@huggingface/transformers 3.3.3 → 3.4.0

package/dist/transformers.min.cjs

@@ -1,2 +0,0 @@
-(()=>{var e,t,s={fs:e=>{"use strict";e.exports=require("fs")},"onnxruntime-node":e=>{"use strict";e.exports=require("onnxruntime-node")},path:e=>{"use strict";e.exports=require("path")},sharp:e=>{"use strict";e.exports=require("sharp")},url:e=>{"use strict";e.exports=require("url")},"?8b6b":()=>{},"./node_modules/@huggingface/jinja/dist/index.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Environment:()=>Y,Interpreter:()=>K,Template:()=>te,parse:()=>N,tokenize:()=>d});var r=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"}),n=Object.freeze({set:r.Set,for:r.For,in:r.In,is:r.Is,if:r.If,else:r.Else,endif:r.EndIf,elif:r.ElseIf,endfor:r.EndFor,and:r.And,or:r.Or,not:r.Not,"not in":r.NotIn,macro:r.Macro,endmacro:r.EndMacro,true:r.BooleanLiteral,false:r.BooleanLiteral,none:r.NullLiteral,True:r.BooleanLiteral,False:r.BooleanLiteral,None:r.NullLiteral}),o=class{constructor(e,t){this.value=e,this.type=t}};function i(e){return/\w/.test(e)}function a(e){return/[0-9]/.test(e)}var l=[["{%",r.OpenStatement],["%}",r.CloseStatement],["{{",r.OpenExpression],["}}",r.CloseExpression],["(",r.OpenParen],[")",r.CloseParen],["{",r.OpenCurlyBracket],["}",r.CloseCurlyBracket],["[",r.OpenSquareBracket],["]",r.CloseSquareBracket],[",",r.Comma],[".",r.Dot],[":",r.Colon],["|",r.Pipe],["<=",r.ComparisonBinaryOperator],[">=",r.ComparisonBinaryOperator],["==",r.ComparisonBinaryOperator],["!=",r.ComparisonBinaryOperator],["<",r.ComparisonBinaryOperator],[">",r.ComparisonBinaryOperator],["+",r.AdditiveBinaryOperator],["-",r.AdditiveBinaryOperator],["*",r.MultiplicativeBinaryOperator],["/",r.MultiplicativeBinaryOperator],["%",r.MultiplicativeBinaryOperator],["=",r.Equals]],c=new Map([["n","\n"],["t","\t"],["r","\r"],["b","\b"],["f","\f"],["v","\v"],["'","'"],['"','"'],["\\","\\"]]);function d(e,t={}){const s=[],d=function(e,t={}){return e.endsWith("\n")&&(e=e.slice(0,-1)),e=e.replace(/{#.*?#}/gs,"{##}"),t.lstrip_blocks&&(e=e.replace(/^[ \t]*({[#%])/gm,"$1")),t.trim_blocks&&(e=e.replace(/([#%]})\n/g,"$1")),e.replace(/{##}/g,"").replace(/-%}\s*/g,"%}").replace(/\s*{%-/g,"{%").replace(/-}}\s*/g,"}}").replace(/\s*{{-/g,"{{")}(e,t);let u=0;const _=e=>{let t="";for(;e(d[u]);)if("\\"!==d[u]){if(t+=d[u++],u>=d.length)throw new SyntaxError("Unexpected end of input")}else{if(++u,u>=d.length)throw new SyntaxError("Unexpected end of input");const e=d[u++],s=c.get(e);if(void 0===s)throw new SyntaxError(`Unexpected escaped character: ${e}`);t+=s}return t};e:for(;u<d.length;){const e=s.at(-1)?.type;if(void 0===e||e===r.CloseStatement||e===r.CloseExpression){let e="";for(;u<d.length&&("{"!==d[u]||"%"!==d[u+1]&&"{"!==d[u+1]);)e+=d[u++];if(e.length>0){s.push(new o(e,r.Text));continue}}_((e=>/\s/.test(e)));const t=d[u];if("-"===t||"+"===t){const e=s.at(-1)?.type;if(e===r.Text||void 0===e)throw new SyntaxError(`Unexpected character: ${t}`);switch(e){case r.Identifier:case r.NumericLiteral:case r.BooleanLiteral:case r.NullLiteral:case r.StringLiteral:case r.CloseParen:case r.CloseSquareBracket:break;default:{++u;const e=_(a);s.push(new o(`${t}${e}`,e.length>0?r.NumericLiteral:r.UnaryOperator));continue}}}for(const[e,t]of l){if(d.slice(u,u+e.length)===e){s.push(new o(e,t)),u+=e.length;continue e}}if("'"!==t&&'"'!==t)if(a(t)){const e=_(a);s.push(new o(e,r.NumericLiteral))}else{if(!i(t))throw new SyntaxError(`Unexpected character: ${t}`);{const e=_(i),t=Object.hasOwn(n,e)?n[e]:r.Identifier;t===r.In&&s.at(-1)?.type===r.Not?(s.pop(),s.push(new o("not in",r.NotIn))):s.push(new o(e,t))}}else{++u;const e=_((e=>e!==t));s.push(new o(e,r.StringLiteral)),++u}}return s}var u=class{type="Statement"},_=class extends u{constructor(e){super(),this.body=e}type="Program"},p=class extends u{constructor(e,t,s){super(),this.test=e,this.body=t,this.alternate=s}type="If"},m=class extends u{constructor(e,t,s,r){super(),this.loopvar=e,this.iterable=t,this.body=s,this.defaultBlock=r}type="For"},h=class extends u{constructor(e,t){super(),this.assignee=e,this.value=t}type="Set"},f=class extends u{constructor(e,t,s){super(),this.name=e,this.args=t,this.body=s}type="Macro"},g=class extends u{type="Expression"},w=class extends g{constructor(e,t,s){super(),this.object=e,this.property=t,this.computed=s}type="MemberExpression"},M=class extends g{constructor(e,t){super(),this.callee=e,this.args=t}type="CallExpression"},b=class extends g{constructor(e){super(),this.value=e}type="Identifier"},x=class extends g{constructor(e){super(),this.value=e}type="Literal"},y=class extends x{type="NumericLiteral"},k=class extends x{type="StringLiteral"},T=class extends x{type="BooleanLiteral"},v=class extends x{type="NullLiteral"},P=class extends x{type="ArrayLiteral"},F=class extends x{type="TupleLiteral"},C=class extends x{type="ObjectLiteral"},A=class extends g{constructor(e,t,s){super(),this.operator=e,this.left=t,this.right=s}type="BinaryExpression"},S=class extends g{constructor(e,t){super(),this.operand=e,this.filter=t}type="FilterExpression"},E=class extends g{constructor(e,t){super(),this.iterable=e,this.test=t}type="SelectExpression"},I=class extends g{constructor(e,t,s){super(),this.operand=e,this.negate=t,this.test=s}type="TestExpression"},L=class extends g{constructor(e,t){super(),this.operator=e,this.argument=t}type="UnaryExpression"},z=class extends g{constructor(e=void 0,t=void 0,s=void 0){super(),this.start=e,this.stop=t,this.step=s}type="SliceExpression"},j=class extends g{constructor(e,t){super(),this.key=e,this.value=t}type="KeywordArgumentExpression"};function N(e){const t=new _([]);let s=0;function n(t,r){const n=e[s++];if(!n||n.type!==t)throw new Error(`Parser Error: ${r}. ${n.type} !== ${t}.`);return n}function o(){switch(e[s].type){case r.Text:return new k(n(r.Text,"Expected text token").value);case r.OpenStatement:return function(){let t;switch(n(r.OpenStatement,"Expected opening statement token"),e[s].type){case r.Set:++s,t=l(),n(r.CloseStatement,"Expected closing statement token");break;case r.If:++s,t=c(),n(r.OpenStatement,"Expected {% token"),n(r.EndIf,"Expected endif token"),n(r.CloseStatement,"Expected %} token");break;case r.Macro:++s,t=function(){const e=q();if("Identifier"!==e.type)throw new SyntaxError("Expected identifier following macro statement");const t=V();n(r.CloseStatement,"Expected closing statement token");const s=[];for(;i(r.OpenStatement,r.EndMacro);)s.push(o());return new f(e,t,s)}(),n(r.OpenStatement,"Expected {% token"),n(r.EndMacro,"Expected endmacro token"),n(r.CloseStatement,"Expected %} token");break;case r.For:++s,t=function(){const e=d(!0);if(!(e instanceof b||e instanceof F))throw new SyntaxError(`Expected identifier/tuple for the loop variable, got ${e.type} instead`);n(r.In,"Expected `in` keyword following loop variable");const t=u();n(r.CloseStatement,"Expected closing statement token");const l=[];for(;i(r.OpenStatement,r.EndFor)&&i(r.OpenStatement,r.Else);)l.push(o());const c=[];if(a(r.OpenStatement,r.Else))for(++s,++s,n(r.CloseStatement,"Expected closing statement token");i(r.OpenStatement,r.EndFor);)c.push(o());return new m(e,t,l,c)}(),n(r.OpenStatement,"Expected {% token"),n(r.EndFor,"Expected endfor token"),n(r.CloseStatement,"Expected %} token");break;default:throw new SyntaxError(`Unknown statement type: ${e[s].type}`)}return t}();case r.OpenExpression:return function(){n(r.OpenExpression,"Expected opening expression token");const e=u();return n(r.CloseExpression,"Expected closing expression token"),e}();default:throw new SyntaxError(`Unexpected token type: ${e[s].type}`)}}function i(...t){return s+t.length<=e.length&&t.some(((t,r)=>t!==e[s+r].type))}function a(...t){return s+t.length<=e.length&&t.every(((t,r)=>t===e[s+r].type))}function l(){const e=u();if(a(r.Equals)){++s;const t=l();return new h(e,t)}return e}function c(){const t=u();n(r.CloseStatement,"Expected closing statement token");const i=[],l=[];for(;e[s]?.type!==r.OpenStatement||e[s+1]?.type!==r.ElseIf&&e[s+1]?.type!==r.Else&&e[s+1]?.type!==r.EndIf;)i.push(o());if(e[s]?.type===r.OpenStatement&&e[s+1]?.type!==r.EndIf)if(++s,a(r.ElseIf))n(r.ElseIf,"Expected elseif token"),l.push(c());else for(n(r.Else,"Expected else token"),n(r.CloseStatement,"Expected closing statement token");e[s]?.type!==r.OpenStatement||e[s+1]?.type!==r.EndIf;)l.push(o());return new p(t,i,l)}function d(e=!1){const t=e?q:u,n=[t()],o=a(r.Comma);for(;o&&(++s,n.push(t()),a(r.Comma)););return o?new F(n):n[0]}function u(){return function(){const e=g();if(a(r.If)){++s;const t=g();if(a(r.Else)){++s;const r=g();return new p(t,[e],[r])}return new E(e,t)}return e}()}function g(){let t=x();for(;a(r.Or);){const r=e[s];++s;const n=x();t=new A(r,t,n)}return t}function x(){let t=N();for(;a(r.And);){const r=e[s];++s;const n=N();t=new A(r,t,n)}return t}function N(){let t;for(;a(r.Not);){const r=e[s];++s;const n=N();t=new L(r,n)}return t??function(){let t=O();for(;a(r.ComparisonBinaryOperator)||a(r.In)||a(r.NotIn);){const r=e[s];++s;const n=O();t=new A(r,t,n)}return t}()}function O(){let t=G();for(;a(r.AdditiveBinaryOperator);){const r=e[s];++s;const n=G();t=new A(r,t,n)}return t}function B(e){let t=new M(e,V());return t=R(t),a(r.OpenParen)&&(t=B(t)),t}function V(){n(r.OpenParen,"Expected opening parenthesis for arguments list");const e=function(){const e=[];for(;!a(r.CloseParen);){let t=u();if(a(r.Equals)){if(++s,!(t instanceof b))throw new SyntaxError("Expected identifier for keyword argument");const e=u();t=new j(t,e)}e.push(t),a(r.Comma)&&++s}return e}();return n(r.CloseParen,"Expected closing parenthesis for arguments list"),e}function D(){const e=[];let t=!1;for(;!a(r.CloseSquareBracket);)a(r.Colon)?(e.push(void 0),++s,t=!0):(e.push(u()),a(r.Colon)&&(++s,t=!0));if(0===e.length)throw new SyntaxError("Expected at least one argument for member/slice expression");if(t){if(e.length>3)throw new SyntaxError("Expected 0-3 arguments for slice expression");return new z(...e)}return e[0]}function R(t){for(;a(r.Dot)||a(r.OpenSquareBracket);){const o=e[s];let i;++s;const a=o.type!==r.Dot;if(a)i=D(),n(r.CloseSquareBracket,"Expected closing square bracket");else if(i=q(),"Identifier"!==i.type)throw new SyntaxError("Expected identifier following dot operator");t=new w(t,i,a)}return t}function G(){let t=$();for(;a(r.MultiplicativeBinaryOperator);){const r=e[s];++s;const n=$();t=new A(r,t,n)}return t}function $(){let e=function(){let e=function(){const e=R(q());return a(r.OpenParen)?B(e):e}();for(;a(r.Pipe);){++s;let t=q();if(!(t instanceof b))throw new SyntaxError("Expected identifier for the filter");a(r.OpenParen)&&(t=B(t)),e=new S(e,t)}return e}();for(;a(r.Is);){++s;const t=a(r.Not);t&&++s;let n=q();if(n instanceof T?n=new b(n.value.toString()):n instanceof v&&(n=new b("none")),!(n instanceof b))throw new SyntaxError("Expected identifier for the test");e=new I(e,t,n)}return e}function q(){const t=e[s];switch(t.type){case r.NumericLiteral:return++s,new y(Number(t.value));case r.StringLiteral:return++s,new k(t.value);case r.BooleanLiteral:return++s,new T("true"===t.value.toLowerCase());case r.NullLiteral:return++s,new v(null);case r.Identifier:return++s,new b(t.value);case r.OpenParen:{++s;const t=d();if(e[s].type!==r.CloseParen)throw new SyntaxError(`Expected closing parenthesis, got ${e[s].type} instead`);return++s,t}case r.OpenSquareBracket:{++s;const e=[];for(;!a(r.CloseSquareBracket);)e.push(u()),a(r.Comma)&&++s;return++s,new P(e)}case r.OpenCurlyBracket:{++s;const e=new Map;for(;!a(r.CloseCurlyBracket);){const t=u();n(r.Colon,"Expected colon between key and value in object literal");const o=u();e.set(t,o),a(r.Comma)&&++s}return++s,new C(e)}default:throw new SyntaxError(`Unexpected token: ${t.type}`)}}for(;s<e.length;)t.body.push(o());return t}function O(e,t,s=1){void 0===t&&(t=e,e=0);const r=[];for(let n=e;n<t;n+=s)r.push(n);return r}function B(e,t,s,r=1){const n=Math.sign(r);n>=0?(t=(t??=0)<0?Math.max(e.length+t,0):Math.min(t,e.length),s=(s??=e.length)<0?Math.max(e.length+s,0):Math.min(s,e.length)):(t=(t??=e.length-1)<0?Math.max(e.length+t,-1):Math.min(t,e.length-1),s=(s??=-1)<-1?Math.max(e.length+s,-1):Math.min(s,e.length-1));const o=[];for(let i=t;n*i<n*s;i+=r)o.push(e[i]);return o}function V(e){return e.replace(/\b\w/g,(e=>e.toUpperCase()))}var D=class{type="RuntimeValue";value;builtins=new Map;constructor(e=void 0){this.value=e}__bool__(){return new $(!!this.value)}},R=class extends D{type="NumericValue"},G=class extends D{type="StringValue";builtins=new Map([["upper",new Q((()=>new G(this.value.toUpperCase())))],["lower",new Q((()=>new G(this.value.toLowerCase())))],["strip",new Q((()=>new G(this.value.trim())))],["title",new Q((()=>new G(V(this.value))))],["length",new R(this.value.length)],["rstrip",new Q((()=>new G(this.value.trimEnd())))],["lstrip",new Q((()=>new G(this.value.trimStart())))],["split",new Q((e=>{const t=e[0]??new H;if(!(t instanceof G||t instanceof H))throw new Error("sep argument must be a string or null");const s=e[1]??new R(-1);if(!(s instanceof R))throw new Error("maxsplit argument must be a number");let r=[];if(t instanceof H){const e=this.value.trimStart();for(const{0:t,index:n}of e.matchAll(/\S+/g)){if(-1!==s.value&&r.length>=s.value&&void 0!==n){r.push(t+e.slice(n+t.length));break}r.push(t)}}else{if(""===t.value)throw new Error("empty separator");r=this.value.split(t.value),-1!==s.value&&r.length>s.value&&r.push(r.splice(s.value).join(t.value))}return new W(r.map((e=>new G(e))))}))]])},$=class extends D{type="BooleanValue"},q=class extends D{type="ObjectValue";__bool__(){return new $(this.value.size>0)}builtins=new Map([["get",new Q((([e,t])=>{if(!(e instanceof G))throw new Error(`Object key must be a string: got ${e.type}`);return this.value.get(e.value)??t??new H}))],["items",new Q((()=>new W(Array.from(this.value.entries()).map((([e,t])=>new W([new G(e),t]))))))]])},U=class extends q{type="KeywordArgumentsValue"},W=class extends D{type="ArrayValue";builtins=new Map([["length",new R(this.value.length)]]);__bool__(){return new $(this.value.length>0)}},X=class extends W{type="TupleValue"},Q=class extends D{type="FunctionValue"},H=class extends D{type="NullValue"},J=class extends D{type="UndefinedValue"},Y=class{constructor(e){this.parent=e}variables=new Map([["namespace",new Q((e=>{if(0===e.length)return new q(new Map);if(1!==e.length||!(e[0]instanceof q))throw new Error("`namespace` expects either zero arguments or a single object argument");return e[0]}))]]);tests=new Map([["boolean",e=>"BooleanValue"===e.type],["callable",e=>e instanceof Q],["odd",e=>{if("NumericValue"!==e.type)throw new Error(`Cannot apply test "odd" to type: ${e.type}`);return e.value%2!=0}],["even",e=>{if("NumericValue"!==e.type)throw new Error(`Cannot apply test "even" to type: ${e.type}`);return e.value%2==0}],["false",e=>"BooleanValue"===e.type&&!e.value],["true",e=>"BooleanValue"===e.type&&e.value],["none",e=>"NullValue"===e.type],["string",e=>"StringValue"===e.type],["number",e=>"NumericValue"===e.type],["integer",e=>"NumericValue"===e.type&&Number.isInteger(e.value)],["iterable",e=>"ArrayValue"===e.type||"StringValue"===e.type],["mapping",e=>"ObjectValue"===e.type],["lower",e=>{const t=e.value;return"StringValue"===e.type&&t===t.toLowerCase()}],["upper",e=>{const t=e.value;return"StringValue"===e.type&&t===t.toUpperCase()}],["none",e=>"NullValue"===e.type],["defined",e=>"UndefinedValue"!==e.type],["undefined",e=>"UndefinedValue"===e.type],["equalto",(e,t)=>e.value===t.value],["eq",(e,t)=>e.value===t.value]]);set(e,t){return this.declareVariable(e,Z(t))}declareVariable(e,t){if(this.variables.has(e))throw new SyntaxError(`Variable already declared: ${e}`);return this.variables.set(e,t),t}setVariable(e,t){return this.variables.set(e,t),t}resolve(e){if(this.variables.has(e))return this;if(this.parent)return this.parent.resolve(e);throw new Error(`Unknown variable: ${e}`)}lookupVariable(e){try{return this.resolve(e).variables.get(e)??new J}catch{return new J}}},K=class{global;constructor(e){this.global=e??new Y}run(e){return this.evaluate(e,this.global)}evaluateBinaryExpression(e,t){const s=this.evaluate(e.left,t);switch(e.operator.value){case"and":return s.__bool__().value?this.evaluate(e.right,t):s;case"or":return s.__bool__().value?s:this.evaluate(e.right,t)}const r=this.evaluate(e.right,t);switch(e.operator.value){case"==":return new $(s.value==r.value);case"!=":return new $(s.value!=r.value)}if(s instanceof J||r instanceof J)throw new Error("Cannot perform operation on undefined values");if(s instanceof H||r instanceof H)throw new Error("Cannot perform operation on null values");if(s instanceof R&&r instanceof R)switch(e.operator.value){case"+":return new R(s.value+r.value);case"-":return new R(s.value-r.value);case"*":return new R(s.value*r.value);case"/":return new R(s.value/r.value);case"%":return new R(s.value%r.value);case"<":return new $(s.value<r.value);case">":return new $(s.value>r.value);case">=":return new $(s.value>=r.value);case"<=":return new $(s.value<=r.value)}else if(s instanceof W&&r instanceof W){if("+"===e.operator.value)return new W(s.value.concat(r.value))}else if(r instanceof W){const t=void 0!==r.value.find((e=>e.value===s.value));switch(e.operator.value){case"in":return new $(t);case"not in":return new $(!t)}}if((s instanceof G||r instanceof G)&&"+"===e.operator.value)return new G(s.value.toString()+r.value.toString());if(s instanceof G&&r instanceof G)switch(e.operator.value){case"in":return new $(r.value.includes(s.value));case"not in":return new $(!r.value.includes(s.value))}if(s instanceof G&&r instanceof q)switch(e.operator.value){case"in":return new $(r.value.has(s.value));case"not in":return new $(!r.value.has(s.value))}throw new SyntaxError(`Unknown operator "${e.operator.value}" between ${s.type} and ${r.type}`)}evaluateArguments(e,t){const s=[],r=new Map;for(const n of e)if("KeywordArgumentExpression"===n.type){const e=n;r.set(e.key.value,this.evaluate(e.value,t))}else{if(r.size>0)throw new Error("Positional arguments must come before keyword arguments");s.push(this.evaluate(n,t))}return[s,r]}evaluateFilterExpression(e,t){const s=this.evaluate(e.operand,t);if("Identifier"===e.filter.type){const t=e.filter;if("tojson"===t.value)return new G(ee(s));if(s instanceof W)switch(t.value){case"list":return s;case"first":return s.value[0];case"last":return s.value[s.value.length-1];case"length":return new R(s.value.length);case"reverse":return new W(s.value.reverse());case"sort":return new W(s.value.sort(((e,t)=>{if(e.type!==t.type)throw new Error(`Cannot compare different types: ${e.type} and ${t.type}`);switch(e.type){case"NumericValue":return e.value-t.value;case"StringValue":return e.value.localeCompare(t.value);default:throw new Error(`Cannot compare type: ${e.type}`)}})));case"join":return new G(s.value.map((e=>e.value)).join(""));default:throw new Error(`Unknown ArrayValue filter: ${t.value}`)}else if(s instanceof G)switch(t.value){case"length":return new R(s.value.length);case"upper":return new G(s.value.toUpperCase());case"lower":return new G(s.value.toLowerCase());case"title":return new G(V(s.value));case"capitalize":return new G(s.value.charAt(0).toUpperCase()+s.value.slice(1));case"trim":return new G(s.value.trim());case"indent":return new G(s.value.split("\n").map(((e,t)=>0===t||0===e.length?e:"    "+e)).join("\n"));case"join":case"string":return s;default:throw new Error(`Unknown StringValue filter: ${t.value}`)}else{if(s instanceof R){if("abs"===t.value)return new R(Math.abs(s.value));throw new Error(`Unknown NumericValue filter: ${t.value}`)}if(s instanceof q)switch(t.value){case"items":return new W(Array.from(s.value.entries()).map((([e,t])=>new W([new G(e),t]))));case"length":return new R(s.value.size);default:throw new Error(`Unknown ObjectValue filter: ${t.value}`)}}throw new Error(`Cannot apply filter "${t.value}" to type: ${s.type}`)}if("CallExpression"===e.filter.type){const r=e.filter;if("Identifier"!==r.callee.type)throw new Error(`Unknown filter: ${r.callee.type}`);const n=r.callee.value;if("tojson"===n){const[,e]=this.evaluateArguments(r.args,t),n=e.get("indent")??new H;if(!(n instanceof R||n instanceof H))throw new Error("If set, indent must be a number");return new G(ee(s,n.value))}if("join"===n){let e;if(s instanceof G)e=Array.from(s.value);else{if(!(s instanceof W))throw new Error(`Cannot apply filter "${n}" to type: ${s.type}`);e=s.value.map((e=>e.value))}const[o,i]=this.evaluateArguments(r.args,t),a=o.at(0)??i.get("separator")??new G("");if(!(a instanceof G))throw new Error("separator must be a string");return new G(e.join(a.value))}if(s instanceof W){switch(n){case"selectattr":case"rejectattr":{const e="selectattr"===n;if(s.value.some((e=>!(e instanceof q))))throw new Error(`\`${n}\` can only be applied to array of objects`);if(r.args.some((e=>"StringLiteral"!==e.type)))throw new Error(`arguments of \`${n}\` must be strings`);const[o,i,a]=r.args.map((e=>this.evaluate(e,t)));let l;if(i){const e=t.tests.get(i.value);if(!e)throw new Error(`Unknown test: ${i.value}`);l=e}else l=(...e)=>e[0].__bool__().value;const c=s.value.filter((t=>{const s=t.value.get(o.value),r=!!s&&l(s,a);return e?r:!r}));return new W(c)}case"map":{const[,e]=this.evaluateArguments(r.args,t);if(e.has("attribute")){const t=e.get("attribute");if(!(t instanceof G))throw new Error("attribute must be a string");const r=e.get("default"),n=s.value.map((e=>{if(!(e instanceof q))throw new Error("items in map must be an object");return e.value.get(t.value)??r??new J}));return new W(n)}throw new Error("`map` expressions without `attribute` set are not currently supported.")}}throw new Error(`Unknown ArrayValue filter: ${n}`)}if(s instanceof G){if("indent"===n){const[e,n]=this.evaluateArguments(r.args,t),o=e.at(0)??n.get("width")??new R(4);if(!(o instanceof R))throw new Error("width must be a number");const i=e.at(1)??n.get("first")??new $(!1),a=e.at(2)??n.get("blank")??new $(!1),l=s.value.split("\n"),c=" ".repeat(o.value),d=l.map(((e,t)=>!i.value&&0===t||!a.value&&0===e.length?e:c+e));return new G(d.join("\n"))}throw new Error(`Unknown StringValue filter: ${n}`)}throw new Error(`Cannot apply filter "${n}" to type: ${s.type}`)}throw new Error(`Unknown filter: ${e.filter.type}`)}evaluateTestExpression(e,t){const s=this.evaluate(e.operand,t),r=t.tests.get(e.test.value);if(!r)throw new Error(`Unknown test: ${e.test.value}`);const n=r(s);return new $(e.negate?!n:n)}evaluateUnaryExpression(e,t){const s=this.evaluate(e.argument,t);if("not"===e.operator.value)return new $(!s.value);throw new SyntaxError(`Unknown operator: ${e.operator.value}`)}evalProgram(e,t){return this.evaluateBlock(e.body,t)}evaluateBlock(e,t){let s="";for(const r of e){const e=this.evaluate(r,t);"NullValue"!==e.type&&"UndefinedValue"!==e.type&&(s+=e.value)}return new G(s)}evaluateIdentifier(e,t){return t.lookupVariable(e.value)}evaluateCallExpression(e,t){const[s,r]=this.evaluateArguments(e.args,t);r.size>0&&s.push(new U(r));const n=this.evaluate(e.callee,t);if("FunctionValue"!==n.type)throw new Error(`Cannot call something that is not a function: got ${n.type}`);return n.value(s,t)}evaluateSliceExpression(e,t,s){if(!(e instanceof W||e instanceof G))throw new Error("Slice object must be an array or string");const r=this.evaluate(t.start,s),n=this.evaluate(t.stop,s),o=this.evaluate(t.step,s);if(!(r instanceof R||r instanceof J))throw new Error("Slice start must be numeric or undefined");if(!(n instanceof R||n instanceof J))throw new Error("Slice stop must be numeric or undefined");if(!(o instanceof R||o instanceof J))throw new Error("Slice step must be numeric or undefined");return e instanceof W?new W(B(e.value,r.value,n.value,o.value)):new G(B(Array.from(e.value),r.value,n.value,o.value).join(""))}evaluateMemberExpression(e,t){const s=this.evaluate(e.object,t);let r,n;if(e.computed){if("SliceExpression"===e.property.type)return this.evaluateSliceExpression(s,e.property,t);r=this.evaluate(e.property,t)}else r=new G(e.property.value);if(s instanceof q){if(!(r instanceof G))throw new Error(`Cannot access property with non-string: got ${r.type}`);n=s.value.get(r.value)??s.builtins.get(r.value)}else if(s instanceof W||s instanceof G)if(r instanceof R)n=s.value.at(r.value),s instanceof G&&(n=new G(s.value.at(r.value)));else{if(!(r instanceof G))throw new Error(`Cannot access property with non-string/non-number: got ${r.type}`);n=s.builtins.get(r.value)}else{if(!(r instanceof G))throw new Error(`Cannot access property with non-string: got ${r.type}`);n=s.builtins.get(r.value)}return n instanceof D?n:new J}evaluateSet(e,t){const s=this.evaluate(e.value,t);if("Identifier"===e.assignee.type){const r=e.assignee.value;t.setVariable(r,s)}else{if("MemberExpression"!==e.assignee.type)throw new Error(`Invalid LHS inside assignment expression: ${JSON.stringify(e.assignee)}`);{const r=e.assignee,n=this.evaluate(r.object,t);if(!(n instanceof q))throw new Error("Cannot assign to member of non-object");if("Identifier"!==r.property.type)throw new Error("Cannot assign to member with non-identifier property");n.value.set(r.property.value,s)}}return new H}evaluateIf(e,t){const s=this.evaluate(e.test,t);return this.evaluateBlock(s.__bool__().value?e.body:e.alternate,t)}evaluateFor(e,t){const s=new Y(t);let r,n;if("SelectExpression"===e.iterable.type){const t=e.iterable;n=this.evaluate(t.iterable,s),r=t.test}else n=this.evaluate(e.iterable,s);if(!(n instanceof W))throw new Error(`Expected iterable type in for loop: got ${n.type}`);const o=[],i=[];for(let t=0;t<n.value.length;++t){const a=new Y(s),l=n.value[t];let c;if("Identifier"===e.loopvar.type)c=t=>t.setVariable(e.loopvar.value,l);else{if("TupleLiteral"!==e.loopvar.type)throw new Error(`Invalid loop variable(s): ${e.loopvar.type}`);{const t=e.loopvar;if("ArrayValue"!==l.type)throw new Error(`Cannot unpack non-iterable type: ${l.type}`);const s=l;if(t.value.length!==s.value.length)throw new Error(`Too ${t.value.length>s.value.length?"few":"many"} items to unpack`);c=e=>{for(let r=0;r<t.value.length;++r){if("Identifier"!==t.value[r].type)throw new Error(`Cannot unpack non-identifier type: ${t.value[r].type}`);e.setVariable(t.value[r].value,s.value[r])}}}}if(r){c(a);if(!this.evaluate(r,a).__bool__().value)continue}o.push(l),i.push(c)}let a="",l=!0;for(let t=0;t<o.length;++t){const r=new Map([["index",new R(t+1)],["index0",new R(t)],["revindex",new R(o.length-t)],["revindex0",new R(o.length-t-1)],["first",new $(0===t)],["last",new $(t===o.length-1)],["length",new R(o.length)],["previtem",t>0?o[t-1]:new J],["nextitem",t<o.length-1?o[t+1]:new J]]);s.setVariable("loop",new q(r)),i[t](s);a+=this.evaluateBlock(e.body,s).value,l=!1}if(l){a+=this.evaluateBlock(e.defaultBlock,s).value}return new G(a)}evaluateMacro(e,t){return t.setVariable(e.name.value,new Q(((t,s)=>{const r=new Y(s);let n;t=t.slice(),"KeywordArgumentsValue"===t.at(-1)?.type&&(n=t.pop());for(let s=0;s<e.args.length;++s){const o=e.args[s],i=t[s];if("Identifier"===o.type){const e=o;if(!i)throw new Error(`Missing positional argument: ${e.value}`);r.setVariable(e.value,i)}else{if("KeywordArgumentExpression"!==o.type)throw new Error(`Unknown argument type: ${o.type}`);{const e=o,t=i??n?.value.get(e.key.value)??this.evaluate(e.value,r);r.setVariable(e.key.value,t)}}}return this.evaluateBlock(e.body,r)}))),new H}evaluate(e,t){if(void 0===e)return new J;switch(e.type){case"Program":return this.evalProgram(e,t);case"Set":return this.evaluateSet(e,t);case"If":return this.evaluateIf(e,t);case"For":return this.evaluateFor(e,t);case"Macro":return this.evaluateMacro(e,t);case"NumericLiteral":return new R(Number(e.value));case"StringLiteral":return new G(e.value);case"BooleanLiteral":return new $(e.value);case"NullLiteral":return new H(e.value);case"ArrayLiteral":return new W(e.value.map((e=>this.evaluate(e,t))));case"TupleLiteral":return new X(e.value.map((e=>this.evaluate(e,t))));case"ObjectLiteral":{const s=new Map;for(const[r,n]of e.value){const e=this.evaluate(r,t);if(!(e instanceof G))throw new Error(`Object keys must be strings: got ${e.type}`);s.set(e.value,this.evaluate(n,t))}return new q(s)}case"Identifier":return this.evaluateIdentifier(e,t);case"CallExpression":return this.evaluateCallExpression(e,t);case"MemberExpression":return this.evaluateMemberExpression(e,t);case"UnaryExpression":return this.evaluateUnaryExpression(e,t);case"BinaryExpression":return this.evaluateBinaryExpression(e,t);case"FilterExpression":return this.evaluateFilterExpression(e,t);case"TestExpression":return this.evaluateTestExpression(e,t);default:throw new SyntaxError(`Unknown node type: ${e.type}`)}}};function Z(e){switch(typeof e){case"number":return new R(e);case"string":return new G(e);case"boolean":return new $(e);case"undefined":return new J;case"object":return null===e?new H:Array.isArray(e)?new W(e.map(Z)):new q(new Map(Object.entries(e).map((([e,t])=>[e,Z(t)]))));case"function":return new Q(((t,s)=>Z(e(...t.map((e=>e.value)))??null)));default:throw new Error(`Cannot convert to runtime value: ${e}`)}}function ee(e,t,s){const r=s??0;switch(e.type){case"NullValue":case"UndefinedValue":return"null";case"NumericValue":case"StringValue":case"BooleanValue":return JSON.stringify(e.value);case"ArrayValue":case"ObjectValue":{const s=t?" ".repeat(t):"",n="\n"+s.repeat(r),o=n+s;if("ArrayValue"===e.type){const s=e.value.map((e=>ee(e,t,r+1)));return t?`[${o}${s.join(`,${o}`)}${n}]`:`[${s.join(", ")}]`}{const s=Array.from(e.value.entries()).map((([e,s])=>{const n=`"${e}": ${ee(s,t,r+1)}`;return t?`${o}${n}`:n}));return t?`{${s.join(",")}${n}}`:`{${s.join(", ")}}`}}default:throw new Error(`Cannot convert to JSON: ${e.type}`)}}var te=class{parsed;constructor(e){const t=d(e,{lstrip_blocks:!0,trim_blocks:!0});this.parsed=N(t)}render(e){const t=new Y;if(t.set("false",!1),t.set("true",!0),t.set("raise_exception",(e=>{throw new Error(e)})),t.set("range",O),e)for(const[s,r]of Object.entries(e))t.set(s,r);return new K(t).run(this.parsed).value}}},"./node_modules/onnxruntime-common/dist/esm/backend-impl.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{registerBackend:()=>o,resolveBackendAndExecutionProviders:()=>a});const r=new Map,n=[],o=(e,t,s)=>{if(!t||"function"!=typeof t.init||"function"!=typeof t.createInferenceSessionHandler)throw new TypeError("not a valid backend");{const o=r.get(e);if(void 0===o)r.set(e,{backend:t,priority:s});else{if(o.priority>s)return;if(o.priority===s&&o.backend!==t)throw new Error(`cannot register backend "${e}" using priority ${s}`)}if(s>=0){const t=n.indexOf(e);-1!==t&&n.splice(t,1);for(let t=0;t<n.length;t++)if(r.get(n[t]).priority<=s)return void n.splice(t,0,e);n.push(e)}}},i=async e=>{const t=r.get(e);if(!t)return"backend not found.";if(t.initialized)return t.backend;if(t.aborted)return t.error;{const s=!!t.initPromise;try{return s||(t.initPromise=t.backend.init(e)),await t.initPromise,t.initialized=!0,t.backend}catch(e){return s||(t.error=`${e}`,t.aborted=!0),t.error}finally{delete t.initPromise}}},a=async e=>{const t=e.executionProviders||[],s=t.map((e=>"string"==typeof e?e:e.name)),r=0===s.length?n:s;let o;const a=[],l=new Set;for(const e of r){const t=await i(e);"string"==typeof t?a.push({name:e,err:t}):(o||(o=t),o===t&&l.add(e))}if(!o)throw new Error(`no available backend found. ERR: ${a.map((e=>`[${e.name}] ${e.err}`)).join(", ")}`);for(const{name:e,err:t}of a)s.includes(e)&&console.warn(`removing requested execution provider "${e}" from session options because it is not available: ${t}`);const c=t.filter((e=>l.has("string"==typeof e?e:e.name)));return[o,new Proxy(e,{get:(e,t)=>"executionProviders"===t?c:Reflect.get(e,t)})]}},"./node_modules/onnxruntime-common/dist/esm/backend.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{registerBackend:()=>r.registerBackend});var r=s("./node_modules/onnxruntime-common/dist/esm/backend-impl.js")},"./node_modules/onnxruntime-common/dist/esm/env-impl.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{env:()=>o});var r=s("./node_modules/onnxruntime-common/dist/esm/version.js");let n="warning";const o={wasm:{},webgl:{},webgpu:{},versions:{common:r.version},set logLevel(e){if(void 0!==e){if("string"!=typeof e||-1===["verbose","info","warning","error","fatal"].indexOf(e))throw new Error(`Unsupported logging level: ${e}`);n=e}},get logLevel(){return n}};Object.defineProperty(o,"logLevel",{enumerable:!0})},"./node_modules/onnxruntime-common/dist/esm/env.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{env:()=>r});const r=s("./node_modules/onnxruntime-common/dist/esm/env-impl.js").env},"./node_modules/onnxruntime-common/dist/esm/index.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{InferenceSession:()=>o.InferenceSession,TRACE:()=>a.TRACE,TRACE_FUNC_BEGIN:()=>a.TRACE_FUNC_BEGIN,TRACE_FUNC_END:()=>a.TRACE_FUNC_END,Tensor:()=>i.Tensor,TrainingSession:()=>l.TrainingSession,env:()=>n.env,registerBackend:()=>r.registerBackend});var r=s("./node_modules/onnxruntime-common/dist/esm/backend.js"),n=s("./node_modules/onnxruntime-common/dist/esm/env.js"),o=s("./node_modules/onnxruntime-common/dist/esm/inference-session.js"),i=s("./node_modules/onnxruntime-common/dist/esm/tensor.js"),a=(s("./node_modules/onnxruntime-common/dist/esm/tensor-conversion.js"),s("./node_modules/onnxruntime-common/dist/esm/tensor-factory.js"),s("./node_modules/onnxruntime-common/dist/esm/trace.js")),l=(s("./node_modules/onnxruntime-common/dist/esm/onnx-model.js"),s("./node_modules/onnxruntime-common/dist/esm/onnx-value.js"),s("./node_modules/onnxruntime-common/dist/esm/training-session.js"))},"./node_modules/onnxruntime-common/dist/esm/inference-session-impl.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{InferenceSession:()=>i});var r=s("./node_modules/onnxruntime-common/dist/esm/backend-impl.js"),n=s("./node_modules/onnxruntime-common/dist/esm/tensor.js"),o=s("./node_modules/onnxruntime-common/dist/esm/trace.js");class i{constructor(e){this.handler=e}async run(e,t,s){(0,o.TRACE_FUNC_BEGIN)();const r={};let i={};if("object"!=typeof e||null===e||e instanceof n.Tensor||Array.isArray(e))throw new TypeError("'feeds' must be an object that use input names as keys and OnnxValue as corresponding values.");let a=!0;if("object"==typeof t){if(null===t)throw new TypeError("Unexpected argument[1]: cannot be null.");if(t instanceof n.Tensor)throw new TypeError("'fetches' cannot be a Tensor");if(Array.isArray(t)){if(0===t.length)throw new TypeError("'fetches' cannot be an empty array.");a=!1;for(const e of t){if("string"!=typeof e)throw new TypeError("'fetches' must be a string array or an object.");if(-1===this.outputNames.indexOf(e))throw new RangeError(`'fetches' contains invalid output name: ${e}.`);r[e]=null}if("object"==typeof s&&null!==s)i=s;else if(void 0!==s)throw new TypeError("'options' must be an object.")}else{let e=!1;const o=Object.getOwnPropertyNames(t);for(const s of this.outputNames)if(-1!==o.indexOf(s)){const o=t[s];(null===o||o instanceof n.Tensor)&&(e=!0,a=!1,r[s]=o)}if(e){if("object"==typeof s&&null!==s)i=s;else if(void 0!==s)throw new TypeError("'options' must be an object.")}else i=t}}else if(void 0!==t)throw new TypeError("Unexpected argument[1]: must be 'fetches' or 'options'.");for(const t of this.inputNames)if(void 0===e[t])throw new Error(`input '${t}' is missing in 'feeds'.`);if(a)for(const e of this.outputNames)r[e]=null;const l=await this.handler.run(e,r,i),c={};for(const e in l)if(Object.hasOwnProperty.call(l,e)){const t=l[e];t instanceof n.Tensor?c[e]=t:c[e]=new n.Tensor(t.type,t.data,t.dims)}return(0,o.TRACE_FUNC_END)(),c}async release(){return this.handler.dispose()}static async create(e,t,s,n){let a;(0,o.TRACE_FUNC_BEGIN)();let l={};if("string"==typeof e){if(a=e,"object"==typeof t&&null!==t)l=t;else if(void 0!==t)throw new TypeError("'options' must be an object.")}else if(e instanceof Uint8Array){if(a=e,"object"==typeof t&&null!==t)l=t;else if(void 0!==t)throw new TypeError("'options' must be an object.")}else{if(!(e instanceof ArrayBuffer||"undefined"!=typeof SharedArrayBuffer&&e instanceof SharedArrayBuffer))throw new TypeError("Unexpected argument[0]: must be 'path' or 'buffer'.");{const r=e;let o=0,i=e.byteLength;if("object"==typeof t&&null!==t)l=t;else if("number"==typeof t){if(o=t,!Number.isSafeInteger(o))throw new RangeError("'byteOffset' must be an integer.");if(o<0||o>=r.byteLength)throw new RangeError(`'byteOffset' is out of range [0, ${r.byteLength}).`);if(i=e.byteLength-o,"number"==typeof s){if(i=s,!Number.isSafeInteger(i))throw new RangeError("'byteLength' must be an integer.");if(i<=0||o+i>r.byteLength)throw new RangeError(`'byteLength' is out of range (0, ${r.byteLength-o}].`);if("object"==typeof n&&null!==n)l=n;else if(void 0!==n)throw new TypeError("'options' must be an object.")}else if(void 0!==s)throw new TypeError("'byteLength' must be a number.")}else if(void 0!==t)throw new TypeError("'options' must be an object.");a=new Uint8Array(r,o,i)}}const[c,d]=await(0,r.resolveBackendAndExecutionProviders)(l),u=await c.createInferenceSessionHandler(a,d);return(0,o.TRACE_FUNC_END)(),new i(u)}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":(e,t,s)=>{"use strict";s.r(t),s.d(t,{InferenceSession:()=>r});const r=s("./node_modules/onnxruntime-common/dist/esm/inference-session-impl.js").InferenceSession},"./node_modules/onnxruntime-common/dist/esm/onnx-model.js":(e,t,s)=>{"use strict";s.r(t)},"./node_modules/onnxruntime-common/dist/esm/onnx-value.js":(e,t,s)=>{"use strict";s.r(t)},"./node_modules/onnxruntime-common/dist/esm/tensor-conversion-impl.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{tensorToDataURL:()=>r,tensorToImageData:()=>n});const r=(e,t)=>{const s="undefined"!=typeof document?document.createElement("canvas"):new OffscreenCanvas(1,1);s.width=e.dims[3],s.height=e.dims[2];const r=s.getContext("2d");if(null!=r){let n,o;void 0!==t?.tensorLayout&&"NHWC"===t.tensorLayout?(n=e.dims[2],o=e.dims[3]):(n=e.dims[3],o=e.dims[2]);const i=void 0!==t?.format?t.format:"RGB",a=t?.norm;let l,c;void 0===a||void 0===a.mean?l=[255,255,255,255]:"number"==typeof a.mean?l=[a.mean,a.mean,a.mean,a.mean]:(l=[a.mean[0],a.mean[1],a.mean[2],0],void 0!==a.mean[3]&&(l[3]=a.mean[3])),void 0===a||void 0===a.bias?c=[0,0,0,0]:"number"==typeof a.bias?c=[a.bias,a.bias,a.bias,a.bias]:(c=[a.bias[0],a.bias[1],a.bias[2],0],void 0!==a.bias[3]&&(c[3]=a.bias[3]));const d=o*n;let u=0,_=d,p=2*d,m=-1;"RGBA"===i?(u=0,_=d,p=2*d,m=3*d):"RGB"===i?(u=0,_=d,p=2*d):"RBG"===i&&(u=0,p=d,_=2*d);for(let t=0;t<o;t++)for(let s=0;s<n;s++){const n=(e.data[u++]-c[0])*l[0],o=(e.data[_++]-c[1])*l[1],i=(e.data[p++]-c[2])*l[2],a=-1===m?255:(e.data[m++]-c[3])*l[3];r.fillStyle="rgba("+n+","+o+","+i+","+a+")",r.fillRect(s,t,1,1)}if("toDataURL"in s)return s.toDataURL();throw new Error("toDataURL is not supported")}throw new Error("Can not access image data")},n=(e,t)=>{const s="undefined"!=typeof document?document.createElement("canvas").getContext("2d"):new OffscreenCanvas(1,1).getContext("2d");let r;if(null==s)throw new Error("Can not access image data");{let n,o,i;void 0!==t?.tensorLayout&&"NHWC"===t.tensorLayout?(n=e.dims[2],o=e.dims[1],i=e.dims[3]):(n=e.dims[3],o=e.dims[2],i=e.dims[1]);const a=void 0!==t&&void 0!==t.format?t.format:"RGB",l=t?.norm;let c,d;void 0===l||void 0===l.mean?c=[255,255,255,255]:"number"==typeof l.mean?c=[l.mean,l.mean,l.mean,l.mean]:(c=[l.mean[0],l.mean[1],l.mean[2],255],void 0!==l.mean[3]&&(c[3]=l.mean[3])),void 0===l||void 0===l.bias?d=[0,0,0,0]:"number"==typeof l.bias?d=[l.bias,l.bias,l.bias,l.bias]:(d=[l.bias[0],l.bias[1],l.bias[2],0],void 0!==l.bias[3]&&(d[3]=l.bias[3]));const u=o*n;if(void 0!==t&&(void 0!==t.format&&4===i&&"RGBA"!==t.format||3===i&&"RGB"!==t.format&&"BGR"!==t.format))throw new Error("Tensor format doesn't match input tensor dims");const _=4;let p=0,m=1,h=2,f=3,g=0,w=u,M=2*u,b=-1;"RGBA"===a?(g=0,w=u,M=2*u,b=3*u):"RGB"===a?(g=0,w=u,M=2*u):"RBG"===a&&(g=0,M=u,w=2*u),r=s.createImageData(n,o);for(let t=0;t<o*n;p+=_,m+=_,h+=_,f+=_,t++)r.data[p]=(e.data[g++]-d[0])*c[0],r.data[m]=(e.data[w++]-d[1])*c[1],r.data[h]=(e.data[M++]-d[2])*c[2],r.data[f]=-1===b?255:(e.data[b++]-d[3])*c[3]}return r}},"./node_modules/onnxruntime-common/dist/esm/tensor-conversion.js":(e,t,s)=>{"use strict";s.r(t)},"./node_modules/onnxruntime-common/dist/esm/tensor-factory-impl.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{bufferToTensor:()=>n,tensorFromGpuBuffer:()=>a,tensorFromImage:()=>o,tensorFromMLTensor:()=>l,tensorFromPinnedBuffer:()=>c,tensorFromTexture:()=>i});var r=s("./node_modules/onnxruntime-common/dist/esm/tensor-impl.js");const n=(e,t)=>{if(void 0===e)throw new Error("Image buffer must be defined");if(void 0===t.height||void 0===t.width)throw new Error("Image height and width must be defined");if("NHWC"===t.tensorLayout)throw new Error("NHWC Tensor layout is not supported yet");const{height:s,width:n}=t,o=t.norm??{mean:255,bias:0};let i,a;i="number"==typeof o.mean?[o.mean,o.mean,o.mean,o.mean]:[o.mean[0],o.mean[1],o.mean[2],o.mean[3]??255],a="number"==typeof o.bias?[o.bias,o.bias,o.bias,o.bias]:[o.bias[0],o.bias[1],o.bias[2],o.bias[3]??0];const l=void 0!==t.format?t.format:"RGBA",c=void 0!==t.tensorFormat&&void 0!==t.tensorFormat?t.tensorFormat:"RGB",d=s*n,u="RGBA"===c?new Float32Array(4*d):new Float32Array(3*d);let _=4,p=0,m=1,h=2,f=3,g=0,w=d,M=2*d,b=-1;"RGB"===l&&(_=3,p=0,m=1,h=2,f=-1),"RGBA"===c?b=3*d:"RBG"===c?(g=0,M=d,w=2*d):"BGR"===c&&(M=0,w=d,g=2*d);for(let t=0;t<d;t++,p+=_,h+=_,m+=_,f+=_)u[g++]=(e[p]+a[0])/i[0],u[w++]=(e[m]+a[1])/i[1],u[M++]=(e[h]+a[2])/i[2],-1!==b&&-1!==f&&(u[b++]=(e[f]+a[3])/i[3]);return"RGBA"===c?new r.Tensor("float32",u,[1,4,s,n]):new r.Tensor("float32",u,[1,3,s,n])},o=async(e,t)=>{const s="undefined"!=typeof HTMLImageElement&&e instanceof HTMLImageElement,r="undefined"!=typeof ImageData&&e instanceof ImageData,o="undefined"!=typeof ImageBitmap&&e instanceof ImageBitmap,i="string"==typeof e;let a,l=t??{};const c=()=>{if("undefined"!=typeof document)return document.createElement("canvas");if("undefined"!=typeof OffscreenCanvas)return new OffscreenCanvas(1,1);throw new Error("Canvas is not supported")},d=e=>"undefined"!=typeof HTMLCanvasElement&&e instanceof HTMLCanvasElement||e instanceof OffscreenCanvas?e.getContext("2d"):null;if(s){const s=c();s.width=e.width,s.height=e.height;const r=d(s);if(null==r)throw new Error("Can not access image data");{let s=e.height,n=e.width;if(void 0!==t&&void 0!==t.resizedHeight&&void 0!==t.resizedWidth&&(s=t.resizedHeight,n=t.resizedWidth),void 0!==t){if(l=t,void 0!==t.tensorFormat)throw new Error("Image input config format must be RGBA for HTMLImageElement");l.tensorFormat="RGBA",l.height=s,l.width=n}else l.tensorFormat="RGBA",l.height=s,l.width=n;r.drawImage(e,0,0),a=r.getImageData(0,0,n,s).data}}else{if(!r){if(o){if(void 0===t)throw new Error("Please provide image config with format for Imagebitmap");const s=c();s.width=e.width,s.height=e.height;const r=d(s);if(null!=r){const t=e.height,s=e.width;return r.drawImage(e,0,0,s,t),a=r.getImageData(0,0,s,t).data,l.height=t,l.width=s,n(a,l)}throw new Error("Can not access image data")}if(i)return new Promise(((t,s)=>{const r=c(),o=d(r);if(!e||!o)return s();const i=new Image;i.crossOrigin="Anonymous",i.src=e,i.onload=()=>{r.width=i.width,r.height=i.height,o.drawImage(i,0,0,r.width,r.height);const e=o.getImageData(0,0,r.width,r.height);l.height=r.height,l.width=r.width,t(n(e.data,l))}}));throw new Error("Input data provided is not supported - aborted tensor creation")}{let s,r;if(void 0!==t&&void 0!==t.resizedWidth&&void 0!==t.resizedHeight?(s=t.resizedHeight,r=t.resizedWidth):(s=e.height,r=e.width),void 0!==t&&(l=t),l.format="RGBA",l.height=s,l.width=r,void 0!==t){const t=c();t.width=r,t.height=s;const n=d(t);if(null==n)throw new Error("Can not access image data");n.putImageData(e,0,0),a=n.getImageData(0,0,r,s).data}else a=e.data}}if(void 0!==a)return n(a,l);throw new Error("Input data provided is not supported - aborted tensor creation")},i=(e,t)=>{const{width:s,height:n,download:o,dispose:i}=t,a=[1,n,s,4];return new r.Tensor({location:"texture",type:"float32",texture:e,dims:a,download:o,dispose:i})},a=(e,t)=>{const{dataType:s,dims:n,download:o,dispose:i}=t;return new r.Tensor({location:"gpu-buffer",type:s??"float32",gpuBuffer:e,dims:n,download:o,dispose:i})},l=(e,t)=>{const{dataType:s,dims:n,download:o,dispose:i}=t;return new r.Tensor({location:"ml-tensor",type:s??"float32",mlTensor:e,dims:n,download:o,dispose:i})},c=(e,t,s)=>new r.Tensor({location:"cpu-pinned",type:e,data:t,dims:s??[t.length]})},"./node_modules/onnxruntime-common/dist/esm/tensor-factory.js":(e,t,s)=>{"use strict";s.r(t)},"./node_modules/onnxruntime-common/dist/esm/tensor-impl-type-mapping.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{NUMERIC_TENSOR_TYPEDARRAY_TO_TYPE_MAP:()=>n,NUMERIC_TENSOR_TYPE_TO_TYPEDARRAY_MAP:()=>r,checkTypedArray:()=>i});const r=new Map([["float32",Float32Array],["uint8",Uint8Array],["int8",Int8Array],["uint16",Uint16Array],["int16",Int16Array],["int32",Int32Array],["bool",Uint8Array],["float64",Float64Array],["uint32",Uint32Array],["int4",Uint8Array],["uint4",Uint8Array]]),n=new Map([[Float32Array,"float32"],[Uint8Array,"uint8"],[Int8Array,"int8"],[Uint16Array,"uint16"],[Int16Array,"int16"],[Int32Array,"int32"],[Float64Array,"float64"],[Uint32Array,"uint32"]]);let o=!1;const i=()=>{if(!o){o=!0;const e="undefined"!=typeof BigInt64Array&&BigInt64Array.from,t="undefined"!=typeof BigUint64Array&&BigUint64Array.from,s="undefined"!=typeof Float16Array&&Float16Array.from;e&&(r.set("int64",BigInt64Array),n.set(BigInt64Array,"int64")),t&&(r.set("uint64",BigUint64Array),n.set(BigUint64Array,"uint64")),s?(r.set("float16",Float16Array),n.set(Float16Array,"float16")):r.set("float16",Uint16Array)}}},"./node_modules/onnxruntime-common/dist/esm/tensor-impl.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Tensor:()=>a});var r=s("./node_modules/onnxruntime-common/dist/esm/tensor-conversion-impl.js"),n=s("./node_modules/onnxruntime-common/dist/esm/tensor-factory-impl.js"),o=s("./node_modules/onnxruntime-common/dist/esm/tensor-impl-type-mapping.js"),i=s("./node_modules/onnxruntime-common/dist/esm/tensor-utils-impl.js");class a{constructor(e,t,s){let r,n;if((0,o.checkTypedArray)(),"object"==typeof e&&"location"in e)switch(this.dataLocation=e.location,r=e.type,n=e.dims,e.location){case"cpu-pinned":{const t=o.NUMERIC_TENSOR_TYPE_TO_TYPEDARRAY_MAP.get(r);if(!t)throw new TypeError(`unsupported type "${r}" to create tensor from pinned buffer`);if(!(e.data instanceof t))throw new TypeError(`buffer should be of type ${t.name}`);this.cpuData=e.data;break}case"texture":if("float32"!==r)throw new TypeError(`unsupported type "${r}" to create tensor from texture`);this.gpuTextureData=e.texture,this.downloader=e.download,this.disposer=e.dispose;break;case"gpu-buffer":if("float32"!==r&&"float16"!==r&&"int32"!==r&&"int64"!==r&&"uint32"!==r&&"uint8"!==r&&"bool"!==r&&"uint4"!==r&&"int4"!==r)throw new TypeError(`unsupported type "${r}" to create tensor from gpu buffer`);this.gpuBufferData=e.gpuBuffer,this.downloader=e.download,this.disposer=e.dispose;break;case"ml-tensor":if("float32"!==r&&"float16"!==r&&"int32"!==r&&"int64"!==r&&"uint32"!==r&&"uint64"!==r&&"int8"!==r&&"uint8"!==r&&"bool"!==r)throw new TypeError(`unsupported type "${r}" 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 i,a;if("string"==typeof e)if(r=e,a=s,"string"===e){if(!Array.isArray(t))throw new TypeError("A string tensor's data must be a string array.");i=t}else{const s=o.NUMERIC_TENSOR_TYPE_TO_TYPEDARRAY_MAP.get(e);if(void 0===s)throw new TypeError(`Unsupported tensor type: ${e}.`);if(Array.isArray(t)){if("float16"===e&&s===Uint16Array||"uint4"===e||"int4"===e)throw new TypeError(`Creating a ${e} tensor from number array is not supported. Please use ${s.name} as data.`);i="uint64"===e||"int64"===e?s.from(t,BigInt):s.from(t)}else if(t instanceof s)i=t;else{if(!(t instanceof Uint8ClampedArray))throw new TypeError(`A ${r} tensor's data must be type of ${s}`);if("uint8"!==e)throw new TypeError("A Uint8ClampedArray tensor's data must be type of uint8");i=Uint8Array.from(t)}}else if(a=t,Array.isArray(e)){if(0===e.length)throw new TypeError("Tensor type cannot be inferred from an empty array.");const t=typeof e[0];if("string"===t)r="string",i=e;else{if("boolean"!==t)throw new TypeError(`Invalid element type of data array: ${t}.`);r="bool",i=Uint8Array.from(e)}}else if(e instanceof Uint8ClampedArray)r="uint8",i=Uint8Array.from(e);else{const t=o.NUMERIC_TENSOR_TYPEDARRAY_TO_TYPE_MAP.get(e.constructor);if(void 0===t)throw new TypeError(`Unsupported type for tensor data: ${e.constructor}.`);r=t,i=e}if(void 0===a)a=[i.length];else if(!Array.isArray(a))throw new TypeError("A tensor's dims must be a number array");n=a,this.cpuData=i,this.dataLocation="cpu"}const a=(0,i.calculateSize)(n);if(this.cpuData&&a!==this.cpuData.length&&("uint4"!==r&&"int4"!==r||Math.ceil(a/2)!==this.cpuData.length))throw new Error(`Tensor's size(${a}) does not match data length(${this.cpuData.length}).`);this.type=r,this.dims=n,this.size=a}static async fromImage(e,t){return(0,n.tensorFromImage)(e,t)}static fromTexture(e,t){return(0,n.tensorFromTexture)(e,t)}static fromGpuBuffer(e,t){return(0,n.tensorFromGpuBuffer)(e,t)}static fromMLTensor(e,t){return(0,n.tensorFromMLTensor)(e,t)}static fromPinnedBuffer(e,t,s){return(0,n.tensorFromPinnedBuffer)(e,t,s)}toDataURL(e){return(0,r.tensorToDataURL)(this,e)}toImageData(e){return(0,r.tensorToImageData)(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;const 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: ${this.dataLocation}`)}}dispose(){if(this.isDownloading)throw new Error("The current tensor is being downloaded.");this.disposer&&(this.disposer(),this.disposer=void 0),this.cpuData=void 0,this.gpuTextureData=void 0,this.gpuBufferData=void 0,this.mlTensorData=void 0,this.downloader=void 0,this.isDownloading=void 0,this.dataLocation="none"}ensureValid(){if("none"===this.dataLocation)throw new Error("The tensor is disposed.")}reshape(e){if(this.ensureValid(),this.downloader||this.disposer)throw new Error("Cannot reshape a tensor that owns GPU resource.");return(0,i.tensorReshape)(this,e)}}},"./node_modules/onnxruntime-common/dist/esm/tensor-utils-impl.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{calculateSize:()=>n,tensorReshape:()=>o});var r=s("./node_modules/onnxruntime-common/dist/esm/tensor-impl.js");const n=e=>{let t=1;for(let s=0;s<e.length;s++){const r=e[s];if("number"!=typeof r||!Number.isSafeInteger(r))throw new TypeError(`dims[${s}] must be an integer, got: ${r}`);if(r<0)throw new RangeError(`dims[${s}] must be a non-negative integer, got: ${r}`);t*=r}return t},o=(e,t)=>{switch(e.location){case"cpu":return new r.Tensor(e.type,e.data,t);case"cpu-pinned":return new r.Tensor({location:"cpu-pinned",data:e.data,type:e.type,dims:t});case"texture":return new r.Tensor({location:"texture",texture:e.texture,type:e.type,dims:t});case"gpu-buffer":return new r.Tensor({location:"gpu-buffer",gpuBuffer:e.gpuBuffer,type:e.type,dims:t});case"ml-tensor":return new r.Tensor({location:"ml-tensor",mlTensor:e.mlTensor,type:e.type,dims:t});default:throw new Error(`tensorReshape: tensor location ${e.location} is not supported`)}}},"./node_modules/onnxruntime-common/dist/esm/tensor.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Tensor:()=>r});const r=s("./node_modules/onnxruntime-common/dist/esm/tensor-impl.js").Tensor},"./node_modules/onnxruntime-common/dist/esm/trace.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{TRACE:()=>n,TRACE_FUNC_BEGIN:()=>i,TRACE_FUNC_END:()=>a});var r=s("./node_modules/onnxruntime-common/dist/esm/env-impl.js");const n=(e,t)=>{(void 0===r.env.trace?r.env.wasm.trace:r.env.trace)&&console.timeStamp(`${e}::ORT::${t}`)},o=(e,t)=>{const s=(new Error).stack?.split(/\r\n|\r|\n/g)||[];let r=!1;for(let o=0;o<s.length;o++){if(r&&!s[o].includes("TRACE_FUNC")){let r=`FUNC_${e}::${s[o].trim().split(" ")[1]}`;return t&&(r+=`::${t}`),void n("CPU",r)}s[o].includes("TRACE_FUNC")&&(r=!0)}},i=e=>{(void 0===r.env.trace?r.env.wasm.trace:r.env.trace)&&o("BEGIN",e)},a=e=>{(void 0===r.env.trace?r.env.wasm.trace:r.env.trace)&&o("END",e)}},"./node_modules/onnxruntime-common/dist/esm/training-session-impl.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{TrainingSession:()=>o});var r=s("./node_modules/onnxruntime-common/dist/esm/backend-impl.js"),n=s("./node_modules/onnxruntime-common/dist/esm/tensor.js");class o{constructor(e,t,s){this.handler=e,this.hasOptimizerModel=t,this.hasEvalModel=s}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(e,t){const s=e.evalModel||"",n=e.optimizerModel||"",i=t||{},[a,l]=await(0,r.resolveBackendAndExecutionProviders)(i);if(a.createTrainingSessionHandler){const t=await a.createTrainingSessionHandler(e.checkpointState,e.trainModel,s,n,l);return new o(t,!!e.optimizerModel,!!e.evalModel)}throw new Error("Training backend could not be resolved. Make sure you're using the correct configuration & WebAssembly files.")}typeNarrowingForRunStep(e,t,s,r,o){const i={};let a={};if("object"!=typeof s||null===s||s instanceof n.Tensor||Array.isArray(s))throw new TypeError("'feeds' must be an object that use input names as keys and OnnxValue as corresponding values.");let l=!0;if("object"==typeof r){if(null===r)throw new TypeError("Unexpected argument[1]: cannot be null.");if(r instanceof n.Tensor)throw new TypeError("'fetches' cannot be a Tensor");if(Array.isArray(r)){if(0===r.length)throw new TypeError("'fetches' cannot be an empty array.");l=!1;for(const e of r){if("string"!=typeof e)throw new TypeError("'fetches' must be a string array or an object.");if(-1===t.indexOf(e))throw new RangeError(`'fetches' contains invalid output name: ${e}.`);i[e]=null}if("object"==typeof o&&null!==o)a=o;else if(void 0!==o)throw new TypeError("'options' must be an object.")}else{let e=!1;const s=Object.getOwnPropertyNames(r);for(const o of t)if(-1!==s.indexOf(o)){const t=r[o];(null===t||t instanceof n.Tensor)&&(e=!0,l=!1,i[o]=t)}if(e){if("object"==typeof o&&null!==o)a=o;else if(void 0!==o)throw new TypeError("'options' must be an object.")}else a=r}}else if(void 0!==r)throw new TypeError("Unexpected argument[1]: must be 'fetches' or 'options'.");for(const t of e)if(void 0===s[t])throw new Error(`input '${t}' is missing in 'feeds'.`);if(l)for(const e of t)i[e]=null;return[i,a]}convertHandlerReturnTypeToMapOfTensors(e){const t={};for(const s in e)if(Object.hasOwnProperty.call(e,s)){const r=e[s];r instanceof n.Tensor?t[s]=r:t[s]=new n.Tensor(r.type,r.data,r.dims)}return t}async lazyResetGrad(){await this.handler.lazyResetGrad()}async runTrainStep(e,t,s){const[r,n]=this.typeNarrowingForRunStep(this.trainingInputNames,this.trainingOutputNames,e,t,s),o=await this.handler.runTrainStep(e,r,n);return this.convertHandlerReturnTypeToMapOfTensors(o)}async runOptimizerStep(e){if(!this.hasOptimizerModel)throw new Error("This TrainingSession has no OptimizerModel loaded.");await this.handler.runOptimizerStep(e||{})}async runEvalStep(e,t,s){if(this.hasEvalModel){const[r,n]=this.typeNarrowingForRunStep(this.evalInputNames,this.evalOutputNames,e,t,s),o=await this.handler.runEvalStep(e,r,n);return this.convertHandlerReturnTypeToMapOfTensors(o)}throw new Error("This TrainingSession has no EvalModel loaded.")}async getParametersSize(e=!0){return this.handler.getParametersSize(e)}async loadParametersBuffer(e,t=!0){const s=await this.getParametersSize(t);if(e.length!==4*s)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(e,t)}async getContiguousParameters(e=!0){return this.handler.getContiguousParameters(e)}async release(){return this.handler.dispose()}}},"./node_modules/onnxruntime-common/dist/esm/training-session.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{TrainingSession:()=>r});const r=s("./node_modules/onnxruntime-common/dist/esm/training-session-impl.js").TrainingSession},"./node_modules/onnxruntime-common/dist/esm/version.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{version:()=>r});const r="1.20.1"},"./src/backends/onnx.js":(e,t,s)=>{"use strict";var r,n;s.r(t),s.d(t,{Tensor:()=>l.Tensor,createInferenceSession:()=>g,deviceToExecutionProviders:()=>h,isONNXProxy:()=>b,isONNXTensor:()=>w});var o=s("./src/env.js"),i=s("onnxruntime-node"),a=s("?8b6b"),l=s("./node_modules/onnxruntime-common/dist/esm/index.js");const c=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"}}),d=[];let u,_;const p=Symbol.for("onnxruntime");if(p in globalThis)_=globalThis[p];else if(o.apis.IS_NODE_ENV){switch(_=i??(r||(r=s.t(i,2))),process.platform){case"win32":d.push("dml");break;case"linux":"x64"===process.arch&&d.push("cuda")}d.push("cpu"),u=["cpu"]}else _=n||(n=s.t(a,2)),o.apis.IS_WEBNN_AVAILABLE&&d.push("webnn-npu","webnn-gpu","webnn-cpu","webnn"),o.apis.IS_WEBGPU_AVAILABLE&&d.push("webgpu"),d.push("wasm"),u=["wasm"];const m=_.InferenceSession;function h(e=null){if(!e)return u;switch(e){case"auto":return d;case"gpu":return d.filter((e=>["webgpu","cuda","dml","webnn-gpu"].includes(e)))}if(d.includes(e))return[c[e]??e];throw new Error(`Unsupported device: "${e}". Should be one of: ${d.join(", ")}.`)}let f=null;async function g(e,t,s){f&&await f;const r=m.create(e,t);f??=r;const n=await r;return n.config=s,n}function w(e){return e instanceof _.Tensor}const M=_?.env;function b(){return M?.wasm?.proxy}M?.wasm&&(M.wasm.wasmPaths=`https://cdn.jsdelivr.net/npm/@huggingface/transformers@${o.env.version}/dist/`,M.wasm.proxy=!1,"undefined"!=typeof crossOriginIsolated&&crossOriginIsolated||(M.wasm.numThreads=1)),M?.webgpu&&(M.webgpu.powerPreference="high-performance"),o.env.backends.onnx=M},"./src/base/feature_extraction_utils.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{FeatureExtractor:()=>i,validate_audio_inputs:()=>a});var r=s("./src/utils/constants.js"),n=s("./src/utils/generic.js"),o=s("./src/utils/hub.js");class i extends n.Callable{constructor(e){super(),this.config=e}static async from_pretrained(e,t){return new this(await(0,o.getModelJSON)(e,r.FEATURE_EXTRACTOR_NAME,!0,t))}}function a(e,t){if(!(e instanceof Float32Array||e instanceof Float64Array))throw new Error(`${t} expects input to be a Float32Array or a Float64Array, but got ${e?.constructor?.name??typeof e} 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":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ImageProcessor:()=>M,center_to_corners_format:()=>u,post_process_instance_segmentation:()=>w,post_process_object_detection:()=>_,post_process_panoptic_segmentation:()=>g,post_process_semantic_segmentation:()=>p});var r=s("./src/utils/generic.js"),n=s("./src/utils/tensor.js"),o=s("./src/utils/maths.js"),i=(s("./src/utils/image.js"),s("./src/utils/core.js")),a=s("./src/utils/hub.js"),l=s("./src/utils/constants.js");function c(e,t,s=0,r=null){const n=e/t;let i=(0,o.bankers_round)(n)*t;return null!==r&&i>r&&(i=Math.floor(n)*t),i<s&&(i=Math.ceil(n)*t),i}function d([e,t],s){return[Math.max(Math.floor(e/s),1)*s,Math.max(Math.floor(t/s),1)*s]}function u([e,t,s,r]){return[e-s/2,t-r/2,e+s/2,t+r/2]}function _(e,t=.5,s=null,r=!1){const n=e.logits,i=e.pred_boxes,[a,l,c]=n.dims;if(null!==s&&s.length!==a)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");let d=[];for(let e=0;e<a;++e){let a=null!==s?s[e]:null,_={boxes:[],classes:[],scores:[]},p=n[e],m=i[e];for(let e=0;e<l;++e){let s,n=p[e],i=[];if(r){s=n.sigmoid().data;for(let e=0;e<s.length;++e)s[e]>t&&i.push(e)}else{let e=(0,o.max)(n.data)[1];if(e===c-1)continue;if(s=(0,o.softmax)(n.data),s[e]<t)continue;i.push(e)}for(const t of i){let r=m[e].data;r=u(r),null!==a&&(r=r.map(((e,t)=>e*a[(t+1)%2]))),_.boxes.push(r),_.classes.push(t),_.scores.push(s[t])}}d.push(_)}return d}function p(e,t=null){const s=e.logits,r=s.dims[0];if(null!==t&&t.length!==r)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");const o=[];for(let e=0;e<r;++e){const r=null!==t?t[e]:null;let i=s[e];null!==r&&(i=(0,n.interpolate)(i,r,"bilinear",!1));const[a,l]=r??i.dims.slice(-2),c=new n.Tensor("int32",new Int32Array(a*l),[a,l]),d=i[0].data,u=c.data;for(let e=1;e<i.dims[0];++e){const t=i[e].data;for(let s=0;s<t.length;++s)t[s]>d[s]&&(d[s]=t[s],u[s]=e)}const _=new Array(i.dims[0]);for(let e=0;e<u.length;++e){const t=u[e];_[t]=t}const p=_.filter((e=>void 0!==e));o.push({segmentation:c,labels:p})}return o}function m(e,t,s,r){const n=[],i=[],a=[];for(let l=0;l<e.dims[0];++l){const c=e[l],d=t[l],u=(0,o.max)(c.data)[1];if(u===r)continue;const _=(0,o.softmax)(c.data)[u];_>s&&(n.push(d),i.push(_),a.push(u))}return[n,i,a]}function h(e,t,s,r=.5,n=.8){const o=[];let i=0,a=0;const l=t[s].data;for(let t=0;t<e.length;++t)e[t]===s&&(o.push(t),++i),l[t]>=r&&++a;let c=i>0&&a>0;if(c){c=i/a>n}return[c,o]}function f(e,t,s,r,o,i=null,a=null){const[l,c]=a??e[0].dims,d=new n.Tensor("int32",new Int32Array(l*c),[l,c]),u=[];if(null!==a)for(let t=0;t<e.length;++t)e[t]=(0,n.interpolate)(e[t],a,"bilinear",!1);const _=new Int32Array(e[0].data.length),p=new Float32Array(e[0].data.length);for(let s=0;s<e.length;++s){let r=t[s];const n=e[s].data;for(let e=0;e<n.length;++e)n[e]*=r,n[e]>p[e]&&(_[e]=s,p[e]=n[e])}let m=0;const f=d.data;for(let n=0;n<s.length;++n){const i=s[n],[a,l]=h(_,e,n,r,o);if(a){++m;for(const e of l)f[e]=m;u.push({id:m,label_id:i,score:t[n]})}}return[d,u]}function g(e,t=.5,s=.5,r=.8,o=null,i=null){null===o&&(console.warn("`label_ids_to_fuse` unset. No instance will be fused."),o=new Set);const a=e.class_queries_logits??e.logits,l=(e.masks_queries_logits??e.pred_masks).sigmoid();let[c,d,u]=a.dims;if(u-=1,null!==i&&i.length!==c)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");let _=[];for(let e=0;e<c;++e){let c=null!==i?i[e]:null,d=a[e],p=l[e],[h,g,w]=m(d,p,t,u);if(0===w.length){let[e,t]=c??p.dims.slice(-2),s=new n.Tensor("int32",new Int32Array(e*t).fill(-1),[e,t]);_.push({segmentation:s,segments_info:[]});continue}let[M,b]=f(h,g,w,s,r,o,c);_.push({segmentation:M,segments_info:b})}return _}function w(e,t=.5,s=null){throw new Error("`post_process_instance_segmentation` is not yet implemented.")}class M extends r.Callable{constructor(e){super(),this.image_mean=e.image_mean??e.mean,this.image_std=e.image_std??e.std,this.resample=e.resample??2,this.do_rescale=e.do_rescale??!0,this.rescale_factor=e.rescale_factor??1/255,this.do_normalize=e.do_normalize,this.do_thumbnail=e.do_thumbnail,this.size=e.size??e.image_size,this.do_resize=e.do_resize??void 0!==this.size,this.size_divisibility=e.size_divisibility??e.size_divisor,this.do_center_crop=e.do_center_crop,this.crop_size=e.crop_size,this.do_convert_rgb=e.do_convert_rgb??!0,this.do_crop_margin=e.do_crop_margin,this.pad_size=e.pad_size,this.do_pad=e.do_pad,this.do_pad&&!this.pad_size&&this.size&&void 0!==this.size.width&&void 0!==this.size.height&&(this.pad_size=this.size),this.do_flip_channel_order=e.do_flip_channel_order??!1,this.config=e}async thumbnail(e,t,s=2){const r=e.height,n=e.width,o=t.height,i=t.width;let a=Math.min(r,o),l=Math.min(n,i);return a===r&&l===n?e:(r>n?l=Math.floor(n*a/r):n>r&&(a=Math.floor(r*l/n)),await e.resize(l,a,{resample:s}))}async crop_margin(e,t=200){const s=e.clone().grayscale(),r=(0,o.min)(s.data)[0],n=(0,o.max)(s.data)[0]-r;if(0===n)return e;const i=t/255;let a=s.width,l=s.height,c=0,d=0;const u=s.data;for(let e=0;e<s.height;++e){const t=e*s.width;for(let o=0;o<s.width;++o)(u[t+o]-r)/n<i&&(a=Math.min(a,o),l=Math.min(l,e),c=Math.max(c,o),d=Math.max(d,e))}return e=await e.crop([a,l,c,d])}pad_image(e,t,s,{mode:r="constant",center:n=!1,constant_values:o=0}={}){const[a,l,c]=t;let d,u;if("number"==typeof s?(d=s,u=s):"square"===s?d=u=Math.max(a,l):(d=s.width,u=s.height),d!==l||u!==a){const s=new Float32Array(d*u*c);if(Array.isArray(o))for(let e=0;e<s.length;++e)s[e]=o[e%c];else 0!==o&&s.fill(o);const[_,p]=n?[Math.floor((d-l)/2),Math.floor((u-a)/2)]:[0,0];for(let t=0;t<a;++t){const r=(t+p)*d,n=t*l;for(let t=0;t<l;++t){const o=(r+t+_)*c,i=(n+t)*c;for(let t=0;t<c;++t)s[o+t]=e[i+t]}}if("symmetric"===r){if(n)throw new Error("`center` padding is not supported when `mode` is set to `symmetric`.");const t=a-1,r=l-1;for(let n=0;n<u;++n){const o=n*d,u=(0,i.calculateReflectOffset)(n,t)*l;for(let t=0;t<d;++t){if(n<a&&t<l)continue;const d=(o+t)*c,_=(u+(0,i.calculateReflectOffset)(t,r))*c;for(let t=0;t<c;++t)s[d+t]=e[_+t]}}}e=s,t=[u,d,c]}return[e,t]}rescale(e){for(let t=0;t<e.length;++t)e[t]=this.rescale_factor*e[t]}get_resize_output_image_size(e,t){const[s,r]=e.size;let n,o;if(this.do_thumbnail){const{height:e,width:s}=t;n=Math.min(e,s)}else Number.isInteger(t)?(n=t,o=this.config.max_size??n):void 0!==t&&(n=t.shortest_edge,o=t.longest_edge);if(void 0!==n||void 0!==o){const e=void 0===n?1:Math.max(n/s,n/r),t=s*e,i=r*e,a=void 0===o?1:Math.min(o/t,o/i);let l=Math.floor(Number((t*a).toFixed(2))),c=Math.floor(Number((i*a).toFixed(2)));return void 0!==this.size_divisibility&&([l,c]=d([l,c],this.size_divisibility)),[l,c]}if(void 0!==t&&void 0!==t.width&&void 0!==t.height){let e=t.width,n=t.height;if(this.config.keep_aspect_ratio&&this.config.ensure_multiple_of){let t=n/r,o=e/s;Math.abs(1-o)<Math.abs(1-t)?t=o:o=t,n=c(t*r,this.config.ensure_multiple_of),e=c(o*s,this.config.ensure_multiple_of)}return[e,n]}if(void 0!==this.size_divisibility)return d([s,r],this.size_divisibility);if(void 0!==t.min_pixels&&void 0!==t.max_pixels){const{min_pixels:e,max_pixels:n}=t;return function(e,t,s=28,r=3136,n=1003520){if(e<s||t<s)throw new Error(`height:${e} or width:${t} must be larger than factor:${s}`);if(Math.max(e,t)/Math.min(e,t)>200)throw new Error("absolute aspect ratio must be smaller than 200, got "+Math.max(e,t)/Math.min(e,t));let o=Math.round(e/s)*s,i=Math.round(t/s)*s;if(o*i>n){const r=Math.sqrt(e*t/n);o=Math.floor(e/r/s)*s,i=Math.floor(t/r/s)*s}else if(o*i<r){const n=Math.sqrt(r/(e*t));o=Math.ceil(e*n/s)*s,i=Math.ceil(t*n/s)*s}return[o,i]}(r,s,this.config.patch_size*this.config.merge_size,e,n)}throw new Error(`Could not resize image due to unsupported \`this.size\` option in config: ${JSON.stringify(t)}`)}async resize(e){const[t,s]=this.get_resize_output_image_size(e,this.size);return await e.resize(t,s,{resample:this.resample})}async preprocess(e,{do_normalize:t=null,do_pad:s=null,do_convert_rgb:r=null,do_convert_grayscale:o=null,do_flip_channel_order:i=null}={}){this.do_crop_margin&&(e=await this.crop_margin(e));const[a,l]=e.size;if(r??this.do_convert_rgb?e=e.rgb():o&&(e=e.grayscale()),this.do_resize&&(e=await this.resize(e)),this.do_thumbnail&&(e=await this.thumbnail(e,this.size,this.resample)),this.do_center_crop){let t,s;Number.isInteger(this.crop_size)?(t=this.crop_size,s=this.crop_size):(t=this.crop_size.width,s=this.crop_size.height),e=await e.center_crop(t,s)}const c=[e.height,e.width];let u=Float32Array.from(e.data),_=[e.height,e.width,e.channels];if(this.do_rescale&&this.rescale(u),t??this.do_normalize){let t=this.image_mean;Array.isArray(this.image_mean)||(t=new Array(e.channels).fill(t));let s=this.image_std;if(Array.isArray(this.image_std)||(s=new Array(e.channels).fill(t)),t.length!==e.channels||s.length!==e.channels)throw new Error(`When set to arrays, the length of \`image_mean\` (${t.length}) and \`image_std\` (${s.length}) must match the number of channels in the image (${e.channels}).`);for(let r=0;r<u.length;r+=e.channels)for(let n=0;n<e.channels;++n)u[r+n]=(u[r+n]-t[n])/s[n]}if(s??this.do_pad)if(this.pad_size){const t=this.pad_image(u,[e.height,e.width,e.channels],this.pad_size);[u,_]=t}else if(this.size_divisibility){const[e,t]=d([_[1],_[0]],this.size_divisibility);[u,_]=this.pad_image(u,_,{width:e,height:t})}if(i??this.do_flip_channel_order){if(3!==_[2])throw new Error("Flipping channel order is only supported for RGB images.");for(let e=0;e<u.length;e+=3){const t=u[e];u[e]=u[e+2],u[e+2]=t}}return{original_size:[l,a],reshaped_input_size:c,pixel_values:new n.Tensor("float32",u,_).permute(2,0,1)}}async _call(e,...t){Array.isArray(e)||(e=[e]);const s=await Promise.all(e.map((e=>this.preprocess(e))));return{pixel_values:(0,n.stack)(s.map((e=>e.pixel_values)),0),original_sizes:s.map((e=>e.original_size)),reshaped_input_sizes:s.map((e=>e.reshaped_input_size))}}static async from_pretrained(e,t){return new this(await(0,a.getModelJSON)(e,l.IMAGE_PROCESSOR_NAME,!0,t))}}},"./src/base/processing_utils.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Processor:()=>i});var r=s("./src/utils/constants.js"),n=s("./src/utils/generic.js"),o=s("./src/utils/hub.js");class i extends n.Callable{static classes=["image_processor_class","tokenizer_class","feature_extractor_class"];static uses_processor_config=!1;constructor(e,t){super(),this.config=e,this.components=t}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(e,t={}){if(!this.tokenizer)throw new Error("Unable to apply chat template without a tokenizer.");return this.tokenizer.apply_chat_template(e,{tokenize:!1,...t})}batch_decode(...e){if(!this.tokenizer)throw new Error("Unable to decode without a tokenizer.");return this.tokenizer.batch_decode(...e)}decode(...e){if(!this.tokenizer)throw new Error("Unable to decode without a tokenizer.");return this.tokenizer.decode(...e)}async _call(e,...t){for(const s of[this.image_processor,this.feature_extractor,this.tokenizer])if(s)return s(e,...t);throw new Error("No image processor, feature extractor, or tokenizer found.")}static async from_pretrained(e,t){const[s,n]=await Promise.all([this.uses_processor_config?(0,o.getModelJSON)(e,r.PROCESSOR_NAME,!0,t):{},Promise.all(this.classes.filter((e=>e in this)).map((async s=>{const r=await this[s].from_pretrained(e,t);return[s.replace(/_class$/,""),r]}))).then(Object.fromEntries)]);return new this(s,n)}}},"./src/configs.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{AutoConfig:()=>l,PretrainedConfig:()=>a,getKeyValueShapes:()=>i});var r=s("./src/utils/core.js"),n=s("./src/utils/hub.js");function o(e){const t={};let s={};switch(e.model_type){case"llava":case"paligemma":case"florence2":case"llava_onevision":case"idefics3":s=o(e.text_config);break;case"moondream1":s=o(e.phi_config);break;case"musicgen":s=o(e.decoder);break;case"multi_modality":s=o(e.language_config);break;case"gpt2":case"gptj":case"jais":case"codegen":case"gpt_bigcode":t.num_heads="n_head",t.num_layers="n_layer",t.hidden_size="n_embd";break;case"gpt_neox":case"stablelm":case"opt":case"falcon":t.num_heads="num_attention_heads",t.num_layers="num_hidden_layers",t.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":t.num_heads="num_key_value_heads",t.num_layers="num_hidden_layers",t.hidden_size="hidden_size",t.num_attention_heads="num_attention_heads";break;case"gemma":case"gemma2":case"glm":case"helium":t.num_heads="num_key_value_heads",t.num_layers="num_hidden_layers",t.dim_kv="head_dim";break;case"openelm":t.num_heads="num_kv_heads",t.num_layers="num_transformer_layers",t.dim_kv="head_dim";break;case"gpt_neo":case"donut-swin":t.num_heads="num_heads",t.num_layers="num_layers",t.hidden_size="hidden_size";break;case"bloom":t.num_heads="n_head",t.num_layers="n_layer",t.hidden_size="hidden_size";break;case"mpt":t.num_heads="n_heads",t.num_layers="n_layers",t.hidden_size="d_model";break;case"exaone":t.num_heads="num_key_value_heads",t.num_layers="num_layers",t.dim_kv="head_dim",t.num_attention_heads="num_attention_heads";break;case"t5":case"mt5":case"longt5":t.num_decoder_layers="num_decoder_layers",t.num_decoder_heads="num_heads",t.decoder_dim_kv="d_kv",t.num_encoder_layers="num_layers",t.num_encoder_heads="num_heads",t.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":t.num_decoder_layers="decoder_layers",t.num_decoder_heads="decoder_attention_heads",t.decoder_hidden_size="d_model",t.num_encoder_layers="encoder_layers",t.num_encoder_heads="encoder_attention_heads",t.encoder_hidden_size="d_model";break;case"speecht5":t.num_decoder_layers="decoder_layers",t.num_decoder_heads="decoder_attention_heads",t.decoder_hidden_size="hidden_size",t.num_encoder_layers="encoder_layers",t.num_encoder_heads="encoder_attention_heads",t.encoder_hidden_size="hidden_size";break;case"trocr":t.num_encoder_layers=t.num_decoder_layers="decoder_layers",t.num_encoder_heads=t.num_decoder_heads="decoder_attention_heads",t.encoder_hidden_size=t.decoder_hidden_size="d_model";break;case"musicgen_decoder":t.num_encoder_layers=t.num_decoder_layers="num_hidden_layers",t.num_encoder_heads=t.num_decoder_heads="num_attention_heads",t.encoder_hidden_size=t.decoder_hidden_size="hidden_size";break;case"moonshine":t.num_decoder_layers="decoder_num_hidden_layers",t.num_decoder_heads="decoder_num_key_value_heads",t.num_encoder_layers="encoder_num_hidden_layers",t.num_encoder_heads="encoder_num_key_value_heads",t.encoder_hidden_size=t.decoder_hidden_size="hidden_size";break;case"vision-encoder-decoder":const n=o(e.decoder),i="num_decoder_layers"in n,a=(0,r.pick)(e,["model_type","is_encoder_decoder"]);return i?(a.num_decoder_layers=n.num_decoder_layers,a.num_decoder_heads=n.num_decoder_heads,a.decoder_hidden_size=n.decoder_hidden_size,a.num_encoder_layers=n.num_encoder_layers,a.num_encoder_heads=n.num_encoder_heads,a.encoder_hidden_size=n.encoder_hidden_size):(a.num_layers=n.num_layers,a.num_heads=n.num_heads,a.hidden_size=n.hidden_size),a}const n={...s,...(0,r.pick)(e,["model_type","multi_query","is_encoder_decoder"])};for(const s in t)n[s]=e[t[s]];return n}function i(e,{prefix:t="past_key_values",batch_size:s=1}={}){const r={},n=e.normalized_config;if(n.is_encoder_decoder&&"num_encoder_heads"in n&&"num_decoder_heads"in n){const e=n.encoder_dim_kv??n.encoder_hidden_size/n.num_encoder_heads,o=n.decoder_dim_kv??n.decoder_hidden_size/n.num_decoder_heads,i=[s,n.num_encoder_heads,0,e],a=[s,n.num_decoder_heads,0,o];for(let e=0;e<n.num_decoder_layers;++e)r[`${t}.${e}.encoder.key`]=i,r[`${t}.${e}.encoder.value`]=i,r[`${t}.${e}.decoder.key`]=a,r[`${t}.${e}.decoder.value`]=a}else{const e=n.num_heads,o=n.num_layers,i=n.dim_kv??n.hidden_size/(n.num_attention_heads??e);if("falcon"===n.model_type){const n=[s*e,0,i];for(let e=0;e<o;++e)r[`${t}.${e}.key`]=n,r[`${t}.${e}.value`]=n}else if(n.multi_query){const n=[s*e,0,2*i];for(let e=0;e<o;++e)r[`${t}.${e}.key_value`]=n}else if("bloom"===n.model_type){const n=[s*e,i,0],a=[s*e,0,i];for(let e=0;e<o;++e)r[`${t}.${e}.key`]=n,r[`${t}.${e}.value`]=a}else if("openelm"===n.model_type)for(let n=0;n<o;++n){const o=[s,e[n],0,i];r[`${t}.${n}.key`]=o,r[`${t}.${n}.value`]=o}else{const n=[s,e,0,i];for(let e=0;e<o;++e)r[`${t}.${e}.key`]=n,r[`${t}.${e}.value`]=n}}return r}class a{model_type=null;is_encoder_decoder=!1;max_position_embeddings;"transformers.js_config";constructor(e){Object.assign(this,e),this.normalized_config=o(this)}static async from_pretrained(e,{progress_callback:t=null,config:s=null,cache_dir:r=null,local_files_only:o=!1,revision:i="main"}={}){!s||s instanceof a||(s=new a(s));const l=s??await async function(e,t){return await(0,n.getModelJSON)(e,"config.json",!0,t)}(e,{progress_callback:t,config:s,cache_dir:r,local_files_only:o,revision:i});return new this(l)}}class l{static async from_pretrained(...e){return a.from_pretrained(...e)}}},"./src/env.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{apis:()=>h,env:()=>b});var r=s("fs"),n=s("path"),o=s("url");const i="undefined"!=typeof window&&void 0!==window.document,a="undefined"!=typeof self&&"DedicatedWorkerGlobalScope"===self.constructor?.name,l="undefined"!=typeof self&&"caches"in self,c="undefined"!=typeof navigator&&"gpu"in navigator,d="undefined"!=typeof navigator&&"ml"in navigator,u="undefined"!=typeof process,_=u&&"node"===process?.release?.name,p=!x(r),m=!x(n),h=Object.freeze({IS_BROWSER_ENV:i,IS_WEBWORKER_ENV:a,IS_WEB_CACHE_AVAILABLE:l,IS_WEBGPU_AVAILABLE:c,IS_WEBNN_AVAILABLE:d,IS_PROCESS_AVAILABLE:u,IS_NODE_ENV:_,IS_FS_AVAILABLE:p,IS_PATH_AVAILABLE:m}),f=p&&m;let g="./";if(f){const e=Object({}).url;e?g=n.dirname(n.dirname(o.fileURLToPath(e))):"undefined"!=typeof __dirname&&(g=n.dirname(__dirname))}const w=f?n.join(g,"/.cache/"):null,M="/models/",b={version:"3.3.3",backends:{onnx:{}},allowRemoteModels:!0,remoteHost:"https://huggingface.co/",remotePathTemplate:"{model}/resolve/{revision}/",allowLocalModels:!(i||a),localModelPath:f?n.join(g,M):M,useFS:p,useBrowserCache:l,useFSCache:p,cacheDir:w,useCustomCache:!1,customCache:null};function x(e){return 0===Object.keys(e).length}},"./src/generation/configuration_utils.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{GenerationConfig:()=>n});var r=s("./src/utils/core.js");class n{max_length=20;max_new_tokens=null;min_length=0;min_new_tokens=null;early_stopping=!1;max_time=null;do_sample=!1;num_beams=1;num_beam_groups=1;penalty_alpha=null;use_cache=!0;temperature=1;top_k=50;top_p=1;typical_p=1;epsilon_cutoff=0;eta_cutoff=0;diversity_penalty=0;repetition_penalty=1;encoder_repetition_penalty=1;length_penalty=1;no_repeat_ngram_size=0;bad_words_ids=null;force_words_ids=null;renormalize_logits=!1;constraints=null;forced_bos_token_id=null;forced_eos_token_id=null;remove_invalid_values=!1;exponential_decay_length_penalty=null;suppress_tokens=null;streamer=null;begin_suppress_tokens=null;forced_decoder_ids=null;guidance_scale=null;num_return_sequences=1;output_attentions=!1;output_hidden_states=!1;output_scores=!1;return_dict_in_generate=!1;pad_token_id=null;bos_token_id=null;eos_token_id=null;encoder_no_repeat_ngram_size=0;decoder_start_token_id=null;generation_kwargs={};constructor(e){Object.assign(this,(0,r.pick)(e,Object.getOwnPropertyNames(this)))}}},"./src/generation/logits_process.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ClassifierFreeGuidanceLogitsProcessor:()=>g,ForcedBOSTokenLogitsProcessor:()=>l,ForcedEOSTokenLogitsProcessor:()=>c,LogitsProcessor:()=>o,LogitsProcessorList:()=>a,LogitsWarper:()=>i,MinLengthLogitsProcessor:()=>m,MinNewTokensLengthLogitsProcessor:()=>h,NoBadWordsLogitsProcessor:()=>f,NoRepeatNGramLogitsProcessor:()=>_,RepetitionPenaltyLogitsProcessor:()=>p,SuppressTokensAtBeginLogitsProcessor:()=>d,TemperatureLogitsWarper:()=>w,TopKLogitsWarper:()=>b,TopPLogitsWarper:()=>M,WhisperTimeStampLogitsProcessor:()=>u});var r=s("./src/utils/generic.js"),n=(s("./src/utils/tensor.js"),s("./src/utils/maths.js"));class o extends r.Callable{_call(e,t){throw Error("`_call` should be implemented in a subclass")}}class i extends r.Callable{_call(e,t){throw Error("`_call` should be implemented in a subclass")}}class a extends r.Callable{constructor(){super(),this.processors=[]}push(e){this.processors.push(e)}extend(e){this.processors.push(...e)}_call(e,t){let s=t;for(const t of this.processors)s=t(e,s);return s}[Symbol.iterator](){return this.processors.values()}}class l extends o{constructor(e){super(),this.bos_token_id=e}_call(e,t){for(let s=0;s<e.length;++s)if(1===e[s].length){const e=t[s].data;e.fill(-1/0),e[this.bos_token_id]=0}return t}}class c extends o{constructor(e,t){super(),this.max_length=e,this.eos_token_id=Array.isArray(t)?t:[t]}_call(e,t){for(let s=0;s<e.length;++s)if(e[s].length===this.max_length-1){const e=t[s].data;e.fill(-1/0);for(const t of this.eos_token_id)e[t]=0}return t}}class d extends o{constructor(e,t){super(),this.begin_suppress_tokens=e,this.begin_index=t}_call(e,t){for(let s=0;s<e.length;++s)if(e[s].length===this.begin_index){const e=t[s].data;for(const t of this.begin_suppress_tokens)e[t]=-1/0}return t}}class u extends o{constructor(e,t){super(),this.eos_token_id=Array.isArray(e.eos_token_id)?e.eos_token_id[0]:e.eos_token_id,this.no_timestamps_token_id=e.no_timestamps_token_id,this.timestamp_begin=this.no_timestamps_token_id+1,this.begin_index=t.length,t.at(-1)===this.no_timestamps_token_id&&(this.begin_index-=1),this.max_initial_timestamp_index=e.max_initial_timestamp_index}_call(e,t){for(let s=0;s<e.length;++s){const r=t[s].data;if(r[this.no_timestamps_token_id]=-1/0,e[s].length===this.begin_index-1){r.fill(-1/0),r[this.timestamp_begin]=0;continue}const o=e[s].slice(this.begin_index),i=o.length>=1&&o[o.length-1]>=this.timestamp_begin,a=o.length<2||o[o.length-2]>=this.timestamp_begin;if(i&&(a?r.subarray(this.timestamp_begin).fill(-1/0):r.subarray(0,this.eos_token_id).fill(-1/0)),e[s].length===this.begin_index&&null!==this.max_initial_timestamp_index){const e=this.timestamp_begin+this.max_initial_timestamp_index;r.subarray(e+1).fill(-1/0)}const l=(0,n.log_softmax)(r);Math.log(l.subarray(this.timestamp_begin).map(Math.exp).reduce(((e,t)=>e+t)))>(0,n.max)(l.subarray(0,this.timestamp_begin))[0]&&r.subarray(0,this.timestamp_begin).fill(-1/0)}return t}}class _ extends o{constructor(e){super(),this.no_repeat_ngram_size=e}getNgrams(e){const t=e.length,s=[];for(let r=0;r<t+1-this.no_repeat_ngram_size;++r){const t=[];for(let s=0;s<this.no_repeat_ngram_size;++s)t.push(e[r+s]);s.push(t.map(Number))}const r=new Map;for(const e of s){const t=e.slice(0,e.length-1),s=JSON.stringify(t),n=r.get(s)??[];n.push(e[e.length-1]),r.set(s,n)}return r}getGeneratedNgrams(e,t){const s=t.slice(t.length+1-this.no_repeat_ngram_size,t.length);return e.get(JSON.stringify(s.map(Number)))??[]}calcBannedNgramTokens(e){const t=[];if(e.length+1<this.no_repeat_ngram_size)return t;{const t=this.getNgrams(e);return this.getGeneratedNgrams(t,e)}}_call(e,t){for(let s=0;s<e.length;++s){const r=t[s].data,n=this.calcBannedNgramTokens(e[s]);for(const e of n)r[e]=-1/0}return t}}class p extends o{constructor(e){super(),this.penalty=e}_call(e,t){for(let s=0;s<e.length;++s){const r=t[s].data;for(const t of new Set(e[s])){const e=Number(t);r[e]<0?r[e]*=this.penalty:r[e]/=this.penalty}}return t}}class m extends o{constructor(e,t){super(),this.min_length=e,this.eos_token_id=Array.isArray(t)?t:[t]}_call(e,t){for(let s=0;s<e.length;++s)if(e[s].length<this.min_length){const e=t[s].data;for(const t of this.eos_token_id)e[t]=-1/0}return t}}class h extends o{constructor(e,t,s){super(),this.prompt_length_to_skip=e,this.min_new_tokens=t,this.eos_token_id=Array.isArray(s)?s:[s]}_call(e,t){for(let s=0;s<e.length;++s){if(e[s].length-this.prompt_length_to_skip<this.min_new_tokens){const e=t[s].data;for(const t of this.eos_token_id)e[t]=-1/0}}return t}}class f extends o{constructor(e,t){super(),this.bad_words_ids=e,this.eos_token_id=Array.isArray(t)?t:[t]}_call(e,t){for(let s=0;s<e.length;++s){const r=t[s].data,n=e[s];for(const e of this.bad_words_ids){let t=!0;for(let s=1;s<=e.length-1&&e.length<n.length;++s)if(e.at(-s-1)!=n.at(-s)){t=!1;break}t&&(r[e.at(-1)]=-1/0)}}return t}}class g extends o{constructor(e){if(super(),e<=1)throw new Error(`Require guidance scale >1 to use the classifier free guidance processor, got guidance scale ${e}.`);this.guidance_scale=e}_call(e,t){if(t.dims[0]!==2*e.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 ${t.dims[0]} for the logits and ${e.length} for the input ids.`);const s=e.length,r=t.slice([0,s],null),n=t.slice([s,t.dims[0]],null);for(let e=0;e<n.data.length;++e)n.data[e]+=(r.data[e]-n.data[e])*this.guidance_scale;return n}}class w extends i{constructor(e){if(super(),"number"!=typeof e||e<=0){let t=`\`temperature\` (=${e}) must be a strictly positive float, otherwise your next token scores will be invalid.`;0===e&&(t+=" If you're looking for greedy decoding strategies, set `do_sample=false`.")}this.temperature=e}_call(e,t){const s=t.data;for(let e=0;e<s.length;++e)s[e]/=this.temperature;return t}}class M extends i{constructor(e,{filter_value:t=-1/0,min_tokens_to_keep:s=1}={}){if(super(),e<0||e>1)throw new Error(`\`top_p\` must be a float > 0 and < 1, but is ${e}`);if(!Number.isInteger(s)||s<1)throw new Error(`\`min_tokens_to_keep\` must be a positive integer, but is ${s}`);this.top_p=e,this.filter_value=t,this.min_tokens_to_keep=s}}class b extends i{constructor(e,{filter_value:t=-1/0,min_tokens_to_keep:s=1}={}){if(super(),!Number.isInteger(e)||e<0)throw new Error(`\`top_k\` must be a positive integer, but is ${e}`);this.top_k=Math.max(e,s),this.filter_value=t}}},"./src/generation/logits_sampler.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{LogitsSampler:()=>i});var r=s("./src/utils/generic.js"),n=s("./src/utils/tensor.js"),o=s("./src/utils/maths.js");s("./src/generation/configuration_utils.js");class i extends r.Callable{constructor(e){super(),this.generation_config=e}async _call(e){return this.sample(e)}async sample(e){throw Error("sample should be implemented in subclasses.")}getLogits(e,t){let s=e.dims.at(-1),r=e.data;if(-1===t)r=r.slice(-s);else{let e=t*s;r=r.slice(e,e+s)}return r}randomSelect(e){let t=0;for(let s=0;s<e.length;++s)t+=e[s];let s=Math.random()*t;for(let t=0;t<e.length;++t)if(s-=e[t],s<=0)return t;return 0}static getSampler(e){if(e.do_sample)return new l(e);if(e.num_beams>1)return new c(e);if(e.num_return_sequences>1)throw Error(`num_return_sequences has to be 1 when doing greedy search, but is ${e.num_return_sequences}.`);return new a(e)}}class a extends i{async sample(e){const t=(0,o.max)(e.data)[1];return[[BigInt(t),0]]}}class l extends i{async sample(e){let t=e.dims.at(-1);this.generation_config.top_k>0&&(t=Math.min(this.generation_config.top_k,t));const[s,r]=await(0,n.topk)(e,t),i=(0,o.softmax)(s.data);return Array.from({length:this.generation_config.num_beams},(()=>{const e=this.randomSelect(i);return[r.data[e],Math.log(i[e])]}))}}class c extends i{async sample(e){let t=e.dims.at(-1);this.generation_config.top_k>0&&(t=Math.min(this.generation_config.top_k,t));const[s,r]=await(0,n.topk)(e,t),i=(0,o.softmax)(s.data);return Array.from({length:this.generation_config.num_beams},((e,t)=>[r.data[t],Math.log(i[t])]))}}},"./src/generation/stopping_criteria.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{EosTokenCriteria:()=>a,InterruptableStoppingCriteria:()=>l,MaxLengthCriteria:()=>i,StoppingCriteria:()=>n,StoppingCriteriaList:()=>o});var r=s("./src/utils/generic.js");class n extends r.Callable{_call(e,t){throw Error("StoppingCriteria needs to be subclassed")}}class o extends r.Callable{constructor(){super(),this.criteria=[]}push(e){this.criteria.push(e)}extend(e){e instanceof o?e=e.criteria:e instanceof n&&(e=[e]),this.criteria.push(...e)}_call(e,t){const s=new Array(e.length).fill(!1);for(const r of this.criteria){const n=r(e,t);for(let e=0;e<s.length;++e)s[e]||=n[e]}return s}[Symbol.iterator](){return this.criteria.values()}}class i extends n{constructor(e,t=null){super(),this.max_length=e,this.max_position_embeddings=t}_call(e){return e.map((e=>e.length>=this.max_length))}}class a extends n{constructor(e){super(),Array.isArray(e)||(e=[e]),this.eos_token_id=e}_call(e,t){return e.map((e=>{const t=e.at(-1);return this.eos_token_id.some((e=>t==e))}))}}class l extends n{constructor(){super(),this.interrupted=!1}interrupt(){this.interrupted=!0}reset(){this.interrupted=!1}_call(e,t){return new Array(e.length).fill(this.interrupted)}}},"./src/generation/streamers.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{BaseStreamer:()=>i,TextStreamer:()=>l,WhisperTextStreamer:()=>c});var r=s("./src/utils/core.js"),n=s("./src/tokenizers.js"),o=s("./src/env.js");class i{put(e){throw Error("Not implemented")}end(){throw Error("Not implemented")}}const a=o.apis.IS_PROCESS_AVAILABLE?e=>process.stdout.write(e):e=>console.log(e);class l extends i{constructor(e,{skip_prompt:t=!1,callback_function:s=null,token_callback_function:r=null,skip_special_tokens:n=!0,decode_kwargs:o={},...i}={}){super(),this.tokenizer=e,this.skip_prompt=t,this.callback_function=s??a,this.token_callback_function=r,this.decode_kwargs={skip_special_tokens:n,...o,...i},this.token_cache=[],this.print_len=0,this.next_tokens_are_prompt=!0}put(e){if(e.length>1)throw Error("TextStreamer only supports batch size of 1");if(this.skip_prompt&&this.next_tokens_are_prompt)return void(this.next_tokens_are_prompt=!1);const t=e[0];this.token_callback_function?.(t),this.token_cache=(0,r.mergeArrays)(this.token_cache,t);const s=this.tokenizer.decode(this.token_cache,this.decode_kwargs);let o;s.endsWith("\n")?(o=s.slice(this.print_len),this.token_cache=[],this.print_len=0):s.length>0&&(0,n.is_chinese_char)(s.charCodeAt(s.length-1))?(o=s.slice(this.print_len),this.print_len+=o.length):(o=s.slice(this.print_len,s.lastIndexOf(" ")+1),this.print_len+=o.length),this.on_finalized_text(o,!1)}end(){let e;if(this.token_cache.length>0){e=this.tokenizer.decode(this.token_cache,this.decode_kwargs).slice(this.print_len),this.token_cache=[],this.print_len=0}else e="";this.next_tokens_are_prompt=!0,this.on_finalized_text(e,!0)}on_finalized_text(e,t){e.length>0&&this.callback_function?.(e),t&&this.callback_function===a&&o.apis.IS_PROCESS_AVAILABLE&&this.callback_function?.("\n")}}class c extends l{constructor(e,{skip_prompt:t=!1,callback_function:s=null,token_callback_function:r=null,on_chunk_start:n=null,on_chunk_end:o=null,on_finalize:i=null,time_precision:a=.02,skip_special_tokens:l=!0,decode_kwargs:c={}}={}){super(e,{skip_prompt:t,skip_special_tokens:l,callback_function:s,token_callback_function:r,decode_kwargs:c}),this.timestamp_begin=e.timestamp_begin,this.on_chunk_start=n,this.on_chunk_end=o,this.on_finalize=i,this.time_precision=a,this.waiting_for_timestamp=!1}put(e){if(e.length>1)throw Error("WhisperTextStreamer only supports batch size of 1");const t=e[0];if(1===t.length){const s=Number(t[0])-this.timestamp_begin;if(s>=0){const t=s*this.time_precision;this.waiting_for_timestamp?this.on_chunk_end?.(t):this.on_chunk_start?.(t),this.waiting_for_timestamp=!this.waiting_for_timestamp,e=[[]]}}return super.put(e)}end(){super.end(),this.on_finalize?.()}}},"./src/models.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ASTForAudioClassification:()=>hs,ASTModel:()=>ms,ASTPreTrainedModel:()=>ps,AlbertForMaskedLM:()=>Tt,AlbertForQuestionAnswering:()=>kt,AlbertForSequenceClassification:()=>yt,AlbertModel:()=>xt,AlbertPreTrainedModel:()=>bt,AutoModel:()=>tc,AutoModelForAudioClassification:()=>bc,AutoModelForAudioFrameClassification:()=>yc,AutoModelForCTC:()=>Mc,AutoModelForCausalLM:()=>lc,AutoModelForDepthEstimation:()=>Pc,AutoModelForDocumentQuestionAnswering:()=>kc,AutoModelForImageClassification:()=>_c,AutoModelForImageFeatureExtraction:()=>Ac,AutoModelForImageMatting:()=>Tc,AutoModelForImageSegmentation:()=>pc,AutoModelForImageToImage:()=>vc,AutoModelForMaskGeneration:()=>wc,AutoModelForMaskedLM:()=>cc,AutoModelForNormalEstimation:()=>Fc,AutoModelForObjectDetection:()=>fc,AutoModelForPoseEstimation:()=>Cc,AutoModelForQuestionAnswering:()=>dc,AutoModelForSemanticSegmentation:()=>mc,AutoModelForSeq2SeqLM:()=>nc,AutoModelForSequenceClassification:()=>sc,AutoModelForSpeechSeq2Seq:()=>oc,AutoModelForTextToSpectrogram:()=>ic,AutoModelForTextToWaveform:()=>ac,AutoModelForTokenClassification:()=>rc,AutoModelForUniversalSegmentation:()=>hc,AutoModelForVision2Seq:()=>uc,AutoModelForXVector:()=>xc,AutoModelForZeroShotObjectDetection:()=>gc,BartForConditionalGeneration:()=>Nt,BartForSequenceClassification:()=>Ot,BartModel:()=>jt,BartPretrainedModel:()=>zt,BaseModelOutput:()=>Y,BeitForImageClassification:()=>ao,BeitModel:()=>io,BeitPreTrainedModel:()=>oo,BertForMaskedLM:()=>ee,BertForQuestionAnswering:()=>re,BertForSequenceClassification:()=>te,BertForTokenClassification:()=>se,BertModel:()=>Z,BertPreTrainedModel:()=>K,BlenderbotForConditionalGeneration:()=>Ut,BlenderbotModel:()=>qt,BlenderbotPreTrainedModel:()=>$t,BlenderbotSmallForConditionalGeneration:()=>Qt,BlenderbotSmallModel:()=>Xt,BlenderbotSmallPreTrainedModel:()=>Wt,BloomForCausalLM:()=>gn,BloomModel:()=>fn,BloomPreTrainedModel:()=>hn,CLIPModel:()=>Ns,CLIPPreTrainedModel:()=>js,CLIPSegForImageSegmentation:()=>Zs,CLIPSegModel:()=>Ks,CLIPSegPreTrainedModel:()=>Ys,CLIPTextModel:()=>Os,CLIPTextModelWithProjection:()=>Bs,CLIPVisionModel:()=>Vs,CLIPVisionModelWithProjection:()=>Ds,CamembertForMaskedLM:()=>Ee,CamembertForQuestionAnswering:()=>ze,CamembertForSequenceClassification:()=>Ie,CamembertForTokenClassification:()=>Le,CamembertModel:()=>Se,CamembertPreTrainedModel:()=>Ae,CausalLMOutput:()=>Nc,CausalLMOutputWithPast:()=>Oc,ChineseCLIPModel:()=>Ws,ChineseCLIPPreTrainedModel:()=>Us,ClapAudioModelWithProjection:()=>Sa,ClapModel:()=>Ca,ClapPreTrainedModel:()=>Fa,ClapTextModelWithProjection:()=>Aa,CodeGenForCausalLM:()=>br,CodeGenModel:()=>Mr,CodeGenPreTrainedModel:()=>wr,CohereForCausalLM:()=>Xr,CohereModel:()=>Wr,CoherePreTrainedModel:()=>Ur,ConvBertForMaskedLM:()=>Me,ConvBertForQuestionAnswering:()=>ye,ConvBertForSequenceClassification:()=>be,ConvBertForTokenClassification:()=>xe,ConvBertModel:()=>we,ConvBertPreTrainedModel:()=>ge,ConvNextForImageClassification:()=>oi,ConvNextModel:()=>ni,ConvNextPreTrainedModel:()=>ri,ConvNextV2ForImageClassification:()=>li,ConvNextV2Model:()=>ai,ConvNextV2PreTrainedModel:()=>ii,DPTForDepthEstimation:()=>Do,DPTModel:()=>Vo,DPTPreTrainedModel:()=>Bo,DebertaForMaskedLM:()=>Oe,DebertaForQuestionAnswering:()=>De,DebertaForSequenceClassification:()=>Be,DebertaForTokenClassification:()=>Ve,DebertaModel:()=>Ne,DebertaPreTrainedModel:()=>je,DebertaV2ForMaskedLM:()=>$e,DebertaV2ForQuestionAnswering:()=>We,DebertaV2ForSequenceClassification:()=>qe,DebertaV2ForTokenClassification:()=>Ue,DebertaV2Model:()=>Ge,DebertaV2PreTrainedModel:()=>Re,DecisionTransformerModel:()=>il,DecisionTransformerPreTrainedModel:()=>ol,DeiTForImageClassification:()=>vo,DeiTModel:()=>To,DeiTPreTrainedModel:()=>ko,DepthAnythingForDepthEstimation:()=>Go,DepthAnythingPreTrainedModel:()=>Ro,DepthProForDepthEstimation:()=>Qo,DepthProPreTrainedModel:()=>Xo,DetrForObjectDetection:()=>uo,DetrForSegmentation:()=>_o,DetrModel:()=>co,DetrObjectDetectionOutput:()=>po,DetrPreTrainedModel:()=>lo,DetrSegmentationOutput:()=>mo,Dinov2ForImageClassification:()=>ui,Dinov2Model:()=>di,Dinov2PreTrainedModel:()=>ci,Dinov2WithRegistersForImageClassification:()=>mi,Dinov2WithRegistersModel:()=>pi,Dinov2WithRegistersPreTrainedModel:()=>_i,DistilBertForMaskedLM:()=>Ke,DistilBertForQuestionAnswering:()=>Ye,DistilBertForSequenceClassification:()=>He,DistilBertForTokenClassification:()=>Je,DistilBertModel:()=>Qe,DistilBertPreTrainedModel:()=>Xe,DonutSwinModel:()=>si,DonutSwinPreTrainedModel:()=>ti,EfficientNetForImageClassification:()=>Ga,EfficientNetModel:()=>Ra,EfficientNetPreTrainedModel:()=>Da,ElectraForMaskedLM:()=>ve,ElectraForQuestionAnswering:()=>Ce,ElectraForSequenceClassification:()=>Pe,ElectraForTokenClassification:()=>Fe,ElectraModel:()=>Te,ElectraPreTrainedModel:()=>ke,EsmForMaskedLM:()=>tt,EsmForSequenceClassification:()=>st,EsmForTokenClassification:()=>rt,EsmModel:()=>et,EsmPreTrainedModel:()=>Ze,ExaoneForCausalLM:()=>Ir,ExaoneModel:()=>Er,ExaonePreTrainedModel:()=>Sr,FalconForCausalLM:()=>Pa,FalconModel:()=>va,FalconPreTrainedModel:()=>Ta,FastViTForImageClassification:()=>qn,FastViTModel:()=>$n,FastViTPreTrainedModel:()=>Gn,Florence2ForConditionalGeneration:()=>Cs,Florence2PreTrainedModel:()=>Fs,GLPNForDepthEstimation:()=>ei,GLPNModel:()=>Zo,GLPNPreTrainedModel:()=>Ko,GPT2LMHeadModel:()=>sr,GPT2Model:()=>tr,GPT2PreTrainedModel:()=>er,GPTBigCodeForCausalLM:()=>gr,GPTBigCodeModel:()=>fr,GPTBigCodePreTrainedModel:()=>hr,GPTJForCausalLM:()=>mr,GPTJModel:()=>pr,GPTJPreTrainedModel:()=>_r,GPTNeoForCausalLM:()=>lr,GPTNeoModel:()=>ar,GPTNeoPreTrainedModel:()=>ir,GPTNeoXForCausalLM:()=>ur,GPTNeoXModel:()=>dr,GPTNeoXPreTrainedModel:()=>cr,Gemma2ForCausalLM:()=>Zr,Gemma2Model:()=>Kr,Gemma2PreTrainedModel:()=>Yr,GemmaForCausalLM:()=>Jr,GemmaModel:()=>Hr,GemmaPreTrainedModel:()=>Qr,GlmForCausalLM:()=>Ar,GlmModel:()=>Cr,GlmPreTrainedModel:()=>Fr,GraniteForCausalLM:()=>qr,GraniteModel:()=>$r,GranitePreTrainedModel:()=>Gr,GroundingDinoForObjectDetection:()=>fi,GroundingDinoPreTrainedModel:()=>hi,GroupViTModel:()=>Rn,GroupViTPreTrainedModel:()=>Dn,HeliumForCausalLM:()=>Pr,HeliumModel:()=>vr,HeliumPreTrainedModel:()=>Tr,HieraForImageClassification:()=>Co,HieraModel:()=>Fo,HieraPreTrainedModel:()=>Po,HubertForCTC:()=>ta,HubertForSequenceClassification:()=>sa,HubertModel:()=>ea,HubertPreTrainedModel:()=>Zi,IJepaForImageClassification:()=>An,IJepaModel:()=>Cn,IJepaPreTrainedModel:()=>Fn,Idefics3ForConditionalGeneration:()=>Is,Idefics3PreTrainedModel:()=>Es,ImageMattingOutput:()=>Bc,JAISLMHeadModel:()=>or,JAISModel:()=>nr,JAISPreTrainedModel:()=>rr,JinaCLIPModel:()=>Qs,JinaCLIPPreTrainedModel:()=>Xs,JinaCLIPTextModel:()=>Hs,JinaCLIPVisionModel:()=>Js,LlamaForCausalLM:()=>kr,LlamaModel:()=>yr,LlamaPreTrainedModel:()=>xr,LlavaForConditionalGeneration:()=>Ts,LlavaOnevisionForConditionalGeneration:()=>vs,LlavaPreTrainedModel:()=>ks,LongT5ForConditionalGeneration:()=>St,LongT5Model:()=>At,LongT5PreTrainedModel:()=>Ct,M2M100ForConditionalGeneration:()=>Ai,M2M100Model:()=>Ci,M2M100PreTrainedModel:()=>Fi,MBartForCausalLM:()=>Gt,MBartForConditionalGeneration:()=>Dt,MBartForSequenceClassification:()=>Rt,MBartModel:()=>Vt,MBartPreTrainedModel:()=>Bt,MPNetForMaskedLM:()=>ut,MPNetForQuestionAnswering:()=>mt,MPNetForSequenceClassification:()=>_t,MPNetForTokenClassification:()=>pt,MPNetModel:()=>dt,MPNetPreTrainedModel:()=>ct,MT5ForConditionalGeneration:()=>Lt,MT5Model:()=>It,MT5PreTrainedModel:()=>Et,MarianMTModel:()=>Pi,MarianModel:()=>vi,MarianPreTrainedModel:()=>Ti,MaskFormerForInstanceSegmentation:()=>Yo,MaskFormerModel:()=>Jo,MaskFormerPreTrainedModel:()=>Ho,MaskedLMOutput:()=>zc,MgpstrForSceneTextRecognition:()=>ul,MgpstrModelOutput:()=>cl,MgpstrPreTrainedModel:()=>dl,MistralForCausalLM:()=>ba,MistralModel:()=>Ma,MistralPreTrainedModel:()=>wa,MobileBertForMaskedLM:()=>it,MobileBertForQuestionAnswering:()=>lt,MobileBertForSequenceClassification:()=>at,MobileBertModel:()=>ot,MobileBertPreTrainedModel:()=>nt,MobileLLMForCausalLM:()=>jr,MobileLLMModel:()=>zr,MobileLLMPreTrainedModel:()=>Lr,MobileNetV1ForImageClassification:()=>Ha,MobileNetV1Model:()=>Qa,MobileNetV1PreTrainedModel:()=>Xa,MobileNetV2ForImageClassification:()=>Ka,MobileNetV2Model:()=>Ya,MobileNetV2PreTrainedModel:()=>Ja,MobileNetV3ForImageClassification:()=>tl,MobileNetV3Model:()=>el,MobileNetV3PreTrainedModel:()=>Za,MobileNetV4ForImageClassification:()=>nl,MobileNetV4Model:()=>rl,MobileNetV4PreTrainedModel:()=>sl,MobileViTForImageClassification:()=>Hn,MobileViTModel:()=>Qn,MobileViTPreTrainedModel:()=>Xn,MobileViTV2ForImageClassification:()=>Kn,MobileViTV2Model:()=>Yn,MobileViTV2PreTrainedModel:()=>Jn,ModelOutput:()=>J,ModernBertForMaskedLM:()=>ie,ModernBertForSequenceClassification:()=>ae,ModernBertForTokenClassification:()=>le,ModernBertModel:()=>oe,ModernBertPreTrainedModel:()=>ne,Moondream1ForConditionalGeneration:()=>Ps,MoonshineForConditionalGeneration:()=>xs,MoonshineModel:()=>bs,MoonshinePreTrainedModel:()=>Ms,MptForCausalLM:()=>bn,MptModel:()=>Mn,MptPreTrainedModel:()=>wn,MultiModalityCausalLM:()=>ll,MultiModalityPreTrainedModel:()=>al,MusicgenForCausalLM:()=>Ua,MusicgenForConditionalGeneration:()=>Wa,MusicgenModel:()=>qa,MusicgenPreTrainedModel:()=>$a,NomicBertModel:()=>de,NomicBertPreTrainedModel:()=>ce,OPTForCausalLM:()=>kn,OPTModel:()=>yn,OPTPreTrainedModel:()=>xn,Olmo2ForCausalLM:()=>Rr,Olmo2Model:()=>Dr,Olmo2PreTrainedModel:()=>Vr,OlmoForCausalLM:()=>Br,OlmoModel:()=>Or,OlmoPreTrainedModel:()=>Nr,OpenELMForCausalLM:()=>sn,OpenELMModel:()=>tn,OpenELMPreTrainedModel:()=>en,OwlViTForObjectDetection:()=>to,OwlViTModel:()=>eo,OwlViTPreTrainedModel:()=>Zn,Owlv2ForObjectDetection:()=>no,Owlv2Model:()=>ro,Owlv2PreTrainedModel:()=>so,PaliGemmaForConditionalGeneration:()=>Ss,PaliGemmaPreTrainedModel:()=>As,PatchTSMixerForPrediction:()=>gl,PatchTSMixerModel:()=>fl,PatchTSMixerPreTrainedModel:()=>hl,PatchTSTForPrediction:()=>ml,PatchTSTModel:()=>pl,PatchTSTPreTrainedModel:()=>_l,Phi3ForCausalLM:()=>mn,Phi3Model:()=>pn,Phi3PreTrainedModel:()=>_n,Phi3VForCausalLM:()=>zs,Phi3VPreTrainedModel:()=>Ls,PhiForCausalLM:()=>un,PhiModel:()=>dn,PhiPreTrainedModel:()=>cn,PreTrainedModel:()=>H,PretrainedMixin:()=>wl,PvtForImageClassification:()=>zn,PvtModel:()=>Ln,PvtPreTrainedModel:()=>In,PyAnnoteForAudioFrameClassification:()=>Oi,PyAnnoteModel:()=>Ni,PyAnnotePreTrainedModel:()=>ji,QuestionAnsweringModelOutput:()=>jc,Qwen2ForCausalLM:()=>on,Qwen2Model:()=>nn,Qwen2PreTrainedModel:()=>rn,Qwen2VLForConditionalGeneration:()=>ln,Qwen2VLPreTrainedModel:()=>an,RTDetrForObjectDetection:()=>go,RTDetrModel:()=>fo,RTDetrObjectDetectionOutput:()=>wo,RTDetrPreTrainedModel:()=>ho,ResNetForImageClassification:()=>Eo,ResNetModel:()=>So,ResNetPreTrainedModel:()=>Ao,RoFormerForMaskedLM:()=>pe,RoFormerForQuestionAnswering:()=>fe,RoFormerForSequenceClassification:()=>me,RoFormerForTokenClassification:()=>he,RoFormerModel:()=>_e,RoFormerPreTrainedModel:()=>ue,RobertaForMaskedLM:()=>Yt,RobertaForQuestionAnswering:()=>es,RobertaForSequenceClassification:()=>Kt,RobertaForTokenClassification:()=>Zt,RobertaModel:()=>Jt,RobertaPreTrainedModel:()=>Ht,SamImageSegmentationOutput:()=>ki,SamModel:()=>yi,SamPreTrainedModel:()=>xi,SapiensForDepthEstimation:()=>Uo,SapiensForNormalEstimation:()=>Wo,SapiensForSemanticSegmentation:()=>qo,SapiensPreTrainedModel:()=>$o,SegformerForImageClassification:()=>ja,SegformerForSemanticSegmentation:()=>Na,SegformerModel:()=>za,SegformerPreTrainedModel:()=>La,Seq2SeqLMOutput:()=>Sc,SequenceClassifierOutput:()=>Ec,SiglipModel:()=>Gs,SiglipPreTrainedModel:()=>Rs,SiglipTextModel:()=>$s,SiglipVisionModel:()=>qs,SpeechT5ForSpeechToText:()=>pa,SpeechT5ForTextToSpeech:()=>ma,SpeechT5HifiGan:()=>ha,SpeechT5Model:()=>_a,SpeechT5PreTrainedModel:()=>ua,SqueezeBertForMaskedLM:()=>gt,SqueezeBertForQuestionAnswering:()=>Mt,SqueezeBertForSequenceClassification:()=>wt,SqueezeBertModel:()=>ft,SqueezeBertPreTrainedModel:()=>ht,StableLmForCausalLM:()=>Va,StableLmModel:()=>Ba,StableLmPreTrainedModel:()=>Oa,Starcoder2ForCausalLM:()=>ka,Starcoder2Model:()=>ya,Starcoder2PreTrainedModel:()=>xa,StyleTextToSpeech2Model:()=>da,StyleTextToSpeech2PreTrainedModel:()=>ca,Swin2SRForImageSuperResolution:()=>Oo,Swin2SRModel:()=>No,Swin2SRPreTrainedModel:()=>jo,SwinForImageClassification:()=>zo,SwinModel:()=>Lo,SwinPreTrainedModel:()=>Io,T5ForConditionalGeneration:()=>Ft,T5Model:()=>Pt,T5PreTrainedModel:()=>vt,TableTransformerForObjectDetection:()=>xo,TableTransformerModel:()=>bo,TableTransformerObjectDetectionOutput:()=>yo,TableTransformerPreTrainedModel:()=>Mo,TokenClassifierOutput:()=>Lc,TrOCRForCausalLM:()=>ga,TrOCRPreTrainedModel:()=>fa,UniSpeechForCTC:()=>Gi,UniSpeechForSequenceClassification:()=>$i,UniSpeechModel:()=>Ri,UniSpeechPreTrainedModel:()=>Di,UniSpeechSatForAudioFrameClassification:()=>Qi,UniSpeechSatForCTC:()=>Wi,UniSpeechSatForSequenceClassification:()=>Xi,UniSpeechSatModel:()=>Ui,UniSpeechSatPreTrainedModel:()=>qi,ViTForImageClassification:()=>Pn,ViTMAEModel:()=>Nn,ViTMAEPreTrainedModel:()=>jn,ViTMSNForImageClassification:()=>Vn,ViTMSNModel:()=>Bn,ViTMSNPreTrainedModel:()=>On,ViTModel:()=>vn,ViTPreTrainedModel:()=>Tn,VisionEncoderDecoderModel:()=>ys,VitMatteForImageMatting:()=>Wn,VitMattePreTrainedModel:()=>Un,VitPoseForPoseEstimation:()=>En,VitPosePreTrainedModel:()=>Sn,VitsModel:()=>Ia,VitsModelOutput:()=>Vc,VitsPreTrainedModel:()=>Ea,Wav2Vec2BertForCTC:()=>Yi,Wav2Vec2BertForSequenceClassification:()=>Ki,Wav2Vec2BertModel:()=>Ji,Wav2Vec2BertPreTrainedModel:()=>Hi,Wav2Vec2ForAudioFrameClassification:()=>zi,Wav2Vec2ForCTC:()=>Ii,Wav2Vec2ForSequenceClassification:()=>Li,Wav2Vec2Model:()=>Ei,Wav2Vec2PreTrainedModel:()=>Si,WavLMForAudioFrameClassification:()=>la,WavLMForCTC:()=>oa,WavLMForSequenceClassification:()=>ia,WavLMForXVector:()=>aa,WavLMModel:()=>na,WavLMPreTrainedModel:()=>ra,WeSpeakerResNetModel:()=>Vi,WeSpeakerResNetPreTrainedModel:()=>Bi,WhisperForConditionalGeneration:()=>ws,WhisperModel:()=>gs,WhisperPreTrainedModel:()=>fs,XLMForQuestionAnswering:()=>is,XLMForSequenceClassification:()=>ns,XLMForTokenClassification:()=>os,XLMModel:()=>ss,XLMPreTrainedModel:()=>ts,XLMRobertaForMaskedLM:()=>cs,XLMRobertaForQuestionAnswering:()=>_s,XLMRobertaForSequenceClassification:()=>ds,XLMRobertaForTokenClassification:()=>us,XLMRobertaModel:()=>ls,XLMRobertaPreTrainedModel:()=>as,XLMWithLMHeadModel:()=>rs,XVectorOutput:()=>Ic,YolosForObjectDetection:()=>Mi,YolosModel:()=>wi,YolosObjectDetectionOutput:()=>bi,YolosPreTrainedModel:()=>gi});var r=s("./src/configs.js"),n=s("./src/backends/onnx.js"),o=s("./src/utils/dtypes.js"),i=s("./src/utils/generic.js"),a=s("./src/utils/core.js"),l=s("./src/utils/hub.js"),c=s("./src/utils/constants.js"),d=s("./src/generation/logits_process.js"),u=s("./src/generation/configuration_utils.js"),_=s("./src/utils/tensor.js"),p=s("./src/utils/image.js"),m=s("./src/utils/maths.js"),h=s("./src/generation/stopping_criteria.js"),f=s("./src/generation/logits_sampler.js"),g=s("./src/env.js"),w=s("./src/models/whisper/generation_whisper.js"),M=s("./src/models/whisper/common_whisper.js");const b=0,x=1,y=2,k=3,T=4,v=5,P=6,F=7,C=8,A=9,S=new Map,E=new Map,I=new Map;async function L(e,t,s){return Object.fromEntries(await Promise.all(Object.keys(t).map((async i=>{const{buffer:a,session_options:c,session_config:d}=await async function(e,t,s){const i=s.config?.["transformers.js_config"]??{};let a=s.device??i.device;a&&"string"!=typeof a&&(a.hasOwnProperty(t)?a=a[t]:(console.warn(`device not specified for "${t}". Using the default device.`),a=null));const c=a??(g.apis.IS_NODE_ENV?"cpu":"wasm"),d=(0,n.deviceToExecutionProviders)(c);let u=s.dtype??i.dtype;if("string"!=typeof u&&(u&&u.hasOwnProperty(t)?u=u[t]:(u=o.DEFAULT_DEVICE_DTYPE_MAPPING[c]??o.DATA_TYPES.fp32,console.warn(`dtype not specified for "${t}". Using the default dtype (${u}) for this device (${c}).`))),u===o.DATA_TYPES.auto){let e=i.dtype;"string"!=typeof e&&(e=e[t]),u=e&&e!==o.DATA_TYPES.auto&&o.DATA_TYPES.hasOwnProperty(e)?e:o.DEFAULT_DEVICE_DTYPE_MAPPING[c]??o.DATA_TYPES.fp32}const _=u;if(!o.DEFAULT_DTYPE_SUFFIX_MAPPING.hasOwnProperty(_))throw new Error(`Invalid dtype: ${_}. Should be one of: ${Object.keys(o.DATA_TYPES).join(", ")}`);if(_===o.DATA_TYPES.fp16&&"webgpu"===c&&!await(0,o.isWebGpuFp16Supported)())throw new Error(`The device (${c}) does not support fp16.`);const p=i.kv_cache_dtype?"string"==typeof i.kv_cache_dtype?i.kv_cache_dtype:i.kv_cache_dtype[_]??"float32":void 0;if(p&&!["float32","float16"].includes(p))throw new Error(`Invalid kv_cache_dtype: ${p}. Should be one of: float32, float16`);const m={dtype:_,kv_cache_dtype:p},h=o.DEFAULT_DTYPE_SUFFIX_MAPPING[_],f=`${s.subfolder??""}/${t}${h}.onnx`,w={...s.session_options};w.executionProviders??=d;const M=i.free_dimension_overrides;M?w.freeDimensionOverrides??=M:c.startsWith("webnn")&&!w.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 b=(0,l.getModelFile)(e,f,!0,s),x=s.use_external_data_format??i.use_external_data_format;let y=[];if(x&&(!0===x||"object"==typeof x&&x.hasOwnProperty(t)&&!0===x[t])){if(g.apis.IS_NODE_ENV)throw new Error("External data format is not yet supported in Node.js");const r=`${t}${h}.onnx_data`,n=`${s.subfolder??""}/${r}`;y.push(new Promise((async(t,o)=>{const i=await(0,l.getModelFile)(e,n,!0,s);t({path:r,data:i})})))}else void 0!==w.externalData&&(y=w.externalData.map((async t=>{if("string"==typeof t.data){const r=await(0,l.getModelFile)(e,t.data,!0,s);return{...t,data:r}}return t})));if(y.length>0&&(w.externalData=await Promise.all(y)),"webgpu"===c){const e=(0,r.getKeyValueShapes)(s.config,{prefix:"present"});if(Object.keys(e).length>0&&!(0,n.isONNXProxy)()){const t={};for(const s in e)t[s]="gpu-buffer";w.preferredOutputLocation=t}}return{buffer:await b,session_options:w,session_config:m}}(e,t[i],s);return[i,await(0,n.createInferenceSession)(a,c,d)]}))))}async function z(e,t,s){return Object.fromEntries(await Promise.all(Object.keys(t).map((async r=>[r,await(0,l.getModelJSON)(e,t[r],!1,s)]))))}async function j(e,t){const s=function(e,t){const s=Object.create(null),r=[];for(const o of e.inputNames){const e=t[o];e instanceof _.Tensor?s[o]=(0,n.isONNXProxy)()?e.clone():e:r.push(o)}if(r.length>0)throw new Error(`An error occurred during model execution: "Missing the following inputs: ${r.join(", ")}.`);const o=Object.keys(t).length,i=e.inputNames.length;if(o>i){let s=Object.keys(t).filter((t=>!e.inputNames.includes(t)));console.warn(`WARNING: Too many inputs were provided (${o} > ${i}). The following inputs will be ignored: "${s.join(", ")}".`)}return s}(e,t);try{const t=Object.fromEntries(Object.entries(s).map((([e,t])=>[e,t.ort_tensor])));let r=await e.run(t);return r=N(r),r}catch(e){const t=Object.fromEntries(Object.entries(s).map((([e,{type:t,dims:s,data:r}])=>[e,{type:t,dims:s,data:r}])));throw console.error(`An error occurred during model execution: "${e}".`),console.error("Inputs given to model:",t),e}}function N(e){for(let t in e)(0,n.isONNXTensor)(e[t])?e[t]=new _.Tensor(e[t]):"object"==typeof e[t]&&N(e[t]);return e}function O(e){if(e instanceof _.Tensor)return e;if(0===e.length)throw Error("items must be non-empty");if(Array.isArray(e[0])){if(e.some((t=>t.length!==e[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 _.Tensor("int64",BigInt64Array.from(e.flat().map((e=>BigInt(e)))),[e.length,e[0].length])}return new _.Tensor("int64",BigInt64Array.from(e.map((e=>BigInt(e)))),[1,e.length])}function B(e){return new _.Tensor("bool",[e],[1])}async function V(e,t){let{encoder_outputs:s,input_ids:r,decoder_input_ids:n,...o}=t;if(!s){const r=(0,a.pick)(t,e.sessions.model.inputNames);s=(await D(e,r)).last_hidden_state}o.input_ids=n,o.encoder_hidden_states=s,e.sessions.decoder_model_merged.inputNames.includes("encoder_attention_mask")&&(o.encoder_attention_mask=t.attention_mask);return await R(e,o,!0)}async function D(e,t){const s=e.sessions.model,r=(0,a.pick)(t,s.inputNames);if(s.inputNames.includes("inputs_embeds")&&!r.inputs_embeds){if(!t.input_ids)throw new Error("Both `input_ids` and `inputs_embeds` are missing in the model inputs.");r.inputs_embeds=await e.encode_text({input_ids:t.input_ids})}if(s.inputNames.includes("token_type_ids")&&!r.token_type_ids){if(!r.input_ids)throw new Error("Both `input_ids` and `token_type_ids` are missing in the model inputs.");r.token_type_ids=(0,_.zeros_like)(r.input_ids)}if(s.inputNames.includes("pixel_mask")&&!r.pixel_mask){if(!r.pixel_values)throw new Error("Both `pixel_values` and `pixel_mask` are missing in the model inputs.");const e=r.pixel_values.dims;r.pixel_mask=(0,_.ones)([e[0],e[2],e[3]])}return await j(s,r)}async function R(e,t,s=!1){const r=e.sessions[s?"decoder_model_merged":"model"],{past_key_values:n,...o}=t;if(r.inputNames.includes("use_cache_branch")&&(o.use_cache_branch=B(!!n)),r.inputNames.includes("position_ids")&&o.attention_mask&&!o.position_ids){const t="paligemma"===e.config.model_type?1:0;o.position_ids=function(e,t=null,s=0){const{input_ids:r,inputs_embeds:n,attention_mask:o}=e,{data:i,dims:a}=q(o,s);let l=new _.Tensor("int64",i,a);if(t){const e=-(r??n).dims.at(1);l=l.slice(null,[e,null])}return l}(o,n,t)}e.addPastKeyValues(o,n);const i=(0,a.pick)(o,r.inputNames);return await j(r,i)}function G({image_token_id:e,inputs_embeds:t,image_features:s,input_ids:r,attention_mask:n}){const o=r.tolist().map((t=>t.reduce(((t,s,r)=>(s==e&&t.push(r),t)),[]))),i=o.reduce(((e,t)=>e+t.length),0),a=s.dims[0];if(i!==a)throw new Error(`Image features and image tokens do not match: tokens: ${i}, features ${a}`);let l=0;for(let e=0;e<o.length;++e){const r=o[e],n=t[e];for(let e=0;e<r.length;++e)n[r[e]].data.set(s[l++].data)}return{inputs_embeds:t,attention_mask:n}}async function $(e,{input_ids:t=null,attention_mask:s=null,pixel_values:r=null,position_ids:n=null,inputs_embeds:o=null,past_key_values:i=null,generation_config:a=null,logits_processor:l=null,...c}){if(!o)if(o=await e.encode_text({input_ids:t,...c}),r&&1!==t.dims[1]){const n=await e.encode_image({pixel_values:r,...c});({inputs_embeds:o,attention_mask:s}=e._merge_input_ids_with_image_features({image_features:n,inputs_embeds:o,input_ids:t,attention_mask:s}))}else if(i&&r&&1===t.dims[1]){const e=t.dims[1],r=Object.values(i)[0].dims.at(-2);s=(0,_.cat)([(0,_.ones)([t.dims[0],r]),s.slice(null,[s.dims[1]-e,s.dims[1]])],1)}if(!n&&"qwen2_vl"===e.config.model_type){const{image_grid_thw:r,video_grid_thw:o}=c;[n]=e.get_rope_index(t,r,o,s)}return await R(e,{inputs_embeds:o,past_key_values:i,attention_mask:s,position_ids:n,generation_config:a,logits_processor:l},!0)}function q(e,t=0){const[s,r]=e.dims,n=e.data,o=new BigInt64Array(n.length);for(let e=0;e<s;++e){const s=e*r;let i=BigInt(t);for(let e=0;e<r;++e){const t=s+e;0n===n[t]?o[t]=BigInt(1):(o[t]=i,i+=n[t])}}return{data:o,dims:e.dims}}function U(e,t,s,r){if(s.past_key_values){const t=Object.values(s.past_key_values)[0].dims.at(-2),{input_ids:r,attention_mask:n}=s;if(n&&n.dims[1]>r.dims[1]);else if(t<r.dims[1])s.input_ids=r.slice(null,[t,null]);else if(null!=e.config.image_token_index&&r.data.some((t=>t==e.config.image_token_index))){const n=e.config.num_image_tokens;if(!n)throw new Error("`num_image_tokens` is missing in the model configuration.");const o=r.dims[1]-(t-n);s.input_ids=r.slice(null,[-o,null]),s.attention_mask=(0,_.ones)([1,t+o])}}return s}function W(e,t,s,r){return s.past_key_values&&(t=t.map((e=>[e.at(-1)]))),{...s,decoder_input_ids:O(t)}}function X(e,...t){return e.config.is_encoder_decoder?W(e,...t):U(e,...t)}function Q(e,t,s,r){const n=!!s.past_key_values;if(null!==r.guidance_scale&&r.guidance_scale>1&&(n?s.input_ids=(0,_.cat)([s.input_ids,s.input_ids],0):(s.input_ids=(0,_.cat)([s.input_ids,(0,_.full_like)(s.input_ids,BigInt(r.pad_token_id))],0),s.attention_mask=(0,_.cat)([s.attention_mask,(0,_.full_like)(s.attention_mask,0n)],0))),!n&&s.pixel_values||(s.pixel_values=(0,_.full)([0,0,3,384,384],1)),n){const e=0,t=1,r=e>0?1:0,n=1;s.images_seq_mask=new _.Tensor("bool",new Array(e+t).fill(!0).fill(!1,0,t),[n,e+t]),s.images_emb_mask=new _.Tensor("bool",new Array(e).fill(!!r),[n,1,e])}return s}class H extends i.Callable{main_input_name="input_ids";forward_params=["input_ids","attention_mask"];constructor(e,t,s){super(),this.config=e,this.sessions=t,this.configs=s;const r=I.get(this.constructor),n=S.get(r);switch(this.can_generate=!1,this._forward=null,this._prepare_inputs_for_generation=null,n){case T:this.can_generate=!0,this._forward=R,this._prepare_inputs_for_generation=U;break;case y:case k:case F:this.can_generate=!0,this._forward=V,this._prepare_inputs_for_generation=W;break;case x:this._forward=V;break;case P:this.can_generate=!0,this._forward=$,this._prepare_inputs_for_generation=X;break;case A:this.can_generate=!0,this._prepare_inputs_for_generation=X;break;case C:this.can_generate=!0,this._prepare_inputs_for_generation=Q;break;default:this._forward=D}this.can_generate&&this.forward_params.push("past_key_values"),this.custom_config=this.config["transformers.js_config"]??{}}async dispose(){const e=[];for(const t of Object.values(this.sessions))t?.handler?.dispose&&e.push(t.handler.dispose());return await Promise.all(e)}static async from_pretrained(e,{progress_callback:t=null,config:s=null,cache_dir:n=null,local_files_only:o=!1,revision:i="main",model_file_name:a=null,subfolder:l="onnx",device:d=null,dtype:u=null,use_external_data_format:_=null,session_options:p={}}={}){let m={progress_callback:t,config:s,cache_dir:n,local_files_only:o,revision:i,model_file_name:a,subfolder:l,device:d,dtype:u,use_external_data_format:_,session_options:p};const h=I.get(this),f=S.get(h);let g;if(s=m.config=await r.AutoConfig.from_pretrained(e,m),f===T)g=await Promise.all([L(e,{model:m.model_file_name??"model"},m),z(e,{generation_config:"generation_config.json"},m)]);else if(f===y||f===k)g=await Promise.all([L(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},m),z(e,{generation_config:"generation_config.json"},m)]);else if(f===v)g=await Promise.all([L(e,{model:"vision_encoder",prompt_encoder_mask_decoder:"prompt_encoder_mask_decoder"},m)]);else if(f===x)g=await Promise.all([L(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},m)]);else if(f===P){const t={embed_tokens:"embed_tokens",vision_encoder:"vision_encoder",decoder_model_merged:"decoder_model_merged"};s.is_encoder_decoder&&(t.model="encoder_model"),g=await Promise.all([L(e,t,m),z(e,{generation_config:"generation_config.json"},m)])}else if(f===F)g=await Promise.all([L(e,{model:"text_encoder",decoder_model_merged:"decoder_model_merged",encodec_decode:"encodec_decode"},m),z(e,{generation_config:"generation_config.json"},m)]);else if(f===C)g=await Promise.all([L(e,{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"},m),z(e,{generation_config:"generation_config.json"},m)]);else if(f===A)g=await Promise.all([L(e,{prepare_inputs_embeds:"prepare_inputs_embeds",model:"model",vision_encoder:"vision_encoder"},m),z(e,{generation_config:"generation_config.json"},m)]);else{if(f!==b){const e=h??s?.model_type;"custom"!==e&&console.warn(`Model type for '${e}' not found, assuming encoder-only architecture. Please report this at ${c.GITHUB_ISSUE_URL}.`)}g=await Promise.all([L(e,{model:m.model_file_name??"model"},m)])}return new this(s,...g)}async _call(e){return await this.forward(e)}async forward(e){return await this._forward(this,e)}get generation_config(){return this.configs?.generation_config??null}_get_logits_warper(e){const t=new d.LogitsProcessorList;return null!==e.temperature&&1!==e.temperature&&t.push(new d.TemperatureLogitsWarper(e.temperature)),null!==e.top_k&&0!==e.top_k&&t.push(new d.TopKLogitsWarper(e.top_k)),null!==e.top_p&&e.top_p<1&&t.push(new d.TopPLogitsWarper(e.top_p)),t}_get_logits_processor(e,t,s=null){const r=new d.LogitsProcessorList;if(null!==e.repetition_penalty&&1!==e.repetition_penalty&&r.push(new d.RepetitionPenaltyLogitsProcessor(e.repetition_penalty)),null!==e.no_repeat_ngram_size&&e.no_repeat_ngram_size>0&&r.push(new d.NoRepeatNGramLogitsProcessor(e.no_repeat_ngram_size)),null!==e.bad_words_ids&&r.push(new d.NoBadWordsLogitsProcessor(e.bad_words_ids,e.eos_token_id)),null!==e.min_length&&null!==e.eos_token_id&&e.min_length>0&&r.push(new d.MinLengthLogitsProcessor(e.min_length,e.eos_token_id)),null!==e.min_new_tokens&&null!==e.eos_token_id&&e.min_new_tokens>0&&r.push(new d.MinNewTokensLengthLogitsProcessor(t,e.min_new_tokens,e.eos_token_id)),null!==e.forced_bos_token_id&&r.push(new d.ForcedBOSTokenLogitsProcessor(e.forced_bos_token_id)),null!==e.forced_eos_token_id&&r.push(new d.ForcedEOSTokenLogitsProcessor(e.max_length,e.forced_eos_token_id)),null!==e.begin_suppress_tokens){const s=t>1||null===e.forced_bos_token_id?t:t+1;r.push(new d.SuppressTokensAtBeginLogitsProcessor(e.begin_suppress_tokens,s))}return null!==e.guidance_scale&&e.guidance_scale>1&&r.push(new d.ClassifierFreeGuidanceLogitsProcessor(e.guidance_scale)),null!==s&&r.extend(s),r}_prepare_generation_config(e,t,s=u.GenerationConfig){const r={...this.config};for(const e of["decoder","generator","text_config"])e in r&&Object.assign(r,r[e]);const n=new s(r);return Object.assign(n,this.generation_config??{}),e&&Object.assign(n,e),t&&Object.assign(n,(0,a.pick)(t,Object.getOwnPropertyNames(n))),n}_get_stopping_criteria(e,t=null){const s=new h.StoppingCriteriaList;return null!==e.max_length&&s.push(new h.MaxLengthCriteria(e.max_length,this.config.max_position_embeddings??null)),null!==e.eos_token_id&&s.push(new h.EosTokenCriteria(e.eos_token_id)),t&&s.extend(t),s}_validate_model_class(){if(!this.can_generate){const e=[Cl,Il,Fl,yl],t=I.get(this.constructor),s=new Set,r=this.config.model_type;for(const t of e){const e=t.get(r);e&&s.add(e[0])}let n=`The current model class (${t}) is not compatible with \`.generate()\`, as it doesn't have a language model head.`;throw s.size>0&&(n+=` Please use the following class instead: ${[...s].join(", ")}`),Error(n)}}prepare_inputs_for_generation(...e){return this._prepare_inputs_for_generation(this,...e)}_update_model_kwargs_for_generation({generated_input_ids:e,outputs:t,model_inputs:s,is_encoder_decoder:r}){return s.past_key_values=this.getPastKeyValues(t,s.past_key_values),s.input_ids=new _.Tensor("int64",e.flat(),[e.length,1]),r||(s.attention_mask=(0,_.cat)([s.attention_mask,(0,_.ones)([s.attention_mask.dims[0],1])],1)),s.position_ids=null,s}_prepare_model_inputs({inputs:e,bos_token_id:t,model_kwargs:s}){const r=(0,a.pick)(s,this.forward_params),n=this.main_input_name;if(n in r){if(e)throw new Error("`inputs`: {inputs}` were passed alongside {input_name} which is not allowed. Make sure to either pass {inputs} or {input_name}=...")}else r[n]=e;return{inputs_tensor:r[n],model_inputs:r,model_input_name:n}}async _prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:e,model_inputs:t,model_input_name:s,generation_config:r}){if(this.sessions.model.inputNames.includes("inputs_embeds")&&!t.inputs_embeds&&"_prepare_inputs_embeds"in this){const{input_ids:e,pixel_values:s,attention_mask:r,...n}=t,o=await this._prepare_inputs_embeds(t);t={...n,...(0,a.pick)(o,["inputs_embeds","attention_mask"])}}let{last_hidden_state:n}=await D(this,t);if(null!==r.guidance_scale&&r.guidance_scale>1)n=(0,_.cat)([n,(0,_.full_like)(n,0)],0),"attention_mask"in t&&(t.attention_mask=(0,_.cat)([t.attention_mask,(0,_.zeros_like)(t.attention_mask)],0));else if(t.decoder_input_ids){const e=O(t.decoder_input_ids).dims[0];if(e!==n.dims[0]){if(1!==n.dims[0])throw new Error(`The encoder outputs have a different batch size (${n.dims[0]}) than the decoder inputs (${e}).`);n=(0,_.cat)(Array.from({length:e},(()=>n)),0)}}return t.encoder_outputs=n,t}_prepare_decoder_input_ids_for_generation({batch_size:e,model_input_name:t,model_kwargs:s,decoder_start_token_id:r,bos_token_id:n,generation_config:o}){let{decoder_input_ids:i,...a}=s;if(!(i instanceof _.Tensor)){if(i)Array.isArray(i[0])||(i=Array.from({length:e},(()=>i)));else if(r??=n,"musicgen"===this.config.model_type)i=Array.from({length:e*this.config.decoder.num_codebooks},(()=>[r]));else if(Array.isArray(r)){if(r.length!==e)throw new Error(`\`decoder_start_token_id\` expcted to have length ${e} but got ${r.length}`);i=r}else i=Array.from({length:e},(()=>[r]));i=O(i)}return s.decoder_attention_mask=(0,_.ones_like)(i),{input_ids:i,model_inputs:a}}async generate({inputs:e=null,generation_config:t=null,logits_processor:s=null,stopping_criteria:r=null,streamer:n=null,...o}){this._validate_model_class(),t=this._prepare_generation_config(t,o);let{inputs_tensor:i,model_inputs:a,model_input_name:l}=this._prepare_model_inputs({inputs:e,model_kwargs:o});const c=this.config.is_encoder_decoder;let d;c&&("encoder_outputs"in a||(a=await this._prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:i,model_inputs:a,model_input_name:l,generation_config:t}))),c?({input_ids:d,model_inputs:a}=this._prepare_decoder_input_ids_for_generation({batch_size:a[l].dims.at(0),model_input_name:l,model_kwargs:a,decoder_start_token_id:t.decoder_start_token_id,bos_token_id:t.bos_token_id,generation_config:t})):d=a[l];let u=d.dims.at(-1);null!==t.max_new_tokens&&(t.max_length=u+t.max_new_tokens);const p=this._get_logits_processor(t,u,s),m=this._get_stopping_criteria(t,r),h=a[l].dims.at(0),g=f.LogitsSampler.getSampler(t),w=new Array(h).fill(0),M=d.tolist();let b;n&&n.put(M);let x={};for(;;){if(a=this.prepare_inputs_for_generation(M,a,t),b=await this.forward(a),t.output_attentions&&t.return_dict_in_generate){const e=this.getAttentions(b);for(const t in e)t in x||(x[t]=[]),x[t].push(e[t])}const e=p(M,b.logits.slice(null,-1,null)),s=[];for(let t=0;t<e.dims.at(0);++t){const r=e[t],n=await g(r);for(const[e,r]of n){const n=BigInt(e);w[t]+=r,M[t].push(n),s.push([n]);break}}n&&n.put(s);if(m(M).every((e=>e)))break;a=this._update_model_kwargs_for_generation({generated_input_ids:s,outputs:b,model_inputs:a,is_encoder_decoder:c})}n&&n.end();const y=this.getPastKeyValues(b,a.past_key_values,!0),k=new _.Tensor("int64",M.flat(),[M.length,M[0].length]);if(t.return_dict_in_generate)return{sequences:k,past_key_values:y,...x};for(const e of Object.values(b))"gpu-buffer"===e.location&&e.dispose();return k}getPastKeyValues(e,t,s=!1){const r=Object.create(null);for(const n in e)if(n.startsWith("present")){const o=n.replace("present","past_key_values"),i=n.includes("encoder");if(r[o]=i&&t?t[o]:e[n],t&&(!i||s)){const e=t[o];"gpu-buffer"===e.location&&e.dispose()}}return r}getAttentions(e){const t={};for(const s of["cross_attentions","encoder_attentions","decoder_attentions"])for(const r in e)r.startsWith(s)&&(s in t||(t[s]=[]),t[s].push(e[r]));return t}addPastKeyValues(e,t){if(t)Object.assign(e,t);else{const t=this.sessions.decoder_model_merged??this.sessions.model,s=t?.config?.kv_cache_dtype??"float32",n="float16"===s?new Uint16Array:[],o=(e[this.main_input_name]??e.attention_mask)?.dims?.[0]??1,i=(0,r.getKeyValueShapes)(this.config,{batch_size:o});for(const t in i)e[t]=new _.Tensor(s,n,i[t])}}async encode_image({pixel_values:e}){const t=(await j(this.sessions.vision_encoder,{pixel_values:e})).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 (${t.dims[1]}).`),this.config.num_image_tokens=t.dims[1]),t}async encode_text({input_ids:e}){return(await j(this.sessions.embed_tokens,{input_ids:e})).inputs_embeds}}class J{}class Y extends J{constructor({last_hidden_state:e,hidden_states:t=null,attentions:s=null}){super(),this.last_hidden_state=e,this.hidden_states=t,this.attentions=s}}class K extends H{}class Z extends K{}class ee extends K{async _call(e){return new zc(await super._call(e))}}class te extends K{async _call(e){return new Ec(await super._call(e))}}class se extends K{async _call(e){return new Lc(await super._call(e))}}class re extends K{async _call(e){return new jc(await super._call(e))}}class ne extends H{}class oe extends ne{}class ie extends ne{async _call(e){return new zc(await super._call(e))}}class ae extends ne{async _call(e){return new Ec(await super._call(e))}}class le extends ne{async _call(e){return new Lc(await super._call(e))}}class ce extends H{}class de extends ce{}class ue extends H{}class _e extends ue{}class pe extends ue{async _call(e){return new zc(await super._call(e))}}class me extends ue{async _call(e){return new Ec(await super._call(e))}}class he extends ue{async _call(e){return new Lc(await super._call(e))}}class fe extends ue{async _call(e){return new jc(await super._call(e))}}class ge extends H{}class we extends ge{}class Me extends ge{async _call(e){return new zc(await super._call(e))}}class be extends ge{async _call(e){return new Ec(await super._call(e))}}class xe extends ge{async _call(e){return new Lc(await super._call(e))}}class ye extends ge{async _call(e){return new jc(await super._call(e))}}class ke extends H{}class Te extends ke{}class ve extends ke{async _call(e){return new zc(await super._call(e))}}class Pe extends ke{async _call(e){return new Ec(await super._call(e))}}class Fe extends ke{async _call(e){return new Lc(await super._call(e))}}class Ce extends ke{async _call(e){return new jc(await super._call(e))}}class Ae extends H{}class Se extends Ae{}class Ee extends Ae{async _call(e){return new zc(await super._call(e))}}class Ie extends Ae{async _call(e){return new Ec(await super._call(e))}}class Le extends Ae{async _call(e){return new Lc(await super._call(e))}}class ze extends Ae{async _call(e){return new jc(await super._call(e))}}class je extends H{}class Ne extends je{}class Oe extends je{async _call(e){return new zc(await super._call(e))}}class Be extends je{async _call(e){return new Ec(await super._call(e))}}class Ve extends je{async _call(e){return new Lc(await super._call(e))}}class De extends je{async _call(e){return new jc(await super._call(e))}}class Re extends H{}class Ge extends Re{}class $e extends Re{async _call(e){return new zc(await super._call(e))}}class qe extends Re{async _call(e){return new Ec(await super._call(e))}}class Ue extends Re{async _call(e){return new Lc(await super._call(e))}}class We extends Re{async _call(e){return new jc(await super._call(e))}}class Xe extends H{}class Qe extends Xe{}class He extends Xe{async _call(e){return new Ec(await super._call(e))}}class Je extends Xe{async _call(e){return new Lc(await super._call(e))}}class Ye extends Xe{async _call(e){return new jc(await super._call(e))}}class Ke extends Xe{async _call(e){return new zc(await super._call(e))}}class Ze extends H{}class et extends Ze{}class tt extends Ze{async _call(e){return new zc(await super._call(e))}}class st extends Ze{async _call(e){return new Ec(await super._call(e))}}class rt extends Ze{async _call(e){return new Lc(await super._call(e))}}class nt extends H{}class ot extends nt{}class it extends nt{async _call(e){return new zc(await super._call(e))}}class at extends nt{async _call(e){return new Ec(await super._call(e))}}class lt extends nt{async _call(e){return new jc(await super._call(e))}}class ct extends H{}class dt extends ct{}class ut extends ct{async _call(e){return new zc(await super._call(e))}}class _t extends ct{async _call(e){return new Ec(await super._call(e))}}class pt extends ct{async _call(e){return new Lc(await super._call(e))}}class mt extends ct{async _call(e){return new jc(await super._call(e))}}class ht extends H{}class ft extends ht{}class gt extends ht{async _call(e){return new zc(await super._call(e))}}class wt extends ht{async _call(e){return new Ec(await super._call(e))}}class Mt extends ht{async _call(e){return new jc(await super._call(e))}}class bt extends H{}class xt extends bt{}class yt extends bt{async _call(e){return new Ec(await super._call(e))}}class kt extends bt{async _call(e){return new jc(await super._call(e))}}class Tt extends bt{async _call(e){return new zc(await super._call(e))}}class vt extends H{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"]}class Pt extends vt{}class Ft extends vt{}class Ct extends H{}class At extends Ct{}class St extends Ct{}class Et extends H{}class It extends Et{}class Lt extends Et{}class zt extends H{}class jt extends zt{}class Nt extends zt{}class Ot extends zt{async _call(e){return new Ec(await super._call(e))}}class Bt extends H{}class Vt extends Bt{}class Dt extends Bt{}class Rt extends Bt{async _call(e){return new Ec(await super._call(e))}}class Gt extends Bt{}class $t extends H{}class qt extends $t{}class Ut extends $t{}class Wt extends H{}class Xt extends Wt{}class Qt extends Wt{}class Ht extends H{}class Jt extends Ht{}class Yt extends Ht{async _call(e){return new zc(await super._call(e))}}class Kt extends Ht{async _call(e){return new Ec(await super._call(e))}}class Zt extends Ht{async _call(e){return new Lc(await super._call(e))}}class es extends Ht{async _call(e){return new jc(await super._call(e))}}class ts extends H{}class ss extends ts{}class rs extends ts{async _call(e){return new zc(await super._call(e))}}class ns extends ts{async _call(e){return new Ec(await super._call(e))}}class os extends ts{async _call(e){return new Lc(await super._call(e))}}class is extends ts{async _call(e){return new jc(await super._call(e))}}class as extends H{}class ls extends as{}class cs extends as{async _call(e){return new zc(await super._call(e))}}class ds extends as{async _call(e){return new Ec(await super._call(e))}}class us extends as{async _call(e){return new Lc(await super._call(e))}}class _s extends as{async _call(e){return new jc(await super._call(e))}}class ps extends H{}class ms extends ps{}class hs extends ps{}class fs extends H{requires_attention_mask=!1;main_input_name="input_features";forward_params=["input_features","attention_mask","decoder_input_ids","decoder_attention_mask","past_key_values"]}class gs extends fs{}class ws extends fs{_prepare_generation_config(e,t){return super._prepare_generation_config(e,t,w.WhisperGenerationConfig)}_retrieve_init_tokens(e){const t=[e.decoder_start_token_id];let s=e.language;const r=e.task;if(e.is_multilingual){s||(console.warn("No language specified - defaulting to English (en)."),s="en");const n=`<|${(0,M.whisper_language_to_code)(s)}|>`;t.push(e.lang_to_id[n]),t.push(e.task_to_id[r??"transcribe"])}else if(s||r)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!e.return_timestamps&&e.no_timestamps_token_id&&t.at(-1)!==e.no_timestamps_token_id?t.push(e.no_timestamps_token_id):e.return_timestamps&&t.at(-1)===e.no_timestamps_token_id&&(console.warn("<|notimestamps|> prompt token is removed from generation_config since `return_timestamps` is set to `true`."),t.pop()),t.filter((e=>null!=e))}async generate({inputs:e=null,generation_config:t=null,logits_processor:s=null,stopping_criteria:r=null,...n}){t=this._prepare_generation_config(t,n);const o=n.decoder_input_ids??this._retrieve_init_tokens(t);if(t.return_timestamps&&(s??=new d.LogitsProcessorList,s.push(new d.WhisperTimeStampLogitsProcessor(t,o))),t.begin_suppress_tokens&&(s??=new d.LogitsProcessorList,s.push(new d.SuppressTokensAtBeginLogitsProcessor(t.begin_suppress_tokens,o.length))),t.return_token_timestamps){if(!t.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.");"translate"===t.task&&console.warn("Token-level timestamps may not be reliable for task 'translate'."),t.output_attentions=!0,t.return_dict_in_generate=!0}const i=await super.generate({inputs:e,generation_config:t,logits_processor:s,decoder_input_ids:o,...n});return t.return_token_timestamps&&(i.token_timestamps=this._extract_token_timestamps(i,t.alignment_heads,t.num_frames)),i}_extract_token_timestamps(e,t,s=null,r=.02){if(!e.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`.");null==s&&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 n=this.config.median_filter_width;void 0===n&&(console.warn("Model config has no `median_filter_width`, using default value of 7."),n=7);const o=e.cross_attentions,i=Array.from({length:this.config.decoder_layers},((e,t)=>(0,_.cat)(o.map((e=>e[t])),2))),l=(0,_.stack)(t.map((([e,t])=>{if(e>=i.length)throw new Error(`Layer index ${e} is out of bounds for cross attentions (length ${i.length}).`);return s?i[e].slice(null,t,null,[0,s]):i[e].slice(null,t)}))).transpose(1,0,2,3),[c,d]=(0,_.std_mean)(l,-2,0,!0),u=l.clone();for(let e=0;e<u.dims[0];++e){const t=u[e];for(let s=0;s<t.dims[0];++s){const r=t[s],o=c[e][s][0].data,i=d[e][s][0].data;for(let e=0;e<r.dims[0];++e){let t=r[e].data;for(let e=0;e<t.length;++e)t[e]=(t[e]-i[e])/o[e];t.set((0,m.medianFilter)(t,n))}}}const p=[(0,_.mean)(u,1)],h=e.sequences.dims,f=new _.Tensor("float32",new Float32Array(h[0]*h[1]),h);for(let e=0;e<h[0];++e){const t=p[e].neg().squeeze_(0),[s,n]=(0,m.dynamic_time_warping)(t.tolist()),o=Array.from({length:s.length-1},((e,t)=>s[t+1]-s[t])),i=(0,a.mergeArrays)([1],o).map((e=>!!e)),l=[];for(let e=0;e<i.length;++e)i[e]&&l.push(n[e]*r);f[e].data.set(l,1)}return f}}class Ms extends H{requires_attention_mask=!1;main_input_name="input_values";forward_params=["input_values","decoder_input_ids","past_key_values"]}class bs extends Ms{}class xs extends Ms{}class ys extends H{main_input_name="pixel_values";forward_params=["pixel_values","decoder_input_ids","encoder_hidden_states","past_key_values"]}class ks extends H{forward_params=["input_ids","attention_mask","pixel_values","position_ids","past_key_values"]}class Ts extends ks{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:s,attention_mask:r}){const n=this.config.image_token_index,o=s.tolist().map((e=>e.findIndex((e=>e==n)))),i=o.every((e=>-1===e)),a=o.every((e=>-1!==e));if(!i&&!a)throw new Error("Every input should contain either 0 or 1 image token.");if(i)return{inputs_embeds:e,attention_mask:r};const l=[],c=[];for(let s=0;s<o.length;++s){const n=o[s],i=e[s],a=t[s],d=r[s];l.push((0,_.cat)([i.slice([0,n]),a,i.slice([n+1,i.dims[0]])],0)),c.push((0,_.cat)([d.slice([0,n]),(0,_.ones)([a.dims[0]]),d.slice([n+1,d.dims[0]])],0))}return{inputs_embeds:(0,_.stack)(l,0),attention_mask:(0,_.stack)(c,0)}}}class vs extends Ts{}class Ps extends Ts{}class Fs extends H{forward_params=["input_ids","inputs_embeds","attention_mask","pixel_values","encoder_outputs","decoder_input_ids","decoder_inputs_embeds","decoder_attention_mask","past_key_values"];main_input_name="inputs_embeds"}class Cs extends Fs{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:s,attention_mask:r}){return{inputs_embeds:(0,_.cat)([t,e],1),attention_mask:(0,_.cat)([(0,_.ones)(t.dims.slice(0,2)),r],1)}}async _prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:s,attention_mask:r}){if(!e&&!t)throw new Error("Either `input_ids` or `pixel_values` should be provided.");let n,o;return e&&(n=await this.encode_text({input_ids:e})),t&&(o=await this.encode_image({pixel_values:t})),n&&o?({inputs_embeds:s,attention_mask:r}=this._merge_input_ids_with_image_features({inputs_embeds:n,image_features:o,input_ids:e,attention_mask:r})):s=n||o,{inputs_embeds:s,attention_mask:r}}async forward({input_ids:e,pixel_values:t,attention_mask:s,decoder_input_ids:r,decoder_attention_mask:n,encoder_outputs:o,past_key_values:i,inputs_embeds:a,decoder_inputs_embeds:l}){if(a||({inputs_embeds:a,attention_mask:s}=await this._prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:a,attention_mask:s})),!o){let{last_hidden_state:e}=await D(this,{inputs_embeds:a,attention_mask:s});o=e}if(!l){if(!r)throw new Error("Either `decoder_input_ids` or `decoder_inputs_embeds` should be provided.");l=await this.encode_text({input_ids:r})}const c={inputs_embeds:l,attention_mask:n,encoder_attention_mask:s,encoder_hidden_states:o,past_key_values:i};return await R(this,c,!0)}}class As extends H{forward_params=["input_ids","attention_mask","pixel_values","position_ids","past_key_values"]}class Ss extends As{_merge_input_ids_with_image_features(e){const t=e.image_features.dims.at(-1),s=e.image_features.view(-1,t);return G({image_token_id:this.config.image_token_index,...e,image_features:s})}}class Es extends H{forward_params=["input_ids","attention_mask","pixel_values","pixel_attention_mask","position_ids","past_key_values"]}class Is extends Es{async encode_image({pixel_values:e,pixel_attention_mask:t}){return(await j(this.sessions.vision_encoder,{pixel_values:e,pixel_attention_mask:t})).image_features}_merge_input_ids_with_image_features(e){const t=e.image_features.dims.at(-1),s=e.image_features.view(-1,t);return G({image_token_id:this.config.image_token_id,...e,image_features:s})}}class Ls extends H{forward_params=["input_ids","inputs_embeds","attention_mask","position_ids","pixel_values","image_sizes","past_key_values"]}class zs extends Ls{async forward({input_ids:e=null,attention_mask:t=null,pixel_values:s=null,image_sizes:r=null,position_ids:n=null,inputs_embeds:o=null,past_key_values:i=null,generation_config:a=null,logits_processor:l=null,...c}){if(!o){let t;if(s&&1!==e.dims[1]){if(!r)throw new Error("`image_sizes` must be provided when `pixel_values` is provided.");({image_features:t}=await j(this.sessions.vision_encoder,{pixel_values:s,image_sizes:r}))}else{const e=this.config.normalized_config.hidden_size;t=new _.Tensor("float32",[],[0,e])}({inputs_embeds:o}=await j(this.sessions.prepare_inputs_embeds,{input_ids:e,image_features:t}))}return await R(this,{inputs_embeds:o,past_key_values:i,attention_mask:t,position_ids:n,generation_config:a,logits_processor:l},!1)}}class js extends H{}class Ns extends js{}class Os extends js{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"text_model"})}}class Bs extends js{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"text_model"})}}class Vs extends js{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"vision_model"})}}class Ds extends js{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"vision_model"})}}class Rs extends H{}class Gs extends Rs{}class $s extends Rs{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"text_model"})}}class qs extends js{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"vision_model"})}}class Us extends H{}class Ws extends Us{}class Xs extends H{}class Qs extends Xs{async forward(e){const t=!e.input_ids,s=!e.pixel_values;if(t&&s)throw new Error("Either `input_ids` or `pixel_values` should be provided.");if(t&&(e.input_ids=(0,_.ones)([e.pixel_values.dims[0],1])),s){const{image_size:t}=this.config.vision_config;e.pixel_values=(0,_.full)([0,3,t,t],0)}const{text_embeddings:r,image_embeddings:n,l2norm_text_embeddings:o,l2norm_image_embeddings:i}=await super.forward(e),a={};return t||(a.text_embeddings=r,a.l2norm_text_embeddings=o),s||(a.image_embeddings=n,a.l2norm_image_embeddings=i),a}}class Hs extends Xs{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"text_model"})}}class Js extends Xs{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"vision_model"})}}class Ys extends H{}class Ks extends Ys{}class Zs extends Ys{}class er extends H{}class tr extends er{}class sr extends er{}class rr extends H{}class nr extends rr{}class or extends rr{}class ir extends H{}class ar extends ir{}class lr extends ir{}class cr extends H{}class dr extends cr{}class ur extends cr{}class _r extends H{}class pr extends _r{}class mr extends _r{}class hr extends H{}class fr extends hr{}class gr extends hr{}class wr extends H{}class Mr extends wr{}class br extends wr{}class xr extends H{}class yr extends xr{}class kr extends xr{}class Tr extends H{}class vr extends Tr{}class Pr extends Tr{}class Fr extends H{}class Cr extends Fr{}class Ar extends Fr{}class Sr extends H{}class Er extends Sr{}class Ir extends Sr{}class Lr extends H{}class zr extends Lr{}class jr extends Lr{}class Nr extends H{}class Or extends Nr{}class Br extends Nr{}class Vr extends H{}class Dr extends Vr{}class Rr extends Vr{}class Gr extends H{}class $r extends Gr{}class qr extends Gr{}class Ur extends H{}class Wr extends Ur{}class Xr extends Ur{}class Qr extends H{}class Hr extends Qr{}class Jr extends Qr{}class Yr extends H{}class Kr extends Yr{}class Zr extends Yr{}class en extends H{}class tn extends en{}class sn extends en{}class rn extends H{}class nn extends rn{}class on extends rn{}class an extends H{forward_params=["input_ids","attention_mask","position_ids","past_key_values","pixel_values","image_grid_thw"]}class ln extends an{get_rope_index(e,t,s,r){const{vision_config:n,image_token_id:o,video_token_id:i,vision_start_token_id:a}=this.config,l=n.spatial_merge_size??2,c=[];if(t||s){let n=e.tolist();r||(r=(0,_.ones_like)(e));const d=r.tolist(),u=Array.from({length:3},(t=>Array.from({length:e.dims[0]},(t=>Array.from({length:e.dims[1]},(e=>1)))))),p=t?t.tolist():[],h=s?s.tolist():[];let f=0,g=0;for(let e=0;e<n.length;++e){const t=n[e].filter(((t,s)=>1==d[e][s])),s=t.reduce(((e,t,s)=>(t==a&&e.push(s),e)),[]).map((e=>t[e+1])),r=s.filter((e=>e==o)).length,_=s.filter((e=>e==i)).length;let w=[],M=0,b=r,x=_;for(let e=0;e<s.length;++e){const e=t.findIndex(((e,t)=>t>M&&e==o)),s=t.findIndex(((e,t)=>t>M&&e==i)),r=b>0&&-1!==e?e:t.length+1,n=x>0&&-1!==s?s:t.length+1;let a,c,d,u;r<n?([c,d,u]=p[f],++f,--b,a=r):([c,d,u]=h[g],++g,--x,a=n);const[_,y,k]=[Number(c),Math.floor(Number(d)/l),Math.floor(Number(u)/l)],T=a-M,v=w.length>0?(0,m.max)(w.at(-1))[0]+1:0;w.push(Array.from({length:3*T},((e,t)=>v+t%T)));const P=T+v,F=_*y*k,C=Array.from({length:F},((e,t)=>P+Math.floor(t/(y*k)))),A=Array.from({length:F},((e,t)=>P+Math.floor(t/k)%y)),S=Array.from({length:F},((e,t)=>P+t%k));w.push([C,A,S].flat()),M=a+F}if(M<t.length){const e=w.length>0?(0,m.max)(w.at(-1))[0]+1:0,s=t.length-M;w.push(Array.from({length:3*s},((t,r)=>e+r%s)))}const y=w.reduce(((e,t)=>e+t.length),0),k=new Array(y);let T=0;for(let e=0;e<3;++e)for(let t=0;t<w.length;++t){const s=w[t],r=s.length/3;for(let t=e*r;t<(e+1)*r;++t)k[T++]=s[t]}let v=0;const P=d[e];for(let t=0;t<P.length;++t)if(1==P[t]){for(let s=0;s<3;++s)u[s][e][t]=k[s*y/3+v];++v}const F=(0,m.max)(k)[0];c.push(F+1-n[e].length)}return[new _.Tensor("int64",u.flat(1/0),[3,e.dims[0],e.dims[1]]),new _.Tensor("int64",c,[c.length,1])]}if(r){const{data:e,dims:t}=q(r),s=BigInt64Array.from({length:3*e.length},((t,s)=>e[s%e.length])),n=Array.from({length:t[0]},((s,r)=>(0,m.max)(e.subarray(t[1]*r,t[1]*(r+1)))[0]+1n+BigInt(t[1])));return[new _.Tensor("int64",s,[3,...t]),new _.Tensor("int64",n,[n.length,1])]}{const[t,s]=e.dims,r=BigInt64Array.from({length:3*t*s},((e,r)=>BigInt(Math.floor(r%s/t))));return[new _.Tensor("int64",r,[3,...e.dims]),(0,_.zeros)([t,1])]}}async encode_image({pixel_values:e,image_grid_thw:t}){return(await j(this.sessions.vision_encoder,{pixel_values:e,grid_thw:t})).image_features}_merge_input_ids_with_image_features(e){return G({image_token_id:this.config.image_token_id,...e})}prepare_inputs_for_generation(e,t,s){if(t.attention_mask&&!t.position_ids)if(t.past_key_values){t.pixel_values=null;const e=BigInt(Object.values(t.past_key_values)[0].dims.at(-2)),s=t.rope_deltas.map((t=>e+t));t.position_ids=(0,_.stack)([s,s,s],0)}else[t.position_ids,t.rope_deltas]=this.get_rope_index(t.input_ids,t.image_grid_thw,t.video_grid_thw,t.attention_mask);return t}}class cn extends H{}class dn extends cn{}class un extends cn{}class _n extends H{}class pn extends _n{}class mn extends _n{}class hn extends H{}class fn extends hn{}class gn extends hn{}class wn extends H{}class Mn extends wn{}class bn extends wn{}class xn extends H{}class yn extends xn{}class kn extends xn{}class Tn extends H{}class vn extends Tn{}class Pn extends Tn{async _call(e){return new Ec(await super._call(e))}}class Fn extends H{}class Cn extends Fn{}class An extends Fn{async _call(e){return new Ec(await super._call(e))}}class Sn extends H{}class En extends Sn{}class In extends H{}class Ln extends In{}class zn extends In{async _call(e){return new Ec(await super._call(e))}}class jn extends H{}class Nn extends jn{}class On extends H{}class Bn extends On{}class Vn extends On{async _call(e){return new Ec(await super._call(e))}}class Dn extends H{}class Rn extends Dn{}class Gn extends H{}class $n extends Gn{}class qn extends Gn{async _call(e){return new Ec(await super._call(e))}}class Un extends H{}class Wn extends Un{async _call(e){return new Bc(await super._call(e))}}class Xn extends H{}class Qn extends Xn{}class Hn extends Xn{async _call(e){return new Ec(await super._call(e))}}class Jn extends H{}class Yn extends Jn{}class Kn extends Jn{async _call(e){return new Ec(await super._call(e))}}class Zn extends H{}class eo extends Zn{}class to extends Zn{}class so extends H{}class ro extends so{}class no extends so{}class oo extends H{}class io extends oo{}class ao extends oo{async _call(e){return new Ec(await super._call(e))}}class lo extends H{}class co extends lo{}class uo extends lo{async _call(e){return new po(await super._call(e))}}class _o extends lo{async _call(e){return new mo(await super._call(e))}}class po extends J{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class mo extends J{constructor({logits:e,pred_boxes:t,pred_masks:s}){super(),this.logits=e,this.pred_boxes=t,this.pred_masks=s}}class ho extends H{}class fo extends ho{}class go extends ho{async _call(e){return new wo(await super._call(e))}}class wo extends J{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class Mo extends H{}class bo extends Mo{}class xo extends Mo{async _call(e){return new yo(await super._call(e))}}class yo extends po{}class ko extends H{}class To extends ko{}class vo extends ko{async _call(e){return new Ec(await super._call(e))}}class Po extends H{}class Fo extends Po{}class Co extends Po{async _call(e){return new Ec(await super._call(e))}}class Ao extends H{}class So extends Ao{}class Eo extends Ao{async _call(e){return new Ec(await super._call(e))}}class Io extends H{}class Lo extends Io{}class zo extends Io{async _call(e){return new Ec(await super._call(e))}}class jo extends H{}class No extends jo{}class Oo extends jo{}class Bo extends H{}class Vo extends Bo{}class Do extends Bo{}class Ro extends H{}class Go extends Ro{}class $o extends H{}class qo extends $o{}class Uo extends $o{}class Wo extends $o{}class Xo extends H{}class Qo extends Xo{}class Ho extends H{}class Jo extends Ho{}class Yo extends Ho{}class Ko extends H{}class Zo extends Ko{}class ei extends Ko{}class ti extends H{}class si extends ti{}class ri extends H{}class ni extends ri{}class oi extends ri{async _call(e){return new Ec(await super._call(e))}}class ii extends H{}class ai extends ii{}class li extends ii{async _call(e){return new Ec(await super._call(e))}}class ci extends H{}class di extends ci{}class ui extends ci{async _call(e){return new Ec(await super._call(e))}}class _i extends H{}class pi extends _i{}class mi extends _i{async _call(e){return new Ec(await super._call(e))}}class hi extends H{}class fi extends hi{}class gi extends H{}class wi extends gi{}class Mi extends gi{async _call(e){return new bi(await super._call(e))}}class bi extends J{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class xi extends H{}class yi extends xi{async get_image_embeddings({pixel_values:e}){return await D(this,{pixel_values:e})}async forward(e){if(e.image_embeddings&&e.image_positional_embeddings||(e={...e,...await this.get_image_embeddings(e)}),!e.input_labels&&e.input_points){const t=e.input_points.dims.slice(0,-1),s=t.reduce(((e,t)=>e*t),1);e.input_labels=new _.Tensor("int64",new BigInt64Array(s).fill(1n),t)}const t={image_embeddings:e.image_embeddings,image_positional_embeddings:e.image_positional_embeddings};return e.input_points&&(t.input_points=e.input_points),e.input_labels&&(t.input_labels=e.input_labels),e.input_boxes&&(t.input_boxes=e.input_boxes),await j(this.sessions.prompt_encoder_mask_decoder,t)}async _call(e){return new ki(await super._call(e))}}class ki extends J{constructor({iou_scores:e,pred_masks:t}){super(),this.iou_scores=e,this.pred_masks=t}}class Ti extends H{}class vi extends Ti{}class Pi extends Ti{}class Fi extends H{}class Ci extends Fi{}class Ai extends Fi{}class Si extends H{}class Ei extends Si{}class Ii extends Si{async _call(e){return new Nc(await super._call(e))}}class Li extends Si{async _call(e){return new Ec(await super._call(e))}}class zi extends Si{async _call(e){return new Lc(await super._call(e))}}class ji extends H{}class Ni extends ji{}class Oi extends ji{async _call(e){return new Lc(await super._call(e))}}class Bi extends H{}class Vi extends Bi{}class Di extends H{}class Ri extends Di{}class Gi extends Di{async _call(e){return new Nc(await super._call(e))}}class $i extends Di{async _call(e){return new Ec(await super._call(e))}}class qi extends H{}class Ui extends qi{}class Wi extends qi{async _call(e){return new Nc(await super._call(e))}}class Xi extends qi{async _call(e){return new Ec(await super._call(e))}}class Qi extends qi{async _call(e){return new Lc(await super._call(e))}}class Hi extends H{}class Ji extends Hi{}class Yi extends Hi{async _call(e){return new Nc(await super._call(e))}}class Ki extends Hi{async _call(e){return new Ec(await super._call(e))}}class Zi extends H{}class ea extends Si{}class ta extends Si{async _call(e){return new Nc(await super._call(e))}}class sa extends Si{async _call(e){return new Ec(await super._call(e))}}class ra extends H{}class na extends ra{}class oa extends ra{async _call(e){return new Nc(await super._call(e))}}class ia extends ra{async _call(e){return new Ec(await super._call(e))}}class aa extends ra{async _call(e){return new Ic(await super._call(e))}}class la extends ra{async _call(e){return new Lc(await super._call(e))}}class ca extends H{}class da extends ca{}class ua extends H{}class _a extends ua{}class pa extends ua{}class ma extends ua{async generate_speech(e,t,{threshold:s=.5,minlenratio:r=0,maxlenratio:n=20,vocoder:o=null}={}){const i={input_ids:e},{encoder_outputs:a,encoder_attention_mask:l}=await D(this,i),c=a.dims[1]/this.config.reduction_factor,d=Math.floor(c*n),u=Math.floor(c*r),p=this.config.num_mel_bins;let m=[],h=null,f=null,g=0;for(;;){++g;const e=B(!!f);let r;r=f?f.output_sequence_out:new _.Tensor("float32",new Float32Array(p),[1,1,p]);let n={use_cache_branch:e,output_sequence:r,encoder_attention_mask:l,speaker_embeddings:t,encoder_hidden_states:a};this.addPastKeyValues(n,h),f=await j(this.sessions.decoder_model_merged,n),h=this.getPastKeyValues(f,h);const{prob:o,spectrum:i}=f;if(m.push(i),g>=u&&(Array.from(o.data).filter((e=>e>=s)).length>0||g>=d))break}const w=(0,_.cat)(m),{waveform:M}=await j(o.sessions.model,{spectrogram:w});return{spectrogram:w,waveform:M}}}class ha extends H{main_input_name="spectrogram"}class fa extends H{}class ga extends fa{}class wa extends H{}class Ma extends wa{}class ba extends wa{}class xa extends H{}class ya extends xa{}class ka extends xa{}class Ta extends H{}class va extends Ta{}class Pa extends Ta{}class Fa extends H{}class Ca extends Fa{}class Aa extends Fa{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"text_model"})}}class Sa extends Fa{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"audio_model"})}}class Ea extends H{}class Ia extends Ea{async _call(e){return new Vc(await super._call(e))}}class La extends H{}class za extends La{}class ja extends La{}class Na extends La{}class Oa extends H{}class Ba extends Oa{}class Va extends Oa{}class Da extends H{}class Ra extends Da{}class Ga extends Da{async _call(e){return new Ec(await super._call(e))}}class $a extends H{}class qa extends $a{}class Ua extends $a{}class Wa extends H{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"];_apply_and_filter_by_delay_pattern_mask(e){const[t,s]=e.dims,r=this.config.decoder.num_codebooks,n=s-r;let o=0;for(let t=0;t<e.size;++t){if(e.data[t]===this.config.decoder.pad_token_id)continue;const i=t%s-Math.floor(t/s)%r;i>0&&i<=n&&(e.data[o++]=e.data[t])}const i=Math.floor(t/r),a=o/(i*r);return new _.Tensor(e.type,e.data.slice(0,o),[i,r,a])}prepare_inputs_for_generation(e,t,s){let r=structuredClone(e);for(let e=0;e<r.length;++e)for(let t=0;t<r[e].length;++t)e%this.config.decoder.num_codebooks>=t&&(r[e][t]=BigInt(this.config.decoder.pad_token_id));null!==s.guidance_scale&&s.guidance_scale>1&&(r=r.concat(r));return super.prepare_inputs_for_generation(r,t,s)}async generate(e){const t=await super.generate(e),s=this._apply_and_filter_by_delay_pattern_mask(t).unsqueeze_(0),{audio_values:r}=await j(this.sessions.encodec_decode,{audio_codes:s});return r}}class Xa extends H{}class Qa extends Xa{}class Ha extends Xa{async _call(e){return new Ec(await super._call(e))}}class Ja extends H{}class Ya extends Ja{}class Ka extends Ja{async _call(e){return new Ec(await super._call(e))}}class Za extends H{}class el extends Za{}class tl extends Za{async _call(e){return new Ec(await super._call(e))}}class sl extends H{}class rl extends sl{}class nl extends sl{async _call(e){return new Ec(await super._call(e))}}class ol extends H{}class il extends ol{}class al extends H{}class ll extends al{forward_params=["input_ids","pixel_values","images_seq_mask","images_emb_mask","attention_mask","position_ids","past_key_values"];constructor(...e){super(...e),this._generation_mode="text"}async forward(e){const t=this._generation_mode??"text";let s;if("text"!==t&&e.past_key_values){const t=this.sessions.gen_img_embeds,r=(0,a.pick)({image_ids:e.input_ids},t.inputNames);s=await j(t,r)}else{const t=this.sessions.prepare_inputs_embeds,r=(0,a.pick)(e,t.inputNames);s=await j(t,r)}const r={...e,...s},n=await R(this,r),o=this.sessions["text"===t?"lm_head":"gen_head"];if(!o)throw new Error(`Unable to find "${o}" generation head`);const i=await j(o,(0,a.pick)(n,o.inputNames));return{...s,...n,...i}}async generate(e){return this._generation_mode="text",super.generate(e)}async generate_images(e){this._generation_mode="image";const t=(e.inputs??e[this.main_input_name]).dims[1],s=(await super.generate(e)).slice(null,[t,null]),r=this.sessions.image_decode,{decoded_image:n}=await j(r,{generated_tokens:s}),o=n.add_(1).mul_(127.5).clamp_(0,255).to("uint8"),i=[];for(const e of o){const t=p.RawImage.fromTensor(e);i.push(t)}return i}}class cl extends J{constructor({char_logits:e,bpe_logits:t,wp_logits:s}){super(),this.char_logits=e,this.bpe_logits=t,this.wp_logits=s}get logits(){return[this.char_logits,this.bpe_logits,this.wp_logits]}}class dl extends H{}class ul extends dl{async _call(e){return new cl(await super._call(e))}}class _l extends H{}class pl extends _l{}class ml extends _l{}class hl extends H{}class fl extends hl{}class gl extends hl{}class wl{static MODEL_CLASS_MAPPINGS=null;static BASE_IF_FAIL=!1;static async from_pretrained(e,{progress_callback:t=null,config:s=null,cache_dir:n=null,local_files_only:o=!1,revision:i="main",model_file_name:a=null,subfolder:l="onnx",device:c=null,dtype:d=null,use_external_data_format:u=null,session_options:_={}}={}){const p={progress_callback:t,config:s,cache_dir:n,local_files_only:o,revision:i,model_file_name:a,subfolder:l,device:c,dtype:d,use_external_data_format:u,session_options:_};if(p.config=await r.AutoConfig.from_pretrained(e,p),!this.MODEL_CLASS_MAPPINGS)throw new Error("`MODEL_CLASS_MAPPINGS` not implemented for this type of `AutoClass`: "+this.name);for(const t of this.MODEL_CLASS_MAPPINGS){const s=t.get(p.config.model_type);if(s)return await s[1].from_pretrained(e,p)}if(this.BASE_IF_FAIL)return console.warn(`Unknown model class "${p.config.model_type}", attempting to construct from base class.`),await H.from_pretrained(e,p);throw Error(`Unsupported model type: ${p.config.model_type}`)}}const Ml=new Map([["bert",["BertModel",Z]],["modernbert",["ModernBertModel",oe]],["nomic_bert",["NomicBertModel",de]],["roformer",["RoFormerModel",_e]],["electra",["ElectraModel",Te]],["esm",["EsmModel",et]],["convbert",["ConvBertModel",we]],["camembert",["CamembertModel",Se]],["deberta",["DebertaModel",Ne]],["deberta-v2",["DebertaV2Model",Ge]],["mpnet",["MPNetModel",dt]],["albert",["AlbertModel",xt]],["distilbert",["DistilBertModel",Qe]],["roberta",["RobertaModel",Jt]],["xlm",["XLMModel",ss]],["xlm-roberta",["XLMRobertaModel",ls]],["clap",["ClapModel",Ca]],["clip",["CLIPModel",Ns]],["clipseg",["CLIPSegModel",Ks]],["chinese_clip",["ChineseCLIPModel",Ws]],["siglip",["SiglipModel",Gs]],["jina_clip",["JinaCLIPModel",Qs]],["mobilebert",["MobileBertModel",ot]],["squeezebert",["SqueezeBertModel",ft]],["wav2vec2",["Wav2Vec2Model",Ei]],["wav2vec2-bert",["Wav2Vec2BertModel",Ji]],["unispeech",["UniSpeechModel",Ri]],["unispeech-sat",["UniSpeechSatModel",Ui]],["hubert",["HubertModel",ea]],["wavlm",["WavLMModel",na]],["audio-spectrogram-transformer",["ASTModel",ms]],["vits",["VitsModel",Ia]],["pyannote",["PyAnnoteModel",Ni]],["wespeaker-resnet",["WeSpeakerResNetModel",Vi]],["detr",["DetrModel",co]],["rt_detr",["RTDetrModel",fo]],["table-transformer",["TableTransformerModel",bo]],["vit",["ViTModel",vn]],["ijepa",["IJepaModel",Cn]],["pvt",["PvtModel",Ln]],["vit_msn",["ViTMSNModel",Bn]],["vit_mae",["ViTMAEModel",Nn]],["groupvit",["GroupViTModel",Rn]],["fastvit",["FastViTModel",$n]],["mobilevit",["MobileViTModel",Qn]],["mobilevitv2",["MobileViTV2Model",Yn]],["owlvit",["OwlViTModel",eo]],["owlv2",["Owlv2Model",ro]],["beit",["BeitModel",io]],["deit",["DeiTModel",To]],["hiera",["HieraModel",Fo]],["convnext",["ConvNextModel",ni]],["convnextv2",["ConvNextV2Model",ai]],["dinov2",["Dinov2Model",di]],["dinov2_with_registers",["Dinov2WithRegistersModel",pi]],["resnet",["ResNetModel",So]],["swin",["SwinModel",Lo]],["swin2sr",["Swin2SRModel",No]],["donut-swin",["DonutSwinModel",si]],["yolos",["YolosModel",wi]],["dpt",["DPTModel",Vo]],["glpn",["GLPNModel",Zo]],["hifigan",["SpeechT5HifiGan",ha]],["efficientnet",["EfficientNetModel",Ra]],["decision_transformer",["DecisionTransformerModel",il]],["patchtst",["PatchTSTForPrediction",pl]],["patchtsmixer",["PatchTSMixerForPrediction",fl]],["mobilenet_v1",["MobileNetV1Model",Qa]],["mobilenet_v2",["MobileNetV2Model",Ya]],["mobilenet_v3",["MobileNetV3Model",el]],["mobilenet_v4",["MobileNetV4Model",rl]],["maskformer",["MaskFormerModel",Jo]],["mgp-str",["MgpstrForSceneTextRecognition",ul]],["style_text_to_speech_2",["StyleTextToSpeech2Model",da]]]),bl=new Map([["t5",["T5Model",Pt]],["longt5",["LongT5Model",At]],["mt5",["MT5Model",It]],["bart",["BartModel",jt]],["mbart",["MBartModel",Vt]],["marian",["MarianModel",vi]],["whisper",["WhisperModel",gs]],["m2m_100",["M2M100Model",Ci]],["blenderbot",["BlenderbotModel",qt]],["blenderbot-small",["BlenderbotSmallModel",Xt]]]),xl=new Map([["bloom",["BloomModel",fn]],["jais",["JAISModel",nr]],["gpt2",["GPT2Model",tr]],["gptj",["GPTJModel",pr]],["gpt_bigcode",["GPTBigCodeModel",fr]],["gpt_neo",["GPTNeoModel",ar]],["gpt_neox",["GPTNeoXModel",dr]],["codegen",["CodeGenModel",Mr]],["llama",["LlamaModel",yr]],["exaone",["ExaoneModel",Er]],["olmo",["OlmoModel",Or]],["olmo2",["Olmo2Model",Dr]],["mobilellm",["MobileLLMModel",zr]],["granite",["GraniteModel",$r]],["cohere",["CohereModel",Wr]],["gemma",["GemmaModel",Hr]],["gemma2",["Gemma2Model",Kr]],["helium",["HeliumModel",vr]],["glm",["GlmModel",Cr]],["openelm",["OpenELMModel",tn]],["qwen2",["Qwen2Model",nn]],["phi",["PhiModel",dn]],["phi3",["Phi3Model",pn]],["mpt",["MptModel",Mn]],["opt",["OPTModel",yn]],["mistral",["MistralModel",Ma]],["starcoder2",["Starcoder2Model",ya]],["falcon",["FalconModel",va]],["stablelm",["StableLmModel",Ba]]]),yl=new Map([["speecht5",["SpeechT5ForSpeechToText",pa]],["whisper",["WhisperForConditionalGeneration",ws]],["moonshine",["MoonshineForConditionalGeneration",xs]]]),kl=new Map([["speecht5",["SpeechT5ForTextToSpeech",ma]]]),Tl=new Map([["vits",["VitsModel",Ia]],["musicgen",["MusicgenForConditionalGeneration",Wa]]]),vl=new Map([["bert",["BertForSequenceClassification",te]],["modernbert",["ModernBertForSequenceClassification",ae]],["roformer",["RoFormerForSequenceClassification",me]],["electra",["ElectraForSequenceClassification",Pe]],["esm",["EsmForSequenceClassification",st]],["convbert",["ConvBertForSequenceClassification",be]],["camembert",["CamembertForSequenceClassification",Ie]],["deberta",["DebertaForSequenceClassification",Be]],["deberta-v2",["DebertaV2ForSequenceClassification",qe]],["mpnet",["MPNetForSequenceClassification",_t]],["albert",["AlbertForSequenceClassification",yt]],["distilbert",["DistilBertForSequenceClassification",He]],["roberta",["RobertaForSequenceClassification",Kt]],["xlm",["XLMForSequenceClassification",ns]],["xlm-roberta",["XLMRobertaForSequenceClassification",ds]],["bart",["BartForSequenceClassification",Ot]],["mbart",["MBartForSequenceClassification",Rt]],["mobilebert",["MobileBertForSequenceClassification",at]],["squeezebert",["SqueezeBertForSequenceClassification",wt]]]),Pl=new Map([["bert",["BertForTokenClassification",se]],["modernbert",["ModernBertForTokenClassification",le]],["roformer",["RoFormerForTokenClassification",he]],["electra",["ElectraForTokenClassification",Fe]],["esm",["EsmForTokenClassification",rt]],["convbert",["ConvBertForTokenClassification",xe]],["camembert",["CamembertForTokenClassification",Le]],["deberta",["DebertaForTokenClassification",Ve]],["deberta-v2",["DebertaV2ForTokenClassification",Ue]],["mpnet",["MPNetForTokenClassification",pt]],["distilbert",["DistilBertForTokenClassification",Je]],["roberta",["RobertaForTokenClassification",Zt]],["xlm",["XLMForTokenClassification",os]],["xlm-roberta",["XLMRobertaForTokenClassification",us]]]),Fl=new Map([["t5",["T5ForConditionalGeneration",Ft]],["longt5",["LongT5ForConditionalGeneration",St]],["mt5",["MT5ForConditionalGeneration",Lt]],["bart",["BartForConditionalGeneration",Nt]],["mbart",["MBartForConditionalGeneration",Dt]],["marian",["MarianMTModel",Pi]],["m2m_100",["M2M100ForConditionalGeneration",Ai]],["blenderbot",["BlenderbotForConditionalGeneration",Ut]],["blenderbot-small",["BlenderbotSmallForConditionalGeneration",Qt]]]),Cl=new Map([["bloom",["BloomForCausalLM",gn]],["gpt2",["GPT2LMHeadModel",sr]],["jais",["JAISLMHeadModel",or]],["gptj",["GPTJForCausalLM",mr]],["gpt_bigcode",["GPTBigCodeForCausalLM",gr]],["gpt_neo",["GPTNeoForCausalLM",lr]],["gpt_neox",["GPTNeoXForCausalLM",ur]],["codegen",["CodeGenForCausalLM",br]],["llama",["LlamaForCausalLM",kr]],["exaone",["ExaoneForCausalLM",Ir]],["olmo",["OlmoForCausalLM",Br]],["olmo2",["Olmo2ForCausalLM",Rr]],["mobilellm",["MobileLLMForCausalLM",jr]],["granite",["GraniteForCausalLM",qr]],["cohere",["CohereForCausalLM",Xr]],["gemma",["GemmaForCausalLM",Jr]],["gemma2",["Gemma2ForCausalLM",Zr]],["helium",["HeliumForCausalLM",Pr]],["glm",["GlmForCausalLM",Ar]],["openelm",["OpenELMForCausalLM",sn]],["qwen2",["Qwen2ForCausalLM",on]],["phi",["PhiForCausalLM",un]],["phi3",["Phi3ForCausalLM",mn]],["mpt",["MptForCausalLM",bn]],["opt",["OPTForCausalLM",kn]],["mbart",["MBartForCausalLM",Gt]],["mistral",["MistralForCausalLM",ba]],["starcoder2",["Starcoder2ForCausalLM",ka]],["falcon",["FalconForCausalLM",Pa]],["trocr",["TrOCRForCausalLM",ga]],["stablelm",["StableLmForCausalLM",Va]],["phi3_v",["Phi3VForCausalLM",zs]]]),Al=new Map([["multi_modality",["MultiModalityCausalLM",ll]]]),Sl=new Map([["bert",["BertForMaskedLM",ee]],["modernbert",["ModernBertForMaskedLM",ie]],["roformer",["RoFormerForMaskedLM",pe]],["electra",["ElectraForMaskedLM",ve]],["esm",["EsmForMaskedLM",tt]],["convbert",["ConvBertForMaskedLM",Me]],["camembert",["CamembertForMaskedLM",Ee]],["deberta",["DebertaForMaskedLM",Oe]],["deberta-v2",["DebertaV2ForMaskedLM",$e]],["mpnet",["MPNetForMaskedLM",ut]],["albert",["AlbertForMaskedLM",Tt]],["distilbert",["DistilBertForMaskedLM",Ke]],["roberta",["RobertaForMaskedLM",Yt]],["xlm",["XLMWithLMHeadModel",rs]],["xlm-roberta",["XLMRobertaForMaskedLM",cs]],["mobilebert",["MobileBertForMaskedLM",it]],["squeezebert",["SqueezeBertForMaskedLM",gt]]]),El=new Map([["bert",["BertForQuestionAnswering",re]],["roformer",["RoFormerForQuestionAnswering",fe]],["electra",["ElectraForQuestionAnswering",Ce]],["convbert",["ConvBertForQuestionAnswering",ye]],["camembert",["CamembertForQuestionAnswering",ze]],["deberta",["DebertaForQuestionAnswering",De]],["deberta-v2",["DebertaV2ForQuestionAnswering",We]],["mpnet",["MPNetForQuestionAnswering",mt]],["albert",["AlbertForQuestionAnswering",kt]],["distilbert",["DistilBertForQuestionAnswering",Ye]],["roberta",["RobertaForQuestionAnswering",es]],["xlm",["XLMForQuestionAnswering",is]],["xlm-roberta",["XLMRobertaForQuestionAnswering",_s]],["mobilebert",["MobileBertForQuestionAnswering",lt]],["squeezebert",["SqueezeBertForQuestionAnswering",Mt]]]),Il=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",ys]],["idefics3",["Idefics3ForConditionalGeneration",Is]]]),Ll=new Map([["llava",["LlavaForConditionalGeneration",Ts]],["llava_onevision",["LlavaOnevisionForConditionalGeneration",vs]],["moondream1",["Moondream1ForConditionalGeneration",Ps]],["florence2",["Florence2ForConditionalGeneration",Cs]],["qwen2-vl",["Qwen2VLForConditionalGeneration",ln]],["idefics3",["Idefics3ForConditionalGeneration",Is]],["paligemma",["PaliGemmaForConditionalGeneration",Ss]]]),zl=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",ys]]]),jl=new Map([["vit",["ViTForImageClassification",Pn]],["ijepa",["IJepaForImageClassification",An]],["pvt",["PvtForImageClassification",zn]],["vit_msn",["ViTMSNForImageClassification",Vn]],["fastvit",["FastViTForImageClassification",qn]],["mobilevit",["MobileViTForImageClassification",Hn]],["mobilevitv2",["MobileViTV2ForImageClassification",Kn]],["beit",["BeitForImageClassification",ao]],["deit",["DeiTForImageClassification",vo]],["hiera",["HieraForImageClassification",Co]],["convnext",["ConvNextForImageClassification",oi]],["convnextv2",["ConvNextV2ForImageClassification",li]],["dinov2",["Dinov2ForImageClassification",ui]],["dinov2_with_registers",["Dinov2WithRegistersForImageClassification",mi]],["resnet",["ResNetForImageClassification",Eo]],["swin",["SwinForImageClassification",zo]],["segformer",["SegformerForImageClassification",ja]],["efficientnet",["EfficientNetForImageClassification",Ga]],["mobilenet_v1",["MobileNetV1ForImageClassification",Ha]],["mobilenet_v2",["MobileNetV2ForImageClassification",Ka]],["mobilenet_v3",["MobileNetV3ForImageClassification",tl]],["mobilenet_v4",["MobileNetV4ForImageClassification",nl]]]),Nl=new Map([["detr",["DetrForObjectDetection",uo]],["rt_detr",["RTDetrForObjectDetection",go]],["table-transformer",["TableTransformerForObjectDetection",xo]],["yolos",["YolosForObjectDetection",Mi]]]),Ol=new Map([["owlvit",["OwlViTForObjectDetection",to]],["owlv2",["Owlv2ForObjectDetection",no]],["grounding-dino",["GroundingDinoForObjectDetection",fi]]]),Bl=new Map([["detr",["DetrForSegmentation",_o]],["clipseg",["CLIPSegForImageSegmentation",Zs]]]),Vl=new Map([["segformer",["SegformerForSemanticSegmentation",Na]],["sapiens",["SapiensForSemanticSegmentation",qo]]]),Dl=new Map([["detr",["DetrForSegmentation",_o]],["maskformer",["MaskFormerForInstanceSegmentation",Yo]]]),Rl=new Map([["sam",["SamModel",yi]]]),Gl=new Map([["wav2vec2",["Wav2Vec2ForCTC",Ii]],["wav2vec2-bert",["Wav2Vec2BertForCTC",Yi]],["unispeech",["UniSpeechForCTC",Gi]],["unispeech-sat",["UniSpeechSatForCTC",Wi]],["wavlm",["WavLMForCTC",oa]],["hubert",["HubertForCTC",ta]]]),$l=new Map([["wav2vec2",["Wav2Vec2ForSequenceClassification",Li]],["wav2vec2-bert",["Wav2Vec2BertForSequenceClassification",Ki]],["unispeech",["UniSpeechForSequenceClassification",$i]],["unispeech-sat",["UniSpeechSatForSequenceClassification",Xi]],["wavlm",["WavLMForSequenceClassification",ia]],["hubert",["HubertForSequenceClassification",sa]],["audio-spectrogram-transformer",["ASTForAudioClassification",hs]]]),ql=new Map([["wavlm",["WavLMForXVector",aa]]]),Ul=new Map([["unispeech-sat",["UniSpeechSatForAudioFrameClassification",Qi]],["wavlm",["WavLMForAudioFrameClassification",la]],["wav2vec2",["Wav2Vec2ForAudioFrameClassification",zi]],["pyannote",["PyAnnoteForAudioFrameClassification",Oi]]]),Wl=new Map([["vitmatte",["VitMatteForImageMatting",Wn]]]),Xl=new Map([["patchtst",["PatchTSTForPrediction",ml]],["patchtsmixer",["PatchTSMixerForPrediction",gl]]]),Ql=new Map([["swin2sr",["Swin2SRForImageSuperResolution",Oo]]]),Hl=new Map([["dpt",["DPTForDepthEstimation",Do]],["depth_anything",["DepthAnythingForDepthEstimation",Go]],["glpn",["GLPNForDepthEstimation",ei]],["sapiens",["SapiensForDepthEstimation",Uo]],["depth_pro",["DepthProForDepthEstimation",Qo]]]),Jl=new Map([["sapiens",["SapiensForNormalEstimation",Wo]]]),Yl=new Map([["vitpose",["VitPoseForPoseEstimation",En]]]),Kl=new Map([["clip",["CLIPVisionModelWithProjection",Ds]],["siglip",["SiglipVisionModel",qs]],["jina_clip",["JinaCLIPVisionModel",Js]]]),Zl=[[Ml,b],[bl,x],[xl,T],[vl,b],[Pl,b],[Fl,y],[yl,y],[Cl,T],[Al,C],[Sl,b],[El,b],[Il,k],[Ll,P],[jl,b],[Bl,b],[Dl,b],[Vl,b],[Wl,b],[Xl,b],[Ql,b],[Hl,b],[Jl,b],[Yl,b],[Nl,b],[Ol,b],[Rl,v],[Gl,b],[$l,b],[kl,y],[Tl,b],[ql,b],[Ul,b],[Kl,b]];for(const[e,t]of Zl)for(const[s,r]of e.values())S.set(s,t),I.set(r,s),E.set(s,r);const ec=[["MusicgenForConditionalGeneration",Wa,F],["Phi3VForCausalLM",zs,A],["CLIPTextModelWithProjection",Bs,b],["SiglipTextModel",$s,b],["JinaCLIPTextModel",Hs,b],["ClapTextModelWithProjection",Aa,b],["ClapAudioModelWithProjection",Sa,b]];for(const[e,t,s]of ec)S.set(e,s),I.set(t,e),E.set(e,t);class tc extends wl{static MODEL_CLASS_MAPPINGS=Zl.map((e=>e[0]));static BASE_IF_FAIL=!0}class sc extends wl{static MODEL_CLASS_MAPPINGS=[vl]}class rc extends wl{static MODEL_CLASS_MAPPINGS=[Pl]}class nc extends wl{static MODEL_CLASS_MAPPINGS=[Fl]}class oc extends wl{static MODEL_CLASS_MAPPINGS=[yl]}class ic extends wl{static MODEL_CLASS_MAPPINGS=[kl]}class ac extends wl{static MODEL_CLASS_MAPPINGS=[Tl]}class lc extends wl{static MODEL_CLASS_MAPPINGS=[Cl]}class cc extends wl{static MODEL_CLASS_MAPPINGS=[Sl]}class dc extends wl{static MODEL_CLASS_MAPPINGS=[El]}class uc extends wl{static MODEL_CLASS_MAPPINGS=[Il]}class _c extends wl{static MODEL_CLASS_MAPPINGS=[jl]}class pc extends wl{static MODEL_CLASS_MAPPINGS=[Bl]}class mc extends wl{static MODEL_CLASS_MAPPINGS=[Vl]}class hc extends wl{static MODEL_CLASS_MAPPINGS=[Dl]}class fc extends wl{static MODEL_CLASS_MAPPINGS=[Nl]}class gc extends wl{static MODEL_CLASS_MAPPINGS=[Ol]}class wc extends wl{static MODEL_CLASS_MAPPINGS=[Rl]}class Mc extends wl{static MODEL_CLASS_MAPPINGS=[Gl]}class bc extends wl{static MODEL_CLASS_MAPPINGS=[$l]}class xc extends wl{static MODEL_CLASS_MAPPINGS=[ql]}class yc extends wl{static MODEL_CLASS_MAPPINGS=[Ul]}class kc extends wl{static MODEL_CLASS_MAPPINGS=[zl]}class Tc extends wl{static MODEL_CLASS_MAPPINGS=[Wl]}class vc extends wl{static MODEL_CLASS_MAPPINGS=[Ql]}class Pc extends wl{static MODEL_CLASS_MAPPINGS=[Hl]}class Fc extends wl{static MODEL_CLASS_MAPPINGS=[Jl]}class Cc extends wl{static MODEL_CLASS_MAPPINGS=[Yl]}class Ac extends wl{static MODEL_CLASS_MAPPINGS=[Kl]}class Sc extends J{constructor({logits:e,past_key_values:t,encoder_outputs:s,decoder_attentions:r=null,cross_attentions:n=null}){super(),this.logits=e,this.past_key_values=t,this.encoder_outputs=s,this.decoder_attentions=r,this.cross_attentions=n}}class Ec extends J{constructor({logits:e,...t}){super(),this.logits=e;const s=Object.values(t);s.length>0&&(this.attentions=s)}}class Ic extends J{constructor({logits:e,embeddings:t}){super(),this.logits=e,this.embeddings=t}}class Lc extends J{constructor({logits:e}){super(),this.logits=e}}class zc extends J{constructor({logits:e}){super(),this.logits=e}}class jc extends J{constructor({start_logits:e,end_logits:t}){super(),this.start_logits=e,this.end_logits=t}}class Nc extends J{constructor({logits:e}){super(),this.logits=e}}class Oc extends J{constructor({logits:e,past_key_values:t}){super(),this.logits=e,this.past_key_values=t}}class Bc extends J{constructor({alphas:e}){super(),this.alphas=e}}class Vc extends J{constructor({waveform:e,spectrogram:t}){super(),this.waveform=e,this.spectrogram=t}}},"./src/models/audio_spectrogram_transformer/feature_extraction_audio_spectrogram_transformer.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ASTFeatureExtractor:()=>o});var r=s("./src/base/feature_extraction_utils.js"),n=(s("./src/utils/tensor.js"),s("./src/utils/audio.js"));class o extends r.FeatureExtractor{constructor(e){super(e);const t=this.config.sampling_rate,s=(0,n.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<s.length;++e)s[e].push(0);this.mel_filters=s,this.window=(0,n.window_function)(400,"hann",{periodic:!1}),this.mean=this.config.mean,this.std=this.config.std}async _extract_fbank_features(e,t){return(0,n.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,max_num_frames:t,transpose:!0})}async _call(e){(0,r.validate_audio_inputs)(e,"ASTFeatureExtractor");const t=await this._extract_fbank_features(e,this.config.max_length);if(this.config.do_normalize){const e=2*this.std,s=t.data;for(let t=0;t<s.length;++t)s[t]=(s[t]-this.mean)/e}return{input_values:t.unsqueeze_(0)}}}},"./src/models/auto/feature_extraction_auto.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{AutoFeatureExtractor:()=>i});var r=s("./src/utils/constants.js"),n=s("./src/utils/hub.js"),o=(s("./src/base/feature_extraction_utils.js"),s("./src/models/feature_extractors.js"));class i{static async from_pretrained(e,t={}){const s=await(0,n.getModelJSON)(e,r.FEATURE_EXTRACTOR_NAME,!0,t),i=s.feature_extractor_type,a=o[i];if(!a)throw new Error(`Unknown feature_extractor_type: '${i}'. Please report this at ${r.GITHUB_ISSUE_URL}.`);return new a(s)}}},"./src/models/auto/image_processing_auto.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{AutoImageProcessor:()=>a});var r=s("./src/utils/constants.js"),n=s("./src/utils/hub.js"),o=s("./src/base/image_processors_utils.js"),i=s("./src/models/image_processors.js");class a{static async from_pretrained(e,t={}){const s=await(0,n.getModelJSON)(e,r.IMAGE_PROCESSOR_NAME,!0,t),a=s.image_processor_type??s.feature_extractor_type;let l=i[a];return l||(void 0!==a&&console.warn(`Image processor type '${a}' not found, assuming base ImageProcessor. Please report this at ${r.GITHUB_ISSUE_URL}.`),l=o.ImageProcessor),new l(s)}}},"./src/models/auto/processing_auto.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{AutoProcessor:()=>c});var r=s("./src/utils/constants.js"),n=s("./src/utils/hub.js"),o=s("./src/base/processing_utils.js"),i=s("./src/models/processors.js"),a=s("./src/models/image_processors.js"),l=s("./src/models/feature_extractors.js");class c{static async from_pretrained(e,t={}){const s=await(0,n.getModelJSON)(e,r.IMAGE_PROCESSOR_NAME,!0,t),{image_processor_type:c,feature_extractor_type:d,processor_class:u}=s;if(u&&i[u])return i[u].from_pretrained(e,t);if(!c&&!d)throw new Error("No `image_processor_type` or `feature_extractor_type` found in the config.");const _={};if(c){const e=a[c];if(!e)throw new Error(`Unknown image_processor_type: '${c}'.`);_.image_processor=new e(s)}if(d){const e=a[d];if(e)_.image_processor=new e(s);else{const e=l[d];if(!e)throw new Error(`Unknown feature_extractor_type: '${d}'.`);_.feature_extractor=new e(s)}}return new o.Processor({},_)}}},"./src/models/beit/image_processing_beit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{BeitFeatureExtractor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{}},"./src/models/bit/image_processing_bit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{BitImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{}},"./src/models/chinese_clip/image_processing_chinese_clip.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ChineseCLIPFeatureExtractor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{}},"./src/models/clap/feature_extraction_clap.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ClapFeatureExtractor:()=>o});var r=s("./src/base/feature_extraction_utils.js"),n=(s("./src/utils/tensor.js"),s("./src/utils/audio.js"));class o extends r.FeatureExtractor{constructor(e){super(e),this.mel_filters=(0,n.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,null,"htk"),this.mel_filters_slaney=(0,n.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,"slaney","slaney"),this.window=(0,n.window_function)(this.config.fft_window_size,"hann")}async _get_input_mel(e,t,s,r){let n,o=!1;const i=e.length-t;if(i>0){if("rand_trunc"!==s)throw new Error(`Truncation strategy "${s}" not implemented`);{o=!0;const s=Math.floor(Math.random()*(i+1));e=e.subarray(s,s+t),n=await this._extract_fbank_features(e,this.mel_filters_slaney,this.config.nb_max_samples)}}else{if(i<0){let s=new Float64Array(t);if(s.set(e),"repeat"===r)for(let r=e.length;r<t;r+=e.length)s.set(e.subarray(0,Math.min(e.length,t-r)),r);else if("repeatpad"===r)for(let t=e.length;t<-i;t+=e.length)s.set(e,t);e=s}if("fusion"===s)throw new Error(`Truncation strategy "${s}" not implemented`);n=await this._extract_fbank_features(e,this.mel_filters_slaney,this.config.nb_max_samples)}return n.unsqueeze_(0)}async _extract_fbank_features(e,t,s=null){return(0,n.spectrogram)(e,this.window,this.config.fft_window_size,this.config.hop_length,{power:2,mel_filters:t,log_mel:"dB",max_num_frames:s,do_pad:!1,transpose:!0})}async _call(e,{max_length:t=null}={}){(0,r.validate_audio_inputs)(e,"ClapFeatureExtractor");return{input_features:(await this._get_input_mel(e,t??this.config.nb_max_samples,this.config.truncation,this.config.padding)).unsqueeze_(0)}}}},"./src/models/clip/image_processing_clip.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{CLIPFeatureExtractor:()=>o,CLIPImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{}class o extends n{}},"./src/models/convnext/image_processing_convnext.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ConvNextFeatureExtractor:()=>o,ConvNextImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{constructor(e){super(e),this.crop_pct=this.config.crop_pct??.875}async resize(e){const t=this.size?.shortest_edge;if(void 0===t)throw new Error("Size dictionary must contain 'shortest_edge' key.");if(t<384){const s=Math.floor(t/this.crop_pct),[r,n]=this.get_resize_output_image_size(e,{shortest_edge:s});e=await e.resize(r,n,{resample:this.resample}),e=await e.center_crop(t,t)}else e=await e.resize(t,t,{resample:this.resample});return e}}class o extends n{}},"./src/models/deit/image_processing_deit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DeiTFeatureExtractor:()=>o,DeiTImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{}class o extends n{}},"./src/models/detr/image_processing_detr.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DetrFeatureExtractor:()=>i,DetrImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js"),n=s("./src/utils/tensor.js");class o extends r.ImageProcessor{async _call(e){const t=await super._call(e),s=[t.pixel_values.dims[0],64,64],r=(0,n.full)(s,1n);return{...t,pixel_mask:r}}post_process_object_detection(...e){return(0,r.post_process_object_detection)(...e)}post_process_panoptic_segmentation(...e){return(0,r.post_process_panoptic_segmentation)(...e)}post_process_instance_segmentation(...e){return(0,r.post_process_instance_segmentation)(...e)}}class i extends o{}},"./src/models/donut/image_processing_donut.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DonutFeatureExtractor:()=>o,DonutImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{pad_image(e,t,s,r={}){const[n,o,i]=t;let a=this.image_mean;Array.isArray(this.image_mean)||(a=new Array(i).fill(a));let l=this.image_std;Array.isArray(l)||(l=new Array(i).fill(a));const c=a.map(((e,t)=>-e/l[t]));return super.pad_image(e,t,s,{center:!0,constant_values:c,...r})}}class o extends n{}},"./src/models/dpt/image_processing_dpt.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DPTFeatureExtractor:()=>o,DPTImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{}class o extends n{}},"./src/models/efficientnet/image_processing_efficientnet.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{EfficientNetImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{constructor(e){super(e),this.include_top=this.config.include_top??!0,this.include_top&&(this.image_std=this.image_std.map((e=>e*e)))}}},"./src/models/feature_extractors.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ASTFeatureExtractor:()=>r.ASTFeatureExtractor,ClapFeatureExtractor:()=>n.ClapFeatureExtractor,ImageFeatureExtractor:()=>_.ImageProcessor,MoonshineFeatureExtractor:()=>o.MoonshineFeatureExtractor,PyAnnoteFeatureExtractor:()=>i.PyAnnoteFeatureExtractor,SeamlessM4TFeatureExtractor:()=>a.SeamlessM4TFeatureExtractor,SpeechT5FeatureExtractor:()=>l.SpeechT5FeatureExtractor,Wav2Vec2FeatureExtractor:()=>c.Wav2Vec2FeatureExtractor,WeSpeakerFeatureExtractor:()=>d.WeSpeakerFeatureExtractor,WhisperFeatureExtractor:()=>u.WhisperFeatureExtractor});var r=s("./src/models/audio_spectrogram_transformer/feature_extraction_audio_spectrogram_transformer.js"),n=s("./src/models/clap/feature_extraction_clap.js"),o=s("./src/models/moonshine/feature_extraction_moonshine.js"),i=s("./src/models/pyannote/feature_extraction_pyannote.js"),a=s("./src/models/seamless_m4t/feature_extraction_seamless_m4t.js"),l=s("./src/models/speecht5/feature_extraction_speecht5.js"),c=s("./src/models/wav2vec2/feature_extraction_wav2vec2.js"),d=s("./src/models/wespeaker/feature_extraction_wespeaker.js"),u=s("./src/models/whisper/feature_extraction_whisper.js"),_=s("./src/base/image_processors_utils.js")},"./src/models/florence2/processing_florence2.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Florence2Processor:()=>i});var r=s("./src/base/processing_utils.js"),n=s("./src/models/auto/image_processing_auto.js"),o=s("./src/tokenizers.js");class i extends r.Processor{static tokenizer_class=o.AutoTokenizer;static image_processor_class=n.AutoImageProcessor;constructor(e,t){super(e,t);const{tasks_answer_post_processing_type:s,task_prompts_without_inputs:r,task_prompts_with_input:n}=this.image_processor.config;this.tasks_answer_post_processing_type=new Map(Object.entries(s??{})),this.task_prompts_without_inputs=new Map(Object.entries(r??{})),this.task_prompts_with_input=new Map(Object.entries(n??{})),this.regexes={quad_boxes:/(.+?)<loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)>/gm,bboxes:/([^<]+)?<loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)>/gm},this.size_per_bin=1e3}construct_prompts(e){"string"==typeof e&&(e=[e]);const t=[];for(const s of e)if(this.task_prompts_without_inputs.has(s))t.push(this.task_prompts_without_inputs.get(s));else{for(const[e,r]of this.task_prompts_with_input)if(s.includes(e)){t.push(r.replaceAll("{input}",s).replaceAll(e,""));break}t.length!==e.length&&t.push(s)}return t}post_process_generation(e,t,s){const r=this.tasks_answer_post_processing_type.get(t)??"pure_text";let n;switch(e=e.replaceAll("<s>","").replaceAll("</s>",""),r){case"pure_text":n=e;break;case"description_with_bboxes":case"bboxes":case"phrase_grounding":case"ocr":const o="ocr"===r?"quad_boxes":"bboxes",i=e.matchAll(this.regexes[o]),a=[],l=[];for(const[e,t,...r]of i)a.push(t?t.trim():a.at(-1)??""),l.push(r.map(((e,t)=>(Number(e)+.5)/this.size_per_bin*s[t%2])));n={labels:a,[o]:l};break;default:throw new Error(`Task "${t}" (of type "${r}") not yet implemented.`)}return{[t]:n}}async _call(e,t=null,s={}){if(!e&&!t)throw new Error("Either text or images must be provided");return{...await this.image_processor(e,s),...t?this.tokenizer(t,s):{}}}}},"./src/models/glpn/image_processing_glpn.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{GLPNFeatureExtractor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{}},"./src/models/grounding_dino/image_processing_grounding_dino.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{GroundingDinoImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js"),n=s("./src/utils/tensor.js");class o extends r.ImageProcessor{async _call(e){const t=await super._call(e),s=t.pixel_values.dims,r=(0,n.ones)([s[0],s[2],s[3]]);return{...t,pixel_mask:r}}}},"./src/models/grounding_dino/processing_grounding_dino.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{GroundingDinoProcessor:()=>l});var r=s("./src/base/processing_utils.js"),n=s("./src/models/auto/image_processing_auto.js"),o=s("./src/tokenizers.js"),i=s("./src/base/image_processors_utils.js");function a(e,t){const s=e.dims.at(-1)-1,r=e.tolist();r.fill(!1,0,1),r.fill(!1,s);const n=t.tolist();return r.map(((e,t)=>e?t:null)).filter((e=>null!==e)).map((e=>n[e]))}class l extends r.Processor{static tokenizer_class=o.AutoTokenizer;static image_processor_class=n.AutoImageProcessor;async _call(e,t,s={}){const r=e?await this.image_processor(e,s):{};return{...t?this.tokenizer(t,s):{},...r}}post_process_grounded_object_detection(e,t,{box_threshold:s=.25,text_threshold:r=.25,target_sizes:n=null}={}){const{logits:o,pred_boxes:l}=e,c=o.dims[0];if(null!==n&&n.length!==c)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");const d=o.dims.at(1),u=o.sigmoid(),_=u.max(-1).tolist(),p=l.tolist().map((e=>e.map((e=>(0,i.center_to_corners_format)(e))))),m=[];for(let e=0;e<c;++e){const o=null!==n?n[e]:null;null!==o&&(p[e]=p[e].map((e=>e.map(((e,t)=>e*o[(t+1)%2])))));const i=_[e],l=[],c=[],h=[];for(let n=0;n<d;++n){const o=i[n];if(o<=s)continue;const d=p[e][n],_=u[e][n];l.push(o),h.push(d);const m=a(_.gt(r),t[e]);c.push(m)}m.push({scores:l,boxes:h,labels:this.batch_decode(c)})}return m}}},"./src/models/idefics3/image_processing_idefics3.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Idefics3ImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js"),n=s("./src/utils/tensor.js");class o extends r.ImageProcessor{constructor(e){super(e),this.do_image_splitting=e.do_image_splitting??!0,this.max_image_size=e.max_image_size}get_resize_for_vision_encoder(e,t){let[s,r]=e.dims.slice(-2);const n=r/s;return r>=s?(r=Math.ceil(r/t)*t,s=Math.floor(r/n),s=Math.ceil(s/t)*t):(s=Math.ceil(s/t)*t,r=Math.floor(s*n),r=Math.ceil(r/t)*t),{height:s,width:r}}async _call(e,{do_image_splitting:t=null,return_row_col_info:s=!1}={}){let r;if(Array.isArray(e)){if(0===e.length||!e[0])throw new Error("No images provided.");r=Array.isArray(e[0])?e:[e]}else r=[[e]];let o=[],i=[],a=[];const l=[],c=[];for(const e of r){let s=await Promise.all(e.map((e=>this.preprocess(e))));l.push(...s.map((e=>e.original_size))),c.push(...s.map((e=>e.reshaped_input_size))),s.forEach((e=>e.pixel_values.unsqueeze_(0)));const{longest_edge:r}=this.max_image_size;let d;if(t??this.do_image_splitting){let e=new Array(s.length),t=new Array(s.length);d=await Promise.all(s.map((async(s,o)=>{const i=this.get_resize_for_vision_encoder(s.pixel_values,r),a=await(0,n.interpolate_4d)(s.pixel_values,{size:[i.height,i.width]}),{frames:l,num_splits_h:c,num_splits_w:d}=await this.split_image(a,this.max_image_size);return e[o]=c,t[o]=d,(0,n.cat)(l,0)}))),i.push(e),a.push(t)}else{const e=[r,r];d=await Promise.all(s.map((t=>(0,n.interpolate_4d)(t.pixel_values,{size:e})))),i.push(new Array(s.length).fill(0)),a.push(new Array(s.length).fill(0))}o.push((0,n.cat)(d,0))}const d=o.length,[u,_,p,m]=o[0].dims;let h,f;if(1===d)h=o[0].unsqueeze_(0),f=(0,n.full)([d,u,p,m],!0);else{const e=Math.max(...o.map((e=>e.dims.at(0))));f=(0,n.full)([d,e,p,m],!0);const t=f.data,s=e*p*m;for(let r=0;r<d;++r){const i=o[r].dims[0];if(i<e){o[r]=(0,n.cat)([o[r],(0,n.full)([e-i,_,p,m],0)],0);const a=r*s+i*p*m,l=(r+1)*s;t.fill(!1,a,l)}}h=(0,n.stack)(o,0)}return{pixel_values:h,pixel_attention_mask:f,original_sizes:l,reshaped_input_sizes:c,...s?{rows:i,cols:a}:{}}}async split_image(e,{longest_edge:t}){const s=t,r=t,o=[],[i,a]=e.dims.slice(-2);let l=0,c=0;if(i>s||a>r){l=Math.ceil(i/s),c=Math.ceil(a/r);const t=Math.ceil(i/l),d=Math.ceil(a/c);for(let s=0;s<l;++s)for(let r=0;r<c;++r){let u,_,p,m;s===l-1?(_=i-t,m=i):(_=s*t,m=(s+1)*t),r===c-1?(u=a-d,p=a):(u=r*d,p=(r+1)*d);const h=[_,u],f=[m,p],g=await(0,n.slice)(e,h,f,[2,3]);o.push(g)}const u=s,_=r;i===u&&a===_||(e=await(0,n.interpolate_4d)(e,{size:[u,_]}))}return o.push(e),{frames:o,num_splits_h:l,num_splits_w:c}}}},"./src/models/idefics3/processing_idefics3.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Idefics3Processor:()=>l});var r=s("./src/base/processing_utils.js"),n=s("./src/models/auto/image_processing_auto.js"),o=s("./src/tokenizers.js"),i=(s("./src/utils/image.js"),s("./src/utils/core.js"));function a(e,t,s,r,n,o){return 0===e&&0===t?function(e,t,s,r){return`${t}${r}`+s.repeat(e)+`${t}`}(s,r,n,o):function(e,t,s,r,n,o){let i="";for(let o=0;o<t;++o){for(let t=0;t<s;++t)i+=r+`<row_${o+1}_col_${t+1}>`+n.repeat(e);i+="\n"}return i+=`\n${r}${o}`+n.repeat(e)+`${r}`,i}(s,e,t,r,n,o)}class l extends r.Processor{static image_processor_class=n.AutoImageProcessor;static tokenizer_class=o.AutoTokenizer;static uses_processor_config=!0;fake_image_token="<fake_token_around_image>";image_token="<image>";global_img_token="<global-img>";async _call(e,t=null,s={}){let r;s.return_row_col_info??=!0,t&&(r=await this.image_processor(t,s)),Array.isArray(e)||(e=[e]);const n=r.rows??[new Array(e.length).fill(0)],o=r.cols??[new Array(e.length).fill(0)],l=this.config.image_seq_len,c=[],d=[];for(let t=0;t<e.length;++t){const s=e[t],r=n[t],u=o[t];c.push((0,i.count)(s,this.image_token));const _=r.map(((e,t)=>a(e,u[t],l,this.fake_image_token,this.image_token,this.global_img_token))),p=s.split(this.image_token);if(0===p.length)throw new Error("The image token should be present in the text.");let m=p[0];for(let e=0;e<_.length;++e)m+=_[e]+p[e+1];d.push(m)}return{...this.tokenizer(d),...r}}}},"./src/models/image_processors.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{BeitFeatureExtractor:()=>r.BeitFeatureExtractor,BitImageProcessor:()=>n.BitImageProcessor,CLIPFeatureExtractor:()=>i.CLIPFeatureExtractor,CLIPImageProcessor:()=>i.CLIPImageProcessor,ChineseCLIPFeatureExtractor:()=>o.ChineseCLIPFeatureExtractor,ConvNextFeatureExtractor:()=>a.ConvNextFeatureExtractor,ConvNextImageProcessor:()=>a.ConvNextImageProcessor,DPTFeatureExtractor:()=>u.DPTFeatureExtractor,DPTImageProcessor:()=>u.DPTImageProcessor,DeiTFeatureExtractor:()=>l.DeiTFeatureExtractor,DeiTImageProcessor:()=>l.DeiTImageProcessor,DetrFeatureExtractor:()=>c.DetrFeatureExtractor,DetrImageProcessor:()=>c.DetrImageProcessor,DonutFeatureExtractor:()=>d.DonutFeatureExtractor,DonutImageProcessor:()=>d.DonutImageProcessor,EfficientNetImageProcessor:()=>_.EfficientNetImageProcessor,GLPNFeatureExtractor:()=>p.GLPNFeatureExtractor,GroundingDinoImageProcessor:()=>m.GroundingDinoImageProcessor,Idefics3ImageProcessor:()=>h.Idefics3ImageProcessor,JinaCLIPImageProcessor:()=>g.JinaCLIPImageProcessor,LlavaOnevisionImageProcessor:()=>w.LlavaOnevisionImageProcessor,Mask2FormerImageProcessor:()=>M.Mask2FormerImageProcessor,MaskFormerFeatureExtractor:()=>b.MaskFormerFeatureExtractor,MaskFormerImageProcessor:()=>b.MaskFormerImageProcessor,MobileNetV1FeatureExtractor:()=>x.MobileNetV1FeatureExtractor,MobileNetV1ImageProcessor:()=>x.MobileNetV1ImageProcessor,MobileNetV2FeatureExtractor:()=>y.MobileNetV2FeatureExtractor,MobileNetV2ImageProcessor:()=>y.MobileNetV2ImageProcessor,MobileNetV3FeatureExtractor:()=>k.MobileNetV3FeatureExtractor,MobileNetV3ImageProcessor:()=>k.MobileNetV3ImageProcessor,MobileNetV4FeatureExtractor:()=>T.MobileNetV4FeatureExtractor,MobileNetV4ImageProcessor:()=>T.MobileNetV4ImageProcessor,MobileViTFeatureExtractor:()=>v.MobileViTFeatureExtractor,MobileViTImageProcessor:()=>v.MobileViTImageProcessor,NougatImageProcessor:()=>P.NougatImageProcessor,OwlViTFeatureExtractor:()=>C.OwlViTFeatureExtractor,OwlViTImageProcessor:()=>C.OwlViTImageProcessor,Owlv2ImageProcessor:()=>F.Owlv2ImageProcessor,Phi3VImageProcessor:()=>A.Phi3VImageProcessor,PvtImageProcessor:()=>S.PvtImageProcessor,Qwen2VLImageProcessor:()=>E.Qwen2VLImageProcessor,RTDetrImageProcessor:()=>I.RTDetrImageProcessor,SamImageProcessor:()=>L.SamImageProcessor,SegformerFeatureExtractor:()=>z.SegformerFeatureExtractor,SegformerImageProcessor:()=>z.SegformerImageProcessor,SiglipImageProcessor:()=>j.SiglipImageProcessor,Swin2SRImageProcessor:()=>N.Swin2SRImageProcessor,VLMImageProcessor:()=>f.VLMImageProcessor,ViTFeatureExtractor:()=>O.ViTFeatureExtractor,ViTImageProcessor:()=>O.ViTImageProcessor,VitMatteImageProcessor:()=>B.VitMatteImageProcessor,VitPoseImageProcessor:()=>V.VitPoseImageProcessor,YolosFeatureExtractor:()=>D.YolosFeatureExtractor,YolosImageProcessor:()=>D.YolosImageProcessor});var r=s("./src/models/beit/image_processing_beit.js"),n=s("./src/models/bit/image_processing_bit.js"),o=s("./src/models/chinese_clip/image_processing_chinese_clip.js"),i=s("./src/models/clip/image_processing_clip.js"),a=s("./src/models/convnext/image_processing_convnext.js"),l=s("./src/models/deit/image_processing_deit.js"),c=s("./src/models/detr/image_processing_detr.js"),d=s("./src/models/donut/image_processing_donut.js"),u=s("./src/models/dpt/image_processing_dpt.js"),_=s("./src/models/efficientnet/image_processing_efficientnet.js"),p=s("./src/models/glpn/image_processing_glpn.js"),m=s("./src/models/grounding_dino/image_processing_grounding_dino.js"),h=s("./src/models/idefics3/image_processing_idefics3.js"),f=s("./src/models/janus/image_processing_janus.js"),g=s("./src/models/jina_clip/image_processing_jina_clip.js"),w=s("./src/models/llava_onevision/image_processing_llava_onevision.js"),M=s("./src/models/mask2former/image_processing_mask2former.js"),b=s("./src/models/maskformer/image_processing_maskformer.js"),x=s("./src/models/mobilenet_v1/image_processing_mobilenet_v1.js"),y=s("./src/models/mobilenet_v2/image_processing_mobilenet_v2.js"),k=s("./src/models/mobilenet_v3/image_processing_mobilenet_v3.js"),T=s("./src/models/mobilenet_v4/image_processing_mobilenet_v4.js"),v=s("./src/models/mobilevit/image_processing_mobilevit.js"),P=s("./src/models/nougat/image_processing_nougat.js"),F=s("./src/models/owlv2/image_processing_owlv2.js"),C=s("./src/models/owlvit/image_processing_owlvit.js"),A=s("./src/models/phi3_v/image_processing_phi3_v.js"),S=s("./src/models/pvt/image_processing_pvt.js"),E=s("./src/models/qwen2_vl/image_processing_qwen2_vl.js"),I=s("./src/models/rt_detr/image_processing_rt_detr.js"),L=s("./src/models/sam/image_processing_sam.js"),z=s("./src/models/segformer/image_processing_segformer.js"),j=s("./src/models/siglip/image_processing_siglip.js"),N=s("./src/models/swin2sr/image_processing_swin2sr.js"),O=s("./src/models/vit/image_processing_vit.js"),B=s("./src/models/vitmatte/image_processing_vitmatte.js"),V=s("./src/models/vitpose/image_processing_vitpose.js"),D=s("./src/models/yolos/image_processing_yolos.js")},"./src/models/janus/image_processing_janus.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{VLMImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{constructor(e){super({do_pad:!0,pad_size:{width:e.image_size,height:e.image_size},...e}),this.constant_values=this.config.background_color.map((e=>e*this.rescale_factor))}pad_image(e,t,s,r){return super.pad_image(e,t,s,{constant_values:this.constant_values,center:!0,...r})}}},"./src/models/janus/processing_janus.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{VLChatProcessor:()=>c});var r=s("./src/base/processing_utils.js"),n=s("./src/models/auto/image_processing_auto.js"),o=s("./src/tokenizers.js"),i=s("./src/utils/core.js"),a=s("./src/utils/tensor.js"),l=s("./src/utils/image.js");class c extends r.Processor{static image_processor_class=n.AutoImageProcessor;static tokenizer_class=o.AutoTokenizer;static uses_processor_config=!0;constructor(e,t){super(e,t),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(e,{images:t=null,chat_template:s="default"}={}){t?Array.isArray(t)||(t=[t]):t=await Promise.all(e.filter((e=>e.images)).flatMap((e=>e.images)).map((e=>l.RawImage.read(e))));const r=this.tokenizer,n=e=>r.encode(e,{add_special_tokens:!1}),o=r.apply_chat_template(e,{tokenize:!1,add_generation_prompt:!0,chat_template:s}).split(this.image_tag),c=o.length-1;if(t.length!==c)throw new Error(`Number of images provided (${t.length}) does not match number of "${this.image_tag}" image tags (${c})`);const[d,u,_]=r.model.convert_tokens_to_ids([this.image_tag,this.image_start_tag,this.image_end_tag]);let p=n(o[0]),m=new Array(p.length).fill(!1);for(let e=1;e<o.length;++e){const t=new Array(this.num_image_tokens).fill(d),s=n(o[e]);p=(0,i.mergeArrays)(p,[u],t,[_],s);const r=new Array(this.num_image_tokens).fill(!0);m=(0,i.mergeArrays)(m,[!1],r,[!1],new Array(s.length).fill(!1))}const h=[1,p.length],f={input_ids:new a.Tensor("int64",p,h),attention_mask:new a.Tensor("int64",new Array(p.length).fill(1),h),images_seq_mask:new a.Tensor("bool",m,h),images_emb_mask:new a.Tensor("bool",new Array(c*this.num_image_tokens).fill(!0),[1,c,this.num_image_tokens])};if(t&&t.length>0){const e=await this.image_processor(t);return e.pixel_values.unsqueeze_(0),{...f,...e}}return f}}},"./src/models/jina_clip/image_processing_jina_clip.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{JinaCLIPImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{constructor(e){const{resize_mode:t,fill_color:s,interpolation:r,size:n,...o}=e;super({...o,size:"squash"===t?{width:n,height:n}:"shortest"===t?{shortest_edge:n}:{longest_edge:n},resample:"bicubic"===r?3:2,do_center_crop:!0,crop_size:n,do_normalize:!0})}}},"./src/models/jina_clip/processing_jina_clip.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{JinaCLIPProcessor:()=>i});var r=s("./src/base/processing_utils.js"),n=s("./src/models/auto/image_processing_auto.js"),o=s("./src/tokenizers.js");class i extends r.Processor{static tokenizer_class=o.AutoTokenizer;static image_processor_class=n.AutoImageProcessor;async _call(e=null,t=null,s={}){if(!e&&!t)throw new Error("Either text or images must be provided");return{...e?this.tokenizer(e,s):{},...t?await this.image_processor(t,s):{}}}}},"./src/models/llava_onevision/image_processing_llava_onevision.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{LlavaOnevisionImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{}},"./src/models/mask2former/image_processing_mask2former.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Mask2FormerImageProcessor:()=>n});var r=s("./src/models/maskformer/image_processing_maskformer.js");class n extends r.MaskFormerImageProcessor{}},"./src/models/maskformer/image_processing_maskformer.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MaskFormerFeatureExtractor:()=>o,MaskFormerImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{post_process_panoptic_segmentation(...e){return(0,r.post_process_panoptic_segmentation)(...e)}post_process_instance_segmentation(...e){return(0,r.post_process_instance_segmentation)(...e)}}class o extends n{}},"./src/models/mgp_str/processing_mgp_str.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MgpstrProcessor:()=>l});var r=s("./src/base/processing_utils.js"),n=s("./src/models/auto/image_processing_auto.js"),o=s("./src/tokenizers.js"),i=s("./src/utils/maths.js");const a={char:["char_decode",1],bpe:["bpe_decode",2],wp:["wp_decode",102]};class l extends r.Processor{static tokenizer_class=o.AutoTokenizer;static image_processor_class=n.AutoImageProcessor;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(e,t){if(!a.hasOwnProperty(t))throw new Error(`Format ${t} is not supported.`);const[s,r]=a[t],n=this[s].bind(this),[o,l]=e.dims,c=[],d=[],u=e.tolist();for(let e=0;e<o;++e){const t=u[e],s=[],n=[];for(let e=1;e<l;++e){const[o,a]=(0,i.max)((0,i.softmax)(t[e]));if(n.push(o),a==r)break;s.push(a)}const o=n.length>0?n.reduce(((e,t)=>e*t),1):0;d.push(s),c.push(o)}return[n(d),c]}char_decode(e){return this.char_tokenizer.batch_decode(e).map((e=>e.replaceAll(" ","")))}bpe_decode(e){return this.bpe_tokenizer.batch_decode(e)}wp_decode(e){return this.wp_tokenizer.batch_decode(e).map((e=>e.replaceAll(" ","")))}batch_decode([e,t,s]){const[r,n]=this._decode_helper(e,"char"),[o,a]=this._decode_helper(t,"bpe"),[l,c]=this._decode_helper(s,"wp"),d=[],u=[];for(let e=0;e<r.length;++e){const[t,s]=(0,i.max)([n[e],a[e],c[e]]);d.push([r[e],o[e],l[e]][s]),u.push(t)}return{generated_text:d,scores:u,char_preds:r,bpe_preds:o,wp_preds:l}}static async from_pretrained(...e){const t=await super.from_pretrained(...e),s=await o.AutoTokenizer.from_pretrained("Xenova/gpt2"),r=await o.AutoTokenizer.from_pretrained("Xenova/bert-base-uncased");return t.components={image_processor:t.image_processor,char_tokenizer:t.tokenizer,bpe_tokenizer:s,wp_tokenizer:r},t}async _call(e,t=null){const s=await this.image_processor(e);return t&&(s.labels=this.tokenizer(t).input_ids),s}}},"./src/models/mobilenet_v1/image_processing_mobilenet_v1.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MobileNetV1FeatureExtractor:()=>o,MobileNetV1ImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{}class o extends n{}},"./src/models/mobilenet_v2/image_processing_mobilenet_v2.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MobileNetV2FeatureExtractor:()=>o,MobileNetV2ImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{}class o extends n{}},"./src/models/mobilenet_v3/image_processing_mobilenet_v3.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MobileNetV3FeatureExtractor:()=>o,MobileNetV3ImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{}class o extends n{}},"./src/models/mobilenet_v4/image_processing_mobilenet_v4.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MobileNetV4FeatureExtractor:()=>o,MobileNetV4ImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{}class o extends n{}},"./src/models/mobilevit/image_processing_mobilevit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MobileViTFeatureExtractor:()=>o,MobileViTImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{}class o extends n{}},"./src/models/moonshine/feature_extraction_moonshine.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MoonshineFeatureExtractor:()=>o});var r=s("./src/base/feature_extraction_utils.js"),n=s("./src/utils/tensor.js");class o extends r.FeatureExtractor{async _call(e){(0,r.validate_audio_inputs)(e,"MoonshineFeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));const t=[1,e.length];return{input_values:new n.Tensor("float32",e,t)}}}},"./src/models/moonshine/processing_moonshine.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MoonshineProcessor:()=>i});var r=s("./src/models/auto/feature_extraction_auto.js"),n=s("./src/tokenizers.js"),o=s("./src/base/processing_utils.js");class i extends o.Processor{static tokenizer_class=n.AutoTokenizer;static feature_extractor_class=r.AutoFeatureExtractor;async _call(e){return await this.feature_extractor(e)}}},"./src/models/nougat/image_processing_nougat.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{NougatImageProcessor:()=>n});var r=s("./src/models/donut/image_processing_donut.js");class n extends r.DonutImageProcessor{}},"./src/models/owlv2/image_processing_owlv2.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Owlv2ImageProcessor:()=>n});var r=s("./src/models/owlvit/image_processing_owlvit.js");class n extends r.OwlViTImageProcessor{}},"./src/models/owlvit/image_processing_owlvit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{OwlViTFeatureExtractor:()=>o,OwlViTImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{post_process_object_detection(...e){return(0,r.post_process_object_detection)(...e)}}class o extends n{}},"./src/models/owlvit/processing_owlvit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{OwlViTProcessor:()=>i});var r=s("./src/base/processing_utils.js"),n=s("./src/models/auto/image_processing_auto.js"),o=s("./src/tokenizers.js");class i extends r.Processor{static tokenizer_class=o.AutoTokenizer;static image_processor_class=n.AutoImageProcessor}},"./src/models/paligemma/processing_paligemma.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{PaliGemmaProcessor:()=>a});var r=s("./src/base/processing_utils.js"),n=s("./src/models/auto/image_processing_auto.js"),o=s("./src/tokenizers.js");const i="<image>";class a extends r.Processor{static tokenizer_class=o.AutoTokenizer;static image_processor_class=n.AutoImageProcessor;static uses_processor_config=!1;async _call(e,t=null,s={}){t||(console.warn("You are using PaliGemma without a text prefix. It will perform as a picture-captioning model."),t=""),Array.isArray(e)||(e=[e]),Array.isArray(t)||(t=[t]);const r=this.tokenizer.bos_token,n=this.image_processor.config.image_seq_length;let o;t.some((e=>e.includes(i)))?o=t.map((e=>{const t=e.replaceAll(i,i.repeat(n)),s=t.lastIndexOf(i),o=-1===s?0:s+7;return t.slice(0,o)+r+t.slice(o)+"\n"})):(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 `<image>` tokens in the very beginning of your text. For this call, we will infer how many images each text has and add special tokens."),o=t.map((t=>function(e,t,s,r,n){return`${r.repeat(s*n)}${t}${e}\n`}(t,r,n,i,e.length))));const a=this.tokenizer(o,s);return{...await this.image_processor(e,s),...a}}}},"./src/models/phi3_v/image_processing_phi3_v.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Phi3VImageProcessor:()=>d});var r=s("./src/base/image_processors_utils.js"),n=s("./src/utils/tensor.js");const o=336,i=[2,3],{ceil:a,floor:l,sqrt:c}=Math;class d extends r.ImageProcessor{constructor(e){super({...e,do_normalize:!0,do_pad:!0,pad_size:"custom",do_convert_rgb:!0,do_resize:!0}),this._num_crops=e.num_crops}calc_num_image_tokens_from_image_size(e,t){const{num_img_tokens:s}=this.config;return l((l(t/o)*l(e/o)+1)*s+1+(l(t/o)+1)*c(s))}get_resize_output_image_size(e,t){const s=this._num_crops,[r,n]=e.size;let o=r/n,i=1;for(;i*Math.ceil(i/o)<=s;)i+=1;i-=1;const a=Math.floor(336*i);return[a,Math.floor(a/o)]}pad_image(e,t,s,r={}){const[n,i]=t,l=o*a(n/o),c=o*a(i/o),d=[1,1,1].map(((e,t)=>(e-this.image_mean[t])/this.image_std[t]));return super.pad_image(e,t,{width:c,height:l},{center:!0,constant_values:d,...r})}async _call(e,{num_crops:t=null}={}){if(this._num_crops=t??=this.config.num_crops,t<4||c(t)%1!=0)throw new Error("num_crops must be a square number >= 4");Array.isArray(e)||(e=[e]);const s=e.length,r=await Promise.all(e.map((e=>this.preprocess(e)))),d=r.map((e=>e.original_size)),u=r.map((e=>e.reshaped_input_size)),_=[];for(const{pixel_values:e}of r){e.unsqueeze_(0);const[s,r]=e.dims.slice(-2),a=await(0,n.interpolate_4d)(e,{size:[o,o],mode:"bicubic"});if(t>0){const d=[],u=c(t),p=l(r/u),m=l(s/u);for(let t=0;t<u;++t)for(let o=0;o<u;++o){let a,l,c,_;t===u-1?(l=s-m,_=s):(l=t*m,_=(t+1)*m),o===u-1?(a=r-p,c=r):(a=o*p,c=(o+1)*p);const h=[l,a],f=[_,c],g=await(0,n.slice)(e,h,f,i);d.push(g)}const h=await(0,n.interpolate_4d)((0,n.cat)(d,0),{size:[o,o],mode:"bicubic"});_.push((0,n.cat)([a,h],0))}else _.push(a)}const p=(0,n.stack)(_,0),m=u.map((e=>e.map((e=>o*a(e/o)))));return{pixel_values:p,original_sizes:d,reshaped_input_sizes:u,image_sizes:new n.Tensor("int64",m.flat(),[s,2]),num_img_tokens:m.map((([e,t])=>this.calc_num_image_tokens_from_image_size(t,e)))}}}},"./src/models/phi3_v/processing_phi3_v.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Phi3VProcessor:()=>l});var r=s("./src/base/processing_utils.js"),n=s("./src/models/auto/image_processing_auto.js"),o=s("./src/tokenizers.js");s("./src/utils/image.js");const i="<|image|>",a=/<\|image_\d+\|>/g;class l extends r.Processor{static image_processor_class=n.AutoImageProcessor;static tokenizer_class=o.AutoTokenizer;async _call(e,t=null,{padding:s=!0,truncation:r=!0,num_crops:n=null}={}){let o,l;if(Array.isArray(e)||(e=[e]),t){l=await this.image_processor(t,{num_crops:n});const{num_img_tokens:c}=l,d=e.map(((e,t)=>e.split(a).join(i.repeat(c[t]))));o=this.tokenizer(d,{padding:s,truncation:r});const u=this.tokenizer.model.convert_tokens_to_ids([i])[0];o.input_ids.map_((e=>e==u?-e:e))}else o=this.tokenizer(e);return{...o,...l}}}},"./src/models/processors.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Florence2Processor:()=>r.Florence2Processor,GroundingDinoProcessor:()=>n.GroundingDinoProcessor,Idefics3Processor:()=>o.Idefics3Processor,JinaCLIPProcessor:()=>a.JinaCLIPProcessor,MgpstrProcessor:()=>l.MgpstrProcessor,MoonshineProcessor:()=>c.MoonshineProcessor,OwlViTProcessor:()=>d.OwlViTProcessor,PaliGemmaProcessor:()=>_.PaliGemmaProcessor,Phi3VProcessor:()=>u.Phi3VProcessor,PyAnnoteProcessor:()=>p.PyAnnoteProcessor,Qwen2VLProcessor:()=>m.Qwen2VLProcessor,SamProcessor:()=>h.SamProcessor,SpeechT5Processor:()=>f.SpeechT5Processor,VLChatProcessor:()=>i.VLChatProcessor,Wav2Vec2Processor:()=>g.Wav2Vec2Processor,Wav2Vec2ProcessorWithLM:()=>w.Wav2Vec2ProcessorWithLM,WhisperProcessor:()=>M.WhisperProcessor});var r=s("./src/models/florence2/processing_florence2.js"),n=s("./src/models/grounding_dino/processing_grounding_dino.js"),o=s("./src/models/idefics3/processing_idefics3.js"),i=s("./src/models/janus/processing_janus.js"),a=s("./src/models/jina_clip/processing_jina_clip.js"),l=s("./src/models/mgp_str/processing_mgp_str.js"),c=s("./src/models/moonshine/processing_moonshine.js"),d=s("./src/models/owlvit/processing_owlvit.js"),u=s("./src/models/phi3_v/processing_phi3_v.js"),_=s("./src/models/paligemma/processing_paligemma.js"),p=s("./src/models/pyannote/processing_pyannote.js"),m=s("./src/models/qwen2_vl/processing_qwen2_vl.js"),h=s("./src/models/sam/processing_sam.js"),f=s("./src/models/speecht5/processing_speecht5.js"),g=s("./src/models/wav2vec2/processing_wav2vec2.js"),w=s("./src/models/wav2vec2_with_lm/processing_wav2vec2_with_lm.js"),M=s("./src/models/whisper/processing_whisper.js")},"./src/models/pvt/image_processing_pvt.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{PvtImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{}},"./src/models/pyannote/feature_extraction_pyannote.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{PyAnnoteFeatureExtractor:()=>i});var r=s("./src/base/feature_extraction_utils.js"),n=s("./src/utils/tensor.js"),o=s("./src/utils/maths.js");class i extends r.FeatureExtractor{async _call(e){(0,r.validate_audio_inputs)(e,"PyAnnoteFeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));const t=[1,1,e.length];return{input_values:new n.Tensor("float32",e,t)}}samples_to_frames(e){return(e-this.config.offset)/this.config.step}post_process_speaker_diarization(e,t){const s=t/this.samples_to_frames(t)/this.config.sampling_rate,r=[];for(const t of e.tolist()){const e=[];let n=-1;for(let s=0;s<t.length;++s){const r=(0,o.softmax)(t[s]),[i,a]=(0,o.max)(r),[l,c]=[s,s+1];a!==n?(n=a,e.push({id:a,start:l,end:c,score:i})):(e.at(-1).end=c,e.at(-1).score+=i)}r.push(e.map((({id:e,start:t,end:r,score:n})=>({id:e,start:t*s,end:r*s,confidence:n/(r-t)}))))}return r}}},"./src/models/pyannote/processing_pyannote.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{PyAnnoteProcessor:()=>o});var r=s("./src/base/processing_utils.js"),n=s("./src/models/pyannote/feature_extraction_pyannote.js");class o extends r.Processor{static feature_extractor_class=n.PyAnnoteFeatureExtractor;async _call(e){return await this.feature_extractor(e)}post_process_speaker_diarization(...e){return this.feature_extractor.post_process_speaker_diarization(...e)}get sampling_rate(){return this.feature_extractor.config.sampling_rate}}},"./src/models/qwen2_vl/image_processing_qwen2_vl.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Qwen2VLImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js"),n=s("./src/utils/tensor.js");class o extends r.ImageProcessor{async _call(e,...t){const{pixel_values:s,original_sizes:r,reshaped_input_sizes:o}=await super._call(e,...t);let i=s;const{temporal_patch_size:a,merge_size:l,patch_size:c}=this.config;1===i.dims[0]&&(i=(0,n.cat)(Array.from({length:a},(()=>i)),0));const d=i.dims[0]/a,u=i.dims[1],_=Math.floor(i.dims[2]/c),p=Math.floor(i.dims[3]/c);return{pixel_values:i.view(d,a,u,Math.floor(_/l),l,c,Math.floor(p/l),l,c).permute(0,3,6,4,7,2,1,5,8).view(d*_*p,u*a*c*c),image_grid_thw:new n.Tensor("int64",[d,_,p],[1,3]),original_sizes:r,reshaped_input_sizes:o}}}},"./src/models/qwen2_vl/processing_qwen2_vl.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Qwen2VLProcessor:()=>i});var r=s("./src/base/processing_utils.js"),n=s("./src/models/auto/image_processing_auto.js"),o=s("./src/tokenizers.js");s("./src/utils/image.js");class i extends r.Processor{static image_processor_class=n.AutoImageProcessor;static tokenizer_class=o.AutoTokenizer;async _call(e,t=null,...s){let r,n;if(Array.isArray(e)||(e=[e]),t&&(r=await this.image_processor(t),n=r.image_grid_thw),n){let t=this.image_processor.config.merge_size**2,s=0;const r=n.tolist();e=e.map((e=>{for(;e.includes("<|image_pad|>");){const n=Number(r[s++].reduce(((e,t)=>e*t),1n));e=e.replace("<|image_pad|>","<|placeholder|>".repeat(Math.floor(n/t)))}return e.replaceAll("<|placeholder|>","<|image_pad|>")}))}return{...this.tokenizer(e),...r}}}},"./src/models/rt_detr/image_processing_rt_detr.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{RTDetrImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{post_process_object_detection(...e){return(0,r.post_process_object_detection)(...e)}}},"./src/models/sam/image_processing_sam.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SamImageProcessor:()=>i});var r=s("./src/base/image_processors_utils.js"),n=s("./src/utils/core.js"),o=s("./src/utils/tensor.js");class i extends r.ImageProcessor{reshape_input_points(e,t,s,r=!1){e=structuredClone(e);let i=(0,n.calculateDimensions)(e);if(3===i.length)r||(i=[1,...i]),e=[e];else if(4!==i.length)throw Error("The input_points must be a 4D tensor of shape `batch_size`, `point_batch_size`, `nb_points_per_image`, `2`.");for(let r=0;r<e.length;++r){let n=t[r],o=s[r],i=[o[0]/n[0],o[1]/n[1]];for(let t=0;t<e[r].length;++t)for(let s=0;s<e[r][t].length;++s)for(let n=0;n<e[r][t][s].length;++n)e[r][t][s][n]*=i[n%2]}return new o.Tensor("float32",Float32Array.from(e.flat(1/0)),i)}add_input_labels(e,t){let s=(0,n.calculateDimensions)(e);if(2===s.length)s=[1,...s],e=[e];else if(3!==s.length)throw Error("The input_points must be a 4D tensor of shape `batch_size`, `point_batch_size`, `nb_points_per_image`, `2`.");if(s.some(((e,s)=>e!==t.dims[s])))throw Error(`The first ${s.length} dimensions of 'input_points' and 'input_labels' must be the same.`);return new o.Tensor("int64",e.flat(1/0).map(BigInt),s)}async _call(e,{input_points:t=null,input_labels:s=null,input_boxes:r=null}={}){const n=await super._call(e);if(t&&(n.input_points=this.reshape_input_points(t,n.original_sizes,n.reshaped_input_sizes)),s){if(!n.input_points)throw Error("`input_points` must be provided if `input_labels` are provided.");n.input_labels=this.add_input_labels(s,n.input_points)}return r&&(n.input_boxes=this.reshape_input_points(r,n.original_sizes,n.reshaped_input_sizes,!0)),n}async post_process_masks(e,t,s,{mask_threshold:r=0,binarize:n=!0,pad_size:i=null}={}){const a=[],l=[(i=i??this.pad_size).height,i.width];for(let i=0;i<t.length;++i){const c=t[i],d=s[i];let u=await(0,o.interpolate_4d)(e[i],{mode:"bilinear",size:l});if(u=u.slice(null,null,[0,d[0]],[0,d[1]]),u=await(0,o.interpolate_4d)(u,{mode:"bilinear",size:c}),n){const e=u.data,t=new Uint8Array(e.length);for(let s=0;s<e.length;++s)e[s]>r&&(t[s]=1);u=new o.Tensor("bool",t,u.dims)}a.push(u)}return a}generate_crop_boxes(e,t,{crop_n_layers:s=0,overlap_ratio:r=512/1500,points_per_crop:n=32,crop_n_points_downscale_factor:o=1}={}){}}},"./src/models/sam/processing_sam.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SamProcessor:()=>o});var r=s("./src/base/processing_utils.js"),n=s("./src/models/auto/image_processing_auto.js");class o extends r.Processor{static image_processor_class=n.AutoImageProcessor;async _call(...e){return await this.image_processor(...e)}post_process_masks(...e){return this.image_processor.post_process_masks(...e)}reshape_input_points(...e){return this.image_processor.reshape_input_points(...e)}}},"./src/models/seamless_m4t/feature_extraction_seamless_m4t.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SeamlessM4TFeatureExtractor:()=>i});var r=s("./src/base/feature_extraction_utils.js"),n=s("./src/utils/tensor.js"),o=s("./src/utils/audio.js");class i extends r.FeatureExtractor{constructor(e){super(e);const t=this.config.sampling_rate,s=(0,o.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<s.length;++e)s[e].push(0);this.mel_filters=s,this.window=(0,o.window_function)(400,"povey",{periodic:!1})}async _extract_fbank_features(e,t){return e=e.map((e=>32768*e)),(0,o.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,max_num_frames:t,transpose:!0})}async _call(e,{padding:t=!0,pad_to_multiple_of:s=2,do_normalize_per_mel_bins:o=!0,return_attention_mask:i=!0}={}){(0,r.validate_audio_inputs)(e,"SeamlessM4TFeatureExtractor");let a,l=await this._extract_fbank_features(e,this.config.max_length);if(o){const[e,t]=l.dims,s=l.data;for(let r=0;r<t;++r){let n=0;for(let o=0;o<e;++o)n+=s[o*t+r];const o=n/e;let i=0;for(let n=0;n<e;++n)i+=(s[n*t+r]-o)**2;i/=e-1;const a=Math.sqrt(i+1e-7);for(let n=0;n<e;++n){const e=n*t+r;s[e]=(s[e]-o)/a}}}if(t){const[e,t]=l.dims,r=l.data,o=e%s;if(o>0){const s=new Float32Array(t*(e+o));s.set(r),s.fill(this.config.padding_value,r.length);const c=e+o;l=new n.Tensor(l.type,s,[c,t]),i&&(a=new n.Tensor("int64",new BigInt64Array(c),[1,c]),a.data.fill(1n,0,e))}}const[c,d]=l.dims,u=this.config.stride;if(0!==c%u)throw new Error(`The number of frames (${c}) must be a multiple of the stride (${u}).`);const _=l.view(1,Math.floor(c/u),d*u),p={input_features:_};if(i){const e=_.dims[1],t=new BigInt64Array(e);if(a){const e=a.data;for(let s=1,r=0;s<c;s+=u,++r)t[r]=e[s]}else t.fill(1n);p.attention_mask=new n.Tensor("int64",t,[1,e])}return p}}},"./src/models/segformer/image_processing_segformer.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SegformerFeatureExtractor:()=>o,SegformerImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{post_process_semantic_segmentation(...e){return(0,r.post_process_semantic_segmentation)(...e)}}class o extends n{}},"./src/models/siglip/image_processing_siglip.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SiglipImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{}},"./src/models/speecht5/feature_extraction_speecht5.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SpeechT5FeatureExtractor:()=>n});var r=s("./src/base/feature_extraction_utils.js");class n extends r.FeatureExtractor{}},"./src/models/speecht5/processing_speecht5.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SpeechT5Processor:()=>i});var r=s("./src/base/processing_utils.js"),n=s("./src/tokenizers.js"),o=s("./src/models/auto/feature_extraction_auto.js");class i extends r.Processor{static tokenizer_class=n.AutoTokenizer;static feature_extractor_class=o.AutoFeatureExtractor;async _call(e){return await this.feature_extractor(e)}}},"./src/models/swin2sr/image_processing_swin2sr.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Swin2SRImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{pad_image(e,t,s,r={}){const[n,o,i]=t;return super.pad_image(e,t,{width:o+(s-o%s)%s,height:n+(s-n%s)%s},{mode:"symmetric",center:!1,constant_values:-1,...r})}}},"./src/models/vit/image_processing_vit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ViTFeatureExtractor:()=>o,ViTImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{}class o extends n{}},"./src/models/vitmatte/image_processing_vitmatte.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{VitMatteImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js"),n=s("./src/utils/tensor.js");class o extends r.ImageProcessor{async _call(e,t){Array.isArray(e)||(e=[e]),Array.isArray(t)||(t=[t]);const s=await Promise.all(e.map((e=>this.preprocess(e)))),r=await Promise.all(t.map((e=>this.preprocess(e,{do_normalize:!1,do_convert_rgb:!1,do_convert_grayscale:!0}))));return{pixel_values:(0,n.stack)(s.map(((e,t)=>(0,n.cat)([e.pixel_values,r[t].pixel_values],0))),0),original_sizes:s.map((e=>e.original_size)),reshaped_input_sizes:s.map((e=>e.reshaped_input_size))}}}},"./src/models/vitpose/image_processing_vitpose.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{VitPoseImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{post_process_pose_estimation(e,t,{threshold:s=null}={}){const r=e.tolist(),[n,o,i,a]=e.dims,l=[];for(let e=0;e<n;++e){const n=r[e],o=t[e],c=[];for(let e=0;e<o.length;++e){const t=o[e],r=[],l=[],d=[],u=t.at(-2)/a,_=t.at(-1)/i;for(let e=0;e<n.length;++e){let[t,o]=[0,0],i=0,a=-1/0;const c=n[e];for(let e=0;e<c.length;++e){const s=c[e];for(let r=0;r<s.length;++r){const n=s[r];i+=n,a=Math.max(a,n),t+=(r+.5)*n,o+=e*n}}if(null!=s&&a<s)continue;const p=[u*t/i,_*o/i];r.push(p),d.push(e),l.push(a)}c.push({bbox:t,scores:l,labels:d,keypoints:r})}l.push(c)}return l}}},"./src/models/wav2vec2/feature_extraction_wav2vec2.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Wav2Vec2FeatureExtractor:()=>o});var r=s("./src/base/feature_extraction_utils.js"),n=s("./src/utils/tensor.js");class o extends r.FeatureExtractor{_zero_mean_unit_var_norm(e){const t=e.reduce(((e,t)=>e+t),0)/e.length,s=e.reduce(((e,s)=>e+(s-t)**2),0)/e.length;return e.map((e=>(e-t)/Math.sqrt(s+1e-7)))}async _call(e){(0,r.validate_audio_inputs)(e,"Wav2Vec2FeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));let t=e;this.config.do_normalize&&(t=this._zero_mean_unit_var_norm(t));const s=[1,t.length];return{input_values:new n.Tensor("float32",t,s),attention_mask:new n.Tensor("int64",new BigInt64Array(t.length).fill(1n),s)}}}},"./src/models/wav2vec2/processing_wav2vec2.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Wav2Vec2Processor:()=>i});var r=s("./src/tokenizers.js"),n=s("./src/models/auto/feature_extraction_auto.js"),o=s("./src/base/processing_utils.js");class i extends o.Processor{static tokenizer_class=r.AutoTokenizer;static feature_extractor_class=n.AutoFeatureExtractor;async _call(e){return await this.feature_extractor(e)}}},"./src/models/wav2vec2_with_lm/processing_wav2vec2_with_lm.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Wav2Vec2ProcessorWithLM:()=>i});var r=s("./src/tokenizers.js"),n=s("./src/models/auto/feature_extraction_auto.js"),o=s("./src/base/processing_utils.js");class i extends o.Processor{static tokenizer_class=r.AutoTokenizer;static feature_extractor_class=n.AutoFeatureExtractor;async _call(e){return await this.feature_extractor(e)}}},"./src/models/wespeaker/feature_extraction_wespeaker.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{WeSpeakerFeatureExtractor:()=>o});var r=s("./src/base/feature_extraction_utils.js"),n=(s("./src/utils/tensor.js"),s("./src/utils/audio.js"));class o extends r.FeatureExtractor{constructor(e){super(e);const t=this.config.sampling_rate,s=(0,n.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<s.length;++e)s[e].push(0);this.mel_filters=s,this.window=(0,n.window_function)(400,"hamming",{periodic:!1}),this.min_num_frames=this.config.min_num_frames}async _extract_fbank_features(e){return e=e.map((e=>32768*e)),(0,n.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,transpose:!0,min_num_frames:this.min_num_frames})}async _call(e){(0,r.validate_audio_inputs)(e,"WeSpeakerFeatureExtractor");const t=(await this._extract_fbank_features(e)).unsqueeze_(0);if(null===this.config.fbank_centering_span){const e=t.mean(1).data,s=t.data,[r,n,o]=t.dims;for(let t=0;t<r;++t){const r=t*n*o,i=t*o;for(let t=0;t<n;++t){const n=r+t*o;for(let t=0;t<o;++t)s[n+t]-=e[i+t]}}}return{input_features:t}}}},"./src/models/whisper/common_whisper.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{WHISPER_LANGUAGE_MAPPING:()=>n,WHISPER_TO_LANGUAGE_CODE_MAPPING:()=>o,whisper_language_to_code:()=>i});const r=[["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"]],n=new Map(r),o=new Map([...r.map((([e,t])=>[t,e])),["burmese","my"],["valencian","ca"],["flemish","nl"],["haitian","ht"],["letzeburgesch","lb"],["pushto","ps"],["panjabi","pa"],["moldavian","ro"],["moldovan","ro"],["sinhalese","si"],["castilian","es"]]);function i(e){e=e.toLowerCase();let t=o.get(e);if(void 0===t){if(!n.has(e)){const t=2===e.length?n.keys():n.values();throw new Error(`Language "${e}" is not supported. Must be one of: ${JSON.stringify(t)}`)}t=e}return t}},"./src/models/whisper/feature_extraction_whisper.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{WhisperFeatureExtractor:()=>i});var r=s("./src/base/feature_extraction_utils.js"),n=(s("./src/utils/tensor.js"),s("./src/utils/audio.js")),o=s("./src/utils/maths.js");class i extends r.FeatureExtractor{constructor(e){super(e),this.config.mel_filters??=(0,n.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,n.window_function)(this.config.n_fft,"hann")}async _extract_fbank_features(e){const t=await(0,n.spectrogram)(e,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}),s=t.data,r=(0,o.max)(s)[0];for(let e=0;e<s.length;++e)s[e]=(Math.max(s[e],r-8)+4)/4;return t}async _call(e){let t;(0,r.validate_audio_inputs)(e,"WhisperFeatureExtractor"),e.length>this.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`."),t=e.slice(0,this.config.n_samples)):(t=new Float32Array(this.config.n_samples),t.set(e));return{input_features:(await this._extract_fbank_features(t)).unsqueeze_(0)}}}},"./src/models/whisper/generation_whisper.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{WhisperGenerationConfig:()=>n});var r=s("./src/generation/configuration_utils.js");class n extends r.GenerationConfig{return_timestamps=null;return_token_timestamps=null;num_frames=null;alignment_heads=null;task=null;language=null;no_timestamps_token_id=null;prompt_ids=null;is_multilingual=null;lang_to_id=null;task_to_id=null;max_initial_timestamp_index=1}},"./src/models/whisper/processing_whisper.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{WhisperProcessor:()=>i});var r=s("./src/models/auto/feature_extraction_auto.js"),n=s("./src/tokenizers.js"),o=s("./src/base/processing_utils.js");class i extends o.Processor{static tokenizer_class=n.AutoTokenizer;static feature_extractor_class=r.AutoFeatureExtractor;async _call(e){return await this.feature_extractor(e)}}},"./src/models/yolos/image_processing_yolos.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{YolosFeatureExtractor:()=>o,YolosImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js");class n extends r.ImageProcessor{post_process_object_detection(...e){return(0,r.post_process_object_detection)(...e)}}class o extends n{}},"./src/ops/registry.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{TensorOpRegistry:()=>l});var r=s("./src/backends/onnx.js"),n=s("./src/utils/tensor.js"),o=s("./src/env.js");const i=o.apis.IS_BROWSER_ENV||o.apis.IS_WEBWORKER_ENV,a=async(e,t,s)=>{const o=await(0,r.createInferenceSession)(new Uint8Array(e),t);let a=Promise.resolve();return async e=>{const t=(0,r.isONNXProxy)(),l=Object.fromEntries(Object.entries(e).map((([e,s])=>[e,(t?s.clone():s).ort_tensor]))),c=await(a=i?a.then((()=>o.run(l))):o.run(l));return Array.isArray(s)?s.map((e=>new n.Tensor(c[e]))):new n.Tensor(c[s])}};class l{static session_options={};static get nearest_interpolate_4d(){return this._nearest_interpolate_4d||(this._nearest_interpolate_4d=a([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=a([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=a([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=a([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=a([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=a([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=a([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=a([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}}},"./src/pipelines.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{AudioClassificationPipeline:()=>C,AutomaticSpeechRecognitionPipeline:()=>S,DepthEstimationPipeline:()=>D,DocumentQuestionAnsweringPipeline:()=>O,FeatureExtractionPipeline:()=>P,FillMaskPipeline:()=>M,ImageClassificationPipeline:()=>I,ImageFeatureExtractionPipeline:()=>F,ImageSegmentationPipeline:()=>L,ImageToImagePipeline:()=>V,ImageToTextPipeline:()=>E,ObjectDetectionPipeline:()=>j,Pipeline:()=>h,QuestionAnsweringPipeline:()=>w,SummarizationPipeline:()=>x,Text2TextGenerationPipeline:()=>b,TextClassificationPipeline:()=>f,TextGenerationPipeline:()=>T,TextToAudioPipeline:()=>B,TokenClassificationPipeline:()=>g,TranslationPipeline:()=>y,ZeroShotAudioClassificationPipeline:()=>A,ZeroShotClassificationPipeline:()=>v,ZeroShotImageClassificationPipeline:()=>z,ZeroShotObjectDetectionPipeline:()=>N,pipeline:()=>$});var r=s("./src/tokenizers.js"),n=s("./src/models.js"),o=s("./src/models/auto/processing_auto.js"),i=(s("./src/base/processing_utils.js"),s("./src/utils/generic.js")),a=s("./src/utils/core.js"),l=s("./src/utils/maths.js"),c=s("./src/utils/audio.js"),d=s("./src/utils/tensor.js"),u=s("./src/utils/image.js");async function _(e){return Array.isArray(e)||(e=[e]),await Promise.all(e.map((e=>u.RawImage.read(e))))}async function p(e,t){return Array.isArray(e)||(e=[e]),await Promise.all(e.map((e=>"string"==typeof e||e instanceof URL?(0,c.read_audio)(e,t):e instanceof Float64Array?new Float32Array(e):e)))}function m(e,t){t&&(e=e.map((e=>0|e)));const[s,r,n,o]=e;return{xmin:s,ymin:r,xmax:n,ymax:o}}class h extends i.Callable{constructor({task:e,model:t,tokenizer:s=null,processor:r=null}){super(),this.task=e,this.model=t,this.tokenizer=s,this.processor=r}async dispose(){await this.model.dispose()}}class f extends h{constructor(e){super(e)}async _call(e,{top_k:t=1}={}){const s=this.tokenizer(e,{padding:!0,truncation:!0}),r=await this.model(s),n="multi_label_classification"===this.model.config.problem_type?e=>e.sigmoid():e=>new d.Tensor("float32",(0,l.softmax)(e.data),e.dims),o=this.model.config.id2label,i=[];for(const e of r.logits){const s=n(e),r=await(0,d.topk)(s,t),a=r[0].tolist(),l=r[1].tolist().map(((e,t)=>({label:o?o[e]:`LABEL_${e}`,score:a[t]})));1===t?i.push(...l):i.push(l)}return Array.isArray(e)||1===t?i:i[0]}}class g extends h{constructor(e){super(e)}async _call(e,{ignore_labels:t=["O"]}={}){const s=Array.isArray(e),r=this.tokenizer(s?e:[e],{padding:!0,truncation:!0}),n=(await this.model(r)).logits,o=this.model.config.id2label,i=[];for(let e=0;e<n.dims[0];++e){const s=r.input_ids[e],a=n[e],c=[];for(let e=0;e<a.dims[0];++e){const r=a[e],n=(0,l.max)(r.data)[1],i=o?o[n]:`LABEL_${n}`;if(t.includes(i))continue;const d=this.tokenizer.decode([s[e].item()],{skip_special_tokens:!0});if(""===d)continue;const u=(0,l.softmax)(r.data);c.push({entity:i,score:u[n],index:e,word:d})}i.push(c)}return s?i:i[0]}}class w extends h{constructor(e){super(e)}async _call(e,t,{top_k:s=1}={}){const r=this.tokenizer(e,{text_pair:t,padding:!0,truncation:!0}),{start_logits:n,end_logits:o}=await this.model(r),i=r.input_ids.tolist(),c=r.attention_mask.tolist(),d=this.tokenizer.all_special_ids,u=[];for(let e=0;e<n.dims[0];++e){const t=i[e],r=t.findIndex((e=>e==this.tokenizer.sep_token_id)),_=(c[e].map(((e,s)=>1==e&&(0===s||s>r&&-1===d.findIndex((e=>e==t[s]))))),n[e].tolist()),p=o[e].tolist();for(let s=1;s<_.length;++s)(0==c[e]||s<=r||-1!==d.findIndex((e=>e==t[s])))&&(_[s]=-1/0,p[s]=-1/0);const m=(0,l.softmax)(_).map(((e,t)=>[e,t])),h=(0,l.softmax)(p).map(((e,t)=>[e,t]));m[0][0]=0,h[0][0]=0;const f=(0,a.product)(m,h).filter((e=>e[0][1]<=e[1][1])).map((e=>[e[0][1],e[1][1],e[0][0]*e[1][0]])).sort(((e,t)=>t[2]-e[2]));for(let e=0;e<Math.min(f.length,s);++e){const[s,r,n]=f[e],o=t.slice(s,r+1),i=this.tokenizer.decode(o,{skip_special_tokens:!0});u.push({answer:i,score:n})}}return 1===s?u[0]:u}}class M extends h{constructor(e){super(e)}async _call(e,{top_k:t=5}={}){const s=this.tokenizer(e,{padding:!0,truncation:!0}),{logits:r}=await this.model(s),n=[],o=s.input_ids.tolist();for(let e=0;e<o.length;++e){const s=o[e],i=s.findIndex((e=>e==this.tokenizer.mask_token_id));if(-1===i)throw Error(`Mask token (${this.tokenizer.mask_token}) not found in text.`);const a=r[e][i],c=await(0,d.topk)(new d.Tensor("float32",(0,l.softmax)(a.data),a.dims),t),u=c[0].tolist(),_=c[1].tolist();n.push(_.map(((e,t)=>{const r=s.slice();return r[i]=e,{score:u[t],token:Number(e),token_str:this.tokenizer.decode([e]),sequence:this.tokenizer.decode(r,{skip_special_tokens:!0})}})))}return Array.isArray(e)?n:n[0]}}class b extends h{_key="generated_text";constructor(e){super(e)}async _call(e,t={}){Array.isArray(e)||(e=[e]),this.model.config.prefix&&(e=e.map((e=>this.model.config.prefix+e)));const s=this.model.config.task_specific_params;s&&s[this.task]&&s[this.task].prefix&&(e=e.map((e=>s[this.task].prefix+e)));const r=this.tokenizer,n={padding:!0,truncation:!0};let o;o=this instanceof y&&"_build_translation_inputs"in r?r._build_translation_inputs(e,n,t):r(e,n);const i=await this.model.generate({...o,...t});return r.batch_decode(i,{skip_special_tokens:!0}).map((e=>({[this._key]:e})))}}class x extends b{_key="summary_text";constructor(e){super(e)}}class y extends b{_key="translation_text";constructor(e){super(e)}}function k(e){return Array.isArray(e)&&e.every((e=>"role"in e&&"content"in e))}class T extends h{constructor(e){super(e)}async _call(e,t={}){let s,r=!1,n=!1;if("string"==typeof e)s=e=[e];else if(Array.isArray(e)&&e.every((e=>"string"==typeof e)))r=!0,s=e;else{if(k(e))e=[e];else{if(!Array.isArray(e)||!e.every(k))throw new Error("Input must be a string, an array of strings, a Chat, or an array of Chats");r=!0}n=!0,s=e.map((e=>this.tokenizer.apply_chat_template(e,{tokenize:!1,add_generation_prompt:!0})))}const o=t.add_special_tokens??!1,i=!n&&(t.return_full_text??!0);this.tokenizer.padding_side="left";const a=this.tokenizer(s,{add_special_tokens:o,padding:!0,truncation:!0}),l=await this.model.generate({...a,...t}),c=this.tokenizer.batch_decode(l,{skip_special_tokens:!0});let d;!i&&a.input_ids.dims.at(-1)>0&&(d=this.tokenizer.batch_decode(a.input_ids,{skip_special_tokens:!0}).map((e=>e.length)));const u=Array.from({length:e.length},(e=>[]));for(let t=0;t<c.length;++t){const s=Math.floor(t/l.dims[0]*e.length);d&&(c[t]=c[t].slice(d[s])),u[s].push({generated_text:n?[...e[s],{role:"assistant",content:c[t]}]:c[t]})}return r||1!==u.length?u:u[0]}}class v extends h{constructor(e){super(e),this.label2id=Object.fromEntries(Object.entries(this.model.config.label2id).map((([e,t])=>[e.toLowerCase(),t]))),this.entailment_id=this.label2id.entailment,void 0===this.entailment_id&&(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,void 0===this.contradiction_id&&(console.warn("Could not find 'contradiction' in label2id mapping. Using 0 as contradiction_id."),this.contradiction_id=0)}async _call(e,t,{hypothesis_template:s="This example is {}.",multi_label:r=!1}={}){const n=Array.isArray(e);n||(e=[e]),Array.isArray(t)||(t=[t]);const o=t.map((e=>s.replace("{}",e))),i=r||1===t.length,a=[];for(const s of e){const e=[];for(const t of o){const r=this.tokenizer(s,{text_pair:t,padding:!0,truncation:!0}),n=await this.model(r);i?e.push([n.logits.data[this.contradiction_id],n.logits.data[this.entailment_id]]):e.push(n.logits.data[this.entailment_id])}const r=(i?e.map((e=>(0,l.softmax)(e)[1])):(0,l.softmax)(e)).map(((e,t)=>[e,t])).sort(((e,t)=>t[0]-e[0]));a.push({sequence:s,labels:r.map((e=>t[e[1]])),scores:r.map((e=>e[0]))})}return n?a:a[0]}}class P extends h{constructor(e){super(e)}async _call(e,{pooling:t="none",normalize:s=!1,quantize:r=!1,precision:n="binary"}={}){const o=this.tokenizer(e,{padding:!0,truncation:!0}),i=await this.model(o);let a=i.last_hidden_state??i.logits??i.token_embeddings;if("none"===t);else if("mean"===t)a=(0,d.mean_pooling)(a,o.attention_mask);else{if("cls"!==t)throw Error(`Pooling method '${t}' not supported.`);a=a.slice(null,0)}return s&&(a=a.normalize(2,-1)),r&&(a=(0,d.quantize_embeddings)(a,n)),a}}class F extends h{constructor(e){super(e)}async _call(e,{pool:t=null}={}){const s=await _(e),{pixel_values:r}=await this.processor(s),n=await this.model({pixel_values:r});let o;if(t){if(!("pooler_output"in n))throw Error("No pooled output was returned. Make sure the model has a 'pooler' layer when using the 'pool' option.");o=n.pooler_output}else o=n.last_hidden_state??n.logits??n.image_embeds;return o}}class C extends h{constructor(e){super(e)}async _call(e,{top_k:t=5}={}){const s=this.processor.feature_extractor.config.sampling_rate,r=await p(e,s),n=this.model.config.id2label,o=[];for(const e of r){const s=await this.processor(e),r=(await this.model(s)).logits[0],i=await(0,d.topk)(new d.Tensor("float32",(0,l.softmax)(r.data),r.dims),t),a=i[0].tolist(),c=i[1].tolist().map(((e,t)=>({label:n?n[e]:`LABEL_${e}`,score:a[t]})));o.push(c)}return Array.isArray(e)?o:o[0]}}class A extends h{constructor(e){super(e)}async _call(e,t,{hypothesis_template:s="This is a sound of {}."}={}){const r=!Array.isArray(e);r&&(e=[e]);const n=t.map((e=>s.replace("{}",e))),o=this.tokenizer(n,{padding:!0,truncation:!0}),i=this.processor.feature_extractor.config.sampling_rate,a=await p(e,i),c=[];for(const e of a){const s=await this.processor(e),r=await this.model({...o,...s}),n=(0,l.softmax)(r.logits_per_audio.data);c.push([...n].map(((e,s)=>({score:e,label:t[s]}))))}return r?c[0]:c}}class S extends h{constructor(e){super(e)}async _call(e,t={}){switch(this.model.config.model_type){case"whisper":return this._call_whisper(e,t);case"wav2vec2":case"wav2vec2-bert":case"unispeech":case"unispeech-sat":case"hubert":return this._call_wav2vec2(e,t);case"moonshine":return this._call_moonshine(e,t);default:throw new Error(`AutomaticSpeechRecognitionPipeline does not support model type '${this.model.config.model_type}'.`)}}async _call_wav2vec2(e,t){t.language&&console.warn('`language` parameter is not yet supported for `wav2vec2` models, defaulting to "English".'),t.task&&console.warn('`task` parameter is not yet supported for `wav2vec2` models, defaulting to "transcribe".');const s=!Array.isArray(e);s&&(e=[e]);const r=this.processor.feature_extractor.config.sampling_rate,n=await p(e,r),o=[];for(const e of n){const t=await this.processor(e),s=(await this.model(t)).logits[0],r=[];for(const e of s)r.push((0,l.max)(e.data)[1]);const n=this.tokenizer.decode(r);o.push({text:n})}return s?o[0]:o}async _call_whisper(e,t){const s=t.return_timestamps??!1,r=t.chunk_length_s??0,n=t.force_full_sequences??!1;let o=t.stride_length_s??null;const i={...t};"word"===s&&(i.return_token_timestamps=!0,i.return_timestamps=!1);const a=!Array.isArray(e);a&&(e=[e]);const c=this.processor.feature_extractor.config.chunk_length/this.model.config.max_source_positions,d=this.processor.feature_extractor.config.hop_length,u=this.processor.feature_extractor.config.sampling_rate,_=await p(e,u),m=[];for(const e of _){let t=[];if(r>0){if(null===o)o=r/6;else if(r<=o)throw Error("`chunk_length_s` must be larger than `stride_length_s`.");const s=u*r,n=u*o,i=s-2*n;let a=0;for(;;){const r=a+s,o=e.subarray(a,r),l=await this.processor(o),c=0===a,d=r>=e.length;if(t.push({stride:[o.length,c?0:n,d?0:n],input_features:l.input_features,is_last:d}),d)break;a+=i}}else t=[{stride:[e.length,0,0],input_features:(await this.processor(e)).input_features,is_last:!0}];for(const e of t){i.num_frames=Math.floor(e.stride[0]/d);const t=await this.model.generate({inputs:e.input_features,...i});"word"===s?(e.tokens=t.sequences.tolist()[0],e.token_timestamps=t.token_timestamps.tolist()[0].map((e=>(0,l.round)(e,2)))):e.tokens=t[0].tolist(),e.stride=e.stride.map((e=>e/u))}const[a,_]=this.tokenizer._decode_asr(t,{time_precision:c,return_timestamps:s,force_full_sequences:n});m.push({text:a,..._})}return a?m[0]:m}async _call_moonshine(e,t){const s=!Array.isArray(e);s&&(e=[e]);const r=this.processor.feature_extractor.config.sampling_rate,n=await p(e,r),o=[];for(const e of n){const s=await this.processor(e),n=6*Math.floor(e.length/r),i=await this.model.generate({max_new_tokens:n,...t,...s}),a=this.processor.batch_decode(i,{skip_special_tokens:!0})[0];o.push({text:a})}return s?o[0]:o}}class E extends h{constructor(e){super(e)}async _call(e,t={}){const s=Array.isArray(e),r=await _(e),{pixel_values:n}=await this.processor(r),o=[];for(const e of n){e.dims=[1,...e.dims];const s=await this.model.generate({inputs:e,...t}),r=this.tokenizer.batch_decode(s,{skip_special_tokens:!0}).map((e=>({generated_text:e.trim()})));o.push(r)}return s?o:o[0]}}class I extends h{constructor(e){super(e)}async _call(e,{top_k:t=5}={}){const s=await _(e),{pixel_values:r}=await this.processor(s),n=await this.model({pixel_values:r}),o=this.model.config.id2label,i=[];for(const e of n.logits){const s=await(0,d.topk)(new d.Tensor("float32",(0,l.softmax)(e.data),e.dims),t),r=s[0].tolist(),n=s[1].tolist().map(((e,t)=>({label:o?o[e]:`LABEL_${e}`,score:r[t]})));i.push(n)}return Array.isArray(e)?i:i[0]}}class L extends h{constructor(e){super(e),this.subtasks_mapping={panoptic:"post_process_panoptic_segmentation",instance:"post_process_instance_segmentation",semantic:"post_process_semantic_segmentation"}}async _call(e,{threshold:t=.5,mask_threshold:s=.5,overlap_mask_area_threshold:r=.8,label_ids_to_fuse:n=null,target_sizes:o=null,subtask:i=null}={}){if(Array.isArray(e)&&1!==e.length)throw Error("Image segmentation pipeline currently only supports a batch size of 1.");const a=await _(e),l=a.map((e=>[e.height,e.width])),{pixel_values:c,pixel_mask:d}=await this.processor(a),p=await this.model({pixel_values:c,pixel_mask:d});let m=null;if(null!==i)m=this.subtasks_mapping[i];else for(let[e,t]of Object.entries(this.subtasks_mapping))if(t in this.processor.image_processor){m=this.processor.image_processor[t].bind(this.processor.image_processor),i=e;break}const h=this.model.config.id2label,f=[];if("panoptic"===i||"instance"===i){const e=m(p,t,s,r,n,o??l)[0],i=e.segmentation;for(const t of e.segments_info){const e=new Uint8ClampedArray(i.data.length);for(let s=0;s<i.data.length;++s)i.data[s]===t.id&&(e[s]=255);const s=new u.RawImage(e,i.dims[1],i.dims[0],1);f.push({score:t.score,label:h[t.label_id],mask:s})}}else{if("semantic"!==i)throw Error(`Subtask ${i} not supported.`);{const{segmentation:e,labels:t}=m(p,o??l)[0];for(const s of t){const t=new Uint8ClampedArray(e.data.length);for(let r=0;r<e.data.length;++r)e.data[r]===s&&(t[r]=255);const r=new u.RawImage(t,e.dims[1],e.dims[0],1);f.push({score:null,label:h[s],mask:r})}}}return f}}class z extends h{constructor(e){super(e)}async _call(e,t,{hypothesis_template:s="This is a photo of {}"}={}){const r=Array.isArray(e),n=await _(e),o=t.map((e=>s.replace("{}",e))),i=this.tokenizer(o,{padding:"siglip"!==this.model.config.model_type||"max_length",truncation:!0}),{pixel_values:a}=await this.processor(n),c=await this.model({...i,pixel_values:a}),d="siglip"===this.model.config.model_type?e=>e.sigmoid().data:e=>(0,l.softmax)(e.data),u=[];for(const e of c.logits_per_image){const s=[...d(e)].map(((e,s)=>({score:e,label:t[s]})));s.sort(((e,t)=>t.score-e.score)),u.push(s)}return r?u:u[0]}}class j extends h{constructor(e){super(e)}async _call(e,{threshold:t=.9,percentage:s=!1}={}){const r=Array.isArray(e);if(r&&1!==e.length)throw Error("Object detection pipeline currently only supports a batch size of 1.");const n=await _(e),o=s?null:n.map((e=>[e.height,e.width])),{pixel_values:i,pixel_mask:a}=await this.processor(n),l=await this.model({pixel_values:i,pixel_mask:a}),c=this.processor.image_processor.post_process_object_detection(l,t,o),d=this.model.config.id2label,u=c.map((e=>e.boxes.map(((t,r)=>({score:e.scores[r],label:d[e.classes[r]],box:m(t,!s)})))));return r?u:u[0]}}class N extends h{constructor(e){super(e)}async _call(e,t,{threshold:s=.1,top_k:r=null,percentage:n=!1}={}){const o=Array.isArray(e),i=await _(e),a=this.tokenizer(t,{padding:!0,truncation:!0}),l=await this.processor(i),c=[];for(let e=0;e<i.length;++e){const o=i[e],d=n?null:[[o.height,o.width]],u=l.pixel_values[e].unsqueeze_(0),_=await this.model({...a,pixel_values:u});let p;if("post_process_grounded_object_detection"in this.processor){const e=this.processor.post_process_grounded_object_detection(_,a.input_ids,{box_threshold:s,text_threshold:s,target_sizes:d})[0];p=e.boxes.map(((t,s)=>({score:e.scores[s],label:e.labels[s],box:m(t,!n)})))}else{const e=this.processor.image_processor.post_process_object_detection(_,s,d,!0)[0];p=e.boxes.map(((s,r)=>({score:e.scores[r],label:t[e.classes[r]],box:m(s,!n)})))}p.sort(((e,t)=>t.score-e.score)),null!==r&&(p=p.slice(0,r)),c.push(p)}return o?c:c[0]}}class O extends h{constructor(e){super(e)}async _call(e,t,s={}){const r=(await _(e))[0],{pixel_values:n}=await this.processor(r),o=`<s_docvqa><s_question>${t}</s_question><s_answer>`,i=this.tokenizer(o,{add_special_tokens:!1,padding:!0,truncation:!0}).input_ids,a=await this.model.generate({inputs:n,max_length:this.model.config.decoder.max_position_embeddings,decoder_input_ids:i,...s}),l=this.tokenizer.batch_decode(a)[0].match(/<s_answer>(.*?)<\/s_answer>/);let c=null;return l&&l.length>=2&&(c=l[1].trim()),[{answer:c}]}}class B extends h{DEFAULT_VOCODER_ID="Xenova/speecht5_hifigan";constructor(e){super(e),this.vocoder=e.vocoder??null}async _call(e,{speaker_embeddings:t=null}={}){return this.processor?this._call_text_to_spectrogram(e,{speaker_embeddings:t}):this._call_text_to_waveform(e)}async _call_text_to_waveform(e){const t=this.tokenizer(e,{padding:!0,truncation:!0}),{waveform:s}=await this.model(t),r=this.model.config.sampling_rate;return new c.RawAudio(s.data,r)}async _call_text_to_spectrogram(e,{speaker_embeddings:t}){if(this.vocoder||(console.log("No vocoder specified, using default HifiGan vocoder."),this.vocoder=await n.AutoModel.from_pretrained(this.DEFAULT_VOCODER_ID,{dtype:"fp32"})),("string"==typeof t||t instanceof URL)&&(t=new Float32Array(await(await fetch(t)).arrayBuffer())),t instanceof Float32Array)t=new d.Tensor("float32",t,[1,t.length]);else if(!(t instanceof d.Tensor))throw new Error("Speaker embeddings must be a `Tensor`, `Float32Array`, `string`, or `URL`.");const{input_ids:s}=this.tokenizer(e,{padding:!0,truncation:!0}),{waveform:r}=await this.model.generate_speech(s,t,{vocoder:this.vocoder}),o=this.processor.feature_extractor.config.sampling_rate;return new c.RawAudio(r.data,o)}}class V extends h{constructor(e){super(e)}async _call(e){const t=await _(e),s=await this.processor(t),r=await this.model(s),n=[];for(const e of r.reconstruction){const t=e.squeeze().clamp_(0,1).mul_(255).round_().to("uint8");n.push(u.RawImage.fromTensor(t))}return n.length>1?n:n[0]}}class D extends h{constructor(e){super(e)}async _call(e){const t=await _(e),s=await this.processor(t),{predicted_depth:r}=await this.model(s),n=[];for(let e=0;e<t.length;++e){const s=r[e],[o,i]=s.dims.slice(-2),[a,l]=t[e].size,c=(await(0,d.interpolate_4d)(s.view(1,1,o,i),{size:[l,a],mode:"bilinear"})).view(l,a),_=c.min().item(),p=c.max().item(),m=c.sub(_).div_(p-_).mul_(255).to("uint8").unsqueeze(0),h=u.RawImage.fromTensor(m);n.push({predicted_depth:c,depth:h})}return n.length>1?n:n[0]}}const R=Object.freeze({"text-classification":{tokenizer:r.AutoTokenizer,pipeline:f,model:n.AutoModelForSequenceClassification,default:{model:"Xenova/distilbert-base-uncased-finetuned-sst-2-english"},type:"text"},"token-classification":{tokenizer:r.AutoTokenizer,pipeline:g,model:n.AutoModelForTokenClassification,default:{model:"Xenova/bert-base-multilingual-cased-ner-hrl"},type:"text"},"question-answering":{tokenizer:r.AutoTokenizer,pipeline:w,model:n.AutoModelForQuestionAnswering,default:{model:"Xenova/distilbert-base-cased-distilled-squad"},type:"text"},"fill-mask":{tokenizer:r.AutoTokenizer,pipeline:M,model:n.AutoModelForMaskedLM,default:{model:"Xenova/bert-base-uncased"},type:"text"},summarization:{tokenizer:r.AutoTokenizer,pipeline:x,model:n.AutoModelForSeq2SeqLM,default:{model:"Xenova/distilbart-cnn-6-6"},type:"text"},translation:{tokenizer:r.AutoTokenizer,pipeline:y,model:n.AutoModelForSeq2SeqLM,default:{model:"Xenova/t5-small"},type:"text"},"text2text-generation":{tokenizer:r.AutoTokenizer,pipeline:b,model:n.AutoModelForSeq2SeqLM,default:{model:"Xenova/flan-t5-small"},type:"text"},"text-generation":{tokenizer:r.AutoTokenizer,pipeline:T,model:n.AutoModelForCausalLM,default:{model:"Xenova/gpt2"},type:"text"},"zero-shot-classification":{tokenizer:r.AutoTokenizer,pipeline:v,model:n.AutoModelForSequenceClassification,default:{model:"Xenova/distilbert-base-uncased-mnli"},type:"text"},"audio-classification":{pipeline:C,model:n.AutoModelForAudioClassification,processor:o.AutoProcessor,default:{model:"Xenova/wav2vec2-base-superb-ks"},type:"audio"},"zero-shot-audio-classification":{tokenizer:r.AutoTokenizer,pipeline:A,model:n.AutoModel,processor:o.AutoProcessor,default:{model:"Xenova/clap-htsat-unfused"},type:"multimodal"},"automatic-speech-recognition":{tokenizer:r.AutoTokenizer,pipeline:S,model:[n.AutoModelForSpeechSeq2Seq,n.AutoModelForCTC],processor:o.AutoProcessor,default:{model:"Xenova/whisper-tiny.en"},type:"multimodal"},"text-to-audio":{tokenizer:r.AutoTokenizer,pipeline:B,model:[n.AutoModelForTextToWaveform,n.AutoModelForTextToSpectrogram],processor:[o.AutoProcessor,null],default:{model:"Xenova/speecht5_tts"},type:"text"},"image-to-text":{tokenizer:r.AutoTokenizer,pipeline:E,model:n.AutoModelForVision2Seq,processor:o.AutoProcessor,default:{model:"Xenova/vit-gpt2-image-captioning"},type:"multimodal"},"image-classification":{pipeline:I,model:n.AutoModelForImageClassification,processor:o.AutoProcessor,default:{model:"Xenova/vit-base-patch16-224"},type:"multimodal"},"image-segmentation":{pipeline:L,model:[n.AutoModelForImageSegmentation,n.AutoModelForSemanticSegmentation,n.AutoModelForUniversalSegmentation],processor:o.AutoProcessor,default:{model:"Xenova/detr-resnet-50-panoptic"},type:"multimodal"},"zero-shot-image-classification":{tokenizer:r.AutoTokenizer,pipeline:z,model:n.AutoModel,processor:o.AutoProcessor,default:{model:"Xenova/clip-vit-base-patch32"},type:"multimodal"},"object-detection":{pipeline:j,model:n.AutoModelForObjectDetection,processor:o.AutoProcessor,default:{model:"Xenova/detr-resnet-50"},type:"multimodal"},"zero-shot-object-detection":{tokenizer:r.AutoTokenizer,pipeline:N,model:n.AutoModelForZeroShotObjectDetection,processor:o.AutoProcessor,default:{model:"Xenova/owlvit-base-patch32"},type:"multimodal"},"document-question-answering":{tokenizer:r.AutoTokenizer,pipeline:O,model:n.AutoModelForDocumentQuestionAnswering,processor:o.AutoProcessor,default:{model:"Xenova/donut-base-finetuned-docvqa"},type:"multimodal"},"image-to-image":{pipeline:V,model:n.AutoModelForImageToImage,processor:o.AutoProcessor,default:{model:"Xenova/swin2SR-classical-sr-x2-64"},type:"image"},"depth-estimation":{pipeline:D,model:n.AutoModelForDepthEstimation,processor:o.AutoProcessor,default:{model:"Xenova/dpt-large"},type:"image"},"feature-extraction":{tokenizer:r.AutoTokenizer,pipeline:P,model:n.AutoModel,default:{model:"Xenova/all-MiniLM-L6-v2"},type:"text"},"image-feature-extraction":{processor:o.AutoProcessor,pipeline:F,model:[n.AutoModelForImageFeatureExtraction,n.AutoModel],default:{model:"Xenova/vit-base-patch16-224-in21k"},type:"image"}}),G=Object.freeze({"sentiment-analysis":"text-classification",ner:"token-classification",asr:"automatic-speech-recognition","text-to-speech":"text-to-audio",embeddings:"feature-extraction"});async function $(e,t=null,{progress_callback:s=null,config:r=null,cache_dir:n=null,local_files_only:o=!1,revision:i="main",device:l=null,dtype:c=null,model_file_name:d=null,session_options:u={}}={}){e=G[e]??e;const _=R[e.split("_",1)[0]];if(!_)throw Error(`Unsupported pipeline: ${e}. Must be one of [${Object.keys(R)}]`);t||(t=_.default.model,console.log(`No model specified. Using default model: "${t}".`));const p={progress_callback:s,config:r,cache_dir:n,local_files_only:o,revision:i,device:l,dtype:c,model_file_name:d,session_options:u},m=new Map([["tokenizer",_.tokenizer],["model",_.model],["processor",_.processor]]),h=await async function(e,t,s){const r=Object.create(null),n=[];for(const[o,i]of e.entries()){if(!i)continue;let e;e=Array.isArray(i)?new Promise((async(e,r)=>{let n;for(const o of i){if(null===o)return void e(null);try{return void e(await o.from_pretrained(t,s))}catch(e){if(e.message?.includes("Unsupported model type"))n=e;else{if(!e.message?.includes("Could not locate file"))return void r(e);n=e}}}r(n)})):i.from_pretrained(t,s),r[o]=e,n.push(e)}await Promise.all(n);for(const[e,t]of Object.entries(r))r[e]=await t;return r}(m,t,p);h.task=e,(0,a.dispatchCallback)(s,{status:"ready",task:e,model:t});return new(0,_.pipeline)(h)}},"./src/tokenizers.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{AlbertTokenizer:()=>ye,AutoTokenizer:()=>mt,BartTokenizer:()=>Ne,BertTokenizer:()=>xe,BlenderbotSmallTokenizer:()=>lt,BlenderbotTokenizer:()=>at,BloomTokenizer:()=>De,CLIPTokenizer:()=>rt,CamembertTokenizer:()=>Ee,CodeGenTokenizer:()=>st,CodeLlamaTokenizer:()=>$e,CohereTokenizer:()=>_t,ConvBertTokenizer:()=>Ce,DebertaTokenizer:()=>ve,DebertaV2Tokenizer:()=>Pe,DistilBertTokenizer:()=>Se,ElectraTokenizer:()=>Le,EsmTokenizer:()=>Qe,FalconTokenizer:()=>We,GPT2Tokenizer:()=>je,GPTNeoXTokenizer:()=>Xe,GemmaTokenizer:()=>Je,Grok1Tokenizer:()=>Ye,HerbertTokenizer:()=>Fe,LlamaTokenizer:()=>Ge,M2M100Tokenizer:()=>et,MBart50Tokenizer:()=>Be,MBartTokenizer:()=>Oe,MPNetTokenizer:()=>Ue,MarianTokenizer:()=>ot,MgpstrTokenizer:()=>pt,MobileBertTokenizer:()=>ke,NllbTokenizer:()=>Ze,NougatTokenizer:()=>dt,PreTrainedTokenizer:()=>be,Qwen2Tokenizer:()=>He,RoFormerTokenizer:()=>Ae,RobertaTokenizer:()=>Ve,SiglipTokenizer:()=>nt,SpeechT5Tokenizer:()=>ct,SqueezeBertTokenizer:()=>Te,T5Tokenizer:()=>ze,TokenizerModel:()=>k,VitsTokenizer:()=>ut,Wav2Vec2CTCTokenizer:()=>it,WhisperTokenizer:()=>tt,XLMRobertaTokenizer:()=>qe,XLMTokenizer:()=>Ie,is_chinese_char:()=>g});var r=s("./src/utils/generic.js"),n=s("./src/utils/core.js"),o=s("./src/utils/hub.js"),i=s("./src/utils/maths.js"),a=s("./src/utils/tensor.js"),l=s("./src/utils/data-structures.js"),c=s("./node_modules/@huggingface/jinja/dist/index.js"),d=s("./src/models/whisper/common_whisper.js");async function u(e,t){const s=await Promise.all([(0,o.getModelJSON)(e,"tokenizer.json",!0,t),(0,o.getModelJSON)(e,"tokenizer_config.json",!0,t)]);return null!==t.legacy&&(s[1].legacy=t.legacy),s}function _(e,t=!0){if(void 0!==e.Regex){let t=e.Regex.replace(/\\([#&~])/g,"$1");for(const[e,s]of x)t=t.replaceAll(e,s);return new RegExp(t,"gu")}if(void 0!==e.String){const s=(0,n.escapeRegExp)(e.String);return new RegExp(t?s:`(${s})`,"gu")}return console.warn("Unknown pattern type:",e),null}function p(e){return new Map(Object.entries(e))}function m(e){const t=e.dims;switch(t.length){case 1:return e.tolist();case 2:if(1!==t[0])throw new Error("Unable to decode tensor with `batch size !== 1`. Use `tokenizer.batch_decode(...)` for batched inputs.");return e.tolist()[0];default:throw new Error(`Expected tensor to have 1-2 dimensions, got ${t.length}.`)}}function h(e){return e.replace(/ \./g,".").replace(/ \?/g,"?").replace(/ \!/g,"!").replace(/ ,/g,",").replace(/ \' /g,"'").replace(/ n\'t/g,"n't").replace(/ \'m/g,"'m").replace(/ \'s/g,"'s").replace(/ \'ve/g,"'ve").replace(/ \'re/g,"'re")}function f(e){return e.replace(/\p{M}/gu,"")}function g(e){return e>=19968&&e<=40959||e>=13312&&e<=19903||e>=131072&&e<=173791||e>=173824&&e<=177983||e>=177984&&e<=178207||e>=178208&&e<=183983||e>=63744&&e<=64255||e>=194560&&e<=195103}const w="\\p{P}\\u0021-\\u002F\\u003A-\\u0040\\u005B-\\u0060\\u007B-\\u007E",M=new RegExp(`^[${w}]+$`,"gu"),b=".,!?…。，、।۔،",x=new Map([["(?i:'s|'t|'re|'ve|'m|'ll|'d)","(?:'([sS]|[tT]|[rR][eE]|[vV][eE]|[mM]|[lL][lL]|[dD]))"],[` ?[^(\\s|[${b}])]+`,` ?[^\\s${b}]+`]]);class y{constructor(e){this.content=e.content,this.id=e.id,this.single_word=e.single_word??!1,this.lstrip=e.lstrip??!1,this.rstrip=e.rstrip??!1,this.special=e.special??!1,this.normalized=e.normalized??null}}class k extends r.Callable{constructor(e){super(),this.config=e,this.vocab=[],this.tokens_to_ids=new Map,this.unk_token_id=void 0,this.unk_token=void 0,this.end_of_word_suffix=void 0,this.fuse_unk=this.config.fuse_unk??!1}static fromConfig(e,...t){switch(e.type){case"WordPiece":return new T(e);case"Unigram":return new v(e,...t);case"BPE":return new C(e);default:if(e.vocab)return Array.isArray(e.vocab)?new v(e,...t):"object"==typeof e.vocab&&e.continuing_subword_prefix&&e.unk_token?new T(e):new A(e,...t);throw new Error(`Unknown TokenizerModel type: ${e.type}`)}}_call(e){return e=this.encode(e),this.fuse_unk&&(e=function(e,t,s){const r=[];let n=0;for(;n<e.length;)if(r.push(e[n]),(t.get(e[n])??s)===s)for(;++n<e.length&&(t.get(e[n])??s)===s;)t.get(r.at(-1))!==s&&(r[r.length-1]+=e[n]);else++n;return r}(e,this.tokens_to_ids,this.unk_token_id)),e}encode(e){throw Error("encode should be implemented in subclass.")}convert_tokens_to_ids(e){return e.map((e=>this.tokens_to_ids.get(e)??this.unk_token_id))}convert_ids_to_tokens(e){return e.map((e=>this.vocab[e]??this.unk_token))}}class T extends k{constructor(e){super(e),this.tokens_to_ids=p(e.vocab),this.unk_token_id=this.tokens_to_ids.get(e.unk_token),this.unk_token=e.unk_token,this.max_input_chars_per_word=e.max_input_chars_per_word??100,this.vocab=new Array(this.tokens_to_ids.size);for(const[e,t]of this.tokens_to_ids)this.vocab[t]=e}encode(e){const t=[];for(const s of e){const e=[...s];if(e.length>this.max_input_chars_per_word){t.push(this.unk_token);continue}let r=!1,n=0;const o=[];for(;n<e.length;){let t=e.length,s=null;for(;n<t;){let r=e.slice(n,t).join("");if(n>0&&(r=this.config.continuing_subword_prefix+r),this.tokens_to_ids.has(r)){s=r;break}--t}if(null===s){r=!0;break}o.push(s),n=t}r?t.push(this.unk_token):t.push(...o)}return t}}class v extends k{constructor(e,t){super(e);const s=e.vocab.length;this.vocab=new Array(s),this.scores=new Array(s);for(let t=0;t<s;++t)[this.vocab[t],this.scores[t]]=e.vocab[t];this.unk_token_id=e.unk_id,this.unk_token=this.vocab[e.unk_id],this.tokens_to_ids=new Map(this.vocab.map(((e,t)=>[e,t]))),this.bos_token=" ",this.bos_token_id=this.tokens_to_ids.get(this.bos_token),this.eos_token=t.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,i.min)(this.scores)[0],this.unk_score=this.minScore-10,this.scores[this.unk_token_id]=this.unk_score,this.trie=new l.CharTrie,this.trie.extend(this.vocab),this.fuse_unk=!0}populateNodes(e){const t=e.chars;let s=0;for(;s<t.length;){let r=!1;const o=[],i=t.slice(s).join(""),a=this.trie.commonPrefixSearch(i);for(const t of a){o.push(t);const i=this.tokens_to_ids.get(t),a=this.scores[i],l=(0,n.len)(t);e.insert(s,l,a,i),r||1!==l||(r=!0)}r||e.insert(s,1,this.unk_score,this.unk_token_id),s+=1}}tokenize(e){const t=new l.TokenLattice(e,this.bos_token_id,this.eos_token_id);return this.populateNodes(t),t.tokens()}encode(e){const t=[];for(const s of e){const e=this.tokenize(s);t.push(...e)}return t}}const P=(()=>{const e=[...Array.from({length:"~".charCodeAt(0)-"!".charCodeAt(0)+1},((e,t)=>t+"!".charCodeAt(0))),...Array.from({length:"¬".charCodeAt(0)-"¡".charCodeAt(0)+1},((e,t)=>t+"¡".charCodeAt(0))),...Array.from({length:"ÿ".charCodeAt(0)-"®".charCodeAt(0)+1},((e,t)=>t+"®".charCodeAt(0)))],t=e.slice();let s=0;for(let r=0;r<256;++r)e.includes(r)||(e.push(r),t.push(256+s),s+=1);const r=t.map((e=>String.fromCharCode(e)));return Object.fromEntries(e.map(((e,t)=>[e,r[t]])))})(),F=(0,n.reverseDictionary)(P);class C extends k{constructor(e){super(e),this.tokens_to_ids=p(e.vocab),this.unk_token_id=this.tokens_to_ids.get(e.unk_token),this.unk_token=e.unk_token,this.vocab=new Array(this.tokens_to_ids.size);for(const[e,t]of this.tokens_to_ids)this.vocab[t]=e;const t=Array.isArray(e.merges[0]);this.merges=t?e.merges:e.merges.map((e=>e.split(" ",2))),this.bpe_ranks=new Map(this.merges.map(((e,t)=>[JSON.stringify(e),t]))),this.end_of_word_suffix=e.end_of_word_suffix,this.continuing_subword_suffix=e.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(e){if(0===e.length)return[];const t=this.cache.get(e);if(void 0!==t)return t;const s=Array.from(e);this.end_of_word_suffix&&(s[s.length-1]+=this.end_of_word_suffix);let r=[];if(s.length>1){const e=new l.PriorityQueue(((e,t)=>e.score<t.score));let t={token:s[0],bias:0,prev:null,next:null},n=t;for(let t=1;t<s.length;++t){const r={bias:t/s.length,token:s[t],prev:n,next:null};n.next=r,this._add_node(e,n),n=r}for(;!e.isEmpty();){const s=e.pop();if(s.deleted||!s.next||s.next.deleted)continue;if(s.deleted=!0,s.next.deleted=!0,s.prev){const e={...s.prev};s.prev.deleted=!0,s.prev=e,e.prev?e.prev.next=e:t=e}const r={token:s.token+s.next.token,bias:s.bias,prev:s.prev,next:s.next.next};r.prev?(r.prev.next=r,this._add_node(e,r.prev)):t=r,r.next&&(r.next.prev=r,this._add_node(e,r))}for(let e=t;null!==e;e=e.next)r.push(e.token)}else r=s;if(this.continuing_subword_suffix)for(let e=0;e<r.length-1;++e)r[e]+=this.continuing_subword_suffix;return this.cache.set(e,r),r}_add_node(e,t){const s=this.bpe_ranks.get(JSON.stringify([t.token,t.next.token]));void 0!==s&&(t.score=s+t.bias,e.push(t))}encode(e){const t=[];for(const s of e){if(this.ignore_merges&&this.tokens_to_ids.has(s)){t.push(s);continue}const e=this.bpe(s);for(const s of e)if(this.tokens_to_ids.has(s))t.push(s);else if(this.byte_fallback){const e=Array.from(this.text_encoder.encode(s)).map((e=>`<0x${e.toString(16).toUpperCase().padStart(2,"0")}>`));e.every((e=>this.tokens_to_ids.has(e)))?t.push(...e):t.push(this.unk_token)}else t.push(this.unk_token)}return t}}class A extends k{constructor(e,t){super(e),this.tokens_to_ids=p(t.target_lang?e.vocab[t.target_lang]:e.vocab),this.bos_token=t.bos_token,this.bos_token_id=this.tokens_to_ids.get(this.bos_token),this.eos_token=t.eos_token,this.eos_token_id=this.tokens_to_ids.get(this.eos_token),this.pad_token=t.pad_token,this.pad_token_id=this.tokens_to_ids.get(this.pad_token),this.unk_token=t.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[e,t]of this.tokens_to_ids)this.vocab[t]=e}encode(e){return e}}class S extends r.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"BertNormalizer":return new D(e);case"Precompiled":return new _e(e);case"Sequence":return new V(e);case"Replace":return new E(e);case"NFC":return new I(e);case"NFKC":return new L(e);case"NFKD":return new z(e);case"Strip":return new j(e);case"StripAccents":return new N(e);case"Lowercase":return new O(e);case"Prepend":return new B(e);default:throw new Error(`Unknown Normalizer type: ${e.type}`)}}normalize(e){throw Error("normalize should be implemented in subclass.")}_call(e){return this.normalize(e)}}class E extends S{normalize(e){const t=_(this.config.pattern);return null===t?e:e.replaceAll(t,this.config.content)}}class I extends S{normalize(e){return e=e.normalize("NFC")}}class L extends S{normalize(e){return e=e.normalize("NFKC")}}class z extends S{normalize(e){return e=e.normalize("NFKD")}}class j extends S{normalize(e){return this.config.strip_left&&this.config.strip_right?e=e.trim():(this.config.strip_left&&(e=e.trimStart()),this.config.strip_right&&(e=e.trimEnd())),e}}class N extends S{normalize(e){return e=f(e)}}class O extends S{normalize(e){return e=e.toLowerCase()}}class B extends S{normalize(e){return e=this.config.prepend+e}}class V extends S{constructor(e){super(e),this.normalizers=e.normalizers.map((e=>S.fromConfig(e)))}normalize(e){return this.normalizers.reduce(((e,t)=>t.normalize(e)),e)}}class D extends S{_tokenize_chinese_chars(e){const t=[];for(let s=0;s<e.length;++s){const r=e[s];g(r.charCodeAt(0))?(t.push(" "),t.push(r),t.push(" ")):t.push(r)}return t.join("")}stripAccents(e){return e.normalize("NFD").replace(/\p{Mn}/gu,"")}_is_control(e){switch(e){case"\t":case"\n":case"\r":return!1;default:return/^\p{Cc}|\p{Cf}|\p{Co}|\p{Cs}$/u.test(e)}}_clean_text(e){const t=[];for(const s of e){const e=s.charCodeAt(0);0===e||65533===e||this._is_control(s)||(/^\s$/.test(s)?t.push(" "):t.push(s))}return t.join("")}normalize(e){return this.config.clean_text&&(e=this._clean_text(e)),this.config.handle_chinese_chars&&(e=this._tokenize_chinese_chars(e)),this.config.lowercase?(e=e.toLowerCase(),!1!==this.config.strip_accents&&(e=this.stripAccents(e))):this.config.strip_accents&&(e=this.stripAccents(e)),e}}class R extends r.Callable{static fromConfig(e){if(null===e)return null;switch(e.type){case"BertPreTokenizer":return new G(e);case"Sequence":return new pe(e);case"Whitespace":return new me(e);case"WhitespaceSplit":return new he(e);case"Metaspace":return new de(e);case"ByteLevel":return new $(e);case"Split":return new q(e);case"Punctuation":return new U(e);case"Digits":return new W(e);case"Replace":return new fe(e);default:throw new Error(`Unknown PreTokenizer type: ${e.type}`)}}pre_tokenize_text(e,t){throw Error("pre_tokenize_text should be implemented in subclass.")}pre_tokenize(e,t){return(Array.isArray(e)?e.map((e=>this.pre_tokenize_text(e,t))):this.pre_tokenize_text(e,t)).flat()}_call(e,t){return this.pre_tokenize(e,t)}}class G extends R{constructor(e){super(),this.pattern=new RegExp(`[^\\s${w}]+|[${w}]`,"gu")}pre_tokenize_text(e,t){return e.trim().match(this.pattern)||[]}}class $ extends R{constructor(e){super(),this.config=e,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=/'s|'t|'re|'ve|'m|'ll|'d| ?\p{L}+| ?\p{N}+| ?[^\s\p{L}\p{N}]+|\s+(?!\S)|\s+/gu,this.byte_encoder=P,this.text_encoder=new TextEncoder}pre_tokenize_text(e,t){this.add_prefix_space&&!e.startsWith(" ")&&(e=" "+e);return(this.use_regex?e.match(this.pattern)||[]:[e]).map((e=>Array.from(this.text_encoder.encode(e),(e=>this.byte_encoder[e])).join("")))}}class q extends R{constructor(e){super(),this.config=e,this.pattern=_(this.config.pattern,this.config.invert)}pre_tokenize_text(e,t){return null===this.pattern?[]:this.config.invert?e.match(this.pattern)||[]:"removed"===this.config.behavior?.toLowerCase()?e.split(this.pattern).filter((e=>e)):function(e,t){const s=[];let r=0;for(const n of e.matchAll(t)){const t=n[0];r<n.index&&s.push(e.slice(r,n.index)),t.length>0&&s.push(t),r=n.index+t.length}return r<e.length&&s.push(e.slice(r)),s}(e,this.pattern)}}class U extends R{constructor(e){super(),this.config=e,this.pattern=new RegExp(`[^${w}]+|[${w}]+`,"gu")}pre_tokenize_text(e,t){return e.match(this.pattern)||[]}}class W extends R{constructor(e){super(),this.config=e;const t="[^\\d]+|\\d"+(this.config.individual_digits?"":"+");this.pattern=new RegExp(t,"gu")}pre_tokenize_text(e,t){return e.match(this.pattern)||[]}}class X extends r.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"TemplateProcessing":return new J(e);case"ByteLevel":return new Y(e);case"RobertaProcessing":return new H(e);case"BertProcessing":return new Q(e);case"Sequence":return new K(e);default:throw new Error(`Unknown PostProcessor type: ${e.type}`)}}post_process(e,...t){throw Error("post_process should be implemented in subclass.")}_call(e,...t){return this.post_process(e,...t)}}class Q extends X{constructor(e){super(e),this.cls=e.cls[0],this.sep=e.sep[0]}post_process(e,t=null,{add_special_tokens:s=!0}={}){s&&(e=(0,n.mergeArrays)([this.cls],e,[this.sep]));let r=new Array(e.length).fill(0);if(null!==t){const o=s&&this instanceof H?[this.sep]:[],i=s?[this.sep]:[];e=(0,n.mergeArrays)(e,o,t,i),r=(0,n.mergeArrays)(r,new Array(t.length+o.length+i.length).fill(1))}return{tokens:e,token_type_ids:r}}}class H extends Q{}class J extends X{constructor(e){super(e),this.single=e.single,this.pair=e.pair}post_process(e,t=null,{add_special_tokens:s=!0}={}){const r=null===t?this.single:this.pair;let o=[],i=[];for(const a of r)"SpecialToken"in a?s&&(o.push(a.SpecialToken.id),i.push(a.SpecialToken.type_id)):"Sequence"in a&&("A"===a.Sequence.id?(o=(0,n.mergeArrays)(o,e),i=(0,n.mergeArrays)(i,new Array(e.length).fill(a.Sequence.type_id))):"B"===a.Sequence.id&&(o=(0,n.mergeArrays)(o,t),i=(0,n.mergeArrays)(i,new Array(t.length).fill(a.Sequence.type_id))));return{tokens:o,token_type_ids:i}}}class Y extends X{post_process(e,t=null){return t&&(e=(0,n.mergeArrays)(e,t)),{tokens:e}}}class K extends X{constructor(e){super(e),this.processors=e.processors.map((e=>X.fromConfig(e)))}post_process(e,t=null,s={}){let r;for(const n of this.processors)if(n instanceof Y){if(e=n.post_process(e).tokens,t){t=n.post_process(t).tokens}}else{const o=n.post_process(e,t,s);e=o.tokens,r=o.token_type_ids}return{tokens:e,token_type_ids:r}}}class Z extends r.Callable{constructor(e){super(),this.config=e,this.added_tokens=[],this.end_of_word_suffix=null,this.trim_offsets=e.trim_offsets}static fromConfig(e){if(null===e)return null;switch(e.type){case"WordPiece":return new ne(e);case"Metaspace":return new ue(e);case"ByteLevel":return new oe(e);case"Replace":return new ee(e);case"ByteFallback":return new te(e);case"Fuse":return new se(e);case"Strip":return new re(e);case"Sequence":return new ae(e);case"CTC":return new ie(e);case"BPEDecoder":return new le(e);default:throw new Error(`Unknown Decoder type: ${e.type}`)}}_call(e){return this.decode(e)}decode(e){return this.decode_chain(e).join("")}decode_chain(e){throw Error("`decode_chain` should be implemented in subclass.")}}class ee extends Z{decode_chain(e){const t=_(this.config.pattern);return null===t?e:e.map((e=>e.replaceAll(t,this.config.content)))}}class te extends Z{constructor(e){super(e),this.text_decoder=new TextDecoder}decode_chain(e){const t=[];let s=[];for(const r of e){let e=null;if(6===r.length&&r.startsWith("<0x")&&r.endsWith(">")){const t=parseInt(r.slice(3,5),16);isNaN(t)||(e=t)}if(null!==e)s.push(e);else{if(s.length>0){const e=this.text_decoder.decode(Uint8Array.from(s));t.push(e),s=[]}t.push(r)}}if(s.length>0){const e=this.text_decoder.decode(Uint8Array.from(s));t.push(e),s=[]}return t}}class se extends Z{decode_chain(e){return[e.join("")]}}class re extends Z{constructor(e){super(e),this.content=this.config.content,this.start=this.config.start,this.stop=this.config.stop}decode_chain(e){return e.map((e=>{let t=0;for(let s=0;s<this.start&&e[s]===this.content;++s)t=s+1;let s=e.length;for(let t=0;t<this.stop;++t){const r=e.length-t-1;if(e[r]!==this.content)break;s=r}return e.slice(t,s)}))}}class ne extends Z{constructor(e){super(e),this.cleanup=e.cleanup}decode_chain(e){return e.map(((e,t)=>(0!==t&&(e=e.startsWith(this.config.prefix)?e.replace(this.config.prefix,""):" "+e),this.cleanup&&(e=h(e)),e)))}}class oe extends Z{constructor(e){super(e),this.byte_decoder=F,this.text_decoder=new TextDecoder("utf-8",{fatal:!1,ignoreBOM:!0}),this.end_of_word_suffix=null}convert_tokens_to_string(e){const t=e.join(""),s=new Uint8Array([...t].map((e=>this.byte_decoder[e])));return this.text_decoder.decode(s)}decode_chain(e){const t=[];let s=[];for(const r of e)void 0!==this.added_tokens.find((e=>e.content===r))?(s.length>0&&(t.push(this.convert_tokens_to_string(s)),s=[]),t.push(r)):s.push(r);return s.length>0&&t.push(this.convert_tokens_to_string(s)),t}}class ie extends Z{constructor(e){super(e),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(e){if(0===e.length)return"";const t=[e[0]];for(let s=1;s<e.length;++s)e[s]!==t.at(-1)&&t.push(e[s]);let s=t.filter((e=>e!==this.pad_token)).join("");return this.cleanup&&(s=h(s).replaceAll(this.word_delimiter_token," ").trim()),s}decode_chain(e){return[this.convert_tokens_to_string(e)]}}class ae extends Z{constructor(e){super(e),this.decoders=e.decoders.map((e=>Z.fromConfig(e)))}decode_chain(e){return this.decoders.reduce(((e,t)=>t.decode_chain(e)),e)}}class le extends Z{constructor(e){super(e),this.suffix=this.config.suffix}decode_chain(e){return e.map(((t,s)=>t.replaceAll(this.suffix,s===e.length-1?"":" ")))}}class ce extends Z{decode_chain(e){let t="";for(let s=1;s<e.length;s+=2)t+=e[s];return[t]}}class de extends R{constructor(e){super(),this.addPrefixSpace=e.add_prefix_space,this.replacement=e.replacement,this.strRep=e.str_rep||this.replacement,this.prepend_scheme=e.prepend_scheme??"always"}pre_tokenize_text(e,{section_index:t}={}){let s=e.replaceAll(" ",this.strRep);return this.addPrefixSpace&&!s.startsWith(this.replacement)&&("always"===this.prepend_scheme||"first"===this.prepend_scheme&&0===t)&&(s=this.strRep+s),[s]}}class ue extends Z{constructor(e){super(e),this.addPrefixSpace=e.add_prefix_space,this.replacement=e.replacement}decode_chain(e){const t=[];for(let s=0;s<e.length;++s){let r=e[s].replaceAll(this.replacement," ");this.addPrefixSpace&&0==s&&r.startsWith(" ")&&(r=r.substring(1)),t.push(r)}return t}}class _e extends S{constructor(e){super(e),this.charsmap=e.precompiled_charsmap}normalize(e){if((e=(e=e.replace(/[\u0001-\u0008\u000B\u000E-\u001F\u007F\u008F\u009F]/gm,"")).replace(/[\u0009\u000A\u000C\u000D\u00A0\u1680\u2000-\u200F\u2028\u2029\u202F\u205F\u2581\u3000\uFEFF\uFFFD]/gm," ")).includes("～")){const t=e.split("～");e=t.map((e=>e.normalize("NFKC"))).join("～")}else e=e.normalize("NFKC");return e}}class pe extends R{constructor(e){super(),this.tokenizers=e.pretokenizers.map((e=>R.fromConfig(e)))}pre_tokenize_text(e,t){return this.tokenizers.reduce(((e,s)=>s.pre_tokenize(e,t)),[e])}}class me extends R{constructor(e){super()}pre_tokenize_text(e,t){return e.match(/\w+|[^\w\s]+/g)||[]}}class he extends R{constructor(e){super()}pre_tokenize_text(e,t){return function(e){return e.match(/\S+/g)||[]}(e)}}class fe extends R{constructor(e){super(),this.config=e,this.pattern=_(this.config.pattern),this.content=this.config.content}pre_tokenize_text(e,t){return null===this.pattern?[e]:[e.replaceAll(this.pattern,this.config.content)]}}const ge=["bos_token","eos_token","unk_token","sep_token","pad_token","cls_token","mask_token"];function we(e,t,s,r){for(const o of Object.keys(e)){const i=t-e[o].length,a=s(o),l=new Array(i).fill(a);e[o]="right"===r?(0,n.mergeArrays)(e[o],l):(0,n.mergeArrays)(l,e[o])}}function Me(e,t){for(const s of Object.keys(e))e[s].length=t}class be extends r.Callable{return_token_type_ids=!1;padding_side="right";constructor(e,t){super(),this._tokenizer_config=t,this.normalizer=S.fromConfig(e.normalizer),this.pre_tokenizer=R.fromConfig(e.pre_tokenizer),this.model=k.fromConfig(e.model,t),this.post_processor=X.fromConfig(e.post_processor),this.decoder=Z.fromConfig(e.decoder),this.special_tokens=[],this.all_special_ids=[],this.added_tokens=[];for(const t of e.added_tokens){const e=new y(t);this.added_tokens.push(e),this.model.tokens_to_ids.set(e.content,e.id),this.model.vocab[e.id]=e.content,e.special&&(this.special_tokens.push(e.content),this.all_special_ids.push(e.id))}if(this.additional_special_tokens=t.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(((e,t)=>t.content.length-e.content.length)).map((e=>`${e.lstrip?"\\s*":""}(${(0,n.escapeRegExp)(e.content)})${e.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=t.model_max_length,this.remove_space=t.remove_space,this.clean_up_tokenization_spaces=t.clean_up_tokenization_spaces??!0,this.do_lowercase_and_remove_accent=t.do_lowercase_and_remove_accent??!1,t.padding_side&&(this.padding_side=t.padding_side),this.legacy=!1,this.chat_template=t.chat_template??null,Array.isArray(this.chat_template)){const e=Object.create(null);for(const{name:t,template:s}of this.chat_template){if("string"!=typeof t||"string"!=typeof s)throw new Error('Chat template must be a list of objects with "name" and "template" properties');e[t]=s}this.chat_template=e}this._compiled_template_cache=new Map}getToken(...e){for(const t of e){const e=this._tokenizer_config[t];if(e){if("object"==typeof e){if("AddedToken"===e.__type)return e.content;throw Error(`Unknown token: ${e}`)}return e}}return null}static async from_pretrained(e,{progress_callback:t=null,config:s=null,cache_dir:r=null,local_files_only:n=!1,revision:o="main",legacy:i=null}={}){return new this(...await u(e,{progress_callback:t,config:s,cache_dir:r,local_files_only:n,revision:o,legacy:i}))}_call(e,{text_pair:t=null,add_special_tokens:s=!0,padding:r=!1,truncation:n=null,max_length:o=null,return_tensor:l=!0,return_token_type_ids:c=null}={}){const d=Array.isArray(e);let u;if(d){if(0===e.length)throw Error("text array must be non-empty");if(null!==t){if(!Array.isArray(t))throw Error("text_pair must also be an array");if(e.length!==t.length)throw Error("text and text_pair must have the same length");u=e.map(((e,r)=>this._encode_plus(e,{text_pair:t[r],add_special_tokens:s,return_token_type_ids:c})))}else u=e.map((e=>this._encode_plus(e,{add_special_tokens:s,return_token_type_ids:c})))}else{if(null==e)throw Error("text may not be null or undefined");if(Array.isArray(t))throw Error("When specifying `text_pair`, since `text` is a string, `text_pair` must also be a string (i.e., not an array).");u=[this._encode_plus(e,{text_pair:t,add_special_tokens:s,return_token_type_ids:c})]}if(null===o?o="max_length"===r?this.model_max_length:(0,i.max)(u.map((e=>e.input_ids.length)))[0]:n||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."),o=Math.min(o,this.model_max_length??1/0),r||n)for(let e=0;e<u.length;++e)u[e].input_ids.length!==o&&(u[e].input_ids.length>o?n&&Me(u[e],o):r&&we(u[e],o,(e=>"input_ids"===e?this.pad_token_id:0),this.padding_side));const _={};if(l){if((!r||!n)&&u.some((e=>{for(const t of Object.keys(e))if(e[t].length!==u[0][t]?.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 e=[u.length,u[0].input_ids.length];for(const t of Object.keys(u[0]))_[t]=new a.Tensor("int64",BigInt64Array.from(u.flatMap((e=>e[t])).map(BigInt)),e)}else{for(const e of Object.keys(u[0]))_[e]=u.map((t=>t[e]));if(!d)for(const e of Object.keys(_))_[e]=_[e][0]}return _}_encode_text(e){if(null===e)return null;const t=(this.added_tokens_regex?e.split(this.added_tokens_regex).filter((e=>e)):[e]).map(((e,t)=>{if(void 0!==this.added_tokens.find((t=>t.content===e)))return e;{if(!0===this.remove_space&&(e=e.trim().split(/\s+/).join(" ")),this.do_lowercase_and_remove_accent&&(e=function(e){return f(e.toLowerCase())}(e)),null!==this.normalizer&&(e=this.normalizer(e)),0===e.length)return[];const s=null!==this.pre_tokenizer?this.pre_tokenizer(e,{section_index:t}):[e];return this.model(s)}})).flat();return t}_encode_plus(e,{text_pair:t=null,add_special_tokens:s=!0,return_token_type_ids:r=null}={}){const{tokens:n,token_type_ids:o}=this._tokenize_helper(e,{pair:t,add_special_tokens:s}),i=this.model.convert_tokens_to_ids(n),a={input_ids:i,attention_mask:new Array(i.length).fill(1)};return(r??this.return_token_type_ids)&&o&&(a.token_type_ids=o),a}_tokenize_helper(e,{pair:t=null,add_special_tokens:s=!1}={}){const r=this._encode_text(e),o=this._encode_text(t);return this.post_processor?this.post_processor(r,o,{add_special_tokens:s}):{tokens:(0,n.mergeArrays)(r??[],o??[])}}tokenize(e,{pair:t=null,add_special_tokens:s=!1}={}){return this._tokenize_helper(e,{pair:t,add_special_tokens:s}).tokens}encode(e,{text_pair:t=null,add_special_tokens:s=!0,return_token_type_ids:r=null}={}){return this._encode_plus(e,{text_pair:t,add_special_tokens:s,return_token_type_ids:r}).input_ids}batch_decode(e,t={}){return e instanceof a.Tensor&&(e=e.tolist()),e.map((e=>this.decode(e,t)))}decode(e,t={}){if(e instanceof a.Tensor&&(e=m(e)),!Array.isArray(e)||0===e.length||!(0,n.isIntegralNumber)(e[0]))throw Error("token_ids must be a non-empty array of integers.");return this.decode_single(e,t)}decode_single(e,{skip_special_tokens:t=!1,clean_up_tokenization_spaces:s=null}){let r=this.model.convert_ids_to_tokens(e);t&&(r=r.filter((e=>!this.special_tokens.includes(e))));let n=this.decoder?this.decoder(r):r.join(" ");return this.decoder&&this.decoder.end_of_word_suffix&&(n=n.replaceAll(this.decoder.end_of_word_suffix," "),t&&(n=n.trim())),(s??this.clean_up_tokenization_spaces)&&(n=h(n)),n}get_chat_template({chat_template:e=null,tools:t=null}={}){if(this.chat_template&&"object"==typeof this.chat_template){const s=this.chat_template;if(null!==e&&Object.hasOwn(s,e))e=s[e];else if(null===e)if(null!==t&&"tool_use"in s)e=s.tool_use;else{if(!("default"in s))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(s).sort()}.`);e=s.default}}else if(null===e){if(!this.chat_template)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");e=this.chat_template}return e}apply_chat_template(e,{tools:t=null,documents:s=null,chat_template:r=null,add_generation_prompt:n=!1,tokenize:o=!0,padding:i=!1,truncation:a=!1,max_length:l=null,return_tensor:d=!0,return_dict:u=!1,tokenizer_kwargs:_={},...p}={}){if("string"!=typeof(r=this.get_chat_template({chat_template:r,tools:t})))throw Error("chat_template must be a string, but got "+typeof r);let m=this._compiled_template_cache.get(r);void 0===m&&(m=new c.Template(r),this._compiled_template_cache.set(r,m));const h=Object.create(null);for(const e of ge){const t=this.getToken(e);t&&(h[e]=t)}const f=m.render({messages:e,add_generation_prompt:n,tools:t,documents:s,...h,...p});if(o){const e=this._call(f,{add_special_tokens:!1,padding:i,truncation:a,max_length:l,return_tensor:d,..._});return u?e:e.input_ids}return f}}class xe extends be{return_token_type_ids=!0}class ye extends be{return_token_type_ids=!0}class ke extends be{return_token_type_ids=!0}class Te extends be{return_token_type_ids=!0}class ve extends be{return_token_type_ids=!0}class Pe extends be{return_token_type_ids=!0}class Fe extends be{return_token_type_ids=!0}class Ce extends be{return_token_type_ids=!0}class Ae extends be{return_token_type_ids=!0}class Se extends be{}class Ee extends be{}class Ie extends be{return_token_type_ids=!0;constructor(e,t){super(e,t),console.warn('WARNING: `XLMTokenizer` is not yet supported by Hugging Face\'s "fast" tokenizers library. Therefore, you may experience slightly inaccurate results.')}}class Le extends be{return_token_type_ids=!0}class ze extends be{}class je extends be{}class Ne extends be{}class Oe extends be{constructor(e,t){super(e,t),this.languageRegex=/^[a-z]{2}_[A-Z]{2}$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))),this.lang_to_token=e=>e}_build_translation_inputs(e,t,s){return Ke(this,e,t,s)}}class Be extends Oe{}class Ve extends be{}class De extends be{}const Re="▁";class Ge extends be{padding_side="left";constructor(e,t){super(e,t),this.legacy=t.legacy??!0,this.legacy||(this.normalizer=null,this.pre_tokenizer=new de({replacement:Re,add_prefix_space:!0,prepend_scheme:"first"}))}_encode_text(e){if(null===e)return null;if(this.legacy||0===e.length)return super._encode_text(e);let t=super._encode_text(Re+e.replaceAll(Re," "));return t.length>1&&t[0]===Re&&this.special_tokens.includes(t[1])&&(t=t.slice(1)),t}}class $e extends be{}class qe extends be{}class Ue extends be{}class We extends be{}class Xe extends be{}class Qe extends be{}class He extends be{}class Je extends be{}class Ye extends be{}function Ke(e,t,s,r){if(!("language_codes"in e)||!Array.isArray(e.language_codes))throw new Error("Tokenizer must have `language_codes` attribute set and it should be an array of language ids.");if(!("languageRegex"in e&&e.languageRegex instanceof RegExp))throw new Error("Tokenizer must have `languageRegex` attribute set and it should be a regular expression.");if(!("lang_to_token"in e)||"function"!=typeof e.lang_to_token)throw new Error("Tokenizer must have `lang_to_token` attribute set and it should be a function.");const n=r.src_lang,o=r.tgt_lang;if(!e.language_codes.includes(o))throw new Error(`Target language code "${o}" is not valid. Must be one of: {${e.language_codes.join(", ")}}`);if(void 0!==n){if(!e.language_codes.includes(n))throw new Error(`Source language code "${n}" is not valid. Must be one of: {${e.language_codes.join(", ")}}`);for(const t of e.post_processor.config.single)if("SpecialToken"in t&&e.languageRegex.test(t.SpecialToken.id)){t.SpecialToken.id=e.lang_to_token(n);break}}return r.forced_bos_token_id=e.model.convert_tokens_to_ids([e.lang_to_token(o)])[0],e._call(t,s)}class Ze extends be{constructor(e,t){super(e,t),this.languageRegex=/^[a-z]{3}_[A-Z][a-z]{3}$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))),this.lang_to_token=e=>e}_build_translation_inputs(e,t,s){return Ke(this,e,t,s)}}class et extends be{constructor(e,t){super(e,t),this.languageRegex=/^__[a-z]{2,3}__$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))).map((e=>e.slice(2,-2))),this.lang_to_token=e=>`__${e}__`}_build_translation_inputs(e,t,s){return Ke(this,e,t,s)}}class tt extends be{get timestamp_begin(){return this.model.convert_tokens_to_ids(["<|notimestamps|>"])[0]+1}_decode_asr(e,{return_timestamps:t=!1,return_language:s=!1,time_precision:r=null,force_full_sequences:n=!0}={}){if(null===r)throw Error("Must specify time_precision");let o=null;const a="word"===t;function l(){return{language:o,timestamp:[null,null],text:""}}const c=[];let u=l(),_=0;const p=this.timestamp_begin,m=p+1500;let h=[],f=[],g=!1,w=null;const b=new Set(this.all_special_ids);for(const s of e){const e=s.tokens,n=a?s.token_timestamps:null;let x=null,y=p;if("stride"in s){const[t,n,o]=s.stride;if(_-=n,w=t-o,n&&(y=n/r+p),o)for(let t=e.length-1;t>=0;--t){const s=Number(e[t]);if(s>=p){if(null!==x&&(s-p)*r<w)break;x=s}}}let k=[],T=[];for(let s=0;s<e.length;++s){const w=Number(e[s]);if(b.has(w)){const e=this.decode([w]),s=d.WHISPER_LANGUAGE_MAPPING.get(e.slice(2,-2));if(void 0!==s){if(null!==o&&s!==o&&!t){h.push(k);const e=this.findLongestCommonSequence(h)[0],t=this.decode(e);u.text=t,c.push(u),h=[],k=[],u=l()}o=u.language=s}}else if(w>=p&&w<=m){const e=(w-p)*r+_,t=(0,i.round)(e,2);if(null!==x&&w>=x)g=!0;else if(g||h.length>0&&w<y)g=!1;else if(null===u.timestamp[0])u.timestamp[0]=t;else if(t===u.timestamp[0]);else{u.timestamp[1]=t,h.push(k),a&&f.push(T);const[e,s]=this.findLongestCommonSequence(h,f),r=this.decode(e);u.text=r,a&&(u.words=this.collateWordTimestamps(e,s,o)),c.push(u),h=[],k=[],f=[],T=[],u=l()}}else if(k.push(w),a){let e,t=(0,i.round)(n[s]+_,2);if(s+1<n.length){e=(0,i.round)(n[s+1]+_,2);const o=this.decode([w]);M.test(o)&&(e=(0,i.round)(Math.min(t+r,e),2))}else e=null;T.push([t,e])}}if("stride"in s){const[e,t,r]=s.stride;_+=e-r}k.length>0?(h.push(k),a&&f.push(T)):h.every((e=>0===e.length))&&(u=l(),h=[],k=[],f=[],T=[])}if(h.length>0){if(n&&t)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[e,s]=this.findLongestCommonSequence(h,f),r=this.decode(e);u.text=r,a&&(u.words=this.collateWordTimestamps(e,s,o)),c.push(u)}let x=Object.create(null);const y=c.map((e=>e.text)).join("");if(t||s){for(let e=0;e<c.length;++e){const r=c[e];t||delete r.timestamp,s||delete r.language}if(a){const e=[];for(const t of c)for(const s of t.words)e.push(s);x={chunks:e}}else x={chunks:c}}return[y,x]}findLongestCommonSequence(e,t=null){let s=e[0],r=s.length,n=[];const o=Array.isArray(t)&&t.length>0;let i=o?[]:null,a=o?t[0]:null;for(let l=1;l<e.length;++l){const c=e[l];let d=0,u=[r,r,0,0];const _=c.length;for(let e=1;e<r+_;++e){const n=Math.max(0,r-e),i=Math.min(r,r+_-e),p=s.slice(n,i),m=Math.max(0,e-r),h=Math.min(_,e),f=c.slice(m,h);if(p.length!==f.length)throw new Error("There is a bug within whisper `decode_asr` function, please report it. Dropping to prevent bad inference.");let g;g=o?p.filter(((e,s)=>e===f[s]&&a[n+s]<=t[l][m+s])).length:p.filter(((e,t)=>e===f[t])).length;const w=g/e+e/1e4;g>1&&w>d&&(d=w,u=[n,i,m,h])}const[p,m,h,f]=u,g=Math.floor((m+p)/2),w=Math.floor((f+h)/2);n.push(...s.slice(0,g)),s=c.slice(w),r=s.length,o&&(i.push(...a.slice(0,g)),a=t[l].slice(w))}return n.push(...s),o?(i.push(...a),[n,i]):[n,[]]}collateWordTimestamps(e,t,s){const[r,n,o]=this.combineTokensIntoWords(e,s),i=[];for(let e=0;e<r.length;++e){const s=o[e];i.push({text:r[e],timestamp:[t[s.at(0)][0],t[s.at(-1)][1]]})}return i}combineTokensIntoWords(e,t,s="\"'“¡¿([{-",r="\"'.。,，!！?？:：”)]}、"){let n,o,i;return["chinese","japanese","thai","lao","myanmar"].includes(t=t??"english")?[n,o,i]=this.splitTokensOnUnicode(e):[n,o,i]=this.splitTokensOnSpaces(e),this.mergePunctuations(n,o,i,s,r)}decode(e,t){let s;return t?.decode_with_timestamps?(e instanceof a.Tensor&&(e=m(e)),s=this.decodeWithTimestamps(e,t)):s=super.decode(e,t),s}decodeWithTimestamps(e,t){const s=t?.time_precision??.02,r=Array.from(this.all_special_ids).at(-1)+1;let n=[[]];for(let t of e)if(t=Number(t),t>=r){const e=((t-r)*s).toFixed(2);n.push(`<|${e}|>`),n.push([])}else n[n.length-1].push(t);return n=n.map((e=>"string"==typeof e?e:super.decode(e,t))),n.join("")}splitTokensOnUnicode(e){const t=this.decode(e,{decode_with_timestamps:!0}),s=[],r=[],n=[];let o=[],i=[],a=0;for(let l=0;l<e.length;++l){const c=e[l];o.push(c),i.push(l);const d=this.decode(o,{decode_with_timestamps:!0});d.includes("�")&&"�"!==t[a+d.indexOf("�")]||(s.push(d),r.push(o),n.push(i),o=[],i=[],a+=d.length)}return[s,r,n]}splitTokensOnSpaces(e){const[t,s,r]=this.splitTokensOnUnicode(e),n=[],o=[],i=[],a=new RegExp(`^[${w}]$`,"gu");for(let e=0;e<t.length;++e){const l=t[e],c=s[e],d=r[e],u=c[0]>=this.model.tokens_to_ids.get("<|endoftext|>"),_=l.startsWith(" "),p=l.trim(),m=a.test(p);if(u||_||m||0===n.length)n.push(l),o.push(c),i.push(d);else{const e=n.length-1;n[e]+=l,o[e].push(...c),i[e].push(...d)}}return[n,o,i]}mergePunctuations(e,t,s,r,o){const i=structuredClone(e),a=structuredClone(t),l=structuredClone(s);let c=i.length-2,d=i.length-1;for(;c>=0;)i[c].startsWith(" ")&&r.includes(i[c].trim())?(i[d]=i[c]+i[d],a[d]=(0,n.mergeArrays)(a[c],a[d]),l[d]=(0,n.mergeArrays)(l[c],l[d]),i[c]="",a[c]=[],l[c]=[]):d=c,--c;for(c=0,d=1;d<i.length;)!i[c].endsWith(" ")&&o.includes(i[d])?(i[c]+=i[d],a[c]=(0,n.mergeArrays)(a[c],a[d]),l[c]=(0,n.mergeArrays)(l[c],l[d]),i[d]="",a[d]=[],l[d]=[]):c=d,++d;return[i.filter((e=>e)),a.filter((e=>e.length>0)),l.filter((e=>e.length>0))]}}class st extends be{}class rt extends be{}class nt extends be{}class ot extends be{constructor(e,t){super(e,t),this.languageRegex=/^(>>\w+<<)\s*/g,this.supported_language_codes=this.model.vocab.filter((e=>this.languageRegex.test(e))),console.warn('WARNING: `MarianTokenizer` is not yet supported by Hugging Face\'s "fast" tokenizers library. Therefore, you may experience slightly inaccurate results.')}_encode_text(e){if(null===e)return null;const[t,...s]=e.trim().split(this.languageRegex);if(0===s.length)return super._encode_text(t);if(2===s.length){const[e,t]=s;return this.supported_language_codes.includes(e)||console.warn(`Unsupported language code "${e}" detected, which may lead to unexpected behavior. Should be one of: ${JSON.stringify(this.supported_language_codes)}`),(0,n.mergeArrays)([e],super._encode_text(t))}}}class it extends be{}class at extends be{}class lt extends be{}class ct extends be{}class dt extends be{}class ut extends be{constructor(e,t){super(e,t),this.decoder=new ce({})}}class _t extends be{}class pt extends be{}class mt{static TOKENIZER_CLASS_MAPPING={T5Tokenizer:ze,DistilBertTokenizer:Se,CamembertTokenizer:Ee,DebertaTokenizer:ve,DebertaV2Tokenizer:Pe,BertTokenizer:xe,HerbertTokenizer:Fe,ConvBertTokenizer:Ce,RoFormerTokenizer:Ae,XLMTokenizer:Ie,ElectraTokenizer:Le,MobileBertTokenizer:ke,SqueezeBertTokenizer:Te,AlbertTokenizer:ye,GPT2Tokenizer:je,BartTokenizer:Ne,MBartTokenizer:Oe,MBart50Tokenizer:Be,RobertaTokenizer:Ve,WhisperTokenizer:tt,CodeGenTokenizer:st,CLIPTokenizer:rt,SiglipTokenizer:nt,MarianTokenizer:ot,BloomTokenizer:De,NllbTokenizer:Ze,M2M100Tokenizer:et,LlamaTokenizer:Ge,CodeLlamaTokenizer:$e,XLMRobertaTokenizer:qe,MPNetTokenizer:Ue,FalconTokenizer:We,GPTNeoXTokenizer:Xe,EsmTokenizer:Qe,Wav2Vec2CTCTokenizer:it,BlenderbotTokenizer:at,BlenderbotSmallTokenizer:lt,SpeechT5Tokenizer:ct,NougatTokenizer:dt,VitsTokenizer:ut,Qwen2Tokenizer:He,GemmaTokenizer:Je,Grok1Tokenizer:Ye,CohereTokenizer:_t,MgpstrTokenizer:pt,PreTrainedTokenizer:be};static async from_pretrained(e,{progress_callback:t=null,config:s=null,cache_dir:r=null,local_files_only:n=!1,revision:o="main",legacy:i=null}={}){const[a,l]=await u(e,{progress_callback:t,config:s,cache_dir:r,local_files_only:n,revision:o,legacy:i}),c=l.tokenizer_class?.replace(/Fast$/,"")??"PreTrainedTokenizer";let d=this.TOKENIZER_CLASS_MAPPING[c];return d||(console.warn(`Unknown tokenizer class "${c}", attempting to construct from base class.`),d=be),new d(a,l)}}},"./src/utils/audio.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{RawAudio:()=>y,hamming:()=>_,hanning:()=>u,mel_filter_bank:()=>g,read_audio:()=>c,spectrogram:()=>M,window_function:()=>b});var r=s("./src/utils/hub.js"),n=s("./src/utils/maths.js"),o=s("./src/utils/core.js"),i=s("./src/env.js"),a=s("fs"),l=s("./src/utils/tensor.js");async function c(e,t){if("undefined"==typeof AudioContext)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 s=await(await(0,r.getFile)(e)).arrayBuffer(),n=new AudioContext({sampleRate:t});void 0===t&&console.warn(`No sampling rate provided, using default of ${n.sampleRate}Hz.`);const o=await n.decodeAudioData(s);let i;if(2===o.numberOfChannels){const e=Math.sqrt(2),t=o.getChannelData(0),s=o.getChannelData(1);i=new Float32Array(t.length);for(let r=0;r<o.length;++r)i[r]=e*(t[r]+s[r])/2}else i=o.getChannelData(0);return i}function d(e,t){if(e<1)return new Float64Array;if(1===e)return new Float64Array([1]);const s=1-t,r=2*Math.PI/(e-1),n=new Float64Array(e);for(let o=0;o<e;++o)n[o]=t-s*Math.cos(o*r);return n}function u(e){return d(e,.5)}function _(e){return d(e,.54)}const p={htk:e=>2595*Math.log10(1+e/700),kaldi:e=>1127*Math.log(1+e/700),slaney:(e,t=1e3,s=15,r=27/Math.log(6.4))=>e>=t?s+Math.log(e/t)*r:3*e/200};function m(e,t="htk"){const s=p[t];if(!s)throw new Error('mel_scale should be one of "htk", "slaney" or "kaldi".');return"number"==typeof e?s(e):e.map((e=>s(e)))}const h={htk:e=>700*(10**(e/2595)-1),kaldi:e=>700*(Math.exp(e/1127)-1),slaney:(e,t=1e3,s=15,r=Math.log(6.4)/27)=>e>=s?t*Math.exp(r*(e-s)):200*e/3};function f(e,t,s){const r=(t-e)/(s-1);return Float64Array.from({length:s},((t,s)=>e+r*s))}function g(e,t,s,r,n,o=null,i="htk",a=!1){if(null!==o&&"slaney"!==o)throw new Error('norm must be one of null or "slaney"');const l=f(m(s,i),m(r,i),t+2);let c,d=function(e,t="htk"){const s=h[t];if(!s)throw new Error('mel_scale should be one of "htk", "slaney" or "kaldi".');return"number"==typeof e?s(e):e.map((e=>s(e)))}(l,i);if(a){const t=n/(2*e);c=m(Float64Array.from({length:e},((e,s)=>s*t)),i),d=l}else c=f(0,Math.floor(n/2),e);const u=function(e,t){const s=Float64Array.from({length:t.length-1},((e,s)=>t[s+1]-t[s])),r=Array.from({length:e.length},(()=>new Array(t.length)));for(let s=0;s<e.length;++s){const n=r[s];for(let r=0;r<t.length;++r)n[r]=t[r]-e[s]}const n=t.length-2,o=Array.from({length:n},(()=>new Array(e.length)));for(let t=0;t<e.length;++t){const e=r[t];for(let r=0;r<n;++r){const n=-e[r]/s[r],i=e[r+2]/s[r+1];o[r][t]=Math.max(0,Math.min(n,i))}}return o}(c,d);if(null!==o&&"slaney"===o)for(let s=0;s<t;++s){const t=u[s],r=2/(d[s+2]-d[s]);for(let s=0;s<e;++s)t[s]*=r}return u}function w(e,t,s,r,o){if(s<=0)throw new Error("reference must be greater than zero");if(r<=0)throw new Error("min_value must be greater than zero");s=Math.max(r,s);const i=Math.log10(s);for(let s=0;s<e.length;++s)e[s]=t*Math.log10(Math.max(r,e[s])-i);if(null!==o){if(o<=0)throw new Error("db_range must be greater than zero");const t=(0,n.max)(e)[0]-o;for(let s=0;s<e.length;++s)e[s]=Math.max(e[s],t)}return e}async function M(e,t,s,r,{fft_length:i=null,power:a=1,center:c=!0,pad_mode:d="reflect",onesided:u=!0,preemphasis:_=null,mel_filters:p=null,mel_floor:m=1e-10,log_mel:h=null,reference:f=1,min_value:g=1e-10,db_range:M=null,remove_dc_offset:b=null,min_num_frames:x=null,max_num_frames:y=null,do_pad:k=!0,transpose:T=!1}={}){const v=t.length;if(null===i&&(i=s),s>i)throw Error(`frame_length (${s}) may not be larger than fft_length (${i})`);if(v!==s)throw new Error(`Length of the window (${v}) must equal frame_length (${s})`);if(r<=0)throw new Error("hop_length must be greater than zero");if(null===a&&null!==p)throw new Error("You have provided `mel_filters` but `power` is `None`. Mel spectrogram computation is not yet supported for complex-valued spectrogram. Specify `power` to fix this issue.");if(c){if("reflect"!==d)throw new Error(`pad_mode="${d}" not implemented yet.`);const t=Math.floor((i-1)/2)+1;e=function(e,t,s){const r=new e.constructor(e.length+t+s),n=e.length-1;for(let s=0;s<e.length;++s)r[t+s]=e[s];for(let s=1;s<=t;++s)r[t-s]=e[(0,o.calculateReflectOffset)(s,n)];for(let i=1;i<=s;++i)r[n+t+i]=e[(0,o.calculateReflectOffset)(n-i,n)];return r}(e,t,t)}let P=Math.floor(1+Math.floor((e.length-s)/r));null!==x&&P<x&&(P=x);const F=u?Math.floor(i/2)+1:i;let C=P,A=P;null!==y&&(y>P?k&&(A=y):A=C=y);const S=new n.FFT(i),E=new Float64Array(i),I=new Float64Array(S.outputBufferSize),L=new Float32Array(F*A);for(let n=0;n<C;++n){const o=n*r,i=Math.min(e.length-o,s);i!==s&&E.fill(0,0,s);for(let t=0;t<i;++t)E[t]=e[o+t];if(b){let e=0;for(let t=0;t<i;++t)e+=E[t];const t=e/i;for(let e=0;e<i;++e)E[e]-=t}if(null!==_){for(let e=i-1;e>=1;--e)E[e]-=_*E[e-1];E[0]*=1-_}for(let e=0;e<t.length;++e)E[e]*=t[e];S.realTransform(I,E);for(let e=0;e<F;++e){const t=e<<1;L[e*A+n]=I[t]**2+I[t+1]**2}}if(null!==a&&2!==a){const e=2/a;for(let t=0;t<L.length;++t)L[t]**=e}const z=p.length;let j=await(0,l.matmul)(new l.Tensor("float32",p.flat(),[z,F]),new l.Tensor("float32",L,[F,A]));T&&(j=j.transpose(1,0));const N=j.data;for(let e=0;e<N.length;++e)N[e]=Math.max(m,N[e]);if(null!==a&&null!==h){const e=Math.min(N.length,C*z);switch(h){case"log":for(let t=0;t<e;++t)N[t]=Math.log(N[t]);break;case"log10":for(let t=0;t<e;++t)N[t]=Math.log10(N[t]);break;case"dB":if(1===a)!function(e,t=1,s=1e-5,r=null){w(e,20,t,s,r)}(N,f,g,M);else{if(2!==a)throw new Error(`Cannot use log_mel option '${h}' with power ${a}`);!function(e,t=1,s=1e-10,r=null){w(e,10,t,s,r)}(N,f,g,M)}break;default:throw new Error(`log_mel must be one of null, 'log', 'log10' or 'dB'. Got '${h}'`)}}return j}function b(e,t,{periodic:s=!0,frame_length:r=null,center:n=!0}={}){const o=s?e+1:e;let i;switch(t){case"boxcar":i=new Float64Array(o).fill(1);break;case"hann":case"hann_window":i=u(o);break;case"hamming":i=_(o);break;case"povey":i=u(o).map((e=>Math.pow(e,.85)));break;default:throw new Error(`Unknown window type ${t}.`)}if(s&&(i=i.subarray(0,e)),null===r)return i;if(e>r)throw new Error(`Length of the window (${e}) may not be larger than frame_length (${r})`);return i}function x(e,t,s){for(let r=0;r<s.length;++r)e.setUint8(t+r,s.charCodeAt(r))}class y{constructor(e,t){this.audio=e,this.sampling_rate=t}toWav(){return function(e,t){let s=44;const r=new ArrayBuffer(s+4*e.length),n=new DataView(r);x(n,0,"RIFF"),n.setUint32(4,36+4*e.length,!0),x(n,8,"WAVE"),x(n,12,"fmt "),n.setUint32(16,16,!0),n.setUint16(20,3,!0),n.setUint16(22,1,!0),n.setUint32(24,t,!0),n.setUint32(28,4*t,!0),n.setUint16(32,4,!0),n.setUint16(34,32,!0),x(n,36,"data"),n.setUint32(40,4*e.length,!0);for(let t=0;t<e.length;++t,s+=4)n.setFloat32(s,e[t],!0);return r}(this.audio,this.sampling_rate)}toBlob(){const e=this.toWav();return new Blob([e],{type:"audio/wav"})}async save(e){let t;if(i.apis.IS_BROWSER_ENV){if(i.apis.IS_WEBWORKER_ENV)throw new Error("Unable to save a file from a Web Worker.");t=o.saveBlob}else{if(!i.apis.IS_FS_AVAILABLE)throw new Error("Unable to save because filesystem is disabled in this environment.");t=async(e,t)=>{let s=await t.arrayBuffer();a.writeFileSync(e,Buffer.from(s))}}await t(e,this.toBlob())}}},"./src/utils/constants.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{CHAT_TEMPLATE_NAME:()=>l,CONFIG_NAME:()=>n,FEATURE_EXTRACTOR_NAME:()=>o,GENERATION_CONFIG_NAME:()=>c,GITHUB_ISSUE_URL:()=>r,IMAGE_PROCESSOR_NAME:()=>i,PROCESSOR_NAME:()=>a});const r="https://github.com/huggingface/transformers.js/issues/new/choose",n="config.json",o="preprocessor_config.json",i=o,a="processor_config.json",l="chat_template.json",c="generation_config.json"},"./src/utils/core.js":(e,t,s)=>{"use strict";function r(e,t){e&&e(t)}function n(e){return Object.fromEntries(Object.entries(e).map((([e,t])=>[t,e])))}function o(e){return e.replace(/[.*+?^${}()|[\]\\]/g,"\\$&")}function i(e){return"TypedArray"===e?.prototype?.__proto__?.constructor?.name}function a(e){return Number.isInteger(e)||"bigint"==typeof e}function l(e){return null==e||-1===e}function c(e){const t=[];let s=e;for(;Array.isArray(s);)t.push(s.length),s=s[0];return t}function d(e,t,s=void 0){const r=e[t];if(void 0!==r)return delete e[t],r;if(void 0===s)throw Error(`Key ${t} does not exist in object.`);return s}function u(...e){return Array.prototype.concat.apply([],e)}function _(...e){return e.reduce(((e,t)=>e.flatMap((e=>t.map((t=>[e,t]))))))}function p(e,t){return Math.abs((e+t)%(2*t)-t)}function m(e,t){const s=URL.createObjectURL(t),r=document.createElement("a");r.href=s,r.download=e,r.click(),r.remove(),URL.revokeObjectURL(s)}function h(e,t){return Object.assign({},...t.map((t=>{if(void 0!==e[t])return{[t]:e[t]}})))}function f(e){let t=0;for(const s of e)++t;return t}function g(e,t){let s=0;for(const r of e)r===t&&++s;return s}s.r(t),s.d(t,{calculateDimensions:()=>c,calculateReflectOffset:()=>p,count:()=>g,dispatchCallback:()=>r,escapeRegExp:()=>o,isIntegralNumber:()=>a,isNullishDimension:()=>l,isTypedArray:()=>i,len:()=>f,mergeArrays:()=>u,pick:()=>h,pop:()=>d,product:()=>_,reverseDictionary:()=>n,saveBlob:()=>m})},"./src/utils/data-structures.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{CharTrie:()=>n,PriorityQueue:()=>r,TokenLattice:()=>i});class r{constructor(e=(e,t)=>e>t,t=1/0){this._heap=[],this._comparator=e,this._maxSize=t}get size(){return this._heap.length}isEmpty(){return 0===this.size}peek(){return this._heap[0]}push(...e){return this.extend(e)}extend(e){for(const t of e)if(this.size<this._maxSize)this._heap.push(t),this._siftUp();else{const e=this._smallest();this._comparator(t,this._heap[e])&&(this._heap[e]=t,this._siftUpFrom(e))}return this.size}pop(){const e=this.peek(),t=this.size-1;return t>0&&this._swap(0,t),this._heap.pop(),this._siftDown(),e}replace(e){const t=this.peek();return this._heap[0]=e,this._siftDown(),t}_parent(e){return(e+1>>>1)-1}_left(e){return 1+(e<<1)}_right(e){return e+1<<1}_greater(e,t){return this._comparator(this._heap[e],this._heap[t])}_swap(e,t){const s=this._heap[e];this._heap[e]=this._heap[t],this._heap[t]=s}_siftUp(){this._siftUpFrom(this.size-1)}_siftUpFrom(e){for(;e>0&&this._greater(e,this._parent(e));)this._swap(e,this._parent(e)),e=this._parent(e)}_siftDown(){let e=0;for(;this._left(e)<this.size&&this._greater(this._left(e),e)||this._right(e)<this.size&&this._greater(this._right(e),e);){const t=this._right(e)<this.size&&this._greater(this._right(e),this._left(e))?this._right(e):this._left(e);this._swap(e,t),e=t}}_smallest(){return 2**Math.floor(Math.log2(this.size))-1}}class n{constructor(){this.root=o.default()}extend(e){for(const t of e)this.push(t)}push(e){let t=this.root;for(const s of e){let e=t.children.get(s);void 0===e&&(e=o.default(),t.children.set(s,e)),t=e}t.isLeaf=!0}*commonPrefixSearch(e){let t=this.root;if(void 0===t)return;let s="";for(const r of e){if(s+=r,t=t.children.get(r),void 0===t)return;t.isLeaf&&(yield s)}}}class o{constructor(e,t){this.isLeaf=e,this.children=t}static default(){return new o(!1,new Map)}}class i{constructor(e,t,s){this.chars=Array.from(e),this.len=this.chars.length,this.bosTokenId=t,this.eosTokenId=s,this.nodes=[],this.beginNodes=Array.from({length:this.len+1},(()=>[])),this.endNodes=Array.from({length:this.len+1},(()=>[]));const r=new a(this.bosTokenId,0,0,0,0),n=new a(this.eosTokenId,1,this.len,0,0);this.nodes.push(r.clone()),this.nodes.push(n.clone()),this.beginNodes[this.len].push(n),this.endNodes[0].push(r)}insert(e,t,s,r){const n=this.nodes.length,o=new a(r,n,e,t,s);this.beginNodes[e].push(o),this.endNodes[e+t].push(o),this.nodes.push(o)}viterbi(){const e=this.len;let t=0;for(;t<=e;){if(0==this.beginNodes[t].length)return[];for(let e of this.beginNodes[t]){e.prev=null;let s=0,r=null;for(let n of this.endNodes[t]){const t=n.backtraceScore+e.score;(null===r||t>s)&&(r=n.clone(),s=t)}if(null===r)return[];e.prev=r,e.backtraceScore=s}++t}const s=[],r=this.beginNodes[e][0].prev;if(null===r)return[];let n=r.clone();for(;null!==n.prev;){s.push(n.clone());const e=n.clone();n=e.prev.clone()}return s.reverse(),s}piece(e){return this.chars.slice(e.pos,e.pos+e.length).join("")}tokens(){return this.viterbi().map((e=>this.piece(e)))}tokenIds(){return this.viterbi().map((e=>e.tokenId))}}class a{constructor(e,t,s,r,n){this.tokenId=e,this.nodeId=t,this.pos=s,this.length=r,this.score=n,this.prev=null,this.backtraceScore=0}clone(){const e=new a(this.tokenId,this.nodeId,this.pos,this.length,this.score);return e.prev=this.prev,e.backtraceScore=this.backtraceScore,e}}},"./src/utils/devices.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DEVICE_TYPES:()=>r});const r=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":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DATA_TYPES:()=>i,DEFAULT_DEVICE_DTYPE_MAPPING:()=>a,DEFAULT_DTYPE_SUFFIX_MAPPING:()=>l,isWebGpuFp16Supported:()=>o});var r=s("./src/env.js"),n=s("./src/utils/devices.js");const o=function(){let e;return async function(){if(void 0===e)if(r.apis.IS_WEBGPU_AVAILABLE)try{const t=await navigator.gpu.requestAdapter();e=t.features.has("shader-f16")}catch(t){e=!1}else e=!1;return e}}(),i=Object.freeze({auto:"auto",fp32:"fp32",fp16:"fp16",q8:"q8",int8:"int8",uint8:"uint8",q4:"q4",bnb4:"bnb4",q4f16:"q4f16"}),a=Object.freeze({[n.DEVICE_TYPES.wasm]:i.q8}),l=Object.freeze({[i.fp32]:"",[i.fp16]:"_fp16",[i.int8]:"_int8",[i.uint8]:"_uint8",[i.q8]:"_quantized",[i.q4]:"_q4",[i.q4f16]:"_q4f16",[i.bnb4]:"_bnb4"})},"./src/utils/generic.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Callable:()=>r});const r=class{constructor(){let e=function(...t){return e._call(...t)};return Object.setPrototypeOf(e,new.target.prototype)}_call(...e){throw Error("Must implement _call method in subclass")}}},"./src/utils/hub.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{getFile:()=>d,getModelFile:()=>p,getModelJSON:()=>m});var r=s("fs"),n=s("path"),o=s("./src/env.js"),i=s("./src/utils/core.js");const a={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 l{constructor(e){if(this.filePath=e,this.headers=new Headers,this.exists=r.existsSync(e),this.exists){this.status=200,this.statusText="OK";let t=r.statSync(e);this.headers.set("content-length",t.size.toString()),this.updateContentType();let s=this;this.body=new ReadableStream({start(e){s.arrayBuffer().then((t=>{e.enqueue(new Uint8Array(t)),e.close()}))}})}else this.status=404,this.statusText="Not Found",this.body=null}updateContentType(){const e=this.filePath.toString().split(".").pop().toLowerCase();this.headers.set("content-type",a[e]??"application/octet-stream")}clone(){let e=new l(this.filePath);return e.exists=this.exists,e.status=this.status,e.statusText=this.statusText,e.headers=new Headers(this.headers),e}async arrayBuffer(){return(await r.promises.readFile(this.filePath)).buffer}async blob(){const e=await r.promises.readFile(this.filePath);return new Blob([e],{type:this.headers.get("content-type")})}async text(){return await r.promises.readFile(this.filePath,"utf8")}async json(){return JSON.parse(await this.text())}}function c(e,t=null,s=null){let r;try{r=new URL(e)}catch(e){return!1}return!(t&&!t.includes(r.protocol))&&!(s&&!s.includes(r.hostname))}async function d(e){if(o.env.useFS&&!c(e,["http:","https:","blob:"]))return new l(e);if("undefined"!=typeof process&&"node"===process?.release?.name){const t=!!process.env?.TESTING_REMOTELY,s=o.env.version,r=new Headers;r.set("User-Agent",`transformers.js/${s}; is_ci/${t};`);if(c(e,["http:","https:"],["huggingface.co","hf.co"])){const e=process.env?.HF_TOKEN??process.env?.HF_ACCESS_TOKEN;e&&r.set("Authorization",`Bearer ${e}`)}return fetch(e,{headers:r})}return fetch(e)}const u={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"};class _{constructor(e){this.path=e}async match(e){let t=n.join(this.path,e),s=new l(t);return s.exists?s:void 0}async put(e,t){const s=Buffer.from(await t.arrayBuffer());let o=n.join(this.path,e);try{await r.promises.mkdir(n.dirname(o),{recursive:!0}),await r.promises.writeFile(o,s)}catch(e){console.warn("An error occurred while writing the file to cache:",e)}}}async function p(e,t,s=!0,r={}){if(!o.env.allowLocalModels){if(r.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(!o.env.allowRemoteModels)throw Error("Invalid configuration detected: both local and remote models are disabled. Fix by setting `env.allowLocalModels` or `env.allowRemoteModels` to `true`.")}let n;if((0,i.dispatchCallback)(r.progress_callback,{status:"initiate",name:e,file:t}),!n&&o.env.useBrowserCache){if("undefined"==typeof caches)throw Error("Browser cache is not available in this environment.");try{n=await caches.open("transformers-cache")}catch(e){console.warn("An error occurred while opening the browser cache:",e)}}if(!n&&o.env.useFSCache&&(n=new _(r.cache_dir??o.env.cacheDir)),!n&&o.env.useCustomCache){if(!o.env.customCache)throw Error("`env.useCustomCache=true`, but `env.customCache` is not defined.");if(!o.env.customCache.match||!o.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");n=o.env.customCache}const a=r.revision??"main";let l,p,m=h(e,t),f=h(o.env.localModelPath,m),g=h(o.env.remoteHost,o.env.remotePathTemplate.replaceAll("{model}",e).replaceAll("{revision}",encodeURIComponent(a)),t),w="main"===a?m:h(e,a,t),M=n instanceof _?w:g,b=!1;n&&(p=await async function(e,...t){for(let s of t)try{let t=await e.match(s);if(t)return t}catch(e){continue}}(n,f,M));const x=void 0!==p;if(void 0===p){if(o.env.allowLocalModels){if(c(m,["http:","https:"])){if(r.local_files_only)throw new Error(`\`local_files_only=true\`, but attempted to load a remote file from: ${m}.`);if(!o.env.allowRemoteModels)throw new Error(`\`env.allowRemoteModels=false\`, but attempted to load a remote file from: ${m}.`)}else try{p=await d(f),l=f}catch(e){console.warn(`Unable to load from local path "${f}": "${e}"`)}}if(void 0===p||404===p.status){if(r.local_files_only||!o.env.allowRemoteModels){if(s)throw Error(`\`local_files_only=true\` or \`env.allowRemoteModels=false\` and file was not found locally at "${f}".`);return null}if(p=await d(g),200!==p.status)return function(e,t,s){if(!s)return null;const r=u[e]??`Error (${e}) occurred while trying to load file`;throw Error(`${r}: "${t}".`)}(p.status,g,s);l=M}b=n&&"undefined"!=typeof Response&&p instanceof Response&&200===p.status}let y;return(0,i.dispatchCallback)(r.progress_callback,{status:"download",name:e,file:t}),r.progress_callback?x&&"undefined"!=typeof navigator&&/firefox/i.test(navigator.userAgent)?(y=new Uint8Array(await p.arrayBuffer()),(0,i.dispatchCallback)(r.progress_callback,{status:"progress",name:e,file:t,progress:100,loaded:y.length,total:y.length})):y=await async function(e,t){const s=e.headers.get("Content-Length");null===s&&console.warn("Unable to determine content-length from response headers. Will expand buffer when needed.");let r=parseInt(s??"0"),n=new Uint8Array(r),o=0;const i=e.body.getReader();async function a(){const{done:e,value:s}=await i.read();if(e)return;let l=o+s.length;if(l>r){r=l;let e=new Uint8Array(r);e.set(n),n=e}n.set(s,o),o=l;return t({progress:o/r*100,loaded:o,total:r}),a()}return await a(),n}(p,(s=>{(0,i.dispatchCallback)(r.progress_callback,{status:"progress",name:e,file:t,...s})})):y=new Uint8Array(await p.arrayBuffer()),b&&l&&void 0===await n.match(l)&&await n.put(l,new Response(y,{headers:p.headers})).catch((e=>{console.warn(`Unable to add response to browser cache: ${e}.`)})),(0,i.dispatchCallback)(r.progress_callback,{status:"done",name:e,file:t}),y}async function m(e,t,s=!0,r={}){let n=await p(e,t,s,r);if(null===n)return{};let o=new TextDecoder("utf-8").decode(n);return JSON.parse(o)}function h(...e){return(e=e.map(((t,s)=>(s&&(t=t.replace(new RegExp("^/"),"")),s!==e.length-1&&(t=t.replace(new RegExp("/$"),"")),t)))).join("/")}},"./src/utils/image.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{RawImage:()=>m,load_image:()=>h});var r=s("./src/utils/core.js"),n=s("./src/utils/hub.js"),o=s("./src/env.js"),i=s("./src/utils/tensor.js"),a=s("sharp");let l,c,d;const u=o.apis.IS_BROWSER_ENV||o.apis.IS_WEBWORKER_ENV;if(u)l=(e,t)=>{if(!self.OffscreenCanvas)throw new Error("OffscreenCanvas not supported by this browser.");return new self.OffscreenCanvas(e,t)},d=self.createImageBitmap,c=self.ImageData;else{if(!a)throw new Error("Unable to load image processing library.");d=async e=>{const t=(await e.metadata()).channels,{data:s,info:r}=await e.rotate().raw().toBuffer({resolveWithObject:!0}),n=new m(new Uint8ClampedArray(s),r.width,r.height,r.channels);return void 0!==t&&t!==r.channels&&n.convert(t),n}}const _={0:"nearest",1:"lanczos",2:"bilinear",3:"bicubic",4:"box",5:"hamming"},p=new Map([["png","image/png"],["jpg","image/jpeg"],["jpeg","image/jpeg"],["gif","image/gif"]]);class m{constructor(e,t,s,r){this.data=e,this.width=t,this.height=s,this.channels=r}get size(){return[this.width,this.height]}static async read(e){if(e instanceof m)return e;if("string"==typeof e||e instanceof URL)return await this.fromURL(e);throw new Error("Unsupported input type: "+typeof e)}static fromCanvas(e){if(!u)throw new Error("fromCanvas() is only supported in browser environments.");const t=e.getContext("2d").getImageData(0,0,e.width,e.height).data;return new m(t,e.width,e.height,4)}static async fromURL(e){const t=await(0,n.getFile)(e);if(200!==t.status)throw new Error(`Unable to read image from "${e}" (${t.status} ${t.statusText})`);const s=await t.blob();return this.fromBlob(s)}static async fromBlob(e){if(u){const t=await d(e),s=l(t.width,t.height).getContext("2d");return s.drawImage(t,0,0),new this(s.getImageData(0,0,t.width,t.height).data,t.width,t.height,4)}{const t=a(await e.arrayBuffer());return await d(t)}}static fromTensor(e,t="CHW"){if(3!==e.dims.length)throw new Error(`Tensor should have 3 dimensions, but has ${e.dims.length} dimensions.`);if("CHW"===t)e=e.transpose(1,2,0);else if("HWC"!==t)throw new Error(`Unsupported channel format: ${t}`);if(!(e.data instanceof Uint8ClampedArray||e.data instanceof Uint8Array))throw new Error(`Unsupported tensor type: ${e.type}`);switch(e.dims[2]){case 1:case 2:case 3:case 4:return new m(e.data,e.dims[1],e.dims[0],e.dims[2]);default:throw new Error(`Unsupported number of channels: ${e.dims[2]}`)}}grayscale(){if(1===this.channels)return this;const e=new Uint8ClampedArray(this.width*this.height*1);switch(this.channels){case 3:case 4:for(let t=0,s=0;t<this.data.length;t+=this.channels){const r=this.data[t],n=this.data[t+1],o=this.data[t+2];e[s++]=Math.round(.2989*r+.587*n+.114*o)}break;default:throw new Error(`Conversion failed due to unsupported number of channels: ${this.channels}`)}return this._update(e,this.width,this.height,1)}rgb(){if(3===this.channels)return this;const e=new Uint8ClampedArray(this.width*this.height*3);switch(this.channels){case 1:for(let t=0,s=0;t<this.data.length;++t)e[s++]=this.data[t],e[s++]=this.data[t],e[s++]=this.data[t];break;case 4:for(let t=0,s=0;t<this.data.length;t+=4)e[s++]=this.data[t],e[s++]=this.data[t+1],e[s++]=this.data[t+2];break;default:throw new Error(`Conversion failed due to unsupported number of channels: ${this.channels}`)}return this._update(e,this.width,this.height,3)}rgba(){if(4===this.channels)return this;const e=new Uint8ClampedArray(this.width*this.height*4);switch(this.channels){case 1:for(let t=0,s=0;t<this.data.length;++t)e[s++]=this.data[t],e[s++]=this.data[t],e[s++]=this.data[t],e[s++]=255;break;case 3:for(let t=0,s=0;t<this.data.length;t+=3)e[s++]=this.data[t],e[s++]=this.data[t+1],e[s++]=this.data[t+2],e[s++]=255;break;default:throw new Error(`Conversion failed due to unsupported number of channels: ${this.channels}`)}return this._update(e,this.width,this.height,4)}putAlpha(e){if(e.width!==this.width||e.height!==this.height)throw new Error(`Expected mask size to be ${this.width}x${this.height}, but got ${e.width}x${e.height}`);if(1!==e.channels)throw new Error(`Expected mask to have 1 channel, but got ${e.channels}`);const t=this.data,s=e.data,r=this.width*this.height;if(3===this.channels){const e=new Uint8ClampedArray(4*r);for(let n=0,o=0,i=0;n<r;++n)e[i++]=t[o++],e[i++]=t[o++],e[i++]=t[o++],e[i++]=s[n];return this._update(e,this.width,this.height,4)}if(4===this.channels){for(let e=0;e<r;++e)t[4*e+3]=s[e];return this}throw new Error(`Expected image to have 3 or 4 channels, but got ${this.channels}`)}async resize(e,t,{resample:s=2}={}){if(this.width===e&&this.height===t)return this;let n=_[s]??s;const o=(0,r.isNullishDimension)(e),i=(0,r.isNullishDimension)(t);if(o&&i)return this;if(o?e=t/this.height*this.width:i&&(t=e/this.width*this.height),u){const s=this.channels,r=this.toCanvas(),n=l(e,t).getContext("2d");n.drawImage(r,0,0,e,t);return new m(n.getImageData(0,0,e,t).data,e,t,4).convert(s)}{let s=this.toSharp();switch(n){case"box":case"hamming":"box"!==n&&"hamming"!==n||(console.warn(`Resampling method ${n} is not yet supported. Using bilinear instead.`),n="bilinear");case"nearest":case"bilinear":case"bicubic":s=s.affine([e/this.width,0,0,t/this.height],{interpolator:n});break;case"lanczos":s=s.resize({width:e,height:t,fit:"fill",kernel:"lanczos3"});break;default:throw new Error(`Resampling method ${n} is not supported.`)}return await d(s)}}async pad([e,t,s,r]){if(e=Math.max(e,0),t=Math.max(t,0),s=Math.max(s,0),r=Math.max(r,0),0===e&&0===t&&0===s&&0===r)return this;if(u){const n=this.channels,o=this.toCanvas(),i=this.width+e+t,a=this.height+s+r,c=l(i,a).getContext("2d");c.drawImage(o,0,0,this.width,this.height,e,s,this.width,this.height);return new m(c.getImageData(0,0,i,a).data,i,a,4).convert(n)}{const n=this.toSharp().extend({left:e,right:t,top:s,bottom:r});return await d(n)}}async crop([e,t,s,r]){if(e=Math.max(e,0),t=Math.max(t,0),s=Math.min(s,this.width-1),r=Math.min(r,this.height-1),0===e&&0===t&&s===this.width-1&&r===this.height-1)return this;const n=s-e+1,o=r-t+1;if(u){const s=this.channels,r=this.toCanvas(),i=l(n,o).getContext("2d");i.drawImage(r,e,t,n,o,0,0,n,o);return new m(i.getImageData(0,0,n,o).data,n,o,4).convert(s)}{const s=this.toSharp().extract({left:e,top:t,width:n,height:o});return await d(s)}}async center_crop(e,t){if(this.width===e&&this.height===t)return this;const s=(this.width-e)/2,r=(this.height-t)/2;if(u){const n=this.channels,o=this.toCanvas(),i=l(e,t).getContext("2d");let a=0,c=0,d=0,u=0;s>=0?a=s:d=-s,r>=0?c=r:u=-r,i.drawImage(o,a,c,e,t,d,u,e,t);return new m(i.getImageData(0,0,e,t).data,e,t,4).convert(n)}{let n=this.toSharp();if(s>=0&&r>=0)n=n.extract({left:Math.floor(s),top:Math.floor(r),width:e,height:t});else if(s<=0&&r<=0){const o=Math.floor(-r),i=Math.floor(-s);n=n.extend({top:o,left:i,right:e-this.width-i,bottom:t-this.height-o})}else{let o=[0,0],i=0;r<0?(o[0]=Math.floor(-r),o[1]=t-this.height-o[0]):i=Math.floor(r);let a=[0,0],l=0;s<0?(a[0]=Math.floor(-s),a[1]=e-this.width-a[0]):l=Math.floor(s),n=n.extend({top:o[0],bottom:o[1],left:a[0],right:a[1]}).extract({left:l,top:i,width:e,height:t})}return await d(n)}}async toBlob(e="image/png",t=1){if(!u)throw new Error("toBlob() is only supported in browser environments.");const s=this.toCanvas();return await s.convertToBlob({type:e,quality:t})}toTensor(e="CHW"){let t=new i.Tensor("uint8",new Uint8Array(this.data),[this.height,this.width,this.channels]);if("HWC"===e);else{if("CHW"!==e)throw new Error(`Unsupported channel format: ${e}`);t=t.permute(2,0,1)}return t}toCanvas(){if(!u)throw new Error("toCanvas() is only supported in browser environments.");const e=this.clone().rgba(),t=l(e.width,e.height),s=new c(e.data,e.width,e.height);return t.getContext("2d").putImageData(s,0,0),t}split(){const{data:e,width:t,height:s,channels:r}=this,n=e.constructor,o=e.length/r,i=Array.from({length:r},(()=>new n(o)));for(let t=0;t<o;++t){const s=r*t;for(let n=0;n<r;++n)i[n][t]=e[s+n]}return i.map((e=>new m(e,t,s,1)))}_update(e,t,s,r=null){return this.data=e,this.width=t,this.height=s,null!==r&&(this.channels=r),this}clone(){return new m(this.data.slice(),this.width,this.height,this.channels)}convert(e){if(this.channels===e)return this;switch(e){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(e){if(!u){if(o.apis.IS_FS_AVAILABLE){const t=this.toSharp();return await t.toFile(e)}throw new Error("Unable to save the image because filesystem is disabled in this environment.")}{if(o.apis.IS_WEBWORKER_ENV)throw new Error("Unable to save an image from a Web Worker.");const t=e.split(".").pop().toLowerCase(),s=p.get(t)??"image/png",n=await this.toBlob(s);(0,r.saveBlob)(e,n)}}toSharp(){if(u)throw new Error("toSharp() is only supported in server-side environments.");return a(this.data,{raw:{width:this.width,height:this.height,channels:this.channels}})}}const h=m.read.bind(m)},"./src/utils/maths.js":(e,t,s)=>{"use strict";function r(e,[t,s,r],[n,o],i="bilinear",a=!1){const l=o/r,c=n/s,d=new e.constructor(n*o*t),u=s*r,_=n*o;for(let i=0;i<n;++i)for(let n=0;n<o;++n){const a=i*o+n,p=(n+.5)/l-.5,m=(i+.5)/c-.5;let h=Math.floor(p),f=Math.floor(m);const g=Math.min(h+1,r-1),w=Math.min(f+1,s-1);h=Math.max(h,0),f=Math.max(f,0);const M=p-h,b=m-f,x=(1-M)*(1-b),y=M*(1-b),k=(1-M)*b,T=M*b,v=f*r,P=w*r,F=v+h,C=v+g,A=P+h,S=P+g;for(let s=0;s<t;++s){const t=s*u;d[s*_+a]=x*e[t+F]+y*e[t+C]+k*e[t+A]+T*e[t+S]}}return d}function n(e,t,s){const r=new Array(s.length),n=new Array(s.length);for(let e=s.length-1,o=1;e>=0;--e)n[e]=o,r[e]=t[s[e]],o*=r[e];const o=s.map(((e,t)=>n[s.indexOf(t)])),i=new e.constructor(e.length);for(let s=0;s<e.length;++s){let r=0;for(let e=t.length-1,n=s;e>=0;--e)r+=n%t[e]*o[e],n=Math.floor(n/t[e]);i[r]=e[s]}return[i,r]}function o(e){const t=u(e)[0],s=e.map((e=>Math.exp(e-t))),r=s.reduce(((e,t)=>e+t),0);return s.map((e=>e/r))}function i(e){const t=u(e)[0];let s=0;for(let r=0;r<e.length;++r)s+=Math.exp(e[r]-t);const r=Math.log(s);return e.map((e=>e-t-r))}function a(e,t){let s=0;for(let r=0;r<e.length;++r)s+=e[r]*t[r];return s}function l(e,t){return a(e,t)/(c(e)*c(t))}function c(e){return Math.sqrt(e.reduce(((e,t)=>e+t*t),0))}function d(e){if(0===e.length)throw Error("Array must not be empty");let t=e[0],s=0;for(let r=1;r<e.length;++r)e[r]<t&&(t=e[r],s=r);return[t,s]}function u(e){if(0===e.length)throw Error("Array must not be empty");let t=e[0],s=0;for(let r=1;r<e.length;++r)e[r]>t&&(t=e[r],s=r);return[t,s]}function _(e){return e>0&&!(e&e-1)}s.r(t),s.d(t,{FFT:()=>h,bankers_round:()=>w,cos_sim:()=>l,dot:()=>a,dynamic_time_warping:()=>M,interpolate_data:()=>r,log_softmax:()=>i,magnitude:()=>c,max:()=>u,medianFilter:()=>f,min:()=>d,permute_data:()=>n,round:()=>g,softmax:()=>o});class p{constructor(e){if(this.size=0|e,this.size<=1||!_(this.size))throw new Error("FFT size must be a power of two larger than 1");this._csize=e<<1,this.table=new Float64Array(2*this.size);for(let e=0;e<this.table.length;e+=2){const t=Math.PI*e/this.size;this.table[e]=Math.cos(t),this.table[e+1]=-Math.sin(t)}let t=0;for(let e=1;this.size>e;e<<=1)++t;this._width=t%2==0?t-1:t,this._bitrev=new Int32Array(1<<this._width);for(let e=0;e<this._bitrev.length;++e){this._bitrev[e]=0;for(let t=0;t<this._width;t+=2){const s=this._width-t-2;this._bitrev[e]|=(e>>>t&3)<<s}}}createComplexArray(){return new Float64Array(this._csize)}fromComplexArray(e,t){const s=t||new Array(e.length>>>1);for(let t=0;t<e.length;t+=2)s[t>>>1]=e[t];return s}toComplexArray(e,t){const s=t||this.createComplexArray();for(let t=0;t<s.length;t+=2)s[t]=e[t>>>1],s[t+1]=0;return s}transform(e,t){if(e===t)throw new Error("Input and output buffers must be different");this._transform4(e,t,1)}realTransform(e,t){if(e===t)throw new Error("Input and output buffers must be different");this._realTransform4(e,t,1)}inverseTransform(e,t){if(e===t)throw new Error("Input and output buffers must be different");this._transform4(e,t,-1);for(let t=0;t<e.length;++t)e[t]/=this.size}_transform4(e,t,s){const r=this._csize;let n,o,i=1<<this._width,a=r/i<<1;const l=this._bitrev;if(4===a)for(n=0,o=0;n<r;n+=a,++o){const s=l[o];this._singleTransform2(t,e,n,s,i)}else for(n=0,o=0;n<r;n+=a,++o){const r=l[o];this._singleTransform4(t,e,n,r,i,s)}const c=this.table;for(i>>=2;i>=2;i>>=2){a=r/i<<1;const t=a>>>2;for(n=0;n<r;n+=a){const r=n+t-1;for(let o=n,a=0;o<r;o+=2,a+=i){const r=o,n=r+t,i=n+t,l=i+t,d=e[r],u=e[r+1],_=e[n],p=e[n+1],m=e[i],h=e[i+1],f=e[l],g=e[l+1],w=c[a],M=s*c[a+1],b=_*w-p*M,x=_*M+p*w,y=c[2*a],k=s*c[2*a+1],T=m*y-h*k,v=m*k+h*y,P=c[3*a],F=s*c[3*a+1],C=f*P-g*F,A=f*F+g*P,S=d+T,E=u+v,I=d-T,L=u-v,z=b+C,j=x+A,N=s*(b-C),O=s*(x-A);e[r]=S+z,e[r+1]=E+j,e[n]=I+O,e[n+1]=L-N,e[i]=S-z,e[i+1]=E-j,e[l]=I-O,e[l+1]=L+N}}}}_singleTransform2(e,t,s,r,n){const o=e[r],i=e[r+1],a=e[r+n],l=e[r+n+1];t[s]=o+a,t[s+1]=i+l,t[s+2]=o-a,t[s+3]=i-l}_singleTransform4(e,t,s,r,n,o){const i=2*n,a=3*n,l=e[r],c=e[r+1],d=e[r+n],u=e[r+n+1],_=e[r+i],p=e[r+i+1],m=e[r+a],h=e[r+a+1],f=l+_,g=c+p,w=l-_,M=c-p,b=d+m,x=u+h,y=o*(d-m),k=o*(u-h);t[s]=f+b,t[s+1]=g+x,t[s+2]=w+k,t[s+3]=M-y,t[s+4]=f-b,t[s+5]=g-x,t[s+6]=w-k,t[s+7]=M+y}_realTransform4(e,t,s){const r=this._csize;let n,o,i=1<<this._width,a=r/i<<1;const l=this._bitrev;if(4===a)for(n=0,o=0;n<r;n+=a,++o){const s=l[o];this._singleRealTransform2(t,e,n,s>>>1,i>>>1)}else for(n=0,o=0;n<r;n+=a,++o){const r=l[o];this._singleRealTransform4(t,e,n,r>>>1,i>>>1,s)}const c=this.table;for(i>>=2;i>=2;i>>=2){a=r/i<<1;const t=a>>>1,o=t>>>1,l=o>>>1;for(n=0;n<r;n+=a)for(let r=0,a=0;r<=l;r+=2,a+=i){const i=n+r,d=i+o,u=d+o,_=u+o,p=e[i],m=e[i+1],h=e[d],f=e[d+1],g=e[u],w=e[u+1],M=e[_],b=e[_+1],x=p,y=m,k=c[a],T=s*c[a+1],v=h*k-f*T,P=h*T+f*k,F=c[2*a],C=s*c[2*a+1],A=g*F-w*C,S=g*C+w*F,E=c[3*a],I=s*c[3*a+1],L=M*E-b*I,z=M*I+b*E,j=x+A,N=y+S,O=x-A,B=y-S,V=v+L,D=P+z,R=s*(v-L),G=s*(P-z);if(e[i]=j+V,e[i+1]=N+D,e[d]=O+G,e[d+1]=B-R,0===r){e[u]=j-V,e[u+1]=N-D;continue}if(r===l)continue;const $=n+o-r,q=n+t-r;e[$]=O-s*G,e[$+1]=-B-s*R,e[q]=j-s*V,e[q+1]=s*D-N}}const d=r>>>1;for(let t=2;t<d;t+=2)e[r-t]=e[t],e[r-t+1]=-e[t+1]}_singleRealTransform2(e,t,s,r,n){const o=e[r],i=e[r+n];t[s]=o+i,t[s+1]=0,t[s+2]=o-i,t[s+3]=0}_singleRealTransform4(e,t,s,r,n,o){const i=2*n,a=3*n,l=e[r],c=e[r+n],d=e[r+i],u=e[r+a],_=l+d,p=l-d,m=c+u,h=o*(c-u);t[s]=_+m,t[s+1]=0,t[s+2]=p,t[s+3]=-h,t[s+4]=_-m,t[s+5]=0,t[s+6]=p,t[s+7]=h}}class m{constructor(e){const t=2*(e-1),s=2*(2*e-1),r=2**Math.ceil(Math.log2(s));this.bufferSize=r,this._a=t;const n=new Float64Array(s),o=new Float64Array(r);this._chirpBuffer=new Float64Array(r),this._buffer1=new Float64Array(r),this._buffer2=new Float64Array(r),this._outBuffer1=new Float64Array(r),this._outBuffer2=new Float64Array(r);const i=-2*Math.PI/e,a=Math.cos(i),l=Math.sin(i);for(let t=0;t<s>>1;++t){const s=(t+1-e)**2/2,r=Math.sqrt(a**2+l**2)**s,i=s*Math.atan2(l,a),c=2*t;n[c]=r*Math.cos(i),n[c+1]=r*Math.sin(i),o[c]=n[c],o[c+1]=-n[c+1]}this._slicedChirpBuffer=n.subarray(t,s),this._f=new p(r>>1),this._f.transform(this._chirpBuffer,o)}_transform(e,t,s){const r=this._buffer1,n=this._buffer2,o=this._outBuffer1,i=this._outBuffer2,a=this._chirpBuffer,l=this._slicedChirpBuffer,c=this._a;if(s)for(let e=0;e<l.length;e+=2){const s=e+1,n=t[e>>1];r[e]=n*l[e],r[s]=n*l[s]}else for(let e=0;e<l.length;e+=2){const s=e+1;r[e]=t[e]*l[e]-t[s]*l[s],r[s]=t[e]*l[s]+t[s]*l[e]}this._f.transform(o,r);for(let e=0;e<a.length;e+=2){const t=e+1;n[e]=o[e]*a[e]-o[t]*a[t],n[t]=o[e]*a[t]+o[t]*a[e]}this._f.inverseTransform(i,n);for(let t=0;t<i.length;t+=2){const s=i[t+c],r=i[t+c+1],n=l[t],o=l[t+1];e[t]=s*n-r*o,e[t+1]=s*o+r*n}}transform(e,t){this._transform(e,t,!1)}realTransform(e,t){this._transform(e,t,!0)}}class h{constructor(e){this.fft_length=e,this.isPowerOfTwo=_(e),this.isPowerOfTwo?(this.fft=new p(e),this.outputBufferSize=2*e):(this.fft=new m(e),this.outputBufferSize=this.fft.bufferSize)}realTransform(e,t){this.fft.realTransform(e,t)}transform(e,t){this.fft.transform(e,t)}}function f(e,t){if(t%2==0||t<=0)throw new Error("Window size must be a positive odd number");const s=new e.constructor(e.length),r=new e.constructor(t),n=Math.floor(t/2);for(let t=0;t<e.length;++t){let o=0;for(let s=-n;s<=n;++s){let n=t+s;n<0?n=Math.abs(n):n>=e.length&&(n=2*(e.length-1)-n),r[o++]=e[n]}r.sort(),s[t]=r[n]}return s}function g(e,t){const s=Math.pow(10,t);return Math.round(e*s)/s}function w(e){const t=Math.round(e);return Math.abs(e)%1==.5?t%2==0?t:t-1:t}function M(e){const t=e.length,s=e[0].length,r=[t+1,s+1],n=Array.from({length:r[0]},(()=>Array(r[1]).fill(1/0)));n[0][0]=0;const o=Array.from({length:r[0]},(()=>Array(r[1]).fill(-1)));for(let t=1;t<r[1];++t)for(let s=1;s<r[0];++s){const r=n[s-1][t-1],i=n[s-1][t],a=n[s][t-1];let l,c;r<i&&r<a?(l=r,c=0):i<r&&i<a?(l=i,c=1):(l=a,c=2),n[s][t]=e[s-1][t-1]+l,o[s][t]=c}for(let e=0;e<r[1];++e)o[0][e]=2;for(let e=0;e<r[0];++e)o[e][0]=1;let i=t,a=s,l=[],c=[];for(;i>0||a>0;)switch(l.push(i-1),c.push(a-1),o[i][a]){case 0:--i,--a;break;case 1:--i;break;case 2:--a;break;default:throw new Error(`Internal error in dynamic time warping. Unexpected trace[${i}, ${a}]. Please file a bug report.`)}return l.reverse(),c.reverse(),[l,c]}},"./src/utils/tensor.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Tensor:()=>a,cat:()=>x,full:()=>F,full_like:()=>C,interpolate:()=>c,interpolate_4d:()=>d,layer_norm:()=>g,matmul:()=>u,mean:()=>v,mean_pooling:()=>f,ones:()=>A,ones_like:()=>S,permute:()=>l,quantize_embeddings:()=>z,rand:()=>L,rfft:()=>_,slice:()=>h,stack:()=>y,std_mean:()=>T,topk:()=>p,zeros:()=>E,zeros_like:()=>I});var r=s("./src/utils/maths.js"),n=s("./src/backends/onnx.js"),o=s("./src/ops/registry.js");const i=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 a{get dims(){return this.ort_tensor.dims}set dims(e){this.ort_tensor.dims=e}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}ort_tensor;constructor(...e){return(0,n.isONNXTensor)(e[0])?this.ort_tensor=e[0]:this.ort_tensor=new n.Tensor(e[0],e[1],e[2]),new Proxy(this,{get:(e,t)=>{if("string"==typeof t){let s=Number(t);if(Number.isInteger(s))return e._getitem(s)}return e[t]},set:(e,t,s)=>e[t]=s})}dispose(){this.ort_tensor.dispose()}*[Symbol.iterator](){const[e,...t]=this.dims;if(t.length>0){const s=t.reduce(((e,t)=>e*t));for(let r=0;r<e;++r)yield this._subarray(r,s,t)}else yield*this.data}_getitem(e){const[t,...s]=this.dims;if(e=b(e,t),s.length>0){const t=s.reduce(((e,t)=>e*t));return this._subarray(e,t,s)}return new a(this.type,[this.data[e]],s)}indexOf(e){const t=this.data;for(let s=0;s<t.length;++s)if(t[s]==e)return s;return-1}_subarray(e,t,s){const r=e*t,n=(e+1)*t,o="subarray"in this.data?this.data.subarray(r,n):this.data.slice(r,n);return new a(this.type,o,s)}item(){const e=this.data;if(1!==e.length)throw new Error(`a Tensor with ${e.length} elements cannot be converted to Scalar`);return e[0]}tolist(){return function(e,t){const s=e.length,r=t.reduce(((e,t)=>e*t));if(s!==r)throw Error(`cannot reshape array of size ${s} into shape (${t})`);let n=e;for(let e=t.length-1;e>=0;e--)n=n.reduce(((s,r)=>{let n=s[s.length-1];return n.length<t[e]?n.push(r):s.push([r]),s}),[[]]);return n[0]}(this.data,this.dims)}sigmoid(){return this.clone().sigmoid_()}sigmoid_(){const e=this.data;for(let t=0;t<e.length;++t)e[t]=1/(1+Math.exp(-e[t]));return this}map(e){return this.clone().map_(e)}map_(e){const t=this.data;for(let s=0;s<t.length;++s)t[s]=e(t[s],s,t);return this}mul(e){return this.clone().mul_(e)}mul_(e){const t=this.data;for(let s=0;s<t.length;++s)t[s]*=e;return this}div(e){return this.clone().div_(e)}div_(e){const t=this.data;for(let s=0;s<t.length;++s)t[s]/=e;return this}add(e){return this.clone().add_(e)}add_(e){const t=this.data;for(let s=0;s<t.length;++s)t[s]+=e;return this}sub(e){return this.clone().sub_(e)}sub_(e){const t=this.data;for(let s=0;s<t.length;++s)t[s]-=e;return this}clone(){return new a(this.type,this.data.slice(),this.dims.slice())}slice(...e){const t=[],s=[];for(let r=0;r<this.dims.length;++r){let n=e[r];if(null==n)s.push([0,this.dims[r]]),t.push(this.dims[r]);else if("number"==typeof n)n=b(n,this.dims[r],r),s.push([n,n+1]);else{if(!Array.isArray(n)||2!==n.length)throw new Error(`Invalid slice: ${n}`);{let[e,o]=n;if(e=null===e?0:b(e,this.dims[r],r,!1),o=null===o?this.dims[r]:b(o,this.dims[r],r,!1),e>o)throw new Error(`Invalid slice: ${n}`);const i=[Math.max(e,0),Math.min(o,this.dims[r])];s.push(i),t.push(i[1]-i[0])}}}const r=s.map((([e,t])=>t-e)),n=r.reduce(((e,t)=>e*t)),o=this.data,i=new o.constructor(n),l=this.stride();for(let e=0;e<n;++e){let t=0;for(let n=r.length-1,o=e;n>=0;--n){const e=r[n];t+=(o%e+s[n][0])*l[n],o=Math.floor(o/e)}i[e]=o[t]}return new a(this.type,i,t)}permute(...e){return l(this,e)}transpose(...e){return this.permute(...e)}sum(e=null,t=!1){return this.norm(1,e,t)}norm(e="fro",t=null,s=!1){if("fro"===e)e=2;else if("string"==typeof e)throw Error(`Unsupported norm: ${e}`);const r=this.data,n=(t,s)=>t+s**e;if(null===t){const t=r.reduce(n,0)**(1/e);return new a(this.type,[t],[])}const[o,i,l]=k(n,this,t,s);if(1!==e)for(let t=0;t<i.length;++t)i[t]=i[t]**(1/e);return new a(o,i,l)}normalize_(e=2,t=1){t=b(t,this.dims.length);const s=this.norm(e,t,!0),r=this.data,n=s.data;for(let e=0;e<r.length;++e){let s=0;for(let r=this.dims.length-1,n=e,o=1;r>=0;--r){const e=this.dims[r];if(r!==t){s+=n%e*o,o*=this.dims[r]}n=Math.floor(n/e)}r[e]/=n[s]}return this}normalize(e=2,t=1){return this.clone().normalize_(e,t)}stride(){return function(e){const t=new Array(e.length);for(let s=e.length-1,r=1;s>=0;--s)t[s]=r,r*=e[s];return t}(this.dims)}squeeze(e=null){return new a(this.type,this.data,w(this.dims,e))}squeeze_(e=null){return this.dims=w(this.dims,e),this}unsqueeze(e=null){return new a(this.type,this.data,M(this.dims,e))}unsqueeze_(e=null){return this.dims=M(this.dims,e),this}flatten_(e=0,t=-1){t=(t+this.dims.length)%this.dims.length;let s=this.dims.slice(0,e),r=this.dims.slice(e,t+1),n=this.dims.slice(t+1);return this.dims=[...s,r.reduce(((e,t)=>e*t),1),...n],this}flatten(e=0,t=-1){return this.clone().flatten_(e,t)}view(...e){let t=-1;for(let s=0;s<e.length;++s)if(-1===e[s]){if(-1!==t)throw new Error("Only one dimension can be inferred");t=s}const s=this.data;if(-1!==t){const r=e.reduce(((e,s,r)=>r!==t?e*s:e),1);e[t]=s.length/r}return new a(this.type,s,e)}neg_(){const e=this.data;for(let t=0;t<e.length;++t)e[t]=-e[t];return this}neg(){return this.clone().neg_()}gt(e){const t=new Uint8Array(this.data.length),s=this.data;for(let r=0;r<s.length;++r)t[r]=s[r]>e?1:0;return new a("bool",t,this.dims)}lt(e){const t=new Uint8Array(this.data.length),s=this.data;for(let r=0;r<s.length;++r)t[r]=s[r]<e?1:0;return new a("bool",t,this.dims)}clamp_(e,t){const s=this.data;for(let r=0;r<s.length;++r)s[r]=Math.min(Math.max(s[r],e),t);return this}clamp(e,t){return this.clone().clamp_(e,t)}round_(){const e=this.data;for(let t=0;t<e.length;++t)e[t]=Math.round(e[t]);return this}round(){return this.clone().round_()}mean(e=null,t=!1){return v(this,e,t)}min(e=null,t=!1){if(null===e){const e=(0,r.min)(this.data)[0];return new a(this.type,[e],[])}const[s,n,o]=k(((e,t)=>Math.min(e,t)),this,e,t,1/0);return new a(s,n,o)}max(e=null,t=!1){if(null===e){const e=(0,r.max)(this.data)[0];return new a(this.type,[e],[])}const[s,n,o]=k(((e,t)=>Math.max(e,t)),this,e,t,-1/0);return new a(s,n,o)}argmin(e=null,t=!1){if(null!==e)throw new Error("`dim !== null` not yet implemented.");const s=(0,r.min)(this.data)[1];return new a("int64",[BigInt(s)],[])}argmax(e=null,t=!1){if(null!==e)throw new Error("`dim !== null` not yet implemented.");const s=(0,r.max)(this.data)[1];return new a("int64",[BigInt(s)],[])}to(e){if(this.type===e)return this;if(!i.hasOwnProperty(e))throw new Error(`Unsupported type: ${e}`);let t;const s=["int64","uint64"].includes(this.type),r=["int64","uint64"].includes(e);return s&&!r?t=Number:!s&&r&&(t=BigInt),new a(e,i[e].from(this.data,t),this.dims)}}function l(e,t){const[s,n]=(0,r.permute_data)(e.data,e.dims,t);return new a(e.type,s,n)}function c(e,[t,s],n="bilinear",o=!1){const i=e.dims.at(-3)??1,l=e.dims.at(-2),c=e.dims.at(-1);let d=(0,r.interpolate_data)(e.data,[i,l,c],[t,s],n,o);return new a(e.type,d,[i,t,s])}async function d(e,{size:t=null,mode:s="bilinear"}={}){if(4!==e.dims.length)throw new Error("`interpolate_4d` currently only supports 4D input.");if(!t)throw new Error("`interpolate_4d` requires a `size` argument.");let r,n;if(2===t.length)r=[...e.dims.slice(0,2),...t];else if(3===t.length)r=[e.dims[0],...t];else{if(4!==t.length)throw new Error("`size` must be of length 2, 3, or 4.");r=t}if("nearest"===s)n=await o.TensorOpRegistry.nearest_interpolate_4d;else if("bilinear"===s)n=await o.TensorOpRegistry.bilinear_interpolate_4d;else{if("bicubic"!==s)throw new Error(`Unsupported mode: ${s}`);n=await o.TensorOpRegistry.bicubic_interpolate_4d}const i=new a("int64",new BigInt64Array(r.map(BigInt)),[r.length]);return await n({x:e,s:i})}async function u(e,t){const s=await o.TensorOpRegistry.matmul;return await s({a:e,b:t})}async function _(e,t){const s=await o.TensorOpRegistry.rfft;return await s({x:e,a:t})}async function p(e,t){const s=await o.TensorOpRegistry.top_k;return t=null==t?e.dims.at(-1):Math.min(t,e.dims.at(-1)),await s({x:e,k:new a("int64",[BigInt(t)],[1])})}const m=e=>new a("int64",e,[e.length]);async function h(e,t,s,r,n){const i=await o.TensorOpRegistry.slice;return await i({x:e,s:m(t),e:m(s),a:m(r),t:m(n??new Array(r.length).fill(1))})}function f(e,t){const s=e.data,r=t.data,n=[e.dims[0],e.dims[2]],o=new s.constructor(n[0]*n[1]),[i,l,c]=e.dims;let d=0;for(let e=0;e<i;++e){const t=e*c*l;for(let n=0;n<c;++n){let i=0,a=0;const u=e*l,_=t+n;for(let e=0;e<l;++e){const t=Number(r[u+e]);a+=t,i+=s[_+e*c]*t}const p=i/a;o[d++]=p}}return new a(e.type,o,n)}function g(e,t,{eps:s=1e-5}={}){if(2!==e.dims.length)throw new Error("`layer_norm` currently only supports 2D input.");const[r,n]=e.dims;if(1!==t.length&&t[0]!==n)throw new Error("`normalized_shape` must be a 1D array with shape `[input.dims[1]]`.");const[o,i]=T(e,1,0,!0),l=o.data,c=i.data,d=e.data,u=new d.constructor(d.length);for(let e=0;e<r;++e){const t=e*n;for(let r=0;r<n;++r){const n=t+r;u[n]=(d[n]-c[e])/(l[e]+s)}}return new a(e.type,u,e.dims)}function w(e,t){return e=e.slice(),null===t?e=e.filter((e=>1!==e)):"number"==typeof t?1===e[t]&&e.splice(t,1):Array.isArray(t)&&(e=e.filter(((e,s)=>1!==e||!t.includes(s)))),e}function M(e,t){return t=b(t,e.length+1),(e=e.slice()).splice(t,0,1),e}function b(e,t,s=null,r=!0){if(r&&(e<-t||e>=t))throw new Error(`IndexError: index ${e} is out of bounds for dimension${null===s?"":" "+s} with size ${t}`);return e<0&&(e=(e%t+t)%t),e}function x(e,t=0){t=b(t,e[0].dims.length);const s=e[0].dims.slice();s[t]=e.reduce(((e,s)=>e+s.dims[t]),0);const r=s.reduce(((e,t)=>e*t),1),n=new e[0].data.constructor(r),o=e[0].type;if(0===t){let t=0;for(const s of e){const e=s.data;n.set(e,t),t+=e.length}}else{let r=0;for(let o=0;o<e.length;++o){const{data:i,dims:a}=e[o];for(let e=0;e<i.length;++e){let o=0;for(let n=a.length-1,i=e,l=1;n>=0;--n){const e=a[n];let c=i%e;n===t&&(c+=r),o+=c*l,l*=s[n],i=Math.floor(i/e)}n[o]=i[e]}r+=a[t]}}return new a(o,n,s)}function y(e,t=0){return x(e.map((e=>e.unsqueeze(t))),t)}function k(e,t,s=null,r=!1,n=null){const o=t.data,i=t.dims;s=b(s,i.length);const a=i.slice();a[s]=1;const l=new o.constructor(o.length/i[s]);null!==n&&l.fill(n);for(let t=0;t<o.length;++t){let r=0;for(let e=i.length-1,n=t,o=1;e>=0;--e){const t=i[e];if(e!==s){r+=n%t*o,o*=a[e]}n=Math.floor(n/t)}l[r]=e(l[r],o[t],t,r)}return r||a.splice(s,1),[t.type,l,a]}function T(e,t=null,s=1,r=!1){const n=e.data,o=e.dims;if(null===t){const t=n.reduce(((e,t)=>e+t),0)/n.length,r=Math.sqrt(n.reduce(((e,s)=>e+(s-t)**2),0)/(n.length-s)),o=new a(e.type,[t],[]);return[new a(e.type,[r],[]),o]}const i=v(e,t=b(t,o.length),r),l=i.data,[c,d,u]=k(((e,t,s,r)=>e+(t-l[r])**2),e,t,r);for(let e=0;e<d.length;++e)d[e]=Math.sqrt(d[e]/(o[t]-s));return[new a(c,d,u),i]}function v(e,t=null,s=!1){const r=e.dims,n=e.data;if(null===t){const t=n.reduce(((e,t)=>e+t),0);return new a(e.type,[t/n.length],[])}t=b(t,r.length);const[o,i,l]=k(((e,t)=>e+t),e,t,s);if(1!==r[t])for(let e=0;e<i.length;++e)i[e]/=r[t];return new a(o,i,l)}function P(e,t,s,r){const n=e.reduce(((e,t)=>e*t),1);return new a(s,new r(n).fill(t),e)}function F(e,t){let s,r;if("number"==typeof t)s="float32",r=Float32Array;else if("bigint"==typeof t)s="int64",r=BigInt64Array;else{if("boolean"!=typeof t)throw new Error("Unsupported data type: "+typeof t);s="bool",r=Uint8Array}return P(e,t,s,r)}function C(e,t){return F(e.dims,t)}function A(e){return P(e,1n,"int64",BigInt64Array)}function S(e){return A(e.dims)}function E(e){return P(e,0n,"int64",BigInt64Array)}function I(e){return E(e.dims)}function L(e){const t=e.reduce(((e,t)=>e*t),1);return new a("float32",Float32Array.from({length:t},(()=>Math.random())),e)}function z(e,t){if(2!==e.dims.length)throw new Error("The tensor must have 2 dimensions");if(e.dims.at(-1)%8!=0)throw new Error("The last dimension of the tensor must be a multiple of 8");if(!["binary","ubinary"].includes(t))throw new Error("The precision must be either 'binary' or 'ubinary'");const s="binary"===t,r=s?"int8":"uint8",n=s?Int8Array:Uint8Array,o=e.data,i=new n(o.length/8);for(let e=0;e<o.length;++e){const t=o[e]>0?1:0,r=Math.floor(e/8),n=e%8;i[r]|=t<<7-n,s&&0===n&&(i[r]-=128)}return new a(r,i,[e.dims[0],e.dims[1]/8])}}},r={};function n(e){var t=r[e];if(void 0!==t)return t.exports;var o=r[e]={exports:{}};return s[e](o,o.exports,n),o.exports}t=Object.getPrototypeOf?e=>Object.getPrototypeOf(e):e=>e.__proto__,n.t=function(s,r){if(1&r&&(s=this(s)),8&r)return s;if("object"==typeof s&&s){if(4&r&&s.__esModule)return s;if(16&r&&"function"==typeof s.then)return s}var o=Object.create(null);n.r(o);var i={};e=e||[null,t({}),t([]),t(t)];for(var a=2&r&&s;"object"==typeof a&&!~e.indexOf(a);a=t(a))Object.getOwnPropertyNames(a).forEach((e=>i[e]=()=>s[e]));return i.default=()=>s,n.d(o,i),o},n.d=(e,t)=>{for(var s in t)n.o(t,s)&&!n.o(e,s)&&Object.defineProperty(e,s,{enumerable:!0,get:t[s]})},n.o=(e,t)=>Object.prototype.hasOwnProperty.call(e,t),n.r=e=>{"undefined"!=typeof Symbol&&Symbol.toStringTag&&Object.defineProperty(e,Symbol.toStringTag,{value:"Module"}),Object.defineProperty(e,"__esModule",{value:!0})};var o={};(()=>{"use strict";n.r(o),n.d(o,{ASTFeatureExtractor:()=>_.ASTFeatureExtractor,ASTForAudioClassification:()=>s.ASTForAudioClassification,ASTModel:()=>s.ASTModel,ASTPreTrainedModel:()=>s.ASTPreTrainedModel,AlbertForMaskedLM:()=>s.AlbertForMaskedLM,AlbertForQuestionAnswering:()=>s.AlbertForQuestionAnswering,AlbertForSequenceClassification:()=>s.AlbertForSequenceClassification,AlbertModel:()=>s.AlbertModel,AlbertPreTrainedModel:()=>s.AlbertPreTrainedModel,AlbertTokenizer:()=>r.AlbertTokenizer,AudioClassificationPipeline:()=>t.AudioClassificationPipeline,AutoConfig:()=>i.AutoConfig,AutoFeatureExtractor:()=>p.AutoFeatureExtractor,AutoImageProcessor:()=>f.AutoImageProcessor,AutoModel:()=>s.AutoModel,AutoModelForAudioClassification:()=>s.AutoModelForAudioClassification,AutoModelForAudioFrameClassification:()=>s.AutoModelForAudioFrameClassification,AutoModelForCTC:()=>s.AutoModelForCTC,AutoModelForCausalLM:()=>s.AutoModelForCausalLM,AutoModelForDepthEstimation:()=>s.AutoModelForDepthEstimation,AutoModelForDocumentQuestionAnswering:()=>s.AutoModelForDocumentQuestionAnswering,AutoModelForImageClassification:()=>s.AutoModelForImageClassification,AutoModelForImageFeatureExtraction:()=>s.AutoModelForImageFeatureExtraction,AutoModelForImageMatting:()=>s.AutoModelForImageMatting,AutoModelForImageSegmentation:()=>s.AutoModelForImageSegmentation,AutoModelForImageToImage:()=>s.AutoModelForImageToImage,AutoModelForMaskGeneration:()=>s.AutoModelForMaskGeneration,AutoModelForMaskedLM:()=>s.AutoModelForMaskedLM,AutoModelForNormalEstimation:()=>s.AutoModelForNormalEstimation,AutoModelForObjectDetection:()=>s.AutoModelForObjectDetection,AutoModelForPoseEstimation:()=>s.AutoModelForPoseEstimation,AutoModelForQuestionAnswering:()=>s.AutoModelForQuestionAnswering,AutoModelForSemanticSegmentation:()=>s.AutoModelForSemanticSegmentation,AutoModelForSeq2SeqLM:()=>s.AutoModelForSeq2SeqLM,AutoModelForSequenceClassification:()=>s.AutoModelForSequenceClassification,AutoModelForSpeechSeq2Seq:()=>s.AutoModelForSpeechSeq2Seq,AutoModelForTextToSpectrogram:()=>s.AutoModelForTextToSpectrogram,AutoModelForTextToWaveform:()=>s.AutoModelForTextToWaveform,AutoModelForTokenClassification:()=>s.AutoModelForTokenClassification,AutoModelForUniversalSegmentation:()=>s.AutoModelForUniversalSegmentation,AutoModelForVision2Seq:()=>s.AutoModelForVision2Seq,AutoModelForXVector:()=>s.AutoModelForXVector,AutoModelForZeroShotObjectDetection:()=>s.AutoModelForZeroShotObjectDetection,AutoProcessor:()=>M.AutoProcessor,AutoTokenizer:()=>r.AutoTokenizer,AutomaticSpeechRecognitionPipeline:()=>t.AutomaticSpeechRecognitionPipeline,BartForConditionalGeneration:()=>s.BartForConditionalGeneration,BartForSequenceClassification:()=>s.BartForSequenceClassification,BartModel:()=>s.BartModel,BartPretrainedModel:()=>s.BartPretrainedModel,BartTokenizer:()=>r.BartTokenizer,BaseModelOutput:()=>s.BaseModelOutput,BaseStreamer:()=>b.BaseStreamer,BeitFeatureExtractor:()=>h.BeitFeatureExtractor,BeitForImageClassification:()=>s.BeitForImageClassification,BeitModel:()=>s.BeitModel,BeitPreTrainedModel:()=>s.BeitPreTrainedModel,BertForMaskedLM:()=>s.BertForMaskedLM,BertForQuestionAnswering:()=>s.BertForQuestionAnswering,BertForSequenceClassification:()=>s.BertForSequenceClassification,BertForTokenClassification:()=>s.BertForTokenClassification,BertModel:()=>s.BertModel,BertPreTrainedModel:()=>s.BertPreTrainedModel,BertTokenizer:()=>r.BertTokenizer,BitImageProcessor:()=>h.BitImageProcessor,BlenderbotForConditionalGeneration:()=>s.BlenderbotForConditionalGeneration,BlenderbotModel:()=>s.BlenderbotModel,BlenderbotPreTrainedModel:()=>s.BlenderbotPreTrainedModel,BlenderbotSmallForConditionalGeneration:()=>s.BlenderbotSmallForConditionalGeneration,BlenderbotSmallModel:()=>s.BlenderbotSmallModel,BlenderbotSmallPreTrainedModel:()=>s.BlenderbotSmallPreTrainedModel,BlenderbotSmallTokenizer:()=>r.BlenderbotSmallTokenizer,BlenderbotTokenizer:()=>r.BlenderbotTokenizer,BloomForCausalLM:()=>s.BloomForCausalLM,BloomModel:()=>s.BloomModel,BloomPreTrainedModel:()=>s.BloomPreTrainedModel,BloomTokenizer:()=>r.BloomTokenizer,CLIPFeatureExtractor:()=>h.CLIPFeatureExtractor,CLIPImageProcessor:()=>h.CLIPImageProcessor,CLIPModel:()=>s.CLIPModel,CLIPPreTrainedModel:()=>s.CLIPPreTrainedModel,CLIPSegForImageSegmentation:()=>s.CLIPSegForImageSegmentation,CLIPSegModel:()=>s.CLIPSegModel,CLIPSegPreTrainedModel:()=>s.CLIPSegPreTrainedModel,CLIPTextModel:()=>s.CLIPTextModel,CLIPTextModelWithProjection:()=>s.CLIPTextModelWithProjection,CLIPTokenizer:()=>r.CLIPTokenizer,CLIPVisionModel:()=>s.CLIPVisionModel,CLIPVisionModelWithProjection:()=>s.CLIPVisionModelWithProjection,CamembertForMaskedLM:()=>s.CamembertForMaskedLM,CamembertForQuestionAnswering:()=>s.CamembertForQuestionAnswering,CamembertForSequenceClassification:()=>s.CamembertForSequenceClassification,CamembertForTokenClassification:()=>s.CamembertForTokenClassification,CamembertModel:()=>s.CamembertModel,CamembertPreTrainedModel:()=>s.CamembertPreTrainedModel,CamembertTokenizer:()=>r.CamembertTokenizer,CausalLMOutput:()=>s.CausalLMOutput,CausalLMOutputWithPast:()=>s.CausalLMOutputWithPast,ChineseCLIPFeatureExtractor:()=>h.ChineseCLIPFeatureExtractor,ChineseCLIPModel:()=>s.ChineseCLIPModel,ChineseCLIPPreTrainedModel:()=>s.ChineseCLIPPreTrainedModel,ClapAudioModelWithProjection:()=>s.ClapAudioModelWithProjection,ClapFeatureExtractor:()=>_.ClapFeatureExtractor,ClapModel:()=>s.ClapModel,ClapPreTrainedModel:()=>s.ClapPreTrainedModel,ClapTextModelWithProjection:()=>s.ClapTextModelWithProjection,ClassifierFreeGuidanceLogitsProcessor:()=>y.ClassifierFreeGuidanceLogitsProcessor,CodeGenForCausalLM:()=>s.CodeGenForCausalLM,CodeGenModel:()=>s.CodeGenModel,CodeGenPreTrainedModel:()=>s.CodeGenPreTrainedModel,CodeGenTokenizer:()=>r.CodeGenTokenizer,CodeLlamaTokenizer:()=>r.CodeLlamaTokenizer,CohereForCausalLM:()=>s.CohereForCausalLM,CohereModel:()=>s.CohereModel,CoherePreTrainedModel:()=>s.CoherePreTrainedModel,CohereTokenizer:()=>r.CohereTokenizer,ConvBertForMaskedLM:()=>s.ConvBertForMaskedLM,ConvBertForQuestionAnswering:()=>s.ConvBertForQuestionAnswering,ConvBertForSequenceClassification:()=>s.ConvBertForSequenceClassification,ConvBertForTokenClassification:()=>s.ConvBertForTokenClassification,ConvBertModel:()=>s.ConvBertModel,ConvBertPreTrainedModel:()=>s.ConvBertPreTrainedModel,ConvBertTokenizer:()=>r.ConvBertTokenizer,ConvNextFeatureExtractor:()=>h.ConvNextFeatureExtractor,ConvNextForImageClassification:()=>s.ConvNextForImageClassification,ConvNextImageProcessor:()=>h.ConvNextImageProcessor,ConvNextModel:()=>s.ConvNextModel,ConvNextPreTrainedModel:()=>s.ConvNextPreTrainedModel,ConvNextV2ForImageClassification:()=>s.ConvNextV2ForImageClassification,ConvNextV2Model:()=>s.ConvNextV2Model,ConvNextV2PreTrainedModel:()=>s.ConvNextV2PreTrainedModel,DPTFeatureExtractor:()=>h.DPTFeatureExtractor,DPTForDepthEstimation:()=>s.DPTForDepthEstimation,DPTImageProcessor:()=>h.DPTImageProcessor,DPTModel:()=>s.DPTModel,DPTPreTrainedModel:()=>s.DPTPreTrainedModel,DebertaForMaskedLM:()=>s.DebertaForMaskedLM,DebertaForQuestionAnswering:()=>s.DebertaForQuestionAnswering,DebertaForSequenceClassification:()=>s.DebertaForSequenceClassification,DebertaForTokenClassification:()=>s.DebertaForTokenClassification,DebertaModel:()=>s.DebertaModel,DebertaPreTrainedModel:()=>s.DebertaPreTrainedModel,DebertaTokenizer:()=>r.DebertaTokenizer,DebertaV2ForMaskedLM:()=>s.DebertaV2ForMaskedLM,DebertaV2ForQuestionAnswering:()=>s.DebertaV2ForQuestionAnswering,DebertaV2ForSequenceClassification:()=>s.DebertaV2ForSequenceClassification,DebertaV2ForTokenClassification:()=>s.DebertaV2ForTokenClassification,DebertaV2Model:()=>s.DebertaV2Model,DebertaV2PreTrainedModel:()=>s.DebertaV2PreTrainedModel,DebertaV2Tokenizer:()=>r.DebertaV2Tokenizer,DecisionTransformerModel:()=>s.DecisionTransformerModel,DecisionTransformerPreTrainedModel:()=>s.DecisionTransformerPreTrainedModel,DeiTFeatureExtractor:()=>h.DeiTFeatureExtractor,DeiTForImageClassification:()=>s.DeiTForImageClassification,DeiTImageProcessor:()=>h.DeiTImageProcessor,DeiTModel:()=>s.DeiTModel,DeiTPreTrainedModel:()=>s.DeiTPreTrainedModel,DepthAnythingForDepthEstimation:()=>s.DepthAnythingForDepthEstimation,DepthAnythingPreTrainedModel:()=>s.DepthAnythingPreTrainedModel,DepthEstimationPipeline:()=>t.DepthEstimationPipeline,DepthProForDepthEstimation:()=>s.DepthProForDepthEstimation,DepthProPreTrainedModel:()=>s.DepthProPreTrainedModel,DetrFeatureExtractor:()=>h.DetrFeatureExtractor,DetrForObjectDetection:()=>s.DetrForObjectDetection,DetrForSegmentation:()=>s.DetrForSegmentation,DetrImageProcessor:()=>h.DetrImageProcessor,DetrModel:()=>s.DetrModel,DetrObjectDetectionOutput:()=>s.DetrObjectDetectionOutput,DetrPreTrainedModel:()=>s.DetrPreTrainedModel,DetrSegmentationOutput:()=>s.DetrSegmentationOutput,Dinov2ForImageClassification:()=>s.Dinov2ForImageClassification,Dinov2Model:()=>s.Dinov2Model,Dinov2PreTrainedModel:()=>s.Dinov2PreTrainedModel,Dinov2WithRegistersForImageClassification:()=>s.Dinov2WithRegistersForImageClassification,Dinov2WithRegistersModel:()=>s.Dinov2WithRegistersModel,Dinov2WithRegistersPreTrainedModel:()=>s.Dinov2WithRegistersPreTrainedModel,DistilBertForMaskedLM:()=>s.DistilBertForMaskedLM,DistilBertForQuestionAnswering:()=>s.DistilBertForQuestionAnswering,DistilBertForSequenceClassification:()=>s.DistilBertForSequenceClassification,DistilBertForTokenClassification:()=>s.DistilBertForTokenClassification,DistilBertModel:()=>s.DistilBertModel,DistilBertPreTrainedModel:()=>s.DistilBertPreTrainedModel,DistilBertTokenizer:()=>r.DistilBertTokenizer,DocumentQuestionAnsweringPipeline:()=>t.DocumentQuestionAnsweringPipeline,DonutFeatureExtractor:()=>h.DonutFeatureExtractor,DonutImageProcessor:()=>h.DonutImageProcessor,DonutSwinModel:()=>s.DonutSwinModel,DonutSwinPreTrainedModel:()=>s.DonutSwinPreTrainedModel,EfficientNetForImageClassification:()=>s.EfficientNetForImageClassification,EfficientNetImageProcessor:()=>h.EfficientNetImageProcessor,EfficientNetModel:()=>s.EfficientNetModel,EfficientNetPreTrainedModel:()=>s.EfficientNetPreTrainedModel,ElectraForMaskedLM:()=>s.ElectraForMaskedLM,ElectraForQuestionAnswering:()=>s.ElectraForQuestionAnswering,ElectraForSequenceClassification:()=>s.ElectraForSequenceClassification,ElectraForTokenClassification:()=>s.ElectraForTokenClassification,ElectraModel:()=>s.ElectraModel,ElectraPreTrainedModel:()=>s.ElectraPreTrainedModel,ElectraTokenizer:()=>r.ElectraTokenizer,EosTokenCriteria:()=>x.EosTokenCriteria,EsmForMaskedLM:()=>s.EsmForMaskedLM,EsmForSequenceClassification:()=>s.EsmForSequenceClassification,EsmForTokenClassification:()=>s.EsmForTokenClassification,EsmModel:()=>s.EsmModel,EsmPreTrainedModel:()=>s.EsmPreTrainedModel,EsmTokenizer:()=>r.EsmTokenizer,ExaoneForCausalLM:()=>s.ExaoneForCausalLM,ExaoneModel:()=>s.ExaoneModel,ExaonePreTrainedModel:()=>s.ExaonePreTrainedModel,FFT:()=>d.FFT,FalconForCausalLM:()=>s.FalconForCausalLM,FalconModel:()=>s.FalconModel,FalconPreTrainedModel:()=>s.FalconPreTrainedModel,FalconTokenizer:()=>r.FalconTokenizer,FastViTForImageClassification:()=>s.FastViTForImageClassification,FastViTModel:()=>s.FastViTModel,FastViTPreTrainedModel:()=>s.FastViTPreTrainedModel,FeatureExtractionPipeline:()=>t.FeatureExtractionPipeline,FeatureExtractor:()=>u.FeatureExtractor,FillMaskPipeline:()=>t.FillMaskPipeline,Florence2ForConditionalGeneration:()=>s.Florence2ForConditionalGeneration,Florence2PreTrainedModel:()=>s.Florence2PreTrainedModel,Florence2Processor:()=>w.Florence2Processor,ForcedBOSTokenLogitsProcessor:()=>y.ForcedBOSTokenLogitsProcessor,ForcedEOSTokenLogitsProcessor:()=>y.ForcedEOSTokenLogitsProcessor,GLPNFeatureExtractor:()=>h.GLPNFeatureExtractor,GLPNForDepthEstimation:()=>s.GLPNForDepthEstimation,GLPNModel:()=>s.GLPNModel,GLPNPreTrainedModel:()=>s.GLPNPreTrainedModel,GPT2LMHeadModel:()=>s.GPT2LMHeadModel,GPT2Model:()=>s.GPT2Model,GPT2PreTrainedModel:()=>s.GPT2PreTrainedModel,GPT2Tokenizer:()=>r.GPT2Tokenizer,GPTBigCodeForCausalLM:()=>s.GPTBigCodeForCausalLM,GPTBigCodeModel:()=>s.GPTBigCodeModel,GPTBigCodePreTrainedModel:()=>s.GPTBigCodePreTrainedModel,GPTJForCausalLM:()=>s.GPTJForCausalLM,GPTJModel:()=>s.GPTJModel,GPTJPreTrainedModel:()=>s.GPTJPreTrainedModel,GPTNeoForCausalLM:()=>s.GPTNeoForCausalLM,GPTNeoModel:()=>s.GPTNeoModel,GPTNeoPreTrainedModel:()=>s.GPTNeoPreTrainedModel,GPTNeoXForCausalLM:()=>s.GPTNeoXForCausalLM,GPTNeoXModel:()=>s.GPTNeoXModel,GPTNeoXPreTrainedModel:()=>s.GPTNeoXPreTrainedModel,GPTNeoXTokenizer:()=>r.GPTNeoXTokenizer,Gemma2ForCausalLM:()=>s.Gemma2ForCausalLM,Gemma2Model:()=>s.Gemma2Model,Gemma2PreTrainedModel:()=>s.Gemma2PreTrainedModel,GemmaForCausalLM:()=>s.GemmaForCausalLM,GemmaModel:()=>s.GemmaModel,GemmaPreTrainedModel:()=>s.GemmaPreTrainedModel,GemmaTokenizer:()=>r.GemmaTokenizer,GlmForCausalLM:()=>s.GlmForCausalLM,GlmModel:()=>s.GlmModel,GlmPreTrainedModel:()=>s.GlmPreTrainedModel,GraniteForCausalLM:()=>s.GraniteForCausalLM,GraniteModel:()=>s.GraniteModel,GranitePreTrainedModel:()=>s.GranitePreTrainedModel,Grok1Tokenizer:()=>r.Grok1Tokenizer,GroundingDinoForObjectDetection:()=>s.GroundingDinoForObjectDetection,GroundingDinoImageProcessor:()=>h.GroundingDinoImageProcessor,GroundingDinoPreTrainedModel:()=>s.GroundingDinoPreTrainedModel,GroundingDinoProcessor:()=>w.GroundingDinoProcessor,GroupViTModel:()=>s.GroupViTModel,GroupViTPreTrainedModel:()=>s.GroupViTPreTrainedModel,HeliumForCausalLM:()=>s.HeliumForCausalLM,HeliumModel:()=>s.HeliumModel,HeliumPreTrainedModel:()=>s.HeliumPreTrainedModel,HerbertTokenizer:()=>r.HerbertTokenizer,HieraForImageClassification:()=>s.HieraForImageClassification,HieraModel:()=>s.HieraModel,HieraPreTrainedModel:()=>s.HieraPreTrainedModel,HubertForCTC:()=>s.HubertForCTC,HubertForSequenceClassification:()=>s.HubertForSequenceClassification,HubertModel:()=>s.HubertModel,HubertPreTrainedModel:()=>s.HubertPreTrainedModel,IJepaForImageClassification:()=>s.IJepaForImageClassification,IJepaModel:()=>s.IJepaModel,IJepaPreTrainedModel:()=>s.IJepaPreTrainedModel,Idefics3ForConditionalGeneration:()=>s.Idefics3ForConditionalGeneration,Idefics3ImageProcessor:()=>h.Idefics3ImageProcessor,Idefics3PreTrainedModel:()=>s.Idefics3PreTrainedModel,Idefics3Processor:()=>w.Idefics3Processor,ImageClassificationPipeline:()=>t.ImageClassificationPipeline,ImageFeatureExtractionPipeline:()=>t.ImageFeatureExtractionPipeline,ImageFeatureExtractor:()=>_.ImageFeatureExtractor,ImageMattingOutput:()=>s.ImageMattingOutput,ImageProcessor:()=>m.ImageProcessor,ImageSegmentationPipeline:()=>t.ImageSegmentationPipeline,ImageToImagePipeline:()=>t.ImageToImagePipeline,ImageToTextPipeline:()=>t.ImageToTextPipeline,InterruptableStoppingCriteria:()=>x.InterruptableStoppingCriteria,JAISLMHeadModel:()=>s.JAISLMHeadModel,JAISModel:()=>s.JAISModel,JAISPreTrainedModel:()=>s.JAISPreTrainedModel,JinaCLIPImageProcessor:()=>h.JinaCLIPImageProcessor,JinaCLIPModel:()=>s.JinaCLIPModel,JinaCLIPPreTrainedModel:()=>s.JinaCLIPPreTrainedModel,JinaCLIPProcessor:()=>w.JinaCLIPProcessor,JinaCLIPTextModel:()=>s.JinaCLIPTextModel,JinaCLIPVisionModel:()=>s.JinaCLIPVisionModel,LlamaForCausalLM:()=>s.LlamaForCausalLM,LlamaModel:()=>s.LlamaModel,LlamaPreTrainedModel:()=>s.LlamaPreTrainedModel,LlamaTokenizer:()=>r.LlamaTokenizer,LlavaForConditionalGeneration:()=>s.LlavaForConditionalGeneration,LlavaOnevisionForConditionalGeneration:()=>s.LlavaOnevisionForConditionalGeneration,LlavaOnevisionImageProcessor:()=>h.LlavaOnevisionImageProcessor,LlavaPreTrainedModel:()=>s.LlavaPreTrainedModel,LogitsProcessor:()=>y.LogitsProcessor,LogitsProcessorList:()=>y.LogitsProcessorList,LogitsWarper:()=>y.LogitsWarper,LongT5ForConditionalGeneration:()=>s.LongT5ForConditionalGeneration,LongT5Model:()=>s.LongT5Model,LongT5PreTrainedModel:()=>s.LongT5PreTrainedModel,M2M100ForConditionalGeneration:()=>s.M2M100ForConditionalGeneration,M2M100Model:()=>s.M2M100Model,M2M100PreTrainedModel:()=>s.M2M100PreTrainedModel,M2M100Tokenizer:()=>r.M2M100Tokenizer,MBart50Tokenizer:()=>r.MBart50Tokenizer,MBartForCausalLM:()=>s.MBartForCausalLM,MBartForConditionalGeneration:()=>s.MBartForConditionalGeneration,MBartForSequenceClassification:()=>s.MBartForSequenceClassification,MBartModel:()=>s.MBartModel,MBartPreTrainedModel:()=>s.MBartPreTrainedModel,MBartTokenizer:()=>r.MBartTokenizer,MPNetForMaskedLM:()=>s.MPNetForMaskedLM,MPNetForQuestionAnswering:()=>s.MPNetForQuestionAnswering,MPNetForSequenceClassification:()=>s.MPNetForSequenceClassification,MPNetForTokenClassification:()=>s.MPNetForTokenClassification,MPNetModel:()=>s.MPNetModel,MPNetPreTrainedModel:()=>s.MPNetPreTrainedModel,MPNetTokenizer:()=>r.MPNetTokenizer,MT5ForConditionalGeneration:()=>s.MT5ForConditionalGeneration,MT5Model:()=>s.MT5Model,MT5PreTrainedModel:()=>s.MT5PreTrainedModel,MarianMTModel:()=>s.MarianMTModel,MarianModel:()=>s.MarianModel,MarianPreTrainedModel:()=>s.MarianPreTrainedModel,MarianTokenizer:()=>r.MarianTokenizer,Mask2FormerImageProcessor:()=>h.Mask2FormerImageProcessor,MaskFormerFeatureExtractor:()=>h.MaskFormerFeatureExtractor,MaskFormerForInstanceSegmentation:()=>s.MaskFormerForInstanceSegmentation,MaskFormerImageProcessor:()=>h.MaskFormerImageProcessor,MaskFormerModel:()=>s.MaskFormerModel,MaskFormerPreTrainedModel:()=>s.MaskFormerPreTrainedModel,MaskedLMOutput:()=>s.MaskedLMOutput,MaxLengthCriteria:()=>x.MaxLengthCriteria,MgpstrForSceneTextRecognition:()=>s.MgpstrForSceneTextRecognition,MgpstrModelOutput:()=>s.MgpstrModelOutput,MgpstrPreTrainedModel:()=>s.MgpstrPreTrainedModel,MgpstrProcessor:()=>w.MgpstrProcessor,MgpstrTokenizer:()=>r.MgpstrTokenizer,MinLengthLogitsProcessor:()=>y.MinLengthLogitsProcessor,MinNewTokensLengthLogitsProcessor:()=>y.MinNewTokensLengthLogitsProcessor,MistralForCausalLM:()=>s.MistralForCausalLM,MistralModel:()=>s.MistralModel,MistralPreTrainedModel:()=>s.MistralPreTrainedModel,MobileBertForMaskedLM:()=>s.MobileBertForMaskedLM,MobileBertForQuestionAnswering:()=>s.MobileBertForQuestionAnswering,MobileBertForSequenceClassification:()=>s.MobileBertForSequenceClassification,MobileBertModel:()=>s.MobileBertModel,MobileBertPreTrainedModel:()=>s.MobileBertPreTrainedModel,MobileBertTokenizer:()=>r.MobileBertTokenizer,MobileLLMForCausalLM:()=>s.MobileLLMForCausalLM,MobileLLMModel:()=>s.MobileLLMModel,MobileLLMPreTrainedModel:()=>s.MobileLLMPreTrainedModel,MobileNetV1FeatureExtractor:()=>h.MobileNetV1FeatureExtractor,MobileNetV1ForImageClassification:()=>s.MobileNetV1ForImageClassification,MobileNetV1ImageProcessor:()=>h.MobileNetV1ImageProcessor,MobileNetV1Model:()=>s.MobileNetV1Model,MobileNetV1PreTrainedModel:()=>s.MobileNetV1PreTrainedModel,MobileNetV2FeatureExtractor:()=>h.MobileNetV2FeatureExtractor,MobileNetV2ForImageClassification:()=>s.MobileNetV2ForImageClassification,MobileNetV2ImageProcessor:()=>h.MobileNetV2ImageProcessor,MobileNetV2Model:()=>s.MobileNetV2Model,MobileNetV2PreTrainedModel:()=>s.MobileNetV2PreTrainedModel,MobileNetV3FeatureExtractor:()=>h.MobileNetV3FeatureExtractor,MobileNetV3ForImageClassification:()=>s.MobileNetV3ForImageClassification,MobileNetV3ImageProcessor:()=>h.MobileNetV3ImageProcessor,MobileNetV3Model:()=>s.MobileNetV3Model,MobileNetV3PreTrainedModel:()=>s.MobileNetV3PreTrainedModel,MobileNetV4FeatureExtractor:()=>h.MobileNetV4FeatureExtractor,MobileNetV4ForImageClassification:()=>s.MobileNetV4ForImageClassification,MobileNetV4ImageProcessor:()=>h.MobileNetV4ImageProcessor,MobileNetV4Model:()=>s.MobileNetV4Model,MobileNetV4PreTrainedModel:()=>s.MobileNetV4PreTrainedModel,MobileViTFeatureExtractor:()=>h.MobileViTFeatureExtractor,MobileViTForImageClassification:()=>s.MobileViTForImageClassification,MobileViTImageProcessor:()=>h.MobileViTImageProcessor,MobileViTModel:()=>s.MobileViTModel,MobileViTPreTrainedModel:()=>s.MobileViTPreTrainedModel,MobileViTV2ForImageClassification:()=>s.MobileViTV2ForImageClassification,MobileViTV2Model:()=>s.MobileViTV2Model,MobileViTV2PreTrainedModel:()=>s.MobileViTV2PreTrainedModel,ModelOutput:()=>s.ModelOutput,ModernBertForMaskedLM:()=>s.ModernBertForMaskedLM,ModernBertForSequenceClassification:()=>s.ModernBertForSequenceClassification,ModernBertForTokenClassification:()=>s.ModernBertForTokenClassification,ModernBertModel:()=>s.ModernBertModel,ModernBertPreTrainedModel:()=>s.ModernBertPreTrainedModel,Moondream1ForConditionalGeneration:()=>s.Moondream1ForConditionalGeneration,MoonshineFeatureExtractor:()=>_.MoonshineFeatureExtractor,MoonshineForConditionalGeneration:()=>s.MoonshineForConditionalGeneration,MoonshineModel:()=>s.MoonshineModel,MoonshinePreTrainedModel:()=>s.MoonshinePreTrainedModel,MoonshineProcessor:()=>w.MoonshineProcessor,MptForCausalLM:()=>s.MptForCausalLM,MptModel:()=>s.MptModel,MptPreTrainedModel:()=>s.MptPreTrainedModel,MultiModalityCausalLM:()=>s.MultiModalityCausalLM,MultiModalityPreTrainedModel:()=>s.MultiModalityPreTrainedModel,MusicgenForCausalLM:()=>s.MusicgenForCausalLM,MusicgenForConditionalGeneration:()=>s.MusicgenForConditionalGeneration,MusicgenModel:()=>s.MusicgenModel,MusicgenPreTrainedModel:()=>s.MusicgenPreTrainedModel,NllbTokenizer:()=>r.NllbTokenizer,NoBadWordsLogitsProcessor:()=>y.NoBadWordsLogitsProcessor,NoRepeatNGramLogitsProcessor:()=>y.NoRepeatNGramLogitsProcessor,NomicBertModel:()=>s.NomicBertModel,NomicBertPreTrainedModel:()=>s.NomicBertPreTrainedModel,NougatImageProcessor:()=>h.NougatImageProcessor,NougatTokenizer:()=>r.NougatTokenizer,OPTForCausalLM:()=>s.OPTForCausalLM,OPTModel:()=>s.OPTModel,OPTPreTrainedModel:()=>s.OPTPreTrainedModel,ObjectDetectionPipeline:()=>t.ObjectDetectionPipeline,Olmo2ForCausalLM:()=>s.Olmo2ForCausalLM,Olmo2Model:()=>s.Olmo2Model,Olmo2PreTrainedModel:()=>s.Olmo2PreTrainedModel,OlmoForCausalLM:()=>s.OlmoForCausalLM,OlmoModel:()=>s.OlmoModel,OlmoPreTrainedModel:()=>s.OlmoPreTrainedModel,OpenELMForCausalLM:()=>s.OpenELMForCausalLM,OpenELMModel:()=>s.OpenELMModel,OpenELMPreTrainedModel:()=>s.OpenELMPreTrainedModel,OwlViTFeatureExtractor:()=>h.OwlViTFeatureExtractor,OwlViTForObjectDetection:()=>s.OwlViTForObjectDetection,OwlViTImageProcessor:()=>h.OwlViTImageProcessor,OwlViTModel:()=>s.OwlViTModel,OwlViTPreTrainedModel:()=>s.OwlViTPreTrainedModel,OwlViTProcessor:()=>w.OwlViTProcessor,Owlv2ForObjectDetection:()=>s.Owlv2ForObjectDetection,Owlv2ImageProcessor:()=>h.Owlv2ImageProcessor,Owlv2Model:()=>s.Owlv2Model,Owlv2PreTrainedModel:()=>s.Owlv2PreTrainedModel,PaliGemmaForConditionalGeneration:()=>s.PaliGemmaForConditionalGeneration,PaliGemmaPreTrainedModel:()=>s.PaliGemmaPreTrainedModel,PaliGemmaProcessor:()=>w.PaliGemmaProcessor,PatchTSMixerForPrediction:()=>s.PatchTSMixerForPrediction,PatchTSMixerModel:()=>s.PatchTSMixerModel,PatchTSMixerPreTrainedModel:()=>s.PatchTSMixerPreTrainedModel,PatchTSTForPrediction:()=>s.PatchTSTForPrediction,PatchTSTModel:()=>s.PatchTSTModel,PatchTSTPreTrainedModel:()=>s.PatchTSTPreTrainedModel,Phi3ForCausalLM:()=>s.Phi3ForCausalLM,Phi3Model:()=>s.Phi3Model,Phi3PreTrainedModel:()=>s.Phi3PreTrainedModel,Phi3VForCausalLM:()=>s.Phi3VForCausalLM,Phi3VImageProcessor:()=>h.Phi3VImageProcessor,Phi3VPreTrainedModel:()=>s.Phi3VPreTrainedModel,Phi3VProcessor:()=>w.Phi3VProcessor,PhiForCausalLM:()=>s.PhiForCausalLM,PhiModel:()=>s.PhiModel,PhiPreTrainedModel:()=>s.PhiPreTrainedModel,Pipeline:()=>t.Pipeline,PreTrainedModel:()=>s.PreTrainedModel,PreTrainedTokenizer:()=>r.PreTrainedTokenizer,PretrainedConfig:()=>i.PretrainedConfig,PretrainedMixin:()=>s.PretrainedMixin,Processor:()=>g.Processor,PvtForImageClassification:()=>s.PvtForImageClassification,PvtImageProcessor:()=>h.PvtImageProcessor,PvtModel:()=>s.PvtModel,PvtPreTrainedModel:()=>s.PvtPreTrainedModel,PyAnnoteFeatureExtractor:()=>_.PyAnnoteFeatureExtractor,PyAnnoteForAudioFrameClassification:()=>s.PyAnnoteForAudioFrameClassification,PyAnnoteModel:()=>s.PyAnnoteModel,PyAnnotePreTrainedModel:()=>s.PyAnnotePreTrainedModel,PyAnnoteProcessor:()=>w.PyAnnoteProcessor,QuestionAnsweringModelOutput:()=>s.QuestionAnsweringModelOutput,QuestionAnsweringPipeline:()=>t.QuestionAnsweringPipeline,Qwen2ForCausalLM:()=>s.Qwen2ForCausalLM,Qwen2Model:()=>s.Qwen2Model,Qwen2PreTrainedModel:()=>s.Qwen2PreTrainedModel,Qwen2Tokenizer:()=>r.Qwen2Tokenizer,Qwen2VLForConditionalGeneration:()=>s.Qwen2VLForConditionalGeneration,Qwen2VLImageProcessor:()=>h.Qwen2VLImageProcessor,Qwen2VLPreTrainedModel:()=>s.Qwen2VLPreTrainedModel,Qwen2VLProcessor:()=>w.Qwen2VLProcessor,RTDetrForObjectDetection:()=>s.RTDetrForObjectDetection,RTDetrImageProcessor:()=>h.RTDetrImageProcessor,RTDetrModel:()=>s.RTDetrModel,RTDetrObjectDetectionOutput:()=>s.RTDetrObjectDetectionOutput,RTDetrPreTrainedModel:()=>s.RTDetrPreTrainedModel,RawAudio:()=>a.RawAudio,RawImage:()=>l.RawImage,RepetitionPenaltyLogitsProcessor:()=>y.RepetitionPenaltyLogitsProcessor,ResNetForImageClassification:()=>s.ResNetForImageClassification,ResNetModel:()=>s.ResNetModel,ResNetPreTrainedModel:()=>s.ResNetPreTrainedModel,RoFormerForMaskedLM:()=>s.RoFormerForMaskedLM,RoFormerForQuestionAnswering:()=>s.RoFormerForQuestionAnswering,RoFormerForSequenceClassification:()=>s.RoFormerForSequenceClassification,RoFormerForTokenClassification:()=>s.RoFormerForTokenClassification,RoFormerModel:()=>s.RoFormerModel,RoFormerPreTrainedModel:()=>s.RoFormerPreTrainedModel,RoFormerTokenizer:()=>r.RoFormerTokenizer,RobertaForMaskedLM:()=>s.RobertaForMaskedLM,RobertaForQuestionAnswering:()=>s.RobertaForQuestionAnswering,RobertaForSequenceClassification:()=>s.RobertaForSequenceClassification,RobertaForTokenClassification:()=>s.RobertaForTokenClassification,RobertaModel:()=>s.RobertaModel,RobertaPreTrainedModel:()=>s.RobertaPreTrainedModel,RobertaTokenizer:()=>r.RobertaTokenizer,SamImageProcessor:()=>h.SamImageProcessor,SamImageSegmentationOutput:()=>s.SamImageSegmentationOutput,SamModel:()=>s.SamModel,SamPreTrainedModel:()=>s.SamPreTrainedModel,SamProcessor:()=>w.SamProcessor,SapiensForDepthEstimation:()=>s.SapiensForDepthEstimation,SapiensForNormalEstimation:()=>s.SapiensForNormalEstimation,SapiensForSemanticSegmentation:()=>s.SapiensForSemanticSegmentation,SapiensPreTrainedModel:()=>s.SapiensPreTrainedModel,SeamlessM4TFeatureExtractor:()=>_.SeamlessM4TFeatureExtractor,SegformerFeatureExtractor:()=>h.SegformerFeatureExtractor,SegformerForImageClassification:()=>s.SegformerForImageClassification,SegformerForSemanticSegmentation:()=>s.SegformerForSemanticSegmentation,SegformerImageProcessor:()=>h.SegformerImageProcessor,SegformerModel:()=>s.SegformerModel,SegformerPreTrainedModel:()=>s.SegformerPreTrainedModel,Seq2SeqLMOutput:()=>s.Seq2SeqLMOutput,SequenceClassifierOutput:()=>s.SequenceClassifierOutput,SiglipImageProcessor:()=>h.SiglipImageProcessor,SiglipModel:()=>s.SiglipModel,SiglipPreTrainedModel:()=>s.SiglipPreTrainedModel,SiglipTextModel:()=>s.SiglipTextModel,SiglipTokenizer:()=>r.SiglipTokenizer,SiglipVisionModel:()=>s.SiglipVisionModel,SpeechT5FeatureExtractor:()=>_.SpeechT5FeatureExtractor,SpeechT5ForSpeechToText:()=>s.SpeechT5ForSpeechToText,SpeechT5ForTextToSpeech:()=>s.SpeechT5ForTextToSpeech,SpeechT5HifiGan:()=>s.SpeechT5HifiGan,SpeechT5Model:()=>s.SpeechT5Model,SpeechT5PreTrainedModel:()=>s.SpeechT5PreTrainedModel,SpeechT5Processor:()=>w.SpeechT5Processor,SpeechT5Tokenizer:()=>r.SpeechT5Tokenizer,SqueezeBertForMaskedLM:()=>s.SqueezeBertForMaskedLM,SqueezeBertForQuestionAnswering:()=>s.SqueezeBertForQuestionAnswering,SqueezeBertForSequenceClassification:()=>s.SqueezeBertForSequenceClassification,SqueezeBertModel:()=>s.SqueezeBertModel,SqueezeBertPreTrainedModel:()=>s.SqueezeBertPreTrainedModel,SqueezeBertTokenizer:()=>r.SqueezeBertTokenizer,StableLmForCausalLM:()=>s.StableLmForCausalLM,StableLmModel:()=>s.StableLmModel,StableLmPreTrainedModel:()=>s.StableLmPreTrainedModel,Starcoder2ForCausalLM:()=>s.Starcoder2ForCausalLM,Starcoder2Model:()=>s.Starcoder2Model,Starcoder2PreTrainedModel:()=>s.Starcoder2PreTrainedModel,StoppingCriteria:()=>x.StoppingCriteria,StoppingCriteriaList:()=>x.StoppingCriteriaList,StyleTextToSpeech2Model:()=>s.StyleTextToSpeech2Model,StyleTextToSpeech2PreTrainedModel:()=>s.StyleTextToSpeech2PreTrainedModel,SummarizationPipeline:()=>t.SummarizationPipeline,SuppressTokensAtBeginLogitsProcessor:()=>y.SuppressTokensAtBeginLogitsProcessor,Swin2SRForImageSuperResolution:()=>s.Swin2SRForImageSuperResolution,Swin2SRImageProcessor:()=>h.Swin2SRImageProcessor,Swin2SRModel:()=>s.Swin2SRModel,Swin2SRPreTrainedModel:()=>s.Swin2SRPreTrainedModel,SwinForImageClassification:()=>s.SwinForImageClassification,SwinModel:()=>s.SwinModel,SwinPreTrainedModel:()=>s.SwinPreTrainedModel,T5ForConditionalGeneration:()=>s.T5ForConditionalGeneration,T5Model:()=>s.T5Model,T5PreTrainedModel:()=>s.T5PreTrainedModel,T5Tokenizer:()=>r.T5Tokenizer,TableTransformerForObjectDetection:()=>s.TableTransformerForObjectDetection,TableTransformerModel:()=>s.TableTransformerModel,TableTransformerObjectDetectionOutput:()=>s.TableTransformerObjectDetectionOutput,TableTransformerPreTrainedModel:()=>s.TableTransformerPreTrainedModel,TemperatureLogitsWarper:()=>y.TemperatureLogitsWarper,Tensor:()=>c.Tensor,Text2TextGenerationPipeline:()=>t.Text2TextGenerationPipeline,TextClassificationPipeline:()=>t.TextClassificationPipeline,TextGenerationPipeline:()=>t.TextGenerationPipeline,TextStreamer:()=>b.TextStreamer,TextToAudioPipeline:()=>t.TextToAudioPipeline,TokenClassificationPipeline:()=>t.TokenClassificationPipeline,TokenClassifierOutput:()=>s.TokenClassifierOutput,TokenizerModel:()=>r.TokenizerModel,TopKLogitsWarper:()=>y.TopKLogitsWarper,TopPLogitsWarper:()=>y.TopPLogitsWarper,TrOCRForCausalLM:()=>s.TrOCRForCausalLM,TrOCRPreTrainedModel:()=>s.TrOCRPreTrainedModel,TranslationPipeline:()=>t.TranslationPipeline,UniSpeechForCTC:()=>s.UniSpeechForCTC,UniSpeechForSequenceClassification:()=>s.UniSpeechForSequenceClassification,UniSpeechModel:()=>s.UniSpeechModel,UniSpeechPreTrainedModel:()=>s.UniSpeechPreTrainedModel,UniSpeechSatForAudioFrameClassification:()=>s.UniSpeechSatForAudioFrameClassification,UniSpeechSatForCTC:()=>s.UniSpeechSatForCTC,UniSpeechSatForSequenceClassification:()=>s.UniSpeechSatForSequenceClassification,UniSpeechSatModel:()=>s.UniSpeechSatModel,UniSpeechSatPreTrainedModel:()=>s.UniSpeechSatPreTrainedModel,VLChatProcessor:()=>w.VLChatProcessor,VLMImageProcessor:()=>h.VLMImageProcessor,ViTFeatureExtractor:()=>h.ViTFeatureExtractor,ViTForImageClassification:()=>s.ViTForImageClassification,ViTImageProcessor:()=>h.ViTImageProcessor,ViTMAEModel:()=>s.ViTMAEModel,ViTMAEPreTrainedModel:()=>s.ViTMAEPreTrainedModel,ViTMSNForImageClassification:()=>s.ViTMSNForImageClassification,ViTMSNModel:()=>s.ViTMSNModel,ViTMSNPreTrainedModel:()=>s.ViTMSNPreTrainedModel,ViTModel:()=>s.ViTModel,ViTPreTrainedModel:()=>s.ViTPreTrainedModel,VisionEncoderDecoderModel:()=>s.VisionEncoderDecoderModel,VitMatteForImageMatting:()=>s.VitMatteForImageMatting,VitMatteImageProcessor:()=>h.VitMatteImageProcessor,VitMattePreTrainedModel:()=>s.VitMattePreTrainedModel,VitPoseForPoseEstimation:()=>s.VitPoseForPoseEstimation,VitPoseImageProcessor:()=>h.VitPoseImageProcessor,VitPosePreTrainedModel:()=>s.VitPosePreTrainedModel,VitsModel:()=>s.VitsModel,VitsModelOutput:()=>s.VitsModelOutput,VitsPreTrainedModel:()=>s.VitsPreTrainedModel,VitsTokenizer:()=>r.VitsTokenizer,Wav2Vec2BertForCTC:()=>s.Wav2Vec2BertForCTC,Wav2Vec2BertForSequenceClassification:()=>s.Wav2Vec2BertForSequenceClassification,Wav2Vec2BertModel:()=>s.Wav2Vec2BertModel,Wav2Vec2BertPreTrainedModel:()=>s.Wav2Vec2BertPreTrainedModel,Wav2Vec2CTCTokenizer:()=>r.Wav2Vec2CTCTokenizer,Wav2Vec2FeatureExtractor:()=>_.Wav2Vec2FeatureExtractor,Wav2Vec2ForAudioFrameClassification:()=>s.Wav2Vec2ForAudioFrameClassification,Wav2Vec2ForCTC:()=>s.Wav2Vec2ForCTC,Wav2Vec2ForSequenceClassification:()=>s.Wav2Vec2ForSequenceClassification,Wav2Vec2Model:()=>s.Wav2Vec2Model,Wav2Vec2PreTrainedModel:()=>s.Wav2Vec2PreTrainedModel,Wav2Vec2Processor:()=>w.Wav2Vec2Processor,Wav2Vec2ProcessorWithLM:()=>w.Wav2Vec2ProcessorWithLM,WavLMForAudioFrameClassification:()=>s.WavLMForAudioFrameClassification,WavLMForCTC:()=>s.WavLMForCTC,WavLMForSequenceClassification:()=>s.WavLMForSequenceClassification,WavLMForXVector:()=>s.WavLMForXVector,WavLMModel:()=>s.WavLMModel,WavLMPreTrainedModel:()=>s.WavLMPreTrainedModel,WeSpeakerFeatureExtractor:()=>_.WeSpeakerFeatureExtractor,WeSpeakerResNetModel:()=>s.WeSpeakerResNetModel,WeSpeakerResNetPreTrainedModel:()=>s.WeSpeakerResNetPreTrainedModel,WhisperFeatureExtractor:()=>_.WhisperFeatureExtractor,WhisperForConditionalGeneration:()=>s.WhisperForConditionalGeneration,WhisperModel:()=>s.WhisperModel,WhisperPreTrainedModel:()=>s.WhisperPreTrainedModel,WhisperProcessor:()=>w.WhisperProcessor,WhisperTextStreamer:()=>b.WhisperTextStreamer,WhisperTimeStampLogitsProcessor:()=>y.WhisperTimeStampLogitsProcessor,WhisperTokenizer:()=>r.WhisperTokenizer,XLMForQuestionAnswering:()=>s.XLMForQuestionAnswering,XLMForSequenceClassification:()=>s.XLMForSequenceClassification,XLMForTokenClassification:()=>s.XLMForTokenClassification,XLMModel:()=>s.XLMModel,XLMPreTrainedModel:()=>s.XLMPreTrainedModel,XLMRobertaForMaskedLM:()=>s.XLMRobertaForMaskedLM,XLMRobertaForQuestionAnswering:()=>s.XLMRobertaForQuestionAnswering,XLMRobertaForSequenceClassification:()=>s.XLMRobertaForSequenceClassification,XLMRobertaForTokenClassification:()=>s.XLMRobertaForTokenClassification,XLMRobertaModel:()=>s.XLMRobertaModel,XLMRobertaPreTrainedModel:()=>s.XLMRobertaPreTrainedModel,XLMRobertaTokenizer:()=>r.XLMRobertaTokenizer,XLMTokenizer:()=>r.XLMTokenizer,XLMWithLMHeadModel:()=>s.XLMWithLMHeadModel,XVectorOutput:()=>s.XVectorOutput,YolosFeatureExtractor:()=>h.YolosFeatureExtractor,YolosForObjectDetection:()=>s.YolosForObjectDetection,YolosImageProcessor:()=>h.YolosImageProcessor,YolosModel:()=>s.YolosModel,YolosObjectDetectionOutput:()=>s.YolosObjectDetectionOutput,YolosPreTrainedModel:()=>s.YolosPreTrainedModel,ZeroShotAudioClassificationPipeline:()=>t.ZeroShotAudioClassificationPipeline,ZeroShotClassificationPipeline:()=>t.ZeroShotClassificationPipeline,ZeroShotImageClassificationPipeline:()=>t.ZeroShotImageClassificationPipeline,ZeroShotObjectDetectionPipeline:()=>t.ZeroShotObjectDetectionPipeline,bankers_round:()=>d.bankers_round,cat:()=>c.cat,cos_sim:()=>d.cos_sim,dot:()=>d.dot,dynamic_time_warping:()=>d.dynamic_time_warping,env:()=>e.env,full:()=>c.full,full_like:()=>c.full_like,getKeyValueShapes:()=>i.getKeyValueShapes,hamming:()=>a.hamming,hanning:()=>a.hanning,interpolate:()=>c.interpolate,interpolate_4d:()=>c.interpolate_4d,interpolate_data:()=>d.interpolate_data,is_chinese_char:()=>r.is_chinese_char,layer_norm:()=>c.layer_norm,load_image:()=>l.load_image,log_softmax:()=>d.log_softmax,magnitude:()=>d.magnitude,matmul:()=>c.matmul,max:()=>d.max,mean:()=>c.mean,mean_pooling:()=>c.mean_pooling,medianFilter:()=>d.medianFilter,mel_filter_bank:()=>a.mel_filter_bank,min:()=>d.min,ones:()=>c.ones,ones_like:()=>c.ones_like,permute:()=>c.permute,permute_data:()=>d.permute_data,pipeline:()=>t.pipeline,quantize_embeddings:()=>c.quantize_embeddings,rand:()=>c.rand,read_audio:()=>a.read_audio,rfft:()=>c.rfft,round:()=>d.round,slice:()=>c.slice,softmax:()=>d.softmax,spectrogram:()=>a.spectrogram,stack:()=>c.stack,std_mean:()=>c.std_mean,topk:()=>c.topk,window_function:()=>a.window_function,zeros:()=>c.zeros,zeros_like:()=>c.zeros_like});var e=n("./src/env.js"),t=n("./src/pipelines.js"),s=n("./src/models.js"),r=n("./src/tokenizers.js"),i=n("./src/configs.js"),a=n("./src/utils/audio.js"),l=n("./src/utils/image.js"),c=n("./src/utils/tensor.js"),d=n("./src/utils/maths.js"),u=n("./src/base/feature_extraction_utils.js"),_=n("./src/models/feature_extractors.js"),p=n("./src/models/auto/feature_extraction_auto.js"),m=n("./src/base/image_processors_utils.js"),h=n("./src/models/image_processors.js"),f=n("./src/models/auto/image_processing_auto.js"),g=n("./src/base/processing_utils.js"),w=n("./src/models/processors.js"),M=n("./src/models/auto/processing_auto.js"),b=n("./src/generation/streamers.js"),x=n("./src/generation/stopping_criteria.js"),y=n("./src/generation/logits_process.js")})();var i=exports;for(var a in o)i[a]=o[a];o.__esModule&&Object.defineProperty(i,"__esModule",{value:!0})})();
-//# sourceMappingURL=transformers.min.cjs.map