@huggingface/transformers 3.0.0 → 3.0.1

package/dist/transformers.min.mjs

@@ -19,7 +19,7 @@ import*as e from"fs";import*as t from"onnxruntime-node";import*as r from"path";i
   \*************************************/()=>{},"./node_modules/@huggingface/jinja/dist/index.js":
 /*!*******************************************************!*\
   !*** ./node_modules/@huggingface/jinja/dist/index.js ***!
-  \*******************************************************/(e,t,r)=>{r.r(t),r.d(t,{Environment:()=>Y,Interpreter:()=>J,Template:()=>ee,parse:()=>B,tokenize:()=>d});var n=Object.freeze({Text:"Text",NumericLiteral:"NumericLiteral",BooleanLiteral:"BooleanLiteral",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"}),o=Object.freeze({set:n.Set,for:n.For,in:n.In,is:n.Is,if:n.If,else:n.Else,endif:n.EndIf,elif:n.ElseIf,endfor:n.EndFor,and:n.And,or:n.Or,not:n.Not,"not in":n.NotIn,macro:n.Macro,endmacro:n.EndMacro,true:n.BooleanLiteral,false:n.BooleanLiteral,True:n.BooleanLiteral,False:n.BooleanLiteral}),s=class{constructor(e,t){this.value=e,this.type=t}};function a(e){return/\w/.test(e)}function i(e){return/[0-9]/.test(e)}var l=[["{%",n.OpenStatement],["%}",n.CloseStatement],["{{",n.OpenExpression],["}}",n.CloseExpression],["(",n.OpenParen],[")",n.CloseParen],["{",n.OpenCurlyBracket],["}",n.CloseCurlyBracket],["[",n.OpenSquareBracket],["]",n.CloseSquareBracket],[",",n.Comma],[".",n.Dot],[":",n.Colon],["|",n.Pipe],["<=",n.ComparisonBinaryOperator],[">=",n.ComparisonBinaryOperator],["==",n.ComparisonBinaryOperator],["!=",n.ComparisonBinaryOperator],["<",n.ComparisonBinaryOperator],[">",n.ComparisonBinaryOperator],["+",n.AdditiveBinaryOperator],["-",n.AdditiveBinaryOperator],["*",n.MultiplicativeBinaryOperator],["/",n.MultiplicativeBinaryOperator],["%",n.MultiplicativeBinaryOperator],["=",n.Equals]],c=new Map([["n","\n"],["t","\t"],["r","\r"],["b","\b"],["f","\f"],["v","\v"],["'","'"],['"','"'],["\\","\\"]]);function d(e,t={}){const r=[],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 p=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++],r=c.get(e);if(void 0===r)throw new SyntaxError(`Unexpected escaped character: ${e}`);t+=r}return t};e:for(;u<d.length;){const e=r.at(-1)?.type;if(void 0===e||e===n.CloseStatement||e===n.CloseExpression){let e="";for(;u<d.length&&("{"!==d[u]||"%"!==d[u+1]&&"{"!==d[u+1]);)e+=d[u++];if(e.length>0){r.push(new s(e,n.Text));continue}}p((e=>/\s/.test(e)));const t=d[u];if("-"===t||"+"===t){const e=r.at(-1)?.type;if(e===n.Text||void 0===e)throw new SyntaxError(`Unexpected character: ${t}`);switch(e){case n.Identifier:case n.NumericLiteral:case n.BooleanLiteral:case n.StringLiteral:case n.CloseParen:case n.CloseSquareBracket:break;default:{++u;const e=p(i);r.push(new s(`${t}${e}`,e.length>0?n.NumericLiteral:n.UnaryOperator));continue}}}for(const[e,t]of l){if(d.slice(u,u+e.length)===e){r.push(new s(e,t)),u+=e.length;continue e}}if("'"!==t&&'"'!==t)if(i(t)){const e=p(i);r.push(new s(e,n.NumericLiteral))}else{if(!a(t))throw new SyntaxError(`Unexpected character: ${t}`);{const e=p(a),t=Object.hasOwn(o,e)?o[e]:n.Identifier;t===n.In&&r.at(-1)?.type===n.Not?(r.pop(),r.push(new s("not in",n.NotIn))):r.push(new s(e,t))}}else{++u;const e=p((e=>e!==t));r.push(new s(e,n.StringLiteral)),++u}}return r}var u=class{type="Statement"},p=class extends u{constructor(e){super(),this.body=e}type="Program"},h=class extends u{constructor(e,t,r){super(),this.test=e,this.body=t,this.alternate=r}type="If"},m=class extends u{constructor(e,t,r,n){super(),this.loopvar=e,this.iterable=t,this.body=r,this.defaultBlock=n}type="For"},_=class extends u{constructor(e,t){super(),this.assignee=e,this.value=t}type="Set"},f=class extends u{constructor(e,t,r){super(),this.name=e,this.args=t,this.body=r}type="Macro"},g=class extends u{type="Expression"},M=class extends g{constructor(e,t,r){super(),this.object=e,this.property=t,this.computed=r}type="MemberExpression"},w=class extends g{constructor(e,t){super(),this.callee=e,this.args=t}type="CallExpression"},T=class extends g{constructor(e){super(),this.value=e}type="Identifier"},b=class extends g{constructor(e){super(),this.value=e}type="Literal"},x=class extends b{type="NumericLiteral"},y=class extends b{type="StringLiteral"},k=class extends b{type="BooleanLiteral"},F=class extends b{type="ArrayLiteral"},P=class extends b{type="TupleLiteral"},C=class extends b{type="ObjectLiteral"},v=class extends g{constructor(e,t,r){super(),this.operator=e,this.left=t,this.right=r}type="BinaryExpression"},S=class extends g{constructor(e,t){super(),this.operand=e,this.filter=t}type="FilterExpression"},A=class extends g{constructor(e,t){super(),this.iterable=e,this.test=t}type="SelectExpression"},E=class extends g{constructor(e,t,r){super(),this.operand=e,this.negate=t,this.test=r}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,r=void 0){super(),this.start=e,this.stop=t,this.step=r}type="SliceExpression"},I=class extends g{constructor(e,t){super(),this.key=e,this.value=t}type="KeywordArgumentExpression"};function B(e){const t=new p([]);let r=0;function o(t,n){const o=e[r++];if(!o||o.type!==t)throw new Error(`Parser Error: ${n}. ${o.type} !== ${t}.`);return o}function s(){switch(e[r].type){case n.Text:return new y(o(n.Text,"Expected text token").value);case n.OpenStatement:return function(){let t;switch(o(n.OpenStatement,"Expected opening statement token"),e[r].type){case n.Set:++r,t=l(),o(n.CloseStatement,"Expected closing statement token");break;case n.If:++r,t=c(),o(n.OpenStatement,"Expected {% token"),o(n.EndIf,"Expected endif token"),o(n.CloseStatement,"Expected %} token");break;case n.Macro:++r,t=function(){const e=q();if("Identifier"!==e.type)throw new SyntaxError("Expected identifier following macro statement");const t=V();o(n.CloseStatement,"Expected closing statement token");const r=[];for(;a(n.OpenStatement,n.EndMacro);)r.push(s());return new f(e,t,r)}(),o(n.OpenStatement,"Expected {% token"),o(n.EndMacro,"Expected endmacro token"),o(n.CloseStatement,"Expected %} token");break;case n.For:++r,t=function(){const e=d(!0);if(!(e instanceof T||e instanceof P))throw new SyntaxError(`Expected identifier/tuple for the loop variable, got ${e.type} instead`);o(n.In,"Expected `in` keyword following loop variable");const t=u();o(n.CloseStatement,"Expected closing statement token");const l=[];for(;a(n.OpenStatement,n.EndFor)&&a(n.OpenStatement,n.Else);)l.push(s());const c=[];if(i(n.OpenStatement,n.Else))for(++r,++r,o(n.CloseStatement,"Expected closing statement token");a(n.OpenStatement,n.EndFor);)c.push(s());return new m(e,t,l,c)}(),o(n.OpenStatement,"Expected {% token"),o(n.EndFor,"Expected endfor token"),o(n.CloseStatement,"Expected %} token");break;default:throw new SyntaxError(`Unknown statement type: ${e[r].type}`)}return t}();case n.OpenExpression:return function(){o(n.OpenExpression,"Expected opening expression token");const e=u();return o(n.CloseExpression,"Expected closing expression token"),e}();default:throw new SyntaxError(`Unexpected token type: ${e[r].type}`)}}function a(...t){return r+t.length<=e.length&&t.some(((t,n)=>t!==e[r+n].type))}function i(...t){return r+t.length<=e.length&&t.every(((t,n)=>t===e[r+n].type))}function l(){const e=u();if(i(n.Equals)){++r;const t=l();return new _(e,t)}return e}function c(){const t=u();o(n.CloseStatement,"Expected closing statement token");const a=[],l=[];for(;e[r]?.type!==n.OpenStatement||e[r+1]?.type!==n.ElseIf&&e[r+1]?.type!==n.Else&&e[r+1]?.type!==n.EndIf;)a.push(s());if(e[r]?.type===n.OpenStatement&&e[r+1]?.type!==n.EndIf)if(++r,i(n.ElseIf))o(n.ElseIf,"Expected elseif token"),l.push(c());else for(o(n.Else,"Expected else token"),o(n.CloseStatement,"Expected closing statement token");e[r]?.type!==n.OpenStatement||e[r+1]?.type!==n.EndIf;)l.push(s());return new h(t,a,l)}function d(e=!1){const t=e?q:u,o=[t()],s=i(n.Comma);for(;s&&(++r,o.push(t()),i(n.Comma)););return s?new P(o):o[0]}function u(){return function(){const e=g();if(i(n.If)){++r;const t=g();if(i(n.Else)){++r;const n=g();return new h(t,[e],[n])}return new A(e,t)}return e}()}function g(){let t=b();for(;i(n.Or);){const n=e[r];++r;const o=b();t=new v(n,t,o)}return t}function b(){let t=B();for(;i(n.And);){const n=e[r];++r;const o=B();t=new v(n,t,o)}return t}function B(){let t;for(;i(n.Not);){const n=e[r];++r;const o=B();t=new L(n,o)}return t??function(){let t=N();for(;i(n.ComparisonBinaryOperator)||i(n.In)||i(n.NotIn);){const n=e[r];++r;const o=N();t=new v(n,t,o)}return t}()}function N(){let t=R();for(;i(n.AdditiveBinaryOperator);){const n=e[r];++r;const o=R();t=new v(n,t,o)}return t}function D(){const t=function(){let t=q();for(;i(n.Dot)||i(n.OpenSquareBracket);){const s=e[r];let a;++r;const i=s.type!==n.Dot;if(i)a=j(),o(n.CloseSquareBracket,"Expected closing square bracket");else if(a=q(),"Identifier"!==a.type)throw new SyntaxError("Expected identifier following dot operator");t=new M(t,a,i)}return t}();return i(n.OpenParen)?O(t):t}function O(e){let t=new w(e,V());return i(n.OpenParen)&&(t=O(t)),t}function V(){o(n.OpenParen,"Expected opening parenthesis for arguments list");const e=function(){const e=[];for(;!i(n.CloseParen);){let t=u();if(i(n.Equals)){if(++r,!(t instanceof T))throw new SyntaxError("Expected identifier for keyword argument");const e=u();t=new I(t,e)}e.push(t),i(n.Comma)&&++r}return e}();return o(n.CloseParen,"Expected closing parenthesis for arguments list"),e}function j(){const e=[];let t=!1;for(;!i(n.CloseSquareBracket);)i(n.Colon)?(e.push(void 0),++r,t=!0):(e.push(u()),i(n.Colon)&&(++r,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(){let t=G();for(;i(n.MultiplicativeBinaryOperator);){const n=e[r];++r;const o=G();t=new v(n,t,o)}return t}function G(){let e=function(){let e=D();for(;i(n.Pipe);){++r;let t=q();if(!(t instanceof T))throw new SyntaxError("Expected identifier for the filter");i(n.OpenParen)&&(t=O(t)),e=new S(e,t)}return e}();for(;i(n.Is);){++r;const t=i(n.Not);t&&++r;let o=q();if(o instanceof k&&(o=new T(o.value.toString())),!(o instanceof T))throw new SyntaxError("Expected identifier for the test");e=new E(e,t,o)}return e}function q(){const t=e[r];switch(t.type){case n.NumericLiteral:return++r,new x(Number(t.value));case n.StringLiteral:return++r,new y(t.value);case n.BooleanLiteral:return++r,new k("true"===t.value.toLowerCase());case n.Identifier:return++r,new T(t.value);case n.OpenParen:{++r;const t=d();if(e[r].type!==n.CloseParen)throw new SyntaxError(`Expected closing parenthesis, got ${e[r].type} instead`);return++r,t}case n.OpenSquareBracket:{++r;const e=[];for(;!i(n.CloseSquareBracket);)e.push(u()),i(n.Comma)&&++r;return++r,new F(e)}case n.OpenCurlyBracket:{++r;const e=new Map;for(;!i(n.CloseCurlyBracket);){const t=u();o(n.Colon,"Expected colon between key and value in object literal");const s=u();e.set(t,s),i(n.Comma)&&++r}return++r,new C(e)}default:throw new SyntaxError(`Unexpected token: ${t.type}`)}}for(;r<e.length;)t.body.push(s());return t}function N(e,t,r=1){void 0===t&&(t=e,e=0);const n=[];for(let o=e;o<t;o+=r)n.push(o);return n}function D(e,t,r,n=1){const o=Math.sign(n);o>=0?(t=(t??=0)<0?Math.max(e.length+t,0):Math.min(t,e.length),r=(r??=e.length)<0?Math.max(e.length+r,0):Math.min(r,e.length)):(t=(t??=e.length-1)<0?Math.max(e.length+t,-1):Math.min(t,e.length-1),r=(r??=-1)<-1?Math.max(e.length+r,-1):Math.min(r,e.length-1));const s=[];for(let a=t;o*a<o*r;a+=n)s.push(e[a]);return s}function O(e){return e.replace(/\b\w/g,(e=>e.toUpperCase()))}var V=class{type="RuntimeValue";value;builtins=new Map;constructor(e=void 0){this.value=e}__bool__(){return new G(!!this.value)}},j=class extends V{type="NumericValue"},R=class extends V{type="StringValue";builtins=new Map([["upper",new X((()=>new R(this.value.toUpperCase())))],["lower",new X((()=>new R(this.value.toLowerCase())))],["strip",new X((()=>new R(this.value.trim())))],["title",new X((()=>new R(O(this.value))))],["length",new j(this.value.length)]])},G=class extends V{type="BooleanValue"},q=class extends V{type="ObjectValue";__bool__(){return new G(this.value.size>0)}builtins=new Map([["get",new X((([e,t])=>{if(!(e instanceof R))throw new Error(`Object key must be a string: got ${e.type}`);return this.value.get(e.value)??t??new Q}))],["items",new X((()=>new $(Array.from(this.value.entries()).map((([e,t])=>new $([new R(e),t]))))))]])},W=class extends q{type="KeywordArgumentsValue"},$=class extends V{type="ArrayValue";builtins=new Map([["length",new j(this.value.length)]]);__bool__(){return new G(this.value.length>0)}},U=class extends ${type="TupleValue"},X=class extends V{type="FunctionValue"},Q=class extends V{type="NullValue"},H=class extends V{type="UndefinedValue"},Y=class{constructor(e){this.parent=e}variables=new Map([["namespace",new X((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 X],["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],["string",e=>"StringValue"===e.type],["number",e=>"NumericValue"===e.type],["integer",e=>"NumericValue"===e.type&&Number.isInteger(e.value)],["iterable",e=>e instanceof $||e instanceof R],["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,K(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 H}catch{return new H}}},J=class{global;constructor(e){this.global=e??new Y}run(e){return this.evaluate(e,this.global)}evaluateBinaryExpression(e,t){const r=this.evaluate(e.left,t);switch(e.operator.value){case"and":return r.__bool__().value?this.evaluate(e.right,t):r;case"or":return r.__bool__().value?r:this.evaluate(e.right,t)}const n=this.evaluate(e.right,t);switch(e.operator.value){case"==":return new G(r.value==n.value);case"!=":return new G(r.value!=n.value)}if(r instanceof H||n instanceof H)throw new Error("Cannot perform operation on undefined values");if(r instanceof Q||n instanceof Q)throw new Error("Cannot perform operation on null values");if(r instanceof j&&n instanceof j)switch(e.operator.value){case"+":return new j(r.value+n.value);case"-":return new j(r.value-n.value);case"*":return new j(r.value*n.value);case"/":return new j(r.value/n.value);case"%":return new j(r.value%n.value);case"<":return new G(r.value<n.value);case">":return new G(r.value>n.value);case">=":return new G(r.value>=n.value);case"<=":return new G(r.value<=n.value)}else if(r instanceof $&&n instanceof $){if("+"===e.operator.value)return new $(r.value.concat(n.value))}else if(n instanceof $){const t=void 0!==n.value.find((e=>e.value===r.value));switch(e.operator.value){case"in":return new G(t);case"not in":return new G(!t)}}if((r instanceof R||n instanceof R)&&"+"===e.operator.value)return new R(r.value.toString()+n.value.toString());if(r instanceof R&&n instanceof R)switch(e.operator.value){case"in":return new G(n.value.includes(r.value));case"not in":return new G(!n.value.includes(r.value))}if(r instanceof R&&n instanceof q)switch(e.operator.value){case"in":return new G(n.value.has(r.value));case"not in":return new G(!n.value.has(r.value))}throw new SyntaxError(`Unknown operator "${e.operator.value}" between ${r.type} and ${n.type}`)}evaluateArguments(e,t){const r=[],n=new Map;for(const o of e)if("KeywordArgumentExpression"===o.type){const e=o;n.set(e.key.value,this.evaluate(e.value,t))}else{if(n.size>0)throw new Error("Positional arguments must come before keyword arguments");r.push(this.evaluate(o,t))}return[r,n]}evaluateFilterExpression(e,t){const r=this.evaluate(e.operand,t);if("Identifier"===e.filter.type){const t=e.filter;if("tojson"===t.value)return new R(Z(r));if(r instanceof $)switch(t.value){case"list":return r;case"first":return r.value[0];case"last":return r.value[r.value.length-1];case"length":return new j(r.value.length);case"reverse":return new $(r.value.reverse());case"sort":return new $(r.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}`)}})));default:throw new Error(`Unknown ArrayValue filter: ${t.value}`)}else if(r instanceof R)switch(t.value){case"length":return new j(r.value.length);case"upper":return new R(r.value.toUpperCase());case"lower":return new R(r.value.toLowerCase());case"title":return new R(O(r.value));case"capitalize":return new R(r.value.charAt(0).toUpperCase()+r.value.slice(1));case"trim":return new R(r.value.trim());case"indent":return new R(r.value.split("\n").map(((e,t)=>0===t||0===e.length?e:"    "+e)).join("\n"));case"string":return r;default:throw new Error(`Unknown StringValue filter: ${t.value}`)}else{if(r instanceof j){if("abs"===t.value)return new j(Math.abs(r.value));throw new Error(`Unknown NumericValue filter: ${t.value}`)}if(r instanceof q)switch(t.value){case"items":return new $(Array.from(r.value.entries()).map((([e,t])=>new $([new R(e),t]))));case"length":return new j(r.value.size);default:throw new Error(`Unknown ObjectValue filter: ${t.value}`)}}throw new Error(`Cannot apply filter "${t.value}" to type: ${r.type}`)}if("CallExpression"===e.filter.type){const n=e.filter;if("Identifier"!==n.callee.type)throw new Error(`Unknown filter: ${n.callee.type}`);const o=n.callee.value;if("tojson"===o){const[,e]=this.evaluateArguments(n.args,t),o=e.get("indent")??new Q;if(!(o instanceof j||o instanceof Q))throw new Error("If set, indent must be a number");return new R(Z(r,o.value))}if(r instanceof $){switch(o){case"selectattr":{if(r.value.some((e=>!(e instanceof q))))throw new Error("`selectattr` can only be applied to array of objects");if(n.args.some((e=>"StringLiteral"!==e.type)))throw new Error("arguments of `selectattr` must be strings");const[e,o,s]=n.args.map((e=>this.evaluate(e,t)));let a;if(o){const e=t.tests.get(o.value);if(!e)throw new Error(`Unknown test: ${o.value}`);a=e}else a=(...e)=>e[0].__bool__().value;const i=r.value.filter((t=>{const r=t.value.get(e.value);return!!r&&a(r,s)}));return new $(i)}case"map":{const[,e]=this.evaluateArguments(n.args,t);if(e.has("attribute")){const t=e.get("attribute");if(!(t instanceof R))throw new Error("attribute must be a string");const n=e.get("default"),o=r.value.map((e=>{if(!(e instanceof q))throw new Error("items in map must be an object");return e.value.get(t.value)??n??new H}));return new $(o)}throw new Error("`map` expressions without `attribute` set are not currently supported.")}}throw new Error(`Unknown ArrayValue filter: ${o}`)}if(r instanceof R){if("indent"===o){const[e,o]=this.evaluateArguments(n.args,t),s=e.at(0)??o.get("width")??new j(4);if(!(s instanceof j))throw new Error("width must be a number");const a=e.at(1)??o.get("first")??new G(!1),i=e.at(2)??o.get("blank")??new G(!1),l=r.value.split("\n"),c=" ".repeat(s.value),d=l.map(((e,t)=>!a.value&&0===t||!i.value&&0===e.length?e:c+e));return new R(d.join("\n"))}throw new Error(`Unknown StringValue filter: ${o}`)}throw new Error(`Cannot apply filter "${o}" to type: ${r.type}`)}throw new Error(`Unknown filter: ${e.filter.type}`)}evaluateTestExpression(e,t){const r=this.evaluate(e.operand,t),n=t.tests.get(e.test.value);if(!n)throw new Error(`Unknown test: ${e.test.value}`);const o=n(r);return new G(e.negate?!o:o)}evaluateUnaryExpression(e,t){const r=this.evaluate(e.argument,t);if("not"===e.operator.value)return new G(!r.value);throw new SyntaxError(`Unknown operator: ${e.operator.value}`)}evalProgram(e,t){return this.evaluateBlock(e.body,t)}evaluateBlock(e,t){let r="";for(const n of e){const e=this.evaluate(n,t);"NullValue"!==e.type&&"UndefinedValue"!==e.type&&(r+=e.value)}return new R(r)}evaluateIdentifier(e,t){return t.lookupVariable(e.value)}evaluateCallExpression(e,t){const[r,n]=this.evaluateArguments(e.args,t);n.size>0&&r.push(new W(n));const o=this.evaluate(e.callee,t);if("FunctionValue"!==o.type)throw new Error(`Cannot call something that is not a function: got ${o.type}`);return o.value(r,t)}evaluateSliceExpression(e,t,r){if(!(e instanceof $||e instanceof R))throw new Error("Slice object must be an array or string");const n=this.evaluate(t.start,r),o=this.evaluate(t.stop,r),s=this.evaluate(t.step,r);if(!(n instanceof j||n instanceof H))throw new Error("Slice start must be numeric or undefined");if(!(o instanceof j||o instanceof H))throw new Error("Slice stop must be numeric or undefined");if(!(s instanceof j||s instanceof H))throw new Error("Slice step must be numeric or undefined");return e instanceof $?new $(D(e.value,n.value,o.value,s.value)):new R(D(Array.from(e.value),n.value,o.value,s.value).join(""))}evaluateMemberExpression(e,t){const r=this.evaluate(e.object,t);let n,o;if(e.computed){if("SliceExpression"===e.property.type)return this.evaluateSliceExpression(r,e.property,t);n=this.evaluate(e.property,t)}else n=new R(e.property.value);if(r instanceof q){if(!(n instanceof R))throw new Error(`Cannot access property with non-string: got ${n.type}`);o=r.value.get(n.value)??r.builtins.get(n.value)}else if(r instanceof $||r instanceof R)if(n instanceof j)o=r.value.at(n.value),r instanceof R&&(o=new R(r.value.at(n.value)));else{if(!(n instanceof R))throw new Error(`Cannot access property with non-string/non-number: got ${n.type}`);o=r.builtins.get(n.value)}else{if(!(n instanceof R))throw new Error(`Cannot access property with non-string: got ${n.type}`);o=r.builtins.get(n.value)}return o instanceof V?o:new H}evaluateSet(e,t){const r=this.evaluate(e.value,t);if("Identifier"===e.assignee.type){const n=e.assignee.value;t.setVariable(n,r)}else{if("MemberExpression"!==e.assignee.type)throw new Error(`Invalid LHS inside assignment expression: ${JSON.stringify(e.assignee)}`);{const n=e.assignee,o=this.evaluate(n.object,t);if(!(o instanceof q))throw new Error("Cannot assign to member of non-object");if("Identifier"!==n.property.type)throw new Error("Cannot assign to member with non-identifier property");o.value.set(n.property.value,r)}}return new Q}evaluateIf(e,t){const r=this.evaluate(e.test,t);return this.evaluateBlock(r.__bool__().value?e.body:e.alternate,t)}evaluateFor(e,t){const r=new Y(t);let n,o;if("SelectExpression"===e.iterable.type){const t=e.iterable;o=this.evaluate(t.iterable,r),n=t.test}else o=this.evaluate(e.iterable,r);if(!(o instanceof $))throw new Error(`Expected iterable type in for loop: got ${o.type}`);const s=[],a=[];for(let t=0;t<o.value.length;++t){const i=new Y(r),l=o.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 r=l;if(t.value.length!==r.value.length)throw new Error(`Too ${t.value.length>r.value.length?"few":"many"} items to unpack`);c=e=>{for(let n=0;n<t.value.length;++n){if("Identifier"!==t.value[n].type)throw new Error(`Cannot unpack non-identifier type: ${t.value[n].type}`);e.setVariable(t.value[n].value,r.value[n])}}}}if(n){c(i);if(!this.evaluate(n,i).__bool__().value)continue}s.push(l),a.push(c)}let i="",l=!0;for(let t=0;t<s.length;++t){const n=new Map([["index",new j(t+1)],["index0",new j(t)],["revindex",new j(s.length-t)],["revindex0",new j(s.length-t-1)],["first",new G(0===t)],["last",new G(t===s.length-1)],["length",new j(s.length)],["previtem",t>0?s[t-1]:new H],["nextitem",t<s.length-1?s[t+1]:new H]]);r.setVariable("loop",new q(n)),a[t](r);i+=this.evaluateBlock(e.body,r).value,l=!1}if(l){i+=this.evaluateBlock(e.defaultBlock,r).value}return new R(i)}evaluateMacro(e,t){return t.setVariable(e.name.value,new X(((t,r)=>{const n=new Y(r);let o;t=t.slice(),"KeywordArgumentsValue"===t.at(-1)?.type&&(o=t.pop());for(let r=0;r<e.args.length;++r){const s=e.args[r],a=t[r];if("Identifier"===s.type){const e=s;if(!a)throw new Error(`Missing positional argument: ${e.value}`);n.setVariable(e.value,a)}else{if("KeywordArgumentExpression"!==s.type)throw new Error(`Unknown argument type: ${s.type}`);{const e=s,t=a??o?.value.get(e.key.value)??this.evaluate(e.value,n);n.setVariable(e.key.value,t)}}}return this.evaluateBlock(e.body,n)}))),new Q}evaluate(e,t){if(void 0===e)return new H;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 j(Number(e.value));case"StringLiteral":return new R(e.value);case"BooleanLiteral":return new G(e.value);case"ArrayLiteral":return new $(e.value.map((e=>this.evaluate(e,t))));case"TupleLiteral":return new U(e.value.map((e=>this.evaluate(e,t))));case"ObjectLiteral":{const r=new Map;for(const[n,o]of e.value){const e=this.evaluate(n,t);if(!(e instanceof R))throw new Error(`Object keys must be strings: got ${e.type}`);r.set(e.value,this.evaluate(o,t))}return new q(r)}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 K(e){switch(typeof e){case"number":return new j(e);case"string":return new R(e);case"boolean":return new G(e);case"undefined":return new H;case"object":return null===e?new Q:Array.isArray(e)?new $(e.map(K)):new q(new Map(Object.entries(e).map((([e,t])=>[e,K(t)]))));case"function":return new X(((t,r)=>K(e(...t.map((e=>e.value)))??null)));default:throw new Error(`Cannot convert to runtime value: ${e}`)}}function Z(e,t,r){const n=r??0;switch(e.type){case"NullValue":case"UndefinedValue":return"null";case"NumericValue":case"StringValue":case"BooleanValue":return JSON.stringify(e.value);case"ArrayValue":case"ObjectValue":{const r=t?" ".repeat(t):"",o="\n"+r.repeat(n),s=o+r;if("ArrayValue"===e.type){const r=e.value.map((e=>Z(e,t,n+1)));return t?`[${s}${r.join(`,${s}`)}${o}]`:`[${r.join(", ")}]`}{const r=Array.from(e.value.entries()).map((([e,r])=>{const o=`"${e}": ${Z(r,t,n+1)}`;return t?`${s}${o}`:o}));return t?`{${r.join(",")}${o}}`:`{${r.join(", ")}}`}}default:throw new Error(`Cannot convert to JSON: ${e.type}`)}}var ee=class{parsed;constructor(e){const t=d(e,{lstrip_blocks:!0,trim_blocks:!0});this.parsed=B(t)}render(e){const t=new Y;t.set("false",!1),t.set("true",!0),t.set("raise_exception",(e=>{throw new Error(e)})),t.set("range",N);for(const[r,n]of Object.entries(e))t.set(r,n);return new J(t).run(this.parsed).value}}},"./node_modules/onnxruntime-common/dist/esm/backend-impl.js":
+  \*******************************************************/(e,t,r)=>{r.r(t),r.d(t,{Environment:()=>Y,Interpreter:()=>J,Template:()=>ee,parse:()=>B,tokenize:()=>d});var n=Object.freeze({Text:"Text",NumericLiteral:"NumericLiteral",BooleanLiteral:"BooleanLiteral",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"}),o=Object.freeze({set:n.Set,for:n.For,in:n.In,is:n.Is,if:n.If,else:n.Else,endif:n.EndIf,elif:n.ElseIf,endfor:n.EndFor,and:n.And,or:n.Or,not:n.Not,"not in":n.NotIn,macro:n.Macro,endmacro:n.EndMacro,true:n.BooleanLiteral,false:n.BooleanLiteral,True:n.BooleanLiteral,False:n.BooleanLiteral}),s=class{constructor(e,t){this.value=e,this.type=t}};function a(e){return/\w/.test(e)}function i(e){return/[0-9]/.test(e)}var l=[["{%",n.OpenStatement],["%}",n.CloseStatement],["{{",n.OpenExpression],["}}",n.CloseExpression],["(",n.OpenParen],[")",n.CloseParen],["{",n.OpenCurlyBracket],["}",n.CloseCurlyBracket],["[",n.OpenSquareBracket],["]",n.CloseSquareBracket],[",",n.Comma],[".",n.Dot],[":",n.Colon],["|",n.Pipe],["<=",n.ComparisonBinaryOperator],[">=",n.ComparisonBinaryOperator],["==",n.ComparisonBinaryOperator],["!=",n.ComparisonBinaryOperator],["<",n.ComparisonBinaryOperator],[">",n.ComparisonBinaryOperator],["+",n.AdditiveBinaryOperator],["-",n.AdditiveBinaryOperator],["*",n.MultiplicativeBinaryOperator],["/",n.MultiplicativeBinaryOperator],["%",n.MultiplicativeBinaryOperator],["=",n.Equals]],c=new Map([["n","\n"],["t","\t"],["r","\r"],["b","\b"],["f","\f"],["v","\v"],["'","'"],['"','"'],["\\","\\"]]);function d(e,t={}){const r=[],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 p=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++],r=c.get(e);if(void 0===r)throw new SyntaxError(`Unexpected escaped character: ${e}`);t+=r}return t};e:for(;u<d.length;){const e=r.at(-1)?.type;if(void 0===e||e===n.CloseStatement||e===n.CloseExpression){let e="";for(;u<d.length&&("{"!==d[u]||"%"!==d[u+1]&&"{"!==d[u+1]);)e+=d[u++];if(e.length>0){r.push(new s(e,n.Text));continue}}p((e=>/\s/.test(e)));const t=d[u];if("-"===t||"+"===t){const e=r.at(-1)?.type;if(e===n.Text||void 0===e)throw new SyntaxError(`Unexpected character: ${t}`);switch(e){case n.Identifier:case n.NumericLiteral:case n.BooleanLiteral:case n.StringLiteral:case n.CloseParen:case n.CloseSquareBracket:break;default:{++u;const e=p(i);r.push(new s(`${t}${e}`,e.length>0?n.NumericLiteral:n.UnaryOperator));continue}}}for(const[e,t]of l){if(d.slice(u,u+e.length)===e){r.push(new s(e,t)),u+=e.length;continue e}}if("'"!==t&&'"'!==t)if(i(t)){const e=p(i);r.push(new s(e,n.NumericLiteral))}else{if(!a(t))throw new SyntaxError(`Unexpected character: ${t}`);{const e=p(a),t=Object.hasOwn(o,e)?o[e]:n.Identifier;t===n.In&&r.at(-1)?.type===n.Not?(r.pop(),r.push(new s("not in",n.NotIn))):r.push(new s(e,t))}}else{++u;const e=p((e=>e!==t));r.push(new s(e,n.StringLiteral)),++u}}return r}var u=class{type="Statement"},p=class extends u{constructor(e){super(),this.body=e}type="Program"},h=class extends u{constructor(e,t,r){super(),this.test=e,this.body=t,this.alternate=r}type="If"},m=class extends u{constructor(e,t,r,n){super(),this.loopvar=e,this.iterable=t,this.body=r,this.defaultBlock=n}type="For"},_=class extends u{constructor(e,t){super(),this.assignee=e,this.value=t}type="Set"},f=class extends u{constructor(e,t,r){super(),this.name=e,this.args=t,this.body=r}type="Macro"},g=class extends u{type="Expression"},M=class extends g{constructor(e,t,r){super(),this.object=e,this.property=t,this.computed=r}type="MemberExpression"},w=class extends g{constructor(e,t){super(),this.callee=e,this.args=t}type="CallExpression"},T=class extends g{constructor(e){super(),this.value=e}type="Identifier"},b=class extends g{constructor(e){super(),this.value=e}type="Literal"},x=class extends b{type="NumericLiteral"},y=class extends b{type="StringLiteral"},k=class extends b{type="BooleanLiteral"},F=class extends b{type="ArrayLiteral"},P=class extends b{type="TupleLiteral"},C=class extends b{type="ObjectLiteral"},v=class extends g{constructor(e,t,r){super(),this.operator=e,this.left=t,this.right=r}type="BinaryExpression"},S=class extends g{constructor(e,t){super(),this.operand=e,this.filter=t}type="FilterExpression"},A=class extends g{constructor(e,t){super(),this.iterable=e,this.test=t}type="SelectExpression"},E=class extends g{constructor(e,t,r){super(),this.operand=e,this.negate=t,this.test=r}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,r=void 0){super(),this.start=e,this.stop=t,this.step=r}type="SliceExpression"},I=class extends g{constructor(e,t){super(),this.key=e,this.value=t}type="KeywordArgumentExpression"};function B(e){const t=new p([]);let r=0;function o(t,n){const o=e[r++];if(!o||o.type!==t)throw new Error(`Parser Error: ${n}. ${o.type} !== ${t}.`);return o}function s(){switch(e[r].type){case n.Text:return new y(o(n.Text,"Expected text token").value);case n.OpenStatement:return function(){let t;switch(o(n.OpenStatement,"Expected opening statement token"),e[r].type){case n.Set:++r,t=l(),o(n.CloseStatement,"Expected closing statement token");break;case n.If:++r,t=c(),o(n.OpenStatement,"Expected {% token"),o(n.EndIf,"Expected endif token"),o(n.CloseStatement,"Expected %} token");break;case n.Macro:++r,t=function(){const e=q();if("Identifier"!==e.type)throw new SyntaxError("Expected identifier following macro statement");const t=V();o(n.CloseStatement,"Expected closing statement token");const r=[];for(;a(n.OpenStatement,n.EndMacro);)r.push(s());return new f(e,t,r)}(),o(n.OpenStatement,"Expected {% token"),o(n.EndMacro,"Expected endmacro token"),o(n.CloseStatement,"Expected %} token");break;case n.For:++r,t=function(){const e=d(!0);if(!(e instanceof T||e instanceof P))throw new SyntaxError(`Expected identifier/tuple for the loop variable, got ${e.type} instead`);o(n.In,"Expected `in` keyword following loop variable");const t=u();o(n.CloseStatement,"Expected closing statement token");const l=[];for(;a(n.OpenStatement,n.EndFor)&&a(n.OpenStatement,n.Else);)l.push(s());const c=[];if(i(n.OpenStatement,n.Else))for(++r,++r,o(n.CloseStatement,"Expected closing statement token");a(n.OpenStatement,n.EndFor);)c.push(s());return new m(e,t,l,c)}(),o(n.OpenStatement,"Expected {% token"),o(n.EndFor,"Expected endfor token"),o(n.CloseStatement,"Expected %} token");break;default:throw new SyntaxError(`Unknown statement type: ${e[r].type}`)}return t}();case n.OpenExpression:return function(){o(n.OpenExpression,"Expected opening expression token");const e=u();return o(n.CloseExpression,"Expected closing expression token"),e}();default:throw new SyntaxError(`Unexpected token type: ${e[r].type}`)}}function a(...t){return r+t.length<=e.length&&t.some(((t,n)=>t!==e[r+n].type))}function i(...t){return r+t.length<=e.length&&t.every(((t,n)=>t===e[r+n].type))}function l(){const e=u();if(i(n.Equals)){++r;const t=l();return new _(e,t)}return e}function c(){const t=u();o(n.CloseStatement,"Expected closing statement token");const a=[],l=[];for(;e[r]?.type!==n.OpenStatement||e[r+1]?.type!==n.ElseIf&&e[r+1]?.type!==n.Else&&e[r+1]?.type!==n.EndIf;)a.push(s());if(e[r]?.type===n.OpenStatement&&e[r+1]?.type!==n.EndIf)if(++r,i(n.ElseIf))o(n.ElseIf,"Expected elseif token"),l.push(c());else for(o(n.Else,"Expected else token"),o(n.CloseStatement,"Expected closing statement token");e[r]?.type!==n.OpenStatement||e[r+1]?.type!==n.EndIf;)l.push(s());return new h(t,a,l)}function d(e=!1){const t=e?q:u,o=[t()],s=i(n.Comma);for(;s&&(++r,o.push(t()),i(n.Comma)););return s?new P(o):o[0]}function u(){return function(){const e=g();if(i(n.If)){++r;const t=g();if(i(n.Else)){++r;const n=g();return new h(t,[e],[n])}return new A(e,t)}return e}()}function g(){let t=b();for(;i(n.Or);){const n=e[r];++r;const o=b();t=new v(n,t,o)}return t}function b(){let t=B();for(;i(n.And);){const n=e[r];++r;const o=B();t=new v(n,t,o)}return t}function B(){let t;for(;i(n.Not);){const n=e[r];++r;const o=B();t=new L(n,o)}return t??function(){let t=N();for(;i(n.ComparisonBinaryOperator)||i(n.In)||i(n.NotIn);){const n=e[r];++r;const o=N();t=new v(n,t,o)}return t}()}function N(){let t=R();for(;i(n.AdditiveBinaryOperator);){const n=e[r];++r;const o=R();t=new v(n,t,o)}return t}function D(){const t=function(){let t=q();for(;i(n.Dot)||i(n.OpenSquareBracket);){const s=e[r];let a;++r;const i=s.type!==n.Dot;if(i)a=j(),o(n.CloseSquareBracket,"Expected closing square bracket");else if(a=q(),"Identifier"!==a.type)throw new SyntaxError("Expected identifier following dot operator");t=new M(t,a,i)}return t}();return i(n.OpenParen)?O(t):t}function O(e){let t=new w(e,V());return i(n.OpenParen)&&(t=O(t)),t}function V(){o(n.OpenParen,"Expected opening parenthesis for arguments list");const e=function(){const e=[];for(;!i(n.CloseParen);){let t=u();if(i(n.Equals)){if(++r,!(t instanceof T))throw new SyntaxError("Expected identifier for keyword argument");const e=u();t=new I(t,e)}e.push(t),i(n.Comma)&&++r}return e}();return o(n.CloseParen,"Expected closing parenthesis for arguments list"),e}function j(){const e=[];let t=!1;for(;!i(n.CloseSquareBracket);)i(n.Colon)?(e.push(void 0),++r,t=!0):(e.push(u()),i(n.Colon)&&(++r,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(){let t=G();for(;i(n.MultiplicativeBinaryOperator);){const n=e[r];++r;const o=G();t=new v(n,t,o)}return t}function G(){let e=function(){let e=D();for(;i(n.Pipe);){++r;let t=q();if(!(t instanceof T))throw new SyntaxError("Expected identifier for the filter");i(n.OpenParen)&&(t=O(t)),e=new S(e,t)}return e}();for(;i(n.Is);){++r;const t=i(n.Not);t&&++r;let o=q();if(o instanceof k&&(o=new T(o.value.toString())),!(o instanceof T))throw new SyntaxError("Expected identifier for the test");e=new E(e,t,o)}return e}function q(){const t=e[r];switch(t.type){case n.NumericLiteral:return++r,new x(Number(t.value));case n.StringLiteral:return++r,new y(t.value);case n.BooleanLiteral:return++r,new k("true"===t.value.toLowerCase());case n.Identifier:return++r,new T(t.value);case n.OpenParen:{++r;const t=d();if(e[r].type!==n.CloseParen)throw new SyntaxError(`Expected closing parenthesis, got ${e[r].type} instead`);return++r,t}case n.OpenSquareBracket:{++r;const e=[];for(;!i(n.CloseSquareBracket);)e.push(u()),i(n.Comma)&&++r;return++r,new F(e)}case n.OpenCurlyBracket:{++r;const e=new Map;for(;!i(n.CloseCurlyBracket);){const t=u();o(n.Colon,"Expected colon between key and value in object literal");const s=u();e.set(t,s),i(n.Comma)&&++r}return++r,new C(e)}default:throw new SyntaxError(`Unexpected token: ${t.type}`)}}for(;r<e.length;)t.body.push(s());return t}function N(e,t,r=1){void 0===t&&(t=e,e=0);const n=[];for(let o=e;o<t;o+=r)n.push(o);return n}function D(e,t,r,n=1){const o=Math.sign(n);o>=0?(t=(t??=0)<0?Math.max(e.length+t,0):Math.min(t,e.length),r=(r??=e.length)<0?Math.max(e.length+r,0):Math.min(r,e.length)):(t=(t??=e.length-1)<0?Math.max(e.length+t,-1):Math.min(t,e.length-1),r=(r??=-1)<-1?Math.max(e.length+r,-1):Math.min(r,e.length-1));const s=[];for(let a=t;o*a<o*r;a+=n)s.push(e[a]);return s}function O(e){return e.replace(/\b\w/g,(e=>e.toUpperCase()))}var V=class{type="RuntimeValue";value;builtins=new Map;constructor(e=void 0){this.value=e}__bool__(){return new G(!!this.value)}},j=class extends V{type="NumericValue"},R=class extends V{type="StringValue";builtins=new Map([["upper",new X((()=>new R(this.value.toUpperCase())))],["lower",new X((()=>new R(this.value.toLowerCase())))],["strip",new X((()=>new R(this.value.trim())))],["title",new X((()=>new R(O(this.value))))],["length",new j(this.value.length)]])},G=class extends V{type="BooleanValue"},q=class extends V{type="ObjectValue";__bool__(){return new G(this.value.size>0)}builtins=new Map([["get",new X((([e,t])=>{if(!(e instanceof R))throw new Error(`Object key must be a string: got ${e.type}`);return this.value.get(e.value)??t??new Q}))],["items",new X((()=>new W(Array.from(this.value.entries()).map((([e,t])=>new W([new R(e),t]))))))]])},$=class extends q{type="KeywordArgumentsValue"},W=class extends V{type="ArrayValue";builtins=new Map([["length",new j(this.value.length)]]);__bool__(){return new G(this.value.length>0)}},U=class extends W{type="TupleValue"},X=class extends V{type="FunctionValue"},Q=class extends V{type="NullValue"},H=class extends V{type="UndefinedValue"},Y=class{constructor(e){this.parent=e}variables=new Map([["namespace",new X((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 X],["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],["string",e=>"StringValue"===e.type],["number",e=>"NumericValue"===e.type],["integer",e=>"NumericValue"===e.type&&Number.isInteger(e.value)],["iterable",e=>e instanceof W||e instanceof R],["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,K(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: ${e}`)}lookupVariable(e){try{return this.resolve(e).variables.get(e)??new H}catch{return new H}}},J=class{global;constructor(e){this.global=e??new Y}run(e){return this.evaluate(e,this.global)}evaluateBinaryExpression(e,t){const r=this.evaluate(e.left,t);switch(e.operator.value){case"and":return r.__bool__().value?this.evaluate(e.right,t):r;case"or":return r.__bool__().value?r:this.evaluate(e.right,t)}const n=this.evaluate(e.right,t);switch(e.operator.value){case"==":return new G(r.value==n.value);case"!=":return new G(r.value!=n.value)}if(r instanceof H||n instanceof H)throw new Error("Cannot perform operation on undefined values");if(r instanceof Q||n instanceof Q)throw new Error("Cannot perform operation on null values");if(r instanceof j&&n instanceof j)switch(e.operator.value){case"+":return new j(r.value+n.value);case"-":return new j(r.value-n.value);case"*":return new j(r.value*n.value);case"/":return new j(r.value/n.value);case"%":return new j(r.value%n.value);case"<":return new G(r.value<n.value);case">":return new G(r.value>n.value);case">=":return new G(r.value>=n.value);case"<=":return new G(r.value<=n.value)}else if(r instanceof W&&n instanceof W){if("+"===e.operator.value)return new W(r.value.concat(n.value))}else if(n instanceof W){const t=void 0!==n.value.find((e=>e.value===r.value));switch(e.operator.value){case"in":return new G(t);case"not in":return new G(!t)}}if((r instanceof R||n instanceof R)&&"+"===e.operator.value)return new R(r.value.toString()+n.value.toString());if(r instanceof R&&n instanceof R)switch(e.operator.value){case"in":return new G(n.value.includes(r.value));case"not in":return new G(!n.value.includes(r.value))}if(r instanceof R&&n instanceof q)switch(e.operator.value){case"in":return new G(n.value.has(r.value));case"not in":return new G(!n.value.has(r.value))}throw new SyntaxError(`Unknown operator "${e.operator.value}" between ${r.type} and ${n.type}`)}evaluateArguments(e,t){const r=[],n=new Map;for(const o of e)if("KeywordArgumentExpression"===o.type){const e=o;n.set(e.key.value,this.evaluate(e.value,t))}else{if(n.size>0)throw new Error("Positional arguments must come before keyword arguments");r.push(this.evaluate(o,t))}return[r,n]}evaluateFilterExpression(e,t){const r=this.evaluate(e.operand,t);if("Identifier"===e.filter.type){const t=e.filter;if("tojson"===t.value)return new R(Z(r));if(r instanceof W)switch(t.value){case"list":return r;case"first":return r.value[0];case"last":return r.value[r.value.length-1];case"length":return new j(r.value.length);case"reverse":return new W(r.value.reverse());case"sort":return new W(r.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}`)}})));default:throw new Error(`Unknown ArrayValue filter: ${t.value}`)}else if(r instanceof R)switch(t.value){case"length":return new j(r.value.length);case"upper":return new R(r.value.toUpperCase());case"lower":return new R(r.value.toLowerCase());case"title":return new R(O(r.value));case"capitalize":return new R(r.value.charAt(0).toUpperCase()+r.value.slice(1));case"trim":return new R(r.value.trim());case"indent":return new R(r.value.split("\n").map(((e,t)=>0===t||0===e.length?e:"    "+e)).join("\n"));case"string":return r;default:throw new Error(`Unknown StringValue filter: ${t.value}`)}else{if(r instanceof j){if("abs"===t.value)return new j(Math.abs(r.value));throw new Error(`Unknown NumericValue filter: ${t.value}`)}if(r instanceof q)switch(t.value){case"items":return new W(Array.from(r.value.entries()).map((([e,t])=>new W([new R(e),t]))));case"length":return new j(r.value.size);default:throw new Error(`Unknown ObjectValue filter: ${t.value}`)}}throw new Error(`Cannot apply filter "${t.value}" to type: ${r.type}`)}if("CallExpression"===e.filter.type){const n=e.filter;if("Identifier"!==n.callee.type)throw new Error(`Unknown filter: ${n.callee.type}`);const o=n.callee.value;if("tojson"===o){const[,e]=this.evaluateArguments(n.args,t),o=e.get("indent")??new Q;if(!(o instanceof j||o instanceof Q))throw new Error("If set, indent must be a number");return new R(Z(r,o.value))}if(r instanceof W){switch(o){case"selectattr":{if(r.value.some((e=>!(e instanceof q))))throw new Error("`selectattr` can only be applied to array of objects");if(n.args.some((e=>"StringLiteral"!==e.type)))throw new Error("arguments of `selectattr` must be strings");const[e,o,s]=n.args.map((e=>this.evaluate(e,t)));let a;if(o){const e=t.tests.get(o.value);if(!e)throw new Error(`Unknown test: ${o.value}`);a=e}else a=(...e)=>e[0].__bool__().value;const i=r.value.filter((t=>{const r=t.value.get(e.value);return!!r&&a(r,s)}));return new W(i)}case"map":{const[,e]=this.evaluateArguments(n.args,t);if(e.has("attribute")){const t=e.get("attribute");if(!(t instanceof R))throw new Error("attribute must be a string");const n=e.get("default"),o=r.value.map((e=>{if(!(e instanceof q))throw new Error("items in map must be an object");return e.value.get(t.value)??n??new H}));return new W(o)}throw new Error("`map` expressions without `attribute` set are not currently supported.")}}throw new Error(`Unknown ArrayValue filter: ${o}`)}if(r instanceof R){if("indent"===o){const[e,o]=this.evaluateArguments(n.args,t),s=e.at(0)??o.get("width")??new j(4);if(!(s instanceof j))throw new Error("width must be a number");const a=e.at(1)??o.get("first")??new G(!1),i=e.at(2)??o.get("blank")??new G(!1),l=r.value.split("\n"),c=" ".repeat(s.value),d=l.map(((e,t)=>!a.value&&0===t||!i.value&&0===e.length?e:c+e));return new R(d.join("\n"))}throw new Error(`Unknown StringValue filter: ${o}`)}throw new Error(`Cannot apply filter "${o}" to type: ${r.type}`)}throw new Error(`Unknown filter: ${e.filter.type}`)}evaluateTestExpression(e,t){const r=this.evaluate(e.operand,t),n=t.tests.get(e.test.value);if(!n)throw new Error(`Unknown test: ${e.test.value}`);const o=n(r);return new G(e.negate?!o:o)}evaluateUnaryExpression(e,t){const r=this.evaluate(e.argument,t);if("not"===e.operator.value)return new G(!r.value);throw new SyntaxError(`Unknown operator: ${e.operator.value}`)}evalProgram(e,t){return this.evaluateBlock(e.body,t)}evaluateBlock(e,t){let r="";for(const n of e){const e=this.evaluate(n,t);"NullValue"!==e.type&&"UndefinedValue"!==e.type&&(r+=e.value)}return new R(r)}evaluateIdentifier(e,t){return t.lookupVariable(e.value)}evaluateCallExpression(e,t){const[r,n]=this.evaluateArguments(e.args,t);n.size>0&&r.push(new $(n));const o=this.evaluate(e.callee,t);if("FunctionValue"!==o.type)throw new Error(`Cannot call something that is not a function: got ${o.type}`);return o.value(r,t)}evaluateSliceExpression(e,t,r){if(!(e instanceof W||e instanceof R))throw new Error("Slice object must be an array or string");const n=this.evaluate(t.start,r),o=this.evaluate(t.stop,r),s=this.evaluate(t.step,r);if(!(n instanceof j||n instanceof H))throw new Error("Slice start must be numeric or undefined");if(!(o instanceof j||o instanceof H))throw new Error("Slice stop must be numeric or undefined");if(!(s instanceof j||s instanceof H))throw new Error("Slice step must be numeric or undefined");return e instanceof W?new W(D(e.value,n.value,o.value,s.value)):new R(D(Array.from(e.value),n.value,o.value,s.value).join(""))}evaluateMemberExpression(e,t){const r=this.evaluate(e.object,t);let n,o;if(e.computed){if("SliceExpression"===e.property.type)return this.evaluateSliceExpression(r,e.property,t);n=this.evaluate(e.property,t)}else n=new R(e.property.value);if(r instanceof q){if(!(n instanceof R))throw new Error(`Cannot access property with non-string: got ${n.type}`);o=r.value.get(n.value)??r.builtins.get(n.value)}else if(r instanceof W||r instanceof R)if(n instanceof j)o=r.value.at(n.value),r instanceof R&&(o=new R(r.value.at(n.value)));else{if(!(n instanceof R))throw new Error(`Cannot access property with non-string/non-number: got ${n.type}`);o=r.builtins.get(n.value)}else{if(!(n instanceof R))throw new Error(`Cannot access property with non-string: got ${n.type}`);o=r.builtins.get(n.value)}return o instanceof V?o:new H}evaluateSet(e,t){const r=this.evaluate(e.value,t);if("Identifier"===e.assignee.type){const n=e.assignee.value;t.setVariable(n,r)}else{if("MemberExpression"!==e.assignee.type)throw new Error(`Invalid LHS inside assignment expression: ${JSON.stringify(e.assignee)}`);{const n=e.assignee,o=this.evaluate(n.object,t);if(!(o instanceof q))throw new Error("Cannot assign to member of non-object");if("Identifier"!==n.property.type)throw new Error("Cannot assign to member with non-identifier property");o.value.set(n.property.value,r)}}return new Q}evaluateIf(e,t){const r=this.evaluate(e.test,t);return this.evaluateBlock(r.__bool__().value?e.body:e.alternate,t)}evaluateFor(e,t){const r=new Y(t);let n,o;if("SelectExpression"===e.iterable.type){const t=e.iterable;o=this.evaluate(t.iterable,r),n=t.test}else o=this.evaluate(e.iterable,r);if(!(o instanceof W))throw new Error(`Expected iterable type in for loop: got ${o.type}`);const s=[],a=[];for(let t=0;t<o.value.length;++t){const i=new Y(r),l=o.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 r=l;if(t.value.length!==r.value.length)throw new Error(`Too ${t.value.length>r.value.length?"few":"many"} items to unpack`);c=e=>{for(let n=0;n<t.value.length;++n){if("Identifier"!==t.value[n].type)throw new Error(`Cannot unpack non-identifier type: ${t.value[n].type}`);e.setVariable(t.value[n].value,r.value[n])}}}}if(n){c(i);if(!this.evaluate(n,i).__bool__().value)continue}s.push(l),a.push(c)}let i="",l=!0;for(let t=0;t<s.length;++t){const n=new Map([["index",new j(t+1)],["index0",new j(t)],["revindex",new j(s.length-t)],["revindex0",new j(s.length-t-1)],["first",new G(0===t)],["last",new G(t===s.length-1)],["length",new j(s.length)],["previtem",t>0?s[t-1]:new H],["nextitem",t<s.length-1?s[t+1]:new H]]);r.setVariable("loop",new q(n)),a[t](r);i+=this.evaluateBlock(e.body,r).value,l=!1}if(l){i+=this.evaluateBlock(e.defaultBlock,r).value}return new R(i)}evaluateMacro(e,t){return t.setVariable(e.name.value,new X(((t,r)=>{const n=new Y(r);let o;t=t.slice(),"KeywordArgumentsValue"===t.at(-1)?.type&&(o=t.pop());for(let r=0;r<e.args.length;++r){const s=e.args[r],a=t[r];if("Identifier"===s.type){const e=s;if(!a)throw new Error(`Missing positional argument: ${e.value}`);n.setVariable(e.value,a)}else{if("KeywordArgumentExpression"!==s.type)throw new Error(`Unknown argument type: ${s.type}`);{const e=s,t=a??o?.value.get(e.key.value)??this.evaluate(e.value,n);n.setVariable(e.key.value,t)}}}return this.evaluateBlock(e.body,n)}))),new Q}evaluate(e,t){if(void 0===e)return new H;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 j(Number(e.value));case"StringLiteral":return new R(e.value);case"BooleanLiteral":return new G(e.value);case"ArrayLiteral":return new W(e.value.map((e=>this.evaluate(e,t))));case"TupleLiteral":return new U(e.value.map((e=>this.evaluate(e,t))));case"ObjectLiteral":{const r=new Map;for(const[n,o]of e.value){const e=this.evaluate(n,t);if(!(e instanceof R))throw new Error(`Object keys must be strings: got ${e.type}`);r.set(e.value,this.evaluate(o,t))}return new q(r)}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 K(e){switch(typeof e){case"number":return new j(e);case"string":return new R(e);case"boolean":return new G(e);case"undefined":return new H;case"object":return null===e?new Q:Array.isArray(e)?new W(e.map(K)):new q(new Map(Object.entries(e).map((([e,t])=>[e,K(t)]))));case"function":return new X(((t,r)=>K(e(...t.map((e=>e.value)))??null)));default:throw new Error(`Cannot convert to runtime value: ${e}`)}}function Z(e,t,r){const n=r??0;switch(e.type){case"NullValue":case"UndefinedValue":return"null";case"NumericValue":case"StringValue":case"BooleanValue":return JSON.stringify(e.value);case"ArrayValue":case"ObjectValue":{const r=t?" ".repeat(t):"",o="\n"+r.repeat(n),s=o+r;if("ArrayValue"===e.type){const r=e.value.map((e=>Z(e,t,n+1)));return t?`[${s}${r.join(`,${s}`)}${o}]`:`[${r.join(", ")}]`}{const r=Array.from(e.value.entries()).map((([e,r])=>{const o=`"${e}": ${Z(r,t,n+1)}`;return t?`${s}${o}`:o}));return t?`{${r.join(",")}${o}}`:`{${r.join(", ")}}`}}default:throw new Error(`Cannot convert to JSON: ${e.type}`)}}var ee=class{parsed;constructor(e){const t=d(e,{lstrip_blocks:!0,trim_blocks:!0});this.parsed=B(t)}render(e){const t=new Y;t.set("false",!1),t.set("true",!0),t.set("raise_exception",(e=>{throw new Error(e)})),t.set("range",N);for(const[r,n]of Object.entries(e))t.set(r,n);return new J(t).run(this.parsed).value}}},"./node_modules/onnxruntime-common/dist/esm/backend-impl.js":
 /*!******************************************************************!*\
   !*** ./node_modules/onnxruntime-common/dist/esm/backend-impl.js ***!
   \******************************************************************/(e,t,r)=>{r.r(t),r.d(t,{registerBackend:()=>s,resolveBackendAndExecutionProviders:()=>i});const n=new Map,o=[],s=(e,t,r)=>{if(!t||"function"!=typeof t.init||"function"!=typeof t.createInferenceSessionHandler)throw new TypeError("not a valid backend");{const s=n.get(e);if(void 0===s)n.set(e,{backend:t,priority:r});else{if(s.priority>r)return;if(s.priority===r&&s.backend!==t)throw new Error(`cannot register backend "${e}" using priority ${r}`)}if(r>=0){const t=o.indexOf(e);-1!==t&&o.splice(t,1);for(let t=0;t<o.length;t++)if(n.get(o[t]).priority<=r)return void o.splice(t,0,e);o.push(e)}}},a=async e=>{const t=n.get(e);if(!t)return"backend not found.";if(t.initialized)return t.backend;if(t.aborted)return t.error;{const r=!!t.initPromise;try{return r||(t.initPromise=t.backend.init(e)),await t.initPromise,t.initialized=!0,t.backend}catch(e){return r||(t.error=`${e}`,t.aborted=!0),t.error}finally{delete t.initPromise}}},i=async e=>{const t=e.executionProviders||[],r=t.map((e=>"string"==typeof e?e:e.name)),n=0===r.length?o:r;let s;const i=[],l=new Set;for(const e of n){const t=await a(e);"string"==typeof t?i.push({name:e,err:t}):(s||(s=t),s===t&&l.add(e))}if(!s)throw new Error(`no available backend found. ERR: ${i.map((e=>`[${e.name}] ${e.err}`)).join(", ")}`);for(const{name:e,err:t}of i)r.includes(e)&&console.warn(`removing requested execution provider "${e}" from session options because it is not available: ${t}`);const c=t.filter((e=>l.has("string"==typeof e?e:e.name)));return[s,new Proxy(e,{get:(e,t)=>"executionProviders"===t?c:Reflect.get(e,t)})]}},"./node_modules/onnxruntime-common/dist/esm/backend.js":
@@ -91,7 +91,7 @@ import*as e from"fs";import*as t from"onnxruntime-node";import*as r from"path";i
   \************************/(e,t,r)=>{r.r(t),r.d(t,{AutoConfig:()=>l,PretrainedConfig:()=>i,getKeyValueShapes:()=>a});var n=r(/*! ./utils/core.js */"./src/utils/core.js"),o=r(/*! ./utils/hub.js */"./src/utils/hub.js");function s(e){const t={};let r={};switch(e.model_type){case"llava":case"paligemma":case"florence2":r=s(e.text_config);break;case"moondream1":r=s(e.phi_config);break;case"musicgen":r=s(e.decoder);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"phi":case"phi3":case"falcon":t.num_heads="num_attention_heads",t.num_layers="num_hidden_layers",t.hidden_size="hidden_size";break;case"llama":case"granite":case"cohere":case"mistral":case"starcoder2":case"qwen2":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":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"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"vision-encoder-decoder":const o=s(e.decoder),a="num_decoder_layers"in o,i=(0,n.pick)(e,["model_type","is_encoder_decoder"]);return a?(i.num_decoder_layers=o.num_decoder_layers,i.num_decoder_heads=o.num_decoder_heads,i.decoder_hidden_size=o.decoder_hidden_size,i.num_encoder_layers=o.num_encoder_layers,i.num_encoder_heads=o.num_encoder_heads,i.encoder_hidden_size=o.encoder_hidden_size):(i.num_layers=o.num_layers,i.num_heads=o.num_heads,i.hidden_size=o.hidden_size),i}const o={...r,...(0,n.pick)(e,["model_type","multi_query","is_encoder_decoder"])};for(const r in t)o[r]=e[t[r]];return o}function a(e,{prefix:t="past_key_values"}={}){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,s=[1,n.num_encoder_heads,0,e],a=[1,n.num_decoder_heads,0,o];for(let e=0;e<n.num_decoder_layers;++e)r[`${t}.${e}.encoder.key`]=s,r[`${t}.${e}.encoder.value`]=s,r[`${t}.${e}.decoder.key`]=a,r[`${t}.${e}.decoder.value`]=a}else{const e=n.num_heads,o=n.num_layers,s=n.dim_kv??n.hidden_size/(n.num_attention_heads??e);if("falcon"===n.model_type){const n=[1*e,0,s];for(let e=0;e<o;++e)r[`${t}.${e}.key`]=n,r[`${t}.${e}.value`]=n}else if(n.multi_query){const n=[1*e,0,2*s];for(let e=0;e<o;++e)r[`${t}.${e}.key_value`]=n}else if("bloom"===n.model_type){const n=[1*e,s,0],a=[1*e,0,s];for(let e=0;e<o;++e)r[`${t}.${e}.key`]=n,r[`${t}.${e}.value`]=a}else if("openelm"===n.model_type)for(let n=0;n<o;++n){const o=[1,e[n],0,s];r[`${t}.${n}.key`]=o,r[`${t}.${n}.value`]=o}else{const n=[1,e,0,s];for(let e=0;e<o;++e)r[`${t}.${e}.key`]=n,r[`${t}.${e}.value`]=n}}return r}class i{model_type=null;is_encoder_decoder=!1;max_position_embeddings;"transformers.js_config";constructor(e){Object.assign(this,e),this.normalized_config=s(this)}static async from_pretrained(e,{progress_callback:t=null,config:r=null,cache_dir:n=null,local_files_only:s=!1,revision:a="main"}={}){!r||r instanceof i||(r=new i(r));const l=r??await async function(e,t){return await(0,o.getModelJSON)(e,"config.json",!0,t)}(e,{progress_callback:t,config:r,cache_dir:n,local_files_only:s,revision:a});return new this(l)}}class l{static async from_pretrained(...e){return i.from_pretrained(...e)}}},"./src/env.js":
 /*!********************!*\
   !*** ./src/env.js ***!
-  \********************/(e,t,r)=>{r.r(t),r.d(t,{apis:()=>_,env:()=>T});var n=r(/*! fs */"fs"),o=r(/*! path */"path"),s=r(/*! url */"url");const a="undefined"!=typeof self,i=a&&"DedicatedWorkerGlobalScope"===self.constructor.name,l=a&&"caches"in self,c="undefined"!=typeof navigator&&"gpu"in navigator,d="undefined"!=typeof navigator&&"ml"in navigator,u="undefined"!=typeof process,p=u&&"node"===process?.release?.name,h=!b(n.default),m=!b(o.default),_=Object.freeze({IS_BROWSER_ENV:a,IS_WEBWORKER_ENV:i,IS_WEB_CACHE_AVAILABLE:l,IS_WEBGPU_AVAILABLE:c,IS_WEBNN_AVAILABLE:d,IS_PROCESS_AVAILABLE:u,IS_NODE_ENV:p,IS_FS_AVAILABLE:h,IS_PATH_AVAILABLE:m}),f=h&&m,g=f?o.default.dirname(o.default.dirname(s.default.fileURLToPath(import.meta.url))):"./",M=f?o.default.join(g,"/.cache/"):null,w="/models/",T={version:"3.0.0",backends:{onnx:{}},allowRemoteModels:!0,remoteHost:"https://huggingface.co/",remotePathTemplate:"{model}/resolve/{revision}/",allowLocalModels:!a,localModelPath:f?o.default.join(g,w):w,useFS:h,useBrowserCache:l,useFSCache:h,cacheDir:M,useCustomCache:!1,customCache:null};function b(e){return 0===Object.keys(e).length}},"./src/generation/configuration_utils.js":
+  \********************/(e,t,r)=>{r.r(t),r.d(t,{apis:()=>_,env:()=>T});var n=r(/*! fs */"fs"),o=r(/*! path */"path"),s=r(/*! url */"url");const a="undefined"!=typeof self,i=a&&"DedicatedWorkerGlobalScope"===self.constructor.name,l=a&&"caches"in self,c="undefined"!=typeof navigator&&"gpu"in navigator,d="undefined"!=typeof navigator&&"ml"in navigator,u="undefined"!=typeof process,p=u&&"node"===process?.release?.name,h=!b(n.default),m=!b(o.default),_=Object.freeze({IS_BROWSER_ENV:a,IS_WEBWORKER_ENV:i,IS_WEB_CACHE_AVAILABLE:l,IS_WEBGPU_AVAILABLE:c,IS_WEBNN_AVAILABLE:d,IS_PROCESS_AVAILABLE:u,IS_NODE_ENV:p,IS_FS_AVAILABLE:h,IS_PATH_AVAILABLE:m}),f=h&&m,g=f?o.default.dirname(o.default.dirname(s.default.fileURLToPath(import.meta.url))):"./",M=f?o.default.join(g,"/.cache/"):null,w="/models/",T={version:"3.0.1",backends:{onnx:{}},allowRemoteModels:!0,remoteHost:"https://huggingface.co/",remotePathTemplate:"{model}/resolve/{revision}/",allowLocalModels:!a,localModelPath:f?o.default.join(g,w):w,useFS:h,useBrowserCache:l,useFSCache:h,cacheDir:M,useCustomCache:!1,customCache:null};function b(e){return 0===Object.keys(e).length}},"./src/generation/configuration_utils.js":
 /*!***********************************************!*\
   !*** ./src/generation/configuration_utils.js ***!
   \***********************************************/(e,t,r)=>{r.r(t),r.d(t,{GenerationConfig:()=>o});var n=r(/*! ../utils/core.js */"./src/utils/core.js");class o{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;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,n.pick)(e,Object.getOwnPropertyNames(this)))}}},"./src/generation/logits_process.js":
@@ -109,7 +109,7 @@ import*as e from"fs";import*as t from"onnxruntime-node";import*as r from"path";i
   \*************************************/(e,t,r)=>{r.r(t),r.d(t,{BaseStreamer:()=>a,TextStreamer:()=>l,WhisperTextStreamer:()=>c});var n=r(/*! ../utils/core.js */"./src/utils/core.js"),o=r(/*! ../tokenizers.js */"./src/tokenizers.js"),s=r(/*! ../env.js */"./src/env.js");class a{put(e){throw Error("Not implemented")}end(){throw Error("Not implemented")}}const i=s.apis.IS_PROCESS_AVAILABLE?e=>process.stdout.write(e):e=>console.log(e);class l extends a{constructor(e,{skip_prompt:t=!1,callback_function:r=null,token_callback_function:n=null,decode_kwargs:o={},...s}={}){super(),this.tokenizer=e,this.skip_prompt=t,this.callback_function=r??i,this.token_callback_function=n,this.decode_kwargs={...o,...s},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,n.mergeArrays)(this.token_cache,t);const r=this.tokenizer.decode(this.token_cache,this.decode_kwargs);let s;r.endsWith("\n")?(s=r.slice(this.print_len),this.token_cache=[],this.print_len=0):r.length>0&&(0,o.is_chinese_char)(r.charCodeAt(r.length-1))?(s=r.slice(this.print_len),this.print_len+=s.length):(s=r.slice(this.print_len,r.lastIndexOf(" ")+1),this.print_len+=s.length),this.on_finalized_text(s,!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===i&&s.apis.IS_PROCESS_AVAILABLE&&this.callback_function?.("\n")}}class c extends l{constructor(e,{skip_prompt:t=!1,callback_function:r=null,token_callback_function:n=null,on_chunk_start:o=null,on_chunk_end:s=null,on_finalize:a=null,time_precision:i=.02,skip_special_tokens:l=!0,decode_kwargs:c={}}={}){super(e,{skip_prompt:t,callback_function:r,token_callback_function:n,decode_kwargs:{skip_special_tokens:l,...c}}),this.timestamp_begin=e.timestamp_begin,this.on_chunk_start=o,this.on_chunk_end=s,this.on_finalize=a,this.time_precision=i,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 r=Number(t[0])-this.timestamp_begin;if(r>=0){const t=r*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":
 /*!***********************!*\
   !*** ./src/models.js ***!
-  \***********************/(e,t,r)=>{r.r(t),r.d(t,{ASTForAudioClassification:()=>or,ASTModel:()=>nr,ASTPreTrainedModel:()=>rr,AlbertForMaskedLM:()=>ht,AlbertForQuestionAnswering:()=>pt,AlbertForSequenceClassification:()=>ut,AlbertModel:()=>dt,AlbertPreTrainedModel:()=>ct,AutoModel:()=>Oi,AutoModelForAudioClassification:()=>nl,AutoModelForAudioFrameClassification:()=>sl,AutoModelForCTC:()=>rl,AutoModelForCausalLM:()=>$i,AutoModelForDepthEstimation:()=>cl,AutoModelForDocumentQuestionAnswering:()=>al,AutoModelForImageClassification:()=>Hi,AutoModelForImageFeatureExtraction:()=>ul,AutoModelForImageMatting:()=>il,AutoModelForImageSegmentation:()=>Yi,AutoModelForImageToImage:()=>ll,AutoModelForMaskGeneration:()=>tl,AutoModelForMaskedLM:()=>Ui,AutoModelForNormalEstimation:()=>dl,AutoModelForObjectDetection:()=>Zi,AutoModelForQuestionAnswering:()=>Xi,AutoModelForSemanticSegmentation:()=>Ji,AutoModelForSeq2SeqLM:()=>Ri,AutoModelForSequenceClassification:()=>Vi,AutoModelForSpeechSeq2Seq:()=>Gi,AutoModelForTextToSpectrogram:()=>qi,AutoModelForTextToWaveform:()=>Wi,AutoModelForTokenClassification:()=>ji,AutoModelForUniversalSegmentation:()=>Ki,AutoModelForVision2Seq:()=>Qi,AutoModelForXVector:()=>ol,AutoModelForZeroShotObjectDetection:()=>el,BartForConditionalGeneration:()=>Ft,BartForSequenceClassification:()=>Pt,BartModel:()=>kt,BartPretrainedModel:()=>yt,BaseModelOutput:()=>$,BeitForImageClassification:()=>po,BeitModel:()=>uo,BeitPreTrainedModel:()=>co,BertForMaskedLM:()=>Q,BertForQuestionAnswering:()=>J,BertForSequenceClassification:()=>H,BertForTokenClassification:()=>Y,BertModel:()=>X,BertPreTrainedModel:()=>U,BlenderbotForConditionalGeneration:()=>It,BlenderbotModel:()=>zt,BlenderbotPreTrainedModel:()=>Lt,BlenderbotSmallForConditionalGeneration:()=>Dt,BlenderbotSmallModel:()=>Nt,BlenderbotSmallPreTrainedModel:()=>Bt,BloomForCausalLM:()=>Pn,BloomModel:()=>Fn,BloomPreTrainedModel:()=>kn,CLIPModel:()=>_r,CLIPPreTrainedModel:()=>mr,CLIPSegForImageSegmentation:()=>vr,CLIPSegModel:()=>Cr,CLIPSegPreTrainedModel:()=>Pr,CLIPTextModel:()=>fr,CLIPTextModelWithProjection:()=>gr,CLIPVisionModel:()=>Mr,CLIPVisionModelWithProjection:()=>wr,CamembertForMaskedLM:()=>Te,CamembertForQuestionAnswering:()=>ye,CamembertForSequenceClassification:()=>be,CamembertForTokenClassification:()=>xe,CamembertModel:()=>we,CamembertPreTrainedModel:()=>Me,CausalLMOutput:()=>Ml,CausalLMOutputWithPast:()=>wl,ChineseCLIPModel:()=>Fr,ChineseCLIPPreTrainedModel:()=>kr,ClapAudioModelWithProjection:()=>Ca,ClapModel:()=>Fa,ClapPreTrainedModel:()=>ka,ClapTextModelWithProjection:()=>Pa,CodeGenForCausalLM:()=>Hr,CodeGenModel:()=>Qr,CodeGenPreTrainedModel:()=>Xr,CohereForCausalLM:()=>on,CohereModel:()=>nn,CoherePreTrainedModel:()=>rn,ConvBertForMaskedLM:()=>le,ConvBertForQuestionAnswering:()=>ue,ConvBertForSequenceClassification:()=>ce,ConvBertForTokenClassification:()=>de,ConvBertModel:()=>ie,ConvBertPreTrainedModel:()=>ae,ConvNextForImageClassification:()=>cs,ConvNextModel:()=>ls,ConvNextPreTrainedModel:()=>is,ConvNextV2ForImageClassification:()=>ps,ConvNextV2Model:()=>us,ConvNextV2PreTrainedModel:()=>ds,DPTForDepthEstimation:()=>Wo,DPTModel:()=>qo,DPTPreTrainedModel:()=>Go,DebertaForMaskedLM:()=>Pe,DebertaForQuestionAnswering:()=>Se,DebertaForSequenceClassification:()=>Ce,DebertaForTokenClassification:()=>ve,DebertaModel:()=>Fe,DebertaPreTrainedModel:()=>ke,DebertaV2ForMaskedLM:()=>Le,DebertaV2ForQuestionAnswering:()=>Be,DebertaV2ForSequenceClassification:()=>ze,DebertaV2ForTokenClassification:()=>Ie,DebertaV2Model:()=>Ee,DebertaV2PreTrainedModel:()=>Ae,DecisionTransformerModel:()=>ni,DecisionTransformerPreTrainedModel:()=>ri,DeiTForImageClassification:()=>So,DeiTModel:()=>vo,DeiTPreTrainedModel:()=>Co,DepthAnythingForDepthEstimation:()=>Uo,DepthAnythingPreTrainedModel:()=>$o,DepthProForDepthEstimation:()=>Ko,DepthProPreTrainedModel:()=>Jo,DetrForObjectDetection:()=>_o,DetrForSegmentation:()=>fo,DetrModel:()=>mo,DetrObjectDetectionOutput:()=>go,DetrPreTrainedModel:()=>ho,DetrSegmentationOutput:()=>Mo,Dinov2ForImageClassification:()=>_s,Dinov2Model:()=>ms,Dinov2PreTrainedModel:()=>hs,DistilBertForMaskedLM:()=>Re,DistilBertForQuestionAnswering:()=>je,DistilBertForSequenceClassification:()=>Oe,DistilBertForTokenClassification:()=>Ve,DistilBertModel:()=>De,DistilBertPreTrainedModel:()=>Ne,DonutSwinModel:()=>as,DonutSwinPreTrainedModel:()=>ss,EfficientNetForImageClassification:()=>Va,EfficientNetModel:()=>Oa,EfficientNetPreTrainedModel:()=>Da,ElectraForMaskedLM:()=>me,ElectraForQuestionAnswering:()=>ge,ElectraForSequenceClassification:()=>_e,ElectraForTokenClassification:()=>fe,ElectraModel:()=>he,ElectraPreTrainedModel:()=>pe,EsmForMaskedLM:()=>We,EsmForSequenceClassification:()=>$e,EsmForTokenClassification:()=>Ue,EsmModel:()=>qe,EsmPreTrainedModel:()=>Ge,FalconForCausalLM:()=>ya,FalconModel:()=>xa,FalconPreTrainedModel:()=>ba,FastViTForImageClassification:()=>Qn,FastViTModel:()=>Xn,FastViTPreTrainedModel:()=>Un,Florence2ForConditionalGeneration:()=>hr,Florence2PreTrainedModel:()=>pr,GLPNForDepthEstimation:()=>os,GLPNModel:()=>ns,GLPNPreTrainedModel:()=>rs,GPT2LMHeadModel:()=>Er,GPT2Model:()=>Ar,GPT2PreTrainedModel:()=>Sr,GPTBigCodeForCausalLM:()=>Ur,GPTBigCodeModel:()=>$r,GPTBigCodePreTrainedModel:()=>Wr,GPTJForCausalLM:()=>qr,GPTJModel:()=>Gr,GPTJPreTrainedModel:()=>Rr,GPTNeoForCausalLM:()=>Dr,GPTNeoModel:()=>Nr,GPTNeoPreTrainedModel:()=>Br,GPTNeoXForCausalLM:()=>jr,GPTNeoXModel:()=>Vr,GPTNeoXPreTrainedModel:()=>Or,Gemma2ForCausalLM:()=>un,Gemma2Model:()=>dn,Gemma2PreTrainedModel:()=>cn,GemmaForCausalLM:()=>ln,GemmaModel:()=>an,GemmaPreTrainedModel:()=>sn,GraniteForCausalLM:()=>tn,GraniteModel:()=>en,GranitePreTrainedModel:()=>Zr,GroupViTModel:()=>$n,GroupViTPreTrainedModel:()=>Wn,HieraForImageClassification:()=>Lo,HieraModel:()=>Eo,HieraPreTrainedModel:()=>Ao,HubertForCTC:()=>ea,HubertForSequenceClassification:()=>ta,HubertModel:()=>Zs,HubertPreTrainedModel:()=>Ks,ImageMattingOutput:()=>Tl,JAISLMHeadModel:()=>Ir,JAISModel:()=>zr,JAISPreTrainedModel:()=>Lr,LlamaForCausalLM:()=>Kr,LlamaModel:()=>Jr,LlamaPreTrainedModel:()=>Yr,LlavaForConditionalGeneration:()=>dr,LlavaPreTrainedModel:()=>cr,LongT5ForConditionalGeneration:()=>wt,LongT5Model:()=>Mt,LongT5PreTrainedModel:()=>gt,M2M100ForConditionalGeneration:()=>vs,M2M100Model:()=>Cs,M2M100PreTrainedModel:()=>Ps,MBartForCausalLM:()=>Et,MBartForConditionalGeneration:()=>St,MBartForSequenceClassification:()=>At,MBartModel:()=>vt,MBartPreTrainedModel:()=>Ct,MPNetForMaskedLM:()=>et,MPNetForQuestionAnswering:()=>nt,MPNetForSequenceClassification:()=>tt,MPNetForTokenClassification:()=>rt,MPNetModel:()=>Ze,MPNetPreTrainedModel:()=>Ke,MT5ForConditionalGeneration:()=>xt,MT5Model:()=>bt,MT5PreTrainedModel:()=>Tt,MarianMTModel:()=>Fs,MarianModel:()=>ks,MarianPreTrainedModel:()=>ys,MaskFormerForInstanceSegmentation:()=>ts,MaskFormerModel:()=>es,MaskFormerPreTrainedModel:()=>Zo,MaskedLMOutput:()=>fl,MistralForCausalLM:()=>ga,MistralModel:()=>fa,MistralPreTrainedModel:()=>_a,MobileBertForMaskedLM:()=>He,MobileBertForQuestionAnswering:()=>Je,MobileBertForSequenceClassification:()=>Ye,MobileBertModel:()=>Qe,MobileBertPreTrainedModel:()=>Xe,MobileNetV1ForImageClassification:()=>Ua,MobileNetV1Model:()=>$a,MobileNetV1PreTrainedModel:()=>Wa,MobileNetV2ForImageClassification:()=>Ha,MobileNetV2Model:()=>Qa,MobileNetV2PreTrainedModel:()=>Xa,MobileNetV3ForImageClassification:()=>Ka,MobileNetV3Model:()=>Ja,MobileNetV3PreTrainedModel:()=>Ya,MobileNetV4ForImageClassification:()=>ti,MobileNetV4Model:()=>ei,MobileNetV4PreTrainedModel:()=>Za,MobileViTForImageClassification:()=>Zn,MobileViTModel:()=>Kn,MobileViTPreTrainedModel:()=>Jn,MobileViTV2ForImageClassification:()=>ro,MobileViTV2Model:()=>to,MobileViTV2PreTrainedModel:()=>eo,ModelOutput:()=>W,Moondream1ForConditionalGeneration:()=>ur,MptForCausalLM:()=>Sn,MptModel:()=>vn,MptPreTrainedModel:()=>Cn,MusicgenForCausalLM:()=>Ga,MusicgenForConditionalGeneration:()=>qa,MusicgenModel:()=>Ra,MusicgenPreTrainedModel:()=>ja,NomicBertModel:()=>Z,NomicBertPreTrainedModel:()=>K,OPTForCausalLM:()=>Ln,OPTModel:()=>En,OPTPreTrainedModel:()=>An,OpenELMForCausalLM:()=>mn,OpenELMModel:()=>hn,OpenELMPreTrainedModel:()=>pn,OwlViTForObjectDetection:()=>so,OwlViTModel:()=>oo,OwlViTPreTrainedModel:()=>no,Owlv2ForObjectDetection:()=>lo,Owlv2Model:()=>io,Owlv2PreTrainedModel:()=>ao,Phi3ForCausalLM:()=>yn,Phi3Model:()=>xn,Phi3PreTrainedModel:()=>bn,PhiForCausalLM:()=>Tn,PhiModel:()=>wn,PhiPreTrainedModel:()=>Mn,PreTrainedModel:()=>q,PretrainedMixin:()=>oi,PvtForImageClassification:()=>On,PvtModel:()=>Dn,PvtPreTrainedModel:()=>Nn,PyAnnoteForAudioFrameClassification:()=>Ns,PyAnnoteModel:()=>Bs,PyAnnotePreTrainedModel:()=>Is,QuestionAnsweringModelOutput:()=>gl,Qwen2ForCausalLM:()=>gn,Qwen2Model:()=>fn,Qwen2PreTrainedModel:()=>_n,RTDetrForObjectDetection:()=>bo,RTDetrModel:()=>To,RTDetrObjectDetectionOutput:()=>xo,RTDetrPreTrainedModel:()=>wo,ResNetForImageClassification:()=>Bo,ResNetModel:()=>Io,ResNetPreTrainedModel:()=>zo,RoFormerForMaskedLM:()=>re,RoFormerForQuestionAnswering:()=>se,RoFormerForSequenceClassification:()=>ne,RoFormerForTokenClassification:()=>oe,RoFormerModel:()=>te,RoFormerPreTrainedModel:()=>ee,RobertaForMaskedLM:()=>jt,RobertaForQuestionAnswering:()=>qt,RobertaForSequenceClassification:()=>Rt,RobertaForTokenClassification:()=>Gt,RobertaModel:()=>Vt,RobertaPreTrainedModel:()=>Ot,SamImageSegmentationOutput:()=>xs,SamModel:()=>bs,SamPreTrainedModel:()=>Ts,SapiensForDepthEstimation:()=>Ho,SapiensForNormalEstimation:()=>Yo,SapiensForSemanticSegmentation:()=>Qo,SapiensPreTrainedModel:()=>Xo,SegformerForImageClassification:()=>La,SegformerForSemanticSegmentation:()=>za,SegformerModel:()=>Ea,SegformerPreTrainedModel:()=>Aa,Seq2SeqLMOutput:()=>pl,SequenceClassifierOutput:()=>hl,SiglipModel:()=>br,SiglipPreTrainedModel:()=>Tr,SiglipTextModel:()=>xr,SiglipVisionModel:()=>yr,SpeechT5ForSpeechToText:()=>da,SpeechT5ForTextToSpeech:()=>ua,SpeechT5HifiGan:()=>pa,SpeechT5Model:()=>ca,SpeechT5PreTrainedModel:()=>la,SqueezeBertForMaskedLM:()=>at,SqueezeBertForQuestionAnswering:()=>lt,SqueezeBertForSequenceClassification:()=>it,SqueezeBertModel:()=>st,SqueezeBertPreTrainedModel:()=>ot,StableLmForCausalLM:()=>Na,StableLmModel:()=>Ba,StableLmPreTrainedModel:()=>Ia,Starcoder2ForCausalLM:()=>Ta,Starcoder2Model:()=>wa,Starcoder2PreTrainedModel:()=>Ma,Swin2SRForImageSuperResolution:()=>Ro,Swin2SRModel:()=>jo,Swin2SRPreTrainedModel:()=>Vo,SwinForImageClassification:()=>Oo,SwinModel:()=>Do,SwinPreTrainedModel:()=>No,T5ForConditionalGeneration:()=>ft,T5Model:()=>_t,T5PreTrainedModel:()=>mt,TableTransformerForObjectDetection:()=>Fo,TableTransformerModel:()=>ko,TableTransformerObjectDetectionOutput:()=>Po,TableTransformerPreTrainedModel:()=>yo,TokenClassifierOutput:()=>_l,TrOCRForCausalLM:()=>ma,TrOCRPreTrainedModel:()=>ha,UniSpeechForCTC:()=>Rs,UniSpeechForSequenceClassification:()=>Gs,UniSpeechModel:()=>js,UniSpeechPreTrainedModel:()=>Vs,UniSpeechSatForAudioFrameClassification:()=>Xs,UniSpeechSatForCTC:()=>$s,UniSpeechSatForSequenceClassification:()=>Us,UniSpeechSatModel:()=>Ws,UniSpeechSatPreTrainedModel:()=>qs,ViTForImageClassification:()=>Bn,ViTMAEModel:()=>jn,ViTMAEPreTrainedModel:()=>Vn,ViTMSNForImageClassification:()=>qn,ViTMSNModel:()=>Gn,ViTMSNPreTrainedModel:()=>Rn,ViTModel:()=>In,ViTPreTrainedModel:()=>zn,VisionEncoderDecoderModel:()=>lr,VitMatteForImageMatting:()=>Yn,VitMattePreTrainedModel:()=>Hn,VitsModel:()=>Sa,VitsModelOutput:()=>bl,VitsPreTrainedModel:()=>va,Wav2Vec2BertForCTC:()=>Ys,Wav2Vec2BertForSequenceClassification:()=>Js,Wav2Vec2BertModel:()=>Hs,Wav2Vec2BertPreTrainedModel:()=>Qs,Wav2Vec2ForAudioFrameClassification:()=>zs,Wav2Vec2ForCTC:()=>Es,Wav2Vec2ForSequenceClassification:()=>Ls,Wav2Vec2Model:()=>As,Wav2Vec2PreTrainedModel:()=>Ss,WavLMForAudioFrameClassification:()=>ia,WavLMForCTC:()=>oa,WavLMForSequenceClassification:()=>sa,WavLMForXVector:()=>aa,WavLMModel:()=>na,WavLMPreTrainedModel:()=>ra,WeSpeakerResNetModel:()=>Os,WeSpeakerResNetPreTrainedModel:()=>Ds,WhisperForConditionalGeneration:()=>ir,WhisperModel:()=>ar,WhisperPreTrainedModel:()=>sr,XLMForQuestionAnswering:()=>Ht,XLMForSequenceClassification:()=>Xt,XLMForTokenClassification:()=>Qt,XLMModel:()=>$t,XLMPreTrainedModel:()=>Wt,XLMRobertaForMaskedLM:()=>Kt,XLMRobertaForQuestionAnswering:()=>tr,XLMRobertaForSequenceClassification:()=>Zt,XLMRobertaForTokenClassification:()=>er,XLMRobertaModel:()=>Jt,XLMRobertaPreTrainedModel:()=>Yt,XLMWithLMHeadModel:()=>Ut,XVectorOutput:()=>ml,YolosForObjectDetection:()=>Ms,YolosModel:()=>gs,YolosObjectDetectionOutput:()=>ws,YolosPreTrainedModel:()=>fs});var n=r(/*! ./configs.js */"./src/configs.js"),o=r(/*! ./backends/onnx.js */"./src/backends/onnx.js"),s=r(/*! ./utils/dtypes.js */"./src/utils/dtypes.js"),a=r(/*! ./utils/generic.js */"./src/utils/generic.js"),i=r(/*! ./utils/core.js */"./src/utils/core.js"),l=r(/*! ./utils/hub.js */"./src/utils/hub.js"),c=r(/*! ./utils/constants.js */"./src/utils/constants.js"),d=r(/*! ./generation/logits_process.js */"./src/generation/logits_process.js"),u=r(/*! ./generation/configuration_utils.js */"./src/generation/configuration_utils.js"),p=r(/*! ./utils/tensor.js */"./src/utils/tensor.js"),h=r(/*! ./utils/maths.js */"./src/utils/maths.js"),m=r(/*! ./generation/stopping_criteria.js */"./src/generation/stopping_criteria.js"),_=r(/*! ./generation/logits_sampler.js */"./src/generation/logits_sampler.js"),f=r(/*! ./env.js */"./src/env.js"),g=r(/*! ./models/whisper/generation_whisper.js */"./src/models/whisper/generation_whisper.js"),M=r(/*! ./models/whisper/common_whisper.js */"./src/models/whisper/common_whisper.js");const w=0,T=1,b=2,x=3,y=4,k=5,F=6,P=7,C=new Map,v=new Map,S=new Map;async function A(e,t,r){return Object.fromEntries(await Promise.all(Object.keys(t).map((async a=>{const{buffer:i,session_options:c,session_config:d}=await async function(e,t,r){const a=r.config?.["transformers.js_config"]??{};let i=r.device??a.device;i&&"string"!=typeof i&&(i.hasOwnProperty(t)?i=i[t]:(console.warn(`device not specified for "${t}". Using the default device.`),i=null));const c=i??(f.apis.IS_NODE_ENV?"cpu":"wasm"),d=(0,o.deviceToExecutionProviders)(c);let u=r.dtype??a.dtype;"string"!=typeof u&&(u&&u.hasOwnProperty(t)?u=u[t]:(u=s.DEFAULT_DEVICE_DTYPE_MAPPING[c]??s.DATA_TYPES.fp32,console.warn(`dtype not specified for "${t}". Using the default dtype (${u}) for this device (${c}).`)));const p=u;if(!s.DEFAULT_DTYPE_SUFFIX_MAPPING.hasOwnProperty(p))throw new Error(`Invalid dtype: ${p}. Should be one of: ${Object.keys(s.DATA_TYPES).join(", ")}`);if(p===s.DATA_TYPES.fp16&&"webgpu"===c&&!await(0,s.isWebGpuFp16Supported)())throw new Error(`The device (${c}) does not support fp16.`);const h=a.kv_cache_dtype?"string"==typeof a.kv_cache_dtype?a.kv_cache_dtype:a.kv_cache_dtype[p]??"float32":void 0;if(h&&!["float32","float16"].includes(h))throw new Error(`Invalid kv_cache_dtype: ${h}. Should be one of: float32, float16`);const m={dtype:p,kv_cache_dtype:h},_=s.DEFAULT_DTYPE_SUFFIX_MAPPING[p],g=`${r.subfolder??""}/${t}${_}.onnx`,M={...r.session_options};M.executionProviders??=d;const w=a.free_dimension_overrides;w?M.freeDimensionOverrides??=w:c.startsWith("webnn")&&!M.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 T=(0,l.getModelFile)(e,g,!0,r),b=r.use_external_data_format??a.use_external_data_format;let x=[];if(b&&(!0===b||"object"==typeof b&&b.hasOwnProperty(t)&&!0===b[t])){if(f.apis.IS_NODE_ENV)throw new Error("External data format is not yet supported in Node.js");const n=`${t}${_}.onnx_data`,o=`${r.subfolder??""}/${n}`;x.push(new Promise((async(t,s)=>{const a=await(0,l.getModelFile)(e,o,!0,r);t({path:n,data:a})})))}else void 0!==M.externalData&&(x=M.externalData.map((async t=>{if("string"==typeof t.data){const n=await(0,l.getModelFile)(e,t.data,!0,r);return{...t,data:n}}return t})));if(x.length>0&&(M.externalData=await Promise.all(x)),"webgpu"===c){const e=(0,n.getKeyValueShapes)(r.config,{prefix:"present"});if(Object.keys(e).length>0&&!(0,o.isONNXProxy)()){const t={};for(const r in e)t[r]="gpu-buffer";M.preferredOutputLocation=t}}return{buffer:await T,session_options:M,session_config:m}}(e,t[a],r);return[a,await(0,o.createInferenceSession)(i,c,d)]}))))}async function E(e,t,r){return Object.fromEntries(await Promise.all(Object.keys(t).map((async n=>[n,await(0,l.getModelJSON)(e,t[n],!1,r)]))))}async function L(e,t){const r=function(e,t){const r=Object.create(null),n=[];for(const s of e.inputNames){const e=t[s];e instanceof p.Tensor?r[s]=(0,o.isONNXProxy)()?e.clone():e:n.push(s)}if(n.length>0)throw new Error(`An error occurred during model execution: "Missing the following inputs: ${n.join(", ")}.`);const s=Object.keys(t).length,a=e.inputNames.length;if(s>a){let r=Object.keys(t).filter((t=>!e.inputNames.includes(t)));console.warn(`WARNING: Too many inputs were provided (${s} > ${a}). The following inputs will be ignored: "${r.join(", ")}".`)}return r}(e,t);try{const t=Object.fromEntries(Object.entries(r).map((([e,t])=>[e,t.ort_tensor])));let n=await e.run(t);return n=z(n),n}catch(e){throw console.error(`An error occurred during model execution: "${e}".`),console.error("Inputs given to model:",r),e}}function z(e){for(let t in e)(0,o.isONNXTensor)(e[t])?e[t]=new p.Tensor(e[t]):"object"==typeof e[t]&&z(e[t]);return e}function I(e){if(e instanceof p.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 p.Tensor("int64",BigInt64Array.from(e.flat().map((e=>BigInt(e)))),[e.length,e[0].length])}return new p.Tensor("int64",BigInt64Array.from(e.map((e=>BigInt(e)))),[1,e.length])}function B(e){return new p.Tensor("bool",[e],[1])}async function N(e,t){let{encoder_outputs:r,input_ids:n,decoder_input_ids:o,...s}=t;if(!r){const n=(0,i.pick)(t,e.sessions.model.inputNames);r=(await D(e,n)).last_hidden_state}s.input_ids=o,s.encoder_hidden_states=r,e.sessions.decoder_model_merged.inputNames.includes("encoder_attention_mask")&&(s.encoder_attention_mask=t.attention_mask);return await O(e,s,!0)}async function D(e,t){const r=e.sessions.model,n=(0,i.pick)(t,r.inputNames);if(r.inputNames.includes("inputs_embeds")&&!n.inputs_embeds){if(!t.input_ids)throw new Error("Both `input_ids` and `inputs_embeds` are missing in the model inputs.");n.inputs_embeds=await e.encode_text({input_ids:t.input_ids})}return r.inputNames.includes("token_type_ids")&&!n.token_type_ids&&(n.token_type_ids=new p.Tensor("int64",new BigInt64Array(n.input_ids.data.length),n.input_ids.dims)),await L(r,n)}async function O(e,t,r=!1){const n=e.sessions[r?"decoder_model_merged":"model"],{past_key_values:o,...s}=t;n.inputNames.includes("use_cache_branch")&&(s.use_cache_branch=B(!!o)),n.inputNames.includes("position_ids")&&s.attention_mask&&!s.position_ids&&(s.position_ids=function(e,t=null){const{input_ids:r,inputs_embeds:n,attention_mask:o}=e,[s,a]=o.dims,i=new BigInt64Array(o.data.length);for(let e=0;e<s;++e){const t=e*a;let r=BigInt(0);for(let e=0;e<a;++e){const n=t+e;0n===o.data[n]?i[n]=BigInt(1):(i[n]=r,r+=o.data[n])}}let l=new p.Tensor("int64",i,o.dims);if(t){const e=-(r??n).dims.at(1);l=l.slice(null,[e,null])}return l}(s,o)),e.addPastKeyValues(s,o);const a=(0,i.pick)(s,n.inputNames);return await L(n,a)}async function V(e,{input_ids:t=null,attention_mask:r=null,pixel_values:n=null,position_ids:o=null,inputs_embeds:s=null,past_key_values:a=null,generation_config:i=null,logits_processor:l=null,...c}){if(!s)if(s=await e.encode_text({input_ids:t}),n&&1!==t.dims[1]){const o=await e.encode_image({pixel_values:n});({inputs_embeds:s,attention_mask:r}=e._merge_input_ids_with_image_features({image_features:o,inputs_embeds:s,input_ids:t,attention_mask:r}))}else if(a&&n&&1===t.dims[1]){const e=t.dims[1],n=Object.values(a)[0].dims.at(-2);r=(0,p.cat)([(0,p.ones)([t.dims[0],n]),r.slice(null,[r.dims[1]-e,r.dims[1]])],1)}return await O(e,{inputs_embeds:s,past_key_values:a,attention_mask:r,position_ids:o,generation_config:i,logits_processor:l},!0)}function j(e,t,r,n){if(r.past_key_values){const t=Object.values(r.past_key_values)[0].dims.at(-2),{input_ids:n,attention_mask:o}=r;if(o&&o.dims[1]>n.dims[1]);else if(t<n.dims[1])r.input_ids=n.slice(null,[t,null]);else if(null!=e.config.image_token_index&&n.data.some((t=>t==e.config.image_token_index))){const o=e.config.num_image_tokens;if(!o)throw new Error("`num_image_tokens` is missing in the model configuration.");const s=n.dims[1]-(t-o);r.input_ids=n.slice(null,[-s,null]),r.attention_mask=(0,p.ones)([1,t+s])}}return r}function R(e,t,r,n){return r.past_key_values&&(t=t.map((e=>[e.at(-1)]))),{...r,decoder_input_ids:I(t)}}function G(e,...t){return e.config.is_encoder_decoder?R(e,...t):j(e,...t)}class q extends a.Callable{main_input_name="input_ids";forward_params=["input_ids","attention_mask"];constructor(e,t,r){super(),this.config=e,this.sessions=t,this.configs=r;const n=S.get(this.constructor),o=C.get(n);switch(this.can_generate=!1,this._forward=null,this._prepare_inputs_for_generation=null,o){case y:this.can_generate=!0,this._forward=O,this._prepare_inputs_for_generation=j;break;case b:case x:case P:this.can_generate=!0,this._forward=N,this._prepare_inputs_for_generation=R;break;case T:this._forward=N;break;case F:this.can_generate=!0,this._forward=V,this._prepare_inputs_for_generation=G;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:r=null,cache_dir:o=null,local_files_only:s=!1,revision:a="main",model_file_name:i=null,subfolder:l="onnx",device:d=null,dtype:u=null,use_external_data_format:p=null,session_options:h={}}={}){let m={progress_callback:t,config:r,cache_dir:o,local_files_only:s,revision:a,model_file_name:i,subfolder:l,device:d,dtype:u,use_external_data_format:p,session_options:h};const _=S.get(this),f=C.get(_);let g;if(r=m.config=await n.AutoConfig.from_pretrained(e,m),f===y)g=await Promise.all([A(e,{model:m.model_file_name??"model"},m),E(e,{generation_config:"generation_config.json"},m)]);else if(f===b||f===x)g=await Promise.all([A(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},m),E(e,{generation_config:"generation_config.json"},m)]);else if(f===k)g=await Promise.all([A(e,{model:"vision_encoder",prompt_encoder_mask_decoder:"prompt_encoder_mask_decoder"},m)]);else if(f===T)g=await Promise.all([A(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},m)]);else if(f===F){const t={embed_tokens:"embed_tokens",vision_encoder:"vision_encoder",decoder_model_merged:"decoder_model_merged"};r.is_encoder_decoder&&(t.model="encoder_model"),g=await Promise.all([A(e,t,m),E(e,{generation_config:"generation_config.json"},m)])}else f===P?g=await Promise.all([A(e,{model:"text_encoder",decoder_model_merged:"decoder_model_merged",encodec_decode:"encodec_decode"},m),E(e,{generation_config:"generation_config.json"},m)]):(f!==w&&console.warn(`Model type for '${_??r?.model_type}' not found, assuming encoder-only architecture. Please report this at ${c.GITHUB_ISSUE_URL}.`),g=await Promise.all([A(e,{model:m.model_file_name??"model"},m)]));return new this(r,...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,r=null){const n=new d.LogitsProcessorList;if(null!==e.repetition_penalty&&1!==e.repetition_penalty&&n.push(new d.RepetitionPenaltyLogitsProcessor(e.repetition_penalty)),null!==e.no_repeat_ngram_size&&e.no_repeat_ngram_size>0&&n.push(new d.NoRepeatNGramLogitsProcessor(e.no_repeat_ngram_size)),null!==e.bad_words_ids&&n.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&&n.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&&n.push(new d.MinNewTokensLengthLogitsProcessor(t,e.min_new_tokens,e.eos_token_id)),null!==e.forced_bos_token_id&&n.push(new d.ForcedBOSTokenLogitsProcessor(e.forced_bos_token_id)),null!==e.forced_eos_token_id&&n.push(new d.ForcedEOSTokenLogitsProcessor(e.max_length,e.forced_eos_token_id)),null!==e.begin_suppress_tokens){const r=t>1||null===e.forced_bos_token_id?t:t+1;n.push(new d.SuppressTokensAtBeginLogitsProcessor(e.begin_suppress_tokens,r))}return null!==e.guidance_scale&&e.guidance_scale>1&&n.push(new d.ClassifierFreeGuidanceLogitsProcessor(e.guidance_scale)),null!==r&&n.extend(r),n}_prepare_generation_config(e,t,r=u.GenerationConfig){const n={...this.config};for(const e of["decoder","generator","text_config"])e in n&&Object.assign(n,n[e]);const o=new r(n);return Object.assign(o,this.generation_config??{}),e&&Object.assign(o,e),t&&Object.assign(o,(0,i.pick)(t,Object.getOwnPropertyNames(o))),o}_get_stopping_criteria(e,t=null){const r=new m.StoppingCriteriaList;return null!==e.max_length&&r.push(new m.MaxLengthCriteria(e.max_length,this.config.max_position_embeddings??null)),null!==e.eos_token_id&&r.push(new m.EosTokenCriteria(e.eos_token_id)),t&&r.extend(t),r}_validate_model_class(){if(!this.can_generate){const e=[mi,gi,hi,li],t=S.get(this.constructor),r=new Set,n=this.config.model_type;for(const t of e){const e=t.get(n);e&&r.add(e[0])}let o=`The current model class (${t}) is not compatible with \`.generate()\`, as it doesn't have a language model head.`;throw r.size>0&&(o+=` Please use the following class instead: ${[...r].join(", ")}`),Error(o)}}prepare_inputs_for_generation(...e){return this._prepare_inputs_for_generation(this,...e)}_update_model_kwargs_for_generation({generated_input_ids:e,outputs:t,model_inputs:r,is_encoder_decoder:n}){return r.past_key_values=this.getPastKeyValues(t,r.past_key_values),r.input_ids=new p.Tensor("int64",e.flat(),[e.length,1]),n||(r.attention_mask=(0,p.cat)([r.attention_mask,(0,p.ones)([r.attention_mask.dims[0],1])],1)),r.position_ids=null,r}_prepare_model_inputs({inputs:e,bos_token_id:t,model_kwargs:r}){const n=(0,i.pick)(r,this.forward_params),o=this.main_input_name;if(o in n){if(e)throw new Error("`inputs`: {inputs}` were passed alongside {input_name} which is not allowed. Make sure to either pass {inputs} or {input_name}=...")}else n[o]=e;return{inputs_tensor:n[o],model_inputs:n,model_input_name:o}}async _prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:e,model_inputs:t,model_input_name:r,generation_config:n}){if(this.sessions.model.inputNames.includes("inputs_embeds")&&!t.inputs_embeds&&"_prepare_inputs_embeds"in this){const{input_ids:e,pixel_values:r,attention_mask:n,...o}=t,s=await this._prepare_inputs_embeds(t);t={...o,...(0,i.pick)(s,["inputs_embeds","attention_mask"])}}let{last_hidden_state:o}=await D(this,t);if(null!==n.guidance_scale&&n.guidance_scale>1)o=(0,p.cat)([o,(0,p.full_like)(o,0)],0),"attention_mask"in t&&(t.attention_mask=(0,p.cat)([t.attention_mask,(0,p.zeros_like)(t.attention_mask)],0));else if(t.decoder_input_ids){const e=I(t.decoder_input_ids).dims[0];if(e!==o.dims[0]){if(1!==o.dims[0])throw new Error(`The encoder outputs have a different batch size (${o.dims[0]}) than the decoder inputs (${e}).`);o=(0,p.cat)(Array.from({length:e},(()=>o)),0)}}return t.encoder_outputs=o,t}_prepare_decoder_input_ids_for_generation({batch_size:e,model_input_name:t,model_kwargs:r,decoder_start_token_id:n,bos_token_id:o,generation_config:s}){let{decoder_input_ids:a,...i}=r;if(a)Array.isArray(a[0])||(a=Array.from({length:e},(()=>a)));else if(n??=o,"musicgen"===this.config.model_type)a=Array.from({length:e*this.config.decoder.num_codebooks},(()=>[n]));else if(Array.isArray(n)){if(n.length!==e)throw new Error(`\`decoder_start_token_id\` expcted to have length ${e} but got ${n.length}`);a=n}else a=Array.from({length:e},(()=>[n]));return a=I(a),r.decoder_attention_mask=(0,p.ones_like)(a),{input_ids:a,model_inputs:i}}async generate({inputs:e=null,generation_config:t=null,logits_processor:r=null,stopping_criteria:n=null,streamer:o=null,...s}){this._validate_model_class(),t=this._prepare_generation_config(t,s);let{inputs_tensor:a,model_inputs:i,model_input_name:l}=this._prepare_model_inputs({inputs:e,model_kwargs:s});const c=this.config.is_encoder_decoder;let d;c&&("encoder_outputs"in i||(i=await this._prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:a,model_inputs:i,model_input_name:l,generation_config:t}))),c?({input_ids:d,model_inputs:i}=this._prepare_decoder_input_ids_for_generation({batch_size:i[l].dims.at(0),model_input_name:l,model_kwargs:i,decoder_start_token_id:t.decoder_start_token_id,bos_token_id:t.bos_token_id,generation_config:t})):d=i[l];let u=d.dims.at(-1);null!==t.max_new_tokens&&(t.max_length=u+t.max_new_tokens);const h=this._get_logits_processor(t,u,r),m=this._get_stopping_criteria(t,n),f=i[l].dims.at(0),g=_.LogitsSampler.getSampler(t),M=new Array(f).fill(0),w=d.tolist();let T;o&&o.put(w);let b={};for(;;){if(i=this.prepare_inputs_for_generation(w,i,t),T=await this.forward(i),t.output_attentions&&t.return_dict_in_generate){const e=this.getAttentions(T);for(const t in e)t in b||(b[t]=[]),b[t].push(e[t])}const e=h(w,T.logits.slice(null,-1,null)),r=[];for(let t=0;t<e.dims.at(0);++t){const n=e[t],o=await g(n);for(const[e,n]of o){const o=BigInt(e);M[t]+=n,w[t].push(o),r.push([o]);break}}o&&o.put(r);if(m(w).every((e=>e)))break;i=this._update_model_kwargs_for_generation({generated_input_ids:r,outputs:T,model_inputs:i,is_encoder_decoder:c})}o&&o.end();const x=this.getPastKeyValues(T,i.past_key_values,!0),y=new p.Tensor("int64",w.flat(),[w.length,w[0].length]);if(t.return_dict_in_generate)return{sequences:y,past_key_values:x,...b};for(const e of Object.values(T))"gpu-buffer"===e.location&&e.dispose();return y}getPastKeyValues(e,t,r=!1){const n=Object.create(null);for(const o in e)if(o.startsWith("present")){const s=o.replace("present","past_key_values"),a=o.includes("encoder");if(n[s]=a&&t?t[s]:e[o],t&&(!a||r)){const e=t[s];"gpu-buffer"===e.location&&e.dispose()}}return n}getAttentions(e){const t={};for(const r of["cross_attentions","encoder_attentions","decoder_attentions"])for(const n in e)n.startsWith(r)&&(r in t||(t[r]=[]),t[r].push(e[n]));return t}addPastKeyValues(e,t){if(t)Object.assign(e,t);else{const t=this.sessions.decoder_model_merged??this.sessions.model,r=t?.config?.kv_cache_dtype??"float32",o="float16"===r?new Uint16Array:[],s=(0,n.getKeyValueShapes)(this.config);for(const t in s)e[t]=new p.Tensor(r,o,s[t])}}async encode_image({pixel_values:e}){const t=(await L(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 L(this.sessions.embed_tokens,{input_ids:e})).inputs_embeds}}class W{}class $ extends W{constructor({last_hidden_state:e,hidden_states:t=null,attentions:r=null}){super(),this.last_hidden_state=e,this.hidden_states=t,this.attentions=r}}class U extends q{}class X extends U{}class Q extends U{async _call(e){return new fl(await super._call(e))}}class H extends U{async _call(e){return new hl(await super._call(e))}}class Y extends U{async _call(e){return new _l(await super._call(e))}}class J extends U{async _call(e){return new gl(await super._call(e))}}class K extends q{}class Z extends K{}class ee extends q{}class te extends ee{}class re extends ee{async _call(e){return new fl(await super._call(e))}}class ne extends ee{async _call(e){return new hl(await super._call(e))}}class oe extends ee{async _call(e){return new _l(await super._call(e))}}class se extends ee{async _call(e){return new gl(await super._call(e))}}class ae extends q{}class ie extends ae{}class le extends ae{async _call(e){return new fl(await super._call(e))}}class ce extends ae{async _call(e){return new hl(await super._call(e))}}class de extends ae{async _call(e){return new _l(await super._call(e))}}class ue extends ae{async _call(e){return new gl(await super._call(e))}}class pe extends q{}class he extends pe{}class me extends pe{async _call(e){return new fl(await super._call(e))}}class _e extends pe{async _call(e){return new hl(await super._call(e))}}class fe extends pe{async _call(e){return new _l(await super._call(e))}}class ge extends pe{async _call(e){return new gl(await super._call(e))}}class Me extends q{}class we extends Me{}class Te extends Me{async _call(e){return new fl(await super._call(e))}}class be extends Me{async _call(e){return new hl(await super._call(e))}}class xe extends Me{async _call(e){return new _l(await super._call(e))}}class ye extends Me{async _call(e){return new gl(await super._call(e))}}class ke extends q{}class Fe extends ke{}class Pe extends ke{async _call(e){return new fl(await super._call(e))}}class Ce extends ke{async _call(e){return new hl(await super._call(e))}}class ve extends ke{async _call(e){return new _l(await super._call(e))}}class Se extends ke{async _call(e){return new gl(await super._call(e))}}class Ae extends q{}class Ee extends Ae{}class Le extends Ae{async _call(e){return new fl(await super._call(e))}}class ze extends Ae{async _call(e){return new hl(await super._call(e))}}class Ie extends Ae{async _call(e){return new _l(await super._call(e))}}class Be extends Ae{async _call(e){return new gl(await super._call(e))}}class Ne extends q{}class De extends Ne{}class Oe extends Ne{async _call(e){return new hl(await super._call(e))}}class Ve extends Ne{async _call(e){return new _l(await super._call(e))}}class je extends Ne{async _call(e){return new gl(await super._call(e))}}class Re extends Ne{async _call(e){return new fl(await super._call(e))}}class Ge extends q{}class qe extends Ge{}class We extends Ge{async _call(e){return new fl(await super._call(e))}}class $e extends Ge{async _call(e){return new hl(await super._call(e))}}class Ue extends Ge{async _call(e){return new _l(await super._call(e))}}class Xe extends q{}class Qe extends Xe{}class He extends Xe{async _call(e){return new fl(await super._call(e))}}class Ye extends Xe{async _call(e){return new hl(await super._call(e))}}class Je extends Xe{async _call(e){return new gl(await super._call(e))}}class Ke extends q{}class Ze extends Ke{}class et extends Ke{async _call(e){return new fl(await super._call(e))}}class tt extends Ke{async _call(e){return new hl(await super._call(e))}}class rt extends Ke{async _call(e){return new _l(await super._call(e))}}class nt extends Ke{async _call(e){return new gl(await super._call(e))}}class ot extends q{}class st extends ot{}class at extends ot{async _call(e){return new fl(await super._call(e))}}class it extends ot{async _call(e){return new hl(await super._call(e))}}class lt extends ot{async _call(e){return new gl(await super._call(e))}}class ct extends q{}class dt extends ct{}class ut extends ct{async _call(e){return new hl(await super._call(e))}}class pt extends ct{async _call(e){return new gl(await super._call(e))}}class ht extends ct{async _call(e){return new fl(await super._call(e))}}class mt extends q{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"]}class _t extends mt{}class ft extends mt{}class gt extends q{}class Mt extends gt{}class wt extends gt{}class Tt extends q{}class bt extends Tt{}class xt extends Tt{}class yt extends q{}class kt extends yt{}class Ft extends yt{}class Pt extends yt{async _call(e){return new hl(await super._call(e))}}class Ct extends q{}class vt extends Ct{}class St extends Ct{}class At extends Ct{async _call(e){return new hl(await super._call(e))}}class Et extends Ct{}class Lt extends q{}class zt extends Lt{}class It extends Lt{}class Bt extends q{}class Nt extends Bt{}class Dt extends Bt{}class Ot extends q{}class Vt extends Ot{}class jt extends Ot{async _call(e){return new fl(await super._call(e))}}class Rt extends Ot{async _call(e){return new hl(await super._call(e))}}class Gt extends Ot{async _call(e){return new _l(await super._call(e))}}class qt extends Ot{async _call(e){return new gl(await super._call(e))}}class Wt extends q{}class $t extends Wt{}class Ut extends Wt{async _call(e){return new fl(await super._call(e))}}class Xt extends Wt{async _call(e){return new hl(await super._call(e))}}class Qt extends Wt{async _call(e){return new _l(await super._call(e))}}class Ht extends Wt{async _call(e){return new gl(await super._call(e))}}class Yt extends q{}class Jt extends Yt{}class Kt extends Yt{async _call(e){return new fl(await super._call(e))}}class Zt extends Yt{async _call(e){return new hl(await super._call(e))}}class er extends Yt{async _call(e){return new _l(await super._call(e))}}class tr extends Yt{async _call(e){return new gl(await super._call(e))}}class rr extends q{}class nr extends rr{}class or extends rr{}class sr extends q{requires_attention_mask=!1;main_input_name="input_features";forward_params=["input_features","attention_mask","decoder_input_ids","decoder_attention_mask","past_key_values"]}class ar extends sr{}class ir extends sr{_prepare_generation_config(e,t){return super._prepare_generation_config(e,t,g.WhisperGenerationConfig)}_retrieve_init_tokens(e){const t=[e.decoder_start_token_id];let r=e.language;const n=e.task;if(e.is_multilingual){r||(console.warn("No language specified - defaulting to English (en)."),r="en");const o=`<|${(0,M.whisper_language_to_code)(r)}|>`;t.push(e.lang_to_id[o]),t.push(e.task_to_id[n??"transcribe"])}else if(r||n)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:r=null,stopping_criteria:n=null,...o}){t=this._prepare_generation_config(t,o);const s=o.decoder_input_ids??this._retrieve_init_tokens(t);if(t.return_timestamps&&(r??=new d.LogitsProcessorList,r.push(new d.WhisperTimeStampLogitsProcessor(t,s))),t.begin_suppress_tokens&&(r??=new d.LogitsProcessorList,r.push(new d.SuppressTokensAtBeginLogitsProcessor(t.begin_suppress_tokens,s.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 a=await super.generate({inputs:e,generation_config:t,logits_processor:r,decoder_input_ids:s,...o});return t.return_token_timestamps&&(a.token_timestamps=this._extract_token_timestamps(a,t.alignment_heads,t.num_frames)),a}_extract_token_timestamps(e,t,r=null,n=.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==r&&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 o=this.config.median_filter_width;void 0===o&&(console.warn("Model config has no `median_filter_width`, using default value of 7."),o=7);const s=e.cross_attentions,a=Array.from({length:this.config.decoder_layers},((e,t)=>(0,p.cat)(s.map((e=>e[t])),2))),l=(0,p.stack)(t.map((([e,t])=>{if(e>=a.length)throw new Error(`Layer index ${e} is out of bounds for cross attentions (length ${a.length}).`);return r?a[e].slice(null,t,null,[0,r]):a[e].slice(null,t)}))).transpose(1,0,2,3),[c,d]=(0,p.std_mean)(l,-2,0,!0),u=l.clone();for(let e=0;e<u.dims[0];++e){const t=u[e];for(let r=0;r<t.dims[0];++r){const n=t[r],s=c[e][r][0].data,a=d[e][r][0].data;for(let e=0;e<n.dims[0];++e){let t=n[e].data;for(let e=0;e<t.length;++e)t[e]=(t[e]-a[e])/s[e];t.set((0,h.medianFilter)(t,o))}}}const m=[(0,p.mean)(u,1)],_=e.sequences.dims,f=new p.Tensor("float32",new Float32Array(_[0]*_[1]),_);for(let e=0;e<_[0];++e){const t=m[e].neg().squeeze_(0),[r,o]=(0,h.dynamic_time_warping)(t.tolist()),s=Array.from({length:r.length-1},((e,t)=>r[t+1]-r[t])),a=(0,i.mergeArrays)([1],s).map((e=>!!e)),l=[];for(let e=0;e<a.length;++e)a[e]&&l.push(o[e]*n);f[e].data.set(l,1)}return f}}class lr extends q{main_input_name="pixel_values";forward_params=["pixel_values","input_ids","encoder_hidden_states","past_key_values"]}class cr extends q{forward_params=["input_ids","pixel_values","attention_mask","position_ids","past_key_values"]}class dr extends cr{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:r,attention_mask:n}){const o=this.config.image_token_index,s=r.tolist().map((e=>e.findIndex((e=>e==o)))),a=s.every((e=>-1===e)),i=s.every((e=>-1!==e));if(!a&&!i)throw new Error("Every input should contain either 0 or 1 image token.");if(a)return{inputs_embeds:e,attention_mask:n};const l=[],c=[];for(let r=0;r<s.length;++r){const o=s[r],a=e[r],i=t[r],d=n[r];l.push((0,p.cat)([a.slice([0,o]),i,a.slice([o+1,a.dims[0]])],0)),c.push((0,p.cat)([d.slice([0,o]),(0,p.ones)([i.dims[0]]),d.slice([o+1,d.dims[0]])],0))}return{inputs_embeds:(0,p.stack)(l,0),attention_mask:(0,p.stack)(c,0)}}}class ur extends dr{}class pr extends q{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 hr extends pr{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:r,attention_mask:n}){return{inputs_embeds:(0,p.cat)([t,e],1),attention_mask:(0,p.cat)([(0,p.ones)(t.dims.slice(0,2)),n],1)}}async _prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:r,attention_mask:n}){if(!e&&!t)throw new Error("Either `input_ids` or `pixel_values` should be provided.");let o,s;return e&&(o=await this.encode_text({input_ids:e})),t&&(s=await this.encode_image({pixel_values:t})),o&&s?({inputs_embeds:r,attention_mask:n}=this._merge_input_ids_with_image_features({inputs_embeds:o,image_features:s,input_ids:e,attention_mask:n})):r=o||s,{inputs_embeds:r,attention_mask:n}}async forward({input_ids:e,pixel_values:t,attention_mask:r,decoder_input_ids:n,decoder_attention_mask:o,encoder_outputs:s,past_key_values:a,inputs_embeds:i,decoder_inputs_embeds:l}){if(i||({inputs_embeds:i,attention_mask:r}=await this._prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:i,attention_mask:r})),!s){let{last_hidden_state:e}=await D(this,{inputs_embeds:i,attention_mask:r});s=e}if(!l){if(!n)throw new Error("Either `decoder_input_ids` or `decoder_inputs_embeds` should be provided.");l=await this.encode_text({input_ids:n})}const c={inputs_embeds:l,attention_mask:o,encoder_attention_mask:r,encoder_hidden_states:s,past_key_values:a};return await O(this,c,!0)}}class mr extends q{}class _r extends mr{}class fr extends mr{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class gr extends mr{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class Mr extends mr{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class wr extends mr{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class Tr extends q{}class br extends Tr{}class xr extends Tr{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class yr extends mr{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class kr extends q{}class Fr extends kr{}class Pr extends q{}class Cr extends Pr{}class vr extends Pr{}class Sr extends q{}class Ar extends Sr{}class Er extends Sr{}class Lr extends q{}class zr extends Lr{}class Ir extends Lr{}class Br extends q{}class Nr extends Br{}class Dr extends Br{}class Or extends q{}class Vr extends Or{}class jr extends Or{}class Rr extends q{}class Gr extends Rr{}class qr extends Rr{}class Wr extends q{}class $r extends Wr{}class Ur extends Wr{}class Xr extends q{}class Qr extends Xr{}class Hr extends Xr{}class Yr extends q{}class Jr extends Yr{}class Kr extends Yr{}class Zr extends q{}class en extends Zr{}class tn extends Zr{}class rn extends q{}class nn extends rn{}class on extends rn{}class sn extends q{}class an extends sn{}class ln extends sn{}class cn extends q{}class dn extends cn{}class un extends cn{}class pn extends q{}class hn extends pn{}class mn extends pn{}class _n extends q{}class fn extends _n{}class gn extends _n{}class Mn extends q{}class wn extends Mn{}class Tn extends Mn{}class bn extends q{}class xn extends bn{}class yn extends bn{}class kn extends q{}class Fn extends kn{}class Pn extends kn{}class Cn extends q{}class vn extends Cn{}class Sn extends Cn{}class An extends q{}class En extends An{}class Ln extends An{}class zn extends q{}class In extends zn{}class Bn extends zn{async _call(e){return new hl(await super._call(e))}}class Nn extends q{}class Dn extends Nn{}class On extends Nn{async _call(e){return new hl(await super._call(e))}}class Vn extends q{}class jn extends Vn{}class Rn extends q{}class Gn extends Rn{}class qn extends Rn{async _call(e){return new hl(await super._call(e))}}class Wn extends q{}class $n extends Wn{}class Un extends q{}class Xn extends Un{}class Qn extends Un{async _call(e){return new hl(await super._call(e))}}class Hn extends q{}class Yn extends Hn{async _call(e){return new Tl(await super._call(e))}}class Jn extends q{}class Kn extends Jn{}class Zn extends Jn{async _call(e){return new hl(await super._call(e))}}class eo extends q{}class to extends eo{}class ro extends eo{async _call(e){return new hl(await super._call(e))}}class no extends q{}class oo extends no{}class so extends no{}class ao extends q{}class io extends ao{}class lo extends ao{}class co extends q{}class uo extends co{}class po extends co{async _call(e){return new hl(await super._call(e))}}class ho extends q{}class mo extends ho{}class _o extends ho{async _call(e){return new go(await super._call(e))}}class fo extends ho{async _call(e){return new Mo(await super._call(e))}}class go extends W{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class Mo extends W{constructor({logits:e,pred_boxes:t,pred_masks:r}){super(),this.logits=e,this.pred_boxes=t,this.pred_masks=r}}class wo extends q{}class To extends wo{}class bo extends wo{async _call(e){return new xo(await super._call(e))}}class xo extends W{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class yo extends q{}class ko extends yo{}class Fo extends yo{async _call(e){return new Po(await super._call(e))}}class Po extends go{}class Co extends q{}class vo extends Co{}class So extends Co{async _call(e){return new hl(await super._call(e))}}class Ao extends q{}class Eo extends Ao{}class Lo extends Ao{async _call(e){return new hl(await super._call(e))}}class zo extends q{}class Io extends zo{}class Bo extends zo{async _call(e){return new hl(await super._call(e))}}class No extends q{}class Do extends No{}class Oo extends No{async _call(e){return new hl(await super._call(e))}}class Vo extends q{}class jo extends Vo{}class Ro extends Vo{}class Go extends q{}class qo extends Go{}class Wo extends Go{}class $o extends q{}class Uo extends $o{}class Xo extends q{}class Qo extends Xo{}class Ho extends Xo{}class Yo extends Xo{}class Jo extends q{}class Ko extends Jo{}class Zo extends q{}class es extends Zo{}class ts extends Zo{}class rs extends q{}class ns extends rs{}class os extends rs{}class ss extends q{}class as extends ss{}class is extends q{}class ls extends is{}class cs extends is{async _call(e){return new hl(await super._call(e))}}class ds extends q{}class us extends ds{}class ps extends ds{async _call(e){return new hl(await super._call(e))}}class hs extends q{}class ms extends hs{}class _s extends hs{async _call(e){return new hl(await super._call(e))}}class fs extends q{}class gs extends fs{}class Ms extends fs{async _call(e){return new ws(await super._call(e))}}class ws extends W{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class Ts extends q{}class bs extends Ts{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),r=t.reduce(((e,t)=>e*t),1);e.input_labels=new p.Tensor("int64",new BigInt64Array(r).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 L(this.sessions.prompt_encoder_mask_decoder,t)}async _call(e){return new xs(await super._call(e))}}class xs extends W{constructor({iou_scores:e,pred_masks:t}){super(),this.iou_scores=e,this.pred_masks=t}}class ys extends q{}class ks extends ys{}class Fs extends ys{}class Ps extends q{}class Cs extends Ps{}class vs extends Ps{}class Ss extends q{}class As extends Ss{}class Es extends Ss{async _call(e){return new Ml(await super._call(e))}}class Ls extends Ss{async _call(e){return new hl(await super._call(e))}}class zs extends Ss{async _call(e){return new _l(await super._call(e))}}class Is extends q{}class Bs extends Is{}class Ns extends Is{async _call(e){return new _l(await super._call(e))}}class Ds extends q{}class Os extends Ds{}class Vs extends q{}class js extends Vs{}class Rs extends Vs{async _call(e){return new Ml(await super._call(e))}}class Gs extends Vs{async _call(e){return new hl(await super._call(e))}}class qs extends q{}class Ws extends qs{}class $s extends qs{async _call(e){return new Ml(await super._call(e))}}class Us extends qs{async _call(e){return new hl(await super._call(e))}}class Xs extends qs{async _call(e){return new _l(await super._call(e))}}class Qs extends q{}class Hs extends Qs{}class Ys extends Qs{async _call(e){return new Ml(await super._call(e))}}class Js extends Qs{async _call(e){return new hl(await super._call(e))}}class Ks extends q{}class Zs extends Ss{}class ea extends Ss{async _call(e){return new Ml(await super._call(e))}}class ta extends Ss{async _call(e){return new hl(await super._call(e))}}class ra extends q{}class na extends ra{}class oa extends ra{async _call(e){return new Ml(await super._call(e))}}class sa extends ra{async _call(e){return new hl(await super._call(e))}}class aa extends ra{async _call(e){return new ml(await super._call(e))}}class ia extends ra{async _call(e){return new _l(await super._call(e))}}class la extends q{}class ca extends la{}class da extends la{}class ua extends la{async generate_speech(e,t,{threshold:r=.5,minlenratio:n=0,maxlenratio:o=20,vocoder:s=null}={}){const a={input_ids:e},{encoder_outputs:i,encoder_attention_mask:l}=await D(this,a),c=i.dims[1]/this.config.reduction_factor,d=Math.floor(c*o),u=Math.floor(c*n),h=this.config.num_mel_bins;let m=[],_=null,f=null,g=0;for(;;){++g;const e=B(!!f);let n;n=f?f.output_sequence_out:new p.Tensor("float32",new Float32Array(h),[1,1,h]);let o={use_cache_branch:e,output_sequence:n,encoder_attention_mask:l,speaker_embeddings:t,encoder_hidden_states:i};this.addPastKeyValues(o,_),f=await L(this.sessions.decoder_model_merged,o),_=this.getPastKeyValues(f,_);const{prob:s,spectrum:a}=f;if(m.push(a),g>=u&&(Array.from(s.data).filter((e=>e>=r)).length>0||g>=d))break}const M=(0,p.cat)(m),{waveform:w}=await L(s.sessions.model,{spectrogram:M});return{spectrogram:M,waveform:w}}}class pa extends q{main_input_name="spectrogram"}class ha extends q{}class ma extends ha{}class _a extends q{}class fa extends _a{}class ga extends _a{}class Ma extends q{}class wa extends Ma{}class Ta extends Ma{}class ba extends q{}class xa extends ba{}class ya extends ba{}class ka extends q{}class Fa extends ka{}class Pa extends ka{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class Ca extends ka{static async from_pretrained(e,t={}){return t.model_file_name??="audio_model",super.from_pretrained(e,t)}}class va extends q{}class Sa extends va{async _call(e){return new bl(await super._call(e))}}class Aa extends q{}class Ea extends Aa{}class La extends Aa{}class za extends Aa{}class Ia extends q{}class Ba extends Ia{}class Na extends Ia{}class Da extends q{}class Oa extends Da{}class Va extends Da{async _call(e){return new hl(await super._call(e))}}class ja extends q{}class Ra extends ja{}class Ga extends ja{}class qa extends q{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,r]=e.dims,n=this.config.decoder.num_codebooks,o=r-n;let s=0;for(let t=0;t<e.size;++t){if(e.data[t]===this.config.decoder.pad_token_id)continue;const a=t%r-Math.floor(t/r)%n;a>0&&a<=o&&(e.data[s++]=e.data[t])}const a=Math.floor(t/n),i=s/(a*n);return new p.Tensor(e.type,e.data.slice(0,s),[a,n,i])}prepare_inputs_for_generation(e,t,r){let n=structuredClone(e);for(let e=0;e<n.length;++e)for(let t=0;t<n[e].length;++t)e%this.config.decoder.num_codebooks>=t&&(n[e][t]=BigInt(this.config.decoder.pad_token_id));null!==r.guidance_scale&&r.guidance_scale>1&&(n=n.concat(n));return super.prepare_inputs_for_generation(n,t,r)}async generate(e){const t=await super.generate(e),r=this._apply_and_filter_by_delay_pattern_mask(t).unsqueeze_(0),{audio_values:n}=await L(this.sessions.encodec_decode,{audio_codes:r});return n}}class Wa extends q{}class $a extends Wa{}class Ua extends Wa{async _call(e){return new hl(await super._call(e))}}class Xa extends q{}class Qa extends Xa{}class Ha extends Xa{async _call(e){return new hl(await super._call(e))}}class Ya extends q{}class Ja extends Ya{}class Ka extends Ya{async _call(e){return new hl(await super._call(e))}}class Za extends q{}class ei extends Za{}class ti extends Za{async _call(e){return new hl(await super._call(e))}}class ri extends q{}class ni extends ri{}class oi{static MODEL_CLASS_MAPPINGS=null;static BASE_IF_FAIL=!1;static async from_pretrained(e,{progress_callback:t=null,config:r=null,cache_dir:o=null,local_files_only:s=!1,revision:a="main",model_file_name:i=null,subfolder:l="onnx",device:c=null,dtype:d=null,use_external_data_format:u=null,session_options:p={}}={}){const h={progress_callback:t,config:r,cache_dir:o,local_files_only:s,revision:a,model_file_name:i,subfolder:l,device:c,dtype:d,use_external_data_format:u,session_options:p};if(h.config=await n.AutoConfig.from_pretrained(e,h),!this.MODEL_CLASS_MAPPINGS)throw new Error("`MODEL_CLASS_MAPPINGS` not implemented for this type of `AutoClass`: "+this.name);for(const t of this.MODEL_CLASS_MAPPINGS){const r=t.get(h.config.model_type);if(r)return await r[1].from_pretrained(e,h)}if(this.BASE_IF_FAIL)return console.warn(`Unknown model class "${h.config.model_type}", attempting to construct from base class.`),await q.from_pretrained(e,h);throw Error(`Unsupported model type: ${h.config.model_type}`)}}const si=new Map([["bert",["BertModel",X]],["nomic_bert",["NomicBertModel",Z]],["roformer",["RoFormerModel",te]],["electra",["ElectraModel",he]],["esm",["EsmModel",qe]],["convbert",["ConvBertModel",ie]],["camembert",["CamembertModel",we]],["deberta",["DebertaModel",Fe]],["deberta-v2",["DebertaV2Model",Ee]],["mpnet",["MPNetModel",Ze]],["albert",["AlbertModel",dt]],["distilbert",["DistilBertModel",De]],["roberta",["RobertaModel",Vt]],["xlm",["XLMModel",$t]],["xlm-roberta",["XLMRobertaModel",Jt]],["clap",["ClapModel",Fa]],["clip",["CLIPModel",_r]],["clipseg",["CLIPSegModel",Cr]],["chinese_clip",["ChineseCLIPModel",Fr]],["siglip",["SiglipModel",br]],["mobilebert",["MobileBertModel",Qe]],["squeezebert",["SqueezeBertModel",st]],["wav2vec2",["Wav2Vec2Model",As]],["wav2vec2-bert",["Wav2Vec2BertModel",Hs]],["unispeech",["UniSpeechModel",js]],["unispeech-sat",["UniSpeechSatModel",Ws]],["hubert",["HubertModel",Zs]],["wavlm",["WavLMModel",na]],["audio-spectrogram-transformer",["ASTModel",nr]],["vits",["VitsModel",Sa]],["pyannote",["PyAnnoteModel",Bs]],["wespeaker-resnet",["WeSpeakerResNetModel",Os]],["detr",["DetrModel",mo]],["rt_detr",["RTDetrModel",To]],["table-transformer",["TableTransformerModel",ko]],["vit",["ViTModel",In]],["pvt",["PvtModel",Dn]],["vit_msn",["ViTMSNModel",Gn]],["vit_mae",["ViTMAEModel",jn]],["groupvit",["GroupViTModel",$n]],["fastvit",["FastViTModel",Xn]],["mobilevit",["MobileViTModel",Kn]],["mobilevitv2",["MobileViTV2Model",to]],["owlvit",["OwlViTModel",oo]],["owlv2",["Owlv2Model",io]],["beit",["BeitModel",uo]],["deit",["DeiTModel",vo]],["hiera",["HieraModel",Eo]],["convnext",["ConvNextModel",ls]],["convnextv2",["ConvNextV2Model",us]],["dinov2",["Dinov2Model",ms]],["resnet",["ResNetModel",Io]],["swin",["SwinModel",Do]],["swin2sr",["Swin2SRModel",jo]],["donut-swin",["DonutSwinModel",as]],["yolos",["YolosModel",gs]],["dpt",["DPTModel",qo]],["glpn",["GLPNModel",ns]],["hifigan",["SpeechT5HifiGan",pa]],["efficientnet",["EfficientNetModel",Oa]],["decision_transformer",["DecisionTransformerModel",ni]],["mobilenet_v1",["MobileNetV1Model",$a]],["mobilenet_v2",["MobileNetV2Model",Qa]],["mobilenet_v3",["MobileNetV3Model",Ja]],["mobilenet_v4",["MobileNetV4Model",ei]],["maskformer",["MaskFormerModel",es]]]),ai=new Map([["t5",["T5Model",_t]],["longt5",["LongT5Model",Mt]],["mt5",["MT5Model",bt]],["bart",["BartModel",kt]],["mbart",["MBartModel",vt]],["marian",["MarianModel",ks]],["whisper",["WhisperModel",ar]],["m2m_100",["M2M100Model",Cs]],["blenderbot",["BlenderbotModel",zt]],["blenderbot-small",["BlenderbotSmallModel",Nt]]]),ii=new Map([["bloom",["BloomModel",Fn]],["jais",["JAISModel",zr]],["gpt2",["GPT2Model",Ar]],["gptj",["GPTJModel",Gr]],["gpt_bigcode",["GPTBigCodeModel",$r]],["gpt_neo",["GPTNeoModel",Nr]],["gpt_neox",["GPTNeoXModel",Vr]],["codegen",["CodeGenModel",Qr]],["llama",["LlamaModel",Jr]],["granite",["GraniteModel",en]],["cohere",["CohereModel",nn]],["gemma",["GemmaModel",an]],["gemma2",["Gemma2Model",dn]],["openelm",["OpenELMModel",hn]],["qwen2",["Qwen2Model",fn]],["phi",["PhiModel",wn]],["phi3",["Phi3Model",xn]],["mpt",["MptModel",vn]],["opt",["OPTModel",En]],["mistral",["MistralModel",fa]],["starcoder2",["Starcoder2Model",wa]],["falcon",["FalconModel",xa]],["stablelm",["StableLmModel",Ba]]]),li=new Map([["speecht5",["SpeechT5ForSpeechToText",da]],["whisper",["WhisperForConditionalGeneration",ir]]]),ci=new Map([["speecht5",["SpeechT5ForTextToSpeech",ua]]]),di=new Map([["vits",["VitsModel",Sa]],["musicgen",["MusicgenForConditionalGeneration",qa]]]),ui=new Map([["bert",["BertForSequenceClassification",H]],["roformer",["RoFormerForSequenceClassification",ne]],["electra",["ElectraForSequenceClassification",_e]],["esm",["EsmForSequenceClassification",$e]],["convbert",["ConvBertForSequenceClassification",ce]],["camembert",["CamembertForSequenceClassification",be]],["deberta",["DebertaForSequenceClassification",Ce]],["deberta-v2",["DebertaV2ForSequenceClassification",ze]],["mpnet",["MPNetForSequenceClassification",tt]],["albert",["AlbertForSequenceClassification",ut]],["distilbert",["DistilBertForSequenceClassification",Oe]],["roberta",["RobertaForSequenceClassification",Rt]],["xlm",["XLMForSequenceClassification",Xt]],["xlm-roberta",["XLMRobertaForSequenceClassification",Zt]],["bart",["BartForSequenceClassification",Pt]],["mbart",["MBartForSequenceClassification",At]],["mobilebert",["MobileBertForSequenceClassification",Ye]],["squeezebert",["SqueezeBertForSequenceClassification",it]]]),pi=new Map([["bert",["BertForTokenClassification",Y]],["roformer",["RoFormerForTokenClassification",oe]],["electra",["ElectraForTokenClassification",fe]],["esm",["EsmForTokenClassification",Ue]],["convbert",["ConvBertForTokenClassification",de]],["camembert",["CamembertForTokenClassification",xe]],["deberta",["DebertaForTokenClassification",ve]],["deberta-v2",["DebertaV2ForTokenClassification",Ie]],["mpnet",["MPNetForTokenClassification",rt]],["distilbert",["DistilBertForTokenClassification",Ve]],["roberta",["RobertaForTokenClassification",Gt]],["xlm",["XLMForTokenClassification",Qt]],["xlm-roberta",["XLMRobertaForTokenClassification",er]]]),hi=new Map([["t5",["T5ForConditionalGeneration",ft]],["longt5",["LongT5ForConditionalGeneration",wt]],["mt5",["MT5ForConditionalGeneration",xt]],["bart",["BartForConditionalGeneration",Ft]],["mbart",["MBartForConditionalGeneration",St]],["marian",["MarianMTModel",Fs]],["m2m_100",["M2M100ForConditionalGeneration",vs]],["blenderbot",["BlenderbotForConditionalGeneration",It]],["blenderbot-small",["BlenderbotSmallForConditionalGeneration",Dt]]]),mi=new Map([["bloom",["BloomForCausalLM",Pn]],["gpt2",["GPT2LMHeadModel",Er]],["jais",["JAISLMHeadModel",Ir]],["gptj",["GPTJForCausalLM",qr]],["gpt_bigcode",["GPTBigCodeForCausalLM",Ur]],["gpt_neo",["GPTNeoForCausalLM",Dr]],["gpt_neox",["GPTNeoXForCausalLM",jr]],["codegen",["CodeGenForCausalLM",Hr]],["llama",["LlamaForCausalLM",Kr]],["granite",["GraniteForCausalLM",tn]],["cohere",["CohereForCausalLM",on]],["gemma",["GemmaForCausalLM",ln]],["gemma2",["Gemma2ForCausalLM",un]],["openelm",["OpenELMForCausalLM",mn]],["qwen2",["Qwen2ForCausalLM",gn]],["phi",["PhiForCausalLM",Tn]],["phi3",["Phi3ForCausalLM",yn]],["mpt",["MptForCausalLM",Sn]],["opt",["OPTForCausalLM",Ln]],["mbart",["MBartForCausalLM",Et]],["mistral",["MistralForCausalLM",ga]],["starcoder2",["Starcoder2ForCausalLM",Ta]],["falcon",["FalconForCausalLM",ya]],["trocr",["TrOCRForCausalLM",ma]],["stablelm",["StableLmForCausalLM",Na]]]),_i=new Map([["bert",["BertForMaskedLM",Q]],["roformer",["RoFormerForMaskedLM",re]],["electra",["ElectraForMaskedLM",me]],["esm",["EsmForMaskedLM",We]],["convbert",["ConvBertForMaskedLM",le]],["camembert",["CamembertForMaskedLM",Te]],["deberta",["DebertaForMaskedLM",Pe]],["deberta-v2",["DebertaV2ForMaskedLM",Le]],["mpnet",["MPNetForMaskedLM",et]],["albert",["AlbertForMaskedLM",ht]],["distilbert",["DistilBertForMaskedLM",Re]],["roberta",["RobertaForMaskedLM",jt]],["xlm",["XLMWithLMHeadModel",Ut]],["xlm-roberta",["XLMRobertaForMaskedLM",Kt]],["mobilebert",["MobileBertForMaskedLM",He]],["squeezebert",["SqueezeBertForMaskedLM",at]]]),fi=new Map([["bert",["BertForQuestionAnswering",J]],["roformer",["RoFormerForQuestionAnswering",se]],["electra",["ElectraForQuestionAnswering",ge]],["convbert",["ConvBertForQuestionAnswering",ue]],["camembert",["CamembertForQuestionAnswering",ye]],["deberta",["DebertaForQuestionAnswering",Se]],["deberta-v2",["DebertaV2ForQuestionAnswering",Be]],["mpnet",["MPNetForQuestionAnswering",nt]],["albert",["AlbertForQuestionAnswering",pt]],["distilbert",["DistilBertForQuestionAnswering",je]],["roberta",["RobertaForQuestionAnswering",qt]],["xlm",["XLMForQuestionAnswering",Ht]],["xlm-roberta",["XLMRobertaForQuestionAnswering",tr]],["mobilebert",["MobileBertForQuestionAnswering",Je]],["squeezebert",["SqueezeBertForQuestionAnswering",lt]]]),gi=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",lr]]]),Mi=new Map([["llava",["LlavaForConditionalGeneration",dr]],["moondream1",["Moondream1ForConditionalGeneration",ur]],["florence2",["Florence2ForConditionalGeneration",hr]]]),wi=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",lr]]]),Ti=new Map([["vit",["ViTForImageClassification",Bn]],["pvt",["PvtForImageClassification",On]],["vit_msn",["ViTMSNForImageClassification",qn]],["fastvit",["FastViTForImageClassification",Qn]],["mobilevit",["MobileViTForImageClassification",Zn]],["mobilevitv2",["MobileViTV2ForImageClassification",ro]],["beit",["BeitForImageClassification",po]],["deit",["DeiTForImageClassification",So]],["hiera",["HieraForImageClassification",Lo]],["convnext",["ConvNextForImageClassification",cs]],["convnextv2",["ConvNextV2ForImageClassification",ps]],["dinov2",["Dinov2ForImageClassification",_s]],["resnet",["ResNetForImageClassification",Bo]],["swin",["SwinForImageClassification",Oo]],["segformer",["SegformerForImageClassification",La]],["efficientnet",["EfficientNetForImageClassification",Va]],["mobilenet_v1",["MobileNetV1ForImageClassification",Ua]],["mobilenet_v2",["MobileNetV2ForImageClassification",Ha]],["mobilenet_v3",["MobileNetV3ForImageClassification",Ka]],["mobilenet_v4",["MobileNetV4ForImageClassification",ti]]]),bi=new Map([["detr",["DetrForObjectDetection",_o]],["rt_detr",["RTDetrForObjectDetection",bo]],["table-transformer",["TableTransformerForObjectDetection",Fo]],["yolos",["YolosForObjectDetection",Ms]]]),xi=new Map([["owlvit",["OwlViTForObjectDetection",so]],["owlv2",["Owlv2ForObjectDetection",lo]]]),yi=new Map([["detr",["DetrForSegmentation",fo]],["clipseg",["CLIPSegForImageSegmentation",vr]]]),ki=new Map([["segformer",["SegformerForSemanticSegmentation",za]],["sapiens",["SapiensForSemanticSegmentation",Qo]]]),Fi=new Map([["detr",["DetrForSegmentation",fo]],["maskformer",["MaskFormerForInstanceSegmentation",ts]]]),Pi=new Map([["sam",["SamModel",bs]]]),Ci=new Map([["wav2vec2",["Wav2Vec2ForCTC",Es]],["wav2vec2-bert",["Wav2Vec2BertForCTC",Ys]],["unispeech",["UniSpeechForCTC",Rs]],["unispeech-sat",["UniSpeechSatForCTC",$s]],["wavlm",["WavLMForCTC",oa]],["hubert",["HubertForCTC",ea]]]),vi=new Map([["wav2vec2",["Wav2Vec2ForSequenceClassification",Ls]],["wav2vec2-bert",["Wav2Vec2BertForSequenceClassification",Js]],["unispeech",["UniSpeechForSequenceClassification",Gs]],["unispeech-sat",["UniSpeechSatForSequenceClassification",Us]],["wavlm",["WavLMForSequenceClassification",sa]],["hubert",["HubertForSequenceClassification",ta]],["audio-spectrogram-transformer",["ASTForAudioClassification",or]]]),Si=new Map([["wavlm",["WavLMForXVector",aa]]]),Ai=new Map([["unispeech-sat",["UniSpeechSatForAudioFrameClassification",Xs]],["wavlm",["WavLMForAudioFrameClassification",ia]],["wav2vec2",["Wav2Vec2ForAudioFrameClassification",zs]],["pyannote",["PyAnnoteForAudioFrameClassification",Ns]]]),Ei=new Map([["vitmatte",["VitMatteForImageMatting",Yn]]]),Li=new Map([["swin2sr",["Swin2SRForImageSuperResolution",Ro]]]),zi=new Map([["dpt",["DPTForDepthEstimation",Wo]],["depth_anything",["DepthAnythingForDepthEstimation",Uo]],["glpn",["GLPNForDepthEstimation",os]],["sapiens",["SapiensForDepthEstimation",Ho]],["depth_pro",["DepthProForDepthEstimation",Ko]]]),Ii=new Map([["sapiens",["SapiensForNormalEstimation",Yo]]]),Bi=new Map([["clip",["CLIPVisionModelWithProjection",wr]],["siglip",["SiglipVisionModel",yr]]]),Ni=[[si,w],[ai,T],[ii,y],[ui,w],[pi,w],[hi,b],[li,b],[mi,y],[_i,w],[fi,w],[gi,x],[Mi,F],[Ti,w],[yi,w],[Fi,w],[ki,w],[Ei,w],[Li,w],[zi,w],[Ii,w],[bi,w],[xi,w],[Pi,k],[Ci,w],[vi,w],[ci,b],[di,w],[Si,w],[Ai,w],[Bi,w]];for(const[e,t]of Ni)for(const[r,n]of e.values())C.set(r,t),S.set(n,r),v.set(r,n);const Di=[["MusicgenForConditionalGeneration",qa,P],["CLIPTextModelWithProjection",gr,w],["SiglipTextModel",xr,w],["ClapTextModelWithProjection",Pa,w],["ClapAudioModelWithProjection",Ca,w]];for(const[e,t,r]of Di)C.set(e,r),S.set(t,e),v.set(e,t);class Oi extends oi{static MODEL_CLASS_MAPPINGS=Ni.map((e=>e[0]));static BASE_IF_FAIL=!0}class Vi extends oi{static MODEL_CLASS_MAPPINGS=[ui]}class ji extends oi{static MODEL_CLASS_MAPPINGS=[pi]}class Ri extends oi{static MODEL_CLASS_MAPPINGS=[hi]}class Gi extends oi{static MODEL_CLASS_MAPPINGS=[li]}class qi extends oi{static MODEL_CLASS_MAPPINGS=[ci]}class Wi extends oi{static MODEL_CLASS_MAPPINGS=[di]}class $i extends oi{static MODEL_CLASS_MAPPINGS=[mi]}class Ui extends oi{static MODEL_CLASS_MAPPINGS=[_i]}class Xi extends oi{static MODEL_CLASS_MAPPINGS=[fi]}class Qi extends oi{static MODEL_CLASS_MAPPINGS=[gi]}class Hi extends oi{static MODEL_CLASS_MAPPINGS=[Ti]}class Yi extends oi{static MODEL_CLASS_MAPPINGS=[yi]}class Ji extends oi{static MODEL_CLASS_MAPPINGS=[ki]}class Ki extends oi{static MODEL_CLASS_MAPPINGS=[Fi]}class Zi extends oi{static MODEL_CLASS_MAPPINGS=[bi]}class el extends oi{static MODEL_CLASS_MAPPINGS=[xi]}class tl extends oi{static MODEL_CLASS_MAPPINGS=[Pi]}class rl extends oi{static MODEL_CLASS_MAPPINGS=[Ci]}class nl extends oi{static MODEL_CLASS_MAPPINGS=[vi]}class ol extends oi{static MODEL_CLASS_MAPPINGS=[Si]}class sl extends oi{static MODEL_CLASS_MAPPINGS=[Ai]}class al extends oi{static MODEL_CLASS_MAPPINGS=[wi]}class il extends oi{static MODEL_CLASS_MAPPINGS=[Ei]}class ll extends oi{static MODEL_CLASS_MAPPINGS=[Li]}class cl extends oi{static MODEL_CLASS_MAPPINGS=[zi]}class dl extends oi{static MODEL_CLASS_MAPPINGS=[Ii]}class ul extends oi{static MODEL_CLASS_MAPPINGS=[Bi]}class pl extends W{constructor({logits:e,past_key_values:t,encoder_outputs:r,decoder_attentions:n=null,cross_attentions:o=null}){super(),this.logits=e,this.past_key_values=t,this.encoder_outputs=r,this.decoder_attentions=n,this.cross_attentions=o}}class hl extends W{constructor({logits:e}){super(),this.logits=e}}class ml extends W{constructor({logits:e,embeddings:t}){super(),this.logits=e,this.embeddings=t}}class _l extends W{constructor({logits:e}){super(),this.logits=e}}class fl extends W{constructor({logits:e}){super(),this.logits=e}}class gl extends W{constructor({start_logits:e,end_logits:t}){super(),this.start_logits=e,this.end_logits=t}}class Ml extends W{constructor({logits:e}){super(),this.logits=e}}class wl extends W{constructor({logits:e,past_key_values:t}){super(),this.logits=e,this.past_key_values=t}}class Tl extends W{constructor({alphas:e}){super(),this.alphas=e}}class bl extends W{constructor({waveform:e,spectrogram:t}){super(),this.waveform=e,this.spectrogram=t}}},"./src/models/whisper/common_whisper.js":
+  \***********************/(e,t,r)=>{r.r(t),r.d(t,{ASTForAudioClassification:()=>or,ASTModel:()=>nr,ASTPreTrainedModel:()=>rr,AlbertForMaskedLM:()=>ht,AlbertForQuestionAnswering:()=>pt,AlbertForSequenceClassification:()=>ut,AlbertModel:()=>dt,AlbertPreTrainedModel:()=>ct,AutoModel:()=>Oi,AutoModelForAudioClassification:()=>nl,AutoModelForAudioFrameClassification:()=>sl,AutoModelForCTC:()=>rl,AutoModelForCausalLM:()=>Wi,AutoModelForDepthEstimation:()=>cl,AutoModelForDocumentQuestionAnswering:()=>al,AutoModelForImageClassification:()=>Hi,AutoModelForImageFeatureExtraction:()=>ul,AutoModelForImageMatting:()=>il,AutoModelForImageSegmentation:()=>Yi,AutoModelForImageToImage:()=>ll,AutoModelForMaskGeneration:()=>tl,AutoModelForMaskedLM:()=>Ui,AutoModelForNormalEstimation:()=>dl,AutoModelForObjectDetection:()=>Zi,AutoModelForQuestionAnswering:()=>Xi,AutoModelForSemanticSegmentation:()=>Ji,AutoModelForSeq2SeqLM:()=>Ri,AutoModelForSequenceClassification:()=>Vi,AutoModelForSpeechSeq2Seq:()=>Gi,AutoModelForTextToSpectrogram:()=>qi,AutoModelForTextToWaveform:()=>$i,AutoModelForTokenClassification:()=>ji,AutoModelForUniversalSegmentation:()=>Ki,AutoModelForVision2Seq:()=>Qi,AutoModelForXVector:()=>ol,AutoModelForZeroShotObjectDetection:()=>el,BartForConditionalGeneration:()=>Ft,BartForSequenceClassification:()=>Pt,BartModel:()=>kt,BartPretrainedModel:()=>yt,BaseModelOutput:()=>W,BeitForImageClassification:()=>po,BeitModel:()=>uo,BeitPreTrainedModel:()=>co,BertForMaskedLM:()=>Q,BertForQuestionAnswering:()=>J,BertForSequenceClassification:()=>H,BertForTokenClassification:()=>Y,BertModel:()=>X,BertPreTrainedModel:()=>U,BlenderbotForConditionalGeneration:()=>It,BlenderbotModel:()=>zt,BlenderbotPreTrainedModel:()=>Lt,BlenderbotSmallForConditionalGeneration:()=>Dt,BlenderbotSmallModel:()=>Nt,BlenderbotSmallPreTrainedModel:()=>Bt,BloomForCausalLM:()=>Pn,BloomModel:()=>Fn,BloomPreTrainedModel:()=>kn,CLIPModel:()=>_r,CLIPPreTrainedModel:()=>mr,CLIPSegForImageSegmentation:()=>vr,CLIPSegModel:()=>Cr,CLIPSegPreTrainedModel:()=>Pr,CLIPTextModel:()=>fr,CLIPTextModelWithProjection:()=>gr,CLIPVisionModel:()=>Mr,CLIPVisionModelWithProjection:()=>wr,CamembertForMaskedLM:()=>Te,CamembertForQuestionAnswering:()=>ye,CamembertForSequenceClassification:()=>be,CamembertForTokenClassification:()=>xe,CamembertModel:()=>we,CamembertPreTrainedModel:()=>Me,CausalLMOutput:()=>Ml,CausalLMOutputWithPast:()=>wl,ChineseCLIPModel:()=>Fr,ChineseCLIPPreTrainedModel:()=>kr,ClapAudioModelWithProjection:()=>Ca,ClapModel:()=>Fa,ClapPreTrainedModel:()=>ka,ClapTextModelWithProjection:()=>Pa,CodeGenForCausalLM:()=>Hr,CodeGenModel:()=>Qr,CodeGenPreTrainedModel:()=>Xr,CohereForCausalLM:()=>on,CohereModel:()=>nn,CoherePreTrainedModel:()=>rn,ConvBertForMaskedLM:()=>le,ConvBertForQuestionAnswering:()=>ue,ConvBertForSequenceClassification:()=>ce,ConvBertForTokenClassification:()=>de,ConvBertModel:()=>ie,ConvBertPreTrainedModel:()=>ae,ConvNextForImageClassification:()=>cs,ConvNextModel:()=>ls,ConvNextPreTrainedModel:()=>is,ConvNextV2ForImageClassification:()=>ps,ConvNextV2Model:()=>us,ConvNextV2PreTrainedModel:()=>ds,DPTForDepthEstimation:()=>$o,DPTModel:()=>qo,DPTPreTrainedModel:()=>Go,DebertaForMaskedLM:()=>Pe,DebertaForQuestionAnswering:()=>Se,DebertaForSequenceClassification:()=>Ce,DebertaForTokenClassification:()=>ve,DebertaModel:()=>Fe,DebertaPreTrainedModel:()=>ke,DebertaV2ForMaskedLM:()=>Le,DebertaV2ForQuestionAnswering:()=>Be,DebertaV2ForSequenceClassification:()=>ze,DebertaV2ForTokenClassification:()=>Ie,DebertaV2Model:()=>Ee,DebertaV2PreTrainedModel:()=>Ae,DecisionTransformerModel:()=>ni,DecisionTransformerPreTrainedModel:()=>ri,DeiTForImageClassification:()=>So,DeiTModel:()=>vo,DeiTPreTrainedModel:()=>Co,DepthAnythingForDepthEstimation:()=>Uo,DepthAnythingPreTrainedModel:()=>Wo,DepthProForDepthEstimation:()=>Ko,DepthProPreTrainedModel:()=>Jo,DetrForObjectDetection:()=>_o,DetrForSegmentation:()=>fo,DetrModel:()=>mo,DetrObjectDetectionOutput:()=>go,DetrPreTrainedModel:()=>ho,DetrSegmentationOutput:()=>Mo,Dinov2ForImageClassification:()=>_s,Dinov2Model:()=>ms,Dinov2PreTrainedModel:()=>hs,DistilBertForMaskedLM:()=>Re,DistilBertForQuestionAnswering:()=>je,DistilBertForSequenceClassification:()=>Oe,DistilBertForTokenClassification:()=>Ve,DistilBertModel:()=>De,DistilBertPreTrainedModel:()=>Ne,DonutSwinModel:()=>as,DonutSwinPreTrainedModel:()=>ss,EfficientNetForImageClassification:()=>Va,EfficientNetModel:()=>Oa,EfficientNetPreTrainedModel:()=>Da,ElectraForMaskedLM:()=>me,ElectraForQuestionAnswering:()=>ge,ElectraForSequenceClassification:()=>_e,ElectraForTokenClassification:()=>fe,ElectraModel:()=>he,ElectraPreTrainedModel:()=>pe,EsmForMaskedLM:()=>$e,EsmForSequenceClassification:()=>We,EsmForTokenClassification:()=>Ue,EsmModel:()=>qe,EsmPreTrainedModel:()=>Ge,FalconForCausalLM:()=>ya,FalconModel:()=>xa,FalconPreTrainedModel:()=>ba,FastViTForImageClassification:()=>Qn,FastViTModel:()=>Xn,FastViTPreTrainedModel:()=>Un,Florence2ForConditionalGeneration:()=>hr,Florence2PreTrainedModel:()=>pr,GLPNForDepthEstimation:()=>os,GLPNModel:()=>ns,GLPNPreTrainedModel:()=>rs,GPT2LMHeadModel:()=>Er,GPT2Model:()=>Ar,GPT2PreTrainedModel:()=>Sr,GPTBigCodeForCausalLM:()=>Ur,GPTBigCodeModel:()=>Wr,GPTBigCodePreTrainedModel:()=>$r,GPTJForCausalLM:()=>qr,GPTJModel:()=>Gr,GPTJPreTrainedModel:()=>Rr,GPTNeoForCausalLM:()=>Dr,GPTNeoModel:()=>Nr,GPTNeoPreTrainedModel:()=>Br,GPTNeoXForCausalLM:()=>jr,GPTNeoXModel:()=>Vr,GPTNeoXPreTrainedModel:()=>Or,Gemma2ForCausalLM:()=>un,Gemma2Model:()=>dn,Gemma2PreTrainedModel:()=>cn,GemmaForCausalLM:()=>ln,GemmaModel:()=>an,GemmaPreTrainedModel:()=>sn,GraniteForCausalLM:()=>tn,GraniteModel:()=>en,GranitePreTrainedModel:()=>Zr,GroupViTModel:()=>Wn,GroupViTPreTrainedModel:()=>$n,HieraForImageClassification:()=>Lo,HieraModel:()=>Eo,HieraPreTrainedModel:()=>Ao,HubertForCTC:()=>ea,HubertForSequenceClassification:()=>ta,HubertModel:()=>Zs,HubertPreTrainedModel:()=>Ks,ImageMattingOutput:()=>Tl,JAISLMHeadModel:()=>Ir,JAISModel:()=>zr,JAISPreTrainedModel:()=>Lr,LlamaForCausalLM:()=>Kr,LlamaModel:()=>Jr,LlamaPreTrainedModel:()=>Yr,LlavaForConditionalGeneration:()=>dr,LlavaPreTrainedModel:()=>cr,LongT5ForConditionalGeneration:()=>wt,LongT5Model:()=>Mt,LongT5PreTrainedModel:()=>gt,M2M100ForConditionalGeneration:()=>vs,M2M100Model:()=>Cs,M2M100PreTrainedModel:()=>Ps,MBartForCausalLM:()=>Et,MBartForConditionalGeneration:()=>St,MBartForSequenceClassification:()=>At,MBartModel:()=>vt,MBartPreTrainedModel:()=>Ct,MPNetForMaskedLM:()=>et,MPNetForQuestionAnswering:()=>nt,MPNetForSequenceClassification:()=>tt,MPNetForTokenClassification:()=>rt,MPNetModel:()=>Ze,MPNetPreTrainedModel:()=>Ke,MT5ForConditionalGeneration:()=>xt,MT5Model:()=>bt,MT5PreTrainedModel:()=>Tt,MarianMTModel:()=>Fs,MarianModel:()=>ks,MarianPreTrainedModel:()=>ys,MaskFormerForInstanceSegmentation:()=>ts,MaskFormerModel:()=>es,MaskFormerPreTrainedModel:()=>Zo,MaskedLMOutput:()=>fl,MistralForCausalLM:()=>ga,MistralModel:()=>fa,MistralPreTrainedModel:()=>_a,MobileBertForMaskedLM:()=>He,MobileBertForQuestionAnswering:()=>Je,MobileBertForSequenceClassification:()=>Ye,MobileBertModel:()=>Qe,MobileBertPreTrainedModel:()=>Xe,MobileNetV1ForImageClassification:()=>Ua,MobileNetV1Model:()=>Wa,MobileNetV1PreTrainedModel:()=>$a,MobileNetV2ForImageClassification:()=>Ha,MobileNetV2Model:()=>Qa,MobileNetV2PreTrainedModel:()=>Xa,MobileNetV3ForImageClassification:()=>Ka,MobileNetV3Model:()=>Ja,MobileNetV3PreTrainedModel:()=>Ya,MobileNetV4ForImageClassification:()=>ti,MobileNetV4Model:()=>ei,MobileNetV4PreTrainedModel:()=>Za,MobileViTForImageClassification:()=>Zn,MobileViTModel:()=>Kn,MobileViTPreTrainedModel:()=>Jn,MobileViTV2ForImageClassification:()=>ro,MobileViTV2Model:()=>to,MobileViTV2PreTrainedModel:()=>eo,ModelOutput:()=>$,Moondream1ForConditionalGeneration:()=>ur,MptForCausalLM:()=>Sn,MptModel:()=>vn,MptPreTrainedModel:()=>Cn,MusicgenForCausalLM:()=>Ga,MusicgenForConditionalGeneration:()=>qa,MusicgenModel:()=>Ra,MusicgenPreTrainedModel:()=>ja,NomicBertModel:()=>Z,NomicBertPreTrainedModel:()=>K,OPTForCausalLM:()=>Ln,OPTModel:()=>En,OPTPreTrainedModel:()=>An,OpenELMForCausalLM:()=>mn,OpenELMModel:()=>hn,OpenELMPreTrainedModel:()=>pn,OwlViTForObjectDetection:()=>so,OwlViTModel:()=>oo,OwlViTPreTrainedModel:()=>no,Owlv2ForObjectDetection:()=>lo,Owlv2Model:()=>io,Owlv2PreTrainedModel:()=>ao,Phi3ForCausalLM:()=>yn,Phi3Model:()=>xn,Phi3PreTrainedModel:()=>bn,PhiForCausalLM:()=>Tn,PhiModel:()=>wn,PhiPreTrainedModel:()=>Mn,PreTrainedModel:()=>q,PretrainedMixin:()=>oi,PvtForImageClassification:()=>On,PvtModel:()=>Dn,PvtPreTrainedModel:()=>Nn,PyAnnoteForAudioFrameClassification:()=>Ns,PyAnnoteModel:()=>Bs,PyAnnotePreTrainedModel:()=>Is,QuestionAnsweringModelOutput:()=>gl,Qwen2ForCausalLM:()=>gn,Qwen2Model:()=>fn,Qwen2PreTrainedModel:()=>_n,RTDetrForObjectDetection:()=>bo,RTDetrModel:()=>To,RTDetrObjectDetectionOutput:()=>xo,RTDetrPreTrainedModel:()=>wo,ResNetForImageClassification:()=>Bo,ResNetModel:()=>Io,ResNetPreTrainedModel:()=>zo,RoFormerForMaskedLM:()=>re,RoFormerForQuestionAnswering:()=>se,RoFormerForSequenceClassification:()=>ne,RoFormerForTokenClassification:()=>oe,RoFormerModel:()=>te,RoFormerPreTrainedModel:()=>ee,RobertaForMaskedLM:()=>jt,RobertaForQuestionAnswering:()=>qt,RobertaForSequenceClassification:()=>Rt,RobertaForTokenClassification:()=>Gt,RobertaModel:()=>Vt,RobertaPreTrainedModel:()=>Ot,SamImageSegmentationOutput:()=>xs,SamModel:()=>bs,SamPreTrainedModel:()=>Ts,SapiensForDepthEstimation:()=>Ho,SapiensForNormalEstimation:()=>Yo,SapiensForSemanticSegmentation:()=>Qo,SapiensPreTrainedModel:()=>Xo,SegformerForImageClassification:()=>La,SegformerForSemanticSegmentation:()=>za,SegformerModel:()=>Ea,SegformerPreTrainedModel:()=>Aa,Seq2SeqLMOutput:()=>pl,SequenceClassifierOutput:()=>hl,SiglipModel:()=>br,SiglipPreTrainedModel:()=>Tr,SiglipTextModel:()=>xr,SiglipVisionModel:()=>yr,SpeechT5ForSpeechToText:()=>da,SpeechT5ForTextToSpeech:()=>ua,SpeechT5HifiGan:()=>pa,SpeechT5Model:()=>ca,SpeechT5PreTrainedModel:()=>la,SqueezeBertForMaskedLM:()=>at,SqueezeBertForQuestionAnswering:()=>lt,SqueezeBertForSequenceClassification:()=>it,SqueezeBertModel:()=>st,SqueezeBertPreTrainedModel:()=>ot,StableLmForCausalLM:()=>Na,StableLmModel:()=>Ba,StableLmPreTrainedModel:()=>Ia,Starcoder2ForCausalLM:()=>Ta,Starcoder2Model:()=>wa,Starcoder2PreTrainedModel:()=>Ma,Swin2SRForImageSuperResolution:()=>Ro,Swin2SRModel:()=>jo,Swin2SRPreTrainedModel:()=>Vo,SwinForImageClassification:()=>Oo,SwinModel:()=>Do,SwinPreTrainedModel:()=>No,T5ForConditionalGeneration:()=>ft,T5Model:()=>_t,T5PreTrainedModel:()=>mt,TableTransformerForObjectDetection:()=>Fo,TableTransformerModel:()=>ko,TableTransformerObjectDetectionOutput:()=>Po,TableTransformerPreTrainedModel:()=>yo,TokenClassifierOutput:()=>_l,TrOCRForCausalLM:()=>ma,TrOCRPreTrainedModel:()=>ha,UniSpeechForCTC:()=>Rs,UniSpeechForSequenceClassification:()=>Gs,UniSpeechModel:()=>js,UniSpeechPreTrainedModel:()=>Vs,UniSpeechSatForAudioFrameClassification:()=>Xs,UniSpeechSatForCTC:()=>Ws,UniSpeechSatForSequenceClassification:()=>Us,UniSpeechSatModel:()=>$s,UniSpeechSatPreTrainedModel:()=>qs,ViTForImageClassification:()=>Bn,ViTMAEModel:()=>jn,ViTMAEPreTrainedModel:()=>Vn,ViTMSNForImageClassification:()=>qn,ViTMSNModel:()=>Gn,ViTMSNPreTrainedModel:()=>Rn,ViTModel:()=>In,ViTPreTrainedModel:()=>zn,VisionEncoderDecoderModel:()=>lr,VitMatteForImageMatting:()=>Yn,VitMattePreTrainedModel:()=>Hn,VitsModel:()=>Sa,VitsModelOutput:()=>bl,VitsPreTrainedModel:()=>va,Wav2Vec2BertForCTC:()=>Ys,Wav2Vec2BertForSequenceClassification:()=>Js,Wav2Vec2BertModel:()=>Hs,Wav2Vec2BertPreTrainedModel:()=>Qs,Wav2Vec2ForAudioFrameClassification:()=>zs,Wav2Vec2ForCTC:()=>Es,Wav2Vec2ForSequenceClassification:()=>Ls,Wav2Vec2Model:()=>As,Wav2Vec2PreTrainedModel:()=>Ss,WavLMForAudioFrameClassification:()=>ia,WavLMForCTC:()=>oa,WavLMForSequenceClassification:()=>sa,WavLMForXVector:()=>aa,WavLMModel:()=>na,WavLMPreTrainedModel:()=>ra,WeSpeakerResNetModel:()=>Os,WeSpeakerResNetPreTrainedModel:()=>Ds,WhisperForConditionalGeneration:()=>ir,WhisperModel:()=>ar,WhisperPreTrainedModel:()=>sr,XLMForQuestionAnswering:()=>Ht,XLMForSequenceClassification:()=>Xt,XLMForTokenClassification:()=>Qt,XLMModel:()=>Wt,XLMPreTrainedModel:()=>$t,XLMRobertaForMaskedLM:()=>Kt,XLMRobertaForQuestionAnswering:()=>tr,XLMRobertaForSequenceClassification:()=>Zt,XLMRobertaForTokenClassification:()=>er,XLMRobertaModel:()=>Jt,XLMRobertaPreTrainedModel:()=>Yt,XLMWithLMHeadModel:()=>Ut,XVectorOutput:()=>ml,YolosForObjectDetection:()=>Ms,YolosModel:()=>gs,YolosObjectDetectionOutput:()=>ws,YolosPreTrainedModel:()=>fs});var n=r(/*! ./configs.js */"./src/configs.js"),o=r(/*! ./backends/onnx.js */"./src/backends/onnx.js"),s=r(/*! ./utils/dtypes.js */"./src/utils/dtypes.js"),a=r(/*! ./utils/generic.js */"./src/utils/generic.js"),i=r(/*! ./utils/core.js */"./src/utils/core.js"),l=r(/*! ./utils/hub.js */"./src/utils/hub.js"),c=r(/*! ./utils/constants.js */"./src/utils/constants.js"),d=r(/*! ./generation/logits_process.js */"./src/generation/logits_process.js"),u=r(/*! ./generation/configuration_utils.js */"./src/generation/configuration_utils.js"),p=r(/*! ./utils/tensor.js */"./src/utils/tensor.js"),h=r(/*! ./utils/maths.js */"./src/utils/maths.js"),m=r(/*! ./generation/stopping_criteria.js */"./src/generation/stopping_criteria.js"),_=r(/*! ./generation/logits_sampler.js */"./src/generation/logits_sampler.js"),f=r(/*! ./env.js */"./src/env.js"),g=r(/*! ./models/whisper/generation_whisper.js */"./src/models/whisper/generation_whisper.js"),M=r(/*! ./models/whisper/common_whisper.js */"./src/models/whisper/common_whisper.js");const w=0,T=1,b=2,x=3,y=4,k=5,F=6,P=7,C=new Map,v=new Map,S=new Map;async function A(e,t,r){return Object.fromEntries(await Promise.all(Object.keys(t).map((async a=>{const{buffer:i,session_options:c,session_config:d}=await async function(e,t,r){const a=r.config?.["transformers.js_config"]??{};let i=r.device??a.device;i&&"string"!=typeof i&&(i.hasOwnProperty(t)?i=i[t]:(console.warn(`device not specified for "${t}". Using the default device.`),i=null));const c=i??(f.apis.IS_NODE_ENV?"cpu":"wasm"),d=(0,o.deviceToExecutionProviders)(c);let u=r.dtype??a.dtype;"string"!=typeof u&&(u&&u.hasOwnProperty(t)?u=u[t]:(u=s.DEFAULT_DEVICE_DTYPE_MAPPING[c]??s.DATA_TYPES.fp32,console.warn(`dtype not specified for "${t}". Using the default dtype (${u}) for this device (${c}).`)));const p=u;if(!s.DEFAULT_DTYPE_SUFFIX_MAPPING.hasOwnProperty(p))throw new Error(`Invalid dtype: ${p}. Should be one of: ${Object.keys(s.DATA_TYPES).join(", ")}`);if(p===s.DATA_TYPES.fp16&&"webgpu"===c&&!await(0,s.isWebGpuFp16Supported)())throw new Error(`The device (${c}) does not support fp16.`);const h=a.kv_cache_dtype?"string"==typeof a.kv_cache_dtype?a.kv_cache_dtype:a.kv_cache_dtype[p]??"float32":void 0;if(h&&!["float32","float16"].includes(h))throw new Error(`Invalid kv_cache_dtype: ${h}. Should be one of: float32, float16`);const m={dtype:p,kv_cache_dtype:h},_=s.DEFAULT_DTYPE_SUFFIX_MAPPING[p],g=`${r.subfolder??""}/${t}${_}.onnx`,M={...r.session_options};M.executionProviders??=d;const w=a.free_dimension_overrides;w?M.freeDimensionOverrides??=w:c.startsWith("webnn")&&!M.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 T=(0,l.getModelFile)(e,g,!0,r),b=r.use_external_data_format??a.use_external_data_format;let x=[];if(b&&(!0===b||"object"==typeof b&&b.hasOwnProperty(t)&&!0===b[t])){if(f.apis.IS_NODE_ENV)throw new Error("External data format is not yet supported in Node.js");const n=`${t}${_}.onnx_data`,o=`${r.subfolder??""}/${n}`;x.push(new Promise((async(t,s)=>{const a=await(0,l.getModelFile)(e,o,!0,r);t({path:n,data:a})})))}else void 0!==M.externalData&&(x=M.externalData.map((async t=>{if("string"==typeof t.data){const n=await(0,l.getModelFile)(e,t.data,!0,r);return{...t,data:n}}return t})));if(x.length>0&&(M.externalData=await Promise.all(x)),"webgpu"===c){const e=(0,n.getKeyValueShapes)(r.config,{prefix:"present"});if(Object.keys(e).length>0&&!(0,o.isONNXProxy)()){const t={};for(const r in e)t[r]="gpu-buffer";M.preferredOutputLocation=t}}return{buffer:await T,session_options:M,session_config:m}}(e,t[a],r);return[a,await(0,o.createInferenceSession)(i,c,d)]}))))}async function E(e,t,r){return Object.fromEntries(await Promise.all(Object.keys(t).map((async n=>[n,await(0,l.getModelJSON)(e,t[n],!1,r)]))))}async function L(e,t){const r=function(e,t){const r=Object.create(null),n=[];for(const s of e.inputNames){const e=t[s];e instanceof p.Tensor?r[s]=(0,o.isONNXProxy)()?e.clone():e:n.push(s)}if(n.length>0)throw new Error(`An error occurred during model execution: "Missing the following inputs: ${n.join(", ")}.`);const s=Object.keys(t).length,a=e.inputNames.length;if(s>a){let r=Object.keys(t).filter((t=>!e.inputNames.includes(t)));console.warn(`WARNING: Too many inputs were provided (${s} > ${a}). The following inputs will be ignored: "${r.join(", ")}".`)}return r}(e,t);try{const t=Object.fromEntries(Object.entries(r).map((([e,t])=>[e,t.ort_tensor])));let n=await e.run(t);return n=z(n),n}catch(e){throw console.error(`An error occurred during model execution: "${e}".`),console.error("Inputs given to model:",r),e}}function z(e){for(let t in e)(0,o.isONNXTensor)(e[t])?e[t]=new p.Tensor(e[t]):"object"==typeof e[t]&&z(e[t]);return e}function I(e){if(e instanceof p.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 p.Tensor("int64",BigInt64Array.from(e.flat().map((e=>BigInt(e)))),[e.length,e[0].length])}return new p.Tensor("int64",BigInt64Array.from(e.map((e=>BigInt(e)))),[1,e.length])}function B(e){return new p.Tensor("bool",[e],[1])}async function N(e,t){let{encoder_outputs:r,input_ids:n,decoder_input_ids:o,...s}=t;if(!r){const n=(0,i.pick)(t,e.sessions.model.inputNames);r=(await D(e,n)).last_hidden_state}s.input_ids=o,s.encoder_hidden_states=r,e.sessions.decoder_model_merged.inputNames.includes("encoder_attention_mask")&&(s.encoder_attention_mask=t.attention_mask);return await O(e,s,!0)}async function D(e,t){const r=e.sessions.model,n=(0,i.pick)(t,r.inputNames);if(r.inputNames.includes("inputs_embeds")&&!n.inputs_embeds){if(!t.input_ids)throw new Error("Both `input_ids` and `inputs_embeds` are missing in the model inputs.");n.inputs_embeds=await e.encode_text({input_ids:t.input_ids})}return r.inputNames.includes("token_type_ids")&&!n.token_type_ids&&(n.token_type_ids=new p.Tensor("int64",new BigInt64Array(n.input_ids.data.length),n.input_ids.dims)),await L(r,n)}async function O(e,t,r=!1){const n=e.sessions[r?"decoder_model_merged":"model"],{past_key_values:o,...s}=t;n.inputNames.includes("use_cache_branch")&&(s.use_cache_branch=B(!!o)),n.inputNames.includes("position_ids")&&s.attention_mask&&!s.position_ids&&(s.position_ids=function(e,t=null){const{input_ids:r,inputs_embeds:n,attention_mask:o}=e,[s,a]=o.dims,i=new BigInt64Array(o.data.length);for(let e=0;e<s;++e){const t=e*a;let r=BigInt(0);for(let e=0;e<a;++e){const n=t+e;0n===o.data[n]?i[n]=BigInt(1):(i[n]=r,r+=o.data[n])}}let l=new p.Tensor("int64",i,o.dims);if(t){const e=-(r??n).dims.at(1);l=l.slice(null,[e,null])}return l}(s,o)),e.addPastKeyValues(s,o);const a=(0,i.pick)(s,n.inputNames);return await L(n,a)}async function V(e,{input_ids:t=null,attention_mask:r=null,pixel_values:n=null,position_ids:o=null,inputs_embeds:s=null,past_key_values:a=null,generation_config:i=null,logits_processor:l=null,...c}){if(!s)if(s=await e.encode_text({input_ids:t}),n&&1!==t.dims[1]){const o=await e.encode_image({pixel_values:n});({inputs_embeds:s,attention_mask:r}=e._merge_input_ids_with_image_features({image_features:o,inputs_embeds:s,input_ids:t,attention_mask:r}))}else if(a&&n&&1===t.dims[1]){const e=t.dims[1],n=Object.values(a)[0].dims.at(-2);r=(0,p.cat)([(0,p.ones)([t.dims[0],n]),r.slice(null,[r.dims[1]-e,r.dims[1]])],1)}return await O(e,{inputs_embeds:s,past_key_values:a,attention_mask:r,position_ids:o,generation_config:i,logits_processor:l},!0)}function j(e,t,r,n){if(r.past_key_values){const t=Object.values(r.past_key_values)[0].dims.at(-2),{input_ids:n,attention_mask:o}=r;if(o&&o.dims[1]>n.dims[1]);else if(t<n.dims[1])r.input_ids=n.slice(null,[t,null]);else if(null!=e.config.image_token_index&&n.data.some((t=>t==e.config.image_token_index))){const o=e.config.num_image_tokens;if(!o)throw new Error("`num_image_tokens` is missing in the model configuration.");const s=n.dims[1]-(t-o);r.input_ids=n.slice(null,[-s,null]),r.attention_mask=(0,p.ones)([1,t+s])}}return r}function R(e,t,r,n){return r.past_key_values&&(t=t.map((e=>[e.at(-1)]))),{...r,decoder_input_ids:I(t)}}function G(e,...t){return e.config.is_encoder_decoder?R(e,...t):j(e,...t)}class q extends a.Callable{main_input_name="input_ids";forward_params=["input_ids","attention_mask"];constructor(e,t,r){super(),this.config=e,this.sessions=t,this.configs=r;const n=S.get(this.constructor),o=C.get(n);switch(this.can_generate=!1,this._forward=null,this._prepare_inputs_for_generation=null,o){case y:this.can_generate=!0,this._forward=O,this._prepare_inputs_for_generation=j;break;case b:case x:case P:this.can_generate=!0,this._forward=N,this._prepare_inputs_for_generation=R;break;case T:this._forward=N;break;case F:this.can_generate=!0,this._forward=V,this._prepare_inputs_for_generation=G;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:r=null,cache_dir:o=null,local_files_only:s=!1,revision:a="main",model_file_name:i=null,subfolder:l="onnx",device:d=null,dtype:u=null,use_external_data_format:p=null,session_options:h={}}={}){let m={progress_callback:t,config:r,cache_dir:o,local_files_only:s,revision:a,model_file_name:i,subfolder:l,device:d,dtype:u,use_external_data_format:p,session_options:h};const _=S.get(this),f=C.get(_);let g;if(r=m.config=await n.AutoConfig.from_pretrained(e,m),f===y)g=await Promise.all([A(e,{model:m.model_file_name??"model"},m),E(e,{generation_config:"generation_config.json"},m)]);else if(f===b||f===x)g=await Promise.all([A(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},m),E(e,{generation_config:"generation_config.json"},m)]);else if(f===k)g=await Promise.all([A(e,{model:"vision_encoder",prompt_encoder_mask_decoder:"prompt_encoder_mask_decoder"},m)]);else if(f===T)g=await Promise.all([A(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},m)]);else if(f===F){const t={embed_tokens:"embed_tokens",vision_encoder:"vision_encoder",decoder_model_merged:"decoder_model_merged"};r.is_encoder_decoder&&(t.model="encoder_model"),g=await Promise.all([A(e,t,m),E(e,{generation_config:"generation_config.json"},m)])}else f===P?g=await Promise.all([A(e,{model:"text_encoder",decoder_model_merged:"decoder_model_merged",encodec_decode:"encodec_decode"},m),E(e,{generation_config:"generation_config.json"},m)]):(f!==w&&console.warn(`Model type for '${_??r?.model_type}' not found, assuming encoder-only architecture. Please report this at ${c.GITHUB_ISSUE_URL}.`),g=await Promise.all([A(e,{model:m.model_file_name??"model"},m)]));return new this(r,...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,r=null){const n=new d.LogitsProcessorList;if(null!==e.repetition_penalty&&1!==e.repetition_penalty&&n.push(new d.RepetitionPenaltyLogitsProcessor(e.repetition_penalty)),null!==e.no_repeat_ngram_size&&e.no_repeat_ngram_size>0&&n.push(new d.NoRepeatNGramLogitsProcessor(e.no_repeat_ngram_size)),null!==e.bad_words_ids&&n.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&&n.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&&n.push(new d.MinNewTokensLengthLogitsProcessor(t,e.min_new_tokens,e.eos_token_id)),null!==e.forced_bos_token_id&&n.push(new d.ForcedBOSTokenLogitsProcessor(e.forced_bos_token_id)),null!==e.forced_eos_token_id&&n.push(new d.ForcedEOSTokenLogitsProcessor(e.max_length,e.forced_eos_token_id)),null!==e.begin_suppress_tokens){const r=t>1||null===e.forced_bos_token_id?t:t+1;n.push(new d.SuppressTokensAtBeginLogitsProcessor(e.begin_suppress_tokens,r))}return null!==e.guidance_scale&&e.guidance_scale>1&&n.push(new d.ClassifierFreeGuidanceLogitsProcessor(e.guidance_scale)),null!==r&&n.extend(r),n}_prepare_generation_config(e,t,r=u.GenerationConfig){const n={...this.config};for(const e of["decoder","generator","text_config"])e in n&&Object.assign(n,n[e]);const o=new r(n);return Object.assign(o,this.generation_config??{}),e&&Object.assign(o,e),t&&Object.assign(o,(0,i.pick)(t,Object.getOwnPropertyNames(o))),o}_get_stopping_criteria(e,t=null){const r=new m.StoppingCriteriaList;return null!==e.max_length&&r.push(new m.MaxLengthCriteria(e.max_length,this.config.max_position_embeddings??null)),null!==e.eos_token_id&&r.push(new m.EosTokenCriteria(e.eos_token_id)),t&&r.extend(t),r}_validate_model_class(){if(!this.can_generate){const e=[mi,gi,hi,li],t=S.get(this.constructor),r=new Set,n=this.config.model_type;for(const t of e){const e=t.get(n);e&&r.add(e[0])}let o=`The current model class (${t}) is not compatible with \`.generate()\`, as it doesn't have a language model head.`;throw r.size>0&&(o+=` Please use the following class instead: ${[...r].join(", ")}`),Error(o)}}prepare_inputs_for_generation(...e){return this._prepare_inputs_for_generation(this,...e)}_update_model_kwargs_for_generation({generated_input_ids:e,outputs:t,model_inputs:r,is_encoder_decoder:n}){return r.past_key_values=this.getPastKeyValues(t,r.past_key_values),r.input_ids=new p.Tensor("int64",e.flat(),[e.length,1]),n||(r.attention_mask=(0,p.cat)([r.attention_mask,(0,p.ones)([r.attention_mask.dims[0],1])],1)),r.position_ids=null,r}_prepare_model_inputs({inputs:e,bos_token_id:t,model_kwargs:r}){const n=(0,i.pick)(r,this.forward_params),o=this.main_input_name;if(o in n){if(e)throw new Error("`inputs`: {inputs}` were passed alongside {input_name} which is not allowed. Make sure to either pass {inputs} or {input_name}=...")}else n[o]=e;return{inputs_tensor:n[o],model_inputs:n,model_input_name:o}}async _prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:e,model_inputs:t,model_input_name:r,generation_config:n}){if(this.sessions.model.inputNames.includes("inputs_embeds")&&!t.inputs_embeds&&"_prepare_inputs_embeds"in this){const{input_ids:e,pixel_values:r,attention_mask:n,...o}=t,s=await this._prepare_inputs_embeds(t);t={...o,...(0,i.pick)(s,["inputs_embeds","attention_mask"])}}let{last_hidden_state:o}=await D(this,t);if(null!==n.guidance_scale&&n.guidance_scale>1)o=(0,p.cat)([o,(0,p.full_like)(o,0)],0),"attention_mask"in t&&(t.attention_mask=(0,p.cat)([t.attention_mask,(0,p.zeros_like)(t.attention_mask)],0));else if(t.decoder_input_ids){const e=I(t.decoder_input_ids).dims[0];if(e!==o.dims[0]){if(1!==o.dims[0])throw new Error(`The encoder outputs have a different batch size (${o.dims[0]}) than the decoder inputs (${e}).`);o=(0,p.cat)(Array.from({length:e},(()=>o)),0)}}return t.encoder_outputs=o,t}_prepare_decoder_input_ids_for_generation({batch_size:e,model_input_name:t,model_kwargs:r,decoder_start_token_id:n,bos_token_id:o,generation_config:s}){let{decoder_input_ids:a,...i}=r;if(!(a instanceof p.Tensor)){if(a)Array.isArray(a[0])||(a=Array.from({length:e},(()=>a)));else if(n??=o,"musicgen"===this.config.model_type)a=Array.from({length:e*this.config.decoder.num_codebooks},(()=>[n]));else if(Array.isArray(n)){if(n.length!==e)throw new Error(`\`decoder_start_token_id\` expcted to have length ${e} but got ${n.length}`);a=n}else a=Array.from({length:e},(()=>[n]));a=I(a)}return r.decoder_attention_mask=(0,p.ones_like)(a),{input_ids:a,model_inputs:i}}async generate({inputs:e=null,generation_config:t=null,logits_processor:r=null,stopping_criteria:n=null,streamer:o=null,...s}){this._validate_model_class(),t=this._prepare_generation_config(t,s);let{inputs_tensor:a,model_inputs:i,model_input_name:l}=this._prepare_model_inputs({inputs:e,model_kwargs:s});const c=this.config.is_encoder_decoder;let d;c&&("encoder_outputs"in i||(i=await this._prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:a,model_inputs:i,model_input_name:l,generation_config:t}))),c?({input_ids:d,model_inputs:i}=this._prepare_decoder_input_ids_for_generation({batch_size:i[l].dims.at(0),model_input_name:l,model_kwargs:i,decoder_start_token_id:t.decoder_start_token_id,bos_token_id:t.bos_token_id,generation_config:t})):d=i[l];let u=d.dims.at(-1);null!==t.max_new_tokens&&(t.max_length=u+t.max_new_tokens);const h=this._get_logits_processor(t,u,r),m=this._get_stopping_criteria(t,n),f=i[l].dims.at(0),g=_.LogitsSampler.getSampler(t),M=new Array(f).fill(0),w=d.tolist();let T;o&&o.put(w);let b={};for(;;){if(i=this.prepare_inputs_for_generation(w,i,t),T=await this.forward(i),t.output_attentions&&t.return_dict_in_generate){const e=this.getAttentions(T);for(const t in e)t in b||(b[t]=[]),b[t].push(e[t])}const e=h(w,T.logits.slice(null,-1,null)),r=[];for(let t=0;t<e.dims.at(0);++t){const n=e[t],o=await g(n);for(const[e,n]of o){const o=BigInt(e);M[t]+=n,w[t].push(o),r.push([o]);break}}o&&o.put(r);if(m(w).every((e=>e)))break;i=this._update_model_kwargs_for_generation({generated_input_ids:r,outputs:T,model_inputs:i,is_encoder_decoder:c})}o&&o.end();const x=this.getPastKeyValues(T,i.past_key_values,!0),y=new p.Tensor("int64",w.flat(),[w.length,w[0].length]);if(t.return_dict_in_generate)return{sequences:y,past_key_values:x,...b};for(const e of Object.values(T))"gpu-buffer"===e.location&&e.dispose();return y}getPastKeyValues(e,t,r=!1){const n=Object.create(null);for(const o in e)if(o.startsWith("present")){const s=o.replace("present","past_key_values"),a=o.includes("encoder");if(n[s]=a&&t?t[s]:e[o],t&&(!a||r)){const e=t[s];"gpu-buffer"===e.location&&e.dispose()}}return n}getAttentions(e){const t={};for(const r of["cross_attentions","encoder_attentions","decoder_attentions"])for(const n in e)n.startsWith(r)&&(r in t||(t[r]=[]),t[r].push(e[n]));return t}addPastKeyValues(e,t){if(t)Object.assign(e,t);else{const t=this.sessions.decoder_model_merged??this.sessions.model,r=t?.config?.kv_cache_dtype??"float32",o="float16"===r?new Uint16Array:[],s=(0,n.getKeyValueShapes)(this.config);for(const t in s)e[t]=new p.Tensor(r,o,s[t])}}async encode_image({pixel_values:e}){const t=(await L(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 L(this.sessions.embed_tokens,{input_ids:e})).inputs_embeds}}class ${}class W extends ${constructor({last_hidden_state:e,hidden_states:t=null,attentions:r=null}){super(),this.last_hidden_state=e,this.hidden_states=t,this.attentions=r}}class U extends q{}class X extends U{}class Q extends U{async _call(e){return new fl(await super._call(e))}}class H extends U{async _call(e){return new hl(await super._call(e))}}class Y extends U{async _call(e){return new _l(await super._call(e))}}class J extends U{async _call(e){return new gl(await super._call(e))}}class K extends q{}class Z extends K{}class ee extends q{}class te extends ee{}class re extends ee{async _call(e){return new fl(await super._call(e))}}class ne extends ee{async _call(e){return new hl(await super._call(e))}}class oe extends ee{async _call(e){return new _l(await super._call(e))}}class se extends ee{async _call(e){return new gl(await super._call(e))}}class ae extends q{}class ie extends ae{}class le extends ae{async _call(e){return new fl(await super._call(e))}}class ce extends ae{async _call(e){return new hl(await super._call(e))}}class de extends ae{async _call(e){return new _l(await super._call(e))}}class ue extends ae{async _call(e){return new gl(await super._call(e))}}class pe extends q{}class he extends pe{}class me extends pe{async _call(e){return new fl(await super._call(e))}}class _e extends pe{async _call(e){return new hl(await super._call(e))}}class fe extends pe{async _call(e){return new _l(await super._call(e))}}class ge extends pe{async _call(e){return new gl(await super._call(e))}}class Me extends q{}class we extends Me{}class Te extends Me{async _call(e){return new fl(await super._call(e))}}class be extends Me{async _call(e){return new hl(await super._call(e))}}class xe extends Me{async _call(e){return new _l(await super._call(e))}}class ye extends Me{async _call(e){return new gl(await super._call(e))}}class ke extends q{}class Fe extends ke{}class Pe extends ke{async _call(e){return new fl(await super._call(e))}}class Ce extends ke{async _call(e){return new hl(await super._call(e))}}class ve extends ke{async _call(e){return new _l(await super._call(e))}}class Se extends ke{async _call(e){return new gl(await super._call(e))}}class Ae extends q{}class Ee extends Ae{}class Le extends Ae{async _call(e){return new fl(await super._call(e))}}class ze extends Ae{async _call(e){return new hl(await super._call(e))}}class Ie extends Ae{async _call(e){return new _l(await super._call(e))}}class Be extends Ae{async _call(e){return new gl(await super._call(e))}}class Ne extends q{}class De extends Ne{}class Oe extends Ne{async _call(e){return new hl(await super._call(e))}}class Ve extends Ne{async _call(e){return new _l(await super._call(e))}}class je extends Ne{async _call(e){return new gl(await super._call(e))}}class Re extends Ne{async _call(e){return new fl(await super._call(e))}}class Ge extends q{}class qe extends Ge{}class $e extends Ge{async _call(e){return new fl(await super._call(e))}}class We extends Ge{async _call(e){return new hl(await super._call(e))}}class Ue extends Ge{async _call(e){return new _l(await super._call(e))}}class Xe extends q{}class Qe extends Xe{}class He extends Xe{async _call(e){return new fl(await super._call(e))}}class Ye extends Xe{async _call(e){return new hl(await super._call(e))}}class Je extends Xe{async _call(e){return new gl(await super._call(e))}}class Ke extends q{}class Ze extends Ke{}class et extends Ke{async _call(e){return new fl(await super._call(e))}}class tt extends Ke{async _call(e){return new hl(await super._call(e))}}class rt extends Ke{async _call(e){return new _l(await super._call(e))}}class nt extends Ke{async _call(e){return new gl(await super._call(e))}}class ot extends q{}class st extends ot{}class at extends ot{async _call(e){return new fl(await super._call(e))}}class it extends ot{async _call(e){return new hl(await super._call(e))}}class lt extends ot{async _call(e){return new gl(await super._call(e))}}class ct extends q{}class dt extends ct{}class ut extends ct{async _call(e){return new hl(await super._call(e))}}class pt extends ct{async _call(e){return new gl(await super._call(e))}}class ht extends ct{async _call(e){return new fl(await super._call(e))}}class mt extends q{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"]}class _t extends mt{}class ft extends mt{}class gt extends q{}class Mt extends gt{}class wt extends gt{}class Tt extends q{}class bt extends Tt{}class xt extends Tt{}class yt extends q{}class kt extends yt{}class Ft extends yt{}class Pt extends yt{async _call(e){return new hl(await super._call(e))}}class Ct extends q{}class vt extends Ct{}class St extends Ct{}class At extends Ct{async _call(e){return new hl(await super._call(e))}}class Et extends Ct{}class Lt extends q{}class zt extends Lt{}class It extends Lt{}class Bt extends q{}class Nt extends Bt{}class Dt extends Bt{}class Ot extends q{}class Vt extends Ot{}class jt extends Ot{async _call(e){return new fl(await super._call(e))}}class Rt extends Ot{async _call(e){return new hl(await super._call(e))}}class Gt extends Ot{async _call(e){return new _l(await super._call(e))}}class qt extends Ot{async _call(e){return new gl(await super._call(e))}}class $t extends q{}class Wt extends $t{}class Ut extends $t{async _call(e){return new fl(await super._call(e))}}class Xt extends $t{async _call(e){return new hl(await super._call(e))}}class Qt extends $t{async _call(e){return new _l(await super._call(e))}}class Ht extends $t{async _call(e){return new gl(await super._call(e))}}class Yt extends q{}class Jt extends Yt{}class Kt extends Yt{async _call(e){return new fl(await super._call(e))}}class Zt extends Yt{async _call(e){return new hl(await super._call(e))}}class er extends Yt{async _call(e){return new _l(await super._call(e))}}class tr extends Yt{async _call(e){return new gl(await super._call(e))}}class rr extends q{}class nr extends rr{}class or extends rr{}class sr extends q{requires_attention_mask=!1;main_input_name="input_features";forward_params=["input_features","attention_mask","decoder_input_ids","decoder_attention_mask","past_key_values"]}class ar extends sr{}class ir extends sr{_prepare_generation_config(e,t){return super._prepare_generation_config(e,t,g.WhisperGenerationConfig)}_retrieve_init_tokens(e){const t=[e.decoder_start_token_id];let r=e.language;const n=e.task;if(e.is_multilingual){r||(console.warn("No language specified - defaulting to English (en)."),r="en");const o=`<|${(0,M.whisper_language_to_code)(r)}|>`;t.push(e.lang_to_id[o]),t.push(e.task_to_id[n??"transcribe"])}else if(r||n)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:r=null,stopping_criteria:n=null,...o}){t=this._prepare_generation_config(t,o);const s=o.decoder_input_ids??this._retrieve_init_tokens(t);if(t.return_timestamps&&(r??=new d.LogitsProcessorList,r.push(new d.WhisperTimeStampLogitsProcessor(t,s))),t.begin_suppress_tokens&&(r??=new d.LogitsProcessorList,r.push(new d.SuppressTokensAtBeginLogitsProcessor(t.begin_suppress_tokens,s.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 a=await super.generate({inputs:e,generation_config:t,logits_processor:r,decoder_input_ids:s,...o});return t.return_token_timestamps&&(a.token_timestamps=this._extract_token_timestamps(a,t.alignment_heads,t.num_frames)),a}_extract_token_timestamps(e,t,r=null,n=.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==r&&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 o=this.config.median_filter_width;void 0===o&&(console.warn("Model config has no `median_filter_width`, using default value of 7."),o=7);const s=e.cross_attentions,a=Array.from({length:this.config.decoder_layers},((e,t)=>(0,p.cat)(s.map((e=>e[t])),2))),l=(0,p.stack)(t.map((([e,t])=>{if(e>=a.length)throw new Error(`Layer index ${e} is out of bounds for cross attentions (length ${a.length}).`);return r?a[e].slice(null,t,null,[0,r]):a[e].slice(null,t)}))).transpose(1,0,2,3),[c,d]=(0,p.std_mean)(l,-2,0,!0),u=l.clone();for(let e=0;e<u.dims[0];++e){const t=u[e];for(let r=0;r<t.dims[0];++r){const n=t[r],s=c[e][r][0].data,a=d[e][r][0].data;for(let e=0;e<n.dims[0];++e){let t=n[e].data;for(let e=0;e<t.length;++e)t[e]=(t[e]-a[e])/s[e];t.set((0,h.medianFilter)(t,o))}}}const m=[(0,p.mean)(u,1)],_=e.sequences.dims,f=new p.Tensor("float32",new Float32Array(_[0]*_[1]),_);for(let e=0;e<_[0];++e){const t=m[e].neg().squeeze_(0),[r,o]=(0,h.dynamic_time_warping)(t.tolist()),s=Array.from({length:r.length-1},((e,t)=>r[t+1]-r[t])),a=(0,i.mergeArrays)([1],s).map((e=>!!e)),l=[];for(let e=0;e<a.length;++e)a[e]&&l.push(o[e]*n);f[e].data.set(l,1)}return f}}class lr extends q{main_input_name="pixel_values";forward_params=["pixel_values","decoder_input_ids","encoder_hidden_states","past_key_values"]}class cr extends q{forward_params=["input_ids","pixel_values","attention_mask","position_ids","past_key_values"]}class dr extends cr{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:r,attention_mask:n}){const o=this.config.image_token_index,s=r.tolist().map((e=>e.findIndex((e=>e==o)))),a=s.every((e=>-1===e)),i=s.every((e=>-1!==e));if(!a&&!i)throw new Error("Every input should contain either 0 or 1 image token.");if(a)return{inputs_embeds:e,attention_mask:n};const l=[],c=[];for(let r=0;r<s.length;++r){const o=s[r],a=e[r],i=t[r],d=n[r];l.push((0,p.cat)([a.slice([0,o]),i,a.slice([o+1,a.dims[0]])],0)),c.push((0,p.cat)([d.slice([0,o]),(0,p.ones)([i.dims[0]]),d.slice([o+1,d.dims[0]])],0))}return{inputs_embeds:(0,p.stack)(l,0),attention_mask:(0,p.stack)(c,0)}}}class ur extends dr{}class pr extends q{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 hr extends pr{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:r,attention_mask:n}){return{inputs_embeds:(0,p.cat)([t,e],1),attention_mask:(0,p.cat)([(0,p.ones)(t.dims.slice(0,2)),n],1)}}async _prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:r,attention_mask:n}){if(!e&&!t)throw new Error("Either `input_ids` or `pixel_values` should be provided.");let o,s;return e&&(o=await this.encode_text({input_ids:e})),t&&(s=await this.encode_image({pixel_values:t})),o&&s?({inputs_embeds:r,attention_mask:n}=this._merge_input_ids_with_image_features({inputs_embeds:o,image_features:s,input_ids:e,attention_mask:n})):r=o||s,{inputs_embeds:r,attention_mask:n}}async forward({input_ids:e,pixel_values:t,attention_mask:r,decoder_input_ids:n,decoder_attention_mask:o,encoder_outputs:s,past_key_values:a,inputs_embeds:i,decoder_inputs_embeds:l}){if(i||({inputs_embeds:i,attention_mask:r}=await this._prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:i,attention_mask:r})),!s){let{last_hidden_state:e}=await D(this,{inputs_embeds:i,attention_mask:r});s=e}if(!l){if(!n)throw new Error("Either `decoder_input_ids` or `decoder_inputs_embeds` should be provided.");l=await this.encode_text({input_ids:n})}const c={inputs_embeds:l,attention_mask:o,encoder_attention_mask:r,encoder_hidden_states:s,past_key_values:a};return await O(this,c,!0)}}class mr extends q{}class _r extends mr{}class fr extends mr{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class gr extends mr{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class Mr extends mr{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class wr extends mr{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class Tr extends q{}class br extends Tr{}class xr extends Tr{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class yr extends mr{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class kr extends q{}class Fr extends kr{}class Pr extends q{}class Cr extends Pr{}class vr extends Pr{}class Sr extends q{}class Ar extends Sr{}class Er extends Sr{}class Lr extends q{}class zr extends Lr{}class Ir extends Lr{}class Br extends q{}class Nr extends Br{}class Dr extends Br{}class Or extends q{}class Vr extends Or{}class jr extends Or{}class Rr extends q{}class Gr extends Rr{}class qr extends Rr{}class $r extends q{}class Wr extends $r{}class Ur extends $r{}class Xr extends q{}class Qr extends Xr{}class Hr extends Xr{}class Yr extends q{}class Jr extends Yr{}class Kr extends Yr{}class Zr extends q{}class en extends Zr{}class tn extends Zr{}class rn extends q{}class nn extends rn{}class on extends rn{}class sn extends q{}class an extends sn{}class ln extends sn{}class cn extends q{}class dn extends cn{}class un extends cn{}class pn extends q{}class hn extends pn{}class mn extends pn{}class _n extends q{}class fn extends _n{}class gn extends _n{}class Mn extends q{}class wn extends Mn{}class Tn extends Mn{}class bn extends q{}class xn extends bn{}class yn extends bn{}class kn extends q{}class Fn extends kn{}class Pn extends kn{}class Cn extends q{}class vn extends Cn{}class Sn extends Cn{}class An extends q{}class En extends An{}class Ln extends An{}class zn extends q{}class In extends zn{}class Bn extends zn{async _call(e){return new hl(await super._call(e))}}class Nn extends q{}class Dn extends Nn{}class On extends Nn{async _call(e){return new hl(await super._call(e))}}class Vn extends q{}class jn extends Vn{}class Rn extends q{}class Gn extends Rn{}class qn extends Rn{async _call(e){return new hl(await super._call(e))}}class $n extends q{}class Wn extends $n{}class Un extends q{}class Xn extends Un{}class Qn extends Un{async _call(e){return new hl(await super._call(e))}}class Hn extends q{}class Yn extends Hn{async _call(e){return new Tl(await super._call(e))}}class Jn extends q{}class Kn extends Jn{}class Zn extends Jn{async _call(e){return new hl(await super._call(e))}}class eo extends q{}class to extends eo{}class ro extends eo{async _call(e){return new hl(await super._call(e))}}class no extends q{}class oo extends no{}class so extends no{}class ao extends q{}class io extends ao{}class lo extends ao{}class co extends q{}class uo extends co{}class po extends co{async _call(e){return new hl(await super._call(e))}}class ho extends q{}class mo extends ho{}class _o extends ho{async _call(e){return new go(await super._call(e))}}class fo extends ho{async _call(e){return new Mo(await super._call(e))}}class go extends ${constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class Mo extends ${constructor({logits:e,pred_boxes:t,pred_masks:r}){super(),this.logits=e,this.pred_boxes=t,this.pred_masks=r}}class wo extends q{}class To extends wo{}class bo extends wo{async _call(e){return new xo(await super._call(e))}}class xo extends ${constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class yo extends q{}class ko extends yo{}class Fo extends yo{async _call(e){return new Po(await super._call(e))}}class Po extends go{}class Co extends q{}class vo extends Co{}class So extends Co{async _call(e){return new hl(await super._call(e))}}class Ao extends q{}class Eo extends Ao{}class Lo extends Ao{async _call(e){return new hl(await super._call(e))}}class zo extends q{}class Io extends zo{}class Bo extends zo{async _call(e){return new hl(await super._call(e))}}class No extends q{}class Do extends No{}class Oo extends No{async _call(e){return new hl(await super._call(e))}}class Vo extends q{}class jo extends Vo{}class Ro extends Vo{}class Go extends q{}class qo extends Go{}class $o extends Go{}class Wo extends q{}class Uo extends Wo{}class Xo extends q{}class Qo extends Xo{}class Ho extends Xo{}class Yo extends Xo{}class Jo extends q{}class Ko extends Jo{}class Zo extends q{}class es extends Zo{}class ts extends Zo{}class rs extends q{}class ns extends rs{}class os extends rs{}class ss extends q{}class as extends ss{}class is extends q{}class ls extends is{}class cs extends is{async _call(e){return new hl(await super._call(e))}}class ds extends q{}class us extends ds{}class ps extends ds{async _call(e){return new hl(await super._call(e))}}class hs extends q{}class ms extends hs{}class _s extends hs{async _call(e){return new hl(await super._call(e))}}class fs extends q{}class gs extends fs{}class Ms extends fs{async _call(e){return new ws(await super._call(e))}}class ws extends ${constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class Ts extends q{}class bs extends Ts{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),r=t.reduce(((e,t)=>e*t),1);e.input_labels=new p.Tensor("int64",new BigInt64Array(r).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 L(this.sessions.prompt_encoder_mask_decoder,t)}async _call(e){return new xs(await super._call(e))}}class xs extends ${constructor({iou_scores:e,pred_masks:t}){super(),this.iou_scores=e,this.pred_masks=t}}class ys extends q{}class ks extends ys{}class Fs extends ys{}class Ps extends q{}class Cs extends Ps{}class vs extends Ps{}class Ss extends q{}class As extends Ss{}class Es extends Ss{async _call(e){return new Ml(await super._call(e))}}class Ls extends Ss{async _call(e){return new hl(await super._call(e))}}class zs extends Ss{async _call(e){return new _l(await super._call(e))}}class Is extends q{}class Bs extends Is{}class Ns extends Is{async _call(e){return new _l(await super._call(e))}}class Ds extends q{}class Os extends Ds{}class Vs extends q{}class js extends Vs{}class Rs extends Vs{async _call(e){return new Ml(await super._call(e))}}class Gs extends Vs{async _call(e){return new hl(await super._call(e))}}class qs extends q{}class $s extends qs{}class Ws extends qs{async _call(e){return new Ml(await super._call(e))}}class Us extends qs{async _call(e){return new hl(await super._call(e))}}class Xs extends qs{async _call(e){return new _l(await super._call(e))}}class Qs extends q{}class Hs extends Qs{}class Ys extends Qs{async _call(e){return new Ml(await super._call(e))}}class Js extends Qs{async _call(e){return new hl(await super._call(e))}}class Ks extends q{}class Zs extends Ss{}class ea extends Ss{async _call(e){return new Ml(await super._call(e))}}class ta extends Ss{async _call(e){return new hl(await super._call(e))}}class ra extends q{}class na extends ra{}class oa extends ra{async _call(e){return new Ml(await super._call(e))}}class sa extends ra{async _call(e){return new hl(await super._call(e))}}class aa extends ra{async _call(e){return new ml(await super._call(e))}}class ia extends ra{async _call(e){return new _l(await super._call(e))}}class la extends q{}class ca extends la{}class da extends la{}class ua extends la{async generate_speech(e,t,{threshold:r=.5,minlenratio:n=0,maxlenratio:o=20,vocoder:s=null}={}){const a={input_ids:e},{encoder_outputs:i,encoder_attention_mask:l}=await D(this,a),c=i.dims[1]/this.config.reduction_factor,d=Math.floor(c*o),u=Math.floor(c*n),h=this.config.num_mel_bins;let m=[],_=null,f=null,g=0;for(;;){++g;const e=B(!!f);let n;n=f?f.output_sequence_out:new p.Tensor("float32",new Float32Array(h),[1,1,h]);let o={use_cache_branch:e,output_sequence:n,encoder_attention_mask:l,speaker_embeddings:t,encoder_hidden_states:i};this.addPastKeyValues(o,_),f=await L(this.sessions.decoder_model_merged,o),_=this.getPastKeyValues(f,_);const{prob:s,spectrum:a}=f;if(m.push(a),g>=u&&(Array.from(s.data).filter((e=>e>=r)).length>0||g>=d))break}const M=(0,p.cat)(m),{waveform:w}=await L(s.sessions.model,{spectrogram:M});return{spectrogram:M,waveform:w}}}class pa extends q{main_input_name="spectrogram"}class ha extends q{}class ma extends ha{}class _a extends q{}class fa extends _a{}class ga extends _a{}class Ma extends q{}class wa extends Ma{}class Ta extends Ma{}class ba extends q{}class xa extends ba{}class ya extends ba{}class ka extends q{}class Fa extends ka{}class Pa extends ka{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class Ca extends ka{static async from_pretrained(e,t={}){return t.model_file_name??="audio_model",super.from_pretrained(e,t)}}class va extends q{}class Sa extends va{async _call(e){return new bl(await super._call(e))}}class Aa extends q{}class Ea extends Aa{}class La extends Aa{}class za extends Aa{}class Ia extends q{}class Ba extends Ia{}class Na extends Ia{}class Da extends q{}class Oa extends Da{}class Va extends Da{async _call(e){return new hl(await super._call(e))}}class ja extends q{}class Ra extends ja{}class Ga extends ja{}class qa extends q{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,r]=e.dims,n=this.config.decoder.num_codebooks,o=r-n;let s=0;for(let t=0;t<e.size;++t){if(e.data[t]===this.config.decoder.pad_token_id)continue;const a=t%r-Math.floor(t/r)%n;a>0&&a<=o&&(e.data[s++]=e.data[t])}const a=Math.floor(t/n),i=s/(a*n);return new p.Tensor(e.type,e.data.slice(0,s),[a,n,i])}prepare_inputs_for_generation(e,t,r){let n=structuredClone(e);for(let e=0;e<n.length;++e)for(let t=0;t<n[e].length;++t)e%this.config.decoder.num_codebooks>=t&&(n[e][t]=BigInt(this.config.decoder.pad_token_id));null!==r.guidance_scale&&r.guidance_scale>1&&(n=n.concat(n));return super.prepare_inputs_for_generation(n,t,r)}async generate(e){const t=await super.generate(e),r=this._apply_and_filter_by_delay_pattern_mask(t).unsqueeze_(0),{audio_values:n}=await L(this.sessions.encodec_decode,{audio_codes:r});return n}}class $a extends q{}class Wa extends $a{}class Ua extends $a{async _call(e){return new hl(await super._call(e))}}class Xa extends q{}class Qa extends Xa{}class Ha extends Xa{async _call(e){return new hl(await super._call(e))}}class Ya extends q{}class Ja extends Ya{}class Ka extends Ya{async _call(e){return new hl(await super._call(e))}}class Za extends q{}class ei extends Za{}class ti extends Za{async _call(e){return new hl(await super._call(e))}}class ri extends q{}class ni extends ri{}class oi{static MODEL_CLASS_MAPPINGS=null;static BASE_IF_FAIL=!1;static async from_pretrained(e,{progress_callback:t=null,config:r=null,cache_dir:o=null,local_files_only:s=!1,revision:a="main",model_file_name:i=null,subfolder:l="onnx",device:c=null,dtype:d=null,use_external_data_format:u=null,session_options:p={}}={}){const h={progress_callback:t,config:r,cache_dir:o,local_files_only:s,revision:a,model_file_name:i,subfolder:l,device:c,dtype:d,use_external_data_format:u,session_options:p};if(h.config=await n.AutoConfig.from_pretrained(e,h),!this.MODEL_CLASS_MAPPINGS)throw new Error("`MODEL_CLASS_MAPPINGS` not implemented for this type of `AutoClass`: "+this.name);for(const t of this.MODEL_CLASS_MAPPINGS){const r=t.get(h.config.model_type);if(r)return await r[1].from_pretrained(e,h)}if(this.BASE_IF_FAIL)return console.warn(`Unknown model class "${h.config.model_type}", attempting to construct from base class.`),await q.from_pretrained(e,h);throw Error(`Unsupported model type: ${h.config.model_type}`)}}const si=new Map([["bert",["BertModel",X]],["nomic_bert",["NomicBertModel",Z]],["roformer",["RoFormerModel",te]],["electra",["ElectraModel",he]],["esm",["EsmModel",qe]],["convbert",["ConvBertModel",ie]],["camembert",["CamembertModel",we]],["deberta",["DebertaModel",Fe]],["deberta-v2",["DebertaV2Model",Ee]],["mpnet",["MPNetModel",Ze]],["albert",["AlbertModel",dt]],["distilbert",["DistilBertModel",De]],["roberta",["RobertaModel",Vt]],["xlm",["XLMModel",Wt]],["xlm-roberta",["XLMRobertaModel",Jt]],["clap",["ClapModel",Fa]],["clip",["CLIPModel",_r]],["clipseg",["CLIPSegModel",Cr]],["chinese_clip",["ChineseCLIPModel",Fr]],["siglip",["SiglipModel",br]],["mobilebert",["MobileBertModel",Qe]],["squeezebert",["SqueezeBertModel",st]],["wav2vec2",["Wav2Vec2Model",As]],["wav2vec2-bert",["Wav2Vec2BertModel",Hs]],["unispeech",["UniSpeechModel",js]],["unispeech-sat",["UniSpeechSatModel",$s]],["hubert",["HubertModel",Zs]],["wavlm",["WavLMModel",na]],["audio-spectrogram-transformer",["ASTModel",nr]],["vits",["VitsModel",Sa]],["pyannote",["PyAnnoteModel",Bs]],["wespeaker-resnet",["WeSpeakerResNetModel",Os]],["detr",["DetrModel",mo]],["rt_detr",["RTDetrModel",To]],["table-transformer",["TableTransformerModel",ko]],["vit",["ViTModel",In]],["pvt",["PvtModel",Dn]],["vit_msn",["ViTMSNModel",Gn]],["vit_mae",["ViTMAEModel",jn]],["groupvit",["GroupViTModel",Wn]],["fastvit",["FastViTModel",Xn]],["mobilevit",["MobileViTModel",Kn]],["mobilevitv2",["MobileViTV2Model",to]],["owlvit",["OwlViTModel",oo]],["owlv2",["Owlv2Model",io]],["beit",["BeitModel",uo]],["deit",["DeiTModel",vo]],["hiera",["HieraModel",Eo]],["convnext",["ConvNextModel",ls]],["convnextv2",["ConvNextV2Model",us]],["dinov2",["Dinov2Model",ms]],["resnet",["ResNetModel",Io]],["swin",["SwinModel",Do]],["swin2sr",["Swin2SRModel",jo]],["donut-swin",["DonutSwinModel",as]],["yolos",["YolosModel",gs]],["dpt",["DPTModel",qo]],["glpn",["GLPNModel",ns]],["hifigan",["SpeechT5HifiGan",pa]],["efficientnet",["EfficientNetModel",Oa]],["decision_transformer",["DecisionTransformerModel",ni]],["mobilenet_v1",["MobileNetV1Model",Wa]],["mobilenet_v2",["MobileNetV2Model",Qa]],["mobilenet_v3",["MobileNetV3Model",Ja]],["mobilenet_v4",["MobileNetV4Model",ei]],["maskformer",["MaskFormerModel",es]]]),ai=new Map([["t5",["T5Model",_t]],["longt5",["LongT5Model",Mt]],["mt5",["MT5Model",bt]],["bart",["BartModel",kt]],["mbart",["MBartModel",vt]],["marian",["MarianModel",ks]],["whisper",["WhisperModel",ar]],["m2m_100",["M2M100Model",Cs]],["blenderbot",["BlenderbotModel",zt]],["blenderbot-small",["BlenderbotSmallModel",Nt]]]),ii=new Map([["bloom",["BloomModel",Fn]],["jais",["JAISModel",zr]],["gpt2",["GPT2Model",Ar]],["gptj",["GPTJModel",Gr]],["gpt_bigcode",["GPTBigCodeModel",Wr]],["gpt_neo",["GPTNeoModel",Nr]],["gpt_neox",["GPTNeoXModel",Vr]],["codegen",["CodeGenModel",Qr]],["llama",["LlamaModel",Jr]],["granite",["GraniteModel",en]],["cohere",["CohereModel",nn]],["gemma",["GemmaModel",an]],["gemma2",["Gemma2Model",dn]],["openelm",["OpenELMModel",hn]],["qwen2",["Qwen2Model",fn]],["phi",["PhiModel",wn]],["phi3",["Phi3Model",xn]],["mpt",["MptModel",vn]],["opt",["OPTModel",En]],["mistral",["MistralModel",fa]],["starcoder2",["Starcoder2Model",wa]],["falcon",["FalconModel",xa]],["stablelm",["StableLmModel",Ba]]]),li=new Map([["speecht5",["SpeechT5ForSpeechToText",da]],["whisper",["WhisperForConditionalGeneration",ir]]]),ci=new Map([["speecht5",["SpeechT5ForTextToSpeech",ua]]]),di=new Map([["vits",["VitsModel",Sa]],["musicgen",["MusicgenForConditionalGeneration",qa]]]),ui=new Map([["bert",["BertForSequenceClassification",H]],["roformer",["RoFormerForSequenceClassification",ne]],["electra",["ElectraForSequenceClassification",_e]],["esm",["EsmForSequenceClassification",We]],["convbert",["ConvBertForSequenceClassification",ce]],["camembert",["CamembertForSequenceClassification",be]],["deberta",["DebertaForSequenceClassification",Ce]],["deberta-v2",["DebertaV2ForSequenceClassification",ze]],["mpnet",["MPNetForSequenceClassification",tt]],["albert",["AlbertForSequenceClassification",ut]],["distilbert",["DistilBertForSequenceClassification",Oe]],["roberta",["RobertaForSequenceClassification",Rt]],["xlm",["XLMForSequenceClassification",Xt]],["xlm-roberta",["XLMRobertaForSequenceClassification",Zt]],["bart",["BartForSequenceClassification",Pt]],["mbart",["MBartForSequenceClassification",At]],["mobilebert",["MobileBertForSequenceClassification",Ye]],["squeezebert",["SqueezeBertForSequenceClassification",it]]]),pi=new Map([["bert",["BertForTokenClassification",Y]],["roformer",["RoFormerForTokenClassification",oe]],["electra",["ElectraForTokenClassification",fe]],["esm",["EsmForTokenClassification",Ue]],["convbert",["ConvBertForTokenClassification",de]],["camembert",["CamembertForTokenClassification",xe]],["deberta",["DebertaForTokenClassification",ve]],["deberta-v2",["DebertaV2ForTokenClassification",Ie]],["mpnet",["MPNetForTokenClassification",rt]],["distilbert",["DistilBertForTokenClassification",Ve]],["roberta",["RobertaForTokenClassification",Gt]],["xlm",["XLMForTokenClassification",Qt]],["xlm-roberta",["XLMRobertaForTokenClassification",er]]]),hi=new Map([["t5",["T5ForConditionalGeneration",ft]],["longt5",["LongT5ForConditionalGeneration",wt]],["mt5",["MT5ForConditionalGeneration",xt]],["bart",["BartForConditionalGeneration",Ft]],["mbart",["MBartForConditionalGeneration",St]],["marian",["MarianMTModel",Fs]],["m2m_100",["M2M100ForConditionalGeneration",vs]],["blenderbot",["BlenderbotForConditionalGeneration",It]],["blenderbot-small",["BlenderbotSmallForConditionalGeneration",Dt]]]),mi=new Map([["bloom",["BloomForCausalLM",Pn]],["gpt2",["GPT2LMHeadModel",Er]],["jais",["JAISLMHeadModel",Ir]],["gptj",["GPTJForCausalLM",qr]],["gpt_bigcode",["GPTBigCodeForCausalLM",Ur]],["gpt_neo",["GPTNeoForCausalLM",Dr]],["gpt_neox",["GPTNeoXForCausalLM",jr]],["codegen",["CodeGenForCausalLM",Hr]],["llama",["LlamaForCausalLM",Kr]],["granite",["GraniteForCausalLM",tn]],["cohere",["CohereForCausalLM",on]],["gemma",["GemmaForCausalLM",ln]],["gemma2",["Gemma2ForCausalLM",un]],["openelm",["OpenELMForCausalLM",mn]],["qwen2",["Qwen2ForCausalLM",gn]],["phi",["PhiForCausalLM",Tn]],["phi3",["Phi3ForCausalLM",yn]],["mpt",["MptForCausalLM",Sn]],["opt",["OPTForCausalLM",Ln]],["mbart",["MBartForCausalLM",Et]],["mistral",["MistralForCausalLM",ga]],["starcoder2",["Starcoder2ForCausalLM",Ta]],["falcon",["FalconForCausalLM",ya]],["trocr",["TrOCRForCausalLM",ma]],["stablelm",["StableLmForCausalLM",Na]]]),_i=new Map([["bert",["BertForMaskedLM",Q]],["roformer",["RoFormerForMaskedLM",re]],["electra",["ElectraForMaskedLM",me]],["esm",["EsmForMaskedLM",$e]],["convbert",["ConvBertForMaskedLM",le]],["camembert",["CamembertForMaskedLM",Te]],["deberta",["DebertaForMaskedLM",Pe]],["deberta-v2",["DebertaV2ForMaskedLM",Le]],["mpnet",["MPNetForMaskedLM",et]],["albert",["AlbertForMaskedLM",ht]],["distilbert",["DistilBertForMaskedLM",Re]],["roberta",["RobertaForMaskedLM",jt]],["xlm",["XLMWithLMHeadModel",Ut]],["xlm-roberta",["XLMRobertaForMaskedLM",Kt]],["mobilebert",["MobileBertForMaskedLM",He]],["squeezebert",["SqueezeBertForMaskedLM",at]]]),fi=new Map([["bert",["BertForQuestionAnswering",J]],["roformer",["RoFormerForQuestionAnswering",se]],["electra",["ElectraForQuestionAnswering",ge]],["convbert",["ConvBertForQuestionAnswering",ue]],["camembert",["CamembertForQuestionAnswering",ye]],["deberta",["DebertaForQuestionAnswering",Se]],["deberta-v2",["DebertaV2ForQuestionAnswering",Be]],["mpnet",["MPNetForQuestionAnswering",nt]],["albert",["AlbertForQuestionAnswering",pt]],["distilbert",["DistilBertForQuestionAnswering",je]],["roberta",["RobertaForQuestionAnswering",qt]],["xlm",["XLMForQuestionAnswering",Ht]],["xlm-roberta",["XLMRobertaForQuestionAnswering",tr]],["mobilebert",["MobileBertForQuestionAnswering",Je]],["squeezebert",["SqueezeBertForQuestionAnswering",lt]]]),gi=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",lr]]]),Mi=new Map([["llava",["LlavaForConditionalGeneration",dr]],["moondream1",["Moondream1ForConditionalGeneration",ur]],["florence2",["Florence2ForConditionalGeneration",hr]]]),wi=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",lr]]]),Ti=new Map([["vit",["ViTForImageClassification",Bn]],["pvt",["PvtForImageClassification",On]],["vit_msn",["ViTMSNForImageClassification",qn]],["fastvit",["FastViTForImageClassification",Qn]],["mobilevit",["MobileViTForImageClassification",Zn]],["mobilevitv2",["MobileViTV2ForImageClassification",ro]],["beit",["BeitForImageClassification",po]],["deit",["DeiTForImageClassification",So]],["hiera",["HieraForImageClassification",Lo]],["convnext",["ConvNextForImageClassification",cs]],["convnextv2",["ConvNextV2ForImageClassification",ps]],["dinov2",["Dinov2ForImageClassification",_s]],["resnet",["ResNetForImageClassification",Bo]],["swin",["SwinForImageClassification",Oo]],["segformer",["SegformerForImageClassification",La]],["efficientnet",["EfficientNetForImageClassification",Va]],["mobilenet_v1",["MobileNetV1ForImageClassification",Ua]],["mobilenet_v2",["MobileNetV2ForImageClassification",Ha]],["mobilenet_v3",["MobileNetV3ForImageClassification",Ka]],["mobilenet_v4",["MobileNetV4ForImageClassification",ti]]]),bi=new Map([["detr",["DetrForObjectDetection",_o]],["rt_detr",["RTDetrForObjectDetection",bo]],["table-transformer",["TableTransformerForObjectDetection",Fo]],["yolos",["YolosForObjectDetection",Ms]]]),xi=new Map([["owlvit",["OwlViTForObjectDetection",so]],["owlv2",["Owlv2ForObjectDetection",lo]]]),yi=new Map([["detr",["DetrForSegmentation",fo]],["clipseg",["CLIPSegForImageSegmentation",vr]]]),ki=new Map([["segformer",["SegformerForSemanticSegmentation",za]],["sapiens",["SapiensForSemanticSegmentation",Qo]]]),Fi=new Map([["detr",["DetrForSegmentation",fo]],["maskformer",["MaskFormerForInstanceSegmentation",ts]]]),Pi=new Map([["sam",["SamModel",bs]]]),Ci=new Map([["wav2vec2",["Wav2Vec2ForCTC",Es]],["wav2vec2-bert",["Wav2Vec2BertForCTC",Ys]],["unispeech",["UniSpeechForCTC",Rs]],["unispeech-sat",["UniSpeechSatForCTC",Ws]],["wavlm",["WavLMForCTC",oa]],["hubert",["HubertForCTC",ea]]]),vi=new Map([["wav2vec2",["Wav2Vec2ForSequenceClassification",Ls]],["wav2vec2-bert",["Wav2Vec2BertForSequenceClassification",Js]],["unispeech",["UniSpeechForSequenceClassification",Gs]],["unispeech-sat",["UniSpeechSatForSequenceClassification",Us]],["wavlm",["WavLMForSequenceClassification",sa]],["hubert",["HubertForSequenceClassification",ta]],["audio-spectrogram-transformer",["ASTForAudioClassification",or]]]),Si=new Map([["wavlm",["WavLMForXVector",aa]]]),Ai=new Map([["unispeech-sat",["UniSpeechSatForAudioFrameClassification",Xs]],["wavlm",["WavLMForAudioFrameClassification",ia]],["wav2vec2",["Wav2Vec2ForAudioFrameClassification",zs]],["pyannote",["PyAnnoteForAudioFrameClassification",Ns]]]),Ei=new Map([["vitmatte",["VitMatteForImageMatting",Yn]]]),Li=new Map([["swin2sr",["Swin2SRForImageSuperResolution",Ro]]]),zi=new Map([["dpt",["DPTForDepthEstimation",$o]],["depth_anything",["DepthAnythingForDepthEstimation",Uo]],["glpn",["GLPNForDepthEstimation",os]],["sapiens",["SapiensForDepthEstimation",Ho]],["depth_pro",["DepthProForDepthEstimation",Ko]]]),Ii=new Map([["sapiens",["SapiensForNormalEstimation",Yo]]]),Bi=new Map([["clip",["CLIPVisionModelWithProjection",wr]],["siglip",["SiglipVisionModel",yr]]]),Ni=[[si,w],[ai,T],[ii,y],[ui,w],[pi,w],[hi,b],[li,b],[mi,y],[_i,w],[fi,w],[gi,x],[Mi,F],[Ti,w],[yi,w],[Fi,w],[ki,w],[Ei,w],[Li,w],[zi,w],[Ii,w],[bi,w],[xi,w],[Pi,k],[Ci,w],[vi,w],[ci,b],[di,w],[Si,w],[Ai,w],[Bi,w]];for(const[e,t]of Ni)for(const[r,n]of e.values())C.set(r,t),S.set(n,r),v.set(r,n);const Di=[["MusicgenForConditionalGeneration",qa,P],["CLIPTextModelWithProjection",gr,w],["SiglipTextModel",xr,w],["ClapTextModelWithProjection",Pa,w],["ClapAudioModelWithProjection",Ca,w]];for(const[e,t,r]of Di)C.set(e,r),S.set(t,e),v.set(e,t);class Oi extends oi{static MODEL_CLASS_MAPPINGS=Ni.map((e=>e[0]));static BASE_IF_FAIL=!0}class Vi extends oi{static MODEL_CLASS_MAPPINGS=[ui]}class ji extends oi{static MODEL_CLASS_MAPPINGS=[pi]}class Ri extends oi{static MODEL_CLASS_MAPPINGS=[hi]}class Gi extends oi{static MODEL_CLASS_MAPPINGS=[li]}class qi extends oi{static MODEL_CLASS_MAPPINGS=[ci]}class $i extends oi{static MODEL_CLASS_MAPPINGS=[di]}class Wi extends oi{static MODEL_CLASS_MAPPINGS=[mi]}class Ui extends oi{static MODEL_CLASS_MAPPINGS=[_i]}class Xi extends oi{static MODEL_CLASS_MAPPINGS=[fi]}class Qi extends oi{static MODEL_CLASS_MAPPINGS=[gi]}class Hi extends oi{static MODEL_CLASS_MAPPINGS=[Ti]}class Yi extends oi{static MODEL_CLASS_MAPPINGS=[yi]}class Ji extends oi{static MODEL_CLASS_MAPPINGS=[ki]}class Ki extends oi{static MODEL_CLASS_MAPPINGS=[Fi]}class Zi extends oi{static MODEL_CLASS_MAPPINGS=[bi]}class el extends oi{static MODEL_CLASS_MAPPINGS=[xi]}class tl extends oi{static MODEL_CLASS_MAPPINGS=[Pi]}class rl extends oi{static MODEL_CLASS_MAPPINGS=[Ci]}class nl extends oi{static MODEL_CLASS_MAPPINGS=[vi]}class ol extends oi{static MODEL_CLASS_MAPPINGS=[Si]}class sl extends oi{static MODEL_CLASS_MAPPINGS=[Ai]}class al extends oi{static MODEL_CLASS_MAPPINGS=[wi]}class il extends oi{static MODEL_CLASS_MAPPINGS=[Ei]}class ll extends oi{static MODEL_CLASS_MAPPINGS=[Li]}class cl extends oi{static MODEL_CLASS_MAPPINGS=[zi]}class dl extends oi{static MODEL_CLASS_MAPPINGS=[Ii]}class ul extends oi{static MODEL_CLASS_MAPPINGS=[Bi]}class pl extends ${constructor({logits:e,past_key_values:t,encoder_outputs:r,decoder_attentions:n=null,cross_attentions:o=null}){super(),this.logits=e,this.past_key_values=t,this.encoder_outputs=r,this.decoder_attentions=n,this.cross_attentions=o}}class hl extends ${constructor({logits:e}){super(),this.logits=e}}class ml extends ${constructor({logits:e,embeddings:t}){super(),this.logits=e,this.embeddings=t}}class _l extends ${constructor({logits:e}){super(),this.logits=e}}class fl extends ${constructor({logits:e}){super(),this.logits=e}}class gl extends ${constructor({start_logits:e,end_logits:t}){super(),this.start_logits=e,this.end_logits=t}}class Ml extends ${constructor({logits:e}){super(),this.logits=e}}class wl extends ${constructor({logits:e,past_key_values:t}){super(),this.logits=e,this.past_key_values=t}}class Tl extends ${constructor({alphas:e}){super(),this.alphas=e}}class bl extends ${constructor({waveform:e,spectrogram:t}){super(),this.waveform=e,this.spectrogram=t}}},"./src/models/whisper/common_whisper.js":
 /*!**********************************************!*\
   !*** ./src/models/whisper/common_whisper.js ***!
   \**********************************************/(e,t,r)=>{r.r(t),r.d(t,{WHISPER_LANGUAGE_MAPPING:()=>o,WHISPER_TO_LANGUAGE_CODE_MAPPING:()=>s,whisper_language_to_code:()=>a});const n=[["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"]],o=new Map(n),s=new Map([...n.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 a(e){e=e.toLowerCase();let t=s.get(e);if(void 0===t){if(!o.has(e)){const t=2===e.length?o.keys():o.values();throw new Error(`Language "${e}" is not supported. Must be one of: ${JSON.stringify(t)}`)}t=e}return t}},"./src/models/whisper/generation_whisper.js":
@@ -121,13 +121,13 @@ import*as e from"fs";import*as t from"onnxruntime-node";import*as r from"path";i
   \*****************************/(e,t,r)=>{r.r(t),r.d(t,{TensorOpRegistry:()=>a});var n=r(/*! ../backends/onnx.js */"./src/backends/onnx.js"),o=r(/*! ../utils/tensor.js */"./src/utils/tensor.js");const s=async(e,t,r)=>{const s=await(0,n.createInferenceSession)(new Uint8Array(e),t);return async e=>{const t=Object.fromEntries(Object.entries(e).map((([e,t])=>[e,t.ort_tensor]))),n=await s.run(t);return Array.isArray(r)?r.map((e=>new o.Tensor(n[e]))):new o.Tensor(n[r])}};class a{static session_options={};static get bilinear_interpolate_4d(){return this._bilinear_interpolate_4d||(this._bilinear_interpolate_4d=s([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=s([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=s([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=s([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=s([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=s([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}}},"./src/pipelines.js":
 /*!**************************!*\
   !*** ./src/pipelines.js ***!
-  \**************************/(e,t,r)=>{r.r(t),r.d(t,{AudioClassificationPipeline:()=>v,AutomaticSpeechRecognitionPipeline:()=>A,DepthEstimationPipeline:()=>j,DocumentQuestionAnsweringPipeline:()=>D,FeatureExtractionPipeline:()=>P,FillMaskPipeline:()=>w,ImageClassificationPipeline:()=>L,ImageFeatureExtractionPipeline:()=>C,ImageSegmentationPipeline:()=>z,ImageToImagePipeline:()=>V,ImageToTextPipeline:()=>E,ObjectDetectionPipeline:()=>B,Pipeline:()=>_,QuestionAnsweringPipeline:()=>M,SummarizationPipeline:()=>b,Text2TextGenerationPipeline:()=>T,TextClassificationPipeline:()=>f,TextGenerationPipeline:()=>k,TextToAudioPipeline:()=>O,TokenClassificationPipeline:()=>g,TranslationPipeline:()=>x,ZeroShotAudioClassificationPipeline:()=>S,ZeroShotClassificationPipeline:()=>F,ZeroShotImageClassificationPipeline:()=>I,ZeroShotObjectDetectionPipeline:()=>N,pipeline:()=>q});var n=r(/*! ./tokenizers.js */"./src/tokenizers.js"),o=r(/*! ./models.js */"./src/models.js"),s=r(/*! ./processors.js */"./src/processors.js"),a=r(/*! ./utils/generic.js */"./src/utils/generic.js"),i=r(/*! ./utils/core.js */"./src/utils/core.js"),l=r(/*! ./utils/maths.js */"./src/utils/maths.js"),c=r(/*! ./utils/audio.js */"./src/utils/audio.js"),d=r(/*! ./utils/tensor.js */"./src/utils/tensor.js"),u=r(/*! ./utils/image.js */"./src/utils/image.js");async function p(e){return Array.isArray(e)||(e=[e]),await Promise.all(e.map((e=>u.RawImage.read(e))))}async function h(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[r,n,o,s]=e;return{xmin:r,ymin:n,xmax:o,ymax:s}}class _ extends a.Callable{constructor({task:e,model:t,tokenizer:r=null,processor:n=null}){super(),this.task=e,this.model=t,this.tokenizer=r,this.processor=n}async dispose(){await this.model.dispose()}}class f extends _{constructor(e){super(e)}async _call(e,{top_k:t=1}={}){const r=this.tokenizer(e,{padding:!0,truncation:!0}),n=await this.model(r),o="multi_label_classification"===this.model.config.problem_type?e=>e.sigmoid():e=>new d.Tensor("float32",(0,l.softmax)(e.data),e.dims),s=this.model.config.id2label,a=[];for(const e of n.logits){const r=o(e),n=await(0,d.topk)(r,t),i=n[0].tolist(),l=n[1].tolist().map(((e,t)=>({label:s?s[e]:`LABEL_${e}`,score:i[t]})));1===t?a.push(...l):a.push(l)}return Array.isArray(e)||1===t?a:a[0]}}class g extends _{constructor(e){super(e)}async _call(e,{ignore_labels:t=["O"]}={}){const r=Array.isArray(e),n=this.tokenizer(r?e:[e],{padding:!0,truncation:!0}),o=(await this.model(n)).logits,s=this.model.config.id2label,a=[];for(let e=0;e<o.dims[0];++e){const r=n.input_ids[e],i=o[e],c=[];for(let e=0;e<i.dims[0];++e){const n=i[e],o=(0,l.max)(n.data)[1],a=s?s[o]:`LABEL_${o}`;if(t.includes(a))continue;const d=this.tokenizer.decode([r[e].item()],{skip_special_tokens:!0});if(""===d)continue;const u=(0,l.softmax)(n.data);c.push({entity:a,score:u[o],index:e,word:d})}a.push(c)}return r?a:a[0]}}class M extends _{constructor(e){super(e)}async _call(e,t,{top_k:r=1}={}){const n=this.tokenizer(e,{text_pair:t,padding:!0,truncation:!0}),{start_logits:o,end_logits:s}=await this.model(n),a=n.input_ids.tolist(),c=n.attention_mask.tolist(),d=this.tokenizer.all_special_ids,u=[];for(let e=0;e<o.dims[0];++e){const t=a[e],n=t.findIndex((e=>e==this.tokenizer.sep_token_id)),p=(c[e].map(((e,r)=>1==e&&(0===r||r>n&&-1===d.findIndex((e=>e==t[r]))))),o[e].tolist()),h=s[e].tolist();for(let r=1;r<p.length;++r)(0==c[e]||r<=n||-1!==d.findIndex((e=>e==t[r])))&&(p[r]=-1/0,h[r]=-1/0);const m=(0,l.softmax)(p).map(((e,t)=>[e,t])),_=(0,l.softmax)(h).map(((e,t)=>[e,t]));m[0][0]=0,_[0][0]=0;const f=(0,i.product)(m,_).filter((e=>e[0][1]<=e[1][1])).map((e=>[e[0][1],e[1][1],e[0][0]*e[1][0]])).sort(((e,t)=>t[2]-e[2]));for(let e=0;e<Math.min(f.length,r);++e){const[r,n,o]=f[e],s=t.slice(r,n+1),a=this.tokenizer.decode(s,{skip_special_tokens:!0});u.push({answer:a,score:o})}}return 1===r?u[0]:u}}class w extends _{constructor(e){super(e)}async _call(e,{top_k:t=5}={}){const r=this.tokenizer(e,{padding:!0,truncation:!0}),{logits:n}=await this.model(r),o=[],s=r.input_ids.tolist();for(let e=0;e<s.length;++e){const r=s[e],a=r.findIndex((e=>e==this.tokenizer.mask_token_id));if(-1===a)throw Error(`Mask token (${this.tokenizer.mask_token}) not found in text.`);const i=n[e][a],c=await(0,d.topk)(new d.Tensor("float32",(0,l.softmax)(i.data),i.dims),t),u=c[0].tolist(),p=c[1].tolist();o.push(p.map(((e,t)=>{const n=r.slice();return n[a]=e,{score:u[t],token:Number(e),token_str:this.tokenizer.model.vocab[e],sequence:this.tokenizer.decode(n,{skip_special_tokens:!0})}})))}return Array.isArray(e)?o:o[0]}}class T extends _{_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 r=this.model.config.task_specific_params;r&&r[this.task]&&r[this.task].prefix&&(e=e.map((e=>r[this.task].prefix+e)));const n=this.tokenizer,o={padding:!0,truncation:!0};let s;s=this instanceof x&&"_build_translation_inputs"in n?n._build_translation_inputs(e,o,t):n(e,o);const a=await this.model.generate({...s,...t});return n.batch_decode(a,{skip_special_tokens:!0}).map((e=>({[this._key]:e})))}}class b extends T{_key="summary_text";constructor(e){super(e)}}class x extends T{_key="translation_text";constructor(e){super(e)}}function y(e){return Array.isArray(e)&&e.every((e=>"role"in e&&"content"in e))}class k extends _{constructor(e){super(e)}async _call(e,t={}){let r,n=!1,o=!1;if("string"==typeof e)r=e=[e];else if(Array.isArray(e)&&e.every((e=>"string"==typeof e)))n=!0,r=e;else{if(y(e))e=[e];else{if(!Array.isArray(e)||!e.every(y))throw new Error("Input must be a string, an array of strings, a Chat, or an array of Chats");n=!0}o=!0,r=e.map((e=>this.tokenizer.apply_chat_template(e,{tokenize:!1,add_generation_prompt:!0})))}const s=t.add_special_tokens??!1,a=!o&&(t.return_full_text??!0);this.tokenizer.padding_side="left";const i=this.tokenizer(r,{add_special_tokens:s,padding:!0,truncation:!0}),l=await this.model.generate({...i,...t}),c=this.tokenizer.batch_decode(l,{skip_special_tokens:!0});let d;!a&&i.input_ids.dims.at(-1)>0&&(d=this.tokenizer.batch_decode(i.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 r=Math.floor(t/l.dims[0]*e.length);d&&(c[t]=c[t].slice(d[r])),u[r].push({generated_text:o?[...e[r],{role:"assistant",content:c[t]}]:c[t]})}return n||1!==u.length?u:u[0]}}class F extends _{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:r="This example is {}.",multi_label:n=!1}={}){const o=Array.isArray(e);o||(e=[e]),Array.isArray(t)||(t=[t]);const s=t.map((e=>r.replace("{}",e))),a=n||1===t.length,i=[];for(const r of e){const e=[];for(const t of s){const n=this.tokenizer(r,{text_pair:t,padding:!0,truncation:!0}),o=await this.model(n);a?e.push([o.logits.data[this.contradiction_id],o.logits.data[this.entailment_id]]):e.push(o.logits.data[this.entailment_id])}const n=(a?e.map((e=>(0,l.softmax)(e)[1])):(0,l.softmax)(e)).map(((e,t)=>[e,t])).sort(((e,t)=>t[0]-e[0]));i.push({sequence:r,labels:n.map((e=>t[e[1]])),scores:n.map((e=>e[0]))})}return o?i:i[0]}}class P extends _{constructor(e){super(e)}async _call(e,{pooling:t="none",normalize:r=!1,quantize:n=!1,precision:o="binary"}={}){const s=this.tokenizer(e,{padding:!0,truncation:!0}),a=await this.model(s);let i=a.last_hidden_state??a.logits??a.token_embeddings;if("none"===t);else if("mean"===t)i=(0,d.mean_pooling)(i,s.attention_mask);else{if("cls"!==t)throw Error(`Pooling method '${t}' not supported.`);i=i.slice(null,0)}return r&&(i=i.normalize(2,-1)),n&&(i=(0,d.quantize_embeddings)(i,o)),i}}class C extends _{constructor(e){super(e)}async _call(e,{pool:t=null}={}){const r=await p(e),{pixel_values:n}=await this.processor(r),o=await this.model({pixel_values:n});let s;if(t){if(!("pooler_output"in o))throw Error("No pooled output was returned. Make sure the model has a 'pooler' layer when using the 'pool' option.");s=o.pooler_output}else s=o.last_hidden_state??o.logits??o.image_embeds;return s}}class v extends _{constructor(e){super(e)}async _call(e,{top_k:t=5}={}){const r=this.processor.feature_extractor.config.sampling_rate,n=await h(e,r),o=this.model.config.id2label,s=[];for(const e of n){const r=await this.processor(e),n=(await this.model(r)).logits[0],a=await(0,d.topk)(new d.Tensor("float32",(0,l.softmax)(n.data),n.dims),t),i=a[0].tolist(),c=a[1].tolist().map(((e,t)=>({label:o?o[e]:`LABEL_${e}`,score:i[t]})));s.push(c)}return Array.isArray(e)?s:s[0]}}class S extends _{constructor(e){super(e)}async _call(e,t,{hypothesis_template:r="This is a sound of {}."}={}){const n=!Array.isArray(e);n&&(e=[e]);const o=t.map((e=>r.replace("{}",e))),s=this.tokenizer(o,{padding:!0,truncation:!0}),a=this.processor.feature_extractor.config.sampling_rate,i=await h(e,a),c=[];for(const e of i){const r=await this.processor(e),n=await this.model({...s,...r}),o=(0,l.softmax)(n.logits_per_audio.data);c.push([...o].map(((e,r)=>({score:e,label:t[r]}))))}return n?c[0]:c}}class A extends _{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);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 r=!Array.isArray(e);r&&(e=[e]);const n=this.processor.feature_extractor.config.sampling_rate,o=await h(e,n),s=[];for(const e of o){const t=await this.processor(e),r=(await this.model(t)).logits[0],n=[];for(const e of r)n.push((0,l.max)(e.data)[1]);const o=this.tokenizer.decode(n);s.push({text:o})}return r?s[0]:s}async _call_whisper(e,t){const r=t.return_timestamps??!1,n=t.chunk_length_s??0,o=t.force_full_sequences??!1;let s=t.stride_length_s??null;const a={...t};"word"===r&&(a.return_token_timestamps=!0,a.return_timestamps=!1);const i=!Array.isArray(e);i&&(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,p=await h(e,u),m=[];for(const e of p){let t=[];if(n>0){if(null===s)s=n/6;else if(n<=s)throw Error("`chunk_length_s` must be larger than `stride_length_s`.");const r=u*n,o=u*s,a=r-2*o;let i=0;for(;;){const n=i+r,s=e.subarray(i,n),l=await this.processor(s),c=0===i,d=n>=e.length;if(t.push({stride:[s.length,c?0:o,d?0:o],input_features:l.input_features,is_last:d}),d)break;i+=a}}else t=[{stride:[e.length,0,0],input_features:(await this.processor(e)).input_features,is_last:!0}];for(const e of t){a.num_frames=Math.floor(e.stride[0]/d);const t=await this.model.generate({inputs:e.input_features,...a});"word"===r?(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[i,p]=this.tokenizer._decode_asr(t,{time_precision:c,return_timestamps:r,force_full_sequences:o});m.push({text:i,...p})}return i?m[0]:m}}class E extends _{constructor(e){super(e)}async _call(e,t={}){const r=Array.isArray(e),n=await p(e),{pixel_values:o}=await this.processor(n),s=[];for(const e of o){e.dims=[1,...e.dims];const r=await this.model.generate({inputs:e,...t}),n=this.tokenizer.batch_decode(r,{skip_special_tokens:!0}).map((e=>({generated_text:e.trim()})));s.push(n)}return r?s:s[0]}}class L extends _{constructor(e){super(e)}async _call(e,{top_k:t=5}={}){const r=await p(e),{pixel_values:n}=await this.processor(r),o=await this.model({pixel_values:n}),s=this.model.config.id2label,a=[];for(const e of o.logits){const r=await(0,d.topk)(new d.Tensor("float32",(0,l.softmax)(e.data),e.dims),t),n=r[0].tolist(),o=r[1].tolist().map(((e,t)=>({label:s?s[e]:`LABEL_${e}`,score:n[t]})));a.push(o)}return Array.isArray(e)?a:a[0]}}class z extends _{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:r=.5,overlap_mask_area_threshold:n=.8,label_ids_to_fuse:o=null,target_sizes:s=null,subtask:a=null}={}){if(Array.isArray(e)&&1!==e.length)throw Error("Image segmentation pipeline currently only supports a batch size of 1.");const i=await p(e),l=i.map((e=>[e.height,e.width])),{pixel_values:c,pixel_mask:d}=await this.processor(i),h=await this.model({pixel_values:c,pixel_mask:d});let m=null;if(null!==a)m=this.subtasks_mapping[a];else for(let[e,t]of Object.entries(this.subtasks_mapping))if(t in this.processor.feature_extractor){m=this.processor.feature_extractor[t].bind(this.processor.feature_extractor),a=e;break}const _=this.model.config.id2label,f=[];if("panoptic"===a||"instance"===a){const e=m(h,t,r,n,o,s??l)[0],a=e.segmentation;for(const t of e.segments_info){const e=new Uint8ClampedArray(a.data.length);for(let r=0;r<a.data.length;++r)a.data[r]===t.id&&(e[r]=255);const r=new u.RawImage(e,a.dims[1],a.dims[0],1);f.push({score:t.score,label:_[t.label_id],mask:r})}}else{if("semantic"!==a)throw Error(`Subtask ${a} not supported.`);{const{segmentation:e,labels:t}=m(h,s??l)[0];for(const r of t){const t=new Uint8ClampedArray(e.data.length);for(let n=0;n<e.data.length;++n)e.data[n]===r&&(t[n]=255);const n=new u.RawImage(t,e.dims[1],e.dims[0],1);f.push({score:null,label:_[r],mask:n})}}}return f}}class I extends _{constructor(e){super(e)}async _call(e,t,{hypothesis_template:r="This is a photo of {}"}={}){const n=Array.isArray(e),o=await p(e),s=t.map((e=>r.replace("{}",e))),a=this.tokenizer(s,{padding:"siglip"!==this.model.config.model_type||"max_length",truncation:!0}),{pixel_values:i}=await this.processor(o),c=await this.model({...a,pixel_values:i}),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 r=[...d(e)].map(((e,r)=>({score:e,label:t[r]})));r.sort(((e,t)=>t.score-e.score)),u.push(r)}return n?u:u[0]}}class B extends _{constructor(e){super(e)}async _call(e,{threshold:t=.9,percentage:r=!1}={}){const n=Array.isArray(e);if(n&&1!==e.length)throw Error("Object detection pipeline currently only supports a batch size of 1.");const o=await p(e),s=r?null:o.map((e=>[e.height,e.width])),{pixel_values:a,pixel_mask:i}=await this.processor(o),l=await this.model({pixel_values:a,pixel_mask:i}),c=this.processor.feature_extractor.post_process_object_detection(l,t,s),d=this.model.config.id2label,u=c.map((e=>e.boxes.map(((t,n)=>({score:e.scores[n],label:d[e.classes[n]],box:m(t,!r)})))));return n?u:u[0]}}class N extends _{constructor(e){super(e)}async _call(e,t,{threshold:r=.1,top_k:n=null,percentage:o=!1}={}){const s=Array.isArray(e),a=await p(e),i=this.tokenizer(t,{padding:!0,truncation:!0}),l=await this.processor(a),c=[];for(let e=0;e<a.length;++e){const s=a[e],d=o?null:[[s.height,s.width]],u=l.pixel_values[e].unsqueeze_(0),p=await this.model({...i,pixel_values:u}),h=this.processor.feature_extractor.post_process_object_detection(p,r,d,!0)[0];let _=h.boxes.map(((e,r)=>({score:h.scores[r],label:t[h.classes[r]],box:m(e,!o)}))).sort(((e,t)=>t.score-e.score));null!==n&&(_=_.slice(0,n)),c.push(_)}return s?c:c[0]}}class D extends _{constructor(e){super(e)}async _call(e,t,r={}){throw new Error("This pipeline is not yet supported in Transformers.js v3.")}}class O extends _{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:r}=await this.model(t),n=this.model.config.sampling_rate;return{audio:r.data,sampling_rate:n}}async _call_text_to_spectrogram(e,{speaker_embeddings:t}){if(this.vocoder||(console.log("No vocoder specified, using default HifiGan vocoder."),this.vocoder=await o.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:r}=this.tokenizer(e,{padding:!0,truncation:!0}),{waveform:n}=await this.model.generate_speech(r,t,{vocoder:this.vocoder}),s=this.processor.feature_extractor.config.sampling_rate;return{audio:n.data,sampling_rate:s}}}class V extends _{constructor(e){super(e)}async _call(e){const t=await p(e),r=await this.processor(t),n=await this.model(r),o=[];for(const e of n.reconstruction){const t=e.squeeze().clamp_(0,1).mul_(255).round_().to("uint8");o.push(u.RawImage.fromTensor(t))}return o.length>1?o:o[0]}}class j extends _{constructor(e){super(e)}async _call(e){const t=await p(e),r=await this.processor(t),{predicted_depth:n}=await this.model(r),o=[];for(let e=0;e<t.length;++e){const r=(0,d.interpolate)(n[e],t[e].size.reverse(),"bilinear",!1),s=r.mul_(255/(0,l.max)(r.data)[0]).to("uint8");o.push({predicted_depth:n[e],depth:u.RawImage.fromTensor(s)})}return o.length>1?o:o[0]}}const R=Object.freeze({"text-classification":{tokenizer:n.AutoTokenizer,pipeline:f,model:o.AutoModelForSequenceClassification,default:{model:"Xenova/distilbert-base-uncased-finetuned-sst-2-english"},type:"text"},"token-classification":{tokenizer:n.AutoTokenizer,pipeline:g,model:o.AutoModelForTokenClassification,default:{model:"Xenova/bert-base-multilingual-cased-ner-hrl"},type:"text"},"question-answering":{tokenizer:n.AutoTokenizer,pipeline:M,model:o.AutoModelForQuestionAnswering,default:{model:"Xenova/distilbert-base-cased-distilled-squad"},type:"text"},"fill-mask":{tokenizer:n.AutoTokenizer,pipeline:w,model:o.AutoModelForMaskedLM,default:{model:"Xenova/bert-base-uncased"},type:"text"},summarization:{tokenizer:n.AutoTokenizer,pipeline:b,model:o.AutoModelForSeq2SeqLM,default:{model:"Xenova/distilbart-cnn-6-6"},type:"text"},translation:{tokenizer:n.AutoTokenizer,pipeline:x,model:o.AutoModelForSeq2SeqLM,default:{model:"Xenova/t5-small"},type:"text"},"text2text-generation":{tokenizer:n.AutoTokenizer,pipeline:T,model:o.AutoModelForSeq2SeqLM,default:{model:"Xenova/flan-t5-small"},type:"text"},"text-generation":{tokenizer:n.AutoTokenizer,pipeline:k,model:o.AutoModelForCausalLM,default:{model:"Xenova/gpt2"},type:"text"},"zero-shot-classification":{tokenizer:n.AutoTokenizer,pipeline:F,model:o.AutoModelForSequenceClassification,default:{model:"Xenova/distilbert-base-uncased-mnli"},type:"text"},"audio-classification":{pipeline:v,model:o.AutoModelForAudioClassification,processor:s.AutoProcessor,default:{model:"Xenova/wav2vec2-base-superb-ks"},type:"audio"},"zero-shot-audio-classification":{tokenizer:n.AutoTokenizer,pipeline:S,model:o.AutoModel,processor:s.AutoProcessor,default:{model:"Xenova/clap-htsat-unfused"},type:"multimodal"},"automatic-speech-recognition":{tokenizer:n.AutoTokenizer,pipeline:A,model:[o.AutoModelForSpeechSeq2Seq,o.AutoModelForCTC],processor:s.AutoProcessor,default:{model:"Xenova/whisper-tiny.en"},type:"multimodal"},"text-to-audio":{tokenizer:n.AutoTokenizer,pipeline:O,model:[o.AutoModelForTextToWaveform,o.AutoModelForTextToSpectrogram],processor:[s.AutoProcessor,null],default:{model:"Xenova/speecht5_tts"},type:"text"},"image-to-text":{tokenizer:n.AutoTokenizer,pipeline:E,model:o.AutoModelForVision2Seq,processor:s.AutoProcessor,default:{model:"Xenova/vit-gpt2-image-captioning"},type:"multimodal"},"image-classification":{pipeline:L,model:o.AutoModelForImageClassification,processor:s.AutoProcessor,default:{model:"Xenova/vit-base-patch16-224"},type:"multimodal"},"image-segmentation":{pipeline:z,model:[o.AutoModelForImageSegmentation,o.AutoModelForSemanticSegmentation,o.AutoModelForUniversalSegmentation],processor:s.AutoProcessor,default:{model:"Xenova/detr-resnet-50-panoptic"},type:"multimodal"},"zero-shot-image-classification":{tokenizer:n.AutoTokenizer,pipeline:I,model:o.AutoModel,processor:s.AutoProcessor,default:{model:"Xenova/clip-vit-base-patch32"},type:"multimodal"},"object-detection":{pipeline:B,model:o.AutoModelForObjectDetection,processor:s.AutoProcessor,default:{model:"Xenova/detr-resnet-50"},type:"multimodal"},"zero-shot-object-detection":{tokenizer:n.AutoTokenizer,pipeline:N,model:o.AutoModelForZeroShotObjectDetection,processor:s.AutoProcessor,default:{model:"Xenova/owlvit-base-patch32"},type:"multimodal"},"document-question-answering":{tokenizer:n.AutoTokenizer,pipeline:D,model:o.AutoModelForDocumentQuestionAnswering,processor:s.AutoProcessor,default:{model:"Xenova/donut-base-finetuned-docvqa"},type:"multimodal"},"image-to-image":{pipeline:V,model:o.AutoModelForImageToImage,processor:s.AutoProcessor,default:{model:"Xenova/swin2SR-classical-sr-x2-64"},type:"image"},"depth-estimation":{pipeline:j,model:o.AutoModelForDepthEstimation,processor:s.AutoProcessor,default:{model:"Xenova/dpt-large"},type:"image"},"feature-extraction":{tokenizer:n.AutoTokenizer,pipeline:P,model:o.AutoModel,default:{model:"Xenova/all-MiniLM-L6-v2"},type:"text"},"image-feature-extraction":{processor:s.AutoProcessor,pipeline:C,model:[o.AutoModelForImageFeatureExtraction,o.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 q(e,t=null,{progress_callback:r=null,config:n=null,cache_dir:o=null,local_files_only:s=!1,revision:a="main",device:l=null,dtype:c=null,model_file_name:d=null,session_options:u={}}={}){e=G[e]??e;const p=R[e.split("_",1)[0]];if(!p)throw Error(`Unsupported pipeline: ${e}. Must be one of [${Object.keys(R)}]`);t||(t=p.default.model,console.log(`No model specified. Using default model: "${t}".`));const h={progress_callback:r,config:n,cache_dir:o,local_files_only:s,revision:a,device:l,dtype:c,model_file_name:d,session_options:u},m=new Map([["tokenizer",p.tokenizer],["model",p.model],["processor",p.processor]]),_=await async function(e,t,r){const n=Object.create(null),o=[];for(const[s,a]of e.entries()){if(!a)continue;let e;e=Array.isArray(a)?new Promise((async(e,n)=>{let o;for(const s of a){if(null===s)return void e(null);try{return void e(await s.from_pretrained(t,r))}catch(e){if(e.message?.includes("Unsupported model type"))o=e;else{if(!e.message?.includes("Could not locate file"))return void n(e);o=e}}}n(o)})):a.from_pretrained(t,r),n[s]=e,o.push(e)}await Promise.all(o);for(const[e,t]of Object.entries(n))n[e]=await t;return n}(m,t,h);_.task=e,(0,i.dispatchCallback)(r,{status:"ready",task:e,model:t});return new(0,p.pipeline)(_)}},"./src/processors.js":
+  \**************************/(e,t,r)=>{r.r(t),r.d(t,{AudioClassificationPipeline:()=>v,AutomaticSpeechRecognitionPipeline:()=>A,DepthEstimationPipeline:()=>j,DocumentQuestionAnsweringPipeline:()=>D,FeatureExtractionPipeline:()=>P,FillMaskPipeline:()=>w,ImageClassificationPipeline:()=>L,ImageFeatureExtractionPipeline:()=>C,ImageSegmentationPipeline:()=>z,ImageToImagePipeline:()=>V,ImageToTextPipeline:()=>E,ObjectDetectionPipeline:()=>B,Pipeline:()=>_,QuestionAnsweringPipeline:()=>M,SummarizationPipeline:()=>b,Text2TextGenerationPipeline:()=>T,TextClassificationPipeline:()=>f,TextGenerationPipeline:()=>k,TextToAudioPipeline:()=>O,TokenClassificationPipeline:()=>g,TranslationPipeline:()=>x,ZeroShotAudioClassificationPipeline:()=>S,ZeroShotClassificationPipeline:()=>F,ZeroShotImageClassificationPipeline:()=>I,ZeroShotObjectDetectionPipeline:()=>N,pipeline:()=>q});var n=r(/*! ./tokenizers.js */"./src/tokenizers.js"),o=r(/*! ./models.js */"./src/models.js"),s=r(/*! ./processors.js */"./src/processors.js"),a=r(/*! ./utils/generic.js */"./src/utils/generic.js"),i=r(/*! ./utils/core.js */"./src/utils/core.js"),l=r(/*! ./utils/maths.js */"./src/utils/maths.js"),c=r(/*! ./utils/audio.js */"./src/utils/audio.js"),d=r(/*! ./utils/tensor.js */"./src/utils/tensor.js"),u=r(/*! ./utils/image.js */"./src/utils/image.js");async function p(e){return Array.isArray(e)||(e=[e]),await Promise.all(e.map((e=>u.RawImage.read(e))))}async function h(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[r,n,o,s]=e;return{xmin:r,ymin:n,xmax:o,ymax:s}}class _ extends a.Callable{constructor({task:e,model:t,tokenizer:r=null,processor:n=null}){super(),this.task=e,this.model=t,this.tokenizer=r,this.processor=n}async dispose(){await this.model.dispose()}}class f extends _{constructor(e){super(e)}async _call(e,{top_k:t=1}={}){const r=this.tokenizer(e,{padding:!0,truncation:!0}),n=await this.model(r),o="multi_label_classification"===this.model.config.problem_type?e=>e.sigmoid():e=>new d.Tensor("float32",(0,l.softmax)(e.data),e.dims),s=this.model.config.id2label,a=[];for(const e of n.logits){const r=o(e),n=await(0,d.topk)(r,t),i=n[0].tolist(),l=n[1].tolist().map(((e,t)=>({label:s?s[e]:`LABEL_${e}`,score:i[t]})));1===t?a.push(...l):a.push(l)}return Array.isArray(e)||1===t?a:a[0]}}class g extends _{constructor(e){super(e)}async _call(e,{ignore_labels:t=["O"]}={}){const r=Array.isArray(e),n=this.tokenizer(r?e:[e],{padding:!0,truncation:!0}),o=(await this.model(n)).logits,s=this.model.config.id2label,a=[];for(let e=0;e<o.dims[0];++e){const r=n.input_ids[e],i=o[e],c=[];for(let e=0;e<i.dims[0];++e){const n=i[e],o=(0,l.max)(n.data)[1],a=s?s[o]:`LABEL_${o}`;if(t.includes(a))continue;const d=this.tokenizer.decode([r[e].item()],{skip_special_tokens:!0});if(""===d)continue;const u=(0,l.softmax)(n.data);c.push({entity:a,score:u[o],index:e,word:d})}a.push(c)}return r?a:a[0]}}class M extends _{constructor(e){super(e)}async _call(e,t,{top_k:r=1}={}){const n=this.tokenizer(e,{text_pair:t,padding:!0,truncation:!0}),{start_logits:o,end_logits:s}=await this.model(n),a=n.input_ids.tolist(),c=n.attention_mask.tolist(),d=this.tokenizer.all_special_ids,u=[];for(let e=0;e<o.dims[0];++e){const t=a[e],n=t.findIndex((e=>e==this.tokenizer.sep_token_id)),p=(c[e].map(((e,r)=>1==e&&(0===r||r>n&&-1===d.findIndex((e=>e==t[r]))))),o[e].tolist()),h=s[e].tolist();for(let r=1;r<p.length;++r)(0==c[e]||r<=n||-1!==d.findIndex((e=>e==t[r])))&&(p[r]=-1/0,h[r]=-1/0);const m=(0,l.softmax)(p).map(((e,t)=>[e,t])),_=(0,l.softmax)(h).map(((e,t)=>[e,t]));m[0][0]=0,_[0][0]=0;const f=(0,i.product)(m,_).filter((e=>e[0][1]<=e[1][1])).map((e=>[e[0][1],e[1][1],e[0][0]*e[1][0]])).sort(((e,t)=>t[2]-e[2]));for(let e=0;e<Math.min(f.length,r);++e){const[r,n,o]=f[e],s=t.slice(r,n+1),a=this.tokenizer.decode(s,{skip_special_tokens:!0});u.push({answer:a,score:o})}}return 1===r?u[0]:u}}class w extends _{constructor(e){super(e)}async _call(e,{top_k:t=5}={}){const r=this.tokenizer(e,{padding:!0,truncation:!0}),{logits:n}=await this.model(r),o=[],s=r.input_ids.tolist();for(let e=0;e<s.length;++e){const r=s[e],a=r.findIndex((e=>e==this.tokenizer.mask_token_id));if(-1===a)throw Error(`Mask token (${this.tokenizer.mask_token}) not found in text.`);const i=n[e][a],c=await(0,d.topk)(new d.Tensor("float32",(0,l.softmax)(i.data),i.dims),t),u=c[0].tolist(),p=c[1].tolist();o.push(p.map(((e,t)=>{const n=r.slice();return n[a]=e,{score:u[t],token:Number(e),token_str:this.tokenizer.model.vocab[e],sequence:this.tokenizer.decode(n,{skip_special_tokens:!0})}})))}return Array.isArray(e)?o:o[0]}}class T extends _{_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 r=this.model.config.task_specific_params;r&&r[this.task]&&r[this.task].prefix&&(e=e.map((e=>r[this.task].prefix+e)));const n=this.tokenizer,o={padding:!0,truncation:!0};let s;s=this instanceof x&&"_build_translation_inputs"in n?n._build_translation_inputs(e,o,t):n(e,o);const a=await this.model.generate({...s,...t});return n.batch_decode(a,{skip_special_tokens:!0}).map((e=>({[this._key]:e})))}}class b extends T{_key="summary_text";constructor(e){super(e)}}class x extends T{_key="translation_text";constructor(e){super(e)}}function y(e){return Array.isArray(e)&&e.every((e=>"role"in e&&"content"in e))}class k extends _{constructor(e){super(e)}async _call(e,t={}){let r,n=!1,o=!1;if("string"==typeof e)r=e=[e];else if(Array.isArray(e)&&e.every((e=>"string"==typeof e)))n=!0,r=e;else{if(y(e))e=[e];else{if(!Array.isArray(e)||!e.every(y))throw new Error("Input must be a string, an array of strings, a Chat, or an array of Chats");n=!0}o=!0,r=e.map((e=>this.tokenizer.apply_chat_template(e,{tokenize:!1,add_generation_prompt:!0})))}const s=t.add_special_tokens??!1,a=!o&&(t.return_full_text??!0);this.tokenizer.padding_side="left";const i=this.tokenizer(r,{add_special_tokens:s,padding:!0,truncation:!0}),l=await this.model.generate({...i,...t}),c=this.tokenizer.batch_decode(l,{skip_special_tokens:!0});let d;!a&&i.input_ids.dims.at(-1)>0&&(d=this.tokenizer.batch_decode(i.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 r=Math.floor(t/l.dims[0]*e.length);d&&(c[t]=c[t].slice(d[r])),u[r].push({generated_text:o?[...e[r],{role:"assistant",content:c[t]}]:c[t]})}return n||1!==u.length?u:u[0]}}class F extends _{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:r="This example is {}.",multi_label:n=!1}={}){const o=Array.isArray(e);o||(e=[e]),Array.isArray(t)||(t=[t]);const s=t.map((e=>r.replace("{}",e))),a=n||1===t.length,i=[];for(const r of e){const e=[];for(const t of s){const n=this.tokenizer(r,{text_pair:t,padding:!0,truncation:!0}),o=await this.model(n);a?e.push([o.logits.data[this.contradiction_id],o.logits.data[this.entailment_id]]):e.push(o.logits.data[this.entailment_id])}const n=(a?e.map((e=>(0,l.softmax)(e)[1])):(0,l.softmax)(e)).map(((e,t)=>[e,t])).sort(((e,t)=>t[0]-e[0]));i.push({sequence:r,labels:n.map((e=>t[e[1]])),scores:n.map((e=>e[0]))})}return o?i:i[0]}}class P extends _{constructor(e){super(e)}async _call(e,{pooling:t="none",normalize:r=!1,quantize:n=!1,precision:o="binary"}={}){const s=this.tokenizer(e,{padding:!0,truncation:!0}),a=await this.model(s);let i=a.last_hidden_state??a.logits??a.token_embeddings;if("none"===t);else if("mean"===t)i=(0,d.mean_pooling)(i,s.attention_mask);else{if("cls"!==t)throw Error(`Pooling method '${t}' not supported.`);i=i.slice(null,0)}return r&&(i=i.normalize(2,-1)),n&&(i=(0,d.quantize_embeddings)(i,o)),i}}class C extends _{constructor(e){super(e)}async _call(e,{pool:t=null}={}){const r=await p(e),{pixel_values:n}=await this.processor(r),o=await this.model({pixel_values:n});let s;if(t){if(!("pooler_output"in o))throw Error("No pooled output was returned. Make sure the model has a 'pooler' layer when using the 'pool' option.");s=o.pooler_output}else s=o.last_hidden_state??o.logits??o.image_embeds;return s}}class v extends _{constructor(e){super(e)}async _call(e,{top_k:t=5}={}){const r=this.processor.feature_extractor.config.sampling_rate,n=await h(e,r),o=this.model.config.id2label,s=[];for(const e of n){const r=await this.processor(e),n=(await this.model(r)).logits[0],a=await(0,d.topk)(new d.Tensor("float32",(0,l.softmax)(n.data),n.dims),t),i=a[0].tolist(),c=a[1].tolist().map(((e,t)=>({label:o?o[e]:`LABEL_${e}`,score:i[t]})));s.push(c)}return Array.isArray(e)?s:s[0]}}class S extends _{constructor(e){super(e)}async _call(e,t,{hypothesis_template:r="This is a sound of {}."}={}){const n=!Array.isArray(e);n&&(e=[e]);const o=t.map((e=>r.replace("{}",e))),s=this.tokenizer(o,{padding:!0,truncation:!0}),a=this.processor.feature_extractor.config.sampling_rate,i=await h(e,a),c=[];for(const e of i){const r=await this.processor(e),n=await this.model({...s,...r}),o=(0,l.softmax)(n.logits_per_audio.data);c.push([...o].map(((e,r)=>({score:e,label:t[r]}))))}return n?c[0]:c}}class A extends _{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);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 r=!Array.isArray(e);r&&(e=[e]);const n=this.processor.feature_extractor.config.sampling_rate,o=await h(e,n),s=[];for(const e of o){const t=await this.processor(e),r=(await this.model(t)).logits[0],n=[];for(const e of r)n.push((0,l.max)(e.data)[1]);const o=this.tokenizer.decode(n);s.push({text:o})}return r?s[0]:s}async _call_whisper(e,t){const r=t.return_timestamps??!1,n=t.chunk_length_s??0,o=t.force_full_sequences??!1;let s=t.stride_length_s??null;const a={...t};"word"===r&&(a.return_token_timestamps=!0,a.return_timestamps=!1);const i=!Array.isArray(e);i&&(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,p=await h(e,u),m=[];for(const e of p){let t=[];if(n>0){if(null===s)s=n/6;else if(n<=s)throw Error("`chunk_length_s` must be larger than `stride_length_s`.");const r=u*n,o=u*s,a=r-2*o;let i=0;for(;;){const n=i+r,s=e.subarray(i,n),l=await this.processor(s),c=0===i,d=n>=e.length;if(t.push({stride:[s.length,c?0:o,d?0:o],input_features:l.input_features,is_last:d}),d)break;i+=a}}else t=[{stride:[e.length,0,0],input_features:(await this.processor(e)).input_features,is_last:!0}];for(const e of t){a.num_frames=Math.floor(e.stride[0]/d);const t=await this.model.generate({inputs:e.input_features,...a});"word"===r?(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[i,p]=this.tokenizer._decode_asr(t,{time_precision:c,return_timestamps:r,force_full_sequences:o});m.push({text:i,...p})}return i?m[0]:m}}class E extends _{constructor(e){super(e)}async _call(e,t={}){const r=Array.isArray(e),n=await p(e),{pixel_values:o}=await this.processor(n),s=[];for(const e of o){e.dims=[1,...e.dims];const r=await this.model.generate({inputs:e,...t}),n=this.tokenizer.batch_decode(r,{skip_special_tokens:!0}).map((e=>({generated_text:e.trim()})));s.push(n)}return r?s:s[0]}}class L extends _{constructor(e){super(e)}async _call(e,{top_k:t=5}={}){const r=await p(e),{pixel_values:n}=await this.processor(r),o=await this.model({pixel_values:n}),s=this.model.config.id2label,a=[];for(const e of o.logits){const r=await(0,d.topk)(new d.Tensor("float32",(0,l.softmax)(e.data),e.dims),t),n=r[0].tolist(),o=r[1].tolist().map(((e,t)=>({label:s?s[e]:`LABEL_${e}`,score:n[t]})));a.push(o)}return Array.isArray(e)?a:a[0]}}class z extends _{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:r=.5,overlap_mask_area_threshold:n=.8,label_ids_to_fuse:o=null,target_sizes:s=null,subtask:a=null}={}){if(Array.isArray(e)&&1!==e.length)throw Error("Image segmentation pipeline currently only supports a batch size of 1.");const i=await p(e),l=i.map((e=>[e.height,e.width])),{pixel_values:c,pixel_mask:d}=await this.processor(i),h=await this.model({pixel_values:c,pixel_mask:d});let m=null;if(null!==a)m=this.subtasks_mapping[a];else for(let[e,t]of Object.entries(this.subtasks_mapping))if(t in this.processor.feature_extractor){m=this.processor.feature_extractor[t].bind(this.processor.feature_extractor),a=e;break}const _=this.model.config.id2label,f=[];if("panoptic"===a||"instance"===a){const e=m(h,t,r,n,o,s??l)[0],a=e.segmentation;for(const t of e.segments_info){const e=new Uint8ClampedArray(a.data.length);for(let r=0;r<a.data.length;++r)a.data[r]===t.id&&(e[r]=255);const r=new u.RawImage(e,a.dims[1],a.dims[0],1);f.push({score:t.score,label:_[t.label_id],mask:r})}}else{if("semantic"!==a)throw Error(`Subtask ${a} not supported.`);{const{segmentation:e,labels:t}=m(h,s??l)[0];for(const r of t){const t=new Uint8ClampedArray(e.data.length);for(let n=0;n<e.data.length;++n)e.data[n]===r&&(t[n]=255);const n=new u.RawImage(t,e.dims[1],e.dims[0],1);f.push({score:null,label:_[r],mask:n})}}}return f}}class I extends _{constructor(e){super(e)}async _call(e,t,{hypothesis_template:r="This is a photo of {}"}={}){const n=Array.isArray(e),o=await p(e),s=t.map((e=>r.replace("{}",e))),a=this.tokenizer(s,{padding:"siglip"!==this.model.config.model_type||"max_length",truncation:!0}),{pixel_values:i}=await this.processor(o),c=await this.model({...a,pixel_values:i}),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 r=[...d(e)].map(((e,r)=>({score:e,label:t[r]})));r.sort(((e,t)=>t.score-e.score)),u.push(r)}return n?u:u[0]}}class B extends _{constructor(e){super(e)}async _call(e,{threshold:t=.9,percentage:r=!1}={}){const n=Array.isArray(e);if(n&&1!==e.length)throw Error("Object detection pipeline currently only supports a batch size of 1.");const o=await p(e),s=r?null:o.map((e=>[e.height,e.width])),{pixel_values:a,pixel_mask:i}=await this.processor(o),l=await this.model({pixel_values:a,pixel_mask:i}),c=this.processor.feature_extractor.post_process_object_detection(l,t,s),d=this.model.config.id2label,u=c.map((e=>e.boxes.map(((t,n)=>({score:e.scores[n],label:d[e.classes[n]],box:m(t,!r)})))));return n?u:u[0]}}class N extends _{constructor(e){super(e)}async _call(e,t,{threshold:r=.1,top_k:n=null,percentage:o=!1}={}){const s=Array.isArray(e),a=await p(e),i=this.tokenizer(t,{padding:!0,truncation:!0}),l=await this.processor(a),c=[];for(let e=0;e<a.length;++e){const s=a[e],d=o?null:[[s.height,s.width]],u=l.pixel_values[e].unsqueeze_(0),p=await this.model({...i,pixel_values:u}),h=this.processor.feature_extractor.post_process_object_detection(p,r,d,!0)[0];let _=h.boxes.map(((e,r)=>({score:h.scores[r],label:t[h.classes[r]],box:m(e,!o)}))).sort(((e,t)=>t.score-e.score));null!==n&&(_=_.slice(0,n)),c.push(_)}return s?c:c[0]}}class D extends _{constructor(e){super(e)}async _call(e,t,r={}){const n=(await p(e))[0],{pixel_values:o}=await this.processor(n),s=`<s_docvqa><s_question>${t}</s_question><s_answer>`,a=this.tokenizer(s,{add_special_tokens:!1,padding:!0,truncation:!0}).input_ids,i=await this.model.generate({inputs:o,max_length:this.model.config.decoder.max_position_embeddings,decoder_input_ids:a,...r}),l=this.tokenizer.batch_decode(i)[0].match(/<s_answer>(.*?)<\/s_answer>/);let c=null;return l&&l.length>=2&&(c=l[1].trim()),[{answer:c}]}}class O extends _{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:r}=await this.model(t),n=this.model.config.sampling_rate;return{audio:r.data,sampling_rate:n}}async _call_text_to_spectrogram(e,{speaker_embeddings:t}){if(this.vocoder||(console.log("No vocoder specified, using default HifiGan vocoder."),this.vocoder=await o.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:r}=this.tokenizer(e,{padding:!0,truncation:!0}),{waveform:n}=await this.model.generate_speech(r,t,{vocoder:this.vocoder}),s=this.processor.feature_extractor.config.sampling_rate;return{audio:n.data,sampling_rate:s}}}class V extends _{constructor(e){super(e)}async _call(e){const t=await p(e),r=await this.processor(t),n=await this.model(r),o=[];for(const e of n.reconstruction){const t=e.squeeze().clamp_(0,1).mul_(255).round_().to("uint8");o.push(u.RawImage.fromTensor(t))}return o.length>1?o:o[0]}}class j extends _{constructor(e){super(e)}async _call(e){const t=await p(e),r=await this.processor(t),{predicted_depth:n}=await this.model(r),o=[];for(let e=0;e<t.length;++e){const r=(0,d.interpolate)(n[e],t[e].size.reverse(),"bilinear",!1),s=r.mul_(255/(0,l.max)(r.data)[0]).to("uint8");o.push({predicted_depth:n[e],depth:u.RawImage.fromTensor(s)})}return o.length>1?o:o[0]}}const R=Object.freeze({"text-classification":{tokenizer:n.AutoTokenizer,pipeline:f,model:o.AutoModelForSequenceClassification,default:{model:"Xenova/distilbert-base-uncased-finetuned-sst-2-english"},type:"text"},"token-classification":{tokenizer:n.AutoTokenizer,pipeline:g,model:o.AutoModelForTokenClassification,default:{model:"Xenova/bert-base-multilingual-cased-ner-hrl"},type:"text"},"question-answering":{tokenizer:n.AutoTokenizer,pipeline:M,model:o.AutoModelForQuestionAnswering,default:{model:"Xenova/distilbert-base-cased-distilled-squad"},type:"text"},"fill-mask":{tokenizer:n.AutoTokenizer,pipeline:w,model:o.AutoModelForMaskedLM,default:{model:"Xenova/bert-base-uncased"},type:"text"},summarization:{tokenizer:n.AutoTokenizer,pipeline:b,model:o.AutoModelForSeq2SeqLM,default:{model:"Xenova/distilbart-cnn-6-6"},type:"text"},translation:{tokenizer:n.AutoTokenizer,pipeline:x,model:o.AutoModelForSeq2SeqLM,default:{model:"Xenova/t5-small"},type:"text"},"text2text-generation":{tokenizer:n.AutoTokenizer,pipeline:T,model:o.AutoModelForSeq2SeqLM,default:{model:"Xenova/flan-t5-small"},type:"text"},"text-generation":{tokenizer:n.AutoTokenizer,pipeline:k,model:o.AutoModelForCausalLM,default:{model:"Xenova/gpt2"},type:"text"},"zero-shot-classification":{tokenizer:n.AutoTokenizer,pipeline:F,model:o.AutoModelForSequenceClassification,default:{model:"Xenova/distilbert-base-uncased-mnli"},type:"text"},"audio-classification":{pipeline:v,model:o.AutoModelForAudioClassification,processor:s.AutoProcessor,default:{model:"Xenova/wav2vec2-base-superb-ks"},type:"audio"},"zero-shot-audio-classification":{tokenizer:n.AutoTokenizer,pipeline:S,model:o.AutoModel,processor:s.AutoProcessor,default:{model:"Xenova/clap-htsat-unfused"},type:"multimodal"},"automatic-speech-recognition":{tokenizer:n.AutoTokenizer,pipeline:A,model:[o.AutoModelForSpeechSeq2Seq,o.AutoModelForCTC],processor:s.AutoProcessor,default:{model:"Xenova/whisper-tiny.en"},type:"multimodal"},"text-to-audio":{tokenizer:n.AutoTokenizer,pipeline:O,model:[o.AutoModelForTextToWaveform,o.AutoModelForTextToSpectrogram],processor:[s.AutoProcessor,null],default:{model:"Xenova/speecht5_tts"},type:"text"},"image-to-text":{tokenizer:n.AutoTokenizer,pipeline:E,model:o.AutoModelForVision2Seq,processor:s.AutoProcessor,default:{model:"Xenova/vit-gpt2-image-captioning"},type:"multimodal"},"image-classification":{pipeline:L,model:o.AutoModelForImageClassification,processor:s.AutoProcessor,default:{model:"Xenova/vit-base-patch16-224"},type:"multimodal"},"image-segmentation":{pipeline:z,model:[o.AutoModelForImageSegmentation,o.AutoModelForSemanticSegmentation,o.AutoModelForUniversalSegmentation],processor:s.AutoProcessor,default:{model:"Xenova/detr-resnet-50-panoptic"},type:"multimodal"},"zero-shot-image-classification":{tokenizer:n.AutoTokenizer,pipeline:I,model:o.AutoModel,processor:s.AutoProcessor,default:{model:"Xenova/clip-vit-base-patch32"},type:"multimodal"},"object-detection":{pipeline:B,model:o.AutoModelForObjectDetection,processor:s.AutoProcessor,default:{model:"Xenova/detr-resnet-50"},type:"multimodal"},"zero-shot-object-detection":{tokenizer:n.AutoTokenizer,pipeline:N,model:o.AutoModelForZeroShotObjectDetection,processor:s.AutoProcessor,default:{model:"Xenova/owlvit-base-patch32"},type:"multimodal"},"document-question-answering":{tokenizer:n.AutoTokenizer,pipeline:D,model:o.AutoModelForDocumentQuestionAnswering,processor:s.AutoProcessor,default:{model:"Xenova/donut-base-finetuned-docvqa"},type:"multimodal"},"image-to-image":{pipeline:V,model:o.AutoModelForImageToImage,processor:s.AutoProcessor,default:{model:"Xenova/swin2SR-classical-sr-x2-64"},type:"image"},"depth-estimation":{pipeline:j,model:o.AutoModelForDepthEstimation,processor:s.AutoProcessor,default:{model:"Xenova/dpt-large"},type:"image"},"feature-extraction":{tokenizer:n.AutoTokenizer,pipeline:P,model:o.AutoModel,default:{model:"Xenova/all-MiniLM-L6-v2"},type:"text"},"image-feature-extraction":{processor:s.AutoProcessor,pipeline:C,model:[o.AutoModelForImageFeatureExtraction,o.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 q(e,t=null,{progress_callback:r=null,config:n=null,cache_dir:o=null,local_files_only:s=!1,revision:a="main",device:l=null,dtype:c=null,model_file_name:d=null,session_options:u={}}={}){e=G[e]??e;const p=R[e.split("_",1)[0]];if(!p)throw Error(`Unsupported pipeline: ${e}. Must be one of [${Object.keys(R)}]`);t||(t=p.default.model,console.log(`No model specified. Using default model: "${t}".`));const h={progress_callback:r,config:n,cache_dir:o,local_files_only:s,revision:a,device:l,dtype:c,model_file_name:d,session_options:u},m=new Map([["tokenizer",p.tokenizer],["model",p.model],["processor",p.processor]]),_=await async function(e,t,r){const n=Object.create(null),o=[];for(const[s,a]of e.entries()){if(!a)continue;let e;e=Array.isArray(a)?new Promise((async(e,n)=>{let o;for(const s of a){if(null===s)return void e(null);try{return void e(await s.from_pretrained(t,r))}catch(e){if(e.message?.includes("Unsupported model type"))o=e;else{if(!e.message?.includes("Could not locate file"))return void n(e);o=e}}}n(o)})):a.from_pretrained(t,r),n[s]=e,o.push(e)}await Promise.all(o);for(const[e,t]of Object.entries(n))n[e]=await t;return n}(m,t,h);_.task=e,(0,i.dispatchCallback)(r,{status:"ready",task:e,model:t});return new(0,p.pipeline)(_)}},"./src/processors.js":
 /*!***************************!*\
   !*** ./src/processors.js ***!
-  \***************************/(e,t,r)=>{r.r(t),r.d(t,{ASTFeatureExtractor:()=>se,AutoProcessor:()=>Me,BeitFeatureExtractor:()=>X,BitImageProcessor:()=>P,CLIPFeatureExtractor:()=>v,CLIPImageProcessor:()=>S,ChineseCLIPFeatureExtractor:()=>A,ClapFeatureExtractor:()=>ae,ConvNextFeatureExtractor:()=>L,ConvNextImageProcessor:()=>z,DPTFeatureExtractor:()=>k,DPTImageProcessor:()=>F,DeiTFeatureExtractor:()=>U,DetrFeatureExtractor:()=>Y,DonutFeatureExtractor:()=>Q,EfficientNetImageProcessor:()=>N,FeatureExtractor:()=>w,Florence2Processor:()=>ge,GLPNFeatureExtractor:()=>C,ImageFeatureExtractor:()=>T,MaskFormerFeatureExtractor:()=>J,MobileNetV1FeatureExtractor:()=>D,MobileNetV2FeatureExtractor:()=>O,MobileNetV3FeatureExtractor:()=>V,MobileNetV4FeatureExtractor:()=>j,MobileViTFeatureExtractor:()=>R,MobileViTImageProcessor:()=>G,NougatImageProcessor:()=>H,OwlViTFeatureExtractor:()=>q,OwlViTProcessor:()=>fe,Owlv2ImageProcessor:()=>W,Processor:()=>de,PvtImageProcessor:()=>y,PyAnnoteFeatureExtractor:()=>ie,PyAnnoteProcessor:()=>me,RTDetrImageProcessor:()=>$,SamImageProcessor:()=>Z,SamProcessor:()=>ue,SapiensFeatureExtractor:()=>b,SeamlessM4TFeatureExtractor:()=>oe,SegformerFeatureExtractor:()=>x,SiglipImageProcessor:()=>E,SpeechT5FeatureExtractor:()=>ce,SpeechT5Processor:()=>_e,Swin2SRImageProcessor:()=>ee,ViTFeatureExtractor:()=>I,ViTImageProcessor:()=>B,VitMatteImageProcessor:()=>te,Wav2Vec2FeatureExtractor:()=>ne,Wav2Vec2ProcessorWithLM:()=>he,WeSpeakerFeatureExtractor:()=>le,WhisperFeatureExtractor:()=>re,WhisperProcessor:()=>pe,YolosFeatureExtractor:()=>K});var n=r(/*! ./utils/generic.js */"./src/utils/generic.js"),o=r(/*! ./utils/core.js */"./src/utils/core.js"),s=r(/*! ./utils/hub.js */"./src/utils/hub.js"),a=r(/*! ./utils/maths.js */"./src/utils/maths.js"),i=r(/*! ./utils/tensor.js */"./src/utils/tensor.js"),l=(r(/*! ./utils/image.js */"./src/utils/image.js"),r(/*! ./utils/audio.js */"./src/utils/audio.js"));function c([e,t,r,n]){return[e-r/2,t-n/2,e+r/2,t+n/2]}function d(e,t=.5,r=null,n=!1){const o=e.logits,s=e.pred_boxes,[i,l,d]=o.dims;if(null!==r&&r.length!==i)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");let u=[];for(let e=0;e<i;++e){let i=null!==r?r[e]:null,p={boxes:[],classes:[],scores:[]},h=o[e],m=s[e];for(let e=0;e<l;++e){let r,o=h[e],s=[];if(n){r=o.sigmoid().data;for(let e=0;e<r.length;++e)r[e]>t&&s.push(e)}else{let e=(0,a.max)(o.data)[1];if(e===d-1)continue;if(r=(0,a.softmax)(o.data),r[e]<t)continue;s.push(e)}for(const t of s){let n=m[e].data;n=c(n),null!==i&&(n=n.map(((e,t)=>e*i[(t+1)%2]))),p.boxes.push(n),p.classes.push(t),p.scores.push(r[t])}}u.push(p)}return u}function u(e,t=null){const r=e.logits,n=r.dims[0];if(null!==t&&t.length!==n)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<n;++e){const n=null!==t?t[e]:null;let s=r[e];null!==n&&(s=(0,i.interpolate)(s,n,"bilinear",!1));const[a,l]=n??s.dims.slice(-2),c=new i.Tensor("int32",new Int32Array(a*l),[a,l]),d=s[0].data,u=c.data;for(let e=1;e<s.dims[0];++e){const t=s[e].data;for(let r=0;r<t.length;++r)t[r]>d[r]&&(d[r]=t[r],u[r]=e)}const p=new Array(s.dims[0]);for(let e=0;e<u.length;++e){const t=u[e];p[t]=t}const h=p.filter((e=>void 0!==e));o.push({segmentation:c,labels:h})}return o}function p(e,t,r,n){const o=[],s=[],i=[];for(let l=0;l<e.dims[0];++l){const c=e[l],d=t[l],u=(0,a.max)(c.data)[1];if(u===n)continue;const p=(0,a.softmax)(c.data)[u];p>r&&(o.push(d),s.push(p),i.push(u))}return[o,s,i]}function h(e,t,r,n=.5,o=.8){const s=[];let a=0,i=0;const l=t[r].data;for(let t=0;t<e.length;++t)e[t]===r&&(s.push(t),++a),l[t]>=n&&++i;let c=a>0&&i>0;if(c){c=a/i>o}return[c,s]}function m(e,t,r,n,o,s=null,a=null){const[l,c]=a??e[0].dims,d=new i.Tensor("int32",new Int32Array(l*c),[l,c]),u=[];if(null!==a)for(let t=0;t<e.length;++t)e[t]=(0,i.interpolate)(e[t],a,"bilinear",!1);const p=new Int32Array(e[0].data.length),m=new Float32Array(e[0].data.length);for(let r=0;r<e.length;++r){let n=t[r];const o=e[r].data;for(let e=0;e<o.length;++e)o[e]*=n,o[e]>m[e]&&(p[e]=r,m[e]=o[e])}let _=0;const f=d.data;for(let s=0;s<r.length;++s){const a=r[s],[i,l]=h(p,e,s,n,o);if(i){++_;for(const e of l)f[e]=_;u.push({id:_,label_id:a,score:t[s]})}}return[d,u]}function _(e,t=.5,r=.5,n=.8,o=null,s=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!==s&&s.length!==c)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");let h=[];for(let e=0;e<c;++e){let c=null!==s?s[e]:null,d=a[e],_=l[e],[f,g,M]=p(d,_,t,u);if(0===M.length){let[e,t]=c??_.dims.slice(-2),r=new i.Tensor("int32",new Int32Array(e*t).fill(-1),[e,t]);h.push({segmentation:r,segments_info:[]});continue}let[w,T]=m(f,g,M,r,n,o,c);h.push({segmentation:w,segments_info:T})}return h}function f(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.`)}function g(e,t,r=0,n=null){const o=e/t;let s=(0,a.bankers_round)(o)*t;return null!==n&&s>n&&(s=Math.floor(o)*t),s<r&&(s=Math.ceil(o)*t),s}function M([e,t],r){return[Math.max(Math.floor(e/r),1)*r,Math.max(Math.floor(t/r),1)*r]}class w extends n.Callable{constructor(e){super(),this.config=e}}class T extends w{constructor(e){super(e),this.image_mean=this.config.image_mean??this.config.mean,this.image_std=this.config.image_std??this.config.std,this.resample=this.config.resample??2,this.do_rescale=this.config.do_rescale??!0,this.rescale_factor=this.config.rescale_factor??1/255,this.do_normalize=this.config.do_normalize,this.do_resize=this.config.do_resize,this.do_thumbnail=this.config.do_thumbnail,this.size=this.config.size,this.size_divisibility=this.config.size_divisibility??this.config.size_divisor,this.do_center_crop=this.config.do_center_crop,this.crop_size=this.config.crop_size,this.do_convert_rgb=this.config.do_convert_rgb??!0,this.do_crop_margin=this.config.do_crop_margin,this.pad_size=this.config.pad_size,this.do_pad=this.config.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=this.config.do_flip_channel_order??!1}async thumbnail(e,t,r=2){const n=e.height,o=e.width,s=t.height,a=t.width;let i=Math.min(n,s),l=Math.min(o,a);return i===n&&l===o?e:(n>o?l=Math.floor(o*i/n):o>n&&(i=Math.floor(n*l/o)),await e.resize(l,i,{resample:r}))}async crop_margin(e,t=200){const r=e.clone().grayscale(),n=(0,a.min)(r.data)[0],o=(0,a.max)(r.data)[0]-n;if(0===o)return e;const s=t/255;let i=r.width,l=r.height,c=0,d=0;const u=r.data;for(let e=0;e<r.height;++e){const t=e*r.width;for(let a=0;a<r.width;++a)(u[t+a]-n)/o<s&&(i=Math.min(i,a),l=Math.min(l,e),c=Math.max(c,a),d=Math.max(d,e))}return e=await e.crop([i,l,c,d])}pad_image(e,t,r,{mode:n="constant",center:s=!1,constant_values:a=0}={}){const[i,l,c]=t;let d,u;if("number"==typeof r?(d=r,u=r):(d=r.width,u=r.height),d!==l||u!==i){const r=new Float32Array(d*u*c);if(Array.isArray(a))for(let e=0;e<r.length;++e)r[e]=a[e%c];else 0!==a&&r.fill(a);const[p,h]=s?[Math.floor((d-l)/2),Math.floor((u-i)/2)]:[0,0];for(let t=0;t<i;++t){const n=(t+h)*d,o=t*l;for(let t=0;t<l;++t){const s=(n+t+p)*c,a=(o+t)*c;for(let t=0;t<c;++t)r[s+t]=e[a+t]}}if("symmetric"===n){if(s)throw new Error("`center` padding is not supported when `mode` is set to `symmetric`.");const t=i-1,n=l-1;for(let s=0;s<u;++s){const a=s*d,u=(0,o.calculateReflectOffset)(s,t)*l;for(let t=0;t<d;++t){if(s<i&&t<l)continue;const d=(a+t)*c,p=(u+(0,o.calculateReflectOffset)(t,n))*c;for(let t=0;t<c;++t)r[d+t]=e[p+t]}}}e=r,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[r,n]=e.size;let o,s;if(this.do_thumbnail){const{height:e,width:r}=t;o=Math.min(e,r)}else Number.isInteger(t)?(o=t,s=this.config.max_size??o):void 0!==t&&(o=t.shortest_edge,s=t.longest_edge);if(void 0!==o||void 0!==s){const e=void 0===o?1:Math.max(o/r,o/n),t=r*e,a=n*e,i=void 0===s?1:Math.min(s/t,s/a);let l=Math.floor(Number((t*i).toFixed(2))),c=Math.floor(Number((a*i).toFixed(2)));return void 0!==this.size_divisibility&&([l,c]=M([l,c],this.size_divisibility)),[l,c]}if(void 0!==t&&void 0!==t.width&&void 0!==t.height){let e=t.width,o=t.height;if(this.config.keep_aspect_ratio&&this.config.ensure_multiple_of){let t=o/n,s=e/r;Math.abs(1-s)<Math.abs(1-t)?t=s:s=t,o=g(t*n,this.config.ensure_multiple_of),e=g(s*r,this.config.ensure_multiple_of)}return[e,o]}if(void 0!==this.size_divisibility)return M([r,n],this.size_divisibility);throw new Error(`Could not resize image due to unsupported \`this.size\` option in config: ${JSON.stringify(t)}`)}async resize(e){const[t,r]=this.get_resize_output_image_size(e,this.size);return await e.resize(t,r,{resample:this.resample})}async preprocess(e,{do_normalize:t=null,do_pad:r=null,do_convert_rgb:n=null,do_convert_grayscale:o=null,do_flip_channel_order:s=null}={}){this.do_crop_margin&&(e=await this.crop_margin(e));const[a,l]=e.size;if(n??this.do_convert_rgb?e=e.rgb():o&&(e=e.grayscale()),this.do_resize&&(e=await this.resize(e)),this.do_thumbnail&&(e=await this.thumbnail(e,this.size,this.resample)),this.do_center_crop){let t,r;Number.isInteger(this.crop_size)?(t=this.crop_size,r=this.crop_size):(t=this.crop_size.width,r=this.crop_size.height),e=await e.center_crop(t,r)}const c=[e.height,e.width];let d=Float32Array.from(e.data),u=[e.height,e.width,e.channels];if(this.do_rescale&&this.rescale(d),t??this.do_normalize){let t=this.image_mean;Array.isArray(this.image_mean)||(t=new Array(e.channels).fill(t));let r=this.image_std;if(Array.isArray(this.image_std)||(r=new Array(e.channels).fill(t)),t.length!==e.channels||r.length!==e.channels)throw new Error(`When set to arrays, the length of \`image_mean\` (${t.length}) and \`image_std\` (${r.length}) must match the number of channels in the image (${e.channels}).`);for(let n=0;n<d.length;n+=e.channels)for(let o=0;o<e.channels;++o)d[n+o]=(d[n+o]-t[o])/r[o]}if(r??this.do_pad)if(this.pad_size){const t=this.pad_image(d,[e.height,e.width,e.channels],this.pad_size);[d,u]=t}else if(this.size_divisibility){const[e,t]=M([u[1],u[0]],this.size_divisibility);[d,u]=this.pad_image(d,u,{width:e,height:t})}if(s??this.do_flip_channel_order){if(3!==u[2])throw new Error("Flipping channel order is only supported for RGB images.");for(let e=0;e<d.length;e+=3){const t=d[e];d[e]=d[e+2],d[e+2]=t}}return{original_size:[l,a],reshaped_input_size:c,pixel_values:new i.Tensor("float32",d,u).permute(2,0,1)}}async _call(e,...t){Array.isArray(e)||(e=[e]);const r=await Promise.all(e.map((e=>this.preprocess(e))));return{pixel_values:(0,i.stack)(r.map((e=>e.pixel_values)),0),original_sizes:r.map((e=>e.original_size)),reshaped_input_sizes:r.map((e=>e.reshaped_input_size))}}}class b extends T{post_process_semantic_segmentation(...e){return u(...e)}}class x extends T{post_process_semantic_segmentation(...e){return u(...e)}}class y extends T{}class k extends T{}class F extends k{}class P extends T{}class C extends T{}class v extends T{}class S extends v{}class A extends T{}class E extends T{}class L extends T{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 r=Math.floor(t/this.crop_pct),[n,o]=this.get_resize_output_image_size(e,{shortest_edge:r});e=await e.resize(n,o,{resample:this.resample}),e=await e.center_crop(t,t)}else e=await e.resize(t,t,{resample:this.resample});return e}}class z extends L{}class I extends T{}class B extends T{}class N extends T{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)))}}class D extends T{}class O extends T{}class V extends T{}class j extends T{}class R extends T{}class G extends R{}class q extends T{post_process_object_detection(...e){return d(...e)}}class W extends q{}class $ extends T{post_process_object_detection(...e){return d(...e)}}class U extends T{}class X extends T{}class Q extends T{pad_image(e,t,r,n={}){const[o,s,a]=t;let i=this.image_mean;Array.isArray(this.image_mean)||(i=new Array(a).fill(i));let l=this.image_std;Array.isArray(l)||(l=new Array(a).fill(i));const c=i.map(((e,t)=>-e/l[t]));return super.pad_image(e,t,r,{center:!0,constant_values:c,...n})}}class H extends Q{}class Y extends T{async _call(e){const t=await super._call(e),r=[t.pixel_values.dims[0],64,64],n=(0,i.full)(r,1n);return{...t,pixel_mask:n}}post_process_object_detection(...e){return d(...e)}post_process_panoptic_segmentation(...e){return _(...e)}post_process_instance_segmentation(){throw Error("Not implemented yet")}}class J extends T{post_process_panoptic_segmentation(...e){return _(...e)}post_process_instance_segmentation(){throw Error("Not implemented yet")}}class K extends T{post_process_object_detection(...e){return d(...e)}}class Z extends T{reshape_input_points(e,t,r,n=!1){e=structuredClone(e);let s=(0,o.calculateDimensions)(e);if(3===s.length)n||(s=[1,...s]),e=[e];else if(4!==s.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 n=0;n<e.length;++n){let o=t[n],s=r[n],a=[s[0]/o[0],s[1]/o[1]];for(let t=0;t<e[n].length;++t)for(let r=0;r<e[n][t].length;++r)for(let o=0;o<e[n][t][r].length;++o)e[n][t][r][o]*=a[o%2]}return new i.Tensor("float32",Float32Array.from(e.flat(1/0)),s)}add_input_labels(e,t){let r=(0,o.calculateDimensions)(e);if(2===r.length)r=[1,...r],e=[e];else if(3!==r.length)throw Error("The input_points must be a 4D tensor of shape `batch_size`, `point_batch_size`, `nb_points_per_image`, `2`.");if(r.some(((e,r)=>e!==t.dims[r])))throw Error(`The first ${r.length} dimensions of 'input_points' and 'input_labels' must be the same.`);return new i.Tensor("int64",e.flat(1/0).map(BigInt),r)}async _call(e,{input_points:t=null,input_labels:r=null,input_boxes:n=null}={}){const o=await super._call(e);if(t&&(o.input_points=this.reshape_input_points(t,o.original_sizes,o.reshaped_input_sizes)),r){if(!o.input_points)throw Error("`input_points` must be provided if `input_labels` are provided.");o.input_labels=this.add_input_labels(r,o.input_points)}return n&&(o.input_boxes=this.reshape_input_points(n,o.original_sizes,o.reshaped_input_sizes,!0)),o}async post_process_masks(e,t,r,{mask_threshold:n=0,binarize:o=!0,pad_size:s=null}={}){const a=[],l=[(s=s??this.pad_size).height,s.width];for(let s=0;s<t.length;++s){const c=t[s],d=r[s];let u=await(0,i.interpolate_4d)(e[s],{mode:"bilinear",size:l});if(u=u.slice(null,null,[0,d[0]],[0,d[1]]),u=await(0,i.interpolate_4d)(u,{mode:"bilinear",size:c}),o){const e=u.data,t=new Uint8Array(e.length);for(let r=0;r<e.length;++r)e[r]>n&&(t[r]=1);u=new i.Tensor("bool",t,u.dims)}a.push(u)}return a}generate_crop_boxes(e,t,{crop_n_layers:r=0,overlap_ratio:n=512/1500,points_per_crop:o=32,crop_n_points_downscale_factor:s=1}={}){}}class ee extends T{pad_image(e,t,r,n={}){const[o,s,a]=t;return super.pad_image(e,t,{width:s+(r-s%r)%r,height:o+(r-o%r)%r},{mode:"symmetric",center:!1,constant_values:-1,...n})}}class te extends T{async _call(e,t){Array.isArray(e)||(e=[e]),Array.isArray(t)||(t=[t]);const r=await Promise.all(e.map((e=>this.preprocess(e)))),n=await Promise.all(t.map((e=>this.preprocess(e,{do_normalize:!1,do_convert_rgb:!1,do_convert_grayscale:!0}))));return{pixel_values:(0,i.stack)(r.map(((e,t)=>(0,i.cat)([e.pixel_values,n[t].pixel_values],0))),0),original_sizes:r.map((e=>e.original_size)),reshaped_input_sizes:r.map((e=>e.reshaped_input_size))}}}class re extends w{constructor(e){super(e),this.config.mel_filters??=(0,l.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,l.window_function)(this.config.n_fft,"hann")}async _extract_fbank_features(e){const t=await(0,l.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}),r=t.data,n=(0,a.max)(r)[0];for(let e=0;e<r.length;++e)r[e]=(Math.max(r[e],n-8)+4)/4;return t}async _call(e){let t;f(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)}}}class ne extends w{_zero_mean_unit_var_norm(e){const t=e.reduce(((e,t)=>e+t),0)/e.length,r=e.reduce(((e,r)=>e+(r-t)**2),0)/e.length;return e.map((e=>(e-t)/Math.sqrt(r+1e-7)))}async _call(e){f(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 r=[1,t.length];return{input_values:new i.Tensor("float32",t,r),attention_mask:new i.Tensor("int64",new BigInt64Array(t.length).fill(1n),r)}}}class oe extends w{constructor(e){super(e);const t=this.config.sampling_rate,r=(0,l.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<r.length;++e)r[e].push(0);this.mel_filters=r,this.window=(0,l.window_function)(400,"povey",{periodic:!1})}async _extract_fbank_features(e,t){return e=e.map((e=>32768*e)),(0,l.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:r=2,do_normalize_per_mel_bins:n=!0,return_attention_mask:o=!0}={}){f(e,"SeamlessM4TFeatureExtractor");let s,a=await this._extract_fbank_features(e,this.config.max_length);if(n){const[e,t]=a.dims,r=a.data;for(let n=0;n<t;++n){let o=0;for(let s=0;s<e;++s)o+=r[s*t+n];const s=o/e;let a=0;for(let o=0;o<e;++o)a+=(r[o*t+n]-s)**2;a/=e-1;const i=Math.sqrt(a+1e-7);for(let o=0;o<e;++o){const e=o*t+n;r[e]=(r[e]-s)/i}}}if(t){const[e,t]=a.dims,n=a.data,l=e%r;if(l>0){const r=new Float32Array(t*(e+l));r.set(n),r.fill(this.config.padding_value,n.length);const c=e+l;a=new i.Tensor(a.type,r,[c,t]),o&&(s=new i.Tensor("int64",new BigInt64Array(c),[1,c]),s.data.fill(1n,0,e))}}const[l,c]=a.dims,d=this.config.stride;if(0!==l%d)throw new Error(`The number of frames (${l}) must be a multiple of the stride (${d}).`);const u=a.view(1,Math.floor(l/d),c*d),p={input_features:u};if(o){const e=u.dims[1],t=new BigInt64Array(e);if(s){const e=s.data;for(let r=1,n=0;r<l;r+=d,++n)t[n]=e[r]}else t.fill(1n);p.attention_mask=new i.Tensor("int64",t,[1,e])}return p}}class se extends w{constructor(e){super(e);const t=this.config.sampling_rate,r=(0,l.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<r.length;++e)r[e].push(0);this.mel_filters=r,this.window=(0,l.window_function)(400,"hann",{periodic:!1}),this.mean=this.config.mean,this.std=this.config.std}async _extract_fbank_features(e,t){return(0,l.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,max_num_frames:t,transpose:!0})}async _call(e){f(e,"ASTFeatureExtractor");const t=await this._extract_fbank_features(e,this.config.max_length);if(this.config.do_normalize){const e=2*this.std,r=t.data;for(let t=0;t<r.length;++t)r[t]=(r[t]-this.mean)/e}return{input_values:t.unsqueeze_(0)}}}class ae extends w{constructor(e){super(e),this.mel_filters=(0,l.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,null,"htk"),this.mel_filters_slaney=(0,l.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,"slaney","slaney"),this.window=(0,l.window_function)(this.config.fft_window_size,"hann")}async _get_input_mel(e,t,r,n){let o,s=!1;const a=e.length-t;if(a>0){if("rand_trunc"!==r)throw new Error(`Truncation strategy "${r}" not implemented`);{s=!0;const r=Math.floor(Math.random()*(a+1));e=e.subarray(r,r+t),o=await this._extract_fbank_features(e,this.mel_filters_slaney,this.config.nb_max_samples)}}else{if(a<0){let r=new Float64Array(t);if(r.set(e),"repeat"===n)for(let n=e.length;n<t;n+=e.length)r.set(e.subarray(0,Math.min(e.length,t-n)),n);else if("repeatpad"===n)for(let t=e.length;t<-a;t+=e.length)r.set(e,t);e=r}if("fusion"===r)throw new Error(`Truncation strategy "${r}" not implemented`);o=await this._extract_fbank_features(e,this.mel_filters_slaney,this.config.nb_max_samples)}return o.unsqueeze_(0)}async _extract_fbank_features(e,t,r=null){return(0,l.spectrogram)(e,this.window,this.config.fft_window_size,this.config.hop_length,{power:2,mel_filters:t,log_mel:"dB",max_num_frames:r,do_pad:!1,transpose:!0})}async _call(e,{max_length:t=null}={}){f(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)}}}class ie extends w{async _call(e){f(e,"PyAnnoteFeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));const t=[1,1,e.length];return{input_values:new i.Tensor("float32",e,t)}}samples_to_frames(e){return(e-this.config.offset)/this.config.step}post_process_speaker_diarization(e,t){const r=t/this.samples_to_frames(t)/this.config.sampling_rate,n=[];for(const t of e.tolist()){const e=[];let o=-1;for(let r=0;r<t.length;++r){const n=(0,a.softmax)(t[r]),[s,i]=(0,a.max)(n),[l,c]=[r,r+1];i!==o?(o=i,e.push({id:i,start:l,end:c,score:s})):(e.at(-1).end=c,e.at(-1).score+=s)}n.push(e.map((({id:e,start:t,end:n,score:o})=>({id:e,start:t*r,end:n*r,confidence:o/(n-t)}))))}return n}}class le extends w{constructor(e){super(e);const t=this.config.sampling_rate,r=(0,l.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<r.length;++e)r[e].push(0);this.mel_filters=r,this.window=(0,l.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,l.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){f(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,r=t.data,[n,o,s]=t.dims;for(let t=0;t<n;++t){const n=t*o*s,a=t*s;for(let t=0;t<o;++t){const o=n+t*s;for(let t=0;t<s;++t)r[o+t]-=e[a+t]}}}return{input_features:t}}}class ce extends w{}class de extends n.Callable{constructor(e){super(),this.feature_extractor=e}async _call(e,...t){return await this.feature_extractor(e,...t)}}class ue extends de{async _call(...e){return await this.feature_extractor(...e)}post_process_masks(...e){return this.feature_extractor.post_process_masks(...e)}reshape_input_points(...e){return this.feature_extractor.reshape_input_points(...e)}}class pe extends de{async _call(e){return await this.feature_extractor(e)}}class he extends de{async _call(e){return await this.feature_extractor(e)}}class me extends de{async _call(e){return await this.feature_extractor(e)}post_process_speaker_diarization(...e){return this.feature_extractor.post_process_speaker_diarization(...e)}}class _e extends de{async _call(e){return await this.feature_extractor(e)}}class fe extends de{}class ge extends de{constructor(e){super(e);const{tasks_answer_post_processing_type:t,task_prompts_without_inputs:r,task_prompts_with_input:n}=e.config;this.tasks_answer_post_processing_type=new Map(Object.entries(t??{})),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 r of e)if(this.task_prompts_without_inputs.has(r))t.push(this.task_prompts_without_inputs.get(r));else{for(const[e,n]of this.task_prompts_with_input)if(r.includes(e)){t.push(n.replaceAll("{input}",r).replaceAll(e,""));break}t.length!==e.length&&t.push(r)}return t}post_process_generation(e,t,r){const n=this.tasks_answer_post_processing_type.get(t)??"pure_text";let o;switch(e=e.replaceAll("<s>","").replaceAll("</s>",""),n){case"pure_text":o=e;break;case"description_with_bboxes":case"bboxes":case"phrase_grounding":case"ocr":const s="ocr"===n?"quad_boxes":"bboxes",a=e.matchAll(this.regexes[s]),i=[],l=[];for(const[e,t,...n]of a)i.push(t?t.trim():i.at(-1)??""),l.push(n.map(((e,t)=>(Number(e)+.5)/this.size_per_bin*r[t%2])));o={labels:i,[s]:l};break;default:throw new Error(`Task "${t}" (of type "${n}") not yet implemented.`)}return{[t]:o}}}class Me{static FEATURE_EXTRACTOR_CLASS_MAPPING={ImageFeatureExtractor:T,WhisperFeatureExtractor:re,ViTFeatureExtractor:I,MobileViTFeatureExtractor:R,MobileViTImageProcessor:G,MobileNetV1FeatureExtractor:D,MobileNetV2FeatureExtractor:O,MobileNetV3FeatureExtractor:V,MobileNetV4FeatureExtractor:j,OwlViTFeatureExtractor:q,Owlv2ImageProcessor:W,CLIPFeatureExtractor:v,CLIPImageProcessor:S,Florence2Processor:ge,ChineseCLIPFeatureExtractor:A,SiglipImageProcessor:E,ConvNextFeatureExtractor:L,ConvNextImageProcessor:z,SegformerFeatureExtractor:x,SapiensFeatureExtractor:b,BitImageProcessor:P,DPTImageProcessor:F,DPTFeatureExtractor:k,PvtImageProcessor:y,GLPNFeatureExtractor:C,BeitFeatureExtractor:X,DeiTFeatureExtractor:U,DetrFeatureExtractor:Y,RTDetrImageProcessor:$,MaskFormerFeatureExtractor:J,YolosFeatureExtractor:K,DonutFeatureExtractor:Q,NougatImageProcessor:H,EfficientNetImageProcessor:N,ViTImageProcessor:B,VitMatteImageProcessor:te,SamImageProcessor:Z,Swin2SRImageProcessor:ee,Wav2Vec2FeatureExtractor:ne,SeamlessM4TFeatureExtractor:oe,SpeechT5FeatureExtractor:ce,ASTFeatureExtractor:se,ClapFeatureExtractor:ae,PyAnnoteFeatureExtractor:ie,WeSpeakerFeatureExtractor:le};static PROCESSOR_CLASS_MAPPING={WhisperProcessor:pe,Wav2Vec2ProcessorWithLM:he,PyAnnoteProcessor:me,SamProcessor:ue,SpeechT5Processor:_e,OwlViTProcessor:fe,Florence2Processor:ge};static async from_pretrained(e,{progress_callback:t=null,config:r=null,cache_dir:n=null,local_files_only:o=!1,revision:a="main"}={}){let i=r??await(0,s.getModelJSON)(e,"preprocessor_config.json",!0,{progress_callback:t,config:r,cache_dir:n,local_files_only:o,revision:a}),l=i.feature_extractor_type??i.image_processor_type,c=this.FEATURE_EXTRACTOR_CLASS_MAPPING[l];if(!c){if(void 0===i.size)throw new Error(`Unknown Feature Extractor type: ${l}`);console.warn(`Feature extractor type "${l}" not found, assuming ImageFeatureExtractor due to size parameter in config.`),c=T}return new(this.PROCESSOR_CLASS_MAPPING[i.processor_class]??de)(new c(i))}}},"./src/tokenizers.js":
+  \***************************/(e,t,r)=>{r.r(t),r.d(t,{ASTFeatureExtractor:()=>ae,AutoProcessor:()=>we,BeitFeatureExtractor:()=>X,BitImageProcessor:()=>P,CLIPFeatureExtractor:()=>v,CLIPImageProcessor:()=>S,ChineseCLIPFeatureExtractor:()=>A,ClapFeatureExtractor:()=>ie,ConvNextFeatureExtractor:()=>L,ConvNextImageProcessor:()=>z,DPTFeatureExtractor:()=>k,DPTImageProcessor:()=>F,DeiTFeatureExtractor:()=>U,DetrFeatureExtractor:()=>J,DonutFeatureExtractor:()=>Q,DonutImageProcessor:()=>H,EfficientNetImageProcessor:()=>N,FeatureExtractor:()=>w,Florence2Processor:()=>Me,GLPNFeatureExtractor:()=>C,ImageFeatureExtractor:()=>T,MaskFormerFeatureExtractor:()=>K,MobileNetV1FeatureExtractor:()=>D,MobileNetV2FeatureExtractor:()=>O,MobileNetV3FeatureExtractor:()=>V,MobileNetV4FeatureExtractor:()=>j,MobileViTFeatureExtractor:()=>R,MobileViTImageProcessor:()=>G,NougatImageProcessor:()=>Y,OwlViTFeatureExtractor:()=>q,OwlViTProcessor:()=>ge,Owlv2ImageProcessor:()=>$,Processor:()=>ue,PvtImageProcessor:()=>y,PyAnnoteFeatureExtractor:()=>le,PyAnnoteProcessor:()=>_e,RTDetrImageProcessor:()=>W,SamImageProcessor:()=>ee,SamProcessor:()=>pe,SapiensFeatureExtractor:()=>b,SeamlessM4TFeatureExtractor:()=>se,SegformerFeatureExtractor:()=>x,SiglipImageProcessor:()=>E,SpeechT5FeatureExtractor:()=>de,SpeechT5Processor:()=>fe,Swin2SRImageProcessor:()=>te,ViTFeatureExtractor:()=>I,ViTImageProcessor:()=>B,VitMatteImageProcessor:()=>re,Wav2Vec2FeatureExtractor:()=>oe,Wav2Vec2ProcessorWithLM:()=>me,WeSpeakerFeatureExtractor:()=>ce,WhisperFeatureExtractor:()=>ne,WhisperProcessor:()=>he,YolosFeatureExtractor:()=>Z});var n=r(/*! ./utils/generic.js */"./src/utils/generic.js"),o=r(/*! ./utils/core.js */"./src/utils/core.js"),s=r(/*! ./utils/hub.js */"./src/utils/hub.js"),a=r(/*! ./utils/maths.js */"./src/utils/maths.js"),i=r(/*! ./utils/tensor.js */"./src/utils/tensor.js"),l=(r(/*! ./utils/image.js */"./src/utils/image.js"),r(/*! ./utils/audio.js */"./src/utils/audio.js"));function c([e,t,r,n]){return[e-r/2,t-n/2,e+r/2,t+n/2]}function d(e,t=.5,r=null,n=!1){const o=e.logits,s=e.pred_boxes,[i,l,d]=o.dims;if(null!==r&&r.length!==i)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");let u=[];for(let e=0;e<i;++e){let i=null!==r?r[e]:null,p={boxes:[],classes:[],scores:[]},h=o[e],m=s[e];for(let e=0;e<l;++e){let r,o=h[e],s=[];if(n){r=o.sigmoid().data;for(let e=0;e<r.length;++e)r[e]>t&&s.push(e)}else{let e=(0,a.max)(o.data)[1];if(e===d-1)continue;if(r=(0,a.softmax)(o.data),r[e]<t)continue;s.push(e)}for(const t of s){let n=m[e].data;n=c(n),null!==i&&(n=n.map(((e,t)=>e*i[(t+1)%2]))),p.boxes.push(n),p.classes.push(t),p.scores.push(r[t])}}u.push(p)}return u}function u(e,t=null){const r=e.logits,n=r.dims[0];if(null!==t&&t.length!==n)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<n;++e){const n=null!==t?t[e]:null;let s=r[e];null!==n&&(s=(0,i.interpolate)(s,n,"bilinear",!1));const[a,l]=n??s.dims.slice(-2),c=new i.Tensor("int32",new Int32Array(a*l),[a,l]),d=s[0].data,u=c.data;for(let e=1;e<s.dims[0];++e){const t=s[e].data;for(let r=0;r<t.length;++r)t[r]>d[r]&&(d[r]=t[r],u[r]=e)}const p=new Array(s.dims[0]);for(let e=0;e<u.length;++e){const t=u[e];p[t]=t}const h=p.filter((e=>void 0!==e));o.push({segmentation:c,labels:h})}return o}function p(e,t,r,n){const o=[],s=[],i=[];for(let l=0;l<e.dims[0];++l){const c=e[l],d=t[l],u=(0,a.max)(c.data)[1];if(u===n)continue;const p=(0,a.softmax)(c.data)[u];p>r&&(o.push(d),s.push(p),i.push(u))}return[o,s,i]}function h(e,t,r,n=.5,o=.8){const s=[];let a=0,i=0;const l=t[r].data;for(let t=0;t<e.length;++t)e[t]===r&&(s.push(t),++a),l[t]>=n&&++i;let c=a>0&&i>0;if(c){c=a/i>o}return[c,s]}function m(e,t,r,n,o,s=null,a=null){const[l,c]=a??e[0].dims,d=new i.Tensor("int32",new Int32Array(l*c),[l,c]),u=[];if(null!==a)for(let t=0;t<e.length;++t)e[t]=(0,i.interpolate)(e[t],a,"bilinear",!1);const p=new Int32Array(e[0].data.length),m=new Float32Array(e[0].data.length);for(let r=0;r<e.length;++r){let n=t[r];const o=e[r].data;for(let e=0;e<o.length;++e)o[e]*=n,o[e]>m[e]&&(p[e]=r,m[e]=o[e])}let _=0;const f=d.data;for(let s=0;s<r.length;++s){const a=r[s],[i,l]=h(p,e,s,n,o);if(i){++_;for(const e of l)f[e]=_;u.push({id:_,label_id:a,score:t[s]})}}return[d,u]}function _(e,t=.5,r=.5,n=.8,o=null,s=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!==s&&s.length!==c)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");let h=[];for(let e=0;e<c;++e){let c=null!==s?s[e]:null,d=a[e],_=l[e],[f,g,M]=p(d,_,t,u);if(0===M.length){let[e,t]=c??_.dims.slice(-2),r=new i.Tensor("int32",new Int32Array(e*t).fill(-1),[e,t]);h.push({segmentation:r,segments_info:[]});continue}let[w,T]=m(f,g,M,r,n,o,c);h.push({segmentation:w,segments_info:T})}return h}function f(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.`)}function g(e,t,r=0,n=null){const o=e/t;let s=(0,a.bankers_round)(o)*t;return null!==n&&s>n&&(s=Math.floor(o)*t),s<r&&(s=Math.ceil(o)*t),s}function M([e,t],r){return[Math.max(Math.floor(e/r),1)*r,Math.max(Math.floor(t/r),1)*r]}class w extends n.Callable{constructor(e){super(),this.config=e}}class T extends w{constructor(e){super(e),this.image_mean=this.config.image_mean??this.config.mean,this.image_std=this.config.image_std??this.config.std,this.resample=this.config.resample??2,this.do_rescale=this.config.do_rescale??!0,this.rescale_factor=this.config.rescale_factor??1/255,this.do_normalize=this.config.do_normalize,this.do_resize=this.config.do_resize,this.do_thumbnail=this.config.do_thumbnail,this.size=this.config.size,this.size_divisibility=this.config.size_divisibility??this.config.size_divisor,this.do_center_crop=this.config.do_center_crop,this.crop_size=this.config.crop_size,this.do_convert_rgb=this.config.do_convert_rgb??!0,this.do_crop_margin=this.config.do_crop_margin,this.pad_size=this.config.pad_size,this.do_pad=this.config.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=this.config.do_flip_channel_order??!1}async thumbnail(e,t,r=2){const n=e.height,o=e.width,s=t.height,a=t.width;let i=Math.min(n,s),l=Math.min(o,a);return i===n&&l===o?e:(n>o?l=Math.floor(o*i/n):o>n&&(i=Math.floor(n*l/o)),await e.resize(l,i,{resample:r}))}async crop_margin(e,t=200){const r=e.clone().grayscale(),n=(0,a.min)(r.data)[0],o=(0,a.max)(r.data)[0]-n;if(0===o)return e;const s=t/255;let i=r.width,l=r.height,c=0,d=0;const u=r.data;for(let e=0;e<r.height;++e){const t=e*r.width;for(let a=0;a<r.width;++a)(u[t+a]-n)/o<s&&(i=Math.min(i,a),l=Math.min(l,e),c=Math.max(c,a),d=Math.max(d,e))}return e=await e.crop([i,l,c,d])}pad_image(e,t,r,{mode:n="constant",center:s=!1,constant_values:a=0}={}){const[i,l,c]=t;let d,u;if("number"==typeof r?(d=r,u=r):(d=r.width,u=r.height),d!==l||u!==i){const r=new Float32Array(d*u*c);if(Array.isArray(a))for(let e=0;e<r.length;++e)r[e]=a[e%c];else 0!==a&&r.fill(a);const[p,h]=s?[Math.floor((d-l)/2),Math.floor((u-i)/2)]:[0,0];for(let t=0;t<i;++t){const n=(t+h)*d,o=t*l;for(let t=0;t<l;++t){const s=(n+t+p)*c,a=(o+t)*c;for(let t=0;t<c;++t)r[s+t]=e[a+t]}}if("symmetric"===n){if(s)throw new Error("`center` padding is not supported when `mode` is set to `symmetric`.");const t=i-1,n=l-1;for(let s=0;s<u;++s){const a=s*d,u=(0,o.calculateReflectOffset)(s,t)*l;for(let t=0;t<d;++t){if(s<i&&t<l)continue;const d=(a+t)*c,p=(u+(0,o.calculateReflectOffset)(t,n))*c;for(let t=0;t<c;++t)r[d+t]=e[p+t]}}}e=r,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[r,n]=e.size;let o,s;if(this.do_thumbnail){const{height:e,width:r}=t;o=Math.min(e,r)}else Number.isInteger(t)?(o=t,s=this.config.max_size??o):void 0!==t&&(o=t.shortest_edge,s=t.longest_edge);if(void 0!==o||void 0!==s){const e=void 0===o?1:Math.max(o/r,o/n),t=r*e,a=n*e,i=void 0===s?1:Math.min(s/t,s/a);let l=Math.floor(Number((t*i).toFixed(2))),c=Math.floor(Number((a*i).toFixed(2)));return void 0!==this.size_divisibility&&([l,c]=M([l,c],this.size_divisibility)),[l,c]}if(void 0!==t&&void 0!==t.width&&void 0!==t.height){let e=t.width,o=t.height;if(this.config.keep_aspect_ratio&&this.config.ensure_multiple_of){let t=o/n,s=e/r;Math.abs(1-s)<Math.abs(1-t)?t=s:s=t,o=g(t*n,this.config.ensure_multiple_of),e=g(s*r,this.config.ensure_multiple_of)}return[e,o]}if(void 0!==this.size_divisibility)return M([r,n],this.size_divisibility);throw new Error(`Could not resize image due to unsupported \`this.size\` option in config: ${JSON.stringify(t)}`)}async resize(e){const[t,r]=this.get_resize_output_image_size(e,this.size);return await e.resize(t,r,{resample:this.resample})}async preprocess(e,{do_normalize:t=null,do_pad:r=null,do_convert_rgb:n=null,do_convert_grayscale:o=null,do_flip_channel_order:s=null}={}){this.do_crop_margin&&(e=await this.crop_margin(e));const[a,l]=e.size;if(n??this.do_convert_rgb?e=e.rgb():o&&(e=e.grayscale()),this.do_resize&&(e=await this.resize(e)),this.do_thumbnail&&(e=await this.thumbnail(e,this.size,this.resample)),this.do_center_crop){let t,r;Number.isInteger(this.crop_size)?(t=this.crop_size,r=this.crop_size):(t=this.crop_size.width,r=this.crop_size.height),e=await e.center_crop(t,r)}const c=[e.height,e.width];let d=Float32Array.from(e.data),u=[e.height,e.width,e.channels];if(this.do_rescale&&this.rescale(d),t??this.do_normalize){let t=this.image_mean;Array.isArray(this.image_mean)||(t=new Array(e.channels).fill(t));let r=this.image_std;if(Array.isArray(this.image_std)||(r=new Array(e.channels).fill(t)),t.length!==e.channels||r.length!==e.channels)throw new Error(`When set to arrays, the length of \`image_mean\` (${t.length}) and \`image_std\` (${r.length}) must match the number of channels in the image (${e.channels}).`);for(let n=0;n<d.length;n+=e.channels)for(let o=0;o<e.channels;++o)d[n+o]=(d[n+o]-t[o])/r[o]}if(r??this.do_pad)if(this.pad_size){const t=this.pad_image(d,[e.height,e.width,e.channels],this.pad_size);[d,u]=t}else if(this.size_divisibility){const[e,t]=M([u[1],u[0]],this.size_divisibility);[d,u]=this.pad_image(d,u,{width:e,height:t})}if(s??this.do_flip_channel_order){if(3!==u[2])throw new Error("Flipping channel order is only supported for RGB images.");for(let e=0;e<d.length;e+=3){const t=d[e];d[e]=d[e+2],d[e+2]=t}}return{original_size:[l,a],reshaped_input_size:c,pixel_values:new i.Tensor("float32",d,u).permute(2,0,1)}}async _call(e,...t){Array.isArray(e)||(e=[e]);const r=await Promise.all(e.map((e=>this.preprocess(e))));return{pixel_values:(0,i.stack)(r.map((e=>e.pixel_values)),0),original_sizes:r.map((e=>e.original_size)),reshaped_input_sizes:r.map((e=>e.reshaped_input_size))}}}class b extends T{post_process_semantic_segmentation(...e){return u(...e)}}class x extends T{post_process_semantic_segmentation(...e){return u(...e)}}class y extends T{}class k extends T{}class F extends k{}class P extends T{}class C extends T{}class v extends T{}class S extends v{}class A extends T{}class E extends T{}class L extends T{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 r=Math.floor(t/this.crop_pct),[n,o]=this.get_resize_output_image_size(e,{shortest_edge:r});e=await e.resize(n,o,{resample:this.resample}),e=await e.center_crop(t,t)}else e=await e.resize(t,t,{resample:this.resample});return e}}class z extends L{}class I extends T{}class B extends T{}class N extends T{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)))}}class D extends T{}class O extends T{}class V extends T{}class j extends T{}class R extends T{}class G extends R{}class q extends T{post_process_object_detection(...e){return d(...e)}}class $ extends q{}class W extends T{post_process_object_detection(...e){return d(...e)}}class U extends T{}class X extends T{}class Q extends T{pad_image(e,t,r,n={}){const[o,s,a]=t;let i=this.image_mean;Array.isArray(this.image_mean)||(i=new Array(a).fill(i));let l=this.image_std;Array.isArray(l)||(l=new Array(a).fill(i));const c=i.map(((e,t)=>-e/l[t]));return super.pad_image(e,t,r,{center:!0,constant_values:c,...n})}}class H extends Q{}class Y extends Q{}class J extends T{async _call(e){const t=await super._call(e),r=[t.pixel_values.dims[0],64,64],n=(0,i.full)(r,1n);return{...t,pixel_mask:n}}post_process_object_detection(...e){return d(...e)}post_process_panoptic_segmentation(...e){return _(...e)}post_process_instance_segmentation(){throw Error("Not implemented yet")}}class K extends T{post_process_panoptic_segmentation(...e){return _(...e)}post_process_instance_segmentation(){throw Error("Not implemented yet")}}class Z extends T{post_process_object_detection(...e){return d(...e)}}class ee extends T{reshape_input_points(e,t,r,n=!1){e=structuredClone(e);let s=(0,o.calculateDimensions)(e);if(3===s.length)n||(s=[1,...s]),e=[e];else if(4!==s.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 n=0;n<e.length;++n){let o=t[n],s=r[n],a=[s[0]/o[0],s[1]/o[1]];for(let t=0;t<e[n].length;++t)for(let r=0;r<e[n][t].length;++r)for(let o=0;o<e[n][t][r].length;++o)e[n][t][r][o]*=a[o%2]}return new i.Tensor("float32",Float32Array.from(e.flat(1/0)),s)}add_input_labels(e,t){let r=(0,o.calculateDimensions)(e);if(2===r.length)r=[1,...r],e=[e];else if(3!==r.length)throw Error("The input_points must be a 4D tensor of shape `batch_size`, `point_batch_size`, `nb_points_per_image`, `2`.");if(r.some(((e,r)=>e!==t.dims[r])))throw Error(`The first ${r.length} dimensions of 'input_points' and 'input_labels' must be the same.`);return new i.Tensor("int64",e.flat(1/0).map(BigInt),r)}async _call(e,{input_points:t=null,input_labels:r=null,input_boxes:n=null}={}){const o=await super._call(e);if(t&&(o.input_points=this.reshape_input_points(t,o.original_sizes,o.reshaped_input_sizes)),r){if(!o.input_points)throw Error("`input_points` must be provided if `input_labels` are provided.");o.input_labels=this.add_input_labels(r,o.input_points)}return n&&(o.input_boxes=this.reshape_input_points(n,o.original_sizes,o.reshaped_input_sizes,!0)),o}async post_process_masks(e,t,r,{mask_threshold:n=0,binarize:o=!0,pad_size:s=null}={}){const a=[],l=[(s=s??this.pad_size).height,s.width];for(let s=0;s<t.length;++s){const c=t[s],d=r[s];let u=await(0,i.interpolate_4d)(e[s],{mode:"bilinear",size:l});if(u=u.slice(null,null,[0,d[0]],[0,d[1]]),u=await(0,i.interpolate_4d)(u,{mode:"bilinear",size:c}),o){const e=u.data,t=new Uint8Array(e.length);for(let r=0;r<e.length;++r)e[r]>n&&(t[r]=1);u=new i.Tensor("bool",t,u.dims)}a.push(u)}return a}generate_crop_boxes(e,t,{crop_n_layers:r=0,overlap_ratio:n=512/1500,points_per_crop:o=32,crop_n_points_downscale_factor:s=1}={}){}}class te extends T{pad_image(e,t,r,n={}){const[o,s,a]=t;return super.pad_image(e,t,{width:s+(r-s%r)%r,height:o+(r-o%r)%r},{mode:"symmetric",center:!1,constant_values:-1,...n})}}class re extends T{async _call(e,t){Array.isArray(e)||(e=[e]),Array.isArray(t)||(t=[t]);const r=await Promise.all(e.map((e=>this.preprocess(e)))),n=await Promise.all(t.map((e=>this.preprocess(e,{do_normalize:!1,do_convert_rgb:!1,do_convert_grayscale:!0}))));return{pixel_values:(0,i.stack)(r.map(((e,t)=>(0,i.cat)([e.pixel_values,n[t].pixel_values],0))),0),original_sizes:r.map((e=>e.original_size)),reshaped_input_sizes:r.map((e=>e.reshaped_input_size))}}}class ne extends w{constructor(e){super(e),this.config.mel_filters??=(0,l.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,l.window_function)(this.config.n_fft,"hann")}async _extract_fbank_features(e){const t=await(0,l.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}),r=t.data,n=(0,a.max)(r)[0];for(let e=0;e<r.length;++e)r[e]=(Math.max(r[e],n-8)+4)/4;return t}async _call(e){let t;f(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)}}}class oe extends w{_zero_mean_unit_var_norm(e){const t=e.reduce(((e,t)=>e+t),0)/e.length,r=e.reduce(((e,r)=>e+(r-t)**2),0)/e.length;return e.map((e=>(e-t)/Math.sqrt(r+1e-7)))}async _call(e){f(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 r=[1,t.length];return{input_values:new i.Tensor("float32",t,r),attention_mask:new i.Tensor("int64",new BigInt64Array(t.length).fill(1n),r)}}}class se extends w{constructor(e){super(e);const t=this.config.sampling_rate,r=(0,l.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<r.length;++e)r[e].push(0);this.mel_filters=r,this.window=(0,l.window_function)(400,"povey",{periodic:!1})}async _extract_fbank_features(e,t){return e=e.map((e=>32768*e)),(0,l.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:r=2,do_normalize_per_mel_bins:n=!0,return_attention_mask:o=!0}={}){f(e,"SeamlessM4TFeatureExtractor");let s,a=await this._extract_fbank_features(e,this.config.max_length);if(n){const[e,t]=a.dims,r=a.data;for(let n=0;n<t;++n){let o=0;for(let s=0;s<e;++s)o+=r[s*t+n];const s=o/e;let a=0;for(let o=0;o<e;++o)a+=(r[o*t+n]-s)**2;a/=e-1;const i=Math.sqrt(a+1e-7);for(let o=0;o<e;++o){const e=o*t+n;r[e]=(r[e]-s)/i}}}if(t){const[e,t]=a.dims,n=a.data,l=e%r;if(l>0){const r=new Float32Array(t*(e+l));r.set(n),r.fill(this.config.padding_value,n.length);const c=e+l;a=new i.Tensor(a.type,r,[c,t]),o&&(s=new i.Tensor("int64",new BigInt64Array(c),[1,c]),s.data.fill(1n,0,e))}}const[l,c]=a.dims,d=this.config.stride;if(0!==l%d)throw new Error(`The number of frames (${l}) must be a multiple of the stride (${d}).`);const u=a.view(1,Math.floor(l/d),c*d),p={input_features:u};if(o){const e=u.dims[1],t=new BigInt64Array(e);if(s){const e=s.data;for(let r=1,n=0;r<l;r+=d,++n)t[n]=e[r]}else t.fill(1n);p.attention_mask=new i.Tensor("int64",t,[1,e])}return p}}class ae extends w{constructor(e){super(e);const t=this.config.sampling_rate,r=(0,l.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<r.length;++e)r[e].push(0);this.mel_filters=r,this.window=(0,l.window_function)(400,"hann",{periodic:!1}),this.mean=this.config.mean,this.std=this.config.std}async _extract_fbank_features(e,t){return(0,l.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,max_num_frames:t,transpose:!0})}async _call(e){f(e,"ASTFeatureExtractor");const t=await this._extract_fbank_features(e,this.config.max_length);if(this.config.do_normalize){const e=2*this.std,r=t.data;for(let t=0;t<r.length;++t)r[t]=(r[t]-this.mean)/e}return{input_values:t.unsqueeze_(0)}}}class ie extends w{constructor(e){super(e),this.mel_filters=(0,l.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,null,"htk"),this.mel_filters_slaney=(0,l.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,"slaney","slaney"),this.window=(0,l.window_function)(this.config.fft_window_size,"hann")}async _get_input_mel(e,t,r,n){let o,s=!1;const a=e.length-t;if(a>0){if("rand_trunc"!==r)throw new Error(`Truncation strategy "${r}" not implemented`);{s=!0;const r=Math.floor(Math.random()*(a+1));e=e.subarray(r,r+t),o=await this._extract_fbank_features(e,this.mel_filters_slaney,this.config.nb_max_samples)}}else{if(a<0){let r=new Float64Array(t);if(r.set(e),"repeat"===n)for(let n=e.length;n<t;n+=e.length)r.set(e.subarray(0,Math.min(e.length,t-n)),n);else if("repeatpad"===n)for(let t=e.length;t<-a;t+=e.length)r.set(e,t);e=r}if("fusion"===r)throw new Error(`Truncation strategy "${r}" not implemented`);o=await this._extract_fbank_features(e,this.mel_filters_slaney,this.config.nb_max_samples)}return o.unsqueeze_(0)}async _extract_fbank_features(e,t,r=null){return(0,l.spectrogram)(e,this.window,this.config.fft_window_size,this.config.hop_length,{power:2,mel_filters:t,log_mel:"dB",max_num_frames:r,do_pad:!1,transpose:!0})}async _call(e,{max_length:t=null}={}){f(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)}}}class le extends w{async _call(e){f(e,"PyAnnoteFeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));const t=[1,1,e.length];return{input_values:new i.Tensor("float32",e,t)}}samples_to_frames(e){return(e-this.config.offset)/this.config.step}post_process_speaker_diarization(e,t){const r=t/this.samples_to_frames(t)/this.config.sampling_rate,n=[];for(const t of e.tolist()){const e=[];let o=-1;for(let r=0;r<t.length;++r){const n=(0,a.softmax)(t[r]),[s,i]=(0,a.max)(n),[l,c]=[r,r+1];i!==o?(o=i,e.push({id:i,start:l,end:c,score:s})):(e.at(-1).end=c,e.at(-1).score+=s)}n.push(e.map((({id:e,start:t,end:n,score:o})=>({id:e,start:t*r,end:n*r,confidence:o/(n-t)}))))}return n}}class ce extends w{constructor(e){super(e);const t=this.config.sampling_rate,r=(0,l.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<r.length;++e)r[e].push(0);this.mel_filters=r,this.window=(0,l.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,l.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){f(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,r=t.data,[n,o,s]=t.dims;for(let t=0;t<n;++t){const n=t*o*s,a=t*s;for(let t=0;t<o;++t){const o=n+t*s;for(let t=0;t<s;++t)r[o+t]-=e[a+t]}}}return{input_features:t}}}class de extends w{}class ue extends n.Callable{constructor(e){super(),this.feature_extractor=e}async _call(e,...t){return await this.feature_extractor(e,...t)}}class pe extends ue{async _call(...e){return await this.feature_extractor(...e)}post_process_masks(...e){return this.feature_extractor.post_process_masks(...e)}reshape_input_points(...e){return this.feature_extractor.reshape_input_points(...e)}}class he extends ue{async _call(e){return await this.feature_extractor(e)}}class me extends ue{async _call(e){return await this.feature_extractor(e)}}class _e extends ue{async _call(e){return await this.feature_extractor(e)}post_process_speaker_diarization(...e){return this.feature_extractor.post_process_speaker_diarization(...e)}}class fe extends ue{async _call(e){return await this.feature_extractor(e)}}class ge extends ue{}class Me extends ue{constructor(e){super(e);const{tasks_answer_post_processing_type:t,task_prompts_without_inputs:r,task_prompts_with_input:n}=e.config;this.tasks_answer_post_processing_type=new Map(Object.entries(t??{})),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 r of e)if(this.task_prompts_without_inputs.has(r))t.push(this.task_prompts_without_inputs.get(r));else{for(const[e,n]of this.task_prompts_with_input)if(r.includes(e)){t.push(n.replaceAll("{input}",r).replaceAll(e,""));break}t.length!==e.length&&t.push(r)}return t}post_process_generation(e,t,r){const n=this.tasks_answer_post_processing_type.get(t)??"pure_text";let o;switch(e=e.replaceAll("<s>","").replaceAll("</s>",""),n){case"pure_text":o=e;break;case"description_with_bboxes":case"bboxes":case"phrase_grounding":case"ocr":const s="ocr"===n?"quad_boxes":"bboxes",a=e.matchAll(this.regexes[s]),i=[],l=[];for(const[e,t,...n]of a)i.push(t?t.trim():i.at(-1)??""),l.push(n.map(((e,t)=>(Number(e)+.5)/this.size_per_bin*r[t%2])));o={labels:i,[s]:l};break;default:throw new Error(`Task "${t}" (of type "${n}") not yet implemented.`)}return{[t]:o}}}class we{static FEATURE_EXTRACTOR_CLASS_MAPPING={ImageFeatureExtractor:T,WhisperFeatureExtractor:ne,ViTFeatureExtractor:I,MobileViTFeatureExtractor:R,MobileViTImageProcessor:G,MobileNetV1FeatureExtractor:D,MobileNetV2FeatureExtractor:O,MobileNetV3FeatureExtractor:V,MobileNetV4FeatureExtractor:j,OwlViTFeatureExtractor:q,Owlv2ImageProcessor:$,CLIPFeatureExtractor:v,CLIPImageProcessor:S,Florence2Processor:Me,ChineseCLIPFeatureExtractor:A,SiglipImageProcessor:E,ConvNextFeatureExtractor:L,ConvNextImageProcessor:z,SegformerFeatureExtractor:x,SapiensFeatureExtractor:b,BitImageProcessor:P,DPTImageProcessor:F,DPTFeatureExtractor:k,PvtImageProcessor:y,GLPNFeatureExtractor:C,BeitFeatureExtractor:X,DeiTFeatureExtractor:U,DetrFeatureExtractor:J,RTDetrImageProcessor:W,MaskFormerFeatureExtractor:K,YolosFeatureExtractor:Z,DonutFeatureExtractor:Q,DonutImageProcessor:H,NougatImageProcessor:Y,EfficientNetImageProcessor:N,ViTImageProcessor:B,VitMatteImageProcessor:re,SamImageProcessor:ee,Swin2SRImageProcessor:te,Wav2Vec2FeatureExtractor:oe,SeamlessM4TFeatureExtractor:se,SpeechT5FeatureExtractor:de,ASTFeatureExtractor:ae,ClapFeatureExtractor:ie,PyAnnoteFeatureExtractor:le,WeSpeakerFeatureExtractor:ce};static PROCESSOR_CLASS_MAPPING={WhisperProcessor:he,Wav2Vec2ProcessorWithLM:me,PyAnnoteProcessor:_e,SamProcessor:pe,SpeechT5Processor:fe,OwlViTProcessor:ge,Florence2Processor:Me};static async from_pretrained(e,{progress_callback:t=null,config:r=null,cache_dir:n=null,local_files_only:o=!1,revision:a="main"}={}){let i=r??await(0,s.getModelJSON)(e,"preprocessor_config.json",!0,{progress_callback:t,config:r,cache_dir:n,local_files_only:o,revision:a}),l=i.feature_extractor_type??i.image_processor_type,c=this.FEATURE_EXTRACTOR_CLASS_MAPPING[l];if(!c){if(void 0===i.size)throw new Error(`Unknown Feature Extractor type: ${l}`);console.warn(`Feature extractor type "${l}" not found, assuming ImageFeatureExtractor due to size parameter in config.`),c=T}return new(this.PROCESSOR_CLASS_MAPPING[i.processor_class]??ue)(new c(i))}}},"./src/tokenizers.js":
 /*!***************************!*\
   !*** ./src/tokenizers.js ***!
-  \***************************/(e,t,r)=>{r.r(t),r.d(t,{AlbertTokenizer:()=>xe,AutoTokenizer:()=>ht,BartTokenizer:()=>Ne,BertTokenizer:()=>be,BlenderbotSmallTokenizer:()=>lt,BlenderbotTokenizer:()=>it,BloomTokenizer:()=>je,CLIPTokenizer:()=>nt,CamembertTokenizer:()=>Ee,CodeGenTokenizer:()=>rt,CodeLlamaTokenizer:()=>qe,CohereTokenizer:()=>pt,ConvBertTokenizer:()=>ve,DebertaTokenizer:()=>Fe,DebertaV2Tokenizer:()=>Pe,DistilBertTokenizer:()=>Ae,ElectraTokenizer:()=>ze,EsmTokenizer:()=>Qe,FalconTokenizer:()=>Ue,GPT2Tokenizer:()=>Be,GPTNeoXTokenizer:()=>Xe,GemmaTokenizer:()=>Ye,Grok1Tokenizer:()=>Je,HerbertTokenizer:()=>Ce,LlamaTokenizer:()=>Ge,M2M100Tokenizer:()=>et,MBart50Tokenizer:()=>Oe,MBartTokenizer:()=>De,MPNetTokenizer:()=>$e,MarianTokenizer:()=>st,MobileBertTokenizer:()=>ye,NllbTokenizer:()=>Ze,NougatTokenizer:()=>dt,PreTrainedTokenizer:()=>Te,Qwen2Tokenizer:()=>He,RoFormerTokenizer:()=>Se,RobertaTokenizer:()=>Ve,SiglipTokenizer:()=>ot,SpeechT5Tokenizer:()=>ct,SqueezeBertTokenizer:()=>ke,T5Tokenizer:()=>Ie,TokenizerModel:()=>y,VitsTokenizer:()=>ut,Wav2Vec2CTCTokenizer:()=>at,WhisperTokenizer:()=>tt,XLMRobertaTokenizer:()=>We,XLMTokenizer:()=>Le,is_chinese_char:()=>g});var n=r(/*! ./utils/generic.js */"./src/utils/generic.js"),o=r(/*! ./utils/core.js */"./src/utils/core.js"),s=r(/*! ./utils/hub.js */"./src/utils/hub.js"),a=r(/*! ./utils/maths.js */"./src/utils/maths.js"),i=r(/*! ./utils/tensor.js */"./src/utils/tensor.js"),l=r(/*! ./utils/data-structures.js */"./src/utils/data-structures.js"),c=r(/*! @huggingface/jinja */"./node_modules/@huggingface/jinja/dist/index.js"),d=r(/*! ./models/whisper/common_whisper.js */"./src/models/whisper/common_whisper.js");r(/*! ./utils/constants.js */"./src/utils/constants.js");async function u(e,t){const r=await Promise.all([(0,s.getModelJSON)(e,"tokenizer.json",!0,t),(0,s.getModelJSON)(e,"tokenizer_config.json",!0,t)]);return null!==t.legacy&&(r[1].legacy=t.legacy),r}function p(e,t=!0){if(void 0!==e.Regex){let t=e.Regex.replace(/\\([#&~])/g,"$1");for(const[e,r]of b)t=t.replaceAll(e,r);return new RegExp(t,"gu")}if(void 0!==e.String){const r=(0,o.escapeRegExp)(e.String);return new RegExp(t?r:`(${r})`,"gu")}return console.warn("Unknown pattern type:",e),null}function h(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 ${t.length}.`)}}function _(e){return e.replace(/ \./g,".").replace(/ \?/g,"?").replace(/ \!/g,"!").replace(/ ,/g,",").replace(/ \' /g,"'").replace(/ n\'t/g,"n't").replace(/ \'m/g,"'m").replace(/ \'s/g,"'s").replace(/ \'ve/g,"'ve").replace(/ \'re/g,"'re")}function f(e){return e.replace(/\p{M}/gu,"")}function g(e){return e>=19968&&e<=40959||e>=13312&&e<=19903||e>=131072&&e<=173791||e>=173824&&e<=177983||e>=177984&&e<=178207||e>=178208&&e<=183983||e>=63744&&e<=64255||e>=194560&&e<=195103}const M="\\p{P}\\u0021-\\u002F\\u003A-\\u0040\\u005B-\\u0060\\u007B-\\u007E",w=new RegExp(`^[${M}]+$`,"gu"),T=".,!?…。，、।۔،",b=new Map([["(?i:'s|'t|'re|'ve|'m|'ll|'d)","(?:'([sS]|[tT]|[rR][eE]|[vV][eE]|[mM]|[lL][lL]|[dD]))"],[` ?[^(\\s|[${T}])]+`,` ?[^\\s${T}]+`]]);class x{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 y extends n.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 k(e);case"Unigram":return new F(e,...t);case"BPE":return new v(e);default:if(e.vocab)return Array.isArray(e.vocab)?new F(e,...t):new S(e,...t);throw new Error(`Unknown TokenizerModel type: ${e.type}`)}}_call(e){return e=this.encode(e),this.fuse_unk&&(e=function(e,t,r){const n=[];let o=0;for(;o<e.length;)if(n.push(e[o]),(t.get(e[o])??r)===r)for(;++o<e.length&&(t.get(e[o])??r)===r;)t.get(n.at(-1))!==r&&(n[n.length-1]+=e[o]);else++o;return n}(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 k extends y{constructor(e){super(e),this.tokens_to_ids=h(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 r of e){const e=[...r];if(e.length>this.max_input_chars_per_word){t.push(this.unk_token);continue}let n=!1,o=0;const s=[];for(;o<e.length;){let t=e.length,r=null;for(;o<t;){let n=e.slice(o,t).join("");if(o>0&&(n=this.config.continuing_subword_prefix+n),this.tokens_to_ids.has(n)){r=n;break}--t}if(null===r){n=!0;break}s.push(r),o=t}n?t.push(this.unk_token):t.push(...s)}return t}}class F extends y{constructor(e,t){super(e);const r=e.vocab.length;this.vocab=new Array(r),this.scores=new Array(r);for(let t=0;t<r;++t){const r=e.vocab[t];this.vocab[t]=r[0],this.scores[t]=r[1]}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,a.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 r=0;for(;r<t.length;){let n=!1;const s=[],a=t.slice(r).join(""),i=this.trie.commonPrefixSearch(a);for(const t of i){s.push(t);const a=this.tokens_to_ids.get(t),i=this.scores[a],l=(0,o.len)(t);e.insert(r,l,i,a),n||1!==l||(n=!0)}n||e.insert(r,1,this.unk_score,this.unk_token_id),r+=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 r of e){const e=this.tokenize(r);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 r=0;for(let n=0;n<256;++n)e.includes(n)||(e.push(n),t.push(256+r),r+=1);const n=t.map((e=>String.fromCharCode(e)));return Object.fromEntries(e.map(((e,t)=>[e,n[t]])))})(),C=(0,o.reverseDictionary)(P);class v extends y{constructor(e){super(e),this.tokens_to_ids=h(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 r=Array.from(e);this.end_of_word_suffix&&(r[r.length-1]+=this.end_of_word_suffix);let n=[];if(r.length>1){const e=new l.PriorityQueue(((e,t)=>e.score<t.score));let t={token:r[0],bias:0,prev:null,next:null},o=t;for(let t=1;t<r.length;++t){const n={bias:t/r.length,token:r[t],prev:o,next:null};o.next=n,this._add_node(e,o),o=n}for(;!e.isEmpty();){const r=e.pop();if(r.deleted||!r.next||r.next.deleted)continue;if(r.deleted=!0,r.next.deleted=!0,r.prev){const e={...r.prev};r.prev.deleted=!0,r.prev=e,e.prev?e.prev.next=e:t=e}const n={token:r.token+r.next.token,bias:r.bias,prev:r.prev,next:r.next.next};n.prev?(n.prev.next=n,this._add_node(e,n.prev)):t=n,n.next&&(n.next.prev=n,this._add_node(e,n))}for(let e=t;null!==e;e=e.next)n.push(e.token)}else n=r;if(this.continuing_subword_suffix)for(let e=0;e<n.length-1;++e)n[e]+=this.continuing_subword_suffix;return this.cache.set(e,n),n}_add_node(e,t){const r=this.bpe_ranks.get(JSON.stringify([t.token,t.next.token]));void 0!==r&&(t.score=r+t.bias,e.push(t))}encode(e){const t=[];for(const r of e){if(this.ignore_merges&&this.tokens_to_ids.has(r)){t.push(r);continue}const e=this.bpe(r);for(const r of e)if(this.tokens_to_ids.has(r))t.push(r);else if(this.byte_fallback){const e=Array.from(this.text_encoder.encode(r)).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 S extends y{constructor(e,t){super(e),this.tokens_to_ids=h(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 A extends n.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"BertNormalizer":return new j(e);case"Precompiled":return new pe(e);case"Sequence":return new V(e);case"Replace":return new E(e);case"NFC":return new L(e);case"NFKC":return new z(e);case"NFKD":return new I(e);case"Strip":return new B(e);case"StripAccents":return new N(e);case"Lowercase":return new D(e);case"Prepend":return new O(e);default:throw new Error(`Unknown Normalizer type: ${e.type}`)}}normalize(e){throw Error("normalize should be implemented in subclass.")}_call(e){return this.normalize(e)}}class E extends A{normalize(e){const t=p(this.config.pattern);return null===t?e:e.replaceAll(t,this.config.content)}}class L extends A{normalize(e){return e=e.normalize("NFC")}}class z extends A{normalize(e){return e=e.normalize("NFKC")}}class I extends A{normalize(e){return e=e.normalize("NFKD")}}class B extends A{normalize(e){return this.config.strip_left&&this.config.strip_right?e=e.trim():(this.config.strip_left&&(e=e.trimStart()),this.config.strip_right&&(e=e.trimEnd())),e}}class N extends A{normalize(e){return e=f(e)}}class D extends A{normalize(e){return e=e.toLowerCase()}}class O extends A{normalize(e){return e=this.config.prepend+e}}class V extends A{constructor(e){super(e),this.normalizers=e.normalizers.map((e=>A.fromConfig(e)))}normalize(e){return this.normalizers.reduce(((e,t)=>t.normalize(e)),e)}}class j extends A{_tokenize_chinese_chars(e){const t=[];for(let r=0;r<e.length;++r){const n=e[r];g(n.charCodeAt(0))?(t.push(" "),t.push(n),t.push(" ")):t.push(n)}return t.join("")}stripAccents(e){return e.normalize("NFD").replace(/\p{Mn}/gu,"")}_is_control(e){switch(e){case"\t":case"\n":case"\r":return!1;default:return/^\p{Cc}|\p{Cf}|\p{Co}|\p{Cs}$/u.test(e)}}_clean_text(e){const t=[];for(const r of e){const e=r.charCodeAt(0);0===e||65533===e||this._is_control(r)||(/^\s$/.test(r)?t.push(" "):t.push(r))}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 n.Callable{static fromConfig(e){if(null===e)return null;switch(e.type){case"BertPreTokenizer":return new G(e);case"Sequence":return new he(e);case"Whitespace":return new me(e);case"WhitespaceSplit":return new _e(e);case"Metaspace":return new de(e);case"ByteLevel":return new q(e);case"Split":return new W(e);case"Punctuation":return new $(e);case"Digits":return new U(e);case"Replace":return new fe(e);default:throw new Error(`Unknown PreTokenizer type: ${e.type}`)}}pre_tokenize_text(e,t){throw Error("pre_tokenize_text should be implemented in subclass.")}pre_tokenize(e,t){return(Array.isArray(e)?e.map((e=>this.pre_tokenize_text(e,t))):this.pre_tokenize_text(e,t)).flat()}_call(e,t){return this.pre_tokenize(e,t)}}class G extends R{constructor(e){super(),this.pattern=new RegExp(`[^\\s${M}]+|[${M}]`,"gu")}pre_tokenize_text(e,t){return e.trim().match(this.pattern)||[]}}class q 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 W extends R{constructor(e){super(),this.config=e,this.pattern=p(this.config.pattern,this.config.invert)}pre_tokenize_text(e,t){return null===this.pattern?[]:this.config.invert?e.match(this.pattern)||[]:function(e,t){const r=[];let n=0;for(const o of e.matchAll(t)){const t=o[0];n<o.index&&r.push(e.slice(n,o.index)),t.length>0&&r.push(t),n=o.index+t.length}return n<e.length&&r.push(e.slice(n)),r}(e,this.pattern)}}class $ extends R{constructor(e){super(),this.config=e,this.pattern=new RegExp(`[^${M}]+|[${M}]+`,"gu")}pre_tokenize_text(e,t){return e.match(this.pattern)||[]}}class U 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 n.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"TemplateProcessing":return new Y(e);case"ByteLevel":return new J(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:r=!0}={}){r&&(e=(0,o.mergeArrays)([this.cls],e,[this.sep]));let n=new Array(e.length).fill(0);if(null!==t){const s=r&&this instanceof H?[this.sep]:[],a=r?[this.sep]:[];e=(0,o.mergeArrays)(e,s,t,a),n=(0,o.mergeArrays)(n,new Array(t.length+s.length+a.length).fill(1))}return{tokens:e,token_type_ids:n}}}class H extends Q{}class Y extends X{constructor(e){super(e),this.single=e.single,this.pair=e.pair}post_process(e,t=null,{add_special_tokens:r=!0}={}){const n=null===t?this.single:this.pair;let s=[],a=[];for(const i of n)"SpecialToken"in i?r&&(s.push(i.SpecialToken.id),a.push(i.SpecialToken.type_id)):"Sequence"in i&&("A"===i.Sequence.id?(s=(0,o.mergeArrays)(s,e),a=(0,o.mergeArrays)(a,new Array(e.length).fill(i.Sequence.type_id))):"B"===i.Sequence.id&&(s=(0,o.mergeArrays)(s,t),a=(0,o.mergeArrays)(a,new Array(t.length).fill(i.Sequence.type_id))));return{tokens:s,token_type_ids:a}}}class J extends X{post_process(e,t=null){return t&&(e=(0,o.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,r={}){let n;for(const o of this.processors)if(o instanceof J){if(e=o.post_process(e).tokens,t){t=o.post_process(t).tokens}}else{const s=o.post_process(e,t,r);e=s.tokens,n=s.token_type_ids}return{tokens:e,token_type_ids:n}}}class Z extends n.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 oe(e);case"Metaspace":return new ue(e);case"ByteLevel":return new se(e);case"Replace":return new ee(e);case"ByteFallback":return new te(e);case"Fuse":return new re(e);case"Strip":return new ne(e);case"Sequence":return new ie(e);case"CTC":return new ae(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=p(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 r=[];for(const n of e){let e=null;if(6===n.length&&n.startsWith("<0x")&&n.endsWith(">")){const t=parseInt(n.slice(3,5),16);isNaN(t)||(e=t)}if(null!==e)r.push(e);else{if(r.length>0){const e=this.text_decoder.decode(Uint8Array.from(r));t.push(e),r=[]}t.push(n)}}if(r.length>0){const e=this.text_decoder.decode(Uint8Array.from(r));t.push(e),r=[]}return t}}class re extends Z{decode_chain(e){return[e.join("")]}}class ne 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 r=0;r<this.start&&e[r]===this.content;++r)t=r+1;let r=e.length;for(let t=0;t<this.stop;++t){const n=e.length-t-1;if(e[n]!==this.content)break;r=n}return e.slice(t,r)}))}}class oe 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=_(e)),e)))}}class se extends Z{constructor(e){super(e),this.byte_decoder=C,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(""),r=new Uint8Array([...t].map((e=>this.byte_decoder[e])));return this.text_decoder.decode(r)}decode_chain(e){const t=[];let r=[];for(const n of e)void 0!==this.added_tokens.find((e=>e.content===n))?(r.length>0&&(t.push(this.convert_tokens_to_string(r)),r=[]),t.push(n)):r.push(n);return r.length>0&&t.push(this.convert_tokens_to_string(r)),t}}class ae 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 r=1;r<e.length;++r)e[r]!==t.at(-1)&&t.push(e[r]);let r=t.filter((e=>e!==this.pad_token)).join("");return this.cleanup&&(r=_(r).replaceAll(this.word_delimiter_token," ").trim()),r}decode_chain(e){return[this.convert_tokens_to_string(e)]}}class ie 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,r)=>t.replaceAll(this.suffix,r===e.length-1?"":" ")))}}class ce extends Z{decode_chain(e){let t="";for(let r=1;r<e.length;r+=2)t+=e[r];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 r=e.replaceAll(" ",this.strRep);return this.addPrefixSpace&&!r.startsWith(this.replacement)&&("always"===this.prepend_scheme||"first"===this.prepend_scheme&&0===t)&&(r=this.strRep+r),[r]}}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 r=0;r<e.length;++r){let n=e[r].replaceAll(this.replacement," ");this.addPrefixSpace&&0==r&&n.startsWith(" ")&&(n=n.substring(1)),t.push(n)}return t}}class pe extends A{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 he extends R{constructor(e){super(),this.tokenizers=e.pretokenizers.map((e=>R.fromConfig(e)))}pre_tokenize_text(e,t){return this.tokenizers.reduce(((e,r)=>r.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 _e 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=p(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 Me(e,t,r,n){for(const s of Object.keys(e)){const a=t-e[s].length,i=r(s),l=new Array(a).fill(i);e[s]="right"===n?(0,o.mergeArrays)(e[s],l):(0,o.mergeArrays)(l,e[s])}}function we(e,t){for(const r of Object.keys(e))e[r].length=t}class Te extends n.Callable{return_token_type_ids=!1;padding_side="right";constructor(e,t){super(),this._tokenizer_config=t,this.normalizer=A.fromConfig(e.normalizer),this.pre_tokenizer=R.fromConfig(e.pre_tokenizer),this.model=y.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 x(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*":""}(${(0,o.escapeRegExp)(e.content)})${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.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:r}of this.chat_template){if("string"!=typeof t||"string"!=typeof r)throw new Error('Chat template must be a list of objects with "name" and "template" properties');e[t]=r}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:r=null,cache_dir:n=null,local_files_only:o=!1,revision:s="main",legacy:a=null}={}){return new this(...await u(e,{progress_callback:t,config:r,cache_dir:n,local_files_only:o,revision:s,legacy:a}))}_call(e,{text_pair:t=null,add_special_tokens:r=!0,padding:n=!1,truncation:o=null,max_length:s=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,n)=>this._encode_plus(e,{text_pair:t[n],add_special_tokens:r,return_token_type_ids:c})))}else u=e.map((e=>this._encode_plus(e,{add_special_tokens:r,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:r,return_token_type_ids:c})]}if(null===s?s="max_length"===n?this.model_max_length:(0,a.max)(u.map((e=>e.input_ids.length)))[0]:o||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."),s=Math.min(s,this.model_max_length??1/0),n||o)for(let e=0;e<u.length;++e)u[e].input_ids.length!==s&&(u[e].input_ids.length>s?o&&we(u[e],s):n&&Me(u[e],s,(e=>"input_ids"===e?this.pad_token_id:0),this.padding_side));const p={};if(l){if((!n||!o)&&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]))p[t]=new i.Tensor("int64",BigInt64Array.from(u.flatMap((e=>e[t])).map(BigInt)),e)}else{for(const e of Object.keys(u[0]))p[e]=u.map((t=>t[e]));if(!d)for(const e of Object.keys(p))p[e]=p[e][0]}return p}_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 r=null!==this.pre_tokenizer?this.pre_tokenizer(e,{section_index:t}):[e];return this.model(r)}})).flat();return t}_encode_plus(e,{text_pair:t=null,add_special_tokens:r=!0,return_token_type_ids:n=null}={}){const{tokens:o,token_type_ids:s}=this._tokenize_helper(e,{pair:t,add_special_tokens:r}),a=this.model.convert_tokens_to_ids(o),i={input_ids:a,attention_mask:new Array(a.length).fill(1)};return(n??this.return_token_type_ids)&&s&&(i.token_type_ids=s),i}_tokenize_helper(e,{pair:t=null,add_special_tokens:r=!1}={}){const n=this._encode_text(e),s=this._encode_text(t);return this.post_processor?this.post_processor(n,s,{add_special_tokens:r}):{tokens:(0,o.mergeArrays)(n??[],s??[])}}tokenize(e,{pair:t=null,add_special_tokens:r=!1}={}){return this._tokenize_helper(e,{pair:t,add_special_tokens:r}).tokens}encode(e,{text_pair:t=null,add_special_tokens:r=!0,return_token_type_ids:n=null}={}){return this._encode_plus(e,{text_pair:t,add_special_tokens:r,return_token_type_ids:n}).input_ids}batch_decode(e,t={}){return e instanceof i.Tensor&&(e=e.tolist()),e.map((e=>this.decode(e,t)))}decode(e,t={}){if(e instanceof i.Tensor&&(e=m(e)),!Array.isArray(e)||0===e.length||!(0,o.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:r=null}){let n=this.model.convert_ids_to_tokens(e);t&&(n=n.filter((e=>!this.special_tokens.includes(e))));let o=this.decoder?this.decoder(n):n.join(" ");return this.decoder&&this.decoder.end_of_word_suffix&&(o=o.replaceAll(this.decoder.end_of_word_suffix," "),t&&(o=o.trim())),(r??this.clean_up_tokenization_spaces)&&(o=_(o)),o}get_chat_template({chat_template:e=null,tools:t=null}={}){if(this.chat_template&&"object"==typeof this.chat_template){const r=this.chat_template;if(null!==e&&Object.hasOwn(r,e))e=r[e];else if(null===e)if(null!==t&&"tool_use"in r)e=r.tool_use;else{if(!("default"in r))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(r).sort()}.`);e=r.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:r=null,chat_template:n=null,add_generation_prompt:o=!1,tokenize:s=!0,padding:a=!1,truncation:i=!1,max_length:l=null,return_tensor:d=!0,return_dict:u=!1,tokenizer_kwargs:p={},...h}={}){if("string"!=typeof(n=this.get_chat_template({chat_template:n,tools:t})))throw Error("chat_template must be a string, but got "+typeof n);let m=this._compiled_template_cache.get(n);void 0===m&&(m=new c.Template(n),this._compiled_template_cache.set(n,m));const _=Object.create(null);for(const e of ge){const t=this.getToken(e);t&&(_[e]=t)}const f=m.render({messages:e,add_generation_prompt:o,tools:t,documents:r,..._,...h});if(s){const e=this._call(f,{add_special_tokens:!1,padding:a,truncation:i,max_length:l,return_tensor:d,...p});return u?e:e.input_ids}return f}}class be extends Te{return_token_type_ids=!0}class xe extends Te{return_token_type_ids=!0}class ye extends Te{return_token_type_ids=!0}class ke extends Te{return_token_type_ids=!0}class Fe extends Te{return_token_type_ids=!0}class Pe extends Te{return_token_type_ids=!0}class Ce extends Te{return_token_type_ids=!0}class ve extends Te{return_token_type_ids=!0}class Se extends Te{return_token_type_ids=!0}class Ae extends Te{}class Ee extends Te{}class Le extends Te{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 ze extends Te{return_token_type_ids=!0}class Ie extends Te{}class Be extends Te{}class Ne extends Te{}class De extends Te{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,r){return Ke(this,e,t,r)}}class Oe extends De{}class Ve extends Te{}class je extends Te{}const Re="▁";class Ge extends Te{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 qe extends Te{}class We extends Te{}class $e extends Te{}class Ue extends Te{}class Xe extends Te{}class Qe extends Te{}class He extends Te{}class Ye extends Te{}class Je extends Te{}function Ke(e,t,r,n){if(!("language_codes"in e)||!Array.isArray(e.language_codes))throw new Error("Tokenizer must have `language_codes` attribute set and it should be an array of language ids.");if(!("languageRegex"in e&&e.languageRegex instanceof RegExp))throw new Error("Tokenizer must have `languageRegex` attribute set and it should be a regular expression.");if(!("lang_to_token"in e)||"function"!=typeof e.lang_to_token)throw new Error("Tokenizer must have `lang_to_token` attribute set and it should be a function.");const o=n.src_lang,s=n.tgt_lang;if(!e.language_codes.includes(s))throw new Error(`Target language code "${s}" is not valid. Must be one of: {${e.language_codes.join(", ")}}`);if(void 0!==o){if(!e.language_codes.includes(o))throw new Error(`Source language code "${o}" is not valid. Must be one of: {${e.language_codes.join(", ")}}`);for(const t of e.post_processor.config.single)if("SpecialToken"in t&&e.languageRegex.test(t.SpecialToken.id)){t.SpecialToken.id=e.lang_to_token(o);break}}return n.forced_bos_token_id=e.model.convert_tokens_to_ids([e.lang_to_token(s)])[0],e._call(t,r)}class Ze extends Te{constructor(e,t){super(e,t),this.languageRegex=/^[a-z]{3}_[A-Z][a-z]{3}$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))),this.lang_to_token=e=>e}_build_translation_inputs(e,t,r){return Ke(this,e,t,r)}}class et extends Te{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,r){return Ke(this,e,t,r)}}class tt extends Te{get timestamp_begin(){return this.model.convert_tokens_to_ids(["<|notimestamps|>"])[0]+1}_decode_asr(e,{return_timestamps:t=!1,return_language:r=!1,time_precision:n=null,force_full_sequences:o=!0}={}){if(null===n)throw Error("Must specify time_precision");let s=null;const i="word"===t;function l(){return{language:s,timestamp:[null,null],text:""}}const c=[];let u=l(),p=0;const h=this.timestamp_begin;let m=[],_=[],f=!1,g=null;const M=new Set(this.all_special_ids);for(const r of e){const e=r.tokens,o=i?r.token_timestamps:null;let T=null,b=h;if("stride"in r){const[t,o,s]=r.stride;if(p-=o,g=t-s,o&&(b=o/n+h),s)for(let t=e.length-1;t>=0;--t){const r=Number(e[t]);if(r>=h){if(null!==T&&(r-h)*n<g)break;T=r}}}let x=[],y=[];for(let r=0;r<e.length;++r){const g=Number(e[r]);if(M.has(g)){const e=this.decode([g]),r=d.WHISPER_LANGUAGE_MAPPING.get(e.slice(2,-2));if(void 0!==r){if(null!==s&&r!==s&&!t){m.push(x);const e=this.findLongestCommonSequence(m)[0],t=this.decode(e);u.text=t,c.push(u),m=[],x=[],u=l()}s=u.language=r}}else if(g>=h){const e=(g-h)*n+p,t=(0,a.round)(e,2);if(null!==T&&g>=T)f=!0;else if(f||m.length>0&&g<b)f=!1;else if(null===u.timestamp[0])u.timestamp[0]=t;else if(t===u.timestamp[0]);else{u.timestamp[1]=t,m.push(x),i&&_.push(y);const[e,r]=this.findLongestCommonSequence(m,_),n=this.decode(e);u.text=n,i&&(u.words=this.collateWordTimestamps(e,r,s)),c.push(u),m=[],x=[],_=[],y=[],u=l()}}else if(x.push(g),i){let e,t=(0,a.round)(o[r]+p,2);if(r+1<o.length){e=(0,a.round)(o[r+1]+p,2);const s=this.decode([g]);w.test(s)&&(e=(0,a.round)(Math.min(t+n,e),2))}else e=null;y.push([t,e])}}if("stride"in r){const[e,t,n]=r.stride;p+=e-n}x.length>0?(m.push(x),i&&_.push(y)):m.every((e=>0===e.length))&&(u=l(),m=[],x=[],_=[],y=[])}if(m.length>0){if(o&&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,r]=this.findLongestCommonSequence(m,_),n=this.decode(e);u.text=n,i&&(u.words=this.collateWordTimestamps(e,r,s)),c.push(u)}let T=Object.create(null);const b=c.map((e=>e.text)).join("");if(t||r){for(let e=0;e<c.length;++e){const n=c[e];t||delete n.timestamp,r||delete n.language}if(i){const e=[];for(const t of c)for(const r of t.words)e.push(r);T={chunks:e}}else T={chunks:c}}return[b,T]}findLongestCommonSequence(e,t=null){let r=e[0],n=r.length,o=[];const s=Array.isArray(t)&&t.length>0;let a=s?[]:null,i=s?t[0]:null;for(let l=1;l<e.length;++l){const c=e[l];let d=0,u=[n,n,0,0];const p=c.length;for(let e=1;e<n+p;++e){const o=Math.max(0,n-e),a=Math.min(n,n+p-e),h=r.slice(o,a),m=Math.max(0,e-n),_=Math.min(p,e),f=c.slice(m,_);if(h.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=s?h.filter(((e,r)=>e===f[r]&&i[o+r]<=t[l][m+r])).length:h.filter(((e,t)=>e===f[t])).length;const M=g/e+e/1e4;g>1&&M>d&&(d=M,u=[o,a,m,_])}const[h,m,_,f]=u,g=Math.floor((m+h)/2),M=Math.floor((f+_)/2);o.push(...r.slice(0,g)),r=c.slice(M),n=r.length,s&&(a.push(...i.slice(0,g)),i=t[l].slice(M))}return o.push(...r),s?(a.push(...i),[o,a]):[o,[]]}collateWordTimestamps(e,t,r){const[n,o,s]=this.combineTokensIntoWords(e,r),a=[];for(let e=0;e<n.length;++e){const r=s[e];a.push({text:n[e],timestamp:[t[r.at(0)][0],t[r.at(-1)][1]]})}return a}combineTokensIntoWords(e,t,r="\"'“¡¿([{-",n="\"'.。,，!！?？:：”)]}、"){let o,s,a;return["chinese","japanese","thai","lao","myanmar"].includes(t=t??"english")?[o,s,a]=this.splitTokensOnUnicode(e):[o,s,a]=this.splitTokensOnSpaces(e),this.mergePunctuations(o,s,a,r,n)}decode(e,t){let r;return t?.decode_with_timestamps?(e instanceof i.Tensor&&(e=m(e)),r=this.decodeWithTimestamps(e,t)):r=super.decode(e,t),r}decodeWithTimestamps(e,t){const r=t?.time_precision??.02,n=Array.from(this.all_special_ids).at(-1)+1;let o=[[]];for(let t of e)if(t=Number(t),t>=n){const e=((t-n)*r).toFixed(2);o.push(`<|${e}|>`),o.push([])}else o[o.length-1].push(t);return o=o.map((e=>"string"==typeof e?e:super.decode(e,t))),o.join("")}splitTokensOnUnicode(e){const t=this.decode(e,{decode_with_timestamps:!0}),r=[],n=[],o=[];let s=[],a=[],i=0;for(let l=0;l<e.length;++l){const c=e[l];s.push(c),a.push(l);const d=this.decode(s,{decode_with_timestamps:!0});d.includes("�")&&"�"!==t[i+d.indexOf("�")]||(r.push(d),n.push(s),o.push(a),s=[],a=[],i+=d.length)}return[r,n,o]}splitTokensOnSpaces(e){const[t,r,n]=this.splitTokensOnUnicode(e),o=[],s=[],a=[],i=new RegExp(`^[${M}]$`,"gu");for(let e=0;e<t.length;++e){const l=t[e],c=r[e],d=n[e],u=c[0]>=this.model.tokens_to_ids.get("<|endoftext|>"),p=l.startsWith(" "),h=l.trim(),m=i.test(h);if(u||p||m||0===o.length)o.push(l),s.push(c),a.push(d);else{const e=o.length-1;o[e]+=l,s[e].push(...c),a[e].push(...d)}}return[o,s,a]}mergePunctuations(e,t,r,n,s){const a=structuredClone(e),i=structuredClone(t),l=structuredClone(r);let c=a.length-2,d=a.length-1;for(;c>=0;)a[c].startsWith(" ")&&n.includes(a[c].trim())?(a[d]=a[c]+a[d],i[d]=(0,o.mergeArrays)(i[c],i[d]),l[d]=(0,o.mergeArrays)(l[c],l[d]),a[c]="",i[c]=[],l[c]=[]):d=c,--c;for(c=0,d=1;d<a.length;)!a[c].endsWith(" ")&&s.includes(a[d])?(a[c]+=a[d],i[c]=(0,o.mergeArrays)(i[c],i[d]),l[c]=(0,o.mergeArrays)(l[c],l[d]),a[d]="",i[d]=[],l[d]=[]):c=d,++d;return[a.filter((e=>e)),i.filter((e=>e.length>0)),l.filter((e=>e.length>0))]}}class rt extends Te{}class nt extends Te{}class ot extends Te{}class st extends Te{constructor(e,t){super(e,t),this.languageRegex=/^(>>\w+<<)\s*/g,this.supported_language_codes=this.model.vocab.filter((e=>this.languageRegex.test(e))),console.warn('WARNING: `MarianTokenizer` is not yet supported by Hugging Face\'s "fast" tokenizers library. Therefore, you may experience slightly inaccurate results.')}_encode_text(e){if(null===e)return null;const[t,...r]=e.trim().split(this.languageRegex);if(0===r.length)return super._encode_text(t);if(2===r.length){const[e,t]=r;return this.supported_language_codes.includes(e)||console.warn(`Unsupported language code "${e}" detected, which may lead to unexpected behavior. Should be one of: ${JSON.stringify(this.supported_language_codes)}`),(0,o.mergeArrays)([e],super._encode_text(t))}}}class at extends Te{}class it extends Te{}class lt extends Te{}class ct extends Te{}class dt extends Te{}class ut extends Te{constructor(e,t){super(e,t),this.decoder=new ce({})}}class pt extends Te{}class ht{static TOKENIZER_CLASS_MAPPING={T5Tokenizer:Ie,DistilBertTokenizer:Ae,CamembertTokenizer:Ee,DebertaTokenizer:Fe,DebertaV2Tokenizer:Pe,BertTokenizer:be,HerbertTokenizer:Ce,ConvBertTokenizer:ve,RoFormerTokenizer:Se,XLMTokenizer:Le,ElectraTokenizer:ze,MobileBertTokenizer:ye,SqueezeBertTokenizer:ke,AlbertTokenizer:xe,GPT2Tokenizer:Be,BartTokenizer:Ne,MBartTokenizer:De,MBart50Tokenizer:Oe,RobertaTokenizer:Ve,WhisperTokenizer:tt,CodeGenTokenizer:rt,CLIPTokenizer:nt,SiglipTokenizer:ot,MarianTokenizer:st,BloomTokenizer:je,NllbTokenizer:Ze,M2M100Tokenizer:et,LlamaTokenizer:Ge,CodeLlamaTokenizer:qe,XLMRobertaTokenizer:We,MPNetTokenizer:$e,FalconTokenizer:Ue,GPTNeoXTokenizer:Xe,EsmTokenizer:Qe,Wav2Vec2CTCTokenizer:at,BlenderbotTokenizer:it,BlenderbotSmallTokenizer:lt,SpeechT5Tokenizer:ct,NougatTokenizer:dt,VitsTokenizer:ut,Qwen2Tokenizer:He,GemmaTokenizer:Ye,Grok1Tokenizer:Je,CohereTokenizer:pt,PreTrainedTokenizer:Te};static async from_pretrained(e,{progress_callback:t=null,config:r=null,cache_dir:n=null,local_files_only:o=!1,revision:s="main",legacy:a=null}={}){const[i,l]=await u(e,{progress_callback:t,config:r,cache_dir:n,local_files_only:o,revision:s,legacy:a}),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=Te),new d(i,l)}}},"./src/utils/audio.js":
+  \***************************/(e,t,r)=>{r.r(t),r.d(t,{AlbertTokenizer:()=>xe,AutoTokenizer:()=>ht,BartTokenizer:()=>Ne,BertTokenizer:()=>be,BlenderbotSmallTokenizer:()=>lt,BlenderbotTokenizer:()=>it,BloomTokenizer:()=>je,CLIPTokenizer:()=>nt,CamembertTokenizer:()=>Ee,CodeGenTokenizer:()=>rt,CodeLlamaTokenizer:()=>qe,CohereTokenizer:()=>pt,ConvBertTokenizer:()=>ve,DebertaTokenizer:()=>Fe,DebertaV2Tokenizer:()=>Pe,DistilBertTokenizer:()=>Ae,ElectraTokenizer:()=>ze,EsmTokenizer:()=>Qe,FalconTokenizer:()=>Ue,GPT2Tokenizer:()=>Be,GPTNeoXTokenizer:()=>Xe,GemmaTokenizer:()=>Ye,Grok1Tokenizer:()=>Je,HerbertTokenizer:()=>Ce,LlamaTokenizer:()=>Ge,M2M100Tokenizer:()=>et,MBart50Tokenizer:()=>Oe,MBartTokenizer:()=>De,MPNetTokenizer:()=>We,MarianTokenizer:()=>st,MobileBertTokenizer:()=>ye,NllbTokenizer:()=>Ze,NougatTokenizer:()=>dt,PreTrainedTokenizer:()=>Te,Qwen2Tokenizer:()=>He,RoFormerTokenizer:()=>Se,RobertaTokenizer:()=>Ve,SiglipTokenizer:()=>ot,SpeechT5Tokenizer:()=>ct,SqueezeBertTokenizer:()=>ke,T5Tokenizer:()=>Ie,TokenizerModel:()=>y,VitsTokenizer:()=>ut,Wav2Vec2CTCTokenizer:()=>at,WhisperTokenizer:()=>tt,XLMRobertaTokenizer:()=>$e,XLMTokenizer:()=>Le,is_chinese_char:()=>g});var n=r(/*! ./utils/generic.js */"./src/utils/generic.js"),o=r(/*! ./utils/core.js */"./src/utils/core.js"),s=r(/*! ./utils/hub.js */"./src/utils/hub.js"),a=r(/*! ./utils/maths.js */"./src/utils/maths.js"),i=r(/*! ./utils/tensor.js */"./src/utils/tensor.js"),l=r(/*! ./utils/data-structures.js */"./src/utils/data-structures.js"),c=r(/*! @huggingface/jinja */"./node_modules/@huggingface/jinja/dist/index.js"),d=r(/*! ./models/whisper/common_whisper.js */"./src/models/whisper/common_whisper.js");r(/*! ./utils/constants.js */"./src/utils/constants.js");async function u(e,t){const r=await Promise.all([(0,s.getModelJSON)(e,"tokenizer.json",!0,t),(0,s.getModelJSON)(e,"tokenizer_config.json",!0,t)]);return null!==t.legacy&&(r[1].legacy=t.legacy),r}function p(e,t=!0){if(void 0!==e.Regex){let t=e.Regex.replace(/\\([#&~])/g,"$1");for(const[e,r]of b)t=t.replaceAll(e,r);return new RegExp(t,"gu")}if(void 0!==e.String){const r=(0,o.escapeRegExp)(e.String);return new RegExp(t?r:`(${r})`,"gu")}return console.warn("Unknown pattern type:",e),null}function h(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 ${t.length}.`)}}function _(e){return e.replace(/ \./g,".").replace(/ \?/g,"?").replace(/ \!/g,"!").replace(/ ,/g,",").replace(/ \' /g,"'").replace(/ n\'t/g,"n't").replace(/ \'m/g,"'m").replace(/ \'s/g,"'s").replace(/ \'ve/g,"'ve").replace(/ \'re/g,"'re")}function f(e){return e.replace(/\p{M}/gu,"")}function g(e){return e>=19968&&e<=40959||e>=13312&&e<=19903||e>=131072&&e<=173791||e>=173824&&e<=177983||e>=177984&&e<=178207||e>=178208&&e<=183983||e>=63744&&e<=64255||e>=194560&&e<=195103}const M="\\p{P}\\u0021-\\u002F\\u003A-\\u0040\\u005B-\\u0060\\u007B-\\u007E",w=new RegExp(`^[${M}]+$`,"gu"),T=".,!?…。，、।۔،",b=new Map([["(?i:'s|'t|'re|'ve|'m|'ll|'d)","(?:'([sS]|[tT]|[rR][eE]|[vV][eE]|[mM]|[lL][lL]|[dD]))"],[` ?[^(\\s|[${T}])]+`,` ?[^\\s${T}]+`]]);class x{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 y extends n.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 k(e);case"Unigram":return new F(e,...t);case"BPE":return new v(e);default:if(e.vocab)return Array.isArray(e.vocab)?new F(e,...t):new S(e,...t);throw new Error(`Unknown TokenizerModel type: ${e.type}`)}}_call(e){return e=this.encode(e),this.fuse_unk&&(e=function(e,t,r){const n=[];let o=0;for(;o<e.length;)if(n.push(e[o]),(t.get(e[o])??r)===r)for(;++o<e.length&&(t.get(e[o])??r)===r;)t.get(n.at(-1))!==r&&(n[n.length-1]+=e[o]);else++o;return n}(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 k extends y{constructor(e){super(e),this.tokens_to_ids=h(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 r of e){const e=[...r];if(e.length>this.max_input_chars_per_word){t.push(this.unk_token);continue}let n=!1,o=0;const s=[];for(;o<e.length;){let t=e.length,r=null;for(;o<t;){let n=e.slice(o,t).join("");if(o>0&&(n=this.config.continuing_subword_prefix+n),this.tokens_to_ids.has(n)){r=n;break}--t}if(null===r){n=!0;break}s.push(r),o=t}n?t.push(this.unk_token):t.push(...s)}return t}}class F extends y{constructor(e,t){super(e);const r=e.vocab.length;this.vocab=new Array(r),this.scores=new Array(r);for(let t=0;t<r;++t){const r=e.vocab[t];this.vocab[t]=r[0],this.scores[t]=r[1]}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,a.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 r=0;for(;r<t.length;){let n=!1;const s=[],a=t.slice(r).join(""),i=this.trie.commonPrefixSearch(a);for(const t of i){s.push(t);const a=this.tokens_to_ids.get(t),i=this.scores[a],l=(0,o.len)(t);e.insert(r,l,i,a),n||1!==l||(n=!0)}n||e.insert(r,1,this.unk_score,this.unk_token_id),r+=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 r of e){const e=this.tokenize(r);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 r=0;for(let n=0;n<256;++n)e.includes(n)||(e.push(n),t.push(256+r),r+=1);const n=t.map((e=>String.fromCharCode(e)));return Object.fromEntries(e.map(((e,t)=>[e,n[t]])))})(),C=(0,o.reverseDictionary)(P);class v extends y{constructor(e){super(e),this.tokens_to_ids=h(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 r=Array.from(e);this.end_of_word_suffix&&(r[r.length-1]+=this.end_of_word_suffix);let n=[];if(r.length>1){const e=new l.PriorityQueue(((e,t)=>e.score<t.score));let t={token:r[0],bias:0,prev:null,next:null},o=t;for(let t=1;t<r.length;++t){const n={bias:t/r.length,token:r[t],prev:o,next:null};o.next=n,this._add_node(e,o),o=n}for(;!e.isEmpty();){const r=e.pop();if(r.deleted||!r.next||r.next.deleted)continue;if(r.deleted=!0,r.next.deleted=!0,r.prev){const e={...r.prev};r.prev.deleted=!0,r.prev=e,e.prev?e.prev.next=e:t=e}const n={token:r.token+r.next.token,bias:r.bias,prev:r.prev,next:r.next.next};n.prev?(n.prev.next=n,this._add_node(e,n.prev)):t=n,n.next&&(n.next.prev=n,this._add_node(e,n))}for(let e=t;null!==e;e=e.next)n.push(e.token)}else n=r;if(this.continuing_subword_suffix)for(let e=0;e<n.length-1;++e)n[e]+=this.continuing_subword_suffix;return this.cache.set(e,n),n}_add_node(e,t){const r=this.bpe_ranks.get(JSON.stringify([t.token,t.next.token]));void 0!==r&&(t.score=r+t.bias,e.push(t))}encode(e){const t=[];for(const r of e){if(this.ignore_merges&&this.tokens_to_ids.has(r)){t.push(r);continue}const e=this.bpe(r);for(const r of e)if(this.tokens_to_ids.has(r))t.push(r);else if(this.byte_fallback){const e=Array.from(this.text_encoder.encode(r)).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 S extends y{constructor(e,t){super(e),this.tokens_to_ids=h(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 A extends n.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"BertNormalizer":return new j(e);case"Precompiled":return new pe(e);case"Sequence":return new V(e);case"Replace":return new E(e);case"NFC":return new L(e);case"NFKC":return new z(e);case"NFKD":return new I(e);case"Strip":return new B(e);case"StripAccents":return new N(e);case"Lowercase":return new D(e);case"Prepend":return new O(e);default:throw new Error(`Unknown Normalizer type: ${e.type}`)}}normalize(e){throw Error("normalize should be implemented in subclass.")}_call(e){return this.normalize(e)}}class E extends A{normalize(e){const t=p(this.config.pattern);return null===t?e:e.replaceAll(t,this.config.content)}}class L extends A{normalize(e){return e=e.normalize("NFC")}}class z extends A{normalize(e){return e=e.normalize("NFKC")}}class I extends A{normalize(e){return e=e.normalize("NFKD")}}class B extends A{normalize(e){return this.config.strip_left&&this.config.strip_right?e=e.trim():(this.config.strip_left&&(e=e.trimStart()),this.config.strip_right&&(e=e.trimEnd())),e}}class N extends A{normalize(e){return e=f(e)}}class D extends A{normalize(e){return e=e.toLowerCase()}}class O extends A{normalize(e){return e=this.config.prepend+e}}class V extends A{constructor(e){super(e),this.normalizers=e.normalizers.map((e=>A.fromConfig(e)))}normalize(e){return this.normalizers.reduce(((e,t)=>t.normalize(e)),e)}}class j extends A{_tokenize_chinese_chars(e){const t=[];for(let r=0;r<e.length;++r){const n=e[r];g(n.charCodeAt(0))?(t.push(" "),t.push(n),t.push(" ")):t.push(n)}return t.join("")}stripAccents(e){return e.normalize("NFD").replace(/\p{Mn}/gu,"")}_is_control(e){switch(e){case"\t":case"\n":case"\r":return!1;default:return/^\p{Cc}|\p{Cf}|\p{Co}|\p{Cs}$/u.test(e)}}_clean_text(e){const t=[];for(const r of e){const e=r.charCodeAt(0);0===e||65533===e||this._is_control(r)||(/^\s$/.test(r)?t.push(" "):t.push(r))}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 n.Callable{static fromConfig(e){if(null===e)return null;switch(e.type){case"BertPreTokenizer":return new G(e);case"Sequence":return new he(e);case"Whitespace":return new me(e);case"WhitespaceSplit":return new _e(e);case"Metaspace":return new de(e);case"ByteLevel":return new q(e);case"Split":return new $(e);case"Punctuation":return new W(e);case"Digits":return new U(e);case"Replace":return new fe(e);default:throw new Error(`Unknown PreTokenizer type: ${e.type}`)}}pre_tokenize_text(e,t){throw Error("pre_tokenize_text should be implemented in subclass.")}pre_tokenize(e,t){return(Array.isArray(e)?e.map((e=>this.pre_tokenize_text(e,t))):this.pre_tokenize_text(e,t)).flat()}_call(e,t){return this.pre_tokenize(e,t)}}class G extends R{constructor(e){super(),this.pattern=new RegExp(`[^\\s${M}]+|[${M}]`,"gu")}pre_tokenize_text(e,t){return e.trim().match(this.pattern)||[]}}class q 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 $ extends R{constructor(e){super(),this.config=e,this.pattern=p(this.config.pattern,this.config.invert)}pre_tokenize_text(e,t){return null===this.pattern?[]:this.config.invert?e.match(this.pattern)||[]:function(e,t){const r=[];let n=0;for(const o of e.matchAll(t)){const t=o[0];n<o.index&&r.push(e.slice(n,o.index)),t.length>0&&r.push(t),n=o.index+t.length}return n<e.length&&r.push(e.slice(n)),r}(e,this.pattern)}}class W extends R{constructor(e){super(),this.config=e,this.pattern=new RegExp(`[^${M}]+|[${M}]+`,"gu")}pre_tokenize_text(e,t){return e.match(this.pattern)||[]}}class U 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 n.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"TemplateProcessing":return new Y(e);case"ByteLevel":return new J(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:r=!0}={}){r&&(e=(0,o.mergeArrays)([this.cls],e,[this.sep]));let n=new Array(e.length).fill(0);if(null!==t){const s=r&&this instanceof H?[this.sep]:[],a=r?[this.sep]:[];e=(0,o.mergeArrays)(e,s,t,a),n=(0,o.mergeArrays)(n,new Array(t.length+s.length+a.length).fill(1))}return{tokens:e,token_type_ids:n}}}class H extends Q{}class Y extends X{constructor(e){super(e),this.single=e.single,this.pair=e.pair}post_process(e,t=null,{add_special_tokens:r=!0}={}){const n=null===t?this.single:this.pair;let s=[],a=[];for(const i of n)"SpecialToken"in i?r&&(s.push(i.SpecialToken.id),a.push(i.SpecialToken.type_id)):"Sequence"in i&&("A"===i.Sequence.id?(s=(0,o.mergeArrays)(s,e),a=(0,o.mergeArrays)(a,new Array(e.length).fill(i.Sequence.type_id))):"B"===i.Sequence.id&&(s=(0,o.mergeArrays)(s,t),a=(0,o.mergeArrays)(a,new Array(t.length).fill(i.Sequence.type_id))));return{tokens:s,token_type_ids:a}}}class J extends X{post_process(e,t=null){return t&&(e=(0,o.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,r={}){let n;for(const o of this.processors)if(o instanceof J){if(e=o.post_process(e).tokens,t){t=o.post_process(t).tokens}}else{const s=o.post_process(e,t,r);e=s.tokens,n=s.token_type_ids}return{tokens:e,token_type_ids:n}}}class Z extends n.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 oe(e);case"Metaspace":return new ue(e);case"ByteLevel":return new se(e);case"Replace":return new ee(e);case"ByteFallback":return new te(e);case"Fuse":return new re(e);case"Strip":return new ne(e);case"Sequence":return new ie(e);case"CTC":return new ae(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=p(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 r=[];for(const n of e){let e=null;if(6===n.length&&n.startsWith("<0x")&&n.endsWith(">")){const t=parseInt(n.slice(3,5),16);isNaN(t)||(e=t)}if(null!==e)r.push(e);else{if(r.length>0){const e=this.text_decoder.decode(Uint8Array.from(r));t.push(e),r=[]}t.push(n)}}if(r.length>0){const e=this.text_decoder.decode(Uint8Array.from(r));t.push(e),r=[]}return t}}class re extends Z{decode_chain(e){return[e.join("")]}}class ne 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 r=0;r<this.start&&e[r]===this.content;++r)t=r+1;let r=e.length;for(let t=0;t<this.stop;++t){const n=e.length-t-1;if(e[n]!==this.content)break;r=n}return e.slice(t,r)}))}}class oe 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=_(e)),e)))}}class se extends Z{constructor(e){super(e),this.byte_decoder=C,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(""),r=new Uint8Array([...t].map((e=>this.byte_decoder[e])));return this.text_decoder.decode(r)}decode_chain(e){const t=[];let r=[];for(const n of e)void 0!==this.added_tokens.find((e=>e.content===n))?(r.length>0&&(t.push(this.convert_tokens_to_string(r)),r=[]),t.push(n)):r.push(n);return r.length>0&&t.push(this.convert_tokens_to_string(r)),t}}class ae 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 r=1;r<e.length;++r)e[r]!==t.at(-1)&&t.push(e[r]);let r=t.filter((e=>e!==this.pad_token)).join("");return this.cleanup&&(r=_(r).replaceAll(this.word_delimiter_token," ").trim()),r}decode_chain(e){return[this.convert_tokens_to_string(e)]}}class ie 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,r)=>t.replaceAll(this.suffix,r===e.length-1?"":" ")))}}class ce extends Z{decode_chain(e){let t="";for(let r=1;r<e.length;r+=2)t+=e[r];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 r=e.replaceAll(" ",this.strRep);return this.addPrefixSpace&&!r.startsWith(this.replacement)&&("always"===this.prepend_scheme||"first"===this.prepend_scheme&&0===t)&&(r=this.strRep+r),[r]}}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 r=0;r<e.length;++r){let n=e[r].replaceAll(this.replacement," ");this.addPrefixSpace&&0==r&&n.startsWith(" ")&&(n=n.substring(1)),t.push(n)}return t}}class pe extends A{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 he extends R{constructor(e){super(),this.tokenizers=e.pretokenizers.map((e=>R.fromConfig(e)))}pre_tokenize_text(e,t){return this.tokenizers.reduce(((e,r)=>r.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 _e 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=p(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 Me(e,t,r,n){for(const s of Object.keys(e)){const a=t-e[s].length,i=r(s),l=new Array(a).fill(i);e[s]="right"===n?(0,o.mergeArrays)(e[s],l):(0,o.mergeArrays)(l,e[s])}}function we(e,t){for(const r of Object.keys(e))e[r].length=t}class Te extends n.Callable{return_token_type_ids=!1;padding_side="right";constructor(e,t){super(),this._tokenizer_config=t,this.normalizer=A.fromConfig(e.normalizer),this.pre_tokenizer=R.fromConfig(e.pre_tokenizer),this.model=y.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 x(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*":""}(${(0,o.escapeRegExp)(e.content)})${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.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:r}of this.chat_template){if("string"!=typeof t||"string"!=typeof r)throw new Error('Chat template must be a list of objects with "name" and "template" properties');e[t]=r}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:r=null,cache_dir:n=null,local_files_only:o=!1,revision:s="main",legacy:a=null}={}){return new this(...await u(e,{progress_callback:t,config:r,cache_dir:n,local_files_only:o,revision:s,legacy:a}))}_call(e,{text_pair:t=null,add_special_tokens:r=!0,padding:n=!1,truncation:o=null,max_length:s=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,n)=>this._encode_plus(e,{text_pair:t[n],add_special_tokens:r,return_token_type_ids:c})))}else u=e.map((e=>this._encode_plus(e,{add_special_tokens:r,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:r,return_token_type_ids:c})]}if(null===s?s="max_length"===n?this.model_max_length:(0,a.max)(u.map((e=>e.input_ids.length)))[0]:o||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."),s=Math.min(s,this.model_max_length??1/0),n||o)for(let e=0;e<u.length;++e)u[e].input_ids.length!==s&&(u[e].input_ids.length>s?o&&we(u[e],s):n&&Me(u[e],s,(e=>"input_ids"===e?this.pad_token_id:0),this.padding_side));const p={};if(l){if((!n||!o)&&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]))p[t]=new i.Tensor("int64",BigInt64Array.from(u.flatMap((e=>e[t])).map(BigInt)),e)}else{for(const e of Object.keys(u[0]))p[e]=u.map((t=>t[e]));if(!d)for(const e of Object.keys(p))p[e]=p[e][0]}return p}_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 r=null!==this.pre_tokenizer?this.pre_tokenizer(e,{section_index:t}):[e];return this.model(r)}})).flat();return t}_encode_plus(e,{text_pair:t=null,add_special_tokens:r=!0,return_token_type_ids:n=null}={}){const{tokens:o,token_type_ids:s}=this._tokenize_helper(e,{pair:t,add_special_tokens:r}),a=this.model.convert_tokens_to_ids(o),i={input_ids:a,attention_mask:new Array(a.length).fill(1)};return(n??this.return_token_type_ids)&&s&&(i.token_type_ids=s),i}_tokenize_helper(e,{pair:t=null,add_special_tokens:r=!1}={}){const n=this._encode_text(e),s=this._encode_text(t);return this.post_processor?this.post_processor(n,s,{add_special_tokens:r}):{tokens:(0,o.mergeArrays)(n??[],s??[])}}tokenize(e,{pair:t=null,add_special_tokens:r=!1}={}){return this._tokenize_helper(e,{pair:t,add_special_tokens:r}).tokens}encode(e,{text_pair:t=null,add_special_tokens:r=!0,return_token_type_ids:n=null}={}){return this._encode_plus(e,{text_pair:t,add_special_tokens:r,return_token_type_ids:n}).input_ids}batch_decode(e,t={}){return e instanceof i.Tensor&&(e=e.tolist()),e.map((e=>this.decode(e,t)))}decode(e,t={}){if(e instanceof i.Tensor&&(e=m(e)),!Array.isArray(e)||0===e.length||!(0,o.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:r=null}){let n=this.model.convert_ids_to_tokens(e);t&&(n=n.filter((e=>!this.special_tokens.includes(e))));let o=this.decoder?this.decoder(n):n.join(" ");return this.decoder&&this.decoder.end_of_word_suffix&&(o=o.replaceAll(this.decoder.end_of_word_suffix," "),t&&(o=o.trim())),(r??this.clean_up_tokenization_spaces)&&(o=_(o)),o}get_chat_template({chat_template:e=null,tools:t=null}={}){if(this.chat_template&&"object"==typeof this.chat_template){const r=this.chat_template;if(null!==e&&Object.hasOwn(r,e))e=r[e];else if(null===e)if(null!==t&&"tool_use"in r)e=r.tool_use;else{if(!("default"in r))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(r).sort()}.`);e=r.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:r=null,chat_template:n=null,add_generation_prompt:o=!1,tokenize:s=!0,padding:a=!1,truncation:i=!1,max_length:l=null,return_tensor:d=!0,return_dict:u=!1,tokenizer_kwargs:p={},...h}={}){if("string"!=typeof(n=this.get_chat_template({chat_template:n,tools:t})))throw Error("chat_template must be a string, but got "+typeof n);let m=this._compiled_template_cache.get(n);void 0===m&&(m=new c.Template(n),this._compiled_template_cache.set(n,m));const _=Object.create(null);for(const e of ge){const t=this.getToken(e);t&&(_[e]=t)}const f=m.render({messages:e,add_generation_prompt:o,tools:t,documents:r,..._,...h});if(s){const e=this._call(f,{add_special_tokens:!1,padding:a,truncation:i,max_length:l,return_tensor:d,...p});return u?e:e.input_ids}return f}}class be extends Te{return_token_type_ids=!0}class xe extends Te{return_token_type_ids=!0}class ye extends Te{return_token_type_ids=!0}class ke extends Te{return_token_type_ids=!0}class Fe extends Te{return_token_type_ids=!0}class Pe extends Te{return_token_type_ids=!0}class Ce extends Te{return_token_type_ids=!0}class ve extends Te{return_token_type_ids=!0}class Se extends Te{return_token_type_ids=!0}class Ae extends Te{}class Ee extends Te{}class Le extends Te{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 ze extends Te{return_token_type_ids=!0}class Ie extends Te{}class Be extends Te{}class Ne extends Te{}class De extends Te{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,r){return Ke(this,e,t,r)}}class Oe extends De{}class Ve extends Te{}class je extends Te{}const Re="▁";class Ge extends Te{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 qe extends Te{}class $e extends Te{}class We extends Te{}class Ue extends Te{}class Xe extends Te{}class Qe extends Te{}class He extends Te{}class Ye extends Te{}class Je extends Te{}function Ke(e,t,r,n){if(!("language_codes"in e)||!Array.isArray(e.language_codes))throw new Error("Tokenizer must have `language_codes` attribute set and it should be an array of language ids.");if(!("languageRegex"in e&&e.languageRegex instanceof RegExp))throw new Error("Tokenizer must have `languageRegex` attribute set and it should be a regular expression.");if(!("lang_to_token"in e)||"function"!=typeof e.lang_to_token)throw new Error("Tokenizer must have `lang_to_token` attribute set and it should be a function.");const o=n.src_lang,s=n.tgt_lang;if(!e.language_codes.includes(s))throw new Error(`Target language code "${s}" is not valid. Must be one of: {${e.language_codes.join(", ")}}`);if(void 0!==o){if(!e.language_codes.includes(o))throw new Error(`Source language code "${o}" is not valid. Must be one of: {${e.language_codes.join(", ")}}`);for(const t of e.post_processor.config.single)if("SpecialToken"in t&&e.languageRegex.test(t.SpecialToken.id)){t.SpecialToken.id=e.lang_to_token(o);break}}return n.forced_bos_token_id=e.model.convert_tokens_to_ids([e.lang_to_token(s)])[0],e._call(t,r)}class Ze extends Te{constructor(e,t){super(e,t),this.languageRegex=/^[a-z]{3}_[A-Z][a-z]{3}$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))),this.lang_to_token=e=>e}_build_translation_inputs(e,t,r){return Ke(this,e,t,r)}}class et extends Te{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,r){return Ke(this,e,t,r)}}class tt extends Te{get timestamp_begin(){return this.model.convert_tokens_to_ids(["<|notimestamps|>"])[0]+1}_decode_asr(e,{return_timestamps:t=!1,return_language:r=!1,time_precision:n=null,force_full_sequences:o=!0}={}){if(null===n)throw Error("Must specify time_precision");let s=null;const i="word"===t;function l(){return{language:s,timestamp:[null,null],text:""}}const c=[];let u=l(),p=0;const h=this.timestamp_begin;let m=[],_=[],f=!1,g=null;const M=new Set(this.all_special_ids);for(const r of e){const e=r.tokens,o=i?r.token_timestamps:null;let T=null,b=h;if("stride"in r){const[t,o,s]=r.stride;if(p-=o,g=t-s,o&&(b=o/n+h),s)for(let t=e.length-1;t>=0;--t){const r=Number(e[t]);if(r>=h){if(null!==T&&(r-h)*n<g)break;T=r}}}let x=[],y=[];for(let r=0;r<e.length;++r){const g=Number(e[r]);if(M.has(g)){const e=this.decode([g]),r=d.WHISPER_LANGUAGE_MAPPING.get(e.slice(2,-2));if(void 0!==r){if(null!==s&&r!==s&&!t){m.push(x);const e=this.findLongestCommonSequence(m)[0],t=this.decode(e);u.text=t,c.push(u),m=[],x=[],u=l()}s=u.language=r}}else if(g>=h){const e=(g-h)*n+p,t=(0,a.round)(e,2);if(null!==T&&g>=T)f=!0;else if(f||m.length>0&&g<b)f=!1;else if(null===u.timestamp[0])u.timestamp[0]=t;else if(t===u.timestamp[0]);else{u.timestamp[1]=t,m.push(x),i&&_.push(y);const[e,r]=this.findLongestCommonSequence(m,_),n=this.decode(e);u.text=n,i&&(u.words=this.collateWordTimestamps(e,r,s)),c.push(u),m=[],x=[],_=[],y=[],u=l()}}else if(x.push(g),i){let e,t=(0,a.round)(o[r]+p,2);if(r+1<o.length){e=(0,a.round)(o[r+1]+p,2);const s=this.decode([g]);w.test(s)&&(e=(0,a.round)(Math.min(t+n,e),2))}else e=null;y.push([t,e])}}if("stride"in r){const[e,t,n]=r.stride;p+=e-n}x.length>0?(m.push(x),i&&_.push(y)):m.every((e=>0===e.length))&&(u=l(),m=[],x=[],_=[],y=[])}if(m.length>0){if(o&&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,r]=this.findLongestCommonSequence(m,_),n=this.decode(e);u.text=n,i&&(u.words=this.collateWordTimestamps(e,r,s)),c.push(u)}let T=Object.create(null);const b=c.map((e=>e.text)).join("");if(t||r){for(let e=0;e<c.length;++e){const n=c[e];t||delete n.timestamp,r||delete n.language}if(i){const e=[];for(const t of c)for(const r of t.words)e.push(r);T={chunks:e}}else T={chunks:c}}return[b,T]}findLongestCommonSequence(e,t=null){let r=e[0],n=r.length,o=[];const s=Array.isArray(t)&&t.length>0;let a=s?[]:null,i=s?t[0]:null;for(let l=1;l<e.length;++l){const c=e[l];let d=0,u=[n,n,0,0];const p=c.length;for(let e=1;e<n+p;++e){const o=Math.max(0,n-e),a=Math.min(n,n+p-e),h=r.slice(o,a),m=Math.max(0,e-n),_=Math.min(p,e),f=c.slice(m,_);if(h.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=s?h.filter(((e,r)=>e===f[r]&&i[o+r]<=t[l][m+r])).length:h.filter(((e,t)=>e===f[t])).length;const M=g/e+e/1e4;g>1&&M>d&&(d=M,u=[o,a,m,_])}const[h,m,_,f]=u,g=Math.floor((m+h)/2),M=Math.floor((f+_)/2);o.push(...r.slice(0,g)),r=c.slice(M),n=r.length,s&&(a.push(...i.slice(0,g)),i=t[l].slice(M))}return o.push(...r),s?(a.push(...i),[o,a]):[o,[]]}collateWordTimestamps(e,t,r){const[n,o,s]=this.combineTokensIntoWords(e,r),a=[];for(let e=0;e<n.length;++e){const r=s[e];a.push({text:n[e],timestamp:[t[r.at(0)][0],t[r.at(-1)][1]]})}return a}combineTokensIntoWords(e,t,r="\"'“¡¿([{-",n="\"'.。,，!！?？:：”)]}、"){let o,s,a;return["chinese","japanese","thai","lao","myanmar"].includes(t=t??"english")?[o,s,a]=this.splitTokensOnUnicode(e):[o,s,a]=this.splitTokensOnSpaces(e),this.mergePunctuations(o,s,a,r,n)}decode(e,t){let r;return t?.decode_with_timestamps?(e instanceof i.Tensor&&(e=m(e)),r=this.decodeWithTimestamps(e,t)):r=super.decode(e,t),r}decodeWithTimestamps(e,t){const r=t?.time_precision??.02,n=Array.from(this.all_special_ids).at(-1)+1;let o=[[]];for(let t of e)if(t=Number(t),t>=n){const e=((t-n)*r).toFixed(2);o.push(`<|${e}|>`),o.push([])}else o[o.length-1].push(t);return o=o.map((e=>"string"==typeof e?e:super.decode(e,t))),o.join("")}splitTokensOnUnicode(e){const t=this.decode(e,{decode_with_timestamps:!0}),r=[],n=[],o=[];let s=[],a=[],i=0;for(let l=0;l<e.length;++l){const c=e[l];s.push(c),a.push(l);const d=this.decode(s,{decode_with_timestamps:!0});d.includes("�")&&"�"!==t[i+d.indexOf("�")]||(r.push(d),n.push(s),o.push(a),s=[],a=[],i+=d.length)}return[r,n,o]}splitTokensOnSpaces(e){const[t,r,n]=this.splitTokensOnUnicode(e),o=[],s=[],a=[],i=new RegExp(`^[${M}]$`,"gu");for(let e=0;e<t.length;++e){const l=t[e],c=r[e],d=n[e],u=c[0]>=this.model.tokens_to_ids.get("<|endoftext|>"),p=l.startsWith(" "),h=l.trim(),m=i.test(h);if(u||p||m||0===o.length)o.push(l),s.push(c),a.push(d);else{const e=o.length-1;o[e]+=l,s[e].push(...c),a[e].push(...d)}}return[o,s,a]}mergePunctuations(e,t,r,n,s){const a=structuredClone(e),i=structuredClone(t),l=structuredClone(r);let c=a.length-2,d=a.length-1;for(;c>=0;)a[c].startsWith(" ")&&n.includes(a[c].trim())?(a[d]=a[c]+a[d],i[d]=(0,o.mergeArrays)(i[c],i[d]),l[d]=(0,o.mergeArrays)(l[c],l[d]),a[c]="",i[c]=[],l[c]=[]):d=c,--c;for(c=0,d=1;d<a.length;)!a[c].endsWith(" ")&&s.includes(a[d])?(a[c]+=a[d],i[c]=(0,o.mergeArrays)(i[c],i[d]),l[c]=(0,o.mergeArrays)(l[c],l[d]),a[d]="",i[d]=[],l[d]=[]):c=d,++d;return[a.filter((e=>e)),i.filter((e=>e.length>0)),l.filter((e=>e.length>0))]}}class rt extends Te{}class nt extends Te{}class ot extends Te{}class st extends Te{constructor(e,t){super(e,t),this.languageRegex=/^(>>\w+<<)\s*/g,this.supported_language_codes=this.model.vocab.filter((e=>this.languageRegex.test(e))),console.warn('WARNING: `MarianTokenizer` is not yet supported by Hugging Face\'s "fast" tokenizers library. Therefore, you may experience slightly inaccurate results.')}_encode_text(e){if(null===e)return null;const[t,...r]=e.trim().split(this.languageRegex);if(0===r.length)return super._encode_text(t);if(2===r.length){const[e,t]=r;return this.supported_language_codes.includes(e)||console.warn(`Unsupported language code "${e}" detected, which may lead to unexpected behavior. Should be one of: ${JSON.stringify(this.supported_language_codes)}`),(0,o.mergeArrays)([e],super._encode_text(t))}}}class at extends Te{}class it extends Te{}class lt extends Te{}class ct extends Te{}class dt extends Te{}class ut extends Te{constructor(e,t){super(e,t),this.decoder=new ce({})}}class pt extends Te{}class ht{static TOKENIZER_CLASS_MAPPING={T5Tokenizer:Ie,DistilBertTokenizer:Ae,CamembertTokenizer:Ee,DebertaTokenizer:Fe,DebertaV2Tokenizer:Pe,BertTokenizer:be,HerbertTokenizer:Ce,ConvBertTokenizer:ve,RoFormerTokenizer:Se,XLMTokenizer:Le,ElectraTokenizer:ze,MobileBertTokenizer:ye,SqueezeBertTokenizer:ke,AlbertTokenizer:xe,GPT2Tokenizer:Be,BartTokenizer:Ne,MBartTokenizer:De,MBart50Tokenizer:Oe,RobertaTokenizer:Ve,WhisperTokenizer:tt,CodeGenTokenizer:rt,CLIPTokenizer:nt,SiglipTokenizer:ot,MarianTokenizer:st,BloomTokenizer:je,NllbTokenizer:Ze,M2M100Tokenizer:et,LlamaTokenizer:Ge,CodeLlamaTokenizer:qe,XLMRobertaTokenizer:$e,MPNetTokenizer:We,FalconTokenizer:Ue,GPTNeoXTokenizer:Xe,EsmTokenizer:Qe,Wav2Vec2CTCTokenizer:at,BlenderbotTokenizer:it,BlenderbotSmallTokenizer:lt,SpeechT5Tokenizer:ct,NougatTokenizer:dt,VitsTokenizer:ut,Qwen2Tokenizer:He,GemmaTokenizer:Ye,Grok1Tokenizer:Je,CohereTokenizer:pt,PreTrainedTokenizer:Te};static async from_pretrained(e,{progress_callback:t=null,config:r=null,cache_dir:n=null,local_files_only:o=!1,revision:s="main",legacy:a=null}={}){const[i,l]=await u(e,{progress_callback:t,config:r,cache_dir:n,local_files_only:o,revision:s,legacy:a}),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=Te),new d(i,l)}}},"./src/utils/audio.js":
 /*!****************************!*\
   !*** ./src/utils/audio.js ***!
   \****************************/(e,t,r)=>{r.r(t),r.d(t,{hamming:()=>d,hanning:()=>c,mel_filter_bank:()=>_,read_audio:()=>i,spectrogram:()=>g,window_function:()=>M});var n=r(/*! ./hub.js */"./src/utils/hub.js"),o=r(/*! ./maths.js */"./src/utils/maths.js"),s=r(/*! ./core.js */"./src/utils/core.js"),a=r(/*! ./tensor.js */"./src/utils/tensor.js");async function i(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 r=await(await(0,n.getFile)(e)).arrayBuffer(),o=new AudioContext({sampleRate:t});void 0===t&&console.warn(`No sampling rate provided, using default of ${o.sampleRate}Hz.`);const s=await o.decodeAudioData(r);let a;if(2===s.numberOfChannels){const e=Math.sqrt(2),t=s.getChannelData(0),r=s.getChannelData(1);a=new Float32Array(t.length);for(let n=0;n<s.length;++n)a[n]=e*(t[n]+r[n])/2}else a=s.getChannelData(0);return a}function l(e,t){if(e<1)return new Float64Array;if(1===e)return new Float64Array([1]);const r=1-t,n=2*Math.PI/(e-1),o=new Float64Array(e);for(let s=0;s<e;++s)o[s]=t-r*Math.cos(s*n);return o}function c(e){return l(e,.5)}function d(e){return l(e,.54)}const u={htk:e=>2595*Math.log10(1+e/700),kaldi:e=>1127*Math.log(1+e/700),slaney:(e,t=1e3,r=15,n=27/Math.log(6.4))=>e>=t?r+Math.log(e/t)*n:3*e/200};function p(e,t="htk"){const r=u[t];if(!r)throw new Error('mel_scale should be one of "htk", "slaney" or "kaldi".');return"number"==typeof e?r(e):e.map((e=>r(e)))}const h={htk:e=>700*(10**(e/2595)-1),kaldi:e=>700*(Math.exp(e/1127)-1),slaney:(e,t=1e3,r=15,n=Math.log(6.4)/27)=>e>=r?t*Math.exp(n*(e-r)):200*e/3};function m(e,t,r){const n=(t-e)/(r-1);return Float64Array.from({length:r},((t,r)=>e+n*r))}function _(e,t,r,n,o,s=null,a="htk",i=!1){if(null!==s&&"slaney"!==s)throw new Error('norm must be one of null or "slaney"');const l=m(p(r,a),p(n,a),t+2);let c,d=function(e,t="htk"){const r=h[t];if(!r)throw new Error('mel_scale should be one of "htk", "slaney" or "kaldi".');return"number"==typeof e?r(e):e.map((e=>r(e)))}(l,a);if(i){const t=o/(2*e);c=p(Float64Array.from({length:e},((e,r)=>r*t)),a),d=l}else c=m(0,Math.floor(o/2),e);const u=function(e,t){const r=Float64Array.from({length:t.length-1},((e,r)=>t[r+1]-t[r])),n=Array.from({length:e.length},(()=>new Array(t.length)));for(let r=0;r<e.length;++r){const o=n[r];for(let n=0;n<t.length;++n)o[n]=t[n]-e[r]}const o=t.length-2,s=Array.from({length:o},(()=>new Array(e.length)));for(let t=0;t<e.length;++t){const e=n[t];for(let n=0;n<o;++n){const o=-e[n]/r[n],a=e[n+2]/r[n+1];s[n][t]=Math.max(0,Math.min(o,a))}}return s}(c,d);if(null!==s&&"slaney"===s)for(let r=0;r<t;++r){const t=u[r],n=2/(d[r+2]-d[r]);for(let r=0;r<e;++r)t[r]*=n}return u}function f(e,t,r,n,s){if(r<=0)throw new Error("reference must be greater than zero");if(n<=0)throw new Error("min_value must be greater than zero");r=Math.max(n,r);const a=Math.log10(r);for(let r=0;r<e.length;++r)e[r]=t*Math.log10(Math.max(n,e[r])-a);if(null!==s){if(s<=0)throw new Error("db_range must be greater than zero");const t=(0,o.max)(e)[0]-s;for(let r=0;r<e.length;++r)e[r]=Math.max(e[r],t)}return e}async function g(e,t,r,n,{fft_length:i=null,power:l=1,center:c=!0,pad_mode:d="reflect",onesided:u=!0,preemphasis:p=null,mel_filters:h=null,mel_floor:m=1e-10,log_mel:_=null,reference:g=1,min_value:M=1e-10,db_range:w=null,remove_dc_offset:T=null,min_num_frames:b=null,max_num_frames:x=null,do_pad:y=!0,transpose:k=!1}={}){const F=t.length;if(null===i&&(i=r),r>i)throw Error(`frame_length (${r}) may not be larger than fft_length (${i})`);if(F!==r)throw new Error(`Length of the window (${F}) must equal frame_length (${r})`);if(n<=0)throw new Error("hop_length must be greater than zero");if(null===l&&null!==h)throw new Error("You have provided `mel_filters` but `power` is `None`. Mel spectrogram computation is not yet supported for complex-valued spectrogram. Specify `power` to fix this issue.");if(c){if("reflect"!==d)throw new Error(`pad_mode="${d}" not implemented yet.`);const t=Math.floor((i-1)/2)+1;e=function(e,t,r){const n=new e.constructor(e.length+t+r),o=e.length-1;for(let r=0;r<e.length;++r)n[t+r]=e[r];for(let r=1;r<=t;++r)n[t-r]=e[(0,s.calculateReflectOffset)(r,o)];for(let a=1;a<=r;++a)n[o+t+a]=e[(0,s.calculateReflectOffset)(o-a,o)];return n}(e,t,t)}let P=Math.floor(1+Math.floor((e.length-r)/n));null!==b&&P<b&&(P=b);const C=u?Math.floor(i/2)+1:i;let v=P,S=P;null!==x&&(x>P?y&&(S=x):S=v=x);const A=new o.FFT(i),E=new Float64Array(i),L=new Float64Array(A.outputBufferSize),z=new Float32Array(C*S);for(let o=0;o<v;++o){const s=o*n,a=Math.min(e.length-s,r);a!==r&&E.fill(0,0,r);for(let t=0;t<a;++t)E[t]=e[s+t];if(T){let e=0;for(let t=0;t<a;++t)e+=E[t];const t=e/a;for(let e=0;e<a;++e)E[e]-=t}if(null!==p){for(let e=a-1;e>=1;--e)E[e]-=p*E[e-1];E[0]*=1-p}for(let e=0;e<t.length;++e)E[e]*=t[e];A.realTransform(L,E);for(let e=0;e<C;++e){const t=e<<1;z[e*S+o]=L[t]**2+L[t+1]**2}}if(null!==l&&2!==l){const e=2/l;for(let t=0;t<z.length;++t)z[t]**=e}const I=h.length;let B=await(0,a.matmul)(new a.Tensor("float32",h.flat(),[I,C]),new a.Tensor("float32",z,[C,S]));k&&(B=B.transpose(1,0));const N=B.data;for(let e=0;e<N.length;++e)N[e]=Math.max(m,N[e]);if(null!==l&&null!==_){const e=Math.min(N.length,v*I);switch(_){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===l)!function(e,t=1,r=1e-5,n=null){f(e,20,t,r,n)}(N,g,M,w);else{if(2!==l)throw new Error(`Cannot use log_mel option '${_}' with power ${l}`);!function(e,t=1,r=1e-10,n=null){f(e,10,t,r,n)}(N,g,M,w)}break;default:throw new Error(`log_mel must be one of null, 'log', 'log10' or 'dB'. Got '${_}'`)}}return B}function M(e,t,{periodic:r=!0,frame_length:n=null,center:o=!0}={}){const s=r?e+1:e;let a;switch(t){case"boxcar":a=new Float64Array(s).fill(1);break;case"hann":case"hann_window":a=c(s);break;case"hamming":a=d(s);break;case"povey":a=c(s).map((e=>Math.pow(e,.85)));break;default:throw new Error(`Unknown window type ${t}.`)}if(r&&(a=a.subarray(0,e)),null===n)return a;if(e>n)throw new Error(`Length of the window (${e}) may not be larger than frame_length (${n})`);return a}},"./src/utils/constants.js":
@@ -157,11 +157,11 @@ import*as e from"fs";import*as t from"onnxruntime-node";import*as r from"path";i
   \****************************/(e,t,r)=>{r.r(t),r.d(t,{RawImage:()=>m});var n=r(/*! ./hub.js */"./src/utils/hub.js"),o=r(/*! ../env.js */"./src/env.js"),s=r(/*! ./tensor.js */"./src/utils/tensor.js"),a=r(/*! sharp */"sharp");const i="undefined"!=typeof self,l=i&&"DedicatedWorkerGlobalScope"===self.constructor.name;let c,d,u;if(i)c=(e,t)=>{if(!self.OffscreenCanvas)throw new Error("OffscreenCanvas not supported by this browser.");return new self.OffscreenCanvas(e,t)},u=self.createImageBitmap,d=self.ImageData;else{if(!a.default)throw new Error("Unable to load image processing library.");u=async e=>{const t=(await e.metadata()).channels,{data:r,info:n}=await e.rotate().raw().toBuffer({resolveWithObject:!0}),o=new m(new Uint8ClampedArray(r),n.width,n.height,n.channels);return void 0!==t&&t!==n.channels&&o.convert(t),o}}const p={0:"nearest",1:"lanczos",2:"bilinear",3:"bicubic",4:"box",5:"hamming"},h=new Map([["png","image/png"],["jpg","image/jpeg"],["jpeg","image/jpeg"],["gif","image/gif"]]);class m{constructor(e,t,r,n){this.data=e,this.width=t,this.height=r,this.channels=n}get size(){return[this.width,this.height]}static async read(e){if(e instanceof m)return e;if("string"==typeof e||e instanceof URL)return await this.fromURL(e);throw new Error("Unsupported input type: "+typeof e)}static fromCanvas(e){if(!i)throw new Error("fromCanvas() is only supported in browser environments.");const t=e.getContext("2d").getImageData(0,0,e.width,e.height).data;return new m(t,e.width,e.height,4)}static async fromURL(e){const t=await(0,n.getFile)(e);if(200!==t.status)throw new Error(`Unable to read image from "${e}" (${t.status} ${t.statusText})`);const r=await t.blob();return this.fromBlob(r)}static async fromBlob(e){if(i){const t=await u(e),r=c(t.width,t.height).getContext("2d");return r.drawImage(t,0,0),new this(r.getImageData(0,0,t.width,t.height).data,t.width,t.height,4)}{const t=(0,a.default)(await e.arrayBuffer());return await u(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,r=0;t<this.data.length;t+=this.channels){const n=this.data[t],o=this.data[t+1],s=this.data[t+2];e[r++]=Math.round(.2989*n+.587*o+.114*s)}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,r=0;t<this.data.length;++t)e[r++]=this.data[t],e[r++]=this.data[t],e[r++]=this.data[t];break;case 4:for(let t=0,r=0;t<this.data.length;t+=4)e[r++]=this.data[t],e[r++]=this.data[t+1],e[r++]=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,r=0;t<this.data.length;++t)e[r++]=this.data[t],e[r++]=this.data[t],e[r++]=this.data[t],e[r++]=255;break;case 3:for(let t=0,r=0;t<this.data.length;t+=3)e[r++]=this.data[t],e[r++]=this.data[t+1],e[r++]=this.data[t+2],e[r++]=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)}async resize(e,t,{resample:r=2}={}){let n=p[r]??r;if(i){const r=this.channels,n=this.toCanvas(),o=c(e,t).getContext("2d");o.drawImage(n,0,0,e,t);return new m(o.getImageData(0,0,e,t).data,e,t,4).convert(r)}{let r=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":r=r.affine([e/this.width,0,0,t/this.height],{interpolator:n});break;case"lanczos":r=r.resize({width:e,height:t,fit:"fill",kernel:"lanczos3"});break;default:throw new Error(`Resampling method ${n} is not supported.`)}return await u(r)}}async pad([e,t,r,n]){if(e=Math.max(e,0),t=Math.max(t,0),r=Math.max(r,0),n=Math.max(n,0),0===e&&0===t&&0===r&&0===n)return this;if(i){const o=this.channels,s=this.toCanvas(),a=this.width+e+t,i=this.height+r+n,l=c(a,i).getContext("2d");l.drawImage(s,0,0,this.width,this.height,e,r,a,i);return new m(l.getImageData(0,0,a,i).data,a,i,4).convert(o)}{const o=this.toSharp().extend({left:e,right:t,top:r,bottom:n});return await u(o)}}async crop([e,t,r,n]){if(e=Math.max(e,0),t=Math.max(t,0),r=Math.min(r,this.width-1),n=Math.min(n,this.height-1),0===e&&0===t&&r===this.width-1&&n===this.height-1)return this;const o=r-e+1,s=n-t+1;if(i){const r=this.channels,n=this.toCanvas(),a=c(o,s).getContext("2d");a.drawImage(n,e,t,o,s,0,0,o,s);return new m(a.getImageData(0,0,o,s).data,o,s,4).convert(r)}{const r=this.toSharp().extract({left:e,top:t,width:o,height:s});return await u(r)}}async center_crop(e,t){if(this.width===e&&this.height===t)return this;const r=(this.width-e)/2,n=(this.height-t)/2;if(i){const o=this.channels,s=this.toCanvas(),a=c(e,t).getContext("2d");let i=0,l=0,d=0,u=0;r>=0?i=r:d=-r,n>=0?l=n:u=-n,a.drawImage(s,i,l,e,t,d,u,e,t);return new m(a.getImageData(0,0,e,t).data,e,t,4).convert(o)}{let o=this.toSharp();if(r>=0&&n>=0)o=o.extract({left:Math.floor(r),top:Math.floor(n),width:e,height:t});else if(r<=0&&n<=0){const s=Math.floor(-n),a=Math.floor(-r);o=o.extend({top:s,left:a,right:e-this.width-a,bottom:t-this.height-s})}else{let s=[0,0],a=0;n<0?(s[0]=Math.floor(-n),s[1]=t-this.height-s[0]):a=Math.floor(n);let i=[0,0],l=0;r<0?(i[0]=Math.floor(-r),i[1]=e-this.width-i[0]):l=Math.floor(r),o=o.extend({top:s[0],bottom:s[1],left:i[0],right:i[1]}).extract({left:l,top:a,width:e,height:t})}return await u(o)}}async toBlob(e="image/png",t=1){if(!i)throw new Error("toBlob() is only supported in browser environments.");const r=this.toCanvas();return await r.convertToBlob({type:e,quality:t})}toTensor(e="CHW"){let t=new s.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(!i)throw new Error("toCanvas() is only supported in browser environments.");const e=this.clone().rgba(),t=c(e.width,e.height),r=new d(e.data,e.width,e.height);return t.getContext("2d").putImageData(r,0,0),t}_update(e,t,r,n=null){return this.data=e,this.width=t,this.height=r,null!==n&&(this.channels=n),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(!i){if(o.env.useFS){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(l)throw new Error("Unable to save an image from a Web Worker.");const t=e.split(".").pop().toLowerCase(),r=h.get(t)??"image/png",n=await this.toBlob(r),o=URL.createObjectURL(n),s=document.createElement("a");s.href=o,s.download=e,s.click(),s.remove()}}toSharp(){if(i)throw new Error("toSharp() is only supported in server-side environments.");return(0,a.default)(this.data,{raw:{width:this.width,height:this.height,channels:this.channels}})}}},"./src/utils/maths.js":
 /*!****************************!*\
   !*** ./src/utils/maths.js ***!
-  \****************************/(e,t,r)=>{function n(e,[t,r,n],[o,s],a="bilinear",i=!1){const l=s/n,c=o/r,d=new e.constructor(o*s*t),u=r*n,p=o*s;for(let a=0;a<o;++a)for(let o=0;o<s;++o){const i=a*s+o,h=(o+.5)/l-.5,m=(a+.5)/c-.5;let _=Math.floor(h),f=Math.floor(m);const g=Math.min(_+1,n-1),M=Math.min(f+1,r-1);_=Math.max(_,0),f=Math.max(f,0);const w=h-_,T=m-f,b=(1-w)*(1-T),x=w*(1-T),y=(1-w)*T,k=w*T,F=f*n,P=M*n,C=F+_,v=F+g,S=P+_,A=P+g;for(let r=0;r<t;++r){const t=r*u;d[r*p+i]=b*e[t+C]+x*e[t+v]+y*e[t+S]+k*e[t+A]}}return d}function o(e,t,r){const n=new Array(r.length),o=new Array(r.length);for(let e=r.length-1,s=1;e>=0;--e)o[e]=s,n[e]=t[r[e]],s*=n[e];const s=r.map(((e,t)=>o[r.indexOf(t)])),a=new e.constructor(e.length);for(let r=0;r<e.length;++r){let n=0;for(let e=t.length-1,o=r;e>=0;--e)n+=o%t[e]*s[e],o=Math.floor(o/t[e]);a[n]=e[r]}return[a,n]}function s(e){const t=u(e)[0],r=e.map((e=>Math.exp(e-t))),n=r.reduce(((e,t)=>e+t),0);return r.map((e=>e/n))}function a(e){const t=u(e)[0];let r=0;for(let n=0;n<e.length;++n)r+=Math.exp(e[n]-t);const n=Math.log(r);return e.map((e=>e-t-n))}function i(e,t){let r=0;for(let n=0;n<e.length;++n)r+=e[n]*t[n];return r}function l(e,t){return i(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],r=0;for(let n=1;n<e.length;++n)e[n]<t&&(t=e[n],r=n);return[t,r]}function u(e){if(0===e.length)throw Error("Array must not be empty");let t=e[0],r=0;for(let n=1;n<e.length;++n)e[n]>t&&(t=e[n],r=n);return[Number(t),r]}function p(e){return e>0&&!(e&e-1)}r.r(t),r.d(t,{FFT:()=>_,bankers_round:()=>M,cos_sim:()=>l,dot:()=>i,dynamic_time_warping:()=>w,interpolate_data:()=>n,log_softmax:()=>a,magnitude:()=>c,max:()=>u,medianFilter:()=>f,min:()=>d,permute_data:()=>o,round:()=>g,softmax:()=>s});class h{constructor(e){if(this.size=0|e,this.size<=1||!p(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 r=this._width-t-2;this._bitrev[e]|=(e>>>t&3)<<r}}}createComplexArray(){return new Float64Array(this._csize)}fromComplexArray(e,t){const r=t||new Array(e.length>>>1);for(let t=0;t<e.length;t+=2)r[t>>>1]=e[t];return r}toComplexArray(e,t){const r=t||this.createComplexArray();for(let t=0;t<r.length;t+=2)r[t]=e[t>>>1],r[t+1]=0;return r}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,r){const n=this._csize;let o,s,a=1<<this._width,i=n/a<<1;const l=this._bitrev;if(4===i)for(o=0,s=0;o<n;o+=i,++s){const r=l[s];this._singleTransform2(t,e,o,r,a)}else for(o=0,s=0;o<n;o+=i,++s){const n=l[s];this._singleTransform4(t,e,o,n,a,r)}const c=this.table;for(a>>=2;a>=2;a>>=2){i=n/a<<1;const t=i>>>2;for(o=0;o<n;o+=i){const n=o+t-1;for(let s=o,i=0;s<n;s+=2,i+=a){const n=s,o=n+t,a=o+t,l=a+t,d=e[n],u=e[n+1],p=e[o],h=e[o+1],m=e[a],_=e[a+1],f=e[l],g=e[l+1],M=c[i],w=r*c[i+1],T=p*M-h*w,b=p*w+h*M,x=c[2*i],y=r*c[2*i+1],k=m*x-_*y,F=m*y+_*x,P=c[3*i],C=r*c[3*i+1],v=f*P-g*C,S=f*C+g*P,A=d+k,E=u+F,L=d-k,z=u-F,I=T+v,B=b+S,N=r*(T-v),D=r*(b-S);e[n]=A+I,e[n+1]=E+B,e[o]=L+D,e[o+1]=z-N,e[a]=A-I,e[a+1]=E-B,e[l]=L-D,e[l+1]=z+N}}}}_singleTransform2(e,t,r,n,o){const s=e[n],a=e[n+1],i=e[n+o],l=e[n+o+1];t[r]=s+i,t[r+1]=a+l,t[r+2]=s-i,t[r+3]=a-l}_singleTransform4(e,t,r,n,o,s){const a=2*o,i=3*o,l=e[n],c=e[n+1],d=e[n+o],u=e[n+o+1],p=e[n+a],h=e[n+a+1],m=e[n+i],_=e[n+i+1],f=l+p,g=c+h,M=l-p,w=c-h,T=d+m,b=u+_,x=s*(d-m),y=s*(u-_);t[r]=f+T,t[r+1]=g+b,t[r+2]=M+y,t[r+3]=w-x,t[r+4]=f-T,t[r+5]=g-b,t[r+6]=M-y,t[r+7]=w+x}_realTransform4(e,t,r){const n=this._csize;let o,s,a=1<<this._width,i=n/a<<1;const l=this._bitrev;if(4===i)for(o=0,s=0;o<n;o+=i,++s){const r=l[s];this._singleRealTransform2(t,e,o,r>>>1,a>>>1)}else for(o=0,s=0;o<n;o+=i,++s){const n=l[s];this._singleRealTransform4(t,e,o,n>>>1,a>>>1,r)}const c=this.table;for(a>>=2;a>=2;a>>=2){i=n/a<<1;const t=i>>>1,s=t>>>1,l=s>>>1;for(o=0;o<n;o+=i)for(let n=0,i=0;n<=l;n+=2,i+=a){const a=o+n,d=a+s,u=d+s,p=u+s,h=e[a],m=e[a+1],_=e[d],f=e[d+1],g=e[u],M=e[u+1],w=e[p],T=e[p+1],b=h,x=m,y=c[i],k=r*c[i+1],F=_*y-f*k,P=_*k+f*y,C=c[2*i],v=r*c[2*i+1],S=g*C-M*v,A=g*v+M*C,E=c[3*i],L=r*c[3*i+1],z=w*E-T*L,I=w*L+T*E,B=b+S,N=x+A,D=b-S,O=x-A,V=F+z,j=P+I,R=r*(F-z),G=r*(P-I);if(e[a]=B+V,e[a+1]=N+j,e[d]=D+G,e[d+1]=O-R,0===n){e[u]=B-V,e[u+1]=N-j;continue}if(n===l)continue;const q=o+s-n,W=o+t-n;e[q]=D-r*G,e[q+1]=-O-r*R,e[W]=B-r*V,e[W+1]=r*j-N}}const d=n>>>1;for(let t=2;t<d;t+=2)e[n-t]=e[t],e[n-t+1]=-e[t+1]}_singleRealTransform2(e,t,r,n,o){const s=e[n],a=e[n+o];t[r]=s+a,t[r+1]=0,t[r+2]=s-a,t[r+3]=0}_singleRealTransform4(e,t,r,n,o,s){const a=2*o,i=3*o,l=e[n],c=e[n+o],d=e[n+a],u=e[n+i],p=l+d,h=l-d,m=c+u,_=s*(c-u);t[r]=p+m,t[r+1]=0,t[r+2]=h,t[r+3]=-_,t[r+4]=p-m,t[r+5]=0,t[r+6]=h,t[r+7]=_}}class m{constructor(e){const t=2*(e-1),r=2*(2*e-1),n=2**Math.ceil(Math.log2(r));this.bufferSize=n,this._a=t;const o=new Float64Array(r),s=new Float64Array(n);this._chirpBuffer=new Float64Array(n),this._buffer1=new Float64Array(n),this._buffer2=new Float64Array(n),this._outBuffer1=new Float64Array(n),this._outBuffer2=new Float64Array(n);const a=-2*Math.PI/e,i=Math.cos(a),l=Math.sin(a);for(let t=0;t<r>>1;++t){const r=(t+1-e)**2/2,n=Math.sqrt(i**2+l**2)**r,a=r*Math.atan2(l,i),c=2*t;o[c]=n*Math.cos(a),o[c+1]=n*Math.sin(a),s[c]=o[c],s[c+1]=-o[c+1]}this._slicedChirpBuffer=o.subarray(t,r),this._f=new h(n>>1),this._f.transform(this._chirpBuffer,s)}_transform(e,t,r){const n=this._buffer1,o=this._buffer2,s=this._outBuffer1,a=this._outBuffer2,i=this._chirpBuffer,l=this._slicedChirpBuffer,c=this._a;if(r)for(let e=0;e<l.length;e+=2){const r=e+1,o=t[e>>1];n[e]=o*l[e],n[r]=o*l[r]}else for(let e=0;e<l.length;e+=2){const r=e+1;n[e]=t[e]*l[e]-t[r]*l[r],n[r]=t[e]*l[r]+t[r]*l[e]}this._f.transform(s,n);for(let e=0;e<i.length;e+=2){const t=e+1;o[e]=s[e]*i[e]-s[t]*i[t],o[t]=s[e]*i[t]+s[t]*i[e]}this._f.inverseTransform(a,o);for(let t=0;t<a.length;t+=2){const r=a[t+c],n=a[t+c+1],o=l[t],s=l[t+1];e[t]=r*o-n*s,e[t+1]=r*s+n*o}}transform(e,t){this._transform(e,t,!1)}realTransform(e,t){this._transform(e,t,!0)}}class _{constructor(e){this.fft_length=e,this.isPowerOfTwo=p(e),this.isPowerOfTwo?(this.fft=new h(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 r=new e.constructor(e.length),n=new e.constructor(t),o=Math.floor(t/2);for(let t=0;t<e.length;++t){let s=0;for(let r=-o;r<=o;++r){let o=t+r;o<0?o=Math.abs(o):o>=e.length&&(o=2*(e.length-1)-o),n[s++]=e[o]}n.sort(),r[t]=n[o]}return r}function g(e,t){const r=Math.pow(10,t);return Math.round(e*r)/r}function M(e){const t=Math.round(e);return Math.abs(e)%1==.5?t%2==0?t:t-1:t}function w(e){const t=e.length,r=e[0].length,n=[t+1,r+1],o=Array.from({length:n[0]},(()=>Array(n[1]).fill(1/0)));o[0][0]=0;const s=Array.from({length:n[0]},(()=>Array(n[1]).fill(-1)));for(let t=1;t<n[1];++t)for(let r=1;r<n[0];++r){const n=o[r-1][t-1],a=o[r-1][t],i=o[r][t-1];let l,c;n<a&&n<i?(l=n,c=0):a<n&&a<i?(l=a,c=1):(l=i,c=2),o[r][t]=e[r-1][t-1]+l,s[r][t]=c}for(let e=0;e<n[1];++e)s[0][e]=2;for(let e=0;e<n[0];++e)s[e][0]=1;let a=t,i=r,l=[],c=[];for(;a>0||i>0;)switch(l.push(a-1),c.push(i-1),s[a][i]){case 0:--a,--i;break;case 1:--a;break;case 2:--i;break;default:throw new Error(`Internal error in dynamic time warping. Unexpected trace[${a}, ${i}]. Please file a bug report.`)}return l.reverse(),c.reverse(),[l,c]}},"./src/utils/tensor.js":
+  \****************************/(e,t,r)=>{function n(e,[t,r,n],[o,s],a="bilinear",i=!1){const l=s/n,c=o/r,d=new e.constructor(o*s*t),u=r*n,p=o*s;for(let a=0;a<o;++a)for(let o=0;o<s;++o){const i=a*s+o,h=(o+.5)/l-.5,m=(a+.5)/c-.5;let _=Math.floor(h),f=Math.floor(m);const g=Math.min(_+1,n-1),M=Math.min(f+1,r-1);_=Math.max(_,0),f=Math.max(f,0);const w=h-_,T=m-f,b=(1-w)*(1-T),x=w*(1-T),y=(1-w)*T,k=w*T,F=f*n,P=M*n,C=F+_,v=F+g,S=P+_,A=P+g;for(let r=0;r<t;++r){const t=r*u;d[r*p+i]=b*e[t+C]+x*e[t+v]+y*e[t+S]+k*e[t+A]}}return d}function o(e,t,r){const n=new Array(r.length),o=new Array(r.length);for(let e=r.length-1,s=1;e>=0;--e)o[e]=s,n[e]=t[r[e]],s*=n[e];const s=r.map(((e,t)=>o[r.indexOf(t)])),a=new e.constructor(e.length);for(let r=0;r<e.length;++r){let n=0;for(let e=t.length-1,o=r;e>=0;--e)n+=o%t[e]*s[e],o=Math.floor(o/t[e]);a[n]=e[r]}return[a,n]}function s(e){const t=u(e)[0],r=e.map((e=>Math.exp(e-t))),n=r.reduce(((e,t)=>e+t),0);return r.map((e=>e/n))}function a(e){const t=u(e)[0];let r=0;for(let n=0;n<e.length;++n)r+=Math.exp(e[n]-t);const n=Math.log(r);return e.map((e=>e-t-n))}function i(e,t){let r=0;for(let n=0;n<e.length;++n)r+=e[n]*t[n];return r}function l(e,t){return i(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],r=0;for(let n=1;n<e.length;++n)e[n]<t&&(t=e[n],r=n);return[t,r]}function u(e){if(0===e.length)throw Error("Array must not be empty");let t=e[0],r=0;for(let n=1;n<e.length;++n)e[n]>t&&(t=e[n],r=n);return[Number(t),r]}function p(e){return e>0&&!(e&e-1)}r.r(t),r.d(t,{FFT:()=>_,bankers_round:()=>M,cos_sim:()=>l,dot:()=>i,dynamic_time_warping:()=>w,interpolate_data:()=>n,log_softmax:()=>a,magnitude:()=>c,max:()=>u,medianFilter:()=>f,min:()=>d,permute_data:()=>o,round:()=>g,softmax:()=>s});class h{constructor(e){if(this.size=0|e,this.size<=1||!p(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 r=this._width-t-2;this._bitrev[e]|=(e>>>t&3)<<r}}}createComplexArray(){return new Float64Array(this._csize)}fromComplexArray(e,t){const r=t||new Array(e.length>>>1);for(let t=0;t<e.length;t+=2)r[t>>>1]=e[t];return r}toComplexArray(e,t){const r=t||this.createComplexArray();for(let t=0;t<r.length;t+=2)r[t]=e[t>>>1],r[t+1]=0;return r}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,r){const n=this._csize;let o,s,a=1<<this._width,i=n/a<<1;const l=this._bitrev;if(4===i)for(o=0,s=0;o<n;o+=i,++s){const r=l[s];this._singleTransform2(t,e,o,r,a)}else for(o=0,s=0;o<n;o+=i,++s){const n=l[s];this._singleTransform4(t,e,o,n,a,r)}const c=this.table;for(a>>=2;a>=2;a>>=2){i=n/a<<1;const t=i>>>2;for(o=0;o<n;o+=i){const n=o+t-1;for(let s=o,i=0;s<n;s+=2,i+=a){const n=s,o=n+t,a=o+t,l=a+t,d=e[n],u=e[n+1],p=e[o],h=e[o+1],m=e[a],_=e[a+1],f=e[l],g=e[l+1],M=c[i],w=r*c[i+1],T=p*M-h*w,b=p*w+h*M,x=c[2*i],y=r*c[2*i+1],k=m*x-_*y,F=m*y+_*x,P=c[3*i],C=r*c[3*i+1],v=f*P-g*C,S=f*C+g*P,A=d+k,E=u+F,L=d-k,z=u-F,I=T+v,B=b+S,N=r*(T-v),D=r*(b-S);e[n]=A+I,e[n+1]=E+B,e[o]=L+D,e[o+1]=z-N,e[a]=A-I,e[a+1]=E-B,e[l]=L-D,e[l+1]=z+N}}}}_singleTransform2(e,t,r,n,o){const s=e[n],a=e[n+1],i=e[n+o],l=e[n+o+1];t[r]=s+i,t[r+1]=a+l,t[r+2]=s-i,t[r+3]=a-l}_singleTransform4(e,t,r,n,o,s){const a=2*o,i=3*o,l=e[n],c=e[n+1],d=e[n+o],u=e[n+o+1],p=e[n+a],h=e[n+a+1],m=e[n+i],_=e[n+i+1],f=l+p,g=c+h,M=l-p,w=c-h,T=d+m,b=u+_,x=s*(d-m),y=s*(u-_);t[r]=f+T,t[r+1]=g+b,t[r+2]=M+y,t[r+3]=w-x,t[r+4]=f-T,t[r+5]=g-b,t[r+6]=M-y,t[r+7]=w+x}_realTransform4(e,t,r){const n=this._csize;let o,s,a=1<<this._width,i=n/a<<1;const l=this._bitrev;if(4===i)for(o=0,s=0;o<n;o+=i,++s){const r=l[s];this._singleRealTransform2(t,e,o,r>>>1,a>>>1)}else for(o=0,s=0;o<n;o+=i,++s){const n=l[s];this._singleRealTransform4(t,e,o,n>>>1,a>>>1,r)}const c=this.table;for(a>>=2;a>=2;a>>=2){i=n/a<<1;const t=i>>>1,s=t>>>1,l=s>>>1;for(o=0;o<n;o+=i)for(let n=0,i=0;n<=l;n+=2,i+=a){const a=o+n,d=a+s,u=d+s,p=u+s,h=e[a],m=e[a+1],_=e[d],f=e[d+1],g=e[u],M=e[u+1],w=e[p],T=e[p+1],b=h,x=m,y=c[i],k=r*c[i+1],F=_*y-f*k,P=_*k+f*y,C=c[2*i],v=r*c[2*i+1],S=g*C-M*v,A=g*v+M*C,E=c[3*i],L=r*c[3*i+1],z=w*E-T*L,I=w*L+T*E,B=b+S,N=x+A,D=b-S,O=x-A,V=F+z,j=P+I,R=r*(F-z),G=r*(P-I);if(e[a]=B+V,e[a+1]=N+j,e[d]=D+G,e[d+1]=O-R,0===n){e[u]=B-V,e[u+1]=N-j;continue}if(n===l)continue;const q=o+s-n,$=o+t-n;e[q]=D-r*G,e[q+1]=-O-r*R,e[$]=B-r*V,e[$+1]=r*j-N}}const d=n>>>1;for(let t=2;t<d;t+=2)e[n-t]=e[t],e[n-t+1]=-e[t+1]}_singleRealTransform2(e,t,r,n,o){const s=e[n],a=e[n+o];t[r]=s+a,t[r+1]=0,t[r+2]=s-a,t[r+3]=0}_singleRealTransform4(e,t,r,n,o,s){const a=2*o,i=3*o,l=e[n],c=e[n+o],d=e[n+a],u=e[n+i],p=l+d,h=l-d,m=c+u,_=s*(c-u);t[r]=p+m,t[r+1]=0,t[r+2]=h,t[r+3]=-_,t[r+4]=p-m,t[r+5]=0,t[r+6]=h,t[r+7]=_}}class m{constructor(e){const t=2*(e-1),r=2*(2*e-1),n=2**Math.ceil(Math.log2(r));this.bufferSize=n,this._a=t;const o=new Float64Array(r),s=new Float64Array(n);this._chirpBuffer=new Float64Array(n),this._buffer1=new Float64Array(n),this._buffer2=new Float64Array(n),this._outBuffer1=new Float64Array(n),this._outBuffer2=new Float64Array(n);const a=-2*Math.PI/e,i=Math.cos(a),l=Math.sin(a);for(let t=0;t<r>>1;++t){const r=(t+1-e)**2/2,n=Math.sqrt(i**2+l**2)**r,a=r*Math.atan2(l,i),c=2*t;o[c]=n*Math.cos(a),o[c+1]=n*Math.sin(a),s[c]=o[c],s[c+1]=-o[c+1]}this._slicedChirpBuffer=o.subarray(t,r),this._f=new h(n>>1),this._f.transform(this._chirpBuffer,s)}_transform(e,t,r){const n=this._buffer1,o=this._buffer2,s=this._outBuffer1,a=this._outBuffer2,i=this._chirpBuffer,l=this._slicedChirpBuffer,c=this._a;if(r)for(let e=0;e<l.length;e+=2){const r=e+1,o=t[e>>1];n[e]=o*l[e],n[r]=o*l[r]}else for(let e=0;e<l.length;e+=2){const r=e+1;n[e]=t[e]*l[e]-t[r]*l[r],n[r]=t[e]*l[r]+t[r]*l[e]}this._f.transform(s,n);for(let e=0;e<i.length;e+=2){const t=e+1;o[e]=s[e]*i[e]-s[t]*i[t],o[t]=s[e]*i[t]+s[t]*i[e]}this._f.inverseTransform(a,o);for(let t=0;t<a.length;t+=2){const r=a[t+c],n=a[t+c+1],o=l[t],s=l[t+1];e[t]=r*o-n*s,e[t+1]=r*s+n*o}}transform(e,t){this._transform(e,t,!1)}realTransform(e,t){this._transform(e,t,!0)}}class _{constructor(e){this.fft_length=e,this.isPowerOfTwo=p(e),this.isPowerOfTwo?(this.fft=new h(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 r=new e.constructor(e.length),n=new e.constructor(t),o=Math.floor(t/2);for(let t=0;t<e.length;++t){let s=0;for(let r=-o;r<=o;++r){let o=t+r;o<0?o=Math.abs(o):o>=e.length&&(o=2*(e.length-1)-o),n[s++]=e[o]}n.sort(),r[t]=n[o]}return r}function g(e,t){const r=Math.pow(10,t);return Math.round(e*r)/r}function M(e){const t=Math.round(e);return Math.abs(e)%1==.5?t%2==0?t:t-1:t}function w(e){const t=e.length,r=e[0].length,n=[t+1,r+1],o=Array.from({length:n[0]},(()=>Array(n[1]).fill(1/0)));o[0][0]=0;const s=Array.from({length:n[0]},(()=>Array(n[1]).fill(-1)));for(let t=1;t<n[1];++t)for(let r=1;r<n[0];++r){const n=o[r-1][t-1],a=o[r-1][t],i=o[r][t-1];let l,c;n<a&&n<i?(l=n,c=0):a<n&&a<i?(l=a,c=1):(l=i,c=2),o[r][t]=e[r-1][t-1]+l,s[r][t]=c}for(let e=0;e<n[1];++e)s[0][e]=2;for(let e=0;e<n[0];++e)s[e][0]=1;let a=t,i=r,l=[],c=[];for(;a>0||i>0;)switch(l.push(a-1),c.push(i-1),s[a][i]){case 0:--a,--i;break;case 1:--a;break;case 2:--i;break;default:throw new Error(`Internal error in dynamic time warping. Unexpected trace[${a}, ${i}]. Please file a bug report.`)}return l.reverse(),c.reverse(),[l,c]}},"./src/utils/tensor.js":
 /*!*****************************!*\
   !*** ./src/utils/tensor.js ***!
   \*****************************/(e,t,r)=>{r.r(t),r.d(t,{Tensor:()=>i,cat:()=>w,full:()=>k,full_like:()=>F,interpolate:()=>c,interpolate_4d:()=>d,layer_norm:()=>_,matmul:()=>u,mean:()=>x,mean_pooling:()=>m,ones:()=>P,ones_like:()=>C,permute:()=>l,quantize_embeddings:()=>A,rfft:()=>p,stack:()=>T,std_mean:()=>b,topk:()=>h,zeros:()=>v,zeros_like:()=>S});var n=r(/*! ./maths.js */"./src/utils/maths.js"),o=r(/*! ../backends/onnx.js */"./src/backends/onnx.js"),s=r(/*! ../ops/registry.js */"./src/ops/registry.js");const a=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});class i{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,o.isONNXTensor)(e[0])?this.ort_tensor=e[0]:this.ort_tensor=new o.Tensor(e[0],e[1],e[2]),new Proxy(this,{get:(e,t)=>{if("string"==typeof t){let r=Number(t);if(Number.isInteger(r))return e._getitem(r)}return e[t]},set:(e,t,r)=>e[t]=r})}dispose(){this.ort_tensor.dispose()}*[Symbol.iterator](){const[e,...t]=this.dims;if(t.length>0){const r=t.reduce(((e,t)=>e*t));for(let n=0;n<e;++n)yield this._subarray(n,r,t)}else yield*this.data}_getitem(e){const[t,...r]=this.dims;if(e=M(e,t),r.length>0){const t=r.reduce(((e,t)=>e*t));return this._subarray(e,t,r)}return new i(this.type,[this.data[e]],r)}indexOf(e){const t=this.data;for(let r=0;r<t.length;++r)if(t[r]==e)return r;return-1}_subarray(e,t,r){const n=e*t,o=(e+1)*t,s="subarray"in this.data?this.data.subarray(n,o):this.data.slice(n,o);return new i(this.type,s,r)}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 r=e.length,n=t.reduce(((e,t)=>e*t));if(r!==n)throw Error(`cannot reshape array of size ${r} into shape (${t})`);let o=e;for(let e=t.length-1;e>=0;e--)o=o.reduce(((r,n)=>{let o=r[r.length-1];return o.length<t[e]?o.push(n):r.push([n]),r}),[[]]);return o[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 r=0;r<t.length;++r)t[r]=e(t[r],r,t);return this}mul(e){return this.clone().mul_(e)}mul_(e){const t=this.data;for(let r=0;r<t.length;++r)t[r]*=e;return this}div(e){return this.clone().div_(e)}div_(e){const t=this.data;for(let r=0;r<t.length;++r)t[r]/=e;return this}add(e){return this.clone().add_(e)}add_(e){const t=this.data;for(let r=0;r<t.length;++r)t[r]+=e;return this}sub(e){return this.clone().sub_(e)}sub_(e){const t=this.data;for(let r=0;r<t.length;++r)t[r]-=e;return this}clone(){return new i(this.type,this.data.slice(),this.dims.slice())}slice(...e){const t=[],r=[];for(let n=0;n<this.dims.length;++n){let o=e[n];if(null==o)r.push([0,this.dims[n]]),t.push(this.dims[n]);else if("number"==typeof o)o=M(o,this.dims[n],n),r.push([o,o+1]);else{if(!Array.isArray(o)||2!==o.length)throw new Error(`Invalid slice: ${o}`);{let[e,s]=o;if(e=null===e?0:M(e,this.dims[n],n,!1),s=null===s?this.dims[n]:M(s,this.dims[n],n,!1),e>s)throw new Error(`Invalid slice: ${o}`);const a=[Math.max(e,0),Math.min(s,this.dims[n])];r.push(a),t.push(a[1]-a[0])}}}const n=r.map((([e,t])=>t-e)),o=n.reduce(((e,t)=>e*t)),s=this.data,a=new s.constructor(o),l=this.stride();for(let e=0;e<o;++e){let t=0;for(let o=n.length-1,s=e;o>=0;--o){const e=n[o];t+=(s%e+r[o][0])*l[o],s=Math.floor(s/e)}a[e]=s[t]}return new i(this.type,a,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,r=!1){if("fro"===e)e=2;else if("string"==typeof e)throw Error(`Unsupported norm: ${e}`);const n=this.data;if(null===t){let t=n.reduce(((t,r)=>t+r**e),0)**(1/e);return new i(this.type,[t],[])}t=M(t,this.dims.length);const o=this.dims.slice();o[t]=1;const s=new n.constructor(n.length/this.dims[t]);for(let r=0;r<n.length;++r){let a=0;for(let e=this.dims.length-1,n=r,s=1;e>=0;--e){const r=this.dims[e];if(e!==t){a+=n%r*s,s*=o[e]}n=Math.floor(n/r)}s[a]+=n[r]**e}if(1!==e)for(let t=0;t<s.length;++t)s[t]=s[t]**(1/e);return r||o.splice(t,1),new i(this.type,s,o)}normalize_(e=2,t=1){t=M(t,this.dims.length);const r=this.norm(e,t,!0),n=this.data,o=r.data;for(let e=0;e<n.length;++e){let r=0;for(let n=this.dims.length-1,o=e,s=1;n>=0;--n){const e=this.dims[n];if(n!==t){r+=o%e*s,s*=this.dims[n]}o=Math.floor(o/e)}n[e]/=o[r]}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 r=e.length-1,n=1;r>=0;--r)t[r]=n,n*=e[r];return t}(this.dims)}squeeze(e=null){return new i(this.type,this.data,f(this.dims,e))}squeeze_(e=null){return this.dims=f(this.dims,e),this}unsqueeze(e=null){return new i(this.type,this.data,g(this.dims,e))}unsqueeze_(e=null){return this.dims=g(this.dims,e),this}flatten_(e=0,t=-1){t=(t+this.dims.length)%this.dims.length;let r=this.dims.slice(0,e),n=this.dims.slice(e,t+1),o=this.dims.slice(t+1);return this.dims=[...r,n.reduce(((e,t)=>e*t),1),...o],this}flatten(e=0,t=-1){return this.clone().flatten_(e,t)}view(...e){let t=-1;for(let r=0;r<e.length;++r)if(-1===e[r]){if(-1!==t)throw new Error("Only one dimension can be inferred");t=r}const r=this.data;if(-1!==t){const n=e.reduce(((e,r,n)=>n!==t?e*r:e),1);e[t]=r.length/n}return new i(this.type,r,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_()}clamp_(e,t){const r=this.data;for(let n=0;n<r.length;++n)r[n]=Math.min(Math.max(r[n],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 x(this,e,t)}to(e){if(this.type===e)return this;if(!a.hasOwnProperty(e))throw new Error(`Unsupported type: ${e}`);return new i(e,a[e].from(this.data),this.dims)}}function l(e,t){const[r,o]=(0,n.permute_data)(e.data,e.dims,t);return new i(e.type,r,o)}function c(e,[t,r],o="bilinear",s=!1){const a=e.dims.at(-3)??1,l=e.dims.at(-2),c=e.dims.at(-1);let d=(0,n.interpolate_data)(e.data,[a,l,c],[t,r],o,s);return new i(e.type,d,[a,t,r])}async function d(e,{size:t=null,mode:r="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 n,o;if(2===t.length)n=[...e.dims.slice(0,2),...t];else if(3===t.length)n=[e.dims[0],...t];else{if(4!==t.length)throw new Error("`size` must be of length 2, 3, or 4.");n=t}if("bilinear"===r)o=await s.TensorOpRegistry.bilinear_interpolate_4d;else{if("bicubic"!==r)throw new Error(`Unsupported mode: ${r}`);o=await s.TensorOpRegistry.bicubic_interpolate_4d}const a=new i("int64",new BigInt64Array(n.map(BigInt)),[n.length]);return await o({x:e,s:a})}async function u(e,t){const r=await s.TensorOpRegistry.matmul;return await r({a:e,b:t})}async function p(e,t){const r=await s.TensorOpRegistry.rfft;return await r({x:e,a:t})}async function h(e,t){const r=await s.TensorOpRegistry.top_k;return t=null===t?e.dims.at(-1):Math.min(t,e.dims.at(-1)),await r({x:e,k:new i("int64",[BigInt(t)],[1])})}function m(e,t){const r=e.data,n=t.data,o=[e.dims[0],e.dims[2]],s=new r.constructor(o[0]*o[1]),[a,l,c]=e.dims;let d=0;for(let e=0;e<a;++e){const t=e*c*l;for(let o=0;o<c;++o){let a=0,i=0;const u=e*l,p=t+o;for(let e=0;e<l;++e){const t=Number(n[u+e]);i+=t,a+=r[p+e*c]*t}const h=a/i;s[d++]=h}}return new i(e.type,s,o)}function _(e,t,{eps:r=1e-5}={}){if(2!==e.dims.length)throw new Error("`layer_norm` currently only supports 2D input.");const[n,o]=e.dims;if(1!==t.length&&t[0]!==o)throw new Error("`normalized_shape` must be a 1D array with shape `[input.dims[1]]`.");const[s,a]=b(e,1,0,!0),l=s.data,c=a.data,d=e.data,u=new d.constructor(d.length);for(let e=0;e<n;++e){const t=e*o;for(let n=0;n<o;++n){const o=t+n;u[o]=(d[o]-c[e])/(l[e]+r)}}return new i(e.type,u,e.dims)}function f(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,r)=>1!==e||!t.includes(r)))),e}function g(e,t){return t=M(t,e.length+1),(e=e.slice()).splice(t,0,1),e}function M(e,t,r=null,n=!0){if(n&&(e<-t||e>=t))throw new Error(`IndexError: index ${e} is out of bounds for dimension${null===r?"":" "+r} with size ${t}`);return e<0&&(e=(e%t+t)%t),e}function w(e,t=0){t=M(t,e[0].dims.length);const r=e[0].dims.slice();r[t]=e.reduce(((e,r)=>e+r.dims[t]),0);const n=r.reduce(((e,t)=>e*t),1),o=new e[0].data.constructor(n),s=e[0].type;if(0===t){let t=0;for(const r of e){const e=r.data;o.set(e,t),t+=e.length}}else{let n=0;for(let s=0;s<e.length;++s){const{data:a,dims:i}=e[s];for(let e=0;e<a.length;++e){let s=0;for(let o=i.length-1,a=e,l=1;o>=0;--o){const e=i[o];let c=a%e;o===t&&(c+=n),s+=c*l,l*=r[o],a=Math.floor(a/e)}o[s]=a[e]}n+=i[t]}}return new i(s,o,r)}function T(e,t=0){return w(e.map((e=>e.unsqueeze(t))),t)}function b(e,t=null,r=1,n=!1){const o=e.data,s=e.dims;if(null===t){const t=o.reduce(((e,t)=>e+t),0)/o.length,n=Math.sqrt(o.reduce(((e,r)=>e+(r-t)**2),0)/(o.length-r)),s=new i(e.type,[t],[]);return[new i(e.type,[n],[]),s]}const a=x(e,t=M(t,s.length),n),l=a.data,c=s.slice();c[t]=1;const d=new o.constructor(o.length/s[t]);for(let e=0;e<o.length;++e){let r=0;for(let n=s.length-1,o=e,a=1;n>=0;--n){const e=s[n];if(n!==t){r+=o%e*a,a*=c[n]}o=Math.floor(o/e)}d[r]+=(o[e]-l[r])**2}for(let e=0;e<d.length;++e)d[e]=Math.sqrt(d[e]/(s[t]-r));n||c.splice(t,1);return[new i(e.type,d,c),a]}function x(e,t=null,r=!1){const n=e.data;if(null===t){const t=n.reduce(((e,t)=>e+t),0);return new i(e.type,[t/n.length],[])}const o=e.dims;t=M(t,o.length);const s=o.slice();s[t]=1;const a=new n.constructor(n.length/o[t]);for(let e=0;e<n.length;++e){let r=0;for(let n=o.length-1,a=e,i=1;n>=0;--n){const e=o[n];if(n!==t){r+=a%e*i,i*=s[n]}a=Math.floor(a/e)}a[r]+=n[e]}if(1!==o[t])for(let e=0;e<a.length;++e)a[e]=a[e]/o[t];return r||s.splice(t,1),new i(e.type,a,s)}function y(e,t,r,n){const o=e.reduce(((e,t)=>e*t),1);return new i(r,new n(o).fill(t),e)}function k(e,t){let r,n;if("number"==typeof t)r="float32",n=Float32Array;else{if("bigint"!=typeof t)throw new Error("Unsupported data type: "+typeof t);r="int64",n=BigInt64Array}return y(e,t,r,n)}function F(e,t){return k(e.dims,t)}function P(e){return y(e,1n,"int64",BigInt64Array)}function C(e){return P(e.dims)}function v(e){return y(e,0n,"int64",BigInt64Array)}function S(e){return v(e.dims)}function A(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 r="binary"===t,n=r?"int8":"uint8",o=r?Int8Array:Uint8Array,s=e.data,a=new o(s.length/8);for(let e=0;e<s.length;++e){const t=s[e]>0?1:0,n=Math.floor(e/8),o=e%8;a[n]|=t<<7-o,r&&0===o&&(a[n]-=128)}return new i(n,a,[e.dims[0],e.dims[1]/8])}}},l={};function c(e){var t=l[e];if(void 0!==t)return t.exports;var r=l[e]={exports:{}};return i[e](r,r.exports,c),r.exports}a=Object.getPrototypeOf?e=>Object.getPrototypeOf(e):e=>e.__proto__,c.t=function(e,t){if(1&t&&(e=this(e)),8&t)return e;if("object"==typeof e&&e){if(4&t&&e.__esModule)return e;if(16&t&&"function"==typeof e.then)return e}var r=Object.create(null);c.r(r);var n={};s=s||[null,a({}),a([]),a(a)];for(var o=2&t&&e;"object"==typeof o&&!~s.indexOf(o);o=a(o))Object.getOwnPropertyNames(o).forEach((t=>n[t]=()=>e[t]));return n.default=()=>e,c.d(r,n),r},c.d=(e,t)=>{for(var r in t)c.o(t,r)&&!c.o(e,r)&&Object.defineProperty(e,r,{enumerable:!0,get:t[r]})},c.o=(e,t)=>Object.prototype.hasOwnProperty.call(e,t),c.r=e=>{"undefined"!=typeof Symbol&&Symbol.toStringTag&&Object.defineProperty(e,Symbol.toStringTag,{value:"Module"}),Object.defineProperty(e,"__esModule",{value:!0})};var d={};
 /*!*****************************!*\
   !*** ./src/transformers.js ***!
-  \*****************************/c.r(d),c.d(d,{ASTFeatureExtractor:()=>_.ASTFeatureExtractor,ASTForAudioClassification:()=>h.ASTForAudioClassification,ASTModel:()=>h.ASTModel,ASTPreTrainedModel:()=>h.ASTPreTrainedModel,AlbertForMaskedLM:()=>h.AlbertForMaskedLM,AlbertForQuestionAnswering:()=>h.AlbertForQuestionAnswering,AlbertForSequenceClassification:()=>h.AlbertForSequenceClassification,AlbertModel:()=>h.AlbertModel,AlbertPreTrainedModel:()=>h.AlbertPreTrainedModel,AlbertTokenizer:()=>m.AlbertTokenizer,AudioClassificationPipeline:()=>p.AudioClassificationPipeline,AutoConfig:()=>f.AutoConfig,AutoModel:()=>h.AutoModel,AutoModelForAudioClassification:()=>h.AutoModelForAudioClassification,AutoModelForAudioFrameClassification:()=>h.AutoModelForAudioFrameClassification,AutoModelForCTC:()=>h.AutoModelForCTC,AutoModelForCausalLM:()=>h.AutoModelForCausalLM,AutoModelForDepthEstimation:()=>h.AutoModelForDepthEstimation,AutoModelForDocumentQuestionAnswering:()=>h.AutoModelForDocumentQuestionAnswering,AutoModelForImageClassification:()=>h.AutoModelForImageClassification,AutoModelForImageFeatureExtraction:()=>h.AutoModelForImageFeatureExtraction,AutoModelForImageMatting:()=>h.AutoModelForImageMatting,AutoModelForImageSegmentation:()=>h.AutoModelForImageSegmentation,AutoModelForImageToImage:()=>h.AutoModelForImageToImage,AutoModelForMaskGeneration:()=>h.AutoModelForMaskGeneration,AutoModelForMaskedLM:()=>h.AutoModelForMaskedLM,AutoModelForNormalEstimation:()=>h.AutoModelForNormalEstimation,AutoModelForObjectDetection:()=>h.AutoModelForObjectDetection,AutoModelForQuestionAnswering:()=>h.AutoModelForQuestionAnswering,AutoModelForSemanticSegmentation:()=>h.AutoModelForSemanticSegmentation,AutoModelForSeq2SeqLM:()=>h.AutoModelForSeq2SeqLM,AutoModelForSequenceClassification:()=>h.AutoModelForSequenceClassification,AutoModelForSpeechSeq2Seq:()=>h.AutoModelForSpeechSeq2Seq,AutoModelForTextToSpectrogram:()=>h.AutoModelForTextToSpectrogram,AutoModelForTextToWaveform:()=>h.AutoModelForTextToWaveform,AutoModelForTokenClassification:()=>h.AutoModelForTokenClassification,AutoModelForUniversalSegmentation:()=>h.AutoModelForUniversalSegmentation,AutoModelForVision2Seq:()=>h.AutoModelForVision2Seq,AutoModelForXVector:()=>h.AutoModelForXVector,AutoModelForZeroShotObjectDetection:()=>h.AutoModelForZeroShotObjectDetection,AutoProcessor:()=>_.AutoProcessor,AutoTokenizer:()=>m.AutoTokenizer,AutomaticSpeechRecognitionPipeline:()=>p.AutomaticSpeechRecognitionPipeline,BartForConditionalGeneration:()=>h.BartForConditionalGeneration,BartForSequenceClassification:()=>h.BartForSequenceClassification,BartModel:()=>h.BartModel,BartPretrainedModel:()=>h.BartPretrainedModel,BartTokenizer:()=>m.BartTokenizer,BaseModelOutput:()=>h.BaseModelOutput,BaseStreamer:()=>b.BaseStreamer,BeitFeatureExtractor:()=>_.BeitFeatureExtractor,BeitForImageClassification:()=>h.BeitForImageClassification,BeitModel:()=>h.BeitModel,BeitPreTrainedModel:()=>h.BeitPreTrainedModel,BertForMaskedLM:()=>h.BertForMaskedLM,BertForQuestionAnswering:()=>h.BertForQuestionAnswering,BertForSequenceClassification:()=>h.BertForSequenceClassification,BertForTokenClassification:()=>h.BertForTokenClassification,BertModel:()=>h.BertModel,BertPreTrainedModel:()=>h.BertPreTrainedModel,BertTokenizer:()=>m.BertTokenizer,BitImageProcessor:()=>_.BitImageProcessor,BlenderbotForConditionalGeneration:()=>h.BlenderbotForConditionalGeneration,BlenderbotModel:()=>h.BlenderbotModel,BlenderbotPreTrainedModel:()=>h.BlenderbotPreTrainedModel,BlenderbotSmallForConditionalGeneration:()=>h.BlenderbotSmallForConditionalGeneration,BlenderbotSmallModel:()=>h.BlenderbotSmallModel,BlenderbotSmallPreTrainedModel:()=>h.BlenderbotSmallPreTrainedModel,BlenderbotSmallTokenizer:()=>m.BlenderbotSmallTokenizer,BlenderbotTokenizer:()=>m.BlenderbotTokenizer,BloomForCausalLM:()=>h.BloomForCausalLM,BloomModel:()=>h.BloomModel,BloomPreTrainedModel:()=>h.BloomPreTrainedModel,BloomTokenizer:()=>m.BloomTokenizer,CLIPFeatureExtractor:()=>_.CLIPFeatureExtractor,CLIPImageProcessor:()=>_.CLIPImageProcessor,CLIPModel:()=>h.CLIPModel,CLIPPreTrainedModel:()=>h.CLIPPreTrainedModel,CLIPSegForImageSegmentation:()=>h.CLIPSegForImageSegmentation,CLIPSegModel:()=>h.CLIPSegModel,CLIPSegPreTrainedModel:()=>h.CLIPSegPreTrainedModel,CLIPTextModel:()=>h.CLIPTextModel,CLIPTextModelWithProjection:()=>h.CLIPTextModelWithProjection,CLIPTokenizer:()=>m.CLIPTokenizer,CLIPVisionModel:()=>h.CLIPVisionModel,CLIPVisionModelWithProjection:()=>h.CLIPVisionModelWithProjection,CamembertForMaskedLM:()=>h.CamembertForMaskedLM,CamembertForQuestionAnswering:()=>h.CamembertForQuestionAnswering,CamembertForSequenceClassification:()=>h.CamembertForSequenceClassification,CamembertForTokenClassification:()=>h.CamembertForTokenClassification,CamembertModel:()=>h.CamembertModel,CamembertPreTrainedModel:()=>h.CamembertPreTrainedModel,CamembertTokenizer:()=>m.CamembertTokenizer,CausalLMOutput:()=>h.CausalLMOutput,CausalLMOutputWithPast:()=>h.CausalLMOutputWithPast,ChineseCLIPFeatureExtractor:()=>_.ChineseCLIPFeatureExtractor,ChineseCLIPModel:()=>h.ChineseCLIPModel,ChineseCLIPPreTrainedModel:()=>h.ChineseCLIPPreTrainedModel,ClapAudioModelWithProjection:()=>h.ClapAudioModelWithProjection,ClapFeatureExtractor:()=>_.ClapFeatureExtractor,ClapModel:()=>h.ClapModel,ClapPreTrainedModel:()=>h.ClapPreTrainedModel,ClapTextModelWithProjection:()=>h.ClapTextModelWithProjection,CodeGenForCausalLM:()=>h.CodeGenForCausalLM,CodeGenModel:()=>h.CodeGenModel,CodeGenPreTrainedModel:()=>h.CodeGenPreTrainedModel,CodeGenTokenizer:()=>m.CodeGenTokenizer,CodeLlamaTokenizer:()=>m.CodeLlamaTokenizer,CohereForCausalLM:()=>h.CohereForCausalLM,CohereModel:()=>h.CohereModel,CoherePreTrainedModel:()=>h.CoherePreTrainedModel,CohereTokenizer:()=>m.CohereTokenizer,ConvBertForMaskedLM:()=>h.ConvBertForMaskedLM,ConvBertForQuestionAnswering:()=>h.ConvBertForQuestionAnswering,ConvBertForSequenceClassification:()=>h.ConvBertForSequenceClassification,ConvBertForTokenClassification:()=>h.ConvBertForTokenClassification,ConvBertModel:()=>h.ConvBertModel,ConvBertPreTrainedModel:()=>h.ConvBertPreTrainedModel,ConvBertTokenizer:()=>m.ConvBertTokenizer,ConvNextFeatureExtractor:()=>_.ConvNextFeatureExtractor,ConvNextForImageClassification:()=>h.ConvNextForImageClassification,ConvNextImageProcessor:()=>_.ConvNextImageProcessor,ConvNextModel:()=>h.ConvNextModel,ConvNextPreTrainedModel:()=>h.ConvNextPreTrainedModel,ConvNextV2ForImageClassification:()=>h.ConvNextV2ForImageClassification,ConvNextV2Model:()=>h.ConvNextV2Model,ConvNextV2PreTrainedModel:()=>h.ConvNextV2PreTrainedModel,DPTFeatureExtractor:()=>_.DPTFeatureExtractor,DPTForDepthEstimation:()=>h.DPTForDepthEstimation,DPTImageProcessor:()=>_.DPTImageProcessor,DPTModel:()=>h.DPTModel,DPTPreTrainedModel:()=>h.DPTPreTrainedModel,DebertaForMaskedLM:()=>h.DebertaForMaskedLM,DebertaForQuestionAnswering:()=>h.DebertaForQuestionAnswering,DebertaForSequenceClassification:()=>h.DebertaForSequenceClassification,DebertaForTokenClassification:()=>h.DebertaForTokenClassification,DebertaModel:()=>h.DebertaModel,DebertaPreTrainedModel:()=>h.DebertaPreTrainedModel,DebertaTokenizer:()=>m.DebertaTokenizer,DebertaV2ForMaskedLM:()=>h.DebertaV2ForMaskedLM,DebertaV2ForQuestionAnswering:()=>h.DebertaV2ForQuestionAnswering,DebertaV2ForSequenceClassification:()=>h.DebertaV2ForSequenceClassification,DebertaV2ForTokenClassification:()=>h.DebertaV2ForTokenClassification,DebertaV2Model:()=>h.DebertaV2Model,DebertaV2PreTrainedModel:()=>h.DebertaV2PreTrainedModel,DebertaV2Tokenizer:()=>m.DebertaV2Tokenizer,DecisionTransformerModel:()=>h.DecisionTransformerModel,DecisionTransformerPreTrainedModel:()=>h.DecisionTransformerPreTrainedModel,DeiTFeatureExtractor:()=>_.DeiTFeatureExtractor,DeiTForImageClassification:()=>h.DeiTForImageClassification,DeiTModel:()=>h.DeiTModel,DeiTPreTrainedModel:()=>h.DeiTPreTrainedModel,DepthAnythingForDepthEstimation:()=>h.DepthAnythingForDepthEstimation,DepthAnythingPreTrainedModel:()=>h.DepthAnythingPreTrainedModel,DepthEstimationPipeline:()=>p.DepthEstimationPipeline,DepthProForDepthEstimation:()=>h.DepthProForDepthEstimation,DepthProPreTrainedModel:()=>h.DepthProPreTrainedModel,DetrFeatureExtractor:()=>_.DetrFeatureExtractor,DetrForObjectDetection:()=>h.DetrForObjectDetection,DetrForSegmentation:()=>h.DetrForSegmentation,DetrModel:()=>h.DetrModel,DetrObjectDetectionOutput:()=>h.DetrObjectDetectionOutput,DetrPreTrainedModel:()=>h.DetrPreTrainedModel,DetrSegmentationOutput:()=>h.DetrSegmentationOutput,Dinov2ForImageClassification:()=>h.Dinov2ForImageClassification,Dinov2Model:()=>h.Dinov2Model,Dinov2PreTrainedModel:()=>h.Dinov2PreTrainedModel,DistilBertForMaskedLM:()=>h.DistilBertForMaskedLM,DistilBertForQuestionAnswering:()=>h.DistilBertForQuestionAnswering,DistilBertForSequenceClassification:()=>h.DistilBertForSequenceClassification,DistilBertForTokenClassification:()=>h.DistilBertForTokenClassification,DistilBertModel:()=>h.DistilBertModel,DistilBertPreTrainedModel:()=>h.DistilBertPreTrainedModel,DistilBertTokenizer:()=>m.DistilBertTokenizer,DocumentQuestionAnsweringPipeline:()=>p.DocumentQuestionAnsweringPipeline,DonutFeatureExtractor:()=>_.DonutFeatureExtractor,DonutSwinModel:()=>h.DonutSwinModel,DonutSwinPreTrainedModel:()=>h.DonutSwinPreTrainedModel,EfficientNetForImageClassification:()=>h.EfficientNetForImageClassification,EfficientNetImageProcessor:()=>_.EfficientNetImageProcessor,EfficientNetModel:()=>h.EfficientNetModel,EfficientNetPreTrainedModel:()=>h.EfficientNetPreTrainedModel,ElectraForMaskedLM:()=>h.ElectraForMaskedLM,ElectraForQuestionAnswering:()=>h.ElectraForQuestionAnswering,ElectraForSequenceClassification:()=>h.ElectraForSequenceClassification,ElectraForTokenClassification:()=>h.ElectraForTokenClassification,ElectraModel:()=>h.ElectraModel,ElectraPreTrainedModel:()=>h.ElectraPreTrainedModel,ElectraTokenizer:()=>m.ElectraTokenizer,EosTokenCriteria:()=>x.EosTokenCriteria,EsmForMaskedLM:()=>h.EsmForMaskedLM,EsmForSequenceClassification:()=>h.EsmForSequenceClassification,EsmForTokenClassification:()=>h.EsmForTokenClassification,EsmModel:()=>h.EsmModel,EsmPreTrainedModel:()=>h.EsmPreTrainedModel,EsmTokenizer:()=>m.EsmTokenizer,FFT:()=>T.FFT,FalconForCausalLM:()=>h.FalconForCausalLM,FalconModel:()=>h.FalconModel,FalconPreTrainedModel:()=>h.FalconPreTrainedModel,FalconTokenizer:()=>m.FalconTokenizer,FastViTForImageClassification:()=>h.FastViTForImageClassification,FastViTModel:()=>h.FastViTModel,FastViTPreTrainedModel:()=>h.FastViTPreTrainedModel,FeatureExtractionPipeline:()=>p.FeatureExtractionPipeline,FeatureExtractor:()=>_.FeatureExtractor,FillMaskPipeline:()=>p.FillMaskPipeline,Florence2ForConditionalGeneration:()=>h.Florence2ForConditionalGeneration,Florence2PreTrainedModel:()=>h.Florence2PreTrainedModel,Florence2Processor:()=>_.Florence2Processor,GLPNFeatureExtractor:()=>_.GLPNFeatureExtractor,GLPNForDepthEstimation:()=>h.GLPNForDepthEstimation,GLPNModel:()=>h.GLPNModel,GLPNPreTrainedModel:()=>h.GLPNPreTrainedModel,GPT2LMHeadModel:()=>h.GPT2LMHeadModel,GPT2Model:()=>h.GPT2Model,GPT2PreTrainedModel:()=>h.GPT2PreTrainedModel,GPT2Tokenizer:()=>m.GPT2Tokenizer,GPTBigCodeForCausalLM:()=>h.GPTBigCodeForCausalLM,GPTBigCodeModel:()=>h.GPTBigCodeModel,GPTBigCodePreTrainedModel:()=>h.GPTBigCodePreTrainedModel,GPTJForCausalLM:()=>h.GPTJForCausalLM,GPTJModel:()=>h.GPTJModel,GPTJPreTrainedModel:()=>h.GPTJPreTrainedModel,GPTNeoForCausalLM:()=>h.GPTNeoForCausalLM,GPTNeoModel:()=>h.GPTNeoModel,GPTNeoPreTrainedModel:()=>h.GPTNeoPreTrainedModel,GPTNeoXForCausalLM:()=>h.GPTNeoXForCausalLM,GPTNeoXModel:()=>h.GPTNeoXModel,GPTNeoXPreTrainedModel:()=>h.GPTNeoXPreTrainedModel,GPTNeoXTokenizer:()=>m.GPTNeoXTokenizer,Gemma2ForCausalLM:()=>h.Gemma2ForCausalLM,Gemma2Model:()=>h.Gemma2Model,Gemma2PreTrainedModel:()=>h.Gemma2PreTrainedModel,GemmaForCausalLM:()=>h.GemmaForCausalLM,GemmaModel:()=>h.GemmaModel,GemmaPreTrainedModel:()=>h.GemmaPreTrainedModel,GemmaTokenizer:()=>m.GemmaTokenizer,GraniteForCausalLM:()=>h.GraniteForCausalLM,GraniteModel:()=>h.GraniteModel,GranitePreTrainedModel:()=>h.GranitePreTrainedModel,Grok1Tokenizer:()=>m.Grok1Tokenizer,GroupViTModel:()=>h.GroupViTModel,GroupViTPreTrainedModel:()=>h.GroupViTPreTrainedModel,HerbertTokenizer:()=>m.HerbertTokenizer,HieraForImageClassification:()=>h.HieraForImageClassification,HieraModel:()=>h.HieraModel,HieraPreTrainedModel:()=>h.HieraPreTrainedModel,HubertForCTC:()=>h.HubertForCTC,HubertForSequenceClassification:()=>h.HubertForSequenceClassification,HubertModel:()=>h.HubertModel,HubertPreTrainedModel:()=>h.HubertPreTrainedModel,ImageClassificationPipeline:()=>p.ImageClassificationPipeline,ImageFeatureExtractionPipeline:()=>p.ImageFeatureExtractionPipeline,ImageFeatureExtractor:()=>_.ImageFeatureExtractor,ImageMattingOutput:()=>h.ImageMattingOutput,ImageSegmentationPipeline:()=>p.ImageSegmentationPipeline,ImageToImagePipeline:()=>p.ImageToImagePipeline,ImageToTextPipeline:()=>p.ImageToTextPipeline,InterruptableStoppingCriteria:()=>x.InterruptableStoppingCriteria,JAISLMHeadModel:()=>h.JAISLMHeadModel,JAISModel:()=>h.JAISModel,JAISPreTrainedModel:()=>h.JAISPreTrainedModel,LlamaForCausalLM:()=>h.LlamaForCausalLM,LlamaModel:()=>h.LlamaModel,LlamaPreTrainedModel:()=>h.LlamaPreTrainedModel,LlamaTokenizer:()=>m.LlamaTokenizer,LlavaForConditionalGeneration:()=>h.LlavaForConditionalGeneration,LlavaPreTrainedModel:()=>h.LlavaPreTrainedModel,LongT5ForConditionalGeneration:()=>h.LongT5ForConditionalGeneration,LongT5Model:()=>h.LongT5Model,LongT5PreTrainedModel:()=>h.LongT5PreTrainedModel,M2M100ForConditionalGeneration:()=>h.M2M100ForConditionalGeneration,M2M100Model:()=>h.M2M100Model,M2M100PreTrainedModel:()=>h.M2M100PreTrainedModel,M2M100Tokenizer:()=>m.M2M100Tokenizer,MBart50Tokenizer:()=>m.MBart50Tokenizer,MBartForCausalLM:()=>h.MBartForCausalLM,MBartForConditionalGeneration:()=>h.MBartForConditionalGeneration,MBartForSequenceClassification:()=>h.MBartForSequenceClassification,MBartModel:()=>h.MBartModel,MBartPreTrainedModel:()=>h.MBartPreTrainedModel,MBartTokenizer:()=>m.MBartTokenizer,MPNetForMaskedLM:()=>h.MPNetForMaskedLM,MPNetForQuestionAnswering:()=>h.MPNetForQuestionAnswering,MPNetForSequenceClassification:()=>h.MPNetForSequenceClassification,MPNetForTokenClassification:()=>h.MPNetForTokenClassification,MPNetModel:()=>h.MPNetModel,MPNetPreTrainedModel:()=>h.MPNetPreTrainedModel,MPNetTokenizer:()=>m.MPNetTokenizer,MT5ForConditionalGeneration:()=>h.MT5ForConditionalGeneration,MT5Model:()=>h.MT5Model,MT5PreTrainedModel:()=>h.MT5PreTrainedModel,MarianMTModel:()=>h.MarianMTModel,MarianModel:()=>h.MarianModel,MarianPreTrainedModel:()=>h.MarianPreTrainedModel,MarianTokenizer:()=>m.MarianTokenizer,MaskFormerFeatureExtractor:()=>_.MaskFormerFeatureExtractor,MaskFormerForInstanceSegmentation:()=>h.MaskFormerForInstanceSegmentation,MaskFormerModel:()=>h.MaskFormerModel,MaskFormerPreTrainedModel:()=>h.MaskFormerPreTrainedModel,MaskedLMOutput:()=>h.MaskedLMOutput,MaxLengthCriteria:()=>x.MaxLengthCriteria,MistralForCausalLM:()=>h.MistralForCausalLM,MistralModel:()=>h.MistralModel,MistralPreTrainedModel:()=>h.MistralPreTrainedModel,MobileBertForMaskedLM:()=>h.MobileBertForMaskedLM,MobileBertForQuestionAnswering:()=>h.MobileBertForQuestionAnswering,MobileBertForSequenceClassification:()=>h.MobileBertForSequenceClassification,MobileBertModel:()=>h.MobileBertModel,MobileBertPreTrainedModel:()=>h.MobileBertPreTrainedModel,MobileBertTokenizer:()=>m.MobileBertTokenizer,MobileNetV1FeatureExtractor:()=>_.MobileNetV1FeatureExtractor,MobileNetV1ForImageClassification:()=>h.MobileNetV1ForImageClassification,MobileNetV1Model:()=>h.MobileNetV1Model,MobileNetV1PreTrainedModel:()=>h.MobileNetV1PreTrainedModel,MobileNetV2FeatureExtractor:()=>_.MobileNetV2FeatureExtractor,MobileNetV2ForImageClassification:()=>h.MobileNetV2ForImageClassification,MobileNetV2Model:()=>h.MobileNetV2Model,MobileNetV2PreTrainedModel:()=>h.MobileNetV2PreTrainedModel,MobileNetV3FeatureExtractor:()=>_.MobileNetV3FeatureExtractor,MobileNetV3ForImageClassification:()=>h.MobileNetV3ForImageClassification,MobileNetV3Model:()=>h.MobileNetV3Model,MobileNetV3PreTrainedModel:()=>h.MobileNetV3PreTrainedModel,MobileNetV4FeatureExtractor:()=>_.MobileNetV4FeatureExtractor,MobileNetV4ForImageClassification:()=>h.MobileNetV4ForImageClassification,MobileNetV4Model:()=>h.MobileNetV4Model,MobileNetV4PreTrainedModel:()=>h.MobileNetV4PreTrainedModel,MobileViTFeatureExtractor:()=>_.MobileViTFeatureExtractor,MobileViTForImageClassification:()=>h.MobileViTForImageClassification,MobileViTImageProcessor:()=>_.MobileViTImageProcessor,MobileViTModel:()=>h.MobileViTModel,MobileViTPreTrainedModel:()=>h.MobileViTPreTrainedModel,MobileViTV2ForImageClassification:()=>h.MobileViTV2ForImageClassification,MobileViTV2Model:()=>h.MobileViTV2Model,MobileViTV2PreTrainedModel:()=>h.MobileViTV2PreTrainedModel,ModelOutput:()=>h.ModelOutput,Moondream1ForConditionalGeneration:()=>h.Moondream1ForConditionalGeneration,MptForCausalLM:()=>h.MptForCausalLM,MptModel:()=>h.MptModel,MptPreTrainedModel:()=>h.MptPreTrainedModel,MusicgenForCausalLM:()=>h.MusicgenForCausalLM,MusicgenForConditionalGeneration:()=>h.MusicgenForConditionalGeneration,MusicgenModel:()=>h.MusicgenModel,MusicgenPreTrainedModel:()=>h.MusicgenPreTrainedModel,NllbTokenizer:()=>m.NllbTokenizer,NomicBertModel:()=>h.NomicBertModel,NomicBertPreTrainedModel:()=>h.NomicBertPreTrainedModel,NougatImageProcessor:()=>_.NougatImageProcessor,NougatTokenizer:()=>m.NougatTokenizer,OPTForCausalLM:()=>h.OPTForCausalLM,OPTModel:()=>h.OPTModel,OPTPreTrainedModel:()=>h.OPTPreTrainedModel,ObjectDetectionPipeline:()=>p.ObjectDetectionPipeline,OpenELMForCausalLM:()=>h.OpenELMForCausalLM,OpenELMModel:()=>h.OpenELMModel,OpenELMPreTrainedModel:()=>h.OpenELMPreTrainedModel,OwlViTFeatureExtractor:()=>_.OwlViTFeatureExtractor,OwlViTForObjectDetection:()=>h.OwlViTForObjectDetection,OwlViTModel:()=>h.OwlViTModel,OwlViTPreTrainedModel:()=>h.OwlViTPreTrainedModel,OwlViTProcessor:()=>_.OwlViTProcessor,Owlv2ForObjectDetection:()=>h.Owlv2ForObjectDetection,Owlv2ImageProcessor:()=>_.Owlv2ImageProcessor,Owlv2Model:()=>h.Owlv2Model,Owlv2PreTrainedModel:()=>h.Owlv2PreTrainedModel,Phi3ForCausalLM:()=>h.Phi3ForCausalLM,Phi3Model:()=>h.Phi3Model,Phi3PreTrainedModel:()=>h.Phi3PreTrainedModel,PhiForCausalLM:()=>h.PhiForCausalLM,PhiModel:()=>h.PhiModel,PhiPreTrainedModel:()=>h.PhiPreTrainedModel,Pipeline:()=>p.Pipeline,PreTrainedModel:()=>h.PreTrainedModel,PreTrainedTokenizer:()=>m.PreTrainedTokenizer,PretrainedConfig:()=>f.PretrainedConfig,PretrainedMixin:()=>h.PretrainedMixin,Processor:()=>_.Processor,PvtForImageClassification:()=>h.PvtForImageClassification,PvtImageProcessor:()=>_.PvtImageProcessor,PvtModel:()=>h.PvtModel,PvtPreTrainedModel:()=>h.PvtPreTrainedModel,PyAnnoteFeatureExtractor:()=>_.PyAnnoteFeatureExtractor,PyAnnoteForAudioFrameClassification:()=>h.PyAnnoteForAudioFrameClassification,PyAnnoteModel:()=>h.PyAnnoteModel,PyAnnotePreTrainedModel:()=>h.PyAnnotePreTrainedModel,PyAnnoteProcessor:()=>_.PyAnnoteProcessor,QuestionAnsweringModelOutput:()=>h.QuestionAnsweringModelOutput,QuestionAnsweringPipeline:()=>p.QuestionAnsweringPipeline,Qwen2ForCausalLM:()=>h.Qwen2ForCausalLM,Qwen2Model:()=>h.Qwen2Model,Qwen2PreTrainedModel:()=>h.Qwen2PreTrainedModel,Qwen2Tokenizer:()=>m.Qwen2Tokenizer,RTDetrForObjectDetection:()=>h.RTDetrForObjectDetection,RTDetrImageProcessor:()=>_.RTDetrImageProcessor,RTDetrModel:()=>h.RTDetrModel,RTDetrObjectDetectionOutput:()=>h.RTDetrObjectDetectionOutput,RTDetrPreTrainedModel:()=>h.RTDetrPreTrainedModel,RawImage:()=>M.RawImage,ResNetForImageClassification:()=>h.ResNetForImageClassification,ResNetModel:()=>h.ResNetModel,ResNetPreTrainedModel:()=>h.ResNetPreTrainedModel,RoFormerForMaskedLM:()=>h.RoFormerForMaskedLM,RoFormerForQuestionAnswering:()=>h.RoFormerForQuestionAnswering,RoFormerForSequenceClassification:()=>h.RoFormerForSequenceClassification,RoFormerForTokenClassification:()=>h.RoFormerForTokenClassification,RoFormerModel:()=>h.RoFormerModel,RoFormerPreTrainedModel:()=>h.RoFormerPreTrainedModel,RoFormerTokenizer:()=>m.RoFormerTokenizer,RobertaForMaskedLM:()=>h.RobertaForMaskedLM,RobertaForQuestionAnswering:()=>h.RobertaForQuestionAnswering,RobertaForSequenceClassification:()=>h.RobertaForSequenceClassification,RobertaForTokenClassification:()=>h.RobertaForTokenClassification,RobertaModel:()=>h.RobertaModel,RobertaPreTrainedModel:()=>h.RobertaPreTrainedModel,RobertaTokenizer:()=>m.RobertaTokenizer,SamImageProcessor:()=>_.SamImageProcessor,SamImageSegmentationOutput:()=>h.SamImageSegmentationOutput,SamModel:()=>h.SamModel,SamPreTrainedModel:()=>h.SamPreTrainedModel,SamProcessor:()=>_.SamProcessor,SapiensFeatureExtractor:()=>_.SapiensFeatureExtractor,SapiensForDepthEstimation:()=>h.SapiensForDepthEstimation,SapiensForNormalEstimation:()=>h.SapiensForNormalEstimation,SapiensForSemanticSegmentation:()=>h.SapiensForSemanticSegmentation,SapiensPreTrainedModel:()=>h.SapiensPreTrainedModel,SeamlessM4TFeatureExtractor:()=>_.SeamlessM4TFeatureExtractor,SegformerFeatureExtractor:()=>_.SegformerFeatureExtractor,SegformerForImageClassification:()=>h.SegformerForImageClassification,SegformerForSemanticSegmentation:()=>h.SegformerForSemanticSegmentation,SegformerModel:()=>h.SegformerModel,SegformerPreTrainedModel:()=>h.SegformerPreTrainedModel,Seq2SeqLMOutput:()=>h.Seq2SeqLMOutput,SequenceClassifierOutput:()=>h.SequenceClassifierOutput,SiglipImageProcessor:()=>_.SiglipImageProcessor,SiglipModel:()=>h.SiglipModel,SiglipPreTrainedModel:()=>h.SiglipPreTrainedModel,SiglipTextModel:()=>h.SiglipTextModel,SiglipTokenizer:()=>m.SiglipTokenizer,SiglipVisionModel:()=>h.SiglipVisionModel,SpeechT5FeatureExtractor:()=>_.SpeechT5FeatureExtractor,SpeechT5ForSpeechToText:()=>h.SpeechT5ForSpeechToText,SpeechT5ForTextToSpeech:()=>h.SpeechT5ForTextToSpeech,SpeechT5HifiGan:()=>h.SpeechT5HifiGan,SpeechT5Model:()=>h.SpeechT5Model,SpeechT5PreTrainedModel:()=>h.SpeechT5PreTrainedModel,SpeechT5Processor:()=>_.SpeechT5Processor,SpeechT5Tokenizer:()=>m.SpeechT5Tokenizer,SqueezeBertForMaskedLM:()=>h.SqueezeBertForMaskedLM,SqueezeBertForQuestionAnswering:()=>h.SqueezeBertForQuestionAnswering,SqueezeBertForSequenceClassification:()=>h.SqueezeBertForSequenceClassification,SqueezeBertModel:()=>h.SqueezeBertModel,SqueezeBertPreTrainedModel:()=>h.SqueezeBertPreTrainedModel,SqueezeBertTokenizer:()=>m.SqueezeBertTokenizer,StableLmForCausalLM:()=>h.StableLmForCausalLM,StableLmModel:()=>h.StableLmModel,StableLmPreTrainedModel:()=>h.StableLmPreTrainedModel,Starcoder2ForCausalLM:()=>h.Starcoder2ForCausalLM,Starcoder2Model:()=>h.Starcoder2Model,Starcoder2PreTrainedModel:()=>h.Starcoder2PreTrainedModel,StoppingCriteria:()=>x.StoppingCriteria,StoppingCriteriaList:()=>x.StoppingCriteriaList,SummarizationPipeline:()=>p.SummarizationPipeline,Swin2SRForImageSuperResolution:()=>h.Swin2SRForImageSuperResolution,Swin2SRImageProcessor:()=>_.Swin2SRImageProcessor,Swin2SRModel:()=>h.Swin2SRModel,Swin2SRPreTrainedModel:()=>h.Swin2SRPreTrainedModel,SwinForImageClassification:()=>h.SwinForImageClassification,SwinModel:()=>h.SwinModel,SwinPreTrainedModel:()=>h.SwinPreTrainedModel,T5ForConditionalGeneration:()=>h.T5ForConditionalGeneration,T5Model:()=>h.T5Model,T5PreTrainedModel:()=>h.T5PreTrainedModel,T5Tokenizer:()=>m.T5Tokenizer,TableTransformerForObjectDetection:()=>h.TableTransformerForObjectDetection,TableTransformerModel:()=>h.TableTransformerModel,TableTransformerObjectDetectionOutput:()=>h.TableTransformerObjectDetectionOutput,TableTransformerPreTrainedModel:()=>h.TableTransformerPreTrainedModel,Tensor:()=>w.Tensor,Text2TextGenerationPipeline:()=>p.Text2TextGenerationPipeline,TextClassificationPipeline:()=>p.TextClassificationPipeline,TextGenerationPipeline:()=>p.TextGenerationPipeline,TextStreamer:()=>b.TextStreamer,TextToAudioPipeline:()=>p.TextToAudioPipeline,TokenClassificationPipeline:()=>p.TokenClassificationPipeline,TokenClassifierOutput:()=>h.TokenClassifierOutput,TokenizerModel:()=>m.TokenizerModel,TrOCRForCausalLM:()=>h.TrOCRForCausalLM,TrOCRPreTrainedModel:()=>h.TrOCRPreTrainedModel,TranslationPipeline:()=>p.TranslationPipeline,UniSpeechForCTC:()=>h.UniSpeechForCTC,UniSpeechForSequenceClassification:()=>h.UniSpeechForSequenceClassification,UniSpeechModel:()=>h.UniSpeechModel,UniSpeechPreTrainedModel:()=>h.UniSpeechPreTrainedModel,UniSpeechSatForAudioFrameClassification:()=>h.UniSpeechSatForAudioFrameClassification,UniSpeechSatForCTC:()=>h.UniSpeechSatForCTC,UniSpeechSatForSequenceClassification:()=>h.UniSpeechSatForSequenceClassification,UniSpeechSatModel:()=>h.UniSpeechSatModel,UniSpeechSatPreTrainedModel:()=>h.UniSpeechSatPreTrainedModel,ViTFeatureExtractor:()=>_.ViTFeatureExtractor,ViTForImageClassification:()=>h.ViTForImageClassification,ViTImageProcessor:()=>_.ViTImageProcessor,ViTMAEModel:()=>h.ViTMAEModel,ViTMAEPreTrainedModel:()=>h.ViTMAEPreTrainedModel,ViTMSNForImageClassification:()=>h.ViTMSNForImageClassification,ViTMSNModel:()=>h.ViTMSNModel,ViTMSNPreTrainedModel:()=>h.ViTMSNPreTrainedModel,ViTModel:()=>h.ViTModel,ViTPreTrainedModel:()=>h.ViTPreTrainedModel,VisionEncoderDecoderModel:()=>h.VisionEncoderDecoderModel,VitMatteForImageMatting:()=>h.VitMatteForImageMatting,VitMatteImageProcessor:()=>_.VitMatteImageProcessor,VitMattePreTrainedModel:()=>h.VitMattePreTrainedModel,VitsModel:()=>h.VitsModel,VitsModelOutput:()=>h.VitsModelOutput,VitsPreTrainedModel:()=>h.VitsPreTrainedModel,VitsTokenizer:()=>m.VitsTokenizer,Wav2Vec2BertForCTC:()=>h.Wav2Vec2BertForCTC,Wav2Vec2BertForSequenceClassification:()=>h.Wav2Vec2BertForSequenceClassification,Wav2Vec2BertModel:()=>h.Wav2Vec2BertModel,Wav2Vec2BertPreTrainedModel:()=>h.Wav2Vec2BertPreTrainedModel,Wav2Vec2CTCTokenizer:()=>m.Wav2Vec2CTCTokenizer,Wav2Vec2FeatureExtractor:()=>_.Wav2Vec2FeatureExtractor,Wav2Vec2ForAudioFrameClassification:()=>h.Wav2Vec2ForAudioFrameClassification,Wav2Vec2ForCTC:()=>h.Wav2Vec2ForCTC,Wav2Vec2ForSequenceClassification:()=>h.Wav2Vec2ForSequenceClassification,Wav2Vec2Model:()=>h.Wav2Vec2Model,Wav2Vec2PreTrainedModel:()=>h.Wav2Vec2PreTrainedModel,Wav2Vec2ProcessorWithLM:()=>_.Wav2Vec2ProcessorWithLM,WavLMForAudioFrameClassification:()=>h.WavLMForAudioFrameClassification,WavLMForCTC:()=>h.WavLMForCTC,WavLMForSequenceClassification:()=>h.WavLMForSequenceClassification,WavLMForXVector:()=>h.WavLMForXVector,WavLMModel:()=>h.WavLMModel,WavLMPreTrainedModel:()=>h.WavLMPreTrainedModel,WeSpeakerFeatureExtractor:()=>_.WeSpeakerFeatureExtractor,WeSpeakerResNetModel:()=>h.WeSpeakerResNetModel,WeSpeakerResNetPreTrainedModel:()=>h.WeSpeakerResNetPreTrainedModel,WhisperFeatureExtractor:()=>_.WhisperFeatureExtractor,WhisperForConditionalGeneration:()=>h.WhisperForConditionalGeneration,WhisperModel:()=>h.WhisperModel,WhisperPreTrainedModel:()=>h.WhisperPreTrainedModel,WhisperProcessor:()=>_.WhisperProcessor,WhisperTextStreamer:()=>b.WhisperTextStreamer,WhisperTokenizer:()=>m.WhisperTokenizer,XLMForQuestionAnswering:()=>h.XLMForQuestionAnswering,XLMForSequenceClassification:()=>h.XLMForSequenceClassification,XLMForTokenClassification:()=>h.XLMForTokenClassification,XLMModel:()=>h.XLMModel,XLMPreTrainedModel:()=>h.XLMPreTrainedModel,XLMRobertaForMaskedLM:()=>h.XLMRobertaForMaskedLM,XLMRobertaForQuestionAnswering:()=>h.XLMRobertaForQuestionAnswering,XLMRobertaForSequenceClassification:()=>h.XLMRobertaForSequenceClassification,XLMRobertaForTokenClassification:()=>h.XLMRobertaForTokenClassification,XLMRobertaModel:()=>h.XLMRobertaModel,XLMRobertaPreTrainedModel:()=>h.XLMRobertaPreTrainedModel,XLMRobertaTokenizer:()=>m.XLMRobertaTokenizer,XLMTokenizer:()=>m.XLMTokenizer,XLMWithLMHeadModel:()=>h.XLMWithLMHeadModel,XVectorOutput:()=>h.XVectorOutput,YolosFeatureExtractor:()=>_.YolosFeatureExtractor,YolosForObjectDetection:()=>h.YolosForObjectDetection,YolosModel:()=>h.YolosModel,YolosObjectDetectionOutput:()=>h.YolosObjectDetectionOutput,YolosPreTrainedModel:()=>h.YolosPreTrainedModel,ZeroShotAudioClassificationPipeline:()=>p.ZeroShotAudioClassificationPipeline,ZeroShotClassificationPipeline:()=>p.ZeroShotClassificationPipeline,ZeroShotImageClassificationPipeline:()=>p.ZeroShotImageClassificationPipeline,ZeroShotObjectDetectionPipeline:()=>p.ZeroShotObjectDetectionPipeline,bankers_round:()=>T.bankers_round,cat:()=>w.cat,cos_sim:()=>T.cos_sim,dot:()=>T.dot,dynamic_time_warping:()=>T.dynamic_time_warping,env:()=>u.env,full:()=>w.full,full_like:()=>w.full_like,getKeyValueShapes:()=>f.getKeyValueShapes,hamming:()=>g.hamming,hanning:()=>g.hanning,interpolate:()=>w.interpolate,interpolate_4d:()=>w.interpolate_4d,interpolate_data:()=>T.interpolate_data,is_chinese_char:()=>m.is_chinese_char,layer_norm:()=>w.layer_norm,log_softmax:()=>T.log_softmax,magnitude:()=>T.magnitude,matmul:()=>w.matmul,max:()=>T.max,mean:()=>w.mean,mean_pooling:()=>w.mean_pooling,medianFilter:()=>T.medianFilter,mel_filter_bank:()=>g.mel_filter_bank,min:()=>T.min,ones:()=>w.ones,ones_like:()=>w.ones_like,permute:()=>w.permute,permute_data:()=>T.permute_data,pipeline:()=>p.pipeline,quantize_embeddings:()=>w.quantize_embeddings,read_audio:()=>g.read_audio,rfft:()=>w.rfft,round:()=>T.round,softmax:()=>T.softmax,spectrogram:()=>g.spectrogram,stack:()=>w.stack,std_mean:()=>w.std_mean,topk:()=>w.topk,window_function:()=>g.window_function,zeros:()=>w.zeros,zeros_like:()=>w.zeros_like});var u=c(/*! ./env.js */"./src/env.js"),p=c(/*! ./pipelines.js */"./src/pipelines.js"),h=c(/*! ./models.js */"./src/models.js"),m=c(/*! ./tokenizers.js */"./src/tokenizers.js"),_=c(/*! ./processors.js */"./src/processors.js"),f=c(/*! ./configs.js */"./src/configs.js"),g=c(/*! ./utils/audio.js */"./src/utils/audio.js"),M=c(/*! ./utils/image.js */"./src/utils/image.js"),w=c(/*! ./utils/tensor.js */"./src/utils/tensor.js"),T=c(/*! ./utils/maths.js */"./src/utils/maths.js"),b=c(/*! ./generation/streamers.js */"./src/generation/streamers.js"),x=c(/*! ./generation/stopping_criteria.js */"./src/generation/stopping_criteria.js"),y=d.ASTFeatureExtractor,k=d.ASTForAudioClassification,F=d.ASTModel,P=d.ASTPreTrainedModel,C=d.AlbertForMaskedLM,v=d.AlbertForQuestionAnswering,S=d.AlbertForSequenceClassification,A=d.AlbertModel,E=d.AlbertPreTrainedModel,L=d.AlbertTokenizer,z=d.AudioClassificationPipeline,I=d.AutoConfig,B=d.AutoModel,N=d.AutoModelForAudioClassification,D=d.AutoModelForAudioFrameClassification,O=d.AutoModelForCTC,V=d.AutoModelForCausalLM,j=d.AutoModelForDepthEstimation,R=d.AutoModelForDocumentQuestionAnswering,G=d.AutoModelForImageClassification,q=d.AutoModelForImageFeatureExtraction,W=d.AutoModelForImageMatting,$=d.AutoModelForImageSegmentation,U=d.AutoModelForImageToImage,X=d.AutoModelForMaskGeneration,Q=d.AutoModelForMaskedLM,H=d.AutoModelForNormalEstimation,Y=d.AutoModelForObjectDetection,J=d.AutoModelForQuestionAnswering,K=d.AutoModelForSemanticSegmentation,Z=d.AutoModelForSeq2SeqLM,ee=d.AutoModelForSequenceClassification,te=d.AutoModelForSpeechSeq2Seq,re=d.AutoModelForTextToSpectrogram,ne=d.AutoModelForTextToWaveform,oe=d.AutoModelForTokenClassification,se=d.AutoModelForUniversalSegmentation,ae=d.AutoModelForVision2Seq,ie=d.AutoModelForXVector,le=d.AutoModelForZeroShotObjectDetection,ce=d.AutoProcessor,de=d.AutoTokenizer,ue=d.AutomaticSpeechRecognitionPipeline,pe=d.BartForConditionalGeneration,he=d.BartForSequenceClassification,me=d.BartModel,_e=d.BartPretrainedModel,fe=d.BartTokenizer,ge=d.BaseModelOutput,Me=d.BaseStreamer,we=d.BeitFeatureExtractor,Te=d.BeitForImageClassification,be=d.BeitModel,xe=d.BeitPreTrainedModel,ye=d.BertForMaskedLM,ke=d.BertForQuestionAnswering,Fe=d.BertForSequenceClassification,Pe=d.BertForTokenClassification,Ce=d.BertModel,ve=d.BertPreTrainedModel,Se=d.BertTokenizer,Ae=d.BitImageProcessor,Ee=d.BlenderbotForConditionalGeneration,Le=d.BlenderbotModel,ze=d.BlenderbotPreTrainedModel,Ie=d.BlenderbotSmallForConditionalGeneration,Be=d.BlenderbotSmallModel,Ne=d.BlenderbotSmallPreTrainedModel,De=d.BlenderbotSmallTokenizer,Oe=d.BlenderbotTokenizer,Ve=d.BloomForCausalLM,je=d.BloomModel,Re=d.BloomPreTrainedModel,Ge=d.BloomTokenizer,qe=d.CLIPFeatureExtractor,We=d.CLIPImageProcessor,$e=d.CLIPModel,Ue=d.CLIPPreTrainedModel,Xe=d.CLIPSegForImageSegmentation,Qe=d.CLIPSegModel,He=d.CLIPSegPreTrainedModel,Ye=d.CLIPTextModel,Je=d.CLIPTextModelWithProjection,Ke=d.CLIPTokenizer,Ze=d.CLIPVisionModel,et=d.CLIPVisionModelWithProjection,tt=d.CamembertForMaskedLM,rt=d.CamembertForQuestionAnswering,nt=d.CamembertForSequenceClassification,ot=d.CamembertForTokenClassification,st=d.CamembertModel,at=d.CamembertPreTrainedModel,it=d.CamembertTokenizer,lt=d.CausalLMOutput,ct=d.CausalLMOutputWithPast,dt=d.ChineseCLIPFeatureExtractor,ut=d.ChineseCLIPModel,pt=d.ChineseCLIPPreTrainedModel,ht=d.ClapAudioModelWithProjection,mt=d.ClapFeatureExtractor,_t=d.ClapModel,ft=d.ClapPreTrainedModel,gt=d.ClapTextModelWithProjection,Mt=d.CodeGenForCausalLM,wt=d.CodeGenModel,Tt=d.CodeGenPreTrainedModel,bt=d.CodeGenTokenizer,xt=d.CodeLlamaTokenizer,yt=d.CohereForCausalLM,kt=d.CohereModel,Ft=d.CoherePreTrainedModel,Pt=d.CohereTokenizer,Ct=d.ConvBertForMaskedLM,vt=d.ConvBertForQuestionAnswering,St=d.ConvBertForSequenceClassification,At=d.ConvBertForTokenClassification,Et=d.ConvBertModel,Lt=d.ConvBertPreTrainedModel,zt=d.ConvBertTokenizer,It=d.ConvNextFeatureExtractor,Bt=d.ConvNextForImageClassification,Nt=d.ConvNextImageProcessor,Dt=d.ConvNextModel,Ot=d.ConvNextPreTrainedModel,Vt=d.ConvNextV2ForImageClassification,jt=d.ConvNextV2Model,Rt=d.ConvNextV2PreTrainedModel,Gt=d.DPTFeatureExtractor,qt=d.DPTForDepthEstimation,Wt=d.DPTImageProcessor,$t=d.DPTModel,Ut=d.DPTPreTrainedModel,Xt=d.DebertaForMaskedLM,Qt=d.DebertaForQuestionAnswering,Ht=d.DebertaForSequenceClassification,Yt=d.DebertaForTokenClassification,Jt=d.DebertaModel,Kt=d.DebertaPreTrainedModel,Zt=d.DebertaTokenizer,er=d.DebertaV2ForMaskedLM,tr=d.DebertaV2ForQuestionAnswering,rr=d.DebertaV2ForSequenceClassification,nr=d.DebertaV2ForTokenClassification,or=d.DebertaV2Model,sr=d.DebertaV2PreTrainedModel,ar=d.DebertaV2Tokenizer,ir=d.DecisionTransformerModel,lr=d.DecisionTransformerPreTrainedModel,cr=d.DeiTFeatureExtractor,dr=d.DeiTForImageClassification,ur=d.DeiTModel,pr=d.DeiTPreTrainedModel,hr=d.DepthAnythingForDepthEstimation,mr=d.DepthAnythingPreTrainedModel,_r=d.DepthEstimationPipeline,fr=d.DepthProForDepthEstimation,gr=d.DepthProPreTrainedModel,Mr=d.DetrFeatureExtractor,wr=d.DetrForObjectDetection,Tr=d.DetrForSegmentation,br=d.DetrModel,xr=d.DetrObjectDetectionOutput,yr=d.DetrPreTrainedModel,kr=d.DetrSegmentationOutput,Fr=d.Dinov2ForImageClassification,Pr=d.Dinov2Model,Cr=d.Dinov2PreTrainedModel,vr=d.DistilBertForMaskedLM,Sr=d.DistilBertForQuestionAnswering,Ar=d.DistilBertForSequenceClassification,Er=d.DistilBertForTokenClassification,Lr=d.DistilBertModel,zr=d.DistilBertPreTrainedModel,Ir=d.DistilBertTokenizer,Br=d.DocumentQuestionAnsweringPipeline,Nr=d.DonutFeatureExtractor,Dr=d.DonutSwinModel,Or=d.DonutSwinPreTrainedModel,Vr=d.EfficientNetForImageClassification,jr=d.EfficientNetImageProcessor,Rr=d.EfficientNetModel,Gr=d.EfficientNetPreTrainedModel,qr=d.ElectraForMaskedLM,Wr=d.ElectraForQuestionAnswering,$r=d.ElectraForSequenceClassification,Ur=d.ElectraForTokenClassification,Xr=d.ElectraModel,Qr=d.ElectraPreTrainedModel,Hr=d.ElectraTokenizer,Yr=d.EosTokenCriteria,Jr=d.EsmForMaskedLM,Kr=d.EsmForSequenceClassification,Zr=d.EsmForTokenClassification,en=d.EsmModel,tn=d.EsmPreTrainedModel,rn=d.EsmTokenizer,nn=d.FFT,on=d.FalconForCausalLM,sn=d.FalconModel,an=d.FalconPreTrainedModel,ln=d.FalconTokenizer,cn=d.FastViTForImageClassification,dn=d.FastViTModel,un=d.FastViTPreTrainedModel,pn=d.FeatureExtractionPipeline,hn=d.FeatureExtractor,mn=d.FillMaskPipeline,_n=d.Florence2ForConditionalGeneration,fn=d.Florence2PreTrainedModel,gn=d.Florence2Processor,Mn=d.GLPNFeatureExtractor,wn=d.GLPNForDepthEstimation,Tn=d.GLPNModel,bn=d.GLPNPreTrainedModel,xn=d.GPT2LMHeadModel,yn=d.GPT2Model,kn=d.GPT2PreTrainedModel,Fn=d.GPT2Tokenizer,Pn=d.GPTBigCodeForCausalLM,Cn=d.GPTBigCodeModel,vn=d.GPTBigCodePreTrainedModel,Sn=d.GPTJForCausalLM,An=d.GPTJModel,En=d.GPTJPreTrainedModel,Ln=d.GPTNeoForCausalLM,zn=d.GPTNeoModel,In=d.GPTNeoPreTrainedModel,Bn=d.GPTNeoXForCausalLM,Nn=d.GPTNeoXModel,Dn=d.GPTNeoXPreTrainedModel,On=d.GPTNeoXTokenizer,Vn=d.Gemma2ForCausalLM,jn=d.Gemma2Model,Rn=d.Gemma2PreTrainedModel,Gn=d.GemmaForCausalLM,qn=d.GemmaModel,Wn=d.GemmaPreTrainedModel,$n=d.GemmaTokenizer,Un=d.GraniteForCausalLM,Xn=d.GraniteModel,Qn=d.GranitePreTrainedModel,Hn=d.Grok1Tokenizer,Yn=d.GroupViTModel,Jn=d.GroupViTPreTrainedModel,Kn=d.HerbertTokenizer,Zn=d.HieraForImageClassification,eo=d.HieraModel,to=d.HieraPreTrainedModel,ro=d.HubertForCTC,no=d.HubertForSequenceClassification,oo=d.HubertModel,so=d.HubertPreTrainedModel,ao=d.ImageClassificationPipeline,io=d.ImageFeatureExtractionPipeline,lo=d.ImageFeatureExtractor,co=d.ImageMattingOutput,uo=d.ImageSegmentationPipeline,po=d.ImageToImagePipeline,ho=d.ImageToTextPipeline,mo=d.InterruptableStoppingCriteria,_o=d.JAISLMHeadModel,fo=d.JAISModel,go=d.JAISPreTrainedModel,Mo=d.LlamaForCausalLM,wo=d.LlamaModel,To=d.LlamaPreTrainedModel,bo=d.LlamaTokenizer,xo=d.LlavaForConditionalGeneration,yo=d.LlavaPreTrainedModel,ko=d.LongT5ForConditionalGeneration,Fo=d.LongT5Model,Po=d.LongT5PreTrainedModel,Co=d.M2M100ForConditionalGeneration,vo=d.M2M100Model,So=d.M2M100PreTrainedModel,Ao=d.M2M100Tokenizer,Eo=d.MBart50Tokenizer,Lo=d.MBartForCausalLM,zo=d.MBartForConditionalGeneration,Io=d.MBartForSequenceClassification,Bo=d.MBartModel,No=d.MBartPreTrainedModel,Do=d.MBartTokenizer,Oo=d.MPNetForMaskedLM,Vo=d.MPNetForQuestionAnswering,jo=d.MPNetForSequenceClassification,Ro=d.MPNetForTokenClassification,Go=d.MPNetModel,qo=d.MPNetPreTrainedModel,Wo=d.MPNetTokenizer,$o=d.MT5ForConditionalGeneration,Uo=d.MT5Model,Xo=d.MT5PreTrainedModel,Qo=d.MarianMTModel,Ho=d.MarianModel,Yo=d.MarianPreTrainedModel,Jo=d.MarianTokenizer,Ko=d.MaskFormerFeatureExtractor,Zo=d.MaskFormerForInstanceSegmentation,es=d.MaskFormerModel,ts=d.MaskFormerPreTrainedModel,rs=d.MaskedLMOutput,ns=d.MaxLengthCriteria,os=d.MistralForCausalLM,ss=d.MistralModel,as=d.MistralPreTrainedModel,is=d.MobileBertForMaskedLM,ls=d.MobileBertForQuestionAnswering,cs=d.MobileBertForSequenceClassification,ds=d.MobileBertModel,us=d.MobileBertPreTrainedModel,ps=d.MobileBertTokenizer,hs=d.MobileNetV1FeatureExtractor,ms=d.MobileNetV1ForImageClassification,_s=d.MobileNetV1Model,fs=d.MobileNetV1PreTrainedModel,gs=d.MobileNetV2FeatureExtractor,Ms=d.MobileNetV2ForImageClassification,ws=d.MobileNetV2Model,Ts=d.MobileNetV2PreTrainedModel,bs=d.MobileNetV3FeatureExtractor,xs=d.MobileNetV3ForImageClassification,ys=d.MobileNetV3Model,ks=d.MobileNetV3PreTrainedModel,Fs=d.MobileNetV4FeatureExtractor,Ps=d.MobileNetV4ForImageClassification,Cs=d.MobileNetV4Model,vs=d.MobileNetV4PreTrainedModel,Ss=d.MobileViTFeatureExtractor,As=d.MobileViTForImageClassification,Es=d.MobileViTImageProcessor,Ls=d.MobileViTModel,zs=d.MobileViTPreTrainedModel,Is=d.MobileViTV2ForImageClassification,Bs=d.MobileViTV2Model,Ns=d.MobileViTV2PreTrainedModel,Ds=d.ModelOutput,Os=d.Moondream1ForConditionalGeneration,Vs=d.MptForCausalLM,js=d.MptModel,Rs=d.MptPreTrainedModel,Gs=d.MusicgenForCausalLM,qs=d.MusicgenForConditionalGeneration,Ws=d.MusicgenModel,$s=d.MusicgenPreTrainedModel,Us=d.NllbTokenizer,Xs=d.NomicBertModel,Qs=d.NomicBertPreTrainedModel,Hs=d.NougatImageProcessor,Ys=d.NougatTokenizer,Js=d.OPTForCausalLM,Ks=d.OPTModel,Zs=d.OPTPreTrainedModel,ea=d.ObjectDetectionPipeline,ta=d.OpenELMForCausalLM,ra=d.OpenELMModel,na=d.OpenELMPreTrainedModel,oa=d.OwlViTFeatureExtractor,sa=d.OwlViTForObjectDetection,aa=d.OwlViTModel,ia=d.OwlViTPreTrainedModel,la=d.OwlViTProcessor,ca=d.Owlv2ForObjectDetection,da=d.Owlv2ImageProcessor,ua=d.Owlv2Model,pa=d.Owlv2PreTrainedModel,ha=d.Phi3ForCausalLM,ma=d.Phi3Model,_a=d.Phi3PreTrainedModel,fa=d.PhiForCausalLM,ga=d.PhiModel,Ma=d.PhiPreTrainedModel,wa=d.Pipeline,Ta=d.PreTrainedModel,ba=d.PreTrainedTokenizer,xa=d.PretrainedConfig,ya=d.PretrainedMixin,ka=d.Processor,Fa=d.PvtForImageClassification,Pa=d.PvtImageProcessor,Ca=d.PvtModel,va=d.PvtPreTrainedModel,Sa=d.PyAnnoteFeatureExtractor,Aa=d.PyAnnoteForAudioFrameClassification,Ea=d.PyAnnoteModel,La=d.PyAnnotePreTrainedModel,za=d.PyAnnoteProcessor,Ia=d.QuestionAnsweringModelOutput,Ba=d.QuestionAnsweringPipeline,Na=d.Qwen2ForCausalLM,Da=d.Qwen2Model,Oa=d.Qwen2PreTrainedModel,Va=d.Qwen2Tokenizer,ja=d.RTDetrForObjectDetection,Ra=d.RTDetrImageProcessor,Ga=d.RTDetrModel,qa=d.RTDetrObjectDetectionOutput,Wa=d.RTDetrPreTrainedModel,$a=d.RawImage,Ua=d.ResNetForImageClassification,Xa=d.ResNetModel,Qa=d.ResNetPreTrainedModel,Ha=d.RoFormerForMaskedLM,Ya=d.RoFormerForQuestionAnswering,Ja=d.RoFormerForSequenceClassification,Ka=d.RoFormerForTokenClassification,Za=d.RoFormerModel,ei=d.RoFormerPreTrainedModel,ti=d.RoFormerTokenizer,ri=d.RobertaForMaskedLM,ni=d.RobertaForQuestionAnswering,oi=d.RobertaForSequenceClassification,si=d.RobertaForTokenClassification,ai=d.RobertaModel,ii=d.RobertaPreTrainedModel,li=d.RobertaTokenizer,ci=d.SamImageProcessor,di=d.SamImageSegmentationOutput,ui=d.SamModel,pi=d.SamPreTrainedModel,hi=d.SamProcessor,mi=d.SapiensFeatureExtractor,_i=d.SapiensForDepthEstimation,fi=d.SapiensForNormalEstimation,gi=d.SapiensForSemanticSegmentation,Mi=d.SapiensPreTrainedModel,wi=d.SeamlessM4TFeatureExtractor,Ti=d.SegformerFeatureExtractor,bi=d.SegformerForImageClassification,xi=d.SegformerForSemanticSegmentation,yi=d.SegformerModel,ki=d.SegformerPreTrainedModel,Fi=d.Seq2SeqLMOutput,Pi=d.SequenceClassifierOutput,Ci=d.SiglipImageProcessor,vi=d.SiglipModel,Si=d.SiglipPreTrainedModel,Ai=d.SiglipTextModel,Ei=d.SiglipTokenizer,Li=d.SiglipVisionModel,zi=d.SpeechT5FeatureExtractor,Ii=d.SpeechT5ForSpeechToText,Bi=d.SpeechT5ForTextToSpeech,Ni=d.SpeechT5HifiGan,Di=d.SpeechT5Model,Oi=d.SpeechT5PreTrainedModel,Vi=d.SpeechT5Processor,ji=d.SpeechT5Tokenizer,Ri=d.SqueezeBertForMaskedLM,Gi=d.SqueezeBertForQuestionAnswering,qi=d.SqueezeBertForSequenceClassification,Wi=d.SqueezeBertModel,$i=d.SqueezeBertPreTrainedModel,Ui=d.SqueezeBertTokenizer,Xi=d.StableLmForCausalLM,Qi=d.StableLmModel,Hi=d.StableLmPreTrainedModel,Yi=d.Starcoder2ForCausalLM,Ji=d.Starcoder2Model,Ki=d.Starcoder2PreTrainedModel,Zi=d.StoppingCriteria,el=d.StoppingCriteriaList,tl=d.SummarizationPipeline,rl=d.Swin2SRForImageSuperResolution,nl=d.Swin2SRImageProcessor,ol=d.Swin2SRModel,sl=d.Swin2SRPreTrainedModel,al=d.SwinForImageClassification,il=d.SwinModel,ll=d.SwinPreTrainedModel,cl=d.T5ForConditionalGeneration,dl=d.T5Model,ul=d.T5PreTrainedModel,pl=d.T5Tokenizer,hl=d.TableTransformerForObjectDetection,ml=d.TableTransformerModel,_l=d.TableTransformerObjectDetectionOutput,fl=d.TableTransformerPreTrainedModel,gl=d.Tensor,Ml=d.Text2TextGenerationPipeline,wl=d.TextClassificationPipeline,Tl=d.TextGenerationPipeline,bl=d.TextStreamer,xl=d.TextToAudioPipeline,yl=d.TokenClassificationPipeline,kl=d.TokenClassifierOutput,Fl=d.TokenizerModel,Pl=d.TrOCRForCausalLM,Cl=d.TrOCRPreTrainedModel,vl=d.TranslationPipeline,Sl=d.UniSpeechForCTC,Al=d.UniSpeechForSequenceClassification,El=d.UniSpeechModel,Ll=d.UniSpeechPreTrainedModel,zl=d.UniSpeechSatForAudioFrameClassification,Il=d.UniSpeechSatForCTC,Bl=d.UniSpeechSatForSequenceClassification,Nl=d.UniSpeechSatModel,Dl=d.UniSpeechSatPreTrainedModel,Ol=d.ViTFeatureExtractor,Vl=d.ViTForImageClassification,jl=d.ViTImageProcessor,Rl=d.ViTMAEModel,Gl=d.ViTMAEPreTrainedModel,ql=d.ViTMSNForImageClassification,Wl=d.ViTMSNModel,$l=d.ViTMSNPreTrainedModel,Ul=d.ViTModel,Xl=d.ViTPreTrainedModel,Ql=d.VisionEncoderDecoderModel,Hl=d.VitMatteForImageMatting,Yl=d.VitMatteImageProcessor,Jl=d.VitMattePreTrainedModel,Kl=d.VitsModel,Zl=d.VitsModelOutput,ec=d.VitsPreTrainedModel,tc=d.VitsTokenizer,rc=d.Wav2Vec2BertForCTC,nc=d.Wav2Vec2BertForSequenceClassification,oc=d.Wav2Vec2BertModel,sc=d.Wav2Vec2BertPreTrainedModel,ac=d.Wav2Vec2CTCTokenizer,ic=d.Wav2Vec2FeatureExtractor,lc=d.Wav2Vec2ForAudioFrameClassification,cc=d.Wav2Vec2ForCTC,dc=d.Wav2Vec2ForSequenceClassification,uc=d.Wav2Vec2Model,pc=d.Wav2Vec2PreTrainedModel,hc=d.Wav2Vec2ProcessorWithLM,mc=d.WavLMForAudioFrameClassification,_c=d.WavLMForCTC,fc=d.WavLMForSequenceClassification,gc=d.WavLMForXVector,Mc=d.WavLMModel,wc=d.WavLMPreTrainedModel,Tc=d.WeSpeakerFeatureExtractor,bc=d.WeSpeakerResNetModel,xc=d.WeSpeakerResNetPreTrainedModel,yc=d.WhisperFeatureExtractor,kc=d.WhisperForConditionalGeneration,Fc=d.WhisperModel,Pc=d.WhisperPreTrainedModel,Cc=d.WhisperProcessor,vc=d.WhisperTextStreamer,Sc=d.WhisperTokenizer,Ac=d.XLMForQuestionAnswering,Ec=d.XLMForSequenceClassification,Lc=d.XLMForTokenClassification,zc=d.XLMModel,Ic=d.XLMPreTrainedModel,Bc=d.XLMRobertaForMaskedLM,Nc=d.XLMRobertaForQuestionAnswering,Dc=d.XLMRobertaForSequenceClassification,Oc=d.XLMRobertaForTokenClassification,Vc=d.XLMRobertaModel,jc=d.XLMRobertaPreTrainedModel,Rc=d.XLMRobertaTokenizer,Gc=d.XLMTokenizer,qc=d.XLMWithLMHeadModel,Wc=d.XVectorOutput,$c=d.YolosFeatureExtractor,Uc=d.YolosForObjectDetection,Xc=d.YolosModel,Qc=d.YolosObjectDetectionOutput,Hc=d.YolosPreTrainedModel,Yc=d.ZeroShotAudioClassificationPipeline,Jc=d.ZeroShotClassificationPipeline,Kc=d.ZeroShotImageClassificationPipeline,Zc=d.ZeroShotObjectDetectionPipeline,ed=d.bankers_round,td=d.cat,rd=d.cos_sim,nd=d.dot,od=d.dynamic_time_warping,sd=d.env,ad=d.full,id=d.full_like,ld=d.getKeyValueShapes,cd=d.hamming,dd=d.hanning,ud=d.interpolate,pd=d.interpolate_4d,hd=d.interpolate_data,md=d.is_chinese_char,_d=d.layer_norm,fd=d.log_softmax,gd=d.magnitude,Md=d.matmul,wd=d.max,Td=d.mean,bd=d.mean_pooling,xd=d.medianFilter,yd=d.mel_filter_bank,kd=d.min,Fd=d.ones,Pd=d.ones_like,Cd=d.permute,vd=d.permute_data,Sd=d.pipeline,Ad=d.quantize_embeddings,Ed=d.read_audio,Ld=d.rfft,zd=d.round,Id=d.softmax,Bd=d.spectrogram,Nd=d.stack,Dd=d.std_mean,Od=d.topk,Vd=d.window_function,jd=d.zeros,Rd=d.zeros_like;export{y as ASTFeatureExtractor,k as ASTForAudioClassification,F as ASTModel,P as ASTPreTrainedModel,C as AlbertForMaskedLM,v as AlbertForQuestionAnswering,S as AlbertForSequenceClassification,A as AlbertModel,E as AlbertPreTrainedModel,L as AlbertTokenizer,z as AudioClassificationPipeline,I as AutoConfig,B as AutoModel,N as AutoModelForAudioClassification,D as AutoModelForAudioFrameClassification,O as AutoModelForCTC,V as AutoModelForCausalLM,j as AutoModelForDepthEstimation,R as AutoModelForDocumentQuestionAnswering,G as AutoModelForImageClassification,q as AutoModelForImageFeatureExtraction,W as AutoModelForImageMatting,$ as AutoModelForImageSegmentation,U as AutoModelForImageToImage,X as AutoModelForMaskGeneration,Q as AutoModelForMaskedLM,H as AutoModelForNormalEstimation,Y as AutoModelForObjectDetection,J as AutoModelForQuestionAnswering,K as AutoModelForSemanticSegmentation,Z as AutoModelForSeq2SeqLM,ee as AutoModelForSequenceClassification,te as AutoModelForSpeechSeq2Seq,re as AutoModelForTextToSpectrogram,ne as AutoModelForTextToWaveform,oe as AutoModelForTokenClassification,se as AutoModelForUniversalSegmentation,ae as AutoModelForVision2Seq,ie as AutoModelForXVector,le as AutoModelForZeroShotObjectDetection,ce as AutoProcessor,de as AutoTokenizer,ue as AutomaticSpeechRecognitionPipeline,pe as BartForConditionalGeneration,he as BartForSequenceClassification,me as BartModel,_e as BartPretrainedModel,fe as BartTokenizer,ge as BaseModelOutput,Me as BaseStreamer,we as BeitFeatureExtractor,Te as BeitForImageClassification,be as BeitModel,xe as BeitPreTrainedModel,ye as BertForMaskedLM,ke as BertForQuestionAnswering,Fe as BertForSequenceClassification,Pe as BertForTokenClassification,Ce as BertModel,ve as BertPreTrainedModel,Se as BertTokenizer,Ae as BitImageProcessor,Ee as BlenderbotForConditionalGeneration,Le as BlenderbotModel,ze as BlenderbotPreTrainedModel,Ie as BlenderbotSmallForConditionalGeneration,Be as BlenderbotSmallModel,Ne as BlenderbotSmallPreTrainedModel,De as BlenderbotSmallTokenizer,Oe as BlenderbotTokenizer,Ve as BloomForCausalLM,je as BloomModel,Re as BloomPreTrainedModel,Ge as BloomTokenizer,qe as CLIPFeatureExtractor,We as CLIPImageProcessor,$e as CLIPModel,Ue as CLIPPreTrainedModel,Xe as CLIPSegForImageSegmentation,Qe as CLIPSegModel,He as CLIPSegPreTrainedModel,Ye as CLIPTextModel,Je as CLIPTextModelWithProjection,Ke as CLIPTokenizer,Ze as CLIPVisionModel,et as CLIPVisionModelWithProjection,tt as CamembertForMaskedLM,rt as CamembertForQuestionAnswering,nt as CamembertForSequenceClassification,ot as CamembertForTokenClassification,st as CamembertModel,at as CamembertPreTrainedModel,it as CamembertTokenizer,lt as CausalLMOutput,ct as CausalLMOutputWithPast,dt as ChineseCLIPFeatureExtractor,ut as ChineseCLIPModel,pt as ChineseCLIPPreTrainedModel,ht as ClapAudioModelWithProjection,mt as ClapFeatureExtractor,_t as ClapModel,ft as ClapPreTrainedModel,gt as ClapTextModelWithProjection,Mt as CodeGenForCausalLM,wt as CodeGenModel,Tt as CodeGenPreTrainedModel,bt as CodeGenTokenizer,xt as CodeLlamaTokenizer,yt as CohereForCausalLM,kt as CohereModel,Ft as CoherePreTrainedModel,Pt as CohereTokenizer,Ct as ConvBertForMaskedLM,vt as ConvBertForQuestionAnswering,St as ConvBertForSequenceClassification,At as ConvBertForTokenClassification,Et as ConvBertModel,Lt as ConvBertPreTrainedModel,zt as ConvBertTokenizer,It as ConvNextFeatureExtractor,Bt as ConvNextForImageClassification,Nt as ConvNextImageProcessor,Dt as ConvNextModel,Ot as ConvNextPreTrainedModel,Vt as ConvNextV2ForImageClassification,jt as ConvNextV2Model,Rt as ConvNextV2PreTrainedModel,Gt as DPTFeatureExtractor,qt as DPTForDepthEstimation,Wt as DPTImageProcessor,$t as DPTModel,Ut as DPTPreTrainedModel,Xt as DebertaForMaskedLM,Qt as DebertaForQuestionAnswering,Ht as DebertaForSequenceClassification,Yt as DebertaForTokenClassification,Jt as DebertaModel,Kt as DebertaPreTrainedModel,Zt as DebertaTokenizer,er as DebertaV2ForMaskedLM,tr as DebertaV2ForQuestionAnswering,rr as DebertaV2ForSequenceClassification,nr as DebertaV2ForTokenClassification,or as DebertaV2Model,sr as DebertaV2PreTrainedModel,ar as DebertaV2Tokenizer,ir as DecisionTransformerModel,lr as DecisionTransformerPreTrainedModel,cr as DeiTFeatureExtractor,dr as DeiTForImageClassification,ur as DeiTModel,pr as DeiTPreTrainedModel,hr as DepthAnythingForDepthEstimation,mr as DepthAnythingPreTrainedModel,_r as DepthEstimationPipeline,fr as DepthProForDepthEstimation,gr as DepthProPreTrainedModel,Mr as DetrFeatureExtractor,wr as DetrForObjectDetection,Tr as DetrForSegmentation,br as DetrModel,xr as DetrObjectDetectionOutput,yr as DetrPreTrainedModel,kr as DetrSegmentationOutput,Fr as Dinov2ForImageClassification,Pr as Dinov2Model,Cr as Dinov2PreTrainedModel,vr as DistilBertForMaskedLM,Sr as DistilBertForQuestionAnswering,Ar as DistilBertForSequenceClassification,Er as DistilBertForTokenClassification,Lr as DistilBertModel,zr as DistilBertPreTrainedModel,Ir as DistilBertTokenizer,Br as DocumentQuestionAnsweringPipeline,Nr as DonutFeatureExtractor,Dr as DonutSwinModel,Or as DonutSwinPreTrainedModel,Vr as EfficientNetForImageClassification,jr as EfficientNetImageProcessor,Rr as EfficientNetModel,Gr as EfficientNetPreTrainedModel,qr as ElectraForMaskedLM,Wr as ElectraForQuestionAnswering,$r as ElectraForSequenceClassification,Ur as ElectraForTokenClassification,Xr as ElectraModel,Qr as ElectraPreTrainedModel,Hr as ElectraTokenizer,Yr as EosTokenCriteria,Jr as EsmForMaskedLM,Kr as EsmForSequenceClassification,Zr as EsmForTokenClassification,en as EsmModel,tn as EsmPreTrainedModel,rn as EsmTokenizer,nn as FFT,on as FalconForCausalLM,sn as FalconModel,an as FalconPreTrainedModel,ln as FalconTokenizer,cn as FastViTForImageClassification,dn as FastViTModel,un as FastViTPreTrainedModel,pn as FeatureExtractionPipeline,hn as FeatureExtractor,mn as FillMaskPipeline,_n as Florence2ForConditionalGeneration,fn as Florence2PreTrainedModel,gn as Florence2Processor,Mn as GLPNFeatureExtractor,wn as GLPNForDepthEstimation,Tn as GLPNModel,bn as GLPNPreTrainedModel,xn as GPT2LMHeadModel,yn as GPT2Model,kn as GPT2PreTrainedModel,Fn as GPT2Tokenizer,Pn as GPTBigCodeForCausalLM,Cn as GPTBigCodeModel,vn as GPTBigCodePreTrainedModel,Sn as GPTJForCausalLM,An as GPTJModel,En as GPTJPreTrainedModel,Ln as GPTNeoForCausalLM,zn as GPTNeoModel,In as GPTNeoPreTrainedModel,Bn as GPTNeoXForCausalLM,Nn as GPTNeoXModel,Dn as GPTNeoXPreTrainedModel,On as GPTNeoXTokenizer,Vn as Gemma2ForCausalLM,jn as Gemma2Model,Rn as Gemma2PreTrainedModel,Gn as GemmaForCausalLM,qn as GemmaModel,Wn as GemmaPreTrainedModel,$n as GemmaTokenizer,Un as GraniteForCausalLM,Xn as GraniteModel,Qn as GranitePreTrainedModel,Hn as Grok1Tokenizer,Yn as GroupViTModel,Jn as GroupViTPreTrainedModel,Kn as HerbertTokenizer,Zn as HieraForImageClassification,eo as HieraModel,to as HieraPreTrainedModel,ro as HubertForCTC,no as HubertForSequenceClassification,oo as HubertModel,so as HubertPreTrainedModel,ao as ImageClassificationPipeline,io as ImageFeatureExtractionPipeline,lo as ImageFeatureExtractor,co as ImageMattingOutput,uo as ImageSegmentationPipeline,po as ImageToImagePipeline,ho as ImageToTextPipeline,mo as InterruptableStoppingCriteria,_o as JAISLMHeadModel,fo as JAISModel,go as JAISPreTrainedModel,Mo as LlamaForCausalLM,wo as LlamaModel,To as LlamaPreTrainedModel,bo as LlamaTokenizer,xo as LlavaForConditionalGeneration,yo as LlavaPreTrainedModel,ko as LongT5ForConditionalGeneration,Fo as LongT5Model,Po as LongT5PreTrainedModel,Co as M2M100ForConditionalGeneration,vo as M2M100Model,So as M2M100PreTrainedModel,Ao as M2M100Tokenizer,Eo as MBart50Tokenizer,Lo as MBartForCausalLM,zo as MBartForConditionalGeneration,Io as MBartForSequenceClassification,Bo as MBartModel,No as MBartPreTrainedModel,Do as MBartTokenizer,Oo as MPNetForMaskedLM,Vo as MPNetForQuestionAnswering,jo as MPNetForSequenceClassification,Ro as MPNetForTokenClassification,Go as MPNetModel,qo as MPNetPreTrainedModel,Wo as MPNetTokenizer,$o as MT5ForConditionalGeneration,Uo as MT5Model,Xo as MT5PreTrainedModel,Qo as MarianMTModel,Ho as MarianModel,Yo as MarianPreTrainedModel,Jo as MarianTokenizer,Ko as MaskFormerFeatureExtractor,Zo as MaskFormerForInstanceSegmentation,es as MaskFormerModel,ts as MaskFormerPreTrainedModel,rs as MaskedLMOutput,ns as MaxLengthCriteria,os as MistralForCausalLM,ss as MistralModel,as as MistralPreTrainedModel,is as MobileBertForMaskedLM,ls as MobileBertForQuestionAnswering,cs as MobileBertForSequenceClassification,ds as MobileBertModel,us as MobileBertPreTrainedModel,ps as MobileBertTokenizer,hs as MobileNetV1FeatureExtractor,ms as MobileNetV1ForImageClassification,_s as MobileNetV1Model,fs as MobileNetV1PreTrainedModel,gs as MobileNetV2FeatureExtractor,Ms as MobileNetV2ForImageClassification,ws as MobileNetV2Model,Ts as MobileNetV2PreTrainedModel,bs as MobileNetV3FeatureExtractor,xs as MobileNetV3ForImageClassification,ys as MobileNetV3Model,ks as MobileNetV3PreTrainedModel,Fs as MobileNetV4FeatureExtractor,Ps as MobileNetV4ForImageClassification,Cs as MobileNetV4Model,vs as MobileNetV4PreTrainedModel,Ss as MobileViTFeatureExtractor,As as MobileViTForImageClassification,Es as MobileViTImageProcessor,Ls as MobileViTModel,zs as MobileViTPreTrainedModel,Is as MobileViTV2ForImageClassification,Bs as MobileViTV2Model,Ns as MobileViTV2PreTrainedModel,Ds as ModelOutput,Os as Moondream1ForConditionalGeneration,Vs as MptForCausalLM,js as MptModel,Rs as MptPreTrainedModel,Gs as MusicgenForCausalLM,qs as MusicgenForConditionalGeneration,Ws as MusicgenModel,$s as MusicgenPreTrainedModel,Us as NllbTokenizer,Xs as NomicBertModel,Qs as NomicBertPreTrainedModel,Hs as NougatImageProcessor,Ys as NougatTokenizer,Js as OPTForCausalLM,Ks as OPTModel,Zs as OPTPreTrainedModel,ea as ObjectDetectionPipeline,ta as OpenELMForCausalLM,ra as OpenELMModel,na as OpenELMPreTrainedModel,oa as OwlViTFeatureExtractor,sa as OwlViTForObjectDetection,aa as OwlViTModel,ia as OwlViTPreTrainedModel,la as OwlViTProcessor,ca as Owlv2ForObjectDetection,da as Owlv2ImageProcessor,ua as Owlv2Model,pa as Owlv2PreTrainedModel,ha as Phi3ForCausalLM,ma as Phi3Model,_a as Phi3PreTrainedModel,fa as PhiForCausalLM,ga as PhiModel,Ma as PhiPreTrainedModel,wa as Pipeline,Ta as PreTrainedModel,ba as PreTrainedTokenizer,xa as PretrainedConfig,ya as PretrainedMixin,ka as Processor,Fa as PvtForImageClassification,Pa as PvtImageProcessor,Ca as PvtModel,va as PvtPreTrainedModel,Sa as PyAnnoteFeatureExtractor,Aa as PyAnnoteForAudioFrameClassification,Ea as PyAnnoteModel,La as PyAnnotePreTrainedModel,za as PyAnnoteProcessor,Ia as QuestionAnsweringModelOutput,Ba as QuestionAnsweringPipeline,Na as Qwen2ForCausalLM,Da as Qwen2Model,Oa as Qwen2PreTrainedModel,Va as Qwen2Tokenizer,ja as RTDetrForObjectDetection,Ra as RTDetrImageProcessor,Ga as RTDetrModel,qa as RTDetrObjectDetectionOutput,Wa as RTDetrPreTrainedModel,$a as RawImage,Ua as ResNetForImageClassification,Xa as ResNetModel,Qa as ResNetPreTrainedModel,Ha as RoFormerForMaskedLM,Ya as RoFormerForQuestionAnswering,Ja as RoFormerForSequenceClassification,Ka as RoFormerForTokenClassification,Za as RoFormerModel,ei as RoFormerPreTrainedModel,ti as RoFormerTokenizer,ri as RobertaForMaskedLM,ni as RobertaForQuestionAnswering,oi as RobertaForSequenceClassification,si as RobertaForTokenClassification,ai as RobertaModel,ii as RobertaPreTrainedModel,li as RobertaTokenizer,ci as SamImageProcessor,di as SamImageSegmentationOutput,ui as SamModel,pi as SamPreTrainedModel,hi as SamProcessor,mi as SapiensFeatureExtractor,_i as SapiensForDepthEstimation,fi as SapiensForNormalEstimation,gi as SapiensForSemanticSegmentation,Mi as SapiensPreTrainedModel,wi as SeamlessM4TFeatureExtractor,Ti as SegformerFeatureExtractor,bi as SegformerForImageClassification,xi as SegformerForSemanticSegmentation,yi as SegformerModel,ki as SegformerPreTrainedModel,Fi as Seq2SeqLMOutput,Pi as SequenceClassifierOutput,Ci as SiglipImageProcessor,vi as SiglipModel,Si as SiglipPreTrainedModel,Ai as SiglipTextModel,Ei as SiglipTokenizer,Li as SiglipVisionModel,zi as SpeechT5FeatureExtractor,Ii as SpeechT5ForSpeechToText,Bi as SpeechT5ForTextToSpeech,Ni as SpeechT5HifiGan,Di as SpeechT5Model,Oi as SpeechT5PreTrainedModel,Vi as SpeechT5Processor,ji as SpeechT5Tokenizer,Ri as SqueezeBertForMaskedLM,Gi as SqueezeBertForQuestionAnswering,qi as SqueezeBertForSequenceClassification,Wi as SqueezeBertModel,$i as SqueezeBertPreTrainedModel,Ui as SqueezeBertTokenizer,Xi as StableLmForCausalLM,Qi as StableLmModel,Hi as StableLmPreTrainedModel,Yi as Starcoder2ForCausalLM,Ji as Starcoder2Model,Ki as Starcoder2PreTrainedModel,Zi as StoppingCriteria,el as StoppingCriteriaList,tl as SummarizationPipeline,rl as Swin2SRForImageSuperResolution,nl as Swin2SRImageProcessor,ol as Swin2SRModel,sl as Swin2SRPreTrainedModel,al as SwinForImageClassification,il as SwinModel,ll as SwinPreTrainedModel,cl as T5ForConditionalGeneration,dl as T5Model,ul as T5PreTrainedModel,pl as T5Tokenizer,hl as TableTransformerForObjectDetection,ml as TableTransformerModel,_l as TableTransformerObjectDetectionOutput,fl as TableTransformerPreTrainedModel,gl as Tensor,Ml as Text2TextGenerationPipeline,wl as TextClassificationPipeline,Tl as TextGenerationPipeline,bl as TextStreamer,xl as TextToAudioPipeline,yl as TokenClassificationPipeline,kl as TokenClassifierOutput,Fl as TokenizerModel,Pl as TrOCRForCausalLM,Cl as TrOCRPreTrainedModel,vl as TranslationPipeline,Sl as UniSpeechForCTC,Al as UniSpeechForSequenceClassification,El as UniSpeechModel,Ll as UniSpeechPreTrainedModel,zl as UniSpeechSatForAudioFrameClassification,Il as UniSpeechSatForCTC,Bl as UniSpeechSatForSequenceClassification,Nl as UniSpeechSatModel,Dl as UniSpeechSatPreTrainedModel,Ol as ViTFeatureExtractor,Vl as ViTForImageClassification,jl as ViTImageProcessor,Rl as ViTMAEModel,Gl as ViTMAEPreTrainedModel,ql as ViTMSNForImageClassification,Wl as ViTMSNModel,$l as ViTMSNPreTrainedModel,Ul as ViTModel,Xl as ViTPreTrainedModel,Ql as VisionEncoderDecoderModel,Hl as VitMatteForImageMatting,Yl as VitMatteImageProcessor,Jl as VitMattePreTrainedModel,Kl as VitsModel,Zl as VitsModelOutput,ec as VitsPreTrainedModel,tc as VitsTokenizer,rc as Wav2Vec2BertForCTC,nc as Wav2Vec2BertForSequenceClassification,oc as Wav2Vec2BertModel,sc as Wav2Vec2BertPreTrainedModel,ac as Wav2Vec2CTCTokenizer,ic as Wav2Vec2FeatureExtractor,lc as Wav2Vec2ForAudioFrameClassification,cc as Wav2Vec2ForCTC,dc as Wav2Vec2ForSequenceClassification,uc as Wav2Vec2Model,pc as Wav2Vec2PreTrainedModel,hc as Wav2Vec2ProcessorWithLM,mc as WavLMForAudioFrameClassification,_c as WavLMForCTC,fc as WavLMForSequenceClassification,gc as WavLMForXVector,Mc as WavLMModel,wc as WavLMPreTrainedModel,Tc as WeSpeakerFeatureExtractor,bc as WeSpeakerResNetModel,xc as WeSpeakerResNetPreTrainedModel,yc as WhisperFeatureExtractor,kc as WhisperForConditionalGeneration,Fc as WhisperModel,Pc as WhisperPreTrainedModel,Cc as WhisperProcessor,vc as WhisperTextStreamer,Sc as WhisperTokenizer,Ac as XLMForQuestionAnswering,Ec as XLMForSequenceClassification,Lc as XLMForTokenClassification,zc as XLMModel,Ic as XLMPreTrainedModel,Bc as XLMRobertaForMaskedLM,Nc as XLMRobertaForQuestionAnswering,Dc as XLMRobertaForSequenceClassification,Oc as XLMRobertaForTokenClassification,Vc as XLMRobertaModel,jc as XLMRobertaPreTrainedModel,Rc as XLMRobertaTokenizer,Gc as XLMTokenizer,qc as XLMWithLMHeadModel,Wc as XVectorOutput,$c as YolosFeatureExtractor,Uc as YolosForObjectDetection,Xc as YolosModel,Qc as YolosObjectDetectionOutput,Hc as YolosPreTrainedModel,Yc as ZeroShotAudioClassificationPipeline,Jc as ZeroShotClassificationPipeline,Kc as ZeroShotImageClassificationPipeline,Zc as ZeroShotObjectDetectionPipeline,ed as bankers_round,td as cat,rd as cos_sim,nd as dot,od as dynamic_time_warping,sd as env,ad as full,id as full_like,ld as getKeyValueShapes,cd as hamming,dd as hanning,ud as interpolate,pd as interpolate_4d,hd as interpolate_data,md as is_chinese_char,_d as layer_norm,fd as log_softmax,gd as magnitude,Md as matmul,wd as max,Td as mean,bd as mean_pooling,xd as medianFilter,yd as mel_filter_bank,kd as min,Fd as ones,Pd as ones_like,Cd as permute,vd as permute_data,Sd as pipeline,Ad as quantize_embeddings,Ed as read_audio,Ld as rfft,zd as round,Id as softmax,Bd as spectrogram,Nd as stack,Dd as std_mean,Od as topk,Vd as window_function,jd as zeros,Rd as zeros_like};
+  \*****************************/c.r(d),c.d(d,{ASTFeatureExtractor:()=>_.ASTFeatureExtractor,ASTForAudioClassification:()=>h.ASTForAudioClassification,ASTModel:()=>h.ASTModel,ASTPreTrainedModel:()=>h.ASTPreTrainedModel,AlbertForMaskedLM:()=>h.AlbertForMaskedLM,AlbertForQuestionAnswering:()=>h.AlbertForQuestionAnswering,AlbertForSequenceClassification:()=>h.AlbertForSequenceClassification,AlbertModel:()=>h.AlbertModel,AlbertPreTrainedModel:()=>h.AlbertPreTrainedModel,AlbertTokenizer:()=>m.AlbertTokenizer,AudioClassificationPipeline:()=>p.AudioClassificationPipeline,AutoConfig:()=>f.AutoConfig,AutoModel:()=>h.AutoModel,AutoModelForAudioClassification:()=>h.AutoModelForAudioClassification,AutoModelForAudioFrameClassification:()=>h.AutoModelForAudioFrameClassification,AutoModelForCTC:()=>h.AutoModelForCTC,AutoModelForCausalLM:()=>h.AutoModelForCausalLM,AutoModelForDepthEstimation:()=>h.AutoModelForDepthEstimation,AutoModelForDocumentQuestionAnswering:()=>h.AutoModelForDocumentQuestionAnswering,AutoModelForImageClassification:()=>h.AutoModelForImageClassification,AutoModelForImageFeatureExtraction:()=>h.AutoModelForImageFeatureExtraction,AutoModelForImageMatting:()=>h.AutoModelForImageMatting,AutoModelForImageSegmentation:()=>h.AutoModelForImageSegmentation,AutoModelForImageToImage:()=>h.AutoModelForImageToImage,AutoModelForMaskGeneration:()=>h.AutoModelForMaskGeneration,AutoModelForMaskedLM:()=>h.AutoModelForMaskedLM,AutoModelForNormalEstimation:()=>h.AutoModelForNormalEstimation,AutoModelForObjectDetection:()=>h.AutoModelForObjectDetection,AutoModelForQuestionAnswering:()=>h.AutoModelForQuestionAnswering,AutoModelForSemanticSegmentation:()=>h.AutoModelForSemanticSegmentation,AutoModelForSeq2SeqLM:()=>h.AutoModelForSeq2SeqLM,AutoModelForSequenceClassification:()=>h.AutoModelForSequenceClassification,AutoModelForSpeechSeq2Seq:()=>h.AutoModelForSpeechSeq2Seq,AutoModelForTextToSpectrogram:()=>h.AutoModelForTextToSpectrogram,AutoModelForTextToWaveform:()=>h.AutoModelForTextToWaveform,AutoModelForTokenClassification:()=>h.AutoModelForTokenClassification,AutoModelForUniversalSegmentation:()=>h.AutoModelForUniversalSegmentation,AutoModelForVision2Seq:()=>h.AutoModelForVision2Seq,AutoModelForXVector:()=>h.AutoModelForXVector,AutoModelForZeroShotObjectDetection:()=>h.AutoModelForZeroShotObjectDetection,AutoProcessor:()=>_.AutoProcessor,AutoTokenizer:()=>m.AutoTokenizer,AutomaticSpeechRecognitionPipeline:()=>p.AutomaticSpeechRecognitionPipeline,BartForConditionalGeneration:()=>h.BartForConditionalGeneration,BartForSequenceClassification:()=>h.BartForSequenceClassification,BartModel:()=>h.BartModel,BartPretrainedModel:()=>h.BartPretrainedModel,BartTokenizer:()=>m.BartTokenizer,BaseModelOutput:()=>h.BaseModelOutput,BaseStreamer:()=>b.BaseStreamer,BeitFeatureExtractor:()=>_.BeitFeatureExtractor,BeitForImageClassification:()=>h.BeitForImageClassification,BeitModel:()=>h.BeitModel,BeitPreTrainedModel:()=>h.BeitPreTrainedModel,BertForMaskedLM:()=>h.BertForMaskedLM,BertForQuestionAnswering:()=>h.BertForQuestionAnswering,BertForSequenceClassification:()=>h.BertForSequenceClassification,BertForTokenClassification:()=>h.BertForTokenClassification,BertModel:()=>h.BertModel,BertPreTrainedModel:()=>h.BertPreTrainedModel,BertTokenizer:()=>m.BertTokenizer,BitImageProcessor:()=>_.BitImageProcessor,BlenderbotForConditionalGeneration:()=>h.BlenderbotForConditionalGeneration,BlenderbotModel:()=>h.BlenderbotModel,BlenderbotPreTrainedModel:()=>h.BlenderbotPreTrainedModel,BlenderbotSmallForConditionalGeneration:()=>h.BlenderbotSmallForConditionalGeneration,BlenderbotSmallModel:()=>h.BlenderbotSmallModel,BlenderbotSmallPreTrainedModel:()=>h.BlenderbotSmallPreTrainedModel,BlenderbotSmallTokenizer:()=>m.BlenderbotSmallTokenizer,BlenderbotTokenizer:()=>m.BlenderbotTokenizer,BloomForCausalLM:()=>h.BloomForCausalLM,BloomModel:()=>h.BloomModel,BloomPreTrainedModel:()=>h.BloomPreTrainedModel,BloomTokenizer:()=>m.BloomTokenizer,CLIPFeatureExtractor:()=>_.CLIPFeatureExtractor,CLIPImageProcessor:()=>_.CLIPImageProcessor,CLIPModel:()=>h.CLIPModel,CLIPPreTrainedModel:()=>h.CLIPPreTrainedModel,CLIPSegForImageSegmentation:()=>h.CLIPSegForImageSegmentation,CLIPSegModel:()=>h.CLIPSegModel,CLIPSegPreTrainedModel:()=>h.CLIPSegPreTrainedModel,CLIPTextModel:()=>h.CLIPTextModel,CLIPTextModelWithProjection:()=>h.CLIPTextModelWithProjection,CLIPTokenizer:()=>m.CLIPTokenizer,CLIPVisionModel:()=>h.CLIPVisionModel,CLIPVisionModelWithProjection:()=>h.CLIPVisionModelWithProjection,CamembertForMaskedLM:()=>h.CamembertForMaskedLM,CamembertForQuestionAnswering:()=>h.CamembertForQuestionAnswering,CamembertForSequenceClassification:()=>h.CamembertForSequenceClassification,CamembertForTokenClassification:()=>h.CamembertForTokenClassification,CamembertModel:()=>h.CamembertModel,CamembertPreTrainedModel:()=>h.CamembertPreTrainedModel,CamembertTokenizer:()=>m.CamembertTokenizer,CausalLMOutput:()=>h.CausalLMOutput,CausalLMOutputWithPast:()=>h.CausalLMOutputWithPast,ChineseCLIPFeatureExtractor:()=>_.ChineseCLIPFeatureExtractor,ChineseCLIPModel:()=>h.ChineseCLIPModel,ChineseCLIPPreTrainedModel:()=>h.ChineseCLIPPreTrainedModel,ClapAudioModelWithProjection:()=>h.ClapAudioModelWithProjection,ClapFeatureExtractor:()=>_.ClapFeatureExtractor,ClapModel:()=>h.ClapModel,ClapPreTrainedModel:()=>h.ClapPreTrainedModel,ClapTextModelWithProjection:()=>h.ClapTextModelWithProjection,CodeGenForCausalLM:()=>h.CodeGenForCausalLM,CodeGenModel:()=>h.CodeGenModel,CodeGenPreTrainedModel:()=>h.CodeGenPreTrainedModel,CodeGenTokenizer:()=>m.CodeGenTokenizer,CodeLlamaTokenizer:()=>m.CodeLlamaTokenizer,CohereForCausalLM:()=>h.CohereForCausalLM,CohereModel:()=>h.CohereModel,CoherePreTrainedModel:()=>h.CoherePreTrainedModel,CohereTokenizer:()=>m.CohereTokenizer,ConvBertForMaskedLM:()=>h.ConvBertForMaskedLM,ConvBertForQuestionAnswering:()=>h.ConvBertForQuestionAnswering,ConvBertForSequenceClassification:()=>h.ConvBertForSequenceClassification,ConvBertForTokenClassification:()=>h.ConvBertForTokenClassification,ConvBertModel:()=>h.ConvBertModel,ConvBertPreTrainedModel:()=>h.ConvBertPreTrainedModel,ConvBertTokenizer:()=>m.ConvBertTokenizer,ConvNextFeatureExtractor:()=>_.ConvNextFeatureExtractor,ConvNextForImageClassification:()=>h.ConvNextForImageClassification,ConvNextImageProcessor:()=>_.ConvNextImageProcessor,ConvNextModel:()=>h.ConvNextModel,ConvNextPreTrainedModel:()=>h.ConvNextPreTrainedModel,ConvNextV2ForImageClassification:()=>h.ConvNextV2ForImageClassification,ConvNextV2Model:()=>h.ConvNextV2Model,ConvNextV2PreTrainedModel:()=>h.ConvNextV2PreTrainedModel,DPTFeatureExtractor:()=>_.DPTFeatureExtractor,DPTForDepthEstimation:()=>h.DPTForDepthEstimation,DPTImageProcessor:()=>_.DPTImageProcessor,DPTModel:()=>h.DPTModel,DPTPreTrainedModel:()=>h.DPTPreTrainedModel,DebertaForMaskedLM:()=>h.DebertaForMaskedLM,DebertaForQuestionAnswering:()=>h.DebertaForQuestionAnswering,DebertaForSequenceClassification:()=>h.DebertaForSequenceClassification,DebertaForTokenClassification:()=>h.DebertaForTokenClassification,DebertaModel:()=>h.DebertaModel,DebertaPreTrainedModel:()=>h.DebertaPreTrainedModel,DebertaTokenizer:()=>m.DebertaTokenizer,DebertaV2ForMaskedLM:()=>h.DebertaV2ForMaskedLM,DebertaV2ForQuestionAnswering:()=>h.DebertaV2ForQuestionAnswering,DebertaV2ForSequenceClassification:()=>h.DebertaV2ForSequenceClassification,DebertaV2ForTokenClassification:()=>h.DebertaV2ForTokenClassification,DebertaV2Model:()=>h.DebertaV2Model,DebertaV2PreTrainedModel:()=>h.DebertaV2PreTrainedModel,DebertaV2Tokenizer:()=>m.DebertaV2Tokenizer,DecisionTransformerModel:()=>h.DecisionTransformerModel,DecisionTransformerPreTrainedModel:()=>h.DecisionTransformerPreTrainedModel,DeiTFeatureExtractor:()=>_.DeiTFeatureExtractor,DeiTForImageClassification:()=>h.DeiTForImageClassification,DeiTModel:()=>h.DeiTModel,DeiTPreTrainedModel:()=>h.DeiTPreTrainedModel,DepthAnythingForDepthEstimation:()=>h.DepthAnythingForDepthEstimation,DepthAnythingPreTrainedModel:()=>h.DepthAnythingPreTrainedModel,DepthEstimationPipeline:()=>p.DepthEstimationPipeline,DepthProForDepthEstimation:()=>h.DepthProForDepthEstimation,DepthProPreTrainedModel:()=>h.DepthProPreTrainedModel,DetrFeatureExtractor:()=>_.DetrFeatureExtractor,DetrForObjectDetection:()=>h.DetrForObjectDetection,DetrForSegmentation:()=>h.DetrForSegmentation,DetrModel:()=>h.DetrModel,DetrObjectDetectionOutput:()=>h.DetrObjectDetectionOutput,DetrPreTrainedModel:()=>h.DetrPreTrainedModel,DetrSegmentationOutput:()=>h.DetrSegmentationOutput,Dinov2ForImageClassification:()=>h.Dinov2ForImageClassification,Dinov2Model:()=>h.Dinov2Model,Dinov2PreTrainedModel:()=>h.Dinov2PreTrainedModel,DistilBertForMaskedLM:()=>h.DistilBertForMaskedLM,DistilBertForQuestionAnswering:()=>h.DistilBertForQuestionAnswering,DistilBertForSequenceClassification:()=>h.DistilBertForSequenceClassification,DistilBertForTokenClassification:()=>h.DistilBertForTokenClassification,DistilBertModel:()=>h.DistilBertModel,DistilBertPreTrainedModel:()=>h.DistilBertPreTrainedModel,DistilBertTokenizer:()=>m.DistilBertTokenizer,DocumentQuestionAnsweringPipeline:()=>p.DocumentQuestionAnsweringPipeline,DonutFeatureExtractor:()=>_.DonutFeatureExtractor,DonutImageProcessor:()=>_.DonutImageProcessor,DonutSwinModel:()=>h.DonutSwinModel,DonutSwinPreTrainedModel:()=>h.DonutSwinPreTrainedModel,EfficientNetForImageClassification:()=>h.EfficientNetForImageClassification,EfficientNetImageProcessor:()=>_.EfficientNetImageProcessor,EfficientNetModel:()=>h.EfficientNetModel,EfficientNetPreTrainedModel:()=>h.EfficientNetPreTrainedModel,ElectraForMaskedLM:()=>h.ElectraForMaskedLM,ElectraForQuestionAnswering:()=>h.ElectraForQuestionAnswering,ElectraForSequenceClassification:()=>h.ElectraForSequenceClassification,ElectraForTokenClassification:()=>h.ElectraForTokenClassification,ElectraModel:()=>h.ElectraModel,ElectraPreTrainedModel:()=>h.ElectraPreTrainedModel,ElectraTokenizer:()=>m.ElectraTokenizer,EosTokenCriteria:()=>x.EosTokenCriteria,EsmForMaskedLM:()=>h.EsmForMaskedLM,EsmForSequenceClassification:()=>h.EsmForSequenceClassification,EsmForTokenClassification:()=>h.EsmForTokenClassification,EsmModel:()=>h.EsmModel,EsmPreTrainedModel:()=>h.EsmPreTrainedModel,EsmTokenizer:()=>m.EsmTokenizer,FFT:()=>T.FFT,FalconForCausalLM:()=>h.FalconForCausalLM,FalconModel:()=>h.FalconModel,FalconPreTrainedModel:()=>h.FalconPreTrainedModel,FalconTokenizer:()=>m.FalconTokenizer,FastViTForImageClassification:()=>h.FastViTForImageClassification,FastViTModel:()=>h.FastViTModel,FastViTPreTrainedModel:()=>h.FastViTPreTrainedModel,FeatureExtractionPipeline:()=>p.FeatureExtractionPipeline,FeatureExtractor:()=>_.FeatureExtractor,FillMaskPipeline:()=>p.FillMaskPipeline,Florence2ForConditionalGeneration:()=>h.Florence2ForConditionalGeneration,Florence2PreTrainedModel:()=>h.Florence2PreTrainedModel,Florence2Processor:()=>_.Florence2Processor,GLPNFeatureExtractor:()=>_.GLPNFeatureExtractor,GLPNForDepthEstimation:()=>h.GLPNForDepthEstimation,GLPNModel:()=>h.GLPNModel,GLPNPreTrainedModel:()=>h.GLPNPreTrainedModel,GPT2LMHeadModel:()=>h.GPT2LMHeadModel,GPT2Model:()=>h.GPT2Model,GPT2PreTrainedModel:()=>h.GPT2PreTrainedModel,GPT2Tokenizer:()=>m.GPT2Tokenizer,GPTBigCodeForCausalLM:()=>h.GPTBigCodeForCausalLM,GPTBigCodeModel:()=>h.GPTBigCodeModel,GPTBigCodePreTrainedModel:()=>h.GPTBigCodePreTrainedModel,GPTJForCausalLM:()=>h.GPTJForCausalLM,GPTJModel:()=>h.GPTJModel,GPTJPreTrainedModel:()=>h.GPTJPreTrainedModel,GPTNeoForCausalLM:()=>h.GPTNeoForCausalLM,GPTNeoModel:()=>h.GPTNeoModel,GPTNeoPreTrainedModel:()=>h.GPTNeoPreTrainedModel,GPTNeoXForCausalLM:()=>h.GPTNeoXForCausalLM,GPTNeoXModel:()=>h.GPTNeoXModel,GPTNeoXPreTrainedModel:()=>h.GPTNeoXPreTrainedModel,GPTNeoXTokenizer:()=>m.GPTNeoXTokenizer,Gemma2ForCausalLM:()=>h.Gemma2ForCausalLM,Gemma2Model:()=>h.Gemma2Model,Gemma2PreTrainedModel:()=>h.Gemma2PreTrainedModel,GemmaForCausalLM:()=>h.GemmaForCausalLM,GemmaModel:()=>h.GemmaModel,GemmaPreTrainedModel:()=>h.GemmaPreTrainedModel,GemmaTokenizer:()=>m.GemmaTokenizer,GraniteForCausalLM:()=>h.GraniteForCausalLM,GraniteModel:()=>h.GraniteModel,GranitePreTrainedModel:()=>h.GranitePreTrainedModel,Grok1Tokenizer:()=>m.Grok1Tokenizer,GroupViTModel:()=>h.GroupViTModel,GroupViTPreTrainedModel:()=>h.GroupViTPreTrainedModel,HerbertTokenizer:()=>m.HerbertTokenizer,HieraForImageClassification:()=>h.HieraForImageClassification,HieraModel:()=>h.HieraModel,HieraPreTrainedModel:()=>h.HieraPreTrainedModel,HubertForCTC:()=>h.HubertForCTC,HubertForSequenceClassification:()=>h.HubertForSequenceClassification,HubertModel:()=>h.HubertModel,HubertPreTrainedModel:()=>h.HubertPreTrainedModel,ImageClassificationPipeline:()=>p.ImageClassificationPipeline,ImageFeatureExtractionPipeline:()=>p.ImageFeatureExtractionPipeline,ImageFeatureExtractor:()=>_.ImageFeatureExtractor,ImageMattingOutput:()=>h.ImageMattingOutput,ImageSegmentationPipeline:()=>p.ImageSegmentationPipeline,ImageToImagePipeline:()=>p.ImageToImagePipeline,ImageToTextPipeline:()=>p.ImageToTextPipeline,InterruptableStoppingCriteria:()=>x.InterruptableStoppingCriteria,JAISLMHeadModel:()=>h.JAISLMHeadModel,JAISModel:()=>h.JAISModel,JAISPreTrainedModel:()=>h.JAISPreTrainedModel,LlamaForCausalLM:()=>h.LlamaForCausalLM,LlamaModel:()=>h.LlamaModel,LlamaPreTrainedModel:()=>h.LlamaPreTrainedModel,LlamaTokenizer:()=>m.LlamaTokenizer,LlavaForConditionalGeneration:()=>h.LlavaForConditionalGeneration,LlavaPreTrainedModel:()=>h.LlavaPreTrainedModel,LongT5ForConditionalGeneration:()=>h.LongT5ForConditionalGeneration,LongT5Model:()=>h.LongT5Model,LongT5PreTrainedModel:()=>h.LongT5PreTrainedModel,M2M100ForConditionalGeneration:()=>h.M2M100ForConditionalGeneration,M2M100Model:()=>h.M2M100Model,M2M100PreTrainedModel:()=>h.M2M100PreTrainedModel,M2M100Tokenizer:()=>m.M2M100Tokenizer,MBart50Tokenizer:()=>m.MBart50Tokenizer,MBartForCausalLM:()=>h.MBartForCausalLM,MBartForConditionalGeneration:()=>h.MBartForConditionalGeneration,MBartForSequenceClassification:()=>h.MBartForSequenceClassification,MBartModel:()=>h.MBartModel,MBartPreTrainedModel:()=>h.MBartPreTrainedModel,MBartTokenizer:()=>m.MBartTokenizer,MPNetForMaskedLM:()=>h.MPNetForMaskedLM,MPNetForQuestionAnswering:()=>h.MPNetForQuestionAnswering,MPNetForSequenceClassification:()=>h.MPNetForSequenceClassification,MPNetForTokenClassification:()=>h.MPNetForTokenClassification,MPNetModel:()=>h.MPNetModel,MPNetPreTrainedModel:()=>h.MPNetPreTrainedModel,MPNetTokenizer:()=>m.MPNetTokenizer,MT5ForConditionalGeneration:()=>h.MT5ForConditionalGeneration,MT5Model:()=>h.MT5Model,MT5PreTrainedModel:()=>h.MT5PreTrainedModel,MarianMTModel:()=>h.MarianMTModel,MarianModel:()=>h.MarianModel,MarianPreTrainedModel:()=>h.MarianPreTrainedModel,MarianTokenizer:()=>m.MarianTokenizer,MaskFormerFeatureExtractor:()=>_.MaskFormerFeatureExtractor,MaskFormerForInstanceSegmentation:()=>h.MaskFormerForInstanceSegmentation,MaskFormerModel:()=>h.MaskFormerModel,MaskFormerPreTrainedModel:()=>h.MaskFormerPreTrainedModel,MaskedLMOutput:()=>h.MaskedLMOutput,MaxLengthCriteria:()=>x.MaxLengthCriteria,MistralForCausalLM:()=>h.MistralForCausalLM,MistralModel:()=>h.MistralModel,MistralPreTrainedModel:()=>h.MistralPreTrainedModel,MobileBertForMaskedLM:()=>h.MobileBertForMaskedLM,MobileBertForQuestionAnswering:()=>h.MobileBertForQuestionAnswering,MobileBertForSequenceClassification:()=>h.MobileBertForSequenceClassification,MobileBertModel:()=>h.MobileBertModel,MobileBertPreTrainedModel:()=>h.MobileBertPreTrainedModel,MobileBertTokenizer:()=>m.MobileBertTokenizer,MobileNetV1FeatureExtractor:()=>_.MobileNetV1FeatureExtractor,MobileNetV1ForImageClassification:()=>h.MobileNetV1ForImageClassification,MobileNetV1Model:()=>h.MobileNetV1Model,MobileNetV1PreTrainedModel:()=>h.MobileNetV1PreTrainedModel,MobileNetV2FeatureExtractor:()=>_.MobileNetV2FeatureExtractor,MobileNetV2ForImageClassification:()=>h.MobileNetV2ForImageClassification,MobileNetV2Model:()=>h.MobileNetV2Model,MobileNetV2PreTrainedModel:()=>h.MobileNetV2PreTrainedModel,MobileNetV3FeatureExtractor:()=>_.MobileNetV3FeatureExtractor,MobileNetV3ForImageClassification:()=>h.MobileNetV3ForImageClassification,MobileNetV3Model:()=>h.MobileNetV3Model,MobileNetV3PreTrainedModel:()=>h.MobileNetV3PreTrainedModel,MobileNetV4FeatureExtractor:()=>_.MobileNetV4FeatureExtractor,MobileNetV4ForImageClassification:()=>h.MobileNetV4ForImageClassification,MobileNetV4Model:()=>h.MobileNetV4Model,MobileNetV4PreTrainedModel:()=>h.MobileNetV4PreTrainedModel,MobileViTFeatureExtractor:()=>_.MobileViTFeatureExtractor,MobileViTForImageClassification:()=>h.MobileViTForImageClassification,MobileViTImageProcessor:()=>_.MobileViTImageProcessor,MobileViTModel:()=>h.MobileViTModel,MobileViTPreTrainedModel:()=>h.MobileViTPreTrainedModel,MobileViTV2ForImageClassification:()=>h.MobileViTV2ForImageClassification,MobileViTV2Model:()=>h.MobileViTV2Model,MobileViTV2PreTrainedModel:()=>h.MobileViTV2PreTrainedModel,ModelOutput:()=>h.ModelOutput,Moondream1ForConditionalGeneration:()=>h.Moondream1ForConditionalGeneration,MptForCausalLM:()=>h.MptForCausalLM,MptModel:()=>h.MptModel,MptPreTrainedModel:()=>h.MptPreTrainedModel,MusicgenForCausalLM:()=>h.MusicgenForCausalLM,MusicgenForConditionalGeneration:()=>h.MusicgenForConditionalGeneration,MusicgenModel:()=>h.MusicgenModel,MusicgenPreTrainedModel:()=>h.MusicgenPreTrainedModel,NllbTokenizer:()=>m.NllbTokenizer,NomicBertModel:()=>h.NomicBertModel,NomicBertPreTrainedModel:()=>h.NomicBertPreTrainedModel,NougatImageProcessor:()=>_.NougatImageProcessor,NougatTokenizer:()=>m.NougatTokenizer,OPTForCausalLM:()=>h.OPTForCausalLM,OPTModel:()=>h.OPTModel,OPTPreTrainedModel:()=>h.OPTPreTrainedModel,ObjectDetectionPipeline:()=>p.ObjectDetectionPipeline,OpenELMForCausalLM:()=>h.OpenELMForCausalLM,OpenELMModel:()=>h.OpenELMModel,OpenELMPreTrainedModel:()=>h.OpenELMPreTrainedModel,OwlViTFeatureExtractor:()=>_.OwlViTFeatureExtractor,OwlViTForObjectDetection:()=>h.OwlViTForObjectDetection,OwlViTModel:()=>h.OwlViTModel,OwlViTPreTrainedModel:()=>h.OwlViTPreTrainedModel,OwlViTProcessor:()=>_.OwlViTProcessor,Owlv2ForObjectDetection:()=>h.Owlv2ForObjectDetection,Owlv2ImageProcessor:()=>_.Owlv2ImageProcessor,Owlv2Model:()=>h.Owlv2Model,Owlv2PreTrainedModel:()=>h.Owlv2PreTrainedModel,Phi3ForCausalLM:()=>h.Phi3ForCausalLM,Phi3Model:()=>h.Phi3Model,Phi3PreTrainedModel:()=>h.Phi3PreTrainedModel,PhiForCausalLM:()=>h.PhiForCausalLM,PhiModel:()=>h.PhiModel,PhiPreTrainedModel:()=>h.PhiPreTrainedModel,Pipeline:()=>p.Pipeline,PreTrainedModel:()=>h.PreTrainedModel,PreTrainedTokenizer:()=>m.PreTrainedTokenizer,PretrainedConfig:()=>f.PretrainedConfig,PretrainedMixin:()=>h.PretrainedMixin,Processor:()=>_.Processor,PvtForImageClassification:()=>h.PvtForImageClassification,PvtImageProcessor:()=>_.PvtImageProcessor,PvtModel:()=>h.PvtModel,PvtPreTrainedModel:()=>h.PvtPreTrainedModel,PyAnnoteFeatureExtractor:()=>_.PyAnnoteFeatureExtractor,PyAnnoteForAudioFrameClassification:()=>h.PyAnnoteForAudioFrameClassification,PyAnnoteModel:()=>h.PyAnnoteModel,PyAnnotePreTrainedModel:()=>h.PyAnnotePreTrainedModel,PyAnnoteProcessor:()=>_.PyAnnoteProcessor,QuestionAnsweringModelOutput:()=>h.QuestionAnsweringModelOutput,QuestionAnsweringPipeline:()=>p.QuestionAnsweringPipeline,Qwen2ForCausalLM:()=>h.Qwen2ForCausalLM,Qwen2Model:()=>h.Qwen2Model,Qwen2PreTrainedModel:()=>h.Qwen2PreTrainedModel,Qwen2Tokenizer:()=>m.Qwen2Tokenizer,RTDetrForObjectDetection:()=>h.RTDetrForObjectDetection,RTDetrImageProcessor:()=>_.RTDetrImageProcessor,RTDetrModel:()=>h.RTDetrModel,RTDetrObjectDetectionOutput:()=>h.RTDetrObjectDetectionOutput,RTDetrPreTrainedModel:()=>h.RTDetrPreTrainedModel,RawImage:()=>M.RawImage,ResNetForImageClassification:()=>h.ResNetForImageClassification,ResNetModel:()=>h.ResNetModel,ResNetPreTrainedModel:()=>h.ResNetPreTrainedModel,RoFormerForMaskedLM:()=>h.RoFormerForMaskedLM,RoFormerForQuestionAnswering:()=>h.RoFormerForQuestionAnswering,RoFormerForSequenceClassification:()=>h.RoFormerForSequenceClassification,RoFormerForTokenClassification:()=>h.RoFormerForTokenClassification,RoFormerModel:()=>h.RoFormerModel,RoFormerPreTrainedModel:()=>h.RoFormerPreTrainedModel,RoFormerTokenizer:()=>m.RoFormerTokenizer,RobertaForMaskedLM:()=>h.RobertaForMaskedLM,RobertaForQuestionAnswering:()=>h.RobertaForQuestionAnswering,RobertaForSequenceClassification:()=>h.RobertaForSequenceClassification,RobertaForTokenClassification:()=>h.RobertaForTokenClassification,RobertaModel:()=>h.RobertaModel,RobertaPreTrainedModel:()=>h.RobertaPreTrainedModel,RobertaTokenizer:()=>m.RobertaTokenizer,SamImageProcessor:()=>_.SamImageProcessor,SamImageSegmentationOutput:()=>h.SamImageSegmentationOutput,SamModel:()=>h.SamModel,SamPreTrainedModel:()=>h.SamPreTrainedModel,SamProcessor:()=>_.SamProcessor,SapiensFeatureExtractor:()=>_.SapiensFeatureExtractor,SapiensForDepthEstimation:()=>h.SapiensForDepthEstimation,SapiensForNormalEstimation:()=>h.SapiensForNormalEstimation,SapiensForSemanticSegmentation:()=>h.SapiensForSemanticSegmentation,SapiensPreTrainedModel:()=>h.SapiensPreTrainedModel,SeamlessM4TFeatureExtractor:()=>_.SeamlessM4TFeatureExtractor,SegformerFeatureExtractor:()=>_.SegformerFeatureExtractor,SegformerForImageClassification:()=>h.SegformerForImageClassification,SegformerForSemanticSegmentation:()=>h.SegformerForSemanticSegmentation,SegformerModel:()=>h.SegformerModel,SegformerPreTrainedModel:()=>h.SegformerPreTrainedModel,Seq2SeqLMOutput:()=>h.Seq2SeqLMOutput,SequenceClassifierOutput:()=>h.SequenceClassifierOutput,SiglipImageProcessor:()=>_.SiglipImageProcessor,SiglipModel:()=>h.SiglipModel,SiglipPreTrainedModel:()=>h.SiglipPreTrainedModel,SiglipTextModel:()=>h.SiglipTextModel,SiglipTokenizer:()=>m.SiglipTokenizer,SiglipVisionModel:()=>h.SiglipVisionModel,SpeechT5FeatureExtractor:()=>_.SpeechT5FeatureExtractor,SpeechT5ForSpeechToText:()=>h.SpeechT5ForSpeechToText,SpeechT5ForTextToSpeech:()=>h.SpeechT5ForTextToSpeech,SpeechT5HifiGan:()=>h.SpeechT5HifiGan,SpeechT5Model:()=>h.SpeechT5Model,SpeechT5PreTrainedModel:()=>h.SpeechT5PreTrainedModel,SpeechT5Processor:()=>_.SpeechT5Processor,SpeechT5Tokenizer:()=>m.SpeechT5Tokenizer,SqueezeBertForMaskedLM:()=>h.SqueezeBertForMaskedLM,SqueezeBertForQuestionAnswering:()=>h.SqueezeBertForQuestionAnswering,SqueezeBertForSequenceClassification:()=>h.SqueezeBertForSequenceClassification,SqueezeBertModel:()=>h.SqueezeBertModel,SqueezeBertPreTrainedModel:()=>h.SqueezeBertPreTrainedModel,SqueezeBertTokenizer:()=>m.SqueezeBertTokenizer,StableLmForCausalLM:()=>h.StableLmForCausalLM,StableLmModel:()=>h.StableLmModel,StableLmPreTrainedModel:()=>h.StableLmPreTrainedModel,Starcoder2ForCausalLM:()=>h.Starcoder2ForCausalLM,Starcoder2Model:()=>h.Starcoder2Model,Starcoder2PreTrainedModel:()=>h.Starcoder2PreTrainedModel,StoppingCriteria:()=>x.StoppingCriteria,StoppingCriteriaList:()=>x.StoppingCriteriaList,SummarizationPipeline:()=>p.SummarizationPipeline,Swin2SRForImageSuperResolution:()=>h.Swin2SRForImageSuperResolution,Swin2SRImageProcessor:()=>_.Swin2SRImageProcessor,Swin2SRModel:()=>h.Swin2SRModel,Swin2SRPreTrainedModel:()=>h.Swin2SRPreTrainedModel,SwinForImageClassification:()=>h.SwinForImageClassification,SwinModel:()=>h.SwinModel,SwinPreTrainedModel:()=>h.SwinPreTrainedModel,T5ForConditionalGeneration:()=>h.T5ForConditionalGeneration,T5Model:()=>h.T5Model,T5PreTrainedModel:()=>h.T5PreTrainedModel,T5Tokenizer:()=>m.T5Tokenizer,TableTransformerForObjectDetection:()=>h.TableTransformerForObjectDetection,TableTransformerModel:()=>h.TableTransformerModel,TableTransformerObjectDetectionOutput:()=>h.TableTransformerObjectDetectionOutput,TableTransformerPreTrainedModel:()=>h.TableTransformerPreTrainedModel,Tensor:()=>w.Tensor,Text2TextGenerationPipeline:()=>p.Text2TextGenerationPipeline,TextClassificationPipeline:()=>p.TextClassificationPipeline,TextGenerationPipeline:()=>p.TextGenerationPipeline,TextStreamer:()=>b.TextStreamer,TextToAudioPipeline:()=>p.TextToAudioPipeline,TokenClassificationPipeline:()=>p.TokenClassificationPipeline,TokenClassifierOutput:()=>h.TokenClassifierOutput,TokenizerModel:()=>m.TokenizerModel,TrOCRForCausalLM:()=>h.TrOCRForCausalLM,TrOCRPreTrainedModel:()=>h.TrOCRPreTrainedModel,TranslationPipeline:()=>p.TranslationPipeline,UniSpeechForCTC:()=>h.UniSpeechForCTC,UniSpeechForSequenceClassification:()=>h.UniSpeechForSequenceClassification,UniSpeechModel:()=>h.UniSpeechModel,UniSpeechPreTrainedModel:()=>h.UniSpeechPreTrainedModel,UniSpeechSatForAudioFrameClassification:()=>h.UniSpeechSatForAudioFrameClassification,UniSpeechSatForCTC:()=>h.UniSpeechSatForCTC,UniSpeechSatForSequenceClassification:()=>h.UniSpeechSatForSequenceClassification,UniSpeechSatModel:()=>h.UniSpeechSatModel,UniSpeechSatPreTrainedModel:()=>h.UniSpeechSatPreTrainedModel,ViTFeatureExtractor:()=>_.ViTFeatureExtractor,ViTForImageClassification:()=>h.ViTForImageClassification,ViTImageProcessor:()=>_.ViTImageProcessor,ViTMAEModel:()=>h.ViTMAEModel,ViTMAEPreTrainedModel:()=>h.ViTMAEPreTrainedModel,ViTMSNForImageClassification:()=>h.ViTMSNForImageClassification,ViTMSNModel:()=>h.ViTMSNModel,ViTMSNPreTrainedModel:()=>h.ViTMSNPreTrainedModel,ViTModel:()=>h.ViTModel,ViTPreTrainedModel:()=>h.ViTPreTrainedModel,VisionEncoderDecoderModel:()=>h.VisionEncoderDecoderModel,VitMatteForImageMatting:()=>h.VitMatteForImageMatting,VitMatteImageProcessor:()=>_.VitMatteImageProcessor,VitMattePreTrainedModel:()=>h.VitMattePreTrainedModel,VitsModel:()=>h.VitsModel,VitsModelOutput:()=>h.VitsModelOutput,VitsPreTrainedModel:()=>h.VitsPreTrainedModel,VitsTokenizer:()=>m.VitsTokenizer,Wav2Vec2BertForCTC:()=>h.Wav2Vec2BertForCTC,Wav2Vec2BertForSequenceClassification:()=>h.Wav2Vec2BertForSequenceClassification,Wav2Vec2BertModel:()=>h.Wav2Vec2BertModel,Wav2Vec2BertPreTrainedModel:()=>h.Wav2Vec2BertPreTrainedModel,Wav2Vec2CTCTokenizer:()=>m.Wav2Vec2CTCTokenizer,Wav2Vec2FeatureExtractor:()=>_.Wav2Vec2FeatureExtractor,Wav2Vec2ForAudioFrameClassification:()=>h.Wav2Vec2ForAudioFrameClassification,Wav2Vec2ForCTC:()=>h.Wav2Vec2ForCTC,Wav2Vec2ForSequenceClassification:()=>h.Wav2Vec2ForSequenceClassification,Wav2Vec2Model:()=>h.Wav2Vec2Model,Wav2Vec2PreTrainedModel:()=>h.Wav2Vec2PreTrainedModel,Wav2Vec2ProcessorWithLM:()=>_.Wav2Vec2ProcessorWithLM,WavLMForAudioFrameClassification:()=>h.WavLMForAudioFrameClassification,WavLMForCTC:()=>h.WavLMForCTC,WavLMForSequenceClassification:()=>h.WavLMForSequenceClassification,WavLMForXVector:()=>h.WavLMForXVector,WavLMModel:()=>h.WavLMModel,WavLMPreTrainedModel:()=>h.WavLMPreTrainedModel,WeSpeakerFeatureExtractor:()=>_.WeSpeakerFeatureExtractor,WeSpeakerResNetModel:()=>h.WeSpeakerResNetModel,WeSpeakerResNetPreTrainedModel:()=>h.WeSpeakerResNetPreTrainedModel,WhisperFeatureExtractor:()=>_.WhisperFeatureExtractor,WhisperForConditionalGeneration:()=>h.WhisperForConditionalGeneration,WhisperModel:()=>h.WhisperModel,WhisperPreTrainedModel:()=>h.WhisperPreTrainedModel,WhisperProcessor:()=>_.WhisperProcessor,WhisperTextStreamer:()=>b.WhisperTextStreamer,WhisperTokenizer:()=>m.WhisperTokenizer,XLMForQuestionAnswering:()=>h.XLMForQuestionAnswering,XLMForSequenceClassification:()=>h.XLMForSequenceClassification,XLMForTokenClassification:()=>h.XLMForTokenClassification,XLMModel:()=>h.XLMModel,XLMPreTrainedModel:()=>h.XLMPreTrainedModel,XLMRobertaForMaskedLM:()=>h.XLMRobertaForMaskedLM,XLMRobertaForQuestionAnswering:()=>h.XLMRobertaForQuestionAnswering,XLMRobertaForSequenceClassification:()=>h.XLMRobertaForSequenceClassification,XLMRobertaForTokenClassification:()=>h.XLMRobertaForTokenClassification,XLMRobertaModel:()=>h.XLMRobertaModel,XLMRobertaPreTrainedModel:()=>h.XLMRobertaPreTrainedModel,XLMRobertaTokenizer:()=>m.XLMRobertaTokenizer,XLMTokenizer:()=>m.XLMTokenizer,XLMWithLMHeadModel:()=>h.XLMWithLMHeadModel,XVectorOutput:()=>h.XVectorOutput,YolosFeatureExtractor:()=>_.YolosFeatureExtractor,YolosForObjectDetection:()=>h.YolosForObjectDetection,YolosModel:()=>h.YolosModel,YolosObjectDetectionOutput:()=>h.YolosObjectDetectionOutput,YolosPreTrainedModel:()=>h.YolosPreTrainedModel,ZeroShotAudioClassificationPipeline:()=>p.ZeroShotAudioClassificationPipeline,ZeroShotClassificationPipeline:()=>p.ZeroShotClassificationPipeline,ZeroShotImageClassificationPipeline:()=>p.ZeroShotImageClassificationPipeline,ZeroShotObjectDetectionPipeline:()=>p.ZeroShotObjectDetectionPipeline,bankers_round:()=>T.bankers_round,cat:()=>w.cat,cos_sim:()=>T.cos_sim,dot:()=>T.dot,dynamic_time_warping:()=>T.dynamic_time_warping,env:()=>u.env,full:()=>w.full,full_like:()=>w.full_like,getKeyValueShapes:()=>f.getKeyValueShapes,hamming:()=>g.hamming,hanning:()=>g.hanning,interpolate:()=>w.interpolate,interpolate_4d:()=>w.interpolate_4d,interpolate_data:()=>T.interpolate_data,is_chinese_char:()=>m.is_chinese_char,layer_norm:()=>w.layer_norm,log_softmax:()=>T.log_softmax,magnitude:()=>T.magnitude,matmul:()=>w.matmul,max:()=>T.max,mean:()=>w.mean,mean_pooling:()=>w.mean_pooling,medianFilter:()=>T.medianFilter,mel_filter_bank:()=>g.mel_filter_bank,min:()=>T.min,ones:()=>w.ones,ones_like:()=>w.ones_like,permute:()=>w.permute,permute_data:()=>T.permute_data,pipeline:()=>p.pipeline,quantize_embeddings:()=>w.quantize_embeddings,read_audio:()=>g.read_audio,rfft:()=>w.rfft,round:()=>T.round,softmax:()=>T.softmax,spectrogram:()=>g.spectrogram,stack:()=>w.stack,std_mean:()=>w.std_mean,topk:()=>w.topk,window_function:()=>g.window_function,zeros:()=>w.zeros,zeros_like:()=>w.zeros_like});var u=c(/*! ./env.js */"./src/env.js"),p=c(/*! ./pipelines.js */"./src/pipelines.js"),h=c(/*! ./models.js */"./src/models.js"),m=c(/*! ./tokenizers.js */"./src/tokenizers.js"),_=c(/*! ./processors.js */"./src/processors.js"),f=c(/*! ./configs.js */"./src/configs.js"),g=c(/*! ./utils/audio.js */"./src/utils/audio.js"),M=c(/*! ./utils/image.js */"./src/utils/image.js"),w=c(/*! ./utils/tensor.js */"./src/utils/tensor.js"),T=c(/*! ./utils/maths.js */"./src/utils/maths.js"),b=c(/*! ./generation/streamers.js */"./src/generation/streamers.js"),x=c(/*! ./generation/stopping_criteria.js */"./src/generation/stopping_criteria.js"),y=d.ASTFeatureExtractor,k=d.ASTForAudioClassification,F=d.ASTModel,P=d.ASTPreTrainedModel,C=d.AlbertForMaskedLM,v=d.AlbertForQuestionAnswering,S=d.AlbertForSequenceClassification,A=d.AlbertModel,E=d.AlbertPreTrainedModel,L=d.AlbertTokenizer,z=d.AudioClassificationPipeline,I=d.AutoConfig,B=d.AutoModel,N=d.AutoModelForAudioClassification,D=d.AutoModelForAudioFrameClassification,O=d.AutoModelForCTC,V=d.AutoModelForCausalLM,j=d.AutoModelForDepthEstimation,R=d.AutoModelForDocumentQuestionAnswering,G=d.AutoModelForImageClassification,q=d.AutoModelForImageFeatureExtraction,$=d.AutoModelForImageMatting,W=d.AutoModelForImageSegmentation,U=d.AutoModelForImageToImage,X=d.AutoModelForMaskGeneration,Q=d.AutoModelForMaskedLM,H=d.AutoModelForNormalEstimation,Y=d.AutoModelForObjectDetection,J=d.AutoModelForQuestionAnswering,K=d.AutoModelForSemanticSegmentation,Z=d.AutoModelForSeq2SeqLM,ee=d.AutoModelForSequenceClassification,te=d.AutoModelForSpeechSeq2Seq,re=d.AutoModelForTextToSpectrogram,ne=d.AutoModelForTextToWaveform,oe=d.AutoModelForTokenClassification,se=d.AutoModelForUniversalSegmentation,ae=d.AutoModelForVision2Seq,ie=d.AutoModelForXVector,le=d.AutoModelForZeroShotObjectDetection,ce=d.AutoProcessor,de=d.AutoTokenizer,ue=d.AutomaticSpeechRecognitionPipeline,pe=d.BartForConditionalGeneration,he=d.BartForSequenceClassification,me=d.BartModel,_e=d.BartPretrainedModel,fe=d.BartTokenizer,ge=d.BaseModelOutput,Me=d.BaseStreamer,we=d.BeitFeatureExtractor,Te=d.BeitForImageClassification,be=d.BeitModel,xe=d.BeitPreTrainedModel,ye=d.BertForMaskedLM,ke=d.BertForQuestionAnswering,Fe=d.BertForSequenceClassification,Pe=d.BertForTokenClassification,Ce=d.BertModel,ve=d.BertPreTrainedModel,Se=d.BertTokenizer,Ae=d.BitImageProcessor,Ee=d.BlenderbotForConditionalGeneration,Le=d.BlenderbotModel,ze=d.BlenderbotPreTrainedModel,Ie=d.BlenderbotSmallForConditionalGeneration,Be=d.BlenderbotSmallModel,Ne=d.BlenderbotSmallPreTrainedModel,De=d.BlenderbotSmallTokenizer,Oe=d.BlenderbotTokenizer,Ve=d.BloomForCausalLM,je=d.BloomModel,Re=d.BloomPreTrainedModel,Ge=d.BloomTokenizer,qe=d.CLIPFeatureExtractor,$e=d.CLIPImageProcessor,We=d.CLIPModel,Ue=d.CLIPPreTrainedModel,Xe=d.CLIPSegForImageSegmentation,Qe=d.CLIPSegModel,He=d.CLIPSegPreTrainedModel,Ye=d.CLIPTextModel,Je=d.CLIPTextModelWithProjection,Ke=d.CLIPTokenizer,Ze=d.CLIPVisionModel,et=d.CLIPVisionModelWithProjection,tt=d.CamembertForMaskedLM,rt=d.CamembertForQuestionAnswering,nt=d.CamembertForSequenceClassification,ot=d.CamembertForTokenClassification,st=d.CamembertModel,at=d.CamembertPreTrainedModel,it=d.CamembertTokenizer,lt=d.CausalLMOutput,ct=d.CausalLMOutputWithPast,dt=d.ChineseCLIPFeatureExtractor,ut=d.ChineseCLIPModel,pt=d.ChineseCLIPPreTrainedModel,ht=d.ClapAudioModelWithProjection,mt=d.ClapFeatureExtractor,_t=d.ClapModel,ft=d.ClapPreTrainedModel,gt=d.ClapTextModelWithProjection,Mt=d.CodeGenForCausalLM,wt=d.CodeGenModel,Tt=d.CodeGenPreTrainedModel,bt=d.CodeGenTokenizer,xt=d.CodeLlamaTokenizer,yt=d.CohereForCausalLM,kt=d.CohereModel,Ft=d.CoherePreTrainedModel,Pt=d.CohereTokenizer,Ct=d.ConvBertForMaskedLM,vt=d.ConvBertForQuestionAnswering,St=d.ConvBertForSequenceClassification,At=d.ConvBertForTokenClassification,Et=d.ConvBertModel,Lt=d.ConvBertPreTrainedModel,zt=d.ConvBertTokenizer,It=d.ConvNextFeatureExtractor,Bt=d.ConvNextForImageClassification,Nt=d.ConvNextImageProcessor,Dt=d.ConvNextModel,Ot=d.ConvNextPreTrainedModel,Vt=d.ConvNextV2ForImageClassification,jt=d.ConvNextV2Model,Rt=d.ConvNextV2PreTrainedModel,Gt=d.DPTFeatureExtractor,qt=d.DPTForDepthEstimation,$t=d.DPTImageProcessor,Wt=d.DPTModel,Ut=d.DPTPreTrainedModel,Xt=d.DebertaForMaskedLM,Qt=d.DebertaForQuestionAnswering,Ht=d.DebertaForSequenceClassification,Yt=d.DebertaForTokenClassification,Jt=d.DebertaModel,Kt=d.DebertaPreTrainedModel,Zt=d.DebertaTokenizer,er=d.DebertaV2ForMaskedLM,tr=d.DebertaV2ForQuestionAnswering,rr=d.DebertaV2ForSequenceClassification,nr=d.DebertaV2ForTokenClassification,or=d.DebertaV2Model,sr=d.DebertaV2PreTrainedModel,ar=d.DebertaV2Tokenizer,ir=d.DecisionTransformerModel,lr=d.DecisionTransformerPreTrainedModel,cr=d.DeiTFeatureExtractor,dr=d.DeiTForImageClassification,ur=d.DeiTModel,pr=d.DeiTPreTrainedModel,hr=d.DepthAnythingForDepthEstimation,mr=d.DepthAnythingPreTrainedModel,_r=d.DepthEstimationPipeline,fr=d.DepthProForDepthEstimation,gr=d.DepthProPreTrainedModel,Mr=d.DetrFeatureExtractor,wr=d.DetrForObjectDetection,Tr=d.DetrForSegmentation,br=d.DetrModel,xr=d.DetrObjectDetectionOutput,yr=d.DetrPreTrainedModel,kr=d.DetrSegmentationOutput,Fr=d.Dinov2ForImageClassification,Pr=d.Dinov2Model,Cr=d.Dinov2PreTrainedModel,vr=d.DistilBertForMaskedLM,Sr=d.DistilBertForQuestionAnswering,Ar=d.DistilBertForSequenceClassification,Er=d.DistilBertForTokenClassification,Lr=d.DistilBertModel,zr=d.DistilBertPreTrainedModel,Ir=d.DistilBertTokenizer,Br=d.DocumentQuestionAnsweringPipeline,Nr=d.DonutFeatureExtractor,Dr=d.DonutImageProcessor,Or=d.DonutSwinModel,Vr=d.DonutSwinPreTrainedModel,jr=d.EfficientNetForImageClassification,Rr=d.EfficientNetImageProcessor,Gr=d.EfficientNetModel,qr=d.EfficientNetPreTrainedModel,$r=d.ElectraForMaskedLM,Wr=d.ElectraForQuestionAnswering,Ur=d.ElectraForSequenceClassification,Xr=d.ElectraForTokenClassification,Qr=d.ElectraModel,Hr=d.ElectraPreTrainedModel,Yr=d.ElectraTokenizer,Jr=d.EosTokenCriteria,Kr=d.EsmForMaskedLM,Zr=d.EsmForSequenceClassification,en=d.EsmForTokenClassification,tn=d.EsmModel,rn=d.EsmPreTrainedModel,nn=d.EsmTokenizer,on=d.FFT,sn=d.FalconForCausalLM,an=d.FalconModel,ln=d.FalconPreTrainedModel,cn=d.FalconTokenizer,dn=d.FastViTForImageClassification,un=d.FastViTModel,pn=d.FastViTPreTrainedModel,hn=d.FeatureExtractionPipeline,mn=d.FeatureExtractor,_n=d.FillMaskPipeline,fn=d.Florence2ForConditionalGeneration,gn=d.Florence2PreTrainedModel,Mn=d.Florence2Processor,wn=d.GLPNFeatureExtractor,Tn=d.GLPNForDepthEstimation,bn=d.GLPNModel,xn=d.GLPNPreTrainedModel,yn=d.GPT2LMHeadModel,kn=d.GPT2Model,Fn=d.GPT2PreTrainedModel,Pn=d.GPT2Tokenizer,Cn=d.GPTBigCodeForCausalLM,vn=d.GPTBigCodeModel,Sn=d.GPTBigCodePreTrainedModel,An=d.GPTJForCausalLM,En=d.GPTJModel,Ln=d.GPTJPreTrainedModel,zn=d.GPTNeoForCausalLM,In=d.GPTNeoModel,Bn=d.GPTNeoPreTrainedModel,Nn=d.GPTNeoXForCausalLM,Dn=d.GPTNeoXModel,On=d.GPTNeoXPreTrainedModel,Vn=d.GPTNeoXTokenizer,jn=d.Gemma2ForCausalLM,Rn=d.Gemma2Model,Gn=d.Gemma2PreTrainedModel,qn=d.GemmaForCausalLM,$n=d.GemmaModel,Wn=d.GemmaPreTrainedModel,Un=d.GemmaTokenizer,Xn=d.GraniteForCausalLM,Qn=d.GraniteModel,Hn=d.GranitePreTrainedModel,Yn=d.Grok1Tokenizer,Jn=d.GroupViTModel,Kn=d.GroupViTPreTrainedModel,Zn=d.HerbertTokenizer,eo=d.HieraForImageClassification,to=d.HieraModel,ro=d.HieraPreTrainedModel,no=d.HubertForCTC,oo=d.HubertForSequenceClassification,so=d.HubertModel,ao=d.HubertPreTrainedModel,io=d.ImageClassificationPipeline,lo=d.ImageFeatureExtractionPipeline,co=d.ImageFeatureExtractor,uo=d.ImageMattingOutput,po=d.ImageSegmentationPipeline,ho=d.ImageToImagePipeline,mo=d.ImageToTextPipeline,_o=d.InterruptableStoppingCriteria,fo=d.JAISLMHeadModel,go=d.JAISModel,Mo=d.JAISPreTrainedModel,wo=d.LlamaForCausalLM,To=d.LlamaModel,bo=d.LlamaPreTrainedModel,xo=d.LlamaTokenizer,yo=d.LlavaForConditionalGeneration,ko=d.LlavaPreTrainedModel,Fo=d.LongT5ForConditionalGeneration,Po=d.LongT5Model,Co=d.LongT5PreTrainedModel,vo=d.M2M100ForConditionalGeneration,So=d.M2M100Model,Ao=d.M2M100PreTrainedModel,Eo=d.M2M100Tokenizer,Lo=d.MBart50Tokenizer,zo=d.MBartForCausalLM,Io=d.MBartForConditionalGeneration,Bo=d.MBartForSequenceClassification,No=d.MBartModel,Do=d.MBartPreTrainedModel,Oo=d.MBartTokenizer,Vo=d.MPNetForMaskedLM,jo=d.MPNetForQuestionAnswering,Ro=d.MPNetForSequenceClassification,Go=d.MPNetForTokenClassification,qo=d.MPNetModel,$o=d.MPNetPreTrainedModel,Wo=d.MPNetTokenizer,Uo=d.MT5ForConditionalGeneration,Xo=d.MT5Model,Qo=d.MT5PreTrainedModel,Ho=d.MarianMTModel,Yo=d.MarianModel,Jo=d.MarianPreTrainedModel,Ko=d.MarianTokenizer,Zo=d.MaskFormerFeatureExtractor,es=d.MaskFormerForInstanceSegmentation,ts=d.MaskFormerModel,rs=d.MaskFormerPreTrainedModel,ns=d.MaskedLMOutput,os=d.MaxLengthCriteria,ss=d.MistralForCausalLM,as=d.MistralModel,is=d.MistralPreTrainedModel,ls=d.MobileBertForMaskedLM,cs=d.MobileBertForQuestionAnswering,ds=d.MobileBertForSequenceClassification,us=d.MobileBertModel,ps=d.MobileBertPreTrainedModel,hs=d.MobileBertTokenizer,ms=d.MobileNetV1FeatureExtractor,_s=d.MobileNetV1ForImageClassification,fs=d.MobileNetV1Model,gs=d.MobileNetV1PreTrainedModel,Ms=d.MobileNetV2FeatureExtractor,ws=d.MobileNetV2ForImageClassification,Ts=d.MobileNetV2Model,bs=d.MobileNetV2PreTrainedModel,xs=d.MobileNetV3FeatureExtractor,ys=d.MobileNetV3ForImageClassification,ks=d.MobileNetV3Model,Fs=d.MobileNetV3PreTrainedModel,Ps=d.MobileNetV4FeatureExtractor,Cs=d.MobileNetV4ForImageClassification,vs=d.MobileNetV4Model,Ss=d.MobileNetV4PreTrainedModel,As=d.MobileViTFeatureExtractor,Es=d.MobileViTForImageClassification,Ls=d.MobileViTImageProcessor,zs=d.MobileViTModel,Is=d.MobileViTPreTrainedModel,Bs=d.MobileViTV2ForImageClassification,Ns=d.MobileViTV2Model,Ds=d.MobileViTV2PreTrainedModel,Os=d.ModelOutput,Vs=d.Moondream1ForConditionalGeneration,js=d.MptForCausalLM,Rs=d.MptModel,Gs=d.MptPreTrainedModel,qs=d.MusicgenForCausalLM,$s=d.MusicgenForConditionalGeneration,Ws=d.MusicgenModel,Us=d.MusicgenPreTrainedModel,Xs=d.NllbTokenizer,Qs=d.NomicBertModel,Hs=d.NomicBertPreTrainedModel,Ys=d.NougatImageProcessor,Js=d.NougatTokenizer,Ks=d.OPTForCausalLM,Zs=d.OPTModel,ea=d.OPTPreTrainedModel,ta=d.ObjectDetectionPipeline,ra=d.OpenELMForCausalLM,na=d.OpenELMModel,oa=d.OpenELMPreTrainedModel,sa=d.OwlViTFeatureExtractor,aa=d.OwlViTForObjectDetection,ia=d.OwlViTModel,la=d.OwlViTPreTrainedModel,ca=d.OwlViTProcessor,da=d.Owlv2ForObjectDetection,ua=d.Owlv2ImageProcessor,pa=d.Owlv2Model,ha=d.Owlv2PreTrainedModel,ma=d.Phi3ForCausalLM,_a=d.Phi3Model,fa=d.Phi3PreTrainedModel,ga=d.PhiForCausalLM,Ma=d.PhiModel,wa=d.PhiPreTrainedModel,Ta=d.Pipeline,ba=d.PreTrainedModel,xa=d.PreTrainedTokenizer,ya=d.PretrainedConfig,ka=d.PretrainedMixin,Fa=d.Processor,Pa=d.PvtForImageClassification,Ca=d.PvtImageProcessor,va=d.PvtModel,Sa=d.PvtPreTrainedModel,Aa=d.PyAnnoteFeatureExtractor,Ea=d.PyAnnoteForAudioFrameClassification,La=d.PyAnnoteModel,za=d.PyAnnotePreTrainedModel,Ia=d.PyAnnoteProcessor,Ba=d.QuestionAnsweringModelOutput,Na=d.QuestionAnsweringPipeline,Da=d.Qwen2ForCausalLM,Oa=d.Qwen2Model,Va=d.Qwen2PreTrainedModel,ja=d.Qwen2Tokenizer,Ra=d.RTDetrForObjectDetection,Ga=d.RTDetrImageProcessor,qa=d.RTDetrModel,$a=d.RTDetrObjectDetectionOutput,Wa=d.RTDetrPreTrainedModel,Ua=d.RawImage,Xa=d.ResNetForImageClassification,Qa=d.ResNetModel,Ha=d.ResNetPreTrainedModel,Ya=d.RoFormerForMaskedLM,Ja=d.RoFormerForQuestionAnswering,Ka=d.RoFormerForSequenceClassification,Za=d.RoFormerForTokenClassification,ei=d.RoFormerModel,ti=d.RoFormerPreTrainedModel,ri=d.RoFormerTokenizer,ni=d.RobertaForMaskedLM,oi=d.RobertaForQuestionAnswering,si=d.RobertaForSequenceClassification,ai=d.RobertaForTokenClassification,ii=d.RobertaModel,li=d.RobertaPreTrainedModel,ci=d.RobertaTokenizer,di=d.SamImageProcessor,ui=d.SamImageSegmentationOutput,pi=d.SamModel,hi=d.SamPreTrainedModel,mi=d.SamProcessor,_i=d.SapiensFeatureExtractor,fi=d.SapiensForDepthEstimation,gi=d.SapiensForNormalEstimation,Mi=d.SapiensForSemanticSegmentation,wi=d.SapiensPreTrainedModel,Ti=d.SeamlessM4TFeatureExtractor,bi=d.SegformerFeatureExtractor,xi=d.SegformerForImageClassification,yi=d.SegformerForSemanticSegmentation,ki=d.SegformerModel,Fi=d.SegformerPreTrainedModel,Pi=d.Seq2SeqLMOutput,Ci=d.SequenceClassifierOutput,vi=d.SiglipImageProcessor,Si=d.SiglipModel,Ai=d.SiglipPreTrainedModel,Ei=d.SiglipTextModel,Li=d.SiglipTokenizer,zi=d.SiglipVisionModel,Ii=d.SpeechT5FeatureExtractor,Bi=d.SpeechT5ForSpeechToText,Ni=d.SpeechT5ForTextToSpeech,Di=d.SpeechT5HifiGan,Oi=d.SpeechT5Model,Vi=d.SpeechT5PreTrainedModel,ji=d.SpeechT5Processor,Ri=d.SpeechT5Tokenizer,Gi=d.SqueezeBertForMaskedLM,qi=d.SqueezeBertForQuestionAnswering,$i=d.SqueezeBertForSequenceClassification,Wi=d.SqueezeBertModel,Ui=d.SqueezeBertPreTrainedModel,Xi=d.SqueezeBertTokenizer,Qi=d.StableLmForCausalLM,Hi=d.StableLmModel,Yi=d.StableLmPreTrainedModel,Ji=d.Starcoder2ForCausalLM,Ki=d.Starcoder2Model,Zi=d.Starcoder2PreTrainedModel,el=d.StoppingCriteria,tl=d.StoppingCriteriaList,rl=d.SummarizationPipeline,nl=d.Swin2SRForImageSuperResolution,ol=d.Swin2SRImageProcessor,sl=d.Swin2SRModel,al=d.Swin2SRPreTrainedModel,il=d.SwinForImageClassification,ll=d.SwinModel,cl=d.SwinPreTrainedModel,dl=d.T5ForConditionalGeneration,ul=d.T5Model,pl=d.T5PreTrainedModel,hl=d.T5Tokenizer,ml=d.TableTransformerForObjectDetection,_l=d.TableTransformerModel,fl=d.TableTransformerObjectDetectionOutput,gl=d.TableTransformerPreTrainedModel,Ml=d.Tensor,wl=d.Text2TextGenerationPipeline,Tl=d.TextClassificationPipeline,bl=d.TextGenerationPipeline,xl=d.TextStreamer,yl=d.TextToAudioPipeline,kl=d.TokenClassificationPipeline,Fl=d.TokenClassifierOutput,Pl=d.TokenizerModel,Cl=d.TrOCRForCausalLM,vl=d.TrOCRPreTrainedModel,Sl=d.TranslationPipeline,Al=d.UniSpeechForCTC,El=d.UniSpeechForSequenceClassification,Ll=d.UniSpeechModel,zl=d.UniSpeechPreTrainedModel,Il=d.UniSpeechSatForAudioFrameClassification,Bl=d.UniSpeechSatForCTC,Nl=d.UniSpeechSatForSequenceClassification,Dl=d.UniSpeechSatModel,Ol=d.UniSpeechSatPreTrainedModel,Vl=d.ViTFeatureExtractor,jl=d.ViTForImageClassification,Rl=d.ViTImageProcessor,Gl=d.ViTMAEModel,ql=d.ViTMAEPreTrainedModel,$l=d.ViTMSNForImageClassification,Wl=d.ViTMSNModel,Ul=d.ViTMSNPreTrainedModel,Xl=d.ViTModel,Ql=d.ViTPreTrainedModel,Hl=d.VisionEncoderDecoderModel,Yl=d.VitMatteForImageMatting,Jl=d.VitMatteImageProcessor,Kl=d.VitMattePreTrainedModel,Zl=d.VitsModel,ec=d.VitsModelOutput,tc=d.VitsPreTrainedModel,rc=d.VitsTokenizer,nc=d.Wav2Vec2BertForCTC,oc=d.Wav2Vec2BertForSequenceClassification,sc=d.Wav2Vec2BertModel,ac=d.Wav2Vec2BertPreTrainedModel,ic=d.Wav2Vec2CTCTokenizer,lc=d.Wav2Vec2FeatureExtractor,cc=d.Wav2Vec2ForAudioFrameClassification,dc=d.Wav2Vec2ForCTC,uc=d.Wav2Vec2ForSequenceClassification,pc=d.Wav2Vec2Model,hc=d.Wav2Vec2PreTrainedModel,mc=d.Wav2Vec2ProcessorWithLM,_c=d.WavLMForAudioFrameClassification,fc=d.WavLMForCTC,gc=d.WavLMForSequenceClassification,Mc=d.WavLMForXVector,wc=d.WavLMModel,Tc=d.WavLMPreTrainedModel,bc=d.WeSpeakerFeatureExtractor,xc=d.WeSpeakerResNetModel,yc=d.WeSpeakerResNetPreTrainedModel,kc=d.WhisperFeatureExtractor,Fc=d.WhisperForConditionalGeneration,Pc=d.WhisperModel,Cc=d.WhisperPreTrainedModel,vc=d.WhisperProcessor,Sc=d.WhisperTextStreamer,Ac=d.WhisperTokenizer,Ec=d.XLMForQuestionAnswering,Lc=d.XLMForSequenceClassification,zc=d.XLMForTokenClassification,Ic=d.XLMModel,Bc=d.XLMPreTrainedModel,Nc=d.XLMRobertaForMaskedLM,Dc=d.XLMRobertaForQuestionAnswering,Oc=d.XLMRobertaForSequenceClassification,Vc=d.XLMRobertaForTokenClassification,jc=d.XLMRobertaModel,Rc=d.XLMRobertaPreTrainedModel,Gc=d.XLMRobertaTokenizer,qc=d.XLMTokenizer,$c=d.XLMWithLMHeadModel,Wc=d.XVectorOutput,Uc=d.YolosFeatureExtractor,Xc=d.YolosForObjectDetection,Qc=d.YolosModel,Hc=d.YolosObjectDetectionOutput,Yc=d.YolosPreTrainedModel,Jc=d.ZeroShotAudioClassificationPipeline,Kc=d.ZeroShotClassificationPipeline,Zc=d.ZeroShotImageClassificationPipeline,ed=d.ZeroShotObjectDetectionPipeline,td=d.bankers_round,rd=d.cat,nd=d.cos_sim,od=d.dot,sd=d.dynamic_time_warping,ad=d.env,id=d.full,ld=d.full_like,cd=d.getKeyValueShapes,dd=d.hamming,ud=d.hanning,pd=d.interpolate,hd=d.interpolate_4d,md=d.interpolate_data,_d=d.is_chinese_char,fd=d.layer_norm,gd=d.log_softmax,Md=d.magnitude,wd=d.matmul,Td=d.max,bd=d.mean,xd=d.mean_pooling,yd=d.medianFilter,kd=d.mel_filter_bank,Fd=d.min,Pd=d.ones,Cd=d.ones_like,vd=d.permute,Sd=d.permute_data,Ad=d.pipeline,Ed=d.quantize_embeddings,Ld=d.read_audio,zd=d.rfft,Id=d.round,Bd=d.softmax,Nd=d.spectrogram,Dd=d.stack,Od=d.std_mean,Vd=d.topk,jd=d.window_function,Rd=d.zeros,Gd=d.zeros_like;export{y as ASTFeatureExtractor,k as ASTForAudioClassification,F as ASTModel,P as ASTPreTrainedModel,C as AlbertForMaskedLM,v as AlbertForQuestionAnswering,S as AlbertForSequenceClassification,A as AlbertModel,E as AlbertPreTrainedModel,L as AlbertTokenizer,z as AudioClassificationPipeline,I as AutoConfig,B as AutoModel,N as AutoModelForAudioClassification,D as AutoModelForAudioFrameClassification,O as AutoModelForCTC,V as AutoModelForCausalLM,j as AutoModelForDepthEstimation,R as AutoModelForDocumentQuestionAnswering,G as AutoModelForImageClassification,q as AutoModelForImageFeatureExtraction,$ as AutoModelForImageMatting,W as AutoModelForImageSegmentation,U as AutoModelForImageToImage,X as AutoModelForMaskGeneration,Q as AutoModelForMaskedLM,H as AutoModelForNormalEstimation,Y as AutoModelForObjectDetection,J as AutoModelForQuestionAnswering,K as AutoModelForSemanticSegmentation,Z as AutoModelForSeq2SeqLM,ee as AutoModelForSequenceClassification,te as AutoModelForSpeechSeq2Seq,re as AutoModelForTextToSpectrogram,ne as AutoModelForTextToWaveform,oe as AutoModelForTokenClassification,se as AutoModelForUniversalSegmentation,ae as AutoModelForVision2Seq,ie as AutoModelForXVector,le as AutoModelForZeroShotObjectDetection,ce as AutoProcessor,de as AutoTokenizer,ue as AutomaticSpeechRecognitionPipeline,pe as BartForConditionalGeneration,he as BartForSequenceClassification,me as BartModel,_e as BartPretrainedModel,fe as BartTokenizer,ge as BaseModelOutput,Me as BaseStreamer,we as BeitFeatureExtractor,Te as BeitForImageClassification,be as BeitModel,xe as BeitPreTrainedModel,ye as BertForMaskedLM,ke as BertForQuestionAnswering,Fe as BertForSequenceClassification,Pe as BertForTokenClassification,Ce as BertModel,ve as BertPreTrainedModel,Se as BertTokenizer,Ae as BitImageProcessor,Ee as BlenderbotForConditionalGeneration,Le as BlenderbotModel,ze as BlenderbotPreTrainedModel,Ie as BlenderbotSmallForConditionalGeneration,Be as BlenderbotSmallModel,Ne as BlenderbotSmallPreTrainedModel,De as BlenderbotSmallTokenizer,Oe as BlenderbotTokenizer,Ve as BloomForCausalLM,je as BloomModel,Re as BloomPreTrainedModel,Ge as BloomTokenizer,qe as CLIPFeatureExtractor,$e as CLIPImageProcessor,We as CLIPModel,Ue as CLIPPreTrainedModel,Xe as CLIPSegForImageSegmentation,Qe as CLIPSegModel,He as CLIPSegPreTrainedModel,Ye as CLIPTextModel,Je as CLIPTextModelWithProjection,Ke as CLIPTokenizer,Ze as CLIPVisionModel,et as CLIPVisionModelWithProjection,tt as CamembertForMaskedLM,rt as CamembertForQuestionAnswering,nt as CamembertForSequenceClassification,ot as CamembertForTokenClassification,st as CamembertModel,at as CamembertPreTrainedModel,it as CamembertTokenizer,lt as CausalLMOutput,ct as CausalLMOutputWithPast,dt as ChineseCLIPFeatureExtractor,ut as ChineseCLIPModel,pt as ChineseCLIPPreTrainedModel,ht as ClapAudioModelWithProjection,mt as ClapFeatureExtractor,_t as ClapModel,ft as ClapPreTrainedModel,gt as ClapTextModelWithProjection,Mt as CodeGenForCausalLM,wt as CodeGenModel,Tt as CodeGenPreTrainedModel,bt as CodeGenTokenizer,xt as CodeLlamaTokenizer,yt as CohereForCausalLM,kt as CohereModel,Ft as CoherePreTrainedModel,Pt as CohereTokenizer,Ct as ConvBertForMaskedLM,vt as ConvBertForQuestionAnswering,St as ConvBertForSequenceClassification,At as ConvBertForTokenClassification,Et as ConvBertModel,Lt as ConvBertPreTrainedModel,zt as ConvBertTokenizer,It as ConvNextFeatureExtractor,Bt as ConvNextForImageClassification,Nt as ConvNextImageProcessor,Dt as ConvNextModel,Ot as ConvNextPreTrainedModel,Vt as ConvNextV2ForImageClassification,jt as ConvNextV2Model,Rt as ConvNextV2PreTrainedModel,Gt as DPTFeatureExtractor,qt as DPTForDepthEstimation,$t as DPTImageProcessor,Wt as DPTModel,Ut as DPTPreTrainedModel,Xt as DebertaForMaskedLM,Qt as DebertaForQuestionAnswering,Ht as DebertaForSequenceClassification,Yt as DebertaForTokenClassification,Jt as DebertaModel,Kt as DebertaPreTrainedModel,Zt as DebertaTokenizer,er as DebertaV2ForMaskedLM,tr as DebertaV2ForQuestionAnswering,rr as DebertaV2ForSequenceClassification,nr as DebertaV2ForTokenClassification,or as DebertaV2Model,sr as DebertaV2PreTrainedModel,ar as DebertaV2Tokenizer,ir as DecisionTransformerModel,lr as DecisionTransformerPreTrainedModel,cr as DeiTFeatureExtractor,dr as DeiTForImageClassification,ur as DeiTModel,pr as DeiTPreTrainedModel,hr as DepthAnythingForDepthEstimation,mr as DepthAnythingPreTrainedModel,_r as DepthEstimationPipeline,fr as DepthProForDepthEstimation,gr as DepthProPreTrainedModel,Mr as DetrFeatureExtractor,wr as DetrForObjectDetection,Tr as DetrForSegmentation,br as DetrModel,xr as DetrObjectDetectionOutput,yr as DetrPreTrainedModel,kr as DetrSegmentationOutput,Fr as Dinov2ForImageClassification,Pr as Dinov2Model,Cr as Dinov2PreTrainedModel,vr as DistilBertForMaskedLM,Sr as DistilBertForQuestionAnswering,Ar as DistilBertForSequenceClassification,Er as DistilBertForTokenClassification,Lr as DistilBertModel,zr as DistilBertPreTrainedModel,Ir as DistilBertTokenizer,Br as DocumentQuestionAnsweringPipeline,Nr as DonutFeatureExtractor,Dr as DonutImageProcessor,Or as DonutSwinModel,Vr as DonutSwinPreTrainedModel,jr as EfficientNetForImageClassification,Rr as EfficientNetImageProcessor,Gr as EfficientNetModel,qr as EfficientNetPreTrainedModel,$r as ElectraForMaskedLM,Wr as ElectraForQuestionAnswering,Ur as ElectraForSequenceClassification,Xr as ElectraForTokenClassification,Qr as ElectraModel,Hr as ElectraPreTrainedModel,Yr as ElectraTokenizer,Jr as EosTokenCriteria,Kr as EsmForMaskedLM,Zr as EsmForSequenceClassification,en as EsmForTokenClassification,tn as EsmModel,rn as EsmPreTrainedModel,nn as EsmTokenizer,on as FFT,sn as FalconForCausalLM,an as FalconModel,ln as FalconPreTrainedModel,cn as FalconTokenizer,dn as FastViTForImageClassification,un as FastViTModel,pn as FastViTPreTrainedModel,hn as FeatureExtractionPipeline,mn as FeatureExtractor,_n as FillMaskPipeline,fn as Florence2ForConditionalGeneration,gn as Florence2PreTrainedModel,Mn as Florence2Processor,wn as GLPNFeatureExtractor,Tn as GLPNForDepthEstimation,bn as GLPNModel,xn as GLPNPreTrainedModel,yn as GPT2LMHeadModel,kn as GPT2Model,Fn as GPT2PreTrainedModel,Pn as GPT2Tokenizer,Cn as GPTBigCodeForCausalLM,vn as GPTBigCodeModel,Sn as GPTBigCodePreTrainedModel,An as GPTJForCausalLM,En as GPTJModel,Ln as GPTJPreTrainedModel,zn as GPTNeoForCausalLM,In as GPTNeoModel,Bn as GPTNeoPreTrainedModel,Nn as GPTNeoXForCausalLM,Dn as GPTNeoXModel,On as GPTNeoXPreTrainedModel,Vn as GPTNeoXTokenizer,jn as Gemma2ForCausalLM,Rn as Gemma2Model,Gn as Gemma2PreTrainedModel,qn as GemmaForCausalLM,$n as GemmaModel,Wn as GemmaPreTrainedModel,Un as GemmaTokenizer,Xn as GraniteForCausalLM,Qn as GraniteModel,Hn as GranitePreTrainedModel,Yn as Grok1Tokenizer,Jn as GroupViTModel,Kn as GroupViTPreTrainedModel,Zn as HerbertTokenizer,eo as HieraForImageClassification,to as HieraModel,ro as HieraPreTrainedModel,no as HubertForCTC,oo as HubertForSequenceClassification,so as HubertModel,ao as HubertPreTrainedModel,io as ImageClassificationPipeline,lo as ImageFeatureExtractionPipeline,co as ImageFeatureExtractor,uo as ImageMattingOutput,po as ImageSegmentationPipeline,ho as ImageToImagePipeline,mo as ImageToTextPipeline,_o as InterruptableStoppingCriteria,fo as JAISLMHeadModel,go as JAISModel,Mo as JAISPreTrainedModel,wo as LlamaForCausalLM,To as LlamaModel,bo as LlamaPreTrainedModel,xo as LlamaTokenizer,yo as LlavaForConditionalGeneration,ko as LlavaPreTrainedModel,Fo as LongT5ForConditionalGeneration,Po as LongT5Model,Co as LongT5PreTrainedModel,vo as M2M100ForConditionalGeneration,So as M2M100Model,Ao as M2M100PreTrainedModel,Eo as M2M100Tokenizer,Lo as MBart50Tokenizer,zo as MBartForCausalLM,Io as MBartForConditionalGeneration,Bo as MBartForSequenceClassification,No as MBartModel,Do as MBartPreTrainedModel,Oo as MBartTokenizer,Vo as MPNetForMaskedLM,jo as MPNetForQuestionAnswering,Ro as MPNetForSequenceClassification,Go as MPNetForTokenClassification,qo as MPNetModel,$o as MPNetPreTrainedModel,Wo as MPNetTokenizer,Uo as MT5ForConditionalGeneration,Xo as MT5Model,Qo as MT5PreTrainedModel,Ho as MarianMTModel,Yo as MarianModel,Jo as MarianPreTrainedModel,Ko as MarianTokenizer,Zo as MaskFormerFeatureExtractor,es as MaskFormerForInstanceSegmentation,ts as MaskFormerModel,rs as MaskFormerPreTrainedModel,ns as MaskedLMOutput,os as MaxLengthCriteria,ss as MistralForCausalLM,as as MistralModel,is as MistralPreTrainedModel,ls as MobileBertForMaskedLM,cs as MobileBertForQuestionAnswering,ds as MobileBertForSequenceClassification,us as MobileBertModel,ps as MobileBertPreTrainedModel,hs as MobileBertTokenizer,ms as MobileNetV1FeatureExtractor,_s as MobileNetV1ForImageClassification,fs as MobileNetV1Model,gs as MobileNetV1PreTrainedModel,Ms as MobileNetV2FeatureExtractor,ws as MobileNetV2ForImageClassification,Ts as MobileNetV2Model,bs as MobileNetV2PreTrainedModel,xs as MobileNetV3FeatureExtractor,ys as MobileNetV3ForImageClassification,ks as MobileNetV3Model,Fs as MobileNetV3PreTrainedModel,Ps as MobileNetV4FeatureExtractor,Cs as MobileNetV4ForImageClassification,vs as MobileNetV4Model,Ss as MobileNetV4PreTrainedModel,As as MobileViTFeatureExtractor,Es as MobileViTForImageClassification,Ls as MobileViTImageProcessor,zs as MobileViTModel,Is as MobileViTPreTrainedModel,Bs as MobileViTV2ForImageClassification,Ns as MobileViTV2Model,Ds as MobileViTV2PreTrainedModel,Os as ModelOutput,Vs as Moondream1ForConditionalGeneration,js as MptForCausalLM,Rs as MptModel,Gs as MptPreTrainedModel,qs as MusicgenForCausalLM,$s as MusicgenForConditionalGeneration,Ws as MusicgenModel,Us as MusicgenPreTrainedModel,Xs as NllbTokenizer,Qs as NomicBertModel,Hs as NomicBertPreTrainedModel,Ys as NougatImageProcessor,Js as NougatTokenizer,Ks as OPTForCausalLM,Zs as OPTModel,ea as OPTPreTrainedModel,ta as ObjectDetectionPipeline,ra as OpenELMForCausalLM,na as OpenELMModel,oa as OpenELMPreTrainedModel,sa as OwlViTFeatureExtractor,aa as OwlViTForObjectDetection,ia as OwlViTModel,la as OwlViTPreTrainedModel,ca as OwlViTProcessor,da as Owlv2ForObjectDetection,ua as Owlv2ImageProcessor,pa as Owlv2Model,ha as Owlv2PreTrainedModel,ma as Phi3ForCausalLM,_a as Phi3Model,fa as Phi3PreTrainedModel,ga as PhiForCausalLM,Ma as PhiModel,wa as PhiPreTrainedModel,Ta as Pipeline,ba as PreTrainedModel,xa as PreTrainedTokenizer,ya as PretrainedConfig,ka as PretrainedMixin,Fa as Processor,Pa as PvtForImageClassification,Ca as PvtImageProcessor,va as PvtModel,Sa as PvtPreTrainedModel,Aa as PyAnnoteFeatureExtractor,Ea as PyAnnoteForAudioFrameClassification,La as PyAnnoteModel,za as PyAnnotePreTrainedModel,Ia as PyAnnoteProcessor,Ba as QuestionAnsweringModelOutput,Na as QuestionAnsweringPipeline,Da as Qwen2ForCausalLM,Oa as Qwen2Model,Va as Qwen2PreTrainedModel,ja as Qwen2Tokenizer,Ra as RTDetrForObjectDetection,Ga as RTDetrImageProcessor,qa as RTDetrModel,$a as RTDetrObjectDetectionOutput,Wa as RTDetrPreTrainedModel,Ua as RawImage,Xa as ResNetForImageClassification,Qa as ResNetModel,Ha as ResNetPreTrainedModel,Ya as RoFormerForMaskedLM,Ja as RoFormerForQuestionAnswering,Ka as RoFormerForSequenceClassification,Za as RoFormerForTokenClassification,ei as RoFormerModel,ti as RoFormerPreTrainedModel,ri as RoFormerTokenizer,ni as RobertaForMaskedLM,oi as RobertaForQuestionAnswering,si as RobertaForSequenceClassification,ai as RobertaForTokenClassification,ii as RobertaModel,li as RobertaPreTrainedModel,ci as RobertaTokenizer,di as SamImageProcessor,ui as SamImageSegmentationOutput,pi as SamModel,hi as SamPreTrainedModel,mi as SamProcessor,_i as SapiensFeatureExtractor,fi as SapiensForDepthEstimation,gi as SapiensForNormalEstimation,Mi as SapiensForSemanticSegmentation,wi as SapiensPreTrainedModel,Ti as SeamlessM4TFeatureExtractor,bi as SegformerFeatureExtractor,xi as SegformerForImageClassification,yi as SegformerForSemanticSegmentation,ki as SegformerModel,Fi as SegformerPreTrainedModel,Pi as Seq2SeqLMOutput,Ci as SequenceClassifierOutput,vi as SiglipImageProcessor,Si as SiglipModel,Ai as SiglipPreTrainedModel,Ei as SiglipTextModel,Li as SiglipTokenizer,zi as SiglipVisionModel,Ii as SpeechT5FeatureExtractor,Bi as SpeechT5ForSpeechToText,Ni as SpeechT5ForTextToSpeech,Di as SpeechT5HifiGan,Oi as SpeechT5Model,Vi as SpeechT5PreTrainedModel,ji as SpeechT5Processor,Ri as SpeechT5Tokenizer,Gi as SqueezeBertForMaskedLM,qi as SqueezeBertForQuestionAnswering,$i as SqueezeBertForSequenceClassification,Wi as SqueezeBertModel,Ui as SqueezeBertPreTrainedModel,Xi as SqueezeBertTokenizer,Qi as StableLmForCausalLM,Hi as StableLmModel,Yi as StableLmPreTrainedModel,Ji as Starcoder2ForCausalLM,Ki as Starcoder2Model,Zi as Starcoder2PreTrainedModel,el as StoppingCriteria,tl as StoppingCriteriaList,rl as SummarizationPipeline,nl as Swin2SRForImageSuperResolution,ol as Swin2SRImageProcessor,sl as Swin2SRModel,al as Swin2SRPreTrainedModel,il as SwinForImageClassification,ll as SwinModel,cl as SwinPreTrainedModel,dl as T5ForConditionalGeneration,ul as T5Model,pl as T5PreTrainedModel,hl as T5Tokenizer,ml as TableTransformerForObjectDetection,_l as TableTransformerModel,fl as TableTransformerObjectDetectionOutput,gl as TableTransformerPreTrainedModel,Ml as Tensor,wl as Text2TextGenerationPipeline,Tl as TextClassificationPipeline,bl as TextGenerationPipeline,xl as TextStreamer,yl as TextToAudioPipeline,kl as TokenClassificationPipeline,Fl as TokenClassifierOutput,Pl as TokenizerModel,Cl as TrOCRForCausalLM,vl as TrOCRPreTrainedModel,Sl as TranslationPipeline,Al as UniSpeechForCTC,El as UniSpeechForSequenceClassification,Ll as UniSpeechModel,zl as UniSpeechPreTrainedModel,Il as UniSpeechSatForAudioFrameClassification,Bl as UniSpeechSatForCTC,Nl as UniSpeechSatForSequenceClassification,Dl as UniSpeechSatModel,Ol as UniSpeechSatPreTrainedModel,Vl as ViTFeatureExtractor,jl as ViTForImageClassification,Rl as ViTImageProcessor,Gl as ViTMAEModel,ql as ViTMAEPreTrainedModel,$l as ViTMSNForImageClassification,Wl as ViTMSNModel,Ul as ViTMSNPreTrainedModel,Xl as ViTModel,Ql as ViTPreTrainedModel,Hl as VisionEncoderDecoderModel,Yl as VitMatteForImageMatting,Jl as VitMatteImageProcessor,Kl as VitMattePreTrainedModel,Zl as VitsModel,ec as VitsModelOutput,tc as VitsPreTrainedModel,rc as VitsTokenizer,nc as Wav2Vec2BertForCTC,oc as Wav2Vec2BertForSequenceClassification,sc as Wav2Vec2BertModel,ac as Wav2Vec2BertPreTrainedModel,ic as Wav2Vec2CTCTokenizer,lc as Wav2Vec2FeatureExtractor,cc as Wav2Vec2ForAudioFrameClassification,dc as Wav2Vec2ForCTC,uc as Wav2Vec2ForSequenceClassification,pc as Wav2Vec2Model,hc as Wav2Vec2PreTrainedModel,mc as Wav2Vec2ProcessorWithLM,_c as WavLMForAudioFrameClassification,fc as WavLMForCTC,gc as WavLMForSequenceClassification,Mc as WavLMForXVector,wc as WavLMModel,Tc as WavLMPreTrainedModel,bc as WeSpeakerFeatureExtractor,xc as WeSpeakerResNetModel,yc as WeSpeakerResNetPreTrainedModel,kc as WhisperFeatureExtractor,Fc as WhisperForConditionalGeneration,Pc as WhisperModel,Cc as WhisperPreTrainedModel,vc as WhisperProcessor,Sc as WhisperTextStreamer,Ac as WhisperTokenizer,Ec as XLMForQuestionAnswering,Lc as XLMForSequenceClassification,zc as XLMForTokenClassification,Ic as XLMModel,Bc as XLMPreTrainedModel,Nc as XLMRobertaForMaskedLM,Dc as XLMRobertaForQuestionAnswering,Oc as XLMRobertaForSequenceClassification,Vc as XLMRobertaForTokenClassification,jc as XLMRobertaModel,Rc as XLMRobertaPreTrainedModel,Gc as XLMRobertaTokenizer,qc as XLMTokenizer,$c as XLMWithLMHeadModel,Wc as XVectorOutput,Uc as YolosFeatureExtractor,Xc as YolosForObjectDetection,Qc as YolosModel,Hc as YolosObjectDetectionOutput,Yc as YolosPreTrainedModel,Jc as ZeroShotAudioClassificationPipeline,Kc as ZeroShotClassificationPipeline,Zc as ZeroShotImageClassificationPipeline,ed as ZeroShotObjectDetectionPipeline,td as bankers_round,rd as cat,nd as cos_sim,od as dot,sd as dynamic_time_warping,ad as env,id as full,ld as full_like,cd as getKeyValueShapes,dd as hamming,ud as hanning,pd as interpolate,hd as interpolate_4d,md as interpolate_data,_d as is_chinese_char,fd as layer_norm,gd as log_softmax,Md as magnitude,wd as matmul,Td as max,bd as mean,xd as mean_pooling,yd as medianFilter,kd as mel_filter_bank,Fd as min,Pd as ones,Cd as ones_like,vd as permute,Sd as permute_data,Ad as pipeline,Ed as quantize_embeddings,Ld as read_audio,zd as rfft,Id as round,Bd as softmax,Nd as spectrogram,Dd as stack,Od as std_mean,Vd as topk,jd as window_function,Rd as zeros,Gd as zeros_like};
 //# sourceMappingURL=transformers.min.mjs.map