ambiser
/
face-api
mirror of https://github.com/vladmandic/face-api.git
9
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
face-api/dist/tfjs.esm.js

4912 lines
1.1 MiB
Raw Blame History

/*
Face-API
homepage: <https://github.com/vladmandic/face-api>
author: <https://github.com/vladmandic>'
*/
var mU=Object.create;var Iv=Object.defineProperty;var fU=Object.getOwnPropertyDescriptor;var dU=Object.getOwnPropertyNames;var hU=Object.getPrototypeOf,gU=Object.prototype.hasOwnProperty;var Oe=(r,t)=>()=>(t||r((t={exports:{}}).exports,t),t.exports),Yt=(r,t)=>{for(var e in t)Iv(r,e,{get:t[e],enumerable:!0})},xU=(r,t,e,n)=>{if(t&&typeof t=="object"||typeof t=="function")for(let o of dU(t))!gU.call(r,o)&&o!==e&&Iv(r,o,{get:()=>t[o],enumerable:!(n=fU(t,o))||n.enumerable});return r};var xu=(r,t,e)=>(e=r!=null?mU(hU(r)):{},xU(t||!r||!r.__esModule?Iv(e,"default",{value:r,enumerable:!0}):e,r));var c1=Oe((Dat,u1)=>{u1.exports=Xe;var lo=null;try{lo=new WebAssembly.Instance(new WebAssembly.Module(new Uint8Array([0,97,115,109,1,0,0,0,1,13,2,96,0,1,127,96,4,127,127,127,127,1,127,3,7,6,0,1,1,1,1,1,6,6,1,127,1,65,0,11,7,50,6,3,109,117,108,0,1,5,100,105,118,95,115,0,2,5,100,105,118,95,117,0,3,5,114,101,109,95,115,0,4,5,114,101,109,95,117,0,5,8,103,101,116,95,104,105,103,104,0,0,10,191,1,6,4,0,35,0,11,36,1,1,126,32,0,173,32,1,173,66,32,134,132,32,2,173,32,3,173,66,32,134,132,126,34,4,66,32,135,167,36,0,32,4,167,11,36,1,1,126,32,0,173,32,1,173,66,32,134,132,32,2,173,32,3,173,66,32,134,132,127,34,4,66,32,135,167,36,0,32,4,167,11,36,1,1,126,32,0,173,32,1,173,66,32,134,132,32,2,173,32,3,173,66,32,134,132,128,34,4,66,32,135,167,36,0,32,4,167,11,36,1,1,126,32,0,173,32,1,173,66,32,134,132,32,2,173,32,3,173,66,32,134,132,129,34,4,66,32,135,167,36,0,32,4,167,11,36,1,1,126,32,0,173,32,1,173,66,32,134,132,32,2,173,32,3,173,66,32,134,132,130,34,4,66,32,135,167,36,0,32,4,167,11])),{}).exports}catch(r){}function Xe(r,t,e){this.low=r|0,this.high=t|0,this.unsigned=!!e}Xe.prototype.__isLong__;Object.defineProperty(Xe.prototype,"__isLong__",{value:!0});function Pn(r){return(r&&r.__isLong__)===!0}Xe.isLong=Pn;var t1={},e1={};function Su(r,t){var e,n,o;return t?(r>>>=0,(o=0<=r&&r<256)&&(n=e1[r],n)?n:(e=Ye(r,(r|0)<0?-1:0,!0),o&&(e1[r]=e),e)):(r|=0,(o=-128<=r&&r<128)&&(n=t1[r],n)?n:(e=Ye(r,r<0?-1:0,!1),o&&(t1[r]=e),e))}Xe.fromInt=Su;function uo(r,t){if(isNaN(r))return t?Iu:co;if(t){if(r<0)return Iu;if(r>=s1)return l1}else{if(r<=-n1)return On;if(r+1>=n1)return a1}return r<0?uo(-r,t).neg():Ye(r%Vp|0,r/Vp|0,t)}Xe.fromNumber=uo;function Ye(r,t,e){return new Xe(r,t,e)}Xe.fromBits=Ye;var nx=Math.pow;function Pv(r,t,e){if(r.length===0)throw Error("empty string");if(r==="NaN"||r==="Infinity"||r==="+Infinity"||r==="-Infinity")return co;if(typeof t=="number"?(e=t,t=!1):t=!!t,e=e||10,e<2||36<e)throw RangeError("radix");var n;if((n=r.indexOf("-"))>0)throw Error("interior hyphen");if(n===0)return Pv(r.substring(1),t,e).neg();for(var o=uo(nx(e,8)),s=co,i=0;i<r.length;i+=8){var a=Math.min(8,r.length-i),u=parseInt(r.substring(i,i+a),e);if(a<8){var l=uo(nx(e,a));s=s.mul(l).add(uo(u))}else s=s.mul(o),s=s.add(uo(u))}return s.unsigned=t,s}Xe.fromString=Pv;function Gs(r,t){return typeof r=="number"?uo(r,t):typeof r=="string"?Pv(r,t):Ye(r.low,r.high,typeof t=="boolean"?t:r.unsigned)}Xe.fromValue=Gs;var r1=1<<16,zU=1<<24,Vp=r1*r1,s1=Vp*Vp,n1=s1/2,o1=Su(zU),co=Su(0);Xe.ZERO=co;var Iu=Su(0,!0);Xe.UZERO=Iu;var Bp=Su(1);Xe.ONE=Bp;var i1=Su(1,!0);Xe.UONE=i1;var Ov=Su(-1);Xe.NEG_ONE=Ov;var a1=Ye(-1,2147483647,!1);Xe.MAX_VALUE=a1;var l1=Ye(-1,-1,!0);Xe.MAX_UNSIGNED_VALUE=l1;var On=Ye(0,-2147483648,!1);Xe.MIN_VALUE=On;var yt=Xe.prototype;yt.toInt=function(){return this.unsigned?this.low>>>0:this.low};yt.toNumber=function(){return this.unsigned?(this.high>>>0)*Vp+(this.low>>>0):this.high*Vp+(this.low>>>0)};yt.toString=function(t){if(t=t||10,t<2||36<t)throw RangeError("radix");if(this.isZero())return"0";if(this.isNegative())if(this.eq(On)){var e=uo(t),n=this.div(e),o=n.mul(e).sub(this);return n.toString(t)+o.toInt().toString(t)}else return"-"+this.neg().toString(t);for(var s=uo(nx(t,6),this.unsigned),i=this,a="";;){var u=i.div(s),l=i.sub(u.mul(s)).toInt()>>>0,c=l.toString(t);if(i=u,i.isZero())return c+a;for(;c.length<6;)c="0"+c;a=""+c+a}};yt.getHighBits=function(){return this.high};yt.getHighBitsUnsigned=function(){return this.high>>>0};yt.getLowBits=function(){return this.low};yt.getLowBitsUnsigned=function(){return this.low>>>0};yt.getNumBitsAbs=function(){if(this.isNegative())return this.eq(On)?64:this.neg().getNumBitsAbs();for(var t=this.high!=0?this.high:this.low,e=31;e>0&&(t&1<<e)==0;e--);return this.high!=0?e+33:e+1};yt.isZero=function(){return this.high===0&&this.low===0};yt.eqz=yt.isZero;yt.isNegative=function(){return!this.unsigned&&this.high<0};yt.isPositive=function(){return this.unsigned||this.high>=0};yt.isOdd=function(){return(this.low&1)===1};yt.isEven=function(){return(this.low&1)===0};yt.equals=function(t){return Pn(t)||(t=Gs(t)),this.unsigned!==t.unsigned&&this.high>>>31===1&&t.high>>>31===1?!1:this.high===t.high&&this.low===t.low};yt.eq=yt.equals;yt.notEquals=function(t){return!this.eq(t)};yt.neq=yt.notEquals;yt.ne=yt.notEquals;yt.lessThan=function(t){return this.comp(t)<0};yt.lt=yt.lessThan;yt.lessThanOrEqual=function(t){return this.comp(t)<=0};yt.lte=yt.lessThanOrEqual;yt.le=yt.lessThanOrEqual;yt.greaterThan=function(t){return this.comp(t)>0};yt.gt=yt.greaterThan;yt.greaterThanOrEqual=function(t){return this.comp(t)>=0};yt.gte=yt.greaterThanOrEqual;yt.ge=yt.greaterThanOrEqual;yt.compare=function(t){if(Pn(t)||(t=Gs(t)),this.eq(t))return 0;var e=this.isNegative(),n=t.isNegative();return e&&!n?-1:!e&&n?1:this.unsigned?t.high>>>0>this.high>>>0||t.high===this.high&&t.low>>>0>this.low>>>0?-1:1:this.sub(t).isNegative()?-1:1};yt.comp=yt.compare;yt.negate=function(){return!this.unsigned&&this.eq(On)?On:this.not().add(Bp)};yt.neg=yt.negate;yt.add=function(t){Pn(t)||(t=Gs(t));var e=this.high>>>16,n=this.high&65535,o=this.low>>>16,s=this.low&65535,i=t.high>>>16,a=t.high&65535,u=t.low>>>16,l=t.low&65535,c=0,p=0,m=0,f=0;return f+=s+l,m+=f>>>16,f&=65535,m+=o+u,p+=m>>>16,m&=65535,p+=n+a,c+=p>>>16,p&=65535,c+=e+i,c&=65535,Ye(m<<16|f,c<<16|p,this.unsigned)};yt.subtract=function(t){return Pn(t)||(t=Gs(t)),this.add(t.neg())};yt.sub=yt.subtract;yt.multiply=function(t){if(this.isZero())return co;if(Pn(t)||(t=Gs(t)),lo){var e=lo.mul(this.low,this.high,t.low,t.high);return Ye(e,lo.get_high(),this.unsigned)}if(t.isZero())return co;if(this.eq(On))return t.isOdd()?On:co;if(t.eq(On))return this.isOdd()?On:co;if(this.isNegative())return t.isNegative()?this.neg().mul(t.neg()):this.neg().mul(t).neg();if(t.isNegative())return this.mul(t.neg()).neg();if(this.lt(o1)&&t.lt(o1))return uo(this.toNumber()*t.toNumber(),this.unsigned);var n=this.high>>>16,o=this.high&65535,s=this.low>>>16,i=this.low&65535,a=t.high>>>16,u=t.high&65535,l=t.low>>>16,c=t.low&65535,p=0,m=0,f=0,d=0;return d+=i*c,f+=d>>>16,d&=65535,f+=s*c,m+=f>>>16,f&=65535,f+=i*l,m+=f>>>16,f&=65535,m+=o*c,p+=m>>>16,m&=65535,m+=s*l,p+=m>>>16,m&=65535,m+=i*u,p+=m>>>16,m&=65535,p+=n*c+o*l+s*u+i*a,p&=65535,Ye(f<<16|d,p<<16|m,this.unsigned)};yt.mul=yt.multiply;yt.divide=function(t){if(Pn(t)||(t=Gs(t)),t.isZero())throw Error("division by zero");if(lo){if(!this.unsigned&&this.high===-2147483648&&t.low===-1&&t.high===-1)return this;var e=(this.unsigned?lo.div_u:lo.div_s)(this.low,this.high,t.low,t.high);return Ye(e,lo.get_high(),this.unsigned)}if(this.isZero())return this.unsigned?Iu:co;var n,o,s;if(this.unsigned){if(t.unsigned||(t=t.toUnsigned()),t.gt(this))return Iu;if(t.gt(this.shru(1)))return i1;s=Iu}else{if(this.eq(On)){if(t.eq(Bp)||t.eq(Ov))return On;if(t.eq(On))return Bp;var i=this.shr(1);return n=i.div(t).shl(1),n.eq(co)?t.isNegative()?Bp:Ov:(o=this.sub(t.mul(n)),s=n.add(o.div(t)),s)}else if(t.eq(On))return this.unsigned?Iu:co;if(this.isNegative())return t.isNegative()?this.neg().div(t.neg()):this.neg().div(t).neg();if(t.isNegative())return this.div(t.neg()).neg();s=co}for(o=this;o.gte(t);){n=Math.max(1,Math.floor(o.toNumber()/t.toNumber()));for(var a=Math.ceil(Math.log(n)/Math.LN2),u=a<=48?1:nx(2,a-48),l=uo(n),c=l.mul(t);c.isNegative()||c.gt(o);)n-=u,l=uo(n,this.unsigned),c=l.mul(t);l.isZero()&&(l=Bp),s=s.add(l),o=o.sub(c)}return s};yt.div=yt.divide;yt.modulo=function(t){if(Pn(t)||(t=Gs(t)),lo){var e=(this.unsigned?lo.rem_u:lo.rem_s)(this.low,this.high,t.low,t.high);return Ye(e,lo.get_high(),this.unsigned)}return this.sub(this.div(t).mul(t))};yt.mod=yt.modulo;yt.rem=yt.modulo;yt.not=function(){return Ye(~this.low,~this.high,this.unsigned)};yt.and=function(t){return Pn(t)||(t=Gs(t)),Ye(this.low&t.low,this.high&t.high,this.unsigned)};yt.or=function(t){return Pn(t)||(t=Gs(t)),Ye(this.low|t.low,this.high|t.high,this.unsigned)};yt.xor=function(t){return Pn(t)||(t=Gs(t)),Ye(this.low^t.low,this.high^t.high,this.unsigned)};yt.shiftLeft=function(t){return Pn(t)&&(t=t.toInt()),(t&=63)===0?this:t<32?Ye(this.low<<t,this.high<<t|this.low>>>32-t,this.unsigned):Ye(0,this.low<<t-32,this.unsigned)};yt.shl=yt.shiftLeft;yt.shiftRight=function(t){return Pn(t)&&(t=t.toInt()),(t&=63)===0?this:t<32?Ye(this.low>>>t|this.high<<32-t,this.high>>t,this.unsigned):Ye(this.high>>t-32,this.high>=0?0:-1,this.unsigned)};yt.shr=yt.shiftRight;yt.shiftRightUnsigned=function(t){if(Pn(t)&&(t=t.toInt()),t&=63,t===0)return this;var e=this.high;if(t<32){var n=this.low;return Ye(n>>>t|e<<32-t,e>>>t,this.unsigned)}else return t===32?Ye(e,0,this.unsigned):Ye(e>>>t-32,0,this.unsigned)};yt.shru=yt.shiftRightUnsigned;yt.shr_u=yt.shiftRightUnsigned;yt.toSigned=function(){return this.unsigned?Ye(this.low,this.high,!1):this};yt.toUnsigned=function(){return this.unsigned?this:Ye(this.low,this.high,!0)};yt.toBytes=function(t){return t?this.toBytesLE():this.toBytesBE()};yt.toBytesLE=function(){var t=this.high,e=this.low;return[e&255,e>>>8&255,e>>>16&255,e>>>24,t&255,t>>>8&255,t>>>16&255,t>>>24]};yt.toBytesBE=function(){var t=this.high,e=this.low;return[t>>>24,t>>>16&255,t>>>8&255,t&255,e>>>24,e>>>16&255,e>>>8&255,e&255]};Xe.fromBytes=function(t,e,n){return n?Xe.fromBytesLE(t,e):Xe.fromBytesBE(t,e)};Xe.fromBytesLE=function(t,e){return new Xe(t[0]|t[1]<<8|t[2]<<16|t[3]<<24,t[4]|t[5]<<8|t[6]<<16|t[7]<<24,e)};Xe.fromBytesBE=function(t,e){return new Xe(t[4]<<24|t[5]<<16|t[6]<<8|t[7],t[0]<<24|t[1]<<16|t[2]<<8|t[3],e)}});var K1=Oe(()=>{});var j1=Oe(()=>{});var D_=Oe((A_,H0)=>{(function(r,t,e){function n(a){var u=this,l=i();u.next=function(){var c=2091639*u.s0+u.c*23283064365386963e-26;return u.s0=u.s1,u.s1=u.s2,u.s2=c-(u.c=c|0)},u.c=1,u.s0=l(" "),u.s1=l(" "),u.s2=l(" "),u.s0-=l(a),u.s0<0&&(u.s0+=1),u.s1-=l(a),u.s1<0&&(u.s1+=1),u.s2-=l(a),u.s2<0&&(u.s2+=1),l=null}function o(a,u){return u.c=a.c,u.s0=a.s0,u.s1=a.s1,u.s2=a.s2,u}function s(a,u){var l=new n(a),c=u&&u.state,p=l.next;return p.int32=function(){return l.next()*4294967296|0},p.double=function(){return p()+(p()*2097152|0)*11102230246251565e-32},p.quick=p,c&&(typeof c=="object"&&o(c,l),p.state=function(){return o(l,{})}),p}function i(){var a=4022871197,u=function(l){l=l.toString();for(var c=0;c<l.length;c++){a+=l.charCodeAt(c);var p=.02519603282416938*a;a=p>>>0,p-=a,p*=a,a=p>>>0,p-=a,a+=p*4294967296}return(a>>>0)*23283064365386963e-26};return u}t&&t.exports?t.exports=s:e&&e.amd?e(function(){return s}):this.alea=s})(A_,typeof H0=="object"&&H0,typeof define=="function"&&define)});var R_=Oe(($_,q0)=>{(function(r,t,e){function n(i){var a=this,u="";a.x=0,a.y=0,a.z=0,a.w=0,a.next=function(){var c=a.x^a.x<<11;return a.x=a.y,a.y=a.z,a.z=a.w,a.w^=a.w>>>19^c^c>>>8},i===(i|0)?a.x=i:u+=i;for(var l=0;l<u.length+64;l++)a.x^=u.charCodeAt(l)|0,a.next()}function o(i,a){return a.x=i.x,a.y=i.y,a.z=i.z,a.w=i.w,a}function s(i,a){var u=new n(i),l=a&&a.state,c=function(){return(u.next()>>>0)/4294967296};return c.double=function(){do var p=u.next()>>>11,m=(u.next()>>>0)/4294967296,f=(p+m)/(1<<21);while(f===0);return f},c.int32=u.next,c.quick=c,l&&(typeof l=="object"&&o(l,u),c.state=function(){return o(u,{})}),c}t&&t.exports?t.exports=s:e&&e.amd?e(function(){return s}):this.xor128=s})($_,typeof q0=="object"&&q0,typeof define=="function"&&define)});var O_=Oe((F_,K0)=>{(function(r,t,e){function n(i){var a=this,u="";a.next=function(){var c=a.x^a.x>>>2;return a.x=a.y,a.y=a.z,a.z=a.w,a.w=a.v,(a.d=a.d+362437|0)+(a.v=a.v^a.v<<4^(c^c<<1))|0},a.x=0,a.y=0,a.z=0,a.w=0,a.v=0,i===(i|0)?a.x=i:u+=i;for(var l=0;l<u.length+64;l++)a.x^=u.charCodeAt(l)|0,l==u.length&&(a.d=a.x<<10^a.x>>>4),a.next()}function o(i,a){return a.x=i.x,a.y=i.y,a.z=i.z,a.w=i.w,a.v=i.v,a.d=i.d,a}function s(i,a){var u=new n(i),l=a&&a.state,c=function(){return(u.next()>>>0)/4294967296};return c.double=function(){do var p=u.next()>>>11,m=(u.next()>>>0)/4294967296,f=(p+m)/(1<<21);while(f===0);return f},c.int32=u.next,c.quick=c,l&&(typeof l=="object"&&o(l,u),c.state=function(){return o(u,{})}),c}t&&t.exports?t.exports=s:e&&e.amd?e(function(){return s}):this.xorwow=s})(F_,typeof K0=="object"&&K0,typeof define=="function"&&define)});var M_=Oe((P_,j0)=>{(function(r,t,e){function n(i){var a=this;a.next=function(){var l=a.x,c=a.i,p,m,f;return p=l[c],p^=p>>>7,m=p^p<<24,p=l[c+1&7],m^=p^p>>>10,p=l[c+3&7],m^=p^p>>>3,p=l[c+4&7],m^=p^p<<7,p=l[c+7&7],p=p^p<<13,m^=p^p<<9,l[c]=m,a.i=c+1&7,m};function u(l,c){var p,m,f=[];if(c===(c|0))m=f[0]=c;else for(c=""+c,p=0;p<c.length;++p)f[p&7]=f[p&7]<<15^c.charCodeAt(p)+f[p+1&7]<<13;for(;f.length<8;)f.push(0);for(p=0;p<8&&f[p]===0;++p);for(p==8?m=f[7]=-1:m=f[p],l.x=f,l.i=0,p=256;p>0;--p)l.next()}u(a,i)}function o(i,a){return a.x=i.x.slice(),a.i=i.i,a}function s(i,a){i==null&&(i=+new Date);var u=new n(i),l=a&&a.state,c=function(){return(u.next()>>>0)/4294967296};return c.double=function(){do var p=u.next()>>>11,m=(u.next()>>>0)/4294967296,f=(p+m)/(1<<21);while(f===0);return f},c.int32=u.next,c.quick=c,l&&(l.x&&o(l,u),c.state=function(){return o(u,{})}),c}t&&t.exports?t.exports=s:e&&e.amd?e(function(){return s}):this.xorshift7=s})(P_,typeof j0=="object"&&j0,typeof define=="function"&&define)});var z_=Oe((L_,X0)=>{(function(r,t,e){function n(i){var a=this;a.next=function(){var l=a.w,c=a.X,p=a.i,m,f;return a.w=l=l+1640531527|0,f=c[p+34&127],m=c[p=p+1&127],f^=f<<13,m^=m<<17,f^=f>>>15,m^=m>>>12,f=c[p]=f^m,a.i=p,f+(l^l>>>16)|0};function u(l,c){var p,m,f,d,h,g=[],y=128;for(c===(c|0)?(m=c,c=null):(c=c+"\0",m=0,y=Math.max(y,c.length)),f=0,d=-32;d<y;++d)c&&(m^=c.charCodeAt((d+32)%c.length)),d===0&&(h=m),m^=m<<10,m^=m>>>15,m^=m<<4,m^=m>>>13,d>=0&&(h=h+1640531527|0,p=g[d&127]^=m+h,f=p==0?f+1:0);for(f>=128&&(g[(c&&c.length||0)&127]=-1),f=127,d=4*128;d>0;--d)m=g[f+34&127],p=g[f=f+1&127],m^=m<<13,p^=p<<17,m^=m>>>15,p^=p>>>12,g[f]=m^p;l.w=h,l.X=g,l.i=f}u(a,i)}function o(i,a){return a.i=i.i,a.w=i.w,a.X=i.X.slice(),a}function s(i,a){i==null&&(i=+new Date);var u=new n(i),l=a&&a.state,c=function(){return(u.next()>>>0)/4294967296};return c.double=function(){do var p=u.next()>>>11,m=(u.next()>>>0)/4294967296,f=(p+m)/(1<<21);while(f===0);return f},c.int32=u.next,c.quick=c,l&&(l.X&&o(l,u),c.state=function(){return o(u,{})}),c}t&&t.exports?t.exports=s:e&&e.amd?e(function(){return s}):this.xor4096=s})(L_,typeof X0=="object"&&X0,typeof define=="function"&&define)});var V_=Oe((B_,Y0)=>{(function(r,t,e){function n(i){var a=this,u="";a.next=function(){var c=a.b,p=a.c,m=a.d,f=a.a;return c=c<<25^c>>>7^p,p=p-m|0,m=m<<24^m>>>8^f,f=f-c|0,a.b=c=c<<20^c>>>12^p,a.c=p=p-m|0,a.d=m<<16^p>>>16^f,a.a=f-c|0},a.a=0,a.b=0,a.c=-1640531527,a.d=1367130551,i===Math.floor(i)?(a.a=i/4294967296|0,a.b=i|0):u+=i;for(var l=0;l<u.length+20;l++)a.b^=u.charCodeAt(l)|0,a.next()}function o(i,a){return a.a=i.a,a.b=i.b,a.c=i.c,a.d=i.d,a}function s(i,a){var u=new n(i),l=a&&a.state,c=function(){return(u.next()>>>0)/4294967296};return c.double=function(){do var p=u.next()>>>11,m=(u.next()>>>0)/4294967296,f=(p+m)/(1<<21);while(f===0);return f},c.int32=u.next,c.quick=c,l&&(typeof l=="object"&&o(l,u),c.state=function(){return o(u,{})}),c}t&&t.exports?t.exports=s:e&&e.amd?e(function(){return s}):this.tychei=s})(B_,typeof Y0=="object"&&Y0,typeof define=="function"&&define)});var Z0=Oe(()=>{});var G_=Oe((pIt,Ix)=>{(function(r,t){var e=this,n=256,o=6,s=52,i="random",a=t.pow(n,o),u=t.pow(2,s),l=u*2,c=n-1,p;function m(w,v,k){var E=[];v=v==!0?{entropy:!0}:v||{};var D=g(h(v.entropy?[w,b(r)]:w==null?y():w,3),E),$=new f(E),R=function(){for(var M=$.g(o),W=a,U=0;M<u;)M=(M+U)*n,W*=n,U=$.g(1);for(;M>=l;)M/=2,W/=2,U>>>=1;return(M+U)/W};return R.int32=function(){return $.g(4)|0},R.quick=function(){return $.g(4)/4294967296},R.double=R,g(b($.S),r),(v.pass||k||function(M,W,U,q){return q&&(q.S&&d(q,$),M.state=function(){return d($,{})}),U?(t[i]=M,W):M})(R,D,"global"in v?v.global:this==t,v.state)}t["seed"+i]=m;function f(w){var v,k=w.length,E=this,D=0,$=E.i=E.j=0,R=E.S=[];for(k||(w=[k++]);D<n;)R[D]=D++;for(D=0;D<n;D++)R[D]=R[$=c&$+w[D%k]+(v=R[D])],R[$]=v;(E.g=function(M){for(var W,U=0,q=E.i,K=E.j,j=E.S;M--;)W=j[q=c&q+1],U=U*n+j[c&(j[q]=j[K=c&K+W])+(j[K]=W)];return E.i=q,E.j=K,U})(n)}function d(w,v){return v.i=w.i,v.j=w.j,v.S=w.S.slice(),v}function h(w,v){var k=[],E=typeof w,D;if(v&&E=="object")for(D in w)try{k.push(h(w[D],v-1))}catch($){}return k.length?k:E=="string"?w:w+"\0"}function g(w,v){for(var k=w+"",E,D=0;D<k.length;)v[c&D]=c&(E^=v[c&D]*19)+k.charCodeAt(D++);return b(v)}function y(){try{var w;return p&&(w=p.randomBytes)?w=w(n):(w=new Uint8Array(n),(e.crypto||e.msCrypto).getRandomValues(w)),b(w)}catch(E){var v=e.navigator,k=v&&v.plugins;return[+new Date,e,k,e.screen,b(r)]}}function b(w){return String.fromCharCode.apply(0,w)}if(g(t.random(),r),typeof Ix=="object"&&Ix.exports){Ix.exports=m;try{p=Z0()}catch(w){}}else typeof define=="function"&&define.amd&&define(function(){return m})})([],Math)});var J0=Oe((mIt,W_)=>{var GK=D_(),WK=R_(),UK=O_(),HK=M_(),qK=z_(),KK=V_(),Hu=G_();Hu.alea=GK;Hu.xor128=WK;Hu.xorwow=UK;Hu.xorshift7=HK;Hu.xor4096=qK;Hu.tychei=KK;W_.exports=Hu});var k$=Oe((S$,VI)=>{(function(r,t,e){function n(a){var u=this,l=i();u.next=function(){var c=2091639*u.s0+u.c*23283064365386963e-26;return u.s0=u.s1,u.s1=u.s2,u.s2=c-(u.c=c|0)},u.c=1,u.s0=l(" "),u.s1=l(" "),u.s2=l(" "),u.s0-=l(a),u.s0<0&&(u.s0+=1),u.s1-=l(a),u.s1<0&&(u.s1+=1),u.s2-=l(a),u.s2<0&&(u.s2+=1),l=null}function o(a,u){return u.c=a.c,u.s0=a.s0,u.s1=a.s1,u.s2=a.s2,u}function s(a,u){var l=new n(a),c=u&&u.state,p=l.next;return p.int32=function(){return l.next()*4294967296|0},p.double=function(){return p()+(p()*2097152|0)*11102230246251565e-32},p.quick=p,c&&(typeof c=="object"&&o(c,l),p.state=function(){return o(l,{})}),p}function i(){var a=4022871197,u=function(l){l=String(l);for(var c=0;c<l.length;c++){a+=l.charCodeAt(c);var p=.02519603282416938*a;a=p>>>0,p-=a,p*=a,a=p>>>0,p-=a,a+=p*4294967296}return(a>>>0)*23283064365386963e-26};return u}t&&t.exports?t.exports=s:e&&e.amd?e(function(){return s}):this.alea=s})(S$,typeof VI=="object"&&VI,typeof define=="function"&&define)});var T$=Oe((N$,GI)=>{(function(r,t,e){function n(i){var a=this,u="";a.x=0,a.y=0,a.z=0,a.w=0,a.next=function(){var c=a.x^a.x<<11;return a.x=a.y,a.y=a.z,a.z=a.w,a.w^=a.w>>>19^c^c>>>8},i===(i|0)?a.x=i:u+=i;for(var l=0;l<u.length+64;l++)a.x^=u.charCodeAt(l)|0,a.next()}function o(i,a){return a.x=i.x,a.y=i.y,a.z=i.z,a.w=i.w,a}function s(i,a){var u=new n(i),l=a&&a.state,c=function(){return(u.next()>>>0)/4294967296};return c.double=function(){do var p=u.next()>>>11,m=(u.next()>>>0)/4294967296,f=(p+m)/(1<<21);while(f===0);return f},c.int32=u.next,c.quick=c,l&&(typeof l=="object"&&o(l,u),c.state=function(){return o(u,{})}),c}t&&t.exports?t.exports=s:e&&e.amd?e(function(){return s}):this.xor128=s})(N$,typeof GI=="object"&&GI,typeof define=="function"&&define)});var E$=Oe((_$,WI)=>{(function(r,t,e){function n(i){var a=this,u="";a.next=function(){var c=a.x^a.x>>>2;return a.x=a.y,a.y=a.z,a.z=a.w,a.w=a.v,(a.d=a.d+362437|0)+(a.v=a.v^a.v<<4^(c^c<<1))|0},a.x=0,a.y=0,a.z=0,a.w=0,a.v=0,i===(i|0)?a.x=i:u+=i;for(var l=0;l<u.length+64;l++)a.x^=u.charCodeAt(l)|0,l==u.length&&(a.d=a.x<<10^a.x>>>4),a.next()}function o(i,a){return a.x=i.x,a.y=i.y,a.z=i.z,a.w=i.w,a.v=i.v,a.d=i.d,a}function s(i,a){var u=new n(i),l=a&&a.state,c=function(){return(u.next()>>>0)/4294967296};return c.double=function(){do var p=u.next()>>>11,m=(u.next()>>>0)/4294967296,f=(p+m)/(1<<21);while(f===0);return f},c.int32=u.next,c.quick=c,l&&(typeof l=="object"&&o(l,u),c.state=function(){return o(u,{})}),c}t&&t.exports?t.exports=s:e&&e.amd?e(function(){return s}):this.xorwow=s})(_$,typeof WI=="object"&&WI,typeof define=="function"&&define)});var D$=Oe((A$,UI)=>{(function(r,t,e){function n(i){var a=this;a.next=function(){var l=a.x,c=a.i,p,m,f;return p=l[c],p^=p>>>7,m=p^p<<24,p=l[c+1&7],m^=p^p>>>10,p=l[c+3&7],m^=p^p>>>3,p=l[c+4&7],m^=p^p<<7,p=l[c+7&7],p=p^p<<13,m^=p^p<<9,l[c]=m,a.i=c+1&7,m};function u(l,c){var p,m,f=[];if(c===(c|0))m=f[0]=c;else for(c=""+c,p=0;p<c.length;++p)f[p&7]=f[p&7]<<15^c.charCodeAt(p)+f[p+1&7]<<13;for(;f.length<8;)f.push(0);for(p=0;p<8&&f[p]===0;++p);for(p==8?m=f[7]=-1:m=f[p],l.x=f,l.i=0,p=256;p>0;--p)l.next()}u(a,i)}function o(i,a){return a.x=i.x.slice(),a.i=i.i,a}function s(i,a){i==null&&(i=+new Date);var u=new n(i),l=a&&a.state,c=function(){return(u.next()>>>0)/4294967296};return c.double=function(){do var p=u.next()>>>11,m=(u.next()>>>0)/4294967296,f=(p+m)/(1<<21);while(f===0);return f},c.int32=u.next,c.quick=c,l&&(l.x&&o(l,u),c.state=function(){return o(u,{})}),c}t&&t.exports?t.exports=s:e&&e.amd?e(function(){return s}):this.xorshift7=s})(A$,typeof UI=="object"&&UI,typeof define=="function"&&define)});var R$=Oe(($$,HI)=>{(function(r,t,e){function n(i){var a=this;a.next=function(){var l=a.w,c=a.X,p=a.i,m,f;return a.w=l=l+1640531527|0,f=c[p+34&127],m=c[p=p+1&127],f^=f<<13,m^=m<<17,f^=f>>>15,m^=m>>>12,f=c[p]=f^m,a.i=p,f+(l^l>>>16)|0};function u(l,c){var p,m,f,d,h,g=[],y=128;for(c===(c|0)?(m=c,c=null):(c=c+"\0",m=0,y=Math.max(y,c.length)),f=0,d=-32;d<y;++d)c&&(m^=c.charCodeAt((d+32)%c.length)),d===0&&(h=m),m^=m<<10,m^=m>>>15,m^=m<<4,m^=m>>>13,d>=0&&(h=h+1640531527|0,p=g[d&127]^=m+h,f=p==0?f+1:0);for(f>=128&&(g[(c&&c.length||0)&127]=-1),f=127,d=4*128;d>0;--d)m=g[f+34&127],p=g[f=f+1&127],m^=m<<13,p^=p<<17,m^=m>>>15,p^=p>>>12,g[f]=m^p;l.w=h,l.X=g,l.i=f}u(a,i)}function o(i,a){return a.i=i.i,a.w=i.w,a.X=i.X.slice(),a}function s(i,a){i==null&&(i=+new Date);var u=new n(i),l=a&&a.state,c=function(){return(u.next()>>>0)/4294967296};return c.double=function(){do var p=u.next()>>>11,m=(u.next()>>>0)/4294967296,f=(p+m)/(1<<21);while(f===0);return f},c.int32=u.next,c.quick=c,l&&(l.X&&o(l,u),c.state=function(){return o(u,{})}),c}t&&t.exports?t.exports=s:e&&e.amd?e(function(){return s}):this.xor4096=s})($$,typeof HI=="object"&&HI,typeof define=="function"&&define)});var O$=Oe((F$,qI)=>{(function(r,t,e){function n(i){var a=this,u="";a.next=function(){var c=a.b,p=a.c,m=a.d,f=a.a;return c=c<<25^c>>>7^p,p=p-m|0,m=m<<24^m>>>8^f,f=f-c|0,a.b=c=c<<20^c>>>12^p,a.c=p=p-m|0,a.d=m<<16^p>>>16^f,a.a=f-c|0},a.a=0,a.b=0,a.c=-1640531527,a.d=1367130551,i===Math.floor(i)?(a.a=i/4294967296|0,a.b=i|0):u+=i;for(var l=0;l<u.length+20;l++)a.b^=u.charCodeAt(l)|0,a.next()}function o(i,a){return a.a=i.a,a.b=i.b,a.c=i.c,a.d=i.d,a}function s(i,a){var u=new n(i),l=a&&a.state,c=function(){return(u.next()>>>0)/4294967296};return c.double=function(){do var p=u.next()>>>11,m=(u.next()>>>0)/4294967296,f=(p+m)/(1<<21);while(f===0);return f},c.int32=u.next,c.quick=c,l&&(typeof l=="object"&&o(l,u),c.state=function(){return o(u,{})}),c}t&&t.exports?t.exports=s:e&&e.amd?e(function(){return s}):this.tychei=s})(F$,typeof qI=="object"&&qI,typeof define=="function"&&define)});var M$=Oe((P$,Ry)=>{(function(r,t,e){var n=256,o=6,s=52,i="random",a=e.pow(n,o),u=e.pow(2,s),l=u*2,c=n-1,p;function m(w,v,k){var E=[];v=v==!0?{entropy:!0}:v||{};var D=g(h(v.entropy?[w,b(t)]:w==null?y():w,3),E),$=new f(E),R=function(){for(var M=$.g(o),W=a,U=0;M<u;)M=(M+U)*n,W*=n,U=$.g(1);for(;M>=l;)M/=2,W/=2,U>>>=1;return(M+U)/W};return R.int32=function(){return $.g(4)|0},R.quick=function(){return $.g(4)/4294967296},R.double=R,g(b($.S),t),(v.pass||k||function(M,W,U,q){return q&&(q.S&&d(q,$),M.state=function(){return d($,{})}),U?(e[i]=M,W):M})(R,D,"global"in v?v.global:this==e,v.state)}function f(w){var v,k=w.length,E=this,D=0,$=E.i=E.j=0,R=E.S=[];for(k||(w=[k++]);D<n;)R[D]=D++;for(D=0;D<n;D++)R[D]=R[$=c&$+w[D%k]+(v=R[D])],R[$]=v;(E.g=function(M){for(var W,U=0,q=E.i,K=E.j,j=E.S;M--;)W=j[q=c&q+1],U=U*n+j[c&(j[q]=j[K=c&K+W])+(j[K]=W)];return E.i=q,E.j=K,U})(n)}function d(w,v){return v.i=w.i,v.j=w.j,v.S=w.S.slice(),v}function h(w,v){var k=[],E=typeof w,D;if(v&&E=="object")for(D in w)try{k.push(h(w[D],v-1))}catch($){}return k.length?k:E=="string"?w:w+"\0"}function g(w,v){for(var k=w+"",E,D=0;D<k.length;)v[c&D]=c&(E^=v[c&D]*19)+k.charCodeAt(D++);return b(v)}function y(){try{var w;return p&&(w=p.randomBytes)?w=w(n):(w=new Uint8Array(n),(r.crypto||r.msCrypto).getRandomValues(w)),b(w)}catch(E){var v=r.navigator,k=v&&v.plugins;return[+new Date,r,k,r.screen,b(t)]}}function b(w){return String.fromCharCode.apply(0,w)}if(g(e.random(),t),typeof Ry=="object"&&Ry.exports){Ry.exports=m;try{p=Z0()}catch(w){}}else typeof define=="function"&&define.amd?define(function(){return m}):e["seed"+i]=m})(typeof self!="undefined"?self:P$,[],Math)});var KI=Oe((Iie,L$)=>{var aZ=k$(),lZ=T$(),uZ=E$(),cZ=D$(),pZ=R$(),mZ=O$(),Ic=M$();Ic.alea=aZ;Ic.xor128=lZ;Ic.xorwow=uZ;Ic.xorshift7=cZ;Ic.xor4096=pZ;Ic.tychei=mZ;L$.exports=Ic});var jI=Oe(()=>{});var By=Oe(()=>{});var xg=Oe(()=>{});var uW=Oe(()=>{});var cW=Oe(()=>{});var pW=Oe(()=>{});var mW=Oe((Gb,xT)=>{var gT=(()=>{var r=typeof document!="undefined"&&document.currentScript?document.currentScript.src:void 0;return typeof __filename!="undefined"&&(r=r||__filename),function(t){t=t||{};function e(){return Ot.buffer!=gr&&Bo(Ot.buffer),Uc}function n(){return Ot.buffer!=gr&&Bo(Ot.buffer),Hc}function o(){return Ot.buffer!=gr&&Bo(Ot.buffer),ef}function s(){return Ot.buffer!=gr&&Bo(Ot.buffer),vg}function i(){return Ot.buffer!=gr&&Bo(Ot.buffer),Cg}function a(){return Ot.buffer!=gr&&Bo(Ot.buffer),Ig}function u(){return Ot.buffer!=gr&&Bo(Ot.buffer),Sg}var l=typeof t!="undefined"?t:{},c,p;l.ready=new Promise(function(T,F){c=T,p=F});var m;typeof process!="undefined"&&process.listeners&&(m={uncaughtException:process.listeners("uncaughtException"),unhandledRejection:process.listeners("unhandledRejection")});var f=Object.assign({},l),d=[],h="./this.program",g=(T,F)=>{throw F},y=typeof window=="object",b=typeof importScripts=="function",w=typeof process=="object"&&typeof process.versions=="object"&&typeof process.versions.node=="string",v=l.ENVIRONMENT_IS_PTHREAD||!1,k="";function E(T){return l.locateFile?l.locateFile(T,k):k+T}var D,$,R,M;function W(T){if(T instanceof mf)return;X("exiting due to exception: "+T)}var U,q,K;if(w){b?k=xg().dirname(k)+"/":k=__dirname+"/",K=()=>{q||(U=By(),q=xg())},D=function(G,Y){return K(),G=q.normalize(G),U.readFileSync(G,Y?void 0:"utf8")},R=F=>{var G=D(F,!0);return G.buffer||(G=new Uint8Array(G)),G},$=(F,G,Y)=>{K(),F=q.normalize(F),U.readFile(F,function(ht,wt){ht?Y(ht):G(wt.buffer)})},process.argv.length>1&&(h=process.argv[1].replace(/\\/g,"/")),d=process.argv.slice(2),process.on("uncaughtException",function(F){if(!(F instanceof mf))throw F}),process.on("unhandledRejection",function(F){throw F}),g=(F,G)=>{if(mu())throw process.exitCode=F,G;W(G),process.exit(F)},l.inspect=function(){return"[Emscripten Module object]"};let T;try{T=uW()}catch(F){throw console.error('The "worker_threads" module is not supported in this node.js build - perhaps a newer version is needed?'),F}global.Worker=T.Worker}else(y||b)&&(b?k=self.location.href:typeof document!="undefined"&&document.currentScript&&(k=document.currentScript.src),typeof r!="undefined"&&r&&(k=r),k.indexOf("blob:")!==0?k=k.substr(0,k.replace(/[?#].*/,"").lastIndexOf("/")+1):k="",w||(D=T=>{var F=new XMLHttpRequest;return F.open("GET",T,!1),F.send(null),F.responseText},b&&(R=T=>{var F=new XMLHttpRequest;return F.open("GET",T,!1),F.responseType="arraybuffer",F.send(null),new Uint8Array(F.response)}),$=(T,F,G)=>{var Y=new XMLHttpRequest;Y.open("GET",T,!0),Y.responseType="arraybuffer",Y.onload=()=>{if(Y.status==200||Y.status==0&&Y.response){F(Y.response);return}G()},Y.onerror=G,Y.send(null)}),M=T=>document.title=T);w&&typeof performance=="undefined"&&(global.performance=cW().performance);var j=console.log.bind(console),Q=console.warn.bind(console);w&&(K(),j=T=>U.writeSync(1,T+`
`),Q=T=>U.writeSync(2,T+`
`));var rt=l.print||j,X=l.printErr||Q;Object.assign(l,f),f=null,l.arguments&&(d=l.arguments),l.thisProgram&&(h=l.thisProgram),l.quit&&(g=l.quit);var ot=4;function st(T){st.shown||(st.shown={}),st.shown[T]||(st.shown[T]=1,X(T))}function it(T,F){if(typeof WebAssembly.Function=="function"){for(var G={i:"i32",j:"i64",f:"f32",d:"f64"},Y={parameters:[],results:F[0]=="v"?[]:[G[F[0]]]},ht=1;ht<F.length;++ht)Y.parameters.push(G[F[ht]]);return new WebAssembly.Function(Y,T)}var wt=[1,0,1,96],Nt=F.slice(0,1),Bt=F.slice(1),or={i:127,j:126,f:125,d:124};wt.push(Bt.length);for(var ht=0;ht<Bt.length;++ht)wt.push(or[Bt[ht]]);Nt=="v"?wt.push(0):wt=wt.concat([1,or[Nt]]),wt[1]=wt.length-2;var Uo=new Uint8Array([0,97,115,109,1,0,0,0].concat(wt,[2,7,1,1,101,1,102,0,0,7,5,1,1,102,0,0])),Ho=new WebAssembly.Module(Uo),Jg=new WebAssembly.Instance(Ho,{e:{f:T}}),ff=Jg.exports.f;return ff}var ft=[],at;function xt(){if(ft.length)return ft.pop();try{Un.grow(1)}catch(T){throw T instanceof RangeError?"Unable to grow wasm table. Set ALLOW_TABLE_GROWTH.":T}return Un.length-1}function dt(T,F){for(var G=T;G<T+F;G++){var Y=Yc(G);Y&&at.set(Y,G)}}var bt=0,kt=T=>{bt=T},At=Atomics.load,$t=Atomics.store,qt=Atomics.compareExchange,Kt;l.wasmBinary&&(Kt=l.wasmBinary);var ce=l.noExitRuntime||!0;typeof WebAssembly!="object"&&Kc("no native wasm support detected");var Ot,Ae,Ie=!1,ae;function je(T,F){T||Kc(F)}function Fe(T){var F=l["_"+T];return F}function Jr(T,F,G,Y,ht){var wt={string:function(Rn){var np=0;if(Rn!=null&&Rn!==0){var WT=(Rn.length<<2)+1;np=rp(WT),$n(Rn,np,WT)}return np},array:function(Rn){var np=rp(Rn.length);return gl(Rn,np),np}};function Nt(Rn){return F==="string"?Hr(Rn):F==="boolean"?Boolean(Rn):Rn}var Bt=Fe(T),or=[],Uo=0;if(Y)for(var Ho=0;Ho<Y.length;Ho++){var Jg=wt[G[Ho]];Jg?(Uo===0&&(Uo=vv()),or[Ho]=Jg(Y[Ho])):or[Ho]=Y[Ho]}var ff=Bt.apply(null,or);function pU(Rn){return Uo!==0&&jg(Uo),Nt(Rn)}return ff=pU(ff),ff}function ze(T,F,G,Y){G=G||[];var ht=G.every(function(Nt){return Nt==="number"}),wt=F!=="string";return wt&&ht&&!Y?Fe(T):function(){return Jr(T,F,G,arguments,Y)}}var Lo=1;function Fr(T){var F=new TextDecoder(T);this.decode=G=>(G.buffer instanceof SharedArrayBuffer&&(G=new Uint8Array(G)),F.decode.call(F,G))}var Qr=typeof TextDecoder!="undefined"?new Fr("utf8"):void 0;function tn(T,F,G){for(var Y=F+G,ht=F;T[ht]&&!(ht>=Y);)++ht;if(ht-F>16&&T.subarray&&Qr)return Qr.decode(T.subarray(F,ht));for(var wt="";F<ht;){var Nt=T[F++];if(!(Nt&128)){wt+=String.fromCharCode(Nt);continue}var Bt=T[F++]&63;if((Nt&224)==192){wt+=String.fromCharCode((Nt&31)<<6|Bt);continue}var or=T[F++]&63;if((Nt&240)==224?Nt=(Nt&15)<<12|Bt<<6|or:Nt=(Nt&7)<<18|Bt<<12|or<<6|T[F++]&63,Nt<65536)wt+=String.fromCharCode(Nt);else{var Uo=Nt-65536;wt+=String.fromCharCode(55296|Uo>>10,56320|Uo&1023)}}return wt}function Hr(T,F){return T?tn(n(),T,F):""}function ro(T,F,G,Y){if(!(Y>0))return 0;for(var ht=G,wt=G+Y-1,Nt=0;Nt<T.length;++Nt){var Bt=T.charCodeAt(Nt);if(Bt>=55296&&Bt<=57343){var or=T.charCodeAt(++Nt);Bt=65536+((Bt&1023)<<10)|or&1023}if(Bt<=127){if(G>=wt)break;F[G++]=Bt}else if(Bt<=2047){if(G+1>=wt)break;F[G++]=192|Bt>>6,F[G++]=128|Bt&63}else if(Bt<=65535){if(G+2>=wt)break;F[G++]=224|Bt>>12,F[G++]=128|Bt>>6&63,F[G++]=128|Bt&63}else{if(G+3>=wt)break;F[G++]=240|Bt>>18,F[G++]=128|Bt>>12&63,F[G++]=128|Bt>>6&63,F[G++]=128|Bt&63}}return F[G]=0,G-ht}function $n(T,F,G){return ro(T,n(),F,G)}function zo(T){for(var F=0,G=0;G<T.length;++G){var Y=T.charCodeAt(G);Y>=55296&&Y<=57343&&(Y=65536+((Y&1023)<<10)|T.charCodeAt(++G)&1023),Y<=127?++F:Y<=2047?F+=2:Y<=65535?F+=3:F+=4}return F}var no=typeof TextDecoder!="undefined"?new Fr("utf-16le"):void 0;function gl(T,F){e().set(T,F)}function cu(T,F,G){for(var Y=0;Y<T.length;++Y)e()[F++>>0]=T.charCodeAt(Y);G||(e()[F>>0]=0)}function Wc(T,F){return T%F>0&&(T+=F-T%F),T}var gr,Uc,Hc,ef,vg,Cg,IT,Ig,Sg;v&&(gr=l.buffer);function Bo(T){gr=T,l.HEAP8=Uc=new Int8Array(T),l.HEAP16=ef=new Int16Array(T),l.HEAP32=Cg=new Int32Array(T),l.HEAPU8=Hc=new Uint8Array(T),l.HEAPU16=vg=new Uint16Array(T),l.HEAPU32=IT=new Uint32Array(T),l.HEAPF32=Ig=new Float32Array(T),l.HEAPF64=Sg=new Float64Array(T)}var kg=l.INITIAL_MEMORY||16777216;if(v)Ot=l.wasmMemory,gr=l.buffer;else if(l.wasmMemory)Ot=l.wasmMemory;else if(Ot=new WebAssembly.Memory({initial:kg/65536,maximum:32768,shared:!0}),!(Ot.buffer instanceof SharedArrayBuffer))throw X("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"),w&&console.log("(on node you may need: --experimental-wasm-threads --experimental-wasm-bulk-memory and also use a recent version)"),Error("bad memory");Ot&&(gr=Ot.buffer),kg=gr.byteLength,Bo(gr);var Un,qc=[],xl=[],qb=[],Ng=[],pu=!1,Kb=!1,Tg=0;function mu(){return ce||Tg>0}function en(){if(l.preRun)for(typeof l.preRun=="function"&&(l.preRun=[l.preRun]);l.preRun.length;)ST(l.preRun.shift());Dg(qc)}function rf(){pu=!0,!v&&Dg(xl)}function jb(){v||(Ht.terminateAllThreads(),Kb=!0)}function Xb(){if(!v){if(l.postRun)for(typeof l.postRun=="function"&&(l.postRun=[l.postRun]);l.postRun.length;)nf(l.postRun.shift());Dg(Ng)}}function ST(T){qc.unshift(T)}function kT(T){xl.unshift(T)}function nf(T){Ng.unshift(T)}var yl=0,_g=null,Vo=null;function of(T){yl++,l.monitorRunDependencies&&l.monitorRunDependencies(yl)}function NT(T){if(yl--,l.monitorRunDependencies&&l.monitorRunDependencies(yl),yl==0&&(_g!==null&&(clearInterval(_g),_g=null),Vo)){var F=Vo;Vo=null,F()}}l.preloadedImages={},l.preloadedAudios={};function Kc(T){v?postMessage({cmd:"onAbort",arg:T}):l.onAbort&&l.onAbort(T),T="Aborted("+T+")",X(T),Ie=!0,ae=1,T+=". Build with -s ASSERTIONS=1 for more info.";var F=new WebAssembly.RuntimeError(T);throw p(F),F}var Yb="data:application/octet-stream;base64,";function sf(T){return T.startsWith(Yb)}function Eg(T){return T.startsWith("file://")}var rn;rn="tfjs-backend-wasm-threaded-simd.wasm",sf(rn)||(rn=E(rn));function Ag(T){try{if(T==rn&&Kt)return new Uint8Array(Kt);if(R)return R(T);throw"both async and sync fetching of the wasm failed"}catch(F){Kc(F)}}function jc(){if(!Kt&&(y||b)){if(typeof fetch=="function"&&!Eg(rn))return fetch(rn,{credentials:"same-origin"}).then(function(T){if(!T.ok)throw"failed to load wasm binary file at '"+rn+"'";return T.arrayBuffer()}).catch(function(){return Ag(rn)});if($)return new Promise(function(T,F){$(rn,function(G){T(new Uint8Array(G))},F)})}return Promise.resolve().then(function(){return Ag(rn)})}function Zb(){var T={env:Wg,wasi_snapshot_preview1:Wg};function F(Nt,Bt){var or=Nt.exports;if(l.asm=or,ow(l.asm.emscripten_tls_init),Un=l.asm.__indirect_function_table,kT(l.asm.__wasm_call_ctors),Ae=Bt,!v){var Uo=Ht.unusedWorkers.length;Ht.unusedWorkers.forEach(function(Ho){Ht.loadWasmModuleToWorker(Ho,function(){--Uo||NT("wasm-instantiate")})})}}v||of("wasm-instantiate");function G(Nt){F(Nt.instance,Nt.module)}function Y(Nt){return jc().then(function(Bt){return WebAssembly.instantiate(Bt,T)}).then(function(Bt){return Bt}).then(Nt,function(Bt){X("failed to asynchronously prepare wasm: "+Bt),Kc(Bt)})}function ht(){return!Kt&&typeof WebAssembly.instantiateStreaming=="function"&&!sf(rn)&&!Eg(rn)&&typeof fetch=="function"?fetch(rn,{credentials:"same-origin"}).then(function(Nt){var Bt=WebAssembly.instantiateStreaming(Nt,T);return Bt.then(G,function(or){return X("wasm streaming compile failed: "+or),X("falling back to ArrayBuffer instantiation"),Y(G)})}):Y(G)}if(l.instantiateWasm)try{var wt=l.instantiateWasm(T,F);return wt}catch(Nt){return X("Module.instantiateWasm callback failed with error: "+Nt),!1}return ht().catch(p),{}}var TT,_T,Jb={};function Dg(T){for(;T.length>0;){var F=T.shift();if(typeof F=="function"){F(l);continue}var G=F.func;typeof G=="number"?F.arg===void 0?Yc(G)():Yc(G)(F.arg):G(F.arg===void 0?null:F.arg)}}function Xc(T){var F=vv(),G=T();return jg(F),G}function bW(T){return T}function ET(T){var F=/\b_Z[\w\d_]+/g;return T.replace(F,function(G){var Y=G;return G===Y?G:Y+" ["+G+"]"})}function Qb(T){i()[T>>2]=0;var F=Ht.pthreads[T];delete Ht.pthreads[T],F.worker.terminate(),wv(T),Ht.runningWorkers.splice(Ht.runningWorkers.indexOf(F.worker),1),F.worker.pthread=void 0}function tw(T){var F=Ht.pthreads[T];F.worker.postMessage({cmd:"cancel"})}function $g(T){var F=Ht.pthreads[T];if(F){i()[T>>2]=0;var G=F.worker;Ht.returnWorkerToPool(G)}}function Rg(T){lU(T)}function ew(T){if(T instanceof mf||T=="unwind")return ae;g(1,T)}var Ht={unusedWorkers:[],runningWorkers:[],tlsInitFunctions:[],init:function(){v?Ht.initWorker():Ht.initMainThread()},initMainThread:function(){for(var T=8,F=0;F<T;++F)Ht.allocateUnusedWorker()},initWorker:function(){ce=!1},pthreads:{},setExitStatus:function(T){ae=T},terminateAllThreads:function(){for(var T in Ht.pthreads){var F=Ht.pthreads[T];F&&F.worker&&Ht.returnWorkerToPool(F.worker)}for(var G=0;G<Ht.unusedWorkers.length;++G){var Y=Ht.unusedWorkers[G];Y.terminate()}Ht.unusedWorkers=[]},returnWorkerToPool:function(T){Ht.runWithoutMainThreadQueuedCalls(function(){delete Ht.pthreads[T.pthread.threadInfoStruct],Ht.unusedWorkers.push(T),Ht.runningWorkers.splice(Ht.runningWorkers.indexOf(T),1),wv(T.pthread.threadInfoStruct),T.pthread=void 0})},runWithoutMainThreadQueuedCalls:function(T){i()[GT>>2]=0;try{T()}finally{i()[GT>>2]=1}},receiveObjectTransfer:function(T){},threadInit:function(){for(var T in Ht.tlsInitFunctions)Ht.tlsInitFunctions[T]()},loadWasmModuleToWorker:function(T,F){T.onmessage=G=>{var Y=G.data,ht=Y.cmd;if(T.pthread&&(Ht.currentProxiedOperationCallerThread=T.pthread.threadInfoStruct),Y.targetThread&&Y.targetThread!=Kg()){var wt=Ht.pthreads[Y.targetThread];wt?wt.worker.postMessage(Y,Y.transferList):X('Internal error! Worker sent a message "'+ht+'" to target pthread '+Y.targetThread+", but that thread no longer exists!"),Ht.currentProxiedOperationCallerThread=void 0;return}ht==="processQueuedMainThreadWork"?MT():ht==="spawnThread"?Og(Y):ht==="cleanupThread"?$g(Y.thread):ht==="killThread"?Qb(Y.thread):ht==="cancelThread"?tw(Y.thread):ht==="loaded"?(T.loaded=!0,F&&F(T),T.runPthread&&(T.runPthread(),delete T.runPthread)):ht==="print"?rt("Thread "+Y.threadId+": "+Y.text):ht==="printErr"?X("Thread "+Y.threadId+": "+Y.text):ht==="alert"?alert("Thread "+Y.threadId+": "+Y.text):Y.target==="setimmediate"?T.postMessage(Y):ht==="onAbort"?l.onAbort&&l.onAbort(Y.arg):X("worker sent an unknown command "+ht),Ht.currentProxiedOperationCallerThread=void 0},T.onerror=G=>{var Y="worker sent an error!";throw X(Y+" "+G.filename+":"+G.lineno+": "+G.message),G},w&&(T.on("message",function(G){T.onmessage({data:G})}),T.on("error",function(G){T.onerror(G)}),T.on("detachedExit",function(){})),T.postMessage({cmd:"load",urlOrBlob:l.mainScriptUrlOrBlob||r,wasmMemory:Ot,wasmModule:Ae})},allocateUnusedWorker:function(){var T=E("tfjs-backend-wasm-threaded-simd.worker.js");Ht.unusedWorkers.push(new Worker(T))},getNewWorker:function(){return Ht.unusedWorkers.length==0&&(Ht.allocateUnusedWorker(),Ht.loadWasmModuleToWorker(Ht.unusedWorkers[0])),Ht.unusedWorkers.pop()}};function rw(){var T=Kg(),F=i()[T+44>>2],G=i()[T+48>>2],Y=F-G;VT(F,Y),jg(F)}l.establishStackSpace=rw;function Fg(T){if(v)return hu(1,0,T);try{Rg(T)}catch(F){ew(F)}}var fu=[];function Yc(T){var F=fu[T];return F||(T>=fu.length&&(fu.length=T+1),fu[T]=F=Un.get(T)),F}function nw(T,F){return Yc(T)(F)}l.invokeEntryPoint=nw;function AT(){var T=new Error;if(!T.stack){try{throw new Error}catch(F){T=F}if(!T.stack)return"(no stack trace available)"}return T.stack.toString()}function ow(T,F,G){Ht.tlsInitFunctions.push(T)}function DT(T,F){Un.set(T,F),fu[T]=F}var du;w?du=()=>{var T=process.hrtime();return T[0]*1e3+T[1]/1e6}:v?du=()=>performance.now()-l.__performance_now_clock_drift:du=()=>performance.now();var sw=!0;function iw(T){return i()[PT()>>2]=T,T}function aw(T,F){var G;if(T===0)G=Date.now();else if((T===1||T===4)&&sw)G=du();else return iw(28),-1;return i()[F>>2]=G/1e3|0,i()[F+4>>2]=G%1e3*1e3*1e3|0,0}function lw(T,F){return aw(T,F)}function uw(T){LT(T,!b,1,!y),Ht.threadInit()}function cw(T){v?postMessage({cmd:"cleanupThread",thread:T}):$g(T)}function Og(T){var F=Ht.getNewWorker();if(!F)return 6;Ht.runningWorkers.push(F);var G=Ht.pthreads[T.pthread_ptr]={worker:F,threadInfoStruct:T.pthread_ptr};F.pthread=G;var Y={cmd:"run",start_routine:T.startRoutine,arg:T.arg,threadInfoStruct:T.pthread_ptr};return F.runPthread=()=>{Y.time=performance.now(),F.postMessage(Y,T.transferList)},F.loaded&&(F.runPthread(),delete F.runPthread),0}function pw(T,F,G,Y){if(typeof SharedArrayBuffer=="undefined")return X("Current environment does not support SharedArrayBuffer, pthreads are not available!"),6;var ht=[],wt=0;if(v&&(ht.length===0||wt))return zT(687865856,T,F,G,Y);if(wt)return wt;var Nt={startRoutine:G,pthread_ptr:T,arg:Y,transferList:ht};return v?(Nt.cmd="spawnThread",postMessage(Nt,ht),0):Og(Nt)}function mw(){return 2097152}function fw(T,F){if(T==F)postMessage({cmd:"processQueuedMainThreadWork"});else if(v)postMessage({targetThread:T,cmd:"processThreadQueue"});else{var G=Ht.pthreads[T],Y=G&&G.worker;if(!Y)return;Y.postMessage({cmd:"processThreadQueue"})}return 1}function dw(){Kc("")}function hw(){w||b||st("Blocking on the main thread is very dangerous, see https://emscripten.org/docs/porting/pthreads.html#blocking-on-the-main-browser-thread")}function Pg(){return 2147483648}function gw(T,F,G){n().copyWithin(T,F,F+G)}function xw(){return w?pW().cpus().length:navigator.hardwareConcurrency}function hu(T,F){var G=arguments.length-2,Y=arguments;return Xc(function(){for(var ht=G,wt=rp(ht*8),Nt=wt>>3,Bt=0;Bt<G;Bt++){var or=Y[2+Bt];u()[Nt+Bt]=or}return BT(T,ht,wt,F)})}var af=[];function yw(T,F,G){af.length=F;for(var Y=G>>3,ht=0;ht<F;ht++)af[ht]=u()[Y+ht];var wt=T<0,Nt=wt?Jb[-T-1]:Mw[T];return Nt.apply(null,af)}function bw(T){try{return Ot.grow(T-gr.byteLength+65535>>>16),Bo(Ot.buffer),1}catch(F){}}function ww(T){var F=n().length;if(T=T>>>0,T<=F)return!1;var G=Pg();if(T>G)return!1;for(var Y=1;Y<=4;Y*=2){var ht=F*(1+.2/Y);ht=Math.min(ht,T+100663296);var wt=Math.min(G,Wc(Math.max(T,ht),65536)),Nt=bw(wt);if(Nt)return!0}return!1}var ee={inEventHandler:0,removeAllEventListeners:function(){for(var T=ee.eventHandlers.length-1;T>=0;--T)ee._removeHandler(T);ee.eventHandlers=[],ee.deferredCalls=[]},registerRemoveEventListeners:function(){ee.removeEventListenersRegistered||(qb.push(ee.removeAllEventListeners),ee.removeEventListenersRegistered=!0)},deferredCalls:[],deferCall:function(T,F,G){function Y(Nt,Bt){if(Nt.length!=Bt.length)return!1;for(var or in Nt)if(Nt[or]!=Bt[or])return!1;return!0}for(var ht in ee.deferredCalls){var wt=ee.deferredCalls[ht];if(wt.targetFunction==T&&Y(wt.argsList,G))return}ee.deferredCalls.push({targetFunction:T,precedence:F,argsList:G}),ee.deferredCalls.sort(function(Nt,Bt){return Nt.precedence<Bt.precedence})},removeDeferredCalls:function(T){for(var F=0;F<ee.deferredCalls.length;++F)ee.deferredCalls[F].targetFunction==T&&(ee.deferredCalls.splice(F,1),--F)},canPerformEventHandlerRequests:function(){return ee.inEventHandler&&ee.currentEventHandler.allowsDeferredCalls},runDeferredCalls:function(){if(!!ee.canPerformEventHandlerRequests())for(var T=0;T<ee.deferredCalls.length;++T){var F=ee.deferredCalls[T];ee.deferredCalls.splice(T,1),--T,F.targetFunction.apply(null,F.argsList)}},eventHandlers:[],removeAllHandlersOnTarget:function(T,F){for(var G=0;G<ee.eventHandlers.length;++G)ee.eventHandlers[G].target==T&&(!F||F==ee.eventHandlers[G].eventTypeString)&&ee._removeHandler(G--)},_removeHandler:function(T){var F=ee.eventHandlers[T];F.target.removeEventListener(F.eventTypeString,F.eventListenerFunc,F.useCapture),ee.eventHandlers.splice(T,1)},registerOrRemoveHandler:function(T){var F=function(ht){++ee.inEventHandler,ee.currentEventHandler=T,ee.runDeferredCalls(),T.handlerFunc(ht),ee.runDeferredCalls(),--ee.inEventHandler};if(T.callbackfunc)T.eventListenerFunc=F,T.target.addEventListener(T.eventTypeString,F,T.useCapture),ee.eventHandlers.push(T),ee.registerRemoveEventListeners();else for(var G=0;G<ee.eventHandlers.length;++G)ee.eventHandlers[G].target==T.target&&ee.eventHandlers[G].eventTypeString==T.eventTypeString&&ee._removeHandler(G--)},queueEventHandlerOnThread_iiii:function(T,F,G,Y,ht){Xc(function(){var wt=rp(12);i()[wt>>2]=G,i()[wt+4>>2]=Y,i()[wt+8>>2]=ht,bv(T,637534208,F,Y,wt)})},getTargetThreadForEventCallback:function(T){switch(T){case 1:return 0;case 2:return Ht.currentProxiedOperationCallerThread;default:return T}},getNodeNameForTarget:function(T){return T?T==window?"#window":T==screen?"#screen":T&&T.nodeName?T.nodeName:"":""},fullscreenEnabled:function(){return document.fullscreenEnabled||document.webkitFullscreenEnabled}};function vw(T){var F=zo(T)+1,G=yv(F);return $n(T,G,F),G}function Cw(T,F,G,Y){Xc(function(){var ht=rp(12),wt=0;F&&(wt=vw(F)),i()[ht>>2]=wt,i()[ht+4>>2]=G,i()[ht+8>>2]=Y,bv(T,657457152,0,wt,ht)})}function Iw(T,F,G,Y){F=F?Hr(F):"",Cw(T,F,G,Y)}function Sw(T){return T>2?Hr(T):T}var kw=[0,typeof document!="undefined"?document:0,typeof window!="undefined"?window:0];function Nw(T){T=Sw(T);var F=kw[T]||(typeof document!="undefined"?document.querySelector(T):void 0);return F}function lf(T){return Nw(T)}function Mg(T,F,G){var Y=lf(T);if(!Y)return-4;if(Y.canvasSharedPtr&&(i()[Y.canvasSharedPtr>>2]=F,i()[Y.canvasSharedPtr+4>>2]=G),Y.offscreenCanvas||!Y.controlTransferredOffscreen){Y.offscreenCanvas&&(Y=Y.offscreenCanvas);var ht=!1;if(Y.GLctxObject&&Y.GLctxObject.GLctx){var wt=Y.GLctxObject.GLctx.getParameter(2978);ht=wt[0]===0&&wt[1]===0&&wt[2]===Y.width&&wt[3]===Y.height}Y.width=F,Y.height=G,ht&&Y.GLctxObject.GLctx.viewport(0,0,F,G)}else if(Y.canvasSharedPtr){var Nt=i()[Y.canvasSharedPtr+8>>2];return Iw(Nt,T,F,G),1}else return-4;return 0}function Lg(T,F,G){return v?hu(2,1,T,F,G):Mg(T,F,G)}function Tw(T,F,G){var Y=lf(T);return Y?Mg(T,F,G):Lg(T,F,G)}function _w(){throw"unwind"}function Ew(T){var F=T.getExtension("ANGLE_instanced_arrays");if(F)return T.vertexAttribDivisor=function(G,Y){F.vertexAttribDivisorANGLE(G,Y)},T.drawArraysInstanced=function(G,Y,ht,wt){F.drawArraysInstancedANGLE(G,Y,ht,wt)},T.drawElementsInstanced=function(G,Y,ht,wt,Nt){F.drawElementsInstancedANGLE(G,Y,ht,wt,Nt)},1}function Aw(T){var F=T.getExtension("OES_vertex_array_object");if(F)return T.createVertexArray=function(){return F.createVertexArrayOES()},T.deleteVertexArray=function(G){F.deleteVertexArrayOES(G)},T.bindVertexArray=function(G){F.bindVertexArrayOES(G)},T.isVertexArray=function(G){return F.isVertexArrayOES(G)},1}function Dw(T){var F=T.getExtension("WEBGL_draw_buffers");if(F)return T.drawBuffers=function(G,Y){F.drawBuffersWEBGL(G,Y)},1}function $w(T){return!!(T.multiDrawWebgl=T.getExtension("WEBGL_multi_draw"))}var nr={counter:1,buffers:[],programs:[],framebuffers:[],renderbuffers:[],textures:[],shaders:[],vaos:[],contexts:{},offscreenCanvases:{},queries:[],stringCache:{},unpackAlignment:4,recordError:function(F){nr.lastError||(nr.lastError=F)},getNewId:function(T){for(var F=nr.counter++,G=T.length;G<F;G++)T[G]=null;return F},getSource:function(T,F,G,Y){for(var ht="",wt=0;wt<F;++wt){var Nt=Y?i()[Y+wt*4>>2]:-1;ht+=Hr(i()[G+wt*4>>2],Nt<0?void 0:Nt)}return ht},createContext:function(T,F){T.getContextSafariWebGL2Fixed||(T.getContextSafariWebGL2Fixed=T.getContext,T.getContext=function(ht,wt){var Nt=T.getContextSafariWebGL2Fixed(ht,wt);return ht=="webgl"==Nt instanceof WebGLRenderingContext?Nt:null});var G=T.getContext("webgl",F);if(!G)return 0;var Y=nr.registerContext(G,F);return Y},registerContext:function(T,F){var G=yv(8);i()[G+4>>2]=Kg();var Y={handle:G,attributes:F,version:F.majorVersion,GLctx:T};return T.canvas&&(T.canvas.GLctxObject=Y),nr.contexts[G]=Y,(typeof F.enableExtensionsByDefault=="undefined"||F.enableExtensionsByDefault)&&nr.initExtensions(Y),G},makeContextCurrent:function(T){return nr.currentContext=nr.contexts[T],l.ctx=Gg=nr.currentContext&&nr.currentContext.GLctx,!(T&&!Gg)},getContext:function(T){return nr.contexts[T]},deleteContext:function(T){nr.currentContext===nr.contexts[T]&&(nr.currentContext=null),typeof ee=="object"&&ee.removeAllHandlersOnTarget(nr.contexts[T].GLctx.canvas),nr.contexts[T]&&nr.contexts[T].GLctx.canvas&&(nr.contexts[T].GLctx.canvas.GLctxObject=void 0),OT(nr.contexts[T].handle),nr.contexts[T]=null},initExtensions:function(T){if(T||(T=nr.currentContext),!T.initExtensionsDone){T.initExtensionsDone=!0;var F=T.GLctx;Ew(F),Aw(F),Dw(F),F.disjointTimerQueryExt=F.getExtension("EXT_disjoint_timer_query"),$w(F);var G=F.getSupportedExtensions()||[];G.forEach(function(Y){!Y.includes("lose_context")&&!Y.includes("debug")&&F.getExtension(Y)})}}},Rw=["default","low-power","high-performance"];function Fw(T,F){var G=F>>2,Y=i()[G+6],ht={alpha:!!i()[G+0],depth:!!i()[G+1],stencil:!!i()[G+2],antialias:!!i()[G+3],premultipliedAlpha:!!i()[G+4],preserveDrawingBuffer:!!i()[G+5],powerPreference:Rw[Y],failIfMajorPerformanceCaveat:!!i()[G+7],majorVersion:i()[G+8],minorVersion:i()[G+9],enableExtensionsByDefault:i()[G+10],explicitSwapControl:i()[G+11],proxyContextToMainThread:i()[G+12],renderViaOffscreenBackBuffer:i()[G+13]},wt=lf(T);if(!wt||ht.explicitSwapControl)return 0;var Nt=nr.createContext(wt,ht);return Nt}function Ow(T,F){return Fw(T,F)}var Zc={mappings:{},buffers:[null,[],[]],printChar:function(T,F){var G=Zc.buffers[T];F===0||F===10?((T===1?rt:X)(tn(G,0)),G.length=0):G.push(F)},varargs:void 0,get:function(){Zc.varargs+=4;var T=i()[Zc.varargs-4>>2];return T},getStr:function(T){var F=Hr(T);return F},get64:function(T,F){return T}};function zg(T){return v?hu(3,1,T):0}function Bg(T,F,G,Y,ht){if(v)return hu(4,1,T,F,G,Y,ht)}function Vg(T,F,G,Y){if(v)return hu(5,1,T,F,G,Y);for(var ht=0,wt=0;wt<G;wt++){var Nt=i()[F>>2],Bt=i()[F+4>>2];F+=8;for(var or=0;or<Bt;or++)Zc.printChar(T,n()[Nt+or]);ht+=Bt}return i()[Y>>2]=ht,0}function Pw(T){kt(T)}Ht.init();var Gg,Mw=[null,Fg,Lg,zg,Bg,Vg],$T=!1,Wg={__clock_gettime:lw,__emscripten_init_main_thread_js:uw,__emscripten_thread_cleanup:cw,__pthread_create_js:pw,_emscripten_default_pthread_stack_size:mw,_emscripten_notify_thread_queue:fw,abort:dw,emscripten_check_blocking_allowed:hw,emscripten_get_heap_max:Pg,emscripten_get_now:du,emscripten_memcpy_big:gw,emscripten_num_logical_cores:xw,emscripten_receive_on_main_thread_js:yw,emscripten_resize_heap:ww,emscripten_set_canvas_element_size:Tw,emscripten_unwind_to_js_event_loop:_w,emscripten_webgl_create_context:Ow,exit:Rg,fd_close:zg,fd_seek:Bg,fd_write:Vg,memory:Ot||l.wasmMemory,setTempRet0:Pw},RT=Zb(),Lw=l.___wasm_call_ctors=function(){return(Lw=l.___wasm_call_ctors=l.asm.__wasm_call_ctors).apply(null,arguments)},zw=l._init=function(){return(zw=l._init=l.asm.init).apply(null,arguments)},Bw=l._init_with_threads_count=function(){return(Bw=l._init_with_threads_count=l.asm.init_with_threads_count).apply(null,arguments)},Vw=l._get_threads_count=function(){return(Vw=l._get_threads_count=l.asm.get_threads_count).apply(null,arguments)},Gw=l._register_tensor=function(){return(Gw=l._register_tensor=l.asm.register_tensor).apply(null,arguments)},Ww=l._dispose_data=function(){return(Ww=l._dispose_data=l.asm.dispose_data).apply(null,arguments)},Uw=l._dispose=function(){return(Uw=l._dispose=l.asm.dispose).apply(null,arguments)},Hw=l._Abs=function(){return(Hw=l._Abs=l.asm.Abs).apply(null,arguments)},qw=l._Add=function(){return(qw=l._Add=l.asm.Add).apply(null,arguments)},Kw=l._AddN=function(){return(Kw=l._AddN=l.asm.AddN).apply(null,arguments)},jw=l._All=function(){return(jw=l._All=l.asm.All).apply(null,arguments)},Xw=l._Any=function(){return(Xw=l._Any=l.asm.Any).apply(null,arguments)},Yw=l._ArgMax=function(){return(Yw=l._ArgMax=l.asm.ArgMax).apply(null,arguments)},Zw=l._AvgPool=function(){return(Zw=l._AvgPool=l.asm.AvgPool).apply(null,arguments)},Jw=l._BatchMatMul=function(){return(Jw=l._BatchMatMul=l.asm.BatchMatMul).apply(null,arguments)},Qw=l._Ceil=function(){return(Qw=l._Ceil=l.asm.Ceil).apply(null,arguments)},tv=l._ClipByValue=function(){return(tv=l._ClipByValue=l.asm.ClipByValue).apply(null,arguments)},ev=l._Conv2D=function(){return(ev=l._Conv2D=l.asm.Conv2D).apply(null,arguments)},rv=l._Conv2DBackpropInput=function(){return(rv=l._Conv2DBackpropInput=l.asm.Conv2DBackpropInput).apply(null,arguments)},nv=l._Cos=function(){return(nv=l._Cos=l.asm.Cos).apply(null,arguments)},ov=l._Cosh=function(){return(ov=l._Cosh=l.asm.Cosh).apply(null,arguments)},sv=l._CropAndResize=function(){return(sv=l._CropAndResize=l.asm.CropAndResize).apply(null,arguments)},iv=l._Cumprod=function(){return(iv=l._Cumprod=l.asm.Cumprod).apply(null,arguments)},av=l._Cumsum=function(){return(av=l._Cumsum=l.asm.Cumsum).apply(null,arguments)},lv=l._DepthToSpace=function(){return(lv=l._DepthToSpace=l.asm.DepthToSpace).apply(null,arguments)},uv=l._DepthwiseConv2dNative=function(){return(uv=l._DepthwiseConv2dNative=l.asm.DepthwiseConv2dNative).apply(null,arguments)},cv=l._Elu=function(){return(cv=l._Elu=l.asm.Elu).apply(null,arguments)},pv=l._Equal=function(){return(pv=l._Equal=l.asm.Equal).apply(null,arguments)},mv=l._Exp=function(){return(mv=l._Exp=l.asm.Exp).apply(null,arguments)},fv=l._FlipLeftRight=function(){return(fv=l._FlipLeftRight=l.asm.FlipLeftRight).apply(null,arguments)},Ug=l._Floor=function(){return(Ug=l._Floor=l.asm.Floor).apply(null,arguments)},Hg=l._FloorDiv=function(){return(Hg=l._FloorDiv=l.asm.FloorDiv).apply(null,arguments)},uf=l._FusedBatchNorm=function(){return(uf=l._FusedBatchNorm=l.asm.FusedBatchNorm).apply(null,arguments)},dv=l._FusedConv2D=function(){return(dv=l._FusedConv2D=l.asm.FusedConv2D).apply(null,arguments)},hv=l._FusedDepthwiseConv2D=function(){return(hv=l._FusedDepthwiseConv2D=l.asm.FusedDepthwiseConv2D).apply(null,arguments)},Jc=l._Gather=function(){return(Jc=l._Gather=l.asm.Gather).apply(null,arguments)},cf=l._GatherNd=function(){return(cf=l._GatherNd=l.asm.GatherNd).apply(null,arguments)},pf=l._Greater=function(){return(pf=l._Greater=l.asm.Greater).apply(null,arguments)},FT=l._GreaterEqual=function(){return(FT=l._GreaterEqual=l.asm.GreaterEqual).apply(null,arguments)},Qc=l._LeakyRelu=function(){return(Qc=l._LeakyRelu=l.asm.LeakyRelu).apply(null,arguments)},tp=l._Less=function(){return(tp=l._Less=l.asm.Less).apply(null,arguments)},gv=l._LessEqual=function(){return(gv=l._LessEqual=l.asm.LessEqual).apply(null,arguments)},H=l._Log=function(){return(H=l._Log=l.asm.Log).apply(null,arguments)},tt=l._LogicalAnd=function(){return(tt=l._LogicalAnd=l.asm.LogicalAnd).apply(null,arguments)},gt=l._Max=function(){return(gt=l._Max=l.asm.Max).apply(null,arguments)},Dt=l._MaxPool=function(){return(Dt=l._MaxPool=l.asm.MaxPool).apply(null,arguments)},fe=l._Maximum=function(){return(fe=l._Maximum=l.asm.Maximum).apply(null,arguments)},he=l._Mean=function(){return(he=l._Mean=l.asm.Mean).apply(null,arguments)},oe=l._Min=function(){return(oe=l._Min=l.asm.Min).apply(null,arguments)},Qt=l._Minimum=function(){return(Qt=l._Minimum=l.asm.Minimum).apply(null,arguments)},xr=l._MirrorPad=function(){return(xr=l._MirrorPad=l.asm.MirrorPad).apply(null,arguments)},Go=l._Multiply=function(){return(Go=l._Multiply=l.asm.Multiply).apply(null,arguments)},Wo=l._Neg=function(){return(Wo=l._Neg=l.asm.Neg).apply(null,arguments)},ep=l._NonMaxSuppressionV3=function(){return(ep=l._NonMaxSuppressionV3=l.asm.NonMaxSuppressionV3).apply(null,arguments)},gu=l._NonMaxSuppressionV4=function(){return(gu=l._NonMaxSuppressionV4=l.asm.NonMaxSuppressionV4).apply(null,arguments)},xv=l._NonMaxSuppressionV5=function(){return(xv=l._NonMaxSuppressionV5=l.asm.NonMaxSuppressionV5).apply(null,arguments)},nn=l._NotEqual=function(){return(nn=l._NotEqual=l.asm.NotEqual).apply(null,arguments)},bl=l._OneHot=function(){return(bl=l._OneHot=l.asm.OneHot).apply(null,arguments)},qg=l._PadV2=function(){return(qg=l._PadV2=l.asm.PadV2).apply(null,arguments)},wW=l._Pow=function(){return(wW=l._Pow=l.asm.Pow).apply(null,arguments)},vW=l._Prelu=function(){return(vW=l._Prelu=l.asm.Prelu).apply(null,arguments)},CW=l._Prod=function(){return(CW=l._Prod=l.asm.Prod).apply(null,arguments)},IW=l._RealDiv=function(){return(IW=l._RealDiv=l.asm.RealDiv).apply(null,arguments)},SW=l._Relu=function(){return(SW=l._Relu=l.asm.Relu).apply(null,arguments)},kW=l._Relu6=function(){return(kW=l._Relu6=l.asm.Relu6).apply(null,arguments)},NW=l._ResizeBilinear=function(){return(NW=l._ResizeBilinear=l.asm.ResizeBilinear).apply(null,arguments)},TW=l._Reverse=function(){return(TW=l._Reverse=l.asm.Reverse).apply(null,arguments)},_W=l._RotateWithOffset=function(){return(_W=l._RotateWithOffset=l.asm.RotateWithOffset).apply(null,arguments)},EW=l._Round=function(){return(EW=l._Round=l.asm.Round).apply(null,arguments)},AW=l._Rsqrt=function(){return(AW=l._Rsqrt=l.asm.Rsqrt).apply(null,arguments)},DW=l._ScatterNd=function(){return(DW=l._ScatterNd=l.asm.ScatterNd).apply(null,arguments)},$W=l._SelectV2=function(){return($W=l._SelectV2=l.asm.SelectV2).apply(null,arguments)},RW=l._Sigmoid=function(){return(RW=l._Sigmoid=l.asm.Sigmoid).apply(null,arguments)},FW=l._Sin=function(){return(FW=l._Sin=l.asm.Sin).apply(null,arguments)},OW=l._Softmax=function(){return(OW=l._Softmax=l.asm.Softmax).apply(null,arguments)},PW=l._SparseFillEmptyRows=function(){return(PW=l._SparseFillEmptyRows=l.asm.SparseFillEmptyRows).apply(null,arguments)},MW=l._SparseReshape=function(){return(MW=l._SparseReshape=l.asm.SparseReshape).apply(null,arguments)},LW=l._SparseSegmentReduction=function(){return(LW=l._SparseSegmentReduction=l.asm.SparseSegmentReduction).apply(null,arguments)},zW=l._Sqrt=function(){return(zW=l._Sqrt=l.asm.Sqrt).apply(null,arguments)},BW=l._Square=function(){return(BW=l._Square=l.asm.Square).apply(null,arguments)},VW=l._SquaredDifference=function(){return(VW=l._SquaredDifference=l.asm.SquaredDifference).apply(null,arguments)},GW=l._Step=function(){return(GW=l._Step=l.asm.Step).apply(null,arguments)},WW=l._StridedSlice=function(){return(WW=l._StridedSlice=l.asm.StridedSlice).apply(null,arguments)},UW=l._Sub=function(){return(UW=l._Sub=l.asm.Sub).apply(null,arguments)},HW=l._Sum=function(){return(HW=l._Sum=l.asm.Sum).apply(null,arguments)},qW=l._Tan=function(){return(qW=l._Tan=l.asm.Tan).apply(null,arguments)},KW=l._Tanh=function(){return(KW=l._Tanh=l.asm.Tanh).apply(null,arguments)},jW=l._Tile=function(){return(jW=l._Tile=l.asm.Tile).apply(null,arguments)},XW=l._TopK=function(){return(XW=l._TopK=l.asm.TopK).apply(null,arguments)},YW=l._Transform=function(){return(YW=l._Transform=l.asm.Transform).apply(null,arguments)},ZW=l._Transpose=function(){return(ZW=l._Transpose=l.asm.Transpose).apply(null,arguments)},JW=l.__FusedMatMul=function(){return(JW=l.__FusedMatMul=l.asm._FusedMatMul).apply(null,arguments)},yv=l._malloc=function(){return(yv=l._malloc=l.asm.malloc).apply(null,arguments)},OT=l._free=function(){return(OT=l._free=l.asm.free).apply(null,arguments)},QW=l._emscripten_tls_init=function(){return(QW=l._emscripten_tls_init=l.asm.emscripten_tls_init).apply(null,arguments)},PT=l.___errno_location=function(){return(PT=l.___errno_location=l.asm.__errno_location).apply(null,arguments)},Kg=l._pthread_self=function(){return(Kg=l._pthread_self=l.asm.pthread_self).apply(null,arguments)},MT=l._emscripten_main_thread_process_queued_calls=function(){return(MT=l._emscripten_main_thread_process_queued_calls=l.asm.emscripten_main_thread_process_queued_calls).apply(null,arguments)},tU=l.__emscripten_thread_crashed=function(){return(tU=l.__emscripten_thread_crashed=l.asm._emscripten_thread_crashed).apply(null,arguments)},LT=l.__emscripten_thread_init=function(){return(LT=l.__emscripten_thread_init=l.asm._emscripten_thread_init).apply(null,arguments)},eU=l._emscripten_current_thread_process_queued_calls=function(){return(eU=l._emscripten_current_thread_process_queued_calls=l.asm.emscripten_current_thread_process_queued_calls).apply(null,arguments)},rU=l._emscripten_main_browser_thread_id=function(){return(rU=l._emscripten_main_browser_thread_id=l.asm.emscripten_main_browser_thread_id).apply(null,arguments)},nU=l._emscripten_sync_run_in_main_thread_2=function(){return(nU=l._emscripten_sync_run_in_main_thread_2=l.asm.emscripten_sync_run_in_main_thread_2).apply(null,arguments)},zT=l._emscripten_sync_run_in_main_thread_4=function(){return(zT=l._emscripten_sync_run_in_main_thread_4=l.asm.emscripten_sync_run_in_main_thread_4).apply(null,arguments)},BT=l._emscripten_run_in_main_runtime_thread_js=function(){return(BT=l._emscripten_run_in_main_runtime_thread_js=l.asm.emscripten_run_in_main_runtime_thread_js).apply(null,arguments)},bv=l._emscripten_dispatch_to_thread_=function(){return(bv=l._emscripten_dispatch_to_thread_=l.asm.emscripten_dispatch_to_thread_).apply(null,arguments)},wv=l.__emscripten_thread_free_data=function(){return(wv=l.__emscripten_thread_free_data=l.asm._emscripten_thread_free_data).apply(null,arguments)},oU=l.__emscripten_thread_exit=function(){return(oU=l.__emscripten_thread_exit=l.asm._emscripten_thread_exit).apply(null,arguments)},sU=l._memalign=function(){return(sU=l._memalign=l.asm.memalign).apply(null,arguments)},VT=l._emscripten_stack_set_limits=function(){return(VT=l._emscripten_stack_set_limits=l.asm.emscripten_stack_set_limits).apply(null,arguments)},vv=l.stackSave=function(){return(vv=l.stackSave=l.asm.stackSave).apply(null,arguments)},jg=l.stackRestore=function(){return(jg=l.stackRestore=l.asm.stackRestore).apply(null,arguments)},rp=l.stackAlloc=function(){return(rp=l.stackAlloc=l.asm.stackAlloc).apply(null,arguments)},iU=l.dynCall_iijjiiii=function(){return(iU=l.dynCall_iijjiiii=l.asm.dynCall_iijjiiii).apply(null,arguments)},aU=l.dynCall_jiji=function(){return(aU=l.dynCall_jiji=l.asm.dynCall_jiji).apply(null,arguments)},GT=l.__emscripten_allow_main_runtime_queued_calls=21464;l.cwrap=ze,l.keepRuntimeAlive=mu,l.PThread=Ht,l.PThread=Ht,l.wasmMemory=Ot,l.ExitStatus=mf;var Xg;function mf(T){this.name="ExitStatus",this.message="Program terminated with exit("+T+")",this.status=T}Vo=function T(){Xg||Cv(),Xg||(Vo=T)};function Cv(T){if(T=T||d,yl>0)return;if(v){c(l),rf(),postMessage({cmd:"loaded"});return}if(en(),yl>0)return;function F(){Xg||(Xg=!0,l.calledRun=!0,!Ie&&(rf(),c(l),l.onRuntimeInitialized&&l.onRuntimeInitialized(),Xb()))}l.setStatus?(l.setStatus("Running..."),setTimeout(function(){setTimeout(function(){l.setStatus("")},1),F()},1)):F()}l.run=Cv;function lU(T,F){if(ae=T,!F&&v)throw Fg(T),"unwind";mu()||jb(),uU(T)}function uU(T){ae=T,mu()||(Ht.terminateAllThreads(),l.onExit&&l.onExit(T),Ie=!0),g(T,new mf(T))}if(l.preInit)for(typeof l.preInit=="function"&&(l.preInit=[l.preInit]);l.preInit.length>0;)l.preInit.pop()();Cv();var Yg;m&&(Yg={uncaughtException:process.listeners("uncaughtException").filter(function(T){return!m.uncaughtException.indexOf(T)>-1}),unhandledRejection:process.listeners("unhandledRejection").filter(function(T){return!m.unhandledRejection.indexOf(T)>-1})});var Zg;if(typeof WasmBackendModule!="undefined")Zg=WasmBackendModule;else if(typeof t!="undefined")Zg=t;else throw new Error("Could not find wasm module in post.js");if(Yg){var cU=Zg._dispose;Zg._dispose=function(){cU(),Yg.uncaughtException.forEach(function(T){process.removeListener("uncaughtException",T)}),Yg.unhandledRejection.forEach(function(T){process.removeListener("unhandledRejection",T)})}}return t.ready}})();typeof Gb=="object"&&typeof xT=="object"?xT.exports=gT:typeof define=="function"&&define.amd?define([],function(){return gT}):typeof Gb=="object"&&(Gb.WasmBackendModuleThreadedSimd=gT)});var dW=Oe((Wb,bT)=>{var yT=(()=>{var r=typeof document!="undefined"&&document.currentScript?document.currentScript.src:void 0;return typeof __filename!="undefined"&&(r=r||__filename),function(t){t=t||{};var e=typeof t!="undefined"?t:{},n,o;e.ready=new Promise(function(H,tt){n=H,o=tt});var s;typeof process!="undefined"&&process.listeners&&(s={uncaughtException:process.listeners("uncaughtException"),unhandledRejection:process.listeners("unhandledRejection")});var i=Object.assign({},e),a=[],u="./this.program",l=(H,tt)=>{throw tt},c=typeof window=="object",p=typeof importScripts=="function",m=typeof process=="object"&&typeof process.versions=="object"&&typeof process.versions.node=="string",f="";function d(H){return e.locateFile?e.locateFile(H,f):f+H}var h,g,y,b;function w(H){if(H instanceof cf)return;$("exiting due to exception: "+H)}var v,k,E;m?(p?f=xg().dirname(f)+"/":f=__dirname+"/",E=()=>{k||(v=By(),k=xg())},h=function(tt,gt){return E(),tt=k.normalize(tt),v.readFileSync(tt,gt?void 0:"utf8")},y=H=>{var tt=h(H,!0);return tt.buffer||(tt=new Uint8Array(tt)),tt},g=(H,tt,gt)=>{E(),H=k.normalize(H),v.readFile(H,function(Dt,fe){Dt?gt(Dt):tt(fe.buffer)})},process.argv.length>1&&(u=process.argv[1].replace(/\\/g,"/")),a=process.argv.slice(2),process.on("uncaughtException",function(H){if(!(H instanceof cf))throw H}),process.on("unhandledRejection",function(H){throw H}),l=(H,tt)=>{if(ef())throw process.exitCode=H,tt;w(tt),process.exit(H)},e.inspect=function(){return"[Emscripten Module object]"}):(c||p)&&(p?f=self.location.href:typeof document!="undefined"&&document.currentScript&&(f=document.currentScript.src),r&&(f=r),f.indexOf("blob:")!==0?f=f.substr(0,f.replace(/[?#].*/,"").lastIndexOf("/")+1):f="",h=H=>{var tt=new XMLHttpRequest;return tt.open("GET",H,!1),tt.send(null),tt.responseText},p&&(y=H=>{var tt=new XMLHttpRequest;return tt.open("GET",H,!1),tt.responseType="arraybuffer",tt.send(null),new Uint8Array(tt.response)}),g=(H,tt,gt)=>{var Dt=new XMLHttpRequest;Dt.open("GET",H,!0),Dt.responseType="arraybuffer",Dt.onload=()=>{if(Dt.status==200||Dt.status==0&&Dt.response){tt(Dt.response);return}gt()},Dt.onerror=gt,Dt.send(null)},b=H=>document.title=H);var D=e.print||console.log.bind(console),$=e.printErr||console.warn.bind(console);Object.assign(e,i),i=null,e.arguments&&(a=e.arguments),e.thisProgram&&(u=e.thisProgram),e.quit&&(l=e.quit);var R=4;function M(H){M.shown||(M.shown={}),M.shown[H]||(M.shown[H]=1,$(H))}function W(H,tt){if(typeof WebAssembly.Function=="function"){for(var gt={i:"i32",j:"i64",f:"f32",d:"f64"},Dt={parameters:[],results:tt[0]=="v"?[]:[gt[tt[0]]]},fe=1;fe<tt.length;++fe)Dt.parameters.push(gt[tt[fe]]);return new WebAssembly.Function(Dt,H)}var he=[1,0,1,96],oe=tt.slice(0,1),Qt=tt.slice(1),xr={i:127,j:126,f:125,d:124};he.push(Qt.length);for(var fe=0;fe<Qt.length;++fe)he.push(xr[Qt[fe]]);oe=="v"?he.push(0):he=he.concat([1,xr[oe]]),he[1]=he.length-2;var Go=new Uint8Array([0,97,115,109,1,0,0,0].concat(he,[2,7,1,1,101,1,102,0,0,7,5,1,1,102,0,0])),Wo=new WebAssembly.Module(Go),ep=new WebAssembly.Instance(Wo,{e:{f:H}}),gu=ep.exports.f;return gu}var U=[],q;function K(){if(U.length)return U.pop();try{no.grow(1)}catch(H){throw H instanceof RangeError?"Unable to grow wasm table. Set ALLOW_TABLE_GROWTH.":H}return no.length-1}function j(H,tt){for(var gt=H;gt<H+tt;gt++){var Dt=of(gt);Dt&&q.set(Dt,gt)}}var Q=0,rt=H=>{Q=H},X;e.wasmBinary&&(X=e.wasmBinary);var ot=e.noExitRuntime||!0;typeof WebAssembly!="object"&&pu("no native wasm support detected");var st,it=!1,ft;function at(H,tt){H||pu(tt)}function xt(H){var tt=e["_"+H];return tt}function dt(H,tt,gt,Dt,fe){var he={string:function(nn){var bl=0;if(nn!=null&&nn!==0){var qg=(nn.length<<2)+1;bl=uf(qg),ce(nn,bl,qg)}return bl},array:function(nn){var bl=uf(nn.length);return Ie(nn,bl),bl}};function oe(nn){return tt==="string"?qt(nn):tt==="boolean"?Boolean(nn):nn}var Qt=xt(H),xr=[],Go=0;if(Dt)for(var Wo=0;Wo<Dt.length;Wo++){var ep=he[gt[Wo]];ep?(Go===0&&(Go=Ug()),xr[Wo]=ep(Dt[Wo])):xr[Wo]=Dt[Wo]}var gu=Qt.apply(null,xr);function xv(nn){return Go!==0&&Hg(Go),oe(nn)}return gu=xv(gu),gu}function bt(H,tt,gt,Dt){gt=gt||[];var fe=gt.every(function(oe){return oe==="number"}),he=tt!=="string";return he&&fe&&!Dt?xt(H):function(){return dt(H,tt,gt,arguments,Dt)}}var kt=1,At=typeof TextDecoder!="undefined"?new TextDecoder("utf8"):void 0;function $t(H,tt,gt){for(var Dt=tt+gt,fe=tt;H[fe]&&!(fe>=Dt);)++fe;if(fe-tt>16&&H.subarray&&At)return At.decode(H.subarray(tt,fe));for(var he="";tt<fe;){var oe=H[tt++];if(!(oe&128)){he+=String.fromCharCode(oe);continue}var Qt=H[tt++]&63;if((oe&224)==192){he+=String.fromCharCode((oe&31)<<6|Qt);continue}var xr=H[tt++]&63;if((oe&240)==224?oe=(oe&15)<<12|Qt<<6|xr:oe=(oe&7)<<18|Qt<<12|xr<<6|H[tt++]&63,oe<65536)he+=String.fromCharCode(oe);else{var Go=oe-65536;he+=String.fromCharCode(55296|Go>>10,56320|Go&1023)}}return he}function qt(H,tt){return H?$t(ze,H,tt):""}function Kt(H,tt,gt,Dt){if(!(Dt>0))return 0;for(var fe=gt,he=gt+Dt-1,oe=0;oe<H.length;++oe){var Qt=H.charCodeAt(oe);if(Qt>=55296&&Qt<=57343){var xr=H.charCodeAt(++oe);Qt=65536+((Qt&1023)<<10)|xr&1023}if(Qt<=127){if(gt>=he)break;tt[gt++]=Qt}else if(Qt<=2047){if(gt+1>=he)break;tt[gt++]=192|Qt>>6,tt[gt++]=128|Qt&63}else if(Qt<=65535){if(gt+2>=he)break;tt[gt++]=224|Qt>>12,tt[gt++]=128|Qt>>6&63,tt[gt++]=128|Qt&63}else{if(gt+3>=he)break;tt[gt++]=240|Qt>>18,tt[gt++]=128|Qt>>12&63,tt[gt++]=128|Qt>>6&63,tt[gt++]=128|Qt&63}}return tt[gt]=0,gt-fe}function ce(H,tt,gt){return Kt(H,ze,tt,gt)}function Ot(H){for(var tt=0,gt=0;gt<H.length;++gt){var Dt=H.charCodeAt(gt);Dt>=55296&&Dt<=57343&&(Dt=65536+((Dt&1023)<<10)|H.charCodeAt(++gt)&1023),Dt<=127?++tt:Dt<=2047?tt+=2:Dt<=65535?tt+=3:tt+=4}return tt}var Ae=typeof TextDecoder!="undefined"?new TextDecoder("utf-16le"):void 0;function Ie(H,tt){Jr.set(H,tt)}function ae(H,tt,gt){for(var Dt=0;Dt<H.length;++Dt)Jr[tt++>>0]=H.charCodeAt(Dt);gt||(Jr[tt>>0]=0)}function je(H,tt){return H%tt>0&&(H+=tt-H%tt),H}var Fe,Jr,ze,Lo,Fr,Qr,tn,Hr,ro;function $n(H){Fe=H,e.HEAP8=Jr=new Int8Array(H),e.HEAP16=Lo=new Int16Array(H),e.HEAP32=Qr=new Int32Array(H),e.HEAPU8=ze=new Uint8Array(H),e.HEAPU16=Fr=new Uint16Array(H),e.HEAPU32=tn=new Uint32Array(H),e.HEAPF32=Hr=new Float32Array(H),e.HEAPF64=ro=new Float64Array(H)}var zo=e.INITIAL_MEMORY||16777216,no,gl=[],cu=[],Wc=[],gr=!1,Uc=!1,Hc=0;function ef(){return ot||Hc>0}function vg(){if(e.preRun)for(typeof e.preRun=="function"&&(e.preRun=[e.preRun]);e.preRun.length;)Sg(e.preRun.shift());nf(gl)}function Cg(){gr=!0,nf(cu)}function IT(){Uc=!0}function Ig(){if(e.postRun)for(typeof e.postRun=="function"&&(e.postRun=[e.postRun]);e.postRun.length;)kg(e.postRun.shift());nf(Wc)}function Sg(H){gl.unshift(H)}function Bo(H){cu.unshift(H)}function kg(H){Wc.unshift(H)}var Un=0,qc=null,xl=null;function qb(H){Un++,e.monitorRunDependencies&&e.monitorRunDependencies(Un)}function Ng(H){if(Un--,e.monitorRunDependencies&&e.monitorRunDependencies(Un),Un==0&&(qc!==null&&(clearInterval(qc),qc=null),xl)){var tt=xl;xl=null,tt()}}e.preloadedImages={},e.preloadedAudios={};function pu(H){e.onAbort&&e.onAbort(H),H="Aborted("+H+")",$(H),it=!0,ft=1,H+=". Build with -s ASSERTIONS=1 for more info.";var tt=new WebAssembly.RuntimeError(H);throw o(tt),tt}var Kb="data:application/octet-stream;base64,";function Tg(H){return H.startsWith(Kb)}function mu(H){return H.startsWith("file://")}var en;en="tfjs-backend-wasm.wasm",Tg(en)||(en=d(en));function rf(H){try{if(H==en&&X)return new Uint8Array(X);if(y)return y(H);throw"both async and sync fetching of the wasm failed"}catch(tt){pu(tt)}}function jb(){if(!X&&(c||p)){if(typeof fetch=="function"&&!mu(en))return fetch(en,{credentials:"same-origin"}).then(function(H){if(!H.ok)throw"failed to load wasm binary file at '"+en+"'";return H.arrayBuffer()}).catch(function(){return rf(en)});if(g)return new Promise(function(H,tt){g(en,function(gt){H(new Uint8Array(gt))},tt)})}return Promise.resolve().then(function(){return rf(en)})}function Xb(){var H={env:Xc,wasi_snapshot_preview1:Xc};function tt(oe,Qt){var xr=oe.exports;e.asm=xr,st=e.asm.memory,$n(st.buffer),no=e.asm.__indirect_function_table,Bo(e.asm.__wasm_call_ctors),Ng("wasm-instantiate")}qb("wasm-instantiate");function gt(oe){tt(oe.instance)}function Dt(oe){return jb().then(function(Qt){return WebAssembly.instantiate(Qt,H)}).then(function(Qt){return Qt}).then(oe,function(Qt){$("failed to asynchronously prepare wasm: "+Qt),pu(Qt)})}function fe(){return!X&&typeof WebAssembly.instantiateStreaming=="function"&&!Tg(en)&&!mu(en)&&typeof fetch=="function"?fetch(en,{credentials:"same-origin"}).then(function(oe){var Qt=WebAssembly.instantiateStreaming(oe,H);return Qt.then(gt,function(xr){return $("wasm streaming compile failed: "+xr),$("falling back to ArrayBuffer instantiation"),Dt(gt)})}):Dt(gt)}if(e.instantiateWasm)try{var he=e.instantiateWasm(H,tt);return he}catch(oe){return $("Module.instantiateWasm callback failed with error: "+oe),!1}return fe().catch(o),{}}var ST,kT;function nf(H){for(;H.length>0;){var tt=H.shift();if(typeof tt=="function"){tt(e);continue}var gt=tt.func;typeof gt=="number"?tt.arg===void 0?of(gt)():of(gt)(tt.arg):gt(tt.arg===void 0?null:tt.arg)}}function yl(H){return H}function _g(H){var tt=/\b_Z[\w\d_]+/g;return H.replace(tt,function(gt){var Dt=gt;return gt===Dt?gt:Dt+" ["+gt+"]"})}var Vo=[];function of(H){var tt=Vo[H];return tt||(H>=Vo.length&&(Vo.length=H+1),Vo[H]=tt=no.get(H)),tt}function NT(){var H=new Error;if(!H.stack){try{throw new Error}catch(tt){H=tt}if(!H.stack)return"(no stack trace available)"}return H.stack.toString()}function Kc(H,tt){no.set(H,tt),Vo[H]=tt}function Yb(){pu("")}function sf(){return 2147483648}function Eg(H,tt,gt){ze.copyWithin(H,tt,tt+gt)}function rn(H){try{return st.grow(H-Fe.byteLength+65535>>>16),$n(st.buffer),1}catch(tt){}}function Ag(H){var tt=ze.length;H=H>>>0;var gt=sf();if(H>gt)return!1;for(var Dt=1;Dt<=4;Dt*=2){var fe=tt*(1+.2/Dt);fe=Math.min(fe,H+100663296);var he=Math.min(gt,je(Math.max(H,fe),65536)),oe=rn(he);if(oe)return!0}return!1}var jc={mappings:{},buffers:[null,[],[]],printChar:function(H,tt){var gt=jc.buffers[H];tt===0||tt===10?((H===1?D:$)($t(gt,0)),gt.length=0):gt.push(tt)},varargs:void 0,get:function(){jc.varargs+=4;var H=Qr[jc.varargs-4>>2];return H},getStr:function(H){var tt=qt(H);return tt},get64:function(H,tt){return H}};function Zb(H){return 0}function TT(H,tt,gt,Dt,fe){}function _T(H,tt,gt,Dt){for(var fe=0,he=0;he<gt;he++){var oe=Qr[tt>>2],Qt=Qr[tt+4>>2];tt+=8;for(var xr=0;xr<Qt;xr++)jc.printChar(H,ze[oe+xr]);fe+=Qt}return Qr[Dt>>2]=fe,0}function Jb(H){rt(H)}var Dg=!1,Xc={abort:Yb,emscripten_get_heap_max:sf,emscripten_memcpy_big:Eg,emscripten_resize_heap:Ag,fd_close:Zb,fd_seek:TT,fd_write:_T,setTempRet0:Jb},bW=Xb(),ET=e.___wasm_call_ctors=function(){return(ET=e.___wasm_call_ctors=e.asm.__wasm_call_ctors).apply(null,arguments)},Qb=e._init=function(){return(Qb=e._init=e.asm.init).apply(null,arguments)},tw=e._init_with_threads_count=function(){return(tw=e._init_with_threads_count=e.asm.init_with_threads_count).apply(null,arguments)},$g=e._get_threads_count=function(){return($g=e._get_threads_count=e.asm.get_threads_count).apply(null,arguments)},Rg=e._register_tensor=function(){return(Rg=e._register_tensor=e.asm.register_tensor).apply(null,arguments)},ew=e._dispose_data=function(){return(ew=e._dispose_data=e.asm.dispose_data).apply(null,arguments)},Ht=e._dispose=function(){return(Ht=e._dispose=e.asm.dispose).apply(null,arguments)},rw=e._Abs=function(){return(rw=e._Abs=e.asm.Abs).apply(null,arguments)},Fg=e._Add=function(){return(Fg=e._Add=e.asm.Add).apply(null,arguments)},fu=e._AddN=function(){return(fu=e._AddN=e.asm.AddN).apply(null,arguments)},Yc=e._All=function(){return(Yc=e._All=e.asm.All).apply(null,arguments)},nw=e._Any=function(){return(nw=e._Any=e.asm.Any).apply(null,arguments)},AT=e._ArgMax=function(){return(AT=e._ArgMax=e.asm.ArgMax).apply(null,arguments)},ow=e._AvgPool=function(){return(ow=e._AvgPool=e.asm.AvgPool).apply(null,arguments)},DT=e._BatchMatMul=function(){return(DT=e._BatchMatMul=e.asm.BatchMatMul).apply(null,arguments)},du=e._Ceil=function(){return(du=e._Ceil=e.asm.Ceil).apply(null,arguments)},sw=e._ClipByValue=function(){return(sw=e._ClipByValue=e.asm.ClipByValue).apply(null,arguments)},iw=e._Conv2D=function(){return(iw=e._Conv2D=e.asm.Conv2D).apply(null,arguments)},aw=e._Conv2DBackpropInput=function(){return(aw=e._Conv2DBackpropInput=e.asm.Conv2DBackpropInput).apply(null,arguments)},lw=e._Cos=function(){return(lw=e._Cos=e.asm.Cos).apply(null,arguments)},uw=e._Cosh=function(){return(uw=e._Cosh=e.asm.Cosh).apply(null,arguments)},cw=e._CropAndResize=function(){return(cw=e._CropAndResize=e.asm.CropAndResize).apply(null,arguments)},Og=e._Cumprod=function(){return(Og=e._Cumprod=e.asm.Cumprod).apply(null,arguments)},pw=e._Cumsum=function(){return(pw=e._Cumsum=e.asm.Cumsum).apply(null,arguments)},mw=e._DepthToSpace=function(){return(mw=e._DepthToSpace=e.asm.DepthToSpace).apply(null,arguments)},fw=e._DepthwiseConv2dNative=function(){return(fw=e._DepthwiseConv2dNative=e.asm.DepthwiseConv2dNative).apply(null,arguments)},dw=e._Elu=function(){return(dw=e._Elu=e.asm.Elu).apply(null,arguments)},hw=e._Equal=function(){return(hw=e._Equal=e.asm.Equal).apply(null,arguments)},Pg=e._Exp=function(){return(Pg=e._Exp=e.asm.Exp).apply(null,arguments)},gw=e._FlipLeftRight=function(){return(gw=e._FlipLeftRight=e.asm.FlipLeftRight).apply(null,arguments)},xw=e._Floor=function(){return(xw=e._Floor=e.asm.Floor).apply(null,arguments)},hu=e._FloorDiv=function(){return(hu=e._FloorDiv=e.asm.FloorDiv).apply(null,arguments)},af=e._FusedBatchNorm=function(){return(af=e._FusedBatchNorm=e.asm.FusedBatchNorm).apply(null,arguments)},yw=e._FusedConv2D=function(){return(yw=e._FusedConv2D=e.asm.FusedConv2D).apply(null,arguments)},bw=e._FusedDepthwiseConv2D=function(){return(bw=e._FusedDepthwiseConv2D=e.asm.FusedDepthwiseConv2D).apply(null,arguments)},ww=e._Gather=function(){return(ww=e._Gather=e.asm.Gather).apply(null,arguments)},ee=e._GatherNd=function(){return(ee=e._GatherNd=e.asm.GatherNd).apply(null,arguments)},vw=e._Greater=function(){return(vw=e._Greater=e.asm.Greater).apply(null,arguments)},Cw=e._GreaterEqual=function(){return(Cw=e._GreaterEqual=e.asm.GreaterEqual).apply(null,arguments)},Iw=e._LeakyRelu=function(){return(Iw=e._LeakyRelu=e.asm.LeakyRelu).apply(null,arguments)},Sw=e._Less=function(){return(Sw=e._Less=e.asm.Less).apply(null,arguments)},kw=e._LessEqual=function(){return(kw=e._LessEqual=e.asm.LessEqual).apply(null,arguments)},Nw=e._Log=function(){return(Nw=e._Log=e.asm.Log).apply(null,arguments)},lf=e._LogicalAnd=function(){return(lf=e._LogicalAnd=e.asm.LogicalAnd).apply(null,arguments)},Mg=e._Max=function(){return(Mg=e._Max=e.asm.Max).apply(null,arguments)},Lg=e._MaxPool=function(){return(Lg=e._MaxPool=e.asm.MaxPool).apply(null,arguments)},Tw=e._Maximum=function(){return(Tw=e._Maximum=e.asm.Maximum).apply(null,arguments)},_w=e._Mean=function(){return(_w=e._Mean=e.asm.Mean).apply(null,arguments)},Ew=e._Min=function(){return(Ew=e._Min=e.asm.Min).apply(null,arguments)},Aw=e._Minimum=function(){return(Aw=e._Minimum=e.asm.Minimum).apply(null,arguments)},Dw=e._MirrorPad=function(){return(Dw=e._MirrorPad=e.asm.MirrorPad).apply(null,arguments)},$w=e._Multiply=function(){return($w=e._Multiply=e.asm.Multiply).apply(null,arguments)},nr=e._Neg=function(){return(nr=e._Neg=e.asm.Neg).apply(null,arguments)},Rw=e._NonMaxSuppressionV3=function(){return(Rw=e._NonMaxSuppressionV3=e.asm.NonMaxSuppressionV3).apply(null,arguments)},Fw=e._NonMaxSuppressionV4=function(){return(Fw=e._NonMaxSuppressionV4=e.asm.NonMaxSuppressionV4).apply(null,arguments)},Ow=e._NonMaxSuppressionV5=function(){return(Ow=e._NonMaxSuppressionV5=e.asm.NonMaxSuppressionV5).apply(null,arguments)},Zc=e._NotEqual=function(){return(Zc=e._NotEqual=e.asm.NotEqual).apply(null,arguments)},zg=e._OneHot=function(){return(zg=e._OneHot=e.asm.OneHot).apply(null,arguments)},Bg=e._PadV2=function(){return(Bg=e._PadV2=e.asm.PadV2).apply(null,arguments)},Vg=e._Pow=function(){return(Vg=e._Pow=e.asm.Pow).apply(null,arguments)},Pw=e._Prelu=function(){return(Pw=e._Prelu=e.asm.Prelu).apply(null,arguments)},Gg=e._Prod=function(){return(Gg=e._Prod=e.asm.Prod).apply(null,arguments)},Mw=e._RealDiv=function(){return(Mw=e._RealDiv=e.asm.RealDiv).apply(null,arguments)},$T=e._Relu=function(){return($T=e._Relu=e.asm.Relu).apply(null,arguments)},Wg=e._Relu6=function(){return(Wg=e._Relu6=e.asm.Relu6).apply(null,arguments)},RT=e._ResizeBilinear=function(){return(RT=e._ResizeBilinear=e.asm.ResizeBilinear).apply(null,arguments)},Lw=e._Reverse=function(){return(Lw=e._Reverse=e.asm.Reverse).apply(null,arguments)},zw=e._RotateWithOffset=function(){return(zw=e._RotateWithOffset=e.asm.RotateWithOffset).apply(null,arguments)},Bw=e._Round=function(){return(Bw=e._Round=e.asm.Round).apply(null,arguments)},Vw=e._Rsqrt=function(){return(Vw=e._Rsqrt=e.asm.Rsqrt).apply(null,arguments)},Gw=e._ScatterNd=function(){return(Gw=e._ScatterNd=e.asm.ScatterNd).apply(null,arguments)},Ww=e._SelectV2=function(){return(Ww=e._SelectV2=e.asm.SelectV2).apply(null,arguments)},Uw=e._Sigmoid=function(){return(Uw=e._Sigmoid=e.asm.Sigmoid).apply(null,arguments)},Hw=e._Sin=function(){return(Hw=e._Sin=e.asm.Sin).apply(null,arguments)},qw=e._Softmax=function(){return(qw=e._Softmax=e.asm.Softmax).apply(null,arguments)},Kw=e._SparseFillEmptyRows=function(){return(Kw=e._SparseFillEmptyRows=e.asm.SparseFillEmptyRows).apply(null,arguments)},jw=e._SparseReshape=function(){return(jw=e._SparseReshape=e.asm.SparseReshape).apply(null,arguments)},Xw=e._SparseSegmentReduction=function(){return(Xw=e._SparseSegmentReduction=e.asm.SparseSegmentReduction).apply(null,arguments)},Yw=e._Sqrt=function(){return(Yw=e._Sqrt=e.asm.Sqrt).apply(null,arguments)},Zw=e._Square=function(){return(Zw=e._Square=e.asm.Square).apply(null,arguments)},Jw=e._SquaredDifference=function(){return(Jw=e._SquaredDifference=e.asm.SquaredDifference).apply(null,arguments)},Qw=e._Step=function(){return(Qw=e._Step=e.asm.Step).apply(null,arguments)},tv=e._StridedSlice=function(){return(tv=e._StridedSlice=e.asm.StridedSlice).apply(null,arguments)},ev=e._Sub=function(){return(ev=e._Sub=e.asm.Sub).apply(null,arguments)},rv=e._Sum=function(){return(rv=e._Sum=e.asm.Sum).apply(null,arguments)},nv=e._Tan=function(){return(nv=e._Tan=e.asm.Tan).apply(null,arguments)},ov=e._Tanh=function(){return(ov=e._Tanh=e.asm.Tanh).apply(null,arguments)},sv=e._Tile=function(){return(sv=e._Tile=e.asm.Tile).apply(null,arguments)},iv=e._TopK=function(){return(iv=e._TopK=e.asm.TopK).apply(null,arguments)},av=e._Transform=function(){return(av=e._Transform=e.asm.Transform).apply(null,arguments)},lv=e._Transpose=function(){return(lv=e._Transpose=e.asm.Transpose).apply(null,arguments)},uv=e.__FusedMatMul=function(){return(uv=e.__FusedMatMul=e.asm._FusedMatMul).apply(null,arguments)},cv=e._malloc=function(){return(cv=e._malloc=e.asm.malloc).apply(null,arguments)},pv=e._free=function(){return(pv=e._free=e.asm.free).apply(null,arguments)},mv=e.___errno_location=function(){return(mv=e.___errno_location=e.asm.__errno_location).apply(null,arguments)},fv=e._emscripten_main_thread_process_queued_calls=function(){return(fv=e._emscripten_main_thread_process_queued_calls=e.asm.emscripten_main_thread_process_queued_calls).apply(null,arguments)},Ug=e.stackSave=function(){return(Ug=e.stackSave=e.asm.stackSave).apply(null,arguments)},Hg=e.stackRestore=function(){return(Hg=e.stackRestore=e.asm.stackRestore).apply(null,arguments)},uf=e.stackAlloc=function(){return(uf=e.stackAlloc=e.asm.stackAlloc).apply(null,arguments)},dv=e.dynCall_iijjiiii=function(){return(dv=e.dynCall_iijjiiii=e.asm.dynCall_iijjiiii).apply(null,arguments)},hv=e.dynCall_jiji=function(){return(hv=e.dynCall_jiji=e.asm.dynCall_jiji).apply(null,arguments)};e.cwrap=bt;var Jc;function cf(H){this.name="ExitStatus",this.message="Program terminated with exit("+H+")",this.status=H}xl=function H(){Jc||pf(),Jc||(xl=H)};function pf(H){if(H=H||a,Un>0||(vg(),Un>0))return;function tt(){Jc||(Jc=!0,e.calledRun=!0,!it&&(Cg(),n(e),e.onRuntimeInitialized&&e.onRuntimeInitialized(),Ig()))}e.setStatus?(e.setStatus("Running..."),setTimeout(function(){setTimeout(function(){e.setStatus("")},1),tt()},1)):tt()}e.run=pf;function FT(H){ft=H,ef()||(e.onExit&&e.onExit(H),it=!0),l(H,new cf(H))}if(e.preInit)for(typeof e.preInit=="function"&&(e.preInit=[e.preInit]);e.preInit.length>0;)e.preInit.pop()();pf();var Qc;s&&(Qc={uncaughtException:process.listeners("uncaughtException").filter(function(H){return!s.uncaughtException.indexOf(H)>-1}),unhandledRejection:process.listeners("unhandledRejection").filter(function(H){return!s.unhandledRejection.indexOf(H)>-1})});var tp;if(typeof t!="undefined")tp=t;else if(typeof WasmBackendModuleThreadedSimd!="undefined")tp=WasmBackendModuleThreadedSimd;else throw new Error("Could not find wasm module in post.js");if(Qc){var gv=tp._dispose;tp._dispose=function(){gv(),Qc.uncaughtException.forEach(function(H){process.removeListener("uncaughtException",H)}),Qc.unhandledRejection.forEach(function(H){process.removeListener("unhandledRejection",H)})}}return t.ready}})();typeof Wb=="object"&&typeof bT=="object"?bT.exports=yT:typeof define=="function"&&define.amd?define([],function(){return yT}):typeof Wb=="object"&&(Wb.WasmBackendModule=yT)});var wl=class{constructor(t,e){this.backend=t,this.dataMover=e,this.data=new WeakMap,this.dataIdsCount=0}get(t){return this.data.has(t)||this.dataMover.moveData(this.backend,t),this.data.get(t)}set(t,e){this.dataIdsCount++,this.data.set(t,e)}has(t){return this.data.has(t)}delete(t){return this.dataIdsCount--,this.data.delete(t)}numDataIds(){return this.dataIdsCount}},mi=class{refCount(t){return Hn("refCount")}incRef(t){return Hn("incRef")}timerAvailable(){return!0}time(t){return Hn("time")}read(t){return Hn("read")}readSync(t){return Hn("readSync")}readToGPU(t,e){return Hn("readToGPU")}numDataIds(){return Hn("numDataIds")}disposeData(t,e){return Hn("disposeData")}write(t,e,n){return Hn("write")}move(t,e,n,o,s){return Hn("move")}memory(){return Hn("memory")}floatPrecision(){return Hn("floatPrecision")}epsilon(){return this.floatPrecision()===32?1e-7:1e-4}dispose(){return Hn("dispose")}};function Hn(r){throw new Error(`'${r}' not yet implemented or not found in the registry. This kernel may not be supported by the tfjs backend you have chosen`)}function UT(r){let t=r.length,e=0;for(;t>0;)e=Math.random()*t|0,t--,Qg(r,t,e)}function yU(r,t){if(r.length!==t.length)throw new Error(`Array sizes must match to be shuffled together First array length was ${r.length}Second array length was ${t.length}`);let e=r.length,n=0;for(;e>0;)n=Math.random()*e|0,e--,Qg(r,e,n),Qg(t,e,n)}function op(r,t,e){return Math.max(r,Math.min(t,e))}function bU(r){return r%2===0?r:r+1}function Qg(r,t,e){let n=r[t];r[t]=r[e],r[e]=n}function wU(r){let t=0;for(let e=0;e<r.length;e++)t+=r[e];return t}function vU(r,t){let e=Math.random();return t*e+(1-e)*r}function CU(r,t){let e=0;for(let n=0;n<r.length;n++){let o=Number(r[n])-Number(t[n]);e+=o*o}return e}function A(r,t){if(!r)throw new Error(typeof t=="string"?t:t())}function Pe(r,t,e=""){A(Fn(r,t),()=>e+` Shapes ${r} and ${t} must match`)}function qn(r){A(r!=null,()=>"The input to the tensor constructor must be a non-null value.")}function qo(r,t=[],e=!1){if(t==null&&(t=[]),Array.isArray(r)||yr(r)&&!e)for(let n=0;n<r.length;++n)qo(r[n],t,e);else t.push(r);return t}function re(r){if(r.length===0)return 1;let t=r[0];for(let e=1;e<r.length;e++)t*=r[e];return t}function IU(r){return r.length===0}function Fn(r,t){if(r===t)return!0;if(r==null||t==null||r.length!==t.length)return!1;for(let e=0;e<r.length;e++)if(r[e]!==t[e])return!1;return!0}function Ki(r){return r%1===0}function SU(r){if(Math.tanh!=null)return Math.tanh(r);if(r===1/0)return 1;if(r===-1/0)return-1;{let t=Math.exp(2*r);return(t-1)/(t+1)}}function kU(r){let t=Math.ceil(Math.sqrt(r));return[t,Math.ceil(r/t)]}function NU(r){let t=new Uint32Array(r);for(let e=0;e<r;++e)t[e]=e;return UT(t),t}function bu(r,t){return t<=r.length?r:r+" ".repeat(t-r.length)}function TU(r,t=n=>0,e){return new Promise((n,o)=>{let s=0,i=()=>{if(r()){n();return}s++;let a=t(s);if(e!=null&&s>=e){o();return}setTimeout(i,a)};i()})}function _U(r,t){let e=1,n=-1;for(let s=0;s<r.length;++s)if(r[s]>=0)e*=r[s];else if(r[s]===-1){if(n!==-1)throw Error(`Shapes can only have 1 implicit size. Found -1 at dim ${n} and dim ${s}`);n=s}else if(r[s]<0)throw Error(`Shapes can not be < 0. Found ${r[s]} at dim ${s}`);if(n===-1){if(t>0&&t!==e)throw Error(`Size(${t}) must match the product of shape ${r}`);return r}if(e===0)throw Error(`Cannot infer the missing size in [${r}] when there are 0 elements`);if(t%e!==0)throw Error(`The implicit shape can't be a fractional number. Got ${t} / ${e}`);let o=r.slice();return o[n]=t/e,o}function mr(r,t){let e=t.length;return r=r==null?t.map((n,o)=>o):[].concat(r),A(r.every(n=>n>=-e&&n<e),()=>`All values in axis param must be in range [-${e}, ${e}) but got axis ${r}`),A(r.every(n=>Ki(n)),()=>`All values in axis param must be integers but got axis ${r}`),r.map(n=>n<0?e+n:n)}function Sv(r,t){let e=[],n=[],o=t!=null&&Array.isArray(t)&&t.length===0,s=t==null||o?null:mr(t,r).sort(),i=0;for(let a=0;a<r.length;++a){if(s!=null){if(s[i]===a&&r[a]!==1)throw new Error(`Can't squeeze axis ${a} since its dim '${r[a]}' is not 1`);(s[i]==null||s[i]>a)&&r[a]===1&&(e.push(r[a]),n.push(a)),s[i]<=a&&i++}r[a]!==1&&(e.push(r[a]),n.push(a))}return{newShape:e,keptDims:n}}function kv(r,t){let e=null;if(r==null||r==="float32")e=new Float32Array(t);else if(r==="int32")e=new Int32Array(t);else if(r==="bool")e=new Uint8Array(t);else throw new Error(`Unknown data type ${r}`);return e}function Nv(r,t){let e=null;if(r==null||r==="float32")e=new Float32Array(t);else if(r==="int32")e=new Int32Array(t);else if(r==="bool")e=new Uint8Array(t);else if(r==="string")e=new Array(t);else throw new Error(`Unknown data type ${r}`);return e}function Tv(r,t){for(let e=0;e<r.length;e++){let n=r[e];if(isNaN(n)||!isFinite(n))throw Error(`A tensor of type ${t} being uploaded contains ${n}.`)}}function _v(r){return r==="bool"||r==="complex64"||r==="float32"||r==="int32"||r==="string"}function EU(r,t){return!(t==="complex64"||t==="float32"&&r!=="complex64"||t==="int32"&&r!=="float32"&&r!=="complex64"||t==="bool"&&r==="bool")}function yr(r){return r instanceof Float32Array||r instanceof Int32Array||r instanceof Uint8Array||r instanceof Uint8ClampedArray}function tx(r){if(r==="float32"||r==="int32")return 4;if(r==="complex64")return 8;if(r==="bool")return 1;throw new Error(`Unknown dtype ${r}`)}function Ev(r){if(r==null)return 0;let t=0;return r.forEach(e=>t+=e.length),t}function Ko(r){return typeof r=="string"||r instanceof String}function HT(r){return typeof r=="boolean"}function qT(r){return typeof r=="number"}function sp(r){return Array.isArray(r)?sp(r[0]):r instanceof Float32Array?"float32":r instanceof Int32Array||r instanceof Uint8Array||r instanceof Uint8ClampedArray?"int32":qT(r)?"float32":Ko(r)?"string":HT(r)?"bool":"float32"}function fi(r){return!!(r&&r.constructor&&r.call&&r.apply)}function ip(r,t){for(let e=t;e<r;++e)if(r%e===0)return e;return r}function di(r){let t=r.length;if(t<2)return[];let e=new Array(t-1);e[t-2]=r[t-1];for(let n=t-3;n>=0;--n)e[n]=e[n+1]*r[n+1];return e}function KT(r,t,e,n=!1){let o=new Array;if(t.length===1){let s=t[0]*(n?2:1);for(let i=0;i<s;i++)o[i]=e[r+i]}else{let s=t[0],i=t.slice(1),a=i.reduce((u,l)=>u*l)*(n?2:1);for(let u=0;u<s;u++)o[u]=KT(r+u*a,i,e,n)}return o}function yu(r,t,e=!1){if(r.length===0)return t[0];let n=r.reduce((o,s)=>o*s)*(e?2:1);if(n===0)return[];if(n!==t.length)throw new Error(`[${r}] does not match the input size ${t.length}${e?" for a complex tensor":""}.`);return KT(0,r,t,e)}function df(r,t){let e=ap(r,t);for(let n=0;n<e.length;n++)e[n]=1;return e}function ap(r,t){if(t==null||t==="float32"||t==="complex64")return new Float32Array(r);if(t==="int32")return new Int32Array(r);if(t==="bool")return new Uint8Array(r);throw new Error(`Unknown data type ${t}`)}function AU(r,t){let e=r.reduce((n,o)=>n*o,1);if(t==null||t==="float32")return yu(r,new Float32Array(e));if(t==="int32")return yu(r,new Int32Array(e));if(t==="bool")return yu(r,new Uint8Array(e));throw new Error(`Unknown data type ${t}`)}function hf(r){r.forEach(t=>{A(Number.isInteger(t)&&t>=0,()=>`Tensor must have a shape comprised of positive integers but got shape [${r}].`)})}function DU(r,t,e){if(t===0)return 0;if(t===1)return r[0];let n=r[r.length-1];for(let o=0;o<r.length-1;++o)n+=e[o]*r[o];return n}function $U(r,t,e){if(t===0)return[];if(t===1)return[r];let n=new Array(t);for(let o=0;o<n.length-1;++o)n[o]=Math.floor(r/e[o]),r-=n[o]*e[o];return n[n.length-1]=r,n}function gf(r){return r&&r.then&&typeof r.then=="function"}var jT="tfjsflags",ex=class{constructor(t){this.global=t,this.flags={},this.flagRegistry={},this.urlFlags={},this.getQueryParams=FU,this.populateURLFlags()}setPlatform(t,e){this.platform!=null&&(V().getBool("IS_TEST")||V().getBool("PROD")||console.warn(`Platform ${this.platformName} has already been set. Overwriting the platform with ${t}.`)),this.platformName=t,this.platform=e}registerFlag(t,e,n){if(this.flagRegistry[t]={evaluationFn:e,setHook:n},this.urlFlags[t]!=null){let o=this.urlFlags[t];V().getBool("IS_TEST")||V().getBool("PROD")||console.warn(`Setting feature override from URL ${t}: ${o}.`),this.set(t,o)}}async getAsync(t){return t in this.flags?this.flags[t]:(this.flags[t]=await this.evaluateFlag(t),this.flags[t])}get(t){if(t in this.flags)return this.flags[t];let e=this.evaluateFlag(t);if(gf(e))throw new Error(`Flag ${t} cannot be synchronously evaluated. Please use getAsync() instead.`);return this.flags[t]=e,this.flags[t]}getNumber(t){return this.get(t)}getBool(t){return this.get(t)}getFlags(){return this.flags}get features(){return this.flags}set(t,e){if(this.flagRegistry[t]==null)throw new Error(`Cannot set flag ${t} as it has not been registered.`);this.flags[t]=e,this.flagRegistry[t].setHook!=null&&this.flagRegistry[t].setHook(e)}evaluateFlag(t){if(this.flagRegistry[t]==null)throw new Error(`Cannot evaluate flag '${t}': no evaluation function found.`);return this.flagRegistry[t].evaluationFn()}setFlags(t){this.flags=Object.assign({},t)}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 t=this.getQueryParams(this.global.location.search);jT in t&&t[jT].split(",").forEach(n=>{let[o,s]=n.split(":");this.urlFlags[o]=PU(o,s)})}};function FU(r){let t={};return r.replace(/[?&]([^=?&]+)(?:=([^&]*))?/g,(e,...n)=>(OU(t,n[0],n[1]),n.join("="))),t}function OU(r,t,e){r[decodeURIComponent(t)]=decodeURIComponent(e||"")}function PU(r,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 ${r}.`)}function V(){return Av}var Av=null;function XT(r){Av=r}var Dv;function $v(){if(Dv==null){let r;if(typeof window!="undefined")r=window;else if(typeof global!="undefined")r=global;else if(typeof process!="undefined")r=process;else if(typeof self!="undefined")r=self;else throw new Error("Could not find a global object");Dv=r}return Dv}function MU(){let r=$v();return r._tfGlobals==null&&(r._tfGlobals=new Map),r._tfGlobals}function xf(r,t){let e=MU();if(e.has(r))return e.get(r);{let n=t();return e.set(r,n),e.get(r)}}var hi="Abs",ji="Acos",Xi="Acosh",Kn="Add",jo="AddN",Yi="All",Zi="Any",Xo="ArgMax",vl="ArgMin",Ji="Asin",Qi="Asinh",ta="Atan",ea="Atanh",ra="Atan2",Yo="AvgPool",lp="AvgPoolGrad",Cl="AvgPool3D",up="AvgPool3DGrad",Zo="BatchMatMul",gi="BatchToSpaceND",cp="Bincount",YT="BroadcastTo",pp="BroadcastArgs",oo="Cast",Jo="Ceil",so="ClipByValue",mp="Complex",Il="ComplexAbs",xi="Concat",Qo="Conv2D",fp="Conv2DBackpropFilter",ts="Conv2DBackpropInput",Sl="Conv3D",dp="Conv3DBackpropFilterV2",hp="Conv3DBackpropInputV2",es="Cos",rs="Cosh",na="Cumprod",ns="Cumsum",oa="CropAndResize",gp="DenseBincount",sa="DepthToSpace",os="DepthwiseConv2dNative",xp="DepthwiseConv2dNativeBackpropFilter",yp="DepthwiseConv2dNativeBackpropInput",bp="Diag",kl="Dilation2D",yf="Dilation2DBackpropInput",bf="Dilation2DBackpropFilter",ss="RealDiv",wp="Einsum",is="Elu",vp="EluGrad",ia="Erf",aa="Equal",as="Exp",yi="ExpandDims",la="Expm1",Cp="FFT",Nl="Fill",ua="FlipLeftRight",ls="Floor",us="FloorDiv",cs="FusedBatchNorm",bi="GatherV2",ca="GatherNd",pa="Greater",ps="GreaterEqual",io="Identity",Ip="IFFT",Sp="Imag",ma="IsFinite",fa="IsInf",da="IsNan",ms="LeakyRelu",ha="Less",ga="LessEqual",kp="LinSpace",fs="Log",xa="Log1p",ya="LogicalAnd",wu="LogicalNot",vu="LogicalOr",ZT="LogSoftmax",Tl="LRN",Np="LRNGrad",ds="Max",hs="Maximum",gs="MaxPool",Tp="MaxPoolGrad",_l="MaxPool3D",_p="MaxPool3DGrad",Ep="MaxPoolWithArgmax",xs="Mean",ys="Min",bs="Minimum",ws="MirrorPad",ba="Mod",Ap="Multinomial",vs="Multiply",wi="Neg",wa="NotEqual",va="NonMaxSuppressionV3",Ca="NonMaxSuppressionV4",Ia="NonMaxSuppressionV5",vi="OnesLike",Cs="OneHot",Ci="Pack",Is="PadV2",Cat="Pool",Ss="Pow",ks="Prelu",Sa="Prod",El="Range",Dp="Real",ka="Reciprocal",Ns="Relu",Ii="Reshape",Al="ResizeNearestNeighbor",$p="ResizeNearestNeighborGrad",Ts="ResizeBilinear",Rp="ResizeBilinearGrad",_s="Relu6",Es="Reverse",As="Round",Ds="Rsqrt",Na="ScatterNd",Si="Select",Ta="Selu",ki="Slice",$s="Sin",_a="Sinh",Ea="Sign",Rs="Sigmoid",Aa="Softplus",Fs="Sqrt",Os="Sum",Ni="SpaceToBatchND",Ti="SplitV",Ps="Softmax",Dl="SparseFillEmptyRows",Da="SparseReshape",$l="SparseSegmentMean",Rl="SparseSegmentSum",Fp="SparseToDense",Ms="SquaredDifference",Fl="Square",$a="StridedSlice",Op="StringNGrams",Pp="StringSplit",Mp="StringToHashBucketFast",Ls="Sub",zs="Tan",Bs="Tanh",jn="Tile",Ra="TopK",Fa="Transform",Vs="Transpose",Lp="Unique",_i="Unpack",Ol="UnsortedSegmentSum",Ei="ZerosLike",ao="Step",wf="FromPixels",Oa="RotateWithOffset",Ai="_FusedMatMul",Di="FusedConv2D",$i="FusedDepthwiseConv2D";function Ri(...r){V().getBool("IS_TEST")||V().getBool("PROD")||console.warn(...r)}function LU(...r){V().getBool("IS_TEST")||V().getBool("PROD")||console.log(...r)}var zp=xf("kernelRegistry",()=>new Map),vf=xf("gradRegistry",()=>new Map);function Cf(r,t){let e=Fv(r,t);return zp.get(e)}function Rv(r){return vf.get(r)}function rx(r){let t=zp.entries(),e=[];for(;;){let{done:n,value:o}=t.next();if(n)break;let[s,i]=o,[a]=s.split("_");a===r&&e.push(i)}return e}function Cu(r){let{kernelName:t,backendName:e}=r,n=Fv(t,e);zp.has(n)&&Ri(`The kernel '${t}' for backend '${e}' is already registered`),zp.set(n,r)}function QT(r){let{kernelName:t}=r;vf.has(t)&&V().getBool("DEBUG")&&Ri(`Overriding the gradient for '${t}'`),vf.set(t,r)}function Tat(r,t){let e=Fv(r,t);if(!zp.has(e))throw new Error(`The kernel '${r}' for backend '${t}' is not registered`);zp.delete(e)}function _at(r){if(!vf.has(r))throw new Error(`The gradient '${r}' for backend is not registered`);vf.delete(r)}function Eat(r,t){rx(r).forEach(n=>{let o=Object.assign({},n,{backendName:t});Cu(o)})}function Fv(r,t){return`${t}_${r}`}var x={};Yt(x,{arraysEqual:()=>Fn,assert:()=>A,assertNonNegativeIntegerDimensions:()=>hf,assertNonNull:()=>qn,assertShapesMatch:()=>Pe,bytesFromStringArray:()=>Ev,bytesPerElement:()=>tx,checkConversionForErrors:()=>Tv,clamp:()=>op,computeStrides:()=>di,createScalarValue:()=>HU,createShuffledIndices:()=>NU,decodeString:()=>Wp,distSquared:()=>CU,encodeString:()=>Ml,fetch:()=>KU,fingerPrint64:()=>UU,flatten:()=>qo,getArrayFromDType:()=>Nv,getTypedArrayFromDType:()=>kv,hasEncodingLoss:()=>EU,hexToLong:()=>If,indexToLoc:()=>$U,inferDtype:()=>sp,inferFromImplicitShape:()=>_U,isBoolean:()=>HT,isFunction:()=>fi,isInt:()=>Ki,isNumber:()=>qT,isPromise:()=>gf,isScalarShape:()=>IU,isString:()=>Ko,isTypedArray:()=>yr,isValidDtype:()=>_v,locToIndex:()=>DU,makeOnesTypedArray:()=>df,makeZerosNestedTypedArray:()=>AU,makeZerosTypedArray:()=>ap,nearestDivisor:()=>ip,nearestLargerEven:()=>bU,now:()=>Tu,parseAxisParam:()=>mr,randUniform:()=>vU,repeatedTry:()=>TU,rightPad:()=>bu,shuffle:()=>UT,shuffleCombo:()=>yU,sizeFromShape:()=>re,sizeToSquarishShape:()=>kU,squeezeShape:()=>Sv,sum:()=>wU,swap:()=>Qg,tanh:()=>SU,toNestedArray:()=>yu,toTypedArray:()=>Gp});var Lv=xu(c1());var Nu=Lv.default||Lv;function If(r){return Nu.fromString(r,!0,16)}var m1=If("c3a5c85c97cb3127"),ku=If("b492b66fbe98f273"),on=If("9ae16a3b2f90404f");function Mv(r){return r.xor(r.shru(47))}function f1(r,t,e){let n=r.slice(t,t+e);return Nu.fromBytes(Array.from(n),!0,!0)}function Be(r,t){return f1(r,t,8)}function p1(r,t){return f1(r,t,4)}function Er(r,t){return t===0?r:r.shru(t).or(r.shl(64-t))}function Pl(r,t,e=If("9ddfea08eb382d69")){let n=r.xor(t).mul(e);n=n.xor(n.shru(47));let o=t.xor(n).mul(e);return o=o.xor(o.shru(47)),o=o.mul(e),o}function BU(r,t,e,n,o,s){o=o.add(r),s=Er(s.add(o).add(n),21);let i=o;return o=o.add(t),o=o.add(e),s=s.add(Er(o,44)),[o.add(n),s.add(i)]}function ox(r,t,e,n){return BU(Be(r,t),Be(r,t+8),Be(r,t+16),Be(r,t+24),e,n)}function VU(r,t=r.length){if(t>=8){let e=on.add(t*2),n=Be(r,0).add(on),o=Be(r,t-8),s=Er(o,37).mul(e).add(n),i=Er(n,25).add(o).mul(e);return Pl(s,i,e)}if(t>=4){let e=on.add(t*2),n=p1(r,0);return Pl(n.shl(3).add(t),p1(r,t-4),e)}if(t>0){let e=r[0],n=r[t>>1],o=r[t-1],s=e+(n<<8),i=t+(o<<2);return Mv(on.mul(s).xor(m1.mul(i))).mul(on)}return on}function GU(r,t=r.length){let e=on.add(t*2),n=Be(r,0).mul(ku),o=Be(r,8),s=Be(r,t-8).mul(e),i=Be(r,t-16).mul(on);return Pl(Er(n.add(o),43).add(Er(s,30)).add(i),n.add(Er(o.add(on),18)).add(s),e)}function WU(r,t=r.length){let e=on.add(t*2),n=Be(r,0).mul(on),o=Be(r,8),s=Be(r,t-8).mul(e),i=Be(r,t-16).mul(on),a=Er(n.add(o),43).add(Er(s,30)).add(i),u=Pl(a,n.add(Er(o.add(on),18)).add(s),e),l=Be(r,16).mul(e),c=Be(r,24),p=a.add(Be(r,t-32)).mul(e),m=u.add(Be(r,t-24)).mul(e);return Pl(Er(l.add(c),43).add(Er(p,30)).add(m),l.add(Er(c.add(n),18)).add(p),e)}function UU(r,t=r.length){let e=Nu.fromNumber(81,!0);if(t<=32)return t<=16?VU(r,t):GU(r,t);if(t<=64)return WU(r,t);let n=e,o=e.mul(ku).add(113),s=Mv(o.mul(on).add(113)).mul(on),i=[Nu.UZERO,Nu.UZERO],a=[Nu.UZERO,Nu.UZERO];n=n.mul(on).add(Be(r,0));let u=0,l=(t-1>>6)*64,c=l+(t-1&63)-63;do n=Er(n.add(o).add(i[0]).add(Be(r,u+8)),37).mul(ku),o=Er(o.add(i[1]).add(Be(r,u+48)),42).mul(ku),n=n.xor(a[1]),o=o.add(i[0]).add(Be(r,u+40)),s=Er(s.add(a[0]),33).mul(ku),i=ox(r,u,i[1].mul(ku),n.add(a[0])),a=ox(r,u+32,s.add(a[1]),o.add(Be(r,u+16))),[s,n]=[n,s],u+=64;while(u!==l);let p=ku.add(s.and(255).shl(1));return u=c,a[0]=a[0].add(t-1&63),i[0]=i[0].add(a[0]),a[0]=a[0].add(i[0]),n=Er(n.add(o).add(i[0]).add(Be(r,u+8)),37).mul(p),o=Er(o.add(i[1]).add(Be(r,u+48)),42).mul(p),n=n.xor(a[1].mul(9)),o=o.add(i[0].mul(9).add(Be(r,u+40))),s=Er(s.add(a[0]),33).mul(p),i=ox(r,u,i[1].mul(p),n.add(a[0])),a=ox(r,u+32,s.add(a[1]),o.add(Be(r,u+16))),[s,n]=[n,s],Pl(Pl(i[0],a[0],p).add(Mv(o).mul(m1)).add(s),Pl(i[1],a[1],p).add(n),p)}function HU(r,t){return t==="string"?Ml(r):Gp([r],t)}function qU(r,t){return r instanceof Float32Array&&t==="float32"||r instanceof Int32Array&&t==="int32"||r instanceof Uint8Array&&t==="bool"}function Gp(r,t){if(t==="string")throw new Error("Cannot convert a string[] to a TypedArray");if(Array.isArray(r)&&(r=qo(r)),V().getBool("DEBUG")&&Tv(r,t),qU(r,t))return r;if(t==null||t==="float32"||t==="complex64")return new Float32Array(r);if(t==="int32")return new Int32Array(r);if(t==="bool"){let e=new Uint8Array(r.length);for(let n=0;n<e.length;++n)Math.round(r[n])!==0&&(e[n]=1);return e}else throw new Error(`Unknown data type ${t}`)}function Tu(){return V().platform.now()}function KU(r,t){return V().platform.fetch(r,t)}function Ml(r,t="utf-8"){return t=t||"utf-8",V().platform.encode(r,t)}function Wp(r,t="utf-8"){return t=t||"utf-8",V().platform.decode(r,t)}var zv=class{constructor(t,e){this.backendTimer=t,this.logger=e,e==null&&(this.logger=new d1)}profileKernel(t,e,n){let o,s=()=>{o=n()},i,a=Tu();if(this.backendTimer.timerAvailable())i=this.backendTimer.time(s);else{s();for(let l of o)l.dataSync();i=Promise.resolve({kernelMs:Tu()-a})}if(V().getBool("CHECK_COMPUTATION_FOR_ERRORS"))for(let l=0;l<o.length;l++){let c=o[l];c.data().then(p=>{jU(p,c.dtype,t)})}return{kernelName:t,outputs:o,inputs:e,timeMs:i.then(l=>l.kernelMs),extraInfo:i.then(l=>l.getExtraProfileInfo!=null?l.getExtraProfileInfo():"")}}logKernelProfile(t){let{kernelName:e,outputs:n,timeMs:o,inputs:s,extraInfo:i}=t;n.forEach(a=>{Promise.all([a.data(),o,i]).then(u=>{this.logger.logKernelProfile(e,a,u[0],u[1],s,u[2])})})}};function jU(r,t,e){if(t!=="float32")return!1;for(let n=0;n<r.length;n++){let o=r[n];if(isNaN(o)||!isFinite(o))return console.warn(`Found ${o} in the result of '${e}'`),!0}return!1}var d1=class{logKernelProfile(t,e,n,o,s,i){let a=typeof o=="number"?bu(`${o}ms`,9):o.error,u=bu(t,25),l=e.rank,c=e.size,p=bu(e.shape.toString(),14),m="";for(let f in s){let d=s[f];if(d!=null){let h=d.shape||e.shape,g=h.length;m+=`${f}: ${g}D ${g>0?h:""} `}}console.log(`%c${u} %c${a} %c${l}D ${p} %c${c} %c${m} %c${i}`,"font-weight:bold","color:red","color:blue","color: orange","color: green","color: steelblue")}};function h1(r,t,e){let n={},o={};for(let u=0;u<t.length;u++)n[t[u].id]=!0;for(let u=0;u<r.length;u++){let l=r[u],c=l.inputs;for(let p in c){let m=c[p],f=!1;for(let d=0;d<t.length;d++)if(n[m.id]){l.outputs.forEach(h=>n[h.id]=!0),f=!0,o[l.id]=!0;break}if(f)break}}let s={};s[e.id]=!0;let i={};for(let u=r.length-1;u>=0;u--){let l=r[u],c=l.inputs;for(let p=0;p<l.outputs.length;p++)if(s[l.outputs[p].id]){for(let m in c)s[c[m].id]=!0,i[l.id]=!0;break}}let a=[];for(let u=0;u<r.length;u++){let l=r[u];if(o[l.id]&&i[l.id]){let c={};for(let m in l.inputs){let f=l.inputs[m];n[f.id]&&(c[m]=f)}let p=Object.assign({},l);p.inputs=c,p.outputs=l.outputs,a.push(p)}}return a}function g1(r,t,e,n){for(let o=t.length-1;o>=0;o--){let s=t[o],i=[];if(s.outputs.forEach(u=>{let l=r[u.id];l!=null?i.push(l):i.push(null)}),s.gradient==null)throw new Error(`Cannot compute gradient: gradient function not found for ${s.kernelName}.`);let a=s.gradient(i);for(let u in s.inputs){if(!(u in a))throw new Error(`Cannot backprop through input ${u}. Available gradients found: ${Object.keys(a)}.`);let l=e(()=>a[u]());if(l.dtype!=="float32")throw new Error(`Error in gradient for op ${s.kernelName}. The gradient of input ${u} must have 'float32' dtype, but has '${l.dtype}'`);let c=s.inputs[u];if(!Fn(l.shape,c.shape))throw new Error(`Error in gradient for op ${s.kernelName}. The gradient of input '${u}' has shape '${l.shape}', which does not match the shape of the input '${c.shape}'`);if(r[c.id]==null)r[c.id]=l;else{let p=r[c.id];r[c.id]=n(p,l),p.dispose()}}}}var x1=20,Sf=3,Bv=7;function y1(r,t,e,n){let o=di(t),s=XU(r,t,e,o),i=t.length,a=sx(r,t,e,o,s),u=["Tensor"];return n&&(u.push(` dtype: ${e}`),u.push(` rank: ${i}`),u.push(` shape: [${t}]`),u.push(" values:")),u.push(a.map(l=>" "+l).join(`
`)),u.join(`
`)}function XU(r,t,e,n){let o=re(t),s=n[n.length-1],i=new Array(s).fill(0),a=t.length,u=e==="complex64"?Nf(r):r;if(a>1)for(let l=0;l<o/s;l++){let c=l*s;for(let p=0;p<s;p++)i[p]=Math.max(i[p],kf(u[c+p],0,e).length)}return i}function kf(r,t,e){let n;return Array.isArray(r)?n=`${parseFloat(r[0].toFixed(Bv))} + ${parseFloat(r[1].toFixed(Bv))}j`:Ko(r)?n=`'${r}'`:e==="bool"?n=b1(r):n=parseFloat(r.toFixed(Bv)).toString(),bu(n,t)}function b1(r){return r===0?"false":"true"}function sx(r,t,e,n,o,s=!0){let i=e==="complex64"?2:1,a=t[0],u=t.length;if(u===0){if(e==="complex64"){let h=Nf(r);return[kf(h[0],0,e)]}return e==="bool"?[b1(r[0])]:[r[0].toString()]}if(u===1){if(a>x1){let g=Sf*i,y=Array.from(r.slice(0,g)),b=Array.from(r.slice((a-Sf)*i,a*i));return e==="complex64"&&(y=Nf(y),b=Nf(b)),["["+y.map((w,v)=>kf(w,o[v],e)).join(", ")+", ..., "+b.map((w,v)=>kf(w,o[a-Sf+v],e)).join(", ")+"]"]}let h=e==="complex64"?Nf(r):Array.from(r);return["["+h.map((g,y)=>kf(g,o[y],e)).join(", ")+"]"]}let l=t.slice(1),c=n.slice(1),p=n[0]*i,m=[];if(a>x1){for(let h=0;h<Sf;h++){let g=h*p,y=g+p;m.push(...sx(r.slice(g,y),l,e,c,o,!1))}m.push("...");for(let h=a-Sf;h<a;h++){let g=h*p,y=g+p;m.push(...sx(r.slice(g,y),l,e,c,o,h===a-1))}}else for(let h=0;h<a;h++){let g=h*p,y=g+p;m.push(...sx(r.slice(g,y),l,e,c,o,h===a-1))}let f=u===2?",":"";m[0]="["+m[0]+f;for(let h=1;h<m.length-1;h++)m[h]=" "+m[h]+f;let d=`,
`;for(let h=2;h<u;h++)d+=`
`;return m[m.length-1]=" "+m[m.length-1]+"]"+(s?"":d),m}function Nf(r){let t=[];for(let e=0;e<r.length;e+=2)t.push([r[e],r[e+1]]);return t}var me=class{constructor(t,e,n){if(this.dtype=e,this.shape=t.slice(),this.size=re(t),n!=null){let o=n.length;A(o===this.size,()=>`Length of values '${o}' does not match the size inferred by the shape '${this.size}'.`)}if(e==="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||Nv(e,this.size),this.strides=di(t)}set(t,...e){e.length===0&&(e=[0]),A(e.length===this.rank,()=>`The number of provided coordinates (${e.length}) must match the rank (${this.rank})`);let n=this.locToIndex(e);this.values[n]=t}get(...t){t.length===0&&(t=[0]);let e=0;for(let o of t){if(o<0||o>=this.shape[e]){let s=`Requested out of range element at ${t}. Buffer shape=${this.shape}`;throw new Error(s)}e++}let n=t[t.length-1];for(let o=0;o<t.length-1;++o)n+=this.strides[o]*t[o];return this.values[n]}locToIndex(t){if(this.rank===0)return 0;if(this.rank===1)return t[0];let e=t[t.length-1];for(let n=0;n<t.length-1;++n)e+=this.strides[n]*t[n];return e}indexToLoc(t){if(this.rank===0)return[];if(this.rank===1)return[t];let e=new Array(this.shape.length);for(let n=0;n<e.length-1;++n)e[n]=Math.floor(t/this.strides[n]),t-=e[n]*this.strides[n];return e[e.length-1]=t,e}get rank(){return this.shape.length}toTensor(){return Ws().makeTensor(this.values,this.shape,this.dtype)}},Ws=null,Up=null,YU=null;function w1(r){Ws=r}function v1(r){Up=r}function C1(r){YU=r}var Vt=class{constructor(t,e,n,o){this.kept=!1,this.isDisposedInternal=!1,this.shape=t.slice(),this.dtype=e||"float32",this.size=re(t),this.strides=di(t),this.dataId=n,this.id=o,this.rankType=this.rank<5?this.rank.toString():"higher"}get rank(){return this.shape.length}async buffer(){let t=await this.data();return Up.buffer(this.shape,this.dtype,t)}bufferSync(){return Up.buffer(this.shape,this.dtype,this.dataSync())}async array(){let t=await this.data();return yu(this.shape,t,this.dtype==="complex64")}arraySync(){return yu(this.shape,this.dataSync(),this.dtype==="complex64")}async data(){this.throwIfDisposed();let t=Ws().read(this.dataId);if(this.dtype==="string"){let e=await t;try{return e.map(n=>Wp(n))}catch(n){throw new Error("Failed to decode the string bytes into utf-8. To get the original bytes, call tensor.bytes().")}}return t}dataToGPU(t){return this.throwIfDisposed(),Ws().readToGPU(this.dataId,t)}dataSync(){this.throwIfDisposed();let t=Ws().readSync(this.dataId);if(this.dtype==="string")try{return t.map(e=>Wp(e))}catch(e){throw new Error("Failed to decode the string bytes into utf-8. To get the original bytes, call tensor.bytes().")}return t}async bytes(){this.throwIfDisposed();let t=await Ws().read(this.dataId);return this.dtype==="string"?t:new Uint8Array(t.buffer)}dispose(){this.isDisposed||(Ws().disposeTensor(this),this.isDisposedInternal=!0)}get isDisposed(){return this.isDisposedInternal}throwIfDisposed(){if(this.isDisposed)throw new Error("Tensor is disposed.")}print(t=!1){return Up.print(this,t)}clone(){return this.throwIfDisposed(),Up.clone(this)}toString(t=!1){let e=this.dataSync();return y1(e,this.shape,this.dtype,t)}cast(t){return this.throwIfDisposed(),Up.cast(this,t)}variable(t=!0,e,n){return this.throwIfDisposed(),Ws().makeVariable(this,t,e,n)}};Object.defineProperty(Vt,Symbol.hasInstance,{value:r=>!!r&&r.data!=null&&r.dataSync!=null&&r.throwIfDisposed!=null});function L(){return xf("Tensor",()=>Vt)}L();var Ll=class extends Vt{constructor(t,e,n,o){super(t.shape,t.dtype,t.dataId,o);this.trainable=e,this.name=n}assign(t){if(t.dtype!==this.dtype)throw new Error(`dtype of the new value (${t.dtype}) and previous value (${this.dtype}) must match`);if(!Fn(t.shape,this.shape))throw new Error(`shape of the new value (${t.shape}) and previous value (${this.shape}) must match`);Ws().disposeTensor(this),this.dataId=t.dataId,Ws().incRef(this,null)}dispose(){Ws().disposeVariable(this),this.isDisposedInternal=!0}};Object.defineProperty(Ll,Symbol.hasInstance,{value:r=>r instanceof Vt&&r.assign!=null&&r.assign instanceof Function});var po={};Yt(po,{assertTypesMatch:()=>qv,getTensorsInContainer:()=>Tf,isTensorInList:()=>JU,makeTypesMatch:()=>jt});var Vv;(function(r){r.R0="R0",r.R1="R1",r.R2="R2",r.R3="R3",r.R4="R4",r.R5="R5",r.R6="R6"})(Vv||(Vv={}));var Gv;(function(r){r.float32="float32",r.int32="int32",r.bool="int32",r.complex64="complex64"})(Gv||(Gv={}));var Wv;(function(r){r.float32="float32",r.int32="int32",r.bool="bool",r.complex64="complex64"})(Wv||(Wv={}));var Uv;(function(r){r.float32="float32",r.int32="float32",r.bool="float32",r.complex64="complex64"})(Uv||(Uv={}));var Hv;(function(r){r.float32="complex64",r.int32="complex64",r.bool="complex64",r.complex64="complex64"})(Hv||(Hv={}));var ZU={float32:Uv,int32:Gv,bool:Wv,complex64:Hv};function ar(r,t){if(r==="string"||t==="string"){if(r==="string"&&t==="string")return"string";throw new Error(`Can not upcast ${r} with ${t}`)}return ZU[r][t]}function _u(r){return ar(r,"int32")}function jt(r,t){if(r.dtype===t.dtype)return[r,t];let e=ar(r.dtype,t.dtype);return[r.cast(e),t.cast(e)]}function qv(r,t){A(r.dtype===t.dtype,()=>`The dtypes of the first(${r.dtype}) and second(${t.dtype}) input must match`)}function JU(r,t){return t.some(e=>e.id===r.id)}function Tf(r){let t=[];return I1(r,t,new Set),t}function I1(r,t,e){if(r==null)return;if(r instanceof Vt){t.push(r);return}if(!QU(r))return;let n=r;for(let o in n){let s=n[o];e.has(s)||(e.add(s),I1(s,t,e))}}function QU(r){return Array.isArray(r)||typeof r=="object"}function Kv(r){return r.kernelName!=null}var jv=class{constructor(){this.registeredVariables={},this.nextTapeNodeId=0,this.numBytes=0,this.numTensors=0,this.numStringTensors=0,this.numDataBuffers=0,this.gradientDepth=0,this.kernelDepth=0,this.scopeStack=[],this.numDataMovesStack=[],this.nextScopeId=0,this.tensorInfo=new WeakMap,this.profiling=!1,this.activeProfile={newBytes:0,newTensors:0,peakBytes:0,kernels:[],result:null,get kernelNames(){return Array.from(new Set(this.kernels.map(t=>t.name)))}}}dispose(){for(let t in this.registeredVariables)this.registeredVariables[t].dispose()}},Eu=class{constructor(t){this.ENV=t,this.registry={},this.registryFactory={},this.pendingBackendInitId=0,this.state=new jv}async ready(){if(this.pendingBackendInit!=null)return this.pendingBackendInit.then(()=>{});if(this.backendInstance!=null)return;let t=this.getSortedBackends();for(let e=0;e<t.length;e++){let n=t[e];if(await this.initializeBackend(n).success){await this.setBackend(n);return}}throw new Error("Could not initialize any backends, all backend initializations failed.")}get backend(){if(this.pendingBackendInit!=null)throw new Error(`Backend '${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:t,asyncInit:e}=this.initializeBackendsAndReturnBest();if(e)throw new Error(`The highest priority backend '${t}' has not yet been initialized. Make sure to await tf.ready() or await tf.setBackend() before calling other methods`);this.setBackend(t)}return this.backendInstance}backendNames(){return Object.keys(this.registryFactory)}findBackend(t){if(!(t in this.registry))if(t in this.registryFactory){let{asyncInit:e}=this.initializeBackend(t);if(e)return null}else return null;return this.registry[t]}findBackendFactory(t){return t in this.registryFactory?this.registryFactory[t].factory:null}registerBackend(t,e,n=1){return t in this.registryFactory?(Ri(`${t} backend was already registered. Reusing existing backend factory.`),!1):(this.registryFactory[t]={factory:e,priority:n},!0)}async setBackend(t){if(this.registryFactory[t]==null)throw new Error(`Backend name '${t}' not found in registry`);if(this.backendName=t,this.registry[t]==null){this.backendInstance=null;let{success:e,asyncInit:n}=this.initializeBackend(t);if(!(n?await e:e))return!1}return this.backendInstance=this.registry[t],this.setupRegisteredKernels(),this.profiler=new zv(this.backendInstance),!0}setupRegisteredKernels(){rx(this.backendName).forEach(e=>{e.setupFunc!=null&&e.setupFunc(this.backendInstance)})}disposeRegisteredKernels(t){rx(t).forEach(n=>{n.disposeFunc!=null&&n.disposeFunc(this.registry[t])})}initializeBackend(t){let e=this.registryFactory[t];if(e==null)throw new Error(`Cannot initialize backend ${t}, no registration found.`);try{let n=e.factory();if(n&&!(n instanceof mi)&&typeof n.then=="function"){let o=++this.pendingBackendInitId,s=n.then(i=>o<this.pendingBackendInitId?!1:(this.registry[t]=i,this.pendingBackendInit=null,!0)).catch(i=>(o<this.pendingBackendInitId||(this.pendingBackendInit=null,Ri(`Initialization of backend ${t} failed`),Ri(i.stack||i.message)),!1));return this.pendingBackendInit=s,{success:s,asyncInit:!0}}else return this.registry[t]=n,{success:!0,asyncInit:!1}}catch(n){return Ri(`Initialization of backend ${t} failed`),Ri(n.stack||n.message),{success:!1,asyncInit:!1}}}removeBackend(t){if(!(t in this.registryFactory))throw new Error(`${t} backend not found in registry`);this.backendName===t&&this.pendingBackendInit!=null&&this.pendingBackendInitId++,t in this.registry&&(this.disposeRegisteredKernels(t),this.registry[t].dispose(),delete this.registry[t]),delete this.registryFactory[t],this.backendName===t&&(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((t,e)=>this.registryFactory[e].priority-this.registryFactory[t].priority)}initializeBackendsAndReturnBest(){let t=this.getSortedBackends();for(let e=0;e<t.length;e++){let n=t[e],{success:o,asyncInit:s}=this.initializeBackend(n);if(s||o)return{name:n,asyncInit:s}}throw new Error("Could not initialize any backends, all backend initializations failed.")}moveData(t,e){let n=this.state.tensorInfo.get(e),o=n.backend,s=this.readSync(e),i=o.refCount(e);o.disposeData(e,!0),n.backend=t,t.move(e,s,n.shape,n.dtype,i),this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack[this.state.numDataMovesStack.length-1]++}tidy(t,e){let n=null;if(e==null){if(typeof t!="function")throw new Error("Please provide a function to tidy()");e=t}else{if(typeof t!="string"&&!(t instanceof String))throw new Error("When calling with two arguments, the first argument to tidy() must be a string");if(typeof e!="function")throw new Error("When calling with two arguments, the 2nd argument to tidy() must be a function");n=t}let o;return this.scopedRun(()=>this.startScope(n),()=>this.endScope(o),()=>(o=e(),o instanceof Promise&&console.error("Cannot return a Promise inside of tidy."),o))}scopedRun(t,e,n){t();try{let o=n();return e(),o}catch(o){throw e(),o}}nextTensorId(){return Eu.nextTensorId++}nextVariableId(){return Eu.nextVariableId++}clone(t){let e=_.runKernel(io,{x:t}),n={x:t},o=i=>({x:()=>{let a="float32",u={x:i},l={dtype:a};return _.runKernel(oo,u,l)}}),s=[];return this.addTapeNode(this.state.activeScope.name,n,[e],o,s,{}),e}runKernel(t,e,n){if(this.backendName==null&&this.backend,!(Cf(t,this.backendName)!=null))throw new Error(`Kernel '${t}' not registered for backend '${this.backendName}'`);return this.runKernelFunc({kernelName:t,inputs:e,attrs:n})}shouldCheckForMemLeaks(){return this.ENV.getBool("IS_TEST")}checkKernelForMemLeak(t,e,n){let o=this.backend.numDataIds(),s=0;n.forEach(u=>{s+=u.dtype==="complex64"?3:1});let i=this.state.numDataMovesStack[this.state.numDataMovesStack.length-1],a=o-e-s-i;if(a>0)throw new Error(`Backend '${this.backendName}' has an internal memory leak (${a} data ids) after running '${t}'`)}runKernelFunc(t){let e,n=[],o=this.isTapeOn(),s=this.state.numBytes,i=this.state.numTensors;this.shouldCheckForMemLeaks()&&this.state.numDataMovesStack.push(0);let a;this.backendName==null&&this.backend;let u,l=Kv(t)?t.kernelName:this.state.activeScope!=null?this.state.activeScope.name:"";if(Kv(t)){let{kernelName:d,inputs:h,attrs:g}=t;this.backendName==null&&this.backend;let y=Cf(d,this.backendName);A(y!=null,()=>`Cannot find registered kernel '${d}' for backend '${this.backendName}'`),a=()=>{let b=this.backend.numDataIds();u=y.kernelFunc({inputs:h,attrs:g,backend:this.backend});let w=Array.isArray(u)?u:[u];this.shouldCheckForMemLeaks()&&this.checkKernelForMemLeak(d,b,w);let v=w.map(k=>{if(k.rank!=null)return k;let{dataId:E,shape:D,dtype:$}=k;return this.makeTensorFromDataId(E,D,$)});if(o){let k=this.getTensorsForGradient(d,h,v);n=this.saveTensorsForBackwardMode(k)}return v}}else{let{forwardFunc:d}=t,h=g=>{!o||(n=g.map(y=>this.keep(this.clone(y))))};a=()=>{let g=this.backend.numDataIds();u=this.tidy(()=>d(this.backend,h));let y=Array.isArray(u)?u:[u];return this.shouldCheckForMemLeaks()&&this.checkKernelForMemLeak(l,g,y),y}}let{inputs:c,attrs:p}=t,m=Kv(t)?null:t.backwardsFunc,f;return this.scopedRun(()=>this.state.kernelDepth++,()=>this.state.kernelDepth--,()=>{!this.ENV.getBool("DEBUG")&&!this.state.profiling?e=a():(f=this.profiler.profileKernel(l,c,()=>a()),this.ENV.getBool("DEBUG")&&this.profiler.logKernelProfile(f),e=f.outputs)}),o&&this.addTapeNode(l,c,e,m,n,p),this.state.profiling&&this.state.activeProfile.kernels.push({name:l,bytesAdded:this.state.numBytes-s,totalBytesSnapshot:this.state.numBytes,tensorsAdded:this.state.numTensors-i,totalTensorsSnapshot:this.state.numTensors,inputShapes:Object.keys(c).map(d=>c[d]!=null?c[d].shape:null),outputShapes:e.map(d=>d.shape),kernelTimeMs:f.timeMs,extraInfo:f.extraInfo}),Array.isArray(u)?e:e[0]}saveTensorsForBackwardMode(t){return t.map(n=>this.keep(this.clone(n)))}getTensorsForGradient(t,e,n){let o=Rv(t);if(o!=null){let s=o.inputsToSave||[],i=o.outputsToSave||[],a;o.saveAllInputs?(A(Array.isArray(e),()=>"saveAllInputs is true, expected inputs to be an array."),a=Object.keys(e).map(l=>e[l])):a=s.map(l=>e[l]);let u=n.filter((l,c)=>i[c]);return a.concat(u)}return[]}makeTensor(t,e,n,o){if(t==null)throw new Error("Values passed to engine.makeTensor() are null");n=n||"float32",o=o||this.backend;let s=t;n==="string"&&Ko(t[0])&&(s=t.map(u=>Ml(u)));let i=o.write(s,e,n),a=new Vt(e,n,i,this.nextTensorId());if(this.trackTensor(a,o),n==="string"){let u=this.state.tensorInfo.get(i),l=Ev(s);this.state.numBytes+=l-u.bytes,u.bytes=l}return a}makeTensorFromDataId(t,e,n,o){n=n||"float32";let s=new Vt(e,n,t,this.nextTensorId());return this.trackTensor(s,o),s}makeVariable(t,e=!0,n,o){n=n||this.nextVariableId().toString(),o!=null&&o!==t.dtype&&(t=t.cast(o));let s=new Ll(t,e,n,this.nextTensorId());if(this.state.registeredVariables[s.name]!=null)throw new Error(`Variable with name ${s.name} was already registered`);return this.state.registeredVariables[s.name]=s,this.incRef(s,this.backend),s}trackTensor(t,e){this.state.numTensors++,t.dtype==="string"&&this.state.numStringTensors++;let n=0;t.dtype!=="complex64"&&t.dtype!=="string"&&(n=t.size*tx(t.dtype)),this.state.numBytes+=n,this.state.tensorInfo.has(t.dataId)||(this.state.numDataBuffers++,this.state.tensorInfo.set(t.dataId,{backend:e||this.backend,dtype:t.dtype,shape:t.shape,bytes:n})),t instanceof Ll||this.track(t)}incRef(t,e){this.trackTensor(t,e),this.backend.incRef(t.dataId)}removeDataId(t,e){this.state.tensorInfo.has(t)&&this.state.tensorInfo.get(t).backend===e&&(this.state.tensorInfo.delete(t),this.state.numDataBuffers--)}disposeTensor(t){if(!this.state.tensorInfo.has(t.dataId))return;let e=this.state.tensorInfo.get(t.dataId);if(this.state.numTensors--,t.dtype==="string"&&(this.state.numStringTensors--,this.state.numBytes-=e.bytes),t.dtype!=="complex64"&&t.dtype!=="string"){let n=t.size*tx(t.dtype);this.state.numBytes-=n}e.backend.disposeData(t.dataId)&&this.removeDataId(t.dataId,e.backend)}disposeVariables(){for(let t in this.state.registeredVariables){let e=this.state.registeredVariables[t];this.disposeVariable(e)}}disposeVariable(t){this.disposeTensor(t),this.state.registeredVariables[t.name]!=null&&delete this.state.registeredVariables[t.name]}memory(){let t=this.backend.memory();return t.numTensors=this.state.numTensors,t.numDataBuffers=this.state.numDataBuffers,t.numBytes=this.state.numBytes,this.state.numStringTensors>0&&(t.unreliable=!0,t.reasons==null&&(t.reasons=[]),t.reasons.push("Memory usage by string tensors is approximate (2 bytes per character)")),t}async profile(t){this.state.profiling=!0;let e=this.state.numBytes,n=this.state.numTensors;this.state.activeProfile.kernels=[],this.state.activeProfile.result=await t(),this.state.profiling=!1,this.state.activeProfile.peakBytes=Math.max(...this.state.activeProfile.kernels.map(o=>o.totalBytesSnapshot)),this.state.activeProfile.newBytes=this.state.numBytes-e,this.state.activeProfile.newTensors=this.state.numTensors-n;for(let o of this.state.activeProfile.kernels)o.kernelTimeMs=await o.kernelTimeMs,o.extraInfo=await o.extraInfo;return this.state.activeProfile}isTapeOn(){return this.state.gradientDepth>0&&this.state.kernelDepth===0}addTapeNode(t,e,n,o,s,i){let a={id:this.state.nextTapeNodeId++,kernelName:t,inputs:e,outputs:n,saved:s},u=Rv(t);u!=null&&(o=u.gradFunc),o!=null&&(a.gradient=l=>(l=l.map((c,p)=>{if(c==null){let m=n[p],f=ap(m.size,m.dtype);return this.makeTensor(f,m.shape,m.dtype)}return c}),o(l.length>1?l:l[0],s,i))),this.state.activeTape.push(a)}keep(t){return t.kept=!0,t}startTape(){this.state.gradientDepth===0&&(this.state.activeTape=[]),this.state.gradientDepth++}endTape(){this.state.gradientDepth--}startScope(t){let e={track:[],name:"unnamed scope",id:this.state.nextScopeId++};t&&(e.name=t),this.state.scopeStack.push(e),this.state.activeScope=e}endScope(t){let e=Tf(t),n=new Set(e.map(s=>s.id));for(let s=0;s<this.state.activeScope.track.length;s++){let i=this.state.activeScope.track[s];!i.kept&&!n.has(i.id)&&i.dispose()}let o=this.state.scopeStack.pop();this.state.activeScope=this.state.scopeStack.length===0?null:this.state.scopeStack[this.state.scopeStack.length-1],e.forEach(s=>{!s.kept&&s.scopeId===o.id&&this.track(s)})}gradients(t,e,n,o=!1){if(A(e.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 s=this.scopedRun(()=>this.startTape(),()=>this.endTape(),()=>this.tidy("forward",t));A(s instanceof Vt,()=>"The result y returned by f() must be a tensor.");let i=h1(this.state.activeTape,e,s);if(!o&&i.length===0&&e.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 a={};a[s.id]=n==null?t4(s.shape):n,g1(a,i,l=>this.tidy(l),e4);let u=e.map(l=>a[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:s,grads:u}})}customGrad(t){return A(fi(t),()=>"The f passed in customGrad(f) must be a function."),(...e)=>{A(e.every(a=>a instanceof Vt),()=>"The args passed in customGrad(f)(x1, x2,...) must all be tensors");let n,o={};e.forEach((a,u)=>{o[u]=a});let s=(a,u)=>(n=t(...e,u),A(n.value instanceof Vt,()=>"The function f passed in customGrad(f) must return an object where `obj.value` is a tensor"),A(fi(n.gradFunc),()=>"The function f passed in customGrad(f) must return an object where `obj.gradFunc` is a function."),n.value),i=(a,u)=>{let l=n.gradFunc(a,u),c=Array.isArray(l)?l:[l];A(c.length===e.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(m=>m instanceof Vt),()=>"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 p={};return c.forEach((m,f)=>{p[f]=()=>m}),p};return this.runKernelFunc({forwardFunc:s,backwardsFunc:i,inputs:o})}}readSync(t){return this.state.tensorInfo.get(t).backend.readSync(t)}read(t){return this.state.tensorInfo.get(t).backend.read(t)}readToGPU(t,e){return this.state.tensorInfo.get(t).backend.readToGPU(t,e)}async time(t){let e=Tu(),n=await this.backend.time(t);return n.wallMs=Tu()-e,n}track(t){return this.state.activeScope!=null&&(t.scopeId=this.state.activeScope.id,this.state.activeScope.track.push(t)),t}get registeredVariables(){return this.state.registeredVariables}reset(){this.pendingBackendInitId++,this.state.dispose(),this.ENV.reset(),this.state=new jv;for(let t in this.registry)this.disposeRegisteredKernels(t),this.registry[t].dispose(),delete this.registry[t];this.backendName=null,this.backendInstance=null,this.pendingBackendInit=null}};Eu.nextTensorId=0;Eu.nextVariableId=0;function t4(r){let t=df(re(r),"float32");return _.makeTensor(t,r,"float32")}function Xv(){let r=$v();if(r._tfengine==null){let t=new ex(r);r._tfengine=new Eu(t)}return XT(r._tfengine.ENV),w1(()=>r._tfengine),r._tfengine}var _=Xv();function e4(r,t){let e={a:r,b:t};return _.runKernel(Kn,e)}var zl={};Yt(zl,{isBrowser:()=>Zv,isMobile:()=>o4,mockIsMobile:()=>n4});function r4(){return typeof navigator!="undefined"&&navigator!=null}var Yv;function n4(r){Yv=r}function o4(r){if(Yv!==void 0)return Yv;if(r||r4()){if(r||(r=navigator),r.product==="ReactNative")return!0;let t=r.userAgent||r.vendor||(typeof window!="undefined"?window.opera:"");if(!t){let e=r;return e.userAgentData&&e.userAgentData.mobile}return/(android|bb\d+|meego).+mobile|avantgo|bada\/|blackberry|blazer|compal|elaine|fennec|hiptop|iemobile|ip(hone|od)|iris|kindle|lge |maemo|midp|mmp|mobile.+firefox|netfront|opera m(ob|in)i|palm( os)?|phone|p(ixi|re)\/|plucker|pocket|psp|series(4|6)0|symbian|treo|up\.(browser|link)|vodafone|wap|windows ce|xda|xiino/i.test(t)||/1207|6310|6590|3gso|4thp|50[1-6]i|770s|802s|a wa|abac|ac(er|oo|s\-)|ai(ko|rn)|al(av|ca|co)|amoi|an(ex|ny|yw)|aptu|ar(ch|go)|as(te|us)|attw|au(di|\-m|r |s )|avan|be(ck|ll|nq)|bi(lb|rd)|bl(ac|az)|br(e|v)w|bumb|bw\-(n|u)|c55\/|capi|ccwa|cdm\-|cell|chtm|cldc|cmd\-|co(mp|nd)|craw|da(it|ll|ng)|dbte|dc\-s|devi|dica|dmob|do(c|p)o|ds(12|\-d)|el(49|ai)|em(l2|ul)|er(ic|k0)|esl8|ez([4-7]0|os|wa|ze)|fetc|fly(\-|_)|g1 u|g560|gene|gf\-5|g\-mo|go(\.w|od)|gr(ad|un)|haie|hcit|hd\-(m|p|t)|hei\-|hi(pt|ta)|hp( i|ip)|hs\-c|ht(c(\-| |_|a|g|p|s|t)|tp)|hu(aw|tc)|i\-(20|go|ma)|i230|iac( |\-|\/)|ibro|idea|ig01|ikom|im1k|inno|ipaq|iris|ja(t|v)a|jbro|jemu|jigs|kddi|keji|kgt( |\/)|klon|kpt |kwc\-|kyo(c|k)|le(no|xi)|lg( g|\/(k|l|u)|50|54|\-[a-w])|libw|lynx|m1\-w|m3ga|m50\/|ma(te|ui|xo)|mc(01|21|ca)|m\-cr|me(rc|ri)|mi(o8|oa|ts)|mmef|mo(01|02|bi|de|do|t(\-| |o|v)|zz)|mt(50|p1|v )|mwbp|mywa|n10[0-2]|n20[2-3]|n30(0|2)|n50(0|2|5)|n7(0(0|1)|10)|ne((c|m)\-|on|tf|wf|wg|wt)|nok(6|i)|nzph|o2im|op(ti|wv)|oran|owg1|p800|pan(a|d|t)|pdxg|pg(13|\-([1-8]|c))|phil|pire|pl(ay|uc)|pn\-2|po(ck|rt|se)|prox|psio|pt\-g|qa\-a|qc(07|12|21|32|60|\-[2-7]|i\-)|qtek|r380|r600|raks|rim9|ro(ve|zo)|s55\/|sa(ge|ma|mm|ms|ny|va)|sc(01|h\-|oo|p\-)|sdk\/|se(c(\-|0|1)|47|mc|nd|ri)|sgh\-|shar|sie(\-|m)|sk\-0|sl(45|id)|sm(al|ar|b3|it|t5)|so(ft|ny)|sp(01|h\-|v\-|v )|sy(01|mb)|t2(18|50)|t6(00|10|18)|ta(gt|lk)|tcl\-|tdg\-|tel(i|m)|tim\-|t\-mo|to(pl|sh)|ts(70|m\-|m3|m5)|tx\-9|up(\.b|g1|si)|utst|v400|v750|veri|vi(rg|te)|vk(40|5[0-3]|\-v)|vm40|voda|vulc|vx(52|53|60|61|70|80|81|83|85|98)|w3c(\-| )|webc|whit|wi(g |nc|nw)|wmlb|wonu|x700|yas\-|your|zeto|zte\-/i.test(t.substr(0,4))}return!1}function Zv(){return typeof window!="undefined"&&window.document!=null||typeof WorkerGlobalScope!="undefined"}var mo=V();mo.registerFlag("DEBUG",()=>!1,r=>{r&&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.")});mo.registerFlag("IS_BROWSER",()=>Zv());mo.registerFlag("IS_NODE",()=>typeof process!="undefined"&&typeof process.versions!="undefined"&&typeof process.versions.node!="undefined");mo.registerFlag("IS_CHROME",()=>typeof navigator!="undefined"&&navigator!=null&&navigator.userAgent!=null&&/Chrome/.test(navigator.userAgent)&&/Google Inc/.test(navigator.vendor));mo.registerFlag("PROD",()=>!1);mo.registerFlag("TENSORLIKE_CHECK_SHAPE_CONSISTENCY",()=>mo.getBool("DEBUG"));mo.registerFlag("DEPRECATION_WARNINGS_ENABLED",()=>!0);mo.registerFlag("IS_TEST",()=>!1);mo.registerFlag("CHECK_COMPUTATION_FOR_ERRORS",()=>!0);mo.registerFlag("WRAP_TO_IMAGEBITMAP",()=>!1);mo.registerFlag("ENGINE_COMPILE_ONLY",()=>!1);function Or(r,t){let e=r;if(yr(r))return t==="string"?[]:[r.length];if(!Array.isArray(r))return[];let n=[];for(;Array.isArray(e)||yr(e)&&t!=="string";)n.push(e.length),e=e[0];return Array.isArray(r)&&V().getBool("TENSORLIKE_CHECK_SHAPE_CONSISTENCY")&&k1(r,n,[]),n}function k1(r,t,e){if(e=e||[],!Array.isArray(r)&&!yr(r)){A(t.length===0,()=>`Element arr[${e.join("][")}] is a primitive, but should be an array/TypedArray of ${t[0]} elements`);return}A(t.length>0,()=>`Element arr[${e.join("][")}] should be a primitive, but is an array of ${r.length} elements`),A(r.length===t[0],()=>`Element arr[${e.join("][")}] should have ${t[0]} elements, but has ${r.length} elements`);let n=t.slice(1);for(let o=0;o<r.length;++o)k1(r[o],n,e.concat(o))}function S1(r,t,e,n){if(r!=="string_or_numeric"){if(r==null)throw new Error("Expected dtype cannot be null.");if(r!=="numeric"&&r!==t||r==="numeric"&&t==="string")throw new Error(`Argument '${e}' passed to '${n}' must be ${r} tensor, but got ${t} tensor`)}}function C(r,t,e,n="numeric"){if(r instanceof Vt)return S1(n,r.dtype,t,e),r;let o=sp(r);if(o!=="string"&&["bool","int32","float32"].indexOf(n)>=0&&(o=n),S1(n,o,t,e),r==null||!yr(r)&&!Array.isArray(r)&&typeof r!="number"&&typeof r!="boolean"&&typeof r!="string"){let u=r==null?"null":r.constructor.name;throw new Error(`Argument '${t}' passed to '${e}' must be a Tensor or TensorLike, but got '${u}'`)}let s=Or(r,o);!yr(r)&&!Array.isArray(r)&&(r=[r]);let a=o!=="string"?Gp(r,o):qo(r,[],!0);return _.makeTensor(a,s,o)}function Pa(r,t,e,n="numeric"){if(!Array.isArray(r))throw new Error(`Argument ${t} passed to ${e} must be a \`Tensor[]\` or \`TensorLike[]\``);return r.map((s,i)=>C(s,`${t}[${i}]`,e,n))}var N1="__op";function N(r){let t=Object.keys(r);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 e=t[0],n=r[e];e.endsWith("_")&&(e=e.substring(0,e.length-1)),e=e+N1;let o=(...s)=>{_.startScope(e);try{let i=n(...s);return gf(i)&&console.error("Cannot return a Promise inside of tidy."),_.endScope(i),i}catch(i){throw _.endScope(null),i}};return Object.defineProperty(o,"name",{value:e,configurable:!0}),o}function s4(r,t){let e=C(r,"real","complex"),n=C(t,"imag","complex");Pe(e.shape,n.shape,`real and imag shapes, ${e.shape} and ${n.shape}, must match in call to tf.complex().`);let o={real:e,imag:n};return _.runKernel(mp,o)}var Mn=N({complex_:s4});function sn(r,t,e,n){if(n==null&&(n=sp(r)),n==="complex64")throw new Error("Cannot construct a complex64 tensor directly. Please use tf.complex(real, imag).");if(!yr(r)&&!Array.isArray(r)&&typeof r!="number"&&typeof r!="boolean"&&typeof r!="string")throw new Error("values passed to tensor(values) must be a number/boolean/string or an array of numbers/booleans/strings, or a TypedArray");if(t!=null){hf(t);let o=re(t),s=re(e);A(o===s,()=>`Based on the provided shape, [${t}], the tensor should have ${o} values but has ${s}`);for(let i=0;i<e.length;++i){let a=e[i],u=i===e.length-1?a!==re(t.slice(i)):!0;A(e[i]===t[i]||!u,()=>`Error creating a new Tensor. Inferred shape (${e}) does not match the provided shape (${t}). `)}}return!yr(r)&&!Array.isArray(r)&&(r=[r]),t=t||e,r=n!=="string"?Gp(r,n):qo(r,[],!0),_.makeTensor(r,t,n)}function Ar(r,t,e){let n=Or(r,e);return sn(r,t,n,e)}var _f={float32:4,float16:2,int32:4,uint16:2,uint8:1,bool:1,complex64:8};var ix=4;async function _1(r,t){let e=[],n=[],o=Array.isArray(r)?r.map(i=>i.name):Object.keys(r);for(let i=0;i<o.length;++i){let a=o[i],u=Array.isArray(r)?r[i].tensor:r[a];if(u.dtype!=="float32"&&u.dtype!=="int32"&&u.dtype!=="bool"&&u.dtype!=="string"&&u.dtype!=="complex64")throw new Error(`Unsupported dtype in weight '${a}': ${u.dtype}`);let l={name:a,shape:u.shape,dtype:u.dtype};if(u.dtype==="string"){let c=new Promise(async p=>{let m=await u.bytes(),f=m.reduce((g,y)=>g+y.length,0)+ix*m.length,d=new Uint8Array(f),h=0;for(let g=0;g<m.length;g++){let y=m[g],b=new Uint8Array(new Uint32Array([y.length]).buffer);d.set(b,h),h+=ix,d.set(y,h),h+=y.length}p(d)});n.push(c)}else n.push(u.data());t!=null&&(l.group=t),e.push(l)}let s=await Promise.all(n);return{data:i4(s),specs:e}}function ax(r,t){let e={},n,o=0;for(let s of t){let i=s.name,a=s.dtype,u=s.shape,l=re(u),c;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(a!=="float32")throw new Error(`Weight ${s.name} is quantized with ${p.dtype} which only supports weights of type float32 not ${a}.`)}else throw new Error(`Weight ${s.name} has unknown quantization dtype ${p.dtype}. Supported quantization dtypes are: 'uint8', 'uint16', and 'float16'.`);let m=_f[p.dtype],f=r.slice(o,o+l*m),d=p.dtype==="uint8"?new Uint8Array(f):new Uint16Array(f);if(a==="float32")if(p.dtype==="uint8"||p.dtype==="uint16"){c=new Float32Array(d.length);for(let h=0;h<d.length;h++){let g=d[h];c[h]=g*p.scale+p.min}}else if(p.dtype==="float16")n===void 0&&(n=c4()),c=n(d);else throw new Error(`Unsupported quantization type ${p.dtype} for weight type float32.`);else if(a==="int32"){if(p.dtype!=="uint8"&&p.dtype!=="uint16")throw new Error(`Unsupported quantization type ${p.dtype} for weight type int32.`);c=new Int32Array(d.length);for(let h=0;h<d.length;h++){let g=d[h];c[h]=Math.round(g*p.scale+p.min)}}else throw new Error(`Unsupported dtype in weight '${i}': ${a}`);o+=l*m}else if(a==="string"){let p=re(s.shape);c=[];for(let m=0;m<p;m++){let f=new Uint32Array(r.slice(o,o+ix))[0];o+=ix;let d=new Uint8Array(r.slice(o,o+f));c.push(d),o+=f}}else{let p=_f[a],m=r.slice(o,o+l*p);if(a==="float32")c=new Float32Array(m);else if(a==="int32")c=new Int32Array(m);else if(a==="bool")c=new Uint8Array(m);else if(a==="complex64"){c=new Float32Array(m);let f=new Float32Array(c.length/2),d=new Float32Array(c.length/2);for(let y=0;y<f.length;y++)f[y]=c[y*2],d[y]=c[y*2+1];let h=Ar(f,u,"float32"),g=Ar(d,u,"float32");e[i]=Mn(h,g),h.dispose(),g.dispose()}else throw new Error(`Unsupported dtype in weight '${i}': ${a}`);o+=l*p}a!=="complex64"&&(e[i]=Ar(c,u,a))}return e}function i4(r){if(r===null)throw new Error(`Invalid input value: ${JSON.stringify(r)}`);let t=0,e=[];r.forEach(s=>{if(t+=s.byteLength,e.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 n=new Uint8Array(t),o=0;return e.forEach(s=>{n.set(new Uint8Array(s.buffer),o),o+=s.byteLength}),n.buffer}var Jv=typeof Buffer!="undefined"&&(typeof Blob=="undefined"||typeof atob=="undefined"||typeof btoa=="undefined");function T1(r){return Jv?Buffer.byteLength(r):new Blob([r]).size}function E1(r){if(Jv)return Buffer.from(r).toString("base64");let t=new Uint8Array(r),e="";for(let n=0,o=t.length;n<o;n++)e+=String.fromCharCode(t[n]);return btoa(e)}function A1(r){if(Jv){let n=Buffer.from(r,"base64");return n.buffer.slice(n.byteOffset,n.byteOffset+n.byteLength)}let t=atob(r),e=new Uint8Array(t.length);for(let n=0;n<t.length;++n)e.set([t.charCodeAt(n)],n);return e.buffer}function Hp(r){if(r.length===1)return r[0];let t=0;r.forEach(o=>{t+=o.byteLength});let e=new Uint8Array(t),n=0;return r.forEach(o=>{e.set(new Uint8Array(o),n),n+=o.byteLength}),e.buffer}function Qv(r){let t="/";for(r=r.trim();r.endsWith(t);)r=r.slice(0,r.length-1);let e=r.split(t);return e[e.length-1]}function lx(r,t){let e={modelTopology:r.modelTopology,format:r.format,generatedBy:r.generatedBy,convertedBy:r.convertedBy,weightsManifest:t};return r.signature!=null&&(e.signature=r.signature),r.userDefinedMetadata!=null&&(e.userDefinedMetadata=r.userDefinedMetadata),r.modelInitializer!=null&&(e.modelInitializer=r.modelInitializer),r.trainingConfig!=null&&(e.trainingConfig=r.trainingConfig),e}async function qp(r,t){let e={modelTopology:r.modelTopology,format:r.format,generatedBy:r.generatedBy,convertedBy:r.convertedBy};if(r.trainingConfig!=null&&(e.trainingConfig=r.trainingConfig),r.weightsManifest!=null){let[n,o]=await t(r.weightsManifest);e.weightSpecs=n,e.weightData=o}return r.signature!=null&&(e.signature=r.signature),r.userDefinedMetadata!=null&&(e.userDefinedMetadata=r.userDefinedMetadata),r.modelInitializer!=null&&(e.modelInitializer=r.modelInitializer),e}function Fi(r){if(r.modelTopology instanceof ArrayBuffer)throw new Error("Expected JSON model topology, received ArrayBuffer.");return{dateSaved:new Date,modelTopologyType:"JSON",modelTopologyBytes:r.modelTopology==null?0:T1(JSON.stringify(r.modelTopology)),weightSpecsBytes:r.weightSpecs==null?0:T1(JSON.stringify(r.weightSpecs)),weightDataBytes:r.weightData==null?0:r.weightData.byteLength}}function a4(){let r=e=>{let n=e<<13,o=0;for(;(n&8388608)===0;)o-=8388608,n<<=1;return n&=-8388609,o+=947912704,n|o},t=new Uint32Array(2048);t[0]=0;for(let e=1;e<1024;e++)t[e]=r(e);for(let e=1024;e<2048;e++)t[e]=939524096+(e-1024<<13);return t}function l4(){let r=new Uint32Array(64);r[0]=0,r[31]=1199570944,r[32]=2147483648,r[63]=3347054592;for(let t=1;t<31;t++)r[t]=t<<23;for(let t=33;t<63;t++)r[t]=2147483648+(t-32<<23);return r}function u4(){let r=new Uint32Array(64);for(let t=0;t<64;t++)r[t]=1024;return r[0]=r[32]=0,r}function c4(){let r=a4(),t=l4(),e=u4();return n=>{let o=new ArrayBuffer(4*n.length),s=new Uint32Array(o);for(let i=0;i<n.length;i++){let a=n[i],u=r[e[a>>10]+(a&1023)]+t[a>>10];s[i]=u}return new Float32Array(o)}}var De=class{constructor(){this.saveRouters=[],this.loadRouters=[]}static getInstance(){return De.instance==null&&(De.instance=new De),De.instance}static registerSaveRouter(t){De.getInstance().saveRouters.push(t)}static registerLoadRouter(t){De.getInstance().loadRouters.push(t)}static getSaveHandlers(t){return De.getHandlers(t,"save")}static getLoadHandlers(t,e){return De.getHandlers(t,"load",e)}static getHandlers(t,e,n){let o=[];return(e==="load"?De.getInstance().loadRouters:De.getInstance().saveRouters).forEach(i=>{let a=i(t,n);a!==null&&o.push(a)}),o}},D1=r=>De.registerSaveRouter(r),$1=r=>De.registerLoadRouter(r),R1=r=>De.getSaveHandlers(r),F1=(r,t)=>De.getLoadHandlers(r,t);var t0="tensorflowjs",e0=1,Au="models_store",Bl="model_info_store";function O1(){if(!V().getBool("IS_BROWSER"))throw new Error("Failed to obtain IndexedDB factory because the current environmentis not a web browser.");let r=typeof window=="undefined"?self:window,t=r.indexedDB||r.mozIndexedDB||r.webkitIndexedDB||r.msIndexedDB||r.shimIndexedDB;if(t==null)throw new Error("The current browser does not appear to support IndexedDB.");return t}function r0(r){let t=r.result;t.createObjectStore(Au,{keyPath:"modelPath"}),t.createObjectStore(Bl,{keyPath:"modelPath"})}var Ma=class{constructor(t){if(this.indexedDB=O1(),t==null||!t)throw new Error("For IndexedDB, modelPath must not be null, undefined or empty.");this.modelPath=t}async save(t){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");return this.databaseAction(this.modelPath,t)}async load(){return this.databaseAction(this.modelPath)}databaseAction(t,e){return new Promise((n,o)=>{let s=this.indexedDB.open(t0,e0);s.onupgradeneeded=()=>r0(s),s.onsuccess=()=>{let i=s.result;if(e==null){let a=i.transaction(Au,"readonly"),l=a.objectStore(Au).get(this.modelPath);l.onsuccess=()=>{if(l.result==null)return i.close(),o(new Error(`Cannot find model with path '${this.modelPath}' in IndexedDB.`));n(l.result.modelArtifacts)},l.onerror=c=>(i.close(),o(l.error)),a.oncomplete=()=>i.close()}else{let a=Fi(e),u=i.transaction(Bl,"readwrite"),l=u.objectStore(Bl),c=l.put({modelPath:this.modelPath,modelArtifactsInfo:a}),p;c.onsuccess=()=>{p=i.transaction(Au,"readwrite");let f=p.objectStore(Au).put({modelPath:this.modelPath,modelArtifacts:e,modelArtifactsInfo:a});f.onsuccess=()=>n({modelArtifactsInfo:a}),f.onerror=d=>{l=u.objectStore(Bl);let h=l.delete(this.modelPath);h.onsuccess=()=>(i.close(),o(f.error)),h.onerror=g=>(i.close(),o(f.error))}},c.onerror=m=>(i.close(),o(c.error)),u.oncomplete=()=>{p==null?i.close():p.oncomplete=()=>i.close()}}},s.onerror=i=>o(s.error)})}};Ma.URL_SCHEME="indexeddb://";var P1=r=>V().getBool("IS_BROWSER")&&!Array.isArray(r)&&r.startsWith(Ma.URL_SCHEME)?p4(r.slice(Ma.URL_SCHEME.length)):null;De.registerSaveRouter(P1);De.registerLoadRouter(P1);function p4(r){return new Ma(r)}function m4(r){return r.startsWith(Ma.URL_SCHEME)?r.slice(Ma.URL_SCHEME.length):r}var n0=class{constructor(){this.indexedDB=O1()}async listModels(){return new Promise((t,e)=>{let n=this.indexedDB.open(t0,e0);n.onupgradeneeded=()=>r0(n),n.onsuccess=()=>{let o=n.result,s=o.transaction(Bl,"readonly"),a=s.objectStore(Bl).getAll();a.onsuccess=()=>{let u={};for(let l of a.result)u[l.modelPath]=l.modelArtifactsInfo;t(u)},a.onerror=u=>(o.close(),e(a.error)),s.oncomplete=()=>o.close()},n.onerror=o=>e(n.error)})}async removeModel(t){return t=m4(t),new Promise((e,n)=>{let o=this.indexedDB.open(t0,e0);o.onupgradeneeded=()=>r0(o),o.onsuccess=()=>{let s=o.result,i=s.transaction(Bl,"readwrite"),a=i.objectStore(Bl),u=a.get(t),l;u.onsuccess=()=>{if(u.result==null)return s.close(),n(new Error(`Cannot find model with path '${t}' in IndexedDB.`));{let c=a.delete(t),p=()=>{l=s.transaction(Au,"readwrite");let f=l.objectStore(Au).delete(t);f.onsuccess=()=>e(u.result.modelArtifactsInfo),f.onerror=d=>n(u.error)};c.onsuccess=p,c.onerror=m=>(p(),s.close(),n(u.error))}},u.onerror=c=>(s.close(),n(u.error)),i.oncomplete=()=>{l==null?s.close():l.oncomplete=()=>s.close()}},o.onerror=s=>n(o.error)})}};var La="/",Kp="tensorflowjs_models",M1="info",f4="model_topology",d4="weight_specs",h4="weight_data",g4="model_metadata";function L1(r){return{info:[Kp,r,M1].join(La),topology:[Kp,r,f4].join(La),weightSpecs:[Kp,r,d4].join(La),weightData:[Kp,r,h4].join(La),modelMetadata:[Kp,r,g4].join(La)}}function z1(r){for(let t of Object.values(r))window.localStorage.removeItem(t)}function x4(r){let t=r.split(La);if(t.length<3)throw new Error(`Invalid key format: ${r}`);return t.slice(1,t.length-1).join(La)}function y4(r){return r.startsWith(za.URL_SCHEME)?r.slice(za.URL_SCHEME.length):r}var za=class{constructor(t){if(!V().getBool("IS_BROWSER")||typeof window=="undefined"||typeof window.localStorage=="undefined")throw new Error("The current environment does not support local storage.");if(this.LS=window.localStorage,t==null||!t)throw new Error("For local storage, modelPath must not be null, undefined or empty.");this.modelPath=t,this.keys=L1(this.modelPath)}async save(t){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserLocalStorage.save() does not support saving model topology in binary formats yet.");{let e=JSON.stringify(t.modelTopology),n=JSON.stringify(t.weightSpecs),o=Fi(t);try{this.LS.setItem(this.keys.info,JSON.stringify(o)),this.LS.setItem(this.keys.topology,e),this.LS.setItem(this.keys.weightSpecs,n),this.LS.setItem(this.keys.weightData,E1(t.weightData));let s={format:t.format,generatedBy:t.generatedBy,convertedBy:t.convertedBy,signature:t.signature!=null?t.signature:void 0,userDefinedMetadata:t.userDefinedMetadata!=null?t.userDefinedMetadata:void 0,modelInitializer:t.modelInitializer!=null?t.modelInitializer:void 0,trainingConfig:t.trainingConfig!=null?t.trainingConfig:void 0};return this.LS.setItem(this.keys.modelMetadata,JSON.stringify(s)),{modelArtifactsInfo:o}}catch(s){throw z1(this.keys),new Error(`Failed to save model '${this.modelPath}' to local storage: size quota being exceeded is a possible cause of this failure: modelTopologyBytes=${o.modelTopologyBytes}, weightSpecsBytes=${o.weightSpecsBytes}, weightDataBytes=${o.weightDataBytes}.`)}}}async load(){let t=JSON.parse(this.LS.getItem(this.keys.info));if(t==null)throw new Error(`In local storage, there is no model with name '${this.modelPath}'`);if(t.modelTopologyType!=="JSON")throw new Error("BrowserLocalStorage does not support loading non-JSON model topology yet.");let e={},n=JSON.parse(this.LS.getItem(this.keys.topology));if(n==null)throw new Error(`In local storage, the topology of model '${this.modelPath}' is missing.`);e.modelTopology=n;let o=JSON.parse(this.LS.getItem(this.keys.weightSpecs));if(o==null)throw new Error(`In local storage, the weight specs of model '${this.modelPath}' are missing.`);e.weightSpecs=o;let s=this.LS.getItem(this.keys.modelMetadata);if(s!=null){let a=JSON.parse(s);e.format=a.format,e.generatedBy=a.generatedBy,e.convertedBy=a.convertedBy,a.signature!=null&&(e.signature=a.signature),a.userDefinedMetadata!=null&&(e.userDefinedMetadata=a.userDefinedMetadata),a.modelInitializer!=null&&(e.modelInitializer=a.modelInitializer),a.trainingConfig!=null&&(e.trainingConfig=a.trainingConfig)}let i=this.LS.getItem(this.keys.weightData);if(i==null)throw new Error(`In local storage, the binary weight values of model '${this.modelPath}' are missing.`);return e.weightData=A1(i),e}};za.URL_SCHEME="localstorage://";var B1=r=>V().getBool("IS_BROWSER")&&!Array.isArray(r)&&r.startsWith(za.URL_SCHEME)?b4(r.slice(za.URL_SCHEME.length)):null;De.registerSaveRouter(B1);De.registerLoadRouter(B1);function b4(r){return new za(r)}var o0=class{constructor(){A(V().getBool("IS_BROWSER"),()=>"Current environment is not a web browser"),A(typeof window=="undefined"||typeof window.localStorage!="undefined",()=>"Current browser does not appear to support localStorage"),this.LS=window.localStorage}async listModels(){let t={},e=Kp+La,n=La+M1;for(let o=0;o<this.LS.length;++o){let s=this.LS.key(o);if(s.startsWith(e)&&s.endsWith(n)){let i=x4(s);t[i]=JSON.parse(this.LS.getItem(s))}}return t}async removeModel(t){t=y4(t);let e=L1(t);if(this.LS.getItem(e.info)==null)throw new Error(`Cannot find model at path '${t}'`);let n=JSON.parse(this.LS.getItem(e.info));return z1(e),n}};var jp="://",an=class{constructor(){this.managers={}}static getInstance(){return an.instance==null&&(an.instance=new an),an.instance}static registerManager(t,e){A(t!=null,()=>"scheme must not be undefined or null."),t.endsWith(jp)&&(t=t.slice(0,t.indexOf(jp))),A(t.length>0,()=>"scheme must not be an empty string.");let n=an.getInstance();A(n.managers[t]==null,()=>`A model store manager is already registered for scheme '${t}'.`),n.managers[t]=e}static getManager(t){let e=this.getInstance().managers[t];if(e==null)throw new Error(`Cannot find model manager for scheme '${t}'`);return e}static getSchemes(){return Object.keys(this.getInstance().managers)}};function ux(r){if(r.indexOf(jp)===-1)throw new Error(`The url string provided does not contain a scheme. Supported schemes are: ${an.getSchemes().join(",")}`);return{scheme:r.split(jp)[0],path:r.split(jp)[1]}}async function V1(r,t,e=!1){A(r!==t,()=>`Old path and new path are the same: '${r}'`);let n=De.getLoadHandlers(r);A(n.length>0,()=>`Copying failed because no load handler is found for source URL ${r}.`),A(n.length<2,()=>`Copying failed because more than one (${n.length}) load handlers for source URL ${r}.`);let o=n[0],s=De.getSaveHandlers(t);A(s.length>0,()=>`Copying failed because no save handler is found for destination URL ${t}.`),A(s.length<2,()=>`Copying failed because more than one (${n.length}) save handlers for destination URL ${t}.`);let i=s[0],a=ux(r).scheme,u=ux(r).path,l=a===ux(r).scheme,c=await o.load();e&&l&&await an.getManager(a).removeModel(u);let p=await i.save(c);return e&&!l&&await an.getManager(a).removeModel(u),p.modelArtifactsInfo}async function G1(){let r=an.getSchemes(),t={};for(let e of r){let n=await an.getManager(e).listModels();for(let o in n){let s=e+jp+o;t[s]=n[o]}}return t}async function W1(r){let t=ux(r);return an.getManager(t.scheme).removeModel(t.path)}async function U1(r,t){return V1(r,t,!1)}async function H1(r,t){return V1(r,t,!0)}var q1=class{fetch(t,e){return fetch(t,e)}now(){return performance.now()}encode(t,e){if(e!=="utf-8"&&e!=="utf8")throw new Error(`Browser's encoder only supports utf-8, but got ${e}`);return this.textEncoder==null&&(this.textEncoder=new TextEncoder),this.textEncoder.encode(t)}decode(t,e){return new TextDecoder(e).decode(t)}};if(V().get("IS_BROWSER")){V().setPlatform("browser",new q1);try{an.registerManager(za.URL_SCHEME,new o0)}catch(r){}try{an.registerManager(Ma.URL_SCHEME,new n0)}catch(r){}}var w4={importFetch:()=>K1()},s0;var X1=class{constructor(){this.util=j1(),this.textEncoder=new this.util.TextEncoder}fetch(t,e){return V().global.fetch!=null?V().global.fetch(t,e):(s0==null&&(s0=w4.importFetch()),s0(t,e))}now(){let t=process.hrtime();return t[0]*1e3+t[1]/1e6}encode(t,e){if(e!=="utf-8"&&e!=="utf8")throw new Error(`Node built-in encoder only supports utf-8, but got ${e}`);return this.textEncoder.encode(t)}decode(t,e){return t.length===0?"":new this.util.TextDecoder(e).decode(t)}};V().get("IS_NODE")&&!V().get("IS_BROWSER")&&V().setPlatform("node",new X1);function It(r,t="float32",e){return t=t||"float32",hf(r),new me(r,t,e)}function v4(r,t){let e=C(r,"x","cast");if(!_v(t))throw new Error(`Failed to cast to unknown dtype ${t}`);if(t==="string"&&e.dtype!=="string"||t!=="string"&&e.dtype==="string")throw new Error("Only strings can be casted to strings");let n={x:e},o={dtype:t};return _.runKernel(oo,n,o)}var Z=N({cast_:v4});function C4(r){let e={x:C(r,"x","clone","string_or_numeric")};return _.runKernel(io,e)}var In=N({clone_:C4});function i0(r,t=!1){console.log(r.toString(t))}Xv();var I4={buffer:It,cast:Z,clone:In,print:i0};v1(I4);var Pr={};Yt(Pr,{browserFiles:()=>J1,browserHTTPRequest:()=>e_,concatenateArrayBuffers:()=>Hp,copyModel:()=>U1,decodeWeights:()=>ax,encodeWeights:()=>_1,fromMemory:()=>n_,getLoadHandlers:()=>F1,getModelArtifactsForJSON:()=>qp,getModelArtifactsInfoForJSON:()=>Fi,getSaveHandlers:()=>R1,http:()=>mx,isHTTPScheme:()=>cx,listModels:()=>G1,loadWeights:()=>Q1,moveModel:()=>H1,registerLoadRouter:()=>$1,registerSaveRouter:()=>D1,removeModel:()=>W1,weightsLoaderFactory:()=>u0,withSaveHandler:()=>o_});var S4="model",k4=".json",N4=".weights.bin";function Y1(r){return new Promise(t=>setTimeout(t)).then(r)}var Vl=class{constructor(t){if(!V().getBool("IS_BROWSER"))throw new Error("browserDownloads() cannot proceed because the current environment is not a browser.");t.startsWith(Vl.URL_SCHEME)&&(t=t.slice(Vl.URL_SCHEME.length)),(t==null||t.length===0)&&(t=S4),this.modelJsonFileName=t+k4,this.weightDataFileName=t+N4}async save(t){if(typeof document=="undefined")throw new Error("Browser downloads are not supported in this environment since `document` is not present");let e=window.URL.createObjectURL(new Blob([t.weightData],{type:"application/octet-stream"}));if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserDownloads.save() does not support saving model topology in binary formats yet.");{let n=[{paths:["./"+this.weightDataFileName],weights:t.weightSpecs}],o=lx(t,n),s=window.URL.createObjectURL(new Blob([JSON.stringify(o)],{type:"application/json"})),i=this.modelJsonAnchor==null?document.createElement("a"):this.modelJsonAnchor;if(i.download=this.modelJsonFileName,i.href=s,await Y1(()=>i.dispatchEvent(new MouseEvent("click"))),t.weightData!=null){let a=this.weightDataAnchor==null?document.createElement("a"):this.weightDataAnchor;a.download=this.weightDataFileName,a.href=e,await Y1(()=>a.dispatchEvent(new MouseEvent("click")))}return{modelArtifactsInfo:Fi(t)}}}};Vl.URL_SCHEME="downloads://";var Z1=class{constructor(t){if(t==null||t.length<1)throw new Error(`When calling browserFiles, at least 1 file is required, but received ${t}`);this.jsonFile=t[0],this.weightsFiles=t.slice(1)}async load(){return new Promise((t,e)=>{let n=new FileReader;n.onload=o=>{let s=JSON.parse(o.target.result),i=s.modelTopology;if(i==null){e(new Error(`modelTopology field is missing from file ${this.jsonFile.name}`));return}if(s.weightsManifest==null){e(new Error(`weightManifest field is missing from file ${this.jsonFile.name}`));return}if(this.weightsFiles.length===0){t({modelTopology:i});return}let u=qp(s,l=>this.loadWeights(l));t(u)},n.onerror=o=>e(`Failed to read model topology and weights manifest JSON from file '${this.jsonFile.name}'. BrowserFiles supports loading Keras-style tf.Model artifacts only.`),n.readAsText(this.jsonFile)})}loadWeights(t){let e=[],n=[];for(let i of t)e.push(...i.weights),n.push(...i.paths);let o=this.checkManifestAndWeightFiles(t),s=n.map(i=>this.loadWeightsFile(i,o[i]));return Promise.all(s).then(i=>[e,Hp(i)])}loadWeightsFile(t,e){return new Promise((n,o)=>{let s=new FileReader;s.onload=i=>{let a=i.target.result;n(a)},s.onerror=i=>o(`Failed to weights data from file of path '${t}'.`),s.readAsArrayBuffer(e)})}checkManifestAndWeightFiles(t){let e=[],n=this.weightsFiles.map(s=>Qv(s.name)),o={};for(let s of t)s.paths.forEach(i=>{let a=Qv(i);if(e.indexOf(a)!==-1)throw new Error(`Duplicate file basename found in weights manifest: '${a}'`);if(e.push(a),n.indexOf(a)===-1)throw new Error(`Weight file with basename '${a}' is not provided.`);o[i]=this.weightsFiles[n.indexOf(a)]});if(e.length!==this.weightsFiles.length)throw new Error(`Mismatch in the number of files in weights manifest (${e.length}) and the number of weight files provided (${this.weightsFiles.length}).`);return o}},T4=r=>V().getBool("IS_BROWSER")&&!Array.isArray(r)&&r.startsWith(Vl.URL_SCHEME)?_4(r.slice(Vl.URL_SCHEME.length)):null;De.registerSaveRouter(T4);function _4(r="model"){return new Vl(r)}function J1(r){return new Z1(r)}function a0(r,t,e,n){i(r),e=e==null?0:e,n=n==null?1:n,a(e,n);let o=0,s=u=>(u.then(l=>{let c=e+ ++o/r.length*(n-e);return t(c),l}),u);function i(u){A(u!=null&&Array.isArray(u)&&u.length>0,()=>"promises must be a none empty array")}function a(u,l){A(u>=0&&u<=1,()=>`Progress fraction must be in range [0, 1], but got startFraction ${u}`),A(l>=0&&l<=1,()=>`Progress fraction must be in range [0, 1], but got endFraction ${l}`),A(l>=u,()=>`startFraction must be no more than endFraction, but got startFraction ${u} and endFraction ${l}`)}return Promise.all(r.map(s))}async function l0(r,t){t==null&&(t={});let e=t.fetchFunc==null?V().platform.fetch:t.fetchFunc,n=r.map(p=>e(p,t.requestInit,{isBinary:!0})),o=0,s=.5,a=(t.onProgress==null?await Promise.all(n):await a0(n,t.onProgress,o,s)).map(p=>p.arrayBuffer()),u=.5,l=1;return t.onProgress==null?await Promise.all(a):await a0(a,t.onProgress,u,l)}async function Q1(r,t="",e,n){return u0(i=>l0(i,{requestInit:n}))(r,t,e)}function u0(r){return async(t,e="",n)=>{let o=t.map(()=>!1),s={},i=n!=null?n.map(()=>!1):[],a=[];if(t.forEach((f,d)=>{let h=0;f.weights.forEach(g=>{let y="quantization"in g?g.quantization.dtype:g.dtype,b=_f[y]*re(g.shape),w=()=>{o[d]=!0,s[d]==null&&(s[d]=[]),s[d].push({manifestEntry:g,groupOffset:h,sizeBytes:b})};n!=null?n.forEach((v,k)=>{v===g.name&&(w(),i[k]=!0)}):w(),a.push(g.name),h+=b})}),!i.every(f=>f)){let f=n.filter((d,h)=>!i[h]);throw new Error(`Could not find weights in manifest with names: ${f.join(", ")}.
Manifest JSON has weights with names: ${a.join(", ")}.`)}let u=o.reduce((f,d,h)=>(d&&f.push(h),f),[]),l=[];u.forEach(f=>{t[f].paths.forEach(d=>{let h=e+(e.endsWith("/")?"":"/")+d;l.push(h)})});let c=await r(l),p={},m=0;return u.forEach(f=>{let d=t[f].paths.length,h=0;for(let v=0;v<d;v++)h+=c[m+v].byteLength;let g=new ArrayBuffer(h),y=new Uint8Array(g),b=0;for(let v=0;v<d;v++){let k=new Uint8Array(c[m+v]);y.set(k,b),b+=k.byteLength}s[f].forEach(v=>{let k=g.slice(v.groupOffset,v.groupOffset+v.sizeBytes),E=ax(k,[v.manifestEntry]);for(let D in E)p[D]=E[D]}),m+=d}),p}}var E4="application/octet-stream",A4="application/json",px=class{constructor(t,e){if(this.DEFAULT_METHOD="POST",e==null&&(e={}),this.weightPathPrefix=e.weightPathPrefix,this.onProgress=e.onProgress,this.weightUrlConverter=e.weightUrlConverter,e.fetchFunc!=null?(A(typeof e.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=e.fetchFunc):this.fetch=V().platform.fetch,A(t!=null&&t.length>0,()=>"URL path for http must not be null, undefined or empty."),Array.isArray(t)&&A(t.length===2,()=>`URL paths for http must have a length of 2, (actual length is ${t.length}).`),this.path=t,e.requestInit!=null&&e.requestInit.body!=null)throw new Error("requestInit is expected to have no pre-existing body, but has one.");this.requestInit=e.requestInit||{}}async save(t){if(t.modelTopology instanceof ArrayBuffer)throw new Error("BrowserHTTPRequest.save() does not support saving model topology in binary formats yet.");let e=Object.assign({method:this.DEFAULT_METHOD},this.requestInit);e.body=new FormData;let n=[{paths:["./model.weights.bin"],weights:t.weightSpecs}],o=lx(t,n);e.body.append("model.json",new Blob([JSON.stringify(o)],{type:A4}),"model.json"),t.weightData!=null&&e.body.append("model.weights.bin",new Blob([t.weightData],{type:E4}),"model.weights.bin");let s=await this.fetch(this.path,e);if(s.ok)return{modelArtifactsInfo:Fi(t),responses:[s]};throw new Error(`BrowserHTTPRequest.save() failed due to HTTP response status ${s.status}.`)}async load(){let t=await this.fetch(this.path,this.requestInit);if(!t.ok)throw new Error(`Request to ${this.path} failed with status code ${t.status}. Please verify this URL points to the model JSON of the model to load.`);let e;try{e=await t.json()}catch(s){let i=`Failed to parse model JSON of response from ${this.path}.`;throw this.path.endsWith(".pb")?i+=" 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.":i+=" Please make sure the server is serving valid JSON for this request.",new Error(i)}let n=e.modelTopology,o=e.weightsManifest;if(n==null&&o==null)throw new Error(`The JSON from HTTP path ${this.path} contains neither model topology or manifest for weights.`);return qp(e,s=>this.loadWeights(s))}async loadWeights(t){let e=Array.isArray(this.path)?this.path[1]:this.path,[n,o]=D4(e),s=this.weightPathPrefix||n,i=[];for(let c of t)i.push(...c.weights);let a=[],u=[];for(let c of t)for(let p of c.paths)this.weightUrlConverter!=null?u.push(this.weightUrlConverter(p)):a.push(s+p+o);this.weightUrlConverter&&a.push(...await Promise.all(u));let l=await l0(a,{requestInit:this.requestInit,fetchFunc:this.fetch,onProgress:this.onProgress});return[i,Hp(l)]}};px.URL_SCHEME_REGEX=/^https?:\/\//;function D4(r){let t=r.lastIndexOf("/"),e=r.lastIndexOf("?"),n=r.substring(0,t),o=e>t?r.substring(e):"";return[n+"/",o]}function cx(r){return r.match(px.URL_SCHEME_REGEX)!=null}var t_=(r,t)=>{if(typeof fetch=="undefined"&&(t==null||t.fetchFunc==null))return null;{let e=!0;if(Array.isArray(r)?e=r.every(n=>cx(n)):e=cx(r),e)return mx(r,t)}return null};De.registerSaveRouter(t_);De.registerLoadRouter(t_);function mx(r,t){return new px(r,t)}function e_(r,t){return mx(r,t)}var fx=class{constructor(t){this.modelArtifacts=t}async load(){return this.modelArtifacts}},r_=class{constructor(t){this.saveHandler=t}async save(t){return this.saveHandler(t)}};function n_(r,t,e,n){return arguments.length===1?r.modelTopology!=null||r.weightSpecs!=null?new fx(r):(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 fx({modelTopology:r})):(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 fx({modelTopology:r,weightSpecs:t,weightData:e,trainingConfig:n}))}function o_(r){return new r_(r)}var i_={};Yt(i_,{confusionMatrix:()=>s_});function $4(r,t,e=!1,n=!1){let o=C(r,"a","matMul"),s=C(t,"b","matMul");[o,s]=jt(o,s);let i={a:o,b:s},a={transposeA:e,transposeB:n};return _.runKernel(Zo,i,a)}var Mt=N({matMul_:$4});function R4(r,t,e=1,n=0){if(t<2)throw new Error(`Error in oneHot: depth must be >=2, but it is ${t}`);let s={indices:C(r,"indices","oneHot","int32")},i={depth:t,onValue:e,offValue:n};return _.runKernel(Cs,s,i)}var Us=N({oneHot_:R4});function F4(r,t){let e=C(r,"x","transpose");if(t==null&&(t=e.shape.map((s,i)=>i).reverse()),A(e.rank===t.length,()=>`Error in transpose: rank of input ${e.rank} must match length of perm ${t}.`),t.forEach(s=>{A(s>=0&&s<e.rank,()=>`All entries in 'perm' must be between 0 and ${e.rank-1} but got ${t}`)}),e.rank<=1)return e.clone();let n={x:e},o={perm:t};return _.runKernel(Vs,n,o)}var Gt=N({transpose_:F4});function O4(r,t,e){let n=C(r,"labels","confusionMatrix"),o=C(t,"predictions","confusionMatrix");A(e==null||e>0&&Number.isInteger(e),()=>`If provided, numClasses must be a positive integer, but got ${e}`),A(n.rank===1,()=>`Expected the rank of labels to be 1, but got ${n.rank}`),A(o.rank===1,()=>`Expected the rank of predictions to be 1, but got ${o.rank}`),A(n.shape[0]===o.shape[0],()=>`Mismatch in the number of examples: ${n.shape[0]} vs. ${o.shape[0]}. Labels and predictions should have the same number of elements.`),A(e>0&&Number.isInteger(e),()=>`numClasses is required to be a positive integer, but got ${e}`);let s=Us(Z(n,"int32"),e),i=Us(Z(o,"int32"),e),a=Gt(s),u=Mt(a,i);return Z(u,"int32")}var s_=N({confusionMatrix_:O4});var Mr={};Yt(Mr,{assertAndGetBroadcastShape:()=>Lt,getBroadcastDims:()=>a_,getReductionAxes:()=>ge});function a_(r,t){let e=r.length,n=[];for(let o=0;o<e;o++){let s=e-1-o,i=r[s]||1;(t[t.length-1-o]||1)>1&&i===1&&n.unshift(s)}return n}function ge(r,t){let e=[];for(let n=0;n<t.length;n++){let o=r[r.length-n-1],s=t.length-n-1,i=t[s];(o==null||o===1&&i>1)&&e.unshift(s)}return e}function Lt(r,t){let e=[],n=Math.max(r.length,t.length);for(let o=0;o<n;o++){let s=r[r.length-o-1];s==null&&(s=1);let i=t[t.length-o-1];if(i==null&&(i=1),s===1)e.unshift(i);else if(i===1)e.unshift(s);else if(s!==i){let a=`Operands could not be broadcast together with shapes ${r} and ${t}.`;throw Error(a)}else e.unshift(s)}return e}var dx={};Yt(dx,{fromPixels:()=>G4,fromPixelsAsync:()=>B4,toPixels:()=>V4});function c0(r,t,e){if(qn(r),t!=null&&t.length!==3)throw new Error("tensor3d() requires shape to have three numbers");let n=Or(r,e);if(n.length!==3&&n.length!==1)throw new Error("tensor3d() requires values to be number[][][] or flat/TypedArray");if(n.length===1&&t==null)throw new Error("tensor3d() requires shape to be provided when `values` are a flat array");return sn(r,t,n,e)}var Du;function l_(r,t=3){if(t>4)throw new Error("Cannot construct Tensor with more than 4 channels from pixels.");if(r==null)throw new Error("pixels passed to tf.browser.fromPixels() can not be null");let e=!1,n=!1,o=!1,s=!1,i=!1,a=!1;if(r.data instanceof Uint8Array)e=!0;else if(typeof ImageData!="undefined"&&r instanceof ImageData)n=!0;else if(typeof HTMLVideoElement!="undefined"&&r instanceof HTMLVideoElement)o=!0;else if(typeof HTMLImageElement!="undefined"&&r instanceof HTMLImageElement)s=!0;else if(r.getContext!=null)i=!0;else if(typeof ImageBitmap!="undefined"&&r instanceof ImageBitmap)a=!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 ${r.constructor.name}`);if(o&&o&&r.readyState<2)throw new Error("The video element has not loaded data yet. Please wait for `loadeddata` event on the <video> element.");if(Cf(wf,_.backendName)!=null){let d={pixels:r},h={numChannels:t};return _.runKernel(wf,d,h)}let[l,c]=o?[r.videoWidth,r.videoHeight]:[r.width,r.height],p;if(i)p=r.getContext("2d").getImageData(0,0,l,c).data;else if(n||e)p=r.data;else if(s||o||a){if(Du==null)if(typeof document=="undefined")if(typeof OffscreenCanvas!="undefined"&&typeof OffscreenCanvasRenderingContext2D!="undefined")Du=new OffscreenCanvas(1,1).getContext("2d");else throw new Error("Cannot parse input in current context. Reason: OffscreenCanvas Context2D rendering is not supported.");else Du=document.createElement("canvas").getContext("2d");Du.canvas.width=l,Du.canvas.height=c,Du.drawImage(r,0,0,l,c),p=Du.getImageData(0,0,l,c).data}let m;if(t===4)m=new Int32Array(p);else{let d=l*c;m=new Int32Array(d*t);for(let h=0;h<d;h++)for(let g=0;g<t;++g)m[h*t+g]=p[h*4+g]}return c0(m,[c,l,t],"int32")}function P4(r){return r!=null&&r.data instanceof Uint8Array}function M4(){return typeof window!="undefined"&&typeof ImageBitmap!="undefined"&&window.hasOwnProperty("createImageBitmap")}function L4(r){return r!=null&&r.width!==0&&r.height!==0}function z4(r){return M4()&&!(r instanceof ImageBitmap)&&L4(r)&&!P4(r)}async function B4(r,t=3){let e=null;if(V().getBool("WRAP_TO_IMAGEBITMAP")&&z4(r)){let n;try{n=await createImageBitmap(r,{premultiplyAlpha:"none"})}catch(o){n=null}n!=null&&n.width===r.width&&n.height===r.height?e=n:e=r}else e=r;return l_(e,t)}async function V4(r,t){let e=C(r,"img","toPixels");if(!(r instanceof Vt)){let l=e;e=Z(l,"int32"),l.dispose()}if(e.rank!==2&&e.rank!==3)throw new Error(`toPixels only supports rank 2 or 3 tensors, got rank ${e.rank}.`);let[n,o]=e.shape.slice(0,2),s=e.rank===2?1:e.shape[2];if(s>4||s===2)throw new Error(`toPixels only supports depth of size 1, 3 or 4 but got ${s}`);if(e.dtype!=="float32"&&e.dtype!=="int32")throw new Error(`Unsupported type for toPixels: ${e.dtype}. Please use float32 or int32 tensors.`);let i=await e.data(),a=e.dtype==="float32"?255:1,u=new Uint8ClampedArray(o*n*4);for(let l=0;l<n*o;++l){let c=[0,0,0,255];for(let m=0;m<s;m++){let f=i[l*s+m];if(e.dtype==="float32"){if(f<0||f>1)throw new Error(`Tensor values for a float32 Tensor must be in the range [0 - 1] but encountered ${f}.`)}else if(e.dtype==="int32"&&(f<0||f>255))throw new Error(`Tensor values for a int32 Tensor must be in the range [0 - 255] but encountered ${f}.`);s===1?(c[0]=f*a,c[1]=f*a,c[2]=f*a):c[m]=f*a}let p=l*4;u[p+0]=Math.round(c[0]),u[p+1]=Math.round(c[1]),u[p+2]=Math.round(c[2]),u[p+3]=Math.round(c[3])}if(t!=null){t.width=o,t.height=n;let l=t.getContext("2d"),c=new ImageData(u,o,n);l.putImageData(c,0,0)}return e!==r&&e.dispose(),u}var G4=N({fromPixels_:l_});var hx={};Yt(hx,{prepareAndValidate:()=>u_});function u_(r,t){let e=r.shape.length,n=t.shape.length;if(e<1)throw new Error(`tf.gatherND() expects the input to be rank 1 or higher, but the rank was ${e}.`);if(n<1)throw new Error(`tf.gatherND() expects the indices to be rank 1 or higher, but the rank was ${n}.`);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[n-1]>e)throw new Error(`index innermost dimension length must be <= tensor rank; saw: ${t.shape[n-1]} vs. ${e}`);if(re(r.shape)===0)throw new Error(`Requested more than 0 entries, but input is empty. Input shape: ${r.shape}.`);let o=t.shape,s=o[o.length-1],i=1;for(let p=0;p<o.length-1;++p)i*=o[p];let a=r.shape,u=o.slice();u.pop();let l=1;for(let p=s;p<e;++p)l*=a[p],u.push(a[p]);let c=[...di(r.shape).map(p=>p/l),1].slice(0,s);return[u,i,l,c]}var Ef={};Yt(Ef,{calculateShapes:()=>c_,validateInput:()=>gx,validateUpdateShape:()=>p0});function p0(r,t,e){let n=t.rank>1?t.shape[t.rank-1]:1,o=t.rank>1?t.rank-1:1,s=`Must have updates.shape = indices.shape[:batchDim] + shape[sliceDim:], got updates.shape: ${e.shape}, indices.shape: ${t.shape}, shape: ${r}, sliceDim: ${n}, and batchDim: ${o}.`;if(e.rank<o)throw new Error(s+` update.rank < ${o}. `);if(r.length<n+(e.rank-o))throw new Error(s+` Output shape length < ${n+(e.rank-o)}`);if(e.rank!==o+r.length-n)throw new Error(s+` update.rank != ${o+r.length-n}`);for(let i=0;i<o;++i)if(e.shape[i]!==t.shape[i])throw new Error(s+` updates.shape[${i}] (${e.shape[i]}) != indices.shape[${i}] (${t.shape[i]}).`);for(let i=0;i<e.rank-o;++i)if(e.shape[i+o]!==r[i+n])throw new Error(s+` updates.shape[${i+o}] (${e.shape[i+o]}) != shape[${i+o}] (${r[i+o]})`)}function gx(r,t,e){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(r.rank<1)throw new Error(`tf.scatterND() expects the updates to be rank 1 or higher, but the rank was ${r.rank}.`);if(t.dtype!=="int32")throw new Error(`The dtype of 'indices' should be int32, but got dtype: ${t.dtype}`);if(e.length<1)throw new Error(`Output rank must be greater or equal to 1, but got shape: ${e}`);if(e.length===0){if(t.size===0)throw new Error(`Indices specified for empty output. indices shape: ${t.shape}`);if(r.size===0)throw new Error(`Updates specified for empty output. updates shape: ${r.shape}`)}p0(e,t,r)}function c_(r,t,e){let n=t.shape.length,o=n>1?t.shape[n-1]:1,s=e.length,i=1;for(let p=o;p<s;++p)i*=e[p];let a=o<1?1:o,u=re(t.shape)/a,l=[...di(e.slice(0,o)),1],c=re(e);return{sliceRank:o,numUpdates:u,sliceSize:i,strides:l,outputSize:c}}var Ve={};Yt(Ve,{assertParamsValid:()=>U4,computeFlatOffset:()=>X4,computeOutShape:()=>q4,getNormalizedAxes:()=>K4,isSliceContinous:()=>j4,maskToAxes:()=>H4,parseSliceParams:()=>f0,sliceInfo:()=>Y4,startForAxis:()=>y_,startIndicesWithElidedDims:()=>h_,stopForAxis:()=>b_,stopIndicesWithElidedDims:()=>g_,stridesForAxis:()=>x_,stridesWithElidedDims:()=>m_});var m0=-2,W4=-1;function U4(r,t,e){let n=r.shape.length;A(n===t.length,()=>`Error in slice${n}D: Length of begin ${t} must match the rank of the array (${n}).`),A(n===e.length,()=>`Error in slice${n}D: Length of size ${e} must match the rank of the array (${n}).`);for(let o=0;o<n;++o)A(t[o]+e[o]<=r.shape[o],()=>`Error in slice${n}D: begin[${o}] + size[${o}] (${t[o]+e[o]}) would overflow input.shape[${o}] (${r.shape[o]})`)}function H4(r){let t=[],e=0;for(;r>0;)r&1&&t.push(e),r/=2,e++;return t}function q4(r,t,e){let n=[];for(let o=0;o<r.length;o++)n[o]=Math.ceil((t[o]-r[o])/e[o]);return n}function m_(r,t,e,n){let o=[...r];for(let s=o.length;s<n.length;s++)o.push(1);for(let s=0;s<e;s++)s===0?o[t]=1:(o.splice(t,0,1),o.pop());return o}function f_(r,t,e){return e<=r?e:e-(t-1)}function d_(r,t){let e=[];for(let n=0;n<r;n++)e.push(t+n);return e}function K4(r,t,e,n,o,s,i,a,u){let l=r.length,c=new Array(l),p=new Array(l),m=new Array(l);if(t.length&&e>0){let f=t[0],d=e+1;c=h_(i,f,d,n,r),p=g_(a,f,d,o,r),m=m_(s,f,d,r)}else for(let f=0;f<l;f++)c[f]=y_(i,n,s,r,f,u),p[f]=b_(a,o,s,r,f,u),m[f]=x_(s,f,u);return{begin:c,end:p,strides:m}}function h_(r,t,e,n,o){let s=[...o],i=d_(e,t);for(let a=0;a<s.length;a++)if(i.indexOf(a)>-1)s[a]=0;else{let u=f_(t,e,a),l=n[u];r&1<<u&&(l=0),s[a]=l}return s}function g_(r,t,e,n,o){let s=[...o],i=d_(e,t);for(let a=0;a<s.length;a++)if(i.indexOf(a)>-1)s[a]=Number.MAX_SAFE_INTEGER;else{let u=f_(t,e,a),l=n[u];r&1<<u&&(l=Number.MAX_SAFE_INTEGER),s[a]=l}for(let a=0;a<s.length;a++){let u=o[a];s[a]<0&&(s[a]+=u),s[a]=op(0,s[a],o[a])}return s}function x_(r,t,e){let n=r[t];return(e&1<<t||n==null)&&(n=1),n}function y_(r,t,e,n,o,s){let i=t[o],a=e[o]||1;(r&1<<o||s&1<<o||i==null)&&(a>0?i=Number.MIN_SAFE_INTEGER:i=Number.MAX_SAFE_INTEGER);let u=n[o];return i<0&&(i+=u),i=op(0,i,u-1),i}function b_(r,t,e,n,o,s){let i=t[o],a=e[o]||1;(r&1<<o||s&1<<o||i==null)&&(a>0?i=Number.MAX_SAFE_INTEGER:i=Number.MIN_SAFE_INTEGER);let u=n[o];return i<0&&(i+=u),a>0?i=op(0,i,u):i=op(-1,i,u-1),i}function j4(r,t,e){let n=e.length;for(let o=0;o<e.length;o++)if(e[o]>1){n=o;break}for(let o=n+1;o<e.length;o++)if(t[o]>0||e[o]!==r[o])return!1;return!0}function X4(r,t){let e=r.length>0?r[r.length-1]:1;for(let n=0;n<r.length-1;n++)e+=r[n]*t[n];return e}function f0(r,t,e){let n,o=r.shape.length;typeof t=="number"?n=[t,...new Array(o-1).fill(0)]:t.length<o?n=t.concat(new Array(o-t.length).fill(0)):n=t.slice(),n.forEach(i=>{A(i!==-1,()=>"slice() does not support negative begin indexing.")});let s;return e==null?s=new Array(o).fill(-1):typeof e=="number"?s=[e,...new Array(o-1).fill(-1)]:e.length<o?s=e.concat(new Array(o-e.length).fill(-1)):s=e,s=s.map((i,a)=>i>=0?i:(A(i===-1,()=>`Negative size values should be exactly -1 but got ${i} for the slice() size at index ${a}.`),r.shape[a]-n[a])),[n,s]}function Y4(r,t,e,n,o,s,i,a,u){let l;if(n==null?(l=new Array(t.length),l.fill(1)):l=n,i!=null&&(i&i-1)!==0)throw new Error("Multiple ellipses in slice is not allowed.");let c=!1,p={dims:l.length,numAddAxisAfterEllipsis:0,begin:t.slice(),end:e.slice(),strides:l.slice(),beginMask:o,endMask:s,ellipsisMask:i,newAxisMask:a,shrinkAxisMask:u};for(let w=0;w<p.dims;w++)c&&(1<<w&a)!==0&&p.numAddAxisAfterEllipsis++,1<<w&i&&(c=!0);c||(p.ellipsisMask|=1<<p.dims,p.dims++);let m={dims:r.length,beginMask:0,endMask:0,beginValid:!1,endValid:!1};Z4(p,m);let f=!0,d=!0,h=!0,g=[],y=[];for(let w=0;w<r.length;++w){if(m.strides[w]===0)throw Error(`strides[${w}] must be non-zero`);let v=!!(m.shrinkAxisMask&1<<w),k=r[w];if(k===-1){g.push(v?1:-1);continue}let E=[m.beginMask&1<<w,m.endMask&1<<w],D=[m.strides[w]>0?0:-1,m.strides[w]>0?k:k-1];if(v&&m.strides[w]<=0)throw Error("only stride 1 allowed on non-range indexing.");h=h&&m.strides[w]===1;let $=!!(m.beginMask&1<<w&&m.endMask&1<<w);if(m.beginValid&&m.endValid){if(v){let U=m.begin[w]<0?k+m.begin[w]:m.begin[w];if(m.begin[w]=U,m.end[w]=m.begin[w]+1,U<0||U>=k)throw Error(`slice index ${m.begin[w]} of dimension ${w} out of bounds.`)}else m.begin[w]=p_(m.begin[w],0,m.strides[w],k,E,D),m.end[w]=p_(m.end[w],1,m.strides[w],k,E,D);let W=m.strides[w]===1&&m.begin[w]===0&&m.end[w]===k;f=f&&W,d=d&&(w===0&&m.strides[w]===1||W)}else f=f&&m.strides[w]===1&&$,d=d&&(w===0&&m.strides[w]===1||$);let R,M=!1;if(m.beginValid&&m.endValid?(R=m.end[w]-m.begin[w],M=!0):v?(R=1,M=!0):$&&k>=0&&(m.strides[w]<0?R=-k:R=k,M=!0),M){let W;R===0||R<0!=m.strides[w]<0?W=0:W=Math.trunc(R/m.strides[w])+(R%m.strides[w]!==0?1:0),g.push(W)}else g.push(-1)}for(let w=0;w<m.finalShapeGatherIndices.length;++w){let v=m.finalShapeGatherIndices[w];v>=0?y.push(g[v]):v===m0&&y.push(1)}return{finalShapeSparse:y.filter((w,v)=>m.finalShapeGatherIndices[v]!==m0),finalShape:y,isIdentity:f,sliceDim0:d,isSimpleSlice:h,begin:m.begin,end:m.end,strides:m.strides}}function Z4(r,t){t.beginMask=0,t.endMask=0,t.shrinkAxisMask=0;let e=0;t.beginValid=r.begin!=null,t.endValid=r.end!=null,t.begin=new Array(t.dims),t.end=new Array(t.dims),t.strides=new Array(t.dims),t.finalShapeGatherIndices=[],t.finalShapeGatherIndicesSparse=[],t.inputShapeGatherIndicesSparse=new Array(t.dims);for(let n=0;n<r.dims;n++)if(1<<n&r.ellipsisMask){let o=Math.min(t.dims-(r.dims-n)+1+r.numAddAxisAfterEllipsis,t.dims);for(;e<o;e++)t.begin[e]=0,t.end[e]=0,t.strides[e]=1,t.beginMask|=1<<e,t.endMask|=1<<e,t.finalShapeGatherIndices.push(e),t.finalShapeGatherIndicesSparse.push(-1),t.inputShapeGatherIndicesSparse[e]=n}else if(1<<n&r.newAxisMask)t.finalShapeGatherIndices.push(m0),t.finalShapeGatherIndicesSparse.push(-1);else{if(e===t.begin.length)throw Error(`Index out of range using input dim ${e}; input has only ${t.dims} dims, ${t.begin.length}.`);r.begin!=null&&(t.begin[e]=r.begin[n]),r.end!=null&&(t.end[e]=r.end[n]),t.strides[e]=r.strides[n],r.beginMask&1<<n&&(t.beginMask|=1<<e),r.endMask&1<<n&&(t.endMask|=1<<e),r.shrinkAxisMask&1<<n?(t.finalShapeGatherIndices.push(W4),t.finalShapeGatherIndicesSparse.push(-1),t.shrinkAxisMask|=1<<e):(t.finalShapeGatherIndices.push(e),t.finalShapeGatherIndicesSparse.push(n)),t.inputShapeGatherIndicesSparse[e]=n,e++}}function p_(r,t,e,n,o,s){if(o[t])return e>0?s[t]:s[t+1&1];{let i=r<0?n+r:r;return i<s[0]?s[0]:i>s[1]?s[1]:i}}var et={};Yt(et,{Serializable:()=>xx,SerializationMap:()=>Ba,registerClass:()=>Sn});var xx=class{getClassName(){return this.constructor.className}static fromConfig(t,e){return new t(e)}},Ba=class{constructor(){this.classNameMap={}}static getMap(){return Ba.instance==null&&(Ba.instance=new Ba),Ba.instance}static register(t){Ba.getMap().classNameMap[t.className]=[t,t.fromConfig]}};function Sn(r){A(r.className!=null,()=>"Class being registered does not have the static className property defined."),A(typeof r.className=="string",()=>"className is required to be a string, but got type "+typeof r.className),A(r.className.length>0,()=>"Class being registered has an empty-string as its className, which is disallowed."),Ba.register(r)}var I_={};Yt(I_,{TEST_EPSILON_FLOAT16:()=>w_,encodeStrings:()=>C_,expectArrayBuffersEqual:()=>nH,expectArraysClose:()=>Q4,expectArraysEqual:()=>eH,expectNumbersClose:()=>v_,expectPromiseToFail:()=>tH,expectValuesInRange:()=>rH,testEpsilon:()=>yx});var J4=.001,w_=.1;function Q4(r,t,e){return e==null&&(e=yx()),d0(r,t,(n,o)=>h0(n,o,e))}function yx(){return _.backend.floatPrecision()===32?J4:w_}function d0(r,t,e){let n=!0;if((yr(r)||yr(t))&&(n=!1),yr(r)&&yr(t)&&(n=!0),n){let i=r.constructor.name,a=t.constructor.name;if(i!==a)throw new Error(`Arrays are of different type. Actual: ${i}. Expected: ${a}`)}if(Array.isArray(r)&&Array.isArray(t)){let i=Or(r),a=Or(t);if(!Fn(i,a))throw new Error(`Arrays have different shapes. Actual: [${i}]. Expected: [${a}]`)}let o=yr(r)?r:qo(r),s=yr(t)?t:qo(t);if(o.length!==s.length)throw new Error(`Arrays have different lengths actual: ${o.length} vs expected: ${s.length}.
Actual: ${o}.
Expected: ${s}.`);for(let i=0;i<s.length;++i){let a=o[i],u=s[i];if(!e(a,u))throw new Error(`Arrays differ: actual[${i}] = ${a}, expected[${i}] = ${u}.
Actual: ${o}.
Expected: ${s}.`)}}function tH(r,t){r().then(()=>t.fail(),()=>t())}function eH(r,t){let e=typeof t=="string"||typeof t=="number"||typeof t=="boolean"?[t]:t;return Ko(r)||Ko(r[0])||Ko(t)||Ko(t[0])?d0(r,e,(n,o)=>n==o):d0(r,t,(n,o)=>h0(n,o,0))}function v_(r,t,e){if(e==null&&(e=yx()),!h0(r,t,e))throw new Error(`Numbers differ: actual === ${r}, expected === ${t}`)}function h0(r,t,e){return!isFinite(r)&&!isFinite(t)?!0:!(isNaN(r)||isNaN(t)||Math.abs(r-t)>e)}function rH(r,t,e){for(let n=0;n<r.length;n++)if(r[n]<t||r[n]>e)throw new Error(`Value out of range:${r[n]} low: ${t}, high: ${e}`)}function nH(r,t){let e=new Float32Array(r),n=new Float32Array(t);if(e.length!==n.length)throw new Error(`Expected ArrayBuffer to be of length ${n.length}, but it was ${e.length}`);for(let o=0;o<n.length;o++)if(e[o]!==n[o])throw new Error(`Expected ArrayBuffer value at ${o} to be ${n[o]} but got ${e[o]} instead`)}function C_(r){for(let t=0;t<r.length;t++){let e=r[t];Array.isArray(e)?C_(e):r[t]=Ml(e)}return r}var S_="3.15.0";function Yct(){V().set("PROD",!0)}function Zct(){V().set("DEBUG",!0)}function Jct(){V().set("DEPRECATION_WARNINGS_ENABLED",!1),console.warn("TensorFlow.js deprecation warnings have been disabled.")}function g0(r){V().getBool("DEPRECATION_WARNINGS_ENABLED")&&console.warn(r+" You can disable deprecation warnings with tf.disableDeprecationWarnings().")}C1(g0);function Qct(){_.disposeVariables()}function fo(){return _}function Af(){return _.memory()}function tpt(r){return _.profile(r)}function B(r,t){return _.tidy(r,t)}function _t(r){Tf(r).forEach(e=>e.dispose())}function Me(r){return _.keep(r)}function ept(r){return _.time(r)}function oH(r){return _.setBackend(r)}function rpt(){return _.ready()}function npt(){return _.backendName}function opt(r){_.removeBackend(r)}function spt(r){return _.findBackend(r)}function ipt(r){return _.findBackendFactory(r)}function Xp(r,t,e=1){return _.registerBackend(r,t,e)}function k_(){return _.backend}function apt(r,t){V().setPlatform(r,t)}function sH(r,t){let e=C(r,"a","add"),n=C(t,"b","add");[e,n]=jt(e,n);let o={a:e,b:n};return _.runKernel(Kn,o)}var J=N({add_:sH});function iH(r,t){let e=C(r,"a","floorDiv"),n=C(t,"b","floorDiv");[e,n]=jt(e,n);let o={a:e,b:n};return _.runKernel(us,o)}var $u=N({floorDiv_:iH});function aH(r,t){let e=C(r,"a","div"),n=C(t,"b","div");if([e,n]=jt(e,n),e.dtype==="int32"&&n.dtype==="int32")return $u(e,n);let o={a:e,b:n},s={};return _.runKernel(ss,o,s)}var ct=N({div_:aH});function lH(r,t){let e=C(r,"a","mul"),n=C(t,"b","mul");[e,n]=jt(e,n);let o={a:e,b:n};return _.runKernel(vs,o)}var P=N({mul_:lH});function uH(r){let t=C(r,"x","abs");if(t.dtype==="complex64"){let e={x:t};return _.runKernel(Il,e)}else{let e={x:t};return _.runKernel(hi,e)}}var Se=N({abs_:uH});function cH(r){let e={x:C(r,"x","acos")};return _.runKernel(ji,e)}var Df=N({acos_:cH});function pH(r){let e={x:C(r,"x","acosh")};return _.runKernel(Xi,e)}var $f=N({acosh_:pH});function mH(r){A(Array.isArray(r),()=>"The argument passed to tf.addN() must be a list of tensors"),A(r.length>=1,()=>`Must pass at least one tensor to tf.addN(), but got ${r.length}`);let t=r.map((o,s)=>C(o,`tensors${s}`,"addN")),e=t[0];t.forEach(o=>{if(o.dtype!==e.dtype)throw new Error("All tensors passed to tf.addN() must have the same dtype")}),t.forEach(o=>{if(!Fn(o.shape,e.shape))throw new Error("All tensors passed to tf.addN() must have the same shape")});let n=t;return _.runKernel(jo,n)}var x0=N({addN_:mH});function fH(r,t=null,e=!1){let o={x:C(r,"x","all","bool")},s={axis:t,keepDims:e};return _.runKernel(Yi,o,s)}var Ru=N({all_:fH});function dH(r,t=null,e=!1){let o={x:C(r,"x","any","bool")},s={axis:t,keepDims:e};return _.runKernel(Zi,o,s)}var Gl=N({any_:dH});function hH(r,t=0){let n={x:C(r,"x","argMax")},o={axis:t};return _.runKernel(Xo,n,o)}var Hs=N({argMax_:hH});function gH(r,t=0){let n={x:C(r,"x","argMin")},o={axis:t};return _.runKernel(vl,n,o)}var Rf=N({argMin_:gH});function xH(r){let e={x:C(r,"x","asin")};return _.runKernel(Ji,e)}var Ff=N({asin_:xH});function yH(r){let e={x:C(r,"x","asinh")};return _.runKernel(Qi,e)}var Of=N({asinh_:yH});function bH(r){let e={x:C(r,"x","atan")};return _.runKernel(ta,e)}var Pf=N({atan_:bH});function wH(r,t){let e=C(r,"a","atan2"),n=C(t,"b","atan2");[e,n]=jt(e,n);let o={a:e,b:n};return _.runKernel(ra,o)}var Mf=N({atan2_:wH});function vH(r){let e={x:C(r,"x","atanh")};return _.runKernel(ea,e)}var Lf=N({atanh_:vH});function CH(r,t,e,n,o="NHWC",s){let i=r[3],a=[...t,i],u=T_(o);return Ou(r,a,e,s,n,null,null,u)}function b0(r,t,e,n,o,s,i="channelsLast"){let[a,u]=bx(t),l;if(i==="channelsLast")l=[a,u,r[3],r[3]];else if(i==="channelsFirst")l=[a,u,r[1],r[1]];else throw new Error(`Unknown dataFormat ${i}`);return Ou(r,l,e,n,o,s,!1,i)}function IH(r,t,e,n,o,s,i="NDHWC"){let[a,u,l]=y0(t),c,p;if(i==="NDHWC")p="channelsLast",c=[a,u,l,r[4],r[4]];else if(i==="NCDHW")p="channelsFirst",c=[a,u,l,r[1],r[1]];else throw new Error(`Unknown dataFormat ${i}`);return N_(r,c,e,n,o,!1,p,s)}function Ou(r,t,e,n,o,s,i=!1,a="channelsLast"){let[u,l,c,p]=[-1,-1,-1,-1];if(a==="channelsLast")[u,l,c,p]=r;else if(a==="channelsFirst")[u,p,l,c]=r;else throw new Error(`Unknown dataFormat ${a}`);let[m,f,,d]=t,[h,g]=bx(e),[y,b]=bx(n),w=Yp(m,y),v=Yp(f,b),{padInfo:k,outHeight:E,outWidth:D}=NH(o,l,c,h,g,w,v,s,a),$=i?d*p:d,R;return a==="channelsFirst"?R=[u,$,E,D]:a==="channelsLast"&&(R=[u,E,D,$]),{batchSize:u,dataFormat:a,inHeight:l,inWidth:c,inChannels:p,outHeight:E,outWidth:D,outChannels:$,padInfo:k,strideHeight:h,strideWidth:g,filterHeight:m,filterWidth:f,effectiveFilterHeight:w,effectiveFilterWidth:v,dilationHeight:y,dilationWidth:b,inShape:r,outShape:R,filterShape:t}}function N_(r,t,e,n,o,s=!1,i="channelsLast",a){let[u,l,c,p,m]=[-1,-1,-1,-1,-1];if(i==="channelsLast")[u,l,c,p,m]=r;else if(i==="channelsFirst")[u,m,l,c,p]=r;else throw new Error(`Unknown dataFormat ${i}`);let[f,d,h,,g]=t,[y,b,w]=y0(e),[v,k,E]=y0(n),D=Yp(f,v),$=Yp(d,k),R=Yp(h,E),{padInfo:M,outDepth:W,outHeight:U,outWidth:q}=TH(o,l,c,p,y,b,w,D,$,R,a),K=s?g*m:g,j;return i==="channelsFirst"?j=[u,K,W,U,q]:i==="channelsLast"&&(j=[u,W,U,q,K]),{batchSize:u,dataFormat:i,inDepth:l,inHeight:c,inWidth:p,inChannels:m,outDepth:W,outHeight:U,outWidth:q,outChannels:K,padInfo:M,strideDepth:y,strideHeight:b,strideWidth:w,filterDepth:f,filterHeight:d,filterWidth:h,effectiveFilterDepth:D,effectiveFilterHeight:$,effectiveFilterWidth:R,dilationDepth:v,dilationHeight:k,dilationWidth:E,inShape:r,outShape:j,filterShape:t}}function SH(r,t,e,n,o){n==null&&(n=w0(r,t,e));let s=r[0],i=r[1],a=Fu((s-t+2*n)/e+1,o),u=Fu((i-t+2*n)/e+1,o);return[a,u]}function kH(r,t,e,n,o,s){o==null&&(o=w0(r,t,n));let i=r[0],a=r[1],u=r[2],l=Fu((i-t+2*o)/n+1,s),c=Fu((a-t+2*o)/n+1,s),p=Fu((u-t+2*o)/n+1,s);return[l,c,p,e]}function w0(r,t,e,n=1){let o=Yp(t,n);return Math.floor((r[0]*(e-1)-e+o)/2)}function bx(r){return typeof r=="number"?[r,r,r]:r.length===2?[r[0],r[1],1]:r}function y0(r){return typeof r=="number"?[r,r,r]:r}function Yp(r,t){return t<=1?r:r+(r-1)*(t-1)}function NH(r,t,e,n,o,s,i,a,u){let l,c,p;if(typeof r=="number"){l={top:r,bottom:r,left:r,right:r,type:r===0?"VALID":"NUMBER"};let f=SH([t,e],s,n,r,a);c=f[0],p=f[1]}else if(r==="same"){c=Math.ceil(t/n),p=Math.ceil(e/o);let m=Math.max(0,(c-1)*n+s-t),f=Math.max(0,(p-1)*o+i-e),d=Math.floor(m/2),h=m-d,g=Math.floor(f/2),y=f-g;l={top:d,bottom:h,left:g,right:y,type:"SAME"}}else if(r==="valid")l={top:0,bottom:0,left:0,right:0,type:"VALID"},c=Math.ceil((t-s+1)/n),p=Math.ceil((e-i+1)/o);else if(typeof r=="object"){let m=u==="channelsLast"?r[1][0]:r[2][0],f=u==="channelsLast"?r[1][1]:r[2][1],d=u==="channelsLast"?r[2][0]:r[3][0],h=u==="channelsLast"?r[2][1]:r[3][1];l={top:m,bottom:f,left:d,right:h,type:m===0&&f===0&&d===0&&h===0?"VALID":"EXPLICIT"},c=Fu((t-s+m+f)/n+1,a),p=Fu((e-i+d+h)/o+1,a)}else throw Error(`Unknown padding parameter: ${r}`);return{padInfo:l,outHeight:c,outWidth:p}}function TH(r,t,e,n,o,s,i,a,u,l,c){let p,m,f,d;if(typeof r=="number"){p={top:r,bottom:r,left:r,right:r,front:r,back:r,type:r===0?"VALID":"NUMBER"};let g=kH([t,e,n,1],a,1,o,r,c);m=g[0],f=g[1],d=g[2]}else if(r==="same"){m=Math.ceil(t/o),f=Math.ceil(e/s),d=Math.ceil(n/i);let h=(m-1)*o+a-t,g=(f-1)*s+u-e,y=(d-1)*i+l-n,b=Math.floor(h/2),w=h-b,v=Math.floor(g/2),k=g-v,E=Math.floor(y/2),D=y-E;p={top:v,bottom:k,left:E,right:D,front:b,back:w,type:"SAME"}}else if(r==="valid")p={top:0,bottom:0,left:0,right:0,front:0,back:0,type:"VALID"},m=Math.ceil((t-a+1)/o),f=Math.ceil((e-u+1)/s),d=Math.ceil((n-l+1)/i);else throw Error(`Unknown padding parameter: ${r}`);return{padInfo:p,outDepth:m,outHeight:f,outWidth:d}}function Fu(r,t){if(!t)return Math.trunc(r);switch(t){case"round":return Math.round(r);case"ceil":return Math.ceil(r);case"floor":return Math.floor(r);default:throw new Error(`Unknown roundingMode ${t}`)}}function Xn(r){let[t,e,n]=bx(r);return t===1&&e===1&&n===1}function Dr(r,t){return Xn(r)||Xn(t)}function T_(r){if(r==="NHWC")return"channelsLast";if(r==="NCHW")return"channelsFirst";throw new Error(`Unknown dataFormat ${r}`)}function ke(r,t,e){if(e!=null){if(typeof t=="string")throw Error(`Error in ${r}: pad must be an integer when using dimRoundingMode ${e} but got pad ${t}.`);if(typeof t=="number")A(Ki(t),()=>`Error in ${r}: pad must be an integer when using dimRoundingMode ${e} but got pad ${t}.`);else if(typeof t=="object")t.forEach(n=>{n.forEach(o=>{A(Ki(o),()=>`Error in ${r}: pad must be an integer when using dimRoundingMode ${e} but got pad ${o}.`)})});else throw Error(`Error in ${r}: Unknown padding parameter: ${t}`)}}function _H(r,t){let n={x:C(r,"x","reshape","string_or_numeric")},o={shape:t};return _.runKernel(Ii,n,o)}var O=N({reshape_:_H});function EH(r,t,e,n,o){let s=C(r,"x","avgPool","float32"),i=1;A(Dr(e,i),()=>`Error in avgPool: Either strides or dilations must be 1. Got strides ${e} and dilations '${i}'`);let a=s,u=!1;s.rank===3&&(u=!0,a=O(s,[1,s.shape[0],s.shape[1],s.shape[2]])),A(a.rank===4,()=>`Error in avgPool: x must be rank 4 but got rank ${a.rank}.`),ke("avgPool",n,o);let l={x:a},c={filterSize:t,strides:e,pad:n,dimRoundingMode:o},p=_.runKernel(Yo,l,c);return p=Z(p,s.dtype),u?O(p,[p.shape[1],p.shape[2],p.shape[3]]):p}var Va=N({avgPool_:EH});function AH(r,t,e,n,o,s="NDHWC"){let i=C(r,"x","avgPool3d","float32"),a=i,u=!1;i.rank===4&&(u=!0,a=O(i,[1,i.shape[0],i.shape[1],i.shape[2],i.shape[3]])),A(a.rank===5,()=>`Error in avgPool3d: x must be rank 5 but got rank ${a.rank}.`),A(s==="NDHWC",()=>`Error in avgPool3d: Only NDHWC is currently supported, but got dataFormat of ${s}`),ke("avgPool3d",n,o);let l={x:a},c={filterSize:t,strides:e,pad:n,dimRoundingMode:o,dataFormat:s},p=_.runKernel(Cl,l,c);return p=Z(p,a.dtype),u?O(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var zf=N({avgPool3d_:AH});function DH(r,t=0){A(r.length>=1,()=>"Pass at least one tensor to concat");let e=Pa(r,"tensors","concat","string_or_numeric");if(e[0].dtype==="complex64"&&e.forEach(s=>{if(s.dtype!=="complex64")throw new Error(`Cannot concatenate complex64 tensors with a tensor
with dtype ${s.dtype}. `)}),e.length===1)return In(e[0]);let n=e,o={axis:t};return _.runKernel(xi,n,o)}var se=N({concat_:DH});function $H(r){let e={x:C(r,"x","sigmoid","float32")};return _.runKernel(Rs,e)}var Lr=N({sigmoid_:$H});function RH(r,t,e){let n=C(r,"x","slice","string_or_numeric");if(n.rank===0)throw new Error("Slicing scalar is not possible");let o={x:n},s={begin:t,size:e};return _.runKernel(ki,o,s)}var Rt=N({slice_:RH});function FH(r){let e={x:C(r,"x","tanh","float32")};return _.runKernel(Bs,e)}var qs=N({tanh_:FH});function OH(r,t,e,n,o,s){let i=C(r,"forgetBias","basicLSTMCell"),a=C(t,"lstmKernel","basicLSTMCell"),u=C(e,"lstmBias","basicLSTMCell"),l=C(n,"data","basicLSTMCell"),c=C(o,"c","basicLSTMCell"),p=C(s,"h","basicLSTMCell"),m=se([l,p],1),f=Mt(m,a),d=J(f,u),h=d.shape[0],g=d.shape[1]/4,y=[h,g],b=Rt(d,[0,0],y),w=Rt(d,[0,g],y),v=Rt(d,[0,g*2],y),k=Rt(d,[0,g*3],y),E=J(P(Lr(b),qs(w)),P(c,Lr(J(i,v)))),D=P(qs(E),Lr(k));return[E,D]}var PH=N({basicLSTMCell_:OH});function MH(r,t,e){let n=C(r,"x","batchToSpaceND"),o=t.reduce((a,u)=>a*u);A(n.rank>=1+t.length,()=>`input rank is ${n.rank} but should be > than blockShape.length ${t.length}`),A(e.length===t.length,()=>`crops.length is ${e.length} but should be equal to blockShape.length ${t.length}`),A(n.shape[0]%o===0,()=>`input tensor batch is ${n.shape[0]} but is not divisible by the product of the elements of blockShape ${t.join(" * ")} === ${o}`);let s={x:n},i={blockShape:t,crops:e};return _.runKernel(gi,s,i)}var Ga=N({batchToSpaceND_:MH});function __(r){let t;return r.rank===0||r.rank===1?t=O(r,[1,1,1,r.size]):r.rank===2?t=O(r,[1,1,r.shape[0],r.shape[1]]):r.rank===3?t=O(r,[1,r.shape[0],r.shape[1],r.shape[2]]):t=r,t}function LH(r,t,e,n,o,s){s==null&&(s=.001);let i=C(r,"x","batchNorm"),a=C(t,"mean","batchNorm"),u=C(e,"variance","batchNorm"),l;o!=null&&(l=C(o,"scale","batchNorm"));let c;n!=null&&(c=C(n,"offset","batchNorm")),A(a.rank===u.rank,()=>"Batch normalization gradient requires mean and variance to have equal ranks."),A(c==null||a.rank===c.rank,()=>"Batch normalization gradient requires mean and offset to have equal ranks."),A(l==null||a.rank===l.rank,()=>"Batch normalization gradient requires mean and scale to have equal ranks.");let m={x:__(i),scale:l,offset:c,mean:a,variance:u},f={varianceEpsilon:s},d=_.runKernel(cs,m,f);return O(d,i.shape)}var ho=N({batchNorm_:LH});function zH(r,t,e,n,o,s){let i=C(r,"x","batchNorm"),a=C(t,"mean","batchNorm"),u=C(e,"variance","batchNorm"),l;o!=null&&(l=C(o,"scale","batchNorm"));let c;return n!=null&&(c=C(n,"offset","batchNorm")),A(i.rank===2,()=>`Error in batchNorm2D: x must be rank 2 but got rank ${i.rank}.`),A(a.rank===2||a.rank===1,()=>`Error in batchNorm2D: mean must be rank 2 or rank 1 but got rank ${a.rank}.`),A(u.rank===2||u.rank===1,()=>`Error in batchNorm2D: variance must be rank 2 or rank 1 but got rank ${u.rank}.`),l!=null&&A(l.rank===2||l.rank===1,()=>`Error in batchNorm2D: scale must be rank 2 or rank 1 but got rank ${l.rank}.`),c!=null&&A(c.rank===2||c.rank===1,()=>`Error in batchNorm2D: offset must be rank 2 or rank 1 but got rank ${c.rank}.`),ho(i,a,u,c,l,s)}var v0=N({batchNorm2d_:zH});function BH(r,t,e,n,o,s){let i=C(r,"x","batchNorm"),a=C(t,"mean","batchNorm"),u=C(e,"variance","batchNorm"),l;o!=null&&(l=C(o,"scale","batchNorm"));let c;return n!=null&&(c=C(n,"offset","batchNorm")),A(i.rank===3,()=>`Error in batchNorm3D: x must be rank 3 but got rank ${i.rank}.`),A(a.rank===3||a.rank===1,()=>`Error in batchNorm3D: mean must be rank 3 or rank 1 but got rank ${a.rank}.`),A(u.rank===3||u.rank===1,()=>`Error in batchNorm3D: variance must be rank 3 or rank 1 but got rank ${u.rank}.`),l!=null&&A(l.rank===3||l.rank===1,()=>`Error in batchNorm3D: scale must be rank 3 or rank 1 but got rank ${l.rank}.`),c!=null&&A(c.rank===3||c.rank===1,()=>`Error in batchNorm3D: offset must be rank 3 or rank 1 but got rank ${c.rank}.`),ho(i,a,u,c,l,s)}var C0=N({batchNorm3d_:BH});function VH(r,t,e,n,o,s){let i=C(r,"x","batchNorm"),a=C(t,"mean","batchNorm"),u=C(e,"variance","batchNorm"),l;o!=null&&(l=C(o,"scale","batchNorm"));let c;return n!=null&&(c=C(n,"offset","batchNorm")),A(i.rank===4,()=>`Error in batchNorm4D: x must be rank 4 but got rank ${i.rank}.`),A(a.rank===4||a.rank===1,()=>`Error in batchNorm4D: mean must be rank 4 or rank 1 but got rank ${a.rank}.`),A(u.rank===4||u.rank===1,()=>`Error in batchNorm4D: variance must be rank 4 or rank 1 but got rank ${u.rank}.`),l!=null&&A(l.rank===4||l.rank===1,()=>`Error in batchNorm4D: scale must be rank 4 or rank 1 but got rank ${l.rank}.`),c!=null&&A(c.rank===4||c.rank===1,()=>`Error in batchNorm4D: offset must be rank 4 or rank 1 but got rank ${c.rank}.`),ho(i,a,u,c,l,s)}var I0=N({batchNorm4d_:VH});function GH(r,t,e){let n=C(r,"x","bincount"),o=C(t,"weights","bincount");A(n.dtype==="int32",()=>`Error in bincount: input dtype must be int32, but got ${n.dtype}`),A(e>=0,()=>`size must be non-negative, but got ${e}.`),A(o.size===n.size||o.size===0,()=>`Error in bincount: weights must have the same size as input or0-length, but got input shape: ${n.shape}, weights shape: ${o.shape}.`);let s={x:n,weights:o},i={size:e};return _.runKernel(cp,s,i)}var Bf=N({bincount_:GH});function WH(r,t){let e=C(r,"s0","broadcastArgs","int32"),n=C(t,"s1","broadcastArgs","int32");if(e.rank!==1)throw new Error(`broadcastArgs(): first input must be a vector (rank=1). Has rank ${e.rank}`);if(n.rank!==1)throw new Error(`broadcastArgs(): second input must be a vector (rank=1). Has rank ${n.rank}`);let o={s0:e,s1:n};return _.runKernel(pp,o)}var S0=N({broadcastArgs_:WH});function UH(r,t){let e=C(r,"broadcastTo","x"),n=e.shape;if(t.some(l=>!(l>0)||l%1!==0))throw new Error(`broadcastTo(): Invalid broadcast shape [${t}].`);if(t.length<e.rank)throw new Error(`broadcastTo(): shape.length=${t.length} < input.rank=${e.rank}.`);if(t.length>e.rank){let l=e.shape.slice();for(;l.length<t.length;)l.unshift(1);e=O(e,l)}let o=e.shape,s=Array.from(t);for(let l=t.length-1;l>=0;l--)if(o[l]===t[l])s[l]=1;else if(e.shape[l]!==1)throw new Error(`broadcastTo(): [${n}] cannot be broadcast to [${t}].`);if(s.map((l,c)=>l>1?c:-1).filter(l=>l>=0).length===0)return In(e);let a={x:e},u={reps:s};return _.runKernel(jn,a,u)}var Wa=N({broadcastTo_:UH});function HH(r){let e={x:C(r,"x","ceil","float32")};return _.runKernel(Jo,e)}var Vf=N({ceil_:HH});function qH(r,t,e){let n=C(r,"x","clipByValue");A(t<=e,()=>`Error in clip: min (${t}) must be less than or equal to max (${e}).`);let o={x:n},s={clipValueMin:t,clipValueMax:e};return _.runKernel(so,o,s)}var br=N({clipByValue_:qH});function KH(r){return se(r,0)}var k0=N({concat1d_:KH});function jH(r,t){return se(r,t)}var N0=N({concat2d_:jH});function XH(r,t){return se(r,t)}var T0=N({concat3d_:XH});function YH(r,t){return se(r,t)}var _0=N({concat4d_:YH});function ZH(r,t,e,n,o="NHWC",s=[1,1],i){let a=C(r,"x","conv2d","float32"),u=C(t,"filter","conv2d","float32"),l=a,c=!1;a.rank===3&&(c=!0,l=O(a,[1,a.shape[0],a.shape[1],a.shape[2]])),A(l.rank===4,()=>`Error in conv2d: input must be rank 4, but got rank ${l.rank}.`),A(u.rank===4,()=>`Error in conv2d: filter must be rank 4, but got rank ${u.rank}.`),ke("conv2d",n,i);let p=o==="NHWC"?l.shape[3]:l.shape[1];A(p===u.shape[2],()=>`Error in conv2d: depth of input (${p}) must match input depth for filter ${u.shape[2]}.`),A(Dr(e,s),()=>`Error in conv2D: Either strides or dilations must be 1. Got strides ${e} and dilations '${s}'`);let m={x:l,filter:u},f={strides:e,pad:n,dataFormat:o,dilations:s,dimRoundingMode:i},d=_.runKernel(Qo,m,f);return c?O(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var ln=N({conv2d_:ZH});function JH(r,t,e,n,o="NWC",s=1,i){let a=C(r,"x","conv1d"),u=C(t,"filter","conv1d"),l=a,c=!1;a.rank===2&&(c=!0,l=O(a,[1,a.shape[0],a.shape[1]])),A(l.rank===3,()=>`Error in conv1d: input must be rank 3, but got rank ${l.rank}.`),A(u.rank===3,()=>`Error in conv1d: filter must be rank 3, but got rank ${u.rank}.`),ke("conv1d",n,i),A(l.shape[2]===u.shape[1],()=>`Error in conv1d: depth of input (${l.shape[2]}) must match input depth for filter ${u.shape[1]}.`),A(Dr(e,s),()=>`Error in conv1D: Either stride or dilation must be 1. Got stride ${e} and dilation '${s}'`),A(o==="NWC",()=>`Error in conv1d: got dataFormat of ${o} but only NWC is currently supported.`);let p=O(u,[1,u.shape[0],u.shape[1],u.shape[2]]),m=O(l,[l.shape[0],1,l.shape[1],l.shape[2]]),g=ln(m,p,[1,e],n,"NHWC",[1,s],i);return c?O(g,[g.shape[2],g.shape[3]]):O(g,[g.shape[0],g.shape[2],g.shape[3]])}var Pu=N({conv1d_:JH});function QH(r,t,e,n,o,s="NHWC",i){A(r.length===t.rank,()=>`Length of inShape (${r.length}) and rank of dy (${t.rank}) must match`);let a=r,u=t,l=!1;t.rank===3&&(l=!0,u=O(t,[1,t.shape[0],t.shape[1],t.shape[2]]),a=[1,r[0],r[1],r[2]]),A(a.length===4,()=>`Error in conv2dDerInput: inShape must be length 4, but got length ${a.length}.`),A(u.rank===4,()=>`Error in conv2dDerInput: dy must be rank 4, but got rank ${u.rank}`),A(e.rank===4,()=>`Error in conv2dDerInput: filter must be rank 4, but got rank ${e.rank}`);let c=s==="NHWC"?a[3]:a[1],p=s==="NHWC"?u.shape[3]:u.shape[1];A(c===e.shape[2],()=>`Error in conv2dDerInput: depth of input (${c}) must match input depth for filter ${e.shape[2]}.`),A(p===e.shape[3],()=>`Error in conv2dDerInput: depth of output (${p}) must match output depth for filter ${e.shape[3]}.`),ke("conv2dDerInput",o,i);let m={dy:u,filter:e},f={strides:n,pad:o,dataFormat:s,dimRoundingMode:i,inputShape:a},d=_.runKernel(ts,m,f);return l?O(d,[d.shape[1],d.shape[2],d.shape[3]]):d}var Zp=N({conv2DBackpropInput_:QH});function tq(r,t,e,n,o,s){let i=C(r,"x","conv2dTranspose"),a=C(t,"filter","conv2dTranspose");return Zp(e,i,a,n,o,"NHWC",s)}var Mu=N({conv2dTranspose_:tq});function eq(r,t,e,n,o="NDHWC",s=[1,1,1]){let i=C(r,"x","conv3d"),a=C(t,"filter","conv3d"),u=i,l=!1;i.rank===4&&(l=!0,u=O(i,[1,i.shape[0],i.shape[1],i.shape[2],i.shape[3]])),A(u.rank===5,()=>`Error in conv3d: input must be rank 5, but got rank ${u.rank}.`),A(a.rank===5,()=>`Error in conv3d: filter must be rank 5, but got rank ${a.rank}.`),A(u.shape[4]===a.shape[3],()=>`Error in conv3d: depth of input (${u.shape[4]}) must match input depth for filter ${a.shape[3]}.`),A(Dr(e,s),()=>`Error in conv3D: Either strides or dilations must be 1. Got strides ${e} and dilations '${s}'`),A(o==="NDHWC",()=>`Error in conv3d: got dataFormat of ${o} but only NDHWC is currently supported.`);let c={x:u,filter:a},p={strides:e,pad:n,dataFormat:o,dilations:s},m=_.runKernel(Sl,c,p);return l?O(m,[m.shape[1],m.shape[2],m.shape[3],m.shape[4]]):m}var Gf=N({conv3d_:eq});function rq(r,t,e,n,o){A(r.length===t.rank,()=>`Length of inShape (${r.length}) and rank of dy (${t.rank}) must match`);let s=r,i=t,a=!1;t.rank===4&&(a=!0,i=O(t,[1,t.shape[0],t.shape[1],t.shape[2],t.shape[3]]),s=[1,r[0],r[1],r[2],r[3]]);let u=s[4],l=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(e.rank===5,()=>`Error in conv3dDerInput: filter must be rank 5, but got rank ${e.rank}`),A(u===e.shape[3],()=>`Error in conv3dDerInput: depth of input (${u}) must match input depth for filter ${e.shape[3]}.`),A(l===e.shape[4],()=>`Error in conv3dDerInput: depth of output (${l}) must match output depth for filter ${e.shape[4]}.`);let c={dy:i,filter:e},p={pad:o,strides:n,inputShape:s},m=_.runKernel(hp,c,p);return a?O(m,[m.shape[1],m.shape[2],m.shape[3],m.shape[4]]):m}var wx=N({conv3DBackpropInput_:rq});function nq(r,t,e,n,o){let s=C(r,"x","conv3dTranspose"),i=C(t,"filter","conv3dTranspose");return wx(e,s,i,n,o)}var E0=N({conv3dTranspose_:nq});function oq(r){let e={x:C(r,"x","cos","float32")};return _.runKernel(es,e)}var Ua=N({cos_:oq});function sq(r){let e={x:C(r,"x","cosh","float32")};return _.runKernel(rs,e)}var Lu=N({cosh_:sq});function iq(r,t=0,e=!1,n=!1){let s={x:C(r,"x","cumprod")},i={axis:t,exclusive:e,reverse:n};return _.runKernel(na,s,i)}var Wf=N({cumprod_:iq});function aq(r,t=0,e=!1,n=!1){let s={x:C(r,"x","cumsum")},i={axis:t,exclusive:e,reverse:n};return _.runKernel(ns,s,i)}var zu=N({cumsum_:aq});function lq(r,t,e,n=!1){let o=C(r,"x","denseBincount"),s=C(t,"weights","denseBincount");A(o.dtype==="int32",()=>`Error in denseBincount: input dtype must be int32, but got ${o.dtype}`),A(o.rank<=2,()=>`Error in denseBincount: input must be at most rank 2, but got rank ${o.rank}.`),A(e>=0,()=>`size must be non-negative, but got ${e}.`),A(s.size===o.size||s.size===0,()=>`Error in denseBincount: weights must have the same shape as x or 0-length, but got x shape: ${o.shape}, weights shape: ${s.shape}.`);let i={x:o,weights:s},a={size:e,binaryOutput:n};return _.runKernel(gp,i,a)}var A0=N({denseBincount_:lq});function uq(r,t,e="NHWC"){let n=C(r,"x","depthToSpace","float32"),o=e==="NHWC"?n.shape[1]:n.shape[2],s=e==="NHWC"?n.shape[2]:n.shape[3],i=e==="NHWC"?n.shape[3]:n.shape[1];A(t>1,()=>`blockSize should be > 1 for depthToSpace, but was: ${t}`),A(o*t>=0,()=>`Negative dimension size caused by overflow when multiplying
${o} and ${t} for depthToSpace with input shape
${n.shape}`),A(s*t>=0,()=>`Negative dimension size caused by overflow when multiplying
${s} and ${t} for depthToSpace with input shape
${n.shape}`),A(i%(t*t)===0,()=>`Dimension size must be evenly divisible by ${t*t} but is ${i} for depthToSpace with input shape ${n.shape}`);let a={x:n},u={blockSize:t,dataFormat:e};return _.runKernel(sa,a,u)}var Uf=N({depthToSpace_:uq});function cq(r,t,e,n,o="NHWC",s=[1,1],i){let a=C(r,"x","depthwiseConv2d","float32"),u=C(t,"filter","depthwiseConv2d","float32"),l=a,c=!1;a.rank===3&&(c=!0,l=O(a,[1,a.shape[0],a.shape[1],a.shape[2]])),A(l.rank===4,()=>`Error in depthwiseConv2d: input must be rank 4, but got rank ${l.rank}.`),A(u.rank===4,()=>`Error in depthwiseConv2d: filter must be rank 4, but got rank ${u.rank}.`),A(l.shape[3]===u.shape[2],()=>`Error in depthwiseConv2d: number of input channels (${l.shape[3]}) must match the inChannels dimension in filter ${u.shape[2]}.`),ke("depthwiseConv2d",n,i);let p={x:l,filter:u},m={strides:e,pad:n,dataFormat:o,dilations:s,dimRoundingMode:i},f=_.runKernel(os,p,m);return c?O(f,[f.shape[1],f.shape[2],f.shape[3]]):f}var Ks=N({depthwiseConv2d_:cq});function pq(r){let e={x:C(r,"x","diag")};return _.runKernel(bp,e)}var mq=N({diag_:pq});function fq(r,t,e,n,o=[1,1],s="NHWC"){let i=C(r,"x","dilation2d"),a=C(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(a.rank===3,()=>`Error in dilation2d: filter must be rank 3, but got rank ${a.rank}.`),A(s==="NHWC",()=>`Error in dilation2d: Only NHWC is currently supported, but got dataFormat of ${s}`);let u=i,l=!1;i.rank===3&&(u=O(i,[1,i.shape[0],i.shape[1],i.shape[2]]),l=!0);let c={x:u,filter:a},p={strides:e,pad:n,dilations:o},m=_.runKernel(kl,c,p);return l?O(m,[m.shape[1],m.shape[2],m.shape[3]]):m}var Hf=N({dilation2d_:fq});function dq(r,t){let e=C(r,"a","equal","string_or_numeric"),n=C(t,"b","equal","string_or_numeric");[e,n]=jt(e,n),Lt(e.shape,n.shape);let o={a:e,b:n};return _.runKernel(aa,o)}var Sr=N({equal_:dq});function hq(r,t,e){let n=C(t,"a","where"),o=C(e,"b","where"),s=C(r,"condition","where","bool"),i=Lt(Lt(s.shape,n.shape),o.shape),a=Wa(s,i),u=Wa(n,i),l=Wa(o,i),c={condition:a,t:u,e:l};return _.runKernel(Si,c)}var Ne=N({where_:hq});function gq(r){let e={x:C(r,"x","zerosLike")};return _.runKernel(Ei,e)}var St=N({zerosLike_:gq});function xq(r,t){let e=C(r,"a","div"),n=C(t,"b","div");[e,n]=jt(e,n);let o=ct(e,n),s=St(o),i=Sr(n,s);return Ne(i,s,o)}var qf=N({divNoNan_:xq});function yq(r,t){let e=C(r,"t1","dot"),n=C(t,"t2","dot");A((e.rank===1||e.rank===2)&&(n.rank===1||n.rank===2),()=>`Error in dot: inputs must all be rank 1 or 2, but got ranks ${e.rank} and ${n.rank}.`);let o=e.rank===1?e.size:e.shape[1],s=n.rank===1?n.size:n.shape[0];if(A(o===s,()=>`Error in dot: inner dimensions of inputs must match, but got ${o} and ${s}.`),e.rank===1&&n.rank===1){let i=O(e,[1,-1]),a=O(n,[-1,1]),u=Mt(i,a);return O(u,[])}else if(e.rank===1&&n.rank===2){let i=O(e,[1,-1]),a=O(n,[n.shape[0],n.shape[1]]),u=Mt(i,a);return O(u,[u.size])}else if(e.rank===2&&n.rank===1){let i=O(n,[-1,1]),a=Mt(e,i);return O(a,[a.size])}else{let i=O(n,[n.shape[0],n.shape[1]]);return Mt(e,i)}}var D0=N({dot_:yq});function bq(r,...t){let e=t.map((o,s)=>C(o,`tensors${s}`,"einsum")),n={equation:r};return _.runKernel(wp,e,n)}var $0=N({einsum_:bq});function wq(r){let e={x:C(r,"x","elu","float32")};return _.runKernel(is,e)}var js=N({elu_:wq});function vq(r){let t=C(r,"x","erf");A(t.dtype==="int32"||t.dtype==="float32",()=>"Input dtype must be `int32` or `float32`."),t.dtype==="int32"&&(t=Z(t,"float32"));let e={x:t};return _.runKernel(ia,e)}var Kf=N({erf_:vq});function Cq(r){let e={x:C(r,"x","exp")};return _.runKernel(as,e)}var Ze=N({exp_:Cq});function Iq(r,t=0){let e=C(r,"x","expandDims","string_or_numeric");A(t<=e.rank,()=>"Axis must be <= rank of the tensor");let n={input:e},o={dim:t};return _.runKernel(yi,n,o)}var fr=N({expandDims_:Iq});function Sq(r){let e={x:C(r,"x","expm1")};return _.runKernel(la,e)}var jf=N({expm1_:Sq});function kq(r,t){let e=C(r,"x","tile","string_or_numeric");A(e.rank===t.length,()=>`Error in transpose: rank of input ${e.rank} must match length of reps ${t}.`);let n={x:e},o={reps:t};return _.runKernel(jn,n,o)}var kr=N({tile_:kq});function Nq(r,t,e,n="float32"){t==null&&(t=r);let o=It([r,t],n),s=r<=t?r:t;for(let a=0;a<s;++a)o.set(1,a,a);let i=O(o.toTensor(),[r,t]);if(e==null)return i;if(e.length===1)return kr(fr(i,0),[e[0],1,1]);if(e.length===2)return kr(fr(fr(i,0),0),[e[0],e[1],1,1]);if(e.length===3)return kr(fr(fr(fr(i,0),0),0),[e[0],e[1],e[2],1,1]);throw new Error(`eye() currently supports only 1D and 2D batchShapes, but received ${e.length}D.`)}var Jp=N({eye_:Nq});function Xs(r,t,e){let n={shape:r,value:t,dtype:e};return _.runKernel(Nl,{},n)}function Tq(r){let e={x:C(r,"x","floor","float32")};return _.runKernel(ls,e)}var Ys=N({floor_:Tq});function _q(r,t,e=0,n=0){let o=C(r,"x","gather"),s=C(t,"indices","gather","int32"),i={x:o,indices:s},a={axis:e,batchDims:n};return _.runKernel(bi,i,a)}var go=N({gather_:_q});function Eq(r,t){let e=C(r,"a","greater","string_or_numeric"),n=C(t,"b","greater","string_or_numeric");[e,n]=jt(e,n),Lt(e.shape,n.shape);let o={a:e,b:n};return _.runKernel(pa,o)}var Ge=N({greater_:Eq});function Aq(r,t){let e=C(r,"a","greaterEqual","string_or_numeric"),n=C(t,"b","greaterEqual","string_or_numeric");[e,n]=jt(e,n),Lt(e.shape,n.shape);let o={a:e,b:n};return _.runKernel(ps,o)}var kn=N({greaterEqual_:Aq});function Dq(r){let e={input:C(r,"input","imag")};return _.runKernel(Sp,e)}var Bu=N({imag_:Dq});function $q(r){let e={x:C(r,"x","isFinite")};return _.runKernel(ma,e)}var R0=N({isFinite_:$q});function Rq(r){let e={x:C(r,"x","isInf")};return _.runKernel(fa,e)}var F0=N({isInf_:Rq});function Fq(r){let e={x:C(r,"x","isNaN")};return _.runKernel(da,e)}var Xf=N({isNaN_:Fq});function Oq(r,t=.2){let n={x:C(r,"x","leakyRelu")},o={alpha:t};return _.runKernel(ms,n,o)}var Ha=N({leakyRelu_:Oq});function Pq(r,t){let e=C(r,"a","less","string_or_numeric"),n=C(t,"b","less","string_or_numeric");[e,n]=jt(e,n),Lt(e.shape,n.shape);let o={a:e,b:n};return _.runKernel(ha,o)}var Vu=N({less_:Pq});function Mq(r,t){let e=C(r,"a","lessEqual","string_or_numeric"),n=C(t,"b","lessEqual","string_or_numeric");[e,n]=jt(e,n),Lt(e.shape,n.shape);let o={a:e,b:n};return _.runKernel(ga,o)}var Nn=N({lessEqual_:Mq});function O0(r,t,e){if(e<=0)throw new Error("The number of values should be positive.");let n={start:r,stop:t,num:e};return _.runKernel(kp,{},n)}function Lq(r,t=5,e=1,n=1,o=.5){let s=C(r,"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(Ki(t),()=>`Error in localResponseNormalization: depthRadius must be an integer but got depthRadius ${t}.`);let i=s,a=!1;s.rank===3&&(a=!0,i=O(s,[1,s.shape[0],s.shape[1],s.shape[2]]));let u={x:i},l={depthRadius:t,bias:e,alpha:n,beta:o},c=_.runKernel(Tl,u,l);return a?O(c,[c.shape[1],c.shape[2],c.shape[3]]):c}var Yf=N({localResponseNormalization_:Lq});function zq(r){let e={x:C(r,"x","log","float32")};return _.runKernel(fs,e)}var wr=N({log_:zq});function Bq(r){let e={x:C(r,"x","log1p")};return _.runKernel(xa,e)}var qa=N({log1p_:Bq});function Vq(r){return A(fi(r),()=>"The f passed in grad(f) must be a function"),(t,e)=>{let n=C(t,"x","tf.grad","string_or_numeric"),o=e!=null?C(e,"dy","tf.grad"):null;return _.tidy(()=>{let{value:s,grads:i}=_.gradients(()=>r(n),[n],o);return o!=null&&Pe(s.shape,o.shape,"The shape of dy passed in grad(f)(x, dy) must match the shape returned by f(x)"),Cx(i),i[0]})}}function Gq(r){return A(fi(r),()=>"The f passed in grads(f) must be a function"),(t,e)=>{A(Array.isArray(t),()=>"The args passed in grads(f)(args) must be an array of `Tensor`s or `TensorLike`s");let n=Pa(t,"args","tf.grads","string_or_numeric"),o=e!=null?C(e,"dy","tf.grads"):null;return _.tidy(()=>{let{value:s,grads:i}=_.gradients(()=>r(...n),n,o);return o!=null&&Pe(s.shape,o.shape,"The shape of dy passed in grads(f)([x1,...], dy) must match the shape returned by f([x1,...])"),Cx(i),i})}}function Wq(r){return A(fi(r),()=>"The f passed in valueAndGrad(f) must be a function"),(t,e)=>{A(t instanceof Vt,()=>"The x passed in valueAndGrad(f)(x) must be a tensor"),A(e==null||e instanceof Vt,()=>"The dy passed in valueAndGrad(f)(x, dy) must be a tensor");let{grads:n,value:o}=_.gradients(()=>r(t),[t],e);return Cx(n),{grad:n[0],value:o}}}function Uq(r){return A(fi(r),()=>"The f passed in valueAndGrads(f) must be a function"),(t,e)=>{A(Array.isArray(t)&&t.every(o=>o instanceof Vt),()=>"The args passed in valueAndGrads(f)(args) must be array of tensors"),A(e==null||e instanceof Vt,()=>"The dy passed in valueAndGrads(f)(args, dy) must be a tensor");let n=_.gradients(()=>r(...t),t,e);return e!=null&&Pe(n.value.shape,e.shape,"The shape of dy passed in valueAndGrads(f)([x1,...], dy) must match the shape returned by f([x1,...])"),Cx(n.grads),n}}function vx(r,t){A(fi(r),()=>"The f passed in variableGrads(f) must be a function"),A(t==null||Array.isArray(t)&&t.every(l=>l instanceof Ll),()=>"The varList passed in variableGrads(f, varList) must be an array of variables");let e=t!=null;if(!e){t=[];for(let l in _.registeredVariables)t.push(_.registeredVariables[l])}let n=e?t.filter(l=>!l.trainable):null,o=t.length;t=t.filter(l=>l.trainable),A(t.length>0,()=>`variableGrads() expects at least one of the input variables to be trainable, but none of the ${o} variables is trainable.`);let s=!0,{value:i,grads:a}=_.gradients(r,t,null,s);A(a.some(l=>l!=null),()=>"Cannot find a connection between any variable and the result of the loss function y=f(x). Please make sure the operations that use variables are inside the function f passed to minimize()."),A(i.rank===0,()=>`The f passed in variableGrads(f) must return a scalar, but it returned a rank-${i.rank} tensor`);let u={};return t.forEach((l,c)=>{a[c]!=null&&(u[l.name]=a[c])}),n!=null&&n.forEach(l=>u[l.name]=null),{value:i,grads:u}}function un(r){return _.customGrad(r)}function Cx(r){if(r.filter(e=>e==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 Hq(r){let e={x:C(r,"x","neg")};return _.runKernel(wi,e)}var Zt=N({neg_:Hq});function qq(r){let e={x:C(r,"x","softplus")};return _.runKernel(Aa,e)}var xo=N({softplus_:qq});function Kq(r){let t=C(r,"x","logSigmoid");return un(n=>({value:Zt(xo(Zt(n))),gradFunc:i=>P(i,Lr(Zt(n)))}))(t)}var P0=N({logSigmoid_:Kq});function jq(r,t=null,e=!1){let o={x:C(r,"x","max")},s={reductionIndices:t,keepDims:e};return _.runKernel(ds,o,s)}var $r=N({max_:jq});function Xq(r,t){let e=C(r,"a","sub"),n=C(t,"b","sub");[e,n]=jt(e,n);let o={a:e,b:n};return _.runKernel(Ls,o)}var lt=N({sub_:Xq});function Yq(r,t=null,e=!1){let n=C(r,"x","sum");n.dtype==="bool"&&(n=Z(n,"int32"));let o={x:n},s={axis:t,keepDims:e};return _.runKernel(Os,o,s)}var mt=N({sum_:Yq});function Zq(r,t=-1){let e=C(r,"logits","logSoftmax");if(t===-1&&(t=e.rank-1),t!==e.rank-1)throw Error(`Log Softmax along a non-last dimension is not yet supported. Logits was rank ${e.rank} and axis was ${t}`);return un((o,s)=>{let a=$r(o,t,!0),u=lt(o,a),l=lt(Z(u,"float32"),wr(mt(Ze(u),t,!0)));return s([l]),{value:l,gradFunc:(p,m)=>{let[f]=m,d=!0,h=Ze(f);return lt(p,P(mt(p,t,d),h))}}})(e)}var Gu=N({logSoftmax_:Zq});function M0(r,t){for(let e=0;e<r.length;++e)if(r[r.length-e-1]!==t-1-e)return!1;return!0}function E_(r,t,e){let n=r.length+t.length,o=[],s=0,i=0;for(let a=0;a<n;a++)e.indexOf(a)===-1?o.push(r[s++]):o.push(t[i++]);return o}function L0(r,t){let e=[],n=r.length;for(let s=0;s<n;s++)t.indexOf(s)===-1&&e.push(r[s]);let o=t.map(s=>r[s]);return[e,o]}function yo(r,t){let e=t.map(n=>1);return E_(r,e,t)}function Jq(r,t,e){A(M0(t,e),()=>`${r} supports only inner-most axes for now. Got axes ${t} and rank-${e} input.`)}function z0(r,t){if(M0(r,t))return null;let e=[];for(let n=0;n<t;++n)r.indexOf(n)===-1&&e.push(n);return r.forEach(n=>e.push(n)),e}function Zf(r){return r.map((t,e)=>[e,t]).sort((t,e)=>t[1]-e[1]).map(t=>t[0])}function Qq(r,t){let e=[];for(let n=t-r;n<t;++n)e.push(n);return e}function tK(r,t=null,e=!1){let n=C(r,"x","logSumExp"),o=mr(t,n.shape),s=$r(n,o,!0),i=lt(n,s),a=Ze(i),u=mt(a,o),l=wr(u),c=J(O(s,l.shape),l);if(e){let p=yo(c.shape,o);return O(c,p)}return c}var Jf=N({logSumExp_:tK});function eK(r,t){let e=C(r,"a","logicalAnd","bool"),n=C(t,"b","logicalAnd","bool");Lt(e.shape,n.shape);let o={a:e,b:n};return _.runKernel(ya,o)}var Nr=N({logicalAnd_:eK});function rK(r){let e={x:C(r,"x","logicalNot","bool")};return _.runKernel(wu,e)}var Ka=N({logicalNot_:rK});function nK(r,t){let e=C(r,"a","logicalOr","bool"),n=C(t,"b","logicalOr","bool");Lt(e.shape,n.shape);let o={a:e,b:n};return _.runKernel(vu,o)}var Wu=N({logicalOr_:nK});function oK(r,t){let e=C(r,"a","logicalXor","bool"),n=C(t,"b","logicalXor","bool");return Lt(e.shape,n.shape),Nr(Wu(r,t),Ka(Nr(r,t)))}var V0=N({logicalXor_:oK});function sK(r,t,e,n,o){let s=C(r,"x","maxPool"),i=1,a=s,u=!1;s.rank===3&&(u=!0,a=O(s,[1,s.shape[0],s.shape[1],s.shape[2]])),A(a.rank===4,()=>`Error in maxPool: input must be rank 4 but got rank ${a.rank}.`),A(Dr(e,i),()=>`Error in maxPool: Either strides or dilations must be 1. Got strides ${e} and dilations '${i}'`),ke("maxPool",n,o);let l={x:a},c={filterSize:t,strides:e,pad:n,dimRoundingMode:o},p=_.runKernel(gs,l,c);return u?O(p,[p.shape[1],p.shape[2],p.shape[3]]):p}var ja=N({maxPool_:sK});function iK(r,t=[1,1,1],e,n,o,s="NDHWC"){let i=C(r,"x","maxPool3d"),a=i,u=!1;i.rank===4&&(u=!0,a=O(i,[1,i.shape[0],i.shape[1],i.shape[2],i.shape[3]])),A(a.rank===5,()=>`Error in maxPool3d: x must be rank 5 but got rank ${a.rank}.`),A(s==="NDHWC",()=>`Error in maxPool3d: Only NDHWC is currently supported, but got dataFormat of ${s}`),ke("maxPool3d",n,o);let l={x:a},c={filterSize:t,strides:e,pad:n,dimRoundingMode:o,dataFormat:s},p=_.runKernel(_l,l,c);return u?O(p,[p.shape[1],p.shape[2],p.shape[3],p.shape[4]]):p}var Qf=N({maxPool3d_:iK});function aK(r,t,e,n,o=!1){let i={x:C(r,"x","maxPoolWithArgmax")},a={filterSize:t,strides:e,pad:n,includeBatchInIndex:o},u=_.runKernel(Ep,i,a);return{result:u[0],indexes:u[1]}}var G0=N({maxPoolWithArgmax_:aK});function lK(r,t){let e=C(r,"a","maximum"),n=C(t,"b","maximum");[e,n]=jt(e,n),e.dtype==="bool"&&(e=Z(e,"int32"),n=Z(n,"int32")),Lt(e.shape,n.shape);let o={a:e,b:n};return _.runKernel(hs,o)}var cn=N({maximum_:lK});function uK(r,t=null,e=!1){let o={x:C(r,"x","mean")},s={axis:t,keepDims:e};return _.runKernel(xs,o,s)}var xe=N({mean_:uK});function ye(r,t="float32"){if(t==="complex64"){let n=ye(r,"float32"),o=ye(r,"float32");return Mn(n,o)}let e=ap(re(r),t);return _.makeTensor(e,r,t)}function lr(r,t="float32"){if(t==="complex64"){let n=lr(r,"float32"),o=ye(r,"float32");return Mn(n,o)}let e=df(re(r),t);return _.makeTensor(e,r,t)}function cK(r,t,{indexing:e="xy"}={}){if(e!=="xy"&&e!=="ij")throw new TypeError(`${e} is not a valid third argument to meshgrid`);if(r===void 0)return[];let n=C(r,"x","meshgrid",r instanceof Vt?r.dtype:"float32");if(t===void 0)return[n];let o=C(t,"y","meshgrid",t instanceof Vt?t.dtype:"float32"),s=re(n.shape),i=re(o.shape);return e==="xy"?(n=O(n,[1,-1]),o=O(o,[-1,1]),[Mt(lr([i,1],n.dtype),n),Mt(o,lr([1,s],o.dtype))]):(n=O(n,[-1,1]),o=O(o,[1,-1]),[Mt(n,lr([1,i],n.dtype)),Mt(lr([s,1],o.dtype),o)])}function pK(r,t=null,e=!1){let o={x:C(r,"x","min")},s={axis:t,keepDims:e};return _.runKernel(ys,o,s)}var Wl=N({min_:pK});function mK(r,t){let e=C(r,"a","minimum"),n=C(t,"b","minimum");[e,n]=jt(e,n),e.dtype==="bool"&&(e=Z(e,"int32"),n=Z(n,"int32")),Lt(e.shape,n.shape);let o={a:e,b:n};return _.runKernel(bs,o)}var Zs=N({minimum_:mK});function fK(r,t,e){A(e==="reflect"||e==="symmetric",()=>`Invalid mode. Mode must be either reflect or symmetric. Got ${e}.`);let n=C(r,"x","mirrorPad");if(n.rank===0)throw new Error("mirrorPad(scalar) is not defined. Pass non-scalar to mirrorPad");A(t.length===n.rank,()=>`Padding doesn't match input. Must be ${n.rank}. Got ${t.length}.`);let o=e==="reflect"?1:0;for(let a=0;a<n.rank;a++)A(t[a].length===2,()=>"Invalid number of paddings. Must be length of 2 each."),A(t[a][0]>=0&&t[a][0]<=n.shape[a]-o&&t[a][1]>=0&&t[a][1]<=n.shape[a]-o,()=>`Padding in dimension ${a} cannot be greater than or equal to ${n.shape[a]-o} or less than 0 for input of shape ${n.shape}`);let s={paddings:t,mode:e},i={x:n};return _.runKernel(ws,i,s)}var td=N({mirrorPad_:fK});function dK(r,t){let e=C(r,"a","mod"),n=C(t,"b","mod");[e,n]=jt(e,n);let o={a:e,b:n};return _.runKernel(ba,o)}var ed=N({mod_:dK});function hK(r){let t=C(r,"x","square"),e={};return _.runKernel("Square",{x:t},e)}var Wt=N({square_:hK});function gK(r,t=null,e=!1){r=C(r,"x","moments");let n=mr(t,r.shape),o=xe(r,n,e),s=o.shape;e||(s=yo(o.shape,n));let i=Wt(lt(Z(r,"float32"),O(o,s))),a=xe(i,n,e);return{mean:o,variance:a}}var Qp=N({moments_:gK});function xK(r,t,e,n){let o=C(t,"data","multiRNNCell"),s=Pa(e,"c","multiRNNCell"),i=Pa(n,"h","multiRNNCell"),a=o,u=[];for(let p=0;p<r.length;p++){let m=r[p](a,s[p],i[p]);u.push(m[0]),u.push(m[1]),a=m[1]}let l=[],c=[];for(let p=0;p<u.length;p+=2)l.push(u[p]),c.push(u[p+1]);return[l,c]}var yK=N({multiRNNCell_:xK});function bK(r,t,e,n=!1){let o=C(r,"logits","multinomial"),s=o.size,i=o.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}`);e=e||Math.random();let u={logits:i===1?O(o,[1,-1]):o},l={numSamples:t,seed:e,normalized:n},c=_.runKernel(Ap,u,l);return i===1?O(c,[c.size]):c}var W0=N({multinomial_:bK});function wK(r,t){let e=C(r,"a","notEqual","string_or_numeric"),n=C(t,"b","notEqual","string_or_numeric");[e,n]=jt(e,n),Lt(e.shape,n.shape);let o={a:e,b:n};return _.runKernel(wa,o)}var bo=N({notEqual_:wK});function vK(r){let e={x:C(r,"x","onesLike")};return _.runKernel(vi,e)}var dr=N({onesLike_:vK});function CK(r,t){let e=C(r,"v1","outerProduct"),n=C(t,"v2","outerProduct");A(e.rank===1&&n.rank===1,()=>`Error in outerProduct: inputs must be rank 1, but got ranks ${e.rank} and ${n.rank}.`);let o=O(e,[-1,1]),s=O(n,[1,-1]);return Mt(o,s)}var IK=N({outerProduct_:CK});function SK(r,t,e=0){let n=C(r,"x","pad");if(n.rank===0)throw new Error("pad(scalar) is not defined. Pass non-scalar to pad");let o={paddings:t,constantValue:e},s={x:n};return _.runKernel(Is,s,o)}var qr=N({pad_:SK});function kK(r,t,e=0){return A(t.length===2,()=>"Invalid number of paddings. Must be length of 2."),qr(r,[t],e)}var NK=N({pad1d_:kK});function TK(r,t,e=0){return A(t.length===2&&t[0].length===2&&t[1].length===2,()=>"Invalid number of paddings. Must be length of 2 each."),qr(r,t,e)}var _K=N({pad2d_:TK});function EK(r,t,e=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."),qr(r,t,e)}var AK=N({pad3d_:EK});function DK(r,t,e=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."),qr(r,t,e)}var $K=N({pad4d_:DK});function RK(r,t,e){let n=C(r,"x","spaceToBatchND");A(n.rank>=1+t.length,()=>`input rank ${n.rank} should be > than [blockShape] ${t.length}`),A(e.length===t.length,()=>`paddings.shape[0] ${e.length} must be equal to [blockShape] ${t.length}`),A(n.shape.reduce((i,a,u)=>u>0&&u<=t.length?i&&(a+e[u-1][0]+e[u-1][1])%t[u-1]===0:i,!0),()=>`input spatial dimensions ${n.shape.slice(1)} with paddings ${e.toString()} must be divisible by blockShapes ${t.toString()}`);let o={x:n},s={blockShape:t,paddings:e};return _.runKernel(Ni,o,s)}var Xa=N({spaceToBatchND_:RK});function FK(r,t,e,n,o,s,i){o==null&&(o=[1,1]),s==null&&(s=1),n===0&&(n="valid");let a=C(r,"x","maxPool"),u=a,l=!1;a.rank===3&&(l=!0,u=O(a,[1,a.shape[0],a.shape[1],a.shape[2]])),A(Dr(s,o),()=>`Error in pool: Either strides or dilations must be 1. Got strides ${s} and dilations '${o}'`);let c=b0(u.shape,t,s,o,n),p=[c.dilationHeight,c.dilationWidth],m;n==="same"?m=PK([c.filterHeight,c.filterWidth],p):m=[[0,0],[0,0]];let f=p[0]===1&&p[1]===1,[d,h]=OK([c.inHeight,c.inWidth],p,m),g=f?n:"valid",y=f?u:Xa(u,p,d),w=(e==="avg"?()=>Va(y,t,s,g,i):()=>ja(y,t,s,g,i))(),v=f?w:Ga(w,p,h);return l?O(v,[v.shape[1],v.shape[2],v.shape[3]]):v}function OK(r,t,e){let n=e.map(c=>c[0]),o=e.map(c=>c[1]),s=r.concat(n,o),i=t.map((c,p)=>(c-s[p]%c)%c),a=o.map((c,p)=>c+i[p]),u=t.map((c,p)=>[n[p],a[p]]),l=t.map((c,p)=>[0,i[p]]);return[u,l]}function PK(r,t){let n=r.map((i,a)=>i+(i-1)*(t[a]-1)).map(i=>i-1),o=n.map(i=>Math.floor(i/2)),s=n.map((i,a)=>i-o[a]);return n.map((i,a)=>[o[a],s[a]])}var U0=N({pool_:FK});function MK(r,t){let e=C(r,"base","pow"),n=C(t,"exp","pow");[e,n]=jt(e,n);let o={a:e,b:n};return _.runKernel(Ss,o)}var Kr=N({pow_:MK});function LK(r,t){let e=C(r,"x","prelu"),n=C(t,"alpha","prelu"),o={x:e,alpha:n};return _.runKernel(ks,o)}var Ya=N({prelu_:LK});function zK(r,t=null,e=!1){let n=C(r,"x","prod");n.dtype==="bool"&&(n=Z(n,"int32"));let o={x:n},s={axis:t,keepDims:e};return _.runKernel(Sa,o,s)}var Uu=N({prod_:zK});function BK(r,t,e){let n=re(r),o=null;if(e==null||e==="float32")o=new Float32Array(n);else if(e==="int32")o=new Int32Array(n);else if(e==="bool")o=new Uint8Array(n);else throw new Error(`Unknown data type ${e}`);for(let s=0;s<n;s++)o[s]=t();return _.makeTensor(o,r,e)}var VK=N({rand_:BK});var Sx=xu(J0());var tm=class{constructor(t,e,n,o,s){this.mean=t,this.stdDev=e,this.dtype=n,this.nextVal=NaN,this.truncated=o,this.truncated&&(this.upper=this.mean+this.stdDev*2,this.lower=this.mean-this.stdDev*2);let i=s||Math.random();this.random=Sx.alea(i.toString())}nextValue(){if(!isNaN(this.nextVal)){let o=this.nextVal;return this.nextVal=NaN,o}let t,e,n=!1;for(;!n;){let o,s,i;do o=2*this.random()-1,s=2*this.random()-1,i=o*o+s*s;while(i>=1||i===0);let a=Math.sqrt(-2*Math.log(i)/i);t=this.mean+this.stdDev*o*a,e=this.mean+this.stdDev*s*a,(!this.truncated||this.isValidTruncated(t))&&(n=!0)}return(!this.truncated||this.isValidTruncated(e))&&(this.nextVal=this.convertValue(e)),this.convertValue(t)}convertValue(t){return this.dtype==null||this.dtype==="float32"?t:Math.round(t)}isValidTruncated(t){return t<=this.upper&&t>=this.lower}},Q0=class{constructor(t,e,n,o){this.alpha=t,this.beta=1/e,this.dtype=n;let s=o||Math.random();this.randu=Sx.alea(s.toString()),this.randn=new tm(0,1,n,!1,this.randu()),t<1?this.d=t+2/3:this.d=t-1/3,this.c=1/Math.sqrt(9*this.d)}nextValue(){let t,e,n,o,s,i;for(;;){do o=this.randn.nextValue(),i=1+this.c*o;while(i<=0);if(i*=i*i,t=o*o,e=1-.331*t*t,n=.5*t+this.d*(1-i+Math.log(i)),s=this.randu(),s<e||Math.log(s)<n)break}return i=1/this.beta*this.d*i,this.alpha<1&&(i*=Math.pow(this.randu(),1/this.alpha)),this.convertValue(i)}convertValue(t){return this.dtype==="float32"?t:Math.round(t)}},tC=class{constructor(t=0,e=1,n,o){if(this.canReturnFloat=()=>this.dtype==null||this.dtype==="float32",this.min=t,this.range=e-t,this.dtype=n,o==null&&(o=Math.random()),typeof o=="number"&&(o=o.toString()),!this.canReturnFloat()&&this.range<=1)throw new Error(`The difference between ${t} - ${e} <= 1 and dtype is not float`);this.random=Sx.alea(o)}convertValue(t){return this.canReturnFloat()?t:Math.round(t)}nextValue(){return this.convertValue(this.min+this.range*this.random())}};function jK(r,t,e=1,n="float32",o){if(e==null&&(e=1),n==null&&(n="float32"),n!=="float32"&&n!=="int32")throw new Error(`Unsupported data type ${n}`);let s=new Q0(t,e,n,o),i=It(r,n);for(let a=0;a<i.values.length;a++)i.values[a]=s.nextValue();return i.toTensor()}var XK=N({randomGamma_:jK});function YK(r,t=0,e=1,n,o){if(n!=null&&n==="bool")throw new Error(`Unsupported data type ${n}`);let s=new tm(t,e,n,!1,o),i=It(r,n);for(let a=0;a<i.values.length;a++)i.values[a]=s.nextValue();return i.toTensor()}var kx=N({randomNormal_:YK});function ZK(r,t=0,e=1,n="float32",o){let s=It(r,n),i=new tC(t,e,null,o);for(let a=0;a<s.values.length;a++)s.values[a]=i.nextValue();return s.toTensor()}var Js=N({randomUniform_:ZK});function Za(r,t,e=1,n="float32"){if(e===0)throw new Error("Cannot have a step of zero");let o={start:r,stop:t,step:e,dtype:n};return _.runKernel(El,{},o)}function JK(r){let e={input:C(r,"input","real")};return _.runKernel(Dp,e)}var Ul=N({real_:JK});function QK(r){let e={x:C(r,"x","reciprocal")};return _.runKernel(ka,e)}var rd=N({reciprocal_:QK});function tj(r){let e={x:C(r,"x","relu")};return _.runKernel(Ns,e)}var Tr=N({relu_:tj});function ej(r){let e={x:C(r,"x","relu6")};return _.runKernel(_s,e)}var qu=N({relu6_:ej});function rj(r,t){let n={x:C(r,"x","reverse")},o={dims:t};return _.runKernel(Es,n,o)}var sr=N({reverse_:rj});function nj(r){let t=C(r,"x","reverse");return A(t.rank===1,()=>`Error in reverse1D: x must be rank 1 but got rank ${t.rank}.`),sr(t,0)}var oj=N({reverse1d_:nj});function sj(r,t){let e=C(r,"x","reverse");return A(e.rank===2,()=>`Error in reverse2D: x must be rank 2 but got rank ${e.rank}.`),sr(e,t)}var ij=N({reverse2d_:sj});function aj(r,t){let e=C(r,"x","reverse");return A(e.rank===3,()=>`Error in reverse3D: x must be rank 3 but got rank ${e.rank}.`),sr(e,t)}var lj=N({reverse3d_:aj});function uj(r,t){let e=C(r,"x","reverse");return A(e.rank===4,()=>`Error in reverse4D: x must be rank 4 but got rank ${e.rank}.`),sr(e,t)}var cj=N({reverse4d_:uj});function pj(r){let e={x:C(r,"x","round")};return _.runKernel(As,e)}var Ku=N({round_:pj});function mj(r){let e={x:C(r,"x","rsqrt","float32")};return _.runKernel(Ds,e)}var ju=N({rsqrt_:mj});function pt(r,t){if((yr(r)&&t!=="string"||Array.isArray(r))&&t!=="complex64")throw new Error("Error creating a new Scalar: value must be a primitive (number|boolean|string)");if(t==="string"&&yr(r)&&!(r instanceof Uint8Array))throw new Error("When making a scalar from encoded string, the value must be `Uint8Array`.");return sn(r,[],[],t)}function fj(r){let e={x:C(r,"x","selu")};return _.runKernel(Ta,e)}var Xu=N({selu_:fj});function dj(r,t,e,n,o,s=[1,1],i="NHWC"){let a=C(r,"x","separableConv2d"),u=C(t,"depthwiseFilter","separableConv2d"),l=C(e,"pointwiseFilter","separableConv2d"),c=a,p=!1;if(a.rank===3&&(p=!0,c=O(a,[1,a.shape[0],a.shape[1],a.shape[2]])),i==="NCHW")throw new Error("separableConv2d currently does not support dataFormat NCHW; only NHWC is supported");A(c.rank===4,()=>`Error in separableConv2d: input must be rank 4, but got rank ${c.rank}.`),A(u.rank===4,()=>`Error in separableConv2d: depthwise filter must be rank 4, but got rank ${u.rank}.`),A(l.rank===4,()=>`Error in separableConv2d: pointwise filter must be rank 4, but got rank ${u.rank}.`),A(l.shape[0]===1,()=>`Error in separableConv2d: the first dimension of pointwise filter must be 1, but got ${l.shape[0]}.`),A(l.shape[1]===1,()=>`Error in separableConv2d: the second dimension of pointwise filter must be 1, but got ${l.shape[1]}.`);let m=u.shape[2],f=u.shape[3];A(l.shape[2]===m*f,()=>`Error in separableConv2d: the third dimension of pointwise filter must be ${m*f}, but got ${l.shape[2]}.`);let d=Ks(c,u,n,o,i,s),g=ln(d,l,1,"valid",i);return p?O(g,[g.shape[1],g.shape[2],g.shape[3]]):g}var nd=N({separableConv2d_:dj});async function hj(r,t){let e=C(r,"x","setdiff1d"),n=C(t,"y","setdiff1d");A(e.dtype===n.dtype,()=>`x and y should have the same dtype, but got x (${e.dtype}) and y (${n.dtype}).`),A(e.rank===1,()=>`x should be 1D tensor, but got x (${e.shape}).`),A(n.rank===1,()=>`y should be 1D tensor, but got y (${n.shape}).`);let o=await e.data(),s=await n.data(),i=new Set(s),a=0;for(let c=0;c<o.length;c++)i.has(o[c])||a++;let u=new me([a],e.dtype),l=new me([a],"int32");for(let c=0,p=0;c<o.length;c++)i.has(o[c])||(u.values[p]=o[c],l.values[p]=c,p++);return[u.toTensor(),l.toTensor()]}var eC=hj;function gj(r){let e={x:C(r,"x","sign")};return _.runKernel(Ea,e)}var od=N({sign_:gj});function xj(r){let e={x:C(r,"x","sin","float32")};return _.runKernel($s,e)}var Yu=N({sin_:xj});function yj(r){let e={x:C(r,"x","sinh")};return _.runKernel(_a,e)}var Zu=N({sinh_:yj});function bj(r,t,e){let n=C(r,"x","slice1d");return A(n.rank===1,()=>`slice1d expects a rank-1 tensor, but got a rank-${n.rank} tensor`),Rt(n,[t],[e])}var sd=N({slice1d_:bj});function wj(r,t,e){let n=C(r,"x","slice2d");return A(n.rank===2,()=>`slice2d expects a rank-2 tensor, but got a rank-${n.rank} tensor`),Rt(n,t,e)}var Nx=N({slice2d_:wj});function vj(r,t,e){let n=C(r,"x","slice3d");return A(n.rank===3,()=>`slice3d expects a rank-3 tensor, but got a rank-${n.rank} tensor`),Rt(n,t,e)}var id=N({slice3d_:vj});function Cj(r,t,e){let n=C(r,"x","slice4d");return A(n.rank===4,()=>`slice4d expects a rank-4 tensor, but got a rank-${n.rank} tensor`),Rt(n,t,e)}var em=N({slice4d_:Cj});function Ij(r,t=-1){let e=C(r,"logits","softmax","float32");if(t===-1&&(t=e.rank-1),t!==e.rank-1)throw Error(`Softmax along a non-last dimension is not yet supported. Logits was rank ${e.rank} and dim was ${t}`);let n={logits:e},o={dim:t};return _.runKernel(Ps,n,o)}var Ja=N({softmax_:Ij});function Sj(r){A(r.dtype==="complex64",()=>`The dtype for tf.spectral.fft() must be complex64 but got ${r.dtype}.`);let t={input:r};return _.runKernel(Cp,t)}var Qa=N({fft_:Sj});function kj(r){A(r.dtype==="complex64",()=>`The dtype for tf.spectral.ifft() must be complex64 but got ${r.dtype}.`);let t={input:r};return _.runKernel(Ip,t)}var Oi=N({ifft_:kj});function Nj(r){let t=r.shape[r.shape.length-1],e=r.size/t,n;if(t<=2){let o=O(r,[e,t]);n=Oi(o)}else{let o=[e,2*(t-1)],s=O(Ul(r),[e,t]),i=O(Bu(r),[e,t]),a=sr(Rt(s,[0,1],[e,t-2]),1),u=P(sr(Rt(i,[0,1],[e,t-2]),1),pt(-1)),l=se([s,a],1),c=se([i,u],1),p=O(Mn(l,c),[o[0],o[1]]);n=Oi(p)}if(n=Ul(n),r.rank===3&&r.shape[0]!==0){let o=n,s=r.shape[0];n=O(n,[s,n.shape[0]/s,n.shape[1]]),o.dispose()}return n}var Ju=N({irfft_:Nj});function Tj(r,t,e=0){let o={x:C(r,"x","split")},s={numOrSizeSplits:t,axis:e};return _.runKernel(Ti,o,s)}var ur=N({split_:Tj});function _j(r,t){A(r.dtype==="float32",()=>`The dtype for rfft() must be real value but got ${r.dtype}`);let e=r.shape[r.shape.length-1],n=r.size/e,o;if(t!=null&&t<e){let d=r.shape.map(g=>0),h=r.shape.map(g=>g);h[r.shape.length-1]=t,o=Rt(r,d,h),e=t}else if(t!=null&&t>e){let d=r.shape.map(h=>h);d[r.shape.length-1]=t-e,o=se([r,ye(d)],r.shape.length-1),e=t}else o=r;let s=St(o),i=O(Mn(o,s),[n,e]),a=Qa(i),u=Math.floor(e/2)+1,l=Ul(a),c=Bu(a),p=ur(l,[u,e-u],l.shape.length-1),m=ur(c,[u,e-u],c.shape.length-1),f=o.shape.slice();return f[o.shape.length-1]=u,O(Mn(p[0],m[0]),f)}var tl=N({rfft_:_j});function Ej(r){let e={x:C(r,"x","sqrt","float32")};return _.runKernel(Fs,e)}var be=N({sqrt_:Ej});function Aj(r,t){let e=C(r,"a","squaredDifference"),n=C(t,"b","squaredDifference");[e,n]=jt(e,n),Lt(e.shape,n.shape);let o={a:e,b:n},s={};return _.runKernel(Ms,o,s)}var Qu=N({squaredDifference_:Aj});function Dj(r,t){let e=C(r,"x","squeeze");return O(e,Sv(e.shape,t).newShape)}var zr=N({squeeze_:Dj});function $j(r,t=0){let e=Pa(r,"tensors","stack","string_or_numeric");A(e.length>=1,()=>"Pass at least one tensor to tf.stack"),e.length>0&&A(t<=e[0].rank,()=>"Axis must be <= rank of the tensor");let n=e,o={axis:t};return _.runKernel(Ci,n,o)}var Je=N({stack_:$j});function Rj(r,t=0){let n={x:C(r,"x","step")},o={alpha:t};return _.runKernel(ao,n,o)}var Qs=N({step_:Rj});function Fj(r,t,e,n,o=0,s=0,i=0,a=0,u=0){let c={x:C(r,"x","stridedSlice","string_or_numeric")},p={begin:t,end:e,strides:n,beginMask:o,endMask:s,ellipsisMask:i,newAxisMask:a,shrinkAxisMask:u};return _.runKernel($a,c,p)}var ad=N({stridedSlice_:Fj});function Oj(r){let e={x:C(r,"x","tan","float32")};return _.runKernel(zs,e)}var ld=N({tan_:Oj});function Re(r,t){qn(r);let e=Or(r,t);if(e.length!==1)throw new Error("tensor1d() requires values to be a flat/TypedArray");return sn(r,null,e,t)}function Pi(r,t,e){if(qn(r),t!=null&&t.length!==2)throw new Error("tensor2d() requires shape to have two numbers");let n=Or(r,e);if(n.length!==2&&n.length!==1)throw new Error("tensor2d() requires values to be number[][] or flat/TypedArray");if(n.length===1&&t==null)throw new Error("tensor2d() requires shape to be provided when `values` are a flat/TypedArray");return sn(r,t,n,e)}function Pj(r,t,e){if(qn(r),t!=null&&t.length!==4)throw new Error("tensor4d() requires shape to have four numbers");let n=Or(r,e);if(n.length!==4&&n.length!==1)throw new Error("tensor4d() requires values to be number[][][][] or flat/TypedArray");if(n.length===1&&t==null)throw new Error("tensor4d() requires shape to be provided when `values` are a flat array");return sn(r,t,n,e)}function Mj(r,t,e){if(qn(r),t!=null&&t.length!==5)throw new Error("tensor5d() requires shape to have five numbers");let n=Or(r,e);if(n.length!==5&&n.length!==1)throw new Error("tensor5d() requires values to be number[][][][][] or flat/TypedArray");if(n.length===1&&t==null)throw new Error("tensor5d() requires shape to be provided when `values` are a flat array");return sn(r,t,n,e)}function Lj(r,t,e){if(qn(r),t!=null&&t.length!==6)throw new Error("tensor6d() requires shape to have six numbers");let n=Or(r,e);if(n.length!==6&&n.length!==1)throw new Error("tensor6d() requires values to be number[][][][][][] or flat/TypedArray");if(n.length===1&&t==null)throw new Error("tensor6d() requires shape to be provided when `values` are a flat array");return t=t||n,sn(r,t,n,e)}function zj(r,t=1,e=!0){let n=C(r,"x","topk");if(n.rank===0)throw new Error("topk() expects the input to be of rank 1 or higher");let o=n.shape[n.shape.length-1];if(t<0)throw new Error(`'k' passed to topk() must be >= 0 but got ${t}`);if(t>o)throw new Error(`'k' passed to topk() must be <= the last dimension (${o}) but got ${t}`);let s={x:n},i={k:t,sorted:e},[a,u]=_.runKernel(Ra,s,i);return{values:a,indices:u}}var ud=N({topk_:zj});function Bj(r,t=0,e=1,n,o){if(n!=null&&n==="bool")throw new Error("Unsupported data type $ { dtype }");let s=new tm(t,e,n,!0,o),i=It(r,n);for(let a=0;a<i.values.length;a++)i.values[a]=s.nextValue();return i.toTensor()}var tc=N({truncatedNormal_:Bj});function Vj(r,t=0){let e=C(r,"x","unique","string_or_numeric");A(e.rank>0,()=>"The input tensor must be at least 1D");let n={x:e},o={axis:t},[s,i]=_.runKernel(Lp,n,o);return{values:s,indices:i}}var rm=N({unique_:Vj});function Gj(r,t,e){let n=C(r,"x","unsortedSegmentSum"),o=C(t,"segmentIds","unsortedSegmentSum","int32");A(Ki(e),()=>"numSegments must be of dtype int");let s={x:n,segmentIds:o},i={numSegments:e};return _.runKernel(Ol,s,i)}var cd=N({unsortedSegmentSum_:Gj});function Wj(r,t=0){let e=C(r,"x","unstack","string_or_numeric");A(t>=-e.shape.length&&t<e.shape.length,()=>`Axis = ${t} is not in [-${e.shape.length}, ${e.shape.length})`);let n={value:e},o={axis:t};return _.runKernel(_i,n,o)}var vr=N({unstack_:Wj});function rC(r,t=!0,e,n){return _.makeVariable(r,t,e,n)}function Tx(r,t){let e=[];for(let s=0;s<t.length;s++)t[s]&&e.push(s);let n=It(r,"int32"),o=It([e.length,r.length],"int32");for(let s=0;s<e.length;s++){let i=n.indexToLoc(e[s]),a=s*r.length;o.values.set(i,a)}return o.toTensor()}async function Uj(r){let t=C(r,"condition","whereAsync","bool"),e=await t.data(),n=Tx(t.shape,e);return r!==t&&t.dispose(),n}var pd=Uj;async function Hj(r,t,e){let n=C(r,"tensor","boolMask"),o=C(t,"mask","boolMask","bool"),s=e==null?0:e,i=o.rank,a=n.shape;A(i>0,()=>"mask cannot be scalar"),Pe(a.slice(s,s+i),o.shape,"mask's shape must match the first K dimensions of tensor's shape,");let u=1;for(let h=s;h<s+i;h++)u*=a[h];let l=a.slice(0,s).concat([u],a.slice(s+i)),c=O(n,l),p=O(o,[-1]),m=await pd(p),f=zr(m,[1]),d=go(c,f,s);return r!==n&&n.dispose(),t!==o&&o.dispose(),f.dispose(),c.dispose(),p.dispose(),m.dispose(),d}var QTt=Hj;function qj(r,t="euclidean",e=null,n=!1){r=C(r,"x","norm");let o=U_(r,t,e),s=o.shape;if(n){let i=mr(e,r.shape);s=yo(o.shape,i)}return O(o,s)}function U_(r,t,e=null){if(r.rank===0)return Se(r);if(r.rank!==1&&e===null)return U_(O(r,[-1]),t,e);if(r.rank===1||typeof e=="number"||Array.isArray(e)&&e.length===1){if(t===1)return mt(Se(r),e);if(t===1/0)return $r(Se(r),e);if(t===-1/0)return Wl(Se(r),e);if(t==="euclidean"||t===2)return be(mt(Kr(Se(r),pt(2,"int32")),e));throw new Error(`Error in norm: invalid ord value: ${t}`)}if(Array.isArray(e)&&e.length===2){if(t===1)return $r(mt(Se(r),e[0]),e[1]-1);if(t===1/0)return $r(mt(Se(r),e[1]),e[0]);if(t===-1/0)return Wl(mt(Se(r),e[1]),e[0]);if(t==="fro"||t==="euclidean")return be(mt(Wt(r),e));throw new Error(`Error in norm: invalid ord value: ${t}`)}throw new Error(`Error in norm: invalid axis: ${e}`)}var nm=N({norm_:qj});function Kj(r,t,e,n,o=!0){let s=C(r,"v","movingAverage"),i=C(t,"x","movingAverage"),a=C(e,"decay","movingAverage");qv(s,i),A(Fn(s.shape,i.shape),()=>"Shape mismatch in v and x");let u=pt(1),l=lt(u,a),c=P(lt(i,s),l);if(o){A(n!=null,()=>"When using zeroDebias: true, step is required.");let p=C(n,"step","movingAverage");c=ct(c,lt(u,Kr(a,p)))}return J(s,c)}var I1t=N({movingAverage_:Kj});function jj(r,t,e){let n=C(r,"indices","scatterND","int32"),o=C(t,"updates","scatterND");gx(o,n,e);let s={indices:n,updates:o},i={shape:e};return _.runKernel(Na,s,i)}var H_=N({scatterND_:jj});function q_(r,t,e,n){if(r.dtype!=="int32")throw new Error(`tf.sparseToDense() expects the indices to be int32 type, but the dtype was ${r.dtype}.`);if(r.rank>2)throw new Error(`sparseIndices should be a scalar, vector, or matrix, but got shape ${r.shape}.`);let o=r.rank>0?r.shape[0]:1,s=r.rank>1?r.shape[1]:1;if(e.length!==s)throw new Error(`outputShape has incorrect number of elements:, ${e.length}, should be: ${s}.`);let i=t.size;if(!(t.rank===0||t.rank===1&&i===o))throw new Error(`sparseValues has incorrect shape ${t.shape}, should be [] or [${o}]`);if(t.dtype!==n.dtype)throw new Error("sparseValues.dtype must match defaultValues.dtype")}function Yj(r,t,e,n=0){let o=C(r,"sparseIndices","sparseToDense","int32"),s=C(t,"sparseValues","sparseToDense"),i=C(n,"defaultValue","sparseToDense",s.dtype);q_(o,s,e,i);let a={sparseIndices:o,sparseValues:s,defaultValue:i},u={outputShape:e};return _.runKernel(Fp,a,u)}var _x=N({sparseToDense_:Yj});function Zj(r,t){let e=C(t,"indices","gatherND","int32"),o={params:C(r,"x","gatherND","string_or_numeric"),indices:e};return _.runKernel(ca,o)}var K_=N({gatherND_:Zj});function j_(r,t){if(t==null)return r.shape.slice();if(Fn(r.shape,t))return t;if(r.shape.length===t.length){let e=[];for(let n=0;n<r.shape.length;n++)t[n]==null&&r.shape[n]!=null?e.push(r.shape[n]):e.push(t[n]);return e}return t}function Jj(r,t,e,n){let o=C(r,"x","dropout");if(A(o.dtype==="float32",()=>`x has to be a floating point tensor since it's going to be scaled, but got a ${o.dtype} tensor instead.`),A(t>=0&&t<1,()=>`rate must be a float in the range [0, 1), but got ${t}.`),t===0)return r instanceof Vt?o.clone():o;let s=j_(o,e),i=1-t,a=ct(Ys(J(Js(s,0,1,"float32",n),i)),i);return P(o,a)}var X_=N({dropout_:Jj});function Y_(r){return Math.floor(Math.pow(2,Math.ceil(Math.log(r)/Math.log(2))))}function Ex(r,t,e){let n=1-r%2,o=new Float32Array(r);for(let s=0;s<r;++s){let i=2*Math.PI*s/(r+n-1);o[s]=t-e*Math.cos(i)}return Re(o,"float32")}async function Qj(r,t,e=1){let n=C(r,"predictions","inTopK"),o=C(t,"targets","inTopK");A(n.rank>1,()=>`inTopK() expects the predictions to be of rank 2 or higher, but got ${n.rank}`),A(n.rank-1===o.rank,()=>`predictions rank should be 1 larger than targets rank, but got predictions rank ${n.rank} and targets rank ${o.rank}`),Pe(n.shape.slice(0,n.shape.length-1),o.shape,"predictions's shape should be align with the targets' shape, except the last dimension.");let s=n.shape[n.shape.length-1];A(e>0&&e<=s,()=>`'k' passed to inTopK() must be > 0 && <= the predictions last dimension (${s}), but got ${e}`);let i=await n.data(),a=await o.data(),[u,l]=[i.length/s,s],c=kv("bool",u);for(let p=0;p<u;p++){let m=p*l,f=i.subarray(m,m+l),d=[];for(let h=0;h<f.length;h++)d.push({value:f[h],index:h});d.sort((h,g)=>g.value-h.value),c[p]=0;for(let h=0;h<e;h++)if(d[h].index===a[p]){c[p]=1;break}}return r!==n&&n.dispose(),t!==o&&o.dispose(),Ar(c,o.shape,"bool")}var r_t=Qj;var wo={};Yt(wo,{conv2d:()=>Z_,depthwiseConv2d:()=>J_,matMul:()=>Q_});function t6(r,t,e,n,o,s="NHWC",i){let a=r;r.rank===3&&(a=O(r,[1,r.shape[0],r.shape[1],r.shape[2]]));let u=t;u.rank===3&&(u=O(t,[1,t.shape[0],t.shape[1],t.shape[2]])),A(a.rank===4,()=>`Error in conv2dDerFilter: input must be rank 4, but got shape ${a.shape}.`),A(u.rank===4,()=>`Error in conv2dDerFilter: dy must be rank 4, but got shape ${u.shape}.`),A(e.length===4,()=>`Error in conv2dDerFilter: filterShape must be length 4, but got ${e}.`);let l=s==="NHWC"?a.shape[3]:a.shape[1],c=s==="NHWC"?u.shape[3]:u.shape[1];A(l===e[2],()=>`Error in conv2dDerFilter: depth of input ${l}) must match input depth in filter (${e[2]}.`),A(c===e[3],()=>`Error in conv2dDerFilter: depth of dy (${c}) must match output depth for filter (${e[3]}).`),ke("conv2dDerFilter",o,i);let p={x:a,dy:u},m={strides:n,pad:o,dataFormat:s,dimRoundingMode:i,filterShape:e};return _.runKernel(fp,p,m)}var om=N({conv2DBackpropFilter_:t6});function ec(r,t,e){if(e==null||e==="linear")return r;if(e==="relu")return P(r,Qs(t));throw new Error(`Cannot compute gradient for fused activation ${e}.`)}function rc(r,t){let e=t,n=ge(r.shape,t.shape);return n.length>0&&(e=mt(e,n)),O(e,r.shape)}function nc(r,t,e,n){if(t==="linear")return r;if(t==="relu")return Tr(r);if(t==="elu")return js(r);if(t==="relu6")return qu(r);if(t==="prelu")return Ya(r,e);if(t==="leakyrelu")return Ha(r,n);if(t==="sigmoid")return Lr(r);throw new Error(`Unknown fused activation ${t}.`)}var oc=(r,t)=>!(r>0)||t==="linear";function e6({x:r,filter:t,strides:e,pad:n,dataFormat:o="NHWC",dilations:s=[1,1],dimRoundingMode:i,bias:a,activation:u="linear",preluActivationWeights:l,leakyreluAlpha:c}){if(u=u||"linear",oc(_.state.gradientDepth,u)===!1){let k=ln(r,t,e,n,o,s,i);return a!=null&&(k=J(k,a)),nc(k,u,l,c)}let p=C(r,"x","conv2d","float32"),m=C(t,"filter","conv2d","float32"),f=p,d=!1;p.rank===3&&(d=!0,f=O(p,[1,p.shape[0],p.shape[1],p.shape[2]])),A(f.rank===4,()=>`Error in fused conv2d: input must be rank 4, but got rank ${f.rank}.`),A(m.rank===4,()=>`Error in fused conv2d: filter must be rank 4, but got rank ${m.rank}.`),ke("fused conv2d",n,i),A(f.shape[3]===m.shape[2],()=>`Error in conv2d: depth of input (${f.shape[3]}) must match input depth for filter ${m.shape[2]}.`),A(Dr(e,s),()=>`Error in conv2D: Either strides or dilations must be 1. Got strides ${e} and dilations '${s}'`),A(o==="NHWC",()=>`Error in conv2d: got dataFormat of ${o} but only NHWC is currently supported.`);let h=Ou(f.shape,m.shape,e,s,n,i),g;a!=null&&(g=C(a,"bias","fused conv2d"),[g]=jt(g,p),Lt(h.outShape,g.shape));let y;l!=null&&(y=C(l,"prelu weights","fused conv2d"));let b=(k,E)=>{let[D,$,R,M]=E,W=ec(k,R,u);A(Xn(s),()=>`Error in gradient of fused conv2D: dilation rates greater than 1 are not yet supported in gradients. Got dilations '${s}'`);let U=Zp($.shape,W,D,e,n),q=om($,W,D.shape,e,n),K=[U,q];if(M!=null){let j=rc(M,W);K.push(j)}return K},w={x:f,filter:m,bias:g,preluActivationWeights:y},v={strides:e,pad:n,dataFormat:o,dilations:s,dimRoundingMode:i,activation:u,leakyreluAlpha:c};return a==null?un((E,D,$)=>{let R=_.runKernel(Di,w,v);return $([D,E,R]),d&&(R=O(R,[R.shape[1],R.shape[2],R.shape[3]])),{value:R,gradFunc:b}})(f,m):un((E,D,$,R)=>{let M=_.runKernel(Di,w,v);return R([D,E,M,$]),d&&(M=O(M,[M.shape[1],M.shape[2],M.shape[3]])),{value:M,gradFunc:b}})(f,m,g)}var Z_=N({fusedConv2d_:e6});function r6(r,t,e,n,o,s=[1,1],i){let a=r;r.rank===3&&(a=O(r,[1,r.shape[0],r.shape[1],r.shape[2]]));let u=t;u.rank===3&&(u=O(t,[1,t.shape[0],t.shape[1],t.shape[2]]));let l={x:a,dy:u},c={strides:n,pad:o,dimRoundingMode:i,dilations:s,filterShape:e};return _.runKernel(xp,l,c)}var Ax=N({depthwiseConv2dNativeBackpropFilter_:r6});function n6(r,t,e,n,o,s=[1,1],i){let a=t,u=!1;t.rank===3&&(u=!0,a=O(t,[1,t.shape[0],t.shape[1],t.shape[2]]));let l={dy:a,filter:e},c={strides:n,pad:o,dimRoundingMode:i,dilations:s,inputShape:r},p=_.runKernel(yp,l,c);return u?O(p,[p.shape[1],p.shape[2],p.shape[3]]):p}var Dx=N({depthwiseConv2dNativeBackpropInput_:n6});function o6({x:r,filter:t,strides:e,pad:n,dataFormat:o="NHWC",dilations:s=[1,1],dimRoundingMode:i,bias:a,activation:u="linear",preluActivationWeights:l,leakyreluAlpha:c}){if(oc(_.state.gradientDepth,u)===!1){let k=Ks(r,t,e,n,o,s,i);return a!=null&&(k=J(k,a)),nc(k,u,l,c)}let p=C(r,"x","depthwiseConv2d","float32"),m=C(t,"filter","depthwiseConv2d","float32"),f=p,d=!1;p.rank===3&&(d=!0,f=O(p,[1,p.shape[0],p.shape[1],p.shape[2]])),A(f.rank===4,()=>`Error in fused depthwiseConv2d: input must be rank 4, but got rank ${f.rank}.`),A(m.rank===4,()=>`Error in fused depthwiseConv2d: filter must be rank 4, but got rank ${m.rank}.`),A(f.shape[3]===m.shape[2],()=>`Error in fused depthwiseConv2d: number of input channels (${f.shape[3]}) must match the inChannels dimension in filter ${m.shape[2]}.`),s==null&&(s=[1,1]),A(Dr(e,s),()=>`Error in fused depthwiseConv2d: Either strides or dilations must be 1. Got strides ${e} and dilations '${s}'`),ke("fused depthwiseConv2d",n,i);let h=Ou(f.shape,m.shape,e,s,n,i,!0),g;a!=null&&(g=C(a,"bias","fused conv2d"),[g]=jt(g,p),Lt(h.outShape,g.shape));let y;l!=null&&(y=C(l,"prelu weights","fused depthwiseConv2d"));let b=(k,E)=>{A(Xn(s),()=>`Error in gradient of fused depthwiseConv2d: dilation rates greater than 1 are not yet supported. Got dilations '${s}'`);let[D,$,R,M]=E,W=ec(k,R,u),U=Dx($.shape,W,D,e,n,s,i),q=Ax($,W,D.shape,e,n,s,i);if(M!=null){let K=rc(g,W);return[U,q,K]}return[U,q]},w={x:f,filter:m,bias:g,preluActivationWeights:y},v={strides:e,pad:n,dataFormat:o,dilations:s,dimRoundingMode:i,activation:u,leakyreluAlpha:c};return a==null?un((E,D,$)=>{let R=_.runKernel($i,w,v);return $([D,E,R]),d&&(R=O(R,[R.shape[1],R.shape[2],R.shape[3]])),{value:R,gradFunc:b}})(f,m):un((E,D,$,R)=>{let M=_.runKernel($i,w,v);return R([D,E,M,$]),d&&(M=O(M,[M.shape[1],M.shape[2],M.shape[3]])),{value:M,gradFunc:b}})(f,m,g)}var J_=N({fusedDepthwiseConv2d_:o6});function s6({a:r,b:t,transposeA:e=!1,transposeB:n=!1,bias:o,activation:s="linear",preluActivationWeights:i,leakyreluAlpha:a}){if(oc(_.state.gradientDepth,s)===!1){let W=Mt(r,t,e,n);return o!=null&&(W=J(W,o)),nc(W,s,i,a)}let u=C(r,"a","fused matMul"),l=C(t,"b","fused matMul");[u,l]=jt(u,l);let c=e?u.shape[u.rank-2]:u.shape[u.rank-1],p=n?l.shape[l.rank-1]:l.shape[l.rank-2],m=e?u.shape[u.rank-1]:u.shape[u.rank-2],f=n?l.shape[l.rank-2]:l.shape[l.rank-1],d=u.shape.slice(0,-2),h=l.shape.slice(0,-2),g=re(d),y=re(h);A(c===p,()=>`Error in fused matMul: inner shapes (${c}) and (${p}) of Tensors with shapes ${u.shape} and ${l.shape} and transposeA=${e} and transposeB=${n} must match.`);let w=Lt(u.shape.slice(0,-2),l.shape.slice(0,-2)).concat([m,f]),v=e?O(u,[g,c,m]):O(u,[g,m,c]),k=n?O(l,[y,f,p]):O(l,[y,p,f]),E;o!=null&&(E=C(o,"bias","fused matMul"),[E]=jt(E,u),Lt(w,E.shape));let D;i!=null&&(D=C(i,"prelu weights","fused matMul"));let $=(W,U)=>{let[q,K,j,Q]=U,rt=ec(O(W,j.shape),j,s),X,ot;if(!e&&!n?(X=Mt(rt,K,!1,!0),ot=Mt(q,rt,!0,!1)):!e&&n?(X=Mt(rt,K,!1,!1),ot=Mt(rt,q,!0,!1)):e&&!n?(X=Mt(K,rt,!1,!0),ot=Mt(q,rt,!1,!1)):(X=Mt(K,rt,!0,!0),ot=Mt(rt,q,!0,!0)),o!=null){let st=rc(Q,rt);return[X,ot,st]}else return[X,ot]},R={a:v,b:k,bias:E,preluActivationWeights:D},M={transposeA:e,transposeB:n,activation:s,leakyreluAlpha:a};return o==null?un((U,q,K)=>{let j=_.runKernel(Ai,R,M);return K([U,q,j]),{value:O(j,w),gradFunc:$}})(v,k):un((U,q,K,j)=>{let Q=_.runKernel(Ai,R,M);return j([U,q,Q,K]),{value:O(Q,w),gradFunc:$}})(v,k,E)}var Q_=N({fusedMatMul_:s6});function i6(r){return Ex(r,.54,.46)}var tE=N({hammingWindow_:i6});function a6(r){return Ex(r,.5,.5)}var $x=N({hannWindow_:a6});function l6(r,t,e,n=!1,o=0){let s=0,i=[];for(;s+t<=r.size;)i.push(Rt(r,s,t)),s+=e;if(n)for(;s<r.size;){let a=s+t-r.size,u=se([Rt(r,s,t-a),Xs([a],o)]);i.push(u),s+=e}return i.length===0?Pi([],[0,t]):O(se(i),[i.length,t])}var Rx=N({frame_:l6});function u6(r,t,e,n,o=$x){n==null&&(n=Y_(t));let s=Rx(r,t,e),i=P(s,o(t));return tl(i,n)}var eE=N({stft_:u6});function c6(r,t,e,n,o="bilinear",s=0){let i=C(r,"image","cropAndResize"),a=C(t,"boxes","cropAndResize","float32"),u=C(e,"boxInd","cropAndResize","int32"),l=a.shape[0];A(i.rank===4,()=>`Error in cropAndResize: image must be rank 4,but got rank ${i.rank}.`),A(a.rank===2&&a.shape[1]===4,()=>`Error in cropAndResize: boxes must be have size [${l},4] but had shape ${a.shape}.`),A(u.rank===1&&u.shape[0]===l,()=>`Error in cropAndResize: boxInd must be have size [${l}] but had shape ${a.shape}.`),A(n.length===2,()=>`Error in cropAndResize: cropSize must be of length 2, but got length ${n.length}.`),A(n[0]>=1&&n[1]>=1,()=>`cropSize must be atleast [1,1], but was ${n}`),A(o==="bilinear"||o==="nearest",()=>`method must be bilinear or nearest, but was ${o}`);let c={image:i,boxes:a,boxInd:u},p={method:o,extrapolationValue:s,cropSize:n};return _.runKernel(oa,c,p)}var rE=N({cropAndResize_:c6});function p6(r){let t=C(r,"image","flipLeftRight","float32");A(t.rank===4,()=>`Error in flipLeftRight: image must be rank 4,but got rank ${t.rank}.`);let e={image:t};return _.runKernel(ua,e,{})}var nE=N({flipLeftRight_:p6});function m6(r){let t=C(r,"image","grayscaleToRGB"),e=t.rank-1,n=t.shape[e];A(t.rank>=2,()=>`Error in grayscaleToRGB: images must be at least rank 2, but got rank ${t.rank}.`),A(n===1,()=>`Error in grayscaleToRGB: last dimension of a grayscale image should be size 1, but got size ${n}.`);let o=new Array(t.rank);return o.fill(1,0,e),o[e]=3,kr(t,o)}var oE=N({grayscaleToRGB_:m6});function f6(r,t,e=0,n=.5){let o=C(r,"image","rotateWithOffset","float32");A(o.rank===4,()=>`Error in rotateWithOffset: image must be rank 4,but got rank ${o.rank}.`);let s={image:o},i={radians:t,fillValue:e,center:n};return _.runKernel(Oa,s,i)}var sE=N({rotateWithOffset_:f6});function vo(r,t,e,n,o,s){n==null&&(n=.5),o==null&&(o=Number.NEGATIVE_INFINITY),s==null&&(s=0);let i=r.shape[0];return e=Math.min(e,i),A(0<=n&&n<=1,()=>`iouThreshold must be in [0, 1], but was '${n}'`),A(r.rank===2,()=>`boxes must be a 2D tensor, but was of rank '${r.rank}'`),A(r.shape[1]===4,()=>`boxes must have 4 columns, but 2nd dimension was ${r.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:e,iouThreshold:n,scoreThreshold:o,softNmsSigma:s}}function d6(r,t,e,n=.5,o=Number.NEGATIVE_INFINITY){let s=C(r,"boxes","nonMaxSuppression","float32"),i=C(t,"scores","nonMaxSuppression","float32"),a=vo(s,i,e,n,o);e=a.maxOutputSize,n=a.iouThreshold,o=a.scoreThreshold;let u={maxOutputSize:e,iouThreshold:n,scoreThreshold:o};return _.runKernel(va,{boxes:s,scores:i},u)}var iE=N({nonMaxSuppression_:d6});function aE(r,t,e){let n=h6(r,t,e),o=n<0?-(n+1):n;r.splice(o,0,t)}function h6(r,t,e){return x6(r,t,e||g6)}function g6(r,t){return r>t?1:r<t?-1:0}function x6(r,t,e){let n=0,o=r.length,s=0,i=!1;for(;n<o;){s=n+(o-n>>>1);let a=e(t,r[s]);a>0?n=s+1:(o=s,i=!a)}return i?n:-n-1}function Fx(r,t,e,n,o){return nC(r,t,e,n,o,0)}function Ox(r,t,e,n,o,s){return nC(r,t,e,n,o,0,!1,s,!0)}function Px(r,t,e,n,o,s){return nC(r,t,e,n,o,s,!0)}function nC(r,t,e,n,o,s,i=!1,a=!1,u=!1){let l=[];for(let g=0;g<t.length;g++)t[g]>o&&l.push({score:t[g],boxIndex:g,suppressBeginIndex:0});l.sort(lE);let c=s>0?-.5/s:0,p=[],m=[];for(;p.length<e&&l.length>0;){let g=l.pop(),{score:y,boxIndex:b,suppressBeginIndex:w}=g;if(y<o)break;let v=!1;for(let k=p.length-1;k>=w;--k){let E=y6(r,b,p[k]);if(E>=n){v=!0;break}if(g.score=g.score*b6(n,c,E),g.score<=o)break}g.suppressBeginIndex=p.length,v||(g.score===y?(p.push(b),m.push(g.score)):g.score>o&&aE(l,g,lE))}let f=p.length,d=e-f;a&&d>0&&(p.push(...new Array(d).fill(0)),m.push(...new Array(d).fill(0)));let h={selectedIndices:p};return i&&(h.selectedScores=m),u&&(h.validOutputs=f),h}function y6(r,t,e){let n=r.subarray(t*4,t*4+4),o=r.subarray(e*4,e*4+4),s=Math.min(n[0],n[2]),i=Math.min(n[1],n[3]),a=Math.max(n[0],n[2]),u=Math.max(n[1],n[3]),l=Math.min(o[0],o[2]),c=Math.min(o[1],o[3]),p=Math.max(o[0],o[2]),m=Math.max(o[1],o[3]),f=(a-s)*(u-i),d=(p-l)*(m-c);if(f<=0||d<=0)return 0;let h=Math.max(s,l),g=Math.max(i,c),y=Math.min(a,p),b=Math.min(u,m),w=Math.max(y-h,0)*Math.max(b-g,0);return w/(f+d-w)}function b6(r,t,e){let n=Math.exp(t*e*e);return e<=r?n:0}function lE(r,t){return r.score-t.score||r.score===t.score&&t.boxIndex-r.boxIndex}async function w6(r,t,e,n=.5,o=Number.NEGATIVE_INFINITY){let s=C(r,"boxes","nonMaxSuppressionAsync"),i=C(t,"scores","nonMaxSuppressionAsync"),a=vo(s,i,e,n,o);e=a.maxOutputSize,n=a.iouThreshold,o=a.scoreThreshold;let u=await Promise.all([s.data(),i.data()]),l=u[0],c=u[1],{selectedIndices:p}=Fx(l,c,e,n,o);return s!==r&&s.dispose(),i!==t&&i.dispose(),Re(p,"int32")}var uE=w6;function v6(r,t,e,n=.5,o=Number.NEGATIVE_INFINITY,s=0){let i=C(r,"boxes","nonMaxSuppression"),a=C(t,"scores","nonMaxSuppression"),u=vo(i,a,e,n,o,s);e=u.maxOutputSize,n=u.iouThreshold,o=u.scoreThreshold,s=u.softNmsSigma;let l={boxes:i,scores:a},c={maxOutputSize:e,iouThreshold:n,scoreThreshold:o,softNmsSigma:s},p=_.runKernel(Ia,l,c);return{selectedIndices:p[0],selectedScores:p[1]}}var cE=N({nonMaxSuppressionWithScore_:v6});async function C6(r,t,e,n=.5,o=Number.NEGATIVE_INFINITY,s=0){let i=C(r,"boxes","nonMaxSuppressionAsync"),a=C(t,"scores","nonMaxSuppressionAsync"),u=vo(i,a,e,n,o,s);e=u.maxOutputSize,n=u.iouThreshold,o=u.scoreThreshold,s=u.softNmsSigma;let l=await Promise.all([i.data(),a.data()]),c=l[0],p=l[1],{selectedIndices:m,selectedScores:f}=Px(c,p,e,n,o,s);return i!==r&&i.dispose(),a!==t&&a.dispose(),{selectedIndices:Re(m,"int32"),selectedScores:Re(f)}}var pE=C6;function I6(r,t,e,n=.5,o=Number.NEGATIVE_INFINITY,s=!1){let i=C(r,"boxes","nonMaxSuppression"),a=C(t,"scores","nonMaxSuppression"),u=vo(i,a,e,n,o,null),l=u.maxOutputSize,c=u.iouThreshold,p=u.scoreThreshold,m={boxes:i,scores:a},f={maxOutputSize:l,iouThreshold:c,scoreThreshold:p,padToMaxOutputSize:s},d=_.runKernel(Ca,m,f);return{selectedIndices:d[0],validOutputs:d[1]}}var mE=N({nonMaxSuppressionPadded_:I6});async function S6(r,t,e,n=.5,o=Number.NEGATIVE_INFINITY,s=!1){let i=C(r,"boxes","nonMaxSuppressionAsync"),a=C(t,"scores","nonMaxSuppressionAsync"),u=vo(i,a,e,n,o,null),l=u.maxOutputSize,c=u.iouThreshold,p=u.scoreThreshold,[m,f]=await Promise.all([i.data(),a.data()]),{selectedIndices:d,validOutputs:h}=Ox(m,f,l,c,p,s);return i!==r&&i.dispose(),a!==t&&a.dispose(),{selectedIndices:Re(d,"int32"),validOutputs:pt(h,"int32")}}var fE=S6;function k6(r,t,e=!1,n=!1){let o=C(r,"images","resizeBilinear");A(o.rank===3||o.rank===4,()=>`Error in resizeBilinear: x must be rank 3 or 4, but got rank ${o.rank}.`),A(t.length===2,()=>`Error in resizeBilinear: new shape must 2D, but got shape ${t}.`),A(n===!1||e===!1,()=>"Error in resizeBilinear: If halfPixelCenters is true, alignCorners must be false.");let s=o,i=!1;o.rank===3&&(i=!0,s=O(o,[1,o.shape[0],o.shape[1],o.shape[2]]));let[]=t,a={images:s},u={alignCorners:e,halfPixelCenters:n,size:t},l=_.runKernel(Ts,a,u);return i?O(l,[l.shape[1],l.shape[2],l.shape[3]]):l}var Mx=N({resizeBilinear_:k6});function N6(r,t,e=!1,n=!1){let o=C(r,"images","resizeNearestNeighbor");A(o.rank===3||o.rank===4,()=>`Error in resizeNearestNeighbor: x must be rank 3 or 4, but got rank ${o.rank}.`),A(t.length===2,()=>`Error in resizeNearestNeighbor: new shape must 2D, but got shape ${t}.`),A(o.dtype==="float32"||o.dtype==="int32",()=>"`images` must have `int32` or `float32` as dtype"),A(n===!1||e===!1,()=>"Error in resizeNearestNeighbor: If halfPixelCenters is true, alignCorners must be false.");let s=o,i=!1;o.rank===3&&(i=!0,s=O(o,[1,o.shape[0],o.shape[1],o.shape[2]]));let[]=t,a={images:s},u={alignCorners:e,halfPixelCenters:n,size:t},l=_.runKernel(Al,a,u);return i?O(l,[l.shape[1],l.shape[2],l.shape[3]]):l}var Lx=N({resizeNearestNeighbor_:N6});function T6(r,t="binary",e=!1,n=.5){let o=C(r,"image","threshold"),s=.2989,i=.587,a=.114,u=o.shape[0]*o.shape[1],l=P(Re([n]),255),c,p,m,f;if(A(o.rank===3,()=>`Error in threshold: image must be rank 3,but got rank ${o.rank}.`),A(o.shape[2]===3||o.shape[2]===1,()=>`Error in threshold: image color channel must be equal to 3 or 1but got ${o.shape[2]}.`),A(o.dtype==="int32"||o.dtype==="float32",()=>`Error in dtype: image dtype must be int32 or float32,but got dtype ${o.dtype}.`),A(t==="otsu"||t==="binary",()=>`Method must be binary or otsu, but was ${t}`),o.shape[2]===3){[c,p,m]=ur(o,[1,1,1],-1);let g=P(c,s),y=P(p,i),b=P(m,a);f=J(J(g,y),b)}else f=r;if(t==="otsu"){let g=Bf(Z(Ku(f),"int32"),Ar([]),256);l=_6(g,u)}let d=e?Nn(f,l):Ge(f,l);return Z(P(d,255),"int32")}function _6(r,t){let e=Re([-1]),n=Re([0]),o=Re([0]),s,i,a,u,l,c;for(let p=0;p<r.size-1;p++){s=Rt(r,0,p+1),i=Rt(r,p+1),l=ct(mt(s),t),c=ct(mt(i),t);let m=mt(P(s,Za(0,s.size)));a=ct(m,mt(s));let f=Xs(i.shape,s.size),d=J(Za(0,i.size),f),h=P(i,d);u=ct(mt(h),mt(i));let g=lt(a,u),y=lt(a,u),b=P(l,c);o=P(P(b,g),y);let w=Ge(o,n);n=Ne(w,o,n),e=Ne(w,Re([p]),e)}return e}var dE=N({threshold_:T6});function E6(r,t,e="nearest",n="constant",o=0,s){let i=C(r,"image","transform","float32"),a=C(t,"transforms","transform","float32");A(i.rank===4,()=>`Error in transform: image must be rank 4,but got rank ${i.rank}.`),A(a.rank===2&&(a.shape[0]===i.shape[0]||a.shape[0]===1)&&a.shape[1]===8,()=>"Error in transform: Input transform should be batch x 8 or 1 x 8"),A(s==null||s.length===2,()=>`Error in transform: outputShape must be [height, width] or null, but got ${s}.`);let u={image:i,transforms:a},l={interpolation:e,fillMode:n,fillValue:o,outputShape:s};return _.runKernel(Fa,u,l)}var hE=N({transform_:E6});function A6(r,t,e){A(t%1===0,()=>`bandPart(): numLower must be an integer, got ${t}.`),A(e%1===0,()=>`bandPart(): numUpper must be an integer, got ${e}.`);let n=C(r,"a","bandPart");A(n.rank>=2,()=>`bandPart(): Rank must be at least 2, got ${n.rank}.`);let o=n.shape,[s,i]=n.shape.slice(-2);if(!(t<=s))throw new Error(`bandPart(): numLower (${t}) must not be greater than the number of rows (${s}).`);if(!(e<=i))throw new Error(`bandPart(): numUpper (${e}) must not be greater than the number of columns (${i}).`);t<0&&(t=s),e<0&&(e=i);let a=O(Za(0,s,1,"int32"),[-1,1]),u=Za(0,i,1,"int32"),l=lt(a,u),c=Nr(Nn(l,pt(+t,"int32")),kn(l,pt(-e,"int32"))),p=ye([s,i],n.dtype);return O(Je(vr(O(n,[-1,s,i])).map(m=>Ne(c,m,p))),o)}var gE=N({bandPart_:A6});function D6(r){let t;if(Array.isArray(r)){t=!1,A(r!=null&&r.length>0,()=>"Gram-Schmidt process: input must not be null, undefined, or empty");let o=r[0].shape[0];for(let s=1;s<r.length;++s)A(r[s].shape[0]===o,()=>`Gram-Schmidt: Non-unique lengths found in the input vectors: (${r[s].shape[0]} vs. ${o})`)}else t=!0,r=ur(r,r.shape[0],0).map(o=>zr(o,[0]));A(r.length<=r[0].shape[0],()=>`Gram-Schmidt: Number of vectors (${r.length}) exceeds number of dimensions (${r[0].shape[0]}).`);let e=[],n=r;for(let o=0;o<r.length;++o)e.push(_.tidy(()=>{let s=n[o];if(o>0)for(let i=0;i<o;++i){let a=P(mt(P(e[i],s)),e[i]);s=lt(s,a)}return ct(s,nm(s,"euclidean"))}));return t?Je(e,0):e}var xE=N({gramSchmidt_:D6});function $6(r,t=!1){if(A(r.rank>=2,()=>`qr() requires input tensor to have a rank >= 2, but got rank ${r.rank}`),r.rank===2)return yE(r,t);{let e=r.shape.slice(0,r.shape.length-2).reduce((u,l)=>u*l),n=vr(O(r,[e,r.shape[r.shape.length-2],r.shape[r.shape.length-1]]),0),o=[],s=[];n.forEach(u=>{let[l,c]=yE(u,t);o.push(l),s.push(c)});let i=O(Je(o,0),r.shape),a=O(Je(s,0),r.shape);return[i,a]}}function yE(r,t=!1){return _.tidy(()=>{A(r.shape.length===2,()=>`qr2d() requires a 2D Tensor, but got a ${r.shape.length}D Tensor.`);let e=r.shape[0],n=r.shape[1],o=Jp(e),s=In(r),i=Pi([[1]],[1,1]),a=In(i),u=e>=n?n:e;for(let l=0;l<u;++l){let c=s,p=a,m=o;[a,s,o]=_.tidy(()=>{let f=Rt(s,[l,l],[e-l,1]),d=nm(f),h=Rt(s,[l,l],[1,1]),g=Ne(Ge(h,0),Pi([[-1]]),Pi([[1]])),y=lt(h,P(g,d)),b=ct(f,y);b.shape[0]===1?a=In(i):a=se([i,Rt(b,[1,0],[b.shape[0]-1,b.shape[1]])],0);let w=Zt(ct(Mt(g,y),d)),v=Rt(s,[l,0],[e-l,n]),k=P(w,a),E=Gt(a);if(l===0)s=lt(v,Mt(k,Mt(E,v)));else{let R=lt(v,Mt(k,Mt(E,v)));s=se([Rt(s,[0,0],[l,n]),R],0)}let D=Gt(k),$=Rt(o,[0,l],[e,o.shape[1]-l]);if(l===0)o=lt($,Mt(Mt($,a),D));else{let R=lt($,Mt(Mt($,a),D));o=se([Rt(o,[0,0],[e,l]),R],1)}return[a,s,o]}),_t([c,p,m])}return!t&&e>n&&(o=Rt(o,[0,0],[e,n]),s=Rt(s,[0,0],[n,n])),[o,s]})}var bE=N({qr_:$6});var Qe;(function(r){r[r.NONE=0]="NONE",r[r.MEAN=1]="MEAN",r[r.SUM=2]="SUM",r[r.SUM_BY_NONZERO_WEIGHTS=3]="SUM_BY_NONZERO_WEIGHTS"})(Qe||(Qe={}));function R6(r,t,e=Qe.SUM_BY_NONZERO_WEIGHTS){let n=C(r,"losses","computeWeightedLoss"),o=null;t!=null&&(o=C(t,"weights","computeWeightedLoss"));let s=o==null?n:P(n,o);if(e===Qe.NONE)return s;if(e===Qe.SUM)return mt(s);if(e===Qe.MEAN){if(o==null)return xe(s);{let i=n.size/o.size,a=ct(mt(s),mt(o));return i>1?ct(a,pt(i)):a}}if(e===Qe.SUM_BY_NONZERO_WEIGHTS){if(o==null)return ct(mt(s),pt(n.size));{let i=P(o,lr(n.shape)),a=Z(mt(bo(i,pt(0))),"float32");return ct(mt(s),a)}}throw Error(`Unknown reduction: ${e}`)}var Br=N({computeWeightedLoss_:R6});function F6(r,t,e,n=Qe.SUM_BY_NONZERO_WEIGHTS){let o=C(r,"labels","absoluteDifference"),s=C(t,"predictions","absoluteDifference"),i=null;e!=null&&(i=C(e,"weights","absoluteDifference")),Pe(o.shape,s.shape,"Error in absoluteDifference: ");let a=Se(lt(o,s));return Br(a,i,n)}var wE=N({absoluteDifference_:F6});function O6(r,t,e,n,o=Qe.SUM_BY_NONZERO_WEIGHTS){let s=C(r,"labels","cosineDistance"),i=C(t,"predictions","cosineDistance"),a=null;n!=null&&(a=C(n,"weights","cosineDistance")),Pe(s.shape,i.shape,"Error in cosineDistance: ");let u=pt(1),l=lt(u,mt(P(s,i),e,!0));return Br(l,a,o)}var vE=N({cosineDistance_:O6});function P6(r,t,e,n=Qe.SUM_BY_NONZERO_WEIGHTS){let o=C(r,"labels","hingeLoss"),s=C(t,"predictions","hingeLoss"),i=null;e!=null&&(i=C(e,"weights","hingeLoss")),Pe(o.shape,s.shape,"Error in hingeLoss: ");let a=pt(1);o=lt(P(pt(2),o),a);let u=Tr(lt(a,P(o,s)));return Br(u,i,n)}var CE=N({hingeLoss_:P6});function M6(r,t,e,n=1,o=Qe.SUM_BY_NONZERO_WEIGHTS){let s=C(r,"labels","huberLoss"),i=C(t,"predictions","huberLoss"),a=null;e!=null&&(a=C(e,"weights","huberLoss")),Pe(s.shape,i.shape,"Error in huberLoss: ");let u=pt(n),l=Se(lt(i,s)),c=Zs(l,u),p=lt(l,c),m=J(P(pt(.5),Wt(c)),P(u,p));return Br(m,a,o)}var IE=N({huberLoss_:M6});function L6(r,t,e,n=1e-7,o=Qe.SUM_BY_NONZERO_WEIGHTS){let s=C(r,"labels","logLoss"),i=C(t,"predictions","logLoss"),a=null;e!=null&&(a=C(e,"weights","logLoss")),Pe(s.shape,i.shape,"Error in logLoss: ");let u=pt(1),l=pt(n),c=Zt(P(s,wr(J(i,l)))),p=P(lt(u,s),wr(J(lt(u,i),l))),m=lt(c,p);return Br(m,a,o)}var SE=N({logLoss_:L6});function z6(r,t,e,n=Qe.SUM_BY_NONZERO_WEIGHTS){let o=C(r,"labels","meanSquaredError"),s=C(t,"predictions","meanSquaredError"),i=null;e!=null&&(i=C(e,"weights","meanSquaredError")),Pe(o.shape,s.shape,"Error in meanSquaredError: ");let a=Qu(o,s);return Br(a,i,n)}var kE=N({meanSquaredError_:z6});function B6(r,t){let e=C(r,"labels","sigmoidCrossEntropyWithLogits"),n=C(t,"logits","sigmoidCrossEntropyWithLogits");Pe(e.shape,n.shape,"Error in sigmoidCrossEntropyWithLogits: ");let o=Tr(n),s=P(n,e),i=qa(Ze(Zt(Se(n))));return J(lt(o,s),i)}function V6(r,t,e,n=0,o=Qe.SUM_BY_NONZERO_WEIGHTS){let s=C(r,"multiClassLabels","sigmoidCrossEntropy"),i=C(t,"logits","sigmoidCrossEntropy"),a=null;if(e!=null&&(a=C(e,"weights","sigmoidCrossEntropy")),Pe(s.shape,i.shape,"Error in sigmoidCrossEntropy: "),n>0){let l=pt(n),c=pt(1),p=pt(.5);s=J(P(s,lt(c,l)),P(p,l))}let u=B6(s,i);return Br(u,a,o)}var NE=N({sigmoidCrossEntropy_:V6});function G6(r,t,e=-1){if(e===-1&&(e=t.rank-1),e!==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 ${e}`);return un((o,s,i)=>{let u=Jf(s,[e],!0),l=lt(Z(s,"float32"),u);i([o,l]);let c=Zt(P(l,o));return{value:mt(c,[e]),gradFunc:(f,d)=>{let[h,g]=d,y=yo(f.shape,[e]);return[P(O(f,y),lt(Z(h,"float32"),Ze(g))),P(O(f,y),lt(Ze(g),Z(h,"float32")))]}}})(r,t)}function W6(r,t,e,n=0,o=Qe.SUM_BY_NONZERO_WEIGHTS){let s=C(r,"onehotLabels","softmaxCrossEntropy"),i=C(t,"logits","softmaxCrossEntropy"),a=null;if(e!=null&&(a=C(e,"weights","softmaxCrossEntropy")),Pe(s.shape,i.shape,"Error in softmaxCrossEntropy: "),n>0){let l=pt(n),c=pt(1),p=pt(s.shape[1]);s=J(P(s,lt(c,l)),ct(l,p))}let u=G6(s,i);return Br(u,a,o)}var TE=N({softmaxCrossEntropy_:W6});function U6(r,t,e,n){let o=C(r,"indices","sparseFillEmptyRows","int32"),s=C(t,"values","sparseFillEmptyRows"),i=C(e,"denseShape","sparseFillEmptyRows","int32"),a=C(n,"defaultValue","sparseFillEmptyRows",s.dtype);if(o.rank!==2)throw new Error(`Indices should be Tensor2D but received shape
${o.shape}`);if(s.rank!==1)throw new Error(`Values should be Tensor1D but received shape ${s.shape}`);if(i.rank!==1)throw new Error(`Dense shape should be Tensor1D but received shape ${i.shape}`);if(a.rank!==0)throw new Error(`Default value should be a scalar but received shape ${a.shape}`);let u={indices:o,values:s,denseShape:i,defaultValue:a},l=_.runKernel(Dl,u);return{outputIndices:l[0],outputValues:l[1],emptyRowIndicator:l[2],reverseIndexMap:l[3]}}var _E=N({sparseFillEmptyRows_:U6});function H6(r,t,e){let n=C(r,"inputIndices","sparseReshape","int32"),o=C(t,"inputShape","sparseReshape","int32"),s=C(e,"newShape","sparseReshape","int32");if(n.rank!==2)throw new Error(`Input indices should be Tensor2D but received shape
${n.shape}`);if(o.rank!==1)throw new Error(`Input shape should be Tensor1D but received shape ${o.shape}`);if(s.rank!==1)throw new Error(`New shape should be Tensor1D but received shape ${s.shape}`);let i={inputIndices:n,inputShape:o,newShape:s},a=_.runKernel(Da,i);return{outputIndices:a[0],outputShape:a[1]}}var EE=N({sparseReshape_:H6});function q6(r,t,e){let n=C(r,"data","sparseSegmentMean"),o=C(t,"indices","sparseSegmentMean","int32"),s=C(e,"segmentIds","sparseSegmentMean","int32");if(n.rank<1)throw new Error("Data should be at least 1 dimensional but received scalar");if(o.rank!==1)throw new Error(`Indices should be Tensor1D but received shape
${o.shape}`);if(s.rank!==1)throw new Error(`Segment ids should be Tensor1D but received shape
${s.shape}`);let i={data:n,indices:o,segmentIds:s};return _.runKernel($l,i)}var AE=N({sparseSegmentMean_:q6});function K6(r,t,e){let n=C(r,"data","sparseSegmentSum"),o=C(t,"indices","sparseSegmentSum","int32"),s=C(e,"segmentIds","sparseSegmentSum","int32");if(n.rank<1)throw new Error("Data should be at least 1 dimensional but received scalar");if(o.rank!==1)throw new Error(`Indices should be Tensor1D but received shape
${o.shape}`);if(s.rank!==1)throw new Error(`Segment ids should be Tensor1D but received shape
${s.shape}`);let i={data:n,indices:o,segmentIds:s};return _.runKernel(Rl,i)}var DE=N({sparseSegmentSum_:K6});function j6(r,t,e,n,o,s,i,a){let u=C(r,"data","stringNGrams","string");if(u.dtype!=="string")throw new Error("Data must be of datatype string");if(u.shape.length!==1)throw new Error(`Data must be a vector, saw: ${u.shape}`);let l=C(t,"dataSplits","stringNGrams");if(l.dtype!=="int32")throw new Error("Data splits must be of datatype int32");let c={separator:e,nGramWidths:n,leftPad:o,rightPad:s,padWidth:i,preserveShortSequences:a},p={data:u,dataSplits:l},m=_.runKernel(Op,p,c);return{nGrams:m[0],nGramsSplits:m[1]}}var $E=N({stringNGrams_:j6});function X6(r,t,e=!0){let n=C(r,"input","stringSplit","string"),o=C(t,"delimiter","stringSplit","string");if(n.rank!==1)throw new Error(`Input should be Tensor1D but received shape ${n.shape}`);if(o.rank!==0)throw new Error(`Delimiter should be a scalar but received shape ${o.shape}`);let s={skipEmpty:e},i={input:n,delimiter:o},a=_.runKernel(Pp,i,s);return{indices:a[0],values:a[1],shape:a[2]}}var RE=N({stringSplit_:X6});function Y6(r,t){let e=C(r,"input","stringToHashBucketFast","string"),n={numBuckets:t};if(t<=0)throw new Error("Number of buckets must be at least 1");let o={input:e};return _.runKernel(Mp,o,n)}var FE=N({stringToHashBucketFast_:Y6});var eFt={fft:Qa,ifft:Oi,rfft:tl,irfft:Ju},iFt={hammingWindow:tE,hannWindow:$x,frame:Rx,stft:eE},pn={flipLeftRight:nE,grayscaleToRGB:oE,resizeNearestNeighbor:Lx,resizeBilinear:Mx,rotateWithOffset:sE,cropAndResize:rE,nonMaxSuppression:iE,nonMaxSuppressionAsync:uE,nonMaxSuppressionWithScore:cE,nonMaxSuppressionWithScoreAsync:pE,nonMaxSuppressionPadded:mE,nonMaxSuppressionPaddedAsync:fE,threshold:dE,transform:hE},OE={bandPart:gE,gramSchmidt:xE,qr:bE},RFt={absoluteDifference:wE,computeWeightedLoss:Br,cosineDistance:vE,hingeLoss:CE,huberLoss:IE,logLoss:SE,meanSquaredError:kE,sigmoidCrossEntropy:NE,softmaxCrossEntropy:TE},md={sparseFillEmptyRows:_E,sparseReshape:EE,sparseSegmentMean:AE,sparseSegmentSum:DE},zx={stringNGrams:$E,stringSplit:RE,stringToHashBucketFast:FE};var jr=class extends xx{minimize(t,e=!1,n){let{value:o,grads:s}=this.computeGradients(t,n);if(n!=null){let i=n.map(a=>({name:a.name,tensor:s[a.name]}));this.applyGradients(i)}else this.applyGradients(s);return _t(s),e?o:(o.dispose(),null)}get iterations(){return this.iterations_==null&&(this.iterations_=0),this.iterations_}incrementIterations(){this.iterations_=this.iterations+1}computeGradients(t,e){return vx(t,e)}dispose(){this.iterations_!=null&&_t(this.iterations_)}async saveIterations(){return this.iterations_==null&&(this.iterations_=0),{name:"iter",tensor:pt(this.iterations_,"int32")}}async getWeights(){throw new Error("getWeights() is not implemented for this optimizer yet.")}async setWeights(t){throw new Error(`setWeights() is not implemented for this optimizer class ${this.getClassName()}`)}async extractIterations(t){return this.iterations_=(await t[0].tensor.data())[0],t.slice(1)}};Object.defineProperty(jr,Symbol.hasInstance,{value:r=>r.minimize!=null&&r.computeGradients!=null&&r.applyGradients!=null});var sc=class extends jr{constructor(t,e,n=null){super();this.learningRate=t,this.rho=e,this.epsilon=n,this.accumulatedGrads=[],this.accumulatedUpdates=[],n==null&&(this.epsilon=_.backend.epsilon())}applyGradients(t){(Array.isArray(t)?t.map(n=>n.name):Object.keys(t)).forEach((n,o)=>{let s=_.registeredVariables[n],i=!1;this.accumulatedGrads[o]==null&&(this.accumulatedGrads[o]={originalName:`${n}/accum_grad`,variable:B(()=>St(s).variable(i))}),this.accumulatedUpdates[o]==null&&(this.accumulatedUpdates[o]={originalName:`${n}/accum_var`,variable:B(()=>St(s).variable(i))});let a=Array.isArray(t)?t[o].tensor:t[n];if(a==null)return;let u=this.accumulatedGrads[o].variable,l=this.accumulatedUpdates[o].variable;B(()=>{let c=J(P(u,this.rho),P(Wt(a),1-this.rho)),p=P(ct(be(J(l,this.epsilon)),be(J(u,this.epsilon))),a),m=J(P(l,this.rho),P(Wt(p),1-this.rho));u.assign(c),l.assign(m);let f=J(P(p,-this.learningRate),s);s.assign(f)})}),this.incrementIterations()}dispose(){this.accumulatedUpdates!=null&&(_t(this.accumulatedGrads.map(t=>t.variable)),_t(this.accumulatedUpdates.map(t=>t.variable)))}async getWeights(){let t=[...this.accumulatedGrads,...this.accumulatedUpdates];return[await this.saveIterations()].concat(t.map(e=>({name:e.originalName,tensor:e.variable})))}async setWeights(t){t=await this.extractIterations(t);let e=t.length/2,n=!1;this.accumulatedGrads=t.slice(0,e).map(o=>({originalName:o.name,variable:o.tensor.variable(n)})),this.accumulatedUpdates=t.slice(e,e*2).map(o=>({originalName:o.name,variable:o.tensor.variable(n)}))}getConfig(){return{learningRate:this.learningRate,rho:this.rho,epsilon:this.epsilon}}static fromConfig(t,e){return new t(e.learningRate,e.rho,e.epsilon)}};sc.className="Adadelta";Sn(sc);var ic=class extends jr{constructor(t,e=.1){super();this.learningRate=t,this.initialAccumulatorValue=e,this.accumulatedGrads=[]}applyGradients(t){(Array.isArray(t)?t.map(n=>n.name):Object.keys(t)).forEach((n,o)=>{let s=_.registeredVariables[n];this.accumulatedGrads[o]==null&&(this.accumulatedGrads[o]={originalName:`${n}/accumulator`,variable:B(()=>Xs(s.shape,this.initialAccumulatorValue).variable(!1))});let i=Array.isArray(t)?t[o].tensor:t[n];if(i==null)return;let a=this.accumulatedGrads[o].variable;B(()=>{let u=J(a,Wt(i));a.assign(u);let l=J(P(ct(i,be(J(u,_.backend.epsilon()))),-this.learningRate),s);s.assign(l)})}),this.incrementIterations()}dispose(){this.accumulatedGrads!=null&&_t(this.accumulatedGrads.map(t=>t.variable))}async getWeights(){return[await this.saveIterations()].concat(this.accumulatedGrads.map(t=>({name:t.originalName,tensor:t.variable})))}async setWeights(t){t=await this.extractIterations(t);let e=!1;this.accumulatedGrads=t.map(n=>({originalName:n.name,variable:n.tensor.variable(e)}))}getConfig(){return{learningRate:this.learningRate,initialAccumulatorValue:this.initialAccumulatorValue}}static fromConfig(t,e){return new t(e.learningRate,e.initialAccumulatorValue)}};ic.className="Adagrad";Sn(ic);var ac=class extends jr{constructor(t,e,n,o=null){super();this.learningRate=t,this.beta1=e,this.beta2=n,this.epsilon=o,this.accumulatedFirstMoment=[],this.accumulatedSecondMoment=[],B(()=>{this.accBeta1=pt(e).variable(),this.accBeta2=pt(n).variable()}),o==null&&(this.epsilon=_.backend.epsilon())}applyGradients(t){let e=Array.isArray(t)?t.map(n=>n.name):Object.keys(t);B(()=>{let n=lt(1,this.accBeta1),o=lt(1,this.accBeta2);e.forEach((s,i)=>{let a=_.registeredVariables[s],u=!1;this.accumulatedFirstMoment[i]==null&&(this.accumulatedFirstMoment[i]={originalName:`${s}/m`,variable:B(()=>St(a).variable(u))}),this.accumulatedSecondMoment[i]==null&&(this.accumulatedSecondMoment[i]={originalName:`${s}/v`,variable:B(()=>St(a).variable(u))});let l=Array.isArray(t)?t[i].tensor:t[s];if(l==null)return;let c=this.accumulatedFirstMoment[i].variable,p=this.accumulatedSecondMoment[i].variable,m=J(P(c,this.beta1),P(l,1-this.beta1)),f=J(P(p,this.beta2),P(Wt(l),1-this.beta2)),d=ct(m,n),h=ct(f,o);c.assign(m),p.assign(f);let g=J(P(ct(d,J(be(h),this.epsilon)),-this.learningRate),a);a.assign(g)}),this.accBeta1.assign(P(this.accBeta1,this.beta1)),this.accBeta2.assign(P(this.accBeta2,this.beta2))}),this.incrementIterations()}dispose(){this.accBeta1.dispose(),this.accBeta2.dispose(),this.accumulatedFirstMoment!=null&&_t(this.accumulatedFirstMoment.map(t=>t.variable)),this.accumulatedSecondMoment!=null&&_t(this.accumulatedSecondMoment.map(t=>t.variable))}async getWeights(){let t=[...this.accumulatedFirstMoment,...this.accumulatedSecondMoment];return[await this.saveIterations()].concat(t.map(e=>({name:e.originalName,tensor:e.variable})))}async setWeights(t){t=await this.extractIterations(t),B(()=>{this.accBeta1.assign(Kr(this.beta1,this.iterations_+1)),this.accBeta2.assign(Kr(this.beta2,this.iterations_+1))});let e=t.length/2,n=!1;this.accumulatedFirstMoment=t.slice(0,e).map(o=>({originalName:o.name,variable:o.tensor.variable(n)})),this.accumulatedSecondMoment=t.slice(e,e*2).map(o=>({originalName:o.name,variable:o.tensor.variable(n)}))}getConfig(){return{learningRate:this.learningRate,beta1:this.beta1,beta2:this.beta2,epsilon:this.epsilon}}static fromConfig(t,e){return new t(e.learningRate,e.beta1,e.beta2,e.epsilon)}};ac.className="Adam";Sn(ac);var lc=class extends jr{constructor(t,e,n,o=null,s=0){super();this.learningRate=t,this.beta1=e,this.beta2=n,this.epsilon=o,this.decay=s,this.accumulatedFirstMoment=[],this.accumulatedWeightedInfNorm=[],B(()=>{this.iteration=pt(0).variable(),this.accBeta1=pt(e).variable()}),o==null&&(this.epsilon=_.backend.epsilon())}applyGradients(t){let e=Array.isArray(t)?t.map(n=>n.name):Object.keys(t);B(()=>{let n=lt(1,this.accBeta1),o=ct(-this.learningRate,J(P(this.iteration,this.decay),1));e.forEach((s,i)=>{let a=_.registeredVariables[s],u=!1;this.accumulatedFirstMoment[i]==null&&(this.accumulatedFirstMoment[i]={originalName:`${s}/m`,variable:St(a).variable(u)}),this.accumulatedWeightedInfNorm[i]==null&&(this.accumulatedWeightedInfNorm[i]={originalName:`${s}/v`,variable:St(a).variable(u)});let l=Array.isArray(t)?t[i].tensor:t[s];if(l==null)return;let c=this.accumulatedFirstMoment[i].variable,p=this.accumulatedWeightedInfNorm[i].variable,m=J(P(c,this.beta1),P(l,1-this.beta1)),f=P(p,this.beta2),d=Se(l),h=cn(f,d);c.assign(m),p.assign(h);let g=J(P(ct(o,n),ct(m,J(h,this.epsilon))),a);a.assign(g)}),this.iteration.assign(J(this.iteration,1)),this.accBeta1.assign(P(this.accBeta1,this.beta1))}),this.incrementIterations()}dispose(){this.accBeta1.dispose(),this.iteration.dispose(),this.accumulatedFirstMoment!=null&&_t(this.accumulatedFirstMoment.map(t=>t.variable)),this.accumulatedWeightedInfNorm!=null&&_t(this.accumulatedWeightedInfNorm.map(t=>t.variable))}async getWeights(){throw new Error("getWeights() is not implemented for Adamax yet.")}async setWeights(t){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(t,e){return new t(e.learningRate,e.beta1,e.beta2,e.epsilon,e.decay)}};lc.className="Adamax";Sn(lc);var el=class extends jr{constructor(t){super();this.learningRate=t,this.setLearningRate(t)}applyGradients(t){(Array.isArray(t)?t.map(n=>n.name):Object.keys(t)).forEach((n,o)=>{let s=Array.isArray(t)?t[o].tensor:t[n];if(s==null)return;let i=_.registeredVariables[n];B(()=>{let a=J(P(this.c,s),i);i.assign(a)})}),this.incrementIterations()}setLearningRate(t){this.learningRate=t,this.c!=null&&this.c.dispose(),this.c=Me(pt(-t))}dispose(){this.c.dispose()}async getWeights(){return[await this.saveIterations()]}async setWeights(t){if(t=await this.extractIterations(t),t.length!==0)throw new Error("SGD optimizer does not have settable weights.")}getConfig(){return{learningRate:this.learningRate}}static fromConfig(t,e){return new t(e.learningRate)}};el.className="SGD";Sn(el);var uc=class extends el{constructor(t,e,n=!1){super(t);this.learningRate=t,this.momentum=e,this.useNesterov=n,this.accumulations=[],this.m=pt(this.momentum)}applyGradients(t){(Array.isArray(t)?t.map(n=>n.name):Object.keys(t)).forEach((n,o)=>{let s=_.registeredVariables[n];this.accumulations[o]==null&&(this.accumulations[o]={originalName:`${n}/momentum`,variable:B(()=>St(s).variable(!1))});let i=this.accumulations[o].variable,a=Array.isArray(t)?t[o].tensor:t[n];a!=null&&B(()=>{let u,l=J(P(this.m,i),a);this.useNesterov?u=J(P(this.c,J(a,P(l,this.m))),s):u=J(P(this.c,l),s),i.assign(l),s.assign(u)})}),this.incrementIterations()}dispose(){this.m.dispose(),this.accumulations!=null&&_t(this.accumulations.map(t=>t.variable))}setMomentum(t){this.momentum=t}async getWeights(){return[await this.saveIterations()].concat(this.accumulations.map(t=>({name:t.originalName,tensor:t.variable})))}async setWeights(t){t=await this.extractIterations(t);let e=!1;this.accumulations=t.map(n=>({originalName:n.name,variable:n.tensor.variable(e)}))}getConfig(){return{learningRate:this.learningRate,momentum:this.momentum,useNesterov:this.useNesterov}}static fromConfig(t,e){return new t(e.learningRate,e.momentum,e.useNesterov)}};uc.className="Momentum";Sn(uc);var cc=class extends jr{constructor(t,e=.9,n=0,o=null,s=!1){super();if(this.learningRate=t,this.decay=e,this.momentum=n,this.epsilon=o,this.accumulatedMeanSquares=[],this.accumulatedMoments=[],this.accumulatedMeanGrads=[],this.centered=s,o==null&&(this.epsilon=_.backend.epsilon()),t==null)throw new Error("learningRate for RMSPropOptimizer must be defined.")}applyGradients(t){(Array.isArray(t)?t.map(n=>n.name):Object.keys(t)).forEach((n,o)=>{let s=_.registeredVariables[n],i=!1;this.accumulatedMeanSquares[o]==null&&(this.accumulatedMeanSquares[o]={originalName:`${n}/rms`,variable:B(()=>St(s).variable(i))}),this.accumulatedMoments[o]==null&&(this.accumulatedMoments[o]={originalName:`${n}/momentum`,variable:B(()=>St(s).variable(i))}),this.accumulatedMeanGrads[o]==null&&this.centered&&(this.accumulatedMeanGrads[o]={originalName:`${n}/mg`,variable:B(()=>St(s).variable(i))});let a=Array.isArray(t)?t[o].tensor:t[n];if(a==null)return;let u=this.accumulatedMeanSquares[o].variable,l=this.accumulatedMoments[o].variable;B(()=>{let c=J(P(u,this.decay),P(Wt(a),1-this.decay));if(this.centered){let p=this.accumulatedMeanGrads[o].variable,m=J(P(p,this.decay),P(a,1-this.decay)),f=ct(P(a,this.learningRate),be(lt(c,J(Wt(m),this.epsilon)))),d=J(P(l,this.momentum),f);u.assign(c),p.assign(m),l.assign(d);let h=lt(s,d);s.assign(h)}else{let p=J(P(u,this.decay),P(Wt(a),1-this.decay)),m=J(P(l,this.momentum),ct(P(a,this.learningRate),be(J(p,this.epsilon))));u.assign(p),l.assign(m);let f=lt(s,m);s.assign(f)}})}),this.incrementIterations()}dispose(){this.accumulatedMeanSquares!=null&&_t(this.accumulatedMeanSquares.map(t=>t.variable)),this.accumulatedMeanGrads!=null&&this.centered&&_t(this.accumulatedMeanGrads.map(t=>t.variable)),this.accumulatedMoments!=null&&_t(this.accumulatedMoments.map(t=>t.variable))}async getWeights(){let t=[...this.accumulatedMeanSquares,...this.accumulatedMoments];return this.centered&&t.push(...this.accumulatedMeanGrads),[await this.saveIterations()].concat(t.map(e=>({name:e.originalName,tensor:e.variable})))}async setWeights(t){t=await this.extractIterations(t);let e=this.centered?t.length/3:t.length/2,n=!1;this.accumulatedMeanSquares=t.slice(0,e).map(o=>({originalName:o.name,variable:o.tensor.variable(n)})),this.accumulatedMoments=t.slice(e,e*2).map(o=>({originalName:o.name,variable:o.tensor.variable(n)})),this.centered&&(this.accumulatedMeanGrads=t.slice(e*2,e*3).map(o=>({originalName:o.name,variable:o.tensor.variable(n)})))}getConfig(){return{learningRate:this.learningRate,decay:this.decay,momentum:this.momentum,epsilon:this.epsilon,centered:this.centered}}static fromConfig(t,e){return new t(e.learningRate,e.decay,e.momentum,e.epsilon,e.centered)}};cc.className="RMSProp";Sn(cc);var Mi=class{static sgd(t){return new el(t)}static momentum(t,e,n=!1){return new uc(t,e,n)}static rmsprop(t,e=.9,n=0,o=null,s=!1){return new cc(t,e,n,o,s)}static adam(t=.001,e=.9,n=.999,o=null){return new ac(t,e,n,o)}static adadelta(t=.001,e=.95,n=null){return new sc(t,e,n)}static adamax(t=.002,e=.9,n=.999,o=null,s=0){return new lc(t,e,n,o,s)}static adagrad(t,e=.1){return new ic(t,e)}};var pc={sgd:Mi.sgd,momentum:Mi.momentum,adadelta:Mi.adadelta,adagrad:Mi.adagrad,rmsprop:Mi.rmsprop,adamax:Mi.adamax,adam:Mi.adam};var Z6=(()=>typeof requestAnimationFrame!="undefined"?requestAnimationFrame:typeof setImmediate!="undefined"?setImmediate:r=>r())();function fd(){return new Promise(r=>Z6(()=>r()))}var S={};Yt(S,{ERF_A1:()=>l5,ERF_A2:()=>u5,ERF_A3:()=>c5,ERF_A4:()=>p5,ERF_A5:()=>m5,ERF_P:()=>a5,PARALLELIZE_THRESHOLD:()=>Bx,SELU_SCALE:()=>sC,SELU_SCALEALPHA:()=>oC,applyActivation:()=>nc,assertAndGetBroadcastShape:()=>Lt,assertAxesAreInnerMostDims:()=>Jq,assertParamsConsistent:()=>J6,assignToTypedArray:()=>y5,axesAreInnerMostDims:()=>M0,calculateShapes:()=>c_,checkEinsumDimSizes:()=>S5,checkPadOnDimRoundingMode:()=>ke,combineLocations:()=>E_,complexWithEvenIndex:()=>h5,complexWithOddIndex:()=>g5,computeConv2DInfo:()=>Ou,computeConv3DInfo:()=>N_,computeDefaultPad:()=>w0,computeDilation2DInfo:()=>CH,computeOptimalWindowSize:()=>t5,computeOutAndReduceShapes:()=>L0,computeOutShape:()=>Q6,computePool2DInfo:()=>b0,computePool3DInfo:()=>IH,convertConv2DDataFormat:()=>T_,decodeEinsumEquation:()=>C5,eitherStridesOrDilationsAreOne:()=>Dr,expandShapeToKeepDim:()=>yo,exponent:()=>w5,exponents:()=>b5,fromStringArrayToUint8:()=>H5,fromUint8ToStringArray:()=>U5,getAxesPermutation:()=>z0,getBroadcastDims:()=>a_,getComplexWithIndex:()=>x5,getEinsumComputePath:()=>k5,getEinsumPermutation:()=>I5,getFusedBiasGradient:()=>rc,getFusedDyActivation:()=>ec,getImageCenter:()=>e5,getInnerMostAxes:()=>Qq,getPermuted:()=>n5,getReductionAxes:()=>ge,getReshaped:()=>r5,getReshapedPermuted:()=>o5,getSliceBeginCoords:()=>s5,getSliceSize:()=>i5,getSparseFillEmptyRowsIndicesDenseShapeMismatch:()=>E5,getSparseFillEmptyRowsNegativeIndexErrorMessage:()=>A5,getSparseFillEmptyRowsOutOfRangeIndexErrorMessage:()=>D5,getSparseReshapeEmptyTensorZeroOutputDimErrorMessage:()=>F5,getSparseReshapeInputOutputMismatchErrorMessage:()=>P5,getSparseReshapeInputOutputMultipleErrorMessage:()=>O5,getSparseReshapeMultipleNegativeOneOutputDimErrorMessage:()=>$5,getSparseReshapeNegativeOutputDimErrorMessage:()=>R5,getSparseSegmentReductionIndicesOutOfRangeErrorMessage:()=>B5,getSparseSegmentReductionNegativeSegmentIdsErrorMessage:()=>M5,getSparseSegmentReductionNonIncreasingSegmentIdsErrorMessage:()=>L5,getSparseSegmentReductionSegmentIdOutOfRangeErrorMessage:()=>z5,getUndoAxesPermutation:()=>Zf,isIdentityPermutation:()=>N5,log:()=>LU,mergeRealAndImagArrays:()=>f5,prepareAndValidate:()=>u_,prepareSplitSize:()=>_5,segment_util:()=>aC,shouldFuse:()=>oc,slice_util:()=>Ve,splitRealAndImagArrays:()=>d5,tupleValuesAreOne:()=>Xn,upcastType:()=>ar,validateInput:()=>gx,validateUpdateShape:()=>p0,warn:()=>Ri});function J6(r,t){let e=r[0].length;r.forEach((o,s)=>{A(o.length===e,()=>`Error in concat${e}D: rank of tensors[${s}] must be the same as the rank of the rest (${e})`)}),A(t>=0&&t<e,()=>`Error in concat${e}D: axis must be between 0 and ${e-1}.`);let n=r[0];r.forEach((o,s)=>{for(let i=0;i<e;i++)A(i===t||o[i]===n[i],()=>`Error in concat${e}D: Shape of tensors[${s}] (${o}) does not match the shape of the rest (${n}) along the non-concatenated axis ${s}.`)})}function Q6(r,t){let e=r[0].slice();for(let n=1;n<r.length;n++)e[t]+=r[n][t];return e}var Bx=30;function t5(r){return r<=Bx?r:ip(r,Math.floor(Math.sqrt(r)))}function e5(r,t,e){let n=e*(typeof r=="number"?r:r[0]),o=t*(typeof r=="number"?r:r[1]);return[n,o]}function r5(r,t,e,n=!0){let o=[];if(n)o=o.concat(t.slice(0)),o.push(r[0]/e),o=o.concat(r.slice(1));else{o=o.concat(r[0]);let s=t.length;for(let i=0;i<s;++i)o=o.concat([r[i+1]/t[i],t[i]]);o=o.concat(r.slice(s+1))}return o}function n5(r,t,e=!0){let n=[];if(e){n.push(t);for(let o=t+1;o<r;++o)o<=2*t?(n.push(o),n.push(o-(t+1))):n.push(o)}else{let o=[],s=[];for(let i=1;i<r;++i)i>=t*2+1||i%2===1?s.push(i):o.push(i);n.push(...o),n.push(0),n.push(...s)}return n}function o5(r,t,e,n=!0){let o=[];n?o.push(r[0]/e):o.push(r[0]*e);for(let s=1;s<r.length;++s)s<=t.length?n?o.push(t[s-1]*r[s]):o.push(r[s]/t[s-1]):o.push(r[s]);return o}function s5(r,t){let e=[0];for(let n=0;n<t;++n)e.push(r[n][0]);return e}function i5(r,t,e){let n=r.slice(0,1);for(let o=0;o<e;++o)n.push(r[o+1]-t[o][0]-t[o][1]);return n}var oC=1.7580993408473768,sC=1.0507009873554805;var a5=.3275911,l5=.254829592,u5=-.284496736,c5=1.421413741,p5=-1.453152027,m5=1.061405429;function f5(r,t){if(r.length!==t.length)throw new Error(`Cannot merge real and imag arrays of different lengths. real:${r.length}, imag: ${t.length}.`);let e=new Float32Array(r.length*2);for(let n=0;n<e.length;n+=2)e[n]=r[n/2],e[n+1]=t[n/2];return e}function d5(r){let t=new Float32Array(r.length/2),e=new Float32Array(r.length/2);for(let n=0;n<r.length;n+=2)t[n/2]=r[n],e[n/2]=r[n+1];return{real:t,imag:e}}function h5(r){let t=Math.ceil(r.length/4),e=new Float32Array(t),n=new Float32Array(t);for(let o=0;o<r.length;o+=4)e[Math.floor(o/4)]=r[o],n[Math.floor(o/4)]=r[o+1];return{real:e,imag:n}}function g5(r){let t=Math.floor(r.length/4),e=new Float32Array(t),n=new Float32Array(t);for(let o=2;o<r.length;o+=4)e[Math.floor(o/4)]=r[o],n[Math.floor(o/4)]=r[o+1];return{real:e,imag:n}}function x5(r,t){let e=r[t*2],n=r[t*2+1];return{real:e,imag:n}}function y5(r,t,e,n){r[n*2]=t,r[n*2+1]=e}function b5(r,t){let e=new Float32Array(r/2),n=new Float32Array(r/2);for(let o=0;o<Math.ceil(r/2);o++){let s=(t?2:-2)*Math.PI*(o/r);e[o]=Math.cos(s),n[o]=Math.sin(s)}return{real:e,imag:n}}function w5(r,t,e){let n=(e?2:-2)*Math.PI*(r/t),o=Math.cos(n),s=Math.sin(n);return{real:o,imag:s}}var iC="->",v5=/->/g,PE=",",ME="...";function C5(r,t){r=r.replace(/\s/g,"");let e=(r.length-r.replace(v5,"").length)/iC.length;if(e<1)throw new Error("Equations without an arrow are not supported.");if(e>1)throw new Error(`Equation must contain exactly one arrow ("${iC}").`);let[n,o]=r.split(iC);A(n.indexOf(ME)===-1,()=>`The ellipsis notation ("${ME}") is not supported yet.`);let s=n.split(PE),i=s.length;if(t!==i)throw new Error(`Expected ${i} input tensors, received ${t}`);if(i>2)throw new Error("Support for more than 2 input tensors is not implemented yet.");let a=[];for(let m=0;m<o.length;++m){let f=o[m];if(!s.some(d=>d.indexOf(f)!==-1))throw new Error(`Output subscripts contain the label ${f} not present in the input subscripts.`);a.indexOf(f)===-1&&a.push(f)}for(let m=0;m<n.length;++m){let f=n[m];a.indexOf(f)===-1&&f!==PE&&a.push(f)}let u=new Array(s.length);for(let m=0;m<i;++m){if(new Set(s[m].split("")).size!==s[m].length)throw new Error(`Found duplicate axes in input component ${s[m]}. Support for duplicate axes in input is not implemented yet.`);u[m]=[];for(let f=0;f<s[m].length;++f)u[m].push(a.indexOf(s[m][f]))}let l=a.length,c=o.length,p=[];for(let m=c;m<l;++m)p.push(m);return{allDims:a,summedDims:p,idDims:u}}function I5(r,t){let e=new Array(r);e.fill(-1);for(let o=0;o<t.length;++o)e[t[o]]=o;let n=[];for(let o=0;o<r;++o)e[o]===-1&&n.push(o);return e=e.filter(o=>o!==-1),{permutationIndices:e,expandDims:n}}function S5(r,t,e){let n=new Array(r);for(let o=0;o<e.length;++o){let s=e[o].shape;for(let i=0;i<t[o].length;++i)n[t[o][i]]===void 0?n[t[o][i]]=s[i]:A(n[t[o][i]]===s[i],()=>`Expected dimension ${n[t[o][i]]} at axis ${i} of input shaped ${JSON.stringify(s)}, but got dimension ${s[i]}`)}}function k5(r,t){let e=r,n=[],o=0;r.length===0&&e.push(-1),o=r.length+1;for(let i=0;i<o;++i)n.push([]);let s=[];for(let i=0;i<e.length;++i){let a=e[i],u=T5(t,a);for(let l of u)s.indexOf(l)===-1&&(n[i].push(l),s.push(l))}return{path:e,steps:n}}function N5(r){return r.every((t,e)=>t===e)}function T5(r,t){let e=[];for(let n=0;n<r.length;++n)(r[n].length===0||r[n].indexOf(t)!==-1||t===-1)&&e.push(n);return e}function _5(r,t,e=0){let n=[];if(typeof t=="number")A(r.shape[e]%t===0,()=>"Number of splits must evenly divide the axis."),n=new Array(t).fill(r.shape[e]/t);else{let o=t.reduce((i,a)=>(a===-1&&(i+=1),i),0);A(o<=1,()=>"There should be only one negative value in split array.");let s=t.indexOf(-1);if(s!==-1){let i=t.reduce((a,u)=>u>0?a+u:a);t[s]=r.shape[e]-i}A(r.shape[e]===t.reduce((i,a)=>i+a),()=>"The sum of sizes must match the size of the axis dimension."),n=t}return n}function E5(r){return`Received SparseTensor with denseShape[0] = 0 but
indices.shape[0] = ${r}`}function A5(r,t){return`indices(${r}, 0) is invalid: ${t} < 0`}function D5(r,t,e){return`indices(${r}, 0) is invalid: ${t} >= ${e}`}function $5(r,t){return`only one output dimension may be -1, not both ${r} and ${t}`}function R5(r,t){return`size ${r} must be non-negative, not ${t}`}function F5(){return"reshape cannot infer the missing input size for an empty tensor unless all specified input sizes are non-zero"}function O5(r,t){let e=re(r),n=re(t);return`Input to reshape is a SparseTensor with ${e}
dense values, but the requested shape requires a multiple of ${n}. inputShape=${r} outputShape= ${t}`}function P5(r,t){let e=re(r),n=re(t);return`Input to reshape is a tensor with ${e} dense values, but the requested shape has ${n}. inputShape=${r} outputShape=${t}`}function M5(){return"segment ids must be >= 0"}function L5(){return"segment ids are not increasing"}function z5(r,t){return`Segment id ${r} out of range [0, ${t}), possibly because segmentIds input is not sorted.`}function B5(r,t,e){return`Bad: indices[${r}] == ${t} out of range [0, ${e})`}var aC={};Yt(aC,{collectGatherOpShapeInfo:()=>W5,computeOutShape:()=>G5,segOpComputeOptimalWindowSize:()=>V5});function V5(r,t){let e=!1,n;for(r<=Bx?(n=r,e=!0):n=ip(r,Math.floor(Math.sqrt(r)));!e;)n>t||n===r?e=!0:n=ip(r,n+1);return n}function G5(r,t,e){let n=[],o=r.length;for(let s=0;s<o;s++)s!==t?n.push(r[s]):n.push(e);return n}function W5(r,t,e,n){let o=t.shape.length,s=r.shape.length;if(n!==0&&(n<-o||n>o))throw new Error(`Expect batchDims in the range of [-${o}, ${o}], but got ${n}`);if(n<0&&(n+=o),n>s)throw new Error(`batchDims (${n}) must be less than rank(x) (
${s}).`);if(e<n)throw new Error(`batchDims (${n}) must be less than or equal to axis (${e}).`);for(let p=0;p<n;++p)if(r.shape[p]!==t.shape[p])throw new Error(`x.shape[${p}]: ${r.shape[p]} should be equal to indices.shape[${p}]: ${t.shape[p]}.`);let i=r.shape[e],a=[],u=1,l=1,c=1;for(let p=0;p<n;++p)a.push(r.shape[p]),u*=r.shape[p];for(let p=n;p<e;p++)a.push(r.shape[p]),l*=r.shape[p];for(let p=n;p<o;p++)a.push(t.shape[p]);for(let p=e+1;p<s;p++)a.push(r.shape[p]),c*=r.shape[p];return{batchSize:u,sliceSize:c,outerSize:l,dimSize:i,outputShape:a}}function U5(r){try{return r.map(t=>Wp(t))}catch(t){throw new Error(`Failed to decode encoded string bytes into utf-8, error: ${t}`)}}function H5(r){return r.map(t=>Ml(t))}var Vr={};Yt(Vr,{nonMaxSuppressionV3Impl:()=>Fx,nonMaxSuppressionV4Impl:()=>Ox,nonMaxSuppressionV5Impl:()=>Px,whereImpl:()=>Tx});var Vx={kernelName:hi,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>P(r,Qs(Z(e,"float32"),-1))}}};var LE={kernelName:ji,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>{let n=Wt(Z(e,"float32")),o=be(lt(pt(1),n));return Zt(ct(r,o))}}}};var zE={kernelName:Xi,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>{let n=be(lt(Wt(Z(e,"float32")),1));return ct(r,n)}}}};var BE={kernelName:Kn,inputsToSave:["a","b"],gradFunc:(r,t)=>{let[e,n]=t,o=Lt(e.shape,n.shape);return{a:()=>{let a=r,u=ge(e.shape,o);return u.length>0&&(a=mt(a,u)),O(a,e.shape)},b:()=>{let a=r,u=ge(n.shape,o);return u.length>0&&(a=mt(a,u)),O(a,n.shape)}}}};var VE={kernelName:jo,saveAllInputs:!0,gradFunc:(r,t)=>{let e={};return t.forEach((n,o)=>{e[o]=()=>r.clone()}),e}};var GE={kernelName:Xo,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>St(e)}}};var WE={kernelName:vl,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>St(e)}}};var UE={kernelName:Ji,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>ct(r,be(lt(pt(1),Wt(Z(e,"float32")))))}}};var HE={kernelName:Qi,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>{let n=be(J(pt(1),Wt(Z(e,"float32"))));return ct(r,n)}}}};var qE={kernelName:ra,inputsToSave:["a","b"],gradFunc:(r,t)=>{let[e,n]=t,o=Lt(e.shape,n.shape);return{a:()=>{let a=J(Wt(e),Wt(n)),u=P(r,ct(n,a)),l=ge(e.shape,o);return l.length>0&&(u=mt(u,l)),O(u,e.shape)},b:()=>{let a=J(Wt(e),Wt(n)),u=Zt(P(r,ct(e,a))),l=ge(n.shape,o);return l.length>0&&(u=mt(u,l)),O(u,n.shape)}}}};var KE={kernelName:ta,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>ct(r,J(Wt(Z(e,"float32")),1))}}};var jE={kernelName:ea,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>ct(r,lt(pt(1),Wt(Z(e,"float32"))))}}};function q5(r,t,e,n,o,s){let i=C(r,"dy","avgPool3dGrad"),a=C(t,"input","avgPool3dGrad"),u=i,l=a,c=!1;a.rank===4&&(c=!0,u=O(i,[1,i.shape[0],i.shape[1],i.shape[2],i.shape[3]]),l=O(a,[1,a.shape[0],a.shape[1],a.shape[2],a.shape[3]])),A(u.rank===5,()=>`Error in avgPool3dGrad: dy must be rank 5 but got rank ${u.rank}.`),A(l.rank===5,()=>`Error in avgPool3dGrad: input must be rank 5 but got rank ${l.rank}.`),ke("avgPool3dGrad",o,s);let p={dy:u,input:l},m={filterSize:e,strides:n,pad:o,dimRoundingMode:s},f=_.runKernel(up,p,m);return c?O(f,[f.shape[1],f.shape[2],f.shape[3],f.shape[4]]):f}var XE=N({avgPool3dGrad_:q5});var YE={kernelName:Cl,inputsToSave:["x"],gradFunc:(r,t,e)=>{let[n]=t,{filterSize:o,strides:s,pad:i,dimRoundingMode:a}=e;return{x:()=>XE(r,n,o,s,i,a)}}};function K5(r,t,e,n,o){let s=C(r,"dy","avgPoolGrad"),i=C(t,"input","avgPoolGrad");A(i.rank===s.rank,()=>`Rank of input (${i.rank}) does not match rank of dy (${s.rank})`);let a=i,u=s,l=!1;i.rank===3&&(l=!0,a=O(i,[1,i.shape[0],i.shape[1],i.shape[2]]),u=O(s,[1,s.shape[0],s.shape[1],s.shape[2]])),A(u.rank===4,()=>`Error in avgPoolGrad: dy must be rank 4 but got rank ${u.rank}.`),A(a.rank===4,()=>`Error in avgPoolGrad: input must be rank 4 but got rank ${a.rank}.`);let c={dy:u,input:a},p={filterSize:e,strides:n,pad:o},m=_.runKernel(lp,c,p);return l?O(m,[m.shape[1],m.shape[2],m.shape[3]]):m}var ZE=N({avgPoolGrad_:K5});var JE={kernelName:Yo,inputsToSave:["x"],gradFunc:(r,t,e)=>{let[n]=t,{filterSize:o,strides:s,pad:i}=e;return{x:()=>ZE(r,n,o,s,i)}}};var QE={kernelName:Zo,inputsToSave:["a","b"],gradFunc:(r,t,e)=>{let[n,o]=t,{transposeA:s,transposeB:i}=e;return!s&&!i?{a:()=>Mt(r,o,!1,!0),b:()=>Mt(n,r,!0,!1)}:!s&&i?{a:()=>Mt(r,o,!1,!1),b:()=>Mt(r,n,!0,!1)}:s&&!i?{a:()=>Mt(o,r,!1,!0),b:()=>Mt(n,r,!1,!1)}:{a:()=>Mt(o,r,!0,!0),b:()=>Mt(r,n,!0,!0)}}};var tA={kernelName:gi,gradFunc:(r,t,e)=>{let{blockShape:n,crops:o}=e;return{x:()=>Xa(r,n,o)}}};var eA={kernelName:YT,gradFunc:(r,t,e)=>{let n=e,o=n.inputShape,s=n.shape,i=Array.from(s);for(let u=o.length-1;u>=0;u--)if(o[u]===s[u])i[u]=1;else if(o[u]!==1)throw new Error(`broadcastTo(): [${o}] cannot be broadcast to [${s}].`);let a=[];for(let u=0;u<i.length;u++)i[u]>1&&a.push(u);return{x:()=>mt(r,a,!0)}}};var rA={kernelName:oo,gradFunc:r=>({x:()=>r.clone()})};var nA={kernelName:Jo,gradFunc:r=>({x:()=>St(r)})};var oA={kernelName:so,inputsToSave:["x"],gradFunc:(r,t,e)=>{let[n]=t,{clipValueMin:o,clipValueMax:s}=e;return{x:()=>Ne(Nr(kn(n,o),Nn(n,s)),r,St(r))}}};var sA={kernelName:Il,inputsToSave:["x"],gradFunc:Vx.gradFunc};var iA={kernelName:xi,saveAllInputs:!0,gradFunc:(r,t,e)=>{let n=t.map(u=>u.shape),{axis:o}=e,s=mr(o,t[0].shape)[0],i=n.map(u=>u[s]);return ur(r,i,s).map(u=>()=>u)}};var aA={kernelName:Qo,inputsToSave:["x","filter"],gradFunc:(r,t,e)=>{let[n,o]=t,{dilations:s,strides:i,pad:a,dataFormat:u}=e;return A(Xn(s),()=>`Error in gradient of conv2D: dilation rates greater than 1 are not yet supported in gradients. Got dilations '${s}'`),{x:()=>Zp(n.shape,r,o,i,a,u),filter:()=>om(n,r,o.shape,i,a,u)}}};var lA={kernelName:ts,inputsToSave:["dy","filter"],gradFunc:(r,t,e)=>{let[n,o]=t,{strides:s,pad:i,dataFormat:a,dimRoundingMode:u}=e;return{dy:()=>ln(r,o,s,i,a,1,u),filter:()=>om(r,n,o.shape,s,i,a,u)}}};function j5(r,t,e,n,o){let s=r;r.rank===4&&(s=O(r,[1,r.shape[0],r.shape[1],r.shape[2],r.shape[3]]));let i=t;i.rank===4&&(i=O(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(e.length===5,()=>`Error in conv3dDerFilter: filterShape must be length 5, but got ${e}.`),A(s.shape[4]===e[3],()=>`Error in conv3dDerFilter: depth of input ${s.shape[4]}) must match input depth in filter (${e[3]}.`),A(i.shape[4]===e[4],()=>`Error in conv3dDerFilter: depth of dy (${i.shape[4]}) must match output depth for filter (${e[4]}).`);let a={x:s,dy:i},u={strides:n,pad:o,filterShape:e};return _.runKernel(dp,a,u)}var uA=N({conv3DBackpropFilter_:j5});var cA={kernelName:Sl,inputsToSave:["x","filter"],gradFunc:(r,t,e)=>{let{dilations:n,strides:o,pad:s}=e;A(Xn(n),()=>`Error in gradient of conv3D: dilation rates greater than 1 are not yet supported in gradients. Got dilations '${n}'`);let[i,a]=t;return{x:()=>wx(i.shape,r,a,o,s),filter:()=>uA(i,r,a.shape,o,s)}}};var pA={kernelName:es,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>P(Zt(Yu(Z(e,"float32"))),r)}}};var mA={kernelName:rs,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>P(Zu(Z(e,"float32")),r)}}};var fA={kernelName:ns,inputsToSave:["x"],gradFunc:(r,t,e)=>{let[n]=t,{axis:o,exclusive:s,reverse:i}=e;return{x:()=>{let a=z0([o],n.rank),u=zu(r,o,s,!i);return a!=null&&(u=Gt(u,a)),u}}}};var dA={kernelName:os,inputsToSave:["x","filter"],gradFunc:(r,t,e)=>{let{dilations:n,strides:o,pad:s,dimRoundingMode:i}=e,a=n==null?[1,1]:n;A(Xn(a),()=>`Error in gradient of depthwiseConv2dNative: dilation rates greater than 1 are not yet supported. Got dilations '${a}'`);let[u,l]=t;return A(u.rank===4,()=>`Error in gradient of depthwiseConv2dNative: input must be rank 4, but got rank ${u.rank}.`),A(l.rank===4,()=>`Error in gradient of depthwiseConv2dNative: filter must be rank 4, but got rank ${l.rank}.`),A(u.shape[3]===l.shape[2],()=>`Error in gradient of depthwiseConv2d: number of input channels (${u.shape[3]}) must match the inChannels dimension in filter ${l.shape[2]}.`),A(Dr(o,a),()=>`Error in gradient of depthwiseConv2d: Either strides or dilations must be 1. Got strides ${o} and dilations '${a}'.`),ke("depthwiseConv2d",s,i),{x:()=>Dx(u.shape,r,l,o,s,a,i),filter:()=>Ax(u,r,l.shape,o,s,a,i)}}};var hA={kernelName:kl,inputsToSave:["x","filter"],gradFunc:(r,t,e)=>{let[n,o]=t,s={x:n,filter:o,dy:r},i={x:n,filter:o,dy:r};return{x:()=>_.runKernel(yf,s,e),filter:()=>_.runKernel(bf,i,e)}}};var gA={kernelName:is,outputsToSave:[!0],gradFunc:(r,t)=>{let[e]=t,n={dy:r,y:e};return{x:()=>_.runKernel(vp,n)}}};var xA={kernelName:ia,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t,n=P(Ze(Zt(Wt(e))),2/Math.sqrt(Math.PI));return{x:()=>P(r,n)}}};var yA={kernelName:as,outputsToSave:[!0],gradFunc:(r,t)=>{let[e]=t;return{x:()=>P(r,e)}}};var bA={kernelName:yi,inputsToSave:["input"],gradFunc:(r,t)=>{let[e]=t;return{input:()=>O(r,e.shape)}}};var wA={kernelName:la,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>P(r,Ze(e))}}};var vA={kernelName:ls,gradFunc:r=>({x:()=>St(r)})};var CA={kernelName:us,inputsToSave:["a","b"],gradFunc:(r,t)=>{let[e,n]=t,o=Lt(e.shape,n.shape);return{a:()=>{let a=ct(r,Z(n,"float32")),u=ge(e.shape,o);return u.length>0?O(mt(a,u),e.shape):a},b:()=>{let a=P(r,Z(e,"float32")),u=ge(n.shape,o);u.length>0&&(a=O(mt(a,u),n.shape));let l=Wt(n);return Zt(ct(a,Z(l,"float32")))}}}};var IA={kernelName:cs,inputsToSave:["x","mean","variance","scale"],gradFunc:(r,t,e)=>{let{varianceEpsilon:n}=e,[o,s,i,a]=t,u=a==null?pt(1):a,l=ge(s.shape,o.shape),c=[];if(s.rank===1){for(let v=0;v<o.shape.length-1;++v)c.push(o.shape[v]);c.push(1)}let p=lt(o,s),m=P(r,u),f=ju(J(i,pt(n))),d=P(P(P(f,f),f),pt(-.5));return{x:()=>s.rank===1?O(P(P(r,kr(O(f,[1,1,1,s.shape[0]]),c)),u),o.shape):O(P(P(r,f),u),o.shape),mean:()=>{let v=P(P(f,pt(-1)),m);return s.rank===1&&(v=mt(v,l)),O(v,s.shape)},variance:()=>{let v=P(P(d,p),m);return s.rank===1&&(v=mt(v,l)),O(v,s.shape)},scale:()=>{let v=P(p,f),k=P(r,v);return s.rank===1&&(k=mt(k,l)),O(k,s.shape)},offset:()=>{let v=r;return s.rank===1&&(v=mt(v,l)),O(v,s.shape)}}}};var NA={kernelName:bi,inputsToSave:["x","indices"],gradFunc:(r,t,e)=>{let[n,o]=t,{axis:s}=e,i=mr(s,n.shape)[0];return{x:()=>{let u=n.shape,l=o.size,c=u.slice(0,i),p=c.length,m=u.slice(s,u.length).slice(1),f=m.length,d=SA(0,p),h=SA(p+1,p+1+f),g=kA([c,[l],m]),y=O(r,g),b=O(o,[l]),w=kA([[p],d,h]),v=Gt(y,w),k=cd(v,b,n.shape[i]),E=Zf(w);return k=Gt(k,E),k},indices:()=>o}}};function SA(r,t){let e=[];for(let n=r;n<t;++n)e.push(n);return e}function kA(r){let t=[];for(let e=0;e<r.length;++e)for(let n=0;n<r[e].length;++n)t.push(r[e][n]);return t}var TA={kernelName:ps,inputsToSave:["a","b"],gradFunc:(r,t)=>{let[e,n]=t;return{a:()=>St(e),b:()=>St(n)}}};var _A={kernelName:io,gradFunc:r=>({x:()=>Z(r,"float32")})};var EA={kernelName:ma,gradFunc:r=>({x:()=>St(r)})};var AA={kernelName:fa,gradFunc:r=>({x:()=>St(r)})};var DA={kernelName:da,gradFunc:r=>({x:()=>St(r)})};var $A={kernelName:ms,inputsToSave:["x"],gradFunc:(r,t,e)=>{let[n]=t,{alpha:o}=e,s=Ge(n,0);return{x:()=>Ne(s,r,P(r,o))}}};var RA={kernelName:xa,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>ct(r,J(e,1))}}};var FA={kernelName:fs,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>ct(r,Z(e,"float32"))}}};var OA={kernelName:ZT,inputsToSave:[],outputsToSave:[!0],gradFunc:(r,t,e)=>{let[n]=t,{axis:o}=e;return{logits:()=>{let i=Ze(n);return lt(r,P(mt(r,o,!0),i))}}}};function X5(r,t,e,n=5,o=1,s=1,i=.5){let a={x:r,y:t,dy:e},u={depthRadius:n,bias:o,alpha:s,beta:i};return _.runKernel(Np,a,u)}var PA=N({localResponseNormalizationBackprop_:X5});var MA={kernelName:Tl,inputsToSave:["x"],outputsToSave:[!0],gradFunc:(r,t,e)=>{let[n,o]=t,{depthRadius:s,bias:i,alpha:a,beta:u}=e;return{x:()=>PA(n,o,r,s,i,a,u)}}};function Gx(r,t,e,n){return t.rank<e.rank&&(t=O(t,yo(t.shape,n))),r.rank<e.rank&&(r=O(r,yo(r.shape,n))),{x:()=>P(r,Z(Sr(e,t),r.dtype))}}var lC={kernelName:ds,inputsToSave:["x"],outputsToSave:[!0],gradFunc:(r,t,e)=>{let n=e,{reductionIndices:o}=n,s=t[0],i=t[1],a=mr(o,s.shape),u=Gx(r,i,s,a);return{x:()=>u.x()}}};var LA={kernelName:hs,inputsToSave:["a","b"],gradFunc:(r,t)=>{let[e,n]=t;return{a:()=>P(r,Z(kn(e,n),"float32")),b:()=>P(r,Z(Vu(e,n),"float32"))}}};function Y5(r,t,e,n,o,s,i){let a=C(r,"dy","maxPool3dGrad"),u=C(t,"input","maxPool3dGrad"),l=C(e,"output","maxPool3dGrad"),c=a,p=u,m=l,f=!1;u.rank===4&&(f=!0,c=O(a,[1,a.shape[0],a.shape[1],a.shape[2],a.shape[3]]),p=O(u,[1,u.shape[0],u.shape[1],u.shape[2],u.shape[3]]),m=O(l,[1,l.shape[0],l.shape[1],l.shape[2],l.shape[3]])),A(c.rank===5,()=>`Error in maxPool3dGrad: dy must be rank 5 but got rank ${c.rank}.`),A(p.rank===5,()=>`Error in maxPool3dGrad: input must be rank 5 but got rank ${p.rank}.`),A(m.rank===5,()=>`Error in maxPool3dGrad: output must be rank 5 but got rank ${m.rank}.`),ke("maxPool3dGrad",s,i);let d={dy:c,input:p,output:m},h={filterSize:n,strides:o,pad:s,dimRoundingMode:i},g=_.runKernel(_p,d,h);return f?O(g,[g.shape[1],g.shape[2],g.shape[3],g.shape[4]]):g}var zA=N({maxPool3dGrad_:Y5});var BA={kernelName:_l,inputsToSave:["x"],outputsToSave:[!0],gradFunc:(r,t,e)=>{let[n,o]=t,{filterSize:s,strides:i,pad:a,dimRoundingMode:u}=e;return{x:()=>zA(r,n,o,s,i,a,u)}}};function Z5(r,t,e,n,o,s,i){let a=C(r,"dy","maxPoolGrad"),u=C(t,"input","maxPoolGrad"),l=C(e,"output","maxPoolGrad");A(u.rank===a.rank,()=>`Rank of input (${u.rank}) does not match rank of dy (${a.rank})`),A(a.rank===4,()=>`Error in maxPoolGrad: dy must be rank 4 but got rank ${a.rank}.`),A(u.rank===4,()=>`Error in maxPoolGrad: input must be rank 4 but got rank ${u.rank}.`),ke("maxPoolGrad",s,i);let c={dy:a,input:u,output:l},p={filterSize:n,strides:o,pad:s,dimRoundingMode:i};return _.runKernel(Tp,c,p)}var VA=N({maxPoolGrad_:Z5});var GA={kernelName:gs,inputsToSave:["x"],outputsToSave:[!0],gradFunc:(r,t,e)=>{let[n,o]=t,{filterSize:s,strides:i,pad:a}=e;return{x:()=>VA(r,n,o,s,i,a)}}};var WA={kernelName:xs,inputsToSave:["x"],gradFunc:(r,t,e)=>{let[n]=t,{axis:o}=e,s=mr(o,n.shape),a=L0(n.shape,s)[1],u=re(a);return{x:()=>{let c=n.shape.slice();s.forEach(f=>{c[f]=1});let p=O(r,c);return ct(P(p,lr(n.shape,"float32")),u)}}}};var UA={kernelName:ys,inputsToSave:["x"],outputsToSave:[!0],gradFunc:(r,t,e)=>{let n=e,{axis:o}=n,[s,i]=t,a=mr(o,s.shape),u=Gx(r,i,s,a);return{x:()=>u.x()}}};var HA={kernelName:bs,inputsToSave:["a","b"],gradFunc:(r,t)=>{let[e,n]=t;return{a:()=>P(r,Z(Nn(e,n),"float32")),b:()=>P(r,Z(Ge(e,n),"float32"))}}};var qA={kernelName:ws,inputsToSave:["x"],gradFunc:(r,t,e)=>{let n=t[0],{paddings:o}=e,s=o.map(i=>i[0]);return{x:()=>Rt(r,s,n.shape)}}};var KA={kernelName:ba,inputsToSave:["a","b"],gradFunc:(r,t)=>{let[e,n]=t,o=Lt(e.shape,n.shape);return{a:()=>{let a=ge(e.shape,o);return a.length>0?O(mt(r,a),e.shape):r},b:()=>{let a=P(r,Zt(Ys(ct(e,n)))),u=ge(n.shape,o);return u.length>0?O(mt(a,u),n.shape):a}}}};var jA={kernelName:vs,inputsToSave:["a","b"],gradFunc:(r,t)=>{let[e,n]=t,o=Lt(e.shape,n.shape);return{a:()=>{let a=P(r,Z(n,"float32")),u=ge(e.shape,o);return u.length>0?O(mt(a,u),e.shape):a},b:()=>{let a=P(r,Z(e,"float32")),u=ge(n.shape,o);return u.length>0?O(mt(a,u),n.shape):a}}}};var XA={kernelName:wi,gradFunc:r=>({x:()=>Zt(r)})};var YA={kernelName:Cs,inputsToSave:["indices"],gradFunc:(r,t)=>{let e=t[0];return{indices:()=>ye(e.shape,"float32")}}};var ZA={kernelName:vi,gradFunc:r=>({x:()=>St(r)})};var JA={kernelName:Ci,saveAllInputs:!0,gradFunc:(r,t,e)=>{let{axis:n}=e;return vr(r,n).map(s=>()=>s)}};var uC={kernelName:Is,inputsToSave:["x"],gradFunc:(r,t,e)=>{let n=t[0],{paddings:o}=e,s=o.map(i=>i[0]);return{x:()=>Rt(r,s,n.shape)}}};var QA={kernelName:Ss,inputsToSave:["a","b"],outputsToSave:[!0],gradFunc:(r,t)=>{let[e,n,o]=t,s=e,i=n,a=Lt(s.shape,i.shape);return{a:()=>{let c=Z(i,"float32"),p=P(r,P(c,Kr(s,lt(c,pt(1))))),m=ge(s.shape,a);return m.length>0&&(p=mt(p,m)),O(p,s.shape)},b:()=>{let c=Ge(s,0),p=Ne(c,wr(s),St(s)),m=P(r,P(o,p)),f=ge(i.shape,a);return f.length>0&&(m=mt(m,f)),O(m,i.shape)}}}};var t2={kernelName:ks,inputsToSave:["x","alpha"],gradFunc:(r,t)=>{let[e,n]=t,o=Ge(e,0);return{x:()=>Ne(o,r,P(r,n)),alpha:()=>{let s=Ne(o,St(r),P(r,e)),i=ge(n.shape,r.shape);return i.length>0&&(s=mt(s,i)),O(s,n.shape)}}}};var e2={kernelName:ss,inputsToSave:["a","b"],gradFunc:(r,t)=>{let[e,n]=t,o=Lt(e.shape,n.shape);return{a:()=>{let a=ct(r,Z(n,"float32")),u=ge(e.shape,o);return u.length>0?O(mt(a,u),e.shape):a},b:()=>{let a=P(r,Z(e,"float32")),u=ge(n.shape,o);u.length>0&&(a=O(mt(a,u),n.shape));let l=Wt(n);return Zt(ct(a,Z(l,"float32")))}}}};var r2={kernelName:ka,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>ct(r,Zt(Wt(e)))}}};var n2={kernelName:_s,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t,n=P(Nn(e,6),Qs(e));return{x:()=>P(r,Z(n,"float32"))}}};var o2={kernelName:Ns,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>P(r,Z(Qs(e),"float32"))}}};var s2={kernelName:Ii,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>O(r,e.shape)}}};var i2={kernelName:Ts,inputsToSave:["images"],gradFunc:(r,t,e)=>{let[n]=t,o={dy:r,images:n};return{images:()=>_.runKernel(Rp,o,e)}}};var a2={kernelName:Al,inputsToSave:["images"],gradFunc:(r,t,e)=>{let[n]=t,o={dy:r,images:n};return{images:()=>_.runKernel($p,o,e)}}};var l2={kernelName:Es,gradFunc:(r,t,e)=>{let{dims:n}=e,o=mr(n,r.shape);return{x:()=>sr(r,o)}}};var u2={kernelName:As,gradFunc:r=>({x:()=>St(r)})};var c2={kernelName:Ds,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>Zt(ct(r,P(Kr(e,1.5),2)))}}};var p2={kernelName:Si,inputsToSave:["condition"],gradFunc:(r,t)=>{let[e]=t;return{condition:()=>Z(St(e),"float32"),t:()=>P(r,Z(e,r.dtype)),e:()=>P(r,Z(Ka(e),r.dtype))}}};var m2={kernelName:Ta,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>{let n=Ge(e,pt(0)),o=pt(oC),s=pt(sC),i=P(r,s),a=P(P(r,o),Ze(Z(e,"float32")));return Ne(n,i,a)}}}};var f2={kernelName:Rs,outputsToSave:[!0],gradFunc:(r,t)=>{let[e]=t;return{x:()=>P(r,P(e,lt(pt(1),e)))}}};var d2={kernelName:Ea,gradFunc:r=>({x:()=>St(r)})};var h2={kernelName:$s,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>P(Ua(Z(e,"float32")),r)}}};var g2={kernelName:_a,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>P(Lu(Z(e,"float32")),r)}}};var x2={kernelName:ki,inputsToSave:["x"],gradFunc:(r,t,e)=>{let[n]=t,{begin:o,size:s}=e,i=n.shape,[a,u]=f0(n,o,s),l=[];for(let c=0;c<r.rank;c++)l.push([a[c],i[c]-a[c]-u[c]]);return{x:()=>qr(r,l)}}};var y2={kernelName:Ps,outputsToSave:[!0],gradFunc:(r,t,e)=>{let[n]=t,{dim:o}=e,s=!0,i=P(r,n);return{logits:()=>lt(i,P(mt(i,[o],s),n))}}};var b2={kernelName:Aa,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>P(r,Lr(e))}}};var cC={kernelName:Ni,gradFunc:(r,t,e)=>{let{blockShape:n,paddings:o}=e;return{x:()=>Ga(r,n,o)}}};var pC={kernelName:Ti,gradFunc:(r,t,e)=>{let{axis:n}=e;return{x:()=>se(r,n)}}};var w2={kernelName:Fs,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>ct(r,P(be(Z(e,"float32")),2))}}};var v2={kernelName:Fl,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>P(r,P(Z(e,"float32"),2))}}};var C2={kernelName:Ms,inputsToSave:["a","b"],gradFunc:(r,t)=>{let[e,n]=t,o=pt(2);return{a:()=>P(r,P(o,lt(e,n))),b:()=>P(r,P(o,lt(n,e)))}}};var I2={kernelName:ao,gradFunc:r=>({x:()=>St(r)})};var S2={kernelName:Ls,inputsToSave:["a","b"],gradFunc:(r,t)=>{let[e,n]=t,o=Lt(e.shape,n.shape);return{a:()=>{let a=r,u=ge(e.shape,o);return u.length>0&&(a=mt(a,u)),O(a,e.shape)},b:()=>{let a=r,u=ge(n.shape,o);return u.length>0&&(a=mt(a,u)),O(Zt(a),n.shape)}}}};var k2={kernelName:Os,inputsToSave:["x"],gradFunc:(r,t,e)=>{let[n]=t,o=n.shape.slice(),{axis:s}=e;mr(s,n.shape).forEach(l=>{o[l]=1});let a=O(r,o),u=P(a,lr(n.shape,"float32"));return{x:()=>u}}};var N2={kernelName:zs,inputsToSave:["x"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>ct(r,Wt(Ua(e)))}}};var T2={kernelName:Bs,outputsToSave:[!0],gradFunc:(r,t)=>{let[e]=t;return{x:()=>P(lt(pt(1),Wt(e)),r)}}};var _2={kernelName:jn,inputsToSave:["x"],gradFunc:(r,t,e)=>{let[n]=t,{reps:o}=e;return{x:()=>{let i=St(n);if(n.rank===1)for(let a=0;a<o[0];++a)i=J(i,Rt(r,[a*n.shape[0]],[n.shape[0]]));else if(n.rank===2)for(let a=0;a<o[0];++a)for(let u=0;u<o[1];++u)i=J(i,Rt(r,[a*n.shape[0],u*n.shape[1]],[n.shape[0],n.shape[1]]));else if(n.rank===3)for(let a=0;a<o[0];++a)for(let u=0;u<o[1];++u)for(let l=0;l<o[2];++l)i=J(i,Rt(r,[a*n.shape[0],u*n.shape[1],l*n.shape[2]],[n.shape[0],n.shape[1],n.shape[2]]));else if(n.rank===4)for(let a=0;a<o[0];++a)for(let u=0;u<o[1];++u)for(let l=0;l<o[2];++l)for(let c=0;c<o[3];++c)i=J(i,Rt(r,[a*n.shape[0],u*n.shape[1],l*n.shape[2],c*n.shape[3]],[n.shape[0],n.shape[1],n.shape[2],n.shape[3]]));else throw new Error(`Gradient for tile operation is not implemented for rank-${n.rank} tensors yet.`);return i}}}};var E2={kernelName:Vs,gradFunc:(r,t,e)=>{let n=e,{perm:o}=n,s=Zf(o);return{x:()=>Gt(r,s)}}};var A2={kernelName:_i,gradFunc:(r,t,e)=>{let n=e,{axis:o}=n;return{value:()=>Je(r,o)}}};var D2={kernelName:Ol,inputsToSave:["segmentIds"],gradFunc:(r,t)=>{let[e]=t;return{x:()=>J5(r,e)}}};function J5(r,t){let e=cn(t,St(t)),n=go(r,e),o=kn(t,pt(0,"int32")),s=n.rank-o.rank;for(let a=0;a<s;++a)o=fr(o,a+1);o=Nr(o,lr(n.shape,"bool"));let i=St(n);return Ne(o,n,i)}var $2={kernelName:Ei,gradFunc:r=>({x:()=>St(r)})};var Q5=[Vx,LE,zE,BE,VE,GE,WE,UE,HE,qE,KE,jE,YE,JE,QE,tA,eA,rA,nA,oA,sA,iA,lA,aA,cA,pA,mA,fA,dA,hA,e2,gA,xA,yA,bA,wA,CA,vA,IA,NA,TA,_A,EA,AA,DA,$A,RA,FA,OA,MA,lC,lC,LA,BA,GA,WA,UA,HA,qA,KA,jA,XA,YA,ZA,JA,uC,uC,QA,t2,r2,n2,o2,s2,i2,a2,l2,u2,c2,p2,m2,f2,d2,h2,g2,x2,y2,b2,cC,cC,pC,pC,w2,C2,v2,I2,S2,k2,N2,T2,_2,E2,A2,D2,$2];for(let r of Q5)QT(r);L().prototype.abs=function(){return this.throwIfDisposed(),Se(this)};L().prototype.acos=function(){return this.throwIfDisposed(),Df(this)};L().prototype.acosh=function(){return this.throwIfDisposed(),$f(this)};L().prototype.add=function(r){return this.throwIfDisposed(),J(this,r)};L().prototype.all=function(r,t){return this.throwIfDisposed(),Ru(this,r,t)};L().prototype.any=function(r,t){return this.throwIfDisposed(),Gl(this,r,t)};L().prototype.argMax=function(r){return this.throwIfDisposed(),Hs(this,r)};L().prototype.argMin=function(r){return this.throwIfDisposed(),Rf(this,r)};L().prototype.asScalar=function(){return this.throwIfDisposed(),A(this.size===1,()=>"The array must have only 1 element."),O(this,[])};L().prototype.asType=function(r){return this.throwIfDisposed(),Z(this,r)};L().prototype.as1D=function(){return this.throwIfDisposed(),O(this,[this.size])};L().prototype.as2D=function(r,t){return this.throwIfDisposed(),O(this,[r,t])};L().prototype.as3D=function(r,t,e){return this.throwIfDisposed(),O(this,[r,t,e])};L().prototype.as4D=function(r,t,e,n){return this.throwIfDisposed(),O(this,[r,t,e,n])};L().prototype.as5D=function(r,t,e,n,o){return this.throwIfDisposed(),O(this,[r,t,e,n,o])};L().prototype.asin=function(){return this.throwIfDisposed(),Ff(this)};L().prototype.asinh=function(){return this.throwIfDisposed(),Of(this)};L().prototype.atan=function(){return this.throwIfDisposed(),Pf(this)};L().prototype.atan2=function(r){return this.throwIfDisposed(),Mf(this,r)};L().prototype.atanh=function(){return this.throwIfDisposed(),Lf(this)};L().prototype.avgPool=function(r,t,e,n){return this.throwIfDisposed(),Va(this,r,t,e,n)};L().prototype.batchToSpaceND=function(r,t){return this.throwIfDisposed(),Ga(this,r,t)};L().prototype.batchNorm=function(r,t,e,n,o){return this.throwIfDisposed(),ho(this,r,t,e,n,o)};L().prototype.broadcastTo=function(r){return this.throwIfDisposed(),Wa(this,r)};L().prototype.cast=function(r){return this.throwIfDisposed(),Z(this,r)};L().prototype.ceil=function(){return this.throwIfDisposed(),Vf(this)};L().prototype.clipByValue=function(r,t){return this.throwIfDisposed(),br(this,r,t)};L().prototype.concat=function(r,t){return this.throwIfDisposed(),r instanceof Vt&&(r=[r]),se([this,...r],t)};L().prototype.conv1d=function(r,t,e,n,o,s){return this.throwIfDisposed(),Pu(this,r,t,e,n,o,s)};L().prototype.conv2dTranspose=function(r,t,e,n,o){return this.throwIfDisposed(),Mu(this,r,t,e,n,o)};L().prototype.conv2d=function(r,t,e,n,o,s){return this.throwIfDisposed(),ln(this,r,t,e,n,o,s)};L().prototype.cos=function(){return this.throwIfDisposed(),Ua(this)};L().prototype.cosh=function(){return this.throwIfDisposed(),Lu(this)};L().prototype.cumprod=function(r,t,e){return this.throwIfDisposed(),Wf(this,r,t,e)};L().prototype.cumsum=function(r,t,e){return this.throwIfDisposed(),zu(this,r,t,e)};L().prototype.depthToSpace=function(r,t){return this.throwIfDisposed(),Uf(this,r,t)};L().prototype.depthwiseConv2d=function(r,t,e,n,o,s){return this.throwIfDisposed(),Ks(this,r,t,e,n,o,s)};L().prototype.dilation2d=function(r,t,e,n,o){return this.throwIfDisposed(),Hf(this,r,t,e,n,o)};L().prototype.divNoNan=function(r){return this.throwIfDisposed(),qf(this,r)};L().prototype.div=function(r){return this.throwIfDisposed(),ct(this,r)};L().prototype.dot=function(r){return this.throwIfDisposed(),D0(this,r)};L().prototype.elu=function(){return this.throwIfDisposed(),js(this)};L().prototype.equal=function(r){return this.throwIfDisposed(),Sr(this,r)};L().prototype.erf=function(){return this.throwIfDisposed(),Kf(this)};L().prototype.exp=function(){return this.throwIfDisposed(),Ze(this)};L().prototype.expandDims=function(r){return this.throwIfDisposed(),fr(this,r)};L().prototype.expm1=function(){return this.throwIfDisposed(),jf(this)};L().prototype.fft=function(){return this.throwIfDisposed(),Qa(this)};L().prototype.flatten=function(){return this.throwIfDisposed(),O(this,[this.size])};L().prototype.floor=function(){return this.throwIfDisposed(),Ys(this)};L().prototype.floorDiv=function(r){return this.throwIfDisposed(),$u(this,r)};L().prototype.gather=function(r,t){return this.throwIfDisposed(),go(this,r,t)};L().prototype.greaterEqual=function(r){return this.throwIfDisposed(),kn(this,r)};L().prototype.greater=function(r){return this.throwIfDisposed(),Ge(this,r)};L().prototype.ifft=function(){return this.throwIfDisposed(),Oi(this)};L().prototype.irfft=function(){return this.throwIfDisposed(),Ju(this)};L().prototype.isFinite=function(){return this.throwIfDisposed(),R0(this)};L().prototype.isInf=function(){return this.throwIfDisposed(),F0(this)};L().prototype.isNaN=function(){return this.throwIfDisposed(),Xf(this)};L().prototype.leakyRelu=function(r){return this.throwIfDisposed(),Ha(this,r)};L().prototype.lessEqual=function(r){return this.throwIfDisposed(),Nn(this,r)};L().prototype.less=function(r){return this.throwIfDisposed(),Vu(this,r)};L().prototype.localResponseNormalization=function(r,t,e,n){return this.throwIfDisposed(),Yf(this,r,t,e,n)};L().prototype.logSigmoid=function(){return this.throwIfDisposed(),P0(this)};L().prototype.logSoftmax=function(r){return this.throwIfDisposed(),Gu(this,r)};L().prototype.logSumExp=function(r,t){return this.throwIfDisposed(),Jf(this,r,t)};L().prototype.log=function(){return this.throwIfDisposed(),wr(this)};L().prototype.log1p=function(){return this.throwIfDisposed(),qa(this)};L().prototype.logicalAnd=function(r){return this.throwIfDisposed(),Nr(this,r)};L().prototype.logicalNot=function(){return this.throwIfDisposed(),Ka(this)};L().prototype.logicalOr=function(r){return this.throwIfDisposed(),Wu(this,r)};L().prototype.logicalXor=function(r){return this.throwIfDisposed(),V0(this,r)};L().prototype.matMul=function(r,t,e){return this.throwIfDisposed(),Mt(this,r,t,e)};L().prototype.maxPool=function(r,t,e,n){return this.throwIfDisposed(),ja(this,r,t,e,n)};L().prototype.max=function(r,t){return this.throwIfDisposed(),$r(this,r,t)};L().prototype.maximum=function(r){return this.throwIfDisposed(),cn(this,r)};L().prototype.mean=function(r,t){return this.throwIfDisposed(),xe(this,r,t)};L().prototype.min=function(r,t){return this.throwIfDisposed(),Wl(this,r,t)};L().prototype.minimum=function(r){return this.throwIfDisposed(),Zs(this,r)};L().prototype.mirrorPad=function(r,t){return this.throwIfDisposed(),td(this,r,t)};L().prototype.mod=function(r){return this.throwIfDisposed(),ed(this,r)};L().prototype.mul=function(r){return this.throwIfDisposed(),P(this,r)};L().prototype.neg=function(){return this.throwIfDisposed(),Zt(this)};L().prototype.norm=function(r,t,e){return this.throwIfDisposed(),nm(this,r,t,e)};L().prototype.notEqual=function(r){return this.throwIfDisposed(),bo(this,r)};L().prototype.oneHot=function(r,t=1,e=0){return this.throwIfDisposed(),Us(this,r,t,e)};L().prototype.onesLike=function(){return this.throwIfDisposed(),dr(this)};L().prototype.pad=function(r,t){return this.throwIfDisposed(),qr(this,r,t)};L().prototype.pool=function(r,t,e,n,o,s){return this.throwIfDisposed(),U0(this,r,t,e,n,o,s)};L().prototype.pow=function(r){return this.throwIfDisposed(),Kr(this,r)};L().prototype.prelu=function(r){return this.throwIfDisposed(),Ya(this,r)};L().prototype.prod=function(r,t){return this.throwIfDisposed(),Uu(this,r,t)};L().prototype.reciprocal=function(){return this.throwIfDisposed(),rd(this)};L().prototype.relu=function(){return this.throwIfDisposed(),Tr(this)};L().prototype.relu6=function(){return this.throwIfDisposed(),qu(this)};L().prototype.reshapeAs=function(r){return this.throwIfDisposed(),O(this,r.shape)};L().prototype.reshape=function(r){return this.throwIfDisposed(),O(this,r)};L().prototype.resizeBilinear=function(r,t,e){return this.throwIfDisposed(),Mx(this,r,t,e)};L().prototype.resizeNearestNeighbor=function(r,t,e){return this.throwIfDisposed(),Lx(this,r,t,e)};L().prototype.reverse=function(r){return this.throwIfDisposed(),sr(this,r)};L().prototype.rfft=function(){return this.throwIfDisposed(),tl(this)};L().prototype.round=function(){return this.throwIfDisposed(),Ku(this)};L().prototype.rsqrt=function(){return this.throwIfDisposed(),ju(this)};L().prototype.selu=function(){return this.throwIfDisposed(),Xu(this)};L().prototype.separableConv2d=function(r,t,e,n,o,s){return this.throwIfDisposed(),nd(this,r,t,e,n,o,s)};L().prototype.sigmoid=function(){return this.throwIfDisposed(),Lr(this)};L().prototype.sign=function(){return this.throwIfDisposed(),od(this)};L().prototype.sin=function(){return this.throwIfDisposed(),Yu(this)};L().prototype.sinh=function(){return this.throwIfDisposed(),Zu(this)};L().prototype.slice=function(r,t){return this.throwIfDisposed(),Rt(this,r,t)};L().prototype.softmax=function(r){return this.throwIfDisposed(),Ja(this,r)};L().prototype.softplus=function(){return this.throwIfDisposed(),xo(this)};L().prototype.spaceToBatchND=function(r,t){return this.throwIfDisposed(),Xa(this,r,t)};L().prototype.split=function(r,t){return this.throwIfDisposed(),ur(this,r,t)};L().prototype.sqrt=function(){return this.throwIfDisposed(),be(this)};L().prototype.square=function(){return this.throwIfDisposed(),Wt(this)};L().prototype.squaredDifference=function(r){return this.throwIfDisposed(),Qu(this,r)};L().prototype.squeeze=function(r){return this.throwIfDisposed(),zr(this,r)};L().prototype.stack=function(r,t){this.throwIfDisposed();let e=r instanceof Vt?[this,r]:[this,...r];return Je(e,t)};L().prototype.step=function(r){return this.throwIfDisposed(),Qs(this,r)};L().prototype.stridedSlice=function(r,t,e,n,o,s,i,a){return this.throwIfDisposed(),ad(this,r,t,e,n,o,s,i,a)};L().prototype.sub=function(r){return this.throwIfDisposed(),lt(this,r)};L().prototype.sum=function(r,t){return this.throwIfDisposed(),mt(this,r,t)};L().prototype.tan=function(){return this.throwIfDisposed(),ld(this)};L().prototype.tanh=function(){return this.throwIfDisposed(),qs(this)};L().prototype.tile=function(r){return this.throwIfDisposed(),kr(this,r)};L().prototype.toBool=function(){return this.throwIfDisposed(),Z(this,"bool")};L().prototype.toFloat=function(){return this.throwIfDisposed(),Z(this,"float32")};L().prototype.toInt=function(){return this.throwIfDisposed(),Z(this,"int32")};L().prototype.topk=function(r,t){return this.throwIfDisposed(),ud(this,r,t)};L().prototype.transpose=function(r){return this.throwIfDisposed(),Gt(this,r)};L().prototype.unique=function(r){return this.throwIfDisposed(),rm(this,r)};L().prototype.unsortedSegmentSum=function(r,t){return this.throwIfDisposed(),cd(this,r,t)};L().prototype.unstack=function(r){return this.throwIfDisposed(),vr(this,r)};L().prototype.where=function(r,t){return this.throwIfDisposed(),Ne(r,this,t)};L().prototype.zerosLike=function(){return this.throwIfDisposed(),St(this)};var L2={};Yt(L2,{maxNorm:()=>eX,minMaxNorm:()=>oX,nonNeg:()=>nX,unitNorm:()=>rX});var mC;function cr(){return mC==null&&(mC=k_().epsilon()),mC}function mn(){return"channelsLast"}var Ln=class extends Error{constructor(t){super(t);Object.setPrototypeOf(this,Ln.prototype)}},Xr=class extends Error{constructor(t){super(t);Object.setPrototypeOf(this,Xr.prototype)}},z=class extends Error{constructor(t){super(t);Object.setPrototypeOf(this,z.prototype)}},Tt=class extends Error{constructor(t){super(t);Object.setPrototypeOf(this,Tt.prototype)}},dd=class extends Error{constructor(t){super(t);Object.setPrototypeOf(this,dd.prototype)}};function Io(r,t){if(Array.isArray(r)){let e=[];for(let n=0;n<t;n++)e=e.concat(r);return e}else{let e=new Array(t);return e.fill(r),e}}function Zn(r,t){if(!r)throw new dd(t)}function dC(r,t){let e=0;for(let n of r)n===t&&e++;return e}function _r(r){return r.length===1?r[0]:r}function we(r){return Array.isArray(r)?r:[r]}function So(r){let e=r.replace(/(.)([A-Z][a-z0-9]+)/g,"$1_$2").replace(/([a-z])([A-Z])/g,"$1_$2").toLowerCase();return e[0]!=="_"?e:"private"+e}function rl(r){return r.length<=1||r.indexOf("_")===-1?r:r.replace(/[_]+(\w|$)/g,(t,e)=>e.toUpperCase())}var Co={};function sm(r){if(r==null)return null;let t={};return t.className=r.getClassName(),t.config=r.getConfig(),t}function fC(r){if(!(r==null||typeof r!="object"))if(Array.isArray(r))r.forEach(t=>fC(t));else{let t=Object.keys(r);for(let e of t){let n=r[e];n!=null&&typeof n=="object"&&(!Array.isArray(n)&&n.type==="ndarray"&&typeof n.value=="number"?r[e]=n.value:fC(n))}}}function Li(r,t={},e={},n="object",o=!1){if(typeof r=="string"){let s=r,i;if(s in e)i=e[s];else if(s in Co)i=Co[s];else if(i=t[s],i==null)throw new z(`Unknown ${n}: ${r}. This may be due to one of the following reasons:
1. The ${n} is defined in Python, in which case it needs to be ported to TensorFlow.js or your JavaScript code.
2. The custom ${n} is defined in JavaScript, but is not registered properly with tf.serialization.registerClass().`);return i}else{let s=r;if(s.className==null||s.config==null)throw new z(`${n}: Improper config format: ${JSON.stringify(s)}.
'className' and 'config' must set.`);let i=s.className,a,u;if(i in e?[a,u]=e[i]:i in Co?[a,u]=Co.className:i in t&&([a,u]=t[i]),a==null)throw new z(`Unknown ${n}: ${i}. This may be due to one of the following reasons:
1. The ${n} is defined in Python, in which case it needs to be ported to TensorFlow.js or your JavaScript code.
2. The custom ${n} is defined in JavaScript, but is not registered properly with tf.serialization.registerClass().`);if(u!=null){let l={};for(let f of Object.keys(Co))l[f]=Co[f];for(let f of Object.keys(e))l[f]=e[f];let c=s.config;c.customObjects=l;let p=Object.assign({},Co);for(let f of Object.keys(e))Co[f]=e[f];fC(s.config);let m=u(a,s.config,e,o);return Co=Object.assign({},p),m}else{let l=Object.assign({},Co);for(let p of Object.keys(e))Co[p]=e[p];let c=new a(s.config);return Co=Object.assign({},l),c}}}function tX(r,t){return r<t?-1:r>t?1:0}function hd(r,t){return-1*tX(r,t)}function ko(r){if(r==null)return r;let t=[];for(let e of r)t.indexOf(e)===-1&&t.push(e);return t}function R2(r){if(r==null)throw new z(`Invalid value in obj: ${JSON.stringify(r)}`);for(let t in r)if(r.hasOwnProperty(t))return!1;return!0}function zi(r,t,e){if(e!=null&&r.indexOf(e)<0)throw new z(`${e} is not a valid ${t}. Valid values are ${r} or null/undefined.`)}function Wx(r,t,e=0,n=1/0){return Zn(e>=0),Zn(n>=e),Array.isArray(r)&&r.length>=e&&r.length<=n&&r.every(o=>typeof o===t)}function er(r,t){Array.isArray(r)?(x.assert(r.length>0,()=>`${t} is unexpectedly an empty array.`),r.forEach((e,n)=>er(e,`element ${n+1} of ${t}`))):x.assert(Number.isInteger(r)&&r>0,()=>`Expected ${t} to be a positive integer, but got ${F2(r)}.`)}function F2(r){return r===null?"null":Array.isArray(r)?"["+r.map(t=>F2(t)).join(",")+"]":typeof r=="string"?`"${r}"`:`${r}`}function O2(r,t,e){let n=e!=null?e():x.now(),o;return(...i)=>{let a=e!=null?e():x.now();return a-n<t||(n=a,o=r(...i)),o}}function Ux(r){return r==="relu"?"relu":r==="linear"?"linear":r==="elu"?"elu":null}function hC(r,t){return B(()=>be(mt(P(r,r),t,!0)))}var im=class extends et.Serializable{getConfig(){return{}}},gd=class extends im{constructor(t){super();this.defaultMaxValue=2,this.defaultAxis=0,this.maxValue=t.maxValue!=null?t.maxValue:this.defaultMaxValue,this.axis=t.axis!=null?t.axis:this.defaultAxis}apply(t){return B(()=>{let e=hC(t,this.axis),n=br(e,0,this.maxValue);return P(t,ct(n,J(cr(),e)))})}getConfig(){return{maxValue:this.maxValue,axis:this.axis}}};gd.className="MaxNorm";et.registerClass(gd);var xd=class extends im{constructor(t){super();this.defaultAxis=0,this.axis=t.axis!=null?t.axis:this.defaultAxis}apply(t){return B(()=>ct(t,J(cr(),hC(t,this.axis))))}getConfig(){return{axis:this.axis}}};xd.className="UnitNorm";et.registerClass(xd);var yd=class extends im{apply(t){return Tr(t)}};yd.className="NonNeg";et.registerClass(yd);var bd=class extends im{constructor(t){super();this.defaultMinValue=0,this.defaultMaxValue=1,this.defaultRate=1,this.defaultAxis=0,this.minValue=t.minValue!=null?t.minValue:this.defaultMinValue,this.maxValue=t.maxValue!=null?t.maxValue:this.defaultMaxValue,this.rate=t.rate!=null?t.rate:this.defaultRate,this.axis=t.axis!=null?t.axis:this.defaultAxis}apply(t){return B(()=>{let e=hC(t,this.axis),n=J(P(this.rate,br(e,this.minValue,this.maxValue)),P(1-this.rate,e));return P(t,ct(n,J(cr(),e)))})}getConfig(){return{minValue:this.minValue,maxValue:this.maxValue,rate:this.rate,axis:this.axis}}};bd.className="MinMaxNorm";et.registerClass(bd);var P2={maxNorm:"MaxNorm",minMaxNorm:"MinMaxNorm",nonNeg:"NonNeg",unitNorm:"UnitNorm"};function We(r){return sm(r)}function M2(r,t={}){return Li(r,et.SerializationMap.getMap().classNameMap,t,"constraint")}function Ue(r){if(r==null)return null;if(typeof r=="string"){let e={className:r in P2?P2[r]:r,config:{}};return M2(e)}else return r instanceof im?r:M2(r)}function eX(r){return new gd(r)}function rX(r){return new xd(r)}function nX(){return new yd}function oX(r){return new bd(r)}var oD={};Yt(oD,{constant:()=>mX,glorotNormal:()=>bX,glorotUniform:()=>yX,heNormal:()=>wX,heUniform:()=>vX,identity:()=>gX,leCunNormal:()=>CX,leCunUniform:()=>IX,ones:()=>pX,orthogonal:()=>SX,randomNormal:()=>dX,randomUniform:()=>fX,truncatedNormal:()=>hX,varianceScaling:()=>xX,zeros:()=>cX});var z2=["channelsFirst","channelsLast"],B2=["nearest","bilinear"],V2=["valid","same","causal"],G2=["max","avg"],W2=["sum","mul","concat","ave"];var am=new Map;function Le(r){zi(z2,"DataFormat",r)}function H2(r){zi(B2,"InterpolationFormat",r)}function fn(r){zi(V2,"PaddingMode",r)}function gC(r){zi(G2,"PoolMode",r)}var wd=[],U2="/";function ei(r,t){wd.push(r);try{let e=t();return wd.pop(),e}catch(e){throw wd.pop(),e}}function sX(){return wd.length===0?"":wd.join(U2)+U2}function Hx(r){if(!q2(r))throw new Error("Not a valid tensor name: '"+r+"'");return sX()+r}function qx(r){if(!q2(r))throw new Error("Not a valid tensor name: '"+r+"'");am.has(r)||am.set(r,0);let t=am.get(r);if(am.set(r,am.get(r)+1),t>0){let e=`${r}_${t}`;return am.set(e,1),e}else return r}var iX=new RegExp(/^[A-Za-z0-9][-A-Za-z0-9\._\/]*$/);function q2(r){return!!r.match(iX)}function K2(r){return r===parseInt(r.toString(),10)}function No(r,t,e){t==null&&(t=0),e==null&&(e=r.length);let n=1;for(let o=t;o<e;++o)n*=r[o];return n}function mc(r){if(r.length===0)return Number.NaN;let t=Number.POSITIVE_INFINITY;for(let e=0;e<r.length;e++){let n=r[e];n<t&&(t=n)}return t}function ri(r){if(r.length===0)return Number.NaN;let t=Number.NEGATIVE_INFINITY;for(let e=0;e<r.length;e++){let n=r[e];n>t&&(t=n)}return t}function Yr(r,t){if(t<r)throw new z(`end (${t}) < begin (${r}) is forbidden.`);let e=[];for(let n=r;n<t;++n)e.push(n);return e}function fc(r,t){return Z(r,t)}function ol(r,t=-1){let e=r.shape.slice();return t<0&&(t=e.length+t+1),e.splice(t,0,1),O(r,e)}function j2(r,t){return B(()=>{if(r.shape.length!==2)throw new z(`repeat() expects a rank-2 tensor, but received a rank-${r.shape.length} tensor.`);let e=ol(r,1);return jx(e,[1,t,1])})}function X2(r){let t=[No(r.shape)];return O(r,t)}function Y2(r){if(r.rank<=1)throw new z(`batchFlatten requires a minimum rank of 2. Got rank: ${r.rank}.`);let t=[r.shape[0],No(r.shape,1)];return O(r,t)}function nl(r,t,e){return B(()=>{switch(r.rank){case 1:return sd(r,t,e);case 2:return Nx(r,[t,0],[e,r.shape[1]]);case 3:return id(r,[t,0,0],[e,r.shape[1],r.shape[2]]);case 4:return em(r,[t,0,0,0],[e,r.shape[1],r.shape[2],r.shape[3]]);case 5:return Rt(r,[t,0,0,0,0],[e,r.shape[1],r.shape[2],r.shape[3],r.shape[4]]);case 6:return Rt(r,[t,0,0,0,0,0],[e,r.shape[1],r.shape[2],r.shape[3],r.shape[4],r.shape[5]]);default:throw new z(`sliceAlongFirstAxis() received an unsupported tensor rank: ${r.rank}`)}})}function xC(r,t,e){return B(()=>{switch(r.rank){case 1:return sd(r,t,e);case 2:return Nx(r,[0,t],[r.shape[0],e]);case 3:return id(r,[0,0,t],[r.shape[0],r.shape[1],e]);case 4:return em(r,[0,0,0,t],[r.shape[0],r.shape[1],r.shape[2],e]);default:throw new z(`sliceAlongLastAxis() received an unsupported tensor rank: ${r.rank}`)}})}function vd(r,t,e,n){return B(()=>{switch(r.rank){case 1:return sd(r,t,e);case 2:switch(n){case 1:return nl(r,t,e);case 2:return xC(r,t,e);default:throw new z(`The axis is not within the rank of the tensor ${n}`)}case 3:switch(n){case 1:return nl(r,t,e);case 2:return id(r,[0,t,0],[r.shape[0],e,r.shape[2]]);case 3:return xC(r,t,e);default:throw new z(`The axis is not within the rank of the tensor ${n}`)}case 4:switch(n){case 1:return nl(r,t,e);case 2:return em(r,[0,t,0,0],[r.shape[0],e,r.shape[2],r.shape[3]]);case 3:return em(r,[0,0,t,0],[r.shape[0],r.shape[1],e,r.shape[3]]);case 4:return xC(r,t,e);default:throw new z(`The axis is not within the rank of the tensor ${n}`)}default:throw new z(`sliceAlongLastAxis() received an unsupported tensor rank: ${r.rank}`)}})}function lm(r,t=-1){let e;return t<0&&(e=r[0].rank,e!==0?t=e:t=0),t===r[0].rank&&(t=-1),se(r,t)}function bC(r,t){switch(r.rank){case 1:return k0([r,t]);case 2:return N0([r,t],0);case 3:return T0([r,t],0);case 4:return _0([r,t],0);default:throw new z(`concatAlongFirstAxis() received an unsupported tensor rank: ${r.rank}`)}}function jx(r,t){if(Array.isArray(t)||(t=[t]),r.rank!==t.length)throw new z(`The length of input n (${t.length}) does not match the number of dimensions in input x (${r.rank})`);return kr(r,t)}function um(r,t=0,e=1,n,o){return kx(r,t,e,n,o)}function To(r,t,e,n){if(r.rank<2||t.rank<2)throw new Tt(`dot requires both inputs to be rank >= 2 but got x shape = ${r.shape} and y shape = ${t.shape}`);if(t.rank>=3){let o=r.shape.slice(-1)[0],s=t.shape.slice(-2)[0];if(o!==s)throw new Tt(`If rank y >= 3, then the second last dim of y must equal the last dim of x but got x shape = ${r.shape} and y shape = ${t.shape}`)}if(r.rank===2&&t.rank===2)return wo.matMul({a:r,b:t,transposeA:!1,transposeB:!1,bias:n?yC(r.rank,n,mn()):null,activation:e});{let o=r.shape.slice(),s=o.pop();r=O(r,[-1,s]);let i=t.shape.slice(),a=i.pop(),u=i.pop(),l=[...i,a],c=Array.from({length:t.rank},(d,h)=>h===0?t.rank-2:h<=t.rank-2?h-1:h);t=O(Gt(t,c),[u,-1]);let p=[...o,...l],m=!1,f=!1;return O(wo.matMul({a:r,b:t,transposeA:m,transposeB:f,bias:n?yC(r.rank,n,mn()):null,activation:e}),p)}}function Xx(r,t,e){return B(()=>(Array.isArray(t)?t=Re(t,"int32"):t=Z(t,"int32"),go(r,t,e)))}function dc(r){return P(r,r)}function yC(r,t,e){let n=t.shape;if(t.rank!==1&&t.rank!==r)throw new z(`Unexpected bias dimensions: ${t.rank}; expected it to be 1 or ${r}`);if(r===5){if(e==="channelsFirst")return n.length===1?O(t,[1,n[0],1,1,1]):O(t,[1,n[3],n[0],n[1],n[2]]);if(e==="channelsLast")return n.length===1?O(t,[1,1,1,1,n[0]]):O(t,[1].concat(n))}else if(r===4){if(e==="channelsFirst")return n.length===1?O(t,[1,n[0],1,1]):O(t,[1,n[2],n[0],n[1]]);if(e==="channelsLast")return n.length===1?O(t,[1,1,1,n[0]]):O(t,[1].concat(n))}else if(r===3){if(e==="channelsFirst")return n.length===1?O(t,[1,n[0],1]):O(t,[1,n[1],n[0]]);if(e==="channelsLast")return n.length===1?O(t,[1,1,n[0]]):O(t,[1].concat(n))}else if(r<3)return t;throw new z(`Unsupported input rank by biasAdd: ${t.rank}`)}function dn(r,t,e){return B(()=>(e==null&&(e=mn()),Le(e),J(r,yC(r.rank,t,e))))}function Z2(r,t=1){if(t!==1)throw new Tt(`Support for alpha values other than 1 (${t}) is not implemented yet.`);return js(r)}function J2(r){return B(()=>ct(r,J(Se(r),1)))}function Yx(r,t,e,n){return B(()=>X_(r,t,e,n))}function Q2(r){return B(()=>{let t=J(.5,P(.2,r));return br(t,0,1)})}function Hl(r,t,e=!1){return e?r():t()}var tD=["fanIn","fanOut","fanAvg"],eD=["normal","uniform","truncatedNormal"];function aX(r){zi(tD,"FanMode",r)}function lX(r){zi(eD,"Distribution",r)}var Tn=class extends et.Serializable{fromConfigUsesCustomObjects(){return!1}getConfig(){return{}}},Cd=class extends Tn{apply(t,e){return ye(t,e)}};Cd.className="Zeros";et.registerClass(Cd);var hc=class extends Tn{apply(t,e){return lr(t,e)}};hc.className="Ones";et.registerClass(hc);var Id=class extends Tn{constructor(t){super();if(typeof t!="object")throw new z(`Expected argument of type ConstantConfig but got ${t}`);if(t.value===void 0)throw new z(`config must have value set but got ${t}`);this.value=t.value}apply(t,e){return B(()=>P(pt(this.value),lr(t,e)))}getConfig(){return{value:this.value}}};Id.className="Constant";et.registerClass(Id);var Sd=class extends Tn{constructor(t){super();this.DEFAULT_MINVAL=-.05,this.DEFAULT_MAXVAL=.05,this.minval=t.minval||this.DEFAULT_MINVAL,this.maxval=t.maxval||this.DEFAULT_MAXVAL,this.seed=t.seed}apply(t,e){return Js(t,this.minval,this.maxval,e)}getConfig(){return{minval:this.minval,maxval:this.maxval,seed:this.seed}}};Sd.className="RandomUniform";et.registerClass(Sd);var kd=class extends Tn{constructor(t){super();this.DEFAULT_MEAN=0,this.DEFAULT_STDDEV=.05,this.mean=t.mean||this.DEFAULT_MEAN,this.stddev=t.stddev||this.DEFAULT_STDDEV,this.seed=t.seed}apply(t,e){if(e=e||"float32",e!=="float32"&&e!=="int32")throw new Tt(`randomNormal does not support dType ${e}.`);return um(t,this.mean,this.stddev,e,this.seed)}getConfig(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}}};kd.className="RandomNormal";et.registerClass(kd);var Nd=class extends Tn{constructor(t){super();this.DEFAULT_MEAN=0,this.DEFAULT_STDDEV=.05,this.mean=t.mean||this.DEFAULT_MEAN,this.stddev=t.stddev||this.DEFAULT_STDDEV,this.seed=t.seed}apply(t,e){if(e=e||"float32",e!=="float32"&&e!=="int32")throw new Tt(`truncatedNormal does not support dType ${e}.`);return tc(t,this.mean,this.stddev,e,this.seed)}getConfig(){return{mean:this.mean,stddev:this.stddev,seed:this.seed}}};Nd.className="TruncatedNormal";et.registerClass(Nd);var Td=class extends Tn{constructor(t){super();this.gain=t.gain!=null?t.gain:1}apply(t,e){return B(()=>{if(t.length!==2||t[0]!==t[1])throw new z("Identity matrix initializer can only be used for 2D square matrices.");return P(this.gain,Jp(t[0]))})}getConfig(){return{gain:this.gain}}};Td.className="Identity";et.registerClass(Td);function uX(r,t="channelsLast"){let e,n;if(Le(t),r.length===2)e=r[0],n=r[1];else if([3,4,5].indexOf(r.length)!==-1){if(t==="channelsFirst"){let o=No(r,2);e=r[1]*o,n=r[0]*o}else if(t==="channelsLast"){let o=No(r,0,r.length-2);e=r[r.length-2]*o,n=r[r.length-1]*o}}else{let o=No(r);e=Math.sqrt(o),n=Math.sqrt(o)}return[e,n]}var Zr=class extends Tn{constructor(t){super();if(t.scale<0)throw new z(`scale must be a positive float. Got: ${t.scale}`);this.scale=t.scale==null?1:t.scale,this.mode=t.mode==null?"fanIn":t.mode,aX(this.mode),this.distribution=t.distribution==null?"normal":t.distribution,lX(this.distribution),this.seed=t.seed}apply(t,e){let n=uX(t),o=n[0],s=n[1],i=this.scale;if(this.mode==="fanIn"?i/=Math.max(1,o):this.mode==="fanOut"?i/=Math.max(1,s):i/=Math.max(1,(o+s)/2),this.distribution==="normal"){let a=Math.sqrt(i);if(e=e||"float32",e!=="float32"&&e!=="int32")throw new Tt(`${this.getClassName()} does not support dType ${e}.`);return tc(t,0,a,e,this.seed)}else{let a=Math.sqrt(3*i);return Js(t,-a,a,e)}}getConfig(){return{scale:this.scale,mode:this.mode,distribution:this.distribution,seed:this.seed}}};Zr.className="VarianceScaling";et.registerClass(Zr);var cm=class extends Zr{constructor(t){super({scale:1,mode:"fanAvg",distribution:"uniform",seed:t==null?null:t.seed})}getClassName(){return Zr.className}};cm.className="GlorotUniform";et.registerClass(cm);var pm=class extends Zr{constructor(t){super({scale:1,mode:"fanAvg",distribution:"normal",seed:t==null?null:t.seed})}getClassName(){return Zr.className}};pm.className="GlorotNormal";et.registerClass(pm);var mm=class extends Zr{constructor(t){super({scale:2,mode:"fanIn",distribution:"normal",seed:t==null?null:t.seed})}getClassName(){return Zr.className}};mm.className="HeNormal";et.registerClass(mm);var fm=class extends Zr{constructor(t){super({scale:2,mode:"fanIn",distribution:"uniform",seed:t==null?null:t.seed})}getClassName(){return Zr.className}};fm.className="HeUniform";et.registerClass(fm);var dm=class extends Zr{constructor(t){super({scale:1,mode:"fanIn",distribution:"normal",seed:t==null?null:t.seed})}getClassName(){return Zr.className}};dm.className="LeCunNormal";et.registerClass(dm);var hm=class extends Zr{constructor(t){super({scale:1,mode:"fanIn",distribution:"uniform",seed:t==null?null:t.seed})}getClassName(){return Zr.className}};hm.className="LeCunNormal";et.registerClass(hm);var _d=class extends Tn{constructor(t){super();if(this.DEFAULT_GAIN=1,this.gain=t.gain==null?this.DEFAULT_GAIN:t.gain,this.seed=t.seed,this.seed!=null)throw new Tt("Random seed is not implemented for Orthogonal Initializer yet.")}apply(t,e){return B(()=>{if(t.length<2)throw new Tt("Shape must be at least 2D.");t[0]*t[1]>2e3&&console.warn(`Orthogonal initializer is being called on a matrix with more than 2000 (${t[0]*t[1]}) elements: Slowness may result.`);let n=t[0]>t[1]?[t[1],t[0]]:t,o=um(n,0,1,"float32"),s=OE.gramSchmidt(o);return t[0]>t[1]&&(s=Gt(s)),P(this.gain,s)})}getConfig(){return{gain:this.gain,seed:this.seed}}};_d.className="Orthogonal";et.registerClass(_d);var rD={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 nD(r,t={}){return Li(r,et.SerializationMap.getMap().classNameMap,t,"initializer")}function Te(r){return sm(r)}function de(r){if(typeof r=="string"){let t=r in rD?rD[r]:r;if(t==="GlorotNormal")return new pm;if(t==="GlorotUniform")return new cm;if(t==="HeNormal")return new mm;if(t==="HeUniform")return new fm;if(t==="LeCunNormal")return new dm;if(t==="LeCunUniform")return new hm;{let e={};return e.className=t,e.config={},nD(e)}}else return r instanceof Tn?r:nD(r)}function cX(){return new Cd}function pX(){return new hc}function mX(r){return new Id(r)}function fX(r){return new Sd(r)}function dX(r){return new kd(r)}function hX(r){return new Nd(r)}function gX(r){return new Td(r)}function xX(r){return new Zr(r)}function yX(r){return new cm(r)}function bX(r){return new pm(r)}function wX(r){return new mm(r)}function vX(r){return new fm(r)}function CX(r){return new dm(r)}function IX(r){return new hm(r)}function SX(r){return new _d(r)}var VD={};Yt(VD,{Layer:()=>Ut,RNN:()=>zn,RNNCell:()=>Ql,activation:()=>Y8,add:()=>sY,alphaDropout:()=>GY,average:()=>iY,averagePooling1d:()=>iI,averagePooling2d:()=>aI,averagePooling3d:()=>lI,avgPool1d:()=>hY,avgPool2d:()=>xY,avgPool3d:()=>bY,avgPooling1d:()=>gY,avgPooling2d:()=>yY,avgPooling3d:()=>wY,batchNormalization:()=>mY,bidirectional:()=>FY,concatenate:()=>aY,conv1d:()=>V8,conv2d:()=>G8,conv2dTranspose:()=>W8,conv3d:()=>U8,conv3dTranspose:()=>H8,convLstm2d:()=>AY,convLstm2dCell:()=>DY,cropping2D:()=>K8,dense:()=>Z8,depthwiseConv2d:()=>X8,dot:()=>pY,dropout:()=>J8,elu:()=>O8,embedding:()=>oY,flatten:()=>tY,gaussianDropout:()=>VY,gaussianNoise:()=>BY,globalAveragePooling1d:()=>vY,globalAveragePooling2d:()=>CY,globalMaxPool1d:()=>PY,globalMaxPool2d:()=>MY,globalMaxPooling1d:()=>MD,globalMaxPooling2d:()=>LD,gru:()=>SY,gruCell:()=>kY,input:()=>$C,inputLayer:()=>F8,layerNormalization:()=>fY,leakyReLU:()=>M8,lstm:()=>NY,lstmCell:()=>TY,masking:()=>WY,maxPool1d:()=>LY,maxPool2d:()=>zY,maxPooling1d:()=>zD,maxPooling2d:()=>BD,maxPooling3d:()=>IY,maximum:()=>lY,minimum:()=>uY,multiply:()=>cY,permute:()=>nY,prelu:()=>L8,reLU:()=>P8,repeatVector:()=>eY,reshape:()=>rY,rnn:()=>$Y,separableConv2d:()=>q8,simpleRNN:()=>_Y,simpleRNNCell:()=>EY,softmax:()=>z8,spatialDropout1d:()=>Q8,stackedRNNCells:()=>RY,thresholdedReLU:()=>B8,timeDistributed:()=>OY,upSampling2d:()=>j8,zeroPadding2d:()=>dY});var kX=0;function Jx(){return kX++}var Zx={};function ql(r=""){return r in Zx||(Zx[r]=0),Zx[r]+=1,r+Zx[r].toString()}function Qx(r){return Array.isArray(r)&&Array.isArray(r[0])}function gm(r){return r.length===0?[]:Array.isArray(r[0])?r:[r]}function Pt(r){let t;if(Array.isArray(r)){if(r.length!==1)throw new z(`Expected Tensor length to be 1; got ${r.length}`);t=r[0]}else t=r;return t}function Jt(r){if(Array.isArray(r)&&Array.isArray(r[0])){if(r.length===1)return r=r,r[0];throw new z(`Expected exactly 1 Shape; got ${r.length}`)}else return r}function xm(r){let t=0;for(let e of r)e.shape.length===0?t+=1:t+=e.shape.reduce((n,o)=>n*o);return t}var iD="Variable",ty=class{constructor(t,e="float32",n=iD,o=!0,s=null){this.dtype=e==null?"float32":e,this.shape=t.shape,this.id=Jx(),n=n==null?iD:n,this.originalName=Hx(n),this.name=qx(this.originalName),this.trainable_=o,this.constraint=s,this.val=rC(t,this.trainable_,this.name,this.dtype)}read(){return this.assertNotDisposed(),this.val}write(t){return this.assertNotDisposed(),TX(this.val,t),this.val.id!==t.id&&(this.val.assign(t),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(t){this.trainable_=t,this.val.trainable=t}};function TX(r,t){if(r.shape.toString()!==t.shape.toString())throw new Error("Shape mismatch: "+JSON.stringify(r.shape)+" vs. "+JSON.stringify(t.shape))}function Ed(r){return r.map(t=>t.read())}function ym(r){r.forEach(t=>{t[0].write(t[1])})}var _e=class{constructor(t){this.dtype=t.dtype,this.shape=t.shape,t.shape!=null?this.ndim=t.shape.length:this.ndim=t.ndim,this.maxNDim=t.maxNDim,this.minNDim=t.minNDim,this.axes=t.axes||{}}},gn=class{constructor(t,e,n,o,s,i,a){this.dtype=t,this.shape=e,this.sourceLayer=n,this.inputs=o,this.callArgs=s,this.outputTensorIndex=a,this.id=Jx(),i!=null&&(this.originalName=Hx(i),this.name=qx(this.originalName)),this.rank=e.length}},_X=0,Kl=class{constructor(t,e){this.callArgs=e,this.id=_X++,this.outboundLayer=t.outboundLayer,this.inboundLayers=t.inboundLayers,this.nodeIndices=t.nodeIndices,this.tensorIndices=t.tensorIndices,this.inputTensors=t.inputTensors,this.outputTensors=t.outputTensors,this.inputMasks=t.inputMasks,this.outputMasks=t.outputMasks,this.inputShapes=t.inputShapes,this.outputShapes=t.outputShapes;for(let n of t.inboundLayers)n!=null&&n.outboundNodes.push(this);t.outboundLayer.inboundNodes.push(this)}getConfig(){let t=[];for(let e of this.inboundLayers)e!=null?t.push(e.name):t.push(null);return{outboundLayer:this.outboundLayer?this.outboundLayer.name:null,inboundLayers:t,nodeIndices:this.nodeIndices,tensorIndices:this.tensorIndices}}},EX=0,Ut=class extends et.Serializable{constructor(t={}){super();this._callHook=null,this._addedWeightNames=[],this._stateful=!1,this.id=EX++,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 e=t.name;if(!e){let n=this.getClassName();e=So(n)+"_"+ql(n)}if(this.name=e,this.trainable_=t.trainable==null?!0:t.trainable,t.inputShape!=null||t.batchInputShape!=null){let n;if(t.batchInputShape!=null)n=t.batchInputShape;else if(t.inputShape!=null){let s=null;t.batchSize!=null&&(s=t.batchSize),n=[s].concat(t.inputShape)}this.batchInputShape=n;let o=t.dtype;o==null&&(o=t.inputDType),o==null&&(o="float32"),this.dtype=o}t.weights!=null?this.initialWeights=t.weights:this.initialWeights=null,this._refCount=null,this.fastWeightInitDuringBuild=!1}static nodeKey(t,e){return t.name+"_ib-"+e.toString()}getNodeAtIndex(t,e){if(this.inboundNodes.length===0)throw new Xr(`The layer has never been called and thus has no defined ${e}.`);if(this.inboundNodes.length<=t)throw new z(`Asked to get ${e} at node ${t}, but the layer has only ${this.inboundNodes.length} inbound nodes.`);return this.inboundNodes[t]}getInputAt(t){return _r(this.getNodeAtIndex(t,"input").inputTensors)}getOutputAt(t){return _r(this.getNodeAtIndex(t,"output").outputTensors)}get input(){if(this.inboundNodes.length>1)throw new Ln(`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 Ln(`Layer ${this.name} is not connected, no input to return.`);return _r(this.getNodeAtIndex(0,"input").inputTensors)}get output(){if(this.inboundNodes.length===0)throw new Ln(`Layer ${this.name} has no inbound nodes.`);if(this.inboundNodes.length>1)throw new Ln(`Layer ${this.name} has multiple inbound nodes, hence the notion of "layer output" is ill-defined. Use \`getOutputAt(nodeIndex)\` instead.`);return _r(this.getNodeAtIndex(0,"output").outputTensors)}get losses(){return this._losses}calculateLosses(){return this.losses.map(t=>t())}get updates(){return this._updates}get built(){return this._built}set built(t){this._built=t}get trainable(){return this.trainable_}set trainable(t){this._trainableWeights.forEach(e=>e.trainable=t),this.trainable_=t}get trainableWeights(){return this.trainable_?this._trainableWeights.filter(t=>t.trainable):[]}set trainableWeights(t){this._trainableWeights=t}get nonTrainableWeights(){return this.trainable?this._trainableWeights.filter(t=>!t.trainable).concat(this._nonTrainableWeights):this._trainableWeights.concat(this._nonTrainableWeights)}set nonTrainableWeights(t){this._nonTrainableWeights=t}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(t){if(t=we(t),this.inputSpec==null||this.inputSpec.length===0)return;let e=we(this.inputSpec);if(t.length!==e.length)throw new z(`Layer ${this.name} expects ${e.length} inputs, but it received ${t.length} input tensors. Input received: ${t}`);for(let n=0;n<t.length;n++){let o=t[n],s=e[n];if(s==null)continue;let i=o.rank;if(s.ndim!=null&&i!==s.ndim)throw new z(`Input ${n} is incompatible with layer ${this.name}: expected ndim=${s.ndim}, found ndim=${i}`);if(s.maxNDim!=null&&i>s.maxNDim)throw new z(`Input ${n} is incompatible with layer ${this.name}: expected max_ndim=${s.maxNDim}, found ndim=${i}`);if(s.minNDim!=null&&i<s.minNDim)throw new z(`Input ${n} is incompatible with layer ${this.name}: expected min_ndim=${s.minNDim}, found ndim=${i}.`);if(s.dtype!=null&&o.dtype!==s.dtype)throw new z(`Input ${n} is incompatible with layer ${this.name} : expected dtype=${s.dtype}, found dtype=${o.dtype}.`);if(s.axes){let a=o.shape;for(let u in s.axes){let l=Number(u),c=s.axes[u],p=l>=0?a[l]:a[a.length+l];if(c!=null&&[c,null].indexOf(p)===-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 ${a}.`)}}if(s.shape!=null)for(let a=0;a<s.shape.length;++a){let u=s.shape[a],l=o.shape[a];if(u!=null&&l!=null&&u!==l)throw new z(`Input ${n} is incompatible with layer ${this.name}: expected shape=${s.shape}, found shape=${o.shape}.`)}}}call(t,e){return t}invokeCallHook(t,e){this._callHook!=null&&this._callHook(t,e)}setCallHook(t){this._callHook=t}clearCallHook(){this._callHook=null}apply(t,e){e=e||{},this.assertNotDisposed();let n=we(t),o=!0;for(let i of n)if(!(i instanceof gn)){o=!1;break}let s=!0;for(let i of n)if(i instanceof gn){s=!1;break}if(o===s)throw new z("Arguments to apply() must be all SymbolicTensors or all Tensors");return ei(this.name,()=>{if(!this.built){this.assertInputCompatibility(t);let i=[];for(let a of we(t))i.push(a.shape);this.build(_r(i)),this.built=!0,this.initialWeights&&this.setWeights(this.initialWeights),this._refCount===null&&s&&(this._refCount=1)}if(this.assertInputCompatibility(t),s){let i=this.call(t,e),a=we(i),u=[];for(let l of a)n.indexOf(l)!==-1&&(l=l.clone()),u.push(l);if(i=_r(u),this.activityRegularizer!=null)throw new Tt("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return i}else{let i=AX(t),a=this.computeOutputShape(i),u,l=DX(t);if(this.warnOnIncompatibleInputShape(Array.isArray(t)?i[0]:i),a!=null&&a.length>0&&Array.isArray(a[0])?u=a.map((c,p)=>new gn(l,c,this,we(t),e,this.name,p)):u=new gn(l,a,this,we(t),e,this.name),this.addInboundNode(t,u,null,null,i,a,e),this._refCount++,this.activityRegularizer!=null)throw new Tt("Layer invocation in the presence of activity regularizer(s) is not supported yet.");return u}})}warnOnIncompatibleInputShape(t){if(this.batchInputShape!=null)if(t.length!==this.batchInputShape.length)console.warn(`The rank of the input tensor provided (shape: ${JSON.stringify(t)}) does not match that of the batchInputShape (${JSON.stringify(this.batchInputShape)}) of the layer ${this.name}`);else{let e=!1;this.batchInputShape.forEach((n,o)=>{n!=null&&t[o]!=null&&t[o]!==n&&(e=!0)}),e&&console.warn(`The shape of the input tensor (${JSON.stringify(t)}) 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 Ln(`The layer ${this.name} has never been called and thus has no defined output shape.`);let t=[];for(let e of this.inboundNodes){let n=JSON.stringify(e.outputShapes);t.indexOf(n)===-1&&t.push(n)}if(t.length===1){let e=this.inboundNodes[0].outputShapes;return Array.isArray(e)&&Array.isArray(e[0])&&e.length===1?e[0]:e}else throw new Ln(`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 Xr(`You tried to call countParams() on ${this.name}, but the layer is not built yet. Build it first by calling build(batchInputShape).`);return xm(this.weights)}build(t){this.built=!0}getWeights(t=!1){return Ed(t?this.trainableWeights:this.weights)}setWeights(t){B(()=>{let e=this.weights;if(e.length!==t.length)throw new z(`You called setWeights(weights) on layer "${this.name}" with a weight list of length ${t.length}, but the layer was expecting ${e.length} weights. Provided weights: ${t}...`);if(e.length===0)return;let n=[],o=Ed(e);for(let s=0;s<o.length;++s){let i=o[s],a=e[s],u=t[s];if(!x.arraysEqual(i.shape,u.shape))throw new z(`Layer weight shape ${i.shape} not compatible with provided weight shape ${u.shape}`);n.push([a,u])}ym(n)})}addWeight(t,e,n,o,s,i,a,u){if(this._addedWeightNames.indexOf(t)!==-1)throw new z(`Duplicate weight name ${t} for layer ${this.name}`);this._addedWeightNames.push(t),n==null&&(n="float32"),this.fastWeightInitDuringBuild&&(o=u!=null?u():de("zeros"));let l=o.apply(e,n),c=new ty(l,n,t,i,a);return l.dispose(),s!=null&&this.addLoss(()=>s.apply(c.read())),i==null&&(i=!0),i?this._trainableWeights.push(c):this._nonTrainableWeights.push(c),c}setFastWeightInitDuringBuild(t){this.fastWeightInitDuringBuild=t}addLoss(t){t==null||Array.isArray(t)&&t.length===0||(t=we(t),this._losses!==void 0&&this._losses!==null&&this.losses.push(...t))}computeOutputShape(t){return t}computeMask(t,e){if(!this.supportsMasking){if(e!=null)if(Array.isArray(e))e.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 ${this.name} does not support masking, but was passed an inputMask.`);return null}return e}addInboundNode(t,e,n,o,s,i,a=null){let u=we(t);e=we(e),n=we(n),o=we(o),s=gm(s),i=gm(i);let l=[],c=[],p=[];for(let m of u)l.push(m.sourceLayer),c.push(m.nodeIndex),p.push(m.tensorIndex);new Kl({outboundLayer:this,inboundLayers:l,nodeIndices:c,tensorIndices:p,inputTensors:u,outputTensors:e,inputMasks:n,outputMasks:o,inputShapes:s,outputShapes:i},a);for(let m=0;m<e.length;m++)e[m].sourceLayer=this,e[m].nodeIndex=this.inboundNodes.length-1,e[m].tensorIndex=m}getConfig(){let t={name:this.name,trainable:this.trainable};return this.batchInputShape!=null&&(t.batchInputShape=this.batchInputShape),this.dtype!=null&&(t.dtype=this.dtype),t}disposeWeights(){return this.weights.forEach(t=>t.dispose()),this.weights.length}assertNotDisposed(){if(this._refCount===0)throw new Error(`Layer '${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 t=0;return--this._refCount===0&&(t=this.disposeWeights()),{refCountAfterDispose:this._refCount,numDisposedVariables:t}}};function AX(r){r=we(r);let t=[];for(let e of r)t.push(e.shape);return _r(t)}function DX(r){return"float32"}function wC(r,t,e){if((t==null||e!=null&&e>0)&&(t=r.sourceLayer,e=r.nodeIndex),t.inboundNodes.length===0)return[r];{let n=t.inboundNodes[e];if(n.inboundLayers.length===0)return n.inputTensors;{let o=[];for(let s=0;s<n.inboundLayers.length;s++){let i=n.inputTensors[s],a=n.inboundLayers[s],u=n.nodeIndices[s],l=wC(i,a,u);for(let c of l)o.indexOf(c)===-1&&o.push(c)}return o}}}var Bi=class extends Ut{constructor(t){super({dtype:t.dtype,name:t.name!=null?t.name:ql("input").toString()});if(t.batchSize==null&&(t.batchSize=null),t.sparse==null&&(t.sparse=!1),this.trainable=!1,this.built=!0,this.sparse=t.sparse,t.inputShape!=null&&t.batchInputShape!=null)throw new z("Only provide the inputShape OR batchInputShape argument to inputLayer, not both at the same time.");let e=t.batchInputShape;if(e==null){if(t.inputShape==null)throw new z("An InputLayer should be passed either a `batchInputShape` or an `inputShape`.");e=[t.batchSize].concat(t.inputShape)}else if(t.batchSize!=null)throw new z("Cannot specify batchSize if batchInputShape is specified when creating an InputLayer.");let n=t.dtype||"float32";this.batchInputShape=e,this.dtype=n,this.inputSpec=[{shape:e}];let o=new gn(this.dtype,this.batchInputShape,this,[],{},this.name);o.nodeIndex=0,o.tensorIndex=0,new Kl({outboundLayer:this,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:[o],outputTensors:[o],inputMasks:[null],outputMasks:[null],inputShapes:[e],outputShapes:[e]})}apply(t,e){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}}};Bi.className="InputLayer";et.registerClass(Bi);function ey(r){if(r.batchShape==null&&r.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(r.batchShape!=null&&r.shape!=null)throw new z("Please provide either a `shape` or `batchShape` argument to Input, but not both.");let t=r.batchShape;r.shape!=null&&t==null&&(t=[null].concat(r.shape));let e=r.dtype;return e==null&&(e="float32"),new Bi({batchInputShape:t,name:r.name,dtype:e,sparse:r.sparse}).inboundNodes[0].outputTensors[0]}async function Vi(r){if(r==null)return;let t=[],e=[],n=[];for(let o in r){let s=r[o];if(typeof s!="number"){let i=s;t.push(i.data()),e.push(o),n.push(i)}}if(t.length>0){let o=await Promise.all(t);for(let s=0;s<o.length;++s)r[e[s]]=o[s][0];_t(n)}}function ry(r){if(r!=null)for(let t in r){let e=r[t];typeof e!="number"&&e.dispose()}}var aD;(function(r){r[r.SILENT=0]="SILENT",r[r.VERBOSE=1]="VERBOSE"})(aD||(aD={}));var $X=125,jl=class{constructor(){this.validationData=null}setParams(t){this.params=t}async onEpochBegin(t,e){}async onEpochEnd(t,e){}async onBatchBegin(t,e){}async onBatchEnd(t,e){}async onTrainBegin(t){}async onTrainEnd(t){}setModel(t){}},vC=class{constructor(t,e=10){t==null&&(t=[]),this.callbacks=t,this.queueLength=e}append(t){this.callbacks.push(t)}setParams(t){for(let e of this.callbacks)e.setParams(t)}setModel(t){for(let e of this.callbacks)e.setModel(t)}async onEpochBegin(t,e){e==null&&(e={});for(let n of this.callbacks)await n.onEpochBegin(t,e)}async onEpochEnd(t,e){e==null&&(e={});for(let n of this.callbacks)await n.onEpochEnd(t,e)}async onBatchBegin(t,e){e==null&&(e={});for(let n of this.callbacks)await n.onBatchBegin(t,e)}async onBatchEnd(t,e){e==null&&(e={});for(let n of this.callbacks)await n.onBatchEnd(t,e)}async onTrainBegin(t){t==null&&(t={});for(let e of this.callbacks)await e.onTrainBegin(t)}async onTrainEnd(t){t==null&&(t={});for(let e of this.callbacks)await e.onTrainEnd(t)}},lD=class extends jl{constructor(){super()}async onEpochBegin(t){this.seen=0,this.totals={}}async onBatchEnd(t,e){e==null&&(e={});let n=e.size==null?0:e.size;this.seen+=n;for(let o in e){let s=e[o];if(typeof s=="number")this.totals.hasOwnProperty(o)||(this.totals[o]=0),this.totals[o]=this.totals[o]+s*n;else{let i;o in this.totals?i=this.totals[o]:this.totals[o]=0;let a=B(()=>J(this.totals[o],P(s,n)));this.totals[o]=a,i!=null&&i.dispose()}}}async onEpochEnd(t,e){if(e!=null)for(let n of this.params.metrics)this.totals[n]!=null&&(typeof this.totals[n]=="number"?e[n]=this.totals[n]/this.seen:B(()=>{let o=P(ct(1,this.seen),this.totals[n]);e[n]=o,this.totals[n].dispose(),Me(e[n])}))}},CC=class extends jl{async onTrainBegin(t){this.epoch=[],this.history={}}async onEpochEnd(t,e){e==null&&(e={}),this.epoch.push(t);for(let n in e)this.history[n]==null&&(this.history[n]=[]),this.history[n].push(e[n])}async syncData(){let t=[],e=[],n=[];for(let s in this.history){let i=this.history[s];for(let a=0;a<i.length;++a)if(typeof i[a]!="number"){let u=i[a];t.push(u.data()),e.push(s),n.push(a)}}let o=await Promise.all(t);for(let s=0;s<o.length;++s)this.history[e[s]][n[s]].dispose(),this.history[e[s]][n[s]]=o[s][0]}},IC=class extends jl{constructor(t,e){super();if(this.currentEpoch=0,this.nowFunc=t.nowFunc,this.nextFrameFunc=t.nextFrameFunc||fd,this.yieldEvery=e||"auto",this.yieldEvery==="auto"&&(this.yieldEvery=$X),this.yieldEvery==="never"&&t.onYield!=null)throw new Error("yieldEvery is `never` but you provided an `onYield` callback. Either change `yieldEvery` or remove the callback");x.isNumber(this.yieldEvery)&&(this.maybeWait=O2(this.maybeWait.bind(this),this.yieldEvery,this.nowFunc)),this.trainBegin=t.onTrainBegin,this.trainEnd=t.onTrainEnd,this.epochBegin=t.onEpochBegin,this.epochEnd=t.onEpochEnd,this.batchBegin=t.onBatchBegin,this.batchEnd=t.onBatchEnd,this.yield=t.onYield}async maybeWait(t,e,n){let o=[];this.yield!=null&&(await Vi(n),o.push(this.yield(t,e,n))),o.push(this.nextFrameFunc()),await Promise.all(o)}async onEpochBegin(t,e){this.currentEpoch=t,this.epochBegin!=null&&(await Vi(e),await this.epochBegin(t,e))}async onEpochEnd(t,e){let n=[];this.epochEnd!=null&&(await Vi(e),n.push(this.epochEnd(t,e))),this.yieldEvery==="epoch"&&n.push(this.nextFrameFunc()),await Promise.all(n)}async onBatchBegin(t,e){this.batchBegin!=null&&(await Vi(e),await this.batchBegin(t,e))}async onBatchEnd(t,e){let n=[];this.batchEnd!=null&&(await Vi(e),n.push(this.batchEnd(t,e))),this.yieldEvery==="batch"?n.push(this.nextFrameFunc()):x.isNumber(this.yieldEvery)&&n.push(this.maybeWait(this.currentEpoch,t,e)),await Promise.all(n)}async onTrainBegin(t){this.trainBegin!=null&&(await Vi(t),await this.trainBegin(t))}async onTrainEnd(t){this.trainEnd!=null&&(await Vi(t),await this.trainEnd(t))}};function ny(r,t){return r==null&&(r={}),r instanceof jl?[r]:Array.isArray(r)&&r[0]instanceof jl?r:we(r).map(n=>new IC(n,t))}var _n=class{constructor(){}static registerCallbackConstructor(t,e){x.assert(t>=0&&Number.isInteger(t),()=>`Verbosity level is expected to be an integer >= 0, but got ${t}`),_n.checkForDuplicate(e),_n.constructors[t]==null&&(_n.constructors[t]=[]),_n.constructors[t].push(e)}static checkForDuplicate(t){for(let e in _n.constructors)_n.constructors[+e].forEach(o=>{if(o===t)throw new z("Duplicate callback constructor.")})}static clear(){_n.constructors={}}static createCallbacks(t){let e=[];for(let n in _n.constructors){let o=+n;t>=o&&e.push(..._n.constructors[o])}return e.map(n=>new n)}};_n.constructors={};function oy(r,t,e,n,o,s,i,a,u){let l=new CC,c=[new lD,..._n.createCallbacks(t)];r!=null&&c.push(...r),c.push(l);let p=new vC(c);return p.setParams({epochs:e,initialEpoch:n,samples:o,steps:s,batchSize:i,verbose:t,doValidation:a,metrics:u}),{callbackList:p,history:l}}function xn(r,t={},e=!1){return Li(r,et.SerializationMap.getMap().classNameMap,t,"layer",e)}function Ad(r,t){return B(()=>{r.dtype!=="float32"&&(r=Z(r,"float32"));let e=mt(dc(r),t,!0),n=Xs(e.shape,cr()),o=be(cn(e,n));return ct(r,o)})}function Gi(r,t){return B(()=>xe(dc(lt(t,r)),-1))}function bm(r,t){return B(()=>xe(Se(lt(t,r)),-1))}function Xl(r,t){return B(()=>{let e=lt(r,t),n=br(Se(r),cr(),Number.MAX_VALUE),o=Se(ct(e,n));return P(100,xe(o,-1))})}function RX(r,t){return B(()=>{let e=br(t,cr(),Number.MAX_VALUE),n=wr(J(1,e)),o=br(r,cr(),Number.MAX_VALUE),s=wr(J(1,o));return xe(dc(lt(n,s)),-1)})}function FX(r,t){return B(()=>{let e=cn(0,lt(1,P(r,t)));return xe(dc(e),-1)})}function OX(r,t){return B(()=>{let e=cn(0,lt(1,P(r,t)));return xe(e,-1)})}function PX(r,t){return B(()=>{let e=mt(P(r,t),-1),n=$r(P(lt(1,r),t),-1);return cn(0,J(1,lt(n,e)))})}function MX(r,t){return B(()=>{let e=Math.log(2),n=lt(t,r),o=lt(J(n,xo(P(-2,n))),e);return xe(o,-1)})}function gc(r,t,e=!1){return B(()=>{if(e)t=Ja(t);else{let n=mt(t,t.shape.length-1,!0);t=ct(t,n)}return t=br(t,cr(),1-cr()),Zt(mt(P(Z(r,"float32"),wr(t)),t.shape.length-1))})}function wm(r,t,e=!1){return B(()=>{let n=Z(Ys(X2(r)),"int32");t=br(t,cr(),1-cr());let o=t.shape,s=O(Us(n,o[o.length-1]),o);return gc(s,t,e)})}function LX(r,t){if(!x.arraysEqual(r.shape,t.shape))throw new z(`logits and labels must have the same shape, but got shapes ${JSON.stringify(r.shape)} and ${JSON.stringify(t.shape)}`);return B(()=>{let e=Tr(t),n=Zt(Se(t));return J(lt(e,P(t,r)),qa(Ze(n)))})}function vm(r,t){return B(()=>{let e;return e=br(t,cr(),1-cr()),e=wr(ct(e,lt(1,e))),xe(LX(r,e),-1)})}function zX(r,t){return B(()=>{let e=br(r,cr(),1),n=br(t,cr(),1);return mt(P(r,wr(ct(e,n))),-1)})}function BX(r,t){return B(()=>{let e=wr(J(cr(),t));return xe(lt(t,P(r,e)),-1)})}function $d(r,t){return B(()=>{let e=Ad(r,-1),n=Ad(t,-1),o=P(e,n);return Zt(mt(o,-1))})}var Dd={meanSquaredError:Gi,meanAbsoluteError:bm,meanAbsolutePercentageError:Xl,meanSquaredLogarithmicError:RX,squaredHinge:FX,hinge:OX,categoricalHinge:PX,logcosh:MX,categoricalCrossentropy:gc,sparseCategoricalCrossentropy:wm,binaryCrossentropy:vm,kullbackLeiblerDivergence:zX,poisson:BX,cosineProximity:$d};function sy(r){if(typeof r=="string"){if(r in Dd)return Dd[r];let t=`Unknown loss ${r}`;throw r.toLowerCase().includes("softmaxcrossentropy")&&(t=`Unknown loss ${r}. Use "categoricalCrossentropy" as the string name for tf.losses.softmaxCrossEntropy`),new z(t)}else return r}function Rd(r,t){return B(()=>{let e=P(.5,dr(t)),n=fc(Ge(t,e),r.dtype);return xe(Sr(r,n),-1)})}function Fd(r,t){return B(()=>fc(Sr(Hs(r,-1),Hs(t,-1)),"float32"))}function cD(r,t){return B(()=>Z(mt(Nr(Sr(r,1),Sr(t,1))),"float32"))}function VX(r,t){return B(()=>Z(mt(Nr(Sr(r,1),Sr(t,0))),"float32"))}function GX(r,t){return B(()=>Z(mt(Nr(Sr(r,0),Sr(t,1))),"float32"))}function SC(r,t){return B(()=>{let e=cD(r,t),n=GX(r,t),o=J(e,n);return Z(Ne(Ge(o,0),ct(e,o),0),"float32")})}function pD(r,t){return B(()=>{let e=cD(r,t),n=VX(r,t),o=J(e,n);return Z(Ne(Ge(o,0),ct(e,o),0),"float32")})}function ay(r,t){return vm(r,t)}function ly(r,t){return r.rank===t.rank&&(r=zr(r,[r.rank-1])),t=Hs(t,-1),t.dtype!==r.dtype&&(t=Z(t,r.dtype)),Z(Sr(r,t),"float32")}var WX=Gi,UX=Gi,HX=bm,qX=bm,KX=Xl,jX=Xl,Od=gc,XX=$d,kC=wm,iy={binaryAccuracy:Rd,categoricalAccuracy:Fd,precision:SC,categoricalCrossentropy:Od,sparseCategoricalCrossentropy:kC,mse:WX,MSE:UX,mae:HX,MAE:qX,mape:KX,MAPE:jX,cosine:XX};function mD(r){if(typeof r=="string"&&r in iy)return iy[r];if(typeof r!="string"&&r!=null)return r;throw new z(`Unknown metric ${r}`)}function Pd(r){if(Zn(r!==null,`Unknown LossOrMetricFn ${r}`),typeof r=="string")return r;{let t;for(let e of Object.keys(Dd))if(Dd[e]===r){t=e;break}if(t!==void 0)return t;for(let e of Object.keys(iy))if(iy[e]===r){t=e;break}return t!==void 0?t:r.name}}function dD(r){let t={Adagrad:()=>pc.adagrad(.01),Adadelta:()=>pc.adadelta(1,.95,cr()),Adam:()=>pc.adam(.001,.9,.999,cr()),Adamax:()=>pc.adamax(.002,.9,.999,cr(),0),RMSProp:()=>pc.rmsprop(.001,.9,0,cr()),SGD:()=>pc.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,r in t)return t[r]();throw new z(`Unknown Optimizer ${r}`)}function TC(r,t,e=!1){if(r==null||typeof r!="object"||Object.getPrototypeOf(r)!==Object.prototype||!NC(r))throw new Error("User-defined metadata is expected to be a JSON object, but is not.");if(e){let n=JSON.stringify(r);n.length>1048576&&console.warn(`User-defined metadata of model "${t}" is too large in size (length=${n.length} when serialized). It is not recommended to store such large objects in user-defined metadata. Please make sure its serialized length is <= ${1048576}.`)}}function NC(r){if(r===null)return!0;if(typeof r=="object")if(Object.getPrototypeOf(r)===Object.prototype){let t=Object.keys(r);for(let e of t)if(typeof e!="string"||!NC(r[e]))return!1;return!0}else if(Array.isArray(r)){for(let t of r)if(!NC(t))return!1;return!0}else return!1;else{let t=typeof r;return t==="string"||t==="number"||t==="boolean"}}function hD(r,t,e,n=console.log){let o=JX(r),s=["Layer (type)","Input Shape","Output shape","Param #"];o?(t=t||90,e=e||[.32,.61,.89,1]):(t=t||115,e=e||[.24,.48,.7,.8,1]),e[e.length-1]<=1&&(e=e.map(c=>Math.floor(t*c)));let i;if(!o){s.push("Receives inputs"),i=[];for(let c in r.nodesByDepth)i.push(...r.nodesByDepth[c])}n("_".repeat(t)),uy(s,e,n),n("=".repeat(t));let a=r.layers;for(let c=0;c<a.length;++c)o?QX(a[c],e,n):t8(a[c],e,i,n),n((c===a.length-1?"=":"_").repeat(t));r.checkTrainableWeightsConsistency();let u=ZX(r),l=xm(r.nonTrainableWeights);n(`Total params: ${u+l}`),n(`Trainable params: ${u}`),n(`Non-trainable params: ${l}`),n("_".repeat(t))}function ZX(r){let t;return r.collectedTrainableWeights!=null?t=xm(r.collectedTrainableWeights):t=xm(r.trainableWeights),t}function JX(r){let t=!0,e=[],n=[];for(let o in r.nodesByDepth)e.push(r.nodesByDepth[o]);for(let o of e){if(o.length>1||o.length===1&&o[0].inboundLayers.length>1){t=!1;break}n.push(...o)}if(t)for(let o of r.layers){let s=!1;for(let i of o.inboundNodes)if(n.indexOf(i)!==-1)if(s){t=!1;break}else s=!0;if(!t)break}return t}function uy(r,t,e=console.log){let n="";for(let o=0;o<r.length;++o)o>0&&(n=n.slice(0,n.length-1)+" "),n+=r[o],n=n.slice(0,t[o]),n+=" ".repeat(t[o]-n.length);e(n)}function QX(r,t,e){let n,o;try{o=r.inboundNodes.map(u=>JSON.stringify(u.inputShapes)).join(",")}catch(u){o="multiple"}try{n=JSON.stringify(r.outputShape)}catch(u){n="multiple"}let s=r.name,i=r.getClassName(),a=[`${s} (${i})`,o,n,r.countParams().toString()];uy(a,t,e)}function t8(r,t,e,n){let o,s;try{s=r.inboundNodes.map(p=>JSON.stringify(p.inputShapes)).join(",")}catch(p){s="multiple"}try{o=JSON.stringify(r.outputShape)}catch(p){o="multiple"}let i=[];for(let p of r.inboundNodes)if(!(e!=null&&e.length>0&&e.indexOf(p)===-1))for(let m=0;m<p.inboundLayers.length;++m){let f=p.inboundLayers[m].name,d=p.nodeIndices[m],h=p.tensorIndices[m];i.push(`${f}[${d}][${h}]`)}let a=r.name,u=r.getClassName(),l=i.length===0?"":i[0],c=[`${a} (${u})`,s,o,r.countParams().toString(),l];uy(c,t,n);for(let p=1;p<i.length;++p)uy(["","","","",i[p]],t,n)}function gD(r,t,e){return(r==="inboundNodes"||r==="outputLayers"||r==="inputLayers")&&t===0&&typeof e=="string"}function xc(r,t){if(r===null)return null;if(typeof r=="string")return rl(r);if(typeof r=="number"||typeof r=="boolean")return r;if(r instanceof Array){let e=[],n=r.length;for(let o=0;o<n;++o){let s=r[o];gD(t,o,s)?e.push(s):e.push(xc(s,t))}return e}else{let e={};for(let n of Object.keys(r)){let o=r[n];if(n==="name"&&typeof o=="string")e[n]=o;else{let s=rl(n);e[s]=xc(o,s)}}return e}}function cy(r,t){if(r==null)return null;if(typeof r=="string")return So(r);if(typeof r=="number"||typeof r=="boolean")return r;if(r instanceof Array){let e=[],n=r.length;for(let o=0;o<n;++o){let s=r[o];gD(t,o,s)?e.push(s):e.push(cy(s,t))}return e}else{let e={};for(let n of Object.keys(r)){let o=r[n],s=So(n);(n==="name"||n==="className")&&typeof o=="string"?e[s]=o:e[s]=cy(o,n)}return e}}var Cm="3.15.0";function e8(r,t){if(r.dtype==null||r.dtype===t.dtype)return t;try{return Z(t,r.dtype)}catch(e){throw new z(`The dtype of the feed (${t.dtype}) can not be cast to the dtype of the key '${r.name}' (${r.dtype}).`)}}var ni=class{constructor(t){if(this.id2Value={},this.id2Mask={},this.name2Id={},t instanceof ni)for(let e in t.id2Value)this.id2Value[e]=t.id2Value[e],e in t.id2Mask&&(this.id2Mask[e]=t.id2Mask[e]);else{if(t==null)return;for(let e of t)this.add(e.key,e.value)}}add(t,e,n){if(this.id2Value[t.id]==null)this.id2Value[t.id]=e8(t,e),this.name2Id[t.name]=t.id,n!=null&&(this.id2Mask[t.id]=n);else throw new z(`Duplicate key: name=${t.name}, id=${t.id}`);return this}addFeed(t){this.add(t.key,t.value)}hasKey(t){return this.id2Value[t.id]!=null}names(){return Object.keys(this.name2Id)}getValue(t){if(t instanceof gn){if(this.id2Value[t.id]==null)throw new z(`Nonexistent key: ${t.name}`);return this.id2Value[t.id]}else{let e=this.name2Id[t];if(e==null)throw new z(`Feed dict has no SymbolicTensor name: ${t}`);return this.id2Value[e]}}getMask(t){if(t instanceof gn){if(this.id2Value[t.id]==null)throw new z(`Nonexistent key: ${t.name}`);return this.id2Mask[t.id]}else{let e=this.name2Id[t];if(e==null)throw new z(`Feed dict has no SymbolicTensor name: ${t}`);return this.id2Mask[e]}}disposeMasks(){this.id2Mask!=null&&_t(this.id2Mask)}},_C={},xD={};function yc(r,t,e,n){let o=e==null?!1:e.training,s=Array.isArray(r),i=s?r:[r],a=i.map(d=>d.name),u=[],l=t.names();for(let d of a)l.indexOf(d)!==-1?u.push(t.getValue(d)):u.push(null);n!=null&&(n.maxNumTensors=-1/0,n.minNumTensors=1/0);let c=a.join(",")+"|"+t.names().join(","),p,m;if(_C[c]==null){let d=r8(i,t);p=d.sorted,m=d.recipientCounts,_C[c]=p,xD[c]=m}p=_C[c],m={},o||Object.assign(m,xD[c]);let f=new ni(t);for(let d=0;d<p.length;++d){if(n!=null){let R=Af().numTensors;R>n.maxNumTensors&&(n.maxNumTensors=R),R<n.minNumTensors&&(n.minNumTensors=R)}let h=p[d],g=h.sourceLayer;if(g instanceof Bi)continue;let y=[],b=[],w=[],v=!1;for(let R of h.inputs){let M=f.getValue(R),W=f.getMask(R);y.push(M),b.push(W),W!=null&&(v=!0),o||(m[R.name]--,m[R.name]===0&&!t.hasKey(R)&&a.indexOf(R.name)===-1&&!M.isDisposed&&R.sourceLayer.stateful!==!0&&w.push(M))}v&&(e=e||{},e.mask=b[0]);let k=we(g.apply(y,e)),E=null;g.supportsMasking&&(E=g.computeMask(y,b));let D=o8(h),$=Array.isArray(D)?D:[D];for(let R=0;R<$.length;++R){f.hasKey($[R])||f.add($[R],k[R],Array.isArray(E)?E[0]:E);let M=a.indexOf($[R].name);M!==-1&&(u[M]=k[R])}o||_t(w)}return f.disposeMasks(),s?u:u[0]}function r8(r,t){x.assert(r!=null&&r.length>0,()=>"Expected at least one fetch, got none");let e=[],n={};if(r.length===1){let o=yD(r[0],t);e=o.sorted,n=o.recipientMap}else{let o=new Set;for(let s of r){let{sorted:i,recipientMap:a}=yD(s,t);for(let u of i)o.has(u.name)||(e.push(u),o.add(u.name));for(let u in a)n[u]==null&&(n[u]=new Set),a[u].forEach(l=>n[u].add(l))}}return{sorted:e,recipientCounts:n8(n)}}function n8(r){let t={};for(let e in r)t[e]=r[e].size;return t}function yD(r,t){let e=new Set,n=[],o={};for(let a of t.names())e.add(a);let s=[],i=[];for(s.push(r);s.length>0;){let a=s[s.length-1];if(e.has(a.name)){s.pop();continue}let u=i[i.length-1]===s.length-1;if(a.inputs.length===0||u)s.pop(),n.push(a),e.add(a.name),u&&i.pop();else{i.push(s.length-1);for(let l of a.inputs)o[l.name]==null&&(o[l.name]=new Set),o[l.name].add(a.name),!e.has(l.name)&&s.push(l)}}return{sorted:n,recipientMap:o}}function o8(r){let t;if(r.sourceLayer.inboundNodes.length===1)t=r.sourceLayer.output;else{let e=null;for(let n=0;n<r.sourceLayer.inboundNodes.length;++n)for(let o of r.sourceLayer.inboundNodes[n].outputTensors)if(o.id===r.id){e=n;break}t=r.sourceLayer.getOutputAt(e)}return t}var Jn=class extends Ut{constructor(t){super({});if(this.containerNodes=new Set,this.name=t.name,this.name==null){let b=this.getClassName().toLowerCase();this.name=ql(b)}if(this.supportsMasking=!1,this.trainable_=!0,Array.isArray(t.inputs)?this.inputs=t.inputs.slice():this.inputs=[t.inputs],Array.isArray(t.outputs)?this.outputs=t.outputs.slice():this.outputs=[t.outputs],ko(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(b=>b.name)}`);ko(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(b=>b.name)}`),this.inputLayers=[],this.inputLayersNodeIndices=[],this.inputLayersTensorIndices=[],this.outputLayers=[],this.outputLayersNodeIndices=[],this.outputLayersTensorIndices=[],this.layers=[],this.internalContainerRefs=[];for(let b of this.outputs){let w=b.sourceLayer,v=b.nodeIndex,k=b.tensorIndex;this.outputLayers.push(w),this.outputLayersNodeIndices.push(v),this.outputLayersTensorIndices.push(k)}for(let b of this.inputs){let w=b.sourceLayer,v=b.nodeIndex,k=b.tensorIndex;Zn(v===0,"input layer has >1 nodes"),Zn(k===0,"input layer has >1 tensors"),this.inputLayers.push(w),this.inputLayersNodeIndices.push(v),this.inputLayersTensorIndices.push(k)}this.inputNames=[],this.outputNames=[],this.feedInputShapes=[],this.feedInputNames=[],this.feedOutputNames=[];for(let b=0;b<this.inputLayers.length;b++){let w=this.inputLayers[b];if(!(w instanceof Bi))throw new TypeError(`Input layers to a LayersModel must be InputLayer objects. Received inputs: ${t.inputs}. Input ${b} (0-based) originates from layer type ${w.getClassName()}.`);this.inputNames.push(w.name),this.feedInputShapes.push(w.batchInputShape),this.feedInputNames.push(w.name)}for(let b of this.outputLayers)this.outputNames.push(b.name);this.internalInputShapes=this.inputs.map(b=>b.shape),this.internalOutputShapes=this.outputs.map(b=>b.shape);let e={},n={},o={},s={},i={},a=[],u=(b,w,v,k,E,D)=>{(k==null||E==null||D==null)&&(k=b.sourceLayer,E=b.nodeIndex,D=b.tensorIndex);let $=k.inboundNodes[E];if(v.indexOf($)!==-1)throw new Xr(`The tensor ${b.name} at layer "${k.name}" is part of a cycle.`);if(w.indexOf($)!==-1)return;this.containerNodes.add(Jn.nodeKey(k,E)),k.id in i||(i[k.id]=Object.keys(i).length),v.indexOf($)===-1&&v.push($);let R=$.inboundLayers.length;for(let M=0;M<R;M++){let W=$.inputTensors[M],U=$.inboundLayers[M],q=$.nodeIndices[M],K=$.tensorIndices[M];u(W,w,v,U,q,K)}for(w.push($);v.indexOf($)>=0;)v.splice(v.indexOf($),1);a.push($)},l=[],c=[];for(let b of this.outputs)u(b,l,c);let p=a.slice().reverse();for(let b of p){n[b.id]=b,b.id in e||(e[b.id]=0);let w=e[b.id],v=o[b.outboundLayer.id]==null?0:o[b.outboundLayer.id];w=Math.max(w,v),o[b.outboundLayer.id]=w,s[b.outboundLayer.id]=b.outboundLayer,e[b.id]=w;for(let k=0;k<b.inboundLayers.length;k++){let E=b.inboundLayers[k],D=b.nodeIndices[k],$=E.inboundNodes[D],R=e[$.id]==null?0:e[$.id];e[$.id]=Math.max(w+1,R),n[$.id]=$}}let m={};for(let b in e){let w=e[b];w in m||(m[w]=[]),m[w].push(n[b])}let f={};for(let b in o){let w=o[b];w in f||(f[w]=[]),f[w].push(s[b])}let d=Object.keys(f).map(b=>parseInt(b,10)).sort(hd);this.layers=[];for(let b of d){let w=f[b];w.sort((v,k)=>{let E=i[v.id],D=i[k.id];return E<D?-1:E>D?1:0});for(let v of w)v instanceof Jn&&this.internalContainerRefs.push(v),this.layers.push(v)}this.layersByDepth=f,d=Object.keys(m).map(b=>parseInt(b,10)).sort(hd);let h=this.inputs.slice(),g=[];for(let b of d)for(let w of m[b]){let v=w.outboundLayer;if(v!=null){for(let k of w.inputTensors)if(h.indexOf(k)===-1)throw new Xr(`Graph disconnected: cannot obtain value for tensor ${k} at layer "${v.name}". The following previous layers were accessed without issue: ${g}`);for(let k of w.outputTensors)h.push(k);g.push(v.name)}}this.nodesByDepth=m;let y=this.layers.map(b=>b.name);for(let b of y){let w=y.filter(v=>v===b).length;if(w!==1)throw new Xr(`The name "${b}" is used ${w} times in the model. All layer names should be unique. Layer names: `+JSON.stringify(y))}this.outboundNodes=[],this.inboundNodes=[],new Kl({outboundLayer:this,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:this.inputs,outputTensors:this.outputs,inputMasks:this.inputs.map(b=>null),outputMasks:this.outputs.map(b=>null),inputShapes:this.inputs.map(b=>b.shape),outputShapes:this.outputs.map(b=>b.shape)}),this.built=!0,this._refCount=1}assertNotDisposed(){if(this._refCount===0)throw new Error(`Container '${this.name}' is already disposed.`)}dispose(){this.assertNotDisposed();let t={refCountAfterDispose:null,numDisposedVariables:0};if(--this._refCount===0){for(let e of this.layers)t.numDisposedVariables+=e.dispose().numDisposedVariables;for(let e of this.internalContainerRefs)t.numDisposedVariables+=e.dispose().numDisposedVariables}return t.refCountAfterDispose=this._refCount,t}get trainable(){return this.trainable_}set trainable(t){this.layers.forEach(e=>{e._trainableWeights.forEach(n=>n.trainable=t)}),this.trainable_=t}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 t=[];for(let e of this.layers)t=t.concat(e.trainableWeights);return t}get nonTrainableWeights(){let t=[];for(let e of this.layers)t.push(...e.nonTrainableWeights);if(!this.trainable){let e=[];for(let n of this.layers)e.push(...n.trainableWeights);return e.concat(t)}return t}get weights(){return this.trainableWeights.concat(this.nonTrainableWeights)}loadWeights(t,e=!0){let n={},o=0;for(let i of this.layers)for(let a of i.weights){if(n[a.originalName]!=null)throw new z(`Duplicate weight name: ${a.originalName}`);n[a.originalName]=a,o++}let s=[];for(let i in t){let a=i;if(n[i]==null){let u=i.split("/");a=u.slice(0,-2).concat([u[u.length-1]]).join("/")}if(n[a]!=null)s.push([n[a],t[i]]);else if(e)throw new z(`Provided weight data has no target variable: ${i}`);delete n[a]}if(e){let i=[];for(let a in n)i.push(a);if(i.length>0)throw new z(`${i.length} of ${o} weights are not set: ${i}`)}ym(s)}updatedConfig(){let t=this.getConfig(),e={};return e.className=this.getClassName(),e.config=t,e.kerasVersion=`tfjs-layers ${Cm}`,e.backend="TensorFlow.js",e}toJSON(t,e=!0){let n=cy(this.updatedConfig());return e?JSON.stringify(n):n}call(t,e){return B(()=>{t=we(t);let n=new ni;for(let o=0;o<this.inputs.length;++o)n.add(this.inputs[o],t[o]);return yc(this.outputs,n,e)})}computeMask(t,e){return B(()=>{t=we(t);let n;return e==null?n=Io(null,t.length):n=we(e),this.runInternalGraph(t,n)[1]})}computeOutputShape(t){let e=gm(t);if(e.length!==this.inputLayers.length)throw new z(`Invalid inputShape argument ${t}: model has ${this.inputLayers.length} tensor inputs.`);let n={};for(let a=0;a<e.length;a++){let u=this.inputLayers[a],l=e[a],c=u.name+"_0_0";n[c]=l}let o=Object.keys(this.nodesByDepth).map(a=>parseInt(a,10)).sort(hd);if(o.length>1)for(let a of o){let u=this.nodesByDepth[a];for(let l of u){let c=l.outboundLayer;if(this.inputLayers.map(h=>h.id).indexOf(c.id)!==-1)continue;let p=[];for(let h=0;h<l.inboundLayers.length;h++){let g=l.inboundLayers[h],y=l.nodeIndices[h],b=l.tensorIndices[h],w=`${g.name}_${y}_${b}`,v=n[w];p.push(v)}let m=c.computeOutputShape(_r(p)),f=gm(m),d=c.inboundNodes.indexOf(l);for(let h=0;h<f.length;h++){let g=`${c.name}_${d}_${h}`;n[g]=f[h]}}}let s=[],i=[];for(let a=0;a<this.outputLayers.length;a++){let u=this.outputLayers[a],l=this.outputLayersNodeIndices[a],c=this.outputLayersTensorIndices[a],p=`${u.name}_${l}_${c}`;i.push(p)}for(let a=0;a<i.length;a++){let u=i[a];Zn(u in n),s.push(n[u])}return _r(s)}runInternalGraph(t,e){e==null&&(e=Io(null,t.length));let n={};for(let u=0;u<this.inputs.length;++u){let l=this.inputs[u],c=t[u],p=e[u];n[l.id]=[c,p]}let o=Object.keys(this.nodesByDepth).map(u=>parseInt(u,10)).sort(hd);for(let u of o){let l=this.nodesByDepth[u];for(let c of l){let p=c.outboundLayer,m=c.inputTensors,f=c.outputTensors,d=new Array;for(let h of m)h.id in n&&d.push(n[h.id]);if(d.length===m.length){let h={},g,y,b,w;if(c.callArgs!=null&&(h=c.callArgs),d.length===1){let[v,k]=d[0];h.mask==null&&(h.mask=k),b=we(p.call(v,h)),w=we(p.computeMask(v,k)),g=[v],y=[k]}else g=d.map(v=>v[0]),y=d.map(v=>v[1]),h.mask==null&&(h.mask=y),b=we(p.call(g,h)),w=we(p.computeMask(g,y));if(p.activityRegularizer)throw new Tt("LayersModel invocation with concrete Tensor value(s) in the presence of activity regularizer(s) is not supported yet.");for(let v=0;v<f.length;++v){let k=f[v],E=b[v],D=w[v];n[k.id]=[E,D]}}}}let s=[],i=[],a=[];for(let u of this.outputs){Zn(u.id in n,`Could not compute output ${u.name} : ${u.id}`);let[l,c]=n[u.id];a.push(l.shape),s.push(l),i.push(c)}return[s,i,a]}buildNodeConversionMap(t){let e={},n;for(let o of this.layers){n=o instanceof Jn?1:0;for(let s=0;s<o.inboundNodes.length;s++){let i=Jn.nodeKey(o,s);this.containerNodes.has(i)&&(e[i]=n,n+=1)}}return e}getLayer(t,e){if(e!=null){if(this.layers.length<=e)throw new z(`Was asked to retrieve layer at index ${e}, but model only has ${this.layers.length} layer(s).`);return this.layers[e]}else if(t==null)throw new z("Provide either a layer name or layer index");for(let n of this.layers)if(n.name===t)return n;throw new z(`No such layer: ${t}`)}calculateLosses(){return B(()=>{let t=[];for(let e of this.layers)for(let n=0;n<e.inboundNodes.length;++n){let o=Jn.nodeKey(e,n);this.containerNodes.has(o)&&t.push(...e.calculateLosses())}return t})}getConfig(){let t={name:this.name},e=this.buildNodeConversionMap(this.layers),n=[];for(let i of this.layers){let a=i.getClassName(),u=i.getConfig(),l=[];for(let p=0;p<i.inboundNodes.length;p++){let m=i.inboundNodes[p],f=Jn.nodeKey(i,p),d={};if(this.containerNodes.has(f)){if(m.callArgs)try{JSON.stringify(m.callArgs),d=m.callArgs}catch(h){console.warn(`Layer ${i.name} was passed non-serializable keyword arguments: ${m.callArgs}. They will not be included in the serialized model (and thus will be missing at deserialization time).`),d={}}if(m.inboundLayers.length>0){let h=[];for(let g=0;g<m.inboundLayers.length;g++){let y=m.inboundLayers[g],b=m.nodeIndices[g],w=m.tensorIndices[g],v=Jn.nodeKey(y,b),k=e[v];k==null&&(k=0),h.push([y.name,k,w,d])}l.push(h)}}}let c={};c.name=i.name,c.className=a,c.config=u,c.inboundNodes=l,n.push(c)}t.layers=n;let o=[];for(let i=0;i<this.inputLayers.length;i++){let a=this.inputLayers[i],u=this.inputLayersNodeIndices[i],l=Jn.nodeKey(a,u);if(!this.containerNodes.has(l))continue;let c=e[l];c==null&&(c=0);let p=this.inputLayersTensorIndices[i];o.push([a.name,c,p])}t.inputLayers=o;let s=[];for(let i=0;i<this.outputLayers.length;i++){let a=this.outputLayers[i],u=this.outputLayersNodeIndices[i],l=Jn.nodeKey(a,u);if(!this.containerNodes.has(l))continue;let c=e[l];c==null&&(c=0);let p=this.outputLayersTensorIndices[i];s.push([a.name,c,p])}return t.outputLayers=s,t}static fromConfig(t,e,n={},o=!1){let s={},i={};function a(g,y){g.name in i?i[g.name].push(y):i[g.name]=[y]}function u(g,y){let b=[],w;for(let v of y){let k=v[0],E=v[1],D=v[2];if(w=v[3]==null?{}:v[3],!(k in s)){a(g,y);return}let $=s[k];if($.inboundNodes.length<=E){a(g,y);return}let R=$.inboundNodes[E];b.push(R.outputTensors[D])}b.length>0&&g.apply(_r(b),w)}function l(g){let y=g.name,b=xn(g,e.customObjects!=null?e.customObjects:{});b.setFastWeightInitDuringBuild(o),s[y]=b,g.inboundNodes.forEach(v=>{if(!(v instanceof Array))throw new z(`Corrupted configuration, expected array for nodeData: ${v}`);a(b,v)})}let c=e.name,p=e.layers;for(let g of p)l(g);for(;!R2(i);)for(let g of p){let y=s[g.name];if(y.name in i){let b=i[y.name];delete i[y.name];for(let w of b)u(y,w)}}let m=[],f=[],d=e.inputLayers;for(let g of d){let y=g[0],b=g[1],w=g[2];Zn(y in s);let k=s[y].inboundNodes[b].outputTensors;m.push(k[w])}let h=e.outputLayers;for(let g of h){let y=g[0],b=g[1],w=g[2];Zn(y in s);let k=s[y].inboundNodes[b].outputTensors;f.push(k[w])}return new t({inputs:m,outputs:f,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 t of this.layers)if(t.stateful)return!0;return!1}resetStates(){B(()=>{this.layers.forEach(t=>{t.stateful&&t.resetStates()})})}};function s8(r,t,e){let n=t.length;if(r==null||Array.isArray(r)&&r.length===0)return t.map(o=>null);if(n===1)return Array.isArray(r)&&r.length===1?r:typeof r=="object"&&t[0]in r?[r[t[0]]]:[r];if(Array.isArray(r)){if(r.length!==n)throw new Error(`Provided ${e} is an array of ${r.length} element(s), but the model has ${n} outputs. Make sure a set of weights is provided for each model output.`);return r}else if(typeof r=="object"&&Object.keys(r).length>0&&typeof r[Object.keys(r)[0]]=="object"){let o=[];return t.forEach(s=>{s in r?o.push(r[s]):o.push(null)}),o}else throw new Error(`The model has multiple (${n}) outputs, so ${e} must be either an array with ${n} elements or an object with ${t} keys. Provided ${e} not understood: ${JSON.stringify(r)}`)}function py(r,t){return s8(r,t,"classWeight")}async function my(r,t,e,n){if(t!=null||n!=null)throw new Error("Support sampleWeight is not implemented yet");if(e!=null){let o=B(()=>{if(r.shape.length===1)return In(r);if(r.shape.length===2){if(r.shape[1]>1)return Hs(r,1);if(r.shape[1]===1)return O(r,[r.shape[0]]);throw new Error(`Encountered unexpected last-dimension size (${r.shape[1]}) during handling of class weights. The size is expected to be >= 1.`)}else throw new Error(`Unexpected rank of target (y) tensor (${r.rank}) during handling of class weights. The rank is expected to be 1 or 2.`)}),s=Array.from(await o.data());_t(o);let i=[];return s.forEach(a=>{if(e[a]==null)throw new Error(`classWeight must contain all classes in the training data. The class ${a} exists in the data but not in classWeight`);i.push(e[a])}),Re(i,"float32")}else return null}function bD(r,t){return P(r,t)}var i8=32;function CD(r,t){let e,n,o=t;e=o.xs,n=o.ys,x.assert(e!=null&&n!=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=wD("input",r.inputNames,e),i=wD("output",r.outputNames,n),a=s[0].shape[0];x.assert(s.length===r.inputs.length,()=>`LayersModel has ${r.inputs.length} inputs, but the dataset provides ${s.length} inputs. (Expected input keys: ${JSON.stringify(r.inputNames)})`),x.assert(i.length===r.outputs.length,()=>`LayersModel has ${r.outputs.length} outputs, but the dataset provides ${i.length} outputs. (Expected output keys: ${JSON.stringify(r.outputNames)})`);for(let u=0;u<s.length;u++)x.assert(s[u].shape[0]===a,()=>`Batch size mismatch: input ${r.inputNames[u]} has ${s[u].shape[0]}; expected ${a} based on input ${r.inputNames[0]}.`);for(let u=0;u<i.length;u++)x.assert(i[u].shape[0]===a,()=>`Batch size mismatch: output ${r.outputNames[u]} has ${i[u].shape[0]}; expected ${a} based on input ${r.inputNames[0]}.`);return{xs:s,ys:i}}function wD(r,t,e){if(e instanceof Vt)return[e];if(Array.isArray(e))return x.assert(e.length===t.length,()=>`Received an array of ${e.length} Tensors, but expected ${t.length} to match the ${r} keys ${t}.`),e;{let n=[];for(let o of t){if(e[o]==null)throw new z(`The feature data generated by the dataset lacks the required ${r} key '${o}'.`);n.push(e[o])}return n}}function a8(r){if(r.length===3)throw new Tt("Validation with sample weights is not implemented yet.");return{xs:r[0],ys:r[1]}}async function ID(r,t,e){let n=e.batchesPerEpoch!=null;if(x.assert(r.optimizer!=null,()=>"You must compile a model before training/testing. Use LayersModel.compile(modelCompileConfig)."),x.assert(e!=null,()=>"For fitDataset(), the 2nd argument (config) is required, but it is not provided in this call."),x.assert(e.epochs!=null&&e.epochs>0&&Number.isInteger(e.epochs),()=>`For fitDataset(), config.epochs is expected to be a positive integer, but got ${e.epochs}`),x.assert(!n||e.batchesPerEpoch>0&&Number.isInteger(e.batchesPerEpoch),()=>`For fitDataset(), config.batchesPerEpoch is expected to be a positive integer if specified, but got ${e.batchesPerEpoch}`),x.assert(e.validationSplit==null,()=>"`validationSplit` is not supported by `fitDataset()`. Use validationData instead."),r.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");r.isTraining=!0;try{let o=e.validationData!=null,s,i;if(o)if(vD(e.validationData))x.assert(e.validationBatches==null||e.validationBatches>0&&Number.isInteger(e.validationBatches),()=>`For fitDataset() with dataset-based validation, config.validationBatches is expected not to be provided, or to be a positive integer, but got ${e.validationBatches}`);else{let g=a8(e.validationData);s=g.xs,i=g.ys}let a=r.makeTrainFunction(),u=r.getDedupedMetricsNames(),l;o?l=u.slice().concat(u.map(g=>"val_"+g)):l=u.slice();let c=ny(e.callbacks,e.yieldEvery),p=e.verbose==null?1:e.verbose,{callbackList:m,history:f}=oy(c,p,e.epochs,null,null,l8(t,e),null,o,l);m.setModel(r),r.history=f,await m.onTrainBegin(),r.stopTraining_=!1;let d=e.initialEpoch==null?0:e.initialEpoch,h=await t.iterator();for(;d<e.epochs;){let g={};await m.onEpochBegin(d);let y=0,b=0;for(n||(h=await t.iterator());!n||y<e.batchesPerEpoch;){let w=await h.next();if(n&&w.done){console.warn(`You provided \`batchesPerEpoch\` as ${e.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, ${e.batchesPerEpoch*e.epochs} batches). You may need to use the repeat() function when building your dataset.`);break}if(w.value!=null){let{xs:v,ys:k}=CD(r,w.value),E={};E.batch=b,E.size=v[0].shape[0],await m.onBatchBegin(b,E);let D=[];if(e.classWeight!=null){let M=py(e.classWeight,r.outputNames);for(let W=0;W<M.length;++W)D.push(await my(k[W],null,M[W]))}let $=v.concat(k).concat(D),R=a($);_t($);for(let M=0;M<u.length;++M){let W=u[M],U=R[M];E[W]=U,Me(U)}await m.onBatchEnd(b,E),ry(E),b++,y++}if(n?y>=e.batchesPerEpoch:w.done){if(o){let v;vD(e.validationData)?v=we(await r.evaluateDataset(e.validationData,{batches:e.validationBatches})):v=we(r.evaluate(s,i,{batchSize:e.validationBatchSize==null?i8:e.validationBatchSize,verbose:0}));for(let k=0;k<r.metricsNames.length;++k)g[`val_${r.metricsNames[k]}`]=v[k]}break}if(r.stopTraining_)break}if(await m.onEpochEnd(d,g),d++,r.stopTraining_)break}return await m.onTrainEnd(),await r.history.syncData(),r.history}finally{r.isTraining=!1}}function l8(r,t){let e=null;return t.batchesPerEpoch!=null?e=t.batchesPerEpoch:Number.isFinite(r.size)&&(e=r.size),e}function vD(r){return typeof r.iterator=="function"}function u8(r){return typeof r.next=="function"}async function SD(r,t,e){e=e||{};let n=e.batches!=null,o=r.testFunction,s=[];if(e.verbose>0)throw new Tt("Verbose mode is not implemented yet.");x.assert(!n||e.batches>0&&Number.isInteger(e.batches),()=>`Test loop expects \`batches\` to be a positive integer, but received ${JSON.stringify(e.batches)}`);let i=u8(t)?t:await t.iterator(),a=0,u=0;for(;!n||u<e.batches;){let l=await i.next();if(s=B(()=>{if(l.value){let{xs:c,ys:p}=CD(r,l.value),m=c.concat(p),f=B(()=>o(m));if(_t(m),u===0)for(let h=0;h<f.length;++h)s.push(pt(0));let d=m[0].shape[0];for(let h=0;h<f.length;++h){let g=f[h],y=s[h];s[h]=B(()=>J(s[h],P(d,g))),u>0&&_t(y)}_t(f),a+=d,++u}return s}),l.done){n&&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, ${e.batches} batches). You may need to use the repeat() function when building your dataset.`);break}}for(let l=0;l<s.length;++l){let c=s[l];s[l]=ct(s[l],a),_t(c)}return _r(s)}function fy(r){x.assert(r>0&&Number.isInteger(r),()=>`batchSize is required to be a positive integer, but got ${r}`)}function Im(r,t,e){return r==null?[null]:Array.isArray(r)?r.map(n=>nl(n,t,e-t)):nl(r,t,e-t)}function dy(r,t){return B(()=>r==null?null:Array.isArray(r)?r.map(e=>dy(e,t)):Xx(r,t.dtype==="int32"?t:Z(t,"int32")))}function hy(r,t){let e=[],n=0,o=null;for(;n<r;)o=n+t,o>=r&&(o=r),e.push([n,o]),n=o;return e}async function c8(r,t,e,n,o,s,i,a,u,l,c,p,m,f,d){o==null&&(o=32),s==null&&(s=1),c==null&&(c=!0),m==null&&(m=0);let h=!1;if(u!=null&&l!=null&&(h=!0),d!=null&&(h=!0,f==null))throw new z("Can only use `validationSteps` when doing step-wise training, i.e., `stepsPerEpoch` must be set.");let g=r.checkNumSamples(e,o,f,"steps_per_epoch"),y;g!=null&&(y=Yr(0,g)),i==null&&(i=1);let{callbackList:b,history:w}=oy(a,i,s,m,g,f,o,h,p);b.setModel(r),r.history=w,await b.onTrainBegin(),r.stopTraining_=!1;for(let v=m;v<s;++v){await b.onEpochBegin(v);let k={};if(f!=null)throw new Tt("stepsPerEpoch mode is not implemented yet.");{if(c==="batch")throw new Tt("batch shuffling is not implemneted yet");c&&x.shuffle(y);let E=Re(y),D=hy(g,o);for(let $=0;$<D.length;++$){let R={};if(await b.onBatchBegin($,R),B(()=>{let M=D[$][0],W=D[$][1],U=nl(E,M,W-M);R.batch=$,R.size=W-M;let q=dy(e,U),K=t(q);for(let j=0;j<n.length;++j){let Q=n[j],rt=K[j];R[Q]=rt,Me(rt)}if($===D.length-1&&h){let j=r.testLoop(u,l,o);for(let Q=0;Q<n.length;++Q){let rt=n[Q],X=j[Q];Me(X),k["val_"+rt]=X}}}),await b.onBatchEnd($,R),ry(R),r.stopTraining_)break}E.dispose()}if(await b.onEpochEnd(v,k),r.stopTraining_)break}return await b.onTrainEnd(),await r.history.syncData(),r.history}async function kD(r,t,e,n={}){if(r.isTraining)throw new Error("Cannot start training because another fit() call is ongoing.");r.isTraining=!0;let o,s,i,a,u,l,c,p,m;try{let f=n.batchSize==null?32:n.batchSize;fy(f);let d=!1,h=await r.standardizeUserData(t,e,n.sampleWeight,n.classWeight,d,f);o=h[0],s=h[1],m=h[2];let g=!1,y;if(n.validationData!=null&&n.validationData.length>0){if(g=!0,n.validationData.length===2)u=n.validationData[0],l=n.validationData[1];else throw n.validationData.length===3?new Tt("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; ${n.validationData} is invalid.`);let R=!0,M=await r.standardizeUserData(u,l,null,null,R,f);c=M[0],p=M[1],y=c.concat(p)}else if(n.validationSplit!=null&&n.validationSplit>0&&n.validationSplit<1){g=!0;let R=Math.floor(o[0].shape[0]*(1-n.validationSplit)),M=o[0].shape[0];c=Im(o,R,M),i=o,o=Im(o,0,R),p=Im(s,R,M),a=s,s=Im(s,0,R),y=c.concat(p)}else n.validationSteps!=null&&(g=!0);let b=o.concat(s).concat(m);r.checkTrainableWeightsConsistency();let w=r.makeTrainFunction(),v=r.getDedupedMetricsNames(),k,E;g?(r.makeTestFunction(),k=r.testFunction,E=v.slice().concat(v.map(R=>"val_"+R))):(k=null,y=[],E=v.slice());let D=ny(n.callbacks,n.yieldEvery);return await c8(r,w,b,v,f,n.epochs,n.verbose,D,k,y,n.shuffle,E,n.initialEpoch,null,null)}finally{r.isTraining=!1,_o(o,t),_o(s,e),_o(i,t),_o(a,e),_o(c,u),_o(p,l),m!=null&&_t(m)}}function EC(r){let t=[];r instanceof Vt&&(r=[r]);for(let e=0;e<r.length;++e){let n=r[e];if(n.rank===1)t.push(ol(n,1));else{if(n.rank===0)throw new Error("Expected tensor to be at least 1D, but received a 0D tensor (scalar).");t.push(n)}}return t}function _o(r,t){if(r==null)return;let e=[];if(t instanceof Vt)e.push(t.id);else if(Array.isArray(t))t.forEach(o=>e.push(o.id));else if(t!=null)for(let o in t){let s=t[o];e.push(s.id)}let n=[];if(r instanceof Vt)e.indexOf(r.id)===-1&&n.push(r);else if(Array.isArray(r))r.forEach(o=>{e.indexOf(o.id)===-1&&n.push(o)});else if(r!=null)for(let o in r){let s=r[o];e.indexOf(s.id)===-1&&n.push(s)}n.forEach(o=>{o.isDisposed||o.dispose()})}function p8(r){return r instanceof Vt}function AC(r){return Array.isArray(r)}function ND(r){return!p8(r)&&!AC(r)}function TD(r,t,e,n=!0,o=""){if(t==null||t.length===0){if(r!=null){let i=!1;if(AC(r)&&r.length>0)i=!0;else if(ND(r)){for(let a in r)if(r.hasOwnProperty(a)){i=!0;break}}else i=!0;if(i)throw new z(`Error when checking model ${o} expected no data, but got ${r}`)}return[]}if(r==null)return t.map(i=>null);let s;if(ND(r)){r=r,s=[];for(let i of t){if(r[i]==null)throw new z(`No data provided for "${i}". Need data for each key in: ${t}`);s.push(r[i])}}else if(AC(r)){if(r=r,r.length!==t.length)throw new z(`Error when checking model ${o}: 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): ${r}`);s=r}else{if(r=r,t.length>1)throw new z(`The model ${o} expects ${t.length} Tensor(s), but only received one Tensor. Found: Tensor with shape ${r.shape}`);s=[r]}if(s=EC(s),e!=null)for(let i=0;i<t.length;++i){if(e[i]==null)continue;let a=s[i];if(a.shape.length!==e[i].length)throw new z(`Error when checking ${o}: expected ${t[i]} to have ${e[i].length} dimension(s). but got array with shape ${a.shape}`);for(let u=0;u<e[i].length;++u){if(u===0&&!n)continue;let l=a.shape[u],c=e[i][u];if(c!=null&&c>=0&&l!==c)throw new z(`${o} expected a batch of elements where each example has shape [${e[i].slice(1,e[i].length)}] (i.e.,tensor shape [*,${e[i].slice(1,e[i].length)}]) but the ${o} received an input with ${a.shape[0]} examples, each with shape [${a.shape.slice(1,a.shape.length)}] (tensor shape [${a.shape}])`)}}return s}function m8(r,t,e){let n=ko(r.map(s=>s.shape[0]));n.sort();let o=ko(t.map(s=>s.shape[0]));if(o.sort(),n.length>1)throw new z(`All input Tensors (x) should have the same number of samples. Got array shapes: ${JSON.stringify(r.map(s=>s.shape))}`);if(o.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(n.length>0&&o.length>0&&!x.arraysEqual(n,o))throw new z(`Input Tensors should have the same number of samples as target Tensors. Found ${n[0]} input sample(s) and ${o[0]} target sample(s).`)}function f8(r,t,e){let n=[Gi,vm,gc];for(let o=0;o<r.length;++o){let s=r[o],i=t[o],a=e[o];if(i!=null){if(i===gc&&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(n.indexOf(i)!==-1){let u=s.shape.slice(1),l=a.slice(1);for(let c=0;c<u.length;++c){let p=u[c],m=l[c];if(m!=null&&p!==m)throw new z(`A target Tensor with shape ${s.shape} was passed for an output of shape ${a}, while using a loss function that expects targets to have the same shape as the output.`)}}}}}function _D(r,t,e,n=!0,o=""){let s;if(Array.isArray(r)){if(r.length!==t.length)throw new z(`Error when checking model ${o}: 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 ${r.length} Tensors(s).`);s=r}else{if(t.length>1)throw new z(`The model expects ${t.length} ${o} Tensors, but only received one Tensor. Found: array with shape ${JSON.stringify(r.shape)}.`);s=[r]}if(e!=null)for(let i=0;i<t.length;++i){if(e[i]==null)continue;let a=s[i];if(a.shape.length!==e[i].length)throw new z(`Error when checking ${o}: expected ${t[i]} to have ${e[i].length} dimension(s), but got array with shape ${JSON.stringify(a.shape)}`);for(let u=0;u<e[i].length;++u){if(u===0&&!n)continue;let l=a.shape[u],c=e[i][u];if(c!=null&&c!==l)throw new z(`Error when checking ${o}: expected ${t[i]} to have shape ${JSON.stringify(e[i])} but got array with shape ${JSON.stringify(a.shape)}.`)}}}function d8(r,t){if(r==null||Array.isArray(r)&&r.length===0)return t.map(n=>[]);let e;if(typeof r=="string"||typeof r=="function")e=[r];else if(Array.isArray(r)||typeof r=="object")e=r;else throw new TypeError(`Type of metrics argument not understood. Expected an string,function, Array, or Object, found: ${r}`);if(Array.isArray(e))return t.map(n=>e);{let n=[];for(let o of t){let s=e.hasOwnProperty(o)?e[o]:[];Array.isArray(s)||(s=[s]),n.push(s)}return n}}var h8="layers-model",Qn=class extends Jn{constructor(t){super(t);this.isTraining=!1}summary(t,e,n=console.log){if(!this.built)throw new z("This model has never been called, thus its weights have not been created yet. So no summary can be displayed. Build the model first (e.g., by calling it on some test data).");hD(this,t,e,n)}compile(t){if(t.loss==null&&(t.loss=[]),this.loss=t.loss,typeof t.optimizer=="string")this.optimizer_=dD(t.optimizer),this.isOptimizerOwned=!0;else{if(!(t.optimizer instanceof jr))throw new z("User-defined optimizer must be an instance of tf.Optimizer.");this.optimizer_=t.optimizer,this.isOptimizerOwned=!1}let e=[];if(!Array.isArray(t.loss)&&typeof t.loss!="string"&&typeof t.loss!="function"){t.loss=t.loss;for(let i in t.loss)if(this.outputNames.indexOf(i)===-1)throw new z(`Unknown entry in loss dictionary: "${i}". Only expected the following keys: ${this.outputNames}`);for(let i of this.outputNames)t.loss[i]==null&&console.warn(`Output "${i}" is missing from loss dictionary. We assume this was done on purpose, and we will not be expecting data to be passed to ${i} during training`),e.push(sy(t.loss[i]))}else if(Array.isArray(t.loss)){if(t.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 ${this.outputs.length} output(s), but you passed loss=${t.loss}.`);e=t.loss.map(a=>sy(a))}else{let i=sy(t.loss);this.outputs.forEach(a=>{e.push(i)})}this.lossFunctions=e,this.feedOutputNames=[],this.feedOutputShapes=[],this.feedLossFns=[];for(let i=0;i<this.outputs.length;++i){let a=this.internalOutputShapes[i],u=this.outputNames[i];this.feedOutputNames.push(u),this.feedOutputShapes.push(a),this.feedLossFns.push(this.lossFunctions[i])}let n=[];this.metrics=t.metrics,this.metricsNames=["loss"],this.metricsTensors=[],ei("loss",()=>{for(let i=0;i<this.outputs.length;++i){if(n.indexOf(i)!==-1)continue;let a=this.lossFunctions[i];this.outputs.length>1&&(this.metricsTensors.push([a,i]),this.metricsNames.push(this.outputNames[i]+"_loss"))}});let o=d8(t.metrics,this.outputNames),s=(i,a,u)=>{this.outputNames.length>1&&(a=this.outputNames[i]+"_"+a),this.metricsNames.push(a),this.metricsTensors.push([u,i])};ei("metric",()=>{for(let i=0;i<this.outputs.length;++i){if(n.indexOf(i)!==-1)continue;let a=o[i];(l=>{let c="",p,m,f;for(let d of l){if(typeof d=="string"&&["accuracy","acc","crossentropy","ce"].indexOf(d)!==-1){let g=this.internalOutputShapes[i];g[g.length-1]===1||this.lossFunctions[i]===vm?["accuracy","acc"].indexOf(d)!==-1?m=Rd:["crossentropy","ce"].indexOf(d)!==-1&&(m=ay):this.lossFunctions[i]===wm?["accuracy","acc"].indexOf(d)!==-1?m=ly:["crossentropy","ce"].indexOf(d)!==-1&&(m=kC):["accuracy","acc"].indexOf(d)!==-1?m=Fd:["crossentropy","ce"].indexOf(d)!==-1&&(m=Od);let y;["accuracy","acc"].indexOf(d)!==-1?y="acc":["crossentropy","ce"].indexOf(d)!==-1&&(y="ce"),f=m,p=c+y}else f=mD(d),p=c+Pd(d);let h;ei(p,()=>{h=f}),s(i,p,h)}})(a)}}),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(t,e,n={}){let o=n.batchSize==null?32:n.batchSize;fy(o);let s=!0,i=this.standardizeUserDataXY(t,e,s,o);try{let a=i[0].concat(i[1]);this.makeTestFunction();let u=this.testFunction,l=this.testLoop(u,a,o,n.verbose,n.steps);return _r(l)}finally{_o(i[0],t),_o(i[1],e)}}async evaluateDataset(t,e){return this.makeTestFunction(),SD(this,t,e)}checkNumSamples(t,e,n,o="steps"){let s;if(n!=null){if(s=null,e!=null)throw new z(`If ${o} is set, batchSize must be null or undefined.Got batchSize = ${e}`)}else if(t!=null)Array.isArray(t)?s=t[0].shape[0]:s=t.shape[0];else throw new z(`Either the input data should have a defined shape, or ${o} shoud be specified.`);return s}execute(t,e){if(Array.isArray(e)&&e.length===0)throw new z("`outputs` is an empty Array, which is not allowed.");let n=Array.isArray(e),o=n?e:[e],s=this.retrieveSymbolicTensors(o),i=new ni;if(t instanceof Vt&&(t=[t]),Array.isArray(t)){if(t.length!==this.inputs.length)throw new z(`The number of inputs provided (${t.length}) does not match the number of inputs of this model (${this.inputs.length}).`);for(let u=0;u<this.inputs.length;++u)i.add(this.inputs[u],t[u])}else for(let u of this.inputs){let l=t[u.name];if(l==null)throw new z(`No value is provided for the model's input ${u.name}`);i.add(u,l)}let a=yc(s,i);return n?a:a[0]}retrieveSymbolicTensors(t){let e=Io(null,t.length),n=t.length;for(let o of this.layers){let s=Array.isArray(o.output)?o.output:[o.output],i=s.map(a=>a.name);for(let a=0;a<t.length;++a){let u=i.indexOf(t[a]);if(u!==-1&&(e[a]=s[u],n--),n===0)break}if(n===0)break}if(n>0){let o=[];throw e.forEach((s,i)=>{s==null&&o.push(t[i])}),new z(`Cannot find SymbolicTensors for output name(s): ${JSON.stringify(o)}`)}return e}predictLoop(t,e=32,n=!1){return B(()=>{let o=this.checkNumSamples(t);if(n)throw new Tt("Verbose predictLoop() is not implemented yet.");let s=hy(o,e),i=this.outputs.map(a=>[]);for(let a=0;a<s.length;++a)B(()=>{let l=s[a][0],c=s[a][1],p=Im(t,l,c),m=[];if(Array.isArray(p))for(let d=0;d<p.length;++d)m.push({key:this.inputs[d],value:p[d]});else m.push({key:this.inputs[0],value:p});let f=new ni(m);return yc(this.outputs,f)}).forEach((l,c)=>i[c].push(l));return _r(i.map(a=>se(a,0)))})}predict(t,e={}){let n=EC(t);_D(n,this.inputNames,this.feedInputShapes,!1);try{let o=e.batchSize==null?32:e.batchSize;return fy(o),this.predictLoop(n,o)}finally{_o(n,t)}}predictOnBatch(t){_D(t,this.inputNames,this.feedInputShapes,!0);let e=(Array.isArray(t)?t[0]:t).shape[0];return this.predictLoop(t,e)}standardizeUserDataXY(t,e,n=!0,o){if(this.optimizer_==null)throw new Xr("You must compile a model before training/testing. Use LayersModel.compile(modelCompileArgs).");let s=[];for(let i=0;i<this.feedOutputShapes.length;++i){let a=this.feedOutputShapes[i];this.feedLossFns[i]===wm?s.push(a.slice(0,a.length-1).concat([1])):s.push(a)}if(t=TD(t,this.feedInputNames,this.feedInputShapes,!1,"input"),e=TD(e,this.feedOutputNames,s,!1,"target"),m8(t,e,null),f8(e,this.feedLossFns,this.feedOutputShapes),this.stateful&&o!=null&&o>0&&t[0].shape[0]%o!==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 ${o}. Found: ${t[0].shape[0]} sample(s).`);return[t,e]}async standardizeUserData(t,e,n,o,s=!0,i){let[a,u]=this.standardizeUserDataXY(t,e,s,i);if(n!=null)throw new Error("sample weight is not supported yet.");let l=null;if(o!=null){let c=py(o,this.outputNames);l=[];for(let p=0;p<c.length;++p)l.push(await my(u[p],null,c[p]))}return[a,u,l]}testLoop(t,e,n,o=0,s){return B(()=>{let i=this.checkNumSamples(e,n,s,"steps"),a=[];if(o>0)throw new Tt("Verbose mode is not implemented yet.");if(s!=null)throw new Tt("steps mode in testLoop() is not implemented yet");{let u=hy(i,n),l=Re(Yr(0,i));for(let c=0;c<u.length;++c){let p=u[c][0],m=u[c][1],f=nl(l,p,m-p),d=dy(e,f),h=t(d);if(c===0)for(let g=0;g<h.length;++g)a.push(pt(0));for(let g=0;g<h.length;++g){let y=h[g];a[g]=J(a[g],P(m-p,y))}}for(let c=0;c<a.length;++c)a[c]=ct(a[c],i)}return a})}getDedupedMetricsNames(){let t=this.metricsNames,e=[];for(let n=0;n<t.length;++n){let o=t[n],s=o;dC(t,o)>1&&(s+=`_${dC(t.slice(0,n),o)}`),e.push(s)}return e}makeTrainFunction(){return t=>{let e=[],n=t.slice(0,this.inputs.length),o=t.slice(this.inputs.length,this.inputs.length+this.outputs.length),s=t.slice(this.inputs.length+this.outputs.length,this.inputs.length+this.outputs.length*2),i=[],a=()=>{let p=[];for(let h=0;h<this.inputs.length;++h)p.push({key:this.inputs[h],value:n[h]});let m=new ni(p),f=yc(this.outputs,m,{training:!0}),d;for(let h=0;h<this.lossFunctions.length;++h){let y=this.lossFunctions[h](o[h],f[h]);s[h]!=null&&(y=bD(y,s[h]));let b=xe(y);e.push(b),h===0?d=y:d=J(d,y)}for(let h=0;h<this.metricsTensors.length;++h){let g;if(this.outputs.length>1&&h<this.outputs.length)g=e[h];else{let y=this.metricsTensors[h][0],b=this.metricsTensors[h][1];g=xe(y(o[b],f[b]))}Me(g),i.push(g)}return d=xe(d),this.calculateLosses().forEach(h=>{d=J(d,h)}),d},u=this.collectedTrainableWeights.map(p=>p.read()),l=!0;return[this.optimizer_.minimize(a,l,u)].concat(i)}}makeTestFunction(){this.testFunction=t=>B(()=>{let e=[],n,o=t.slice(0,this.inputs.length),s=t.slice(this.inputs.length,this.inputs.length+this.outputs.length),i=[];for(let l=0;l<this.inputs.length;++l)i.push({key:this.inputs[l],value:o[l]});let a=new ni(i),u=yc(this.outputs,a);for(let l=0;l<this.lossFunctions.length;++l){let c=this.lossFunctions[l],p=xe(c(s[l],u[l]));l===0?n=p:n=J(n,p),e.push(n)}for(let l=0;l<this.metricsTensors.length;++l){let c=this.metricsTensors[l][0],p=this.metricsTensors[l][1],m=xe(c(s[p],u[p]));e.push(m)}return e})}async fit(t,e,n={}){return kD(this,t,e,n)}async fitDataset(t,e){return ID(this,t,e)}async trainOnBatch(t,e){let n=await this.standardizeUserData(t,e),o=n[0],s=n[1],a=this.makeTrainFunction()(o.concat(s)),u=[];for(let l of a){let c=await l.data();u.push(c[0])}return _t(a),_o(n[0],t),_o(n[1],e),_r(u)}getNamedWeights(t){let e=[],n=t!=null&&t.trainableOnly,o=n?this.trainableWeights:this.weights,s=this.getWeights(n);for(let i=0;i<o.length;++i)n&&!o[i].trainable||e.push({name:o[i].originalName,tensor:s[i]});return e}set stopTraining(t){this.stopTraining_=t}get stopTraining(){return this.stopTraining_}get optimizer(){return this.optimizer_}set optimizer(t){this.optimizer_!==t&&(this.optimizer_=t,this.isOptimizerOwned=!1)}dispose(){let t=super.dispose();if(t.refCountAfterDispose===0&&this.optimizer!=null&&this.isOptimizerOwned){let e=Af().numTensors;this.optimizer_.dispose(),t.numDisposedVariables+=e-Af().numTensors}return t}getLossIdentifiers(){let t;if(typeof this.loss=="string")t=So(this.loss);else if(Array.isArray(this.loss)){for(let e of this.loss)if(typeof e!="string")throw new Error("Serialization of non-string loss is not supported.");t=this.loss.map(e=>So(e))}else{let e=Object.keys(this.loss);t={};let n=this.loss;for(let o of e)if(typeof n[o]=="string")t[o]=So(n[o]);else throw new Error("Serialization of non-string loss is not supported.")}return t}getMetricIdentifiers(){if(typeof this.metrics=="string"||typeof this.metrics=="function")return[So(Pd(this.metrics))];if(Array.isArray(this.metrics))return this.metrics.map(t=>So(Pd(t)));{let t={};for(let e in this.metrics)t[e]=So(Pd(this.metrics[e]));return t}}getTrainingConfig(){return{loss:this.getLossIdentifiers(),metrics:this.getMetricIdentifiers(),optimizer_config:{class_name:this.optimizer.getClassName(),config:this.optimizer.getConfig()}}}loadTrainingConfig(t){if(t.weighted_metrics!=null)throw new Error("Loading weight_metrics is not supported yet.");if(t.loss_weights!=null)throw new Error("Loading loss_weights is not supported yet.");if(t.sample_weight_mode!=null)throw new Error("Loading sample_weight_mode is not supported yet.");let e=xc(t.optimizer_config),n=xn(e),o;if(typeof t.loss=="string")o=rl(t.loss);else if(Array.isArray(t.loss))o=t.loss.map(i=>rl(i));else if(t.loss!=null){o={};for(let i in t.loss)o[i]=rl(t.loss[i])}let s;if(Array.isArray(t.metrics))s=t.metrics.map(i=>rl(i));else if(t.metrics!=null){s={};for(let i in t.metrics)s[i]=rl(t.metrics[i])}this.compile({loss:o,metrics:s,optimizer:n})}async save(t,e){if(typeof t=="string"){let l=Pr.getSaveHandlers(t);if(l.length===0)throw new z(`Cannot find any save handlers for URL '${t}'`);if(l.length>1)throw new z(`Found more than one (${l.length}) save handlers for URL '${t}'`);t=l[0]}if(t.save==null)throw new z("LayersModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");let n=await Pr.encodeWeights(this.getNamedWeights(e)),o=!1,s=null,a={modelTopology:this.toJSON(s,o),format:h8,generatedBy:`TensorFlow.js tfjs-layers v${Cm}`,convertedBy:null};if((e==null?!1:e.includeOptimizer)&&this.optimizer!=null){a.trainingConfig=this.getTrainingConfig();let l="optimizer",{data:c,specs:p}=await Pr.encodeWeights(await this.optimizer.getWeights(),l);n.specs.push(...p),n.data=Pr.concatenateArrayBuffers([n.data,c])}return this.userDefinedMetadata!=null&&(TC(this.userDefinedMetadata,this.name,!0),a.userDefinedMetadata=this.userDefinedMetadata),a.weightData=n.data,a.weightSpecs=n.specs,t.save(a)}setUserDefinedMetadata(t){TC(t,this.name),this.userDefinedMetadata=t}getUserDefinedMetadata(){return this.userDefinedMetadata}};Qn.className="Model";et.registerClass(Qn);var DC=class extends Qn{};DC.className="Functional";et.registerClass(DC);async function ED(r,t){"modelTopology"in r||(r={modelTopology:r}),r=r;let e=r.modelTopology;e.model_config!=null&&(e=e.model_config);let n=xc(e),o=xn(n,t);if(r.weightsManifest!=null){let s=await Pr.loadWeights(r.weightsManifest,r.pathPrefix,o.weights.map(a=>a.originalName)),i={};for(let a of o.weights)i[a.originalName]=s[a.originalName];o.loadWeights(i),_t(s)}return o}async function AD(r,t){if(t==null&&(t={}),typeof r=="string"){let e=Pr.getLoadHandlers(r,t);if(e.length===0)e.push(Pr.browserHTTPRequest(r,t));else if(e.length>1)throw new z(`Found more than one (${e.length}) load handlers for URL '${r}'`);r=e[0]}return g8(r,void 0,t)}async function g8(r,t,e){if(e==null&&(e={}),r.load==null)throw new z("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");let n=await r.load(),o=n.modelTopology;o.model_config!=null&&(o=o.model_config);let s=e.strict==null?!0:e.strict,i=n.weightData!=null&&n.weightSpecs!=null&&s,a=xn(xc(o),t,i),u=n.trainingConfig;if(u!=null&&a.loadTrainingConfig(u),n.userDefinedMetadata!=null&&a.setUserDefinedMetadata(n.userDefinedMetadata),n.weightData!=null){if(n.weightSpecs==null)throw new z("LayersModel artifacts contains weight data, but not weight specs. Therefore loading of weights cannot proceed.");let{modelWeights:l,optimizerWeights:c}=x8(n.weightData,n.weightSpecs);a.loadWeights(l,s),a.optimizer!=null&&c.length>0&&await a.optimizer.setWeights(c),_t(l),_t(c.map(p=>p.tensor))}return a}function x8(r,t){let e=Pr.decodeWeights(r,t),n={},o=[];return t.forEach(s=>{s.group==="optimizer"?o.push({name:s.name,tensor:e[s.name]}):n[s.name]=e[s.name]}),{modelWeights:n,optimizerWeights:o}}var sl=class extends Qn{constructor(t){super({inputs:[],outputs:[]});if(t=t||{},this.trainable=!0,this.built=!1,this.name=t.name!=null?t.name:ql("sequential_"),t.layers!=null)for(let e of t.layers)this.add(e)}checkShape(t){if(t.inboundNodes[0].outputTensors[0].shape.some(n=>n<0))throw new z(`Negative dimension size caused by adding layer ${t.name} with input shape [${t.inboundNodes[0].inputTensors[0].shape}]`)}add(t){let e=t instanceof sl||t instanceof Qn,n;if(e){if(n=t,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(t.inboundNodes.length===0){if(t.batchInputShape==null)throw new z("The first layer in a Sequential model must get an `inputShape` or `batchInputShape` argument.");let o=ey({batchShape:t.batchInputShape,dtype:t.dtype,name:t.name+"_input"});t.apply(o)}if(e)this.outputs=n.outputs,this.inputs=n.inputs;else{if(t.inboundNodes.length!==1)throw new z(`A layer added to a Sequential model must not already be connected somewhere else. LayersModel received layer ${t.name} which has ${t.inboundNodes.length} pre-existing inbound connections.`);if(t.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(t),this.outputs=[t.inboundNodes[0].outputTensors[0]],this.inputs=wC(this.outputs[0])}this.inboundNodes=[],new Kl({outboundLayer:this,inboundLayers:[],nodeIndices:[],tensorIndices:[],inputTensors:this.inputs,outputTensors:this.outputs,inputMasks:Io(null,this.inputs.length),outputMasks:[null],inputShapes:this.inputs.map(o=>o.shape),outputShapes:this.outputs[0].shape})}else{let o=t.apply(this.outputs[0]);if(Array.isArray(o))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(t),this.outputs=[o],this.inboundNodes[0].outputTensors=this.outputs,this.inboundNodes[0].outputShapes=[this.outputs[0].shape]}this.layers.push(t),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 t=this.layers.length-1;this.layers[t].outboundNodes=[],this.outputs=[this.layers[t].output],this.inboundNodes[0].outputTensors=this.outputs,this.inboundNodes[0].outputShapes=[this.outputs[0].shape]}}call(t,e){return this.model==null&&this.build(),this.model.call(t,e)}build(t){if(Jt(t),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 Qn({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(t,e,n=console.log){this.built||this.build(),super.summary(t,e,n)}setWeights(t){this.model==null&&this.build(),this.model.setWeights(t)}evaluate(t,e,n={}){if(!this.built)throw new Xr("The model needs to be compiled before being used.");return this.model.evaluate(t,e,n)}async evaluateDataset(t,e){if(!this.built)throw new Xr("The model needs to be compiled before being used.");return this.model.evaluateDataset(t,e)}predict(t,e={}){return this.model==null&&this.build(),this.model.predict(t,e)}predictOnBatch(t){return this.model==null&&this.build(),this.model.predictOnBatch(t)}compile(t){this.build(),this.model.compile(t),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(t){this.model.optimizer=t}async fit(t,e,n={}){if(!this.built)throw new Xr("The model needs to be compiled before being used.");return this.model.fit(t,e,n)}async fitDataset(t,e){if(!this.built)throw new Xr("The model needs to be compiled before being used.");return this.model.fitDataset(t,e)}async trainOnBatch(t,e){return this.model.trainOnBatch(t,e)}static fromConfig(t,e,n={},o=!1){let s,i={};if(e instanceof Array){if(e[0].className==null||e[0].className==="Merge")throw new z("Legacy serialization format not supported yet.");s=e}else x.assert(e.layers!=null,()=>"When the config data for a Sequential model is not an Array, it must be an Object that contains the 'layers' field."),s=e.layers,delete e.layers,i=e;let a=new t(i);if(!(a instanceof sl))throw new Tt(`Sequential.fromConfig called on non-Sequential input: ${a}`);for(let u of s){let c=xn(u,void 0,o);o&&c.setFastWeightInitDuringBuild(!0),a.add(c)}return a}set stopTraining(t){if(this.model==null)throw new z("Cannot set the stopTraining property of a sequential model before it is compiled.");this.model.stopTraining=t}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 t=[];for(let e of this.layers){let n={};n.className=e.getClassName(),n.config=e.getConfig(),t.push(n)}return{name:this.name,layers:t}}};sl.className="Sequential";et.registerClass(sl);function y8(r){return new Qn(r)}function b8(r){return new sl(r)}function w8(r,t){return t==null&&(t={}),AD(r,t)}function $C(r){return ey(r)}function v8(r,t){_n.registerCallbackConstructor(r,t)}var yn=class extends et.Serializable{getConfig(){return{}}},FC=class extends yn{apply(t,e=1){return Z2(t,e)}};FC.className="elu";et.registerClass(FC);var OC=class extends yn{apply(t){return Xu(t)}};OC.className="selu";et.registerClass(OC);var PC=class extends yn{apply(t){return Tr(t)}};PC.className="relu";et.registerClass(PC);var MC=class extends yn{apply(t){return B(()=>Zs(6,Tr(t)))}};MC.className="relu6";et.registerClass(MC);var LC=class extends yn{apply(t){return t}};LC.className="linear";et.registerClass(LC);var zC=class extends yn{apply(t){return Lr(t)}};zC.className="sigmoid";et.registerClass(zC);var BC=class extends yn{apply(t){return Q2(t)}};BC.className="hardSigmoid";et.registerClass(BC);var VC=class extends yn{apply(t){return xo(t)}};VC.className="softplus";et.registerClass(VC);var GC=class extends yn{apply(t){return J2(t)}};GC.className="softsign";et.registerClass(GC);var WC=class extends yn{apply(t){return qs(t)}};WC.className="tanh";et.registerClass(WC);var Md=class extends yn{apply(t,e=-1){return Ja(t,e)}};Md.className="softmax";et.registerClass(Md);var UC=class extends yn{apply(t,e=-1){return Gu(t,e)}};UC.className="logSoftmax";et.registerClass(UC);var HC=class extends yn{apply(t,e=1){return B(()=>P(Lr(P(t,e)),t))}};HC.className="swish";et.registerClass(HC);var qC=class extends yn{apply(t){return B(()=>P(t,qs(xo(t))))}};qC.className="mish";et.registerClass(qC);function oi(r){return r.getClassName()}function RC(r,t={}){return Li(r,et.SerializationMap.getMap().classNameMap,t,"activation")}function si(r){if(r==null){let t={};return t.className="linear",t.config={},RC(t)}if(typeof r=="string"){let t={};return t.className=r,t.config={},RC(t)}else return r instanceof yn?r:RC(r)}function KC(r){if(r!=null&&typeof r!="object")throw new Error(`Argument to L1L2 regularizer's constructor is expected to be an object, but received: ${r}`)}var jC=class extends et.Serializable{},bc=class extends jC{constructor(t){super();KC(t),this.l1=t==null||t.l1==null?.01:t.l1,this.l2=t==null||t.l2==null?.01:t.l2,this.hasL1=this.l1!==0,this.hasL2=this.l2!==0}apply(t){return B(()=>{let e=ye([1]);return this.hasL1&&(e=J(e,mt(P(this.l1,Se(t))))),this.hasL2&&(e=J(e,mt(P(this.l2,dc(t))))),O(e,[])})}getConfig(){return{l1:this.l1,l2:this.l2}}static fromConfig(t,e){return new t({l1:e.l1,l2:e.l2})}};bc.className="L1L2";et.registerClass(bc);function RD(r){return KC(r),new bc({l1:r!=null?r.l1:null,l2:0})}function FD(r){return KC(r),new bc({l2:r!=null?r.l2:null,l1:0})}var DD={l1l2:"L1L2"};function pe(r){return sm(r)}function $D(r,t={}){return Li(r,et.SerializationMap.getMap().classNameMap,t,"regularizer")}function ve(r){if(r==null)return null;if(typeof r=="string"){let e={className:r in DD?DD[r]:r,config:{}};return $D(e)}else return r instanceof jC?r:$D(r)}var Ld=class extends Ut{constructor(t){super(t==null?{}:t);this.supportsMasking=!0,t!=null&&(this.maxValue=t.maxValue)}call(t,e){t=Pt(t);let n=Tr(t);return this.maxValue!=null&&(n=br(n,0,this.maxValue)),n}computeOutputShape(t){return t}getConfig(){let t={maxValue:this.maxValue},e=super.getConfig();return Object.assign(t,e),t}};Ld.className="ReLU";et.registerClass(Ld);var zd=class extends Ut{constructor(t){super(t==null?{}:t);this.DEFAULT_ALPHA=.3,t==null&&(t={}),this.alpha=t.alpha==null?this.DEFAULT_ALPHA:t.alpha}call(t,e){let n=Pt(t);return Ha(n,this.alpha)}computeOutputShape(t){return t}getConfig(){let t={alpha:this.alpha},e=super.getConfig();return Object.assign(t,e),t}};zd.className="LeakyReLU";et.registerClass(zd);var Bd=class extends Ut{constructor(t){super(t==null?{}:t);if(this.DEFAULT_ALPHA_INITIALIZER="zeros",t==null&&(t={}),this.supportsMasking=!0,this.alphaInitializer=de(t.alphaInitializer||this.DEFAULT_ALPHA_INITIALIZER),this.alphaRegularizer=ve(t.alphaRegularizer),this.alphaConstraint=Ue(t.alphaConstraint),t.sharedAxes==null)this.sharedAxes=null;else if(Array.isArray(t.sharedAxes))this.sharedAxes=t.sharedAxes;else if(typeof t.sharedAxes=="number")this.sharedAxes=[t.sharedAxes];else throw new z(`Expected sharedAxes to be a number or an array of numbers, but got ${t.sharedAxes}`)}build(t){t=Jt(t);let e=t.slice(1);if(this.sharedAxes!=null)for(let o of this.sharedAxes)e[o-1]=1;this.alpha=this.addWeight("alpha",e,"float32",this.alphaInitializer,this.alphaRegularizer,!0,this.alphaConstraint);let n={};if(this.sharedAxes!=null)for(let o=1;o<t.length;++o)n[o]=t[o];this.inputSpec=[new _e({ndim:t.length,axes:n})],this.built=!0}call(t,e){return t=Pt(t),Ya(t,this.alpha.read())}getConfig(){let t={alphaInitializer:Te(this.alphaInitializer),alphaRegularizer:pe(this.alphaRegularizer),alphaConstraint:We(this.alphaConstraint),sharedAxes:this.sharedAxes},e=super.getConfig();return Object.assign(t,e),t}};Bd.className="PReLU";et.registerClass(Bd);var Vd=class extends Ut{constructor(t){super(t==null?{}:t);if(this.DEFAULT_ALPHA=1,t==null&&(t={}),t.alpha!=null&&t.alpha!==this.DEFAULT_ALPHA)throw new Tt(`Non-default alpha value (${t.alpha}) is not supported by the ELU layer yet.`);this.alpha=t.alpha==null?this.DEFAULT_ALPHA:t.alpha}call(t,e){let n=Pt(t);return js(n)}computeOutputShape(t){return t}getConfig(){let t={alpha:this.alpha},e=super.getConfig();return Object.assign(t,e),t}};Vd.className="ELU";et.registerClass(Vd);var Gd=class extends Ut{constructor(t){super(t==null?{}:t);this.DEFAULT_THETA=1,t==null&&(t={}),this.theta=t.theta==null?this.DEFAULT_THETA:t.theta}call(t,e){let n=Pt(t);return P(n,Z(Ge(n,this.theta),"float32"))}computeOutputShape(t){return t}getConfig(){let t={theta:this.theta},e=super.getConfig();return Object.assign(t,e),t}};Gd.className="ThresholdedReLU";et.registerClass(Gd);var Wd=class extends Ut{constructor(t){super(t==null?{}:t);this.DEFAULT_AXIS=1,t==null&&(t={}),this.softmax=new Md().apply,this.axis=t.axis==null?this.DEFAULT_AXIS:t.axis}call(t,e){let n=Pt(t);return this.softmax(n,this.axis)}computeOutputShape(t){return t}getConfig(){let t={axis:this.axis},e=super.getConfig();return Object.assign(t,e),t}};Wd.className="Softmax";et.registerClass(Wd);function Yl(r,t,e){if(typeof r=="number")return Io(r,t);if(r.length!==t)throw new z(`The ${e} argument must be an integer or tuple of ${t} integers. Received: ${r.length} elements.`);for(let n=0;n<t;++n){let o=r[n];if(!K2(o))throw new z(`The ${e} argument must be an integer or tuple of ${t} integers. Received: ${JSON.stringify(r)} including a non-integer number ${o}`)}return r}function En(r,t,e,n,o=1){if(r==null)return r;let s=t+(t-1)*(o-1),i;return e==="same"?i=r:i=r-s+1,Math.floor((i+n-1)/n)}function ii(r,t,e,n){if(r==null)return null;if(n==="valid")r=r*t+ri([e-t,0]);else if(n==="same")r=r*t;else throw new z(`Unsupport padding mode: ${n}.`);return r}function Ud(r,t){return B(()=>(Le(t),t==="channelsFirst"?Gt(r,[0,2,3,1]):r))}function XC(r,t){return B(()=>(Le(t),t==="channelsFirst"?Gt(r,[0,2,3,4,1]):r))}function I8(r,t,e,n=1,o="valid",s,i=1){return B(()=>{if(s==null&&(s=mn()),Le(s),r.shape.length!==3)throw new z(`The input of a conv1dWithBias operation should be 3, but is ${r.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(e!=null&&e.shape.length!==1)throw new z(`The bias for a conv1dWithBias operation should be 1, but is ${t.shape.length} instead`);if(s==="channelsFirst"&&(r=Gt(r,[0,2,1])),o==="causal")throw new Tt("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");let a=Pu(r,t,n,o==="same"?"same":"valid","NWC",i);return e!=null&&(a=dn(a,e)),a})}function OD(r,t,e,n=[1,1],o="valid",s,i,a=null){return B(()=>{if(s==null&&(s=mn()),Le(s),r.rank!==3&&r.rank!==4)throw new z(`conv2dWithBiasActivation expects input to be of rank 3 or 4, but received ${r.rank}.`);if(t.rank!==3&&t.rank!==4)throw new z(`conv2dWithBiasActivation expects kernel to be of rank 3 or 4, but received ${r.rank}.`);let u=Ud(r,s);if(o==="causal")throw new Tt("The support for CAUSAL padding mode in conv1dWithBias is not implemented yet.");return u=wo.conv2d({x:u,filter:t,strides:n,pad:o==="same"?"same":"valid",dilations:i,dataFormat:"NHWC",bias:e,activation:a}),s==="channelsFirst"&&(u=Gt(u,[0,3,1,2])),u})}function S8(r,t,e,n=[1,1,1],o="valid",s,i){return B(()=>{if(s==null&&(s=mn()),Le(s),r.rank!==4&&r.rank!==5)throw new z(`conv3dWithBias expects input to be of rank 4 or 5, but received ${r.rank}.`);if(t.rank!==4&&t.rank!==5)throw new z(`conv3dWithBias expects kernel to be of rank 4 or 5, but received ${r.rank}.`);let a=XC(r,s);if(o==="causal")throw new Tt("The support for CAUSAL padding mode in conv3dWithBias is not implemented yet.");return a=Gf(a,t,n,o==="same"?"same":"valid","NDHWC",i),e!=null&&(a=dn(a,e)),s==="channelsFirst"&&(a=Gt(a,[0,4,1,2,3])),a})}var Sm=class extends Ut{constructor(t,e){super(e);if(this.bias=null,this.DEFAULT_KERNEL_INITIALIZER="glorotNormal",this.DEFAULT_BIAS_INITIALIZER="zeros",Sm.verifyArgs(e),this.rank=t,er(this.rank,"rank"),this.rank!==1&&this.rank!==2&&this.rank!==3)throw new Tt(`Convolution layer for rank other than 1, 2, or 3 (${this.rank}) is not implemented yet.`);if(this.kernelSize=Yl(e.kernelSize,t,"kernelSize"),this.strides=Yl(e.strides==null?1:e.strides,t,"strides"),this.padding=e.padding==null?"valid":e.padding,fn(this.padding),this.dataFormat=e.dataFormat==null?"channelsLast":e.dataFormat,Le(this.dataFormat),this.activation=si(e.activation),this.useBias=e.useBias==null?!0:e.useBias,this.biasInitializer=de(e.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.biasConstraint=Ue(e.biasConstraint),this.biasRegularizer=ve(e.biasRegularizer),this.activityRegularizer=ve(e.activityRegularizer),this.dilationRate=Yl(e.dilationRate==null?1:e.dilationRate,t,"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(t){if(Zn("kernelSize"in t,"required key 'kernelSize' not in config"),typeof t.kernelSize!="number"&&!Wx(t.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(t.kernelSize)}.`)}getConfig(){let t={kernelSize:this.kernelSize,strides:this.strides,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,activation:oi(this.activation),useBias:this.useBias,biasInitializer:Te(this.biasInitializer),biasRegularizer:pe(this.biasRegularizer),activityRegularizer:pe(this.activityRegularizer),biasConstraint:We(this.biasConstraint)},e=super.getConfig();return Object.assign(t,e),t}},wc=class extends Sm{constructor(t,e){super(t,e);this.kernel=null,wc.verifyArgs(e),this.filters=e.filters,er(this.filters,"filters"),this.kernelInitializer=de(e.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.kernelConstraint=Ue(e.kernelConstraint),this.kernelRegularizer=ve(e.kernelRegularizer)}build(t){t=Jt(t);let e=this.dataFormat==="channelsFirst"?1:t.length-1;if(t[e]==null)throw new z(`The channel dimension of the input should be defined. Found ${t[e]}`);let n=t[e],o=this.kernelSize.concat([n,this.filters]);this.kernel=this.addWeight("kernel",o,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:{[e]:n}}],this.built=!0}call(t,e){return B(()=>{t=Pt(t);let n,o=this.bias==null?null:this.bias.read(),s=Ux(this.activation.getClassName());if(s!=null&&this.rank===2)n=OD(t,this.kernel.read(),o,this.strides,this.padding,this.dataFormat,this.dilationRate,s);else{if(this.rank===1)n=I8(t,this.kernel.read(),o,this.strides[0],this.padding,this.dataFormat,this.dilationRate[0]);else if(this.rank===2)n=OD(t,this.kernel.read(),o,this.strides,this.padding,this.dataFormat,this.dilationRate);else if(this.rank===3)n=S8(t,this.kernel.read(),o,this.strides,this.padding,this.dataFormat,this.dilationRate);else throw new Tt("convolutions greater than 3D are not implemented yet.");this.activation!=null&&(n=this.activation.apply(n))}return n})}computeOutputShape(t){t=Jt(t);let e=[],n=this.dataFormat==="channelsLast"?t.slice(1,t.length-1):t.slice(2);for(let s=0;s<n.length;++s){let i=En(n[s],this.kernelSize[s],this.padding,this.strides[s],typeof this.dilationRate=="number"?this.dilationRate:this.dilationRate[s]);e.push(i)}let o=[t[0]];return this.dataFormat==="channelsLast"?(o=o.concat(e),o.push(this.filters)):(o.push(this.filters),o=o.concat(e)),o}getConfig(){let t={filters:this.filters,kernelInitializer:Te(this.kernelInitializer),kernelRegularizer:pe(this.kernelRegularizer),kernelConstraint:We(this.kernelConstraint)},e=super.getConfig();return Object.assign(t,e),t}static verifyArgs(t){if(!("filters"in t)||typeof t.filters!="number"||t.filters<1)throw new z(`Convolution layer expected config.filters to be a 'number' > 0 but got ${JSON.stringify(t.filters)}`)}},Zl=class extends wc{constructor(t){super(2,t);Zl.verifyArgs(t)}getConfig(){let t=super.getConfig();return delete t.rank,t}static verifyArgs(t){if(typeof t.kernelSize!="number"&&!Wx(t.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(t.kernelSize)}.`)}};Zl.className="Conv2D";et.registerClass(Zl);var Jl=class extends wc{constructor(t){super(3,t);Jl.verifyArgs(t)}getConfig(){let t=super.getConfig();return delete t.rank,t}static verifyArgs(t){if(typeof t.kernelSize!="number"&&!(Array.isArray(t.kernelSize)&&(t.kernelSize.length===1||t.kernelSize.length===3)))throw new z(`Conv3D expects config.kernelSize to be number or [number, number, number], but received ${JSON.stringify(t.kernelSize)}.`)}};Jl.className="Conv3D";et.registerClass(Jl);var Hd=class extends Zl{constructor(t){super(t);if(this.inputSpec=[new _e({ndim:4})],this.padding!=="same"&&this.padding!=="valid")throw new z(`Conv2DTranspose currently supports only padding modes 'same' and 'valid', but received padding mode ${this.padding}`)}build(t){if(t=Jt(t),t.length!==4)throw new z("Input should have rank 4; Received input shape: "+JSON.stringify(t));let e=this.dataFormat==="channelsFirst"?1:t.length-1;if(t[e]==null)throw new z("The channel dimension of the inputs should be defined. Found `None`.");let n=t[e],o=this.kernelSize.concat([this.filters,n]);this.kernel=this.addWeight("kernel",o,"float32",this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[new _e({ndim:4,axes:{[e]:n}})],this.built=!0}call(t,e){return B(()=>{let n=Pt(t);if(n.shape.length!==4)throw new z(`Conv2DTranspose.call() expects input tensor to be rank-4, but received a tensor of rank-${n.shape.length}`);let o=n.shape,s=o[0],i,a;this.dataFormat==="channelsFirst"?(i=2,a=3):(i=1,a=2);let u=o[i],l=o[a],c=this.kernelSize[0],p=this.kernelSize[1],m=this.strides[0],f=this.strides[1],d=ii(u,m,c,this.padding),h=ii(l,f,p,this.padding),g=[s,d,h,this.filters];this.dataFormat!=="channelsLast"&&(n=Gt(n,[0,2,3,1]));let y=Mu(n,this.kernel.read(),g,this.strides,this.padding);return this.dataFormat!=="channelsLast"&&(y=Gt(y,[0,3,1,2])),this.bias!=null&&(y=dn(y,this.bias.read(),this.dataFormat)),this.activation!=null&&(y=this.activation.apply(y)),y})}computeOutputShape(t){t=Jt(t);let e=t.slice(),n,o,s;this.dataFormat==="channelsFirst"?(n=1,o=2,s=3):(n=3,o=1,s=2);let i=this.kernelSize[0],a=this.kernelSize[1],u=this.strides[0],l=this.strides[1];return e[n]=this.filters,e[o]=ii(e[o],u,i,this.padding),e[s]=ii(e[s],l,a,this.padding),e}getConfig(){let t=super.getConfig();return delete t.dilationRate,t}};Hd.className="Conv2DTranspose";et.registerClass(Hd);var qd=class extends Jl{constructor(t){super(t);if(this.inputSpec=[new _e({ndim:5})],this.padding!=="same"&&this.padding!=="valid")throw new z(`Conv3DTranspose currently supports only padding modes 'same' and 'valid', but received padding mode ${this.padding}`)}build(t){if(t=Jt(t),t.length!==5)throw new z("Input should have rank 5; Received input shape: "+JSON.stringify(t));let e=this.dataFormat==="channelsFirst"?1:t.length-1;if(t[e]==null)throw new z("The channel dimension of the inputs should be defined. Found `None`.");let n=t[e],o=this.kernelSize.concat([this.filters,n]);this.kernel=this.addWeight("kernel",o,"float32",this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint),this.useBias&&(this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,!0,this.biasConstraint)),this.inputSpec=[new _e({ndim:5,axes:{[e]:n}})],this.built=!0}call(t,e){return B(()=>{let n=Pt(t);if(n.shape.length!==5)throw new z(`Conv3DTranspose.call() expects input tensor to be rank-4, but received a tensor of rank-${n.shape.length}`);let o=n.shape,s=o[0],i,a,u;this.dataFormat==="channelsFirst"?(u=2,i=3,a=4):(u=1,i=2,a=3);let l=o[u],c=o[i],p=o[a],m=this.kernelSize[0],f=this.kernelSize[1],d=this.kernelSize[2],h=this.strides[0],g=this.strides[1],y=this.strides[2],b=ii(l,h,m,this.padding),w=ii(c,g,f,this.padding),v=ii(p,y,d,this.padding),k=[s,b,w,v,this.filters];this.dataFormat!=="channelsLast"&&(n=Gt(n,[0,2,3,4,1]));let E=E0(n,this.kernel.read(),k,this.strides,this.padding);return this.dataFormat!=="channelsLast"&&(E=Gt(E,[0,4,1,2,3])),this.bias!==null&&(E=dn(E,this.bias.read(),this.dataFormat)),this.activation!==null&&(E=this.activation.apply(E)),E})}computeOutputShape(t){t=Jt(t);let e=t.slice(),n,o,s,i;this.dataFormat==="channelsFirst"?(n=1,o=2,s=3,i=4):(n=4,o=1,s=2,i=3);let a=this.kernelSize[0],u=this.kernelSize[1],l=this.kernelSize[2],c=this.strides[0],p=this.strides[1],m=this.strides[2];return e[n]=this.filters,e[o]=ii(e[o],c,a,this.padding),e[s]=ii(e[s],p,u,this.padding),e[i]=ii(e[i],m,l,this.padding),e}getConfig(){let t=super.getConfig();return delete t.dilationRate,t}};qd.className="Conv3DTranspose";et.registerClass(qd);var YC=class extends wc{constructor(t,e){super(t,e);if(this.DEFAULT_DEPTHWISE_INITIALIZER="glorotUniform",this.DEFAULT_POINTWISE_INITIALIZER="glorotUniform",this.depthwiseKernel=null,this.pointwiseKernel=null,e.filters==null)throw new z("The `filters` configuration field is required by SeparableConv, but is unspecified.");if(e.kernelInitializer!=null||e.kernelRegularizer!=null||e.kernelConstraint!=null)throw new z("Fields kernelInitializer, kernelRegularizer and kernelConstraint are invalid for SeparableConv2D. Use depthwiseInitializer, depthwiseRegularizer, depthwiseConstraint, pointwiseInitializer, pointwiseRegularizer and pointwiseConstraint instead.");if(e.padding!=null&&e.padding!=="same"&&e.padding!=="valid")throw new z(`SeparableConv${this.rank}D supports only padding modes: 'same' and 'valid', but received ${JSON.stringify(e.padding)}`);this.depthMultiplier=e.depthMultiplier==null?1:e.depthMultiplier,this.depthwiseInitializer=de(e.depthwiseInitializer||this.DEFAULT_DEPTHWISE_INITIALIZER),this.depthwiseRegularizer=ve(e.depthwiseRegularizer),this.depthwiseConstraint=Ue(e.depthwiseConstraint),this.pointwiseInitializer=de(e.depthwiseInitializer||this.DEFAULT_POINTWISE_INITIALIZER),this.pointwiseRegularizer=ve(e.pointwiseRegularizer),this.pointwiseConstraint=Ue(e.pointwiseConstraint)}build(t){if(t=Jt(t),t.length<this.rank+2)throw new z(`Inputs to SeparableConv${this.rank}D should have rank ${this.rank+2}, but received input shape: ${JSON.stringify(t)}`);let e=this.dataFormat==="channelsFirst"?1:t.length-1;if(t[e]==null||t[e]<0)throw new z(`The channel dimension of the inputs should be defined, but found ${JSON.stringify(t[e])}`);let n=t[e],o=this.kernelSize.concat([n,this.depthMultiplier]),s=[];for(let a=0;a<this.rank;++a)s.push(1);s.push(n*this.depthMultiplier,this.filters);let i=!0;this.depthwiseKernel=this.addWeight("depthwise_kernel",o,"float32",this.depthwiseInitializer,this.depthwiseRegularizer,i,this.depthwiseConstraint),this.pointwiseKernel=this.addWeight("pointwise_kernel",s,"float32",this.pointwiseInitializer,this.pointwiseRegularizer,i,this.pointwiseConstraint),this.useBias?this.bias=this.addWeight("bias",[this.filters],"float32",this.biasInitializer,this.biasRegularizer,i,this.biasConstraint):this.bias=null,this.inputSpec=[new _e({ndim:this.rank+2,axes:{[e]:n}})],this.built=!0}call(t,e){return B(()=>{t=Pt(t);let n;if(this.rank===1)throw new Tt("1D separable convolution is not implemented yet.");return this.rank===2&&(this.dataFormat==="channelsFirst"&&(t=Gt(t,[0,2,3,1])),n=nd(t,this.depthwiseKernel.read(),this.pointwiseKernel.read(),this.strides,this.padding,this.dilationRate,"NHWC")),this.useBias&&(n=dn(n,this.bias.read(),this.dataFormat)),this.activation!=null&&(n=this.activation.apply(n)),this.dataFormat==="channelsFirst"&&(n=Gt(n,[0,3,1,2])),n})}getConfig(){let t=super.getConfig();return delete t.rank,delete t.kernelInitializer,delete t.kernelRegularizer,delete t.kernelConstraint,t.depthwiseInitializer=Te(this.depthwiseInitializer),t.pointwiseInitializer=Te(this.pointwiseInitializer),t.depthwiseRegularizer=pe(this.depthwiseRegularizer),t.pointwiseRegularizer=pe(this.pointwiseRegularizer),t.depthwiseConstraint=We(this.depthwiseConstraint),t.pointwiseConstraint=We(this.pointwiseConstraint),t}};YC.className="SeparableConv";var Kd=class extends YC{constructor(t){super(2,t)}};Kd.className="SeparableConv2D";et.registerClass(Kd);var vc=class extends wc{constructor(t){super(1,t);vc.verifyArgs(t),this.inputSpec=[{ndim:3}]}getConfig(){let t=super.getConfig();return delete t.rank,delete t.dataFormat,t}static verifyArgs(t){if(typeof t.kernelSize!="number"&&!Wx(t.kernelSize,"number",1,1))throw new z(`Conv1D expects config.kernelSize to be number or number[] with length 1, but received ${JSON.stringify(t.kernelSize)}.`)}};vc.className="Conv1D";et.registerClass(vc);var jd=class extends Ut{constructor(t){super(t);typeof t.cropping=="number"?this.cropping=[[t.cropping,t.cropping],[t.cropping,t.cropping]]:typeof t.cropping[0]=="number"?this.cropping=[[t.cropping[0],t.cropping[0]],[t.cropping[1],t.cropping[1]]]:this.cropping=t.cropping,this.dataFormat=t.dataFormat===void 0?"channelsLast":t.dataFormat,this.inputSpec=[{ndim:4}]}computeOutputShape(t){return this.dataFormat==="channelsFirst"?[t[0],t[1],t[2]-this.cropping[0][0]-this.cropping[0][1],t[3]-this.cropping[1][0]-this.cropping[1][1]]:[t[0],t[1]-this.cropping[0][0]-this.cropping[0][1],t[2]-this.cropping[1][0]-this.cropping[1][1],t[3]]}call(t,e){return B(()=>{if(t=Pt(t),this.dataFormat==="channelsLast"){let n=vd(t,this.cropping[0][0],t.shape[1]-this.cropping[0][0]-this.cropping[0][1],2);return vd(n,this.cropping[1][0],t.shape[2]-this.cropping[1][1]-this.cropping[1][0],3)}else{let n=vd(t,this.cropping[0][0],t.shape[2]-this.cropping[0][0]-this.cropping[0][1],3);return vd(n,this.cropping[1][0],t.shape[3]-this.cropping[1][1]-this.cropping[1][0],4)}})}getConfig(){let t={cropping:this.cropping,dataFormat:this.dataFormat},e=super.getConfig();return Object.assign(t,e),t}};jd.className="Cropping2D";et.registerClass(jd);var Xd=class extends Ut{constructor(t){super(t);this.DEFAULT_SIZE=[2,2],this.inputSpec=[{ndim:4}],this.size=t.size==null?this.DEFAULT_SIZE:t.size,this.dataFormat=t.dataFormat==null?"channelsLast":t.dataFormat,Le(this.dataFormat),this.interpolation=t.interpolation==null?"nearest":t.interpolation,H2(this.interpolation)}computeOutputShape(t){if(this.dataFormat==="channelsFirst"){let e=t[2]==null?null:this.size[0]*t[2],n=t[3]==null?null:this.size[1]*t[3];return[t[0],t[1],e,n]}else{let e=t[1]==null?null:this.size[0]*t[1],n=t[2]==null?null:this.size[1]*t[2];return[t[0],e,n,t[3]]}}call(t,e){return B(()=>{let n=Pt(t),o=n.shape;if(this.dataFormat==="channelsFirst"){n=Gt(n,[0,2,3,1]);let s=this.size[0]*o[2],i=this.size[1]*o[3],a=this.interpolation==="nearest"?pn.resizeNearestNeighbor(n,[s,i]):pn.resizeBilinear(n,[s,i]);return Gt(a,[0,3,1,2])}else{let s=this.size[0]*o[1],i=this.size[1]*o[2];return this.interpolation==="nearest"?pn.resizeNearestNeighbor(n,[s,i]):pn.resizeBilinear(n,[s,i])}})}getConfig(){let t={size:this.size,dataFormat:this.dataFormat},e=super.getConfig();return Object.assign(t,e),t}};Xd.className="UpSampling2D";et.registerClass(Xd);function k8(r,t,e=[1,1],n="valid",o,s){return B(()=>{o==null&&(o=mn()),Le(o);let i=Ud(r,o);if(r.rank!==4)throw new z(`Input for depthwiseConv2d is required to be 4-D, but is instead ${r.rank}-D`);if(t.rank!==4)throw new z(`depthwiseKernel is required to be 4-D, but is instead ${t.rank}-D`);return i=Ks(i,t,e,n==="same"?"same":"valid","NHWC",s),o==="channelsFirst"&&(i=Gt(i,[0,3,1,2])),i})}var Yd=class extends Sm{constructor(t){super(2,t);this.depthwiseKernel=null,this.depthMultiplier=t.depthMultiplier==null?1:t.depthMultiplier,this.depthwiseInitializer=de(t.depthwiseInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.depthwiseConstraint=Ue(t.depthwiseConstraint),this.depthwiseRegularizer=ve(t.depthwiseRegularizer)}build(t){if(t=Jt(t),t.length<4)throw new z(`Inputs to DepthwiseConv2D should have rank 4. Received input shape: ${JSON.stringify(t)}.`);let e=this.dataFormat==="channelsFirst"?1:3;if(t[e]==null||t[e]<0)throw new z(`The channel dimension of the inputs to DepthwiseConv2D should be defined, but is not (${t[e]}).`);let n=t[e],o=[this.kernelSize[0],this.kernelSize[1],n,this.depthMultiplier];this.depthwiseKernel=this.addWeight("depthwise_kernel",o,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(t,e){return B(()=>{t=Pt(t);let n=k8(t,this.depthwiseKernel.read(),this.strides,this.padding,this.dataFormat,null);return this.useBias&&(n=dn(n,this.bias.read(),this.dataFormat)),this.activation!=null&&(n=this.activation.apply(n)),n})}computeOutputShape(t){t=Jt(t);let e=this.dataFormat==="channelsFirst"?t[2]:t[1],n=this.dataFormat==="channelsFirst"?t[3]:t[2],o=this.dataFormat==="channelsFirst"?t[1]*this.depthMultiplier:t[3]*this.depthMultiplier,s=En(e,this.kernelSize[0],this.padding,this.strides[0]),i=En(n,this.kernelSize[1],this.padding,this.strides[1]);return this.dataFormat==="channelsFirst"?[t[0],o,s,i]:[t[0],s,i,o]}getConfig(){let t=super.getConfig();return t.depthMultiplier=this.depthMultiplier,t.depthwiseInitializer=Te(this.depthwiseInitializer),t.depthwiseRegularizer=pe(this.depthwiseRegularizer),t.depthwiseConstraint=We(this.depthwiseRegularizer),t}};Yd.className="DepthwiseConv2D";et.registerClass(Yd);function ZC(r,t,e,n){if(Array.isArray(r)){if(t!=null||e!=null)throw new z("When inputs is an array, neither initialState or constants should be provided");n!=null&&(e=r.slice(r.length-n,r.length),r=r.slice(0,r.length-n)),r.length>1&&(t=r.slice(1,r.length)),r=r[0]}function o(s){return s==null||Array.isArray(s)?s:[s]}return t=o(t),e=o(e),{inputs:r,initialState:t,constants:e}}function JC(r,t,e,n=!1,o,s,i=!1,a=!1){return B(()=>{let u=t.shape.length;if(u<3)throw new z(`Input should be at least 3D, but is ${u}D.`);let l=[1,0].concat(Yr(2,u));if(t=Gt(t,l),s!=null)throw new Tt("The rnn() functoin of the deeplearn.js backend does not support constants yet.");i&&console.warn("Backend rnn(): the unroll = true option is not applicable to the imperative deeplearn.js backend."),o!=null&&(o=Z(Z(o,"bool"),"float32"),o.rank===u-1&&(o=fr(o,-1)),o=Gt(o,l)),n&&(t=sr(t,0),o!=null&&(o=sr(o,0)));let c=[],p,m=e,f=t.shape[0],d=vr(t),h;o!=null&&(h=vr(o));for(let y=0;y<f;++y){let b=d[y],w=B(()=>r(b,m));if(o==null)p=w[0],m=w[1];else{let v=B(()=>{let k=h[y],E=lt(dr(k),k),D=J(P(w[0],k),P(m[0],E)),$=m.map((R,M)=>J(P(w[1][M],k),P(R,E)));return{output:D,newStates:$}});p=v.output,m=v.newStates}a&&c.push(p)}let g;return a&&(g=Je(c,1)),[p,g,m]})}var zn=class extends Ut{constructor(t){super(t);let e;if(t.cell==null)throw new z("cell property is missing for the constructor of RNN.");if(Array.isArray(t.cell)?e=new Tm({cells:t.cell}):e=t.cell,e.stateSize==null)throw new z("The RNN cell should have an attribute `stateSize` (tuple of integers, one integer per RNN state).");this.cell=e,this.returnSequences=t.returnSequences==null?!1:t.returnSequences,this.returnState=t.returnState==null?!1:t.returnState,this.goBackwards=t.goBackwards==null?!1:t.goBackwards,this._stateful=t.stateful==null?!1:t.stateful,this.unroll=t.unroll==null?!1:t.unroll,this.supportsMasking=!0,this.inputSpec=[new _e({ndim:3})],this.stateSpec=null,this.states_=null,this.numConstants=null,this.keptStates=[]}getStates(){if(this.states_==null){let t=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1;return Yr(0,t).map(e=>null)}else return this.states_}setStates(t){this.states_=t}computeOutputShape(t){Qx(t)&&(t=t[0]),t=t;let e=this.cell.stateSize;Array.isArray(e)||(e=[e]);let n=e[0],o;if(this.returnSequences?o=[t[0],t[1],n]:o=[t[0],n],this.returnState){let s=[];for(let i of e)s.push([t[0],i]);return[o].concat(s)}else return o}computeMask(t,e){return B(()=>{Array.isArray(e)&&(e=e[0]);let n=this.returnSequences?e:null;if(this.returnState){let o=this.states.map(s=>null);return[n].concat(o)}else return n})}get states(){if(this.states_==null){let t=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1,e=[];for(let n=0;n<t;++n)e.push(null);return e}else return this.states_}set states(t){this.states_=t}build(t){if(this.numConstants!=null)throw new Tt("Constants support is not implemented in RNN yet.");Qx(t)&&(t=t[0]),t=t;let n=this.stateful?t[0]:null,o=t.slice(2);this.inputSpec[0]=new _e({shape:[n,null,...o]});let s=[t[0]].concat(t.slice(2));this.cell.build(s);let i;if(Array.isArray(this.cell.stateSize)?i=this.cell.stateSize:i=[this.cell.stateSize],this.stateSpec!=null){if(!x.arraysEqual(this.stateSpec.map(a=>a.shape[a.shape.length-1]),i))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=i.map(a=>new _e({shape:[null,a]}));this.stateful&&this.resetStates()}resetStates(t,e=!1){B(()=>{if(!this.stateful)throw new Ln("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(o=>ye([n,o])):this.states_=[ye([n,this.cell.stateSize])];else if(t==null)_t(this.states_),this.keptStates!=null&&(_t(this.keptStates),this.keptStates=[]),Array.isArray(this.cell.stateSize)?this.states_=this.cell.stateSize.map(o=>ye([n,o])):this.states_[0]=ye([n,this.cell.stateSize]);else{if(Array.isArray(t)||(t=[t]),t.length!==this.states_.length)throw new z(`Layer ${this.name} expects ${this.states_.length} state(s), but it received ${t.length} state value(s). Input received: ${t}`);e===!0?this.keptStates.push(this.states_.slice()):_t(this.states_);for(let o=0;o<this.states_.length;++o){let s=t[o],i=Array.isArray(this.cell.stateSize)?this.cell.stateSize[o]:this.cell.stateSize,a=[n,i];if(!x.arraysEqual(s.shape,a))throw new z(`State ${o} is incompatible with layer ${this.name}: expected shape=${a}, received shape=${s.shape}`);this.states_[o]=s}}this.states_=this.states_.map(o=>Me(o.clone()))})}apply(t,e){let n=e==null?null:e.initialState,o=e==null?null:e.constants;e==null&&(e={});let s=ZC(t,n,o,this.numConstants);t=s.inputs,n=s.initialState,o=s.constants;let i=[],a=[];if(n!=null){e.initialState=n,i=i.concat(n),this.stateSpec=[];for(let l of n)this.stateSpec.push(new _e({shape:l.shape}));a=a.concat(this.stateSpec)}if(o!=null&&(e.constants=o,i=i.concat(o),this.numConstants=o.length),i[0]instanceof gn){let l=[t].concat(i),c=this.inputSpec.concat(a),p=this.inputSpec;this.inputSpec=c;let m=super.apply(l,e);return this.inputSpec=p,m}else return super.apply(t,e)}call(t,e){return B(()=>{let n=e==null?null:e.mask,o=e==null?null:e.training,s=e==null?null:e.initialState;t=Pt(t),s==null&&(this.stateful?s=this.states_:s=this.getInitialState(t));let i=Array.isArray(this.cell.stateSize)?this.cell.stateSize.length:1;if(s.length!==i)throw new z(`RNN Layer has ${i} state(s) but was passed ${s.length} initial state(s).`);this.unroll&&console.warn("Ignoring unroll = true for RNN layer, due to imperative backend.");let a={training:o},l=JC((d,h)=>{let g=this.cell.call([d].concat(h),a);return[g[0],g.slice(1)]},t,s,this.goBackwards,n,null,this.unroll,this.returnSequences),c=l[0],p=l[1],m=l[2];this.stateful&&this.resetStates(m,o);let f=this.returnSequences?p:c;return this.returnState?[f].concat(m):f})}getInitialState(t){return B(()=>{let e=ye(t.shape);return e=mt(e,[1,2]),e=ol(e),Array.isArray(this.cell.stateSize)?this.cell.stateSize.map(n=>n>1?jx(e,[1,n]):e):this.cell.stateSize>1?[jx(e,[1,this.cell.stateSize])]:[e]})}get trainableWeights(){return this.trainable?this.cell.trainableWeights:[]}get nonTrainableWeights(){return this.trainable?this.cell.nonTrainableWeights:this.cell.weights}setFastWeightInitDuringBuild(t){super.setFastWeightInitDuringBuild(t),this.cell!=null&&this.cell.setFastWeightInitDuringBuild(t)}getConfig(){let t=super.getConfig(),e={returnSequences:this.returnSequences,returnState:this.returnState,goBackwards:this.goBackwards,stateful:this.stateful,unroll:this.unroll};this.numConstants!=null&&(e.numConstants=this.numConstants);let n=this.cell.getConfig();return this.getClassName()===zn.className&&(e.cell={className:this.cell.getClassName(),config:n}),Object.assign({},n,t,e)}static fromConfig(t,e,n={}){let o=e.cell,s=xn(o,n);return new t(Object.assign(e,{cell:s}))}};zn.className="RNN";et.registerClass(zn);var Ql=class extends Ut{},km=class extends Ql{constructor(t){super(t);this.DEFAULT_ACTIVATION="tanh",this.DEFAULT_KERNEL_INITIALIZER="glorotNormal",this.DEFAULT_RECURRENT_INITIALIZER="orthogonal",this.DEFAULT_BIAS_INITIALIZER="zeros",this.units=t.units,er(this.units,"units"),this.activation=si(t.activation==null?this.DEFAULT_ACTIVATION:t.activation),this.useBias=t.useBias==null?!0:t.useBias,this.kernelInitializer=de(t.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.recurrentInitializer=de(t.recurrentInitializer||this.DEFAULT_RECURRENT_INITIALIZER),this.biasInitializer=de(t.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.kernelRegularizer=ve(t.kernelRegularizer),this.recurrentRegularizer=ve(t.recurrentRegularizer),this.biasRegularizer=ve(t.biasRegularizer),this.kernelConstraint=Ue(t.kernelConstraint),this.recurrentConstraint=Ue(t.recurrentConstraint),this.biasConstraint=Ue(t.biasConstraint),this.dropout=mc([1,ri([0,t.dropout==null?0:t.dropout])]),this.recurrentDropout=mc([1,ri([0,t.recurrentDropout==null?0:t.recurrentDropout])]),this.dropoutFunc=t.dropoutFunc,this.stateSize=this.units,this.dropoutMask=null,this.recurrentDropoutMask=null}build(t){t=Jt(t),this.kernel=this.addWeight("kernel",[t[t.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(t,e){return B(()=>{if(t=t,t.length!==2)throw new z(`SimpleRNNCell expects 2 input Tensors, got ${t.length}.`);let n=t[1];t=t[0];let o=e.training==null?!1:e.training;0<this.dropout&&this.dropout<1&&this.dropoutMask==null&&(this.dropoutMask=il({ones:()=>dr(t),rate:this.dropout,training:o,dropoutFunc:this.dropoutFunc})),0<this.recurrentDropout&&this.recurrentDropout<1&&this.recurrentDropoutMask==null&&(this.recurrentDropoutMask=il({ones:()=>dr(n),rate:this.recurrentDropout,training:o,dropoutFunc:this.dropoutFunc}));let s,i=this.dropoutMask,a=this.recurrentDropoutMask;i!=null?s=To(P(t,i),this.kernel.read()):s=To(t,this.kernel.read()),this.bias!=null&&(s=dn(s,this.bias.read())),a!=null&&(n=P(n,a));let u=J(s,To(n,this.recurrentKernel.read()));return this.activation!=null&&(u=this.activation.apply(u)),[u,u]})}getConfig(){let t=super.getConfig(),e={units:this.units,activation:oi(this.activation),useBias:this.useBias,kernelInitializer:Te(this.kernelInitializer),recurrentInitializer:Te(this.recurrentInitializer),biasInitializer:Te(this.biasInitializer),kernelRegularizer:pe(this.kernelRegularizer),recurrentRegularizer:pe(this.recurrentRegularizer),biasRegularizer:pe(this.biasRegularizer),activityRegularizer:pe(this.activityRegularizer),kernelConstraint:We(this.kernelConstraint),recurrentConstraint:We(this.recurrentConstraint),biasConstraint:We(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout};return Object.assign({},t,e)}};km.className="SimpleRNNCell";et.registerClass(km);var Zd=class extends zn{constructor(t){t.cell=new km(t);super(t)}call(t,e){return B(()=>{this.cell.dropoutMask!=null&&(_t(this.cell.dropoutMask),this.cell.dropoutMask=null),this.cell.recurrentDropoutMask!=null&&(_t(this.cell.recurrentDropoutMask),this.cell.recurrentDropoutMask=null);let n=e==null?null:e.mask,o=e==null?null:e.training,s=e==null?null:e.initialState;return super.call(t,{mask:n,training:o,initialState:s})})}static fromConfig(t,e){return new t(e)}};Zd.className="SimpleRNN";et.registerClass(Zd);var Nm=class extends Ql{constructor(t){super(t);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",t.resetAfter)throw new z("GRUCell does not support reset_after parameter set to true.");this.units=t.units,er(this.units,"units"),this.activation=si(t.activation===void 0?this.DEFAULT_ACTIVATION:t.activation),this.recurrentActivation=si(t.recurrentActivation===void 0?this.DEFAULT_RECURRENT_ACTIVATION:t.recurrentActivation),this.useBias=t.useBias==null?!0:t.useBias,this.kernelInitializer=de(t.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.recurrentInitializer=de(t.recurrentInitializer||this.DEFAULT_RECURRENT_INITIALIZER),this.biasInitializer=de(t.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.kernelRegularizer=ve(t.kernelRegularizer),this.recurrentRegularizer=ve(t.recurrentRegularizer),this.biasRegularizer=ve(t.biasRegularizer),this.kernelConstraint=Ue(t.kernelConstraint),this.recurrentConstraint=Ue(t.recurrentConstraint),this.biasConstraint=Ue(t.biasConstraint),this.dropout=mc([1,ri([0,t.dropout==null?0:t.dropout])]),this.recurrentDropout=mc([1,ri([0,t.recurrentDropout==null?0:t.recurrentDropout])]),this.dropoutFunc=t.dropoutFunc,this.implementation=t.implementation,this.stateSize=this.units,this.dropoutMask=null,this.recurrentDropoutMask=null}build(t){t=Jt(t);let e=t[t.length-1];this.kernel=this.addWeight("kernel",[e,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(t,e){return B(()=>{if(t=t,t.length!==2)throw new z(`GRUCell expects 2 input Tensors (inputs, h, c), got ${t.length}.`);let n=e.training==null?!1:e.training,o=t[1];t=t[0],0<this.dropout&&this.dropout<1&&this.dropoutMask==null&&(this.dropoutMask=il({ones:()=>dr(t),rate:this.dropout,training:n,count:3,dropoutFunc:this.dropoutFunc})),0<this.recurrentDropout&&this.recurrentDropout<1&&this.recurrentDropoutMask==null&&(this.recurrentDropoutMask=il({ones:()=>dr(o),rate:this.recurrentDropout,training:n,count:3,dropoutFunc:this.dropoutFunc}));let s=this.dropoutMask,i=this.recurrentDropoutMask,a,u,l;0<this.dropout&&this.dropout<1&&(t=P(t,s[0]));let c=To(t,this.kernel.read());this.useBias&&(c=dn(c,this.bias.read())),0<this.recurrentDropout&&this.recurrentDropout<1&&(o=P(o,i[0]));let p=this.recurrentKernel.read(),[m,f]=ur(p,[2*this.units,this.units],p.rank-1),d=To(o,m),[h,g,y]=ur(c,3,c.rank-1),[b,w]=ur(d,2,d.rank-1);a=this.recurrentActivation.apply(J(h,b)),u=this.recurrentActivation.apply(J(g,w));let v=To(P(u,o),f);l=this.activation.apply(J(y,v));let k=J(P(a,o),P(J(1,Zt(a)),l));return[k,k]})}getConfig(){let t=super.getConfig(),e={units:this.units,activation:oi(this.activation),recurrentActivation:oi(this.recurrentActivation),useBias:this.useBias,kernelInitializer:Te(this.kernelInitializer),recurrentInitializer:Te(this.recurrentInitializer),biasInitializer:Te(this.biasInitializer),kernelRegularizer:pe(this.kernelRegularizer),recurrentRegularizer:pe(this.recurrentRegularizer),biasRegularizer:pe(this.biasRegularizer),activityRegularizer:pe(this.activityRegularizer),kernelConstraint:We(this.kernelConstraint),recurrentConstraint:We(this.recurrentConstraint),biasConstraint:We(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation,resetAfter:!1};return Object.assign({},t,e)}};Nm.className="GRUCell";et.registerClass(Nm);var Jd=class extends zn{constructor(t){t.implementation===0&&console.warn("`implementation=0` has been deprecated, and now defaults to `implementation=1`. Please update your layer call."),t.cell=new Nm(t);super(t)}call(t,e){return B(()=>{this.cell.dropoutMask!=null&&(_t(this.cell.dropoutMask),this.cell.dropoutMask=null),this.cell.recurrentDropoutMask!=null&&(_t(this.cell.recurrentDropoutMask),this.cell.recurrentDropoutMask=null);let n=e==null?null:e.mask,o=e==null?null:e.training,s=e==null?null:e.initialState;return super.call(t,{mask:n,training:o,initialState:s})})}static fromConfig(t,e){return e.implmentation===0&&(e.implementation=1),new t(e)}};Jd.className="GRU";et.registerClass(Jd);var tu=class extends Ql{constructor(t){super(t);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=t.units,er(this.units,"units"),this.activation=si(t.activation===void 0?this.DEFAULT_ACTIVATION:t.activation),this.recurrentActivation=si(t.recurrentActivation===void 0?this.DEFAULT_RECURRENT_ACTIVATION:t.recurrentActivation),this.useBias=t.useBias==null?!0:t.useBias,this.kernelInitializer=de(t.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.recurrentInitializer=de(t.recurrentInitializer||this.DEFAULT_RECURRENT_INITIALIZER),this.biasInitializer=de(t.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.unitForgetBias=t.unitForgetBias,this.kernelRegularizer=ve(t.kernelRegularizer),this.recurrentRegularizer=ve(t.recurrentRegularizer),this.biasRegularizer=ve(t.biasRegularizer),this.kernelConstraint=Ue(t.kernelConstraint),this.recurrentConstraint=Ue(t.recurrentConstraint),this.biasConstraint=Ue(t.biasConstraint),this.dropout=mc([1,ri([0,t.dropout==null?0:t.dropout])]),this.recurrentDropout=mc([1,ri([0,t.recurrentDropout==null?0:t.recurrentDropout])]),this.dropoutFunc=t.dropoutFunc,this.implementation=t.implementation,this.stateSize=[this.units,this.units],this.dropoutMask=null,this.recurrentDropoutMask=null}build(t){var e;t=Jt(t);let n=t[t.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 o;if(this.useBias){if(this.unitForgetBias){let s=this.biasInitializer,i=this.units;o=new(e=class extends Tn{apply(u,l){let c=s.apply([i]),p=new hc().apply([i]),m=s.apply([i*2]);return bC(bC(c,p),m)}},e.className="CustomInit",e)}else o=this.biasInitializer;this.bias=this.addWeight("bias",[this.units*4],null,o,this.biasRegularizer,!0,this.biasConstraint)}else this.bias=null;this.built=!0}call(t,e){return B(()=>{let n=e.training==null?!1:e.training;if(t=t,t.length!==3)throw new z(`LSTMCell expects 3 input Tensors (inputs, h, c), got ${t.length}.`);let o=t[1],s=t[2];t=t[0],0<this.dropout&&this.dropout<1&&this.dropoutMask==null&&(this.dropoutMask=il({ones:()=>dr(t),rate:this.dropout,training:n,count:4,dropoutFunc:this.dropoutFunc})),0<this.recurrentDropout&&this.recurrentDropout<1&&this.recurrentDropoutMask==null&&(this.recurrentDropoutMask=il({ones:()=>dr(o),rate:this.recurrentDropout,training:n,count:4,dropoutFunc:this.dropoutFunc}));let i=this.dropoutMask,a=this.recurrentDropoutMask,u,l,c,p;0<this.dropout&&this.dropout<1&&(t=P(t,i[0]));let m=To(t,this.kernel.read());0<this.recurrentDropout&&this.recurrentDropout<1&&(o=P(o,a[0])),m=J(m,To(o,this.recurrentKernel.read())),this.useBias&&(m=dn(m,this.bias.read()));let[f,d,h,g]=ur(m,4,m.rank-1);u=this.recurrentActivation.apply(f),l=this.recurrentActivation.apply(d),c=J(P(l,s),P(u,this.activation.apply(h))),p=this.recurrentActivation.apply(g);let y=P(p,this.activation.apply(c));return[y,y,c]})}getConfig(){let t=super.getConfig(),e={units:this.units,activation:oi(this.activation),recurrentActivation:oi(this.recurrentActivation),useBias:this.useBias,kernelInitializer:Te(this.kernelInitializer),recurrentInitializer:Te(this.recurrentInitializer),biasInitializer:Te(this.biasInitializer),unitForgetBias:this.unitForgetBias,kernelRegularizer:pe(this.kernelRegularizer),recurrentRegularizer:pe(this.recurrentRegularizer),biasRegularizer:pe(this.biasRegularizer),activityRegularizer:pe(this.activityRegularizer),kernelConstraint:We(this.kernelConstraint),recurrentConstraint:We(this.recurrentConstraint),biasConstraint:We(this.biasConstraint),dropout:this.dropout,recurrentDropout:this.recurrentDropout,implementation:this.implementation};return Object.assign({},t,e)}};tu.className="LSTMCell";et.registerClass(tu);var Qd=class extends zn{constructor(t){t.implementation===0&&console.warn("`implementation=0` has been deprecated, and now defaults to `implementation=1`. Please update your layer call."),t.cell=new tu(t);super(t)}call(t,e){return B(()=>{this.cell.dropoutMask!=null&&(_t(this.cell.dropoutMask),this.cell.dropoutMask=null),this.cell.recurrentDropoutMask!=null&&(_t(this.cell.recurrentDropoutMask),this.cell.recurrentDropoutMask=null);let n=e==null?null:e.mask,o=e==null?null:e.training,s=e==null?null:e.initialState;return super.call(t,{mask:n,training:o,initialState:s})})}static fromConfig(t,e){return e.implmentation===0&&(e.implementation=1),new t(e)}};Qd.className="LSTM";et.registerClass(Qd);var Tm=class extends Ql{constructor(t){super(t);this.cells=t.cells}get stateSize(){let t=[];for(let e of this.cells.slice().reverse())Array.isArray(e.stateSize)?t.push(...e.stateSize):t.push(e.stateSize);return t}call(t,e){return B(()=>{t=t;let n=t.slice(1),o=[];for(let a of this.cells.slice().reverse())Array.isArray(a.stateSize)?o.push(n.splice(0,a.stateSize.length)):o.push(n.splice(0,1));o.reverse();let s=[],i;for(let a=0;a<this.cells.length;++a){let u=this.cells[a];n=o[a],a===0?i=[t[0]].concat(n):i=[i[0]].concat(n),i=u.call(i,e),s.push(i.slice(1))}n=[];for(let a of s.slice().reverse())n.push(...a);return[i[0]].concat(n)})}build(t){Qx(t)&&(t=t[0]),t=t;let e;this.cells.forEach((n,o)=>{ei(`RNNCell_${o}`,()=>{n.build(t),Array.isArray(n.stateSize)?e=n.stateSize[0]:e=n.stateSize,t=[t[0],e]})}),this.built=!0}getConfig(){let t=super.getConfig(),e=s=>({className:s.getClassName(),config:s.getConfig()}),o={cells:this.cells.map(e)};return Object.assign({},t,o)}static fromConfig(t,e,n={}){let o=[];for(let s of e.cells)o.push(xn(s,n));return new t({cells:o})}get trainableWeights(){if(!this.trainable)return[];let t=[];for(let e of this.cells)t.push(...e.trainableWeights);return t}get nonTrainableWeights(){let t=[];for(let e of this.cells)t.push(...e.nonTrainableWeights);if(!this.trainable){let e=[];for(let n of this.cells)e.push(...n.trainableWeights);return e.concat(t)}return t}getWeights(){let t=[];for(let e of this.cells)t.push(...e.weights);return Ed(t)}setWeights(t){let e=[];for(let n of this.cells){let o=n.weights.length,s=t.splice(o);for(let i=0;i<n.weights.length;++i)e.push([n.weights[i],s[i]])}ym(e)}};Tm.className="StackedRNNCells";et.registerClass(Tm);function il(r){let{ones:t,rate:e,training:n=!1,count:o=1,dropoutFunc:s}=r,i=()=>s!=null?s(t(),e):Yx(t(),e),a=()=>Hl(i,t,n);return!o||o<=1?Me(a().clone()):Array(o).fill(void 0).map(a).map(l=>Me(l.clone()))}var N8=function(r,t){var e={};for(var n in r)Object.prototype.hasOwnProperty.call(r,n)&&t.indexOf(n)<0&&(e[n]=r[n]);if(r!=null&&typeof Object.getOwnPropertySymbols=="function")for(var o=0,n=Object.getOwnPropertySymbols(r);o<n.length;o++)t.indexOf(n[o])<0&&Object.prototype.propertyIsEnumerable.call(r,n[o])&&(e[n[o]]=r[n[o]]);return e};var QC=class extends zn{constructor(t){if(t.unroll)throw new Tt("Unrolling is not possible with convolutional RNNs.");if(Array.isArray(t.cell))throw new Tt("It is not possible at the moment to stack convolutional cells.");super(t);this.inputSpec=[new _e({ndim:5})]}call(t,e){return B(()=>{if(this.cell.dropoutMask!=null&&(_t(this.cell.dropoutMask),this.cell.dropoutMask=null),this.cell.recurrentDropoutMask!=null&&(_t(this.cell.recurrentDropoutMask),this.cell.recurrentDropoutMask=null),e&&e.constants)throw new z("ConvRNN2D cell does not support constants");let n=e==null?null:e.mask,o=e==null?null:e.training,s=e==null?null:e.initialState;return super.call(t,{mask:n,training:o,initialState:s})})}computeOutputShape(t){let e=this.computeSingleOutputShape(t);return this.returnSequences||(e=[e[0],...e.slice(2)]),this.returnState&&(e=[e,...Array(2).fill([t[0],...e.slice(-3)])]),e}getInitialState(t){return B(()=>{let{stateSize:e}=this.cell,n=t.shape,o=this.computeSingleOutputShape(n),s=[o[0],...o.slice(2)],i=ye(s);return Array.isArray(e)?Array(e.length).fill(i):[i]})}resetStates(t,e=!1){B(()=>{if(!this.stateful)throw new Ln("Cannot call resetStates() on an RNN Layer that is not stateful.");let n=this.inputSpec[0].shape,o=this.computeSingleOutputShape(n),s=[o[0],...o.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(()=>ye(s)):this.states_=[ye(s)];else if(t==null)_t(this.states_),this.keptStates!=null&&(_t(this.keptStates),this.keptStates=[]),Array.isArray(this.cell.stateSize)?this.states_=this.cell.stateSize.map(()=>ye(s)):this.states_[0]=ye(s);else{if(Array.isArray(t)||(t=[t]),t.length!==this.states_.length)throw new z(`Layer ${this.name} expects ${this.states_.length} state(s), but it received ${t.length} state value(s). Input received: ${t}`);e?this.keptStates.push(this.states_.slice()):_t(this.states_);for(let a=0;a<this.states_.length;++a){let u=t[a],l=s;if(!x.arraysEqual(u.shape,l))throw new z(`State ${a} is incompatible with layer ${this.name}: expected shape=${l}, received shape=${u.shape}`);this.states_[a]=u}}this.states_=this.states_.map(a=>Me(a.clone()))})}computeSingleOutputShape(t){let{dataFormat:e,filters:n,kernelSize:o,padding:s,strides:i,dilationRate:a}=this.cell,u=e==="channelsFirst",l=t[u?3:2],c=t[u?4:3],p=En(l,o[0],s,i[0],a[0]),m=En(c,o[1],s,i[1],a[1]);return[...t.slice(0,2),...u?[n,p,m]:[p,m,n]]}};QC.className="ConvRNN2D";var _m=class extends tu{constructor(t){let{filters:e,kernelSize:n,strides:o,padding:s,dataFormat:i,dilationRate:a}=t;super(Object.assign({},t,{units:e}));this.filters=e,er(this.filters,"filters"),this.kernelSize=Yl(n,2,"kernelSize"),this.kernelSize.forEach(u=>er(u,"kernelSize")),this.strides=Yl(o||1,2,"strides"),this.strides.forEach(u=>er(u,"strides")),this.padding=s||"valid",fn(this.padding),this.dataFormat=i||"channelsLast",Le(this.dataFormat),this.dilationRate=Yl(a||1,2,"dilationRate"),this.dilationRate.forEach(u=>er(u,"dilationRate"))}build(t){var e;t=Jt(t);let n=this.dataFormat==="channelsFirst"?1:t.length-1;if(t[n]==null)throw new z(`The channel dimension of the input should be defined. Found ${t[n]}`);let o=t[n],s=4,i=this.kernelSize.concat([o,this.filters*s]);this.kernel=this.addWeight("kernel",i,null,this.kernelInitializer,this.kernelRegularizer,!0,this.kernelConstraint);let a=this.kernelSize.concat([this.filters,this.filters*s]);if(this.recurrentKernel=this.addWeight("recurrent_kernel",a,null,this.recurrentInitializer,this.recurrentRegularizer,!0,this.recurrentConstraint),this.useBias){let u;if(this.unitForgetBias){let l=this.biasInitializer,c=this.filters;u=new(e=class extends Tn{apply(m,f){let d=l.apply([c]),h=lr([c]),g=l.apply([c*2]);return lm([d,h,g])}},e.className="CustomInit",e)}else u=this.biasInitializer;this.bias=this.addWeight("bias",[this.filters*s],null,u,this.biasRegularizer,!0,this.biasConstraint)}this.built=!0}call(t,e){return B(()=>{if(t.length!==3)throw new z(`ConvLSTM2DCell expects 3 input Tensors (inputs, h, c), got ${t.length}.`);let n=e.training||!1,o=t[0],s=t[1],i=t[2],a=4;0<this.dropout&&this.dropout<1&&this.dropoutMask==null&&(this.dropoutMask=il({ones:()=>dr(o),rate:this.dropout,training:n,count:a,dropoutFunc:this.dropoutFunc}));let u=this.dropoutMask,l=(st,it,ft)=>!it||!it[ft]?st:P(it[ft],st),c=l(o,u,0),p=l(o,u,1),m=l(o,u,2),f=l(o,u,3);0<this.recurrentDropout&&this.recurrentDropout<1&&this.recurrentDropoutMask==null&&(this.recurrentDropoutMask=il({ones:()=>dr(s),rate:this.recurrentDropout,training:n,count:a,dropoutFunc:this.dropoutFunc}));let d=this.recurrentDropoutMask,h=l(s,d,0),g=l(s,d,1),y=l(s,d,2),b=l(s,d,3),w=3,[v,k,E,D]=ur(this.kernel.read(),a,w),[$,R,M,W]=this.useBias?ur(this.bias.read(),a):[null,null,null,null];c=this.inputConv(c,v,$,this.padding),p=this.inputConv(p,k,R,this.padding),m=this.inputConv(m,E,M,this.padding),f=this.inputConv(f,D,W,this.padding);let[U,q,K,j]=ur(this.recurrentKernel.read(),a,w);h=this.recurrentConv(h,U),g=this.recurrentConv(g,q),y=this.recurrentConv(y,K),b=this.recurrentConv(b,j);let Q=this.recurrentActivation.apply(J(c,h)),rt=this.recurrentActivation.apply(J(p,g)),X=J(P(rt,i),P(Q,this.activation.apply(J(m,y)))),ot=P(this.recurrentActivation.apply(J(f,b)),this.activation.apply(X));return[ot,ot,X]})}getConfig(){let t=super.getConfig(),{units:e}=t,n=N8(t,["units"]),o={filters:this.filters,kernelSize:this.kernelSize,padding:this.padding,dataFormat:this.dataFormat,dilationRate:this.dilationRate,strides:this.strides};return Object.assign({},n,o)}inputConv(t,e,n,o){let s=ln(t,e,this.strides,o||"valid",this.dataFormat==="channelsFirst"?"NCHW":"NHWC",this.dilationRate);return n?dn(s,n,this.dataFormat):s}recurrentConv(t,e){return ln(t,e,1,"same",this.dataFormat==="channelsFirst"?"NCHW":"NHWC")}};_m.className="ConvLSTM2DCell";et.registerClass(_m);var th=class extends QC{constructor(t){let e=new _m(t);super(Object.assign({},t,{cell:e}))}static fromConfig(t,e){return new t(e)}};th.className="ConvLSTM2D";et.registerClass(th);var Em=class extends Ut{constructor(t){super(t);this.rate=Math.max(Math.min(t.rate,1),0),this.noiseShape=t.noiseShape,this.seed=t.seed,this.supportsMasking=!0}getNoiseShape(t){if(this.noiseShape==null)return this.noiseShape;let e=t.shape,n=[];for(let o=0;o<this.noiseShape.length;++o)n.push(this.noiseShape[o]==null?e[o]:this.noiseShape[o]);return n}call(t,e){return B(()=>{this.invokeCallHook(t,e);let n=Pt(t);if(0<this.rate&&this.rate<1){let o=e.training==null?!1:e.training,s=this.getNoiseShape(n);return Hl(()=>Yx(n,this.rate,s,this.seed),()=>n,o)}return t})}getConfig(){let t={rate:this.rate,noiseShape:this.noiseShape,seed:this.seed},e=super.getConfig();return Object.assign(t,e),t}dispose(){return super.dispose()}};Em.className="Dropout";et.registerClass(Em);var eh=class extends Em{constructor(t){super(t);this.inputSpec=[{ndim:3}]}getNoiseShape(t){let e=t.shape;return[e[0],1,e[2]]}};eh.className="SpatialDropout1D";et.registerClass(eh);var rh=class extends Ut{constructor(t){super(t);if(this.activation=null,this.useBias=!0,this.kernel=null,this.bias=null,this.DEFAULT_KERNEL_INITIALIZER="glorotNormal",this.DEFAULT_BIAS_INITIALIZER="zeros",t.batchInputShape==null&&t.inputShape==null&&t.inputDim!=null){let e=null;t.batchSize!=null&&(e=t.batchSize),this.batchInputShape=[e,t.inputDim]}this.units=t.units,er(this.units,"units"),this.activation=si(t.activation),t.useBias!=null&&(this.useBias=t.useBias),this.kernelInitializer=de(t.kernelInitializer||this.DEFAULT_KERNEL_INITIALIZER),this.biasInitializer=de(t.biasInitializer||this.DEFAULT_BIAS_INITIALIZER),this.kernelConstraint=Ue(t.kernelConstraint),this.biasConstraint=Ue(t.biasConstraint),this.kernelRegularizer=ve(t.kernelRegularizer),this.biasRegularizer=ve(t.biasRegularizer),this.activityRegularizer=ve(t.activityRegularizer),this.supportsMasking=!0,this.inputSpec=[{minNDim:2}]}build(t){t=Jt(t);let e=t[t.length-1];this.kernel==null&&(this.kernel=this.addWeight("kernel",[e,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]:e}}],this.built=!0}computeOutputShape(t){t=Jt(t);let e=t.slice();return e[e.length-1]=this.units,e}call(t,e){return B(()=>{this.invokeCallHook(t,e);let n=Pt(t),o=Ux(this.activation.getClassName()),s;return o!=null?s=To(n,this.kernel.read(),o,this.bias?this.bias.read():null):(s=To(n,this.kernel.read()),this.bias!=null&&(s=dn(s,this.bias.read())),this.activation!=null&&(s=this.activation.apply(s))),s})}getConfig(){let t={units:this.units,activation:oi(this.activation),useBias:this.useBias,kernelInitializer:Te(this.kernelInitializer),biasInitializer:Te(this.biasInitializer),kernelRegularizer:pe(this.kernelRegularizer),biasRegularizer:pe(this.biasRegularizer),activityRegularizer:pe(this.activityRegularizer),kernelConstraint:We(this.kernelConstraint),biasConstraint:We(this.biasConstraint)},e=super.getConfig();return Object.assign(t,e),t}};rh.className="Dense";et.registerClass(rh);var nh=class extends Ut{constructor(t){t=t||{};super(t);this.inputSpec=[{minNDim:3}],this.dataFormat=t.dataFormat}computeOutputShape(t){t=Jt(t);for(let e of t.slice(1))if(e==null)throw new z(`The shape of the input to "Flatten" is not fully defined (got ${t.slice(1)}). Make sure to pass a complete "input_shape" or "batch_input_shape" argument to the first layer in your model.`);return[t[0],No(t,1)]}call(t,e){return B(()=>{this.invokeCallHook(t,e);let n=Pt(t);if(this.dataFormat==="channelsFirst"&&n.rank>1){let o=[0];for(let s=2;s<n.rank;++s)o.push(s);o.push(1),n=Gt(n,o)}return Y2(n)})}getConfig(){let t={};this.dataFormat!=null&&(t.dataFormat=this.dataFormat);let e=super.getConfig();return Object.assign(t,e),t}};nh.className="Flatten";et.registerClass(nh);var oh=class extends Ut{constructor(t){super(t);this.supportsMasking=!0,this.activation=si(t.activation)}call(t,e){return B(()=>{this.invokeCallHook(t,e);let n=Pt(t);return this.activation.apply(n)})}getConfig(){let t={activation:oi(this.activation)},e=super.getConfig();return Object.assign(t,e),t}};oh.className="Activation";et.registerClass(oh);var sh=class extends Ut{constructor(t){super(t);this.n=t.n,this.inputSpec=[{ndim:2}]}computeOutputShape(t){return[t[0],this.n,t[1]]}call(t,e){return B(()=>(t=Pt(t),j2(t,this.n)))}getConfig(){let t={n:this.n},e=super.getConfig();return Object.assign(t,e),t}};sh.className="RepeatVector";et.registerClass(sh);var ih=class extends Ut{constructor(t){super(t);this.targetShape=t.targetShape;for(let e=0;e<this.targetShape.length;++e)this.isUnknown(this.targetShape[e])&&(this.targetShape[e]=null)}isUnknown(t){return t<0||t==null}fixUnknownDimension(t,e){let n="Total size of new array must be unchanged.",o=e.slice(),s=1,i=null;for(let u=0;u<o.length;++u){let l=o[u];if(this.isUnknown(l))if(i===null)i=u;else throw new z("Can only specifiy one unknown dimension.");else s*=l}let a=No(t);if(i!==null){if(s===0||a%s!==0)throw new z(n);o[i]=a/s}else if(a!==s)throw new z(n);return o}computeOutputShape(t){let e=!1;for(let n=0;n<t.length;++n)if(this.isUnknown(t[n])){e=!0;break}return e?t.slice(0,1).concat(this.targetShape):t.slice(0,1).concat(this.fixUnknownDimension(t.slice(1),this.targetShape))}call(t,e){return B(()=>{this.invokeCallHook(t,e);let n=Pt(t),o=n.shape,s=o.slice(0,1).concat(this.fixUnknownDimension(o.slice(1),this.targetShape));return O(n,s)})}getConfig(){let t={targetShape:this.targetShape},e=super.getConfig();return Object.assign(t,e),t}};ih.className="Reshape";et.registerClass(ih);var ah=class extends Ut{constructor(t){super(t);if(t.dims==null)throw new Error("Required configuration field `dims` is missing during Permute constructor call.");if(!Array.isArray(t.dims))throw new Error(`Permute constructor requires \`dims\` to be an Array, but received ${t.dims} instead.`);let e=Yr(1,t.dims.length+1);if(!x.arraysEqual(t.dims.slice().sort(),e))throw new Error("Invalid permutation `dims`: "+JSON.stringify(t.dims)+" `dims` must contain consecutive integers starting from 1.");this.dims=t.dims,this.dimsIncludingBatch=[0].concat(this.dims),this.inputSpec=[new _e({ndim:this.dims.length+1})]}computeOutputShape(t){t=Jt(t);let e=t.slice();return this.dims.forEach((n,o)=>{e[o+1]=t[n]}),e}call(t,e){return Gt(Pt(t),this.dimsIncludingBatch)}getConfig(){let t={dims:this.dims},e=super.getConfig();return Object.assign(t,e),t}};ah.className="Permute";et.registerClass(ah);var lh=class extends Ut{constructor(t){super(t==null?{}:t);this.supportsMasking=!0,t!=null?this.maskValue=t.maskValue==null?0:t.maskValue:this.maskValue=0}computeOutputShape(t){return t}getConfig(){let t=super.getConfig(),e={maskValue:this.maskValue};return Object.assign(e,t),e}computeMask(t,e){let n=Pt(t),o=-1;return Gl(bo(n,this.maskValue),o)}call(t,e){return B(()=>{this.invokeCallHook(t,e);let n=Pt(t),o=-1,s=!0,i=Gl(bo(n,this.maskValue),o,s);return P(n,Z(i,n.dtype))})}};lh.className="Masking";et.registerClass(lh);var uh=class extends Ut{constructor(t){super(t);if(this.embeddings=null,this.DEFAULT_EMBEDDINGS_INITIALIZER="randomUniform",t.batchInputShape==null&&t.inputShape==null){let e=null;t.batchSize!=null&&(e=t.batchSize),t.inputLength==null?this.batchInputShape=[e,null]:this.batchInputShape=[e].concat(we(t.inputLength))}this.inputDim=t.inputDim,er(this.inputDim,"inputDim"),this.outputDim=t.outputDim,er(this.outputDim,"outputDim"),this.embeddingsInitializer=de(t.embeddingsInitializer||this.DEFAULT_EMBEDDINGS_INITIALIZER),this.embeddingsRegularizer=ve(t.embeddingsRegularizer),this.activityRegularizer=ve(t.activityRegularizer),this.embeddingsConstraint=Ue(t.embeddingsConstraint),this.maskZero=t.maskZero,this.supportsMasking=t.maskZero,this.inputLength=t.inputLength}build(t){this.embeddings=this.addWeight("embeddings",[this.inputDim,this.outputDim],this.dtype,this.embeddingsInitializer,this.embeddingsRegularizer,!0,this.embeddingsConstraint),this.built=!0}warnOnIncompatibleInputShape(t){}computeMask(t,e){return B(()=>this.maskZero?(t=Pt(t),bo(t,St(t))):null)}computeOutputShape(t){if(t=Jt(t),this.inputLength==null)return[...t,this.outputDim];let e=we(this.inputLength);if(e.length!==t.length-1)throw new z(`"inputLength" is ${this.inputLength}, but received input shape has shape ${t}`);{let n=0;for(let o=0;o<e.length;++o){let s=e[o],i=t[o+1];if(s!=null&&i!=null&&s!==i)throw new z(`"inputLength" is ${this.inputLength}, but received input shape has shape ${t}`);s==null&&(e[n]=i),n++}}return[t[0],...e,this.outputDim]}call(t,e){return B(()=>{this.invokeCallHook(t,e);let n=Pt(t);n.dtype!=="int32"&&(n=fc(n,"int32"));let o=Xx(this.embeddings.read(),O(n,[n.size]));return O(o,Jt(this.computeOutputShape(n.shape)))})}getConfig(){let t={inputDim:this.inputDim,outputDim:this.outputDim,embeddingsInitializer:Te(this.embeddingsInitializer),embeddingsRegularizer:pe(this.embeddingsRegularizer),activityRegularizer:pe(this.activityRegularizer),embeddingsConstraint:We(this.embeddingsConstraint),maskZero:this.maskZero,inputLength:this.inputLength},e=super.getConfig();return Object.assign(t,e),t}};uh.className="Embedding";et.registerClass(uh);var eu=class extends Ut{constructor(t){super(t||{});this.supportsMasking=!0}mergeFunction(t){throw new Tt}computeElementwiseOpOutputShape(t,e){if(t==null||e==null)return null;if(t.length<e.length)return this.computeElementwiseOpOutputShape(e,t);if(e.length===0)return t;let n=t.slice(0,t.length-e.length);for(let o=0;o<e.length;++o){let s=t[t.length-e.length+o],i=e[o];if(s==null||i==null||s<0||i<0)n.push(null);else if(s===1)n.push(i);else if(i===1)n.push(s);else{if(s!==i)throw new z("Operands could not be broadcast together with shapes "+JSON.stringify(t)+" "+JSON.stringify(e));n.push(s)}}return n}build(t){if(Array.isArray(t)&&!Array.isArray(t[0])&&(t=[Jt(t)]),t=t,t.length<2)throw new z(`A merge layer should be called on an Array of at least 2 inputs. Got ${t.length} input(s).`);let e=[];for(let s of t)s!=null&&s[0]!==null&&e.push(s[0]);if(e=ko(e),e.length>1)throw new z(`Can not merge tensors with different batch sizes. Got tensors with shapes: ${JSON.stringify(t)}.`);let n=t[0]==null?null:t[0].slice(1);for(let s=1;s<t.length;++s){let i=t[s]==null?null:t[s].slice(1);n=this.computeElementwiseOpOutputShape(n,i)}let o=t.map(s=>s.length);t.indexOf(null)===-1&&ko(o).length===1?this.reshapeRequired=!1:this.reshapeRequired=!0}call(t,e){return B(()=>{if(t=t,this.reshapeRequired){let n=[],o=t.map(s=>s.rank);if(o.indexOf(null)===-1){let s=ri(o);for(let i of t){let a=i.rank;for(let u=0;u<s-a;++u)i=ol(i,1);n.push(i)}return this.mergeFunction(n)}else{let s=!1;for(let u of t){let l=u.rank;if(l==null){let c=u.shape,p=c[0],m=c.slice(1).concat([p]),f=O(u,[p].concat(No(c.slice(1))));f=Gt(f,[1,0]),f=O(f,m),n.push(f),s=!0}else if(l>1){let c=Yr(1,l).concat([0]);n.push(Gt(u,c)),s=!0}else n.push(u)}let i=this.mergeFunction(n),a=i.rank;if(s){if(a==null){let u=i.shape,l=u.length,c=u[l-1],p=[c].concat(u.slice(0,u.length-1));i=O(Gt(O(i,[-1,c]),[1,0]),p)}else if(a>1){let u=[a-1].concat(Yr(0,a-1));i=Gt(i,u)}}return i}}else return this.mergeFunction(t)})}computeOutputShape(t){t=t;let e;t[0]==null?e=null:e=t[0].slice(1);for(let o=1;o<t.length;++o){let s=t[o]==null?null:t[o].slice(1);e=this.computeElementwiseOpOutputShape(e,s)}let n=[];for(let o of t)o!=null&&o[0]!==null&&n.push(o[0]);return n=ko(n),n.length===1?e=n.concat(e):e=[null].concat(e),e}computeMask(t,e){return B(()=>{if(e==null)return null;if(!Array.isArray(e))throw new z("`mask` should be an Array");if(!Array.isArray(t))throw new z("`inputs` should be an Array");if(e.length!==t.length)throw new z(`The Array 'inputs' and 'mask' are expected to have the same length, but have different lengths (${t.length} vs ${e.length})`);if(e.every(o=>o==null))return null;e=e.map(o=>o==null?o:fr(o,0));let n=e[0];for(let o=1;o<e.length-1;++o)n=Nr(n,e[o]);return n})}},ph=class extends eu{constructor(t){super(t)}mergeFunction(t){return B(()=>{let e=t[0].clone();for(let n=1;n<t.length;++n)e=J(e,t[n]);return e})}};ph.className="Add";et.registerClass(ph);var mh=class extends eu{constructor(t){super(t)}mergeFunction(t){return B(()=>{let e=t[0].clone();for(let n=1;n<t.length;++n)e=P(e,t[n]);return e})}};mh.className="Multiply";et.registerClass(mh);var fh=class extends eu{constructor(t){super(t)}mergeFunction(t){return B(()=>{let e=t[0].clone();for(let n=1;n<t.length;++n)e=J(e,t[n]);return P(1/t.length,e)})}};fh.className="Average";et.registerClass(fh);var dh=class extends eu{constructor(t){super(t)}mergeFunction(t){return B(()=>{let e=t[0];for(let n=1;n<t.length;++n)e=cn(e,t[n]);return e})}};dh.className="Maximum";et.registerClass(dh);var hh=class extends eu{constructor(t){super(t)}mergeFunction(t){return B(()=>{let e=t[0];for(let n=1;n<t.length;++n)e=Zs(e,t[n]);return e})}};hh.className="Minimum";et.registerClass(hh);var gh=class extends eu{constructor(t){super(t);this.DEFAULT_AXIS=-1,t==null&&(t={}),this.axis=t.axis==null?this.DEFAULT_AXIS:t.axis,this.supportsMasking=!0,this.reshapeRequired=!1}build(t){if(!(Array.isArray(t)&&Array.isArray(t[0]))||t.length===1)throw new z("A `Concatenate` layer should be called on a list of at least 2 inputs");t=t;let e=!0;for(let o of t)if(o!=null){e=!1;break}if(e)return;let n=[];for(let o=0;o<t.length;++o){let s=t[o].slice();s.splice(this.axis,1);let i=!1;for(let a of n)if(x.arraysEqual(a,s)){i=!0;break}i||n.push(s)}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(t))}mergeFunction(t){return B(()=>lm(t,this.axis))}computeOutputShape(t){if(!(Array.isArray(t)&&Array.isArray(t[0])))throw new z("A `Concatenate` layer should be called on a list of inputs.");let e=t,n=e[0].slice(),o=this.axis<0?n.length+this.axis:this.axis;for(let s of e.slice(1)){if(n[o]==null||s[o]==null){n[o]=null;break}n[o]+=s[o]}return n}computeMask(t,e){if(e==null)return null;if(!Array.isArray(e))throw new z("`mask` should be an array for Concatenate");if(!Array.isArray(t))throw new z("`inputs` should be an array for Concatenate");if(e.length!==t.length)throw new z(`Mismatch in the length of mask (${e.length}) and the legnth of inputs (${t.length})`);return B(()=>{let n=!0;if(e.forEach(i=>{if(i!=null){n=!1;return}}),n)return null;let o=[];for(let i=0;i<t.length;++i)e[i]==null?o.push(Z(dr(t[i]),"bool")):e[i].rank<t[i].rank?o.push(fr(e[i],-1)):o.push(e[i]);let s=se(o,this.axis);return Ru(s,-1,!1)})}getConfig(){let t={axis:this.axis},e=super.getConfig();return Object.assign(t,e),t}};gh.className="Concatenate";et.registerClass(gh);function ch(r,t){for(;r<0;)r+=t;return r}function T8(r,t,e){if(r.shape.length>3||t.shape.length>3)throw new Tt("batchDot is not implemented for tensors of 4D or higher rank yet");if(x.assert(r.shape.length>=2,()=>`batchDot requires the rank of x to be >= 2, but got ${r.shape.length}`),x.assert(r.shape.length>=2,()=>`batchDot requires the rank of y to be >= 2, but got ${t.shape.length}`),typeof e=="number"&&(e=[e,e]),r.dtype==="complex64"||t.dtype==="complex64")throw new Tt("batchDot is not implemented for complex64-type Tensors yet.");let n=r.shape.length,o=t.shape.length;e==null&&(e=[n-1,o-2]);let s=e;return B(()=>{let i;if(n>o){i=n-o;let u=[];for(let l=0;l<i;++l)u.push(1);t=O(t,t.shape.concat(u))}else if(o>n){i=o-n;let u=[];for(let l=0;l<i;++l)u.push(1);r=O(r,r.shape.concat(u))}else i=0;let a;if(r.shape.length===2&&t.shape.length===2)s[0]===s[1]?a=mt(P(r,t),s[0]):a=mt(P(Gt(r,[1,0]),t),s[1]);else{let u=s[0]!==r.shape.length-1,l=s[1]===t.shape.length-1;a=Mt(r,t,u,l)}if(i>0){let u;n>o?u=n+o-3:u=n-1;let l=[];for(let c=u;c<u+i;++c)l.push(c);a=zr(a,l)}return a.shape.length===1&&(a=fr(a,1)),a})}var xh=class extends eu{constructor(t){super(t);this.axes=t.axes,this.normalize=t.normalize==null?!1:t.normalize,this.supportsMasking=!0,this.reshapeRequired=!1}build(t){x.assert(Array.isArray(t)&&t.length===2&&Array.isArray(t[0])&&Array.isArray(t[1]),()=>"A `Dot` layer should be called on a list of exactly 2 inputs.");let e=t[0],n=t[1];if(e.length>3||n.length>3)throw new Tt("Dot layer does not support tensors of 4D or higher rank yet.");let o=this.interpretAxes(e,n);if(e[o[0]]!==n[o[1]])throw new z(`Dimension incompatibility: ${e[o[0]]} !== ${n[o[1]]}`)}mergeFunction(t){if(t.length!==2)throw new z(`A \`Dot\` layer must be called on exactly 2 inputs, but received ${t.length} input(s).`);let e=t[0],n=t[1],o;return Array.isArray(this.axes)?o=this.axes.map((s,i)=>ch(s,t[i].shape.length)):o=[ch(this.axes,e.shape.length),ch(this.axes,n.shape.length)],this.normalize&&(e=Ad(e,o[0]),n=Ad(n,o[1])),T8(e,n,o)}interpretAxes(t,e){let n;return Array.isArray(this.axes)?n=this.axes:n=[ch(this.axes,t.length),ch(this.axes,e.length)],n}computeOutputShape(t){x.assert(Array.isArray(t)&&t.length===2&&Array.isArray(t[0])&&Array.isArray(t[1]),()=>"A `Dot` layer should be called on a list of exactly 2 inputs.");let e=t[0].slice(),n=t[1].slice();if(e.length>3||n.length>3)throw new Tt("Dot layer does not support tensors of 4D or higher rank yet.");let o=this.interpretAxes(e,n);e.splice(o[0],1),n.splice(o[1],1),n.splice(0,1);let s=e.concat(n);return s.length===1&&s.push(1),s}computeMask(t,e){return null}getConfig(){let t={axes:this.axes,normalize:this.normalize},e=super.getConfig();return Object.assign(t,e),t}};xh.className="Dot";et.registerClass(xh);var yh=class extends Ut{constructor(t){super(t);this.supportsMasking=!0,this.stddev=t.stddev}computeOutputShape(t){return t}getConfig(){let t=super.getConfig(),e={stddev:this.stddev};return Object.assign(e,t),e}call(t,e){return B(()=>{this.invokeCallHook(t,e);let n=Pt(t);return Hl(()=>J(um(n.shape,0,this.stddev),n),()=>n,e.training||!1)})}};yh.className="GaussianNoise";et.registerClass(yh);var bh=class extends Ut{constructor(t){super(t);this.supportsMasking=!0,this.rate=t.rate}computeOutputShape(t){return t}getConfig(){let t=super.getConfig(),e={rate:this.rate};return Object.assign(e,t),e}call(t,e){return B(()=>{this.invokeCallHook(t,e);let n=Pt(t);return this.rate>0&&this.rate<1?Hl(()=>{let s=Math.sqrt(this.rate/(1-this.rate));return P(n,um(n.shape,1,s))},()=>n,e.training||!1):n})}};bh.className="GaussianDropout";et.registerClass(bh);var wh=class extends Ut{constructor(t){super(t);this.supportsMasking=!0,this.rate=t.rate,this.noiseShape=t.noiseShape}_getNoiseShape(t){return this.noiseShape||Pt(t).shape}computeOutputShape(t){return t}getConfig(){let t=super.getConfig(),e={rate:this.rate};return Object.assign(e,t),e}call(t,e){return B(()=>{if(this.rate<1&&this.rate>0){let n=this._getNoiseShape(t);return Hl(()=>{let s=Pt(t),i=1.6732632423543772,a=1.0507009873554805,u=-i*a,l=kn(Js(n),this.rate);l=fc(l,"float32");let c=((1-this.rate)*(1+this.rate*u**2))**-.5,p=-c*u*this.rate,m=J(P(s,l),P(J(l,-1),u));return J(P(m,c),p)},()=>Pt(t),e.training||!1)}return t})}};wh.className="AlphaDropout";et.registerClass(wh);function vh(r,t,e,n,o,s=.001){let i;if(r.rank===2)i=v0(r,t,e,n,o,s);else if(r.rank===3)i=C0(r,t,e,n,o,s);else if(r.rank===4)i=I0(r,t,e,n,o,s);else throw new Tt(`batchNormalization is not implemented for array of rank ${r.rank} yet`);return i}function _8(r,t,e,n,o=.001){return B(()=>{let s=Qp(r,n),i=s.mean,a=s.variance;return[vh(r,i,a,e,t,o),i,a]})}function E8(r,t,e,n,o=.001){return B(()=>{let s=Qp(r,n),i=s.mean,a=s.variance,u=[];for(let d of Yr(0,r.rank))n.indexOf(d)!==-1?u.push(1):u.push(r.shape[d]);let l=O(i,u),c=O(a,u),p=t==null?null:O(t,u),m=e==null?null:O(e,u);return[vh(r,l,c,m,p,o),i,a]})}function A8(r,t,e,n,o=.001){return x.arraysEqual(n.slice().sort(),Yr(0,r.rank-1))?_8(r,t,e,n,o):E8(r,t,e,n,o)}var Ch=class extends Ut{constructor(t){t==null&&(t={});super(t);this.supportsMasking=!0,this.axis=t.axis==null?-1:t.axis,this.momentum=t.momentum==null?.99:t.momentum,this.epsilon=t.epsilon==null?.001:t.epsilon,this.center=t.center==null?!0:t.center,this.scale=t.scale==null?!0:t.scale,this.betaInitializer=de(t.betaInitializer||"zeros"),this.gammaInitializer=de(t.gammaInitializer||"ones"),this.movingMeanInitializer=de(t.movingMeanInitializer||"zeros"),this.movingVarianceInitializer=de(t.movingVarianceInitializer||"ones"),this.betaConstraint=Ue(t.betaConstraint),this.gammaConstraint=Ue(t.gammaConstraint),this.betaRegularizer=ve(t.betaRegularizer),this.gammaRegularizer=ve(t.gammaRegularizer)}build(t){t=Jt(t);let e=this.axis>=0?this.axis:this.axis+t.length,n=t[e];if(n==null)throw new z(`Axis ${e} of input tensor should have a defined dimension but the layer received an input with shape ${JSON.stringify(t)}.`);this.inputSpec=[new _e({ndim:t.length,axes:{[e]:n}})];let o=[n];this.scale&&(this.gamma=this.addWeight("gamma",o,null,this.gammaInitializer,this.gammaRegularizer,!0,this.gammaConstraint)),this.center&&(this.beta=this.addWeight("beta",o,null,this.betaInitializer,this.betaRegularizer,!0,this.betaConstraint)),this.movingMean=this.addWeight("moving_mean",o,null,this.movingMeanInitializer,null,!1),this.movingVariance=this.addWeight("moving_variance",o,null,this.movingVarianceInitializer,null,!1),this.built=!0}call(t,e){return B(()=>{let n=e.training==null?!1:e.training,o=Pt(t),s=o.shape,i=s.length,a=Yr(0,i),u=this.axis>=0?this.axis:this.axis+i;a.splice(u,1);let l=Io(1,i);l[u]=s[u];let c=a.slice();c.sort();let p=!x.arraysEqual(c,Yr(0,i).slice(0,i-1)),m=()=>{if(p){let b=O(this.movingMean.read(),l),w=O(this.movingVariance.read(),l),v=this.center?O(this.beta.read(),l):null,k=this.scale?O(this.gamma.read(),l):null;return vh(o,b,w,v,k,this.epsilon)}else return vh(o,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 m();let[f,d,h]=A8(o,this.gamma.read(),this.beta.read(),a,this.epsilon),g=(b,w,v)=>{B(()=>{let k=1-v,E=b.read(),D=P(lt(E,w),k);b.write(lt(E,D))})};return(()=>{g(this.movingMean,d,this.momentum),g(this.movingVariance,h,this.momentum)})(),f})}getConfig(){let t={axis:this.axis,momentum:this.momentum,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:Te(this.betaInitializer),gammaInitializer:Te(this.gammaInitializer),movingMeanInitializer:Te(this.movingMeanInitializer),movingVarianceInitializer:Te(this.movingVarianceInitializer),betaRegularizer:pe(this.betaRegularizer),gammaRegularizer:pe(this.gammaRegularizer),betaConstraint:We(this.betaConstraint),gammaConstraint:We(this.gammaConstraint)},e=super.getConfig();return Object.assign(t,e),t}};Ch.className="BatchNormalization";et.registerClass(Ch);var Ih=class extends Ut{constructor(t){t==null&&(t={});super(t);if(this.axis=t.axis==null?-1:t.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 e of this.axis)if(!Number.isInteger(e))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=t.epsilon==null?.001:t.epsilon,this.center=t.center==null?!0:t.center,this.scale=t.scale==null?!0:t.scale,this.betaInitializer=de(t.betaInitializer||"zeros"),this.gammaInitializer=de(t.gammaInitializer||"ones"),this.betaRegularizer=ve(t.betaRegularizer),this.gammaRegularizer=ve(t.gammaRegularizer),this.supportsMasking=!0}build(t){t=Jt(t);let e=t.length;typeof this.axis=="number"&&(this.axis=[this.axis]);for(let s=0;s<this.axis.length;++s)this.axis[s]<0&&(this.axis[s]+=e);for(let s of this.axis)if(s<0||s>=e)throw new Error(`Invalid axis: ${s}`);if(this.axis.length!==ko(this.axis).length)throw new Error(`Found duplicate axes in: ${this.axis}`);let n=this.axis.map(s=>t[s]),o=!0;this.scale?this.gamma=this.addWeight("gamma",n,"float32",this.gammaInitializer,this.gammaRegularizer,o):this.gamma=null,this.center?this.beta=this.addWeight("beta",n,"float32",this.betaInitializer,this.betaRegularizer,o):this.beta=null,this.built=!0}call(t,e){let n=Pt(t),o=n.shape,s=o.length;return B(()=>{let{mean:a,variance:u}=Qp(n,this.axis,!0),l=Io(1,s);for(let h of this.axis)l[h]=o[h];let c=h=>h!=null&&h.shape.length!==s?O(h,l):h,p=c(this.gamma.read()),m=c(this.beta.read()),f=[],d=[];for(let h=0;h<s;++h)this.axis.indexOf(h)!==-1?(f.push(o[h]),d.push(1)):(f.push(1),d.push(o[h]));return a=kr(a,f),u=kr(u,f),p=kr(p,d),m=kr(m,d),vh(n,a,u,m,p,this.epsilon)})}getConfig(){let t={axis:this.axis,epsilon:this.epsilon,center:this.center,scale:this.scale,betaInitializer:Te(this.betaInitializer),gammaInitializer:Te(this.gammaInitializer),betaRegularizer:pe(this.betaRegularizer),gammaRegularizer:pe(this.gammaRegularizer)},e=super.getConfig();return Object.assign(t,e),t}};Ih.className="LayerNormalization";et.registerClass(Ih);function D8(r,t,e){return B(()=>{if(r.rank!==4)throw new z(`temporalPadding expects input tensor to be 4-D, but received a ${r.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(e==null&&(e=mn()),e!=="channelsLast"&&e!=="channelsFirst")throw new z(`Unknown data format: ${e}. Supported data formats are 'channelsLast' and 'channelsFirst.`);let n;return e==="channelsFirst"?n=[[0,0],[0,0],t[0],t[1]]:n=[[0,0],t[0],t[1],[0,0]],qr(r,n)})}var Sh=class extends Ut{constructor(t){t==null&&(t={});super(t);if(this.dataFormat=t.dataFormat==null?mn():t.dataFormat,t.padding==null)this.padding=[[1,1],[1,1]];else if(typeof t.padding=="number")this.padding=[[t.padding,t.padding],[t.padding,t.padding]];else{if(t.padding=t.padding,t.padding.length!==2)throw new z(`ZeroPadding2D expects padding to be a length-2 array, but received a length-${t.padding.length} array.`);let e,n;if(typeof t.padding[0]=="number")e=[t.padding[0],t.padding[0]],n=[t.padding[1],t.padding[1]];else{if(t.padding=t.padding,t.padding[0].length!==2)throw new z(`ZeroPadding2D expects height padding to be a length-2 array, but received a length-${t.padding[0].length} array.`);if(e=t.padding[0],t.padding[1].length!==2)throw new z(`ZeroPadding2D expects width padding to be a length-2 array, but received a length-${t.padding[1].length} array.`);n=t.padding[1]}this.padding=[e,n]}this.inputSpec=[new _e({ndim:4})]}computeOutputShape(t){t=Jt(t);let e,n;return this.dataFormat==="channelsFirst"?(t[2]!=null&&t[2]>=0?e=t[2]+this.padding[0][0]+this.padding[0][1]:e=null,t[3]!=null&&t[3]>=0?n=t[3]+this.padding[1][0]+this.padding[1][1]:n=null,[t[0],t[1],e,n]):(t[1]!=null&&t[1]>=0?e=t[1]+this.padding[0][0]+this.padding[0][1]:e=null,t[2]!=null&&t[2]>=0?n=t[2]+this.padding[1][0]+this.padding[1][1]:n=null,[t[0],e,n,t[3]])}call(t,e){return B(()=>D8(Pt(t),this.padding,this.dataFormat))}getConfig(){let t={padding:this.padding,dataFormat:this.dataFormat},e=super.getConfig();return Object.assign(t,e),t}};Sh.className="ZeroPadding2D";et.registerClass(Sh);function gy(r,t,e,n,o,s){return B(()=>{Le(o),gC(s),fn(n),e==null&&(e=[1,1]),n==null&&(n="valid"),o==null&&(o=mn()),s==null&&(s="max"),r=Ud(r,o);let i,a=n==="same"?"same":"valid";return s==="max"?i=ja(r,t,e,a):i=Va(r,t,e,a),o==="channelsFirst"&&(i=Gt(i,[0,3,1,2])),i})}function PD(r,t,e,n,o,s){return B(()=>{Le(o),gC(s),fn(n),e==null&&(e=[1,1,1]),n==null&&(n="valid"),o==null&&(o=mn()),s==null&&(s="max"),r=XC(r,o);let i,a=n==="same"?"same":"valid";return s==="max"?i=Qf(r,t,e,a):i=zf(r,t,e,a),o==="channelsFirst"&&(i=Gt(i,[0,4,1,2,3])),i})}var tI=class extends Ut{constructor(t){t.poolSize==null&&(t.poolSize=2);super(t);if(typeof t.poolSize=="number")this.poolSize=[t.poolSize];else if(Array.isArray(t.poolSize)&&t.poolSize.length===1&&typeof t.poolSize[0]=="number")this.poolSize=t.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(t.poolSize)}`);if(er(this.poolSize,"poolSize"),t.strides==null)this.strides=this.poolSize;else if(typeof t.strides=="number")this.strides=[t.strides];else if(Array.isArray(t.strides)&&t.strides.length===1&&typeof t.strides[0]=="number")this.strides=t.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(t.strides)}`);er(this.strides,"strides"),this.padding=t.padding==null?"valid":t.padding,fn(this.padding),this.inputSpec=[new _e({ndim:3})]}computeOutputShape(t){t=Jt(t);let e=En(t[1],this.poolSize[0],this.padding,this.strides[0]);return[t[0],e,t[2]]}call(t,e){return B(()=>{this.invokeCallHook(t,e),t=ol(Pt(t),2);let n=this.poolingFunction(Pt(t),[this.poolSize[0],1],[this.strides[0],1],this.padding,"channelsLast");return zr(n,[2])})}getConfig(){let t={poolSize:this.poolSize,padding:this.padding,strides:this.strides},e=super.getConfig();return Object.assign(t,e),t}},kh=class extends tI{constructor(t){super(t)}poolingFunction(t,e,n,o,s){return Le(s),fn(o),gy(t,e,n,o,s,"max")}};kh.className="MaxPooling1D";et.registerClass(kh);var Nh=class extends tI{constructor(t){super(t)}poolingFunction(t,e,n,o,s){return Le(s),fn(o),gy(t,e,n,o,s,"avg")}};Nh.className="AveragePooling1D";et.registerClass(Nh);var eI=class extends Ut{constructor(t){t.poolSize==null&&(t.poolSize=[2,2]);super(t);if(this.poolSize=Array.isArray(t.poolSize)?t.poolSize:[t.poolSize,t.poolSize],t.strides==null)this.strides=this.poolSize;else if(Array.isArray(t.strides)){if(t.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 ${t.strides.length}.`);this.strides=t.strides}else this.strides=[t.strides,t.strides];er(this.poolSize,"poolSize"),er(this.strides,"strides"),this.padding=t.padding==null?"valid":t.padding,this.dataFormat=t.dataFormat==null?"channelsLast":t.dataFormat,Le(this.dataFormat),fn(this.padding),this.inputSpec=[new _e({ndim:4})]}computeOutputShape(t){t=Jt(t);let e=this.dataFormat==="channelsFirst"?t[2]:t[1],n=this.dataFormat==="channelsFirst"?t[3]:t[2];return e=En(e,this.poolSize[0],this.padding,this.strides[0]),n=En(n,this.poolSize[1],this.padding,this.strides[1]),this.dataFormat==="channelsFirst"?[t[0],t[1],e,n]:[t[0],e,n,t[3]]}call(t,e){return B(()=>(this.invokeCallHook(t,e),this.poolingFunction(Pt(t),this.poolSize,this.strides,this.padding,this.dataFormat)))}getConfig(){let t={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},e=super.getConfig();return Object.assign(t,e),t}},Th=class extends eI{constructor(t){super(t)}poolingFunction(t,e,n,o,s){return Le(s),fn(o),gy(t,e,n,o,s,"max")}};Th.className="MaxPooling2D";et.registerClass(Th);var _h=class extends eI{constructor(t){super(t)}poolingFunction(t,e,n,o,s){return Le(s),fn(o),gy(t,e,n,o,s,"avg")}};_h.className="AveragePooling2D";et.registerClass(_h);var rI=class extends Ut{constructor(t){t.poolSize==null&&(t.poolSize=[2,2,2]);super(t);if(this.poolSize=Array.isArray(t.poolSize)?t.poolSize:[t.poolSize,t.poolSize,t.poolSize],t.strides==null)this.strides=this.poolSize;else if(Array.isArray(t.strides)){if(t.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 ${t.strides.length}.`);this.strides=t.strides}else this.strides=[t.strides,t.strides,t.strides];er(this.poolSize,"poolSize"),er(this.strides,"strides"),this.padding=t.padding==null?"valid":t.padding,this.dataFormat=t.dataFormat==null?"channelsLast":t.dataFormat,Le(this.dataFormat),fn(this.padding),this.inputSpec=[new _e({ndim:5})]}computeOutputShape(t){t=Jt(t);let e=this.dataFormat==="channelsFirst"?t[2]:t[1],n=this.dataFormat==="channelsFirst"?t[3]:t[2],o=this.dataFormat==="channelsFirst"?t[4]:t[3];return e=En(e,this.poolSize[0],this.padding,this.strides[0]),n=En(n,this.poolSize[1],this.padding,this.strides[1]),o=En(o,this.poolSize[2],this.padding,this.strides[2]),this.dataFormat==="channelsFirst"?[t[0],t[1],e,n,o]:[t[0],e,n,o,t[4]]}call(t,e){return B(()=>(this.invokeCallHook(t,e),this.poolingFunction(Pt(t),this.poolSize,this.strides,this.padding,this.dataFormat)))}getConfig(){let t={poolSize:this.poolSize,padding:this.padding,strides:this.strides,dataFormat:this.dataFormat},e=super.getConfig();return Object.assign(t,e),t}},Eh=class extends rI{constructor(t){super(t)}poolingFunction(t,e,n,o,s){return Le(s),fn(o),PD(t,e,n,o,s,"max")}};Eh.className="MaxPooling3D";et.registerClass(Eh);var Ah=class extends rI{constructor(t){super(t)}poolingFunction(t,e,n,o,s){return Le(s),fn(o),PD(t,e,n,o,s,"avg")}};Ah.className="AveragePooling3D";et.registerClass(Ah);var nI=class extends Ut{constructor(t){super(t);this.inputSpec=[new _e({ndim:3})]}computeOutputShape(t){return[t[0],t[2]]}call(t,e){throw new Tt}},Dh=class extends nI{constructor(t){super(t||{})}call(t,e){return B(()=>{let n=Pt(t);return xe(n,1)})}};Dh.className="GlobalAveragePooling1D";et.registerClass(Dh);var $h=class extends nI{constructor(t){super(t||{})}call(t,e){return B(()=>{let n=Pt(t);return $r(n,1)})}};$h.className="GlobalMaxPooling1D";et.registerClass($h);var oI=class extends Ut{constructor(t){super(t);this.dataFormat=t.dataFormat==null?"channelsLast":t.dataFormat,Le(this.dataFormat),this.inputSpec=[new _e({ndim:4})]}computeOutputShape(t){return t=t,this.dataFormat==="channelsLast"?[t[0],t[3]]:[t[0],t[1]]}call(t,e){throw new Tt}getConfig(){let t={dataFormat:this.dataFormat},e=super.getConfig();return Object.assign(t,e),t}},Rh=class extends oI{call(t,e){return B(()=>{let n=Pt(t);return this.dataFormat==="channelsLast"?xe(n,[1,2]):xe(n,[2,3])})}};Rh.className="GlobalAveragePooling2D";et.registerClass(Rh);var Fh=class extends oI{call(t,e){return B(()=>{let n=Pt(t);return this.dataFormat==="channelsLast"?$r(n,[1,2]):$r(n,[2,3])})}};Fh.className="GlobalMaxPooling2D";et.registerClass(Fh);var sI=class extends Ut{constructor(t){super(t);this.layer=t.layer}build(t){this.built=!0}get trainable(){return this.layer!=null?this.layer.trainable:!1}set trainable(t){this.layer!=null&&(this.layer.trainable=t)}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(t){this.layer.setWeights(t)}getConfig(){let t={layer:{className:this.layer.getClassName(),config:this.layer.getConfig()}},e=super.getConfig();return Object.assign(t,e),t}setFastWeightInitDuringBuild(t){super.setFastWeightInitDuringBuild(t),this.layer!=null&&this.layer.setFastWeightInitDuringBuild(t)}static fromConfig(t,e,n={}){let o=e.layer,s=xn(o,n);delete e.layer;let i={layer:s};return Object.assign(i,e),new t(i)}},Oh=class extends sI{constructor(t){super(t);this.supportsMasking=!0}build(t){if(t=Jt(t),t.length<3)throw new z(`TimeDistributed layer expects an input shape >= 3D, but received input shape ${JSON.stringify(t)}`);this.inputSpec=[{shape:t}];let e=[t[0]].concat(t.slice(2));this.layer.built||(this.layer.build(e),this.layer.built=!0),super.build(t)}computeOutputShape(t){t=Jt(t);let e=[t[0]].concat(t.slice(2)),n=this.layer.computeOutputShape(e),o=t[1];return[n[0],o].concat(n.slice(1))}call(t,e){return B(()=>(t=Pt(t),JC((i,a)=>[Pt(this.layer.call(i,e)),[]],t,[],!1,null,null,!1,!0)[1]))}};Oh.className="TimeDistributed";et.registerClass(Oh);function $8(r){zi(W2,"BidirectionalMergeMode",r)}var R8="concat",Ph=class extends sI{constructor(t){super(t);let e=t.layer.getConfig(),n={};n.className=t.layer.getClassName(),n.config=e,this.forwardLayer=xn(n),e.goBackwards=e.goBackwards!==!0;let o={};if(o.className=t.layer.getClassName(),o.config=e,this.backwardLayer=xn(o),this.forwardLayer.name="forward_"+this.forwardLayer.name,this.backwardLayer.name="backward_"+this.backwardLayer.name,this.mergeMode=t.mergeMode===void 0?R8:t.mergeMode,$8(this.mergeMode),t.weights)throw new Tt("weights support is not implemented for Bidirectional layer yet.");this._stateful=t.layer.stateful,this.returnSequences=t.layer.returnSequences,this.returnState=t.layer.returnState,this.supportsMasking=!0,this._trainable=!0,this.inputSpec=t.layer.inputSpec,this.numConstants=null}get trainable(){return this._trainable}set trainable(t){this._trainable=t,this.forwardLayer!=null&&(this.forwardLayer.trainable=t),this.backwardLayer!=null&&(this.backwardLayer.trainable=t)}getWeights(){return this.forwardLayer.getWeights().concat(this.backwardLayer.getWeights())}setWeights(t){let e=t.length,n=Math.floor(e/2);this.forwardLayer.setWeights(t.slice(0,n)),this.backwardLayer.setWeights(t.slice(n))}computeOutputShape(t){let e=this.forwardLayer.computeOutputShape(t);Array.isArray(e)&&Array.isArray(e[0])||(e=[e]),e=e;let n,o,s;return this.returnState&&(s=e.slice(1)),n=e[0],n=n,this.mergeMode==="concat"?(n[n.length-1]*=2,o=[n]):this.mergeMode==null?o=[n,n.slice()]:o=[n],this.returnState?this.mergeMode==null?o.concat(s).concat(s.slice()):[n].concat(s).concat(s.slice()):_r(o)}apply(t,e){let n=e==null?null:e.initialState,o=e==null?null:e.constants;e==null&&(e={});let s=ZC(t,n,o,this.numConstants);if(t=s.inputs,n=s.initialState,o=s.constants,Array.isArray(t)&&(n=t.slice(1),t=t[0]),(n==null||n.length===0)&&o==null)return super.apply(t,e);let i=[],a=[];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.");e.initialState=n,i.push(...n);let c=n.map(p=>new _e({shape:p.shape}));this.forwardLayer.stateSpec=c.slice(0,l/2),this.backwardLayer.stateSpec=c.slice(l/2),a.push(...c)}if(o!=null)throw new Tt("Support for constants in Bidirectional layers is not implemented yet.");let u=i[0]instanceof gn;for(let l of i)if(l instanceof gn!==u)throw new z("The initial state of a Bidirectional layer cannot be specified as a mix of symbolic and non-symbolic tensors");if(u){let l=[t].concat(i),c=this.inputSpec.concat(a),p=this.inputSpec;this.inputSpec=c;let m=super.apply(l,e);return this.inputSpec=p,m}else return super.apply(t,e)}call(t,e){return B(()=>{let n=e.initialState,o,s;if(n==null)o=this.forwardLayer.call(t,e),s=this.backwardLayer.call(t,e);else{let u=n.slice(0,n.length/2),l=n.slice(n.length/2);o=this.forwardLayer.call(t,Object.assign(e,{initialState:u})),s=this.backwardLayer.call(t,Object.assign(e,{initialState:l}))}let i;this.returnState&&(Array.isArray(o)&&(i=o.slice(1).concat(s.slice(1))),o=o[0],s=s[0]),this.returnSequences&&(s=sr(s,1));let a;return this.mergeMode==="concat"?a=lm([o,s]):this.mergeMode==="sum"?a=J(o,s):this.mergeMode==="ave"?a=P(.5,J(o,s)):this.mergeMode==="mul"?a=P(o,s):this.mergeMode==null&&(a=[o,s]),this.returnState?this.mergeMode==null?a.concat(i):[a].concat(i):a})}resetStates(t){this.forwardLayer.resetStates(),this.backwardLayer.resetStates()}build(t){ei(this.forwardLayer.name,()=>{this.forwardLayer.build(t)}),ei(this.backwardLayer.name,()=>{this.backwardLayer.build(t)}),this.built=!0}computeMask(t,e){Array.isArray(e)&&(e=e[0]);let n;if(this.returnSequences?this.mergeMode==null?n=[e,e]:n=e:this.mergeMode==null?n=[null,null]:n=null,this.returnState){let s=this.forwardLayer.states.map(i=>null);return Array.isArray(n)?n.concat(s).concat(s):[n].concat(s).concat(s)}else return n}get trainableWeights(){return this.forwardLayer.trainableWeights.concat(this.backwardLayer.trainableWeights)}get nonTrainableWeights(){return this.forwardLayer.nonTrainableWeights.concat(this.backwardLayer.nonTrainableWeights)}setFastWeightInitDuringBuild(t){super.setFastWeightInitDuringBuild(t),this.forwardLayer!=null&&this.forwardLayer.setFastWeightInitDuringBuild(t),this.backwardLayer!=null&&this.backwardLayer.setFastWeightInitDuringBuild(t)}getConfig(){let t={mergeMode:this.mergeMode},e=super.getConfig();return Object.assign(t,e),t}static fromConfig(t,e){let n=xn(e.layer);if(delete e.layer,e.numConstants!=null)throw new Tt("Deserialization of a Bidirectional layer with numConstants present is not supported yet.");let o=e;return o.layer=n,new t(o)}};Ph.className="Bidirectional";et.registerClass(Ph);function F8(r){return new Bi(r)}function O8(r){return new Vd(r)}function P8(r){return new Ld(r)}function M8(r){return new zd(r)}function L8(r){return new Bd(r)}function z8(r){return new Wd(r)}function B8(r){return new Gd(r)}function V8(r){return new vc(r)}function G8(r){return new Zl(r)}function W8(r){return new Hd(r)}function U8(r){return new Jl(r)}function H8(r){return new qd(r)}function q8(r){return new Kd(r)}function K8(r){return new jd(r)}function j8(r){return new Xd(r)}function X8(r){return new Yd(r)}function Y8(r){return new oh(r)}function Z8(r){return new rh(r)}function J8(r){return new Em(r)}function Q8(r){return new eh(r)}function tY(r){return new nh(r)}function eY(r){return new sh(r)}function rY(r){return new ih(r)}function nY(r){return new ah(r)}function oY(r){return new uh(r)}function sY(r){return new ph(r)}function iY(r){return new fh(r)}function aY(r){return new gh(r)}function lY(r){return new dh(r)}function uY(r){return new hh(r)}function cY(r){return new mh(r)}function pY(r){return new xh(r)}function mY(r){return new Ch(r)}function fY(r){return new Ih(r)}function dY(r){return new Sh(r)}function iI(r){return new Nh(r)}function hY(r){return iI(r)}function gY(r){return iI(r)}function aI(r){return new _h(r)}function xY(r){return aI(r)}function yY(r){return aI(r)}function lI(r){return new Ah(r)}function bY(r){return lI(r)}function wY(r){return lI(r)}function vY(r){return new Dh(r)}function CY(r){return new Rh(r)}function MD(r){return new $h(r)}function LD(r){return new Fh(r)}function zD(r){return new kh(r)}function BD(r){return new Th(r)}function IY(r){return new Eh(r)}function SY(r){return new Jd(r)}function kY(r){return new Nm(r)}function NY(r){return new Qd(r)}function TY(r){return new tu(r)}function _Y(r){return new Zd(r)}function EY(r){return new km(r)}function AY(r){return new th(r)}function DY(r){return new _m(r)}function $Y(r){return new zn(r)}function RY(r){return new Tm(r)}function FY(r){return new Ph(r)}function OY(r){return new Oh(r)}var PY=MD,MY=LD,LY=zD,zY=BD;function BY(r){return new yh(r)}function VY(r){return new bh(r)}function GY(r){return new wh(r)}function WY(r){return new lh(r)}var GD={};Yt(GD,{MAPE:()=>t7,MSE:()=>n7,binaryAccuracy:()=>UY,binaryCrossentropy:()=>HY,categoricalAccuracy:()=>KY,categoricalCrossentropy:()=>jY,cosineProximity:()=>ZY,mape:()=>e7,meanAbsoluteError:()=>JY,meanAbsolutePercentageError:()=>QY,meanSquaredError:()=>r7,mse:()=>o7,precision:()=>XY,recall:()=>YY,sparseCategoricalAccuracy:()=>qY});function UY(r,t){return Rd(r,t)}function HY(r,t){return ay(r,t)}function qY(r,t){return ly(r,t)}function KY(r,t){return Fd(r,t)}function jY(r,t){return Od(r,t)}function XY(r,t){return SC(r,t)}function YY(r,t){return pD(r,t)}function ZY(r,t){return $d(r,t)}function JY(r,t){return bm(r,t)}function QY(r,t){return Xl(r,t)}function t7(r,t){return Xl(r,t)}function e7(r,t){return Xl(r,t)}function r7(r,t){return Gi(r,t)}function n7(r,t){return Gi(r,t)}function o7(r,t){return Gi(r,t)}var WD={};Yt(WD,{modelFromJSON:()=>ED});var UD={};Yt(UD,{l1:()=>i7,l1l2:()=>s7,l2:()=>a7});function s7(r){return new bc(r)}function i7(r){return RD(r)}function a7(r){return FD(r)}var uI=class extends jl{constructor(){super(...arguments);this.model=null}setModel(t){if(!(t instanceof Qn))throw new Error("model must be a LayersModel, not some other Container");this.model=t}};function xy(r,t){return r<t}function HD(r,t){return r>t}var cI=class extends uI{constructor(t){super();if(t==null&&(t={}),t.restoreBestWeights)throw new Tt("restoreBestWeights = True is not implemented in EarlyStopping yet.");this.monitor=t.monitor||"val_loss",this.minDelta=Math.abs(t.minDelta||0),this.patience=t.patience||0,this.verbose=t.verbose||0,this.mode=t.mode||"auto",this.baseline=t.baseline,["auto","min","max"].indexOf(this.mode)===-1&&(console.warn(`EarlyStopping mode '${this.mode}' is invalid. Falling back to mode 'auto'.`),this.mode="auto"),this.mode==="min"?this.monitorFunc=xy:this.mode==="max"?this.monitorFunc=HD:this.monitor.indexOf("acc")!==-1?this.monitorFunc=HD:this.monitorFunc=xy,this.monitorFunc===xy&&(this.minDelta*=-1)}async onTrainBegin(t){this.wait=0,this.stoppedEpoch=0,this.baseline!=null?this.best=this.baseline:this.best=this.monitorFunc===xy?1/0:-1/0}async onEpochEnd(t,e){await Vi(e);let n=this.getMonitorValue(e);n!=null&&(this.monitorFunc(n-this.minDelta,this.best)?(this.best=n,this.wait=0):(this.wait++,this.wait>=this.patience&&(this.stoppedEpoch=t,this.model.stopTraining=!0)))}async onTrainEnd(t){this.stoppedEpoch>0&&this.verbose&&console.log(`Epoch ${this.stoppedEpoch}: early stopping.`)}getMonitorValue(t){t==null&&(t={});let e=t[this.monitor];return e==null&&console.warn(`Metric for EarlyStopping ${this.monitor} is not available. Available metrics are: ${Object.keys(t)}`),e}};function l7(r){return new cI(r)}var u7={earlyStopping:l7};var c7=V();c7.registerFlag("KEEP_INTERMEDIATE_TENSORS",()=>!1,r=>{r&&console.warn("Keep intermediate tensors is ON. This will print the values of all intermediate tensors during model inference. Not all models support this mode. For details, check e2e/benchmarks/ model_config.js. This significantly impacts performance.")});var to;(function(r){r[r.DT_INVALID=0]="DT_INVALID",r[r.DT_FLOAT=1]="DT_FLOAT",r[r.DT_DOUBLE=2]="DT_DOUBLE",r[r.DT_INT32=3]="DT_INT32",r[r.DT_UINT8=4]="DT_UINT8",r[r.DT_INT16=5]="DT_INT16",r[r.DT_INT8=6]="DT_INT8",r[r.DT_STRING=7]="DT_STRING",r[r.DT_COMPLEX64=8]="DT_COMPLEX64",r[r.DT_INT64=9]="DT_INT64",r[r.DT_BOOL=10]="DT_BOOL",r[r.DT_QINT8=11]="DT_QINT8",r[r.DT_QUINT8=12]="DT_QUINT8",r[r.DT_QINT32=13]="DT_QINT32",r[r.DT_BFLOAT16=14]="DT_BFLOAT16",r[r.DT_QINT16=15]="DT_QINT16",r[r.DT_QUINT16=16]="DT_QUINT16",r[r.DT_UINT16=17]="DT_UINT16",r[r.DT_COMPLEX128=18]="DT_COMPLEX128",r[r.DT_HALF=19]="DT_HALF",r[r.DT_RESOURCE=20]="DT_RESOURCE",r[r.DT_VARIANT=21]="DT_VARIANT",r[r.DT_UINT32=22]="DT_UINT32",r[r.DT_UINT64=23]="DT_UINT64",r[r.DT_FLOAT_REF=101]="DT_FLOAT_REF",r[r.DT_DOUBLE_REF=102]="DT_DOUBLE_REF",r[r.DT_INT32_REF=103]="DT_INT32_REF",r[r.DT_UINT8_REF=104]="DT_UINT8_REF",r[r.DT_INT16_REF=105]="DT_INT16_REF",r[r.DT_INT8_REF=106]="DT_INT8_REF",r[r.DT_STRING_REF=107]="DT_STRING_REF",r[r.DT_COMPLEX64_REF=108]="DT_COMPLEX64_REF",r[r.DT_INT64_REF=109]="DT_INT64_REF",r[r.DT_BOOL_REF=110]="DT_BOOL_REF",r[r.DT_QINT8_REF=111]="DT_QINT8_REF",r[r.DT_QUINT8_REF=112]="DT_QUINT8_REF",r[r.DT_QINT32_REF=113]="DT_QINT32_REF",r[r.DT_BFLOAT16_REF=114]="DT_BFLOAT16_REF",r[r.DT_QINT16_REF=115]="DT_QINT16_REF",r[r.DT_QUINT16_REF=116]="DT_QUINT16_REF",r[r.DT_UINT16_REF=117]="DT_UINT16_REF",r[r.DT_COMPLEX128_REF=118]="DT_COMPLEX128_REF",r[r.DT_HALF_REF=119]="DT_HALF_REF",r[r.DT_RESOURCE_REF=120]="DT_RESOURCE_REF",r[r.DT_VARIANT_REF=121]="DT_VARIANT_REF",r[r.DT_UINT32_REF=122]="DT_UINT32_REF",r[r.DT_UINT64_REF=123]="DT_UINT64_REF"})(to||(to={}));var qD;(function(r){let t;(function(e){e[e.LEGACY=0]="LEGACY",e[e.V1=1]="V1",e[e.V2=2]="V2"})(t=r.CheckpointFormatVersion||(r.CheckpointFormatVersion={}))})(qD||(qD={}));var pI={};function m7(r,t){let e={tfOpName:r,category:"custom",inputs:[],attrs:[],customExecutor:t};pI[r]=e}function yy(r){return pI[r]}function f7(r){delete pI[r]}function I(r,t,e,n,o){let s=t.inputParams[r];if(s&&s.inputIndexStart!==void 0){let a=s.inputIndexStart,u=s.inputIndexEnd===0?void 0:s.inputIndexEnd===void 0?a+1:s.inputIndexEnd;if(s.type==="tensor")return Cr(t.inputNames[s.inputIndexStart],e,n,o);if(s.type==="tensors")return t.inputNames.slice(a,u).map(m=>Cr(m,e,n,o));let l=Cr(t.inputNames.slice(a)[0],e,n,o),c=l.dataSync();return s.type==="number"?c[0]:x.toNestedArray(l.shape,c)}let i=t.attrParams[r];return i&&i.value}function Cr(r,t,e,n){let[o,s]=bn(r);if(n!=null){let a=n.getHashTableHandleByName(o);if(a!=null)return a}let i=e.currentContextIds.find(a=>!!t[by(o,a)]);return i!==void 0?t[by(o,i)][s]:void 0}function KD(r,t,e){return t[by(r,e.currentContextId)]}function Eo(r,t){let[e,n,o]=bn(r);return[by(e,t&&t.currentContextId),n,o]}function by(r,t){return t?`${r}-${t}`:r}function bn(r){let t=r.split(":");if(t.length===1)return[r,0,void 0];let e=t[0],n=t.length===3?t[1]:void 0,o=Number(t[t.length-1]);return[e,o,n]}function Mh(r,t,e){let n=I("pad",r,t,e);if(n==="explicit"){n=I("explicitPaddings",r,t,e);let o=[[0,0],[0,0],[0,0],[0,0]];for(let s=0;s<4;s++)o[s][0]=n[s*2],o[s][1]=n[s*2+1];return o}return n}function ai(r){return r.kept?r:In(r)}var mI={};Yt(mI,{json:()=>d7});var d7=[{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}]}];var fI={};Yt(fI,{json:()=>h7});var h7=[{tfOpName:"Abs",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Acos",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Asin",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atan",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atan2",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"y",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Ceil",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ClipByValue",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"clipValueMin",type:"number"},{start:2,name:"clipValueMax",type:"number"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Complex",category:"basic_math",inputs:[{start:0,name:"real",type:"tensor"},{start:1,name:"imag",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"ComplexAbs",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Cos",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Cosh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Elu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Exp",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Floor",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Log",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Imag",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"Tout",name:"outputType",type:"dtype",notSupported:!0}]},{tfOpName:"Neg",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Real",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0},{tfName:"Tout",name:"outputType",type:"dtype",notSupported:!0}]},{tfOpName:"Prelu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"alpha",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Relu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Relu6",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Selu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sigmoid",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sin",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sinh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sqrt",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Rsqrt",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Square",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Tan",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Tanh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Sign",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Round",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Expm1",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Log1p",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Reciprocal",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Softplus",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Asinh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Acosh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Atanh",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Erf",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Prod",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axes",type:"number[]"}],attrs:[{tfName:"keep_dims",name:"keepDims",type:"bool",notSupported:!0},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"LeakyRelu",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"alpha",name:"alpha",type:"number",defaultValue:.2},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"IsNan",category:"basic_math",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]}];var dI={};Yt(dI,{json:()=>g7});var g7=[{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"}]}];var hI={};Yt(hI,{json:()=>x7});var x7=[{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"}]}];var gI={};Yt(gI,{json:()=>y7});var y7=[{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"}]}];var xI={};Yt(xI,{json:()=>b7});var b7=[{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}]}];var yI={};Yt(yI,{json:()=>w7});var w7=[{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"}]}];var bI={};Yt(bI,{json:()=>v7});var v7=[{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"}]}];var wI={};Yt(wI,{json:()=>C7});var C7=[{tfOpName:"HashTable",category:"hash_table",inputs:[],attrs:[{tfName:"shared_name",name:"sharedName",type:"string"},{tfName:"use_node_name_sharing",name:"useNodeNameSharing",type:"bool"},{tfName:"key_dtype",name:"keyDType",type:"dtype"},{tfName:"value_dtype",name:"valueDType",type:"dtype"}]},{tfOpName:"HashTableV2",category:"hash_table",inputs:[],attrs:[{tfName:"shared_name",name:"sharedName",type:"string"},{tfName:"use_node_name_sharing",name:"useNodeNameSharing",type:"bool"},{tfName:"key_dtype",name:"keyDType",type:"dtype"},{tfName:"value_dtype",name:"valueDType",type:"dtype"}]},{tfOpName:"LookupTableImport",category:"hash_table",inputs:[{start:0,name:"tableHandle",type:"tensor"},{start:1,name:"keys",type:"tensor"},{start:2,name:"values",type:"tensor"}],attrs:[{tfName:"Tin",name:"tIn",type:"dtype",notSupported:!0},{tfName:"Tout",name:"tOut",type:"dtype",notSupported:!0}]},{tfOpName:"LookupTableImportV2",category:"hash_table",inputs:[{start:0,name:"tableHandle",type:"tensor"},{start:1,name:"keys",type:"tensor"},{start:2,name:"values",type:"tensor"}],attrs:[{tfName:"Tin",name:"tIn",type:"dtype",notSupported:!0},{tfName:"Tout",name:"tOut",type:"dtype",notSupported:!0}]},{tfOpName:"LookupTableFind",category:"hash_table",inputs:[{start:0,name:"tableHandle",type:"tensor"},{start:1,name:"keys",type:"tensor"},{start:2,name:"defaultValue",type:"tensor"}],attrs:[{tfName:"Tin",name:"tIn",type:"dtype",notSupported:!0},{tfName:"Tout",name:"tOut",type:"dtype",notSupported:!0}]},{tfOpName:"LookupTableFindV2",category:"hash_table",inputs:[{start:0,name:"tableHandle",type:"tensor"},{start:1,name:"keys",type:"tensor"},{start:2,name:"defaultValue",type:"tensor"}],attrs:[{tfName:"Tin",name:"tIn",type:"dtype",notSupported:!0},{tfName:"Tout",name:"tOut",type:"dtype",notSupported:!0}]},{tfOpName:"LookupTableSize",category:"hash_table",inputs:[{start:0,name:"tableHandle",type:"tensor"}]},{tfOpName:"LookupTableSizeV2",category:"hash_table",inputs:[{start:0,name:"tableHandle",type:"tensor"}]}];var vI={};Yt(vI,{json:()=>I7});var I7=[{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"}]},{tfOpName:"ImageProjectiveTransformV3",category:"image",inputs:[{start:0,name:"images",type:"tensor"},{start:1,name:"transforms",type:"tensor"},{start:2,name:"outputShape",type:"number[]"},{start:3,name:"fillValue",type:"number"}],attrs:[{tfName:"interpolation",name:"interpolation",type:"string"},{tfName:"fill_mode",name:"fillMode",type:"string"}]}];var CI={};Yt(CI,{json:()=>S7});var S7=[{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}]}];var II={};Yt(II,{json:()=>k7});var k7=[{tfOpName:"_FusedMatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"},{start:2,end:0,name:"args",type:"tensors"}],attrs:[{tfName:"num_args",name:"numArgs",type:"number"},{tfName:"fused_ops",name:"fusedOps",type:"string[]",defaultValue:[]},{tfName:"epsilon",name:"epsilon",type:"number",defaultValue:1e-4},{tfName:"transpose_a",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"transpose_b",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"MatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"transpose_a",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"transpose_b",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"BatchMatMul",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"adj_x",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"adj_y",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"BatchMatMulV2",category:"matrices",inputs:[{start:0,name:"a",type:"tensor"},{start:1,name:"b",type:"tensor"}],attrs:[{tfName:"adj_x",name:"transposeA",type:"bool",defaultValue:!1},{tfName:"adj_y",name:"transposeB",type:"bool",defaultValue:!1},{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Transpose",category:"matrices",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"perm",type:"number[]"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"Einsum",category:"matrices",inputs:[{start:0,end:0,name:"tensors",type:"tensors"}],attrs:[{tfName:"equation",name:"equation",type:"string"},{tfName:"N",name:"n",type:"number",defaultValue:2},{tfName:"T",name:"dtype",type:"dtype"}]}];var SI={};Yt(SI,{json:()=>N7});var N7=[{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}]}];var kI={};Yt(kI,{json:()=>T7});var T7=[{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:"Cumprod",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"}]},{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"}]}];var NI={};Yt(NI,{json:()=>_7});var _7=[{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}]}];var TI={};Yt(TI,{json:()=>E7});var E7=[{tfOpName:"SparseFillEmptyRows",category:"sparse",inputs:[{start:0,name:"indices",type:"tensor"},{start:1,name:"values",type:"tensor"},{start:2,name:"denseShape",type:"tensor"},{start:3,name:"defaultValue",type:"tensor"}]},{tfOpName:"SparseReshape",category:"sparse",inputs:[{start:0,name:"inputIndices",type:"tensor"},{start:1,name:"inputShape",type:"tensor"},{start:2,name:"newShape",type:"tensor"}],attrs:[{tfName:"T",name:"dtype",type:"dtype",notSupported:!0}]},{tfOpName:"SparseSegmentMean",category:"sparse",inputs:[{start:0,name:"data",type:"tensor"},{start:1,name:"indices",type:"tensor"},{start:2,name:"segmentIds",type:"tensor"}]},{tfOpName:"SparseSegmentSum",category:"sparse",inputs:[{start:0,name:"data",type:"tensor"},{start:1,name:"indices",type:"tensor"},{start:2,name:"segmentIds",type:"tensor"}]}];var _I={};Yt(_I,{json:()=>A7});var A7=[{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}]}];var EI={};Yt(EI,{json:()=>D7});var D7=[{tfOpName:"StringNGrams",category:"string",inputs:[{start:0,name:"data",type:"tensor"},{start:1,name:"dataSplits",type:"tensor"}],attrs:[{tfName:"separator",name:"separator",type:"string"},{tfName:"ngram_widths",name:"nGramWidths",type:"number[]"},{tfName:"left_pad",name:"leftPad",type:"string"},{tfName:"right_pad",name:"rightPad",type:"string"},{tfName:"pad_width",name:"padWidth",type:"number"},{tfName:"preserve_short_sequences",name:"preserveShortSequences",type:"bool"}],outputs:["ngrams","ngrams_splits"]},{tfOpName:"StringSplit",category:"string",inputs:[{start:0,name:"input",type:"tensor"},{start:1,name:"delimiter",type:"tensor"}],attrs:[{tfName:"skip_empty",name:"skipEmpty",type:"bool"}],outputs:["indices","values","shape"]},{tfOpName:"StringToHashBucketFast",category:"string",inputs:[{start:0,name:"input",type:"tensor"}],attrs:[{tfName:"num_buckets",name:"numBuckets",type:"number"}]}];var AI={};Yt(AI,{json:()=>$7});var $7=[{tfOpName:"Cast",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"SrcT",name:"sdtype",type:"dtype",notSupported:!0},{tfName:"DstT",name:"dtype",type:"dtype"}]},{tfOpName:"ExpandDims",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"axis",type:"number"}]},{tfOpName:"MirrorPad",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"padding",type:"number[]"}],attrs:[{tfName:"mode",name:"mode",type:"string"}]},{tfOpName:"Pad",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"padding",type:"number[]"}],attrs:[{tfName:"constant_value",name:"constantValue",type:"number",defaultValue:0}]},{tfOpName:"PadV2",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"padding",type:"number[]"},{start:2,name:"constantValue",type:"number",defaultValue:0}]},{tfOpName:"Reshape",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"shape",type:"number[]"}]},{tfOpName:"Squeeze",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"axis",tfDeprecatedName:"squeeze_dims",name:"axis",type:"number[]"}]},{tfOpName:"SpaceToBatchND",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"blockShape",type:"number[]"},{start:2,name:"paddings",type:"number[]"}]},{tfOpName:"BatchToSpaceND",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"blockShape",type:"number[]"},{start:2,name:"crops",type:"number[]"}]},{tfOpName:"DepthToSpace",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"}],attrs:[{tfName:"block_size",name:"blockSize",type:"number"},{tfName:"data_format",name:"dataFormat",type:"string"}]},{tfOpName:"BroadcastTo",category:"transformation",inputs:[{start:0,name:"x",type:"tensor"},{start:1,name:"shape",type:"number[]"}],attrs:[]},{tfOpName:"BroadcastArgs",category:"transformation",inputs:[{start:0,name:"s0",type:"tensor"},{start:1,name:"s1",type:"tensor"}],attrs:[]}];var Ay=class{static get Instance(){return this._instance||(this._instance=new this)}constructor(){let t=[mI,fI,dI,hI,gI,xI,yI,bI,wI,vI,CI,II,SI,kI,NI,TI,_I,EI,AI],e=[].concat(...t.map(n=>n.json));this.opMappers=e.reduce((n,o)=>(n[o.tfOpName]=o,n),{})}transformGraph(t,e={}){let n=t.node,o=[],s=[],i=[],a=n.reduce((h,g)=>(h[g.name]=this.mapNode(g),g.op.startsWith("Placeholder")?o.push(h[g.name]):g.op==="Const"?s.push(h[g.name]):(g.input==null||g.input.length===0)&&i.push(h[g.name]),h),{}),u=[],l=[],c={},p={};e!=null&&(c=this.mapSignatureEntries(e.inputs),p=this.mapSignatureEntries(e.outputs));let m=Object.keys(a);m.forEach(h=>{let g=a[h];g.inputNames.forEach((y,b)=>{let[w,,v]=Eo(y),k=a[w];if(k.outputs!=null){let E=k.outputs.indexOf(v);if(E!==-1){let D=`${w}:${E}`;g.inputNames[b]=D}}g.inputs.push(k),k.children.push(g)})}),Object.keys(p).length===0?m.forEach(h=>{let g=a[h];g.children.length===0&&l.push(g)}):Object.keys(p).forEach(h=>{let[g]=Eo(h),y=a[g];y!=null&&(y.signatureKey=p[h],l.push(y))}),Object.keys(c).length>0?Object.keys(c).forEach(h=>{let[g]=Eo(h),y=a[g];y&&(y.signatureKey=c[h],u.push(y))}):u=o;let f={};t.library!=null&&t.library.function!=null&&(f=t.library.function.reduce((h,g)=>(h[g.signature.name]=this.mapFunction(g),h),{}));let d={nodes:a,inputs:u,outputs:l,weights:s,placeholders:o,signature:e,functions:f};return i.length>0&&(d.initNodes=i),d}mapSignatureEntries(t){return Object.keys(t||{}).reduce((e,n)=>(e[t[n].name]=n,e),{})}mapNode(t){let e=yy(t.op)||this.opMappers[t.op]||{};t.attr==null&&(t.attr={});let n={name:t.name,op:t.op,category:e.category,inputNames:(t.input||[]).map(o=>o.startsWith("^")?o.substr(1):o),inputs:[],children:[],inputParams:{},attrParams:{},rawAttrs:t.attr,outputs:e.outputs};return e.inputs!=null&&(n.inputParams=e.inputs.reduce((o,s)=>(o[s.name]={type:s.type,inputIndexStart:s.start,inputIndexEnd:s.end},o),{})),e.attrs!=null&&(n.attrParams=e.attrs.reduce((o,s)=>{let i=s.type,a;switch(s.type){case"string":a=wy(t.attr,s.tfName,s.defaultValue),a===void 0&&!!s.tfDeprecatedName&&(a=wy(t.attr,s.tfDeprecatedName,s.defaultValue));break;case"string[]":a=Ty(t.attr,s.tfName,s.defaultValue),a===void 0&&!!s.tfDeprecatedName&&(a=Ty(t.attr,s.tfDeprecatedName,s.defaultValue));break;case"number":a=Cy(t.attr,s.tfName,s.defaultValue||0),a===void 0&&!!s.tfDeprecatedName&&(a=Cy(t.attr,s.tfDeprecatedName,s.defaultValue));break;case"number[]":a=Ny(t.attr,s.tfName,s.defaultValue),a===void 0&&!!s.tfDeprecatedName&&(a=Ny(t.attr,s.tfDeprecatedName,s.defaultValue));break;case"bool":a=vy(t.attr,s.tfName,s.defaultValue),a===void 0&&!!s.tfDeprecatedName&&(a=vy(t.attr,s.tfDeprecatedName,s.defaultValue));break;case"bool[]":a=Ey(t.attr,s.tfName,s.defaultValue),a===void 0&&!!s.tfDeprecatedName&&(a=Ey(t.attr,s.tfDeprecatedName,s.defaultValue));break;case"shape":a=ky(t.attr,s.tfName,s.defaultValue),a===void 0&&!!s.tfDeprecatedName&&(a=ky(t.attr,s.tfDeprecatedName,s.defaultValue));break;case"shape[]":a=_y(t.attr,s.tfName,s.defaultValue),a===void 0&&!!s.tfDeprecatedName&&(a=_y(t.attr,s.tfDeprecatedName,s.defaultValue));break;case"dtype":a=Iy(t.attr,s.tfName,s.defaultValue),a===void 0&&!!s.tfDeprecatedName&&(a=Iy(t.attr,s.tfDeprecatedName,s.defaultValue));break;case"dtype[]":a=Sy(t.attr,s.tfName,s.defaultValue),a===void 0&&!!s.tfDeprecatedName&&(a=Sy(t.attr,s.tfDeprecatedName,s.defaultValue));break;case"func":a=jD(t.attr,s.tfName,s.defaultValue),a===void 0&&!!s.tfDeprecatedName&&(a=jD(t.attr,s.tfDeprecatedName,s.defaultValue));break;case"tensor":case"tensors":break;default:throw new Error(`Unsupported param type: ${s.type} for op: ${t.op}`)}return o[s.name]={value:a,type:i},o},{})),n}mapFunction(t){let e=t.nodeDef,n=[],o=[],s={};e!=null&&(s=e.reduce((p,m)=>(p[m.name]=this.mapNode(m),m.op==="Const"&&o.push(p[m.name]),p),{}));let i=[],a=[];t.signature.inputArg.forEach(p=>{let[m]=Eo(p.name),f={name:m,op:"Placeholder",inputs:[],inputNames:[],category:"graph",inputParams:{},attrParams:{dtype:{value:DI(p.type),type:"dtype"}},children:[]};f.signatureKey=p.name,i.push(f),s[m]=f}),Object.keys(s).forEach(p=>{let m=s[p];m.inputNames.forEach((f,d)=>{let[h,,g]=Eo(f),y=s[h];if(y.outputs!=null){let b=y.outputs.indexOf(g);if(b!==-1){let w=`${h}:${b}`;m.inputNames[d]=w}}m.inputs.push(y),y.children.push(m)})});let l=t.ret;t.signature.outputArg.forEach(p=>{let[m,f]=Eo(l[p.name]),d=s[m];d!=null&&(d.defaultOutput=f,a.push(d))});let c=this.mapArgsToSignature(t);return{nodes:s,inputs:i,outputs:a,weights:o,placeholders:n,signature:c}}mapArgsToSignature(t){return{methodName:t.signature.name,inputs:t.signature.inputArg.reduce((e,n)=>(e[n.name]=this.mapArgToTensorInfo(n),e),{}),outputs:t.signature.outputArg.reduce((e,n)=>(e[n.name]=this.mapArgToTensorInfo(n,t.ret),e),{})}}mapArgToTensorInfo(t,e){let n=t.name;return e!=null&&(n=e[n]),{name:n,dtype:t.type}}};function R7(r){let t=V().global;if(typeof t.atob!="undefined")return t.atob(r);if(typeof Buffer!="undefined")return new Buffer(r,"base64").toString();throw new Error("Unable to decode base64 in this environment. Missing built-in atob() or Buffer()")}function XD(r,t){let e=Array.isArray(r)?String.fromCharCode.apply(null,r):R7(r);return t?e:e.toLowerCase()}function wy(r,t,e,n=!1){let o=r[t];return o!=null?XD(o.s,n):e}function vy(r,t,e){let n=r[t];return n?n.b:e}function Cy(r,t,e){let n=r[t]||{},o=n.i!=null?n.i:n.f!=null?n.f:e;return typeof o=="number"?o:parseInt(o,10)}function DI(r){switch(typeof r=="string"&&(r=to[r]),r){case to.DT_FLOAT:case to.DT_HALF:return"float32";case to.DT_INT32:case to.DT_INT64:case to.DT_INT8:case to.DT_UINT8:return"int32";case to.DT_BOOL:return"bool";case to.DT_DOUBLE:return"float32";case to.DT_STRING:return"string";default:return null}}function jD(r,t,e){let n=r[t];return n&&n.func?n.func.name:e}function Iy(r,t,e){let n=r[t];return n&&n.type?DI(n.type):e}function Sy(r,t,e){let n=r[t];return n&&n.list&&n.list.type?n.list.type.map(o=>DI(o)):e}function YD(r){if(!r.unknownRank)return r.dim!=null?r.dim.map(t=>typeof t.size=="number"?t.size:parseInt(t.size,10)):[]}function ky(r,t,e){let n=r[t];return n&&n.shape?YD(n.shape):e}function Ny(r,t,e){let n=r[t];return n?((n.list.f&&n.list.f.length?n.list.f:n.list.i)||[]).map(o=>typeof o=="number"?o:parseInt(o,10)):e}function Ty(r,t,e,n=!1){let o=r[t];return o&&o.list&&o.list.s?o.list.s.map(s=>XD(s,n)):e}function _y(r,t,e){let n=r[t];return n&&n.list&&n.list.shape?n.list.shape.map(o=>YD(o)):e}function Ey(r,t,e){let n=r[t];return n&&n.list&&n.list.b?n.list.b:e}var $I=class{constructor(t,e,n){this.node=t,this.tensorMap=e,this.context=n,this.inputs=[],this.attrs={},this.inputs=t.inputNames.map(o=>this.getInput(o)),t.rawAttrs!=null&&(this.attrs=Object.keys(t.rawAttrs).reduce((o,s)=>(o[s]=this.getAttr(s),o),{}))}getInput(t){return Cr(t,this.tensorMap,this.context)}getAttr(t,e){let n=this.node.rawAttrs[t];if(n.tensor!=null)return Cr(t,this.tensorMap,this.context);if(n.i!=null||n.f!=null)return Cy(this.node.rawAttrs,t,e);if(n.s!=null)return wy(this.node.rawAttrs,t,e);if(n.b!=null)return vy(this.node.rawAttrs,t,e);if(n.shape!=null)return ky(this.node.rawAttrs,t,e);if(n.type!=null)return Iy(this.node.rawAttrs,t,e);if(n.list!=null){if(n.list.i!=null||n.list.f!=null)return Ny(this.node.rawAttrs,t,e);if(n.list.s!=null)return Ty(this.node.rawAttrs,t,e);if(n.list.shape!=null)return _y(this.node.rawAttrs,t,e);if(n.list.b!=null)return Ey(this.node.rawAttrs,t,e);if(n.list.type!=null)return Sy(this.node.rawAttrs,t,e)}return e}};var ZD=(r,t,e)=>{switch(r.op){case"BiasAdd":case"AddV2":case"Add":return[J(I("a",r,t,e),I("b",r,t,e))];case"AddN":return[x0(I("tensors",r,t,e))];case"FloorMod":case"Mod":return[ed(I("a",r,t,e),I("b",r,t,e))];case"Mul":return[P(I("a",r,t,e),I("b",r,t,e))];case"RealDiv":case"Div":return[ct(I("a",r,t,e),I("b",r,t,e))];case"DivNoNan":return[qf(I("a",r,t,e),I("b",r,t,e))];case"FloorDiv":return[$u(I("a",r,t,e),I("b",r,t,e))];case"Sub":return[lt(I("a",r,t,e),I("b",r,t,e))];case"Minimum":return[Zs(I("a",r,t,e),I("b",r,t,e))];case"Maximum":return[cn(I("a",r,t,e),I("b",r,t,e))];case"Pow":return[Kr(I("a",r,t,e),I("b",r,t,e))];case"SquaredDifference":return[Qu(I("a",r,t,e),I("b",r,t,e))];default:throw TypeError(`Node type ${r.op} is not implemented`)}};var JD=(r,t,e)=>{switch(r.op){case"Abs":case"ComplexAbs":return[Se(I("x",r,t,e))];case"Acos":return[Df(I("x",r,t,e))];case"Acosh":return[$f(I("x",r,t,e))];case"Asin":return[Ff(I("x",r,t,e))];case"Asinh":return[Of(I("x",r,t,e))];case"Atan":return[Pf(I("x",r,t,e))];case"Atan2":return[Mf(I("x",r,t,e),I("y",r,t,e))];case"Atanh":return[Lf(I("x",r,t,e))];case"Ceil":return[Vf(I("x",r,t,e))];case"Complex":return[Mn(I("real",r,t,e),I("imag",r,t,e))];case"Cos":return[Ua(I("x",r,t,e))];case"Cosh":return[Lu(I("x",r,t,e))];case"Elu":return[js(I("x",r,t,e))];case"Erf":return[Kf(I("x",r,t,e))];case"Exp":return[Ze(I("x",r,t,e))];case"Expm1":return[jf(I("x",r,t,e))];case"Floor":return[Ys(I("x",r,t,e))];case"Log":return[wr(I("x",r,t,e))];case"Log1p":return[qa(I("x",r,t,e))];case"Imag":return[Bu(I("x",r,t,e))];case"Neg":return[Zt(I("x",r,t,e))];case"Reciprocal":return[rd(I("x",r,t,e))];case"Real":return[Ul(I("x",r,t,e))];case"Relu":return[Tr(I("x",r,t,e))];case"Round":return[Ku(I("x",r,t,e))];case"Selu":return[Xu(I("x",r,t,e))];case"Sigmoid":return[Lr(I("x",r,t,e))];case"Sin":return[Yu(I("x",r,t,e))];case"Sign":return[od(I("x",r,t,e))];case"Sinh":return[Zu(I("x",r,t,e))];case"Softplus":return[xo(I("x",r,t,e))];case"Sqrt":return[be(I("x",r,t,e))];case"Square":return[Wt(I("x",r,t,e))];case"Tanh":return[qs(I("x",r,t,e))];case"Tan":return[ld(I("x",r,t,e))];case"ClipByValue":return[br(I("x",r,t,e),I("clipValueMin",r,t,e),I("clipValueMax",r,t,e))];case"Relu6":return[qu(I("x",r,t,e))];case"Rsqrt":return[ju(Cr(r.inputNames[0],t,e))];case"Prod":return[Uu(I("x",r,t,e),I("axes",r,t,e))];case"LeakyRelu":return[Ha(I("x",r,t,e),I("alpha",r,t,e))];case"Prelu":return[Ya(I("x",r,t,e),I("alpha",r,t,e))];case"IsNan":return[Xf(Cr(r.inputNames[0],t,e))];default:throw TypeError(`Node type ${r.op} is not implemented`)}};function Bn(r,t,e=""){if(!(typeof r=="number"||typeof t=="number")){x.assert(r.length===t.length,()=>e+` Shapes ${r} and ${t} must match`);for(let n=0;n<r.length;n++){let o=r[n],s=t[n];x.assert(o<0||s<0||o===s,()=>e+` Shapes ${r} and ${t} must match`)}}}function QD(r){return!(typeof r=="number"||r.some(t=>t<0))}function Am(r,t,e){let n=Dy(r,e),o=!QD(n);if(o&&t.length===0)throw new Error(`Tried to calculate elements of an empty list with non-fully-defined elementShape: ${n}`);if(o&&t.forEach(s=>{n=Dy(s.shape,n)}),!QD(n))throw new Error(`Non-fully-defined elementShape: ${n}`);return n}function Dy(r,t){if(typeof r=="number")return t;if(typeof t=="number")return r;if(r.length!==t.length)throw new Error(`Incompatible ranks during merge: ${r} vs. ${t}`);let e=[];for(let n=0;n<r.length;++n){let o=r[n],s=t[n];if(o>=0&&s>=0&&o!==s)throw new Error(`Incompatible shape during merge: ${r} vs. ${t}`);e[n]=o>=0?o:s}return e}var RI=class{constructor(t,e,n,o,s,i,a){this.name=t,this.dtype=e,this.maxSize=n,this.elementShape=o,this.identicalElementShapes=s,this.dynamicSize=i,this.clearAfterRead=a,this.tensors=[],this.closed_=!1,this.idTensor=pt(0),Me(this.idTensor)}get id(){return this.idTensor.id}get closed(){return this.closed_}clearAndClose(t){this.tensors.forEach(e=>{(t==null||!t.has(e.tensor.id))&&e.tensor.dispose()}),this.tensors=[],this.closed_=!0,this.idTensor.dispose()}size(){return this.tensors.length}read(t){if(this.closed_)throw new Error(`TensorArray ${this.name} has already been closed.`);if(t<0||t>=this.size())throw new Error(`Tried to read from index ${t}, but array size is: ${this.size()}`);let e=this.tensors[t];if(e.cleared)throw new Error(`TensorArray ${this.name}: Could not read index ${t} twice because it was cleared after a previous read (perhaps try setting clear_after_read = false?).`);return this.clearAfterRead&&(e.cleared=!0),e.read=!0,e.tensor}readMany(t){return t.map(e=>this.read(e))}write(t,e){if(this.closed_)throw new Error(`TensorArray ${this.name} has already been closed.`);if(t<0||!this.dynamicSize&&t>=this.maxSize)throw new Error(`Tried to write to index ${t}, but array is not resizeable and size is: ${this.maxSize}`);let n=this.tensors[t]||{};if(e.dtype!==this.dtype)throw new Error(`TensorArray ${this.name}: Could not write to TensorArray index ${t},
because the value dtype is ${e.dtype}, but TensorArray dtype is ${this.dtype}.`);if(this.size()===0&&(this.elementShape==null||this.elementShape.length===0)&&(this.elementShape=e.shape),Bn(this.elementShape,e.shape,`TensorArray ${this.name}: Could not write to TensorArray index ${t}.`),n.read)throw new Error(`TensorArray ${this.name}: Could not write to TensorArray index ${t}, because it has already been read.`);if(n.written)throw new Error(`TensorArray ${this.name}: Could not write to TensorArray index ${t}, because it has already been written.`);n.tensor=e,Me(e),n.written=!0,this.tensors[t]=n}writeMany(t,e){if(t.length!==e.length)throw new Error(`TensorArray ${this.name}: could not write multiple tensors,because the index size: ${t.length} is not the same as tensors size: ${e.length}.`);t.forEach((n,o)=>this.write(n,e[o]))}gather(t,e){if(!!e&&e!==this.dtype)throw new Error(`TensorArray dtype is ${this.dtype} but gather requested dtype ${e}`);if(t)t=t.slice(0,this.size());else{t=[];for(let o=0;o<this.size();o++)t.push(o)}if(t.length===0)return Ar([],[0].concat(this.elementShape));let n=this.readMany(t);return Bn(this.elementShape,n[0].shape,"TensorArray shape mismatch: "),Je(n,0)}concat(t){if(!!t&&t!==this.dtype)throw new Error(`TensorArray dtype is ${this.dtype} but concat requested dtype ${t}`);if(this.size()===0)return Ar([],[0].concat(this.elementShape));let e=[];for(let o=0;o<this.size();o++)e.push(o);let n=this.readMany(e);return Bn(this.elementShape,n[0].shape,`TensorArray shape mismatch: tensor array shape (${this.elementShape}) vs first tensor shape (${n[0].shape})`),se(n,0)}scatter(t,e){if(e.dtype!==this.dtype)throw new Error(`TensorArray dtype is ${this.dtype} but tensor has dtype ${e.dtype}`);if(t.length!==e.shape[0])throw new Error(`Expected len(indices) == tensor.shape[0], but saw: ${t.length} vs. ${e.shape[0]}`);let n=Math.max(...t);if(!this.dynamicSize&&n>=this.maxSize)throw new Error(`Max index must be < array size (${n} vs. ${this.maxSize})`);this.writeMany(t,vr(e,0))}split(t,e){if(e.dtype!==this.dtype)throw new Error(`TensorArray dtype is ${this.dtype} but tensor has dtype ${e.dtype}`);let n=0,o=t.map(u=>(n+=u,n));if(n!==e.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: ${e.shape}`);if(!this.dynamicSize&&t.length!==this.maxSize)throw new Error(`TensorArray's size is not equal to the size of lengths (${this.maxSize} vs. ${t.length}), and the TensorArray is not marked as dynamically resizeable`);let s=n===0?0:e.size/n,i=[];B(()=>{e=O(e,[1,n,s]);for(let u=0;u<t.length;++u){let l=u===0?0:o[u-1],c=[0,l,0],p=[1,t[u],s];i[u]=O(Rt(e,c,p),this.elementShape)}return i});let a=[];for(let u=0;u<t.length;u++)a[u]=u;this.writeMany(a,i)}};var Cc=class{constructor(t,e,n,o=-1){this.tensors=t,this.elementShape=e,this.elementDtype=n,t!=null&&t.forEach(s=>{if(n!==s.dtype)throw new Error(`Invalid data types; op elements ${n}, but list elements ${s.dtype}`);Bn(e,s.shape,"TensorList shape mismatch: "),Me(s)}),this.idTensor=pt(0),this.maxNumElements=o,Me(this.idTensor)}get id(){return this.idTensor.id}copy(){return new Cc([...this.tensors],this.elementShape,this.elementDtype)}clearAndClose(t){this.tensors.forEach(e=>{(t==null||!t.has(e.id))&&e.dispose()}),this.tensors.length=0,this.idTensor.dispose()}size(){return this.tensors.length}stack(t,e,n=-1){if(e!==this.elementDtype)throw new Error(`Invalid data types; op elements ${e}, 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.`);Bn(t,this.elementShape,"TensorList shape mismatch: ");let o=Am(this.elementShape,this.tensors,t);return B(()=>{let s=this.tensors.map(i=>O(i,o));return Je(s,0)})}popBack(t,e){if(e!==this.elementDtype)throw new Error(`Invalid data types; op elements ${e}, but list elements ${this.elementDtype}`);if(this.size()===0)throw new Error("Trying to pop from an empty list.");let n=Am(this.elementShape,this.tensors,t),o=this.tensors.pop();return Bn(o.shape,t,"TensorList shape mismatch: "),O(o,n)}pushBack(t){if(t.dtype!==this.elementDtype)throw new Error(`Invalid data types; op elements ${t.dtype}, but list elements ${this.elementDtype}`);if(Bn(t.shape,this.elementShape,"TensorList shape mismatch: "),this.maxNumElements===this.size())throw new Error("Trying to push element into a full list.");Me(t),this.tensors.push(t)}resize(t){if(t<0)throw new Error(`TensorListResize expects size to be non-negative. Got: ${t}`);if(this.maxNumElements!==-1&&t>this.maxNumElements)throw new Error(`TensorListResize input size ${t} is greater maxNumElement ${this.maxNumElements}.`);this.tensors.length=t}getItem(t,e,n){if(n!==this.elementDtype)throw new Error(`Invalid data types; op elements ${n}, but list elements ${this.elementDtype}`);if(t<0||t>this.tensors.length)throw new Error(`Trying to access element ${t} in a list with ${this.tensors.length} elements.`);if(this.tensors[t]==null)throw new Error(`element at index ${t} is null.`);Bn(this.tensors[t].shape,e,"TensorList shape mismatch: ");let o=Am(this.elementShape,this.tensors,e);return O(this.tensors[t],o)}setItem(t,e){if(e.dtype!==this.elementDtype)throw new Error(`Invalid data types; op elements ${e.dtype}, but list elements ${this.elementDtype}`);if(t<0||this.maxNumElements!==-1&&t>=this.maxNumElements)throw new Error(`Trying to set element ${t} in a list with max ${this.maxNumElements} elements.`);Bn(this.elementShape,e.shape,"TensorList shape mismatch: "),Me(e),this.tensors[t]=e}gather(t,e,n){if(e!==this.elementDtype)throw new Error(`Invalid data types; op elements ${e}, but list elements ${this.elementDtype}`);Bn(this.elementShape,n,"TensorList shape mismatch: "),t=t.slice(0,this.size());let o=Am(this.elementShape,this.tensors,n);return t.length===0?Ar([],[0].concat(o)):B(()=>{let s=t.map(i=>O(this.tensors[i],o));return Je(s,0)})}concat(t,e){if(!!t&&t!==this.elementDtype)throw new Error(`TensorList dtype is ${this.elementDtype} but concat requested dtype ${t}`);Bn(this.elementShape,e,"TensorList shape mismatch: ");let n=Am(this.elementShape,this.tensors,e);return this.size()===0?Ar([],[0].concat(n)):B(()=>{let o=this.tensors.map(s=>O(s,n));return se(o,0)})}};function t$(r,t,e){let n=r.dtype;if(r.shape.length<1)throw new Error(`Tensor must be at least a vector, but saw shape: ${r.shape}`);if(r.dtype!==e)throw new Error(`Invalid data types; op elements ${r.dtype}, but list elements ${e}`);let o=r.shape.slice(1);Bn(o,t,"TensorList shape mismatch: ");let s=vr(r);return new Cc(s,t,n)}function e$(r,t,e){return new Cc([],r,t,e)}function r$(r,t,e,n){if(t.length!==r.shape[0])throw new Error(`Expected len(indices) == tensor.shape[0], but saw: ${t.length} vs. ${r.shape[0]}`);let o=Math.max(...t);if(n!=null&&n!==-1&&o>=n)throw new Error(`Max index must be < array size (${o} vs. ${n})`);let s=new Cc([],e,r.dtype,n),i=vr(r,0);return t.forEach((a,u)=>{s.setItem(a,i[u])}),s}function n$(r,t,e){let n=0,o=t.map(c=>(n+=c,n));if(n!==r.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: ${r.shape}`);let s=r.shape.slice(1),i=Dy(s,e),a=n===0?0:r.size/n,u=B(()=>{let c=[];r=O(r,[1,n,a]);for(let p=0;p<t.length;++p){let m=p===0?0:o[p-1],f=[0,m,0],d=[1,t[p],a];c[p]=O(Rt(r,f,d),i)}return r.dispose(),c}),l=new Cc([],e,r.dtype,t.length);for(let c=0;c<u.length;c++)l.setItem(c,u[c]);return l}var o$=async(r,t,e)=>{switch(r.op){case"If":case"StatelessIf":{let n=I("thenBranch",r,t,e),o=I("elseBranch",r,t,e),s=I("cond",r,t,e),i=I("args",r,t,e);return(await s.data())[0]?e.functionMap[n].executeFunctionAsync(i,e.tensorArrayMap,e.tensorListMap):e.functionMap[o].executeFunctionAsync(i,e.tensorArrayMap,e.tensorListMap)}case"While":case"StatelessWhile":{let n=I("body",r,t,e),o=I("cond",r,t,e),s=I("args",r,t,e),i=await e.functionMap[o].executeFunctionAsync(s,e.tensorArrayMap,e.tensorListMap),a=s.map(c=>c.id),u=await i[0].data();i.forEach(c=>{!c.kept&&a.indexOf(c.id)===-1&&c.dispose()});let l=s;for(;u[0];){let c=l;l=await e.functionMap[n].executeFunctionAsync(l,e.tensorArrayMap,e.tensorListMap);let p=l.map(f=>f.id);c.forEach(f=>{!f.kept&&a.indexOf(f.id)===-1&&p.indexOf(f.id)===-1&&f.dispose()});let m=await e.functionMap[o].executeFunctionAsync(l,e.tensorArrayMap,e.tensorListMap);u=await m[0].data(),m.forEach(f=>{!f.kept&&a.indexOf(f.id)===-1&&p.indexOf(f.id)===-1&&f.dispose()})}return l}case"LoopCond":{let n=I("pred",r,t,e);return[ai(n)]}case"Switch":{let n=I("pred",r,t,e),o=I("data",r,t,e);return o.kept||(o=ai(o)),(await n.data())[0]?[void 0,o]:[o,void 0]}case"Merge":{let n=r.inputNames.find(o=>Cr(o,t,e)!==void 0);if(n){let o=Cr(n,t,e);return[ai(o)]}return}case"Enter":{let n=I("frameName",r,t,e),o=I("tensor",r,t,e);return e.enterFrame(n),[ai(o)]}case"Exit":{let n=I("tensor",r,t,e);return e.exitFrame(),[ai(n)]}case"NextIteration":{let n=I("tensor",r,t,e);return e.nextIteration(),[ai(n)]}case"TensorArrayV3":{let n=I("size",r,t,e),o=I("dtype",r,t,e),s=I("elementShape",r,t,e),i=I("dynamicSize",r,t,e),a=I("clearAfterRead",r,t,e),u=I("identicalElementShapes",r,t,e),l=I("name",r,t,e),c=new RI(l,o,n,s,u,i,a);return e.addTensorArray(c),[c.idTensor,pt(1)]}case"TensorArrayWriteV3":{let n=I("tensorArrayId",r,t,e),o=I("index",r,t,e),s=I("tensor",r,t,e),i=e.getTensorArray(n.id);return i.write(o,s),[i.idTensor]}case"TensorArrayReadV3":{let n=I("tensorArrayId",r,t,e),o=I("index",r,t,e);return[e.getTensorArray(n.id).read(o)]}case"TensorArrayGatherV3":{let n=I("tensorArrayId",r,t,e),o=I("indices",r,t,e),s=I("dtype",r,t,e);return[e.getTensorArray(n.id).gather(o,s)]}case"TensorArrayScatterV3":{let n=I("tensorArrayId",r,t,e),o=I("indices",r,t,e),s=I("tensor",r,t,e),i=e.getTensorArray(n.id);return i.scatter(o,s),[i.idTensor]}case"TensorArrayConcatV3":{let n=I("tensorArrayId",r,t,e),o=e.getTensorArray(n.id),s=I("dtype",r,t,e);return[o.concat(s)]}case"TensorArraySplitV3":{let n=I("tensorArrayId",r,t,e),o=I("tensor",r,t,e),s=I("lengths",r,t,e),i=e.getTensorArray(n.id);return i.split(s,o),[i.idTensor]}case"TensorArraySizeV3":{let n=I("tensorArrayId",r,t,e),o=e.getTensorArray(n.id);return[pt(o.size(),"int32")]}case"TensorArrayCloseV3":{let n=I("tensorArrayId",r,t,e),o=e.getTensorArray(n.id);return o.clearAndClose(),[o.idTensor]}case"TensorListSetItem":{let n=I("tensorListId",r,t,e),o=I("index",r,t,e),s=I("tensor",r,t,e),i=e.getTensorList(n.id);return i.setItem(o,s),[i.idTensor]}case"TensorListGetItem":{let n=I("tensorListId",r,t,e),o=I("index",r,t,e),s=I("elementShape",r,t,e),i=I("elementDType",r,t,e);return[e.getTensorList(n.id).getItem(o,s,i)]}case"TensorListScatterV2":case"TensorListScatter":{let n=I("indices",r,t,e),o=I("tensor",r,t,e),s=I("elementShape",r,t,e),i=I("numElements",r,t,e),a=r$(o,n,s,i);return e.addTensorList(a),[a.idTensor]}case"TensorListReserve":case"EmptyTensorList":{let n=I("elementShape",r,t,e),o=I("elementDType",r,t,e),s;r.op==="TensorListReserve"?s="numElements":s="maxNumElements";let i=I(s,r,t,e),a=e$(n,o,i);return e.addTensorList(a),[a.idTensor]}case"TensorListGather":{let n=I("tensorListId",r,t,e),o=I("indices",r,t,e),s=I("elementShape",r,t,e),i=I("elementDType",r,t,e);return[e.getTensorList(n.id).gather(o,i,s)]}case"TensorListStack":{let n=I("tensorListId",r,t,e),o=I("elementShape",r,t,e),s=I("elementDType",r,t,e),i=I("numElements",r,t,e);return[e.getTensorList(n.id).stack(o,s,i)]}case"TensorListFromTensor":{let n=I("tensor",r,t,e),o=I("elementShape",r,t,e),s=I("elementDType",r,t,e),i=t$(n,o,s);return e.addTensorList(i),[i.idTensor]}case"TensorListConcat":{let n=I("tensorListId",r,t,e),o=e.getTensorList(n.id),s=I("dtype",r,t,e),i=I("elementShape",r,t,e);return[o.concat(s,i)]}case"TensorListPushBack":{let n=I("tensorListId",r,t,e),o=I("tensor",r,t,e),s=e.getTensorList(n.id);return s.pushBack(o),[s.idTensor]}case"TensorListPopBack":{let n=I("tensorListId",r,t,e),o=I("elementShape",r,t,e),s=I("elementDType",r,t,e);return[e.getTensorList(n.id).popBack(o,s)]}case"TensorListSplit":{let n=I("tensor",r,t,e),o=I("elementShape",r,t,e),s=I("lengths",r,t,e),i=n$(n,s,o);return e.addTensorList(i),[i.idTensor]}default:throw TypeError(`Node type ${r.op} is not implemented`)}};function s$(r,t,e){let[n,o]=I("fusedOps",r,t,e),s=n==="biasadd",i=!s,a=o==="prelu",u=n==="fusedbatchnorm",l=I("numArgs",r,t,e);if(s){if(a&&l!==2)throw new Error("FusedConv2d and DepthwiseConv2d with BiasAdd and Prelu must have two extra arguments: bias and alpha.");if(!a&&s&&l!==1)throw new Error("FusedConv2d and DepthwiseConv2d with BiasAdd must have one extra argument: bias.")}if(u)throw new Error("FusedConv2d and DepthwiseConv2d with FusedBatchNorm is not supported");let c=I("strides",r,t,e),p=Mh(r,t,e),m=I("dataFormat",r,t,e).toUpperCase(),f=I("dilations",r,t,e),[d,h]=I("args",r,t,e);i&&(h=d,d=void 0);let g=I("leakyreluAlpha",r,t,e);return{stride:c,pad:p,dataFormat:m,dilations:f,biasArg:d,preluArg:h,activationFunc:o,leakyreluAlpha:g}}var i$=(r,t,e)=>{switch(r.op){case"Conv1D":{let n=I("stride",r,t,e),o=I("pad",r,t,e),s=I("dataFormat",r,t,e).toUpperCase(),i=I("dilation",r,t,e);return[Pu(I("x",r,t,e),I("filter",r,t,e),n,o,s,i)]}case"Conv2D":{let n=I("strides",r,t,e),o=Mh(r,t,e),s=I("dataFormat",r,t,e).toUpperCase(),i=I("dilations",r,t,e);return[ln(I("x",r,t,e),I("filter",r,t,e),[n[1],n[2]],o,s,[i[1],i[2]])]}case"_FusedConv2D":{let{stride:n,pad:o,dataFormat:s,dilations:i,biasArg:a,preluArg:u,activationFunc:l,leakyreluAlpha:c}=s$(r,t,e);return[wo.conv2d({x:I("x",r,t,e),filter:I("filter",r,t,e),strides:[n[1],n[2]],pad:o,dataFormat:s,dilations:[i[1],i[2]],bias:a,activation:l,preluActivationWeights:u,leakyreluAlpha:c})]}case"FusedDepthwiseConv2dNative":{let{stride:n,pad:o,dataFormat:s,dilations:i,biasArg:a,preluArg:u,activationFunc:l,leakyreluAlpha:c}=s$(r,t,e);return[wo.depthwiseConv2d({x:I("x",r,t,e),filter:I("filter",r,t,e),strides:[n[1],n[2]],pad:o,dataFormat:s,dilations:[i[1],i[2]],bias:a,activation:l,preluActivationWeights:u,leakyreluAlpha:c})]}case"Conv2DBackpropInput":case"Conv2dTranspose":{let n=I("outputShape",r,t,e),o=I("strides",r,t,e),s=Mh(r,t,e);return[Mu(I("x",r,t,e),I("filter",r,t,e),n,[o[1],o[2]],s)]}case"DepthwiseConv2dNative":case"DepthwiseConv2d":{let n=I("strides",r,t,e),o=Mh(r,t,e),s=I("dilations",r,t,e),i=I("dataFormat",r,t,e).toUpperCase();return[Ks(I("input",r,t,e),I("filter",r,t,e),[n[1],n[2]],o,i,[s[1],s[2]])]}case"Conv3D":{let n=I("strides",r,t,e),o=I("pad",r,t,e),s=I("dataFormat",r,t,e).toUpperCase(),i=I("dilations",r,t,e);return[Gf(I("x",r,t,e),I("filter",r,t,e),[n[1],n[2],n[3]],o,s,[i[1],i[2],i[3]])]}case"AvgPool":{let n=I("strides",r,t,e),o=I("pad",r,t,e),s=I("kernelSize",r,t,e);return[Va(I("x",r,t,e),[s[1],s[2]],[n[1],n[2]],o)]}case"MaxPool":{let n=I("strides",r,t,e),o=I("pad",r,t,e),s=I("kernelSize",r,t,e);return[ja(I("x",r,t,e),[s[1],s[2]],[n[1],n[2]],o)]}case"MaxPoolWithArgmax":{let n=I("strides",r,t,e),o=I("pad",r,t,e),s=I("kernelSize",r,t,e),i=I("includeBatchInIndex",r,t,e),{result:a,indexes:u}=G0(I("x",r,t,e),[s[1],s[2]],[n[1],n[2]],o,i);return[a,u]}case"AvgPool3D":{let n=I("strides",r,t,e),o=I("pad",r,t,e),s=I("kernelSize",r,t,e);return[zf(I("x",r,t,e),[s[1],s[2],s[3]],[n[1],n[2],n[3]],o)]}case"MaxPool3D":{let n=I("strides",r,t,e),o=I("pad",r,t,e),s=I("kernelSize",r,t,e);return[Qf(I("x",r,t,e),[s[1],s[2],s[3]],[n[1],n[2],n[3]],o)]}case"Dilation2D":{let n=I("strides",r,t,e),o=I("pad",r,t,e),s=I("dilations",r,t,e),i=n[1],a=n[2],u=s[1],l=s[2];return[Hf(I("x",r,t,e),I("filter",r,t,e),[i,a],o,[u,l],"NHWC")]}default:throw TypeError(`Node type ${r.op} is not implemented`)}};var a$=(r,t,e)=>{switch(r.op){case"Fill":{let n=I("shape",r,t,e),o=I("dtype",r,t,e),s=I("value",r,t,e);return[Xs(n,s,o)]}case"LinSpace":{let n=I("start",r,t,e),o=I("stop",r,t,e),s=I("num",r,t,e);return[O0(n,o,s)]}case"Multinomial":{let n=I("logits",r,t,e),o=I("numSamples",r,t,e),s=I("seed",r,t,e);return[W0(n,o,s)]}case"OneHot":{let n=I("indices",r,t,e),o=I("depth",r,t,e),s=I("onValue",r,t,e),i=I("offValue",r,t,e);return[Us(n,o,s,i)]}case"Ones":return[lr(I("shape",r,t,e),I("dtype",r,t,e))];case"OnesLike":return[dr(I("x",r,t,e))];case"RandomUniform":return[Js(I("shape",r,t,e),I("minval",r,t,e),I("maxval",r,t,e),I("dtype",r,t,e))];case"Range":{let n=I("start",r,t,e),o=I("stop",r,t,e),s=I("step",r,t,e);return[Za(n,o,s,I("dtype",r,t,e))]}case"TruncatedNormal":{let n=I("shape",r,t,e),o=I("mean",r,t,e),s=I("stdDev",r,t,e),i=I("seed",r,t,e);return[tc(n,o,s,I("dtype",r,t,e),i)]}case"Zeros":return[ye(I("shape",r,t,e),I("dtype",r,t,e))];case"ZerosLike":return[St(I("x",r,t,e))];default:throw TypeError(`Node type ${r.op} is not implemented`)}};function FI(r,t,e){let n=I("boxes",r,t,e),o=I("scores",r,t,e),s=I("maxOutputSize",r,t,e),i=I("iouThreshold",r,t,e),a=I("scoreThreshold",r,t,e),u=I("softNmsSigma",r,t,e);return{boxes:n,scores:o,maxOutputSize:s,iouThreshold:i,scoreThreshold:a,softNmsSigma:u}}var l$=async(r,t,e)=>{switch(r.op){case"NonMaxSuppressionV5":{let{boxes:n,scores:o,maxOutputSize:s,iouThreshold:i,scoreThreshold:a,softNmsSigma:u}=FI(r,t,e),l=await pn.nonMaxSuppressionWithScoreAsync(n,o,s,i,a,u);return[l.selectedIndices,l.selectedScores]}case"NonMaxSuppressionV4":{let{boxes:n,scores:o,maxOutputSize:s,iouThreshold:i,scoreThreshold:a}=FI(r,t,e),u=I("padToMaxOutputSize",r,t,e),l=await pn.nonMaxSuppressionPaddedAsync(n,o,s,i,a,u);return[l.selectedIndices,l.validOutputs]}case"NonMaxSuppressionV3":case"NonMaxSuppressionV2":{let{boxes:n,scores:o,maxOutputSize:s,iouThreshold:i,scoreThreshold:a}=FI(r,t,e);return[await pn.nonMaxSuppressionAsync(n,o,s,i,a)]}case"Where":{let n=Z(I("condition",r,t,e),"bool"),o=[await pd(n)];return n.dispose(),o}case"ListDiff":return eC(I("x",r,t,e),I("y",r,t,e));default:throw TypeError(`Node type ${r.op} is not implemented`)}};var u$=(r,t,e)=>{switch(r.op){case"TopKV2":{let n=I("x",r,t,e),o=I("k",r,t,e),s=I("sorted",r,t,e),i=ud(n,o,s);return[i.values,i.indices]}case"Unique":{let n=I("x",r,t,e),o=rm(n);return[o.values,o.indices]}case"UniqueV2":{let n=I("x",r,t,e),o=I("axis",r,t,e),s=rm(n,o);return[s.values,s.indices]}default:throw TypeError(`Node type ${r.op} is not implemented`)}};var c$=(r,t,e)=>{switch(r.op){case"Const":return t[r.name];case"PlaceholderWithDefault":let n=I("default",r,t,e);return[Cr(r.name,t,e)||n];case"Placeholder":return[Cr(r.name,t,e)];case"Identity":case"StopGradient":case"FakeQuantWithMinMaxVars":{let l=I("x",r,t,e);return[ai(l)]}case"IdentityN":return I("x",r,t,e).map(l=>ai(l));case"Snapshot":let o=I("x",r,t,e);return[ai(o)];case"Shape":return[Re(I("x",r,t,e).shape,"int32")];case"ShapeN":return I("x",r,t,e).map(l=>Re(l.shape));case"Size":return[pt(I("x",r,t,e).size,"int32")];case"Rank":return[pt(I("x",r,t,e).rank,"int32")];case"NoOp":return[pt(1)];case"Print":let s=I("x",r,t,e),i=I("data",r,t,e),a=I("message",r,t,e),u=I("summarize",r,t,e);console.warn("The graph has a tf.print() operation,usually used for debugging, which slows down performance."),console.log(a);for(let l=0;l<i.length;l++)console.log(Array.prototype.slice.call(i[l].dataSync()).slice(0,u));return[s];default:throw TypeError(`Node type ${r.op} is not implemented`)}};var OI=class{constructor(t,e){this.keyDType=t,this.valueDType=e,this.handle=pt(0),this.tensorMap=new Map,Me(this.handle)}get id(){return this.handle.id}clearAndClose(){this.tensorMap.forEach(t=>t.dispose()),this.tensorMap.clear(),this.handle.dispose()}size(){return this.tensorMap.size}tensorSize(){return pt(this.size(),"int32")}async import(t,e){this.checkKeyAndValueTensor(t,e);let n=await t.data();return this.tensorMap.forEach(o=>o.dispose()),this.tensorMap.clear(),B(()=>{let o=vr(e),s=n.length,i=o.length;x.assert(s===i,()=>`The number of elements doesn't match, keys has ${s} elements, the values has ${i} elements.`);for(let a=0;a<s;a++){let u=n[a],l=o[a];Me(l),this.tensorMap.set(u,l)}return this.handle})}async find(t,e){this.checkKeyAndValueTensor(t,e);let n=await t.data();return B(()=>{let o=[];for(let s=0;s<n.length;s++){let i=n[s],a=this.findWithDefault(i,e);o.push(a)}return Je(o)})}findWithDefault(t,e){let n=this.tensorMap.get(t);return n!=null?n:e}checkKeyAndValueTensor(t,e){if(t.dtype!==this.keyDType)throw new Error(`Expect key dtype ${this.keyDType}, but got ${t.dtype}`);if(e.dtype!==this.valueDType)throw new Error(`Expect value dtype ${this.valueDType}, but got ${e.dtype}`)}};var p$=async(r,t,e,n)=>{switch(r.op){case"HashTable":case"HashTableV2":{let o=I("keyDType",r,t,e),s=I("valueDType",r,t,e),i=new OI(o,s);return n.addHashTable(r.name,i),[i.handle]}case"LookupTableImport":case"LookupTableImportV2":{let o=I("tableHandle",r,t,e,n),s=I("keys",r,t,e),i=I("values",r,t,e);return[await n.getHashTableById(o.id).import(s,i)]}case"LookupTableFind":case"LookupTableFindV2":{let o=I("tableHandle",r,t,e,n),s=I("keys",r,t,e),i=I("defaultValue",r,t,e);return[await n.getHashTableById(o.id).find(s,i)]}case"LookupTableSize":case"LookupTableSizeV2":{let o=I("tableHandle",r,t,e,n);return[n.getHashTableById(o.id).tensorSize()]}default:throw TypeError(`Node type ${r.op} is not implemented`)}};var m$=(r,t,e)=>{switch(r.op){case"ResizeBilinear":{let n=I("images",r,t,e),o=I("size",r,t,e),s=I("alignCorners",r,t,e),i=I("halfPixelCenters",r,t,e);return[pn.resizeBilinear(n,[o[0],o[1]],s,i)]}case"ResizeNearestNeighbor":{let n=I("images",r,t,e),o=I("size",r,t,e),s=I("alignCorners",r,t,e),i=I("halfPixelCenters",r,t,e);return[pn.resizeNearestNeighbor(n,[o[0],o[1]],s,i)]}case"CropAndResize":{let n=I("image",r,t,e),o=I("boxes",r,t,e),s=I("boxInd",r,t,e),i=I("cropSize",r,t,e),a=I("method",r,t,e),u=I("extrapolationValue",r,t,e);return[pn.cropAndResize(n,o,s,i,a,u)]}case"ImageProjectiveTransformV3":{let n=I("images",r,t,e),o=I("transforms",r,t,e),s=I("outputShape",r,t,e),i=I("fillValue",r,t,e),a=I("interpolation",r,t,e),u=I("fillMode",r,t,e);return[pn.transform(n,o,a.toLowerCase(),u.toLowerCase(),i,s)]}default:throw TypeError(`Node type ${r.op} is not implemented`)}};var f$=(r,t,e)=>{switch(r.op){case"Equal":return[Sr(I("a",r,t,e),I("b",r,t,e))];case"NotEqual":return[bo(I("a",r,t,e),I("b",r,t,e))];case"Greater":return[Ge(I("a",r,t,e),I("b",r,t,e))];case"GreaterEqual":return[kn(I("a",r,t,e),I("b",r,t,e))];case"Less":return[Vu(I("a",r,t,e),I("b",r,t,e))];case"LessEqual":return[Nn(I("a",r,t,e),I("b",r,t,e))];case"LogicalAnd":return[Nr(I("a",r,t,e),I("b",r,t,e))];case"LogicalNot":return[Ka(I("a",r,t,e))];case"LogicalOr":return[Wu(I("a",r,t,e),I("b",r,t,e))];case"Select":case"SelectV2":return[Ne(I("condition",r,t,e),I("a",r,t,e),I("b",r,t,e))];default:throw TypeError(`Node type ${r.op} is not implemented`)}};var d$=(r,t,e)=>{switch(r.op){case"BatchMatMul":case"BatchMatMulV2":case"MatMul":return[Mt(I("a",r,t,e),I("b",r,t,e),I("transposeA",r,t,e),I("transposeB",r,t,e))];case"Einsum":return[$0(I("equation",r,t,e),...I("tensors",r,t,e))];case"Transpose":return[Gt(I("x",r,t,e),I("perm",r,t,e))];case"_FusedMatMul":let[n,o]=I("fusedOps",r,t,e),s=n==="biasadd",i=o==="prelu",a=I("numArgs",r,t,e),u=I("leakyreluAlpha",r,t,e);if(s){if(i&&a!==2)throw new Error("Fused MatMul with BiasAdd and Prelu must have two extra arguments: bias and alpha.");if(!i&&a!==1)throw new Error("Fused MatMul with BiasAdd must have one extra argument: bias.")}let[l,c]=I("args",r,t,e);return[wo.matMul({a:I("a",r,t,e),b:I("b",r,t,e),transposeA:I("transposeA",r,t,e),transposeB:I("transposeB",r,t,e),bias:l,activation:o,preluActivationWeights:c,leakyreluAlpha:u})];default:throw TypeError(`Node type ${r.op} is not implemented`)}};var h$=(r,t,e)=>{switch(r.op){case"FusedBatchNorm":case"FusedBatchNormV2":return[ho(I("x",r,t,e),I("mean",r,t,e),I("variance",r,t,e),I("offset",r,t,e),I("scale",r,t,e),I("epsilon",r,t,e))];case"FusedBatchNormV3":return[ho(I("x",r,t,e),I("mean",r,t,e),I("variance",r,t,e),I("offset",r,t,e),I("scale",r,t,e),I("epsilon",r,t,e))];case"LRN":return[Yf(I("x",r,t,e),I("radius",r,t,e),I("bias",r,t,e),I("alpha",r,t,e),I("beta",r,t,e))];case"Softmax":return[Ja(I("x",r,t,e))];case"LogSoftmax":return[Gu(I("x",r,t,e))];case"SparseToDense":return[_x(I("sparseIndices",r,t,e),I("outputShape",r,t,e),I("sparseValues",r,t,e),I("defaultValue",r,t,e))];default:throw TypeError(`Node type ${r.op} is not implemented`)}};var g$=(r,t,e)=>{switch(r.op){case"Max":{let i=I("axis",r,t,e),a=I("keepDims",r,t,e);return[$r(I("x",r,t,e),i,a)]}case"Mean":{let i=I("axis",r,t,e),a=I("keepDims",r,t,e);return[xe(I("x",r,t,e),i,a)]}case"Min":{let i=I("axis",r,t,e),a=I("keepDims",r,t,e);return[Wl(I("x",r,t,e),i,a)]}case"Sum":{let i=I("axis",r,t,e),a=I("keepDims",r,t,e);return[mt(I("x",r,t,e),i,a)]}case"All":{let i=I("axis",r,t,e),a=I("keepDims",r,t,e);return[Ru(I("x",r,t,e),i,a)]}case"Any":{let i=I("axis",r,t,e),a=I("keepDims",r,t,e);return[Gl(I("x",r,t,e),i,a)]}case"ArgMax":{let i=I("axis",r,t,e);return[Hs(I("x",r,t,e),i)]}case"ArgMin":{let i=I("axis",r,t,e);return[Rf(I("x",r,t,e),i)]}case"Prod":{let i=I("axis",r,t,e),a=I("keepDims",r,t,e);return[Uu(I("x",r,t,e),i,a)]}case"Cumprod":{let i=I("axis",r,t,e),a=I("exclusive",r,t,e),u=I("reverse",r,t,e);return[Wf(I("x",r,t,e),i,a,u)]}case"Cumsum":{let i=I("axis",r,t,e),a=I("exclusive",r,t,e),u=I("reverse",r,t,e);return[zu(I("x",r,t,e),i,a,u)]}case"Bincount":let n=I("x",r,t,e),o=I("weights",r,t,e),s=I("size",r,t,e);return[Bf(n,o,s)];case"DenseBincount":{let i=I("x",r,t,e),a=I("weights",r,t,e),u=I("size",r,t,e),l=I("binaryOutput",r,t,e);return[A0(i,a,u,l)]}default:throw TypeError(`Node type ${r.op} is not implemented`)}};var x$=(r,t,e)=>{switch(r.op){case"ConcatV2":case"Concat":{let n=I("n",r,t,e),o=I("axis",r,t,e),s=I("tensors",r,t,e);return s=s.slice(0,n),[se(s,o)]}case"Gather":{let n=I("x",r,t,e),o=I("indices",r,t,e);return[go(n,Z(o,"int32"),0)]}case"GatherV2":{let n=I("axis",r,t,e),o=I("batchDims",r,t,e),s=I("x",r,t,e),i=I("indices",r,t,e);return[go(s,Z(i,"int32"),n,o)]}case"Reverse":{let n=I("dims",r,t,e),o=[];for(let i=0;i<n.length;i++)n[i]&&o.push(i);let s=I("x",r,t,e);return[sr(s,o)]}case"ReverseV2":{let n=I("axis",r,t,e),o=I("x",r,t,e);return[sr(o,n)]}case"Slice":{let n=I("begin",r,t,e),o=I("size",r,t,e);return[Rt(I("x",r,t,e),n,o)]}case"StridedSlice":{let n=I("begin",r,t,e),o=I("end",r,t,e),s=I("strides",r,t,e),i=I("beginMask",r,t,e),a=I("endMask",r,t,e),u=I("ellipsisMask",r,t,e),l=I("newAxisMask",r,t,e),c=I("shrinkAxisMask",r,t,e),p=I("x",r,t,e);return[ad(p,n,o,s,i,a,u,l,c)]}case"Pack":return B(()=>{let n=I("axis",r,t,e),o=I("tensors",r,t,e),s=o[0].shape,i=zr(o[0]).shape,a=o.map(u=>{let l=x.arraysEqual(u.shape,s);if(!l&&!x.arraysEqual(zr(u).shape,i))throw new Error("the input tensors shape does not match");return l?u:O(u,s)});return[Je(a,n)]});case"Unpack":{let n=I("axis",r,t,e),o=I("tensor",r,t,e);return vr(o,n)}case"Tile":{let n=I("reps",r,t,e);return[kr(I("x",r,t,e),n)]}case"Split":case"SplitV":{let n=I("axis",r,t,e),o=I("numOrSizeSplits",r,t,e),s=I("x",r,t,e);return ur(s,o,n)}case"ScatterNd":{let n=I("indices",r,t,e),o=I("values",r,t,e),s=I("shape",r,t,e);return[H_(n,o,s)]}case"GatherNd":{let n=I("x",r,t,e),o=I("indices",r,t,e);return[K_(n,o)]}case"SparseToDense":{let n=I("sparseIndices",r,t,e),o=I("outputShape",r,t,e),s=I("sparseValues",r,t,e),i=I("defaultValue",r,t,e);return[_x(n,s,o,s.dtype===i.dtype?i:Z(i,s.dtype))]}default:throw TypeError(`Node type ${r.op} is not implemented`)}};var y$=(r,t,e)=>{switch(r.op){case"SparseFillEmptyRows":{let{outputIndices:n,outputValues:o,emptyRowIndicator:s,reverseIndexMap:i}=md.sparseFillEmptyRows(I("indices",r,t,e),I("values",r,t,e),I("denseShape",r,t,e),I("defaultValue",r,t,e));return[n,o,s,i]}case"SparseReshape":{let{outputIndices:n,outputShape:o}=md.sparseReshape(I("inputIndices",r,t,e),I("inputShape",r,t,e),I("newShape",r,t,e));return[n,o]}case"SparseSegmentMean":return[md.sparseSegmentMean(I("data",r,t,e),I("indices",r,t,e),I("segmentIds",r,t,e))];case"SparseSegmentSum":return[md.sparseSegmentSum(I("data",r,t,e),I("indices",r,t,e),I("segmentIds",r,t,e))];default:throw TypeError(`Node type ${r.op} is not implemented`)}};var b$=(r,t,e)=>{switch(r.op){case"FFT":return[Qa(I("x",r,t,e))];case"IFFT":return[Oi(I("x",r,t,e))];case"RFFT":return[tl(I("x",r,t,e))];case"IRFFT":return[Ju(I("x",r,t,e))];default:throw TypeError(`Node type ${r.op} is not implemented`)}};var w$=(r,t,e)=>{switch(r.op){case"StringNGrams":{let{nGrams:n,nGramsSplits:o}=zx.stringNGrams(I("data",r,t,e),I("dataSplits",r,t,e),I("separator",r,t,e),I("nGramWidths",r,t,e),I("leftPad",r,t,e),I("rightPad",r,t,e),I("padWidth",r,t,e),I("preserveShortSequences",r,t,e));return[n,o]}case"StringSplit":{let{indices:n,values:o,shape:s}=zx.stringSplit(I("input",r,t,e),I("delimiter",r,t,e),I("skipEmpty",r,t,e));return[n,o,s]}case"StringToHashBucketFast":return[zx.stringToHashBucketFast(I("input",r,t,e),I("numBuckets",r,t,e))];default:throw TypeError(`Node type ${r.op} is not implemented`)}};var v$=(r,t,e)=>{switch(r.op){case"Cast":return[Z(I("x",r,t,e),I("dtype",r,t,e))];case"ExpandDims":{let n=I("axis",r,t,e);return[fr(I("x",r,t,e),n)]}case"Squeeze":{let n=I("axis",r,t,e);return[zr(I("x",r,t,e),n)]}case"Reshape":return[O(I("x",r,t,e),I("shape",r,t,e))];case"MirrorPad":return[td(I("x",r,t,e),I("padding",r,t,e),I("mode",r,t,e))];case"PadV2":case"Pad":return[qr(I("x",r,t,e),I("padding",r,t,e),I("constantValue",r,t,e))];case"SpaceToBatchND":{let n=I("blockShape",r,t,e),o=I("paddings",r,t,e);return[Xa(I("x",r,t,e),n,o)]}case"BatchToSpaceND":{let n=I("blockShape",r,t,e),o=I("crops",r,t,e);return[Ga(I("x",r,t,e),n,o)]}case"DepthToSpace":{let n=I("blockSize",r,t,e),o=I("dataFormat",r,t,e).toUpperCase();return[Uf(I("x",r,t,e),n,o)]}case"BroadcastTo":return[Wa(I("x",r,t,e),I("shape",r,t,e))];case"BroadcastArgs":return[S0(I("s0",r,t,e),I("s1",r,t,e))];default:throw TypeError(`Node type ${r.op} is not implemented`)}};function PI(r,t,e,n){let o=((s,i,a)=>{switch(s.category){case"arithmetic":return B(()=>ZD(s,i,a));case"basic_math":return B(()=>JD(s,i,a));case"control":return o$(s,i,a);case"convolution":return B(()=>i$(s,i,a));case"creation":return B(()=>a$(s,i,a));case"dynamic":return l$(s,i,a);case"evaluation":return B(()=>u$(s,i,a));case"image":return B(()=>m$(s,i,a));case"graph":return B(()=>c$(s,i,a));case"logical":return B(()=>f$(s,i,a));case"matrices":return B(()=>d$(s,i,a));case"normalization":return B(()=>h$(s,i,a));case"reduction":return B(()=>g$(s,i,a));case"slice_join":return B(()=>x$(s,i,a));case"sparse":return B(()=>y$(s,i,a));case"spectral":return B(()=>b$(s,i,a));case"string":return B(()=>w$(s,i,a));case"transformation":return B(()=>v$(s,i,a));case"hash_table":return p$(s,i,a,n);case"custom":let u=yy(s.op);if(u&&u.customExecutor)return u.customExecutor(new $I(s,i,a));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()`)}})(r,t,e);return x.isPromise(o)?o.then(s=>[].concat(s)):[].concat(o)}var $y=class{constructor(t={},e={},n={},o={}){this.weightMap=t,this.tensorArrayMap=e,this.tensorListMap=n,this.functionMap=o,this.rootContext={id:0,frameName:"",iterationId:0},this.contexts=[this.rootContext],this.lastId=0,this.generateCurrentContextIds()}newFrame(t,e){return{id:t,frameName:e,iterationId:0}}set currentContext(t){this.contexts!==t&&(this.contexts=t,this.generateCurrentContextIds())}get currentContext(){return this.contexts}get currentContextId(){return this._currentContextIds[0]}get currentContextIds(){return this._currentContextIds}generateCurrentContextIds(){let t=[];for(let e=0;e<this.contexts.length-1;e++){let n=this.contexts.slice(0,this.contexts.length-e);t.push(this.contextIdforContexts(n))}t.push(""),this._currentContextIds=t}contextIdforContexts(t){return t?t.map(e=>e.id===0&&e.iterationId===0?"":`${e.frameName}-${e.iterationId}`).join("/"):""}enterFrame(t){this.contexts&&(this.lastId++,this.contexts=this.contexts.slice(),this.contexts.push(this.newFrame(this.lastId,t)),this._currentContextIds.unshift(this.contextIdforContexts(this.contexts)))}exitFrame(){if(this.contexts&&this.contexts.length>1)this.contexts=this.contexts.slice(),this.contexts.splice(-1),this.currentContextIds.shift();else throw new Error("Cannot exit frame, the context is empty")}nextIteration(){if(this.contexts&&this.contexts.length>0){this.contexts=this.contexts.slice(),this.lastId++;let t=Object.assign({},this.contexts[this.contexts.length-1]);t.iterationId+=1,t.id=this.lastId,this.contexts.splice(-1,1,t),this._currentContextIds.splice(0,1,this.contextIdforContexts(this.contexts))}else throw new Error("Cannot increase frame iteration, the context is empty")}getWeight(t){return this.weightMap[t]}addTensorArray(t){this.tensorArrayMap[t.id]=t}getTensorArray(t){return this.tensorArrayMap[t]}addTensorList(t){this.tensorListMap[t.id]=t}getTensorList(t){return this.tensorListMap[t]}dispose(t){for(let e in this.tensorArrayMap)this.tensorArrayMap[e].clearAndClose(t);for(let e in this.tensorListMap)this.tensorListMap[e].clearAndClose(t)}};function MI(r,t,e,n){let o=new Set,s=[],i=null,a=null,u=new Set,l=Object.keys(r).map(m=>bn(m)[0]),c=[];n!=null&&(c=n.map(m=>bn(m.name)[0]));let p=[...t];for(;p.length>0;){let m=p.pop();if((LI(m)||rZ(m)||nZ(m))&&i==null&&(i=m,a=i.children.map(f=>f.name).filter(f=>o.has(f))),o.add(m.name),e[m.name]==null&&l.indexOf(m.name)===-1&&c.indexOf(m.name)===-1){if(m.inputs.length===0){s.push(m.name);continue}m.inputs.forEach(f=>{u.has(f.name)||(u.add(f.name),p.push(f))})}}return{inputs:r,outputs:t,usedNodes:o,missingInputs:s,dynamicNode:i,syncInputs:a}}function C$(r,t,e){let{usedNodes:n,inputs:o}=e,s=[],i=Object.keys(o).map(c=>bn(c)[0]).map(c=>r.nodes[c]),a=r.initNodes;i.forEach(c=>{n.has(c.name)&&s.push(c)}),r.weights.forEach(c=>{n.has(c.name)&&s.push(c)}),a!=null&&a.forEach(c=>{n.has(c.name)&&s.push(c)});let u=new Set,l=[];for(;s.length>0;){let c=s.pop();u.add(c.name),t[c.name]||l.push(c),c.children.forEach(p=>{!u.has(p.name)&&n.has(p.name)&&p.inputs.every(m=>u.has(m.name))&&s.push(p)})}return l}var Q7=["Switch","Merge","Enter","Exit","NextIteration","StatelessIf","StatelessWhile","if","While"],tZ=["NonMaxSuppressionV2","NonMaxSuppressionV3","NonMaxSuppressionV5","Where"],eZ=["HashTable","HashTableV2","LookupTableImport","LookupTableImportV2","LookupTableFind","LookupTableFindV2","LookupTableSize","LookupTableSizeV2"];function LI(r){return Q7.indexOf(r.op)>=0}function rZ(r){return tZ.indexOf(r.op)>=0}function nZ(r){return eZ.indexOf(r.op)>=0}var Dm=class{constructor(t,e){this.graph=t,this.parent=e,this.compiledMap=new Map,this._weightMap={},this.SEPERATOR=",",this._functions={},this._functionExecutorMap={},this.intermediateTensors={},this.keepTensorForDebug=!1,this._outputs=t.outputs,this._inputs=t.inputs,this._initNodes=t.initNodes,this._signature=t.signature,this._functions=t.functions,t.functions!=null&&Object.keys(t.functions).forEach(n=>{this._functionExecutorMap[n]=new Dm(t.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(t){let e=Object.keys(t).map(n=>t[n].map(o=>o.id));this._weightIds=[].concat(...e),this._weightMap=t}set resourceManager(t){this._resourceManager=t}get inputs(){return this._inputs.map(t=>({name:t.name,shape:t.attrParams.shape?t.attrParams.shape.value:void 0,dtype:t.attrParams.dtype?t.attrParams.dtype.value:void 0}))}get outputs(){return this._outputs.map(t=>({name:t.name,shape:t.attrParams.shape?t.attrParams.shape.value:void 0,dtype:t.attrParams.dtype?t.attrParams.dtype.value:void 0}))}get inputNodes(){return this._inputs.map(t=>t.signatureKey||t.name)}get outputNodes(){return this._outputs.map(t=>{let e=t.signatureKey||t.name;return t.defaultOutput?`${e}:${t.defaultOutput}`:e})}get functions(){return Object.keys(this._functions).reduce((t,e)=>(t[e]=this._functions[e].signature,t),{})}getCompilationKey(t,e){let n=t.map(s=>s.name).sort(),o=e.map(s=>s.name).sort();return n.join(this.SEPERATOR)+"--"+o.join(this.SEPERATOR)}compile(t,e){let n=MI(t,e,this.weightMap,this._initNodes),{missingInputs:o,dynamicNode:s,syncInputs:i}=n;if(s!=null)throw new Error(`This execution contains the node '${s.name}', which has the dynamic op '${s.op}'. Please use model.executeAsync() instead. Alternatively, to avoid the dynamic ops, specify the inputs [${i}]`);if(o.length>0){let a=e.map(l=>l.name),u=Object.keys(t);throw new Error(`Cannot compute the outputs [${a}] from the provided inputs [${u}]. Missing the following inputs: [${o}]`)}return C$(this.graph,this.weightMap,n)}execute(t,e){t=this.mapInputs(t);let n=Object.keys(t).sort();this.checkInputs(t),this.checkInputShapeAndType(t),e=this.mapOutputs(e),this.checkOutputs(e);let o=n.map(p=>this.graph.nodes[bn(p)[0]]),s=e.map(p=>bn(p)[0]),i=s.map(p=>this.graph.nodes[p]);this.resetIntermediateTensors(),i.length===0&&(i=this._outputs);let a=this.getCompilationKey(o,i),u=this.compiledMap.get(a);u==null&&(u=this.compile(t,i),this.compiledMap.set(a,u));let l={},c={};return B(()=>{let p=new $y(this.weightMap,l,c,this.functionExecutorMap),m=Object.assign({},this.weightMap);Object.keys(t).forEach(h=>{let[g,y]=bn(h),b=[];b[y]=t[h],m[g]=b});let f=this.getFrozenTensorIds(m),d={};for(let h=0;h<u.length;h++){let g=u[h];if(!m[g.name]){let y=PI(g,m,p,this._resourceManager);if(x.isPromise(y))throw new Error(`The execution of the op '${g.op}' returned a promise. Please use model.executeAsync() instead.`);m[g.name]=y,this.checkTensorForDisposal(g.name,g,m,p,f,s,d)}}return this.parent==null&&p.dispose(f),e.map(h=>Cr(h,m,p))})}getFrozenTensorIds(t){let e=[].concat.apply([],Object.keys(t).map(n=>t[n]).map(n=>n.map(o=>o.id)));return new Set(e)}checkTensorForDisposal(t,e,n,o,s,i,a){e.category==="control"||i.indexOf(t)!==-1||(n[t].forEach(u=>{u!=null&&(a[u.id]=(a[u.id]||0)+e.children.length)}),e.inputs.forEach(u=>{if(u.category!=="control"){let l=KD(u.name,n,o);l!=null&&l.forEach(c=>{if(c&&!c.kept&&!s.has(c.id)){let p=a[c.id];if(p===1){if(!this.keepTensorForDebug)c.dispose();else{let[m,f]=Eo(e.name,o);this.intermediateTensors[m]?this.intermediateTensors[m][f]=c:(this.intermediateTensors[m]=[],this.intermediateTensors[m][f]=c)}delete a[c.id]}else p!=null&&a[c.id]--}})}}))}async executeAsync(t,e){return this._executeAsync(t,e)}disposeIntermediateTensors(){!this.intermediateTensors||(Object.keys(this.intermediateTensors).forEach(t=>this.intermediateTensors[t].forEach(e=>e.dispose())),this.disposeTensorsMap())}disposeTensorsMap(){!this.tensorsMap||Object.keys(this.tensorsMap).forEach(t=>{this.tensorsMap[t].forEach(n=>{n&&!n.kept&&!n.isDisposed&&!this.keepIds.has(n.id)&&n.dispose()})})}getIntermediateTensors(){return this.tensorsMap}resetIntermediateTensors(){for(let t in this.intermediateTensors)this.intermediateTensors[t].forEach(e=>e.dispose()),delete this.intermediateTensors[t]}async _executeAsync(t,e,n=!1,o={},s={}){n||(t=this.mapInputs(t),this.checkInputs(t),this.checkInputShapeAndType(t),e=this.mapOutputs(e),this.checkOutputs(e));try{this.keepTensorForDebug=V().getBool("KEEP_INTERMEDIATE_TENSORS")}catch(c){console.warn(c.message)}this.resetIntermediateTensors();let i=new $y(this.weightMap,o,s,this.functionExecutorMap);this.tensorsMap=await this.executeWithControlFlow(t,i,e,n);let a=e.map(c=>Cr(c,this.tensorsMap,i)),u=a.map(c=>c.id),l=Object.keys(t).map(c=>t[c].id);return this.keepIds=new Set([...u,...l,...this.weightIds]),this.keepTensorForDebug||this.disposeTensorsMap(),this.parent==null&&i.dispose(this.keepIds),a}async executeFunctionAsync(t,e,n){let o=t.reduce((s,i,a)=>(s[this.inputs[a].name]=i,s),{});return this._executeAsync(o,this.outputNodes,!0,e,n)}async executeWithControlFlow(t,e,n,o){let s=Object.keys(t),i=s.map(w=>this.graph.nodes[bn(w)[0]]),a=n.map(w=>bn(w)[0]),u=a.map(w=>this.graph.nodes[w]);u.length===0&&(u=this._outputs);let{usedNodes:l,missingInputs:c,dynamicNode:p,syncInputs:m}=MI(t,u,this.weightMap,this._initNodes),f=[...i,...this.graph.weights,...this._initNodes||[]].map(w=>({node:w,contexts:e.currentContext})),d=Object.assign({},this.weightMap);Object.keys(t).forEach(w=>{let[v,k]=bn(w),E=[];E[k]=t[w],d[v]=E});let h={},g=this.getFrozenTensorIds(d),y={};for(;f.length>0;){let w=this.processStack(i,f,e,d,y,g,a,h,l);await Promise.all(w)}p==null&&!o&&console.warn("This model execution did not contain any nodes with control flow or dynamic output shapes. You can use model.execute() instead.");let b=u.filter(w=>!LI(w)&&!Cr(w.name,d,e)).map(w=>w.name);if(b.length>0){let w="";throw p!=null&&(w=`Alternatively, to avoid the dynamic ops, use model.execute() and specify the inputs [${m}]`),new Error(`Cannot compute the outputs [${b}] from the provided inputs [${s}]. Consider providing the following inputs: [${c}]. ${w}`)}return d}processStack(t,e,n,o,s,i,a,u,l){let c=[];for(;e.length>0;){let p=e.pop();n.currentContext=p.contexts;let m="";if(p.node.op==="Enter"&&I("isConstant",p.node,o,n)&&([m]=Eo(p.node.name,n)),o[p.node.name]==null){let f=PI(p.node,o,n,this._resourceManager);m||([m]=Eo(p.node.name,n));let d=n.currentContext;x.isPromise(f)?c.push(f.then(h=>(o[m]=h,n.currentContext=d,this.checkTensorForDisposal(m,p.node,o,n,i,a,u),this.processChildNodes(p.node,e,n,o,s,l),h))):(o[m]=f,this.checkTensorForDisposal(m,p.node,o,n,i,a,u),this.processChildNodes(p.node,e,n,o,s,l))}else this.processChildNodes(p.node,e,n,o,s,l)}return c}processChildNodes(t,e,n,o,s,i){t.children.forEach(a=>{let[u]=Eo(a.name,n);s[u]||!i.has(a.name)||(a.op==="Merge"?a.inputNames.some(l=>!!Cr(l,o,n))&&(s[u]=!0,e.push({contexts:n.currentContext,node:a})):a.inputNames.every(l=>!!Cr(l,o,n))&&(s[u]=!0,e.push({contexts:n.currentContext,node:a})))})}dispose(){Object.keys(this.weightMap).forEach(t=>this.weightMap[t].forEach(e=>e.dispose()))}checkInputShapeAndType(t){Object.keys(t).forEach(e=>{let n=t[e],[o]=bn(e),s=this.graph.nodes[o];if(s.attrParams.shape&&s.attrParams.shape.value){let i=s.attrParams.shape.value,a=i.length===n.shape.length&&n.shape.every((u,l)=>i[l]===-1||i[l]===u);x.assert(a,()=>`The shape of dict['${s.name}'] provided in model.execute(dict) must be [${i}], but was [${n.shape}]`)}s.attrParams.dtype&&s.attrParams.dtype.value&&x.assert(n.dtype===s.attrParams.dtype.value,()=>`The dtype of dict['${s.name}'] provided in model.execute(dict) must be ${s.attrParams.dtype.value}, but was ${n.dtype}`)})}mapInputs(t){let e={};for(let n in t)if(this._signature!=null&&this._signature.inputs!=null&&this._signature.inputs[n]!=null){let o=this._signature.inputs[n];e[o.name]=t[n]}else e[n]=t[n];return e}checkInputs(t){let e=Object.keys(t).filter(n=>{let[o]=bn(n);return this.graph.nodes[o]==null});if(e.length>0)throw new Error(`The dict provided in model.execute(dict) has keys: [${e}] that are not part of graph`)}mapOutputs(t){return t.map(e=>this._signature!=null&&this._signature.outputs!=null&&this._signature.outputs[e]!=null?this._signature.outputs[e].name:e,{})}checkOutputs(t){t.forEach(e=>{let[n]=bn(e);if(!this.graph.nodes[n])throw new Error(`The output '${e}' is not found in the graph`)})}};var zI=class{constructor(t={},e={}){this.hashTableNameToHandle=t,this.hashTableMap=e}addHashTable(t,e){this.hashTableNameToHandle[t]=e.handle,this.hashTableMap[e.id]=e}getHashTableHandleByName(t){return this.hashTableNameToHandle[t]}getHashTableById(t){return this.hashTableMap[t]}dispose(){for(let t in this.hashTableMap)this.hashTableMap[t].clearAndClose(),delete this.hashTableMap[t];for(let t in this.hashTableNameToHandle)this.hashTableNameToHandle[t].dispose(),delete this.hashTableNameToHandle[t]}};var oZ="?tfjs-format=file",sZ="model.json",BI=class{constructor(t,e={}){this.modelUrl=t,this.loadOptions=e,this.version="n/a",e==null&&(this.loadOptions={}),this.resourceManager=new zI}get modelVersion(){return this.version}get inputNodes(){return this.executor.inputNodes}get outputNodes(){return this.executor.outputNodes}get inputs(){return this.executor.inputs}get outputs(){return this.executor.outputs}get weights(){return this.executor.weightMap}get metadata(){return this.artifacts.userDefinedMetadata}get modelSignature(){return this.signature}findIOHandler(){let t=this.modelUrl;if(t.load!=null)this.handler=t;else if(this.loadOptions.requestInit!=null)this.handler=Pr.browserHTTPRequest(t,this.loadOptions);else{let e=Pr.getLoadHandlers(t,this.loadOptions);if(e.length===0)e.push(Pr.browserHTTPRequest(t,this.loadOptions));else if(e.length>1)throw new Error(`Found more than one (${e.length}) load handlers for URL '${[t]}'`);this.handler=e[0]}}async load(){if(this.findIOHandler(),this.handler.load==null)throw new Error("Cannot proceed with model loading because the IOHandler provided does not have the `load` method implemented.");let t=await this.handler.load();return this.loadSync(t)}loadSync(t){this.artifacts=t;let e=this.artifacts.modelTopology,n;this.artifacts.userDefinedMetadata!=null&&this.artifacts.userDefinedMetadata.signature!=null?n=this.artifacts.userDefinedMetadata.signature:n=this.artifacts.signature,this.signature=n,this.version=`${e.versions.producer}.${e.versions.minConsumer}`;let o=Pr.decodeWeights(this.artifacts.weightData,this.artifacts.weightSpecs);if(this.executor=new Dm(Ay.Instance.transformGraph(e,this.signature)),this.executor.weightMap=this.convertTensorMapToTensorsMap(o),this.executor.resourceManager=this.resourceManager,t.modelInitializer!=null&&t.modelInitializer.node!=null){let s=Ay.Instance.transformGraph(t.modelInitializer);this.initializer=new Dm(s),this.initializer.weightMap=this.executor.weightMap,this.initializer.resourceManager=this.resourceManager,this.initializer.executeAsync({},[])}return!0}async save(t,e){if(typeof t=="string"){let n=Pr.getSaveHandlers(t);if(n.length===0)throw new Error(`Cannot find any save handlers for URL '${t}'`);if(n.length>1)throw new Error(`Found more than one (${n.length}) save handlers for URL '${t}'`);t=n[0]}if(t.save==null)throw new Error("GraphModel.save() cannot proceed because the IOHandler provided does not have the `save` attribute defined.");return t.save(this.artifacts)}predict(t,e){return this.execute(t,this.outputNodes)}normalizeInputs(t){if(!(t instanceof Vt)&&!Array.isArray(t))return t;if(t=Array.isArray(t)?t:[t],t.length!==this.inputNodes.length)throw new Error(`Input tensor count mismatch,the graph model has ${this.inputNodes.length} placeholders, while there are ${t.length} input tensors.`);return this.inputNodes.reduce((e,n,o)=>(e[n]=t[o],e),{})}normalizeOutputs(t){return t=t||this.outputNodes,Array.isArray(t)?t:[t]}execute(t,e){t=this.normalizeInputs(t),e=this.normalizeOutputs(e);let n=this.executor.execute(t,e);return n.length>1?n:n[0]}async executeAsync(t,e){t=this.normalizeInputs(t),e=this.normalizeOutputs(e);let n=await this.executor.executeAsync(t,e);return n.length>1?n:n[0]}getIntermediateTensors(){return this.executor.getIntermediateTensors()}disposeIntermediateTensors(){this.executor.disposeIntermediateTensors()}convertTensorMapToTensorsMap(t){return Object.keys(t).reduce((e,n)=>(e[n]=[t[n]],e),{})}dispose(){this.executor.dispose(),this.initializer&&this.initializer.dispose(),this.resourceManager.dispose()}};async function iZ(r,t={}){if(r==null)throw new Error("modelUrl in loadGraphModel() cannot be null. Please provide a url or an IOHandler that loads the model");t==null&&(t={}),t.fromTFHub&&r.load==null&&(r.endsWith("/")||(r=r+"/"),r=`${r}${sZ}${oZ}`);let e=new BI(r,t);return await e.load(),e}var I$="3.15.0";var wR={};Yt(wR,{CSVDataset:()=>Gh,Dataset:()=>Wi,FileDataSource:()=>Kh,TextLineDataset:()=>Bh,URLDataSource:()=>jh,array:()=>iR,csv:()=>hR,func:()=>gR,generator:()=>xR,microphone:()=>bR,version_data:()=>rS,webcam:()=>yR,zip:()=>aR});var sR=xu(KI());var U$=xu(KI());function z$(r,t){return Fy(r,t)}function Fy(r,t,e=new Map,n=new Set){if(r==null)return null;if(typeof Blob=="function"&&r instanceof Blob)return r.slice();if(n.has(r))throw new Error("Circular references are not supported.");if(e.has(r))return e.get(r);let o=t(r);if(o.recurse&&o.value!==null)throw new Error("A deep map function may not return both a value and recurse=true.");if(o.recurse)if(ru(r)){let s=Array.isArray(r)?[]:{};n.add(r);for(let i in r){let a=r[i],u=Fy(a,t,e,n);s[i]=u}return n.delete(r),r.__proto__&&(s.__proto__=r.__proto__),s}else throw new Error(`Can't recurse into non-iterable type: ${r}`);else return e.set(r,o.value),o.value}function B$(r,t=XI){return V$(r,t)}function V$(r,t,e=new Set){let n=r[0];if(e.has(n))throw new Error("Circular references are not supported.");let o=t(r);if(o.recurse&&o.value!==null)throw new Error("A deep zip function may not return both a value and recurse=true.");if(o.recurse)if(ru(n)){let s=Array.isArray(n)?[]:{};e.add(n);for(let i in n){let a=r.map(l=>l[i]),u=V$(a,t,e);s[i]=u}return e.delete(n),s}else throw new Error(`Can't recurse into non-iterable type: ${n}`);else return o.value}function XI(r){return r===null?null:ru(r[0])?{value:null,recurse:!0}:{value:r,recurse:!1}}async function Oy(r,t){let e=new Map;Fy(r,t,e);for(let o of Array.from(e.keys())){let s=e.get(o);if(x.isPromise(s)){let i=await s;e.set(o,i)}}return Fy(r,t,e)}function ru(r){let t=!1;if(V().get("IS_BROWSER"))t=r instanceof TextDecoder;else{let{StringDecoder:e}=jI();t=r instanceof e}return r!=null&&!ArrayBuffer.isView(r)&&(Array.isArray(r)||typeof r=="object"&&!(r instanceof Vt)&&!(r instanceof Promise)&&!t)}function G$(r){return r==null||fZ(r)||Array.isArray(r)||typeof r=="object"&&r instanceof Vt||x.isTypedArray(r)}function fZ(r){return r===null||typeof r!="object"&&typeof r!="function"}function W$(r){return z$(r,dZ)}function dZ(r){return r instanceof Vt?{value:r.clone(),recurse:!1}:ru(r)?{value:null,recurse:!0}:{value:r,recurse:!1}}var Lh=class{constructor(t){if(this.capacity=t,this.begin=0,this.end=0,t==null)throw new RangeError("Can't create a ring buffer of unknown capacity.");if(t<1)throw new RangeError("Can't create ring buffer of capacity < 1.");this.data=new Array(t),this.doubledCapacity=2*t}wrap(t){for(;t<0;)t+=this.doubledCapacity;return t%this.doubledCapacity}get(t){if(t<0)throw new RangeError("Can't get item at a negative index.");return this.data[t%this.capacity]}set(t,e){if(t<0)throw new RangeError("Can't set item at a negative index.");this.data[t%this.capacity]=e}length(){let t=this.end-this.begin;return t<0&&(t=this.doubledCapacity+t),t}isFull(){return this.length()===this.capacity}isEmpty(){return this.length()===0}push(t){if(this.isFull())throw new RangeError("Ring buffer is full.");this.set(this.end,t),this.end=this.wrap(this.end+1)}pushAll(t){for(let e of t)this.push(e)}pop(){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");this.end=this.wrap(this.end-1);let t=this.get(this.end);return this.set(this.end,void 0),t}unshift(t){if(this.isFull())throw new RangeError("Ring buffer is full.");this.begin=this.wrap(this.begin-1),this.set(this.begin,t)}shift(){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");let t=this.get(this.begin);return this.set(this.begin,void 0),this.begin=this.wrap(this.begin+1),t}shuffleExcise(t){if(this.isEmpty())throw new RangeError("Ring buffer is empty.");let e=this.wrap(this.begin+t),n=this.get(e);return this.set(e,this.pop()),n}};var $m=class extends Lh{constructor(){super($m.INITIAL_CAPACITY)}isFull(){return!1}push(t){super.isFull()&&this.expand(),super.push(t)}unshift(t){super.isFull()&&this.expand(),super.unshift(t)}expand(){let t=this.capacity*2,e=new Array(t),n=this.length();for(let o=0;o<n;o++)e[o]=this.get(this.wrap(this.begin+o));this.data=e,this.capacity=t,this.doubledCapacity=2*this.capacity,this.begin=0,this.end=n}};$m.INITIAL_CAPACITY=32;function ZI(r){return new K$(r)}function zh(r){return new j$(r)}function H$(r,t){return new JI(r,t)}function q$(r,t=al.FAIL){return new nR(r,t)}var ir=class{async toArray(){let t=[],e=await this.next();for(;!e.done;)t.push(e.value),e=await this.next();return t}async toArrayForTest(){let t=this.prefetch(100),e=[],n=await t.next();for(;!n.done;)e.push(n.value),n=await t.next();return e}async resolveFully(){let t=await this.next();for(;!t.done;)t=await this.next()}async resolveWhile(t){let e=await this.next(),n=t(e.value);for(;!e.done&&n;)e=await this.next(),n=t(e.value)}handleErrors(t){return new eR(this,t)}filter(t){return new Q$(this,t)}map(t){return new tR(this,t)}mapAsync(t){return new YI(this,t)}serialMapAsync(t){return new YI(this,t).serial()}flatmap(t){return new rR(this,t)}async forEachAsync(t){return this.map(t).resolveFully()}async serialForEach(t){return this.serialMapAsync(t).resolveWhile(e=>e===!0)}rowMajorBatch(t,e=!0){return new J$(this,t,e)}columnMajorBatch(t,e=!0,n=XI){return this.rowMajorBatch(t,e).map(s=>B$(s,n))}concatenate(t,e){return new JI(ZI([this,t]),e)}take(t){return t<0||t==null?this:new Z$(this,t)}skip(t){return t<0||t==null?this:new Y$(this,t)}prefetch(t){return new QI(this,t)}shuffle(t,e){return new oR(this,t,e)}serial(){return new X$(this)}},K$=class extends ir{constructor(t){super();this.items=t,this.trav=0}summary(){return`Array of ${this.items.length} items`}async next(){if(this.trav>=this.items.length)return{value:null,done:!0};let t=this.items[this.trav];return this.trav++,{value:W$(t),done:!1}}},j$=class extends ir{constructor(t){super();this.nextFn=t}summary(){return"Function call"}async next(){try{return this.nextFn()}catch(t){throw t.message=`Error thrown while iterating through a dataset: ${t.message}`,t}}},X$=class extends ir{constructor(t){super();this.upstream=t,this.lastRead=Promise.resolve({value:null,done:!1})}summary(){return`${this.upstream.summary()} -> Serial`}async next(){return this.lastRead=this.lastRead.then(()=>this.serialNext()),this.lastRead}async serialNext(){return this.upstream.next()}},Y$=class extends ir{constructor(t,e){super();this.upstream=t,this.maxCount=e,this.count=0,this.lastRead=Promise.resolve({value:null,done:!1})}summary(){return`${this.upstream.summary()} -> Skip`}async next(){return this.lastRead=this.lastRead.then(()=>this.serialNext()),this.lastRead}async serialNext(){for(;this.count++<this.maxCount;){let t=await this.upstream.next();if(t.done)return t;_t(t.value)}return this.upstream.next()}},Z$=class extends ir{constructor(t,e){super();this.upstream=t,this.maxCount=e,this.count=0}summary(){return`${this.upstream.summary()} -> Take`}async next(){return this.count++>=this.maxCount?{value:null,done:!0}:this.upstream.next()}},J$=class extends ir{constructor(t,e,n=!0){super();this.upstream=t,this.batchSize=e,this.enableSmallLastBatch=n,this.lastRead=Promise.resolve({value:null,done:!1})}summary(){return`${this.upstream.summary()} -> RowMajorBatch`}async next(){return this.lastRead=this.lastRead.then(()=>this.serialNext()),this.lastRead}async serialNext(){let t=[];for(;t.length<this.batchSize;){let e=await this.upstream.next();if(e.done)return this.enableSmallLastBatch&&t.length>0?{value:t,done:!1}:{value:null,done:!0};t.push(e.value)}return{value:t,done:!1}}},Q$=class extends ir{constructor(t,e){super();this.upstream=t,this.predicate=e,this.lastRead=Promise.resolve({value:null,done:!1})}summary(){return`${this.upstream.summary()} -> Filter`}async next(){return this.lastRead=this.lastRead.then(()=>this.serialNext()),this.lastRead}async serialNext(){for(;;){let t=await this.upstream.next();if(t.done||this.predicate(t.value))return t;_t(t.value)}}},tR=class extends ir{constructor(t,e){super();this.upstream=t,this.transform=e}summary(){return`${this.upstream.summary()} -> Map`}async next(){let t=await this.upstream.next();if(t.done)return{value:null,done:!0};let e=po.getTensorsInContainer(t.value),n=this.transform(t.value),o=po.getTensorsInContainer(n);for(let s of e)po.isTensorInList(s,o)||s.dispose();return{value:n,done:!1}}},eR=class extends ir{constructor(t,e){super();this.upstream=t,this.handler=e,this.count=0,this.lastRead=Promise.resolve({value:null,done:!1})}summary(){return`${this.upstream.summary()} -> handleErrors`}async next(){return this.lastRead=this.lastRead.then(()=>this.serialNext()),this.lastRead}async serialNext(){for(;;)try{return await this.upstream.next()}catch(t){if(!this.handler(t))return{value:null,done:!0}}}},YI=class extends ir{constructor(t,e){super();this.upstream=t,this.transform=e}summary(){return`${this.upstream.summary()} -> AsyncMap`}async next(){let t=await this.upstream.next();if(t.done)return{value:null,done:!0};let e=po.getTensorsInContainer(t.value),n=await this.transform(t.value),o=po.getTensorsInContainer(n);for(let s of e)po.isTensorInList(s,o)||s.dispose();return{value:n,done:!1}}},Rm=class extends ir{constructor(){super();this.outputQueue=new $m,this.lastRead=Promise.resolve({value:null,done:!1})}async next(){return this.lastRead=this.lastRead.then(()=>this.serialNext()),this.lastRead}async serialNext(){for(;this.outputQueue.length()===0;)if(!await this.pump())return{value:null,done:!0};return{value:this.outputQueue.shift(),done:!1}}},rR=class extends Rm{constructor(t,e){super();this.upstream=t,this.transform=e}summary(){return`${this.upstream.summary()} -> Flatmap`}async pump(){let t=await this.upstream.next();if(t.done)return!1;let e=po.getTensorsInContainer(t.value),n=this.transform(t.value),o=po.getTensorsInContainer(n);this.outputQueue.pushAll(n);for(let s of e)po.isTensorInList(s,o)||s.dispose();return!0}},JI=class extends ir{constructor(t,e){super();this.baseErrorHandler=e,this.lastRead=null,this.iterator=null,this.moreIterators=t}summary(){return"TODO: fill in upstream of chained summaries -> Chained"}async next(){return this.lastRead=this.readFromChain(this.lastRead),this.lastRead}async readFromChain(t){if(await t,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 e=await this.iterator.next();return e.done?(this.iterator=null,this.readFromChain(t)):e}},al;(function(r){r[r.FAIL=0]="FAIL",r[r.SHORTEST=1]="SHORTEST",r[r.LONGEST=2]="LONGEST"})(al||(al={}));var nR=class extends ir{constructor(t,e=al.FAIL){super();this.iterators=t,this.mismatchMode=e,this.count=0,this.currentPromise=null}summary(){return"{TODO: fill in upstream of zip summaries} -> Zip"}async nextState(t){await t;let e=0,n=0;function o(i){return i instanceof ir?{value:i.next().then(u=>(e++,u.done&&n++,u.value)),recurse:!1}:{value:null,recurse:!0}}let s=await Oy(this.iterators,o);if(e===n)return{value:null,done:!0};if(n>0)switch(this.mismatchMode){case al.FAIL:throw new Error(`Zipped streams should have the same length. Mismatched at element ${this.count}.`);case al.SHORTEST:return{value:null,done:!0};case al.LONGEST:default:}return this.count++,{value:s,done:!1}}async next(){return this.currentPromise=this.nextState(this.currentPromise),this.currentPromise}},QI=class extends ir{constructor(t,e){super();this.upstream=t,this.bufferSize=e,this.buffer=new Lh(e)}summary(){return`${this.upstream.summary()} -> Prefetch`}refill(){for(;!this.buffer.isFull();){let t=this.upstream.next();this.buffer.push(t)}}next(){return this.refill(),this.buffer.shift()}},oR=class extends QI{constructor(t,e,n){super(t,e);this.upstream=t,this.windowSize=e,this.upstreamExhausted=!1,this.random=U$.alea(n||x.now().toString()),this.lastRead=Promise.resolve({value:null,done:!1})}async next(){return this.lastRead=this.lastRead.then(()=>this.serialNext()),this.lastRead}randomInt(t){return Math.floor(this.random()*t)}chooseIndex(){return this.randomInt(this.buffer.length())}async serialNext(){for(this.upstreamExhausted||this.refill();!this.buffer.isEmpty();){let t=this.chooseIndex(),e=await this.buffer.shuffleExcise(t);if(e.done)this.upstreamExhausted=!0;else return this.refill(),e}return{value:null,done:!0}}};var Wi=class{constructor(){this.size=null}batch(t,e=!0){let n=this;x.assert(t>0,()=>`batchSize needs to be positive, but it is
${t}`);let o;return this.size===1/0||this.size==null?o=this.size:e?o=Math.ceil(this.size/t):o=Math.floor(this.size/t),An(async()=>(await n.iterator()).columnMajorBatch(t,e,hZ),o)}concatenate(t){let e=this,n;return this.size===1/0||t.size===1/0?n=1/0:this.size!=null&&t.size!=null?n=this.size+t.size:n=null,An(async()=>(await e.iterator()).concatenate(await t.iterator()),n)}filter(t){let e=this,n;return this.size===1/0?n=1/0:n=null,An(async()=>(await e.iterator()).filter(o=>B(()=>t(o))),n)}async forEachAsync(t){return(await this.iterator()).forEachAsync(t)}map(t){let e=this;return An(async()=>(await e.iterator()).map(n=>B(()=>t(n))),this.size)}mapAsync(t){let e=this;return An(async()=>(await e.iterator()).mapAsync(t),this.size)}prefetch(t){if(t==null)throw new RangeError("`Dataset.prefetch()` requires bufferSize to be specified.");let e=this;return An(async()=>(await e.iterator()).prefetch(t),this.size)}repeat(t){let e=this,n;return this.size!=null&&t>0?n=this.size*t:t===0?n=0:this.size!=null&&(t===void 0||t<0)?n=1/0:n=null,An(async()=>{let o=zh(async()=>({value:await e.iterator(),done:!1}));return H$(o.take(t))},n)}skip(t){let e=this,n;return this.size!=null&&t>=0&&this.size>=t?n=this.size-t:this.size!=null&&(this.size<t||t===void 0||t<0)?n=0:n=null,An(async()=>(await e.iterator()).skip(t),n)}shuffle(t,e,n=!0){if(t==null||t<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 (${this.size} elements)`);let o=this,s=sR.alea(e||x.now().toString());return An(async()=>{let i=s.int32();return n&&(i+=s.int32()),(await o.iterator()).shuffle(t,i.toString())},this.size)}take(t){let e=this,n;return this.size!=null&&this.size>t?n=t:this.size!=null&&this.size<=t?n=this.size:n=null,An(async()=>(await e.iterator()).take(t),n)}async toArray(){if(this.size===1/0)throw new Error("Can not convert infinite data stream to array.");return(await this.iterator()).toArray()}async toArrayForTest(){if(this.size===1/0)throw new Error("Can not convert infinite data stream to array.");return(await this.iterator()).toArrayForTest()}};Wi.MAX_BUFFER_SIZE=1e4;function An(r,t=null){return new class extends Wi{constructor(){super(...arguments);this.size=t}async iterator(){return r()}}}function iR(r){return An(async()=>ZI(r),r.length)}function aR(r){if(!ru(r))throw new Error("The argument to zip() must be an object or array.");let t;if(Array.isArray(r))for(let e=0;e<r.length;e++)t=t==null?r[e].size:Math.min(t,r[e].size);else if(r instanceof Object)for(let e in r)t=t==null?r[e].size:Math.min(t,r[e].size);return An(async()=>{let e=await Oy(r,n=>{if(n instanceof Wi)return{value:n.iterator(),recurse:!1};if(ru(n))return{value:null,recurse:!0};throw new Error("Leaves of the structure passed to zip() must be Datasets, not primitives.")});return q$(e,al.SHORTEST)},t)}function hZ(r){if(r===null)return null;let t=r[0];return G$(t)?{value:gZ(r),recurse:!1}:{value:null,recurse:!0}}function gZ(r){if(r.length===0)throw new Error("Can't make a batch of zero elements.");return r[0]instanceof Vt?Je(r):Ar(r)}var Bh=class extends Wi{constructor(t){super();this.input=t}async iterator(){return(await this.input.iterator()).decodeUTF8().split(`
`).map(o=>(o.endsWith("\r")&&(o=o.slice(0,-1)),o))}};var Py='"',Vh=Symbol("out"),lR=Symbol("field"),My=Symbol("quote"),tS=Symbol("quoteafterquote"),uR=Symbol("quoteinquote"),Gh=class extends Wi{constructor(t,e){super();this.input=t,this.hasHeader=!0,this.fullColumnNames=null,this.columnNamesValidated=!1,this.columnConfigs=null,this.configuredColumnsOnly=!1,this.delimiter=",",this.delimWhitespace=!1,this.base=new Bh(t),e||(e={}),this.hasHeader=e.hasHeader!==!1,this.fullColumnNames=e.columnNames,this.columnConfigs=e.columnConfigs,this.configuredColumnsOnly=e.configuredColumnsOnly,e.delimWhitespace?(x.assert(e.delimiter==null,()=>"Delimiter should not be provided when delimWhitespace is true."),this.delimWhitespace=!0,this.delimiter=" "):this.delimiter=e.delimiter?e.delimiter:","}async columnNames(){return this.columnNamesValidated||await this.setColumnNames(),this.configuredColumnsOnly?Object.keys(this.columnConfigs):this.fullColumnNames}async setColumnNames(){let t=await this.maybeReadHeaderLine();if(!this.fullColumnNames&&!t)throw new Error("Column names must be provided if there is no header line.");this.fullColumnNames&&t&&x.assert(t.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 ("+t.length.toString()+")."),this.fullColumnNames||(this.fullColumnNames=t);let e=this.fullColumnNames.reduce((o,s)=>(o[s]=o[s]+1||1,o),{}),n=Object.keys(e).filter(o=>e[o]>1);if(x.assert(n.length===0,()=>"Duplicate column names found: "+n.toString()),this.columnConfigs){for(let o of Object.keys(this.columnConfigs))if(this.fullColumnNames.indexOf(o)===-1)throw new Error('The key "'+o+'" 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 n=e.value;return this.parseRow(n,!1)}else return null}async iterator(){this.columnNamesValidated||await this.setColumnNames();let t=await this.base.iterator();return this.hasHeader&&(t=t.skip(1)),t.map(e=>this.makeDataElement(e))}makeDataElement(t){let e=this.parseRow(t),n={},o={};for(let s=0;s<this.fullColumnNames.length;s++){let i=this.fullColumnNames[s],a=this.columnConfigs?this.columnConfigs[i]:null;if(!(this.configuredColumnsOnly&&!a)){let u=e[s],l=null;if(u==="")if(a&&a.default!==void 0)l=a.default;else{if(a&&(a.required||a.isLabel))throw new Error(`Required column ${i} is empty in this line: ${t}`);l=void 0}else{let c=Number(u);if(isNaN(c))a&&a.dtype==="bool"?l=this.getBoolean(u):l=u;else if(!a||!a.dtype)l=c;else switch(a.dtype){case"float32":l=c;break;case"int32":l=Math.floor(c);break;case"bool":l=this.getBoolean(u);break;default:l=c}}a&&a.isLabel?o[i]=l:n[i]=l}}return Object.keys(o).length===0?n:{xs:n,ys:o}}getBoolean(t){return t==="1"||t.toLowerCase()==="true"?1:0}parseRow(t,e=!0){let n=[],o=0,s=t.length,i=Vh;for(let a=0;a<s;a++)switch(i){case Vh:switch(t.charAt(a)){case Py:o=a+1,i=My;break;case this.delimiter:if(o=a+1,this.delimiter===" "&&this.delimWhitespace)break;n.push(""),i=Vh;break;default:i=lR,o=a;break}break;case lR:switch(t.charAt(a)){case this.delimiter:n.push(t.substring(o,a)),i=Vh,o=a+1;break;default:}break;case My:switch(t.charAt(a)){case Py:i=tS;break;default:}break;case tS:switch(t.charAt(a)){case this.delimiter:n.push(t.substring(o,a-1)),i=Vh,o=a+1;break;case Py:i=My;break;default:i=uR;break}break;case uR:switch(t.charAt(a)){case Py:i=My;break;default:}break;default:}if(i===tS?n.push(t.substring(o,s-1)):n.push(t.substring(o)),e&&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}};var Wh=class extends ir{constructor(t){super();this.microphoneConfig=t,this.isClosed=!1,this.fftSize=t.fftSize||1024;let e=Math.log2(this.fftSize);if(this.fftSize<0||e<4||e>14||!Number.isInteger(e))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=t.numFramesPerSpectrogram||43,this.sampleRateHz=t.sampleRateHz,this.columnTruncateLength=t.columnTruncateLength||this.fftSize,this.audioTrackConstraints=t.audioTrackConstraints,this.smoothingTimeConstant=t.smoothingTimeConstant||0,this.includeSpectrogram=t.includeSpectrogram!==!1,this.includeWaveform=t.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(t={}){if(!V().get("IS_BROWSER"))throw new Error("microphone API is only supported in browser environment.");let e=new Wh(t);return await e.start(),e}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 t=window.AudioContext||window.webkitAudioContext;if(this.audioContext=new t,!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 e=this.audioContext.createMediaStreamSource(this.stream);this.analyser=this.audioContext.createAnalyser(),this.analyser.fftSize=this.fftSize*2,this.analyser.smoothingTimeConstant=this.smoothingTimeConstant,e.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 t,e,n=await this.getAudioData();if(this.includeSpectrogram){let o=this.flattenQueue(n.freqDataQueue);t=this.getTensorFromAudioDataArray(o,[this.numFrames,this.columnTruncateLength,1])}if(this.includeWaveform){let o=this.flattenQueue(n.timeDataQueue);e=this.getTensorFromAudioDataArray(o,[this.numFrames*this.fftSize,1])}return{value:{spectrogram:t,waveform:e},done:!1}}async capture(){return(await this.next()).value}async getAudioData(){let t=[],e=[],n=0;return new Promise(o=>{let s=setInterval(()=>{this.includeSpectrogram&&(this.analyser.getFloatFrequencyData(this.freqData),this.freqData[0]===-1/0&&o({freqDataQueue:t,timeDataQueue:e}),t.push(this.freqData.slice(0,this.columnTruncateLength))),this.includeWaveform&&(this.analyser.getFloatTimeDomainData(this.timeData),e.push(this.timeData.slice())),++n===this.numFrames&&(clearInterval(s),o({freqDataQueue:t,timeDataQueue:e}))},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(t){let e=t[0].length,n=new Float32Array(t.length*e);return t.forEach((o,s)=>n.set(o,s*e)),n}getTensorFromAudioDataArray(t,e){let n=new Float32Array(x.sizeFromShape(e));return n.set(t,n.length-t.length),Ar(n,e)}};var Uh=class extends ir{constructor(t,e){super();if(this.webcamVideoElement=t,this.webcamConfig=e,this.isClosed=!0,this.resize=!1,this.needToResize())if(this.resize=!0,this.cropSize=[this.webcamConfig.resizeHeight,this.webcamConfig.resizeWidth],this.cropBoxInd=Re([0],"int32"),this.webcamConfig.centerCrop){let n=this.webcamConfig.resizeWidth*1/this.webcamVideoElement.width,o=this.webcamConfig.resizeHeight*1/this.webcamVideoElement.height,s=(1-n)/2,i=(1-o)/2,a=s+n,u=o+i;this.cropBox=Pi([i,s,u,a],[1,4])}else this.cropBox=Pi([0,0,1,1],[1,4])}summary(){return"webcam"}static async create(t,e={}){if(!V().get("IS_BROWSER"))throw new Error("tf.data.webcam is only supported in browser environment.");if(!t){if(t=document.createElement("video"),!e.resizeWidth||!e.resizeHeight)throw new Error("Please provide webcam video element, or resizeWidth and resizeHeight to create a hidden video element.");t.width=e.resizeWidth,t.height=e.resizeHeight}let n=new Uh(t,e);return await n.start(),n}async start(){this.webcamConfig.facingMode&&x.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(t){throw t.message=`Error thrown while initializing video stream: ${t.message}`,t}if(!this.stream)throw new Error("Could not obtain video from webcam.");try{this.webcamVideoElement.srcObject=this.stream}catch(t){console.log(t),this.webcamVideoElement.src=window.URL.createObjectURL(this.stream)}return this.webcamVideoElement.play(),this.isClosed=!1,new Promise(t=>{this.webcamVideoElement.onloadedmetadata=()=>{t()}})}async next(){if(this.isClosed)return{value:null,done:!0};let t;try{t=dx.fromPixels(this.webcamVideoElement)}catch(e){throw new Error(`Error thrown converting video to pixels: ${JSON.stringify(e)}`)}if(this.resize)try{return{value:this.cropAndResizeFrame(t),done:!1}}catch(e){throw new Error(`Error thrown cropping the video: ${e.message}`)}finally{t.dispose()}else return{value:t,done:!1}}needToResize(){return!!(this.webcamConfig.resizeWidth&&this.webcamConfig.resizeHeight&&(this.webcamVideoElement.width!==this.webcamConfig.resizeWidth||this.webcamVideoElement.height!==this.webcamConfig.resizeHeight))}cropAndResizeFrame(t){return B(()=>{let e=fr(Z(t,"float32"),0),n;n=pn.cropAndResize(e,this.cropBox,this.cropBoxInd,this.cropSize,"bilinear");let o=n.shape;return O(n,o.slice(1))})}async capture(){return(await this.next()).value}stop(){this.stream.getTracks().forEach(e=>e.stop());try{this.webcamVideoElement.srcObject=null}catch(e){console.log(e),this.webcamVideoElement.src=null}this.isClosed=!0}toArray(){throw new Error("Can not convert infinite video stream to array.")}};var Hh=class{};var Ly=class extends ir{split(t){return new cR(this,t)}},cR=class extends Ly{constructor(t,e){super();this.upstream=t,this.impl=new pR(t,e)}summary(){return this.impl.summary()}async next(){return this.impl.next()}},pR=class extends Rm{constructor(t,e){super();this.upstream=t,this.separator=e,this.carryover=""}summary(){return`${this.upstream.summary()} -> Split('${this.separator}')`}async pump(){let t=await this.upstream.next();if(t.done)return this.carryover===""?!1:(this.outputQueue.push(this.carryover),this.carryover="",!0);let e=t.value.split(this.separator);e[0]=this.carryover+e[0];for(let n of e.slice(0,-1))this.outputQueue.push(n);return this.carryover=e[e.length-1],!0}};var eS=class extends ir{decodeUTF8(){return new mR(this)}},mR=class extends Ly{constructor(t){super();this.upstream=t,this.impl=new fR(t)}summary(){return this.impl.summary()}async next(){return this.impl.next()}},fR=class extends Rm{constructor(t){super();if(this.upstream=t,V().get("IS_BROWSER"))this.decoder=new TextDecoder("utf-8");else{let{StringDecoder:e}=jI();this.decoder=new e("utf8")}}summary(){return`${this.upstream.summary()} -> Utf8`}async pump(){let t=await this.upstream.next(),e;if(t.done)return!1;e=t.value;let n;return V().get("IS_BROWSER")?n=this.decoder.decode(e,{stream:!0}):n=this.decoder.write(Buffer.from(e.buffer)),this.outputQueue.push(n),!0}};var qh=class extends eS{constructor(t,e={}){super();this.file=t,this.options=e,x.assert(t instanceof Uint8Array||(V().get("IS_BROWSER")?t instanceof File||t instanceof Blob:!1),()=>"FileChunkIterator only supports File, Blob and Uint8Array right now."),this.offset=e.offset||0,this.chunkSize=e.chunkSize||1024*1024}summary(){return`FileChunks ${this.file}`}async next(){return this.offset>=(this.file instanceof Uint8Array?this.file.byteLength:this.file.size)?{value:null,done:!0}:{value:await new Promise((e,n)=>{let o=this.offset+this.chunkSize;if(this.file instanceof Uint8Array)e(new Uint8Array(this.file.slice(this.offset,o)));else{let s=new FileReader;s.onload=a=>{let u=s.result;if(u instanceof ArrayBuffer&&(u=new Uint8Array(u)),!(u instanceof Uint8Array))return n(new TypeError("FileReader returned unknown type."));e(u)},s.onabort=a=>n(new Error("Aborted")),s.onerror=a=>n(new Error(a.type));let i=this.file.slice(this.offset,o);s.readAsArrayBuffer(i)}this.offset=o}),done:!1}}};async function dR(r,t={},e){let n,o;typeof r=="string"?n=r:(n=r.url,o=xZ(r));let s=await(e||x.fetch)(n,o);if(s.ok){let i=new Uint8Array(await s.arrayBuffer());return new qh(i,t)}else throw new Error(s.statusText)}var xZ=r=>({method:r.method,headers:r.headers,body:r.body,mode:r.mode,credentials:r.credentials,cache:r.cache,redirect:r.redirect,referrer:r.referrer,integrity:r.integrity});function zy(r){return typeof r=="string"&&r.substr(0,7)==="file://"}var Kh=class extends Hh{constructor(t,e={}){super();this.input=t,this.options=e}async iterator(){if(zy(this.input)&&V().get("IS_NODE")){let t=By();this.input=t.readFileSync(this.input.substr(7))}return new qh(this.input,this.options)}};var jh=class extends Hh{constructor(t,e={}){super();this.url=t,this.fileOptions=e}async iterator(){return zy(this.url)?new Kh(this.url,this.fileOptions).iterator():dR(this.url,this.fileOptions)}};function hR(r,t={}){return new Gh(new jh(r),t)}function gR(r){let t=zh(r);return An(async()=>t)}function xR(r){return An(async()=>{let t=await r();return zh(()=>t.next())})}async function yR(r,t){return Uh.create(r,t)}async function bR(r){return Wh.create(r)}var rS="3.15.0";function nt(r,t){Array.isArray(r)||(r=[r]),r.forEach(e=>{e!=null&&x.assert(e.dtype!=="complex64",()=>`${t} does not support complex64 tensors in the CPU backend.`)})}var yZ=Vr.whereImpl,Sc=class extends mi{constructor(){super();this.blockSize=48,this.firstUse=!0,this.data=new wl(this,fo())}nextDataId(){return Sc.nextDataId++}write(t,e,n){this.firstUse&&(this.firstUse=!1,V().get("IS_NODE")&&S.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 o={id:this.nextDataId()};return this.data.set(o,{values:t,dtype:n,refCount:1}),o}makeTensorInfo(t,e,n){let o;if(e==="string"&&n!=null&&n.length>0&&x.isString(n[0])){let s=n.map(i=>x.encodeString(i));o=this.write(s,t,e)}else o=this.write(n,t,e);return{dataId:o,shape:t,dtype:e}}refCount(t){return this.data.has(t)?this.data.get(t).refCount:0}incRef(t){let e=this.data.get(t);e.refCount++}decRef(t){if(this.data.has(t)){let e=this.data.get(t);e.refCount--}}move(t,e,n,o,s){this.data.set(t,{values:e,dtype:o,refCount:s})}numDataIds(){return this.data.numDataIds()}async read(t){return this.readSync(t)}readSync(t){let{dtype:e,complexTensorInfos:n}=this.data.get(t);if(e==="complex64"){let o=this.readSync(n.real.dataId),s=this.readSync(n.imag.dataId);return S.mergeRealAndImagArrays(o,s)}return this.data.get(t).values}bufferSync(t){let e=this.readSync(t.dataId),n=e;if(t.dtype==="string")try{n=e.map(o=>x.decodeString(o))}catch(o){throw new Error("Failed to decode encoded string bytes into utf-8")}return It(t.shape,t.dtype,n)}makeOutput(t,e,n){let o=this.write(t,e,n);return fo().makeTensorFromDataId(o,e,n,this)}disposeData(t,e=!1){if(this.data.has(t)){if(this.data.get(t).refCount--,!e&&this.data.get(t).refCount>0)return!1;let{complexTensorInfos:n}=this.data.get(t);n!=null&&(this.disposeData(n.real.dataId,!0),this.disposeData(n.imag.dataId,!0)),this.data.delete(t)}return!0}disposeIntermediateTensorInfo(t){this.disposeData(t.dataId)}async time(t){let e=x.now();return t(),{kernelMs:x.now()-e}}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(t){nt([t],"where");let e=this.readSync(t.dataId);return yZ(t.shape,e)}dispose(){}floatPrecision(){return 32}epsilon(){return super.epsilon()}};Sc.nextDataId=0;var eb={};Yt(eb,{addImpl:()=>oS,bincountImpl:()=>Pm,bincountReduceImpl:()=>Vy,ceilImpl:()=>sS,concatImpl:()=>kc,equalImpl:()=>iS,expImpl:()=>lS,expm1Impl:()=>cS,floorImpl:()=>pS,gatherNdImpl:()=>Gy,gatherV2Impl:()=>Wy,greaterEqualImpl:()=>fS,greaterImpl:()=>mS,lessEqualImpl:()=>hS,lessImpl:()=>dS,linSpaceImpl:()=>Uy,logImpl:()=>gS,maxImpl:()=>Hy,maximumImpl:()=>xS,minimumImpl:()=>yS,multiplyImpl:()=>Xh,negImpl:()=>bS,notEqualImpl:()=>wS,prodImpl:()=>vS,rangeImpl:()=>Tc,rsqrtImpl:()=>CS,sigmoidImpl:()=>HR,simpleAbsImpl:()=>nS,sliceImpl:()=>_c,sparseFillEmptyRowsImpl:()=>qy,sparseReshapeImpl:()=>Ky,sparseSegmentReductionImpl:()=>Lm,sqrtImpl:()=>jR,squaredDifferenceImpl:()=>SS,stridedSliceImpl:()=>jy,stringNGramsImpl:()=>Xy,stringSplitImpl:()=>Yy,stringToHashBucketFastImpl:()=>Zy,subImpl:()=>kS,tileImpl:()=>Jy,topKImpl:()=>Qy,transposeImpl:()=>Mm,uniqueImpl:()=>tb});function nS(r){let t=new Float32Array(r.length);for(let e=0;e<r.length;++e)t[e]=Math.abs(r[e]);return t}var bZ=r=>{let{x:t}=r.inputs,e=r.backend;nt(t,"abs");let n=new Float32Array(x.sizeFromShape(t.shape)),o=e.data.get(t.dataId).values;return n=nS(o),e.makeOutput(n,t.shape,t.dtype)},vR={kernelName:hi,backendName:"cpu",kernelFunc:bZ};function te(r){return(t,e,n,o,s)=>{let i=S.assertAndGetBroadcastShape(t,e),a=i.length,u=x.computeStrides(i),l=x.sizeFromShape(i),c=x.getTypedArrayFromDType(s,l),p=t.length,m=e.length,f=x.computeStrides(t),d=x.computeStrides(e),h=S.getBroadcastDims(t,i),g=S.getBroadcastDims(e,i);if(h.length+g.length===0)for(let y=0;y<c.length;++y)c[y]=r(n[y%n.length],o[y%o.length]);else for(let y=0;y<c.length;++y){let b=x.indexToLoc(y,a,u),w=b.slice(-p);h.forEach(D=>w[D]=0);let v=x.locToIndex(w,p,f),k=b.slice(-m);g.forEach(D=>k[D]=0);let E=x.locToIndex(k,m,d);c[y]=r(n[v],o[E])}return[c,i]}}function Ir(r){let{inputs:t,backend:e}=r,{real:n,imag:o}=t,s=e.data.get(n.dataId).values,i=e.data.get(o.dataId).values,a=e.makeTensorInfo(n.shape,"complex64"),u=e.data.get(a.dataId);return u.complexTensorInfos={real:e.makeTensorInfo(n.shape,"float32",s),imag:e.makeTensorInfo(o.shape,"float32",i)},a}var CR={kernelName:mp,backendName:"cpu",kernelFunc:Ir};function Fm(r,t,e="float32"){if(e==="complex64"){let o=Fm(r,t,"float32"),s=Fm(r,t,"float32");return Ir({inputs:{real:o,imag:s},backend:r})}let n=x.makeZerosTypedArray(x.sizeFromShape(t),e);return r.makeTensorInfo(t,e,n)}function Wr(r){let{inputs:t,backend:e}=r,{x:n}=t;return e.incRef(n.dataId),{dataId:n.dataId,shape:n.shape,dtype:n.dtype}}var IR={kernelName:io,backendName:"cpu",kernelFunc:Wr};function Ao(r){let{inputs:t,backend:e}=r,{input:n}=t,o=e.data.get(n.dataId).complexTensorInfos.real,s=e.data.get(o.dataId).values;return e.makeTensorInfo(o.shape,o.dtype,s)}var SR={kernelName:Dp,backendName:"cpu",kernelFunc:Ao};function Do(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{dtype:s}=n;if(s==="complex64"){if(o.dtype==="complex64")return Wr({inputs:{x:o},backend:e});let i=Fm(e,o.shape,o.dtype),a=Do({inputs:{x:o},backend:e,attrs:{dtype:"float32"}}),u=Ir({inputs:{real:a,imag:i},backend:e});return e.disposeIntermediateTensorInfo(i),e.disposeIntermediateTensorInfo(a),u}if(o.dtype==="complex64"){let i=Ao({inputs:{input:o},backend:e}),a=Do({inputs:{x:i},backend:e,attrs:{dtype:s}});return e.disposeIntermediateTensorInfo(i),a}if(!x.hasEncodingLoss(o.dtype,s)){let i=Wr({inputs:{x:o},backend:e});return{dataId:i.dataId,shape:i.shape,dtype:s}}if(s==="int32"){let i=e.data.get(o.dataId).values,a=Int32Array.from(i);return e.makeTensorInfo(o.shape,"int32",a)}if(s==="bool"){let i=e.data.get(o.dataId).values,a=x.toTypedArray([0],o.dtype),[u,l]=te((c,p)=>c!==p?1:0)(o.shape,[],i,a,"bool");return e.makeTensorInfo(l,"bool",u)}throw new Error(`Error in Cast: failed to cast ${o.dtype} to ${s}`)}var kR={kernelName:oo,backendName:"cpu",kernelFunc:Do};function ie(r,t,e,n){return e==null?({inputs:o,backend:s})=>{let{a:i,b:a}=o,u=s;nt([i,a],r);let l=u.data.get(i.dataId).values,c=u.data.get(a.dataId).values,p=i.dtype==="string"?S.fromUint8ToStringArray(l):l,m=i.dtype==="string"?S.fromUint8ToStringArray(c):c,f=n||i.dtype,[d,h]=t(i.shape,a.shape,p,m,f);return u.makeTensorInfo(h,f,d)}:({inputs:o,backend:s})=>{let{a:i,b:a}=o,u=s;if(i.dtype==="complex64"||a.dtype==="complex64"){let l=Do({inputs:{x:i},backend:u,attrs:{dtype:"complex64"}}),c=u.data.get(l.dataId),p=c.complexTensorInfos.real,m=c.complexTensorInfos.imag,f=u.data.get(p.dataId).values,d=u.data.get(m.dataId).values,h=Do({inputs:{x:a},backend:u,attrs:{dtype:"complex64"}}),g=u.data.get(h.dataId),y=g.complexTensorInfos.real,b=g.complexTensorInfos.imag,w=u.data.get(y.dataId).values,v=u.data.get(b.dataId).values,[k,E,D]=e(i.shape,a.shape,f,d,w,v),$=u.makeTensorInfo(D,"float32",k),R=u.makeTensorInfo(D,"float32",E),M=Ir({inputs:{real:$,imag:R},backend:u});return u.disposeIntermediateTensorInfo(l),u.disposeIntermediateTensorInfo(h),u.disposeIntermediateTensorInfo($),u.disposeIntermediateTensorInfo(R),M}else{let l=u.data.get(i.dataId).values,c=u.data.get(a.dataId).values,p=n||i.dtype,[m,f]=t(i.shape,a.shape,l,c,p);return u.makeTensorInfo(f,p,m)}}}function Om(r){return(t,e,n,o,s,i)=>{let a=S.assertAndGetBroadcastShape(t,e),u=x.sizeFromShape(a),l=a.length,c=x.computeStrides(a),p=x.getTypedArrayFromDType("float32",u),m=x.getTypedArrayFromDType("float32",u),f=S.getBroadcastDims(t,a),d=S.getBroadcastDims(e,a),h=S.mergeRealAndImagArrays(n,o),g=S.mergeRealAndImagArrays(s,i),y=t.length,b=x.computeStrides(t),w=e.length,v=x.computeStrides(e);if(f.length+d.length===0)for(let k=0;k<p.length;k++){let E=k%h.length,D=k%g.length,$=r(h[E*2],h[E*2+1],g[D*2],g[D*2+1]);p[k]=$.real,m[k]=$.imag}else for(let k=0;k<p.length;k++){let E=x.indexToLoc(k,l,c),D=E.slice(-y);f.forEach(U=>D[U]=0);let $=x.locToIndex(D,y,b),R=E.slice(-w);d.forEach(U=>R[U]=0);let M=x.locToIndex(R,w,v),W=r(h[$*2],h[$*2+1],g[M*2],g[M*2+1]);p[k]=W.real,m[k]=W.imag}return[p,m,a]}}var oS=te((r,t)=>r+t),wZ=Om((r,t,e,n)=>({real:r+e,imag:t+n})),ll=ie(Kn,oS,wZ),NR={kernelName:Kn,backendName:"cpu",kernelFunc:ll};function Pm(r,t,e,n,o){let s=x.sizeFromShape(n),i=x.makeZerosTypedArray(o,e);for(let a=0;a<r.length;a++){let u=r[a];if(u<0)throw new Error("Input x must be non-negative!");u>=o||(s>0?i[u]+=t[a]:i[u]+=1)}return i}function Vy(r,t,e,n=!1){let o=r.shape[0],s=r.shape[1],i=It([o,e],t.dtype);for(let a=0;a<o;a++)for(let u=0;u<s;u++){let l=r.get(a,u);if(l<0)throw new Error("Input x must be non-negative!");l>=e||(n?i.set(1,a,l):t.size>0?i.set(i.get(a,l)+t.get(a,u),a,l):i.set(i.get(a,l)+1,a,l))}return i}function wn(r){return(t,e,n)=>{let o=x.getTypedArrayFromDType(e,t.length);for(let s=0;s<t.length;++s)o[s]=r(t[s],n);return o}}function Et(r,t,e){return({inputs:n,attrs:o,backend:s})=>{let{x:i}=n;if(nt(i,r),i.dtype==="string"||e==="string")throw new Error("unaryKernelFunc does not support string input/output");let a=s,u=a.data.get(i.dataId).values,l=x.sizeFromShape(i.shape),c=e||i.dtype,p=x.getArrayFromDType(c,l);for(let m=0;m<l;++m)p[m]=t(u[m],o);return a.makeTensorInfo(i.shape,c,p)}}function $o(r,t,e){return({inputs:n,attrs:o,backend:s})=>{let{x:i}=n;if(nt(i,r),i.dtype==="string"||e==="string")throw new Error("unaryKernelFunc does not support string input/output");let a=s,u=a.data.get(i.dataId).values,l=e||i.dtype,c=t(u,l,o);return a.makeTensorInfo(i.shape,l,c)}}var sS=wn(r=>Math.ceil(r)),vZ=$o(Jo,sS),TR={kernelName:Jo,backendName:"cpu",kernelFunc:vZ};function kc(r,t,e,n){let o=x.getArrayFromDType(e,x.sizeFromShape(t));if(n&&e!=="string"){let s=0;r.forEach(i=>{let a=x.sizeFromShape(i.shape);o.set(i.vals,s),s+=a})}else{let s=0;r.forEach(i=>{let a=e==="string"?S.fromUint8ToStringArray(i.vals):i.vals,u=0;for(let l=0;l<i.shape[0];++l){let c=l*t[1]+s;for(let p=0;p<i.shape[1];++p)o[c+p]=a[u++]}s+=i.shape[1]})}return o}var iS=te((r,t)=>r===t?1:0),aS=ie(aa,iS,null,"bool"),_R={kernelName:aa,backendName:"cpu",kernelFunc:aS};var lS=wn(r=>Math.exp(r)),uS=$o(as,lS,"float32"),ER={kernelName:as,backendName:"cpu",kernelFunc:uS};var cS=wn(r=>Math.expm1(r)),CZ=$o(la,cS),AR={kernelName:la,backendName:"cpu",kernelFunc:CZ};var pS=wn(r=>Math.floor(r)),IZ=$o(ls,pS),DR={kernelName:ls,backendName:"cpu",kernelFunc:IZ};function Gy(r,t,e,n,o,s,i,a,u){let l=It([n,s],e);for(let c=0;c<n;c++){let p=[],m=0;for(let f=0;f<o;f++){let d=r[c*o+f];m+=d*i[f],p.push(d)}if(m<0||m>=u/s)throw new Error(`Invalid indices: ${p} does not index into ${a}`);for(let f=0;f<s;f++)l.values[c*s+f]=t.get(...t.indexToLoc(m*s+f))}return l}function Wy(r,t,e){let n=It(e,r.dtype);for(let o=0;o<n.size;++o){let i=n.indexToLoc(o).slice(),a=i[0],u=i[2],l=t.locToIndex([a,u]);i[2]=t.values[l];let c=r.locToIndex(i);0<=c&&c<r.values.length&&(n.values[o]=r.values[c])}return n}var mS=te((r,t)=>r>t?1:0),SZ=ie(pa,mS,null,"bool"),$R={kernelName:pa,backendName:"cpu",kernelFunc:SZ};var fS=te((r,t)=>r>=t?1:0),kZ=ie(ps,fS,null,"bool"),RR={kernelName:ps,backendName:"cpu",kernelFunc:kZ};var dS=te((r,t)=>r<t?1:0),NZ=ie(ha,dS,null,"bool"),FR={kernelName:ha,backendName:"cpu",kernelFunc:NZ};var hS=te((r,t)=>r<=t?1:0),TZ=ie(ga,hS,null,"bool"),OR={kernelName:ga,backendName:"cpu",kernelFunc:TZ};function Uy(r,t,e){let n=(t-r)/(e-1),o=x.makeZerosTypedArray(e,"float32");o[0]=r;for(let s=1;s<o.length;s++)o[s]=o[s-1]+n;return o}var gS=wn(r=>Math.log(r)),_Z=$o(fs,gS),PR={kernelName:fs,backendName:"cpu",kernelFunc:_Z};function Hy(r,t,e,n){let o=x.getTypedArrayFromDType(n,x.sizeFromShape(e));for(let s=0;s<o.length;++s){let i=s*t,a=r[i];for(let u=0;u<t;++u){let l=r[i+u];(Number.isNaN(l)||l>a)&&(a=l)}o[s]=a}return o}var xS=te((r,t)=>Math.max(r,t)),EZ=ie(hs,xS),MR={kernelName:hs,backendName:"cpu",kernelFunc:EZ};var yS=te((r,t)=>Math.min(r,t)),AZ=ie(bs,yS),LR={kernelName:bs,backendName:"cpu",kernelFunc:AZ};var Xh=te((r,t)=>r*t),DZ=Om((r,t,e,n)=>({real:r*e-t*n,imag:r*n+t*e})),Nc=ie(vs,Xh,DZ),zR={kernelName:vs,backendName:"cpu",kernelFunc:Nc};function bS(r,t,e){let n=x.createScalarValue(-1,e);return Xh([],t,n,r,e)}function $Z(r){let{inputs:t,backend:e}=r,{x:n}=t;nt(n,"neg");let o=e.data.get(n.dataId).values,[s,i]=bS(o,n.shape,n.dtype);return e.makeTensorInfo(i,n.dtype,s)}var BR={kernelName:wi,backendName:"cpu",kernelFunc:$Z};var wS=te((r,t)=>r!==t?1:0),RZ=ie(wa,wS,null,"bool"),VR={kernelName:wa,backendName:"cpu",kernelFunc:RZ};function Mm(r,t,e,n,o){let s=t.length,i=x.sizeFromShape(t),a=x.computeStrides(t),u=x.computeStrides(o),l=x.getTypedArrayFromDType(e,x.sizeFromShape(o));for(let c=0;c<i;++c){let p=x.indexToLoc(c,s,a),m=new Array(p.length);for(let d=0;d<m.length;d++)m[d]=p[n[d]];let f=x.locToIndex(m,s,u);l[f]=r[c]}return l}function He(r){let{inputs:t,attrs:e,backend:n}=r,{x:o}=t,{perm:s}=e;nt(o,"transpose");let i=o.shape.length,a=new Array(i);for(let p=0;p<a.length;p++)a[p]=o.shape[s[p]];let u=n.data.get(o.dataId).values,l=Mm(u,o.shape,o.dtype,s,a);return{dataId:n.write(l,a,o.dtype),shape:a,dtype:o.dtype}}var GR={kernelName:Vs,backendName:"cpu",kernelFunc:He};function vS(r,t,e,n){let[o,s]=S.computeOutAndReduceShapes(r,n),i=ar(t,"int32"),a=x.makeZerosTypedArray(x.sizeFromShape(o),i),u=x.sizeFromShape(s);for(let l=0;l<a.length;++l){let c=l*u,p=1;for(let m=0;m<u;++m)p*=e[c+m];a[l]=p}return{outVals:a,outShape:o,outDtype:i}}function FZ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s,keepDims:i}=n;nt(o,"prod");let a=o.shape.length,u=x.parseAxisParam(s,o.shape),l=S.getAxesPermutation(u,a),c=u,p=o,m=[];l!=null&&(p=He({inputs:{x:o},backend:e,attrs:{perm:l}}),m.push(p),c=S.getInnerMostAxes(c.length,a));let f=e.data.get(p.dataId).values,{outVals:d,outShape:h,outDtype:g}=vS(p.shape,p.dtype,f,c),y=h;return i&&(y=S.expandShapeToKeepDim(h,u)),m.forEach(b=>e.disposeIntermediateTensorInfo(b)),e.makeTensorInfo(y,g,d)}var WR={kernelName:Sa,backendName:"cpu",kernelFunc:FZ};function Tc(r,t,e,n){let o=r===t,s=r<t&&e<0,i=t<r&&e>1;if(o||s||i)return x.makeZerosTypedArray(0,n);let a=Math.abs(Math.ceil((t-r)/e)),u=x.makeZerosTypedArray(a,n);t<r&&e===1&&(e=-1),u[0]=r;for(let l=1;l<u.length;l++)u[l]=u[l-1]+e;return u}var CS=wn(r=>1/Math.sqrt(r)),OZ=$o(Ds,CS),UR={kernelName:Ds,backendName:"cpu",kernelFunc:OZ};var HR=wn(r=>1/(1+Math.exp(-r))),IS=Et(Rs,r=>1/(1+Math.exp(-r))),qR={kernelName:Rs,backendName:"cpu",kernelFunc:IS};function _c(r,t,e,n,o){let s=Ve.isSliceContinous(n,t,e),i=x.sizeFromShape(e),a=x.computeStrides(n);if(s){let p=Ve.computeFlatOffset(t,a);return o==="string"?r.slice(p,p+i):r.subarray(p,p+i)}let u=o==="string"?S.fromUint8ToStringArray(r):r,l=It(n,o,u),c=It(e,o);for(let p=0;p<c.size;++p){let m=c.indexToLoc(p),f=m.map((d,h)=>d+t[h]);c.set(l.get(...f),...m)}return o==="string"?S.fromStringArrayToUint8(c.values):c.values}function Ro(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{begin:s,size:i}=n;nt(o,"slice");let[a,u]=Ve.parseSliceParams(o,s,i);Ve.assertParamsValid(o,a,u);let l=e.data.get(o.dataId).values,c=_c(l,a,u,o.shape,o.dtype);return e.makeTensorInfo(u,o.dtype,c)}var KR={kernelName:ki,backendName:"cpu",kernelFunc:Ro};function qy(r,t,e,n,o,s,i){let a=t[0],u=s[0],l=new Array(u),c=new Array(a),p=t[1];if(u===0){if(a!==0)throw new Error(S.getSparseFillEmptyRowsIndicesDenseShapeMismatch(a));let g=x.getArrayFromDType(e,0),y=x.getArrayFromDType(o,0);return[g,[0,p],y,l,c]}let m=!0,f=0,d=new Array(u).fill(0);for(let g=0;g<a;++g){let y=r[g*p];if(y<0)throw new Error(S.getSparseFillEmptyRowsNegativeIndexErrorMessage(g,y));if(y>=u)throw new Error(S.getSparseFillEmptyRowsOutOfRangeIndexErrorMessage(g,y,u));++d[y],m=m&&y>=f,f=y}let h=!0;for(let g=0;g<u;++g){let y=d[g]===0;l[g]=y,h=h&&!y,d[g]=Math.max(d[g],1),g>0&&(d[g]+=d[g-1])}if(h&&m){let g=r,y=n;for(let b=0;b<a;++b)c[b]=b;return[g,[a,p],y,l,c]}else{let g=d[u-1],y=x.getArrayFromDType(e,g*p),b=x.getArrayFromDType(o,g),w=new Array(u).fill(0);for(let v=0;v<a;++v){let k=r[v*p],E=w[k],D=(k===0?0:d[k-1])+E;w[k]++;for(let $=0;$<p;++$)y[D*p+$]=r[v*p+$];b[D]=n[v],c[v]=D}for(let v=0;v<u;++v)if(w[v]===0){let E=v===0?0:d[v-1];y[E*p+0]=v;for(let D=1;D<p;++D)y[E*p+D]=0;b[E]=i}return[y,[g,p],b,l,c]}}function Ky(r,t,e,n,o){let s=x.sizeFromShape(n),i=t[0],a=o.length,u=[],l=1,c=-1;for(let g=0;g<a;++g){let y=o[g];if(y===-1){if(c!==-1)throw new Error(S.getSparseReshapeMultipleNegativeOneOutputDimErrorMessage(c,g));c=g,u.push(1)}else{if(y<0)throw new Error(S.getSparseReshapeNegativeOutputDimErrorMessage(g,y));l*=y,u.push(y)}}if(c!==-1){if(l<=0)throw new Error(S.getSparseReshapeEmptyTensorZeroOutputDimErrorMessage());let g=Math.trunc(s/l);if(l*g!==s)throw new Error(S.getSparseReshapeInputOutputMultipleErrorMessage(n,u));u[c]=g}if(x.sizeFromShape(u)!==s)throw new Error(S.getSparseReshapeInputOutputMismatchErrorMessage(n,u));let m=n.length,f=[];if(m>0){f[m-1]=1;for(let g=m-2;g>=0;--g)f[g]=f[g+1]*n[g+1]}let d=[];if(a>0){d[a-1]=1;for(let g=a-2;g>=0;--g)d[g]=d[g+1]*u[g+1]}let h=x.getArrayFromDType(e,i*a);for(let g=0;g<i;++g){let y=0;for(let b=0;b<m;++b)y+=r[g*m+b]*f[b];for(let b=0;b<a;++b)h[g*a+b]=Math.trunc(y/d[b]),y%=d[b]}return[h,[i,a],u]}function Lm(r,t,e,n,o,s=!1,i=0){let a=n.length,u=[t[0],r.length/t[0]],l=u[1],p=a>0?o[a-1]+1:0;if(p<0)throw new Error(S.getSparseSegmentReductionNegativeSegmentIdsErrorMessage());let m=t.slice();m[0]=p;let f=m.reduce((w,v)=>w*v,1),d=x.getArrayFromDType(e,f);if(a===0)return p>0&&d.fill(i),[d,m];if(p<=0)throw new Error(S.getSparseSegmentReductionNegativeSegmentIdsErrorMessage());let h=0,g=1,y=0,b=o[h];for(;;){let w=0;if(g<a){if(w=o[g],b===w){++g;continue}if(b>=w)throw new Error(S.getSparseSegmentReductionNonIncreasingSegmentIdsErrorMessage())}if(b<0||b>=p)throw new Error(S.getSparseSegmentReductionSegmentIdOutOfRangeErrorMessage(b,p));b>y&&d.fill(i,y*l,b*l);for(let v=h;v<g;++v){let k=n[v];if(k<0||k>=u[0])throw new Error(S.getSparseSegmentReductionIndicesOutOfRangeErrorMessage(v,n[v],u[0]));for(let E=0;E<l;E++)d[b*l+E]+=r[k*l+E]}if(s)for(let v=0;v<l;v++)d[b*l+v]/=g-h;if(h=g,++g,y=b+1,b=w,g>a)break}return y<p&&d.fill(i,y*l,p*l),[d,m]}var jR=wn(r=>Math.sqrt(r)),PZ=Et(Fs,r=>Math.sqrt(r)),XR={kernelName:Fs,backendName:"cpu",kernelFunc:PZ};var SS=te((r,t)=>{let e=r-t;return e*e}),MZ=ie(Ms,SS),YR={kernelName:Ms,backendName:"cpu",kernelFunc:MZ};function jy(r,t,e,n){let o=It(r,t.dtype);for(let s=0;s<o.size;s++){let i=o.indexToLoc(s),a=new Array(i.length);for(let u=0;u<a.length;u++)a[u]=i[u]*e[u]+n[u];o.set(t.get(...a),...i)}return o}var ZR=class{constructor(t,e,n,o,s,i){this.separator=x.encodeString(t),this.nGramWidths=e,this.leftPad=x.encodeString(n),this.rightPad=x.encodeString(o),this.padWidth=s,this.preserveShort=i}getPadWidth(t){return Math.min(this.padWidth<0?t-1:this.padWidth,t-1)}getNumNGrams(t,e){let n=this.getPadWidth(e);return Math.max(0,t+2*n-e+1)}createNGrams(t,e,n,o,s,i){for(let a=0;a<s;++a){let u=this.getPadWidth(i),l=Math.max(0,u-a),c=Math.max(0,u-(s-(a+1))),p=i-(l+c),m=e+(l>0?0:a-u),f=0;f+=l*this.leftPad.length;for(let b=0;b<p;++b)f+=t[m+b].length;f+=c*this.rightPad.length,f+=(l+c+p-1)*this.separator.length,n[o+a]=new Uint8Array(f);let h=n[o+a],g=0,y=b=>b.forEach(w=>h[g++]=w);for(let b=0;b<l;++b)y(this.leftPad),y(this.separator);for(let b=0;b<p-1;++b)y(t[m+b]),y(this.separator);if(p>0){y(t[m+p-1]);for(let b=0;b<c;++b)y(this.separator),y(this.rightPad)}else{for(let b=0;b<c-1;++b)y(this.rightPad),y(this.separator);y(this.rightPad)}}}compute(t,e){let n=t.length,o=e.length;if(o>0){let u=e[0];if(u!==0)throw new Error(`First split value must be 0, got ${u}`);for(let l=1;l<o;++l){let c=e[l]>=u;if(c=c&&e[l]<=n,!c)throw new Error(`Invalid split value ${e[l]}, must be in [${u}, ${n}]`);u=e[l]}if(u!==n)throw new Error(`Last split value must be data size. Expected ${n}, got ${u}`)}let s=o-1,i=x.getArrayFromDType("int32",o);if(n===0||o===0){let u=new Array(n);for(let l=0;l<=s;++l)i[l]=0;return[u,i]}i[0]=0;for(let u=1;u<=s;++u){let l=e[u]-e[u-1],c=0;this.nGramWidths.forEach(p=>{c+=this.getNumNGrams(l,p)}),this.preserveShort&&l>0&&c===0&&(c=1),i[u]=i[u-1]+c}let a=new Array(i[s]);for(let u=0;u<s;++u){let l=e[u],c=i[u];if(this.nGramWidths.forEach(p=>{let m=e[u+1]-e[u],f=this.getNumNGrams(m,p);this.createNGrams(t,l,a,c,f,p),c+=f}),this.preserveShort&&c===i[u]){let p=e[u+1]-e[u];if(p===0)continue;let m=p+2*this.padWidth,f=1;this.createNGrams(t,l,a,c,f,m)}}return[a,i]}};function Xy(r,t,e,n,o,s,i,a){return new ZR(e,n,o,s,i,a).compute(r,t)}function LZ(r,t,e,n){if(!r.length)return;if(t.length===0){for(let s=0;s<r.length;++s)n.push(r.subarray(s,s+1));return}if(t.length===1){let s=t[0],i=r.indexOf(s);for(;i!==-1;){let a=r.subarray(0,i);(!e||a.length!==0)&&n.push(a),r=r.subarray(i+1),i=r.indexOf(s)}(!e||r.length!==0)&&n.push(r);return}let o=0;for(let s=0;s<r.length+1;s++)if(s===r.length||t.indexOf(r[s])!==-1){let i=r.subarray(o,s);(!e||i.length!==0)&&n.push(i),o=s+1}}function Yy(r,t,e){let n=r.length,o=[],s=0,i=0,a=new Array(n);for(let m=0;m<n;++m){let f=o.length;LZ(r[m],t,e,o);let d=o.length-f;a[m]=d,s+=d,i=Math.max(i,d)}let u=x.getArrayFromDType("int32",s*2),l=new Array(s),c=[n,i],p=0;for(let m=0;m<n;++m)for(let f=0;f<a[m];++f)u[p*2]=m,u[p*2+1]=f,l[p]=o[p],++p;return[u,l,c]}function Zy(r,t){let e=x.getArrayFromDType("int32",r.length);for(let n=0;n<r.length;++n)e[n]=x.fingerPrint64(r[n]).modulo(t).getLowBitsUnsigned();return e}var kS=te((r,t)=>r-t),zZ=Om((r,t,e,n)=>({real:r-e,imag:t-n})),Yh=ie(Ls,kS,zZ),JR={kernelName:Ls,backendName:"cpu",kernelFunc:Yh};function Jy(r,t){let e=new Array(r.rank);for(let o=0;o<e.length;o++)e[o]=r.shape[o]*t[o];let n=It(e,r.dtype);for(let o=0;o<n.values.length;++o){let s=n.indexToLoc(o),i=new Array(r.rank);for(let u=0;u<i.length;u++)i[u]=s[u]%r.shape[u];let a=r.locToIndex(i);n.values[o]=r.values[a]}return n}var Zh=(r,t)=>{let e=t.value-r.value;return e===0?r.index-t.index:e};function QR(r,t,e=0,n=r.length-1){for(;n>e;){if(n-e>600){let a=n-e+1,u=t-e+1,l=Math.log(a),c=.5*Math.exp(2*l/3),p=.5*Math.sqrt(l*c*(a-c)/a)*Math.sign(u-a/2),m=Math.max(e,Math.floor(t-u*c/a+p)),f=Math.min(n,Math.floor(t+(a-u)*c/a+p));QR(r,t,m,f)}let o=r[t],s=e,i=n;for(x.swap(r,e,t),Zh(r[n],o)>0&&x.swap(r,e,n);s<i;){for(x.swap(r,s,i),s++,i--;Zh(r[s],o)<0;)s=s+1;for(;Zh(r[i],o)>0;)i=i-1}Zh(r[e],o)===0?x.swap(r,e,i):(i=i+1,x.swap(r,i,n)),i<=t&&(e=i+1),t<=i&&(n=i-1)}}function Qy(r,t,e,n,o){let s=t[t.length-1],[i,a]=[r.length/s,s],u=x.getTypedArrayFromDType(e,i*n),l=x.getTypedArrayFromDType("int32",i*n);for(let p=0;p<i;p++){let m=p*a,f=r.subarray(m,m+a),d=new Array(f.length);f.forEach((b,w)=>d[w]={value:b,index:w}),n<d.length&&(QR(d,n),d=d.slice(0,n)),o&&d.sort(Zh);let h=p*n,g=u.subarray(h,h+n),y=l.subarray(h,h+n);for(let b=0;b<n;b++)g[b]=d[b].value,y[b]=d[b].index}let c=t.slice();return c[c.length-1]=n,[It(c,e,u),It(c,"int32",l)]}function tb(r,t,e,n){let o=x.parseAxisParam(t,e)[0],s=[1,e[0],1];for(let d=0;d<o;d++)s[0]*=e[d];s[1]=e[o];for(let d=o+1;d<e.length;d++)s[2]*=e[d];let i={},a=new Int32Array(e[o]),u=new me(s,n,r),l=[],c=s[0]===1&&s[2]===1;for(let d=0;d<e[o];d++){let h;if(c)h=r[d].toString();else{let g=[];for(let y=0;y<s[0];y++)for(let b=0;b<s[2];b++)g.push(u.get(y,d,b));h=g.join(",")}if(i[h]!==void 0)a[d]=i[h];else{let g=Object.keys(i).length;i[h]=g,a[d]=g,l.push(d)}}let p=s.slice();p[1]=Object.keys(i).length;let m=new me(p,n);l.forEach((d,h)=>{for(let g=0;g<s[0];g++)for(let y=0;y<s[2];y++)m.set(u.get(g,d,y),g,h,y)});let f=e.slice();return f[o]=p[1],{outputValues:m.values,outputShape:f,indices:a}}Xp("cpu",()=>new Sc,1);var NS=Et(is,r=>r>=0?r:Math.exp(r)-1),tF={kernelName:is,backendName:"cpu",kernelFunc:NS};function TS(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{alpha:s}=n;nt([o],"leakyRelu");let i=x.sizeFromShape(o.shape),a=e.data.get(o.dataId).values,u=x.getTypedArrayFromDType("float32",i);for(let l=0;l<a.length;l++)u[l]=a[l]<0?s*a[l]:a[l];return e.makeTensorInfo(o.shape,"float32",u)}var eF={kernelName:ms,backendName:"cpu",kernelFunc:TS};var VZ=te((r,t)=>r<0?t*r:r);function _S(r){let{inputs:t,backend:e}=r,{x:n,alpha:o}=t;nt([n,o],"prelu");let s=e.data.get(n.dataId).values,i=e.data.get(o.dataId).values,[a,u]=VZ(n.shape,o.shape,s,i,"float32");return e.makeTensorInfo(u,"float32",a)}var rF={kernelName:ks,backendName:"cpu",kernelFunc:_S};var ES=Et(Ns,r=>Math.max(0,r)),nF={kernelName:Ns,backendName:"cpu",kernelFunc:ES};var AS=Et(_s,r=>Math.min(Math.max(0,r),6)),oF={kernelName:_s,backendName:"cpu",kernelFunc:AS};function zm(r,t,e,n,o){if(e==="linear")return Wr({inputs:{x:t},backend:r});if(e==="relu")return ES({inputs:{x:t},backend:r});if(e==="elu")return NS({inputs:{x:t},backend:r});if(e==="relu6")return AS({inputs:{x:t},backend:r});if(e==="prelu")return _S({inputs:{x:t,alpha:n},backend:r});if(e==="leakyrelu")return TS({inputs:{x:t},backend:r,attrs:{alpha:o}});if(e==="sigmoid")return IS({inputs:{x:t},backend:r});throw new Error(`Activation ${e} has not been implemented for the CPU backend.`)}function ne(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{shape:s}=n,i=x.sizeFromShape(o.shape),a=x.inferFromImplicitShape(s,i),u=x.sizeFromShape(a);x.assert(i===u,()=>`The new shape (${a}) has ${u} elements and the old shape (${o.shape}) has ${i} elements. The new shape and old shape must have the same number of elements.`),e.incRef(o.dataId);let l=e.data.get(o.dataId);if(l.complexTensorInfos!=null){let c=l.complexTensorInfos.real,p=l.complexTensorInfos.imag;c.shape=a,p.shape=a}return{dataId:o.dataId,shape:a,dtype:o.dtype}}var sF={kernelName:Ii,backendName:"cpu",kernelFunc:ne};function DS(r){let{inputs:t,backend:e,attrs:n}=r,{a:o,b:s}=t,{transposeA:i,transposeB:a}=n;nt([o,s],"matMul");let u=o.shape.length,l=s.shape.length,c=i?o.shape[u-2]:o.shape[u-1],p=a?s.shape[l-1]:s.shape[l-2],m=i?o.shape[u-1]:o.shape[u-2],f=a?s.shape[l-2]:s.shape[l-1],d=o.shape.slice(0,-2),h=s.shape.slice(0,-2),g=x.sizeFromShape(d),y=x.sizeFromShape(h),w=Mr.assertAndGetBroadcastShape(o.shape.slice(0,-2),s.shape.slice(0,-2)).concat([m,f]);x.assert(c===p,()=>`Error in matMul: inner shapes (${c}) and (${p}) of Tensors with shapes ${o.shape} and ${s.shape} and transposeA=${i} and transposeB=${a} must match.`);let v=i?[g,c,m]:[g,m,c],k=a?[y,f,p]:[y,p,f],E=ne({inputs:{x:o},backend:e,attrs:{shape:v}}),D=ne({inputs:{x:s},backend:e,attrs:{shape:k}}),$=i?E.shape[1]:E.shape[2],R=i?E.shape[2]:E.shape[1],M=a?D.shape[1]:D.shape[2],W=Math.max(g,y),U=e.data.get(E.dataId).values,q=e.data.get(D.dataId).values,K=x.computeStrides(E.shape),j=x.computeStrides(D.shape),[Q,rt,X]=i?[K[0],1,K[1]]:[K[0],K[1],1],[ot,st,it]=a?[1,j[1],j[0]]:[j[1],1,j[0]],ft=R*M,at=It([W,R,M],E.dtype),xt=at.values,dt=e.blockSize;for(let bt=0;bt<W;bt++)for(let kt=0;kt<R;kt+=dt)for(let At=0;At<M;At+=dt)for(let $t=0;$t<$;$t+=dt){let qt=Math.min(kt+dt,R),Kt=Math.min(At+dt,M),ce=Math.min($t+dt,$);for(let Ot=kt;Ot<qt;Ot++)for(let Ae=At;Ae<Kt;Ae++){let Ie=0;for(let ae=$t;ae<ce;ae++){let je=Math.min(bt,g-1)*Q,Fe=Math.min(bt,y-1)*it,Jr=U[je+Ot*rt+ae*X],ze=q[ae*ot+Ae*st+Fe];Ie+=Jr*ze}xt[bt*ft+(Ot*M+Ae)]+=Ie}}return e.disposeIntermediateTensorInfo(E),e.disposeIntermediateTensorInfo(D),e.makeTensorInfo(w,at.dtype,at.values)}var iF={kernelName:Zo,backendName:"cpu",kernelFunc:DS};function GZ(r){let{inputs:t,backend:e,attrs:n}=r,{a:o,b:s,bias:i,preluActivationWeights:a}=t,{transposeA:u,transposeB:l,activation:c,leakyreluAlpha:p}=n,m,f,d,h=[];m=DS({inputs:{a:o,b:s},attrs:{transposeA:u,transposeB:l},backend:e}),i&&(f=ll({inputs:{a:m,b:i},backend:e}),h.push(m),m=f),c&&(d=zm(e,m,c,a,p),h.push(m),m=d);for(let y of h)e.disposeIntermediateTensorInfo(y);return m}var aF={kernelName:Ai,backendName:"cpu",kernelFunc:GZ};var WZ=Et(ji,r=>Math.acos(r)),lF={kernelName:ji,backendName:"cpu",kernelFunc:WZ};var UZ=Et(Xi,r=>Math.acosh(r)),uF={kernelName:Xi,backendName:"cpu",kernelFunc:UZ};function HZ(r){let{inputs:t,backend:e}=r,n=t;nt(t,"addN");let o=n.map(a=>e.data.get(a.dataId).values),s=It(n[0].shape,n[0].dtype),i=s.values;for(let a=0;a<n.length;a++){let u=o[a];for(let l=0;l<i.length;l++)i[l]+=u[l]}return e.makeTensorInfo(s.shape,s.dtype,s.values)}var cF={kernelName:jo,backendName:"cpu",kernelFunc:HZ};function qZ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s,keepDims:i}=n;nt(o,"all");let a=x.parseAxisParam(s,o.shape),u=a,l=S.getAxesPermutation(u,o.shape.length),c=o;l!=null&&(c=He({inputs:{x:o},backend:e,attrs:{perm:l}}),u=S.getInnerMostAxes(u.length,o.shape.length)),S.assertAxesAreInnerMostDims("all",u,c.shape.length);let[p,m]=S.computeOutAndReduceShapes(c.shape,u),f=x.sizeFromShape(m),d=x.makeZerosTypedArray(x.sizeFromShape(p),c.dtype),h=e.data.get(c.dataId).values;for(let y=0;y<d.length;++y){let b=y*f,w=h[b];for(let v=0;v<f;++v){let k=h[b+v];w=w&&k}d[y]=w}l!=null&&e.disposeIntermediateTensorInfo(c);let g=e.makeTensorInfo(p,c.dtype,d);if(i){let y=S.expandShapeToKeepDim(p,a),b=ne({inputs:{x:g},backend:e,attrs:{shape:y}});return e.disposeIntermediateTensorInfo(g),b}return g}var pF={kernelName:Yi,backendName:"cpu",kernelFunc:qZ};function KZ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s,keepDims:i}=n;nt(o,"any");let a=x.parseAxisParam(s,o.shape),u=a,l=S.getAxesPermutation(u,o.shape.length),c=o;l!=null&&(c=He({inputs:{x:o},backend:e,attrs:{perm:l}}),u=S.getInnerMostAxes(u.length,o.shape.length)),S.assertAxesAreInnerMostDims("any",u,c.shape.length);let[p,m]=S.computeOutAndReduceShapes(c.shape,u),f=x.sizeFromShape(m),d=x.makeZerosTypedArray(x.sizeFromShape(p),c.dtype),h=e.data.get(c.dataId).values;for(let y=0;y<d.length;++y){let b=y*f,w=h[b];for(let v=0;v<f;++v){let k=h[b+v];w=w||k}d[y]=w}l!=null&&e.disposeIntermediateTensorInfo(c);let g=e.makeTensorInfo(p,c.dtype,d);if(i){let y=S.expandShapeToKeepDim(p,a),b=ne({inputs:{x:g},backend:e,attrs:{shape:y}});return e.disposeIntermediateTensorInfo(g),b}return g}var mF={kernelName:Zi,backendName:"cpu",kernelFunc:KZ};function jZ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s}=n;nt(o,"argMax");let i=x.parseAxisParam(s,o.shape),a=S.getAxesPermutation(i,o.shape.length),u=o,l=[];a!=null&&(u=He({inputs:{x:o},backend:e,attrs:{perm:a}}),l.push(u),i=S.getInnerMostAxes(i.length,u.shape.length)),i=[i[0]],S.assertAxesAreInnerMostDims("argMax",i,u.shape.length);let[c,p]=S.computeOutAndReduceShapes(u.shape,i),m=x.sizeFromShape(c),f=x.makeZerosTypedArray(m,"int32"),d=x.sizeFromShape(p),h=e.data.get(u.dataId).values;for(let g=0;g<f.length;++g){let y=g*d,b=h[y],w=0;for(let v=0;v<d;++v){let k=h[y+v];k>b&&(b=k,w=v)}f[g]=w}return l.forEach(g=>e.disposeIntermediateTensorInfo(g)),e.makeTensorInfo(c,"int32",f)}var fF={kernelName:Xo,backendName:"cpu",kernelFunc:jZ};function XZ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s}=n;nt(o,"argMin");let i=x.parseAxisParam(s,o.shape),a=S.getAxesPermutation(i,o.shape.length),u=o,l=[];a!=null&&(u=He({inputs:{x:o},backend:e,attrs:{perm:a}}),l.push(u),i=S.getInnerMostAxes(i.length,u.shape.length)),i=[i[0]],S.assertAxesAreInnerMostDims("argMin",i,u.shape.length);let[c,p]=S.computeOutAndReduceShapes(u.shape,i),m=x.sizeFromShape(c),f=x.makeZerosTypedArray(m,"int32"),d=x.sizeFromShape(p),h=e.data.get(u.dataId).values;for(let g=0;g<f.length;++g){let y=g*d,b=h[y],w=0;for(let v=0;v<d;++v){let k=h[y+v];k<b&&(b=k,w=v)}f[g]=w}return l.forEach(g=>e.disposeIntermediateTensorInfo(g)),e.makeTensorInfo(c,"int32",f)}var dF={kernelName:vl,backendName:"cpu",kernelFunc:XZ};var YZ=Et(Ji,r=>Math.asin(r)),hF={kernelName:Ji,backendName:"cpu",kernelFunc:YZ};var ZZ=Et(Qi,r=>Math.asinh(r)),gF={kernelName:Qi,backendName:"cpu",kernelFunc:ZZ};var JZ=Et(ta,r=>Math.atan(r)),xF={kernelName:ta,backendName:"cpu",kernelFunc:JZ};var QZ=te((r,t)=>Math.atan2(r,t)),t9=ie(ra,QZ),yF={kernelName:ra,backendName:"cpu",kernelFunc:t9};var e9=Et(ea,r=>Math.atanh(r)),bF={kernelName:ea,backendName:"cpu",kernelFunc:e9};function Bm(r,t,e,n,o,s){let i=o.strideHeight,a=o.strideWidth,u=o.dilationHeight,l=o.dilationWidth,c=o.effectiveFilterHeight,p=o.effectiveFilterWidth,m=o.padInfo.top,f=o.padInfo.left,d=s==="max"?Number.NEGATIVE_INFINITY:Number.POSITIVE_INFINITY,h=It(o.outShape,e),g=h.values,y=o.outShape[1]*o.outShape[2]*o.outShape[3],b=o.outShape[2]*o.outShape[3],w=o.outShape[3];for(let v=0;v<o.batchSize;++v){let k=v*y,E=v*n[0];for(let D=0;D<o.inChannels;++D)for(let $=0;$<o.outHeight;++$){let R=$*i-m,M=Math.max(0,R),W=Math.min(o.inHeight,c+R),U=k+$*b;for(let q=0;q<o.outWidth;++q){let K=q*a-f,j=Math.max(0,K),Q=Math.min(o.inWidth,p+K),rt=d,X=0,ot=0;for(let it=M;it<W;it+=u){let ft=E+it*n[1];for(let at=j;at<Q;at+=l){let xt=ft+at*n[2],dt=r[xt+D];s==="max"&&dt>rt?rt=dt:s==="avg"&&(X+=dt,ot++)}if(isNaN(rt))break}let st=U+q*w+D;g[st]=s==="avg"?X/ot:rt}}}return h}function rb(r,t,e,n,o=!1,s=!1){let i=It(n.outShape,"int32"),a=n.strideHeight,u=n.strideWidth,l=n.dilationHeight,c=n.dilationWidth,p=n.effectiveFilterHeight,m=n.effectiveFilterWidth,f=n.padInfo.top,d=n.padInfo.left,h=It(t,e,r);for(let g=0;g<n.batchSize;++g)for(let y=0;y<n.inChannels;++y)for(let b=0;b<n.outHeight;++b){let w=b*a-f,v=w;for(;v<0;)v+=l;let k=Math.min(n.inHeight,p+w);for(let E=0;E<n.outWidth;++E){let D=E*u-d,$=D;for(;$<0;)$+=c;let R=Math.min(n.inWidth,m+D),M=Number.NEGATIVE_INFINITY,W=-1;for(let U=v;U<k;U+=l){let q=U-w;for(let K=$;K<R;K+=c){let j=K-D,Q=h.get(g,U,K,y);Q>M&&(M=Q,o?W=s?((g*n.inHeight+U)*n.inWidth+K)*n.inChannels+y:(U*n.inWidth+K)*n.inChannels+y:W=q*m+j)}}i.set(W,g,b,E,y)}}return i}function nb(r,t,e,n,o,s){let i=o.strideDepth,a=o.strideHeight,u=o.strideWidth,l=o.dilationDepth,c=o.dilationHeight,p=o.dilationWidth,m=o.effectiveFilterDepth,f=o.effectiveFilterHeight,d=o.effectiveFilterWidth,h=o.padInfo.front,g=o.padInfo.top,y=o.padInfo.left,b=s==="max"?Number.NEGATIVE_INFINITY:Number.POSITIVE_INFINITY,w=It(o.outShape,e),v=w.values,k=o.outShape[1]*o.outShape[2]*o.outShape[3]*o.outShape[4],E=o.outShape[2]*o.outShape[3]*o.outShape[4],D=o.outShape[3]*o.outShape[4],$=o.outShape[4];for(let R=0;R<o.batchSize;++R){let M=R*k,W=R*n[0];for(let U=0;U<o.inChannels;++U)for(let q=0;q<o.outDepth;++q){let K=q*i-h,j=K;for(;j<0;)j+=l;let Q=Math.min(o.inDepth,m+K),rt=M+q*E;for(let X=0;X<o.outHeight;++X){let ot=X*a-g,st=ot;for(;st<0;)st+=c;let it=Math.min(o.inHeight,f+ot),ft=rt+X*D;for(let at=0;at<o.outWidth;++at){let xt=at*u-y,dt=xt;for(;dt<0;)dt+=p;let bt=Math.min(o.inWidth,d+xt),kt=ft+at*$,At=b,$t=0,qt=0;for(let ce=j;ce<Q;ce+=l){let Ot=W+ce*n[1];for(let Ae=st;Ae<it;Ae+=c){let Ie=Ot+Ae*n[2];for(let ae=dt;ae<bt;ae+=p){let je=Ie+ae*n[3],Fe=r[je+U];if(s==="max"&&Fe>At?At=Fe:s==="avg"&&($t+=Fe,qt++),isNaN(At))break}if(isNaN(At))break}if(isNaN(At))break}let Kt=kt+U;v[Kt]=s==="avg"?$t/qt:At}}}}return w}function wF(r,t){let e=It(t.outShape,"int32"),n=t.strideDepth,o=t.strideHeight,s=t.strideWidth,i=t.dilationDepth,a=t.dilationHeight,u=t.dilationWidth,l=t.effectiveFilterDepth,c=t.effectiveFilterHeight,p=t.effectiveFilterWidth,m=t.padInfo.front,f=t.padInfo.top,d=t.padInfo.left;for(let h=0;h<t.batchSize;++h)for(let g=0;g<t.inChannels;++g)for(let y=0;y<t.outDepth;++y){let b=y*n-m,w=b;for(;w<0;)w+=i;let v=Math.min(t.inDepth,l+b);for(let k=0;k<t.outHeight;++k){let E=k*o-f,D=E;for(;D<0;)D+=a;let $=Math.min(t.inHeight,c+E);for(let R=0;R<t.outWidth;++R){let M=R*s-d,W=M;for(;W<0;)W+=u;let U=Math.min(t.inWidth,p+M),q=Number.NEGATIVE_INFINITY,K=-1;for(let j=w;j<v;j+=i){let Q=j-b;for(let rt=D;rt<$;rt+=a){let X=rt-E;for(let ot=W;ot<U;ot+=u){let st=ot-M,it=r.get(h,j,rt,ot,g);it>=q&&(q=it,K=Q*c*p+X*c+st)}}}e.set(K,h,y,k,R,g)}}}return e}function r9(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t;nt(o,"avgPool");let{filterSize:s,strides:i,pad:a,dimRoundingMode:u}=n,l=1;x.assert(S.eitherStridesOrDilationsAreOne(i,l),()=>`Error in avgPool: Either strides or dilations must be 1. Got strides ${i} and dilations '${l}'`);let c=S.computePool2DInfo(o.shape,s,i,l,a,u),p;if(c.filterWidth===1&&c.filterHeight===1&&x.arraysEqual(c.inShape,c.outShape))p=Wr({inputs:{x:o},backend:e});else{let m=e.data.get(o.dataId).values,f=x.computeStrides(o.shape),d=Bm(m,o.shape,o.dtype,f,c,"avg");p=e.makeTensorInfo(c.outShape,o.dtype,d.values)}return p}var vF={kernelName:Yo,backendName:"cpu",kernelFunc:r9};function n9(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{filterSize:s,strides:i,pad:a,dimRoundingMode:u,dataFormat:l}=n;nt(o,"avgPool3d");let c=S.computePool3DInfo(o.shape,s,i,1,a,u,l),p=e.data.get(o.dataId).values,m=nb(p,o.shape,o.dtype,x.computeStrides(o.shape),c,"avg");return e.makeTensorInfo(m.shape,"float32",m.values)}var CF={kernelName:Cl,backendName:"cpu",kernelFunc:n9};function o9(r){let{inputs:t,backend:e,attrs:n}=r,{dy:o,input:s}=t,{filterSize:i,strides:a,pad:u,dimRoundingMode:l}=n;nt([o,s],"avgPool3DGrad");let c=S.computePool3DInfo(s.shape,i,a,1,u,l),p=c.strideDepth,m=c.strideHeight,f=c.strideWidth,d=c.filterDepth,h=c.filterHeight,g=c.filterWidth,y=c.dilationDepth,b=c.dilationHeight,w=c.dilationWidth,v=c.effectiveFilterDepth,k=c.effectiveFilterHeight,E=c.effectiveFilterWidth,D=v-1-c.padInfo.front,$=E-1-c.padInfo.left,R=k-1-c.padInfo.top,M=It(s.shape,"float32"),W=1/(d*h*g),U=e.bufferSync(o);for(let q=0;q<c.batchSize;++q)for(let K=0;K<c.inChannels;++K)for(let j=0;j<c.inDepth;++j)for(let Q=0;Q<c.inHeight;++Q)for(let rt=0;rt<c.inWidth;++rt){let X=j-D,ot=Q-R,st=rt-$,it=0;for(let ft=0;ft<v;ft+=y){let at=(X+ft)/p;if(!(at<0||at>=c.outDepth||Math.floor(at)!==at))for(let xt=0;xt<k;xt+=b){let dt=(ot+xt)/m;if(!(dt<0||dt>=c.outHeight||Math.floor(dt)!==dt))for(let bt=0;bt<E;bt+=w){let kt=(st+bt)/f;if(kt<0||kt>=c.outWidth||Math.floor(kt)!==kt)continue;it+=U.get(q,at,dt,kt,K)}}}M.set(it*W,q,j,Q,rt,K)}return e.makeTensorInfo(M.shape,M.dtype,M.values)}var IF={kernelName:up,backendName:"cpu",kernelFunc:o9};function s9(r){let{inputs:t,backend:e,attrs:n}=r,{dy:o,input:s}=t,i=s;nt([o,s],"avgPoolGrad");let{filterSize:a,strides:u,pad:l}=n,c=S.computePool2DInfo(i.shape,a,u,1,l),p=c.strideHeight,m=c.strideWidth,f=c.filterHeight,d=c.filterWidth,h=c.dilationHeight,g=c.dilationWidth,y=c.effectiveFilterHeight,b=c.effectiveFilterWidth,w=b-1-c.padInfo.left,v=y-1-c.padInfo.top,k=It(i.shape,"float32"),E=1/(f*d),D=e.data.get(o.dataId).values,$=It(o.shape,"float32",D);for(let R=0;R<c.batchSize;++R)for(let M=0;M<c.inChannels;++M)for(let W=0;W<c.inHeight;++W)for(let U=0;U<c.inWidth;++U){let q=W-v,K=U-w,j=0;for(let Q=0;Q<y;Q+=h){let rt=(q+Q)/p;if(!(rt<0||rt>=c.outHeight||Math.floor(rt)!==rt))for(let X=0;X<b;X+=g){let ot=(K+X)/m;if(ot<0||ot>=c.outWidth||Math.floor(ot)!==ot)continue;j+=$.get(R,rt,ot,M)}}k.set(j*E,R,W,U,M)}return e.makeTensorInfo(k.shape,k.dtype,k.values)}var SF={kernelName:lp,backendName:"cpu",kernelFunc:s9};function i9(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,scale:s,offset:i,mean:a,variance:u}=t;x.assert(a.shape.length===u.shape.length,()=>"Batch normalization gradient requires mean and variance to have equal ranks."),x.assert(i==null||a.shape.length===i.shape.length,()=>"Batch normalization gradient requires mean and offset to have equal ranks."),x.assert(s==null||a.shape.length===s.shape.length,()=>"Batch normalization gradient requires mean and scale to have equal ranks."),nt([o,a,u,s,i],"batchNorm");let{varianceEpsilon:l}=n;l==null&&(l=.001);let c=e.data.get(o.dataId).values,p=e.data.get(a.dataId).values,m=e.data.get(u.dataId).values,f=s?e.data.get(s.dataId).values:new Float32Array([1]),d=i?e.data.get(i.dataId).values:new Float32Array([0]),h=new Float32Array(c.length),g=d.length,y=f.length,b=m.length,w=p.length,v=0,k=0,E=0,D=0;for(let $=0;$<c.length;++$)h[$]=d[v++]+(c[$]-p[k++])*f[E++]/Math.sqrt(m[D++]+l),v>=g&&(v=0),k>=w&&(k=0),E>=y&&(E=0),D>=b&&(D=0);return e.makeTensorInfo(o.shape,o.dtype,h)}var kF={kernelName:cs,backendName:"cpu",kernelFunc:i9};function a9(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{blockShape:s,crops:i}=n;nt([o],"batchToSpaceND");let a=s.reduce((y,b)=>y*b),u=S.getReshaped(o.shape,s,a),l=S.getPermuted(u.length,s.length),c=S.getReshapedPermuted(o.shape,s,a),p=S.getSliceBeginCoords(i,s.length),m=S.getSliceSize(c,i,s.length),f=ne({inputs:{x:o},backend:e,attrs:{shape:u}}),d=He({inputs:{x:f},backend:e,attrs:{perm:l}}),h=ne({inputs:{x:d},backend:e,attrs:{shape:c}}),g=Ro({inputs:{x:h},backend:e,attrs:{begin:p,size:m}});return e.disposeIntermediateTensorInfo(f),e.disposeIntermediateTensorInfo(d),e.disposeIntermediateTensorInfo(h),g}var NF={kernelName:gi,backendName:"cpu",kernelFunc:a9};function l9(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,weights:s}=t,{size:i}=n,a=e.data.get(o.dataId).values,u=e.data.get(s.dataId).values,l=Pm(a,u,s.dtype,s.shape,i);return e.makeTensorInfo([i],s.dtype,l)}var TF={kernelName:cp,backendName:"cpu",kernelFunc:l9};function u9(r){let{inputs:t,backend:e}=r,{s0:n,s1:o}=t,s=e.data.get(n.dataId).values,i=e.data.get(o.dataId).values,a=S.assertAndGetBroadcastShape(Array.from(s),Array.from(i));return e.makeTensorInfo([a.length],"int32",Int32Array.from(a))}var _F={kernelName:pp,backendName:"cpu",kernelFunc:u9};var c9=Et(so,(r,t)=>{let e=t;return r>e.clipValueMax?e.clipValueMax:r<e.clipValueMin?e.clipValueMin:r}),EF={kernelName:so,backendName:"cpu",kernelFunc:c9};var p9=r=>{let{x:t}=r.inputs,e=r.backend,n=new Float32Array(x.sizeFromShape(t.shape)),o=e.data.get(t.dataId),s=o.complexTensorInfos.real,i=o.complexTensorInfos.imag,a=e.data.get(s.dataId).values,u=e.data.get(i.dataId).values;for(let l=0;l<a.length;l++){let c=a[l],p=u[l];n[l]=Math.hypot(c,p)}return e.makeOutput(n,t.shape,"float32")},AF={kernelName:Il,backendName:"cpu",kernelFunc:p9};function Ui(r){let{inputs:t,backend:e}=r,{input:n}=t,o=e.data.get(n.dataId).complexTensorInfos.imag,s=e.data.get(o.dataId).values;return e.makeTensorInfo(o.shape,o.dtype,s)}var DF={kernelName:Sp,backendName:"cpu",kernelFunc:Ui};function nu(r){let{inputs:t,backend:e,attrs:n}=r,{axis:o}=n,s=x.parseAxisParam(o,t[0].shape)[0],i=S.computeOutShape(t.map(h=>h.shape),s);if(x.sizeFromShape(i)===0)return e.makeTensorInfo(i,t[0].dtype,[]);let a=t.filter(h=>x.sizeFromShape(h.shape)>0);if(a.length===1)return Wr({inputs:{x:a[0]},backend:e});let u=a.map(h=>h.shape);if(S.assertParamsConsistent(u,s),a[0].dtype==="complex64"){let h=a.map(v=>Ao({inputs:{input:v},backend:e})),g=a.map(v=>Ui({inputs:{input:v},backend:e})),y=nu({inputs:h,backend:e,attrs:{axis:s}}),b=nu({inputs:g,backend:e,attrs:{axis:s}}),w=Ir({inputs:{real:y,imag:b},backend:e});return h.forEach(v=>e.disposeIntermediateTensorInfo(v)),g.forEach(v=>e.disposeIntermediateTensorInfo(v)),e.disposeIntermediateTensorInfo(y),e.disposeIntermediateTensorInfo(b),w}let l=a.map(h=>{let g=x.sizeFromShape(h.shape.slice(s));return ne({inputs:{x:h},backend:e,attrs:{shape:[-1,g]}})}),c=l.map(h=>({vals:e.data.get(h.dataId).values,shape:h.shape}));i=S.computeOutShape(l.map(h=>h.shape),1);let p=l[0].shape[0]===1,m=kc(c,i,t[0].dtype,p),f=S.computeOutShape(a.map(h=>h.shape),s),d=e.makeTensorInfo(f,t[0].dtype,m);return l.forEach(h=>e.disposeIntermediateTensorInfo(h)),d}var $F={kernelName:xi,backendName:"cpu",kernelFunc:nu};function $S(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,filter:s}=t,{strides:i,pad:a,dataFormat:u,dilations:l,dimRoundingMode:c}=n;nt([o,s],"conv2d");let p=S.convertConv2DDataFormat(u),m=S.computeConv2DInfo(o.shape,s.shape,i,l,a,c,!1,p),f=m.filterHeight,d=m.filterWidth,h=m.dilationHeight,g=m.dilationWidth,y=m.padInfo.left,b=m.padInfo.top,w=m.dataFormat==="channelsLast",v=new me(m.outShape,o.dtype),k=x.computeStrides(o.shape),E=x.computeStrides(s.shape),D=k[0],$=w?k[1]:k[2],R=w?k[2]:1,M=w?1:k[1],W=v.strides[0],U=w?v.strides[1]:v.strides[2],q=w?v.strides[2]:1,K=w?1:v.strides[1],j=e.data.get(o.dataId).values,Q=e.data.get(s.dataId).values,rt=v.values;for(let X=0;X<m.batchSize;++X){let ot=X*D,st=X*W;for(let it=0;it<m.outHeight;++it){let ft=st+it*U,at=it*m.strideHeight-b;for(let xt=0;xt<f;++xt){let dt=at+xt*h;if(dt<0||dt>=m.inHeight)continue;let bt=xt*E[0],kt=ot+dt*$;for(let At=0;At<m.outWidth;++At){let $t=ft+At*q,qt=At*m.strideWidth-y;for(let Kt=0;Kt<d;++Kt){let ce=qt+Kt*g;if(ce<0||ce>=m.inWidth)continue;let Ot=bt+Kt*E[1],Ae=kt+ce*R,Ie=Ot;for(let ae=0;ae<m.inChannels;++ae){let je=j[Ae+ae*M];for(let Fe=0;Fe<m.outChannels;++Fe)rt[$t+Fe*K]+=je*Q[Ie+Fe];Ie+=m.outChannels}}}}}}return e.makeTensorInfo(v.shape,v.dtype,rt)}var RF={kernelName:Qo,backendName:"cpu",kernelFunc:$S};function m9(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,dy:s}=t,{strides:i,pad:a,dataFormat:u,dimRoundingMode:l,filterShape:c}=n;nt([o,s],"conv2dBackpropFilter");let p=S.convertConv2DDataFormat(u),m=S.computeConv2DInfo(o.shape,c,i,1,a,l,!1,p),{strideHeight:f,strideWidth:d,filterHeight:h,filterWidth:g}=m,y=m.dataFormat==="channelsLast",b=new me(m.filterShape,"float32"),w=m.padInfo.left,v=m.padInfo.top,k=e.data.get(o.dataId).values,E=e.data.get(s.dataId).values,D=new me(o.shape,o.dtype,k),$=new me(s.shape,s.dtype,E);for(let R=0;R<h;++R){let M=Math.max(0,Math.ceil((v-R)/f)),W=Math.min(m.outHeight,(m.inHeight+v-R)/f);for(let U=0;U<g;++U){let q=Math.max(0,Math.ceil((w-U)/d)),K=Math.min(m.outWidth,(m.inWidth+w-U)/d);for(let j=0;j<m.inChannels;++j)for(let Q=0;Q<m.outChannels;++Q){let rt=0;for(let X=0;X<m.batchSize;++X)for(let ot=M;ot<W;++ot){let st=R+ot*f-v;for(let it=q;it<K;++it){let ft=U+it*d-w;y?rt+=D.get(X,st,ft,j)*$.get(X,ot,it,Q):rt+=D.get(X,j,st,ft)*$.get(X,Q,ot,it)}}b.set(rt,R,U,j,Q)}}}return e.makeTensorInfo(b.shape,b.dtype,b.values)}var FF={kernelName:fp,backendName:"cpu",kernelFunc:m9};function f9(r){let{inputs:t,backend:e,attrs:n}=r,{dy:o,filter:s}=t,{inputShape:i,strides:a,pad:u,dataFormat:l,dimRoundingMode:c}=n;nt([o,s],"conv2dBackpropInput");let p=x.computeStrides(s.shape),m=x.computeStrides(o.shape),f=S.convertConv2DDataFormat(l),d=S.computeConv2DInfo(i,s.shape,a,1,u,c,!1,f),h=new me(d.inShape,"float32"),g=h.values,y=e.data.get(o.dataId).values,b=e.data.get(s.dataId).values,[w,v,k]=p,{batchSize:E,filterHeight:D,filterWidth:$,inChannels:R,inHeight:M,inWidth:W,outChannels:U,outHeight:q,outWidth:K,strideHeight:j,strideWidth:Q}=d;f=d.dataFormat;let rt=D-1-d.padInfo.top,X=$-1-d.padInfo.left,ot=f==="channelsLast",st=h.strides[0],it=ot?h.strides[1]:h.strides[2],ft=ot?h.strides[2]:1,at=ot?1:h.strides[1],xt=m[0],dt=ot?m[1]:m[2],bt=ot?m[2]:1,kt=ot?1:m[1];for(let At=0;At<E;++At)for(let $t=0;$t<R;++$t)for(let qt=0;qt<M;++qt){let Kt=qt-rt,ce=Math.max(0,Math.ceil(Kt/j)),Ot=Math.min(q,(D+Kt)/j);for(let Ae=0;Ae<W;++Ae){let Ie=Ae-X,ae=Math.max(0,Math.ceil(Ie/Q)),je=Math.min(K,($+Ie)/Q),Fe=0;for(let ze=ce;ze<Ot;++ze){let Lo=ze*j-Kt;for(let Fr=ae;Fr<je;++Fr){let Qr=Fr*Q-Ie,tn=xt*At+dt*ze+bt*Fr,Hr=w*(D-1-Lo)+v*($-1-Qr)+k*$t;for(let ro=0;ro<U;++ro){let $n=y[tn+kt*ro],zo=b[Hr+ro];Fe+=$n*zo}}}let Jr=st*At+it*qt+ft*Ae+at*$t;g[Jr]=Fe}}return e.makeTensorInfo(h.shape,h.dtype,h.values)}var OF={kernelName:ts,backendName:"cpu",kernelFunc:f9};function d9(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,filter:s}=t,{strides:i,pad:a,dilations:u}=n;nt([o,s],"conv3d");let l=S.computeConv3DInfo(o.shape,s.shape,i,u,a),{filterDepth:c,filterHeight:p,filterWidth:m,dilationDepth:f,dilationHeight:d,dilationWidth:h,padInfo:g}=l,y=g.front,b=g.left,w=g.top,v=new me(l.outShape,o.dtype),k=e.data.get(o.dataId).values,E=e.data.get(s.dataId).values,D=v.values,$=x.computeStrides(o.shape),R=x.computeStrides(s.shape);for(let M=0;M<l.batchSize;++M){let W=M*$[0],U=M*v.strides[0];for(let q=0;q<l.outDepth;++q){let K=U+q*v.strides[1],j=q*l.strideDepth-y;for(let Q=0;Q<c;++Q){let rt=j+Q*f;if(rt<0||rt>=l.inDepth)continue;let X=Q*R[0],ot=W+rt*$[1];for(let st=0;st<l.outHeight;++st){let it=K+st*v.strides[2],ft=st*l.strideHeight-w;for(let at=0;at<p;++at){let xt=ft+at*d;if(xt<0||xt>=l.inHeight)continue;let dt=X+at*R[1],bt=ot+xt*$[2];for(let kt=0;kt<l.outWidth;++kt){let At=it+kt*l.outChannels,$t=kt*l.strideWidth-b;for(let qt=0;qt<m;++qt){let Kt=$t+qt*h;if(Kt<0||Kt>=l.inWidth)continue;let ce=dt+qt*R[2],Ot=bt+Kt*l.inChannels,Ae=ce;for(let Ie=0;Ie<l.inChannels;++Ie){let ae=k[Ot+Ie];for(let je=0;je<l.outChannels;++je)D[At+je]+=ae*E[Ae+je];Ae+=l.outChannels}}}}}}}}return e.makeTensorInfo(v.shape,v.dtype,v.values)}var PF={kernelName:Sl,backendName:"cpu",kernelFunc:d9};function h9(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,dy:s}=t,{strides:i,pad:a,filterShape:u}=n;nt([o,s],"conv3dBackpropFilterV2");let l=x.computeStrides(o.shape),c=x.computeStrides(s.shape),p=S.computeConv3DInfo(o.shape,u,i,1,a),m=p.strideDepth,f=p.strideHeight,d=p.strideWidth,h=p.filterDepth,g=p.filterHeight,y=p.filterWidth,b=new me(p.filterShape,"float32"),w=b.values,[v,k,E,D]=b.strides,$=e.data.get(s.dataId).values,[R,M,W,U]=c,q=e.data.get(o.dataId).values,[K,j,Q,rt]=l,X=p.padInfo.front,ot=p.padInfo.left,st=p.padInfo.top;for(let it=0;it<h;++it){let ft=Math.max(0,Math.ceil((X-it)/m)),at=Math.min(p.outDepth,(p.inDepth+X-it)/m),xt=it*v;for(let dt=0;dt<g;++dt){let bt=Math.max(0,Math.ceil((st-dt)/f)),kt=Math.min(p.outHeight,(p.inHeight+st-dt)/f),At=dt*k+xt;for(let $t=0;$t<y;++$t){let qt=Math.max(0,Math.ceil((ot-$t)/d)),Kt=Math.min(p.outWidth,(p.inWidth+ot-$t)/d),ce=$t*E+At;for(let Ot=0;Ot<p.inChannels;++Ot){let Ae=Ot*D+ce;for(let Ie=0;Ie<p.outChannels;++Ie){let ae=0;for(let je=0;je<p.batchSize;++je){let Fe=je*K,Jr=je*R;for(let ze=ft;ze<at;++ze){let Fr=(it+ze*m-X)*j+Fe,Qr=ze*M+Jr;for(let tn=bt;tn<kt;++tn){let ro=(dt+tn*f-st)*Q+Fr,$n=tn*W+Qr;for(let zo=qt;zo<Kt;++zo){let gl=($t+zo*d-ot)*rt+ro,cu=zo*U+$n;ae+=q[gl+Ot]*$[cu+Ie]}}}}w[Ae+Ie]=ae}}}}}return e.makeTensorInfo(b.shape,b.dtype,b.values)}var MF={kernelName:dp,backendName:"cpu",kernelFunc:h9};function g9(r){let{inputs:t,backend:e,attrs:n}=r,{dy:o,filter:s}=t,{pad:i,strides:a,inputShape:u}=n;nt([o],"conv3dBackpropInputV2");let l=x.computeStrides(o.shape),c=x.computeStrides(s.shape),p=S.computeConv3DInfo(u,s.shape,a,1,i),m=new me(p.inShape,"float32"),f=m.values,[d,h,g,y]=m.strides,b=e.data.get(o.dataId).values,[w,v,k,E]=l,D=e.data.get(s.dataId).values,[$,R,M,W]=c,{batchSize:U,filterDepth:q,filterHeight:K,filterWidth:j,inChannels:Q,inDepth:rt,inHeight:X,inWidth:ot,outChannels:st,outDepth:it,outHeight:ft,outWidth:at,strideDepth:xt,strideHeight:dt,strideWidth:bt}=p,kt=q-1-p.padInfo.front,At=K-1-p.padInfo.top,$t=j-1-p.padInfo.left;for(let qt=0;qt<U;++qt)for(let Kt=0;Kt<Q;++Kt)for(let ce=0;ce<rt;++ce){let Ot=ce-kt,Ae=Math.max(0,Math.ceil(Ot/xt)),Ie=Math.min(it,(q+Ot)/xt);for(let ae=0;ae<X;++ae){let je=ae-At,Fe=Math.max(0,Math.ceil(je/dt)),Jr=Math.min(ft,(K+je)/dt);for(let ze=0;ze<ot;++ze){let Lo=ze-$t,Fr=Math.max(0,Math.ceil(Lo/bt)),Qr=Math.min(at,(j+Lo)/bt),tn=0;for(let Hr=Ae;Hr<Ie;++Hr){let ro=Hr*xt-Ot;for(let $n=Fe;$n<Jr;++$n){let zo=$n*dt-je;for(let no=Fr;no<Qr;++no){let gl=no*bt-Lo,cu=w*qt+v*Hr+k*$n+E*no,Wc=$*(q-1-ro)+R*(K-1-zo)+M*(j-1-gl)+W*Kt;for(let gr=0;gr<st;++gr){let Uc=b[cu+gr],Hc=D[Wc+gr];tn+=Uc*Hc}}}}f[d*qt+h*ce+g*ae+y*ze+Kt]=tn}}}return e.makeTensorInfo(m.shape,m.dtype,m.values)}var LF={kernelName:hp,backendName:"cpu",kernelFunc:g9};var x9=Et(es,r=>Math.cos(r)),zF={kernelName:es,backendName:"cpu",kernelFunc:x9};var y9=Et(rs,r=>Math.cosh(r)),BF={kernelName:rs,backendName:"cpu",kernelFunc:y9};function b9(r){let{inputs:t,backend:e,attrs:n}=r,{image:o,boxes:s,boxInd:i}=t,{cropSize:a,method:u,extrapolationValue:l}=n,[c,p,m,f]=o.shape,d=s.shape[0],[h,g]=a,y=It([d,h,g,f],"float32"),b=e.data.get(s.dataId).values,w=e.data.get(i.dataId).values,v=e.data.get(o.dataId).values,k=x.computeStrides(o.shape),E=x.computeStrides(y.shape);for(let D=0;D<d;D++){let $=D*4,R=b[$],M=b[$+1],W=b[$+2],U=b[$+3],q=w[D];if(q>=c)continue;let K=h>1?(W-R)*(p-1)/(h-1):0,j=g>1?(U-M)*(m-1)/(g-1):0;for(let Q=0;Q<h;Q++){let rt=h>1?R*(p-1)+Q*K:.5*(R+W)*(p-1);if(rt<0||rt>p-1){for(let X=0;X<g;X++)for(let ot=0;ot<f;ot++){let st=ot+X*E[2]+Q*E[1]+D*E[0];y.values[st]=l}continue}if(u==="bilinear"){let X=Math.floor(rt),ot=Math.ceil(rt),st=rt-X;for(let it=0;it<g;it++){let ft=g>1?M*(m-1)+it*j:.5*(M+U)*(m-1);if(ft<0||ft>m-1){for(let bt=0;bt<f;bt++){let kt=bt+it*E[2]+Q*E[1]+D*E[0];y.values[kt]=l}continue}let at=Math.floor(ft),xt=Math.ceil(ft),dt=ft-at;for(let bt=0;bt<f;bt++){let kt=bt+at*k[2]+X*k[1]+q*k[0],At=v[kt];kt=bt+xt*k[2]+X*k[1]+q*k[0];let $t=v[kt];kt=bt+at*k[2]+ot*k[1]+q*k[0];let qt=v[kt];kt=bt+xt*k[2]+ot*k[1]+q*k[0];let Kt=v[kt],ce=At+($t-At)*dt,Ot=qt+(Kt-qt)*dt;kt=bt+it*E[2]+Q*E[1]+D*E[0],y.values[kt]=ce+(Ot-ce)*st}}}else for(let X=0;X<g;++X){let ot=g>1?M*(m-1)+X*j:.5*(M+U)*(m-1);if(ot<0||ot>m-1){for(let ft=0;ft<f;ft++){let at=ft+X*E[2]+Q*E[1]+D*E[0];y.values[at]=l}continue}let st=Math.round(ot),it=Math.round(rt);for(let ft=0;ft<f;ft++){let at=ft+st*k[2]+it*k[1]+q*k[0],xt=ft+X*E[2]+Q*E[1]+D*E[0];y.values[xt]=v[at]}}}}return e.makeTensorInfo(y.shape,y.dtype,y.values)}var VF={kernelName:oa,backendName:"cpu",kernelFunc:b9};function w9(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s,exclusive:i,reverse:a}=n;nt(o,"cumprod");let u=S.getAxesPermutation([s],o.shape.length),l=o;u!=null&&(l=He({inputs:{x:o},backend:e,attrs:{perm:u}}));let c=S.getInnerMostAxes(1,o.shape.length)[0];if(c!==l.shape.length-1)throw new Error(`backend.cumprod in CPU expects an inner-most axis=${l.shape.length-1} but got axis=${c}`);let p=ar(l.dtype,"int32"),m=x.makeOnesTypedArray(x.sizeFromShape(l.shape),p),f=e.data.get(l.dataId).values,d=l.shape[l.shape.length-1],h=a?(y,b)=>y+d-b-1:(y,b)=>y+b;for(let y=0;y<f.length;y+=d)for(let b=0;b<d;b++){let w=h(y,b);if(b===0)m[w]=i?1:f[w];else{let v=h(y,b-1);m[w]=i?f[v]*m[v]:f[w]*m[v]}}let g=e.makeTensorInfo(l.shape,p,m);if(u!=null){let y=S.getUndoAxesPermutation(u),b=He({inputs:{x:g},backend:e,attrs:{perm:y}});return e.disposeIntermediateTensorInfo(g),e.disposeIntermediateTensorInfo(l),b}return g}var GF={kernelName:na,backendName:"cpu",kernelFunc:w9};function v9(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s,exclusive:i,reverse:a}=n;nt(o,"cumsum");let u=S.getAxesPermutation([s],o.shape.length),l=o;u!=null&&(l=He({inputs:{x:o},backend:e,attrs:{perm:u}}));let c=S.getInnerMostAxes(1,o.shape.length)[0];if(c!==l.shape.length-1)throw new Error(`backend.cumsum in CPU expects an inner-most axis=${l.shape.length-1} but got axis=${c}`);let p=ar(l.dtype,"int32"),m=x.makeZerosTypedArray(x.sizeFromShape(l.shape),p),f=e.data.get(l.dataId).values,d=l.shape[l.shape.length-1],h=a?(y,b)=>y+d-b-1:(y,b)=>y+b;for(let y=0;y<f.length;y+=d)for(let b=0;b<d;b++){let w=h(y,b);if(b===0)m[w]=i?0:f[w];else{let v=h(y,b-1);m[w]=i?f[v]+m[v]:f[w]+m[v]}}let g=e.makeTensorInfo(l.shape,p,m);if(u!=null){let y=S.getUndoAxesPermutation(u),b=He({inputs:{x:g},backend:e,attrs:{perm:y}});return e.disposeIntermediateTensorInfo(g),e.disposeIntermediateTensorInfo(l),b}return g}var WF={kernelName:ns,backendName:"cpu",kernelFunc:v9};function C9(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,weights:s}=t,{size:i,binaryOutput:a}=n;if(o.shape.length===1){let u=e.data.get(o.dataId).values,l=e.data.get(s.dataId).values,c=Pm(u,l,s.dtype,s.shape,i);return e.makeTensorInfo([i],s.dtype,c)}else if(o.shape.length===2){let u=e.bufferSync(o),l=e.bufferSync(s),c=Vy(u,l,i,a);return e.makeTensorInfo(c.shape,s.dtype,c.values)}throw new Error(`Error in denseBincount: input must be at most rank 2, but got rank${o.shape.length}.`)}var UF={kernelName:gp,backendName:"cpu",kernelFunc:C9};function I9(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{blockSize:s,dataFormat:i}=n;x.assert(i==="NHWC",()=>`Only NHWC dataFormat supported on CPU for depthToSpace. Got ${i}`);let a=o.shape[0],u=o.shape[1],l=o.shape[2],c=o.shape[3],p=u*s,m=l*s,f=c/(s*s),d=e.data.get(o.dataId).values,h=new Float32Array(a*p*m*f),g=0;for(let y=0;y<a;++y)for(let b=0;b<p;++b){let w=Math.floor(b/s),v=b%s;for(let k=0;k<m;++k){let E=Math.floor(k/s),D=k%s,$=(v*s+D)*f;for(let R=0;R<f;++R){let W=R+$+c*(E+l*(w+u*y));h[g++]=d[W]}}}return e.makeTensorInfo([a,p,m,f],o.dtype,h)}var HF={kernelName:sa,backendName:"cpu",kernelFunc:I9};function RS(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,filter:s}=t,{strides:i,pad:a,dilations:u,dimRoundingMode:l}=n;nt([o,s],"depthwiseConv2DNative");let c=x.computeStrides(o.shape),p=x.computeStrides(s.shape),m=u;m==null&&(m=[1,1]),x.assert(S.eitherStridesOrDilationsAreOne(i,m),()=>`Error in depthwiseConv2d: Either strides or dilations must be 1. Got strides ${i} and dilations '${m}'`);let f=S.computeConv2DInfo(o.shape,s.shape,i,m,a,l,!0),{filterHeight:d,filterWidth:h,dilationHeight:g,dilationWidth:y,padInfo:b}=f,w=b.left,v=b.top,k=f.outChannels/f.inChannels,E=new me(f.outShape,o.dtype),D=e.data.get(o.dataId).values,$=e.data.get(s.dataId).values,R=E.values;for(let M=0;M<f.batchSize;++M){let W=M*c[0],U=M*E.strides[0];for(let q=0;q<f.outHeight;++q){let K=U+q*E.strides[1],j=q*f.strideHeight-v;for(let Q=0;Q<d;++Q){let rt=j+Q*g;if(rt<0||rt>=f.inHeight)continue;let X=Q*p[0],ot=W+rt*c[1];for(let st=0;st<f.outWidth;++st){let it=K+st*E.strides[2],ft=st*f.strideWidth-w;for(let at=0;at<h;++at){let xt=ft+at*y;if(xt<0||xt>=f.inWidth)continue;let dt=X+at*p[1],bt=ot+xt*f.inChannels,kt=it,At=dt;for(let $t=0;$t<f.inChannels;++$t){let qt=D[bt+$t];for(let Kt=0;Kt<k;++Kt)R[kt+Kt]+=qt*$[At+Kt];kt+=k,At+=k}}}}}}return e.makeTensorInfo(E.shape,E.dtype,E.values)}var qF={kernelName:os,backendName:"cpu",kernelFunc:RS};function S9(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,dy:s}=t,{strides:i,dilations:a,pad:u,dimRoundingMode:l,filterShape:c}=n;nt([o,s],"depthwiseConv2dNativeBackpropFilter");let p=S.computeConv2DInfo(o.shape,c,i,a,u,l,!0),{strideHeight:m,strideWidth:f,filterHeight:d,filterWidth:h}=p,g=new me(p.filterShape,"float32"),y=p.padInfo.left,b=p.padInfo.top,w=p.outChannels/p.inChannels,v=e.data.get(o.dataId).values,k=new me(o.shape,o.dtype,v),E=e.data.get(s.dataId).values,D=new me(s.shape,s.dtype,E);for(let $=0;$<d;++$){let R=Math.max(0,Math.ceil((b-$)/m)),M=Math.min(p.outHeight,(p.inHeight+b-$)/m);for(let W=0;W<h;++W){let U=Math.max(0,Math.ceil((y-W)/f)),q=Math.min(p.outWidth,(p.inWidth+y-W)/f);for(let K=0;K<p.outChannels;++K){let j=Math.trunc(K/w),Q=K%w,rt=0;for(let X=0;X<p.batchSize;++X)for(let ot=R;ot<M;++ot){let st=$+ot*m-b;for(let it=U;it<q;++it){let ft=W+it*f-y;rt+=k.get(X,st,ft,j)*D.get(X,ot,it,K)}}g.set(rt,$,W,j,Q)}}}return e.makeTensorInfo(g.shape,g.dtype,g.values)}var KF={kernelName:xp,backendName:"cpu",kernelFunc:S9};function k9(r){let{inputs:t,backend:e,attrs:n}=r,{dy:o,filter:s}=t,{strides:i,dilations:a,pad:u,dimRoundingMode:l,inputShape:c}=n;nt([o,s],"depthwiseConv2DNativeBackpropInput");let p=x.computeStrides(o.shape),m=x.computeStrides(s.shape),f=S.computeConv2DInfo(c,s.shape,i,a,u,l,!0),d=new me(f.inShape,"float32"),h=d.values,[g,y,b]=d.strides,w=e.data.get(o.dataId).values,[v,k,E]=p,D=e.data.get(s.dataId).values,[$,R,M]=m,{batchSize:W,filterHeight:U,filterWidth:q,inChannels:K,inHeight:j,inWidth:Q,outChannels:rt,outHeight:X,outWidth:ot,strideHeight:st,strideWidth:it}=f,ft=U-1-f.padInfo.top,at=q-1-f.padInfo.left,xt=rt/K;for(let dt=0;dt<W;++dt)for(let bt=0;bt<K;++bt)for(let kt=0;kt<j;++kt){let At=kt-ft,$t=Math.max(0,Math.ceil(At/st)),qt=Math.min(X,(U+At)/st);for(let Kt=0;Kt<Q;++Kt){let ce=Kt-at,Ot=Math.max(0,Math.ceil(ce/it)),Ae=Math.min(ot,(q+ce)/it),Ie=0;for(let ae=$t;ae<qt;++ae){let je=ae*st-At;for(let Fe=Ot;Fe<Ae;++Fe){let Jr=Fe*it-ce,ze=v*dt+k*ae+E*Fe,Lo=$*(U-1-je)+R*(q-1-Jr)+M*bt;for(let Fr=0;Fr<xt;++Fr){let Qr=bt*xt+Fr,tn=w[ze+Qr],Hr=D[Lo+Fr];Ie+=tn*Hr}}}h[g*dt+y*kt+b*Kt+bt]=Ie}}return e.makeTensorInfo(d.shape,d.dtype,d.values)}var jF={kernelName:yp,backendName:"cpu",kernelFunc:k9};function N9(r){let{inputs:t,backend:e}=r,{x:n}=t,o=x.sizeFromShape(n.shape),s=e.data.get(n.dataId).values,i=It([o,o],n.dtype),a=i.values;for(let l=0;l<s.length;l++)a[l*o+l]=s[l];let u=[...n.shape,...n.shape];return e.makeTensorInfo(u,i.dtype,i.values)}var XF={kernelName:bp,backendName:"cpu",kernelFunc:N9};var YF={kernelName:kl,backendName:"cpu",kernelFunc:({inputs:r,backend:t,attrs:e})=>{let{x:n,filter:o}=r,{strides:s,pad:i,dilations:a}=e,u=t,l=u.data.get(n.dataId).values,c=n.shape.length,p=u.data.get(o.dataId).values,m=o.shape.length,{batchSize:f,inHeight:d,inWidth:h,inChannels:g,outHeight:y,outWidth:b,padInfo:w,strideHeight:v,strideWidth:k,filterHeight:E,filterWidth:D,dilationHeight:$,dilationWidth:R,outShape:M}=S.computeDilation2DInfo(n.shape,o.shape,s,i,"NHWC",a),W=x.sizeFromShape(M),U=M.length,q=x.getArrayFromDType(n.dtype,W);for(let j=0;j<f;++j)for(let Q=0;Q<y;++Q){let rt=Q*v-w.top;for(let X=0;X<b;++X){let ot=X*k-w.left;for(let st=0;st<g;++st){let it=Number.MIN_SAFE_INTEGER;for(let at=0;at<E;++at){let xt=rt+at*$;if(xt>=0&&xt<d)for(let dt=0;dt<D;++dt){let bt=ot+dt*R;if(bt>=0&&bt<h){let kt=x.locToIndex([j,xt,bt,st],c,x.computeStrides(n.shape)),At=x.locToIndex([at,dt,st],m,x.computeStrides(o.shape)),$t=l[kt]+p[At];$t>it&&(it=$t)}}}let ft=x.locToIndex([j,Q,X,st],U,x.computeStrides(M));q[ft]=it}}}return{dataId:u.write(x.toTypedArray(q,n.dtype),M,n.dtype),shape:M,dtype:n.dtype}}};var ZF={kernelName:bf,backendName:"cpu",kernelFunc:({inputs:r,backend:t,attrs:e})=>{let{x:n,filter:o,dy:s}=r,{strides:i,pad:a,dilations:u}=e,l=t,c=x.toNestedArray(n.shape,l.data.get(n.dataId).values),p=x.toNestedArray(o.shape,l.data.get(o.dataId).values),{batchSize:m,inHeight:f,inWidth:d,inChannels:h,outHeight:g,outWidth:y,padInfo:b,strideHeight:w,strideWidth:v,filterHeight:k,filterWidth:E,dilationHeight:D,dilationWidth:$,outShape:R}=S.computeDilation2DInfo(n.shape,o.shape,i,a,"NHWC",u);x.assert(s.rank===R.length,()=>`Error in ${bf}, dy must have the same rank as output ${R.length}, but got ${s.rank}`);let M=x.toNestedArray(R,l.data.get(s.dataId).values),W=x.makeZerosNestedTypedArray(o.shape,o.dtype);for(let q=0;q<m;++q)for(let K=0;K<g;++K){let j=K*w-b.top;for(let Q=0;Q<y;++Q){let rt=Q*v-b.left;for(let X=0;X<h;++X){let ot=Number.MIN_SAFE_INTEGER,st=0,it=0;for(let ft=0;ft<k;++ft){let at=j+ft*D;if(at>=0&&at<f)for(let xt=0;xt<E;++xt){let dt=rt+xt*$;if(dt>=0&&dt<d){let bt=c[q][at][dt][X]+p[ft][xt][X];bt>ot&&(ot=bt,st=ft,it=xt)}}}W[st][it][X]+=M[q][K][Q][X]}}}return{dataId:l.write(x.toTypedArray(W,n.dtype),o.shape,o.dtype),shape:o.shape,dtype:o.dtype}}};var JF={kernelName:yf,backendName:"cpu",kernelFunc:({inputs:r,backend:t,attrs:e})=>{let{x:n,filter:o,dy:s}=r,{strides:i,pad:a,dilations:u}=e,l=t,c=x.toNestedArray(n.shape,l.data.get(n.dataId).values),p=x.toNestedArray(o.shape,l.data.get(o.dataId).values),{batchSize:m,inHeight:f,inWidth:d,inChannels:h,outHeight:g,outWidth:y,padInfo:b,strideHeight:w,strideWidth:v,filterHeight:k,filterWidth:E,dilationHeight:D,dilationWidth:$,outShape:R}=S.computeDilation2DInfo(n.shape,o.shape,i,a,"NHWC",u);x.assert(s.rank===R.length,()=>`Error in ${yf}, dy must have the same rank as output ${R.length}, but got ${s.rank}`);let M=x.toNestedArray(R,l.data.get(s.dataId).values),W=x.makeZerosNestedTypedArray(n.shape,n.dtype);for(let q=0;q<m;++q)for(let K=0;K<g;++K){let j=K*w-b.top;for(let Q=0;Q<y;++Q){let rt=Q*v-b.left;for(let X=0;X<h;++X){let ot=Number.MIN_SAFE_INTEGER,st=j<0?0:j,it=rt<0?0:rt;for(let ft=0;ft<k;++ft){let at=j+ft*D;if(at>=0&&at<f)for(let xt=0;xt<E;++xt){let dt=rt+xt*$;if(dt>=0&&dt<d){let bt=c[q][at][dt][X]+p[ft][xt][X];bt>ot&&(ot=bt,st=at,it=dt)}}}W[q][st][it][X]+=M[q][K][Q][X]}}}return{dataId:l.write(x.toTypedArray(W,n.dtype),n.shape,n.dtype),shape:n.shape,dtype:n.dtype}}};function ul(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s,keepDims:i}=n;nt(o,"sum");let a;o.dtype==="bool"?a=Do({inputs:{x:o},backend:e,attrs:{dtype:"int32"}}):a=Wr({inputs:{x:o},backend:e});let u=a.shape.length,l=x.parseAxisParam(s,a.shape),c=S.getAxesPermutation(l,u),p=l,m=a;c!=null&&(m=He({inputs:{x:a},backend:e,attrs:{perm:c}}),p=S.getInnerMostAxes(p.length,u)),S.assertAxesAreInnerMostDims("sum",p,m.shape.length);let[f,d]=S.computeOutAndReduceShapes(m.shape,p),h=S.upcastType(m.dtype,"int32"),g=Fm(e,f,h),y=x.sizeFromShape(d),b=e.data.get(g.dataId).values,w=e.data.get(m.dataId).values;for(let v=0;v<b.length;++v){let k=v*y,E=0;for(let D=0;D<y;++D)E+=w[k+D];b[v]=E}if(i){let v=S.expandShapeToKeepDim(g.shape,l),k=g;g=ne({inputs:{x:g},backend:e,attrs:{shape:v}}),e.disposeIntermediateTensorInfo(k)}return e.disposeIntermediateTensorInfo(a),c!=null&&e.disposeIntermediateTensorInfo(m),g}var QF={kernelName:Os,backendName:"cpu",kernelFunc:ul};function T9(r){let{inputs:t,backend:e,attrs:n}=r,{equation:o}=n,s=t,{allDims:i,summedDims:a,idDims:u}=S.decodeEinsumEquation(o,s.length);S.checkEinsumDimSizes(i.length,u,s);let{path:l,steps:c}=S.getEinsumComputePath(a,u),p=c.length,m=null,f=i.length,d=[];for(let h=0;h<p;++h){for(let g of c[h]){let{permutationIndices:y,expandDims:b}=S.getEinsumPermutation(f,u[g]),w;S.isIdentityPermutation(y)?w=s[g]:(w=He({inputs:{x:s[g]},backend:e,attrs:{perm:y}}),d.push(w));let v=w.shape.slice();for(let k=0;k<b.length;++k)v.splice(b[k],0,1);x.arraysEqual(w.shape,v)||(w=ne({inputs:{x:w},backend:e,attrs:{shape:v}}),d.push(w)),m===null?m=w:(m=Nc({inputs:{a:w,b:m},backend:e}),d.push(m))}h<p-1&&(l[h]>=0&&(m=ul({inputs:{x:m},backend:e,attrs:{axis:l[h]-(i.length-f),keepDims:!1}}),d.push(m)),f--)}for(let h of d)h!==m&&e.disposeIntermediateTensorInfo(h);return m}var tO={kernelName:wp,backendName:"cpu",kernelFunc:T9};function _9(r){let{inputs:t,backend:e}=r,{dy:n,y:o}=t;nt([n,o],"eluGrad");let s=new Float32Array(x.sizeFromShape(o.shape)),i=e.data.get(o.dataId).values,a=e.data.get(n.dataId).values;for(let u=0;u<i.length;++u){let l=i[u];l>=1?s[u]=a[u]:s[u]=a[u]*(l+1)}return e.makeTensorInfo(o.shape,"float32",s)}var eO={kernelName:vp,backendName:"cpu",kernelFunc:_9};var E9=S.ERF_P,A9=S.ERF_A1,D9=S.ERF_A2,$9=S.ERF_A3,R9=S.ERF_A4,F9=S.ERF_A5,O9=Et(ia,r=>{let t=Math.sign(r),e=Math.abs(r),n=1/(1+E9*e);return t*(1-((((F9*n+R9)*n+$9)*n+D9)*n+A9)*n*Math.exp(-e*e))}),rO={kernelName:ia,backendName:"cpu",kernelFunc:O9};function Vm(r){let{inputs:t,backend:e,attrs:n}=r,{input:o}=t,{dim:s}=n,i=o.shape.length,a=o.shape.slice(),u=s;return s<0&&(x.assert(-(i+1)<=s,()=>`Axis must be in the interval [${-(i+1)}, ${i}]`),u=i+s+1),a.splice(u,0,1),ne({inputs:{x:o},backend:e,attrs:{shape:a}})}var nO={kernelName:yi,backendName:"cpu",kernelFunc:Vm};var P9=te((r,t)=>r/t),Jh=ie(ss,P9),Qh={kernelName:ss,backendName:"cpu",kernelFunc:Jh};function ob(r,t,e){let n=r.shape,o=n[0],s=n[1],i=e.data.get(r.dataId),a=i.complexTensorInfos.real,u=i.complexTensorInfos.imag,l=[o,s],c=x.sizeFromShape(l),p=x.getTypedArrayFromDType("float32",c),m=x.getTypedArrayFromDType("float32",c);for(let g=0;g<o;g++){let y=Ro({inputs:{x:a},backend:e,attrs:{begin:[g,0],size:[1,s]}}),b=Ro({inputs:{x:u},backend:e,attrs:{begin:[g,0],size:[1,s]}}),w=Ir({inputs:{real:y,imag:b},backend:e}),{real:v,imag:k}=M9(w,t,e),E=S.mergeRealAndImagArrays(v,k);for(let D=0;D<s;D++){let $=S.getComplexWithIndex(E,D);p[g*s+D]=$.real,m[g*s+D]=$.imag}e.disposeIntermediateTensorInfo(y),e.disposeIntermediateTensorInfo(b),e.disposeIntermediateTensorInfo(w)}let f=e.makeTensorInfo(l,"float32",p),d=e.makeTensorInfo(l,"float32",m),h=Ir({inputs:{real:f,imag:d},backend:e});return e.disposeIntermediateTensorInfo(f),e.disposeIntermediateTensorInfo(d),h}function M9(r,t,e){let n=x.sizeFromShape(r.shape),o=e.data.get(r.dataId),s=e.data.get(o.complexTensorInfos.real.dataId).values,i=e.data.get(o.complexTensorInfos.imag.dataId).values;if(L9(n)){let a=FS(s,i,n,t,e),u=[r.shape[0],r.shape[1]];if(t){let l=e.makeTensorInfo(u,"float32",a.real),c=e.makeTensorInfo(u,"float32",a.imag),p=e.makeTensorInfo([],"float32",x.createScalarValue(n,"float32")),m=Wr({inputs:{x:p},backend:e}),f=Qh.kernelFunc({inputs:{a:l,b:p},backend:e}),d=Qh.kernelFunc({inputs:{a:c,b:m},backend:e}),h=e.data.get(f.dataId).values,g=e.data.get(d.dataId).values;return e.disposeIntermediateTensorInfo(l),e.disposeIntermediateTensorInfo(c),e.disposeIntermediateTensorInfo(p),e.disposeIntermediateTensorInfo(m),e.disposeIntermediateTensorInfo(f),e.disposeIntermediateTensorInfo(d),{real:h,imag:g}}return a}else{let a=S.mergeRealAndImagArrays(s,i),u=z9(a,n,t);return S.splitRealAndImagArrays(u)}}function L9(r){return(r&r-1)===0}function FS(r,t,e,n,o){if(e===1)return{real:r,imag:t};let s=S.mergeRealAndImagArrays(r,t),i=e/2,a=S.complexWithEvenIndex(s),u=a.real,l=a.imag,c=[u.length],p=o.makeTensorInfo(c,"float32",u),m=o.makeTensorInfo(c,"float32",l),f=Ir({inputs:{real:p,imag:m},backend:o}),d=S.complexWithOddIndex(s),h=d.real,g=d.imag,y=[h.length],b=o.makeTensorInfo(y,"float32",h),w=o.makeTensorInfo(y,"float32",g),v=Ir({inputs:{real:b,imag:w},backend:o}),k=FS(u,l,i,n,o),E=k.real,D=k.imag,$=[E.length],R=o.makeTensorInfo($,"float32",E),M=o.makeTensorInfo($,"float32",D),W=Ir({inputs:{real:R,imag:M},backend:o}),U=FS(h,g,i,n,o),q=U.real,K=U.imag,j=[q.length],Q=o.makeTensorInfo(j,"float32",q),rt=o.makeTensorInfo(j,"float32",K),X=Ir({inputs:{real:Q,imag:rt},backend:o}),ot=S.exponents(e,n),st=[ot.real.length],it=o.makeTensorInfo(st,"float32",ot.real),ft=o.makeTensorInfo(st,"float32",ot.imag),at=Ir({inputs:{real:it,imag:ft},backend:o}),xt=Nc({inputs:{a:at,b:X},backend:o}),dt=ll({inputs:{a:W,b:xt},backend:o}),bt=Yh({inputs:{a:W,b:xt},backend:o}),kt=Ao({inputs:{input:dt},backend:o}),At=Ao({inputs:{input:bt},backend:o}),$t=Ui({inputs:{input:dt},backend:o}),qt=Ui({inputs:{input:bt},backend:o}),Kt=nu({inputs:[kt,At],backend:o,attrs:{axis:0}}),ce=nu({inputs:[$t,qt],backend:o,attrs:{axis:0}}),Ot=o.data.get(Kt.dataId).values,Ae=o.data.get(ce.dataId).values;return o.disposeIntermediateTensorInfo(p),o.disposeIntermediateTensorInfo(m),o.disposeIntermediateTensorInfo(f),o.disposeIntermediateTensorInfo(b),o.disposeIntermediateTensorInfo(w),o.disposeIntermediateTensorInfo(v),o.disposeIntermediateTensorInfo(R),o.disposeIntermediateTensorInfo(M),o.disposeIntermediateTensorInfo(W),o.disposeIntermediateTensorInfo(Q),o.disposeIntermediateTensorInfo(rt),o.disposeIntermediateTensorInfo(X),o.disposeIntermediateTensorInfo(it),o.disposeIntermediateTensorInfo(ft),o.disposeIntermediateTensorInfo(at),o.disposeIntermediateTensorInfo(xt),o.disposeIntermediateTensorInfo(dt),o.disposeIntermediateTensorInfo(bt),o.disposeIntermediateTensorInfo(kt),o.disposeIntermediateTensorInfo($t),o.disposeIntermediateTensorInfo(At),o.disposeIntermediateTensorInfo(qt),o.disposeIntermediateTensorInfo(Kt),o.disposeIntermediateTensorInfo(ce),{real:Ot,imag:Ae}}function z9(r,t,e){let n=new Float32Array(t*2);for(let o=0;o<t;o++){let s=0,i=0;for(let a=0;a<t;a++){let u=S.exponent(o*a,t,e),l=S.getComplexWithIndex(r,a);s+=l.real*u.real-l.imag*u.imag,i+=l.real*u.imag+l.imag*u.real}e&&(s/=t,i/=t),S.assignToTypedArray(n,s,i,o)}return n}function B9(r){let{inputs:t,backend:e}=r,{input:n}=t,o=x.sizeFromShape(n.shape),s=n.shape[n.shape.length-1],i=o/s,a=ne({inputs:{x:n},backend:e,attrs:{shape:[i,s]}}),u=ob(a,!1,e),l=ne({inputs:{x:u},backend:e,attrs:{shape:n.shape}});return e.disposeIntermediateTensorInfo(a),e.disposeIntermediateTensorInfo(u),l}var oO={kernelName:Cp,backendName:"cpu",kernelFunc:B9};function tg(r){let{backend:t,attrs:e}=r,{shape:n,value:o,dtype:s}=e,i=s||x.inferDtype(o),a=x.getArrayFromDType(i,x.sizeFromShape(n));return V9(a,o,i),t.makeTensorInfo(n,i,a)}var sO={kernelName:Nl,backendName:"cpu",kernelFunc:tg};function V9(r,t,e){r.fill(t)}var iO={kernelName:ua,backendName:"cpu",kernelFunc:({inputs:r,attrs:t,backend:e})=>{let{image:n}=r,o=e,s=x.getTypedArrayFromDType(n.dtype,x.sizeFromShape(n.shape)),[i,a,u,l]=n.shape,c=o.data.get(n.dataId).values;for(let m=0;m<i;m++){let f=m*u*a*l;for(let d=0;d<a;d++){let h=d*(u*l);for(let g=0;g<u;g++){let y=g*l;for(let b=0;b<l;b++){let w=Math.round(u-g-1),v=f+h+y+b,k=c[v];if(w>=0&&w<u){let E=w*l,D=f+h+E+b;k=c[D]}s[v]=k}}}}return{dataId:o.write(s,n.shape,n.dtype),shape:n.shape,dtype:n.dtype}}};var G9=te((r,t)=>Math.floor(r/t)),W9=ie(us,G9,null,"int32"),aO={kernelName:us,backendName:"cpu",kernelFunc:W9};function U9(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,filter:s,bias:i,preluActivationWeights:a}=t,{strides:u,pad:l,dataFormat:c,dilations:p,dimRoundingMode:m,activation:f,leakyreluAlpha:d}=n,h=$S({inputs:{x:o,filter:s},backend:e,attrs:{strides:u,pad:l,dataFormat:c,dilations:p,dimRoundingMode:m}});if(i){let g=h;h=ll({inputs:{a:h,b:i},backend:e}),e.disposeIntermediateTensorInfo(g)}if(f){let g=h;h=zm(e,h,f,a,d),e.disposeIntermediateTensorInfo(g)}return h}var lO={kernelName:Di,backendName:"cpu",kernelFunc:U9};function H9(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,filter:s,bias:i,preluActivationWeights:a}=t,{strides:u,pad:l,dataFormat:c,dilations:p,dimRoundingMode:m,activation:f,leakyreluAlpha:d}=n,h=RS({inputs:{x:o,filter:s},backend:e,attrs:{strides:u,pad:l,dataFormat:c,dilations:p,dimRoundingMode:m}});if(i){let g=h;h=ll({inputs:{a:h,b:i},backend:e}),e.disposeIntermediateTensorInfo(g)}if(f){let g=h;h=zm(e,h,f,a,d),e.disposeIntermediateTensorInfo(g)}return h}var uO={kernelName:$i,backendName:"cpu",kernelFunc:H9};function q9(r){let{inputs:t,backend:e}=r,{params:n,indices:o}=t,s=x.sizeFromShape(n.shape),i=o.shape,a=i[i.length-1],[u,l,c,p]=S.prepareAndValidate(n,o);if(l===0)return e.makeTensorInfo(u,n.dtype,[]);let m=e.data.get(o.dataId).values,f=e.bufferSync(n),d=Gy(m,f,n.dtype,l,a,c,p,n.shape,s);return e.makeTensorInfo(u,n.dtype,d.values)}var cO={kernelName:ca,backendName:"cpu",kernelFunc:q9};function K9(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,indices:s}=t,{axis:i,batchDims:a}=n;nt([o,s],"gatherV2");let u=x.parseAxisParam(i,o.shape)[0],l=e.data.get(s.dataId).values,c=o.shape[u];for(let v=0;v<l.length;++v){let k=l[v];x.assert(k<=c-1&&k>=0,()=>`GatherV2: the index value ${k} is not in [0, ${c-1}]`)}let p=a;a==null&&(p=0);let m=x.sizeFromShape(s.shape),f=S.segment_util.collectGatherOpShapeInfo(o,s,u,p),d=ne({inputs:{x:o},backend:e,attrs:{shape:[f.batchSize,f.outerSize,f.dimSize,f.sliceSize]}}),h=ne({inputs:{x:s},backend:e,attrs:{shape:[f.batchSize,m/f.batchSize]}}),g=[f.batchSize,f.outerSize,m/f.batchSize,f.sliceSize],y=e.bufferSync(h),b=e.bufferSync(d),w=Wy(b,y,g);return e.disposeIntermediateTensorInfo(d),e.disposeIntermediateTensorInfo(h),e.makeTensorInfo(f.outputShape,w.dtype,w.values)}var pO={kernelName:bi,backendName:"cpu",kernelFunc:K9};function j9(r){let{inputs:t,backend:e}=r,{input:n}=t,o=x.sizeFromShape(n.shape),s=n.shape[n.shape.length-1],i=o/s,a=ne({inputs:{x:n},backend:e,attrs:{shape:[i,s]}}),u=ob(a,!0,e),l=ne({inputs:{x:u},backend:e,attrs:{shape:n.shape}});return e.disposeIntermediateTensorInfo(a),e.disposeIntermediateTensorInfo(u),l}var mO={kernelName:Ip,backendName:"cpu",kernelFunc:j9};var X9=Et(ma,r=>Number.isFinite(r)?1:0,"bool"),fO={kernelName:ma,backendName:"cpu",kernelFunc:X9};var Y9=Et(fa,r=>Math.abs(r)===1/0?1:0,"bool"),dO={kernelName:fa,backendName:"cpu",kernelFunc:Y9};var Z9=Et(da,r=>Number.isNaN(r)?1:0,"bool"),hO={kernelName:da,backendName:"cpu",kernelFunc:Z9};function J9(r){let{backend:t,attrs:e}=r,{start:n,stop:o,num:s}=e,i=Uy(n,o,s);return t.makeTensorInfo([i.length],"float32",i)}var gO={kernelName:kp,backendName:"cpu",kernelFunc:J9};var Q9=Et(xa,r=>Math.log1p(r)),xO={kernelName:xa,backendName:"cpu",kernelFunc:Q9};var tJ=te((r,t)=>r&&t),eJ=ie(ya,tJ,null,"bool"),yO={kernelName:ya,backendName:"cpu",kernelFunc:eJ};var rJ=Et(wu,r=>r?0:1,"bool"),bO={kernelName:wu,backendName:"cpu",kernelFunc:rJ};var nJ=te((r,t)=>r||t),oJ=ie(vu,nJ,null,"bool"),wO={kernelName:vu,backendName:"cpu",kernelFunc:oJ};function sJ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{depthRadius:s,bias:i,alpha:a,beta:u}=n;nt(o,"LRN");let l=o.shape[3],c=l-1,p=e.data.get(o.dataId).values,m=x.sizeFromShape(o.shape),f=new Float32Array(m);function d(h){let g=h%l,y=h-g+Math.max(0,g-s),b=h-g+Math.min(g+s,c),w=0;for(;y<=b;y++){let v=p[y];w+=v*v}return w}for(let h=0;h<m;h++){let g=d(h),y=p[h]*Math.pow(i+a*g,-u);f[h]=y}return e.makeTensorInfo(o.shape,o.dtype,f)}var vO={kernelName:Tl,backendName:"cpu",kernelFunc:sJ};function iJ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,y:s,dy:i}=t,{depthRadius:a,bias:u,alpha:l,beta:c}=n;nt(i,"LRNGrad");let p=x.sizeFromShape(i.shape),m=i.shape[3],f=e.data.get(i.dataId).values,d=e.data.get(o.dataId).values,h=e.data.get(s.dataId).values,g=new Float32Array(p),y=p;for(let b=0;b<y;b++){let w=b%m,v=b-w+Math.max(0,w-a),k=b-w+Math.min(m,w+a+1),E=0;for(let D=v;D<k;D++)E+=Math.pow(d[D],2);E=l*E+u;for(let D=v;D<k;D++){let $=-2*l*c*d[D]*h[b]/E;b===D&&($+=Math.pow(E,-c)),$*=f[b],g[D]+=$}}return e.makeTensorInfo(i.shape,o.dtype,g)}var CO={kernelName:Np,backendName:"cpu",kernelFunc:iJ};function OS(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{reductionIndices:s,keepDims:i}=n,a=e,u=o.shape,l=u.length,c=x.parseAxisParam(s,u),p=c,m=S.getAxesPermutation(p,l),f=a.data.get(o.dataId).values;if(m!=null){let v=new Array(l);for(let k=0;k<v.length;k++)v[k]=u[m[k]];f=Mm(f,u,o.dtype,m,v),p=S.getInnerMostAxes(p.length,l),u=v}nt(o,"max"),S.assertAxesAreInnerMostDims("max",p,l);let[d,h]=S.computeOutAndReduceShapes(u,p),g=x.sizeFromShape(h),y=Hy(f,g,d,o.dtype),b=a.write(y,d,o.dtype),w=d;return i&&(w=S.expandShapeToKeepDim(d,c)),{dataId:b,shape:w,dtype:o.dtype}}var IO={kernelName:ds,backendName:"cpu",kernelFunc:OS};function aJ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t;nt(o,"maxPool");let{filterSize:s,strides:i,pad:a,dimRoundingMode:u}=n,l=1;x.assert(S.eitherStridesOrDilationsAreOne(i,l),()=>`Error in maxPool: Either strides or dilations must be 1. Got strides ${i} and dilations '${l}'`);let c=S.computePool2DInfo(o.shape,s,i,l,a,u),p;if(c.filterWidth===1&&c.filterHeight===1&&x.arraysEqual(c.inShape,c.outShape))p=Wr({inputs:{x:o},backend:e});else{let m=e.data.get(o.dataId).values,f=x.computeStrides(o.shape),d=Bm(m,o.shape,o.dtype,f,c,"max");p=e.makeTensorInfo(c.outShape,o.dtype,d.values)}return p}var SO={kernelName:gs,backendName:"cpu",kernelFunc:aJ};function lJ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{filterSize:s,strides:i,pad:a,dimRoundingMode:u,dataFormat:l}=n;nt(o,"maxPool3d");let c=S.computePool3DInfo(o.shape,s,i,1,a,u,l),p=e.data.get(o.dataId).values,m=nb(p,o.shape,o.dtype,x.computeStrides(o.shape),c,"max");return e.makeTensorInfo(m.shape,"float32",m.values)}var kO={kernelName:_l,backendName:"cpu",kernelFunc:lJ};function uJ(r){let{inputs:t,backend:e,attrs:n}=r,{dy:o,input:s}=t,{filterSize:i,strides:a,pad:u,dimRoundingMode:l}=n;nt([o,s],"maxPool3DGrad");let c=S.computePool3DInfo(s.shape,i,a,1,u,l),p=e.bufferSync(s),m=wF(p,c),f=c.strideDepth,d=c.strideHeight,h=c.strideWidth,g=c.dilationDepth,y=c.dilationHeight,b=c.dilationWidth,w=c.effectiveFilterDepth,v=c.effectiveFilterHeight,k=c.effectiveFilterWidth,E=w-1-c.padInfo.front,D=k-1-c.padInfo.left,$=v-1-c.padInfo.top,R=It(s.shape,"float32"),M=e.bufferSync(o);for(let W=0;W<c.batchSize;++W)for(let U=0;U<c.inChannels;++U)for(let q=0;q<c.inDepth;++q)for(let K=0;K<c.inHeight;++K)for(let j=0;j<c.inWidth;++j){let Q=q-E,rt=K-$,X=j-D,ot=0;for(let st=0;st<w;st+=g){let it=(Q+st)/f;if(!(it<0||it>=c.outDepth||Math.floor(it)!==it))for(let ft=0;ft<v;ft+=y){let at=(rt+ft)/d;if(!(at<0||at>=c.outHeight||Math.floor(at)!==at))for(let xt=0;xt<k;xt+=b){let dt=(X+xt)/h;if(dt<0||dt>=c.outWidth||Math.floor(dt)!==dt)continue;let bt=w*v*k-1-m.get(W,it,at,dt,U),kt=st*v*k+ft*k+xt,At=bt===kt?1:0;if(At===0)continue;ot+=M.get(W,it,at,dt,U)*At}}}R.set(ot,W,q,K,j,U)}return e.makeTensorInfo(R.shape,R.dtype,R.values)}var NO={kernelName:_p,backendName:"cpu",kernelFunc:uJ};function cJ(r){let{inputs:t,backend:e,attrs:n}=r,{dy:o,input:s,output:i}=t,a=s;nt([s,i],"maxPoolGrad");let{filterSize:u,strides:l,pad:c,dimRoundingMode:p}=n,m=S.computePool2DInfo(a.shape,u,l,1,c,p),f=e.data.get(a.dataId).values,d=It(m.outShape,a.dtype,rb(f,a.shape,a.dtype,m).values),h=m.strideHeight,g=m.strideWidth,y=m.dilationHeight,b=m.dilationWidth,w=m.effectiveFilterHeight,v=m.effectiveFilterWidth,k=v-1-m.padInfo.left,E=w-1-m.padInfo.top,D=It(a.shape,"float32"),$=e.data.get(o.dataId).values,R=It(o.shape,"float32",$);for(let M=0;M<m.batchSize;++M)for(let W=0;W<m.inChannels;++W)for(let U=0;U<m.inHeight;++U)for(let q=0;q<m.inWidth;++q){let K=U-E,j=q-k,Q=0;for(let rt=0;rt<w;rt+=y){let X=(K+rt)/h;if(!(X<0||X>=m.outHeight||Math.floor(X)!==X))for(let ot=0;ot<v;ot+=b){let st=(j+ot)/g;if(st<0||st>=m.outWidth||Math.floor(st)!==st)continue;let it=w*v-1-d.get(M,X,st,W),ft=rt*v+ot,at=it===ft?1:0;if(at===0)continue;Q+=R.get(M,X,st,W)*at}}D.set(Q,M,U,q,W)}return e.makeTensorInfo(D.shape,D.dtype,D.values)}var TO={kernelName:Tp,backendName:"cpu",kernelFunc:cJ};function _O(r,t,e,n,o){let s=x.computeStrides(t),i=Bm(r,t,e,s,o,"max"),a=rb(r,t,e,o,!0,n);return[i.values,a.values]}var EO={kernelName:Ep,backendName:"cpu",kernelFunc:({inputs:r,attrs:t,backend:e})=>{let{x:n}=r,{filterSize:o,strides:s,pad:i,includeBatchInIndex:a}=t,u=e;nt(n,"MaxPoolWithArgmax");let l=u.data.get(n.dataId).values,c=S.computePool2DInfo(n.shape,o,s,[1,1],i),[p,m]=_O(l,n.shape,n.dtype,a,c),f=u.write(p,c.outShape,n.dtype),d=u.write(m,c.outShape,n.dtype);return[{dataId:f,shape:c.outShape,dtype:n.dtype},{dataId:d,shape:c.outShape,dtype:"int32"}]}};function pJ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s,keepDims:i}=n,a=x.parseAxisParam(s,o.shape),l=S.computeOutAndReduceShapes(o.shape,a)[1],c=x.sizeFromShape(l),p=[],m=e.makeTensorInfo([],"float32",new Float32Array([c]));p.push(m);let f=Do({inputs:{x:o},backend:e,attrs:{dtype:"float32"}});p.push(f);let d=Jh({inputs:{a:f,b:m},backend:e});p.push(d);let h=ul({inputs:{x:d},backend:e,attrs:{axis:s,keepDims:i}});return p.forEach(g=>e.disposeIntermediateTensorInfo(g)),h}var AO={kernelName:xs,backendName:"cpu",kernelFunc:pJ};function mJ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s,keepDims:i}=n;nt(o,"min");let a=x.parseAxisParam(s,o.shape),u=a,l=S.getAxesPermutation(u,o.shape.length),c=o;l!=null&&(c=He({inputs:{x:o},backend:e,attrs:{perm:l}}),u=S.getInnerMostAxes(u.length,o.shape.length)),S.assertAxesAreInnerMostDims("min",u,c.shape.length);let[p,m]=S.computeOutAndReduceShapes(c.shape,u),f=x.sizeFromShape(m),d=x.makeZerosTypedArray(x.sizeFromShape(p),c.dtype),h=e.data.get(c.dataId).values;for(let y=0;y<d.length;++y){let b=y*f,w=h[b];for(let v=0;v<f;++v){let k=h[b+v];(Number.isNaN(k)||k<w)&&(w=k)}d[y]=w}l!=null&&e.disposeIntermediateTensorInfo(c);let g=e.makeTensorInfo(p,c.dtype,d);if(i){let y=S.expandShapeToKeepDim(p,a),b=ne({inputs:{x:g},backend:e,attrs:{shape:y}});return e.disposeIntermediateTensorInfo(g),b}return g}var DO={kernelName:ys,backendName:"cpu",kernelFunc:mJ};function fJ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{paddings:s,mode:i}=n;nt(o,"mirrorPad");let a=s.map((w,v)=>w[0]+o.shape[v]+w[1]),u=s.map(w=>w[0]),l=s.map((w,v)=>w[0]+o.shape[v]),c=i==="reflect"?0:1,p=e.data.get(o.dataId).values,m=o.shape.length,f=x.computeStrides(o.shape),d=x.sizeFromShape(a),h=a.length,g=x.computeStrides(a),y=x.getTypedArrayFromDType(o.dtype,d);for(let w=0;w<d;w++){let v=x.indexToLoc(w,h,g);for(let E=0;E<h;E++)v[E]<u[E]?v[E]=u[E]*2-v[E]-c:v[E]>=l[E]&&(v[E]=(l[E]-1)*2-v[E]+c);v=v.map((E,D)=>E-u[D]);let k=x.locToIndex(v,m,f);y[w]=p[k]}return{dataId:e.write(y,a,o.dtype),shape:a,dtype:o.dtype}}var $O={kernelName:ws,backendName:"cpu",kernelFunc:fJ};var dJ=te((r,t)=>{let e=r%t;return r<0&&t<0||r>=0&&t>=0?e:(e+t)%t}),hJ=ie(ba,dJ),RO={kernelName:ba,backendName:"cpu",kernelFunc:hJ};var OO=xu(J0());function PS(r){let{inputs:t,backend:e,attrs:n}=r,{logits:o}=t,{dim:s}=n,i=o.shape.length,a=s;if(a===-1&&(a=i-1),a!==i-1)throw Error(`Softmax along a non-last dimension is not yet supported. Logits was rank ${i} and dim was ${a}`);let u=x.parseAxisParam([a],o.shape),l=OS({inputs:{x:o},backend:e,attrs:{reductionIndices:u,keepDims:!1}}),c=S.expandShapeToKeepDim(l.shape,u),p=ne({inputs:{x:l},backend:e,attrs:{shape:c}}),m=Yh({inputs:{a:o,b:p},backend:e}),f=uS({inputs:{x:m},backend:e}),d=ul({inputs:{x:f},backend:e,attrs:{axis:u,keepDims:!1}}),h=ne({inputs:{x:d},backend:e,attrs:{shape:c}}),g=Jh({inputs:{a:f,b:h},backend:e});return e.disposeIntermediateTensorInfo(l),e.disposeIntermediateTensorInfo(p),e.disposeIntermediateTensorInfo(m),e.disposeIntermediateTensorInfo(f),e.disposeIntermediateTensorInfo(d),e.disposeIntermediateTensorInfo(h),g}var FO={kernelName:Ps,backendName:"cpu",kernelFunc:PS};function gJ(r){let{inputs:t,backend:e,attrs:n}=r,{logits:o}=t,{numSamples:s,seed:i,normalized:a}=n;nt(o,"multinomial");let u=a?o:PS({inputs:{logits:o},backend:e,attrs:{dim:-1}}),l=u.shape[0],c=u.shape[1],p=e.data.get(u.dataId).values,m=[l,s],f=x.makeZerosTypedArray(x.sizeFromShape(m),"int32");for(let d=0;d<l;++d){let h=d*c,g=new Float32Array(c-1);g[0]=p[h];for(let w=1;w<g.length;++w)g[w]=g[w-1]+p[h+w];let y=OO.alea(i.toString()),b=d*s;for(let w=0;w<s;++w){let v=y();f[b+w]=g.length;for(let k=0;k<g.length;k++)if(v<g[k]){f[b+w]=k;break}}}return a||e.disposeIntermediateTensorInfo(u),e.makeTensorInfo(m,"int32",f)}var PO={kernelName:Ap,backendName:"cpu",kernelFunc:gJ};var xJ=Vr.nonMaxSuppressionV3Impl;function yJ(r){let{inputs:t,backend:e,attrs:n}=r,{boxes:o,scores:s}=t,{maxOutputSize:i,iouThreshold:a,scoreThreshold:u}=n;nt(o,"NonMaxSuppression");let l=e.data.get(o.dataId).values,c=e.data.get(s.dataId).values,{selectedIndices:p}=xJ(l,c,i,a,u);return e.makeTensorInfo([p.length],"int32",new Int32Array(p))}var MO={kernelName:va,backendName:"cpu",kernelFunc:yJ};var bJ=Vr.nonMaxSuppressionV4Impl;function wJ(r){let{inputs:t,backend:e,attrs:n}=r,{boxes:o,scores:s}=t,{maxOutputSize:i,iouThreshold:a,scoreThreshold:u,padToMaxOutputSize:l}=n;nt(o,"NonMaxSuppressionPadded");let c=e.data.get(o.dataId).values,p=e.data.get(s.dataId).values,{selectedIndices:m,validOutputs:f}=bJ(c,p,i,a,u,l);return[e.makeTensorInfo([m.length],"int32",new Int32Array(m)),e.makeTensorInfo([],"int32",new Int32Array([f]))]}var LO={kernelName:Ca,backendName:"cpu",kernelFunc:wJ};var vJ=Vr.nonMaxSuppressionV5Impl;function CJ(r){let{inputs:t,backend:e,attrs:n}=r,{boxes:o,scores:s}=t,{maxOutputSize:i,iouThreshold:a,scoreThreshold:u,softNmsSigma:l}=n;nt(o,"NonMaxSuppressionWithScore");let c=e.data.get(o.dataId).values,p=e.data.get(s.dataId).values,m=i,f=a,d=u,h=l,{selectedIndices:g,selectedScores:y}=vJ(c,p,m,f,d,h);return[e.makeTensorInfo([g.length],"int32",new Int32Array(g)),e.makeTensorInfo([y.length],"float32",new Float32Array(y))]}var zO={kernelName:Ia,backendName:"cpu",kernelFunc:CJ};function IJ(r){let{inputs:t,backend:e,attrs:n}=r,{indices:o}=t,{depth:s,onValue:i,offValue:a}=n;nt(o,"oneHot");let u=x.sizeFromShape(o.shape),l=new Float32Array(u*s);l.fill(a);let c=e.data.get(o.dataId).values;for(let p=0;p<u;++p)c[p]>=0&&c[p]<s&&(l[p*s+c[p]]=i);return e.makeTensorInfo([...o.shape,s],"int32",l)}var BO={kernelName:Cs,backendName:"cpu",kernelFunc:IJ};function eg(r){let{inputs:t,backend:e}=r,{x:n}=t;if(n.dtype==="string")throw new Error("zerosLike is not supported for string tensors");if(n.dtype==="complex64"){let o=Ao({inputs:{input:n},backend:e}),s=eg({inputs:{x:o},backend:e}),i=Ui({inputs:{input:n},backend:e}),a=eg({inputs:{x:i},backend:e}),u=Ir({inputs:{real:s,imag:a},backend:e});return e.disposeIntermediateTensorInfo(o),e.disposeIntermediateTensorInfo(s),e.disposeIntermediateTensorInfo(i),e.disposeIntermediateTensorInfo(a),u}else return tg({backend:e,attrs:{shape:n.shape,value:0,dtype:n.dtype}})}var VO={kernelName:Ei,backendName:"cpu",kernelFunc:eg};function GO(r){let{inputs:t,backend:e}=r,{x:n}=t;if(n.dtype==="string")throw new Error("onesLike is not supported for string tensors");if(n.dtype==="complex64"){let o=Ao({inputs:{input:n},backend:e}),s=GO({inputs:{x:o},backend:e}),i=Ui({inputs:{input:n},backend:e}),a=eg({inputs:{x:i},backend:e}),u=Ir({inputs:{real:s,imag:a},backend:e});return e.disposeIntermediateTensorInfo(o),e.disposeIntermediateTensorInfo(s),e.disposeIntermediateTensorInfo(i),e.disposeIntermediateTensorInfo(a),u}else return tg({backend:e,attrs:{shape:n.shape,value:1,dtype:n.dtype}})}var WO={kernelName:vi,backendName:"cpu",kernelFunc:GO};function MS(r){let{inputs:t,backend:e,attrs:n}=r,{axis:o}=n;if(t.length===1)return Vm({inputs:{input:t[0]},backend:e,attrs:{dim:o}});let s=t[0].shape,i=t[0].dtype;t.forEach(c=>{x.assertShapesMatch(s,c.shape,"All tensors passed to stack must have matching shapes"),x.assert(i===c.dtype,()=>"All tensors passed to stack must have matching dtypes")});let a=[],u=t.map(c=>{let p=Vm({inputs:{input:c},backend:e,attrs:{dim:o}});return a.push(p),p}),l=nu({inputs:u,backend:e,attrs:{axis:o}});return a.forEach(c=>e.disposeIntermediateTensorInfo(c)),l}var UO={kernelName:Ci,backendName:"cpu",kernelFunc:MS};function SJ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{paddings:s,constantValue:i}=n;nt(o,"pad");let a=s.map((b,w)=>b[0]+o.shape[w]+b[1]),u=s.map(b=>b[0]),l=e.data.get(o.dataId).values,c=x.sizeFromShape(o.shape),p=o.shape.length,m=x.computeStrides(o.shape),f=x.sizeFromShape(a),d=a.length,h=x.computeStrides(a),g=x.getTypedArrayFromDType(o.dtype,f);i!==0&&g.fill(i);for(let b=0;b<c;b++){let v=x.indexToLoc(b,p,m).map((E,D)=>E+u[D]),k=x.locToIndex(v,d,h);g[k]=l[b]}return{dataId:e.write(g,a,o.dtype),shape:a,dtype:o.dtype}}var sb={kernelName:Is,backendName:"cpu",kernelFunc:SJ};var kJ=te((r,t)=>Math.pow(r,t)),NJ=ie(Ss,kJ),HO={kernelName:Ss,backendName:"cpu",kernelFunc:NJ};function TJ(r){let{backend:t,attrs:e}=r,{start:n,stop:o,dtype:s,step:i}=e,a=Tc(n,o,i,s);return t.makeTensorInfo([a.length],s,a)}var qO={kernelName:El,backendName:"cpu",kernelFunc:TJ};var _J=Et(ka,r=>1/r),KO={kernelName:ka,backendName:"cpu",kernelFunc:_J};function EJ(r){let{inputs:t,backend:e,attrs:n}=r,{images:o}=t,{alignCorners:s,halfPixelCenters:i,size:a}=n;nt(o,"resizeBilinear");let u=x.computeStrides(o.shape),[l,c]=a,[p,m,f,d]=o.shape,h=e.data.get(o.dataId).values,g=new Float32Array(x.sizeFromShape([p,l,c,d])),y=[s&&l>1?m-1:m,s&&c>1?f-1:f],b=[s&&l>1?l-1:l,s&&c>1?c-1:c],w=0,v=y[0]/b[0],k=y[1]/b[1];for(let E=0;E<p;E++)for(let D=0;D<l;D++){let $;i?$=v*(D+.5)-.5:$=v*D;let R=Math.max(0,Math.floor($)),M=$-R,W=Math.min(m-1,Math.ceil($)),U=E*u[0]+R*u[1],q=E*u[0]+W*u[1];for(let K=0;K<c;K++){let j;i?j=k*(K+.5)-.5:j=k*K;let Q=Math.max(0,Math.floor(j)),rt=j-Q,X=Math.min(f-1,Math.ceil(j)),ot=U+Q*u[2],st=q+Q*u[2],it=U+X*u[2],ft=q+X*u[2];for(let at=0;at<d;at++){let xt=h[ot+at],dt=h[st+at],bt=h[it+at],kt=h[ft+at],At=xt+(bt-xt)*rt,$t=dt+(kt-dt)*rt,qt=At+($t-At)*M;g[w++]=qt}}}return e.makeTensorInfo([p,l,c,d],"float32",g)}var jO={kernelName:Ts,backendName:"cpu",kernelFunc:EJ};function AJ(r){let{inputs:t,backend:e,attrs:n}=r,{images:o,dy:s}=t,{alignCorners:i}=n;nt([s,o],"resizeBilinearGrad");let a=x.computeStrides(o.shape),[u,l,c,p]=o.shape,[,m,f]=s.shape,d=new Float32Array(u*l*c*p),h=[i&&m>1?l-1:l,i&&f>1?c-1:c],g=[i&&m>1?m-1:m,i&&f>1?f-1:f],y=h[0]/g[0],b=h[1]/g[1],w=e.data.get(s.dataId).values,v=0;for(let k=0;k<u;k++){let E=k*a[0];for(let D=0;D<m;D++){let $=D*y,R=Math.floor($),M=Math.min(Math.ceil($),l-1),W=E+R*a[1],U=E+M*a[1],q=$-R,K=1-q;for(let j=0;j<f;j++){let Q=j*b,rt=Math.floor(Q),X=Math.min(Math.ceil(Q),c-1),ot=Q-rt,st=1-ot,it=W+rt*a[2],ft=W+X*a[2],at=U+rt*a[2],xt=U+X*a[2],dt=K*st,bt=K*ot,kt=q*st,At=q*ot;for(let $t=0;$t<p;$t++){let qt=w[v++];d[it+$t]+=qt*dt,d[ft+$t]+=qt*bt,d[at+$t]+=qt*kt,d[xt+$t]+=qt*At}}}}return e.makeTensorInfo([u,c,l,p],"float32",d)}var XO={kernelName:Rp,backendName:"cpu",kernelFunc:AJ};function DJ(r){let{inputs:t,backend:e,attrs:n}=r,{images:o}=t,{alignCorners:s,halfPixelCenters:i,size:a}=n;nt(o,"resizeNearestNeighbor");let u=x.computeStrides(o.shape),[l,c]=a,[p,m,f,d]=o.shape,h=e.data.get(o.dataId).values,g=new Float32Array(p*l*c*d),y=[s&&l>1?m-1:m,s&&c>1?f-1:f],b=[s&&l>1?l-1:l,s&&c>1?c-1:c],w=y[0]/b[0],v=y[1]/b[1],k=0;for(let E=0;E<p;E++){let D=E*u[0];for(let $=0;$<l;$++){let R=i?w*($+.5):w*$,M=Math.min(m-1,s?Math.round(R):Math.floor(R));i&&(M=Math.max(0,M));let W=D+M*u[1];for(let U=0;U<c;U++){let q=i?v*(U+.5):v*U,K=Math.min(f-1,s?Math.round(q):Math.floor(q));i&&(K=Math.max(0,K));let j=W+K*u[2];for(let Q=0;Q<d;Q++){let rt=h[j+Q];g[k++]=rt}}}}return e.makeTensorInfo([p,l,c,d],o.dtype,g)}var YO={kernelName:Al,backendName:"cpu",kernelFunc:DJ};function $J(r){let{inputs:t,backend:e,attrs:n}=r,{images:o,dy:s}=t,{alignCorners:i}=n;nt([s,o],"resizeNearestNeighborGrad");let a=x.computeStrides(o.shape),u=x.computeStrides(s.shape),[l,c,p,m]=o.shape,[,f,d]=s.shape,h=new Float32Array(l*c*p*m),g=e.data.get(s.dataId).values,y=[i&&f>1?c-1:c,i&&d>1?p-1:p],b=[i&&f>1?f-1:f,i&&d>1?d-1:d],w=y[0]/b[0],v=y[1]/b[1],k=1/w,E=1/v,D=Math.ceil(k)*2+2,$=Math.ceil(E)*2+2;for(let R=0;R<l;R++){let M=R*a[0];for(let W=0;W<c;W++){let U=M+W*a[1],q=Math.floor(W*k),K=Math.floor(q-D/2);for(let j=0;j<p;j++){let Q=U+j*a[2],rt=Math.floor(j*E),X=Math.floor(rt-$/2);for(let ot=0;ot<m;ot++){let st=0;for(let it=0;it<D;it++){let ft=it+K;if(ft<0||ft>=f)continue;let at=M+ft*u[1],xt=ft*w,dt=Math.min(c-1,i?Math.round(xt):Math.floor(xt));if(W===dt)for(let bt=0;bt<$;bt++){let kt=bt+X;if(kt<0||kt>=d)continue;let At=at+kt*u[2],$t=kt*v,qt=Math.min(p-1,i?Math.round($t):Math.floor($t));j===qt&&(st+=g[At+ot])}}h[Q+ot]=st}}}}return e.makeTensorInfo(o.shape,o.dtype,h)}var ZO={kernelName:$p,backendName:"cpu",kernelFunc:$J};function RJ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{dims:s}=n;nt(o,"reverse");let i=o.shape.length,a=x.parseAxisParam(s,o.shape);if(i===0)return Wr({inputs:{x:o},backend:e});let u=new me(o.shape,o.dtype),l=e.bufferSync(o);for(let c=0;c<u.size;c++){let p=u.indexToLoc(c),m=p.slice();a.forEach(f=>m[f]=o.shape[f]-1-m[f]),u.set(l.get(...m),...p)}return e.makeTensorInfo(u.shape,u.dtype,u.values)}var JO={kernelName:Es,backendName:"cpu",kernelFunc:RJ};var QO={kernelName:Oa,backendName:"cpu",kernelFunc:({inputs:r,attrs:t,backend:e})=>{let{image:n}=r,{radians:o,fillValue:s,center:i}=t,a=e,u=x.getTypedArrayFromDType(n.dtype,x.sizeFromShape(n.shape)),[l,c,p,m]=n.shape,[f,d]=S.getImageCenter(i,c,p),h=255,g=Math.sin(o),y=Math.cos(o),b=a.data.get(n.dataId).values;for(let v=0;v<l;v++){let k=v*p*c*m;for(let E=0;E<c;E++){let D=E*(p*m);for(let $=0;$<p;$++){let R=$*m;for(let M=0;M<m;M++){let W=[l,E,$,M],U=W[2],q=W[1],K=(U-f)*y-(q-d)*g,j=(U-f)*g+(q-d)*y;K=Math.round(K+f),j=Math.round(j+d);let Q=s;if(typeof s!="number"&&(M===3?Q=h:Q=s[M]),K>=0&&K<p&&j>=0&&j<c){let X=j*(p*m),ot=K*m,st=k+X+ot+M;Q=b[st]}let rt=k+D+R+M;u[rt]=Q}}}}return{dataId:a.write(u,n.shape,n.dtype),shape:n.shape,dtype:n.dtype}}};var FJ=Et(As,r=>{let t=Math.floor(r);return r-t<.5?Math.floor(r):r-t>.5?Math.ceil(r):t%2===0?t:t+1}),tP={kernelName:As,backendName:"cpu",kernelFunc:FJ};function ib(r,t,e,n,o,s,i,a,u,l){let c=[n/o,o],p=r.values,m=t.values;if(n===0)return It(e,t.dtype);let f=It(c,t.dtype);f.values.fill(u);for(let d=0;d<s;d++){let h=[],g=0;for(let y=0;y<i;y++){let b=p[d*i+y];h.push(b),g+=b*a[y]}if(g<0||g>=n/o)throw new Error(`Invalid indices: ${h} does not index into ${e}`);for(let y=0;y<o;y++)l?f.values[g*o+y]+=m[d*o+y]:f.values[g*o+y]=t.rank===0?m[0]:m[d*o+y]}return f}function OJ(r){let{inputs:t,backend:e,attrs:n}=r,{indices:o,updates:s}=t,{shape:i}=n,{sliceRank:a,numUpdates:u,sliceSize:l,strides:c,outputSize:p}=S.calculateShapes(s,o,i),m=!0,f=e.bufferSync(o),d=e.bufferSync(s),h=ib(f,d,i,p,l,u,a,c,0,m);return e.makeTensorInfo(i,h.dtype,h.values)}var eP={kernelName:Na,backendName:"cpu",kernelFunc:OJ};function PJ(r){let{inputs:t,backend:e}=r,{condition:n,t:o,e:s}=t;nt([n,o,s],"select");let i=n.shape.length,a=e.data.get(n.dataId).values,u=e.data.get(o.dataId).values,l=e.data.get(s.dataId).values,c=ar(o.dtype,s.dtype),p=x.makeZerosTypedArray(x.sizeFromShape(o.shape),c),m=0,f=i===0||i>1||o.shape.length===1?1:x.sizeFromShape(o.shape.slice(1));for(let d=0;d<a.length;d++)for(let h=0;h<f;h++)a[d]===1?p[m++]=u[d]:p[m++]=l[d];return e.makeTensorInfo(o.shape,c,p)}var rP={kernelName:Si,backendName:"cpu",kernelFunc:PJ};var MJ=S.SELU_SCALEALPHA,LJ=S.SELU_SCALE,zJ=Et(Ta,r=>r>=0?LJ*r:MJ*(Math.exp(r)-1)),nP={kernelName:Ta,backendName:"cpu",kernelFunc:zJ};var BJ=Et(Ea,r=>r<0?-1:r>0?1:0),oP={kernelName:Ea,backendName:"cpu",kernelFunc:BJ};var VJ=Et($s,r=>Math.sin(r)),sP={kernelName:$s,backendName:"cpu",kernelFunc:VJ};var GJ=Et(_a,r=>Math.sinh(r)),iP={kernelName:_a,backendName:"cpu",kernelFunc:GJ};var WJ=11920928955078125e-23,aP=Math.log(WJ)+2,UJ=Et(Aa,r=>{let t=r>-aP,e=r<aP,n=Math.exp(r),o;return e?o=n:t?o=r:o=Math.log(1+n),o}),lP={kernelName:Aa,backendName:"cpu",kernelFunc:UJ};function HJ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{blockShape:s,paddings:i}=n;nt([o],"spaceToBatchND");let a=x.sizeFromShape(s),u=[[0,0]];u.push(...i);for(let E=1+s.length;E<o.shape.length;++E)u.push([0,0]);let l=sb.kernelFunc({inputs:{x:o},backend:e,attrs:{paddings:u,constantValue:0}}),c=S.getReshaped(l.shape,s,a,!1),p=S.getPermuted(c.length,s.length,!1),m=S.getReshapedPermuted(l.shape,s,a,!1),h=ne({inputs:{x:l},backend:e,attrs:{shape:c}}),b=He({inputs:{x:h},backend:e,attrs:{perm:p}}),k=ne({inputs:{x:b},backend:e,attrs:{shape:m}});return e.disposeIntermediateTensorInfo(l),e.disposeIntermediateTensorInfo(h),e.disposeIntermediateTensorInfo(b),k}var uP={kernelName:Ni,backendName:"cpu",kernelFunc:HJ};function qJ(r){let{inputs:t,backend:e}=r,{indices:n,values:o,denseShape:s,defaultValue:i}=t;if(s.shape.length!==1)throw new Error(`Dense shape must be a vector, saw:
${s.shape}`);if(n.shape.length!==2)throw new Error(`Indices must be a matrix, saw:
${n.shape}`);if(o.shape.length!==1)throw new Error(`Values must be a vector, saw:
${o.shape}`);if(i.shape.length!==0)throw new Error(`Default value must be a scalar, saw:
${i.shape}`);let a=e.data.get(n.dataId).values,u=e.data.get(o.dataId).values,l=e.data.get(s.dataId).values,c=e.data.get(i.dataId).values[0],[p,m,f,d,h]=qy(a,n.shape,n.dtype,u,o.dtype,l,c);return[e.makeTensorInfo(m,n.dtype,p),e.makeTensorInfo([m[0]],o.dtype,f),e.makeTensorInfo([d.length],"bool",new Uint8Array(d.map(g=>Number(g)))),e.makeTensorInfo([h.length],n.dtype,new Int32Array(h))]}var cP={kernelName:Dl,backendName:"cpu",kernelFunc:qJ};function KJ(r){let{inputs:t,backend:e}=r,{inputIndices:n,inputShape:o,newShape:s}=t;if(n.shape.length!==2)throw new Error(`Input indices should be a matrix but received shape
${n.shape}`);if(o.shape.length!==1)throw new Error(`Input shape should be a vector but received shape
${o.shape}`);if(s.shape.length!==1)throw new Error(`Target shape should be a vector but received shape ${s.shape}`);let i=Array.from(e.data.get(o.dataId).values),a=e.data.get(n.dataId).values,u=Array.from(e.data.get(s.dataId).values),[l,c,p]=Ky(a,n.shape,n.dtype,i,u);return[e.makeTensorInfo(c,n.dtype,l),e.makeTensorInfo([p.length],s.dtype,new Int32Array(p))]}var pP={kernelName:Da,backendName:"cpu",kernelFunc:KJ};function jJ(r){let{inputs:t,backend:e}=r,{data:n,indices:o,segmentIds:s}=t;if(n.shape.length<1)throw new Error("Data should be at least 1 dimensional but received scalar");if(o.shape.length!==1)throw new Error(`Indices should be a vector but received shape
${o.shape}`);if(s.shape.length!==1)throw new Error(`Segment ids should be a vector but received shape
${s.shape}`);if(o.shape[0]!==s.shape[0])throw new Error("segmentIds and indices should have same size.");let i=e.data.get(n.dataId).values,a=e.data.get(o.dataId).values,u=e.data.get(s.dataId).values,[l,c]=Lm(i,n.shape,n.dtype,a,u,!0);return e.makeTensorInfo(c,n.dtype,l)}var mP={kernelName:$l,backendName:"cpu",kernelFunc:jJ};function XJ(r){let{inputs:t,backend:e}=r,{data:n,indices:o,segmentIds:s}=t;if(n.shape.length<1)throw new Error("Data should be at least 1 dimensional but received scalar");if(o.shape.length!==1)throw new Error(`Indices should be a vector but received shape
${o.shape}`);if(s.shape.length!==1)throw new Error(`Segment ids should be a vector but received shape
${s.shape}`);if(o.shape[0]!==s.shape[0])throw new Error("segmentIds and indices should have same size.");let i=e.data.get(n.dataId).values,a=e.data.get(o.dataId).values,u=e.data.get(s.dataId).values,[l,c]=Lm(i,n.shape,n.dtype,a,u);return e.makeTensorInfo(c,n.dtype,l)}var fP={kernelName:Rl,backendName:"cpu",kernelFunc:XJ};function YJ(r){let{inputs:t,backend:e,attrs:n}=r,{sparseIndices:o,sparseValues:s,defaultValue:i}=t,{outputShape:a}=n,{sliceRank:u,numUpdates:l,sliceSize:c,strides:p,outputSize:m}=S.calculateShapes(s,o,a),f=!1,d=e.bufferSync(o),h=e.bufferSync(s),g=e.data.get(i.dataId).values[0],y=ib(d,h,a,m,c,l,u,p,g,f);return e.makeTensorInfo(a,y.dtype,y.values)}var dP={kernelName:Fp,backendName:"cpu",kernelFunc:YJ};function ZJ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{numOrSizeSplits:s,axis:i}=n,a=x.parseAxisParam(i,o.shape)[0],u=S.prepareSplitSize(o,s,a),l=new Array(o.shape.length).fill(0),c=o.shape.slice();return u.map(p=>{let m=[...c];m[a]=p;let f=Ro({inputs:{x:o},backend:e,attrs:{begin:l,size:m}});return l[a]+=p,f})}var hP={kernelName:Ti,backendName:"cpu",kernelFunc:ZJ};var gP={kernelName:Fl,backendName:"cpu",kernelFunc:({inputs:r,backend:t})=>{let{x:e}=r,n=t;nt(e,"square");let o=n.data.get(e.dataId).values,s=new Float32Array(o.length);for(let a=0;a<o.length;++a){let u=o[a];s[a]=u*u}return{dataId:n.write(s,e.shape,e.dtype),shape:e.shape,dtype:e.dtype}}};var JJ=Et(ao,(r,t)=>{let e=t;return isNaN(r)?NaN:r>0?1:e.alpha}),xP={kernelName:ao,backendName:"cpu",kernelFunc:JJ};function QJ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{begin:s,end:i,strides:a,beginMask:u,endMask:l,ellipsisMask:c,newAxisMask:p,shrinkAxisMask:m}=n;nt(o,"stridedSlice");let{finalShapeSparse:f,finalShape:d,isIdentity:h,sliceDim0:g,isSimpleSlice:y,begin:b,end:w,strides:v}=Ve.sliceInfo(o.shape,s,i,a,u,l,c,p,m),k;if(h)k=ne({inputs:{x:o},backend:e,attrs:{shape:d}});else if(g||y){x.assert(o.shape.length>=1,()=>`Input must have rank at least 1, got: ${o.shape.length}`);let E=Ve.computeOutShape(b,w,v),D=Ro({inputs:{x:o},backend:e,attrs:{begin:b,size:E}});k=ne({inputs:{x:D},backend:e,attrs:{shape:d}}),e.disposeIntermediateTensorInfo(D)}else{let E=e.bufferSync(o),D=jy(f,E,v,b);k=e.makeTensorInfo(d,D.dtype,D.values)}return k}var yP={kernelName:$a,backendName:"cpu",kernelFunc:QJ};function tQ(r){let{inputs:t,backend:e,attrs:n}=r,{separator:o,nGramWidths:s,leftPad:i,rightPad:a,padWidth:u,preserveShortSequences:l}=n,{data:c,dataSplits:p}=t,m=e.data.get(c.dataId).values,f=e.data.get(p.dataId).values,[d,h]=Xy(m,f,o,s,i,a,u,l);return[e.makeTensorInfo([d.length],"string",d),e.makeTensorInfo(p.shape,"int32",h)]}var bP={kernelName:Op,backendName:"cpu",kernelFunc:tQ};function eQ(r){let{inputs:t,backend:e,attrs:n}=r,{skipEmpty:o}=n,{input:s,delimiter:i}=t;if(s.dtype!=="string")throw new Error("Input must be of datatype string");if(s.shape.length!==1)throw new Error(`Input must be a vector, got shape: ${s.shape}`);if(i.shape.length!==0)throw new Error(`Delimiter must be a scalar, got shape: ${i.shape}`);let a=e.data.get(s.dataId).values,u=e.data.get(i.dataId).values[0],[l,c,p]=Yy(a,u,o),m=c.length;return[e.makeTensorInfo([m,2],"int32",l),e.makeTensorInfo([m],"string",c),e.makeTensorInfo([2],"int32",new Int32Array(p))]}var wP={kernelName:Pp,backendName:"cpu",kernelFunc:eQ};function rQ(r){let{inputs:t,backend:e,attrs:n}=r,{numBuckets:o}=n,{input:s}=t;if(s.dtype!=="string")throw new Error("Input must be of datatype string");if(o<=0)throw new Error("Number of buckets must be at least 1");let i=e.data.get(s.dataId).values,a=Zy(i,o);return e.makeTensorInfo(s.shape,"int32",a)}var vP={kernelName:Mp,backendName:"cpu",kernelFunc:rQ};var nQ=Et(zs,r=>Math.tan(r)),CP={kernelName:zs,backendName:"cpu",kernelFunc:nQ};var oQ=Et(Bs,r=>Math.tanh(r)),IP={kernelName:Bs,backendName:"cpu",kernelFunc:oQ};function sQ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{reps:s}=n;nt(o,"tile");let i=Jy(e.bufferSync(o),s);return e.makeTensorInfo(i.shape,i.dtype,i.values)}var SP={kernelName:jn,backendName:"cpu",kernelFunc:sQ};function iQ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{k:s,sorted:i}=n;nt(o,"topk");let a=e.data.get(o.dataId).values,[u,l]=Qy(a,o.shape,o.dtype,s,i);return[e.makeTensorInfo(u.shape,u.dtype,u.values),e.makeTensorInfo(l.shape,l.dtype,l.values)]}var kP={kernelName:Ra,backendName:"cpu",kernelFunc:iQ};function aQ(r){let{inputs:t,attrs:e,backend:n}=r,{image:o,transforms:s}=t,{interpolation:i,fillMode:a,fillValue:u,outputShape:l}=e,[c,p,m,f]=o.shape,[d,h]=l!=null?l:[p,m],g=[c,d,h,f],y=x.computeStrides(o.shape),b=y[0],w=y[1],v=y[2],k=x.getTypedArrayFromDType(o.dtype,x.sizeFromShape(g));k.fill(u);let E=n.data.get(o.dataId).values,D=n.data.get(s.dataId).values;for(let R=0;R<c;++R){let M=s.shape[0]===1?D:D.subarray(R*8,R*8+8);for(let W=0;W<d;++W)for(let U=0;U<h;++U)for(let q=0;q<f;++q){let K,j=M[6]*U+M[7]*W+1;if(j===0)continue;let Q=(M[0]*U+M[1]*W+M[2])/j,rt=(M[3]*U+M[4]*W+M[5])/j,X=NP(Q,m,a),ot=NP(rt,p,a);switch(i){case"nearest":K=mQ(E,p,m,b,w,v,R,ot,X,q,u);break;case"bilinear":K=fQ(E,p,m,b,w,v,R,ot,X,q,u);break;default:throw new Error(`Error in Transform: Expect 'nearest' or 'bilinear', but got ${i}`)}let st=R*b+W*w+U*v+q;k[st]=K}return n.makeTensorInfo(g,o.dtype,k)}return{dataId:n.write(k,g,o.dtype),shape:o.shape,dtype:o.dtype}}var TP={kernelName:Fa,backendName:"cpu",kernelFunc:aQ};function NP(r,t,e){switch(e){case"reflect":return lQ(r,t);case"wrap":return uQ(r,t);case"nearest":return pQ(r,t);case"constant":default:return cQ(r,t)}}function lQ(r,t){let e=r;if(e<0)if(t<=1)e=0;else{let n=2*t;e<n&&(e=n*Math.trunc(-e/n)+e),e=e<-t?e+n:-e-1}else if(e>t-1)if(t<=1)e=0;else{let n=2*t;e-=n*Math.trunc(e/n),e>=t&&(e=n-e-1)}return x.clamp(0,e,t-1)}function uQ(r,t){let e=r;if(e<0)if(t<=1)e=0;else{let n=t-1;e+=t*(Math.trunc(-e/n)+1)}else if(e>t-1)if(t<=1)e=0;else{let n=t-1;e-=t*Math.trunc(e/n)}return x.clamp(0,e,t-1)}function cQ(r,t){return r}function pQ(r,t){return x.clamp(0,r,t-1)}function rg(r,t,e,n,o,s,i,a,u,l,c){let p=i*n+a*o+u*s+l;return 0<=a&&a<t&&0<=u&&u<e?r[p]:c}function mQ(r,t,e,n,o,s,i,a,u,l,c){let p=Math.round(a),m=Math.round(u);return rg(r,t,e,n,o,s,i,p,m,l,c)}function fQ(r,t,e,n,o,s,i,a,u,l,c){let p=Math.floor(a),m=Math.floor(u),f=p+1,d=m+1,h=(d-u)*rg(r,t,e,n,o,s,i,p,m,l,c)+(u-m)*rg(r,t,e,n,o,s,i,p,d,l,c),g=(d-u)*rg(r,t,e,n,o,s,i,f,m,l,c)+(u-m)*rg(r,t,e,n,o,s,i,f,d,l,c);return(f-a)*h+(a-p)*g}function dQ(r){let{inputs:t,attrs:e,backend:n}=r,{axis:o}=e,{x:s}=t;nt(s,"unique");let i=n.data.get(s.dataId).values,{outputValues:a,outputShape:u,indices:l}=tb(i,o,s.shape,s.dtype);return[n.makeTensorInfo(u,s.dtype,a),n.makeTensorInfo([l.length],"int32",l)]}var _P={kernelName:Lp,backendName:"cpu",kernelFunc:dQ};function hQ(r){let{inputs:t,backend:e,attrs:n}=r,{value:o}=t,{axis:s}=n;s<0&&(s+=o.shape.length);let i=o.shape.length,a=o.shape[s],u=new Array(i-1),l=0;for(let f=0;f<i;f++)f!==s&&(u[l++]=o.shape[f]);let c=new Array(i).fill(0),p=o.shape.slice();p[s]=1;let m=new Array(a);for(let f=0;f<m.length;f++){c[s]=f;let d=Ro({inputs:{x:o},backend:e,attrs:{begin:c,size:p}});m[f]=ne({inputs:{x:d},backend:e,attrs:{shape:u}}),e.disposeIntermediateTensorInfo(d)}return m}var EP={kernelName:_i,backendName:"cpu",kernelFunc:hQ};function gQ(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,segmentIds:s}=t,{numSegments:i}=n;nt(o,"unsortedSegmentSum");let a=o.shape.length,u=s.shape.length,l=[],c=[],p=a-u,m=s;for(let d=0;d<p;++d){let h=Vm({inputs:{input:m},backend:e,attrs:{dim:d+1}});m=h,c.push(h)}for(let d=0;d<i;++d){let h=x.createScalarValue(d,"int32"),g=e.makeTensorInfo([],"int32",h),y=aS({inputs:{a:g,b:m},backend:e}),b=Do({inputs:{x:y},backend:e,attrs:{dtype:"float32"}}),w=Nc({inputs:{a:b,b:o},backend:e}),v=ul({inputs:{x:w},backend:e,attrs:{axis:0,keepDims:!1}});l.push(v),c.push(g),c.push(y),c.push(b),c.push(w),c.push(v)}let f=MS({inputs:l,backend:e,attrs:{axis:0}});return c.forEach(d=>e.disposeIntermediateTensorInfo(d)),f}var AP={kernelName:Ol,backendName:"cpu",kernelFunc:gQ};var xQ=[aF,vR,lF,uF,NR,cF,pF,mF,fF,dF,hF,gF,xF,yF,bF,vF,CF,IF,SF,iF,kF,NF,TF,_F,kR,TR,EF,CR,AF,$F,RF,FF,OF,PF,MF,LF,zF,BF,VF,GF,WF,UF,HF,qF,KF,jF,XF,YF,ZF,JF,tO,tF,eO,_R,rO,ER,nO,AR,oO,sO,iO,DR,aO,lO,uO,cO,pO,$R,RR,IR,mO,DF,fO,dO,hO,eF,FR,OR,gO,PR,xO,yO,bO,wO,vO,CO,IO,MR,SO,kO,NO,TO,EO,AO,DO,LR,$O,RO,PO,zR,BR,MO,LO,zO,VR,BO,WO,UO,sb,HO,rF,WR,qO,SR,Qh,KO,nF,oF,sF,jO,XO,YO,ZO,JO,QO,tP,UR,eP,rP,nP,qR,oP,sP,iP,KR,FO,lP,uP,cP,pP,mP,fP,dP,hP,XR,gP,YR,xP,yP,bP,wP,vP,JR,QF,CP,IP,SP,kP,TP,GR,_P,EP,AP,VO];for(let r of xQ)Cu(r);var Um={};Yt(Um,{assertNotComplex:()=>li,bindCanvasToFramebuffer:()=>NQ,bindColorTextureToFramebuffer:()=>ig,bindTextureToProgramUniformSampler:()=>tk,bindTextureUnit:()=>FP,bindVertexBufferToProgramAttribute:()=>pb,callAndCheck:()=>vt,canBeRepresented:()=>GS,createFragmentShader:()=>US,createFramebuffer:()=>ZS,createProgram:()=>HS,createStaticIndexBuffer:()=>jS,createStaticVertexBuffer:()=>KS,createTexture:()=>XS,createVertexShader:()=>WS,getBatchDim:()=>pl,getExtensionOrThrow:()=>Gm,getFramebufferErrorMessage:()=>OP,getMaxTexturesInShader:()=>nk,getNumChannels:()=>SQ,getProgramUniformLocation:()=>QS,getProgramUniformLocationOrThrow:()=>JS,getRowsCols:()=>ml,getShapeAs3D:()=>ag,getTextureShapeFromLogicalShape:()=>ek,getWebGLDisjointQueryTimerVersion:()=>ok,getWebGLErrorMessage:()=>RP,getWebGLMaxTextureSize:()=>rk,hasExtension:()=>Gn,isCapableOfRenderingToFloatTexture:()=>sk,isDownloadFloatTextureEnabled:()=>ik,isReshapeFree:()=>su,isWebGLFenceEnabled:()=>ak,isWebGLVersionEnabled:()=>fb,linkProgram:()=>qS,logShaderSourceAndInfoLog:()=>cb,resetMaxTextureSize:()=>TQ,resetMaxTexturesInShader:()=>_Q,unbindColorTextureFromFramebuffer:()=>mb,unbindTextureUnit:()=>kQ,validateFramebuffer:()=>Wm,validateProgram:()=>sg,validateTextureSize:()=>YS});var Ec={},LS={alpha:!1,antialias:!1,premultipliedAlpha:!1,preserveDrawingBuffer:!1,depth:!1,stencil:!1,failIfMajorPerformanceCaveat:!0};function zS(r,t){Ec[r]=t}function Vn(r,t){if(!(r in Ec)||t!=null){let n=bQ(r,t);if(n!==null)Ec[r]=n;else return console.log("Could not get context for WebGL version",r),null}let e=Ec[r];return e==null||e.isContextLost()?(delete Ec[r],Vn(r)):(e.disable(e.DEPTH_TEST),e.disable(e.STENCIL_TEST),e.disable(e.BLEND),e.disable(e.DITHER),e.disable(e.POLYGON_OFFSET_FILL),e.disable(e.SAMPLE_COVERAGE),e.enable(e.SCISSOR_TEST),e.enable(e.CULL_FACE),e.cullFace(e.BACK),Ec[r])}function yQ(r){if(typeof OffscreenCanvas!="undefined"&&r===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 bQ(r,t){if(r!==1&&r!==2)throw new Error("Cannot get WebGL rendering context, WebGL is disabled.");let e=t==null?yQ(r):t;return e.addEventListener("webglcontextlost",n=>{n.preventDefault(),delete Ec[r]},!1),r===1?e.getContext("webgl",LS)||e.getContext("experimental-webgl",LS):e.getContext("webgl2",LS)}var ou;(function(r){r[r.DENSE=0]="DENSE",r[r.SHARED_BATCH=1]="SHARED_BATCH"})(ou||(ou={}));var Ur;(function(r){r[r.RENDER=0]="RENDER",r[r.UPLOAD=1]="UPLOAD",r[r.PIXELS=2]="PIXELS",r[r.DOWNLOAD=3]="DOWNLOAD"})(Ur||(Ur={}));var Rr;(function(r){r[r.UNPACKED_FLOAT16=0]="UNPACKED_FLOAT16",r[r.UNPACKED_FLOAT32=1]="UNPACKED_FLOAT32",r[r.PACKED_4X1_UNSIGNED_BYTE=2]="PACKED_4X1_UNSIGNED_BYTE",r[r.PACKED_2X2_FLOAT32=3]="PACKED_2X2_FLOAT32",r[r.PACKED_2X2_FLOAT16=4]="PACKED_2X2_FLOAT16"})(Rr||(Rr={}));function Ac(r,t){return[t,r]}function DP(r,t){return r*t}function ng(r){let t=x.sizeFromShape(r),e=Math.ceil(t/4);return x.sizeToSquarishShape(e)}function Hi(r,t){return[Math.max(1,Math.ceil(t/2)),Math.max(1,Math.ceil(r/2))]}function $P(r,t){let[e,n]=Hi(r,t);return e*n*4}function og(r,t){let e=r,n,o,s,i,a,u,l,c,p,m;return V().getNumber("WEBGL_VERSION")===2?(n=e.R32F,o=e.R16F,s=e.RGBA16F,i=e.RGBA32F,a=e.RED,l=4,c=1,p=e.HALF_FLOAT,m=e.FLOAT,u=e.RGBA8):(n=r.RGBA,o=r.RGBA,s=r.RGBA,i=e.RGBA,a=r.RGBA,l=4,c=4,p=t!=null?t.HALF_FLOAT_OES:null,m=r.FLOAT,u=r.RGBA),{internalFormatFloat:n,internalFormatHalfFloat:o,internalFormatPackedHalfFloat:s,internalFormatPackedFloat:i,textureFormatFloat:a,downloadTextureFormat:u,downloadUnpackNumChannels:l,defaultNumChannels:c,textureTypeHalfFloat:p,textureTypeFloat:m}}function vt(r,t){let e=t();return V().getBool("DEBUG")&&wQ(r),e}function wQ(r){let t=r.getError();if(t!==r.NO_ERROR)throw new Error("WebGL Error: "+RP(r,t))}var vQ=596e-10,CQ=65504;function GS(r){return!!(V().getBool("WEBGL_RENDER_FLOAT32_ENABLED")||r===0||vQ<Math.abs(r)&&Math.abs(r)<CQ)}function RP(r,t){switch(t){case r.NO_ERROR:return"NO_ERROR";case r.INVALID_ENUM:return"INVALID_ENUM";case r.INVALID_VALUE:return"INVALID_VALUE";case r.INVALID_OPERATION:return"INVALID_OPERATION";case r.INVALID_FRAMEBUFFER_OPERATION:return"INVALID_FRAMEBUFFER_OPERATION";case r.OUT_OF_MEMORY:return"OUT_OF_MEMORY";case r.CONTEXT_LOST_WEBGL:return"CONTEXT_LOST_WEBGL";default:return`Unknown error code ${t}`}}function Gm(r,t){return cl(r,()=>r.getExtension(t),'Extension "'+t+'" not supported on this browser.')}function WS(r,t){let e=cl(r,()=>r.createShader(r.VERTEX_SHADER),"Unable to create vertex WebGLShader.");if(vt(r,()=>r.shaderSource(e,t)),vt(r,()=>r.compileShader(e)),r.getShaderParameter(e,r.COMPILE_STATUS)===!1)throw console.log(r.getShaderInfoLog(e)),new Error("Failed to compile vertex shader.");return e}function US(r,t){let e=cl(r,()=>r.createShader(r.FRAGMENT_SHADER),"Unable to create fragment WebGLShader.");if(vt(r,()=>r.shaderSource(e,t)),vt(r,()=>r.compileShader(e)),V().get("ENGINE_COMPILE_ONLY"))return e;if(r.getShaderParameter(e,r.COMPILE_STATUS)===!1)throw cb(t,r.getShaderInfoLog(e)),new Error("Failed to compile fragment shader.");return e}var IQ=/ERROR: [0-9]+:([0-9]+):/g;function cb(r,t){let e=IQ.exec(t);if(e==null){console.log(`Couldn't parse line number in error: ${t}`),console.log(r);return}let n=+e[1],o=r.split(`
`),s=o.length.toString().length+2,i=o.map((p,m)=>x.rightPad((m+1).toString(),s)+p),a=0;for(let p=0;p<i.length;p++)a=Math.max(i[p].length,a);let u=i.slice(0,n-1),l=i.slice(n-1,n),c=i.slice(n);console.log(u.join(`
`)),console.log(t.split(`
`)[0]),console.log(`%c ${x.rightPad(l[0],a)}`,"border:1px solid red; background-color:#e3d2d2; color:#a61717"),console.log(c.join(`
`))}function HS(r){return cl(r,()=>r.createProgram(),"Unable to create WebGLProgram.")}function qS(r,t){if(vt(r,()=>r.linkProgram(t)),!V().get("ENGINE_COMPILE_ONLY")&&r.getProgramParameter(t,r.LINK_STATUS)===!1)throw console.log(r.getProgramInfoLog(t)),new Error("Failed to link vertex and fragment shaders.")}function sg(r,t){if(vt(r,()=>r.validateProgram(t)),r.getProgramParameter(t,r.VALIDATE_STATUS)===!1)throw console.log(r.getProgramInfoLog(t)),new Error("Shader program validation failed.")}function KS(r,t){let e=cl(r,()=>r.createBuffer(),"Unable to create WebGLBuffer");return vt(r,()=>r.bindBuffer(r.ARRAY_BUFFER,e)),vt(r,()=>r.bufferData(r.ARRAY_BUFFER,t,r.STATIC_DRAW)),e}function jS(r,t){let e=cl(r,()=>r.createBuffer(),"Unable to create WebGLBuffer");return vt(r,()=>r.bindBuffer(r.ELEMENT_ARRAY_BUFFER,e)),vt(r,()=>r.bufferData(r.ELEMENT_ARRAY_BUFFER,t,r.STATIC_DRAW)),e}function SQ(){return V().getNumber("WEBGL_VERSION")===2?1:4}function XS(r){return cl(r,()=>r.createTexture(),"Unable to create WebGLTexture.")}function YS(r,t){let e=V().getNumber("WEBGL_MAX_TEXTURE_SIZE");if(r<=0||t<=0){let n=`[${r}x${t}]`;throw new Error("Requested texture size "+n+" is invalid.")}if(r>e||t>e){let n=`[${r}x${t}]`,o=`[${e}x${e}]`;throw new Error("Requested texture size "+n+" greater than WebGL maximum on this browser / GPU "+o+".")}}function ZS(r){return cl(r,()=>r.createFramebuffer(),"Unable to create WebGLFramebuffer.")}function pb(r,t,e,n,o,s,i){let a=r.getAttribLocation(t,e);return a===-1?!1:(vt(r,()=>r.bindBuffer(r.ARRAY_BUFFER,n)),vt(r,()=>r.vertexAttribPointer(a,o,r.FLOAT,!1,s,i)),vt(r,()=>r.enableVertexAttribArray(a)),!0)}function FP(r,t,e){PP(r,e),vt(r,()=>r.activeTexture(r.TEXTURE0+e)),vt(r,()=>r.bindTexture(r.TEXTURE_2D,t))}function kQ(r,t){PP(r,t),vt(r,()=>r.activeTexture(r.TEXTURE0+t)),vt(r,()=>r.bindTexture(r.TEXTURE_2D,null))}function JS(r,t,e){return cl(r,()=>r.getUniformLocation(t,e),'uniform "'+e+'" not present in program.')}function QS(r,t,e){return r.getUniformLocation(t,e)}function tk(r,t,e,n){vt(r,()=>FP(r,t,n)),vt(r,()=>r.uniform1i(e,n))}function NQ(r){vt(r,()=>r.bindFramebuffer(r.FRAMEBUFFER,null)),vt(r,()=>r.viewport(0,0,r.canvas.width,r.canvas.height)),vt(r,()=>r.scissor(0,0,r.canvas.width,r.canvas.height))}function ig(r,t,e){vt(r,()=>r.bindFramebuffer(r.FRAMEBUFFER,e)),vt(r,()=>r.framebufferTexture2D(r.FRAMEBUFFER,r.COLOR_ATTACHMENT0,r.TEXTURE_2D,t,0))}function mb(r,t){vt(r,()=>r.bindFramebuffer(r.FRAMEBUFFER,t)),vt(r,()=>r.framebufferTexture2D(r.FRAMEBUFFER,r.COLOR_ATTACHMENT0,r.TEXTURE_2D,null,0))}function Wm(r){let t=r.checkFramebufferStatus(r.FRAMEBUFFER);if(t!==r.FRAMEBUFFER_COMPLETE)throw new Error("Error binding framebuffer: "+OP(r,t))}function OP(r,t){switch(t){case r.FRAMEBUFFER_INCOMPLETE_ATTACHMENT:return"FRAMEBUFFER_INCOMPLETE_ATTACHMENT";case r.FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT:return"FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT";case r.FRAMEBUFFER_INCOMPLETE_DIMENSIONS:return"FRAMEBUFFER_INCOMPLETE_DIMENSIONS";case r.FRAMEBUFFER_UNSUPPORTED:return"FRAMEBUFFER_UNSUPPORTED";default:return`unknown error ${t}`}}function cl(r,t,e){let n=vt(r,()=>t());if(n==null)throw new Error(e);return n}function PP(r,t){let e=r.MAX_COMBINED_TEXTURE_IMAGE_UNITS-1,n=t+r.TEXTURE0;if(n<r.TEXTURE0||n>e){let o=`[gl.TEXTURE0, gl.TEXTURE${e}]`;throw new Error(`textureUnit must be in ${o}.`)}}function pl(r,t=2){return x.sizeFromShape(r.slice(0,r.length-t))}function ml(r){if(r.length===0)throw Error("Cannot get rows and columns of an empty shape array.");return[r.length>1?r[r.length-2]:1,r[r.length-1]]}function ag(r){let t=[1,1,1];return r.length===0||r.length===1&&r[0]===1||(t=[pl(r),...ml(r)]),t}function ek(r,t=!1){let e=V().getNumber("WEBGL_MAX_TEXTURE_SIZE");t&&(e=e*2,r=r.map((o,s)=>s>=r.length-2?x.nearestLargerEven(r[s]):r[s]),r.length===1&&(r=[2,r[0]])),r.length!==2&&(r=x.squeezeShape(r).newShape);let n=x.sizeFromShape(r);if(r.length<=1&&n<=e)return[1,n];if(r.length===2&&r[0]<=e&&r[1]<=e)return r;if(r.length===3&&r[0]*r[1]<=e&&r[2]<=e)return[r[0]*r[1],r[2]];if(r.length===3&&r[0]<=e&&r[1]*r[2]<=e)return[r[0],r[1]*r[2]];if(r.length===4&&r[0]*r[1]*r[2]<=e&&r[3]<=e)return[r[0]*r[1]*r[2],r[3]];if(r.length===4&&r[0]<=e&&r[1]*r[2]*r[3]<=e)return[r[0],r[1]*r[2]*r[3]];if(t){let o=pl(r),s=2,i=2;return r.length&&([s,i]=ml(r)),n=o*(s/2)*(i/2),x.sizeToSquarishShape(n).map(a=>a*2)}return x.sizeToSquarishShape(n)}function ab(r){return r%2===0}function su(r,t){if(r=r.slice(-2),t=t.slice(-2),x.arraysEqual(r,t)||!r.length||!t.length||r[0]===0||r[1]===0||t[0]===0||t[1]===0)return!0;if(r.length!==t.length){let e=r.slice(-1)[0],n=t.slice(-1)[0];if(e===n||ab(e)&&ab(n)&&(r[0]===1||t[0]===1))return!0}return r[1]===t[1]&&ab(r[0])&&ab(t[0])}var lb,ub;function rk(r){if(lb==null){let t=Vn(r);lb=t.getParameter(t.MAX_TEXTURE_SIZE)}return lb}function TQ(){lb=null}function _Q(){ub=null}function nk(r){if(ub==null){let t=Vn(r);ub=t.getParameter(t.MAX_TEXTURE_IMAGE_UNITS)}return Math.min(16,ub)}function ok(r){if(r===0)return 0;let t,e=Vn(r);return Gn(e,"EXT_disjoint_timer_query_webgl2")&&r===2?t=2:Gn(e,"EXT_disjoint_timer_query")?t=1:t=0,t}function Gn(r,t){return r.getExtension(t)!=null}function fb(r){try{if(Vn(r)!=null)return!0}catch(t){return console.log("Error when getting WebGL context: ",t),!1}return!1}function sk(r){if(r===0)return!1;let t=Vn(r);if(r===1){if(!Gn(t,"OES_texture_float"))return!1}else if(!Gn(t,"EXT_color_buffer_float"))return!1;return VS(t)}function ik(r){if(r===0)return!1;let t=Vn(r);if(r===1){if(!Gn(t,"OES_texture_float")||!Gn(t,"WEBGL_color_buffer_float"))return!1}else{if(Gn(t,"EXT_color_buffer_float"))return VS(t);let n="EXT_color_buffer_half_float";if(Gn(t,n)){let o=t.getExtension(n);return EQ(t,o)}return!1}return VS(t)}function VS(r){let t=og(r),e=r.createTexture();r.bindTexture(r.TEXTURE_2D,e);let n=1,o=1;r.texImage2D(r.TEXTURE_2D,0,t.internalFormatFloat,n,o,0,t.textureFormatFloat,t.textureTypeFloat,null);let s=r.createFramebuffer();r.bindFramebuffer(r.FRAMEBUFFER,s),r.framebufferTexture2D(r.FRAMEBUFFER,r.COLOR_ATTACHMENT0,r.TEXTURE_2D,e,0);let i=r.checkFramebufferStatus(r.FRAMEBUFFER)===r.FRAMEBUFFER_COMPLETE;return r.bindTexture(r.TEXTURE_2D,null),r.bindFramebuffer(r.FRAMEBUFFER,null),r.deleteTexture(e),r.deleteFramebuffer(s),i}function EQ(r,t){let e=og(r,t),n=r.createTexture();r.bindTexture(r.TEXTURE_2D,n);let o=1,s=1;r.texImage2D(r.TEXTURE_2D,0,e.internalFormatHalfFloat,o,s,0,e.textureFormatFloat,e.textureTypeHalfFloat,null);let i=r.createFramebuffer();r.bindFramebuffer(r.FRAMEBUFFER,i),r.framebufferTexture2D(r.FRAMEBUFFER,r.COLOR_ATTACHMENT0,r.TEXTURE_2D,n,0);let a=r.checkFramebufferStatus(r.FRAMEBUFFER)===r.FRAMEBUFFER_COMPLETE;return r.bindTexture(r.TEXTURE_2D,null),r.bindFramebuffer(r.FRAMEBUFFER,null),r.deleteTexture(n),r.deleteFramebuffer(i),a}function ak(r){return r!==2?!1:Vn(r).fenceSync!=null}function li(r,t){Array.isArray(r)||(r=[r]),r.forEach(e=>{e!=null&&x.assert(e.dtype!=="complex64",()=>`${t} does not support complex64 tensors in the WebGL backend.`)})}var Ft=V();Ft.registerFlag("HAS_WEBGL",()=>Ft.getNumber("WEBGL_VERSION")>0);Ft.registerFlag("WEBGL_VERSION",()=>fb(2)?2:fb(1)?1:0);Ft.registerFlag("WEBGL_CHECK_NUMERICAL_PROBLEMS",()=>!1);Ft.registerFlag("WEBGL_BUFFER_SUPPORTED",()=>Ft.get("WEBGL_VERSION")===2);Ft.registerFlag("WEBGL_CPU_FORWARD",()=>!0);Ft.registerFlag("WEBGL_FORCE_F16_TEXTURES",()=>!1);Ft.registerFlag("WEBGL_PACK",()=>Ft.getBool("HAS_WEBGL"));Ft.registerFlag("WEBGL_PACK_NORMALIZATION",()=>Ft.getBool("WEBGL_PACK"));Ft.registerFlag("WEBGL_PACK_CLIP",()=>Ft.getBool("WEBGL_PACK"));Ft.registerFlag("WEBGL_PACK_DEPTHWISECONV",()=>Ft.getBool("WEBGL_PACK"));Ft.registerFlag("WEBGL_PACK_BINARY_OPERATIONS",()=>Ft.getBool("WEBGL_PACK"));Ft.registerFlag("WEBGL_PACK_UNARY_OPERATIONS",()=>Ft.getBool("WEBGL_PACK"));Ft.registerFlag("WEBGL_PACK_ARRAY_OPERATIONS",()=>Ft.getBool("WEBGL_PACK"));Ft.registerFlag("WEBGL_PACK_IMAGE_OPERATIONS",()=>Ft.getBool("WEBGL_PACK"));Ft.registerFlag("WEBGL_PACK_REDUCE",()=>Ft.getBool("WEBGL_PACK"));Ft.registerFlag("WEBGL_LAZILY_UNPACK",()=>Ft.getBool("WEBGL_PACK"));Ft.registerFlag("WEBGL_CONV_IM2COL",()=>Ft.getBool("WEBGL_PACK"));Ft.registerFlag("WEBGL_MAX_TEXTURE_SIZE",()=>rk(Ft.getNumber("WEBGL_VERSION")));Ft.registerFlag("WEBGL_MAX_TEXTURES_IN_SHADER",()=>nk(Ft.getNumber("WEBGL_VERSION")));Ft.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION",()=>{let r=Ft.getNumber("WEBGL_VERSION");return r===0?0:ok(r)});Ft.registerFlag("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE",()=>Ft.getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0&&!zl.isMobile());Ft.registerFlag("WEBGL_RENDER_FLOAT32_CAPABLE",()=>sk(Ft.getNumber("WEBGL_VERSION")));Ft.registerFlag("WEBGL_RENDER_FLOAT32_ENABLED",()=>Ft.getBool("WEBGL_FORCE_F16_TEXTURES")?!1:Ft.getBool("WEBGL_RENDER_FLOAT32_CAPABLE"));Ft.registerFlag("WEBGL_DOWNLOAD_FLOAT_ENABLED",()=>ik(Ft.getNumber("WEBGL_VERSION")));Ft.registerFlag("WEBGL_FENCE_API_ENABLED",()=>ak(Ft.getNumber("WEBGL_VERSION")));Ft.registerFlag("WEBGL_SIZE_UPLOAD_UNIFORM",()=>Ft.getBool("WEBGL_RENDER_FLOAT32_ENABLED")?4:0);Ft.registerFlag("WEBGL_DELETE_TEXTURE_THRESHOLD",()=>-1,r=>{if(r<0&&r!==-1)throw new Error(`WEBGL_DELETE_TEXTURE_THRESHOLD must be -1 (indicating never delete) or at least 0, but got ${r}.`)});Ft.registerFlag("WEBGL_FLUSH_THRESHOLD",()=>zl.isMobile()?1:-1,r=>{if(r<0&&r!==-1)throw new Error(`WEBGL_FLUSH_THRESHOLD must be -1 (indicating never manual flush) or at least 0, but got ${r}.`)});Ft.registerFlag("CPU_HANDOFF_SIZE_THRESHOLD",()=>128);Ft.registerFlag("WEBGL_USE_SHAPES_UNIFORMS",()=>!1);Ft.registerFlag("TOPK_LAST_DIM_CPU_HANDOFF_SIZE_THRESHOLD",()=>1e5);Ft.registerFlag("TOPK_K_CPU_HANDOFF_THRESHOLD",()=>128);function qe(){let r,t,e,n,o,s,i,a,u,l;return V().getNumber("WEBGL_VERSION")===2?(r="#version 300 es",t="in",e="out",n="in",o="texture",s="outputColor",i="out vec4 outputColor;",a=`
bool isnan_custom(float val) {
uint floatToUint = floatBitsToUint(val);
return (floatToUint & 0x7fffffffu) > 0x7f800000u;
}
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)
`,u="",l=`
#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)));
}
`):(r="",t="attribute",e="varying",n="varying",o="texture2D",s="gl_FragColor",i="",a=`
#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));
}
`,u=`
uniform float INFINITY;
bool isinf(float val) {
return abs(val) == INFINITY;
}
bvec4 isinf(vec4 val) {
return equal(abs(val), vec4(INFINITY));
}
`,l=`
int round(float value) {
return int(floor(value + 0.5));
}
ivec4 round(vec4 value) {
return ivec4(floor(value + vec4(0.5)));
}
`),{version:r,attribute:t,varyingVs:e,varyingFs:n,texture2D:o,output:s,defineOutput:i,defineSpecialNaN:a,defineSpecialInf:u,defineRound:l}}function ui(r,t,e="index"){let n=x.computeStrides(t);return n.map((o,s)=>{let i=`int ${r[s]} = ${e} / ${o}`,a=s===n.length-1?`int ${r[s+1]} = ${e} - ${r[s]} * ${o}`:`index -= ${r[s]} * ${o}`;return`${i}; ${a};`}).join("")}function Dc(r,t,e="index"){let n=x.computeStrides(t);return n.map((o,s)=>{let i=`int ${r[s]} = ${e} / outShapeStrides[${s}]`,a=s===n.length-1?`int ${r[s+1]} = ${e} - ${r[s]} * outShapeStrides[${s}]`:`index -= ${r[s]} * outShapeStrides[${s}]`;return`${i}; ${a};`}).join("")}function AQ(r,t){let e=r.length,n=r.map(s=>`${t}[${s}]`),o=new Array(e-1);o[e-2]=n[e-1];for(let s=e-3;s>=0;--s)o[s]=`(${o[s+1]} * ${n[s+1]})`;return o}function MP(r,t,e="index"){let n=r.map((s,i)=>i),o=AQ(n,t);return o.map((s,i)=>{let a=`int ${r[i]} = ${e} / ${o[i]}`,u=i===o.length-1?`int ${r[i+1]} = ${e} - ${r[i]} * ${o[i]}`:`index -= ${r[i]} * ${o[i]}`;return`${a}; ${u};`}).join("")}function Hm(r){let t=x.computeStrides(r).map(e=>e.toString());return`
int getFlatIndex(ivec3 coords) {
return coords.x * ${t[0]} + coords.y * ${t[1]} + coords.z;
}
`}function qm(){return`
int getFlatIndex(ivec3 coords) {
return coords.x * outShapeStrides[0] + coords.y * outShapeStrides[1] + coords.z;
}
`}var db=`
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;
}
`;var{getBroadcastDims:LP}=S;function zP(r,t,e){let n=[];if(r.forEach(f=>{let d=x.sizeFromShape(f.shapeInfo.logicalShape);if(f.shapeInfo.isUniform?n.push(`uniform float ${f.name}${d>1?`[${d}]`:""};`):(n.push(`uniform sampler2D ${f.name};`),n.push(`uniform int offset${f.name};`)),e.enableShapeUniforms){let{uniformShape:h}=hb(e.packedInputs,f.shapeInfo.logicalShape,f.shapeInfo.texShape);switch(h.length){case 1:n.push(`uniform int ${f.name}Shape;`);break;case 2:n.push(`uniform ivec2 ${f.name}Shape;`);break;case 3:n.push(`uniform ivec3 ${f.name}Shape;`);break;case 4:n.push(`uniform ivec4 ${f.name}Shape;`);break;default:break}n.push(`uniform ivec2 ${f.name}TexShape;`)}}),e.enableShapeUniforms){switch(t.logicalShape.length){case 1:n.push("uniform int outShape;");break;case 2:n.push("uniform ivec2 outShape;"),n.push("uniform int outShapeStrides;");break;case 3:n.push("uniform ivec3 outShape;"),n.push("uniform ivec2 outShapeStrides;");break;case 4:n.push("uniform ivec4 outShape;"),n.push("uniform ivec3 outShapeStrides;");break;default:break}n.push("uniform ivec2 outTexShape;")}e.customUniforms&&e.customUniforms.forEach(f=>{n.push(`uniform ${f.type} ${f.name}${f.arrayIndex?`[${f.arrayIndex}]`:""};`)});let o=n.join(`
`),s=r.map(f=>DQ(f,t,e.packedInputs,e.enableShapeUniforms)).join(`
`),i=t.texShape,a=qe(),u=FQ(a),l,c,p=MQ(a);return t.isPacked?(l=$Q(t.logicalShape,i,e.enableShapeUniforms),c=PQ(a)):(l=RQ(t.logicalShape,i,e.enableShapeUniforms),c=OQ(a)),e.packedInputs&&(p+=VQ),[p,u,c,o,l,s,e.userCode].join(`
`)}function jm(r,t=!1){let e=r.shapeInfo.logicalShape;switch(e.length){case 0:return QQ(r,t);case 1:return ett(r,t);case 2:return ntt(r,t);case 3:return stt(r,t);case 4:return att(r,t);case 5:return ltt(r);case 6:return utt(r);default:throw new Error(`${e.length}-D input sampling is not yet supported`)}}function BP(r,t){switch(r.shapeInfo.logicalShape.length){case 0:return JQ(r);case 1:return ttt(r,t);case 2:return rtt(r,t);case 3:return ott(r,t);default:return itt(r,t)}}function DQ(r,t,e=!1,n){let o="";e?o+=BP(r,n):o+=jm(r,n);let s=r.shapeInfo.logicalShape,i=t.logicalShape;return s.length<=i.length&&(e?o+=ctt(r,t):o+=ptt(r,t)),o}function $Q(r,t,e){switch(r.length){case 0:return VP();case 1:return GQ(r,t,e);case 2:return YQ(r,t,e);case 3:return UQ(r,t,e);default:return qQ(r,t,e)}}function RQ(r,t,e){switch(r.length){case 0:return VP();case 1:return WQ(r,t,e);case 2:return ZQ(r,t,e);case 3:return HQ(r,t,e);case 4:return KQ(r,t,e);case 5:return jQ(r,t);case 6:return XQ(r,t);default:throw new Error(`${r.length}-D output sampling is not yet supported`)}}function FQ(r){return`
float sampleTexture(sampler2D textureSampler, vec2 uv) {
return ${r.texture2D}(textureSampler, uv).r;
}
`}function OQ(r){return`
void setOutput(float val) {
${r.output} = vec4(val, 0, 0, 0);
}
`}function PQ(r){return`
void setOutput(vec4 val) {
${r.output} = val;
}
`}function MQ(r){return`${r.version}
precision highp float;
precision highp int;
precision highp sampler2D;
${r.varyingFs} vec2 resultUV;
${r.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;
${r.defineSpecialNaN}
${r.defineSpecialInf}
${r.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);
}
${LQ}
${zQ}
${BQ}
`}var LQ=`
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);
}
`,zQ=`
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);
}
`,BQ=`
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);
}
`,VQ=`
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 VP(){return`
int getOutputCoords() {
return 0;
}
`}function GQ(r,t,e){let n=[Math.ceil(t[0]/2),Math.ceil(t[1]/2)];return n[0]===1?e?`
int getOutputCoords() {
return 2 * int(resultUV.x * ceil(float(outTexShape[1]) / 2.0));
}
`:`
int getOutputCoords() {
return 2 * int(resultUV.x * ${n[1]}.0);
}
`:n[1]===1?e?`
int getOutputCoords() {
return 2 * int(resultUV.y * ceil(float(outTexShape[0]) / 2.0));
}
`:`
int getOutputCoords() {
return 2 * int(resultUV.y * ${n[0]}.0);
}
`:e?`
int getOutputCoords() {
ivec2 packedTexShape = ivec2(ceil(float(outTexShape[0]) / 2.0), ceil(float(outTexShape[1]) / 2.0));
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(packedTexShape[0], packedTexShape[1]));
return 2 * (resTexRC.x * packedTexShape[1] + resTexRC.y);
}
`:`
int getOutputCoords() {
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(${n[0]}, ${n[1]}));
return 2 * (resTexRC.x * ${n[1]} + resTexRC.y);
}
`}function WQ(r,t,e){return t[0]===1?e?`
int getOutputCoords() {
return int(resultUV.x * float(outTexShape[1]));
}
`:`
int getOutputCoords() {
return int(resultUV.x * ${t[1]}.0);
}
`:t[1]===1?e?`
int getOutputCoords() {
return int(resultUV.y * float(outTexShape[0]));
}
`:`
int getOutputCoords() {
return int(resultUV.y * ${t[0]}.0);
}
`:e?`
int getOutputCoords() {
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(outTexShape[0], outTexShape[1]));
return resTexRC.x * outTexShape[1] + resTexRC.y;
}
`:`
int getOutputCoords() {
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(${t[0]}, ${t[1]}));
return resTexRC.x * ${t[1]} + resTexRC.y;
}
`}function UQ(r,t,e){if(e)return`
ivec3 getOutputCoords() {
ivec2 packedTexShape = ivec2(ceil(float(outTexShape[0]) / 2.0), ceil(float(outTexShape[1]) / 2.0));
int texelsInLogicalRow = int(ceil(float(outShape[2]) / 2.0));
int texelsInBatch = texelsInLogicalRow * int(ceil(float(outShape[1]) / 2.0));
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(packedTexShape[0], packedTexShape[1]));
int index = resTexRC.x * packedTexShape[1] + resTexRC.y;
int b = index / texelsInBatch;
index -= b * texelsInBatch;
int r = 2 * (index / texelsInLogicalRow);
int c = imod(index, texelsInLogicalRow) * 2;
return ivec3(b, r, c);
}
`;let n=[Math.ceil(t[0]/2),Math.ceil(t[1]/2)],o=Math.ceil(r[2]/2),s=o*Math.ceil(r[1]/2);return`
ivec3 getOutputCoords() {
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(${n[0]}, ${n[1]}));
int index = resTexRC.x * ${n[1]} + resTexRC.y;
int b = index / ${s};
index -= b * ${s};
int r = 2 * (index / ${o});
int c = imod(index, ${o}) * 2;
return ivec3(b, r, c);
}
`}function HQ(r,t,e){if(e)return`
ivec3 getOutputCoords() {
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(outTexShape[0], outTexShape[1]));
int index = resTexRC.x * outTexShape[1] + resTexRC.y;
${Dc(["r","c","d"],r)}
return ivec3(r, c, d);
}
`;let n=ui(["r","c","d"],r);return`
ivec3 getOutputCoords() {
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(${t[0]}, ${t[1]}));
int index = resTexRC.x * ${t[1]} + resTexRC.y;
${n}
return ivec3(r, c, d);
}
`}function qQ(r,t,e){if(e)return`
ivec4 getOutputCoords() {
ivec2 packedTexShape = ivec2(ceil(float(outTexShape[0]) / 2.0), ceil(float(outTexShape[1]) / 2.0));
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(packedTexShape[0], packedTexShape[1]));
int index = resTexRC.x * packedTexShape[1] + resTexRC.y;
int texelsInLogicalRow = int(ceil(float(outShape[3]) / 2.0));
int texelsInBatch = texelsInLogicalRow * int(ceil(float(outShape[2]) / 2.0));
int texelsInBatchN = texelsInBatch * outShape[1];
int b2 = index / texelsInBatchN;
index -= b2 * texelsInBatchN;
int b = index / texelsInBatch;
index -= b * texelsInBatch;
int r = 2 * (index / texelsInLogicalRow);
int c = imod(index, texelsInLogicalRow) * 2;
return ivec4(b2, b, r, c);
}
`;let n=[Math.ceil(t[0]/2),Math.ceil(t[1]/2)],o=Math.ceil(r[r.length-1]/2),s=o*Math.ceil(r[r.length-2]/2),i=s,a="",u="b, r, c";for(let l=2;l<r.length-1;l++)i*=r[r.length-l-1],a=`
int b${l} = index / ${i};
index -= b${l} * ${i};
`+a,u=`b${l}, `+u;return`
ivec${r.length} getOutputCoords() {
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(${n[0]}, ${n[1]}));
int index = resTexRC.x * ${n[1]} + resTexRC.y;
${a}
int b = index / ${s};
index -= b * ${s};
int r = 2 * (index / ${o});
int c = imod(index, ${o}) * 2;
return ivec${r.length}(${u});
}
`}function KQ(r,t,e){if(e)return`
ivec4 getOutputCoords() {
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(outTexShape[0], outTexShape[1]));
int index = resTexRC.x * outTexShape[1] + resTexRC.y;
${Dc(["r","c","d","d2"],r)}
return ivec4(r, c, d, d2);
}
`;let n=ui(["r","c","d","d2"],r);return`
ivec4 getOutputCoords() {
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(${t[0]}, ${t[1]}));
int index = resTexRC.x * ${t[1]} + resTexRC.y;
${n}
return ivec4(r, c, d, d2);
}
`}function jQ(r,t){let e=ui(["r","c","d","d2","d3"],r);return`
ivec5 getOutputCoords() {
ivec2 resTexRC = ivec2(resultUV.yx * vec2(${t[0]},
${t[1]}));
int index = resTexRC.x * ${t[1]} + resTexRC.y;
${e}
ivec5 outShape = ivec5(r, c, d, d2, d3);
return outShape;
}
`}function XQ(r,t){let e=ui(["r","c","d","d2","d3","d4"],r);return`
ivec6 getOutputCoords() {
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(${t[0]}, ${t[1]}));
int index = resTexRC.x * ${t[1]} + resTexRC.y;
${e}
ivec6 result = ivec6(r, c, d, d2, d3, d4);
return result;
}
`}function YQ(r,t,e){let n=[Math.ceil(t[0]/2),Math.ceil(t[1]/2)];if(x.arraysEqual(r,t))return e?`
ivec2 getOutputCoords() {
ivec2 packedTexShape = ivec2(ceil(float(outTexShape[0]) / 2.0), ceil(float(outTexShape[1]) / 2.0));
return 2 * ivec2(resultUV.yx * vec2(packedTexShape[0], packedTexShape[1]));
}
`:`
ivec2 getOutputCoords() {
return 2 * ivec2(resultUV.yx * vec2(${n[0]}, ${n[1]}));
}
`;let o=Math.ceil(r[1]/2);return e?`
ivec2 getOutputCoords() {
ivec2 packedTexShape = ivec2(ceil(float(outTexShape[0]) / 2.0), ceil(float(outTexShape[1]) / 2.0));
int texelsInLogicalRow = int(ceil(float(outShape[1]) / 2.0));
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(packedTexShape[0], packedTexShape[1]));
int index = resTexRC.x * packedTexShape[1] + resTexRC.y;
int r = 2 * (index / texelsInLogicalRow);
int c = imod(index, texelsInLogicalRow) * 2;
return ivec2(r, c);
}
`:`
ivec2 getOutputCoords() {
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(${n[0]}, ${n[1]}));
int index = resTexRC.x * ${n[1]} + resTexRC.y;
int r = 2 * (index / ${o});
int c = imod(index, ${o}) * 2;
return ivec2(r, c);
}
`}function ZQ(r,t,e){return x.arraysEqual(r,t)?e?`
ivec2 getOutputCoords() {
return ivec2(resultUV.yx * vec2(outTexShape[0], outTexShape[1]));
}
`:`
ivec2 getOutputCoords() {
return ivec2(resultUV.yx * vec2(${t[0]}, ${t[1]}));
}
`:r[1]===1?e?`
ivec2 getOutputCoords() {
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(outTexShape[0], outTexShape[1]));
int index = resTexRC.x * outTexShape[1] + resTexRC.y;
return ivec2(index, 0);
}
`:`
ivec2 getOutputCoords() {
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(${t[0]}, ${t[1]}));
int index = resTexRC.x * ${t[1]} + resTexRC.y;
return ivec2(index, 0);
}
`:r[0]===1?e?`
ivec2 getOutputCoords() {
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(outTexShape[0], outTexShape[1]));
int index = resTexRC.x * outTexShape[1] + resTexRC.y;
return ivec2(0, index);
}
`:`
ivec2 getOutputCoords() {
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(${t[0]}, ${t[1]}));
int index = resTexRC.x * ${t[1]} + resTexRC.y;
return ivec2(0, index);
}
`:e?`
ivec2 getOutputCoords() {
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(outTexShape[0], outTexShape[1]));
int index = resTexRC.x * outTexShape[1] + resTexRC.y;
int r = index / outShape[1];
int c = index - r * outShape[1];
return ivec2(r, c);
}
`:`
ivec2 getOutputCoords() {
ivec2 resTexRC = ivec2(resultUV.yx *
vec2(${t[0]}, ${t[1]}));
int index = resTexRC.x * ${t[1]} + resTexRC.y;
int r = index / ${r[1]};
int c = index - r * ${r[1]};
return ivec2(r, c);
}
`}function $c(r){return`offset${r}`}function JQ(r){let t=r.name,e="get"+t.charAt(0).toUpperCase()+t.slice(1),n=qe();return`
vec4 ${e}() {
return ${n.texture2D}(${t}, halfCR);
}
`}function QQ(r,t){let e=r.name,n="get"+e.charAt(0).toUpperCase()+e.slice(1);if(r.shapeInfo.isUniform)return`float ${n}() {return ${e};}`;let[o,s]=r.shapeInfo.texShape;if(o===1&&s===1)return`
float ${n}() {
return sampleTexture(${e}, halfCR);
}
`;let i=$c(e);if(t)return`
float ${n}() {
vec2 uv = uvFromFlat(${e}TexShape[0], ${e}TexShape[1], ${i});
return sampleTexture(${e}, uv);
}
`;let[a,u]=r.shapeInfo.texShape;return`
float ${n}() {
vec2 uv = uvFromFlat(${a}, ${u}, ${i});
return sampleTexture(${e}, uv);
}
`}function ttt(r,t){let e=r.name,n="get"+e.charAt(0).toUpperCase()+e.slice(1),o=r.shapeInfo.texShape,s=qe();if(t)return`
vec4 ${n}(int index) {
ivec2 packedTexShape = ivec2(ceil(float(${e}TexShape[0]) / 2.0), ceil(float(${e}TexShape[1]) / 2.0));
vec2 uv = packedUVfrom1D(
packedTexShape[0], packedTexShape[1], index);
return ${s.texture2D}(${e}, uv);
}
`;let i=[Math.ceil(o[0]/2),Math.ceil(o[1]/2)];return`
vec4 ${n}(int index) {
vec2 uv = packedUVfrom1D(
${i[0]}, ${i[1]}, index);
return ${s.texture2D}(${e}, uv);
}
`}function ett(r,t){let e=r.name,n="get"+e.charAt(0).toUpperCase()+e.slice(1);if(r.shapeInfo.isUniform)return`
float ${n}(int index) {
${Xm(r)}
}
`;let o=r.shapeInfo.texShape,s=o[0],i=o[1];if(i===1&&s===1)return`
float ${n}(int index) {
return sampleTexture(${e}, halfCR);
}
`;let a=$c(e);return i===1?t?`
float ${n}(int index) {
vec2 uv = vec2(0.5, (float(index + ${a}) + 0.5) / float(${e}TexShape[0]));
return sampleTexture(${e}, uv);
}
`:`
float ${n}(int index) {
vec2 uv = vec2(0.5, (float(index + ${a}) + 0.5) / ${s}.0);
return sampleTexture(${e}, uv);
}
`:s===1?t?`
float ${n}(int index) {
vec2 uv = vec2((float(index + ${a}) + 0.5) / float(${e}TexShape[1]), 0.5);
return sampleTexture(${e}, uv);
}
`:`
float ${n}(int index) {
vec2 uv = vec2((float(index + ${a}) + 0.5) / ${i}.0, 0.5);
return sampleTexture(${e}, uv);
}
`:t?`
float ${n}(int index) {
vec2 uv = uvFromFlat(${e}TexShape[0], ${e}TexShape[1], index + ${a});
return sampleTexture(${e}, uv);
}
`:`
float ${n}(int index) {
vec2 uv = uvFromFlat(${s}, ${i}, index + ${a});
return sampleTexture(${e}, uv);
}
`}function rtt(r,t){let e=r.shapeInfo.logicalShape,n=r.name,o="get"+n.charAt(0).toUpperCase()+n.slice(1),s=r.shapeInfo.texShape,i=s[0],a=s[1],u=qe();if(s!=null&&x.arraysEqual(e,s))return t?`
vec4 ${o}(int row, int col) {
vec2 uv = (vec2(col, row) + halfCR) / vec2(${n}TexShape[1], ${n}TexShape[0]);
return ${u.texture2D}(${n}, uv);
}
`:`
vec4 ${o}(int row, int col) {
vec2 uv = (vec2(col, row) + halfCR) / vec2(${a}.0, ${i}.0);
return ${u.texture2D}(${n}, uv);
}
`;if(t)return`
vec4 ${o}(int row, int col) {
ivec2 packedTexShape = ivec2(ceil(float(${n}TexShape[0]) / 2.0), ceil(float(${n}TexShape[1]) / 2.0));
int valuesPerRow = int(ceil(float(${n}Shape[1]) / 2.0));
vec2 uv = packedUVfrom2D(valuesPerRow, packedTexShape[0], packedTexShape[1], row, col);
return ${u.texture2D}(${n}, uv);
}
`;let l=[Math.ceil(s[0]/2),Math.ceil(s[1]/2)],c=Math.ceil(e[1]/2);return`
vec4 ${o}(int row, int col) {
vec2 uv = packedUVfrom2D(${c}, ${l[0]}, ${l[1]}, row, col);
return ${u.texture2D}(${n}, uv);
}
`}function ntt(r,t){let e=r.shapeInfo.logicalShape,n=r.name,o="get"+n.charAt(0).toUpperCase()+n.slice(1),s=r.shapeInfo.texShape;if(s!=null&&x.arraysEqual(e,s)){if(t)return`
float ${o}(int row, int col) {
vec2 uv = (vec2(col, row) + halfCR) / vec2(${n}TexShape[1], ${n}TexShape[0]);
return sampleTexture(${n}, uv);
}
`;let m=s[0],f=s[1];return`
float ${o}(int row, int col) {
vec2 uv = (vec2(col, row) + halfCR) / vec2(${f}.0, ${m}.0);
return sampleTexture(${n}, uv);
}
`}let{newShape:i,keptDims:a}=x.squeezeShape(e),u=i;if(u.length<e.length){let m=Ym(r,u),f=["row","col"];return`
${jm(m,t)}
float ${o}(int row, int col) {
return ${o}(${Zm(f,a)});
}
`}if(r.shapeInfo.isUniform)return`
float ${o}(int row, int col) {
int index = round(dot(vec2(row, col), vec2(${e[1]}, 1)));
${Xm(r)}
}
`;let l=s[0],c=s[1],p=$c(n);return c===1?t?`
float ${o}(int row, int col) {
float index = dot(vec3(row, col, ${p}), vec3(${n}Shape[1], 1, 1));
vec2 uv = vec2(0.5, (index + 0.5) / float(${n}TexShape[0]));
return sampleTexture(${n}, uv);
}
`:`
float ${o}(int row, int col) {
float index = dot(vec3(row, col, ${p}), vec3(${e[1]}, 1, 1));
vec2 uv = vec2(0.5, (index + 0.5) / ${l}.0);
return sampleTexture(${n}, uv);
}
`:l===1?t?`
float ${o}(int row, int col) {
float index = dot(vec3(row, col, ${p}), vec3(${n}Shape[1], 1, 1));
vec2 uv = vec2((index + 0.5) / float(${n}TexShape[1]), 0.5);
return sampleTexture(${n}, uv);
}
`:`
float ${o}(int row, int col) {
float index = dot(vec3(row, col, ${p}), vec3(${e[1]}, 1, 1));
vec2 uv = vec2((index + 0.5) / ${c}.0, 0.5);
return sampleTexture(${n}, uv);
}
`:t?`
float ${o}(int row, int col) {
// Explicitly use integer operations as dot() only works on floats.
int index = row * ${n}Shape[1] + col + ${p};
vec2 uv = uvFromFlat(${n}TexShape[0], ${n}TexShape[1], index);
return sampleTexture(${n}, uv);
}
`:`
float ${o}(int row, int col) {
// Explicitly use integer operations as dot() only works on floats.
int index = row * ${e[1]} + col + ${p};
vec2 uv = uvFromFlat(${l}, ${c}, index);
return sampleTexture(${n}, uv);
}
`}function ott(r,t){let e=r.shapeInfo.logicalShape,n=r.name,o="get"+n.charAt(0).toUpperCase()+n.slice(1),s=r.shapeInfo.texShape,i=[Math.ceil(s[0]/2),Math.ceil(s[1]/2)];if(e[0]===1){let m=e.slice(1),f=[1,2],d=Ym(r,m),h=["b","row","col"];return`
${BP(d,t)}
vec4 ${o}(int b, int row, int col) {
return ${o}(${Zm(h,f)});
}
`}let a=qe();if(t)return`
vec4 ${o}(int b, int row, int col) {
ivec2 packedTexShape = ivec2(ceil(float(${n}TexShape[0]) / 2.0), ceil(float(${n}TexShape[1]) / 2.0));
int valuesPerRow = int(ceil(float(${n}Shape[2]) / 2.0));
int texelsInBatch = valuesPerRow * int(ceil(float(${n}Shape[1]) / 2.0));
vec2 uv = packedUVfrom3D(
packedTexShape[0], packedTexShape[1], texelsInBatch, valuesPerRow, b, row, col);
return ${a.texture2D}(${n}, uv);
}
`;let u=i[0],l=i[1],c=Math.ceil(e[2]/2),p=c*Math.ceil(e[1]/2);return`
vec4 ${o}(int b, int row, int col) {
vec2 uv = packedUVfrom3D(
${u}, ${l}, ${p}, ${c}, b, row, col);
return ${a.texture2D}(${n}, uv);
}
`}function stt(r,t){let e=r.shapeInfo.logicalShape,n=r.name,o="get"+n.charAt(0).toUpperCase()+n.slice(1),s=e[1]*e[2],i=e[2],{newShape:a,keptDims:u}=x.squeezeShape(e),l=a;if(l.length<e.length){let h=Ym(r,l),g=["row","col","depth"];return`
${jm(h,t)}
float ${o}(int row, int col, int depth) {
return ${o}(${Zm(g,u)});
}
`}if(r.shapeInfo.isUniform)return`
float ${o}(int row, int col, int depth) {
int index = round(dot(vec3(row, col, depth),
vec3(${s}, ${i}, 1)));
${Xm(r)}
}
`;let c=r.shapeInfo.texShape,p=c[0],m=c[1],f=r.shapeInfo.flatOffset;if(m===s&&f==null)return t?`
float ${o}(int row, int col, int depth) {
int stride1 = ${n}Shape[2];
float texR = float(row);
float texC = dot(vec2(col, depth), vec2(stride1, 1));
vec2 uv = (vec2(texC, texR) + halfCR) /
vec2(${n}TexShape[1], ${n}TexShape[0]);
return sampleTexture(${n}, uv);
}
`:`
float ${o}(int row, int col, int depth) {
float texR = float(row);
float texC = dot(vec2(col, depth), vec2(${i}, 1));
vec2 uv = (vec2(texC, texR) + halfCR) /
vec2(${m}.0, ${p}.0);
return sampleTexture(${n}, uv);
}
`;if(m===i&&f==null)return t?`
float ${o}(int row, int col, int depth) {
float texR = dot(vec2(row, col), vec2(${n}Shape[1], 1));
float texC = float(depth);
vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${n}TexShape[1], ${n}TexShape[0]);
return sampleTexture(${n}, uv);
}
`:`
float ${o}(int row, int col, int depth) {
float texR = dot(vec2(row, col), vec2(${e[1]}, 1));
float texC = float(depth);
vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${m}.0, ${p}.0);
return sampleTexture(${n}, uv);
}
`;let d=$c(n);return t?`
float ${o}(int row, int col, int depth) {
// Explicitly use integer operations as dot() only works on floats.
int stride0 = ${n}Shape[1] * ${n}Shape[2];
int stride1 = ${n}Shape[2];
int index = row * ${s} + col * ${i} + depth + ${d};
vec2 uv = uvFromFlat(${n}TexShape[0], ${n}TexShape[1], index);
return sampleTexture(${n}, uv);
}
`:`
float ${o}(int row, int col, int depth) {
// Explicitly use integer operations as dot() only works on floats.
int index = row * ${s} + col * ${i} + depth + ${d};
vec2 uv = uvFromFlat(${p}, ${m}, index);
return sampleTexture(${n}, uv);
}
`}function itt(r,t){let e=r.name,n="get"+e.charAt(0).toUpperCase()+e.slice(1),o=qe();if(t)return`
vec4 ${n}(int b2, int b, int row, int col) {
int valuesPerRow = int(ceil(float(${e}Shape[3]) / 2.0));
int texelsInBatch = valuesPerRow * int(ceil(float(${e}Shape[2]) / 2.0));
int index = b * texelsInBatch + (row / 2) * valuesPerRow + (col / 2);
texelsInBatch *= ${e}Shape[1];
index = b2 * texelsInBatch + index;
ivec2 packedTexShape = ivec2(ceil(float(${e}TexShape[0]) / 2.0), ceil(float(${e}TexShape[1]) / 2.0));
int texR = index / packedTexShape[1];
int texC = index - texR * packedTexShape[1];
vec2 uv = (vec2(texC, texR) + halfCR) / vec2(packedTexShape[1], packedTexShape[0]); return ${o.texture2D}(${e}, uv);
}
`;let s=r.shapeInfo.logicalShape,i=s.length,a=r.shapeInfo.texShape,u=[Math.ceil(a[0]/2),Math.ceil(a[1]/2)],l=u[0],c=u[1],p=Math.ceil(s[i-1]/2),m=p*Math.ceil(s[i-2]/2),f="int b, int row, int col",d=`b * ${m} + (row / 2) * ${p} + (col / 2)`;for(let h=2;h<i-1;h++)f=`int b${h}, `+f,m*=s[i-h-1],d=`b${h} * ${m} + `+d;return`
vec4 ${n}(${f}) {
int index = ${d};
int texR = index / ${c};
int texC = index - texR * ${c};
vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${c}, ${l});
return ${o.texture2D}(${e}, uv);
}
`}function att(r,t){let e=r.shapeInfo.logicalShape,n=r.name,o="get"+n.charAt(0).toUpperCase()+n.slice(1),s=e[3],i=e[2]*s,a=e[1]*i,{newShape:u,keptDims:l}=x.squeezeShape(e);if(u.length<e.length){let b=Ym(r,u),w=["row","col","depth","depth2"];return`
${jm(b,t)}
float ${o}(int row, int col, int depth, int depth2) {
return ${o}(${Zm(w,l)});
}
`}if(r.shapeInfo.isUniform)return`
float ${o}(int row, int col, int depth, int depth2) {
int index = round(dot(vec4(row, col, depth, depth2),
vec4(${a}, ${i}, ${s}, 1)));
${Xm(r)}
}
`;let c=r.shapeInfo.flatOffset,p=r.shapeInfo.texShape,m=p[0],f=p[1],d=`int stride2 = ${n}Shape[3];`,h=`int stride1 = ${n}Shape[2] * stride2;`,g=`int stride0 = ${n}Shape[1] * stride1;`;if(f===a&&c==null)return t?`
float ${o}(int row, int col, int depth, int depth2) {
${d}
${h}
float texR = float(row);
float texC =
dot(vec3(col, depth, depth2),
vec3(stride1, stride2, 1));
vec2 uv = (vec2(texC, texR) + halfCR) /
vec2(${n}TexShape[1], ${n}TexShape[0]);
return sampleTexture(${n}, uv);
}
`:`
float ${o}(int row, int col, int depth, int depth2) {
float texR = float(row);
float texC =
dot(vec3(col, depth, depth2),
vec3(${i}, ${s}, 1));
vec2 uv = (vec2(texC, texR) + halfCR) /
vec2(${f}.0, ${m}.0);
return sampleTexture(${n}, uv);
}
`;if(f===s&&c==null)return t?`
float ${o}(int row, int col, int depth, int depth2) {
float texR = dot(vec3(row, col, depth),
vec3(${n}Shape[1] * ${n}Shape[2], ${n}Shape[2], 1));
float texC = float(depth2);
vec2 uv = (vec2(texC, texR) + halfCR) /
vec2(${n}TexShape[1], ${n}TexShape[0]);
return sampleTexture(${n}, uv);
}
`:`
float ${o}(int row, int col, int depth, int depth2) {
float texR = dot(vec3(row, col, depth),
vec3(${e[1]*e[2]}, ${e[2]}, 1));
float texC = float(depth2);
vec2 uv = (vec2(texC, texR) + halfCR) /
vec2(${f}.0, ${m}.0);
return sampleTexture(${n}, uv);
}
`;let y=$c(n);return t?`
float ${o}(int row, int col, int depth, int depth2) {
// Explicitly use integer operations as dot() only works on floats.
${d}
${h}
${g}
int index = row * stride0 + col * stride1 +
depth * stride2 + depth2;
vec2 uv = uvFromFlat(${n}TexShape[0], ${n}TexShape[1], index + ${y});
return sampleTexture(${n}, uv);
}
`:`
float ${o}(int row, int col, int depth, int depth2) {
// Explicitly use integer operations as dot() only works on floats.
int index = row * ${a} + col * ${i} +
depth * ${s} + depth2;
vec2 uv = uvFromFlat(${m}, ${f}, index + ${y});
return sampleTexture(${n}, uv);
}
`}function ltt(r){let t=r.shapeInfo.logicalShape,e=r.name,n="get"+e.charAt(0).toUpperCase()+e.slice(1),o=t[4],s=t[3]*o,i=t[2]*s,a=t[1]*i,{newShape:u,keptDims:l}=x.squeezeShape(t);if(u.length<t.length){let h=Ym(r,u),g=["row","col","depth","depth2","depth3"];return`
${jm(h)}
float ${n}(int row, int col, int depth, int depth2, int depth3) {
return ${n}(${Zm(g,l)});
}
`}if(r.shapeInfo.isUniform)return`
float ${n}(int row, int col, int depth, int depth2, int depth3) {
float index = dot(
vec4(row, col, depth, depth2),
vec4(${a}, ${i}, ${s}, ${o})) +
depth3;
${Xm(r)}
}
`;let c=r.shapeInfo.flatOffset,p=r.shapeInfo.texShape,m=p[0],f=p[1];if(f===a&&c==null)return`
float ${n}(int row, int col, int depth, int depth2, int depth3) {
int texR = row;
float texC = dot(vec4(col, depth, depth2, depth3),
vec4(${i}, ${s}, ${o}, 1));
vec2 uv = (vec2(texC, texR) + halfCR) /
vec2(${f}.0, ${m}.0);
return sampleTexture(${e}, uv);
}
`;if(f===o&&c==null)return`
float ${n}(int row, int col, int depth, int depth2, int depth3) {
float texR = dot(
vec4(row, col, depth, depth2),
vec4(${t[1]*t[2]*t[3]},
${t[2]*t[3]}, ${t[3]}, 1));
int texC = depth3;
vec2 uv = (vec2(texC, texR) + halfCR) /
vec2(${f}.0, ${m}.0);
return sampleTexture(${e}, uv);
}
`;let d=$c(e);return`
float ${n}(int row, int col, int depth, int depth2, int depth3) {
// Explicitly use integer operations as dot() only works on floats.
int index = row * ${a} + col * ${i} + depth * ${s} +
depth2 * ${o} + depth3 + ${d};
vec2 uv = uvFromFlat(${m}, ${f}, index);
return sampleTexture(${e}, uv);
}
`}function utt(r){let t=r.shapeInfo.logicalShape,e=r.name,n="get"+e.charAt(0).toUpperCase()+e.slice(1),{newShape:o,keptDims:s}=x.squeezeShape(t);if(o.length<t.length){let g=Ym(r,o),y=["row","col","depth","depth2","depth3","depth4"];return`
${jm(g)}
float ${n}(int row, int col, int depth,
int depth2, int depth3, int depth4) {
return ${n}(${Zm(y,s)});
}
`}let i=t[5],a=t[4]*i,u=t[3]*a,l=t[2]*u,c=t[1]*l;if(r.shapeInfo.isUniform)return`
float ${n}(int row, int col, int depth,
int depth2, int depth3, int depth4) {
int index = round(dot(
vec4(row, col, depth, depth2),
vec4(${c}, ${l}, ${u}, ${a})) +
dot(
vec2(depth3, depth4),
vec2(${i}, 1)));
${Xm(r)}
}
`;let p=r.shapeInfo.flatOffset,m=r.shapeInfo.texShape,f=m[0],d=m[1];if(d===c&&p==null)return`
float ${n}(int row, int col, int depth,
int depth2, int depth3, int depth4) {
int texR = row;
float texC = dot(vec4(col, depth, depth2, depth3),
vec4(${l}, ${u}, ${a}, ${i})) +
float(depth4);
vec2 uv = (vec2(texC, texR) + halfCR) /
vec2(${d}.0, ${f}.0);
return sampleTexture(${e}, uv);
}
`;if(d===i&&p==null)return`
float ${n}(int row, int col, int depth,
int depth2, int depth3, int depth4) {
float texR = dot(vec4(row, col, depth, depth2),
vec4(${t[1]*t[2]*t[3]*t[4]},
${t[2]*t[3]*t[4]},
${t[3]*t[4]},
${t[4]})) + float(depth3);
int texC = depth4;
vec2 uv = (vec2(texC, texR) + halfCR) /
vec2(${d}.0, ${f}.0);
return sampleTexture(${e}, uv);
}
`;let h=$c(e);return`
float ${n}(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 * ${c} + col * ${l} + depth * ${u} +
depth2 * ${a} + depth3 * ${i} + depth4 + ${h};
vec2 uv = uvFromFlat(${f}, ${d}, index);
return sampleTexture(${e}, uv);
}
`}function Xm(r){let t=r.name,e=x.sizeFromShape(r.shapeInfo.logicalShape);return e<2?`return ${t};`:`
for (int i = 0; i < ${e}; i++) {
if (i == index) {
return ${t}[i];
}
}
`}function ctt(r,t){let e=r.name,n=e.charAt(0).toUpperCase()+e.slice(1),o="get"+n+"AtOutCoords",s=r.shapeInfo.logicalShape.length,i=t.logicalShape.length,a=LP(r.shapeInfo.logicalShape,t.logicalShape),u=zt(i),l=i-s,c,p=["x","y","z","w","u","v"];s===0?c="":i<2&&a.length>=1?c="coords = 0;":c=a.map(b=>`coords.${p[b+l]} = 0;`).join(`
`);let m="";i<2&&s>0?m="coords":m=r.shapeInfo.logicalShape.map((b,w)=>`coords.${p[w+l]}`).join(", ");let f="return outputValue;",h=x.sizeFromShape(r.shapeInfo.logicalShape)===1,y=x.sizeFromShape(t.logicalShape)===1;if(s===1&&!h&&!y)f=`
return vec4(outputValue.xy, outputValue.xy);
`;else if(h&&!y)i===1?f=`
return vec4(outputValue.x, outputValue.x, 0., 0.);
`:f=`
return vec4(outputValue.x);
`;else if(a.length){let b=s-2,w=s-1;a.indexOf(b)>-1&&a.indexOf(w)>-1?f="return vec4(outputValue.x);":a.indexOf(b)>-1?f="return vec4(outputValue.x, outputValue.y, outputValue.x, outputValue.y);":a.indexOf(w)>-1&&(f="return vec4(outputValue.xx, outputValue.zz);")}return`
vec4 ${o}() {
${u} coords = getOutputCoords();
${c}
vec4 outputValue = get${n}(${m});
${f}
}
`}function ptt(r,t){let e=r.name,n=e.charAt(0).toUpperCase()+e.slice(1),o="get"+n+"AtOutCoords",s=t.texShape,i=r.shapeInfo.texShape,a=r.shapeInfo.logicalShape.length,u=t.logicalShape.length;if(!r.shapeInfo.isUniform&&a===u&&r.shapeInfo.flatOffset==null&&x.arraysEqual(i,s))return`
float ${o}() {
return sampleTexture(${e}, resultUV);
}
`;let l=zt(u),c=LP(r.shapeInfo.logicalShape,t.logicalShape),p=u-a,m,f=["x","y","z","w","u","v"];a===0?m="":u<2&&c.length>=1?m="coords = 0;":m=c.map(h=>`coords.${f[h+p]} = 0;`).join(`
`);let d="";return u<2&&a>0?d="coords":d=r.shapeInfo.logicalShape.map((h,g)=>`coords.${f[g+p]}`).join(", "),`
float ${o}() {
${l} coords = getOutputCoords();
${m}
return get${n}(${d});
}
`}function zt(r){if(r<=1)return"int";if(r===2)return"ivec2";if(r===3)return"ivec3";if(r===4)return"ivec4";if(r===5)return"ivec5";if(r===6)return"ivec6";throw Error(`GPU for rank ${r} is not yet supported`)}function hb(r,t,e){let{newShape:n,keptDims:o}=x.squeezeShape(t),s=t.length,i=r&&s===3&&t[0]===1,a=i?t.slice(1):n,u=!r&&s>1&&!x.arraysEqual(t,e)&&n.length<s||i;return{useSqueezeShape:u,uniformShape:u?a:t,keptDims:o}}function Ym(r,t){let e=JSON.parse(JSON.stringify(r));return e.shapeInfo.logicalShape=t,e}function Zm(r,t){return t.map(e=>r[e]).join(", ")}function WP(r,t,e,n){let o=e.map((c,p)=>{let m={logicalShape:c.shape,texShape:c.isUniform?null:c.texData.texShape,isUniform:c.isUniform,isPacked:c.isUniform?!1:c.texData.isPacked,flatOffset:null};return c.texData!=null&&c.texData.slice!=null&&c.texData.slice.flatOffset>0&&(m.flatOffset=c.texData.slice.flatOffset),{name:t.variableNames[p],shapeInfo:m}}),s=o.map(c=>c.shapeInfo),i={logicalShape:n.shape,texShape:n.texData.texShape,isUniform:!1,isPacked:n.texData.isPacked,flatOffset:null},a=zP(o,i,t),u=US(r.gl,a),l=r.createProgram(u);return V().get("ENGINE_COMPILE_ONLY")?{program:t,fragmentShader:u,source:a,webGLProgram:l,inShapeInfos:s,outShapeInfo:i,uniformLocations:null,customUniformLocations:null,infLoc:null,nanLoc:null,inShapesLocations:null,inTexShapesLocations:null,outShapeLocation:null,outShapeStridesLocation:null,outTexShapeLocation:null}:Object.assign({program:t,fragmentShader:u,source:a,webGLProgram:l,inShapeInfos:s,outShapeInfo:i},lk(r,t,l))}function lk(r,t,e){let n={},o={},s={},i=[],a,u,l,c=null,p=null;p=r.getUniformLocation(e,"NAN",!1),V().getNumber("WEBGL_VERSION")===1&&(c=r.getUniformLocation(e,"INFINITY",!1));let m=!1;for(let f=0;f<t.variableNames.length;f++){let d=t.variableNames[f];n[d]=r.getUniformLocation(e,d,m),n[`offset${d}`]=r.getUniformLocation(e,`offset${d}`,m),t.enableShapeUniforms&&(o[`${d}Shape`]=r.getUniformLocation(e,`${d}Shape`,m),s[`${d}TexShape`]=r.getUniformLocation(e,`${d}TexShape`,m))}return t.enableShapeUniforms&&(a=r.getUniformLocation(e,"outShape",m),l=r.getUniformLocation(e,"outShapeStrides",m),u=r.getUniformLocation(e,"outTexShape",m)),t.customUniforms&&t.customUniforms.forEach((f,d)=>{i[d]=r.getUniformLocation(e,f.name,m)}),{uniformLocations:n,customUniformLocations:i,infLoc:c,nanLoc:p,inShapesLocations:o,inTexShapesLocations:s,outShapeLocation:a,outShapeStridesLocation:l,outTexShapeLocation:u}}function GP(r,t){if(r.length!==t.length)throw Error(`Binary was compiled with ${r.length} inputs, but was executed with ${t.length} inputs`);r.forEach((e,n)=>{let o=e.logicalShape,s=t[n],i=s.shape;if(!x.arraysEqual(o,i))throw Error(`Binary was compiled with different shapes than the current args. Shapes ${o} and ${i} must match`);if(e.isUniform&&s.isUniform)return;let a=e.texShape,u=s.isUniform?null:s.texData.texShape;if(!x.arraysEqual(a,u))throw Error(`Binary was compiled with different texture shapes than the current args. Shape ${a} and ${u} must match`)})}function UP(r,t,e,n,o){t.program.enableShapeUniforms||(GP(t.inShapeInfos,e),GP([t.outShapeInfo],[n]));let s=n.texData.texture,i=n.texData.texShape;n.texData.isPacked?r.setOutputPackedMatrixTexture(s.texture,i[0],i[1]):r.setOutputMatrixTexture(s.texture,i[0],i[1]),r.setProgram(t.webGLProgram),V().getNumber("WEBGL_VERSION")===1&&t.infLoc!==null&&r.gl.uniform1f(t.infLoc,1/0),t.nanLoc!==null&&r.gl.uniform1f(t.nanLoc,NaN),e.forEach((u,l)=>{let c=t.program.variableNames[l],p=t.uniformLocations[c],m=t.uniformLocations[`offset${c}`],f=t.inShapesLocations[`${c}Shape`],d=t.inTexShapesLocations[`${c}TexShape`];if(f){let{uniformShape:h}=hb(t.program.packedInputs,u.shape,u.texData.texShape);switch(h.length){case 1:r.gl.uniform1iv(f,new Int32Array(h));break;case 2:r.gl.uniform2iv(f,new Int32Array(h));break;case 3:r.gl.uniform3iv(f,new Int32Array(h));break;case 4:r.gl.uniform4iv(f,new Int32Array(h));break;default:break}}if(d&&r.gl.uniform2i(d,u.texData.texShape[0],u.texData.texShape[1]),p!=null){if(u.isUniform){if(x.sizeFromShape(u.shape)<2)r.gl.uniform1f(p,u.uniformValues[0]);else{let h=u.uniformValues;h instanceof Float32Array||(h=new Float32Array(h)),r.gl.uniform1fv(p,h)}return}u.texData.slice!=null&&m!=null&&r.gl.uniform1i(m,u.texData.slice.flatOffset),r.setInputMatrixTexture(u.texData.texture.texture,p,l)}});let a=t.outShapeLocation;if(a)switch(n.shape.length){case 1:r.gl.uniform1iv(a,new Int32Array(n.shape));break;case 2:r.gl.uniform2iv(a,new Int32Array(n.shape));break;case 3:r.gl.uniform3iv(a,new Int32Array(n.shape));break;case 4:r.gl.uniform4iv(a,new Int32Array(n.shape));break;default:break}if(t.outShapeStridesLocation){let u=x.computeStrides(n.shape);switch(n.shape.length){case 2:r.gl.uniform1iv(t.outShapeStridesLocation,new Int32Array(u));break;case 3:r.gl.uniform2iv(t.outShapeStridesLocation,new Int32Array(u));break;case 4:r.gl.uniform3iv(t.outShapeStridesLocation,new Int32Array(u));break;default:break}}t.outTexShapeLocation&&r.gl.uniform2i(t.outTexShapeLocation,n.texData.texShape[0],n.texData.texShape[1]),t.program.customUniforms&&o&&t.program.customUniforms.forEach((u,l)=>{let c=t.customUniformLocations[l],p=o[l];if(u.type==="float")r.gl.uniform1fv(c,p);else if(u.type==="vec2")r.gl.uniform2fv(c,p);else if(u.type==="vec3")r.gl.uniform3fv(c,p);else if(u.type==="vec4")r.gl.uniform4fv(c,p);else if(u.type==="int")r.gl.uniform1iv(c,p);else if(u.type==="ivec2")r.gl.uniform2iv(c,p);else if(u.type==="ivec3")r.gl.uniform3iv(c,p);else if(u.type==="ivec4")r.gl.uniform4iv(c,p);else throw Error(`uniform type ${u.type} is not supported yet.`)}),r.executeProgram()}function HP(r,t,e){let n="";t.concat(e).forEach(i=>{let a=i.texData!=null&&i.texData.slice!=null&&i.texData.slice.flatOffset>0;if(r.enableShapeUniforms&&!i.isUniform){let u=i.texData.texShape,{useSqueezeShape:l,uniformShape:c,keptDims:p}=hb(r.packedInputs,i.shape,u),m="",f="",d="";if(c.length===1&&r.packedInputs){let k=[Math.ceil(u[0]/2),Math.ceil(u[1]/2)];m=`${k[0]>1}_${k[1]>1}`}else if(c.length===2&&!r.packedInputs)f=`${c[0]>1}_${c[1]>1}`;else if(c.length>2&&!r.packedInputs){let k=x.computeStrides(c);d=`${k[0]===u[1]}_${k[k.length-1]===u[1]}`}let h=i.shape.length,g=c.length===2&&x.arraysEqual(i.shape,u),y=x.sizeFromShape(i.shape)===1,b=S.getBroadcastDims(i.shape,e.shape),w=!r.packedInputs&&h===e.shape.length&&x.arraysEqual(u,e.texData.texShape),v=r.packedInputs||c.length>2?"":`${u[0]>1}_${u[1]>1}`;n+=`${h}_${w}_${l?p:""}_${c.length}_${y}_${b}_${g}_${m}_${f}_${d}_${v}_${a}`}else{let u=i.isUniform?"uniform":i.texData.texShape;n+=`${i.shape}_${u}_${a}`}});let o=r.userCode,s=r.constructor.name;return s+="_"+n+"_"+o+`${V().getNumber("WEBGL_VERSION")}`,s}function $e(r){return V().getBool("WEBGL_USE_SHAPES_UNIFORMS")&&r<=4}var uk=class{constructor(t){this.variableNames=["A"],this.packedInputs=!1,this.packedOutput=!0,this.outPackingScheme=ou.DENSE,this.customUniforms=[{name:"texShape",type:"ivec2"}];let e=qe();this.outputShape=t,this.enableShapeUniforms=$e(this.outputShape.length),this.userCode=`
ivec3 outCoordsFromFlatIndex(int index) {
${this.enableShapeUniforms?Dc(["r","c","d"],t):ui(["r","c","d"],t)}
return ivec3(r, c, d);
}
void main() {
ivec2 resTexRC = ivec2(resultUV.yx * vec2(texShape[0], texShape[1]));
int index = 4 * (resTexRC.x * texShape[1] + resTexRC.y);
vec4 result = vec4(0.);
for (int i=0; i<4; i++) {
int flatIndex = index + i;
ivec3 rc = outCoordsFromFlatIndex(flatIndex);
result[i] = getA(rc.x, rc.y, rc.z);
}
${e.output} = result;
}
`}};var ck=class{constructor(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outPackingScheme=ou.DENSE,this.customUniforms=[{name:"texShape",type:"ivec2"}];let e=qe();this.outputShape=t,this.enableShapeUniforms=$e(this.outputShape.length),this.userCode=`
ivec3 outCoordsFromFlatIndex(int index) {
${this.enableShapeUniforms?Dc(["r","c","d"],t):ui(["r","c","d"],t)}
return ivec3(r, c, d);
}
void main() {
ivec2 resTexRC = ivec2(resultUV.yx * vec2(texShape[0], texShape[1]));
int index = 4 * (resTexRC.x * texShape[1] + resTexRC.y);
vec4 result = vec4(0.);
for (int i=0; i<4; i++) {
int flatIndex = index + i;
ivec3 rc = outCoordsFromFlatIndex(flatIndex);
result[i] = getChannel(getA(rc.x, rc.y, rc.z), vec2(rc.y, rc.z));
}
${e.output} = result;
}
`}};var pk=class{constructor(t){this.variableNames=["A"],this.outTexUsage=Ur.DOWNLOAD;let e=qe();this.outputShape=t,this.userCode=`
${db}
void main() {
float x = getAAtOutCoords();
${e.output} = encode_float(x);
}
`}};var mk=class{constructor(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outTexUsage=Ur.DOWNLOAD;let e=qe();this.outputShape=t,this.userCode=`
${db}
void main() {
ivec3 coords = getOutputCoords();
float x = getChannel(getAAtOutCoords(), vec2(coords.y, coords.z));
${e.output} = encode_float(x);
}
`}};var fk=class{constructor(t,e=!1){this.variableNames=["A"],this.customUniforms=[{name:"texShape",type:"ivec2"}];let n=qe();this.outputShape=t,this.enableShapeUniforms=$e(this.outputShape.length);let o="result";e&&(o="floor(result * 255. + 0.5)"),this.userCode=`
${this.enableShapeUniforms?qm():Hm(t)}
void main() {
ivec3 coords = getOutputCoords();
int flatIndex = getFlatIndex(coords);
int offset = imod(flatIndex, 4);
flatIndex = idiv(flatIndex, 4, 1.);
int r = flatIndex / texShape[1];
int c = imod(flatIndex, texShape[1]);
vec2 uv = (vec2(c, r) + halfCR) / vec2(texShape[1], texShape[0]);
vec4 values = ${n.texture2D}(A, uv);
float result;
if(offset == 0) {
result = values[0];
} else if(offset == 1) {
result = values[1];
} else if(offset == 2) {
result = values[2];
} else {
result = values[3];
}
${n.output} = vec4(${o}, 0., 0., 0.);
}
`}};var dk=class{constructor(t,e=!1){this.variableNames=["A"],this.packedInputs=!1,this.packedOutput=!0,this.customUniforms=[{name:"texShape",type:"ivec2"}];let n=qe();this.outputShape=t,this.enableShapeUniforms=$e(this.outputShape.length);let o="",s="result";e&&(s="floor(result * 255. + 0.5)");for(let i=0;i<=1;i++)for(let a=0;a<=1;a++){let u=i*2+a;o+=`
localCoords = coords;
if(localCoords[2] + ${a} < ${this.enableShapeUniforms?"outShape[2]":`${t[2]}`}) {
localCoords[2] += ${a};
if (localCoords[1] + ${i} < ${this.enableShapeUniforms?"outShape[1]":`${t[1]}`}) {
localCoords[1] += ${i};
flatIndex = getFlatIndex(localCoords);
offset = imod(flatIndex, 4);
flatIndex = idiv(flatIndex, 4, 1.);
int r = flatIndex / texShape[1];
int c = imod(flatIndex, texShape[1]);
vec2 uv = (vec2(c, r) + halfCR) / vec2(texShape[1], texShape[0]);
values = ${n.texture2D}(A, uv);
if (offset == 0) {
result[${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=`
${this.enableShapeUniforms?qm():Hm(t)}
void main() {
ivec3 coords = getOutputCoords();
vec4 result = vec4(0.);
int flatIndex, r, c, offset;
ivec3 localCoords;
vec2 uv;
vec4 values;
${o}
${n.output} = ${s};
}
`}};var Dk={};Yt(Dk,{bindVertexProgramAttributeStreams:()=>Ik,createBufferFromOutputTexture:()=>Nk,createFloat16MatrixTexture:()=>bk,createFloat16PackedMatrixTexture:()=>Ck,createFloat32MatrixTexture:()=>yk,createIndexBuffer:()=>xk,createPackedMatrixTexture:()=>vk,createUnsignedBytesMatrixTexture:()=>wk,createVertexBuffer:()=>gk,createVertexShader:()=>hk,downloadByteEncodedFloatMatrixFromOutputTexture:()=>_k,downloadFloat32MatrixFromBuffer:()=>Tk,downloadMatrixFromPackedOutputTexture:()=>Ak,downloadPackedMatrixFromBuffer:()=>Ek,getInternalFormatForFloat16MatrixTexture:()=>xb,getInternalFormatForFloat16PackedMatrixTexture:()=>wb,getInternalFormatForFloat32MatrixTexture:()=>gb,getInternalFormatForPackedMatrixTexture:()=>bb,getInternalFormatForUnsignedBytesMatrixTexture:()=>yb,uploadDenseMatrixToTexture:()=>Sk,uploadPixelDataToTexture:()=>kk});function hk(r){let t=qe(),e=`${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 WS(r,e)}function gk(r){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 KS(r,t)}function xk(r){let t=new Uint16Array([0,1,2,2,1,3]);return jS(r,t)}function lg(r,t,e,n,o,s){YS(t,e);let i=XS(r),a=r.TEXTURE_2D;return vt(r,()=>r.bindTexture(a,i)),vt(r,()=>r.texParameteri(a,r.TEXTURE_WRAP_S,r.CLAMP_TO_EDGE)),vt(r,()=>r.texParameteri(a,r.TEXTURE_WRAP_T,r.CLAMP_TO_EDGE)),vt(r,()=>r.texParameteri(a,r.TEXTURE_MIN_FILTER,r.NEAREST)),vt(r,()=>r.texParameteri(a,r.TEXTURE_MAG_FILTER,r.NEAREST)),V().getNumber("WEBGL_VERSION")===1?vt(r,()=>r.texImage2D(a,0,n,t,e,0,o,s,null)):vt(r,()=>r.texStorage2D(a,1,n,t,e)),vt(r,()=>r.bindTexture(r.TEXTURE_2D,null)),{texture:i,texShape:[e,t]}}function gb(r){return r.internalFormatFloat}function yk(r,t,e,n){let[o,s]=Ac(t,e);return lg(r,o,s,gb(n),n.textureFormatFloat,r.FLOAT)}function xb(r){return r.internalFormatHalfFloat}function bk(r,t,e,n){let[o,s]=Ac(t,e);return lg(r,o,s,xb(n),n.textureFormatFloat,n.textureTypeHalfFloat)}function yb(r){return r.downloadTextureFormat}function wk(r,t,e,n){let[o,s]=Ac(t,e);return lg(r,o,s,yb(n),r.RGBA,r.UNSIGNED_BYTE)}function bb(r){return r.internalFormatPackedFloat}function vk(r,t,e,n){let[o,s]=Hi(t,e);return lg(r,o,s,bb(n),r.RGBA,r.FLOAT)}function wb(r){return r.internalFormatPackedHalfFloat}function Ck(r,t,e,n){let[o,s]=Hi(t,e);return lg(r,o,s,wb(n),r.RGBA,n.textureTypeHalfFloat)}function Ik(r,t,e){return vt(r,()=>r.bindBuffer(r.ARRAY_BUFFER,e)),pb(r,t,"clipSpacePos",e,3,20,0)&&pb(r,t,"uv",e,2,20,12)}function Sk(r,t,e,n,o,s){vt(r,()=>r.bindTexture(r.TEXTURE_2D,t));let i,a,u;o instanceof Uint8Array?(i=new Uint8Array(e*n*4),a=r.UNSIGNED_BYTE,u=r.RGBA):(i=new Float32Array(e*n*4),a=r.FLOAT,u=s.internalFormatPackedFloat),i.set(o),V().getNumber("WEBGL_VERSION")===2?vt(r,()=>r.texSubImage2D(r.TEXTURE_2D,0,0,0,e,n,r.RGBA,a,i)):vt(r,()=>r.texImage2D(r.TEXTURE_2D,0,u,e,n,0,r.RGBA,a,i)),vt(r,()=>r.bindTexture(r.TEXTURE_2D,null))}function kk(r,t,e){vt(r,()=>r.bindTexture(r.TEXTURE_2D,t)),e.data instanceof Uint8Array?V().getNumber("WEBGL_VERSION")===2?vt(r,()=>r.texSubImage2D(r.TEXTURE_2D,0,0,0,e.width,e.height,r.RGBA,r.UNSIGNED_BYTE,e.data)):vt(r,()=>r.texImage2D(r.TEXTURE_2D,0,r.RGBA,e.width,e.height,0,r.RGBA,r.UNSIGNED_BYTE,e.data)):V().getNumber("WEBGL_VERSION")===2?vt(r,()=>r.texSubImage2D(r.TEXTURE_2D,0,0,0,r.RGBA,r.UNSIGNED_BYTE,e)):vt(r,()=>r.texImage2D(r.TEXTURE_2D,0,r.RGBA,r.RGBA,r.UNSIGNED_BYTE,e)),vt(r,()=>r.bindTexture(r.TEXTURE_2D,null))}function Nk(r,t,e,n){let o=r.createBuffer();vt(r,()=>r.bindBuffer(r.PIXEL_PACK_BUFFER,o));let a=4*4*t*e;return vt(r,()=>r.bufferData(r.PIXEL_PACK_BUFFER,a,r.STREAM_READ)),vt(r,()=>r.readPixels(0,0,e,t,r.RGBA,r.FLOAT,0)),vt(r,()=>r.bindBuffer(r.PIXEL_PACK_BUFFER,null)),o}function Tk(r,t,e){let n=r,o=new Float32Array(e);return n.bindBuffer(n.PIXEL_PACK_BUFFER,t),n.getBufferSubData(n.PIXEL_PACK_BUFFER,0,o),n.bindBuffer(n.PIXEL_PACK_BUFFER,null),o}function _k(r,t,e,n){let[o,s]=Ac(t,e),i=4,a=new Uint8Array(DP(t*e,i));return vt(r,()=>r.readPixels(0,0,o,s,n.downloadTextureFormat,r.UNSIGNED_BYTE,a)),new Float32Array(a.buffer)}function Ek(r,t,e,n,o,s,i,a){let u=r,l=new Float32Array($P(s,i));return u.bindBuffer(u.PIXEL_PACK_BUFFER,t),u.getBufferSubData(u.PIXEL_PACK_BUFFER,0,l),u.bindBuffer(u.PIXEL_PACK_BUFFER,null),l}function Ak(r,t,e){let n=new Float32Array(t*e*4);return vt(r,()=>r.readPixels(0,0,e,t,r.RGBA,r.FLOAT,n)),n}var Jm=class{constructor(t){this.outputTexture=null,this.program=null,this.disposed=!1,this.vertexAttrsAreBound=!1,this.itemsToPoll=[];let e=V().getNumber("WEBGL_VERSION");t!=null?(this.gl=t,zS(e,t)):this.gl=Vn(e);let n="WEBGL_color_buffer_float",o="EXT_color_buffer_half_float";if(this.parallelCompilationExtension=this.gl.getExtension("KHR_parallel_shader_compile"),V().getNumber("WEBGL_VERSION")===1){let s="OES_texture_float",i="OES_texture_half_float";if(this.textureFloatExtension=Gm(this.gl,s),Gn(this.gl,i))this.textureHalfFloatExtension=Gm(this.gl,i);else if(V().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),Gn(this.gl,o))this.colorBufferHalfFloatExtension=Gm(this.gl,o);else if(V().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",Gn(this.gl,n))this.colorBufferFloatExtension=this.gl.getExtension(n);else if(Gn(this.gl,o))this.colorBufferHalfFloatExtension=this.gl.getExtension(o);else throw new Error("GL context does not support color renderable floats");this.vertexBuffer=gk(this.gl),this.indexBuffer=xk(this.gl),this.framebuffer=ZS(this.gl),this.textureConfig=og(this.gl,this.textureHalfFloatExtension)}get debug(){return V().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 t=this.gl;vt(t,()=>t.finish()),vt(t,()=>t.bindFramebuffer(t.FRAMEBUFFER,null)),vt(t,()=>t.deleteFramebuffer(this.framebuffer)),vt(t,()=>t.bindBuffer(t.ARRAY_BUFFER,null)),vt(t,()=>t.bindBuffer(t.ELEMENT_ARRAY_BUFFER,null)),vt(t,()=>t.deleteBuffer(this.indexBuffer)),this.disposed=!0}createFloat32MatrixTexture(t,e){return this.throwIfDisposed(),yk(this.gl,t,e,this.textureConfig)}createFloat16MatrixTexture(t,e){return this.throwIfDisposed(),bk(this.gl,t,e,this.textureConfig)}createUnsignedBytesMatrixTexture(t,e){return this.throwIfDisposed(),wk(this.gl,t,e,this.textureConfig)}uploadPixelDataToTexture(t,e){this.throwIfDisposed(),kk(this.gl,t,e)}uploadDenseMatrixToTexture(t,e,n,o){this.throwIfDisposed(),Sk(this.gl,t,e,n,o,this.textureConfig)}createFloat16PackedMatrixTexture(t,e){return this.throwIfDisposed(),Ck(this.gl,t,e,this.textureConfig)}createPackedMatrixTexture(t,e){return this.throwIfDisposed(),vk(this.gl,t,e,this.textureConfig)}deleteMatrixTexture(t){this.throwIfDisposed(),this.outputTexture===t&&(mb(this.gl,this.framebuffer),this.outputTexture=null),vt(this.gl,()=>this.gl.deleteTexture(t))}downloadByteEncodedFloatMatrixFromOutputTexture(t,e,n){return this.downloadMatrixDriver(t,()=>_k(this.gl,e,n,this.textureConfig))}downloadPackedMatrixFromBuffer(t,e,n,o,s,i){return Ek(this.gl,t,e,n,o,s,i,this.textureConfig)}downloadFloat32MatrixFromBuffer(t,e){return Tk(this.gl,t,e)}createBufferFromTexture(t,e,n){this.bindTextureToFrameBuffer(t);let o=Nk(this.gl,e,n,this.textureConfig);return this.unbindTextureToFrameBuffer(),o}createAndWaitForFence(){let t=this.createFence(this.gl);return this.pollFence(t)}createFence(t){let e,n;if(V().getBool("WEBGL_FENCE_API_ENABLED")){let o=t,s=o.fenceSync(o.SYNC_GPU_COMMANDS_COMPLETE,0);t.flush(),n=()=>{let i=o.clientWaitSync(s,0,0);return i===o.ALREADY_SIGNALED||i===o.CONDITION_SATISFIED},e=s}else V().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")>0?(e=this.beginQuery(),this.endQuery(),n=()=>this.isQueryAvailable(e,V().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))):n=()=>!0;return{query:e,isFencePassed:n}}downloadMatrixFromPackedTexture(t,e,n){return this.downloadMatrixDriver(t,()=>Ak(this.gl,e,n))}createProgram(t){this.throwIfDisposed();let e=this.gl;this.vertexShader==null&&(this.vertexShader=hk(e));let n=HS(e);return vt(e,()=>e.attachShader(n,this.vertexShader)),vt(e,()=>e.attachShader(n,t)),qS(e,n),this.debug&&sg(e,n),this.vertexAttrsAreBound||(this.setProgram(n),this.vertexAttrsAreBound=Ik(e,this.program,this.vertexBuffer)),n}deleteProgram(t){this.throwIfDisposed(),t===this.program&&(this.program=null),t!=null&&vt(this.gl,()=>this.gl.deleteProgram(t))}setProgram(t){this.throwIfDisposed(),this.program=t,this.program!=null&&this.debug&&sg(this.gl,this.program),vt(this.gl,()=>this.gl.useProgram(t))}getUniformLocation(t,e,n=!0){return this.throwIfDisposed(),n?JS(this.gl,t,e):QS(this.gl,t,e)}getAttributeLocation(t,e){return this.throwIfDisposed(),vt(this.gl,()=>this.gl.getAttribLocation(t,e))}getUniformLocationNoThrow(t,e){return this.throwIfDisposed(),this.gl.getUniformLocation(t,e)}setInputMatrixTexture(t,e,n){this.throwIfDisposed(),this.throwIfNoProgram(),tk(this.gl,t,e,n)}setOutputMatrixTexture(t,e,n){this.setOutputMatrixTextureDriver(t,n,e)}setOutputPackedMatrixTexture(t,e,n){this.throwIfDisposed();let[o,s]=Hi(e,n);this.setOutputMatrixTextureDriver(t,o,s)}setOutputMatrixWriteRegion(t,e,n,o){this.setOutputMatrixWriteRegionDriver(n,t,o,e)}setOutputPackedMatrixWriteRegion(t,e,n,o){throw new Error("setOutputPackedMatrixWriteRegion not implemented.")}debugValidate(){this.program!=null&&sg(this.gl,this.program),Wm(this.gl)}executeProgram(){this.throwIfDisposed(),this.throwIfNoProgram();let t=this.gl;this.debug&&this.debugValidate(),vt(t,()=>t.drawElements(t.TRIANGLES,6,t.UNSIGNED_SHORT,0))}blockUntilAllProgramsCompleted(){this.throwIfDisposed(),vt(this.gl,()=>this.gl.finish())}getQueryTimerExtension(){return this.disjointQueryTimerExtension==null&&(this.disjointQueryTimerExtension=Gm(this.gl,V().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(V().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){let n=this.gl,o=this.getQueryTimerExtensionWebGL2(),s=n.createQuery();return n.beginQuery(o.TIME_ELAPSED_EXT,s),s}let t=this.getQueryTimerExtensionWebGL1(),e=t.createQueryEXT();return t.beginQueryEXT(t.TIME_ELAPSED_EXT,e),e}endQuery(){if(V().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION")===2){let e=this.gl,n=this.getQueryTimerExtensionWebGL2();e.endQuery(n.TIME_ELAPSED_EXT);return}let t=this.getQueryTimerExtensionWebGL1();t.endQueryEXT(t.TIME_ELAPSED_EXT)}async waitForQueryAndGetTime(t){return await x.repeatedTry(()=>this.disposed||this.isQueryAvailable(t,V().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))),this.getQueryTime(t,V().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_VERSION"))}getQueryTime(t,e){if(e===0)return null;if(e===2){let n=this.gl;return n.getQueryParameter(t,n.QUERY_RESULT)/1e6}else{let n=this.getQueryTimerExtensionWebGL1();return n.getQueryObjectEXT(t,n.QUERY_RESULT_EXT)/1e6}}isQueryAvailable(t,e){if(e===0)return!0;if(e===2){let n=this.gl,o=this.getQueryTimerExtensionWebGL2(),s=n.getQueryParameter(t,n.QUERY_RESULT_AVAILABLE);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(o.GPU_DISJOINT_EXT)),s&&!this.disjoint}else{let n=this.getQueryTimerExtensionWebGL1(),o=n.getQueryObjectEXT(t,n.QUERY_RESULT_AVAILABLE_EXT);return this.disjoint==null&&(this.disjoint=this.gl.getParameter(n.GPU_DISJOINT_EXT)),o&&!this.disjoint}}pollFence(t){return new Promise(e=>{this.addItemToPoll(()=>t.isFencePassed(),()=>e())})}pollItems(){let t=dtt(this.itemsToPoll.map(e=>e.isDoneFn));for(let e=0;e<=t;++e){let{resolveFn:n}=this.itemsToPoll[e];n()}this.itemsToPoll=this.itemsToPoll.slice(t+1)}addItemToPoll(t,e){this.itemsToPoll.push({isDoneFn:t,resolveFn:e}),!(this.itemsToPoll.length>1)&&x.repeatedTry(()=>(this.pollItems(),this.itemsToPoll.length===0))}bindTextureToFrameBuffer(t){this.throwIfDisposed(),ig(this.gl,t,this.framebuffer),this.debug&&Wm(this.gl)}unbindTextureToFrameBuffer(){this.outputTexture!=null?(ig(this.gl,this.outputTexture,this.framebuffer),this.debug&&Wm(this.gl)):mb(this.gl,this.framebuffer)}downloadMatrixDriver(t,e){this.bindTextureToFrameBuffer(t);let n=e();return this.unbindTextureToFrameBuffer(),n}setOutputMatrixTextureDriver(t,e,n){this.throwIfDisposed();let o=this.gl;ig(o,t,this.framebuffer),this.debug&&Wm(o),this.outputTexture=t,vt(o,()=>o.viewport(0,0,e,n)),vt(o,()=>o.scissor(0,0,e,n))}setOutputMatrixWriteRegionDriver(t,e,n,o){this.throwIfDisposed(),vt(this.gl,()=>this.gl.scissor(t,e,n,o))}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 dtt(r){let t=0;for(;t<r.length&&r[t]();++t);return t-1}var{addImpl:qP,bincountImpl:vb,bincountReduceImpl:KP,ceilImpl:jP,concatImpl:XP,equalImpl:YP,expImpl:ZP,expm1Impl:JP,floorImpl:QP,gatherNdImpl:tM,gatherV2Impl:eM,greaterImpl:rM,greaterEqualImpl:nM,lessImpl:oM,lessEqualImpl:sM,linSpaceImpl:iM,logImpl:aM,maxImpl:lM,maximumImpl:uM,minimumImpl:cM,multiplyImpl:pM,negImpl:mM,notEqualImpl:fM,prodImpl:dM,rangeImpl:hM,rsqrtImpl:gM,sigmoidImpl:xM,simpleAbsImpl:Cb,sliceImpl:yM,sparseFillEmptyRowsImpl:bM,sparseReshapeImpl:wM,sparseSegmentReductionImpl:Ib,sqrtImpl:vM,stridedSliceImpl:CM,stringNGramsImpl:IM,stringSplitImpl:SM,stringToHashBucketFastImpl:kM,subImpl:NM,tileImpl:TM,topKImpl:_M,transposeImpl:Rc,uniqueImpl:EM}=eb;function $k(r,t){return["x","y","z","w","u","v"].slice(0,t).map(e=>`${r}.${e}`)}function rr(r,t){return t===1?[r]:$k(r,t)}function AM(r,t){if(r===1)return"rc";let e="";for(let n=0;n<r;n++)e+=t[n],n<r-1&&(e+=",");return e}var Rk=class{constructor(t){if(this.variableNames=["A"],this.packedInputs=!1,this.packedOutput=!0,this.outputShape=t,this.rank=t.length,this.enableShapeUniforms=$e(this.outputShape.length),this.rank===0)this.userCode=`
void main() {
setOutput(vec4(getA(), 0., 0., 0.));
}
`;else{let e=rr("rc",this.rank),n=zt(this.rank),o=this.getOutOfBoundsCondition(e),s=this.getSetup(e),i=this.getOutput(e);this.userCode=`
void main() {
${n} rc = getOutputCoords();
if(${o}) {
setOutput(vec4(0));
} else {
${s}
setOutput(vec4(${i}));
}
}
`}}getSourceCoordsArr(t){let e=[];for(let n=0;n<=1;n++)for(let o=0;o<=1;o++){let s=`${n===0?"r":"rp1"}, ${o===0?"c":"cp1"}`;for(let i=2;i<this.rank;i++)s=`${t[t.length-1-i]},`+s;e.push(s)}return e}getOutOfBoundsCondition(t){if(this.rank===1)return`rc > ${this.enableShapeUniforms?"outShape":this.outputShape[0]}`;let e="";for(let n=this.rank-2;n<this.rank;n++)e+=`${t[n]} >= ${this.enableShapeUniforms?`outShape[${n}]`:this.outputShape[n]}`,n<this.rank-1&&(e+="||");return e}getSetup(t){if(this.rank===1)return"";let e=t.slice(-2),n=this.enableShapeUniforms?`outShape[${this.rank} - 1]`:this.outputShape[this.rank-1],o=this.enableShapeUniforms?`outShape[${this.rank} - 2]`:this.outputShape[this.rank-2];return`
int r = ${e[0]};
int c = ${e[1]};
int rp1 = r + 1;
int cp1 = c + 1;
bool cEdge = cp1 >= ${n};
bool rEdge = rp1 >= ${o};
`}getOutput(t){let e=this.getSourceCoordsArr(t);return this.rank===1?`getA(rc), (rc + 1 >= ${this.enableShapeUniforms?"outShape":this.outputShape[0]} ? 0. : getA(rc + 1)), 0, 0`:`getA(${e[0]}),
cEdge ? 0. : getA(${e[1]}),
rEdge ? 0. : getA(${e[2]}),
rEdge || cEdge ? 0. : getA(${e[3]})`}};var ug=class{constructor(t,e){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.customUniforms=[{name:"inputShape",type:"ivec3"}],this.outputShape=t,this.enableShapeUniforms=$e(this.outputShape.length);let n="";for(let o=0;o<4;o++){let s="thisRC = rc;";o%2===1&&(s+="thisRC.z += 1;"),o>1&&(s+="thisRC.y += 1;"),n+=`
${s}
${o>0?"if(thisRC.y < rows && thisRC.z < cols){":""}
int flatIndex = getFlatIndex(thisRC);
ivec3 inputRC = inputCoordsFromReshapedOutCoords(flatIndex);
vec2 inputRCInnerDims = vec2(float(inputRC.y),float(inputRC.z));
result[${o}] =
getChannel(getA(inputRC.x, inputRC.y, inputRC.z), inputRCInnerDims);
${o>0?"}":""}
`}this.userCode=`
${htt(e,this.enableShapeUniforms)}
${this.enableShapeUniforms?qm():Hm(t)}
void main() {
ivec3 rc = getOutputCoords();
vec4 result = vec4(0.);
ivec3 thisRC;
int rows = ${this.enableShapeUniforms?"outShape[1]":t[1]};
int cols = ${this.enableShapeUniforms?"outShape[2]":t[2]};
${n}
setOutput(result);
}
`}};function htt(r,t){return`
ivec3 inputCoordsFromReshapedOutCoords(int index) {
${t?MP(["r","c","d"],"inputShape"):ui(["r","c","d"],r)}
return ivec3(r, c, d);
}
`}var Fk=class{constructor(t){this.gpgpu=t,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0,this.freeTextures={},this.logEnabled=!1,this.usedTextures={}}acquireTexture(t,e,n){let o=$M(e,n),s=RM(t,o,n);s in this.freeTextures||(this.freeTextures[s]=[]),s in this.usedTextures||(this.usedTextures[s]=[]);let i=DM(t,o,this.gpgpu.gl,this.gpgpu.textureConfig,n);if(this.freeTextures[s].length>0){this.numFreeTextures--,this.numUsedTextures++,this._numBytesFree-=i,this.log();let u=this.freeTextures[s].shift();return this.usedTextures[s].push(u),u}let a;return o===Rr.PACKED_2X2_FLOAT32?a=this.gpgpu.createPackedMatrixTexture(t[0],t[1]):o===Rr.PACKED_2X2_FLOAT16?a=this.gpgpu.createFloat16PackedMatrixTexture(t[0],t[1]):o===Rr.UNPACKED_FLOAT32?a=this.gpgpu.createFloat32MatrixTexture(t[0],t[1]):o===Rr.UNPACKED_FLOAT16?a=this.gpgpu.createFloat16MatrixTexture(t[0],t[1]):o===Rr.PACKED_4X1_UNSIGNED_BYTE&&(a=this.gpgpu.createUnsignedBytesMatrixTexture(t[0],t[1])),this.usedTextures[s].push(a),this.numUsedTextures++,this._numBytesAllocated+=i,this.log(),a}releaseTexture(t,e,n,o){if(this.freeTextures==null)return;let s=$M(n,o),i=RM(e,s,o);i in this.freeTextures||(this.freeTextures[i]=[]);let a=DM(e,s,this.gpgpu.gl,this.gpgpu.textureConfig,o),u=V().get("WEBGL_DELETE_TEXTURE_THRESHOLD");u!==-1&&this._numBytesAllocated>u?(this.gpgpu.deleteMatrixTexture(t.texture),this._numBytesAllocated-=a):(this.freeTextures[i].push(t),this.numFreeTextures++,this._numBytesFree+=a),this.numUsedTextures--;let l=this.usedTextures[i],c=l.indexOf(t);if(c<0)throw new Error("Cannot release a texture that was never provided by this texture manager");l.splice(c,1),this.log()}log(){if(!this.logEnabled)return;let t=this.numFreeTextures+this.numUsedTextures;console.log("Free/Used",`${this.numFreeTextures} / ${this.numUsedTextures}`,`(${t})`);let e=this._numBytesFree/this._numBytesAllocated;console.log(`Bytes allocated: ${this._numBytesAllocated}`),console.log(`Bytes unused: ${this._numBytesFree} (${Math.round(100*e)}%)`)}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 t in this.freeTextures)this.freeTextures[t].forEach(e=>{this.gpgpu.deleteMatrixTexture(e.texture)});for(let t in this.usedTextures)this.usedTextures[t].forEach(e=>{this.gpgpu.deleteMatrixTexture(e.texture)});this.freeTextures=null,this.usedTextures=null,this.numUsedTextures=0,this.numFreeTextures=0,this._numBytesAllocated=0,this._numBytesFree=0}}};function gtt(r,t){let e=r;if(t===e.R32F)return 4;if(t===e.R16F)return 2;if(t===e.RGBA32F)return 16;if(t===r.RGBA)return 16;if(t===e.RGBA16F)return 8;if(t===e.RGBA8)return 4;throw new Error(`Unknown internal format ${t}`)}function DM(r,t,e,n,o){let s=xtt(t,n),i;if(o){let[u,l]=Hi(r[0],r[1]);i=u*l}else{let[u,l]=Ac(r[0],r[1]);i=u*l}let a=gtt(e,s);return i*a}function xtt(r,t){switch(r){case Rr.PACKED_2X2_FLOAT32:return bb(t);case Rr.PACKED_2X2_FLOAT16:return wb(t);case Rr.UNPACKED_FLOAT32:return gb(t);case Rr.UNPACKED_FLOAT16:return xb(t);case Rr.PACKED_4X1_UNSIGNED_BYTE:return yb(t);default:throw new Error(`Unknown physical texture type ${r}`)}}function ytt(r){return V().getBool("WEBGL_RENDER_FLOAT32_ENABLED")?r?Rr.PACKED_2X2_FLOAT32:Rr.UNPACKED_FLOAT32:r?Rr.PACKED_2X2_FLOAT16:Rr.UNPACKED_FLOAT16}function $M(r,t){if(r===Ur.UPLOAD)return Rr.PACKED_2X2_FLOAT32;if(r===Ur.RENDER||r==null)return ytt(t);if(r===Ur.DOWNLOAD||r===Ur.PIXELS)return Rr.PACKED_4X1_UNSIGNED_BYTE;throw new Error(`Unknown logical texture type ${r}`)}function RM(r,t,e){return`${r[0]}_${r[1]}_${t}_${e}`}var vn=class{constructor(t,e){this.variableNames=["A"],this.outputShape=t,this.enableShapeUniforms=$e(this.outputShape.length),this.userCode=`
float unaryOperation(float x) {
${e}
}
void main() {
float x = getAAtOutCoords();
float y = unaryOperation(x);
setOutput(y);
}
`}},hr="if (isnan(x)) return x;",FM="return x;",Ok="return abs(x);";var OM="return (x >= 0.0) ? x : (exp(x) - 1.0);",PM=hr+`
return (x < 0.0) ? 0.0 : x;
`,MM=hr+`
return (x < 0.0) ? 0.0 : min(6.0, x);
`,Fc="return x;",LM="return 1.0 / (1.0 + exp(-1.0 * x));";var BM="return x;",VM=`
vec4 result;
result.r = (x.r >= 0.0) ? x.r : (exp(x.r) - 1.0);
result.g = (x.g >= 0.0) ? x.g : (exp(x.g) - 1.0);
result.b = (x.b >= 0.0) ? x.b : (exp(x.b) - 1.0);
result.a = (x.a >= 0.0) ? x.a : (exp(x.a) - 1.0);
return result;
`,GM=`
vec4 result = x * vec4(greaterThanEqual(x, vec4(0.0)));
bvec4 isNaN = isnan(x);
result.r = isNaN.r ? x.r : result.r;
result.g = isNaN.g ? x.g : result.g;
result.b = isNaN.b ? x.b : result.b;
result.a = isNaN.a ? x.a : result.a;
return result;
`,WM=`
vec4 result = min(x, vec4(6.)) * vec4(greaterThanEqual(x, vec4(0.0)));
bvec4 isNaN = isnan(x);
result.r = isNaN.r ? x.r : result.r;
result.g = isNaN.g ? x.g : result.g;
result.b = isNaN.b ? x.b : result.b;
result.a = isNaN.a ? x.a : result.a;
return result;
`,UM="return 1.0 / (1.0 + exp(-1.0 * x));",Fo=class{constructor(t,e){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t,this.enableShapeUniforms=$e(this.outputShape.length),this.userCode=`
vec4 unaryOperation(vec4 x) {
${e}
}
void main() {
vec4 x = getAAtOutCoords();
vec4 y = unaryOperation(x);
setOutput(y);
}
`}};var Pk=class{constructor(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!1,this.outputShape=t,this.enableShapeUniforms=$e(this.outputShape.length);let e=t.length,n=rr("rc",e),o=zt(e),s=AM(e,n),i=n.slice(-2),a=e<=1?"rc":`vec2(${i.join(",")})`;this.userCode=`
void main() {
${o} rc = getOutputCoords();
vec4 packedInput = getA(${s});
setOutput(getChannel(packedInput, ${a}));
}
`}};var wtt=Vr.whereImpl,vtt=1e-7,Ctt=1e-4,Sb={};function Itt(r){return r in Sb||(Sb[r]={}),Sb[r]}var Stt=V().getNumber("CPU_HANDOFF_SIZE_THRESHOLD"),ktt=600;function Ntt(){return V().global.screen==null?1024:V().global.screen.height*V().global.screen.width*window.devicePixelRatio*ktt/1024/1024}var Oc=class extends mi{constructor(t){super();if(this.pendingRead=new WeakMap,this.pendingDisposal=new WeakSet,this.dataRefCount=new WeakMap,this.numBytesInGPU=0,this.uploadWaitMs=0,this.downloadWaitMs=0,this.lastGlFlushTime=0,this.warnedAboutMemory=!1,this.pendingDeletes=0,this.disposed=!1,!V().getBool("HAS_WEBGL"))throw new Error("WebGL is not supported on this device");let e;if(t!=null){if(t instanceof Jm)e=t;else{let n=Vn(V().getNumber("WEBGL_VERSION"),t);e=new Jm(n)}this.binaryCache={},this.gpgpuCreatedLocally=!1}else{let n=Vn(V().getNumber("WEBGL_VERSION"));e=new Jm(n),this.binaryCache=Itt(V().getNumber("WEBGL_VERSION")),this.gpgpuCreatedLocally=!0}this.gpgpu=e,this.canvas=this.gpgpu.gl.canvas,this.textureManager=new Fk(this.gpgpu),this.numMBBeforeWarning=Ntt(),this.texData=new wl(this,fo())}nextDataId(){return Oc.nextDataId++}numDataIds(){return this.texData.numDataIds()-this.pendingDeletes}write(t,e,n){if((V().getBool("WEBGL_CHECK_NUMERICAL_PROBLEMS")||V().getBool("DEBUG"))&&this.checkNumericalProblems(t),n==="complex64"&&t!=null)throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");let o={id:this.nextDataId()};return this.texData.set(o,{shape:e,dtype:n,values:t,usage:Ur.UPLOAD,refCount:1}),o}refCount(t){return this.texData.has(t)?this.texData.get(t).refCount:0}incRef(t){let e=this.texData.get(t);e.refCount++}decRef(t){if(this.texData.has(t)){let e=this.texData.get(t);e.refCount--}}move(t,e,n,o,s){if(V().getBool("DEBUG")&&this.checkNumericalProblems(e),o==="complex64")throw new Error("Cannot write to a complex64 dtype. Please use tf.complex(real, imag).");this.texData.set(t,{shape:n,dtype:o,values:e,usage:Ur.UPLOAD,refCount:s})}disposeIntermediateTensorInfo(t){this.disposeData(t.dataId)}readSync(t){let e=this.texData.get(t),{values:n,dtype:o,complexTensorInfos:s,slice:i,shape:a,isPacked:u}=e;if(i!=null){let m;u?m=new Fo(a,Fc):m=new vn(a,Fc);let f=this.runWebGLProgram(m,[{dataId:t,shape:a,dtype:o}],o),d=this.readSync(f.dataId);return this.disposeIntermediateTensorInfo(f),d}if(n!=null)return this.convertAndCacheOnCPU(t);if(o==="string")return n;let l=this.activeTimers!=null,c;l&&(c=x.now());let p;if(o==="complex64"){let m=this.readSync(s.real.dataId),f=this.readSync(s.imag.dataId);p=S.mergeRealAndImagArrays(m,f)}else p=this.getValuesFromTexture(t);return l&&(this.downloadWaitMs+=x.now()-c),this.convertAndCacheOnCPU(t,p)}async read(t){if(this.pendingRead.has(t)){let d=this.pendingRead.get(t);return new Promise(h=>d.push(h))}let e=this.texData.get(t),{values:n,shape:o,slice:s,dtype:i,complexTensorInfos:a,isPacked:u}=e;if(s!=null){let d;u?d=new Fo(o,Fc):d=new vn(o,Fc);let h=this.runWebGLProgram(d,[{dataId:t,shape:o,dtype:i}],i),g=this.read(h.dataId);return this.disposeIntermediateTensorInfo(h),g}if(n!=null)return this.convertAndCacheOnCPU(t);if(V().getBool("DEBUG")&&!V().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")&&V().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(i!=="complex64"&&V().get("WEBGL_BUFFER_SUPPORTED")){c=this.decode(t);let d=this.texData.get(c.dataId);l=this.gpgpu.createBufferFromTexture(d.texture.texture,...ng(o))}this.pendingRead.set(t,[]),i!=="complex64"&&await this.gpgpu.createAndWaitForFence();let p;if(i==="complex64"){let d=await Promise.all([this.read(a.real.dataId),this.read(a.imag.dataId)]),h=d[0],g=d[1];p=S.mergeRealAndImagArrays(h,g)}else if(l==null)p=this.getValuesFromTexture(t);else{let d=x.sizeFromShape(o);p=this.gpgpu.downloadFloat32MatrixFromBuffer(l,d)}if(c!=null&&this.disposeIntermediateTensorInfo(c),l!=null){let d=this.gpgpu.gl;vt(d,()=>d.deleteBuffer(l))}let m=this.convertAndCacheOnCPU(t,p),f=this.pendingRead.get(t);return this.pendingRead.delete(t),f.forEach(d=>d(m)),this.pendingDisposal.has(t)&&(this.pendingDisposal.delete(t),this.disposeData(t)&&fo().removeDataId(t,this),this.pendingDeletes--),m}readToGPU(t,e={}){let n=this.texData.get(t),{values:o,shape:s,slice:i,dtype:a,isPacked:u,texture:l}=n;if(a==="complex64")throw new Error("Does not support reading texture for complex64 dtype.");if(i!=null){let f;u?f=new Fo(s,Fc):f=new vn(s,Fc);let d=this.runWebGLProgram(f,[{dataId:t,shape:s,dtype:a}],a),h=this.readToGPU(d,e);return this.disposeIntermediateTensorInfo(d),h}if(l==null)throw o!=null?new Error("Data is not on GPU but on CPU."):new Error("There is no data on GPU or CPU.");let c=this.decode(t,e.customTexShape),p=fo().makeTensorFromDataId(c.dataId,c.shape,c.dtype),m=this.texData.get(c.dataId);return Object.assign({tensorRef:p},m.texture)}bufferSync(t){let e=this.readSync(t.dataId),n=e;if(t.dtype==="string")try{n=e.map(o=>x.decodeString(o))}catch(o){throw new Error("Failed to decode encoded string bytes into utf-8")}return It(t.shape,t.dtype,n)}checkNumericalProblems(t){if(t!=null)for(let e=0;e<t.length;e++){let n=t[e];if(!GS(n))throw V().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 ${n} cannot be represented on this device.`)}}getValuesFromTexture(t){let{shape:e,dtype:n,isPacked:o}=this.texData.get(t),s=x.sizeFromShape(e);if(V().getBool("WEBGL_DOWNLOAD_FLOAT_ENABLED")){let m=this.decode(t),f=this.texData.get(m.dataId),d=this.gpgpu.downloadMatrixFromPackedTexture(f.texture.texture,...ng(e)).subarray(0,s);return this.disposeIntermediateTensorInfo(m),d}let i=V().getBool("WEBGL_PACK")&&o===!0,a=i?ag(e):e,u=i?new mk(a):new pk(a),l=this.runWebGLProgram(u,[{shape:a,dtype:n,dataId:t}],"float32"),c=this.texData.get(l.dataId),p=this.gpgpu.downloadByteEncodedFloatMatrixFromOutputTexture(c.texture.texture,c.texShape[0],c.texShape[1]).subarray(0,s);return this.disposeIntermediateTensorInfo(l),p}timerAvailable(){return V().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0}time(t){let e=this.activeTimers,n=[],o=!1;this.programTimersStack==null?(this.programTimersStack=n,o=!0):this.activeTimers.push(n),this.activeTimers=n,t();let s=x.flatten(this.activeTimers.map(u=>u.query)).filter(u=>u!=null),i=x.flatten(this.activeTimers.map(u=>u.name)).filter(u=>u!=null);this.activeTimers=e,o&&(this.programTimersStack=null);let a={uploadWaitMs:this.uploadWaitMs,downloadWaitMs:this.downloadWaitMs,kernelMs:null,wallMs:null};return(async()=>{if(V().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0){let u=await Promise.all(s);a.kernelMs=x.sum(u),a.getExtraProfileInfo=()=>u.map((l,c)=>({name:i[c],ms:l})).map(l=>`${l.name}: ${l.ms}`).join(", ")}else a.kernelMs={error:"WebGL query timers are not supported in this environment."};return this.uploadWaitMs=0,this.downloadWaitMs=0,a})()}memory(){return{unreliable:!1,numBytesInGPU:this.numBytesInGPU,numBytesInGPUAllocated:this.textureManager.numBytesAllocated,numBytesInGPUFree:this.textureManager.numBytesFree}}startTimer(){return V().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?this.gpgpu.beginQuery():{startMs:x.now(),endMs:null}}endTimer(t){return V().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0?(this.gpgpu.endQuery(),t):(t.endMs=x.now(),t)}async getQueryTime(t){if(V().getNumber("WEBGL_DISJOINT_QUERY_TIMER_EXTENSION_RELIABLE")>0)return this.gpgpu.waitForQueryAndGetTime(t);let e=t;return e.endMs-e.startMs}disposeData(t,e=!1){if(this.pendingDisposal.has(t))return!1;if(!this.texData.has(t))return!0;if(e?this.texData.get(t).refCount=0:this.texData.get(t).refCount--,!e&&this.texData.get(t).refCount>0)return!1;if(this.pendingRead.has(t))return this.pendingDisposal.add(t),this.pendingDeletes++,!1;this.releaseGPUData(t);let{complexTensorInfos:n}=this.texData.get(t);return n!=null&&(this.disposeData(n.real.dataId,e),this.disposeData(n.imag.dataId,e)),this.texData.delete(t),!0}releaseGPUData(t){let{texture:e,dtype:n,texShape:o,usage:s,isPacked:i,slice:a}=this.texData.get(t),u=a&&a.origDataId||t,l=this.dataRefCount.get(u);l>1?this.dataRefCount.set(u,l-1):(this.dataRefCount.delete(u),e!=null&&(this.numBytesInGPU-=this.computeBytes(o,n),this.textureManager.releaseTexture(e,o,s,i)));let c=this.texData.get(t);c.texture=null,c.texShape=null,c.isPacked=!1,c.slice=null}getTexture(t){return this.uploadToGPU(t),this.texData.get(t).texture.texture}getDataInfo(t){return this.texData.get(t)}shouldExecuteOnCPU(t,e=Stt){return V().getBool("WEBGL_CPU_FORWARD")&&t.every(n=>this.texData.get(n.dataId).texture==null&&x.sizeFromShape(n.shape)<e)}getGPGPUContext(){return this.gpgpu}where(t){S.warn("tf.where() in webgl locks the UI thread. Call tf.whereAsync() instead");let e=t.dataSync();return wtt(t.shape,e)}packedUnaryOp(t,e,n){let o=new Fo(t.shape,e),s=this.compileAndRun(o,[t],n);return fo().makeTensorFromDataId(s.dataId,s.shape,s.dtype)}abs(t){if(this.shouldExecuteOnCPU([t])&&t.dtype!=="complex64"){let o=Cb(this.texData.get(t.dataId).values);return this.makeOutput(t.shape,t.dtype,o)}if(V().getBool("WEBGL_PACK_UNARY_OPERATIONS"))return this.packedUnaryOp(t,Ok,t.dtype);let e=new vn(t.shape,Ok),n=this.compileAndRun(e,[t]);return fo().makeTensorFromDataId(n.dataId,n.shape,n.dtype)}makeTensorInfo(t,e,n){let o;if(e==="string"&&n!=null&&n.length>0&&x.isString(n[0])){let s=n.map(i=>x.encodeString(i));o=this.write(s,t,e)}else o=this.write(n,t,e);return this.texData.get(o).usage=null,{dataId:o,shape:t,dtype:e}}makeOutput(t,e,n){let{dataId:o}=this.makeTensorInfo(t,e,n);return fo().makeTensorFromDataId(o,t,e,this)}unpackTensor(t){let e=new Pk(t.shape);return this.runWebGLProgram(e,[t],t.dtype)}packTensor(t){let e=new Rk(t.shape),n=!0;return this.runWebGLProgram(e,[t],t.dtype,null,n)}packedReshape(t,e){let n=[pl(t.shape),...ml(t.shape)],o={dtype:t.dtype,shape:n,dataId:t.dataId},s=[pl(e),...ml(e)],i=new ug(s,n),a=!0,u=[n],l=this.runWebGLProgram(i,[o],t.dtype,u,a);return{dataId:l.dataId,shape:e,dtype:l.dtype}}decode(t,e){let n=this.texData.get(t),{isPacked:o,shape:s,dtype:i}=n;if(e!=null){let m=x.sizeFromShape(s),f=e[0]*e[1]*4;x.assert(m<=f,()=>"customTexShape is too small. Row * Column * 4 should be equal or larger than the size of the tensor data.")}let a=ag(s),u;o?u=new ck(a):u=new uk(a);let l=!0,c=[e!=null?e:ng(a)],p=this.runWebGLProgram(u,[{shape:a,dtype:i,dataId:t}],i,c,l,e);return{dtype:i,shape:s,dataId:p.dataId}}runWebGLProgram(t,e,n,o,s=!1,i){let a=this.makeTensorInfo(t.outputShape,n),u=this.texData.get(a.dataId);if(t.packedOutput&&(u.isPacked=!0),t.outPackingScheme===ou.DENSE){let y=i!=null?i:ng(t.outputShape);u.texShape=y.map(b=>b*2)}if(t.outTexUsage!=null&&(u.usage=t.outTexUsage),x.sizeFromShape(a.shape)===0)return u.values=x.getTypedArrayFromDType(a.dtype,0),a;let l=[],c=e.map(y=>{if(y.dtype==="complex64")throw new Error("GPGPUProgram does not support complex64 input. For complex64 dtypes, please separate the program into real and imaginary parts.");let b=this.texData.get(y.dataId);if(b.texture==null){if(!t.packedInputs&&x.sizeFromShape(y.shape)<=V().getNumber("WEBGL_SIZE_UPLOAD_UNIFORM"))return{shape:y.shape,texData:null,isUniform:!0,uniformValues:b.values};t.packedInputs&&(b.isPacked=!0,b.shape=y.shape)}if(this.uploadToGPU(y.dataId),!!b.isPacked!=!!t.packedInputs)y=b.isPacked?this.unpackTensor(y):this.packTensor(y),l.push(y),b=this.texData.get(y.dataId);else if(b.isPacked&&!su(b.shape,y.shape)){let w=y,v=y.shape;y.shape=b.shape,y=this.packedReshape(y,v),l.push(y),b=this.texData.get(y.dataId),w.shape=v}return{shape:y.shape,texData:b,isUniform:!1}});this.uploadToGPU(a.dataId);let p={shape:a.shape,texData:u,isUniform:!1},m=HP(t,c,p),f=this.getAndSaveBinary(m,()=>WP(this.gpgpu,t,c,p)),d=this.activeTimers!=null,h;d&&(h=this.startTimer()),V().get("ENGINE_COMPILE_ONLY")||UP(this.gpgpu,f,c,p,o),l.forEach(y=>this.disposeIntermediateTensorInfo(y)),d&&(h=this.endTimer(h),this.activeTimers.push({name:t.constructor.name,query:this.getQueryTime(h)}));let g=V().get("WEBGL_FLUSH_THRESHOLD");if(g>0){let y=x.now();y-this.lastGlFlushTime>g&&(this.gpgpu.gl.flush(),this.lastGlFlushTime=y)}if(!V().getBool("WEBGL_LAZILY_UNPACK")&&u.isPacked&&s===!1){let y=this.unpackTensor(a);return this.disposeIntermediateTensorInfo(a),y}return a}compileAndRun(t,e,n,o,s=!1){return n=n||e[0].dtype,this.runWebGLProgram(t,e,n,o,s)}getAndSaveBinary(t,e){return t in this.binaryCache||(this.binaryCache[t]=e()),this.binaryCache[t]}getTextureManager(){return this.textureManager}dispose(){this.disposed||(V().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=B(()=>{if(!V().get("WEBGL_RENDER_FLOAT32_ENABLED")){let t=V().getBool("DEBUG");V().set("DEBUG",!1);let e=this.abs(pt(1e-8)).dataSync()[0];if(V().set("DEBUG",t),e>0)return 32}return 16})),this.floatPrecisionValue}epsilon(){return this.floatPrecision()===32?vtt:Ctt}uploadToGPU(t){let e=this.texData.get(t),{shape:n,dtype:o,values:s,texture:i,usage:a,isPacked:u}=e;if(i!=null)return;let l=this.activeTimers!=null,c;l&&(c=x.now());let p=e.texShape;if(p==null&&(p=ek(n,u),e.texShape=p),s!=null){let m=ag(n),f,d=p[1],h=p[0],g=s instanceof Uint8Array||s instanceof Uint8ClampedArray;(u||!g)&&([d,h]=Hi(p[0],p[1])),u?f=new dk(m,g):f=new fk(m,g);let y=g?[h,d]:p,b=this.makeTensorInfo(y,o),w=this.texData.get(b.dataId);g?w.usage=Ur.PIXELS:w.usage=Ur.UPLOAD,w.texShape=y,this.gpgpu.uploadDenseMatrixToTexture(this.getTexture(b.dataId),d,h,s);let v=[[h,d]],k=!0,E=this.runWebGLProgram(f,[b],o,v,k),D=this.texData.get(E.dataId);e.texShape=D.texShape,e.isPacked=D.isPacked,e.usage=D.usage,V().get("ENGINE_COMPILE_ONLY")?this.disposeData(E.dataId):(e.texture=D.texture,e.values=null,this.texData.delete(E.dataId)),this.disposeIntermediateTensorInfo(b),l&&(this.uploadWaitMs+=x.now()-c)}else{let m=this.acquireTexture(p,a,o,u);e.texture=m}}convertAndCacheOnCPU(t,e){let n=this.texData.get(t),{dtype:o}=n;return this.releaseGPUData(t),e!=null&&(n.values=Ttt(e,o)),n.values}acquireTexture(t,e,n,o){if(this.numBytesInGPU+=this.computeBytes(t,n),!this.warnedAboutMemory&&this.numBytesInGPU>this.numMBBeforeWarning*1024*1024){let s=(this.numBytesInGPU/1024/1024).toFixed(2);this.warnedAboutMemory=!0,console.warn(`High memory usage in GPU: ${s} MB, most likely due to a memory leak`)}return this.textureManager.acquireTexture(t,e,o)}computeBytes(t,e){return t[0]*t[1]*x.bytesPerElement(e)}checkCompileCompletion(){for(let[,t]of Object.entries(this.binaryCache))this.checkCompletion_(t)}async checkCompileCompletionAsync(){let t=[];if(this.gpgpu.parallelCompilationExtension){for(let[,e]of Object.entries(this.binaryCache))t.push(this.checkCompletionAsync_(e));return Promise.all(t)}else{for(let[,e]of Object.entries(this.binaryCache)){let n=new Promise(o=>{try{this.checkCompletion_(e),o(!0)}catch(s){throw s}});t.push(n)}return Promise.all(t)}}async checkCompletionAsync_(t){return this.gpgpu.gl.getProgramParameter(t.webGLProgram,this.gpgpu.parallelCompilationExtension.COMPLETION_STATUS_KHR)?this.checkCompletion_(t):(await fd(),this.checkCompletionAsync_(t))}checkCompletion_(t){if(this.gpgpu.gl.getProgramParameter(t.webGLProgram,this.gpgpu.gl.LINK_STATUS)===!1)throw console.log(this.gpgpu.gl.getProgramInfoLog(t.webGLProgram)),this.gpgpu.gl.getShaderParameter(t.fragmentShader,this.gpgpu.gl.COMPILE_STATUS)===!1?(cb(t.source,this.gpgpu.gl.getShaderInfoLog(t.fragmentShader)),new Error("Failed to compile fragment shader.")):new Error("Failed to link vertex and fragment shaders.");return!0}getUniformLocations(){for(let[,t]of Object.entries(this.binaryCache)){let{uniformLocations:e,customUniformLocations:n,infLoc:o,nanLoc:s,inShapesLocations:i,inTexShapesLocations:a,outShapeLocation:u,outShapeStridesLocation:l,outTexShapeLocation:c}=lk(this.gpgpu,t.program,t.webGLProgram);t.uniformLocations=e,t.customUniformLocations=n,t.infLoc=o,t.nanLoc=s,t.inShapesLocations=i,t.inTexShapesLocations=a,t.outShapeLocation=u,t.outShapeStridesLocation=l,t.outTexShapeLocation=c}}};Oc.nextDataId=0;function Ttt(r,t){if(t==="float32"||t==="complex64")return r;if(t==="int32"||t==="bool"){let e=t==="int32"?new Int32Array(r.length):new Uint8Array(r.length);for(let n=0;n<e.length;++n)e[n]=Math.round(r[n]);return e}else throw new Error(`Unknown dtype ${t}`)}var HM="3.15.0";function qM(){V().set("WEBGL_FORCE_F16_TEXTURES",!0)}zl.isBrowser()&&Xp("webgl",()=>new Oc,2);var XSe={forceHalfFloat:qM};var kb=`
if (isnan(a)) return a;
if (isnan(b)) return b;
`;var Oo=class{constructor(t,e,n){this.variableNames=["A","B"],this.outputShape=S.assertAndGetBroadcastShape(e,n),this.enableShapeUniforms=$e(this.outputShape.length),this.userCode=`
float binaryOperation(float a, float b) {
${t}
}
void main() {
float a = getAAtOutCoords();
float b = getBAtOutCoords();
setOutput(binaryOperation(a, b));
}
`}};var iu=`
result.r = isNaN.r > 0. ? NAN : result.r;
result.g = isNaN.g > 0. ? NAN : result.g;
result.b = isNaN.b > 0. ? NAN : result.b;
result.a = isNaN.a > 0. ? NAN : result.a;
`;var ci=class{constructor(t,e,n,o=!1){this.variableNames=["A","B"],this.supportsBroadcasting=!0,this.packedInputs=!0,this.packedOutput=!0,this.outputShape=S.assertAndGetBroadcastShape(e,n);let s=this.outputShape.length;this.enableShapeUniforms=$e(s);let i="";if(o)if(s===0||x.sizeFromShape(this.outputShape)===1)i=`
result.y = 0.;
result.z = 0.;
result.w = 0.;
`;else if(i=`
${zt(s)} coords = getOutputCoords();
`,s===1)this.enableShapeUniforms?i+=`
result.y = (coords + 1) >= outShape ? 0. : result.y;
result.z = 0.;
result.w = 0.;
`:i+=`
result.y = (coords + 1) >= ${this.outputShape[0]} ? 0. : result.y;
result.z = 0.;
result.w = 0.;
`;else{let u=rr("coords",s);this.enableShapeUniforms?i+=`
bool nextRowOutOfBounds =
(${u[s-2]} + 1) >= outShape[${s} - 2];
bool nextColOutOfBounds =
(${u[s-1]} + 1) >= outShape[${s} - 1];
result.y = nextColOutOfBounds ? 0. : result.y;
result.z = nextRowOutOfBounds ? 0. : result.z;
result.w = nextColOutOfBounds || nextRowOutOfBounds ? 0. : result.w;
`:i+=`
bool nextRowOutOfBounds =
(${u[s-2]} + 1) >= ${this.outputShape[s-2]};
bool nextColOutOfBounds =
(${u[s-1]} + 1) >= ${this.outputShape[s-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) {
${t}
}
void main() {
vec4 a = getAAtOutCoords();
vec4 b = getBAtOutCoords();
vec4 result = binaryOperation(a, b);
${i}
setOutput(result);
}
`}};function Ke(r){let{inputs:t,backend:e}=r,{x:n}=t;return e.incRef(n.dataId),{dataId:n.dataId,shape:n.shape,dtype:n.dtype}}var KM={kernelName:io,backendName:"webgl",kernelFunc:Ke};function Dn(r){let{inputs:t,backend:e}=r,{real:n,imag:o}=t,s=e.makeTensorInfo(n.shape,"complex64"),i=e.texData.get(s.dataId),a=Ke({inputs:{x:n},backend:e}),u=Ke({inputs:{x:o},backend:e});return i.complexTensorInfos={real:a,imag:u},s}var jM={kernelName:mp,backendName:"webgl",kernelFunc:Dn};var Mk="return (a < 0.) ? b * a : a;",Lk=`
vec4 aLessThanZero = vec4(lessThan(a, vec4(0.)));
return (aLessThanZero * (b * a)) + ((vec4(1.0) - aLessThanZero) * a);
`;function _tt(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{alpha:s}=n,i=e.makeTensorInfo([],"float32",x.createScalarValue(s,"float32")),a=V().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new ci(Lk,o.shape,i.shape):new Oo(Mk,o.shape,i.shape),u=e.runWebGLProgram(a,[o,i],"float32");return e.disposeIntermediateTensorInfo(i),u}var XM={kernelName:ms,backendName:"webgl",kernelFunc:_tt};var zk="return (a < 0.) ? b * a : a;",Bk=`
vec4 aLessThanZero = vec4(lessThan(a, vec4(0.)));
return (aLessThanZero * (b * a)) + ((vec4(1.0) - aLessThanZero) * a);
`;function Ett(r){let{inputs:t,backend:e}=r,{x:n,alpha:o}=t,s=V().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new ci(Bk,n.shape,o.shape):new Oo(zk,n.shape,o.shape);return e.runWebGLProgram(s,[n,o],"float32")}var YM={kernelName:ks,backendName:"webgl",kernelFunc:Ett};var Po="if (isnan(x)) return x;",ZM=`
if (isnan(a)) return a;
if (isnan(b)) return b;
`,JM=`
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 Ct({opSnippet:r,packedOpSnippet:t,cpuKernelImpl:e,dtype:n}){return({inputs:o,backend:s})=>{let{x:i}=o,a=s,u=n||i.dtype;if(a.shouldExecuteOnCPU([i])&&e!=null){let p=a.texData.get(i.dataId),m=e(p.values,u);return a.makeTensorInfo(i.shape,u,m)}let l=V().getBool("WEBGL_PACK_UNARY_OPERATIONS")&&t!=null,c;return l?c=new Fo(i.shape,t):c=new vn(i.shape,r),a.runWebGLProgram(c,[i],u)}}function le({opSnippet:r,packedOpSnippet:t,checkOutOfBounds:e=!1,supportsComplex:n=!1,cpuKernelImpl:o,dtype:s}){return({inputs:i,backend:a})=>{let{a:u,b:l}=i,c=a;if(n&&u.dtype==="complex64"){let d=c.texData.get(u.dataId),h=c.texData.get(l.dataId),[g,y]=[[d.complexTensorInfos.real,h.complexTensorInfos.real],[d.complexTensorInfos.imag,h.complexTensorInfos.imag]].map(w=>{let[v,k]=w,E={dataId:v.dataId,dtype:v.dtype,shape:u.shape},D={dataId:k.dataId,dtype:k.dtype,shape:l.shape},$=new Oo(r,u.shape,l.shape);return c.runWebGLProgram($,[E,D],ar(v.dtype,k.dtype))}),b=Dn({inputs:{real:g,imag:y},backend:c});return c.disposeIntermediateTensorInfo(g),c.disposeIntermediateTensorInfo(y),b}let p=s||ar(u.dtype,l.dtype);if((u.dtype==="string"||l.dtype==="string"||c.shouldExecuteOnCPU([u,l]))&&o!=null){let d=c.texData.get(u.dataId).values,h=c.texData.get(l.dataId).values,g=u.dtype==="string"?S.fromUint8ToStringArray(d):d,y=u.dtype==="string"?S.fromUint8ToStringArray(h):h,[b,w]=o(u.shape,l.shape,g,y,p),v=c.makeTensorInfo(w,p),k=c.texData.get(v.dataId);return k.values=b,v}let m=V().getBool("WEBGL_PACK_BINARY_OPERATIONS")&&t!=null,f;return m?f=new ci(t,u.shape,l.shape,e):f=new Oo(r,u.shape,l.shape),c.runWebGLProgram(f,[u,l],p)}}function au(r,t=!1){if(r==="linear")return t?BM:FM;if(r==="relu")return t?GM:PM;if(r==="elu")return t?VM:OM;if(r==="relu6")return t?WM:MM;if(r==="prelu")return t?Bk:zk;if(r==="leakyrelu")return t?Lk:Mk;if(r==="sigmoid")return t?UM:LM;throw new Error(`Activation ${r} has not been implemented for the WebGL backend.`)}var cg=class{constructor(t,e,n,o=!1,s=!1,i=!1,a=null,u=!1,l=!1){this.variableNames=["matrixA","matrixB"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=n,this.enableShapeUniforms=$e(this.outputShape.length);let c=o?t[1]:t[2],p=Math.ceil(c/2),m=o?"i * 2, rc.y":"rc.y, i * 2",f=s?"rc.z, i * 2":"i * 2, rc.z",d=o?["a.xxyy","a.zzww"]:["a.xxzz","a.yyww"],h=s?["b.xzxz","b.ywyw"]:["b.xyxy","b.zwzw"],g="",y="";a&&(u?g=`vec4 activation(vec4 a) {
vec4 b = getPreluActivationWeightsAtOutCoords();
${a}
}`:l?g=`vec4 activation(vec4 a) {
vec4 b = getLeakyreluAlphaAtOutCoords();
${a}
}`:g=`vec4 activation(vec4 x) {
${a}
}`,y="result = activation(result);");let b=i?"result += getBiasAtOutCoords();":"";i&&this.variableNames.push("bias"),u&&this.variableNames.push("preluActivationWeights"),l&&this.variableNames.push("leakyreluAlpha");let w="rc.x",v="rc.x";t[0]<e[0]?w=`int(min(float(rc.x), ${t[0]-1}.))`:e[0]<t[0]&&(v=`int(min(float(rc.x), ${e[0]-1}.))`),this.userCode=`
${g}
// Don't use uniform for sharedDimensionPacked for performance.
const float sharedDimension = ${p}.0;
vec4 dot2x2ARowBCol(ivec3 rc) {
vec4 result = vec4(0);
for (int i = 0; i < ${p}; i++) {
int batchA = ${w};
int batchB = ${v};
vec4 a = getMatrixA(batchA, ${m});
vec4 b = getMatrixB(batchB, ${f});
// These swizzled products need to be separately added.
// See: https://github.com/tensorflow/tfjs/issues/1735
result += (${d[0]} * ${h[0]});
result += (${d[1]} * ${h[1]});
}
return result;
}
void main() {
ivec3 rc = getOutputCoords();
vec4 result = dot2x2ARowBCol(rc);
${b}
${y}
setOutput(result);
}
`}};var Vk={REAL:"return areal * breal - aimag * bimag;",IMAG:"return areal * bimag + aimag * breal;"},Nb=class{constructor(t,e,n){this.variableNames=["AReal","AImag","BReal","BImag"],this.outputShape=S.assertAndGetBroadcastShape(e,n),this.userCode=`
float binaryOpComplex(
float areal, float aimag, float breal, float bimag) {
${t}
}
void main() {
float areal = getARealAtOutCoords();
float aimag = getAImagAtOutCoords();
float breal = getBRealAtOutCoords();
float bimag = getBImagAtOutCoords();
setOutput(binaryOpComplex(areal, aimag, breal, bimag));
}
`}};var QM="return a * b;";function pg(r){let{inputs:t,backend:e}=r,{a:n,b:o}=t,s=S.upcastType(n.dtype,o.dtype);if(n.dtype==="complex64"){let a=e.texData.get(n.dataId),u=e.texData.get(o.dataId),l=new Nb(Vk.REAL,n.shape,o.shape),c=new Nb(Vk.IMAG,n.shape,o.shape),p=[{dataId:a.complexTensorInfos.real.dataId,dtype:a.complexTensorInfos.real.dtype,shape:n.shape},{dataId:a.complexTensorInfos.imag.dataId,dtype:a.complexTensorInfos.imag.dtype,shape:n.shape},{dataId:u.complexTensorInfos.real.dataId,dtype:u.complexTensorInfos.real.dtype,shape:o.shape},{dataId:u.complexTensorInfos.imag.dataId,dtype:u.complexTensorInfos.imag.dtype,shape:o.shape}],m=e.runWebGLProgram(l,p,"float32"),f=e.runWebGLProgram(c,p,"float32"),d=Dn({inputs:{real:m,imag:f},backend:e});return e.disposeIntermediateTensorInfo(m),e.disposeIntermediateTensorInfo(f),d}if(e.shouldExecuteOnCPU([n,o])){let a=e.texData.get(n.dataId),u=e.texData.get(o.dataId),[l,c]=pM(n.shape,o.shape,a.values,u.values,s),p=e.makeTensorInfo(c,s),m=e.texData.get(p.dataId);return m.values=l,p}let i;return V().getBool("WEBGL_PACK_BINARY_OPERATIONS")?i=new ci(QM,n.shape,o.shape):i=new Oo(QM,n.shape,o.shape),e.runWebGLProgram(i,[n,o],s)}var tL={kernelName:vs,backendName:"webgl",kernelFunc:pg};function eL(r,t,e){let n=[pl(r.shape),...ml(r.shape)],o={dtype:r.dtype,shape:n,dataId:r.dataId},s=[pl(t),...ml(t)],i=new ug(s,n),a=!0,u=[n],l=e.runWebGLProgram(i,[o],r.dtype,u,a);return{dataId:l.dataId,shape:t,dtype:l.dtype}}function ut(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{shape:s}=n,i=e,a=x.sizeFromShape(o.shape),u=x.inferFromImplicitShape(s,a),l=x.sizeFromShape(u);x.assert(a===l,()=>`The new shape (${u}) has ${l} elements and the old shape (${o.shape}) has ${a} elements. The new shape and old shape must have the same number of elements.`);let c=i.texData.get(o.dataId);return c.isPacked&&!su(o.shape,u)&&!(c.texture!==null&&su(c.shape,u))?eL(o,u,i):(i.incRef(o.dataId),{dataId:o.dataId,shape:u,dtype:o.dtype})}var rL={kernelName:Ii,backendName:"webgl",kernelFunc:ut};var Tb=class{constructor(t,e){this.variableNames=["x"];let{windowSize:n,batchSize:o,inSize:s,outSize:i}=t;this.outputShape=[o,i];let a=Math.floor(n/4)*4,u=n%4,l="sumValue += dot(values, ones);";if(e!=null){let p=1/e;l=`sumValue += dot(values * ${x.isInt(p)?p.toPrecision(2):p}, ones);`}let c="";s%n>0&&(c=`
if (inIdx < 0 || inIdx >= ${s}) {
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 < ${a}; 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 + ${a};
if (${u===1}) {
vec4 values = vec4(getValue(batch, inIdx), 0.0, 0.0, 0.0);
${l}
} else if (${u===2}) {
vec4 values = vec4(
getValue(batch, inIdx),
getValue(batch, inIdx + 1), 0.0, 0.0);
${l}
} else if (${u===3}) {
vec4 values = vec4(
getValue(batch, inIdx),
getValue(batch, inIdx + 1),
getValue(batch, inIdx + 2), 0.0);
${l}
}
setOutput(sumValue);
}
`}};var Gk=class{constructor(t,e){this.variableNames=["x"];let{windowSize:n,batchSize:o,inSize:s,outSize:i}=t;this.outputShape=[o,i];let a="0.0",u="";e==="prod"?a="1.0":e==="min"?(a="1.0 / 1e-20",u="min"):e==="max"&&(a="-1.0 / 1e-20",u="max");let l=`${e}(${e}(${e}(minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])`;e==="sum"?l="sumValue":e==="prod"?l="prodValue":e==="all"?l="allValue":e==="any"&&(l="anyValue");let c=Math.floor(n/4)*4,p=n%4,m=`
if (${e==="sum"}) {
sumValue += dot(values, ones);
} else if (${e==="prod"}) {
vec2 tmp = vec2(values[0], values[1]) * vec2(values[2], values[3]);
prodValue *= tmp[0] * tmp[1];
} else {
minMaxValue = ${u}(values, minMaxValue);
if (${e==="min"} || ${e==="max"}) {
minMaxValue = ${u}(values, minMaxValue);
bvec4 isNaN = isnan(values);
if (isNaN.r || isNaN.g || isNaN.b || isNaN.a) {
minMaxValue = vec4(NAN);
}
}
}
`,f="vec4";e==="all"?(a="1.0",m=`
bool reducedAllValue = all(values);
float floatedReducedAllValue = float(reducedAllValue);
allValue = float(allValue >= 1.0 && floatedReducedAllValue >= 1.0);
`,f="bvec4"):e==="any"&&(a="0.0",m=`
bool reducedAnyValue = any(values);
float floatedReducedAnyValue = float(reducedAnyValue);
anyValue = float(anyValue >= 1.0 || floatedReducedAnyValue >= 1.0);
`,f="bvec4");let d="";s%n>0&&(d=`
if (inIdx < 0 || inIdx >= ${s}) {
return initializationValue;
}
`),this.userCode=`
const float initializationValue = ${a};
const vec4 ones = vec4(1.0, 1.0, 1.0, 1.0);
float getValue(int batch, int inIdx) {
${d}
return getX(batch, inIdx);
}
void main() {
ivec2 coords = getOutputCoords();
int batch = coords[0];
int outIdx = coords[1];
int inOffset = outIdx * ${n};
vec4 minMaxValue = vec4(${a});
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;
${f} values = ${f}(
getValue(batch, inIdx),
getValue(batch, inIdx + 1),
getValue(batch, inIdx + 2),
getValue(batch, inIdx + 3)
);
${m}
}
int inIdx = inOffset + ${c};
if (${p===1}) {
${f} values = ${f}(
getValue(batch, inIdx),
initializationValue,
initializationValue,
initializationValue
);
${m}
} else if (${p===2}) {
${f} values = ${f}(
getValue(batch, inIdx),
getValue(batch, inIdx + 1),
initializationValue,
initializationValue
);
${m}
} else if (${p===3}) {
${f} values = ${f}(
getValue(batch, inIdx),
getValue(batch, inIdx + 1),
getValue(batch, inIdx + 2),
initializationValue
);
${m}
}
setOutput(${l});
}
`}};function Dtt(r){let t=[];for(;t.length===0||t[t.length-1].outSize!==1;){let e=t.length?t[t.length-1].outSize:r[1],n=S.computeOptimalWindowSize(e);t.push({inSize:e,windowSize:n,outSize:Math.ceil(e/n)})}return t}function Wn(r,t,e,n){let o=Dtt(r.shape),s=r;for(let i=0;i<o.length;i++){let{inSize:a,windowSize:u,outSize:l}=o[i],c,p;e==="mean"?c=i===0?new Tb({windowSize:u,inSize:a,batchSize:r.shape[0],outSize:l},a):new Tb({windowSize:u,inSize:a,batchSize:r.shape[0],outSize:l}):c=new Gk({windowSize:u,inSize:a,batchSize:r.shape[0],outSize:l},e),p=s,s=n.runWebGLProgram(c,[s],t),p.dataId!==r.dataId&&n.disposeIntermediateTensorInfo(p)}return s}var Wk=class{constructor(t,e){this.variableNames=["A"];let n=new Array(t.length);for(let i=0;i<n.length;i++)n[i]=t[e[i]];this.outputShape=n,this.rank=n.length;let o=zt(this.rank),s=$tt(e);this.userCode=`
void main() {
${o} resRC = getOutputCoords();
setOutput(getA(${s}));
}
`}};function $tt(r){let t=r.length;if(t>6)throw Error(`Transpose for rank ${t} is not yet supported`);let e=["resRC.x","resRC.y","resRC.z","resRC.w","resRC.u","resRC.v"],n=new Array(t);for(let o=0;o<r.length;o++)n[r[o]]=e[o];return n.join()}var Uk=class{constructor(t,e){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0;let n=new Array(t.length);for(let c=0;c<n.length;c++)n[c]=t[e[c]];if(this.outputShape=n,this.rank=n.length,this.rank>6)throw Error(`Packed transpose for rank ${this.rank} is not yet supported.`);let o=zt(this.rank),s=$k("rc",this.rank),i=new Array(this.rank);for(let c=0;c<e.length;c++)i[e[c]]=s[c];let a=`vec2(${i.slice(-2).join()})`,u=`++${s[this.rank-1]} < ${n[this.rank-1]}`,l=`getChannel(getA(${i.join()}), ${a})`;this.userCode=`
void main() {
${o} rc = getOutputCoords();
vec4 result = vec4(0.);
result[0] = ${l};
if(${u}) {
result[1] = ${l};
}
--${s[this.rank-1]};
if(++${s[this.rank-2]} < ${n[this.rank-2]}) {
result[2] = ${l};
if(${u}) {
result[3] = ${l};
}
}
setOutput(result);
}
`}};function lu(r,t,e){let n=V().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new Uk(r.shape,t):new Wk(r.shape,t);return e.runWebGLProgram(n,[r],r.dtype)}function nL(r,t,e,n){let o=t,s=r.shape.length,i=x.parseAxisParam(o,r.shape),a=i,u=S.getAxesPermutation(a,s),l=u!=null,c=r;l&&(c=lu(r,u,n),a=S.getInnerMostAxes(a.length,s)),S.assertAxesAreInnerMostDims("sum",a,s);let[p,m]=S.computeOutAndReduceShapes(c.shape,a),f=p;e&&(f=S.expandShapeToKeepDim(p,i));let d=x.sizeFromShape(m),g=x.sizeFromShape(r.shape)/d,y=ut({inputs:{x:c},attrs:{shape:[g,d]},backend:n}),b=_u(r.dtype),w=Wn(y,b,"sum",n),v=ut({inputs:{x:w},attrs:{shape:f},backend:n});return n.disposeIntermediateTensorInfo(y),n.disposeIntermediateTensorInfo(w),l&&n.disposeIntermediateTensorInfo(c),v}function Pc(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s,keepDims:i}=n;return nL(o,s,i,e)}var oL={kernelName:Os,backendName:"webgl",kernelFunc:Pc};function Ee(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{perm:s}=n,i=e,a=o.shape.length,u=new Array(a);for(let c=0;c<u.length;c++)u[c]=o.shape[s[c]];let l;if(i.shouldExecuteOnCPU([o])){let p=i.texData.get(o.dataId).values,m=Rc(p,o.shape,o.dtype,s,u);l=i.makeTensorInfo(u,o.dtype);let f=i.texData.get(l.dataId);f.values=m}else l=lu(o,s,i);return l}var sL={kernelName:Vs,backendName:"webgl",kernelFunc:Ee};var Hk=1e3;function Mc({a:r,b:t,transposeA:e,transposeB:n,backend:o,bias:s=null,preluActivationWeights:i=null,leakyreluAlpha:a=0,activation:u=null}){let l=r.shape.length,c=t.shape.length,p=e?r.shape[l-2]:r.shape[l-1],m=n?t.shape[c-1]:t.shape[c-2],f=e?r.shape[l-1]:r.shape[l-2],d=n?t.shape[c-2]:t.shape[c-1],h=r.shape.slice(0,-2),g=t.shape.slice(0,-2),y=x.sizeFromShape(h),b=x.sizeFromShape(g),v=Mr.assertAndGetBroadcastShape(r.shape.slice(0,-2),t.shape.slice(0,-2)).concat([f,d]);x.assert(p===m,()=>`Error in matMul: inner shapes (${p}) and (${m}) of Tensors with shapes ${r.shape} and ${t.shape} and transposeA=${e} and transposeB=${n} must match.`);let k=e?[y,p,f]:[y,f,p],E=n?[b,d,m]:[b,m,d],D=ut({inputs:{x:r},backend:o,attrs:{shape:k}}),$=ut({inputs:{x:t},backend:o,attrs:{shape:E}}),R=[D,$],M=Math.max(y,b),W=e?D.shape[1]:D.shape[2],U=s!=null,q=i!=null,K=u==="leakyrelu",j=u!=null?au(u,!0):null,Q=U||q||K||j!=null,rt;if((f===1||d===1)&&W>Hk&&Q===!1){let ot=D,st=$;e&&(ot=Ee({inputs:{x:D},backend:o,attrs:{perm:[0,2,1]}}),R.push(ot)),n&&(st=Ee({inputs:{x:$},backend:o,attrs:{perm:[0,2,1]}}),R.push(st));let it=d!==1,ft=d===1,at=ot;it&&(at=ut({inputs:{x:ot},backend:o,attrs:{shape:[M,W,1]}}),R.push(at));let xt=d===1?2:1,dt=st;ft&&(dt=ut({inputs:{x:st},backend:o,attrs:{shape:[M,1,W]}}),R.push(dt));let bt=pg({inputs:{a:at,b:dt},backend:o});rt=Pc({inputs:{x:bt},backend:o,attrs:{axis:xt,keepDims:!0}}),R.push(bt)}else{let ot=ar(r.dtype,t.dtype),st=new cg(k,E,[M,f,d],e,n,U,j,q,K),it=[D,$];if(s!=null&&it.push(s),q&&it.push(i),K){let ft=o.makeTensorInfo([],"float32",x.createScalarValue(a,"float32"));it.push(ft),R.push(ft)}rt=o.runWebGLProgram(st,it,ot)}let X=ut({inputs:{x:rt},backend:o,attrs:{shape:v}});R.push(rt);for(let ot of R)o.disposeIntermediateTensorInfo(ot);return X}function Rtt(r){let{inputs:t,backend:e,attrs:n}=r,{a:o,b:s,bias:i,preluActivationWeights:a}=t,{transposeA:u,transposeB:l,activation:c,leakyreluAlpha:p}=n;return Mc({a:o,b:s,transposeA:u,transposeB:l,backend:e,bias:i,preluActivationWeights:a,leakyreluAlpha:p,activation:c})}var iL={kernelName:Ai,backendName:"webgl",kernelFunc:Rtt};var aL="return abs(x);";function Ftt(r){let{inputs:t,backend:e}=r,{x:n}=t;if(e.shouldExecuteOnCPU([n])&&n.dtype!=="complex64"){let s=e.texData.get(n.dataId),i=Cb(s.values);return e.makeTensorInfo(n.shape,n.dtype,i)}let o;return V().getBool("WEBGL_PACK_UNARY_OPERATIONS")?o=new Fo(n.shape,aL):o=new vn(n.shape,aL),e.runWebGLProgram(o,[n],n.dtype)}var lL={kernelName:hi,backendName:"webgl",kernelFunc:Ftt};var Ott=hr+`
if (abs(x) > 1.) {
return NAN;
}
return acos(x);
`,Ptt=Ct({opSnippet:Ott}),uL={kernelName:ji,backendName:"webgl",kernelFunc:Ptt};var Mtt=hr+`
if (x < 1.0) return NAN;
return log(x + sqrt(x * x - 1.0));`,Ltt=Ct({opSnippet:Mtt}),cL={kernelName:Xi,backendName:"webgl",kernelFunc:Ltt};var pL="return a + b;",ztt=le({opSnippet:pL,packedOpSnippet:pL,supportsComplex:!0,cpuKernelImpl:qP}),mL={kernelName:Kn,backendName:"webgl",kernelFunc:ztt};var qk=class{constructor(t,e){this.outputShape=[],this.outputShape=t,this.variableNames=e.map((s,i)=>`T${i}`);let n=[];this.variableNames.forEach(s=>{n.push(`float v${s} = get${s}AtOutCoords();`)});let o=this.variableNames.map(s=>`v${s}`).join(" + ");this.userCode=`
void main() {
${n.join(`
`)}
float result = ${o};
setOutput(result);
}
`}};var Kk=class{constructor(t,e){this.outputShape=[],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t,this.variableNames=e.map((s,i)=>`T${i}`);let n=[];this.variableNames.forEach(s=>{n.push(`vec4 v${s} = get${s}AtOutCoords();`)});let o=this.variableNames.map(s=>`v${s}`).join(" + ");this.userCode=`
void main() {
${n.join(`
`)}
vec4 result = ${o};
setOutput(result);
}
`}};function _b(r){let{inputs:t,backend:e}=r,n=t;if(n.length===1)return Ke({inputs:{x:n[0]},backend:e});if(n.length>V().get("WEBGL_MAX_TEXTURES_IN_SHADER")){let u=Math.floor(n.length/2),l=_b({inputs:n.slice(0,u),backend:e}),c=_b({inputs:n.slice(u),backend:e});return _b({inputs:[l,c],backend:e})}let o=n.map(u=>u.dtype).reduce((u,l)=>ar(u,l)),s=n.map(u=>u.shape),a=V().getBool("WEBGL_PACK")?new Kk(n[0].shape,s):new qk(n[0].shape,s);return e.runWebGLProgram(a,n,o)}var fL={kernelName:jo,backendName:"webgl",kernelFunc:_b};function Btt(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s,keepDims:i}=n,a=o.shape.length,u=x.parseAxisParam(s,o.shape),l=u,c=S.getAxesPermutation(l,a),p=o;c!=null&&(p=Ee({inputs:{x:o},backend:e,attrs:{perm:c}}),l=S.getInnerMostAxes(l.length,a)),S.assertAxesAreInnerMostDims("all",l,a);let[m,f]=S.computeOutAndReduceShapes(p.shape,l),d=x.sizeFromShape(f),h=ut({inputs:{x:p},backend:e,attrs:{shape:[-1,d]}}),g=Wn(h,h.dtype,"all",e),y;if(i){let b=S.expandShapeToKeepDim(m,u);y=ut({inputs:{x:g},backend:e,attrs:{shape:b}})}else y=ut({inputs:{x:g},backend:e,attrs:{shape:m}});return e.disposeIntermediateTensorInfo(h),e.disposeIntermediateTensorInfo(g),c!=null&&e.disposeIntermediateTensorInfo(p),y}var dL={kernelName:Yi,backendName:"webgl",kernelFunc:Btt};function Vtt(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s,keepDims:i}=n,a=o.shape.length,u=x.parseAxisParam(s,o.shape),l=u,c=S.getAxesPermutation(l,a),p=o;c!=null&&(p=Ee({inputs:{x:o},backend:e,attrs:{perm:c}}),l=S.getInnerMostAxes(l.length,a)),S.assertAxesAreInnerMostDims("any",l,a);let[m,f]=S.computeOutAndReduceShapes(p.shape,l),d=x.sizeFromShape(f),h=ut({inputs:{x:p},backend:e,attrs:{shape:[-1,d]}}),g=Wn(h,h.dtype,"any",e),y;if(i){let b=S.expandShapeToKeepDim(m,u);y=ut({inputs:{x:g},backend:e,attrs:{shape:b}})}else y=ut({inputs:{x:g},backend:e,attrs:{shape:m}});return e.disposeIntermediateTensorInfo(h),e.disposeIntermediateTensorInfo(g),c!=null&&e.disposeIntermediateTensorInfo(p),y}var hL={kernelName:Zi,backendName:"webgl",kernelFunc:Vtt};var jk=class{constructor(t,e,n){this.variableNames=["A"];let{windowSize:o,batchSize:s,outSize:i}=t;n||this.variableNames.push("bestIndicesA"),this.outputShape=[s,i];let a=e==="max"?">":"<",u=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 * ${o};
int bestIndex = inOffset;
float bestValue = getA(batch, bestIndex);
for (int i = 0; i < ${o}; i++) {
int inIdx = ${u};
float candidate = getA(batch, inIdx);
if (candidate ${a} bestValue) {
bestValue = candidate;
bestIndex = inIdx;
}
}
setOutput(float(bestIndex));
}
`}};var Xk=class{constructor(t,e,n,o){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,x.assert(t.length>2,()=>`Packed arg${n.charAt(0).toUpperCase()+n.slice(1)} supports only inputs with rank above 2.`);let s=t[t.length-1],i=Math.ceil(s/e);this.outputShape=t.slice(0,-1),i>1&&this.outputShape.push(i),o||this.variableNames.push("bestIndicesA");let a=this.outputShape,u=a.length,l=zt(u),c=rr("coords",u),p,m;if(i===1){m=u+1;let $=zt(m);p=`
${$} sourceLocR = ${$}(${c.join()}, 0);
++${c[u-1]};
${$} sourceLocG = ${$}(${c.join()}, 0);
++${c[u-2]};
${$} sourceLocA = ${$}(${c.join()}, 0);
--${c[u-1]};
${$} sourceLocB = ${$}(${c.join()}, 0);
--${c[u-2]};`}else m=u,p=`
${l} sourceLocR = coords;
++${c[u-1]};
${l} sourceLocG = coords;
++${c[u-2]};
${l} sourceLocA = coords;
--${c[u-1]};
${l} sourceLocB = coords;
--${c[u-2]};`;let f=["x","y","z","w","u","v"].slice(0,m),d="."+f[m-1],h=f.map($=>"int "+$),g=rr("sourceLocR",m-1).concat("inIdx.r"),y=rr("sourceLocG",m-1).concat("inIdx.g"),b=rr("sourceLocB",m-1).concat("inIdx.b"),w=rr("sourceLocA",m-1).concat("inIdx.a"),v=n==="max"?"greaterThan":"lessThan",k=o?"":`
inIdx = round(vec4(getBestIndicesAChannel(${g.join()}),
getBestIndicesAChannel(${y.join()}),
getBestIndicesAChannel(${b.join()}),
getBestIndicesAChannel(${w.join()})));`,E=`vec4(
getAChannel(${g.join()}),
hasNextCol ? getAChannel(${y.join()}) : 0.,
hasNextRow ? getAChannel(${b.join()}) : 0.,
hasNextRow && hasNextCol ? getAChannel(${w.join()}) : 0.)`,D=o?"":`
float getBestIndicesAChannel(${h.join()}) {
return getChannel(getBestIndicesA(${f.join()}),
vec2(${f.slice(-2).join()}));
}`;this.userCode=`
float getAChannel(${h.join()}) {
return getChannel(getA(${f.join()}),
vec2(${f.slice(-2).join()}));
}
${D}
void main() {
${l} coords = getOutputCoords();
bool hasNextCol = ${c[u-1]} < ${a[u-1]-1};
bool hasNextRow = ${c[u-2]} < ${a[u-2]-1};
${p}
ivec4 srcIdx = ivec4(sourceLocR${d}, sourceLocG${d},
sourceLocB${d}, sourceLocA${d}) * ${e};
ivec4 inIdx = srcIdx;
vec4 bestIndex = vec4(inIdx);
vec4 bestValue = ${E};
for (int i = 0; i < ${e}; i++) {
inIdx = srcIdx;
${k}
vec4 candidate = ${E};
bvec4 nan = isnan(candidate);
bvec4 replace = bvec4(
vec4(${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 gL(r,t,e,n=null){let o=t.shape[0],s=t.shape[1];n!=null&&(o=n.shape[0],s=n.shape[1]);let i=S.computeOptimalWindowSize(s),a={windowSize:i,inSize:s,batchSize:o,outSize:Math.ceil(s/i)},u=new jk(a,e,n==null),l=[t];n!=null&&l.push(n);let c=r.runWebGLProgram(u,l,"int32");if(c.shape[1]===1)return c;let p=gL(r,t,e,c);return r.disposeIntermediateTensorInfo(c),p}function xL(r,t,e,n=null){let o=n!=null?n.shape:t.shape,s=o[o.length-1],i=S.computeOptimalWindowSize(s),a=new Xk(o,i,e,n==null),u=n==null?[t]:[t,n],l=r.runWebGLProgram(a,u,"int32");if(l.shape.length===t.shape.length){let c=xL(r,t,e,l);return r.disposeIntermediateTensorInfo(l),c}return l}function Eb(r,t,e,n){let o=[e];if(S.assertAxesAreInnerMostDims("arg"+n.charAt(0).toUpperCase()+n.slice(1),o,t.shape.length),!V().getBool("WEBGL_PACK_REDUCE")||t.shape.length<=2){let s=[],i=r.texData.get(t.dataId),a=i!==null&&i.isPacked,u=t;a&&(u=r.unpackTensor(t),s.push(u));let[l,c]=S.computeOutAndReduceShapes(u.shape,o),p=x.sizeFromShape(c),m=ut({inputs:{x:u},backend:r,attrs:{shape:[-1,p]}});s.push(m);let f=gL(r,m,n);s.push(f);let d=ut({inputs:{x:f},backend:r,attrs:{shape:l}});return s.forEach(h=>r.disposeIntermediateTensorInfo(h)),d}return xL(r,t,n)}function Gtt(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s}=n,i=x.parseAxisParam(s,o.shape),a=S.getAxesPermutation(i,o.shape.length),u=o,l=[];a!=null&&(u=Ee({inputs:{x:o},backend:e,attrs:{perm:a}}),l.push(u),i=S.getInnerMostAxes(i.length,u.shape.length)),S.assertAxesAreInnerMostDims("argMax",[i[0]],u.shape.length);let c=Eb(e,u,i[0],"max");return l.forEach(p=>e.disposeIntermediateTensorInfo(p)),c}var yL={kernelName:Xo,backendName:"webgl",kernelFunc:Gtt};function Wtt(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s}=n,i=x.parseAxisParam(s,o.shape),a=S.getAxesPermutation(i,o.shape.length),u=o,l=[];a!=null&&(u=Ee({inputs:{x:o},backend:e,attrs:{perm:a}}),l.push(u),i=S.getInnerMostAxes(i.length,u.shape.length)),S.assertAxesAreInnerMostDims("argMin",[i[0]],u.shape.length);let c=Eb(e,u,i[0],"min");return l.forEach(p=>e.disposeIntermediateTensorInfo(p)),c}var bL={kernelName:vl,backendName:"webgl",kernelFunc:Wtt};var Utt=hr+`
if (abs(x) > 1.) {
return NAN;
}
return asin(x);
`,Htt=Ct({opSnippet:Utt}),wL={kernelName:Ji,backendName:"webgl",kernelFunc:Htt};var qtt=hr+"return log(x + sqrt(x * x + 1.0));",Ktt=Ct({opSnippet:qtt}),vL={kernelName:Qi,backendName:"webgl",kernelFunc:Ktt};var jtt=hr+`
return atan(x);
`,Xtt=Ct({opSnippet:jtt}),CL={kernelName:ta,backendName:"webgl",kernelFunc:Xtt};var Ytt=ZM+`
return atan(a, b);
`,Ztt=`
vec4 result = atan(a, b);
vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));
`+JM+`
return result;
`,Jtt=le({opSnippet:Ytt,packedOpSnippet:Ztt}),IL={kernelName:ra,backendName:"webgl",kernelFunc:Jtt};var Qtt=hr+`
if ((x < -1.0) || (x > 1.0)) return NAN;
return (log(1.0 + x) - log(1.0 - x)) / 2.0;`,tet=Ct({opSnippet:Qtt}),SL={kernelName:ea,backendName:"webgl",kernelFunc:tet};var qi=class{constructor(t,e,n,o=!1,s=!1){if(this.variableNames=["x"],e==="avg"&&n)throw new Error("Cannot compute positions for average pool.");let i=t.filterWidth,a=t.strideHeight,u=t.strideWidth,l=t.dilationHeight,c=t.dilationWidth,p=t.effectiveFilterHeight,m=t.effectiveFilterWidth,f=t.padInfo.top,d=t.padInfo.left;this.outputShape=t.outShape;let h=e==="avg",g=`((batch * ${t.inHeight} + xR) * ${t.inWidth} + xC) * ${t.inChannels} + d`,y=`(xR * ${t.inWidth} + xC) * ${t.inChannels} + d`,b="0.0";if(h||(b="-1.0 / 1e-20"),n){let $=">=";this.userCode=`
const ivec2 strides = ivec2(${a}, ${u});
const ivec2 pads = ivec2(${f}, ${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
float minMaxValue = 0.0;
float minMaxValueFound = 0.0;
int minMaxPosition = 0;
float avgValue = 0.0;
for (int wR = 0; wR < ${p};
wR += ${l}) {
int xR = xRCorner + wR;
if (xR < 0 || xR >= ${t.inHeight}) {
continue;
}
for (int wC = 0; wC < ${m};
wC += ${c}) {
int xC = xCCorner + wC;
if (xC < 0 || xC >= ${t.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 ${$} currMinMaxValue) {
minMaxValue = value;
minMaxValueFound = 1.0;
minMaxPosition = ${o?s?g:y:`wR * ${m} + wC`};
}
}
}
setOutput(float(minMaxPosition));
}
`;return}let w="max",v=`${e}(${e}(${e}(minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])`;e==="avg"&&(v="avgValue / count");let k=Math.floor(i/4)*4,E=i%4,D=`
if (${h}) {
avgValue += dot(values, ones);
} else {
minMaxValue = ${w}(values, minMaxValue);
}
`;this.userCode=`
const ivec2 strides = ivec2(${a}, ${u});
const ivec2 pads = ivec2(${f}, ${d});
const float initializationValue = ${b};
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 >= ${t.inWidth}) {
return initializationValue;
}
count += 1.0;
return getX(batch, xR, xC, d);
}
void main() {
ivec4 coords = getOutputCoords();
int batch = coords[0];
int d = coords[3];
ivec2 xRCCorner = coords.yz * strides - pads;
int xRCorner = xRCCorner.x;
int xCCorner = xRCCorner.y;
// max/min x(?, ?, d) to get y(yR, yC, d).
// ? = to be determined
vec4 minMaxValue = vec4(${b});
float avgValue = 0.0;
count = 0.0;
for (int wR = 0; wR < ${p};
wR += ${l}) {
int xR = xRCorner + wR;
if (xR < 0 || xR >= ${t.inHeight}) {
continue;
}
for (int wC = 0; wC < ${k}; 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)
);
${D}
}
int xC = xCCorner + ${k};
if (${E===1}) {
vec4 values = vec4(
getValue(batch, xR, xC, d),
initializationValue,
initializationValue,
initializationValue
);
${D}
} else if (${E===2}) {
vec4 values = vec4(
getValue(batch, xR, xC, d),
getValue(batch, xR, xC + ${c}, d),
initializationValue,
initializationValue
);
${D}
} else if (${E===3}) {
vec4 values = vec4(
getValue(batch, xR, xC, d),
getValue(batch, xR, xC + ${c}, d),
getValue(batch, xR, xC + 2 * ${c}, d),
initializationValue
);
${D}
}
}
setOutput(${v});
}
`}},Lc=class{constructor(t,e,n,o=!1,s=!1){if(this.variableNames=["x"],e==="avg"&&n)throw new Error("Cannot compute positions for average pool.");let i=t.filterWidth,a=t.strideDepth,u=t.strideHeight,l=t.strideWidth,c=t.dilationDepth,p=t.dilationHeight,m=t.dilationWidth,f=t.effectiveFilterDepth,d=t.effectiveFilterHeight,h=t.effectiveFilterWidth,g=t.padInfo.front,y=t.padInfo.top,b=t.padInfo.left;this.outputShape=t.outShape;let w=e==="avg",v="0.0";if(w||(v="-1.0 / 1e-20"),n){let M=">=";this.userCode=`
const ivec3 strides =
ivec3(${a}, ${u}, ${l});
const ivec3 pads = ivec3(${g}, ${y}, ${b});
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 < ${f};
wD += ${c}) {
int xD = xDCorner + wD;
if (xD < 0 || xD >= ${t.inDepth}) {
continue;
}
for (int wR = 0; wR < ${d};
wR += ${p}) {
int xR = xRCorner + wR;
if (xR < 0 || xR >= ${t.inHeight}) {
continue;
}
for (int wC = 0; wC < ${h};
wC += ${m}) {
int xC = xCCorner + wC;
if (xC < 0 || xC >= ${t.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 ${M} currMinMaxValue) {
minMaxValue = value;
minMaxValueFound = 1.0;
minMaxPosition = ${o?s?`(((batch * ${t.inDepth} + xD) * ${t.inHeight} + xR) * ${t.inWidth} + xC) * ${t.inChannels} + ch`:`((xD * ${t.inHeight} + xR) * ${t.inWidth} + xC) * ${t.inChannels} + ch`:`wD * ${d} * ${h} +
wR * ${h} + wC`};
}
}
}
}
setOutput(float(minMaxPosition));
}
`;return}let k="max",E=`${e}(${e}(${e}(minMaxValue[0], minMaxValue[1]), minMaxValue[2]), minMaxValue[3])`;e==="avg"&&(E="avgValue / count");let D=Math.floor(i/4)*4,$=i%4,R=`
if (${w}) {
avgValue += dot(values, ones);
} else {
minMaxValue = ${k}(values, minMaxValue);
}
`;this.userCode=`
const ivec3 strides =
ivec3(${a}, ${u}, ${l});
const ivec3 pads = ivec3(${g}, ${y}, ${b});
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 >= ${t.inWidth}) {
return initializationValue;
}
count += 1.0;
return getX(batch, xD, xR, xC, ch);
}
void main() {
ivec5 coords = getOutputCoords();
int batch = coords.x;
int ch = coords.u;
ivec3 xCorner = ivec3(coords.y, coords.z, coords.w) * strides - pads;
int xDCorner = xCorner.x;
int xRCorner = xCorner.y;
int xCCorner = xCorner.z;
// max/min x(?, ?, ?, d) to get y(yD, yR, yC, ch).
// ? = to be determined
vec4 minMaxValue = vec4(${v});
float avgValue = 0.0;
count = 0.0;
for (int wD = 0; wD < ${f};
wD += ${c}) {
int xD = xDCorner + wD;
if (xD < 0 || xD >= ${t.inDepth}) {
continue;
}
for (int wR = 0; wR < ${d};
wR += ${p}) {
int xR = xRCorner + wR;
if (xR < 0 || xR >= ${t.inHeight}) {
continue;
}
for (int wC = 0; wC < ${D}; wC += 4) {
int xC = xCCorner + wC * ${m};
vec4 values = vec4(
getValue(batch, xD, xR, xC, ch),
getValue(batch, xD, xR, xC + ${m}, ch),
getValue(batch, xD, xR, xC + 2 * ${m}, ch),
getValue(batch, xD, xR, xC + 3 * ${m}, ch)
);
${R}
}
int xC = xCCorner + ${D};
if (${$===1}) {
vec4 values = vec4(
getValue(batch, xD, xR, xC, ch),
initializationValue,
initializationValue,
initializationValue
);
${R}
} else if (${$===2}) {
vec4 values = vec4(
getValue(batch, xD, xR, xC, ch),
getValue(batch, xD, xR, xC + ${m}, ch),
initializationValue,
initializationValue
);
${R}
} else if (${$===3}) {
vec4 values = vec4(
getValue(batch, xD, xR, xC, ch),
getValue(batch, xD, xR, xC + ${m}, ch),
getValue(batch, xD, xR, xC + 2 * ${m}, ch),
initializationValue
);
${R}
}
}
setOutput(${E});
}
}
`}};function eet(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t;li(o,"avgPool");let{filterSize:s,strides:i,pad:a,dimRoundingMode:u}=n,l=1;x.assert(S.eitherStridesOrDilationsAreOne(i,l),()=>`Error in avgPool: Either strides or dilations must be 1. Got strides ${i} and dilations '${l}'`);let c=S.computePool2DInfo(o.shape,s,i,l,a,u);if(c.filterWidth===1&&c.filterHeight===1&&x.arraysEqual(c.inShape,c.outShape))return Ke({inputs:{x:o},backend:e});let p=new qi(c,"avg",!1);return e.runWebGLProgram(p,[o],"float32")}var kL={kernelName:Yo,backendName:"webgl",kernelFunc:eet};function ret(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{filterSize:s,strides:i,pad:a,dimRoundingMode:u,dataFormat:l}=n,c=[1,1,1],p=S.computePool3DInfo(o.shape,s,i,c,a,u,l),m=new Lc(p,"avg",!1);return e.runWebGLProgram(m,[o],"float32")}var NL={kernelName:Cl,backendName:"webgl",kernelFunc:ret};var Yk=class{constructor(t){this.variableNames=["dy"],this.outputShape=t.inShape;let e=t.filterHeight,n=t.filterWidth,o=t.strideHeight,s=t.strideWidth,i=t.dilationHeight,a=t.dilationWidth,u=t.effectiveFilterHeight,l=t.effectiveFilterWidth,c=u-1-t.padInfo.top,p=l-1-t.padInfo.left,m=1/(e*n);this.userCode=`
const ivec2 pads = ivec2(${c}, ${p});
const float avgMultiplier = float(${m});
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 < ${u};
wR += ${i}) {
float dyR = float(dyRCorner + wR) / ${o}.0;
if (dyR < 0.0 || dyR >= ${t.outHeight}.0 || fract(dyR) > 0.0) {
continue;
}
int idyR = int(dyR);
for (int wC = 0; wC < ${l};
wC+= ${a}) {
float dyC = float(dyCCorner + wC) / ${s}.0;
if (dyC < 0.0 || dyC >= ${t.outWidth}.0 ||
fract(dyC) > 0.0) {
continue;
}
int idyC = int(dyC);
float dyValue = getDy(b, idyR, idyC, d);
dotProd += dyValue * avgMultiplier;
}
}
setOutput(dotProd);
}
`}},Zk=class{constructor(t){this.variableNames=["dy"],this.outputShape=t.inShape;let e=t.filterDepth,n=t.filterHeight,o=t.filterWidth,s=t.strideDepth,i=t.strideHeight,a=t.strideWidth,u=t.dilationDepth,l=t.dilationHeight,c=t.dilationWidth,p=t.effectiveFilterDepth,m=t.effectiveFilterHeight,f=t.effectiveFilterWidth,d=p-1-t.padInfo.front,h=m-1-t.padInfo.top,g=f-1-t.padInfo.left,y=1/(e*n*o);this.userCode=`
const ivec3 pads = ivec3(${d}, ${h}, ${g});
const float avgMultiplier = float(${y});
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 < ${p};
wD += ${u}) {
float dyD = float(dyDCorner + wD) / ${s}.0;
if (dyD < 0.0 || dyD >= ${t.outDepth}.0 || fract(dyD) > 0.0) {
continue;
}
int idyD = int(dyD);
for (int wR = 0; wR < ${m};
wR += ${l}) {
float dyR = float(dyRCorner + wR) / ${i}.0;
if (dyR < 0.0 || dyR >= ${t.outHeight}.0 ||
fract(dyR) > 0.0) {
continue;
}
int idyR = int(dyR);
for (int wC = 0; wC < ${f};
wC += ${c}) {
float dyC = float(dyCCorner + wC) / ${a}.0;
if (dyC < 0.0 || dyC >= ${t.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 net(r){let{inputs:t,backend:e,attrs:n}=r,{dy:o,input:s}=t,i=s,{filterSize:a,strides:u,pad:l,dimRoundingMode:c}=n,p=[1,1,1],m=S.computePool3DInfo(i.shape,a,u,p,l,c),f=new Zk(m);return e.runWebGLProgram(f,[o],i.dtype)}var TL={kernelName:up,backendName:"webgl",kernelFunc:net};function oet(r){let{inputs:t,backend:e,attrs:n}=r,{dy:o,input:s}=t,i=s;li([o,s],"avgPoolGrad");let{filterSize:a,strides:u,pad:l}=n,c=S.computePool2DInfo(i.shape,a,u,1,l),p=new Yk(c);return e.runWebGLProgram(p,[o],i.dtype)}var _L={kernelName:lp,backendName:"webgl",kernelFunc:oet};function set(r){let{inputs:t,backend:e,attrs:n}=r,{a:o,b:s}=t,{transposeA:i,transposeB:a}=n;return Mc({a:o,b:s,transposeA:i,transposeB:a,backend:e})}var EL={kernelName:Zo,backendName:"webgl",kernelFunc:set};var Jk=class{constructor(t,e,n,o,s,i){this.outputShape=[],this.variableNames=["x","mean","variance"],S.assertAndGetBroadcastShape(t,e),S.assertAndGetBroadcastShape(t,n);let a="0.0";o!=null&&(S.assertAndGetBroadcastShape(t,o),this.variableNames.push("offset"),a="getOffsetAtOutCoords()");let u="1.0";s!=null&&(S.assertAndGetBroadcastShape(t,s),this.variableNames.push("scale"),u="getScaleAtOutCoords()"),this.outputShape=t,this.userCode=`
void main() {
float x = getXAtOutCoords();
float mean = getMeanAtOutCoords();
float variance = getVarianceAtOutCoords();
float offset = ${a};
float scale = ${u};
float inv = scale * inversesqrt(variance + float(${i}));
setOutput(dot(vec3(x, -mean, offset), vec3(inv, inv, 1)));
}
`}};var Qk=class{constructor(t,e,n,o,s,i){this.packedInputs=!0,this.packedOutput=!0,this.variableNames=["x","mean","variance"],S.assertAndGetBroadcastShape(t,e),S.assertAndGetBroadcastShape(t,n);let a="vec4(0.0)";o!=null&&(S.assertAndGetBroadcastShape(t,o),this.variableNames.push("offset"),a="getOffsetAtOutCoords()");let u="vec4(1.0)";s!=null&&(S.assertAndGetBroadcastShape(t,s),this.variableNames.push("scale"),u="getScaleAtOutCoords()"),this.outputShape=t,this.userCode=`
void main() {
vec4 offset = ${a};
vec4 scale = ${u};
vec4 x = getXAtOutCoords();
vec4 mean = getMeanAtOutCoords();
vec4 variance = getVarianceAtOutCoords();
vec4 inv = scale * inversesqrt(variance + vec4(${i}));
setOutput((x - mean) * inv + offset);
}
`}};var iet=({inputs:r,backend:t,attrs:e})=>{let{x:n,mean:o,variance:s,offset:i,scale:a}=r;x.assert(o.shape.length===s.shape.length,()=>"Batch normalization gradient requires mean and variance to have equal ranks."),x.assert(i==null||o.shape.length===i.shape.length,()=>"Batch normalization gradient requires mean and offset to have equal ranks."),x.assert(a==null||o.shape.length===a.shape.length,()=>"Batch normalization gradient requires mean and scale to have equal ranks.");let{varianceEpsilon:u}=e;u==null&&(u=.001);let l=[n,o,s],c=null;i!=null&&(c=i.shape,l.push(i));let p=null;a!=null&&(p=a.shape,l.push(a));let m=V().getBool("WEBGL_PACK_NORMALIZATION")?new Qk(n.shape,o.shape,s.shape,c,p,u):new Jk(n.shape,o.shape,s.shape,c,p,u);return t.runWebGLProgram(m,l,l[0].dtype)},AL={kernelName:cs,backendName:"webgl",kernelFunc:iet};var eN=class{constructor(t){this.variableNames=["source"],this.outputShape=t,this.rank=t.length;let e=zt(this.rank);this.customUniforms=[{name:"start",arrayIndex:this.rank,type:"int"}];let n=aet(this.rank),o,s=t.map((i,a)=>`sourceLoc.${tN[a]} = start[${a}] + coords.${tN[a]};`);o=`
${e} sourceLoc;
${e} coords = getOutputCoords();
${s.join(`
`)}
`,this.userCode=`
void main() {
${o}
setOutput(getSource(${n}));
}
`}},tN=["x","y","z","w","u","v"];function aet(r){if(r===1)return"sourceLoc";if(r<=6)return tN.slice(0,r).map(t=>"sourceLoc."+t).join(",");throw Error(`Slicing for rank ${r} is not yet supported`)}var rN=class{constructor(t){this.variableNames=["source"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=t,this.rank=t.length,this.customUniforms=[{name:"start",arrayIndex:this.rank,type:"int"}];let e=zt(this.rank),n=rr("coords",this.rank),o=rr("sourceLoc",this.rank),s=this.rank===1?"sourceLoc":`vec2(${o.slice(-2).join()})`,i=`getChannel(getSource(${o.join()}), ${s})`,a=`
result.x = ${i};
if (++${n[this.rank-1]} < ${t[this.rank-1]}) {
++${o[this.rank-1]};
result.y = ${i};
--${o[this.rank-1]};
}
`,u=this.rank===1?"":`
--${n[this.rank-1]};
if (++${n[this.rank-2]} < ${t[this.rank-2]}) {
++${o[this.rank-2]};
result.z = ${i};
if (++${n[this.rank-1]} < ${t[this.rank-1]}) {
++${o[this.rank-1]};
result.w = ${i};
}
}
`,l=this.rank<=4?`sourceLoc = coords +
${e}(${t.map((c,p)=>`start[${p}]`).join()});`:t.map((c,p)=>`${o[p]} = ${n[p]} + start[${p}];`).join(`
`);this.userCode=`
void main() {
${e} coords = getOutputCoords();
${e} sourceLoc;
${l}
vec4 result = vec4(0.);
${a}
${u}
setOutput(result);
}
`}};function uet(r,t,e,n){let o=n.texData.get(r.dataId),s=n.makeTensorInfo(e,r.dtype),i=n.texData.get(s.dataId);Object.assign(i,o),i.refCount=1,i.shape=e,i.dtype=r.dtype;let a=Ve.computeFlatOffset(t,x.computeStrides(r.shape));o.slice&&(a+=o.slice.flatOffset),i.slice={flatOffset:a,origDataId:o.slice&&o.slice.origDataId||r.dataId};let u=n.dataRefCount.get(i.slice.origDataId)||1;return n.dataRefCount.set(i.slice.origDataId,u+1),s}function pi(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{begin:s,size:i}=n,[a,u]=Ve.parseSliceParams(o,s,i);if(Ve.assertParamsValid(o,a,u),x.sizeFromShape(u)===0)return e.makeTensorInfo(u,o.dtype,[]);if(e.shouldExecuteOnCPU([o])||o.dtype==="string"){let p=e.texData.get(o.dataId),m=yM(p.values,a,u,o.shape,o.dtype);return e.makeTensorInfo(u,o.dtype,m)}let{isPacked:l}=e.texData.get(o.dataId),c=Ve.isSliceContinous(o.shape,a,u);if(l||!c){let p=V().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new rN(u):new eN(u),m=[a];return e.runWebGLProgram(p,[o],o.dtype,m)}return e.uploadToGPU(o.dataId),uet(o,a,u,e)}var DL={kernelName:ki,backendName:"webgl",kernelFunc:pi};var cet=r=>{let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{blockShape:s,crops:i}=n;x.assert(o.shape.length<=4,()=>"batchToSpaceND for rank > 4 with a WebGL backend not implemented yet");let a=s.reduce((b,w)=>b*w),u=S.getReshaped(o.shape,s,a),l=S.getPermuted(u.length,s.length),c=S.getReshapedPermuted(o.shape,s,a),p=S.getSliceBeginCoords(i,s.length),m=S.getSliceSize(c,i,s.length),f=[],d=ut({inputs:{x:o},backend:e,attrs:{shape:u}}),h=Ee({inputs:{x:d},backend:e,attrs:{perm:l}}),g=ut({inputs:{x:h},backend:e,attrs:{shape:c}}),y=pi({inputs:{x:g},backend:e,attrs:{begin:p,size:m}});return f.push(d),f.push(h),f.push(g),f.forEach(b=>e.disposeIntermediateTensorInfo(b)),y},$L={kernelName:gi,backendName:"webgl",kernelFunc:cet};function pet(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,weights:s}=t,{size:i}=n,a=e.readSync(o.dataId),u=e.readSync(s.dataId),l=vb(a,u,s.dtype,s.shape,i);return e.makeTensorInfo([i],s.dtype,l)}var RL={kernelName:cp,backendName:"webgl",kernelFunc:pet};function met(r){let{inputs:t,backend:e}=r,{s0:n,s1:o}=t,s=e.readSync(n.dataId),i=e.readSync(o.dataId),a=S.assertAndGetBroadcastShape(Array.from(s),Array.from(i));return e.makeTensorInfo([a.length],"int32",Int32Array.from(a))}var FL={kernelName:pp,backendName:"webgl",kernelFunc:met};var fet="return float(a != b);",nN=le({opSnippet:fet,cpuKernelImpl:fM,dtype:"bool"}),OL={kernelName:wa,backendName:"webgl",kernelFunc:nN};function fl(r){let{inputs:t,backend:e}=r,{input:n}=t,o=e.texData.get(n.dataId);return Ke({inputs:{x:o.complexTensorInfos.real},backend:e})}var PL={kernelName:Dp,backendName:"webgl",kernelFunc:fl};var det="return float(int(x));";function ML(r,t){let e=new vn(r.shape,det),n=t.runWebGLProgram(e,[r],"int32");return{dataId:n.dataId,shape:n.shape,dtype:n.dtype}}function oN(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{dtype:s}=n;if(s==="complex64"){if(o.dtype==="complex64")return Ke({inputs:{x:o},backend:e});let i=ye(o.shape),a=oN({inputs:{x:o},backend:e,attrs:{dtype:"float32"}}),u=Dn({inputs:{real:a,imag:i},backend:e});return i.dispose(),e.disposeIntermediateTensorInfo(a),u}if(o.dtype==="complex64"){let i=fl({inputs:{input:o},backend:e}),a=oN({inputs:{x:i},backend:e,attrs:{dtype:s}});return e.disposeIntermediateTensorInfo(i),a}if(!x.hasEncodingLoss(o.dtype,s)){let i=Ke({inputs:{x:o},backend:e});return{dataId:i.dataId,shape:i.shape,dtype:s}}if(s==="int32")return ML(o,e);if(s==="bool"){let i=e.makeTensorInfo([],"bool",x.getTypedArrayFromDType("bool",1)),u=nN({inputs:{a:o,b:i},backend:e});return e.disposeIntermediateTensorInfo(i),u}throw new Error(`Error in Cast: failed to cast ${o.dtype} to ${s}`)}var LL={kernelName:oo,backendName:"webgl",kernelFunc:oN};var zL="return ceil(x);",het=Ct({opSnippet:zL,packedOpSnippet:zL,cpuKernelImpl:jP}),BL={kernelName:Jo,backendName:"webgl",kernelFunc:het};var sN=class{constructor(t){this.variableNames=["A"],this.customUniforms=[{name:"minVal",type:"float"},{name:"maxVal",type:"float"}],this.outputShape=t,this.userCode=`
void main() {
float value = getAAtOutCoords();
if (isnan(value)) {
setOutput(value);
return;
}
setOutput(clamp(value, minVal, maxVal));
}
`}};var iN=class{constructor(t){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.customUniforms=[{name:"minVal",type:"float"},{name:"maxVal",type:"float"}],this.outputShape=t,this.userCode=`
void main() {
vec4 value = getAAtOutCoords();
if (any(isnan(value))) {
setOutput(value);
return;
}
setOutput(clamp(value, vec4(minVal), vec4(maxVal)));
}
`}};function get(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{clipValueMin:s,clipValueMax:i}=n,a;V().getBool("WEBGL_PACK_CLIP")?a=new iN(o.shape):a=new sN(o.shape);let u=[[s],[i]];return e.runWebGLProgram(a,[o],o.dtype,u)}var VL={kernelName:so,backendName:"webgl",kernelFunc:get};var aN=class{constructor(t){this.variableNames=["real","imag"],this.outputShape=t,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 GL(r,t){return{dataId:t.dataId,dtype:t.dtype,shape:r.shape}}function xet(r){let{inputs:t,backend:e}=r,{x:n}=t,o=e.texData.get(n.dataId),s=new aN(n.shape),i=[GL(n,o.complexTensorInfos.real),GL(n,o.complexTensorInfos.imag)];return e.runWebGLProgram(s,i,i[0].dtype)}var WL={kernelName:Il,backendName:"webgl",kernelFunc:xet};var lN=class{constructor(t){this.outputShape=[],this.outputShape=S.computeOutShape(t,1),this.variableNames=t.map((i,a)=>`T${a}`);let e=new Array(t.length-1);e[0]=t[0][1];for(let i=1;i<e.length;i++)e[i]=e[i-1]+t[i][1];let n=[`if (yC < ${e[0]}) setOutput(getT0(yR, yC));`];for(let i=1;i<e.length;i++){let a=e[i-1];n.push(`else if (yC < ${e[i]}) setOutput(getT${i}(yR, yC-${a}));`)}let o=e.length,s=e[e.length-1];n.push(`else setOutput(getT${o}(yR, yC-${s}));`),this.userCode=`
void main() {
ivec2 coords = getOutputCoords();
int yR = coords.x;
int yC = coords.y;
${n.join(`
`)}
}
`}};var uN=class{constructor(t,e){this.packedInputs=!0,this.packedOutput=!0,this.outputShape=[],this.outputShape=S.computeOutShape(t,e);let n=this.outputShape,o=n.length,s=zt(o),i=rr("coords",o),a=["x","y","z","w","u","v"].slice(0,o);this.variableNames=t.map((h,g)=>`T${g}`);let u=new Array(t.length-1);u[0]=t[0][e];for(let h=1;h<u.length;h++)u[h]=u[h-1]+t[h][e];let l=a[e],c=a.slice(-2),p=a.join(),m=`if (${l} < ${u[0]}) {
return getChannel(
getT0(${p}), vec2(${c.join()}));
}`;for(let h=1;h<u.length;h++){let g=u[h-1];m+=`
if (${l} < ${u[h]} && ${l} >= ${u[h-1]}) {
return getChannel(
getT${h}(${Ab(a,l,g)}),
vec2(${Ab(c,l,g)}));
}`}let f=u.length,d=u[u.length-1];m+=`
return getChannel(
getT${f}(${Ab(a,l,d)}),
vec2(${Ab(c,l,d)}));`,this.userCode=`
float getValue(${a.map(h=>"int "+h)}) {
${m}
}
void main() {
${s} coords = getOutputCoords();
vec4 result = vec4(getValue(${i}), 0., 0., 0.);
${i[o-1]} = ${i[o-1]} + 1;
if (${i[o-1]} < ${n[o-1]}) {
result.g = getValue(${i});
}
${i[o-2]} = ${i[o-2]} + 1;
if (${i[o-2]} < ${n[o-2]}) {
result.a = getValue(${i});
}
${i[o-1]} = ${i[o-1]} - 1;
if (${i[o-2]} < ${n[o-2]} &&
${i[o-1]} < ${n[o-1]}) {
result.b = getValue(${i});
}
setOutput(result);
}
`}};function Ab(r,t,e){let n=r.indexOf(t);return r.map((s,i)=>i===n?`${s} - ${e}`:s).join()}function zc(r){let{inputs:t,backend:e}=r,{input:n}=t,o=e.texData.get(n.dataId);return Ke({inputs:{x:o.complexTensorInfos.imag},backend:e})}var UL={kernelName:Sp,backendName:"webgl",kernelFunc:zc};function Bc(r,t,e){let n=r[0].dtype;if(n==="complex64"){let c=r.map(h=>fl({inputs:{input:h},backend:e})),p=r.map(h=>zc({inputs:{input:h},backend:e})),m=Bc(c,t,e),f=Bc(p,t,e),d=Dn({inputs:{real:m,imag:f},backend:e});return c.forEach(h=>e.disposeIntermediateTensorInfo(h)),p.forEach(h=>e.disposeIntermediateTensorInfo(h)),e.disposeIntermediateTensorInfo(m),e.disposeIntermediateTensorInfo(f),d}let o=e.shouldExecuteOnCPU(r);if(n==="string"&&(o=!0),o){let c=r.map(y=>{let b=x.sizeFromShape(y.shape.slice(t));return ut({inputs:{x:y},backend:e,attrs:{shape:[-1,b]}})}),p=c.map(y=>({vals:e.readSync(y.dataId),shape:y.shape})),m=S.computeOutShape(c.map(y=>y.shape),1),f=c[0].shape[0]===1,d=XP(p,m,n,f),h=S.computeOutShape(r.map(y=>y.shape),t),g=e.makeTensorInfo(h,n,d);return c.forEach(y=>e.disposeIntermediateTensorInfo(y)),g}if(r.length>V().getNumber("WEBGL_MAX_TEXTURES_IN_SHADER")){let c=Math.floor(r.length/2),p=Bc(r.slice(0,c),t,e),m=Bc(r.slice(c),t,e),f=Bc([p,m],t,e);return e.disposeIntermediateTensorInfo(p),e.disposeIntermediateTensorInfo(m),f}if(V().getBool("WEBGL_PACK_ARRAY_OPERATIONS")&&r[0].shape.length>1){let c=new uN(r.map(p=>p.shape),t);return e.runWebGLProgram(c,r,n)}let{tensors2D:s,outShape:i}=yet(r,t,e),a=new lN(s.map(c=>c.shape)),u=e.runWebGLProgram(a,s,n);s.forEach(c=>e.disposeIntermediateTensorInfo(c));let l=ut({inputs:{x:u},attrs:{shape:i},backend:e});return e.disposeIntermediateTensorInfo(u),l}function yet(r,t,e){let n=S.computeOutShape(r.map(s=>s.shape),t);return{tensors2D:r.map(s=>ut({inputs:{x:s},attrs:{shape:[-1,x.sizeFromShape(s.shape.slice(t))]},backend:e})),outShape:n}}function cN(r){let{inputs:t,backend:e,attrs:n}=r,{axis:o}=n,s=x.parseAxisParam(o,t[0].shape)[0],i=S.computeOutShape(t.map(l=>l.shape),s);if(x.sizeFromShape(i)===0)return e.makeTensorInfo(i,t[0].dtype,[]);let a=t.filter(l=>x.sizeFromShape(l.shape)>0);if(a.length===1)return Ke({inputs:{x:a[0]},backend:e});let u=a.map(l=>l.shape);return S.assertParamsConsistent(u,s),Bc(a,s,e)}var HL={kernelName:xi,backendName:"webgl",kernelFunc:cN};var mg=class{constructor(t,e=!1,n=null,o=!1,s=!1){this.variableNames=["x","W"],this.outputShape=t.outShape;let i=t.padInfo.top,a=t.padInfo.left,u=t.strideHeight,l=t.strideWidth,c=t.dilationHeight,p=t.dilationWidth,m=t.filterHeight,f=t.filterWidth,d=Math.floor(t.inChannels/4)*4,h=t.inChannels%4,g=t.dataFormat==="channelsLast",y=g?1:2,b=g?2:3,w=g?3:1,v="",k="";n&&(o?v=`float activation(float a) {
float b = getPreluActivationWeightsAtOutCoords();
${n}
}`:s?v=`float activation(float a) {
float b = getLeakyreluAlphaAtOutCoords();
${n}
}`:v=`
float activation(float x) {
${n}
}
`,k="result = activation(result);");let E=e?"result += getBiasAtOutCoords();":"";e&&this.variableNames.push("bias"),o&&this.variableNames.push("preluActivationWeights"),s&&this.variableNames.push("leakyreluAlpha"),this.userCode=`
${v}
const ivec2 strides = ivec2(${u}, ${l});
const ivec2 pads = ivec2(${i}, ${a});
void main() {
ivec4 coords = getOutputCoords();
int batch = coords[0];
int d2 = coords[${w}];
ivec2 xRCCorner =
ivec2(coords[${y}], coords[${b}]) * 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 < ${m}; wR++) {
int xR = xRCorner + wR * ${c};
if (xR < 0 || xR >= ${t.inHeight}) {
continue;
}
for (int wC = 0; wC < ${f}; wC++) {
int xC = xCCorner + wC * ${p};
if (xC < 0 || xC >= ${t.inWidth}) {
continue;
}
for (int d1 = 0; d1 < ${d}; d1 += 4) {
vec4 wValues = vec4(
getW(wR, wC, d1, d2),
getW(wR, wC, d1 + 1, d2),
getW(wR, wC, d1 + 2, d2),
getW(wR, wC, d1 + 3, d2)
);
if (${g}) {
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 (${h===1}) {
if (${g}) {
dotProd +=
getX(batch, xR, xC, ${d}) *
getW(wR, wC, ${d}, d2);
} else {
dotProd +=
getX(batch, ${d}, xR, xC) *
getW(wR, wC, ${d}, d2);
}
} else if (${h===2}) {
vec2 wValues = vec2(
getW(wR, wC, ${d}, d2),
getW(wR, wC, ${d} + 1, d2)
);
if (${g}) {
vec2 xValues = vec2(
getX(batch, xR, xC, ${d}),
getX(batch, xR, xC, ${d} + 1)
);
dotProd += dot(xValues, wValues);
} else {
vec2 xValues = vec2(
getX(batch, ${d}, xR, xC),
getX(batch, ${d} + 1, xR, xC)
);
dotProd += dot(xValues, wValues);
}
} else if (${h===3}) {
vec3 wValues = vec3(
getW(wR, wC, ${d}, d2),
getW(wR, wC, ${d} + 1, d2),
getW(wR, wC, ${d} + 2, d2)
);
if (${g}) {
vec3 xValues = vec3(
getX(batch, xR, xC, ${d}),
getX(batch, xR, xC, ${d} + 1),
getX(batch, xR, xC, ${d} + 2)
);
dotProd += dot(xValues, wValues);
} else {
vec3 xValues = vec3(
getX(batch, ${d}, xR, xC),
getX(batch, ${d} + 1, xR, xC),
getX(batch, ${d} + 2, xR, xC)
);
dotProd += dot(xValues, wValues);
}
}
}
}
float result = dotProd;
${E}
${k}
setOutput(result);
}
`}},pN=class{constructor(t){this.variableNames=["x","W"],this.outputShape=t.outShape;let e=t.padInfo.front,n=t.padInfo.top,o=t.padInfo.left,s=t.strideDepth,i=t.strideHeight,a=t.strideWidth,u=t.dilationDepth,l=t.dilationHeight,c=t.dilationWidth,p=t.filterDepth,m=t.filterHeight,f=t.filterWidth,d=Math.floor(t.inChannels/4)*4,h=t.inChannels%4;this.userCode=`
const ivec3 strides = ivec3(${s}, ${i}, ${a});
const ivec3 pads = ivec3(${e}, ${n}, ${o});
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 < ${p}; wF++) {
int xF = xFCorner + wF * ${u};
if (xF < 0 || xF >= ${t.inDepth}) {
continue;
}
for (int wR = 0; wR < ${m}; wR++) {
int xR = xRCorner + wR * ${l};
if (xR < 0 || xR >= ${t.inHeight}) {
continue;
}
for (int wC = 0; wC < ${f}; wC++) {
int xC = xCCorner + wC * ${c};
if (xC < 0 || xC >= ${t.inWidth}) {
continue;
}
for (int d1 = 0; d1 < ${d}; d1 += 4) {
vec4 xValues = vec4(
getX(batch, xF, xR, xC, d1),
getX(batch, xF, xR, xC, d1 + 1),
getX(batch, xF, xR, xC, d1 + 2),
getX(batch, xF, xR, xC, d1 + 3)
);
vec4 wValues = vec4(
getW(wF, wR, wC, d1, d2),
getW(wF, wR, wC, d1 + 1, d2),
getW(wF, wR, wC, d1 + 2, d2),
getW(wF, wR, wC, d1 + 3, d2)
);
dotProd += dot(xValues, wValues);
}
if (${h===1}) {
dotProd +=
getX(batch, xF, xR, xC, ${d}) *
getW(wF, wR, wC, ${d}, d2);
} else if (${h===2}) {
vec2 xValues = vec2(
getX(batch, xF, xR, xC, ${d}),
getX(batch, xF, xR, xC, ${d} + 1)
);
vec2 wValues = vec2(
getW(wF, wR, wC, ${d}, d2),
getW(wF, wR, wC, ${d} + 1, d2)
);
dotProd += dot(xValues, wValues);
} else if (${h===3}) {
vec3 xValues = vec3(
getX(batch, xF, xR, xC, ${d}),
getX(batch, xF, xR, xC, ${d} + 1),
getX(batch, xF, xR, xC, ${d} + 2)
);
vec3 wValues = vec3(
getW(wF, wR, wC, ${d}, d2),
getW(wF, wR, wC, ${d} + 1, d2),
getW(wF, wR, wC, ${d} + 2, d2)
);
dotProd += dot(xValues, wValues);
}
}
}
}
setOutput(dotProd);
}
`}};var mN=class{constructor(t,e){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.customUniforms=[{name:"inputShape",type:"ivec3"},{name:"pad",type:"ivec2"},{name:"stride",type:"ivec2"},{name:"dilation",type:"ivec2"},{name:"inChannels",type:"int"},{name:"itemsPerBlockRow",type:"int"},{name:"outWidth",type:"int"}],this.outputShape=t,this.enableShapeUniforms=$e(this.outputShape.length);let{dataFormat:n}=e,o=qe(),s=n==="channelsLast",i=s?0:1,a=s?1:2,u=this.enableShapeUniforms?"if(blockIndex < outShape[1] && pos < outShape[0]) {":`if(blockIndex < ${t[1]} && pos < ${t[0]}) {`,l="";for(let c=0;c<=1;c++)for(let p=0;p<=1;p++)l+=`
blockIndex = rc.y + ${p};
pos = rc.x + ${c};
${u}
offsetY = int(blockIndex / outWidth) * stride[0] - pad[0];
d0 = offsetY + dilation[0] * (pos / itemsPerBlockRow);
if(d0 < inputShape[${i}] && d0 >= 0) {
// Use custom imod instead mod. On Intel GPU, mod may generate
// unexpected value.
// https://github.com/tensorflow/tfjs/issues/5447
offsetX = imod(blockIndex, outWidth) * stride[1] - pad[1];
d1 = offsetX + dilation[1] * (imod(pos, itemsPerBlockRow) /
inChannels);
if(d1 < inputShape[${a}] && d1 >= 0) {
ch = imod(pos, inChannels);
if (${s}) {
innerDims = vec2(d1, ch);
result[${c*2+p}] = getChannel(
getA(d0, int(innerDims.x),
int(innerDims.y)), innerDims);
} else {
innerDims = vec2(d0, d1);
result[${c*2+p}] = getChannel(
getA(ch, int(innerDims.x),
int(innerDims.y)), innerDims);
}
}
}
}
`;this.userCode=`
void main() {
ivec2 rc = getOutputCoords();
vec4 result = vec4(0);
int blockIndex, pos, offsetY, d0, offsetX, d1, ch;
vec2 innerDims;
${l}
${o.output} = result;
}
`}};function Db({x:r,filter:t,convInfo:e,backend:n,bias:o=null,preluActivationWeights:s=null,leakyreluAlpha:i=0,activation:a=null}){let u=r.shape,l=n.texData.get(r.dataId),c=e.inChannels,p=u[0]*u[1]*u[2],m=e.outChannels,f=e.dataFormat==="channelsLast",d=!1,h=!1,g,y=[];if(!((p===1||m===1)&&c>Hk)&&l.isPacked&&f&&l.texture!=null&&u[2]%2!==0&&x.arraysEqual(l.shape.slice(-3),u.slice(-3))){let v=u[0]*u[1]*(u[2]+1),k={dataId:r.dataId,shape:[1,v,e.inChannels],dtype:r.dtype},E=l.shape;l.shape=l.shape.slice(),l.shape[l.shape.length-2]++,x.assert(su(l.shape,k.shape),()=>`packed reshape ${l.shape} to ${k.shape} isn't free`);let D=ut({inputs:{x:t},backend:n,attrs:{shape:[1,e.inChannels,e.outChannels]}});y.push(D);let $=Mc({a:k,b:D,backend:n,transposeA:d,transposeB:h,bias:o,activation:a,preluActivationWeights:s,leakyreluAlpha:i}),R=n.texData.get($.dataId);x.assert(R.isPacked,()=>"batchMatMul result is expected to be packed"),l.shape=E,R.shape=e.outShape,g=Ke({inputs:{x:$},backend:n}),g.shape=e.outShape,y.push($)}else{let v=f?u[0]*u[1]*u[2]:u[0]*u[2]*u[3],k=ut({inputs:{x:r},backend:n,attrs:{shape:[1,v,e.inChannels]}}),E=ut({inputs:{x:t},backend:n,attrs:{shape:[1,e.inChannels,e.outChannels]}}),D=Mc({a:k,b:E,transposeA:d,transposeB:h,backend:n,bias:o,activation:a,preluActivationWeights:s,leakyreluAlpha:i});g=ut({inputs:{x:D},backend:n,attrs:{shape:e.outShape}}),y.push(k),y.push(E),y.push(D)}for(let v of y)n.disposeIntermediateTensorInfo(v);return g}function $b({x:r,filter:t,convInfo:e,backend:n,bias:o=null,preluActivationWeights:s=null,leakyreluAlpha:i=0,activation:a=null}){let{filterWidth:u,filterHeight:l,inChannels:c,outWidth:p,outHeight:m,dataFormat:f}=e,d=f==="channelsLast",h=u*l*c,g=m*p,y=[h,g],b=!0,w=!1,v=[],k=ut({inputs:{x:r},backend:n,attrs:{shape:r.shape.slice(1)}}),E=ut({inputs:{x:t},backend:n,attrs:{shape:[1,h,x.sizeFromShape(t.shape)/h]}});v.push(k),v.push(E);let D=new mN(y,e),$=[k.shape,[e.padInfo.top,e.padInfo.left],[e.strideHeight,e.strideWidth],[e.dilationHeight,e.dilationWidth],[e.inChannels],[e.filterWidth*e.inChannels],[e.outWidth]],R=n.runWebGLProgram(D,[k],"float32",$),M=ut({inputs:{x:R},backend:n,attrs:{shape:[1,y[0],y[1]]}});v.push(R),v.push(M);let W=o!=null,U=s!=null,q=a==="leakyrelu",K=a?au(a,!0):null,j=new cg(M.shape,E.shape,[1,g,e.outChannels],b,w,W,K,U,q),Q=[M,E];if(o&&Q.push(o),U&&Q.push(s),q){let st=n.makeTensorInfo([],"float32",x.createScalarValue(i,"float32"));Q.push(st),v.push(st)}let rt=n.runWebGLProgram(j,Q,"float32"),X=d?[1,m,p,e.outChannels]:[1,e.outChannels,m,p],ot=ut({inputs:{x:rt},backend:n,attrs:{shape:X}});v.push(rt);for(let st of v)n.disposeIntermediateTensorInfo(st);return ot}function bet(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,filter:s}=t,{strides:i,pad:a,dataFormat:u,dilations:l,dimRoundingMode:c}=n,p=S.convertConv2DDataFormat(u),m=S.computeConv2DInfo(o.shape,s.shape,i,l,a,c,!1,p),f;if(m.filterHeight===1&&m.filterWidth===1&&m.dilationHeight===1&&m.dilationWidth===1&&m.strideHeight===1&&m.strideWidth===1&&(m.padInfo.type==="SAME"||m.padInfo.type==="VALID"))f=Db({x:o,filter:s,convInfo:m,backend:e});else if(V().getBool("WEBGL_CONV_IM2COL")&&o.shape[0]===1)f=$b({x:o,filter:s,convInfo:m,backend:e});else{let h=new mg(m);f=e.runWebGLProgram(h,[o,s],"float32")}let d=ut({inputs:{x:f},backend:e,attrs:{shape:m.outShape}});return e.disposeIntermediateTensorInfo(f),d}var qL={kernelName:Qo,backendName:"webgl",kernelFunc:bet};var fN=class{constructor(t){this.variableNames=["x","dy"],this.outputShape=t.filterShape;let e=t.strideHeight,n=t.strideWidth,o=t.padInfo.top,s=t.padInfo.left,i=t.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 < ${t.batchSize}; b++) {
for (int yR = 0; yR < ${t.outHeight}; yR++) {
int xR = wR + yR * ${e} - ${o};
if (xR < 0 || xR >= ${t.inHeight}) {
continue;
}
for (int yC = 0; yC < ${t.outWidth}; yC++) {
int xC = wC + yC * ${n} - ${s};
if (xC < 0 || xC >= ${t.inWidth}) {
continue;
}
if (${i}) {
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);
}
`}},dN=class{constructor(t){this.variableNames=["dy","W"],this.outputShape=t.inShape;let e=t.filterHeight,n=t.filterWidth,o=t.strideHeight,s=t.strideWidth,i=t.dataFormat==="channelsLast",a=e-1-t.padInfo.top,u=n-1-t.padInfo.left,l=i?1:2,c=i?2:3,p=i?3:1;this.userCode=`
const ivec2 pads = ivec2(${a}, ${u});
void main() {
ivec4 coords = getOutputCoords();
int batch = coords[0];
int d1 = coords[${p}];
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 < ${e}; wR++) {
float dyR = float(dyRCorner + wR) / ${o}.0;
if (dyR < 0.0 || dyR >= ${t.outHeight}.0 || fract(dyR) > 0.0) {
continue;
}
int idyR = int(dyR);
int wRPerm = ${e} - 1 - wR;
for (int wC = 0; wC < ${n}; wC++) {
float dyC = float(dyCCorner + wC) / ${s}.0;
if (dyC < 0.0 || dyC >= ${t.outWidth}.0 ||
fract(dyC) > 0.0) {
continue;
}
int idyC = int(dyC);
int wCPerm = ${n} - 1 - wC;
for (int d2 = 0; d2 < ${t.outChannels}; d2++) {
if (${i}) {
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);
}
`}},hN=class{constructor(t){this.variableNames=["x","dy"],this.outputShape=t.filterShape;let e=t.strideDepth,n=t.strideHeight,o=t.strideWidth,s=t.padInfo.front,i=t.padInfo.top,a=t.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 < ${t.batchSize}; b++) {
for (int yF = 0; yF < ${t.outDepth}; yF++) {
int xF = wF + yF * ${e} - ${s};
if (xF < 0 || xF >= ${t.inDepth}) {
continue;
}
for (int yR = 0; yR < ${t.outHeight}; yR++) {
int xR = wR + yR * ${n} - ${i};
if (xR < 0 || xR >= ${t.inHeight}) {
continue;
}
for (int yC = 0; yC < ${t.outWidth}; yC++) {
int xC = wC + yC * ${o} - ${a};
if (xC < 0 || xC >= ${t.inWidth}) {
continue;
}
float dyValue = getDy(b, yF, yR, yC, d2);
float xValue = getX(b, xF, xR, xC, d1);
dotProd += (xValue * dyValue);
}
}
}
}
setOutput(dotProd);
}
`}},gN=class{constructor(t){this.variableNames=["dy","W"],this.outputShape=t.inShape;let e=t.filterDepth,n=t.filterHeight,o=t.filterWidth,s=t.strideDepth,i=t.strideHeight,a=t.strideWidth,u=e-1-t.padInfo.front,l=n-1-t.padInfo.top,c=o-1-t.padInfo.left;this.userCode=`
const ivec3 pads = ivec3(${u}, ${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 < ${e}; wF++) {
float dyF = float(dyFCorner + wF) / ${s}.0;
if (dyF < 0.0 || dyF >= ${t.outDepth}.0 || fract(dyF) > 0.0) {
continue;
}
int idyF = int(dyF);
int wFPerm = ${e} - 1 - wF;
for (int wR = 0; wR < ${n}; wR++) {
float dyR = float(dyRCorner + wR) / ${i}.0;
if (dyR < 0.0 || dyR >= ${t.outHeight}.0 ||
fract(dyR) > 0.0) {
continue;
}
int idyR = int(dyR);
int wRPerm = ${n} - 1 - wR;
for (int wC = 0; wC < ${o}; wC++) {
float dyC = float(dyCCorner + wC) / ${a}.0;
if (dyC < 0.0 || dyC >= ${t.outWidth}.0 ||
fract(dyC) > 0.0) {
continue;
}
int idyC = int(dyC);
int wCPerm = ${o} - 1 - wC;
for (int d2 = 0; d2 < ${t.outChannels}; d2++) {
float xValue = getDy(batch, idyF, idyR, idyC, d2);
float wValue = getW(wFPerm, wRPerm, wCPerm, d1, d2);
dotProd += xValue * wValue;
}
}
}
}
setOutput(dotProd);
}
`}};function wet(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,dy:s}=t,{strides:i,pad:a,dataFormat:u,dimRoundingMode:l,filterShape:c}=n,p=S.convertConv2DDataFormat(u),m=S.computeConv2DInfo(o.shape,c,i,1,a,l,!1,p),f=new fN(m);return e.runWebGLProgram(f,[o,s],"float32")}var KL={kernelName:fp,backendName:"webgl",kernelFunc:wet};function vet(r){let{inputs:t,backend:e,attrs:n}=r,{dy:o,filter:s}=t,{inputShape:i,strides:a,pad:u,dataFormat:l,dimRoundingMode:c}=n,p=S.convertConv2DDataFormat(l),m=S.computeConv2DInfo(i,s.shape,a,1,u,c,!1,p),f=new dN(m);return e.runWebGLProgram(f,[o,s],"float32")}var jL={kernelName:ts,backendName:"webgl",kernelFunc:vet};function Cet(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,filter:s}=t,{strides:i,pad:a,dilations:u}=n,l=S.computeConv3DInfo(o.shape,s.shape,i,u,a),c=new pN(l);return e.runWebGLProgram(c,[o,s],"float32")}var XL={kernelName:Sl,backendName:"webgl",kernelFunc:Cet};function Iet(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,dy:s}=t,{strides:i,pad:a,filterShape:u}=n,l=S.computeConv3DInfo(o.shape,u,i,1,a),c=new hN(l);return e.runWebGLProgram(c,[o,s],"float32")}var YL={kernelName:dp,backendName:"webgl",kernelFunc:Iet};function ket(r){let{inputs:t,backend:e,attrs:n}=r,{dy:o,filter:s}=t,{pad:i,strides:a,inputShape:u}=n,l=S.computeConv3DInfo(u,s.shape,a,1,i),c=new gN(l);return e.runWebGLProgram(c,[o,s],"float32")}var ZL={kernelName:hp,backendName:"webgl",kernelFunc:ket};var Net=Po+`
return cos(x);
`,Tet=Ct({opSnippet:Net}),JL={kernelName:es,backendName:"webgl",kernelFunc:Tet};var _et=`
float e2x = exp(-x);
return (e2x + 1.0 / e2x) / 2.0;
`,Eet=Ct({opSnippet:_et}),QL={kernelName:rs,backendName:"webgl",kernelFunc:Eet};var xN=class{constructor(t,e,n,o,s){this.variableNames=["Image","Boxes","BoxInd"],this.outputShape=[];let[i,a,u,l]=t,[c]=e,[p,m]=n;this.outputShape=[c,p,m,l];let f=o==="bilinear"?1:0,[d,h]=[`${a-1}.0`,`${u-1}.0`],[g,y,b]=p>1?[`${(a-1)/(p-1)}`,"(y2-y1) * height_ratio",`y1*${d} + float(y)*(height_scale)`]:["0.0","0.0",`0.5 * (y1+y2) * ${d}`],[w,v,k]=m>1?[`${(u-1)/(m-1)}`,"(x2-x1) * width_ratio",`x1*${h} + float(x)*(width_scale)`]:["0.0","0.0",`0.5 * (x1+x2) * ${h}`];this.userCode=`
const float height_ratio = float(${g});
const float width_ratio = float(${w});
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 >= ${i}) {
return;
}
float height_scale = ${y};
float width_scale = ${v};
float in_y = ${b};
if( in_y < 0.0 || in_y > ${d} ) {
setOutput(float(${s}));
return;
}
float in_x = ${k};
if( in_x < 0.0 || in_x > ${h} ) {
setOutput(float(${s}));
return;
}
vec2 sourceFracIndexCR = vec2(in_x,in_y);
if(${f} == 1) {
// Compute the four integer indices.
ivec2 sourceFloorCR = ivec2(sourceFracIndexCR);
ivec2 sourceCeilCR = ivec2(ceil(sourceFracIndexCR));
float topLeft = getImage(b, sourceFloorCR.y, sourceFloorCR.x, d);
float bottomLeft = getImage(b, sourceCeilCR.y, sourceFloorCR.x, d);
float topRight = getImage(b, sourceFloorCR.y, sourceCeilCR.x, d);
float bottomRight = getImage(b, sourceCeilCR.y, sourceCeilCR.x, d);
vec2 fracCR = sourceFracIndexCR - vec2(sourceFloorCR);
float top = topLeft + (topRight - topLeft) * fracCR.x;
float bottom = bottomLeft + (bottomRight - bottomLeft) * fracCR.x;
float newValue = top + (bottom - top) * fracCR.y;
setOutput(newValue);
} else {
// Compute the coordinators of nearest neighbor point.
ivec2 sourceNearestCR = ivec2(floor(
sourceFracIndexCR + vec2(0.5,0.5)));
float newValue = getImage(b, sourceNearestCR.y, sourceNearestCR.x, d);
setOutput(newValue);
}
}
`}};var Aet=r=>{let{inputs:t,backend:e,attrs:n}=r,{image:o,boxes:s,boxInd:i}=t,{cropSize:a,method:u,extrapolationValue:l}=n,c=new xN(o.shape,s.shape,a,u,l);return e.runWebGLProgram(c,[o,s,i],"float32")},t3={kernelName:oa,backendName:"webgl",kernelFunc:Aet};var Rb=class{constructor(t,e,n){this.variableNames=["x"],this.customUniforms=[{name:"index",type:"float"}],this.outputShape=t;let o=t.length,s=e?"1.0":`getX(${e3(o,"coords")})`,i=t[t.length-1],a="",u="";e?(a=n?`end != ${i-1}`:"end != 0",u=n?"end + 1":"end - 1"):(a=n?`end + pow2 < ${i}`:"end >= pow2",u=n?"end + pow2":"end - pow2"),this.userCode=`
void main() {
${zt(o)} coords = getOutputCoords();
int end = ${r3(o,"coords")};
float val = ${s};
int pow2 = int(pow(2.0, index));
if (${a}) {
int idx = ${u};
${r3(o,"coords")} = idx;
val *= getX(${e3(o,"coords")});
}
setOutput(val);
}
`}};function e3(r,t){if(r===1)return`${t}`;if(r===2)return`${t}.x, ${t}.y`;if(r===3)return`${t}.x, ${t}.y, ${t}.z`;if(r===4)return`${t}.x, ${t}.y, ${t}.z, ${t}.w`;throw Error(`Cumulative product for rank ${r} is not yet supported`)}function r3(r,t){if(r===1)return`${t}`;if(r===2)return`${t}.y`;if(r===3)return`${t}.z`;if(r===4)return`${t}.w`;throw Error(`Cumulative product for rank ${r} is not yet supported`)}function Det(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s,exclusive:i,reverse:a}=n,u=o.shape.length,l=S.getAxesPermutation([s],u),c=o;l!=null&&(c=Ee({inputs:{x:o},backend:e,attrs:{perm:l}}));let p=S.getInnerMostAxes(1,u)[0];if(p!==u-1)throw new Error(`WebGL cumprod shader expects an inner-most axis=${o.shape.length-1} but got axis=${s}`);let m=c.shape[p],f=Ke({inputs:{x:c},backend:e});for(let d=0;d<=Math.ceil(Math.log2(m))-1;d++){let h=new Rb(c.shape,!1,a),g=[[d]],y=f;f=e.runWebGLProgram(h,[f],f.dtype,g),e.disposeIntermediateTensorInfo(y)}if(i){let d=new Rb(c.shape,i,a),h=f;f=e.runWebGLProgram(d,[f],f.dtype),e.disposeIntermediateTensorInfo(h)}if(l!=null){let d=S.getUndoAxesPermutation(l),h=Ee({inputs:{x:f},backend:e,attrs:{perm:d}});return e.disposeIntermediateTensorInfo(f),e.disposeIntermediateTensorInfo(c),h}return f}var n3={kernelName:na,backendName:"webgl",kernelFunc:Det};var Fb=class{constructor(t,e,n){this.variableNames=["x"],this.customUniforms=[{name:"index",type:"float"}],this.outputShape=t;let o=t.length,s=e?"0.0":`getX(${o3(o,"coords")})`,i=t[t.length-1],a="",u="";e?(a=n?`end != ${i-1}`:"end != 0",u=n?"end + 1":"end - 1"):(a=n?`end + pow2 < ${i}`:"end >= pow2",u=n?"end + pow2":"end - pow2"),this.userCode=`
void main() {
${zt(o)} coords = getOutputCoords();
int end = ${s3(o,"coords")};
float val = ${s};
int pow2 = int(pow(2.0, index));
if (${a}) {
int idx = ${u};
${s3(o,"coords")} = idx;
val += getX(${o3(o,"coords")});
}
setOutput(val);
}
`}};function o3(r,t){if(r===1)return`${t}`;if(r===2)return`${t}.x, ${t}.y`;if(r===3)return`${t}.x, ${t}.y, ${t}.z`;if(r===4)return`${t}.x, ${t}.y, ${t}.z, ${t}.w`;throw Error(`Cumulative sum for rank ${r} is not yet supported`)}function s3(r,t){if(r===1)return`${t}`;if(r===2)return`${t}.y`;if(r===3)return`${t}.z`;if(r===4)return`${t}.w`;throw Error(`Cumulative sum for rank ${r} is not yet supported`)}function $et(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s,exclusive:i,reverse:a}=n,u=o.shape.length,l=S.getAxesPermutation([s],u),c=o;l!=null&&(c=Ee({inputs:{x:o},backend:e,attrs:{perm:l}}));let p=S.getInnerMostAxes(1,u)[0];if(p!==u-1)throw new Error(`WebGL cumsum shader expects an inner-most axis=${o.shape.length-1} but got axis=${s}`);let m=c.shape[p],f=Ke({inputs:{x:c},backend:e});for(let d=0;d<=Math.ceil(Math.log2(m))-1;d++){let h=new Fb(c.shape,!1,a),g=[[d]],y=f;f=e.runWebGLProgram(h,[f],f.dtype,g),e.disposeIntermediateTensorInfo(y)}if(i){let d=new Fb(c.shape,i,a),h=f;f=e.runWebGLProgram(d,[f],f.dtype),e.disposeIntermediateTensorInfo(h)}if(l!=null){let d=S.getUndoAxesPermutation(l),h=Ee({inputs:{x:f},backend:e,attrs:{perm:d}});return e.disposeIntermediateTensorInfo(f),e.disposeIntermediateTensorInfo(c),h}return f}var i3={kernelName:ns,backendName:"webgl",kernelFunc:$et};function Ret(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,weights:s}=t,{size:i,binaryOutput:a}=n;if(o.shape.length===1){let u=e.readSync(o.dataId),l=e.readSync(s.dataId),c=vb(u,l,s.dtype,s.shape,i);return e.makeTensorInfo([i],s.dtype,c)}else if(o.shape.length===2){let u=e.bufferSync(o),l=e.bufferSync(s),c=KP(u,l,i,a);return e.makeTensorInfo(c.shape,s.dtype,c.values)}throw new Error(`Error in denseBincount: input must be at most rank 2, but got rank${o.shape.length}.`)}var a3={kernelName:gp,backendName:"webgl",kernelFunc:Ret};var yN=class{constructor(t,e,n){this.variableNames=["x"],this.outputShape=[],this.outputShape=t,this.blockSize=e,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 / ${e};
int offset_h = imod(h, ${e});
int in_w = w / ${e};
int offset_w = imod(w, ${e});
int offset_d = (offset_h * ${e} + offset_w) *
${this.getOutputDepthSize()};
int in_d = d + offset_d;
float result = ${this.getInputSamplingString()};
setOutput(result);
}
`}getHeightCoordString(){return this.dataFormat==="NHWC"?"coords[1]":"coords[2]"}getWidthCoordString(){return this.dataFormat==="NHWC"?"coords[2]":"coords[3]"}getDepthCoordString(){return this.dataFormat==="NHWC"?"coords[3]":"coords[1]"}getOutputDepthSize(){return this.dataFormat==="NHWC"?this.outputShape[3]:this.outputShape[1]}getInputSamplingString(){return this.dataFormat==="NHWC"?"getX(b, in_h, in_w, in_d)":"getX(b, in_d, in_h, in_w)"}};function Fet(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{blockSize:s,dataFormat:i}=n,a=o.shape[0],u=i==="NHWC"?o.shape[1]:o.shape[2],l=i==="NHWC"?o.shape[2]:o.shape[3],c=i==="NHWC"?o.shape[3]:o.shape[1],p=u*s,m=l*s,f=c/(s*s),d=i==="NHWC"?[a,p,m,f]:[a,f,p,m],h=new yN(d,s,i);return e.runWebGLProgram(h,[o],o.dtype)}var l3={kernelName:sa,backendName:"webgl",kernelFunc:Fet};var fg=class{constructor(t,e=!1,n=null,o=!1,s=!1){this.variableNames=["x","W"],this.customUniforms=[{name:"pads",type:"ivec2"},{name:"strides",type:"ivec2"},{name:"dilations",type:"ivec2"},{name:"inDims",type:"ivec2"}],this.outputShape=t.outShape,this.enableShapeUniforms=$e(this.outputShape.length);let i=t.filterHeight,a=t.filterWidth,u=t.outChannels/t.inChannels,l="",c="";n&&(o?l=`float activation(float a) {
float b = getPreluActivationWeightsAtOutCoords();
${n}
}`:s?l=`float activation(float a) {
float b = getLeakyreluAlphaAtOutCoords();
${n}
}`:l=`
float activation(float x) {
${n}
}
`,c="result = activation(result);");let p=e?"result += getBiasAtOutCoords();":"";e&&this.variableNames.push("bias"),o&&this.variableNames.push("preluActivationWeights"),s&&this.variableNames.push("leakyreluAlpha"),this.userCode=`
${l}
void main() {
ivec4 coords = getOutputCoords();
int batch = coords.x;
ivec2 xRCCorner = coords.yz * strides - pads;
int d2 = coords.w;
int d1 = d2 / ${u};
int q = d2 - d1 * ${u};
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 < ${i}; wR++) {
int xR = xRCorner + wR * dilations[0];
if (xR < 0 || xR >= inDims[0]) {
continue;
}
for (int wC = 0; wC < ${a}; wC++) {
int xC = xCCorner + wC * dilations[1];
if (xC < 0 || xC >= inDims[1]) {
continue;
}
float xVal = getX(batch, xR, xC, d1);
float wVal = getW(wR, wC, d1, q);
dotProd += xVal * wVal;
}
}
float result = dotProd;
${p}
${c}
setOutput(result);
}
`}};var dg=class{constructor(t,e=!1,n=null,o=!1,s=!1){this.variableNames=["x","W"],this.packedInputs=!0,this.packedOutput=!0,this.customUniforms=[{name:"pads",type:"ivec2"},{name:"strides",type:"ivec2"},{name:"dilations",type:"ivec2"},{name:"inDims",type:"ivec2"}],this.outputShape=t.outShape,this.enableShapeUniforms=$e(this.outputShape.length);let i=t.outChannels/t.inChannels,a=t.padInfo.left,u=t.strideWidth,l=t.dilationWidth,c=t.filterHeight,p=t.filterWidth,m=p,f=`
int xR; int xC; int xCOffset;
vec4 wTexel; vec4 previous; vec4 final;`;for(let y=0;y<p;y++)f+=`
vec4 xTexelC${y*2};
int xTexelC${y*2}Ready;
vec4 xTexelC${y*2+1};
int xTexelC${y*2+1}Ready;
vec4 xC${y};`;f+=`
for (int r = 0; r < ${c}; r++) {
`;for(let y=0;y<p;y++)f+=`
xTexelC${y*2} = vec4(0.0);
xTexelC${y*2}Ready = 0;
xTexelC${y*2+1} = vec4(0.0);
xTexelC${y*2+1}Ready = 0;
xC${y} = vec4(0.0);`;f+=`
xR = xRCorner + r * dilations[0];
if (xR >=0 && xR < inDims[0]) {
`;for(let y=0;y<(m+1)/2;y++){let b=y*2;if(f+=`
xC = xCCorner + ${b*l};
`,u===1){if(b<p&&(a%2===1?(f+=`
xCOffset = xC + 1;
if (xCOffset >= 0 && xCOffset < inDims[1] && xTexelC${b}Ready == 0) {
xTexelC${b} = getX(batch, xR, xCOffset, d1);
// Need to manually clear unused channels in case
// we're reading from recycled texture.
if (xCOffset + 1 >= inDims[1]) {
xTexelC${b}.zw = vec2(0.0);
}
xTexelC${b}Ready = 1;
}
`,l===1&&b>0?f+=`
xC${b} = vec4(xTexelC${b-2}.zw, xTexelC${b}.xy);
`:f+=`
xCOffset = xC + 1 - 2;
if (xCOffset >= 0 && xCOffset < inDims[1]) {
previous = getX(batch, xR, xCOffset, d1);
// Need to manually clear unused channels in case
// we're reading from recycled texture.
if (xCOffset + 1 >= inDims[1]) {
previous.zw = vec2(0.0);
}
xC${b} = vec4(previous.zw, xTexelC${b}.xy);
} else {
xC${b} = vec4(0.0, 0.0, xTexelC${b}.xy);
}
`):f+=`
if (xC >= 0 && xC < inDims[1] && xTexelC${b}Ready == 0) {
xTexelC${b} = getX(batch, xR, xC, d1);
if (xC + 1 >= inDims[1]) {
xTexelC${b}.zw = vec2(0.0);
}
xTexelC${b}Ready = 1;
}
xC${b} = xTexelC${b};
`,b+1<p)){let w=a%2===0?x.nearestLargerEven(l):l;l%2===0&&a%2===1||l%2!==0&&a%2!==1?(f+=`
xCOffset = xC + imod(pads[1], 2) + ${w};
if (xCOffset >= 0 && xCOffset < inDims[1] && xTexelC${b+1}Ready == 0) {
xTexelC${b+1} = getX(batch, xR, xCOffset, d1);
// Need to manually clear unused channels in case
// we're reading from recycled texture.
if (xCOffset + 1 >= inDims[1]) {
xTexelC${b+1}.zw = vec2(0.0);
}
xTexelC${b+1}Ready = 1;
}
`,l>1&&(f+=`
xCOffset -= 2;
if (xCOffset >= 0 && xCOffset < inDims[1] && xTexelC${b}Ready == 0) {
xTexelC${b} = getX(batch, xR, xCOffset, d1);
xTexelC${b}Ready = 1;
}
`),f+=`
xC${b+1} = vec4(xTexelC${b}.zw, xTexelC${b+1}.xy);
`):w===1?f+=`
xC${b+1} = xTexelC${b};
`:f+=`
xCOffset = xC + ${w};
if (xCOffset >= 0 && xCOffset < inDims[1] && xTexelC${b+1}Ready == 0) {
xTexelC${b+1} = getX(batch, xR, xCOffset, d1);
if (xCOffset + 1 >= inDims[1]) {
xTexelC${b+1}.zw = vec2(0.0);
}
xTexelC${b+1}Ready = 1;
}
xC${b+1} = xTexelC${b+1};
`}}else b<p&&(a%2===1?(f+=`
xCOffset = xC + 1 - strides[1];
if(xCOffset >= 0 && xCOffset < inDims[1] && xTexelC${b}Ready == 0) {
xTexelC${b} = getX(batch, xR, xCOffset, d1);
// Need to manually clear unused channels in case
// we're reading from recycled texture.
if (xCOffset + 1 >= inDims[1]) {
xTexelC${b}.zw = vec2(0.0);
}
xTexelC${b}Ready = 1;
}
if(xC + 1 >= 0 && xC + 1 < inDims[1] && xTexelC${b+1}Ready == 0) {
xTexelC${b+1} = getX(batch, xR, xC + 1, d1);
// Need to manually clear unused channels in case
// we're reading from recycled texture.
if (xC + 2 >= inDims[1]) {
xTexelC${b+1}.zw = vec2(0.0);
}
xTexelC${b+1}Ready = 1;
}
xC${b} = vec4(xTexelC${b}.zw, xTexelC${b+1}.zw);
`,b+1<p&&(f+=`
final = vec4(0.0);
xCOffset = xC + 1 + strides[1];
if(xCOffset >= 0 && xCOffset < inDims[1]) {
final = getX(batch, xR, xCOffset, d1);
}
xC${b+1} = vec4(xTexelC${b+1}.xy, final.xy);
`)):(f+=`
if(xC >= 0 && xC < inDims[1] && xTexelC${b}Ready == 0) {
xTexelC${b} = getX(batch, xR, xC, d1);
if (xC + 1 >= inDims[1]) {
xTexelC${b}.zw = vec2(0.0);
}
xTexelC${b}Ready = 1;
}
xCOffset = xC + strides[1];
if(xCOffset >= 0 && xCOffset < inDims[1] && xTexelC${b+1}Ready == 0) {
xTexelC${b+1} = getX(batch, xR, xCOffset, d1);
if (xCOffset + 1 >= inDims[1]) {
xTexelC${b+1}.zw = vec2(0.);
}
xTexelC${b+1}Ready = 1;
}
xC${b} = vec4(
xTexelC${b}.xy, xTexelC${b+1}.xy);
`,b+1<p&&(f+=`
xC${b+1} = vec4(xTexelC${b}.zw, xTexelC${b+1}.zw);
`)));b<p&&(f+=`
wTexel = getW(r, ${b}, d1, q);
dotProd += xC${b} * vec4(wTexel.xz, wTexel.xz);
`,b+1<p&&(f+=`
wTexel = getW(r, ${b+1}, d1, q);
dotProd += xC${b+1} * vec4(wTexel.xz, wTexel.xz);
`))}f+=`
}
`,f+=`
}
`;let d="",h="";n&&(o?d=`vec4 activation(vec4 a) {
vec4 b = getPreluActivationWeightsAtOutCoords();
${n}
}`:s?d=`vec4 activation(vec4 a) {
vec4 b = getLeakyreluAlphaAtOutCoords();
${n}
}`:d=`vec4 activation(vec4 x) {
${n}
}`,h="result = activation(result);");let g=e?"result += getBiasAtOutCoords();":"";e&&this.variableNames.push("bias"),o&&this.variableNames.push("preluActivationWeights"),s&&this.variableNames.push("leakyreluAlpha"),this.userCode=`
${d}
void main() {
ivec4 coords = getOutputCoords();
int batch = coords.x;
ivec2 xRCCorner = coords.yz * strides - pads;
int d2 = coords.w;
int d1 = d2 / ${i};
int q = d2 - d1 * ${i};
int xRCorner = xRCCorner.x;
int xCCorner = xRCCorner.y;
//intialize dotProd with a small epsilon seems to reduce GPU accuracy loss.
vec4 dotProd = vec4(0.000000000000001);
${f}
vec4 result = dotProd - vec4(0.000000000000001);
${g}
${h}
setOutput(result);
}
`}};function Oet(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,filter:s}=t,{strides:i,pad:a,dilations:u,dimRoundingMode:l}=n,c=u;c==null&&(c=[1,1]),x.assert(S.eitherStridesOrDilationsAreOne(i,c),()=>`Error in depthwiseConv2d: Either strides or dilations must be 1. Got strides ${i} and dilations '${c}'`);let p=S.computeConv2DInfo(o.shape,s.shape,i,c,a,l,!0),m;V().getBool("WEBGL_PACK_DEPTHWISECONV")&&p.strideWidth<=2&&p.outChannels/p.inChannels===1?m=new dg(p):m=new fg(p);let f=[[p.padInfo.top,p.padInfo.left],[p.strideHeight,p.strideWidth],[p.dilationHeight,p.dilationWidth],[p.inHeight,p.inWidth]];return e.runWebGLProgram(m,[o,s],"float32",f)}var u3={kernelName:os,backendName:"webgl",kernelFunc:Oet};var bN=class{constructor(t){this.variableNames=["x","dy"],this.outputShape=t.filterShape;let e=t.strideHeight,n=t.strideWidth,o=t.padInfo.top,s=t.padInfo.left,i=t.outChannels/t.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 * ${i} + dm;
float dotProd = 0.0;
// TO DO: Vec4 over the batch size
for (int b = 0; b < ${t.batchSize}; b++) {
for (int yR = 0; yR < ${t.outHeight}; yR++) {
int xR = wR + yR * ${e} - ${o};
if (xR < 0 || xR >= ${t.inHeight}) {
continue;
}
for (int yC = 0; yC < ${t.outWidth}; yC++) {
int xC = wC + yC * ${n} - ${s};
if (xC < 0 || xC >= ${t.inWidth}) {
continue;
}
float dyValue = getDy(b, yR, yC, d2);
float xValue = getX(b, xR, xC, d1);
dotProd += (xValue * dyValue);
}
}
}
setOutput(dotProd);
}
`}},wN=class{constructor(t){this.variableNames=["dy","W"],this.outputShape=t.inShape;let e=t.filterHeight,n=t.filterWidth,o=t.strideHeight,s=t.strideWidth,i=e-1-t.padInfo.top,a=n-1-t.padInfo.left,u=t.outChannels/t.inChannels;this.userCode=`
const ivec2 pads = ivec2(${i}, ${a});
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 < ${e}; wR++) {
float dyR = float(dyRCorner + wR) / ${o}.0;
if (dyR < 0.0 || dyR >= ${t.outHeight}.0 || fract(dyR) > 0.0) {
continue;
}
int idyR = int(dyR);
int wRPerm = ${e} - 1 - wR;
for (int wC = 0; wC < ${n}; wC++) {
float dyC = float(dyCCorner + wC) / ${s}.0;
if (dyC < 0.0 || dyC >= ${t.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 < ${u}; dm++) {
int d2 = d1 * ${u} + dm;
float xValue = getDy(batch, idyR, idyC, d2);
float wValue = getW(wRPerm, wCPerm, d1, dm);
dotProd += xValue * wValue;
}
}
}
setOutput(dotProd);
}
`}};function Pet(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,dy:s}=t,{strides:i,dilations:a,pad:u,dimRoundingMode:l,filterShape:c}=n,p=S.computeConv2DInfo(o.shape,c,i,a,u,l,!0),m=new bN(p);return e.runWebGLProgram(m,[o,s],"float32")}var c3={kernelName:xp,backendName:"webgl",kernelFunc:Pet};function Met(r){let{inputs:t,backend:e,attrs:n}=r,{dy:o,filter:s}=t,{strides:i,dilations:a,pad:u,dimRoundingMode:l,inputShape:c}=n,p=S.computeConv2DInfo(c,s.shape,i,a,u,l,!0),m=new wN(p);return e.runWebGLProgram(m,[o,s],"float32")}var p3={kernelName:yp,backendName:"webgl",kernelFunc:Met};var vN=class{constructor(t){this.variableNames=["X"],this.outputShape=[t,t],this.userCode=`
void main() {
ivec2 coords = getOutputCoords();
float val = coords[0] == coords[1] ? getX(coords[0]) : 0.0;
setOutput(val);
}
`}};function Let(r){let{inputs:t,backend:e}=r,{x:n}=t,o=[...n.shape,...n.shape],s=x.sizeFromShape(n.shape),i=ut({inputs:{x:n},backend:e,attrs:{shape:[s]}}),a=new vN(s),u=e.runWebGLProgram(a,[i],i.dtype),l=ut({inputs:{x:u},backend:e,attrs:{shape:o}});return e.disposeIntermediateTensorInfo(i),e.disposeIntermediateTensorInfo(u),l}var m3={kernelName:bp,backendName:"webgl",kernelFunc:Let};var CN=class{constructor(t){this.variableNames=["x","W"],this.outputShape=t.outShape;let{inHeight:e,inWidth:n,padInfo:o,strideHeight:s,strideWidth:i,filterHeight:a,filterWidth:u,dilationHeight:l,dilationWidth:c}=t,{top:p,left:m}=o;this.userCode=`
const ivec2 strides = ivec2(${s}, ${i});
const ivec2 pads = ivec2(${p}, ${m});
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 < ${a}; h++) {
int hIn = hBeg + h * ${l};
if (hIn >= 0 && hIn < ${e}) {
for (int w = 0; w < ${u}; 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 zet(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,filter:s}=t,{strides:i,pad:a,dilations:u}=n,l=S.computeDilation2DInfo(o.shape,s.shape,i,a,"NHWC",u),c,p=new CN(l);c=e.runWebGLProgram(p,[o,s],"float32");let m=ut({inputs:{x:c},backend:e,attrs:{shape:l.outShape}});return e.disposeIntermediateTensorInfo(c),m}var f3={kernelName:kl,backendName:"webgl",kernelFunc:zet};function Bet(r){let{inputs:t,backend:e,attrs:n}=r,{equation:o}=n,s=t,{allDims:i,summedDims:a,idDims:u}=S.decodeEinsumEquation(o,s.length);S.checkEinsumDimSizes(i.length,u,s);let{path:l,steps:c}=S.getEinsumComputePath(a,u),p=c.length,m=null,f=i.length,d=[];for(let h=0;h<p;++h){for(let g of c[h]){let{permutationIndices:y,expandDims:b}=S.getEinsumPermutation(f,u[g]),w;S.isIdentityPermutation(y)?w=s[g]:(w=Ee({inputs:{x:s[g]},backend:e,attrs:{perm:y}}),d.push(w));let v=w.shape.slice();for(let k=0;k<b.length;++k)v.splice(b[k],0,1);x.arraysEqual(w.shape,v)||(w=ut({inputs:{x:w},backend:e,attrs:{shape:v}}),d.push(w)),m===null?m=w:(m=pg({inputs:{a:w,b:m},backend:e}),d.push(m))}h<p-1&&(l[h]>=0&&(m=Pc({inputs:{x:m},backend:e,attrs:{axis:l[h]-(i.length-f),keepDims:!1}}),d.push(m)),f--)}for(let h of d)h!==m&&e.disposeIntermediateTensorInfo(h);return m}var d3={kernelName:wp,backendName:"webgl",kernelFunc:Bet};var Vet="return (x >= 0.0) ? x : (exp(x) - 1.0);",Get=`
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;
`,Wet=Ct({opSnippet:Vet,packedOpSnippet:Get}),h3={kernelName:is,backendName:"webgl",kernelFunc:Wet};var Uet="return (b >= 1.0) ? a : a * (b + 1.0);",Het=`
vec4 bGTEZero = vec4(greaterThanEqual(b, vec4(0.)));
return (bGTEZero * a) + ((vec4(1.0) - bGTEZero) * (a * (b + vec4(1.0))));
`,qet=r=>{let{inputs:t,backend:e}=r,{dy:n,y:o}=t,s=V().getBool("WEBGL_PACK_BINARY_OPERATIONS")?new ci(Het,n.shape,o.shape):new Oo(Uet,n.shape,o.shape);return e.runWebGLProgram(s,[n,o],n.dtype)},g3={kernelName:vp,backendName:"webgl",kernelFunc:qet};var Ket=`
return vec4(equal(a, b));
`,jet="return float(a == b);",Xet=le({opSnippet:jet,packedOpSnippet:Ket,dtype:"bool",cpuKernelImpl:YP}),x3={kernelName:aa,backendName:"webgl",kernelFunc:Xet};var Yet=`
// Error function is calculated approximately with elementary function.
// See "Handbook of Mathematical Functions with Formulas,
// Graphs, and Mathematical Tables", Abramowitz and Stegun.
float p = ${S.ERF_P};
float a1 = ${S.ERF_A1};
float a2 = ${S.ERF_A2};
float a3 = ${S.ERF_A3};
float a4 = ${S.ERF_A4};
float a5 = ${S.ERF_A5};
float sign = sign(x);
x = abs(x);
float t = 1.0 / (1.0 + p * x);
return sign * (1.0 - (((((a5*t + a4)*t) + a3)*t + a2)*t + a1)*t*exp(-x*x));
`,Zet=Ct({opSnippet:Yet}),y3={kernelName:ia,backendName:"webgl",kernelFunc:Zet};var Jet=Po+`
return exp(x);
`,Qet=`
vec4 result = exp(x);
bvec4 isNaN = isnan(x);
result.r = isNaN.r ? x.r : result.r;
result.g = isNaN.g ? x.g : result.g;
result.b = isNaN.b ? x.b : result.b;
result.a = isNaN.a ? x.a : result.a;
return result;
`,IN=Ct({opSnippet:Jet,packedOpSnippet:Qet,cpuKernelImpl:ZP,dtype:"float32"}),b3={kernelName:as,backendName:"webgl",kernelFunc:IN};function Ob(r){let{inputs:t,attrs:e,backend:n}=r,{dim:o}=e,{input:s}=t,i=s.shape.length,a=s.shape.slice(),u=o;return o<0&&(x.assert(-(i+1)<=o,()=>`Axis must be in the interval [${-(i+1)}, ${i}]`),u=i+o+1),a.splice(u,0,1),ut({inputs:{x:s},backend:n,attrs:{shape:a}})}var w3={kernelName:yi,backendName:"webgl",kernelFunc:Ob};var v3="return exp(x) - 1.0;",trt=Ct({opSnippet:v3,packedOpSnippet:v3,cpuKernelImpl:JP}),C3={kernelName:la,backendName:"webgl",kernelFunc:trt};var Pb=class{constructor(t,e,n){this.variableNames=["real","imag"];let o=e[1];this.outputShape=e;let s=n?`2.0 * ${Math.PI}`:`-2.0 * ${Math.PI}`,i=n?`${o}.0`:"1.0",a;if(t==="real")a="return real * expR - imag * expI;";else if(t==="imag")a="return real * expI + imag * expR;";else throw new Error(`FFT component must be either "real" or "imag", got ${t}.`);this.userCode=`
const float exponentMultiplier = ${s};
float unaryOpComplex(float real, float expR, float imag, float expI) {
${a}
}
float mulMatDFT(int batch, int index) {
float indexRatio = float(index) / float(${o});
float exponentMultiplierTimesIndexRatio =
exponentMultiplier * indexRatio;
float result = 0.0;
for (int i = 0; i < ${o}; 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) / ${i};
}
return result;
}
void main() {
ivec2 coords = getOutputCoords();
setOutput(mulMatDFT(coords[0], coords[1]));
}
`}};function Mb(r,t,e){let n=e.texData.get(r.dataId),o=x.sizeFromShape(r.shape),s=r.shape[r.shape.length-1],i=o/s,a=ut({inputs:{x:r},backend:e,attrs:{shape:[i,s]}}),u=a.shape,l=new Pb("real",u,t),c=new Pb("imag",u,t),p=[{dataId:n.complexTensorInfos.real.dataId,dtype:n.complexTensorInfos.real.dtype,shape:u},{dataId:n.complexTensorInfos.imag.dataId,dtype:n.complexTensorInfos.imag.dtype,shape:u}],m=e.runWebGLProgram(l,p,"float32"),f=e.runWebGLProgram(c,p,"float32"),d=Dn({inputs:{real:m,imag:f},backend:e});e.disposeIntermediateTensorInfo(m),e.disposeIntermediateTensorInfo(f);let h=ut({inputs:{x:d},backend:e,attrs:{shape:r.shape}});return e.disposeIntermediateTensorInfo(a),e.disposeIntermediateTensorInfo(d),h}function ert(r){let{inputs:t,backend:e}=r,{input:n}=t;return Mb(n,!1,e)}var I3={kernelName:Cp,backendName:"webgl",kernelFunc:ert};var SN=class{constructor(t,e){this.outputShape=[],this.customUniforms=[{name:"value",type:"float"}],this.variableNames=["x"],this.outputShape=t,this.userCode=`
void main() {
// Input can be obtained from uniform value.
setOutput(value);
}
`}};function dl(r){let{backend:t,attrs:e}=r,{shape:n,value:o}=e,{dtype:s}=e;if(s=s||x.inferDtype(o),s==="string"){let i=x.getArrayFromDType(s,x.sizeFromShape(n));return i.fill(o),t.makeTensorInfo(n,s,i)}else{let i=new SN(n,o),a=[[o]];return t.runWebGLProgram(i,[],s,a)}}var S3={kernelName:Nl,backendName:"webgl",kernelFunc:dl};var kN=class{constructor(t){this.variableNames=["Image"],this.outputShape=[];let e=t[2];this.outputShape=t,this.userCode=`
void main() {
ivec4 coords = getOutputCoords();
int x = coords[2];
int coordX = ${e} - x - 1;
float outputValue;
if(coordX >= 0 && coordX < ${e}) {
outputValue = getImage(coords[0], coords[1], coordX, coords[3]);
} else {
outputValue = getImage(coords[0], coords[1], coords[2], coords[3]);
}
setOutput(outputValue);
}
`}};var k3={kernelName:ua,backendName:"webgl",kernelFunc:({inputs:r,backend:t})=>{let{image:e}=r,n=t,o=new kN(e.shape);return n.runWebGLProgram(o,[e],e.dtype)}};var N3="return floor(x);",rrt=Ct({opSnippet:N3,packedOpSnippet:N3,cpuKernelImpl:QP}),T3={kernelName:ls,backendName:"webgl",kernelFunc:rrt};var nrt=`
float s = sign(a) * sign(b);
int ia = round(a);
int ib = round(b);
if (ib != 0) {
// Windows (D3D) wants guaranteed non-zero int division at compile-time.
return float(idiv(ia, ib, s));
} else {
return NAN;
}
`,ort=`
ivec4 ia = round(a);
ivec4 ib = round(b);
bvec4 cond = notEqual(ib, ivec4(0));
ivec4 result = ivec4(0);
vec4 s = sign(a) * sign(b);
// Windows (D3D) wants guaranteed non-zero int division at compile-time.
if (cond[0]) {
result[0] = idiv(ia[0], ib[0], s[0]);
}
if (cond[1]) {
result[1] = idiv(ia[1], ib[1], s[1]);
}
if (cond[2]) {
result[2] = idiv(ia[2], ib[2], s[2]);
}
if (cond[3]) {
result[3] = idiv(ia[3], ib[3], s[3]);
}
return vec4(result);
`,srt=le({opSnippet:nrt,packedOpSnippet:ort,dtype:"int32"}),_3={kernelName:us,backendName:"webgl",kernelFunc:srt};var NN=class{constructor(t){this.variableNames=["A"];let e=qe(),[n,o]=t;this.outputShape=t,this.userCode=`
void main() {
ivec3 coords = getOutputCoords();
int texR = coords[0];
int texC = coords[1];
int depth = coords[2];
vec2 uv = (vec2(texC, texR) + halfCR) / vec2(${o}.0, ${n}.0);
vec4 values = ${e.texture2D}(A, uv);
float value;
if (depth == 0) {
value = values.r;
} else if (depth == 1) {
value = values.g;
} else if (depth == 2) {
value = values.b;
} else if (depth == 3) {
value = values.a;
}
setOutput(floor(value * 255.0 + 0.5));
}
`}};var TN=class{constructor(t){this.variableNames=["A"],this.packedInputs=!1,this.packedOutput=!0;let e=qe(),[n,o]=t;this.outputShape=t,this.userCode=`
void main() {
ivec3 coords = getOutputCoords();
int texR = coords[0];
int texC = coords[1];
int depth = coords[2];
vec4 result = vec4(0.);
for(int row=0; row<=1; row++) {
for(int col=0; col<=1; col++) {
texC = coords[1] + row;
depth = coords[2] + col;
vec2 uv = (vec2(texC, texR) + halfCR) /
vec2(${o}.0, ${n}.0);
vec4 values = ${e.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);
}
}
${e.output} = result;
}
`}};var E3={kernelName:wf,backendName:"webgl",kernelFunc:irt},Qm;function irt(r){let{inputs:t,backend:e,attrs:n}=r,{pixels:o}=t,{numChannels:s}=n,i=typeof HTMLVideoElement!="undefined"&&o instanceof HTMLVideoElement,a=typeof HTMLImageElement!="undefined"&&o instanceof HTMLImageElement,[u,l]=i?[o.videoWidth,o.videoHeight]:[o.width,o.height],c=[l,u],p=[l,u,s];(a||i)&&(Qm==null&&(Qm=document.createElement("canvas").getContext("2d")),Qm.canvas.width=u,Qm.canvas.height=l,Qm.drawImage(o,0,0,u,l),o=Qm.canvas);let m=e.makeTensorInfo(c,"int32");e.texData.get(m.dataId).usage=Ur.PIXELS,e.gpgpu.uploadPixelDataToTexture(e.getTexture(m.dataId),o);let f=V().getBool("WEBGL_PACK")?new TN(p):new NN(p),d=e.runWebGLProgram(f,[m],"int32");return e.disposeData(m.dataId),d}function art(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,filter:s,bias:i,preluActivationWeights:a}=t,{strides:u,pad:l,dataFormat:c,dilations:p,dimRoundingMode:m,activation:f,leakyreluAlpha:d}=n,h=S.convertConv2DDataFormat(c),g=S.computeConv2DInfo(o.shape,s.shape,u,p,l,m,!1,h),y,b=[];if(g.filterHeight===1&&g.filterWidth===1&&g.dilationHeight===1&&g.dilationWidth===1&&g.strideHeight===1&&g.strideWidth===1&&(g.padInfo.type==="SAME"||g.padInfo.type==="VALID"))y=Db({x:o,filter:s,convInfo:g,backend:e,bias:i,activation:f,preluActivationWeights:a,leakyreluAlpha:d});else if(V().getBool("WEBGL_CONV_IM2COL")&&o.shape[0]===1)y=$b({x:o,filter:s,convInfo:g,backend:e,bias:i,activation:f,preluActivationWeights:a,leakyreluAlpha:d});else{let v=i!=null,k=a!=null,E=f==="leakyrelu",D=f?au(f,!1):null,$=new mg(g,v,D,k,E),R=[o,s];if(i&&R.push(i),a&&R.push(a),E){let M=e.makeTensorInfo([],"float32",x.createScalarValue(d,"float32"));R.push(M),b.push(M)}y=e.runWebGLProgram($,R,"float32")}let w=ut({inputs:{x:y},backend:e,attrs:{shape:g.outShape}});return b.push(y),b.forEach(v=>e.disposeIntermediateTensorInfo(v)),w}var A3={kernelName:Di,backendName:"webgl",kernelFunc:art};function lrt(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,filter:s,bias:i,preluActivationWeights:a}=t,{strides:u,pad:l,dilations:c,dimRoundingMode:p,activation:m,leakyreluAlpha:f}=n,d=[],h=c;h==null&&(h=[1,1]),x.assert(S.eitherStridesOrDilationsAreOne(u,h),()=>`Error in depthwiseConv2d: Either strides or dilations must be 1. Got strides ${u} and dilations '${h}'`);let g=S.computeConv2DInfo(o.shape,s.shape,u,h,l,p,!0),y=V().getBool("WEBGL_PACK_DEPTHWISECONV")&&g.strideWidth<=2&&g.outChannels/g.inChannels===1,b=m?au(m,y):null,w=[o,s],v=i!=null,k=a!=null,E=m==="leakyrelu";if(v&&w.push(i),k&&w.push(a),E){let M=e.makeTensorInfo([],"float32",x.createScalarValue(f,"float32"));w.push(M),d.push(M)}let D;y?D=new dg(g,v,b,k,E):D=new fg(g,v,b,k,E);let $=[[g.padInfo.top,g.padInfo.left],[g.strideHeight,g.strideWidth],[g.dilationHeight,g.dilationWidth],[g.inHeight,g.inWidth]],R=e.runWebGLProgram(D,w,"float32",$);return d.forEach(M=>e.disposeIntermediateTensorInfo(M)),R}var D3={kernelName:$i,backendName:"webgl",kernelFunc:lrt};var _N=class{constructor(t,e,n){this.sliceDim=t,this.strides=e,this.variableNames=["x","indices"],this.outputShape=n;let o=zt(e.length),s=zt(n.length),i=this.sliceDim>1?"strides[j]":"strides";this.userCode=`
${o} strides = ${o}(${this.strides});
void main() {
${s} coords = getOutputCoords();
int flattenIndex = 0;
for (int j = 0; j < ${this.sliceDim}; j++) {
int index = round(getIndices(coords[0], j));
flattenIndex += index * ${i};
}
setOutput(getX(flattenIndex, coords[1]));
}
`}};function urt(r){let{inputs:t,backend:e}=r,{params:n,indices:o}=t,s=o.shape,i=s[s.length-1],a=x.sizeFromShape(n.shape),[u,l,c,p]=S.prepareAndValidate(n,o),m=ut({inputs:{x:o},backend:e,attrs:{shape:[l,i]}}),f=ut({inputs:{x:n},backend:e,attrs:{shape:[x.sizeFromShape(n.shape)/c,c]}});if(e.shouldExecuteOnCPU([n,o])||n.dtype==="string"){let y=e.readSync(o.dataId),b=e.bufferSync(n),w=tM(y,b,n.dtype,l,i,c,p,n.shape,a);return e.makeTensorInfo(u,n.dtype,w.values)}let d=new _N(i,p,[l,c]),h=e.runWebGLProgram(d,[f,m],f.dtype),g=ut({inputs:{x:h},backend:e,attrs:{shape:u}});return e.disposeIntermediateTensorInfo(m),e.disposeIntermediateTensorInfo(f),e.disposeIntermediateTensorInfo(h),g}var $3={kernelName:ca,backendName:"webgl",kernelFunc:urt};var EN=class{constructor(t,e){this.variableNames=["A","indices"],this.outputShape=e,this.rank=e.length;let n=zt(this.rank),o=crt(t,2);this.userCode=`
void main() {
${n} resRC = getOutputCoords();
int index = int(getIndices(resRC.x, resRC.z));
float inBounds = (index >= 0) && (index < ${t[2]}) ? 1.0 : 0.0;
setOutput(inBounds * getA(${o}));
}
`}};function crt(r,t){let e=["resRC.x","resRC.y","resRC.z","resRC.w"],n=[];for(let o=0;o<r.length;o++)o===2?n.push("index"):n.push(`${e[o]}`);return n.join()}function AN(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,indices:s}=t,{axis:i,batchDims:a}=n,u=x.parseAxisParam(i,o.shape)[0];if(V().get("DEBUG")){let b=e.readSync(s.dataId),w=o.shape[u];for(let v=0;v<b.length;++v){let k=b[v];x.assert(k<=w-1&&k>=0,()=>`GatherV2: the index value ${k} is not in [0, ${w-1}]`)}}let l=S.segment_util.collectGatherOpShapeInfo(o,s,u,a),c=x.sizeFromShape(s.shape),p=[],m=ut({inputs:{x:o},backend:e,attrs:{shape:[l.batchSize,l.outerSize,l.dimSize,l.sliceSize]}}),f=ut({inputs:{x:s},backend:e,attrs:{shape:[l.batchSize,c/l.batchSize]}});p.push(m),p.push(f);let d=[l.batchSize,l.outerSize,c/l.batchSize,l.sliceSize];if(e.shouldExecuteOnCPU([o,s])||o.dtype==="string"){let b=e.bufferSync(f),w=e.bufferSync(m),v=eM(w,b,d);return p.forEach(k=>e.disposeIntermediateTensorInfo(k)),e.makeTensorInfo(l.outputShape,v.dtype,v.values)}let h=new EN(m.shape,d),g=e.runWebGLProgram(h,[m,f],m.dtype);p.push(g);let y=ut({inputs:{x:g},backend:e,attrs:{shape:l.outputShape}});return p.forEach(b=>e.disposeIntermediateTensorInfo(b)),y}var R3={kernelName:bi,backendName:"webgl",kernelFunc:AN};var prt="return float(a > b);",mrt=`
return vec4(greaterThan(a, b));
`,frt=le({opSnippet:prt,packedOpSnippet:mrt,cpuKernelImpl:rM,dtype:"bool"}),F3={kernelName:pa,backendName:"webgl",kernelFunc:frt};var drt="return float(a >= b);",hrt=`
return vec4(greaterThanEqual(a, b));
`,grt=le({opSnippet:drt,packedOpSnippet:hrt,dtype:"bool",cpuKernelImpl:nM}),O3={kernelName:ps,backendName:"webgl",kernelFunc:grt};function xrt(r){let{inputs:t,backend:e}=r,{input:n}=t;return Mb(n,!0,e)}var P3={kernelName:Ip,backendName:"webgl",kernelFunc:xrt};var yrt="return float(!isnan(x) && !isinf(x));",brt=Ct({opSnippet:yrt,dtype:"bool"}),M3={kernelName:ma,backendName:"webgl",kernelFunc:brt};var wrt="return float(isinf(x));",vrt=Ct({opSnippet:wrt,dtype:"bool"}),L3={kernelName:fa,backendName:"webgl",kernelFunc:vrt};var Crt="return float(isnan(x));",Irt=Ct({opSnippet:Crt,dtype:"bool"}),z3={kernelName:da,backendName:"webgl",kernelFunc:Irt};var Srt="return float(a < b);",krt=`
return vec4(lessThan(a, b));
`,Nrt=le({opSnippet:Srt,packedOpSnippet:krt,cpuKernelImpl:oM,dtype:"bool"}),B3={kernelName:ha,backendName:"webgl",kernelFunc:Nrt};var Trt="return float(a <= b);",_rt=`
return vec4(lessThanEqual(a, b));
`,Ert=le({opSnippet:Trt,packedOpSnippet:_rt,cpuKernelImpl:sM,dtype:"bool"}),V3={kernelName:ga,backendName:"webgl",kernelFunc:Ert};function Art(r){let{backend:t,attrs:e}=r,{start:n,stop:o,num:s}=e,i=iM(n,o,s);return t.makeTensorInfo([i.length],"float32",i)}var G3={kernelName:kp,backendName:"webgl",kernelFunc:Art};var Drt=Po+`
return x < 0.0 ? 0./0. : log(x);
`,$rt=`
vec4 result = log(x);
bvec4 isNaN = isnan(x);
result.r = isNaN.r ? x.r : (x.r < 0.0 ? 0./0. : result.r);
result.g = isNaN.g ? x.g : (x.g < 0.0 ? 0./0. : result.g);
result.b = isNaN.b ? x.b : (x.b < 0.0 ? 0./0. : result.b);
result.a = isNaN.a ? x.a : (x.a < 0.0 ? 0./0. : result.a);
return result;
`,Rrt=Ct({opSnippet:Drt,packedOpSnippet:$rt,cpuKernelImpl:aM}),W3={kernelName:fs,backendName:"webgl",kernelFunc:Rrt};var Frt=Po+`
return log(1.0 + x);
`,Ort=Ct({opSnippet:Frt}),U3={kernelName:xa,backendName:"webgl",kernelFunc:Ort};var Prt="return float(a >= 1.0 && b >= 1.0);",Mrt=`
return vec4(
vec4(greaterThanEqual(a, vec4(1.0))) *
vec4(greaterThanEqual(b, vec4(1.0))));
`,Lrt=le({opSnippet:Prt,packedOpSnippet:Mrt,dtype:"bool"}),H3={kernelName:ya,backendName:"webgl",kernelFunc:Lrt};var zrt="return float(!(x >= 1.0));",Brt=Ct({opSnippet:zrt}),q3={kernelName:wu,backendName:"webgl",kernelFunc:Brt};var Vrt="return float(a >= 1.0 || b >= 1.0);",Grt=`
return min(
vec4(greaterThanEqual(a, vec4(1.0))) +
vec4(greaterThanEqual(b, vec4(1.0))),
vec4(1.0));
`,Wrt=le({opSnippet:Vrt,packedOpSnippet:Grt,dtype:"bool"}),K3={kernelName:vu,backendName:"webgl",kernelFunc:Wrt};var DN=class{constructor(t,e,n,o,s){this.variableNames=["x"],this.outputShape=[];let i=e,a=t[3]-1;this.outputShape=t;let u,l=`float(${n}) + float(${o}) * sum`;s===.5?u=`inversesqrt(${l})`:s===1?u=`1.0/(${l})`:u=`exp(log(${l}) * float(-${s}));`,this.userCode=`
void main() {
ivec4 coords = getOutputCoords();
int b = coords[0];
int r = coords[1];
int c = coords[2];
int d = coords[3];
float x = getX(b, r, c, d);
float sum = 0.0;
for (int j = -${i}; j <= ${i}; j++) {
int idx = d + j;
if (idx >= 0 && idx <= ${a}) {
float z = getX(b, r, c, idx);
sum += z * z;
}
}
float val = x * ${u};
setOutput(val);
}
`}};var $N=class{constructor(t,e,n,o,s){this.variableNames=["x"],this.outputShape=[],this.packedInputs=!0,this.packedOutput=!0;let i=e,a=t[3]-1;this.outputShape=t;let u,l=`float(${n}) + float(${o}) * sum`;s===.5?u=`inversesqrt(${l})`:s===1?u=`1.0/(${l})`:u=`exp(log(${l}) * float(-${s}));`,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 - ${i};
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 = - ${i}; j <= ${i}; j++) {
ivec2 idx = depth + j;
bvec2 aboveLowerBound = greaterThanEqual(idx, ivec2(0));
bvec2 belowUpperBound = lessThanEqual(idx, ivec2(${a}));
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 * ${u};
setOutput(result);
}
`}};var Urt=r=>{let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{depthRadius:s,bias:i,alpha:a,beta:u}=n,l=V().getBool("WEBGL_PACK_NORMALIZATION")?new $N(o.shape,s,i,a,u):new DN(o.shape,s,i,a,u);return e.runWebGLProgram(l,[o],o.dtype)},j3={kernelName:Tl,backendName:"webgl",kernelFunc:Urt};var RN=class{constructor(t,e,n,o,s){this.variableNames=["inputImage","outputImage","dy"],this.outputShape=[],this.outputShape=t,this.depth=t[3],this.depthRadius=e,this.bias=n,this.alpha=o,this.beta=s,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 - ${e})));
int depthEnd = int(min(float(${this.depth}),
float(d + ${e} + 1)));
const int MIN_DEPTH_BEGIN = 0;
const int MAX_DEPTH_END = ${this.depth};
float norm = 0.0;
for (int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k) {
if (k < depthBegin){
continue;
}
else if (k >= depthBegin && k < depthEnd) {
norm += getInputImage(b, r, c, k) * getInputImage(b, r, c, k);
}
else {
break;
}
}
norm = float(${o}) * norm + float(${n});
for(int k = MIN_DEPTH_BEGIN; k < MAX_DEPTH_END; ++k){
if (k < depthBegin){
continue;
}
else if (k >= depthBegin && k < depthEnd){
float dyi = -2.0 * float(${o})
* float(${s})
* getInputImage(b ,r ,c, k) * getOutputImage(b, r, c, d)
/ norm;
if (k == d) {
dyi += pow(norm, -1.0 * ${s});
}
if (k == coords[3]) {
dyi *= getDy(b, r, c, d);
result += dyi;
}
}
else {
break;
}
}
}
setOutput(result);
}
`}};var Hrt=r=>{let{inputs:t,backend:e,attrs:n}=r,{x:o,y:s,dy:i}=t,{depthRadius:a,bias:u,alpha:l,beta:c}=n,p=new RN(o.shape,a,u,l,c);return e.runWebGLProgram(p,[o,s,i],o.dtype)},X3={kernelName:Np,backendName:"webgl",kernelFunc:Hrt};function Y3(r,t,e,n){let o=x.sizeFromShape(t),i=x.sizeFromShape(r.shape)/o,a=ut({inputs:{x:r},attrs:{shape:[i,o]},backend:n}),u=Wn(a,r.dtype,"max",n),l=ut({inputs:{x:u},attrs:{shape:e},backend:n});return n.disposeIntermediateTensorInfo(a),n.disposeIntermediateTensorInfo(u),l}function FN(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{reductionIndices:s,keepDims:i}=n,a=o.shape.length,u=x.parseAxisParam(s,o.shape),l=u,c=S.getAxesPermutation(l,a),p=c!=null,m=e.shouldExecuteOnCPU([o]),f=o;if(p){if(m){let w=e.texData.get(f.dataId).values,v=new Array(a);for(let D=0;D<v.length;D++)v[D]=o.shape[c[D]];let k=Rc(w,o.shape,o.dtype,c,v);f=e.makeTensorInfo(v,o.dtype);let E=e.texData.get(f.dataId);E.values=k}else f=lu(o,c,e);l=S.getInnerMostAxes(l.length,a)}S.assertAxesAreInnerMostDims("max",l,a);let[d,h]=S.computeOutAndReduceShapes(f.shape,l),g=d;i&&(g=S.expandShapeToKeepDim(d,u));let y;if(m){let w=e.texData.get(f.dataId).values,v=lM(w,x.sizeFromShape(h),g,o.dtype);y=e.makeTensorInfo(g,o.dtype);let k=e.texData.get(y.dataId);k.values=v}else y=Y3(f,h,g,e);return p&&e.disposeIntermediateTensorInfo(f),y}var Z3={kernelName:ds,backendName:"webgl",kernelFunc:FN};var qrt=kb+`
return max(a, b);
`,Krt=`
vec4 result = vec4(max(a, b));
vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));
`+iu+`
return result;
`,jrt=le({opSnippet:qrt,packedOpSnippet:Krt,cpuKernelImpl:uM}),J3={kernelName:hs,backendName:"webgl",kernelFunc:jrt};function Xrt(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t;li(o,"maxPool");let{filterSize:s,strides:i,pad:a,dimRoundingMode:u}=n,l=1;x.assert(S.eitherStridesOrDilationsAreOne(i,l),()=>`Error in maxPool: Either strides or dilations must be 1. Got strides ${i} and dilations '${l}'`);let c=S.computePool2DInfo(o.shape,s,i,l,a,u);if(c.filterWidth===1&&c.filterHeight===1&&x.arraysEqual(c.inShape,c.outShape))return Ke({inputs:{x:o},backend:e});let p=new qi(c,"max",!1);return e.runWebGLProgram(p,[o],o.dtype)}var Q3={kernelName:gs,backendName:"webgl",kernelFunc:Xrt};function Yrt(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{filterSize:s,strides:i,pad:a,dataFormat:u,dimRoundingMode:l}=n,c=[1,1,1],p=S.computePool3DInfo(o.shape,s,i,c,a,l,u),m=new Lc(p,"max",!1);return e.runWebGLProgram(m,[o],o.dtype)}var tz={kernelName:_l,backendName:"webgl",kernelFunc:Yrt};var ON=class{constructor(t){this.variableNames=["dy","maxPos"],this.outputShape=t.inShape;let e=t.strideHeight,n=t.strideWidth,o=t.dilationHeight,s=t.effectiveFilterHeight,i=t.effectiveFilterWidth,a=s-1-t.padInfo.top,u=i-1-t.padInfo.left,l=s*i-1;this.userCode=`
const ivec2 pads = ivec2(${a}, ${u});
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 < ${s};
wR += ${o}) {
float dyR = float(dyRCorner + wR) / ${e}.0;
if (dyR < 0.0 || dyR >= ${t.outHeight}.0 || fract(dyR) > 0.0) {
continue;
}
int idyR = int(dyR);
for (int wC = 0; wC < ${i}; wC++) {
float dyC = float(dyCCorner + wC) / ${n}.0;
if (dyC < 0.0 || dyC >= ${t.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 * ${i} + wC;
float mask = float(maxPosValue == curPosValue ? 1.0 : 0.0);
dotProd += dyValue * mask;
}
}
setOutput(dotProd);
}
`}},PN=class{constructor(t){this.variableNames=["dy","maxPos"],this.outputShape=t.inShape;let e=t.strideDepth,n=t.strideHeight,o=t.strideWidth,s=t.dilationDepth,i=t.dilationHeight,a=t.dilationWidth,u=t.effectiveFilterDepth,l=t.effectiveFilterHeight,c=t.effectiveFilterWidth,p=u-1-t.padInfo.front,m=l-1-t.padInfo.top,f=c-1-t.padInfo.left,d=u*l*c-1;this.userCode=`
const ivec3 pads = ivec3(${p}, ${m}, ${f});
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 < ${u};
wD += ${s}) {
float dyD = float(dyDCorner + wD) / ${e}.0;
if (dyD < 0.0 || dyD >= ${t.outDepth}.0 || fract(dyD) > 0.0) {
continue;
}
int idyD = int(dyD);
for (int wR = 0; wR < ${l};
wR += ${i}) {
float dyR = float(dyRCorner + wR) / ${n}.0;
if (dyR < 0.0 || dyR >= ${t.outHeight}.0 ||
fract(dyR) > 0.0) {
continue;
}
int idyR = int(dyR);
for (int wC = 0; wC < ${c};
wC += ${a}) {
float dyC = float(dyCCorner + wC) / ${o}.0;
if (dyC < 0.0 || dyC >= ${t.outWidth}.0 ||
fract(dyC) > 0.0) {
continue;
}
int idyC = int(dyC);
float dyValue = getDy(batch, idyD, idyR, idyC, ch);
int maxPosValue = ${d} -
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 Zrt(r){let{inputs:t,backend:e,attrs:n}=r,{dy:o,input:s}=t,i=s,{filterSize:a,strides:u,pad:l,dimRoundingMode:c}=n,p=[1,1,1],m=S.computePool3DInfo(i.shape,a,u,p,l,c),f=new Lc(m,"max",!0),d=e.runWebGLProgram(f,[i],i.dtype),h=new PN(m),g=e.runWebGLProgram(h,[o,d],i.dtype);return e.disposeIntermediateTensorInfo(d),g}var ez={kernelName:_p,backendName:"webgl",kernelFunc:Zrt};function Jrt(r){let{inputs:t,backend:e,attrs:n}=r,{dy:o,input:s,output:i}=t,a=s;li([s,i],"maxPoolGrad");let{filterSize:u,strides:l,pad:c,dimRoundingMode:p}=n,m=S.computePool2DInfo(a.shape,u,l,1,c,p),f=!0,d=new qi(m,"max",f),h=e.runWebGLProgram(d,[a],a.dtype),g=new ON(m),y=e.runWebGLProgram(g,[o,h],a.dtype);return e.disposeIntermediateTensorInfo(h),y}var rz={kernelName:Tp,backendName:"webgl",kernelFunc:Jrt};function nz(r,t,e,n){let o=new qi(e,"max",!1),s=n.runWebGLProgram(o,[r],"float32");o=new qi(e,"max",!0,!0,t);let i=n.runWebGLProgram(o,[r],"float32");return[s,i]}var oz={kernelName:Ep,backendName:"webgl",kernelFunc:({inputs:r,attrs:t,backend:e})=>{let{x:n}=r,{filterSize:o,strides:s,pad:i,includeBatchInIndex:a}=t,u=e;x.assert(n.shape.length===4,()=>`Error in maxPool: input must be rank 4 but got rank ${n.shape.length}.`);let l=[1,1];x.assert(S.eitherStridesOrDilationsAreOne(s,l),()=>`Error in maxPool: Either strides or dilations must be 1. Got strides ${s} and dilations '${l}'`);let c=S.computePool2DInfo(n.shape,o,s,l,i),[p,m]=nz(n,a,c,u);return[p,m]}};function sz(r,t,e,n){let o=x.sizeFromShape(t),i=x.sizeFromShape(r.shape)/o,a=ut({inputs:{x:r},attrs:{shape:[i,o]},backend:n}),u=Wn(a,"float32","mean",n),l=ut({inputs:{x:u},attrs:{shape:e},backend:n});return n.disposeIntermediateTensorInfo(a),n.disposeIntermediateTensorInfo(u),l}var iz={kernelName:xs,backendName:"webgl",kernelFunc:({inputs:r,attrs:t,backend:e})=>{let{x:n}=r,{keepDims:o,axis:s}=t,i=e,a=n.shape.length,u=x.parseAxisParam(s,n.shape),l=u,c=S.getAxesPermutation(l,a),p=c!=null,m=i.shouldExecuteOnCPU([n]),f=[],d=n;if(p){if(m){let v=i.texData.get(d.dataId).values,k=new Array(a);for(let $=0;$<k.length;$++)k[$]=n.shape[c[$]];let E=Rc(v,n.shape,n.dtype,c,k);d=i.makeTensorInfo(k,n.dtype);let D=i.texData.get(d.dataId);D.values=E}else d=lu(n,c,i);f.push(d),l=S.getInnerMostAxes(l.length,a)}S.assertAxesAreInnerMostDims("sum",l,a);let[h,g]=S.computeOutAndReduceShapes(d.shape,l),y=h;o&&(y=S.expandShapeToKeepDim(h,u));let b=sz(d,g,y,i);for(let w of f)i.disposeIntermediateTensorInfo(w);return b}};function Qrt(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s,keepDims:i}=n,a=o.shape.length,u=x.parseAxisParam(s,o.shape),l=u,c=S.getAxesPermutation(l,a),p=o;c!=null&&(p=Ee({inputs:{x:o},backend:e,attrs:{perm:c}}),l=S.getInnerMostAxes(l.length,o.shape.length)),S.assertAxesAreInnerMostDims("min",l,a);let[m,f]=S.computeOutAndReduceShapes(p.shape,l),d=x.sizeFromShape(f),h=ut({inputs:{x:p},backend:e,attrs:{shape:[-1,d]}}),g=Wn(h,h.dtype,"min",e),y;if(i){let b=S.expandShapeToKeepDim(m,u);y=ut({inputs:{x:g},backend:e,attrs:{shape:b}})}else y=ut({inputs:{x:g},backend:e,attrs:{shape:m}});return e.disposeIntermediateTensorInfo(h),e.disposeIntermediateTensorInfo(g),c!=null&&e.disposeIntermediateTensorInfo(p),y}var az={kernelName:ys,backendName:"webgl",kernelFunc:Qrt};var tnt=kb+`
return min(a, b);
`,ent=`
vec4 result = vec4(min(a, b));
vec4 isNaN = min(vec4(isnan(a)) + vec4(isnan(b)), vec4(1.0));
`+iu+`
return result;
`,rnt=le({opSnippet:tnt,packedOpSnippet:ent,cpuKernelImpl:cM}),lz={kernelName:bs,backendName:"webgl",kernelFunc:rnt};var MN=class{constructor(t,e,n){this.variableNames=["x"],this.outputShape=e.map((c,p)=>c[0]+t[p]+c[1]);let o=t.length,s=zt(o),i=e.map(c=>c[0]).join(","),a=e.map((c,p)=>c[0]+t[p]).join(","),u=["coords[0]","coords[1]","coords[2]","coords[3]"].slice(0,o),l=n==="reflect"?0:1;if(o===1){this.userCode=`
int start = ${i};
int end = ${a};
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=`
${s} start = ${s}(${i});
${s} end = ${s}(${a});
void main() {
${s} outC = getOutputCoords();
for (int i = 0; i < ${o}; 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};
}
}
${s} coords = outC - start;
setOutput(getX(${u}));
}
`}};var LN=class{constructor(t,e,n){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=e.map((d,h)=>d[0]+t[h]+d[1]);let o=t.length,s=zt(o),i=e.map(d=>d[0]).join(","),a=e.map((d,h)=>d[0]+t[h]).join(","),u=rr("rc",o),l=rr("source",o),c=`${u[o-1]} < ${this.outputShape[o-1]}`,p=o===1?"source":`vec2(${l.slice(-2).join()})`,m=n==="reflect"?0:1,f="";if(o===1){let d=`
${s} source = rc;
if (source < start) {
source = start * 2 - source - ${m};
} else if (source >= end) {
source = (end - 1) * 2 - source + ${m};
}
source -= start;
`;f=`
${s} rc = outputLoc;
${d}
result[0] = getChannel(getX(${l.join()}), ${p});
${u[o-1]} += 1;
if(${c}) {
${d}
result[1] = getChannel(getX(${l.join()}), ${p});
}
`}else{let d=`
${s} source = rc;
${s} lt = ${s}(lessThan(source, start));
${s} gte = ${s}(greaterThanEqual(source, end));
${s} orig = 1 - (lt + gte);
source = orig * source +
lt * (start * 2 - source - ${m}) +
gte * ((end - 1) * 2 - source + ${m});
source -= start;
`;f=`
${s} rc = outputLoc;
${d}
result[0] = getChannel(getX(${l.join()}), ${p});
${u[o-1]} += 1;
if(${c}) {
${d}
result[1] = getChannel(getX(${l.join()}), ${p});
}
rc = outputLoc;
${u[o-2]} += 1;
if(${u[o-2]} < ${this.outputShape[o-2]}) {
${d}
result[2] = getChannel(getX(${l.join()}), ${p});
${u[o-1]} += 1;
if(${c}) {
${d}
result[3] = getChannel(getX(${l.join()}), ${p});
}
}
`}this.userCode=`
const ${s} start = ${s}(${i});
const ${s} end = ${s}(${a});
void main() {
${s} outputLoc = getOutputCoords();
vec4 result = vec4(0.);
${f}
setOutput(result);
}
`}};var nnt=({inputs:r,backend:t,attrs:e})=>{let{x:n}=r,{paddings:o,mode:s}=e,i=V().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new LN(n.shape,o,s):new MN(n.shape,o,s);return t.runWebGLProgram(i,[n],n.dtype)},uz={kernelName:ws,backendName:"webgl",kernelFunc:nnt};var ont=`if (b == 0.0) return NAN;
return mod(a, b);`,snt=`
vec4 result = mod(a, b);
vec4 isNaN = vec4(equal(b, vec4(0.0)));
`+iu+`
return result;
`,int=le({opSnippet:ont,packedOpSnippet:snt}),cz={kernelName:ba,backendName:"webgl",kernelFunc:int};var zN=class{constructor(t,e,n){this.variableNames=["probs"],this.customUniforms=[{name:"seed",type:"float"}],this.outputShape=[t,n],this.userCode=`
void main() {
ivec2 coords = getOutputCoords();
int batch = coords[0];
float r = random(seed);
float cdf = 0.0;
for (int i = 0; i < ${e-1}; i++) {
cdf += getProbs(batch, i);
if (r < cdf) {
setOutput(float(i));
return;
}
}
// If no other event happened, last event happened.
setOutput(float(${e-1}));
}
`}};var ant=`
if (a == b) {
return 1.0;
};
return a / b;`,lnt=`
// 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;
`,BN=le({opSnippet:ant,packedOpSnippet:lnt,checkOutOfBounds:!0}),pz={kernelName:ss,backendName:"webgl",kernelFunc:BN};var mz="return a - b;",VN=le({opSnippet:mz,packedOpSnippet:mz,supportsComplex:!0,cpuKernelImpl:NM}),fz={kernelName:Ls,backendName:"webgl",kernelFunc:VN};function GN(r){let{inputs:t,backend:e,attrs:n}=r,{logits:o}=t,{dim:s}=n,i=x.parseAxisParam([s],o.shape),a=FN({inputs:{x:o},backend:e,attrs:{reductionIndices:i,keepDims:!1}}),u=S.expandShapeToKeepDim(a.shape,i),l=ut({inputs:{x:a},backend:e,attrs:{shape:u}}),c=VN({inputs:{a:o,b:l},backend:e}),p=IN({inputs:{x:c},backend:e}),m=Pc({inputs:{x:p},backend:e,attrs:{axis:i,keepDims:!1}}),f=ut({inputs:{x:m},backend:e,attrs:{shape:u}}),d=BN({inputs:{a:p,b:f},backend:e});return e.disposeIntermediateTensorInfo(a),e.disposeIntermediateTensorInfo(l),e.disposeIntermediateTensorInfo(c),e.disposeIntermediateTensorInfo(p),e.disposeIntermediateTensorInfo(m),e.disposeIntermediateTensorInfo(f),d}var dz={kernelName:Ps,backendName:"webgl",kernelFunc:GN};function unt(r){let{inputs:t,backend:e,attrs:n}=r,{logits:o}=t,{numSamples:s,seed:i,normalized:a}=n,u=a?o:GN({inputs:{logits:o},backend:e,attrs:{dim:o.shape.length-1}}),l=u.shape[0],c=u.shape[1],p=new zN(l,c,s),m=[[i]],f=e.runWebGLProgram(p,[u],"int32",m);return a||e.disposeIntermediateTensorInfo(u),f}var hz={kernelName:Ap,backendName:"webgl",kernelFunc:unt};var cnt=hr+`
return -x;
`,pnt=`
vec4 result = -x;
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;
`;function mnt(r){let{inputs:t,backend:e}=r,{x:n}=t;if(e.shouldExecuteOnCPU([n])){let s=e.texData.get(n.dataId),[i,a]=mM(s.values,n.shape,n.dtype);return e.makeTensorInfo(a,n.dtype,i)}let o;return V().getBool("WEBGL_PACK_UNARY_OPERATIONS")?o=new Fo(n.shape,pnt):o=new vn(n.shape,cnt),e.runWebGLProgram(o,[n],n.dtype)}var gz={kernelName:wi,backendName:"webgl",kernelFunc:mnt};var fnt=Vr.nonMaxSuppressionV3Impl;function dnt(r){S.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");let{inputs:t,backend:e,attrs:n}=r,{boxes:o,scores:s}=t,{maxOutputSize:i,iouThreshold:a,scoreThreshold:u}=n,l=e.readSync(o.dataId),c=e.readSync(s.dataId),{selectedIndices:p}=fnt(l,c,i,a,u);return e.makeTensorInfo([p.length],"int32",new Int32Array(p))}var xz={kernelName:va,backendName:"webgl",kernelFunc:dnt};var hnt=Vr.nonMaxSuppressionV4Impl;function gnt(r){S.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");let{inputs:t,backend:e,attrs:n}=r,{boxes:o,scores:s}=t,{maxOutputSize:i,iouThreshold:a,scoreThreshold:u,padToMaxOutputSize:l}=n,c=e.readSync(o.dataId),p=e.readSync(s.dataId),{selectedIndices:m,validOutputs:f}=hnt(c,p,i,a,u,l);return[e.makeTensorInfo([m.length],"int32",new Int32Array(m)),e.makeTensorInfo([],"int32",new Int32Array([f]))]}var yz={kernelName:Ca,backendName:"webgl",kernelFunc:gnt};var xnt=Vr.nonMaxSuppressionV5Impl;function ynt(r){S.warn("tf.nonMaxSuppression() in webgl locks the UI thread. Call tf.nonMaxSuppressionAsync() instead");let{inputs:t,backend:e,attrs:n}=r,{boxes:o,scores:s}=t,{maxOutputSize:i,iouThreshold:a,scoreThreshold:u,softNmsSigma:l}=n,c=e.readSync(o.dataId),p=e.readSync(s.dataId),m=i,f=a,d=u,h=l,{selectedIndices:g,selectedScores:y}=xnt(c,p,m,f,d,h);return[e.makeTensorInfo([g.length],"int32",new Int32Array(g)),e.makeTensorInfo([y.length],"float32",new Float32Array(y))]}var bz={kernelName:Ia,backendName:"webgl",kernelFunc:ynt};var WN=class{constructor(t,e,n,o){this.variableNames=["indices"],this.outputShape=[t,e],this.userCode=`
void main() {
ivec2 coords = getOutputCoords();
int index = round(getIndices(coords.x));
setOutput(mix(float(${o}), float(${n}),
float(index == coords.y)));
}
`}};var bnt=r=>{let{inputs:t,backend:e,attrs:n}=r,{indices:o}=t,{depth:s,onValue:i,offValue:a}=n,u=x.sizeFromShape(o.shape),l=new WN(u,s,i,a),c=ut({inputs:{x:o},backend:e,attrs:{shape:[u]}}),p=e.runWebGLProgram(l,[c],o.dtype);e.disposeIntermediateTensorInfo(c);let m=[...o.shape,s],f=ut({inputs:{x:p},backend:e,attrs:{shape:m}});return e.disposeIntermediateTensorInfo(p),f},wz={kernelName:Cs,backendName:"webgl",kernelFunc:bnt};function hg(r){let{inputs:t,backend:e}=r,{x:n}=t;if(n.dtype==="complex64"){let o=fl({inputs:{input:n},backend:e}),s=hg({inputs:{x:o},backend:e}),i=zc({inputs:{input:n},backend:e}),a=hg({inputs:{x:i},backend:e}),u=Dn({inputs:{real:s,imag:a},backend:e});return e.disposeIntermediateTensorInfo(o),e.disposeIntermediateTensorInfo(s),e.disposeIntermediateTensorInfo(i),e.disposeIntermediateTensorInfo(a),u}else return dl({attrs:{shape:n.shape,dtype:n.dtype,value:n.dtype==="string"?"":0},backend:e})}var vz={kernelName:Ei,backendName:"webgl",kernelFunc:hg};function Cz(r){let{inputs:t,backend:e}=r,{x:n}=t;if(n.dtype==="string")throw new Error("onesLike is not supported under string dtype");if(n.dtype==="complex64"){let o=fl({inputs:{input:n},backend:e}),s=Cz({inputs:{x:o},backend:e}),i=zc({inputs:{input:n},backend:e}),a=hg({inputs:{x:i},backend:e}),u=Dn({inputs:{real:s,imag:a},backend:e});return e.disposeIntermediateTensorInfo(o),e.disposeIntermediateTensorInfo(s),e.disposeIntermediateTensorInfo(i),e.disposeIntermediateTensorInfo(a),u}else return dl({attrs:{shape:n.shape,dtype:n.dtype,value:1},backend:e})}var Iz={kernelName:vi,backendName:"webgl",kernelFunc:Cz};function wnt(r){let{inputs:t,backend:e,attrs:n}=r,{axis:o}=n;if(t.length===1)return Ob({inputs:{input:t[0]},backend:e,attrs:{dim:o}});let s=t[0].shape,i=t[0].dtype;t.forEach(c=>{x.assertShapesMatch(s,c.shape,"All tensors passed to stack must have matching shapes"),x.assert(i===c.dtype,()=>"All tensors passed to stack must have matching dtypes")});let a=[],u=t.map(c=>{let p=Ob({inputs:{input:c},backend:e,attrs:{dim:o}});return a.push(p),p}),l=cN({inputs:u,backend:e,attrs:{axis:o}});return a.forEach(c=>e.disposeIntermediateTensorInfo(c)),l}var Sz={kernelName:Ci,backendName:"webgl",kernelFunc:wnt};var UN=class{constructor(t,e,n){this.variableNames=["x"],this.customUniforms=[{name:"value",type:"float"}],this.outputShape=e.map((l,c)=>l[0]+t[c]+l[1]);let o=t.length,s=zt(o),i=e.map(l=>l[0]).join(","),a=e.map((l,c)=>l[0]+t[c]).join(","),u=["coords[0]","coords[1]","coords[2]","coords[3]"].slice(0,o);if(o===1){this.userCode=`
int start = ${i};
int end = ${a};
void main() {
int outC = getOutputCoords();
if (outC < start || outC >= end) {
setOutput(value);
} else {
setOutput(getX(outC - start));
}
}
`;return}this.userCode=`
${s} start = ${s}(${i});
${s} end = ${s}(${a});
void main() {
${s} outC = getOutputCoords();
if (any(lessThan(outC, start)) || any(greaterThanEqual(outC, end))) {
setOutput(value);
} else {
${s} coords = outC - start;
setOutput(getX(${u}));
}
}
`}};var HN=class{constructor(t,e,n){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0,this.customUniforms=[{name:"value",type:"float"}],this.outputShape=e.map((h,g)=>h[0]+t[g]+h[1]);let o=t.length,s=zt(o),i=e.map(h=>h[0]).join(","),a=e.map((h,g)=>h[0]+t[g]).join(","),u=rr("rc",o),l=rr("source",o),c=`${u[o-1]} < ${this.outputShape[o-1]}`,p=o===1?"source":`vec2(${l.slice(-2).join()})`,m=[`${s} rc = outputLoc;`,`${u[o-1]} += 1;
if(${c}) {
`,o===1?"":`}
rc = outputLoc;
${u[o-2]} += 1;
if(${u[o-2]} < ${this.outputShape[o-2]}) {`,o===1?"":` ${u[o-1]} += 1;
if(${c}) {`],f=o===1?"rc < start || rc >= end":"any(lessThan(rc, start)) || any(greaterThanEqual(rc, end))",d="";for(let h=0,g=o===1?2:4;h<g;h++)d+=`
${m[h]}
if (${f}) {
result[${h}] = float(value);
} else {
${s} source = rc - start;
result[${h}] = getChannel(getX(${l.join()}), ${p});
}
`;d+=o===1?"} ":"}}",this.userCode=`
const ${s} start = ${s}(${i});
const ${s} end = ${s}(${a});
void main() {
${s} outputLoc = getOutputCoords();
vec4 result = vec4(0.);
${d}
setOutput(result);
}
`}};var qN=r=>{let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{paddings:s,constantValue:i}=n;if(x.sizeFromShape(o.shape)===0){let l=s.map((c,p)=>c[0]+o.shape[p]+c[1]);return dl({backend:e,attrs:{shape:l,value:i,dtype:o.dtype}})}let a=V().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new HN(o.shape,s,i):new UN(o.shape,s,i),u=[[i]];return e.runWebGLProgram(a,[o],o.dtype,u)},kz={kernelName:Is,backendName:"webgl",kernelFunc:qN};var vnt=`
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);
`,Cnt=`
// 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));
`+iu+`
return result;
`,Int=le({opSnippet:vnt,packedOpSnippet:Cnt}),Nz={kernelName:Ss,backendName:"webgl",kernelFunc:Int};function Snt(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s,keepDims:i}=n,a=o.shape.length,u=[],l=x.parseAxisParam(s,o.shape),c=l,p=S.getAxesPermutation(c,a),m=o;p!=null&&(m=Ee({inputs:{x:o},backend:e,attrs:{perm:p}}),c=S.getInnerMostAxes(c.length,a),u.push(m)),S.assertAxesAreInnerMostDims("prod",c,a);let f;if(e.shouldExecuteOnCPU([m])){let d=e.texData.get(m.dataId).values,{outVals:h,outShape:g,outDtype:y}=dM(m.shape,m.dtype,d,c);f=e.makeTensorInfo(g,y,h)}else{let[d,h]=S.computeOutAndReduceShapes(m.shape,c),g=x.sizeFromShape(h),y=ut({inputs:{x:m},backend:e,attrs:{shape:[-1,g]}}),b=_u(o.dtype),w=Wn(y,b,"prod",e);f=ut({inputs:{x:w},backend:e,attrs:{shape:d}}),u.push(y),u.push(w)}if(i){u.push(f);let d=S.expandShapeToKeepDim(f.shape,l);f=ut({inputs:{x:f},backend:e,attrs:{shape:d}})}return u.forEach(d=>e.disposeIntermediateTensorInfo(d)),f}var Tz={kernelName:Sa,backendName:"webgl",kernelFunc:Snt};var KN=r=>{let{backend:t,attrs:e}=r,{start:n,stop:o,step:s,dtype:i}=e,a=hM(n,o,s,i);return t.makeTensorInfo([a.length],i,a)},_z={kernelName:El,backendName:"webgl",kernelFunc:KN};var knt="return 1.0 / x;",Nnt=Ct({opSnippet:knt}),Ez={kernelName:ka,backendName:"webgl",kernelFunc:Nnt};var Tnt=hr+`
return (x < 0.0) ? 0.0 : x;
`,_nt=`
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;
`,Ent=Ct({opSnippet:Tnt,packedOpSnippet:_nt}),Az={kernelName:Ns,backendName:"webgl",kernelFunc:Ent};var Ant=hr+`
return (x < 0.0) ? 0.0 : min(6.0, x);
`,Dnt=`
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;
`,$nt=Ct({opSnippet:Ant,packedOpSnippet:Dnt}),Dz={kernelName:_s,backendName:"webgl",kernelFunc:$nt};var jN=class{constructor(t,e,n,o,s){this.variableNames=["A"],this.outputShape=[];let[i,a,u,l]=t;this.outputShape=[i,e,n,l];let c=[o&&e>1?a-1:a,o&&n>1?u-1:u],p=[o&&e>1?e-1:e,o&&n>1?n-1:n],m;s?m="(vec2(yRC) + vec2(0.5)) * effectiveInputOverOutputRatioRC - vec2(0.5)":m="vec2(yRC) * effectiveInputOverOutputRatioRC",this.userCode=`
const vec2 effectiveInputOverOutputRatioRC = vec2(
${c[0]/p[0]},
${c[1]/p[1]});
const vec2 inputShapeRC = vec2(${a}.0, ${u}.0);
void main() {
ivec4 coords = getOutputCoords();
int b = coords[0];
int d = coords[3];
ivec2 yRC = coords.yz;
// Fractional source index.
vec2 sourceFracIndexRC = ${m};
// 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);
}
`}};var XN=class{constructor(t,e,n,o,s){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=[];let[i,a,u,l]=t;this.outputShape=[i,e,n,l];let c=[o&&e>1?a-1:a,o&&n>1?u-1:u],p=[o&&e>1?e-1:e,o&&n>1?n-1:n],m;s?m="(vec3(yRC) + vec3(0.5)) * effectiveInputOverOutputRatioRC - vec3(0.5)":m="vec3(yRC) * effectiveInputOverOutputRatioRC",this.userCode=`
const vec3 effectiveInputOverOutputRatioRC = vec3(
${c[0]/p[0]},
${c[1]/p[1]},
${c[1]/p[1]});
const vec3 inputShapeRC = vec3(${a}.0, ${u}.0,
${u}.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 = ${m};
// 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 Rnt(r){let{inputs:t,backend:e,attrs:n}=r,{images:o}=t,{alignCorners:s,halfPixelCenters:i,size:a}=n,[u,l]=a,c=V().getBool("WEBGL_PACK_IMAGE_OPERATIONS")?new XN(o.shape,u,l,s,i):new jN(o.shape,u,l,s,i);return e.runWebGLProgram(c,[o],"float32")}var $z={kernelName:Ts,backendName:"webgl",kernelFunc:Rnt};var YN=class{constructor(t,e,n){this.variableNames=["dy"],this.outputShape=[],this.outputShape=e;let[,o,s]=e,[,i,a]=t,u=[n&&i>1?o-1:o,n&&a>1?s-1:s],l=[n&&i>1?i-1:i,n&&a>1?a-1:a],c=u[0]/l[0],p=u[1]/l[1],m=1/c,f=1/p,d=Math.ceil(m)*2+2,h=Math.ceil(f)*2+2;this.userCode=`
void main() {
ivec4 coords = getOutputCoords();
int b = coords[0];
int d = coords[3];
int r = coords[1];
int c = coords[2];
float accumulator = 0.0;
const float heightScale = float(${c});
const float widthScale = float(${p});
const float invHeightScale = float(${m});
const float invWidthScale = float(${f});
const int winHeight = int(${d});
const int winWidth = int(${h});
// 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 >= ${i}) {
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 >= ${a}) {
continue;
}
float dxR = float(dyR) * heightScale;
int topDxRIndex = int(floor(dxR));
int bottomDxRIndex = int(min(ceil(dxR), ${o-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), ${s-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 Fnt(r){let{inputs:t,backend:e,attrs:n}=r,{images:o,dy:s}=t,{alignCorners:i}=n,a=new YN(s.shape,o.shape,i);return e.runWebGLProgram(a,[s],s.dtype)}var Rz={kernelName:Rp,backendName:"webgl",kernelFunc:Fnt};var ZN=class{constructor(t,e,n,o,s){this.variableNames=["A"],this.outputShape=[];let[i,a,u,l]=t;this.outputShape=[i,e,n,l];let c=[o&&e>1?a-1:a,o&&n>1?u-1:u],p=[o&&e>1?e-1:e,o&&n>1?n-1:n],m=o?"0.5":"0.0",f;s?f="max((vec2(yRC) + vec2(0.5)) * effectiveInputOverOutputRatioRC, vec2(0.0))":f="vec2(yRC) * effectiveInputOverOutputRatioRC",this.userCode=`
const vec2 effectiveInputOverOutputRatioRC = vec2(
${c[0]/p[0]},
${c[1]/p[1]});
const vec2 inputShapeRC = vec2(${a}.0, ${u}.0);
void main() {
ivec4 coords = getOutputCoords();
int b = coords[0];
int d = coords[3];
ivec2 yRC = coords.yz;
// Fractional source index.
vec2 sourceFracIndexRC = ${f};
// Compute the coordinators of nearest neighbor point.
ivec2 sourceNearestRC = ivec2(
min(inputShapeRC - 1.0, floor(sourceFracIndexRC + ${m})));
float newValue = getA(b, sourceNearestRC.x, sourceNearestRC.y, d);
setOutput(newValue);
}
`}};var JN=class{constructor(t,e,n,o,s){this.variableNames=["A"],this.packedInputs=!0,this.packedOutput=!0,this.outputShape=[];let[i,a,u,l]=t;this.outputShape=[i,e,n,l];let c=[o&&e>1?a-1:a,o&&n>1?u-1:u],p=[o&&e>1?e-1:e,o&&n>1?n-1:n],m=o?"0.5":"0.0",f;s?f="max((vec3(yRC) + vec3(0.5)) * effectiveInputOverOutputRatioRC, vec3(0.0))":f="vec3(yRC) * effectiveInputOverOutputRatioRC",this.userCode=`
const vec3 effectiveInputOverOutputRatioRC = vec3(
${c[0]/p[0]},
${c[1]/p[1]},
${c[1]/p[1]});
const vec3 inputShapeRC = vec3(${a}.0, ${u}.0,
${u}.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 = ${f};
// Compute the coordinators of nearest neighbor point.
ivec3 sourceNearestRC = ivec3(
min(inputShapeRC - 1.0, floor(sourceFracIndexRC + ${m})));
// Should we calculate next column and row elements in 2x2 packed cell.
bool hasNextCol = d < ${l-1};
bool hasNextRow = coords.z < ${n-1};
vec4 newValue = vec4(
getAValue(b, sourceNearestRC.x, sourceNearestRC.y, d),
hasNextCol ? getAValue(b, sourceNearestRC.x, sourceNearestRC.y, d + 1)
: 0.0,
hasNextRow ? getAValue(b, sourceNearestRC.x, sourceNearestRC.z, d)
: 0.0,
(hasNextRow && hasNextCol) ?
getAValue(b, sourceNearestRC.x, sourceNearestRC.z, d + 1) : 0.0);
setOutput(newValue);
}
`}};function Ont(r){let{inputs:t,backend:e,attrs:n}=r,{images:o}=t,{alignCorners:s,halfPixelCenters:i,size:a}=n,[u,l]=a,c=V().getBool("WEBGL_PACK_IMAGE_OPERATIONS")?new JN(o.shape,u,l,s,i):new ZN(o.shape,u,l,s,i);return e.runWebGLProgram(c,[o],o.dtype)}var Fz={kernelName:Al,backendName:"webgl",kernelFunc:Ont};var QN=class{constructor(t,e,n){this.variableNames=["dy"],this.outputShape=[],this.outputShape=e;let[,o,s]=e,[,i,a]=t,u=[n&&i>1?o-1:o,n&&a>1?s-1:s],l=[n&&i>1?i-1:i,n&&a>1?a-1:a],c=u[0]/l[0],p=u[1]/l[1],m=1/c,f=1/p,d=Math.ceil(m)*2+2,h=Math.ceil(f)*2+2;this.userCode=`
void main() {
ivec4 coords = getOutputCoords();
int b = coords[0];
int d = coords[3];
int r = coords[1];
int c = coords[2];
float accumulator = 0.0;
const float heightScale = float(${c});
const float widthScale = float(${p});
const float invHeightScale = float(${m});
const float invWidthScale = float(${f});
const int winHeight = int(${d});
const int winWidth = int(${h});
// 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 >= ${i}) {
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 >= ${a}) {
continue;
}
float sourceFracRow =
float(${u[0]}) *
(float(dyR) / float(${l[0]}));
float sourceFracCol =
float(${u[1]}) *
(float(dyC) / float(${l[1]}));
int sourceNearestRow = int(min(
float(int(${o}) - 1),
${n} ? float(round(sourceFracRow)) :
float(floor(sourceFracRow))));
int sourceNearestCol = int(min(
float(int(${s}) - 1),
${n} ? float(round(sourceFracCol)) :
float(floor(sourceFracCol))));
if (r == sourceNearestRow && c == sourceNearestCol) {
accumulator += getDy(b, dyR, dyC, d);
}
}
}
// End loop over dy
setOutput(accumulator);
}
`}};function Pnt(r){let{inputs:t,backend:e,attrs:n}=r,{images:o,dy:s}=t,{alignCorners:i}=n,a=new QN(s.shape,o.shape,i);return e.runWebGLProgram(a,[s],s.dtype)}var Oz={kernelName:$p,backendName:"webgl",kernelFunc:Pnt};var tT=class{constructor(t,e){this.variableNames=["x"];let n=t.length;if(n>4)throw new Error(`WebGL backend: Reverse of rank-${n} tensor is not yet supported`);if(this.outputShape=t,n===1){this.userCode=`
void main() {
int coord = getOutputCoords();
setOutput(getX(${t[0]} - coord - 1));
}
`;return}let o=a=>e.indexOf(a)!==-1&&t[a]!==1?`${t[a]} - coords[${a}] - 1`:`coords[${a}]`,s=t.map((a,u)=>o(u)).join(","),i=zt(n);this.userCode=`
void main() {
${i} coords = getOutputCoords();
setOutput(getX(${s}));
}
`}};var eT=class{constructor(t,e){this.variableNames=["x"],this.packedInputs=!0,this.packedOutput=!0;let n=t.length;if(n>4)throw new Error(`WebGL backend: Reverse of rank-${n} tensor is not yet supported`);this.outputShape=t;let o=rr("rc",n),s=`${o[n-1]} + 1 < ${this.outputShape[n-1]}`,i=`${o[n-2]} + 1 < ${this.outputShape[n-2]}`,a=zt(n);n===1?this.userCode=`
void main(){
int rc = getOutputCoords();
vec4 result = vec4(0.);
result.r = getChannel(getX(${t[0]} - rc - 1),
${t[0]} - rc - 1);
if(${s}){
result.g = getChannel(getX(${t[0]} - (rc + 1) - 1),
${t[0]} - (rc + 1) - 1);
}
setOutput(result);
}
`:this.userCode=`
void main() {
${a} rc = getOutputCoords();
vec4 result = vec4(0.);
result.r = ${u(o.slice())};
if(${s}){
result.g = ${l(o.slice())};
}
if(${i}) {
result.b = ${c(o.slice())};
if(${s}) {
result.a = ${p(o.slice())};
}
}
setOutput(result);
}
`;function u(d){return m(d)}function l(d){return d[n-1]="("+d[n-1]+" + 1)",m(d)}function c(d){return d[n-2]="("+d[n-2]+" + 1)",m(d)}function p(d){return d[n-1]="("+d[n-1]+" + 1)",d[n-2]="("+d[n-2]+" + 1)",m(d)}function m(d){let h=t.map((b,w)=>f(w,d)),g=h.join(","),y=h.slice(-2).join(",");return`getChannel(getX(${g}), vec2(${y}))`}function f(d,h){return e.indexOf(d)!==-1&&t[d]!==1?`${t[d]} - ${h[d]} - 1`:`${h[d]}`}}};function Mnt(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{dims:s}=n,i=o.shape.length,a=x.parseAxisParam(s,o.shape);if(i===0)return Ke({inputs:{x:o},backend:e});let u=V().getBool("WEBGL_PACK_ARRAY_OPERATIONS")?new eT(o.shape,a):new tT(o.shape,a);return e.runWebGLProgram(u,[o],o.dtype)}var Pz={kernelName:Es,backendName:"webgl",kernelFunc:Mnt};var rT=class{constructor(t,e){this.variableNames=["Image"],this.outputShape=[],this.customUniforms=[{name:"params",type:"vec4"}];let n=t[1],o=t[2];this.outputShape=t;let s="";typeof e=="number"?s=`float outputValue = ${e.toFixed(2)};`:s=`
vec3 fill = vec3(${e.join(",")});
float outputValue = fill[coords[3]];`,this.userCode=`
void main() {
ivec4 coords = getOutputCoords();
int x = coords[2];
int y = coords[1];
float coordXFloat = (float(x) - params[0]) * params[3] -
(float(y) - params[1]) * params[2];
float coordYFloat = (float(x) - params[0]) * params[2] +
(float(y) - params[1]) * params[3];
int coordX = int(round(coordXFloat + params[0]));
int coordY = int(round(coordYFloat + params[1]));
${s}
if(coordX >= 0 && coordX < ${o} && coordY >= 0 && coordY < ${n}) {
outputValue = getImage(coords[0], coordY, coordX, coords[3]);
}
setOutput(outputValue);
}
`}};var Mz={kernelName:Oa,backendName:"webgl",kernelFunc:({inputs:r,attrs:t,backend:e})=>{let{image:n}=r,{radians:o,fillValue:s,center:i}=t,a=e,u=new rT(n.shape,s),[l,c]=S.getImageCenter(i,n.shape[1],n.shape[2]),p=[[l,c,Math.sin(o),Math.cos(o)]];return a.runWebGLProgram(u,[n],n.dtype,p)}};var Lnt=`
// OpenGL ES does not support round function.
// The algorithm is based on banker's rounding.
float base = floor(x);
if ((x - base) < 0.5) {
return floor(x);
} else if ((x - base) > 0.5) {
return ceil(x);
} else {
if (mod(base, 2.0) == 0.0) {
return base;
} else {
return base + 1.0;
}
}
`,znt=Ct({opSnippet:Lnt}),Lz={kernelName:As,backendName:"webgl",kernelFunc:znt};var Bnt="return inversesqrt(x);",Vnt=Ct({opSnippet:Bnt,cpuKernelImpl:gM}),zz={kernelName:Ds,backendName:"webgl",kernelFunc:Vnt};var gg=class{constructor(t,e,n,o,s,i,a=!0){this.variableNames=["updates","indices","defaultValue"],this.outputShape=i;let u=zt(s.length),l=zt(i.length),c="";n===1?c="i":n===2&&(c="i, j");let p=`getIndices(${c})`,m="";o===1?m="i":o===2&&(m="i, coords[1]");let f=`getUpdates(${m})`,d=e>1?"strides[j]":"strides";this.userCode=`
${u} strides = ${u}(${s});
void main() {
${l} coords = getOutputCoords();
float sum = 0.0;
bool found = false;
for (int i = 0; i < ${t}; i++) {
int flattenedIndex = 0;
for (int j = 0; j < ${e}; j++) {
int index = round(${p});
flattenedIndex += index * ${d};
}
if (flattenedIndex == coords[0]) {
sum += ${f};
found = true;
}
}
setOutput(mix(getDefaultValue(), sum, float(found)));
}
`}};function Gnt(r){let{inputs:t,backend:e,attrs:n}=r,{indices:o,updates:s}=t,{shape:i}=n,{sliceRank:a,numUpdates:u,sliceSize:l,strides:c,outputSize:p}=S.calculateShapes(s,o,i),m=[p/l,l];if(p===0)return e.makeTensorInfo(i,o.dtype);let f=ut({inputs:{x:o},backend:e,attrs:{shape:[u,a]}}),d=ut({inputs:{x:s},backend:e,attrs:{shape:[u,l]}}),h=e.makeTensorInfo([],"float32",new Float32Array([0])),g=new gg(u,a,f.shape.length,d.shape.length,c,m),y=e.runWebGLProgram(g,[d,f,h],d.dtype),b=ut({inputs:{x:y},backend:e,attrs:{shape:i}});return e.disposeIntermediateTensorInfo(f),e.disposeIntermediateTensorInfo(d),e.disposeIntermediateTensorInfo(y),e.disposeIntermediateTensorInfo(h),b}var Bz={kernelName:Na,backendName:"webgl",kernelFunc:Gnt};var nT=class{constructor(t,e,n){this.variableNames=["c","a","b"],this.outputShape=e;let o,s;if(n>4)throw Error(`Where for rank ${n} is not yet supported`);if(n===1)s="resRC",o="resRC";else{let a=["resRC.x","resRC.y","resRC.z","resRC.w"],u=[],l=[];for(let c=0;c<e.length;c++)l.push(`${a[c]}`),c<t&&u.push(`${a[c]}`);o=u.join(),s=l.join()}let i=zt(n);this.userCode=`
void main() {
${i} resRC = getOutputCoords();
float cVal = getC(${o});
if (cVal >= 1.0) {
setOutput(getA(${s}));
} else {
setOutput(getB(${s}));
}
}
`}};function Wnt(r){let{inputs:t,backend:e}=r,{condition:n,t:o,e:s}=t,i=new nT(n.shape.length,o.shape,o.shape.length);return e.runWebGLProgram(i,[n,o,s],ar(o.dtype,s.dtype))}var Vz={kernelName:Si,backendName:"webgl",kernelFunc:Wnt};var Unt=`
// Stable and Attracting Fixed Point (0, 1) for Normalized Weights.
// see: https://arxiv.org/abs/1706.02515
float scaleAlpha = ${S.SELU_SCALEALPHA};
float scale = ${S.SELU_SCALE};
return (x >= 0.0) ? scale * x : scaleAlpha * (exp(x) - 1.0);
`,Hnt=Ct({opSnippet:Unt}),Gz={kernelName:Ta,backendName:"webgl",kernelFunc:Hnt};var qnt=Po+`
return 1.0 / (1.0 + exp(-1.0 * x));
`,Knt=`
vec4 result = 1.0 / (1.0 + exp(-1.0 * x));
bvec4 isNaN = isnan(x);
result.r = isNaN.r ? x.r : result.r;
result.g = isNaN.g ? x.g : result.g;
result.b = isNaN.b ? x.b : result.b;
result.a = isNaN.a ? x.a : result.a;
return result;
`,jnt=Ct({opSnippet:qnt,packedOpSnippet:Knt,cpuKernelImpl:xM}),Wz={kernelName:Rs,backendName:"webgl",kernelFunc:jnt};var Xnt=`
if (isnan(x)) { return 0.0; }
return sign(x);
`,Ynt=Ct({opSnippet:Xnt}),Uz={kernelName:Ea,backendName:"webgl",kernelFunc:Ynt};var Znt=Po+`
return sin(x);
`,Jnt=Ct({opSnippet:Znt}),Hz={kernelName:$s,backendName:"webgl",kernelFunc:Jnt};var Qnt=`
float e2x = exp(x);
return (e2x - 1.0 / e2x) / 2.0;
`,tot=Ct({opSnippet:Qnt}),qz={kernelName:_a,backendName:"webgl",kernelFunc:tot};var eot=`
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;
`,rot=Ct({opSnippet:eot}),Kz={kernelName:Aa,backendName:"webgl",kernelFunc:rot};var not=r=>{let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{blockShape:s,paddings:i}=n;x.assert(o.shape.length<=4,()=>"spaceToBatchND for rank > 4 with a WebGL backend not implemented yet");let a=s.reduce((y,b)=>y*b),u=[[0,0]];u.push(...i);for(let y=1+s.length;y<o.shape.length;++y)u.push([0,0]);let l=[],c=qN({inputs:{x:o},backend:e,attrs:{paddings:u,constantValue:0}}),p=S.getReshaped(c.shape,s,a,!1),m=S.getPermuted(p.length,s.length,!1),f=S.getReshapedPermuted(c.shape,s,a,!1),d=ut({inputs:{x:c},backend:e,attrs:{shape:p}}),h=Ee({inputs:{x:d},backend:e,attrs:{perm:m}}),g=ut({inputs:{x:h},backend:e,attrs:{shape:f}});return l.push(c),l.push(d),l.push(h),l.forEach(y=>e.disposeIntermediateTensorInfo(y)),g},jz={kernelName:Ni,backendName:"webgl",kernelFunc:not};function oot(r){let{inputs:t,backend:e}=r,{indices:n,values:o,denseShape:s,defaultValue:i}=t;if(s.shape.length!==1)throw new Error(`Dense shape must be a vector, saw:
${s.shape}`);if(n.shape.length!==2)throw new Error(`Indices must be a matrix, saw:
${n.shape}`);if(o.shape.length!==1)throw new Error(`Values must be a vector, saw:
${o.shape}`);if(i.shape.length!==0)throw new Error(`Default value must be a scalar, saw:
${i.shape}`);let a=e.readSync(n.dataId),u=e.readSync(o.dataId),l=e.readSync(s.dataId),c=e.readSync(i.dataId)[0],[p,m,f,d,h]=bM(a,n.shape,n.dtype,u,o.dtype,l,c);return[e.makeTensorInfo(m,n.dtype,p),e.makeTensorInfo([m[0]],o.dtype,f),e.makeTensorInfo([d.length],"bool",new Uint8Array(d.map(g=>Number(g)))),e.makeTensorInfo([h.length],n.dtype,new Int32Array(h))]}var Xz={kernelName:Dl,backendName:"webgl",kernelFunc:oot};function sot(r){let{inputs:t,backend:e}=r,{inputIndices:n,inputShape:o,newShape:s}=t;if(n.shape.length!==2)throw new Error(`Input indices should be a matrix but received shape ${n.shape}`);if(o.shape.length!==1)throw new Error(`Input shape should be a vector but received shape ${o.shape}`);if(s.shape.length!==1)throw new Error(`Target shape should be a vector but received shape ${s.shape}`);let i=Array.from(e.readSync(o.dataId)),a=e.readSync(n.dataId),u=Array.from(e.readSync(s.dataId)),[l,c,p]=wM(a,n.shape,n.dtype,i,u);return[e.makeTensorInfo(c,n.dtype,l),e.makeTensorInfo([p.length],s.dtype,new Int32Array(p))]}var Yz={kernelName:Da,backendName:"webgl",kernelFunc:sot};function iot(r){let{inputs:t,backend:e}=r,{data:n,indices:o,segmentIds:s}=t;if(n.shape.length<1)throw new Error("Data should be at least 1 dimensional but received scalar");if(o.shape.length!==1)throw new Error(`Indices should be a vector but received shape
${o.shape}`);if(s.shape.length!==1)throw new Error(`Segment ids should be a vector but received shape
${s.shape}`);let i=e.readSync(n.dataId),a=e.readSync(o.dataId),u=e.readSync(s.dataId),[l,c]=Ib(i,n.shape,n.dtype,a,u,!0);return e.makeTensorInfo(c,n.dtype,l)}var Zz={kernelName:$l,backendName:"webgl",kernelFunc:iot};function aot(r){let{inputs:t,backend:e}=r,{data:n,indices:o,segmentIds:s}=t;if(n.shape.length<1)throw new Error("Data should be at least 1 dimensional but received scalar");if(o.shape.length!==1)throw new Error(`Indices should be a vector but received shape
${o.shape}`);if(s.shape.length!==1)throw new Error(`Segment ids should be a vector but received shape
${s.shape}`);let i=e.readSync(n.dataId),a=e.readSync(o.dataId),u=e.readSync(s.dataId),[l,c]=Ib(i,n.shape,n.dtype,a,u);return e.makeTensorInfo(c,n.dtype,l)}var Jz={kernelName:Rl,backendName:"webgl",kernelFunc:aot};function lot(r){let{inputs:t,backend:e,attrs:n}=r,{sparseIndices:o,sparseValues:s,defaultValue:i}=t,{outputShape:a}=n,{sliceRank:u,numUpdates:l,strides:c,outputSize:p}=S.calculateShapes(s,o,a),m=!1,f=new gg(l,u,o.shape.length,s.shape.length,c,[p,1],m),d=e.runWebGLProgram(f,[s,o,i],s.dtype),h=ut({inputs:{x:d},backend:e,attrs:{shape:a}});return e.disposeIntermediateTensorInfo(d),h}var Qz={kernelName:Fp,backendName:"webgl",kernelFunc:lot};function uot(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{numOrSizeSplits:s,axis:i}=n,a=x.parseAxisParam(i,o.shape)[0],u=S.prepareSplitSize(o,s,a),l=o.shape.length,c=new Array(l).fill(0),p=o.shape.slice();return u.map(m=>{let f=[...p];f[a]=m;let d=pi({inputs:{x:o},backend:e,attrs:{begin:c,size:f}});return c[a]+=m,d})}var tB={kernelName:Ti,backendName:"webgl",kernelFunc:uot};var eB="return sqrt(x);",cot=Ct({opSnippet:eB,packedOpSnippet:eB,cpuKernelImpl:vM}),rB={kernelName:Fs,backendName:"webgl",kernelFunc:cot};var pot="return x * x;",mot=Ct({opSnippet:pot}),nB={kernelName:Fl,backendName:"webgl",kernelFunc:mot};var oB="return (a - b) * (a - b);",fot=le({opSnippet:oB,packedOpSnippet:oB}),sB={kernelName:Ms,backendName:"webgl",kernelFunc:fot};function dot({inputs:r,attrs:t,backend:e}){let{x:n}=r,o=hr+`
return x > 0.0 ? 1.0 : float(${t.alpha});
`,s=new vn(n.shape,o);return e.runWebGLProgram(s,[n],n.dtype)}var iB={kernelName:ao,backendName:"webgl",kernelFunc:dot};var oT=class{constructor(t,e,n){this.variableNames=["x"],this.outputShape=n;let o=n.length,s=zt(n.length),i=zt(n.length),a="";if(o===1)a="coords * strides + begin";else{let u=0;a=n.map((l,c)=>(u++,n.length===1?`coords * strides[${c}] + begin[${c}]`:`coords[${u-1}] * strides[${c}] + begin[${c}]`)).join(",")}this.userCode=`
${s} begin = ${s}(${t});
${s} strides = ${s}(${e});
void main() {
${i} coords = getOutputCoords();
setOutput(getX(${a}));
}
`}};function hot(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{begin:s,end:i,strides:a,beginMask:u,endMask:l,ellipsisMask:c,newAxisMask:p,shrinkAxisMask:m}=n,{finalShapeSparse:f,finalShape:d,isIdentity:h,sliceDim0:g,isSimpleSlice:y,begin:b,end:w,strides:v}=Ve.sliceInfo(o.shape,s,i,a,u,l,c,p,m),k;if(h)k=ut({inputs:{x:o},backend:e,attrs:{shape:d}});else if(g||y){x.assert(o.shape.length>=1,()=>`Input must have rank at least 1, got: ${o.shape.length}`);let D=Ve.computeOutShape(b,w,v),$=pi({inputs:{x:o},backend:e,attrs:{begin:b,size:D}});k=ut({inputs:{x:$},backend:e,attrs:{shape:d}}),e.disposeIntermediateTensorInfo($)}else if(e.shouldExecuteOnCPU([o])){let $=e.readSync(o.dataId),R=It(o.shape,o.dtype,$),M=CM(f,R,v,b);k=e.makeTensorInfo(d,o.dtype,M.values)}else{let $=new oT(b,v,f);k=e.runWebGLProgram($,[o],o.dtype)}let E=ut({inputs:{x:k},backend:e,attrs:{shape:d}});return e.disposeIntermediateTensorInfo(k),E}var aB={kernelName:$a,backendName:"webgl",kernelFunc:hot};function got(r){let{inputs:t,backend:e,attrs:n}=r,{separator:o,nGramWidths:s,leftPad:i,rightPad:a,padWidth:u,preserveShortSequences:l}=n,{data:c,dataSplits:p}=t,m=e.readSync(c.dataId),f=e.readSync(p.dataId),[d,h]=IM(m,f,o,s,i,a,u,l);return[e.makeTensorInfo([d.length],"string",d),e.makeTensorInfo(p.shape,"int32",h)]}var lB={kernelName:Op,backendName:"webgl",kernelFunc:got};function xot(r){let{inputs:t,backend:e,attrs:n}=r,{skipEmpty:o}=n,{input:s,delimiter:i}=t;if(s.dtype!=="string")throw new Error("Input must be of datatype string");if(s.shape.length!==1)throw new Error(`Input must be a vector, got shape: ${s.shape}`);if(i.shape.length!==0)throw new Error(`Delimiter must be a scalar, got shape: ${i.shape}`);let a=e.readSync(s.dataId),u=e.readSync(i.dataId)[0],[l,c,p]=SM(a,u,o),m=c.length;return[e.makeTensorInfo([m,2],"int32",l),e.makeTensorInfo([m],"string",c),e.makeTensorInfo([2],"int32",new Int32Array(p))]}var uB={kernelName:Pp,backendName:"webgl",kernelFunc:xot};function yot(r){let{inputs:t,backend:e,attrs:n}=r,{numBuckets:o}=n,{input:s}=t;if(s.dtype!=="string")throw new Error("Input must be of datatype string");if(o<=0)throw new Error("Number of buckets must be at least 1");let i=e.readSync(s.dataId),a=kM(i,o);return e.makeTensorInfo(s.shape,"int32",a)}var cB={kernelName:Mp,backendName:"webgl",kernelFunc:yot};var bot="return tan(x);",wot=Ct({opSnippet:bot}),pB={kernelName:zs,backendName:"webgl",kernelFunc:wot};var vot=`
float e2x = exp(-2.0 * abs(x));
return sign(x) * (1.0 - e2x) / (1.0 + e2x);
`,Cot=Ct({opSnippet:vot}),mB={kernelName:Bs,backendName:"webgl",kernelFunc:Cot};var sT=class{constructor(t,e){this.variableNames=["A"];let n=new Array(t.length);for(let i=0;i<n.length;i++)n[i]=t[i]*e[i];this.outputShape=n,this.rank=n.length;let o=zt(this.rank),s=Iot(t);this.userCode=`
void main() {
${o} resRC = getOutputCoords();
setOutput(getA(${s}));
}
`}};function Iot(r){let t=r.length;if(t>5)throw Error(`Tile for rank ${t} is not yet supported`);if(t===1)return`imod(resRC, ${r[0]})`;let e=["resRC.x","resRC.y","resRC.z","resRC.w","resRC.u"],n=[];for(let o=0;o<r.length;o++)n.push(`imod(${e[o]}, ${r[o]})`);return n.join()}function iT(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{reps:s}=n;if(o.dtype==="string"||o.shape.length>5){let u=e.readSync(o.dataId),l=o.dtype==="string"?u.map(m=>x.decodeString(m)):u,c=It(o.shape,o.dtype,l),p=TM(c,s);return e.makeTensorInfo(p.shape,p.dtype,p.values)}let i=new sT(o.shape,s);return e.runWebGLProgram(i,[o],o.dtype)}var fB={kernelName:jn,backendName:"webgl",kernelFunc:iT};var aT=class{constructor(t){this.variableNames=["x","indices"],this.customUniforms=[{name:"n",type:"int"},{name:"firstPass",type:"int"},{name:"negativeInf",type:"float"},{name:"dir",type:"int"},{name:"inc",type:"int"}],this.outputShape=t,this.userCode=`
void main() {
ivec2 coords = getOutputCoords();
int batch = coords[0];
int elemIdx = coords[1];
// We compare elements pair-wise within a group of size 2 * inc.
// The comparing rule for each group alternates between ascending
// and descending. Within each group, we compare each pair at
// positions i and i+inc. To decide whether an element at position i
// is x0 or x1, we mod it by 2 * inc, if the result is smaller than
// inc, it is in the first half of the group, we denote it as x0,
// otherwise we denote it as x1.
// For example, as shown in the Bitonic top K paper referenced above,
// Figure5(a) shows that element[1] is in the
// second half of the group when group size is 2, but it is in the
// first half of the group when group size is 4.
bool isFirstInPair = imod(elemIdx, 2 * inc) < inc;
int i = isFirstInPair ? elemIdx : elemIdx - inc;
int i0 = firstPass == 1 ? i : int(getIndices(batch, i));
int i1 = firstPass == 1 ? i + inc : int(getIndices(batch, i + inc));
float x0 = i0 < n ? getX(batch, i0) : negativeInf;
float x1 = i1 < n ? getX(batch, i1) : negativeInf;
// Denotes which direction indices are in (ascending or descending).
bool reverse = imod(elemIdx, 2 * dir) >= dir;
bool isGreater = x0 > x1 || (x0 == x1 && i1 > i0);
if (reverse == isGreater) { // Elements in opposite order of direction
int iTemp = i0;
i0 = i1;
i1 = iTemp;
}
if (isFirstInPair) {
setOutput(float(i0));
} else {
setOutput(float(i1));
}
}
`}},lT=class{constructor(t){this.variableNames=["x","indices"],this.customUniforms=[{name:"n",type:"int"},{name:"firstPass",type:"int"},{name:"k",type:"int"}],this.outputShape=t,this.userCode=`
void main() {
// Takes max of indices (0, k), (1, k + 1), (2, k + 2) ...
ivec2 coords = getOutputCoords();
int batch = coords[0];
int elemIdx = coords[1];
// The output size is half of the previous size.
// If the previous sequence is | | | | _ _ _ _ | | | | _ _ _ _ (k=4),
// we only need to output the indices at positions |, the indices at
// positions _ can be thrown away, see Figure5(b) After Phase 2
// (Merge phase) in the Bitonic Top K paper referenced above.
// For example, the paper shows we only need to output the orange bars.
// The output sequence should look like this | | | | | | | |.
// Because the sequence is halved, to map the output index back
// to the previous sequence to find the corresponding value,
// we need to double the index. When we double the index,
// we basically interpolate a position, so 2i looks like
// | _ | _ | _ | _ | _ | _ | _. We move the | to the first k position
// of each 2k positions by - elemIdx % k. E.g. for output at
// index 4,5,6,7, we want to get the corresponding element at
// original index 8,9,10,11, for output at index 8,9,10,11,
// we want to get the corresponding element at original index
// 16,17,18,19, so on and so forth.
int i = elemIdx < k ? elemIdx : (elemIdx * 2 - imod(elemIdx, k));
int i0 = firstPass == 1 ? i : int(getIndices(batch, i));
int i1 = firstPass == 1 ? i + k : int(getIndices(batch, i + k));
float x0 = getX(batch, i0);
float x1 = i1 < n ? getX(batch, i1) : x0;
setOutput(x0 >= x1 ? float(i0) : float(i1));
}
`}};function Vc(r,t){t!==null&&r.disposeIntermediateTensorInfo(t)}function dB(r){let t=1;for(;t<r;)t*=2;return t}function Sot(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{k:s,sorted:i}=n,a=V().getNumber("TOPK_LAST_DIM_CPU_HANDOFF_SIZE_THRESHOLD"),u=V().getNumber("TOPK_K_CPU_HANDOFF_THRESHOLD"),l=o.shape,c=l[l.length-1];if(e.shouldExecuteOnCPU([o])||c<a||s>u){let M=e.readSync(o.dataId),[W,U]=_M(M,l,o.dtype,s,i);return[e.makeTensorInfo(W.shape,W.dtype,W.values),e.makeTensorInfo(U.shape,U.dtype,U.values)]}if(s===0)return l[l.length-1]=0,[e.makeTensorInfo(l,o.dtype,[]),e.makeTensorInfo(l,"int32",[])];if(c===1)return[o,dl({attrs:{shape:l,dtype:"int32",value:0},backend:e})];let p=e.texData.get(o.dataId),m=p!==null&&p.isPacked,f=m?e.unpackTensor(o):o,h=x.sizeFromShape(l)/c,g=ut({inputs:{x:f},attrs:{shape:[h,c]},backend:e});m&&Vc(e,f);let y=dB(s),b=dB(c),w=null,v=()=>w===null?[g,g]:[g,w],k=(M,W,U)=>{let q=v(),K=new aT(U),Q=[[c],[w===null?1:0],[Number.NEGATIVE_INFINITY],[M],[W]],rt=w;w=e.runWebGLProgram(K,q,"int32",Q),Vc(e,rt)};for(let M=1;M<y;M*=2){let W=M*2;for(let U=M;U>=1;U/=2)k(W,U,[h,b])}for(let M=b;M>y;M/=2){let W=v(),U=new lT([h,M/2]),K=[[c],[w===null?1:0],[y]],j=w;w=e.runWebGLProgram(U,W,"int32",K),Vc(e,j);let Q=y/2,rt=Q*2;for(let X=Q;X>=1;X/=2)k(rt,X,w.shape)}let E=w;w=pi({inputs:{x:w},backend:e,attrs:{begin:0,size:[h,s]}}),Vc(e,E);let D=AN({inputs:{x:g,indices:w},backend:e,attrs:{axis:1,batchDims:1}});Vc(e,g);let $=l.slice(0,-1);$.push(s),E=w,w=ut({inputs:{x:w},attrs:{shape:$},backend:e}),Vc(e,E);let R=D;return D=ut({inputs:{x:D},attrs:{shape:$},backend:e}),Vc(e,R),[D,w]}var hB={kernelName:Ra,backendName:"webgl",kernelFunc:Sot};var uT=class{constructor(t,e,n,o,s,i){this.variableNames=["Image","Transforms"],this.outputShape=i;let a=n==="nearest"?1:2,u;switch(o){case"constant":u=1;break;case"reflect":u=2;break;case"wrap":u=3;break;case"nearest":u=4;break;default:u=1;break}this.userCode=`
float mapCoord(float outCoord, float len) {
float inCoord = outCoord;
if(${u} == 2) {
if (inCoord < 0.0) {
if (len <= 1.0) {
inCoord = 0.0;
} else {
float sz2 = 2.0 * len;
if (inCoord < sz2) {
inCoord = sz2 * float(int(float(-inCoord / sz2))) +
inCoord;
}
inCoord = inCoord < -len ? inCoord + sz2 : -inCoord - 1.0;
}
} else if (inCoord > len - 1.0) {
if (len <= 1.0) {
inCoord = 0.0;
} else {
float sz2 = 2.0 * len;
inCoord -= sz2 * float(int(float(inCoord / sz2)));
if (inCoord >= len) {
inCoord = sz2 - inCoord - 1.0;
}
}
}
return clamp(inCoord, 0.0, len - 1.0);
} else if (${u} == 3) {
if (inCoord < 0.0) {
if (len <= 1.0) {
inCoord = 0.0;
} else {
float sz = len - 1.0;
inCoord += len * (float(int(float(-inCoord / sz))) + 1.0);
}
} else if (inCoord > len - 1.0) {
if (len <= 1.0) {
inCoord = 0.0;
} else {
float sz = len - 1.0;
inCoord -= len * float(int(float(inCoord / sz)));
}
}
return clamp(inCoord, 0.0, len - 1.0);
} else if (${u} == 4) {
return clamp(outCoord, 0.0, len - 1.0);
} else {
return outCoord;
}
}
float readWithFillValue(int batch, int coordY, int coordX,
int channel) {
float outputValue;
if (0 <= coordY && coordY < ${t} && 0 <= coordX && coordX < ${e}) {
outputValue = getImage(batch, coordY, coordX, channel);
} else {
outputValue = float(${s});
}
return outputValue;
}
void main() {
ivec4 coords = getOutputCoords();
float outputValue;
int batch = coords[0];
int x = coords[2];
int y = coords[1];
int channel = coords[3];
float xf = float(x);
float yf = float(y);
float a1 = getTransforms(batch, 0);
float a2 = getTransforms(batch, 1);
float a3 = getTransforms(batch, 2);
float b1 = getTransforms(batch, 3);
float b2 = getTransforms(batch, 4);
float b3 = getTransforms(batch, 5);
float c1 = getTransforms(batch, 6);
float c2 = getTransforms(batch, 7);
float projection = c1 * xf + c2 * yf + 1.0;
if (projection == 0.0) {
outputValue = float(${s});
} else {
float inX = (a1 * xf + a2 * yf + a3) / projection;
float inY = (b1 * xf + b2 * yf + b3) / projection;
float mapX = mapCoord(inX, float(${e}));
float mapY = mapCoord(inY, float(${t}));
if (${a} == 1) {
int coordY = int(round(mapY));
int coordX = int(round(mapX));
outputValue = readWithFillValue(batch, coordY, coordX,
channel);
} else {
float yFloor = floor(mapY);
float xFloor = floor(mapX);
float yCeil = yFloor + 1.0;
float xCeil = xFloor + 1.0;
float valueYFloor = (xCeil - mapX) *
readWithFillValue(batch, int(yFloor), int(xFloor), channel) +
(mapX - xFloor) *
readWithFillValue(batch, int(yFloor), int(xCeil), channel);
float valueYCeil = (xCeil - mapX) *
readWithFillValue(batch, int(yCeil), int(xFloor), channel) +
(mapX - xFloor) *
readWithFillValue(batch, int(yCeil), int(xCeil), channel);
outputValue = (yCeil - mapY) * valueYFloor +
(mapY - yFloor) * valueYCeil;
}
}
setOutput(outputValue);
}
`}};function kot(r){let{inputs:t,backend:e,attrs:n}=r,{image:o,transforms:s}=t,{interpolation:i,fillMode:a,fillValue:u,outputShape:l}=n,[c,p,m,f]=o.shape,[d,h]=l!=null?l:[p,m],g=[c,d,h,f],y=new uT(p,m,i,a,u,g);return e.runWebGLProgram(y,[o,s],"float32")}var gB={kernelName:Fa,backendName:"webgl",kernelFunc:kot};function Not(r){let{inputs:t,attrs:e,backend:n}=r,{axis:o}=e,{x:s}=t;li(s,"unique"),console.warn("WARNING: ","UI might be locked temporarily as data is being downloaded");let i=n.readSync(s.dataId),{outputValues:a,outputShape:u,indices:l}=EM(i,o,s.shape,s.dtype);return[n.makeTensorInfo(u,s.dtype,a),n.makeTensorInfo([l.length],"int32",l)]}var xB={kernelName:Lp,backendName:"webgl",kernelFunc:Not};function Tot(r){let{inputs:t,backend:e,attrs:n}=r,{value:o}=t,{axis:s}=n;s<0&&(s+=o.shape.length);let i=o,a=i.shape.length,u=o.shape[s],l=new Array(a-1),c=0;for(let h=0;h<a;h++)h!==s&&(l[c++]=i.shape[h]);let p=[],m=new Array(a).fill(0),f=i.shape.slice();f[s]=1;let d=new Array(u);for(let h=0;h<d.length;h++){m[s]=h;let g=pi({inputs:{x:i},backend:e,attrs:{begin:m,size:f}}),y=ut({inputs:{x:g},backend:e,attrs:{shape:l}});d[h]=y,p.push(g)}return p.forEach(h=>e.disposeIntermediateTensorInfo(h)),d}var yB={kernelName:_i,backendName:"webgl",kernelFunc:Tot};var cT=class{constructor(t,e){this.variableNames=["x","segmentIds"];let n=t.windowSize,o=t.batchSize,s=t.inSize,i=t.numSegments,a=i*Math.ceil(s/n);this.outputShape=[o,a];let u="0.0",l="sumValue",c=Math.floor(n/4)*4,p=n%4,m=`
sumValue += dot(values, segFilter);
`,f="";s%n>0&&(f=`
if (inIdx < 0 || inIdx >= ${s}) {
return initializationValue;
}
`);let d="";s%n>0&&(d=`
if (inIdx < 0 || inIdx >= ${s}) {
return -1.0;
}
`),this.userCode=`
const float initializationValue = ${u};
float getValue(int batch, int inIdx) {
${f}
return getX(batch, inIdx);
}
float getSegmentIdAtIndex(int inIdx) {
${d}
return getSegmentIds(inIdx);
}
void main() {
ivec2 coords = getOutputCoords();
int batch = coords[0];
int outIdx = coords[1];
int inOffset = int(floor(float(outIdx) / float(
${i})) * float(${n}));
int currentSeg = int(mod(float(outIdx), float(${i})));
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
);
${m}
}
int inIdx = inOffset + ${c};
if (${p===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
);
${m}
} else if (${p===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
);
${m}
} else if (${p===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
);
${m}
}
setOutput(${l});
}
`}};function _ot(r){let{inputs:t,backend:e,attrs:n}=r,{x:o,segmentIds:s}=t,{numSegments:i}=n,a=o.shape.length,u=[],l=0,c=S.getAxesPermutation([l],a),p=o;c!=null&&(p=Ee({inputs:{x:o},backend:e,attrs:{perm:c}}),u.push(p),l=S.getInnerMostAxes(1,a)[0]);let m=S.segment_util.computeOutShape(p.shape,l,i),f=x.sizeFromShape([p.shape[l]]),d=ut({inputs:{x:p},backend:e,attrs:{shape:[-1,f]}});u.push(d);let h=_u(o.dtype),g=(v,k,E,D,$)=>{let R=v.shape[0],M=v.shape[1],W=S.segment_util.segOpComputeOptimalWindowSize(M,$),U={windowSize:W,inSize:M,batchSize:R,numSegments:$},q=new cT(U,k),K=e.compileAndRun(q,[v,E],D);if(u.push(K),K.shape[1]===$)return K;let j=KN({backend:e,attrs:{start:0,stop:$,step:1,dtype:"float32"}}),Q=iT({inputs:{x:j},backend:e,attrs:{reps:[M/W]}});return u.push(j),u.push(Q),g(K,k,Q,D,$)},y=g(d,"unsortedSegmentSum",s,h,i),b=ut({inputs:{x:y},backend:e,attrs:{shape:m}}),w=b;if(c!=null){u.push(b);let v=S.getUndoAxesPermutation(c);w=Ee({inputs:{x:w},backend:e,attrs:{perm:v}})}return u.forEach(v=>e.disposeIntermediateTensorInfo(v)),w}var bB={kernelName:Ol,backendName:"webgl",kernelFunc:_ot};var Eot=[iL,lL,uL,cL,mL,fL,dL,hL,yL,bL,wL,vL,CL,IL,SL,kL,NL,TL,_L,EL,AL,$L,RL,FL,LL,BL,VL,jM,WL,HL,qL,KL,jL,XL,YL,ZL,JL,QL,t3,n3,i3,a3,l3,u3,c3,p3,m3,f3,d3,h3,g3,x3,y3,b3,w3,C3,I3,S3,k3,T3,_3,E3,A3,D3,$3,R3,F3,O3,KM,P3,UL,M3,L3,z3,XM,B3,V3,G3,W3,U3,H3,q3,K3,j3,X3,Z3,J3,Q3,tz,ez,rz,oz,iz,az,lz,uz,cz,hz,tL,gz,xz,yz,bz,OL,wz,Iz,Sz,kz,Nz,YM,Tz,_z,PL,pz,Ez,Az,Dz,rL,$z,Rz,Fz,Oz,Pz,Mz,Lz,zz,Bz,Vz,Gz,Wz,Uz,Hz,qz,DL,dz,Kz,jz,Xz,Yz,Zz,Jz,Qz,tB,rB,nB,sB,iB,aB,lB,uB,cB,fz,oL,pB,mB,fB,hB,gB,sL,xB,yB,bB,vz];for(let r of Eot)Cu(r);var Xt;(function(r){r[r.float32=0]="float32",r[r.int32=1]="int32",r[r.bool=2]="bool",r[r.string=3]="string",r[r.complex64=4]="complex64"})(Xt||(Xt={}));var uu;(function(r){r[r.linear=0]="linear",r[r.relu=1]="relu",r[r.relu6=2]="relu6",r[r.prelu=3]="prelu",r[r.leakyrelu=4]="leakyrelu",r[r.sigmoid=5]="sigmoid",r[r.elu=6]="elu"})(uu||(uu={}));var wB;function Aot(r){wB=r.wasm.cwrap(Ai,null,["number","array","number","number","array","number","number","number","number","number","number","number","number"])}function Dot(r){let{inputs:t,backend:e,attrs:n}=r,{a:o,b:s,bias:i,preluActivationWeights:a}=t;if(o.dtype!=="float32"||s.dtype!=="float32")throw new Error("_FusedMatMul for non non-float32 tensors not yet supported.");let{transposeA:u,transposeB:l,activation:c,leakyreluAlpha:p}=n,m=e.dataIdMap.get(o.dataId).id,f=e.dataIdMap.get(s.dataId).id,d=0;if(i!=null){let $=e.dataIdMap.get(i.dataId);if($.shape.length!==1)throw new Error(`_FusedMatMul only supports rank-1 bias but got rank ${$.shape.length}.`);d=$.id}let h=a==null?0:e.dataIdMap.get(a.dataId).id,g=uu[c];if(g==null)throw new Error(`${c} activation not yet supported for FusedConv2D in the wasm backend.`);let y=u?o.shape[2]:o.shape[1],b=l?s.shape[1]:s.shape[2],w=Mr.assertAndGetBroadcastShape(o.shape.slice(0,-2),s.shape.slice(0,-2)),v=e.makeOutput([...w,y,b],o.dtype),k=e.dataIdMap.get(v.dataId).id,E=new Uint8Array(new Int32Array(o.shape).buffer),D=new Uint8Array(new Int32Array(s.shape).buffer);return wB(m,E,o.shape.length,f,D,s.shape.length,u,l,g,d,h,p||0,k),v}var vB={kernelName:Ai,backendName:"wasm",setupFunc:Aot,kernelFunc:Dot};function ue(r,t){let e;function n(s){e=s.wasm.cwrap(r,null,["number","number","number"])}function o(s){let{backend:i,inputs:{x:a}}=s,u=i.dataIdMap.get(a.dataId).id,l=i.makeOutput(a.shape,t||a.dtype),c=i.dataIdMap.get(l.dataId).id;return x.sizeFromShape(l.shape)===0||e(u,Xt[a.dtype],c),l}return{kernelName:r,backendName:"wasm",setupFunc:n,kernelFunc:o}}var CB=ue(hi);function Ce(r,t,e){let n;function o(i){n=i.wasm.cwrap(r,null,["number","array","number","number","array","number","number","number"])}function s(i){let{backend:a,inputs:u}=i,{a:l,b:c}=u,p=a.dataIdMap.get(l.dataId).id,m=a.dataIdMap.get(c.dataId).id,f=e!=null?e:l.dtype,d=S.assertAndGetBroadcastShape(l.shape,c.shape),h=a.makeOutput(d,f);if(x.sizeFromShape(d)===0)return h;let g=new Uint8Array(new Int32Array(l.shape).buffer),y=new Uint8Array(new Int32Array(c.shape).buffer),b=a.dataIdMap.get(h.dataId).id;return(()=>n(p,g,l.shape.length,m,y,c.shape.length,Xt[l.dtype],b))(),h}return{kernelName:r,backendName:"wasm",setupFunc:o,kernelFunc:s}}var $ot=!0,IB=Ce(Kn,$ot);var SB;function Rot(r){SB=r.wasm.cwrap(jo,null,["array","number","number","number"])}function Fot(r){let{inputs:t,backend:e}=r,n=e.makeOutput(t[0].shape,t[0].dtype);if(x.sizeFromShape(n.shape)===0)return n;let o=t.map(a=>e.dataIdMap.get(a.dataId).id),s=new Uint8Array(new Int32Array(o).buffer),i=e.dataIdMap.get(n.dataId).id;return SB(s,o.length,Xt[n.dtype],i),n}var kB={kernelName:jo,backendName:"wasm",setupFunc:Rot,kernelFunc:Fot};function Gc(r){let{inputs:{x:t},backend:e}=r,n=e.makeOutput(t.shape,t.dtype),o=e.typedArrayFromHeap(t);return e.typedArrayFromHeap(n).set(o),n}var NB={kernelName:io,backendName:"wasm",kernelFunc:Gc};var TB;function Oot(r){TB=r.wasm.cwrap(Vs,null,["number","array","number","number","number","array","number"])}function eo(r){let{inputs:t,backend:e,attrs:n}=r,[o,s]=Mot(t.x.shape,n.perm),i=!0;for(let d=0;d<s.length;d++)s[d]!==d&&(i=!1);let a=Pot(t.x.shape,n.perm),u={dataId:t.x.dataId,shape:o,dtype:t.x.dtype};if(i){let d=Gc({inputs:t,backend:e});return d.shape=a,d}let l=e.makeOutput(a,u.dtype),c=e.dataIdMap.get(u.dataId).id,p=e.dataIdMap.get(l.dataId).id,m=new Uint8Array(new Int32Array(s).buffer),f=new Uint8Array(new Int32Array(u.shape).buffer);return TB(c,f,u.shape.length,Xt[u.dtype],p,m,s.length),l}function Pot(r,t){let e=new Array(r.length);for(let n=0;n<e.length;n++)e[n]=r[t[n]];return e}function Mot(r,t){let e=[],n=[];for(let o=0;o<r.length;++o)r[o]!==1&&e.push(r[o]),r[t[o]]!==1&&n.push(t[o]);for(let o=0;o<n.length;++o){let s=-1;for(let i=0;i<n.length;++i)n[i]>=o&&(s===-1||n[s]>n[i])&&(s=i);n[s]=o}return[e,n]}var _B={kernelName:Vs,backendName:"wasm",kernelFunc:eo,setupFunc:Oot};function Cn(r,t,e){let n=r.shape,o=r.shape.length,s=x.parseAxisParam(t,n),i=s,a=S.getAxesPermutation(i,o),u=null,l=!1;if(a!=null){let c=new Array(o);for(let f=0;f<c.length;f++)c[f]=n[a[f]];i=S.getInnerMostAxes(i.length,o),u=eo({inputs:{x:r},attrs:{perm:a},backend:e});let p=e.dataIdMap.get(r.dataId).id;e.dataIdMap.get(u.dataId).id!==p&&(l=!0)}return{transposed:u,originalAxes:s,axes:i,inputWasTransposed:l}}var EB;function Lot(r){EB=r.wasm.cwrap(Yi,null,["number, number, number"])}function zot(r){let{backend:t,inputs:e,attrs:n}=r,{axis:o,keepDims:s}=n,{x:i}=e,u=t.dataIdMap.get(i.dataId).id,l=i,{transposed:c,axes:p,originalAxes:m,inputWasTransposed:f}=Cn(i,o,t);if(f){let w=t.dataIdMap.get(c.dataId).id;l=c,u=w}let d=l.shape.length;S.assertAxesAreInnerMostDims("all",p,d);let[h,g]=S.computeOutAndReduceShapes(l.shape,p),y=x.sizeFromShape(g),b=t.makeOutput(h,i.dtype);if(x.sizeFromShape(l.shape)!==0){let w=t.dataIdMap.get(b.dataId).id;EB(u,y,w)}if(f&&t.disposeData(c.dataId),s){let w=S.expandShapeToKeepDim(b.shape,m);b.shape=w}return b}var AB={kernelName:Yi,backendName:"wasm",setupFunc:Lot,kernelFunc:zot};var DB;function Bot(r){DB=r.wasm.cwrap(Zi,null,["number, number, number"])}function Vot(r){let{backend:t,inputs:e,attrs:n}=r,{axis:o,keepDims:s}=n,{x:i}=e,u=t.dataIdMap.get(i.dataId).id,l=i,{transposed:c,axes:p,originalAxes:m,inputWasTransposed:f}=Cn(i,o,t);if(f){let w=t.dataIdMap.get(c.dataId).id;l=c,u=w}let d=l.shape.length;S.assertAxesAreInnerMostDims("any",p,d);let[h,g]=S.computeOutAndReduceShapes(l.shape,p),y=x.sizeFromShape(g),b=t.makeOutput(h,i.dtype);if(x.sizeFromShape(l.shape)!==0){let w=t.dataIdMap.get(b.dataId).id;DB(u,y,w)}if(f&&t.disposeData(c.dataId),s){let w=S.expandShapeToKeepDim(b.shape,m);b.shape=w}return b}var $B={kernelName:Zi,backendName:"wasm",setupFunc:Bot,kernelFunc:Vot};var RB;function Got(r){RB=r.wasm.cwrap(Xo,null,["number","number","number","number","number"])}function Wot(r){let{backend:t,inputs:e,attrs:n}=r,{axis:o}=n,{x:s}=e,i=t.dataIdMap.get(s.dataId).id,a=i,u=s,{transposed:l,axes:c,inputWasTransposed:p}=Cn(s,o,t);if(p){let y=t.dataIdMap.get(l.dataId).id;y!==i&&(u=l,a=y)}let m=u.shape.slice(0,-1),f=t.makeOutput(m,"int32"),d=t.dataIdMap.get(f.dataId).id,h=x.sizeFromShape(f.shape),g=u.shape[c[0]];return RB(a,Xt[u.dtype],h,g,d),p&&t.disposeData(l.dataId),f}var FB={kernelName:Xo,backendName:"wasm",kernelFunc:Wot,setupFunc:Got};var OB;function Uot(r){OB=r.wasm.cwrap(Yo,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function Hot(r){let{inputs:t,attrs:e,backend:n}=r,o=t.x,s=n.dataIdMap.get(o.dataId).id,{filterSize:i,strides:a,pad:u,dimRoundingMode:l}=e,c=S.computePool2DInfo(o.shape,i,a,1,u,l),p=c.filterHeight,m=c.filterWidth,f=c.padInfo.top,d=c.padInfo.right,h=c.padInfo.bottom,g=c.padInfo.left,y=c.strideHeight,b=c.strideWidth,w=c.inChannels;if(c.dataFormat!=="channelsLast")throw new Error(`wasm backend does not support dataFormat:'${c.dataFormat}'. Please use 'channelsLast'.`);if(c.dilationWidth!==1||c.dilationHeight!==1)throw new Error(`was backend only supports average pooling with dilation = [1, 1], got [${c.dilationHeight}, ${c.dilationWidth}].`);let v=n.makeOutput(c.outShape,"float32"),k=n.dataIdMap.get(v.dataId).id;return OB(s,o.shape[0],o.shape[1],o.shape[2],p,m,f,d,h,g,y,b,w,k),v}var PB={kernelName:Yo,backendName:"wasm",setupFunc:Uot,kernelFunc:Hot};function pr(r){let{inputs:t,attrs:e}=r,{x:n}=t,{shape:o}=e,s=x.sizeFromShape(n.shape),i=x.inferFromImplicitShape(o,s);return x.assert(s===x.sizeFromShape(i),()=>`new shape: ${i}, old shape: ${n.shape}. New shape and old shape must have the same number of elements.`),r.backend.incRef(n.dataId),{dataId:n.dataId,shape:i,dtype:n.dtype}}var MB={kernelName:Ii,backendName:"wasm",kernelFunc:pr};var LB;function qot(r){LB=r.wasm.cwrap(Zo,null,["number","array","number","number","array","number","number","number","number"])}function Kot(r){let{inputs:t,backend:e,attrs:n}=r,{a:o,b:s}=t,{transposeA:i,transposeB:a}=n;if(o.dtype!=="float32"||s.dtype!=="float32")throw new Error("BatchMatMul for non non-float32 tensors not yet supported.");let u=o.shape.length,l=s.shape.length,c=i?o.shape[u-2]:o.shape[u-1],p=a?s.shape[l-1]:s.shape[l-2],m=i?o.shape[u-1]:o.shape[u-2],f=a?s.shape[l-2]:s.shape[l-1],d=o.shape.slice(0,-2),h=s.shape.slice(0,-2),g=x.sizeFromShape(d),y=x.sizeFromShape(h),w=Mr.assertAndGetBroadcastShape(o.shape.slice(0,-2),s.shape.slice(0,-2)).concat([m,f]);x.assert(c===p,()=>`Error in matMul: inner shapes (${c}) and (${p}) of Tensors with shapes ${o.shape} and ${s.shape} and transposeA=${i} and transposeB=${a} must match.`);let v=i?[g,c,m]:[g,m,c],k=a?[y,f,p]:[y,p,f],E=pr({inputs:{x:o},backend:e,attrs:{shape:v}}),D=pr({inputs:{x:s},backend:e,attrs:{shape:k}}),$=e.dataIdMap.get(E.dataId).id,R=e.dataIdMap.get(D.dataId).id,M=i?E.shape[2]:E.shape[1],W=a?D.shape[1]:D.shape[2],U=Math.max(g,y),q=e.makeOutput([U,M,W],E.dtype),K=e.dataIdMap.get(q.dataId).id,j=new Uint8Array(new Int32Array(E.shape).buffer),Q=new Uint8Array(new Int32Array(D.shape).buffer);return LB($,j,E.shape.length,R,Q,D.shape.length,i,a,K),e.disposeData(E.dataId),e.disposeData(D.dataId),q.shape=w,q}var zB={kernelName:Zo,backendName:"wasm",setupFunc:qot,kernelFunc:Kot};function Mo(r){let{inputs:{x:t},attrs:{begin:e,size:n},backend:o}=r,[s,i]=Ve.parseSliceParams(t,e,n),a=Ve.isSliceContinous(t.shape,s,i),u=o.readSync(t.dataId),l=o.makeOutput(i,t.dtype),c=x.computeStrides(t.shape),p=o.dataIdMap.get(l.dataId);if(a){let d=Ve.computeFlatOffset(s,c);return t.dtype==="string"?p.stringBytes=u.slice(d,d+x.sizeFromShape(i)):o.typedArrayFromHeap(l).set(u.subarray(d,d+x.sizeFromShape(i))),l}if(t.dtype==="string"){let d=_c(u,s,i,t.shape,t.dtype);return p.stringBytes=d,l}let m=o.typedArrayFromHeap(l),f=t.shape.length;if(f===2)jot(u,c[0],m,s,i);else if(f===3)Xot(u,c[0],c[1],m,s,i);else if(f===4)Yot(u,c[0],c[1],c[2],m,s,i);else{let d=_c(u,s,i,t.shape,t.dtype);m.set(d)}return l}function jot(r,t,e,n,o){let s=0,i=n[0],a=n[1],u=i+o[0];for(let l=i;l<u;l++){let c=l*t+a;e.set(r.subarray(c,c+o[1]),s),s+=o[1]}}function Xot(r,t,e,n,o,s){let i=0,a=o[0],u=o[1],l=o[2],c=a+s[0],p=u+s[1];for(let m=a;m<c;m++)for(let f=u;f<p;f++){let d=m*t+f*e+l;n.set(r.subarray(d,d+s[2]),i),i+=s[2]}}function Yot(r,t,e,n,o,s,i){let a=0,u=s[0],l=s[1],c=s[2],p=u+i[0],m=l+i[1],f=c+i[2],d=s[3];for(let h=u;h<p;h++)for(let g=l;g<m;g++)for(let y=c;y<f;y++){let b=h*t+g*e+y*n+d;o.set(r.subarray(b,b+i[3]),a),a+=i[3]}}var BB={kernelName:ki,backendName:"wasm",kernelFunc:Mo};function Zot(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{blockShape:s,crops:i}=n,a=s.reduce((y,b)=>y*b),u=S.getReshaped(o.shape,s,a),l=S.getPermuted(u.length,s.length),c=S.getReshapedPermuted(o.shape,s,a),p=S.getSliceBeginCoords(i,s.length),m=S.getSliceSize(c,i,s.length),f=pr({inputs:{x:o},backend:e,attrs:{shape:u}}),d=eo({inputs:{x:f},backend:e,attrs:{perm:l}}),h=pr({inputs:{x:d},backend:e,attrs:{shape:c}}),g=Mo({inputs:{x:h},backend:e,attrs:{begin:p,size:m}});return e.disposeData(f.dataId),e.disposeData(d.dataId),e.disposeData(f.dataId),g}var VB={kernelName:gi,backendName:"wasm",kernelFunc:Zot};function hl(r){let{inputs:{x:t},attrs:{dtype:e},backend:n}=r,o=n.makeOutput(t.shape,e),s=n.typedArrayFromHeap(t);return n.typedArrayFromHeap(o).set(s),o}var GB={kernelName:oo,backendName:"wasm",kernelFunc:hl};var WB=ue(Jo);var UB;function Jot(r){UB=r.wasm.cwrap(so,null,["number","number","number","number"])}function Qot(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{clipValueMin:s,clipValueMax:i}=n,a=e.dataIdMap.get(o.dataId).id,u=e.makeOutput(o.shape,o.dtype),l=e.dataIdMap.get(u.dataId).id;return UB(a,s,i,l),u}var HB={kernelName:so,backendName:"wasm",setupFunc:Jot,kernelFunc:Qot};function pT(r){let{inputs:t,backend:e}=r,n=x.parseAxisParam(r.attrs.axis,t[0].shape)[0],o=S.computeOutShape(t.map(f=>f.shape),n),s=t.filter(f=>x.sizeFromShape(f.shape)>0);if(s.length===1)return Gc({inputs:{x:s[0]},backend:e});let i=e.makeOutput(o,t[0].dtype);if(x.sizeFromShape(o)===0)return i;let a=s.map(f=>f.shape);if(S.assertParamsConsistent(a,n),s[0].dtype==="string"){let f=s.map(w=>{let v=x.sizeFromShape(w.shape.slice(n));return pr({inputs:{x:w},backend:e,attrs:{shape:[-1,v]}})}),d=f.map(w=>({vals:e.readSync(w.dataId),shape:w.shape}));o=S.computeOutShape(f.map(w=>w.shape),1);let h=f[0].shape[0]===1,g=kc(d,o,t[0].dtype,h),y=S.computeOutShape(s.map(w=>w.shape),n);i.shape=y;let b=e.dataIdMap.get(i.dataId);return b.stringBytes=S.fromStringArrayToUint8(g),f.forEach(w=>e.disposeData(w.dataId)),i}let u=x.sizeFromShape(s[0].shape.slice(0,n)),l=0,c=s.map(f=>{let d=x.sizeFromShape(f.shape.slice(n));return l+=d,d}),p=s.map(f=>e.typedArrayFromHeap(f)),m=e.typedArrayFromHeap(i);for(let f=0;f<u;f++){let d=f*l;for(let h=0;h<p.length;h++){let g=c[h],y=f*g,b=p[h].subarray(y,y+g);m.set(b,d),d+=g}}return i}var qB={kernelName:xi,backendName:"wasm",kernelFunc:pT};var KB;function tst(r){KB=r.wasm.cwrap(Qo,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function est(r){let{inputs:t,attrs:e,backend:n}=r,{x:o,filter:s}=t,i=n.dataIdMap.get(o.dataId).id,a=n.dataIdMap.get(s.dataId).id,{strides:u,dilations:l,pad:c,dimRoundingMode:p,dataFormat:m}=e,f=S.convertConv2DDataFormat(m),d=S.computeConv2DInfo(o.shape,s.shape,u,l,c,p,!1,f),h=d.filterHeight,g=d.filterWidth,y=d.padInfo.top,b=d.padInfo.right,w=d.padInfo.bottom,v=d.padInfo.left,k=d.dilationHeight,E=d.dilationWidth,D=d.strideHeight,$=d.strideWidth,R=d.inChannels,M=d.outChannels,W=d.padInfo.type==="SAME"?1:0;if(d.dataFormat!=="channelsLast")throw new Error(`wasm backend Conv2D does not support dataFormat:'${d.dataFormat}'. Please use 'channelsLast'.`);let U=n.makeOutput(d.outShape,"float32"),q=n.dataIdMap.get(U.dataId).id;return KB(i,o.shape[0],o.shape[1],o.shape[2],a,h,g,y,b,w,v,W,k,E,D,$,R,M,q),U}var jB={kernelName:Qo,backendName:"wasm",setupFunc:tst,kernelFunc:est};var XB;function rst(r){XB=r.wasm.cwrap(ts,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 nst(r){let{backend:t,inputs:e,attrs:n}=r,{dy:o,filter:s}=e,{strides:i,pad:a,dataFormat:u,dimRoundingMode:l,inputShape:c}=n,p=1,m=S.convertConv2DDataFormat(u),f=S.computeConv2DInfo(c,s.shape,i,p,a,l,!1,m),{batchSize:d,filterHeight:h,filterWidth:g,inChannels:y,inHeight:b,inWidth:w,outChannels:v,outHeight:k,outWidth:E,strideHeight:D,strideWidth:$}=f,R=h-1-f.padInfo.top,M=g-1-f.padInfo.left,W=f.dataFormat==="channelsLast",U=x.computeStrides(f.inShape),q=x.computeStrides(o.shape),[K,j,Q]=x.computeStrides(s.shape),rt=U[0],X=W?U[1]:U[2],ot=W?U[2]:1,st=W?1:U[1],it=q[0],ft=W?q[1]:q[2],at=W?q[2]:1,xt=W?1:q[1],dt=t.makeOutput(f.inShape,"float32"),bt=t.dataIdMap.get(dt.dataId).id,kt=t.dataIdMap.get(o.dataId).id,At=t.dataIdMap.get(s.dataId).id;return XB(kt,At,d,h,g,b,w,y,k,E,v,D,$,R,M,K,j,Q,rt,X,ot,st,it,ft,at,xt,bt),dt}var YB={kernelName:ts,backendName:"wasm",setupFunc:rst,kernelFunc:nst};var ZB=ue(es);var JB=ue(rs);var mT;(function(r){r[r.bilinear=0]="bilinear",r[r.nearest=1]="nearest"})(mT||(mT={}));var QB;function ost(r){QB=r.wasm.cwrap(oa,null,["number","number","number","number","array","number","number","number","number","number"])}function sst(r){let{backend:t,inputs:e,attrs:n}=r,{method:o,extrapolationValue:s,cropSize:i}=n,{image:a,boxes:u,boxInd:l}=e,c=u.shape[0],[p,m]=i,f=[c,p,m,a.shape[3]],d=t.dataIdMap.get(a.dataId),h;a.dtype!=="float32"&&(h=hl({backend:t,inputs:{x:a},attrs:{dtype:"float32"}}),d=t.dataIdMap.get(h.dataId));let g=d.id,y=t.dataIdMap.get(u.dataId).id,b=t.dataIdMap.get(l.dataId).id,w=t.makeOutput(f,"float32"),v=t.dataIdMap.get(w.dataId).id,k=new Uint8Array(new Int32Array(a.shape).buffer);return QB(g,y,b,c,k,p,m,mT[o],s,v),h!=null&&t.disposeData(h.dataId),w}var tV={kernelName:oa,backendName:"wasm",setupFunc:ost,kernelFunc:sst};var eV;function ist(r){eV=r.wasm.cwrap(na,null,["number","number","number","number","number","number"])}function ast(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s,exclusive:i,reverse:a}=n,u=o.shape.length;x.assert(o.dtype==="float32"||o.dtype==="int32",()=>`cumprod does not support ${o.dtype} tensors in the WASM backend`);let l=S.getAxesPermutation([s],u),c=o;l!==null&&(c=eo({inputs:{x:o},attrs:{perm:l},backend:e}));let p=S.getInnerMostAxes(1,u)[0];S.assertAxesAreInnerMostDims("cumprod",[p],u);let m=e.makeOutput(c.shape,c.dtype),f=c.shape[p],d=e.dataIdMap.get(c.dataId).id,h=e.dataIdMap.get(m.dataId).id;eV(d,i?1:0,a?1:0,f,h,Xt[o.dtype]);let g=m;if(l!==null){let y=S.getUndoAxesPermutation(l);g=eo({inputs:{x:m},attrs:{perm:y},backend:e}),e.disposeData(c.dataId),e.disposeData(m.dataId)}return g}var rV={kernelName:na,backendName:"wasm",setupFunc:ist,kernelFunc:ast};var nV;function lst(r){nV=r.wasm.cwrap(ns,null,["number","number","number","number","number","number"])}function ust(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{axis:s,exclusive:i,reverse:a}=n,u=o.shape.length;x.assert(o.dtype==="float32"||o.dtype==="int32",()=>`cumsum does not support ${o.dtype} tensors in the WASM backend`);let l=S.getAxesPermutation([s],u),c=o;l!==null&&(c=eo({inputs:{x:o},attrs:{perm:l},backend:e}));let p=S.getInnerMostAxes(1,u)[0];S.assertAxesAreInnerMostDims("cumsum",[p],u);let m=e.makeOutput(c.shape,c.dtype),f=c.shape[p],d=e.dataIdMap.get(c.dataId).id,h=e.dataIdMap.get(m.dataId).id;nV(d,i?1:0,a?1:0,f,h,Xt[o.dtype]);let g=m;if(l!==null){let y=S.getUndoAxesPermutation(l);g=eo({inputs:{x:m},attrs:{perm:y},backend:e}),e.disposeData(c.dataId),e.disposeData(m.dataId)}return g}var oV={kernelName:ns,backendName:"wasm",setupFunc:lst,kernelFunc:ust};var sV;function cst(r){sV=r.wasm.cwrap(sa,null,["number","number","number","array","number","array","array","number","number"])}function pst(r){let{backend:t,inputs:e,attrs:n}=r,{x:o}=e,{blockSize:s,dataFormat:i}=n,a=o.shape[0],u=i==="NHWC"?o.shape[1]:o.shape[2],l=i==="NHWC"?o.shape[2]:o.shape[3],c=i==="NHWC"?o.shape[3]:o.shape[1],p=u*s,m=l*s,f=c/(s*s),d=i==="NHWC"?[a,p,m,f]:[a,f,p,m],h=t.makeOutput(d,"float32"),y=t.dataIdMap.get(o.dataId).id,b=new Uint8Array(new Int32Array(x.computeStrides(o.shape)).buffer),w=new Uint8Array(new Int32Array(d).buffer),v=new Uint8Array(new Int32Array(x.computeStrides(d)).buffer),k=t.dataIdMap.get(h.dataId).id;return sV(y,s,i==="NHWC"?1:0,b,o.shape.length-1,w,v,d.length,k),h}var iV={kernelName:sa,backendName:"wasm",setupFunc:cst,kernelFunc:pst};var aV;function mst(r){aV=r.wasm.cwrap(os,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function fst(r){let{inputs:t,attrs:e,backend:n}=r,{x:o,filter:s}=t,i=n.dataIdMap.get(o.dataId).id,a=n.dataIdMap.get(s.dataId).id,{strides:u,dilations:l,pad:c,dimRoundingMode:p}=e,m=l==null?[1,1]:l,f=S.computeConv2DInfo(o.shape,s.shape,u,m,c,p,!0),d=f.filterHeight,h=f.filterWidth,g=f.padInfo.top,y=f.padInfo.right,b=f.padInfo.bottom,w=f.padInfo.left,v=f.dilationHeight,k=f.dilationWidth,E=f.strideHeight,D=f.strideWidth,$=f.inChannels,R=f.outChannels,M=f.padInfo.type==="SAME"?1:0;if(f.dataFormat!=="channelsLast")throw new Error(`wasm backend DepthwiseConv2dNative does not support dataFormat:'${f.dataFormat}'. Please use 'channelsLast'.`);let W=n.makeOutput(f.outShape,"float32"),U=n.dataIdMap.get(W.dataId).id;return aV(i,o.shape[0],o.shape[1],o.shape[2],a,d,h,g,y,b,w,M,v,k,E,D,$,R,U),W}var lV={kernelName:os,backendName:"wasm",setupFunc:mst,kernelFunc:fst};var uV=ue(is);var dst=!1,cV=Ce(aa,dst,"bool");var pV=ue(as,"float32");function Lb(r){let{inputs:t,attrs:e,backend:n}=r,{input:o}=t,{dim:s}=e,i=o.shape.length,a=o.shape.slice(),u=s;return s<0&&(x.assert(-(i+1)<=s,()=>`Axis must be in the interval [${-(i+1)}, ${i}]`),u=i+s+1),a.splice(u,0,1),pr({inputs:{x:o},backend:n,attrs:{shape:a}})}var mV={kernelName:yi,backendName:"wasm",kernelFunc:Lb};function fT(r){let{attrs:{shape:t,value:e,dtype:n},backend:o}=r,s=o.makeOutput(t,n);return o.typedArrayFromHeap(s).fill(e),s}var fV={kernelName:Nl,backendName:"wasm",kernelFunc:fT};var dV;function hst(r){dV=r.wasm.cwrap(ua,null,["number","number","number","number","number","number"])}function gst(r){let{inputs:t,backend:e}=r,{image:n}=t,o=e.makeOutput(n.shape,n.dtype),s=e.dataIdMap.get(n.dataId).id,i=e.dataIdMap.get(o.dataId).id,[a,u,l,c]=n.shape;return dV(s,a,u,l,c,i),o}var hV={kernelName:ua,backendName:"wasm",kernelFunc:gst,setupFunc:hst};var gV=ue(ls);var xst=!1,xV=Ce(us,xst);var yV;function yst(r){yV=r.wasm.cwrap(cs,null,["number","number","number","number","number","number","number"])}function bst(r){let{backend:t,inputs:e,attrs:n}=r,{varianceEpsilon:o}=n,{x:s,mean:i,variance:a,offset:u,scale:l}=e,c=t.dataIdMap.get(s.dataId).id,p=t.dataIdMap.get(i.dataId).id,m=t.dataIdMap.get(a.dataId).id,f=u!=null?t.dataIdMap.get(u.dataId).id:0,d=l!=null?t.dataIdMap.get(l.dataId).id:0,h=t.makeOutput(s.shape,s.dtype);if(x.sizeFromShape(s.shape)===0)return h;let g=t.dataIdMap.get(h.dataId).id;return yV(c,p,m,f,d,o,g),h}var bV={kernelName:cs,backendName:"wasm",setupFunc:yst,kernelFunc:bst};var wV;function wst(r){wV=r.wasm.cwrap(Di,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function vst(r){let{inputs:t,attrs:e,backend:n}=r,{x:o,filter:s,bias:i,preluActivationWeights:a}=t,{strides:u,pad:l,dilations:c,dataFormat:p,dimRoundingMode:m,activation:f,leakyreluAlpha:d}=e,h=S.computeConv2DInfo(o.shape,s.shape,u,c,l,m),g=uu[f];if(g==null)throw new Error(`${f} activation not yet supported for FusedConv2D in the wasm backend.`);let y=n.dataIdMap.get(o.dataId).id,b=n.dataIdMap.get(s.dataId).id,w=h.outChannels,v=0;if(i!=null){let at=n.dataIdMap.get(i.dataId);if(at.shape.length!==1)throw new Error(`FusedConv2D only supports rank-1 bias but got rank ${at.shape.length}.`);if(at.shape[0]!==w)throw new Error(`FusedConv2D bias shape (${at.shape}) does not match the number of output channels (${w})`);v=at.id}let k=h.filterHeight,E=h.filterWidth,D=h.padInfo.top,$=h.padInfo.right,R=h.padInfo.bottom,M=h.padInfo.left,W=h.dilationHeight,U=h.dilationWidth,q=h.strideHeight,K=h.strideWidth,j=h.inChannels,Q=h.padInfo.type==="SAME"?1:0,rt=h.batchSize,X=h.inHeight,ot=h.inWidth;if(p!=="NHWC")throw new Error(`wasm backend FusedConv2D does not support dataFormat:'${p}'. Please use 'NHWC'.`);let st=n.makeOutput(h.outShape,"float32"),it=n.dataIdMap.get(st.dataId).id,ft=a==null?0:n.dataIdMap.get(a.dataId).id;return wV(y,rt,X,ot,b,k,E,v,D,$,R,M,Q,W,U,q,K,j,w,g,ft,d||0,it),st}var vV={kernelName:Di,backendName:"wasm",setupFunc:wst,kernelFunc:vst};var CV;function Cst(r){CV=r.wasm.cwrap($i,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function Ist(r){let{inputs:t,attrs:e,backend:n}=r,{x:o,filter:s,bias:i,preluActivationWeights:a}=t,{strides:u,pad:l,dilations:c,dataFormat:p,dimRoundingMode:m,activation:f,leakyreluAlpha:d}=e,h=S.computeConv2DInfo(o.shape,s.shape,u,c,l,m,!0),g=uu[f];if(g==null)throw new Error(`${f} activation not yet supported for FusedDepthwiseConv2D in the wasm backend.`);let y=n.dataIdMap.get(o.dataId).id,b=n.dataIdMap.get(s.dataId).id,w=h.outChannels,v=0;if(i!=null){let at=n.dataIdMap.get(i.dataId);if(at.shape.length!==1)throw new Error(`FusedDepthwiseConv2D only supports rank-1 bias but got rank ${at.shape.length}.`);if(at.shape[0]!==w)throw new Error(`FusedDepthwiseConv2D bias shape (${at.shape}) does not match the number of output channels (${w})`);v=at.id}let k=h.filterHeight,E=h.filterWidth,D=h.padInfo.top,$=h.padInfo.right,R=h.padInfo.bottom,M=h.padInfo.left,W=h.dilationHeight,U=h.dilationWidth,q=h.strideHeight,K=h.strideWidth,j=h.inChannels,Q=h.padInfo.type==="SAME"?1:0,rt=h.batchSize,X=h.inHeight,ot=h.inWidth;if(p!=="NHWC")throw new Error(`wasm backend FusedDepthwiseConv2D does not support dataFormat:'${p}'. Please use 'NHWC'.`);let st=n.makeOutput(h.outShape,"float32"),it=n.dataIdMap.get(st.dataId).id,ft=a==null?0:n.dataIdMap.get(a.dataId).id;return CV(y,rt,X,ot,b,k,E,v,D,$,R,M,Q,W,U,q,K,j,w,g,ft,d||0,it),st}var IV={kernelName:$i,backendName:"wasm",setupFunc:Cst,kernelFunc:Ist};var SV;function Sst(r){SV=r.wasm.cwrap(ca,null,["number","number","number","number","number","number","array","number"])}function kst(r){let{backend:t,inputs:e}=r,{params:n,indices:o}=e,[s,i,a,u]=hx.prepareAndValidate(n,o),l=t.makeOutput(s,n.dtype);if(i===0)return l;let c=o.shape,p=c[c.length-1],f=t.dataIdMap.get(n.dataId).id,h=t.dataIdMap.get(o.dataId).id,g=new Uint8Array(new Int32Array(u).buffer),y=t.dataIdMap.get(l.dataId).id;return SV(f,Xt[n.dtype],h,i,p,a,g,y),l}var kV={kernelName:ca,backendName:"wasm",setupFunc:Sst,kernelFunc:kst};var NV;function Nst(r){NV=r.wasm.cwrap("Gather",null,["number","number","array","number","number","number","array","number"])}function Tst(r){let{backend:t,inputs:e,attrs:n}=r,{x:o,indices:s}=e,{axis:i,batchDims:a}=n,u=x.parseAxisParam(i,o.shape)[0],l=t.readSync(s.dataId),c=o.shape[u];for(let R=0;R<l.length;++R){let M=l[R];x.assert(M<=c-1&&M>=0,()=>`GatherV2: the index value ${M} is not in [0, ${c-1}]`)}let p=S.segment_util.collectGatherOpShapeInfo(o,s,u,a),m=pr({inputs:{x:o},attrs:{shape:[p.batchSize,p.outerSize,p.dimSize,p.sliceSize]},backend:t}),f=x.sizeFromShape(s.shape),d=pr({inputs:{x:s},attrs:{shape:[p.batchSize,f/p.batchSize]},backend:t}),h=[p.batchSize,p.outerSize,f/p.batchSize,p.sliceSize],g=t.makeOutput(h,o.dtype);if(x.sizeFromShape(o.shape)===0)return g;let y=m.shape.length-1,w=t.dataIdMap.get(m.dataId).id,k=t.dataIdMap.get(d.dataId).id,E=t.dataIdMap.get(g.dataId).id,D=new Uint8Array(new Int32Array(x.computeStrides(m.shape)).buffer),$=new Uint8Array(new Int32Array(x.computeStrides(h)).buffer);return NV(w,Xt[o.dtype],D,y,k,p.batchSize,$,E),t.disposeData(m.dataId),t.disposeData(d.dataId),g.shape=p.outputShape,g}var TV={kernelName:bi,backendName:"wasm",setupFunc:Nst,kernelFunc:Tst};var _st=!1,_V=Ce(pa,_st,"bool");var Est=!1,EV=Ce(ps,Est,"bool");var AV;function Ast(r){AV=r.wasm.cwrap(ms,null,["number","number","number","number"])}function Dst(r){let{inputs:{x:t},attrs:{alpha:e},backend:n}=r,o=n.dataIdMap.get(t.dataId).id,s=n.makeOutput(t.shape,"float32");if(x.sizeFromShape(t.shape)!==0){let i=n.dataIdMap.get(s.dataId).id;AV(o,Xt[t.dtype],e,i)}return s}var DV={kernelName:ms,backendName:"wasm",setupFunc:Ast,kernelFunc:Dst};var $st=!1,$V=Ce(ha,$st,"bool");var Rst=!1,RV=Ce(ga,Rst,"bool");var FV=ue(fs);var Fst=!1,OV=Ce(ya,Fst,"bool");var PV;function Ost(r){PV=r.wasm.cwrap(ds,null,["number","number","number","number"])}function Pst(r){let{backend:t,inputs:e,attrs:n}=r,{reductionIndices:o,keepDims:s}=n,{x:i}=e,u=t.dataIdMap.get(i.dataId).id,l=i,{transposed:c,axes:p,originalAxes:m,inputWasTransposed:f}=Cn(i,o,t);if(f){let w=t.dataIdMap.get(c.dataId).id;l=c,u=w}let d=l.shape.length;S.assertAxesAreInnerMostDims("max",p,d);let[h,g]=S.computeOutAndReduceShapes(l.shape,p),y=x.sizeFromShape(g),b=t.makeOutput(h,i.dtype);if(x.sizeFromShape(l.shape)!==0){let w=t.dataIdMap.get(b.dataId).id;PV(u,Xt[i.dtype],y,w)}if(f&&t.disposeData(c.dataId),s){let w=S.expandShapeToKeepDim(b.shape,m);b.shape=w}return b}var MV={kernelName:ds,backendName:"wasm",setupFunc:Ost,kernelFunc:Pst};var Mst=!1,LV=Ce(hs,Mst);var zV;function Lst(r){zV=r.wasm.cwrap(gs,null,["number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number","number"])}function zst(r){let{inputs:t,attrs:e,backend:n}=r,o=t.x,s=n.dataIdMap.get(o.dataId).id;x.assert(o.dtype==="float32",()=>`Error in MaxPool: only float32 input is supported. Got ${o.dtype}.`);let{filterSize:i,strides:a,pad:u,dimRoundingMode:l}=e,c=S.computePool2DInfo(o.shape,i,a,1,u,l),p=c.filterHeight,m=c.filterWidth,f=c.padInfo.top,d=c.padInfo.right,h=c.padInfo.bottom,g=c.padInfo.left,y=c.dilationHeight,b=c.dilationWidth,w=c.strideHeight,v=c.strideWidth,k=c.inChannels,E=c.outChannels;if(c.dataFormat!=="channelsLast")throw new Error(`wasm backend does not support dataFormat:'${c.dataFormat}'. Please use 'channelsLast'.`);let D=n.makeOutput(c.outShape,"float32"),$=n.dataIdMap.get(D.dataId).id;return zV(s,o.shape[0],o.shape[1],o.shape[2],p,m,f,d,h,g,y,b,w,v,k,E,$),D}var BV={kernelName:gs,backendName:"wasm",setupFunc:Lst,kernelFunc:zst};var VV;function Bst(r){VV=r.wasm.cwrap(xs,null,["number, number, number"])}function Vst(r){let{backend:t,inputs:e,attrs:n}=r,{axis:o,keepDims:s}=n,{x:i}=e,a=t.dataIdMap.get(i.dataId).id,u=a,l=i,{transposed:c,axes:p,originalAxes:m,inputWasTransposed:f}=Cn(i,o,t),d=p;if(f){let v=t.dataIdMap.get(c.dataId).id;v!==a&&(l=c,u=v,d=S.getInnerMostAxes(d.length,l.shape.length))}S.assertAxesAreInnerMostDims("mean",d,l.shape.length);let[h,g]=S.computeOutAndReduceShapes(l.shape,d),y=x.sizeFromShape(g),b=l;l.dtype!=="float32"&&(b=hl({backend:t,inputs:{x:l},attrs:{dtype:"float32"}}),u=t.dataIdMap.get(b.dataId).id);let w=t.makeOutput(h,"float32");if(x.sizeFromShape(l.shape)!==0){let v=t.dataIdMap.get(w.dataId).id;VV(u,y,v)}if(f&&t.disposeData(c.dataId),s){let v=S.expandShapeToKeepDim(w.shape,m);w.shape=v}return l.dtype!=="float32"&&t.disposeData(b.dataId),w}var GV={kernelName:xs,backendName:"wasm",setupFunc:Bst,kernelFunc:Vst};var WV;function Gst(r){WV=r.wasm.cwrap(ys,null,["number","number","number","number"])}function Wst(r){let{backend:t,inputs:e,attrs:n}=r,{axis:o,keepDims:s}=n,{x:i}=e,a=t.dataIdMap.get(i.dataId).id,u=a,l=i,{transposed:c,axes:p,originalAxes:m,inputWasTransposed:f}=Cn(i,o,t);if(f){let w=t.dataIdMap.get(c.dataId).id;w!==a&&(l=c,u=w)}let d=l.shape.length;S.assertAxesAreInnerMostDims("min",p,d);let[h,g]=S.computeOutAndReduceShapes(l.shape,p),y=x.sizeFromShape(g),b=t.makeOutput(h,l.dtype);if(x.sizeFromShape(l.shape)!==0){let w=t.dataIdMap.get(b.dataId).id;WV(u,Xt[i.dtype],y,w)}if(f&&t.disposeData(c.dataId),s){let w=S.expandShapeToKeepDim(b.shape,m);b.shape=w}return b}var UV={kernelName:ys,backendName:"wasm",setupFunc:Gst,kernelFunc:Wst};var Ust=!1,HV=Ce(bs,Ust);var dT;(function(r){r[r.reflect=0]="reflect",r[r.symmetric=1]="symmetric"})(dT||(dT={}));var qV;function Hst(r){qV=r.wasm.cwrap(ws,null,["number","array","number","number","array","array","number","number"])}function qst(r){let{inputs:{x:t},backend:e,attrs:{paddings:n,mode:o}}=r,s=n.map((d,h)=>d[0]+t.shape[h]+d[1]),i=e.dataIdMap.get(t.dataId).id,a=e.makeOutput(s,t.dtype),u=e.dataIdMap.get(a.dataId).id,l=new Uint8Array(new Int32Array(t.shape).buffer),c=n.map(d=>d[0]),p=n.map(d=>d[1]),m=new Uint8Array(new Int32Array(c).buffer),f=new Uint8Array(new Int32Array(p).buffer);return qV(i,l,t.shape.length,Xt[t.dtype],m,f,dT[o],u),a}var KV={kernelName:ws,backendName:"wasm",kernelFunc:qst,setupFunc:Hst};var Kst=!0,jV=Ce(vs,Kst);var XV=ue(wi);function tf(r,t){let e=new Int32Array(r.wasm.HEAPU8.buffer,t,4),n=e[0],o=e[1],s=e[2],i=e[3];return r.wasm._free(t),{pSelectedIndices:n,selectedSize:o,pSelectedScores:s,pValidOutputs:i}}var YV;function jst(r){YV=r.wasm.cwrap(va,"number",["number","number","number","number","number"])}function Xst(r){let{backend:t,inputs:e,attrs:n}=r,{iouThreshold:o,maxOutputSize:s,scoreThreshold:i}=n,{boxes:a,scores:u}=e,l=t.dataIdMap.get(a.dataId).id,c=t.dataIdMap.get(u.dataId).id,p=YV(l,c,s,o,i),{pSelectedIndices:m,selectedSize:f,pSelectedScores:d,pValidOutputs:h}=tf(t,p);return t.wasm._free(d),t.wasm._free(h),t.makeOutput([f],"int32",m)}var ZV={kernelName:va,backendName:"wasm",setupFunc:jst,kernelFunc:Xst};var JV;function Yst(r){JV=r.wasm.cwrap(Ca,"number",["number","number","number","number","number","bool"])}function Zst(r){let{backend:t,inputs:e,attrs:n}=r,{iouThreshold:o,maxOutputSize:s,scoreThreshold:i,padToMaxOutputSize:a}=n,{boxes:u,scores:l}=e,c=t.dataIdMap.get(u.dataId).id,p=t.dataIdMap.get(l.dataId).id,m=JV(c,p,s,o,i,a),{pSelectedIndices:f,selectedSize:d,pSelectedScores:h,pValidOutputs:g}=tf(t,m);t.wasm._free(h);let y=t.makeOutput([d],"int32",f),b=t.makeOutput([],"int32",g);return[y,b]}var QV={kernelName:Ca,backendName:"wasm",setupFunc:Yst,kernelFunc:Zst};var tG;function Jst(r){tG=r.wasm.cwrap(Ia,"number",["number","number","number","number","number","number"])}function Qst(r){let{backend:t,inputs:e,attrs:n}=r,{iouThreshold:o,maxOutputSize:s,scoreThreshold:i,softNmsSigma:a}=n,{boxes:u,scores:l}=e,c=t.dataIdMap.get(u.dataId).id,p=t.dataIdMap.get(l.dataId).id,m=tG(c,p,s,o,i,a),{pSelectedIndices:f,selectedSize:d,pSelectedScores:h,pValidOutputs:g}=tf(t,m);t.wasm._free(g);let y=t.makeOutput([d],"int32",f),b=t.makeOutput([d],"float32",h);return[y,b]}var eG={kernelName:Ia,backendName:"wasm",setupFunc:Jst,kernelFunc:Qst};var tit=!1,rG=Ce(wa,tit,"bool");var nG;function eit(r){nG=r.wasm.cwrap(Cs,null,["number","number","number","number","number"])}function rit(r){let{inputs:t,backend:e,attrs:n}=r,{indices:o}=t,{depth:s,onValue:i,offValue:a}=n,u=e.makeOutput([...o.shape,s],"int32"),l=e.dataIdMap.get(u.dataId).id,p=e.dataIdMap.get(o.dataId).id;return nG(p,s,i,a,l),u}var oG={kernelName:Cs,backendName:"wasm",setupFunc:eit,kernelFunc:rit};function nit(r){let{inputs:{x:t},backend:e}=r,n=e.makeOutput(t.shape,t.dtype);return e.typedArrayFromHeap(n).fill(1),n}var sG={kernelName:vi,backendName:"wasm",kernelFunc:nit};function oit(r){let{inputs:t,backend:e,attrs:n}=r,{axis:o}=n;if(t.length===1)return Lb({inputs:{input:t[0]},backend:e,attrs:{dim:o}});let s=t[0].shape,i=t[0].dtype;t.forEach(c=>{x.assertShapesMatch(s,c.shape,"All tensors passed to stack must have matching shapes"),x.assert(i===c.dtype,()=>"All tensors passed to stack must have matching dtypes")});let a=[],u=t.map(c=>{let p=Lb({inputs:{input:c},backend:e,attrs:{dim:o}});return a.push(p),p}),l=pT({inputs:u,backend:e,attrs:{axis:o}});return a.forEach(c=>e.disposeData(c.dataId)),l}var iG={kernelName:Ci,backendName:"wasm",kernelFunc:oit};var aG;function sit(r){aG=r.wasm.cwrap(Is,null,["number","array","number","number","array","array","number","number"])}function iit(r){let{inputs:{x:t},backend:e,attrs:{paddings:n,constantValue:o}}=r,s=n.map((h,g)=>h[0]+t.shape[g]+h[1]);if(x.sizeFromShape(t.shape)===0)return fT({backend:e,attrs:{shape:s,value:o,dtype:t.dtype}});let i=e.dataIdMap.get(t.dataId).id,a=e.makeOutput(s,t.dtype),l=e.dataIdMap.get(a.dataId).id,c=new Uint8Array(new Int32Array(t.shape).buffer),p=n.map(h=>h[0]),m=n.map(h=>h[1]),f=new Uint8Array(new Int32Array(p).buffer),d=new Uint8Array(new Int32Array(m).buffer);return aG(i,c,t.shape.length,Xt[t.dtype],f,d,o,l),a}var zb={kernelName:Is,backendName:"wasm",kernelFunc:iit,setupFunc:sit};var ait=!1,lG=Ce(Ss,ait);var uG;function lit(r){uG=r.wasm.cwrap(ks,null,["number","number","number"])}function uit(r){let{inputs:t,backend:e}=r,{x:n,alpha:o}=t,s=e.dataIdMap.get(n.dataId).id,i=e.dataIdMap.get(o.dataId).id,a=s,u=n,l=u;u.dtype!=="float32"&&(l=hl({backend:e,inputs:{x:n},attrs:{dtype:"float32"}}),a=e.dataIdMap.get(l.dataId).id);let c=e.makeOutput(n.shape,"float32"),p=e.dataIdMap.get(c.dataId).id;return uG(a,i,p),u.dtype!=="float32"&&e.disposeData(l.dataId),c}var cG={kernelName:ks,backendName:"wasm",setupFunc:lit,kernelFunc:uit};var pG;function cit(r){pG=r.wasm.cwrap(Sa,null,["number","number","number","number"])}function pit(r){let{backend:t,inputs:e,attrs:n}=r,{axis:o,keepDims:s}=n,{x:i}=e,a=t.dataIdMap.get(i.dataId).id,u=a,l=i,{transposed:c,axes:p,originalAxes:m,inputWasTransposed:f}=Cn(i,o,t),d=p;if(f){let w=t.dataIdMap.get(c.dataId).id;w!==a&&(l=c,u=w,d=S.getInnerMostAxes(d.length,l.shape.length))}S.assertAxesAreInnerMostDims("prod",d,l.shape.length);let[h,g]=S.computeOutAndReduceShapes(l.shape,d),y=x.sizeFromShape(g),b=t.makeOutput(h,l.dtype);if(x.sizeFromShape(l.shape)!==0){let w=t.dataIdMap.get(b.dataId).id;pG(u,y,Xt[b.dtype],w)}if(f&&t.disposeData(c.dataId),s){let w=S.expandShapeToKeepDim(b.shape,m);b.shape=w}return b}var mG={kernelName:Sa,backendName:"wasm",setupFunc:cit,kernelFunc:pit};var mit=r=>{let{backend:t,attrs:e}=r,{start:n,stop:o,step:s,dtype:i}=e,a=Tc(n,o,s,i),u=t.makeOutput([a.length],i);return t.typedArrayFromHeap(u).set(a),u},fG={kernelName:El,backendName:"wasm",kernelFunc:mit};var fit=!0,dG=Ce(ss,fit);var hG=ue(Ns);var gG=ue(_s);var xG;function dit(r){xG=r.wasm.cwrap(Ts,null,["number","number","number","number","number","number","number","number","number","number"])}function hit(r){let{backend:t,inputs:e,attrs:n}=r,{images:o}=e,{alignCorners:s,halfPixelCenters:i,size:a}=n,[u,l]=a,[c,p,m,f]=o.shape,d=[c,u,l,f],h=t.dataIdMap.get(o.dataId),g;h.dtype!=="float32"&&(g=hl({backend:t,inputs:{x:o},attrs:{dtype:"float32"}}),h=t.dataIdMap.get(g.dataId));let y=h.id,b=t.makeOutput(d,"float32");if(x.sizeFromShape(o.shape)===0)return b;let w=t.dataIdMap.get(b.dataId).id;return xG(y,c,p,m,f,u,l,s?1:0,i?1:0,w),g!=null&&t.disposeData(g.dataId),b}var yG={kernelName:Ts,backendName:"wasm",setupFunc:dit,kernelFunc:hit};var bG;function git(r){bG=r.wasm.cwrap(Es,null,["number","array","number","array","number","number"])}function xit(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{dims:s}=n,i=x.parseAxisParam(s,o.shape);if(o.shape.length===0)return Gc({inputs:{x:o},backend:e});let a=e.makeOutput(o.shape,o.dtype),u=e.dataIdMap.get(o.dataId).id,l=e.dataIdMap.get(a.dataId).id,c=new Uint8Array(new Int32Array(i).buffer),p=new Uint8Array(new Int32Array(o.shape).buffer);bG(u,c,i.length,p,o.shape.length,l);let m=pr({inputs:{x:a},attrs:{shape:o.shape},backend:e});return e.disposeData(a.dataId),m}var wG={kernelName:Es,backendName:"wasm",kernelFunc:xit,setupFunc:git};var vG;function yit(r){vG=r.wasm.cwrap(Oa,null,["number","number","number","number","number","number","number","number","array","number","number"])}function bit(r){let{inputs:t,backend:e,attrs:n}=r,{image:o}=t,{radians:s,fillValue:i,center:a}=n,u=e.makeOutput(o.shape,o.dtype),l=e.dataIdMap.get(o.dataId).id,c=e.dataIdMap.get(u.dataId).id,[p,m,f,d]=o.shape,[h,g]=S.getImageCenter(a,m,f),y=i===0,b=255,w=typeof i=="number"?[i,i,i,y?0:b]:[...i,b],v=new Uint8Array(new Int32Array(w).buffer);return vG(l,p,m,f,d,s,h,g,v,w.length,c),u}var CG={kernelName:Oa,backendName:"wasm",kernelFunc:bit,setupFunc:yit};var IG=ue(As);var SG=ue(Ds);var kG;function wit(r){kG=r.wasm.cwrap(Na,null,["number","number","number","number","number","number","array","number","number"])}function vit(r){let{backend:t,inputs:e,attrs:n}=r,{indices:o,updates:s}=e,{shape:i}=n,a=t.makeOutput(i,s.dtype);if(x.sizeFromShape(i)===0)return a;let{sliceRank:u,numUpdates:l,sliceSize:c,strides:p,outputSize:m}=Ef.calculateShapes(s,o,i),d=t.dataIdMap.get(o.dataId).id,g=t.dataIdMap.get(s.dataId).id,y=new Uint8Array(new Int32Array(p).buffer),b=t.dataIdMap.get(a.dataId).id;return kG(d,g,Xt[s.dtype],u,l,c,y,m,b),a}var NG={kernelName:Na,backendName:"wasm",setupFunc:wit,kernelFunc:vit};var TG;function Cit(r){TG=r.wasm.cwrap("SelectV2",null,["number","number","number","number","number"])}function Iit(r){let{inputs:t,backend:e}=r,{condition:n,t:o,e:s}=t,i=e.dataIdMap.get(n.dataId).id,a=e.dataIdMap.get(o.dataId).id,u=e.dataIdMap.get(s.dataId).id,l=e.makeOutput(o.shape,o.dtype),c=e.dataIdMap.get(l.dataId).id,p=n.shape.length,m=o.shape.length,f=p===0||p>1||m===1?1:x.sizeFromShape(o.shape.slice(1));return TG(i,a,u,f,c),l}var _G={kernelName:Si,backendName:"wasm",kernelFunc:Iit,setupFunc:Cit};var EG;function Sit(r){EG=r.wasm.cwrap(Rs,null,["number","number"])}function kit(r){let{backend:t,inputs:{x:e}}=r,n=t.dataIdMap.get(e.dataId).id,o=t.makeOutput(e.shape,e.dtype),s=t.dataIdMap.get(o.dataId).id;return x.sizeFromShape(o.shape)===0||EG(n,s),o}var AG={kernelName:"Sigmoid",backendName:"wasm",setupFunc:Sit,kernelFunc:kit};var DG=ue($s);var $G;function Nit(r){$G=r.wasm.cwrap(Ps,null,["number","number","number","number"])}function Tit(r){let{backend:t,inputs:{logits:e},attrs:{dim:n}}=r,o=t.dataIdMap.get(e.dataId).id,s=t.makeOutput(e.shape,e.dtype),i=t.dataIdMap.get(s.dataId).id,a=e.shape[n],u=x.sizeFromShape(e.shape)/a;return x.sizeFromShape(s.shape)===0||$G(o,i,a,u),s}var RG={kernelName:Ps,backendName:"wasm",setupFunc:Nit,kernelFunc:Tit};function _it(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,{blockShape:s,paddings:i}=n,a=x.sizeFromShape(s),u=[[0,0]];u.push(...i);for(let E=1+s.length;E<o.shape.length;++E)u.push([0,0]);let l=zb.kernelFunc({inputs:{x:o},backend:e,attrs:{paddings:u,constantValue:0}}),c=S.getReshaped(l.shape,s,a,!1),p=S.getPermuted(c.length,s.length,!1),m=S.getReshapedPermuted(l.shape,s,a,!1),h=pr({inputs:{x:l},backend:e,attrs:{shape:c}}),b=eo({inputs:{x:h},backend:e,attrs:{perm:p}}),k=pr({inputs:{x:b},backend:e,attrs:{shape:m}});return e.disposeData(l.dataId),e.disposeData(h.dataId),e.disposeData(b.dataId),k}var FG={kernelName:Ni,backendName:"wasm",kernelFunc:_it};var OG;function Eit(r){OG=r.wasm.cwrap("SparseFillEmptyRows","number",["number","number","number","number","number","number","number","number","number","number","number","number"])}function Ait(r){let{backend:t,inputs:e}=r,{indices:n,values:o,denseShape:s,defaultValue:i}=e,a=n.shape[0],u=n.shape[1],l=t.readSync(s.dataId)[0],c=[a+l,u],p=t.dataIdMap.get(n.dataId).id,m=t.dataIdMap.get(o.dataId).id,f=t.dataIdMap.get(i.dataId).id,d=t.makeOutput(c,n.dtype),h=t.dataIdMap.get(d.dataId).id,g=t.makeOutput(c.slice(0,1),o.dtype),y=t.dataIdMap.get(g.dataId).id,b=t.makeOutput([l],"bool"),w=t.dataIdMap.get(b.dataId).id,v=t.makeOutput([a],n.dtype),k=t.dataIdMap.get(v.dataId).id,E=t.makeOutput([4],"int32"),D=t.dataIdMap.get(E.dataId).id,$=OG(p,m,Xt[o.dtype],a,l,u,f,h,y,w,k,D),R=t.readSync(E.dataId),M;switch(R[0]){case 1:{M=S.getSparseFillEmptyRowsIndicesDenseShapeMismatch(R[1]);break}case 2:{M=S.getSparseFillEmptyRowsNegativeIndexErrorMessage(R[1],R[2]);break}case 3:M=S.getSparseFillEmptyRowsOutOfRangeIndexErrorMessage(R[1],R[2],R[3]);break;default:M=""}if(t.disposeData(E.dataId),M)throw t.disposeData(d.dataId),t.disposeData(g.dataId),t.disposeData(b.dataId),t.disposeData(v.dataId),new Error(M);let W=d,U=g;return $!==c[0]&&(W=Mo({inputs:{x:d},attrs:{begin:0,size:[$,u]},backend:t}),U=Mo({inputs:{x:g},attrs:{begin:0,size:$},backend:t}),t.disposeData(d.dataId),t.disposeData(g.dataId)),[W,U,b,v]}var PG={kernelName:Dl,backendName:"wasm",setupFunc:Eit,kernelFunc:Ait};var MG;function Dit(r){MG=r.wasm.cwrap(Da,null,["number","number","number","number","number","number","number"])}function $it(r){let{backend:t,inputs:e}=r,{inputIndices:n,inputShape:o,newShape:s}=e;if(n.shape.length!==2)throw new Error(`Input indices should be a matrix but received shape
${n.shape}`);if(o.shape.length!==1)throw new Error(`Input shape should be a vector but received shape
${o.shape}`);if(s.shape.length!==1)throw new Error(`Target shape should be a vector but received shape ${s.shape}`);let i=t.dataIdMap.get(n.dataId).id,a=t.dataIdMap.get(o.dataId).id,u=t.dataIdMap.get(s.dataId).id,l=n.shape[0],c=x.sizeFromShape(s.shape),p=t.makeOutput([l,c],n.dtype),m=t.dataIdMap.get(p.dataId).id,f=t.makeOutput([c],s.dtype),d=t.dataIdMap.get(f.dataId).id,h=t.makeOutput([3],"int32"),g=t.dataIdMap.get(h.dataId).id;MG(i,a,u,l,m,d,g);let y=t.readSync(h.dataId),b;switch(y[0]){case 0:{b=S.getSparseReshapeMultipleNegativeOneOutputDimErrorMessage(y[1],y[2]);break}case 1:{b=S.getSparseReshapeNegativeOutputDimErrorMessage(y[1],y[2]);break}case 2:b=S.getSparseReshapeEmptyTensorZeroOutputDimErrorMessage();break;case 3:{let w=Array.from(t.readSync(o.dataId)),v=Array.from(t.readSync(f.dataId));b=S.getSparseReshapeInputOutputMultipleErrorMessage(w,v);break}case 4:{let w=Array.from(t.readSync(o.dataId)),v=Array.from(t.readSync(f.dataId));b=S.getSparseReshapeInputOutputMismatchErrorMessage(w,v);break}default:b=""}if(t.disposeData(h.dataId),b)throw t.disposeData(p.dataId),t.disposeData(f.dataId),new Error(b);return[p,f]}var LG={kernelName:Da,backendName:"wasm",setupFunc:Dit,kernelFunc:$it};var zG;function Bb(r){zG=r.wasm.cwrap("SparseSegmentReduction",null,["number","number","number","number","number","number","number","number","number"])}function Vb(r,t){let{backend:e,inputs:n}=r,{data:o,indices:s,segmentIds:i}=n,a=s.shape[0],u=e.readSync(i.dataId,a-1,a)[0],c=a>0?u+1:0;if(c<0)throw new Error(S.getSparseSegmentReductionNegativeSegmentIdsErrorMessage());let p=o.shape.slice();p[0]=c;let m=e.dataIdMap.get(o.dataId).id,f=e.dataIdMap.get(s.dataId).id,d=e.dataIdMap.get(i.dataId).id,h=e.makeOutput(p,o.dtype),g=e.dataIdMap.get(h.dataId).id,y=e.makeOutput([4],"int32"),b=e.dataIdMap.get(y.dataId).id;zG(m,Xt[o.dtype],o.shape[0],f,d,g,b,t,0);let w=e.readSync(y.dataId),v;switch(w[0]){case 0:{v=S.getSparseSegmentReductionNegativeSegmentIdsErrorMessage();break}case 1:{v=S.getSparseSegmentReductionNonIncreasingSegmentIdsErrorMessage();break}case 2:v=S.getSparseSegmentReductionSegmentIdOutOfRangeErrorMessage(w[1],w[2]);break;case 3:v=S.getSparseSegmentReductionIndicesOutOfRangeErrorMessage(w[1],w[2],w[3]);break;default:v=""}if(e.disposeData(y.dataId),v)throw e.disposeData(h.dataId),new Error(v);return h}function Rit(r){return Vb(r,!0)}var BG={kernelName:$l,backendName:"wasm",setupFunc:Bb,kernelFunc:Rit};function Fit(r){return Vb(r,!1)}var VG={kernelName:Rl,backendName:"wasm",setupFunc:Bb,kernelFunc:Fit};function Oit(r){let{inputs:t,attrs:e,backend:n}=r,{x:o}=t,{numOrSizeSplits:s,axis:i}=e,a=x.parseAxisParam(i,o.shape)[0],u=S.prepareSplitSize(o,s,a),l=new Array(o.shape.length).fill(0),c=o.shape.slice();return u.map(p=>{let m=[...c];m[a]=p;let f=Mo({inputs:{x:o},attrs:{begin:l,size:m},backend:n});return l[a]+=p,f})}var GG={kernelName:Ti,backendName:"wasm",kernelFunc:Oit};var WG=ue(Fs);var UG=ue(Fl);var Pit=!0,HG=Ce(Ms,Pit);var qG;function Mit(r){qG=r.wasm.cwrap(ao,null,["number","number","number","number"])}function Lit(r){let{backend:t,inputs:e,attrs:n}=r,{alpha:o}=n,{x:s}=e,i=t.dataIdMap.get(s.dataId).id,a=t.makeOutput(s.shape,s.dtype),u=t.dataIdMap.get(a.dataId).id;return qG(i,o,Xt[s.dtype],u),a}var KG={kernelName:ao,backendName:"wasm",setupFunc:Mit,kernelFunc:Lit};var jG;function zit(r){jG=r.wasm.cwrap($a,null,["number","array","number","array","array","array","array","array","number","number"])}function Bit(r){let{backend:t,inputs:e,attrs:n}=r,{x:o}=e,{begin:s,end:i,strides:a,beginMask:u,endMask:l,ellipsisMask:c,newAxisMask:p,shrinkAxisMask:m}=n,{finalShapeSparse:f,finalShape:d,isIdentity:h,sliceDim0:g,isSimpleSlice:y,begin:b,end:w,strides:v}=Ve.sliceInfo(o.shape,s,i,a,u,l,c,p,m),k;if(h)k=pr({inputs:{x:o},backend:t,attrs:{shape:d}});else if(g||y){x.assert(o.shape.length>=1,()=>`Input must have rank at least 1, got: ${o.shape.length}`);let E=Ve.computeOutShape(b,w,v),D=Mo({inputs:{x:o},backend:t,attrs:{begin:b,size:E}});k=pr({inputs:{x:D},backend:t,attrs:{shape:d}}),t.disposeData(D.dataId)}else{let E=t.makeOutput(f,"float32"),D=t.dataIdMap.get(o.dataId).id,$=new Uint8Array(new Int32Array(x.computeStrides(o.shape)).buffer),R=new Uint8Array(new Int32Array(b).buffer),M=new Uint8Array(new Int32Array(w).buffer),W=new Uint8Array(new Int32Array(v).buffer),U=new Uint8Array(new Int32Array(f).buffer),q=new Uint8Array(new Int32Array(x.computeStrides(f)).buffer),K=t.dataIdMap.get(E.dataId).id;jG(D,$,o.shape.length,R,M,W,U,q,f.length,K),k=pr({inputs:{x:E},backend:t,attrs:{shape:d}}),t.disposeData(E.dataId)}return k}var XG={kernelName:$a,backendName:"wasm",setupFunc:zit,kernelFunc:Bit};var Vit=!0,YG=Ce(Ls,Vit);var ZG;function Git(r){ZG=r.wasm.cwrap(Os,null,["number","number","number","number"])}function Wit(r){let{backend:t,inputs:e,attrs:n}=r,{axis:o,keepDims:s}=n,{x:i}=e,a=t.dataIdMap.get(i.dataId).id,u=a,l=i,{transposed:c,axes:p,originalAxes:m,inputWasTransposed:f}=Cn(i,o,t),d=p;if(f){let w=t.dataIdMap.get(c.dataId).id;w!==a&&(l=c,u=w,d=S.getInnerMostAxes(d.length,l.shape.length))}S.assertAxesAreInnerMostDims("sum",d,l.shape.length);let[h,g]=S.computeOutAndReduceShapes(l.shape,d),y=x.sizeFromShape(g),b=t.makeOutput(h,l.dtype);if(x.sizeFromShape(l.shape)!==0){let w=t.dataIdMap.get(b.dataId).id;ZG(u,y,Xt[b.dtype],w)}if(f&&t.disposeData(c.dataId),s){let w=S.expandShapeToKeepDim(b.shape,m);b.shape=w}return b}var JG={kernelName:Os,backendName:"wasm",setupFunc:Git,kernelFunc:Wit};var QG=ue(zs);var tW=ue(Bs);var eW;function Uit(r){eW=r.wasm.cwrap(jn,null,["number","array","number","array","number","number"])}function Hit(r){let{inputs:t,backend:e,attrs:n}=r,{x:o}=t,s=e.dataIdMap.get(o.dataId).id,{reps:i}=n,a=new Array(o.shape.length);for(let m=0;m<a.length;m++)a[m]=o.shape[m]*i[m];let u=new Uint8Array(new Int32Array(o.shape).buffer),l=new Uint8Array(new Int32Array(a).buffer),c=e.makeOutput(a,o.dtype),p=e.dataIdMap.get(c.dataId).id;return eW(s,u,o.shape.length,l,a.length,Xt[c.dtype],p),c}var rW={kernelName:jn,backendName:"wasm",setupFunc:Uit,kernelFunc:Hit};var nW;function qit(r){nW=r.wasm.cwrap(Ra,null,["number","array","number","number","number","bool","number","number"])}var Kit=({inputs:r,backend:t,attrs:e})=>{let{x:n}=r,{k:o,sorted:s}=e,i=t.dataIdMap.get(n.dataId).id,a=new Uint8Array(new Int32Array(n.shape).buffer),u=n.shape.slice();u[u.length-1]=o;let l=t.makeOutput(u,n.dtype),c=t.dataIdMap.get(l.dataId).id,p=t.makeOutput(u,"int32"),m=t.dataIdMap.get(p.dataId).id;return nW(i,a,n.shape.length,Xt[n.dtype],o,s,c,m),[l,p]},oW={kernelName:Ra,backendName:"wasm",setupFunc:qit,kernelFunc:Kit};var sW;function jit(r){sW=r.wasm.cwrap(Fa,null,["number","number","bool","number","number","number","number","number","number","array","number","number","number","number","number"])}function Xit(r){let{backend:t,inputs:e,attrs:n}=r,{image:o,transforms:s}=e,{interpolation:i,fillMode:a,fillValue:u,outputShape:l}=n,[c,p,m,f]=o.shape,[d,h]=l!=null?l:[p,m],g=[c,d,h,f],y=new Uint8Array(new Int32Array(x.computeStrides(o.shape)).buffer),b=t.makeOutput(g,o.dtype),w=t.dataIdMap.get(b.dataId).id,k=t.dataIdMap.get(o.dataId).id,D=t.dataIdMap.get(s.dataId).id,$=i==="nearest"?1:2,R;switch(a){case"constant":R=1;break;case"reflect":R=2;break;case"wrap":R=3;break;case"nearest":R=4;break;default:R=1;break}return sW(k,D,s.shape[0]>1,c,d,h,f,m,p,y,o.shape.length-1,$,R,u,w),b}var iW={kernelName:Fa,backendName:"wasm",setupFunc:jit,kernelFunc:Xit};function Yit(r){let{inputs:t,backend:e,attrs:n}=r,{value:o}=t,{axis:s}=n;s<0&&(s+=o.shape.length);let i=o.shape[s],a=o.shape.length,u=new Array(a-1),l=0;for(let f=0;f<a;f++)f!==s&&(u[l++]=o.shape[f]);let c=new Array(i),p=new Array(a).fill(0),m=o.shape.slice();m[s]=1;for(let f=0;f<c.length;f++)p[s]=f,c[f]=Mo({inputs:{x:o},attrs:{begin:p,size:m},backend:e});return c.map(({dataId:f,dtype:d})=>({dataId:f,dtype:d,shape:u}))}var aW={kernelName:_i,backendName:"wasm",kernelFunc:Yit};function Zit(r){let{inputs:{x:t},backend:e}=r,n=e.makeOutput(t.shape,t.dtype);return e.typedArrayFromHeap(n).fill(0),n}var lW={kernelName:Ei,backendName:"wasm",kernelFunc:Zit};var Jit=[vB,CB,IB,kB,AB,$B,FB,PB,zB,VB,GB,WB,HB,qB,jB,YB,ZB,JB,tV,rV,oV,iV,lV,uV,cV,pV,mV,fV,hV,gV,xV,bV,vV,IV,kV,TV,_V,EV,NB,DV,$V,RV,FV,OV,MV,LV,BV,GV,UV,HV,KV,jV,XV,ZV,QV,eG,rG,oG,sG,iG,zb,lG,cG,mG,fG,dG,hG,gG,MB,yG,wG,CG,IG,SG,NG,_G,AG,DG,BB,RG,FG,PG,LG,BG,VG,GG,WG,UG,HG,KG,XG,YG,JG,QG,tW,rW,oW,iW,_B,aW,lW];for(let r of Jit)Cu(r);var hT=V();hT.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])));hT.registerFlag("WASM_HAS_MULTITHREAD_SUPPORT",async()=>{if(hT.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(r){return!1}});var wT=xu(mW());var fW=`"use strict";var Module={};var ENVIRONMENT_IS_NODE=typeof process==="object"&&typeof process.versions==="object"&&typeof process.versions.node==="string";if(ENVIRONMENT_IS_NODE){var nodeWorkerThreads=require("worker_threads");var parentPort=nodeWorkerThreads.parentPort;parentPort.on("message",function(data){onmessage({data:data})});var fs=require("fs");Object.assign(global,{self:global,require:require,Module:Module,location:{href:__filename},Worker:nodeWorkerThreads.Worker,importScripts:function(f){(0,eval)(fs.readFileSync(f,"utf8"))},postMessage:function(msg){parentPort.postMessage(msg)},performance:global.performance||{now:function(){return Date.now()}}})}function threadPrintErr(){var text=Array.prototype.slice.call(arguments).join(" ");if(ENVIRONMENT_IS_NODE){fs.writeSync(2,text+"
");return}console.error(text)}function threadAlert(){var text=Array.prototype.slice.call(arguments).join(" ");postMessage({cmd:"alert",text:text,threadId:Module["_pthread_self"]()})}var err=threadPrintErr;self.alert=threadAlert;Module["instantiateWasm"]=((info,receiveInstance)=>{var instance=new WebAssembly.Instance(Module["wasmModule"],info);receiveInstance(instance);Module["wasmModule"]=null;return instance.exports});self.onmessage=(e=>{try{if(e.data.cmd==="load"){Module["wasmModule"]=e.data.wasmModule;Module["wasmMemory"]=e.data.wasmMemory;Module["buffer"]=Module["wasmMemory"].buffer;Module["ENVIRONMENT_IS_PTHREAD"]=true;if(typeof e.data.urlOrBlob==="string"){importScripts(e.data.urlOrBlob)}else{var objectUrl=URL.createObjectURL(e.data.urlOrBlob);importScripts(objectUrl);URL.revokeObjectURL(objectUrl)}WasmBackendModuleThreadedSimd(Module).then(function(instance){Module=instance})}else if(e.data.cmd==="run"){Module["__performance_now_clock_drift"]=performance.now()-e.data.time;Module["__emscripten_thread_init"](e.data.threadInfoStruct,0,0,1);Module["establishStackSpace"]();Module["PThread"].receiveObjectTransfer(e.data);Module["PThread"].threadInit();try{var result=Module["invokeEntryPoint"](e.data.start_routine,e.data.arg);if(Module["keepRuntimeAlive"]()){Module["PThread"].setExitStatus(result)}else{Module["__emscripten_thread_exit"](result)}}catch(ex){if(ex!="unwind"){if(ex instanceof Module["ExitStatus"]){if(Module["keepRuntimeAlive"]()){}else{Module["__emscripten_thread_exit"](ex.status)}}else{throw ex}}}}else if(e.data.cmd==="cancel"){if(Module["_pthread_self"]()){Module["__emscripten_thread_exit"](-1)}}else if(e.data.target==="setimmediate"){}else if(e.data.cmd==="processThreadQueue"){if(Module["_pthread_self"]()){Module["_emscripten_current_thread_process_queued_calls"]()}}else if(e.data.cmd==="processProxyingQueue"){if(Module["_pthread_self"]()){Module["_emscripten_proxy_execute_queue"](e.data.queue)}}else{err("worker.js received unknown command "+e.data.cmd);err(e.data)}}catch(ex){err("worker.js onmessage() captured an uncaught exception: "+ex);if(ex&&ex.stack)err(ex.stack);if(Module["__emscripten_thread_crashed"]){Module["__emscripten_thread_crashed"]()}throw ex}});`;var gW=xu(dW());var Hb=class extends mi{constructor(t){super();this.wasm=t,this.dataIdNextNumber=1,this.wasm.tfjs.initWithThreadsCount(yW),vT=this.wasm.tfjs.getThreadsCount(),this.dataIdMap=new wl(this,fo())}write(t,e,n){let o={id:this.dataIdNextNumber++};return this.move(o,t,e,n,1),o}numDataIds(){return this.dataIdMap.numDataIds()}async time(t){let e=x.now();return t(),{kernelMs:x.now()-e}}move(t,e,n,o,s){let i=this.dataIdNextNumber++;if(o==="string"){let c=e;this.dataIdMap.set(t,{id:i,stringBytes:c,shape:n,dtype:o,memoryOffset:null,refCount:s});return}let a=x.sizeFromShape(n),u=a*x.bytesPerElement(o),l=this.wasm._malloc(u);this.dataIdMap.set(t,{id:i,memoryOffset:l,shape:n,dtype:o,refCount:s}),this.wasm.tfjs.registerTensor(i,a,l),e!=null&&this.wasm.HEAPU8.set(new Uint8Array(e.buffer,e.byteOffset,u),l)}async read(t){return this.readSync(t)}readSync(t,e,n){let{memoryOffset:o,dtype:s,shape:i,stringBytes:a}=this.dataIdMap.get(t);if(s==="string")return(e==null||e===0)&&(n==null||n>=a.length)?a:a.slice(e,n);e=e||0,n=n||x.sizeFromShape(i);let u=x.bytesPerElement(s),l=this.wasm.HEAPU8.slice(o+e*u,o+n*u);return tat(l.buffer,s)}disposeData(t,e=!1){if(this.dataIdMap.has(t)){let n=this.dataIdMap.get(t);if(n.refCount--,!e&&n.refCount>0)return!1;this.wasm._free(n.memoryOffset),this.wasm.tfjs.disposeData(n.id),this.dataIdMap.delete(t)}return!0}refCount(t){return this.dataIdMap.has(t)?this.dataIdMap.get(t).refCount:0}incRef(t){let e=this.dataIdMap.get(t);e!=null&&e.refCount++}floatPrecision(){return 32}getMemoryOffset(t){return this.dataIdMap.get(t).memoryOffset}dispose(){this.wasm.tfjs.dispose(),"PThread"in this.wasm&&this.wasm.PThread.terminateAllThreads(),this.wasm=null}memory(){return{unreliable:!1}}makeOutput(t,e,n){let o;if(n==null)o=this.write(null,t,e);else{let s=this.dataIdNextNumber++;o={id:s},this.dataIdMap.set(o,{id:s,memoryOffset:n,shape:t,dtype:e,refCount:1});let i=x.sizeFromShape(t);this.wasm.tfjs.registerTensor(s,i,n)}return{dataId:o,shape:t,dtype:e}}typedArrayFromHeap({shape:t,dtype:e,dataId:n}){let o=this.wasm.HEAPU8.buffer,{memoryOffset:s}=this.dataIdMap.get(n),i=x.sizeFromShape(t);switch(e){case"float32":return new Float32Array(o,s,i);case"int32":return new Int32Array(o,s,i);case"bool":return new Uint8Array(o,s,i);default:throw new Error(`Unknown dtype ${e}`)}}};function Qit(r){return(t,e)=>(x.fetch(r,{credentials:"same-origin"}).then(n=>{n.ok||t.env.a(`failed to load wasm binary file at '${r}'`),n.arrayBuffer().then(o=>{WebAssembly.instantiate(o,t).then(s=>{e(s.instance,s.module)})})}),{})}function hW(r,t,e){if(Ub!=null)return Ub;let n="tfjs-backend-wasm.wasm";return r&&t?n="tfjs-backend-wasm-threaded-simd.wasm":r&&(n="tfjs-backend-wasm-simd.wasm"),bg!=null&&bg[n]!=null?bg[n]:e+n}async function xW(){let[r,t]=await Promise.all([V().getAsync("WASM_HAS_SIMD_SUPPORT"),V().getAsync("WASM_HAS_MULTITHREAD_SUPPORT")]);return new Promise((e,n)=>{let o={};o.locateFile=(a,u)=>{if(a.endsWith(".worker.js")){let l=fW.replace(/\n/g,"\\n"),c=new Blob([l],{type:"application/javascript"});return URL.createObjectURL(c)}return a.endsWith(".wasm")?hW(r,t,yg!=null?yg:u):u+a},CT&&(o.instantiateWasm=Qit(hW(r,t,yg!=null?yg:"")));let s=!1;o.onAbort=()=>{if(s||wg)return;wg=!0,n({message:"Make sure the server can serve the `.wasm` file relative to the bundled js file. For more details see https://github.com/tensorflow/tfjs/blob/master/tfjs-backend-wasm/README.md#using-bundlers"})};let i;t&&r&&Ub==null?(o.mainScriptUrlOrBlob=new Blob(["var WasmBackendModuleThreadedSimd = "+wT.default.toString()],{type:"text/javascript"}),i=(0,wT.default)(o)):i=(0,gW.default)(o),i.then(a=>{s=!0,wg=!1;let u=null;a.tfjs={init:a.cwrap("init",null,[]),initWithThreadsCount:a.cwrap("init_with_threads_count",null,["number"]),getThreadsCount:a.cwrap("get_threads_count","number",[]),registerTensor:a.cwrap("register_tensor",null,["number","number","number"]),disposeData:a.cwrap("dispose_data",u,["number"]),dispose:a.cwrap("dispose",u,[])},e({wasm:a})})})}function tat(r,t){switch(t){case"float32":return new Float32Array(r);case"int32":return new Int32Array(r);case"bool":return new Uint8Array(r);default:throw new Error(`Unknown dtype ${t}`)}}var eat=["tfjs-backend-wasm.wasm","tfjs-backend-wasm-simd.wasm","tfjs-backend-wasm-threaded-simd.wasm"],Ub=null,yg=null,bg={},wg=!1,CT=!1;function rat(r,t=!1){if(g0("setWasmPath has been deprecated in favor of setWasmPaths and will be removed in a future release."),wg)throw new Error("The WASM backend was already initialized. Make sure you call `setWasmPath()` before you call `tf.setBackend()` or `tf.ready()`");Ub=r,CT=t}function nat(r,t=!1){if(wg)throw new Error("The WASM backend was already initialized. Make sure you call `setWasmPaths()` before you call `tf.setBackend()` or `tf.ready()`");if(typeof r=="string")yg=r;else{bg=r;let e=eat.filter(n=>bg[n]==null);if(e.length>0)throw new Error(`There were no entries found for the following binaries: ${e.join(",")}. Please either call setWasmPaths with a map providing a path for each binary, or with a string indicating the directory where all the binaries can be found.`)}CT=t}var yW=-1,vT=-1;function oat(r){yW=r}function sat(){if(vT===-1)throw new Error("WASM backend not initialized.");return vT}var iat="3.15.0";var aat=2;Xp("wasm",async()=>{let{wasm:r}=await xW();return new Hb(r)},aat);var lat="3.15.0",uat="3.15.0",cat="3.15.0",pat="3.15.0",mat="3.15.0",fat="3.15.0",dat="3.15.0",hat="3.15.0",gat={tfjs:lat,"tfjs-core":uat,"tfjs-data":cat,"tfjs-layers":pat,"tfjs-converter":mat,"tfjs-backend-cpu":fat,"tfjs-backend-webgl":dat,"tfjs-backend-wasm":hat};export{hi as Abs,ji as Acos,Xi as Acosh,sc as AdadeltaOptimizer,ic as AdagradOptimizer,ac as AdamOptimizer,lc as AdamaxOptimizer,Kn as Add,jo as AddN,Yi as All,Zi as Any,Xo as ArgMax,vl as ArgMin,Ji as Asin,Qi as Asinh,ta as Atan,ra as Atan2,ea as Atanh,Yo as AvgPool,Cl as AvgPool3D,up as AvgPool3DGrad,lp as AvgPoolGrad,Hb as BackendWasm,Zo as BatchMatMul,gi as BatchToSpaceND,cp as Bincount,pp as BroadcastArgs,YT as BroadcastTo,uI as Callback,vC as CallbackList,oo as Cast,Jo as Ceil,so as ClipByValue,mp as Complex,Il as ComplexAbs,xi as Concat,Qo as Conv2D,fp as Conv2DBackpropFilter,ts as Conv2DBackpropInput,Sl as Conv3D,dp as Conv3DBackpropFilterV2,hp as Conv3DBackpropInputV2,es as Cos,rs as Cosh,oa as CropAndResize,na as Cumprod,ns as Cumsum,IC as CustomCallback,wl as DataStorage,gp as DenseBincount,sa as DepthToSpace,os as DepthwiseConv2dNative,xp as DepthwiseConv2dNativeBackpropFilter,yp as DepthwiseConv2dNativeBackpropInput,bp as Diag,kl as Dilation2D,bf as Dilation2DBackpropFilter,yf as Dilation2DBackpropInput,Av as ENV,cI as EarlyStopping,wp as Einsum,is as Elu,vp as EluGrad,ex as Environment,aa as Equal,ia as Erf,as as Exp,yi as ExpandDims,la as Expm1,Cp as FFT,Nl as Fill,ua as FlipLeftRight,ls as Floor,us as FloorDiv,wf as FromPixels,cs as FusedBatchNorm,Di as FusedConv2D,$i as FusedDepthwiseConv2D,Jm as GPGPUContext,ca as GatherNd,bi as GatherV2,BI as GraphModel,pa as Greater,ps as GreaterEqual,CC as History,Ip as IFFT,io as Identity,Sp as Imag,_e as InputSpec,ma as IsFinite,fa as IsInf,da as IsNan,mi as KernelBackend,Tl as LRN,Np as LRNGrad,ty as LayerVariable,Qn as LayersModel,ms as LeakyRelu,ha as Less,ga as LessEqual,kp as LinSpace,fs as Log,xa as Log1p,ZT as LogSoftmax,ya as LogicalAnd,wu as LogicalNot,vu as LogicalOr,Oc as MathBackendWebGL,ds as Max,gs as MaxPool,_l as MaxPool3D,_p as MaxPool3DGrad,Tp as MaxPoolGrad,Ep as MaxPoolWithArgmax,hs as Maximum,xs as Mean,ys as Min,bs as Minimum,ws as MirrorPad,ba as Mod,uc as MomentumOptimizer,Ap as Multinomial,vs as Multiply,wi as Neg,va as NonMaxSuppressionV3,Ca as NonMaxSuppressionV4,Ia as NonMaxSuppressionV5,wa as NotEqual,N1 as OP_SCOPE_SUFFIX,Cs as OneHot,vi as OnesLike,jr as Optimizer,Mi as OptimizerConstructors,Ci as Pack,Is as PadV2,Cat as Pool,Ss as Pow,ks as Prelu,Sa as Prod,cc as RMSPropOptimizer,zn as RNN,El as Range,Vv as Rank,Dp as Real,ss as RealDiv,ka as Reciprocal,Qe as Reduction,Ns as Relu,_s as Relu6,Ii as Reshape,Ts as ResizeBilinear,Rp as ResizeBilinearGrad,Al as ResizeNearestNeighbor,$p as ResizeNearestNeighborGrad,Es as Reverse,Oa as RotateWithOffset,As as Round,Ds as Rsqrt,el as SGDOptimizer,Na as ScatterNd,Si as Select,Ta as Selu,sl as Sequential,Rs as Sigmoid,Ea as Sign,$s as Sin,_a as Sinh,ki as Slice,Ps as Softmax,Aa as Softplus,Ni as SpaceToBatchND,Dl as SparseFillEmptyRows,Da as SparseReshape,$l as SparseSegmentMean,Rl as SparseSegmentSum,Fp as SparseToDense,Ti as SplitV,Fs as Sqrt,Fl as Square,Ms as SquaredDifference,ao as Step,$a as StridedSlice,Op as StringNGrams,Pp as StringSplit,Mp as StringToHashBucketFast,Ls as Sub,Os as Sum,gn as SymbolicTensor,zs as Tan,Bs as Tanh,Vt as Tensor,me as TensorBuffer,jn as Tile,Ra as TopK,Fa as Transform,Vs as Transpose,Lp as Unique,_i as Unpack,Ol as UnsortedSegmentSum,Ll as Variable,Ei as ZerosLike,Ai as _FusedMatMul,Se as abs,Df as acos,$f as acosh,J as add,x0 as addN,Ru as all,Gl as any,Hs as argMax,Rf as argMin,Ff as asin,Of as asinh,Pf as atan,Mf as atan2,Lf as atanh,Va as avgPool,zf as avgPool3d,k_ as backend,S as backend_util,PH as basicLSTMCell,ho as batchNorm,v0 as batchNorm2d,C0 as batchNorm3d,I0 as batchNorm4d,Ga as batchToSpaceND,Bf as bincount,QTt as booleanMaskAsync,S0 as broadcastArgs,Wa as broadcastTo,Mr as broadcast_util,dx as browser,It as buffer,u7 as callbacks,Z as cast,Vf as ceil,br as clipByValue,In as clone,Mn as complex,se as concat,k0 as concat1d,N0 as concat2d,T0 as concat3d,_0 as concat4d,L2 as constraints,Pu as conv1d,ln as conv2d,Mu as conv2dTranspose,Gf as conv3d,E0 as conv3dTranspose,Eat as copyRegisteredKernels,Ua as cos,Lu as cosh,Ex as cosineWindow,Wf as cumprod,zu as cumsum,un as customGrad,wR as data,A0 as denseBincount,g0 as deprecationWarn,Uf as depthToSpace,Ks as depthwiseConv2d,f7 as deregisterOp,zl as device_util,mq as diag,Hf as dilation2d,Jct as disableDeprecationWarnings,_t as dispose,Qct as disposeVariables,ct as div,qf as divNoNan,D0 as dot,X_ as dropout,$0 as einsum,js as elu,Zct as enableDebugMode,Yct as enableProdMode,Y_ as enclosingPowerOfTwo,fo as engine,V as env,Sr as equal,Kf as erf,Ze as exp,fr as expandDims,jf as expm1,Jp as eye,Qa as fft,Xs as fill,spt as findBackend,ipt as findBackendFactory,Ys as floor,$u as floorDiv,qM as forceHalfFloat,wo as fused,go as gather,K_ as gatherND,hx as gather_util,npt as getBackend,Rv as getGradient,Cf as getKernel,rx as getKernelsForBackend,sat as getThreadsCount,Dk as gpgpu_util,Vq as grad,Gq as grads,Ge as greater,kn as greaterEqual,Oi as ifft,Bu as imag,pn as image,r_t as inTopKAsync,oD as initializers,$C as input,Pr as io,Ju as irfft,R0 as isFinite,F0 as isInf,Xf as isNaN,Me as keep,Vr as kernel_impls,VD as layers,Ha as leakyRelu,Vu as less,Nn as lessEqual,OE as linalg,O0 as linspace,iZ as loadGraphModel,w8 as loadLayersModel,Yf as localResponseNormalization,wr as log,qa as log1p,P0 as logSigmoid,Gu as logSoftmax,Jf as logSumExp,Nr as logicalAnd,Ka as logicalNot,Wu as logicalOr,V0 as logicalXor,RFt as losses,Mt as matMul,i_ as math,$r as max,ja as maxPool,Qf as maxPool3d,G0 as maxPoolWithArgmax,cn as maximum,xe as mean,Af as memory,cK as meshgrid,GD as metrics,Wl as min,Zs as minimum,td as mirrorPad,ed as mod,y8 as model,WD as models,Qp as moments,I1t as movingAverage,P as mul,yK as multiRNNCell,W0 as multinomial,Zt as neg,fd as nextFrame,nm as norm,bo as notEqual,Us as oneHot,lr as ones,dr as onesLike,N as op,IK as outerProduct,qr as pad,NK as pad1d,_K as pad2d,AK as pad3d,$K as pad4d,U0 as pool,Kr as pow,Ya as prelu,i0 as print,Uu as prod,tpt as profile,VK as rand,XK as randomGamma,kx as randomNormal,Js as randomUniform,Za as range,rpt as ready,Ul as real,rd as reciprocal,Xp as registerBackend,v8 as registerCallbackConstructor,QT as registerGradient,Cu as registerKernel,m7 as registerOp,UD as regularizers,Tr as relu,qu as relu6,opt as removeBackend,O as reshape,sr as reverse,oj as reverse1d,ij as reverse2d,lj as reverse3d,cj as reverse4d,tl as rfft,Ku as round,ju as rsqrt,pt as scalar,H_ as scatterND,Ef as scatter_util,Xu as selu,nd as separableConv2d,b8 as sequential,et as serialization,oH as setBackend,apt as setPlatform,oat as setThreadsCount,rat as setWasmPath,nat as setWasmPaths,zS as setWebGLContext,eC as setdiff1dAsync,Lr as sigmoid,od as sign,iFt as signal,Yu as sin,Zu as sinh,Rt as slice,sd as slice1d,Nx as slice2d,id as slice3d,em as slice4d,Ve as slice_util,Ja as softmax,xo as softplus,Xa as spaceToBatchND,md as sparse,_x as sparseToDense,eFt as spectral,ur as split,be as sqrt,Wt as square,Qu as squaredDifference,zr as squeeze,Je as stack,Qs as step,ad as stridedSlice,zx as string,lt as sub,mt as sum,_u as sumOutType,ld as tan,qs as tanh,Ar as tensor,Re as tensor1d,Pi as tensor2d,c0 as tensor3d,Pj as tensor4d,Mj as tensor5d,Lj as tensor6d,po as tensor_util,I_ as test_util,B as tidy,kr as tile,ept as time,ud as topk,pc as train,Gt as transpose,tc as truncatedNormal,rm as unique,_at as unregisterGradient,Tat as unregisterKernel,cd as unsortedSegmentSum,vr as unstack,ar as upcastType,x as util,Wq as valueAndGrad,Uq as valueAndGrads,rC as variable,vx as variableGrads,gat as version,I$ as version_converter,S_ as version_core,Cm as version_layers,iat as version_wasm,HM as version_webgl,XSe as webgl,Um as webgl_util,Ne as where,pd as whereAsync,ye as zeros,St as zerosLike};
/**
* @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
* Copyright 2021 Google LLC. All Rights Reserved.
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* https://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
* =============================================================================
*/
/**
* @license
* Copyright 2022 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 2022 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 See the LICENSE file. */
Reference in New Issue View Git Blame Copy Permalink