!!this.params}getParamFromPath(t){let{obj:n,objProp:a}=this.traversePropertyPath(t);return n[a]}reassignParamFromPath(t,n){let{obj:a,objProp:r}=this.traversePropertyPath(t);a[r].dispose(),a[r]=n}getParamList(){return this._paramMappings.map(({paramPath:t})=>({path:t,tensor:this.getParamFromPath(t)}))}getTrainableParams(){return this.getParamList().filter(t=>t.tensor instanceof Vr)}getFrozenParams(){return this.getParamList().filter(t=>!(t.tensor instanceof Vr))}variable(){this.getFrozenParams().forEach(({path:t,tensor:n})=>{this.reassignParamFromPath(t,n.variable())})}freeze(){this.getTrainableParams().forEach(({path:t,tensor:n})=>{let a=qn(n.dataSync());n.dispose(),this.reassignParamFromPath(t,a)})}dispose(t=!0){this.getParamList().forEach(n=>{if(t&&n.tensor.isDisposed)throw new Error(`param tensor has already been disposed for path ${n.path}`);n.tensor.dispose()}),this._params=void 0}serializeParams(){return new Float32Array(this.getParamList().map(({tensor:t})=>Array.from(t.dataSync())).reduce((t,n)=>t.concat(n)))}async load(t){if(t instanceof Float32Array){this.extractWeights(t);return}await this.loadFromUri(t)}async loadFromUri(t){if(t&&typeof t!="string")throw new Error(`${this._name}.loadFromUri - expected model uri`);let n=await gw(t,this.getDefaultModelName());this.loadFromWeightMap(n)}async loadFromDisk(t){if(t&&typeof t!="string")throw new Error(`${this._name}.loadFromDisk - expected model file path`);let{readFile:n}=st.getEnv(),{manifestUri:a,modelBaseUri:r}=rf(t,this.getDefaultModelName()),s=c=>Promise.all(c.map(u=>n(u).then(p=>p.buffer))),i=Ht.weightsLoaderFactory(s),o=JSON.parse((await n(a)).toString()),l=await i(o,r);this.loadFromWeightMap(l)}loadFromWeightMap(t){let{paramMappings:n,params:a}=this.extractParamsFromWeightMap(t);this._paramMappings=n,this._params=a}extractWeights(t){let{paramMappings:n,params:a}=this.extractParams(t);this._paramMappings=n,this._params=a}traversePropertyPath(t){if(!this.params)throw new Error("traversePropertyPath - model has no loaded params");let n=t.split("/").reduce((s,i)=>{if(!s.nextObj.hasOwnProperty(i))throw new Error(`traversePropertyPath - object does not have property ${i}, for path ${t}`);return{obj:s.nextObj,objProp:i,nextObj:s.nextObj[i]}},{nextObj:this.params}),{obj:a,objProp:r}=n;if(!a||!r||!(a[r]instanceof Ae))throw new Error(`traversePropertyPath - parameter is not a tensor, for path ${t}`);return{obj:a,objProp:r}}};function Cn(e,t,n){return D(()=>{let a=Ei(e,t.depthwise_filter,t.pointwise_filter,n,"same");return a=Z(a,t.bias),a})}function sf(e,t,n=!1){return D(()=>{let a=Ye(n?Z(At(e,t.conv0.filters,[2,2],"same"),t.conv0.bias):Cn(e,t.conv0,[2,2])),r=Cn(a,t.conv1,[1,1]),s=Ye(Z(a,r)),i=Cn(s,t.conv2,[1,1]);return Ye(Z(a,Z(r,i)))})}function Mp(e,t,n=!1,a=!0){return D(()=>{let r=Ye(n?Z(At(e,t.conv0.filters,a?[2,2]:[1,1],"same"),t.conv0.bias):Cn(e,t.conv0,a?[2,2]:[1,1])),s=Cn(r,t.conv1,[1,1]),i=Ye(Z(r,s)),o=Cn(i,t.conv2,[1,1]),l=Ye(Z(r,Z(s,o))),c=Cn(l,t.conv3,[1,1]);return Ye(Z(r,Z(s,Z(o,c))))})}function eo(e,t,n="same",a=!1){return D(()=>{let r=Z(At(e,t.filters,[1,1],n),t.bias);return a?Ye(r):r})}function yn(e,t){Object.keys(e).forEach(n=>{t.some(a=>a.originalPath===n)||e[n].dispose()})}function Nu(e,t){return(n,a,r,s)=>{let i=Sa(e(n*a*r*r),[r,r,n,a]),o=tt(e(a));return t.push({paramPath:`${s}/filters`},{paramPath:`${s}/bias`}),{filters:i,bias:o}}}function of(e,t){return(n,a,r)=>{let s=Ta(e(n*a),[n,a]),i=tt(e(a));return t.push({paramPath:`${r}/weights`},{paramPath:`${r}/bias`}),{weights:s,bias:i}}}var lf=class{constructor(t,n,a){this.depthwise_filter=t;this.pointwise_filter=n;this.bias=a}};function Tu(e,t){return(n,a,r)=>{let s=Sa(e(3*3*n),[3,3,n,1]),i=Sa(e(n*a),[1,1,n,a]),o=tt(e(a));return t.push({paramPath:`${r}/depthwise_filter`},{paramPath:`${r}/pointwise_filter`},{paramPath:`${r}/bias`}),new lf(s,i,o)}}function Su(e){return t=>{let n=e(`${t}/depthwise_filter`,4),a=e(`${t}/pointwise_filter`,4),r=e(`${t}/bias`,1);return new lf(n,a,r)}}function Un(e,t){return(n,a,r)=>{let s=e[n];if(!qi(s,a))throw new Error(`expected weightMap[${n}] to be a Tensor${a}D, instead have ${s}`);return t.push({originalPath:n,paramPath:r||n}),s}}function bn(e){let t=e;function n(r){let s=t.slice(0,r);return t=t.slice(r),s}function a(){return t}return{extractWeights:n,getRemainingWeights:a}}function uf(e,t){let n=Nu(e,t),a=Tu(e,t);function r(i,o,l,c=!1){let u=c?n(i,o,3,`${l}/conv0`):a(i,o,`${l}/conv0`),p=a(o,o,`${l}/conv1`),d=a(o,o,`${l}/conv2`);return{conv0:u,conv1:p,conv2:d}}function s(i,o,l,c=!1){let{conv0:u,conv1:p,conv2:d}=r(i,o,l,c),h=a(o,o,`${l}/conv3`);return{conv0:u,conv1:p,conv2:d,conv3:h}}return{extractDenseBlock3Params:r,extractDenseBlock4Params:s}}function G2(e){let t=[],{extractWeights:n,getRemainingWeights:a}=bn(e),{extractDenseBlock4Params:r}=uf(n,t),s=r(3,32,"dense0",!0),i=r(32,64,"dense1"),o=r(64,128,"dense2"),l=r(128,256,"dense3");if(a().length!==0)throw new Error(`weights remaing after extract: ${a().length}`);return{paramMappings:t,params:{dense0:s,dense1:i,dense2:o,dense3:l}}}function cf(e){return t=>{let n=e(`${t}/filters`,4),a=e(`${t}/bias`,1);return{filters:n,bias:a}}}function pf(e,t){let n=Un(e,t),a=cf(n),r=Su(n);function s(o,l=!1){let c=l?a(`${o}/conv0`):r(`${o}/conv0`),u=r(`${o}/conv1`),p=r(`${o}/conv2`);return{conv0:c,conv1:u,conv2:p}}function i(o,l=!1){let c=l?a(`${o}/conv0`):r(`${o}/conv0`),u=r(`${o}/conv1`),p=r(`${o}/conv2`),d=r(`${o}/conv3`);return{conv0:c,conv1:u,conv2:p,conv3:d}}return{extractDenseBlock3Params:s,extractDenseBlock4Params:i}}function H2(e){let t=[],{extractDenseBlock4Params:n}=pf(e,t),a={dense0:n("dense0",!0),dense1:n("dense1"),dense2:n("dense2"),dense3:n("dense3")};return yn(e,t),{params:a,paramMappings:t}}var Pp=class extends an{constructor(){super("FaceFeatureExtractor")}forwardInput(t){let{params:n}=this;if(!n)throw new Error("FaceFeatureExtractor - load model before inference");return D(()=>{let a=ce(t.toBatchTensor(112,!0),"float32"),s=Oa(a,[122.782,117.001,104.298]).div(de(255)),i=Mp(s,n.dense0,!0);return i=Mp(i,n.dense1),i=Mp(i,n.dense2),i=Mp(i,n.dense3),i=Xn(i,[7,7],[2,2],"valid"),i})}async forward(t){return this.forwardInput(await gt(t))}getDefaultModelName(){return"face_feature_extractor_model"}extractParamsFromWeightMap(t){return H2(t)}extractParams(t){return G2(t)}};function Op(e,t){return D(()=>Z(ze(e,t.weights),t.bias))}function j2(e,t,n){let a=[],{extractWeights:r,getRemainingWeights:s}=bn(e),o=of(r,a)(t,n,"fc");if(s().length!==0)throw new Error(`weights remaing after extract: ${s().length}`);return{paramMappings:a,params:{fc:o}}}function q2(e){let t=[],n=Un(e,t);function a(s){let i=n(`${s}/weights`,2),o=n(`${s}/bias`,1);return{weights:i,bias:o}}let r={fc:a("fc")};return yn(e,t),{params:r,paramMappings:t}}function df(e){let t={},n={};return Object.keys(e).forEach(a=>{let r=a.startsWith("fc")?n:t;r[a]=e[a]}),{featureExtractorMap:t,classifierMap:n}}var Lp=class extends an{constructor(t,n){super(t);this._faceFeatureExtractor=n}get faceFeatureExtractor(){return this._faceFeatureExtractor}runNet(t){let{params:n}=this;if(!n)throw new Error(`${this._name} - load model before inference`);return D(()=>{let a=t instanceof _r?this.faceFeatureExtractor.forwardInput(t):t;return Op(a.as2D(a.shape[0],-1),n.fc)})}dispose(t=!0){this.faceFeatureExtractor.dispose(t),super.dispose(t)}loadClassifierParams(t){let{params:n,paramMappings:a}=this.extractClassifierParams(t);this._params=n,this._paramMappings=a}extractClassifierParams(t){return j2(t,this.getClassifierChannelsIn(),this.getClassifierChannelsOut())}extractParamsFromWeightMap(t){let{featureExtractorMap:n,classifierMap:a}=df(t);return this.faceFeatureExtractor.loadFromWeightMap(n),q2(a)}extractParams(t){let n=this.getClassifierChannelsIn(),a=this.getClassifierChannelsOut(),r=a*n+a,s=t.slice(0,t.length-r),i=t.slice(t.length-r);return this.faceFeatureExtractor.extractWeights(s),this.extractClassifierParams(i)}};var yw=["neutral","happy","sad","angry","fearful","disgusted","surprised"],ms=class{constructor(t){if(t.length!==7)throw new Error(`FaceExpressions.constructor - expected probabilities.length to be 7, have: ${t.length}`);yw.forEach((n,a)=>{this[n]=t[a]})}asSortedArray(){return yw.map(t=>({expression:t,probability:this[t]})).sort((t,n)=>n.probability-t.probability)}};var hf=class extends Lp{constructor(t=new Pp){super("FaceExpressionNet",t)}forwardInput(t){return D(()=>Na(this.runNet(t)))}async forward(t){return this.forwardInput(await gt(t))}async predictExpressions(t){let n=await gt(t),a=await this.forwardInput(n),r=await Promise.all(dt(a).map(async i=>{let o=await i.data();return i.dispose(),o}));a.dispose();let s=r.map(i=>new ms(i));return n.isBatchInput?s:s[0]}getDefaultModelName(){return"face_expression_model"}getClassifierChannelsIn(){return 256}getClassifierChannelsOut(){return 7}};function bw(e){return e.expressions instanceof ms}function mf(e,t){return{...e,...{expressions:t}}}function yre(e,t,n=.1,a){(Array.isArray(t)?t:[t]).forEach(s=>{let i=s instanceof ms?s:bw(s)?s.expressions:void 0;if(!i)throw new Error("drawFaceExpressions - expected faceExpressions to be FaceExpressions | WithFaceExpressions<{}> or array thereof");let l=i.asSortedArray().filter(p=>p.probability>n),c=ir(s)?s.detection.box.bottomLeft:a||new De(0,0);new hs(l.map(p=>`${p.expression} (${Ki(p.probability)})`),c).draw(e)})}function to(e){return ir(e)&&e.landmarks instanceof ta&&e.unshiftedLandmarks instanceof ta&&e.alignedRect instanceof bt}function Cu(e,t){let{box:n}=e.detection,a=t.shiftBy(n.x,n.y),r=a.align(),{imageDims:s}=e.detection,i=new bt(e.detection.score,r.rescale(s.reverse()),s);return{...e,...{landmarks:a,unshiftedLandmarks:t,alignedRect:i}}}var xw=class{constructor(t={}){let{drawLines:n=!0,drawPoints:a=!0,lineWidth:r,lineColor:s,pointSize:i,pointColor:o}=t;this.drawLines=n,this.drawPoints=a,this.lineWidth=r||1,this.pointSize=i||2,this.lineColor=s||"rgba(0, 255, 255, 1)",this.pointColor=o||"rgba(255, 0, 255, 1)"}},vw=class{constructor(t,n={}){this.faceLandmarks=t,this.options=new xw(n)}draw(t){let n=Sn(t),{drawLines:a,drawPoints:r,lineWidth:s,lineColor:i,pointSize:o,pointColor:l}=this.options;if(a&&this.faceLandmarks instanceof vu&&(n.strokeStyle=i,n.lineWidth=s,Tr(n,this.faceLandmarks.getJawOutline()),Tr(n,this.faceLandmarks.getLeftEyeBrow()),Tr(n,this.faceLandmarks.getRightEyeBrow()),Tr(n,this.faceLandmarks.getNose()),Tr(n,this.faceLandmarks.getLeftEye(),!0),Tr(n,this.faceLandmarks.getRightEye(),!0),Tr(n,this.faceLandmarks.getMouth(),!0)),r){n.strokeStyle=l,n.fillStyle=l;let c=u=>{n.beginPath(),n.arc(u.x,u.y,o,0,2*Math.PI),n.fill()};this.faceLandmarks.positions.forEach(c)}}};function bre(e,t){(Array.isArray(t)?t:[t]).forEach(a=>{let r=a instanceof ta?a:to(a)?a.landmarks:void 0;if(!r)throw new Error("drawFaceLandmarks - expected faceExpressions to be FaceLandmarks | WithFaceLandmarks<WithFaceDetection<{}>> or array thereof");new vw(r).draw(e)})}var K2="0.12.1";function xre(e,t){let n=Nu(e,t),a=Tu(e,t);function r(i,o,l){let c=a(i,o,`${l}/separable_conv0`),u=a(o,o,`${l}/separable_conv1`),p=n(i,o,1,`${l}/expansion_conv`);return{separable_conv0:c,separable_conv1:u,expansion_conv:p}}function s(i,o){let l=a(i,i,`${o}/separable_conv0`),c=a(i,i,`${o}/separable_conv1`),u=a(i,i,`${o}/separable_conv2`);return{separable_conv0:l,separable_conv1:c,separable_conv2:u}}return{extractConvParams:n,extractSeparableConvParams:a,extractReductionBlockParams:r,extractMainBlockParams:s}}function X2(e,t){let n=[],{extractWeights:a,getRemainingWeights:r}=bn(e),{extractConvParams:s,extractSeparableConvParams:i,extractReductionBlockParams:o,extractMainBlockParams:l}=xre(a,n),c=s(3,32,3,"entry_flow/conv_in"),u=o(32,64,"entry_flow/reduction_block_0"),p=o(64,128,"entry_flow/reduction_block_1"),d={conv_in:c,reduction_block_0:u,reduction_block_1:p},h={};sr(t,0,1).forEach(y=>{h[`main_block_${y}`]=l(128,`middle_flow/main_block_${y}`)});let m=o(128,256,"exit_flow/reduction_block"),f=i(256,512,"exit_flow/separable_conv"),g={reduction_block:m,separable_conv:f};if(r().length!==0)throw new Error(`weights remaing after extract: ${r().length}`);return{paramMappings:n,params:{entry_flow:d,middle_flow:h,exit_flow:g}}}function vre(e,t){let n=Un(e,t),a=cf(n),r=Su(n);function s(o){let l=r(`${o}/separable_conv0`),c=r(`${o}/separable_conv1`),u=a(`${o}/expansion_conv`);return{separable_conv0:l,separable_conv1:c,expansion_conv:u}}function i(o){let l=r(`${o}/separable_conv0`),c=r(`${o}/separable_conv1`),u=r(`${o}/separable_conv2`);return{separable_conv0:l,separable_conv1:c,separable_conv2:u}}return{extractConvParams:a,extractSeparableConvParams:r,extractReductionBlockParams:s,extractMainBlockParams:i}}function Y2(e,t){let n=[],{extractConvParams:a,extractSeparableConvParams:r,extractReductionBlockParams:s,extractMainBlockParams:i}=vre(e,n),o=a("entry_flow/conv_in"),l=s("entry_flow/reduction_block_0"),c=s("entry_flow/reduction_block_1"),u={conv_in:o,reduction_block_0:l,reduction_block_1:c},p={};sr(t,0,1).forEach(f=>{p[`main_block_${f}`]=i(`middle_flow/main_block_${f}`)});let d=s("exit_flow/reduction_block"),h=r("exit_flow/separable_conv"),m={reduction_block:d,separable_conv:h};return yn(e,n),{params:{entry_flow:u,middle_flow:p,exit_flow:m},paramMappings:n}}function Z2(e,t,n){return Z(At(e,t.filters,n,"same"),t.bias)}function kw(e,t,n=!0){let a=n?Ye(e):e;return a=Cn(a,t.separable_conv0,[1,1]),a=Cn(Ye(a),t.separable_conv1,[1,1]),a=Ft(a,[3,3],[2,2],"same"),a=Z(a,Z2(e,t.expansion_conv,[2,2])),a}function wre(e,t){let n=Cn(Ye(e),t.separable_conv0,[1,1]);return n=Cn(Ye(n),t.separable_conv1,[1,1]),n=Cn(Ye(n),t.separable_conv2,[1,1]),n=Z(n,e),n}var Iw=class extends an{constructor(t){super("TinyXception");this._numMainBlocks=t}forwardInput(t){let{params:n}=this;if(!n)throw new Error("TinyXception - load model before inference");return D(()=>{let a=ce(t.toBatchTensor(112,!0),"float32"),s=Oa(a,[122.782,117.001,104.298]).div(de(256)),i=Ye(Z2(s,n.entry_flow.conv_in,[2,2]));return i=kw(i,n.entry_flow.reduction_block_0,!1),i=kw(i,n.entry_flow.reduction_block_1),sr(this._numMainBlocks,0,1).forEach(o=>{i=wre(i,n.middle_flow[`main_block_${o}`])}),i=kw(i,n.exit_flow.reduction_block),i=Ye(Cn(i,n.exit_flow.separable_conv,[1,1])),i})}async forward(t){return this.forwardInput(await gt(t))}getDefaultModelName(){return"tiny_xception_model"}extractParamsFromWeightMap(t){return Y2(t,this._numMainBlocks)}extractParams(t){return X2(t,this._numMainBlocks)}};function J2(e){let t=[],{extractWeights:n,getRemainingWeights:a}=bn(e),r=of(n,t),s=r(512,1,"fc/age"),i=r(512,2,"fc/gender");if(a().length!==0)throw new Error(`weights remaing after extract: ${a().length}`);return{paramMappings:t,params:{fc:{age:s,gender:i}}}}function Q2(e){let t=[],n=Un(e,t);function a(s){let i=n(`${s}/weights`,2),o=n(`${s}/bias`,1);return{weights:i,bias:o}}let r={fc:{age:a("fc/age"),gender:a("fc/gender")}};return yn(e,t),{params:r,paramMappings:t}}var Er;(function(e){e.FEMALE="female",e.MALE="male"})(Er||(Er={}));var ff=class extends an{constructor(t=new Iw(2)){super("AgeGenderNet");this._faceFeatureExtractor=t}get faceFeatureExtractor(){return this._faceFeatureExtractor}runNet(t){let{params:n}=this;if(!n)throw new Error(`${this._name} - load model before inference`);return D(()=>{let a=t instanceof _r?this.faceFeatureExtractor.forwardInput(t):t,r=Xn(a,[7,7],[2,2],"valid").as2D(a.shape[0],-1),s=Op(r,n.fc.age).as1D(),i=Op(r,n.fc.gender);return{age:s,gender:i}})}forwardInput(t){return D(()=>{let{age:n,gender:a}=this.runNet(t);return{age:n,gender:Na(a)}})}async forward(t){return this.forwardInput(await gt(t))}async predictAgeAndGender(t){let n=await gt(t),a=await this.forwardInput(n),r=dt(a.age),s=dt(a.gender),i=r.map((l,c)=>({ageTensor:l,genderTensor:s[c]})),o=await Promise.all(i.map(async({ageTensor:l,genderTensor:c})=>{let u=(await l.data())[0],p=(await c.data())[0],d=p>.5,h=d?Er.MALE:Er.FEMALE,m=d?p:1-p;return l.dispose(),c.dispose(),{age:u,gender:h,genderProbability:m}}));return a.age.dispose(),a.gender.dispose(),n.isBatchInput?o:o[0]}getDefaultModelName(){return"age_gender_model"}dispose(t=!0){this.faceFeatureExtractor.dispose(t),super.dispose(t)}loadClassifierParams(t){let{params:n,paramMappings:a}=this.extractClassifierParams(t);this._params=n,this._paramMappings=a}extractClassifierParams(t){return J2(t)}extractParamsFromWeightMap(t){let{featureExtractorMap:n,classifierMap:a}=df(t);return this.faceFeatureExtractor.loadFromWeightMap(n),Q2(a)}extractParams(t){let n=512*1+1+(512*2+2),a=t.slice(0,t.length-n),r=t.slice(t.length-n);return this.faceFeatureExtractor.extractWeights(a),this.extractClassifierParams(r)}};var zp=class extends Lp{postProcess(t,n,a){let r=a.map(({width:i,height:o})=>{let l=n/Math.max(o,i);return{width:i*l,height:o*l}}),s=r.length;return D(()=>{let i=(p,d)=>$t([wn([68],p,"float32"),wn([68],d,"float32")],1).as2D(1,136).as1D(),o=(p,d)=>{let{width:h,height:m}=r[p];return d(h,m)?Math.abs(h-m)/2:0},l=p=>o(p,(d,h)=>d<h),c=p=>o(p,(d,h)=>h<d);return t.mul(wn([s,136],n,"float32")).sub($t(Array.from(Array(s),(p,d)=>i(l(d),c(d))))).div($t(Array.from(Array(s),(p,d)=>i(r[d].width,r[d].height))))})}forwardInput(t){return D(()=>{let n=this.runNet(t);return this.postProcess(n,t.inputSize,t.inputDimensions.map(([a,r])=>({height:a,width:r})))})}async forward(t){return this.forwardInput(await gt(t))}async detectLandmarks(t){let n=await gt(t),a=D(()=>dt(this.forwardInput(n))),r=await Promise.all(a.map(async(s,i)=>{let o=Array.from(await s.data()),l=o.filter((u,p)=>Qm(p)),c=o.filter((u,p)=>!Qm(p));return new vu(Array(68).fill(0).map((u,p)=>new De(l[p],c[p])),{height:n.getInputHeight(i),width:n.getInputWidth(i)})}));return a.forEach(s=>s.dispose()),n.isBatchInput?r:r[0]}getClassifierChannelsOut(){return 136}};var _u=class extends zp{constructor(t=new Pp){super("FaceLandmark68Net",t)}getDefaultModelName(){return"face_landmark_68_model"}getClassifierChannelsIn(){return 256}};function eC(e){let t=[],{extractDenseBlock3Params:n}=pf(e,t),a={dense0:n("dense0",!0),dense1:n("dense1"),dense2:n("dense2")};return yn(e,t),{params:a,paramMappings:t}}function tC(e){let t=[],{extractWeights:n,getRemainingWeights:a}=bn(e),{extractDenseBlock3Params:r}=uf(n,t),s=r(3,32,"dense0",!0),i=r(32,64,"dense1"),o=r(64,128,"dense2");if(a().length!==0)throw new Error(`weights remaing after extract: ${a().length}`);return{paramMappings:t,params:{dense0:s,dense1:i,dense2:o}}}var Nw=class extends an{constructor(){super("TinyFaceFeatureExtractor")}forwardInput(t){let{params:n}=this;if(!n)throw new Error("TinyFaceFeatureExtractor - load model before inference");return D(()=>{let a=ce(t.toBatchTensor(112,!0),"float32"),s=Oa(a,[122.782,117.001,104.298]).div(de(255)),i=sf(s,n.dense0,!0);return i=sf(i,n.dense1),i=sf(i,n.dense2),i=Xn(i,[14,14],[2,2],"valid"),i})}async forward(t){return this.forwardInput(await gt(t))}getDefaultModelName(){return"face_feature_extractor_tiny_model"}extractParamsFromWeightMap(t){return eC(t)}extractParams(t){return tC(t)}};var gf=class extends zp{constructor(t=new Nw){super("FaceLandmark68TinyNet",t)}getDefaultModelName(){return"face_landmark_68_tiny_model"}getClassifierChannelsIn(){return 128}};var nC=class extends _u{};function aC(e,t){return Z(L(e,t.weights),t.biases)}function Tw(e,t,n,a,r="same"){let{filters:s,bias:i}=t.conv,o=At(e,s,n,r);return o=Z(o,i),o=aC(o,t.scale),a?Ye(o):o}function rC(e,t){return Tw(e,t,[1,1],!0)}function Sw(e,t){return Tw(e,t,[1,1],!1)}function yf(e,t){return Tw(e,t,[2,2],!0,"valid")}function kre(e,t){function n(o,l,c){let u=e(o),p=u.length/(l*c*c);if(Jv(p))throw new Error(`depth has to be an integer: ${p}, weights.length: ${u.length}, numFilters: ${l}, filterSize: ${c}`);return D(()=>Ue(Sa(u,[l,p,c,c]),[2,3,1,0]))}function a(o,l,c,u){let p=n(o,l,c),d=tt(e(l));return t.push({paramPath:`${u}/filters`},{paramPath:`${u}/bias`}),{filters:p,bias:d}}function r(o,l){let c=tt(e(o)),u=tt(e(o));return t.push({paramPath:`${l}/weights`},{paramPath:`${l}/biases`}),{weights:c,biases:u}}function s(o,l,c,u){let p=a(o,l,c,`${u}/conv`),d=r(l,`${u}/scale`);return{conv:p,scale:d}}function i(o,l,c,u,p=!1){let d=s((p?.5:1)*o,l,c,`${u}/conv1`),h=s(o,l,c,`${u}/conv2`);return{conv1:d,conv2:h}}return{extractConvLayerParams:s,extractResidualLayerParams:i}}function sC(e){let{extractWeights:t,getRemainingWeights:n}=bn(e),a=[],{extractConvLayerParams:r,extractResidualLayerParams:s}=kre(t,a),i=r(4704,32,7,"conv32_down"),o=s(9216,32,3,"conv32_1"),l=s(9216,32,3,"conv32_2"),c=s(9216,32,3,"conv32_3"),u=s(36864,64,3,"conv64_down",!0),p=s(36864,64,3,"conv64_1"),d=s(36864,64,3,"conv64_2"),h=s(36864,64,3,"conv64_3"),m=s(147456,128,3,"conv128_down",!0),f=s(147456,128,3,"conv128_1"),g=s(147456,128,3,"conv128_2"),y=s(589824,256,3,"conv256_down",!0),b=s(589824,256,3,"conv256_1"),v=s(589824,256,3,"conv256_2"),x=s(589824,256,3,"conv256_down_out"),N=D(()=>Ue(Ta(t(256*128),[128,256]),[1,0]));if(a.push({paramPath:"fc"}),n().length!==0)throw new Error(`weights remaing after extract: ${n().length}`);return{params:{conv32_down:i,conv32_1:o,conv32_2:l,conv32_3:c,conv64_down:u,conv64_1:p,conv64_2:d,conv64_3:h,conv128_down:m,conv128_1:f,conv128_2:g,conv256_down:y,conv256_1:b,conv256_2:v,conv256_down_out:x,fc:N},paramMappings:a}}function Ire(e,t){let n=Un(e,t);function a(i){let o=n(`${i}/scale/weights`,1),l=n(`${i}/scale/biases`,1);return{weights:o,biases:l}}function r(i){let o=n(`${i}/conv/filters`,4),l=n(`${i}/conv/bias`,1),c=a(i);return{conv:{filters:o,bias:l},scale:c}}function s(i){return{conv1:r(`${i}/conv1`),conv2:r(`${i}/conv2`)}}return{extractConvLayerParams:r,extractResidualLayerParams:s}}function iC(e){let t=[],{extractConvLayerParams:n,extractResidualLayerParams:a}=Ire(e,t),r=n("conv32_down"),s=a("conv32_1"),i=a("conv32_2"),o=a("conv32_3"),l=a("conv64_down"),c=a("conv64_1"),u=a("conv64_2"),p=a("conv64_3"),d=a("conv128_down"),h=a("conv128_1"),m=a("conv128_2"),f=a("conv256_down"),g=a("conv256_1"),y=a("conv256_2"),b=a("conv256_down_out"),{fc:v}=e;if(t.push({originalPath:"fc",paramPath:"fc"}),!Zv(v))throw new Error(`expected weightMap[fc] to be a Tensor2D, instead have ${v}`);let x={conv32_down:r,conv32_1:s,conv32_2:i,conv32_3:o,conv64_down:l,conv64_1:c,conv64_2:u,conv64_3:p,conv128_down:d,conv128_1:h,conv128_2:m,conv256_down:f,conv256_1:g,conv256_2:y,conv256_down_out:b,fc:v};return yn(e,t),{params:x,paramMappings:t}}function La(e,t){let n=rC(e,t.conv1);return n=Sw(n,t.conv2),n=Z(n,e),n=Ye(n),n}function Bp(e,t){let n=yf(e,t.conv1);n=Sw(n,t.conv2);let a=Xn(e,2,2,"valid"),r=vt(a.shape),s=a.shape[3]!==n.shape[3];if(a.shape[1]!==n.shape[1]||a.shape[2]!==n.shape[2]){let o=[...n.shape];o[1]=1;let l=vt(o);n=Qe([n,l],1);let c=[...n.shape];c[2]=1;let u=vt(c);n=Qe([n,u],2)}return a=s?Qe([a,r],3):a,n=Z(a,n),n=Ye(n),n}var Eu=class extends an{constructor(){super("FaceRecognitionNet")}forwardInput(t){let{params:n}=this;if(!n)throw new Error("FaceRecognitionNet - load model before inference");return D(()=>{let a=ce(t.toBatchTensor(150,!0),"float32"),s=Oa(a,[122.782,117.001,104.298]).div(de(256)),i=yf(s,n.conv32_down);i=Ft(i,3,2,"valid"),i=La(i,n.conv32_1),i=La(i,n.conv32_2),i=La(i,n.conv32_3),i=Bp(i,n.conv64_down),i=La(i,n.conv64_1),i=La(i,n.conv64_2),i=La(i,n.conv64_3),i=Bp(i,n.conv128_down),i=La(i,n.conv128_1),i=La(i,n.conv128_2),i=Bp(i,n.conv256_down),i=La(i,n.conv256_1),i=La(i,n.conv256_2),i=Bp(i,n.conv256_down_out);let o=i.mean([1,2]);return ze(o,n.fc)})}async forward(t){return this.forwardInput(await gt(t))}async computeFaceDescriptor(t){var s;if((s=t==null?void 0:t.shape)==null?void 0:s.some(i=>i<=0))return new Float32Array(128);let n=await gt(t),a=D(()=>dt(this.forwardInput(n))),r=await Promise.all(a.map(i=>i.data()));return a.forEach(i=>i.dispose()),n.isBatchInput?r:r[0]}getDefaultModelName(){return"face_recognition_model"}extractParamsFromWeightMap(t){return iC(t)}extractParams(t){return sC(t)}};function Nre(e){let t=new Eu;return t.extractWeights(e),t}function bf(e,t){return{...e,...{descriptor:t}}}function Tre(e){return typeof e.age=="number"}function xf(e,t){return{...e,...{age:t}}}function Sre(e){return(e.gender===Er.MALE||e.gender===Er.FEMALE)&&yu(e.genderProbability)}function vf(e,t,n){return{...e,...{gender:t,genderProbability:n}}}function Cre(e,t){function n(l,c){let u=Sa(e(3*3*l),[3,3,l,1]),p=tt(e(l)),d=tt(e(l)),h=tt(e(l)),m=tt(e(l));return t.push({paramPath:`${c}/filters`},{paramPath:`${c}/batch_norm_scale`},{paramPath:`${c}/batch_norm_offset`},{paramPath:`${c}/batch_norm_mean`},{paramPath:`${c}/batch_norm_variance`}),{filters:u,batch_norm_scale:p,batch_norm_offset:d,batch_norm_mean:h,batch_norm_variance:m}}function a(l,c,u,p,d){let h=Sa(e(l*c*u*u),[u,u,l,c]),m=tt(e(c));return t.push({paramPath:`${p}/filters`},{paramPath:`${p}/${d?"batch_norm_offset":"bias"}`}),{filters:h,bias:m}}function r(l,c,u,p){let{filters:d,bias:h}=a(l,c,u,p,!0);return{filters:d,batch_norm_offset:h}}function s(l,c,u){let p=n(l,`${u}/depthwise_conv`),d=r(l,c,1,`${u}/pointwise_conv`);return{depthwise_conv:p,pointwise_conv:d}}function i(){let l=r(3,32,3,"mobilenetv1/conv_0"),c=s(32,64,"mobilenetv1/conv_1"),u=s(64,128,"mobilenetv1/conv_2"),p=s(128,128,"mobilenetv1/conv_3"),d=s(128,256,"mobilenetv1/conv_4"),h=s(256,256,"mobilenetv1/conv_5"),m=s(256,512,"mobilenetv1/conv_6"),f=s(512,512,"mobilenetv1/conv_7"),g=s(512,512,"mobilenetv1/conv_8"),y=s(512,512,"mobilenetv1/conv_9"),b=s(512,512,"mobilenetv1/conv_10"),v=s(512,512,"mobilenetv1/conv_11"),x=s(512,1024,"mobilenetv1/conv_12"),N=s(1024,1024,"mobilenetv1/conv_13");return{conv_0:l,conv_1:c,conv_2:u,conv_3:p,conv_4:d,conv_5:h,conv_6:m,conv_7:f,conv_8:g,conv_9:y,conv_10:b,conv_11:v,conv_12:x,conv_13:N}}function o(){let l=r(1024,256,1,"prediction_layer/conv_0"),c=r(256,512,3,"prediction_layer/conv_1"),u=r(512,128,1,"prediction_layer/conv_2"),p=r(128,256,3,"prediction_layer/conv_3"),d=r(256,128,1,"prediction_layer/conv_4"),h=r(128,256,3,"prediction_layer/conv_5"),m=r(256,64,1,"prediction_layer/conv_6"),f=r(64,128,3,"prediction_layer/conv_7"),g=a(512,12,1,"prediction_layer/box_predictor_0/box_encoding_predictor"),y=a(512,9,1,"prediction_layer/box_predictor_0/class_predictor"),b=a(1024,24,1,"prediction_layer/box_predictor_1/box_encoding_predictor"),v=a(1024,18,1,"prediction_layer/box_predictor_1/class_predictor"),x=a(512,24,1,"prediction_layer/box_predictor_2/box_encoding_predictor"),N=a(512,18,1,"prediction_layer/box_predictor_2/class_predictor"),T=a(256,24,1,"prediction_layer/box_predictor_3/box_encoding_predictor"),C=a(256,18,1,"prediction_layer/box_predictor_3/class_predictor"),$=a(256,24,1,"prediction_layer/box_predictor_4/box_encoding_predictor"),F=a(256,18,1,"prediction_layer/box_predictor_4/class_predictor"),O=a(128,24,1,"prediction_layer/box_predictor_5/box_encoding_predictor"),B=a(128,18,1,"prediction_layer/box_predictor_5/class_predictor");return{conv_0:l,conv_1:c,conv_2:u,conv_3:p,conv_4:d,conv_5:h,conv_6:m,conv_7:f,box_predictor_0:{box_encoding_predictor:g,class_predictor:y},box_predictor_1:{box_encoding_predictor:b,class_predictor:v},box_predictor_2:{box_encoding_predictor:x,class_predictor:N},box_predictor_3:{box_encoding_predictor:T,class_predictor:C},box_predictor_4:{box_encoding_predictor:$,class_predictor:F},box_predictor_5:{box_encoding_predictor:O,class_predictor:B}}}return{extractMobilenetV1Params:i,extractPredictionLayerParams:o}}function oC(e){let t=[],{extractWeights:n,getRemainingWeights:a}=bn(e),{extractMobilenetV1Params:r,extractPredictionLayerParams:s}=Cre(n,t),i=r(),o=s(),c={extra_dim:ah(n(5118*4),[1,5118,4])};if(t.push({paramPath:"output_layer/extra_dim"}),a().length!==0)throw new Error(`weights remaing after extract: ${a().length}`);return{params:{mobilenetv1:i,prediction_layer:o,output_layer:c},paramMappings:t}}function _re(e,t){let n=Un(e,t);function a(c,u,p){let d=n(`${c}/Conv2d_${u}_pointwise/weights`,4,`${p}/filters`),h=n(`${c}/Conv2d_${u}_pointwise/convolution_bn_offset`,1,`${p}/batch_norm_offset`);return{filters:d,batch_norm_offset:h}}function r(c){let u=`mobilenetv1/conv_${c}`,p=`MobilenetV1/Conv2d_${c}_depthwise`,d=`${u}/depthwise_conv`,h=`${u}/pointwise_conv`,m=n(`${p}/depthwise_weights`,4,`${d}/filters`),f=n(`${p}/BatchNorm/gamma`,1,`${d}/batch_norm_scale`),g=n(`${p}/BatchNorm/beta`,1,`${d}/batch_norm_offset`),y=n(`${p}/BatchNorm/moving_mean`,1,`${d}/batch_norm_mean`),b=n(`${p}/BatchNorm/moving_variance`,1,`${d}/batch_norm_variance`);return{depthwise_conv:{filters:m,batch_norm_scale:f,batch_norm_offset:g,batch_norm_mean:y,batch_norm_variance:b},pointwise_conv:a("MobilenetV1",c,h)}}function s(){return{conv_0:a("MobilenetV1",0,"mobilenetv1/conv_0"),conv_1:r(1),conv_2:r(2),conv_3:r(3),conv_4:r(4),conv_5:r(5),conv_6:r(6),conv_7:r(7),conv_8:r(8),conv_9:r(9),conv_10:r(10),conv_11:r(11),conv_12:r(12),conv_13:r(13)}}function i(c,u){let p=n(`${c}/weights`,4,`${u}/filters`),d=n(`${c}/biases`,1,`${u}/bias`);return{filters:p,bias:d}}function o(c){let u=i(`Prediction/BoxPredictor_${c}/BoxEncodingPredictor`,`prediction_layer/box_predictor_${c}/box_encoding_predictor`),p=i(`Prediction/BoxPredictor_${c}/ClassPredictor`,`prediction_layer/box_predictor_${c}/class_predictor`);return{box_encoding_predictor:u,class_predictor:p}}function l(){return{conv_0:a("Prediction",0,"prediction_layer/conv_0"),conv_1:a("Prediction",1,"prediction_layer/conv_1"),conv_2:a("Prediction",2,"prediction_layer/conv_2"),conv_3:a("Prediction",3,"prediction_layer/conv_3"),conv_4:a("Prediction",4,"prediction_layer/conv_4"),conv_5:a("Prediction",5,"prediction_layer/conv_5"),conv_6:a("Prediction",6,"prediction_layer/conv_6"),conv_7:a("Prediction",7,"prediction_layer/conv_7"),box_predictor_0:o(0),box_predictor_1:o(1),box_predictor_2:o(2),box_predictor_3:o(3),box_predictor_4:o(4),box_predictor_5:o(5)}}return{extractMobilenetV1Params:s,extractPredictionLayerParams:l}}function lC(e){let t=[],{extractMobilenetV1Params:n,extractPredictionLayerParams:a}=_re(e,t),r=e["Output/extra_dim"];if(t.push({originalPath:"Output/extra_dim",paramPath:"output_layer/extra_dim"}),!Sr(r))throw new Error(`expected weightMap['Output/extra_dim'] to be a Tensor3D, instead have ${r}`);let s={mobilenetv1:n(),prediction_layer:a(),output_layer:{extra_dim:r}};return yn(e,t),{params:s,paramMappings:t}}function ba(e,t,n){return D(()=>{let a=At(e,t.filters,n,"same");return a=Z(a,t.batch_norm_offset),qt(a,0,6)})}var Ere=.0010000000474974513;function Are(e,t,n){return D(()=>{let a=Yr(e,t.filters,n,"same");return a=fr(a,t.batch_norm_mean,t.batch_norm_variance,t.batch_norm_offset,t.batch_norm_scale,Ere),qt(a,0,6)})}function Fre(e){return[2,4,6,12].some(t=>t===e)?[2,2]:[1,1]}function uC(e,t){return D(()=>{let n,a=ba(e,t.conv_0,[2,2]);if([t.conv_1,t.conv_2,t.conv_3,t.conv_4,t.conv_5,t.conv_6,t.conv_7,t.conv_8,t.conv_9,t.conv_10,t.conv_11,t.conv_12,t.conv_13].forEach((s,i)=>{let o=i+1,l=Fre(o);a=Are(a,s.depthwise_conv,l),a=ba(a,s.pointwise_conv,[1,1]),o===11&&(n=a)}),n===null)throw new Error("mobileNetV1 - output of conv layer 11 is null");return{out:a,conv11:n}})}function $re(e,t,n){let a=e.arraySync(),r=Math.min(a[t][0],a[t][2]),s=Math.min(a[t][1],a[t][3]),i=Math.max(a[t][0],a[t][2]),o=Math.max(a[t][1],a[t][3]),l=Math.min(a[n][0],a[n][2]),c=Math.min(a[n][1],a[n][3]),u=Math.max(a[n][0],a[n][2]),p=Math.max(a[n][1],a[n][3]),d=(i-r)*(o-s),h=(u-l)*(p-c);if(d<=0||h<=0)return 0;let m=Math.max(r,l),f=Math.max(s,c),g=Math.min(i,u),y=Math.min(o,p),b=Math.max(g-m,0)*Math.max(y-f,0);return b/(d+h-b)}function cC(e,t,n,a,r){let s=e.shape[0],i=Math.min(n,s),o=t.map((u,p)=>({score:u,boxIndex:p})).filter(u=>u.score>r).sort((u,p)=>p.score-u.score),l=u=>u<=a?1:0,c=[];return o.forEach(u=>{if(c.length>=i)return;let p=u.score;for(let d=c.length-1;d>=0;--d){let h=$re(e,u.boxIndex,c[d]);if(h!==0&&(u.score*=l(h),u.score<=r))break}p===u.score&&c.push(u.boxIndex)}),c}function Dre(e){let t=dt(Ue(e,[1,0])),n=[fe(t[2],t[0]),fe(t[3],t[1])],a=[Z(t[0],ve(n[0],de(2))),Z(t[1],ve(n[1],de(2)))];return{sizes:n,centers:a}}function Rre(e,t){let{sizes:n,centers:a}=Dre(e),r=dt(Ue(t,[1,0])),s=ve(L(pn(ve(r[2],de(5))),n[0]),de(2)),i=Z(L(ve(r[0],de(10)),n[0]),a[0]),o=ve(L(pn(ve(r[3],de(5))),n[1]),de(2)),l=Z(L(ve(r[1],de(10)),n[1]),a[1]);return Ue($t([fe(i,s),fe(l,o),Z(i,s),Z(l,o)]),[1,0])}function pC(e,t,n){return D(()=>{let a=e.shape[0],r=Rre(j(ja(n.extra_dim,[a,1,1]),[-1,4]),j(e,[-1,4]));r=j(r,[a,r.shape[0]/a,4]);let s=ua(We(t,[0,0,1],[-1,-1,-1])),i=We(s,[0,0,0],[-1,-1,1]);i=j(i,[a,i.shape[1]]);let o=dt(r),l=dt(i);return{boxes:o,scores:l}})}function no(e,t){return D(()=>{let n=e.shape[0],a=j(eo(e,t.box_encoding_predictor),[n,-1,1,4]),r=j(eo(e,t.class_predictor),[n,-1,3]);return{boxPredictionEncoding:a,classPrediction:r}})}function dC(e,t,n){return D(()=>{let a=ba(e,n.conv_0,[1,1]),r=ba(a,n.conv_1,[2,2]),s=ba(r,n.conv_2,[1,1]),i=ba(s,n.conv_3,[2,2]),o=ba(i,n.conv_4,[1,1]),l=ba(o,n.conv_5,[2,2]),c=ba(l,n.conv_6,[1,1]),u=ba(c,n.conv_7,[2,2]),p=no(t,n.box_predictor_0),d=no(e,n.box_predictor_1),h=no(r,n.box_predictor_2),m=no(i,n.box_predictor_3),f=no(l,n.box_predictor_4),g=no(u,n.box_predictor_5),y=Qe([p.boxPredictionEncoding,d.boxPredictionEncoding,h.boxPredictionEncoding,m.boxPredictionEncoding,f.boxPredictionEncoding,g.boxPredictionEncoding],1),b=Qe([p.classPrediction,d.classPrediction,h.classPrediction,m.classPrediction,f.classPrediction,g.classPrediction],1);return{boxPredictions:y,classPredictions:b}})}var xa=class{constructor({minConfidence:t,maxResults:n}={}){this._name="SsdMobilenetv1Options";if(this._minConfidence=t||.5,this._maxResults=n||100,typeof this._minConfidence!="number"||this._minConfidence<=0||this._minConfidence>=1)throw new Error(`${this._name} - expected minConfidence to be a number between 0 and 1`);if(typeof this._maxResults!="number")throw new Error(`${this._name} - expected maxResults to be a number`)}get minConfidence(){return this._minConfidence}get maxResults(){return this._maxResults}};var ao=class extends an{constructor(){super("SsdMobilenetv1")}forwardInput(t){let{params:n}=this;if(!n)throw new Error("SsdMobilenetv1 - load model before inference");return D(()=>{let a=ce(t.toBatchTensor(512,!1),"float32"),r=fe(L(a,de(.007843137718737125)),de(1)),s=uC(r,n.mobilenetv1),{boxPredictions:i,classPredictions:o}=dC(s.out,s.conv11,n.prediction_layer);return pC(i,o,n.output_layer)})}async forward(t){return this.forwardInput(await gt(t))}async locateFaces(t,n={}){let{maxResults:a,minConfidence:r}=new xa(n),s=await gt(t),{boxes:i,scores:o}=this.forwardInput(s),l=i[0],c=o[0];for(let v=1;v<i.length;v++)i[v].dispose(),o[v].dispose();let u=Array.from(await c.data()),d=cC(l,u,a,.5,r),h=s.getReshapedInputDimensions(0),m=s.inputSize,f=m/h.width,g=m/h.height,y=l.arraySync(),b=d.map(v=>{let[x,N]=[Math.max(0,y[v][0]),Math.min(1,y[v][2])].map($=>$*g),[T,C]=[Math.max(0,y[v][1]),Math.min(1,y[v][3])].map($=>$*f);return new bt(u[v],new xu(T,x,C-T,N-x),{height:s.getInputHeight(0),width:s.getInputWidth(0)})});return l.dispose(),c.dispose(),b}getDefaultModelName(){return"ssd_mobilenetv1_model"}extractParamsFromWeightMap(t){return lC(t)}extractParams(t){return oC(t)}};function hC(e){let t=new ao;return t.extractWeights(e),t}function Mre(e){return hC(e)}var mC=class extends ao{};var fC=.4,gC=[new De(.738768,.874946),new De(2.42204,2.65704),new De(4.30971,7.04493),new De(10.246,4.59428),new De(12.6868,11.8741)],yC=[new De(1.603231,2.094468),new De(6.041143,7.080126),new De(2.882459,3.518061),new De(4.266906,5.178857),new De(9.041765,10.66308)],bC=[117.001,114.697,97.404],xC="tiny_yolov2_model",vC="tiny_yolov2_separable_conv_model";var wf=e=>typeof e=="number";function Cw(e){if(!e)throw new Error(`invalid config: ${e}`);if(typeof e.withSeparableConvs!="boolean")throw new Error(`config.withSeparableConvs has to be a boolean, have: ${e.withSeparableConvs}`);if(!wf(e.iouThreshold)||e.iouThreshold<0||e.iouThreshold>1)throw new Error(`config.iouThreshold has to be a number between [0, 1], have: ${e.iouThreshold}`);if(!Array.isArray(e.classes)||!e.classes.length||!e.classes.every(t=>typeof t=="string"))throw new Error(`config.classes has to be an array class names: string[], have: ${JSON.stringify(e.classes)}`);if(!Array.isArray(e.anchors)||!e.anchors.length||!e.anchors.map(t=>t||{}).every(t=>wf(t.x)&&wf(t.y)))throw new Error(`config.anchors has to be an array of { x: number, y: number }, have: ${JSON.stringify(e.anchors)}`);if(e.meanRgb&&(!Array.isArray(e.meanRgb)||e.meanRgb.length!==3||!e.meanRgb.every(wf)))throw new Error(`config.meanRgb has to be an array of shape [number, number, number], have: ${JSON.stringify(e.meanRgb)}`)}function Au(e){return D(()=>{let t=L(e,de(.10000000149011612));return Z(Ye(fe(e,t)),t)})}function Ar(e,t){return D(()=>{let n=Zn(e,[[0,0],[1,1],[1,1],[0,0]]);return n=At(n,t.conv.filters,[1,1],"valid"),n=fe(n,t.bn.sub),n=L(n,t.bn.truediv),n=Z(n,t.conv.bias),Au(n)})}function Fr(e,t){return D(()=>{let n=Zn(e,[[0,0],[1,1],[1,1],[0,0]]);return n=Ei(n,t.depthwise_filter,t.pointwise_filter,[1,1],"valid"),n=Z(n,t.bias),Au(n)})}function Pre(e,t){let n=Nu(e,t);function a(i,o){let l=tt(e(i)),c=tt(e(i));return t.push({paramPath:`${o}/sub`},{paramPath:`${o}/truediv`}),{sub:l,truediv:c}}function r(i,o,l){let c=n(i,o,3,`${l}/conv`),u=a(o,`${l}/bn`);return{conv:c,bn:u}}let s=Tu(e,t);return{extractConvParams:n,extractConvWithBatchNormParams:r,extractSeparableConvParams:s}}function wC(e,t,n,a){let{extractWeights:r,getRemainingWeights:s}=bn(e),i=[],{extractConvParams:o,extractConvWithBatchNormParams:l,extractSeparableConvParams:c}=Pre(r,i),u;if(t.withSeparableConvs){let[p,d,h,m,f,g,y,b,v]=a,x=t.isFirstLayerConv2d?o(p,d,3,"conv0"):c(p,d,"conv0"),N=c(d,h,"conv1"),T=c(h,m,"conv2"),C=c(m,f,"conv3"),$=c(f,g,"conv4"),F=c(g,y,"conv5"),O=b?c(y,b,"conv6"):void 0,B=v?c(b,v,"conv7"):void 0,W=o(v||b||y,5*n,1,"conv8");u={conv0:x,conv1:N,conv2:T,conv3:C,conv4:$,conv5:F,conv6:O,conv7:B,conv8:W}}else{let[p,d,h,m,f,g,y,b,v]=a,x=l(p,d,"conv0"),N=l(d,h,"conv1"),T=l(h,m,"conv2"),C=l(m,f,"conv3"),$=l(f,g,"conv4"),F=l(g,y,"conv5"),O=l(y,b,"conv6"),B=l(b,v,"conv7"),W=o(v,5*n,1,"conv8");u={conv0:x,conv1:N,conv2:T,conv3:C,conv4:$,conv5:F,conv6:O,conv7:B,conv8:W}}if(s().length!==0)throw new Error(`weights remaing after extract: ${s().length}`);return{params:u,paramMappings:i}}function Ore(e,t){let n=Un(e,t);function a(o){let l=n(`${o}/sub`,1),c=n(`${o}/truediv`,1);return{sub:l,truediv:c}}function r(o){let l=n(`${o}/filters`,4),c=n(`${o}/bias`,1);return{filters:l,bias:c}}function s(o){let l=r(`${o}/conv`),c=a(`${o}/bn`);return{conv:l,bn:c}}let i=Su(n);return{extractConvParams:r,extractConvWithBatchNormParams:s,extractSeparableConvParams:i}}function kC(e,t){let n=[],{extractConvParams:a,extractConvWithBatchNormParams:r,extractSeparableConvParams:s}=Ore(e,n),i;if(t.withSeparableConvs){let o=t.filterSizes&&t.filterSizes.length||9;i={conv0:t.isFirstLayerConv2d?a("conv0"):s("conv0"),conv1:s("conv1"),conv2:s("conv2"),conv3:s("conv3"),conv4:s("conv4"),conv5:s("conv5"),conv6:o>7?s("conv6"):void 0,conv7:o>8?s("conv7"):void 0,conv8:a("conv8")}}else i={conv0:r("conv0"),conv1:r("conv1"),conv2:r("conv2"),conv3:r("conv3"),conv4:r("conv4"),conv5:r("conv5"),conv6:r("conv6"),conv7:r("conv7"),conv8:a("conv8")};return yn(e,n),{params:i,paramMappings:n}}var lr=class{constructor({inputSize:t,scoreThreshold:n}={}){this._name="TinyYolov2Options";if(this._inputSize=t||416,this._scoreThreshold=n||.5,typeof this._inputSize!="number"||this._inputSize%32!=0)throw new Error(`${this._name} - expected inputSize to be a number divisible by 32`);if(typeof this._scoreThreshold!="number"||this._scoreThreshold<=0||this._scoreThreshold>=1)throw new Error(`${this._name} - expected scoreThreshold to be a number between 0 and 1`)}get inputSize(){return this._inputSize}get scoreThreshold(){return this._scoreThreshold}};var _w=class extends an{constructor(t){super("TinyYolov2");Cw(t),this._config=t}get config(){return this._config}get withClassScores(){return this.config.withClassScores||this.config.classes.length>1}get boxEncodingSize(){return 5+(this.withClassScores?this.config.classes.length:0)}runTinyYolov2(t,n){let a=Ar(t,n.conv0);return a=Ft(a,[2,2],[2,2],"same"),a=Ar(a,n.conv1),a=Ft(a,[2,2],[2,2],"same"),a=Ar(a,n.conv2),a=Ft(a,[2,2],[2,2],"same"),a=Ar(a,n.conv3),a=Ft(a,[2,2],[2,2],"same"),a=Ar(a,n.conv4),a=Ft(a,[2,2],[2,2],"same"),a=Ar(a,n.conv5),a=Ft(a,[2,2],[1,1],"same"),a=Ar(a,n.conv6),a=Ar(a,n.conv7),eo(a,n.conv8,"valid",!1)}runMobilenet(t,n){let a=this.config.isFirstLayerConv2d?Au(eo(t,n.conv0,"valid",!1)):Fr(t,n.conv0);return a=Ft(a,[2,2],[2,2],"same"),a=Fr(a,n.conv1),a=Ft(a,[2,2],[2,2],"same"),a=Fr(a,n.conv2),a=Ft(a,[2,2],[2,2],"same"),a=Fr(a,n.conv3),a=Ft(a,[2,2],[2,2],"same"),a=Fr(a,n.conv4),a=Ft(a,[2,2],[2,2],"same"),a=Fr(a,n.conv5),a=Ft(a,[2,2],[1,1],"same"),a=n.conv6?Fr(a,n.conv6):a,a=n.conv7?Fr(a,n.conv7):a,eo(a,n.conv8,"valid",!1)}forwardInput(t,n){let{params:a}=this;if(!a)throw new Error("TinyYolov2 - load model before inference");return D(()=>{let r=ce(t.toBatchTensor(n,!1),"float32");return r=this.config.meanRgb?Oa(r,this.config.meanRgb):r,r=r.div(de(256)),this.config.withSeparableConvs?this.runMobilenet(r,a):this.runTinyYolov2(r,a)})}async forward(t,n){return this.forwardInput(await gt(t),n)}async detect(t,n={}){let{inputSize:a,scoreThreshold:r}=new lr(n),s=await gt(t),i=await this.forwardInput(s,a),o=D(()=>dt(i)[0].expandDims()),l={width:s.getInputWidth(0),height:s.getInputHeight(0)},c=await this.extractBoxes(o,s.getReshapedInputDimensions(0),r);i.dispose(),o.dispose();let u=c.map(g=>g.box),p=c.map(g=>g.score),d=c.map(g=>g.classScore),h=c.map(g=>this.config.classes[g.label]);return nw(u.map(g=>g.rescale(a)),p,this.config.iouThreshold,!0).map(g=>new ds(p[g],d[g],h[g],u[g],l))}getDefaultModelName(){return""}extractParamsFromWeightMap(t){return kC(t,this.config)}extractParams(t){let n=this.config.filterSizes||_w.DEFAULT_FILTER_SIZES,a=n?n.length:void 0;if(a!==7&&a!==8&&a!==9)throw new Error(`TinyYolov2 - expected 7 | 8 | 9 convolutional filters, but found ${a} filterSizes in config`);return wC(t,this.config,this.boxEncodingSize,n)}async extractBoxes(t,n,a){let{width:r,height:s}=n,i=Math.max(r,s),o=i/r,l=i/s,c=t.shape[1],u=this.config.anchors.length,[p,d,h]=D(()=>{let y=t.reshape([c,c,u,this.boxEncodingSize]),b=y.slice([0,0,0,0],[c,c,u,4]),v=y.slice([0,0,0,4],[c,c,u,1]),x=this.withClassScores?Na(y.slice([0,0,0,5],[c,c,u,this.config.classes.length]),3):de(0);return[b,v,x]}),m=[],f=await d.array(),g=await p.array();for(let y=0;y<c;y++)for(let b=0;b<c;b++)for(let v=0;v<u;v++){let x=Ep(f[y][b][v][0]);if(!a||x>a){let N=(b+Ep(g[y][b][v][0]))/c*o,T=(y+Ep(g[y][b][v][1]))/c*l,C=Math.exp(g[y][b][v][2])*this.config.anchors[v].x/c*o,$=Math.exp(g[y][b][v][3])*this.config.anchors[v].y/c*l,F=N-C/2,O=T-$/2,B={row:y,col:b,anchor:v},{classScore:W,label:G}=this.withClassScores?await this.extractPredictedClass(h,B):{classScore:1,label:0};m.push({box:new bu(F,O,F+C,O+$),score:x,classScore:x*W,label:G,...B})}}return p.dispose(),d.dispose(),h.dispose(),m}async extractPredictedClass(t,n){let{row:a,col:r,anchor:s}=n,i=await t.array();return Array(this.config.classes.length).fill(0).map((o,l)=>i[a][r][s][l]).map((o,l)=>({classScore:o,label:l})).reduce((o,l)=>o.classScore>l.classScore?o:l)}},Fu=_w;Fu.DEFAULT_FILTER_SIZES=[3,16,32,64,128,256,512,1024,1024];var $u=class extends Fu{constructor(t=!0){let n={withSeparableConvs:t,iouThreshold:fC,classes:["face"],...t?{anchors:yC,meanRgb:bC}:{anchors:gC,withClassScores:!0}};super(n)}get withSeparableConvs(){return this.config.withSeparableConvs}get anchors(){return this.config.anchors}async locateFaces(t,n){return(await this.detect(t,n)).map(r=>new bt(r.score,r.relativeBox,{width:r.imageWidth,height:r.imageHeight}))}getDefaultModelName(){return this.withSeparableConvs?vC:xC}extractParamsFromWeightMap(t){return super.extractParamsFromWeightMap(t)}};function Lre(e,t=!0){let n=new $u(t);return n.extractWeights(e),n}var kf=class extends lr{constructor(){super(...arguments);this._name="TinyFaceDetectorOptions"}};var va=class{async then(t){return t(await this.run())}async run(){throw new Error("ComposableTask - run is not implemented")}};async function ro(e,t,n,a,r=({alignedRect:s})=>s){let s=e.map(l=>to(l)?r(l):l.detection),i=a||(t instanceof Ae?await Iu(t,s):await ku(t,s)),o=await n(i);return i.forEach(l=>l instanceof Ae&&l.dispose()),o}async function Du(e,t,n,a,r){return ro([e],t,async s=>n(s[0]),a,r)}var IC=.4,NC=[new De(1.603231,2.094468),new De(6.041143,7.080126),new De(2.882459,3.518061),new De(4.266906,5.178857),new De(9.041765,10.66308)],TC=[117.001,114.697,97.404];var Ru=class extends Fu{constructor(){let t={withSeparableConvs:!0,iouThreshold:IC,classes:["face"],anchors:NC,meanRgb:TC,isFirstLayerConv2d:!0,filterSizes:[3,16,32,64,128,256,512]};super(t)}get anchors(){return this.config.anchors}async locateFaces(t,n){return(await this.detect(t,n)).map(r=>new bt(r.score,r.relativeBox,{width:r.imageWidth,height:r.imageHeight}))}getDefaultModelName(){return"tiny_face_detector_model"}extractParamsFromWeightMap(t){return super.extractParamsFromWeightMap(t)}};var nt={ssdMobilenetv1:new ao,tinyFaceDetector:new Ru,tinyYolov2:new $u,faceLandmark68Net:new _u,faceLandmark68TinyNet:new gf,faceRecognitionNet:new Eu,faceExpressionNet:new hf,ageGenderNet:new ff},SC=(e,t)=>nt.ssdMobilenetv1.locateFaces(e,t),zre=(e,t)=>nt.tinyFaceDetector.locateFaces(e,t),Bre=(e,t)=>nt.tinyYolov2.locateFaces(e,t),CC=e=>nt.faceLandmark68Net.detectLandmarks(e),Wre=e=>nt.faceLandmark68TinyNet.detectLandmarks(e),Vre=e=>nt.faceRecognitionNet.computeFaceDescriptor(e),Ure=e=>nt.faceExpressionNet.predictExpressions(e),Gre=e=>nt.ageGenderNet.predictAgeAndGender(e),_C=e=>nt.ssdMobilenetv1.load(e),Hre=e=>nt.tinyFaceDetector.load(e),jre=e=>nt.tinyYolov2.load(e),qre=e=>nt.faceLandmark68Net.load(e),Kre=e=>nt.faceLandmark68TinyNet.load(e),Xre=e=>nt.faceRecognitionNet.load(e),Yre=e=>nt.faceExpressionNet.load(e),Zre=e=>nt.ageGenderNet.load(e),Jre=_C,Qre=SC,ese=CC;var Ew=class extends va{constructor(t,n,a){super();this.parentTask=t;this.input=n;this.extractedFaces=a}},Ou=class extends Ew{async run(){let t=await this.parentTask,n=await ro(t,this.input,async a=>Promise.all(a.map(r=>nt.faceExpressionNet.predictExpressions(r))),this.extractedFaces);return t.map((a,r)=>mf(a,n[r]))}withAgeAndGender(){return new Mu(this,this.input)}},Lu=class extends Ew{async run(){let t=await this.parentTask;if(!t)return;let n=await Du(t,this.input,a=>nt.faceExpressionNet.predictExpressions(a),this.extractedFaces);return mf(t,n)}withAgeAndGender(){return new Pu(this,this.input)}},oo=class extends Ou{withAgeAndGender(){return new so(this,this.input)}withFaceDescriptors(){return new fs(this,this.input)}},lo=class extends Lu{withAgeAndGender(){return new io(this,this.input)}withFaceDescriptor(){return new gs(this,this.input)}};var Aw=class extends va{constructor(t,n,a){super();this.parentTask=t;this.input=n;this.extractedFaces=a}},Mu=class extends Aw{async run(){let t=await this.parentTask,n=await ro(t,this.input,async a=>Promise.all(a.map(r=>nt.ageGenderNet.predictAgeAndGender(r))),this.extractedFaces);return t.map((a,r)=>{let{age:s,gender:i,genderProbability:o}=n[r];return xf(vf(a,i,o),s)})}withFaceExpressions(){return new Ou(this,this.input)}},Pu=class extends Aw{async run(){let t=await this.parentTask;if(!t)return;let{age:n,gender:a,genderProbability:r}=await Du(t,this.input,s=>nt.ageGenderNet.predictAgeAndGender(s),this.extractedFaces);return xf(vf(t,a,r),n)}withFaceExpressions(){return new Lu(this,this.input)}},so=class extends Mu{withFaceExpressions(){return new oo(this,this.input)}withFaceDescriptors(){return new fs(this,this.input)}},io=class extends Pu{withFaceExpressions(){return new lo(this,this.input)}withFaceDescriptor(){return new gs(this,this.input)}};var If=class extends va{constructor(t,n){super();this.parentTask=t;this.input=n}},fs=class extends If{async run(){let t=await this.parentTask;return(await ro(t,this.input,a=>Promise.all(a.map(r=>nt.faceRecognitionNet.computeFaceDescriptor(r))),null,a=>a.landmarks.align(null,{useDlibAlignment:!0}))).map((a,r)=>bf(t[r],a))}withFaceExpressions(){return new oo(this,this.input)}withAgeAndGender(){return new so(this,this.input)}},gs=class extends If{async run(){let t=await this.parentTask;if(!t)return;let n=await Du(t,this.input,a=>nt.faceRecognitionNet.computeFaceDescriptor(a),null,a=>a.landmarks.align(null,{useDlibAlignment:!0}));return bf(t,n)}withFaceExpressions(){return new lo(this,this.input)}withAgeAndGender(){return new io(this,this.input)}};var Nf=class extends va{constructor(t,n,a){super();this.parentTask=t;this.input=n;this.useTinyLandmarkNet=a}get landmarkNet(){return this.useTinyLandmarkNet?nt.faceLandmark68TinyNet:nt.faceLandmark68Net}},Tf=class extends Nf{async run(){let t=await this.parentTask,n=t.map(s=>s.detection),a=this.input instanceof Ae?await Iu(this.input,n):await ku(this.input,n),r=await Promise.all(a.map(s=>this.landmarkNet.detectLandmarks(s)));return a.forEach(s=>s instanceof Ae&&s.dispose()),t.map((s,i)=>Cu(s,r[i]))}withFaceExpressions(){return new oo(this,this.input)}withAgeAndGender(){return new so(this,this.input)}withFaceDescriptors(){return new fs(this,this.input)}},Sf=class extends Nf{async run(){let t=await this.parentTask;if(!t)return;let{detection:n}=t,a=this.input instanceof Ae?await Iu(this.input,[n]):await ku(this.input,[n]),r=await this.landmarkNet.detectLandmarks(a[0]);return a.forEach(s=>s instanceof Ae&&s.dispose()),Cu(t,r)}withFaceExpressions(){return new lo(this,this.input)}withAgeAndGender(){return new io(this,this.input)}withFaceDescriptor(){return new gs(this,this.input)}};var Cf=class extends va{constructor(t,n=new xa){super();this.input=t;this.options=n}},Wp=class extends Cf{async run(){let{input:t,options:n}=this,a=n instanceof kf?r=>nt.tinyFaceDetector.locateFaces(r,n):n instanceof xa?r=>nt.ssdMobilenetv1.locateFaces(r,n):n instanceof lr?r=>nt.tinyYolov2.locateFaces(r,n):null;if(!a)throw new Error("detectFaces - expected options to be instance of TinyFaceDetectorOptions | SsdMobilenetv1Options | MtcnnOptions | TinyYolov2Options");return a(t)}runAndExtendWithFaceDetections(){return new Promise(async t=>{let n=await this.run();t(n.map(a=>Yi({},a)))})}withFaceLandmarks(t=!1){return new Tf(this.runAndExtendWithFaceDetections(),this.input,t)}withFaceExpressions(){return new Ou(this.runAndExtendWithFaceDetections(),this.input)}withAgeAndGender(){return new Mu(this.runAndExtendWithFaceDetections(),this.input)}},_f=class extends Cf{async run(){let t=await new Wp(this.input,this.options),n=t[0];return t.forEach(a=>{a.score>n.score&&(n=a)}),n}runAndExtendWithFaceDetection(){return new Promise(async t=>{let n=await this.run();t(n?Yi({},n):void 0)})}withFaceLandmarks(t=!1){return new Sf(this.runAndExtendWithFaceDetection(),this.input,t)}withFaceExpressions(){return new Lu(this.runAndExtendWithFaceDetection(),this.input)}withAgeAndGender(){return new Pu(this.runAndExtendWithFaceDetection(),this.input)}};function tse(e,t=new xa){return new _f(e,t)}function Ef(e,t=new xa){return new Wp(e,t)}async function EC(e,t){return Ef(e,new xa(t?{minConfidence:t}:{})).withFaceLandmarks().withFaceDescriptors()}async function nse(e,t={}){return Ef(e,new lr(t)).withFaceLandmarks().withFaceDescriptors()}var ase=EC;function Fw(e,t){if(e.length!==t.length)throw new Error("euclideanDistance: arr1.length !== arr2.length");let n=Array.from(e),a=Array.from(t);return Math.sqrt(n.map((r,s)=>r-a[s]).reduce((r,s)=>r+s**2,0))}var Af=class{constructor(t,n=.6){this._distanceThreshold=n;let a=Array.isArray(t)?t:[t];if(!a.length)throw new Error("FaceRecognizer.constructor - expected atleast one input");let r=1,s=()=>`person ${r++}`;this._labeledDescriptors=a.map(i=>{if(i instanceof Cr)return i;if(i instanceof Float32Array)return new Cr(s(),[i]);if(i.descriptor&&i.descriptor instanceof Float32Array)return new Cr(s(),[i.descriptor]);throw new Error("FaceRecognizer.constructor - expected inputs to be of type LabeledFaceDescriptors | WithFaceDescriptor<any> | Float32Array | Array<LabeledFaceDescriptors | WithFaceDescriptor<any> | Float32Array>")})}get labeledDescriptors(){return this._labeledDescriptors}get distanceThreshold(){return this._distanceThreshold}computeMeanDistance(t,n){return n.map(a=>Fw(a,t)).reduce((a,r)=>a+r,0)/(n.length||1)}matchDescriptor(t){return this.labeledDescriptors.map(({descriptors:n,label:a})=>new Ap(a,this.computeMeanDistance(t,n))).reduce((n,a)=>n.distance<a.distance?n:a)}findBestMatch(t){let n=this.matchDescriptor(t);return n.distance<this.distanceThreshold?n:new Ap("unknown",n.distance)}toJSON(){return{distanceThreshold:this.distanceThreshold,labeledDescriptors:this.labeledDescriptors.map(t=>t.toJSON())}}static fromJSON(t){let n=t.labeledDescriptors.map(a=>Cr.fromJSON(a));return new Af(n,t.distanceThreshold)}};function rse(e){let t=new Ru;return t.extractWeights(e),t}function AC(e,t){let{width:n,height:a}=new gn(t.width,t.height);if(n<=0||a<=0)throw new Error(`resizeResults - invalid dimensions: ${JSON.stringify({width:n,height:a})}`);if(Array.isArray(e))return e.map(r=>AC(r,{width:n,height:a}));if(to(e)){let r=e.detection.forSize(n,a),s=e.unshiftedLandmarks.forSize(r.box.width,r.box.height);return Cu(Yi(e,r),s)}return ir(e)?Yi(e,e.detection.forSize(n,a)):e instanceof ta||e instanceof bt?e.forSize(n,a):e}var sse=typeof process!="undefined",ise=typeof navigator!="undefined"&&typeof navigator.userAgent!="undefined",ose={faceapi:K2,node:sse,browser:ise};export{ff as AgeGenderNet,bu as BoundingBox,ut as Box,va as ComposableTask,fs as ComputeAllFaceDescriptorsTask,If as ComputeFaceDescriptorsTaskBase,gs as ComputeSingleFaceDescriptorTask,Tf as DetectAllFaceLandmarksTask,Wp as DetectAllFacesTask,Nf as DetectFaceLandmarksTaskBase,Cf as DetectFacesTaskBase,Sf as DetectSingleFaceLandmarksTask,_f as DetectSingleFaceTask,gn as Dimensions,yw as FACE_EXPRESSION_LABELS,bt as FaceDetection,mC as FaceDetectionNet,hf as FaceExpressionNet,ms as FaceExpressions,_u as FaceLandmark68Net,gf as FaceLandmark68TinyNet,nC as FaceLandmarkNet,ta as FaceLandmarks,B2 as FaceLandmarks5,vu as FaceLandmarks68,Ap as FaceMatch,Af as FaceMatcher,Eu as FaceRecognitionNet,Er as Gender,Fp as LabeledBox,Cr as LabeledFaceDescriptors,_r as NetInput,an as NeuralNetwork,ds as ObjectDetection,De as Point,W2 as PredictedBox,xu as Rect,ao as SsdMobilenetv1,xa as SsdMobilenetv1Options,Ru as TinyFaceDetector,kf as TinyFaceDetectorOptions,$u as TinyYolov2,lr as TinyYolov2Options,ase as allFaces,EC as allFacesSsdMobilenetv1,nse as allFacesTinyYolov2,pw as awaitMediaLoaded,dw as bufferToImage,Vre as computeFaceDescriptor,wu as createCanvas,Rp as createCanvasFromMedia,Mre as createFaceDetectionNet,Nre as createFaceRecognitionNet,hC as createSsdMobilenetv1,rse as createTinyFaceDetector,Lre as createTinyYolov2,Ef as detectAllFaces,CC as detectFaceLandmarks,Wre as detectFaceLandmarksTiny,ese as detectLandmarks,tse as detectSingleFace,ww as draw,st as env,Fw as euclideanDistance,xf as extendWithAge,bf as extendWithFaceDescriptor,Yi as extendWithFaceDetection,mf as extendWithFaceExpressions,Cu as extendWithFaceLandmarks,vf as extendWithGender,Iu as extractFaceTensors,ku as extractFaces,mre as fetchImage,fw as fetchJson,fre as fetchNetWeights,Qi as fetchOrThrow,Sn as getContext2dOrThrow,Ji as getMediaDimensions,hw as imageTensorToCanvas,mw as imageToSquare,sre as inverseSigmoid,ew as iou,af as isMediaElement,Dp as isMediaLoaded,Tre as isWithAge,ir as isWithFaceDetection,bw as isWithFaceExpressions,to as isWithFaceLandmarks,Sre as isWithGender,Zre as loadAgeGenderModel,Jre as loadFaceDetectionModel,Yre as loadFaceExpressionModel,qre as loadFaceLandmarkModel,Kre as loadFaceLandmarkTinyModel,Xre as loadFaceRecognitionModel,_C as loadSsdMobilenetv1Model,Hre as loadTinyFaceDetectorModel,jre as loadTinyYolov2Model,gw as loadWeightMap,Qre as locateFaces,gre as matchDimensions,tw as minBbox,nt as nets,nw as nonMaxSuppression,Oa as normalize,aw as padToSquare,Gre as predictAgeAndGender,Ure as recognizeFaceExpressions,AC as resizeResults,Zi as resolveInput,rre as shuffleArray,Ep as sigmoid,SC as ssdMobilenetv1,Bg as tf,zre as tinyFaceDetector,Bre as tinyYolov2,gt as toNetInput,Yv as utils,Cw as validateConfig,ose as version};
|
|
/**
|
|
* @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 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 See the LICENSE file. */
|
|
//# sourceMappingURL=face-api.esm.js.map
|