diff --git "a/assets/worker-tV2Tfrl_.js" "b/assets/worker-tV2Tfrl_.js" new file mode 100644--- /dev/null +++ "b/assets/worker-tV2Tfrl_.js" @@ -0,0 +1,2787 @@ +var F_=Object.defineProperty;var D_=(dr,Xr,ni)=>Xr in dr?F_(dr,Xr,{enumerable:!0,configurable:!0,writable:!0,value:ni}):dr[Xr]=ni;var me=(dr,Xr,ni)=>D_(dr,typeof Xr!="symbol"?Xr+"":Xr,ni);(function(){"use strict";var dr={},Xr={"./node_modules/onnxruntime-web/dist/ort-wasm-simd-threaded.jsep.wasm":(Le,A,r)=>{Le.exports=r.p+"ort-wasm-simd-threaded.jsep.wasm"},"./node_modules/onnxruntime-web/dist/ort.bundle.min.mjs?46eb":(Le,A,r)=>{Le.exports=r.p+"ort.bundle.min.mjs"},"?2ce3":()=>{},"?7a2c":()=>{},"?a42a":()=>{},"?2b25":()=>{},"?569f":()=>{},"?3f59":()=>{},"?154a":()=>{},"./node_modules/@huggingface/jinja/dist/index.js":(Le,A,r)=>{r.r(A),r.d(A,{Environment:()=>st,Interpreter:()=>dt,Template:()=>ht,parse:()=>ye,tokenize:()=>y});var _=Object.freeze({Text:"Text",NumericLiteral:"NumericLiteral",BooleanLiteral:"BooleanLiteral",NullLiteral:"NullLiteral",StringLiteral:"StringLiteral",Identifier:"Identifier",Equals:"Equals",OpenParen:"OpenParen",CloseParen:"CloseParen",OpenStatement:"OpenStatement",CloseStatement:"CloseStatement",OpenExpression:"OpenExpression",CloseExpression:"CloseExpression",OpenSquareBracket:"OpenSquareBracket",CloseSquareBracket:"CloseSquareBracket",OpenCurlyBracket:"OpenCurlyBracket",CloseCurlyBracket:"CloseCurlyBracket",Comma:"Comma",Dot:"Dot",Colon:"Colon",Pipe:"Pipe",CallOperator:"CallOperator",AdditiveBinaryOperator:"AdditiveBinaryOperator",MultiplicativeBinaryOperator:"MultiplicativeBinaryOperator",ComparisonBinaryOperator:"ComparisonBinaryOperator",UnaryOperator:"UnaryOperator",Set:"Set",If:"If",For:"For",In:"In",Is:"Is",NotIn:"NotIn",Else:"Else",EndIf:"EndIf",ElseIf:"ElseIf",EndFor:"EndFor",And:"And",Or:"Or",Not:"UnaryOperator",Macro:"Macro",EndMacro:"EndMacro"}),D=Object.freeze({set:_.Set,for:_.For,in:_.In,is:_.Is,if:_.If,else:_.Else,endif:_.EndIf,elif:_.ElseIf,endfor:_.EndFor,and:_.And,or:_.Or,not:_.Not,"not in":_.NotIn,macro:_.Macro,endmacro:_.EndMacro,true:_.BooleanLiteral,false:_.BooleanLiteral,none:_.NullLiteral,True:_.BooleanLiteral,False:_.BooleanLiteral,None:_.NullLiteral}),j=class{constructor(L,oe){this.value=L,this.type=oe}};function X(L){return/\w/.test(L)}function R(L){return/[0-9]/.test(L)}var g=[["{%",_.OpenStatement],["%}",_.CloseStatement],["{{",_.OpenExpression],["}}",_.CloseExpression],["(",_.OpenParen],[")",_.CloseParen],["{",_.OpenCurlyBracket],["}",_.CloseCurlyBracket],["[",_.OpenSquareBracket],["]",_.CloseSquareBracket],[",",_.Comma],[".",_.Dot],[":",_.Colon],["|",_.Pipe],["<=",_.ComparisonBinaryOperator],[">=",_.ComparisonBinaryOperator],["==",_.ComparisonBinaryOperator],["!=",_.ComparisonBinaryOperator],["<",_.ComparisonBinaryOperator],[">",_.ComparisonBinaryOperator],["+",_.AdditiveBinaryOperator],["-",_.AdditiveBinaryOperator],["*",_.MultiplicativeBinaryOperator],["/",_.MultiplicativeBinaryOperator],["%",_.MultiplicativeBinaryOperator],["=",_.Equals]],v=new Map([["n",` +`],["t"," "],["r","\r"],["b","\b"],["f","\f"],["v","\v"],["'","'"],['"','"'],["\\","\\"]]);function M(L,oe={}){return L.endsWith(` +`)&&(L=L.slice(0,-1)),L=L.replace(/{#.*?#}/gs,"{##}"),oe.lstrip_blocks&&(L=L.replace(/^[ \t]*({[#%])/gm,"$1")),oe.trim_blocks&&(L=L.replace(/([#%]})\n/g,"$1")),L.replace(/{##}/g,"").replace(/-%}\s*/g,"%}").replace(/\s*{%-/g,"{%").replace(/-}}\s*/g,"}}").replace(/\s*{{-/g,"{{")}function y(L,oe={}){var Je,ut,mt;const H=[],fe=M(L,oe);let $e=0;const We=vt=>{let kt="";for(;vt(fe[$e]);){if(fe[$e]==="\\"){if(++$e,$e>=fe.length)throw new SyntaxError("Unexpected end of input");const At=fe[$e++],is=v.get(At);if(is===void 0)throw new SyntaxError(`Unexpected escaped character: ${At}`);kt+=is;continue}if(kt+=fe[$e++],$e>=fe.length)throw new SyntaxError("Unexpected end of input")}return kt};e:for(;$e0){H.push(new j(At,_.Text));continue}}We(At=>/\s/.test(At));const kt=fe[$e];if(kt==="-"||kt==="+"){const At=(ut=H.at(-1))==null?void 0:ut.type;if(At===_.Text||At===void 0)throw new SyntaxError(`Unexpected character: ${kt}`);switch(At){case _.Identifier:case _.NumericLiteral:case _.BooleanLiteral:case _.NullLiteral:case _.StringLiteral:case _.CloseParen:case _.CloseSquareBracket:break;default:{++$e;const is=We(R);H.push(new j(`${kt}${is}`,is.length>0?_.NumericLiteral:_.UnaryOperator));continue}}}for(const[At,is]of g)if(fe.slice($e,$e+At.length)===At){H.push(new j(At,is)),$e+=At.length;continue e}if(kt==="'"||kt==='"'){++$e;const At=We(is=>is!==kt);H.push(new j(At,_.StringLiteral)),++$e;continue}if(R(kt)){const At=We(R);H.push(new j(At,_.NumericLiteral));continue}if(X(kt)){const At=We(X),is=Object.hasOwn(D,At)?D[At]:_.Identifier;is===_.In&&((mt=H.at(-1))==null?void 0:mt.type)===_.Not?(H.pop(),H.push(new j("not in",_.NotIn))):H.push(new j(At,is));continue}throw new SyntaxError(`Unexpected character: ${kt}`)}return H}var b=class{constructor(){me(this,"type","Statement")}},I=class extends b{constructor(oe){super();me(this,"type","Program");this.body=oe}},K=class extends b{constructor(oe,H,fe){super();me(this,"type","If");this.test=oe,this.body=H,this.alternate=fe}},se=class extends b{constructor(oe,H,fe,$e){super();me(this,"type","For");this.loopvar=oe,this.iterable=H,this.body=fe,this.defaultBlock=$e}},ne=class extends b{constructor(oe,H){super();me(this,"type","Set");this.assignee=oe,this.value=H}},W=class extends b{constructor(oe,H,fe){super();me(this,"type","Macro");this.name=oe,this.args=H,this.body=fe}},U=class extends b{constructor(){super(...arguments);me(this,"type","Expression")}},q=class extends U{constructor(oe,H,fe){super();me(this,"type","MemberExpression");this.object=oe,this.property=H,this.computed=fe}},$=class extends U{constructor(oe,H){super();me(this,"type","CallExpression");this.callee=oe,this.args=H}},S=class extends U{constructor(oe){super();me(this,"type","Identifier");this.value=oe}},w=class extends U{constructor(oe){super();me(this,"type","Literal");this.value=oe}},x=class extends w{constructor(){super(...arguments);me(this,"type","NumericLiteral")}},O=class extends w{constructor(){super(...arguments);me(this,"type","StringLiteral")}},ae=class extends w{constructor(){super(...arguments);me(this,"type","BooleanLiteral")}},ie=class extends w{constructor(){super(...arguments);me(this,"type","NullLiteral")}},ve=class extends w{constructor(){super(...arguments);me(this,"type","ArrayLiteral")}},we=class extends w{constructor(){super(...arguments);me(this,"type","TupleLiteral")}},re=class extends w{constructor(){super(...arguments);me(this,"type","ObjectLiteral")}},xe=class extends U{constructor(oe,H,fe){super();me(this,"type","BinaryExpression");this.operator=oe,this.left=H,this.right=fe}},ce=class extends U{constructor(oe,H){super();me(this,"type","FilterExpression");this.operand=oe,this.filter=H}},ke=class extends U{constructor(oe,H){super();me(this,"type","SelectExpression");this.iterable=oe,this.test=H}},Ie=class extends U{constructor(oe,H,fe){super();me(this,"type","TestExpression");this.operand=oe,this.negate=H,this.test=fe}},Ee=class extends U{constructor(oe,H){super();me(this,"type","UnaryExpression");this.operator=oe,this.argument=H}},tt=class extends U{constructor(oe=void 0,H=void 0,fe=void 0){super();me(this,"type","SliceExpression");this.start=oe,this.stop=H,this.step=fe}},Ge=class extends U{constructor(oe,H){super();me(this,"type","KeywordArgumentExpression");this.key=oe,this.value=H}};function ye(L){const oe=new I([]);let H=0;function fe(it,Tt){const Dt=L[H++];if(!Dt||Dt.type!==it)throw new Error(`Parser Error: ${Tt}. ${Dt.type} !== ${it}.`);return Dt}function $e(){switch(L[H].type){case _.Text:return ut();case _.OpenStatement:return mt();case _.OpenExpression:return vt();default:throw new SyntaxError(`Unexpected token type: ${L[H].type}`)}}function We(...it){return H+it.length<=L.length&&it.some((Tt,Dt)=>Tt!==L[H+Dt].type)}function Je(...it){return H+it.length<=L.length&&it.every((Tt,Dt)=>Tt===L[H+Dt].type)}function ut(){return new O(fe(_.Text,"Expected text token").value)}function mt(){fe(_.OpenStatement,"Expected opening statement token");let it;switch(L[H].type){case _.Set:++H,it=kt(),fe(_.CloseStatement,"Expected closing statement token");break;case _.If:++H,it=At(),fe(_.OpenStatement,"Expected {% token"),fe(_.EndIf,"Expected endif token"),fe(_.CloseStatement,"Expected %} token");break;case _.Macro:++H,it=is(),fe(_.OpenStatement,"Expected {% token"),fe(_.EndMacro,"Expected endmacro token"),fe(_.CloseStatement,"Expected %} token");break;case _.For:++H,it=Cs(),fe(_.OpenStatement,"Expected {% token"),fe(_.EndFor,"Expected endfor token"),fe(_.CloseStatement,"Expected %} token");break;default:throw new SyntaxError(`Unknown statement type: ${L[H].type}`)}return it}function vt(){fe(_.OpenExpression,"Expected opening expression token");const it=Ds();return fe(_.CloseExpression,"Expected closing expression token"),it}function kt(){const it=Ds();if(Je(_.Equals)){++H;const Tt=kt();return new ne(it,Tt)}return it}function At(){var Vs,jr,Or,Ms,lr,Os,Er,es;const it=Ds();fe(_.CloseStatement,"Expected closing statement token");const Tt=[],Dt=[];for(;!(((Vs=L[H])==null?void 0:Vs.type)===_.OpenStatement&&(((jr=L[H+1])==null?void 0:jr.type)===_.ElseIf||((Or=L[H+1])==null?void 0:Or.type)===_.Else||((Ms=L[H+1])==null?void 0:Ms.type)===_.EndIf));)Tt.push($e());if(((lr=L[H])==null?void 0:lr.type)===_.OpenStatement&&((Os=L[H+1])==null?void 0:Os.type)!==_.EndIf)if(++H,Je(_.ElseIf))fe(_.ElseIf,"Expected elseif token"),Dt.push(At());else for(fe(_.Else,"Expected else token"),fe(_.CloseStatement,"Expected closing statement token");!(((Er=L[H])==null?void 0:Er.type)===_.OpenStatement&&((es=L[H+1])==null?void 0:es.type)===_.EndIf);)Dt.push($e());return new K(it,Tt,Dt)}function is(){const it=ar();if(it.type!=="Identifier")throw new SyntaxError("Expected identifier following macro statement");const Tt=Ar();fe(_.CloseStatement,"Expected closing statement token");const Dt=[];for(;We(_.OpenStatement,_.EndMacro);)Dt.push($e());return new W(it,Tt,Dt)}function ys(it=!1){const Tt=it?ar:Ds,Dt=[Tt()],Vs=Je(_.Comma);for(;Vs&&(++H,Dt.push(Tt()),!!Je(_.Comma)););return Vs?new we(Dt):Dt[0]}function Cs(){const it=ys(!0);if(!(it instanceof S||it instanceof we))throw new SyntaxError(`Expected identifier/tuple for the loop variable, got ${it.type} instead`);fe(_.In,"Expected `in` keyword following loop variable");const Tt=Ds();fe(_.CloseStatement,"Expected closing statement token");const Dt=[];for(;We(_.OpenStatement,_.EndFor)&&We(_.OpenStatement,_.Else);)Dt.push($e());const Vs=[];if(Je(_.OpenStatement,_.Else))for(++H,++H,fe(_.CloseStatement,"Expected closing statement token");We(_.OpenStatement,_.EndFor);)Vs.push($e());return new se(it,Tt,Dt,Vs)}function Ds(){return sr()}function sr(){const it=Sr();if(Je(_.If)){++H;const Tt=Sr();if(Je(_.Else)){++H;const Dt=Sr();return new K(Tt,[it],[Dt])}else return new ke(it,Tt)}return it}function Sr(){let it=Yr();for(;Je(_.Or);){const Tt=L[H];++H;const Dt=Yr();it=new xe(Tt,it,Dt)}return it}function Yr(){let it=Us();for(;Je(_.And);){const Tt=L[H];++H;const Dt=Us();it=new xe(Tt,it,Dt)}return it}function Us(){let it;for(;Je(_.Not);){const Tt=L[H];++H;const Dt=Us();it=new Ee(Tt,Dt)}return it??Pr()}function Pr(){let it=Nt();for(;Je(_.ComparisonBinaryOperator)||Je(_.In)||Je(_.NotIn);){const Tt=L[H];++H;const Dt=Nt();it=new xe(Tt,it,Dt)}return it}function Nt(){let it=Ir();for(;Je(_.AdditiveBinaryOperator);){const Tt=L[H];++H;const Dt=Ir();it=new xe(Tt,it,Dt)}return it}function Jr(){const it=en();return Je(_.OpenParen)?$r(it):it}function $r(it){let Tt=new $(it,Ar());return Je(_.OpenParen)&&(Tt=$r(Tt)),Tt}function Ar(){fe(_.OpenParen,"Expected opening parenthesis for arguments list");const it=Zr();return fe(_.CloseParen,"Expected closing parenthesis for arguments list"),it}function Zr(){const it=[];for(;!Je(_.CloseParen);){let Tt=Ds();if(Je(_.Equals)){if(++H,!(Tt instanceof S))throw new SyntaxError("Expected identifier for keyword argument");const Dt=Ds();Tt=new Ge(Tt,Dt)}it.push(Tt),Je(_.Comma)&&++H}return it}function cr(){const it=[];let Tt=!1;for(;!Je(_.CloseSquareBracket);)Je(_.Colon)?(it.push(void 0),++H,Tt=!0):(it.push(Ds()),Je(_.Colon)&&(++H,Tt=!0));if(it.length===0)throw new SyntaxError("Expected at least one argument for member/slice expression");if(Tt){if(it.length>3)throw new SyntaxError("Expected 0-3 arguments for slice expression");return new tt(...it)}return it[0]}function en(){let it=ar();for(;Je(_.Dot)||Je(_.OpenSquareBracket);){const Tt=L[H];++H;let Dt;const Vs=Tt.type!==_.Dot;if(Vs)Dt=cr(),fe(_.CloseSquareBracket,"Expected closing square bracket");else if(Dt=ar(),Dt.type!=="Identifier")throw new SyntaxError("Expected identifier following dot operator");it=new q(it,Dt,Vs)}return it}function Ir(){let it=Rr();for(;Je(_.MultiplicativeBinaryOperator);){const Tt=L[H];++H;const Dt=Rr();it=new xe(Tt,it,Dt)}return it}function Rr(){let it=Nr();for(;Je(_.Is);){++H;const Tt=Je(_.Not);Tt&&++H;let Dt=ar();if(Dt instanceof ae?Dt=new S(Dt.value.toString()):Dt instanceof ie&&(Dt=new S("none")),!(Dt instanceof S))throw new SyntaxError("Expected identifier for the test");it=new Ie(it,Tt,Dt)}return it}function Nr(){let it=Jr();for(;Je(_.Pipe);){++H;let Tt=ar();if(!(Tt instanceof S))throw new SyntaxError("Expected identifier for the filter");Je(_.OpenParen)&&(Tt=$r(Tt)),it=new ce(it,Tt)}return it}function ar(){const it=L[H];switch(it.type){case _.NumericLiteral:return++H,new x(Number(it.value));case _.StringLiteral:return++H,new O(it.value);case _.BooleanLiteral:return++H,new ae(it.value.toLowerCase()==="true");case _.NullLiteral:return++H,new ie(null);case _.Identifier:return++H,new S(it.value);case _.OpenParen:{++H;const Tt=ys();if(L[H].type!==_.CloseParen)throw new SyntaxError(`Expected closing parenthesis, got ${L[H].type} instead`);return++H,Tt}case _.OpenSquareBracket:{++H;const Tt=[];for(;!Je(_.CloseSquareBracket);)Tt.push(Ds()),Je(_.Comma)&&++H;return++H,new ve(Tt)}case _.OpenCurlyBracket:{++H;const Tt=new Map;for(;!Je(_.CloseCurlyBracket);){const Dt=Ds();fe(_.Colon,"Expected colon between key and value in object literal");const Vs=Ds();Tt.set(Dt,Vs),Je(_.Comma)&&++H}return++H,new re(Tt)}default:throw new SyntaxError(`Unexpected token: ${it.type}`)}}for(;H=0?(oe=(oe??(oe=0))<0?Math.max(L.length+oe,0):Math.min(oe,L.length),H=(H??(H=L.length))<0?Math.max(L.length+H,0):Math.min(H,L.length)):(oe=(oe??(oe=L.length-1))<0?Math.max(L.length+oe,-1):Math.min(oe,L.length-1),H=(H??(H=-1))<-1?Math.max(L.length+H,-1):Math.min(H,L.length-1));const We=[];for(let Je=oe;$e*Je<$e*H;Je+=fe)We.push(L[Je]);return We}function Ce(L){return L.replace(/\b\w/g,oe=>oe.toUpperCase())}var Be=class{constructor(L=void 0){me(this,"type","RuntimeValue");me(this,"value");me(this,"builtins",new Map);this.value=L}__bool__(){return new Ke(!!this.value)}},Ze=class extends Be{constructor(){super(...arguments);me(this,"type","NumericValue")}},te=class extends Be{constructor(){super(...arguments);me(this,"type","StringValue");me(this,"builtins",new Map([["upper",new Ve(()=>new te(this.value.toUpperCase()))],["lower",new Ve(()=>new te(this.value.toLowerCase()))],["strip",new Ve(()=>new te(this.value.trim()))],["title",new Ve(()=>new te(Ce(this.value)))],["length",new Ze(this.value.length)],["rstrip",new Ve(()=>new te(this.value.trimEnd()))],["lstrip",new Ve(()=>new te(this.value.trimStart()))]]))}},Ke=class extends Be{constructor(){super(...arguments);me(this,"type","BooleanValue")}},je=class extends Be{constructor(){super(...arguments);me(this,"type","ObjectValue");me(this,"builtins",new Map([["get",new Ve(([oe,H])=>{if(!(oe instanceof te))throw new Error(`Object key must be a string: got ${oe.type}`);return this.value.get(oe.value)??H??new Ne})],["items",new Ve(()=>new Te(Array.from(this.value.entries()).map(([oe,H])=>new Te([new te(oe),H]))))]]))}__bool__(){return new Ke(this.value.size>0)}},le=class extends je{constructor(){super(...arguments);me(this,"type","KeywordArgumentsValue")}},Te=class extends Be{constructor(){super(...arguments);me(this,"type","ArrayValue");me(this,"builtins",new Map([["length",new Ze(this.value.length)]]))}__bool__(){return new Ke(this.value.length>0)}},Ue=class extends Te{constructor(){super(...arguments);me(this,"type","TupleValue")}},Ve=class extends Be{constructor(){super(...arguments);me(this,"type","FunctionValue")}},Ne=class extends Be{constructor(){super(...arguments);me(this,"type","NullValue")}},Re=class extends Be{constructor(){super(...arguments);me(this,"type","UndefinedValue")}},st=class{constructor(L){me(this,"variables",new Map([["namespace",new Ve(L=>{if(L.length===0)return new je(new Map);if(L.length!==1||!(L[0]instanceof je))throw new Error("`namespace` expects either zero arguments or a single object argument");return L[0]})]]));me(this,"tests",new Map([["boolean",L=>L.type==="BooleanValue"],["callable",L=>L instanceof Ve],["odd",L=>{if(L.type!=="NumericValue")throw new Error(`Cannot apply test "odd" to type: ${L.type}`);return L.value%2!==0}],["even",L=>{if(L.type!=="NumericValue")throw new Error(`Cannot apply test "even" to type: ${L.type}`);return L.value%2===0}],["false",L=>L.type==="BooleanValue"&&!L.value],["true",L=>L.type==="BooleanValue"&&L.value],["none",L=>L.type==="NullValue"],["string",L=>L.type==="StringValue"],["number",L=>L.type==="NumericValue"],["integer",L=>L.type==="NumericValue"&&Number.isInteger(L.value)],["iterable",L=>L.type==="ArrayValue"||L.type==="StringValue"],["mapping",L=>L.type==="ObjectValue"],["lower",L=>{const oe=L.value;return L.type==="StringValue"&&oe===oe.toLowerCase()}],["upper",L=>{const oe=L.value;return L.type==="StringValue"&&oe===oe.toUpperCase()}],["none",L=>L.type==="NullValue"],["defined",L=>L.type!=="UndefinedValue"],["undefined",L=>L.type==="UndefinedValue"],["equalto",(L,oe)=>L.value===oe.value],["eq",(L,oe)=>L.value===oe.value]]));this.parent=L}set(L,oe){return this.declareVariable(L,ct(oe))}declareVariable(L,oe){if(this.variables.has(L))throw new SyntaxError(`Variable already declared: ${L}`);return this.variables.set(L,oe),oe}setVariable(L,oe){return this.variables.set(L,oe),oe}resolve(L){if(this.variables.has(L))return this;if(this.parent)return this.parent.resolve(L);throw new Error(`Unknown variable: ${L}`)}lookupVariable(L){try{return this.resolve(L).variables.get(L)??new Re}catch{return new Re}}},dt=class{constructor(L){me(this,"global");this.global=L??new st}run(L){return this.evaluate(L,this.global)}evaluateBinaryExpression(L,oe){const H=this.evaluate(L.left,oe);switch(L.operator.value){case"and":return H.__bool__().value?this.evaluate(L.right,oe):H;case"or":return H.__bool__().value?H:this.evaluate(L.right,oe)}const fe=this.evaluate(L.right,oe);switch(L.operator.value){case"==":return new Ke(H.value==fe.value);case"!=":return new Ke(H.value!=fe.value)}if(H instanceof Re||fe instanceof Re)throw new Error("Cannot perform operation on undefined values");if(H instanceof Ne||fe instanceof Ne)throw new Error("Cannot perform operation on null values");if(H instanceof Ze&&fe instanceof Ze)switch(L.operator.value){case"+":return new Ze(H.value+fe.value);case"-":return new Ze(H.value-fe.value);case"*":return new Ze(H.value*fe.value);case"/":return new Ze(H.value/fe.value);case"%":return new Ze(H.value%fe.value);case"<":return new Ke(H.value":return new Ke(H.value>fe.value);case">=":return new Ke(H.value>=fe.value);case"<=":return new Ke(H.value<=fe.value)}else if(H instanceof Te&&fe instanceof Te)switch(L.operator.value){case"+":return new Te(H.value.concat(fe.value))}else if(fe instanceof Te){const $e=fe.value.find(We=>We.value===H.value)!==void 0;switch(L.operator.value){case"in":return new Ke($e);case"not in":return new Ke(!$e)}}if(H instanceof te||fe instanceof te)switch(L.operator.value){case"+":return new te(H.value.toString()+fe.value.toString())}if(H instanceof te&&fe instanceof te)switch(L.operator.value){case"in":return new Ke(fe.value.includes(H.value));case"not in":return new Ke(!fe.value.includes(H.value))}if(H instanceof te&&fe instanceof je)switch(L.operator.value){case"in":return new Ke(fe.value.has(H.value));case"not in":return new Ke(!fe.value.has(H.value))}throw new SyntaxError(`Unknown operator "${L.operator.value}" between ${H.type} and ${fe.type}`)}evaluateArguments(L,oe){const H=[],fe=new Map;for(const $e of L)if($e.type==="KeywordArgumentExpression"){const We=$e;fe.set(We.key.value,this.evaluate(We.value,oe))}else{if(fe.size>0)throw new Error("Positional arguments must come before keyword arguments");H.push(this.evaluate($e,oe))}return[H,fe]}evaluateFilterExpression(L,oe){const H=this.evaluate(L.operand,oe);if(L.filter.type==="Identifier"){const fe=L.filter;if(fe.value==="tojson")return new te(lt(H));if(H instanceof Te)switch(fe.value){case"list":return H;case"first":return H.value[0];case"last":return H.value[H.value.length-1];case"length":return new Ze(H.value.length);case"reverse":return new Te(H.value.reverse());case"sort":return new Te(H.value.sort(($e,We)=>{if($e.type!==We.type)throw new Error(`Cannot compare different types: ${$e.type} and ${We.type}`);switch($e.type){case"NumericValue":return $e.value-We.value;case"StringValue":return $e.value.localeCompare(We.value);default:throw new Error(`Cannot compare type: ${$e.type}`)}}));default:throw new Error(`Unknown ArrayValue filter: ${fe.value}`)}else if(H instanceof te)switch(fe.value){case"length":return new Ze(H.value.length);case"upper":return new te(H.value.toUpperCase());case"lower":return new te(H.value.toLowerCase());case"title":return new te(Ce(H.value));case"capitalize":return new te(H.value.charAt(0).toUpperCase()+H.value.slice(1));case"trim":return new te(H.value.trim());case"indent":return new te(H.value.split(` +`).map(($e,We)=>We===0||$e.length===0?$e:" "+$e).join(` +`));case"string":return H;default:throw new Error(`Unknown StringValue filter: ${fe.value}`)}else if(H instanceof Ze)switch(fe.value){case"abs":return new Ze(Math.abs(H.value));default:throw new Error(`Unknown NumericValue filter: ${fe.value}`)}else if(H instanceof je)switch(fe.value){case"items":return new Te(Array.from(H.value.entries()).map(([$e,We])=>new Te([new te($e),We])));case"length":return new Ze(H.value.size);default:throw new Error(`Unknown ObjectValue filter: ${fe.value}`)}throw new Error(`Cannot apply filter "${fe.value}" to type: ${H.type}`)}else if(L.filter.type==="CallExpression"){const fe=L.filter;if(fe.callee.type!=="Identifier")throw new Error(`Unknown filter: ${fe.callee.type}`);const $e=fe.callee.value;if($e==="tojson"){const[,We]=this.evaluateArguments(fe.args,oe),Je=We.get("indent")??new Ne;if(!(Je instanceof Ze||Je instanceof Ne))throw new Error("If set, indent must be a number");return new te(lt(H,Je.value))}if(H instanceof Te){switch($e){case"selectattr":case"rejectattr":{const We=$e==="selectattr";if(H.value.some(At=>!(At instanceof je)))throw new Error(`\`${$e}\` can only be applied to array of objects`);if(fe.args.some(At=>At.type!=="StringLiteral"))throw new Error(`arguments of \`${$e}\` must be strings`);const[Je,ut,mt]=fe.args.map(At=>this.evaluate(At,oe));let vt;if(ut){const At=oe.tests.get(ut.value);if(!At)throw new Error(`Unknown test: ${ut.value}`);vt=At}else vt=(...At)=>At[0].__bool__().value;const kt=H.value.filter(At=>{const is=At.value.get(Je.value),ys=is?vt(is,mt):!1;return We?ys:!ys});return new Te(kt)}case"map":{const[,We]=this.evaluateArguments(fe.args,oe);if(We.has("attribute")){const Je=We.get("attribute");if(!(Je instanceof te))throw new Error("attribute must be a string");const ut=We.get("default"),mt=H.value.map(vt=>{if(!(vt instanceof je))throw new Error("items in map must be an object");return vt.value.get(Je.value)??ut??new Re});return new Te(mt)}else throw new Error("`map` expressions without `attribute` set are not currently supported.")}}throw new Error(`Unknown ArrayValue filter: ${$e}`)}else if(H instanceof te){switch($e){case"indent":{const[We,Je]=this.evaluateArguments(fe.args,oe),ut=We.at(0)??Je.get("width")??new Ze(4);if(!(ut instanceof Ze))throw new Error("width must be a number");const mt=We.at(1)??Je.get("first")??new Ke(!1),vt=We.at(2)??Je.get("blank")??new Ke(!1),kt=H.value.split(` +`),At=" ".repeat(ut.value),is=kt.map((ys,Cs)=>!mt.value&&Cs===0||!vt.value&&ys.length===0?ys:At+ys);return new te(is.join(` +`))}}throw new Error(`Unknown StringValue filter: ${$e}`)}else throw new Error(`Cannot apply filter "${$e}" to type: ${H.type}`)}throw new Error(`Unknown filter: ${L.filter.type}`)}evaluateTestExpression(L,oe){const H=this.evaluate(L.operand,oe),fe=oe.tests.get(L.test.value);if(!fe)throw new Error(`Unknown test: ${L.test.value}`);const $e=fe(H);return new Ke(L.negate?!$e:$e)}evaluateUnaryExpression(L,oe){const H=this.evaluate(L.argument,oe);switch(L.operator.value){case"not":return new Ke(!H.value);default:throw new SyntaxError(`Unknown operator: ${L.operator.value}`)}}evalProgram(L,oe){return this.evaluateBlock(L.body,oe)}evaluateBlock(L,oe){let H="";for(const fe of L){const $e=this.evaluate(fe,oe);$e.type!=="NullValue"&&$e.type!=="UndefinedValue"&&(H+=$e.value)}return new te(H)}evaluateIdentifier(L,oe){return oe.lookupVariable(L.value)}evaluateCallExpression(L,oe){const[H,fe]=this.evaluateArguments(L.args,oe);fe.size>0&&H.push(new le(fe));const $e=this.evaluate(L.callee,oe);if($e.type!=="FunctionValue")throw new Error(`Cannot call something that is not a function: got ${$e.type}`);return $e.value(H,oe)}evaluateSliceExpression(L,oe,H){if(!(L instanceof Te||L instanceof te))throw new Error("Slice object must be an array or string");const fe=this.evaluate(oe.start,H),$e=this.evaluate(oe.stop,H),We=this.evaluate(oe.step,H);if(!(fe instanceof Ze||fe instanceof Re))throw new Error("Slice start must be numeric or undefined");if(!($e instanceof Ze||$e instanceof Re))throw new Error("Slice stop must be numeric or undefined");if(!(We instanceof Ze||We instanceof Re))throw new Error("Slice step must be numeric or undefined");return L instanceof Te?new Te(de(L.value,fe.value,$e.value,We.value)):new te(de(Array.from(L.value),fe.value,$e.value,We.value).join(""))}evaluateMemberExpression(L,oe){const H=this.evaluate(L.object,oe);let fe;if(L.computed){if(L.property.type==="SliceExpression")return this.evaluateSliceExpression(H,L.property,oe);fe=this.evaluate(L.property,oe)}else fe=new te(L.property.value);let $e;if(H instanceof je){if(!(fe instanceof te))throw new Error(`Cannot access property with non-string: got ${fe.type}`);$e=H.value.get(fe.value)??H.builtins.get(fe.value)}else if(H instanceof Te||H instanceof te)if(fe instanceof Ze)$e=H.value.at(fe.value),H instanceof te&&($e=new te(H.value.at(fe.value)));else if(fe instanceof te)$e=H.builtins.get(fe.value);else throw new Error(`Cannot access property with non-string/non-number: got ${fe.type}`);else{if(!(fe instanceof te))throw new Error(`Cannot access property with non-string: got ${fe.type}`);$e=H.builtins.get(fe.value)}return $e instanceof Be?$e:new Re}evaluateSet(L,oe){const H=this.evaluate(L.value,oe);if(L.assignee.type==="Identifier"){const fe=L.assignee.value;oe.setVariable(fe,H)}else if(L.assignee.type==="MemberExpression"){const fe=L.assignee,$e=this.evaluate(fe.object,oe);if(!($e instanceof je))throw new Error("Cannot assign to member of non-object");if(fe.property.type!=="Identifier")throw new Error("Cannot assign to member with non-identifier property");$e.value.set(fe.property.value,H)}else throw new Error(`Invalid LHS inside assignment expression: ${JSON.stringify(L.assignee)}`);return new Ne}evaluateIf(L,oe){const H=this.evaluate(L.test,oe);return this.evaluateBlock(H.__bool__().value?L.body:L.alternate,oe)}evaluateFor(L,oe){const H=new st(oe);let fe,$e;if(L.iterable.type==="SelectExpression"){const vt=L.iterable;$e=this.evaluate(vt.iterable,H),fe=vt.test}else $e=this.evaluate(L.iterable,H);if(!($e instanceof Te))throw new Error(`Expected iterable type in for loop: got ${$e.type}`);const We=[],Je=[];for(let vt=0;vt<$e.value.length;++vt){const kt=new st(H),At=$e.value[vt];let is;if(L.loopvar.type==="Identifier")is=ys=>ys.setVariable(L.loopvar.value,At);else if(L.loopvar.type==="TupleLiteral"){const ys=L.loopvar;if(At.type!=="ArrayValue")throw new Error(`Cannot unpack non-iterable type: ${At.type}`);const Cs=At;if(ys.value.length!==Cs.value.length)throw new Error(`Too ${ys.value.length>Cs.value.length?"few":"many"} items to unpack`);is=Ds=>{for(let sr=0;sr0?We[vt-1]:new Re],["nextitem",vt{var Je;const $e=new st(fe);H=H.slice();let We;((Je=H.at(-1))==null?void 0:Je.type)==="KeywordArgumentsValue"&&(We=H.pop());for(let ut=0;utthis.evaluate(H,oe)));case"TupleLiteral":return new Ue(L.value.map(H=>this.evaluate(H,oe)));case"ObjectLiteral":{const H=new Map;for(const[fe,$e]of L.value){const We=this.evaluate(fe,oe);if(!(We instanceof te))throw new Error(`Object keys must be strings: got ${We.type}`);H.set(We.value,this.evaluate($e,oe))}return new je(H)}case"Identifier":return this.evaluateIdentifier(L,oe);case"CallExpression":return this.evaluateCallExpression(L,oe);case"MemberExpression":return this.evaluateMemberExpression(L,oe);case"UnaryExpression":return this.evaluateUnaryExpression(L,oe);case"BinaryExpression":return this.evaluateBinaryExpression(L,oe);case"FilterExpression":return this.evaluateFilterExpression(L,oe);case"TestExpression":return this.evaluateTestExpression(L,oe);default:throw new SyntaxError(`Unknown node type: ${L.type}`)}}};function ct(L){switch(typeof L){case"number":return new Ze(L);case"string":return new te(L);case"boolean":return new Ke(L);case"undefined":return new Re;case"object":return L===null?new Ne:Array.isArray(L)?new Te(L.map(ct)):new je(new Map(Object.entries(L).map(([oe,H])=>[oe,ct(H)])));case"function":return new Ve((oe,H)=>{const fe=L(...oe.map($e=>$e.value))??null;return ct(fe)});default:throw new Error(`Cannot convert to runtime value: ${L}`)}}function lt(L,oe,H){const fe=H??0;switch(L.type){case"NullValue":case"UndefinedValue":return"null";case"NumericValue":case"StringValue":case"BooleanValue":return JSON.stringify(L.value);case"ArrayValue":case"ObjectValue":{const $e=oe?" ".repeat(oe):"",We=` +`+$e.repeat(fe),Je=We+$e;if(L.type==="ArrayValue"){const ut=L.value.map(mt=>lt(mt,oe,fe+1));return oe?`[${Je}${ut.join(`,${Je}`)}${We}]`:`[${ut.join(", ")}]`}else{const ut=Array.from(L.value.entries()).map(([mt,vt])=>{const kt=`"${mt}": ${lt(vt,oe,fe+1)}`;return oe?`${Je}${kt}`:kt});return oe?`{${ut.join(",")}${We}}`:`{${ut.join(", ")}}`}}default:throw new Error(`Cannot convert to JSON: ${L.type}`)}}var ht=class{constructor(L){me(this,"parsed");const oe=y(L,{lstrip_blocks:!0,trim_blocks:!0});this.parsed=ye(oe)}render(L){const oe=new st;oe.set("false",!1),oe.set("true",!0),oe.set("raise_exception",$e=>{throw new Error($e)}),oe.set("range",J);for(const[$e,We]of Object.entries(L))oe.set($e,We);return new dt(oe).run(this.parsed).value}}},"./node_modules/onnxruntime-common/dist/esm/backend-impl.js":(Le,A,r)=>{r.r(A),r.d(A,{registerBackend:()=>j,resolveBackendAndExecutionProviders:()=>R});const _=new Map,D=[],j=(g,v,M)=>{if(v&&typeof v.init=="function"&&typeof v.createInferenceSessionHandler=="function"){const y=_.get(g);if(y===void 0)_.set(g,{backend:v,priority:M});else{if(y.priority>M)return;if(y.priority===M&&y.backend!==v)throw new Error(`cannot register backend "${g}" using priority ${M}`)}if(M>=0){const b=D.indexOf(g);b!==-1&&D.splice(b,1);for(let I=0;I{const v=_.get(g);if(!v)return"backend not found.";if(v.initialized)return v.backend;if(v.aborted)return v.error;{const M=!!v.initPromise;try{return M||(v.initPromise=v.backend.init(g)),await v.initPromise,v.initialized=!0,v.backend}catch(y){return M||(v.error=`${y}`,v.aborted=!0),v.error}finally{delete v.initPromise}}},R=async g=>{const v=g.executionProviders||[],M=v.map(ne=>typeof ne=="string"?ne:ne.name),y=M.length===0?D:M;let b;const I=[],K=new Set;for(const ne of y){const W=await X(ne);typeof W=="string"?I.push({name:ne,err:W}):(b||(b=W),b===W&&K.add(ne))}if(!b)throw new Error(`no available backend found. ERR: ${I.map(ne=>`[${ne.name}] ${ne.err}`).join(", ")}`);for(const{name:ne,err:W}of I)M.includes(ne)&&console.warn(`removing requested execution provider "${ne}" from session options because it is not available: ${W}`);const se=v.filter(ne=>K.has(typeof ne=="string"?ne:ne.name));return[b,new Proxy(g,{get:(ne,W)=>W==="executionProviders"?se:Reflect.get(ne,W)})]}},"./node_modules/onnxruntime-common/dist/esm/backend.js":(Le,A,r)=>{r.r(A),r.d(A,{registerBackend:()=>_.registerBackend});var _=r("./node_modules/onnxruntime-common/dist/esm/backend-impl.js")},"./node_modules/onnxruntime-common/dist/esm/env-impl.js":(Le,A,r)=>{r.r(A),r.d(A,{env:()=>j});var _=r("./node_modules/onnxruntime-common/dist/esm/version.js");let D="warning";const j={wasm:{},webgl:{},webgpu:{},versions:{common:_.version},set logLevel(X){if(X!==void 0){if(typeof X!="string"||["verbose","info","warning","error","fatal"].indexOf(X)===-1)throw new Error(`Unsupported logging level: ${X}`);D=X}},get logLevel(){return D}};Object.defineProperty(j,"logLevel",{enumerable:!0})},"./node_modules/onnxruntime-common/dist/esm/env.js":(Le,A,r)=>{r.r(A),r.d(A,{env:()=>D});var _=r("./node_modules/onnxruntime-common/dist/esm/env-impl.js");const D=_.env},"./node_modules/onnxruntime-common/dist/esm/index.js":(Le,A,r)=>{r.r(A),r.d(A,{InferenceSession:()=>j.InferenceSession,TRACE:()=>R.TRACE,TRACE_FUNC_BEGIN:()=>R.TRACE_FUNC_BEGIN,TRACE_FUNC_END:()=>R.TRACE_FUNC_END,Tensor:()=>X.Tensor,TrainingSession:()=>g.TrainingSession,env:()=>D.env,registerBackend:()=>_.registerBackend});var _=r("./node_modules/onnxruntime-common/dist/esm/backend.js"),D=r("./node_modules/onnxruntime-common/dist/esm/env.js"),j=r("./node_modules/onnxruntime-common/dist/esm/inference-session.js"),X=r("./node_modules/onnxruntime-common/dist/esm/tensor.js");r("./node_modules/onnxruntime-common/dist/esm/tensor-conversion.js"),r("./node_modules/onnxruntime-common/dist/esm/tensor-factory.js");var R=r("./node_modules/onnxruntime-common/dist/esm/trace.js");r("./node_modules/onnxruntime-common/dist/esm/onnx-model.js"),r("./node_modules/onnxruntime-common/dist/esm/onnx-value.js");var g=r("./node_modules/onnxruntime-common/dist/esm/training-session.js")},"./node_modules/onnxruntime-common/dist/esm/inference-session-impl.js":(Le,A,r)=>{r.r(A),r.d(A,{InferenceSession:()=>X});var _=r("./node_modules/onnxruntime-common/dist/esm/backend-impl.js"),D=r("./node_modules/onnxruntime-common/dist/esm/tensor.js"),j=r("./node_modules/onnxruntime-common/dist/esm/trace.js");class X{constructor(g){this.handler=g}async run(g,v,M){(0,j.TRACE_FUNC_BEGIN)();const y={};let b={};if(typeof g!="object"||g===null||g instanceof D.Tensor||Array.isArray(g))throw new TypeError("'feeds' must be an object that use input names as keys and OnnxValue as corresponding values.");let I=!0;if(typeof v=="object"){if(v===null)throw new TypeError("Unexpected argument[1]: cannot be null.");if(v instanceof D.Tensor)throw new TypeError("'fetches' cannot be a Tensor");if(Array.isArray(v)){if(v.length===0)throw new TypeError("'fetches' cannot be an empty array.");I=!1;for(const ne of v){if(typeof ne!="string")throw new TypeError("'fetches' must be a string array or an object.");if(this.outputNames.indexOf(ne)===-1)throw new RangeError(`'fetches' contains invalid output name: ${ne}.`);y[ne]=null}if(typeof M=="object"&&M!==null)b=M;else if(typeof M<"u")throw new TypeError("'options' must be an object.")}else{let ne=!1;const W=Object.getOwnPropertyNames(v);for(const U of this.outputNames)if(W.indexOf(U)!==-1){const q=v[U];(q===null||q instanceof D.Tensor)&&(ne=!0,I=!1,y[U]=q)}if(ne){if(typeof M=="object"&&M!==null)b=M;else if(typeof M<"u")throw new TypeError("'options' must be an object.")}else b=v}}else if(typeof v<"u")throw new TypeError("Unexpected argument[1]: must be 'fetches' or 'options'.");for(const ne of this.inputNames)if(typeof g[ne]>"u")throw new Error(`input '${ne}' is missing in 'feeds'.`);if(I)for(const ne of this.outputNames)y[ne]=null;const K=await this.handler.run(g,y,b),se={};for(const ne in K)if(Object.hasOwnProperty.call(K,ne)){const W=K[ne];W instanceof D.Tensor?se[ne]=W:se[ne]=new D.Tensor(W.type,W.data,W.dims)}return(0,j.TRACE_FUNC_END)(),se}async release(){return this.handler.dispose()}static async create(g,v,M,y){(0,j.TRACE_FUNC_BEGIN)();let b,I={};if(typeof g=="string"){if(b=g,typeof v=="object"&&v!==null)I=v;else if(typeof v<"u")throw new TypeError("'options' must be an object.")}else if(g instanceof Uint8Array){if(b=g,typeof v=="object"&&v!==null)I=v;else if(typeof v<"u")throw new TypeError("'options' must be an object.")}else if(g instanceof ArrayBuffer||typeof SharedArrayBuffer<"u"&&g instanceof SharedArrayBuffer){const W=g;let U=0,q=g.byteLength;if(typeof v=="object"&&v!==null)I=v;else if(typeof v=="number"){if(U=v,!Number.isSafeInteger(U))throw new RangeError("'byteOffset' must be an integer.");if(U<0||U>=W.byteLength)throw new RangeError(`'byteOffset' is out of range [0, ${W.byteLength}).`);if(q=g.byteLength-U,typeof M=="number"){if(q=M,!Number.isSafeInteger(q))throw new RangeError("'byteLength' must be an integer.");if(q<=0||U+q>W.byteLength)throw new RangeError(`'byteLength' is out of range (0, ${W.byteLength-U}].`);if(typeof y=="object"&&y!==null)I=y;else if(typeof y<"u")throw new TypeError("'options' must be an object.")}else if(typeof M<"u")throw new TypeError("'byteLength' must be a number.")}else if(typeof v<"u")throw new TypeError("'options' must be an object.");b=new Uint8Array(W,U,q)}else throw new TypeError("Unexpected argument[0]: must be 'path' or 'buffer'.");const[K,se]=await(0,_.resolveBackendAndExecutionProviders)(I),ne=await K.createInferenceSessionHandler(b,se);return(0,j.TRACE_FUNC_END)(),new X(ne)}startProfiling(){this.handler.startProfiling()}endProfiling(){this.handler.endProfiling()}get inputNames(){return this.handler.inputNames}get outputNames(){return this.handler.outputNames}}},"./node_modules/onnxruntime-common/dist/esm/inference-session.js":(Le,A,r)=>{r.r(A),r.d(A,{InferenceSession:()=>D});var _=r("./node_modules/onnxruntime-common/dist/esm/inference-session-impl.js");const D=_.InferenceSession},"./node_modules/onnxruntime-common/dist/esm/onnx-model.js":(Le,A,r)=>{r.r(A)},"./node_modules/onnxruntime-common/dist/esm/onnx-value.js":(Le,A,r)=>{r.r(A)},"./node_modules/onnxruntime-common/dist/esm/tensor-conversion-impl.js":(Le,A,r)=>{r.r(A),r.d(A,{tensorToDataURL:()=>_,tensorToImageData:()=>D});const _=(j,X)=>{const R=typeof document<"u"?document.createElement("canvas"):new OffscreenCanvas(1,1);R.width=j.dims[3],R.height=j.dims[2];const g=R.getContext("2d");if(g!=null){let v,M;(X==null?void 0:X.tensorLayout)!==void 0&&X.tensorLayout==="NHWC"?(v=j.dims[2],M=j.dims[3]):(v=j.dims[3],M=j.dims[2]);const y=(X==null?void 0:X.format)!==void 0?X.format:"RGB",b=X==null?void 0:X.norm;let I,K;b===void 0||b.mean===void 0?I=[255,255,255,255]:typeof b.mean=="number"?I=[b.mean,b.mean,b.mean,b.mean]:(I=[b.mean[0],b.mean[1],b.mean[2],0],b.mean[3]!==void 0&&(I[3]=b.mean[3])),b===void 0||b.bias===void 0?K=[0,0,0,0]:typeof b.bias=="number"?K=[b.bias,b.bias,b.bias,b.bias]:(K=[b.bias[0],b.bias[1],b.bias[2],0],b.bias[3]!==void 0&&(K[3]=b.bias[3]));const se=M*v;let ne=0,W=se,U=se*2,q=-1;y==="RGBA"?(ne=0,W=se,U=se*2,q=se*3):y==="RGB"?(ne=0,W=se,U=se*2):y==="RBG"&&(ne=0,U=se,W=se*2);for(let $=0;${const R=typeof document<"u"?document.createElement("canvas").getContext("2d"):new OffscreenCanvas(1,1).getContext("2d");let g;if(R!=null){let v,M,y;(X==null?void 0:X.tensorLayout)!==void 0&&X.tensorLayout==="NHWC"?(v=j.dims[2],M=j.dims[1],y=j.dims[3]):(v=j.dims[3],M=j.dims[2],y=j.dims[1]);const b=X!==void 0&&X.format!==void 0?X.format:"RGB",I=X==null?void 0:X.norm;let K,se;I===void 0||I.mean===void 0?K=[255,255,255,255]:typeof I.mean=="number"?K=[I.mean,I.mean,I.mean,I.mean]:(K=[I.mean[0],I.mean[1],I.mean[2],255],I.mean[3]!==void 0&&(K[3]=I.mean[3])),I===void 0||I.bias===void 0?se=[0,0,0,0]:typeof I.bias=="number"?se=[I.bias,I.bias,I.bias,I.bias]:(se=[I.bias[0],I.bias[1],I.bias[2],0],I.bias[3]!==void 0&&(se[3]=I.bias[3]));const ne=M*v;if(X!==void 0&&(X.format!==void 0&&y===4&&X.format!=="RGBA"||y===3&&X.format!=="RGB"&&X.format!=="BGR"))throw new Error("Tensor format doesn't match input tensor dims");const W=4;let U=0,q=1,$=2,S=3,w=0,x=ne,O=ne*2,ae=-1;b==="RGBA"?(w=0,x=ne,O=ne*2,ae=ne*3):b==="RGB"?(w=0,x=ne,O=ne*2):b==="RBG"&&(w=0,O=ne,x=ne*2),g=R.createImageData(v,M);for(let ie=0;ie{r.r(A)},"./node_modules/onnxruntime-common/dist/esm/tensor-factory-impl.js":(Le,A,r)=>{r.r(A),r.d(A,{bufferToTensor:()=>D,tensorFromGpuBuffer:()=>R,tensorFromImage:()=>j,tensorFromMLTensor:()=>g,tensorFromPinnedBuffer:()=>v,tensorFromTexture:()=>X});var _=r("./node_modules/onnxruntime-common/dist/esm/tensor-impl.js");const D=(M,y)=>{if(M===void 0)throw new Error("Image buffer must be defined");if(y.height===void 0||y.width===void 0)throw new Error("Image height and width must be defined");if(y.tensorLayout==="NHWC")throw new Error("NHWC Tensor layout is not supported yet");const{height:b,width:I}=y,K=y.norm??{mean:255,bias:0};let se,ne;typeof K.mean=="number"?se=[K.mean,K.mean,K.mean,K.mean]:se=[K.mean[0],K.mean[1],K.mean[2],K.mean[3]??255],typeof K.bias=="number"?ne=[K.bias,K.bias,K.bias,K.bias]:ne=[K.bias[0],K.bias[1],K.bias[2],K.bias[3]??0];const W=y.format!==void 0?y.format:"RGBA",U=y.tensorFormat!==void 0&&y.tensorFormat!==void 0?y.tensorFormat:"RGB",q=b*I,$=U==="RGBA"?new Float32Array(q*4):new Float32Array(q*3);let S=4,w=0,x=1,O=2,ae=3,ie=0,ve=q,we=q*2,re=-1;W==="RGB"&&(S=3,w=0,x=1,O=2,ae=-1),U==="RGBA"?re=q*3:U==="RBG"?(ie=0,we=q,ve=q*2):U==="BGR"&&(we=0,ve=q,ie=q*2);for(let ce=0;ce{const b=typeof HTMLImageElement<"u"&&M instanceof HTMLImageElement,I=typeof ImageData<"u"&&M instanceof ImageData,K=typeof ImageBitmap<"u"&&M instanceof ImageBitmap,se=typeof M=="string";let ne,W=y??{};const U=()=>{if(typeof document<"u")return document.createElement("canvas");if(typeof OffscreenCanvas<"u")return new OffscreenCanvas(1,1);throw new Error("Canvas is not supported")},q=$=>typeof HTMLCanvasElement<"u"&&$ instanceof HTMLCanvasElement||$ instanceof OffscreenCanvas?$.getContext("2d"):null;if(b){const $=U();$.width=M.width,$.height=M.height;const S=q($);if(S!=null){let w=M.height,x=M.width;if(y!==void 0&&y.resizedHeight!==void 0&&y.resizedWidth!==void 0&&(w=y.resizedHeight,x=y.resizedWidth),y!==void 0){if(W=y,y.tensorFormat!==void 0)throw new Error("Image input config format must be RGBA for HTMLImageElement");W.tensorFormat="RGBA",W.height=w,W.width=x}else W.tensorFormat="RGBA",W.height=w,W.width=x;S.drawImage(M,0,0),ne=S.getImageData(0,0,x,w).data}else throw new Error("Can not access image data")}else if(I){let $,S;if(y!==void 0&&y.resizedWidth!==void 0&&y.resizedHeight!==void 0?($=y.resizedHeight,S=y.resizedWidth):($=M.height,S=M.width),y!==void 0&&(W=y),W.format="RGBA",W.height=$,W.width=S,y!==void 0){const w=U();w.width=S,w.height=$;const x=q(w);if(x!=null)x.putImageData(M,0,0),ne=x.getImageData(0,0,S,$).data;else throw new Error("Can not access image data")}else ne=M.data}else if(K){if(y===void 0)throw new Error("Please provide image config with format for Imagebitmap");const $=U();$.width=M.width,$.height=M.height;const S=q($);if(S!=null){const w=M.height,x=M.width;return S.drawImage(M,0,0,x,w),ne=S.getImageData(0,0,x,w).data,W.height=w,W.width=x,D(ne,W)}else throw new Error("Can not access image data")}else{if(se)return new Promise(($,S)=>{const w=U(),x=q(w);if(!M||!x)return S();const O=new Image;O.crossOrigin="Anonymous",O.src=M,O.onload=()=>{w.width=O.width,w.height=O.height,x.drawImage(O,0,0,w.width,w.height);const ae=x.getImageData(0,0,w.width,w.height);W.height=w.height,W.width=w.width,$(D(ae.data,W))}});throw new Error("Input data provided is not supported - aborted tensor creation")}if(ne!==void 0)return D(ne,W);throw new Error("Input data provided is not supported - aborted tensor creation")},X=(M,y)=>{const{width:b,height:I,download:K,dispose:se}=y,ne=[1,I,b,4];return new _.Tensor({location:"texture",type:"float32",texture:M,dims:ne,download:K,dispose:se})},R=(M,y)=>{const{dataType:b,dims:I,download:K,dispose:se}=y;return new _.Tensor({location:"gpu-buffer",type:b??"float32",gpuBuffer:M,dims:I,download:K,dispose:se})},g=(M,y)=>{const{dataType:b,dims:I,download:K,dispose:se}=y;return new _.Tensor({location:"ml-tensor",type:b??"float32",mlTensor:M,dims:I,download:K,dispose:se})},v=(M,y,b)=>new _.Tensor({location:"cpu-pinned",type:M,data:y,dims:b??[y.length]})},"./node_modules/onnxruntime-common/dist/esm/tensor-factory.js":(Le,A,r)=>{r.r(A)},"./node_modules/onnxruntime-common/dist/esm/tensor-impl-type-mapping.js":(Le,A,r)=>{r.r(A),r.d(A,{NUMERIC_TENSOR_TYPEDARRAY_TO_TYPE_MAP:()=>D,NUMERIC_TENSOR_TYPE_TO_TYPEDARRAY_MAP:()=>_,checkTypedArray:()=>X});const _=new Map([["float32",Float32Array],["uint8",Uint8Array],["int8",Int8Array],["uint16",Uint16Array],["int16",Int16Array],["int32",Int32Array],["bool",Uint8Array],["float64",Float64Array],["uint32",Uint32Array],["int4",Uint8Array],["uint4",Uint8Array]]),D=new Map([[Float32Array,"float32"],[Uint8Array,"uint8"],[Int8Array,"int8"],[Uint16Array,"uint16"],[Int16Array,"int16"],[Int32Array,"int32"],[Float64Array,"float64"],[Uint32Array,"uint32"]]);let j=!1;const X=()=>{if(!j){j=!0;const R=typeof BigInt64Array<"u"&&BigInt64Array.from,g=typeof BigUint64Array<"u"&&BigUint64Array.from,v=typeof Float16Array<"u"&&Float16Array.from;R&&(_.set("int64",BigInt64Array),D.set(BigInt64Array,"int64")),g&&(_.set("uint64",BigUint64Array),D.set(BigUint64Array,"uint64")),v?(_.set("float16",Float16Array),D.set(Float16Array,"float16")):_.set("float16",Uint16Array)}}},"./node_modules/onnxruntime-common/dist/esm/tensor-impl.js":(Le,A,r)=>{r.r(A),r.d(A,{Tensor:()=>R});var _=r("./node_modules/onnxruntime-common/dist/esm/tensor-conversion-impl.js"),D=r("./node_modules/onnxruntime-common/dist/esm/tensor-factory-impl.js"),j=r("./node_modules/onnxruntime-common/dist/esm/tensor-impl-type-mapping.js"),X=r("./node_modules/onnxruntime-common/dist/esm/tensor-utils-impl.js");class R{constructor(v,M,y){(0,j.checkTypedArray)();let b,I;if(typeof v=="object"&&"location"in v)switch(this.dataLocation=v.location,b=v.type,I=v.dims,v.location){case"cpu-pinned":{const se=j.NUMERIC_TENSOR_TYPE_TO_TYPEDARRAY_MAP.get(b);if(!se)throw new TypeError(`unsupported type "${b}" to create tensor from pinned buffer`);if(!(v.data instanceof se))throw new TypeError(`buffer should be of type ${se.name}`);this.cpuData=v.data;break}case"texture":{if(b!=="float32")throw new TypeError(`unsupported type "${b}" to create tensor from texture`);this.gpuTextureData=v.texture,this.downloader=v.download,this.disposer=v.dispose;break}case"gpu-buffer":{if(b!=="float32"&&b!=="float16"&&b!=="int32"&&b!=="int64"&&b!=="uint32"&&b!=="uint8"&&b!=="bool"&&b!=="uint4"&&b!=="int4")throw new TypeError(`unsupported type "${b}" to create tensor from gpu buffer`);this.gpuBufferData=v.gpuBuffer,this.downloader=v.download,this.disposer=v.dispose;break}case"ml-tensor":{if(b!=="float32"&&b!=="float16"&&b!=="int32"&&b!=="int64"&&b!=="uint32"&&b!=="uint64"&&b!=="int8"&&b!=="uint8"&&b!=="bool")throw new TypeError(`unsupported type "${b}" to create tensor from MLTensor`);this.mlTensorData=v.mlTensor,this.downloader=v.download,this.disposer=v.dispose;break}default:throw new Error(`Tensor constructor: unsupported location '${this.dataLocation}'`)}else{let se,ne;if(typeof v=="string")if(b=v,ne=y,v==="string"){if(!Array.isArray(M))throw new TypeError("A string tensor's data must be a string array.");se=M}else{const W=j.NUMERIC_TENSOR_TYPE_TO_TYPEDARRAY_MAP.get(v);if(W===void 0)throw new TypeError(`Unsupported tensor type: ${v}.`);if(Array.isArray(M)){if(v==="float16"&&W===Uint16Array||v==="uint4"||v==="int4")throw new TypeError(`Creating a ${v} tensor from number array is not supported. Please use ${W.name} as data.`);v==="uint64"||v==="int64"?se=W.from(M,BigInt):se=W.from(M)}else if(M instanceof W)se=M;else if(M instanceof Uint8ClampedArray)if(v==="uint8")se=Uint8Array.from(M);else throw new TypeError("A Uint8ClampedArray tensor's data must be type of uint8");else throw new TypeError(`A ${b} tensor's data must be type of ${W}`)}else if(ne=M,Array.isArray(v)){if(v.length===0)throw new TypeError("Tensor type cannot be inferred from an empty array.");const W=typeof v[0];if(W==="string")b="string",se=v;else if(W==="boolean")b="bool",se=Uint8Array.from(v);else throw new TypeError(`Invalid element type of data array: ${W}.`)}else if(v instanceof Uint8ClampedArray)b="uint8",se=Uint8Array.from(v);else{const W=j.NUMERIC_TENSOR_TYPEDARRAY_TO_TYPE_MAP.get(v.constructor);if(W===void 0)throw new TypeError(`Unsupported type for tensor data: ${v.constructor}.`);b=W,se=v}if(ne===void 0)ne=[se.length];else if(!Array.isArray(ne))throw new TypeError("A tensor's dims must be a number array");I=ne,this.cpuData=se,this.dataLocation="cpu"}const K=(0,X.calculateSize)(I);if(this.cpuData&&K!==this.cpuData.length&&!((b==="uint4"||b==="int4")&&Math.ceil(K/2)===this.cpuData.length))throw new Error(`Tensor's size(${K}) does not match data length(${this.cpuData.length}).`);this.type=b,this.dims=I,this.size=K}static async fromImage(v,M){return(0,D.tensorFromImage)(v,M)}static fromTexture(v,M){return(0,D.tensorFromTexture)(v,M)}static fromGpuBuffer(v,M){return(0,D.tensorFromGpuBuffer)(v,M)}static fromMLTensor(v,M){return(0,D.tensorFromMLTensor)(v,M)}static fromPinnedBuffer(v,M,y){return(0,D.tensorFromPinnedBuffer)(v,M,y)}toDataURL(v){return(0,_.tensorToDataURL)(this,v)}toImageData(v){return(0,_.tensorToImageData)(this,v)}get data(){if(this.ensureValid(),!this.cpuData)throw new Error("The data is not on CPU. Use `getData()` to download GPU data to CPU, or use `texture` or `gpuBuffer` property to access the GPU data directly.");return this.cpuData}get location(){return this.dataLocation}get texture(){if(this.ensureValid(),!this.gpuTextureData)throw new Error("The data is not stored as a WebGL texture.");return this.gpuTextureData}get gpuBuffer(){if(this.ensureValid(),!this.gpuBufferData)throw new Error("The data is not stored as a WebGPU buffer.");return this.gpuBufferData}get mlTensor(){if(this.ensureValid(),!this.mlTensorData)throw new Error("The data is not stored as a WebNN MLTensor.");return this.mlTensorData}async getData(v){switch(this.ensureValid(),this.dataLocation){case"cpu":case"cpu-pinned":return this.data;case"texture":case"gpu-buffer":case"ml-tensor":{if(!this.downloader)throw new Error("The current tensor is not created with a specified data downloader.");if(this.isDownloading)throw new Error("The current tensor is being downloaded.");try{this.isDownloading=!0;const M=await this.downloader();return this.downloader=void 0,this.dataLocation="cpu",this.cpuData=M,v&&this.disposer&&(this.disposer(),this.disposer=void 0),M}finally{this.isDownloading=!1}}default:throw new Error(`cannot get data from location: ${this.dataLocation}`)}}dispose(){if(this.isDownloading)throw new Error("The current tensor is being downloaded.");this.disposer&&(this.disposer(),this.disposer=void 0),this.cpuData=void 0,this.gpuTextureData=void 0,this.gpuBufferData=void 0,this.mlTensorData=void 0,this.downloader=void 0,this.isDownloading=void 0,this.dataLocation="none"}ensureValid(){if(this.dataLocation==="none")throw new Error("The tensor is disposed.")}reshape(v){if(this.ensureValid(),this.downloader||this.disposer)throw new Error("Cannot reshape a tensor that owns GPU resource.");return(0,X.tensorReshape)(this,v)}}},"./node_modules/onnxruntime-common/dist/esm/tensor-utils-impl.js":(Le,A,r)=>{r.r(A),r.d(A,{calculateSize:()=>D,tensorReshape:()=>j});var _=r("./node_modules/onnxruntime-common/dist/esm/tensor-impl.js");const D=X=>{let R=1;for(let g=0;g{switch(X.location){case"cpu":return new _.Tensor(X.type,X.data,R);case"cpu-pinned":return new _.Tensor({location:"cpu-pinned",data:X.data,type:X.type,dims:R});case"texture":return new _.Tensor({location:"texture",texture:X.texture,type:X.type,dims:R});case"gpu-buffer":return new _.Tensor({location:"gpu-buffer",gpuBuffer:X.gpuBuffer,type:X.type,dims:R});case"ml-tensor":return new _.Tensor({location:"ml-tensor",mlTensor:X.mlTensor,type:X.type,dims:R});default:throw new Error(`tensorReshape: tensor location ${X.location} is not supported`)}}},"./node_modules/onnxruntime-common/dist/esm/tensor.js":(Le,A,r)=>{r.r(A),r.d(A,{Tensor:()=>D});var _=r("./node_modules/onnxruntime-common/dist/esm/tensor-impl.js");const D=_.Tensor},"./node_modules/onnxruntime-common/dist/esm/trace.js":(Le,A,r)=>{r.r(A),r.d(A,{TRACE:()=>D,TRACE_FUNC_BEGIN:()=>X,TRACE_FUNC_END:()=>R});var _=r("./node_modules/onnxruntime-common/dist/esm/env-impl.js");const D=(g,v)=>{(typeof _.env.trace>"u"?!_.env.wasm.trace:!_.env.trace)||console.timeStamp(`${g}::ORT::${v}`)},j=(g,v)=>{var b;const M=((b=new Error().stack)==null?void 0:b.split(/\r\n|\r|\n/g))||[];let y=!1;for(let I=0;I{(typeof _.env.trace>"u"?!_.env.wasm.trace:!_.env.trace)||j("BEGIN",g)},R=g=>{(typeof _.env.trace>"u"?!_.env.wasm.trace:!_.env.trace)||j("END",g)}},"./node_modules/onnxruntime-common/dist/esm/training-session-impl.js":(Le,A,r)=>{r.r(A),r.d(A,{TrainingSession:()=>X});var _=r("./node_modules/onnxruntime-common/dist/esm/backend-impl.js"),D=r("./node_modules/onnxruntime-common/dist/esm/tensor.js");const j="Training backend could not be resolved. Make sure you're using the correct configuration & WebAssembly files.";class X{constructor(g,v,M){this.handler=g,this.hasOptimizerModel=v,this.hasEvalModel=M}get trainingInputNames(){return this.handler.inputNames}get trainingOutputNames(){return this.handler.outputNames}get evalInputNames(){if(this.hasEvalModel)return this.handler.evalInputNames;throw new Error("This training session has no evalModel loaded.")}get evalOutputNames(){if(this.hasEvalModel)return this.handler.evalOutputNames;throw new Error("This training session has no evalModel loaded.")}static async create(g,v){const M=g.evalModel||"",y=g.optimizerModel||"",b=v||{},[I,K]=await(0,_.resolveBackendAndExecutionProviders)(b);if(I.createTrainingSessionHandler){const se=await I.createTrainingSessionHandler(g.checkpointState,g.trainModel,M,y,K);return new X(se,!!g.optimizerModel,!!g.evalModel)}else throw new Error(j)}typeNarrowingForRunStep(g,v,M,y,b){const I={};let K={};if(typeof M!="object"||M===null||M instanceof D.Tensor||Array.isArray(M))throw new TypeError("'feeds' must be an object that use input names as keys and OnnxValue as corresponding values.");let se=!0;if(typeof y=="object"){if(y===null)throw new TypeError("Unexpected argument[1]: cannot be null.");if(y instanceof D.Tensor)throw new TypeError("'fetches' cannot be a Tensor");if(Array.isArray(y)){if(y.length===0)throw new TypeError("'fetches' cannot be an empty array.");se=!1;for(const ne of y){if(typeof ne!="string")throw new TypeError("'fetches' must be a string array or an object.");if(v.indexOf(ne)===-1)throw new RangeError(`'fetches' contains invalid output name: ${ne}.`);I[ne]=null}if(typeof b=="object"&&b!==null)K=b;else if(typeof b<"u")throw new TypeError("'options' must be an object.")}else{let ne=!1;const W=Object.getOwnPropertyNames(y);for(const U of v)if(W.indexOf(U)!==-1){const q=y[U];(q===null||q instanceof D.Tensor)&&(ne=!0,se=!1,I[U]=q)}if(ne){if(typeof b=="object"&&b!==null)K=b;else if(typeof b<"u")throw new TypeError("'options' must be an object.")}else K=y}}else if(typeof y<"u")throw new TypeError("Unexpected argument[1]: must be 'fetches' or 'options'.");for(const ne of g)if(typeof M[ne]>"u")throw new Error(`input '${ne}' is missing in 'feeds'.`);if(se)for(const ne of v)I[ne]=null;return[I,K]}convertHandlerReturnTypeToMapOfTensors(g){const v={};for(const M in g)if(Object.hasOwnProperty.call(g,M)){const y=g[M];y instanceof D.Tensor?v[M]=y:v[M]=new D.Tensor(y.type,y.data,y.dims)}return v}async lazyResetGrad(){await this.handler.lazyResetGrad()}async runTrainStep(g,v,M){const[y,b]=this.typeNarrowingForRunStep(this.trainingInputNames,this.trainingOutputNames,g,v,M),I=await this.handler.runTrainStep(g,y,b);return this.convertHandlerReturnTypeToMapOfTensors(I)}async runOptimizerStep(g){if(this.hasOptimizerModel)await this.handler.runOptimizerStep(g||{});else throw new Error("This TrainingSession has no OptimizerModel loaded.")}async runEvalStep(g,v,M){if(this.hasEvalModel){const[y,b]=this.typeNarrowingForRunStep(this.evalInputNames,this.evalOutputNames,g,v,M),I=await this.handler.runEvalStep(g,y,b);return this.convertHandlerReturnTypeToMapOfTensors(I)}else throw new Error("This TrainingSession has no EvalModel loaded.")}async getParametersSize(g=!0){return this.handler.getParametersSize(g)}async loadParametersBuffer(g,v=!0){const M=await this.getParametersSize(v);if(g.length!==4*M)throw new Error("Size of the buffer passed into loadParametersBuffer must match the number of parameters in the model. Please use getParametersSize method to check.");return this.handler.loadParametersBuffer(g,v)}async getContiguousParameters(g=!0){return this.handler.getContiguousParameters(g)}async release(){return this.handler.dispose()}}},"./node_modules/onnxruntime-common/dist/esm/training-session.js":(Le,A,r)=>{r.r(A),r.d(A,{TrainingSession:()=>D});var _=r("./node_modules/onnxruntime-common/dist/esm/training-session-impl.js");const D=_.TrainingSession},"./node_modules/onnxruntime-common/dist/esm/version.js":(Le,A,r)=>{r.r(A),r.d(A,{version:()=>_});const _="1.20.1"},"./node_modules/onnxruntime-web/dist/ort.bundle.min.mjs?3a96":(Le,A,r)=>{r.r(A),r.d(A,{InferenceSession:()=>lt,TRACE:()=>Ue,TRACE_FUNC_BEGIN:()=>Ne,TRACE_FUNC_END:()=>Re,Tensor:()=>le,default:()=>ff,env:()=>O,registerBackend:()=>K});/*! + * ONNX Runtime Web v1.21.0-dev.20250114-228dd16893 + * Copyright (c) Microsoft Corporation. All rights reserved. + * Licensed under the MIT License. + */var _=Object.defineProperty,D=Object.getOwnPropertyDescriptor,j=Object.getOwnPropertyNames,X=Object.prototype.hasOwnProperty,R=(e=>typeof require<"u"?require:typeof Proxy<"u"?new Proxy(e,{get:(t,s)=>(typeof require<"u"?require:t)[s]}):e)(function(e){if(typeof require<"u")return require.apply(this,arguments);throw Error('Dynamic require of "'+e+'" is not supported')}),g=(e,t)=>()=>(e&&(t=e(e=0)),t),v=(e,t)=>{for(var s in t)_(e,s,{get:t[s],enumerable:!0})},M=(e,t,s,n)=>{if(t&&typeof t=="object"||typeof t=="function")for(let i of j(t))!X.call(e,i)&&i!==s&&_(e,i,{get:()=>t[i],enumerable:!(n=D(t,i))||n.enumerable});return e},y=e=>M(_({},"__esModule",{value:!0}),e),b,I,K,se,ne,W=g(()=>{b=new Map,I=[],K=(e,t,s)=>{if(t&&typeof t.init=="function"&&typeof t.createInferenceSessionHandler=="function"){let n=b.get(e);if(n===void 0)b.set(e,{backend:t,priority:s});else{if(n.priority>s)return;if(n.priority===s&&n.backend!==t)throw new Error(`cannot register backend "${e}" using priority ${s}`)}if(s>=0){let i=I.indexOf(e);i!==-1&&I.splice(i,1);for(let a=0;a{let t=b.get(e);if(!t)return"backend not found.";if(t.initialized)return t.backend;if(t.aborted)return t.error;{let s=!!t.initPromise;try{return s||(t.initPromise=t.backend.init(e)),await t.initPromise,t.initialized=!0,t.backend}catch(n){return s||(t.error=`${n}`,t.aborted=!0),t.error}finally{delete t.initPromise}}},ne=async e=>{let t=e.executionProviders||[],s=t.map(p=>typeof p=="string"?p:p.name),n=s.length===0?I:s,i,a=[],o=new Set;for(let p of n){let h=await se(p);typeof h=="string"?a.push({name:p,err:h}):(i||(i=h),i===h&&o.add(p))}if(!i)throw new Error(`no available backend found. ERR: ${a.map(p=>`[${p.name}] ${p.err}`).join(", ")}`);for(let{name:p,err:h}of a)s.includes(p)&&console.warn(`removing requested execution provider "${p}" from session options because it is not available: ${h}`);let u=t.filter(p=>o.has(typeof p=="string"?p:p.name));return[i,new Proxy(e,{get:(p,h)=>h==="executionProviders"?u:Reflect.get(p,h)})]}}),U=g(()=>{W()}),q,$=g(()=>{q="1.21.0-dev.20241212-1f88284f96"}),S,w,x=g(()=>{$(),S="warning",w={wasm:{},webgl:{},webgpu:{},versions:{common:q},set logLevel(e){if(e!==void 0){if(typeof e!="string"||["verbose","info","warning","error","fatal"].indexOf(e)===-1)throw new Error(`Unsupported logging level: ${e}`);S=e}},get logLevel(){return S}},Object.defineProperty(w,"logLevel",{enumerable:!0})}),O,ae=g(()=>{x(),O=w}),ie,ve,we=g(()=>{ie=(e,t)=>{let s=typeof document<"u"?document.createElement("canvas"):new OffscreenCanvas(1,1);s.width=e.dims[3],s.height=e.dims[2];let n=s.getContext("2d");if(n!=null){let i,a;(t==null?void 0:t.tensorLayout)!==void 0&&t.tensorLayout==="NHWC"?(i=e.dims[2],a=e.dims[3]):(i=e.dims[3],a=e.dims[2]);let o=(t==null?void 0:t.format)!==void 0?t.format:"RGB",u=t==null?void 0:t.norm,p,h;u===void 0||u.mean===void 0?p=[255,255,255,255]:typeof u.mean=="number"?p=[u.mean,u.mean,u.mean,u.mean]:(p=[u.mean[0],u.mean[1],u.mean[2],0],u.mean[3]!==void 0&&(p[3]=u.mean[3])),u===void 0||u.bias===void 0?h=[0,0,0,0]:typeof u.bias=="number"?h=[u.bias,u.bias,u.bias,u.bias]:(h=[u.bias[0],u.bias[1],u.bias[2],0],u.bias[3]!==void 0&&(h[3]=u.bias[3]));let k=a*i,C=0,d=k,z=k*2,B=-1;o==="RGBA"?(C=0,d=k,z=k*2,B=k*3):o==="RGB"?(C=0,d=k,z=k*2):o==="RBG"&&(C=0,z=k,d=k*2);for(let V=0;V{let s=typeof document<"u"?document.createElement("canvas").getContext("2d"):new OffscreenCanvas(1,1).getContext("2d"),n;if(s!=null){let i,a,o;(t==null?void 0:t.tensorLayout)!==void 0&&t.tensorLayout==="NHWC"?(i=e.dims[2],a=e.dims[1],o=e.dims[3]):(i=e.dims[3],a=e.dims[2],o=e.dims[1]);let u=t!==void 0&&t.format!==void 0?t.format:"RGB",p=t==null?void 0:t.norm,h,k;p===void 0||p.mean===void 0?h=[255,255,255,255]:typeof p.mean=="number"?h=[p.mean,p.mean,p.mean,p.mean]:(h=[p.mean[0],p.mean[1],p.mean[2],255],p.mean[3]!==void 0&&(h[3]=p.mean[3])),p===void 0||p.bias===void 0?k=[0,0,0,0]:typeof p.bias=="number"?k=[p.bias,p.bias,p.bias,p.bias]:(k=[p.bias[0],p.bias[1],p.bias[2],0],p.bias[3]!==void 0&&(k[3]=p.bias[3]));let C=a*i;if(t!==void 0&&(t.format!==void 0&&o===4&&t.format!=="RGBA"||o===3&&t.format!=="RGB"&&t.format!=="BGR"))throw new Error("Tensor format doesn't match input tensor dims");let d=4,z=0,B=1,V=2,Z=3,ee=0,Y=C,he=C*2,pe=-1;u==="RGBA"?(ee=0,Y=C,he=C*2,pe=C*3):u==="RGB"?(ee=0,Y=C,he=C*2):u==="RBG"&&(ee=0,he=C,Y=C*2),n=s.createImageData(i,a);for(let Me=0;Me{je(),re=(e,t)=>{if(e===void 0)throw new Error("Image buffer must be defined");if(t.height===void 0||t.width===void 0)throw new Error("Image height and width must be defined");if(t.tensorLayout==="NHWC")throw new Error("NHWC Tensor layout is not supported yet");let{height:s,width:n}=t,i=t.norm??{mean:255,bias:0},a,o;typeof i.mean=="number"?a=[i.mean,i.mean,i.mean,i.mean]:a=[i.mean[0],i.mean[1],i.mean[2],i.mean[3]??255],typeof i.bias=="number"?o=[i.bias,i.bias,i.bias,i.bias]:o=[i.bias[0],i.bias[1],i.bias[2],i.bias[3]??0];let u=t.format!==void 0?t.format:"RGBA",p=t.tensorFormat!==void 0&&t.tensorFormat!==void 0?t.tensorFormat:"RGB",h=s*n,k=p==="RGBA"?new Float32Array(h*4):new Float32Array(h*3),C=4,d=0,z=1,B=2,V=3,Z=0,ee=h,Y=h*2,he=-1;u==="RGB"&&(C=3,d=0,z=1,B=2,V=-1),p==="RGBA"?he=h*3:p==="RBG"?(Z=0,Y=h,ee=h*2):p==="BGR"&&(Y=0,ee=h,Z=h*2);for(let pe=0;pe{let s=typeof HTMLImageElement<"u"&&e instanceof HTMLImageElement,n=typeof ImageData<"u"&&e instanceof ImageData,i=typeof ImageBitmap<"u"&&e instanceof ImageBitmap,a=typeof e=="string",o,u=t??{},p=()=>{if(typeof document<"u")return document.createElement("canvas");if(typeof OffscreenCanvas<"u")return new OffscreenCanvas(1,1);throw new Error("Canvas is not supported")},h=k=>typeof HTMLCanvasElement<"u"&&k instanceof HTMLCanvasElement||k instanceof OffscreenCanvas?k.getContext("2d"):null;if(s){let k=p();k.width=e.width,k.height=e.height;let C=h(k);if(C!=null){let d=e.height,z=e.width;if(t!==void 0&&t.resizedHeight!==void 0&&t.resizedWidth!==void 0&&(d=t.resizedHeight,z=t.resizedWidth),t!==void 0){if(u=t,t.tensorFormat!==void 0)throw new Error("Image input config format must be RGBA for HTMLImageElement");u.tensorFormat="RGBA",u.height=d,u.width=z}else u.tensorFormat="RGBA",u.height=d,u.width=z;C.drawImage(e,0,0),o=C.getImageData(0,0,z,d).data}else throw new Error("Can not access image data")}else if(n){let k,C;if(t!==void 0&&t.resizedWidth!==void 0&&t.resizedHeight!==void 0?(k=t.resizedHeight,C=t.resizedWidth):(k=e.height,C=e.width),t!==void 0&&(u=t),u.format="RGBA",u.height=k,u.width=C,t!==void 0){let d=p();d.width=C,d.height=k;let z=h(d);if(z!=null)z.putImageData(e,0,0),o=z.getImageData(0,0,C,k).data;else throw new Error("Can not access image data")}else o=e.data}else if(i){if(t===void 0)throw new Error("Please provide image config with format for Imagebitmap");let k=p();k.width=e.width,k.height=e.height;let C=h(k);if(C!=null){let d=e.height,z=e.width;return C.drawImage(e,0,0,z,d),o=C.getImageData(0,0,z,d).data,u.height=d,u.width=z,re(o,u)}else throw new Error("Can not access image data")}else{if(a)return new Promise((k,C)=>{let d=p(),z=h(d);if(!e||!z)return C();let B=new Image;B.crossOrigin="Anonymous",B.src=e,B.onload=()=>{d.width=B.width,d.height=B.height,z.drawImage(B,0,0,d.width,d.height);let V=z.getImageData(0,0,d.width,d.height);u.height=d.height,u.width=d.width,k(re(V.data,u))}});throw new Error("Input data provided is not supported - aborted tensor creation")}if(o!==void 0)return re(o,u);throw new Error("Input data provided is not supported - aborted tensor creation")},ce=(e,t)=>{let{width:s,height:n,download:i,dispose:a}=t,o=[1,n,s,4];return new Ke({location:"texture",type:"float32",texture:e,dims:o,download:i,dispose:a})},ke=(e,t)=>{let{dataType:s,dims:n,download:i,dispose:a}=t;return new Ke({location:"gpu-buffer",type:s??"float32",gpuBuffer:e,dims:n,download:i,dispose:a})},Ie=(e,t)=>{let{dataType:s,dims:n,download:i,dispose:a}=t;return new Ke({location:"ml-tensor",type:s??"float32",mlTensor:e,dims:n,download:i,dispose:a})},Ee=(e,t,s)=>new Ke({location:"cpu-pinned",type:e,data:t,dims:s??[t.length]})}),Ge,ye,J,de,Ce=g(()=>{Ge=new Map([["float32",Float32Array],["uint8",Uint8Array],["int8",Int8Array],["uint16",Uint16Array],["int16",Int16Array],["int32",Int32Array],["bool",Uint8Array],["float64",Float64Array],["uint32",Uint32Array],["int4",Uint8Array],["uint4",Uint8Array]]),ye=new Map([[Float32Array,"float32"],[Uint8Array,"uint8"],[Int8Array,"int8"],[Uint16Array,"uint16"],[Int16Array,"int16"],[Int32Array,"int32"],[Float64Array,"float64"],[Uint32Array,"uint32"]]),J=!1,de=()=>{if(!J){J=!0;let e=typeof BigInt64Array<"u"&&BigInt64Array.from,t=typeof BigUint64Array<"u"&&BigUint64Array.from,s=typeof Float16Array<"u"&&Float16Array.from;e&&(Ge.set("int64",BigInt64Array),ye.set(BigInt64Array,"int64")),t&&(Ge.set("uint64",BigUint64Array),ye.set(BigUint64Array,"uint64")),s?(Ge.set("float16",Float16Array),ye.set(Float16Array,"float16")):Ge.set("float16",Uint16Array)}}}),Be,Ze,te=g(()=>{je(),Be=e=>{let t=1;for(let s=0;s{switch(e.location){case"cpu":return new Ke(e.type,e.data,t);case"cpu-pinned":return new Ke({location:"cpu-pinned",data:e.data,type:e.type,dims:t});case"texture":return new Ke({location:"texture",texture:e.texture,type:e.type,dims:t});case"gpu-buffer":return new Ke({location:"gpu-buffer",gpuBuffer:e.gpuBuffer,type:e.type,dims:t});case"ml-tensor":return new Ke({location:"ml-tensor",mlTensor:e.mlTensor,type:e.type,dims:t});default:throw new Error(`tensorReshape: tensor location ${e.location} is not supported`)}}}),Ke,je=g(()=>{we(),tt(),Ce(),te(),Ke=class{constructor(e,t,s){de();let n,i;if(typeof e=="object"&&"location"in e)switch(this.dataLocation=e.location,n=e.type,i=e.dims,e.location){case"cpu-pinned":{let o=Ge.get(n);if(!o)throw new TypeError(`unsupported type "${n}" to create tensor from pinned buffer`);if(!(e.data instanceof o))throw new TypeError(`buffer should be of type ${o.name}`);this.cpuData=e.data;break}case"texture":{if(n!=="float32")throw new TypeError(`unsupported type "${n}" to create tensor from texture`);this.gpuTextureData=e.texture,this.downloader=e.download,this.disposer=e.dispose;break}case"gpu-buffer":{if(n!=="float32"&&n!=="float16"&&n!=="int32"&&n!=="int64"&&n!=="uint32"&&n!=="uint8"&&n!=="bool"&&n!=="uint4"&&n!=="int4")throw new TypeError(`unsupported type "${n}" to create tensor from gpu buffer`);this.gpuBufferData=e.gpuBuffer,this.downloader=e.download,this.disposer=e.dispose;break}case"ml-tensor":{if(n!=="float32"&&n!=="float16"&&n!=="int32"&&n!=="int64"&&n!=="uint32"&&n!=="uint64"&&n!=="int8"&&n!=="uint8"&&n!=="bool"&&n!=="uint4"&&n!=="int4")throw new TypeError(`unsupported type "${n}" to create tensor from MLTensor`);this.mlTensorData=e.mlTensor,this.downloader=e.download,this.disposer=e.dispose;break}default:throw new Error(`Tensor constructor: unsupported location '${this.dataLocation}'`)}else{let o,u;if(typeof e=="string")if(n=e,u=s,e==="string"){if(!Array.isArray(t))throw new TypeError("A string tensor's data must be a string array.");o=t}else{let p=Ge.get(e);if(p===void 0)throw new TypeError(`Unsupported tensor type: ${e}.`);if(Array.isArray(t)){if(e==="float16"&&p===Uint16Array||e==="uint4"||e==="int4")throw new TypeError(`Creating a ${e} tensor from number array is not supported. Please use ${p.name} as data.`);e==="uint64"||e==="int64"?o=p.from(t,BigInt):o=p.from(t)}else if(t instanceof p)o=t;else if(t instanceof Uint8ClampedArray)if(e==="uint8")o=Uint8Array.from(t);else throw new TypeError("A Uint8ClampedArray tensor's data must be type of uint8");else throw new TypeError(`A ${n} tensor's data must be type of ${p}`)}else if(u=t,Array.isArray(e)){if(e.length===0)throw new TypeError("Tensor type cannot be inferred from an empty array.");let p=typeof e[0];if(p==="string")n="string",o=e;else if(p==="boolean")n="bool",o=Uint8Array.from(e);else throw new TypeError(`Invalid element type of data array: ${p}.`)}else if(e instanceof Uint8ClampedArray)n="uint8",o=Uint8Array.from(e);else{let p=ye.get(e.constructor);if(p===void 0)throw new TypeError(`Unsupported type for tensor data: ${e.constructor}.`);n=p,o=e}if(u===void 0)u=[o.length];else if(!Array.isArray(u))throw new TypeError("A tensor's dims must be a number array");i=u,this.cpuData=o,this.dataLocation="cpu"}let a=Be(i);if(this.cpuData&&a!==this.cpuData.length&&!((n==="uint4"||n==="int4")&&Math.ceil(a/2)===this.cpuData.length))throw new Error(`Tensor's size(${a}) does not match data length(${this.cpuData.length}).`);this.type=n,this.dims=i,this.size=a}static async fromImage(e,t){return xe(e,t)}static fromTexture(e,t){return ce(e,t)}static fromGpuBuffer(e,t){return ke(e,t)}static fromMLTensor(e,t){return Ie(e,t)}static fromPinnedBuffer(e,t,s){return Ee(e,t,s)}toDataURL(e){return ie(this,e)}toImageData(e){return ve(this,e)}get data(){if(this.ensureValid(),!this.cpuData)throw new Error("The data is not on CPU. Use `getData()` to download GPU data to CPU, or use `texture` or `gpuBuffer` property to access the GPU data directly.");return this.cpuData}get location(){return this.dataLocation}get texture(){if(this.ensureValid(),!this.gpuTextureData)throw new Error("The data is not stored as a WebGL texture.");return this.gpuTextureData}get gpuBuffer(){if(this.ensureValid(),!this.gpuBufferData)throw new Error("The data is not stored as a WebGPU buffer.");return this.gpuBufferData}get mlTensor(){if(this.ensureValid(),!this.mlTensorData)throw new Error("The data is not stored as a WebNN MLTensor.");return this.mlTensorData}async getData(e){switch(this.ensureValid(),this.dataLocation){case"cpu":case"cpu-pinned":return this.data;case"texture":case"gpu-buffer":case"ml-tensor":{if(!this.downloader)throw new Error("The current tensor is not created with a specified data downloader.");if(this.isDownloading)throw new Error("The current tensor is being downloaded.");try{this.isDownloading=!0;let t=await this.downloader();return this.downloader=void 0,this.dataLocation="cpu",this.cpuData=t,e&&this.disposer&&(this.disposer(),this.disposer=void 0),t}finally{this.isDownloading=!1}}default:throw new Error(`cannot get data from location: ${this.dataLocation}`)}}dispose(){if(this.isDownloading)throw new Error("The current tensor is being downloaded.");this.disposer&&(this.disposer(),this.disposer=void 0),this.cpuData=void 0,this.gpuTextureData=void 0,this.gpuBufferData=void 0,this.mlTensorData=void 0,this.downloader=void 0,this.isDownloading=void 0,this.dataLocation="none"}ensureValid(){if(this.dataLocation==="none")throw new Error("The tensor is disposed.")}reshape(e){if(this.ensureValid(),this.downloader||this.disposer)throw new Error("Cannot reshape a tensor that owns GPU resource.");return Ze(this,e)}}}),le,Te=g(()=>{je(),le=Ke}),Ue,Ve,Ne,Re,st=g(()=>{x(),Ue=(e,t)=>{(typeof w.trace>"u"?!w.wasm.trace:!w.trace)||console.timeStamp(`${e}::ORT::${t}`)},Ve=(e,t)=>{var i;let s=((i=new Error().stack)==null?void 0:i.split(/\r\n|\r|\n/g))||[],n=!1;for(let a=0;a{(typeof w.trace>"u"?!w.wasm.trace:!w.trace)||Ve("BEGIN",e)},Re=e=>{(typeof w.trace>"u"?!w.wasm.trace:!w.trace)||Ve("END",e)}}),dt,ct=g(()=>{W(),Te(),st(),dt=class tf{constructor(t){this.handler=t}async run(t,s,n){Ne();let i={},a={};if(typeof t!="object"||t===null||t instanceof le||Array.isArray(t))throw new TypeError("'feeds' must be an object that use input names as keys and OnnxValue as corresponding values.");let o=!0;if(typeof s=="object"){if(s===null)throw new TypeError("Unexpected argument[1]: cannot be null.");if(s instanceof le)throw new TypeError("'fetches' cannot be a Tensor");if(Array.isArray(s)){if(s.length===0)throw new TypeError("'fetches' cannot be an empty array.");o=!1;for(let h of s){if(typeof h!="string")throw new TypeError("'fetches' must be a string array or an object.");if(this.outputNames.indexOf(h)===-1)throw new RangeError(`'fetches' contains invalid output name: ${h}.`);i[h]=null}if(typeof n=="object"&&n!==null)a=n;else if(typeof n<"u")throw new TypeError("'options' must be an object.")}else{let h=!1,k=Object.getOwnPropertyNames(s);for(let C of this.outputNames)if(k.indexOf(C)!==-1){let d=s[C];(d===null||d instanceof le)&&(h=!0,o=!1,i[C]=d)}if(h){if(typeof n=="object"&&n!==null)a=n;else if(typeof n<"u")throw new TypeError("'options' must be an object.")}else a=s}}else if(typeof s<"u")throw new TypeError("Unexpected argument[1]: must be 'fetches' or 'options'.");for(let h of this.inputNames)if(typeof t[h]>"u")throw new Error(`input '${h}' is missing in 'feeds'.`);if(o)for(let h of this.outputNames)i[h]=null;let u=await this.handler.run(t,i,a),p={};for(let h in u)if(Object.hasOwnProperty.call(u,h)){let k=u[h];k instanceof le?p[h]=k:p[h]=new le(k.type,k.data,k.dims)}return Re(),p}async release(){return this.handler.dispose()}static async create(t,s,n,i){Ne();let a,o={};if(typeof t=="string"){if(a=t,typeof s=="object"&&s!==null)o=s;else if(typeof s<"u")throw new TypeError("'options' must be an object.")}else if(t instanceof Uint8Array){if(a=t,typeof s=="object"&&s!==null)o=s;else if(typeof s<"u")throw new TypeError("'options' must be an object.")}else if(t instanceof ArrayBuffer||typeof SharedArrayBuffer<"u"&&t instanceof SharedArrayBuffer){let k=t,C=0,d=t.byteLength;if(typeof s=="object"&&s!==null)o=s;else if(typeof s=="number"){if(C=s,!Number.isSafeInteger(C))throw new RangeError("'byteOffset' must be an integer.");if(C<0||C>=k.byteLength)throw new RangeError(`'byteOffset' is out of range [0, ${k.byteLength}).`);if(d=t.byteLength-C,typeof n=="number"){if(d=n,!Number.isSafeInteger(d))throw new RangeError("'byteLength' must be an integer.");if(d<=0||C+d>k.byteLength)throw new RangeError(`'byteLength' is out of range (0, ${k.byteLength-C}].`);if(typeof i=="object"&&i!==null)o=i;else if(typeof i<"u")throw new TypeError("'options' must be an object.")}else if(typeof n<"u")throw new TypeError("'byteLength' must be a number.")}else if(typeof s<"u")throw new TypeError("'options' must be an object.");a=new Uint8Array(k,C,d)}else throw new TypeError("Unexpected argument[0]: must be 'path' or 'buffer'.");let[u,p]=await ne(o),h=await u.createInferenceSessionHandler(a,p);return Re(),new tf(h)}startProfiling(){this.handler.startProfiling()}endProfiling(){this.handler.endProfiling()}get inputNames(){return this.handler.inputNames}get outputNames(){return this.handler.outputNames}}}),lt,ht=g(()=>{ct(),lt=dt}),L=g(()=>{}),oe=g(()=>{}),H=g(()=>{}),fe=g(()=>{}),$e={};v($e,{InferenceSession:()=>lt,TRACE:()=>Ue,TRACE_FUNC_BEGIN:()=>Ne,TRACE_FUNC_END:()=>Re,Tensor:()=>le,env:()=>O,registerBackend:()=>K});var We=g(()=>{U(),ae(),ht(),Te(),L(),oe(),st(),H(),fe()}),Je=g(()=>{}),ut={};v(ut,{default:()=>kt});var mt,vt,kt,At=g(()=>{var e;vh(),lr(),Nr(),mt="ort-wasm-proxy-worker",vt=((e=globalThis.self)==null?void 0:e.name)===mt,vt&&(self.onmessage=t=>{let{type:s,in:n}=t.data;try{switch(s){case"init-wasm":Or(n.wasm).then(()=>{Fi(n).then(()=>{postMessage({type:s})},i=>{postMessage({type:s,err:i})})},i=>{postMessage({type:s,err:i})});break;case"init-ep":{let{epName:i,env:a}=n;qn(a,i).then(()=>{postMessage({type:s})},o=>{postMessage({type:s,err:o})});break}case"copy-from":{let{buffer:i}=n,a=Qn(i);postMessage({type:s,out:a});break}case"create":{let{model:i,options:a}=n;Xn(i,a).then(o=>{postMessage({type:s,out:o})},o=>{postMessage({type:s,err:o})});break}case"release":Sc(n),postMessage({type:s});break;case"run":{let{sessionId:i,inputIndices:a,inputs:o,outputIndices:u,options:p}=n;Ac(i,a,o,u,new Array(u.length).fill(null),p).then(h=>{h.some(k=>k[3]!=="cpu")?postMessage({type:s,err:"Proxy does not support non-cpu tensor location."}):postMessage({type:s,out:h},mr([...o,...h]))},h=>{postMessage({type:s,err:h})});break}case"end-profiling":Qr(n),postMessage({type:s});break;default:}}catch(i){postMessage({type:s,err:i})}}),kt=vt?null:t=>new Worker(t??Us,{type:"module",name:mt})}),is={};v(is,{default:()=>Ds});var ys,Cs,Ds,sr=g(()=>{var e;Cs=(ys=self.location.href,async function(t={}){function s(){return Ft.buffer!=Rt.buffer&&Ps(),Rt}function n(){return Ft.buffer!=Rt.buffer&&Ps(),_s}function i(){return Ft.buffer!=Rt.buffer&&Ps(),ot}function a(){return Ft.buffer!=Rt.buffer&&Ps(),Et}function o(){return Ft.buffer!=Rt.buffer&&Ps(),ps}function u(){return Ft.buffer!=Rt.buffer&&Ps(),Ns}function p(){return Ft.buffer!=Rt.buffer&&Ps(),Tr}function h(){return Ft.buffer!=Rt.buffer&&Ps(),zi}var k,C,d=Object.assign({},t),z=new Promise((l,m)=>{k=l,C=m}),B=typeof window=="object",V=typeof importScripts=="function",Z=V&&self.name=="em-pthread";d.mountExternalData=(l,m)=>{l.startsWith("./")&&(l=l.substring(2)),(d.Fb||(d.Fb=new Map)).set(l,m)},d.unmountExternalData=()=>{delete d.Fb};var ee=globalThis.SharedArrayBuffer??new WebAssembly.Memory({initial:0,maximum:0,shared:!0}).buffer.constructor;let Y=()=>{let l=(E,F,G)=>(...ge)=>{let Xe=_n,nt=F==null?void 0:F();ge=E(...ge);let wt=F==null?void 0:F();return nt!==wt&&(E=wt,G(nt),F=G=null),_n!=Xe?new Promise((Ct,Wt)=>{mh={resolve:Ct,reject:Wt}}):ge},m=E=>async(...F)=>{var G;try{if(d.Gb)throw Error("Session already started");let ge=d.Gb={hc:F[0],errors:[]},Xe=await E(...F);if(d.Gb!==ge)throw Error("Session mismatch");(G=d.Hb)==null||G.flush();let nt=ge.errors;if(0Ct),0d._OrtCreateSession,E=>d._OrtCreateSession=E),d._OrtRun=m(l(d._OrtRun,()=>d._OrtRun,E=>d._OrtRun=E)),d._OrtRunWithBinding=m(l(d._OrtRunWithBinding,()=>d._OrtRunWithBinding,E=>d._OrtRunWithBinding=E)),d._OrtBindInput=l(d._OrtBindInput,()=>d._OrtBindInput,E=>d._OrtBindInput=E),Y=void 0};d.jsepInit=(l,m)=>{if(Y==null||Y(),l==="webgpu"){[d.Hb,d.Vb,d.Zb,d.Ob,d.Yb,d.kb,d.$b,d.cc,d.Wb,d.Xb,d.ac]=m;let E=d.Hb;d.jsepRegisterBuffer=(F,G,ge,Xe)=>E.registerBuffer(F,G,ge,Xe),d.jsepGetBuffer=F=>E.getBuffer(F),d.jsepCreateDownloader=(F,G,ge)=>E.createDownloader(F,G,ge),d.jsepOnCreateSession=F=>{E.onCreateSession(F)},d.jsepOnReleaseSession=F=>{E.onReleaseSession(F)},d.jsepOnRunStart=F=>E.onRunStart(F),d.dc=(F,G)=>{E.upload(F,G)}}else if(l==="webnn"){[d.Hb,d.bc,d.Pb,d.jsepEnsureTensor,d.ec,d.jsepDownloadTensor]=m,d.jsepReleaseTensorId=d.Pb;let E=d.Hb;d.jsepOnRunStart=F=>E.onRunStart(F),d.jsepRegisterMLContext=(F,G)=>{E.registerMLContext(F,G)},d.jsepOnReleaseSession=F=>{E.onReleaseSession(F)},d.jsepCreateMLTensorDownloader=(F,G)=>E.createMLTensorDownloader(F,G),d.jsepRegisterMLTensor=(F,G,ge)=>E.registerMLTensor(F,G,ge),d.jsepCreateMLContext=F=>E.createMLContext(F),d.jsepRegisterMLConstant=(F,G,ge,Xe,nt)=>E.registerMLConstant(F,G,ge,Xe,nt,d.Fb)}};var he,pe,Me=Object.assign({},d),Oe="./this.program",De=(l,m)=>{throw m},Ye="";(B||V)&&(V?Ye=self.location.href:typeof document<"u"&&document.currentScript&&(Ye=document.currentScript.src),ys&&(Ye=ys),Ye=Ye.startsWith("blob:")?"":Ye.substr(0,Ye.replace(/[?#].*/,"").lastIndexOf("/")+1),V&&(pe=l=>{var m=new XMLHttpRequest;return m.open("GET",l,!1),m.responseType="arraybuffer",m.send(null),new Uint8Array(m.response)}),he=(l,m,E)=>{var F=new XMLHttpRequest;F.open("GET",l,!0),F.responseType="arraybuffer",F.onload=()=>{F.status==200||F.status==0&&F.response?m(F.response):E()},F.onerror=E,F.send(null)});var at,Pt=console.log.bind(console),Xt=console.error.bind(console),Zt=Pt,bt=Xt;if(Object.assign(d,Me),Me=null,Z){let l=function(m){try{var E=m.data,F=E.cmd;if(F==="load"){let G=[];self.onmessage=ge=>G.push(ge),self.startWorker=()=>{postMessage({cmd:"loaded"});for(let ge of G)l(ge);self.onmessage=l};for(let ge of E.handlers)d[ge]&&!d[ge].proxy||(d[ge]=(...Xe)=>{postMessage({Nb:"callHandler",pc:ge,args:Xe})},ge=="print"&&(Zt=d[ge]),ge=="printErr"&&(bt=d[ge]));Ft=E.wasmMemory,Ps(),ss(E.wasmModule)}else if(F==="run"){wh(E.pthread_ptr,0,0,1,0,0),ph(E.pthread_ptr),wf(),Kh(),St||(Vm(),St=!0);try{yf(E.start_routine,E.arg)}catch(G){if(G!="unwind")throw G}}else F==="cancel"?La()&&Up(-1):E.target!=="setimmediate"&&(F==="checkMailbox"?St&&Ip():F&&(bt(`worker: received unknown command ${F}`),bt(E)))}catch(G){throw Wm(),G}};var ss,St=!1;bt=function(...m){m=m.join(" "),console.error(m)},self.alert=function(...m){postMessage({Nb:"alert",text:m.join(" "),rc:La()})},d.instantiateWasm=(m,E)=>new Promise(F=>{ss=G=>{G=new WebAssembly.Instance(G,jh()),E(G),F()}}),self.onunhandledrejection=m=>{throw m.reason||m},self.onmessage=l}d.wasmBinary&&(at=d.wasmBinary);var Ft,bs,Ht,Rt,_s,ot,Et,ps,Ns,Tr,er,hn,zi,zs=!1;function Ps(){var l=Ft.buffer;d.HEAP8=Rt=new Int8Array(l),d.HEAP16=ot=new Int16Array(l),d.HEAPU8=_s=new Uint8Array(l),d.HEAPU16=Et=new Uint16Array(l),d.HEAP32=ps=new Int32Array(l),d.HEAPU32=Ns=new Uint32Array(l),d.HEAPF32=Tr=new Float32Array(l),d.HEAPF64=zi=new Float64Array(l),d.HEAP64=er=new BigInt64Array(l),d.HEAPU64=hn=new BigUint64Array(l)}if(!Z){if(!((Ft=new WebAssembly.Memory({initial:256,maximum:65536,shared:!0})).buffer instanceof ee))throw bt("requested a shared WebAssembly.Memory but the returned buffer is not a SharedArrayBuffer, indicating that while the browser has SharedArrayBuffer it does not have WebAssembly threads support - you may need to set a flag"),Error("bad memory");Ps()}var Jn=[],Fc=[],Zp=[],Dc=0,Lc=null;function Lh(){if(--Dc==0&&Lc){var l=Lc;Lc=null,l()}}function Zn(l){throw bt(l="Aborted("+l+")"),zs=!0,Ht=1,l=new WebAssembly.RuntimeError(l+". Build with -sASSERTIONS for more info."),C(l),l}var eh,zh=l=>l.startsWith("data:application/octet-stream;base64,"),Bh=l=>l.startsWith("file://");function Rh(l){if(l==eh&&at)return new Uint8Array(at);if(pe)return pe(l);throw"both async and sync fetching of the wasm failed"}function Nh(l,m,E){return function(F){if(!at&&(B||V)){if(typeof fetch=="function"&&!Bh(F))return fetch(F,{credentials:"same-origin"}).then(G=>{if(!G.ok)throw`failed to load wasm binary file at '${F}'`;return G.arrayBuffer()}).catch(()=>Rh(F));if(he)return new Promise((G,ge)=>{he(F,Xe=>G(new Uint8Array(Xe)),ge)})}return Promise.resolve().then(()=>Rh(F))}(l).then(F=>WebAssembly.instantiate(F,m)).then(E,F=>{bt(`failed to asynchronously prepare wasm: ${F}`),Zn(F)})}function jh(){return{a:{O:gf,Aa:_f,b:bf,aa:Xh,B:Zh,qa:em,Y:sm,_:rm,ra:nm,oa:im,ha:om,na:am,L:lm,Z:um,W:dm,pa:cm,X:pm,va:vf,F:xf,Q:Tf,P:Ef,E:kf,u:Sf,q:$f,G:Af,A:Bf,R:Rf,ua:Nf,ka:jf,U:Uf,ba:Vf,H:Wf,ja:ph,ta:Gf,t:Kf,Ba:Hf,x:Xf,n:Yf,l:Zf,c:dh,o:e_,j:r_,w:n_,p:i_,f:o_,s:a_,m:l_,e:u_,k:d_,i:c_,h:p_,d:h_,ea:m_,fa:f_,ga:__,ca:Em,da:Cm,T:g_,g:w_,D:y_,I:M_,M:b_,y:v_,sa:x_,V:T_,v:Sm,z:P_,N:E_,S:C_,za:k_,ya:S_,la:Im,ma:Om,$:ih,C:Fm,K:Dm,ia:Lm,J:zm,a:Ft,xa:nh,wa:Nm,r:I_}}}var th={913700:(l,m,E,F,G)=>{if(d===void 0||!d.Fb)return 1;if((l=Qs(Number(l>>>0))).startsWith("./")&&(l=l.substring(2)),!(l=d.Fb.get(l)))return 2;if(m=Number(m>>>0),E=Number(E>>>0),F=Number(F>>>0),m+E>l.byteLength)return 3;try{let ge=l.subarray(m,m+E);switch(G){case 0:n().set(ge,F>>>0);break;case 1:d.dc(F,ge);break;default:return 4}return 0}catch{return 4}},914415:(l,m,E)=>{d.ec(l,n().subarray(m>>>0,m+E>>>0))},914478:()=>d.bc(),914519:l=>{d.Pb(l)},914555:()=>{d.Wb()},914586:()=>{d.Xb()},914615:()=>{d.ac()},914640:l=>d.Vb(l),914673:l=>d.Zb(l),914705:(l,m,E)=>{d.Ob(Number(l),Number(m),Number(E),!0)},914768:(l,m,E)=>{d.Ob(Number(l),Number(m),Number(E))},914825:()=>typeof wasmOffsetConverter<"u",914882:l=>{d.kb("Abs",l,void 0)},914933:l=>{d.kb("Neg",l,void 0)},914984:l=>{d.kb("Floor",l,void 0)},915037:l=>{d.kb("Ceil",l,void 0)},915089:l=>{d.kb("Reciprocal",l,void 0)},915147:l=>{d.kb("Sqrt",l,void 0)},915199:l=>{d.kb("Exp",l,void 0)},915250:l=>{d.kb("Erf",l,void 0)},915301:l=>{d.kb("Sigmoid",l,void 0)},915356:(l,m,E)=>{d.kb("HardSigmoid",l,{alpha:m,beta:E})},915435:l=>{d.kb("Log",l,void 0)},915486:l=>{d.kb("Sin",l,void 0)},915537:l=>{d.kb("Cos",l,void 0)},915588:l=>{d.kb("Tan",l,void 0)},915639:l=>{d.kb("Asin",l,void 0)},915691:l=>{d.kb("Acos",l,void 0)},915743:l=>{d.kb("Atan",l,void 0)},915795:l=>{d.kb("Sinh",l,void 0)},915847:l=>{d.kb("Cosh",l,void 0)},915899:l=>{d.kb("Asinh",l,void 0)},915952:l=>{d.kb("Acosh",l,void 0)},916005:l=>{d.kb("Atanh",l,void 0)},916058:l=>{d.kb("Tanh",l,void 0)},916110:l=>{d.kb("Not",l,void 0)},916161:(l,m,E)=>{d.kb("Clip",l,{min:m,max:E})},916230:l=>{d.kb("Clip",l,void 0)},916282:(l,m)=>{d.kb("Elu",l,{alpha:m})},916340:l=>{d.kb("Gelu",l,void 0)},916392:l=>{d.kb("Relu",l,void 0)},916444:(l,m)=>{d.kb("LeakyRelu",l,{alpha:m})},916508:(l,m)=>{d.kb("ThresholdedRelu",l,{alpha:m})},916578:(l,m)=>{d.kb("Cast",l,{to:m})},916636:l=>{d.kb("Add",l,void 0)},916687:l=>{d.kb("Sub",l,void 0)},916738:l=>{d.kb("Mul",l,void 0)},916789:l=>{d.kb("Div",l,void 0)},916840:l=>{d.kb("Pow",l,void 0)},916891:l=>{d.kb("Equal",l,void 0)},916944:l=>{d.kb("Greater",l,void 0)},916999:l=>{d.kb("GreaterOrEqual",l,void 0)},917061:l=>{d.kb("Less",l,void 0)},917113:l=>{d.kb("LessOrEqual",l,void 0)},917172:(l,m,E,F,G)=>{d.kb("ReduceMean",l,{keepDims:!!m,noopWithEmptyAxes:!!E,axes:F?Array.from(o().subarray(Number(F)>>>0,Number(G)>>>0)):[]})},917347:(l,m,E,F,G)=>{d.kb("ReduceMax",l,{keepDims:!!m,noopWithEmptyAxes:!!E,axes:F?Array.from(o().subarray(Number(F)>>>0,Number(G)>>>0)):[]})},917521:(l,m,E,F,G)=>{d.kb("ReduceMin",l,{keepDims:!!m,noopWithEmptyAxes:!!E,axes:F?Array.from(o().subarray(Number(F)>>>0,Number(G)>>>0)):[]})},917695:(l,m,E,F,G)=>{d.kb("ReduceProd",l,{keepDims:!!m,noopWithEmptyAxes:!!E,axes:F?Array.from(o().subarray(Number(F)>>>0,Number(G)>>>0)):[]})},917870:(l,m,E,F,G)=>{d.kb("ReduceSum",l,{keepDims:!!m,noopWithEmptyAxes:!!E,axes:F?Array.from(o().subarray(Number(F)>>>0,Number(G)>>>0)):[]})},918044:(l,m,E,F,G)=>{d.kb("ReduceL1",l,{keepDims:!!m,noopWithEmptyAxes:!!E,axes:F?Array.from(o().subarray(Number(F)>>>0,Number(G)>>>0)):[]})},918217:(l,m,E,F,G)=>{d.kb("ReduceL2",l,{keepDims:!!m,noopWithEmptyAxes:!!E,axes:F?Array.from(o().subarray(Number(F)>>>0,Number(G)>>>0)):[]})},918390:(l,m,E,F,G)=>{d.kb("ReduceLogSum",l,{keepDims:!!m,noopWithEmptyAxes:!!E,axes:F?Array.from(o().subarray(Number(F)>>>0,Number(G)>>>0)):[]})},918567:(l,m,E,F,G)=>{d.kb("ReduceSumSquare",l,{keepDims:!!m,noopWithEmptyAxes:!!E,axes:F?Array.from(o().subarray(Number(F)>>>0,Number(G)>>>0)):[]})},918747:(l,m,E,F,G)=>{d.kb("ReduceLogSumExp",l,{keepDims:!!m,noopWithEmptyAxes:!!E,axes:F?Array.from(o().subarray(Number(F)>>>0,Number(G)>>>0)):[]})},918927:l=>{d.kb("Where",l,void 0)},918980:(l,m,E)=>{d.kb("Transpose",l,{perm:m?Array.from(o().subarray(Number(m)>>>0,Number(E)>>>0)):[]})},919104:(l,m,E,F)=>{d.kb("DepthToSpace",l,{blocksize:m,mode:Qs(E),format:F?"NHWC":"NCHW"})},919237:(l,m,E,F)=>{d.kb("DepthToSpace",l,{blocksize:m,mode:Qs(E),format:F?"NHWC":"NCHW"})},919370:(l,m,E,F,G,ge,Xe,nt,wt,Ct,Wt,hs,vs,He,ls)=>{d.kb("ConvTranspose",l,{format:wt?"NHWC":"NCHW",autoPad:m,dilations:[E],group:F,kernelShape:[G],pads:[ge,Xe],strides:[nt],wIsConst:()=>!!s()[Ct>>>0],outputPadding:Wt?Array.from(o().subarray(Number(Wt)>>>0,Number(hs)>>>0)):[],outputShape:vs?Array.from(o().subarray(Number(vs)>>>0,Number(He)>>>0)):[],activation:Qs(ls)})},919803:(l,m,E,F,G,ge,Xe,nt,wt,Ct,Wt,hs,vs,He)=>{d.kb("ConvTranspose",l,{format:nt?"NHWC":"NCHW",autoPad:m,dilations:Array.from(o().subarray(Number(E)>>>0,2+(Number(E)>>>0)>>>0)),group:F,kernelShape:Array.from(o().subarray(Number(G)>>>0,2+(Number(G)>>>0)>>>0)),pads:Array.from(o().subarray(Number(ge)>>>0,4+(Number(ge)>>>0)>>>0)),strides:Array.from(o().subarray(Number(Xe)>>>0,2+(Number(Xe)>>>0)>>>0)),wIsConst:()=>!!s()[wt>>>0],outputPadding:Ct?Array.from(o().subarray(Number(Ct)>>>0,Number(Wt)>>>0)):[],outputShape:hs?Array.from(o().subarray(Number(hs)>>>0,Number(vs)>>>0)):[],activation:Qs(He)})},920464:(l,m,E,F,G,ge,Xe,nt,wt,Ct,Wt,hs,vs,He,ls)=>{d.kb("ConvTranspose",l,{format:wt?"NHWC":"NCHW",autoPad:m,dilations:[E],group:F,kernelShape:[G],pads:[ge,Xe],strides:[nt],wIsConst:()=>!!s()[Ct>>>0],outputPadding:Wt?Array.from(o().subarray(Number(Wt)>>>0,Number(hs)>>>0)):[],outputShape:vs?Array.from(o().subarray(Number(vs)>>>0,Number(He)>>>0)):[],activation:Qs(ls)})},920897:(l,m,E,F,G,ge,Xe,nt,wt,Ct,Wt,hs,vs,He)=>{d.kb("ConvTranspose",l,{format:nt?"NHWC":"NCHW",autoPad:m,dilations:Array.from(o().subarray(Number(E)>>>0,2+(Number(E)>>>0)>>>0)),group:F,kernelShape:Array.from(o().subarray(Number(G)>>>0,2+(Number(G)>>>0)>>>0)),pads:Array.from(o().subarray(Number(ge)>>>0,4+(Number(ge)>>>0)>>>0)),strides:Array.from(o().subarray(Number(Xe)>>>0,2+(Number(Xe)>>>0)>>>0)),wIsConst:()=>!!s()[wt>>>0],outputPadding:Ct?Array.from(o().subarray(Number(Ct)>>>0,Number(Wt)>>>0)):[],outputShape:hs?Array.from(o().subarray(Number(hs)>>>0,Number(vs)>>>0)):[],activation:Qs(He)})},921558:(l,m)=>{d.kb("GlobalAveragePool",l,{format:m?"NHWC":"NCHW"})},921649:(l,m,E,F,G,ge,Xe,nt,wt,Ct,Wt,hs,vs,He)=>{d.kb("AveragePool",l,{format:He?"NHWC":"NCHW",auto_pad:m,ceil_mode:E,count_include_pad:F,storage_order:G,dilations:ge?Array.from(o().subarray(Number(ge)>>>0,Number(Xe)>>>0)):[],kernel_shape:nt?Array.from(o().subarray(Number(nt)>>>0,Number(wt)>>>0)):[],pads:Ct?Array.from(o().subarray(Number(Ct)>>>0,Number(Wt)>>>0)):[],strides:hs?Array.from(o().subarray(Number(hs)>>>0,Number(vs)>>>0)):[]})},922128:(l,m)=>{d.kb("GlobalAveragePool",l,{format:m?"NHWC":"NCHW"})},922219:(l,m,E,F,G,ge,Xe,nt,wt,Ct,Wt,hs,vs,He)=>{d.kb("AveragePool",l,{format:He?"NHWC":"NCHW",auto_pad:m,ceil_mode:E,count_include_pad:F,storage_order:G,dilations:ge?Array.from(o().subarray(Number(ge)>>>0,Number(Xe)>>>0)):[],kernel_shape:nt?Array.from(o().subarray(Number(nt)>>>0,Number(wt)>>>0)):[],pads:Ct?Array.from(o().subarray(Number(Ct)>>>0,Number(Wt)>>>0)):[],strides:hs?Array.from(o().subarray(Number(hs)>>>0,Number(vs)>>>0)):[]})},922698:(l,m)=>{d.kb("GlobalMaxPool",l,{format:m?"NHWC":"NCHW"})},922785:(l,m,E,F,G,ge,Xe,nt,wt,Ct,Wt,hs,vs,He)=>{d.kb("MaxPool",l,{format:He?"NHWC":"NCHW",auto_pad:m,ceil_mode:E,count_include_pad:F,storage_order:G,dilations:ge?Array.from(o().subarray(Number(ge)>>>0,Number(Xe)>>>0)):[],kernel_shape:nt?Array.from(o().subarray(Number(nt)>>>0,Number(wt)>>>0)):[],pads:Ct?Array.from(o().subarray(Number(Ct)>>>0,Number(Wt)>>>0)):[],strides:hs?Array.from(o().subarray(Number(hs)>>>0,Number(vs)>>>0)):[]})},923260:(l,m)=>{d.kb("GlobalMaxPool",l,{format:m?"NHWC":"NCHW"})},923347:(l,m,E,F,G,ge,Xe,nt,wt,Ct,Wt,hs,vs,He)=>{d.kb("MaxPool",l,{format:He?"NHWC":"NCHW",auto_pad:m,ceil_mode:E,count_include_pad:F,storage_order:G,dilations:ge?Array.from(o().subarray(Number(ge)>>>0,Number(Xe)>>>0)):[],kernel_shape:nt?Array.from(o().subarray(Number(nt)>>>0,Number(wt)>>>0)):[],pads:Ct?Array.from(o().subarray(Number(Ct)>>>0,Number(Wt)>>>0)):[],strides:hs?Array.from(o().subarray(Number(hs)>>>0,Number(vs)>>>0)):[]})},923822:(l,m,E,F,G)=>{d.kb("Gemm",l,{alpha:m,beta:E,transA:F,transB:G})},923926:l=>{d.kb("MatMul",l,void 0)},923980:(l,m,E,F)=>{d.kb("ArgMax",l,{keepDims:!!m,selectLastIndex:!!E,axis:F})},924088:(l,m,E,F)=>{d.kb("ArgMin",l,{keepDims:!!m,selectLastIndex:!!E,axis:F})},924196:(l,m)=>{d.kb("Softmax",l,{axis:m})},924259:(l,m)=>{d.kb("Concat",l,{axis:m})},924319:(l,m,E,F,G)=>{d.kb("Split",l,{axis:m,numOutputs:E,splitSizes:F?Array.from(o().subarray(Number(F)>>>0,Number(G)>>>0)):[]})},924475:l=>{d.kb("Expand",l,void 0)},924529:(l,m)=>{d.kb("Gather",l,{axis:Number(m)})},924600:(l,m)=>{d.kb("GatherElements",l,{axis:Number(m)})},924679:(l,m)=>{d.kb("GatherND",l,{batch_dims:Number(m)})},924758:(l,m,E,F,G,ge,Xe,nt,wt,Ct,Wt)=>{d.kb("Resize",l,{antialias:m,axes:E?Array.from(o().subarray(Number(E)>>>0,Number(F)>>>0)):[],coordinateTransformMode:Qs(G),cubicCoeffA:ge,excludeOutside:Xe,extrapolationValue:nt,keepAspectRatioPolicy:Qs(wt),mode:Qs(Ct),nearestMode:Qs(Wt)})},925120:(l,m,E,F,G,ge,Xe)=>{d.kb("Slice",l,{starts:m?Array.from(o().subarray(Number(m)>>>0,Number(E)>>>0)):[],ends:F?Array.from(o().subarray(Number(F)>>>0,Number(G)>>>0)):[],axes:ge?Array.from(o().subarray(Number(ge)>>>0,Number(Xe)>>>0)):[]})},925384:l=>{d.kb("Tile",l,void 0)},925436:(l,m,E)=>{d.kb("InstanceNormalization",l,{epsilon:m,format:E?"NHWC":"NCHW"})},925550:(l,m,E)=>{d.kb("InstanceNormalization",l,{epsilon:m,format:E?"NHWC":"NCHW"})},925664:l=>{d.kb("Range",l,void 0)},925717:(l,m)=>{d.kb("Einsum",l,{equation:Qs(m)})},925798:(l,m,E,F,G)=>{d.kb("Pad",l,{mode:m,value:E,pads:F?Array.from(o().subarray(Number(F)>>>0,Number(G)>>>0)):[]})},925941:(l,m,E,F,G,ge)=>{d.kb("BatchNormalization",l,{epsilon:m,momentum:E,spatial:!!G,trainingMode:!!F,format:ge?"NHWC":"NCHW"})},926110:(l,m,E,F,G,ge)=>{d.kb("BatchNormalization",l,{epsilon:m,momentum:E,spatial:!!G,trainingMode:!!F,format:ge?"NHWC":"NCHW"})},926279:(l,m,E)=>{d.kb("CumSum",l,{exclusive:Number(m),reverse:Number(E)})},926376:(l,m,E)=>{d.kb("DequantizeLinear",l,{axis:m,blockSize:E})},926466:(l,m,E,F,G)=>{d.kb("GridSample",l,{align_corners:m,mode:Qs(E),padding_mode:Qs(F),format:G?"NHWC":"NCHW"})},926636:(l,m,E,F,G)=>{d.kb("GridSample",l,{align_corners:m,mode:Qs(E),padding_mode:Qs(F),format:G?"NHWC":"NCHW"})},926806:(l,m,E,F,G,ge,Xe,nt,wt)=>{d.kb("Attention",l,{numHeads:m,isUnidirectional:E,maskFilterValue:F,scale:G,doRotary:ge,qkvHiddenSizes:Xe?Array.from(o().subarray(Number(nt)>>>0,Number(nt)+Xe>>>0)):[],pastPresentShareBuffer:!!wt})},927078:l=>{d.kb("BiasAdd",l,void 0)},927133:l=>{d.kb("BiasSplitGelu",l,void 0)},927194:l=>{d.kb("FastGelu",l,void 0)},927250:(l,m,E,F,G,ge,Xe,nt,wt,Ct,Wt,hs,vs,He,ls,js)=>{d.kb("Conv",l,{format:hs?"NHWC":"NCHW",auto_pad:m,dilations:E?Array.from(o().subarray(Number(E)>>>0,Number(F)>>>0)):[],group:G,kernel_shape:ge?Array.from(o().subarray(Number(ge)>>>0,Number(Xe)>>>0)):[],pads:nt?Array.from(o().subarray(Number(nt)>>>0,Number(wt)>>>0)):[],strides:Ct?Array.from(o().subarray(Number(Ct)>>>0,Number(Wt)>>>0)):[],w_is_const:()=>!!s()[Number(vs)>>>0],activation:Qs(He),activation_params:ls?Array.from(p().subarray(Number(ls)>>>0,Number(js)>>>0)):[]})},927834:l=>{d.kb("Gelu",l,void 0)},927886:(l,m,E,F,G,ge,Xe,nt,wt)=>{d.kb("GroupQueryAttention",l,{numHeads:m,kvNumHeads:E,scale:F,softcap:G,doRotary:ge,rotaryInterleaved:Xe,smoothSoftmax:nt,localWindowSize:wt})},928103:(l,m,E,F)=>{d.kb("LayerNormalization",l,{axis:m,epsilon:E,simplified:!!F})},928214:(l,m,E,F)=>{d.kb("LayerNormalization",l,{axis:m,epsilon:E,simplified:!!F})},928325:(l,m,E,F,G,ge)=>{d.kb("MatMulNBits",l,{k:m,n:E,accuracyLevel:F,bits:G,blockSize:ge})},928452:(l,m,E,F,G,ge)=>{d.kb("MultiHeadAttention",l,{numHeads:m,isUnidirectional:E,maskFilterValue:F,scale:G,doRotary:ge})},928611:(l,m)=>{d.kb("QuickGelu",l,{alpha:m})},928675:(l,m,E,F,G)=>{d.kb("RotaryEmbedding",l,{interleaved:!!m,numHeads:E,rotaryEmbeddingDim:F,scale:G})},928814:(l,m,E)=>{d.kb("SkipLayerNormalization",l,{epsilon:m,simplified:!!E})},928916:(l,m,E)=>{d.kb("SkipLayerNormalization",l,{epsilon:m,simplified:!!E})},929018:(l,m,E,F)=>{d.kb("GatherBlockQuantized",l,{gatherAxis:m,quantizeAxis:E,blockSize:F})},929139:l=>{d.$b(l)},929173:(l,m)=>d.cc(Number(l),Number(m),d.Gb.hc,d.Gb.errors)};function _f(l,m,E){return bm(async()=>{await d.Yb(Number(l),Number(m),Number(E))})}function gf(){return typeof wasmOffsetConverter<"u"}function sh(l){this.name="ExitStatus",this.message=`Program terminated with exit(${l})`,this.status=l}var rh=l=>{l.terminate(),l.onmessage=()=>{}},Uh=l=>{ei.length==0&&(qh(),Hh(ei[0]));var m=ei.pop();if(!m)return 6;Ri.push(m),mn[l.Bb]=m,m.Bb=l.Bb;var E={cmd:"run",start_routine:l.ic,arg:l.Rb,pthread_ptr:l.Bb};return m.postMessage(E,l.nc),0},Bi=0,Bs=(l,m,...E)=>{for(var F=2*E.length,G=bh(),ge=Mh(8*F),Xe=ge>>>3,nt=0;nt>>0]=wt)}return l=Gm(l,0,F,ge,m),Vp(G),l};function nh(l){if(Z)return Bs(0,1,l);if(Ht=l,!(0{if(Ht=l,Z)throw Vh(l),"unwind";nh(l)},ei=[],Ri=[],Wh=[],mn={},Gh=l=>{var m=l.Bb;delete mn[m],ei.push(l),Ri.splice(Ri.indexOf(l),1),l.Bb=0,yh(m)};function Kh(){Wh.forEach(l=>l())}var Hh=l=>new Promise(m=>{l.onmessage=G=>{var ge=(G=G.data).cmd;if(G.targetThread&&G.targetThread!=La()){var Xe=mn[G.targetThread];Xe?Xe.postMessage(G,G.transferList):bt(`Internal error! Worker sent a message "${ge}" to target pthread ${G.targetThread}, but that thread no longer exists!`)}else ge==="checkMailbox"?Ip():ge==="spawnThread"?Uh(G):ge==="cleanupThread"?Gh(mn[G.thread]):ge==="killThread"?(G=G.thread,ge=mn[G],delete mn[G],rh(ge),yh(G),Ri.splice(Ri.indexOf(ge),1),ge.Bb=0):ge==="cancelThread"?mn[G.thread].postMessage({cmd:"cancel"}):ge==="loaded"?(l.loaded=!0,m(l)):ge==="alert"?alert(`Thread ${G.threadId}: ${G.text}`):G.target==="setimmediate"?l.postMessage(G):ge==="callHandler"?d[G.handler](...G.args):ge&&bt(`worker sent an unknown command ${ge}`)},l.onerror=G=>{throw bt(`worker sent an error! ${G.filename}:${G.lineno}: ${G.message}`),G};var E,F=[];for(E of[])d.hasOwnProperty(E)&&F.push(E);l.postMessage({cmd:"load",handlers:F,wasmMemory:Ft,wasmModule:bs})});function qh(){var l=new Worker(self.location.href.startsWith("file:")?new URL(r("./node_modules/onnxruntime-web/dist/ort.bundle.min.mjs?46eb"),r.b):new URL(self.location.href),{type:"module",workerData:"em-pthread",name:"em-pthread"});ei.push(l)}var Ap=l=>{for(;0{var l=La(),m=u()[l+52>>>2>>>0];l=u()[l+56>>>2>>>0],Hm(m,m-l),Vp(m)},yf=(l,m)=>{Bi=0,l=qm(l,m),0>>=0);throw m>>>=0,E>>>=0,u()[F.Kb+16>>>2>>>0]=0,u()[F.Kb+4>>>2>>>0]=m,u()[F.Kb+8>>>2>>>0]=E,l}function Qh(l,m,E,F){return Z?Bs(2,1,l,m,E,F):Xh(l,m,E,F)}function Xh(l,m,E,F){if(l>>>=0,m>>>=0,E>>>=0,F>>>=0,ee===void 0)return bt("Current environment does not support SharedArrayBuffer, pthreads are not available!"),6;var G=[];return Z&&G.length===0?Qh(l,m,E,F):(l={ic:E,Bb:l,Rb:F,nc:G},Z?(l.Nb="spawnThread",postMessage(l,G),0):Uh(l))}var Yh=typeof TextDecoder<"u"?new TextDecoder("utf8"):void 0,Jh=(l,m,E)=>{var F=(m>>>=0)+E;for(E=m;l[E]&&!(E>=F);)++E;if(16(G=(240&G)==224?(15&G)<<12|ge<<6|Xe:(7&G)<<18|ge<<12|Xe<<6|63&l[m++])?F+=String.fromCharCode(G):(G-=65536,F+=String.fromCharCode(55296|G>>10,56320|1023&G))}}else F+=String.fromCharCode(G)}return F},Qs=(l,m)=>(l>>>=0)?Jh(n(),l,m):"";function Zh(l,m,E){return Z?Bs(3,1,l,m,E):0}function em(l,m){if(Z)return Bs(4,1,l,m)}var oh=l=>{for(var m=0,E=0;E=F?m++:2047>=F?m+=2:55296<=F&&57343>=F?(m+=4,++E):m+=3}return m},tm=(l,m,E,F)=>{if(!(0>>=0;F=E+F-1;for(var ge=0;ge=Xe&&(Xe=65536+((1023&Xe)<<10)|1023&l.charCodeAt(++ge)),127>=Xe){if(E>=F)break;m[E++>>>0]=Xe}else{if(2047>=Xe){if(E+1>=F)break;m[E++>>>0]=192|Xe>>6}else{if(65535>=Xe){if(E+2>=F)break;m[E++>>>0]=224|Xe>>12}else{if(E+3>=F)break;m[E++>>>0]=240|Xe>>18,m[E++>>>0]=128|Xe>>12&63}m[E++>>>0]=128|Xe>>6&63}m[E++>>>0]=128|63&Xe}}return m[E>>>0]=0,E-G},Fa=(l,m,E)=>tm(l,n(),m,E);function sm(l,m){if(Z)return Bs(5,1,l,m)}function rm(l,m,E){if(Z)return Bs(6,1,l,m,E)}function nm(l,m,E){return Z?Bs(7,1,l,m,E):0}function im(l,m){if(Z)return Bs(8,1,l,m)}function om(l,m,E){if(Z)return Bs(9,1,l,m,E)}function am(l,m,E,F){if(Z)return Bs(10,1,l,m,E,F)}function lm(l,m,E,F){if(Z)return Bs(11,1,l,m,E,F)}function um(l,m,E,F){if(Z)return Bs(12,1,l,m,E,F)}function dm(l){if(Z)return Bs(13,1,l)}function cm(l,m){if(Z)return Bs(14,1,l,m)}function pm(l,m,E){if(Z)return Bs(15,1,l,m,E)}var hm,ti,vf=()=>{Zn("")},fn=l=>{for(var m="";n()[l>>>0];)m+=hm[n()[l++>>>0]];return m},ah={},lh={};function Sn(l,m,E={}){if(!("argPackAdvance"in m))throw new TypeError("registerType registeredInstance requires argPackAdvance");return function(F,G,ge={}){var Xe=G.name;if(!F)throw new ti(`type "${Xe}" must have a positive integer typeid pointer`);if(lh.hasOwnProperty(F)){if(ge.Tb)return;throw new ti(`Cannot register type '${Xe}' twice`)}lh[F]=G,ah.hasOwnProperty(F)&&(G=ah[F],delete ah[F],G.forEach(nt=>nt()))}(l,m,E)}var mm=(l,m,E)=>{switch(m){case 1:return E?F=>s()[F>>>0]:F=>n()[F>>>0];case 2:return E?F=>i()[F>>>1>>>0]:F=>a()[F>>>1>>>0];case 4:return E?F=>o()[F>>>2>>>0]:F=>u()[F>>>2>>>0];case 8:return E?F=>er[F>>>3]:F=>hn[F>>>3];default:throw new TypeError(`invalid integer width (${m}): ${l}`)}};function xf(l,m,E){E>>>=0,Sn(l>>>=0,{name:m=fn(m>>>0),fromWireType:F=>F,toWireType:function(F,G){if(typeof G!="bigint"&&typeof G!="number")throw G=G===null?"null":(F=typeof G)=="object"||F==="array"||F==="function"?G.toString():""+G,new TypeError(`Cannot convert "${G}" to ${this.name}`);return typeof G=="number"&&(G=BigInt(G)),G},argPackAdvance:si,readValueFromPointer:mm(m,E,m.indexOf("u")==-1),Eb:null})}var si=8;function Tf(l,m,E,F){Sn(l>>>=0,{name:m=fn(m>>>0),fromWireType:function(G){return!!G},toWireType:function(G,ge){return ge?E:F},argPackAdvance:si,readValueFromPointer:function(G){return this.fromWireType(n()[G>>>0])},Eb:null})}var uh=[],$n=[];function dh(l){9<(l>>>=0)&&--$n[l+1]==0&&($n[l]=void 0,uh.push(l))}var kr=l=>{if(!l)throw new ti("Cannot use deleted val. handle = "+l);return $n[l]},Br=l=>{switch(l){case void 0:return 2;case null:return 4;case!0:return 6;case!1:return 8;default:let m=uh.pop()||$n.length;return $n[m]=l,$n[m+1]=1,m}};function ch(l){return this.fromWireType(u()[l>>>2>>>0])}var Pf={name:"emscripten::val",fromWireType:l=>{var m=kr(l);return dh(l),m},toWireType:(l,m)=>Br(m),argPackAdvance:si,readValueFromPointer:ch,Eb:null};function Ef(l){return Sn(l>>>0,Pf)}var Cf=(l,m)=>{switch(m){case 4:return function(E){return this.fromWireType(p()[E>>>2>>>0])};case 8:return function(E){return this.fromWireType(h()[E>>>3>>>0])};default:throw new TypeError(`invalid float width (${m}): ${l}`)}};function kf(l,m,E){E>>>=0,Sn(l>>>=0,{name:m=fn(m>>>0),fromWireType:F=>F,toWireType:(F,G)=>G,argPackAdvance:si,readValueFromPointer:Cf(m,E),Eb:null})}function Sf(l,m,E,F,G){if(l>>>=0,E>>>=0,m=fn(m>>>0),G===-1&&(G=4294967295),G=nt=>nt,F===0){var ge=32-8*E;G=nt=>nt<>>ge}var Xe=m.includes("unsigned")?function(nt,wt){return wt>>>0}:function(nt,wt){return wt};Sn(l,{name:m,fromWireType:G,toWireType:Xe,argPackAdvance:si,readValueFromPointer:mm(m,E,F!==0),Eb:null})}function $f(l,m,E){function F(ge){var Xe=u()[ge>>>2>>>0];return ge=u()[ge+4>>>2>>>0],new G(s().buffer,ge,Xe)}var G=[Int8Array,Uint8Array,Int16Array,Uint16Array,Int32Array,Uint32Array,Float32Array,Float64Array,BigInt64Array,BigUint64Array][m];Sn(l>>>=0,{name:E=fn(E>>>0),fromWireType:F,argPackAdvance:si,readValueFromPointer:F},{Tb:!0})}function Af(l,m){l>>>=0;var E=(m=fn(m>>>0))==="std::string";Sn(l,{name:m,fromWireType:function(F){var G=u()[F>>>2>>>0],ge=F+4;if(E)for(var Xe=ge,nt=0;nt<=G;++nt){var wt=ge+nt;if(nt==G||n()[wt>>>0]==0){if(Xe=Qs(Xe,wt-Xe),Ct===void 0)var Ct=Xe;else Ct+="\0",Ct+=Xe;Xe=wt+1}}else{for(Ct=Array(G),nt=0;nt>>0]);Ct=Ct.join("")}return gn(F),Ct},toWireType:function(F,G){G instanceof ArrayBuffer&&(G=new Uint8Array(G));var ge=typeof G=="string";if(!(ge||G instanceof Uint8Array||G instanceof Uint8ClampedArray||G instanceof Int8Array))throw new ti("Cannot pass non-string to std::string");var Xe=E&&ge?oh(G):G.length,nt=jp(4+Xe+1),wt=nt+4;if(u()[nt>>>2>>>0]=Xe,E&&ge)Fa(G,wt,Xe+1);else if(ge)for(ge=0;ge>>0]=Ct}else for(ge=0;ge>>0]=G[ge];return F!==null&&F.push(gn,nt),nt},argPackAdvance:si,readValueFromPointer:ch,Eb(F){gn(F)}})}var fm=typeof TextDecoder<"u"?new TextDecoder("utf-16le"):void 0,If=(l,m)=>{for(var E=l>>1,F=E+m/2;!(E>=F)&&a()[E>>>0];)++E;if(32<(E<<=1)-l&&fm)return fm.decode(n().slice(l,E));for(E="",F=0;!(F>=m/2);++F){var G=i()[l+2*F>>>1>>>0];if(G==0)break;E+=String.fromCharCode(G)}return E},Of=(l,m,E)=>{if(E??(E=2147483647),2>E)return 0;var F=m;E=(E-=2)<2*l.length?E/2:l.length;for(var G=0;G>>1>>>0]=ge,m+=2}return i()[m>>>1>>>0]=0,m-F},Ff=l=>2*l.length,Df=(l,m)=>{for(var E=0,F="";!(E>=m/4);){var G=o()[l+4*E>>>2>>>0];if(G==0)break;++E,65536<=G?(G-=65536,F+=String.fromCharCode(55296|G>>10,56320|1023&G)):F+=String.fromCharCode(G)}return F},Lf=(l,m,E)=>{if(m>>>=0,E??(E=2147483647),4>E)return 0;var F=m;E=F+E-4;for(var G=0;G=ge&&(ge=65536+((1023&ge)<<10)|1023&l.charCodeAt(++G)),o()[m>>>2>>>0]=ge,(m+=4)+4>E)break}return o()[m>>>2>>>0]=0,m-F},zf=l=>{for(var m=0,E=0;E=F&&++E,m+=4}return m};function Bf(l,m,E){if(l>>>=0,m>>>=0,E=fn(E>>>=0),m===2)var F=If,G=Of,ge=Ff,Xe=nt=>a()[nt>>>1>>>0];else m===4&&(F=Df,G=Lf,ge=zf,Xe=nt=>u()[nt>>>2>>>0]);Sn(l,{name:E,fromWireType:nt=>{for(var wt,Ct=u()[nt>>>2>>>0],Wt=nt+4,hs=0;hs<=Ct;++hs){var vs=nt+4+hs*m;hs!=Ct&&Xe(vs)!=0||(Wt=F(Wt,vs-Wt),wt===void 0?wt=Wt:(wt+="\0",wt+=Wt),Wt=vs+m)}return gn(nt),wt},toWireType:(nt,wt)=>{if(typeof wt!="string")throw new ti(`Cannot pass non-string to C++ string type ${E}`);var Ct=ge(wt),Wt=jp(4+Ct+m);return u()[Wt>>>2>>>0]=Ct/m,G(wt,Wt+4,Ct+m),nt!==null&&nt.push(gn,Wt),Wt},argPackAdvance:si,readValueFromPointer:ch,Eb(nt){gn(nt)}})}function Rf(l,m){Sn(l>>>=0,{Ub:!0,name:m=fn(m>>>0),argPackAdvance:0,fromWireType:()=>{},toWireType:()=>{}})}var Nf=()=>1;function jf(l){wh(l>>>0,!V,1,!B,131072,!1),Kh()}var _m=l=>{if(!zs)try{if(l(),!(0>>=0,typeof Atomics.oc=="function"&&(Atomics.oc(o(),l>>>2,l).value.then(Ip),l+=128,Atomics.store(o(),l>>>2,1))}var Ip=()=>{var l=La();l&&(ph(l),_m(Km))};function Uf(l,m){(l>>>=0)==m>>>0?setTimeout(Ip):Z?postMessage({targetThread:l,cmd:"checkMailbox"}):(l=mn[l])&&l.postMessage({cmd:"checkMailbox"})}var hh=[];function Vf(l,m,E,F,G){for(m>>>=0,F/=2,hh.length=F,E=G>>>0>>>3,G=0;G>>0];return(m?th[m]:O_[l])(...hh)}function Wf(l){l>>>=0,Z?postMessage({cmd:"cleanupThread",thread:l}):Gh(mn[l])}function Gf(l){}var Op=(l,m)=>{var E=lh[l];if(E===void 0)throw l=Um(l),E=fn(l),gn(l),new ti(`${m} has unknown type ${E}`);return E},gm=(l,m,E)=>{var F=[];return l=l.toWireType(F,E),F.length&&(u()[m>>>2>>>0]=Br(F)),l};function Kf(l,m,E){return m>>>=0,E>>>=0,l=kr(l>>>0),m=Op(m,"emval::as"),gm(m,E,l)}function Hf(l,m){return m>>>=0,l=kr(l>>>0),(m=Op(m,"emval::as")).toWireType(null,l)}var Fp=l=>{try{l()}catch(m){Zn(m)}},ri=0,_n=null,wm=0,Dp=[],ym={},Mm={},qf=0,mh=null,Qf=[];function bm(l){return function(m){if(!zs){if(ri===0){var E=!1,F=!1;m((G=0)=>{if(!zs&&(wm=G,E=!0,F)){ri=2,Fp(()=>Ym(_n)),typeof Browser<"u"&&Browser.Lb.Sb&&Browser.Lb.resume(),G=!1;try{var ge=function(){var wt=o()[_n+8>>>2>>>0];return wt=Ut[Mm[wt]],--Bi,wt()}()}catch(wt){ge=wt,G=!0}var Xe=!1;if(!_n){var nt=mh;nt&&(mh=null,(G?nt.reject:nt.resolve)(ge),Xe=!0)}if(G&&!Xe)throw ge}}),F=!0,E||(ri=1,_n=function(){var G=jp(65548),ge=G+12;u()[G>>>2>>>0]=ge,u()[G+4>>>2>>>0]=ge+65536,ge=Dp[0];var Xe=ym[ge];return Xe===void 0&&(Xe=qf++,ym[ge]=Xe,Mm[Xe]=ge),ge=Xe,o()[G+8>>>2>>>0]=ge,G}(),typeof Browser<"u"&&Browser.Lb.Sb&&Browser.Lb.pause(),Fp(()=>Qm(_n)))}else ri===2?(ri=0,Fp(Jm),gn(_n),_n=null,Qf.forEach(_m)):Zn(`invalid state: ${ri}`);return wm}}(m=>{l().then(m)})}function Xf(l){return l>>>=0,bm(()=>(l=kr(l)).then(Br))}var Lp=[];function Yf(l,m,E,F){return E>>>=0,F>>>=0,(l=Lp[l>>>0])(null,m=kr(m>>>0),E,F)}var Jf={},zp=l=>{var m=Jf[l];return m===void 0?fn(l):m};function Zf(l,m,E,F,G){return E>>>=0,F>>>=0,G>>>=0,(l=Lp[l>>>0])(m=kr(m>>>0),m[E=zp(E)],F,G)}var vm=()=>typeof globalThis=="object"?globalThis:Function("return this")();function e_(l){return(l>>>=0)==0?Br(vm()):(l=zp(l),Br(vm()[l]))}var t_=l=>{var m=Lp.length;return Lp.push(l),m},s_=(l,m)=>{for(var E=Array(l),F=0;F>>2>>>0],"parameter "+F);return E},xm=(l,m)=>Object.defineProperty(m,"name",{value:l});function r_(l,m,E){var F=(m=s_(l,m>>>0)).shift();l--;var G=`return function (obj, func, destructorsRef, args) { +`,ge=0,Xe=[];E===0&&Xe.push("obj");for(var nt=["retType"],wt=[F],Ct=0;CtWt.name).join(", ")}) => ${F.name}>`,t_(xm(E,l))}function n_(l){return l=zp(l>>>0),Br(d[l])}function i_(l,m){return m>>>=0,l=kr(l>>>0),m=kr(m),Br(l[m])}function o_(l){9<(l>>>=0)&&($n[l+1]+=1)}function a_(){return Br([])}function l_(l){l=kr(l>>>0);for(var m=Array(l.length),E=0;E>>0))}function d_(){return Br({})}function c_(l){for(var m=kr(l>>>=0);m.length;){var E=m.pop();m.pop()(E)}dh(l)}function p_(l,m,E){m>>>=0,E>>>=0,l=kr(l>>>0),m=kr(m),E=kr(E),l[m]=E}function h_(l,m){return m>>>=0,l=(l=Op(l>>>0,"_emval_take_value")).readValueFromPointer(m),Br(l)}function m_(l,m){l=-9007199254740992>l||9007199254740992>>=0,l=new Date(1e3*l),o()[m>>>2>>>0]=l.getUTCSeconds(),o()[m+4>>>2>>>0]=l.getUTCMinutes(),o()[m+8>>>2>>>0]=l.getUTCHours(),o()[m+12>>>2>>>0]=l.getUTCDate(),o()[m+16>>>2>>>0]=l.getUTCMonth(),o()[m+20>>>2>>>0]=l.getUTCFullYear()-1900,o()[m+24>>>2>>>0]=l.getUTCDay(),l=(l.getTime()-Date.UTC(l.getUTCFullYear(),0,1,0,0,0,0))/864e5|0,o()[m+28>>>2>>>0]=l}var Da=l=>l%4==0&&(l%100!=0||l%400==0),Tm=[0,31,60,91,121,152,182,213,244,274,305,335],Pm=[0,31,59,90,120,151,181,212,243,273,304,334];function f_(l,m){l=-9007199254740992>l||9007199254740992>>=0,l=new Date(1e3*l),o()[m>>>2>>>0]=l.getSeconds(),o()[m+4>>>2>>>0]=l.getMinutes(),o()[m+8>>>2>>>0]=l.getHours(),o()[m+12>>>2>>>0]=l.getDate(),o()[m+16>>>2>>>0]=l.getMonth(),o()[m+20>>>2>>>0]=l.getFullYear()-1900,o()[m+24>>>2>>>0]=l.getDay();var E=(Da(l.getFullYear())?Tm:Pm)[l.getMonth()]+l.getDate()-1|0;o()[m+28>>>2>>>0]=E,o()[m+36>>>2>>>0]=-60*l.getTimezoneOffset(),E=new Date(l.getFullYear(),6,1).getTimezoneOffset();var F=new Date(l.getFullYear(),0,1).getTimezoneOffset();l=0|(E!=F&&l.getTimezoneOffset()==Math.min(F,E)),o()[m+32>>>2>>>0]=l}function __(l){l>>>=0;var m=new Date(o()[l+20>>>2>>>0]+1900,o()[l+16>>>2>>>0],o()[l+12>>>2>>>0],o()[l+8>>>2>>>0],o()[l+4>>>2>>>0],o()[l>>>2>>>0],0),E=o()[l+32>>>2>>>0],F=m.getTimezoneOffset(),G=new Date(m.getFullYear(),6,1).getTimezoneOffset(),ge=new Date(m.getFullYear(),0,1).getTimezoneOffset(),Xe=Math.min(ge,G);return 0>E?o()[l+32>>>2>>>0]=+(G!=ge&&Xe==F):0>>2>>>0]=m.getDay(),E=(Da(m.getFullYear())?Tm:Pm)[m.getMonth()]+m.getDate()-1|0,o()[l+28>>>2>>>0]=E,o()[l>>>2>>>0]=m.getSeconds(),o()[l+4>>>2>>>0]=m.getMinutes(),o()[l+8>>>2>>>0]=m.getHours(),o()[l+12>>>2>>>0]=m.getDate(),o()[l+16>>>2>>>0]=m.getMonth(),o()[l+20>>>2>>>0]=m.getYear(),l=m.getTime(),BigInt(isNaN(l)?-1:l/1e3)}function Em(l,m,E,F,G,ge,Xe){return Z?Bs(16,1,l,m,E,F,G,ge,Xe):-52}function Cm(l,m,E,F,G,ge){if(Z)return Bs(17,1,l,m,E,F,G,ge)}function g_(l,m,E,F){l>>>=0,m>>>=0,E>>>=0,F>>>=0;var G=new Date().getFullYear(),ge=new Date(G,0,1),Xe=new Date(G,6,1);G=ge.getTimezoneOffset();var nt=Xe.getTimezoneOffset(),wt=Math.max(G,nt);u()[l>>>2>>>0]=60*wt,o()[m>>>2>>>0]=+(G!=nt),ge=(l=Ct=>Ct.toLocaleTimeString(void 0,{hour12:!1,timeZoneName:"short"}).split(" ")[1])(ge),Xe=l(Xe),nt{fh.length=0;for(var E;E=n()[l++>>>0];){var F=E!=105;m+=(F&=E!=112)&&m%8?4:0,fh.push(E==112?u()[m>>>2>>>0]:E==106?er[m>>>3]:E==105?o()[m>>>2>>>0]:h()[m>>>3>>>0]),m+=F?8:4}return fh};function w_(l,m,E){return l>>>=0,m=km(m>>>0,E>>>0),th[l](...m)}function y_(l,m,E){return l>>>=0,m=km(m>>>0,E>>>0),th[l](...m)}var M_=()=>{},b_=()=>Date.now();function v_(l,m){return bt(Qs(l>>>0,m>>>0))}var Sm,x_=()=>{throw Bi+=1,"unwind"};function T_(){return 4294901760}Sm=()=>performance.timeOrigin+performance.now();var P_=()=>navigator.hardwareConcurrency;function E_(){return Zn("Cannot use emscripten_pc_get_function without -sUSE_OFFSET_CONVERTER"),0}function C_(l){l>>>=0;var m=n().length;if(l<=m||4294901760=E;E*=2){var F=m*(1+.2/E);F=Math.min(F,l+100663296);var G=Math;F=Math.max(l,F);e:{G=(G.min.call(G,4294901760,F+(65536-F%65536)%65536)-Ft.buffer.byteLength+65535)/65536;try{Ft.grow(G),Ps();var ge=1;break e}catch{}ge=void 0}if(ge)return!0}return!1}var Bp=()=>(Zn("Cannot use convertFrameToPC (needed by __builtin_return_address) without -sUSE_OFFSET_CONVERTER"),0),zc={},$m=l=>{l.forEach(m=>{Bp()})};function k_(){var l=Error().stack.toString().split(` +`);return l[0]=="Error"&&l.shift(),$m(l),zc.Qb=Bp(),zc.fc=l,zc.Qb}function S_(l,m,E){if(l>>>=0,m>>>=0,zc.Qb==l)var F=zc.fc;else(F=Error().stack.toString().split(` +`))[0]=="Error"&&F.shift(),$m(F);for(var G=3;F[G]&&Bp()!=l;)++G;for(l=0;l>>2>>>0]=Bp();return l}var _h,gh={},Am=()=>{if(!_h){var l,m={USER:"web_user",LOGNAME:"web_user",PATH:"/",PWD:"/",HOME:"/home/web_user",LANG:(typeof navigator=="object"&&navigator.languages&&navigator.languages[0]||"C").replace("-","_")+".UTF-8",_:Oe};for(l in gh)gh[l]===void 0?delete m[l]:m[l]=gh[l];var E=[];for(l in m)E.push(`${l}=${m[l]}`);_h=E}return _h};function Im(l,m){if(Z)return Bs(18,1,l,m);l>>>=0,m>>>=0;var E=0;return Am().forEach((F,G)=>{var ge=m+E;for(G=u()[l+4*G>>>2>>>0]=ge,ge=0;ge>>0]=F.charCodeAt(ge);s()[G>>>0]=0,E+=F.length+1}),0}function Om(l,m){if(Z)return Bs(19,1,l,m);l>>>=0,m>>>=0;var E=Am();u()[l>>>2>>>0]=E.length;var F=0;return E.forEach(G=>F+=G.length+1),u()[m>>>2>>>0]=F,0}function Fm(l){return Z?Bs(20,1,l):52}function Dm(l,m,E,F){return Z?Bs(21,1,l,m,E,F):52}function Lm(l,m,E,F){return Z?Bs(22,1,l,m,E,F):70}var $_=[null,[],[]];function zm(l,m,E,F){if(Z)return Bs(23,1,l,m,E,F);m>>>=0,E>>>=0,F>>>=0;for(var G=0,ge=0;ge>>2>>>0],nt=u()[m+4>>>2>>>0];m+=8;for(var wt=0;wt>>0],Wt=$_[l];Ct===0||Ct===10?((l===1?Zt:bt)(Jh(Wt,0)),Wt.length=0):Wt.push(Ct)}G+=nt}return u()[F>>>2>>>0]=G,0}var Bm=[31,29,31,30,31,30,31,31,30,31,30,31],Rm=[31,28,31,30,31,30,31,31,30,31,30,31],A_=(l,m)=>{s().set(l,m>>>0)};function Nm(l,m,E,F){function G(He,ls,js){for(He=typeof He=="number"?He.toString():He||"";He.lengthef?-1:0Ni-He.getDate())){He.setDate(He.getDate()+ls);break}ls-=Ni-He.getDate()+1,He.setDate(1),11>js?He.setMonth(js+1):(He.setMonth(0),He.setFullYear(He.getFullYear()+1))}return js=new Date(He.getFullYear()+1,0,4),ls=nt(new Date(He.getFullYear(),0,4)),js=nt(js),0>=Xe(ls,He)?0>=Xe(js,He)?He.getFullYear()+1:He.getFullYear():He.getFullYear()-1}l>>>=0,m>>>=0,E>>>=0,F>>>=0;var Ct=u()[F+40>>>2>>>0];for(var Wt in F={lc:o()[F>>>2>>>0],kc:o()[F+4>>>2>>>0],Ib:o()[F+8>>>2>>>0],Mb:o()[F+12>>>2>>>0],Jb:o()[F+16>>>2>>>0],Db:o()[F+20>>>2>>>0],vb:o()[F+24>>>2>>>0],Cb:o()[F+28>>>2>>>0],sc:o()[F+32>>>2>>>0],jc:o()[F+36>>>2>>>0],mc:Ct?Qs(Ct):""},E=Qs(E),Ct={"%c":"%a %b %d %H:%M:%S %Y","%D":"%m/%d/%y","%F":"%Y-%m-%d","%h":"%b","%r":"%I:%M:%S %p","%R":"%H:%M","%T":"%H:%M:%S","%x":"%m/%d/%y","%X":"%H:%M:%S","%Ec":"%c","%EC":"%C","%Ex":"%m/%d/%y","%EX":"%H:%M:%S","%Ey":"%y","%EY":"%Y","%Od":"%d","%Oe":"%e","%OH":"%H","%OI":"%I","%Om":"%m","%OM":"%M","%OS":"%S","%Ou":"%u","%OU":"%U","%OV":"%V","%Ow":"%w","%OW":"%W","%Oy":"%y"})E=E.replace(new RegExp(Wt,"g"),Ct[Wt]);var hs="Sunday Monday Tuesday Wednesday Thursday Friday Saturday".split(" "),vs="January February March April May June July August September October November December".split(" ");for(Wt in Ct={"%a":He=>hs[He.vb].substring(0,3),"%A":He=>hs[He.vb],"%b":He=>vs[He.Jb].substring(0,3),"%B":He=>vs[He.Jb],"%C":He=>ge((He.Db+1900)/100|0,2),"%d":He=>ge(He.Mb,2),"%e":He=>G(He.Mb,2," "),"%g":He=>wt(He).toString().substring(2),"%G":wt,"%H":He=>ge(He.Ib,2),"%I":He=>((He=He.Ib)==0?He=12:12{for(var ls=0,js=0;js<=He.Jb-1;ls+=(Da(He.Db+1900)?Bm:Rm)[js++]);return ge(He.Mb+ls,3)},"%m":He=>ge(He.Jb+1,2),"%M":He=>ge(He.kc,2),"%n":()=>` +`,"%p":He=>0<=He.Ib&&12>He.Ib?"AM":"PM","%S":He=>ge(He.lc,2),"%t":()=>" ","%u":He=>He.vb||7,"%U":He=>ge(Math.floor((He.Cb+7-He.vb)/7),2),"%V":He=>{var ls=Math.floor((He.Cb+7-(He.vb+6)%7)/7);if(2>=(He.vb+371-He.Cb-2)%7&&ls++,ls)ls==53&&((js=(He.vb+371-He.Cb)%7)==4||js==3&&Da(He.Db)||(ls=1));else{ls=52;var js=(He.vb+7-He.Cb-1)%7;(js==4||js==5&&Da(He.Db%400-1))&&ls++}return ge(ls,2)},"%w":He=>He.vb,"%W":He=>ge(Math.floor((He.Cb+7-(He.vb+6)%7)/7),2),"%y":He=>(He.Db+1900).toString().substring(2),"%Y":He=>He.Db+1900,"%z":He=>{var ls=0<=(He=He.jc);return He=Math.abs(He)/60,(ls?"+":"-")+("0000"+(He/60*100+He%60)).slice(-4)},"%Z":He=>He.mc,"%%":()=>"%"},E=E.replace(/%%/g,"\0\0"),Ct)E.includes(Wt)&&(E=E.replace(new RegExp(Wt,"g"),Ct[Wt](F)));return Wt=function(He){var ls=Array(oh(He)+1);return tm(He,ls,0,ls.length),ls}(E=E.replace(/\0\0/g,"%")),Wt.length>m?0:(A_(Wt,l),Wt.length-1)}function I_(l,m,E,F){return Nm(l>>>0,m>>>0,E>>>0,F>>>0)}Z||function(){for(var l=d.numThreads-1;l--;)qh();Jn.unshift(()=>{Dc++,function(m){Z?m():Promise.all(ei.map(Hh)).then(m)}(()=>Lh())})}();for(var jm=Array(256),Rp=0;256>Rp;++Rp)jm[Rp]=String.fromCharCode(Rp);hm=jm,ti=d.BindingError=class extends Error{constructor(l){super(l),this.name="BindingError"}},d.InternalError=class extends Error{constructor(l){super(l),this.name="InternalError"}},$n.push(0,1,void 0,1,null,1,!0,1,!1,1),d.count_emval_handles=()=>$n.length/2-5-uh.length;var O_=[nh,Vh,Qh,Zh,em,sm,rm,nm,im,om,am,lm,um,dm,cm,pm,Em,Cm,Im,Om,Fm,Dm,Lm,zm],Ut=function(){function l(E,F){return Ut=E.exports,Ut=function(){var G=Ut,ge={};for(let[Xe,nt]of Object.entries(G))ge[Xe]=typeof nt=="function"?(...wt)=>{Dp.push(Xe);try{return nt(...wt)}finally{zs||(Dp.pop(),_n&&ri===1&&Dp.length===0&&(ri=0,Bi+=1,Fp(Xm),typeof Fibers<"u"&&Fibers.tc()))}}:nt;return ge}(),Ut=function(){var G=Ut,ge=nt=>wt=>nt(wt)>>>0,Xe=nt=>()=>nt()>>>0;return(G=Object.assign({},G)).Da=ge(G.Da),G.gb=Xe(G.gb),G.ib=ge(G.ib),G.emscripten_main_runtime_thread_id=Xe(G.emscripten_main_runtime_thread_id),G.tb=ge(G.tb),G.ub=Xe(G.ub),G}(),Wh.push(Ut.jb),Fc.unshift(Ut.Ca),bs=F,Lh(),Ut}var m=jh();if(Dc++,d.instantiateWasm)try{return d.instantiateWasm(m,l)}catch(E){bt(`Module.instantiateWasm callback failed with error: ${E}`),C(E)}return eh||(eh=d.locateFile?zh("ort-wasm-simd-threaded.jsep.wasm")?"ort-wasm-simd-threaded.jsep.wasm":d.locateFile?d.locateFile("ort-wasm-simd-threaded.jsep.wasm",Ye):Ye+"ort-wasm-simd-threaded.jsep.wasm":new URL(r("./node_modules/onnxruntime-web/dist/ort-wasm-simd-threaded.jsep.wasm"),r.b).href),function(E,F){var G=eh;return at||typeof WebAssembly.instantiateStreaming!="function"||zh(G)||Bh(G)||typeof fetch!="function"?Nh(G,E,F):fetch(G,{credentials:"same-origin"}).then(ge=>WebAssembly.instantiateStreaming(ge,E).then(F,function(Xe){return bt(`wasm streaming compile failed: ${Xe}`),bt("falling back to ArrayBuffer instantiation"),Nh(G,E,F)}))}(m,function(E){l(E.instance,E.module)}).catch(C),{}}(),Um=l=>(Um=Ut.Da)(l),Vm=()=>(Vm=Ut.Ea)();d._OrtInit=(l,m)=>(d._OrtInit=Ut.Fa)(l,m),d._OrtGetLastError=(l,m)=>(d._OrtGetLastError=Ut.Ga)(l,m),d._OrtCreateSessionOptions=(l,m,E,F,G,ge,Xe,nt,wt,Ct)=>(d._OrtCreateSessionOptions=Ut.Ha)(l,m,E,F,G,ge,Xe,nt,wt,Ct),d._OrtAppendExecutionProvider=(l,m)=>(d._OrtAppendExecutionProvider=Ut.Ia)(l,m),d._OrtAddFreeDimensionOverride=(l,m,E)=>(d._OrtAddFreeDimensionOverride=Ut.Ja)(l,m,E),d._OrtAddSessionConfigEntry=(l,m,E)=>(d._OrtAddSessionConfigEntry=Ut.Ka)(l,m,E),d._OrtReleaseSessionOptions=l=>(d._OrtReleaseSessionOptions=Ut.La)(l),d._OrtCreateSession=(l,m,E)=>(d._OrtCreateSession=Ut.Ma)(l,m,E),d._OrtReleaseSession=l=>(d._OrtReleaseSession=Ut.Na)(l),d._OrtGetInputOutputCount=(l,m,E)=>(d._OrtGetInputOutputCount=Ut.Oa)(l,m,E),d._OrtGetInputName=(l,m)=>(d._OrtGetInputName=Ut.Pa)(l,m),d._OrtGetOutputName=(l,m)=>(d._OrtGetOutputName=Ut.Qa)(l,m),d._OrtFree=l=>(d._OrtFree=Ut.Ra)(l),d._OrtCreateTensor=(l,m,E,F,G,ge)=>(d._OrtCreateTensor=Ut.Sa)(l,m,E,F,G,ge),d._OrtGetTensorData=(l,m,E,F,G)=>(d._OrtGetTensorData=Ut.Ta)(l,m,E,F,G),d._OrtReleaseTensor=l=>(d._OrtReleaseTensor=Ut.Ua)(l),d._OrtCreateRunOptions=(l,m,E,F)=>(d._OrtCreateRunOptions=Ut.Va)(l,m,E,F),d._OrtAddRunConfigEntry=(l,m,E)=>(d._OrtAddRunConfigEntry=Ut.Wa)(l,m,E),d._OrtReleaseRunOptions=l=>(d._OrtReleaseRunOptions=Ut.Xa)(l),d._OrtCreateBinding=l=>(d._OrtCreateBinding=Ut.Ya)(l),d._OrtBindInput=(l,m,E)=>(d._OrtBindInput=Ut.Za)(l,m,E),d._OrtBindOutput=(l,m,E,F)=>(d._OrtBindOutput=Ut._a)(l,m,E,F),d._OrtClearBoundOutputs=l=>(d._OrtClearBoundOutputs=Ut.$a)(l),d._OrtReleaseBinding=l=>(d._OrtReleaseBinding=Ut.ab)(l),d._OrtRunWithBinding=(l,m,E,F,G)=>(d._OrtRunWithBinding=Ut.bb)(l,m,E,F,G),d._OrtRun=(l,m,E,F,G,ge,Xe,nt)=>(d._OrtRun=Ut.cb)(l,m,E,F,G,ge,Xe,nt),d._OrtEndProfiling=l=>(d._OrtEndProfiling=Ut.db)(l),d._JsepOutput=(l,m,E)=>(d._JsepOutput=Ut.eb)(l,m,E),d._JsepGetNodeName=l=>(d._JsepGetNodeName=Ut.fb)(l);var Np,La=()=>(La=Ut.gb)(),gn=d._free=l=>(gn=d._free=Ut.hb)(l),jp=d._malloc=l=>(jp=d._malloc=Ut.ib)(l),wh=(l,m,E,F,G,ge)=>(wh=Ut.lb)(l,m,E,F,G,ge),Wm=()=>(Wm=Ut.mb)(),Gm=(l,m,E,F,G)=>(Gm=Ut.nb)(l,m,E,F,G),yh=l=>(yh=Ut.ob)(l),Up=l=>(Up=Ut.pb)(l),Km=()=>(Km=Ut.qb)(),Hm=(l,m)=>(Hm=Ut.rb)(l,m),Vp=l=>(Vp=Ut.sb)(l),Mh=l=>(Mh=Ut.tb)(l),bh=()=>(bh=Ut.ub)(),qm=d.dynCall_ii=(l,m)=>(qm=d.dynCall_ii=Ut.wb)(l,m),Qm=l=>(Qm=Ut.xb)(l),Xm=()=>(Xm=Ut.yb)(),Ym=l=>(Ym=Ut.zb)(l),Jm=()=>(Jm=Ut.Ab)();function Zm(){0bh(),d.stackRestore=l=>Vp(l),d.stackAlloc=l=>Mh(l),d.setValue=function(l,m,E="i8"){switch(E.endsWith("*")&&(E="*"),E){case"i1":case"i8":s()[l>>>0]=m;break;case"i16":i()[l>>>1>>>0]=m;break;case"i32":o()[l>>>2>>>0]=m;break;case"i64":er[l>>>3]=BigInt(m);break;case"float":p()[l>>>2>>>0]=m;break;case"double":h()[l>>>3>>>0]=m;break;case"*":u()[l>>>2>>>0]=m;break;default:Zn(`invalid type for setValue: ${E}`)}},d.getValue=function(l,m="i8"){switch(m.endsWith("*")&&(m="*"),m){case"i1":case"i8":return s()[l>>>0];case"i16":return i()[l>>>1>>>0];case"i32":return o()[l>>>2>>>0];case"i64":return er[l>>>3];case"float":return p()[l>>>2>>>0];case"double":return h()[l>>>3>>>0];case"*":return u()[l>>>2>>>0];default:Zn(`invalid type for getValue: ${m}`)}},d.UTF8ToString=Qs,d.stringToUTF8=Fa,d.lengthBytesUTF8=oh,Lc=function l(){Np||Zm(),Np||(Lc=l)},Zm(),d.PTR_SIZE=4,z}),Ds=Cs,((e=globalThis.self)==null?void 0:e.name)==="em-pthread"&&Cs()}),Sr,Yr,Us,Pr,Nt,Jr,$r,Ar,Zr,cr,en,Ir,Rr,Nr=g(()=>{Je(),Sr=typeof location>"u"?void 0:location.origin,Yr=()=>{var e;return(e=self.location.href)!=null&&e.startsWith("file:")?new URL(new URL(r("./node_modules/onnxruntime-web/dist/ort.bundle.min.mjs?46eb"),r.b).href,Sr).href:self.location.href},Us=Yr(),Pr=()=>{if(Us&&!Us.startsWith("blob:"))return Us.substring(0,Us.lastIndexOf("/")+1)},Nt=(e,t)=>{try{let s=t??Us;return(s?new URL(e,s):new URL(e)).origin===Sr}catch{return!1}},Jr=(e,t)=>{let s=t??Us;try{return(s?new URL(e,s):new URL(e)).href}catch{return}},$r=(e,t)=>`${t??"./"}${e}`,Ar=async e=>{let t=await(await fetch(e,{credentials:"same-origin"})).blob();return URL.createObjectURL(t)},Zr=async e=>(await import(e)).default,cr=(At(),y(ut)).default,en=async()=>{if(!Us)throw new Error("Failed to load proxy worker: cannot determine the script source URL.");if(Nt(Us))return[void 0,cr()];let e=await Ar(Us);return[e,cr(e)]},Ir=(sr(),y(is)).default,Rr=async(e,t,s)=>{if(!e&&!t&&Ir&&Us&&Nt(Us))return[void 0,Ir];{let n="ort-wasm-simd-threaded.jsep.mjs",i=e??Jr(n,t),a=s&&i&&!Nt(i,t),o=a?await Ar(i):i??$r(n,t);return[a?o:void 0,await Zr(o)]}}}),ar,it,Tt,Dt,Vs,jr,Or,Ms,lr=g(()=>{Nr(),it=!1,Tt=!1,Dt=!1,Vs=()=>{if(typeof SharedArrayBuffer>"u")return!1;try{return typeof MessageChannel<"u"&&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{return!1}},jr=()=>{try{return WebAssembly.validate(new Uint8Array([0,97,115,109,1,0,0,0,1,4,1,96,0,0,3,2,1,0,10,30,1,28,0,65,0,253,15,253,12,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,253,186,1,26,11]))}catch{return!1}},Or=async e=>{if(it)return Promise.resolve();if(Tt)throw new Error("multiple calls to 'initializeWebAssembly()' detected.");if(Dt)throw new Error("previous call to 'initializeWebAssembly()' failed.");Tt=!0;let t=e.initTimeout,s=e.numThreads;if(!jr())throw new Error("WebAssembly SIMD is not supported in the current environment.");let n=Vs();s>1&&!n&&(typeof self<"u"&&!self.crossOriginIsolated&&console.warn("env.wasm.numThreads is set to "+s+", but this will not work unless you enable crossOriginIsolated mode. See https://web.dev/cross-origin-isolation-guide/ for more info."),console.warn("WebAssembly multi-threading is not supported in the current environment. Falling back to single-threading."),e.numThreads=s=1);let i=e.wasmPaths,a=typeof i=="string"?i:void 0,o=i==null?void 0:i.mjs,u=(o==null?void 0:o.href)??o,p=i==null?void 0:i.wasm,h=(p==null?void 0:p.href)??p,k=e.wasmBinary,[C,d]=await Rr(u,a,s>1),z=!1,B=[];if(t>0&&B.push(new Promise(V=>{setTimeout(()=>{z=!0,V()},t)})),B.push(new Promise((V,Z)=>{let ee={numThreads:s};if(k)ee.wasmBinary=k;else if(h||a)ee.locateFile=Y=>h??a+Y;else if(u&&u.indexOf("blob:")!==0)ee.locateFile=Y=>new URL(Y,u).href;else if(C){let Y=Pr();Y&&(ee.locateFile=he=>Y+he)}d(ee).then(Y=>{Tt=!1,it=!0,ar=Y,V(),C&&URL.revokeObjectURL(C)},Y=>{Tt=!1,Dt=!0,Z(Y)})})),await Promise.race(B),z)throw new Error(`WebAssembly backend initializing failed due to timeout: ${t}ms`)},Ms=()=>{if(it&&ar)return ar;throw new Error("WebAssembly is not initialized yet.")}}),Os,Er,es,wn=g(()=>{lr(),Os=(e,t)=>{let s=Ms(),n=s.lengthBytesUTF8(e)+1,i=s._malloc(n);return s.stringToUTF8(e,i,n),t.push(i),i},Er=(e,t,s,n)=>{if(typeof e=="object"&&e!==null){if(s.has(e))throw new Error("Circular reference in options");s.add(e)}Object.entries(e).forEach(([i,a])=>{let o=t?t+i:i;if(typeof a=="object")Er(a,o+".",s,n);else if(typeof a=="string"||typeof a=="number")n(o,a.toString());else if(typeof a=="boolean")n(o,a?"1":"0");else throw new Error(`Can't handle extra config type: ${typeof a}`)})},es=e=>{let t=Ms(),s=t.stackSave();try{let n=t.PTR_SIZE,i=t.stackAlloc(2*n);t._OrtGetLastError(i,i+n);let a=Number(t.getValue(i,n===4?"i32":"i64")),o=t.getValue(i+n,"*"),u=o?t.UTF8ToString(o):"";throw new Error(`${e} ERROR_CODE: ${a}, ERROR_MESSAGE: ${u}`)}finally{t.stackRestore(s)}}}),Ur,ii=g(()=>{lr(),wn(),Ur=e=>{let t=Ms(),s=0,n=[],i=e||{};try{if((e==null?void 0:e.logSeverityLevel)===void 0)i.logSeverityLevel=2;else if(typeof e.logSeverityLevel!="number"||!Number.isInteger(e.logSeverityLevel)||e.logSeverityLevel<0||e.logSeverityLevel>4)throw new Error(`log serverity level is not valid: ${e.logSeverityLevel}`);if((e==null?void 0:e.logVerbosityLevel)===void 0)i.logVerbosityLevel=0;else if(typeof e.logVerbosityLevel!="number"||!Number.isInteger(e.logVerbosityLevel))throw new Error(`log verbosity level is not valid: ${e.logVerbosityLevel}`);(e==null?void 0:e.terminate)===void 0&&(i.terminate=!1);let a=0;return(e==null?void 0:e.tag)!==void 0&&(a=Os(e.tag,n)),s=t._OrtCreateRunOptions(i.logSeverityLevel,i.logVerbosityLevel,!!i.terminate,a),s===0&&es("Can't create run options."),(e==null?void 0:e.extra)!==void 0&&Er(e.extra,"",new WeakSet,(o,u)=>{let p=Os(o,n),h=Os(u,n);t._OrtAddRunConfigEntry(s,p,h)!==0&&es(`Can't set a run config entry: ${o} - ${u}.`)}),[s,n]}catch(a){throw s!==0&&t._OrtReleaseRunOptions(s),n.forEach(o=>t._free(o)),a}}}),An,In,On,Vr,Fn,oi=g(()=>{lr(),wn(),An=e=>{switch(e){case"disabled":return 0;case"basic":return 1;case"extended":return 2;case"all":return 99;default:throw new Error(`unsupported graph optimization level: ${e}`)}},In=e=>{switch(e){case"sequential":return 0;case"parallel":return 1;default:throw new Error(`unsupported execution mode: ${e}`)}},On=e=>{e.extra||(e.extra={}),e.extra.session||(e.extra.session={});let t=e.extra.session;t.use_ort_model_bytes_directly||(t.use_ort_model_bytes_directly="1"),e.executionProviders&&e.executionProviders.some(s=>(typeof s=="string"?s:s.name)==="webgpu")&&(e.enableMemPattern=!1)},Vr=(e,t,s)=>{for(let n of t){let i=typeof n=="string"?n:n.name;switch(i){case"webnn":if(i="WEBNN",typeof n!="string"){let o=n==null?void 0:n.deviceType;if(o){let u=Os("deviceType",s),p=Os(o,s);Ms()._OrtAddSessionConfigEntry(e,u,p)!==0&&es(`Can't set a session config entry: 'deviceType' - ${o}.`)}}break;case"webgpu":if(i="JS",typeof n!="string"){let o=n;if(o!=null&&o.preferredLayout){if(o.preferredLayout!=="NCHW"&&o.preferredLayout!=="NHWC")throw new Error(`preferredLayout must be either 'NCHW' or 'NHWC': ${o.preferredLayout}`);let u=Os("preferredLayout",s),p=Os(o.preferredLayout,s);Ms()._OrtAddSessionConfigEntry(e,u,p)!==0&&es(`Can't set a session config entry: 'preferredLayout' - ${o.preferredLayout}.`)}}break;case"wasm":case"cpu":continue;default:throw new Error(`not supported execution provider: ${i}`)}let a=Os(i,s);Ms()._OrtAppendExecutionProvider(e,a)!==0&&es(`Can't append execution provider: ${i}.`)}},Fn=e=>{let t=Ms(),s=0,n=[],i=e||{};On(i);try{let a=An(i.graphOptimizationLevel??"all"),o=In(i.executionMode??"sequential"),u=typeof i.logId=="string"?Os(i.logId,n):0,p=i.logSeverityLevel??2;if(!Number.isInteger(p)||p<0||p>4)throw new Error(`log serverity level is not valid: ${p}`);let h=i.logVerbosityLevel??0;if(!Number.isInteger(h)||h<0||h>4)throw new Error(`log verbosity level is not valid: ${h}`);let k=typeof i.optimizedModelFilePath=="string"?Os(i.optimizedModelFilePath,n):0;if(s=t._OrtCreateSessionOptions(a,!!i.enableCpuMemArena,!!i.enableMemPattern,o,!!i.enableProfiling,0,u,p,h,k),s===0&&es("Can't create session options."),i.executionProviders&&Vr(s,i.executionProviders,n),i.enableGraphCapture!==void 0){if(typeof i.enableGraphCapture!="boolean")throw new Error(`enableGraphCapture must be a boolean value: ${i.enableGraphCapture}`);let C=Os("enableGraphCapture",n),d=Os(i.enableGraphCapture.toString(),n);t._OrtAddSessionConfigEntry(s,C,d)!==0&&es(`Can't set a session config entry: 'enableGraphCapture' - ${i.enableGraphCapture}.`)}if(i.freeDimensionOverrides)for(let[C,d]of Object.entries(i.freeDimensionOverrides)){if(typeof C!="string")throw new Error(`free dimension override name must be a string: ${C}`);if(typeof d!="number"||!Number.isInteger(d)||d<0)throw new Error(`free dimension override value must be a non-negative integer: ${d}`);let z=Os(C,n);t._OrtAddFreeDimensionOverride(s,z,d)!==0&&es(`Can't set a free dimension override: ${C} - ${d}.`)}return i.extra!==void 0&&Er(i.extra,"",new WeakSet,(C,d)=>{let z=Os(C,n),B=Os(d,n);t._OrtAddSessionConfigEntry(s,z,B)!==0&&es(`Can't set a session config entry: ${C} - ${d}.`)}),[s,n]}catch(a){throw s!==0&&t._OrtReleaseSessionOptions(s)!==0&&es("Can't release session options."),n.forEach(o=>t._free(o)),a}}}),Wr,fr,ur,yn,tn,Mn,bn,vn,zt=g(()=>{Wr=e=>{switch(e){case"int8":return 3;case"uint8":return 2;case"bool":return 9;case"int16":return 5;case"uint16":return 4;case"int32":return 6;case"uint32":return 12;case"float16":return 10;case"float32":return 1;case"float64":return 11;case"string":return 8;case"int64":return 7;case"uint64":return 13;case"int4":return 22;case"uint4":return 21;default:throw new Error(`unsupported data type: ${e}`)}},fr=e=>{switch(e){case 3:return"int8";case 2:return"uint8";case 9:return"bool";case 5:return"int16";case 4:return"uint16";case 6:return"int32";case 12:return"uint32";case 10:return"float16";case 1:return"float32";case 11:return"float64";case 8:return"string";case 7:return"int64";case 13:return"uint64";case 22:return"int4";case 21:return"uint4";default:throw new Error(`unsupported data type: ${e}`)}},ur=(e,t)=>{let s=[-1,4,1,1,2,2,4,8,-1,1,2,8,4,8,-1,-1,-1,-1,-1,-1,-1,.5,.5][e],n=typeof t=="number"?t:t.reduce((i,a)=>i*a,1);return s>0?Math.ceil(n*s):void 0},yn=e=>{switch(e){case"float16":return typeof Float16Array<"u"&&Float16Array.from?Float16Array:Uint16Array;case"float32":return Float32Array;case"uint8":return Uint8Array;case"int8":return Int8Array;case"uint16":return Uint16Array;case"int16":return Int16Array;case"int32":return Int32Array;case"bool":return Uint8Array;case"float64":return Float64Array;case"uint32":return Uint32Array;case"int64":return BigInt64Array;case"uint64":return BigUint64Array;default:throw new Error(`unsupported type: ${e}`)}},tn=e=>{switch(e){case"verbose":return 0;case"info":return 1;case"warning":return 2;case"error":return 3;case"fatal":return 4;default:throw new Error(`unsupported logging level: ${e}`)}},Mn=e=>e==="float32"||e==="float16"||e==="int32"||e==="int64"||e==="uint32"||e==="uint8"||e==="bool"||e==="uint4"||e==="int4",bn=e=>e==="float32"||e==="float16"||e==="int32"||e==="int64"||e==="uint32"||e==="uint64"||e==="int8"||e==="uint8"||e==="bool"||e==="uint4"||e==="int4",vn=e=>{switch(e){case"none":return 0;case"cpu":return 1;case"cpu-pinned":return 2;case"texture":return 3;case"gpu-buffer":return 4;case"ml-tensor":return 5;default:throw new Error(`unsupported data location: ${e}`)}}}),xn,Dn=g(()=>{Je(),xn=async e=>{if(typeof e=="string"){let t=await fetch(e);if(!t.ok)throw new Error(`failed to load external data file: ${e}`);let s=t.headers.get("Content-Length"),n=s?parseInt(s,10):0;if(n<1073741824)return new Uint8Array(await t.arrayBuffer());{if(!t.body)throw new Error(`failed to load external data file: ${e}, no response body.`);let i=t.body.getReader(),a;try{a=new ArrayBuffer(n)}catch(u){if(u instanceof RangeError){let p=Math.ceil(n/65536);a=new WebAssembly.Memory({initial:p,maximum:p}).buffer}else throw u}let o=0;for(;;){let{done:u,value:p}=await i.read();if(u)break;let h=p.byteLength;new Uint8Array(a,o,h).set(p),o+=h}return new Uint8Array(a,0,n)}}else return e instanceof Blob?new Uint8Array(await e.arrayBuffer()):e instanceof Uint8Array?e:new Uint8Array(e)}}),Ln,zn,Gr,Bn,Tn,Rn,as,Pe=g(()=>{zt(),Ln=["V","I","W","E","F"],zn=(e,t)=>{console.log(`[${Ln[e]},${new Date().toISOString()}]${t}`)},Tn=(e,t)=>{Gr=e,Bn=t},Rn=(e,t)=>{let s=tn(e),n=tn(Gr);s>=n&&zn(s,typeof t=="function"?t():t)},as=(...e)=>{Bn&&Rn(...e)}}),P,Q=g(()=>{zt(),P=(e,t)=>new(yn(t))(e)}),ue=g(()=>{}),be,Se,Qe,pt,gt,ft,xt,Kt,ms,us=g(()=>{Pe(),ue(),be=new Map([[64,250],[128,200],[256,200],[512,200],[2048,230],[4096,200],[8192,50],[16384,50],[32768,50],[65536,50],[131072,50],[262144,50],[524288,50],[1048576,50],[2097152,30],[4194304,20],[8388608,10],[12582912,10],[16777216,10],[26214400,15],[33554432,22],[44236800,2],[58982400,6],[67108864,6],[134217728,6],[167772160,6]]),Se=[],Qe=e=>Math.ceil(Number(e)/16)*16,pt=e=>{for(let t=0;tgt++,xt=async(e,t,s,n)=>{let i=Qe(s),a=e.device.createBuffer({size:i,usage:GPUBufferUsage.COPY_DST|GPUBufferUsage.MAP_READ});try{let o=e.getCommandEncoder();e.endComputePass(),o.copyBufferToBuffer(t,0,a,0,i),e.flush(),await a.mapAsync(GPUMapMode.READ);let u=a.getMappedRange();if(n){let p=n();return p.set(new Uint8Array(u,0,s)),p}else return new Uint8Array(u.slice(0,s))}finally{a.destroy()}},Kt=class{constructor(e){this.backend=e,this.storageCache=new Map,this.freeBuffers=new Map,this.freeUniformBuffers=new Map,this.buffersPending=[],this.capturedPendingBuffers=new Map;for(let[t]of be)Se.push(t),this.freeBuffers.set(t,[]),this.freeUniformBuffers.set(t,[]);this.sessionCount=0}upload(e,t){let s=t.buffer,n=t.byteOffset,i=t.byteLength,a=Qe(i),o=this.storageCache.get(e);if(!o)throw new Error("gpu data for uploading does not exist");if(Number(o.originalSize)!==i)throw new Error(`inconsistent data size. gpu data size=${o.originalSize}, data size=${i}`);let u=this.backend.device.createBuffer({mappedAtCreation:!0,size:a,usage:GPUBufferUsage.MAP_WRITE|GPUBufferUsage.COPY_SRC}),p=u.getMappedRange();new Uint8Array(p).set(new Uint8Array(s,n,i)),u.unmap();let h=this.backend.device.createCommandEncoder();h.copyBufferToBuffer(u,0,o.gpuData.buffer,0,a),this.backend.device.queue.submit([h.finish()]),u.destroy(),as("verbose",()=>`[WebGPU] GpuDataManager.upload(id=${e})`)}memcpy(e,t){let s=this.storageCache.get(e);if(!s)throw new Error("source gpu data for memcpy does not exist");let n=this.storageCache.get(t);if(!n)throw new Error("destination gpu data for memcpy does not exist");if(s.originalSize!==n.originalSize)throw new Error("inconsistent source and destination gpu data size");let i=Qe(s.originalSize),a=this.backend.getCommandEncoder();this.backend.endComputePass(),a.copyBufferToBuffer(s.gpuData.buffer,0,n.gpuData.buffer,0,i)}registerExternalBuffer(e,t,s){let n;if(s){if(n=s[0],e===s[1])return as("verbose",()=>`[WebGPU] GpuDataManager.registerExternalBuffer(size=${t}) => id=${n}, buffer is the same, skip.`),n;if(this.backend.capturedCommandList.has(this.backend.currentSessionId))throw new Error(`Registering a different external buffer under graph capture mode is not supported yet. + Please use the previous external buffer!`)}else n=ft();return this.storageCache.set(n,{gpuData:{id:n,type:0,buffer:e},originalSize:t}),as("verbose",()=>`[WebGPU] GpuDataManager.registerExternalBuffer(size=${t}) => id=${n}, registered.`),n}unregisterExternalBuffer(e){e!==void 0&&(this.storageCache.delete(e),as("verbose",()=>`[WebGPU] GpuDataManager.unregisterExternalBuffer() => id=${e}`))}create(e,t=GPUBufferUsage.STORAGE|GPUBufferUsage.COPY_SRC|GPUBufferUsage.COPY_DST){let s=pt(e),n,i=(t&GPUBufferUsage.STORAGE)===GPUBufferUsage.STORAGE,a=(t&GPUBufferUsage.UNIFORM)===GPUBufferUsage.UNIFORM;if(i||a){let u=(i?this.freeBuffers:this.freeUniformBuffers).get(s);u?u.length>0?n=u.pop():n=this.backend.device.createBuffer({size:s,usage:t}):n=this.backend.device.createBuffer({size:s,usage:t})}else n=this.backend.device.createBuffer({size:s,usage:t});let o={id:ft(),type:0,buffer:n};return this.storageCache.set(o.id,{gpuData:o,originalSize:Number(e)}),as("verbose",()=>`[WebGPU] GpuDataManager.create(size=${e}) => id=${o.id}`),o}get(e){var t;return(t=this.storageCache.get(e))==null?void 0:t.gpuData}release(e){let t=typeof e=="bigint"?Number(e):e,s=this.storageCache.get(t);if(!s){if(this.storageCache.size===0)return 0;throw new Error("releasing data does not exist")}return as("verbose",()=>`[WebGPU] GpuDataManager.release(id=${t}), gpuDataId=${s.gpuData.id}`),this.storageCache.delete(t),this.buffersPending.push(s.gpuData.buffer),s.originalSize}async download(e,t){let s=this.storageCache.get(Number(e));if(!s)throw new Error("data does not exist");await xt(this.backend,s.gpuData.buffer,s.originalSize,t)}refreshPendingBuffers(){if(this.buffersPending.length!==0)if(this.backend.sessionStatus==="default"){for(let e of this.buffersPending){let t=be.get(e.size);if((e.usage&GPUBufferUsage.STORAGE)===GPUBufferUsage.STORAGE){let s=this.freeBuffers.get(e.size)||[];t===void 0||s.length>=t?e.destroy():s.push(e)}else if((e.usage&GPUBufferUsage.UNIFORM)===GPUBufferUsage.UNIFORM){let s=this.freeUniformBuffers.get(e.size)||[];t===void 0||s.length>=t?e.destroy():s.push(e)}else e.destroy()}this.buffersPending=[]}else{let e=this.capturedPendingBuffers.get(this.backend.currentSessionId);e||(e=[],this.capturedPendingBuffers.set(this.backend.currentSessionId,e));for(let t of this.buffersPending)e.push(t);this.buffersPending=[]}}dispose(){this.freeBuffers.forEach(e=>{e.forEach(t=>{t.destroy()})}),this.freeUniformBuffers.forEach(e=>{e.forEach(t=>{t.destroy()})}),this.storageCache.forEach(e=>{e.gpuData.buffer.destroy()}),this.capturedPendingBuffers.forEach(e=>{e.forEach(t=>{t.destroy()})}),this.storageCache=new Map,this.freeBuffers=new Map,this.freeUniformBuffers=new Map,this.capturedPendingBuffers=new Map}onCreateSession(){this.sessionCount+=1}onReleaseSession(e){let t=this.capturedPendingBuffers.get(e);t&&(t.forEach(s=>{s.destroy()}),this.capturedPendingBuffers.delete(e)),this.sessionCount-=1,this.sessionCount===0&&(as("warning",()=>"[WebGPU] Clearing webgpu buffer cache"),this.storageCache.forEach(s=>{s.gpuData.buffer.destroy()}),this.storageCache=new Map)}},ms=(...e)=>new Kt(...e)}),Fs,Bt,rs=g(()=>{Fs=class{constructor(e){Object.assign(this,e)}get cacheKey(){return this.key||(this.key=Object.getOwnPropertyNames(this).sort().map(e=>`${this[e]}`).join(";")),this.key}},Bt=e=>new Fs(e)}),rr,Ws,ze,Js,Fr,ks,Xs,Ot=g(()=>{rr=class{static calcMatMulShape(e,t){return e[1]!==t[0]?void 0:[e[0],t[1]]}},Ws=class{static calcShape(e,t,s=!1){let n=e.length,i=t.length;if(n===0)return t;if(i===0)return e;let a=Math.max(e.length,t.length),o=new Array(a);if(s){if(n<2||i<2)return;let u=rr.calcMatMulShape([e[n-2],e[n-1]],[t[i-2],t[i-1]]);if(u===void 0)return;[o[a-2],o[a-1]]=u}for(let u=s?3:1;u<=a;u++){let p=n-u<0?1:e[n-u],h=i-u<0?1:t[i-u];if(p!==h&&p>1&&h>1)return;let k=Math.max(p,h);if(p&&h)o[a-u]=Math.max(p,h);else{if(k>1)return;o[a-u]=0}}return o}static isValidBroadcast(e,t){let s=e.length,n=t.length;if(s>n)return!1;for(let i=1;i<=s;i++)if(e[s-i]!==1&&e[s-i]!==t[n-i])return!1;return!0}},ze=class Wp{static size(t){return Wp.getSizeFromDimensionRange(t,0,t.length)}static convertShape(t,s=4){let n=t.length;if(n===0)return[];let i=new Array(n),a=n-1;for(;a>=0;){if(t[a]%s===0){i[a]=t[a]/s;break}if(s%t[a]!==0)throw new Error("cannot convert shape");i[a]=1,s/=t[a],a--}for(a--;a>=0;a--)i[a]=t[a];return i}static sizeFromDimension(t,s){if(s<0||s>t.length)throw new Error(`invalid dimension of ${s} for sizeFromDimension as Tensor has ${t.length} dimensions.`);return Wp.getSizeFromDimensionRange(t,s,t.length)}static sizeToDimension(t,s){if(s<0||s>t.length)throw new Error(`invalid dimension of ${s} for sizeToDimension as Tensor has ${t.length} dimensions.`);return Wp.getSizeFromDimensionRange(t,0,s)}static getSizeFromDimensionRange(t,s,n){let i=1;for(let a=s;a=0;--i)n[i]=n[i+1]*t[i+1];return n}static normalizeAxis(t,s){if(t<-s&&t>=s)throw new Error("unsupported axis for this operation.");return t<0?t+s:t}static normalizeAxes(t,s){return t.map(n=>this.normalizeAxis(n,s??t.length))}static sortBasedOnPerm(t,s){return s?s.map(n=>t[n]):t.slice().reverse()}static padShape(t,s){let n=t.length;return t.map((i,a)=>i+s[a]+s[a+n])}static areEqual(t,s){return t.length!==s.length?!1:t.every((n,i)=>n===s[i])}},Js=class Bc{static adjustPoolAttributes(t,s,n,i,a,o){if(!t&&n.length!==s.length-2)throw new Error("length of specified kernel shapes should be 2 less than length of input dimensions");if(t)for(let u=0;u=n.length?n.push(s[u+2]):n[u]=s[u+2];for(let u=0;u=n[u]||o[u+n.length]>=n[u])throw new Error("pads should be smaller than kernel")}}static adjustPadsBasedOnAutoPad(t,s,n,i,a,o,u){if(u){if(a.length!==2*(t.length-2))throw new Error("length of pads should be twice the length of data dimensions");if(s.length!==t.length-2)throw new Error("length of strides should be the length of data dimensions");if(i.length!==t.length-2)throw new Error("length of kernel shapes should be the length of data dimensions");for(let p=0;p{zt(),Ot(),ir=64,_r=(e,t)=>{if(t===3)throw new Error("vec3 has same alignment as vec4, use vec4 instead");switch(Number(e)){case 10:return t>1?`vec${t}`:"f16";case 1:return t>1?`vec${t}`:"f32";case 6:return t>1?`vec${t}`:"i32";case 12:return t>1?`vec${t}`:"u32";case 7:if(t>1)throw new Error("currently not supported vecX of uint64 yet");return["vec2","i32"];case 13:if(t>1)throw new Error("currently not supported vecX of uint64 yet");return["vec2","u32"];case 9:if(t!==4)throw new Error("bool must be vec4");return["u32","vec4"];case 22:return"i32";case 21:return"u32";default:throw new Error(`Unknown data type: ${e}`)}},fs=(e,t=1)=>{let s=_r(e,t);return typeof s=="string"?s:s[0]},Ss=(e,t=1)=>{let s=_r(e,t);return typeof s=="string"?s:s[1]},yt=(...e)=>{let t=[];return e.forEach(s=>{s.length!==0&&t.push({type:12,data:s},{type:12,data:ze.computeStrides(s)})}),t},qt=e=>e%4===0?4:e%2===0?2:1,Ls=(e="f32",t,s="0")=>!t||t===1?`${e}(${s})`:`vec${t}<${e}>(${s})`,$s=(e,t,s)=>e==="f32"?s:t===1?`f32(${s})`:`vec${t}(${s})`,Gs=(e,t)=>t===4?`(${e}.x + ${e}.y + ${e}.z + ${e}.w)`:t===2?`(${e}.x + ${e}.y)`:t===3?`(${e}.x + ${e}.y + ${e}.z)`:e,$t=(e,t,s,n)=>e.startsWith("uniforms.")&&s>4?typeof t=="string"?n==="f16"?`${e}[(${t}) / 8][(${t}) % 8 / 4][(${t}) % 8 % 4]`:`${e}[(${t}) / 4][(${t}) % 4]`:n==="f16"?`${e}[${Math.floor(t/8)}][${Math.floor(t%8/4)}][${t%8%4}]`:`${e}[${Math.floor(t/4)}][${t%4}]`:s>1?`${e}[${t}]`:e,sn=(e,t,s,n,i)=>{let a=typeof s=="number",o=a?s:s.length,u=[...new Array(o).keys()],p=o<2?"u32":o<=4?`vec${o}`:`array`,h=_r(t,i),k=typeof h=="string"?h:h[1],C=typeof h=="string"?h:h[0],d={indices:p,value:k,storage:C,tensor:t},z=ot=>typeof ot=="string"?ot:`${ot}u`,B={offsetToIndices:!1,indicesToOffset:!1,broadcastedIndicesToOffset:!1,set:!1,setByIndices:!1,get:!1,getByIndices:!1},V=a?"uniforms.":"",Z=`${V}${e}_shape`,ee=`${V}${e}_strides`,Y="";for(let ot=0;ot ${d.indices} { + var indices: ${d.indices}; + var current = offset; + ${Y} + return indices; + }`,pe=ot=>(B.offsetToIndices=!0,o<2?ot:`o2i_${e}(${ot})`),Me=[];if(o>=2)for(let ot=o-1;ot>=0;ot--)Me.push(`${$t(ee,ot,o)} * (indices[${ot}])`);let Oe=o<2?"":` + fn i2o_${e}(indices: ${d.indices}) -> u32 { + return ${Me.join("+")}; + }`,De=ot=>(B.indicesToOffset=!0,o<2?ot:`i2o_${e}(${ot})`),Ye=(...ot)=>o===0?"0u":`${d.indices}(${ot.map(z).join(",")})`,at=(ot,Et)=>o<2?`${ot}`:`${$t(ot,Et,o)}`,Pt=(ot,Et,ps)=>o<2?`${ot}=${ps};`:`${$t(ot,Et,o)}=${ps};`,Xt={},Zt=(ot,Et)=>{B.broadcastedIndicesToOffset=!0;let ps=`${Et.name}broadcastedIndicesTo${e}Offset`;if(ps in Xt)return`${ps}(${ot})`;let Ns=[];for(let Tr=o-1;Tr>=0;Tr--){let er=Et.indicesGet("outputIndices",Tr+Et.rank-o);Ns.push(`${at(ee,Tr)} * (${er} % ${at(Z,Tr)})`)}return Xt[ps]=`fn ${ps}(outputIndices: ${Et.type.indices}) -> u32 { + return ${Ns.length>0?Ns.join("+"):"0u"}; + }`,`${ps}(${ot})`},bt=(ot,Et)=>(()=>{if(d.storage===d.value)return`${e}[${ot}]=${Et};`;if(d.storage==="vec2"&&d.value==="i32")return`${e}[${ot}]=vec2(u32(${Et}), select(0u, 0xFFFFFFFFu, ${Et} < 0));`;if(d.storage==="vec2"&&d.value==="u32")return`${e}[${ot}]=vec2(u32(${Et}), 0u);`;if(d.storage==="u32"&&d.value==="vec4")return`${e}[${ot}]=dot(vec4(0x1, 0x100, 0x10000, 0x1000000), vec4(${Et}));`;throw new Error(`not supported combination of storage type ${d.storage} and value type ${d.value} yet`)})(),ss=ot=>(()=>{if(d.storage===d.value)return`${e}[${ot}]`;if(d.storage==="vec2"&&d.value==="i32")return`i32(${e}[${ot}].x)`;if(d.storage==="vec2"&&d.value==="u32")return`u32(${e}[${ot}].x)`;if(d.storage==="u32"&&d.value==="vec4")return`vec4(bool(${e}[${ot}] & 0xFFu), bool(${e}[${ot}] & 0xFF00u), bool(${e}[${ot}] & 0xFF0000u), bool(${e}[${ot}] & 0xFF000000u))`;throw new Error(`not supported combination of storage type ${d.storage} and value type ${d.value} yet`)})(),St=o<2?"":` + fn get_${e}ByIndices(indices: ${d.indices}) -> ${k} { + return ${ss(`i2o_${e}(indices)`)}; + }`,Ft=o<2?"":(()=>{let ot=u.map(ps=>`d${ps}: u32`).join(", "),Et=u.map(ps=>`d${ps}`).join(", ");return` + fn get_${e}(${ot}) -> ${k} { + return get_${e}ByIndices(${Ye(Et)}); + }`})(),bs=(...ot)=>{if(ot.length!==o)throw new Error(`indices length must be ${o}`);let Et=ot.map(z).join(",");return o===0?ss("0u"):o===1?ss(Et[0]):(B.get=!0,B.getByIndices=!0,B.indicesToOffset=!0,`get_${e}(${Et})`)},Ht=ot=>o<2?ss(ot):(B.getByIndices=!0,B.indicesToOffset=!0,`get_${e}ByIndices(${ot})`),Rt=o<2?"":` + fn set_${e}ByIndices(indices: ${d.indices}, value: ${k}) { + ${bt(`i2o_${e}(indices)`,"value")} + }`,_s=o<2?"":(()=>{let ot=u.map(ps=>`d${ps}: u32`).join(", "),Et=u.map(ps=>`d${ps}`).join(", ");return` + fn set_${e}(${ot}, value: ${k}) { + set_${e}ByIndices(${Ye(Et)}, value); + }`})();return{impl:()=>{let ot=[],Et=!1;return B.offsetToIndices&&(ot.push(he),Et=!0),B.indicesToOffset&&(ot.push(Oe),Et=!0),B.broadcastedIndicesToOffset&&(Object.values(Xt).forEach(ps=>ot.push(ps)),Et=!0),B.set&&(ot.push(_s),Et=!0),B.setByIndices&&(ot.push(Rt),Et=!0),B.get&&(ot.push(Ft),Et=!0),B.getByIndices&&(ot.push(St),Et=!0),!a&&Et&&ot.unshift(`const ${Z} = ${d.indices}(${s.join(",")});`,`const ${ee} = ${d.indices}(${ze.computeStrides(s).join(",")});`),ot.join(` +`)},type:d,offsetToIndices:pe,indicesToOffset:De,broadcastedIndicesToOffset:Zt,indices:Ye,indicesGet:at,indicesSet:Pt,set:(...ot)=>{if(ot.length!==o+1)throw new Error(`indices length must be ${o}`);let Et=ot[o];if(typeof Et!="string")throw new Error("value must be string");let ps=ot.slice(0,o).map(z).join(",");return o===0?bt("0u",Et):o===1?bt(ps[0],Et):(B.set=!0,B.setByIndices=!0,B.indicesToOffset=!0,`set_${e}(${ps}, ${Et})`)},setByOffset:bt,setByIndices:(ot,Et)=>o<2?bt(ot,Et):(B.setByIndices=!0,B.indicesToOffset=!0,`set_${e}ByIndices(${ot}, ${Et});`),get:bs,getByOffset:ss,getByIndices:Ht,usage:n,name:e,strides:ee,shape:Z,rank:o}},qe=(e,t,s,n=1)=>sn(e,t,s,"input",n),It=(e,t,s,n=1)=>sn(e,t,s,"output",n),za=(e,t,s)=>sn(e,t,s,"atomicOutput",1),Ui=(e,t,s,n=1)=>sn(e,t,s,"internal",n),Ba=class{constructor(e,t){this.normalizedDispatchGroup=e,this.limits=t,this.internalVariables=[],this.variables=[],this.uniforms=[],this.variableIndex=0}guardAgainstOutOfBoundsWorkgroupSizes(e){return`if (global_idx >= ${typeof e=="number"?`${e}u`:e}) { return; }`}mainStart(e=ir){let t=typeof e=="number"?e:e[0],s=typeof e=="number"?1:e[1],n=typeof e=="number"?1:e[2];if(t>this.limits.maxComputeWorkgroupSizeX||s>this.limits.maxComputeWorkgroupSizeY||n>this.limits.maxComputeWorkgroupSizeZ)throw new Error(`workgroup size [${t}, ${s}, ${n}] exceeds the maximum workgroup size [${this.limits.maxComputeWorkgroupSizeX}, ${this.limits.maxComputeWorkgroupSizeY}, ${this.limits.maxComputeWorkgroupSizeZ}].`);if(t*s*n>this.limits.maxComputeInvocationsPerWorkgroup)throw new Error(`workgroup size [${t}, ${s}, ${n}] exceeds the maximum workgroup invocations ${this.limits.maxComputeInvocationsPerWorkgroup}.`);let i=this.normalizedDispatchGroup[1]===1&&this.normalizedDispatchGroup[2]===1,a=i?`@builtin(global_invocation_id) global_id : vec3, + @builtin(workgroup_id) workgroup_id : vec3, + @builtin(local_invocation_index) local_idx : u32, + @builtin(local_invocation_id) local_id : vec3`:`@builtin(global_invocation_id) global_id : vec3, + @builtin(local_invocation_id) local_id : vec3, + @builtin(local_invocation_index) local_idx : u32, + @builtin(workgroup_id) workgroup_id : vec3, + @builtin(num_workgroups) num_workgroups : vec3`,o=i?`let global_idx = global_id.x; + let workgroup_index = workgroup_id.x;`:`let workgroup_index = workgroup_id.z * num_workgroups[0] * num_workgroups[1] + + workgroup_id.y * num_workgroups[0] + workgroup_id.x; + let global_idx = workgroup_index * ${t*s*n}u + local_idx;`;return`@compute @workgroup_size(${t}, ${s}, ${n}) + fn main(${a}) { + ${o} + `}appendVariableUniforms(e){e.rank!==0&&(e.shape.startsWith("uniforms.")&&this.uniforms.push({name:e.shape.replace("uniforms.",""),type:"u32",length:e.rank}),e.strides.startsWith("uniforms.")&&this.uniforms.push({name:e.strides.replace("uniforms.",""),type:"u32",length:e.rank}))}declareVariable(e,t){if(e.usage==="internal")throw new Error("cannot use internal variable with declareVariable(). use registerInternalVariables() instead.");this.variables.push(e),this.appendVariableUniforms(e);let s=e.usage==="input"?"read":"read_write",n=e.usage==="atomicOutput"?"atomic":e.type.storage;return`@group(0) @binding(${t}) var ${e.name}: array<${n}>;`}declareVariables(...e){return e.map(t=>this.declareVariable(t,this.variableIndex++)).join(` +`)}registerInternalVariable(e){if(e.usage!=="internal")throw new Error("cannot use input or output variable with registerInternalVariable(). use declareVariables() instead.");this.internalVariables.push(e),this.appendVariableUniforms(e)}registerInternalVariables(...e){return e.forEach(t=>this.registerInternalVariable(t)),this}registerUniform(e,t,s=1){return this.uniforms.push({name:e,type:t,length:s}),this}registerUniforms(e){return this.uniforms=this.uniforms.concat(e),this}uniformDeclaration(){if(this.uniforms.length===0)return"";let e=[];for(let{name:t,type:s,length:n}of this.uniforms)if(n&&n>4)s==="f16"?e.push(`@align(16) ${t}:array, ${Math.ceil(n/8)}>`):e.push(`${t}:array, ${Math.ceil(n/4)}>`);else{let i=n==null||n===1?s:`vec${n}<${s}>`;e.push(`${t}:${i}`)}return` + struct Uniforms { ${e.join(", ")} }; + @group(0) @binding(${this.variableIndex}) var uniforms: Uniforms;`}get additionalImplementations(){return this.uniformDeclaration()+this.variables.map(e=>e.impl()).join(` +`)+this.internalVariables.map(e=>e.impl()).join(` +`)}get variablesInfo(){if(this.uniforms.length===0)return;let e=t=>[12,10,1,6][["u32","f16","f32","i32"].indexOf(t)];return this.uniforms.map(t=>[e(t.type),t.length??1])}},Ra=(e,t)=>new Ba(e,t)}),Na,Vi,Wi,ja,Ua,Gi,pr,Va,Ki,Kr=g(()=>{zt(),Ot(),rs(),Yt(),Na=(e,t)=>{if(!e||e.length!==1)throw new Error("Transpose requires 1 input.");if(t.length!==0&&t.length!==e[0].dims.length)throw new Error(`perm size ${t.length} does not match input rank ${e[0].dims.length}`)},Vi=(e,t)=>t.length!==0?t:[...new Array(e).keys()].reverse(),Wi=(e,t)=>ze.sortBasedOnPerm(e,Vi(e.length,t)),ja=(e,t,s,n)=>{let i=`fn perm(i: ${n.type.indices}) -> ${s.type.indices} { + var a: ${s.type.indices};`;for(let a=0;a{let s=[],n=[];for(let i=0;i{let s=0;for(let n=0;n{let s=e.dataType,n=e.dims.length,i=Vi(n,t),a=Wi(e.dims,i),o=e.dims,u=a,p=n<2||Gi(i,e.dims),h;if(p)return h=B=>{let V=qe("input",s,o,4),Z=It("output",s,u,4);return` + ${B.registerUniform("output_size","u32").declareVariables(V,Z)} + ${B.mainStart()} + ${B.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + output[global_idx] = input[global_idx]; + }`},{name:"TransposeCopy",shaderCache:{inputDependencies:["type"]},getRunData:()=>{let B=ze.size(a);return{outputs:[{dims:a,dataType:e.dataType}],dispatchGroup:{x:Math.ceil(B/64/4)},programUniforms:[{type:12,data:Math.ceil(B/4)}]}},getShaderSource:h};let{newShape:k,newPerm:C}=Ua(e.dims,i),d=ze.areEqual(C,[2,3,1]),z=ze.areEqual(C,[3,1,2]);if(k.length===2||d||z){o=d?[k[0],k[1]*k[2]]:z?[k[0]*k[1],k[2]]:k,u=[o[1],o[0]];let B=16;return h=V=>{let Z=qe("a",s,o.length),ee=It("output",s,u.length);return` + ${V.registerUniform("output_size","u32").declareVariables(Z,ee)} + var tile : array, ${B}>; + ${V.mainStart([B,B,1])} + let stride = (uniforms.output_shape[1] - 1) / ${B} + 1; + let workgroup_id_x = workgroup_index % stride; + let workgroup_id_y = workgroup_index / stride; + let input_col = workgroup_id_y * ${B}u + local_id.x; + let input_row = workgroup_id_x * ${B}u + local_id.y; + if (input_row < uniforms.a_shape[0] && input_col < uniforms.a_shape[1]) { + tile[local_id.y][local_id.x] = ${Z.getByIndices(`${Z.type.indices}(input_row, input_col)`)}; + } + workgroupBarrier(); + + let output_col = workgroup_id_x * ${B}u + local_id.x; + let output_row = workgroup_id_y * ${B}u + local_id.y; + if (output_row < uniforms.output_shape[0] && output_col < uniforms.output_shape[1]) { + ${ee.setByIndices(`${ee.type.indices}(output_row, output_col)`,"tile[local_id.x][local_id.y]")} + } + }`},{name:"TransposeShared",shaderCache:{inputDependencies:["type"]},getRunData:()=>{let V=ze.size(a);return{outputs:[{dims:a,dataType:e.dataType}],dispatchGroup:{x:Math.ceil(u[1]/B),y:Math.ceil(u[0]/B)},programUniforms:[{type:12,data:V},...yt(o,u)]}},getShaderSource:h}}return h=B=>{let V=qe("a",s,o.length),Z=It("output",s,u.length);return` + ${B.registerUniform("output_size","u32").declareVariables(V,Z)} + + ${ja(i,n,V,Z)} + + ${B.mainStart()} + ${B.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + + let indices = ${Z.offsetToIndices("global_idx")}; + let aIndices = perm(indices); + + ${Z.setByOffset("global_idx",V.getByIndices("aIndices"))} + }`},{name:"Transpose",shaderCache:{hint:`${t}`,inputDependencies:["rank"]},getRunData:()=>{let B=ze.size(a);return{outputs:[{dims:a,dataType:e.dataType}],dispatchGroup:{x:Math.ceil(B/64)},programUniforms:[{type:12,data:B},...yt(o,u)]}},getShaderSource:h}},Va=(e,t)=>{Na(e.inputs,t.perm),e.compute(pr(e.inputs[0],t.perm))},Ki=e=>Bt({perm:e.perm})}),ai,Wa,Ga,Ka,Ha,qa,Qa,Xa,Hi,Ya,hr,rn,Ja,Rc,Za,Nc,el,qi,tl,sl,rl,jc=g(()=>{zt(),Ot(),Yt(),di(),Kr(),ai={max:"select(bestValue, candidate, candidate > bestValue)",min:"select(bestValue, candidate, candidate < bestValue)",mean:"bestValue + candidate",sum:"bestValue + candidate",prod:"bestValue * candidate",sumSquare:"bestValue + candidate * candidate",logSumExp:"bestValue + exp(candidate)",l1:"bestValue + abs(candidate)",l2:"bestValue + candidate * candidate",logSum:"bestValue + candidate"},Wa={max:"select(bestValue, candidate, candidate > bestValue)",min:"select(bestValue, candidate, candidate < bestValue)",mean:"bestValue + candidate",sum:"bestValue + candidate",prod:"bestValue * candidate",sumSquare:"bestValue + candidate",logSumExp:"bestValue + candidate",l1:"bestValue + candidate",l2:"bestValue + candidate",logSum:"bestValue + candidate"},Ga={max:"_A[offset]",min:"_A[offset]",mean:"0",sum:"0",prod:"1",sumSquare:"0",logSumExp:"0",l1:"0",l2:"0",logSum:"0"},Ka={max:"bestValue",min:"bestValue",sum:"bestValue",prod:"bestValue",sumSquare:"bestValue",logSumExp:"log(bestValue)",l1:"bestValue",l2:"sqrt(bestValue)",logSum:"log(bestValue)"},Ha=(e,t)=>{let s=[];for(let n=t-e;n{let s=[],n=e.length;for(let a=0;ae[a]);return[s,i]},Qa=(e,t)=>{let s=e.length+t.length,n=[],i=0;for(let a=0;a{for(let s=0;s{let s=[];if(!Xa(e,t)){for(let n=0;ns.push(n))}return s},Ya=(e,t,s,n,i,a,o)=>{let u=s[0].dims,p=ze.size(a),h=ze.size(o),k=qe("_A",s[0].dataType,u),C=It("output",i,a),d=64;p===1&&(d=256);let z=` + var aBestValues : array; + `,B=V=>` + ${V.registerUniform("reduceSize","u32").declareVariables(k,C)} + ${z} + fn DIV_CEIL(a : u32, b : u32) -> u32 { + return ((a - 1u) / b + 1u); + } + ${V.mainStart(d)} + + let outputIndex = global_idx / ${d}; + let offset = outputIndex * uniforms.reduceSize; + + var bestValue = f32(${Ga[n]}); + let Length = uniforms.reduceSize; + for (var k = local_idx; k < Length; k = k + ${d}) { + let candidate = f32(${k.getByOffset("offset + k")}); + bestValue = ${ai[n]}; + } + aBestValues[local_idx] = bestValue; + workgroupBarrier(); + + var reduceSize = min(Length, ${d}u); + for (var currentSize = reduceSize / 2u; reduceSize > 1u; + currentSize = reduceSize / 2u) { + let interval = DIV_CEIL(reduceSize, 2u); + if (local_idx < currentSize) { + let candidate = aBestValues[local_idx + interval]; + bestValue = ${Wa[n]}; + aBestValues[local_idx] = bestValue; + } + reduceSize = interval; + workgroupBarrier(); + } + + if (local_idx == 0u) { + ${C.setByOffset("outputIndex",`${n==="mean"?`${C.type.storage}(bestValue / f32(uniforms.reduceSize))`:`${C.type.storage}(${Ka[n]})`}`)}; + } + }`;return{name:e,shaderCache:{hint:`${t};${d}`,inputDependencies:["type"]},getShaderSource:B,getRunData:()=>({outputs:[{dims:a,dataType:i}],dispatchGroup:{x:p},programUniforms:[{type:12,data:h}]})}},hr=(e,t,s,n)=>{let i=e.inputs.length===1?s:Qi(e.inputs,s),a=i.axes;a.length===0&&!i.noopWithEmptyAxes&&(a=e.inputs[0].dims.map((z,B)=>B));let o=ze.normalizeAxes(a,e.inputs[0].dims.length),u=o,p=e.inputs[0],h=Hi(u,e.inputs[0].dims.length);h.length>0&&(p=e.compute(pr(e.inputs[0],h),{inputs:[0],outputs:[-1]})[0],u=Ha(u.length,p.dims.length));let[k,C]=qa(p.dims,u),d=k;i.keepDims&&(d=Qa(k,o)),e.compute(Ya(t,i.cacheKey,[p],n,e.inputs[0].dataType,d,C),{inputs:[p]})},rn=(e,t)=>{hr(e,"ReduceMeanShared",t,"mean")},Ja=(e,t)=>{hr(e,"ReduceL1Shared",t,"l1")},Rc=(e,t)=>{hr(e,"ReduceL2Shared",t,"l2")},Za=(e,t)=>{hr(e,"ReduceLogSumExpShared",t,"logSumExp")},Nc=(e,t)=>{hr(e,"ReduceMaxShared",t,"max")},el=(e,t)=>{hr(e,"ReduceMinShared",t,"min")},qi=(e,t)=>{hr(e,"ReduceProdShared",t,"prod")},tl=(e,t)=>{hr(e,"ReduceSumShared",t,"sum")},sl=(e,t)=>{hr(e,"ReduceSumSquareShared",t,"sumSquare")},rl=(e,t)=>{hr(e,"ReduceLogSumShared",t,"logSum")}}),gr,li,ui,Qi,wr,Xi,nl,il,Yi,ol,al,Ji,ll,ul,Zi,yr,dl,eo,cl,pl,to,hl,ml,so,fl,_l,di=g(()=>{zt(),Ot(),rs(),Yt(),jc(),gr=e=>{if(!e||e.length===0||e.length>2)throw new Error("Reduce op requires 1 or 2 inputs.");if(e.length===2&&e[1].dims.length!==1)throw new Error("Invalid axes input dims.")},li=e=>["","",`var value = ${e.getByIndices("input_indices")};`,""],ui=(e,t,s,n,i,a,o=!1,u=!1)=>{let p=[],h=s[0].dims,k=h.length,C=ze.normalizeAxes(i,k),d=!u&&C.length===0;h.forEach((V,Z)=>{d||C.indexOf(Z)>=0?o&&p.push(1):p.push(V)});let z=p.length,B=ze.size(p);return{name:e,shaderCache:t,getShaderSource:V=>{let Z=[],ee=qe("_A",s[0].dataType,k),Y=It("output",a,z),he=n(ee,Y,C),pe=he[2];for(let Me=0,Oe=0;Me=0?(o&&Oe++,pe=`for(var j${Me}: u32 = 0; j${Me} < ${h[Me]}; j${Me}++) { + ${he[2].includes("last_index")?`let last_index = j${Me};`:""} + ${ee.indicesSet("input_indices",Me,`j${Me}`)} + ${pe} + }`):(Z.push(`${ee.indicesSet("input_indices",Me,Y.indicesGet("output_indices",Oe))};`),Oe++);return` + + ${V.registerUniform("output_size","u32").declareVariables(ee,Y)} + + ${V.mainStart()} + ${V.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + var input_indices: ${ee.type.indices}; + let output_indices = ${Y.offsetToIndices("global_idx")}; + + ${Z.join(` +`)} + ${he[0]} // init ops for reduce max/min + ${he[1]} + ${pe} + ${he[3]} + ${he.length===4?Y.setByOffset("global_idx","value"):he.slice(4).join(` +`)} + }`},getRunData:()=>({outputs:[{dims:p,dataType:a}],dispatchGroup:{x:Math.ceil(B/64)},programUniforms:[{type:12,data:B},...yt(h,p)]})}},Qi=(e,t)=>{let s=[];return e[1].dims[0]>0&&e[1].getBigInt64Array().forEach(n=>s.push(Number(n))),Bt({axes:s,keepDims:t.keepDims,noopWithEmptyAxes:t.noopWithEmptyAxes})},wr=(e,t,s,n)=>{let i=e.inputs,a=i.length===1?s:Qi(i,s);e.compute(ui(t,{hint:a.cacheKey,inputDependencies:["rank"]},[i[0]],a.noopWithEmptyAxes&&a.axes.length===0?li:n,a.axes,i[0].dataType,a.keepDims,a.noopWithEmptyAxes),{inputs:[0]})},Xi=(e,t)=>{gr(e.inputs),wr(e,"ReduceLogSum",t,(s,n)=>[`var value = ${n.type.storage}(0);`,"",`value += ${s.getByIndices("input_indices")};`,"value = log(value);"])},nl=(e,t)=>{gr(e.inputs),wr(e,"ReduceL1",t,(s,n)=>[`var value = ${n.type.storage}(0);`,"",`value += abs(${s.getByIndices("input_indices")});`,""])},il=(e,t)=>{gr(e.inputs),wr(e,"ReduceL2",t,(s,n)=>[`var t = ${n.type.value}(0); var value = ${n.type.value}(0);`,"",`t = ${s.getByIndices("input_indices")}; value += (t * t);`,"value = sqrt(value);"])},Yi=(e,t)=>{gr(e.inputs),wr(e,"ReduceLogSumExp",t,(s,n)=>[`var value = ${n.type.storage}(0);`,"",`value += exp(${s.getByIndices("input_indices")});`,"value = log(value);"])},ol=(e,t)=>{gr(e.inputs),wr(e,"ReduceMax",t,(s,n,i)=>{let a=[];for(let o=0;o=0||i.length===0)&&a.push(s.indicesSet("input_indices",o,0));return[`${a.join(` +`)}`,`var value = ${s.getByIndices("input_indices")};`,`value = max(value, ${s.getByIndices("input_indices")});`,""]})},al=(e,t)=>{gr(e.inputs),wr(e,"ReduceMean",t,(s,n,i)=>{let a=1;for(let o=0;o=0||i.length===0)&&(a*=e.inputs[0].dims[o]);return["var sum = f32(0);","",`sum += f32(${s.getByIndices("input_indices")});`,`let value = ${n.type.value}(sum / ${a});`]})},Ji=(e,t)=>{gr(e.inputs),wr(e,"ReduceMin",t,(s,n,i)=>{let a=[];for(let o=0;o=0||i.length===0)&&a.push(`input_indices[${o}] = 0;`);return[`${a.join(` +`)}`,`var value = ${s.getByIndices("input_indices")};`,`value = min(value, ${s.getByIndices("input_indices")});`,""]})},ll=(e,t)=>{gr(e.inputs),wr(e,"ReduceProd",t,(s,n)=>[`var value = ${n.type.storage}(1);`,"",`value *= ${s.getByIndices("input_indices")};`,""])},ul=(e,t)=>{gr(e.inputs),wr(e,"ReduceSum",t,(s,n)=>[`var value = ${n.type.storage}(0);`,"",`value += ${s.getByIndices("input_indices")};`,""])},Zi=(e,t)=>{gr(e.inputs),wr(e,"ReduceSumSquare",t,(s,n)=>[`var t = ${n.type.value}(0); var value = ${n.type.value}(0);`,"",`t = ${s.getByIndices("input_indices")}; value += t * t;`,""])},yr=(e,t,s)=>{if(t.length===0)return s;let n=1,i=1;for(let a=0;a1024},dl=(e,t)=>{yr(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?al(e,t):rn(e,t)},eo=(e,t)=>{yr(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?nl(e,t):Ja(e,t)},cl=(e,t)=>{yr(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?il(e,t):Rc(e,t)},pl=(e,t)=>{yr(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?Yi(e,t):Za(e,t)},to=(e,t)=>{yr(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?ol(e,t):Nc(e,t)},hl=(e,t)=>{yr(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?Ji(e,t):el(e,t)},ml=(e,t)=>{yr(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?ll(e,t):qi(e,t)},so=(e,t)=>{yr(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?ul(e,t):tl(e,t)},fl=(e,t)=>{yr(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?Zi(e,t):sl(e,t)},_l=(e,t)=>{yr(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?Xi(e,t):rl(e,t)}}),ro,gl,no,io,Uc=g(()=>{zt(),rs(),di(),ro=e=>{if(!e||e.length===0||e.length>2)throw new Error("ArgMinMaxOp op requires 1 or 2 inputs.");if(e[0].dataType!==1)throw new Error("Invalid input type.")},gl=(e,t)=>{ro(e.inputs);let s=(n,i,a)=>{let o=[];for(let u=0;u=0||a.length===0)&&o.push(`input_indices[${u}] = 0;`);return[`${o.join(` +`)}`,`var value = ${n.getByIndices("input_indices")}; +var best_index : i32 = 0;`,`if (${n.getByIndices("input_indices")} ${t.selectLastIndex>0?"<=":"<"} value) { + value = ${n.getByIndices("input_indices")}; + best_index = i32(last_index); + }`,"",i.setByOffset("global_idx","best_index")]};e.compute(ui("ArgMin",{hint:t.cacheKey,inputDependencies:["rank"]},[e.inputs[0]],s,[t.axis],7,t.keepDims),{inputs:[0]})},no=(e,t)=>{ro(e.inputs);let s=(n,i,a)=>{let o=[];for(let u=0;u=0||a.length===0)&&o.push(`input_indices[${u}] = 0;`);return[`${o.join(` +`)}`,`var value = ${n.getByIndices("input_indices")}; +var best_index : i32 = 0;`,`if (${n.getByIndices("input_indices")} ${t.selectLastIndex>0?">=":">"} value) { + value = ${n.getByIndices("input_indices")}; + best_index = i32(last_index); + }`,"",i.setByOffset("global_idx","best_index")]};e.compute(ui("argMax",{hint:t.cacheKey,inputDependencies:["rank"]},[e.inputs[0]],s,[t.axis],7,t.keepDims),{inputs:[0]})},io=e=>Bt(e)}),oo,ci,wl,ao,yl,Nn,lo,Ml,uo=g(()=>{zt(),Ot(),ue(),Yt(),oo=(e,t)=>{let s=e[0],n=e[1],i=e[2],a=e[3],o=e[4],u=e[5];if(o&&u)throw new Error("Attention cannot have both past and attention_bias");if(s.dims.length!==3)throw new Error('Input "input" must have 3 dimensions');let p=s.dims[0],h=s.dims[1],k=s.dims[2];if(i.dims.length!==1)throw new Error('Input "bias" is expected to have 1 dimensions');if(n.dims.length!==2)throw new Error('Input "weights" is expected to have 2 dimensions');if(n.dims[0]!==k)throw new Error("Input 1 dimension 0 should have same length as dimension 2 of input 0");if(i.dims[0]!==n.dims[1])throw new Error('Input "bias" dimension 0 should have same length as dimension 1 of input "weights"');let C=i.dims[0]/3,d=C,z=d;if(t.qkvHiddenSizes.length>0){if(t.qkvHiddenSizes.length!==3)throw new Error("qkv_hidden_sizes attribute should have 3 elements");for(let he of t.qkvHiddenSizes)if(he%t.numHeads!==0)throw new Error("qkv_hidden_sizes should be divisible by num_heads");C=t.qkvHiddenSizes[0],d=t.qkvHiddenSizes[1],z=t.qkvHiddenSizes[2]}let B=h;if(C!==d)throw new Error("qkv_hidden_sizes first element should be same as the second");if(i.dims[0]!==C+d+z)throw new Error('Input "bias" dimension 0 should have same length as sum of Q/K/V hidden sizes');let V=0;if(o){if(d!==z)throw new Error('Input "past" expect k_hidden_size == v_hidden_size');if(o.dims.length!==5)throw new Error('Input "past" must have 5 dimensions');if(o.dims[0]!==2)throw new Error('Input "past" first dimension must be 2');if(o.dims[1]!==p)throw new Error('Input "past" second dimension must be batch_size');if(o.dims[2]!==t.numHeads)throw new Error('Input "past" third dimension must be num_heads');if(o.dims[4]!==d/t.numHeads)throw new Error('Input "past" fifth dimension must be k_hidden_size / num_heads');t.pastPresentShareBuffer||(V=o.dims[3])}let Z=B+V,ee=-1,Y=0;if(a)throw new Error("Mask not supported");if(o)throw new Error("past is not supported");if(u){if(u.dims.length!==4)throw new Error('Input "attention_bias" must have 4 dimensions');if(u.dims[0]!==p||u.dims[1]!==t.numHeads||u.dims[2]!==h||u.dims[3]!==Z)throw new Error('Expect "attention_bias" shape (batch_size, num_heads, sequence_length, total_sequence_length)')}return{batchSize:p,sequenceLength:h,pastSequenceLength:V,kvSequenceLength:B,totalSequenceLength:Z,maxSequenceLength:ee,inputHiddenSize:k,hiddenSize:C,vHiddenSize:z,headSize:Math.floor(C/t.numHeads),vHeadSize:Math.floor(z/t.numHeads),numHeads:t.numHeads,isUnidirectional:!1,pastPresentShareBuffer:!1,maskFilterValue:t.maskFilterValue,maskType:Y,scale:t.scale,broadcastResPosBias:!1,passPastInKv:!1,qkvFormat:1}},ci=(e,t,s)=>t&&e?` + let total_sequence_length_input = u32(${t.getByOffset("0")}); + let present_sequence_length = max(total_sequence_length_input, uniforms.past_sequence_length); + let is_subsequent_prompt: bool = sequence_length > 1 && sequence_length != total_sequence_length_input; + let is_first_prompt: bool = is_subsequent_prompt == false && sequence_length == total_sequence_length_input; + total_sequence_length = u32(${e==null?void 0:e.getByOffset("batchIdx")}) + 1; + var past_sequence_length: u32 = 0; + if (is_first_prompt == false) { + past_sequence_length = total_sequence_length - sequence_length; + } + `:` + ${s?"let past_sequence_length = uniforms.past_sequence_length":""}; + let present_sequence_length = total_sequence_length; + `,wl=(e,t,s,n,i,a,o,u)=>{let p=qt(o?1:a),h=64,k=a/p;k{let Y=It("x",e.dataType,e.dims,p),he=[Y],pe=o?qe("seq_lens",o.dataType,o.dims):void 0;pe&&he.push(pe);let Me=u?qe("total_sequence_length_input",u.dataType,u.dims):void 0;Me&&he.push(Me);let Oe=Ss(e.dataType),De=[{name:"batch_size",type:"u32"},{name:"num_heads",type:"u32"},{name:"past_sequence_length",type:"u32"},{name:"sequence_length",type:"u32"},{name:"total_sequence_length",type:"u32"},{name:"elements_per_thread",type:"u32"}];return` + var thread_max: array; + var thread_sum: array; + ${ee.registerUniforms(De).declareVariables(...he)} + ${ee.mainStart([h,1,1])} + let batchIdx = workgroup_id.z / uniforms.num_heads; + let headIdx = workgroup_id.z % uniforms.num_heads; + let sequence_length = uniforms.sequence_length; + var total_sequence_length = uniforms.total_sequence_length; + ${ci(pe,Me,!1)} + let local_offset = local_idx * uniforms.elements_per_thread; + let offset = (global_idx / ${h}) * uniforms.total_sequence_length + local_offset; + let seq_causal_length = ${o?"u32(past_sequence_length + workgroup_id.y + 1)":"total_sequence_length"}; + var thread_max_vector = ${B}(-3.402823e+38f); + for (var i: u32 = 0; i < uniforms.elements_per_thread && i + local_offset < seq_causal_length; i++) { + thread_max_vector = max(${B}(x[offset + i]), thread_max_vector); + } + thread_max[local_idx] = ${(()=>{switch(p){case 1:return"thread_max_vector";case 2:return"max(thread_max_vector.x, thread_max_vector.y)";case 4:return"max(max(thread_max_vector.x, thread_max_vector.y), max(thread_max_vector.z, thread_max_vector.w))";default:throw new Error(`Unsupported components: ${p}`)}})()}; + workgroupBarrier(); + + var max_value = f32(-3.402823e+38f); + for (var i = 0u; i < ${h}; i++) { + max_value = max(thread_max[i], max_value); + } + + var sum_vector = ${B}(0); + for (var i: u32 = 0; i < uniforms.elements_per_thread && i + local_offset < seq_causal_length; i++) { + sum_vector += exp(${B}(x[offset + i]) - max_value); + } + thread_sum[local_idx] = ${(()=>{switch(p){case 1:return"sum_vector";case 2:return"sum_vector.x + sum_vector.y";case 4:return"sum_vector.x + sum_vector.y + sum_vector.z + sum_vector.w";default:throw new Error(`Unsupported components: ${p}`)}})()}; + workgroupBarrier(); + + var sum: f32 = 0; + for (var i = 0u; i < ${h}; i++) { + sum += thread_sum[i]; + } + + if (sum == 0) { + for (var i: u32 = 0; i < uniforms.elements_per_thread && i + local_offset < seq_causal_length; i++) { + x[offset + i] = ${Y.type.value}(${Oe}(1.0) / ${Oe}(seq_causal_length)); + } + } else { + for (var i: u32 = 0; i < uniforms.elements_per_thread && i + local_offset < seq_causal_length; i++) { + var f32input = ${B}(x[offset + i]); + x[offset + i] = ${Y.type.value}(exp(f32input - max_value) / sum); + } + } + ${o?` + for (var total_seq_id: u32 = seq_causal_length; total_seq_id + local_offset < uniforms.total_sequence_length; total_seq_id++) { + x[offset + total_seq_id] = ${Y.type.value}(${Oe}(0)); + }`:""}; + }`};return{name:"AttentionProbsSoftmax",shaderCache:{hint:`${h};${z};${p}`,inputDependencies:V},getShaderSource:Z,getRunData:()=>({outputs:[],dispatchGroup:{x:Math.ceil(a/h),y:i,z:t*s},programUniforms:d})}},ao=(e,t,s,n,i,a,o,u,p)=>{let h=o+a.kvSequenceLength,k=[a.batchSize,a.numHeads,a.sequenceLength,h],C=e>1&&n,d=a.kvNumHeads?a.kvNumHeads:a.numHeads,z=C?[a.batchSize,d,h,a.headSize]:void 0,B=a.nReps?a.nReps:1,V=a.scale===0?1/Math.sqrt(a.headSize):a.scale,Z=qt(a.headSize),ee=a.headSize/Z,Y=12,he={x:Math.ceil(h/Y),y:Math.ceil(a.sequenceLength/Y),z:a.batchSize*a.numHeads},pe=[{type:12,data:a.sequenceLength},{type:12,data:ee},{type:12,data:h},{type:12,data:a.numHeads},{type:12,data:a.headSize},{type:1,data:V},{type:12,data:o},{type:12,data:a.kvSequenceLength},{type:12,data:B}],Me=C&&n&&ze.size(n.dims)>0,Oe=["type","type"];Me&&Oe.push("type"),i&&Oe.push("type"),u&&Oe.push("type"),p&&Oe.push("type");let De=[{dims:k,dataType:t.dataType,gpuDataType:0}];C&&De.push({dims:z,dataType:t.dataType,gpuDataType:0});let Ye=at=>{let Pt=qe("q",t.dataType,t.dims,Z),Xt=qe("key",s.dataType,s.dims,Z),Zt=[Pt,Xt];if(Me){let Rt=qe("past_key",n.dataType,n.dims,Z);Zt.push(Rt)}i&&Zt.push(qe("attention_bias",i.dataType,i.dims));let bt=u?qe("seq_lens",u.dataType,u.dims):void 0;bt&&Zt.push(bt);let ss=p?qe("total_sequence_length_input",p.dataType,p.dims):void 0;ss&&Zt.push(ss);let St=It("output",t.dataType,k),Ft=[St];C&&Ft.push(It("present_key",t.dataType,z,Z));let bs=Ss(1,Z),Ht=[{name:"M",type:"u32"},{name:"K",type:"u32"},{name:"N",type:"u32"},{name:"num_heads",type:"u32"},{name:"head_size",type:"u32"},{name:"alpha",type:"f32"},{name:"past_sequence_length",type:"u32"},{name:"kv_sequence_length",type:"u32"},{name:"n_reps",type:"u32"}];return` + const TILE_SIZE = ${Y}u; + + var tileQ: array<${Pt.type.storage}, ${Y*Y}>; + var tileK: array<${Pt.type.storage}, ${Y*Y}>; + ${at.registerUniforms(Ht).declareVariables(...Zt,...Ft)} + ${at.mainStart([Y,Y,1])} + // x holds the N and y holds the M + let headIdx = workgroup_id.z % uniforms.num_heads; + let kvHeadIdx = ${B===1?"headIdx":"headIdx / uniforms.n_reps"}; + let kv_num_heads = ${B===1?"uniforms.num_heads":"uniforms.num_heads / uniforms.n_reps"}; + let batchIdx = workgroup_id.z / uniforms.num_heads; + let m = workgroup_id.y * TILE_SIZE; + let n = workgroup_id.x * TILE_SIZE; + let sequence_length = uniforms.M; + var total_sequence_length = uniforms.N; + ${ci(bt,ss,!0)} + let absKvHeadIdx = batchIdx * kv_num_heads + kvHeadIdx; + let qOffset = workgroup_id.z * uniforms.M * uniforms.K + m * uniforms.K; + ${Me&&C?"let pastKeyOffset = absKvHeadIdx * uniforms.past_sequence_length * uniforms.K;":""}; + let kOffset = absKvHeadIdx * uniforms.kv_sequence_length * uniforms.K; + ${C?"let presentKeyOffset = absKvHeadIdx * uniforms.N * uniforms.K;":""} + var value = ${bs}(0); + for (var w: u32 = 0u; w < uniforms.K; w += TILE_SIZE) { + if (global_id.y < uniforms.M && w + local_id.x < uniforms.K) { + tileQ[TILE_SIZE * local_id.y + local_id.x] = q[qOffset + local_id.y * uniforms.K + w + local_id.x]; + } + if (n + local_id.y < uniforms.N && w + local_id.x < uniforms.K) { + var idx = TILE_SIZE * local_id.y + local_id.x; + ${Me&&C?` + if (n + local_id.y < past_sequence_length) { + tileK[idx] = past_key[pastKeyOffset + (n + local_id.y) * uniforms.K + w + local_id.x]; + } else if (n + local_id.y - past_sequence_length < uniforms.kv_sequence_length) { + tileK[idx] = key[kOffset + (n + local_id.y - past_sequence_length) * uniforms.K + w + local_id.x]; + }`:` + if (n + local_id.y < uniforms.kv_sequence_length) { + tileK[idx] = key[kOffset + (n + local_id.y) * uniforms.K + w + local_id.x]; + }`} + ${C?`if (n + local_id.y < present_sequence_length) { + present_key[presentKeyOffset + (n + local_id.y) * uniforms.K + w + local_id.x] = tileK[idx]; + }`:""} + } + workgroupBarrier(); + + for (var k: u32 = 0u; k < TILE_SIZE && w+k < uniforms.K; k++) { + value += ${bs}(tileQ[TILE_SIZE * local_id.y + k] * tileK[TILE_SIZE * local_id.x + k]); + } + + workgroupBarrier(); + } + + if (global_id.y < uniforms.M && global_id.x < total_sequence_length) { + let headOffset = workgroup_id.z * uniforms.M * uniforms.N; + let outputIdx = headOffset + global_id.y * uniforms.N + global_id.x; + var sum: f32 = ${(()=>{switch(Z){case 1:return"value";case 2:return"value.x + value.y";case 4:return"value.x + value.y + value.z + value.w";default:throw new Error(`Unsupported components: ${Z}`)}})()}; + output[outputIdx] = ${St.type.value} (sum * uniforms.alpha) + ${i?"attention_bias[outputIdx]":"0.0"}; + } + }`};return{name:"AttentionProbs",shaderCache:{hint:`${Z};${i!==void 0};${n!==void 0};${e}`,inputDependencies:Oe},getRunData:()=>({outputs:De,dispatchGroup:he,programUniforms:pe}),getShaderSource:Ye}},yl=(e,t,s,n,i,a,o=void 0,u=void 0)=>{let p=a+i.kvSequenceLength,h=i.nReps?i.nReps:1,k=i.vHiddenSize*h,C=e>1&&n,d=i.kvNumHeads?i.kvNumHeads:i.numHeads,z=C?[i.batchSize,d,p,i.headSize]:void 0,B=[i.batchSize,i.sequenceLength,k],V=12,Z={x:Math.ceil(i.vHeadSize/V),y:Math.ceil(i.sequenceLength/V),z:i.batchSize*i.numHeads},ee=[{type:12,data:i.sequenceLength},{type:12,data:p},{type:12,data:i.vHeadSize},{type:12,data:i.numHeads},{type:12,data:i.headSize},{type:12,data:k},{type:12,data:a},{type:12,data:i.kvSequenceLength},{type:12,data:h}],Y=C&&n&&ze.size(n.dims)>0,he=["type","type"];Y&&he.push("type"),o&&he.push("type"),u&&he.push("type");let pe=[{dims:B,dataType:t.dataType,gpuDataType:0}];C&&pe.push({dims:z,dataType:t.dataType,gpuDataType:0});let Me=Oe=>{let De=qe("probs",t.dataType,t.dims),Ye=qe("v",s.dataType,s.dims),at=[De,Ye];Y&&at.push(qe("past_value",n.dataType,n.dims));let Pt=o?qe("seq_lens",o.dataType,o.dims):void 0;o&&at.push(Pt);let Xt=u?qe("total_sequence_length_input",u.dataType,u.dims):void 0;u&&at.push(Xt);let Zt=[It("output",t.dataType,B)];C&&Zt.push(It("present_value",t.dataType,z));let bt=[{name:"M",type:"u32"},{name:"K",type:"u32"},{name:"N",type:"u32"},{name:"num_heads",type:"u32"},{name:"head_size",type:"u32"},{name:"v_hidden_size",type:"u32"},{name:"past_sequence_length",type:"u32"},{name:"kv_sequence_length",type:"u32"},{name:"n_reps",type:"u32"}];return` + const TILE_SIZE = ${V}u; + var tileQ: array<${De.type.value}, ${V*V}>; + var tileV: array<${De.type.value}, ${V*V}>; + ${Oe.registerUniforms(bt).declareVariables(...at,...Zt)} + ${Oe.mainStart([V,V,1])} + let headIdx = workgroup_id.z % uniforms.num_heads; + let batchIdx = workgroup_id.z / uniforms.num_heads; + let kvHeadIdx = ${h===1?"headIdx":"headIdx / uniforms.n_reps"}; + let kv_num_heads = ${h===1?"uniforms.num_heads":"uniforms.num_heads / uniforms.n_reps"}; + let m = global_id.y; + let n = global_id.x; + let sequence_length = uniforms.M; + var total_sequence_length = uniforms.K; + ${ci(Pt,Xt,!0)} + let offsetA = workgroup_id.z * uniforms.M * uniforms.K + m * uniforms.K; + let absKvHeadIdx = batchIdx * kv_num_heads + kvHeadIdx; // kvHeadIdx is relative to the batch + ${Y&&C?"let pastValueOffset = absKvHeadIdx * uniforms.N * uniforms.past_sequence_length + n;":""}; + let vOffset = absKvHeadIdx * uniforms.N * uniforms.kv_sequence_length + n; + ${C?"let presentValueOffset = absKvHeadIdx * uniforms.N * uniforms.K + n;":""} + var value = ${De.type.storage}(0); + for (var w: u32 = 0u; w < uniforms.K; w += TILE_SIZE) { + if (m < uniforms.M && w + local_id.x < uniforms.K) { + tileQ[TILE_SIZE * local_id.y + local_id.x] = probs[offsetA + w + local_id.x]; + } + if (n < uniforms.N && w + local_id.y < uniforms.K) { + var idx = TILE_SIZE * local_id.y + local_id.x; + ${Y&&C?` + if (w + local_id.y < past_sequence_length) { + tileV[idx] = past_value[pastValueOffset + (w + local_id.y) * uniforms.N]; + } else if (w + local_id.y - past_sequence_length < uniforms.kv_sequence_length) { + tileV[idx] = v[vOffset + (w + local_id.y - past_sequence_length) * uniforms.N]; + } + `:` + if (w + local_id.y < uniforms.kv_sequence_length) { + tileV[idx] = v[vOffset + (w + local_id.y) * uniforms.N]; + }`} + ${C?` + if (w + local_id.y < present_sequence_length) { + present_value[presentValueOffset + (w + local_id.y) * uniforms.N] = tileV[idx]; + }`:""} + } + workgroupBarrier(); + for (var k: u32 = 0u; k < TILE_SIZE && w+k < total_sequence_length; k++) { + value += tileQ[TILE_SIZE * local_id.y + k] * tileV[TILE_SIZE * k + local_id.x]; + } + workgroupBarrier(); + } + + // we need to transpose output from BNSH_v to BSND_v + if (m < uniforms.M && n < uniforms.N) { + let outputIdx = batchIdx * uniforms.M * uniforms.v_hidden_size + m * uniforms.v_hidden_size + + headIdx * uniforms.N + n; + output[outputIdx] = value; + } + }`};return{name:"AttentionScore",shaderCache:{hint:`${n!==void 0};${e}`,inputDependencies:he},getRunData:()=>({outputs:pe,dispatchGroup:Z,programUniforms:ee}),getShaderSource:Me}},Nn=(e,t,s,n,i,a,o,u,p,h,k=void 0,C=void 0)=>{let d=Math.min(e.outputCount,1+(o?1:0)+(u?1:0)),z=d>1?h.pastSequenceLength:0,B=z+h.kvSequenceLength,V=p&&ze.size(p.dims)>0?p:void 0,Z=[t,s];d>1&&o&&ze.size(o.dims)>0&&Z.push(o),V&&Z.push(V),k&&Z.push(k),C&&Z.push(C);let ee=e.compute(ao(d,t,s,o,V,h,z,k,C),{inputs:Z,outputs:d>1?[-1,1]:[-1]})[0];e.compute(wl(ee,h.batchSize,h.numHeads,z,h.sequenceLength,B,k,C),{inputs:k&&C?[ee,k,C]:[ee],outputs:[]});let Y=[ee,n];d>1&&u&&ze.size(u.dims)>0&&Y.push(u),k&&Y.push(k),C&&Y.push(C),e.compute(yl(d,ee,n,u,h,z,k,C),{inputs:Y,outputs:d>1?[0,2]:[0]})},lo=(e,t)=>{let s=[t.batchSize,t.numHeads,t.sequenceLength,t.headSize],n=t.sequenceLength,i=t.inputHiddenSize,a=t.headSize,o=12,u={x:Math.ceil(t.headSize/o),y:Math.ceil(t.sequenceLength/o),z:t.batchSize*t.numHeads},p=[e.inputs[0],e.inputs[1],e.inputs[2]],h=[{type:12,data:n},{type:12,data:i},{type:12,data:a},{type:12,data:t.numHeads},{type:12,data:t.headSize},{type:12,data:t.hiddenSize},{type:12,data:t.hiddenSize+t.hiddenSize+t.vHiddenSize}],k=C=>{let d=It("output_q",p[0].dataType,s),z=It("output_k",p[0].dataType,s),B=It("output_v",p[0].dataType,s),V=qe("input",p[0].dataType,p[0].dims),Z=qe("weight",p[1].dataType,p[1].dims),ee=qe("bias",p[2].dataType,p[2].dims),Y=V.type.storage,he=[{name:"M",type:"u32"},{name:"K",type:"u32"},{name:"N",type:"u32"},{name:"num_heads",type:"u32"},{name:"head_size",type:"u32"},{name:"hidden_size",type:"u32"},{name:"ldb",type:"u32"}];return` + const TILE_SIZE = ${o}u; + var tileInput: array<${Y}, ${o*o}>; + var tileWeightQ: array<${Y}, ${o*o}>; + var tileWeightK: array<${Y}, ${o*o}>; + var tileWeightV: array<${Y}, ${o*o}>; + ${C.registerUniforms(he).declareVariables(V,Z,ee,d,z,B)} + ${C.mainStart([o,o,1])} + let batchIndex = workgroup_id.z / uniforms.num_heads; + let headNumber = workgroup_id.z % uniforms.num_heads; + let m = global_id.y; + let n = global_id.x; + + let inputOffset = batchIndex * (uniforms.M * uniforms.K) + m * uniforms.K; + let biasOffsetQ = headNumber * uniforms.head_size; + let biasOffsetK = uniforms.hidden_size + biasOffsetQ; + let biasOffsetV = uniforms.hidden_size + biasOffsetK; + + var valueQ = ${Y}(0); + var valueK = ${Y}(0); + var valueV = ${Y}(0); + for (var w: u32 = 0u; w < uniforms.K; w += TILE_SIZE) { + if (m < uniforms.M && w + local_id.x < uniforms.K) { + tileInput[TILE_SIZE * local_id.y + local_id.x] = input[inputOffset + w + local_id.x]; + } + if (n < uniforms.N && w + local_id.y < uniforms.K) { + let offset = n + (w + local_id.y) * uniforms.ldb; + tileWeightQ[TILE_SIZE * local_id.y + local_id.x] = weight[biasOffsetQ + offset]; + tileWeightK[TILE_SIZE * local_id.y + local_id.x] = weight[biasOffsetK + offset]; + tileWeightV[TILE_SIZE * local_id.y + local_id.x] = weight[biasOffsetV + offset]; + } + workgroupBarrier(); + for (var k: u32 = 0u; k({outputs:[{dims:s,dataType:e.inputs[0].dataType,gpuDataType:0},{dims:s,dataType:e.inputs[0].dataType,gpuDataType:0},{dims:s,dataType:e.inputs[0].dataType,gpuDataType:0}],dispatchGroup:u,programUniforms:h}),getShaderSource:k},{inputs:p,outputs:[-1,-1,-1]})},Ml=(e,t)=>{let s=oo(e.inputs,t),[n,i,a]=lo(e,s);return Nn(e,n,i,a,e.inputs[4],void 0,void 0,void 0,e.inputs[5],s)}}),co,bl,vl,po,Vc=g(()=>{We(),zt(),Ot(),rs(),Yt(),co=(e,t)=>{if(!e||e.length!==5)throw new Error("BatchNormalization requires 5 inputs");let s=(n,i,a)=>{let o=i.length;if(o!==n.length)throw new Error(`${a}: num dimensions != ${o}`);i.forEach((u,p)=>{if(u!==n[p])throw new Error(`${a}: dim[${p}] do not match`)})};if(e[0].dims.length>1){let n=t.format==="NHWC"?t.spatial?e[0].dims.slice(-1):e[0].dims.slice(-1).concat(e[0].dims.slice(1,e[0].dims.length-1)):e[0].dims.slice(1,t.spatial?2:void 0);s(e[1].dims,n,"Invalid input scale"),s(e[2].dims,n,"Invalid input B"),s(e[3].dims,n,"Invalid input mean"),s(e[4].dims,n,"Invalid input var")}else s(e[1].dims,[1],"Invalid input scale"),s(e[2].dims,[1],"Invalid input B"),s(e[3].dims,[1],"Invalid input mean"),s(e[4].dims,[1],"Invalid input var")},bl=(e,t)=>{let{epsilon:s,spatial:n,format:i}=t,a=e[0].dims,o=n?qt(a[a.length-1]):1,u=i==="NHWC"&&a.length>1?o:1,p=ze.size(a)/o,h=n,k=h?a.length:a,C=qe("x",e[0].dataType,e[0].dims,o),d=qe("scale",e[1].dataType,e[1].dims,u),z=qe("bias",e[2].dataType,e[2].dims,u),B=qe("inputMean",e[3].dataType,e[3].dims,u),V=qe("inputVar",e[4].dataType,e[4].dims,u),Z=It("y",e[0].dataType,k,o),ee=()=>{let he="";if(n)he=`let cOffset = ${a.length===1?"0u":i==="NHWC"?`outputIndices[${a.length-1}] / ${o}`:"outputIndices[1]"};`;else if(i==="NCHW")he=` + ${Z.indicesSet("outputIndices","0","0")} + let cOffset = ${Z.indicesToOffset("outputIndices")};`;else{he=`var cIndices = ${d.type.indices}(0); + cIndices[0] = outputIndices[${a.length-1}];`;for(let pe=1;pe` + const epsilon = ${s}; + ${he.registerUniform("outputSize","u32").declareVariables(C,d,z,B,V,Z)} + ${he.mainStart()} + ${he.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")} + var outputIndices = ${Z.offsetToIndices(`global_idx * ${o}`)}; + ${ee()} + let scale = ${d.getByOffset("cOffset")}; + let bias = ${z.getByOffset("cOffset")}; + let inputMean = ${B.getByOffset("cOffset")}; + let inputVar = ${V.getByOffset("cOffset")}; + let x = ${C.getByOffset("global_idx")}; + let value = (x - inputMean) * inverseSqrt(inputVar + epsilon) * scale + bias; + ${Z.setByOffset("global_idx","value")} + }`;return{name:"BatchNormalization",shaderCache:{hint:`${t.epsilon}_${t.format}_${n}_${o}`,inputDependencies:h?["rank","type","type","type","type"]:void 0},getShaderSource:Y,getRunData:()=>({outputs:[{dims:e[0].dims,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(p/64)},programUniforms:h?[{type:12,data:p},...yt(a)]:[{type:12,data:p}]})}},vl=e=>Bt(e),po=(e,t)=>{let{inputs:s,outputCount:n}=e,i=vl({...t,outputCount:n});if(O.webgpu.validateInputContent&&co(s,i),t.trainingMode)throw new Error("BatchNormalization trainingMode is not supported yet.");e.compute(bl(s,i))}}),xl,ho,Tl,Wc=g(()=>{Ot(),Yt(),xl=e=>{if(e[0].dims.length!==3)throw new Error("input should have 3 dimensions");if(![320,640,1280].includes(e[0].dims[2]))throw new Error("number of channels should be 320, 640 or 1280");if(e[1].dims.length!==1)throw new Error("bias is expected to have 1 dimensions");if(e[0].dims[2]!==e[1].dims[0])throw new Error("last dimension of input and bias are not the same")},ho=e=>{let t=e[0].dims,s=e[0].dims[2],n=ze.size(t)/4,i=e[0].dataType,a=qe("input",i,t,4),o=qe("bias",i,[s],4),u=qe("residual",i,t,4),p=It("output",i,t,4);return{name:"BiasAdd",getRunData:()=>({outputs:[{dims:t,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(n/64)}}),getShaderSource:h=>` + const channels = ${s}u / 4; + ${h.declareVariables(a,o,u,p)} + + ${h.mainStart()} + ${h.guardAgainstOutOfBoundsWorkgroupSizes(n)} + let value = ${a.getByOffset("global_idx")} + + ${o.getByOffset("global_idx % channels")} + ${u.getByOffset("global_idx")}; + ${p.setByOffset("global_idx","value")} + }`}},Tl=e=>{xl(e.inputs),e.compute(ho(e.inputs))}}),mo,ds,Pl,fo,El,Cl,_o,kl,Sl,go,$l,Al,wo,Il,Ol,yo,jn,Fl,pi,Dl,Mo,Ll,zl,bo,Bl,Rl,vo,Nl,jl,xo,Ul,Vl,To,Wl,Gl,hi,Kl,Po,mi,Hl,ql,Ql,Xl,Eo,Yl,Co=g(()=>{zt(),Ot(),rs(),Yt(),mo=(e,t,s,n,i,a,o)=>{let u=Math.ceil(t/4),p="";typeof i=="string"?p=`${i}(a)`:p=i("a");let h=qe("inputData",s,[u],4),k=It("outputData",n,[u],4),C=[{name:"vec_size",type:"u32"}];return o&&C.push(...o),` + ${e.registerUniforms(C).declareVariables(h,k)} + + ${a??""} + + ${e.mainStart()} + ${e.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.vec_size")} + + let a = ${h.getByOffset("global_idx")}; + ${k.setByOffset("global_idx",p)} + }`},ds=(e,t,s,n,i,a=e.dataType,o,u)=>{let p=[{type:12,data:Math.ceil(ze.size(e.dims)/4)}];return o&&p.push(...o),{name:t,shaderCache:{hint:i,inputDependencies:["type"]},getShaderSource:h=>mo(h,ze.size(e.dims),e.dataType,a,s,n,u),getRunData:h=>({outputs:[{dims:e.dims,dataType:a}],dispatchGroup:{x:Math.ceil(ze.size(h[0].dims)/64/4)},programUniforms:p})}},Pl=e=>{e.compute(ds(e.inputs[0],"Abs","abs"))},fo=e=>{e.compute(ds(e.inputs[0],"Acos","acos"))},El=e=>{e.compute(ds(e.inputs[0],"Acosh","acosh"))},Cl=e=>{e.compute(ds(e.inputs[0],"Asin","asin"))},_o=e=>{e.compute(ds(e.inputs[0],"Asinh","asinh"))},kl=e=>{e.compute(ds(e.inputs[0],"Atan","atan"))},Sl=e=>{e.compute(ds(e.inputs[0],"Atanh","atanh"))},go=e=>Bt(e),$l=(e,t)=>{let s;switch(t.to){case 10:s="vec4";break;case 1:s="vec4";break;case 12:s="vec4";break;case 6:s="vec4";break;case 9:s="vec4";break;default:throw new RangeError(`not supported type (specified in attribute 'to' from 'Cast' operator): ${t.to}`)}e.compute(ds(e.inputs[0],"Cast",s,void 0,t.cacheKey,t.to))},Al=e=>{let t,s,n=e.length>=2&&e[1].data!==0,i=e.length>=3&&e[2].data!==0;switch(e[0].dataType){case 1:t=n?e[1].getFloat32Array()[0]:-34028234663852886e22,s=i?e[2].getFloat32Array()[0]:34028234663852886e22;break;case 10:t=n?e[1].getUint16Array()[0]:64511,s=i?e[2].getUint16Array()[0]:31743;break;default:throw new Error("Unsupport data type")}return Bt({min:t,max:s})},wo=(e,t)=>{let s=t||Al(e.inputs),n=Ss(e.inputs[0].dataType);e.compute(ds(e.inputs[0],"Clip",i=>`clamp(${i}, vec4<${n}>(uniforms.min), vec4<${n}>(uniforms.max))`,void 0,s.cacheKey,void 0,[{type:e.inputs[0].dataType,data:s.min},{type:e.inputs[0].dataType,data:s.max}],[{name:"min",type:n},{name:"max",type:n}]),{inputs:[0]})},Il=e=>{e.compute(ds(e.inputs[0],"Ceil","ceil"))},Ol=e=>{e.compute(ds(e.inputs[0],"Cos","cos"))},yo=e=>{e.compute(ds(e.inputs[0],"Cosh","cosh"))},jn=e=>Bt(e),Fl=(e,t)=>{let s=Ss(e.inputs[0].dataType);e.compute(ds(e.inputs[0],"Elu",n=>`elu_vf32(${n})`,` + const elu_alpha_ = ${s}(${t.alpha}); + + fn elu_f32(a: ${s}) -> ${s} { + return select((exp(a) - 1.0) * elu_alpha_, a, a >= 0.0); + } + + fn elu_vf32(v: vec4<${s}>) -> vec4<${s}> { + return vec4(elu_f32(v.x), elu_f32(v.y), elu_f32(v.z), elu_f32(v.w)); + }`,t.cacheKey))},pi=(e="f32")=>` +const r0: ${e} = 0.3275911; +const r1: ${e} = 0.254829592; +const r2: ${e} = -0.284496736; +const r3: ${e} = 1.421413741; +const r4: ${e} = -1.453152027; +const r5: ${e} = 1.061405429; + +fn erf_vf32(v: vec4<${e}>) -> vec4<${e}> { + let absv = abs(v); + let x = 1.0 / (1.0 + r0 * absv); + return sign(v) * (1.0 - ((((r5 * x + r4) * x + r3) * x + r2) * x + r1) * x * exp(-absv * absv)); +}`,Dl=e=>{let t=Ss(e.inputs[0].dataType);e.compute(ds(e.inputs[0],"Erf",s=>`erf_vf32(${s})`,pi(t)))},Mo=e=>{e.compute(ds(e.inputs[0],"Exp","exp"))},Ll=e=>{e.compute(ds(e.inputs[0],"Floor","floor"))},zl=e=>{let t=Ss(e.inputs[0].dataType);e.compute(ds(e.inputs[0],"Gelu",s=>`0.5 * ${s} * (1.0 + erf_vf32(${s} * 0.7071067811865475))`,pi(t)))},bo=(e,t)=>{let s=Ss(e.inputs[0].dataType);e.compute(ds(e.inputs[0],"LeakyRelu",n=>`select(leaky_relu_alpha_ * ${n}, ${n}, ${n} >= vec4<${s}>(0.0))`,`const leaky_relu_alpha_ = ${s}(${t.alpha});`,t.cacheKey))},Bl=e=>{e.compute(ds(e.inputs[0],"Not",t=>`!${t}`))},Rl=e=>{e.compute(ds(e.inputs[0],"Neg",t=>`-${t}`))},vo=e=>{e.compute(ds(e.inputs[0],"Reciprocal",t=>`1.0/${t}`))},Nl=e=>{let t=Ss(e.inputs[0].dataType);e.compute(ds(e.inputs[0],"Relu",s=>`select(vec4<${t}>(0.0), ${s}, ${s} > vec4<${t}>(0.0))`))},jl=e=>{e.compute(ds(e.inputs[0],"Sigmoid",t=>`(1.0 / (1.0 + exp(-${t})))`))},xo=e=>Bt(e),Ul=(e,t)=>{let s=Ss(e.inputs[0].dataType);e.compute(ds(e.inputs[0],"HardSigmoid",n=>`max(vec4<${s}>(0.0), min(vec4<${s}>(1.0), ${t.alpha} * ${n} + vec4<${s}>(${t.beta})))`,void 0,t.cacheKey))},Vl=e=>{e.compute(ds(e.inputs[0],"Sin","sin"))},To=e=>{e.compute(ds(e.inputs[0],"Sinh","sinh"))},Wl=e=>{e.compute(ds(e.inputs[0],"Sqrt","sqrt"))},Gl=e=>{e.compute(ds(e.inputs[0],"Tan","tan"))},hi=e=>`sign(${e}) * (1 - exp(-2 * abs(${e}))) / (1 + exp(-2 * abs(${e})))`,Kl=e=>{e.compute(ds(e.inputs[0],"Tanh",hi))},Po=(e="f32")=>` +const fast_gelu_a: ${e} = 0.5; +const fast_gelu_b: ${e} = 0.7978845608028654; +const fast_gelu_c: ${e} = 0.035677408136300125; + +fn tanh_v(v: vec4<${e}>) -> vec4<${e}> { + return ${hi("v")}; +} +`,mi=e=>`(fast_gelu_a + fast_gelu_a * tanh_v(${e} * (fast_gelu_c * ${e} * ${e} + fast_gelu_b))) * ${e}`,Hl=e=>{let t=Ss(e.inputs[0].dataType);e.compute(ds(e.inputs[0],"FastGelu",mi,Po(t),void 0,e.inputs[0].dataType))},ql=(e,t)=>{let s=Ss(e.inputs[0].dataType);return e.compute(ds(e.inputs[0],"ThresholdedRelu",n=>`select(vec4<${s}>(0.0), ${n}, ${n} > thresholded_relu_alpha_)`,`const thresholded_relu_alpha_ = vec4<${s}>(${t.alpha});`,t.cacheKey)),0},Ql=e=>{e.compute(ds(e.inputs[0],"Log","log"))},Xl=(e,t)=>` +const alpha = vec4<${e}>(${t}); +const one = ${e}(1.0); +const zero = ${e}(0.0); + +fn quick_gelu_impl(x: vec4<${e}>) -> vec4<${e}> { + let v = x *alpha; + var x1 : vec4<${e}>; + for (var i = 0; i < 4; i = i + 1) { + if (v[i] >= zero) { + x1[i] = one / (one + exp(-v[i])); + } else { + x1[i] = one - one / (one + exp(v[i])); + } + } + return x * x1; +} +`,Eo=e=>`quick_gelu_impl(${e})`,Yl=(e,t)=>{let s=Ss(e.inputs[0].dataType);e.compute(ds(e.inputs[0],"QuickGelu",Eo,Xl(s,t.alpha),t.cacheKey,e.inputs[0].dataType))}}),Jl,Zl,ko,Gc=g(()=>{Ot(),Yt(),Co(),Jl=e=>{if(e[0].dims.length!==3)throw new Error("input should have 3 dimensions");if(![2560,5120,10240].includes(e[0].dims[2]))throw new Error("hidden state should be 2560, 5120 or 10240");if(e[1].dims.length!==1)throw new Error("bias is expected to have 1 dimensions");if(e[0].dims[2]!==e[1].dims[0])throw new Error("last dimension of input and bias are not the same")},Zl=e=>{let t=e[0].dims.slice();t[2]=t[2]/2;let s=qe("input",e[0].dataType,e[0].dims,4),n=qe("bias",e[0].dataType,[e[0].dims[2]],4),i=It("output",e[0].dataType,t,4),a=ze.size(t)/4,o=fs(e[0].dataType);return{name:"BiasSplitGelu",getRunData:()=>({outputs:[{dims:t,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(a/64)}}),getShaderSource:u=>` + const M_SQRT2 = sqrt(2.0); + const halfChannels = ${e[0].dims[2]/4/2}u; + + ${u.declareVariables(s,n,i)} + + ${pi(o)} + + ${u.mainStart()} + ${u.guardAgainstOutOfBoundsWorkgroupSizes(a)} + let biasIdx = global_idx % halfChannels; + let batchIndex = global_idx / halfChannels; + let inputOffset = biasIdx + batchIndex * halfChannels * 2; + let valueLeft = input[inputOffset] + bias[biasIdx]; + let valueRight = input[inputOffset + halfChannels] + bias[biasIdx + halfChannels]; + let geluRight = valueRight * 0.5 * (erf_vf32(valueRight / M_SQRT2) + 1); + + ${i.setByOffset("global_idx","valueLeft * geluRight")} + }`}},ko=e=>{Jl(e.inputs),e.compute(Zl(e.inputs))}}),eu,tu,Mr,So,su,ru,nu,iu,$o,ou,au,Ao,lu,Kc=g(()=>{zt(),Ot(),Yt(),eu=(e,t,s,n,i,a,o,u,p,h,k,C)=>{let d,z;typeof u=="string"?d=z=(Y,he)=>`${u}((${Y}),(${he}))`:typeof u=="function"?d=z=u:(d=u.scalar,z=u.vector);let B=It("outputData",k,n.length,4),V=qe("aData",p,t.length,4),Z=qe("bData",h,s.length,4),ee;if(i)if(a){let Y=ze.size(t)===1,he=ze.size(s)===1,pe=t.length>0&&t[t.length-1]%4===0,Me=s.length>0&&s[s.length-1]%4===0;Y||he?ee=B.setByOffset("global_idx",z(Y?`${V.type.value}(${V.getByOffset("0")}.x)`:V.getByOffset("global_idx"),he?`${Z.type.value}(${Z.getByOffset("0")}.x)`:Z.getByOffset("global_idx"))):ee=` + let outputIndices = ${B.offsetToIndices("global_idx * 4u")}; + let offsetA = ${V.broadcastedIndicesToOffset("outputIndices",B)}; + let offsetB = ${Z.broadcastedIndicesToOffset("outputIndices",B)}; + ${B.setByOffset("global_idx",z(o||pe?V.getByOffset("offsetA / 4u"):`${V.type.value}(${V.getByOffset("offsetA / 4u")}[offsetA % 4u])`,o||Me?Z.getByOffset("offsetB / 4u"):`${Z.type.value}(${Z.getByOffset("offsetB / 4u")}[offsetB % 4u])`))} + `}else ee=B.setByOffset("global_idx",z(V.getByOffset("global_idx"),Z.getByOffset("global_idx")));else{if(!a)throw new Error("no necessary to use scalar implementation for element-wise binary op implementation.");let Y=(he,pe,Me="")=>{let Oe=`aData[indexA${pe}][componentA${pe}]`,De=`bData[indexB${pe}][componentB${pe}]`;return` + let outputIndices${pe} = ${B.offsetToIndices(`global_idx * 4u + ${pe}u`)}; + let offsetA${pe} = ${V.broadcastedIndicesToOffset(`outputIndices${pe}`,B)}; + let offsetB${pe} = ${Z.broadcastedIndicesToOffset(`outputIndices${pe}`,B)}; + let indexA${pe} = offsetA${pe} / 4u; + let indexB${pe} = offsetB${pe} / 4u; + let componentA${pe} = offsetA${pe} % 4u; + let componentB${pe} = offsetB${pe} % 4u; + ${he}[${pe}] = ${Me}(${d(Oe,De)}); + `};k===9?ee=` + var data = vec4(0); + ${Y("data",0,"u32")} + ${Y("data",1,"u32")} + ${Y("data",2,"u32")} + ${Y("data",3,"u32")} + outputData[global_idx] = dot(vec4(0x1, 0x100, 0x10000, 0x1000000), vec4(data));`:ee=` + ${Y("outputData[global_idx]",0)} + ${Y("outputData[global_idx]",1)} + ${Y("outputData[global_idx]",2)} + ${Y("outputData[global_idx]",3)} + `}return` + ${e.registerUniform("vec_size","u32").declareVariables(V,Z,B)} + + ${C??""} + + ${e.mainStart()} + ${e.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.vec_size")} + ${ee} + }`},tu=(e,t,s,n,i,a,o=s.dataType)=>{let u=s.dims.map(V=>Number(V)??1),p=n.dims.map(V=>Number(V)??1),h=!ze.areEqual(u,p),k=u,C=ze.size(u),d=!1,z=!1,B=[h];if(h){let V=Ws.calcShape(u,p,!1);if(!V)throw new Error("Can't perform binary op on the given tensors");k=V.slice(),C=ze.size(k);let Z=ze.size(u)===1,ee=ze.size(p)===1,Y=u.length>0&&u[u.length-1]%4===0,he=p.length>0&&p[p.length-1]%4===0;B.push(Z),B.push(ee),B.push(Y),B.push(he);let pe=1;for(let Me=1;MeV.toString()).join("_"),inputDependencies:["rank","rank"]},getShaderSource:V=>eu(V,u,p,k,d,h,z,i,s.dataType,n.dataType,o,a),getRunData:()=>({outputs:[{dims:k,dataType:o}],dispatchGroup:{x:Math.ceil(C/64/4)},programUniforms:[{type:12,data:Math.ceil(ze.size(k)/4)},...yt(u,p,k)]})}},Mr=(e,t,s,n,i,a)=>{e.compute(tu(t,i??"",e.inputs[0],e.inputs[1],s,n,a))},So=e=>{Mr(e,"Add",(t,s)=>`${t}+${s}`)},su=e=>{Mr(e,"Div",(t,s)=>`${t}/${s}`)},ru=e=>{Mr(e,"Equal",{scalar:(t,s)=>`u32(${t}==${s})`,vector:(t,s)=>`vec4(${t}==${s})`},void 0,void 0,9)},nu=e=>{Mr(e,"Mul",(t,s)=>`${t}*${s}`)},iu=e=>{let t=qe("input",e.inputs[0].dataType,e.inputs[0].dims).type.value;Mr(e,"Pow",{scalar:(s,n)=>`pow_custom(${s},${n})`,vector:(s,n)=>`pow_vector_custom(${s},${n})`},` + fn pow_custom(a : ${t}, b : ${t}) -> ${t} { + if (b == ${t}(0.0)) { + return ${t}(1.0); + } else if (a < ${t}(0.0) && f32(b) != floor(f32(b))) { + return ${t}(pow(f32(a), f32(b))); // NaN + } + return select(sign(a), ${t}(1.0), round(f32(abs(b) % ${t}(2.0))) != 1.0) * ${t}(${t==="i32"?"round":""}(pow(f32(abs(a)), f32(b)))); + } + fn pow_vector_custom(a : vec4<${t}>, b : vec4<${t}>) -> vec4<${t}> { + // TODO: implement vectorized pow + return vec4<${t}>(pow_custom(a.x, b.x), pow_custom(a.y, b.y), pow_custom(a.z, b.z), pow_custom(a.w, b.w)); + } + `)},$o=e=>{Mr(e,"Sub",(t,s)=>`${t}-${s}`)},ou=e=>{Mr(e,"Greater",{scalar:(t,s)=>`u32(${t}>${s})`,vector:(t,s)=>`vec4(${t}>${s})`},void 0,void 0,9)},au=e=>{Mr(e,"Less",{scalar:(t,s)=>`u32(${t}<${s})`,vector:(t,s)=>`vec4(${t}<${s})`},void 0,void 0,9)},Ao=e=>{Mr(e,"GreaterOrEqual",{scalar:(t,s)=>`u32(${t}>=${s})`,vector:(t,s)=>`vec4(${t}>=${s})`},void 0,void 0,9)},lu=e=>{Mr(e,"LessOrEqual",{scalar:(t,s)=>`u32(${t}<=${s})`,vector:(t,s)=>`vec4(${t}<=${s})`},void 0,void 0,9)}}),Io,uu,du,Oo,cu,pu,hu=g(()=>{zt(),Ot(),rs(),Yt(),Io=(e,t)=>{if(!e||e.length<1)throw new Error("too few inputs");let s=0,n=e[s],i=n.dataType,a=n.dims.length;e.forEach((o,u)=>{if(u!==s){if(o.dataType!==i)throw new Error("input tensors should be one type");if(o.dims.length!==a)throw new Error("input tensors should have the same shape");o.dims.forEach((p,h)=>{if(h!==t&&p!==n.dims[h])throw new Error("non concat dimensions must match")})}})},uu=(e,t)=>` + fn calculateInputIndex(index: u32) -> u32 { + let sizeInConcatAxis = array(${t}); + for (var i: u32 = 0u; i < ${e}; i += 1u ) { + if (index < sizeInConcatAxis[i]) { + return i; + } + } + return ${e}u; + }`,du=(e,t)=>{let s=e.length,n=[];for(let i=0;i{let i=ze.size(s),a=new Array(e.length),o=new Array(e.length),u=0,p=[],h=[],k=[{type:12,data:i}];for(let V=0;V`uniforms.sizeInConcatAxis${V}`).join(","),B=V=>` + + ${(()=>{V.registerUniform("outputSize","u32");for(let Z=0;Z(${z}); + ${d} -= sizeInConcatAxis[inputIndex - 1u]; + } + + ${du(o,C)} + }`;return{name:"Concat",shaderCache:{hint:`${t}`,inputDependencies:p},getRunData:()=>({outputs:[{dims:s,dataType:n}],dispatchGroup:{x:Math.ceil(i/64)},programUniforms:k}),getShaderSource:B}},cu=(e,t)=>{let s=e.inputs,n=s[0].dims,i=ze.normalizeAxis(t.axis,n.length);Io(s,i);let a=n.slice();a[i]=s.reduce((u,p)=>u+(p.dims.length>i?p.dims[i]:0),0);let o=s.filter(u=>ze.size(u.dims)>0);e.compute(Oo(o,i,a,s[0].dataType),{inputs:o})},pu=e=>Bt({axis:e.axis})}),nn,on,Dr,Fo,an=g(()=>{zt(),Ot(),nn=(e,t,s="f32")=>{switch(e.activation){case"Relu":return`value = max(value, ${t}(0.0));`;case"Sigmoid":return`value = (${t}(1.0) / (${t}(1.0) + exp(-value)));`;case"Clip":return`value = clamp(value, ${t}(${s}(uniforms.clip_min)), ${t}(${s}(uniforms.clip_max)));`;case"HardSigmoid":return`value = max(${t}(0.0), min(${t}(1.0), ${s}(uniforms.alpha) * value + ${s}(uniforms.beta)));`;case"LeakyRelu":return`value = select(${s}(uniforms.alpha) * value, value, value >= ${t}(0.0));`;case"Tanh":return`let e2x = exp(-2.0 * abs(value)); + value = sign(value) * (1.0 - e2x) / (1.0 + e2x); + `;case"":return"";default:throw new Error(`Unsupported activation ${e.activation}`)}},on=(e,t)=>{e.activation==="Clip"?t.push({type:1,data:e.clipMax},{type:1,data:e.clipMin}):e.activation==="HardSigmoid"?t.push({type:1,data:e.alpha},{type:1,data:e.beta}):e.activation==="LeakyRelu"&&t.push({type:1,data:e.alpha})},Dr=(e,t)=>{e.activation==="Clip"?t.push({name:"clip_max",type:"f32"},{name:"clip_min",type:"f32"}):e.activation==="HardSigmoid"?t.push({name:"alpha",type:"f32"},{name:"beta",type:"f32"}):e.activation==="LeakyRelu"&&t.push({name:"alpha",type:"f32"})},Fo=e=>{let t=(e==null?void 0:e.activation)||"";if(t==="HardSigmoid"){let[s,n]=(e==null?void 0:e.activation_params)||[.2,.5];return{activation:t,alpha:s,beta:n}}else if(t==="Clip"){let[s,n]=(e==null?void 0:e.activation_params)||[ks,Xs];return{activation:t,clipMax:n,clipMin:s}}else if(t==="LeakyRelu"){let[s]=(e==null?void 0:e.activation_params)||[.01];return{activation:t,alpha:s}}return{activation:t}}}),Ks,Do,Lo=g(()=>{Ks=(e,t)=>{switch(e){case 1:return t;case 2:return`vec2<${t}>`;case 3:return`vec3<${t}>`;case 4:return`vec4<${t}>`;default:throw new Error(`${e}-component is not supported.`)}},Do=e=>` + ${e?"value = value + getBiasByOutputCoords(coords);":""} + `}),zo,Hc=g(()=>{zo=e=>` +fn getIndexFromCoords4D(coords : vec4, shape : vec4) -> i32 { + return dot(coords, vec4( + shape.y * shape.z * shape.w, shape.z * shape.w, shape.w, 1)); +} +fn getOutputIndexFromCoords(coords : vec4) -> i32 { + return dot(coords, vec4( + i32(${e}.x), i32(${e}.y), i32(${e}.z), 1)); +} +`}),Un,Bo,fi=g(()=>{zt(),Ot(),Yt(),an(),Un=(e,t,s,n,i)=>{let a=n-s;return` + ${Array.from({length:s}).map((o,u)=>` + if (${$t(t.shape,u,t.rank)} != 1) { + ${t.indicesSet(e,u,$t(i,u+a,n))} + } else { + ${t.indicesSet(e,u,0)} + }`).join("")} +`},Bo=(e,t,s,n,i=!1,a)=>{let o=e[0].dims,u=e[1].dims,p=o[o.length-2],h=u[u.length-1],k=o[o.length-1],C=qt(h),d=qt(k),z=qt(p),B=ze.size(s)/C/z,V=e.length>2,Z=n?n.slice(0,-2):s.slice(0,-2),ee=[ze.size(Z),p,h],Y=[{type:12,data:B},{type:12,data:p},{type:12,data:h},{type:12,data:k}];on(t,Y),Y.push(...yt(Z,o,u)),V&&Y.push(...yt(e[2].dims)),Y.push(...yt(ee));let he=pe=>{let Me=Ui("batch_dims",e[0].dataType,Z.length),Oe=qe("a",e[0].dataType,o.length,d),De=qe("b",e[1].dataType,u.length,C),Ye=It("output",e[0].dataType,ee.length,C),at=fs(Ye.type.tensor),Pt=nn(t,Ye.type.value,at),Xt=[Oe,De],Zt="";if(V){let St=i?C:1;Xt.push(qe("bias",e[2].dataType,e[2].dims.length,St)),Zt=`${i?`value += bias[col / ${St}];`:`value += ${Ye.type.value}(bias[row + i]);`}`}let bt=[{name:"output_size",type:"u32"},{name:"M",type:"u32"},{name:"N",type:"u32"},{name:"K",type:"u32"}];Dr(t,bt);let ss=()=>{let St=`var a_data: ${Oe.type.value};`;for(let Ft=0;Ft; + for (var k: u32 = 0u; k < uniforms.K; k = k + ${d}) { + ${ss()} + } + for (var i = 0u; i < ${z}u; i++) { + var value = values[i]; + ${Zt} + ${Pt} + let cur_indices = ${Ye.type.indices}(batch, row + i, col); + let offset = ${Ye.indicesToOffset("cur_indices")}; + ${Ye.setByOffset(`offset / ${C}`,"value")}; + } + } + `};return{name:"MatMulNaive",shaderCache:{hint:`${t.activation};${C};${d};${z};${i}`,inputDependencies:V?["rank","rank","rank"]:["rank","rank"]},getRunData:()=>({outputs:[{dims:a?a(s):s,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(B/64)},programUniforms:Y}),getShaderSource:he}}}),mu,fu,Ro,_i,_u,No,jo,gi,Uo=g(()=>{zt(),Ot(),Yt(),an(),fi(),Lo(),mu=(e,t)=>e?` + mm_Asub[inputRow][inputCol] = mm_readA(batch, + kStart + inputRow, + globalRowStart / innerElementSize + inputCol${t?", batchIndices":""}); + `:` + mm_Asub[inputRow][inputCol] = mm_readA(batch, + globalRow + innerRow, + kStart / innerElementSize + inputCol${t?", batchIndices":""}); + `,fu=(e,t)=>e?` + let ACached0 = mm_Asub[k * innerElementSize][localRow]; + let ACached1 = mm_Asub[k * innerElementSize + 1][localRow]; + let ACached2 = mm_Asub[k * innerElementSize + 2][localRow]; + ${t===3?"":"let ACached3 = mm_Asub[k * innerElementSize + 3][localRow];"} + for (var i = 0; i < rowPerThread; i = i + 1) { + acc[i] = BCached0 * ACached0[i] + acc[i]; + acc[i] = BCached1 * ACached1[i] + acc[i]; + acc[i] = BCached2 * ACached2[i] + acc[i]; + ${t===3?"":"acc[i] = BCached3 * ACached3[i] + acc[i];"} + }`:` + for (var i = 0; i < rowPerThread; i = i + 1) { + let ACached = mm_Asub[tileRow + i][k]; + acc[i] = BCached0 * ACached.x + acc[i]; + acc[i] = BCached1 * ACached.y + acc[i]; + acc[i] = BCached2 * ACached.z + acc[i]; + ${t===3?"":"acc[i] = BCached3 * ACached.w + acc[i];"} + }`,Ro=(e,t,s="f32",n,i=!1,a=32,o=!1,u=32)=>{let p=t[1]*e[1],h=t[0]*e[0],k=i?p:a,C=i?a:p,d=k/t[0],z=a/t[1];if(!((i&&d===4&&e[1]===4||!i&&(d===3||d===4))&&k%t[0]===0&&a%t[1]===0&&e[0]===4))throw new Error(`If transposeA ${i} is true, innerElementSize ${d} and workPerThread[1] ${e[1]} must be 4. + Otherwise, innerElementSize ${d} must be 3 or 4. + tileAWidth ${k} must be divisible by workgroupSize[0]${t[0]}. tileInner ${a} must be divisible by workgroupSize[1] ${t[1]}. colPerThread ${e[0]} must be 4.`);return` +var mm_Asub: array, ${k/d}>, ${C}>; +var mm_Bsub: array, ${h/e[0]}>, ${a}>; + +const rowPerThread = ${e[1]}; +const colPerThread = ${e[0]}; +const innerElementSize = ${d}; +const tileInner = ${a}; + +@compute @workgroup_size(${t[0]}, ${t[1]}, ${t[2]}) +fn main(@builtin(local_invocation_id) localId : vec3, + @builtin(global_invocation_id) globalId : vec3, + @builtin(workgroup_id) workgroupId : vec3) { + let localRow = i32(localId.y); + let tileRow = localRow * rowPerThread; + let tileCol = i32(localId.x); + + let globalRow =i32(globalId.y) * rowPerThread; + let globalCol = i32(globalId.x); + let batch = ${o?"0":"i32(globalId.z)"}; + ${n?`let batchIndices = ${n.offsetToIndices("u32(batch)")};`:""} + let globalRowStart = i32(workgroupId.y) * ${p}; + + let num_tiles = ${o?`${Math.ceil(u/a)}`:"(uniforms.dim_inner - 1) / tileInner + 1"}; + var kStart = ${o?`i32(globalId.z) * ${u}`:"0"}; + + var acc: array, rowPerThread>; + + // Loop over shared dimension. + let tileRowB = localRow * ${z}; + for (var t = 0; t < num_tiles; t = t + 1) { + // Load one tile of A into local memory. + for (var innerRow = 0; innerRow < rowPerThread; innerRow = innerRow + 1) { + let inputRow = tileRow + innerRow; + let inputCol = tileCol; + ${mu(i,n)} + } + + // Load one tile of B into local memory. + for (var innerRow = 0; innerRow < ${z}; innerRow = innerRow + 1) { + let inputRow = tileRowB + innerRow; + let inputCol = tileCol; + mm_Bsub[inputRow][inputCol] = mm_readB(batch, kStart + inputRow, globalCol${n?", batchIndices":""}); + } + kStart = kStart + tileInner; + workgroupBarrier(); + + // Compute acc values for a single thread. + for (var k = 0; k < tileInner / innerElementSize; k = k + 1) { + let BCached0 = mm_Bsub[k * innerElementSize][tileCol]; + let BCached1 = mm_Bsub[k * innerElementSize + 1][tileCol]; + let BCached2 = mm_Bsub[k * innerElementSize + 2][tileCol]; + ${d===3?"":"let BCached3 = mm_Bsub[k * innerElementSize + 3][tileCol];"} + + ${fu(i,d)} + } + + workgroupBarrier(); + } + + for (var innerRow = 0; innerRow < rowPerThread; innerRow = innerRow + 1) { + mm_write(batch, globalRow + innerRow, globalCol, acc[innerRow]); + } +}`},_i=(e,t)=>e?` + mm_Asub[inputRow][inputCol] = mm_readA(batch, + kStart + inputRow, + globalRowStart + inputCol${t?", batchIndices":""}); + `:` + mm_Asub[inputRow][inputCol] = mm_readA(batch, + globalRowStart + inputRow, + kStart + inputCol${t?", batchIndices":""}); + `,_u=e=>e?"let ACached = mm_Asub[k][tileRow + innerRow];":"let ACached = mm_Asub[tileRow + innerRow][k];",No=(e,t,s="f32",n,i=!1,a=32,o=!1,u=32,p=!1)=>{let h=e[1]*t[1],k=e[0]*t[0],C=i?h:a,d=i?a:h;if(!(d%t[1]===0&&C%t[0]===0&&a%t[1]===0))throw new Error(`tileAHight ${d} must be divisible by workgroupSize[1]${t[1]}, tileAWidth ${C} must be divisible by workgroupSize[0]${t[0]}, tileInner ${a} must be divisible by workgroupSize[1]${t[1]}`);let z=d/t[1],B=C/t[0],V=a/t[1],Z=p?` + let localRow = i32(localId.y); + let localCol = i32(localId.x); + let globalRowStart = i32(workgroupId.y) * ${h}; + let globalColStart = i32(workgroupId.x) * ${k}; + + // Loop over shared dimension. + for (var t = 0; t < num_tiles; t = t + 1) { + // Load one tile of A into local memory. + for (var inputRow = localRow; inputRow < ${d}; inputRow = inputRow + ${t[1]}) { + for (var inputCol = localCol; inputCol < ${C}; inputCol = inputCol + ${t[0]}) { + ${_i(i,n)} + } + } + // Load one tile of B into local memory. + for (var inputRow = localRow; inputRow < ${a}; inputRow = inputRow + ${t[1]}) { + for (var inputCol = localCol; inputCol < ${k}; inputCol = inputCol + ${t[0]}) { + mm_Bsub[inputRow][inputCol] = mm_readB(batch, + kStart + inputRow, + globalColStart + inputCol${n?", batchIndices":""}); + } + } + kStart = kStart + tileInner; + workgroupBarrier(); + + // Compute acc values for a single thread. + var BCached : array<${s}, colPerThread>; + for (var k = 0; k < tileInner; k = k + 1) { + for (var inner = 0; inner < colPerThread; inner = inner + 1) { + BCached[inner] = mm_Bsub[k][localCol + inner * ${t[0]}]; + } + for (var innerRow = 0; innerRow < rowPerThread; innerRow = innerRow + 1) { + let ACached = ${i?`mm_Asub[k][localRow + innerRow * ${t[1]}];`:`mm_Asub[localRow + innerRow * ${t[1]}][k];`} + for (var innerCol = 0; innerCol < colPerThread; innerCol = innerCol + 1) { + acc[innerRow][innerCol] = acc[innerRow][innerCol] + + ACached * BCached[innerCol]; + } + } + } + workgroupBarrier(); + } + for (var innerRow = 0; innerRow < rowPerThread; innerRow = innerRow + 1) { + let gRow = globalRowStart + localRow + innerRow * ${t[1]}; + for (var innerCol = 0; innerCol < colPerThread; innerCol = innerCol + 1) { + let gCol = globalColStart + localCol + innerCol * ${t[0]}; + mm_write(batch, gRow, gCol, acc[innerRow][innerCol]); + } + } + `:` +let tileRow = i32(localId.y) * rowPerThread; +let tileCol = i32(localId.x) * colPerThread; + +let globalRow = i32(globalId.y) * rowPerThread; +let globalCol = i32(globalId.x) * colPerThread; +let globalRowStart = i32(workgroupId.y) * ${h}; + +let tileRowA = i32(localId.y) * ${z}; +let tileColA = i32(localId.x) * ${B}; +let tileRowB = i32(localId.y) * ${V}; +// Loop over shared dimension. +for (var t = 0; t < num_tiles; t = t + 1) { + // Load one tile of A into local memory. + for (var innerRow = 0; innerRow < ${z}; innerRow = innerRow + 1) { + for (var innerCol = 0; innerCol < ${B}; innerCol = innerCol + 1) { + let inputRow = tileRowA + innerRow; + let inputCol = tileColA + innerCol; + ${_i(i,n)} + } + } + + // Load one tile of B into local memory. + for (var innerRow = 0; innerRow < ${V}; innerRow = innerRow + 1) { + for (var innerCol = 0; innerCol < colPerThread; innerCol = innerCol + 1) { + let inputRow = tileRowB + innerRow; + let inputCol = tileCol + innerCol; + mm_Bsub[inputRow][inputCol] = mm_readB(batch, + kStart + inputRow, + globalCol + innerCol${n?", batchIndices":""}); + } + } + kStart = kStart + tileInner; + workgroupBarrier(); + + // Compute acc values for a single thread. + var BCached : array<${s}, colPerThread>; + for (var k = 0; k < tileInner; k = k + 1) { + for (var inner = 0; inner < colPerThread; inner = inner + 1) { + BCached[inner] = mm_Bsub[k][tileCol + inner]; + } + + for (var innerRow = 0; innerRow < rowPerThread; innerRow = innerRow + 1) { + ${_u(i)} + for (var innerCol = 0; innerCol < colPerThread; innerCol = innerCol + 1) { + acc[innerRow][innerCol] = acc[innerRow][innerCol] + ACached * BCached[innerCol]; + } + } + } + + workgroupBarrier(); +} + +for (var innerRow = 0; innerRow < rowPerThread; innerRow = innerRow + 1) { + for (var innerCol = 0; innerCol < colPerThread; innerCol = innerCol + 1) { + mm_write(batch, globalRow + innerRow, globalCol + innerCol, + acc[innerRow][innerCol]); + } +} +`;return` + var mm_Asub : array, ${d}>; + var mm_Bsub : array, ${a}>; + const rowPerThread = ${e[1]}; + const colPerThread = ${e[0]}; + const tileInner = ${a}; + +@compute @workgroup_size(${t[0]}, ${t[1]}, ${t[2]}) +fn main(@builtin(local_invocation_id) localId : vec3, + @builtin(global_invocation_id) globalId : vec3, + @builtin(workgroup_id) workgroupId : vec3) { + let batch = ${o?"0":"i32(globalId.z)"}; + ${n?`let batchIndices = ${n.offsetToIndices("u32(batch)")};`:""} + let num_tiles = ${o?`${Math.ceil(u/a)}`:"(uniforms.dim_inner - 1) / tileInner + 1"}; + var kStart = ${o?`i32(globalId.z) * ${u}`:"0"}; + + var acc : array, rowPerThread>; + ${Z} + } +`},jo=(e,t,s,n,i=!1)=>{let[a,o,u,p]=n,h=fs(n[0].type.tensor);return` + fn mm_readA(batch: i32, row: i32, colIn: i32, batchIndices: ${a.type.indices}) -> ${Ks(e,h)} { + var value = ${Ks(e,h)}(0.0); + let col = colIn * ${e}; + if(row < uniforms.dim_a_outer && col < uniforms.dim_inner) + { + var aIndices: ${o.type.indices}; + ${Un("aIndices",o,o.rank-2,a.rank,"batchIndices")} + ${o.indicesSet("aIndices",o.rank-2,"u32(row)")} + ${o.indicesSet("aIndices",o.rank-1,"u32(colIn)")} + value = ${o.getByIndices("aIndices")}; + } + return value; + } + + fn mm_readB(batch: i32, row: i32, colIn: i32, batchIndices: ${a.type.indices}) -> ${Ks(e,h)} { + var value = ${Ks(e,h)}(0.0); + let col = colIn * ${e}; + if(row < uniforms.dim_inner && col < uniforms.dim_b_outer) + { + var bIndices: ${u.type.indices}; + ${Un("bIndices",u,u.rank-2,a.rank,"batchIndices")} + ${u.indicesSet("bIndices",u.rank-2,"u32(row)")} + ${u.indicesSet("bIndices",u.rank-1,"u32(colIn)")} + value = ${u.getByIndices("bIndices")}; + } + return value; + } + + fn mm_write(batch: i32, row: i32, colIn: i32, valueIn: ${Ks(e,h)}) { + let col = colIn * ${e}; + if (row < uniforms.dim_a_outer && col < uniforms.dim_b_outer) { + var value = valueIn; + let coords = vec3(batch, row, colIn); + ${t?`value = value + ${i?"bias[colIn]":`${Ks(e,h)}(bias[row])`};`:""} + ${s} + ${p.setByIndices("vec3(coords)","value")} + } + } + `},gi=(e,t,s,n,i=!1,a)=>{let o=e[0].dims,u=e[1].dims,p=o.slice(0,-2),h=u.slice(0,-2),k=n?n.slice(0,-2):s.slice(0,-2),C=ze.size(k),d=o[o.length-2],z=o[o.length-1],B=u[u.length-1],V=z%4===0&&B%4===0,Z=d<=8?[4,1,1]:[4,4,1],ee=[8,8,1],Y=[Math.ceil(B/ee[0]/Z[0]),Math.ceil(d/ee[1]/Z[1]),Math.ceil(C/ee[2]/Z[2])],he=V?4:1,pe=[...p,d,z/he],Me=pe.length,Oe=[...h,z,B/he],De=Oe.length,Ye=[C,d,B/he],at=[{type:6,data:d},{type:6,data:B},{type:6,data:z}];on(t,at),at.push(...yt(k,pe,Oe));let Pt=["rank","rank"],Xt=e.length>2;Xt&&(at.push(...yt(e[2].dims)),Pt.push("rank")),at.push(...yt(Ye));let Zt=bt=>{let ss=k.length,St=Ui("batchDims",e[0].dataType,ss,1),Ft=fs(e[0].dataType),bs=qe("a",e[0].dataType,Me,he),Ht=qe("b",e[1].dataType,De,he),Rt=It("result",e[0].dataType,Ye.length,he),_s=[bs,Ht];if(Xt){let Tr=i?he:1;_s.push(qe("bias",e[2].dataType,e[2].dims.length,Tr))}let ot=[{name:"dim_a_outer",type:"i32"},{name:"dim_b_outer",type:"i32"},{name:"dim_inner",type:"i32"}];Dr(t,ot);let Et=fs(Rt.type.tensor),ps=nn(t,Rt.type.value,Et),Ns=jo(he,Xt,ps,[St,bs,Ht,Rt],i);return` + ${bt.registerUniforms(ot).registerInternalVariables(St).declareVariables(..._s,Rt)} + ${Ns} + ${V?Ro(Z,ee,Ft,St):No(Z,ee,Ft,St)} + `};return{name:"MatMul",shaderCache:{hint:`${Z};${t.activation};${V};${i}`,inputDependencies:Pt},getRunData:()=>({outputs:[{dims:a?a(s):s,dataType:e[0].dataType}],dispatchGroup:{x:Y[0],y:Y[1],z:Y[2]},programUniforms:at}),getShaderSource:Zt}}}),Vo,gu,qc=g(()=>{zt(),Pe(),Yt(),an(),Lo(),Hc(),Uo(),Vo=(e,t,s,n,i=!1,a,o=4,u=4,p=4,h="f32")=>{let k=at=>{switch(at){case 1:return"resData = x[xIndex];";case 3:return`resData = vec3<${h}>(x[xIndex], x[xIndex + 1], x[xIndex + 2]);`;case 4:return"resData = x[xIndex / 4];";default:throw new Error(`innerElementSize ${at} is not supported.`)}},C=at=>{switch(at){case 1:return"return w[row * i32(uniforms.w_shape[3]) + colIn];";case 4:return"return w[row * i32(uniforms.w_shape[3]) / 4 + colIn];";default:throw new Error(`innerElementSize ${at} is not supported.`)}},d=e?` + let coord = vec4(batch, xRow, xCol, xCh); + `:` + let coord = vec4(batch, xCh, xRow, xCol); + `,z=e?` + let coords = vec4( + batch, + row / outWidth, + row % outWidth, + col); + `:` + let coords = vec4( + batch, + row, + col / outWidth, + col % outWidth); + `,B=e?"i32(uniforms.x_shape[1])":"i32(uniforms.x_shape[2])",V=e?"i32(uniforms.x_shape[2])":"i32(uniforms.x_shape[3])",Z=e?"row":"col",ee=e?"col":"row",Y=` + let inChannels = i32(uniforms.w_shape[2]); + let outWidth = ${e?"i32(uniforms.result_shape[2])":"i32(uniforms.result_shape[3])"}; + let outRow = ${Z} / outWidth; + let outCol = ${Z} % outWidth; + + let WRow = ${ee} / (i32(uniforms.w_shape[1]) * inChannels); + let WCol = ${ee} / inChannels % i32(uniforms.w_shape[1]); + let xRow = outRow * uniforms.stride[0] + uniforms.dilation[0] * WRow - uniforms.pad[0]; + let xCol = outCol * uniforms.stride[1] + uniforms.dilation[1] * WCol - uniforms.pad[1]; + let xCh = ${ee} % inChannels; + var resData = ${Ks(o,h)}(0.0); + // The bounds checking is always needed since we use it to pad zero for + // the 'same' padding type. + if (xRow >= 0 && xRow < ${B} && xCol >= 0 && xCol < ${V}) { + ${d} + let xIndex = getIndexFromCoords4D(coord, vec4(uniforms.x_shape)); + ${k(o)} + } + return resData;`,he=e?t&&n?` + let col = colIn * ${o}; + ${Y}`:` + let col = colIn * ${o}; + if (row < uniforms.dim_a_outer && col < uniforms.dim_inner) { + ${Y} + } + return ${Ks(o,h)}(0.0);`:n&&s?` + let col = colIn * ${o}; + ${Y}`:` + let col = colIn * ${o}; + if (row < uniforms.dim_inner && col < uniforms.dim_b_outer) { + ${Y} + } + return ${Ks(o,h)}(0.0);`,pe=e?n&&s?C(u):` + let col = colIn * ${u}; + if (row < uniforms.dim_inner && col < uniforms.dim_b_outer) { + ${C(u)} + } + return ${Ks(u,h)}(0.0);`:` + let col = colIn * ${u}; + if (row < uniforms.dim_inner && col < uniforms.dim_a_outer) { + ${C(u)} + } + return ${Ks(u,h)}(0.0);`,Me=Ks(p,h),Oe=Ks(e?o:u,h),De=Ks(e?u:o,h),Ye=nn(a,Me,h);return` + fn mm_readA(batch: i32, row : i32, colIn : i32) -> ${Oe} { + ${e?he:pe} + } + + fn mm_readB(batch: i32, row : i32, colIn : i32) -> ${De} { + ${e?pe:he} + } + + fn mm_write(batch: i32, row : i32, colIn : i32, valueIn : ${Me}) { + let col = colIn * ${p}; + if (row < uniforms.dim_a_outer && col < uniforms.dim_b_outer) + { + var value = valueIn; + let outWidth = ${e?"i32(uniforms.result_shape[2])":"i32(uniforms.result_shape[3])"}; + ${z} + ${Do(i)} + ${Ye} + setOutputAtCoords(coords[0], coords[1], coords[2], coords[3], value); + } + }`},gu=(e,t,s,n,i,a,o,u,p)=>{let h=t.format==="NHWC",k=h?e[0].dims[3]:e[0].dims[1],C=s[0],d=h?s[2]:s[3],z=h?s[1]:s[2],B=h?s[3]:s[1],V=h&&(k%4===0||k%3===0)&&B%4===0,Z=h?B:d*z,ee=h?d*z:B,Y=[8,8,1],he=n<=8?[4,1,1]:[4,4,1],pe=[Math.ceil(Z/Y[0]/he[0]),Math.ceil(ee/Y[1]/he[1]),Math.ceil(C/Y[2]/he[2])];as("verbose",()=>`[conv2d_mm_webgpu] dispatch = ${pe}`);let Me=V?h&&k%4!==0?3:4:1,Oe=Y[1]*he[1],De=Y[0]*he[0],Ye=Math.max(Y[0]*Me,Y[1]),at=n%Oe===0,Pt=i%De===0,Xt=a%Ye===0,Zt=V?[Me,4,4]:[1,1,1],bt=[{type:6,data:n},{type:6,data:i},{type:6,data:a},{type:6,data:[t.pads[0],t.pads[1]]},{type:6,data:t.strides},{type:6,data:t.dilations}];on(t,bt),bt.push(...yt(e[0].dims,e[1].dims));let ss=["rank","rank"];o&&(bt.push(...yt(e[2].dims)),ss.push("rank")),bt.push(...yt(s));let St=Ft=>{let bs=[{name:"dim_a_outer",type:"i32"},{name:"dim_b_outer",type:"i32"},{name:"dim_inner",type:"i32"},{name:"pad",type:"i32",length:2},{name:"stride",type:"i32",length:2},{name:"dilation",type:"i32",length:2}];Dr(t,bs);let Ht=V?4:1,Rt=fs(e[0].dataType),_s=` + fn setOutputAtIndex(flatIndex : i32, value : ${V?`vec4<${Rt}>`:Rt}) { + result[flatIndex] = ${V?`vec4<${Rt}>`:Rt}(value); + } + fn setOutputAtCoords(d0 : i32, d1 : i32, d2 : i32, d3 : i32, value : ${V?`vec4<${Rt}>`:Rt}) { + let flatIndex = getOutputIndexFromCoords(vec4(d0, d1, d2, d3)); + setOutputAtIndex(flatIndex ${V?"/ 4":""}, value); + }`,ot=qe("x",e[0].dataType,e[0].dims.length,Me===3?1:Me),Et=qe("w",e[1].dataType,e[1].dims.length,Ht),ps=[ot,Et],Ns=It("result",e[0].dataType,s.length,Ht);if(o){let Tr=qe("bias",e[2].dataType,e[2].dims.length,Ht);ps.push(Tr),_s+=` + fn getBiasByOutputCoords(coords : vec4) -> ${V?`vec4<${Rt}>`:Rt} { + return bias[coords.${h?"w":"y"}${V?"/ 4":""}]; + }`}return` + ${zo("uniforms.result_strides")} + //struct Uniforms { xShape : vec4, wShape : vec4, outShape : vec4, + // outShapeStrides: vec3, filterDims : vec2, pad : vec2, stride : vec2, + // dilation : vec2, dimAOuter : i32, dimBOuter : i32, dimInner : i32 }; + ${Ft.registerUniforms(bs).declareVariables(...ps,Ns)} + ${_s} + ${Vo(h,at,Pt,Xt,o,t,Zt[0],Zt[1],Zt[2],Rt)} + ${V?Ro(he,Y,Rt,void 0,!h,Ye):No(he,Y,Rt,void 0,!h,Ye,!1,void 0,u)}`};return{name:"Conv2DMatMul",shaderCache:{hint:`${t.cacheKey};${Me};${V};${at};${Pt};${Xt};${Oe};${De};${Ye}`,inputDependencies:ss},getRunData:()=>({outputs:[{dims:p?p(s):s,dataType:e[0].dataType}],dispatchGroup:{x:pe[0],y:pe[1],z:pe[2]},programUniforms:bt}),getShaderSource:St}}}),Wo,Go,Vn,Ko,Ho,wu,qo,yu,Qc=g(()=>{zt(),Pe(),Ot(),Yt(),an(),Lo(),Wo=e=>{let t=1;for(let s=0;stypeof e=="number"?[e,e,e]:e,Vn=(e,t)=>t<=1?e:e+(e-1)*(t-1),Ko=(e,t,s,n=1)=>{let i=Vn(t,n);return Math.floor((e[0]*(s-1)-s+i)/2)},Ho=(e,t,s,n,i)=>{i==null&&(i=Ko(e,t[0],n[0]));let a=[0,0,0,s];for(let o=0;o<3;o++)e[o]+2*i>=t[o]&&(a[o]=Math.trunc((e[o]-t[o]+2*i)/n[o]+1));return a},wu=(e,t,s,n,i,a,o,u,p,h)=>{let k,C,d,z;if(e==="VALID"&&(e=0),typeof e=="number"){k={top:e,bottom:e,left:e,right:e,front:e,back:e};let B=Ho([t,s,n,1],[u,p,h],1,[i,a,o],e);C=B[0],d=B[1],z=B[2]}else if(Array.isArray(e)){if(!e.every((V,Z,ee)=>V===ee[0]))throw Error(`Unsupported padding parameter: ${e}`);k={top:e[0],bottom:e[1],left:e[2],right:e[3],front:e[4],back:e[5]};let B=Ho([t,s,n,1],[u,p,h],1,[i,a,o],e[0]);C=B[0],d=B[1],z=B[2]}else if(e==="SAME_UPPER"){C=Math.ceil(t/i),d=Math.ceil(s/a),z=Math.ceil(n/o);let B=(C-1)*i+u-t,V=(d-1)*a+p-s,Z=(z-1)*o+h-n,ee=Math.floor(B/2),Y=B-ee,he=Math.floor(V/2),pe=V-he,Me=Math.floor(Z/2),Oe=Z-Me;k={top:he,bottom:pe,left:Me,right:Oe,front:ee,back:Y}}else throw Error(`Unknown padding parameter: ${e}`);return{padInfo:k,outDepth:C,outHeight:d,outWidth:z}},qo=(e,t,s,n,i,a=!1,o="channelsLast")=>{let u,p,h,k,C;if(o==="channelsLast")[u,p,h,k,C]=e;else if(o==="channelsFirst")[u,C,p,h,k]=e;else throw new Error(`Unknown dataFormat ${o}`);let[d,,z,B,V]=t,[Z,ee,Y]=Go(s),[he,pe,Me]=Go(n),Oe=Vn(z,he),De=Vn(B,pe),Ye=Vn(V,Me),{padInfo:at,outDepth:Pt,outHeight:Xt,outWidth:Zt}=wu(i,p,h,k,Z,ee,Y,Oe,De,Ye),bt=a?d*C:d,ss=[0,0,0,0,0];return o==="channelsFirst"?ss=[u,bt,Pt,Xt,Zt]:o==="channelsLast"&&(ss=[u,Pt,Xt,Zt,bt]),{batchSize:u,dataFormat:o,inDepth:p,inHeight:h,inWidth:k,inChannels:C,outDepth:Pt,outHeight:Xt,outWidth:Zt,outChannels:bt,padInfo:at,strideDepth:Z,strideHeight:ee,strideWidth:Y,filterDepth:z,filterHeight:B,filterWidth:V,effectiveFilterDepth:Oe,effectiveFilterHeight:De,effectiveFilterWidth:Ye,dilationDepth:he,dilationHeight:pe,dilationWidth:Me,inShape:e,outShape:ss,filterShape:t}},yu=(e,t,s,n,i,a)=>{let o=a==="channelsLast";o?e[0].dims[3]:e[0].dims[1];let u=[64,1,1],p={x:s.map((Z,ee)=>ee)},h=[Math.ceil(Wo(p.x.map(Z=>s[Z]))/u[0]),1,1];as("verbose",()=>`[conv3d_naive_webgpu] dispatch = ${h}`);let k=1,C=ze.size(s),d=[{type:12,data:C},{type:12,data:n},{type:12,data:i},{type:12,data:t.strides},{type:12,data:t.dilations}];on(t,d),d.push(...yt(e[0].dims,e[1].dims));let z=["rank","rank"],B=e.length===3;B&&(d.push(...yt(e[2].dims)),z.push("rank")),d.push(...yt(s));let V=Z=>{let ee=[{name:"output_size",type:"u32"},{name:"filter_dims",type:"u32",length:n.length},{name:"pads",type:"u32",length:i.length},{name:"strides",type:"u32",length:t.strides.length},{name:"dilations",type:"u32",length:t.dilations.length}];Dr(t,ee);let Y=1,he=fs(e[0].dataType),pe=qe("x",e[0].dataType,e[0].dims.length,k),Me=qe("W",e[1].dataType,e[1].dims.length,Y),Oe=[pe,Me],De=It("result",e[0].dataType,s.length,Y),Ye="";if(B){let Xt=qe("bias",e[2].dataType,e[2].dims.length,Y);Oe.push(Xt),Ye+=` + fn getBiasByOutputCoords(coords : array) -> ${he} { + return bias[${o?$t("coords",4,5):$t("coords",1,5)}]; + }`}let at=Ks(k,he),Pt=nn(t,at,he);return` + ${Ye} + fn getX(d0 : u32, d1 : u32, d2 : u32, d3 : u32, d4 : u32) -> f32 { + let aIndices = array(d0, d1, d2, d3, d4); + return ${pe.getByIndices("aIndices")}; + } + fn getW(d0 : u32, d1 : u32, d2 : u32, d3 : u32, d4 : u32) -> f32 { + let aIndices = array(d0, d1, d2, d3, d4); + return ${Me.getByIndices("aIndices")}; + } + ${Z.registerUniforms(ee).declareVariables(...Oe,De)} + ${Z.mainStart()} + ${Z.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let coords = ${De.offsetToIndices("global_idx")}; + let batch = ${$t("coords",0,pe.rank)}; + let d2 = ${o?$t("coords",pe.rank-1,pe.rank):$t("coords",1,pe.rank)}; + let xFRCCorner = vec3(${o?$t("coords",1,pe.rank):$t("coords",2,pe.rank)}, + ${o?$t("coords",2,pe.rank):$t("coords",3,pe.rank)}, + ${o?$t("coords",3,pe.rank):$t("coords",4,pe.rank)}) * uniforms.strides - uniforms.pads; + let xFCorner = xFRCCorner.x; + let xRCorner = xFRCCorner.y; + let xCCorner = xFRCCorner.z; + let xShapeY = ${o?$t("uniforms.x_shape",1,pe.rank):$t("uniforms.x_shape",2,pe.rank)}; + let xShapeZ = ${o?$t("uniforms.x_shape",2,pe.rank):$t("uniforms.x_shape",3,pe.rank)}; + let xShapeW = ${o?$t("uniforms.x_shape",3,pe.rank):$t("uniforms.x_shape",4,pe.rank)}; + let xShapeU = ${o?$t("uniforms.x_shape",4,pe.rank):$t("uniforms.x_shape",1,pe.rank)}; + let inputDepthNearestVec4 = (xShapeU / 4) * 4; + let inputDepthVec4Remainder = xShapeU % 4; + + var value = 0.0; + for (var wF = 0u; wF < uniforms.filter_dims[0]; wF++) { + let xF = xFCorner + wF * uniforms.dilations[0]; + if (xF < 0 || xF >= xShapeY) { + continue; + } + + for (var wR = 0u; wR < uniforms.filter_dims[1]; wR++) { + let xR = xRCorner + wR * uniforms.dilations[1]; + if (xR < 0 || xR >= xShapeZ) { + continue; + } + + for (var wC = 0u; wC < uniforms.filter_dims[2]; wC++) { + let xC = xCCorner + wC * uniforms.dilations[2]; + if (xC < 0 || xC >= xShapeW) { + continue; + } + + for (var d1 = 0u; d1 < inputDepthNearestVec4; d1 += 4) { + ${o?`let 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)); + `:`let xValues = vec4( + getX(batch, d1, xF, xR, xC), + getX(batch, d1 + 1, xF, xR, xC), + getX(batch, d1 + 2, xF, xR, xC), + getX(batch, d1 + 3, xF, xR, xC)); + `} + let wValues = vec4( + getW(d2, d1, wF, wR, wC), + getW(d2, d1 + 1, wF, wR, wC), + getW(d2, d1 + 2, wF, wR, wC), + getW(d2, d1 + 3, wF, wR, wC)); + value += dot(xValues, wValues); + } + if (inputDepthVec4Remainder == 1) { + ${o?`value += getX(batch, xF, xR, xC, inputDepthNearestVec4) + * getW(d2, inputDepthNearestVec4, wF, wR, wC);`:`value += getX(batch, inputDepthNearestVec4, xF, xR, xC) + * getW(d2, inputDepthNearestVec4, wF, wR, wC);`} + } else if (inputDepthVec4Remainder == 2) { + ${o?`let xValues = vec2( + getX(batch, xF, xR, xC, inputDepthNearestVec4), + getX(batch, xF, xR, xC, inputDepthNearestVec4 + 1)); + `:`let xValues = vec2( + getX(batch, inputDepthNearestVec4, xF, xR, xC), + getX(batch, inputDepthNearestVec4 + 1, xF, xR, xC)); + `} + let wValues = vec2( + getW(d2, inputDepthNearestVec4, wF, wR, wC), + getW(d2, inputDepthNearestVec4 + 1, wF, wR, wC)); + value += dot(xValues, wValues); + } else if (inputDepthVec4Remainder == 3) { + ${o?`let xValues = vec3( + getX(batch, xF, xR, xC, inputDepthNearestVec4), + getX(batch, xF, xR, xC, inputDepthNearestVec4 + 1), + getX(batch, xF, xR, xC, inputDepthNearestVec4 + 2)); + `:`let xValues = vec3( + getX(batch, inputDepthNearestVec4, xF, xR, xC), + getX(batch, inputDepthNearestVec4 + 1, xF, xR, xC), + getX(batch, inputDepthNearestVec4 + 2, xF, xR, xC)); + `} + let wValues = vec3( + getW(d2, inputDepthNearestVec4, wF, wR, wC), + getW(d2, inputDepthNearestVec4 + 1, wF, wR, wC), + getW(d2, inputDepthNearestVec4 + 2, wF, wR, wC)); + value += dot(xValues, wValues); + } + } + } + } + ${B?"value = value + getBiasByOutputCoords(coords)":""}; + ${Pt} + result[global_idx] = f32(value); + }`};return{name:"Conv3DNaive",shaderCache:{hint:`${t.cacheKey};${o};${k};${B}`,inputDependencies:z},getRunData:()=>({outputs:[{dims:s,dataType:e[0].dataType}],dispatchGroup:{x:h[0],y:h[1],z:h[2]},programUniforms:d}),getShaderSource:V}}}),Mu,bu,Qo=g(()=>{zt(),Ot(),Yt(),an(),Mu=(e,t,s,n)=>{let i=e.length>2,a=i?"value += b[output_channel];":"",o=e[0].dims,u=e[1].dims,p=t.format==="NHWC",h=p?s[3]:s[1],k=h/t.group,C=p&&k>=4?qt(h):1,d=ze.size(s)/C,z=[{type:12,data:d},{type:12,data:t.dilations},{type:12,data:[t.strides[0],t.strides[1]]},{type:12,data:[t.pads[0],t.pads[1]]},{type:12,data:k}];on(t,z),z.push(...yt(o,[u[0],u[1],u[2],u[3]/C]));let B=i?["rank","rank","rank"]:["rank","rank"];z.push(...yt([s[0],s[1],s[2],s[3]/C]));let V=Z=>{let ee=It("output",e[0].dataType,s.length,C),Y=fs(ee.type.tensor),he=nn(t,ee.type.value,Y),pe=qe("x",e[0].dataType,o.length),Me=qe("w",e[1].dataType,u.length,C),Oe=[pe,Me];i&&Oe.push(qe("b",e[2].dataType,e[2].dims,C));let De=[{name:"output_size",type:"u32"},{name:"dilations",type:"u32",length:t.dilations.length},{name:"strides",type:"u32",length:2},{name:"pads",type:"u32",length:2},{name:"output_channels_per_group",type:"u32"}];Dr(t,De);let Ye=p?` + for (var wHeight: u32 = 0u; wHeight < uniforms.w_shape[0]; wHeight++) { + let xHeight = xRCCorner.x + wHeight * uniforms.dilations[0]; + + if (xHeight < 0u || xHeight >= uniforms.x_shape[1]) { + continue; + } + + for (var wWidth: u32 = 0u; wWidth < uniforms.w_shape[1]; wWidth++) { + let xWidth = xRCCorner.y + wWidth * uniforms.dilations[1]; + if (xWidth < 0u || xWidth >= uniforms.x_shape[2]) { + continue; + } + + for (var wInChannel: u32 = 0u; wInChannel < uniforms.w_shape[2]; wInChannel++) { + let input_channel = in_channel_offset + wInChannel; + let xVal = ${pe.get("batch","xHeight","xWidth","input_channel")}; + let wVal = ${Me.get("wHeight","wWidth","wInChannel","output_channel")}; + value += xVal * wVal; + } + } + } + `:` + for (var wInChannel: u32 = 0u; wInChannel < uniforms.w_shape[1]; wInChannel++) { + let input_channel = in_channel_offset + wInChannel; + for (var wHeight: u32 = 0u; wHeight < uniforms.w_shape[2]; wHeight++) { + let xHeight = xRCCorner.x + wHeight * uniforms.dilations[0]; + + if (xHeight < 0u || xHeight >= uniforms.x_shape[2]) { + continue; + } + + for (var wWidth: u32 = 0u; wWidth < uniforms.w_shape[3]; wWidth++) { + let xWidth = xRCCorner.y + wWidth * uniforms.dilations[1]; + if (xWidth < 0u || xWidth >= uniforms.x_shape[3]) { + continue; + } + + let xVal = ${pe.get("batch","input_channel","xHeight","xWidth")}; + let wVal = ${Me.get("output_channel","wInChannel","wHeight","wWidth")}; + value += xVal * wVal; + } + } + } + `;return` + ${Z.registerUniforms(De).declareVariables(...Oe,ee)} + + ${Z.mainStart()} + ${Z.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + + let outputIndices = ${ee.offsetToIndices("global_idx")}; + let batch: u32 = outputIndices[0]; + let output_channel: u32 = outputIndices[${p?3:1}]; + let xRCCorner: vec2 = vec2(outputIndices[${p?1:2}], outputIndices[${p?2:3}]) * uniforms.strides - uniforms.pads; + let group_id: u32 = output_channel * ${C} / uniforms.output_channels_per_group; + var in_channel_offset = group_id * uniforms.w_shape[${p?2:1}]; + + var value: ${ee.type.value} = ${ee.type.value}(0); + ${Ye} + ${a} + ${he} + ${ee.setByOffset("global_idx","value")} + }`};return{name:"GroupedConv",shaderCache:{hint:`${t.cacheKey}_${C}`,inputDependencies:B},getRunData:()=>({outputs:[{dims:n?n(s):s,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(d/64)},programUniforms:z}),getShaderSource:V}},bu=(e,t,s,n)=>{let i=e.length>2,a=qt(s[3]),o=qt(s[2]),u=ze.size(s)/a/o,p=[e[0].dims[0],e[0].dims[1],e[0].dims[2],e[0].dims[3]/a],h=[e[1].dims[0],e[1].dims[1],e[1].dims[2],e[1].dims[3]/a],k=[s[0],s[1],s[2],s[3]/a],C=[{type:12,data:u},{type:6,data:[t.strides[0],t.strides[1]]},{type:6,data:[t.pads[0],t.pads[1]]}];on(t,C),C.push(...yt(p,h,k));let d=(o-1)*t.strides[1]+h[1],z=B=>{let V=It("output",e[0].dataType,k.length,a),Z=fs(V.type.tensor),ee=nn(t,V.type.value,Z),Y=qe("x",e[0].dataType,p.length,a),he=qe("w",e[1].dataType,h.length,a),pe=[Y,he];i&&pe.push(qe("b",e[2].dataType,e[2].dims,a));let Me=i?"value += b[output_channel];":"",Oe=[{name:"output_size",type:"u32"},{name:"strides",type:"i32",length:2},{name:"pads",type:"i32",length:2}];return Dr(t,Oe),` + ${B.registerUniforms(Oe).declareVariables(...pe,V)} + ${B.mainStart()} + ${B.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let width0 = uniforms.output_shape[3]; + let output_channel = global_idx % width0; + var index1 = global_idx / width0; + let width1 = uniforms.output_shape[2] / ${o}u; + let col = (index1 % width1) * ${o}u; + index1 = index1 / width1; + let row = index1 % uniforms.output_shape[1]; + let batch = index1 / uniforms.output_shape[1]; + + let x_corner = vec2(i32(row), i32(col)) * uniforms.strides - uniforms.pads; + + var x_vals: array<${Y.type.value}, ${d}>; + var values: array<${V.type.value}, ${o}>; + let input_channel = output_channel; + // Use constant instead of uniform can give better performance for w's height/width. + for (var w_height: u32 = 0u; w_height < ${h[0]}; w_height++) { + let x_height = x_corner.x + i32(w_height); + if (x_height >= 0 && u32(x_height) < uniforms.x_shape[1]) { + for (var i = 0; i < ${d}; i++) { + let x_width = x_corner.y + i; + if (x_width >= 0 && u32(x_width) < uniforms.x_shape[2]) { + x_vals[i] = ${Y.get("batch","u32(x_height)","u32(x_width)","input_channel")}; + } else { + x_vals[i] = ${Y.type.value}(0); + } + } + for (var w_width: u32 = 0u; w_width < ${h[1]}; w_width++) { + let w_val = ${he.get("w_height","w_width","0","output_channel")}; + for (var i = 0u; i < ${o}u; i++) { + values[i] = fma(x_vals[i * u32(uniforms.strides[1]) + w_width], w_val, values[i]); + } + } + } + } + + for (var i = 0u; i < ${o}u; i++) { + var value = values[i]; + ${Me} + ${ee} + ${V.set("batch","row","col + i","output_channel","value")}; + } + }`};return{name:"GroupedConv-Vectorize",shaderCache:{hint:`${t.cacheKey};${a};${o};${d};${h[0]};${h[1]}`,inputDependencies:i?["rank","rank","type"]:["rank","rank"]},getRunData:()=>({outputs:[{dims:n?n(s):s,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(u/64)},programUniforms:C}),getShaderSource:z}}}),vu,wi,xu,yi,Xo,Mi,Tu,Pu,bi,Xc=g(()=>{Ot(),qc(),Qc(),Uo(),Qo(),an(),fi(),Kr(),vu=(e,t,s,n,i,a)=>{let o=e[0],u=e.slice(a?1:2,a?3:4),p=u.length,h=t[0],k=t.slice(2).map((d,z)=>d+(d-1)*(s[z]-1)),C=u.map((d,z)=>d+n[z]+n[z+p]).map((d,z)=>Math.floor((d-k[z]+i[z])/i[z]));return C.splice(0,0,o),C.splice(a?3:1,0,h),C},wi=[2,3,1,0],xu=(e,t)=>{if(!e||e.length!==2&&e.length!==3)throw new Error("Conv requires 2 or 3 inputs");if(e[0].dims.length>5)throw new Error("greater than 5D is not supported");if(e[0].dims.length!==e[1].dims.length)throw new Error("filter does not have same dimension as input");let s=e[0].dims[t.format==="NHWC"?e[0].dims.length-1:1],n=e[1].dims[1]*t.group;if(s!==n)throw new Error("FILTER_IN_CHANNEL should be equal to DATA_CHANNEL");if(e.length===3&&(e[2].dims.length!==1||e[1].dims[0]!==e[2].dims[0]))throw new Error("invalid bias");let i=e[0].dims.length-2;if(t.dilations.length!==i)throw new Error(`dilations should be ${i}D`);if(t.strides.length!==i)throw new Error(`strides should be ${i}D`);if(t.pads.length!==i*2)throw new Error(`pads should be ${i*2}D`);if(t.kernelShape.length!==0&&t.kernelShape.length!==e[1].dims.length-2)throw new Error("invalid kernel shape")},yi=(e,t)=>{let s=e.kernelShape.slice();s.length{let t=Fo(e),s=e.format,n=["NOTSET","VALID","SAME_UPPER","SAME_LOWER"][e.auto_pad],i=e.dilations,a=e.group,o=e.kernel_shape,u=e.pads,p=e.strides,h=e.w_is_const();return{autoPad:n,format:s,dilations:i,group:a,kernelShape:o,pads:u,strides:p,wIsConst:h,...t,cacheKey:`${e.format};${t.activation};`}},Mi=(e,t,s,n)=>{let i=s.format==="NHWC",a=vu(t[0].dims,t[1].dims,s.dilations,s.pads,s.strides,i);if(s.group!==1){let Oe=[t[0]];if(i){let De=e.kernelCustomData.wT??e.compute(pr(t[1],wi),{inputs:[1],outputs:[s.wIsConst?-2:-1]})[0];s.wIsConst&&!e.kernelCustomData.wT&&(e.kernelCustomData.wT=De),Oe.push(De)}else Oe.push(t[1]);t.length===3&&Oe.push(t[2]),!e.adapterInfo.isArchitecture("ampere")&&i&&t[1].dims[0]===s.group&&t[1].dims[1]===1&&s.dilations[0]===1&&s.dilations[1]===1?e.compute(bu(Oe,s,a,n),{inputs:Oe}):e.compute(Mu(Oe,s,a,n),{inputs:Oe});return}let o=t.length===3,u=t[0].dims[i?1:2],p=t[0].dims[i?2:3],h=t[0].dims[i?3:1],k=t[1].dims[2],C=t[1].dims[3],d=a[i?1:2],z=a[i?2:3],B=a[i?3:1],V=i&&k===u&&C===p&&s.pads[0]===0&&s.pads[1]===0;if(V||k===1&&C===1&&s.dilations[0]===1&&s.dilations[1]===1&&s.strides[0]===1&&s.strides[1]===1&&s.pads[0]===0&&s.pads[1]===0){let Oe=a[0],De,Ye,at,Pt=[];if(i){let bt=e.kernelCustomData.wT??e.compute(pr(t[1],wi),{inputs:[1],outputs:[s.wIsConst?-2:-1]})[0];if(s.wIsConst&&!e.kernelCustomData.wT&&(e.kernelCustomData.wT=bt),V){let ss=u*p*h;De=t[0].reshape([1,Oe,ss]),Ye=bt.reshape([1,ss,B]),at=[1,Oe,B]}else De=t[0].reshape([Oe,u*p,h]),Ye=bt.reshape([1,h,B]),at=[Oe,d*z,B];Pt.push(De),Pt.push(Ye)}else De=t[0].reshape([Oe,h,u*p]),Ye=t[1].reshape([1,B,h]),at=[Oe,B,d*z],Pt.push(Ye),Pt.push(De);o&&Pt.push(t[2]);let Xt=at[2],Zt=Pt[0].dims[Pt[0].dims.length-1];Xt<8&&Zt<8?e.compute(Bo(Pt,s,a,at,i,n),{inputs:Pt}):e.compute(gi(Pt,s,a,at,i,n),{inputs:Pt});return}let Z=!0,ee=e.kernelCustomData.wT??e.compute(pr(t[1],wi),{inputs:[1],outputs:[s.wIsConst?-2:-1]})[0];s.wIsConst&&!e.kernelCustomData.wT&&(e.kernelCustomData.wT=ee);let Y=[t[0],ee];o&&Y.push(t[2]);let he=i?d*z:B,pe=i?B:d*z,Me=k*C*h;e.compute(gu(Y,s,a,he,pe,Me,o,Z,n),{inputs:Y})},Tu=(e,t)=>{let s=t.format==="NHWC",n=[e.inputs[0].reshape(s?[e.inputs[0].dims[0],1,e.inputs[0].dims[1],e.inputs[0].dims[2]]:[e.inputs[0].dims[0],e.inputs[0].dims[1],1,e.inputs[0].dims[2]]),e.inputs[1].reshape([e.inputs[1].dims[0],e.inputs[1].dims[1],1,e.inputs[1].dims[2]])];e.inputs.length===3&&n.push(e.inputs[2]);let i=[0,t.pads[0],0,t.pads[1]],a=[1].concat(t.strides),o=[1].concat(t.dilations),u=[1].concat(t.kernelShape),p=yi({...t,pads:i,strides:a,dilations:o,kernelShape:u},n);Mi(e,n,p,h=>s?[h[0],h[2],h[3]]:[h[0],h[1],h[3]])},Pu=(e,t,s)=>{let n=s.format==="NHWC"?"channelsLast":"channelsFirst",i=yi(s,t),a=s.autoPad==="NOTSET"?s.pads:s.autoPad,o=qo(t[0].dims,t[1].dims,s.strides,s.dilations,a,!1,n);e.compute(yu(t,i,o.outShape,[o.filterDepth,o.filterHeight,o.filterWidth],[o.padInfo.front,o.padInfo.top,o.padInfo.left],n))},bi=(e,t)=>{if(xu(e.inputs,t),e.inputs[0].dims.length===3)Tu(e,t);else if(e.inputs[0].dims.length===5)Pu(e,e.inputs,t);else{let s=yi(t,e.inputs);Mi(e,e.inputs,s)}}}),Eu,Yc=g(()=>{zt(),Pe(),Ot(),Yt(),Eu=(e,t,s)=>{let n=e.length>2,i=t.outputShape,a=t.format==="NHWC",o=t.group,u=e[1].dims,p=u[2]/o,h=u[3],k=a?qt(p):1,C=a?qt(h):1,d=a?h===1?k:C:1,z=ze.size(i)/C,B=[Math.ceil(z/64),1,1];as("verbose",()=>`[conv2d_backprop_webgpu] dispatch = ${B}`);let V=["rank","rank"],Z=[t.strides[0],t.strides[1]],ee=[t.kernelShape[a?1:2],t.kernelShape[a?2:3]],Y=[t.dilations[0],t.dilations[1]],he=[ee[0]+(t.dilations[0]<=1?0:(t.kernelShape[a?1:2]-1)*(t.dilations[0]-1)),ee[1]+(t.dilations[1]<=1?0:(t.kernelShape[a?2:3]-1)*(t.dilations[1]-1))],pe=[he[0]-1-Math.floor((t.pads[0]+t.pads[2])/2),he[1]-1-Math.floor((t.pads[1]+t.pads[3])/2)],Me=[{type:12,data:z},{type:12,data:Z},{type:12,data:ee},{type:12,data:Y},{type:12,data:he},{type:6,data:pe},{type:12,data:p},{type:12,data:h},...yt(e[0].dims,e[1].dims)];n&&(Me.push(...yt(e[2].dims)),V.push("rank")),Me.push(...yt(i));let Oe=De=>{let Ye=[{name:"output_size",type:"u32"},{name:"strides",type:"u32",length:Z.length},{name:"filter_dims",type:"u32",length:ee.length},{name:"dilations",type:"u32",length:ee.length},{name:"effective_filter_dims",type:"u32",length:he.length},{name:"pads",type:"i32",length:pe.length},{name:"input_channels_per_group",type:"u32"},{name:"output_channels_per_group",type:"u32"}],at=fs(e[0].dataType),Pt=a?1:2,Xt=a?2:3,Zt=a?3:1,bt=qe("W",e[1].dataType,e[1].dims.length,d),ss=qe("Dy",e[0].dataType,e[0].dims.length,k),St=[ss,bt];n&&St.push(qe("bias",e[2].dataType,[i[Zt]].length,C));let Ft=It("result",e[0].dataType,i.length,C),bs=()=>{let Rt="";if(k===1)Rt+=` + let w_offset = ${bt.indicesToOffset(`${bt.type.indices}(u32(wRPerm), u32(wCPerm), inputChannel, wOutChannel)`)}; + let wValue = ${bt.getByOffset(`w_offset / ${d}`)}; + dotProd = dotProd + xValue * wValue;`;else if(h===1)Rt+=` + let wValue = ${bt.getByOffset(`${bt.indicesToOffset(`${bt.type.indices}(u32(wRPerm), u32(wCPerm), inputChannel, wOutChannel)`)} / ${d}`)}; + dotProd = dotProd + dot(xValue, wValue);`;else for(let _s=0;_s(i32(r), i32(c)) - uniforms.pads; + let dyRCorner = dyCorner.x; + let dyCCorner = dyCorner.y; + let groupId = d1 / uniforms.output_channels_per_group; + let wOutChannel = d1 - groupId * uniforms.output_channels_per_group; + // Convolve dy(?, ?, d2) with w(:, :, d1, d2) to compute dx(xR, xC, d1). + // ? = to be determined. : = across all values in that axis. + var dotProd = ${Ft.type.value}(0.0); + for (var wR: u32 = 0; wR < uniforms.effective_filter_dims.x; wR = wR + 1) { + if (wR % uniforms.dilations.x != 0) { + continue; + } + let dyR = (${at}(dyRCorner) + ${at}(wR)) / ${at}(uniforms.strides[0]); + let wRPerm = uniforms.filter_dims.x - 1 - wR / uniforms.dilations.x; + if (dyR < 0.0 || dyR >= ${at}(uniforms.Dy_shape[${Pt}]) || fract(dyR) > 0.0 || + wRPerm < 0) { + continue; + } + wR = wR + uniforms.strides[0] - 1; + let idyR: u32 = u32(dyR); + + for (var wC: u32 = 0; wC < uniforms.effective_filter_dims.y; wC = wC + 1) { + if (wC % uniforms.dilations.y != 0) { + continue; + } + let dyC = (${at}(dyCCorner) + ${at}(wC)) / ${at}(uniforms.strides.y); + let wCPerm = uniforms.filter_dims.y - 1 - wC / uniforms.dilations.y; + if (dyC < 0.0 || dyC >= ${at}(uniforms.Dy_shape[${Xt}]) || + fract(dyC) > 0.0 || wCPerm < 0) { + continue; + } + wC = wC + uniforms.strides.y - 1; + let idyC: u32 = u32(dyC); + var inputChannel = groupId * uniforms.input_channels_per_group; + for (var d2: u32 = 0; d2 < uniforms.input_channels_per_group; d2 = d2 + ${k}) { + let xValue = ${a?ss.getByOffset(`${ss.indicesToOffset(`${ss.type.indices}(batch, idyR, idyC, inputChannel)`)} / ${k}`):ss.get("batch","inputChannel","idyR","idyC")}; + ${bs()} + inputChannel = inputChannel + ${k}; + } + } + } + let value = dotProd${n?` + bias[d1 / ${C}]`:""}; + ${Ft.setByOffset("global_idx","value")}; + `;return` + ${De.registerUniforms(Ye).declareVariables(...St,Ft)} + ${De.mainStart()} + ${De.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")}; + ${Ht}}`};return{name:"ConvTranspose2D",shaderCache:{hint:`${t.cacheKey};${k}${d}${C}${h===1}`,inputDependencies:V},getRunData:()=>({dispatchGroup:{x:B[0],y:B[1],z:B[2]},outputs:[{dims:s?s(i):i,dataType:e[0].dataType}],programUniforms:Me}),getShaderSource:Oe}}}),Cu,Yo,ku,Jo,Zo,Su,ea,ta,$u,Jc=g(()=>{Yc(),an(),Kr(),Cu=(e,t,s,n,i,a)=>(e-1)*t+s+(n-1)*i+1-a,Yo=(e,t,s,n,i)=>{let a=Math.floor(e/2);t==="SAME_UPPER"?(s[n]=a,s[i]=e-a):t==="SAME_LOWER"&&(s[n]=e-a,s[i]=a)},ku=(e,t,s,n,i,a,o,u,p,h)=>{let k=e.length-2,C=h.length===0;p.length{let s=e.kernelShape.slice();if(e.kernelShape.length===0||e.kernelShape.reduce((C,d)=>C*d,1)===0){s.length=0;for(let C=2;CC+d,0)===0){let C=t[0].dims.length-2;p=new Array(C).fill(1)}let h=e.strides.slice();if(h.reduce((C,d)=>C+d,0)===0){let C=t[0].dims.length-2;h=new Array(C).fill(1)}ku(u,s,p,e.autoPad,e.group,i,h,n,o,a);let k=Object.assign({},e);return Object.assign(k,{kernelShape:s,pads:i,outputPadding:o,outputShape:a,dilations:p,strides:h}),k},Zo=e=>{let t=Fo(e),s=e.format,n=["NOTSET","VALID","SAME_UPPER","SAME_LOWER"][typeof e.autoPad>"u"?0:e.autoPad],i=e.dilations,a=e.group,o=e.kernelShape,u=e.pads,p=e.strides,h=e.wIsConst(),k=e.outputPadding,C=e.outputShape;return{autoPad:n,format:s,dilations:i,group:a,kernelShape:o,outputPadding:k,outputShape:C,pads:u,strides:p,wIsConst:h,...t,cacheKey:`${e.format};${t.activation};`}},Su=(e,t)=>{if(!e||e.length!==2&&e.length!==3)throw new Error("Conv requires 2 or 3 inputs");if(e[0].dims.length!==4&&e[0].dims.length!==3)throw new Error("currently only support 2-dimensional conv");if(e[0].dims.length!==e[1].dims.length)throw new Error("filter does not have same dimension as input");let s=e[0].dims[t.format==="NHWC"?e[0].dims.length-1:1],n=e[1].dims[0];if(s!==n)throw new Error("FILTER_IN_CHANNEL should be equal to DATA_CHANNEL");let i=e[1].dims[1]*t.group;if(e.length===3&&(e[2].dims.length!==1||e[2].dims[0]!==i))throw new Error("invalid bias");let a=e[0].dims.length-2;if(t.dilations.reduce((o,u)=>o+u,0)>0&&t.dilations.length!==a)throw new Error(`dilations should be ${a}D`);if(t.strides.reduce((o,u)=>o+u,0)>0&&t.strides.length!==a)throw new Error(`strides should be ${a}D`);if(t.pads.reduce((o,u)=>o+u,0)>0&&t.pads.length!==a*2)throw new Error(`pads should be ${a*2}D`);if(t.outputPadding.length!==a&&t.outputPadding.length!==0)throw new Error(`output_padding should be ${a}D`);if(t.kernelShape.reduce((o,u)=>o+u,0)>0&&t.kernelShape.length!==0&&t.kernelShape.length!==e[1].dims.length-2)throw new Error("invalid kernel shape");if(t.outputShape.length!==0&&t.outputShape.length!==e[0].dims.length-2)throw new Error("invalid output shape")},ea=(e,t,s,n)=>{let i=e.kernelCustomData.wT??e.compute(pr(t[1],[2,3,0,1]),{inputs:[1],outputs:[s.wIsConst?-2:-1]})[0];s.wIsConst&&!e.kernelCustomData.wT&&(e.kernelCustomData.wT=i);let a=[t[0],i];t.length===3&&a.push(t[2]),e.compute(Eu(a,s,n),{inputs:a})},ta=(e,t)=>{let s=t.format==="NHWC",n=[e.inputs[0].reshape(s?[e.inputs[0].dims[0],1,e.inputs[0].dims[1],e.inputs[0].dims[2]]:[e.inputs[0].dims[0],e.inputs[0].dims[1],1,e.inputs[0].dims[2]]),e.inputs[1].reshape([e.inputs[1].dims[0],e.inputs[1].dims[1],1,e.inputs[1].dims[2]])];e.inputs.length===3&&n.push(e.inputs[2]);let i=t.kernelShape;(i.length===0||i[0]===0)&&(i=[e.inputs[1].dims[2]]);let a=t.dilations;(a.length===0||a[0]===0)&&(a=[1]);let o=t.strides;(o.length===0||o[0]===0)&&(o=[1]);let u=t.pads;u.length===0&&(u=[0,0]),u=[0,u[0],0,u[1]],o=[1].concat(o),a=[1].concat(a),i=[1].concat(i);let p=Jo({...t,pads:u,strides:o,dilations:a,kernelShape:i},n);ea(e,n,p,h=>s?[h[0],h[2],h[3]]:[h[0],h[1],h[3]])},$u=(e,t)=>{if(Su(e.inputs,t),e.inputs[0].dims.length===3)ta(e,t);else{let s=Jo(t,e.inputs);ea(e,e.inputs,s)}}}),sa,Au,Iu,Zc=g(()=>{zt(),Ot(),rs(),Yt(),sa=(e,t,s,n)=>{let i=ze.size(t),a=t.length,o=qe("input",e,a),u=It("output",e,a),p=s.dataType===6?s.getInt32Array()[0]:Number(s.getBigInt64Array()[0]),h=ze.normalizeAxis(p,a),k=C=>{let d=` i32(${o.indicesGet("inputIndices","uniforms.axis")}) `,z=$t("uniforms.input_shape","uniforms.axis",a),B=n.reverse?d+(n.exclusive?" + 1":""):"0",V=n.reverse?z:d+(n.exclusive?"":" + 1");return` + ${C.registerUniform("outputSize","u32").registerUniform("axis","u32").declareVariables(o,u)} + ${C.mainStart()} + ${C.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")} + var inputIndices = ${u.offsetToIndices("global_idx")}; + var sum = ${u.type.value}(0); + let first : i32 = ${B}; + let last : i32 = ${V}; + for (var i : i32 = first; i < last; i++) { + ${o.indicesSet("inputIndices","uniforms.axis","u32(i)")}; + sum = sum + ${o.getByIndices("inputIndices")}; + } + ${u.setByOffset("global_idx","sum")}; + }`};return{name:"CumSum",shaderCache:{hint:n.cacheKey,inputDependencies:["rank"]},getRunData:()=>({outputs:[{dims:t,dataType:e}],dispatchGroup:{x:Math.ceil(i/64)},programUniforms:[{type:12,data:i},{type:12,data:h},...yt(t,t)]}),getShaderSource:k}},Au=(e,t)=>{let s=e.inputs[0].dims,n=e.inputs[0].dataType,i=e.inputs[1];e.compute(sa(n,s,i,t),{inputs:[0]})},Iu=e=>{let t=e.exclusive===1,s=e.reverse===1;return Bt({exclusive:t,reverse:s})}}),Ou,ra,Fu,Du,Lu,ep=g(()=>{zt(),Ot(),rs(),Yt(),Ou=e=>{if(!e||e.length!==1)throw new Error("DepthToSpace requires 1 input.");if(e[0].dims.length!==4)throw new Error("DepthToSpace requires 4D input.")},ra=(e,t,s,n)=>{let i=[];i.push(`fn perm(i: ${n.type.indices}) -> ${s.type.indices} { + var a: ${s.type.indices};`);for(let a=0;a{let s,n,i,a,o,u,p=t.format==="NHWC",h=t.blocksize,k=t.mode==="DCR";p?([s,n,i,a]=e.dims,o=k?[s,n,i,h,h,a/h**2]:[s,n,i,a/h**2,h,h],u=k?[0,1,3,2,4,5]:[0,1,4,2,5,3]):([s,n,i,a]=[e.dims[0],e.dims[2],e.dims[3],e.dims[1]],o=k?[s,h,h,a/h**2,n,i]:[s,a/h**2,h,h,n,i],u=k?[0,3,4,1,5,2]:[0,1,4,2,5,3]);let C=e.reshape(o),d=C.dims.length,z=e.dataType,B=qe("a",z,d),V=It("output",z,d),Z=ee=>` + ${ee.registerUniform("output_size","u32").declareVariables(B,V)} + + ${ra(u,d,B,V)} + + ${ee.mainStart()} + ${ee.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + + let indices = ${V.offsetToIndices("global_idx")}; + let aIndices = perm(indices); + + ${V.setByOffset("global_idx",B.getByIndices("aIndices"))} + }`;return{name:"DepthToSpace",shaderCache:{hint:`${e.dims};${t.blocksize};${t.mode}`,inputDependencies:["rank"]},getRunData:ee=>{let Y=p?[s,n*h,i*h,a/h**2]:[s,a/h**2,n*h,i*h],he=ze.size(Y),pe=C.dims,Me=ze.sortBasedOnPerm(pe,u);return{outputs:[{dims:Y,dataType:ee[0].dataType}],dispatchGroup:{x:Math.ceil(he/64)},programUniforms:[{type:12,data:he},...yt(pe,Me)]}},getShaderSource:Z}},Du=(e,t)=>{Ou(e.inputs),e.compute(Fu(e.inputs[0],t))},Lu=e=>Bt({blocksize:e.blocksize,mode:e.mode,format:e.format})}),vi,Wn,xi,zu,Bu,na,Ru,ia,Hr,Nu,ju,tp=g(()=>{zt(),Ot(),rs(),Yt(),vi="[a-zA-Z]|\\.\\.\\.",Wn="("+vi+")+",xi="^"+Wn+"$",zu="("+Wn+",)*"+Wn,Bu="^"+zu+"$",na=class{constructor(e=-1){this.symbolToIndices=new Map,this.inputIndex=e}addSymbol(e,t){let s=this.symbolToIndices.get(e);s===void 0?s=[t]:s.push(t),this.symbolToIndices.set(e,s)}},Ru=class{constructor(e,t){var i;this.equation=t,this.hasEllipsis=!1,this.symbolToInfo=new Map,this.lhs=new Array,this.outputDims=[];let[s,n]=t.includes("->")?t.split("->",2):[t,""];if(!s.match(RegExp(Bu)))throw new Error("Invalid LHS term");if(s.split(",").forEach((a,o)=>{let u=e[o].dims.slice();if(!a.match(RegExp(xi)))throw new Error("Invalid LHS term");let p=this.processTerm(a,!0,u,o);this.lhs.push(p)}),n==="")n+=[...this.symbolToInfo.entries()].filter(([a,o])=>o.count===1||a==="...").map(([a])=>a).join("");else if(!n.match(RegExp(Wn)))throw new Error("Invalid RHS");(i=n.match(RegExp(vi,"g")))==null||i.forEach(a=>{if(a==="...")this.outputDims=this.outputDims.concat(this.ellipsisDims);else{let o=this.symbolToInfo.get(a);if(o===void 0)throw new Error("Invalid RHS symbol");this.outputDims.push(o.dimValue)}}),this.rhs=this.processTerm(n,!1,this.outputDims)}addSymbol(e,t,s){let n=this.symbolToInfo.get(e);if(n!==void 0){if(n.dimValue!==t&&n.count!==1)throw new Error("Dimension mismatch");n.count++,n.inputIndices.push(s)}else n={count:1,dimValue:t,inputIndices:[s]};this.symbolToInfo.set(e,n)}processTerm(e,t,s,n=-1){let i=s.length,a=!1,o=[],u=0;if(!e.match(RegExp(xi))&&!t&&e!=="")throw new Error("Invalid LHS term");let p=e.match(RegExp(vi,"g")),h=new na(n);return p==null||p.forEach((k,C)=>{if(k==="..."){if(a)throw new Error("Only one ellipsis is allowed per input term");a=!0;let d=i-p.length+1;if(d<0)throw new Error("Ellipsis out of bounds");if(o=s.slice(u,u+d),this.hasEllipsis){if(this.ellipsisDims.length!==o.length||this.ellipsisDims.toString()!==o.toString())throw new Error("Ellipsis dimensions mismatch")}else if(t)this.hasEllipsis=!0,this.ellipsisDims=o;else throw new Error("Ellipsis must be specified in the LHS");for(let z=0;ze+"_max",Hr=(e,t,s,n)=>{let i=e.map(h=>h.length).map((h,k)=>qe(`input${k}`,t,h)),a=ze.size(n),o=It("output",t,n.length),u=[...s.symbolToInfo.keys()].filter(h=>!s.rhs.symbolToIndices.has(h)),p=h=>{let k=[],C="var prod = 1.0;",d="var sum = 0.0;",z="sum += prod;",B=[],V=[],Z=[],ee=[],Y=s.symbolToInfo.size===s.rhs.symbolToIndices.size;s.symbolToInfo.forEach((pe,Me)=>{var Oe;if(s.rhs.symbolToIndices.has(Me)){let De=(Oe=s.rhs.symbolToIndices.get(Me))==null?void 0:Oe[0];De!==void 0&&s.lhs.forEach((Ye,at)=>{if(pe.inputIndices.includes(at)){let Pt=Ye.symbolToIndices.get(Me);if(Pt===void 0)throw new Error("Invalid symbol error");Pt.forEach(Xt=>{k.push(`${i[at].indicesSet(`input${at}Indices`,Xt,o.indicesGet("outputIndices",De))}`)})}})}else s.lhs.forEach((De,Ye)=>{if(pe.inputIndices.includes(Ye)){let at=De.symbolToIndices.get(Me);if(at===void 0)throw new Error("Invalid symbol error");at.forEach(Pt=>{B.push(`${i[Ye].indicesSet(`input${Ye}Indices`,Pt,`${Me}`)}`)}),ee.push(`prod *= ${i[Ye].getByIndices(`input${Ye}Indices`)};`)}}),V.push(`for(var ${Me}: u32 = 0; ${Me} < uniforms.${ia(Me)}; ${Me}++) {`),Z.push("}")});let he=Y?[...k,`let sum = ${i.map((pe,Me)=>pe.getByIndices(`input${Me}Indices`)).join(" * ")};`]:[...k,d,...V,...B,C,...ee,z,...Z];return` + ${h.registerUniforms(u.map(pe=>({name:`${ia(pe)}`,type:"u32"}))).registerUniform("outputSize","u32").declareVariables(...i,o)} + + ${h.mainStart()} + ${h.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")} + var outputIndices = ${o.offsetToIndices("global_idx")}; + ${i.map((pe,Me)=>`var input${Me}Indices: ${i[Me].type.indices};`).join(` +`)} + ${he.join(` +`)}; + ${o.setByOffset("global_idx","sum")}; + }`};return{name:"Einsum",shaderCache:{hint:s.equation,inputDependencies:e.map(()=>"rank")},getRunData:()=>{let h=u.filter(C=>s.symbolToInfo.has(C)).map(C=>{var d;return{type:12,data:((d=s.symbolToInfo.get(C))==null?void 0:d.dimValue)||0}});h.push({type:12,data:a});let k=e.map((C,d)=>[...yt(C)]).reduce((C,d)=>C.concat(d),h);return k.push(...yt(n)),{outputs:[{dims:n,dataType:t}],dispatchGroup:{x:Math.ceil(a/64)},programUniforms:k}},getShaderSource:p}},Nu=(e,t)=>{let s=new Ru(e.inputs,t.equation),n=s.outputDims,i=e.inputs.map((a,o)=>a.dims);e.compute(Hr(i,e.inputs[0].dataType,s,n))},ju=e=>{let t=e.equation.replace(/\s+/g,"");return Bt({equation:t})}}),Uu,Ti,Vu,Wu,Gu,sp=g(()=>{zt(),Ot(),Yt(),Uu=e=>{if(!e||e.length!==2)throw new Error("Expand requires 2 input.");let t=e[0].dims,s=Array.from(e[1].getBigInt64Array(),Number),n=s.length{let s=e.length-t.length,n=[];for(let i=0;ie.length>t.length?Ti(e,t):Ti(t,e),Wu=e=>{let t=e[0].dims,s=Array.from(e[1].getBigInt64Array(),Number),n=Vu(t,s),i=e[0].dataType,a=i===9||ze.size(t)===1,o=i===9||t.length>0&&t[t.length-1]%4===0?4:1,u=a||n.length>0&&n[n.length-1]%4===0?4:1,p=Math.ceil(ze.size(n)/u),h=C=>{let d=qe("input",i,t.length,o),z=It("output",i,n.length,u),B;if(i===9){let V=(Z,ee,Y="")=>` + let outputIndices${ee} = ${z.offsetToIndices(`outputOffset + ${ee}u`)}; + let offset${ee} = ${d.broadcastedIndicesToOffset(`outputIndices${ee}`,z)}; + let index${ee} = offset${ee} / 4u; + let component${ee} = offset${ee} % 4u; + ${Z}[${ee}] = ${Y}(${d.getByOffset(`index${ee}`)}[component${ee}]); + `;B=` + let outputOffset = global_idx * ${u}; + var data = vec4(0); + ${V("data",0,"u32")} + ${V("data",1,"u32")} + ${V("data",2,"u32")} + ${V("data",3,"u32")} + ${z.setByOffset("global_idx","data")} + }`}else B=` + let outputIndices = ${z.offsetToIndices(`global_idx * ${u}`)}; + let inputOffset = ${d.broadcastedIndicesToOffset("outputIndices",z)}; + let data = ${z.type.value}(${d.getByOffset(`inputOffset / ${o}`)}); + ${z.setByOffset("global_idx","data")} + }`;return` + ${C.registerUniform("vec_size","u32").declareVariables(d,z)} + ${C.mainStart()} + ${C.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.vec_size")} + ${B}`},k=[{type:12,data:p},...yt(t,n)];return{name:"Expand",shaderCache:{hint:`${n.length};${o}${u}`,inputDependencies:["rank"]},getShaderSource:h,getRunData:()=>({outputs:[{dims:n,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(p/64)},programUniforms:k})}},Gu=e=>{Uu(e.inputs),e.compute(Wu(e.inputs),{inputs:[0]})}}),Pi,Ku,rp=g(()=>{zt(),Ot(),Yt(),Co(),Pi=e=>{let t=e[0].dataType,s=ze.size(e[0].dims),n=ze.size(e[1].dims),i=n%4===0,a=o=>{let u=qe("x",t,[1],4),p=qe("bias",t,[1],4),h=It("y",t,[1],4),k=[{name:"output_vec_size",type:"u32"},{name:"bias_size",type:"u32"}],C=z=>` + let bias${z}_offset: u32 = (global_idx * 4 + ${z}) % uniforms.bias_size; + let bias${z} = ${p.getByOffset(`bias${z}_offset / 4`)}[bias${z}_offset % 4];`,d=i?` + let bias = ${p.getByOffset("global_idx % (uniforms.bias_size / 4)")};`:`${C(0)}${C(1)}${C(2)}${C(3)} + let bias = ${u.type.value}(bias0, bias1, bias2, bias3);`;return`${o.registerUniforms(k).declareVariables(u,p,h)} + + ${Po(Ss(t))} + + ${o.mainStart(ir)} + ${o.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_vec_size")} + + let x = ${u.getByOffset("global_idx")}; + ${d} + let x_in = x + bias; + ${h.setByOffset("global_idx",mi("x_in"))} + }`};return{name:"FastGeluWithBias",shaderCache:{hint:`${i}`,inputDependencies:["type","type"]},getShaderSource:a,getRunData:o=>({outputs:[{dims:o[0].dims,dataType:o[0].dataType}],programUniforms:[{type:12,data:Math.ceil(s/4)},{type:12,data:n}],dispatchGroup:{x:Math.ceil(s/ir/4)}})}},Ku=e=>{e.inputs.length<2||ze.size(e.inputs[1].dims)===0?Hl(e):e.compute(Pi(e.inputs))}}),Hu,Gn,qu,Qu,np=g(()=>{zt(),Ot(),rs(),Yt(),Hu=e=>{if(!e||e.length!==2)throw new Error("Gather requires 2 inputs.")},Gn=(e,t)=>{let s=e[0].dims,n=e[1].dims,i=s.length,a=ze.normalizeAxis(t.axis,i),o=s.slice(0);o.splice(a,1,...n);let u=s[a],p=e[0].dataType===9?4:1,h=Math.ceil(ze.size(o)/p),k=[{type:12,data:h},{type:6,data:u},{type:12,data:a},...yt(e[0].dims,e[1].dims,o)],C=d=>{let z=qe("data",e[0].dataType,e[0].dims.length,p),B=qe("inputIndices",e[1].dataType,e[1].dims.length),V=It("output",e[0].dataType,o.length,p),Z=Y=>{let he=n.length,pe=`var indicesIndices${Y} = ${B.type.indices}(0);`;for(let Me=0;Me1?`indicesIndices${Y}[${Me}]`:`indicesIndices${Y}`} = ${o.length>1?`outputIndices${Y}[uniforms.axis + ${Me}]`:`outputIndices${Y}`};`;pe+=` + var idx${Y} = ${B.getByIndices(`indicesIndices${Y}`)}; + if (idx${Y} < 0) { + idx${Y} = idx${Y} + uniforms.axisDimLimit; + } + var dataIndices${Y} : ${z.type.indices}; + `;for(let Me=0,Oe=0;Me1?`dataIndices${Y}[${Me}]`:`dataIndices${Y}`} = u32(idx${Y});`,Oe+=he):(pe+=`${i>1?`dataIndices${Y}[${Me}]`:`dataIndices${Y}`} = ${o.length>1?`outputIndices${Y}[${Oe}]`:`outputIndices${Y}`};`,Oe++);return pe},ee;if(e[0].dataType===9){let Y=(he,pe,Me="")=>` + let outputIndices${pe} = ${V.offsetToIndices(`outputOffset + ${pe}u`)}; + ${Z(pe)}; + let offset${pe} = ${z.indicesToOffset(`dataIndices${pe}`)}; + let index${pe} = offset${pe} / 4u; + let component${pe} = offset${pe} % 4u; + ${he}[${pe}] = ${Me}(${z.getByOffset(`index${pe}`)}[component${pe}]); + `;ee=` + let outputOffset = global_idx * ${p}; + var value = vec4(0); + ${Y("value",0,"u32")} + ${Y("value",1,"u32")} + ${Y("value",2,"u32")} + ${Y("value",3,"u32")} + ${V.setByOffset("global_idx","value")} + `}else ee=` + let outputIndices = ${V.offsetToIndices("global_idx")}; + ${Z("")}; + let value = ${z.getByIndices("dataIndices")}; + ${V.setByOffset("global_idx","value")}; + `;return` + ${d.registerUniform("outputSize","u32").registerUniform("axisDimLimit","i32").registerUniform("axis","u32").declareVariables(z,B,V)} + ${d.mainStart()} + ${d.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")} + ${ee} + }`};return{name:"Gather",shaderCache:{hint:t.cacheKey,inputDependencies:["rank","rank"]},getRunData:()=>({outputs:[{dims:o,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(h/64)},programUniforms:k}),getShaderSource:C}},qu=e=>Bt({axis:e.axis}),Qu=(e,t)=>{let s=e.inputs;Hu(s),e.compute(Gn(e.inputs,t))}}),Xu,Ei,Yu,ip=g(()=>{zt(),Ot(),Yt(),Xu=(e,t,s,n,i,a,o,u,p)=>{let h=[{type:12,data:a},{type:12,data:n},{type:12,data:i},{type:12,data:s},{type:12,data:o},{type:12,data:u},{type:12,data:p}],k=[a];h.push(...yt(t.dims,k));let C=d=>{let z=qe("indices_data",t.dataType,t.dims.length),B=It("input_slice_offsets_data",12,1,1),V=[z,B],Z=[{name:"output_size",type:"u32"},{name:"batch_dims",type:"u32"},{name:"input_dims",type:"u32",length:i.length},{name:"sizes_from_slice_dims_data",type:"u32",length:s.length},{name:"num_slices_per_batch",type:"u32"},{name:"input_batch_stride",type:"u32"},{name:"num_slice_dims",type:"u32"}];return` + ${d.registerUniforms(Z).declareVariables(...V)} + ${d.mainStart()} + ${d.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let batch_idx = global_idx / uniforms.num_slices_per_batch; + let base_offset = batch_idx * uniforms.input_batch_stride; + + let slice_indices_base_offset = global_idx * uniforms.num_slice_dims; + var relative_slice_offset = 0; + for (var dim_idx = 0u; dim_idx < uniforms.num_slice_dims; dim_idx ++) { + var index = i32(indices_data[dim_idx + slice_indices_base_offset].x); + let input_dim_idx = uniforms.batch_dims + dim_idx; + if (index < 0) { + ${i.length===1?"index += i32(uniforms.input_dims);":"index += i32(uniforms.input_dims[input_dim_idx]);"} + } + ${s.length===1?"relative_slice_offset += index * i32(uniforms.sizes_from_slice_dims_data);":"relative_slice_offset += index * i32(uniforms.sizes_from_slice_dims_data[dim_idx]);"} + } + + input_slice_offsets_data[global_idx] = base_offset + u32(relative_slice_offset); + }`};return e.compute({name:"computeSliceOffsets",shaderCache:{hint:`${i.length}_${s.length}`,inputDependencies:["rank"]},getRunData:()=>({outputs:[{dims:k,dataType:e.inputs[1].dataType}],dispatchGroup:{x:Math.ceil(a/64)},programUniforms:h}),getShaderSource:C},{inputs:[t],outputs:[-1]})[0]},Ei=(e,t)=>{let s=e.inputs,n=s[0].dims,i=s[0].dataType,a=s[1].dims,o=a[a.length-1],u=ze.sizeToDimension(a,a.length-1),p=ze.sizeFromDimension(n,t.batchDims+o),h=ze.sizeToDimension(n,t.batchDims),k=ze.sizeFromDimension(n,t.batchDims),C=u/h,d=new Array(o),z=p;for(let pe=0;pen.length)throw new Error("last dimension of indices must not be larger than rank of input tensor");let Z=a.slice(0,-1).concat(n.slice(V)),ee=ze.size(Z),Y=[{type:12,data:ee},{type:12,data:p},...yt(s[0].dims,B.dims,Z)],he=pe=>{let Me=qe("data",s[0].dataType,s[0].dims.length),Oe=qe("slice_offsets",12,B.dims.length),De=It("output",s[0].dataType,Z.length);return` + ${pe.registerUniform("output_size","u32").registerUniform("slice_size","u32").declareVariables(Me,Oe,De)} + ${pe.mainStart()} + ${pe.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let slice_offset = slice_offsets[global_idx / uniforms.slice_size]; + output[global_idx] = data[u32(slice_offset) + global_idx % uniforms.slice_size]; + }`};e.compute({name:"GatherND",shaderCache:{hint:t.cacheKey,inputDependencies:["rank","rank"]},getRunData:()=>({outputs:[{dims:Z,dataType:i}],dispatchGroup:{x:Math.ceil(ee/64)},programUniforms:Y}),getShaderSource:he},{inputs:[s[0],B]})},Yu=e=>({batchDims:e.batch_dims,cacheKey:""})}),Ju,op,Zu,ed,ap=g(()=>{zt(),Ot(),rs(),Yt(),Ju=(e,t)=>{if(e.length<3||e.length>4)throw new Error("GatherBlockQuantized requires 3 or 4 inputs.");let s=ze.normalizeAxis(t.quantizeAxis,e[0].dims.length),n=t.blockSize,i=e[0],a=e[2],o=e.length===4?e[3]:void 0;if(a.dims.length!==i.dims.length||!i.dims.map((u,p)=>p===s?Math.ceil(u/n)===a.dims[p]:u===a.dims[p]).reduce((u,p)=>u&&p,!0))throw new Error("Scales must have the same rank as the input tensor and the dims should match except on gatherAxis.");if(o){if(o.dataType!==i.dataType)throw new Error("Zero point must have the same data type as the input tensor.");if(o.dims.length!==a.dims.length||!o.dims.map((u,p)=>u===a.dims[p]).reduce((u,p)=>u&&p,!0))throw new Error("Zero point must have the same rank as the input tensor and the dims should match except on quantizeAxis.")}},op=(e,t)=>{let s=e[0].dims,n=e[1].dims,i=s.length,a=ze.normalizeAxis(t.gatherAxis,i),o=ze.normalizeAxis(t.quantizeAxis,i),u=s.slice(0);u.splice(a,1,...n);let p=ze.size(u),h=e[2].dataType,k=e[0].dataType===22,C=[{type:12,data:p},{type:12,data:o},{type:12,data:a},{type:12,data:t.blockSize},...yt(...e.map((z,B)=>z.dims),u)],d=z=>{let B=qe("data",e[0].dataType,e[0].dims.length),V=qe("inputIndices",e[1].dataType,e[1].dims.length),Z=qe("scales",e[2].dataType,e[2].dims.length),ee=e.length>3?qe("zeroPoint",e[3].dataType,e[3].dims.length):void 0,Y=It("output",h,u.length),he=[B,V,Z];ee&&he.push(ee);let pe=[{name:"output_size",type:"u32"},{name:"quantize_axis",type:"u32"},{name:"gather_axis",type:"u32"},{name:"block_size",type:"u32"}];return` + ${z.registerUniforms(pe).declareVariables(...he,Y)} + ${z.mainStart()} + let output_indices = ${Y.offsetToIndices("global_idx")}; + var indices_indices = ${V.type.indices}(0); + ${n.length>1?` + for (var i: u32 = 0; i < ${n.length}; i++) { + let index = ${Y.indicesGet("output_indices","uniforms.gather_axis + i")}; + ${V.indicesSet("indices_indices","i","index")}; + }`:`indices_indices = ${Y.indicesGet("output_indices","uniforms.gather_axis")};`}; + var data_indices = ${B.type.indices}(0); + for (var i: u32 = 0; i < uniforms.gather_axis; i++) { + let index = ${Y.indicesGet("output_indices","i")}; + ${B.indicesSet("data_indices","i","index")}; + } + var index_from_indices = ${V.getByIndices("indices_indices")}; + if (index_from_indices < 0) { + index_from_indices += ${s[a]}; + } + ${B.indicesSet("data_indices","uniforms.gather_axis","u32(index_from_indices)")}; + for (var i = uniforms.gather_axis + 1; i < ${u.length}; i++) { + let index = ${Y.indicesGet("output_indices",`i + ${n.length} - 1`)}; + ${B.indicesSet("data_indices","i","index")}; + } + let data_offset = ${B.indicesToOffset("data_indices")}; + let data_index = data_offset % 8; + // Convert 4-bit packed data to 8-bit packed data. + let packed_4bit_quantized_data = ${B.getByOffset("data_offset / 8")}; + let packed_8bit_quantized_data = (packed_4bit_quantized_data >> (4 * (data_index % 2))) & 0x0f0f0f0f; + let quantized_data_vec = ${k?"unpack4xI8":"unpack4xU8"}(u32(packed_8bit_quantized_data)); + let quantized_data = quantized_data_vec[data_index / 2]; + var scale_indices = data_indices; + let quantize_axis_index = ${Z.indicesGet("data_indices","uniforms.quantize_axis")} / uniforms.block_size; + ${Z.indicesSet("scale_indices","uniforms.quantize_axis","quantize_axis_index")}; + var scale = ${Z.getByIndices("scale_indices")}; + ${ee?` + let zero_point_indices = scale_indices; + let zero_point_offset = ${ee.indicesToOffset("zero_point_indices")}; + let zero_point_index = zero_point_offset % 8; + let packed_4bit_zero_points = ${ee.getByOffset("zero_point_offset / 8")}; + let packed_8bit_zero_points = (packed_4bit_zero_points >> (4 * (zero_point_index % 2))) & 0x0f0f0f0f; + let zero_point_vec = ${k?"unpack4xI8":"unpack4xU8"}(u32(packed_8bit_zero_points)); + let zero_point = zero_point_vec[zero_point_index / 2];`:"var zero_point = 0"}; + let dequantized_data = ${Ss(h)}(quantized_data - zero_point) * scale; + ${Y.setByOffset("global_idx","dequantized_data")}; + }`};return{name:"GatherBlockQuantized",shaderCache:{hint:`${t.cacheKey};${e.filter((z,B)=>B!==1).map(z=>z.dims.join("_")).join(";")}`,inputDependencies:Array.from({length:e.length},(z,B)=>"rank")},getRunData:()=>({outputs:[{dims:u,dataType:h}],dispatchGroup:{x:Math.ceil(p/64)},programUniforms:C}),getShaderSource:d}},Zu=(e,t)=>{let s=e.inputs;Ju(s,t),e.compute(op(e.inputs,t))},ed=e=>Bt({blockSize:e.blockSize,gatherAxis:e.gatherAxis,quantizeAxis:e.quantizeAxis})}),Pn,td,sd,rd,lp=g(()=>{zt(),Ot(),rs(),Yt(),Pn=e=>{if(!e||e.length!==2)throw new Error("GatherElements requires 2 inputs.");if(e[0].dims.length<1)throw new Error("GatherElements requires that the data input be rank >= 1.");if(e[0].dims.length!==e[1].dims.length)throw new Error(`GatherElements requires that the data input and + indices input tensors be of same rank.`)},td=(e,t)=>{let s=e[0].dims,n=e[0].dataType,i=s.length,a=e[1].dims,o=e[1].dataType,u=ze.normalizeAxis(t.axis,i),p=s[u],h=a.slice(0),k=ze.size(h),C=qe("input",n,i),d=qe("indicesInput",o,a.length),z=It("output",n,h.length),B=[{type:12,data:k},{type:6,data:p},{type:12,data:u}];return B.push(...yt(s,a,h)),{name:"GatherElements",shaderCache:{inputDependencies:["rank","rank"]},getRunData:()=>({outputs:[{dims:h,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(k/64)},programUniforms:B}),getShaderSource:V=>` + ${V.registerUniform("outputSize","u32").registerUniform("axisDimLimit","i32").registerUniform("axis","u32").declareVariables(C,d,z)} + ${V.mainStart()} + ${V.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")} + + let outputIndices = ${z.offsetToIndices("global_idx")}; + + var idx = ${d.getByOffset("global_idx")}; + if (idx < 0) { + idx = idx + uniforms.axisDimLimit; + } + var inputIndices = ${C.type.indices}(outputIndices); + ${C.indicesSet("inputIndices","uniforms.axis","u32(idx)")}; + let value = ${C.getByIndices("inputIndices")}; + + ${z.setByOffset("global_idx","value")}; + }`}},sd=e=>Bt({axis:e.axis}),rd=(e,t)=>{let s=e.inputs;Pn(s),e.compute(td(e.inputs,t))}}),nd,id,od,Ci,Kp=g(()=>{zt(),Ot(),Yt(),nd=e=>{if(!e)throw new Error("Input is missing");if(e.length<2||e.length>3)throw new Error("Invaid input number.");if(e.length===3&&e[2].dims.length>2)throw new Error("Invalid input shape of C");if(e[0].dataType!==e[1].dataType||e.length===3&&e[0].dataType!==e[2].dataType)throw new Error("Input types are mismatched")},id=(e,t)=>{let s=e[0].dims.slice(),n=e[1].dims.slice(),[i,a,o]=Fr.getShapeOfGemmResult(s,t.transA,n,t.transB,e.length===3?e[2].dims:void 0),u=[i,a];if(!u)throw new Error("Can't use gemm on the given tensors");let p=16,h=Math.ceil(a/p),k=Math.ceil(i/p),C=!0,d=ze.size(u),z=[{type:12,data:C?h:d},{type:12,data:i},{type:12,data:a},{type:12,data:o},{type:1,data:t.alpha},{type:1,data:t.beta}],B=["type","type"];e.length===3&&(z.push(...yt(e[2].dims)),B.push("rank")),z.push(...yt(u));let V=ee=>{let Y="";t.transA&&t.transB?Y="value += a[k * uniforms.M + m] * b[n * uniforms.K + k];":t.transA&&!t.transB?Y="value += a[k * uniforms.M + m] * b[k * uniforms.N + n];":!t.transA&&t.transB?Y="value += a[m * uniforms.K + k] * b[n * uniforms.K + k];":!t.transA&&!t.transB&&(Y="value += a[m * uniforms.K + k] * b[k * uniforms.N + n];");let he=t.alpha===1?"":"value *= uniforms.alpha;",pe=qe("a",e[0].dataType,e[0].dims),Me=qe("b",e[1].dataType,e[1].dims),Oe=pe.type.value,De=null,Ye=[pe,Me];e.length===3&&(De=qe("c",e[2].dataType,e[2].dims.length),Ye.push(De));let at=It("output",e[0].dataType,u.length);Ye.push(at);let Pt=[{name:"output_size",type:"u32"},{name:"M",type:"u32"},{name:"N",type:"u32"},{name:"K",type:"u32"},{name:"alpha",type:"f32"},{name:"beta",type:"f32"}];return` + ${ee.registerUniforms(Pt).declareVariables(...Ye)} + + ${ee.mainStart()} + ${ee.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + + let m = global_idx / uniforms.N; + let n = global_idx % uniforms.N; + + var value = ${Oe}(0); + for (var k: u32 = 0u; k < uniforms.K; k++) { + ${Y} + } + + ${he} + ${De!=null?`let cOffset = ${De.broadcastedIndicesToOffset("vec2(m, n)",at)}; value += ${Oe}(uniforms.beta) * ${De.getByOffset("cOffset")};`:""} + output[global_idx] = value; + }`},Z=ee=>{let Y=qe("a",e[0].dataType,e[0].dims),he=qe("b",e[1].dataType,e[1].dims),pe=null,Me=[Y,he];e.length===3&&(pe=qe("c",e[2].dataType,e[2].dims.length),Me.push(pe));let Oe=It("output",e[0].dataType,u.length);Me.push(Oe);let De=[{name:"num_tile_n",type:"u32"},{name:"M",type:"u32"},{name:"N",type:"u32"},{name:"K",type:"u32"},{name:"alpha",type:"f32"},{name:"beta",type:"f32"}],Ye="",at="";t.transA&&t.transB?(at=` + var col = tile_row_start + local_id.x; + var row = k_start + local_id.y; + if (col < uniforms.M && row < uniforms.K) { + tile_a[local_id.y][local_id.x] = a[row * uniforms.M + col]; + } else { + tile_a[local_id.y][local_id.x] = ${Y.type.value}(0); + } + + col = k_start + local_id.x; + row = tile_col_start + local_id.y; + if (col < uniforms.K && row < uniforms.N) { + tile_b[local_id.y][local_id.x] = b[row * uniforms.K + col]; + } else { + tile_b[local_id.y][local_id.x] = ${he.type.value}(0); + } + `,Ye="value += tile_a[k][local_id.y] * tile_b[local_id.x][k];"):t.transA&&!t.transB?(at=` + var col = tile_row_start + local_id.x; + var row = k_start + local_id.y; + if (col < uniforms.M && row < uniforms.K) { + tile_a[local_id.y][local_id.x] = a[row * uniforms.M + col]; + } else { + tile_a[local_id.y][local_id.x] = ${Y.type.value}(0); + } + + col = tile_col_start + local_id.x; + row = k_start + local_id.y; + if (col < uniforms.N && row < uniforms.K) { + tile_b[local_id.y][local_id.x] = b[row * uniforms.N + col]; + } else { + tile_b[local_id.y][local_id.x] = ${he.type.value}(0); + } + `,Ye="value += tile_a[k][local_id.y] * tile_b[k][local_id.x];"):!t.transA&&t.transB?(at=` + var col = k_start + local_id.x; + var row = tile_row_start + local_id.y; + if (col < uniforms.K && row < uniforms.M) { + tile_a[local_id.y][local_id.x] = a[row * uniforms.K + col]; + } else { + tile_a[local_id.y][local_id.x] = ${Y.type.value}(0); + } + + col = k_start + local_id.x; + row = tile_col_start + local_id.y; + if (col < uniforms.K && row < uniforms.N) { + tile_b[local_id.y][local_id.x] = b[row * uniforms.K + col]; + } else { + tile_b[local_id.y][local_id.x] = ${he.type.value}(0); + } + `,Ye="value += tile_a[local_id.y][k] * tile_b[local_id.x][k];"):!t.transA&&!t.transB&&(at=` + var col = k_start + local_id.x; + var row = tile_row_start + local_id.y; + if (col < uniforms.K && row < uniforms.M) { + tile_a[local_id.y][local_id.x] = a[row * uniforms.K + col]; + } else { + tile_a[local_id.y][local_id.x] = ${Y.type.value}(0); + } + + col = tile_col_start + local_id.x; + row = k_start + local_id.y; + if (col < uniforms.N && row < uniforms.K) { + tile_b[local_id.y][local_id.x] = b[row * uniforms.N + col]; + } else { + tile_b[local_id.y][local_id.x] = ${he.type.value}(0); + } + `,Ye="value += tile_a[local_id.y][k] * tile_b[k][local_id.x];");let Pt=t.alpha===1?"":"value *= uniforms.alpha;";return` + ${ee.registerUniforms(De).declareVariables(...Me)} + var tile_a: array, ${p}>; + var tile_b: array, ${p}>; + ${ee.mainStart([p,p,1])} + let tile_col_start = (workgroup_index % uniforms.num_tile_n) * ${p}; + let tile_row_start = (workgroup_index / uniforms.num_tile_n) * ${p}; + let num_tiles = (uniforms.K - 1) / ${p} + 1; + var k_start = 0u; + var value = ${Oe.type.value}(0); + for (var t: u32 = 0u; t < num_tiles; t++) { + ${at} + k_start = k_start + ${p}; + workgroupBarrier(); + + for (var k: u32 = 0u; k < ${p}; k++) { + ${Ye} + } + workgroupBarrier(); + } + + ${Pt} + let m = tile_row_start + local_id.y; + let n = tile_col_start + local_id.x; + ${pe!=null?`let cOffset = ${pe.broadcastedIndicesToOffset("vec2(m, n)",Oe)}; value += ${Oe.type.value}(uniforms.beta) * ${pe.getByOffset("cOffset")};`:""} + if (m < uniforms.M && n < uniforms.N) { + output[m * uniforms.N + n] = value; + } + }`};return C?{name:"GemmShared",shaderCache:{hint:`${t.cacheKey}`,inputDependencies:B},getRunData:()=>({outputs:[{dims:u,dataType:e[0].dataType}],dispatchGroup:{x:h*k},programUniforms:z}),getShaderSource:Z}:{name:"Gemm",shaderCache:{hint:`${t.cacheKey}`,inputDependencies:B},getRunData:()=>({outputs:[{dims:u,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(d/64)},programUniforms:z}),getShaderSource:V}},od=e=>{let t=e.transA,s=e.transB,n=e.alpha,i=e.beta;return{transA:t,transB:s,alpha:n,beta:i,cacheKey:`${e.transA};${e.transB};${e.alpha===1}`}},Ci=(e,t)=>{nd(e.inputs),e.compute(id(e.inputs,t))}}),Cr,Lr,ln,un,ad,oa,ld,ud,aa,dd,cd,la,pd,hd,ua=g(()=>{zt(),Ot(),rs(),Yt(),[Cr,Lr,ln,un]=[0,1,2,3],ad=e=>{if(e[0].dims.length!==4)throw new Error("only 4-D tensor is supported.");if(e[0].dims.length!==e[1].dims.length)throw new Error("input dimensions must be equal to grid dimensions");if(e[0].dims.length-2!==e[1].dims[e[1].dims.length-1])throw new Error(`last dimension of grid must be equal to ${e[0].dims.length-2}`);if(e[0].dims[0]!==e[1].dims[0])throw new Error("grid batch size must match input batch size")},oa=` + fn gs_get_cubic_coeffs(x: f32) -> vec4 { + let cubic_alpha = -0.75f; + let x_abs = abs(x); + var coeffs: vec4; + coeffs[0] = (((cubic_alpha * (x_abs + 1) - 5 * cubic_alpha) * (x_abs + 1) + 8 * cubic_alpha) * (x_abs + 1) - 4 * cubic_alpha); + coeffs[1] = (((cubic_alpha + 2) * x_abs - (cubic_alpha + 3)) * x_abs * x_abs + 1); + coeffs[2] = (((cubic_alpha + 2) * (1 - x_abs) - (cubic_alpha + 3)) * (1 - x_abs) * (1 - x_abs) + 1); + coeffs[3] = (((cubic_alpha * (2 - x_abs) - 5 * cubic_alpha) * (2 - x_abs) + 8 * cubic_alpha) * (2 - x_abs) - 4 * cubic_alpha); + return coeffs; + } +`,ld=e=>` + fn gs_bicubic_interpolate(p: mat4x4<${e}>, x: f32, y: f32) -> ${e} { + var v: vec4; + var coeffs = gs_get_cubic_coeffs(x); + for (var i = 0; i < 4; i++) { + v[i] = coeffs[0] * p[i][0] + coeffs[1] * p[i][1] + coeffs[2] * p[i][2] + coeffs[3] * p[i][3]; + } + coeffs = gs_get_cubic_coeffs(y); + let pixel = ${e}(coeffs[0] * v[0] + coeffs[1] * v[1] + coeffs[2] * v[2] + coeffs[3] * v[3]); + return pixel; + } +`,ud=e=>` + fn gs_denormalize(n: f32, length: i32) -> f32 { + ${e.alignCorners===0?` + // alignCorners: false => [-1, 1] to [-0.5, length - 0.5] + return ((n + 1.0) * f32(length) - 1.0) / 2.0; + `:` + // alignCorners: true => [-1, 1] to [0, length - 1] + return (n + 1.0) / 2.0 * (f32(length - 1)); + `} + } +`,aa=e=>` + ${e.paddingMode==="reflection"?` + fn gs_reflect(x: i32, x_min: f32, x_max: f32) -> u32 { + var dx = 0.0; + var fx = f32(x); + let range = x_max - x_min; + if (fx < x_min) { + dx = x_min - fx; + let n = u32(dx / range); + let r = dx - f32(n) * range; + if (n % 2 == 0) { + fx = x_min + r; + } else { + fx = x_max - r; + } + } else if (fx > x_max) { + dx = fx - x_max; + let n = u32(dx / range); + let r = dx - f32(n) * range; + if (n % 2 == 0) { + fx = x_max - r; + } else { + fx = x_min + r; + } + } + return u32(fx); + }`:""} +`,dd=(e,t,s)=>` + fn pixel_at_grid(r: i32, c: i32, H: i32, W: i32, batch: u32, channel: u32, border: vec4) -> ${t} { + var pixel = ${t}(0); + var indices = vec4(0); + indices[${Cr}] = batch; + indices[${Lr}] = channel;`+(()=>{switch(s.paddingMode){case"zeros":return` + if (r >= 0 && r < H && c >=0 && c < W) { + indices[${ln}] = u32(r); + indices[${un}] = u32(c); + } + `;case"border":return` + indices[${ln}] = u32(clamp(r, 0, H - 1)); + indices[${un}] = u32(clamp(c, 0, W - 1)); + `;case"reflection":return` + indices[${ln}] = gs_reflect(r, border[1], border[3]); + indices[${un}] = gs_reflect(c, border[0], border[2]); + `;default:throw new Error(`padding mode ${s.paddingMode} is not supported`)}})()+` + return ${e.getByIndices("indices")}; + } +`,cd=(e,t,s)=>(()=>{switch(s.mode){case"nearest":return` + let result = pixel_at_grid(i32(round(y)), i32(round(x)), H_in, W_in, indices[${Cr}], indices[${Lr}], border); + `;case"bilinear":return` + let x1 = i32(floor(x)); + let y1 = i32(floor(y)); + let x2 = x1 + 1; + let y2 = y1 + 1; + + let p11 = pixel_at_grid(y1, x1, H_in, W_in, indices[${Cr}], indices[${Lr}], border); + let p12 = pixel_at_grid(y1, x2, H_in, W_in, indices[${Cr}], indices[${Lr}], border); + let p21 = pixel_at_grid(y2, x1, H_in, W_in, indices[${Cr}], indices[${Lr}], border); + let p22 = pixel_at_grid(y2, x2, H_in, W_in, indices[${Cr}], indices[${Lr}], border); + + let dx2 = ${t}(f32(x2) - x); + let dx1 = ${t}(x - f32(x1)); + let dy2 = ${t}(f32(y2) - y); + let dy1 = ${t}(y - f32(y1)); + let result = dy2 * (dx2 * p11 + dx1 * p12) + dy1 * (dx2 * p21 + dx1 * p22); + `;case"bicubic":return` + let x0 = i32(floor(x)) - 1; + let y0 = i32(floor(y)) - 1; + var p: mat4x4<${t}>; + for (var h = 0; h < 4; h++) { + for (var w = 0; w < 4; w++) { + p[h][w] = pixel_at_grid(h + y0, w + x0, H_in, W_in, indices[${Cr}], indices[${Lr}], border); + } + } + + let dx = x - f32(x0 + 1); + let dy = y - f32(y0 + 1); + let result = gs_bicubic_interpolate(p, dx, dy); + `;default:throw new Error(`mode ${s.mode} is not supported`)}})()+`${e.setByOffset("global_idx","result")}`,la=(e,t)=>{let s=qe("x",e[0].dataType,e[0].dims.length),n=[e[1].dims[0],e[1].dims[1],e[1].dims[2]],i=qe("grid",e[1].dataType,n.length,2),a=[e[0].dims[0],e[0].dims[1],e[1].dims[1],e[1].dims[2]];t.format==="NHWC"&&(a=[e[0].dims[0],e[1].dims[1],e[1].dims[2],e[0].dims[3]],[Cr,Lr,ln,un]=[0,3,1,2]);let o=It("output",e[0].dataType,a.length),u=s.type.value,p=ze.size(a),h=[{type:12,data:p},...yt(e[0].dims,n,a)],k=C=>` + ${C.registerUniform("output_size","u32").declareVariables(s,i,o)} + ${oa} + ${ld(u)} + ${ud(t)} + ${aa(t)} + ${dd(s,u,t)} + + ${C.mainStart()} + ${C.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let H_in = i32(uniforms.x_shape[${ln}]); + let W_in = i32(uniforms.x_shape[${un}]); + + ${t.alignCorners===0?` + let x_min = -0.5; + let x_max = f32(W_in) - 0.5; + let y_min = -0.5; + let y_max = f32(H_in) - 0.5; + `:` + let x_min = 0.0; + let x_max = f32(W_in) - 1.0; + let y_min = 0.0; + let y_max = f32(H_in) - 1.0; + `}; + let border = vec4(x_min, y_min, x_max, y_max); + + let indices = ${o.offsetToIndices("global_idx")}; + var grid_indices = vec3(indices[${Cr}], indices[${ln}], indices[${un}]); + let nxy = ${i.getByIndices("grid_indices")}; + var x = gs_denormalize(f32(nxy[0]), W_in); + var y = gs_denormalize(f32(nxy[1]), H_in); + + ${cd(o,u,t)} + }`;return{name:"GridSample",shaderCache:{hint:`${t.cacheKey}`,inputDependencies:["type","type"]},getRunData:C=>{let d=ze.size(a);return{outputs:[{dims:a,dataType:C[0].dataType}],dispatchGroup:{x:Math.ceil(d/64)},programUniforms:h}},getShaderSource:k}},pd=(e,t)=>{ad(e.inputs),e.compute(la(e.inputs,t))},hd=e=>Bt({alignCorners:e.align_corners,mode:e.mode,paddingMode:e.padding_mode,format:e.format})}),nr,md,fd,da,ca,dn,up,_d=g(()=>{zt(),Ot(),rs(),ue(),uo(),Yt(),Kr(),nr=(e,t)=>e.length>t&&e[t].dims.length>0?e[t]:void 0,md=(e,t)=>{let s=e[0],n=nr(e,1),i=nr(e,2),a=nr(e,3),o=nr(e,4),u=nr(e,5),p=nr(e,6),h=nr(e,7);if(s.dims.length!==3&&s.dims.length!==5)throw new Error("Input query is expected to have 3 or 5 dimensions");let k=s.dims[0],C=s.dims[1],d=s.dims.length===3?s.dims[2]:t.numHeads*s.dims[4],z=C,B=0,V=0,Z=Math.floor(d/t.numHeads);if(p&&h&&ze.size(p.dims)&&ze.size(h.dims)){if(p.dims.length!==4)throw new Error('Input "past_key" is expected to have 4 dimensions');if(p.dims[0]!==k||p.dims[1]!==t.numHeads||p.dims[3]!==Z)throw new Error('Input "past_key" shape (batch_size, num_heads, past_sequence_length, head_size)');if(h.dims[0]!==k||h.dims[1]!==t.numHeads||h.dims[3]!==Z)throw new Error('Input "past_value" shape (batch_size, num_heads, past_sequence_length, head_size)');if(p.dims[2]!==h.dims[2])throw new Error('Input "past_key" and "past_value" shall have same dim 2 (past_sequence_length)');if(h.dims.length!==4)throw new Error('Input "past_value" is expected to have 4 dimensions');B=p.dims[2],V=p.dims[2]}else if(p&&ze.size(p.dims)||h&&ze.size(h.dims))throw new Error('Input "past_key" and "past_value" shall be both present or both absent');let ee;if(n&&ze.size(n.dims)>0){if(s.dims.length!==3)throw new Error('Input "query" is expected to have 3 dimensions when key is given');if(n.dims.length<3||n.dims.length>5)throw new Error('Input "key" is expected to have 3, 4, or 5 dimensions');if(s.dims[0]!==n.dims[0])throw new Error('Input "query" and "key" shall have same dim 0 (batch size)');if(n.dims.length===3){if(n.dims[2]!==s.dims[2])throw new Error('Input "query" and "key" shall have same dim 2 (hidden_size)');ee=2,z=n.dims[1]}else if(n.dims.length===5){if(n.dims[2]!==t.numHeads||n.dims[3]!==2||n.dims[4]!==Z)throw new Error('Expect "key" shape (batch_size, kv_sequence_length, num_heads, 2, head_size) for packed kv');if(i)throw new Error('Expect "value" be none when "key" has packed kv format.');ee=5,z=n.dims[1]}else{if(n.dims[1]!==t.numHeads||n.dims[3]!==Z)throw new Error('Expect "key" shape (batch_size, num_heads, kv_sequence_length, head_size) for past_key');ee=0,z=n.dims[2]}}else{if(s.dims.length!==5)throw new Error('Input "query" is expected to have 5 dimensions when key is empty');if(s.dims[2]!==t.numHeads||s.dims[3]!==3)throw new Error('Expect "query" shape (batch_size, kv_sequence_length, num_heads, 3, head_size) for packed kv');ee=3}if(a&&ze.size(a.dims)>0){if(a.dims.length!==1)throw new Error('Input "bias" is expected to have 1 dimension');if(n&&n.dims.length===5&&n.dims[3]===2)throw new Error("bias is not allowed for packed kv.")}let Y=B+z,he=0;if(o&&ze.size(o.dims)>0){he=8;let De=o.dims;throw De.length===1?De[0]===k?he=1:De[0]===3*k+2&&(he=3):De.length===2&&De[0]===k&&De[1]===Y&&(he=5),he===8?new Error('Input "key_padding_mask" shape shall be (batch_size) or (batch_size, total_sequence_length)'):new Error("Mask not supported")}let pe=!1,Me=d;if(i&&ze.size(i.dims)>0){if(i.dims.length!==3&&i.dims.length!==4)throw new Error('Input "value" is expected to have 3 or 4 dimensions');if(s.dims[0]!==i.dims[0])throw new Error('Input "query" and "value" shall have same dim 0 (batch_size)');if(i.dims.length===3){if(z!==i.dims[1])throw new Error('Input "key" and "value" shall have the same dim 1 (kv_sequence_length)');Me=i.dims[2]}else{if(z!==i.dims[2])throw new Error('Input "key" and "value" shall have the same dim 2 (kv_sequence_length)');Me=i.dims[1]*i.dims[3],pe=!0}}let Oe=!1;if(o&&ze.size(o.dims)>0)throw new Error("Key padding mask is not supported");if(u&&ze.size(u.dims)>0){if(u.dims.length!==4)throw new Error('Input "attention_bias" is expected to have 4 dimensions');if(u.dims[0]!==k||u.dims[1]!==t.numHeads||u.dims[2]!==C||u.dims[3]!==Y)throw new Error('Expect "attention_bias" shape (batch_size, num_heads, sequence_length, total_sequence_length)')}return{batchSize:k,sequenceLength:C,pastSequenceLength:B,kvSequenceLength:z,totalSequenceLength:Y,maxSequenceLength:V,inputHiddenSize:0,hiddenSize:d,vHiddenSize:Me,headSize:Z,vHeadSize:Math.floor(Me/t.numHeads),numHeads:t.numHeads,isUnidirectional:!1,pastPresentShareBuffer:!1,maskFilterValue:t.maskFilterValue,maskType:he,scale:t.scale,broadcastResPosBias:Oe,passPastInKv:pe,qkvFormat:ee}},fd=e=>Bt({...e}),da=Bt({perm:[0,2,1,3]}),ca=(e,t,s,n,i,a,o)=>{let u=[n,i,a],p=ze.size(u),h=[{type:12,data:p},{type:12,data:o},{type:12,data:a}],k=C=>{let d=It("qkv_with_bias",t.dataType,u),z=qe("qkv",t.dataType,u),B=qe("bias",s.dataType,u),V=[{name:"output_size",type:"u32"},{name:"bias_offset",type:"u32"},{name:"hidden_size",type:"u32"}];return` + ${C.registerUniforms(V).declareVariables(z,B,d)} + ${C.mainStart()} + ${C.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let bias_offset_idx = (global_idx % uniforms.hidden_size) + uniforms.bias_offset; + + qkv_with_bias[global_idx] = qkv[global_idx] + bias[bias_offset_idx]; + }`};return e.compute({name:"MultiHeadAttentionAddBias",shaderCache:{inputDependencies:["type","type"]},getRunData:()=>({outputs:[{dims:u,dataType:t.dataType,gpuDataType:0}],dispatchGroup:{x:Math.ceil(p/64)},programUniforms:h}),getShaderSource:k},{inputs:[t,s],outputs:[-1]})[0]},dn=(e,t,s,n,i,a,o,u)=>{let p=a;if(o&&ze.size(o.dims)>0){if(n===1)throw new Error("AddBiasReshape is not implemented. Please export your model with packed QKV or KV");return p=ca(e,a,o,t,n,s*i,u),p=p.reshape([t,n,s,i]),s===1||n===1?p:e.compute(pr(p,da.perm),{inputs:[p],outputs:[-1]})[0]}else return a.dims.length===3&&(p=a.reshape([t,n,s,i])),s===1||n===1?p:e.compute(pr(p,da.perm),{inputs:[p],outputs:[-1]})[0]},up=(e,t)=>{let s=md(e.inputs,t),n=e.inputs[0],i=nr(e.inputs,1),a=nr(e.inputs,2),o=nr(e.inputs,3),u=nr(e.inputs,4),p=nr(e.inputs,5),h=nr(e.inputs,6),k=nr(e.inputs,7);if(n.dims.length===5)throw new Error("Packed QKV is not implemented");if((i==null?void 0:i.dims.length)===5)throw new Error("Packed KV is not implemented");let C=i&&a&&i.dims.length===4&&a.dims.length===4,d=dn(e,s.batchSize,s.numHeads,s.sequenceLength,s.headSize,n,o,0);if(C)return Nn(e,d,i,a,u,void 0,h,k,p,s);if(!i||!a)throw new Error("key and value must be provided");let z=dn(e,s.batchSize,s.numHeads,s.kvSequenceLength,s.headSize,i,o,s.hiddenSize),B=dn(e,s.batchSize,s.numHeads,s.kvSequenceLength,s.vHeadSize,a,o,2*s.hiddenSize);Nn(e,d,z,B,u,void 0,h,k,p,s)}}),gd,pa,wd,yd,ki,Md,bd,ha=g(()=>{zt(),Ot(),rs(),Yt(),gd=e=>{if(!e||e.length<1)throw new Error("too few inputs")},pa=(e,t)=>{let s=[],n=t.numOutputs;return e[1].dims[0]>0&&(e[1].getBigInt64Array().forEach(i=>s.push(Number(i))),n=s.length),Bt({numOutputs:n,axis:t.axis,splitSizes:s})},wd=e=>` +fn calculateOutputIndex(index: u32) -> u32 { + for (var i: u32 = 0u; i < ${e}u; i += 1u ) { + if (index < ${$t("uniforms.size_in_split_axis","i",e)}) { + return i; + } + } + return ${e}u; +}`,yd=e=>{let t=e.length,s=[];for(let n=0;n{let s=e[0].dims,n=ze.size(s),i=e[0].dataType,a=ze.normalizeAxis(t.axis,s.length),o=new Array(t.numOutputs),u=qe("input",i,s.length),p=new Array(t.numOutputs),h=[],k=[],C=0,d=[{type:12,data:n}];for(let B=0;B` + ${B.registerUniform("input_size","u32").registerUniform("size_in_split_axis","u32",p.length).declareVariables(u,...o)} + ${wd(p.length)} + ${yd(o)} + + ${B.mainStart()} + ${B.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.input_size")} + + var indices = ${u.offsetToIndices("global_idx")}; + var index = ${u.indicesGet("indices",a)}; + let output_number = calculateOutputIndex(index); + if (output_number != 0) { + index -= ${$t("uniforms.size_in_split_axis","output_number - 1u",p.length)}; + ${u.indicesSet("indices",a,"index")}; + } + writeBufferData(output_number, indices, global_idx); + }`;return{name:"Split",shaderCache:{hint:t.cacheKey,inputDependencies:["rank"]},getShaderSource:z,getRunData:()=>({outputs:h,dispatchGroup:{x:Math.ceil(n/64)},programUniforms:d})}},Md=(e,t)=>{gd(e.inputs);let s=e.inputs.length===1?t:pa(e.inputs,t);e.compute(ki(e.inputs,s),{inputs:[0]})},bd=e=>{let t=e.axis,s=e.splitSizes,n=e.numOutputs<0?s.length:e.numOutputs;if(n!==s.length)throw new Error("numOutputs and splitSizes lengh must be equal");return Bt({axis:t,numOutputs:n,splitSizes:s})}}),dp,cp,Si,ma,pp=g(()=>{rs(),uo(),_d(),ha(),Kr(),dp=(e,t)=>{if(t.doRotary)throw new Error("GroupQuerryAttention do_rotary attribute is not supported");if(t.doRotary&&e.length<=7)throw new Error("cos_cache and sin_cache inputs are required if do_rotary is specified");let s=e[0],n=e[1],i=e[2],a=e[3],o=e[4];if(t.localWindowSize!==-1)throw new Error("Local attention is not supported");if(t.softcap!==0)throw new Error("Softcap is not supported");if(t.rotaryInterleaved!==0)throw new Error("Rotary interleaved is not supported");if(t.smoothSoftmax)throw new Error("Smooth softmax is not supported");if(s.dims.length!==3&&s.dims.length!==5)throw new Error("Input query is expected to have 3 or 5 dimensions");let u=!1,p=s.dims[0],h=s.dims[1],k=s.dims.length===3?u?s.dims[2]/3:s.dims[2]:t.numHeads*s.dims[4],C=h,d=0,z=!n||n.dims.length===0,B=Math.floor(z?k/(t.numHeads+2*t.kvNumHeads):k/t.numHeads);z&&(k=B*t.numHeads);let V=a&&a.dims.length!==0,Z=o&&o.dims.length!==0;if(V&&a.dims.length===4&&a.dims[0]===p&&a.dims[1]!==t.kvNumHeads&&a.dims[2]===t.kvNumHeads&&a.dims[3]===B)throw new Error("BSNH pastKey/pastValue is not supported");if(V&&Z){if(a.dims.length!==4)throw new Error('Input "past_key" is expected to have 4 dimensions');if(o.dims.length!==4)throw new Error('Input "past_value" is expected to have 4 dimensions');d=a.dims[2]}else if(V||Z)throw new Error('Input "past_key" and "past_value" shall be both present or both absent');let ee=1;if(n&&n.dims.length>0){if(s.dims.length!==3)throw new Error('Input "query" is expected to have 3 dimensions when key is given');if(n.dims.length<3||n.dims.length>5)throw new Error('Input "key" is expected to have 3, 4, or 5 dimensions');if(s.dims[0]!==n.dims[0])throw new Error('Input "query" and "key" shall have same dim 0 (batch size)');if(n.dims.length===3){if(s.dims[2]%n.dims[2]!==0)throw new Error('Dimension 2 of "query" should be a multiple of "key"');C=n.dims[1]}else if(n.dims.length===5){if(n.dims[2]!==t.numHeads||n.dims[3]!==2||n.dims[4]!==B)throw new Error('Expect "key" shape (batch_size, kv_sequence_length, num_heads, 2, head_size) for packed kv');if(i)throw new Error('Expect "value" be none when "key" has packed kv format.');C=n.dims[1]}else{if(n.dims[1]!==t.numHeads||n.dims[3]!==B)throw new Error('Expect "key" shape (batch_size, num_heads, kv_sequence_length, head_size) for past_key');C=n.dims[2]}}else{if(s.dims.length!==3&&s.dims.length!==5)throw new Error('Input "query" is expected to have 3 or 5 dimensions when key is empty');if(s.dims.length===5&&(s.dims[2]!==t.numHeads||s.dims[3]!==3))throw new Error('Expect "query" shape (batch_size, kv_sequence_length, num_heads, 3, head_size) for packed kv');ee=3}let Y=0,he=!1,pe=t.kvNumHeads?B*t.kvNumHeads:k;if(i&&i.dims.length>0){if(i.dims.length!==3&&i.dims.length!==4)throw new Error('Input "value" is expected to have 3 or 4 dimensions');if(s.dims[0]!==i.dims[0])throw new Error('Input "query" and "value" shall have same dim 0 (batch_size)');if(i.dims.length===3){if(C!==i.dims[1])throw new Error('Input "key" and "value" shall have the same dim 1 (kv_sequence_length)');pe=i.dims[2]}else{if(C!==i.dims[2])throw new Error('Input "past_key" and "past_value" shall have the same dim 2 (kv_sequence_length)');pe=i.dims[1]*i.dims[3],he=!0}}let Me=e.length>4?e[5]:void 0;if(Me&&Me.dims.length!==1&&Me.dims[0]!==p)throw new Error('Input "seqlens" is expected to have 1 dimension and the same dim 0 as batch_size');return{batchSize:p,sequenceLength:h,pastSequenceLength:d,kvSequenceLength:C,totalSequenceLength:-1,maxSequenceLength:-1,inputHiddenSize:0,hiddenSize:k,vHiddenSize:pe,headSize:B,vHeadSize:Math.floor(pe/t.kvNumHeads),numHeads:t.numHeads,kvNumHeads:t.kvNumHeads,nReps:t.numHeads/t.kvNumHeads,pastPresentShareBuffer:!1,maskType:Y,scale:t.scale,broadcastResPosBias:!1,passPastInKv:he,qkvFormat:ee}},cp=Bt({perm:[0,2,1,3]}),Si=(e,t,s)=>{let n=t,i=s.kvNumHeads;return t.dims.length===3&&s.kvSequenceLength!==0&&(n=t.reshape([s.batchSize,s.kvSequenceLength,i,s.headSize]),n=e.compute(pr(n,cp.perm),{inputs:[n],outputs:[-1]})[0]),n},ma=(e,t)=>{var Z;let s=dp(e.inputs,t);if(e.inputs[0].dims.length===5)throw new Error("Packed QKV is not implemented");if(((Z=e.inputs[1])==null?void 0:Z.dims.length)===5)throw new Error("Packed KV is not implemented");let n=e.inputs[0],i=e.inputs[1]&&e.inputs[1].dims.length>0?e.inputs[1]:void 0,a=e.inputs[2]&&e.inputs[2].dims.length>0?e.inputs[2]:void 0,o=e.inputs[3]&&e.inputs[3].dims.length!==0?e.inputs[3]:void 0,u=e.inputs[4]&&e.inputs[4].dims.length!==0?e.inputs[4]:void 0,p=e.inputs.length>4?e.inputs[5]:void 0,h=e.inputs.length>5?e.inputs[6]:void 0,k=s.kvNumHeads?s.kvNumHeads:s.numHeads,C=Bt({axis:2,numOutputs:3,splitSizes:[s.numHeads*s.headSize,k*s.headSize,k*s.headSize]}),[d,z,B]=!i&&!a?e.compute(ki([n],C),{inputs:[n],outputs:[-1,-1,-1]}):[n,i,a],V=dn(e,s.batchSize,s.numHeads,s.sequenceLength,s.headSize,d,void 0,0);Nn(e,V,Si(e,z,s),Si(e,B,s),void 0,void 0,o,u,void 0,s,p,h)}}),fa,_a,vd,xd,Td=g(()=>{zt(),Ot(),Kr(),Yt(),fa=(e,t,s,n,i,a,o,u)=>{let p=qt(a),h=p===1?"f32":`vec${p}f`,k=p===1?"vec2f":`mat2x${p}f`,C=i*o,d=64;C===1&&(d=256);let z=[i,o,a/p],B=[i,o,2],V=["rank","type","type"],Z=[];Z.push(...yt(z,B));let ee=Y=>{let he=qe("x",t.dataType,3,p),pe=qe("scale",s.dataType,s.dims),Me=qe("bias",n.dataType,n.dims),Oe=It("output",1,3,2),De=[he,pe,Me,Oe];return` + var workgroup_shared : array<${k}, ${d}>; + const workgroup_size = ${d}u; + ${Y.declareVariables(...De)} + ${Y.mainStart(d)} + let batch = workgroup_index / uniforms.x_shape[1]; + let channel = workgroup_index % uniforms.x_shape[1]; + let hight = uniforms.x_shape[2]; + // initialize workgroup memory + var sum = ${h}(0); + var squared_sum = ${h}(0); + for (var h = local_idx; h < hight; h += workgroup_size) { + let value = ${h}(${he.get("batch","channel","h")}); + sum += value; + squared_sum += value * value; + } + workgroup_shared[local_idx] = ${k}(sum, squared_sum); + workgroupBarrier(); + + for (var currSize = workgroup_size >> 1; currSize > 0; currSize = currSize >> 1) { + if (local_idx < currSize) { + workgroup_shared[local_idx] = workgroup_shared[local_idx] + workgroup_shared[local_idx + currSize]; + } + workgroupBarrier(); + } + if (local_idx == 0) { + let sum_final = ${Gs("workgroup_shared[0][0]",p)} / f32(hight * ${p}); + let squared_sum_final = ${Gs("workgroup_shared[0][1]",p)} / f32(hight * ${p}); + + let inv_std_dev = inverseSqrt(squared_sum_final - sum_final * sum_final + f32(${u})); + let channel_scale = inv_std_dev * f32(scale[channel]); + let channel_shift = f32(bias[channel]) - sum_final * channel_scale; + output[workgroup_index] = vec2f(channel_scale, channel_shift); + } + }`};return e.compute({name:"InstanceNormComputeChannelScaleShift",shaderCache:{hint:`${p};${u};${d}`,inputDependencies:V},getRunData:()=>({outputs:[{dims:B,dataType:1}],dispatchGroup:{x:C},programUniforms:Z}),getShaderSource:ee},{inputs:[t,s,n],outputs:[-1]})[0]},_a=(e,t,s)=>{let n=t[0].dims,i=n,a=2,o=n[0],u=n[1],p=ze.sizeFromDimension(n,a),h=qt(p),k=ze.size(i)/h,C=fa(e,t[0],t[1],t[2],o,p,u,s.epsilon),d=[o,u,p/h],z=[o,u],B=["type","none"],V=Z=>{let ee=qe("x",t[0].dataType,d.length,h),Y=qe("scale_shift",1,z.length,2),he=It("output",t[0].dataType,d.length,h),pe=[ee,Y,he];return` + ${Z.registerUniform("output_size","u32").declareVariables(...pe)} + ${Z.mainStart()} + ${Z.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let outputIndices = ${he.offsetToIndices("global_idx")}; + let batch = outputIndices[0]; + let channel = outputIndices[1]; + let scale_shift = ${Y.getByIndices("vec2(batch, channel)")}; + let value = ${ee.getByOffset("global_idx")} * ${he.type.value}(scale_shift.x) + ${he.type.value}(scale_shift.y); + ${he.setByOffset("global_idx","value")}; + }`};e.compute({name:"InstanceNormalization",shaderCache:{hint:`${h}`,inputDependencies:B},getRunData:()=>({outputs:[{dims:i,dataType:t[0].dataType}],dispatchGroup:{x:Math.ceil(k/64)},programUniforms:[{type:12,data:k},...yt(d,z,d)]}),getShaderSource:V},{inputs:[t[0],C]})},vd=(e,t,s)=>{let n=t[0].dims,i=n,a=n[0],o=n[n.length-1],u=ze.sizeFromDimension(n,1)/o,p=qt(o),h=ze.size(i)/p,k=[{type:12,data:u},{type:12,data:Math.floor(o/p)}],C=["type","type"],d=!1,z=[0,n.length-1];for(let ee=0;een[z[Y]])),V=fa(e,B,t[1],t[2],a,u,o,s.epsilon),Z=ee=>{let Y=fs(t[0].dataType),he=p===1?"vec2f":`mat${p}x2f`,pe=De=>{let Ye=De===0?"x":"y",at=p===1?"f32":`vec${p}f`;switch(p){case 1:return`${Y}(${at}(scale.${Ye}))`;case 2:return`vec2<${Y}>(${at}(scale[0].${Ye}, scale[1].${Ye}))`;case 4:return`vec4<${Y}>(${at}(scale[0].${Ye}, scale[1].${Ye}, scale[2].${Ye}, scale[3].${Ye}))`;default:throw new Error(`Not supported compoents ${p}`)}},Me=qe("input",t[0].dataType,t[0].dims,p),Oe=It("output",t[0].dataType,i,p);return` + @group(0) @binding(0) var input : array<${Me.type.storage}>; + @group(0) @binding(1) var scale_input : array<${he}>; + @group(0) @binding(2) var output : array<${Oe.type.storage}>; + struct Uniforms {H: u32, C : u32}; + @group(0) @binding(3) var uniforms: Uniforms; + + ${ee.mainStart()} + let current_image_number = global_idx / (uniforms.C * uniforms.H); + let current_channel_number = global_idx % uniforms.C; + + let scale_offset = current_image_number * uniforms.C + current_channel_number; + let scale = scale_input[scale_offset]; + output[global_idx] = fma(input[global_idx], ${pe(0)}, ${pe(1)}); + }`};e.compute({name:"InstanceNormalizationNHWC",shaderCache:{hint:`${p}`,inputDependencies:C},getRunData:()=>({outputs:[{dims:i,dataType:t[0].dataType}],dispatchGroup:{x:Math.ceil(h/64)},programUniforms:k}),getShaderSource:Z},{inputs:[t[0],V]})},xd=(e,t)=>{t.format==="NHWC"?vd(e,e.inputs,t):_a(e,e.inputs,t)}}),Pd,Ed,ga,hp=g(()=>{zt(),Ot(),Yt(),Pd=e=>{if(!e||e.length<2)throw new Error("layerNorm requires at least 2 inputs.")},Ed=(e,t,s)=>{let n=t.simplified,i=e[0].dims,a=e[1],o=!n&&e[2],u=i,p=ze.normalizeAxis(t.axis,i.length),h=ze.sizeToDimension(i,p),k=ze.sizeFromDimension(i,p),C=ze.size(a.dims),d=o?ze.size(o.dims):0;if(C!==k||o&&d!==k)throw new Error(`Size of X.shape()[axis:] == ${k}. + Size of scale and bias (if provided) must match this. + Got scale size of ${C} and bias size of ${d}`);let z=[];for(let Me=0;Me1,Y=s>2,he=Me=>{let Oe=fs(e[0].dataType),De=[qe("x",e[0].dataType,e[0].dims,B),qe("scale",a.dataType,a.dims,B)];o&&De.push(qe("bias",o.dataType,o.dims,B)),De.push(It("output",e[0].dataType,u,B)),ee&&De.push(It("mean_data_output",1,z)),Y&&De.push(It("inv_std_output",1,z));let Ye=[{name:"norm_count",type:"u32"},{name:"norm_size",type:"f32"},{name:"norm_size_vectorized",type:"u32"},{name:"epsilon",type:"f32"}];return` + ${Me.registerUniforms(Ye).declareVariables(...De)} + ${Me.mainStart()} + ${Me.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.norm_count")} + let offset = global_idx * uniforms.norm_size_vectorized; + var mean_vector = ${Ls("f32",B)}; + var mean_square_vector = ${Ls("f32",B)}; + + for (var h: u32 = 0u; h < uniforms.norm_size_vectorized; h++) { + let value = ${$s(Oe,B,"x[h + offset]")}; + mean_vector += value; + mean_square_vector += value * value; + } + let mean = ${Gs("mean_vector",B)} / uniforms.norm_size; + let inv_std_dev = inverseSqrt(${Gs("mean_square_vector",B)} / uniforms.norm_size ${n?"":"- mean * mean"} + uniforms.epsilon); + + for (var j: u32 = 0; j < uniforms.norm_size_vectorized; j++) { + let f32input = ${$s(Oe,B,"x[j + offset]")}; + let f32scale = ${$s(Oe,B,"scale[j]")}; + output[j + offset] = ${De[0].type.value}((f32input ${n?"":"- mean"}) * inv_std_dev * f32scale + ${o?`+ ${$s(Oe,B,"bias[j]")}`:""} + ); + } + + ${ee?"mean_data_output[global_idx] = mean":""}; + ${Y?"inv_std_output[global_idx] = inv_std_dev":""}; + }`},pe=[{dims:u,dataType:e[0].dataType}];return ee&&pe.push({dims:z,dataType:1}),Y&&pe.push({dims:z,dataType:1}),{name:"LayerNormalization",shaderCache:{hint:`${B};${s};${n}`,inputDependencies:V},getRunData:()=>({outputs:pe,dispatchGroup:{x:Math.ceil(h/64)},programUniforms:Z}),getShaderSource:he}},ga=(e,t)=>{Pd(e.inputs),e.compute(Ed(e.inputs,t,e.outputCount))}}),Cd,kd,mp=g(()=>{Ot(),fi(),Uo(),Cd=e=>{if(!e||e.length!==2)throw new Error("MatMul requires 2 inputs.");if(e[0].dims[e[0].dims.length-1]!==e[1].dims[e[1].dims.length-2])throw new Error("shared dimension does not match.")},kd=e=>{Cd(e.inputs);let t=Ws.calcShape(e.inputs[0].dims,e.inputs[1].dims,!0);if(!t)throw new Error("Can't use matmul on the given tensors");let s=t[t.length-1],n=e.inputs[0].dims[e.inputs[0].dims.length-1];if(s<8&&n<8)e.compute(Bo(e.inputs,{activation:""},t));else{let i=t[t.length-2],a=ze.size(e.inputs[0].dims.slice(0,-2)),o=ze.size(e.inputs[1].dims.slice(0,-2));if(a!==1&&i===1&&o===1){let u=e.inputs[0].reshape([1,a,n]),p=e.inputs[1].reshape([1,n,s]),h=[1,a,s],k=[u,p];e.compute(gi(k,{activation:""},t,h),{inputs:k})}else e.compute(gi(e.inputs,{activation:""},t))}}}),Sd,$d,Ad,Id,Od,Fd=g(()=>{zt(),Ot(),rs(),Yt(),Sd=(e,t)=>{if(e.length<3||e.length>4)throw new Error("MatMulNBits requires 3 or 4 inputs");let s=e[0],n=s.dims.length;if(s.dims[n-1]!==t.k)throw new Error("The last dim of input shape does not match the k value");let i=Math.floor((t.k+t.blockSize-1)/t.blockSize),a=t.blockSize/8*t.bits,o=e[1];if(!ze.areEqual(o.dims,[t.n,i,a]))throw new Error("The second inputs must be 3D tensor with shape N X nBlocksPerCol X blobSize");let u=e[2].dims;if(ze.size(u)!==t.n*i)throw new Error("scales input size error.");if(e.length===4){let p=e[3].dims,h=t.bits>4?t.n*i:t.n*Math.floor((i+1)/2);if(ze.size(p)!==h)throw new Error("zeroPoints input size error.")}},$d=(e,t)=>{let s=e[0].dims,n=s.length,i=s[n-2],a=t.k,o=t.n,u=s.slice(0,n-2),p=ze.size(u),h=e[1].dims[2]/4,k=e[0].dataType,C=qt(t.k),d=qt(h),z=qt(o),B=u.concat([i,o]),V=i>1&&o/z%2===0?2:1,Z=ze.size(B)/z/V,ee=64,Y=[],he=[p,i,a/C],pe=ze.convertShape(e[1].dims).slice();pe.splice(-1,1,h/d),Y.push(...yt(he)),Y.push(...yt(pe)),Y.push(...yt(e[2].dims)),e.length===4&&Y.push(...yt(ze.convertShape(e[3].dims)));let Me=[p,i,o/z];Y.push(...yt(Me));let Oe=De=>{let Ye=he.length,at=qe("a",e[0].dataType,Ye,C),Pt=qe("b",12,pe.length,d),Xt=qe("scales",e[2].dataType,e[2].dims.length),Zt=[at,Pt,Xt],bt=e.length===4?qe("zero_points",12,e[3].dims.length):void 0;bt&&Zt.push(bt);let ss=Me.length,St=It("output",e[0].dataType,ss,z),Ft=fs(e[0].dataType),bs=(()=>{switch(C){case 1:return`array<${Ft}, 8>`;case 2:return`mat4x2<${Ft}>`;case 4:return`mat2x4<${Ft}>`;default:throw new Error(`${C}-component is not supported.`)}})(),Ht=()=>{let ot=` + // reuse a data + var input_offset = ${at.indicesToOffset(`${at.type.indices}(batch, row, word_offset)`)}; + var a_data: ${bs}; + for (var j: u32 = 0; j < ${8/C}; j++) { + a_data[j] = ${at.getByOffset("input_offset")}; + input_offset++; + } + `;for(let Et=0;Et> 4) & b_mask); + b_quantized_values = ${bs}(${Array.from({length:4},(ps,Ns)=>`${Ft}(b_value_lower[${Ns}]), ${Ft}(b_value_upper[${Ns}])`).join(", ")}); + b_dequantized_values = ${C===1?`${bs}(${Array.from({length:8},(ps,Ns)=>`(b_quantized_values[${Ns}] - ${bt?`zero_point${Et}`:"zero_point"}) * scale${Et}`).join(", ")});`:`(b_quantized_values - ${bs}(${Array(8).fill(`${bt?`zero_point${Et}`:"zero_point"}`).join(",")})) * scale${Et};`}; + workgroup_shared[local_id.x * ${V} + ${Math.floor(Et/z)}]${z>1?`[${Et%z}]`:""} += ${Array.from({length:8/C},(ps,Ns)=>`${C===1?`a_data[${Ns}] * b_dequantized_values[${Ns}]`:`dot(a_data[${Ns}], b_dequantized_values[${Ns}])`}`).join(" + ")}; + `;return ot},Rt=()=>{let ot=` + var col_index = col * ${z}; + ${bt?` + let zero_point_bytes_per_col = (nBlocksPerCol + 1) / 2; + var zero_point_byte_count: u32; + var zero_point_word_index: u32; + var zero_point_byte_offset: u32; + let zero_point_nibble_offset: u32 = block & 0x1u; + var zero_point_bits_offset: u32; + var zero_point_word: u32;`:` + // The default zero point is 8 for unsigned 4-bit quantization. + let zero_point = ${Ft}(8);`} + `;for(let Et=0;Et> 0x1u); + zero_point_word_index = zero_point_byte_count >> 0x2u; + zero_point_byte_offset = zero_point_byte_count & 0x3u; + zero_point_bits_offset = (zero_point_byte_offset << 3) + (zero_point_nibble_offset << 2); + zero_point_word = ${bt.getByOffset("zero_point_word_index")} >> zero_point_bits_offset; + let zero_point${Et} = ${Ft}((zero_point_word) & 0xFu);`:""} + col_index += 1;`;return ot},_s=()=>{let ot=`col_index = col * ${z};`;for(let Et=0;Et; + var b_value_upper: vec4; + var b_quantized_values: ${bs}; + var b_dequantized_values: ${bs};`,ot};return` + var workgroup_shared: array<${St.type.value}, ${V*ee}>; + ${De.declareVariables(...Zt,St)} + ${De.mainStart([ee,1,1])} + let output_indices = ${St.offsetToIndices(`(global_idx / ${ee}) * ${V}`)}; + let col = output_indices[2]; + let row = output_indices[1]; + let batch = output_indices[0]; + let nBlocksPerCol = uniforms.b_shape[1]; + + for (var block = local_id.x; block < nBlocksPerCol; block += ${ee}) { + //process one block + var word_offset: u32 = block * ${t.blockSize/C}; + ${Rt()} + for (var word: u32 = 0; word < ${h}; word += ${d}) { + ${_s()} + for (var i: u32 = 0; i < ${d}; i++) { + ${Ht()} + word_offset += ${8/C}; + } + } + } + workgroupBarrier(); + + if (local_id.x < ${V}) { + var output_value: ${St.type.value} = ${St.type.value}(0); + var workgroup_shared_offset: u32 = local_id.x; + for (var b: u32 = 0u; b < ${ee}u; b++) { + output_value += workgroup_shared[workgroup_shared_offset]; + workgroup_shared_offset += ${V}; + } + ${St.setByIndices(`${St.type.indices}(batch, row, col + local_id.x)`,"output_value")}; + } + }`};return{name:"MatMulNBits",shaderCache:{hint:`${t.blockSize};${t.bits};${C};${d};${z};${V};${ee}`,inputDependencies:Array(e.length).fill("rank")},getRunData:()=>({outputs:[{dims:B,dataType:k}],dispatchGroup:{x:Z},programUniforms:Y}),getShaderSource:Oe}},Ad=(e,t)=>{let s=e[0].dims,n=s.length,i=s[n-2],a=t.k,o=t.n,u=s.slice(0,n-2),p=ze.size(u),h=e[1].dims[2]/4,k=e[0].dataType,C=qt(t.k),d=qt(h),z=u.concat([i,o]),B=128,V=o%8===0?8:o%4===0?4:1,Z=B/V,ee=Z*d*8,Y=ee/C,he=ee/t.blockSize,pe=ze.size(z)/V,Me=[],Oe=[p,i,a/C],De=ze.convertShape(e[1].dims).slice();De.splice(-1,1,h/d),Me.push(...yt(Oe)),Me.push(...yt(De)),Me.push(...yt(e[2].dims)),e.length===4&&Me.push(...yt(ze.convertShape(e[3].dims)));let Ye=[p,i,o];Me.push(...yt(Ye));let at=Pt=>{let Xt=Oe.length,Zt=qe("a",e[0].dataType,Xt,C),bt=qe("b",12,De.length,d),ss=qe("scales",e[2].dataType,e[2].dims.length),St=[Zt,bt,ss],Ft=e.length===4?qe("zero_points",12,e[3].dims.length):void 0;Ft&&St.push(Ft);let bs=Ye.length,Ht=It("output",e[0].dataType,bs),Rt=fs(e[0].dataType),_s=()=>{switch(C){case 1:return` + let a_data0 = vec4<${Rt}>(sub_a[word_offset], sub_a[word_offset + 1], sub_a[word_offset + 2], sub_a[word_offset + 3]); + let a_data1 = vec4<${Rt}>(sub_a[word_offset + 4], sub_a[word_offset + 5], sub_a[word_offset + 6], sub_a[word_offset + 7]);`;case 2:return` + let a_data0 = vec4<${Rt}>(sub_a[word_offset], sub_a[word_offset + 1]); + let a_data1 = vec4<${Rt}>(sub_a[word_offset + 2], sub_a[word_offset + 3]);`;case 4:return` + let a_data0 = sub_a[word_offset]; + let a_data1 = sub_a[word_offset + 1];`;default:throw new Error(`${C}-component is not supported.`)}};return` + var sub_a: array<${Zt.type.value}, ${Y}>; + var inter_results: array, ${V}>; + ${Pt.declareVariables(...St,Ht)} + ${Pt.mainStart([Z,V,1])} + let output_indices = ${Ht.offsetToIndices(`workgroup_index * ${V}`)}; + let col = output_indices[2]; + let row = output_indices[1]; + let batch = output_indices[0]; + let n_blocks_per_col = uniforms.b_shape[1]; + let num_tiles = (n_blocks_per_col - 1) / ${he} + 1; + + // Loop over shared dimension. + for (var tile: u32 = 0; tile < num_tiles; tile += 1) { + let a_col_start = tile * ${Y}; + // load one tile A data into shared memory. + for (var a_offset = local_idx; a_offset < ${Y}; a_offset += ${B}) + { + let a_col = a_col_start + a_offset; + if (a_col < uniforms.a_shape[2]) + { + sub_a[a_offset] = ${Zt.getByIndices(`${Zt.type.indices}(batch, row, a_col)`)}; + } else { + sub_a[a_offset] = ${Zt.type.value}(0); + } + } + workgroupBarrier(); + + // each thread process one block + let b_row = col + local_id.y; + let block = tile * ${he} + local_id.x; + ${Ft?` + let zero_point_bytes_per_col = (n_blocks_per_col + 1) / 2; + let zero_point_byte_count = b_row * zero_point_bytes_per_col + (block >> 0x1u); + let zero_point_word_index = zero_point_byte_count >> 0x2u; + let zero_point_byte_offset = zero_point_byte_count & 0x3u; + let zero_point_nibble_offset: u32 = block & 0x1u; + let zero_point_bits_offset = (zero_point_byte_offset << 3) + (zero_point_nibble_offset << 2); + let zero_point_word = ${Ft.getByOffset("zero_point_word_index")} >> zero_point_bits_offset; + let zero_point = ${Rt}((zero_point_word) & 0xFu);`:` + // The default zero point is 8 for unsigned 4-bit quantization. + let zero_point = ${Rt}(8);`} + let scale = ${ss.getByOffset("b_row * n_blocks_per_col + block")}; + let b_data = ${bt.getByIndices(`${bt.type.indices}(b_row, block, 0)`)}; + var word_offset = local_id.x * ${t.blockSize/C}; + for (var i: u32 = 0; i < ${d}; i++) { + ${_s()} + let b_value = ${d===1?"b_data":"b_data[i]"}; + let b_value_lower = unpack4xU8(b_value & 0x0F0F0F0Fu); + let b_value_upper = unpack4xU8((b_value >> 4) & 0x0F0F0F0Fu); + let b_quantized_values = mat2x4<${Rt}>(${Array.from({length:4},(ot,Et)=>`${Rt}(b_value_lower[${Et}]), ${Rt}(b_value_upper[${Et}])`).join(", ")}); + let b_dequantized_values = (b_quantized_values - mat2x4<${Rt}>(${Array(8).fill("zero_point").join(",")})) * scale; + inter_results[local_id.y][local_id.x] += ${Array.from({length:2},(ot,Et)=>`${`dot(a_data${Et}, b_dequantized_values[${Et}])`}`).join(" + ")}; + word_offset += ${8/C}; + } + workgroupBarrier(); + } + + if (local_idx < ${V}) { + var output_value: ${Ht.type.value} = ${Ht.type.value}(0); + for (var b = 0u; b < ${Z}; b++) { + output_value += inter_results[local_idx][b]; + } + if (col + local_idx < uniforms.output_shape[2]) + { + ${Ht.setByIndices(`${Ht.type.indices}(batch, row, col + local_idx)`,"output_value")} + } + } + }`};return{name:"BlockwiseMatMulNBits32",shaderCache:{hint:`${t.blockSize};${C};${d};${Z};${V}`,inputDependencies:Array(e.length).fill("rank")},getRunData:()=>({outputs:[{dims:z,dataType:k}],dispatchGroup:{x:pe},programUniforms:Me}),getShaderSource:at}},Id=(e,t)=>{Sd(e.inputs,t),t.blockSize===32&&e.adapterInfo.isVendor("intel")&&e.adapterInfo.isArchitecture("gen-12lp")?e.compute(Ad(e.inputs,t)):e.compute($d(e.inputs,t))},Od=e=>Bt(e)}),Dd,Ld,wa,zd,Bd,ws,fp,_p,gp,Rd=g(()=>{zt(),Ot(),Yt(),Dd=e=>{if(!e||e.length<1)throw new Error("Too few inputs");if(e[0].dataType!==1&&e[0].dataType!==10)throw new Error("Input type must be float or float16.");if(e.length>=2){let t=e[0].dims.length*2===e[1].dims[0];if(e.length===4&&(t=e[3].dims[0]*2===e[1].dims[0]),!t)throw new Error("The pads should be a 1D tensor of shape [2 * input_rank] or [2 * num_axes].")}},Ld=(e,t,s)=>{let n="";for(let i=t-1;i>=0;--i)n+=` + k = i32(${e.indicesGet("indices",i)}) - ${$t("uniforms.pads",i,s)}; + if (k < 0) { + break; + } + if (k >= i32(${$t("uniforms.x_shape",i,t)})) { + break; + } + offset += k * i32(${$t("uniforms.x_strides",i,t)}); + `;return` + value = ${e.type.value}(uniforms.constant_value); + for (var i = 0; i < 1; i++) { + var offset = 0; + var k = 0; + ${n} + value = x[offset]; + } + `},wa=(e,t,s)=>{let n="";for(let i=t-1;i>=0;--i)n+=` + k = i32(${e.indicesGet("indices",i)}) - ${$t("uniforms.pads",i,s)}; + if (k < 0) { + k = -k; + } + { + let _2n_1 = 2 * (i32(${$t("uniforms.x_shape",i,t)}) - 1); + k = k % _2n_1; + if(k >= i32(${$t("uniforms.x_shape",i,t)})) { + k = _2n_1 - k; + } + } + offset += k * i32(${$t("uniforms.x_strides",i,t)}); + `;return` + var offset = 0; + var k = 0; + ${n} + value = x[offset]; + `},zd=(e,t,s)=>{let n="";for(let i=t-1;i>=0;--i)n+=` + k = i32(${e.indicesGet("indices",i)}) - ${$t("uniforms.pads",i,s)}; + if (k < 0) { + k = 0; + } + if (k >= i32(${$t("uniforms.x_shape",i,t)})) { + k = i32(${$t("uniforms.x_shape",i,t)}) - 1; + } + offset += k * i32(${$t("uniforms.x_strides",i,t)}); + `;return` + var offset = 0; + var k = 0; + ${n} + value = x[offset]; + `},Bd=(e,t,s)=>{let n="";for(let i=t-1;i>=0;--i)n+=` + k = i32(${e.indicesGet("indices",i)}) - ${$t("uniforms.pads",i,s)}; + if (k < 0) { + k += i32(${$t("uniforms.x_shape",i,t)}]); + } + if (k >= i32(${$t("uniforms.x_shape",i,t)})) { + k -= i32(${$t("uniforms.x_shape",i,t)}); + } + offset += k * i32(${$t("uniforms.x_strides",i,t)}); + `;return` + var offset = 0; + var k = 0; + ${n} + value = x[offset]; + `},ws=(e,t,s)=>{switch(s.mode){case 0:return Ld(e,t,s.pads.length);case 1:return wa(e,t,s.pads.length);case 2:return zd(e,t,s.pads.length);case 3:return Bd(e,t,s.pads.length);default:throw new Error("Invalid mode")}},fp=(e,t)=>{let s=ze.padShape(e[0].dims.slice(),t.pads),n=e[0].dims,i=ze.size(s),a=[{type:12,data:i},{type:6,data:t.pads}],o=e.length>=3&&e[2].data;t.mode===0&&a.push({type:o?e[2].dataType:1,data:t.value}),a.push(...yt(e[0].dims,s));let u=["rank"],p=h=>{let k=It("output",e[0].dataType,s.length),C=qe("x",e[0].dataType,n.length),d=C.type.value,z=ws(k,n.length,t),B=[{name:"output_size",type:"u32"},{name:"pads",type:"i32",length:t.pads.length}];return t.mode===0&&B.push({name:"constant_value",type:o?d:"f32"}),` + ${h.registerUniforms(B).declareVariables(C,k)} + ${h.mainStart()} + ${h.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + + let indices = ${k.offsetToIndices("global_idx")}; + + var value = ${d}(0); + ${z} + output[global_idx] = value; + }`};return{name:"Pad",shaderCache:{hint:`${t.mode}${o}`,inputDependencies:u},getRunData:()=>({outputs:[{dims:s,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(ze.size(s)/64)},programUniforms:a}),getShaderSource:p}},_p=(e,t)=>{if(e.length>1){let s=e[1].getBigInt64Array(),n=e.length>=3&&e[2].data?e[2].dataType===10?e[2].getUint16Array()[0]:e[2].getFloat32Array()[0]:0,i=e[0].dims.length,a=new Int32Array(2*i).fill(0);if(e.length>=4){let u=e[3].getBigInt64Array();for(let p=0;pa[Number(p)]=Number(u));let o=[];return a.forEach(u=>o.push(u)),{mode:t.mode,value:n,pads:o}}else return t},gp=(e,t)=>{Dd(e.inputs);let s=_p(e.inputs,t);e.compute(fp(e.inputs,s),{inputs:[0]})}}),Kn,ya,Ma,ba,$i,va,wp,xa,Ta,Pa,yp,Nd,jd,Ud,Ea,Vd,Wd,Gd,Kd,Mp=g(()=>{We(),zt(),Ot(),Yt(),Kn=e=>{if(O.webgpu.validateInputContent&&(!e||e.length!==1))throw new Error("Pool ops requires 1 input.")},ya=(e,t,s)=>{let n=t.format==="NHWC",i=e.dims.slice();n&&i.splice(1,0,i.pop());let a=Object.hasOwnProperty.call(t,"dilations"),o=t.kernelShape.slice(),u=t.strides.slice(),p=a?t.dilations.slice():[],h=t.pads.slice();Js.adjustPoolAttributes(s,i,o,u,p,h);let k=Js.computePoolOutputShape(s,i,u,p,o,h,t.autoPad),C=Object.assign({},t);a?Object.assign(C,{kernelShape:o,strides:u,pads:h,dilations:p,cacheKey:t.cacheKey}):Object.assign(C,{kernelShape:o,strides:u,pads:h,cacheKey:t.cacheKey});let d=k.slice();return d.push(d.splice(1,1)[0]),[C,n?d:k]},Ma=(e,t)=>{let s=t.format==="NHWC",n=ze.size(e),i=ze.size(t.kernelShape),a=[{type:12,data:n},{type:12,data:i}],o=[{name:"outputSize",type:"u32"},{name:"kernelSize",type:"u32"}];if(t.kernelShape.length<=2){let u=t.kernelShape[t.kernelShape.length-1],p=t.strides[t.strides.length-1],h=t.pads[t.pads.length/2-1],k=t.pads[t.pads.length-1],C=!!(h+k);a.push({type:12,data:u},{type:12,data:p},{type:12,data:h},{type:12,data:k}),o.push({name:"kw",type:"u32"},{name:"sw",type:"u32"},{name:"pwStart",type:"u32"},{name:"pwEnd",type:"u32"});let d=!1;if(t.kernelShape.length===2){let z=t.kernelShape[t.kernelShape.length-2],B=t.strides[t.strides.length-2],V=t.pads[t.pads.length/2-2],Z=t.pads[t.pads.length-2];d=!!(V+Z),a.push({type:12,data:z},{type:12,data:B},{type:12,data:V},{type:12,data:Z}),o.push({name:"kh",type:"u32"},{name:"sh",type:"u32"},{name:"phStart",type:"u32"},{name:"phEnd",type:"u32"})}return[a,o,!0,C,d]}else{if(s)throw new Error("Pooling with kernelShape.length > 2 is not supported for NHWC format.");let u=ze.computeStrides(t.kernelShape);a.push({type:12,data:u},{type:12,data:t.pads},{type:12,data:t.strides}),o.push({name:"kernelStrides",type:"u32",length:u.length},{name:"pads",type:"u32",length:t.pads.length},{name:"strides",type:"u32",length:t.strides.length});let p=t.pads.reduce((h,k)=>h+k);return[a,o,!!p,!1,!1]}},ba=(e,t,s,n,i,a,o,u,p,h,k,C)=>{let d=i.format==="NHWC",z=t.type.value,B=It("output",t.type.tensor,n);if(i.kernelShape.length<=2){let V="",Z="",ee="",Y=s-(d?2:1);if(k?V=` + for (var i: u32 = 0u; i < uniforms.kw; i++) { + xIndices[${Y}] = indices[${Y}] * uniforms.sw - uniforms.pwStart + i; + if (xIndices[${Y}] < 0 || xIndices[${Y}] + >= uniforms.x_shape[${Y}]) { + pad++; + continue; + } + let x_val = x[${t.indicesToOffset("xIndices")}]; + ${a} + }`:V=` + for (var i: u32 = 0u; i < uniforms.kw; i++) { + xIndices[${Y}] = indices[${Y}] * uniforms.sw - uniforms.pwStart + i; + let x_val = x[${t.indicesToOffset("xIndices")}]; + ${a} + }`,i.kernelShape.length===2){let he=s-(d?3:2);C?Z=` + for (var j: u32 = 0u; j < uniforms.kh; j++) { + xIndices[${he}] = indices[${he}] * uniforms.sh - uniforms.phStart + j; + if (xIndices[${he}] < 0 || xIndices[${he}] >= uniforms.x_shape[${he}]) { + pad += i32(uniforms.kw); + continue; + } + `:Z=` + for (var j: u32 = 0u; j < uniforms.kh; j++) { + xIndices[${he}] = indices[${he}] * uniforms.sh - uniforms.phStart + j; + `,ee=` + } + `}return` + ${e.registerUniforms(p).declareVariables(t,B)} + + ${e.mainStart()} + ${e.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")} + + let indices = ${B.offsetToIndices("global_idx")}; + var xIndices = ${B.offsetToIndices("global_idx")}; + + var value = ${z}(${u}); + var pad = 0; + ${Z} + ${V} + ${ee} + ${o} + + output[global_idx] = value; + }`}else{if(d)throw new Error("Pooling with kernelShape.length > 2 is not supported for NHWC format.");let V=i.kernelShape.length,Z=i.pads.length,ee="";return h?ee=` + if (xIndices[j] >= uniforms.x_shape[j]) { + pad++; + isPad = true; + break; + } + } + if (!isPad) { + let x_val = x[${t.indicesToOffset("xIndices")}]; + ${a} + }`:ee=` + } + let x_val = x[${t.indicesToOffset("xIndices")}]; + ${a} + `,` + ${e.registerUniforms(p).declareVariables(t,B)} + + ${e.mainStart()} + ${e.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")} + let indices = ${B.offsetToIndices("global_idx")}; + var xIndices = ${B.offsetToIndices("global_idx")}; + + var offsets: array; + + var value = ${z}(${u}); + var pad = 0; + var isPad = false; + + for (var i: u32 = 0u; i < uniforms.kernelSize; i++) { + var offset = i; + for (var j = 0u; j < ${V-1}u; j++) { + offsets[j] = offset / ${$t("uniforms.kernelStrides","j",V)}; + offset -= offsets[j] * ${$t("uniforms.kernelStrides","j",V)}; + } + offsets[${V-1}] = offset; + + isPad = false; + for (var j = ${s-V}u; j < ${s}u; j++) { + xIndices[j] = indices[j] * ${$t("uniforms.strides",`j - ${s-V}u`,V)} + + offsets[j - ${s-V}u] - ${$t("uniforms.pads","j - 2u",Z)}; + ${ee} + } + ${o} + + output[global_idx] = value; + }`}},$i=e=>`${e.format};${e.ceilMode};${e.autoPad};${e.kernelShape.length}`,va=e=>`${$i(e)};${e.countIncludePad}`,wp=e=>`${$i(e)};${e.storageOrder};${e.dilations}`,xa=e=>({format:e.format,autoPad:["NOTSET","VALID","SAME_UPPER","SAME_LOWER"][e.auto_pad],ceilMode:e.ceil_mode,kernelShape:e.kernel_shape,strides:e.strides,pads:e.pads}),Ta=(e,t,s,n)=>{let[i,a]=ya(t,n,s),o=qe("x",t.dataType,t.dims.length),u=o.type.value,p="value += x_val;",h="";i.countIncludePad?h+=`value /= ${u}(uniforms.kernelSize);`:h+=`value /= ${u}(i32(uniforms.kernelSize) - pad);`;let[k,C,d,z,B]=Ma(a,i);k.push(...yt(t.dims,a));let V=["rank"];return{name:e,shaderCache:{hint:`${n.cacheKey};${d};${z};${B}`,inputDependencies:V},getRunData:()=>({outputs:[{dims:a,dataType:t.dataType}],dispatchGroup:{x:Math.ceil(ze.size(a)/64)},programUniforms:k}),getShaderSource:Z=>ba(Z,o,t.dims.length,a.length,i,p,h,0,C,d,z,B)}},Pa=e=>{let t=e.count_include_pad!==0,s=xa(e);if(s.ceilMode!==0)throw new Error("using ceil() in shape computation is not yet supported for AveragePool");let n={countIncludePad:t,...s,cacheKey:""};return{...n,cacheKey:va(n)}},yp=(e,t)=>{Kn(e.inputs),e.compute(Ta("AveragePool",e.inputs[0],!1,t))},Nd={autoPad:"",ceilMode:0,countIncludePad:!1,kernelShape:[],strides:[],pads:[],storageOrder:0,dilations:[]},jd=e=>{let t=e.format;return{format:t,...Nd,cacheKey:t}},Ud=(e,t)=>{Kn(e.inputs),e.compute(Ta("GlobalAveragePool",e.inputs[0],!0,t))},Ea=(e,t,s,n)=>{let[i,a]=ya(t,n,s),o=` + value = max(x_val, value); + `,u="",p=qe("x",t.dataType,t.dims.length),h=["rank"],[k,C,d,z,B]=Ma(a,i);return k.push(...yt(t.dims,a)),{name:e,shaderCache:{hint:`${n.cacheKey};${d};${z};${B}`,inputDependencies:h},getRunData:()=>({outputs:[{dims:a,dataType:t.dataType}],dispatchGroup:{x:Math.ceil(ze.size(a)/64)},programUniforms:k}),getShaderSource:V=>ba(V,p,t.dims.length,a.length,i,o,u,t.dataType===10?-65504:-1e5,C,d,z,B)}},Vd=(e,t)=>{Kn(e.inputs),e.compute(Ea("MaxPool",e.inputs[0],!1,t))},Wd=e=>{let t=e.storage_order,s=e.dilations,n=xa(e);if(t!==0)throw new Error("column major storage order is not yet supported for MaxPool");if(n.ceilMode!==0)throw new Error("using ceil() in shape computation is not yet supported for MaxPool");let i={storageOrder:t,dilations:s,...n,cacheKey:""};return{...i,cacheKey:wp(i)}},Gd=e=>{let t=e.format;return{format:t,...Nd,cacheKey:t}},Kd=(e,t)=>{Kn(e.inputs),e.compute(Ea("GlobalMaxPool",e.inputs[0],!0,t))}}),Hd,qd,Qd,Xd,Hp=g(()=>{zt(),Ot(),rs(),Yt(),Hd=(e,t)=>{if(e.length<2||e.length>3)throw new Error("DequantizeLinear requires 2 or 3 inputs.");if(e.length===3&&e[1].dims===e[2].dims)throw new Error("x-scale and x-zero-point must have the same shape.");if(e.length===3&&e[0].dataType!==e[2].dataType)throw new Error("x and x-zero-point must have the same data type.");if(e[0].dataType===6&&e.length>2)throw new Error("In the case of dequantizing int32 there is no zero point.");if(e[1].dims.length!==0&&e[1].dims.length!==1&&e[1].dims.length!==e[0].dims.length)throw new Error("scale input must be a scalar, a 1D tensor, or have the same rank as the input tensor.");if(e.length>2){if(e[0].dataType!==e[2].dataType)throw new Error("x and x-zero-point must have the same data type.");if(e[1].dims.length!==e[2].dims.length)throw new Error("scale and zero-point inputs must have the same rank.");if(!e[1].dims.map((s,n)=>s===e[2].dims[n]).reduce((s,n)=>s&&n,!0))throw new Error("scale and zero-point inputs must have the same shape.")}if(t.blockSize>0){if(e[1].dims.length===0||e[1].dims.length===1&&e[1].dims[0]===1)throw new Error("blockSize must be set only for block quantization.");if(!e[1].dims.map((i,a)=>a===t.axis||i===e[0].dims[a]).reduce((i,a)=>i&&a,!0))throw new Error("For block qunatization, scale input shape to match the input shape except for the axis");if(e[1].dims.length!==e[0].dims.length)throw new Error("For block qunatization the scale input rank must be the same as the x rank.");let s=e[0].dims[t.axis],n=e[1].dims[t.axis];if(t.blockSizeMath.ceil(s/(n-1)-1))throw new Error("blockSize must be with in the range [ceil(dI / Si), ceil(dI / (Si - 1) - 1)].")}},qd=(e,t)=>{let s=ze.normalizeAxis(t.axis,e[0].dims.length),n=e[0].dataType,i=n===3,a=e[0].dims,o=e[1].dataType,u=ze.size(a),p=n===3||n===2,h=p?[Math.ceil(ze.size(e[0].dims)/4)]:e[0].dims,k=e[1].dims,C=e.length>2?e[2]:void 0,d=C?p?[Math.ceil(ze.size(C.dims)/4)]:C.dims:void 0,z=k.length===0||k.length===1&&k[0]===1,B=z===!1&&k.length===1,V=qt(u),Z=z&&(!p||V===4),ee=Z?V:1,Y=Z&&!p?V:1,he=qe("input",p?12:n,h.length,Y),pe=qe("scale",o,k.length),Me=C?qe("zero_point",p?12:n,d.length):void 0,Oe=It("output",o,a.length,ee),De=[he,pe];Me&&De.push(Me);let Ye=[h,k];C&&Ye.push(d);let at=[{type:12,data:u/ee},{type:12,data:s},{type:12,data:t.blockSize},...yt(...Ye,a)],Pt=Xt=>{let Zt=[{name:"output_size",type:"u32"},{name:"axis",type:"u32"},{name:"block_size",type:"u32"}];return` + ${Xt.registerUniforms(Zt).declareVariables(...De,Oe)} + ${Xt.mainStart()} + ${Xt.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let output_indices = ${Oe.offsetToIndices("global_idx")}; + + // Set input x + ${p?` + let input = ${he.getByOffset("global_idx / 4")}; + let x_vec = ${i?"unpack4xI8(input)":"unpack4xU8(input)"}; + let x_value = ${ee===1?"x_vec[global_idx % 4]":"x_vec"};`:`let x_value = ${he.getByOffset("global_idx")};`}; + + // Set scale input + ${z?`let scale_value= ${pe.getByOffset("0")}`:B?` + let scale_index = ${Oe.indicesGet("output_indices","uniforms.axis")}; + let scale_value= ${pe.getByOffset("scale_index")};`:` + var scale_indices: ${pe.type.indices} = output_indices; + let index = ${pe.indicesGet("scale_indices","uniforms.axis")} / uniforms.block_size; + ${pe.indicesSet("scale_indices","uniforms.axis","index")}; + let scale_value= ${pe.getByIndices("scale_indices")};`}; + + // Set zero-point input + ${Me?z?p?` + let zero_point_input = ${Me.getByOffset("0")}; + let zero_point_vec = ${i?"unpack4xI8(zero_point_input)":"unpack4xU8(zero_point_input)"}; + let zero_point_value= zero_point_vec[0]`:`let zero_point_value = ${Me.getByOffset("0")}`:B?p?` + let zero_point_index = ${Oe.indicesGet("output_indices","uniforms.axis")}; + let zero_point_input = ${Me.getByOffset("zero_point_index / 4")}; + let zero_point_vec = ${i?"unpack4xI8(zero_point_input)":"unpack4xU8(zero_point_input)"}; + let zero_point_value = zero_point_vec[zero_point_index % 4]`:` + let zero_point_index = ${Oe.indicesGet("output_indices","uniforms.axis")}; + let zero_point_value = ${Me.getByOffset("zero_point_index")};`:p?` + let zero_point_offset = ${pe.indicesToOffset("scale_indices")}; + let zero_point_input = ${Me.getByOffset("zero_point_offset / 4")}; + let zero_point_vec = ${i?"unpack4xI8(zero_point_input)":"unpack4xU8(zero_point_input)"}; + let zero_point_value = zero_point_vec[zero_point_offset % 4];`:`let zero_point_value = ${Me.getByIndices("scale_indices")};`:`let zero_point_value = ${p?i?"i32":"u32":he.type.value}(0);`}; + // Compute and write output + ${Oe.setByOffset("global_idx",`${Oe.type.value}(x_value - zero_point_value) * scale_value`)}; + }`};return{name:"DequantizeLinear",shaderCache:{hint:t.cacheKey,inputDependencies:Me?["rank","rank","rank"]:["rank","rank"]},getShaderSource:Pt,getRunData:()=>({outputs:[{dims:a,dataType:o}],dispatchGroup:{x:Math.ceil(u/ee/64),y:1,z:1},programUniforms:at})}},Qd=(e,t)=>{Hd(e.inputs,t),e.compute(qd(e.inputs,t))},Xd=e=>Bt({axis:e.axis,blockSize:e.blockSize})}),Yd,Jd,Zd,bp=g(()=>{We(),zt(),Yt(),Yd=(e,t,s)=>{let n=e===t,i=et&&s>0;if(n||i||a)throw new Error("Range these inputs' contents are invalid.")},Jd=(e,t,s,n)=>{let i=Math.abs(Math.ceil((t-e)/s)),a=[i],o=i,u=[{type:12,data:o},{type:n,data:e},{type:n,data:s},...yt(a)],p=h=>{let k=It("output",n,a.length),C=k.type.value,d=[{name:"outputSize",type:"u32"},{name:"start",type:C},{name:"delta",type:C}];return` + ${h.registerUniforms(d).declareVariables(k)} + ${h.mainStart()} + ${h.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")} + output[global_idx] = uniforms.start + ${C}(global_idx) * uniforms.delta; + }`};return{name:"Range",shaderCache:{hint:`${n}`},getShaderSource:p,getRunData:()=>({outputs:[{dims:a,dataType:n}],dispatchGroup:{x:Math.ceil(o/64)},programUniforms:u})}},Zd=e=>{let t=0,s=0,n=0;e.inputs[0].dataType===6?(t=e.inputs[0].getInt32Array()[0],s=e.inputs[1].getInt32Array()[0],n=e.inputs[2].getInt32Array()[0]):e.inputs[0].dataType===1&&(t=e.inputs[0].getFloat32Array()[0],s=e.inputs[1].getFloat32Array()[0],n=e.inputs[2].getFloat32Array()[0]),O.webgpu.validateInputContent&&Yd(t,s,n),e.compute(Jd(t,s,n,e.inputs[0].dataType),{inputs:[]})}}),ec,tc,vp,Ca,xp=g(()=>{zt(),Ot(),rs(),Yt(),ec=(e,t,s,n)=>{if(e!=="none"&&n!=="i32"&&n!=="u32"&&n!=="f32")throw new Error(`Input ${n} is not supported with reduction ${e}.`);let i=`{ + var oldValue = 0; + loop { + let newValueF32 =`,a=`; + let newValue = bitcast(newValueF32); + let res = atomicCompareExchangeWeak(&${t}, oldValue, newValue); + if res.exchanged { + break; + } + oldValue = res.old_value; + } + }`;switch(e){case"none":return`${t}=${s};`;case"add":return n==="i32"||n==="u32"?`atomicAdd(&${t}, bitcast<${n}>(${s}));`:` + ${i}bitcast<${n}>(oldValue) + (${s})${a}`;case"max":return n==="i32"||n==="u32"?`atomicMax(&${t}, bitcast<${n}>(${s}));`:` + ${i}max(bitcast(oldValue), (${s}))${a}`;case"min":return n==="i32"||n==="u32"?`atomicMin(&${t}, bitcast<${n}>(${s}));`:`${i}min(bitcast<${n}>(oldValue), (${s}))${a}`;case"mul":return`${i}(bitcast<${n}>(oldValue) * (${s}))${a}`;default:throw new Error(`Reduction ${e} is not supported.`)}},tc=(e,t)=>{let s=e[0].dims,n=e[1].dims,i=s,a=1,o=Math.ceil(ze.size(n)/a),u=n[n.length-1],p=ze.sizeFromDimension(s,u),h=[{type:12,data:o},{type:12,data:u},{type:12,data:p},...yt(e[1].dims,e[2].dims,i)],k=C=>{let d=qe("indices",e[1].dataType,e[1].dims.length),z=qe("updates",e[2].dataType,e[2].dims.length,a),B=t.reduction!=="none"&&t.reduction!==""?za("output",e[0].dataType,i.length):It("output",e[0].dataType,i.length,a);return` + ${C.registerUniform("output_size","u32").registerUniform("last_index_dimension","u32").registerUniform("num_updates_elements","u32").declareVariables(d,z,B)} + ${C.mainStart()} + ${C.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + var data_offset = 0u; + let indices_start = uniforms.last_index_dimension * global_idx; + let indices_end = indices_start + uniforms.last_index_dimension; + for (var i = indices_start; i < indices_end; i++) { + var index = i32(indices[i].x); + ${e[0].dims.length===1?` + let element_count_dim = uniforms.output_strides; + let dim_value = uniforms.output_shape;`:` + let element_count_dim = uniforms.output_strides[i - indices_start]; + let dim_value = uniforms.output_shape[i - indices_start + uniforms.last_index_dimension];`} + if (index >= 0) { + if (index >= i32(dim_value)) { + index = i32(dim_value - 1); + } + } else { + if (index < -i32(dim_value)) { + index = 0; + } else { + index += i32(dim_value); + } + } + data_offset += u32((u32(index) * element_count_dim)); + } + + for (var i = 0u; i < uniforms.num_updates_elements; i++) { + let value = updates[uniforms.num_updates_elements * global_idx + i]; + ${ec(t.reduction,"output[data_offset + i]","value",B.type.value)} + } + + }`};return{name:"ScatterND",shaderCache:{hint:`${t.cacheKey}_${t.reduction}`,inputDependencies:["rank","rank"]},getRunData:()=>({outputs:[{dims:i,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(o/64)},programUniforms:h}),getShaderSource:k}},vp=e=>Bt({reduction:e.reduction}),Ca=(e,t)=>{e.compute(tc(e.inputs,t),{inputs:[e.inputs[1],e.inputs[2]],outputs:[]})}}),sc,rc,nc,ic,oc,ac,lc,uc,dc,cc,pc,ka,hc,mc,fc,_c,gc,wc,yc,Tp=g(()=>{zt(),Ot(),rs(),Yt(),sc=(e,t)=>{if(e.every(s=>s>0||(()=>{throw new Error("Resize requires scales input values to be positive")})),e.length>0){if(t.mode==="linear"){if(!(e.length===2||e.length===3||e.length===4&&e[0]===1&&e[1]===1||e.length===4&&e[0]===1&&e[3]===1||e.length===5&&e[0]===1&&e[1]===1))throw new Error(`For linear mode, Resize requires scales to be 2D, 3D, 4D with either two outermost or one innermost and + one outermost scale values equal to 1, or 5D with two outermost scale values equal to 1`)}else if(t.mode==="cubic"&&!(e.length===2||e.length===4&&e[0]===1&&e[1]===1||e.length===4&&e[0]===1&&e[3]===1))throw new Error("Resize requires scales input size to be 2 or 4 for cubic mode")}},rc=(e,t,s)=>{t.every(i=>i>=0&&i{throw new Error("Resize requires axes input values to be positive and less than rank")}));let n=new Array(s).fill(1);return t.forEach((i,a)=>n[i]=e[a]),n},nc=(e,t,s,n,i,a)=>{let[o,u,p]=s>10?[1,2,3]:[-1,e.length>1?1:-1,-1],h=e[0].dims.length;if(o>0&&e.length>o&&e[o].dims.length>0)e[o].getFloat32Array().forEach(k=>a.push(k));else if(t.coordinateTransformMode==="tf_crop_and_resize")throw new Error("Resize requires RoI input to be specified when coordinateTransformMode is tfCropAndResize");if(u>0&&e.length>u&&e[u].dims.length===1&&e[u].dims[0]>0){if(e[u].getFloat32Array().forEach(k=>n.push(k)),n.length!==0&&n.length!==h&&s>=18&&n.length!==t.axes.length)throw new Error("Resize requires scales input size to be same as input rank or axes size for opset 18 and up");sc(n,t),t.axes.length>0&&rc(n,t.axes,h).forEach((k,C)=>n[C]=k)}if(p>0&&e.length>p&&e[p].dims.length===1&&e[p].dims[0]>0&&(e[p].getBigInt64Array().forEach(k=>i.push(Number(k))),i.length!==0&&i.length!==h&&s>=18&&i.length!==t.axes.length))throw new Error("Resize requires sizes input size to be same as input rank or axes size for opset 18 and up");if(t.axes.length>0){if(n.length!==0&&n.length!==t.axes.length)throw new Error('Resize requires "scales" input size to be of axes rank when axes attributes is specified');if(i.length!==0&&i.length!==t.axes.length)throw new Error('Resize requires "sizes" input size to be of rank axes rank when axes attributes is specified')}if(typeof n<"u"&&typeof i<"u"&&n.length>0&&i.length>h)throw new Error("Resize requires only of scales or sizes to be specified")},ic=(e,t)=>`fn getOriginalCoordinateFromResizedCoordinate(xResized: u32, xScale: f32, lengthResized: u32, + lengthOriginal: u32, roiStart: f32, roiEnd: f32) -> ${t} { `+(()=>{switch(e){case"asymmetric":return`return ${t}(xResized) / ${t}(xScale);`;case"pytorch_half_pixel":return`if (lengthResized > 1) { + return (${t}(xResized) + 0.5) / ${t}(xScale) - 0.5; + } else { + return 0.0; + }`;case"tf_half_pixel_for_nn":return`return (${t}(xResized) + 0.5) / ${t}(xScale);`;case"align_corners":return`if (lengthResized == 1) { + return 0.0; + } else { + // The whole part and the fractional part are calculated separately due to inaccuracy of floating + // point division. As an example, f32(21) / f32(7) may evaluate to 2.99... instead of 3, causing an + // offset-by-one error later in floor(). + let whole = ${t}(xResized * (lengthOriginal - 1) / (lengthResized - 1)); + let fract = + ${t}(xResized * (lengthOriginal - 1) % (lengthResized - 1)) / ${t}(lengthResized - 1); + return whole + fract; + }`;case"tf_crop_and_resize":return`if (lengthResized > 1) { + return ${t}(roiStart) * ${t}(lengthOriginal - 1) + + (${t}(xResized) * ${t}(roiEnd - roiStart) * ${t}(lengthOriginal - 1)) / + ${t}(lengthResized - 1); + } else { + return 0.5 * ${t}(roiStart + roiEnd) * ${t}(lengthOriginal - 1); + }`;case"half_pixel_symmetric":return`const outputWidth = ${t}xScale * ${t}(lengthResized); + const adjustment = ${t}(lengthResized) / outputWidth; + const center = ${t}(lengthOriginal) / 2; + const offset = center * (1 - adjustment); + return offset + ((${t}(xResized) + 0.5) / ${t}(xScale)) - 0.5;`;case"half_pixel":return`return ((${t}(xResized) + 0.5) / ${t}(xScale)) - 0.5;`;default:throw new Error(`Coordinate transform mode ${e} is not supported`)}})()+"}",oc=(e,t,s)=>`fn getNearestPixelFromOriginal(xOriginal: ${s}, isDownSample: bool) -> ${s} {`+(()=>{switch(e){case"round_prefer_ceil":return"if (fract(xOriginal) == 0.5) { return ceil(xOriginal); } else { return round(xOriginal); }";case"floor":return"return floor(xOriginal);";case"ceil":return"return ceil(xOriginal);";case"round_prefer_floor":return"if (fract(xOriginal) == 0.5) { return floor(xOriginal); } else { return round(xOriginal); }";case"simple":default:if(t<11)return"if (isDownSample) { return ceil(xOriginal); } else { return xOriginal; }";throw new Error(`Nearest mode ${e} is not supported`)}})()+"}",ac=(e,t,s)=>{let n=new Array(s).fill(0).concat(new Array(s).fill(1)),i=e.length===0?n:e.slice();return t.length>0?(t.forEach((a,o)=>{n[a]=i[o],n[o+s]=i[t.length+o]}),n):i},lc=(e,t,s,n)=>{let i=[];if(s.length>0)if(n.length>0){if(e.forEach(a=>i.push(a)),Math.max(...n)>e.length)throw new Error("axes is out of bound");n.forEach((a,o)=>i[a]=s[o])}else s.forEach(a=>i.push(a));else{if(t.length===0)throw new Error("Resize requires either scales or sizes.");i=e.map((a,o)=>Math.round(a*t[o]))}return i},uc=(e,t,s)=>{let n=(()=>{switch(s.keepAspectRatioPolicy){case"not_larger":return s.axes.length>0?Math.min(...s.axes.map(a=>t[a]),Number.MAX_VALUE):Math.min(...t,Number.MAX_VALUE);case"not_smaller":return s.axes.length>0?Math.max(...s.axes.map(a=>t[a]),Number.MIN_VALUE):Math.max(...t,Number.MIN_VALUE);default:throw new Error(`Keep aspect ratio policy ${s.keepAspectRatioPolicy} is not supported`)}})();t.fill(1,0,t.length);let i=e.slice();return s.axes.length>0?(s.axes.forEach(a=>t[a]=n),s.axes.forEach(a=>i[a]=Math.round(e[a]*t[a]))):(t.fill(n,0,t.length),i.forEach((a,o)=>i[o]=Math.round(a*t[o]))),i},dc=(e,t,s,n,i)=>` + fn calculateOriginalIndicesFromOutputIndices(output_indices: ${e.type.indices}) -> array<${e.type.value}, ${s.length}> { + var original_indices: array<${e.type.value}, ${s.length}>; + for (var i:u32 = 0; i < ${s.length}; i++) { + var output_index = ${e.indicesGet("output_indices","i")}; + var scale = ${$t("uniforms.scales","i",n)}; + var roi_low = ${$t("uniforms.roi","i",i)}; + var roi_hi = ${$t("uniforms.roi",`i + ${t.length}`,i)}; + if (scale == 1.0) { + original_indices[i] = ${e.type.value}(output_index); + } else { + var input_shape_i = ${$t("uniforms.input_shape","i",t.length)}; + var output_shape_i = ${$t("uniforms.output_shape","i",s.length)}; + original_indices[i] = getOriginalCoordinateFromResizedCoordinate(output_index, scale, output_shape_i, + input_shape_i, roi_low, roi_hi); + } + } + return original_indices; + }`,cc=(e,t,s,n,i,a,o)=>` + fn calculateInputIndicesFromOutputIndices(output_indices: ${t.type.indices}) -> ${e.type.indices} { + var input_indices: ${e.type.indices}; + for (var i:u32 = 0; i < ${n.length}; i++) { + var output_index = ${t.indicesGet("output_indices","i")}; + var input_index: u32; + var scale = ${$t("uniforms.scales","i",i)}; + if (scale == 1.0) { + input_index = output_index; + } else { + var roi_low = ${$t("uniforms.roi","i",a)}; + var roi_hi = ${$t("uniforms.roi",`i + ${s.length}`,a)}; + var input_shape_i = ${$t("uniforms.input_shape","i",s.length)}; + var output_shape_i = ${$t("uniforms.output_shape","i",n.length)}; + var original_idx = getOriginalCoordinateFromResizedCoordinate(output_index, scale, output_shape_i, + input_shape_i, roi_low, roi_hi); + if (!${o} || (original_idx >= 0 && original_idx < ${t.type.value}(input_shape_i))) { + if (original_idx < 0) { + input_index = 0; + } else if (original_idx > ${t.type.value}(input_shape_i - 1)) { + input_index = input_shape_i - 1; + } else { + input_index = u32(getNearestPixelFromOriginal(original_idx, scale < 1)); + } + } else { + input_index = u32(original_idx); + } + } + ${e.indicesSet("input_indices","i"," input_index")} + } + return input_indices; + }`,pc=(e,t)=>` + fn checkInputIndices(input_indices: ${e.type.indices}) -> bool { + for (var i:u32 = 0; i < ${t.length}; i++) { + var input_index = ${e.indicesGet("input_indices","i")}; + if (input_index < 0 || input_index >= ${$t("uniforms.input_shape","i",t.length)}) { + return false; + } + } + return true; + }`,ka=(e,t,s,n)=>e.rank>n?` + ${e.indicesSet("input_indices",t,"channel")}; + ${e.indicesSet("input_indices",s,"batch")}; +`:"",hc=(e,t,s,n,i)=>{let[a,o,u,p]=s.length===2?[-1,0,1,-1]:[0,2,3,1],h=e.type.value;return` + fn getInputValue(batch: u32, channel: u32, row: u32, col: u32) -> ${h} { + var input_indices: ${e.type.indices}; + ${e.indicesSet("input_indices",o,`max(0, min(row, ${s[o]} - 1))`)}; + ${e.indicesSet("input_indices",u,`max(0, min(col, ${s[u]} - 1))`)}; + ${ka(e,p,a,2)} + return ${e.getByIndices("input_indices")}; + } + + fn bilinearInterpolation(output_indices: ${t.type.indices}) -> ${h} { + var originalIndices = calculateOriginalIndicesFromOutputIndices(output_indices); + var row:${h} = originalIndices[${o}]; + var col:${h} = originalIndices[${u}]; + ${n?`if (row < 0 || row > (${s[o]} - 1) || col < 0 || col > (${s[u]} - 1)) { + return ${i}; + }`:""}; + row = max(0, min(row, ${s[o]} - 1)); + col = max(0, min(col, ${s[u]} - 1)); + var row1: u32 = u32(row); + var col1: u32 = u32(col); + var row2: u32 = u32(row + 1); + var col2: u32 = u32(col + 1); + var channel: u32 = ${s.length>2?`u32(originalIndices[${p}])`:"0"}; + var batch: u32 = ${s.length>2?`u32(originalIndices[${a}])`:"0"}; + var x11: ${h} = getInputValue(batch, channel, row1, col1); + var x12: ${h} = getInputValue(batch, channel, row1, col2); + var x21: ${h} = getInputValue(batch, channel, row2, col1); + var x22: ${h} = getInputValue(batch, channel, row2, col2); + var dx1: ${h} = abs(row - ${h}(row1)); + var dx2: ${h} = abs(${h}(row2) - row); + var dy1: ${h} = abs(col - ${h}(col1)); + var dy2: ${h} = abs(${h}(col2) - col); + if (row1 == row2) { + dx1 = 0.5; + dx2 = 0.5; + } + if (col1 == col2) { + dy1 = 0.5; + dy2 = 0.5; + } + return (x11 * dx2 * dy2 + x12 * dx2 * dy1 + x21 * dx1 * dy2 + x22 * dx1 * dy1); + }`},mc=(e,t,s,n,i,a,o,u,p,h)=>{let k=s.length===2,[C,d]=k?[0,1]:[2,3],z=e.type.value,B=V=>{let Z=V===C?"row":"col";return` + fn ${Z}CubicInterpolation(input_indices: ${e.type.indices}, output_indices: ${t.type.indices}) -> ${z} { + var output_index = ${t.indicesGet("output_indices",V)}; + var originalIdx: ${z} = getOriginalCoordinateFromResizedCoordinate(output_index, ${i[V]}, + ${n[V]}, ${s[V]}, ${a[V]}, ${a[V]} + ${s.length}); + var fractOriginalIdx: ${z} = originalIdx - floor(originalIdx); + var coefs = getCubicInterpolationCoefs(fractOriginalIdx); + + if (${u} && (originalIdx < 0 || originalIdx > (${s[V]} - 1))) { + return ${p}; + } + var data: array<${z}, 4> = array<${z}, 4>(0.0, 0.0, 0.0, 0.0); + for (var i: i32 = -1; i < 3; i++) { + var ${Z}: ${z} = originalIdx + ${z}(i); + if (${Z} < 0 || ${Z} >= ${s[V]}) { + ${h?`coefs[i + 1] = 0.0; + continue;`:u?`return ${p};`:`${Z} = max(0, min(${Z}, ${s[V]} - 1));`}; + } + var input_indices_copy: ${e.type.indices} = input_indices; + ${e.indicesSet("input_indices_copy",V,`u32(${Z})`)}; + data[i + 1] = ${V===C?e.getByIndices("input_indices_copy"):"rowCubicInterpolation(input_indices_copy, output_indices)"}; + } + return cubicInterpolation1D(data, coefs); + }`};return` + ${B(C)}; + ${B(d)}; + fn getCubicInterpolationCoefs(s: ${z}) -> array<${z}, 4> { + var absS = abs(s); + var coeffs: array<${z}, 4> = array<${z}, 4>(0.0, 0.0, 0.0, 0.0); + var oneMinusAbsS: ${z} = 1.0 - absS; + var twoMinusAbsS: ${z} = 2.0 - absS; + var onePlusAbsS: ${z} = 1.0 + absS; + coeffs[0] = ((${o} * onePlusAbsS - 5 * ${o}) * onePlusAbsS + 8 * ${o}) * onePlusAbsS - 4 * ${o}; + coeffs[1] = ((${o} + 2) * absS - (${o} + 3)) * absS * absS + 1; + coeffs[2] = ((${o} + 2) * oneMinusAbsS - (${o} + 3)) * oneMinusAbsS * oneMinusAbsS + 1; + coeffs[3] = ((${o} * twoMinusAbsS - 5 * ${o}) * twoMinusAbsS + 8 * ${o}) * twoMinusAbsS - 4 * ${o}; + return coeffs; + } + + fn cubicInterpolation1D(x: array<${z}, 4>, coefs: array<${z}, 4>) -> ${z} { + var coefsSum: ${z} = coefs[0] + coefs[1] + coefs[2] + coefs[3]; + return (x[0] * coefs[0] + x[1] * coefs[1]+ x[2] * coefs[2]+ x[3] * coefs[3]) / coefsSum; + } + + fn bicubicInterpolation(output_indices: ${t.type.indices}) -> ${z} { + var input_indices: ${e.type.indices} = output_indices; + return colCubicInterpolation(input_indices, output_indices); + } + `},fc=(e,t,s,n,i)=>{let[a,o,u,p,h]=s.length===3?[-1,0,1,2,-1]:[0,2,3,4,1],k=e.type.value;return` + fn getInputValue(batch: u32, channel: u32, depth:u32, height: u32, width: u32) -> ${k} { + var input_indices: ${e.type.indices}; + ${e.indicesSet("input_indices",o,`max(0, min(depth, ${s[o]} - 1))`)}; + ${e.indicesSet("input_indices",u,`max(0, min(height, ${s[u]} - 1))`)}; + ${e.indicesSet("input_indices",p,`max(0, min(width, ${s[p]} - 1))`)}; + ${ka(e,h,a,3)} + return ${e.getByIndices("input_indices")}; + } + + fn trilinearInterpolation(output_indices: ${t.type.indices}) -> ${k} { + var originalIndices = calculateOriginalIndicesFromOutputIndices(output_indices); + var depth:${k} = originalIndices[${o}]; + var height:${k} = originalIndices[${u}]; + var width:${k} = originalIndices[${p}]; + ${n?`if (depth < 0 || depth > (${s[o]} - 1) || height < 0 || height > (${s[u]} - 1) || width < 0 || (width > ${s[p]} - 1)) { + return ${i}; + }`:""}; + + depth = max(0, min(depth, ${s[o]} - 1)); + height = max(0, min(height, ${s[u]} - 1)); + width = max(0, min(width, ${s[p]} - 1)); + var depth1: u32 = u32(depth); + var height1: u32 = u32(height); + var width1: u32 = u32(width); + var depth2: u32 = u32(depth + 1); + var height2: u32 = u32(height + 1); + var width2: u32 = u32(width + 1); + var channel: u32 = ${s.length>3?`u32(originalIndices[${h}])`:"0"}; + var batch: u32 = ${s.length>3?`u32(originalIndices[${a}])`:"0"}; + + var x111: ${k} = getInputValue(batch, channel, depth1, height1, width1); + var x112: ${k} = getInputValue(batch, channel, depth1, height1, width2); + var x121: ${k} = getInputValue(batch, channel, depth1, height2, width1); + var x122: ${k} = getInputValue(batch, channel, depth1, height2, width2); + var x211: ${k} = getInputValue(batch, channel, depth2, height1, width1); + var x212: ${k} = getInputValue(batch, channel, depth2, height1, width2); + var x221: ${k} = getInputValue(batch, channel, depth2, height2, width1); + var x222: ${k} = getInputValue(batch, channel, depth2, height2, width2); + var dx1: ${k} = abs(depth - ${k}(depth1)); + var dx2: ${k} = abs(${k}(depth2) - depth); + var dy1: ${k} = abs(height - ${k}(height1)); + var dy2: ${k} = abs(${k}(height2) - height); + var dz1: ${k} = abs(width - ${k}(width1)); + var dz2: ${k} = abs(${k}(width2) - width); + if (depth1 == depth2) { + dx1 = 0.5; + dx2 = 0.5; + } + if (height1 == height2) { + dy1 = 0.5; + dy2 = 0.5; + } + if (width1 == width2) { + dz1 = 0.5; + dz2 = 0.5; + } + return (x111 * dx2 * dy2 * dz2 + x112 * dx2 * dy2 * dz1 + x121 * dx2 * dy1 *dz2 + x122 * dx2 * dy1 * dz1 + + x211 * dx1 * dy2 * dz2 + x212 * dx1 * dy2 * dz1 + x221 * dx1 * dy1 *dz2 + x222 * dx1 * dy1 * dz1); + }`},_c=(e,t,s,n,i,a)=>{let o=e.dims,u=ac(a,t.axes,o.length),p=lc(o,n,i,t.axes),h=n.slice();n.length===0&&(h=o.map((Y,he)=>Y===0?1:p[he]/Y),t.keepAspectRatioPolicy!=="stretch"&&(p=uc(o,h,t)));let k=It("output",e.dataType,p.length),C=qe("input",e.dataType,o.length),d=ze.size(p),z=o.length===p.length&&o.every((Y,he)=>Y===p[he]),B=t.coordinateTransformMode==="tf_crop_and_resize",V=t.extrapolationValue,Z=C.type.value,ee=Y=>` + ${z?"":` + ${ic(t.coordinateTransformMode,Z)}; + ${(()=>{switch(t.mode){case"nearest":return` + ${pc(C,o)}; + ${oc(t.nearestMode,s,Z)}; + ${cc(C,k,o,p,h.length,u.length,B)}; + `;case"linear":return` + ${dc(k,o,p,h.length,u.length)}; + ${(()=>{if(o.length===2||o.length===4)return`${hc(C,k,o,B,V)}`;if(o.length===3||o.length===5)return`${fc(C,k,o,B,V)}`;throw Error("Linear mode only supports input dims 2, 3, 4 and 5 are supported in linear mode.")})()}; + `;case"cubic":return` + ${(()=>{if(o.length===2||o.length===4)return`${mc(C,k,o,p,h,u,t.cubicCoeffA,B,t.extrapolationValue,t.excludeOutside)}`;throw Error("Cubic mode only supports input dims 2 and 4 are supported in linear mode.")})()}; + `;default:throw Error("Invalid resize mode")}})()}; + `} + ${Y.registerUniform("output_size","u32").registerUniform("scales","f32",h.length).registerUniform("roi","f32",u.length).declareVariables(C,k)} + ${Y.mainStart()} + ${Y.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + ${z?"output[global_idx] = input[global_idx];":` + let output_indices = ${k.offsetToIndices("global_idx")}; + var input_indices: ${C.type.indices}; + ${(()=>{switch(t.mode){case"nearest":return`input_indices = calculateInputIndicesFromOutputIndices(output_indices); + if (checkInputIndices(input_indices)) { + output[global_idx] = ${C.getByIndices("input_indices")}; + } else { + output[global_idx] = ${t.extrapolationValue}; + }`;case"linear":return`output[global_idx] = ${o.length===2||o.length===4?"bilinearInterpolation":"trilinearInterpolation"}(output_indices);`;case"cubic":return"output[global_idx] = bicubicInterpolation(output_indices);";default:throw Error(`Unsupported resize mode: ${t.mode}`)}})()}; +`} + }`;return{name:"Resize",shaderCache:{hint:`${t.cacheKey}|${s}|${h.length>0?h:""}|${i.length>0?i:""}|${u.length>0?u:""}|${z}|${o}`,inputDependencies:["rank"]},getShaderSource:ee,getRunData:()=>({outputs:[{dims:p,dataType:e.dataType}],dispatchGroup:{x:Math.ceil(d/64)},programUniforms:[{type:12,data:d},{type:1,data:h},{type:1,data:u},...yt(o,p)]})}},gc=e=>{let t=e.customDataBuffer;return new Uint32Array(t,t.byteOffset,1)[0]},wc=(e,t)=>{let s=[],n=[],i=[],a=gc(e);if(t.antialias!==0)throw Error("Only default value (0) for Antialias attribute is supported");nc(e.inputs,t,a,s,n,i),e.compute(_c(e.inputs[0],t,a,s,n,i),{inputs:[0]})},yc=e=>{let t=e.antialias,s=e.axes,n=e.coordinateTransformMode,i=e.cubicCoeffA,a=e.excludeOutside!==0,o=e.extrapolationValue,u=e.keepAspectRatioPolicy,p=e.mode,h=e.nearestMode===""?"simple":e.nearestMode;return Bt({antialias:t,axes:s,coordinateTransformMode:n,cubicCoeffA:i,excludeOutside:a,extrapolationValue:o,keepAspectRatioPolicy:u,mode:p,nearestMode:h})}}),Mc,bc,vc,Pp=g(()=>{zt(),Ot(),rs(),Yt(),Mc=(e,t)=>{let[s,n,i,a]=e,{numHeads:o,rotaryEmbeddingDim:u}=t;if(s.dims.length!==3&&s.dims.length!==4)throw new Error(`Input 'x' is expected to have 3 or 4 dimensions, got ${s.dims.length}`);if(!ze.areEqual(n.dims,[])&&!ze.areEqual(n.dims,[1])&&n.dims.length!==2)throw new Error(`Input 'position_ids' is expected to have 0, 1, or 2 dimensions, got ${n.dims.length}`);if(i.dims.length!==2)throw new Error(`Input 'cos_cache' is expected to have 2 dimensions, got ${i.dims.length}`);if(a.dims.length!==2)throw new Error(`Input 'sin_cache' is expected to have 2 dimensions, got ${a.dims.length}`);if(!ze.areEqual(i.dims,a.dims))throw new Error("Inputs 'cos_cache' and 'sin_cache' are expected to have the same shape");if(u>0&&o===0)throw new Error("num_heads must be provided if rotary_embedding_dim is specified");let p=s.dims[0],h=s.dims[s.dims.length-2],k=i.dims[0],C=ze.sizeFromDimension(s.dims,1)/h,d=u===0?i.dims[1]*2:C/o;if(u>d)throw new Error("rotary_embedding_dim must be less than or equal to head_size");if(n.dims.length===2){if(p!==n.dims[0])throw new Error(`Input 'position_ids' dimension 0 should be of size batch_size, got ${n.dims[0]}`);if(h!==n.dims[1])throw new Error(`Input 'position_ids' dimension 1 should be of size sequence_length, got ${n.dims[1]}`)}if(d/2!==i.dims[1]&&u/2!==i.dims[1])throw new Error(`Input 'cos_cache' dimension 1 should be same as head_size / 2 or rotary_embedding_dim / 2, got ${i.dims[1]}`);if(h>k)throw new Error("Updating cos_cache and sin_cache in RotaryEmbedding is not currently supported")},bc=(e,t)=>{let{interleaved:s,numHeads:n,rotaryEmbeddingDim:i,scale:a}=t,o=e[0].dims[0],u=ze.sizeFromDimension(e[0].dims,1),p=e[0].dims[e[0].dims.length-2],h=u/p,k=e[2].dims[1],C=i===0?k*2:h/n,d=new Array(o,p,h/C,C-k),z=ze.computeStrides(d),B=[{type:1,data:a},{type:12,data:d},{type:12,data:z},...e[0].dims.length===3?new Array({type:12,data:[u,h,C,1]}):[],...e[0].dims.length===4?new Array({type:12,data:[u,C,p*C,1]}):[],...yt(e[0].dims,e[1].dims,e[2].dims,e[3].dims,e[0].dims)],V=Z=>{let ee=qe("input",e[0].dataType,e[0].dims.length),Y=qe("position_ids",e[1].dataType,e[1].dims.length),he=qe("cos_cache",e[2].dataType,e[2].dims.length),pe=qe("sin_cache",e[3].dataType,e[3].dims.length),Me=It("output",e[0].dataType,e[0].dims.length);return Z.registerUniforms([{name:"scale",type:"f32"},{name:"global_shape",type:"u32",length:d.length},{name:"global_strides",type:"u32",length:z.length},{name:"input_output_strides",type:"u32",length:z.length}]),` + ${Z.declareVariables(ee,Y,he,pe,Me)} + + ${Z.mainStart(ir)} + let half_rotary_emb_dim = uniforms.${he.name}_shape[1]; + let bsnh = global_idx / uniforms.global_strides % uniforms.global_shape; + let size = uniforms.global_shape[0] * uniforms.global_strides[0]; + ${Z.guardAgainstOutOfBoundsWorkgroupSizes("size")} + + if (bsnh[3] < half_rotary_emb_dim) { + let position_ids_idx = + ${Y.broadcastedIndicesToOffset("bsnh.xy",It("",Y.type.tensor,2))}; + let position_id = + u32(${Y.getByOffset("position_ids_idx")}) + select(0, bsnh[1], position_ids_idx == 0); + let i = dot(bsnh, uniforms.input_output_strides) + select(0, bsnh[3], ${s}); + let j = i + select(half_rotary_emb_dim, 1, ${s}); + let re = ${ee.getByOffset("i")} * ${he.get("position_id","bsnh[3]")} - + ${ee.getByOffset("j")} * ${pe.get("position_id","bsnh[3]")}; + ${Me.setByOffset("i","re")} + let im = ${ee.getByOffset("i")} * ${pe.get("position_id","bsnh[3]")} + + ${ee.getByOffset("j")} * ${he.get("position_id","bsnh[3]")}; + ${Me.setByOffset("j","im")} + } else { + let k = dot(bsnh, uniforms.input_output_strides) + half_rotary_emb_dim; + ${Me.setByOffset("k",ee.getByOffset("k"))} + } + }`};return{name:"RotaryEmbedding",shaderCache:{hint:Bt({interleaved:s}).cacheKey,inputDependencies:["rank","rank","rank","rank"]},getShaderSource:V,getRunData:()=>({outputs:[{dims:e[0].dims,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(ze.size(d)/ir)},programUniforms:B})}},vc=(e,t)=>{Mc(e.inputs,t),e.compute(bc(e.inputs,t))}}),xc,Tc,Pc,qp=g(()=>{zt(),Ot(),Yt(),xc=e=>{if(!e||e.length<3)throw new Error("layerNorm requires at least 3 inputs.");let t=e[0],s=e[1],n=e[2];if(t.dataType!==s.dataType||t.dataType!==n.dataType)throw new Error("All inputs must have the same data type");if(t.dims.length!==3&&t.dims.length!==2)throw new Error("Input must be 2D or 3D");if(s.dims.length!==3&&s.dims.length!==2)throw new Error("Skip must be 2D or 3D");let i=t.dims[t.dims.length-1],a=t.dims[t.dims.length-2];if(s.dims[s.dims.length-1]!==i)throw new Error("Skip must have the same hidden size as input");if(s.dims[s.dims.length-2]!==a)throw new Error("Skip must have the same sequence length as input");if(n.dims.length!==1)throw new Error("Gamma must be 1D");if(n.dims[n.dims.length-1]!==i)throw new Error("Gamma must have the same hidden size as input");if(e.length>3){let o=e[3];if(o.dims.length!==1)throw new Error("Beta must be 1D");if(o.dims[o.dims.length-1]!==i)throw new Error("Beta must have the same hidden size as input")}if(e.length>4){let o=e[4];if(o.dims.length!==1)throw new Error("Bias must be 1D");if(o.dims[o.dims.length-1]!==i)throw new Error("Bias must have the same hidden size as input")}},Tc=(e,t,s,n)=>{let i=t.simplified,a=e[0].dims,o=ze.size(a),u=a,p=o,h=a.slice(-1)[0],k=n?a.slice(0,-1).concat(1):[],C=!i&&e.length>3,d=e.length>4,z=n&&s>1,B=n&&s>2,V=s>3,Z=64,ee=qt(h),Y=[{type:12,data:p},{type:12,data:ee},{type:12,data:h},{type:1,data:t.epsilon}],he=Me=>{let Oe=[{name:"output_size",type:"u32"},{name:"components",type:"u32"},{name:"hidden_size",type:"u32"},{name:"epsilon",type:"f32"}],De=[qe("x",e[0].dataType,e[0].dims,ee),qe("skip",e[1].dataType,e[1].dims,ee),qe("gamma",e[2].dataType,e[2].dims,ee)];C&&De.push(qe("beta",e[3].dataType,e[3].dims,ee)),d&&De.push(qe("bias",e[4].dataType,e[4].dims,ee)),De.push(It("output",e[0].dataType,u,ee)),z&&De.push(It("mean_output",1,k)),B&&De.push(It("inv_std_output",1,k)),V&&De.push(It("input_skip_bias_sum",e[0].dataType,u,ee));let Ye=fs(e[0].dataType),at=fs(1,ee);return` + + ${Me.registerUniforms(Oe).declareVariables(...De)} + var sum_shared : array<${at}, ${Z}>; + var sum_squared_shared : array<${at}, ${Z}>; + + ${Me.mainStart([Z,1,1])} + let ix = local_id.x; + let iy = global_id.x / ${Z}; + + let hidden_size_vectorized: u32 = uniforms.hidden_size / uniforms.components; + var stride = hidden_size_vectorized / ${Z}; + let offset = ix * stride + iy * hidden_size_vectorized; + let offset1d = stride * ix; + if (ix == ${Z-1}) { + stride = hidden_size_vectorized - stride * ix; + } + for (var i: u32 = 0; i < stride; i++) { + let skip_value = skip[offset + i]; + let bias_value = ${d?"bias[offset1d + i]":Ye+"(0.0)"}; + let input_value = x[offset + i]; + let value = input_value + skip_value + bias_value; + ${V?"input_skip_bias_sum[offset + i] = value;":""} + output[offset + i] = value; + let f32_value = ${$s(Ye,ee,"value")}; + sum_shared[ix] += f32_value; + sum_squared_shared[ix] += f32_value * f32_value; + } + workgroupBarrier(); + + var reduce_size : u32 = ${Z}; + for (var curr_size = reduce_size >> 1; curr_size > 0; curr_size = reduce_size >> 1) { + reduce_size = curr_size + (reduce_size & 1); + if (ix < curr_size) { + sum_shared[ix] += sum_shared[ix + reduce_size]; + sum_squared_shared[ix] += sum_squared_shared[ix + reduce_size]; + } + workgroupBarrier(); + } + + let sum = sum_shared[0]; + let square_sum = sum_squared_shared[0]; + let mean = ${Gs("sum",ee)} / f32(uniforms.hidden_size); + let inv_std_dev = inverseSqrt(${Gs("square_sum",ee)} / f32(uniforms.hidden_size) ${i?"":"- mean * mean"} + uniforms.epsilon); + ${z?"mean_output[global_idx] = mean;":""} + ${B?"inv_std_output[global_idx] = inv_std_dev;":""} + + for (var i: u32 = 0; i < stride; i++) { + output[offset + i] = (output[offset + i] ${i?"":`- ${Ye}(mean)`}) * + ${Ye}(inv_std_dev) * gamma[offset1d + i] + ${C?"+ beta[offset1d + i]":""}; + } + }`},pe=[{dims:u,dataType:e[0].dataType}];return s>1&&pe.push({dims:k,dataType:1}),s>2&&pe.push({dims:k,dataType:1}),s>3&&pe.push({dims:a,dataType:e[0].dataType}),{name:"SkipLayerNormalization",shaderCache:{hint:`${ee};${z};${B};${V}`,inputDependencies:e.map((Me,Oe)=>"type")},getShaderSource:he,getRunData:()=>({outputs:pe,dispatchGroup:{x:Math.ceil(p/h)},programUniforms:Y})}},Pc=(e,t)=>{xc(e.inputs);let s=[0];e.outputCount>1&&s.push(-3),e.outputCount>2&&s.push(-3),e.outputCount>3&&s.push(3),e.compute(Tc(e.inputs,t,e.outputCount,!1),{outputs:s})}}),Qt,Hn,Hs,qs,tr,cn,Ep,Ec,Cp=g(()=>{zt(),Ot(),rs(),Yt(),Qt=(e,t)=>{if(!e||e.length<1)throw new Error("too few inputs");if(t.axes.length!==0){if(t.axes.length!==t.starts.length||t.axes.length!==t.ends.length)throw new Error("axes, starts and ends must have the same length")}else if(t.starts.length!==t.ends.length)throw new Error("starts and ends must have the same length");e.slice(1).forEach((s,n)=>{if(e[n+1].dataType!==6&&e[n+1].dataType!==7)throw new Error(`Input ${n} must be an array of int32 or int64`)})},Hn=(e,t)=>{let s=[];if(e.length>t)if(e[t].dataType===7)e[t].getBigInt64Array().forEach(n=>s.push(Number(n)));else if(e[t].dataType===6)e[t].getInt32Array().forEach(n=>s.push(Number(n)));else throw new Error(`Input ${t} must be an array of int32 or int64`);return s},Hs=(e,t)=>{if(e.length>1){let s=Hn(e,1),n=Hn(e,2),i=Hn(e,3);return i.length===0&&(i=[...Array(e[0].dims.length).keys()]),Bt({starts:s,ends:n,axes:i})}else return t},qs=(e,t,s,n,i)=>{let a=e;return e<0&&(a+=s[n[t]]),i[t]<0?Math.max(0,Math.min(a,s[n[t]]-1)):Math.max(0,Math.min(a,s[n[t]]))},tr=(e,t,s)=>`fn calculateInputIndices(output_indices: ${t.type.indices}) -> ${e.type.indices} { + var input_indices: ${e.type.indices}; + var carry = 0u; + for (var i = ${s.length}; i >= 0; i--) { + let input_shape_i = ${$t("uniforms.input_shape","i",s.length)}; + let steps_i = ${$t("uniforms.steps","i",s.length)}; + let signs_i = ${$t("uniforms.signs","i",s.length)}; + let starts_i = ${$t("uniforms.starts","i",s.length)}; + var output_index = ${t.indicesGet("output_indices","i")}; + var input_index = output_index * steps_i + starts_i + carry; + carry = input_index / input_shape_i; + input_index = input_index % input_shape_i; + if (signs_i < 0) { + input_index = input_shape_i - input_index - 1u + starts_i; + } + ${e.indicesSet("input_indices","i","input_index")}; + } + return input_indices; + }`,cn=(e,t)=>{let s=e[0].dims,n=ze.size(s),i=t.axes.length>0?ze.normalizeAxes(t.axes,s.length):[...Array(s.length).keys()],a=Hn(e,4);a.forEach(ee=>ee!==0||(()=>{throw new Error("step cannot be 0")})),a.length===0&&(a=Array(i.length).fill(1));let o=t.starts.map((ee,Y)=>qs(ee,Y,s,i,a)),u=t.ends.map((ee,Y)=>qs(ee,Y,s,i,a));if(i.length!==o.length||i.length!==u.length)throw new Error("start, ends and axes should have the same number of elements");if(i.length!==s.length)for(let ee=0;eeMath.sign(ee));a.forEach((ee,Y,he)=>{if(ee<0){let pe=(u[Y]-o[Y])/ee,Me=o[Y],Oe=Me+pe*a[Y];o[Y]=Oe,u[Y]=Me,he[Y]=-ee}});let h=s.slice(0);i.forEach((ee,Y)=>{h[ee]=Math.ceil((u[ee]-o[ee])/a[ee])});let k={dims:h,dataType:e[0].dataType},C=It("output",e[0].dataType,h.length),d=qe("input",e[0].dataType,e[0].dims.length),z=ze.size(h),B=[{name:"outputSize",type:"u32"},{name:"starts",type:"u32",length:o.length},{name:"signs",type:"i32",length:p.length},{name:"steps",type:"u32",length:a.length}],V=[{type:12,data:z},{type:12,data:o},{type:6,data:p},{type:12,data:a},...yt(e[0].dims,h)],Z=ee=>` + ${ee.registerUniforms(B).declareVariables(d,C)} + ${tr(d,C,s)} + ${ee.mainStart()} + ${ee.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")} + let output_indices = ${C.offsetToIndices("global_idx")}; + let input_indices = calculateInputIndices(output_indices); + ${C.setByOffset("global_idx",d.getByIndices("input_indices"))} + }`;return{name:"Slice",shaderCache:{hint:`${p.length}_${o.length}_${a.length}`,inputDependencies:["rank"]},getShaderSource:Z,getRunData:()=>({outputs:[k],dispatchGroup:{x:Math.ceil(n/64)},programUniforms:V})}},Ep=(e,t)=>{Qt(e.inputs,t);let s=Hs(e.inputs,t);e.compute(cn(e.inputs,s),{inputs:[0]})},Ec=e=>{let t=e.starts,s=e.ends,n=e.axes;return Bt({starts:t,ends:s,axes:n})}}),f,T,N,_e,Fe=g(()=>{zt(),Ot(),rs(),Kr(),Yt(),f=e=>{if(!e||e.length!==1)throw new Error("Softmax op requires 1 input.")},T=(e,t)=>{let s=e.inputs[0],n=s.dims,i=ze.size(n),a=n.length,o=ze.normalizeAxis(t.axis,a),u=oYe),h[o]=a-1,h[a-1]=o,p=e.compute(pr(s,h),{inputs:[s],outputs:[-1]})[0]):p=s;let k=p.dims,C=k[a-1],d=i/C,z=qt(C),B=C/z,V=64;d===1&&(V=256);let Z=(De,Ye)=>Ye===4?`max(max(${De}.x, ${De}.y), max(${De}.z, ${De}.w))`:Ye===2?`max(${De}.x, ${De}.y)`:Ye===3?`max(max(${De}.x, ${De}.y), ${De}.z)`:De,ee=qe("x",p.dataType,p.dims,z),Y=It("result",p.dataType,p.dims,z),he=ee.type.value,pe=fs(p.dataType)==="f32"?`var threadMax = ${he}(-3.402823e+38f);`:`var threadMax = ${he}(-65504.0h);`,Me=De=>` + var rowMaxShared : ${he}; + var rowSumShared : ${he}; + var threadShared : array<${he}, ${V}>; + + fn getValue(row: i32, col: i32, row_stride: i32) -> ${he} { + let index = row * row_stride + col; + return x[index]; + } + + fn setValue(row: i32, col: i32, row_stride: i32, value: ${he}) { + let index = row * row_stride + col; + result[index] = value; + } + ${De.registerUniform("packedCols","i32").declareVariables(ee,Y)} + ${De.mainStart(V)} + let gindex = i32(global_idx); + let lindex = i32(local_idx); + const wg = ${V}; + let row = gindex / wg; + let cols = uniforms.packedCols; + let row_stride : i32 = uniforms.packedCols; + + // find the rows max + ${pe} + for (var col = lindex; col < cols; col += wg) { + let value = getValue(row, col, row_stride); + threadMax = max(threadMax, value); + } + if (lindex < cols) { + threadShared[lindex] = threadMax; + } + workgroupBarrier(); + + var reduceSize = min(cols, wg); + for (var currSize = reduceSize >> 1; currSize > 0; currSize = reduceSize >> 1) { + reduceSize = currSize + (reduceSize & 1); + if (lindex < currSize) { + threadShared[lindex] = max(threadShared[lindex], threadShared[lindex + reduceSize]); + } + workgroupBarrier(); + } + if (lindex == 0) { + rowMaxShared = ${he}(${Z("threadShared[0]",z)}); + } + workgroupBarrier(); + + // find the rows sum + var threadSum = ${he}(0.0); + for (var col = lindex; col < cols; col += wg) { + let subExp = exp(getValue(row, col, row_stride) - rowMaxShared); + threadSum += subExp; + } + threadShared[lindex] = threadSum; + workgroupBarrier(); + + for (var currSize = wg >> 1; currSize > 0; currSize = currSize >> 1) { + if (lindex < currSize) { + threadShared[lindex] = threadShared[lindex] + threadShared[lindex + currSize]; + } + workgroupBarrier(); + } + if (lindex == 0) { + rowSumShared = ${he}(${Gs("threadShared[0]",z)}); + } + workgroupBarrier(); + + // calculate final value for each element in the row + for (var col = lindex; col < cols; col += wg) { + let value = exp(getValue(row, col, row_stride) - rowMaxShared) / rowSumShared; + setValue(row, col, row_stride, value); + } + }`,Oe=e.compute({name:"Softmax",shaderCache:{hint:`${z};${V}`,inputDependencies:["type"]},getRunData:()=>({outputs:[{dims:k,dataType:p.dataType}],dispatchGroup:{x:d},programUniforms:[{type:6,data:B}]}),getShaderSource:Me},{inputs:[p],outputs:[u?-1:0]})[0];u&&e.compute(pr(Oe,h),{inputs:[Oe]})},N=(e,t)=>{f(e.inputs),T(e,t)},_e=e=>Bt({axis:e.axis})}),Ae,et,rt,_t,Mt,jt=g(()=>{zt(),Ot(),Yt(),Ae=e=>Array.from(e.getBigInt64Array(),Number),et=e=>{if(!e||e.length!==2)throw new Error("Tile requires 2 inputs.");if(e[0].dataType!==1&&e[0].dataType!==10&&e[0].dataType!==6&&e[0].dataType!==12)throw new Error("Tile only support float, float16, int32, and uint32 data types");if(e[1].dataType!==7)throw new Error("Tile `repeats` input should be of int64 data type");if(e[1].dims.length!==1)throw new Error("Tile `repeats` input should be 1-D");if(Ae(e[1]).length!==e[0].dims.length)throw new Error("Tile `repeats` input should have same number of elements as rank of input data tensor")},rt=(e,t)=>{let s=[];for(let n=0;n{let s=e[0].dims,n=t??Ae(e[1]),i=rt(s,n),a=ze.size(i),o=e[0].dataType,u=qe("input",o,s.length),p=It("output",o,i.length),h=k=>` + const inputShape = ${u.indices(...s)}; + ${k.registerUniform("output_size","u32").declareVariables(u,p)} + ${k.mainStart()} + ${k.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let output_indices = ${p.offsetToIndices("global_idx")}; + var input_indices: ${u.type.indices}; + for (var i = 0; i < ${s.length}; i++) { + let input_dim_i = ${u.indicesGet("uniforms.input_shape","i")}; + let input_dim_value = ${p.indicesGet("output_indices","i")} % input_dim_i; + + ${u.indicesSet("input_indices","i","input_dim_value")} + } + ${p.setByOffset("global_idx",u.getByIndices("input_indices"))} + }`;return{name:"Tile",shaderCache:{hint:`${n}`,inputDependencies:["rank"]},getRunData:()=>({outputs:[{dims:i,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(a/64)},programUniforms:[{type:12,data:a},...yt(e[0].dims,i)]}),getShaderSource:h}},Mt=e=>{et(e.inputs),e.compute(_t(e.inputs),{inputs:[0]})}}),Vt,Lt,Gt,ts=g(()=>{zt(),Ot(),Yt(),Vt=(e,t,s,n,i)=>{let a=It("output_data",i,s.length,4),o=qe("a_data",t[1].dataType,t[1].dims.length,4),u=qe("b_data",t[2].dataType,t[2].dims.length,4),p=qe("c_data",t[0].dataType,t[0].dims.length,4),h,k=(C,d,z)=>`select(${d}, ${C}, ${z})`;if(!n)h=a.setByOffset("global_idx",k(o.getByOffset("global_idx"),u.getByOffset("global_idx"),p.getByOffset("global_idx")));else{let C=(d,z,B="")=>{let V=`a_data[index_a${z}][component_a${z}]`,Z=`b_data[index_b${z}][component_b${z}]`,ee=`bool(c_data[index_c${z}] & (0xffu << (component_c${z} * 8)))`;return` + let output_indices${z} = ${a.offsetToIndices(`global_idx * 4u + ${z}u`)}; + let offset_a${z} = ${o.broadcastedIndicesToOffset(`output_indices${z}`,a)}; + let offset_b${z} = ${u.broadcastedIndicesToOffset(`output_indices${z}`,a)}; + let offset_c${z} = ${p.broadcastedIndicesToOffset(`output_indices${z}`,a)}; + let index_a${z} = offset_a${z} / 4u; + let index_b${z} = offset_b${z} / 4u; + let index_c${z} = offset_c${z} / 4u; + let component_a${z} = offset_a${z} % 4u; + let component_b${z} = offset_b${z} % 4u; + let component_c${z} = offset_c${z} % 4u; + ${d}[${z}] = ${B}(${k(V,Z,ee)}); + `};i===9?h=` + var data = vec4(0); + ${C("data",0,"u32")} + ${C("data",1,"u32")} + ${C("data",2,"u32")} + ${C("data",3,"u32")} + output_data[global_idx] = dot(vec4(0x1, 0x100, 0x10000, 0x1000000), vec4(data));`:h=` + ${C("output_data[global_idx]",0)} + ${C("output_data[global_idx]",1)} + ${C("output_data[global_idx]",2)} + ${C("output_data[global_idx]",3)} + `}return` + ${e.registerUniform("vec_size","u32").declareVariables(p,o,u,a)} + ${e.mainStart()} + ${e.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.vec_size")} + ${h} + }`},Lt=e=>{let t=e[1].dims,s=e[2].dims,n=e[0].dims,i=e[1].dataType,a=!(ze.areEqual(t,s)&&ze.areEqual(s,n)),o=t,u=ze.size(t);if(a){let h=Ws.calcShape(Ws.calcShape(t,s,!1),n,!1);if(!h)throw new Error("Can't perform where op on the given tensors");o=h,u=ze.size(o)}let p=Math.ceil(u/4);return{name:"Where",shaderCache:{inputDependencies:["rank","rank","rank"]},getShaderSource:h=>Vt(h,e,o,a,i),getRunData:()=>({outputs:[{dims:o,dataType:i}],dispatchGroup:{x:Math.ceil(u/64/4)},programUniforms:[{type:12,data:p},...yt(n,t,s,o)]})}},Gt=e=>{e.compute(Lt(e.inputs))}}),ns,Jt=g(()=>{Uc(),uo(),Vc(),Wc(),Gc(),Kc(),hu(),Xc(),Jc(),Zc(),ep(),tp(),sp(),rp(),np(),ip(),ap(),lp(),Kp(),ua(),pp(),Td(),hp(),mp(),Fd(),_d(),Rd(),Mp(),Hp(),bp(),xp(),di(),Tp(),Pp(),qp(),Cp(),Fe(),ha(),jt(),Kr(),Co(),ts(),ns=new Map([["Abs",[Pl]],["Acos",[fo]],["Acosh",[El]],["Add",[So]],["ArgMax",[no,io]],["ArgMin",[gl,io]],["Asin",[Cl]],["Asinh",[_o]],["Atan",[kl]],["Atanh",[Sl]],["Attention",[Ml]],["AveragePool",[yp,Pa]],["BatchNormalization",[po]],["BiasAdd",[Tl]],["BiasSplitGelu",[ko]],["Cast",[$l,go]],["Ceil",[Il]],["Clip",[wo]],["Concat",[cu,pu]],["Conv",[bi,Xo]],["ConvTranspose",[$u,Zo]],["Cos",[Ol]],["Cosh",[yo]],["CumSum",[Au,Iu]],["DepthToSpace",[Du,Lu]],["DequantizeLinear",[Qd,Xd]],["Div",[su]],["Einsum",[Nu,ju]],["Elu",[Fl,jn]],["Equal",[ru]],["Erf",[Dl]],["Exp",[Mo]],["Expand",[Gu]],["FastGelu",[Ku]],["Floor",[Ll]],["FusedConv",[bi,Xo]],["Gather",[Qu,qu]],["GatherElements",[rd,sd]],["GatherBlockQuantized",[Zu,ed]],["GatherND",[Ei,Yu]],["Gelu",[zl]],["Gemm",[Ci,od]],["GlobalAveragePool",[Ud,jd]],["GlobalMaxPool",[Kd,Gd]],["Greater",[ou]],["GreaterOrEqual",[Ao]],["GridSample",[pd,hd]],["GroupQueryAttention",[ma]],["HardSigmoid",[Ul,xo]],["InstanceNormalization",[xd]],["LayerNormalization",[ga]],["LeakyRelu",[bo,jn]],["Less",[au]],["LessOrEqual",[lu]],["Log",[Ql]],["MatMul",[kd]],["MatMulNBits",[Id,Od]],["MaxPool",[Vd,Wd]],["Mul",[nu]],["MultiHeadAttention",[up,fd]],["Neg",[Rl]],["Not",[Bl]],["Pad",[gp]],["Pow",[iu]],["QuickGelu",[Yl,jn]],["Range",[Zd]],["Reciprocal",[vo]],["ReduceMin",[hl]],["ReduceMean",[dl]],["ReduceMax",[to]],["ReduceSum",[so]],["ReduceProd",[ml]],["ReduceL1",[eo]],["ReduceL2",[cl]],["ReduceLogSum",[_l]],["ReduceLogSumExp",[pl]],["ReduceSumSquare",[fl]],["Relu",[Nl]],["Resize",[wc,yc]],["RotaryEmbedding",[vc]],["ScatterND",[Ca,vp]],["Sigmoid",[jl]],["Sin",[Vl]],["Sinh",[To]],["Slice",[Ep,Ec]],["SkipLayerNormalization",[Pc]],["Split",[Md,bd]],["Sqrt",[Wl]],["Softmax",[N,_e]],["Sub",[$o]],["Tan",[Gl]],["Tanh",[Kl]],["ThresholdedRelu",[ql,jn]],["Tile",[Mt]],["Transpose",[Va,Ki]],["Where",[Gt]]])}),os,As=g(()=>{We(),Pe(),Yt(),os=class{constructor(e){this.backend=e,this.repo=new Map,this.attributesBound=!1}getArtifact(e){return this.repo.get(e)}setArtifact(e,t){this.repo.set(e,t)}run(e,t,s,n,i){Ne(e.programInfo.name);let a=this.backend.device,o=this.backend.getComputePassEncoder();this.backend.writeTimestamp(this.backend.pendingDispatchNumber*2);let u=[];for(let h of t)u.push({binding:u.length,resource:{buffer:h.buffer}});for(let h of s)u.push({binding:u.length,resource:{buffer:h.buffer}});i&&u.push({binding:u.length,resource:i});let p=a.createBindGroup({layout:e.computePipeline.getBindGroupLayout(0),entries:u,label:e.programInfo.name});if(this.backend.sessionStatus==="capturing"){let h={kernelId:this.backend.currentKernelId,computePipeline:e.computePipeline,bindGroup:p,dispatchGroup:n};this.backend.capturedCommandList.get(this.backend.currentSessionId).push(h)}o.setPipeline(e.computePipeline),o.setBindGroup(0,p),o.dispatchWorkgroups(...n),this.backend.writeTimestamp(this.backend.pendingDispatchNumber*2+1),this.backend.pendingDispatchNumber++,(this.backend.pendingDispatchNumber>=this.backend.maxDispatchNumber||this.backend.queryType==="at-passes")&&this.backend.endComputePass(),this.backend.pendingDispatchNumber>=this.backend.maxDispatchNumber&&this.backend.flush(),Re(e.programInfo.name)}dispose(){}build(e,t){Ne(e.name);let s=this.backend.device,n=[];[{feature:"shader-f16",extension:"f16"},{feature:"subgroups",extension:"subgroups"},{feature:"subgroups-f16",extension:"subgroups_f16"}].forEach(h=>{s.features.has(h.feature)&&n.push(`enable ${h.extension};`)});let i=Ra(t,this.backend.device.limits),a=e.getShaderSource(i),o=`${n.join(` +`)} +${i.additionalImplementations} +${a}`,u=s.createShaderModule({code:o,label:e.name});as("verbose",()=>`[WebGPU] ${e.name} shader code: ${o}`);let p=s.createComputePipeline({compute:{module:u,entryPoint:"main"},layout:"auto",label:e.name});return Re(e.name),{programInfo:e,computePipeline:p,uniformVariablesInfo:i.variablesInfo}}normalizeDispatchGroupSize(e){let t=typeof e=="number"?e:e.x,s=typeof e=="number"?1:e.y||1,n=typeof e=="number"?1:e.z||1,i=this.backend.device.limits.maxComputeWorkgroupsPerDimension;if(t<=i&&s<=i&&n<=i)return[t,s,n];let a=t*s*n,o=Math.ceil(Math.sqrt(a));if(o>i){if(o=Math.ceil(Math.cbrt(a)),o>i)throw new Error("Total dispatch size exceeds WebGPU maximum.");return[o,o,o]}else return[o,o,1]}}}),xs,cs,Es,Is,Zs,Ys=g(()=>{We(),zt(),Pe(),Q(),us(),Jt(),As(),xs=(e,t)=>{if(t.length!==e.length)throw new Error(`inputDependencies length ${t.length} is not equal to inputTensors length ${e.length}.`);let s=[];for(let n=0;n{var i,a;let n=e.name;return(i=e.shaderCache)!=null&&i.hint&&(n+="["+e.shaderCache.hint+"]"),n+=":"+s+`:${xs(t,((a=e.shaderCache)==null?void 0:a.inputDependencies)??new Array(t.length).fill("dims"))}`,n},Es=class{constructor(e){e&&(this.architecture=e.architecture,this.vendor=e.vendor)}isArchitecture(e){return this.architecture===e}isVendor(e){return this.vendor===e}},Is=class{constructor(e){this.subgroupsSupported=e.features.has("subgroups"),this.subgroupsF16Supported=e.features.has("subgroups");let t=e.limits;!this.subgroupsSupported||!t.minSubgroupSize||!t.maxSubgroupSize?this.subgroupSizeRange=void 0:this.subgroupSizeRange=[t.minSubgroupSize,t.maxSubgroupSize]}},Zs=class{constructor(){this.currentSessionId=null,this.currentKernelId=null,this.commandEncoder=null,this.computePassEncoder=null,this.maxDispatchNumber=16,this.pendingDispatchNumber=0,this.pendingKernels=[],this.pendingQueries=new Map,this.sessionStatus="default",this.capturedCommandList=new Map,this.capturedPendingKernels=new Map,this.sessionExternalDataMapping=new Map}get currentKernelCustomData(){if(this.currentKernelId===null)throw new Error("currentKernelCustomData(): currentKernelId is null. (should not happen)");let e=this.kernelCustomData.get(this.currentKernelId);return e||(e={},this.kernelCustomData.set(this.currentKernelId,e)),e}async initialize(e,t){this.env=e;let s=[],n={requiredLimits:{maxComputeWorkgroupStorageSize:t.limits.maxComputeWorkgroupStorageSize,maxComputeWorkgroupsPerDimension:t.limits.maxComputeWorkgroupsPerDimension,maxStorageBufferBindingSize:t.limits.maxStorageBufferBindingSize,maxBufferSize:t.limits.maxBufferSize,maxComputeInvocationsPerWorkgroup:t.limits.maxComputeInvocationsPerWorkgroup,maxComputeWorkgroupSizeX:t.limits.maxComputeWorkgroupSizeX,maxComputeWorkgroupSizeY:t.limits.maxComputeWorkgroupSizeY,maxComputeWorkgroupSizeZ:t.limits.maxComputeWorkgroupSizeZ},requiredFeatures:s},i=a=>t.features.has(a)&&s.push(a)&&!0;i("chromium-experimental-timestamp-query-inside-passes")||i("timestamp-query"),i("shader-f16"),i("subgroups")&&i("subgroups-f16"),this.device=await t.requestDevice(n),this.deviceInfo=new Is(this.device),this.adapterInfo=new Es(t.info||await t.requestAdapterInfo()),this.gpuDataManager=ms(this),this.programManager=new os(this),this.kernels=new Map,this.kernelPersistentData=new Map,this.kernelCustomData=new Map,Tn(e.logLevel,!!e.debug),this.device.onuncapturederror=a=>{a.error instanceof GPUValidationError&&console.error(`An uncaught WebGPU validation error was raised: ${a.error.message}`)},Object.defineProperty(this.env.webgpu,"device",{value:this.device,writable:!1,enumerable:!0,configurable:!1}),Object.defineProperty(this.env.webgpu,"adapter",{value:t,writable:!1,enumerable:!0,configurable:!1}),this.setQueryType()}dispose(){typeof this.querySet<"u"&&this.querySet.destroy(),this.gpuDataManager.dispose()}getCommandEncoder(){return this.commandEncoder||(this.commandEncoder=this.device.createCommandEncoder()),this.commandEncoder}getComputePassEncoder(){if(!this.computePassEncoder){let e=this.getCommandEncoder(),t={};this.queryType==="at-passes"&&(t.timestampWrites={querySet:this.querySet,beginningOfPassWriteIndex:this.pendingDispatchNumber*2,endOfPassWriteIndex:this.pendingDispatchNumber*2+1}),this.computePassEncoder=e.beginComputePass(t)}return this.computePassEncoder}endComputePass(){this.computePassEncoder&&(this.computePassEncoder.end(),this.computePassEncoder=null)}flush(){if(!this.commandEncoder)return;Ne(),this.endComputePass();let e;this.queryType!=="none"&&(this.commandEncoder.resolveQuerySet(this.querySet,0,this.pendingDispatchNumber*2,this.queryResolveBuffer,0),e=this.device.createBuffer({size:this.pendingDispatchNumber*2*8,usage:GPUBufferUsage.MAP_READ|GPUBufferUsage.COPY_DST}),this.pendingQueries.set(e,this.pendingKernels),this.pendingKernels=[],this.commandEncoder.copyBufferToBuffer(this.queryResolveBuffer,0,e,0,this.pendingDispatchNumber*2*8)),this.device.queue.submit([this.commandEncoder.finish()]),this.gpuDataManager.refreshPendingBuffers(),this.commandEncoder=null,this.pendingDispatchNumber=0,this.queryType!=="none"&&e.mapAsync(GPUMapMode.READ).then(()=>{var n;let t=new BigUint64Array(e.getMappedRange()),s=this.pendingQueries.get(e);for(let i=0;i"u"&&(this.queryTimeBase=z);let V=Number(z-this.queryTimeBase),Z=Number(B-this.queryTimeBase);if(!Number.isSafeInteger(V)||!Number.isSafeInteger(Z))throw new RangeError("incorrect timestamp range");if((n=this.env.webgpu.profiling)!=null&&n.ondata)this.env.webgpu.profiling.ondata({version:1,inputsMetadata:C.map(ee=>({dims:ee.dims,dataType:fr(ee.dataType)})),outputsMetadata:d.map(ee=>({dims:ee.dims,dataType:fr(ee.dataType)})),kernelId:o,kernelType:p,kernelName:h,programName:k,startTime:V,endTime:Z});else{let ee="";C.forEach((he,pe)=>{ee+=`input[${pe}]: [${he.dims}] | ${fr(he.dataType)}, `});let Y="";d.forEach((he,pe)=>{Y+=`output[${pe}]: [${he.dims}] | ${fr(he.dataType)}, `}),console.log(`[profiling] kernel "${o}|${p}|${h}|${k}" ${ee}${Y}execution time: ${Z-V} ns`)}Ue("GPU",`${k}::${z}::${B}`)}e.unmap(),this.pendingQueries.delete(e)}),Re()}run(e,t,s,n,i,a){Ne(e.name);let o=[];for(let Y=0;Yhe):s;if(k.length!==u.length)throw new Error(`Output size ${k.length} must be equal to ${u.length}.`);let C=[],d=[];for(let Y=0;Y=a)throw new Error(`Invalid output index: ${k[Y]}`);if(k[Y]===-3)continue;let he=k[Y]===-1,pe=k[Y]===-2,Me=he||pe?i(u[Y].dataType,u[Y].dims):n(k[Y],u[Y].dataType,u[Y].dims);if(C.push(Me),Me.data===0)continue;let Oe=this.gpuDataManager.get(Me.data);if(!Oe)throw new Error(`no GPU data for output: ${Me.data}`);if(he&&this.temporaryData.push(Oe),pe){let De=this.kernelPersistentData.get(this.currentKernelId);De||(De=[],this.kernelPersistentData.set(this.currentKernelId,De)),De.push(Oe)}d.push(Oe)}if(o.length!==t.length||d.length!==C.length){if(d.length===0)return Re(e.name),C;throw new Error(`Program ${e.name} has zero-sized tensor(s) in inputs or outputs. This is not supported now.`)}let z;if(h){let Y=0,he=[];h.forEach(De=>{let Ye=typeof De.data=="number"?[De.data]:De.data;if(Ye.length===0)return;let at=De.type===10?2:4,Pt,Xt;De.type===10?(Xt=Ye.length>4?16:Ye.length>2?8:Ye.length*at,Pt=Ye.length>4?16:at*Ye.length):(Xt=Ye.length<=2?Ye.length*at:16,Pt=16),Y=Math.ceil(Y/Xt)*Xt,he.push(Y);let Zt=De.type===10?8:4;Y+=Ye.length>4?Math.ceil(Ye.length/Zt)*Pt:Ye.length*at});let pe=16;Y=Math.ceil(Y/pe)*pe;let Me=new ArrayBuffer(Y);h.forEach((De,Ye)=>{let at=he[Ye],Pt=typeof De.data=="number"?[De.data]:De.data;if(De.type===6)new Int32Array(Me,at,Pt.length).set(Pt);else if(De.type===12)new Uint32Array(Me,at,Pt.length).set(Pt);else if(De.type===10)new Uint16Array(Me,at,Pt.length).set(Pt);else if(De.type===1)new Float32Array(Me,at,Pt.length).set(Pt);else throw new Error(`Unsupported uniform type: ${fr(De.type)}`)});let Oe=this.gpuDataManager.create(Y,GPUBufferUsage.COPY_DST|GPUBufferUsage.UNIFORM);this.device.queue.writeBuffer(Oe.buffer,0,Me,0,Y),this.gpuDataManager.release(Oe.id),z={offset:0,size:Y,buffer:Oe.buffer}}let B=this.programManager.normalizeDispatchGroupSize(p),V=B[1]===1&&B[2]===1,Z=cs(e,t,V),ee=this.programManager.getArtifact(Z);if(ee||(ee=this.programManager.build(e,B),this.programManager.setArtifact(Z,ee),as("info",()=>`[artifact] key: ${Z}, programName: ${e.name}`)),h&&ee.uniformVariablesInfo){if(h.length!==ee.uniformVariablesInfo.length)throw new Error(`Uniform variables count mismatch: expect ${ee.uniformVariablesInfo.length}, got ${h.length} in program "${ee.programInfo.name}".`);for(let Y=0;Y`[ProgramManager] run "${e.name}" (key=${Z}) with ${B[0]}x${B[1]}x${B[2]}`),this.queryType!=="none"||this.sessionStatus==="capturing"){let Y={kernelId:this.currentKernelId,programName:ee.programInfo.name,inputTensorViews:t,outputTensorViews:C};this.pendingKernels.push(Y),this.sessionStatus==="capturing"&&this.capturedPendingKernels.get(this.currentSessionId).push(Y)}return this.programManager.run(ee,o,d,B,z),Re(e.name),C}upload(e,t){this.gpuDataManager.upload(e,t)}memcpy(e,t){this.gpuDataManager.memcpy(e,t)}async download(e,t){await this.gpuDataManager.download(e,t)}alloc(e){return this.gpuDataManager.create(e).id}free(e){return this.gpuDataManager.release(e)}createKernel(e,t,s,n){let i=ns.get(e);if(!i)throw new Error(`kernel not implemented: ${e}`);let a={kernelType:e,kernelName:n,kernelEntry:i[0],attributes:[i[1],s]};this.kernels.set(t,a)}releaseKernel(e){let t=this.kernelPersistentData.get(e);if(t){for(let s of t)this.gpuDataManager.release(s.id);this.kernelPersistentData.delete(e)}this.kernelCustomData.delete(e),this.kernels.delete(e)}computeKernel(e,t,s){let n=this.kernels.get(e);if(!n)throw new Error(`kernel not created: ${e}`);let i=n.kernelType,a=n.kernelName,o=n.kernelEntry,u=n.attributes;if(this.currentKernelId!==null)throw new Error(`kernel "[${i}] ${a}" is not allowed to be called recursively`);this.currentKernelId=e,u[0]&&(u[1]=u[0](u[1]),u[0]=void 0),as("info",()=>`[WebGPU] Start to run kernel "[${i}] ${a}"...`);let p=this.env.debug;this.temporaryData=[];try{return p&&this.device.pushErrorScope("validation"),o(t,u[1]),0}catch(h){return s.push(Promise.resolve(`[WebGPU] Kernel "[${i}] ${a}" failed. ${h}`)),1}finally{p&&s.push(this.device.popErrorScope().then(h=>h?`GPU validation error for kernel "[${i}] ${a}": ${h.message}`:null));for(let h of this.temporaryData)this.gpuDataManager.release(h.id);this.temporaryData=[],this.currentKernelId=null}}registerBuffer(e,t,s,n){let i=this.sessionExternalDataMapping.get(e);i||(i=new Map,this.sessionExternalDataMapping.set(e,i));let a=i.get(t),o=this.gpuDataManager.registerExternalBuffer(s,n,a);return i.set(t,[o,s]),o}unregisterBuffers(e){let t=this.sessionExternalDataMapping.get(e);t&&(t.forEach(s=>this.gpuDataManager.unregisterExternalBuffer(s[0])),this.sessionExternalDataMapping.delete(e))}getBuffer(e){let t=this.gpuDataManager.get(e);if(!t)throw new Error(`no GPU data for buffer: ${e}`);return t.buffer}createDownloader(e,t,s){return async()=>{let n=await xt(this,e,t);return P(n.buffer,s)}}writeTimestamp(e){this.queryType==="inside-passes"&&this.computePassEncoder.writeTimestamp(this.querySet,e)}setQueryType(){var e;this.queryType="none",(((e=this.env.webgpu.profiling)==null?void 0:e.mode)==="default"||(typeof this.env.trace>"u"?this.env.wasm.trace:this.env.trace))&&(this.device.features.has("chromium-experimental-timestamp-query-inside-passes")?this.queryType="inside-passes":this.device.features.has("timestamp-query")&&(this.queryType="at-passes"),this.queryType!=="none"&&typeof this.querySet>"u"&&(this.querySet=this.device.createQuerySet({type:"timestamp",count:this.maxDispatchNumber*2}),this.queryResolveBuffer=this.device.createBuffer({size:this.maxDispatchNumber*2*8,usage:GPUBufferUsage.COPY_SRC|GPUBufferUsage.QUERY_RESOLVE})))}captureBegin(){as("info","captureBegin"),this.capturedCommandList.get(this.currentSessionId)||this.capturedCommandList.set(this.currentSessionId,[]),this.capturedPendingKernels.get(this.currentSessionId)||this.capturedPendingKernels.set(this.currentSessionId,[]),this.flush(),this.sessionStatus="capturing"}captureEnd(){as("info","captureEnd"),this.flush(),this.sessionStatus="default"}replay(){as("info","replay"),this.sessionStatus="replaying";let e=this.capturedCommandList.get(this.currentSessionId),t=this.capturedPendingKernels.get(this.currentSessionId),s=e.length;this.pendingKernels=[];for(let n=0;n=this.maxDispatchNumber||this.queryType==="at-passes")&&this.endComputePass(),this.pendingDispatchNumber>=this.maxDispatchNumber&&this.flush()}this.flush(),this.sessionStatus="default"}onCreateSession(){this.gpuDataManager.onCreateSession()}onReleaseSession(e){this.unregisterBuffers(e),this.capturedCommandList.has(e)&&this.capturedCommandList.delete(e),this.capturedPendingKernels.has(e)&&this.capturedPendingKernels.delete(e),this.gpuDataManager.onReleaseSession(e)}onRunStart(e){this.currentSessionId=e,this.setQueryType()}}}),br,pn,Sa,or,vr,Ai,Ii,Oi,$a=g(()=>{Pe(),br=1,pn=()=>br++,Sa=new Map([["float32",32],["float16",16],["int32",32],["uint32",32],["int64",64],["uint64",64],["int8",8],["uint8",8],["int4",4],["uint4",4]]),or=(e,t)=>{let s=Sa.get(e);if(!s)throw new Error("Unsupported data type.");return t.length>0?Math.ceil(t.reduce((n,i)=>n*i)*s/8):0},vr=class{constructor(e){this.sessionId=e.sessionId,this.mlContext=e.context,this.mlTensor=e.tensor,this.dataType=e.dataType,this.tensorShape=e.shape}get tensor(){return this.mlTensor}get type(){return this.dataType}get shape(){return this.tensorShape}get byteLength(){return or(this.dataType,this.tensorShape)}destroy(){as("verbose",()=>"[WebNN] TensorWrapper.destroy"),this.mlTensor.destroy()}write(e){this.mlContext.writeTensor(this.mlTensor,e)}async read(e){return e?this.mlContext.readTensor(this.mlTensor,e):this.mlContext.readTensor(this.mlTensor)}canReuseTensor(e,t,s){return this.mlContext===e&&this.dataType===t&&this.tensorShape.length===s.length&&this.tensorShape.every((n,i)=>n===s[i])}},Ai=class{constructor(e,t){this.tensorManager=e,this.wrapper=t}get tensorWrapper(){return this.wrapper}releaseTensor(){this.tensorWrapper&&(this.tensorManager.releaseTensor(this.tensorWrapper),this.wrapper=void 0)}async ensureTensor(e,t,s,n){if(this.wrapper){if(this.wrapper.canReuseTensor(e,t,s))return this.wrapper.tensor;if(n){if(this.wrapper.byteLength!==or(t,s))throw new Error("Unable to copy data to tensor with different size.");this.activeUpload=new Uint8Array(await this.wrapper.read())}this.tensorManager.releaseTensor(this.wrapper)}let i=typeof MLTensorUsage>"u"?void 0:MLTensorUsage.READ|MLTensorUsage.WRITE;return this.wrapper=await this.tensorManager.getCachedTensor(t,s,i,!0,!0),n&&this.activeUpload&&(this.wrapper.write(this.activeUpload),this.activeUpload=void 0),this.wrapper.tensor}upload(e){if(this.wrapper)if(e.byteLength===this.wrapper.byteLength){this.wrapper.write(e);return}else as("verbose",()=>"Data size does not match tensor size. Releasing tensor."),this.releaseTensor();this.activeUpload?this.activeUpload.set(e):this.activeUpload=new Uint8Array(e)}async download(e){if(this.activeUpload)if(e){e instanceof ArrayBuffer?new Uint8Array(e).set(this.activeUpload):new Uint8Array(e.buffer,e.byteOffset,e.byteLength).set(this.activeUpload);return}else return this.activeUpload.buffer;if(!this.wrapper)throw new Error("Tensor has not been created.");return e?this.wrapper.read(e):this.wrapper.read()}},Ii=class{constructor(e){this.backend=e,this.tensorTrackersById=new Map,this.freeTensors=[],this.externalTensors=new Set}reserveTensorId(){let e=pn();return this.tensorTrackersById.set(e,new Ai(this)),e}releaseTensorId(e){let t=this.tensorTrackersById.get(e);t&&(this.tensorTrackersById.delete(e),t.tensorWrapper&&this.releaseTensor(t.tensorWrapper))}async ensureTensor(e,t,s,n){as("verbose",()=>`[WebNN] TensorManager.ensureTensor {tensorId: ${e}, dataType: ${t}, shape: ${s}, copyOld: ${n}}`);let i=this.tensorTrackersById.get(e);if(!i)throw new Error("Tensor not found.");return i.ensureTensor(this.backend.currentContext,t,s,n)}upload(e,t){let s=this.tensorTrackersById.get(e);if(!s)throw new Error("Tensor not found.");s.upload(t)}async download(e,t){as("verbose",()=>`[WebNN] TensorManager.download {tensorId: ${e}, dstBuffer: ${t==null?void 0:t.byteLength}}`);let s=this.tensorTrackersById.get(e);if(!s)throw new Error("Tensor not found.");return s.download(t)}releaseTensorsForSession(e){for(let t of this.freeTensors)t.sessionId===e&&t.destroy();this.freeTensors=this.freeTensors.filter(t=>t.sessionId!==e)}registerTensor(e,t,s,n){let i=pn(),a=new vr({sessionId:this.backend.currentSessionId,context:e,tensor:t,dataType:s,shape:n});return this.tensorTrackersById.set(i,new Ai(this,a)),this.externalTensors.add(a),i}async getCachedTensor(e,t,s,n,i){let a=this.backend.currentSessionId,o=this.backend.currentContext;for(let[p,h]of this.freeTensors.entries())if(h.canReuseTensor(o,e,t)){as("verbose",()=>`[WebNN] Reusing tensor {dataType: ${e}, shape: ${t}}`);let k=this.freeTensors.splice(p,1)[0];return k.sessionId=a,k}as("verbose",()=>`[WebNN] MLContext.createTensor {dataType: ${e}, shape: ${t}}`);let u=await o.createTensor({dataType:e,shape:t,dimensions:t,usage:s,writable:n,readable:i});return new vr({sessionId:a,context:o,tensor:u,dataType:e,shape:t})}releaseTensor(e){this.externalTensors.has(e)&&this.externalTensors.delete(e),this.freeTensors.push(e)}},Oi=(...e)=>new Ii(...e)}),Ts,Rs,qr,En=g(()=>{zt(),lr(),Q(),$a(),Pe(),Ts=new Map([[1,"float32"],[10,"float16"],[6,"int32"],[12,"uint32"],[7,"int64"],[13,"uint64"],[22,"int4"],[21,"uint4"],[3,"int8"],[2,"uint8"],[9,"uint8"]]),Rs=(e,t)=>{if(e===t)return!0;if(e===void 0||t===void 0)return!1;let s=Object.keys(e).sort(),n=Object.keys(t).sort();return s.length===n.length&&s.every((i,a)=>i===n[a]&&e[i]===t[i])},qr=class{constructor(e){this.tensorManager=Oi(this),this.mlContextBySessionId=new Map,this.sessionIdsByMLContext=new Map,this.mlContextCache=[],Tn(e.logLevel,!!e.debug)}get currentSessionId(){if(this.activeSessionId===void 0)throw new Error("No active session");return this.activeSessionId}onRunStart(e){this.activeSessionId=e}async createMLContext(e){if(e instanceof GPUDevice){let s=this.mlContextCache.findIndex(n=>n.gpuDevice===e);if(s!==-1)return this.mlContextCache[s].mlContext;{let n=await navigator.ml.createContext(e);return this.mlContextCache.push({gpuDevice:e,mlContext:n}),n}}else if(e===void 0){let s=this.mlContextCache.findIndex(n=>n.options===void 0&&n.gpuDevice===void 0);if(s!==-1)return this.mlContextCache[s].mlContext;{let n=await navigator.ml.createContext();return this.mlContextCache.push({mlContext:n}),n}}let t=this.mlContextCache.findIndex(s=>Rs(s.options,e));if(t!==-1)return this.mlContextCache[t].mlContext;{let s=await navigator.ml.createContext(e);return this.mlContextCache.push({options:e,mlContext:s}),s}}get currentContext(){let e=this.getMLContext(this.currentSessionId);if(!e)throw new Error(`No MLContext found for session ${this.currentSessionId}`);return e}registerMLContext(e,t){this.mlContextBySessionId.set(e,t);let s=this.sessionIdsByMLContext.get(t);s||(s=new Set,this.sessionIdsByMLContext.set(t,s)),s.add(e)}onReleaseSession(e){let t=this.mlContextBySessionId.get(e);if(!t)return;this.tensorManager.releaseTensorsForSession(e),this.mlContextBySessionId.delete(e);let s=this.sessionIdsByMLContext.get(t);if(s.delete(e),s.size===0){this.sessionIdsByMLContext.delete(t);let n=this.mlContextCache.findIndex(i=>i.mlContext===t);n!==-1&&this.mlContextCache.splice(n,1)}}getMLContext(e){return this.mlContextBySessionId.get(e)}reserveTensorId(){return this.tensorManager.reserveTensorId()}releaseTensorId(e){as("verbose",()=>`[WebNN] releaseTensorId {tensorId: ${e}}`),this.tensorManager.releaseTensorId(e)}async ensureTensor(e,t,s,n){let i=Ts.get(t);if(!i)throw new Error(`Unsupported ONNX data type: ${t}`);return this.tensorManager.ensureTensor(e,i,s,n)}uploadTensor(e,t){if(!Ms().shouldTransferToMLTensor)throw new Error("Trying to upload to a MLTensor while shouldTransferToMLTensor is false");as("verbose",()=>`[WebNN] uploadTensor {tensorId: ${e}, data: ${t.byteLength}}`),this.tensorManager.upload(e,t)}async downloadTensor(e,t){return this.tensorManager.download(e,t)}createMLTensorDownloader(e,t){return async()=>{let s=await this.tensorManager.download(e);return P(s,t)}}registerMLTensor(e,t,s){let n=Ts.get(t);if(!n)throw new Error(`Unsupported ONNX data type: ${t}`);let i=this.tensorManager.registerTensor(this.currentContext,e,n,s);return as("verbose",()=>`[WebNN] registerMLTensor {tensor: ${e}, dataType: ${n}, dimensions: ${s}} -> {tensorId: ${i}}`),i}registerMLConstant(e,t,s,n,i,a){if(!a)throw new Error("External mounted files are not available.");let o=e;e.startsWith("./")&&(o=e.substring(2));let u=a.get(o);if(!u)throw new Error(`File with name ${o} not found in preloaded files.`);if(t+s>u.byteLength)throw new Error("Out of bounds: data offset and length exceed the external file data size.");let p=u.slice(t,t+s).buffer,h;switch(i.dataType){case"float32":h=new Float32Array(p);break;case"float16":h=new Uint16Array(p);break;case"int32":h=new Int32Array(p);break;case"uint32":h=new Uint32Array(p);break;case"int64":h=new BigInt64Array(p);break;case"uint64":h=new BigUint64Array(p);break;case"int8":h=new Int8Array(p);break;case"int4":case"uint4":case"uint8":h=new Uint8Array(p);break;default:throw new Error(`Unsupported data type: ${i.dataType} in creating WebNN Constant from external data.`)}return as("verbose",()=>`[WebNN] registerMLConstant {dataType: ${i.dataType}, shape: ${i.shape}}}`),n.constant(i,h)}flush(){}}}),Cn={};v(Cn,{init:()=>Ia});var kn,Aa,Ia,Cc=g(()=>{zt(),Ys(),Pe(),Ot(),En(),kn=class sf{constructor(t,s,n,i){this.module=t,this.dataType=s,this.data=n,this.dims=i}getFloat32Array(){if(this.dataType!==1)throw new Error("Invalid data type");let t=ze.size(this.dims);return t===0?new Float32Array:new Float32Array(this.module.HEAP8.buffer,this.data,t)}getBigInt64Array(){if(this.dataType!==7)throw new Error("Invalid data type");let t=ze.size(this.dims);return t===0?new BigInt64Array:new BigInt64Array(this.module.HEAP8.buffer,this.data,t)}getInt32Array(){if(this.dataType!==6)throw new Error("Invalid data type");let t=ze.size(this.dims);return t===0?new Int32Array:new Int32Array(this.module.HEAP8.buffer,this.data,t)}getUint16Array(){if(this.dataType!==10&&this.dataType!==4)throw new Error("Invalid data type");let t=ze.size(this.dims);return t===0?new Uint16Array:new Uint16Array(this.module.HEAP8.buffer,this.data,t)}reshape(t){if(ze.size(t)!==ze.size(this.dims))throw new Error("Invalid new shape");return new sf(this.module,this.dataType,this.data,t)}},Aa=class{constructor(e,t,s){this.module=e,this.backend=t,this.customDataOffset=0,this.customDataSize=0,this.adapterInfo=t.adapterInfo,this.deviceInfo=t.deviceInfo;let n=e.PTR_SIZE,i=s/e.PTR_SIZE,a=n===4?"i32":"i64";this.opKernelContext=Number(e.getValue(n*i++,a));let o=Number(e.getValue(n*i++,a));this.outputCount=Number(e.getValue(n*i++,a)),this.customDataOffset=Number(e.getValue(n*i++,"*")),this.customDataSize=Number(e.getValue(n*i++,a));let u=[];for(let p=0;ptypeof u=="number"?this.inputs[u]:u))??this.inputs,n=(t==null?void 0:t.outputs)??[],i=(u,p,h)=>new kn(this.module,p,this.output(u,h),h),a=(u,p)=>{let h=ur(u,p);if(!h)throw new Error(`Unsupported data type: ${u}`);let k=h>0?this.backend.gpuDataManager.create(h).id:0;return new kn(this.module,u,k,p)};return this.backend.run(e,s,n,i,a,this.outputCount)}output(e,t){let s=this.module.stackSave();try{let n=this.module.PTR_SIZE,i=n===4?"i32":"i64",a=this.module.stackAlloc((1+t.length)*n);this.module.setValue(a,t.length,i);for(let o=0;o{let i=t.jsepInit;if(!i)throw new Error("Failed to initialize JSEP. The WebAssembly module is not built with JSEP support.");if(e==="webgpu"){let a=new Zs;await a.initialize(s,n),i("webgpu",[a,o=>a.alloc(Number(o)),o=>a.free(o),(o,u,p,h=!1)=>{if(h)as("verbose",()=>`[WebGPU] jsepCopyGpuToGpu: src=${Number(o)}, dst=${Number(u)}, size=${Number(p)}`),a.memcpy(Number(o),Number(u));else{as("verbose",()=>`[WebGPU] jsepCopyCpuToGpu: dataOffset=${Number(o)}, gpuDataId=${Number(u)}, size=${Number(p)}`);let k=t.HEAPU8.subarray(Number(o>>>0),Number(o>>>0)+Number(p));a.upload(Number(u),k)}},async(o,u,p)=>{as("verbose",()=>`[WebGPU] jsepCopyGpuToCpu: gpuDataId=${o}, dataOffset=${u}, size=${p}`),await a.download(Number(o),()=>t.HEAPU8.subarray(Number(u)>>>0,Number(u+p)>>>0))},(o,u,p)=>a.createKernel(o,Number(u),p,t.UTF8ToString(t._JsepGetNodeName(Number(u)))),o=>a.releaseKernel(o),(o,u,p,h)=>{as("verbose",()=>`[WebGPU] jsepRun: sessionHandle=${p}, kernel=${o}, contextDataOffset=${u}`);let k=new Aa(t,a,Number(u));return a.computeKernel(Number(o),k,h)},()=>a.captureBegin(),()=>a.captureEnd(),()=>a.replay()])}else{let a=new qr(s);i("webnn",[a,()=>a.reserveTensorId(),o=>a.releaseTensorId(o),async(o,u,p,h)=>a.ensureTensor(o,u,p,h),(o,u)=>{a.uploadTensor(o,u)},async(o,u)=>a.downloadTensor(o,u)])}}}),kp,Fi,qn,zr,kc,Qn,Xn,Sc,$c,Ac,Qr,mr,vh=g(()=>{ii(),oi(),zt(),lr(),wn(),Dn(),kp=(e,t)=>{Ms()._OrtInit(e,t)!==0&&es("Can't initialize onnxruntime.")},Fi=async e=>{kp(e.wasm.numThreads,tn(e.logLevel))},qn=async(e,t)=>{{let s=(Cc(),y(Cn)).init;if(t==="webgpu"){if(typeof navigator>"u"||!navigator.gpu)throw new Error("WebGPU is not supported in current environment");let n=e.webgpu.adapter;if(n){if(typeof n.limits!="object"||typeof n.features!="object"||typeof n.requestDevice!="function")throw new Error("Invalid GPU adapter set in `env.webgpu.adapter`. It must be a GPUAdapter object.")}else{let i=e.webgpu.powerPreference;if(i!==void 0&&i!=="low-power"&&i!=="high-performance")throw new Error(`Invalid powerPreference setting: "${i}"`);let a=e.webgpu.forceFallbackAdapter;if(a!==void 0&&typeof a!="boolean")throw new Error(`Invalid forceFallbackAdapter setting: "${a}"`);if(n=await navigator.gpu.requestAdapter({powerPreference:i,forceFallbackAdapter:a}),!n)throw new Error('Failed to get GPU adapter. You may need to enable flag "--enable-unsafe-webgpu" if you are using Chrome.')}await s("webgpu",Ms(),e,n)}if(t==="webnn"){if(typeof navigator>"u"||!navigator.ml)throw new Error("WebNN is not supported in current environment");await s("webnn",Ms(),e)}}},zr=new Map,kc=e=>{let t=Ms(),s=t.stackSave();try{let n=t.PTR_SIZE,i=t.stackAlloc(2*n);t._OrtGetInputOutputCount(e,i,i+n)!==0&&es("Can't get session input/output count.");let a=n===4?"i32":"i64";return[Number(t.getValue(i,a)),Number(t.getValue(i+n,a))]}finally{t.stackRestore(s)}},Qn=e=>{let t=Ms(),s=t._malloc(e.byteLength);if(s===0)throw new Error(`Can't create a session. failed to allocate a buffer of size ${e.byteLength}.`);return t.HEAPU8.set(e,s),[s,e.byteLength]},Xn=async(e,t)=>{var C,d,z;let s,n,i=Ms();Array.isArray(e)?[s,n]=e:e.buffer===i.HEAPU8.buffer?[s,n]=[e.byteOffset,e.byteLength]:[s,n]=Qn(e);let a=0,o=0,u=0,p=[],h=[],k=[];try{if([o,p]=Fn(t),(t==null?void 0:t.externalData)&&i.mountExternalData){let Me=[];for(let Oe of t.externalData){let De=typeof Oe=="string"?Oe:Oe.path;Me.push(xn(typeof Oe=="string"?Oe:Oe.data).then(Ye=>{i.mountExternalData(De,Ye)}))}await Promise.all(Me)}for(let Me of(t==null?void 0:t.executionProviders)??[])if((typeof Me=="string"?Me:Me.name)==="webnn"){if(i.shouldTransferToMLTensor=!1,typeof Me!="string"){let Oe=Me,De=Oe==null?void 0:Oe.context,Ye=Oe==null?void 0:Oe.gpuDevice,at=Oe==null?void 0:Oe.deviceType,Pt=Oe==null?void 0:Oe.powerPreference;De?i.currentContext=De:Ye?i.currentContext=await i.jsepCreateMLContext(Ye):i.currentContext=await i.jsepCreateMLContext({deviceType:at,powerPreference:Pt})}else i.currentContext=await i.jsepCreateMLContext();break}a=await i._OrtCreateSession(s,n,o),a===0&&es("Can't create a session."),(C=i.jsepOnCreateSession)==null||C.call(i),i.currentContext&&(i.jsepRegisterMLContext(a,i.currentContext),i.currentContext=void 0,i.shouldTransferToMLTensor=!0);let[B,V]=kc(a),Z=!!(t!=null&&t.enableGraphCapture),ee=[],Y=[],he=[];for(let Me=0;MeMe==="gpu-buffer"||Me==="ml-tensor")&&(u=i._OrtCreateBinding(a),u===0&&es("Can't create IO binding."),pe={handle:u,outputPreferredLocations:he,outputPreferredLocationsEncoded:he.map(Me=>vn(Me))}),zr.set(a,[a,h,k,pe,Z,!1]),[a,ee,Y]}catch(B){throw h.forEach(V=>i._OrtFree(V)),k.forEach(V=>i._OrtFree(V)),u!==0&&i._OrtReleaseBinding(u)!==0&&es("Can't release IO binding."),a!==0&&i._OrtReleaseSession(a)!==0&&es("Can't release session."),B}finally{i._free(s),o!==0&&i._OrtReleaseSessionOptions(o)!==0&&es("Can't release session options."),p.forEach(B=>i._free(B)),(z=i.unmountExternalData)==null||z.call(i)}},Sc=e=>{var p;let t=Ms(),s=zr.get(e);if(!s)throw new Error(`cannot release session. invalid session id: ${e}`);let[n,i,a,o,u]=s;o&&(u&&t._OrtClearBoundOutputs(o.handle)!==0&&es("Can't clear bound outputs."),t._OrtReleaseBinding(o.handle)!==0&&es("Can't release IO binding.")),(p=t.jsepOnReleaseSession)==null||p.call(t,e),i.forEach(h=>t._OrtFree(h)),a.forEach(h=>t._OrtFree(h)),t._OrtReleaseSession(n)!==0&&es("Can't release session."),zr.delete(e)},$c=(e,t,s,n,i,a=!1)=>{if(!e){t.push(0);return}let o=Ms(),u=o.PTR_SIZE,p=e[0],h=e[1],k=e[3],C,d;if(p==="string"&&(k==="gpu-buffer"||k==="ml-tensor"))throw new Error("String tensor is not supported on GPU.");if(a&&k!=="gpu-buffer")throw new Error(`External buffer must be provided for input/output index ${i} when enableGraphCapture is true.`);if(k==="gpu-buffer"){let V=e[2].gpuBuffer;d=ur(Wr(p),h);let Z=o.jsepRegisterBuffer;if(!Z)throw new Error('Tensor location "gpu-buffer" is not supported without using WebGPU.');C=Z(n,i,V,d)}else if(k==="ml-tensor"){let V=e[2].mlTensor;d=ur(Wr(p),h);let Z=o.jsepRegisterMLTensor;if(!Z)throw new Error('Tensor location "ml-tensor" is not supported without using WebNN.');C=Z(V,Wr(p),h)}else{let V=e[2];if(Array.isArray(V)){d=u*V.length,C=o._malloc(d),s.push(C);for(let Z=0;Zo.setValue(B+ee*u,Z,u===4?"i32":"i64"));let V=o._OrtCreateTensor(Wr(p),C,d,B,h.length,vn(k));V===0&&es(`Can't create tensor for input/output. session=${n}, index=${i}.`),t.push(V)}finally{o.stackRestore(z)}},Ac=async(e,t,s,n,i,a)=>{var Xt,Zt;let o=Ms(),u=o.PTR_SIZE,p=zr.get(e);if(!p)throw new Error(`cannot run inference. invalid session id: ${e}`);let h=p[0],k=p[1],C=p[2],d=p[3],z=p[4],B=p[5],V=t.length,Z=n.length,ee=0,Y=[],he=[],pe=[],Me=[],Oe=o.stackSave(),De=o.stackAlloc(V*u),Ye=o.stackAlloc(V*u),at=o.stackAlloc(Z*u),Pt=o.stackAlloc(Z*u);try{(Xt=o.jsepOnRunStart)==null||Xt.call(o,h),[ee,Y]=Ur(a);for(let St=0;Stzs*Ps,1);_s=fr(ps);let zi=d==null?void 0:d.outputPreferredLocations[n[St]];if(_s==="string"){if(zi==="gpu-buffer"||zi==="ml-tensor")throw new Error("String tensor is not supported on GPU.");let zs=[];for(let Ps=0;Ps0){let zs=o.jsepGetBuffer;if(!zs)throw new Error('preferredLocation "gpu-buffer" is not supported without using WebGPU.');let Ps=zs(ot),Jn=ur(ps,hn);if(Jn===void 0||!Mn(_s))throw new Error(`Unsupported data type: ${_s}`);Rt=!0,ss.push([_s,er,{gpuBuffer:Ps,download:o.jsepCreateDownloader(Ps,Jn,_s),dispose:()=>{o._OrtReleaseTensor(Ft)!==0&&es("Can't release tensor.")}},"gpu-buffer"])}else if(zi==="ml-tensor"&&hn>0){let zs=o.jsepEnsureTensor;if(!zs)throw new Error('preferredLocation "ml-tensor" is not supported without using WebNN.');if(ur(ps,hn)===void 0||!bn(_s))throw new Error(`Unsupported data type: ${_s}`);let Ps=await zs(ot,ps,er,!1);Rt=!0,ss.push([_s,er,{mlTensor:Ps,download:o.jsepCreateMLTensorDownloader(ot,_s),dispose:()=>{o.jsepReleaseTensorId(ot),o._OrtReleaseTensor(Ft)}},"ml-tensor"])}else{let zs=yn(_s),Ps=new zs(hn);new Uint8Array(Ps.buffer,Ps.byteOffset,Ps.byteLength).set(o.HEAPU8.subarray(ot,ot+Ps.byteLength)),ss.push([_s,er,Ps,"cpu"])}}finally{o.stackRestore(bs),_s==="string"&&ot&&o._free(ot),Rt||o._OrtReleaseTensor(Ft)}}return d&&!z&&(o._OrtClearBoundOutputs(d.handle)!==0&&es("Can't clear bound outputs."),zr.set(e,[h,k,C,d,z,!1])),ss}finally{o.stackRestore(Oe),he.forEach(bt=>o._OrtReleaseTensor(bt)),pe.forEach(bt=>o._OrtReleaseTensor(bt)),Me.forEach(bt=>o._free(bt)),ee!==0&&o._OrtReleaseRunOptions(ee),Y.forEach(bt=>o._free(bt))}},Qr=e=>{let t=Ms(),s=zr.get(e);if(!s)throw new Error("invalid session id");let n=s[0],i=t._OrtEndProfiling(n);i===0&&es("Can't get an profile file name."),t._OrtFree(i)},mr=e=>{let t=[];for(let s of e){let n=s[2];!Array.isArray(n)&&"buffer"in n&&t.push(n.buffer)}return t}}),Yn,xr,Oa,Ic,Oc,Sp,Qp,$p,Di,Li,xh,Th,Ph,Eh,Ch,kh,Sh,$h,Ah=g(()=>{We(),vh(),lr(),Nr(),Yn=()=>!!O.wasm.proxy&&typeof document<"u",Oa=!1,Ic=!1,Oc=!1,$p=new Map,Di=(e,t)=>{let s=$p.get(e);s?s.push(t):$p.set(e,[t])},Li=()=>{if(Oa||!Ic||Oc||!xr)throw new Error("worker not ready")},xh=e=>{switch(e.data.type){case"init-wasm":Oa=!1,e.data.err?(Oc=!0,Qp[1](e.data.err)):(Ic=!0,Qp[0]()),Sp&&(URL.revokeObjectURL(Sp),Sp=void 0);break;case"init-ep":case"copy-from":case"create":case"release":case"run":case"end-profiling":{let t=$p.get(e.data.type);e.data.err?t.shift()[1](e.data.err):t.shift()[0](e.data.out);break}}},Th=async()=>{if(!Ic){if(Oa)throw new Error("multiple calls to 'initWasm()' detected.");if(Oc)throw new Error("previous call to 'initWasm()' failed.");if(Oa=!0,Yn())return new Promise((e,t)=>{xr==null||xr.terminate(),en().then(([s,n])=>{var i;try{xr=n,xr.onerror=o=>t(o),xr.onmessage=xh,Qp=[e,t];let a={type:"init-wasm",in:O};!a.in.wasm.wasmPaths&&(s||(i=self.location.href)!=null&&i.startsWith("file:"))&&(a.in.wasm.wasmPaths={wasm:new URL(r("./node_modules/onnxruntime-web/dist/ort-wasm-simd-threaded.jsep.wasm"),r.b).href}),xr.postMessage(a),Sp=s}catch(a){t(a)}},t)});try{await Or(O.wasm),await Fi(O),Ic=!0}catch(e){throw Oc=!0,e}finally{Oa=!1}}},Ph=async e=>{if(Yn())return Li(),new Promise((t,s)=>{Di("init-ep",[t,s]);let n={type:"init-ep",in:{epName:e,env:O}};xr.postMessage(n)});await qn(O,e)},Eh=async e=>Yn()?(Li(),new Promise((t,s)=>{Di("copy-from",[t,s]);let n={type:"copy-from",in:{buffer:e}};xr.postMessage(n,[e.buffer])})):Qn(e),Ch=async(e,t)=>{if(Yn()){if(t!=null&&t.preferredOutputLocation)throw new Error('session option "preferredOutputLocation" is not supported for proxy.');return Li(),new Promise((s,n)=>{Di("create",[s,n]);let i={type:"create",in:{model:e,options:{...t}}},a=[];e instanceof Uint8Array&&a.push(e.buffer),xr.postMessage(i,a)})}else return Xn(e,t)},kh=async e=>{if(Yn())return Li(),new Promise((t,s)=>{Di("release",[t,s]);let n={type:"release",in:e};xr.postMessage(n)});Sc(e)},Sh=async(e,t,s,n,i,a)=>{if(Yn()){if(s.some(o=>o[3]!=="cpu"))throw new Error("input tensor on GPU is not supported for proxy.");if(i.some(o=>o))throw new Error("pre-allocated output tensor is not supported for proxy.");return Li(),new Promise((o,u)=>{Di("run",[o,u]);let p=s,h={type:"run",in:{sessionId:e,inputIndices:t,inputs:p,outputIndices:n,options:a}};xr.postMessage(h,mr(p))})}else return Ac(e,t,s,n,i,a)},$h=async e=>{if(Yn())return Li(),new Promise((t,s)=>{Di("end-profiling",[t,s]);let n={type:"end-profiling",in:e};xr.postMessage(n)});Qr(e)}}),Xp,Ih,Oh,pf=g(()=>{We(),Ah(),zt(),Je(),Dn(),Xp=(e,t)=>{switch(e.location){case"cpu":return[e.type,e.dims,e.data,"cpu"];case"gpu-buffer":return[e.type,e.dims,{gpuBuffer:e.gpuBuffer},"gpu-buffer"];case"ml-tensor":return[e.type,e.dims,{mlTensor:e.mlTensor},"ml-tensor"];default:throw new Error(`invalid data location: ${e.location} for ${t()}`)}},Ih=e=>{switch(e[3]){case"cpu":return new le(e[0],e[2],e[1]);case"gpu-buffer":{let t=e[0];if(!Mn(t))throw new Error(`not supported data type: ${t} for deserializing GPU tensor`);let{gpuBuffer:s,download:n,dispose:i}=e[2];return le.fromGpuBuffer(s,{dataType:t,dims:e[1],download:n,dispose:i})}case"ml-tensor":{let t=e[0];if(!bn(t))throw new Error(`not supported data type: ${t} for deserializing MLTensor tensor`);let{mlTensor:s,download:n,dispose:i}=e[2];return le.fromMLTensor(s,{dataType:t,dims:e[1],download:n,dispose:i})}default:throw new Error(`invalid data location: ${e[3]}`)}},Oh=class{async fetchModelAndCopyToWasmMemory(e){return Eh(await xn(e))}async loadModel(e,t){Ne();let s;typeof e=="string"?s=await this.fetchModelAndCopyToWasmMemory(e):s=e,[this.sessionId,this.inputNames,this.outputNames]=await Ch(s,t),Re()}async dispose(){return kh(this.sessionId)}async run(e,t,s){Ne();let n=[],i=[];Object.entries(e).forEach(C=>{let d=C[0],z=C[1],B=this.inputNames.indexOf(d);if(B===-1)throw new Error(`invalid input '${d}'`);n.push(z),i.push(B)});let a=[],o=[];Object.entries(t).forEach(C=>{let d=C[0],z=C[1],B=this.outputNames.indexOf(d);if(B===-1)throw new Error(`invalid output '${d}'`);a.push(z),o.push(B)});let u=n.map((C,d)=>Xp(C,()=>`input "${this.inputNames[i[d]]}"`)),p=a.map((C,d)=>C?Xp(C,()=>`output "${this.outputNames[o[d]]}"`):null),h=await Sh(this.sessionId,i,u,o,p,s),k={};for(let C=0;CJp,initializeFlags:()=>Yp,wasmBackend:()=>Dh});var Yp,Jp,Dh,hf=g(()=>{We(),Ah(),pf(),Yp=()=>{if((typeof O.wasm.initTimeout!="number"||O.wasm.initTimeout<0)&&(O.wasm.initTimeout=0),O.wasm.simd===!1&&console.warn('Deprecated property "env.wasm.simd" is set to false. non-SIMD build is no longer provided, and this setting will be ignored.'),typeof O.wasm.proxy!="boolean"&&(O.wasm.proxy=!1),typeof O.wasm.trace!="boolean"&&(O.wasm.trace=!1),typeof O.wasm.numThreads!="number"||!Number.isInteger(O.wasm.numThreads)||O.wasm.numThreads<=0)if(typeof self<"u"&&!self.crossOriginIsolated)O.wasm.numThreads=1;else{let e=typeof navigator>"u"?R("node:os").cpus().length:navigator.hardwareConcurrency;O.wasm.numThreads=Math.min(4,Math.ceil((e||1)/2))}},Jp=class{async init(e){Yp(),await Th(),await Ph(e)}async createInferenceSessionHandler(e,t){let s=new Oh;return await s.loadModel(e,t),Promise.resolve(s)}},Dh=new Jp});We(),We(),We();var mf="1.21.0-dev.20250114-228dd16893",ff=$e;{let e=(hf(),y(Fh)).wasmBackend;K("webgpu",e,5),K("webnn",e,5),K("cpu",e,10),K("wasm",e,10)}Object.defineProperty(O.versions,"web",{value:mf,enumerable:!0});/** + * @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 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 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. + * ============================================================================= + */},"./src/backends/onnx.js":(Le,A,r)=>{var _;r.r(A),r.d(A,{Tensor:()=>R.Tensor,createInferenceSession:()=>ne,deviceToExecutionProviders:()=>K,isONNXProxy:()=>q,isONNXTensor:()=>W});var D=r("./src/env.js"),j=r("?2ce3"),X=r("./node_modules/onnxruntime-web/dist/ort.bundle.min.mjs?3a96"),R=r("./node_modules/onnxruntime-common/dist/esm/index.js");const g=Object.freeze({auto:null,gpu:null,cpu:"cpu",wasm:"wasm",webgpu:"webgpu",cuda:"cuda",dml:"dml",webnn:{name:"webnn",deviceType:"cpu"},"webnn-npu":{name:"webnn",deviceType:"npu"},"webnn-gpu":{name:"webnn",deviceType:"gpu"},"webnn-cpu":{name:"webnn",deviceType:"cpu"}}),v=[];let M,y;const b=Symbol.for("onnxruntime");if(b in globalThis)y=globalThis[b];else if(D.apis.IS_NODE_ENV){switch(y=j??(_||(_=r.t(j,2))),process.platform){case"win32":v.push("dml");break;case"linux":process.arch==="x64"&&v.push("cuda");break}v.push("cpu"),M=["cpu"]}else y=X,D.apis.IS_WEBNN_AVAILABLE&&v.push("webnn-npu","webnn-gpu","webnn-cpu","webnn"),D.apis.IS_WEBGPU_AVAILABLE&&v.push("webgpu"),v.push("wasm"),M=["wasm"];const I=y.InferenceSession;function K($=null){if(!$)return M;switch($){case"auto":return v;case"gpu":return v.filter(S=>["webgpu","cuda","dml","webnn-gpu"].includes(S))}if(v.includes($))return[g[$]??$];throw new Error(`Unsupported device: "${$}". Should be one of: ${v.join(", ")}.`)}let se=null;async function ne($,S,w){se&&await se;const x=I.create($,S);se??(se=x);const O=await x;return O.config=w,O}function W($){return $ instanceof y.Tensor}const U=y==null?void 0:y.env;U!=null&&U.wasm&&(U.wasm.wasmPaths=`https://cdn.jsdelivr.net/npm/@huggingface/transformers@${D.env.version}/dist/`,U.wasm.proxy=!1,(typeof crossOriginIsolated>"u"||!crossOriginIsolated)&&(U.wasm.numThreads=1)),U!=null&&U.webgpu&&(U.webgpu.powerPreference="high-performance");function q(){var $;return($=U==null?void 0:U.wasm)==null?void 0:$.proxy}D.env.backends.onnx=U},"./src/base/feature_extraction_utils.js":(Le,A,r)=>{r.r(A),r.d(A,{FeatureExtractor:()=>X,validate_audio_inputs:()=>R});var _=r("./src/utils/constants.js"),D=r("./src/utils/generic.js"),j=r("./src/utils/hub.js");class X extends D.Callable{constructor(v){super(),this.config=v}static async from_pretrained(v,M){const y=await(0,j.getModelJSON)(v,_.FEATURE_EXTRACTOR_NAME,!0,M);return new this(y)}}function R(g,v){var M;if(!(g instanceof Float32Array||g instanceof Float64Array))throw new Error(`${v} expects input to be a Float32Array or a Float64Array, but got ${((M=g==null?void 0:g.constructor)==null?void 0:M.name)??typeof g} instead. If using the feature extractor directly, remember to use \`read_audio(url, sampling_rate)\` to obtain the raw audio data of the file/url.`)}},"./src/base/image_processors_utils.js":(Le,A,r)=>{r.r(A),r.d(A,{ImageProcessor:()=>$,center_to_corners_format:()=>y,post_process_instance_segmentation:()=>q,post_process_object_detection:()=>b,post_process_panoptic_segmentation:()=>U,post_process_semantic_segmentation:()=>I});var _=r("./src/utils/generic.js"),D=r("./src/utils/tensor.js"),j=r("./src/utils/maths.js");r("./src/utils/image.js");var X=r("./src/utils/core.js"),R=r("./src/utils/hub.js"),g=r("./src/utils/constants.js");function v(S,w,x=0,O=null){const ae=S/w;let ie=(0,j.bankers_round)(ae)*w;return O!==null&&ie>O&&(ie=Math.floor(ae)*w),iew&&J.push(Ce)}else{let Ce=(0,j.max)(ye.data)[1];if(Ce===re-1||(de=(0,j.softmax)(ye.data),de[Ce]Ze*ke[(te+1)%2])),Ie.boxes.push(Be),Ie.classes.push(Ce),Ie.scores.push(de[Ce])}}xe.push(Ie)}return xe}function I(S,w=null){const x=S.logits,O=x.dims[0];if(w!==null&&w.length!==O)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");const ae=[];for(let ie=0;ieke[J]&&(ke[J]=ye[J],Ie[J]=Ge)}const Ee=new Array(we.dims[0]);for(let Ge=0;GeGe!==void 0);ae.push({segmentation:ce,labels:tt})}return ae}function K(S,w,x,O){const ae=[],ie=[],ve=[];for(let we=0;wex&&(ae.push(xe),ie.push(Ie),ve.push(ce))}return[ae,ie,ve]}function se(S,w,x,O=.5,ae=.8){const ie=[];let ve=0,we=0;const re=w[x].data;for(let ce=0;ce=O&&++we;let xe=ve>0&&we>0;return xe&&(xe=ve/we>ae),[xe,ie]}function ne(S,w,x,O,ae,ie=null,ve=null){const[we,re]=ve??S[0].dims,xe=new D.Tensor("int32",new Int32Array(we*re),[we,re]),ce=[];if(ve!==null)for(let Ge=0;GeIe[de]&&(ke[de]=Ge,Ie[de]=J[de])}let Ee=0;const tt=xe.data;for(let Ge=0;Ge200)throw new Error(`absolute aspect ratio must be smaller than 200, got ${Math.max(S,w)/Math.min(S,w)}`);let ie=Math.round(S/x)*x,ve=Math.round(w/x)*x;if(ie*ve>ae){const we=Math.sqrt(S*w/ae);ie=Math.floor(S/we/x)*x,ve=Math.floor(w/we/x)*x}else if(ie*veie?xe=Math.floor(ie*re/ae):ie>ae&&(re=Math.floor(ae*xe/ie)),await w.resize(xe,re,{resample:O}))}async crop_margin(w,x=200){const O=w.clone().grayscale(),ae=(0,j.min)(O.data)[0],ve=(0,j.max)(O.data)[0]-ae;if(ve===0)return w;const we=x/255;let re=O.width,xe=O.height,ce=0,ke=0;const Ie=O.data;for(let Ee=0;Eethis.preprocess(ie)));return{pixel_values:(0,D.stack)(O.map(ie=>ie.pixel_values),0),original_sizes:O.map(ie=>ie.original_size),reshaped_input_sizes:O.map(ie=>ie.reshaped_input_size)}}static async from_pretrained(w,x){const O=await(0,R.getModelJSON)(w,g.IMAGE_PROCESSOR_NAME,!0,x);return new this(O)}}},"./src/base/processing_utils.js":(Le,A,r)=>{r.r(A),r.d(A,{Processor:()=>X});var _=r("./src/utils/constants.js"),D=r("./src/utils/generic.js"),j=r("./src/utils/hub.js");class X extends D.Callable{constructor(g,v){super(),this.config=g,this.components=v}get image_processor(){return this.components.image_processor}get tokenizer(){return this.components.tokenizer}get feature_extractor(){return this.components.feature_extractor}apply_chat_template(g,v={}){if(!this.tokenizer)throw new Error("Unable to apply chat template without a tokenizer.");return this.tokenizer.apply_chat_template(g,{tokenize:!1,...v})}batch_decode(...g){if(!this.tokenizer)throw new Error("Unable to decode without a tokenizer.");return this.tokenizer.batch_decode(...g)}decode(...g){if(!this.tokenizer)throw new Error("Unable to decode without a tokenizer.");return this.tokenizer.decode(...g)}async _call(g,...v){for(const M of[this.image_processor,this.feature_extractor,this.tokenizer])if(M)return M(g,...v);throw new Error("No image processor, feature extractor, or tokenizer found.")}static async from_pretrained(g,v){const[M,y]=await Promise.all([this.uses_processor_config?(0,j.getModelJSON)(g,_.PROCESSOR_NAME,!0,v):{},Promise.all(this.classes.filter(b=>b in this).map(async b=>{const I=await this[b].from_pretrained(g,v);return[b.replace(/_class$/,""),I]})).then(Object.fromEntries)]);return new this(M,y)}}me(X,"classes",["image_processor_class","tokenizer_class","feature_extractor_class"]),me(X,"uses_processor_config",!1)},"./src/configs.js":(Le,A,r)=>{r.r(A),r.d(A,{AutoConfig:()=>v,PretrainedConfig:()=>g,getKeyValueShapes:()=>R});var _=r("./src/utils/core.js"),D=r("./src/utils/hub.js");async function j(M,y){return await(0,D.getModelJSON)(M,"config.json",!0,y)}function X(M){const y={};let b={};switch(M.model_type){case"llava":case"paligemma":case"florence2":case"llava_onevision":case"idefics3":b=X(M.text_config);break;case"moondream1":b=X(M.phi_config);break;case"musicgen":b=X(M.decoder);break;case"multi_modality":b=X(M.language_config);break;case"gpt2":case"gptj":case"jais":case"codegen":case"gpt_bigcode":y.num_heads="n_head",y.num_layers="n_layer",y.hidden_size="n_embd";break;case"gpt_neox":case"stablelm":case"opt":case"falcon":y.num_heads="num_attention_heads",y.num_layers="num_hidden_layers",y.hidden_size="hidden_size";break;case"llama":case"olmo":case"olmo2":case"mobilellm":case"granite":case"cohere":case"mistral":case"starcoder2":case"qwen2":case"qwen2_vl":case"phi":case"phi3":case"phi3_v":y.num_heads="num_key_value_heads",y.num_layers="num_hidden_layers",y.hidden_size="hidden_size",y.num_attention_heads="num_attention_heads";break;case"gemma":case"gemma2":case"glm":case"helium":y.num_heads="num_key_value_heads",y.num_layers="num_hidden_layers",y.dim_kv="head_dim";break;case"openelm":y.num_heads="num_kv_heads",y.num_layers="num_transformer_layers",y.dim_kv="head_dim";break;case"gpt_neo":case"donut-swin":y.num_heads="num_heads",y.num_layers="num_layers",y.hidden_size="hidden_size";break;case"bloom":y.num_heads="n_head",y.num_layers="n_layer",y.hidden_size="hidden_size";break;case"mpt":y.num_heads="n_heads",y.num_layers="n_layers",y.hidden_size="d_model";break;case"exaone":y.num_heads="num_key_value_heads",y.num_layers="num_layers",y.dim_kv="head_dim",y.num_attention_heads="num_attention_heads";break;case"t5":case"mt5":case"longt5":y.num_decoder_layers="num_decoder_layers",y.num_decoder_heads="num_heads",y.decoder_dim_kv="d_kv",y.num_encoder_layers="num_layers",y.num_encoder_heads="num_heads",y.encoder_dim_kv="d_kv";break;case"bart":case"mbart":case"marian":case"whisper":case"m2m_100":case"blenderbot":case"blenderbot-small":case"florence2_language":y.num_decoder_layers="decoder_layers",y.num_decoder_heads="decoder_attention_heads",y.decoder_hidden_size="d_model",y.num_encoder_layers="encoder_layers",y.num_encoder_heads="encoder_attention_heads",y.encoder_hidden_size="d_model";break;case"speecht5":y.num_decoder_layers="decoder_layers",y.num_decoder_heads="decoder_attention_heads",y.decoder_hidden_size="hidden_size",y.num_encoder_layers="encoder_layers",y.num_encoder_heads="encoder_attention_heads",y.encoder_hidden_size="hidden_size";break;case"trocr":y.num_encoder_layers=y.num_decoder_layers="decoder_layers",y.num_encoder_heads=y.num_decoder_heads="decoder_attention_heads",y.encoder_hidden_size=y.decoder_hidden_size="d_model";break;case"musicgen_decoder":y.num_encoder_layers=y.num_decoder_layers="num_hidden_layers",y.num_encoder_heads=y.num_decoder_heads="num_attention_heads",y.encoder_hidden_size=y.decoder_hidden_size="hidden_size";break;case"moonshine":y.num_decoder_layers="decoder_num_hidden_layers",y.num_decoder_heads="decoder_num_key_value_heads",y.num_encoder_layers="encoder_num_hidden_layers",y.num_encoder_heads="encoder_num_key_value_heads",y.encoder_hidden_size=y.decoder_hidden_size="hidden_size";break;case"vision-encoder-decoder":const K=X(M.decoder),se="num_decoder_layers"in K,ne=(0,_.pick)(M,["model_type","is_encoder_decoder"]);return se?(ne.num_decoder_layers=K.num_decoder_layers,ne.num_decoder_heads=K.num_decoder_heads,ne.decoder_hidden_size=K.decoder_hidden_size,ne.num_encoder_layers=K.num_encoder_layers,ne.num_encoder_heads=K.num_encoder_heads,ne.encoder_hidden_size=K.encoder_hidden_size):(ne.num_layers=K.num_layers,ne.num_heads=K.num_heads,ne.hidden_size=K.hidden_size),ne}const I={...b,...(0,_.pick)(M,["model_type","multi_query","is_encoder_decoder"])};for(const K in y)I[K]=M[y[K]];return I}function R(M,{prefix:y="past_key_values",batch_size:b=1}={}){const I={},K=M.normalized_config;if(K.is_encoder_decoder&&"num_encoder_heads"in K&&"num_decoder_heads"in K){const se=K.encoder_dim_kv??K.encoder_hidden_size/K.num_encoder_heads,ne=K.decoder_dim_kv??K.decoder_hidden_size/K.num_decoder_heads,W=[b,K.num_encoder_heads,0,se],U=[b,K.num_decoder_heads,0,ne];for(let q=0;q{var O,ae;r.r(A),r.d(A,{apis:()=>ne,env:()=>w});var _=r("?569f"),D=r("?3f59"),j=r("?154a");const X="3.3.2",R=typeof window<"u"&&typeof window.document<"u",g=typeof self<"u"&&((O=self.constructor)==null?void 0:O.name)==="DedicatedWorkerGlobalScope",v=typeof self<"u"&&"caches"in self,M=typeof navigator<"u"&&"gpu"in navigator,y=typeof navigator<"u"&&"ml"in navigator,b=typeof process<"u",I=b&&((ae=process==null?void 0:process.release)==null?void 0:ae.name)==="node",K=!x(_),se=!x(D),ne=Object.freeze({IS_BROWSER_ENV:R,IS_WEBWORKER_ENV:g,IS_WEB_CACHE_AVAILABLE:v,IS_WEBGPU_AVAILABLE:M,IS_WEBNN_AVAILABLE:y,IS_PROCESS_AVAILABLE:b,IS_NODE_ENV:I,IS_FS_AVAILABLE:K,IS_PATH_AVAILABLE:se}),W=K&&se;let U="./";if(W){const ie=Object({url:self.location.href}).url;ie?U=D.dirname(D.dirname(j.fileURLToPath(ie))):typeof __dirname<"u"&&(U=D.dirname(__dirname))}const q=W?D.join(U,"/.cache/"):null,$="/models/",S=W?D.join(U,$):$,w={version:X,backends:{onnx:{}},allowRemoteModels:!0,remoteHost:"https://huggingface.co/",remotePathTemplate:"{model}/resolve/{revision}/",allowLocalModels:!(R||g),localModelPath:S,useFS:K,useBrowserCache:v,useFSCache:K,cacheDir:q,useCustomCache:!1,customCache:null};function x(ie){return Object.keys(ie).length===0}},"./src/generation/configuration_utils.js":(Le,A,r)=>{r.r(A),r.d(A,{GenerationConfig:()=>D});var _=r("./src/utils/core.js");class D{constructor(X){me(this,"max_length",20);me(this,"max_new_tokens",null);me(this,"min_length",0);me(this,"min_new_tokens",null);me(this,"early_stopping",!1);me(this,"max_time",null);me(this,"do_sample",!1);me(this,"num_beams",1);me(this,"num_beam_groups",1);me(this,"penalty_alpha",null);me(this,"use_cache",!0);me(this,"temperature",1);me(this,"top_k",50);me(this,"top_p",1);me(this,"typical_p",1);me(this,"epsilon_cutoff",0);me(this,"eta_cutoff",0);me(this,"diversity_penalty",0);me(this,"repetition_penalty",1);me(this,"encoder_repetition_penalty",1);me(this,"length_penalty",1);me(this,"no_repeat_ngram_size",0);me(this,"bad_words_ids",null);me(this,"force_words_ids",null);me(this,"renormalize_logits",!1);me(this,"constraints",null);me(this,"forced_bos_token_id",null);me(this,"forced_eos_token_id",null);me(this,"remove_invalid_values",!1);me(this,"exponential_decay_length_penalty",null);me(this,"suppress_tokens",null);me(this,"streamer",null);me(this,"begin_suppress_tokens",null);me(this,"forced_decoder_ids",null);me(this,"guidance_scale",null);me(this,"num_return_sequences",1);me(this,"output_attentions",!1);me(this,"output_hidden_states",!1);me(this,"output_scores",!1);me(this,"return_dict_in_generate",!1);me(this,"pad_token_id",null);me(this,"bos_token_id",null);me(this,"eos_token_id",null);me(this,"encoder_no_repeat_ngram_size",0);me(this,"decoder_start_token_id",null);me(this,"generation_kwargs",{});Object.assign(this,(0,_.pick)(X,Object.getOwnPropertyNames(this)))}}},"./src/generation/logits_process.js":(Le,A,r)=>{r.r(A),r.d(A,{ClassifierFreeGuidanceLogitsProcessor:()=>W,ForcedBOSTokenLogitsProcessor:()=>g,ForcedEOSTokenLogitsProcessor:()=>v,LogitsProcessor:()=>j,LogitsProcessorList:()=>R,LogitsWarper:()=>X,MinLengthLogitsProcessor:()=>K,MinNewTokensLengthLogitsProcessor:()=>se,NoBadWordsLogitsProcessor:()=>ne,NoRepeatNGramLogitsProcessor:()=>b,RepetitionPenaltyLogitsProcessor:()=>I,SuppressTokensAtBeginLogitsProcessor:()=>M,TemperatureLogitsWarper:()=>U,TopKLogitsWarper:()=>$,TopPLogitsWarper:()=>q,WhisperTimeStampLogitsProcessor:()=>y});var _=r("./src/utils/generic.js");r("./src/utils/tensor.js");var D=r("./src/utils/maths.js");class j extends _.Callable{_call(w,x){throw Error("`_call` should be implemented in a subclass")}}class X extends _.Callable{_call(w,x){throw Error("`_call` should be implemented in a subclass")}}class R extends _.Callable{constructor(){super(),this.processors=[]}push(w){this.processors.push(w)}extend(w){this.processors.push(...w)}_call(w,x){let O=x;for(const ae of this.processors)O=ae(w,O);return O}[Symbol.iterator](){return this.processors.values()}}class g extends j{constructor(w){super(),this.bos_token_id=w}_call(w,x){for(let O=0;O=1&&ie[ie.length-1]>=this.timestamp_begin,we=ie.length<2||ie[ie.length-2]>=this.timestamp_begin;if(ve&&(we?ae.subarray(this.timestamp_begin).fill(-1/0):ae.subarray(0,this.eos_token_id).fill(-1/0)),w[O].length===this.begin_index&&this.max_initial_timestamp_index!==null){const ke=this.timestamp_begin+this.max_initial_timestamp_index;ae.subarray(ke+1).fill(-1/0)}const re=(0,D.log_softmax)(ae),xe=Math.log(re.subarray(this.timestamp_begin).map(Math.exp).reduce((ke,Ie)=>ke+Ie)),ce=(0,D.max)(re.subarray(0,this.timestamp_begin))[0];xe>ce&&ae.subarray(0,this.timestamp_begin).fill(-1/0)}return x}}class b extends j{constructor(w){super(),this.no_repeat_ngram_size=w}getNgrams(w){const x=w.length,O=[];for(let ie=0;ie1 to use the classifier free guidance processor, got guidance scale ${w}.`);this.guidance_scale=w}_call(w,x){if(x.dims[0]!==2*w.length)throw new Error(`Logits should have twice the batch size of the input ids, the first half of batches corresponding to the conditional inputs, and the second half of batches corresponding to the unconditional inputs. Got batch size ${x.dims[0]} for the logits and ${w.length} for the input ids.`);const O=w.length,ae=x.slice([0,O],null),ie=x.slice([O,x.dims[0]],null);for(let ve=0;ve1)throw new Error(`\`top_p\` must be a float > 0 and < 1, but is ${w}`);if(!Number.isInteger(O)||O<1)throw new Error(`\`min_tokens_to_keep\` must be a positive integer, but is ${O}`);this.top_p=w,this.filter_value=x,this.min_tokens_to_keep=O}}class $ extends X{constructor(w,{filter_value:x=-1/0,min_tokens_to_keep:O=1}={}){if(super(),!Number.isInteger(w)||w<0)throw new Error(`\`top_k\` must be a positive integer, but is ${w}`);this.top_k=Math.max(w,O),this.filter_value=x}}},"./src/generation/logits_sampler.js":(Le,A,r)=>{r.r(A),r.d(A,{LogitsSampler:()=>X});var _=r("./src/utils/generic.js"),D=r("./src/utils/tensor.js"),j=r("./src/utils/maths.js");r("./src/generation/configuration_utils.js");class X extends _.Callable{constructor(y){super(),this.generation_config=y}async _call(y){return this.sample(y)}async sample(y){throw Error("sample should be implemented in subclasses.")}getLogits(y,b){let I=y.dims.at(-1),K=y.data;if(b===-1)K=K.slice(-I);else{let se=b*I;K=K.slice(se,se+I)}return K}randomSelect(y){let b=0;for(let K=0;K1)return new v(y);if(y.num_return_sequences>1)throw Error(`num_return_sequences has to be 1 when doing greedy search, but is ${y.num_return_sequences}.`);return new R(y)}}class R extends X{async sample(y){const b=(0,j.max)(y.data)[1];return[[BigInt(b),0]]}}class g extends X{async sample(y){let b=y.dims.at(-1);this.generation_config.top_k>0&&(b=Math.min(this.generation_config.top_k,b));const[I,K]=await(0,D.topk)(y,b),se=(0,j.softmax)(I.data);return Array.from({length:this.generation_config.num_beams},()=>{const ne=this.randomSelect(se);return[K.data[ne],Math.log(se[ne])]})}}class v extends X{async sample(y){let b=y.dims.at(-1);this.generation_config.top_k>0&&(b=Math.min(this.generation_config.top_k,b));const[I,K]=await(0,D.topk)(y,b),se=(0,j.softmax)(I.data);return Array.from({length:this.generation_config.num_beams},(ne,W)=>[K.data[W],Math.log(se[W])])}}},"./src/generation/stopping_criteria.js":(Le,A,r)=>{r.r(A),r.d(A,{EosTokenCriteria:()=>R,InterruptableStoppingCriteria:()=>g,MaxLengthCriteria:()=>X,StoppingCriteria:()=>D,StoppingCriteriaList:()=>j});var _=r("./src/utils/generic.js");class D extends _.Callable{_call(M,y){throw Error("StoppingCriteria needs to be subclassed")}}class j extends _.Callable{constructor(){super(),this.criteria=[]}push(M){this.criteria.push(M)}extend(M){M instanceof j?M=M.criteria:M instanceof D&&(M=[M]),this.criteria.push(...M)}_call(M,y){const b=new Array(M.length).fill(!1);for(const I of this.criteria){const K=I(M,y);for(let se=0;sey.length>=this.max_length)}}class R extends D{constructor(M){super(),Array.isArray(M)||(M=[M]),this.eos_token_id=M}_call(M,y){return M.map(b=>{const I=b.at(-1);return this.eos_token_id.some(K=>I==K)})}}class g extends D{constructor(){super(),this.interrupted=!1}interrupt(){this.interrupted=!0}reset(){this.interrupted=!1}_call(M,y){return new Array(M.length).fill(this.interrupted)}}},"./src/generation/streamers.js":(Le,A,r)=>{r.r(A),r.d(A,{BaseStreamer:()=>X,TextStreamer:()=>g,WhisperTextStreamer:()=>v});var _=r("./src/utils/core.js"),D=r("./src/tokenizers.js"),j=r("./src/env.js");class X{put(y){throw Error("Not implemented")}end(){throw Error("Not implemented")}}const R=j.apis.IS_PROCESS_AVAILABLE?M=>process.stdout.write(M):M=>console.log(M);class g extends X{constructor(y,{skip_prompt:b=!1,callback_function:I=null,token_callback_function:K=null,skip_special_tokens:se=!0,decode_kwargs:ne={},...W}={}){super(),this.tokenizer=y,this.skip_prompt=b,this.callback_function=I??R,this.token_callback_function=K,this.decode_kwargs={skip_special_tokens:se,...ne,...W},this.token_cache=[],this.print_len=0,this.next_tokens_are_prompt=!0}put(y){var se;if(y.length>1)throw Error("TextStreamer only supports batch size of 1");if(this.skip_prompt&&this.next_tokens_are_prompt){this.next_tokens_are_prompt=!1;return}const b=y[0];(se=this.token_callback_function)==null||se.call(this,b),this.token_cache=(0,_.mergeArrays)(this.token_cache,b);const I=this.tokenizer.decode(this.token_cache,this.decode_kwargs);let K;I.endsWith(` +`)?(K=I.slice(this.print_len),this.token_cache=[],this.print_len=0):I.length>0&&(0,D.is_chinese_char)(I.charCodeAt(I.length-1))?(K=I.slice(this.print_len),this.print_len+=K.length):(K=I.slice(this.print_len,I.lastIndexOf(" ")+1),this.print_len+=K.length),this.on_finalized_text(K,!1)}end(){let y;this.token_cache.length>0?(y=this.tokenizer.decode(this.token_cache,this.decode_kwargs).slice(this.print_len),this.token_cache=[],this.print_len=0):y="",this.next_tokens_are_prompt=!0,this.on_finalized_text(y,!0)}on_finalized_text(y,b){var I,K;y.length>0&&((I=this.callback_function)==null||I.call(this,y)),b&&this.callback_function===R&&j.apis.IS_PROCESS_AVAILABLE&&((K=this.callback_function)==null||K.call(this,` +`))}}class v extends g{constructor(y,{skip_prompt:b=!1,callback_function:I=null,token_callback_function:K=null,on_chunk_start:se=null,on_chunk_end:ne=null,on_finalize:W=null,time_precision:U=.02,skip_special_tokens:q=!0,decode_kwargs:$={}}={}){super(y,{skip_prompt:b,skip_special_tokens:q,callback_function:I,token_callback_function:K,decode_kwargs:$}),this.timestamp_begin=y.timestamp_begin,this.on_chunk_start=se,this.on_chunk_end=ne,this.on_finalize=W,this.time_precision=U,this.waiting_for_timestamp=!1}put(y){var I,K;if(y.length>1)throw Error("WhisperTextStreamer only supports batch size of 1");const b=y[0];if(b.length===1){const se=Number(b[0])-this.timestamp_begin;if(se>=0){const ne=se*this.time_precision;this.waiting_for_timestamp?(I=this.on_chunk_end)==null||I.call(this,ne):(K=this.on_chunk_start)==null||K.call(this,ne),this.waiting_for_timestamp=!this.waiting_for_timestamp,y=[[]]}}return super.put(y)}end(){var y;super.end(),(y=this.on_finalize)==null||y.call(this)}}},"./src/models.js":(Le,A,r)=>{r.r(A),r.d(A,{ASTForAudioClassification:()=>Vi,ASTModel:()=>Na,ASTPreTrainedModel:()=>Yt,AlbertForMaskedLM:()=>as,AlbertForQuestionAnswering:()=>Rn,AlbertForSequenceClassification:()=>Tn,AlbertModel:()=>Bn,AlbertPreTrainedModel:()=>Gr,AutoModel:()=>Ca,AutoModelForAudioClassification:()=>wc,AutoModelForAudioFrameClassification:()=>Tp,AutoModelForCTC:()=>gc,AutoModelForCausalLM:()=>ac,AutoModelForDepthEstimation:()=>Pp,AutoModelForDocumentQuestionAnswering:()=>Mc,AutoModelForImageClassification:()=>cc,AutoModelForImageFeatureExtraction:()=>Pc,AutoModelForImageMatting:()=>bc,AutoModelForImageSegmentation:()=>pc,AutoModelForImageToImage:()=>vc,AutoModelForMaskGeneration:()=>_c,AutoModelForMaskedLM:()=>lc,AutoModelForNormalEstimation:()=>xc,AutoModelForObjectDetection:()=>mc,AutoModelForPoseEstimation:()=>Tc,AutoModelForQuestionAnswering:()=>uc,AutoModelForSemanticSegmentation:()=>ka,AutoModelForSeq2SeqLM:()=>rc,AutoModelForSequenceClassification:()=>xp,AutoModelForSpeechSeq2Seq:()=>nc,AutoModelForTextToSpectrogram:()=>ic,AutoModelForTextToWaveform:()=>oc,AutoModelForTokenClassification:()=>sc,AutoModelForUniversalSegmentation:()=>hc,AutoModelForVision2Seq:()=>dc,AutoModelForXVector:()=>yc,AutoModelForZeroShotObjectDetection:()=>fc,BartForConditionalGeneration:()=>Kt,BartForSequenceClassification:()=>ms,BartModel:()=>xt,BartPretrainedModel:()=>ft,BaseModelOutput:()=>je,BeitForImageClassification:()=>on,BeitModel:()=>nn,BeitPreTrainedModel:()=>hu,BertForMaskedLM:()=>Ue,BertForQuestionAnswering:()=>Re,BertForSequenceClassification:()=>Ve,BertForTokenClassification:()=>Ne,BertModel:()=>Te,BertPreTrainedModel:()=>le,BlenderbotForConditionalGeneration:()=>Js,BlenderbotModel:()=>ze,BlenderbotPreTrainedModel:()=>Ws,BlenderbotSmallForConditionalGeneration:()=>Xs,BlenderbotSmallModel:()=>ks,BlenderbotSmallPreTrainedModel:()=>Fr,BloomForCausalLM:()=>jl,BloomModel:()=>Nl,BloomPreTrainedModel:()=>vo,CLIPModel:()=>Ja,CLIPPreTrainedModel:()=>rn,CLIPSegForImageSegmentation:()=>il,CLIPSegModel:()=>nl,CLIPSegPreTrainedModel:()=>Xi,CLIPTextModel:()=>Rc,CLIPTextModelWithProjection:()=>Za,CLIPVisionModel:()=>Nc,CLIPVisionModelWithProjection:()=>el,CamembertForMaskedLM:()=>Jr,CamembertForQuestionAnswering:()=>Zr,CamembertForSequenceClassification:()=>$r,CamembertForTokenClassification:()=>Ar,CamembertModel:()=>Nt,CamembertPreTrainedModel:()=>Pr,CausalLMOutput:()=>cn,CausalLMOutputWithPast:()=>Ep,ChineseCLIPModel:()=>gr,ChineseCLIPPreTrainedModel:()=>jc,ClapAudioModelWithProjection:()=>fd,ClapModel:()=>nr,ClapPreTrainedModel:()=>ua,ClapTextModelWithProjection:()=>md,CodeGenForCausalLM:()=>gl,CodeGenModel:()=>ro,CodeGenPreTrainedModel:()=>di,CohereForCausalLM:()=>Cl,CohereModel:()=>El,CoherePreTrainedModel:()=>fo,ConvBertForMaskedLM:()=>kt,ConvBertForQuestionAnswering:()=>ys,ConvBertForSequenceClassification:()=>At,ConvBertForTokenClassification:()=>is,ConvBertModel:()=>vt,ConvBertPreTrainedModel:()=>mt,ConvNextForImageClassification:()=>Jo,ConvNextModel:()=>ku,ConvNextPreTrainedModel:()=>Yo,ConvNextV2ForImageClassification:()=>ea,ConvNextV2Model:()=>Su,ConvNextV2PreTrainedModel:()=>Zo,DPTForDepthEstimation:()=>Qc,DPTModel:()=>yu,DPTPreTrainedModel:()=>qo,DebertaForMaskedLM:()=>Ir,DebertaForQuestionAnswering:()=>ar,DebertaForSequenceClassification:()=>Rr,DebertaForTokenClassification:()=>Nr,DebertaModel:()=>en,DebertaPreTrainedModel:()=>cr,DebertaV2ForMaskedLM:()=>Dt,DebertaV2ForQuestionAnswering:()=>Or,DebertaV2ForSequenceClassification:()=>Vs,DebertaV2ForTokenClassification:()=>jr,DebertaV2Model:()=>Tt,DebertaV2PreTrainedModel:()=>it,DecisionTransformerModel:()=>mp,DecisionTransformerPreTrainedModel:()=>kd,DeiTForImageClassification:()=>No,DeiTModel:()=>_u,DeiTPreTrainedModel:()=>_i,DepthAnythingForDepthEstimation:()=>bu,DepthAnythingPreTrainedModel:()=>Mu,DepthProForDepthEstimation:()=>Xo,DepthProPreTrainedModel:()=>yi,DetrForObjectDetection:()=>an,DetrForSegmentation:()=>Ks,DetrModel:()=>Fo,DetrObjectDetectionOutput:()=>Do,DetrPreTrainedModel:()=>Dr,DetrSegmentationOutput:()=>Lo,Dinov2ForImageClassification:()=>Jc,Dinov2Model:()=>$u,Dinov2PreTrainedModel:()=>ta,Dinov2WithRegistersForImageClassification:()=>Iu,Dinov2WithRegistersModel:()=>Au,Dinov2WithRegistersPreTrainedModel:()=>sa,DistilBertForMaskedLM:()=>wn,DistilBertForQuestionAnswering:()=>es,DistilBertForSequenceClassification:()=>Os,DistilBertForTokenClassification:()=>Er,DistilBertModel:()=>lr,DistilBertPreTrainedModel:()=>Ms,DonutSwinModel:()=>Cu,DonutSwinPreTrainedModel:()=>Yc,EfficientNetForImageClassification:()=>bd,EfficientNetModel:()=>Md,EfficientNetPreTrainedModel:()=>ki,ElectraForMaskedLM:()=>sr,ElectraForQuestionAnswering:()=>Us,ElectraForSequenceClassification:()=>Sr,ElectraForTokenClassification:()=>Yr,ElectraModel:()=>Ds,ElectraPreTrainedModel:()=>Cs,EsmForMaskedLM:()=>An,EsmForSequenceClassification:()=>In,EsmForTokenClassification:()=>On,EsmModel:()=>ii,EsmPreTrainedModel:()=>Ur,ExaoneForCausalLM:()=>uo,ExaoneModel:()=>Ml,ExaonePreTrainedModel:()=>lo,FalconForCausalLM:()=>hd,FalconModel:()=>pd,FalconPreTrainedModel:()=>la,FastViTForImageClassification:()=>ru,FastViTModel:()=>su,FastViTPreTrainedModel:()=>So,Florence2ForConditionalGeneration:()=>Ha,Florence2PreTrainedModel:()=>Ka,GLPNForDepthEstimation:()=>Eu,GLPNModel:()=>Xc,GLPNPreTrainedModel:()=>bi,GPT2LMHeadModel:()=>al,GPT2Model:()=>ol,GPT2PreTrainedModel:()=>Yi,GPTBigCodeForCausalLM:()=>_l,GPTBigCodeModel:()=>fl,GPTBigCodePreTrainedModel:()=>so,GPTJForCausalLM:()=>ml,GPTJModel:()=>hl,GPTJPreTrainedModel:()=>to,GPTNeoForCausalLM:()=>dl,GPTNeoModel:()=>yr,GPTNeoPreTrainedModel:()=>Zi,GPTNeoXForCausalLM:()=>pl,GPTNeoXModel:()=>cl,GPTNeoXPreTrainedModel:()=>eo,Gemma2ForCausalLM:()=>Al,Gemma2Model:()=>$l,Gemma2PreTrainedModel:()=>go,GemmaForCausalLM:()=>Sl,GemmaModel:()=>kl,GemmaPreTrainedModel:()=>_o,GlmForCausalLM:()=>Nn,GlmModel:()=>yl,GlmPreTrainedModel:()=>ao,GraniteForCausalLM:()=>Pl,GraniteModel:()=>ds,GranitePreTrainedModel:()=>mo,GroundingDinoForObjectDetection:()=>Ou,GroundingDinoPreTrainedModel:()=>Zc,GroupViTModel:()=>Mr,GroupViTPreTrainedModel:()=>tu,HeliumForCausalLM:()=>wl,HeliumModel:()=>ci,HeliumPreTrainedModel:()=>oo,HieraForImageClassification:()=>Uo,HieraModel:()=>gi,HieraPreTrainedModel:()=>jo,HubertForCTC:()=>ed,HubertForSequenceClassification:()=>ap,HubertModel:()=>Zu,HubertPreTrainedModel:()=>op,IJepaForImageClassification:()=>ql,IJepaModel:()=>Hl,IJepaPreTrainedModel:()=>mi,Idefics3ForConditionalGeneration:()=>Hi,Idefics3PreTrainedModel:()=>Xa,ImageMattingOutput:()=>Ec,JAISLMHeadModel:()=>ul,JAISModel:()=>ll,JAISPreTrainedModel:()=>Ji,JinaCLIPModel:()=>ui,JinaCLIPPreTrainedModel:()=>li,JinaCLIPTextModel:()=>Qi,JinaCLIPVisionModel:()=>wr,LlamaForCausalLM:()=>Uc,LlamaModel:()=>io,LlamaPreTrainedModel:()=>no,LlavaForConditionalGeneration:()=>ai,LlavaOnevisionForConditionalGeneration:()=>Wa,LlavaPreTrainedModel:()=>Kr,LongT5ForConditionalGeneration:()=>Se,LongT5Model:()=>be,LongT5PreTrainedModel:()=>ue,M2M100ForConditionalGeneration:()=>ia,M2M100Model:()=>Ru,M2M100PreTrainedModel:()=>na,MBartForCausalLM:()=>rr,MBartForConditionalGeneration:()=>Bt,MBartForSequenceClassification:()=>rs,MBartModel:()=>Fs,MBartPreTrainedModel:()=>us,MPNetForMaskedLM:()=>tn,MPNetForQuestionAnswering:()=>vn,MPNetForSequenceClassification:()=>Mn,MPNetForTokenClassification:()=>bn,MPNetModel:()=>yn,MPNetPreTrainedModel:()=>ur,MT5ForConditionalGeneration:()=>gt,MT5Model:()=>pt,MT5PreTrainedModel:()=>Qe,MarianMTModel:()=>Bu,MarianModel:()=>zu,MarianPreTrainedModel:()=>xi,MaskFormerForInstanceSegmentation:()=>Pu,MaskFormerModel:()=>Tu,MaskFormerPreTrainedModel:()=>Mi,MaskedLMOutput:()=>qs,MgpstrForSceneTextRecognition:()=>Od,MgpstrModelOutput:()=>Ad,MgpstrPreTrainedModel:()=>Id,MistralForCausalLM:()=>ud,MistralModel:()=>ld,MistralPreTrainedModel:()=>oa,MobileBertForMaskedLM:()=>oi,MobileBertForQuestionAnswering:()=>fr,MobileBertForSequenceClassification:()=>Wr,MobileBertModel:()=>Fn,MobileBertPreTrainedModel:()=>Vr,MobileLLMForCausalLM:()=>vl,MobileLLMModel:()=>bl,MobileLLMPreTrainedModel:()=>co,MobileNetV1ForImageClassification:()=>fa,MobileNetV1Model:()=>pp,MobileNetV1PreTrainedModel:()=>ma,MobileNetV2ForImageClassification:()=>xd,MobileNetV2Model:()=>vd,MobileNetV2PreTrainedModel:()=>_a,MobileNetV3ForImageClassification:()=>Ed,MobileNetV3Model:()=>Pd,MobileNetV3PreTrainedModel:()=>Td,MobileNetV4ForImageClassification:()=>Cd,MobileNetV4Model:()=>hp,MobileNetV4PreTrainedModel:()=>ga,MobileViTForImageClassification:()=>au,MobileViTModel:()=>ou,MobileViTPreTrainedModel:()=>$o,MobileViTV2ForImageClassification:()=>Kc,MobileViTV2Model:()=>lu,MobileViTV2PreTrainedModel:()=>Ao,ModelOutput:()=>Ke,ModernBertForMaskedLM:()=>ct,ModernBertForSequenceClassification:()=>lt,ModernBertForTokenClassification:()=>ht,ModernBertModel:()=>dt,ModernBertPreTrainedModel:()=>st,Moondream1ForConditionalGeneration:()=>Ga,MoonshineForConditionalGeneration:()=>Va,MoonshineModel:()=>pr,MoonshinePreTrainedModel:()=>Gi,MptForCausalLM:()=>Vl,MptModel:()=>Ul,MptPreTrainedModel:()=>xo,MultiModalityCausalLM:()=>$d,MultiModalityPreTrainedModel:()=>Sd,MusicgenForCausalLM:()=>cp,MusicgenForConditionalGeneration:()=>Si,MusicgenModel:()=>dp,MusicgenPreTrainedModel:()=>ha,NomicBertModel:()=>oe,NomicBertPreTrainedModel:()=>L,OPTForCausalLM:()=>Gl,OPTModel:()=>Wl,OPTPreTrainedModel:()=>To,Olmo2ForCausalLM:()=>Wc,Olmo2Model:()=>Tl,Olmo2PreTrainedModel:()=>ho,OlmoForCausalLM:()=>xl,OlmoModel:()=>Vc,OlmoPreTrainedModel:()=>po,OpenELMForCausalLM:()=>Ol,OpenELMModel:()=>Il,OpenELMPreTrainedModel:()=>wo,OwlViTForObjectDetection:()=>du,OwlViTModel:()=>uu,OwlViTPreTrainedModel:()=>Io,Owlv2ForObjectDetection:()=>pu,Owlv2Model:()=>cu,Owlv2PreTrainedModel:()=>Oo,PaliGemmaForConditionalGeneration:()=>Qa,PaliGemmaPreTrainedModel:()=>qa,PatchTSMixerForPrediction:()=>Bd,PatchTSMixerModel:()=>zd,PatchTSMixerPreTrainedModel:()=>wa,PatchTSTForPrediction:()=>Ld,PatchTSTModel:()=>Dd,PatchTSTPreTrainedModel:()=>Fd,Phi3ForCausalLM:()=>Rl,Phi3Model:()=>Bl,Phi3PreTrainedModel:()=>bo,Phi3VForCausalLM:()=>hr,Phi3VPreTrainedModel:()=>Ya,PhiForCausalLM:()=>zl,PhiModel:()=>Ll,PhiPreTrainedModel:()=>Mo,PreTrainedModel:()=>te,PretrainedMixin:()=>ws,PvtForImageClassification:()=>Co,PvtModel:()=>Yl,PvtPreTrainedModel:()=>Eo,PyAnnoteForAudioFrameClassification:()=>Wu,PyAnnoteModel:()=>Vu,PyAnnotePreTrainedModel:()=>Ti,QuestionAnsweringModelOutput:()=>tr,Qwen2ForCausalLM:()=>Fl,Qwen2Model:()=>jn,Qwen2PreTrainedModel:()=>yo,Qwen2VLForConditionalGeneration:()=>Dl,Qwen2VLPreTrainedModel:()=>pi,RTDetrForObjectDetection:()=>Un,RTDetrModel:()=>Hc,RTDetrObjectDetectionOutput:()=>Bo,RTDetrPreTrainedModel:()=>zo,ResNetForImageClassification:()=>qc,ResNetModel:()=>gu,ResNetPreTrainedModel:()=>Vo,RoFormerForMaskedLM:()=>$e,RoFormerForQuestionAnswering:()=>ut,RoFormerForSequenceClassification:()=>We,RoFormerForTokenClassification:()=>Je,RoFormerModel:()=>fe,RoFormerPreTrainedModel:()=>H,RobertaForMaskedLM:()=>_r,RobertaForQuestionAnswering:()=>yt,RobertaForSequenceClassification:()=>fs,RobertaForTokenClassification:()=>Ss,RobertaModel:()=>ir,RobertaPreTrainedModel:()=>Ot,SamImageSegmentationOutput:()=>Wn,SamModel:()=>vi,SamPreTrainedModel:()=>ep,SapiensForDepthEstimation:()=>wi,SapiensForNormalEstimation:()=>xu,SapiensForSemanticSegmentation:()=>vu,SapiensPreTrainedModel:()=>Qo,SegformerForImageClassification:()=>_d,SegformerForSemanticSegmentation:()=>gd,SegformerModel:()=>up,SegformerPreTrainedModel:()=>dn,Seq2SeqLMOutput:()=>qp,SequenceClassifierOutput:()=>Qt,SiglipModel:()=>tl,SiglipPreTrainedModel:()=>qi,SiglipTextModel:()=>sl,SiglipVisionModel:()=>rl,SpeechT5ForSpeechToText:()=>Cr,SpeechT5ForTextToSpeech:()=>Lr,SpeechT5HifiGan:()=>ln,SpeechT5Model:()=>Kp,SpeechT5PreTrainedModel:()=>Ci,SqueezeBertForMaskedLM:()=>Dn,SqueezeBertForQuestionAnswering:()=>zn,SqueezeBertForSequenceClassification:()=>Ln,SqueezeBertModel:()=>xn,SqueezeBertPreTrainedModel:()=>zt,StableLmForCausalLM:()=>yd,StableLmModel:()=>wd,StableLmPreTrainedModel:()=>pa,Starcoder2ForCausalLM:()=>cd,Starcoder2Model:()=>dd,Starcoder2PreTrainedModel:()=>aa,StyleTextToSpeech2Model:()=>od,StyleTextToSpeech2PreTrainedModel:()=>id,Swin2SRForImageSuperResolution:()=>wu,Swin2SRModel:()=>Ho,Swin2SRPreTrainedModel:()=>Ko,SwinForImageClassification:()=>Vn,SwinModel:()=>Go,SwinPreTrainedModel:()=>Wo,T5ForConditionalGeneration:()=>Q,T5Model:()=>P,T5PreTrainedModel:()=>Pe,TableTransformerForObjectDetection:()=>fu,TableTransformerModel:()=>mu,TableTransformerObjectDetectionOutput:()=>Ro,TableTransformerPreTrainedModel:()=>fi,TokenClassifierOutput:()=>Hs,TrOCRForCausalLM:()=>ad,TrOCRPreTrainedModel:()=>un,UniSpeechForCTC:()=>rp,UniSpeechForSequenceClassification:()=>Hu,UniSpeechModel:()=>Ku,UniSpeechPreTrainedModel:()=>Pi,UniSpeechSatForAudioFrameClassification:()=>Xu,UniSpeechSatForCTC:()=>Qu,UniSpeechSatForSequenceClassification:()=>np,UniSpeechSatModel:()=>qu,UniSpeechSatPreTrainedModel:()=>Gn,ViTForImageClassification:()=>Po,ViTMAEModel:()=>Zl,ViTMAEPreTrainedModel:()=>Jl,ViTMSNForImageClassification:()=>eu,ViTMSNModel:()=>Gc,ViTMSNPreTrainedModel:()=>ko,ViTModel:()=>Kl,ViTPreTrainedModel:()=>hi,VisionEncoderDecoderModel:()=>Ki,VitMatteForImageMatting:()=>iu,VitMattePreTrainedModel:()=>nu,VitPoseForPoseEstimation:()=>Xl,VitPosePreTrainedModel:()=>Ql,VitsModel:()=>ca,VitsModelOutput:()=>Cp,VitsPreTrainedModel:()=>da,Wav2Vec2BertForCTC:()=>ip,Wav2Vec2BertForSequenceClassification:()=>Ju,Wav2Vec2BertModel:()=>Yu,Wav2Vec2BertPreTrainedModel:()=>Ei,Wav2Vec2ForAudioFrameClassification:()=>Uu,Wav2Vec2ForCTC:()=>ju,Wav2Vec2ForSequenceClassification:()=>tp,Wav2Vec2Model:()=>Nu,Wav2Vec2PreTrainedModel:()=>Hr,WavLMForAudioFrameClassification:()=>nd,WavLMForCTC:()=>sd,WavLMForSequenceClassification:()=>rd,WavLMForXVector:()=>lp,WavLMModel:()=>td,WavLMPreTrainedModel:()=>Pn,WeSpeakerResNetModel:()=>sp,WeSpeakerResNetPreTrainedModel:()=>Gu,WhisperForConditionalGeneration:()=>Ua,WhisperModel:()=>ja,WhisperPreTrainedModel:()=>Wi,XLMForQuestionAnswering:()=>sn,XLMForSequenceClassification:()=>Gs,XLMForTokenClassification:()=>$t,XLMModel:()=>Ls,XLMPreTrainedModel:()=>qt,XLMRobertaForMaskedLM:()=>za,XLMRobertaForQuestionAnswering:()=>Ra,XLMRobertaForSequenceClassification:()=>Ui,XLMRobertaForTokenClassification:()=>Ba,XLMRobertaModel:()=>It,XLMRobertaPreTrainedModel:()=>qe,XLMWithLMHeadModel:()=>$s,XVectorOutput:()=>Hn,YolosForObjectDetection:()=>Du,YolosModel:()=>Fu,YolosObjectDetectionOutput:()=>Lu,YolosPreTrainedModel:()=>ra});var _=r("./src/configs.js"),D=r("./src/backends/onnx.js"),j=r("./src/utils/dtypes.js"),X=r("./src/utils/generic.js"),R=r("./src/utils/core.js"),g=r("./src/utils/hub.js"),v=r("./src/utils/constants.js"),M=r("./src/generation/logits_process.js"),y=r("./src/generation/configuration_utils.js"),b=r("./src/utils/tensor.js"),I=r("./src/utils/image.js"),K=r("./src/utils/maths.js"),se=r("./src/generation/stopping_criteria.js"),ne=r("./src/generation/logits_sampler.js"),W=r("./src/env.js"),U=r("./src/models/whisper/generation_whisper.js"),q=r("./src/models/whisper/common_whisper.js");const $={EncoderOnly:0,EncoderDecoder:1,Seq2Seq:2,Vision2Seq:3,DecoderOnly:4,MaskGeneration:5,ImageTextToText:6,Musicgen:7,MultiModality:8,Phi3V:9},S=new Map,w=new Map,x=new Map;async function O(f,T,N){var xs;const _e=((xs=N.config)==null?void 0:xs["transformers.js_config"])??{};let Fe=N.device??_e.device;Fe&&typeof Fe!="string"&&(Fe.hasOwnProperty(T)?Fe=Fe[T]:(console.warn(`device not specified for "${T}". Using the default device.`),Fe=null));const Ae=Fe??(W.apis.IS_NODE_ENV?"cpu":"wasm"),et=(0,D.deviceToExecutionProviders)(Ae);let rt=N.dtype??_e.dtype;if(typeof rt!="string"&&(rt&&rt.hasOwnProperty(T)?rt=rt[T]:(rt=j.DEFAULT_DEVICE_DTYPE_MAPPING[Ae]??j.DATA_TYPES.fp32,console.warn(`dtype not specified for "${T}". Using the default dtype (${rt}) for this device (${Ae}).`))),rt===j.DATA_TYPES.auto){let cs=_e.dtype;typeof cs!="string"&&(cs=cs[T]),cs&&cs!==j.DATA_TYPES.auto&&j.DATA_TYPES.hasOwnProperty(cs)?rt=cs:rt=j.DEFAULT_DEVICE_DTYPE_MAPPING[Ae]??j.DATA_TYPES.fp32}const _t=rt;if(j.DEFAULT_DTYPE_SUFFIX_MAPPING.hasOwnProperty(_t)){if(_t===j.DATA_TYPES.fp16&&Ae==="webgpu"&&!await(0,j.isWebGpuFp16Supported)())throw new Error(`The device (${Ae}) does not support fp16.`)}else throw new Error(`Invalid dtype: ${_t}. Should be one of: ${Object.keys(j.DATA_TYPES).join(", ")}`);const Mt=_e.kv_cache_dtype?typeof _e.kv_cache_dtype=="string"?_e.kv_cache_dtype:_e.kv_cache_dtype[_t]??"float32":void 0;if(Mt&&!["float32","float16"].includes(Mt))throw new Error(`Invalid kv_cache_dtype: ${Mt}. Should be one of: float32, float16`);const jt={dtype:_t,kv_cache_dtype:Mt},Vt=j.DEFAULT_DTYPE_SUFFIX_MAPPING[_t],Lt=`${N.subfolder??""}/${T}${Vt}.onnx`,Gt={...N.session_options};Gt.executionProviders??(Gt.executionProviders=et);const ts=_e.free_dimension_overrides;ts?Gt.freeDimensionOverrides??(Gt.freeDimensionOverrides=ts):Ae.startsWith("webnn")&&!Gt.freeDimensionOverrides&&console.warn('WebNN does not currently support dynamic shapes and requires `free_dimension_overrides` to be set in config.json as a field within "transformers.js_config". When `free_dimension_overrides` is not set, you may experience significant performance degradation.');const ns=(0,g.getModelFile)(f,Lt,!0,N),Jt=N.use_external_data_format??_e.use_external_data_format;let os=[];if(Jt&&(Jt===!0||typeof Jt=="object"&&Jt.hasOwnProperty(T)&&Jt[T]===!0)){if(W.apis.IS_NODE_ENV)throw new Error("External data format is not yet supported in Node.js");const cs=`${T}${Vt}.onnx_data`,Es=`${N.subfolder??""}/${cs}`;os.push(new Promise(async(Is,Zs)=>{const Ys=await(0,g.getModelFile)(f,Es,!0,N);Is({path:cs,data:Ys})}))}else Gt.externalData!==void 0&&(os=Gt.externalData.map(async cs=>{if(typeof cs.data=="string"){const Es=await(0,g.getModelFile)(f,cs.data,!0,N);return{...cs,data:Es}}return cs}));if(os.length>0&&(Gt.externalData=await Promise.all(os)),Ae==="webgpu"){const cs=(0,_.getKeyValueShapes)(N.config,{prefix:"present"});if(Object.keys(cs).length>0&&!(0,D.isONNXProxy)()){const Es={};for(const Is in cs)Es[Is]="gpu-buffer";Gt.preferredOutputLocation=Es}}return{buffer:await ns,session_options:Gt,session_config:jt}}async function ae(f,T,N){return Object.fromEntries(await Promise.all(Object.keys(T).map(async _e=>{const{buffer:Fe,session_options:Ae,session_config:et}=await O(f,T[_e],N),rt=await(0,D.createInferenceSession)(Fe,Ae,et);return[_e,rt]})))}async function ie(f,T,N){return Object.fromEntries(await Promise.all(Object.keys(T).map(async _e=>{const Fe=await(0,g.getModelJSON)(f,T[_e],!1,N);return[_e,Fe]})))}function ve(f,T){const N=Object.create(null),_e=[];for(const et of f.inputNames){const rt=T[et];if(!(rt instanceof b.Tensor)){_e.push(et);continue}N[et]=(0,D.isONNXProxy)()?rt.clone():rt}if(_e.length>0)throw new Error(`An error occurred during model execution: "Missing the following inputs: ${_e.join(", ")}.`);const Fe=Object.keys(T).length,Ae=f.inputNames.length;if(Fe>Ae){let et=Object.keys(T).filter(rt=>!f.inputNames.includes(rt));console.warn(`WARNING: Too many inputs were provided (${Fe} > ${Ae}). The following inputs will be ignored: "${et.join(", ")}".`)}return N}async function we(f,T){const N=ve(f,T);try{const _e=Object.fromEntries(Object.entries(N).map(([Ae,et])=>[Ae,et.ort_tensor]));let Fe=await f.run(_e);return Fe=re(Fe),Fe}catch(_e){const Fe=Object.fromEntries(Object.entries(N).map(([Ae,{type:et,dims:rt,data:_t}])=>[Ae,{type:et,dims:rt,data:_t}]));throw console.error(`An error occurred during model execution: "${_e}".`),console.error("Inputs given to model:",Fe),_e}}function re(f){for(let T in f)(0,D.isONNXTensor)(f[T])?f[T]=new b.Tensor(f[T]):typeof f[T]=="object"&&re(f[T]);return f}function xe(f){if(f instanceof b.Tensor)return f;if(f.length===0)throw Error("items must be non-empty");if(Array.isArray(f[0])){if(f.some(T=>T.length!==f[0].length))throw Error("Unable to create tensor, you should probably activate truncation and/or padding with 'padding=True' and/or 'truncation=True' to have batched tensors with the same length.");return new b.Tensor("int64",BigInt64Array.from(f.flat().map(T=>BigInt(T))),[f.length,f[0].length])}else return new b.Tensor("int64",BigInt64Array.from(f.map(T=>BigInt(T))),[1,f.length])}function ce(f){return new b.Tensor("bool",[f],[1])}async function ke(f,T){let{encoder_outputs:N,input_ids:_e,decoder_input_ids:Fe,...Ae}=T;if(!N){const rt=(0,R.pick)(T,f.sessions.model.inputNames);N=(await Ie(f,rt)).last_hidden_state}return Ae.input_ids=Fe,Ae.encoder_hidden_states=N,f.sessions.decoder_model_merged.inputNames.includes("encoder_attention_mask")&&(Ae.encoder_attention_mask=T.attention_mask),await Ee(f,Ae,!0)}async function Ie(f,T){const N=f.sessions.model,_e=(0,R.pick)(T,N.inputNames);if(N.inputNames.includes("inputs_embeds")&&!_e.inputs_embeds){if(!T.input_ids)throw new Error("Both `input_ids` and `inputs_embeds` are missing in the model inputs.");_e.inputs_embeds=await f.encode_text({input_ids:T.input_ids})}if(N.inputNames.includes("token_type_ids")&&!_e.token_type_ids){if(!_e.input_ids)throw new Error("Both `input_ids` and `token_type_ids` are missing in the model inputs.");_e.token_type_ids=(0,b.zeros_like)(_e.input_ids)}if(N.inputNames.includes("pixel_mask")&&!_e.pixel_mask){if(!_e.pixel_values)throw new Error("Both `pixel_values` and `pixel_mask` are missing in the model inputs.");const Fe=_e.pixel_values.dims;_e.pixel_mask=(0,b.ones)([Fe[0],Fe[2],Fe[3]])}return await we(N,_e)}async function Ee(f,T,N=!1){const _e=f.sessions[N?"decoder_model_merged":"model"],{past_key_values:Fe,...Ae}=T;if(_e.inputNames.includes("use_cache_branch")&&(Ae.use_cache_branch=ce(!!Fe)),_e.inputNames.includes("position_ids")&&Ae.attention_mask&&!Ae.position_ids){const rt=f.config.model_type==="paligemma"?1:0;Ae.position_ids=J(Ae,Fe,rt)}f.addPastKeyValues(Ae,Fe);const et=(0,R.pick)(Ae,_e.inputNames);return await we(_e,et)}function tt({image_token_id:f,inputs_embeds:T,image_features:N,input_ids:_e,attention_mask:Fe}){const Ae=_e.tolist().map(Mt=>Mt.reduce((jt,Vt,Lt)=>(Vt==f&&jt.push(Lt),jt),[])),et=Ae.reduce((Mt,jt)=>Mt+jt.length,0),rt=N.dims[0];if(et!==rt)throw new Error(`Image features and image tokens do not match: tokens: ${et}, features ${rt}`);let _t=0;for(let Mt=0;MtAe.dims[1])){if(Fert==f.config.image_token_index)){const rt=f.config.num_image_tokens;if(!rt)throw new Error("`num_image_tokens` is missing in the model configuration.");const _t=Ae.dims[1]-(Fe-rt);N.input_ids=Ae.slice(null,[-_t,null]),N.attention_mask=(0,b.ones)([1,Fe+_t])}}}return N}function Ce(f,T,N,_e){return N.past_key_values&&(T=T.map(Fe=>[Fe.at(-1)])),{...N,decoder_input_ids:xe(T)}}function Be(f,...T){return f.config.is_encoder_decoder?Ce(f,...T):de(f,...T)}function Ze(f,T,N,_e){const Fe=!!N.past_key_values;return _e.guidance_scale!==null&&_e.guidance_scale>1&&(Fe?N.input_ids=(0,b.cat)([N.input_ids,N.input_ids],0):(N.input_ids=(0,b.cat)([N.input_ids,(0,b.full_like)(N.input_ids,BigInt(_e.pad_token_id))],0),N.attention_mask=(0,b.cat)([N.attention_mask,(0,b.full_like)(N.attention_mask,0n)],0))),(Fe||!N.pixel_values)&&(N.pixel_values=(0,b.full)([0,0,3,384,384],1)),Fe&&(N.images_seq_mask=new b.Tensor("bool",new Array(1).fill(!0).fill(!1,0,1),[1,1]),N.images_emb_mask=new b.Tensor("bool",new Array(0).fill(!1),[1,1,0])),N}class te extends X.Callable{constructor(N,_e,Fe){super();me(this,"main_input_name","input_ids");me(this,"forward_params",["input_ids","attention_mask"]);this.config=N,this.sessions=_e,this.configs=Fe;const Ae=x.get(this.constructor),et=S.get(Ae);switch(this.can_generate=!1,this._forward=null,this._prepare_inputs_for_generation=null,et){case $.DecoderOnly:this.can_generate=!0,this._forward=Ee,this._prepare_inputs_for_generation=de;break;case $.Seq2Seq:case $.Vision2Seq:case $.Musicgen:this.can_generate=!0,this._forward=ke,this._prepare_inputs_for_generation=Ce;break;case $.EncoderDecoder:this._forward=ke;break;case $.ImageTextToText:this.can_generate=!0,this._forward=Ge,this._prepare_inputs_for_generation=Be;break;case $.Phi3V:this.can_generate=!0,this._prepare_inputs_for_generation=Be;break;case $.MultiModality:this.can_generate=!0,this._prepare_inputs_for_generation=Ze;break;default:this._forward=Ie;break}this.can_generate&&this.forward_params.push("past_key_values"),this.custom_config=this.config["transformers.js_config"]??{}}async dispose(){var _e;const N=[];for(const Fe of Object.values(this.sessions))(_e=Fe==null?void 0:Fe.handler)!=null&&_e.dispose&&N.push(Fe.handler.dispose());return await Promise.all(N)}static async from_pretrained(N,{progress_callback:_e=null,config:Fe=null,cache_dir:Ae=null,local_files_only:et=!1,revision:rt="main",model_file_name:_t=null,subfolder:Mt="onnx",device:jt=null,dtype:Vt=null,use_external_data_format:Lt=null,session_options:Gt={}}={}){let ts={progress_callback:_e,config:Fe,cache_dir:Ae,local_files_only:et,revision:rt,model_file_name:_t,subfolder:Mt,device:jt,dtype:Vt,use_external_data_format:Lt,session_options:Gt};const ns=x.get(this),Jt=S.get(ns);Fe=ts.config=await _.AutoConfig.from_pretrained(N,ts);let os;if(Jt===$.DecoderOnly)os=await Promise.all([ae(N,{model:ts.model_file_name??"model"},ts),ie(N,{generation_config:"generation_config.json"},ts)]);else if(Jt===$.Seq2Seq||Jt===$.Vision2Seq)os=await Promise.all([ae(N,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},ts),ie(N,{generation_config:"generation_config.json"},ts)]);else if(Jt===$.MaskGeneration)os=await Promise.all([ae(N,{model:"vision_encoder",prompt_encoder_mask_decoder:"prompt_encoder_mask_decoder"},ts)]);else if(Jt===$.EncoderDecoder)os=await Promise.all([ae(N,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},ts)]);else if(Jt===$.ImageTextToText){const As={embed_tokens:"embed_tokens",vision_encoder:"vision_encoder",decoder_model_merged:"decoder_model_merged"};Fe.is_encoder_decoder&&(As.model="encoder_model"),os=await Promise.all([ae(N,As,ts),ie(N,{generation_config:"generation_config.json"},ts)])}else if(Jt===$.Musicgen)os=await Promise.all([ae(N,{model:"text_encoder",decoder_model_merged:"decoder_model_merged",encodec_decode:"encodec_decode"},ts),ie(N,{generation_config:"generation_config.json"},ts)]);else if(Jt===$.MultiModality)os=await Promise.all([ae(N,{prepare_inputs_embeds:"prepare_inputs_embeds",model:"language_model",lm_head:"lm_head",gen_head:"gen_head",gen_img_embeds:"gen_img_embeds",image_decode:"image_decode"},ts),ie(N,{generation_config:"generation_config.json"},ts)]);else if(Jt===$.Phi3V)os=await Promise.all([ae(N,{prepare_inputs_embeds:"prepare_inputs_embeds",model:"model",vision_encoder:"vision_encoder"},ts),ie(N,{generation_config:"generation_config.json"},ts)]);else{if(Jt!==$.EncoderOnly){const As=ns??(Fe==null?void 0:Fe.model_type);As!=="custom"&&console.warn(`Model type for '${As}' not found, assuming encoder-only architecture. Please report this at ${v.GITHUB_ISSUE_URL}.`)}os=await Promise.all([ae(N,{model:ts.model_file_name??"model"},ts)])}return new this(Fe,...os)}async _call(N){return await this.forward(N)}async forward(N){return await this._forward(this,N)}get generation_config(){var N;return((N=this.configs)==null?void 0:N.generation_config)??null}_get_logits_warper(N){const _e=new M.LogitsProcessorList;return N.temperature!==null&&N.temperature!==1&&_e.push(new M.TemperatureLogitsWarper(N.temperature)),N.top_k!==null&&N.top_k!==0&&_e.push(new M.TopKLogitsWarper(N.top_k)),N.top_p!==null&&N.top_p<1&&_e.push(new M.TopPLogitsWarper(N.top_p)),_e}_get_logits_processor(N,_e,Fe=null){const Ae=new M.LogitsProcessorList;if(N.repetition_penalty!==null&&N.repetition_penalty!==1&&Ae.push(new M.RepetitionPenaltyLogitsProcessor(N.repetition_penalty)),N.no_repeat_ngram_size!==null&&N.no_repeat_ngram_size>0&&Ae.push(new M.NoRepeatNGramLogitsProcessor(N.no_repeat_ngram_size)),N.bad_words_ids!==null&&Ae.push(new M.NoBadWordsLogitsProcessor(N.bad_words_ids,N.eos_token_id)),N.min_length!==null&&N.eos_token_id!==null&&N.min_length>0&&Ae.push(new M.MinLengthLogitsProcessor(N.min_length,N.eos_token_id)),N.min_new_tokens!==null&&N.eos_token_id!==null&&N.min_new_tokens>0&&Ae.push(new M.MinNewTokensLengthLogitsProcessor(_e,N.min_new_tokens,N.eos_token_id)),N.forced_bos_token_id!==null&&Ae.push(new M.ForcedBOSTokenLogitsProcessor(N.forced_bos_token_id)),N.forced_eos_token_id!==null&&Ae.push(new M.ForcedEOSTokenLogitsProcessor(N.max_length,N.forced_eos_token_id)),N.begin_suppress_tokens!==null){const et=_e>1||N.forced_bos_token_id===null?_e:_e+1;Ae.push(new M.SuppressTokensAtBeginLogitsProcessor(N.begin_suppress_tokens,et))}return N.guidance_scale!==null&&N.guidance_scale>1&&Ae.push(new M.ClassifierFreeGuidanceLogitsProcessor(N.guidance_scale)),Fe!==null&&Ae.extend(Fe),Ae}_prepare_generation_config(N,_e,Fe=y.GenerationConfig){const Ae={...this.config};for(const rt of["decoder","generator","text_config"])rt in Ae&&Object.assign(Ae,Ae[rt]);const et=new Fe(Ae);return Object.assign(et,this.generation_config??{}),N&&Object.assign(et,N),_e&&Object.assign(et,(0,R.pick)(_e,Object.getOwnPropertyNames(et))),et}_get_stopping_criteria(N,_e=null){const Fe=new se.StoppingCriteriaList;return N.max_length!==null&&Fe.push(new se.MaxLengthCriteria(N.max_length,this.config.max_position_embeddings??null)),N.eos_token_id!==null&&Fe.push(new se.EosTokenCriteria(N.eos_token_id)),_e&&Fe.extend(_e),Fe}_validate_model_class(){if(!this.can_generate){const N=[va,Pa,$i,Rd],_e=x.get(this.constructor),Fe=new Set,Ae=this.config.model_type;for(const rt of N){const _t=rt.get(Ae);_t&&Fe.add(_t[0])}let et=`The current model class (${_e}) is not compatible with \`.generate()\`, as it doesn't have a language model head.`;throw Fe.size>0&&(et+=` Please use the following class instead: ${[...Fe].join(", ")}`),Error(et)}}prepare_inputs_for_generation(...N){return this._prepare_inputs_for_generation(this,...N)}_update_model_kwargs_for_generation({generated_input_ids:N,outputs:_e,model_inputs:Fe,is_encoder_decoder:Ae}){return Fe.past_key_values=this.getPastKeyValues(_e,Fe.past_key_values),Fe.input_ids=new b.Tensor("int64",N.flat(),[N.length,1]),Ae||(Fe.attention_mask=(0,b.cat)([Fe.attention_mask,(0,b.ones)([Fe.attention_mask.dims[0],1])],1)),Fe.position_ids=null,Fe}_prepare_model_inputs({inputs:N,bos_token_id:_e,model_kwargs:Fe}){const Ae=(0,R.pick)(Fe,this.forward_params),et=this.main_input_name;if(et in Ae){if(N)throw new Error("`inputs`: {inputs}` were passed alongside {input_name} which is not allowed. Make sure to either pass {inputs} or {input_name}=...")}else Ae[et]=N;return{inputs_tensor:Ae[et],model_inputs:Ae,model_input_name:et}}async _prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:N,model_inputs:_e,model_input_name:Fe,generation_config:Ae}){if(this.sessions.model.inputNames.includes("inputs_embeds")&&!_e.inputs_embeds&&"_prepare_inputs_embeds"in this){const{input_ids:rt,pixel_values:_t,attention_mask:Mt,...jt}=_e,Vt=await this._prepare_inputs_embeds(_e);_e={...jt,...(0,R.pick)(Vt,["inputs_embeds","attention_mask"])}}let{last_hidden_state:et}=await Ie(this,_e);if(Ae.guidance_scale!==null&&Ae.guidance_scale>1)et=(0,b.cat)([et,(0,b.full_like)(et,0)],0),"attention_mask"in _e&&(_e.attention_mask=(0,b.cat)([_e.attention_mask,(0,b.zeros_like)(_e.attention_mask)],0));else if(_e.decoder_input_ids){const rt=xe(_e.decoder_input_ids).dims[0];if(rt!==et.dims[0]){if(et.dims[0]!==1)throw new Error(`The encoder outputs have a different batch size (${et.dims[0]}) than the decoder inputs (${rt}).`);et=(0,b.cat)(Array.from({length:rt},()=>et),0)}}return _e.encoder_outputs=et,_e}_prepare_decoder_input_ids_for_generation({batch_size:N,model_input_name:_e,model_kwargs:Fe,decoder_start_token_id:Ae,bos_token_id:et,generation_config:rt}){let{decoder_input_ids:_t,...Mt}=Fe;if(!(_t instanceof b.Tensor)){if(_t)Array.isArray(_t[0])||(_t=Array.from({length:N},()=>_t));else if(Ae??(Ae=et),this.config.model_type==="musicgen")_t=Array.from({length:N*this.config.decoder.num_codebooks},()=>[Ae]);else if(Array.isArray(Ae)){if(Ae.length!==N)throw new Error(`\`decoder_start_token_id\` expcted to have length ${N} but got ${Ae.length}`);_t=Ae}else _t=Array.from({length:N},()=>[Ae]);_t=xe(_t)}return Fe.decoder_attention_mask=(0,b.ones_like)(_t),{input_ids:_t,model_inputs:Mt}}async generate({inputs:N=null,generation_config:_e=null,logits_processor:Fe=null,stopping_criteria:Ae=null,streamer:et=null,...rt}){this._validate_model_class(),_e=this._prepare_generation_config(_e,rt);let{inputs_tensor:_t,model_inputs:Mt,model_input_name:jt}=this._prepare_model_inputs({inputs:N,model_kwargs:rt});const Vt=this.config.is_encoder_decoder;Vt&&("encoder_outputs"in Mt||(Mt=await this._prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:_t,model_inputs:Mt,model_input_name:jt,generation_config:_e})));let Lt;Vt?{input_ids:Lt,model_inputs:Mt}=this._prepare_decoder_input_ids_for_generation({batch_size:Mt[jt].dims.at(0),model_input_name:jt,model_kwargs:Mt,decoder_start_token_id:_e.decoder_start_token_id,bos_token_id:_e.bos_token_id,generation_config:_e}):Lt=Mt[jt];let Gt=Lt.dims.at(-1);_e.max_new_tokens!==null&&(_e.max_length=Gt+_e.max_new_tokens);const ts=this._get_logits_processor(_e,Gt,Fe),ns=this._get_stopping_criteria(_e,Ae),Jt=Mt[jt].dims.at(0),os=ne.LogitsSampler.getSampler(_e),As=new Array(Jt).fill(0),xs=Lt.tolist();et&&et.put(xs);let cs,Es={};for(;;){if(Mt=this.prepare_inputs_for_generation(xs,Mt,_e),cs=await this.forward(Mt),_e.output_attentions&&_e.return_dict_in_generate){const or=this.getAttentions(cs);for(const vr in or)vr in Es||(Es[vr]=[]),Es[vr].push(or[vr])}const Ys=cs.logits.slice(null,-1,null),br=ts(xs,Ys),pn=[];for(let or=0;oror))break;Mt=this._update_model_kwargs_for_generation({generated_input_ids:pn,outputs:cs,model_inputs:Mt,is_encoder_decoder:Vt})}et&&et.end();const Is=this.getPastKeyValues(cs,Mt.past_key_values,!0),Zs=new b.Tensor("int64",xs.flat(),[xs.length,xs[0].length]);if(_e.return_dict_in_generate)return{sequences:Zs,past_key_values:Is,...Es};for(const Ys of Object.values(cs))Ys.location==="gpu-buffer"&&Ys.dispose();return Zs}getPastKeyValues(N,_e,Fe=!1){const Ae=Object.create(null);for(const et in N)if(et.startsWith("present")){const rt=et.replace("present","past_key_values"),_t=et.includes("encoder");if(_t&&_e?Ae[rt]=_e[rt]:Ae[rt]=N[et],_e&&(!_t||Fe)){const Mt=_e[rt];Mt.location==="gpu-buffer"&&Mt.dispose()}}return Ae}getAttentions(N){const _e={};for(const Fe of["cross_attentions","encoder_attentions","decoder_attentions"])for(const Ae in N)Ae.startsWith(Fe)&&(Fe in _e||(_e[Fe]=[]),_e[Fe].push(N[Ae]));return _e}addPastKeyValues(N,_e){var Fe,Ae,et;if(_e)Object.assign(N,_e);else{const rt=this.sessions.decoder_model_merged??this.sessions.model,_t=((Fe=rt==null?void 0:rt.config)==null?void 0:Fe.kv_cache_dtype)??"float32",Mt=_t==="float16"?new Uint16Array:[],jt=((et=(Ae=N[this.main_input_name]??N.attention_mask)==null?void 0:Ae.dims)==null?void 0:et[0])??1,Vt=(0,_.getKeyValueShapes)(this.config,{batch_size:jt});for(const Lt in Vt)N[Lt]=new b.Tensor(_t,Mt,Vt[Lt])}}async encode_image({pixel_values:N}){const _e=(await we(this.sessions.vision_encoder,{pixel_values:N})).image_features;return this.config.num_image_tokens||(console.warn(`The number of image tokens was not set in the model configuration. Setting it to the number of features detected by the vision encoder (${_e.dims[1]}).`),this.config.num_image_tokens=_e.dims[1]),_e}async encode_text({input_ids:N}){return(await we(this.sessions.embed_tokens,{input_ids:N})).inputs_embeds}}class Ke{}class je extends Ke{constructor({last_hidden_state:T,hidden_states:N=null,attentions:_e=null}){super(),this.last_hidden_state=T,this.hidden_states=N,this.attentions=_e}}class le extends te{}class Te extends le{}class Ue extends le{async _call(T){return new qs(await super._call(T))}}class Ve extends le{async _call(T){return new Qt(await super._call(T))}}class Ne extends le{async _call(T){return new Hs(await super._call(T))}}class Re extends le{async _call(T){return new tr(await super._call(T))}}class st extends te{}class dt extends st{}class ct extends st{async _call(T){return new qs(await super._call(T))}}class lt extends st{async _call(T){return new Qt(await super._call(T))}}class ht extends st{async _call(T){return new Hs(await super._call(T))}}class L extends te{}class oe extends L{}class H extends te{}class fe extends H{}class $e extends H{async _call(T){return new qs(await super._call(T))}}class We extends H{async _call(T){return new Qt(await super._call(T))}}class Je extends H{async _call(T){return new Hs(await super._call(T))}}class ut extends H{async _call(T){return new tr(await super._call(T))}}class mt extends te{}class vt extends mt{}class kt extends mt{async _call(T){return new qs(await super._call(T))}}class At extends mt{async _call(T){return new Qt(await super._call(T))}}class is extends mt{async _call(T){return new Hs(await super._call(T))}}class ys extends mt{async _call(T){return new tr(await super._call(T))}}class Cs extends te{}class Ds extends Cs{}class sr extends Cs{async _call(T){return new qs(await super._call(T))}}class Sr extends Cs{async _call(T){return new Qt(await super._call(T))}}class Yr extends Cs{async _call(T){return new Hs(await super._call(T))}}class Us extends Cs{async _call(T){return new tr(await super._call(T))}}class Pr extends te{}class Nt extends Pr{}class Jr extends Pr{async _call(T){return new qs(await super._call(T))}}class $r extends Pr{async _call(T){return new Qt(await super._call(T))}}class Ar extends Pr{async _call(T){return new Hs(await super._call(T))}}class Zr extends Pr{async _call(T){return new tr(await super._call(T))}}class cr extends te{}class en extends cr{}class Ir extends cr{async _call(T){return new qs(await super._call(T))}}class Rr extends cr{async _call(T){return new Qt(await super._call(T))}}class Nr extends cr{async _call(T){return new Hs(await super._call(T))}}class ar extends cr{async _call(T){return new tr(await super._call(T))}}class it extends te{}class Tt extends it{}class Dt extends it{async _call(T){return new qs(await super._call(T))}}class Vs extends it{async _call(T){return new Qt(await super._call(T))}}class jr extends it{async _call(T){return new Hs(await super._call(T))}}class Or extends it{async _call(T){return new tr(await super._call(T))}}class Ms extends te{}class lr extends Ms{}class Os extends Ms{async _call(T){return new Qt(await super._call(T))}}class Er extends Ms{async _call(T){return new Hs(await super._call(T))}}class es extends Ms{async _call(T){return new tr(await super._call(T))}}class wn extends Ms{async _call(T){return new qs(await super._call(T))}}class Ur extends te{}class ii extends Ur{}class An extends Ur{async _call(T){return new qs(await super._call(T))}}class In extends Ur{async _call(T){return new Qt(await super._call(T))}}class On extends Ur{async _call(T){return new Hs(await super._call(T))}}class Vr extends te{}class Fn extends Vr{}class oi extends Vr{async _call(T){return new qs(await super._call(T))}}class Wr extends Vr{async _call(T){return new Qt(await super._call(T))}}class fr extends Vr{async _call(T){return new tr(await super._call(T))}}class ur extends te{}class yn extends ur{}class tn extends ur{async _call(T){return new qs(await super._call(T))}}class Mn extends ur{async _call(T){return new Qt(await super._call(T))}}class bn extends ur{async _call(T){return new Hs(await super._call(T))}}class vn extends ur{async _call(T){return new tr(await super._call(T))}}class zt extends te{}class xn extends zt{}class Dn extends zt{async _call(T){return new qs(await super._call(T))}}class Ln extends zt{async _call(T){return new Qt(await super._call(T))}}class zn extends zt{async _call(T){return new tr(await super._call(T))}}class Gr extends te{}class Bn extends Gr{}class Tn extends Gr{async _call(T){return new Qt(await super._call(T))}}class Rn extends Gr{async _call(T){return new tr(await super._call(T))}}class as extends Gr{async _call(T){return new qs(await super._call(T))}}class Pe extends te{constructor(){super(...arguments);me(this,"forward_params",["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"])}}class P extends Pe{}class Q extends Pe{}class ue extends te{}class be extends ue{}class Se extends ue{}class Qe extends te{}class pt extends Qe{}class gt extends Qe{}class ft extends te{}class xt extends ft{}class Kt extends ft{}class ms extends ft{async _call(T){return new Qt(await super._call(T))}}class us extends te{}class Fs extends us{}class Bt extends us{}class rs extends us{async _call(T){return new Qt(await super._call(T))}}class rr extends us{}class Ws extends te{}class ze extends Ws{}class Js extends Ws{}class Fr extends te{}class ks extends Fr{}class Xs extends Fr{}class Ot extends te{}class ir extends Ot{}class _r extends Ot{async _call(T){return new qs(await super._call(T))}}class fs extends Ot{async _call(T){return new Qt(await super._call(T))}}class Ss extends Ot{async _call(T){return new Hs(await super._call(T))}}class yt extends Ot{async _call(T){return new tr(await super._call(T))}}class qt extends te{}class Ls extends qt{}class $s extends qt{async _call(T){return new qs(await super._call(T))}}class Gs extends qt{async _call(T){return new Qt(await super._call(T))}}class $t extends qt{async _call(T){return new Hs(await super._call(T))}}class sn extends qt{async _call(T){return new tr(await super._call(T))}}class qe extends te{}class It extends qe{}class za extends qe{async _call(T){return new qs(await super._call(T))}}class Ui extends qe{async _call(T){return new Qt(await super._call(T))}}class Ba extends qe{async _call(T){return new Hs(await super._call(T))}}class Ra extends qe{async _call(T){return new tr(await super._call(T))}}class Yt extends te{}class Na extends Yt{}class Vi extends Yt{}class Wi extends te{constructor(){super(...arguments);me(this,"requires_attention_mask",!1);me(this,"main_input_name","input_features");me(this,"forward_params",["input_features","attention_mask","decoder_input_ids","decoder_attention_mask","past_key_values"])}}class ja extends Wi{}class Ua extends Wi{_prepare_generation_config(T,N){return super._prepare_generation_config(T,N,U.WhisperGenerationConfig)}_retrieve_init_tokens(T){const N=[T.decoder_start_token_id];let _e=T.language;const Fe=T.task;if(T.is_multilingual){_e||(console.warn("No language specified - defaulting to English (en)."),_e="en");const et=`<|${(0,q.whisper_language_to_code)(_e)}|>`;N.push(T.lang_to_id[et]),N.push(T.task_to_id[Fe??"transcribe"])}else if(_e||Fe)throw new Error("Cannot specify `task` or `language` for an English-only model. If the model is intended to be multilingual, pass `is_multilingual=true` to generate, or update the generation config.");return!T.return_timestamps&&T.no_timestamps_token_id&&N.at(-1)!==T.no_timestamps_token_id?N.push(T.no_timestamps_token_id):T.return_timestamps&&N.at(-1)===T.no_timestamps_token_id&&(console.warn("<|notimestamps|> prompt token is removed from generation_config since `return_timestamps` is set to `true`."),N.pop()),N.filter(Ae=>Ae!=null)}async generate({inputs:T=null,generation_config:N=null,logits_processor:_e=null,stopping_criteria:Fe=null,...Ae}){N=this._prepare_generation_config(N,Ae);const et=Ae.decoder_input_ids??this._retrieve_init_tokens(N);if(N.return_timestamps&&(_e??(_e=new M.LogitsProcessorList),_e.push(new M.WhisperTimeStampLogitsProcessor(N,et))),N.begin_suppress_tokens&&(_e??(_e=new M.LogitsProcessorList),_e.push(new M.SuppressTokensAtBeginLogitsProcessor(N.begin_suppress_tokens,et.length))),N.return_token_timestamps){if(!N.alignment_heads)throw new Error("Model generation config has no `alignment_heads`, token-level timestamps not available. See https://gist.github.com/hollance/42e32852f24243b748ae6bc1f985b13a on how to add this property to the generation config.");N.task==="translate"&&console.warn("Token-level timestamps may not be reliable for task 'translate'."),N.output_attentions=!0,N.return_dict_in_generate=!0}const rt=await super.generate({inputs:T,generation_config:N,logits_processor:_e,decoder_input_ids:et,...Ae});return N.return_token_timestamps&&(rt.token_timestamps=this._extract_token_timestamps(rt,N.alignment_heads,N.num_frames)),rt}_extract_token_timestamps(T,N,_e=null,Fe=.02){if(!T.cross_attentions)throw new Error("Model outputs must contain cross attentions to extract timestamps. This is most likely because the model was not exported with `output_attentions=True`.");_e==null&&console.warn("`num_frames` has not been set, meaning the entire audio will be analyzed. This may lead to inaccurate token-level timestamps for short audios (< 30 seconds).");let Ae=this.config.median_filter_width;Ae===void 0&&(console.warn("Model config has no `median_filter_width`, using default value of 7."),Ae=7);const et=T.cross_attentions,rt=Array.from({length:this.config.decoder_layers},(ns,Jt)=>(0,b.cat)(et.map(os=>os[Jt]),2)),_t=(0,b.stack)(N.map(([ns,Jt])=>{if(ns>=rt.length)throw new Error(`Layer index ${ns} is out of bounds for cross attentions (length ${rt.length}).`);return _e?rt[ns].slice(null,Jt,null,[0,_e]):rt[ns].slice(null,Jt)})).transpose(1,0,2,3),[Mt,jt]=(0,b.std_mean)(_t,-2,0,!0),Vt=_t.clone();for(let ns=0;nsos[Zs+1]-os[Zs]),cs=(0,R.mergeArrays)([1],xs).map(Is=>!!Is),Es=[];for(let Is=0;IsLt.findIndex(Gt=>Gt==Ae)),_t=rt.every(Lt=>Lt===-1),Mt=rt.every(Lt=>Lt!==-1);if(!_t&&!Mt)throw new Error("Every input should contain either 0 or 1 image token.");if(_t)return{inputs_embeds:T,attention_mask:Fe};const jt=[],Vt=[];for(let Lt=0;LtArray.from({length:T.dims[0]},xs=>Array.from({length:T.dims[1]},cs=>1))),ts=N?N.tolist():[],ns=_e?_e.tolist():[];let Jt=0,os=0;for(let As=0;AsLt[As][Rs]==1),Es=xs.reduce((Ts,Rs,qr)=>(Rs==_t&&Ts.push(qr),Ts),[]).map(Ts=>xs[Ts+1]),Is=Es.filter(Ts=>Ts==et).length,Zs=Es.filter(Ts=>Ts==rt).length;let Ys=[],br=0,pn=Is,Sa=Zs;for(let Ts=0;Tsmr>br&&Qr==et),qr=xs.findIndex((Qr,mr)=>mr>br&&Qr==rt),En=pn>0&&Rs!==-1?Rs:xs.length+1,Cn=Sa>0&&qr!==-1?qr:xs.length+1;let kn,Aa,Ia,Cc;En0?(0,K.max)(Ys.at(-1))[0]+1:0;Ys.push(Array.from({length:3*zr},(Qr,mr)=>kc+mr%zr));const Qn=zr+kc,Xn=kp*Fi*qn,Sc=Array.from({length:Xn},(Qr,mr)=>Qn+Math.floor(mr/(Fi*qn))),$c=Array.from({length:Xn},(Qr,mr)=>Qn+Math.floor(mr/qn)%Fi),Ac=Array.from({length:Xn},(Qr,mr)=>Qn+mr%qn);Ys.push([Sc,$c,Ac].flat()),br=kn+Xn}if(br0?(0,K.max)(Ys.at(-1))[0]+1:0,Rs=xs.length-br;Ys.push(Array.from({length:3*Rs},(qr,En)=>Ts+En%Rs))}const or=Ys.reduce((Ts,Rs)=>Ts+Rs.length,0),vr=new Array(or);let Ai=0;for(let Ts=0;Ts<3;++Ts)for(let Rs=0;RsVt[Jt%Vt.length]),ts=Array.from({length:Lt[0]},(ns,Jt)=>(0,K.max)(Vt.subarray(Lt[1]*Jt,Lt[1]*(Jt+1)))[0]+1n+BigInt(Lt[1]));return[new b.Tensor("int64",Gt,[3,...Lt]),new b.Tensor("int64",ts,[ts.length,1])]}else{const[Vt,Lt]=T.dims,Gt=BigInt64Array.from({length:3*Vt*Lt},(ts,ns)=>BigInt(Math.floor(ns%Lt/Vt)));return[new b.Tensor("int64",Gt,[3,...T.dims]),(0,b.zeros)([Vt,1])]}}async encode_image({pixel_values:T,image_grid_thw:N}){return(await we(this.sessions.vision_encoder,{pixel_values:T,grid_thw:N})).image_features}_merge_input_ids_with_image_features(T){return tt({image_token_id:this.config.image_token_id,...T})}prepare_inputs_for_generation(T,N,_e){if(N.attention_mask&&!N.position_ids)if(!N.past_key_values)[N.position_ids,N.rope_deltas]=this.get_rope_index(N.input_ids,N.image_grid_thw,N.video_grid_thw,N.attention_mask);else{N.pixel_values=null;const Fe=BigInt(Object.values(N.past_key_values)[0].dims.at(-2)),Ae=N.rope_deltas.map(et=>Fe+et);N.position_ids=(0,b.stack)([Ae,Ae,Ae],0)}return N}}class Mo extends te{}class Ll extends Mo{}class zl extends Mo{}class bo extends te{}class Bl extends bo{}class Rl extends bo{}class vo extends te{}class Nl extends vo{}class jl extends vo{}class xo extends te{}class Ul extends xo{}class Vl extends xo{}class To extends te{}class Wl extends To{}class Gl extends To{}class hi extends te{}class Kl extends hi{}class Po extends hi{async _call(T){return new Qt(await super._call(T))}}class mi extends te{}class Hl extends mi{}class ql extends mi{async _call(T){return new Qt(await super._call(T))}}class Ql extends te{}class Xl extends Ql{}class Eo extends te{}class Yl extends Eo{}class Co extends Eo{async _call(T){return new Qt(await super._call(T))}}class Jl extends te{}class Zl extends Jl{}class ko extends te{}class Gc extends ko{}class eu extends ko{async _call(T){return new Qt(await super._call(T))}}class tu extends te{}class Mr extends tu{}class So extends te{}class su extends So{}class ru extends So{async _call(T){return new Qt(await super._call(T))}}class nu extends te{}class iu extends nu{async _call(T){return new Ec(await super._call(T))}}class $o extends te{}class ou extends $o{}class au extends $o{async _call(T){return new Qt(await super._call(T))}}class Ao extends te{}class lu extends Ao{}class Kc extends Ao{async _call(T){return new Qt(await super._call(T))}}class Io extends te{}class uu extends Io{}class du extends Io{}class Oo extends te{}class cu extends Oo{}class pu extends Oo{}class hu extends te{}class nn extends hu{}class on extends hu{async _call(T){return new Qt(await super._call(T))}}class Dr extends te{}class Fo extends Dr{}class an extends Dr{async _call(T){return new Do(await super._call(T))}}class Ks extends Dr{async _call(T){return new Lo(await super._call(T))}}class Do extends Ke{constructor({logits:T,pred_boxes:N}){super(),this.logits=T,this.pred_boxes=N}}class Lo extends Ke{constructor({logits:T,pred_boxes:N,pred_masks:_e}){super(),this.logits=T,this.pred_boxes=N,this.pred_masks=_e}}class zo extends te{}class Hc extends zo{}class Un extends zo{async _call(T){return new Bo(await super._call(T))}}class Bo extends Ke{constructor({logits:T,pred_boxes:N}){super(),this.logits=T,this.pred_boxes=N}}class fi extends te{}class mu extends fi{}class fu extends fi{async _call(T){return new Ro(await super._call(T))}}class Ro extends Do{}class _i extends te{}class _u extends _i{}class No extends _i{async _call(T){return new Qt(await super._call(T))}}class jo extends te{}class gi extends jo{}class Uo extends jo{async _call(T){return new Qt(await super._call(T))}}class Vo extends te{}class gu extends Vo{}class qc extends Vo{async _call(T){return new Qt(await super._call(T))}}class Wo extends te{}class Go extends Wo{}class Vn extends Wo{async _call(T){return new Qt(await super._call(T))}}class Ko extends te{}class Ho extends Ko{}class wu extends Ko{}class qo extends te{}class yu extends qo{}class Qc extends qo{}class Mu extends te{}class bu extends Mu{}class Qo extends te{}class vu extends Qo{}class wi extends Qo{}class xu extends Qo{}class yi extends te{}class Xo extends yi{}class Mi extends te{}class Tu extends Mi{}class Pu extends Mi{}class bi extends te{}class Xc extends bi{}class Eu extends bi{}class Yc extends te{}class Cu extends Yc{}class Yo extends te{}class ku extends Yo{}class Jo extends Yo{async _call(T){return new Qt(await super._call(T))}}class Zo extends te{}class Su extends Zo{}class ea extends Zo{async _call(T){return new Qt(await super._call(T))}}class ta extends te{}class $u extends ta{}class Jc extends ta{async _call(T){return new Qt(await super._call(T))}}class sa extends te{}class Au extends sa{}class Iu extends sa{async _call(T){return new Qt(await super._call(T))}}class Zc extends te{}class Ou extends Zc{}class ra extends te{}class Fu extends ra{}class Du extends ra{async _call(T){return new Lu(await super._call(T))}}class Lu extends Ke{constructor({logits:T,pred_boxes:N}){super(),this.logits=T,this.pred_boxes=N}}class ep extends te{}class vi extends ep{async get_image_embeddings({pixel_values:T}){return await Ie(this,{pixel_values:T})}async forward(T){if((!T.image_embeddings||!T.image_positional_embeddings)&&(T={...T,...await this.get_image_embeddings(T)}),!T.input_labels&&T.input_points){const _e=T.input_points.dims.slice(0,-1),Fe=_e.reduce((Ae,et)=>Ae*et,1);T.input_labels=new b.Tensor("int64",new BigInt64Array(Fe).fill(1n),_e)}const N={image_embeddings:T.image_embeddings,image_positional_embeddings:T.image_positional_embeddings};return T.input_points&&(N.input_points=T.input_points),T.input_labels&&(N.input_labels=T.input_labels),T.input_boxes&&(N.input_boxes=T.input_boxes),await we(this.sessions.prompt_encoder_mask_decoder,N)}async _call(T){return new Wn(await super._call(T))}}class Wn extends Ke{constructor({iou_scores:T,pred_masks:N}){super(),this.iou_scores=T,this.pred_masks=N}}class xi extends te{}class zu extends xi{}class Bu extends xi{}class na extends te{}class Ru extends na{}class ia extends na{}class Hr extends te{}class Nu extends Hr{}class ju extends Hr{async _call(T){return new cn(await super._call(T))}}class tp extends Hr{async _call(T){return new Qt(await super._call(T))}}class Uu extends Hr{async _call(T){return new Hs(await super._call(T))}}class Ti extends te{}class Vu extends Ti{}class Wu extends Ti{async _call(T){return new Hs(await super._call(T))}}class Gu extends te{}class sp extends Gu{}class Pi extends te{}class Ku extends Pi{}class rp extends Pi{async _call(T){return new cn(await super._call(T))}}class Hu extends Pi{async _call(T){return new Qt(await super._call(T))}}class Gn extends te{}class qu extends Gn{}class Qu extends Gn{async _call(T){return new cn(await super._call(T))}}class np extends Gn{async _call(T){return new Qt(await super._call(T))}}class Xu extends Gn{async _call(T){return new Hs(await super._call(T))}}class Ei extends te{}class Yu extends Ei{}class ip extends Ei{async _call(T){return new cn(await super._call(T))}}class Ju extends Ei{async _call(T){return new Qt(await super._call(T))}}class op extends te{}class Zu extends Hr{}class ed extends Hr{async _call(T){return new cn(await super._call(T))}}class ap extends Hr{async _call(T){return new Qt(await super._call(T))}}class Pn extends te{}class td extends Pn{}class sd extends Pn{async _call(T){return new cn(await super._call(T))}}class rd extends Pn{async _call(T){return new Qt(await super._call(T))}}class lp extends Pn{async _call(T){return new Hn(await super._call(T))}}class nd extends Pn{async _call(T){return new Hs(await super._call(T))}}class id extends te{}class od extends id{}class Ci extends te{}class Kp extends Ci{}class Cr extends Ci{}class Lr extends Ci{async generate_speech(T,N,{threshold:_e=.5,minlenratio:Fe=0,maxlenratio:Ae=20,vocoder:et=null}={}){const rt={input_ids:T},{encoder_outputs:_t,encoder_attention_mask:Mt}=await Ie(this,rt),jt=_t.dims[1]/this.config.reduction_factor,Vt=Math.floor(jt*Ae),Lt=Math.floor(jt*Fe),Gt=this.config.num_mel_bins;let ts=[],ns=null,Jt=null,os=0;for(;;){++os;const cs=ce(!!Jt);let Es;Jt?Es=Jt.output_sequence_out:Es=new b.Tensor("float32",new Float32Array(Gt),[1,1,Gt]);let Is={use_cache_branch:cs,output_sequence:Es,encoder_attention_mask:Mt,speaker_embeddings:N,encoder_hidden_states:_t};this.addPastKeyValues(Is,ns),Jt=await we(this.sessions.decoder_model_merged,Is),ns=this.getPastKeyValues(Jt,ns);const{prob:Zs,spectrum:Ys}=Jt;if(ts.push(Ys),os>=Lt&&(Array.from(Zs.data).filter(br=>br>=_e).length>0||os>=Vt))break}const As=(0,b.cat)(ts),{waveform:xs}=await we(et.sessions.model,{spectrogram:As});return{spectrogram:As,waveform:xs}}}class ln extends te{constructor(){super(...arguments);me(this,"main_input_name","spectrogram")}}class un extends te{}class ad extends un{}class oa extends te{}class ld extends oa{}class ud extends oa{}class aa extends te{}class dd extends aa{}class cd extends aa{}class la extends te{}class pd extends la{}class hd extends la{}class ua extends te{}class nr extends ua{}class md extends ua{static async from_pretrained(T,N={}){return super.from_pretrained(T,{...N,model_file_name:N.model_file_name??"text_model"})}}class fd extends ua{static async from_pretrained(T,N={}){return super.from_pretrained(T,{...N,model_file_name:N.model_file_name??"audio_model"})}}class da extends te{}class ca extends da{async _call(T){return new Cp(await super._call(T))}}class dn extends te{}class up extends dn{}class _d extends dn{}class gd extends dn{}class pa extends te{}class wd extends pa{}class yd extends pa{}class ki extends te{}class Md extends ki{}class bd extends ki{async _call(T){return new Qt(await super._call(T))}}class ha extends te{}class dp extends ha{}class cp extends ha{}class Si extends te{constructor(){super(...arguments);me(this,"forward_params",["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"])}_apply_and_filter_by_delay_pattern_mask(N){const[_e,Fe]=N.dims,Ae=this.config.decoder.num_codebooks,et=Fe-Ae;let rt=0;for(let jt=0;jt0&&Gt<=et&&(N.data[rt++]=N.data[jt])}const _t=Math.floor(_e/Ae),Mt=rt/(_t*Ae);return new b.Tensor(N.type,N.data.slice(0,rt),[_t,Ae,Mt])}prepare_inputs_for_generation(N,_e,Fe){let Ae=structuredClone(N);for(let rt=0;rt=_t&&(Ae[rt][_t]=BigInt(this.config.decoder.pad_token_id));return Fe.guidance_scale!==null&&Fe.guidance_scale>1&&(Ae=Ae.concat(Ae)),super.prepare_inputs_for_generation(Ae,_e,Fe)}async generate(N){const _e=await super.generate(N),Fe=this._apply_and_filter_by_delay_pattern_mask(_e).unsqueeze_(0),{audio_values:Ae}=await we(this.sessions.encodec_decode,{audio_codes:Fe});return Ae}}class ma extends te{}class pp extends ma{}class fa extends ma{async _call(T){return new Qt(await super._call(T))}}class _a extends te{}class vd extends _a{}class xd extends _a{async _call(T){return new Qt(await super._call(T))}}class Td extends te{}class Pd extends Td{}class Ed extends Td{async _call(T){return new Qt(await super._call(T))}}class ga extends te{}class hp extends ga{}class Cd extends ga{async _call(T){return new Qt(await super._call(T))}}class kd extends te{}class mp extends kd{}class Sd extends te{}class $d extends Sd{constructor(...N){super(...N);me(this,"forward_params",["input_ids","pixel_values","images_seq_mask","images_emb_mask","attention_mask","position_ids","past_key_values"]);this._generation_mode="text"}async forward(N){const _e=this._generation_mode??"text";let Fe;if(_e==="text"||!N.past_key_values){const Mt=this.sessions.prepare_inputs_embeds,jt=(0,R.pick)(N,Mt.inputNames);Fe=await we(Mt,jt)}else{const Mt=this.sessions.gen_img_embeds,jt=(0,R.pick)({image_ids:N.input_ids},Mt.inputNames);Fe=await we(Mt,jt)}const Ae={...N,...Fe},et=await Ee(this,Ae),rt=this.sessions[_e==="text"?"lm_head":"gen_head"];if(!rt)throw new Error(`Unable to find "${rt}" generation head`);const _t=await we(rt,(0,R.pick)(et,rt.inputNames));return{...Fe,...et,..._t}}async generate(N){return this._generation_mode="text",super.generate(N)}async generate_images(N){this._generation_mode="image";const _e=(N.inputs??N[this.main_input_name]).dims[1],Ae=(await super.generate(N)).slice(null,[_e,null]),et=this.sessions.image_decode,{decoded_image:rt}=await we(et,{generated_tokens:Ae}),_t=rt.add_(1).mul_(255/2).clamp_(0,255).to("uint8"),Mt=[];for(const jt of _t){const Vt=I.RawImage.fromTensor(jt);Mt.push(Vt)}return Mt}}class Ad extends Ke{constructor({char_logits:T,bpe_logits:N,wp_logits:_e}){super(),this.char_logits=T,this.bpe_logits=N,this.wp_logits=_e}get logits(){return[this.char_logits,this.bpe_logits,this.wp_logits]}}class Id extends te{}class Od extends Id{async _call(T){return new Ad(await super._call(T))}}class Fd extends te{}class Dd extends Fd{}class Ld extends Fd{}class wa extends te{}class zd extends wa{}class Bd extends wa{}class ws{static async from_pretrained(T,{progress_callback:N=null,config:_e=null,cache_dir:Fe=null,local_files_only:Ae=!1,revision:et="main",model_file_name:rt=null,subfolder:_t="onnx",device:Mt=null,dtype:jt=null,use_external_data_format:Vt=null,session_options:Lt={}}={}){const Gt={progress_callback:N,config:_e,cache_dir:Fe,local_files_only:Ae,revision:et,model_file_name:rt,subfolder:_t,device:Mt,dtype:jt,use_external_data_format:Vt,session_options:Lt};if(Gt.config=await _.AutoConfig.from_pretrained(T,Gt),!this.MODEL_CLASS_MAPPINGS)throw new Error("`MODEL_CLASS_MAPPINGS` not implemented for this type of `AutoClass`: "+this.name);for(const ts of this.MODEL_CLASS_MAPPINGS){const ns=ts.get(Gt.config.model_type);if(ns)return await ns[1].from_pretrained(T,Gt)}if(this.BASE_IF_FAIL)return console.warn(`Unknown model class "${Gt.config.model_type}", attempting to construct from base class.`),await te.from_pretrained(T,Gt);throw Error(`Unsupported model type: ${Gt.config.model_type}`)}}me(ws,"MODEL_CLASS_MAPPINGS",null),me(ws,"BASE_IF_FAIL",!1);const fp=new Map([["bert",["BertModel",Te]],["modernbert",["ModernBertModel",dt]],["nomic_bert",["NomicBertModel",oe]],["roformer",["RoFormerModel",fe]],["electra",["ElectraModel",Ds]],["esm",["EsmModel",ii]],["convbert",["ConvBertModel",vt]],["camembert",["CamembertModel",Nt]],["deberta",["DebertaModel",en]],["deberta-v2",["DebertaV2Model",Tt]],["mpnet",["MPNetModel",yn]],["albert",["AlbertModel",Bn]],["distilbert",["DistilBertModel",lr]],["roberta",["RobertaModel",ir]],["xlm",["XLMModel",Ls]],["xlm-roberta",["XLMRobertaModel",It]],["clap",["ClapModel",nr]],["clip",["CLIPModel",Ja]],["clipseg",["CLIPSegModel",nl]],["chinese_clip",["ChineseCLIPModel",gr]],["siglip",["SiglipModel",tl]],["jina_clip",["JinaCLIPModel",ui]],["mobilebert",["MobileBertModel",Fn]],["squeezebert",["SqueezeBertModel",xn]],["wav2vec2",["Wav2Vec2Model",Nu]],["wav2vec2-bert",["Wav2Vec2BertModel",Yu]],["unispeech",["UniSpeechModel",Ku]],["unispeech-sat",["UniSpeechSatModel",qu]],["hubert",["HubertModel",Zu]],["wavlm",["WavLMModel",td]],["audio-spectrogram-transformer",["ASTModel",Na]],["vits",["VitsModel",ca]],["pyannote",["PyAnnoteModel",Vu]],["wespeaker-resnet",["WeSpeakerResNetModel",sp]],["detr",["DetrModel",Fo]],["rt_detr",["RTDetrModel",Hc]],["table-transformer",["TableTransformerModel",mu]],["vit",["ViTModel",Kl]],["ijepa",["IJepaModel",Hl]],["pvt",["PvtModel",Yl]],["vit_msn",["ViTMSNModel",Gc]],["vit_mae",["ViTMAEModel",Zl]],["groupvit",["GroupViTModel",Mr]],["fastvit",["FastViTModel",su]],["mobilevit",["MobileViTModel",ou]],["mobilevitv2",["MobileViTV2Model",lu]],["owlvit",["OwlViTModel",uu]],["owlv2",["Owlv2Model",cu]],["beit",["BeitModel",nn]],["deit",["DeiTModel",_u]],["hiera",["HieraModel",gi]],["convnext",["ConvNextModel",ku]],["convnextv2",["ConvNextV2Model",Su]],["dinov2",["Dinov2Model",$u]],["dinov2_with_registers",["Dinov2WithRegistersModel",Au]],["resnet",["ResNetModel",gu]],["swin",["SwinModel",Go]],["swin2sr",["Swin2SRModel",Ho]],["donut-swin",["DonutSwinModel",Cu]],["yolos",["YolosModel",Fu]],["dpt",["DPTModel",yu]],["glpn",["GLPNModel",Xc]],["hifigan",["SpeechT5HifiGan",ln]],["efficientnet",["EfficientNetModel",Md]],["decision_transformer",["DecisionTransformerModel",mp]],["patchtst",["PatchTSTForPrediction",Dd]],["patchtsmixer",["PatchTSMixerForPrediction",zd]],["mobilenet_v1",["MobileNetV1Model",pp]],["mobilenet_v2",["MobileNetV2Model",vd]],["mobilenet_v3",["MobileNetV3Model",Pd]],["mobilenet_v4",["MobileNetV4Model",hp]],["maskformer",["MaskFormerModel",Tu]],["mgp-str",["MgpstrForSceneTextRecognition",Od]],["style_text_to_speech_2",["StyleTextToSpeech2Model",od]]]),_p=new Map([["t5",["T5Model",P]],["longt5",["LongT5Model",be]],["mt5",["MT5Model",pt]],["bart",["BartModel",xt]],["mbart",["MBartModel",Fs]],["marian",["MarianModel",zu]],["whisper",["WhisperModel",ja]],["m2m_100",["M2M100Model",Ru]],["blenderbot",["BlenderbotModel",ze]],["blenderbot-small",["BlenderbotSmallModel",ks]]]),gp=new Map([["bloom",["BloomModel",Nl]],["jais",["JAISModel",ll]],["gpt2",["GPT2Model",ol]],["gptj",["GPTJModel",hl]],["gpt_bigcode",["GPTBigCodeModel",fl]],["gpt_neo",["GPTNeoModel",yr]],["gpt_neox",["GPTNeoXModel",cl]],["codegen",["CodeGenModel",ro]],["llama",["LlamaModel",io]],["exaone",["ExaoneModel",Ml]],["olmo",["OlmoModel",Vc]],["olmo2",["Olmo2Model",Tl]],["mobilellm",["MobileLLMModel",bl]],["granite",["GraniteModel",ds]],["cohere",["CohereModel",El]],["gemma",["GemmaModel",kl]],["gemma2",["Gemma2Model",$l]],["helium",["HeliumModel",ci]],["glm",["GlmModel",yl]],["openelm",["OpenELMModel",Il]],["qwen2",["Qwen2Model",jn]],["phi",["PhiModel",Ll]],["phi3",["Phi3Model",Bl]],["mpt",["MptModel",Ul]],["opt",["OPTModel",Wl]],["mistral",["MistralModel",ld]],["starcoder2",["Starcoder2Model",dd]],["falcon",["FalconModel",pd]],["stablelm",["StableLmModel",wd]]]),Rd=new Map([["speecht5",["SpeechT5ForSpeechToText",Cr]],["whisper",["WhisperForConditionalGeneration",Ua]],["moonshine",["MoonshineForConditionalGeneration",Va]]]),Kn=new Map([["speecht5",["SpeechT5ForTextToSpeech",Lr]]]),ya=new Map([["vits",["VitsModel",ca]],["musicgen",["MusicgenForConditionalGeneration",Si]]]),Ma=new Map([["bert",["BertForSequenceClassification",Ve]],["modernbert",["ModernBertForSequenceClassification",lt]],["roformer",["RoFormerForSequenceClassification",We]],["electra",["ElectraForSequenceClassification",Sr]],["esm",["EsmForSequenceClassification",In]],["convbert",["ConvBertForSequenceClassification",At]],["camembert",["CamembertForSequenceClassification",$r]],["deberta",["DebertaForSequenceClassification",Rr]],["deberta-v2",["DebertaV2ForSequenceClassification",Vs]],["mpnet",["MPNetForSequenceClassification",Mn]],["albert",["AlbertForSequenceClassification",Tn]],["distilbert",["DistilBertForSequenceClassification",Os]],["roberta",["RobertaForSequenceClassification",fs]],["xlm",["XLMForSequenceClassification",Gs]],["xlm-roberta",["XLMRobertaForSequenceClassification",Ui]],["bart",["BartForSequenceClassification",ms]],["mbart",["MBartForSequenceClassification",rs]],["mobilebert",["MobileBertForSequenceClassification",Wr]],["squeezebert",["SqueezeBertForSequenceClassification",Ln]]]),ba=new Map([["bert",["BertForTokenClassification",Ne]],["modernbert",["ModernBertForTokenClassification",ht]],["roformer",["RoFormerForTokenClassification",Je]],["electra",["ElectraForTokenClassification",Yr]],["esm",["EsmForTokenClassification",On]],["convbert",["ConvBertForTokenClassification",is]],["camembert",["CamembertForTokenClassification",Ar]],["deberta",["DebertaForTokenClassification",Nr]],["deberta-v2",["DebertaV2ForTokenClassification",jr]],["mpnet",["MPNetForTokenClassification",bn]],["distilbert",["DistilBertForTokenClassification",Er]],["roberta",["RobertaForTokenClassification",Ss]],["xlm",["XLMForTokenClassification",$t]],["xlm-roberta",["XLMRobertaForTokenClassification",Ba]]]),$i=new Map([["t5",["T5ForConditionalGeneration",Q]],["longt5",["LongT5ForConditionalGeneration",Se]],["mt5",["MT5ForConditionalGeneration",gt]],["bart",["BartForConditionalGeneration",Kt]],["mbart",["MBartForConditionalGeneration",Bt]],["marian",["MarianMTModel",Bu]],["m2m_100",["M2M100ForConditionalGeneration",ia]],["blenderbot",["BlenderbotForConditionalGeneration",Js]],["blenderbot-small",["BlenderbotSmallForConditionalGeneration",Xs]]]),va=new Map([["bloom",["BloomForCausalLM",jl]],["gpt2",["GPT2LMHeadModel",al]],["jais",["JAISLMHeadModel",ul]],["gptj",["GPTJForCausalLM",ml]],["gpt_bigcode",["GPTBigCodeForCausalLM",_l]],["gpt_neo",["GPTNeoForCausalLM",dl]],["gpt_neox",["GPTNeoXForCausalLM",pl]],["codegen",["CodeGenForCausalLM",gl]],["llama",["LlamaForCausalLM",Uc]],["exaone",["ExaoneForCausalLM",uo]],["olmo",["OlmoForCausalLM",xl]],["olmo2",["Olmo2ForCausalLM",Wc]],["mobilellm",["MobileLLMForCausalLM",vl]],["granite",["GraniteForCausalLM",Pl]],["cohere",["CohereForCausalLM",Cl]],["gemma",["GemmaForCausalLM",Sl]],["gemma2",["Gemma2ForCausalLM",Al]],["helium",["HeliumForCausalLM",wl]],["glm",["GlmForCausalLM",Nn]],["openelm",["OpenELMForCausalLM",Ol]],["qwen2",["Qwen2ForCausalLM",Fl]],["phi",["PhiForCausalLM",zl]],["phi3",["Phi3ForCausalLM",Rl]],["mpt",["MptForCausalLM",Vl]],["opt",["OPTForCausalLM",Gl]],["mbart",["MBartForCausalLM",rr]],["mistral",["MistralForCausalLM",ud]],["starcoder2",["Starcoder2ForCausalLM",cd]],["falcon",["FalconForCausalLM",hd]],["trocr",["TrOCRForCausalLM",ad]],["stablelm",["StableLmForCausalLM",yd]],["phi3_v",["Phi3VForCausalLM",hr]]]),wp=new Map([["multi_modality",["MultiModalityCausalLM",$d]]]),xa=new Map([["bert",["BertForMaskedLM",Ue]],["modernbert",["ModernBertForMaskedLM",ct]],["roformer",["RoFormerForMaskedLM",$e]],["electra",["ElectraForMaskedLM",sr]],["esm",["EsmForMaskedLM",An]],["convbert",["ConvBertForMaskedLM",kt]],["camembert",["CamembertForMaskedLM",Jr]],["deberta",["DebertaForMaskedLM",Ir]],["deberta-v2",["DebertaV2ForMaskedLM",Dt]],["mpnet",["MPNetForMaskedLM",tn]],["albert",["AlbertForMaskedLM",as]],["distilbert",["DistilBertForMaskedLM",wn]],["roberta",["RobertaForMaskedLM",_r]],["xlm",["XLMWithLMHeadModel",$s]],["xlm-roberta",["XLMRobertaForMaskedLM",za]],["mobilebert",["MobileBertForMaskedLM",oi]],["squeezebert",["SqueezeBertForMaskedLM",Dn]]]),Ta=new Map([["bert",["BertForQuestionAnswering",Re]],["roformer",["RoFormerForQuestionAnswering",ut]],["electra",["ElectraForQuestionAnswering",Us]],["convbert",["ConvBertForQuestionAnswering",ys]],["camembert",["CamembertForQuestionAnswering",Zr]],["deberta",["DebertaForQuestionAnswering",ar]],["deberta-v2",["DebertaV2ForQuestionAnswering",Or]],["mpnet",["MPNetForQuestionAnswering",vn]],["albert",["AlbertForQuestionAnswering",Rn]],["distilbert",["DistilBertForQuestionAnswering",es]],["roberta",["RobertaForQuestionAnswering",yt]],["xlm",["XLMForQuestionAnswering",sn]],["xlm-roberta",["XLMRobertaForQuestionAnswering",Ra]],["mobilebert",["MobileBertForQuestionAnswering",fr]],["squeezebert",["SqueezeBertForQuestionAnswering",zn]]]),Pa=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",Ki]],["idefics3",["Idefics3ForConditionalGeneration",Hi]]]),yp=new Map([["llava",["LlavaForConditionalGeneration",ai]],["llava_onevision",["LlavaOnevisionForConditionalGeneration",Wa]],["moondream1",["Moondream1ForConditionalGeneration",Ga]],["florence2",["Florence2ForConditionalGeneration",Ha]],["qwen2-vl",["Qwen2VLForConditionalGeneration",Dl]],["idefics3",["Idefics3ForConditionalGeneration",Hi]],["paligemma",["PaliGemmaForConditionalGeneration",Qa]]]),Nd=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",Ki]]]),jd=new Map([["vit",["ViTForImageClassification",Po]],["ijepa",["IJepaForImageClassification",ql]],["pvt",["PvtForImageClassification",Co]],["vit_msn",["ViTMSNForImageClassification",eu]],["fastvit",["FastViTForImageClassification",ru]],["mobilevit",["MobileViTForImageClassification",au]],["mobilevitv2",["MobileViTV2ForImageClassification",Kc]],["beit",["BeitForImageClassification",on]],["deit",["DeiTForImageClassification",No]],["hiera",["HieraForImageClassification",Uo]],["convnext",["ConvNextForImageClassification",Jo]],["convnextv2",["ConvNextV2ForImageClassification",ea]],["dinov2",["Dinov2ForImageClassification",Jc]],["dinov2_with_registers",["Dinov2WithRegistersForImageClassification",Iu]],["resnet",["ResNetForImageClassification",qc]],["swin",["SwinForImageClassification",Vn]],["segformer",["SegformerForImageClassification",_d]],["efficientnet",["EfficientNetForImageClassification",bd]],["mobilenet_v1",["MobileNetV1ForImageClassification",fa]],["mobilenet_v2",["MobileNetV2ForImageClassification",xd]],["mobilenet_v3",["MobileNetV3ForImageClassification",Ed]],["mobilenet_v4",["MobileNetV4ForImageClassification",Cd]]]),Ud=new Map([["detr",["DetrForObjectDetection",an]],["rt_detr",["RTDetrForObjectDetection",Un]],["table-transformer",["TableTransformerForObjectDetection",fu]],["yolos",["YolosForObjectDetection",Du]]]),Ea=new Map([["owlvit",["OwlViTForObjectDetection",du]],["owlv2",["Owlv2ForObjectDetection",pu]],["grounding-dino",["GroundingDinoForObjectDetection",Ou]]]),Vd=new Map([["detr",["DetrForSegmentation",Ks]],["clipseg",["CLIPSegForImageSegmentation",il]]]),Wd=new Map([["segformer",["SegformerForSemanticSegmentation",gd]],["sapiens",["SapiensForSemanticSegmentation",vu]]]),Gd=new Map([["detr",["DetrForSegmentation",Ks]],["maskformer",["MaskFormerForInstanceSegmentation",Pu]]]),Kd=new Map([["sam",["SamModel",vi]]]),Mp=new Map([["wav2vec2",["Wav2Vec2ForCTC",ju]],["wav2vec2-bert",["Wav2Vec2BertForCTC",ip]],["unispeech",["UniSpeechForCTC",rp]],["unispeech-sat",["UniSpeechSatForCTC",Qu]],["wavlm",["WavLMForCTC",sd]],["hubert",["HubertForCTC",ed]]]),Hd=new Map([["wav2vec2",["Wav2Vec2ForSequenceClassification",tp]],["wav2vec2-bert",["Wav2Vec2BertForSequenceClassification",Ju]],["unispeech",["UniSpeechForSequenceClassification",Hu]],["unispeech-sat",["UniSpeechSatForSequenceClassification",np]],["wavlm",["WavLMForSequenceClassification",rd]],["hubert",["HubertForSequenceClassification",ap]],["audio-spectrogram-transformer",["ASTForAudioClassification",Vi]]]),qd=new Map([["wavlm",["WavLMForXVector",lp]]]),Qd=new Map([["unispeech-sat",["UniSpeechSatForAudioFrameClassification",Xu]],["wavlm",["WavLMForAudioFrameClassification",nd]],["wav2vec2",["Wav2Vec2ForAudioFrameClassification",Uu]],["pyannote",["PyAnnoteForAudioFrameClassification",Wu]]]),Xd=new Map([["vitmatte",["VitMatteForImageMatting",iu]]]),Hp=new Map([["patchtst",["PatchTSTForPrediction",Ld]],["patchtsmixer",["PatchTSMixerForPrediction",Bd]]]),Yd=new Map([["swin2sr",["Swin2SRForImageSuperResolution",wu]]]),Jd=new Map([["dpt",["DPTForDepthEstimation",Qc]],["depth_anything",["DepthAnythingForDepthEstimation",bu]],["glpn",["GLPNForDepthEstimation",Eu]],["sapiens",["SapiensForDepthEstimation",wi]],["depth_pro",["DepthProForDepthEstimation",Xo]]]),Zd=new Map([["sapiens",["SapiensForNormalEstimation",xu]]]),bp=new Map([["vitpose",["VitPoseForPoseEstimation",Xl]]]),ec=new Map([["clip",["CLIPVisionModelWithProjection",el]],["siglip",["SiglipVisionModel",rl]],["jina_clip",["JinaCLIPVisionModel",wr]]]),tc=[[fp,$.EncoderOnly],[_p,$.EncoderDecoder],[gp,$.DecoderOnly],[Ma,$.EncoderOnly],[ba,$.EncoderOnly],[$i,$.Seq2Seq],[Rd,$.Seq2Seq],[va,$.DecoderOnly],[wp,$.MultiModality],[xa,$.EncoderOnly],[Ta,$.EncoderOnly],[Pa,$.Vision2Seq],[yp,$.ImageTextToText],[jd,$.EncoderOnly],[Vd,$.EncoderOnly],[Gd,$.EncoderOnly],[Wd,$.EncoderOnly],[Xd,$.EncoderOnly],[Hp,$.EncoderOnly],[Yd,$.EncoderOnly],[Jd,$.EncoderOnly],[Zd,$.EncoderOnly],[bp,$.EncoderOnly],[Ud,$.EncoderOnly],[Ea,$.EncoderOnly],[Kd,$.MaskGeneration],[Mp,$.EncoderOnly],[Hd,$.EncoderOnly],[Kn,$.Seq2Seq],[ya,$.EncoderOnly],[qd,$.EncoderOnly],[Qd,$.EncoderOnly],[ec,$.EncoderOnly]];for(const[f,T]of tc)for(const[N,_e]of f.values())S.set(N,T),x.set(_e,N),w.set(N,_e);const vp=[["MusicgenForConditionalGeneration",Si,$.Musicgen],["Phi3VForCausalLM",hr,$.Phi3V],["CLIPTextModelWithProjection",Za,$.EncoderOnly],["SiglipTextModel",sl,$.EncoderOnly],["JinaCLIPTextModel",Qi,$.EncoderOnly],["ClapTextModelWithProjection",md,$.EncoderOnly],["ClapAudioModelWithProjection",fd,$.EncoderOnly]];for(const[f,T,N]of vp)S.set(f,N),x.set(T,f),w.set(f,T);class Ca extends ws{}me(Ca,"MODEL_CLASS_MAPPINGS",tc.map(T=>T[0])),me(Ca,"BASE_IF_FAIL",!0);class xp extends ws{}me(xp,"MODEL_CLASS_MAPPINGS",[Ma]);class sc extends ws{}me(sc,"MODEL_CLASS_MAPPINGS",[ba]);class rc extends ws{}me(rc,"MODEL_CLASS_MAPPINGS",[$i]);class nc extends ws{}me(nc,"MODEL_CLASS_MAPPINGS",[Rd]);class ic extends ws{}me(ic,"MODEL_CLASS_MAPPINGS",[Kn]);class oc extends ws{}me(oc,"MODEL_CLASS_MAPPINGS",[ya]);class ac extends ws{}me(ac,"MODEL_CLASS_MAPPINGS",[va]);class lc extends ws{}me(lc,"MODEL_CLASS_MAPPINGS",[xa]);class uc extends ws{}me(uc,"MODEL_CLASS_MAPPINGS",[Ta]);class dc extends ws{}me(dc,"MODEL_CLASS_MAPPINGS",[Pa]);class cc extends ws{}me(cc,"MODEL_CLASS_MAPPINGS",[jd]);class pc extends ws{}me(pc,"MODEL_CLASS_MAPPINGS",[Vd]);class ka extends ws{}me(ka,"MODEL_CLASS_MAPPINGS",[Wd]);class hc extends ws{}me(hc,"MODEL_CLASS_MAPPINGS",[Gd]);class mc extends ws{}me(mc,"MODEL_CLASS_MAPPINGS",[Ud]);class fc extends ws{}me(fc,"MODEL_CLASS_MAPPINGS",[Ea]);class _c extends ws{}me(_c,"MODEL_CLASS_MAPPINGS",[Kd]);class gc extends ws{}me(gc,"MODEL_CLASS_MAPPINGS",[Mp]);class wc extends ws{}me(wc,"MODEL_CLASS_MAPPINGS",[Hd]);class yc extends ws{}me(yc,"MODEL_CLASS_MAPPINGS",[qd]);class Tp extends ws{}me(Tp,"MODEL_CLASS_MAPPINGS",[Qd]);class Mc extends ws{}me(Mc,"MODEL_CLASS_MAPPINGS",[Nd]);class bc extends ws{}me(bc,"MODEL_CLASS_MAPPINGS",[Xd]);class vc extends ws{}me(vc,"MODEL_CLASS_MAPPINGS",[Yd]);class Pp extends ws{}me(Pp,"MODEL_CLASS_MAPPINGS",[Jd]);class xc extends ws{}me(xc,"MODEL_CLASS_MAPPINGS",[Zd]);class Tc extends ws{}me(Tc,"MODEL_CLASS_MAPPINGS",[bp]);class Pc extends ws{}me(Pc,"MODEL_CLASS_MAPPINGS",[ec]);class qp extends Ke{constructor({logits:T,past_key_values:N,encoder_outputs:_e,decoder_attentions:Fe=null,cross_attentions:Ae=null}){super(),this.logits=T,this.past_key_values=N,this.encoder_outputs=_e,this.decoder_attentions=Fe,this.cross_attentions=Ae}}class Qt extends Ke{constructor({logits:T,...N}){super(),this.logits=T;const _e=Object.values(N);_e.length>0&&(this.attentions=_e)}}class Hn extends Ke{constructor({logits:T,embeddings:N}){super(),this.logits=T,this.embeddings=N}}class Hs extends Ke{constructor({logits:T}){super(),this.logits=T}}class qs extends Ke{constructor({logits:T}){super(),this.logits=T}}class tr extends Ke{constructor({start_logits:T,end_logits:N}){super(),this.start_logits=T,this.end_logits=N}}class cn extends Ke{constructor({logits:T}){super(),this.logits=T}}class Ep extends Ke{constructor({logits:T,past_key_values:N}){super(),this.logits=T,this.past_key_values=N}}class Ec extends Ke{constructor({alphas:T}){super(),this.alphas=T}}class Cp extends Ke{constructor({waveform:T,spectrogram:N}){super(),this.waveform=T,this.spectrogram=N}}},"./src/models/audio_spectrogram_transformer/feature_extraction_audio_spectrogram_transformer.js":(Le,A,r)=>{r.r(A),r.d(A,{ASTFeatureExtractor:()=>j});var _=r("./src/base/feature_extraction_utils.js");r("./src/utils/tensor.js");var D=r("./src/utils/audio.js");class j extends _.FeatureExtractor{constructor(R){super(R);const g=this.config.sampling_rate,v=(0,D.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(g/2),g,null,"kaldi",!0);for(let M=0;M{r.r(A),r.d(A,{AutoFeatureExtractor:()=>X});var _=r("./src/utils/constants.js"),D=r("./src/utils/hub.js");r("./src/base/feature_extraction_utils.js");var j=r("./src/models/feature_extractors.js");class X{static async from_pretrained(g,v={}){const M=await(0,D.getModelJSON)(g,_.FEATURE_EXTRACTOR_NAME,!0,v),y=M.feature_extractor_type,b=j[y];if(!b)throw new Error(`Unknown feature_extractor_type: '${y}'. Please report this at ${_.GITHUB_ISSUE_URL}.`);return new b(M)}}},"./src/models/auto/image_processing_auto.js":(Le,A,r)=>{r.r(A),r.d(A,{AutoImageProcessor:()=>R});var _=r("./src/utils/constants.js"),D=r("./src/utils/hub.js"),j=r("./src/base/image_processors_utils.js"),X=r("./src/models/image_processors.js");class R{static async from_pretrained(v,M={}){const y=await(0,D.getModelJSON)(v,_.IMAGE_PROCESSOR_NAME,!0,M),b=y.image_processor_type??y.feature_extractor_type;let I=X[b];return I||(b!==void 0&&console.warn(`Image processor type '${b}' not found, assuming base ImageProcessor. Please report this at ${_.GITHUB_ISSUE_URL}.`),I=j.ImageProcessor),new I(y)}}},"./src/models/auto/processing_auto.js":(Le,A,r)=>{r.r(A),r.d(A,{AutoProcessor:()=>v});var _=r("./src/utils/constants.js"),D=r("./src/utils/hub.js"),j=r("./src/base/processing_utils.js"),X=r("./src/models/processors.js"),R=r("./src/models/image_processors.js"),g=r("./src/models/feature_extractors.js");class v{static async from_pretrained(y,b={}){const I=await(0,D.getModelJSON)(y,_.IMAGE_PROCESSOR_NAME,!0,b),{image_processor_type:K,feature_extractor_type:se,processor_class:ne}=I;if(ne&&X[ne])return X[ne].from_pretrained(y,b);if(!K&&!se)throw new Error("No `image_processor_type` or `feature_extractor_type` found in the config.");const W={};if(K){const q=R[K];if(!q)throw new Error(`Unknown image_processor_type: '${K}'.`);W.image_processor=new q(I)}if(se){const q=R[se];if(q)W.image_processor=new q(I);else{const $=g[se];if(!$)throw new Error(`Unknown feature_extractor_type: '${se}'.`);W.feature_extractor=new $(I)}}const U={};return new j.Processor(U,W)}}},"./src/models/beit/image_processing_beit.js":(Le,A,r)=>{r.r(A),r.d(A,{BeitFeatureExtractor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{}},"./src/models/bit/image_processing_bit.js":(Le,A,r)=>{r.r(A),r.d(A,{BitImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{}},"./src/models/chinese_clip/image_processing_chinese_clip.js":(Le,A,r)=>{r.r(A),r.d(A,{ChineseCLIPFeatureExtractor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{}},"./src/models/clap/feature_extraction_clap.js":(Le,A,r)=>{r.r(A),r.d(A,{ClapFeatureExtractor:()=>j});var _=r("./src/base/feature_extraction_utils.js");r("./src/utils/tensor.js");var D=r("./src/utils/audio.js");class j extends _.FeatureExtractor{constructor(R){super(R),this.mel_filters=(0,D.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,null,"htk"),this.mel_filters_slaney=(0,D.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,"slaney","slaney"),this.window=(0,D.window_function)(this.config.fft_window_size,"hann")}async _get_input_mel(R,g,v,M){let y;const b=R.length-g;if(b>0)if(v==="rand_trunc"){const I=Math.floor(Math.random()*(b+1));R=R.subarray(I,I+g),y=await this._extract_fbank_features(R,this.mel_filters_slaney,this.config.nb_max_samples)}else throw new Error(`Truncation strategy "${v}" not implemented`);else{if(b<0){let I=new Float64Array(g);if(I.set(R),M==="repeat")for(let K=R.length;K{r.r(A),r.d(A,{CLIPFeatureExtractor:()=>j,CLIPImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{}class j extends D{}},"./src/models/convnext/image_processing_convnext.js":(Le,A,r)=>{r.r(A),r.d(A,{ConvNextFeatureExtractor:()=>j,ConvNextImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{constructor(R){super(R),this.crop_pct=this.config.crop_pct??.875}async resize(R){var v;const g=(v=this.size)==null?void 0:v.shortest_edge;if(g===void 0)throw new Error("Size dictionary must contain 'shortest_edge' key.");if(g<384){const M=Math.floor(g/this.crop_pct),[y,b]=this.get_resize_output_image_size(R,{shortest_edge:M});R=await R.resize(y,b,{resample:this.resample}),R=await R.center_crop(g,g)}else R=await R.resize(g,g,{resample:this.resample});return R}}class j extends D{}},"./src/models/deit/image_processing_deit.js":(Le,A,r)=>{r.r(A),r.d(A,{DeiTFeatureExtractor:()=>j,DeiTImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{}class j extends D{}},"./src/models/detr/image_processing_detr.js":(Le,A,r)=>{r.r(A),r.d(A,{DetrFeatureExtractor:()=>X,DetrImageProcessor:()=>j});var _=r("./src/base/image_processors_utils.js"),D=r("./src/utils/tensor.js");class j extends _.ImageProcessor{async _call(g){const v=await super._call(g),M=[v.pixel_values.dims[0],64,64],y=(0,D.full)(M,1n);return{...v,pixel_mask:y}}post_process_object_detection(...g){return(0,_.post_process_object_detection)(...g)}post_process_panoptic_segmentation(...g){return(0,_.post_process_panoptic_segmentation)(...g)}post_process_instance_segmentation(...g){return(0,_.post_process_instance_segmentation)(...g)}}class X extends j{}},"./src/models/donut/image_processing_donut.js":(Le,A,r)=>{r.r(A),r.d(A,{DonutFeatureExtractor:()=>j,DonutImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{pad_image(R,g,v,M={}){const[y,b,I]=g;let K=this.image_mean;Array.isArray(this.image_mean)||(K=new Array(I).fill(K));let se=this.image_std;Array.isArray(se)||(se=new Array(I).fill(K));const ne=K.map((W,U)=>-W/se[U]);return super.pad_image(R,g,v,{center:!0,constant_values:ne,...M})}}class j extends D{}},"./src/models/dpt/image_processing_dpt.js":(Le,A,r)=>{r.r(A),r.d(A,{DPTFeatureExtractor:()=>j,DPTImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{}class j extends D{}},"./src/models/efficientnet/image_processing_efficientnet.js":(Le,A,r)=>{r.r(A),r.d(A,{EfficientNetImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{constructor(X){super(X),this.include_top=this.config.include_top??!0,this.include_top&&(this.image_std=this.image_std.map(R=>R*R))}}},"./src/models/feature_extractors.js":(Le,A,r)=>{r.r(A),r.d(A,{ASTFeatureExtractor:()=>_.ASTFeatureExtractor,ClapFeatureExtractor:()=>D.ClapFeatureExtractor,ImageFeatureExtractor:()=>b.ImageProcessor,MoonshineFeatureExtractor:()=>j.MoonshineFeatureExtractor,PyAnnoteFeatureExtractor:()=>X.PyAnnoteFeatureExtractor,SeamlessM4TFeatureExtractor:()=>R.SeamlessM4TFeatureExtractor,SpeechT5FeatureExtractor:()=>g.SpeechT5FeatureExtractor,Wav2Vec2FeatureExtractor:()=>v.Wav2Vec2FeatureExtractor,WeSpeakerFeatureExtractor:()=>M.WeSpeakerFeatureExtractor,WhisperFeatureExtractor:()=>y.WhisperFeatureExtractor});var _=r("./src/models/audio_spectrogram_transformer/feature_extraction_audio_spectrogram_transformer.js"),D=r("./src/models/clap/feature_extraction_clap.js"),j=r("./src/models/moonshine/feature_extraction_moonshine.js"),X=r("./src/models/pyannote/feature_extraction_pyannote.js"),R=r("./src/models/seamless_m4t/feature_extraction_seamless_m4t.js"),g=r("./src/models/speecht5/feature_extraction_speecht5.js"),v=r("./src/models/wav2vec2/feature_extraction_wav2vec2.js"),M=r("./src/models/wespeaker/feature_extraction_wespeaker.js"),y=r("./src/models/whisper/feature_extraction_whisper.js"),b=r("./src/base/image_processors_utils.js")},"./src/models/florence2/processing_florence2.js":(Le,A,r)=>{r.r(A),r.d(A,{Florence2Processor:()=>X});var _=r("./src/base/processing_utils.js"),D=r("./src/models/auto/image_processing_auto.js"),j=r("./src/tokenizers.js");class X extends _.Processor{constructor(g,v){super(g,v);const{tasks_answer_post_processing_type:M,task_prompts_without_inputs:y,task_prompts_with_input:b}=this.image_processor.config;this.tasks_answer_post_processing_type=new Map(Object.entries(M??{})),this.task_prompts_without_inputs=new Map(Object.entries(y??{})),this.task_prompts_with_input=new Map(Object.entries(b??{})),this.regexes={quad_boxes:/(.+?)/gm,bboxes:/([^<]+)?/gm},this.size_per_bin=1e3}construct_prompts(g){typeof g=="string"&&(g=[g]);const v=[];for(const M of g)if(this.task_prompts_without_inputs.has(M))v.push(this.task_prompts_without_inputs.get(M));else{for(const[y,b]of this.task_prompts_with_input)if(M.includes(y)){v.push(b.replaceAll("{input}",M).replaceAll(y,""));break}v.length!==g.length&&v.push(M)}return v}post_process_generation(g,v,M){const y=this.tasks_answer_post_processing_type.get(v)??"pure_text";g=g.replaceAll("","").replaceAll("","");let b;switch(y){case"pure_text":b=g;break;case"description_with_bboxes":case"bboxes":case"phrase_grounding":case"ocr":const I=y==="ocr"?"quad_boxes":"bboxes",K=g.matchAll(this.regexes[I]),se=[],ne=[];for(const[W,U,...q]of K)se.push(U?U.trim():se.at(-1)??""),ne.push(q.map(($,S)=>(Number($)+.5)/this.size_per_bin*M[S%2]));b={labels:se,[I]:ne};break;default:throw new Error(`Task "${v}" (of type "${y}") not yet implemented.`)}return{[v]:b}}async _call(g,v=null,M={}){if(!g&&!v)throw new Error("Either text or images must be provided");const y=await this.image_processor(g,M),b=v?this.tokenizer(v,M):{};return{...y,...b}}}me(X,"tokenizer_class",j.AutoTokenizer),me(X,"image_processor_class",D.AutoImageProcessor)},"./src/models/glpn/image_processing_glpn.js":(Le,A,r)=>{r.r(A),r.d(A,{GLPNFeatureExtractor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{}},"./src/models/grounding_dino/image_processing_grounding_dino.js":(Le,A,r)=>{r.r(A),r.d(A,{GroundingDinoImageProcessor:()=>j});var _=r("./src/base/image_processors_utils.js"),D=r("./src/utils/tensor.js");class j extends _.ImageProcessor{async _call(R){const g=await super._call(R),v=g.pixel_values.dims,M=(0,D.ones)([v[0],v[2],v[3]]);return{...g,pixel_mask:M}}}},"./src/models/grounding_dino/processing_grounding_dino.js":(Le,A,r)=>{r.r(A),r.d(A,{GroundingDinoProcessor:()=>g});var _=r("./src/base/processing_utils.js"),D=r("./src/models/auto/image_processing_auto.js"),j=r("./src/tokenizers.js"),X=r("./src/base/image_processors_utils.js");function R(v,M){const b=v.dims.at(-1)-1,I=v.tolist();I.fill(!1,0,1),I.fill(!1,b);const K=M.tolist();return I.map((se,ne)=>se?ne:null).filter(se=>se!==null).map(se=>K[se])}class g extends _.Processor{async _call(M,y,b={}){const I=M?await this.image_processor(M,b):{};return{...y?this.tokenizer(y,b):{},...I}}post_process_grounded_object_detection(M,y,{box_threshold:b=.25,text_threshold:I=.25,target_sizes:K=null}={}){const{logits:se,pred_boxes:ne}=M,W=se.dims[0];if(K!==null&&K.length!==W)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");const U=se.dims.at(1),q=se.sigmoid(),$=q.max(-1).tolist(),S=ne.tolist().map(x=>x.map(O=>(0,X.center_to_corners_format)(O))),w=[];for(let x=0;xre.map((xe,ce)=>xe*O[(ce+1)%2])));const ae=$[x],ie=[],ve=[],we=[];for(let re=0;re{r.r(A),r.d(A,{Idefics3ImageProcessor:()=>j});var _=r("./src/base/image_processors_utils.js"),D=r("./src/utils/tensor.js");class j extends _.ImageProcessor{constructor(R){super(R),this.do_image_splitting=R.do_image_splitting??!0,this.max_image_size=R.max_image_size}get_resize_for_vision_encoder(R,g){let[v,M]=R.dims.slice(-2);const y=M/v;return M>=v?(M=Math.ceil(M/g)*g,v=Math.floor(M/y),v=Math.ceil(v/g)*g):(v=Math.ceil(v/g)*g,M=Math.floor(v*y),M=Math.ceil(M/g)*g),{height:v,width:M}}async _call(R,{do_image_splitting:g=null,return_row_col_info:v=!1}={}){let M;if(!Array.isArray(R))M=[[R]];else{if(R.length===0||!R[0])throw new Error("No images provided.");Array.isArray(R[0])?M=R:M=[R]}let y=[],b=[],I=[];const K=[],se=[];for(const x of M){let O=await Promise.all(x.map(ve=>this.preprocess(ve)));K.push(...O.map(ve=>ve.original_size)),se.push(...O.map(ve=>ve.reshaped_input_size)),O.forEach(ve=>ve.pixel_values.unsqueeze_(0));const{longest_edge:ae}=this.max_image_size;let ie;if(g??this.do_image_splitting){let ve=new Array(O.length),we=new Array(O.length);ie=await Promise.all(O.map(async(re,xe)=>{const ce=this.get_resize_for_vision_encoder(re.pixel_values,ae),ke=await(0,D.interpolate_4d)(re.pixel_values,{size:[ce.height,ce.width]}),{frames:Ie,num_splits_h:Ee,num_splits_w:tt}=await this.split_image(ke,this.max_image_size);return ve[xe]=Ee,we[xe]=tt,(0,D.cat)(Ie,0)})),b.push(ve),I.push(we)}else{const ve=[ae,ae];ie=await Promise.all(O.map(we=>(0,D.interpolate_4d)(we.pixel_values,{size:ve}))),b.push(new Array(O.length).fill(0)),I.push(new Array(O.length).fill(0))}y.push((0,D.cat)(ie,0))}const ne=y.length,[W,U,q,$]=y[0].dims;let S,w;if(ne===1)S=y[0].unsqueeze_(0),w=(0,D.full)([ne,W,q,$],!0);else{const x=Math.max(...y.map(ie=>ie.dims.at(0)));w=(0,D.full)([ne,x,q,$],!0);const O=w.data,ae=x*q*$;for(let ie=0;iev||I>M){K=Math.ceil(b/v),se=Math.ceil(I/M);const ne=Math.ceil(b/K),W=Math.ceil(I/se);for(let $=0;${r.r(A),r.d(A,{Idefics3Processor:()=>M});var _=r("./src/base/processing_utils.js"),D=r("./src/models/auto/image_processing_auto.js"),j=r("./src/tokenizers.js");r("./src/utils/image.js");var X=r("./src/utils/core.js");function R(y,b,I,K,se,ne){let W="";for(let U=0;U`+se.repeat(y);W+=` +`}return W+=` +${K}${ne}`+se.repeat(y)+`${K}`,W}function g(y,b,I,K){return`${b}${K}`+I.repeat(y)+`${b}`}function v(y,b,I,K,se,ne){return y===0&&b===0?g(I,K,se,ne):R(I,y,b,K,se,ne)}class M extends _.Processor{constructor(){super(...arguments);me(this,"fake_image_token","");me(this,"image_token","");me(this,"global_img_token","")}async _call(I,K=null,se={}){se.return_row_col_info??(se.return_row_col_info=!0);let ne;K&&(ne=await this.image_processor(K,se)),Array.isArray(I)||(I=[I]);const W=ne.rows??[new Array(I.length).fill(0)],U=ne.cols??[new Array(I.length).fill(0)],q=this.config.image_seq_len,$=[],S=[];for(let x=0;xv(xe,ie[ce],q,this.fake_image_token,this.image_token,this.global_img_token)),we=O.split(this.image_token);if(we.length===0)throw new Error("The image token should be present in the text.");let re=we[0];for(let xe=0;xe{r.r(A),r.d(A,{BeitFeatureExtractor:()=>_.BeitFeatureExtractor,BitImageProcessor:()=>D.BitImageProcessor,CLIPFeatureExtractor:()=>X.CLIPFeatureExtractor,CLIPImageProcessor:()=>X.CLIPImageProcessor,ChineseCLIPFeatureExtractor:()=>j.ChineseCLIPFeatureExtractor,ConvNextFeatureExtractor:()=>R.ConvNextFeatureExtractor,ConvNextImageProcessor:()=>R.ConvNextImageProcessor,DPTFeatureExtractor:()=>y.DPTFeatureExtractor,DPTImageProcessor:()=>y.DPTImageProcessor,DeiTFeatureExtractor:()=>g.DeiTFeatureExtractor,DeiTImageProcessor:()=>g.DeiTImageProcessor,DetrFeatureExtractor:()=>v.DetrFeatureExtractor,DetrImageProcessor:()=>v.DetrImageProcessor,DonutFeatureExtractor:()=>M.DonutFeatureExtractor,DonutImageProcessor:()=>M.DonutImageProcessor,EfficientNetImageProcessor:()=>b.EfficientNetImageProcessor,GLPNFeatureExtractor:()=>I.GLPNFeatureExtractor,GroundingDinoImageProcessor:()=>K.GroundingDinoImageProcessor,Idefics3ImageProcessor:()=>se.Idefics3ImageProcessor,JinaCLIPImageProcessor:()=>W.JinaCLIPImageProcessor,LlavaOnevisionImageProcessor:()=>U.LlavaOnevisionImageProcessor,Mask2FormerImageProcessor:()=>q.Mask2FormerImageProcessor,MaskFormerFeatureExtractor:()=>$.MaskFormerFeatureExtractor,MaskFormerImageProcessor:()=>$.MaskFormerImageProcessor,MobileNetV1FeatureExtractor:()=>S.MobileNetV1FeatureExtractor,MobileNetV1ImageProcessor:()=>S.MobileNetV1ImageProcessor,MobileNetV2FeatureExtractor:()=>w.MobileNetV2FeatureExtractor,MobileNetV2ImageProcessor:()=>w.MobileNetV2ImageProcessor,MobileNetV3FeatureExtractor:()=>x.MobileNetV3FeatureExtractor,MobileNetV3ImageProcessor:()=>x.MobileNetV3ImageProcessor,MobileNetV4FeatureExtractor:()=>O.MobileNetV4FeatureExtractor,MobileNetV4ImageProcessor:()=>O.MobileNetV4ImageProcessor,MobileViTFeatureExtractor:()=>ae.MobileViTFeatureExtractor,MobileViTImageProcessor:()=>ae.MobileViTImageProcessor,NougatImageProcessor:()=>ie.NougatImageProcessor,OwlViTFeatureExtractor:()=>we.OwlViTFeatureExtractor,OwlViTImageProcessor:()=>we.OwlViTImageProcessor,Owlv2ImageProcessor:()=>ve.Owlv2ImageProcessor,Phi3VImageProcessor:()=>re.Phi3VImageProcessor,PvtImageProcessor:()=>xe.PvtImageProcessor,Qwen2VLImageProcessor:()=>ce.Qwen2VLImageProcessor,RTDetrImageProcessor:()=>ke.RTDetrImageProcessor,SamImageProcessor:()=>Ie.SamImageProcessor,SegformerFeatureExtractor:()=>Ee.SegformerFeatureExtractor,SegformerImageProcessor:()=>Ee.SegformerImageProcessor,SiglipImageProcessor:()=>tt.SiglipImageProcessor,Swin2SRImageProcessor:()=>Ge.Swin2SRImageProcessor,VLMImageProcessor:()=>ne.VLMImageProcessor,ViTFeatureExtractor:()=>ye.ViTFeatureExtractor,ViTImageProcessor:()=>ye.ViTImageProcessor,VitMatteImageProcessor:()=>J.VitMatteImageProcessor,VitPoseImageProcessor:()=>de.VitPoseImageProcessor,YolosFeatureExtractor:()=>Ce.YolosFeatureExtractor,YolosImageProcessor:()=>Ce.YolosImageProcessor});var _=r("./src/models/beit/image_processing_beit.js"),D=r("./src/models/bit/image_processing_bit.js"),j=r("./src/models/chinese_clip/image_processing_chinese_clip.js"),X=r("./src/models/clip/image_processing_clip.js"),R=r("./src/models/convnext/image_processing_convnext.js"),g=r("./src/models/deit/image_processing_deit.js"),v=r("./src/models/detr/image_processing_detr.js"),M=r("./src/models/donut/image_processing_donut.js"),y=r("./src/models/dpt/image_processing_dpt.js"),b=r("./src/models/efficientnet/image_processing_efficientnet.js"),I=r("./src/models/glpn/image_processing_glpn.js"),K=r("./src/models/grounding_dino/image_processing_grounding_dino.js"),se=r("./src/models/idefics3/image_processing_idefics3.js"),ne=r("./src/models/janus/image_processing_janus.js"),W=r("./src/models/jina_clip/image_processing_jina_clip.js"),U=r("./src/models/llava_onevision/image_processing_llava_onevision.js"),q=r("./src/models/mask2former/image_processing_mask2former.js"),$=r("./src/models/maskformer/image_processing_maskformer.js"),S=r("./src/models/mobilenet_v1/image_processing_mobilenet_v1.js"),w=r("./src/models/mobilenet_v2/image_processing_mobilenet_v2.js"),x=r("./src/models/mobilenet_v3/image_processing_mobilenet_v3.js"),O=r("./src/models/mobilenet_v4/image_processing_mobilenet_v4.js"),ae=r("./src/models/mobilevit/image_processing_mobilevit.js"),ie=r("./src/models/nougat/image_processing_nougat.js"),ve=r("./src/models/owlv2/image_processing_owlv2.js"),we=r("./src/models/owlvit/image_processing_owlvit.js"),re=r("./src/models/phi3_v/image_processing_phi3_v.js"),xe=r("./src/models/pvt/image_processing_pvt.js"),ce=r("./src/models/qwen2_vl/image_processing_qwen2_vl.js"),ke=r("./src/models/rt_detr/image_processing_rt_detr.js"),Ie=r("./src/models/sam/image_processing_sam.js"),Ee=r("./src/models/segformer/image_processing_segformer.js"),tt=r("./src/models/siglip/image_processing_siglip.js"),Ge=r("./src/models/swin2sr/image_processing_swin2sr.js"),ye=r("./src/models/vit/image_processing_vit.js"),J=r("./src/models/vitmatte/image_processing_vitmatte.js"),de=r("./src/models/vitpose/image_processing_vitpose.js"),Ce=r("./src/models/yolos/image_processing_yolos.js")},"./src/models/janus/image_processing_janus.js":(Le,A,r)=>{r.r(A),r.d(A,{VLMImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{constructor(X){super({do_pad:!0,pad_size:{width:X.image_size,height:X.image_size},...X}),this.constant_values=this.config.background_color.map(R=>R*this.rescale_factor)}pad_image(X,R,g,v){return super.pad_image(X,R,g,{constant_values:this.constant_values,center:!0,...v})}}},"./src/models/janus/processing_janus.js":(Le,A,r)=>{r.r(A),r.d(A,{VLChatProcessor:()=>v});var _=r("./src/base/processing_utils.js"),D=r("./src/models/auto/image_processing_auto.js"),j=r("./src/tokenizers.js"),X=r("./src/utils/core.js"),R=r("./src/utils/tensor.js"),g=r("./src/utils/image.js");class v extends _.Processor{constructor(y,b){super(y,b),this.image_tag=this.config.image_tag,this.image_start_tag=this.config.image_start_tag,this.image_end_tag=this.config.image_end_tag,this.num_image_tokens=this.config.num_image_tokens}async _call(y,{images:b=null,chat_template:I="default"}={}){b?Array.isArray(b)||(b=[b]):b=await Promise.all(y.filter(ie=>ie.images).flatMap(ie=>ie.images).map(ie=>g.RawImage.read(ie)));const K=this.tokenizer,se=K.apply_chat_template(y,{tokenize:!1,add_generation_prompt:!0,chat_template:I}),ne=ie=>K.encode(ie,{add_special_tokens:!1}),W=se.split(this.image_tag),U=W.length-1;if(b.length!==U)throw new Error(`Number of images provided (${b.length}) does not match number of "${this.image_tag}" image tags (${U})`);const[q,$,S]=K.model.convert_tokens_to_ids([this.image_tag,this.image_start_tag,this.image_end_tag]);let w=ne(W[0]),x=new Array(w.length).fill(!1);for(let ie=1;ie0){const ie=await this.image_processor(b);return ie.pixel_values.unsqueeze_(0),{...ae,...ie}}return ae}}me(v,"image_processor_class",D.AutoImageProcessor),me(v,"tokenizer_class",j.AutoTokenizer),me(v,"uses_processor_config",!0)},"./src/models/jina_clip/image_processing_jina_clip.js":(Le,A,r)=>{r.r(A),r.d(A,{JinaCLIPImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{constructor(X){const{resize_mode:R,fill_color:g,interpolation:v,size:M,...y}=X,b=R==="squash"?{width:M,height:M}:R==="shortest"?{shortest_edge:M}:{longest_edge:M},I=v==="bicubic"?3:2;super({...y,size:b,resample:I,do_center_crop:!0,crop_size:M,do_normalize:!0})}}},"./src/models/jina_clip/processing_jina_clip.js":(Le,A,r)=>{r.r(A),r.d(A,{JinaCLIPProcessor:()=>X});var _=r("./src/base/processing_utils.js"),D=r("./src/models/auto/image_processing_auto.js"),j=r("./src/tokenizers.js");class X extends _.Processor{async _call(g=null,v=null,M={}){if(!g&&!v)throw new Error("Either text or images must be provided");const y=g?this.tokenizer(g,M):{},b=v?await this.image_processor(v,M):{};return{...y,...b}}}me(X,"tokenizer_class",j.AutoTokenizer),me(X,"image_processor_class",D.AutoImageProcessor)},"./src/models/llava_onevision/image_processing_llava_onevision.js":(Le,A,r)=>{r.r(A),r.d(A,{LlavaOnevisionImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{}},"./src/models/mask2former/image_processing_mask2former.js":(Le,A,r)=>{r.r(A),r.d(A,{Mask2FormerImageProcessor:()=>D});var _=r("./src/models/maskformer/image_processing_maskformer.js");class D extends _.MaskFormerImageProcessor{}},"./src/models/maskformer/image_processing_maskformer.js":(Le,A,r)=>{r.r(A),r.d(A,{MaskFormerFeatureExtractor:()=>j,MaskFormerImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{post_process_panoptic_segmentation(...R){return(0,_.post_process_panoptic_segmentation)(...R)}post_process_instance_segmentation(...R){return(0,_.post_process_instance_segmentation)(...R)}}class j extends D{}},"./src/models/mgp_str/processing_mgp_str.js":(Le,A,r)=>{r.r(A),r.d(A,{MgpstrProcessor:()=>g});var _=r("./src/base/processing_utils.js"),D=r("./src/models/auto/image_processing_auto.js"),j=r("./src/tokenizers.js"),X=r("./src/utils/maths.js");const R={char:["char_decode",1],bpe:["bpe_decode",2],wp:["wp_decode",102]};class g extends _.Processor{get char_tokenizer(){return this.components.char_tokenizer}get bpe_tokenizer(){return this.components.bpe_tokenizer}get wp_tokenizer(){return this.components.wp_tokenizer}_decode_helper(M,y){if(!R.hasOwnProperty(y))throw new Error(`Format ${y} is not supported.`);const[b,I]=R[y],K=this[b].bind(this),[se,ne]=M.dims,W=[],U=[],q=M.tolist();for(let S=0;S0?O.reduce((ie,ve)=>ie*ve,1):0;U.push(x),W.push(ae)}return[K(U),W]}char_decode(M){return this.char_tokenizer.batch_decode(M).map(y=>y.replaceAll(" ",""))}bpe_decode(M){return this.bpe_tokenizer.batch_decode(M)}wp_decode(M){return this.wp_tokenizer.batch_decode(M).map(y=>y.replaceAll(" ",""))}batch_decode([M,y,b]){const[I,K]=this._decode_helper(M,"char"),[se,ne]=this._decode_helper(y,"bpe"),[W,U]=this._decode_helper(b,"wp"),q=[],$=[];for(let S=0;S{r.r(A),r.d(A,{MobileNetV1FeatureExtractor:()=>j,MobileNetV1ImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{}class j extends D{}},"./src/models/mobilenet_v2/image_processing_mobilenet_v2.js":(Le,A,r)=>{r.r(A),r.d(A,{MobileNetV2FeatureExtractor:()=>j,MobileNetV2ImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{}class j extends D{}},"./src/models/mobilenet_v3/image_processing_mobilenet_v3.js":(Le,A,r)=>{r.r(A),r.d(A,{MobileNetV3FeatureExtractor:()=>j,MobileNetV3ImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{}class j extends D{}},"./src/models/mobilenet_v4/image_processing_mobilenet_v4.js":(Le,A,r)=>{r.r(A),r.d(A,{MobileNetV4FeatureExtractor:()=>j,MobileNetV4ImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{}class j extends D{}},"./src/models/mobilevit/image_processing_mobilevit.js":(Le,A,r)=>{r.r(A),r.d(A,{MobileViTFeatureExtractor:()=>j,MobileViTImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{}class j extends D{}},"./src/models/moonshine/feature_extraction_moonshine.js":(Le,A,r)=>{r.r(A),r.d(A,{MoonshineFeatureExtractor:()=>j});var _=r("./src/base/feature_extraction_utils.js"),D=r("./src/utils/tensor.js");class j extends _.FeatureExtractor{async _call(R){(0,_.validate_audio_inputs)(R,"MoonshineFeatureExtractor"),R instanceof Float64Array&&(R=new Float32Array(R));const g=[1,R.length];return{input_values:new D.Tensor("float32",R,g)}}}},"./src/models/moonshine/processing_moonshine.js":(Le,A,r)=>{r.r(A),r.d(A,{MoonshineProcessor:()=>X});var _=r("./src/models/auto/feature_extraction_auto.js"),D=r("./src/tokenizers.js"),j=r("./src/base/processing_utils.js");class X extends j.Processor{async _call(g){return await this.feature_extractor(g)}}me(X,"tokenizer_class",D.AutoTokenizer),me(X,"feature_extractor_class",_.AutoFeatureExtractor)},"./src/models/nougat/image_processing_nougat.js":(Le,A,r)=>{r.r(A),r.d(A,{NougatImageProcessor:()=>D});var _=r("./src/models/donut/image_processing_donut.js");class D extends _.DonutImageProcessor{}},"./src/models/owlv2/image_processing_owlv2.js":(Le,A,r)=>{r.r(A),r.d(A,{Owlv2ImageProcessor:()=>D});var _=r("./src/models/owlvit/image_processing_owlvit.js");class D extends _.OwlViTImageProcessor{}},"./src/models/owlvit/image_processing_owlvit.js":(Le,A,r)=>{r.r(A),r.d(A,{OwlViTFeatureExtractor:()=>j,OwlViTImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{post_process_object_detection(...R){return(0,_.post_process_object_detection)(...R)}}class j extends D{}},"./src/models/owlvit/processing_owlvit.js":(Le,A,r)=>{r.r(A),r.d(A,{OwlViTProcessor:()=>X});var _=r("./src/base/processing_utils.js"),D=r("./src/models/auto/image_processing_auto.js"),j=r("./src/tokenizers.js");class X extends _.Processor{}me(X,"tokenizer_class",j.AutoTokenizer),me(X,"image_processor_class",D.AutoImageProcessor)},"./src/models/paligemma/processing_paligemma.js":(Le,A,r)=>{r.r(A),r.d(A,{PaliGemmaProcessor:()=>g});var _=r("./src/base/processing_utils.js"),D=r("./src/models/auto/image_processing_auto.js"),j=r("./src/tokenizers.js");const X="";function R(v,M,y,b,I){return`${b.repeat(y*I)}${M}${v} +`}class g extends _.Processor{async _call(M,y=null,b={}){y||(console.warn("You are using PaliGemma without a text prefix. It will perform as a picture-captioning model."),y=""),Array.isArray(M)||(M=[M]),Array.isArray(y)||(y=[y]);const I=this.tokenizer.bos_token,K=this.image_processor.config.image_seq_length;let se;y.some(U=>U.includes(X))?se=y.map(U=>{const q=U.replaceAll(X,X.repeat(K)),$=q.lastIndexOf(X),S=$===-1?0:$+X.length;return q.slice(0,S)+I+q.slice(S)+` +`}):(console.warn("You are passing both `text` and `images` to `PaliGemmaProcessor`. The processor expects special image tokens in the text, as many tokens as there are images per each text. It is recommended to add `` tokens in the very beginning of your text. For this call, we will infer how many images each text has and add special tokens."),se=y.map(U=>R(U,I,K,X,M.length)));const ne=this.tokenizer(se,b);return{...await this.image_processor(M,b),...ne}}}me(g,"tokenizer_class",j.AutoTokenizer),me(g,"image_processor_class",D.AutoImageProcessor),me(g,"uses_processor_config",!1)},"./src/models/phi3_v/image_processing_phi3_v.js":(Le,A,r)=>{r.r(A),r.d(A,{Phi3VImageProcessor:()=>M});var _=r("./src/base/image_processors_utils.js"),D=r("./src/utils/tensor.js");const j=336,X=[2,3],{ceil:R,floor:g,sqrt:v}=Math;class M extends _.ImageProcessor{constructor(b){super({...b,do_normalize:!0,do_pad:!0,pad_size:"custom",do_convert_rgb:!0,do_resize:!0}),this._num_crops=b.num_crops}calc_num_image_tokens_from_image_size(b,I){const{num_img_tokens:K}=this.config;return g((g(I/j)*g(b/j)+1)*K+1+(g(I/j)+1)*v(K))}get_resize_output_image_size(b,I){const K=this._num_crops,[se,ne]=b.size;let W=se/ne,U=1;for(;U*Math.ceil(U/W)<=K;)U+=1;U-=1;const q=Math.floor(U*336),$=Math.floor(q/W);return[q,$]}pad_image(b,I,K,se={}){const[ne,W]=I,U=j*R(ne/j),q=j*R(W/j),$=[1,1,1].map((S,w)=>(S-this.image_mean[w])/this.image_std[w]);return super.pad_image(b,I,{width:q,height:U},{center:!0,constant_values:$,...se})}async _call(b,{num_crops:I=null}={}){if(this._num_crops=I??(I=this.config.num_crops),I<4||v(I)%1!==0)throw new Error("num_crops must be a square number >= 4");Array.isArray(b)||(b=[b]);const K=b.length,se=await Promise.all(b.map(x=>this.preprocess(x))),ne=se.map(x=>x.original_size),W=se.map(x=>x.reshaped_input_size),U=[];for(const{pixel_values:x}of se){x.unsqueeze_(0);const[O,ae]=x.dims.slice(-2),ie=await(0,D.interpolate_4d)(x,{size:[j,j],mode:"bicubic"});if(I>0){const ve=[],we=v(I),re=g(ae/we),xe=g(O/we);for(let ke=0;kex.map(O=>j*R(O/j))),S=new D.Tensor("int64",$.flat(),[K,2]),w=$.map(([x,O])=>this.calc_num_image_tokens_from_image_size(O,x));return{pixel_values:q,original_sizes:ne,reshaped_input_sizes:W,image_sizes:S,num_img_tokens:w}}}},"./src/models/phi3_v/processing_phi3_v.js":(Le,A,r)=>{r.r(A),r.d(A,{Phi3VProcessor:()=>g});var _=r("./src/base/processing_utils.js"),D=r("./src/models/auto/image_processing_auto.js"),j=r("./src/tokenizers.js");r("./src/utils/image.js");const X="<|image|>",R=/<\|image_\d+\|>/g;class g extends _.Processor{async _call(M,y=null,{padding:b=!0,truncation:I=!0,num_crops:K=null}={}){Array.isArray(M)||(M=[M]);let se,ne;if(y){ne=await this.image_processor(y,{num_crops:K});const{num_img_tokens:W}=ne,U=M.map(($,S)=>$.split(R).join(X.repeat(W[S])));se=this.tokenizer(U,{padding:b,truncation:I});const q=this.tokenizer.model.convert_tokens_to_ids([X])[0];se.input_ids.map_($=>$==q?-$:$)}else se=this.tokenizer(M);return{...se,...ne}}}me(g,"image_processor_class",D.AutoImageProcessor),me(g,"tokenizer_class",j.AutoTokenizer)},"./src/models/processors.js":(Le,A,r)=>{r.r(A),r.d(A,{Florence2Processor:()=>_.Florence2Processor,GroundingDinoProcessor:()=>D.GroundingDinoProcessor,Idefics3Processor:()=>j.Idefics3Processor,JinaCLIPProcessor:()=>R.JinaCLIPProcessor,MgpstrProcessor:()=>g.MgpstrProcessor,MoonshineProcessor:()=>v.MoonshineProcessor,OwlViTProcessor:()=>M.OwlViTProcessor,PaliGemmaProcessor:()=>b.PaliGemmaProcessor,Phi3VProcessor:()=>y.Phi3VProcessor,PyAnnoteProcessor:()=>I.PyAnnoteProcessor,Qwen2VLProcessor:()=>K.Qwen2VLProcessor,SamProcessor:()=>se.SamProcessor,SpeechT5Processor:()=>ne.SpeechT5Processor,VLChatProcessor:()=>X.VLChatProcessor,Wav2Vec2Processor:()=>W.Wav2Vec2Processor,Wav2Vec2ProcessorWithLM:()=>U.Wav2Vec2ProcessorWithLM,WhisperProcessor:()=>q.WhisperProcessor});var _=r("./src/models/florence2/processing_florence2.js"),D=r("./src/models/grounding_dino/processing_grounding_dino.js"),j=r("./src/models/idefics3/processing_idefics3.js"),X=r("./src/models/janus/processing_janus.js"),R=r("./src/models/jina_clip/processing_jina_clip.js"),g=r("./src/models/mgp_str/processing_mgp_str.js"),v=r("./src/models/moonshine/processing_moonshine.js"),M=r("./src/models/owlvit/processing_owlvit.js"),y=r("./src/models/phi3_v/processing_phi3_v.js"),b=r("./src/models/paligemma/processing_paligemma.js"),I=r("./src/models/pyannote/processing_pyannote.js"),K=r("./src/models/qwen2_vl/processing_qwen2_vl.js"),se=r("./src/models/sam/processing_sam.js"),ne=r("./src/models/speecht5/processing_speecht5.js"),W=r("./src/models/wav2vec2/processing_wav2vec2.js"),U=r("./src/models/wav2vec2_with_lm/processing_wav2vec2_with_lm.js"),q=r("./src/models/whisper/processing_whisper.js")},"./src/models/pvt/image_processing_pvt.js":(Le,A,r)=>{r.r(A),r.d(A,{PvtImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{}},"./src/models/pyannote/feature_extraction_pyannote.js":(Le,A,r)=>{r.r(A),r.d(A,{PyAnnoteFeatureExtractor:()=>X});var _=r("./src/base/feature_extraction_utils.js"),D=r("./src/utils/tensor.js"),j=r("./src/utils/maths.js");class X extends _.FeatureExtractor{async _call(g){(0,_.validate_audio_inputs)(g,"PyAnnoteFeatureExtractor"),g instanceof Float64Array&&(g=new Float32Array(g));const v=[1,1,g.length];return{input_values:new D.Tensor("float32",g,v)}}samples_to_frames(g){return(g-this.config.offset)/this.config.step}post_process_speaker_diarization(g,v){const M=v/this.samples_to_frames(v)/this.config.sampling_rate,y=[];for(const b of g.tolist()){const I=[];let K=-1;for(let se=0;se({id:se,start:ne*M,end:W*M,confidence:U/(W-ne)})))}return y}}},"./src/models/pyannote/processing_pyannote.js":(Le,A,r)=>{r.r(A),r.d(A,{PyAnnoteProcessor:()=>j});var _=r("./src/base/processing_utils.js"),D=r("./src/models/pyannote/feature_extraction_pyannote.js");class j extends _.Processor{async _call(R){return await this.feature_extractor(R)}post_process_speaker_diarization(...R){return this.feature_extractor.post_process_speaker_diarization(...R)}get sampling_rate(){return this.feature_extractor.config.sampling_rate}}me(j,"feature_extractor_class",D.PyAnnoteFeatureExtractor)},"./src/models/qwen2_vl/image_processing_qwen2_vl.js":(Le,A,r)=>{r.r(A),r.d(A,{Qwen2VLImageProcessor:()=>j});var _=r("./src/base/image_processors_utils.js"),D=r("./src/utils/tensor.js");class j extends _.ImageProcessor{async _call(R,...g){const{pixel_values:v,original_sizes:M,reshaped_input_sizes:y}=await super._call(R,...g);let b=v;const{temporal_patch_size:I,merge_size:K,patch_size:se}=this.config;b.dims[0]===1&&(b=(0,D.cat)(Array.from({length:I},()=>b),0));const ne=b.dims[0]/I,W=b.dims[1],U=Math.floor(b.dims[2]/se),q=Math.floor(b.dims[3]/se),$=b.view(ne,I,W,Math.floor(U/K),K,se,Math.floor(q/K),K,se).permute(0,3,6,4,7,2,1,5,8).view(ne*U*q,W*I*se*se),S=new D.Tensor("int64",[ne,U,q],[1,3]);return{pixel_values:$,image_grid_thw:S,original_sizes:M,reshaped_input_sizes:y}}}},"./src/models/qwen2_vl/processing_qwen2_vl.js":(Le,A,r)=>{r.r(A),r.d(A,{Qwen2VLProcessor:()=>X});var _=r("./src/base/processing_utils.js"),D=r("./src/models/auto/image_processing_auto.js"),j=r("./src/tokenizers.js");r("./src/utils/image.js");class X extends _.Processor{async _call(g,v=null,...M){Array.isArray(g)||(g=[g]);let y,b;if(v&&(y=await this.image_processor(v),b=y.image_grid_thw),b){let K=this.image_processor.config.merge_size**2,se=0;const ne=b.tolist();g=g.map(W=>{for(;W.includes("<|image_pad|>");){const U=Number(ne[se++].reduce((q,$)=>q*$,1n));W=W.replace("<|image_pad|>","<|placeholder|>".repeat(Math.floor(U/K)))}return W.replaceAll("<|placeholder|>","<|image_pad|>")})}return{...this.tokenizer(g),...y}}}me(X,"image_processor_class",D.AutoImageProcessor),me(X,"tokenizer_class",j.AutoTokenizer)},"./src/models/rt_detr/image_processing_rt_detr.js":(Le,A,r)=>{r.r(A),r.d(A,{RTDetrImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{post_process_object_detection(...X){return(0,_.post_process_object_detection)(...X)}}},"./src/models/sam/image_processing_sam.js":(Le,A,r)=>{r.r(A),r.d(A,{SamImageProcessor:()=>X});var _=r("./src/base/image_processors_utils.js"),D=r("./src/utils/core.js"),j=r("./src/utils/tensor.js");class X extends _.ImageProcessor{reshape_input_points(g,v,M,y=!1){g=structuredClone(g);let b=(0,D.calculateDimensions)(g);if(b.length===3)y||(b=[1,...b]),g=[g];else if(b.length!==4)throw Error("The input_points must be a 4D tensor of shape `batch_size`, `point_batch_size`, `nb_points_per_image`, `2`.");for(let I=0;Iy!==v.dims[b]))throw Error(`The first ${M.length} dimensions of 'input_points' and 'input_labels' must be the same.`);return new j.Tensor("int64",g.flat(1/0).map(BigInt),M)}async _call(g,{input_points:v=null,input_labels:M=null,input_boxes:y=null}={}){const b=await super._call(g);if(v&&(b.input_points=this.reshape_input_points(v,b.original_sizes,b.reshaped_input_sizes)),M){if(!b.input_points)throw Error("`input_points` must be provided if `input_labels` are provided.");b.input_labels=this.add_input_labels(M,b.input_points)}return y&&(b.input_boxes=this.reshape_input_points(y,b.original_sizes,b.reshaped_input_sizes,!0)),b}async post_process_masks(g,v,M,{mask_threshold:y=0,binarize:b=!0,pad_size:I=null}={}){const K=[];I=I??this.pad_size;const se=[I.height,I.width];for(let ne=0;ney&&(S[w]=1);q=new j.Tensor("bool",S,q.dims)}K.push(q)}return K}generate_crop_boxes(g,v,{crop_n_layers:M=0,overlap_ratio:y=.3413333333333333,points_per_crop:b=32,crop_n_points_downscale_factor:I=1}={}){}}},"./src/models/sam/processing_sam.js":(Le,A,r)=>{r.r(A),r.d(A,{SamProcessor:()=>j});var _=r("./src/base/processing_utils.js"),D=r("./src/models/auto/image_processing_auto.js");class j extends _.Processor{async _call(...R){return await this.image_processor(...R)}post_process_masks(...R){return this.image_processor.post_process_masks(...R)}reshape_input_points(...R){return this.image_processor.reshape_input_points(...R)}}me(j,"image_processor_class",D.AutoImageProcessor)},"./src/models/seamless_m4t/feature_extraction_seamless_m4t.js":(Le,A,r)=>{r.r(A),r.d(A,{SeamlessM4TFeatureExtractor:()=>X});var _=r("./src/base/feature_extraction_utils.js"),D=r("./src/utils/tensor.js"),j=r("./src/utils/audio.js");class X extends _.FeatureExtractor{constructor(g){super(g);const v=this.config.sampling_rate,M=(0,j.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(v/2),v,null,"kaldi",!0);for(let y=0;yM*32768),(0,j.spectrogram)(g,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1192092955078125e-22,remove_dc_offset:!0,max_num_frames:v,transpose:!0})}async _call(g,{padding:v=!0,pad_to_multiple_of:M=2,do_normalize_per_mel_bins:y=!0,return_attention_mask:b=!0}={}){(0,_.validate_audio_inputs)(g,"SeamlessM4TFeatureExtractor");let I=await this._extract_fbank_features(g,this.config.max_length);if(y){const[S,w]=I.dims,x=I.data;for(let O=0;O0){const ae=new Float32Array(w*(S+O));ae.set(x),ae.fill(this.config.padding_value,x.length);const ie=S+O;I=new D.Tensor(I.type,ae,[ie,w]),b&&(K=new D.Tensor("int64",new BigInt64Array(ie),[1,ie]),K.data.fill(1n,0,S))}}const[se,ne]=I.dims,W=this.config.stride;if(se%W!==0)throw new Error(`The number of frames (${se}) must be a multiple of the stride (${W}).`);const q=I.view(1,Math.floor(se/W),ne*W),$={input_features:q};if(b){const S=q.dims[1],w=new BigInt64Array(S);if(K){const x=K.data;for(let O=1,ae=0;O{r.r(A),r.d(A,{SegformerFeatureExtractor:()=>j,SegformerImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{post_process_semantic_segmentation(...R){return(0,_.post_process_semantic_segmentation)(...R)}}class j extends D{}},"./src/models/siglip/image_processing_siglip.js":(Le,A,r)=>{r.r(A),r.d(A,{SiglipImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{}},"./src/models/speecht5/feature_extraction_speecht5.js":(Le,A,r)=>{r.r(A),r.d(A,{SpeechT5FeatureExtractor:()=>D});var _=r("./src/base/feature_extraction_utils.js");class D extends _.FeatureExtractor{}},"./src/models/speecht5/processing_speecht5.js":(Le,A,r)=>{r.r(A),r.d(A,{SpeechT5Processor:()=>X});var _=r("./src/base/processing_utils.js"),D=r("./src/tokenizers.js"),j=r("./src/models/auto/feature_extraction_auto.js");class X extends _.Processor{async _call(g){return await this.feature_extractor(g)}}me(X,"tokenizer_class",D.AutoTokenizer),me(X,"feature_extractor_class",j.AutoFeatureExtractor)},"./src/models/swin2sr/image_processing_swin2sr.js":(Le,A,r)=>{r.r(A),r.d(A,{Swin2SRImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{pad_image(X,R,g,v={}){const[M,y,b]=R;return super.pad_image(X,R,{width:y+(g-y%g)%g,height:M+(g-M%g)%g},{mode:"symmetric",center:!1,constant_values:-1,...v})}}},"./src/models/vit/image_processing_vit.js":(Le,A,r)=>{r.r(A),r.d(A,{ViTFeatureExtractor:()=>j,ViTImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{}class j extends D{}},"./src/models/vitmatte/image_processing_vitmatte.js":(Le,A,r)=>{r.r(A),r.d(A,{VitMatteImageProcessor:()=>j});var _=r("./src/base/image_processors_utils.js"),D=r("./src/utils/tensor.js");class j extends _.ImageProcessor{async _call(R,g){Array.isArray(R)||(R=[R]),Array.isArray(g)||(g=[g]);const v=await Promise.all(R.map(b=>this.preprocess(b))),M=await Promise.all(g.map(b=>this.preprocess(b,{do_normalize:!1,do_convert_rgb:!1,do_convert_grayscale:!0})));return{pixel_values:(0,D.stack)(v.map((b,I)=>(0,D.cat)([b.pixel_values,M[I].pixel_values],0)),0),original_sizes:v.map(b=>b.original_size),reshaped_input_sizes:v.map(b=>b.reshaped_input_size)}}}},"./src/models/vitpose/image_processing_vitpose.js":(Le,A,r)=>{r.r(A),r.d(A,{VitPoseImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{post_process_pose_estimation(X,R,{threshold:g=null}={}){const v=X.tolist(),[M,y,b,I]=X.dims,K=[];for(let se=0;se{r.r(A),r.d(A,{Wav2Vec2FeatureExtractor:()=>j});var _=r("./src/base/feature_extraction_utils.js"),D=r("./src/utils/tensor.js");class j extends _.FeatureExtractor{_zero_mean_unit_var_norm(R){const v=R.reduce((y,b)=>y+b,0)/R.length,M=R.reduce((y,b)=>y+(b-v)**2,0)/R.length;return R.map(y=>(y-v)/Math.sqrt(M+1e-7))}async _call(R){(0,_.validate_audio_inputs)(R,"Wav2Vec2FeatureExtractor"),R instanceof Float64Array&&(R=new Float32Array(R));let g=R;this.config.do_normalize&&(g=this._zero_mean_unit_var_norm(g));const v=[1,g.length];return{input_values:new D.Tensor("float32",g,v),attention_mask:new D.Tensor("int64",new BigInt64Array(g.length).fill(1n),v)}}}},"./src/models/wav2vec2/processing_wav2vec2.js":(Le,A,r)=>{r.r(A),r.d(A,{Wav2Vec2Processor:()=>X});var _=r("./src/tokenizers.js"),D=r("./src/models/auto/feature_extraction_auto.js"),j=r("./src/base/processing_utils.js");class X extends j.Processor{async _call(g){return await this.feature_extractor(g)}}me(X,"tokenizer_class",_.AutoTokenizer),me(X,"feature_extractor_class",D.AutoFeatureExtractor)},"./src/models/wav2vec2_with_lm/processing_wav2vec2_with_lm.js":(Le,A,r)=>{r.r(A),r.d(A,{Wav2Vec2ProcessorWithLM:()=>X});var _=r("./src/tokenizers.js"),D=r("./src/models/auto/feature_extraction_auto.js"),j=r("./src/base/processing_utils.js");class X extends j.Processor{async _call(g){return await this.feature_extractor(g)}}me(X,"tokenizer_class",_.AutoTokenizer),me(X,"feature_extractor_class",D.AutoFeatureExtractor)},"./src/models/wespeaker/feature_extraction_wespeaker.js":(Le,A,r)=>{r.r(A),r.d(A,{WeSpeakerFeatureExtractor:()=>j});var _=r("./src/base/feature_extraction_utils.js");r("./src/utils/tensor.js");var D=r("./src/utils/audio.js");class j extends _.FeatureExtractor{constructor(R){super(R);const g=this.config.sampling_rate,v=(0,D.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(g/2),g,null,"kaldi",!0);for(let M=0;Mg*32768),(0,D.spectrogram)(R,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1192092955078125e-22,remove_dc_offset:!0,transpose:!0,min_num_frames:this.min_num_frames})}async _call(R){(0,_.validate_audio_inputs)(R,"WeSpeakerFeatureExtractor");const g=(await this._extract_fbank_features(R)).unsqueeze_(0);if(this.config.fbank_centering_span===null){const v=g.mean(1).data,M=g.data,[y,b,I]=g.dims;for(let K=0;K{r.r(A),r.d(A,{WHISPER_LANGUAGE_MAPPING:()=>D,WHISPER_TO_LANGUAGE_CODE_MAPPING:()=>j,whisper_language_to_code:()=>X});const _=[["en","english"],["zh","chinese"],["de","german"],["es","spanish"],["ru","russian"],["ko","korean"],["fr","french"],["ja","japanese"],["pt","portuguese"],["tr","turkish"],["pl","polish"],["ca","catalan"],["nl","dutch"],["ar","arabic"],["sv","swedish"],["it","italian"],["id","indonesian"],["hi","hindi"],["fi","finnish"],["vi","vietnamese"],["he","hebrew"],["uk","ukrainian"],["el","greek"],["ms","malay"],["cs","czech"],["ro","romanian"],["da","danish"],["hu","hungarian"],["ta","tamil"],["no","norwegian"],["th","thai"],["ur","urdu"],["hr","croatian"],["bg","bulgarian"],["lt","lithuanian"],["la","latin"],["mi","maori"],["ml","malayalam"],["cy","welsh"],["sk","slovak"],["te","telugu"],["fa","persian"],["lv","latvian"],["bn","bengali"],["sr","serbian"],["az","azerbaijani"],["sl","slovenian"],["kn","kannada"],["et","estonian"],["mk","macedonian"],["br","breton"],["eu","basque"],["is","icelandic"],["hy","armenian"],["ne","nepali"],["mn","mongolian"],["bs","bosnian"],["kk","kazakh"],["sq","albanian"],["sw","swahili"],["gl","galician"],["mr","marathi"],["pa","punjabi"],["si","sinhala"],["km","khmer"],["sn","shona"],["yo","yoruba"],["so","somali"],["af","afrikaans"],["oc","occitan"],["ka","georgian"],["be","belarusian"],["tg","tajik"],["sd","sindhi"],["gu","gujarati"],["am","amharic"],["yi","yiddish"],["lo","lao"],["uz","uzbek"],["fo","faroese"],["ht","haitian creole"],["ps","pashto"],["tk","turkmen"],["nn","nynorsk"],["mt","maltese"],["sa","sanskrit"],["lb","luxembourgish"],["my","myanmar"],["bo","tibetan"],["tl","tagalog"],["mg","malagasy"],["as","assamese"],["tt","tatar"],["haw","hawaiian"],["ln","lingala"],["ha","hausa"],["ba","bashkir"],["jw","javanese"],["su","sundanese"]],D=new Map(_),j=new Map([..._.map(([R,g])=>[g,R]),["burmese","my"],["valencian","ca"],["flemish","nl"],["haitian","ht"],["letzeburgesch","lb"],["pushto","ps"],["panjabi","pa"],["moldavian","ro"],["moldovan","ro"],["sinhalese","si"],["castilian","es"]]);function X(R){R=R.toLowerCase();let g=j.get(R);if(g===void 0)if(D.has(R))g=R;else{const M=R.length===2?D.keys():D.values();throw new Error(`Language "${R}" is not supported. Must be one of: ${JSON.stringify(M)}`)}return g}},"./src/models/whisper/feature_extraction_whisper.js":(Le,A,r)=>{r.r(A),r.d(A,{WhisperFeatureExtractor:()=>X});var _=r("./src/base/feature_extraction_utils.js");r("./src/utils/tensor.js");var D=r("./src/utils/audio.js"),j=r("./src/utils/maths.js");class X extends _.FeatureExtractor{constructor(g){var v;super(g),(v=this.config).mel_filters??(v.mel_filters=(0,D.mel_filter_bank)(Math.floor(1+this.config.n_fft/2),this.config.feature_size,0,8e3,this.config.sampling_rate,"slaney","slaney")),this.window=(0,D.window_function)(this.config.n_fft,"hann")}async _extract_fbank_features(g){const v=await(0,D.spectrogram)(g,this.window,this.config.n_fft,this.config.hop_length,{power:2,mel_filters:this.config.mel_filters,log_mel:"log10",max_num_frames:this.config.nb_max_frames}),M=v.data,y=(0,j.max)(M)[0];for(let b=0;bthis.config.n_samples?(console.warn("Attempting to extract features for audio longer than 30 seconds. If using a pipeline to extract transcript from a long audio clip, remember to specify `chunk_length_s` and/or `stride_length_s`."),v=g.slice(0,this.config.n_samples)):(v=new Float32Array(this.config.n_samples),v.set(g)),{input_features:(await this._extract_fbank_features(v)).unsqueeze_(0)}}}},"./src/models/whisper/generation_whisper.js":(Le,A,r)=>{r.r(A),r.d(A,{WhisperGenerationConfig:()=>D});var _=r("./src/generation/configuration_utils.js");class D extends _.GenerationConfig{constructor(){super(...arguments);me(this,"return_timestamps",null);me(this,"return_token_timestamps",null);me(this,"num_frames",null);me(this,"alignment_heads",null);me(this,"task",null);me(this,"language",null);me(this,"no_timestamps_token_id",null);me(this,"prompt_ids",null);me(this,"is_multilingual",null);me(this,"lang_to_id",null);me(this,"task_to_id",null);me(this,"max_initial_timestamp_index",1)}}},"./src/models/whisper/processing_whisper.js":(Le,A,r)=>{r.r(A),r.d(A,{WhisperProcessor:()=>X});var _=r("./src/models/auto/feature_extraction_auto.js"),D=r("./src/tokenizers.js"),j=r("./src/base/processing_utils.js");class X extends j.Processor{async _call(g){return await this.feature_extractor(g)}}me(X,"tokenizer_class",D.AutoTokenizer),me(X,"feature_extractor_class",_.AutoFeatureExtractor)},"./src/models/yolos/image_processing_yolos.js":(Le,A,r)=>{r.r(A),r.d(A,{YolosFeatureExtractor:()=>j,YolosImageProcessor:()=>D});var _=r("./src/base/image_processors_utils.js");class D extends _.ImageProcessor{post_process_object_detection(...R){return(0,_.post_process_object_detection)(...R)}}class j extends D{}},"./src/ops/registry.js":(Le,A,r)=>{r.r(A),r.d(A,{TensorOpRegistry:()=>g});var _=r("./src/backends/onnx.js"),D=r("./src/utils/tensor.js"),j=r("./src/env.js");const X=j.apis.IS_BROWSER_ENV||j.apis.IS_WEBWORKER_ENV,R=async(v,M,y)=>{const b=await(0,_.createInferenceSession)(new Uint8Array(v),M);let I=Promise.resolve();return async K=>{const se=(0,_.isONNXProxy)(),ne=Object.fromEntries(Object.entries(K).map(([U,q])=>[U,(se?q.clone():q).ort_tensor])),W=await(I=X?I.then(()=>b.run(ne)):b.run(ne));return Array.isArray(y)?y.map(U=>new D.Tensor(W[U])):new D.Tensor(W[y])}};class g{static get nearest_interpolate_4d(){return this._nearest_interpolate_4d||(this._nearest_interpolate_4d=R([8,10,18,0,58,129,1,10,41,10,1,120,10,0,10,0,10,1,115,18,1,121,34,6,82,101,115,105,122,101,42,18,10,4,109,111,100,101,34,7,110,101,97,114,101,115,116,160,1,3,18,1,114,90,31,10,1,120,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,99,10,3,18,1,104,10,3,18,1,119,90,15,10,1,115,18,10,10,8,8,7,18,4,10,2,8,4,98,31,10,1,121,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,99,10,3,18,1,104,10,3,18,1,119,66,2,16,21],this.session_options,"y")),this._nearest_interpolate_4d}static get bilinear_interpolate_4d(){return this._bilinear_interpolate_4d||(this._bilinear_interpolate_4d=R([8,9,18,0,58,128,1,10,40,10,1,120,10,0,10,0,10,1,115,18,1,121,34,6,82,101,115,105,122,101,42,17,10,4,109,111,100,101,34,6,108,105,110,101,97,114,160,1,3,18,1,114,90,31,10,1,120,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,99,10,3,18,1,104,10,3,18,1,119,90,15,10,1,115,18,10,10,8,8,7,18,4,10,2,8,4,98,31,10,1,121,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,99,10,3,18,1,104,10,3,18,1,119,66,2,16,20],this.session_options,"y")),this._bilinear_interpolate_4d}static get bicubic_interpolate_4d(){return this._bicubic_interpolate_4d||(this._bicubic_interpolate_4d=R([8,9,18,0,58,127,10,39,10,1,120,10,0,10,0,10,1,115,18,1,121,34,6,82,101,115,105,122,101,42,16,10,4,109,111,100,101,34,5,99,117,98,105,99,160,1,3,18,1,114,90,31,10,1,120,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,99,10,3,18,1,104,10,3,18,1,119,90,15,10,1,115,18,10,10,8,8,7,18,4,10,2,8,4,98,31,10,1,121,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,99,10,3,18,1,104,10,3,18,1,119,66,2,16,20],this.session_options,"y")),this._bicubic_interpolate_4d}static get matmul(){return this._matmul||(this._matmul=R([8,9,18,0,58,55,10,17,10,1,97,10,1,98,18,1,99,34,6,77,97,116,77,117,108,18,1,114,90,9,10,1,97,18,4,10,2,8,1,90,9,10,1,98,18,4,10,2,8,1,98,9,10,1,99,18,4,10,2,8,1,66,2,16,20],this.session_options,"c")),this._matmul}static get stft(){return this._stft||(this._stft=R([8,7,18,0,58,148,1,10,38,10,1,115,10,1,106,10,1,119,10,1,108,18,1,111,34,4,83,84,70,84,42,15,10,8,111,110,101,115,105,100,101,100,24,1,160,1,2,18,1,115,90,26,10,1,115,18,21,10,19,8,1,18,15,10,3,18,1,98,10,3,18,1,115,10,3,18,1,99,90,11,10,1,106,18,6,10,4,8,7,18,0,90,16,10,1,119,18,11,10,9,8,1,18,5,10,3,18,1,119,90,11,10,1,108,18,6,10,4,8,7,18,0,98,31,10,1,111,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,102,10,3,18,1,100,10,3,18,1,99,66,2,16,17],this.session_options,"o")),this._stft}static get rfft(){return this._rfft||(this._rfft=R([8,9,18,0,58,97,10,33,10,1,120,10,0,10,1,97,18,1,121,34,3,68,70,84,42,15,10,8,111,110,101,115,105,100,101,100,24,1,160,1,2,18,1,100,90,21,10,1,120,18,16,10,14,8,1,18,10,10,3,18,1,115,10,3,18,1,99,90,11,10,1,97,18,6,10,4,8,7,18,0,98,21,10,1,121,18,16,10,14,8,1,18,10,10,3,18,1,115,10,3,18,1,99,66,2,16,20],this.session_options,"y")),this._rfft}static get top_k(){return this._top_k||(this._top_k=R([8,10,18,0,58,73,10,18,10,1,120,10,1,107,18,1,118,18,1,105,34,4,84,111,112,75,18,1,116,90,9,10,1,120,18,4,10,2,8,1,90,15,10,1,107,18,10,10,8,8,7,18,4,10,2,8,1,98,9,10,1,118,18,4,10,2,8,1,98,9,10,1,105,18,4,10,2,8,7,66,2,16,21],this.session_options,["v","i"])),this._top_k}static get slice(){return this._slice||(this._slice=R([8,7,18,0,58,96,10,25,10,1,120,10,1,115,10,1,101,10,1,97,10,1,116,18,1,121,34,5,83,108,105,99,101,18,1,114,90,9,10,1,120,18,4,10,2,8,1,90,9,10,1,115,18,4,10,2,8,7,90,9,10,1,101,18,4,10,2,8,7,90,9,10,1,97,18,4,10,2,8,7,90,9,10,1,116,18,4,10,2,8,7,98,9,10,1,121,18,4,10,2,8,1,66,2,16,13],this.session_options,"y")),this._slice}}me(g,"session_options",{})},"./src/pipelines.js":(Le,A,r)=>{r.r(A),r.d(A,{AudioClassificationPipeline:()=>we,AutomaticSpeechRecognitionPipeline:()=>xe,DepthEstimationPipeline:()=>Ce,DocumentQuestionAnsweringPipeline:()=>ye,FeatureExtractionPipeline:()=>ie,FillMaskPipeline:()=>q,ImageClassificationPipeline:()=>ke,ImageFeatureExtractionPipeline:()=>ve,ImageSegmentationPipeline:()=>Ie,ImageToImagePipeline:()=>de,ImageToTextPipeline:()=>ce,ObjectDetectionPipeline:()=>tt,Pipeline:()=>se,QuestionAnsweringPipeline:()=>U,SummarizationPipeline:()=>S,Text2TextGenerationPipeline:()=>$,TextClassificationPipeline:()=>ne,TextGenerationPipeline:()=>O,TextToAudioPipeline:()=>J,TokenClassificationPipeline:()=>W,TranslationPipeline:()=>w,ZeroShotAudioClassificationPipeline:()=>re,ZeroShotClassificationPipeline:()=>ae,ZeroShotImageClassificationPipeline:()=>Ee,ZeroShotObjectDetectionPipeline:()=>Ge,pipeline:()=>te});var _=r("./src/tokenizers.js"),D=r("./src/models.js"),j=r("./src/models/auto/processing_auto.js");r("./src/base/processing_utils.js");var X=r("./src/utils/generic.js"),R=r("./src/utils/core.js"),g=r("./src/utils/maths.js"),v=r("./src/utils/audio.js"),M=r("./src/utils/tensor.js"),y=r("./src/utils/image.js");async function b(je){return Array.isArray(je)||(je=[je]),await Promise.all(je.map(le=>y.RawImage.read(le)))}async function I(je,le){return Array.isArray(je)||(je=[je]),await Promise.all(je.map(Te=>typeof Te=="string"||Te instanceof URL?(0,v.read_audio)(Te,le):Te instanceof Float64Array?new Float32Array(Te):Te))}function K(je,le){le&&(je=je.map(Re=>Re|0));const[Te,Ue,Ve,Ne]=je;return{xmin:Te,ymin:Ue,xmax:Ve,ymax:Ne}}class se extends X.Callable{constructor({task:le,model:Te,tokenizer:Ue=null,processor:Ve=null}){super(),this.task=le,this.model=Te,this.tokenizer=Ue,this.processor=Ve}async dispose(){await this.model.dispose()}}class ne extends se{constructor(le){super(le)}async _call(le,{top_k:Te=1}={}){const Ue=this.tokenizer(le,{padding:!0,truncation:!0}),Ve=await this.model(Ue),Ne=this.model.config.problem_type==="multi_label_classification"?dt=>dt.sigmoid():dt=>new M.Tensor("float32",(0,g.softmax)(dt.data),dt.dims),Re=this.model.config.id2label,st=[];for(const dt of Ve.logits){const ct=Ne(dt),lt=await(0,M.topk)(ct,Te),ht=lt[0].tolist(),oe=lt[1].tolist().map((H,fe)=>({label:Re?Re[H]:`LABEL_${H}`,score:ht[fe]}));Te===1?st.push(...oe):st.push(oe)}return Array.isArray(le)||Te===1?st:st[0]}}class W extends se{constructor(le){super(le)}async _call(le,{ignore_labels:Te=["O"]}={}){const Ue=Array.isArray(le),Ve=this.tokenizer(Ue?le:[le],{padding:!0,truncation:!0}),Re=(await this.model(Ve)).logits,st=this.model.config.id2label,dt=[];for(let ct=0;ctut==this.tokenizer.sep_token_id);dt[ht].map((ut,mt)=>ut==1&&(mt===0||mt>oe&&ct.findIndex(vt=>vt==L[mt])===-1));const H=Ne[ht].tolist(),fe=Re[ht].tolist();for(let ut=1;utmt==L[ut])!==-1)&&(H[ut]=-1/0,fe[ut]=-1/0);const $e=(0,g.softmax)(H).map((ut,mt)=>[ut,mt]),We=(0,g.softmax)(fe).map((ut,mt)=>[ut,mt]);$e[0][0]=0,We[0][0]=0;const Je=(0,R.product)($e,We).filter(ut=>ut[0][1]<=ut[1][1]).map(ut=>[ut[0][1],ut[1][1],ut[0][0]*ut[1][0]]).sort((ut,mt)=>mt[2]-ut[2]);for(let ut=0;utH==this.tokenizer.mask_token_id);if(ct===-1)throw Error(`Mask token (${this.tokenizer.mask_token}) not found in text.`);const lt=Ve[st][ct],ht=await(0,M.topk)(new M.Tensor("float32",(0,g.softmax)(lt.data),lt.dims),Te),L=ht[0].tolist(),oe=ht[1].tolist();Ne.push(oe.map((H,fe)=>{const $e=dt.slice();return $e[ct]=H,{score:L[fe],token:Number(H),token_str:this.tokenizer.decode([H]),sequence:this.tokenizer.decode($e,{skip_special_tokens:!0})}}))}return Array.isArray(le)?Ne:Ne[0]}}class $ extends se{constructor(Te){super(Te);me(this,"_key","generated_text")}async _call(Te,Ue={}){Array.isArray(Te)||(Te=[Te]),this.model.config.prefix&&(Te=Te.map(ct=>this.model.config.prefix+ct));const Ve=this.model.config.task_specific_params;Ve&&Ve[this.task]&&Ve[this.task].prefix&&(Te=Te.map(ct=>Ve[this.task].prefix+ct));const Ne=this.tokenizer,Re={padding:!0,truncation:!0};let st;this instanceof w&&"_build_translation_inputs"in Ne?st=Ne._build_translation_inputs(Te,Re,Ue):st=Ne(Te,Re);const dt=await this.model.generate({...st,...Ue});return Ne.batch_decode(dt,{skip_special_tokens:!0}).map(ct=>({[this._key]:ct}))}}class S extends ${constructor(Te){super(Te);me(this,"_key","summary_text")}}class w extends ${constructor(Te){super(Te);me(this,"_key","translation_text")}}function x(je){return Array.isArray(je)&&je.every(le=>"role"in le&&"content"in le)}class O extends se{constructor(le){super(le)}async _call(le,Te={}){let Ue=!1,Ve=!1,Ne;if(typeof le=="string")Ne=le=[le];else if(Array.isArray(le)&&le.every(oe=>typeof oe=="string"))Ue=!0,Ne=le;else{if(x(le))le=[le];else if(Array.isArray(le)&&le.every(x))Ue=!0;else throw new Error("Input must be a string, an array of strings, a Chat, or an array of Chats");Ve=!0,Ne=le.map(oe=>this.tokenizer.apply_chat_template(oe,{tokenize:!1,add_generation_prompt:!0}))}const Re=Te.add_special_tokens??!1,st=Ve?!1:Te.return_full_text??!0;this.tokenizer.padding_side="left";const dt=this.tokenizer(Ne,{add_special_tokens:Re,padding:!0,truncation:!0}),ct=await this.model.generate({...dt,...Te}),lt=this.tokenizer.batch_decode(ct,{skip_special_tokens:!0});let ht;!st&&dt.input_ids.dims.at(-1)>0&&(ht=this.tokenizer.batch_decode(dt.input_ids,{skip_special_tokens:!0}).map(oe=>oe.length));const L=Array.from({length:le.length},oe=>[]);for(let oe=0;oe[Te.toLowerCase(),Ue])),this.entailment_id=this.label2id.entailment,this.entailment_id===void 0&&(console.warn("Could not find 'entailment' in label2id mapping. Using 2 as entailment_id."),this.entailment_id=2),this.contradiction_id=this.label2id.contradiction??this.label2id.not_entailment,this.contradiction_id===void 0&&(console.warn("Could not find 'contradiction' in label2id mapping. Using 0 as contradiction_id."),this.contradiction_id=0)}async _call(le,Te,{hypothesis_template:Ue="This example is {}.",multi_label:Ve=!1}={}){const Ne=Array.isArray(le);Ne||(le=[le]),Array.isArray(Te)||(Te=[Te]);const Re=Te.map(ct=>Ue.replace("{}",ct)),st=Ve||Te.length===1,dt=[];for(const ct of le){const lt=[];for(const oe of Re){const H=this.tokenizer(ct,{text_pair:oe,padding:!0,truncation:!0}),fe=await this.model(H);st?lt.push([fe.logits.data[this.contradiction_id],fe.logits.data[this.entailment_id]]):lt.push(fe.logits.data[this.entailment_id])}const L=(st?lt.map(oe=>(0,g.softmax)(oe)[1]):(0,g.softmax)(lt)).map((oe,H)=>[oe,H]).sort((oe,H)=>H[0]-oe[0]);dt.push({sequence:ct,labels:L.map(oe=>Te[oe[1]]),scores:L.map(oe=>oe[0])})}return Ne?dt:dt[0]}}class ie extends se{constructor(le){super(le)}async _call(le,{pooling:Te="none",normalize:Ue=!1,quantize:Ve=!1,precision:Ne="binary"}={}){const Re=this.tokenizer(le,{padding:!0,truncation:!0}),st=await this.model(Re);let dt=st.last_hidden_state??st.logits??st.token_embeddings;if(Te!=="none")if(Te==="mean")dt=(0,M.mean_pooling)(dt,Re.attention_mask);else if(Te==="cls")dt=dt.slice(null,0);else throw Error(`Pooling method '${Te}' not supported.`);return Ue&&(dt=dt.normalize(2,-1)),Ve&&(dt=(0,M.quantize_embeddings)(dt,Ne)),dt}}class ve extends se{constructor(le){super(le)}async _call(le,{pool:Te=null}={}){const Ue=await b(le),{pixel_values:Ve}=await this.processor(Ue),Ne=await this.model({pixel_values:Ve});let Re;if(Te){if(!("pooler_output"in Ne))throw Error("No pooled output was returned. Make sure the model has a 'pooler' layer when using the 'pool' option.");Re=Ne.pooler_output}else Re=Ne.last_hidden_state??Ne.logits??Ne.image_embeds;return Re}}class we extends se{constructor(le){super(le)}async _call(le,{top_k:Te=5}={}){const Ue=this.processor.feature_extractor.config.sampling_rate,Ve=await I(le,Ue),Ne=this.model.config.id2label,Re=[];for(const st of Ve){const dt=await this.processor(st),lt=(await this.model(dt)).logits[0],ht=await(0,M.topk)(new M.Tensor("float32",(0,g.softmax)(lt.data),lt.dims),Te),L=ht[0].tolist(),H=ht[1].tolist().map((fe,$e)=>({label:Ne?Ne[fe]:`LABEL_${fe}`,score:L[$e]}));Re.push(H)}return Array.isArray(le)?Re:Re[0]}}class re extends se{constructor(le){super(le)}async _call(le,Te,{hypothesis_template:Ue="This is a sound of {}."}={}){const Ve=!Array.isArray(le);Ve&&(le=[le]);const Ne=Te.map(lt=>Ue.replace("{}",lt)),Re=this.tokenizer(Ne,{padding:!0,truncation:!0}),st=this.processor.feature_extractor.config.sampling_rate,dt=await I(le,st),ct=[];for(const lt of dt){const ht=await this.processor(lt),L=await this.model({...Re,...ht}),oe=(0,g.softmax)(L.logits_per_audio.data);ct.push([...oe].map((H,fe)=>({score:H,label:Te[fe]})))}return Ve?ct[0]:ct}}class xe extends se{constructor(le){super(le)}async _call(le,Te={}){switch(this.model.config.model_type){case"whisper":return this._call_whisper(le,Te);case"wav2vec2":case"wav2vec2-bert":case"unispeech":case"unispeech-sat":case"hubert":return this._call_wav2vec2(le,Te);case"moonshine":return this._call_moonshine(le,Te);default:throw new Error(`AutomaticSpeechRecognitionPipeline does not support model type '${this.model.config.model_type}'.`)}}async _call_wav2vec2(le,Te){Te.language&&console.warn('`language` parameter is not yet supported for `wav2vec2` models, defaulting to "English".'),Te.task&&console.warn('`task` parameter is not yet supported for `wav2vec2` models, defaulting to "transcribe".');const Ue=!Array.isArray(le);Ue&&(le=[le]);const Ve=this.processor.feature_extractor.config.sampling_rate,Ne=await I(le,Ve),Re=[];for(const st of Ne){const dt=await this.processor(st),lt=(await this.model(dt)).logits[0],ht=[];for(const oe of lt)ht.push((0,g.max)(oe.data)[1]);const L=this.tokenizer.decode(ht);Re.push({text:L})}return Ue?Re[0]:Re}async _call_whisper(le,Te){const Ue=Te.return_timestamps??!1,Ve=Te.chunk_length_s??0,Ne=Te.force_full_sequences??!1;let Re=Te.stride_length_s??null;const st={...Te};Ue==="word"&&(st.return_token_timestamps=!0,st.return_timestamps=!1);const dt=!Array.isArray(le);dt&&(le=[le]);const ct=this.processor.feature_extractor.config.chunk_length/this.model.config.max_source_positions,lt=this.processor.feature_extractor.config.hop_length,ht=this.processor.feature_extractor.config.sampling_rate,L=await I(le,ht),oe=[];for(const H of L){let fe=[];if(Ve>0){if(Re===null)Re=Ve/6;else if(Ve<=Re)throw Error("`chunk_length_s` must be larger than `stride_length_s`.");const Je=ht*Ve,ut=ht*Re,mt=Je-2*ut;let vt=0;for(;;){const kt=vt+Je,At=H.subarray(vt,kt),is=await this.processor(At),ys=vt===0,Cs=kt>=H.length;if(fe.push({stride:[At.length,ys?0:ut,Cs?0:ut],input_features:is.input_features,is_last:Cs}),Cs)break;vt+=mt}}else fe=[{stride:[H.length,0,0],input_features:(await this.processor(H)).input_features,is_last:!0}];for(const Je of fe){st.num_frames=Math.floor(Je.stride[0]/lt);const ut=await this.model.generate({inputs:Je.input_features,...st});Ue==="word"?(Je.tokens=ut.sequences.tolist()[0],Je.token_timestamps=ut.token_timestamps.tolist()[0].map(mt=>(0,g.round)(mt,2))):Je.tokens=ut[0].tolist(),Je.stride=Je.stride.map(mt=>mt/ht)}const[$e,We]=this.tokenizer._decode_asr(fe,{time_precision:ct,return_timestamps:Ue,force_full_sequences:Ne});oe.push({text:$e,...We})}return dt?oe[0]:oe}async _call_moonshine(le,Te){const Ue=!Array.isArray(le);Ue&&(le=[le]);const Ve=this.processor.feature_extractor.config.sampling_rate,Ne=await I(le,Ve),Re=[];for(const st of Ne){const dt=await this.processor(st),ct=Math.floor(st.length/Ve)*6,lt=await this.model.generate({max_new_tokens:ct,...Te,...dt}),ht=this.processor.batch_decode(lt,{skip_special_tokens:!0})[0];Re.push({text:ht})}return Ue?Re[0]:Re}}class ce extends se{constructor(le){super(le)}async _call(le,Te={}){const Ue=Array.isArray(le),Ve=await b(le),{pixel_values:Ne}=await this.processor(Ve),Re=[];for(const st of Ne){st.dims=[1,...st.dims];const dt=await this.model.generate({inputs:st,...Te}),ct=this.tokenizer.batch_decode(dt,{skip_special_tokens:!0}).map(lt=>({generated_text:lt.trim()}));Re.push(ct)}return Ue?Re:Re[0]}}class ke extends se{constructor(le){super(le)}async _call(le,{top_k:Te=5}={}){const Ue=await b(le),{pixel_values:Ve}=await this.processor(Ue),Ne=await this.model({pixel_values:Ve}),Re=this.model.config.id2label,st=[];for(const dt of Ne.logits){const ct=await(0,M.topk)(new M.Tensor("float32",(0,g.softmax)(dt.data),dt.dims),Te),lt=ct[0].tolist(),L=ct[1].tolist().map((oe,H)=>({label:Re?Re[oe]:`LABEL_${oe}`,score:lt[H]}));st.push(L)}return Array.isArray(le)?st:st[0]}}class Ie extends se{constructor(le){super(le),this.subtasks_mapping={panoptic:"post_process_panoptic_segmentation",instance:"post_process_instance_segmentation",semantic:"post_process_semantic_segmentation"}}async _call(le,{threshold:Te=.5,mask_threshold:Ue=.5,overlap_mask_area_threshold:Ve=.8,label_ids_to_fuse:Ne=null,target_sizes:Re=null,subtask:st=null}={}){if(Array.isArray(le)&&le.length!==1)throw Error("Image segmentation pipeline currently only supports a batch size of 1.");const ct=await b(le),lt=ct.map(We=>[We.height,We.width]),{pixel_values:ht,pixel_mask:L}=await this.processor(ct),oe=await this.model({pixel_values:ht,pixel_mask:L});let H=null;if(st!==null)H=this.subtasks_mapping[st];else for(let[We,Je]of Object.entries(this.subtasks_mapping))if(Je in this.processor.image_processor){H=this.processor.image_processor[Je].bind(this.processor.image_processor),st=We;break}const fe=this.model.config.id2label,$e=[];if(st==="panoptic"||st==="instance"){const We=H(oe,Te,Ue,Ve,Ne,Re??lt)[0],Je=We.segmentation;for(const ut of We.segments_info){const mt=new Uint8ClampedArray(Je.data.length);for(let kt=0;ktUe.replace("{}",L)),st=this.tokenizer(Re,{padding:this.model.config.model_type==="siglip"?"max_length":!0,truncation:!0}),{pixel_values:dt}=await this.processor(Ne),ct=await this.model({...st,pixel_values:dt}),lt=this.model.config.model_type==="siglip"?L=>L.sigmoid().data:L=>(0,g.softmax)(L.data),ht=[];for(const L of ct.logits_per_image){const H=[...lt(L)].map((fe,$e)=>({score:fe,label:Te[$e]}));H.sort((fe,$e)=>$e.score-fe.score),ht.push(H)}return Ve?ht:ht[0]}}class tt extends se{constructor(le){super(le)}async _call(le,{threshold:Te=.9,percentage:Ue=!1}={}){const Ve=Array.isArray(le);if(Ve&&le.length!==1)throw Error("Object detection pipeline currently only supports a batch size of 1.");const Ne=await b(le),Re=Ue?null:Ne.map(oe=>[oe.height,oe.width]),{pixel_values:st,pixel_mask:dt}=await this.processor(Ne),ct=await this.model({pixel_values:st,pixel_mask:dt}),lt=this.processor.image_processor.post_process_object_detection(ct,Te,Re),ht=this.model.config.id2label,L=lt.map(oe=>oe.boxes.map((H,fe)=>({score:oe.scores[fe],label:ht[oe.classes[fe]],box:K(H,!Ue)})));return Ve?L:L[0]}}class Ge extends se{constructor(le){super(le)}async _call(le,Te,{threshold:Ue=.1,top_k:Ve=null,percentage:Ne=!1}={}){const Re=Array.isArray(le),st=await b(le),dt=this.tokenizer(Te,{padding:!0,truncation:!0}),ct=await this.processor(st),lt=[];for(let ht=0;ht({score:We.scores[ut],label:We.labels[ut],box:K(Je,!Ne)}))}else{const We=this.processor.image_processor.post_process_object_detection(fe,Ue,oe,!0)[0];$e=We.boxes.map((Je,ut)=>({score:We.scores[ut],label:Te[We.classes[ut]],box:K(Je,!Ne)}))}$e.sort((We,Je)=>Je.score-We.score),Ve!==null&&($e=$e.slice(0,Ve)),lt.push($e)}return Re?lt:lt[0]}}class ye extends se{constructor(le){super(le)}async _call(le,Te,Ue={}){const Ve=(await b(le))[0],{pixel_values:Ne}=await this.processor(Ve),Re=`${Te}`,st=this.tokenizer(Re,{add_special_tokens:!1,padding:!0,truncation:!0}).input_ids,dt=await this.model.generate({inputs:Ne,max_length:this.model.config.decoder.max_position_embeddings,decoder_input_ids:st,...Ue}),lt=this.tokenizer.batch_decode(dt)[0].match(/(.*?)<\/s_answer>/);let ht=null;return lt&<.length>=2&&(ht=lt[1].trim()),[{answer:ht}]}}class J extends se{constructor(Te){super(Te);me(this,"DEFAULT_VOCODER_ID","Xenova/speecht5_hifigan");this.vocoder=Te.vocoder??null}async _call(Te,{speaker_embeddings:Ue=null}={}){return this.processor?this._call_text_to_spectrogram(Te,{speaker_embeddings:Ue}):this._call_text_to_waveform(Te)}async _call_text_to_waveform(Te){const Ue=this.tokenizer(Te,{padding:!0,truncation:!0}),{waveform:Ve}=await this.model(Ue),Ne=this.model.config.sampling_rate;return new v.RawAudio(Ve.data,Ne)}async _call_text_to_spectrogram(Te,{speaker_embeddings:Ue}){if(this.vocoder||(console.log("No vocoder specified, using default HifiGan vocoder."),this.vocoder=await D.AutoModel.from_pretrained(this.DEFAULT_VOCODER_ID,{dtype:"fp32"})),(typeof Ue=="string"||Ue instanceof URL)&&(Ue=new Float32Array(await(await fetch(Ue)).arrayBuffer())),Ue instanceof Float32Array)Ue=new M.Tensor("float32",Ue,[1,Ue.length]);else if(!(Ue instanceof M.Tensor))throw new Error("Speaker embeddings must be a `Tensor`, `Float32Array`, `string`, or `URL`.");const{input_ids:Ve}=this.tokenizer(Te,{padding:!0,truncation:!0}),{waveform:Ne}=await this.model.generate_speech(Ve,Ue,{vocoder:this.vocoder}),Re=this.processor.feature_extractor.config.sampling_rate;return new v.RawAudio(Ne.data,Re)}}class de extends se{constructor(le){super(le)}async _call(le){const Te=await b(le),Ue=await this.processor(Te),Ve=await this.model(Ue),Ne=[];for(const Re of Ve.reconstruction){const st=Re.squeeze().clamp_(0,1).mul_(255).round_().to("uint8");Ne.push(y.RawImage.fromTensor(st))}return Ne.length>1?Ne:Ne[0]}}class Ce extends se{constructor(le){super(le)}async _call(le){const Te=await b(le),Ue=await this.processor(Te),{predicted_depth:Ve}=await this.model(Ue),Ne=[];for(let Re=0;Re1?Ne:Ne[0]}}const Be=Object.freeze({"text-classification":{tokenizer:_.AutoTokenizer,pipeline:ne,model:D.AutoModelForSequenceClassification,default:{model:"Xenova/distilbert-base-uncased-finetuned-sst-2-english"},type:"text"},"token-classification":{tokenizer:_.AutoTokenizer,pipeline:W,model:D.AutoModelForTokenClassification,default:{model:"Xenova/bert-base-multilingual-cased-ner-hrl"},type:"text"},"question-answering":{tokenizer:_.AutoTokenizer,pipeline:U,model:D.AutoModelForQuestionAnswering,default:{model:"Xenova/distilbert-base-cased-distilled-squad"},type:"text"},"fill-mask":{tokenizer:_.AutoTokenizer,pipeline:q,model:D.AutoModelForMaskedLM,default:{model:"Xenova/bert-base-uncased"},type:"text"},summarization:{tokenizer:_.AutoTokenizer,pipeline:S,model:D.AutoModelForSeq2SeqLM,default:{model:"Xenova/distilbart-cnn-6-6"},type:"text"},translation:{tokenizer:_.AutoTokenizer,pipeline:w,model:D.AutoModelForSeq2SeqLM,default:{model:"Xenova/t5-small"},type:"text"},"text2text-generation":{tokenizer:_.AutoTokenizer,pipeline:$,model:D.AutoModelForSeq2SeqLM,default:{model:"Xenova/flan-t5-small"},type:"text"},"text-generation":{tokenizer:_.AutoTokenizer,pipeline:O,model:D.AutoModelForCausalLM,default:{model:"Xenova/gpt2"},type:"text"},"zero-shot-classification":{tokenizer:_.AutoTokenizer,pipeline:ae,model:D.AutoModelForSequenceClassification,default:{model:"Xenova/distilbert-base-uncased-mnli"},type:"text"},"audio-classification":{pipeline:we,model:D.AutoModelForAudioClassification,processor:j.AutoProcessor,default:{model:"Xenova/wav2vec2-base-superb-ks"},type:"audio"},"zero-shot-audio-classification":{tokenizer:_.AutoTokenizer,pipeline:re,model:D.AutoModel,processor:j.AutoProcessor,default:{model:"Xenova/clap-htsat-unfused"},type:"multimodal"},"automatic-speech-recognition":{tokenizer:_.AutoTokenizer,pipeline:xe,model:[D.AutoModelForSpeechSeq2Seq,D.AutoModelForCTC],processor:j.AutoProcessor,default:{model:"Xenova/whisper-tiny.en"},type:"multimodal"},"text-to-audio":{tokenizer:_.AutoTokenizer,pipeline:J,model:[D.AutoModelForTextToWaveform,D.AutoModelForTextToSpectrogram],processor:[j.AutoProcessor,null],default:{model:"Xenova/speecht5_tts"},type:"text"},"image-to-text":{tokenizer:_.AutoTokenizer,pipeline:ce,model:D.AutoModelForVision2Seq,processor:j.AutoProcessor,default:{model:"Xenova/vit-gpt2-image-captioning"},type:"multimodal"},"image-classification":{pipeline:ke,model:D.AutoModelForImageClassification,processor:j.AutoProcessor,default:{model:"Xenova/vit-base-patch16-224"},type:"multimodal"},"image-segmentation":{pipeline:Ie,model:[D.AutoModelForImageSegmentation,D.AutoModelForSemanticSegmentation,D.AutoModelForUniversalSegmentation],processor:j.AutoProcessor,default:{model:"Xenova/detr-resnet-50-panoptic"},type:"multimodal"},"zero-shot-image-classification":{tokenizer:_.AutoTokenizer,pipeline:Ee,model:D.AutoModel,processor:j.AutoProcessor,default:{model:"Xenova/clip-vit-base-patch32"},type:"multimodal"},"object-detection":{pipeline:tt,model:D.AutoModelForObjectDetection,processor:j.AutoProcessor,default:{model:"Xenova/detr-resnet-50"},type:"multimodal"},"zero-shot-object-detection":{tokenizer:_.AutoTokenizer,pipeline:Ge,model:D.AutoModelForZeroShotObjectDetection,processor:j.AutoProcessor,default:{model:"Xenova/owlvit-base-patch32"},type:"multimodal"},"document-question-answering":{tokenizer:_.AutoTokenizer,pipeline:ye,model:D.AutoModelForDocumentQuestionAnswering,processor:j.AutoProcessor,default:{model:"Xenova/donut-base-finetuned-docvqa"},type:"multimodal"},"image-to-image":{pipeline:de,model:D.AutoModelForImageToImage,processor:j.AutoProcessor,default:{model:"Xenova/swin2SR-classical-sr-x2-64"},type:"image"},"depth-estimation":{pipeline:Ce,model:D.AutoModelForDepthEstimation,processor:j.AutoProcessor,default:{model:"Xenova/dpt-large"},type:"image"},"feature-extraction":{tokenizer:_.AutoTokenizer,pipeline:ie,model:D.AutoModel,default:{model:"Xenova/all-MiniLM-L6-v2"},type:"text"},"image-feature-extraction":{processor:j.AutoProcessor,pipeline:ve,model:[D.AutoModelForImageFeatureExtraction,D.AutoModel],default:{model:"Xenova/vit-base-patch16-224-in21k"},type:"image"}}),Ze=Object.freeze({"sentiment-analysis":"text-classification",ner:"token-classification",asr:"automatic-speech-recognition","text-to-speech":"text-to-audio",embeddings:"feature-extraction"});async function te(je,le=null,{progress_callback:Te=null,config:Ue=null,cache_dir:Ve=null,local_files_only:Ne=!1,revision:Re="main",device:st=null,dtype:dt=null,model_file_name:ct=null,session_options:lt={}}={}){je=Ze[je]??je;const ht=Be[je.split("_",1)[0]];if(!ht)throw Error(`Unsupported pipeline: ${je}. Must be one of [${Object.keys(Be)}]`);le||(le=ht.default.model,console.log(`No model specified. Using default model: "${le}".`));const L={progress_callback:Te,config:Ue,cache_dir:Ve,local_files_only:Ne,revision:Re,device:st,dtype:dt,model_file_name:ct,session_options:lt},oe=new Map([["tokenizer",ht.tokenizer],["model",ht.model],["processor",ht.processor]]),H=await Ke(oe,le,L);H.task=je,(0,R.dispatchCallback)(Te,{status:"ready",task:je,model:le});const fe=ht.pipeline;return new fe(H)}async function Ke(je,le,Te){const Ue=Object.create(null),Ve=[];for(const[Ne,Re]of je.entries()){if(!Re)continue;let st;Array.isArray(Re)?st=new Promise(async(dt,ct)=>{var ht,L;let lt;for(const oe of Re){if(oe===null){dt(null);return}try{dt(await oe.from_pretrained(le,Te));return}catch(H){if((ht=H.message)!=null&&ht.includes("Unsupported model type"))lt=H;else if((L=H.message)!=null&&L.includes("Could not locate file"))lt=H;else{ct(H);return}}}ct(lt)}):st=Re.from_pretrained(le,Te),Ue[Ne]=st,Ve.push(st)}await Promise.all(Ve);for(const[Ne,Re]of Object.entries(Ue))Ue[Ne]=await Re;return Ue}},"./src/tokenizers.js":(Le,A,r)=>{r.r(A),r.d(A,{AlbertTokenizer:()=>$r,AutoTokenizer:()=>as,BartTokenizer:()=>Or,BertTokenizer:()=>Jr,BlenderbotSmallTokenizer:()=>Ln,BlenderbotTokenizer:()=>Dn,BloomTokenizer:()=>Er,CLIPTokenizer:()=>bn,CamembertTokenizer:()=>it,CodeGenTokenizer:()=>Mn,CodeLlamaTokenizer:()=>Ur,CohereTokenizer:()=>Tn,ConvBertTokenizer:()=>Rr,DebertaTokenizer:()=>cr,DebertaV2Tokenizer:()=>en,DistilBertTokenizer:()=>ar,ElectraTokenizer:()=>Dt,EsmTokenizer:()=>Vr,FalconTokenizer:()=>In,GPT2Tokenizer:()=>jr,GPTNeoXTokenizer:()=>On,GemmaTokenizer:()=>oi,Grok1Tokenizer:()=>Wr,HerbertTokenizer:()=>Ir,LlamaTokenizer:()=>wn,M2M100Tokenizer:()=>yn,MBart50Tokenizer:()=>lr,MBartTokenizer:()=>Ms,MPNetTokenizer:()=>An,MarianTokenizer:()=>zt,MgpstrTokenizer:()=>Rn,MobileBertTokenizer:()=>Ar,NllbTokenizer:()=>ur,NougatTokenizer:()=>Gr,PreTrainedTokenizer:()=>Nt,Qwen2Tokenizer:()=>Fn,RoFormerTokenizer:()=>Nr,RobertaTokenizer:()=>Os,SiglipTokenizer:()=>vn,SpeechT5Tokenizer:()=>zn,SqueezeBertTokenizer:()=>Zr,T5Tokenizer:()=>Vs,TokenizerModel:()=>ve,VitsTokenizer:()=>Bn,Wav2Vec2CTCTokenizer:()=>xn,WhisperTokenizer:()=>tn,XLMRobertaTokenizer:()=>ii,XLMTokenizer:()=>Tt,is_chinese_char:()=>q});var _=r("./src/utils/generic.js"),D=r("./src/utils/core.js"),j=r("./src/utils/hub.js"),X=r("./src/utils/maths.js"),R=r("./src/utils/tensor.js"),g=r("./src/utils/data-structures.js"),v=r("./node_modules/@huggingface/jinja/dist/index.js"),M=r("./src/models/whisper/common_whisper.js");async function y(Pe,P){const Q=await Promise.all([(0,j.getModelJSON)(Pe,"tokenizer.json",!0,P),(0,j.getModelJSON)(Pe,"tokenizer_config.json",!0,P)]);return P.legacy!==null&&(Q[1].legacy=P.legacy),Q}function b(Pe,P){const Q=[];let ue=0;for(const be of Pe.matchAll(P)){const Se=be[0];ue0&&Q.push(Se),ue=be.index+Se.length}return ue=19968&&Pe<=40959||Pe>=13312&&Pe<=19903||Pe>=131072&&Pe<=173791||Pe>=173824&&Pe<=177983||Pe>=177984&&Pe<=178207||Pe>=178208&&Pe<=183983||Pe>=63744&&Pe<=64255||Pe>=194560&&Pe<=195103}function $(Pe,P,Q){const ue=[];let be=0;for(;bethis.tokens_to_ids.get(Q)??this.unk_token_id)}convert_ids_to_tokens(P){return P.map(Q=>this.vocab[Q]??this.unk_token)}}class we extends ve{constructor(P){super(P),this.tokens_to_ids=K(P.vocab),this.unk_token_id=this.tokens_to_ids.get(P.unk_token),this.unk_token=P.unk_token,this.max_input_chars_per_word=P.max_input_chars_per_word??100,this.vocab=new Array(this.tokens_to_ids.size);for(const[Q,ue]of this.tokens_to_ids)this.vocab[ue]=Q}encode(P){const Q=[];for(const ue of P){const be=[...ue];if(be.length>this.max_input_chars_per_word){Q.push(this.unk_token);continue}let Se=!1,Qe=0;const pt=[];for(;Qe0&&(xt=this.config.continuing_subword_prefix+xt),this.tokens_to_ids.has(xt)){ft=xt;break}--gt}if(ft===null){Se=!0;break}pt.push(ft),Qe=gt}Se?Q.push(this.unk_token):Q.push(...pt)}return Q}}class re extends ve{constructor(P,Q){super(P);const ue=P.vocab.length;this.vocab=new Array(ue),this.scores=new Array(ue);for(let be=0;be[be,Se])),this.bos_token=" ",this.bos_token_id=this.tokens_to_ids.get(this.bos_token),this.eos_token=Q.eos_token,this.eos_token_id=this.tokens_to_ids.get(this.eos_token),this.unk_token=this.vocab[this.unk_token_id],this.minScore=(0,X.min)(this.scores)[0],this.unk_score=this.minScore-10,this.scores[this.unk_token_id]=this.unk_score,this.trie=new g.CharTrie,this.trie.extend(this.vocab),this.fuse_unk=!0}populateNodes(P){const Q=P.chars,ue=1;let be=0;for(;be{const Pe=[...Array.from({length:94},(be,Se)=>Se+33),...Array.from({length:12},(be,Se)=>Se+161),...Array.from({length:82},(be,Se)=>Se+174)],P=Pe.slice();let Q=0;for(let be=0;be<256;++be)Pe.includes(be)||(Pe.push(be),P.push(256+Q),Q+=1);const ue=P.map(be=>String.fromCharCode(be));return Object.fromEntries(Pe.map((be,Se)=>[be,ue[Se]]))})(),ce=(0,D.reverseDictionary)(xe);class ke extends ve{constructor(P){super(P),this.tokens_to_ids=K(P.vocab),this.unk_token_id=this.tokens_to_ids.get(P.unk_token),this.unk_token=P.unk_token,this.vocab=new Array(this.tokens_to_ids.size);for(const[ue,be]of this.tokens_to_ids)this.vocab[be]=ue;const Q=Array.isArray(P.merges[0]);this.merges=Q?P.merges:P.merges.map(ue=>ue.split(" ",2)),this.bpe_ranks=new Map(this.merges.map((ue,be)=>[JSON.stringify(ue),be])),this.end_of_word_suffix=P.end_of_word_suffix,this.continuing_subword_suffix=P.continuing_subword_suffix??null,this.byte_fallback=this.config.byte_fallback??!1,this.byte_fallback&&(this.text_encoder=new TextEncoder),this.ignore_merges=this.config.ignore_merges??!1,this.cache=new Map}bpe(P){if(P.length===0)return[];const Q=this.cache.get(P);if(Q!==void 0)return Q;const ue=Array.from(P);this.end_of_word_suffix&&(ue[ue.length-1]+=this.end_of_word_suffix);let be=[];if(ue.length>1){const Se=new g.PriorityQueue((gt,ft)=>gt.score`<0x${pt.toString(16).toUpperCase().padStart(2,"0")}>`);Qe.every(pt=>this.tokens_to_ids.has(pt))?Q.push(...Qe):Q.push(this.unk_token)}else Q.push(this.unk_token)}return Q}}class Ie extends ve{constructor(P,Q){super(P),this.tokens_to_ids=K(Q.target_lang?P.vocab[Q.target_lang]:P.vocab),this.bos_token=Q.bos_token,this.bos_token_id=this.tokens_to_ids.get(this.bos_token),this.eos_token=Q.eos_token,this.eos_token_id=this.tokens_to_ids.get(this.eos_token),this.pad_token=Q.pad_token,this.pad_token_id=this.tokens_to_ids.get(this.pad_token),this.unk_token=Q.unk_token,this.unk_token_id=this.tokens_to_ids.get(this.unk_token),this.vocab=new Array(this.tokens_to_ids.size);for(const[ue,be]of this.tokens_to_ids)this.vocab[be]=ue}encode(P){return P}}class Ee extends _.Callable{constructor(P){super(),this.config=P}static fromConfig(P){if(P===null)return null;switch(P.type){case"BertNormalizer":return new Ke(P);case"Precompiled":return new ys(P);case"Sequence":return new te(P);case"Replace":return new tt(P);case"NFC":return new Ge(P);case"NFKC":return new ye(P);case"NFKD":return new J(P);case"Strip":return new de(P);case"StripAccents":return new Ce(P);case"Lowercase":return new Be(P);case"Prepend":return new Ze(P);default:throw new Error(`Unknown Normalizer type: ${P.type}`)}}normalize(P){throw Error("normalize should be implemented in subclass.")}_call(P){return this.normalize(P)}}class tt extends Ee{normalize(P){const Q=I(this.config.pattern);return Q===null?P:P.replaceAll(Q,this.config.content)}}class Ge extends Ee{normalize(P){return P=P.normalize("NFC"),P}}class ye extends Ee{normalize(P){return P=P.normalize("NFKC"),P}}class J extends Ee{normalize(P){return P=P.normalize("NFKD"),P}}class de extends Ee{normalize(P){return this.config.strip_left&&this.config.strip_right?P=P.trim():(this.config.strip_left&&(P=P.trimStart()),this.config.strip_right&&(P=P.trimEnd())),P}}class Ce extends Ee{normalize(P){return P=W(P),P}}class Be extends Ee{normalize(P){return P=P.toLowerCase(),P}}class Ze extends Ee{normalize(P){return P=this.config.prepend+P,P}}class te extends Ee{constructor(P){super(P),this.normalizers=P.normalizers.map(Q=>Ee.fromConfig(Q))}normalize(P){return this.normalizers.reduce((Q,ue)=>ue.normalize(Q),P)}}class Ke extends Ee{_tokenize_chinese_chars(P){const Q=[];for(let ue=0;uethis.pre_tokenize_text(ue,Q)):this.pre_tokenize_text(P,Q)).flat()}_call(P,Q){return this.pre_tokenize(P,Q)}}class le extends je{constructor(P){super(),this.pattern=new RegExp(`[^\\s${w}]+|[${w}]`,"gu")}pre_tokenize_text(P,Q){return P.trim().match(this.pattern)||[]}}class Te extends je{constructor(P){super(),this.config=P,this.add_prefix_space=this.config.add_prefix_space,this.trim_offsets=this.config.trim_offsets,this.use_regex=this.config.use_regex??!0,this.pattern=new RegExp("'s|'t|'re|'ve|'m|'ll|'d| ?\\p{L}+| ?\\p{N}+| ?[^\\s\\p{L}\\p{N}]+|\\s+(?!\\S)|\\s+","gu"),this.byte_encoder=xe,this.text_encoder=new TextEncoder}pre_tokenize_text(P,Q){return this.add_prefix_space&&!P.startsWith(" ")&&(P=" "+P),(this.use_regex?P.match(this.pattern)||[]:[P]).map(be=>Array.from(this.text_encoder.encode(be),Se=>this.byte_encoder[Se]).join(""))}}class Ue extends je{constructor(P){super(),this.config=P,this.pattern=I(this.config.pattern,this.config.invert)}pre_tokenize_text(P,Q){var ue;return this.pattern===null?[]:this.config.invert?P.match(this.pattern)||[]:((ue=this.config.behavior)==null?void 0:ue.toLowerCase())==="removed"?P.split(this.pattern).filter(be=>be):b(P,this.pattern)}}class Ve extends je{constructor(P){super(),this.config=P,this.pattern=new RegExp(`[^${w}]+|[${w}]+`,"gu")}pre_tokenize_text(P,Q){return P.match(this.pattern)||[]}}class Ne extends je{constructor(P){super(),this.config=P;const Q=`[^\\d]+|\\d${this.config.individual_digits?"":"+"}`;this.pattern=new RegExp(Q,"gu")}pre_tokenize_text(P,Q){return P.match(this.pattern)||[]}}class Re extends _.Callable{constructor(P){super(),this.config=P}static fromConfig(P){if(P===null)return null;switch(P.type){case"TemplateProcessing":return new ct(P);case"ByteLevel":return new lt(P);case"RobertaProcessing":return new dt(P);case"BertProcessing":return new st(P);case"Sequence":return new ht(P);default:throw new Error(`Unknown PostProcessor type: ${P.type}`)}}post_process(P,...Q){throw Error("post_process should be implemented in subclass.")}_call(P,...Q){return this.post_process(P,...Q)}}class st extends Re{constructor(P){super(P),this.cls=P.cls[0],this.sep=P.sep[0]}post_process(P,Q=null,{add_special_tokens:ue=!0}={}){ue&&(P=(0,D.mergeArrays)([this.cls],P,[this.sep]));let be=new Array(P.length).fill(0);if(Q!==null){const Se=ue&&this instanceof dt?[this.sep]:[],Qe=ue?[this.sep]:[];P=(0,D.mergeArrays)(P,Se,Q,Qe),be=(0,D.mergeArrays)(be,new Array(Q.length+Se.length+Qe.length).fill(1))}return{tokens:P,token_type_ids:be}}}class dt extends st{}class ct extends Re{constructor(P){super(P),this.single=P.single,this.pair=P.pair}post_process(P,Q=null,{add_special_tokens:ue=!0}={}){const be=Q===null?this.single:this.pair;let Se=[],Qe=[];for(const pt of be)"SpecialToken"in pt?ue&&(Se.push(pt.SpecialToken.id),Qe.push(pt.SpecialToken.type_id)):"Sequence"in pt&&(pt.Sequence.id==="A"?(Se=(0,D.mergeArrays)(Se,P),Qe=(0,D.mergeArrays)(Qe,new Array(P.length).fill(pt.Sequence.type_id))):pt.Sequence.id==="B"&&(Se=(0,D.mergeArrays)(Se,Q),Qe=(0,D.mergeArrays)(Qe,new Array(Q.length).fill(pt.Sequence.type_id))));return{tokens:Se,token_type_ids:Qe}}}class lt extends Re{post_process(P,Q=null){return Q&&(P=(0,D.mergeArrays)(P,Q)),{tokens:P}}}class ht extends Re{constructor(P){super(P),this.processors=P.processors.map(Q=>Re.fromConfig(Q))}post_process(P,Q=null,ue={}){let be;for(const Se of this.processors)if(Se instanceof lt)P=Se.post_process(P).tokens,Q&&(Q=Se.post_process(Q).tokens);else{const Qe=Se.post_process(P,Q,ue);P=Qe.tokens,be=Qe.token_type_ids}return{tokens:P,token_type_ids:be}}}class L extends _.Callable{constructor(P){super(),this.config=P,this.added_tokens=[],this.end_of_word_suffix=null,this.trim_offsets=P.trim_offsets}static fromConfig(P){if(P===null)return null;switch(P.type){case"WordPiece":return new We(P);case"Metaspace":return new is(P);case"ByteLevel":return new Je(P);case"Replace":return new oe(P);case"ByteFallback":return new H(P);case"Fuse":return new fe(P);case"Strip":return new $e(P);case"Sequence":return new mt(P);case"CTC":return new ut(P);case"BPEDecoder":return new vt(P);default:throw new Error(`Unknown Decoder type: ${P.type}`)}}_call(P){return this.decode(P)}decode(P){return this.decode_chain(P).join("")}decode_chain(P){throw Error("`decode_chain` should be implemented in subclass.")}}class oe extends L{decode_chain(P){const Q=I(this.config.pattern);return Q===null?P:P.map(ue=>ue.replaceAll(Q,this.config.content))}}class H extends L{constructor(P){super(P),this.text_decoder=new TextDecoder}decode_chain(P){const Q=[];let ue=[];for(const be of P){let Se=null;if(be.length===6&&be.startsWith("<0x")&&be.endsWith(">")){const Qe=parseInt(be.slice(3,5),16);isNaN(Qe)||(Se=Qe)}if(Se!==null)ue.push(Se);else{if(ue.length>0){const Qe=this.text_decoder.decode(Uint8Array.from(ue));Q.push(Qe),ue=[]}Q.push(be)}}if(ue.length>0){const be=this.text_decoder.decode(Uint8Array.from(ue));Q.push(be),ue=[]}return Q}}class fe extends L{decode_chain(P){return[P.join("")]}}class $e extends L{constructor(P){super(P),this.content=this.config.content,this.start=this.config.start,this.stop=this.config.stop}decode_chain(P){return P.map(Q=>{let ue=0;for(let Se=0;Se(ue!==0&&(Q.startsWith(this.config.prefix)?Q=Q.replace(this.config.prefix,""):Q=" "+Q),this.cleanup&&(Q=ne(Q)),Q))}}class Je extends L{constructor(P){super(P),this.byte_decoder=ce,this.text_decoder=new TextDecoder("utf-8",{fatal:!1,ignoreBOM:!0}),this.end_of_word_suffix=null}convert_tokens_to_string(P){const Q=P.join(""),ue=new Uint8Array([...Q].map(Se=>this.byte_decoder[Se]));return this.text_decoder.decode(ue)}decode_chain(P){const Q=[];let ue=[];for(const be of P)this.added_tokens.find(Se=>Se.content===be)!==void 0?(ue.length>0&&(Q.push(this.convert_tokens_to_string(ue)),ue=[]),Q.push(be)):ue.push(be);return ue.length>0&&Q.push(this.convert_tokens_to_string(ue)),Q}}class ut extends L{constructor(P){super(P),this.pad_token=this.config.pad_token,this.word_delimiter_token=this.config.word_delimiter_token,this.cleanup=this.config.cleanup}convert_tokens_to_string(P){if(P.length===0)return"";const Q=[P[0]];for(let Se=1;SeSe!==this.pad_token).join("");return this.cleanup&&(be=ne(be).replaceAll(this.word_delimiter_token," ").trim()),be}decode_chain(P){return[this.convert_tokens_to_string(P)]}}class mt extends L{constructor(P){super(P),this.decoders=P.decoders.map(Q=>L.fromConfig(Q))}decode_chain(P){return this.decoders.reduce((Q,ue)=>ue.decode_chain(Q),P)}}class vt extends L{constructor(P){super(P),this.suffix=this.config.suffix}decode_chain(P){return P.map((Q,ue)=>Q.replaceAll(this.suffix,ue===P.length-1?"":" "))}}class kt extends L{decode_chain(P){let Q="";for(let ue=1;ueue.normalize("NFKC")).join("～"):P=P.normalize("NFKC"),P}}class Cs extends je{constructor(P){super(),this.tokenizers=P.pretokenizers.map(Q=>je.fromConfig(Q))}pre_tokenize_text(P,Q){return this.tokenizers.reduce((ue,be)=>be.pre_tokenize(ue,Q),[P])}}class Ds extends je{constructor(P){super()}pre_tokenize_text(P,Q){return P.match(/\w+|[^\w\s]+/g)||[]}}class sr extends je{constructor(P){super()}pre_tokenize_text(P,Q){return S(P)}}class Sr extends je{constructor(P){super(),this.config=P,this.pattern=I(this.config.pattern),this.content=this.config.content}pre_tokenize_text(P,Q){return this.pattern===null?[P]:[P.replaceAll(this.pattern,this.config.content)]}}const Yr=["bos_token","eos_token","unk_token","sep_token","pad_token","cls_token","mask_token"];function Us(Pe,P,Q,ue){for(const be of Object.keys(Pe)){const Se=P-Pe[be].length,Qe=Q(be),pt=new Array(Se).fill(Qe);Pe[be]=ue==="right"?(0,D.mergeArrays)(Pe[be],pt):(0,D.mergeArrays)(pt,Pe[be])}}function Pr(Pe,P){for(const Q of Object.keys(Pe))Pe[Q].length=P}class Nt extends _.Callable{constructor(Q,ue){super();me(this,"return_token_type_ids",!1);me(this,"padding_side","right");this._tokenizer_config=ue,this.normalizer=Ee.fromConfig(Q.normalizer),this.pre_tokenizer=je.fromConfig(Q.pre_tokenizer),this.model=ve.fromConfig(Q.model,ue),this.post_processor=Re.fromConfig(Q.post_processor),this.decoder=L.fromConfig(Q.decoder),this.special_tokens=[],this.all_special_ids=[],this.added_tokens=[];for(const be of Q.added_tokens){const Se=new ie(be);this.added_tokens.push(Se),this.model.tokens_to_ids.set(Se.content,Se.id),this.model.vocab[Se.id]=Se.content,Se.special&&(this.special_tokens.push(Se.content),this.all_special_ids.push(Se.id))}if(this.additional_special_tokens=ue.additional_special_tokens??[],this.special_tokens.push(...this.additional_special_tokens),this.special_tokens=[...new Set(this.special_tokens)],this.decoder&&(this.decoder.added_tokens=this.added_tokens,this.decoder.end_of_word_suffix=this.model.end_of_word_suffix),this.added_tokens_regex=this.added_tokens.length>0?new RegExp(this.added_tokens.slice().sort((be,Se)=>Se.content.length-be.content.length).map(be=>`${be.lstrip?"\\s*":""}(${(0,D.escapeRegExp)(be.content)})${be.rstrip?"\\s*":""}`).join("|")):null,this.mask_token=this.getToken("mask_token"),this.mask_token_id=this.model.tokens_to_ids.get(this.mask_token),this.pad_token=this.getToken("pad_token","eos_token"),this.pad_token_id=this.model.tokens_to_ids.get(this.pad_token),this.sep_token=this.getToken("sep_token"),this.sep_token_id=this.model.tokens_to_ids.get(this.sep_token),this.unk_token=this.getToken("unk_token"),this.unk_token_id=this.model.tokens_to_ids.get(this.unk_token),this.bos_token=this.getToken("bos_token"),this.bos_token_id=this.model.tokens_to_ids.get(this.bos_token),this.eos_token=this.getToken("eos_token"),this.eos_token_id=this.model.tokens_to_ids.get(this.eos_token),this.model_max_length=ue.model_max_length,this.remove_space=ue.remove_space,this.clean_up_tokenization_spaces=ue.clean_up_tokenization_spaces??!0,this.do_lowercase_and_remove_accent=ue.do_lowercase_and_remove_accent??!1,ue.padding_side&&(this.padding_side=ue.padding_side),this.legacy=!1,this.chat_template=ue.chat_template??null,Array.isArray(this.chat_template)){const be=Object.create(null);for(const{name:Se,template:Qe}of this.chat_template){if(typeof Se!="string"||typeof Qe!="string")throw new Error('Chat template must be a list of objects with "name" and "template" properties');be[Se]=Qe}this.chat_template=be}this._compiled_template_cache=new Map}getToken(...Q){for(const ue of Q){const be=this._tokenizer_config[ue];if(be)if(typeof be=="object"){if(be.__type==="AddedToken")return be.content;throw Error(`Unknown token: ${be}`)}else return be}return null}static async from_pretrained(Q,{progress_callback:ue=null,config:be=null,cache_dir:Se=null,local_files_only:Qe=!1,revision:pt="main",legacy:gt=null}={}){const ft=await y(Q,{progress_callback:ue,config:be,cache_dir:Se,local_files_only:Qe,revision:pt,legacy:gt});return new this(...ft)}_call(Q,{text_pair:ue=null,add_special_tokens:be=!0,padding:Se=!1,truncation:Qe=null,max_length:pt=null,return_tensor:gt=!0,return_token_type_ids:ft=null}={}){const xt=Array.isArray(Q);let Kt;if(xt){if(Q.length===0)throw Error("text array must be non-empty");if(ue!==null){if(Array.isArray(ue)){if(Q.length!==ue.length)throw Error("text and text_pair must have the same length")}else throw Error("text_pair must also be an array");Kt=Q.map((us,Fs)=>this._encode_plus(us,{text_pair:ue[Fs],add_special_tokens:be,return_token_type_ids:ft}))}else Kt=Q.map(us=>this._encode_plus(us,{add_special_tokens:be,return_token_type_ids:ft}))}else{if(Q==null)throw Error("text may not be null or undefined");if(Array.isArray(ue))throw Error("When specifying `text_pair`, since `text` is a string, `text_pair` must also be a string (i.e., not an array).");Kt=[this._encode_plus(Q,{text_pair:ue,add_special_tokens:be,return_token_type_ids:ft})]}if(pt===null?Se==="max_length"?pt=this.model_max_length:pt=(0,X.max)(Kt.map(us=>us.input_ids.length))[0]:Qe||console.warn("Truncation was not explicitly activated but `max_length` is provided a specific value, please use `truncation=true` to explicitly truncate examples to max length."),pt=Math.min(pt,this.model_max_length??1/0),Se||Qe)for(let us=0;uspt?Qe&&Pr(Kt[us],pt):Se&&Us(Kt[us],pt,Fs=>Fs==="input_ids"?this.pad_token_id:0,this.padding_side));const ms={};if(gt){if(!(Se&&Qe)&&Kt.some(Fs=>{var Bt;for(const rs of Object.keys(Fs))if(Fs[rs].length!==((Bt=Kt[0][rs])==null?void 0:Bt.length))return!0;return!1}))throw Error("Unable to create tensor, you should probably activate truncation and/or padding with 'padding=true' and 'truncation=true' to have batched tensors with the same length.");const us=[Kt.length,Kt[0].input_ids.length];for(const Fs of Object.keys(Kt[0]))ms[Fs]=new R.Tensor("int64",BigInt64Array.from(Kt.flatMap(Bt=>Bt[Fs]).map(BigInt)),us)}else{for(const us of Object.keys(Kt[0]))ms[us]=Kt.map(Fs=>Fs[us]);if(!xt)for(const us of Object.keys(ms))ms[us]=ms[us][0]}return ms}_encode_text(Q){return Q===null?null:(this.added_tokens_regex?Q.split(this.added_tokens_regex).filter(Se=>Se):[Q]).map((Se,Qe)=>{if(this.added_tokens.find(gt=>gt.content===Se)!==void 0)return Se;{if(this.remove_space===!0&&(Se=Se.trim().split(/\s+/).join(" ")),this.do_lowercase_and_remove_accent&&(Se=U(Se)),this.normalizer!==null&&(Se=this.normalizer(Se)),Se.length===0)return[];const gt=this.pre_tokenizer!==null?this.pre_tokenizer(Se,{section_index:Qe}):[Se];return this.model(gt)}}).flat()}_encode_plus(Q,{text_pair:ue=null,add_special_tokens:be=!0,return_token_type_ids:Se=null}={}){const{tokens:Qe,token_type_ids:pt}=this._tokenize_helper(Q,{pair:ue,add_special_tokens:be}),gt=this.model.convert_tokens_to_ids(Qe),ft={input_ids:gt,attention_mask:new Array(gt.length).fill(1)};return(Se??this.return_token_type_ids)&&pt&&(ft.token_type_ids=pt),ft}_tokenize_helper(Q,{pair:ue=null,add_special_tokens:be=!1}={}){const Se=this._encode_text(Q),Qe=this._encode_text(ue);return this.post_processor?this.post_processor(Se,Qe,{add_special_tokens:be}):{tokens:(0,D.mergeArrays)(Se??[],Qe??[])}}tokenize(Q,{pair:ue=null,add_special_tokens:be=!1}={}){return this._tokenize_helper(Q,{pair:ue,add_special_tokens:be}).tokens}encode(Q,{text_pair:ue=null,add_special_tokens:be=!0,return_token_type_ids:Se=null}={}){return this._encode_plus(Q,{text_pair:ue,add_special_tokens:be,return_token_type_ids:Se}).input_ids}batch_decode(Q,ue={}){return Q instanceof R.Tensor&&(Q=Q.tolist()),Q.map(be=>this.decode(be,ue))}decode(Q,ue={}){if(Q instanceof R.Tensor&&(Q=se(Q)),!Array.isArray(Q)||Q.length===0||!(0,D.isIntegralNumber)(Q[0]))throw Error("token_ids must be a non-empty array of integers.");return this.decode_single(Q,ue)}decode_single(Q,{skip_special_tokens:ue=!1,clean_up_tokenization_spaces:be=null}){let Se=this.model.convert_ids_to_tokens(Q);ue&&(Se=Se.filter(pt=>!this.special_tokens.includes(pt)));let Qe=this.decoder?this.decoder(Se):Se.join(" ");return this.decoder&&this.decoder.end_of_word_suffix&&(Qe=Qe.replaceAll(this.decoder.end_of_word_suffix," "),ue&&(Qe=Qe.trim())),(be??this.clean_up_tokenization_spaces)&&(Qe=ne(Qe)),Qe}get_chat_template({chat_template:Q=null,tools:ue=null}={}){if(this.chat_template&&typeof this.chat_template=="object"){const be=this.chat_template;if(Q!==null&&Object.hasOwn(be,Q))Q=be[Q];else if(Q===null)if(ue!==null&&"tool_use"in be)Q=be.tool_use;else if("default"in be)Q=be.default;else throw Error(`This model has multiple chat templates with no default specified! Please either pass a chat template or the name of the template you wish to use to the 'chat_template' argument. Available template names are ${Object.keys(be).sort()}.`)}else if(Q===null)if(this.chat_template)Q=this.chat_template;else throw Error("Cannot use apply_chat_template() because tokenizer.chat_template is not set and no template argument was passed! For information about writing templates and setting the tokenizer.chat_template attribute, please see the documentation at https://huggingface.co/docs/transformers/main/en/chat_templating");return Q}apply_chat_template(Q,{tools:ue=null,documents:be=null,chat_template:Se=null,add_generation_prompt:Qe=!1,tokenize:pt=!0,padding:gt=!1,truncation:ft=!1,max_length:xt=null,return_tensor:Kt=!0,return_dict:ms=!1,tokenizer_kwargs:us={},...Fs}={}){if(Se=this.get_chat_template({chat_template:Se,tools:ue}),typeof Se!="string")throw Error(`chat_template must be a string, but got ${typeof Se}`);let Bt=this._compiled_template_cache.get(Se);Bt===void 0&&(Bt=new v.Template(Se),this._compiled_template_cache.set(Se,Bt));const rs=Object.create(null);for(const Ws of Yr){const ze=this.getToken(Ws);ze&&(rs[Ws]=ze)}const rr=Bt.render({messages:Q,add_generation_prompt:Qe,tools:ue,documents:be,...rs,...Fs});if(pt){const Ws=this._call(rr,{add_special_tokens:!1,padding:gt,truncation:ft,max_length:xt,return_tensor:Kt,...us});return ms?Ws:Ws.input_ids}return rr}}class Jr extends Nt{constructor(){super(...arguments);me(this,"return_token_type_ids",!0)}}class $r extends Nt{constructor(){super(...arguments);me(this,"return_token_type_ids",!0)}}class Ar extends Nt{constructor(){super(...arguments);me(this,"return_token_type_ids",!0)}}class Zr extends Nt{constructor(){super(...arguments);me(this,"return_token_type_ids",!0)}}class cr extends Nt{constructor(){super(...arguments);me(this,"return_token_type_ids",!0)}}class en extends Nt{constructor(){super(...arguments);me(this,"return_token_type_ids",!0)}}class Ir extends Nt{constructor(){super(...arguments);me(this,"return_token_type_ids",!0)}}class Rr extends Nt{constructor(){super(...arguments);me(this,"return_token_type_ids",!0)}}class Nr extends Nt{constructor(){super(...arguments);me(this,"return_token_type_ids",!0)}}class ar extends Nt{}class it extends Nt{}class Tt extends Nt{constructor(Q,ue){super(Q,ue);me(this,"return_token_type_ids",!0);console.warn('WARNING: `XLMTokenizer` is not yet supported by Hugging Face\'s "fast" tokenizers library. Therefore, you may experience slightly inaccurate results.')}}class Dt extends Nt{constructor(){super(...arguments);me(this,"return_token_type_ids",!0)}}class Vs extends Nt{}class jr extends Nt{}class Or extends Nt{}class Ms extends Nt{constructor(P,Q){super(P,Q),this.languageRegex=/^[a-z]{2}_[A-Z]{2}$/,this.language_codes=this.special_tokens.filter(ue=>this.languageRegex.test(ue)),this.lang_to_token=ue=>ue}_build_translation_inputs(P,Q,ue){return fr(this,P,Q,ue)}}class lr extends Ms{}class Os extends Nt{}class Er extends Nt{}const es="▁";class wn extends Nt{constructor(Q,ue){super(Q,ue);me(this,"padding_side","left");this.legacy=ue.legacy??!0,this.legacy||(this.normalizer=null,this.pre_tokenizer=new At({replacement:es,add_prefix_space:!0,prepend_scheme:"first"}))}_encode_text(Q){if(Q===null)return null;if(this.legacy||Q.length===0)return super._encode_text(Q);let ue=super._encode_text(es+Q.replaceAll(es," "));return ue.length>1&&ue[0]===es&&this.special_tokens.includes(ue[1])&&(ue=ue.slice(1)),ue}}class Ur extends Nt{}class ii extends Nt{}class An extends Nt{}class In extends Nt{}class On extends Nt{}class Vr extends Nt{}class Fn extends Nt{}class oi extends Nt{}class Wr extends Nt{}function fr(Pe,P,Q,ue){if(!("language_codes"in Pe)||!Array.isArray(Pe.language_codes))throw new Error("Tokenizer must have `language_codes` attribute set and it should be an array of language ids.");if(!("languageRegex"in Pe)||!(Pe.languageRegex instanceof RegExp))throw new Error("Tokenizer must have `languageRegex` attribute set and it should be a regular expression.");if(!("lang_to_token"in Pe)||typeof Pe.lang_to_token!="function")throw new Error("Tokenizer must have `lang_to_token` attribute set and it should be a function.");const be=ue.src_lang,Se=ue.tgt_lang;if(!Pe.language_codes.includes(Se))throw new Error(`Target language code "${Se}" is not valid. Must be one of: {${Pe.language_codes.join(", ")}}`);if(be!==void 0){if(!Pe.language_codes.includes(be))throw new Error(`Source language code "${be}" is not valid. Must be one of: {${Pe.language_codes.join(", ")}}`);for(const Qe of Pe.post_processor.config.single)if("SpecialToken"in Qe&&Pe.languageRegex.test(Qe.SpecialToken.id)){Qe.SpecialToken.id=Pe.lang_to_token(be);break}}return ue.forced_bos_token_id=Pe.model.convert_tokens_to_ids([Pe.lang_to_token(Se)])[0],Pe._call(P,Q)}class ur extends Nt{constructor(P,Q){super(P,Q),this.languageRegex=/^[a-z]{3}_[A-Z][a-z]{3}$/,this.language_codes=this.special_tokens.filter(ue=>this.languageRegex.test(ue)),this.lang_to_token=ue=>ue}_build_translation_inputs(P,Q,ue){return fr(this,P,Q,ue)}}class yn extends Nt{constructor(P,Q){super(P,Q),this.languageRegex=/^__[a-z]{2,3}__$/,this.language_codes=this.special_tokens.filter(ue=>this.languageRegex.test(ue)).map(ue=>ue.slice(2,-2)),this.lang_to_token=ue=>`__${ue}__`}_build_translation_inputs(P,Q,ue){return fr(this,P,Q,ue)}}class tn extends Nt{get timestamp_begin(){return this.model.convert_tokens_to_ids(["<|notimestamps|>"])[0]+1}_decode_asr(P,{return_timestamps:Q=!1,return_language:ue=!1,time_precision:be=null,force_full_sequences:Se=!0}={}){if(be===null)throw Error("Must specify time_precision");let Qe=null;const pt=Q==="word";function gt(){return{language:Qe,timestamp:[null,null],text:""}}const ft=[];let xt=gt(),Kt=0;const ms=this.timestamp_begin,Fs=ms+1500;let Bt=[],rs=[],rr=!1,Ws=null;const ze=new Set(this.all_special_ids);for(const ks of P){const Xs=ks.tokens,Ot=pt?ks.token_timestamps:null;let ir=null,_r=ms;if("stride"in ks){const[yt,qt,Ls]=ks.stride;if(Kt-=qt,Ws=yt-Ls,qt&&(_r=qt/be+ms),Ls)for(let $s=Xs.length-1;$s>=0;--$s){const Gs=Number(Xs[$s]);if(Gs>=ms){if(ir!==null&&(Gs-ms)*be=ms&&qt<=Fs){const Ls=(qt-ms)*be+Kt,$s=(0,X.round)(Ls,2);if(ir!==null&&qt>=ir)rr=!0;else if(rr||Bt.length>0&&qt<_r)rr=!1;else if(xt.timestamp[0]===null)xt.timestamp[0]=$s;else if($s!==xt.timestamp[0]){xt.timestamp[1]=$s,Bt.push(fs),pt&&rs.push(Ss);const[Gs,$t]=this.findLongestCommonSequence(Bt,rs),sn=this.decode(Gs);xt.text=sn,pt&&(xt.words=this.collateWordTimestamps(Gs,$t,Qe)),ft.push(xt),Bt=[],fs=[],rs=[],Ss=[],xt=gt()}}else if(fs.push(qt),pt){let Ls=(0,X.round)(Ot[yt]+Kt,2),$s;if(yt+10?(Bt.push(fs),pt&&rs.push(Ss)):Bt.every(yt=>yt.length===0)&&(xt=gt(),Bt=[],fs=[],rs=[],Ss=[])}if(Bt.length>0){if(Se&&Q)throw new Error("Whisper did not predict an ending timestamp, which can happen if audio is cut off in the middle of a word. Also make sure WhisperTimeStampLogitsProcessor was used during generation.");const[ks,Xs]=this.findLongestCommonSequence(Bt,rs),Ot=this.decode(ks);xt.text=Ot,pt&&(xt.words=this.collateWordTimestamps(ks,Xs,Qe)),ft.push(xt)}let Js=Object.create(null);const Fr=ft.map(ks=>ks.text).join("");if(Q||ue){for(let ks=0;ks0;let pt=Qe?[]:null,gt=Qe?Q[0]:null;for(let ft=1;ftqt===_r[Ls]&>[Fr+Ls]<=Q[ft][Ot+Ls]).length:fs=Xs.filter((qt,Ls)=>qt===_r[Ls]).length;const Ss=Js/1e4,yt=fs/Js+Ss;fs>1&&yt>Kt&&(Kt=yt,ms=[Fr,ks,Ot,ir])}const[Fs,Bt,rs,rr]=ms,Ws=Math.floor((Bt+Fs)/2),ze=Math.floor((rr+rs)/2);Se.push(...ue.slice(0,Ws)),ue=xt.slice(ze),be=ue.length,Qe&&(pt.push(...gt.slice(0,Ws)),gt=Q[ft].slice(ze))}return Se.push(...ue),Qe?(pt.push(...gt),[Se,pt]):[Se,[]]}collateWordTimestamps(P,Q,ue){const[be,Se,Qe]=this.combineTokensIntoWords(P,ue),pt=[];for(let gt=0;gt=be){const pt=((Qe-be)*ue).toFixed(2);Se.push(`<|${pt}|>`),Se.push([])}else Se[Se.length-1].push(Qe);return Se=Se.map(Qe=>typeof Qe=="string"?Qe:super.decode(Qe,Q)),Se.join("")}splitTokensOnUnicode(P){const Q=this.decode(P,{decode_with_timestamps:!0}),ue="�",be=[],Se=[],Qe=[];let pt=[],gt=[],ft=0;for(let xt=0;xt=this.model.tokens_to_ids.get("<|endoftext|>"),Fs=xt.startsWith(" "),Bt=xt.trim(),rs=gt.test(Bt);if(us||Fs||rs||Se.length===0)Se.push(xt),Qe.push(Kt),pt.push(ms);else{const rr=Se.length-1;Se[rr]+=xt,Qe[rr].push(...Kt),pt[rr].push(...ms)}}return[Se,Qe,pt]}mergePunctuations(P,Q,ue,be,Se){const Qe=structuredClone(P),pt=structuredClone(Q),gt=structuredClone(ue);let ft=Qe.length-2,xt=Qe.length-1;for(;ft>=0;)Qe[ft].startsWith(" ")&&be.includes(Qe[ft].trim())?(Qe[xt]=Qe[ft]+Qe[xt],pt[xt]=(0,D.mergeArrays)(pt[ft],pt[xt]),gt[xt]=(0,D.mergeArrays)(gt[ft],gt[xt]),Qe[ft]="",pt[ft]=[],gt[ft]=[]):xt=ft,--ft;for(ft=0,xt=1;xtKt),pt.filter(Kt=>Kt.length>0),gt.filter(Kt=>Kt.length>0)]}}class Mn extends Nt{}class bn extends Nt{}class vn extends Nt{}class zt extends Nt{constructor(P,Q){super(P,Q),this.languageRegex=/^(>>\w+<<)\s*/g,this.supported_language_codes=this.model.vocab.filter(ue=>this.languageRegex.test(ue)),console.warn('WARNING: `MarianTokenizer` is not yet supported by Hugging Face\'s "fast" tokenizers library. Therefore, you may experience slightly inaccurate results.')}_encode_text(P){if(P===null)return null;const[Q,...ue]=P.trim().split(this.languageRegex);if(ue.length===0)return super._encode_text(Q);if(ue.length===2){const[be,Se]=ue;return this.supported_language_codes.includes(be)||console.warn(`Unsupported language code "${be}" detected, which may lead to unexpected behavior. Should be one of: ${JSON.stringify(this.supported_language_codes)}`),(0,D.mergeArrays)([be],super._encode_text(Se))}}}class xn extends Nt{}class Dn extends Nt{}class Ln extends Nt{}class zn extends Nt{}class Gr extends Nt{}class Bn extends Nt{constructor(P,Q){super(P,Q),this.decoder=new kt({})}}class Tn extends Nt{}class Rn extends Nt{}class as{static async from_pretrained(P,{progress_callback:Q=null,config:ue=null,cache_dir:be=null,local_files_only:Se=!1,revision:Qe="main",legacy:pt=null}={}){var ms;const[gt,ft]=await y(P,{progress_callback:Q,config:ue,cache_dir:be,local_files_only:Se,revision:Qe,legacy:pt}),xt=((ms=ft.tokenizer_class)==null?void 0:ms.replace(/Fast$/,""))??"PreTrainedTokenizer";let Kt=this.TOKENIZER_CLASS_MAPPING[xt];return Kt||(console.warn(`Unknown tokenizer class "${xt}", attempting to construct from base class.`),Kt=Nt),new Kt(gt,ft)}}me(as,"TOKENIZER_CLASS_MAPPING",{T5Tokenizer:Vs,DistilBertTokenizer:ar,CamembertTokenizer:it,DebertaTokenizer:cr,DebertaV2Tokenizer:en,BertTokenizer:Jr,HerbertTokenizer:Ir,ConvBertTokenizer:Rr,RoFormerTokenizer:Nr,XLMTokenizer:Tt,ElectraTokenizer:Dt,MobileBertTokenizer:Ar,SqueezeBertTokenizer:Zr,AlbertTokenizer:$r,GPT2Tokenizer:jr,BartTokenizer:Or,MBartTokenizer:Ms,MBart50Tokenizer:lr,RobertaTokenizer:Os,WhisperTokenizer:tn,CodeGenTokenizer:Mn,CLIPTokenizer:bn,SiglipTokenizer:vn,MarianTokenizer:zt,BloomTokenizer:Er,NllbTokenizer:ur,M2M100Tokenizer:yn,LlamaTokenizer:wn,CodeLlamaTokenizer:Ur,XLMRobertaTokenizer:ii,MPNetTokenizer:An,FalconTokenizer:In,GPTNeoXTokenizer:On,EsmTokenizer:Vr,Wav2Vec2CTCTokenizer:xn,BlenderbotTokenizer:Dn,BlenderbotSmallTokenizer:Ln,SpeechT5Tokenizer:zn,NougatTokenizer:Gr,VitsTokenizer:Bn,Qwen2Tokenizer:Fn,GemmaTokenizer:oi,Grok1Tokenizer:Wr,CohereTokenizer:Tn,MgpstrTokenizer:Rn,PreTrainedTokenizer:Nt})},"./src/utils/audio.js":(Le,A,r)=>{r.r(A),r.d(A,{RawAudio:()=>we,hamming:()=>b,hanning:()=>y,mel_filter_bank:()=>q,read_audio:()=>v,spectrogram:()=>O,window_function:()=>ae});var _=r("./src/utils/hub.js"),D=r("./src/utils/maths.js"),j=r("./src/utils/core.js"),X=r("./src/env.js"),R=r("?7a2c"),g=r("./src/utils/tensor.js");async function v(re,xe){if(typeof AudioContext>"u")throw Error("Unable to load audio from path/URL since `AudioContext` is not available in your environment. Instead, audio data should be passed directly to the pipeline/processor. For more information and some example code, see https://huggingface.co/docs/transformers.js/guides/node-audio-processing.");const ce=await(await(0,_.getFile)(re)).arrayBuffer(),ke=new AudioContext({sampleRate:xe});typeof xe>"u"&&console.warn(`No sampling rate provided, using default of ${ke.sampleRate}Hz.`);const Ie=await ke.decodeAudioData(ce);let Ee;if(Ie.numberOfChannels===2){const tt=Math.sqrt(2),Ge=Ie.getChannelData(0),ye=Ie.getChannelData(1);Ee=new Float32Array(Ge.length);for(let J=0;J2595*Math.log10(1+re/700),kaldi:re=>1127*Math.log(1+re/700),slaney:(re,xe=1e3,ce=15,ke=27/Math.log(6.4))=>re>=xe?ce+Math.log(re/xe)*ke:3*re/200};function K(re,xe="htk"){const ce=I[xe];if(!ce)throw new Error('mel_scale should be one of "htk", "slaney" or "kaldi".');return typeof re=="number"?ce(re):re.map(ke=>ce(ke))}const se={htk:re=>700*(10**(re/2595)-1),kaldi:re=>700*(Math.exp(re/1127)-1),slaney:(re,xe=1e3,ce=15,ke=Math.log(6.4)/27)=>re>=ce?xe*Math.exp(ke*(re-ce)):200*re/3};function ne(re,xe="htk"){const ce=se[xe];if(!ce)throw new Error('mel_scale should be one of "htk", "slaney" or "kaldi".');return typeof re=="number"?ce(re):re.map(ke=>ce(ke))}function W(re,xe){const ce=Float64Array.from({length:xe.length-1},(tt,Ge)=>xe[Ge+1]-xe[Ge]),ke=Array.from({length:re.length},()=>new Array(xe.length));for(let tt=0;ttnew Array(re.length));for(let tt=0;ttre+ke*Ee)}function q(re,xe,ce,ke,Ie,Ee=null,tt="htk",Ge=!1){if(Ee!==null&&Ee!=="slaney")throw new Error('norm must be one of null or "slaney"');const ye=K(ce,tt),J=K(ke,tt),de=U(ye,J,xe+2);let Ce=ne(de,tt),Be;if(Ge){const te=Ie/(re*2);Be=K(Float64Array.from({length:re},(Ke,je)=>je*te),tt),Ce=de}else Be=U(0,Math.floor(Ie/2),re);const Ze=W(Be,Ce);if(Ee!==null&&Ee==="slaney")for(let te=0;teIe)throw Error(`frame_length (${ce}) may not be larger than fft_length (${Ie})`);if(Ne!==ce)throw new Error(`Length of the window (${Ne}) must equal frame_length (${ce})`);if(ke<=0)throw new Error("hop_length must be greater than zero");if(Ee===null&&de!==null)throw new Error("You have provided `mel_filters` but `power` is `None`. Mel spectrogram computation is not yet supported for complex-valued spectrogram. Specify `power` to fix this issue.");if(tt){if(Ge!=="reflect")throw new Error(`pad_mode="${Ge}" not implemented yet.`);const We=Math.floor((Ie-1)/2)+1;re=$(re,We,We)}let Re=Math.floor(1+Math.floor((re.length-ce)/ke));le!==null&&ReRe?Ue&&(ct=Te):ct=dt=Te);const lt=new D.FFT(Ie),ht=new Float64Array(Ie),L=new Float64Array(lt.outputBufferSize),oe=new Float32Array(st*ct);for(let We=0;We=1;--mt)ht[mt]-=J*ht[mt-1];ht[0]*=1-J}for(let mt=0;mtMath.pow(Ge,.85));break;default:throw new Error(`Unknown window type ${xe}.`)}if(ce&&(tt=tt.subarray(0,re)),ke===null)return tt;if(re>ke)throw new Error(`Length of the window (${re}) may not be larger than frame_length (${ke})`);return tt}function ie(re,xe){let ce=44;const ke=new ArrayBuffer(ce+re.length*4),Ie=new DataView(ke);ve(Ie,0,"RIFF"),Ie.setUint32(4,36+re.length*4,!0),ve(Ie,8,"WAVE"),ve(Ie,12,"fmt "),Ie.setUint32(16,16,!0),Ie.setUint16(20,3,!0),Ie.setUint16(22,1,!0),Ie.setUint32(24,xe,!0),Ie.setUint32(28,xe*4,!0),Ie.setUint16(32,4,!0),Ie.setUint16(34,32,!0),ve(Ie,36,"data"),Ie.setUint32(40,re.length*4,!0);for(let Ee=0;Ee{let Ee=await Ie.arrayBuffer();R.writeFileSync(ke,Buffer.from(Ee))};else throw new Error("Unable to save because filesystem is disabled in this environment.");await ce(xe,this.toBlob())}}},"./src/utils/constants.js":(Le,A,r)=>{r.r(A),r.d(A,{CHAT_TEMPLATE_NAME:()=>g,CONFIG_NAME:()=>D,FEATURE_EXTRACTOR_NAME:()=>j,GENERATION_CONFIG_NAME:()=>v,GITHUB_ISSUE_URL:()=>_,IMAGE_PROCESSOR_NAME:()=>X,PROCESSOR_NAME:()=>R});const _="https://github.com/huggingface/transformers.js/issues/new/choose",D="config.json",j="preprocessor_config.json",X=j,R="processor_config.json",g="chat_template.json",v="generation_config.json"},"./src/utils/core.js":(Le,A,r)=>{r.r(A),r.d(A,{calculateDimensions:()=>v,calculateReflectOffset:()=>I,count:()=>W,dispatchCallback:()=>_,escapeRegExp:()=>j,isIntegralNumber:()=>R,isNullishDimension:()=>g,isTypedArray:()=>X,len:()=>ne,mergeArrays:()=>y,pick:()=>se,pop:()=>M,product:()=>b,reverseDictionary:()=>D,saveBlob:()=>K});function _(U,q){U&&U(q)}function D(U){return Object.fromEntries(Object.entries(U).map(([q,$])=>[$,q]))}function j(U){return U.replace(/[.*+?^${}()|[\]\\]/g,"\\$&")}function X(U){var q,$,S;return((S=($=(q=U==null?void 0:U.prototype)==null?void 0:q.__proto__)==null?void 0:$.constructor)==null?void 0:S.name)==="TypedArray"}function R(U){return Number.isInteger(U)||typeof U=="bigint"}function g(U){return U==null||U===-1}function v(U){const q=[];let $=U;for(;Array.isArray($);)q.push($.length),$=$[0];return q}function M(U,q,$=void 0){const S=U[q];if(S!==void 0)return delete U[q],S;if($===void 0)throw Error(`Key ${q} does not exist in object.`);return $}function y(...U){return Array.prototype.concat.apply([],U)}function b(...U){return U.reduce((q,$)=>q.flatMap(S=>$.map(w=>[S,w])))}function I(U,q){return Math.abs((U+q)%(2*q)-q)}function K(U,q){const $=URL.createObjectURL(q),S=document.createElement("a");S.href=$,S.download=U,S.click(),S.remove(),URL.revokeObjectURL($)}function se(U,q){return Object.assign({},...q.map($=>{if(U[$]!==void 0)return{[$]:U[$]}}))}function ne(U){let q=0;for(const $ of U)++q;return q}function W(U,q){let $=0;for(const S of U)S===q&&++$;return $}},"./src/utils/data-structures.js":(Le,A,r)=>{r.r(A),r.d(A,{CharTrie:()=>D,PriorityQueue:()=>_,TokenLattice:()=>X});class _{constructor(v=(y,b)=>y>b,M=1/0){this._heap=[],this._comparator=v,this._maxSize=M}get size(){return this._heap.length}isEmpty(){return this.size===0}peek(){return this._heap[0]}push(...v){return this.extend(v)}extend(v){for(const M of v)if(this.size0&&this._swap(0,M),this._heap.pop(),this._siftDown(),v}replace(v){const M=this.peek();return this._heap[0]=v,this._siftDown(),M}_parent(v){return(v+1>>>1)-1}_left(v){return(v<<1)+1}_right(v){return v+1<<1}_greater(v,M){return this._comparator(this._heap[v],this._heap[M])}_swap(v,M){const y=this._heap[v];this._heap[v]=this._heap[M],this._heap[M]=y}_siftUp(){this._siftUpFrom(this.size-1)}_siftUpFrom(v){for(;v>0&&this._greater(v,this._parent(v));)this._swap(v,this._parent(v)),v=this._parent(v)}_siftDown(){let v=0;for(;this._left(v)[]),this.endNodes=Array.from({length:this.len+1},()=>[]);const b=new R(this.bosTokenId,0,0,0,0),I=new R(this.eosTokenId,1,this.len,0,0);this.nodes.push(b.clone()),this.nodes.push(I.clone()),this.beginNodes[this.len].push(I),this.endNodes[0].push(b)}insert(v,M,y,b){const I=this.nodes.length,K=new R(b,I,v,M,y);this.beginNodes[v].push(K),this.endNodes[v+M].push(K),this.nodes.push(K)}viterbi(){const v=this.len;let M=0;for(;M<=v;){if(this.beginNodes[M].length==0)return[];for(let se of this.beginNodes[M]){se.prev=null;let ne=0,W=null;for(let U of this.endNodes[M]){const q=U.backtraceScore+se.score;(W===null||q>ne)&&(W=U.clone(),ne=q)}if(W!==null)se.prev=W,se.backtraceScore=ne;else return[]}++M}const y=[],I=this.beginNodes[v][0].prev;if(I===null)return[];let K=I.clone();for(;K.prev!==null;)y.push(K.clone()),K=K.clone().prev.clone();return y.reverse(),y}piece(v){return this.chars.slice(v.pos,v.pos+v.length).join("")}tokens(){return this.viterbi().map(M=>this.piece(M))}tokenIds(){return this.viterbi().map(M=>M.tokenId)}}class R{constructor(v,M,y,b,I){this.tokenId=v,this.nodeId=M,this.pos=y,this.length=b,this.score=I,this.prev=null,this.backtraceScore=0}clone(){const v=new R(this.tokenId,this.nodeId,this.pos,this.length,this.score);return v.prev=this.prev,v.backtraceScore=this.backtraceScore,v}}},"./src/utils/devices.js":(Le,A,r)=>{r.r(A),r.d(A,{DEVICE_TYPES:()=>_});const _=Object.freeze({auto:"auto",gpu:"gpu",cpu:"cpu",wasm:"wasm",webgpu:"webgpu",cuda:"cuda",dml:"dml",webnn:"webnn","webnn-npu":"webnn-npu","webnn-gpu":"webnn-gpu","webnn-cpu":"webnn-cpu"})},"./src/utils/dtypes.js":(Le,A,r)=>{r.r(A),r.d(A,{DATA_TYPES:()=>X,DEFAULT_DEVICE_DTYPE_MAPPING:()=>R,DEFAULT_DTYPE_SUFFIX_MAPPING:()=>g,isWebGpuFp16Supported:()=>j});var _=r("./src/env.js"),D=r("./src/utils/devices.js");const j=function(){let v;return async function(){if(v===void 0)if(!_.apis.IS_WEBGPU_AVAILABLE)v=!1;else try{v=(await navigator.gpu.requestAdapter()).features.has("shader-f16")}catch{v=!1}return v}}(),X=Object.freeze({auto:"auto",fp32:"fp32",fp16:"fp16",q8:"q8",int8:"int8",uint8:"uint8",q4:"q4",bnb4:"bnb4",q4f16:"q4f16"}),R=Object.freeze({[D.DEVICE_TYPES.wasm]:X.q8}),g=Object.freeze({[X.fp32]:"",[X.fp16]:"_fp16",[X.int8]:"_int8",[X.uint8]:"_uint8",[X.q8]:"_quantized",[X.q4]:"_q4",[X.q4f16]:"_q4f16",[X.bnb4]:"_bnb4"})},"./src/utils/generic.js":(Le,A,r)=>{r.r(A),r.d(A,{Callable:()=>_});const _=class{constructor(){let D=function(...j){return D._call(...j)};return Object.setPrototypeOf(D,new.target.prototype)}_call(...D){throw Error("Must implement _call method in subclass")}}},"./src/utils/hub.js":(Le,A,r)=>{r.r(A),r.d(A,{getFile:()=>M,getModelFile:()=>se,getModelJSON:()=>ne});var _=r("?7a2c"),D=r("?a42a"),j=r("./src/env.js"),X=r("./src/utils/core.js");const R={txt:"text/plain",html:"text/html",css:"text/css",js:"text/javascript",json:"application/json",png:"image/png",jpg:"image/jpeg",jpeg:"image/jpeg",gif:"image/gif"};class g{constructor($){if(this.filePath=$,this.headers=new Headers,this.exists=_.existsSync($),this.exists){this.status=200,this.statusText="OK";let S=_.statSync($);this.headers.set("content-length",S.size.toString()),this.updateContentType();let w=this;this.body=new ReadableStream({start(x){w.arrayBuffer().then(O=>{x.enqueue(new Uint8Array(O)),x.close()})}})}else this.status=404,this.statusText="Not Found",this.body=null}updateContentType(){const $=this.filePath.toString().split(".").pop().toLowerCase();this.headers.set("content-type",R[$]??"application/octet-stream")}clone(){let $=new g(this.filePath);return $.exists=this.exists,$.status=this.status,$.statusText=this.statusText,$.headers=new Headers(this.headers),$}async arrayBuffer(){return(await _.promises.readFile(this.filePath)).buffer}async blob(){const $=await _.promises.readFile(this.filePath);return new Blob([$],{type:this.headers.get("content-type")})}async text(){return await _.promises.readFile(this.filePath,"utf8")}async json(){return JSON.parse(await this.text())}}function v(q,$=null,S=null){let w;try{w=new URL(q)}catch{return!1}return!($&&!$.includes(w.protocol)||S&&!S.includes(w.hostname))}async function M(q){var $;if(j.env.useFS&&!v(q,["http:","https:","blob:"]))return new g(q);if(typeof process<"u"&&(($=process==null?void 0:process.release)==null?void 0:$.name)==="node"){const S=!!(dr!=null&&dr.TESTING_REMOTELY),w=j.env.version,x=new Headers;if(x.set("User-Agent",`transformers.js/${w}; is_ci/${S};`),v(q,["http:","https:"],["huggingface.co","hf.co"])){const ae=(dr==null?void 0:dr.HF_TOKEN)??(dr==null?void 0:dr.HF_ACCESS_TOKEN);ae&&x.set("Authorization",`Bearer ${ae}`)}return fetch(q,{headers:x})}else return fetch(q)}const y={400:"Bad request error occurred while trying to load file",401:"Unauthorized access to file",403:"Forbidden access to file",404:"Could not locate file",408:"Request timeout error occurred while trying to load file",500:"Internal server error error occurred while trying to load file",502:"Bad gateway error occurred while trying to load file",503:"Service unavailable error occurred while trying to load file",504:"Gateway timeout error occurred while trying to load file"};function b(q,$,S){if(!S)return null;const w=y[q]??`Error (${q}) occurred while trying to load file`;throw Error(`${w}: "${$}".`)}class I{constructor($){this.path=$}async match($){let S=D.join(this.path,$),w=new g(S);if(w.exists)return w}async put($,S){const w=Buffer.from(await S.arrayBuffer());let x=D.join(this.path,$);try{await _.promises.mkdir(D.dirname(x),{recursive:!0}),await _.promises.writeFile(x,w)}catch(O){console.warn("An error occurred while writing the file to cache:",O)}}}async function K(q,...$){for(let S of $)try{let w=await q.match(S);if(w)return w}catch{continue}}async function se(q,$,S=!0,w={}){if(!j.env.allowLocalModels){if(w.local_files_only)throw Error("Invalid configuration detected: local models are disabled (`env.allowLocalModels=false`) but you have requested to only use local models (`local_files_only=true`).");if(!j.env.allowRemoteModels)throw Error("Invalid configuration detected: both local and remote models are disabled. Fix by setting `env.allowLocalModels` or `env.allowRemoteModels` to `true`.")}(0,X.dispatchCallback)(w.progress_callback,{status:"initiate",name:q,file:$});let x;if(!x&&j.env.useBrowserCache){if(typeof caches>"u")throw Error("Browser cache is not available in this environment.");try{x=await caches.open("transformers-cache")}catch(tt){console.warn("An error occurred while opening the browser cache:",tt)}}if(!x&&j.env.useFSCache&&(x=new I(w.cache_dir??j.env.cacheDir)),!x&&j.env.useCustomCache){if(!j.env.customCache)throw Error("`env.useCustomCache=true`, but `env.customCache` is not defined.");if(!j.env.customCache.match||!j.env.customCache.put)throw new Error("`env.customCache` must be an object which implements the `match` and `put` functions of the Web Cache API. For more information, see https://developer.mozilla.org/en-US/docs/Web/API/Cache");x=j.env.customCache}const O=w.revision??"main";let ae=U(q,$),ie=U(j.env.localModelPath,ae),ve=U(j.env.remoteHost,j.env.remotePathTemplate.replaceAll("{model}",q).replaceAll("{revision}",encodeURIComponent(O)),$),we=O==="main"?ae:U(q,O,$),re,xe=x instanceof I?we:ve,ce=!1,ke;x&&(ke=await K(x,ie,xe));const Ie=ke!==void 0;if(ke===void 0){if(j.env.allowLocalModels)if(v(ae,["http:","https:"])){if(w.local_files_only)throw new Error(`\`local_files_only=true\`, but attempted to load a remote file from: ${ae}.`);if(!j.env.allowRemoteModels)throw new Error(`\`env.allowRemoteModels=false\`, but attempted to load a remote file from: ${ae}.`)}else try{ke=await M(ie),re=ie}catch(Ge){console.warn(`Unable to load from local path "${ie}": "${Ge}"`)}if(ke===void 0||ke.status===404){if(w.local_files_only||!j.env.allowRemoteModels){if(S)throw Error(`\`local_files_only=true\` or \`env.allowRemoteModels=false\` and file was not found locally at "${ie}".`);return null}if(ke=await M(ve),ke.status!==200)return b(ke.status,ve,S);re=xe}ce=x&&typeof Response<"u"&&ke instanceof Response&&ke.status===200}(0,X.dispatchCallback)(w.progress_callback,{status:"download",name:q,file:$});let Ee;return w.progress_callback?Ie&&typeof navigator<"u"&&/firefox/i.test(navigator.userAgent)?(Ee=new Uint8Array(await ke.arrayBuffer()),(0,X.dispatchCallback)(w.progress_callback,{status:"progress",name:q,file:$,progress:100,loaded:Ee.length,total:Ee.length})):Ee=await W(ke,tt=>{(0,X.dispatchCallback)(w.progress_callback,{status:"progress",name:q,file:$,...tt})}):Ee=new Uint8Array(await ke.arrayBuffer()),ce&&re&&await x.match(re)===void 0&&await x.put(re,new Response(Ee,{headers:ke.headers})).catch(tt=>{console.warn(`Unable to add response to browser cache: ${tt}.`)}),(0,X.dispatchCallback)(w.progress_callback,{status:"done",name:q,file:$}),Ee}async function ne(q,$,S=!0,w={}){let x=await se(q,$,S,w);if(x===null)return{};let ae=new TextDecoder("utf-8").decode(x);return JSON.parse(ae)}async function W(q,$){const S=q.headers.get("Content-Length");S===null&&console.warn("Unable to determine content-length from response headers. Will expand buffer when needed.");let w=parseInt(S??"0"),x=new Uint8Array(w),O=0;const ae=q.body.getReader();async function ie(){const{done:ve,value:we}=await ae.read();if(ve)return;let re=O+we.length;if(re>w){w=re;let ce=new Uint8Array(w);ce.set(x),x=ce}x.set(we,O),O=re;const xe=O/w*100;return $({progress:xe,loaded:O,total:w}),ie()}return await ie(),x}function U(...q){return q=q.map(($,S)=>(S&&($=$.replace(new RegExp("^/"),"")),S!==q.length-1&&($=$.replace(new RegExp("/$"),"")),$)),q.join("/")}},"./src/utils/image.js":(Le,A,r)=>{r.r(A),r.d(A,{RawImage:()=>K,load_image:()=>se});var _=r("./src/utils/core.js"),D=r("./src/utils/hub.js"),j=r("./src/env.js"),X=r("./src/utils/tensor.js"),R=r("?2b25");let g,v,M;const y=j.apis.IS_BROWSER_ENV||j.apis.IS_WEBWORKER_ENV;if(y)g=(ne,W)=>{if(!self.OffscreenCanvas)throw new Error("OffscreenCanvas not supported by this browser.");return new self.OffscreenCanvas(ne,W)},M=self.createImageBitmap,v=self.ImageData;else if(R)M=async ne=>{const U=(await ne.metadata()).channels,{data:q,info:$}=await ne.rotate().raw().toBuffer({resolveWithObject:!0}),S=new K(new Uint8ClampedArray(q),$.width,$.height,$.channels);return U!==void 0&&U!==$.channels&&S.convert(U),S};else throw new Error("Unable to load image processing library.");const b={0:"nearest",1:"lanczos",2:"bilinear",3:"bicubic",4:"box",5:"hamming"},I=new Map([["png","image/png"],["jpg","image/jpeg"],["jpeg","image/jpeg"],["gif","image/gif"]]);class K{constructor(W,U,q,$){this.data=W,this.width=U,this.height=q,this.channels=$}get size(){return[this.width,this.height]}static async read(W){if(W instanceof K)return W;if(typeof W=="string"||W instanceof URL)return await this.fromURL(W);throw new Error(`Unsupported input type: ${typeof W}`)}static fromCanvas(W){if(!y)throw new Error("fromCanvas() is only supported in browser environments.");const q=W.getContext("2d").getImageData(0,0,W.width,W.height).data;return new K(q,W.width,W.height,4)}static async fromURL(W){const U=await(0,D.getFile)(W);if(U.status!==200)throw new Error(`Unable to read image from "${W}" (${U.status} ${U.statusText})`);const q=await U.blob();return this.fromBlob(q)}static async fromBlob(W){if(y){const U=await M(W),q=g(U.width,U.height).getContext("2d");return q.drawImage(U,0,0),new this(q.getImageData(0,0,U.width,U.height).data,U.width,U.height,4)}else{const U=R(await W.arrayBuffer());return await M(U)}}static fromTensor(W,U="CHW"){if(W.dims.length!==3)throw new Error(`Tensor should have 3 dimensions, but has ${W.dims.length} dimensions.`);if(U==="CHW")W=W.transpose(1,2,0);else if(U!=="HWC")throw new Error(`Unsupported channel format: ${U}`);if(!(W.data instanceof Uint8ClampedArray||W.data instanceof Uint8Array))throw new Error(`Unsupported tensor type: ${W.type}`);switch(W.dims[2]){case 1:case 2:case 3:case 4:return new K(W.data,W.dims[1],W.dims[0],W.dims[2]);default:throw new Error(`Unsupported number of channels: ${W.dims[2]}`)}}grayscale(){if(this.channels===1)return this;const W=new Uint8ClampedArray(this.width*this.height*1);switch(this.channels){case 3:case 4:for(let U=0,q=0;U=0?O=q:ie=-q,$>=0?ae=$:ve=-$,x.drawImage(w,O,ae,W,U,ie,ve,W,U),new K(x.getImageData(0,0,W,U).data,W,U,4).convert(S)}else{let S=this.toSharp();if(q>=0&&$>=0)S=S.extract({left:Math.floor(q),top:Math.floor($),width:W,height:U});else if(q<=0&&$<=0){const w=Math.floor(-$),x=Math.floor(-q);S=S.extend({top:w,left:x,right:W-this.width-x,bottom:U-this.height-w})}else{let w=[0,0],x=0;$<0?(w[0]=Math.floor(-$),w[1]=U-this.height-w[0]):x=Math.floor($);let O=[0,0],ae=0;q<0?(O[0]=Math.floor(-q),O[1]=W-this.width-O[0]):ae=Math.floor(q),S=S.extend({top:w[0],bottom:w[1],left:O[0],right:O[1]}).extract({left:ae,top:x,width:W,height:U})}return await M(S)}}async toBlob(W="image/png",U=1){if(!y)throw new Error("toBlob() is only supported in browser environments.");return await this.toCanvas().convertToBlob({type:W,quality:U})}toTensor(W="CHW"){let U=new X.Tensor("uint8",new Uint8Array(this.data),[this.height,this.width,this.channels]);if(W!=="HWC")if(W==="CHW")U=U.permute(2,0,1);else throw new Error(`Unsupported channel format: ${W}`);return U}toCanvas(){if(!y)throw new Error("toCanvas() is only supported in browser environments.");const W=this.clone().rgba(),U=g(W.width,W.height),q=new v(W.data,W.width,W.height);return U.getContext("2d").putImageData(q,0,0),U}split(){const{data:W,width:U,height:q,channels:$}=this,S=W.constructor,w=W.length/$,x=Array.from({length:$},()=>new S(w));for(let O=0;Onew K(O,U,q,1))}_update(W,U,q,$=null){return this.data=W,this.width=U,this.height=q,$!==null&&(this.channels=$),this}clone(){return new K(this.data.slice(),this.width,this.height,this.channels)}convert(W){if(this.channels===W)return this;switch(W){case 1:this.grayscale();break;case 3:this.rgb();break;case 4:this.rgba();break;default:throw new Error(`Conversion failed due to unsupported number of channels: ${this.channels}`)}return this}async save(W){if(y){if(j.apis.IS_WEBWORKER_ENV)throw new Error("Unable to save an image from a Web Worker.");const U=W.split(".").pop().toLowerCase(),q=I.get(U)??"image/png",$=await this.toBlob(q);(0,_.saveBlob)(W,$)}else{if(j.apis.IS_FS_AVAILABLE)return await this.toSharp().toFile(W);throw new Error("Unable to save the image because filesystem is disabled in this environment.")}}toSharp(){if(y)throw new Error("toSharp() is only supported in server-side environments.");return R(this.data,{raw:{width:this.width,height:this.height,channels:this.channels}})}}const se=K.read.bind(K)},"./src/utils/maths.js":(Le,A,r)=>{r.r(A),r.d(A,{FFT:()=>se,bankers_round:()=>U,cos_sim:()=>g,dot:()=>R,dynamic_time_warping:()=>q,interpolate_data:()=>_,log_softmax:()=>X,magnitude:()=>v,max:()=>y,medianFilter:()=>ne,min:()=>M,permute_data:()=>D,round:()=>W,softmax:()=>j});function _($,[S,w,x],[O,ae],ie="bilinear",ve=!1){const we=ae/x,re=O/w,xe=new $.constructor(O*ae*S),ce=w*x,ke=O*ae;for(let Ie=0;Ie=0;--ve)O[ve]=we,x[ve]=S[w[ve]],we*=x[ve];const ae=w.map((ve,we)=>O[w.indexOf(we)]),ie=new $.constructor($.length);for(let ve=0;ve<$.length;++ve){let we=0;for(let re=S.length-1,xe=ve;re>=0;--re)we+=xe%S[re]*ae[re],xe=Math.floor(xe/S[re]);ie[we]=$[ve]}return[ie,x]}function j($){const S=y($)[0],w=$.map(ae=>Math.exp(ae-S)),x=w.reduce((ae,ie)=>ae+ie,0);return w.map(ae=>ae/x)}function X($){const S=y($)[0];let w=0;for(let ae=0;ae<$.length;++ae)w+=Math.exp($[ae]-S);const x=Math.log(w);return $.map(ae=>ae-S-x)}function R($,S){let w=0;for(let x=0;x<$.length;++x)w+=$[x]*S[x];return w}function g($,S){const w=R($,S),x=v($),O=v(S);return w/(x*O)}function v($){return Math.sqrt($.reduce((S,w)=>S+w*w,0))}function M($){if($.length===0)throw Error("Array must not be empty");let S=$[0],w=0;for(let x=1;x<$.length;++x)$[x]S&&(S=$[x],w=x);return[S,w]}function b($){return $>0&&($&$-1)===0}class I{constructor(S){if(this.size=S|0,this.size<=1||!b(this.size))throw new Error("FFT size must be a power of two larger than 1");this._csize=S<<1,this.table=new Float64Array(this.size*2);for(let x=0;xx;x<<=1)++w;this._width=w%2===0?w-1:w,this._bitrev=new Int32Array(1<>>O&3)<>>1);for(let O=0;O>>1]=S[O];return x}toComplexArray(S,w){const x=w||this.createComplexArray();for(let O=0;O>>1],x[O+1]=0;return x}transform(S,w){if(S===w)throw new Error("Input and output buffers must be different");this._transform4(S,w,1)}realTransform(S,w){if(S===w)throw new Error("Input and output buffers must be different");this._realTransform4(S,w,1)}inverseTransform(S,w){if(S===w)throw new Error("Input and output buffers must be different");this._transform4(S,w,-1);for(let x=0;x>=2;ie>=2;ie>>=2){ve=O/ie<<1;const ke=ve>>>2;for(we=0;we>>1,ie>>>1)}else for(we=0,re=0;we>>1,ie>>>1,x)}const ce=this.table;for(ie>>=2;ie>=2;ie>>=2){ve=O/ie<<1;const Ie=ve>>>1,Ee=Ie>>>1,tt=Ee>>>1;for(we=0;we>>1;for(let Ie=2;Ie>1;++xe){const ce=(xe+1-S)**2/2,ke=Math.sqrt(we**2+re**2)**ce,Ie=ce*Math.atan2(re,we),Ee=2*xe;ae[Ee]=ke*Math.cos(Ie),ae[Ee+1]=ke*Math.sin(Ie),ie[Ee]=ae[Ee],ie[Ee+1]=-ae[Ee+1]}this._slicedChirpBuffer=ae.subarray(w,x),this._f=new I(O>>1),this._f.transform(this._chirpBuffer,ie)}_transform(S,w,x){const O=this._buffer1,ae=this._buffer2,ie=this._outBuffer1,ve=this._outBuffer2,we=this._chirpBuffer,re=this._slicedChirpBuffer,xe=this._a;if(x)for(let ce=0;ce>1,Ee=w[Ie];O[ce]=Ee*re[ce],O[ke]=Ee*re[ke]}else for(let ce=0;ce=$.length&&(we=2*($.length-1)-we),x[ie++]=$[we]}x.sort(),w[ae]=x[O]}return w}function W($,S){const w=Math.pow(10,S);return Math.round($*w)/w}function U($){const S=Math.round($);return Math.abs($)%1===.5?S%2===0?S:S-1:S}function q($){const S=$.length,w=$[0].length,x=[S+1,w+1],O=Array.from({length:x[0]},()=>Array(x[1]).fill(1/0));O[0][0]=0;const ae=Array.from({length:x[0]},()=>Array(x[1]).fill(-1));for(let xe=1;xe0||ve>0;)switch(we.push(ie-1),re.push(ve-1),ae[ie][ve]){case 0:--ie,--ve;break;case 1:--ie;break;case 2:--ve;break;default:throw new Error(`Internal error in dynamic time warping. Unexpected trace[${ie}, ${ve}]. Please file a bug report.`)}return we.reverse(),re.reverse(),[we,re]}},"./src/utils/tensor.js":(Le,A,r)=>{r.r(A),r.d(A,{Tensor:()=>R,cat:()=>w,full:()=>re,full_like:()=>xe,interpolate:()=>M,interpolate_4d:()=>y,layer_norm:()=>U,matmul:()=>b,mean:()=>ie,mean_pooling:()=>W,ones:()=>ce,ones_like:()=>ke,permute:()=>v,quantize_embeddings:()=>Ge,rand:()=>tt,rfft:()=>I,slice:()=>ne,stack:()=>x,std_mean:()=>ae,topk:()=>K,zeros:()=>Ie,zeros_like:()=>Ee});var _=r("./src/utils/maths.js"),D=r("./src/backends/onnx.js"),j=r("./src/ops/registry.js");const X=Object.freeze({float32:Float32Array,float16:Uint16Array,float64:Float64Array,string:Array,int8:Int8Array,uint8:Uint8Array,int16:Int16Array,uint16:Uint16Array,int32:Int32Array,uint32:Uint32Array,int64:BigInt64Array,uint64:BigUint64Array,bool:Uint8Array,uint4:Uint8Array,int4:Int8Array});class R{constructor(...J){me(this,"ort_tensor");return(0,D.isONNXTensor)(J[0])?this.ort_tensor=J[0]:this.ort_tensor=new D.Tensor(J[0],J[1],J[2]),new Proxy(this,{get:(de,Ce)=>{if(typeof Ce=="string"){let Be=Number(Ce);if(Number.isInteger(Be))return de._getitem(Be)}return de[Ce]},set:(de,Ce,Be)=>de[Ce]=Be})}get dims(){return this.ort_tensor.dims}set dims(J){this.ort_tensor.dims=J}get type(){return this.ort_tensor.type}get data(){return this.ort_tensor.data}get size(){return this.ort_tensor.size}get location(){return this.ort_tensor.location}dispose(){this.ort_tensor.dispose()}*[Symbol.iterator](){const[J,...de]=this.dims;if(de.length>0){const Ce=de.reduce((Be,Ze)=>Be*Ze);for(let Be=0;Be0){const Be=Ce.reduce((Ze,te)=>Ze*te);return this._subarray(J,Be,Ce)}else return new R(this.type,[this.data[J]],Ce)}indexOf(J){const de=this.data;for(let Ce=0;CeVe)throw new Error(`Invalid slice: ${Te}`);const Ne=[Math.max(Ue,0),Math.min(Ve,this.dims[le])];Ce.push(Ne),de.push(Ne[1]-Ne[0])}else throw new Error(`Invalid slice: ${Te}`)}const Be=Ce.map(([le,Te])=>Te-le),Ze=Be.reduce((le,Te)=>le*Te),te=this.data,Ke=new te.constructor(Ze),je=this.stride();for(let le=0;le=0;--Ue){const Ne=Be[Ue];Te+=(Ve%Ne+Ce[Ue][0])*je[Ue],Ve=Math.floor(Ve/Ne)}Ke[le]=te[Te]}return new R(this.type,Ke,de)}permute(...J){return v(this,J)}transpose(...J){return this.permute(...J)}sum(J=null,de=!1){return this.norm(1,J,de)}norm(J="fro",de=null,Ce=!1){if(J==="fro")J=2;else if(typeof J=="string")throw Error(`Unsupported norm: ${J}`);const Be=this.data,Ze=(le,Te)=>le+Te**J;if(de===null){const le=Be.reduce(Ze,0)**(1/J);return new R(this.type,[le],[])}const[te,Ke,je]=O(Ze,this,de,Ce);if(J!==1)for(let le=0;le=0;--je){const Ue=this.dims[je];if(je!==de){const Ve=le%Ue;Ke+=Ve*Te,Te*=this.dims[je]}le=Math.floor(le/Ue)}Be[te]/=Ze[Ke]}return this}normalize(J=2,de=1){return this.clone().normalize_(J,de)}stride(){return ve(this.dims)}squeeze(J=null){return new R(this.type,this.data,q(this.dims,J))}squeeze_(J=null){return this.dims=q(this.dims,J),this}unsqueeze(J=null){return new R(this.type,this.data,$(this.dims,J))}unsqueeze_(J=null){return this.dims=$(this.dims,J),this}flatten_(J=0,de=-1){de=(de+this.dims.length)%this.dims.length;let Ce=this.dims.slice(0,J),Be=this.dims.slice(J,de+1),Ze=this.dims.slice(de+1);return this.dims=[...Ce,Be.reduce((te,Ke)=>te*Ke,1),...Ze],this}flatten(J=0,de=-1){return this.clone().flatten_(J,de)}view(...J){let de=-1;for(let Be=0;BeKe!==de?Ze*te:Ze,1);J[de]=Ce.length/Be}return new R(this.type,Ce,J)}neg_(){const J=this.data;for(let de=0;deJ?1:0;return new R("bool",de,this.dims)}lt(J){const de=new Uint8Array(this.data.length),Ce=this.data;for(let Be=0;BeMath.min(te,Ke),this,J,de,1/0);return new R(Ce,Be,Ze)}max(J=null,de=!1){if(J===null){const te=(0,_.max)(this.data)[0];return new R(this.type,[te],[])}const[Ce,Be,Ze]=O((te,Ke)=>Math.max(te,Ke),this,J,de,-1/0);return new R(Ce,Be,Ze)}argmin(J=null,de=!1){if(J!==null)throw new Error("`dim !== null` not yet implemented.");const Ce=(0,_.min)(this.data)[1];return new R("int64",[BigInt(Ce)],[])}argmax(J=null,de=!1){if(J!==null)throw new Error("`dim !== null` not yet implemented.");const Ce=(0,_.max)(this.data)[1];return new R("int64",[BigInt(Ce)],[])}to(J){if(this.type===J)return this;if(!X.hasOwnProperty(J))throw new Error(`Unsupported type: ${J}`);let de;const Ce=["int64","uint64"].includes(this.type),Be=["int64","uint64"].includes(J);return Ce&&!Be?de=Number:!Ce&&Be&&(de=BigInt),new R(J,X[J].from(this.data,de),this.dims)}}function g(ye,J){const de=ye.length,Ce=J.reduce((Ze,te)=>Ze*te);if(de!==Ce)throw Error(`cannot reshape array of size ${de} into shape (${J})`);let Be=ye;for(let Ze=J.length-1;Ze>=0;Ze--)Be=Be.reduce((te,Ke)=>{let je=te[te.length-1];return je.lengthnew R("int64",ye,[ye.length]);async function ne(ye,J,de,Ce,Be){return await(await j.TensorOpRegistry.slice)({x:ye,s:se(J),e:se(de),a:se(Ce),t:se(Be??new Array(Ce.length).fill(1))})}function W(ye,J){const de=ye.data,Ce=J.data,Be=[ye.dims[0],ye.dims[2]],Ze=new de.constructor(Be[0]*Be[1]),[te,Ke,je]=ye.dims;let le=0;for(let Te=0;Tede!==1):typeof J=="number"?ye[J]===1&&ye.splice(J,1):Array.isArray(J)&&(ye=ye.filter((de,Ce)=>de!==1||!J.includes(Ce))),ye}function $(ye,J){return J=S(J,ye.length+1),ye=ye.slice(),ye.splice(J,0,1),ye}function S(ye,J,de=null,Ce=!0){if(Ce&&(ye<-J||ye>=J))throw new Error(`IndexError: index ${ye} is out of bounds for dimension${de===null?"":" "+de} with size ${J}`);return ye<0&&(ye=(ye%J+J)%J),ye}function w(ye,J=0){J=S(J,ye[0].dims.length);const de=ye[0].dims.slice();de[J]=ye.reduce((te,Ke)=>te+Ke.dims[J],0);const Ce=de.reduce((te,Ke)=>te*Ke,1),Be=new ye[0].data.constructor(Ce),Ze=ye[0].type;if(J===0){let te=0;for(const Ke of ye){const je=Ke.data;Be.set(je,te),te+=je.length}}else{let te=0;for(let Ke=0;Ke=0;--Ve){const st=le[Ve];let dt=Ne%st;Ve===J&&(dt+=te),Ue+=dt*Re,Re*=de[Ve],Ne=Math.floor(Ne/st)}Be[Ue]=je[Te]}te+=le[J]}}return new R(Ze,Be,de)}function x(ye,J=0){return w(ye.map(de=>de.unsqueeze(J)),J)}function O(ye,J,de=null,Ce=!1,Be=null){const Ze=J.data,te=J.dims;de=S(de,te.length);const Ke=te.slice();Ke[de]=1;const je=new Ze.constructor(Ze.length/te[de]);Be!==null&&je.fill(Be);for(let le=0;le=0;--Ue){const Re=te[Ue];if(Ue!==de){const st=Ve%Re;Te+=st*Ne,Ne*=Ke[Ue]}Ve=Math.floor(Ve/Re)}je[Te]=ye(je[Te],Ze[le],le,Te)}return Ce||Ke.splice(de,1),[J.type,je,Ke]}function ae(ye,J=null,de=1,Ce=!1){const Be=ye.data,Ze=ye.dims;if(J===null){const Ne=Be.reduce((ct,lt)=>ct+lt,0)/Be.length,Re=Math.sqrt(Be.reduce((ct,lt)=>ct+(lt-Ne)**2,0)/(Be.length-de)),st=new R(ye.type,[Ne],[]);return[new R(ye.type,[Re],[]),st]}J=S(J,Ze.length);const te=ie(ye,J,Ce),Ke=te.data,[je,le,Te]=O((Ve,Ne,Re,st)=>Ve+(Ne-Ke[st])**2,ye,J,Ce);for(let Ve=0;Vele+Te,0);return new R(ye.type,[je/Be.length],[])}J=S(J,Ce.length);const[Ze,te,Ke]=O((je,le)=>je+le,ye,J,de);if(Ce[J]!==1)for(let je=0;je=0;--de)J[de]=Ce,Ce*=ye[de];return J}function we(ye,J,de,Ce){const Be=ye.reduce((Ze,te)=>Ze*te,1);return new R(de,new Ce(Be).fill(J),ye)}function re(ye,J){let de,Ce;if(typeof J=="number")de="float32",Ce=Float32Array;else if(typeof J=="bigint")de="int64",Ce=BigInt64Array;else if(typeof J=="boolean")de="bool",Ce=Uint8Array;else throw new Error(`Unsupported data type: ${typeof J}`);return we(ye,J,de,Ce)}function xe(ye,J){return re(ye.dims,J)}function ce(ye){return we(ye,1n,"int64",BigInt64Array)}function ke(ye){return ce(ye.dims)}function Ie(ye){return we(ye,0n,"int64",BigInt64Array)}function Ee(ye){return Ie(ye.dims)}function tt(ye){const J=ye.reduce((de,Ce)=>de*Ce,1);return new R("float32",Float32Array.from({length:J},()=>Math.random()),ye)}function Ge(ye,J){if(ye.dims.length!==2)throw new Error("The tensor must have 2 dimensions");if(ye.dims.at(-1)%8!==0)throw new Error("The last dimension of the tensor must be a multiple of 8");if(!["binary","ubinary"].includes(J))throw new Error("The precision must be either 'binary' or 'ubinary'");const de=J==="binary",Ce=de?"int8":"uint8",Be=de?Int8Array:Uint8Array,Ze=ye.data,te=new Be(Ze.length/8);for(let Ke=0;Ke0?1:0,le=Math.floor(Ke/8),Te=Ke%8;te[le]|=je<<7-Te,de&&Te===0&&(te[le]-=128)}return new R(Ce,te,[ye.dims[0],ye.dims[1]/8])}}},ni={};function gs(Le){var A=ni[Le];if(A!==void 0)return A.exports;var r=ni[Le]={exports:{}};return Xr[Le](r,r.exports,gs),r.exports}gs.m=Xr,(()=>{var Le=Object.getPrototypeOf?r=>Object.getPrototypeOf(r):r=>r.__proto__,A;gs.t=function(r,_){if(_&1&&(r=this(r)),_&8||typeof r=="object"&&r&&(_&4&&r.__esModule||_&16&&typeof r.then=="function"))return r;var D=Object.create(null);gs.r(D);var j={};A=A||[null,Le({}),Le([]),Le(Le)];for(var X=_&2&&r;typeof X=="object"&&!~A.indexOf(X);X=Le(X))Object.getOwnPropertyNames(X).forEach(R=>j[R]=()=>r[R]);return j.default=()=>r,gs.d(D,j),D}})(),gs.d=(Le,A)=>{for(var r in A)gs.o(A,r)&&!gs.o(Le,r)&&Object.defineProperty(Le,r,{enumerable:!0,get:A[r]})},gs.o=(Le,A)=>Object.prototype.hasOwnProperty.call(Le,A),gs.r=Le=>{typeof Symbol<"u"&&Symbol.toStringTag&&Object.defineProperty(Le,Symbol.toStringTag,{value:"Module"}),Object.defineProperty(Le,"__esModule",{value:!0})},(()=>{var Le;if(typeof self.location.href=="string"&&(Le=self.location.href),!Le)throw new Error("Automatic publicPath is not supported in this browser");Le=Le.replace(/#.*$/,"").replace(/\?.*$/,"").replace(/\/[^\/]+$/,"/"),gs.p=Le})(),gs.b=new URL(self.location.href);var c={};(()=>{/*!*****************************!*\ + !*** ./src/transformers.js ***! + \*****************************/gs.r(c),gs.d(c,{ASTFeatureExtractor:()=>M.ASTFeatureExtractor,ASTForAudioClassification:()=>r.ASTForAudioClassification,ASTModel:()=>r.ASTModel,ASTPreTrainedModel:()=>r.ASTPreTrainedModel,AlbertForMaskedLM:()=>r.AlbertForMaskedLM,AlbertForQuestionAnswering:()=>r.AlbertForQuestionAnswering,AlbertForSequenceClassification:()=>r.AlbertForSequenceClassification,AlbertModel:()=>r.AlbertModel,AlbertPreTrainedModel:()=>r.AlbertPreTrainedModel,AlbertTokenizer:()=>_.AlbertTokenizer,AudioClassificationPipeline:()=>A.AudioClassificationPipeline,AutoConfig:()=>D.AutoConfig,AutoFeatureExtractor:()=>y.AutoFeatureExtractor,AutoImageProcessor:()=>K.AutoImageProcessor,AutoModel:()=>r.AutoModel,AutoModelForAudioClassification:()=>r.AutoModelForAudioClassification,AutoModelForAudioFrameClassification:()=>r.AutoModelForAudioFrameClassification,AutoModelForCTC:()=>r.AutoModelForCTC,AutoModelForCausalLM:()=>r.AutoModelForCausalLM,AutoModelForDepthEstimation:()=>r.AutoModelForDepthEstimation,AutoModelForDocumentQuestionAnswering:()=>r.AutoModelForDocumentQuestionAnswering,AutoModelForImageClassification:()=>r.AutoModelForImageClassification,AutoModelForImageFeatureExtraction:()=>r.AutoModelForImageFeatureExtraction,AutoModelForImageMatting:()=>r.AutoModelForImageMatting,AutoModelForImageSegmentation:()=>r.AutoModelForImageSegmentation,AutoModelForImageToImage:()=>r.AutoModelForImageToImage,AutoModelForMaskGeneration:()=>r.AutoModelForMaskGeneration,AutoModelForMaskedLM:()=>r.AutoModelForMaskedLM,AutoModelForNormalEstimation:()=>r.AutoModelForNormalEstimation,AutoModelForObjectDetection:()=>r.AutoModelForObjectDetection,AutoModelForPoseEstimation:()=>r.AutoModelForPoseEstimation,AutoModelForQuestionAnswering:()=>r.AutoModelForQuestionAnswering,AutoModelForSemanticSegmentation:()=>r.AutoModelForSemanticSegmentation,AutoModelForSeq2SeqLM:()=>r.AutoModelForSeq2SeqLM,AutoModelForSequenceClassification:()=>r.AutoModelForSequenceClassification,AutoModelForSpeechSeq2Seq:()=>r.AutoModelForSpeechSeq2Seq,AutoModelForTextToSpectrogram:()=>r.AutoModelForTextToSpectrogram,AutoModelForTextToWaveform:()=>r.AutoModelForTextToWaveform,AutoModelForTokenClassification:()=>r.AutoModelForTokenClassification,AutoModelForUniversalSegmentation:()=>r.AutoModelForUniversalSegmentation,AutoModelForVision2Seq:()=>r.AutoModelForVision2Seq,AutoModelForXVector:()=>r.AutoModelForXVector,AutoModelForZeroShotObjectDetection:()=>r.AutoModelForZeroShotObjectDetection,AutoProcessor:()=>W.AutoProcessor,AutoTokenizer:()=>_.AutoTokenizer,AutomaticSpeechRecognitionPipeline:()=>A.AutomaticSpeechRecognitionPipeline,BartForConditionalGeneration:()=>r.BartForConditionalGeneration,BartForSequenceClassification:()=>r.BartForSequenceClassification,BartModel:()=>r.BartModel,BartPretrainedModel:()=>r.BartPretrainedModel,BartTokenizer:()=>_.BartTokenizer,BaseModelOutput:()=>r.BaseModelOutput,BaseStreamer:()=>U.BaseStreamer,BeitFeatureExtractor:()=>I.BeitFeatureExtractor,BeitForImageClassification:()=>r.BeitForImageClassification,BeitModel:()=>r.BeitModel,BeitPreTrainedModel:()=>r.BeitPreTrainedModel,BertForMaskedLM:()=>r.BertForMaskedLM,BertForQuestionAnswering:()=>r.BertForQuestionAnswering,BertForSequenceClassification:()=>r.BertForSequenceClassification,BertForTokenClassification:()=>r.BertForTokenClassification,BertModel:()=>r.BertModel,BertPreTrainedModel:()=>r.BertPreTrainedModel,BertTokenizer:()=>_.BertTokenizer,BitImageProcessor:()=>I.BitImageProcessor,BlenderbotForConditionalGeneration:()=>r.BlenderbotForConditionalGeneration,BlenderbotModel:()=>r.BlenderbotModel,BlenderbotPreTrainedModel:()=>r.BlenderbotPreTrainedModel,BlenderbotSmallForConditionalGeneration:()=>r.BlenderbotSmallForConditionalGeneration,BlenderbotSmallModel:()=>r.BlenderbotSmallModel,BlenderbotSmallPreTrainedModel:()=>r.BlenderbotSmallPreTrainedModel,BlenderbotSmallTokenizer:()=>_.BlenderbotSmallTokenizer,BlenderbotTokenizer:()=>_.BlenderbotTokenizer,BloomForCausalLM:()=>r.BloomForCausalLM,BloomModel:()=>r.BloomModel,BloomPreTrainedModel:()=>r.BloomPreTrainedModel,BloomTokenizer:()=>_.BloomTokenizer,CLIPFeatureExtractor:()=>I.CLIPFeatureExtractor,CLIPImageProcessor:()=>I.CLIPImageProcessor,CLIPModel:()=>r.CLIPModel,CLIPPreTrainedModel:()=>r.CLIPPreTrainedModel,CLIPSegForImageSegmentation:()=>r.CLIPSegForImageSegmentation,CLIPSegModel:()=>r.CLIPSegModel,CLIPSegPreTrainedModel:()=>r.CLIPSegPreTrainedModel,CLIPTextModel:()=>r.CLIPTextModel,CLIPTextModelWithProjection:()=>r.CLIPTextModelWithProjection,CLIPTokenizer:()=>_.CLIPTokenizer,CLIPVisionModel:()=>r.CLIPVisionModel,CLIPVisionModelWithProjection:()=>r.CLIPVisionModelWithProjection,CamembertForMaskedLM:()=>r.CamembertForMaskedLM,CamembertForQuestionAnswering:()=>r.CamembertForQuestionAnswering,CamembertForSequenceClassification:()=>r.CamembertForSequenceClassification,CamembertForTokenClassification:()=>r.CamembertForTokenClassification,CamembertModel:()=>r.CamembertModel,CamembertPreTrainedModel:()=>r.CamembertPreTrainedModel,CamembertTokenizer:()=>_.CamembertTokenizer,CausalLMOutput:()=>r.CausalLMOutput,CausalLMOutputWithPast:()=>r.CausalLMOutputWithPast,ChineseCLIPFeatureExtractor:()=>I.ChineseCLIPFeatureExtractor,ChineseCLIPModel:()=>r.ChineseCLIPModel,ChineseCLIPPreTrainedModel:()=>r.ChineseCLIPPreTrainedModel,ClapAudioModelWithProjection:()=>r.ClapAudioModelWithProjection,ClapFeatureExtractor:()=>M.ClapFeatureExtractor,ClapModel:()=>r.ClapModel,ClapPreTrainedModel:()=>r.ClapPreTrainedModel,ClapTextModelWithProjection:()=>r.ClapTextModelWithProjection,ClassifierFreeGuidanceLogitsProcessor:()=>$.ClassifierFreeGuidanceLogitsProcessor,CodeGenForCausalLM:()=>r.CodeGenForCausalLM,CodeGenModel:()=>r.CodeGenModel,CodeGenPreTrainedModel:()=>r.CodeGenPreTrainedModel,CodeGenTokenizer:()=>_.CodeGenTokenizer,CodeLlamaTokenizer:()=>_.CodeLlamaTokenizer,CohereForCausalLM:()=>r.CohereForCausalLM,CohereModel:()=>r.CohereModel,CoherePreTrainedModel:()=>r.CoherePreTrainedModel,CohereTokenizer:()=>_.CohereTokenizer,ConvBertForMaskedLM:()=>r.ConvBertForMaskedLM,ConvBertForQuestionAnswering:()=>r.ConvBertForQuestionAnswering,ConvBertForSequenceClassification:()=>r.ConvBertForSequenceClassification,ConvBertForTokenClassification:()=>r.ConvBertForTokenClassification,ConvBertModel:()=>r.ConvBertModel,ConvBertPreTrainedModel:()=>r.ConvBertPreTrainedModel,ConvBertTokenizer:()=>_.ConvBertTokenizer,ConvNextFeatureExtractor:()=>I.ConvNextFeatureExtractor,ConvNextForImageClassification:()=>r.ConvNextForImageClassification,ConvNextImageProcessor:()=>I.ConvNextImageProcessor,ConvNextModel:()=>r.ConvNextModel,ConvNextPreTrainedModel:()=>r.ConvNextPreTrainedModel,ConvNextV2ForImageClassification:()=>r.ConvNextV2ForImageClassification,ConvNextV2Model:()=>r.ConvNextV2Model,ConvNextV2PreTrainedModel:()=>r.ConvNextV2PreTrainedModel,DPTFeatureExtractor:()=>I.DPTFeatureExtractor,DPTForDepthEstimation:()=>r.DPTForDepthEstimation,DPTImageProcessor:()=>I.DPTImageProcessor,DPTModel:()=>r.DPTModel,DPTPreTrainedModel:()=>r.DPTPreTrainedModel,DebertaForMaskedLM:()=>r.DebertaForMaskedLM,DebertaForQuestionAnswering:()=>r.DebertaForQuestionAnswering,DebertaForSequenceClassification:()=>r.DebertaForSequenceClassification,DebertaForTokenClassification:()=>r.DebertaForTokenClassification,DebertaModel:()=>r.DebertaModel,DebertaPreTrainedModel:()=>r.DebertaPreTrainedModel,DebertaTokenizer:()=>_.DebertaTokenizer,DebertaV2ForMaskedLM:()=>r.DebertaV2ForMaskedLM,DebertaV2ForQuestionAnswering:()=>r.DebertaV2ForQuestionAnswering,DebertaV2ForSequenceClassification:()=>r.DebertaV2ForSequenceClassification,DebertaV2ForTokenClassification:()=>r.DebertaV2ForTokenClassification,DebertaV2Model:()=>r.DebertaV2Model,DebertaV2PreTrainedModel:()=>r.DebertaV2PreTrainedModel,DebertaV2Tokenizer:()=>_.DebertaV2Tokenizer,DecisionTransformerModel:()=>r.DecisionTransformerModel,DecisionTransformerPreTrainedModel:()=>r.DecisionTransformerPreTrainedModel,DeiTFeatureExtractor:()=>I.DeiTFeatureExtractor,DeiTForImageClassification:()=>r.DeiTForImageClassification,DeiTImageProcessor:()=>I.DeiTImageProcessor,DeiTModel:()=>r.DeiTModel,DeiTPreTrainedModel:()=>r.DeiTPreTrainedModel,DepthAnythingForDepthEstimation:()=>r.DepthAnythingForDepthEstimation,DepthAnythingPreTrainedModel:()=>r.DepthAnythingPreTrainedModel,DepthEstimationPipeline:()=>A.DepthEstimationPipeline,DepthProForDepthEstimation:()=>r.DepthProForDepthEstimation,DepthProPreTrainedModel:()=>r.DepthProPreTrainedModel,DetrFeatureExtractor:()=>I.DetrFeatureExtractor,DetrForObjectDetection:()=>r.DetrForObjectDetection,DetrForSegmentation:()=>r.DetrForSegmentation,DetrImageProcessor:()=>I.DetrImageProcessor,DetrModel:()=>r.DetrModel,DetrObjectDetectionOutput:()=>r.DetrObjectDetectionOutput,DetrPreTrainedModel:()=>r.DetrPreTrainedModel,DetrSegmentationOutput:()=>r.DetrSegmentationOutput,Dinov2ForImageClassification:()=>r.Dinov2ForImageClassification,Dinov2Model:()=>r.Dinov2Model,Dinov2PreTrainedModel:()=>r.Dinov2PreTrainedModel,Dinov2WithRegistersForImageClassification:()=>r.Dinov2WithRegistersForImageClassification,Dinov2WithRegistersModel:()=>r.Dinov2WithRegistersModel,Dinov2WithRegistersPreTrainedModel:()=>r.Dinov2WithRegistersPreTrainedModel,DistilBertForMaskedLM:()=>r.DistilBertForMaskedLM,DistilBertForQuestionAnswering:()=>r.DistilBertForQuestionAnswering,DistilBertForSequenceClassification:()=>r.DistilBertForSequenceClassification,DistilBertForTokenClassification:()=>r.DistilBertForTokenClassification,DistilBertModel:()=>r.DistilBertModel,DistilBertPreTrainedModel:()=>r.DistilBertPreTrainedModel,DistilBertTokenizer:()=>_.DistilBertTokenizer,DocumentQuestionAnsweringPipeline:()=>A.DocumentQuestionAnsweringPipeline,DonutFeatureExtractor:()=>I.DonutFeatureExtractor,DonutImageProcessor:()=>I.DonutImageProcessor,DonutSwinModel:()=>r.DonutSwinModel,DonutSwinPreTrainedModel:()=>r.DonutSwinPreTrainedModel,EfficientNetForImageClassification:()=>r.EfficientNetForImageClassification,EfficientNetImageProcessor:()=>I.EfficientNetImageProcessor,EfficientNetModel:()=>r.EfficientNetModel,EfficientNetPreTrainedModel:()=>r.EfficientNetPreTrainedModel,ElectraForMaskedLM:()=>r.ElectraForMaskedLM,ElectraForQuestionAnswering:()=>r.ElectraForQuestionAnswering,ElectraForSequenceClassification:()=>r.ElectraForSequenceClassification,ElectraForTokenClassification:()=>r.ElectraForTokenClassification,ElectraModel:()=>r.ElectraModel,ElectraPreTrainedModel:()=>r.ElectraPreTrainedModel,ElectraTokenizer:()=>_.ElectraTokenizer,EosTokenCriteria:()=>q.EosTokenCriteria,EsmForMaskedLM:()=>r.EsmForMaskedLM,EsmForSequenceClassification:()=>r.EsmForSequenceClassification,EsmForTokenClassification:()=>r.EsmForTokenClassification,EsmModel:()=>r.EsmModel,EsmPreTrainedModel:()=>r.EsmPreTrainedModel,EsmTokenizer:()=>_.EsmTokenizer,ExaoneForCausalLM:()=>r.ExaoneForCausalLM,ExaoneModel:()=>r.ExaoneModel,ExaonePreTrainedModel:()=>r.ExaonePreTrainedModel,FFT:()=>g.FFT,FalconForCausalLM:()=>r.FalconForCausalLM,FalconModel:()=>r.FalconModel,FalconPreTrainedModel:()=>r.FalconPreTrainedModel,FalconTokenizer:()=>_.FalconTokenizer,FastViTForImageClassification:()=>r.FastViTForImageClassification,FastViTModel:()=>r.FastViTModel,FastViTPreTrainedModel:()=>r.FastViTPreTrainedModel,FeatureExtractionPipeline:()=>A.FeatureExtractionPipeline,FeatureExtractor:()=>v.FeatureExtractor,FillMaskPipeline:()=>A.FillMaskPipeline,Florence2ForConditionalGeneration:()=>r.Florence2ForConditionalGeneration,Florence2PreTrainedModel:()=>r.Florence2PreTrainedModel,Florence2Processor:()=>ne.Florence2Processor,ForcedBOSTokenLogitsProcessor:()=>$.ForcedBOSTokenLogitsProcessor,ForcedEOSTokenLogitsProcessor:()=>$.ForcedEOSTokenLogitsProcessor,GLPNFeatureExtractor:()=>I.GLPNFeatureExtractor,GLPNForDepthEstimation:()=>r.GLPNForDepthEstimation,GLPNModel:()=>r.GLPNModel,GLPNPreTrainedModel:()=>r.GLPNPreTrainedModel,GPT2LMHeadModel:()=>r.GPT2LMHeadModel,GPT2Model:()=>r.GPT2Model,GPT2PreTrainedModel:()=>r.GPT2PreTrainedModel,GPT2Tokenizer:()=>_.GPT2Tokenizer,GPTBigCodeForCausalLM:()=>r.GPTBigCodeForCausalLM,GPTBigCodeModel:()=>r.GPTBigCodeModel,GPTBigCodePreTrainedModel:()=>r.GPTBigCodePreTrainedModel,GPTJForCausalLM:()=>r.GPTJForCausalLM,GPTJModel:()=>r.GPTJModel,GPTJPreTrainedModel:()=>r.GPTJPreTrainedModel,GPTNeoForCausalLM:()=>r.GPTNeoForCausalLM,GPTNeoModel:()=>r.GPTNeoModel,GPTNeoPreTrainedModel:()=>r.GPTNeoPreTrainedModel,GPTNeoXForCausalLM:()=>r.GPTNeoXForCausalLM,GPTNeoXModel:()=>r.GPTNeoXModel,GPTNeoXPreTrainedModel:()=>r.GPTNeoXPreTrainedModel,GPTNeoXTokenizer:()=>_.GPTNeoXTokenizer,Gemma2ForCausalLM:()=>r.Gemma2ForCausalLM,Gemma2Model:()=>r.Gemma2Model,Gemma2PreTrainedModel:()=>r.Gemma2PreTrainedModel,GemmaForCausalLM:()=>r.GemmaForCausalLM,GemmaModel:()=>r.GemmaModel,GemmaPreTrainedModel:()=>r.GemmaPreTrainedModel,GemmaTokenizer:()=>_.GemmaTokenizer,GlmForCausalLM:()=>r.GlmForCausalLM,GlmModel:()=>r.GlmModel,GlmPreTrainedModel:()=>r.GlmPreTrainedModel,GraniteForCausalLM:()=>r.GraniteForCausalLM,GraniteModel:()=>r.GraniteModel,GranitePreTrainedModel:()=>r.GranitePreTrainedModel,Grok1Tokenizer:()=>_.Grok1Tokenizer,GroundingDinoForObjectDetection:()=>r.GroundingDinoForObjectDetection,GroundingDinoImageProcessor:()=>I.GroundingDinoImageProcessor,GroundingDinoPreTrainedModel:()=>r.GroundingDinoPreTrainedModel,GroundingDinoProcessor:()=>ne.GroundingDinoProcessor,GroupViTModel:()=>r.GroupViTModel,GroupViTPreTrainedModel:()=>r.GroupViTPreTrainedModel,HeliumForCausalLM:()=>r.HeliumForCausalLM,HeliumModel:()=>r.HeliumModel,HeliumPreTrainedModel:()=>r.HeliumPreTrainedModel,HerbertTokenizer:()=>_.HerbertTokenizer,HieraForImageClassification:()=>r.HieraForImageClassification,HieraModel:()=>r.HieraModel,HieraPreTrainedModel:()=>r.HieraPreTrainedModel,HubertForCTC:()=>r.HubertForCTC,HubertForSequenceClassification:()=>r.HubertForSequenceClassification,HubertModel:()=>r.HubertModel,HubertPreTrainedModel:()=>r.HubertPreTrainedModel,IJepaForImageClassification:()=>r.IJepaForImageClassification,IJepaModel:()=>r.IJepaModel,IJepaPreTrainedModel:()=>r.IJepaPreTrainedModel,Idefics3ForConditionalGeneration:()=>r.Idefics3ForConditionalGeneration,Idefics3ImageProcessor:()=>I.Idefics3ImageProcessor,Idefics3PreTrainedModel:()=>r.Idefics3PreTrainedModel,Idefics3Processor:()=>ne.Idefics3Processor,ImageClassificationPipeline:()=>A.ImageClassificationPipeline,ImageFeatureExtractionPipeline:()=>A.ImageFeatureExtractionPipeline,ImageFeatureExtractor:()=>M.ImageFeatureExtractor,ImageMattingOutput:()=>r.ImageMattingOutput,ImageProcessor:()=>b.ImageProcessor,ImageSegmentationPipeline:()=>A.ImageSegmentationPipeline,ImageToImagePipeline:()=>A.ImageToImagePipeline,ImageToTextPipeline:()=>A.ImageToTextPipeline,InterruptableStoppingCriteria:()=>q.InterruptableStoppingCriteria,JAISLMHeadModel:()=>r.JAISLMHeadModel,JAISModel:()=>r.JAISModel,JAISPreTrainedModel:()=>r.JAISPreTrainedModel,JinaCLIPImageProcessor:()=>I.JinaCLIPImageProcessor,JinaCLIPModel:()=>r.JinaCLIPModel,JinaCLIPPreTrainedModel:()=>r.JinaCLIPPreTrainedModel,JinaCLIPProcessor:()=>ne.JinaCLIPProcessor,JinaCLIPTextModel:()=>r.JinaCLIPTextModel,JinaCLIPVisionModel:()=>r.JinaCLIPVisionModel,LlamaForCausalLM:()=>r.LlamaForCausalLM,LlamaModel:()=>r.LlamaModel,LlamaPreTrainedModel:()=>r.LlamaPreTrainedModel,LlamaTokenizer:()=>_.LlamaTokenizer,LlavaForConditionalGeneration:()=>r.LlavaForConditionalGeneration,LlavaOnevisionForConditionalGeneration:()=>r.LlavaOnevisionForConditionalGeneration,LlavaOnevisionImageProcessor:()=>I.LlavaOnevisionImageProcessor,LlavaPreTrainedModel:()=>r.LlavaPreTrainedModel,LogitsProcessor:()=>$.LogitsProcessor,LogitsProcessorList:()=>$.LogitsProcessorList,LogitsWarper:()=>$.LogitsWarper,LongT5ForConditionalGeneration:()=>r.LongT5ForConditionalGeneration,LongT5Model:()=>r.LongT5Model,LongT5PreTrainedModel:()=>r.LongT5PreTrainedModel,M2M100ForConditionalGeneration:()=>r.M2M100ForConditionalGeneration,M2M100Model:()=>r.M2M100Model,M2M100PreTrainedModel:()=>r.M2M100PreTrainedModel,M2M100Tokenizer:()=>_.M2M100Tokenizer,MBart50Tokenizer:()=>_.MBart50Tokenizer,MBartForCausalLM:()=>r.MBartForCausalLM,MBartForConditionalGeneration:()=>r.MBartForConditionalGeneration,MBartForSequenceClassification:()=>r.MBartForSequenceClassification,MBartModel:()=>r.MBartModel,MBartPreTrainedModel:()=>r.MBartPreTrainedModel,MBartTokenizer:()=>_.MBartTokenizer,MPNetForMaskedLM:()=>r.MPNetForMaskedLM,MPNetForQuestionAnswering:()=>r.MPNetForQuestionAnswering,MPNetForSequenceClassification:()=>r.MPNetForSequenceClassification,MPNetForTokenClassification:()=>r.MPNetForTokenClassification,MPNetModel:()=>r.MPNetModel,MPNetPreTrainedModel:()=>r.MPNetPreTrainedModel,MPNetTokenizer:()=>_.MPNetTokenizer,MT5ForConditionalGeneration:()=>r.MT5ForConditionalGeneration,MT5Model:()=>r.MT5Model,MT5PreTrainedModel:()=>r.MT5PreTrainedModel,MarianMTModel:()=>r.MarianMTModel,MarianModel:()=>r.MarianModel,MarianPreTrainedModel:()=>r.MarianPreTrainedModel,MarianTokenizer:()=>_.MarianTokenizer,Mask2FormerImageProcessor:()=>I.Mask2FormerImageProcessor,MaskFormerFeatureExtractor:()=>I.MaskFormerFeatureExtractor,MaskFormerForInstanceSegmentation:()=>r.MaskFormerForInstanceSegmentation,MaskFormerImageProcessor:()=>I.MaskFormerImageProcessor,MaskFormerModel:()=>r.MaskFormerModel,MaskFormerPreTrainedModel:()=>r.MaskFormerPreTrainedModel,MaskedLMOutput:()=>r.MaskedLMOutput,MaxLengthCriteria:()=>q.MaxLengthCriteria,MgpstrForSceneTextRecognition:()=>r.MgpstrForSceneTextRecognition,MgpstrModelOutput:()=>r.MgpstrModelOutput,MgpstrPreTrainedModel:()=>r.MgpstrPreTrainedModel,MgpstrProcessor:()=>ne.MgpstrProcessor,MgpstrTokenizer:()=>_.MgpstrTokenizer,MinLengthLogitsProcessor:()=>$.MinLengthLogitsProcessor,MinNewTokensLengthLogitsProcessor:()=>$.MinNewTokensLengthLogitsProcessor,MistralForCausalLM:()=>r.MistralForCausalLM,MistralModel:()=>r.MistralModel,MistralPreTrainedModel:()=>r.MistralPreTrainedModel,MobileBertForMaskedLM:()=>r.MobileBertForMaskedLM,MobileBertForQuestionAnswering:()=>r.MobileBertForQuestionAnswering,MobileBertForSequenceClassification:()=>r.MobileBertForSequenceClassification,MobileBertModel:()=>r.MobileBertModel,MobileBertPreTrainedModel:()=>r.MobileBertPreTrainedModel,MobileBertTokenizer:()=>_.MobileBertTokenizer,MobileLLMForCausalLM:()=>r.MobileLLMForCausalLM,MobileLLMModel:()=>r.MobileLLMModel,MobileLLMPreTrainedModel:()=>r.MobileLLMPreTrainedModel,MobileNetV1FeatureExtractor:()=>I.MobileNetV1FeatureExtractor,MobileNetV1ForImageClassification:()=>r.MobileNetV1ForImageClassification,MobileNetV1ImageProcessor:()=>I.MobileNetV1ImageProcessor,MobileNetV1Model:()=>r.MobileNetV1Model,MobileNetV1PreTrainedModel:()=>r.MobileNetV1PreTrainedModel,MobileNetV2FeatureExtractor:()=>I.MobileNetV2FeatureExtractor,MobileNetV2ForImageClassification:()=>r.MobileNetV2ForImageClassification,MobileNetV2ImageProcessor:()=>I.MobileNetV2ImageProcessor,MobileNetV2Model:()=>r.MobileNetV2Model,MobileNetV2PreTrainedModel:()=>r.MobileNetV2PreTrainedModel,MobileNetV3FeatureExtractor:()=>I.MobileNetV3FeatureExtractor,MobileNetV3ForImageClassification:()=>r.MobileNetV3ForImageClassification,MobileNetV3ImageProcessor:()=>I.MobileNetV3ImageProcessor,MobileNetV3Model:()=>r.MobileNetV3Model,MobileNetV3PreTrainedModel:()=>r.MobileNetV3PreTrainedModel,MobileNetV4FeatureExtractor:()=>I.MobileNetV4FeatureExtractor,MobileNetV4ForImageClassification:()=>r.MobileNetV4ForImageClassification,MobileNetV4ImageProcessor:()=>I.MobileNetV4ImageProcessor,MobileNetV4Model:()=>r.MobileNetV4Model,MobileNetV4PreTrainedModel:()=>r.MobileNetV4PreTrainedModel,MobileViTFeatureExtractor:()=>I.MobileViTFeatureExtractor,MobileViTForImageClassification:()=>r.MobileViTForImageClassification,MobileViTImageProcessor:()=>I.MobileViTImageProcessor,MobileViTModel:()=>r.MobileViTModel,MobileViTPreTrainedModel:()=>r.MobileViTPreTrainedModel,MobileViTV2ForImageClassification:()=>r.MobileViTV2ForImageClassification,MobileViTV2Model:()=>r.MobileViTV2Model,MobileViTV2PreTrainedModel:()=>r.MobileViTV2PreTrainedModel,ModelOutput:()=>r.ModelOutput,ModernBertForMaskedLM:()=>r.ModernBertForMaskedLM,ModernBertForSequenceClassification:()=>r.ModernBertForSequenceClassification,ModernBertForTokenClassification:()=>r.ModernBertForTokenClassification,ModernBertModel:()=>r.ModernBertModel,ModernBertPreTrainedModel:()=>r.ModernBertPreTrainedModel,Moondream1ForConditionalGeneration:()=>r.Moondream1ForConditionalGeneration,MoonshineFeatureExtractor:()=>M.MoonshineFeatureExtractor,MoonshineForConditionalGeneration:()=>r.MoonshineForConditionalGeneration,MoonshineModel:()=>r.MoonshineModel,MoonshinePreTrainedModel:()=>r.MoonshinePreTrainedModel,MoonshineProcessor:()=>ne.MoonshineProcessor,MptForCausalLM:()=>r.MptForCausalLM,MptModel:()=>r.MptModel,MptPreTrainedModel:()=>r.MptPreTrainedModel,MultiModalityCausalLM:()=>r.MultiModalityCausalLM,MultiModalityPreTrainedModel:()=>r.MultiModalityPreTrainedModel,MusicgenForCausalLM:()=>r.MusicgenForCausalLM,MusicgenForConditionalGeneration:()=>r.MusicgenForConditionalGeneration,MusicgenModel:()=>r.MusicgenModel,MusicgenPreTrainedModel:()=>r.MusicgenPreTrainedModel,NllbTokenizer:()=>_.NllbTokenizer,NoBadWordsLogitsProcessor:()=>$.NoBadWordsLogitsProcessor,NoRepeatNGramLogitsProcessor:()=>$.NoRepeatNGramLogitsProcessor,NomicBertModel:()=>r.NomicBertModel,NomicBertPreTrainedModel:()=>r.NomicBertPreTrainedModel,NougatImageProcessor:()=>I.NougatImageProcessor,NougatTokenizer:()=>_.NougatTokenizer,OPTForCausalLM:()=>r.OPTForCausalLM,OPTModel:()=>r.OPTModel,OPTPreTrainedModel:()=>r.OPTPreTrainedModel,ObjectDetectionPipeline:()=>A.ObjectDetectionPipeline,Olmo2ForCausalLM:()=>r.Olmo2ForCausalLM,Olmo2Model:()=>r.Olmo2Model,Olmo2PreTrainedModel:()=>r.Olmo2PreTrainedModel,OlmoForCausalLM:()=>r.OlmoForCausalLM,OlmoModel:()=>r.OlmoModel,OlmoPreTrainedModel:()=>r.OlmoPreTrainedModel,OpenELMForCausalLM:()=>r.OpenELMForCausalLM,OpenELMModel:()=>r.OpenELMModel,OpenELMPreTrainedModel:()=>r.OpenELMPreTrainedModel,OwlViTFeatureExtractor:()=>I.OwlViTFeatureExtractor,OwlViTForObjectDetection:()=>r.OwlViTForObjectDetection,OwlViTImageProcessor:()=>I.OwlViTImageProcessor,OwlViTModel:()=>r.OwlViTModel,OwlViTPreTrainedModel:()=>r.OwlViTPreTrainedModel,OwlViTProcessor:()=>ne.OwlViTProcessor,Owlv2ForObjectDetection:()=>r.Owlv2ForObjectDetection,Owlv2ImageProcessor:()=>I.Owlv2ImageProcessor,Owlv2Model:()=>r.Owlv2Model,Owlv2PreTrainedModel:()=>r.Owlv2PreTrainedModel,PaliGemmaForConditionalGeneration:()=>r.PaliGemmaForConditionalGeneration,PaliGemmaPreTrainedModel:()=>r.PaliGemmaPreTrainedModel,PaliGemmaProcessor:()=>ne.PaliGemmaProcessor,PatchTSMixerForPrediction:()=>r.PatchTSMixerForPrediction,PatchTSMixerModel:()=>r.PatchTSMixerModel,PatchTSMixerPreTrainedModel:()=>r.PatchTSMixerPreTrainedModel,PatchTSTForPrediction:()=>r.PatchTSTForPrediction,PatchTSTModel:()=>r.PatchTSTModel,PatchTSTPreTrainedModel:()=>r.PatchTSTPreTrainedModel,Phi3ForCausalLM:()=>r.Phi3ForCausalLM,Phi3Model:()=>r.Phi3Model,Phi3PreTrainedModel:()=>r.Phi3PreTrainedModel,Phi3VForCausalLM:()=>r.Phi3VForCausalLM,Phi3VImageProcessor:()=>I.Phi3VImageProcessor,Phi3VPreTrainedModel:()=>r.Phi3VPreTrainedModel,Phi3VProcessor:()=>ne.Phi3VProcessor,PhiForCausalLM:()=>r.PhiForCausalLM,PhiModel:()=>r.PhiModel,PhiPreTrainedModel:()=>r.PhiPreTrainedModel,Pipeline:()=>A.Pipeline,PreTrainedModel:()=>r.PreTrainedModel,PreTrainedTokenizer:()=>_.PreTrainedTokenizer,PretrainedConfig:()=>D.PretrainedConfig,PretrainedMixin:()=>r.PretrainedMixin,Processor:()=>se.Processor,PvtForImageClassification:()=>r.PvtForImageClassification,PvtImageProcessor:()=>I.PvtImageProcessor,PvtModel:()=>r.PvtModel,PvtPreTrainedModel:()=>r.PvtPreTrainedModel,PyAnnoteFeatureExtractor:()=>M.PyAnnoteFeatureExtractor,PyAnnoteForAudioFrameClassification:()=>r.PyAnnoteForAudioFrameClassification,PyAnnoteModel:()=>r.PyAnnoteModel,PyAnnotePreTrainedModel:()=>r.PyAnnotePreTrainedModel,PyAnnoteProcessor:()=>ne.PyAnnoteProcessor,QuestionAnsweringModelOutput:()=>r.QuestionAnsweringModelOutput,QuestionAnsweringPipeline:()=>A.QuestionAnsweringPipeline,Qwen2ForCausalLM:()=>r.Qwen2ForCausalLM,Qwen2Model:()=>r.Qwen2Model,Qwen2PreTrainedModel:()=>r.Qwen2PreTrainedModel,Qwen2Tokenizer:()=>_.Qwen2Tokenizer,Qwen2VLForConditionalGeneration:()=>r.Qwen2VLForConditionalGeneration,Qwen2VLImageProcessor:()=>I.Qwen2VLImageProcessor,Qwen2VLPreTrainedModel:()=>r.Qwen2VLPreTrainedModel,Qwen2VLProcessor:()=>ne.Qwen2VLProcessor,RTDetrForObjectDetection:()=>r.RTDetrForObjectDetection,RTDetrImageProcessor:()=>I.RTDetrImageProcessor,RTDetrModel:()=>r.RTDetrModel,RTDetrObjectDetectionOutput:()=>r.RTDetrObjectDetectionOutput,RTDetrPreTrainedModel:()=>r.RTDetrPreTrainedModel,RawAudio:()=>j.RawAudio,RawImage:()=>X.RawImage,RepetitionPenaltyLogitsProcessor:()=>$.RepetitionPenaltyLogitsProcessor,ResNetForImageClassification:()=>r.ResNetForImageClassification,ResNetModel:()=>r.ResNetModel,ResNetPreTrainedModel:()=>r.ResNetPreTrainedModel,RoFormerForMaskedLM:()=>r.RoFormerForMaskedLM,RoFormerForQuestionAnswering:()=>r.RoFormerForQuestionAnswering,RoFormerForSequenceClassification:()=>r.RoFormerForSequenceClassification,RoFormerForTokenClassification:()=>r.RoFormerForTokenClassification,RoFormerModel:()=>r.RoFormerModel,RoFormerPreTrainedModel:()=>r.RoFormerPreTrainedModel,RoFormerTokenizer:()=>_.RoFormerTokenizer,RobertaForMaskedLM:()=>r.RobertaForMaskedLM,RobertaForQuestionAnswering:()=>r.RobertaForQuestionAnswering,RobertaForSequenceClassification:()=>r.RobertaForSequenceClassification,RobertaForTokenClassification:()=>r.RobertaForTokenClassification,RobertaModel:()=>r.RobertaModel,RobertaPreTrainedModel:()=>r.RobertaPreTrainedModel,RobertaTokenizer:()=>_.RobertaTokenizer,SamImageProcessor:()=>I.SamImageProcessor,SamImageSegmentationOutput:()=>r.SamImageSegmentationOutput,SamModel:()=>r.SamModel,SamPreTrainedModel:()=>r.SamPreTrainedModel,SamProcessor:()=>ne.SamProcessor,SapiensForDepthEstimation:()=>r.SapiensForDepthEstimation,SapiensForNormalEstimation:()=>r.SapiensForNormalEstimation,SapiensForSemanticSegmentation:()=>r.SapiensForSemanticSegmentation,SapiensPreTrainedModel:()=>r.SapiensPreTrainedModel,SeamlessM4TFeatureExtractor:()=>M.SeamlessM4TFeatureExtractor,SegformerFeatureExtractor:()=>I.SegformerFeatureExtractor,SegformerForImageClassification:()=>r.SegformerForImageClassification,SegformerForSemanticSegmentation:()=>r.SegformerForSemanticSegmentation,SegformerImageProcessor:()=>I.SegformerImageProcessor,SegformerModel:()=>r.SegformerModel,SegformerPreTrainedModel:()=>r.SegformerPreTrainedModel,Seq2SeqLMOutput:()=>r.Seq2SeqLMOutput,SequenceClassifierOutput:()=>r.SequenceClassifierOutput,SiglipImageProcessor:()=>I.SiglipImageProcessor,SiglipModel:()=>r.SiglipModel,SiglipPreTrainedModel:()=>r.SiglipPreTrainedModel,SiglipTextModel:()=>r.SiglipTextModel,SiglipTokenizer:()=>_.SiglipTokenizer,SiglipVisionModel:()=>r.SiglipVisionModel,SpeechT5FeatureExtractor:()=>M.SpeechT5FeatureExtractor,SpeechT5ForSpeechToText:()=>r.SpeechT5ForSpeechToText,SpeechT5ForTextToSpeech:()=>r.SpeechT5ForTextToSpeech,SpeechT5HifiGan:()=>r.SpeechT5HifiGan,SpeechT5Model:()=>r.SpeechT5Model,SpeechT5PreTrainedModel:()=>r.SpeechT5PreTrainedModel,SpeechT5Processor:()=>ne.SpeechT5Processor,SpeechT5Tokenizer:()=>_.SpeechT5Tokenizer,SqueezeBertForMaskedLM:()=>r.SqueezeBertForMaskedLM,SqueezeBertForQuestionAnswering:()=>r.SqueezeBertForQuestionAnswering,SqueezeBertForSequenceClassification:()=>r.SqueezeBertForSequenceClassification,SqueezeBertModel:()=>r.SqueezeBertModel,SqueezeBertPreTrainedModel:()=>r.SqueezeBertPreTrainedModel,SqueezeBertTokenizer:()=>_.SqueezeBertTokenizer,StableLmForCausalLM:()=>r.StableLmForCausalLM,StableLmModel:()=>r.StableLmModel,StableLmPreTrainedModel:()=>r.StableLmPreTrainedModel,Starcoder2ForCausalLM:()=>r.Starcoder2ForCausalLM,Starcoder2Model:()=>r.Starcoder2Model,Starcoder2PreTrainedModel:()=>r.Starcoder2PreTrainedModel,StoppingCriteria:()=>q.StoppingCriteria,StoppingCriteriaList:()=>q.StoppingCriteriaList,StyleTextToSpeech2Model:()=>r.StyleTextToSpeech2Model,StyleTextToSpeech2PreTrainedModel:()=>r.StyleTextToSpeech2PreTrainedModel,SummarizationPipeline:()=>A.SummarizationPipeline,SuppressTokensAtBeginLogitsProcessor:()=>$.SuppressTokensAtBeginLogitsProcessor,Swin2SRForImageSuperResolution:()=>r.Swin2SRForImageSuperResolution,Swin2SRImageProcessor:()=>I.Swin2SRImageProcessor,Swin2SRModel:()=>r.Swin2SRModel,Swin2SRPreTrainedModel:()=>r.Swin2SRPreTrainedModel,SwinForImageClassification:()=>r.SwinForImageClassification,SwinModel:()=>r.SwinModel,SwinPreTrainedModel:()=>r.SwinPreTrainedModel,T5ForConditionalGeneration:()=>r.T5ForConditionalGeneration,T5Model:()=>r.T5Model,T5PreTrainedModel:()=>r.T5PreTrainedModel,T5Tokenizer:()=>_.T5Tokenizer,TableTransformerForObjectDetection:()=>r.TableTransformerForObjectDetection,TableTransformerModel:()=>r.TableTransformerModel,TableTransformerObjectDetectionOutput:()=>r.TableTransformerObjectDetectionOutput,TableTransformerPreTrainedModel:()=>r.TableTransformerPreTrainedModel,TemperatureLogitsWarper:()=>$.TemperatureLogitsWarper,Tensor:()=>R.Tensor,Text2TextGenerationPipeline:()=>A.Text2TextGenerationPipeline,TextClassificationPipeline:()=>A.TextClassificationPipeline,TextGenerationPipeline:()=>A.TextGenerationPipeline,TextStreamer:()=>U.TextStreamer,TextToAudioPipeline:()=>A.TextToAudioPipeline,TokenClassificationPipeline:()=>A.TokenClassificationPipeline,TokenClassifierOutput:()=>r.TokenClassifierOutput,TokenizerModel:()=>_.TokenizerModel,TopKLogitsWarper:()=>$.TopKLogitsWarper,TopPLogitsWarper:()=>$.TopPLogitsWarper,TrOCRForCausalLM:()=>r.TrOCRForCausalLM,TrOCRPreTrainedModel:()=>r.TrOCRPreTrainedModel,TranslationPipeline:()=>A.TranslationPipeline,UniSpeechForCTC:()=>r.UniSpeechForCTC,UniSpeechForSequenceClassification:()=>r.UniSpeechForSequenceClassification,UniSpeechModel:()=>r.UniSpeechModel,UniSpeechPreTrainedModel:()=>r.UniSpeechPreTrainedModel,UniSpeechSatForAudioFrameClassification:()=>r.UniSpeechSatForAudioFrameClassification,UniSpeechSatForCTC:()=>r.UniSpeechSatForCTC,UniSpeechSatForSequenceClassification:()=>r.UniSpeechSatForSequenceClassification,UniSpeechSatModel:()=>r.UniSpeechSatModel,UniSpeechSatPreTrainedModel:()=>r.UniSpeechSatPreTrainedModel,VLChatProcessor:()=>ne.VLChatProcessor,VLMImageProcessor:()=>I.VLMImageProcessor,ViTFeatureExtractor:()=>I.ViTFeatureExtractor,ViTForImageClassification:()=>r.ViTForImageClassification,ViTImageProcessor:()=>I.ViTImageProcessor,ViTMAEModel:()=>r.ViTMAEModel,ViTMAEPreTrainedModel:()=>r.ViTMAEPreTrainedModel,ViTMSNForImageClassification:()=>r.ViTMSNForImageClassification,ViTMSNModel:()=>r.ViTMSNModel,ViTMSNPreTrainedModel:()=>r.ViTMSNPreTrainedModel,ViTModel:()=>r.ViTModel,ViTPreTrainedModel:()=>r.ViTPreTrainedModel,VisionEncoderDecoderModel:()=>r.VisionEncoderDecoderModel,VitMatteForImageMatting:()=>r.VitMatteForImageMatting,VitMatteImageProcessor:()=>I.VitMatteImageProcessor,VitMattePreTrainedModel:()=>r.VitMattePreTrainedModel,VitPoseForPoseEstimation:()=>r.VitPoseForPoseEstimation,VitPoseImageProcessor:()=>I.VitPoseImageProcessor,VitPosePreTrainedModel:()=>r.VitPosePreTrainedModel,VitsModel:()=>r.VitsModel,VitsModelOutput:()=>r.VitsModelOutput,VitsPreTrainedModel:()=>r.VitsPreTrainedModel,VitsTokenizer:()=>_.VitsTokenizer,Wav2Vec2BertForCTC:()=>r.Wav2Vec2BertForCTC,Wav2Vec2BertForSequenceClassification:()=>r.Wav2Vec2BertForSequenceClassification,Wav2Vec2BertModel:()=>r.Wav2Vec2BertModel,Wav2Vec2BertPreTrainedModel:()=>r.Wav2Vec2BertPreTrainedModel,Wav2Vec2CTCTokenizer:()=>_.Wav2Vec2CTCTokenizer,Wav2Vec2FeatureExtractor:()=>M.Wav2Vec2FeatureExtractor,Wav2Vec2ForAudioFrameClassification:()=>r.Wav2Vec2ForAudioFrameClassification,Wav2Vec2ForCTC:()=>r.Wav2Vec2ForCTC,Wav2Vec2ForSequenceClassification:()=>r.Wav2Vec2ForSequenceClassification,Wav2Vec2Model:()=>r.Wav2Vec2Model,Wav2Vec2PreTrainedModel:()=>r.Wav2Vec2PreTrainedModel,Wav2Vec2Processor:()=>ne.Wav2Vec2Processor,Wav2Vec2ProcessorWithLM:()=>ne.Wav2Vec2ProcessorWithLM,WavLMForAudioFrameClassification:()=>r.WavLMForAudioFrameClassification,WavLMForCTC:()=>r.WavLMForCTC,WavLMForSequenceClassification:()=>r.WavLMForSequenceClassification,WavLMForXVector:()=>r.WavLMForXVector,WavLMModel:()=>r.WavLMModel,WavLMPreTrainedModel:()=>r.WavLMPreTrainedModel,WeSpeakerFeatureExtractor:()=>M.WeSpeakerFeatureExtractor,WeSpeakerResNetModel:()=>r.WeSpeakerResNetModel,WeSpeakerResNetPreTrainedModel:()=>r.WeSpeakerResNetPreTrainedModel,WhisperFeatureExtractor:()=>M.WhisperFeatureExtractor,WhisperForConditionalGeneration:()=>r.WhisperForConditionalGeneration,WhisperModel:()=>r.WhisperModel,WhisperPreTrainedModel:()=>r.WhisperPreTrainedModel,WhisperProcessor:()=>ne.WhisperProcessor,WhisperTextStreamer:()=>U.WhisperTextStreamer,WhisperTimeStampLogitsProcessor:()=>$.WhisperTimeStampLogitsProcessor,WhisperTokenizer:()=>_.WhisperTokenizer,XLMForQuestionAnswering:()=>r.XLMForQuestionAnswering,XLMForSequenceClassification:()=>r.XLMForSequenceClassification,XLMForTokenClassification:()=>r.XLMForTokenClassification,XLMModel:()=>r.XLMModel,XLMPreTrainedModel:()=>r.XLMPreTrainedModel,XLMRobertaForMaskedLM:()=>r.XLMRobertaForMaskedLM,XLMRobertaForQuestionAnswering:()=>r.XLMRobertaForQuestionAnswering,XLMRobertaForSequenceClassification:()=>r.XLMRobertaForSequenceClassification,XLMRobertaForTokenClassification:()=>r.XLMRobertaForTokenClassification,XLMRobertaModel:()=>r.XLMRobertaModel,XLMRobertaPreTrainedModel:()=>r.XLMRobertaPreTrainedModel,XLMRobertaTokenizer:()=>_.XLMRobertaTokenizer,XLMTokenizer:()=>_.XLMTokenizer,XLMWithLMHeadModel:()=>r.XLMWithLMHeadModel,XVectorOutput:()=>r.XVectorOutput,YolosFeatureExtractor:()=>I.YolosFeatureExtractor,YolosForObjectDetection:()=>r.YolosForObjectDetection,YolosImageProcessor:()=>I.YolosImageProcessor,YolosModel:()=>r.YolosModel,YolosObjectDetectionOutput:()=>r.YolosObjectDetectionOutput,YolosPreTrainedModel:()=>r.YolosPreTrainedModel,ZeroShotAudioClassificationPipeline:()=>A.ZeroShotAudioClassificationPipeline,ZeroShotClassificationPipeline:()=>A.ZeroShotClassificationPipeline,ZeroShotImageClassificationPipeline:()=>A.ZeroShotImageClassificationPipeline,ZeroShotObjectDetectionPipeline:()=>A.ZeroShotObjectDetectionPipeline,bankers_round:()=>g.bankers_round,cat:()=>R.cat,cos_sim:()=>g.cos_sim,dot:()=>g.dot,dynamic_time_warping:()=>g.dynamic_time_warping,env:()=>Le.env,full:()=>R.full,full_like:()=>R.full_like,getKeyValueShapes:()=>D.getKeyValueShapes,hamming:()=>j.hamming,hanning:()=>j.hanning,interpolate:()=>R.interpolate,interpolate_4d:()=>R.interpolate_4d,interpolate_data:()=>g.interpolate_data,is_chinese_char:()=>_.is_chinese_char,layer_norm:()=>R.layer_norm,load_image:()=>X.load_image,log_softmax:()=>g.log_softmax,magnitude:()=>g.magnitude,matmul:()=>R.matmul,max:()=>g.max,mean:()=>R.mean,mean_pooling:()=>R.mean_pooling,medianFilter:()=>g.medianFilter,mel_filter_bank:()=>j.mel_filter_bank,min:()=>g.min,ones:()=>R.ones,ones_like:()=>R.ones_like,permute:()=>R.permute,permute_data:()=>g.permute_data,pipeline:()=>A.pipeline,quantize_embeddings:()=>R.quantize_embeddings,rand:()=>R.rand,read_audio:()=>j.read_audio,rfft:()=>R.rfft,round:()=>g.round,slice:()=>R.slice,softmax:()=>g.softmax,spectrogram:()=>j.spectrogram,stack:()=>R.stack,std_mean:()=>R.std_mean,topk:()=>R.topk,window_function:()=>j.window_function,zeros:()=>R.zeros,zeros_like:()=>R.zeros_like});var Le=gs("./src/env.js"),A=gs("./src/pipelines.js"),r=gs("./src/models.js"),_=gs("./src/tokenizers.js"),D=gs("./src/configs.js"),j=gs("./src/utils/audio.js"),X=gs("./src/utils/image.js"),R=gs("./src/utils/tensor.js"),g=gs("./src/utils/maths.js"),v=gs("./src/base/feature_extraction_utils.js"),M=gs("./src/models/feature_extractors.js"),y=gs("./src/models/auto/feature_extraction_auto.js"),b=gs("./src/base/image_processors_utils.js"),I=gs("./src/models/image_processors.js"),K=gs("./src/models/auto/image_processing_auto.js"),se=gs("./src/base/processing_utils.js"),ne=gs("./src/models/processors.js"),W=gs("./src/models/auto/processing_auto.js"),U=gs("./src/generation/streamers.js"),q=gs("./src/generation/stopping_criteria.js"),$=gs("./src/generation/logits_process.js")})(),c.ASTFeatureExtractor,c.ASTForAudioClassification,c.ASTModel,c.ASTPreTrainedModel,c.AlbertForMaskedLM,c.AlbertForQuestionAnswering,c.AlbertForSequenceClassification,c.AlbertModel,c.AlbertPreTrainedModel,c.AlbertTokenizer,c.AudioClassificationPipeline,c.AutoConfig,c.AutoFeatureExtractor,c.AutoImageProcessor,c.AutoModel,c.AutoModelForAudioClassification,c.AutoModelForAudioFrameClassification,c.AutoModelForCTC,c.AutoModelForCausalLM,c.AutoModelForDepthEstimation,c.AutoModelForDocumentQuestionAnswering,c.AutoModelForImageClassification,c.AutoModelForImageFeatureExtraction,c.AutoModelForImageMatting,c.AutoModelForImageSegmentation,c.AutoModelForImageToImage,c.AutoModelForMaskGeneration,c.AutoModelForMaskedLM,c.AutoModelForNormalEstimation,c.AutoModelForObjectDetection,c.AutoModelForPoseEstimation,c.AutoModelForQuestionAnswering,c.AutoModelForSemanticSegmentation,c.AutoModelForSeq2SeqLM,c.AutoModelForSequenceClassification,c.AutoModelForSpeechSeq2Seq,c.AutoModelForTextToSpectrogram,c.AutoModelForTextToWaveform,c.AutoModelForTokenClassification,c.AutoModelForUniversalSegmentation,c.AutoModelForVision2Seq,c.AutoModelForXVector,c.AutoModelForZeroShotObjectDetection;var rf=c.AutoProcessor,nf=c.AutoTokenizer;c.AutomaticSpeechRecognitionPipeline,c.BartForConditionalGeneration,c.BartForSequenceClassification,c.BartModel,c.BartPretrainedModel,c.BartTokenizer,c.BaseModelOutput,c.BaseStreamer,c.BeitFeatureExtractor,c.BeitForImageClassification,c.BeitModel,c.BeitPreTrainedModel,c.BertForMaskedLM,c.BertForQuestionAnswering,c.BertForSequenceClassification,c.BertForTokenClassification,c.BertModel,c.BertPreTrainedModel,c.BertTokenizer,c.BitImageProcessor,c.BlenderbotForConditionalGeneration,c.BlenderbotModel,c.BlenderbotPreTrainedModel,c.BlenderbotSmallForConditionalGeneration,c.BlenderbotSmallModel,c.BlenderbotSmallPreTrainedModel,c.BlenderbotSmallTokenizer,c.BlenderbotTokenizer,c.BloomForCausalLM,c.BloomModel,c.BloomPreTrainedModel,c.BloomTokenizer,c.CLIPFeatureExtractor,c.CLIPImageProcessor,c.CLIPModel,c.CLIPPreTrainedModel,c.CLIPSegForImageSegmentation,c.CLIPSegModel,c.CLIPSegPreTrainedModel,c.CLIPTextModel,c.CLIPTextModelWithProjection,c.CLIPTokenizer,c.CLIPVisionModel,c.CLIPVisionModelWithProjection,c.CamembertForMaskedLM,c.CamembertForQuestionAnswering,c.CamembertForSequenceClassification,c.CamembertForTokenClassification,c.CamembertModel,c.CamembertPreTrainedModel,c.CamembertTokenizer,c.CausalLMOutput,c.CausalLMOutputWithPast,c.ChineseCLIPFeatureExtractor,c.ChineseCLIPModel,c.ChineseCLIPPreTrainedModel,c.ClapAudioModelWithProjection,c.ClapFeatureExtractor,c.ClapModel,c.ClapPreTrainedModel,c.ClapTextModelWithProjection,c.ClassifierFreeGuidanceLogitsProcessor,c.CodeGenForCausalLM,c.CodeGenModel,c.CodeGenPreTrainedModel,c.CodeGenTokenizer,c.CodeLlamaTokenizer,c.CohereForCausalLM,c.CohereModel,c.CoherePreTrainedModel,c.CohereTokenizer,c.ConvBertForMaskedLM,c.ConvBertForQuestionAnswering,c.ConvBertForSequenceClassification,c.ConvBertForTokenClassification,c.ConvBertModel,c.ConvBertPreTrainedModel,c.ConvBertTokenizer,c.ConvNextFeatureExtractor,c.ConvNextForImageClassification,c.ConvNextImageProcessor,c.ConvNextModel,c.ConvNextPreTrainedModel,c.ConvNextV2ForImageClassification,c.ConvNextV2Model,c.ConvNextV2PreTrainedModel,c.DPTFeatureExtractor,c.DPTForDepthEstimation,c.DPTImageProcessor,c.DPTModel,c.DPTPreTrainedModel,c.DebertaForMaskedLM,c.DebertaForQuestionAnswering,c.DebertaForSequenceClassification,c.DebertaForTokenClassification,c.DebertaModel,c.DebertaPreTrainedModel,c.DebertaTokenizer,c.DebertaV2ForMaskedLM,c.DebertaV2ForQuestionAnswering,c.DebertaV2ForSequenceClassification,c.DebertaV2ForTokenClassification,c.DebertaV2Model,c.DebertaV2PreTrainedModel,c.DebertaV2Tokenizer,c.DecisionTransformerModel,c.DecisionTransformerPreTrainedModel,c.DeiTFeatureExtractor,c.DeiTForImageClassification,c.DeiTImageProcessor,c.DeiTModel,c.DeiTPreTrainedModel,c.DepthAnythingForDepthEstimation,c.DepthAnythingPreTrainedModel,c.DepthEstimationPipeline,c.DepthProForDepthEstimation,c.DepthProPreTrainedModel,c.DetrFeatureExtractor,c.DetrForObjectDetection,c.DetrForSegmentation,c.DetrImageProcessor,c.DetrModel,c.DetrObjectDetectionOutput,c.DetrPreTrainedModel,c.DetrSegmentationOutput,c.Dinov2ForImageClassification,c.Dinov2Model,c.Dinov2PreTrainedModel,c.Dinov2WithRegistersForImageClassification,c.Dinov2WithRegistersModel,c.Dinov2WithRegistersPreTrainedModel,c.DistilBertForMaskedLM,c.DistilBertForQuestionAnswering,c.DistilBertForSequenceClassification,c.DistilBertForTokenClassification,c.DistilBertModel,c.DistilBertPreTrainedModel,c.DistilBertTokenizer,c.DocumentQuestionAnsweringPipeline,c.DonutFeatureExtractor,c.DonutImageProcessor,c.DonutSwinModel,c.DonutSwinPreTrainedModel,c.EfficientNetForImageClassification,c.EfficientNetImageProcessor,c.EfficientNetModel,c.EfficientNetPreTrainedModel,c.ElectraForMaskedLM,c.ElectraForQuestionAnswering,c.ElectraForSequenceClassification,c.ElectraForTokenClassification,c.ElectraModel,c.ElectraPreTrainedModel,c.ElectraTokenizer,c.EosTokenCriteria,c.EsmForMaskedLM,c.EsmForSequenceClassification,c.EsmForTokenClassification,c.EsmModel,c.EsmPreTrainedModel,c.EsmTokenizer,c.ExaoneForCausalLM,c.ExaoneModel,c.ExaonePreTrainedModel,c.FFT,c.FalconForCausalLM,c.FalconModel,c.FalconPreTrainedModel,c.FalconTokenizer,c.FastViTForImageClassification,c.FastViTModel,c.FastViTPreTrainedModel,c.FeatureExtractionPipeline,c.FeatureExtractor,c.FillMaskPipeline,c.Florence2ForConditionalGeneration,c.Florence2PreTrainedModel,c.Florence2Processor,c.ForcedBOSTokenLogitsProcessor,c.ForcedEOSTokenLogitsProcessor,c.GLPNFeatureExtractor,c.GLPNForDepthEstimation,c.GLPNModel,c.GLPNPreTrainedModel,c.GPT2LMHeadModel,c.GPT2Model,c.GPT2PreTrainedModel,c.GPT2Tokenizer,c.GPTBigCodeForCausalLM,c.GPTBigCodeModel,c.GPTBigCodePreTrainedModel,c.GPTJForCausalLM,c.GPTJModel,c.GPTJPreTrainedModel,c.GPTNeoForCausalLM,c.GPTNeoModel,c.GPTNeoPreTrainedModel,c.GPTNeoXForCausalLM,c.GPTNeoXModel,c.GPTNeoXPreTrainedModel,c.GPTNeoXTokenizer,c.Gemma2ForCausalLM,c.Gemma2Model,c.Gemma2PreTrainedModel,c.GemmaForCausalLM,c.GemmaModel,c.GemmaPreTrainedModel,c.GemmaTokenizer,c.GlmForCausalLM,c.GlmModel,c.GlmPreTrainedModel,c.GraniteForCausalLM,c.GraniteModel,c.GranitePreTrainedModel,c.Grok1Tokenizer,c.GroundingDinoForObjectDetection,c.GroundingDinoImageProcessor,c.GroundingDinoPreTrainedModel,c.GroundingDinoProcessor,c.GroupViTModel,c.GroupViTPreTrainedModel,c.HeliumForCausalLM,c.HeliumModel,c.HeliumPreTrainedModel,c.HerbertTokenizer,c.HieraForImageClassification,c.HieraModel,c.HieraPreTrainedModel,c.HubertForCTC,c.HubertForSequenceClassification,c.HubertModel,c.HubertPreTrainedModel,c.IJepaForImageClassification,c.IJepaModel,c.IJepaPreTrainedModel,c.Idefics3ForConditionalGeneration,c.Idefics3ImageProcessor,c.Idefics3PreTrainedModel,c.Idefics3Processor,c.ImageClassificationPipeline,c.ImageFeatureExtractionPipeline,c.ImageFeatureExtractor,c.ImageMattingOutput,c.ImageProcessor,c.ImageSegmentationPipeline,c.ImageToImagePipeline,c.ImageToTextPipeline,c.InterruptableStoppingCriteria,c.JAISLMHeadModel,c.JAISModel,c.JAISPreTrainedModel,c.JinaCLIPImageProcessor,c.JinaCLIPModel,c.JinaCLIPPreTrainedModel,c.JinaCLIPProcessor,c.JinaCLIPTextModel,c.JinaCLIPVisionModel,c.LlamaForCausalLM,c.LlamaModel,c.LlamaPreTrainedModel,c.LlamaTokenizer,c.LlavaForConditionalGeneration,c.LlavaOnevisionForConditionalGeneration,c.LlavaOnevisionImageProcessor,c.LlavaPreTrainedModel,c.LogitsProcessor,c.LogitsProcessorList,c.LogitsWarper,c.LongT5ForConditionalGeneration,c.LongT5Model,c.LongT5PreTrainedModel,c.M2M100ForConditionalGeneration,c.M2M100Model,c.M2M100PreTrainedModel,c.M2M100Tokenizer,c.MBart50Tokenizer,c.MBartForCausalLM,c.MBartForConditionalGeneration,c.MBartForSequenceClassification,c.MBartModel,c.MBartPreTrainedModel,c.MBartTokenizer,c.MPNetForMaskedLM,c.MPNetForQuestionAnswering,c.MPNetForSequenceClassification,c.MPNetForTokenClassification,c.MPNetModel,c.MPNetPreTrainedModel,c.MPNetTokenizer,c.MT5ForConditionalGeneration,c.MT5Model,c.MT5PreTrainedModel,c.MarianMTModel,c.MarianModel,c.MarianPreTrainedModel,c.MarianTokenizer,c.Mask2FormerImageProcessor,c.MaskFormerFeatureExtractor,c.MaskFormerForInstanceSegmentation,c.MaskFormerImageProcessor,c.MaskFormerModel,c.MaskFormerPreTrainedModel,c.MaskedLMOutput,c.MaxLengthCriteria,c.MgpstrForSceneTextRecognition,c.MgpstrModelOutput,c.MgpstrPreTrainedModel,c.MgpstrProcessor,c.MgpstrTokenizer,c.MinLengthLogitsProcessor,c.MinNewTokensLengthLogitsProcessor,c.MistralForCausalLM,c.MistralModel,c.MistralPreTrainedModel,c.MobileBertForMaskedLM,c.MobileBertForQuestionAnswering,c.MobileBertForSequenceClassification,c.MobileBertModel,c.MobileBertPreTrainedModel,c.MobileBertTokenizer,c.MobileLLMForCausalLM,c.MobileLLMModel,c.MobileLLMPreTrainedModel,c.MobileNetV1FeatureExtractor,c.MobileNetV1ForImageClassification,c.MobileNetV1ImageProcessor,c.MobileNetV1Model,c.MobileNetV1PreTrainedModel,c.MobileNetV2FeatureExtractor,c.MobileNetV2ForImageClassification,c.MobileNetV2ImageProcessor,c.MobileNetV2Model,c.MobileNetV2PreTrainedModel,c.MobileNetV3FeatureExtractor,c.MobileNetV3ForImageClassification,c.MobileNetV3ImageProcessor,c.MobileNetV3Model,c.MobileNetV3PreTrainedModel,c.MobileNetV4FeatureExtractor,c.MobileNetV4ForImageClassification,c.MobileNetV4ImageProcessor,c.MobileNetV4Model,c.MobileNetV4PreTrainedModel,c.MobileViTFeatureExtractor,c.MobileViTForImageClassification,c.MobileViTImageProcessor,c.MobileViTModel,c.MobileViTPreTrainedModel,c.MobileViTV2ForImageClassification,c.MobileViTV2Model,c.MobileViTV2PreTrainedModel,c.ModelOutput,c.ModernBertForMaskedLM,c.ModernBertForSequenceClassification,c.ModernBertForTokenClassification,c.ModernBertModel,c.ModernBertPreTrainedModel,c.Moondream1ForConditionalGeneration,c.MoonshineFeatureExtractor,c.MoonshineForConditionalGeneration,c.MoonshineModel,c.MoonshinePreTrainedModel,c.MoonshineProcessor,c.MptForCausalLM,c.MptModel,c.MptPreTrainedModel,c.MultiModalityCausalLM,c.MultiModalityPreTrainedModel,c.MusicgenForCausalLM,c.MusicgenForConditionalGeneration,c.MusicgenModel,c.MusicgenPreTrainedModel,c.NllbTokenizer,c.NoBadWordsLogitsProcessor,c.NoRepeatNGramLogitsProcessor,c.NomicBertModel,c.NomicBertPreTrainedModel,c.NougatImageProcessor,c.NougatTokenizer,c.OPTForCausalLM,c.OPTModel,c.OPTPreTrainedModel,c.ObjectDetectionPipeline,c.Olmo2ForCausalLM,c.Olmo2Model,c.Olmo2PreTrainedModel,c.OlmoForCausalLM,c.OlmoModel,c.OlmoPreTrainedModel,c.OpenELMForCausalLM,c.OpenELMModel,c.OpenELMPreTrainedModel,c.OwlViTFeatureExtractor,c.OwlViTForObjectDetection,c.OwlViTImageProcessor,c.OwlViTModel,c.OwlViTPreTrainedModel,c.OwlViTProcessor,c.Owlv2ForObjectDetection,c.Owlv2ImageProcessor,c.Owlv2Model,c.Owlv2PreTrainedModel,c.PaliGemmaForConditionalGeneration,c.PaliGemmaPreTrainedModel,c.PaliGemmaProcessor,c.PatchTSMixerForPrediction,c.PatchTSMixerModel,c.PatchTSMixerPreTrainedModel,c.PatchTSTForPrediction,c.PatchTSTModel,c.PatchTSTPreTrainedModel,c.Phi3ForCausalLM,c.Phi3Model,c.Phi3PreTrainedModel,c.Phi3VForCausalLM,c.Phi3VImageProcessor,c.Phi3VPreTrainedModel,c.Phi3VProcessor,c.PhiForCausalLM,c.PhiModel,c.PhiPreTrainedModel,c.Pipeline,c.PreTrainedModel,c.PreTrainedTokenizer,c.PretrainedConfig,c.PretrainedMixin,c.Processor,c.PvtForImageClassification,c.PvtImageProcessor,c.PvtModel,c.PvtPreTrainedModel,c.PyAnnoteFeatureExtractor,c.PyAnnoteForAudioFrameClassification,c.PyAnnoteModel,c.PyAnnotePreTrainedModel,c.PyAnnoteProcessor,c.QuestionAnsweringModelOutput,c.QuestionAnsweringPipeline,c.Qwen2ForCausalLM,c.Qwen2Model,c.Qwen2PreTrainedModel,c.Qwen2Tokenizer,c.Qwen2VLForConditionalGeneration,c.Qwen2VLImageProcessor,c.Qwen2VLPreTrainedModel,c.Qwen2VLProcessor,c.RTDetrForObjectDetection,c.RTDetrImageProcessor,c.RTDetrModel,c.RTDetrObjectDetectionOutput,c.RTDetrPreTrainedModel,c.RawAudio,c.RawImage,c.RepetitionPenaltyLogitsProcessor,c.ResNetForImageClassification,c.ResNetModel,c.ResNetPreTrainedModel,c.RoFormerForMaskedLM,c.RoFormerForQuestionAnswering,c.RoFormerForSequenceClassification,c.RoFormerForTokenClassification,c.RoFormerModel,c.RoFormerPreTrainedModel,c.RoFormerTokenizer,c.RobertaForMaskedLM,c.RobertaForQuestionAnswering,c.RobertaForSequenceClassification,c.RobertaForTokenClassification,c.RobertaModel,c.RobertaPreTrainedModel,c.RobertaTokenizer,c.SamImageProcessor,c.SamImageSegmentationOutput,c.SamModel,c.SamPreTrainedModel,c.SamProcessor,c.SapiensForDepthEstimation,c.SapiensForNormalEstimation,c.SapiensForSemanticSegmentation,c.SapiensPreTrainedModel,c.SeamlessM4TFeatureExtractor,c.SegformerFeatureExtractor,c.SegformerForImageClassification,c.SegformerForSemanticSegmentation,c.SegformerImageProcessor,c.SegformerModel,c.SegformerPreTrainedModel,c.Seq2SeqLMOutput,c.SequenceClassifierOutput,c.SiglipImageProcessor,c.SiglipModel,c.SiglipPreTrainedModel,c.SiglipTextModel,c.SiglipTokenizer,c.SiglipVisionModel,c.SpeechT5FeatureExtractor,c.SpeechT5ForSpeechToText,c.SpeechT5ForTextToSpeech,c.SpeechT5HifiGan,c.SpeechT5Model,c.SpeechT5PreTrainedModel,c.SpeechT5Processor,c.SpeechT5Tokenizer,c.SqueezeBertForMaskedLM,c.SqueezeBertForQuestionAnswering,c.SqueezeBertForSequenceClassification,c.SqueezeBertModel,c.SqueezeBertPreTrainedModel,c.SqueezeBertTokenizer,c.StableLmForCausalLM,c.StableLmModel,c.StableLmPreTrainedModel,c.Starcoder2ForCausalLM,c.Starcoder2Model,c.Starcoder2PreTrainedModel,c.StoppingCriteria,c.StoppingCriteriaList,c.StyleTextToSpeech2Model,c.StyleTextToSpeech2PreTrainedModel,c.SummarizationPipeline,c.SuppressTokensAtBeginLogitsProcessor,c.Swin2SRForImageSuperResolution,c.Swin2SRImageProcessor,c.Swin2SRModel,c.Swin2SRPreTrainedModel,c.SwinForImageClassification,c.SwinModel,c.SwinPreTrainedModel,c.T5ForConditionalGeneration,c.T5Model,c.T5PreTrainedModel,c.T5Tokenizer,c.TableTransformerForObjectDetection,c.TableTransformerModel,c.TableTransformerObjectDetectionOutput,c.TableTransformerPreTrainedModel,c.TemperatureLogitsWarper,c.Tensor,c.Text2TextGenerationPipeline,c.TextClassificationPipeline,c.TextGenerationPipeline;var of=c.TextStreamer;c.TextToAudioPipeline,c.TokenClassificationPipeline,c.TokenClassifierOutput,c.TokenizerModel,c.TopKLogitsWarper,c.TopPLogitsWarper,c.TrOCRForCausalLM,c.TrOCRPreTrainedModel,c.TranslationPipeline,c.UniSpeechForCTC,c.UniSpeechForSequenceClassification,c.UniSpeechModel,c.UniSpeechPreTrainedModel,c.UniSpeechSatForAudioFrameClassification,c.UniSpeechSatForCTC,c.UniSpeechSatForSequenceClassification,c.UniSpeechSatModel,c.UniSpeechSatPreTrainedModel,c.VLChatProcessor,c.VLMImageProcessor,c.ViTFeatureExtractor,c.ViTForImageClassification,c.ViTImageProcessor,c.ViTMAEModel,c.ViTMAEPreTrainedModel,c.ViTMSNForImageClassification,c.ViTMSNModel,c.ViTMSNPreTrainedModel,c.ViTModel,c.ViTPreTrainedModel,c.VisionEncoderDecoderModel,c.VitMatteForImageMatting,c.VitMatteImageProcessor,c.VitMattePreTrainedModel,c.VitPoseForPoseEstimation,c.VitPoseImageProcessor,c.VitPosePreTrainedModel,c.VitsModel,c.VitsModelOutput,c.VitsPreTrainedModel,c.VitsTokenizer,c.Wav2Vec2BertForCTC,c.Wav2Vec2BertForSequenceClassification,c.Wav2Vec2BertModel,c.Wav2Vec2BertPreTrainedModel,c.Wav2Vec2CTCTokenizer,c.Wav2Vec2FeatureExtractor,c.Wav2Vec2ForAudioFrameClassification,c.Wav2Vec2ForCTC,c.Wav2Vec2ForSequenceClassification,c.Wav2Vec2Model,c.Wav2Vec2PreTrainedModel,c.Wav2Vec2Processor,c.Wav2Vec2ProcessorWithLM,c.WavLMForAudioFrameClassification,c.WavLMForCTC,c.WavLMForSequenceClassification,c.WavLMForXVector,c.WavLMModel,c.WavLMPreTrainedModel,c.WeSpeakerFeatureExtractor,c.WeSpeakerResNetModel,c.WeSpeakerResNetPreTrainedModel,c.WhisperFeatureExtractor;var af=c.WhisperForConditionalGeneration;c.WhisperModel,c.WhisperPreTrainedModel,c.WhisperProcessor,c.WhisperTextStreamer,c.WhisperTimeStampLogitsProcessor,c.WhisperTokenizer,c.XLMForQuestionAnswering,c.XLMForSequenceClassification,c.XLMForTokenClassification,c.XLMModel,c.XLMPreTrainedModel,c.XLMRobertaForMaskedLM,c.XLMRobertaForQuestionAnswering,c.XLMRobertaForSequenceClassification,c.XLMRobertaForTokenClassification,c.XLMRobertaModel,c.XLMRobertaPreTrainedModel,c.XLMRobertaTokenizer,c.XLMTokenizer,c.XLMWithLMHeadModel,c.XVectorOutput,c.YolosFeatureExtractor,c.YolosForObjectDetection,c.YolosImageProcessor,c.YolosModel,c.YolosObjectDetectionOutput,c.YolosPreTrainedModel,c.ZeroShotAudioClassificationPipeline,c.ZeroShotClassificationPipeline,c.ZeroShotImageClassificationPipeline,c.ZeroShotObjectDetectionPipeline,c.bankers_round,c.cat,c.cos_sim,c.dot,c.dynamic_time_warping,c.env;var lf=c.full;c.full_like,c.getKeyValueShapes,c.hamming,c.hanning,c.interpolate,c.interpolate_4d,c.interpolate_data,c.is_chinese_char,c.layer_norm,c.load_image,c.log_softmax,c.magnitude,c.matmul,c.max,c.mean,c.mean_pooling,c.medianFilter,c.mel_filter_bank,c.min,c.ones,c.ones_like,c.permute,c.permute_data,c.pipeline,c.quantize_embeddings,c.rand,c.read_audio,c.rfft,c.round,c.slice,c.softmax,c.spectrogram,c.stack,c.std_mean,c.topk,c.window_function,c.zeros,c.zeros_like;const uf=64;class ji{static async getInstance(A){return this.model_id="onnx-community/whisper-base",this.tokenizer??(this.tokenizer=nf.from_pretrained(this.model_id,{progress_callback:A})),this.processor??(this.processor=rf.from_pretrained(this.model_id,{progress_callback:A})),this.model??(this.model=af.from_pretrained(this.model_id,{dtype:{encoder_model:"fp32",decoder_model_merged:"q4"},device:"webgpu",progress_callback:A})),Promise.all([this.tokenizer,this.processor,this.model])}}me(ji,"model_id",null),me(ji,"tokenizer"),me(ji,"processor"),me(ji,"model");let Gp=!1;async function df({audio:Le,language:A}){if(Gp)return;Gp=!0,self.postMessage({status:"start"});const[r,_,D]=await ji.getInstance();let j,X=0;const R=b=>{j??(j=performance.now());let I;X++>0&&(I=X/(performance.now()-j)*1e3),self.postMessage({status:"update",output:b,tps:I,numTokens:X})},g=new of(r,{skip_prompt:!0,decode_kwargs:{skip_special_tokens:!0},callback_function:R}),v=await _(Le),M=await D.generate({...v,max_new_tokens:uf,language:A,streamer:g}),y=r.batch_decode(M,{skip_special_tokens:!0});self.postMessage({status:"complete",output:y}),Gp=!1}async function cf(){self.postMessage({status:"loading",data:"Loading model..."});const[Le,A,r]=await ji.getInstance(_=>{self.postMessage(_)});self.postMessage({status:"loading",data:"Compiling shaders and warming up model..."}),await r.generate({input_features:lf([1,80,3e3],0),max_new_tokens:1}),self.postMessage({status:"ready"})}self.addEventListener("message",async Le=>{const{type:A,data:r}=Le.data;switch(A){case"load":cf();break;case"generate":df(r);break}})})();