@huggingface/transformers 3.5.0 → 3.5.2

package/dist/transformers.node.min.cjs

@@ -1,2 +1,2 @@
-(()=>{var e,t,s={fs:e=>{"use strict";e.exports=require("fs")},"onnxruntime-common":e=>{"use strict";e.exports=require("onnxruntime-common")},"onnxruntime-node":e=>{"use strict";e.exports=require("onnxruntime-node")},path:e=>{"use strict";e.exports=require("path")},sharp:e=>{"use strict";e.exports=require("sharp")},url:e=>{"use strict";e.exports=require("url")},"?8b6b":()=>{},"./node_modules/@huggingface/jinja/dist/index.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Environment:()=>Y,Interpreter:()=>K,Template:()=>te,parse:()=>D,tokenize:()=>d});var r=Object.freeze({Text:"Text",NumericLiteral:"NumericLiteral",BooleanLiteral:"BooleanLiteral",NullLiteral:"NullLiteral",StringLiteral:"StringLiteral",Identifier:"Identifier",Equals:"Equals",OpenParen:"OpenParen",CloseParen:"CloseParen",OpenStatement:"OpenStatement",CloseStatement:"CloseStatement",OpenExpression:"OpenExpression",CloseExpression:"CloseExpression",OpenSquareBracket:"OpenSquareBracket",CloseSquareBracket:"CloseSquareBracket",OpenCurlyBracket:"OpenCurlyBracket",CloseCurlyBracket:"CloseCurlyBracket",Comma:"Comma",Dot:"Dot",Colon:"Colon",Pipe:"Pipe",CallOperator:"CallOperator",AdditiveBinaryOperator:"AdditiveBinaryOperator",MultiplicativeBinaryOperator:"MultiplicativeBinaryOperator",ComparisonBinaryOperator:"ComparisonBinaryOperator",UnaryOperator:"UnaryOperator",Set:"Set",If:"If",For:"For",In:"In",Is:"Is",NotIn:"NotIn",Else:"Else",EndSet:"EndSet",EndIf:"EndIf",ElseIf:"ElseIf",EndFor:"EndFor",And:"And",Or:"Or",Not:"UnaryOperator",Macro:"Macro",EndMacro:"EndMacro"}),o=Object.freeze({set:r.Set,for:r.For,in:r.In,is:r.Is,if:r.If,else:r.Else,endset:r.EndSet,endif:r.EndIf,elif:r.ElseIf,endfor:r.EndFor,and:r.And,or:r.Or,not:r.Not,"not in":r.NotIn,macro:r.Macro,endmacro:r.EndMacro,true:r.BooleanLiteral,false:r.BooleanLiteral,none:r.NullLiteral,True:r.BooleanLiteral,False:r.BooleanLiteral,None:r.NullLiteral}),n=class{constructor(e,t){this.value=e,this.type=t}};function i(e){return/\w/.test(e)}function a(e){return/[0-9]/.test(e)}var l=[["{%",r.OpenStatement],["%}",r.CloseStatement],["{{",r.OpenExpression],["}}",r.CloseExpression],["(",r.OpenParen],[")",r.CloseParen],["{",r.OpenCurlyBracket],["}",r.CloseCurlyBracket],["[",r.OpenSquareBracket],["]",r.CloseSquareBracket],[",",r.Comma],[".",r.Dot],[":",r.Colon],["|",r.Pipe],["<=",r.ComparisonBinaryOperator],[">=",r.ComparisonBinaryOperator],["==",r.ComparisonBinaryOperator],["!=",r.ComparisonBinaryOperator],["<",r.ComparisonBinaryOperator],[">",r.ComparisonBinaryOperator],["+",r.AdditiveBinaryOperator],["-",r.AdditiveBinaryOperator],["*",r.MultiplicativeBinaryOperator],["/",r.MultiplicativeBinaryOperator],["%",r.MultiplicativeBinaryOperator],["=",r.Equals]],c=new Map([["n","\n"],["t","\t"],["r","\r"],["b","\b"],["f","\f"],["v","\v"],["'","'"],['"','"'],["\\","\\"]]);function d(e,t={}){const s=[],d=function(e,t={}){return e.endsWith("\n")&&(e=e.slice(0,-1)),e=e.replace(/{#.*?#}/gs,"{##}"),t.lstrip_blocks&&(e=e.replace(/^[ \t]*({[#%])/gm,"$1")),t.trim_blocks&&(e=e.replace(/([#%]})\n/g,"$1")),e.replace(/{##}/g,"").replace(/-%}\s*/g,"%}").replace(/\s*{%-/g,"{%").replace(/-}}\s*/g,"}}").replace(/\s*{{-/g,"{{")}(e,t);let u=0;const _=e=>{let t="";for(;e(d[u]);)if("\\"!==d[u]){if(t+=d[u++],u>=d.length)throw new SyntaxError("Unexpected end of input")}else{if(++u,u>=d.length)throw new SyntaxError("Unexpected end of input");const e=d[u++],s=c.get(e);if(void 0===s)throw new SyntaxError(`Unexpected escaped character: ${e}`);t+=s}return t};e:for(;u<d.length;){const e=s.at(-1)?.type;if(void 0===e||e===r.CloseStatement||e===r.CloseExpression){let e="";for(;u<d.length&&("{"!==d[u]||"%"!==d[u+1]&&"{"!==d[u+1]);)e+=d[u++];if(e.length>0){s.push(new n(e,r.Text));continue}}_((e=>/\s/.test(e)));const t=d[u];if("-"===t||"+"===t){const e=s.at(-1)?.type;if(e===r.Text||void 0===e)throw new SyntaxError(`Unexpected character: ${t}`);switch(e){case r.Identifier:case r.NumericLiteral:case r.BooleanLiteral:case r.NullLiteral:case r.StringLiteral:case r.CloseParen:case r.CloseSquareBracket:break;default:{++u;const e=_(a);s.push(new n(`${t}${e}`,e.length>0?r.NumericLiteral:r.UnaryOperator));continue}}}for(const[e,t]of l){if(d.slice(u,u+e.length)===e){s.push(new n(e,t)),u+=e.length;continue e}}if("'"!==t&&'"'!==t)if(a(t)){const e=_(a);s.push(new n(e,r.NumericLiteral))}else{if(!i(t))throw new SyntaxError(`Unexpected character: ${t}`);{const e=_(i),t=Object.hasOwn(o,e)?o[e]:r.Identifier;t===r.In&&s.at(-1)?.type===r.Not?(s.pop(),s.push(new n("not in",r.NotIn))):s.push(new n(e,t))}}else{++u;const e=_((e=>e!==t));s.push(new n(e,r.StringLiteral)),++u}}return s}var u=class{type="Statement"},_=class extends u{constructor(e){super(),this.body=e}type="Program"},p=class extends u{constructor(e,t,s){super(),this.test=e,this.body=t,this.alternate=s}type="If"},m=class extends u{constructor(e,t,s,r){super(),this.loopvar=e,this.iterable=t,this.body=s,this.defaultBlock=r}type="For"},h=class extends u{constructor(e,t,s){super(),this.assignee=e,this.value=t,this.body=s}type="Set"},g=class extends u{constructor(e,t,s){super(),this.name=e,this.args=t,this.body=s}type="Macro"},f=class extends u{type="Expression"},w=class extends f{constructor(e,t,s){super(),this.object=e,this.property=t,this.computed=s}type="MemberExpression"},M=class extends f{constructor(e,t){super(),this.callee=e,this.args=t}type="CallExpression"},x=class extends f{constructor(e){super(),this.value=e}type="Identifier"},b=class extends f{constructor(e){super(),this.value=e}type="Literal"},k=class extends b{type="NumericLiteral"},y=class extends b{type="StringLiteral"},v=class extends b{type="BooleanLiteral"},T=class extends b{type="NullLiteral"},P=class extends b{type="ArrayLiteral"},F=class extends b{type="TupleLiteral"},C=class extends b{type="ObjectLiteral"},S=class extends f{constructor(e,t,s){super(),this.operator=e,this.left=t,this.right=s}type="BinaryExpression"},A=class extends f{constructor(e,t){super(),this.operand=e,this.filter=t}type="FilterExpression"},E=class extends f{constructor(e,t){super(),this.iterable=e,this.test=t}type="SelectExpression"},L=class extends f{constructor(e,t,s){super(),this.operand=e,this.negate=t,this.test=s}type="TestExpression"},I=class extends f{constructor(e,t){super(),this.operator=e,this.argument=t}type="UnaryExpression"},z=class extends f{constructor(e=void 0,t=void 0,s=void 0){super(),this.start=e,this.stop=t,this.step=s}type="SliceExpression"},j=class extends f{constructor(e,t){super(),this.key=e,this.value=t}type="KeywordArgumentExpression"};function D(e){const t=new _([]);let s=0;function o(t,r){const o=e[s++];if(!o||o.type!==t)throw new Error(`Parser Error: ${r}. ${o.type} !== ${t}.`);return o}function n(){switch(e[s].type){case r.Text:return new y(o(r.Text,"Expected text token").value);case r.OpenStatement:return function(){let t;switch(o(r.OpenStatement,"Expected opening statement token"),e[s].type){case r.Set:++s,t=function(){const t=d();if(a(r.Equals)){++s;const e=d();return new h(t,e,[])}{const i=[];for(o(r.CloseStatement,"Expected %} token");e[s]?.type!==r.OpenStatement||e[s+1]?.type!==r.EndSet;){const e=n();i.push(e)}return o(r.OpenStatement,"Expected {% token"),o(r.EndSet,"Expected endset token"),new h(t,null,i)}}(),o(r.CloseStatement,"Expected closing statement token");break;case r.If:++s,t=l(),o(r.OpenStatement,"Expected {% token"),o(r.EndIf,"Expected endif token"),o(r.CloseStatement,"Expected %} token");break;case r.Macro:++s,t=function(){const e=q();if("Identifier"!==e.type)throw new SyntaxError("Expected identifier following macro statement");const t=V();o(r.CloseStatement,"Expected closing statement token");const s=[];for(;i(r.OpenStatement,r.EndMacro);)s.push(n());return new g(e,t,s)}(),o(r.OpenStatement,"Expected {% token"),o(r.EndMacro,"Expected endmacro token"),o(r.CloseStatement,"Expected %} token");break;case r.For:++s,t=function(){const e=c(!0);if(!(e instanceof x||e instanceof F))throw new SyntaxError(`Expected identifier/tuple for the loop variable, got ${e.type} instead`);o(r.In,"Expected `in` keyword following loop variable");const t=d();o(r.CloseStatement,"Expected closing statement token");const l=[];for(;i(r.OpenStatement,r.EndFor)&&i(r.OpenStatement,r.Else);)l.push(n());const u=[];if(a(r.OpenStatement,r.Else))for(++s,++s,o(r.CloseStatement,"Expected closing statement token");i(r.OpenStatement,r.EndFor);)u.push(n());return new m(e,t,l,u)}(),o(r.OpenStatement,"Expected {% token"),o(r.EndFor,"Expected endfor token"),o(r.CloseStatement,"Expected %} token");break;default:throw new SyntaxError(`Unknown statement type: ${e[s].type}`)}return t}();case r.OpenExpression:return function(){o(r.OpenExpression,"Expected opening expression token");const e=d();return o(r.CloseExpression,"Expected closing expression token"),e}();default:throw new SyntaxError(`Unexpected token type: ${e[s].type}`)}}function i(...t){return s+t.length<=e.length&&t.some(((t,r)=>t!==e[s+r].type))}function a(...t){return s+t.length<=e.length&&t.every(((t,r)=>t===e[s+r].type))}function l(){const t=d();o(r.CloseStatement,"Expected closing statement token");const i=[],c=[];for(;e[s]?.type!==r.OpenStatement||e[s+1]?.type!==r.ElseIf&&e[s+1]?.type!==r.Else&&e[s+1]?.type!==r.EndIf;)i.push(n());if(e[s]?.type===r.OpenStatement&&e[s+1]?.type!==r.EndIf)if(++s,a(r.ElseIf))o(r.ElseIf,"Expected elseif token"),c.push(l());else for(o(r.Else,"Expected else token"),o(r.CloseStatement,"Expected closing statement token");e[s]?.type!==r.OpenStatement||e[s+1]?.type!==r.EndIf;)c.push(n());return new p(t,i,c)}function c(e=!1){const t=e?q:d,o=[t()],n=a(r.Comma);for(;n&&(++s,o.push(t()),a(r.Comma)););return n?new F(o):o[0]}function d(){return function(){const e=u();if(a(r.If)){++s;const t=u();if(a(r.Else)){++s;const r=u();return new p(t,[e],[r])}return new E(e,t)}return e}()}function u(){let t=f();for(;a(r.Or);){const r=e[s];++s;const o=f();t=new S(r,t,o)}return t}function f(){let t=b();for(;a(r.And);){const r=e[s];++s;const o=b();t=new S(r,t,o)}return t}function b(){let t;for(;a(r.Not);){const r=e[s];++s;const o=b();t=new I(r,o)}return t??function(){let t=D();for(;a(r.ComparisonBinaryOperator)||a(r.In)||a(r.NotIn);){const r=e[s];++s;const o=D();t=new S(r,t,o)}return t}()}function D(){let t=R();for(;a(r.AdditiveBinaryOperator);){const r=e[s];++s;const o=R();t=new S(r,t,o)}return t}function O(e){let t=new M(e,V());return t=B(t),a(r.OpenParen)&&(t=O(t)),t}function V(){o(r.OpenParen,"Expected opening parenthesis for arguments list");const e=function(){const e=[];for(;!a(r.CloseParen);){let t=d();if(a(r.Equals)){if(++s,!(t instanceof x))throw new SyntaxError("Expected identifier for keyword argument");const e=d();t=new j(t,e)}e.push(t),a(r.Comma)&&++s}return e}();return o(r.CloseParen,"Expected closing parenthesis for arguments list"),e}function N(){const e=[];let t=!1;for(;!a(r.CloseSquareBracket);)a(r.Colon)?(e.push(void 0),++s,t=!0):(e.push(d()),a(r.Colon)&&(++s,t=!0));if(0===e.length)throw new SyntaxError("Expected at least one argument for member/slice expression");if(t){if(e.length>3)throw new SyntaxError("Expected 0-3 arguments for slice expression");return new z(...e)}return e[0]}function B(t){for(;a(r.Dot)||a(r.OpenSquareBracket);){const n=e[s];let i;++s;const a=n.type!==r.Dot;if(a)i=N(),o(r.CloseSquareBracket,"Expected closing square bracket");else if(i=q(),"Identifier"!==i.type)throw new SyntaxError("Expected identifier following dot operator");t=new w(t,i,a)}return t}function R(){let t=G();for(;a(r.MultiplicativeBinaryOperator);){const r=e[s];++s;const o=G();t=new S(r,t,o)}return t}function G(){let e=function(){let e=function(){const e=B(q());return a(r.OpenParen)?O(e):e}();for(;a(r.Pipe);){++s;let t=q();if(!(t instanceof x))throw new SyntaxError("Expected identifier for the filter");a(r.OpenParen)&&(t=O(t)),e=new A(e,t)}return e}();for(;a(r.Is);){++s;const t=a(r.Not);t&&++s;let o=q();if(o instanceof v?o=new x(o.value.toString()):o instanceof T&&(o=new x("none")),!(o instanceof x))throw new SyntaxError("Expected identifier for the test");e=new L(e,t,o)}return e}function q(){const t=e[s];switch(t.type){case r.NumericLiteral:return++s,new k(Number(t.value));case r.StringLiteral:return++s,new y(t.value);case r.BooleanLiteral:return++s,new v("true"===t.value.toLowerCase());case r.NullLiteral:return++s,new T(null);case r.Identifier:return++s,new x(t.value);case r.OpenParen:{++s;const t=c();if(e[s].type!==r.CloseParen)throw new SyntaxError(`Expected closing parenthesis, got ${e[s].type} instead`);return++s,t}case r.OpenSquareBracket:{++s;const e=[];for(;!a(r.CloseSquareBracket);)e.push(d()),a(r.Comma)&&++s;return++s,new P(e)}case r.OpenCurlyBracket:{++s;const e=new Map;for(;!a(r.CloseCurlyBracket);){const t=d();o(r.Colon,"Expected colon between key and value in object literal");const n=d();e.set(t,n),a(r.Comma)&&++s}return++s,new C(e)}default:throw new SyntaxError(`Unexpected token: ${t.type}`)}}for(;s<e.length;)t.body.push(n());return t}function O(e,t,s=1){void 0===t&&(t=e,e=0);const r=[];for(let o=e;o<t;o+=s)r.push(o);return r}function V(e,t,s,r=1){const o=Math.sign(r);o>=0?(t=(t??=0)<0?Math.max(e.length+t,0):Math.min(t,e.length),s=(s??=e.length)<0?Math.max(e.length+s,0):Math.min(s,e.length)):(t=(t??=e.length-1)<0?Math.max(e.length+t,-1):Math.min(t,e.length-1),s=(s??=-1)<-1?Math.max(e.length+s,-1):Math.min(s,e.length-1));const n=[];for(let i=t;o*i<o*s;i+=r)n.push(e[i]);return n}function N(e){return e.replace(/\b\w/g,(e=>e.toUpperCase()))}var B=class{type="RuntimeValue";value;builtins=new Map;constructor(e=void 0){this.value=e}__bool__(){return new q(!!this.value)}},R=class extends B{type="NumericValue"},G=class extends B{type="StringValue";builtins=new Map([["upper",new Q((()=>new G(this.value.toUpperCase())))],["lower",new Q((()=>new G(this.value.toLowerCase())))],["strip",new Q((()=>new G(this.value.trim())))],["title",new Q((()=>new G(N(this.value))))],["length",new R(this.value.length)],["rstrip",new Q((()=>new G(this.value.trimEnd())))],["lstrip",new Q((()=>new G(this.value.trimStart())))],["split",new Q((e=>{const t=e[0]??new H;if(!(t instanceof G||t instanceof H))throw new Error("sep argument must be a string or null");const s=e[1]??new R(-1);if(!(s instanceof R))throw new Error("maxsplit argument must be a number");let r=[];if(t instanceof H){const e=this.value.trimStart();for(const{0:t,index:o}of e.matchAll(/\S+/g)){if(-1!==s.value&&r.length>=s.value&&void 0!==o){r.push(t+e.slice(o+t.length));break}r.push(t)}}else{if(""===t.value)throw new Error("empty separator");r=this.value.split(t.value),-1!==s.value&&r.length>s.value&&r.push(r.splice(s.value).join(t.value))}return new U(r.map((e=>new G(e))))}))]])},q=class extends B{type="BooleanValue"},$=class extends B{type="ObjectValue";__bool__(){return new q(this.value.size>0)}builtins=new Map([["get",new Q((([e,t])=>{if(!(e instanceof G))throw new Error(`Object key must be a string: got ${e.type}`);return this.value.get(e.value)??t??new H}))],["items",new Q((()=>new U(Array.from(this.value.entries()).map((([e,t])=>new U([new G(e),t]))))))]])},W=class extends ${type="KeywordArgumentsValue"},U=class extends B{type="ArrayValue";builtins=new Map([["length",new R(this.value.length)]]);__bool__(){return new q(this.value.length>0)}},X=class extends U{type="TupleValue"},Q=class extends B{type="FunctionValue"},H=class extends B{type="NullValue"},J=class extends B{type="UndefinedValue"},Y=class{constructor(e){this.parent=e}variables=new Map([["namespace",new Q((e=>{if(0===e.length)return new $(new Map);if(1!==e.length||!(e[0]instanceof $))throw new Error("`namespace` expects either zero arguments or a single object argument");return e[0]}))]]);tests=new Map([["boolean",e=>"BooleanValue"===e.type],["callable",e=>e instanceof Q],["odd",e=>{if("NumericValue"!==e.type)throw new Error(`Cannot apply test "odd" to type: ${e.type}`);return e.value%2!=0}],["even",e=>{if("NumericValue"!==e.type)throw new Error(`Cannot apply test "even" to type: ${e.type}`);return e.value%2==0}],["false",e=>"BooleanValue"===e.type&&!e.value],["true",e=>"BooleanValue"===e.type&&e.value],["none",e=>"NullValue"===e.type],["string",e=>"StringValue"===e.type],["number",e=>"NumericValue"===e.type],["integer",e=>"NumericValue"===e.type&&Number.isInteger(e.value)],["iterable",e=>"ArrayValue"===e.type||"StringValue"===e.type],["mapping",e=>"ObjectValue"===e.type],["lower",e=>{const t=e.value;return"StringValue"===e.type&&t===t.toLowerCase()}],["upper",e=>{const t=e.value;return"StringValue"===e.type&&t===t.toUpperCase()}],["none",e=>"NullValue"===e.type],["defined",e=>"UndefinedValue"!==e.type],["undefined",e=>"UndefinedValue"===e.type],["equalto",(e,t)=>e.value===t.value],["eq",(e,t)=>e.value===t.value]]);set(e,t){return this.declareVariable(e,Z(t))}declareVariable(e,t){if(this.variables.has(e))throw new SyntaxError(`Variable already declared: ${e}`);return this.variables.set(e,t),t}setVariable(e,t){return this.variables.set(e,t),t}resolve(e){if(this.variables.has(e))return this;if(this.parent)return this.parent.resolve(e);throw new Error(`Unknown variable: ${e}`)}lookupVariable(e){try{return this.resolve(e).variables.get(e)??new J}catch{return new J}}},K=class{global;constructor(e){this.global=e??new Y}run(e){return this.evaluate(e,this.global)}evaluateBinaryExpression(e,t){const s=this.evaluate(e.left,t);switch(e.operator.value){case"and":return s.__bool__().value?this.evaluate(e.right,t):s;case"or":return s.__bool__().value?s:this.evaluate(e.right,t)}const r=this.evaluate(e.right,t);switch(e.operator.value){case"==":return new q(s.value==r.value);case"!=":return new q(s.value!=r.value)}if(s instanceof J||r instanceof J)throw new Error("Cannot perform operation on undefined values");if(s instanceof H||r instanceof H)throw new Error("Cannot perform operation on null values");if(s instanceof R&&r instanceof R)switch(e.operator.value){case"+":return new R(s.value+r.value);case"-":return new R(s.value-r.value);case"*":return new R(s.value*r.value);case"/":return new R(s.value/r.value);case"%":return new R(s.value%r.value);case"<":return new q(s.value<r.value);case">":return new q(s.value>r.value);case">=":return new q(s.value>=r.value);case"<=":return new q(s.value<=r.value)}else if(s instanceof U&&r instanceof U){if("+"===e.operator.value)return new U(s.value.concat(r.value))}else if(r instanceof U){const t=void 0!==r.value.find((e=>e.value===s.value));switch(e.operator.value){case"in":return new q(t);case"not in":return new q(!t)}}if((s instanceof G||r instanceof G)&&"+"===e.operator.value)return new G(s.value.toString()+r.value.toString());if(s instanceof G&&r instanceof G)switch(e.operator.value){case"in":return new q(r.value.includes(s.value));case"not in":return new q(!r.value.includes(s.value))}if(s instanceof G&&r instanceof $)switch(e.operator.value){case"in":return new q(r.value.has(s.value));case"not in":return new q(!r.value.has(s.value))}throw new SyntaxError(`Unknown operator "${e.operator.value}" between ${s.type} and ${r.type}`)}evaluateArguments(e,t){const s=[],r=new Map;for(const o of e)if("KeywordArgumentExpression"===o.type){const e=o;r.set(e.key.value,this.evaluate(e.value,t))}else{if(r.size>0)throw new Error("Positional arguments must come before keyword arguments");s.push(this.evaluate(o,t))}return[s,r]}evaluateFilterExpression(e,t){const s=this.evaluate(e.operand,t);if("Identifier"===e.filter.type){const t=e.filter;if("tojson"===t.value)return new G(ee(s));if(s instanceof U)switch(t.value){case"list":return s;case"first":return s.value[0];case"last":return s.value[s.value.length-1];case"length":return new R(s.value.length);case"reverse":return new U(s.value.reverse());case"sort":return new U(s.value.sort(((e,t)=>{if(e.type!==t.type)throw new Error(`Cannot compare different types: ${e.type} and ${t.type}`);switch(e.type){case"NumericValue":return e.value-t.value;case"StringValue":return e.value.localeCompare(t.value);default:throw new Error(`Cannot compare type: ${e.type}`)}})));case"join":return new G(s.value.map((e=>e.value)).join(""));case"string":return new G(ee(s));default:throw new Error(`Unknown ArrayValue filter: ${t.value}`)}else if(s instanceof G)switch(t.value){case"length":return new R(s.value.length);case"upper":return new G(s.value.toUpperCase());case"lower":return new G(s.value.toLowerCase());case"title":return new G(N(s.value));case"capitalize":return new G(s.value.charAt(0).toUpperCase()+s.value.slice(1));case"trim":return new G(s.value.trim());case"indent":return new G(s.value.split("\n").map(((e,t)=>0===t||0===e.length?e:"    "+e)).join("\n"));case"join":case"string":return s;default:throw new Error(`Unknown StringValue filter: ${t.value}`)}else{if(s instanceof R){if("abs"===t.value)return new R(Math.abs(s.value));throw new Error(`Unknown NumericValue filter: ${t.value}`)}if(s instanceof $)switch(t.value){case"items":return new U(Array.from(s.value.entries()).map((([e,t])=>new U([new G(e),t]))));case"length":return new R(s.value.size);default:throw new Error(`Unknown ObjectValue filter: ${t.value}`)}}throw new Error(`Cannot apply filter "${t.value}" to type: ${s.type}`)}if("CallExpression"===e.filter.type){const r=e.filter;if("Identifier"!==r.callee.type)throw new Error(`Unknown filter: ${r.callee.type}`);const o=r.callee.value;if("tojson"===o){const[,e]=this.evaluateArguments(r.args,t),o=e.get("indent")??new H;if(!(o instanceof R||o instanceof H))throw new Error("If set, indent must be a number");return new G(ee(s,o.value))}if("join"===o){let e;if(s instanceof G)e=Array.from(s.value);else{if(!(s instanceof U))throw new Error(`Cannot apply filter "${o}" to type: ${s.type}`);e=s.value.map((e=>e.value))}const[n,i]=this.evaluateArguments(r.args,t),a=n.at(0)??i.get("separator")??new G("");if(!(a instanceof G))throw new Error("separator must be a string");return new G(e.join(a.value))}if(s instanceof U){switch(o){case"selectattr":case"rejectattr":{const e="selectattr"===o;if(s.value.some((e=>!(e instanceof $))))throw new Error(`\`${o}\` can only be applied to array of objects`);if(r.args.some((e=>"StringLiteral"!==e.type)))throw new Error(`arguments of \`${o}\` must be strings`);const[n,i,a]=r.args.map((e=>this.evaluate(e,t)));let l;if(i){const e=t.tests.get(i.value);if(!e)throw new Error(`Unknown test: ${i.value}`);l=e}else l=(...e)=>e[0].__bool__().value;const c=s.value.filter((t=>{const s=t.value.get(n.value),r=!!s&&l(s,a);return e?r:!r}));return new U(c)}case"map":{const[,e]=this.evaluateArguments(r.args,t);if(e.has("attribute")){const t=e.get("attribute");if(!(t instanceof G))throw new Error("attribute must be a string");const r=e.get("default"),o=s.value.map((e=>{if(!(e instanceof $))throw new Error("items in map must be an object");return e.value.get(t.value)??r??new J}));return new U(o)}throw new Error("`map` expressions without `attribute` set are not currently supported.")}}throw new Error(`Unknown ArrayValue filter: ${o}`)}if(s instanceof G){if("indent"===o){const[e,o]=this.evaluateArguments(r.args,t),n=e.at(0)??o.get("width")??new R(4);if(!(n instanceof R))throw new Error("width must be a number");const i=e.at(1)??o.get("first")??new q(!1),a=e.at(2)??o.get("blank")??new q(!1),l=s.value.split("\n"),c=" ".repeat(n.value),d=l.map(((e,t)=>!i.value&&0===t||!a.value&&0===e.length?e:c+e));return new G(d.join("\n"))}throw new Error(`Unknown StringValue filter: ${o}`)}throw new Error(`Cannot apply filter "${o}" to type: ${s.type}`)}throw new Error(`Unknown filter: ${e.filter.type}`)}evaluateTestExpression(e,t){const s=this.evaluate(e.operand,t),r=t.tests.get(e.test.value);if(!r)throw new Error(`Unknown test: ${e.test.value}`);const o=r(s);return new q(e.negate?!o:o)}evaluateUnaryExpression(e,t){const s=this.evaluate(e.argument,t);if("not"===e.operator.value)return new q(!s.value);throw new SyntaxError(`Unknown operator: ${e.operator.value}`)}evalProgram(e,t){return this.evaluateBlock(e.body,t)}evaluateBlock(e,t){let s="";for(const r of e){const e=this.evaluate(r,t);"NullValue"!==e.type&&"UndefinedValue"!==e.type&&(s+=e.value)}return new G(s)}evaluateIdentifier(e,t){return t.lookupVariable(e.value)}evaluateCallExpression(e,t){const[s,r]=this.evaluateArguments(e.args,t);r.size>0&&s.push(new W(r));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(s,t)}evaluateSliceExpression(e,t,s){if(!(e instanceof U||e instanceof G))throw new Error("Slice object must be an array or string");const r=this.evaluate(t.start,s),o=this.evaluate(t.stop,s),n=this.evaluate(t.step,s);if(!(r instanceof R||r instanceof J))throw new Error("Slice start must be numeric or undefined");if(!(o instanceof R||o instanceof J))throw new Error("Slice stop must be numeric or undefined");if(!(n instanceof R||n instanceof J))throw new Error("Slice step must be numeric or undefined");return e instanceof U?new U(V(e.value,r.value,o.value,n.value)):new G(V(Array.from(e.value),r.value,o.value,n.value).join(""))}evaluateMemberExpression(e,t){const s=this.evaluate(e.object,t);let r,o;if(e.computed){if("SliceExpression"===e.property.type)return this.evaluateSliceExpression(s,e.property,t);r=this.evaluate(e.property,t)}else r=new G(e.property.value);if(s instanceof $){if(!(r instanceof G))throw new Error(`Cannot access property with non-string: got ${r.type}`);o=s.value.get(r.value)??s.builtins.get(r.value)}else if(s instanceof U||s instanceof G)if(r instanceof R)o=s.value.at(r.value),s instanceof G&&(o=new G(s.value.at(r.value)));else{if(!(r instanceof G))throw new Error(`Cannot access property with non-string/non-number: got ${r.type}`);o=s.builtins.get(r.value)}else{if(!(r instanceof G))throw new Error(`Cannot access property with non-string: got ${r.type}`);o=s.builtins.get(r.value)}return o instanceof B?o:new J}evaluateSet(e,t){const s=e.value?this.evaluate(e.value,t):this.evaluateBlock(e.body,t);if("Identifier"===e.assignee.type){const r=e.assignee.value;t.setVariable(r,s)}else{if("MemberExpression"!==e.assignee.type)throw new Error(`Invalid LHS inside assignment expression: ${JSON.stringify(e.assignee)}`);{const r=e.assignee,o=this.evaluate(r.object,t);if(!(o instanceof $))throw new Error("Cannot assign to member of non-object");if("Identifier"!==r.property.type)throw new Error("Cannot assign to member with non-identifier property");o.value.set(r.property.value,s)}}return new H}evaluateIf(e,t){const s=this.evaluate(e.test,t);return this.evaluateBlock(s.__bool__().value?e.body:e.alternate,t)}evaluateFor(e,t){const s=new Y(t);let r,o;if("SelectExpression"===e.iterable.type){const t=e.iterable;o=this.evaluate(t.iterable,s),r=t.test}else o=this.evaluate(e.iterable,s);if(!(o instanceof U))throw new Error(`Expected iterable type in for loop: got ${o.type}`);const n=[],i=[];for(let t=0;t<o.value.length;++t){const a=new Y(s),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 s=l;if(t.value.length!==s.value.length)throw new Error(`Too ${t.value.length>s.value.length?"few":"many"} items to unpack`);c=e=>{for(let r=0;r<t.value.length;++r){if("Identifier"!==t.value[r].type)throw new Error(`Cannot unpack non-identifier type: ${t.value[r].type}`);e.setVariable(t.value[r].value,s.value[r])}}}}if(r){c(a);if(!this.evaluate(r,a).__bool__().value)continue}n.push(l),i.push(c)}let a="",l=!0;for(let t=0;t<n.length;++t){const r=new Map([["index",new R(t+1)],["index0",new R(t)],["revindex",new R(n.length-t)],["revindex0",new R(n.length-t-1)],["first",new q(0===t)],["last",new q(t===n.length-1)],["length",new R(n.length)],["previtem",t>0?n[t-1]:new J],["nextitem",t<n.length-1?n[t+1]:new J]]);s.setVariable("loop",new $(r)),i[t](s);a+=this.evaluateBlock(e.body,s).value,l=!1}if(l){a+=this.evaluateBlock(e.defaultBlock,s).value}return new G(a)}evaluateMacro(e,t){return t.setVariable(e.name.value,new Q(((t,s)=>{const r=new Y(s);let o;t=t.slice(),"KeywordArgumentsValue"===t.at(-1)?.type&&(o=t.pop());for(let s=0;s<e.args.length;++s){const n=e.args[s],i=t[s];if("Identifier"===n.type){const e=n;if(!i)throw new Error(`Missing positional argument: ${e.value}`);r.setVariable(e.value,i)}else{if("KeywordArgumentExpression"!==n.type)throw new Error(`Unknown argument type: ${n.type}`);{const e=n,t=i??o?.value.get(e.key.value)??this.evaluate(e.value,r);r.setVariable(e.key.value,t)}}}return this.evaluateBlock(e.body,r)}))),new H}evaluate(e,t){if(void 0===e)return new J;switch(e.type){case"Program":return this.evalProgram(e,t);case"Set":return this.evaluateSet(e,t);case"If":return this.evaluateIf(e,t);case"For":return this.evaluateFor(e,t);case"Macro":return this.evaluateMacro(e,t);case"NumericLiteral":return new R(Number(e.value));case"StringLiteral":return new G(e.value);case"BooleanLiteral":return new q(e.value);case"NullLiteral":return new H(e.value);case"ArrayLiteral":return new U(e.value.map((e=>this.evaluate(e,t))));case"TupleLiteral":return new X(e.value.map((e=>this.evaluate(e,t))));case"ObjectLiteral":{const s=new Map;for(const[r,o]of e.value){const e=this.evaluate(r,t);if(!(e instanceof G))throw new Error(`Object keys must be strings: got ${e.type}`);s.set(e.value,this.evaluate(o,t))}return new $(s)}case"Identifier":return this.evaluateIdentifier(e,t);case"CallExpression":return this.evaluateCallExpression(e,t);case"MemberExpression":return this.evaluateMemberExpression(e,t);case"UnaryExpression":return this.evaluateUnaryExpression(e,t);case"BinaryExpression":return this.evaluateBinaryExpression(e,t);case"FilterExpression":return this.evaluateFilterExpression(e,t);case"TestExpression":return this.evaluateTestExpression(e,t);default:throw new SyntaxError(`Unknown node type: ${e.type}`)}}};function Z(e){switch(typeof e){case"number":return new R(e);case"string":return new G(e);case"boolean":return new q(e);case"undefined":return new J;case"object":return null===e?new H:Array.isArray(e)?new U(e.map(Z)):new $(new Map(Object.entries(e).map((([e,t])=>[e,Z(t)]))));case"function":return new Q(((t,s)=>Z(e(...t.map((e=>e.value)))??null)));default:throw new Error(`Cannot convert to runtime value: ${e}`)}}function ee(e,t,s){const r=s??0;switch(e.type){case"NullValue":case"UndefinedValue":return"null";case"NumericValue":case"StringValue":case"BooleanValue":return JSON.stringify(e.value);case"ArrayValue":case"ObjectValue":{const s=t?" ".repeat(t):"",o="\n"+s.repeat(r),n=o+s;if("ArrayValue"===e.type){const s=e.value.map((e=>ee(e,t,r+1)));return t?`[${n}${s.join(`,${n}`)}${o}]`:`[${s.join(", ")}]`}{const s=Array.from(e.value.entries()).map((([e,s])=>{const o=`"${e}": ${ee(s,t,r+1)}`;return t?`${n}${o}`:o}));return t?`{${s.join(",")}${o}}`:`{${s.join(", ")}}`}}default:throw new Error(`Cannot convert to JSON: ${e.type}`)}}var te=class{parsed;constructor(e){const t=d(e,{lstrip_blocks:!0,trim_blocks:!0});this.parsed=D(t)}render(e){const t=new Y;if(t.set("false",!1),t.set("true",!0),t.set("raise_exception",(e=>{throw new Error(e)})),t.set("range",O),e)for(const[s,r]of Object.entries(e))t.set(s,r);return new K(t).run(this.parsed).value}}},"./src/backends/onnx.js":(e,t,s)=>{"use strict";var r,o;s.r(t),s.d(t,{Tensor:()=>l.Tensor,createInferenceSession:()=>f,deviceToExecutionProviders:()=>h,isONNXProxy:()=>x,isONNXTensor:()=>w});var n=s("./src/env.js"),i=s("onnxruntime-node"),a=s("?8b6b"),l=s("onnxruntime-common");const c=Object.freeze({auto:null,gpu:null,cpu:"cpu",wasm:"wasm",webgpu:"webgpu",cuda:"cuda",dml:"dml",webnn:{name:"webnn",deviceType:"cpu"},"webnn-npu":{name:"webnn",deviceType:"npu"},"webnn-gpu":{name:"webnn",deviceType:"gpu"},"webnn-cpu":{name:"webnn",deviceType:"cpu"}}),d=[];let u,_;const p=Symbol.for("onnxruntime");if(p in globalThis)_=globalThis[p];else if(n.apis.IS_NODE_ENV){switch(_=i??(r||(r=s.t(i,2))),process.platform){case"win32":d.push("dml");break;case"linux":"x64"===process.arch&&d.push("cuda")}d.push("cpu"),u=["cpu"]}else _=o||(o=s.t(a,2)),n.apis.IS_WEBNN_AVAILABLE&&d.push("webnn-npu","webnn-gpu","webnn-cpu","webnn"),n.apis.IS_WEBGPU_AVAILABLE&&d.push("webgpu"),d.push("wasm"),u=["wasm"];const m=_.InferenceSession;function h(e=null){if(!e)return u;switch(e){case"auto":return d;case"gpu":return d.filter((e=>["webgpu","cuda","dml","webnn-gpu"].includes(e)))}if(d.includes(e))return[c[e]??e];throw new Error(`Unsupported device: "${e}". Should be one of: ${d.join(", ")}.`)}let g=null;async function f(e,t,s){g&&await g;const r=m.create(e,t);g??=r;const o=await r;return o.config=s,o}function w(e){return e instanceof _.Tensor}const M=_?.env;function x(){return M?.wasm?.proxy}M?.wasm&&("undefined"!=typeof ServiceWorkerGlobalScope&&self instanceof ServiceWorkerGlobalScope||M.wasm.wasmPaths||(M.wasm.wasmPaths=`https://cdn.jsdelivr.net/npm/@huggingface/transformers@${n.env.version}/dist/`),M.wasm.proxy=!1),M?.webgpu&&(M.webgpu.powerPreference="high-performance"),n.env.backends.onnx=M},"./src/base/feature_extraction_utils.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{FeatureExtractor:()=>i,validate_audio_inputs:()=>a});var r=s("./src/utils/constants.js"),o=s("./src/utils/generic.js"),n=s("./src/utils/hub.js");class i extends o.Callable{constructor(e){super(),this.config=e}static async from_pretrained(e,t){return new this(await(0,n.getModelJSON)(e,r.FEATURE_EXTRACTOR_NAME,!0,t))}}function a(e,t){if(!(e instanceof Float32Array||e instanceof Float64Array))throw new Error(`${t} expects input to be a Float32Array or a Float64Array, but got ${e?.constructor?.name??typeof e} instead. If using the feature extractor directly, remember to use \`read_audio(url, sampling_rate)\` to obtain the raw audio data of the file/url.`)}},"./src/base/image_processors_utils.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ImageProcessor:()=>M,center_to_corners_format:()=>u,post_process_instance_segmentation:()=>w,post_process_object_detection:()=>_,post_process_panoptic_segmentation:()=>f,post_process_semantic_segmentation:()=>p});var r=s("./src/utils/generic.js"),o=s("./src/utils/tensor.js"),n=s("./src/utils/maths.js"),i=(s("./src/utils/image.js"),s("./src/utils/core.js")),a=s("./src/utils/hub.js"),l=s("./src/utils/constants.js");function c(e,t,s=0,r=null){const o=e/t;let i=(0,n.bankers_round)(o)*t;return null!==r&&i>r&&(i=Math.floor(o)*t),i<s&&(i=Math.ceil(o)*t),i}function d([e,t],s){return[Math.max(Math.floor(e/s),1)*s,Math.max(Math.floor(t/s),1)*s]}function u([e,t,s,r]){return[e-s/2,t-r/2,e+s/2,t+r/2]}function _(e,t=.5,s=null,r=!1){const o=e.logits,i=e.pred_boxes,[a,l,c]=o.dims;if(null!==s&&s.length!==a)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");let d=[];for(let e=0;e<a;++e){let a=null!==s?s[e]:null,_={boxes:[],classes:[],scores:[]},p=o[e],m=i[e];for(let e=0;e<l;++e){let s,o=p[e],i=[];if(r){s=o.sigmoid().data;for(let e=0;e<s.length;++e)s[e]>t&&i.push(e)}else{let e=(0,n.max)(o.data)[1];if(e===c-1)continue;if(s=(0,n.softmax)(o.data),s[e]<t)continue;i.push(e)}for(const t of i){let r=m[e].data;r=u(r),null!==a&&(r=r.map(((e,t)=>e*a[(t+1)%2]))),_.boxes.push(r),_.classes.push(t),_.scores.push(s[t])}}d.push(_)}return d}function p(e,t=null){const s=e.logits,r=s.dims[0];if(null!==t&&t.length!==r)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");const n=[];for(let e=0;e<r;++e){const r=null!==t?t[e]:null;let i=s[e];null!==r&&(i=(0,o.interpolate)(i,r,"bilinear",!1));const[a,l]=r??i.dims.slice(-2),c=new o.Tensor("int32",new Int32Array(a*l),[a,l]),d=i[0].data,u=c.data;for(let e=1;e<i.dims[0];++e){const t=i[e].data;for(let s=0;s<t.length;++s)t[s]>d[s]&&(d[s]=t[s],u[s]=e)}const _=new Array(i.dims[0]);for(let e=0;e<u.length;++e){const t=u[e];_[t]=t}const p=_.filter((e=>void 0!==e));n.push({segmentation:c,labels:p})}return n}function m(e,t,s,r){const o=[],i=[],a=[];for(let l=0;l<e.dims[0];++l){const c=e[l],d=t[l],u=(0,n.max)(c.data)[1];if(u===r)continue;const _=(0,n.softmax)(c.data)[u];_>s&&(o.push(d),i.push(_),a.push(u))}return[o,i,a]}function h(e,t,s,r=.5,o=.8){const n=[];let i=0,a=0;const l=t[s].data;for(let t=0;t<e.length;++t)e[t]===s&&(n.push(t),++i),l[t]>=r&&++a;let c=i>0&&a>0;if(c){c=i/a>o}return[c,n]}function g(e,t,s,r,n,i=null,a=null){const[l,c]=a??e[0].dims,d=new o.Tensor("int32",new Int32Array(l*c),[l,c]),u=[];if(null!==a)for(let t=0;t<e.length;++t)e[t]=(0,o.interpolate)(e[t],a,"bilinear",!1);const _=new Int32Array(e[0].data.length),p=new Float32Array(e[0].data.length);for(let s=0;s<e.length;++s){let r=t[s];const o=e[s].data;for(let e=0;e<o.length;++e)o[e]*=r,o[e]>p[e]&&(_[e]=s,p[e]=o[e])}let m=0;const g=d.data;for(let o=0;o<s.length;++o){const i=s[o],[a,l]=h(_,e,o,r,n);if(a){++m;for(const e of l)g[e]=m;u.push({id:m,label_id:i,score:t[o]})}}return[d,u]}function f(e,t=.5,s=.5,r=.8,n=null,i=null){null===n&&(console.warn("`label_ids_to_fuse` unset. No instance will be fused."),n=new Set);const a=e.class_queries_logits??e.logits,l=(e.masks_queries_logits??e.pred_masks).sigmoid();let[c,d,u]=a.dims;if(u-=1,null!==i&&i.length!==c)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");let _=[];for(let e=0;e<c;++e){let c=null!==i?i[e]:null,d=a[e],p=l[e],[h,f,w]=m(d,p,t,u);if(0===w.length){let[e,t]=c??p.dims.slice(-2),s=new o.Tensor("int32",new Int32Array(e*t).fill(-1),[e,t]);_.push({segmentation:s,segments_info:[]});continue}let[M,x]=g(h,f,w,s,r,n,c);_.push({segmentation:M,segments_info:x})}return _}function w(e,t=.5,s=null){throw new Error("`post_process_instance_segmentation` is not yet implemented.")}class M extends r.Callable{constructor(e){super(),this.image_mean=e.image_mean??e.mean,this.image_std=e.image_std??e.std,this.resample=e.resample??2,this.do_rescale=e.do_rescale??!0,this.rescale_factor=e.rescale_factor??1/255,this.do_normalize=e.do_normalize,this.do_thumbnail=e.do_thumbnail,this.size=e.size??e.image_size,this.do_resize=e.do_resize??void 0!==this.size,this.size_divisibility=e.size_divisibility??e.size_divisor,this.do_center_crop=e.do_center_crop,this.crop_size=e.crop_size,this.do_convert_rgb=e.do_convert_rgb??!0,this.do_crop_margin=e.do_crop_margin,this.pad_size=e.pad_size,this.do_pad=e.do_pad,this.do_pad&&!this.pad_size&&this.size&&void 0!==this.size.width&&void 0!==this.size.height&&(this.pad_size=this.size),this.do_flip_channel_order=e.do_flip_channel_order??!1,this.config=e}async thumbnail(e,t,s=2){const r=e.height,o=e.width,n=t.height,i=t.width;let a=Math.min(r,n),l=Math.min(o,i);return a===r&&l===o?e:(r>o?l=Math.floor(o*a/r):o>r&&(a=Math.floor(r*l/o)),await e.resize(l,a,{resample:s}))}async crop_margin(e,t=200){const s=e.clone().grayscale(),r=(0,n.min)(s.data)[0],o=(0,n.max)(s.data)[0]-r;if(0===o)return e;const i=t/255;let a=s.width,l=s.height,c=0,d=0;const u=s.data;for(let e=0;e<s.height;++e){const t=e*s.width;for(let n=0;n<s.width;++n)(u[t+n]-r)/o<i&&(a=Math.min(a,n),l=Math.min(l,e),c=Math.max(c,n),d=Math.max(d,e))}return e=await e.crop([a,l,c,d])}pad_image(e,t,s,{mode:r="constant",center:o=!1,constant_values:n=0}={}){const[a,l,c]=t;let d,u;if("number"==typeof s?(d=s,u=s):"square"===s?d=u=Math.max(a,l):(d=s.width,u=s.height),d!==l||u!==a){const s=new Float32Array(d*u*c);if(Array.isArray(n))for(let e=0;e<s.length;++e)s[e]=n[e%c];else 0!==n&&s.fill(n);const[_,p]=o?[Math.floor((d-l)/2),Math.floor((u-a)/2)]:[0,0];for(let t=0;t<a;++t){const r=(t+p)*d,o=t*l;for(let t=0;t<l;++t){const n=(r+t+_)*c,i=(o+t)*c;for(let t=0;t<c;++t)s[n+t]=e[i+t]}}if("symmetric"===r){if(o)throw new Error("`center` padding is not supported when `mode` is set to `symmetric`.");const t=a-1,r=l-1;for(let o=0;o<u;++o){const n=o*d,u=(0,i.calculateReflectOffset)(o,t)*l;for(let t=0;t<d;++t){if(o<a&&t<l)continue;const d=(n+t)*c,_=(u+(0,i.calculateReflectOffset)(t,r))*c;for(let t=0;t<c;++t)s[d+t]=e[_+t]}}}e=s,t=[u,d,c]}return[e,t]}rescale(e){for(let t=0;t<e.length;++t)e[t]=this.rescale_factor*e[t]}get_resize_output_image_size(e,t){const[s,r]=e.size;let o,n;if(this.do_thumbnail){const{height:e,width:s}=t;o=Math.min(e,s)}else Number.isInteger(t)?(o=t,n=this.config.max_size??o):void 0!==t&&(o=t.shortest_edge,n=t.longest_edge);if(void 0!==o||void 0!==n){const e=void 0===o?1:Math.max(o/s,o/r),t=s*e,i=r*e,a=void 0===n?1:Math.min(n/t,n/i);let l=Math.floor(Number((t*a).toFixed(2))),c=Math.floor(Number((i*a).toFixed(2)));return void 0!==this.size_divisibility&&([l,c]=d([l,c],this.size_divisibility)),[l,c]}if(void 0!==t&&void 0!==t.width&&void 0!==t.height){let e=t.width,o=t.height;if(this.config.keep_aspect_ratio&&this.config.ensure_multiple_of){let t=o/r,n=e/s;Math.abs(1-n)<Math.abs(1-t)?t=n:n=t,o=c(t*r,this.config.ensure_multiple_of),e=c(n*s,this.config.ensure_multiple_of)}return[e,o]}if(void 0!==this.size_divisibility)return d([s,r],this.size_divisibility);if(void 0!==t.min_pixels&&void 0!==t.max_pixels){const{min_pixels:e,max_pixels:o}=t;return function(e,t,s=28,r=3136,o=1003520){if(e<s||t<s)throw new Error(`height:${e} or width:${t} must be larger than factor:${s}`);if(Math.max(e,t)/Math.min(e,t)>200)throw new Error("absolute aspect ratio must be smaller than 200, got "+Math.max(e,t)/Math.min(e,t));let n=Math.round(e/s)*s,i=Math.round(t/s)*s;if(n*i>o){const r=Math.sqrt(e*t/o);n=Math.floor(e/r/s)*s,i=Math.floor(t/r/s)*s}else if(n*i<r){const o=Math.sqrt(r/(e*t));n=Math.ceil(e*o/s)*s,i=Math.ceil(t*o/s)*s}return[n,i]}(r,s,this.config.patch_size*this.config.merge_size,e,o)}throw new Error(`Could not resize image due to unsupported \`this.size\` option in config: ${JSON.stringify(t)}`)}async resize(e){const[t,s]=this.get_resize_output_image_size(e,this.size);return await e.resize(t,s,{resample:this.resample})}async preprocess(e,{do_normalize:t=null,do_pad:s=null,do_convert_rgb:r=null,do_convert_grayscale:n=null,do_flip_channel_order:i=null}={}){this.do_crop_margin&&(e=await this.crop_margin(e));const[a,l]=e.size;if(r??this.do_convert_rgb?e=e.rgb():n&&(e=e.grayscale()),this.do_resize&&(e=await this.resize(e)),this.do_thumbnail&&(e=await this.thumbnail(e,this.size,this.resample)),this.do_center_crop){let t,s;Number.isInteger(this.crop_size)?(t=this.crop_size,s=this.crop_size):(t=this.crop_size.width,s=this.crop_size.height),e=await e.center_crop(t,s)}const c=[e.height,e.width];let u=Float32Array.from(e.data),_=[e.height,e.width,e.channels];if(this.do_rescale&&this.rescale(u),t??this.do_normalize){let t=this.image_mean;Array.isArray(this.image_mean)||(t=new Array(e.channels).fill(t));let s=this.image_std;if(Array.isArray(this.image_std)||(s=new Array(e.channels).fill(t)),t.length!==e.channels||s.length!==e.channels)throw new Error(`When set to arrays, the length of \`image_mean\` (${t.length}) and \`image_std\` (${s.length}) must match the number of channels in the image (${e.channels}).`);for(let r=0;r<u.length;r+=e.channels)for(let o=0;o<e.channels;++o)u[r+o]=(u[r+o]-t[o])/s[o]}if(s??this.do_pad)if(this.pad_size){const t=this.pad_image(u,[e.height,e.width,e.channels],this.pad_size);[u,_]=t}else if(this.size_divisibility){const[e,t]=d([_[1],_[0]],this.size_divisibility);[u,_]=this.pad_image(u,_,{width:e,height:t})}if(i??this.do_flip_channel_order){if(3!==_[2])throw new Error("Flipping channel order is only supported for RGB images.");for(let e=0;e<u.length;e+=3){const t=u[e];u[e]=u[e+2],u[e+2]=t}}return{original_size:[l,a],reshaped_input_size:c,pixel_values:new o.Tensor("float32",u,_).permute(2,0,1)}}async _call(e,...t){Array.isArray(e)||(e=[e]);const s=await Promise.all(e.map((e=>this.preprocess(e))));return{pixel_values:(0,o.stack)(s.map((e=>e.pixel_values)),0),original_sizes:s.map((e=>e.original_size)),reshaped_input_sizes:s.map((e=>e.reshaped_input_size))}}static async from_pretrained(e,t){return new this(await(0,a.getModelJSON)(e,l.IMAGE_PROCESSOR_NAME,!0,t))}}},"./src/base/processing_utils.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Processor:()=>i});var r=s("./src/utils/constants.js"),o=s("./src/utils/generic.js"),n=s("./src/utils/hub.js");class i extends o.Callable{static classes=["image_processor_class","tokenizer_class","feature_extractor_class"];static uses_processor_config=!1;constructor(e,t){super(),this.config=e,this.components=t}get image_processor(){return this.components.image_processor}get tokenizer(){return this.components.tokenizer}get feature_extractor(){return this.components.feature_extractor}apply_chat_template(e,t={}){if(!this.tokenizer)throw new Error("Unable to apply chat template without a tokenizer.");return this.tokenizer.apply_chat_template(e,{tokenize:!1,...t})}batch_decode(...e){if(!this.tokenizer)throw new Error("Unable to decode without a tokenizer.");return this.tokenizer.batch_decode(...e)}decode(...e){if(!this.tokenizer)throw new Error("Unable to decode without a tokenizer.");return this.tokenizer.decode(...e)}async _call(e,...t){for(const s of[this.image_processor,this.feature_extractor,this.tokenizer])if(s)return s(e,...t);throw new Error("No image processor, feature extractor, or tokenizer found.")}static async from_pretrained(e,t){const[s,o]=await Promise.all([this.uses_processor_config?(0,n.getModelJSON)(e,r.PROCESSOR_NAME,!0,t):{},Promise.all(this.classes.filter((e=>e in this)).map((async s=>{const r=await this[s].from_pretrained(e,t);return[s.replace(/_class$/,""),r]}))).then(Object.fromEntries)]);return new this(s,o)}}},"./src/configs.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{AutoConfig:()=>l,PretrainedConfig:()=>a,getKeyValueShapes:()=>i});var r=s("./src/utils/core.js"),o=s("./src/utils/hub.js");function n(e){const t={};let s={};switch(e.model_type){case"llava":case"paligemma":case"gemma3":case"florence2":case"llava_onevision":case"idefics3":case"ultravox":case"smolvlm":s=n(e.text_config);break;case"moondream1":s=n(e.phi_config);break;case"musicgen":s=n(e.decoder);break;case"multi_modality":s=n(e.language_config);break;case"gpt2":case"gptj":case"jais":case"codegen":case"gpt_bigcode":t.num_heads="n_head",t.num_layers="n_layer",t.hidden_size="n_embd";break;case"gpt_neox":case"stablelm":case"opt":case"falcon":t.num_heads="num_attention_heads",t.num_layers="num_hidden_layers",t.hidden_size="hidden_size";break;case"llama":case"olmo":case"olmo2":case"mobilellm":case"granite":case"cohere":case"mistral":case"starcoder2":case"qwen2":case"qwen2_vl":case"phi":case"phi3":case"phi3_v":t.num_heads="num_key_value_heads",t.num_layers="num_hidden_layers",t.hidden_size="hidden_size",t.num_attention_heads="num_attention_heads";break;case"gemma":case"gemma2":case"gemma3_text":case"glm":case"helium":t.num_heads="num_key_value_heads",t.num_layers="num_hidden_layers",t.dim_kv="head_dim";break;case"openelm":t.num_heads="num_kv_heads",t.num_layers="num_transformer_layers",t.dim_kv="head_dim";break;case"gpt_neo":case"donut-swin":t.num_heads="num_heads",t.num_layers="num_layers",t.hidden_size="hidden_size";break;case"bloom":t.num_heads="n_head",t.num_layers="n_layer",t.hidden_size="hidden_size";break;case"mpt":t.num_heads="n_heads",t.num_layers="n_layers",t.hidden_size="d_model";break;case"exaone":t.num_heads="num_key_value_heads",t.num_layers="num_layers",t.dim_kv="head_dim",t.num_attention_heads="num_attention_heads";break;case"t5":case"mt5":case"longt5":t.num_decoder_layers="num_decoder_layers",t.num_decoder_heads="num_heads",t.decoder_dim_kv="d_kv",t.num_encoder_layers="num_layers",t.num_encoder_heads="num_heads",t.encoder_dim_kv="d_kv";break;case"bart":case"mbart":case"marian":case"whisper":case"lite-whisper":case"m2m_100":case"blenderbot":case"blenderbot-small":case"florence2_language":t.num_decoder_layers="decoder_layers",t.num_decoder_heads="decoder_attention_heads",t.decoder_hidden_size="d_model",t.num_encoder_layers="encoder_layers",t.num_encoder_heads="encoder_attention_heads",t.encoder_hidden_size="d_model";break;case"speecht5":t.num_decoder_layers="decoder_layers",t.num_decoder_heads="decoder_attention_heads",t.decoder_hidden_size="hidden_size",t.num_encoder_layers="encoder_layers",t.num_encoder_heads="encoder_attention_heads",t.encoder_hidden_size="hidden_size";break;case"trocr":t.num_encoder_layers=t.num_decoder_layers="decoder_layers",t.num_encoder_heads=t.num_decoder_heads="decoder_attention_heads",t.encoder_hidden_size=t.decoder_hidden_size="d_model";break;case"musicgen_decoder":t.num_encoder_layers=t.num_decoder_layers="num_hidden_layers",t.num_encoder_heads=t.num_decoder_heads="num_attention_heads",t.encoder_hidden_size=t.decoder_hidden_size="hidden_size";break;case"moonshine":t.num_decoder_layers="decoder_num_hidden_layers",t.num_decoder_heads="decoder_num_key_value_heads",t.num_encoder_layers="encoder_num_hidden_layers",t.num_encoder_heads="encoder_num_key_value_heads",t.encoder_hidden_size=t.decoder_hidden_size="hidden_size";break;case"vision-encoder-decoder":const o=n(e.decoder),i="num_decoder_layers"in o,a=(0,r.pick)(e,["model_type","is_encoder_decoder"]);return i?(a.num_decoder_layers=o.num_decoder_layers,a.num_decoder_heads=o.num_decoder_heads,a.decoder_hidden_size=o.decoder_hidden_size,a.num_encoder_layers=o.num_encoder_layers,a.num_encoder_heads=o.num_encoder_heads,a.encoder_hidden_size=o.encoder_hidden_size):(a.num_layers=o.num_layers,a.num_heads=o.num_heads,a.hidden_size=o.hidden_size),a}const o={...s,...(0,r.pick)(e,["model_type","multi_query","is_encoder_decoder"])};for(const s in t)o[s]=e[t[s]];return o}function i(e,{prefix:t="past_key_values",batch_size:s=1}={}){const r={},o=e.normalized_config;if(o.is_encoder_decoder&&"num_encoder_heads"in o&&"num_decoder_heads"in o){const e=o.encoder_dim_kv??o.encoder_hidden_size/o.num_encoder_heads,n=o.decoder_dim_kv??o.decoder_hidden_size/o.num_decoder_heads,i=[s,o.num_encoder_heads,0,e],a=[s,o.num_decoder_heads,0,n];for(let e=0;e<o.num_decoder_layers;++e)r[`${t}.${e}.encoder.key`]=i,r[`${t}.${e}.encoder.value`]=i,r[`${t}.${e}.decoder.key`]=a,r[`${t}.${e}.decoder.value`]=a}else{const e=o.num_heads,n=o.num_layers,i=o.dim_kv??o.hidden_size/(o.num_attention_heads??e);if("falcon"===o.model_type){const o=[s*e,0,i];for(let e=0;e<n;++e)r[`${t}.${e}.key`]=o,r[`${t}.${e}.value`]=o}else if(o.multi_query){const o=[s*e,0,2*i];for(let e=0;e<n;++e)r[`${t}.${e}.key_value`]=o}else if("bloom"===o.model_type){const o=[s*e,i,0],a=[s*e,0,i];for(let e=0;e<n;++e)r[`${t}.${e}.key`]=o,r[`${t}.${e}.value`]=a}else if("openelm"===o.model_type)for(let o=0;o<n;++o){const n=[s,e[o],0,i];r[`${t}.${o}.key`]=n,r[`${t}.${o}.value`]=n}else{const o=[s,e,0,i];for(let e=0;e<n;++e)r[`${t}.${e}.key`]=o,r[`${t}.${e}.value`]=o}}return r}class a{model_type=null;is_encoder_decoder=!1;max_position_embeddings;"transformers.js_config";constructor(e){Object.assign(this,e),this.normalized_config=n(this)}static async from_pretrained(e,{progress_callback:t=null,config:s=null,cache_dir:r=null,local_files_only:n=!1,revision:i="main"}={}){!s||s instanceof a||(s=new a(s));const l=s??await async function(e,t){return await(0,o.getModelJSON)(e,"config.json",!0,t)}(e,{progress_callback:t,config:s,cache_dir:r,local_files_only:n,revision:i});return new this(l)}}class l{static async from_pretrained(...e){return a.from_pretrained(...e)}}},"./src/env.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{apis:()=>h,env:()=>x});var r=s("fs"),o=s("path"),n=s("url");const i="undefined"!=typeof window&&void 0!==window.document,a="undefined"!=typeof self&&"DedicatedWorkerGlobalScope"===self.constructor?.name,l="undefined"!=typeof self&&"caches"in self,c="undefined"!=typeof navigator&&"gpu"in navigator,d="undefined"!=typeof navigator&&"ml"in navigator,u="undefined"!=typeof process,_=u&&"node"===process?.release?.name,p=!b(r),m=!b(o),h=Object.freeze({IS_BROWSER_ENV:i,IS_WEBWORKER_ENV:a,IS_WEB_CACHE_AVAILABLE:l,IS_WEBGPU_AVAILABLE:c,IS_WEBNN_AVAILABLE:d,IS_PROCESS_AVAILABLE:u,IS_NODE_ENV:_,IS_FS_AVAILABLE:p,IS_PATH_AVAILABLE:m}),g=p&&m;let f="./";if(g){const e=Object({}).url;e?f=o.dirname(o.dirname(n.fileURLToPath(e))):"undefined"!=typeof __dirname&&(f=o.dirname(__dirname))}const w=g?o.join(f,"/.cache/"):null,M="/models/",x={version:"3.5.0",backends:{onnx:{}},allowRemoteModels:!0,remoteHost:"https://huggingface.co/",remotePathTemplate:"{model}/resolve/{revision}/",allowLocalModels:!(i||a),localModelPath:g?o.join(f,M):M,useFS:p,useBrowserCache:l,useFSCache:p,cacheDir:w,useCustomCache:!1,customCache:null};function b(e){return 0===Object.keys(e).length}},"./src/generation/configuration_utils.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{GenerationConfig:()=>o});var r=s("./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;streamer=null;begin_suppress_tokens=null;forced_decoder_ids=null;guidance_scale=null;num_return_sequences=1;output_attentions=!1;output_hidden_states=!1;output_scores=!1;return_dict_in_generate=!1;pad_token_id=null;bos_token_id=null;eos_token_id=null;encoder_no_repeat_ngram_size=0;decoder_start_token_id=null;generation_kwargs={};constructor(e){Object.assign(this,(0,r.pick)(e,Object.getOwnPropertyNames(this)))}}},"./src/generation/logits_process.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ClassifierFreeGuidanceLogitsProcessor:()=>f,ForcedBOSTokenLogitsProcessor:()=>l,ForcedEOSTokenLogitsProcessor:()=>c,LogitsProcessor:()=>n,LogitsProcessorList:()=>a,LogitsWarper:()=>i,MinLengthLogitsProcessor:()=>m,MinNewTokensLengthLogitsProcessor:()=>h,NoBadWordsLogitsProcessor:()=>g,NoRepeatNGramLogitsProcessor:()=>_,RepetitionPenaltyLogitsProcessor:()=>p,SuppressTokensAtBeginLogitsProcessor:()=>d,TemperatureLogitsWarper:()=>w,TopKLogitsWarper:()=>x,TopPLogitsWarper:()=>M,WhisperTimeStampLogitsProcessor:()=>u});var r=s("./src/utils/generic.js"),o=(s("./src/utils/tensor.js"),s("./src/utils/maths.js"));class n extends r.Callable{_call(e,t){throw Error("`_call` should be implemented in a subclass")}}class i extends r.Callable{_call(e,t){throw Error("`_call` should be implemented in a subclass")}}class a extends r.Callable{constructor(){super(),this.processors=[]}push(e){this.processors.push(e)}extend(e){this.processors.push(...e)}_call(e,t){let s=t;for(const t of this.processors)s=t(e,s);return s}[Symbol.iterator](){return this.processors.values()}}class l extends n{constructor(e){super(),this.bos_token_id=e}_call(e,t){for(let s=0;s<e.length;++s)if(1===e[s].length){const e=t[s].data;e.fill(-1/0),e[this.bos_token_id]=0}return t}}class c extends n{constructor(e,t){super(),this.max_length=e,this.eos_token_id=Array.isArray(t)?t:[t]}_call(e,t){for(let s=0;s<e.length;++s)if(e[s].length===this.max_length-1){const e=t[s].data;e.fill(-1/0);for(const t of this.eos_token_id)e[t]=0}return t}}class d extends n{constructor(e,t){super(),this.begin_suppress_tokens=e,this.begin_index=t}_call(e,t){for(let s=0;s<e.length;++s)if(e[s].length===this.begin_index){const e=t[s].data;for(const t of this.begin_suppress_tokens)e[t]=-1/0}return t}}class u extends n{constructor(e,t){super(),this.eos_token_id=Array.isArray(e.eos_token_id)?e.eos_token_id[0]:e.eos_token_id,this.no_timestamps_token_id=e.no_timestamps_token_id,this.timestamp_begin=this.no_timestamps_token_id+1,this.begin_index=t.length,t.at(-1)===this.no_timestamps_token_id&&(this.begin_index-=1),this.max_initial_timestamp_index=e.max_initial_timestamp_index}_call(e,t){for(let s=0;s<e.length;++s){const r=t[s].data;if(r[this.no_timestamps_token_id]=-1/0,e[s].length===this.begin_index-1){r.fill(-1/0),r[this.timestamp_begin]=0;continue}const n=e[s].slice(this.begin_index),i=n.length>=1&&n[n.length-1]>=this.timestamp_begin,a=n.length<2||n[n.length-2]>=this.timestamp_begin;if(i&&(a?r.subarray(this.timestamp_begin).fill(-1/0):r.subarray(0,this.eos_token_id).fill(-1/0)),e[s].length===this.begin_index&&null!==this.max_initial_timestamp_index){const e=this.timestamp_begin+this.max_initial_timestamp_index;r.subarray(e+1).fill(-1/0)}const l=(0,o.log_softmax)(r);Math.log(l.subarray(this.timestamp_begin).map(Math.exp).reduce(((e,t)=>e+t)))>(0,o.max)(l.subarray(0,this.timestamp_begin))[0]&&r.subarray(0,this.timestamp_begin).fill(-1/0)}return t}}class _ extends n{constructor(e){super(),this.no_repeat_ngram_size=e}getNgrams(e){const t=e.length,s=[];for(let r=0;r<t+1-this.no_repeat_ngram_size;++r){const t=[];for(let s=0;s<this.no_repeat_ngram_size;++s)t.push(e[r+s]);s.push(t.map(Number))}const r=new Map;for(const e of s){const t=e.slice(0,e.length-1),s=JSON.stringify(t),o=r.get(s)??[];o.push(e[e.length-1]),r.set(s,o)}return r}getGeneratedNgrams(e,t){const s=t.slice(t.length+1-this.no_repeat_ngram_size,t.length);return e.get(JSON.stringify(s.map(Number)))??[]}calcBannedNgramTokens(e){const t=[];if(e.length+1<this.no_repeat_ngram_size)return t;{const t=this.getNgrams(e);return this.getGeneratedNgrams(t,e)}}_call(e,t){for(let s=0;s<e.length;++s){const r=t[s].data,o=this.calcBannedNgramTokens(e[s]);for(const e of o)r[e]=-1/0}return t}}class p extends n{constructor(e){super(),this.penalty=e}_call(e,t){for(let s=0;s<e.length;++s){const r=t[s].data;for(const t of new Set(e[s])){const e=Number(t);r[e]<0?r[e]*=this.penalty:r[e]/=this.penalty}}return t}}class m extends n{constructor(e,t){super(),this.min_length=e,this.eos_token_id=Array.isArray(t)?t:[t]}_call(e,t){for(let s=0;s<e.length;++s)if(e[s].length<this.min_length){const e=t[s].data;for(const t of this.eos_token_id)e[t]=-1/0}return t}}class h extends n{constructor(e,t,s){super(),this.prompt_length_to_skip=e,this.min_new_tokens=t,this.eos_token_id=Array.isArray(s)?s:[s]}_call(e,t){for(let s=0;s<e.length;++s){if(e[s].length-this.prompt_length_to_skip<this.min_new_tokens){const e=t[s].data;for(const t of this.eos_token_id)e[t]=-1/0}}return t}}class g extends n{constructor(e,t){super(),this.bad_words_ids=e,this.eos_token_id=Array.isArray(t)?t:[t]}_call(e,t){for(let s=0;s<e.length;++s){const r=t[s].data,o=e[s];for(const e of this.bad_words_ids){if(o.length<e.length-1)continue;let t=!0;for(let s=1;s<=e.length-1;++s)if(e.at(-s-1)!=o.at(-s)){t=!1;break}t&&(r[e.at(-1)]=-1/0)}}return t}}class f extends n{constructor(e){if(super(),e<=1)throw new Error(`Require guidance scale >1 to use the classifier free guidance processor, got guidance scale ${e}.`);this.guidance_scale=e}_call(e,t){if(t.dims[0]!==2*e.length)throw new Error(`Logits should have twice the batch size of the input ids, the first half of batches corresponding to the conditional inputs, and the second half of batches corresponding to the unconditional inputs. Got batch size ${t.dims[0]} for the logits and ${e.length} for the input ids.`);const s=e.length,r=t.slice([0,s],null),o=t.slice([s,t.dims[0]],null);for(let e=0;e<o.data.length;++e)o.data[e]+=(r.data[e]-o.data[e])*this.guidance_scale;return o}}class w extends i{constructor(e){if(super(),"number"!=typeof e||e<=0){let t=`\`temperature\` (=${e}) must be a strictly positive float, otherwise your next token scores will be invalid.`;0===e&&(t+=" If you're looking for greedy decoding strategies, set `do_sample=false`.")}this.temperature=e}_call(e,t){const s=t.data;for(let e=0;e<s.length;++e)s[e]/=this.temperature;return t}}class M extends i{constructor(e,{filter_value:t=-1/0,min_tokens_to_keep:s=1}={}){if(super(),e<0||e>1)throw new Error(`\`top_p\` must be a float > 0 and < 1, but is ${e}`);if(!Number.isInteger(s)||s<1)throw new Error(`\`min_tokens_to_keep\` must be a positive integer, but is ${s}`);this.top_p=e,this.filter_value=t,this.min_tokens_to_keep=s}}class x extends i{constructor(e,{filter_value:t=-1/0,min_tokens_to_keep:s=1}={}){if(super(),!Number.isInteger(e)||e<0)throw new Error(`\`top_k\` must be a positive integer, but is ${e}`);this.top_k=Math.max(e,s),this.filter_value=t}}},"./src/generation/logits_sampler.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{LogitsSampler:()=>i});var r=s("./src/utils/generic.js"),o=s("./src/utils/tensor.js"),n=s("./src/utils/maths.js");s("./src/generation/configuration_utils.js");class i extends r.Callable{constructor(e){super(),this.generation_config=e}async _call(e){return this.sample(e)}async sample(e){throw Error("sample should be implemented in subclasses.")}getLogits(e,t){let s=e.dims.at(-1),r=e.data;if(-1===t)r=r.slice(-s);else{let e=t*s;r=r.slice(e,e+s)}return r}randomSelect(e){let t=0;for(let s=0;s<e.length;++s)t+=e[s];let s=Math.random()*t;for(let t=0;t<e.length;++t)if(s-=e[t],s<=0)return t;return 0}static getSampler(e){if(e.do_sample)return new l(e);if(e.num_beams>1)return new c(e);if(e.num_return_sequences>1)throw Error(`num_return_sequences has to be 1 when doing greedy search, but is ${e.num_return_sequences}.`);return new a(e)}}class a extends i{async sample(e){const t=(0,n.max)(e.data)[1];return[[BigInt(t),0]]}}class l extends i{async sample(e){let t=e.dims.at(-1);this.generation_config.top_k>0&&(t=Math.min(this.generation_config.top_k,t));const[s,r]=await(0,o.topk)(e,t),i=(0,n.softmax)(s.data);return Array.from({length:this.generation_config.num_beams},(()=>{const e=this.randomSelect(i);return[r.data[e],Math.log(i[e])]}))}}class c extends i{async sample(e){let t=e.dims.at(-1);this.generation_config.top_k>0&&(t=Math.min(this.generation_config.top_k,t));const[s,r]=await(0,o.topk)(e,t),i=(0,n.softmax)(s.data);return Array.from({length:this.generation_config.num_beams},((e,t)=>[r.data[t],Math.log(i[t])]))}}},"./src/generation/stopping_criteria.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{EosTokenCriteria:()=>a,InterruptableStoppingCriteria:()=>l,MaxLengthCriteria:()=>i,StoppingCriteria:()=>o,StoppingCriteriaList:()=>n});var r=s("./src/utils/generic.js");class o extends r.Callable{_call(e,t){throw Error("StoppingCriteria needs to be subclassed")}}class n extends r.Callable{constructor(){super(),this.criteria=[]}push(e){this.criteria.push(e)}extend(e){e instanceof n?e=e.criteria:e instanceof o&&(e=[e]),this.criteria.push(...e)}_call(e,t){const s=new Array(e.length).fill(!1);for(const r of this.criteria){const o=r(e,t);for(let e=0;e<s.length;++e)s[e]||=o[e]}return s}[Symbol.iterator](){return this.criteria.values()}}class i extends o{constructor(e,t=null){super(),this.max_length=e,this.max_position_embeddings=t}_call(e){return e.map((e=>e.length>=this.max_length))}}class a extends o{constructor(e){super(),Array.isArray(e)||(e=[e]),this.eos_token_id=e}_call(e,t){return e.map((e=>{const t=e.at(-1);return this.eos_token_id.some((e=>t==e))}))}}class l extends o{constructor(){super(),this.interrupted=!1}interrupt(){this.interrupted=!0}reset(){this.interrupted=!1}_call(e,t){return new Array(e.length).fill(this.interrupted)}}},"./src/generation/streamers.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{BaseStreamer:()=>i,TextStreamer:()=>l,WhisperTextStreamer:()=>c});var r=s("./src/utils/core.js"),o=s("./src/tokenizers.js"),n=s("./src/env.js");class i{put(e){throw Error("Not implemented")}end(){throw Error("Not implemented")}}const a=n.apis.IS_PROCESS_AVAILABLE?e=>process.stdout.write(e):e=>console.log(e);class l extends i{constructor(e,{skip_prompt:t=!1,callback_function:s=null,token_callback_function:r=null,skip_special_tokens:o=!0,decode_kwargs:n={},...i}={}){super(),this.tokenizer=e,this.skip_prompt=t,this.callback_function=s??a,this.token_callback_function=r,this.decode_kwargs={skip_special_tokens:o,...n,...i},this.token_cache=[],this.print_len=0,this.next_tokens_are_prompt=!0}put(e){if(e.length>1)throw Error("TextStreamer only supports batch size of 1");const t=this.next_tokens_are_prompt;if(t&&(this.next_tokens_are_prompt=!1,this.skip_prompt))return;const s=e[0];this.token_callback_function?.(s),this.token_cache=(0,r.mergeArrays)(this.token_cache,s);const n=this.tokenizer.decode(this.token_cache,this.decode_kwargs);let i;t||n.endsWith("\n")?(i=n.slice(this.print_len),this.token_cache=[],this.print_len=0):n.length>0&&(0,o.is_chinese_char)(n.charCodeAt(n.length-1))?(i=n.slice(this.print_len),this.print_len+=i.length):(i=n.slice(this.print_len,n.lastIndexOf(" ")+1),this.print_len+=i.length),this.on_finalized_text(i,!1)}end(){let e;if(this.token_cache.length>0){e=this.tokenizer.decode(this.token_cache,this.decode_kwargs).slice(this.print_len),this.token_cache=[],this.print_len=0}else e="";this.next_tokens_are_prompt=!0,this.on_finalized_text(e,!0)}on_finalized_text(e,t){e.length>0&&this.callback_function?.(e),t&&this.callback_function===a&&n.apis.IS_PROCESS_AVAILABLE&&this.callback_function?.("\n")}}class c extends l{constructor(e,{skip_prompt:t=!1,callback_function:s=null,token_callback_function:r=null,on_chunk_start:o=null,on_chunk_end:n=null,on_finalize:i=null,time_precision:a=.02,skip_special_tokens:l=!0,decode_kwargs:c={}}={}){super(e,{skip_prompt:t,skip_special_tokens:l,callback_function:s,token_callback_function:r,decode_kwargs:c}),this.timestamp_begin=e.timestamp_begin,this.on_chunk_start=o,this.on_chunk_end=n,this.on_finalize=i,this.time_precision=a,this.waiting_for_timestamp=!1}put(e){if(e.length>1)throw Error("WhisperTextStreamer only supports batch size of 1");const t=e[0];if(1===t.length){const s=Number(t[0])-this.timestamp_begin;if(s>=0){const t=s*this.time_precision;this.waiting_for_timestamp?this.on_chunk_end?.(t):this.on_chunk_start?.(t),this.waiting_for_timestamp=!this.waiting_for_timestamp,e=[[]]}}return super.put(e)}end(){super.end(),this.on_finalize?.()}}},"./src/models.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ASTForAudioClassification:()=>bs,ASTModel:()=>xs,ASTPreTrainedModel:()=>Ms,AlbertForMaskedLM:()=>At,AlbertForQuestionAnswering:()=>St,AlbertForSequenceClassification:()=>Ct,AlbertModel:()=>Ft,AlbertPreTrainedModel:()=>Pt,AutoModel:()=>Jc,AutoModelForAudioClassification:()=>hd,AutoModelForAudioFrameClassification:()=>fd,AutoModelForAudioTextToText:()=>Pd,AutoModelForCTC:()=>md,AutoModelForCausalLM:()=>rd,AutoModelForDepthEstimation:()=>bd,AutoModelForDocumentQuestionAnswering:()=>wd,AutoModelForImageClassification:()=>ad,AutoModelForImageFeatureExtraction:()=>vd,AutoModelForImageMatting:()=>Md,AutoModelForImageSegmentation:()=>ld,AutoModelForImageTextToText:()=>Td,AutoModelForImageToImage:()=>xd,AutoModelForMaskGeneration:()=>pd,AutoModelForMaskedLM:()=>od,AutoModelForNormalEstimation:()=>kd,AutoModelForObjectDetection:()=>ud,AutoModelForPoseEstimation:()=>yd,AutoModelForQuestionAnswering:()=>nd,AutoModelForSemanticSegmentation:()=>cd,AutoModelForSeq2SeqLM:()=>Zc,AutoModelForSequenceClassification:()=>Yc,AutoModelForSpeechSeq2Seq:()=>ed,AutoModelForTextToSpectrogram:()=>td,AutoModelForTextToWaveform:()=>sd,AutoModelForTokenClassification:()=>Kc,AutoModelForUniversalSegmentation:()=>dd,AutoModelForVision2Seq:()=>id,AutoModelForXVector:()=>gd,AutoModelForZeroShotObjectDetection:()=>_d,BartForConditionalGeneration:()=>Gt,BartForSequenceClassification:()=>qt,BartModel:()=>Rt,BartPretrainedModel:()=>Bt,BaseModelOutput:()=>re,BeitForImageClassification:()=>Mn,BeitModel:()=>wn,BeitPreTrainedModel:()=>fn,BertForMaskedLM:()=>ie,BertForQuestionAnswering:()=>ce,BertForSequenceClassification:()=>ae,BertForTokenClassification:()=>le,BertModel:()=>ne,BertPreTrainedModel:()=>oe,BlenderbotForConditionalGeneration:()=>Yt,BlenderbotModel:()=>Jt,BlenderbotPreTrainedModel:()=>Ht,BlenderbotSmallForConditionalGeneration:()=>es,BlenderbotSmallModel:()=>Zt,BlenderbotSmallPreTrainedModel:()=>Kt,BloomForCausalLM:()=>Co,BloomModel:()=>Fo,BloomPreTrainedModel:()=>Po,CLIPModel:()=>$s,CLIPPreTrainedModel:()=>qs,CLIPSegForImageSegmentation:()=>ar,CLIPSegModel:()=>ir,CLIPSegPreTrainedModel:()=>nr,CLIPTextModel:()=>Ws,CLIPTextModelWithProjection:()=>Us,CLIPVisionModel:()=>Xs,CLIPVisionModelWithProjection:()=>Qs,CamembertForMaskedLM:()=>Oe,CamembertForQuestionAnswering:()=>Be,CamembertForSequenceClassification:()=>Ve,CamembertForTokenClassification:()=>Ne,CamembertModel:()=>De,CamembertPreTrainedModel:()=>je,CausalLMOutput:()=>Id,CausalLMOutputWithPast:()=>zd,ChineseCLIPModel:()=>er,ChineseCLIPPreTrainedModel:()=>Zs,ClapAudioModelWithProjection:()=>el,ClapModel:()=>Ka,ClapPreTrainedModel:()=>Ya,ClapTextModelWithProjection:()=>Za,CodeGenForCausalLM:()=>Cr,CodeGenModel:()=>Fr,CodeGenPreTrainedModel:()=>Pr,CohereForCausalLM:()=>to,CohereModel:()=>eo,CoherePreTrainedModel:()=>Zr,ConvBertForMaskedLM:()=>Te,ConvBertForQuestionAnswering:()=>Ce,ConvBertForSequenceClassification:()=>Pe,ConvBertForTokenClassification:()=>Fe,ConvBertModel:()=>ve,ConvBertPreTrainedModel:()=>ye,ConvNextForImageClassification:()=>Si,ConvNextModel:()=>Ci,ConvNextPreTrainedModel:()=>Fi,ConvNextV2ForImageClassification:()=>Li,ConvNextV2Model:()=>Ei,ConvNextV2PreTrainedModel:()=>Ai,DPTForDepthEstimation:()=>ii,DPTModel:()=>ni,DPTPreTrainedModel:()=>oi,DacDecoderModel:()=>nc,DacDecoderOutput:()=>sc,DacEncoderModel:()=>oc,DacEncoderOutput:()=>tc,DacModel:()=>rc,DacPreTrainedModel:()=>ec,DebertaForMaskedLM:()=>qe,DebertaForQuestionAnswering:()=>Ue,DebertaForSequenceClassification:()=>$e,DebertaForTokenClassification:()=>We,DebertaModel:()=>Ge,DebertaPreTrainedModel:()=>Re,DebertaV2ForMaskedLM:()=>He,DebertaV2ForQuestionAnswering:()=>Ke,DebertaV2ForSequenceClassification:()=>Je,DebertaV2ForTokenClassification:()=>Ye,DebertaV2Model:()=>Qe,DebertaV2PreTrainedModel:()=>Xe,DecisionTransformerModel:()=>zl,DecisionTransformerPreTrainedModel:()=>Il,DeiTForImageClassification:()=>$n,DeiTModel:()=>qn,DeiTPreTrainedModel:()=>Gn,DepthAnythingForDepthEstimation:()=>li,DepthAnythingPreTrainedModel:()=>ai,DepthProForDepthEstimation:()=>mi,DepthProPreTrainedModel:()=>pi,DetrForObjectDetection:()=>kn,DetrForSegmentation:()=>yn,DetrModel:()=>bn,DetrObjectDetectionOutput:()=>vn,DetrPreTrainedModel:()=>xn,DetrSegmentationOutput:()=>Tn,Dinov2ForImageClassification:()=>ji,Dinov2Model:()=>zi,Dinov2PreTrainedModel:()=>Ii,Dinov2WithRegistersForImageClassification:()=>Vi,Dinov2WithRegistersModel:()=>Oi,Dinov2WithRegistersPreTrainedModel:()=>Di,DistilBertForMaskedLM:()=>ot,DistilBertForQuestionAnswering:()=>rt,DistilBertForSequenceClassification:()=>tt,DistilBertForTokenClassification:()=>st,DistilBertModel:()=>et,DistilBertPreTrainedModel:()=>Ze,DonutSwinModel:()=>Pi,DonutSwinPreTrainedModel:()=>Ti,EfficientNetForImageClassification:()=>_l,EfficientNetModel:()=>ul,EfficientNetPreTrainedModel:()=>dl,ElectraForMaskedLM:()=>Ee,ElectraForQuestionAnswering:()=>ze,ElectraForSequenceClassification:()=>Le,ElectraForTokenClassification:()=>Ie,ElectraModel:()=>Ae,ElectraPreTrainedModel:()=>Se,EsmForMaskedLM:()=>at,EsmForSequenceClassification:()=>lt,EsmForTokenClassification:()=>ct,EsmModel:()=>it,EsmPreTrainedModel:()=>nt,ExaoneForCausalLM:()=>Br,ExaoneModel:()=>Nr,ExaonePreTrainedModel:()=>Vr,FalconForCausalLM:()=>Ja,FalconModel:()=>Ha,FalconPreTrainedModel:()=>Qa,FastViTForImageClassification:()=>tn,FastViTModel:()=>en,FastViTPreTrainedModel:()=>Zo,Florence2ForConditionalGeneration:()=>js,Florence2PreTrainedModel:()=>zs,GLPNForDepthEstimation:()=>vi,GLPNModel:()=>yi,GLPNPreTrainedModel:()=>ki,GPT2LMHeadModel:()=>dr,GPT2Model:()=>cr,GPT2PreTrainedModel:()=>lr,GPTBigCodeForCausalLM:()=>Tr,GPTBigCodeModel:()=>vr,GPTBigCodePreTrainedModel:()=>yr,GPTJForCausalLM:()=>kr,GPTJModel:()=>br,GPTJPreTrainedModel:()=>xr,GPTNeoForCausalLM:()=>gr,GPTNeoModel:()=>hr,GPTNeoPreTrainedModel:()=>mr,GPTNeoXForCausalLM:()=>Mr,GPTNeoXModel:()=>wr,GPTNeoXPreTrainedModel:()=>fr,Gemma2ForCausalLM:()=>ao,Gemma2Model:()=>io,Gemma2PreTrainedModel:()=>no,Gemma3ForCausalLM:()=>uo,Gemma3Model:()=>co,Gemma3PreTrainedModel:()=>lo,GemmaForCausalLM:()=>oo,GemmaModel:()=>ro,GemmaPreTrainedModel:()=>so,GlmForCausalLM:()=>Or,GlmModel:()=>Dr,GlmPreTrainedModel:()=>jr,GraniteForCausalLM:()=>Kr,GraniteModel:()=>Yr,GranitePreTrainedModel:()=>Jr,GroundingDinoForObjectDetection:()=>Bi,GroundingDinoPreTrainedModel:()=>Ni,GroupViTModel:()=>Ko,GroupViTPreTrainedModel:()=>Yo,HeliumForCausalLM:()=>zr,HeliumModel:()=>Ir,HeliumPreTrainedModel:()=>Lr,HieraForImageClassification:()=>Xn,HieraModel:()=>Un,HieraPreTrainedModel:()=>Wn,HubertForCTC:()=>Ta,HubertForSequenceClassification:()=>Pa,HubertModel:()=>va,HubertPreTrainedModel:()=>ya,IJepaForImageClassification:()=>Bo,IJepaModel:()=>No,IJepaPreTrainedModel:()=>Vo,Idefics3ForConditionalGeneration:()=>Ns,Idefics3PreTrainedModel:()=>Vs,ImageMattingOutput:()=>jd,JAISLMHeadModel:()=>pr,JAISModel:()=>_r,JAISPreTrainedModel:()=>ur,JinaCLIPModel:()=>sr,JinaCLIPPreTrainedModel:()=>tr,JinaCLIPTextModel:()=>rr,JinaCLIPVisionModel:()=>or,LiteWhisperForConditionalGeneration:()=>Ts,LlamaForCausalLM:()=>Er,LlamaModel:()=>Ar,LlamaPreTrainedModel:()=>Sr,LlavaForConditionalGeneration:()=>Es,LlavaOnevisionForConditionalGeneration:()=>Ls,LlavaPreTrainedModel:()=>As,LongT5ForConditionalGeneration:()=>Dt,LongT5Model:()=>jt,LongT5PreTrainedModel:()=>zt,M2M100ForConditionalGeneration:()=>Zi,M2M100Model:()=>Ki,M2M100PreTrainedModel:()=>Yi,MBartForCausalLM:()=>Qt,MBartForConditionalGeneration:()=>Ut,MBartForSequenceClassification:()=>Xt,MBartModel:()=>Wt,MBartPreTrainedModel:()=>$t,MPNetForMaskedLM:()=>ft,MPNetForQuestionAnswering:()=>xt,MPNetForSequenceClassification:()=>wt,MPNetForTokenClassification:()=>Mt,MPNetModel:()=>gt,MPNetPreTrainedModel:()=>ht,MT5ForConditionalGeneration:()=>Nt,MT5Model:()=>Vt,MT5PreTrainedModel:()=>Ot,MarianMTModel:()=>Ji,MarianModel:()=>Hi,MarianPreTrainedModel:()=>Qi,MaskFormerForInstanceSegmentation:()=>bi,MaskFormerModel:()=>xi,MaskFormerPreTrainedModel:()=>Mi,MaskedLMOutput:()=>Ed,Metric3DForDepthEstimation:()=>gi,Metric3DPreTrainedModel:()=>hi,Metric3Dv2ForDepthEstimation:()=>wi,Metric3Dv2PreTrainedModel:()=>fi,MgpstrForSceneTextRecognition:()=>Nl,MgpstrModelOutput:()=>Ol,MgpstrPreTrainedModel:()=>Vl,MimiDecoderModel:()=>Zl,MimiDecoderOutput:()=>Jl,MimiEncoderModel:()=>Kl,MimiEncoderOutput:()=>Hl,MimiModel:()=>Yl,MimiPreTrainedModel:()=>Ql,MistralForCausalLM:()=>$a,MistralModel:()=>qa,MistralPreTrainedModel:()=>Ga,MobileBertForMaskedLM:()=>_t,MobileBertForQuestionAnswering:()=>mt,MobileBertForSequenceClassification:()=>pt,MobileBertModel:()=>ut,MobileBertPreTrainedModel:()=>dt,MobileLLMForCausalLM:()=>qr,MobileLLMModel:()=>Gr,MobileLLMPreTrainedModel:()=>Rr,MobileNetV1ForImageClassification:()=>Ml,MobileNetV1ForSemanticSegmentation:()=>xl,MobileNetV1Model:()=>wl,MobileNetV1PreTrainedModel:()=>fl,MobileNetV2ForImageClassification:()=>yl,MobileNetV2ForSemanticSegmentation:()=>vl,MobileNetV2Model:()=>kl,MobileNetV2PreTrainedModel:()=>bl,MobileNetV3ForImageClassification:()=>Fl,MobileNetV3ForSemanticSegmentation:()=>Cl,MobileNetV3Model:()=>Pl,MobileNetV3PreTrainedModel:()=>Tl,MobileNetV4ForImageClassification:()=>El,MobileNetV4ForSemanticSegmentation:()=>Ll,MobileNetV4Model:()=>Al,MobileNetV4PreTrainedModel:()=>Sl,MobileViTForImageClassification:()=>an,MobileViTModel:()=>nn,MobileViTPreTrainedModel:()=>on,MobileViTV2ForImageClassification:()=>dn,MobileViTV2Model:()=>cn,MobileViTV2PreTrainedModel:()=>ln,ModelOutput:()=>se,ModernBertForMaskedLM:()=>_e,ModernBertForSequenceClassification:()=>pe,ModernBertForTokenClassification:()=>me,ModernBertModel:()=>ue,ModernBertPreTrainedModel:()=>de,Moondream1ForConditionalGeneration:()=>Is,MoonshineForConditionalGeneration:()=>Cs,MoonshineModel:()=>Fs,MoonshinePreTrainedModel:()=>Ps,MptForCausalLM:()=>Eo,MptModel:()=>Ao,MptPreTrainedModel:()=>So,MultiModalityCausalLM:()=>Dl,MultiModalityPreTrainedModel:()=>jl,MusicgenForCausalLM:()=>hl,MusicgenForConditionalGeneration:()=>gl,MusicgenModel:()=>ml,MusicgenPreTrainedModel:()=>pl,NomicBertModel:()=>ge,NomicBertPreTrainedModel:()=>he,OPTForCausalLM:()=>zo,OPTModel:()=>Io,OPTPreTrainedModel:()=>Lo,Olmo2ForCausalLM:()=>Hr,Olmo2Model:()=>Qr,Olmo2PreTrainedModel:()=>Xr,OlmoForCausalLM:()=>Ur,OlmoModel:()=>Wr,OlmoPreTrainedModel:()=>$r,OpenELMForCausalLM:()=>mo,OpenELMModel:()=>po,OpenELMPreTrainedModel:()=>_o,OwlViTForObjectDetection:()=>pn,OwlViTModel:()=>_n,OwlViTPreTrainedModel:()=>un,Owlv2ForObjectDetection:()=>gn,Owlv2Model:()=>hn,Owlv2PreTrainedModel:()=>mn,PaliGemmaForConditionalGeneration:()=>Os,PaliGemmaPreTrainedModel:()=>Ds,PatchTSMixerForPrediction:()=>Wl,PatchTSMixerModel:()=>$l,PatchTSMixerPreTrainedModel:()=>ql,PatchTSTForPrediction:()=>Gl,PatchTSTModel:()=>Rl,PatchTSTPreTrainedModel:()=>Bl,Phi3ForCausalLM:()=>To,Phi3Model:()=>vo,Phi3PreTrainedModel:()=>yo,Phi3VForCausalLM:()=>Gs,Phi3VPreTrainedModel:()=>Rs,PhiForCausalLM:()=>ko,PhiModel:()=>bo,PhiPreTrainedModel:()=>xo,PreTrainedModel:()=>te,PretrainedMixin:()=>dc,PvtForImageClassification:()=>Wo,PvtModel:()=>$o,PvtPreTrainedModel:()=>qo,PyAnnoteForAudioFrameClassification:()=>aa,PyAnnoteModel:()=>ia,PyAnnotePreTrainedModel:()=>na,QuestionAnsweringModelOutput:()=>Ld,Qwen2ForCausalLM:()=>fo,Qwen2Model:()=>go,Qwen2PreTrainedModel:()=>ho,Qwen2VLForConditionalGeneration:()=>Mo,Qwen2VLPreTrainedModel:()=>wo,RFDetrForObjectDetection:()=>Dn,RFDetrModel:()=>jn,RFDetrObjectDetectionOutput:()=>On,RFDetrPreTrainedModel:()=>zn,RTDetrForObjectDetection:()=>Cn,RTDetrModel:()=>Fn,RTDetrObjectDetectionOutput:()=>Sn,RTDetrPreTrainedModel:()=>Pn,RTDetrV2ForObjectDetection:()=>Ln,RTDetrV2Model:()=>En,RTDetrV2ObjectDetectionOutput:()=>In,RTDetrV2PreTrainedModel:()=>An,ResNetForImageClassification:()=>Jn,ResNetModel:()=>Hn,ResNetPreTrainedModel:()=>Qn,RoFormerForMaskedLM:()=>Me,RoFormerForQuestionAnswering:()=>ke,RoFormerForSequenceClassification:()=>xe,RoFormerForTokenClassification:()=>be,RoFormerModel:()=>we,RoFormerPreTrainedModel:()=>fe,RobertaForMaskedLM:()=>rs,RobertaForQuestionAnswering:()=>is,RobertaForSequenceClassification:()=>os,RobertaForTokenClassification:()=>ns,RobertaModel:()=>ss,RobertaPreTrainedModel:()=>ts,SamImageSegmentationOutput:()=>Xi,SamModel:()=>Ui,SamPreTrainedModel:()=>Wi,SapiensForDepthEstimation:()=>ui,SapiensForNormalEstimation:()=>_i,SapiensForSemanticSegmentation:()=>di,SapiensPreTrainedModel:()=>ci,SegformerForImageClassification:()=>nl,SegformerForSemanticSegmentation:()=>il,SegformerModel:()=>ol,SegformerPreTrainedModel:()=>rl,Seq2SeqLMOutput:()=>Fd,SequenceClassifierOutput:()=>Cd,SiglipModel:()=>Js,SiglipPreTrainedModel:()=>Hs,SiglipTextModel:()=>Ys,SiglipVisionModel:()=>Ks,SmolVLMForConditionalGeneration:()=>Bs,SnacDecoderModel:()=>cc,SnacEncoderModel:()=>lc,SnacModel:()=>ac,SnacPreTrainedModel:()=>ic,SpeechT5ForSpeechToText:()=>Oa,SpeechT5ForTextToSpeech:()=>Va,SpeechT5HifiGan:()=>Na,SpeechT5Model:()=>Da,SpeechT5PreTrainedModel:()=>ja,SqueezeBertForMaskedLM:()=>yt,SqueezeBertForQuestionAnswering:()=>Tt,SqueezeBertForSequenceClassification:()=>vt,SqueezeBertModel:()=>kt,SqueezeBertPreTrainedModel:()=>bt,StableLmForCausalLM:()=>cl,StableLmModel:()=>ll,StableLmPreTrainedModel:()=>al,Starcoder2ForCausalLM:()=>Xa,Starcoder2Model:()=>Ua,Starcoder2PreTrainedModel:()=>Wa,StyleTextToSpeech2Model:()=>za,StyleTextToSpeech2PreTrainedModel:()=>Ia,Swin2SRForImageSuperResolution:()=>ri,Swin2SRModel:()=>si,Swin2SRPreTrainedModel:()=>ti,SwinForImageClassification:()=>Zn,SwinForSemanticSegmentation:()=>ei,SwinModel:()=>Kn,SwinPreTrainedModel:()=>Yn,T5ForConditionalGeneration:()=>It,T5Model:()=>Lt,T5PreTrainedModel:()=>Et,TableTransformerForObjectDetection:()=>Bn,TableTransformerModel:()=>Nn,TableTransformerObjectDetectionOutput:()=>Rn,TableTransformerPreTrainedModel:()=>Vn,TokenClassifierOutput:()=>Ad,TrOCRForCausalLM:()=>Ra,TrOCRPreTrainedModel:()=>Ba,UltravoxModel:()=>Xl,UltravoxPreTrainedModel:()=>Ul,UniSpeechForCTC:()=>_a,UniSpeechForSequenceClassification:()=>pa,UniSpeechModel:()=>ua,UniSpeechPreTrainedModel:()=>da,UniSpeechSatForAudioFrameClassification:()=>wa,UniSpeechSatForCTC:()=>ga,UniSpeechSatForSequenceClassification:()=>fa,UniSpeechSatModel:()=>ha,UniSpeechSatPreTrainedModel:()=>ma,ViTForImageClassification:()=>Oo,ViTMAEModel:()=>Xo,ViTMAEPreTrainedModel:()=>Uo,ViTMSNForImageClassification:()=>Jo,ViTMSNModel:()=>Ho,ViTMSNPreTrainedModel:()=>Qo,ViTModel:()=>Do,ViTPreTrainedModel:()=>jo,VisionEncoderDecoderModel:()=>Ss,VitMatteForImageMatting:()=>rn,VitMattePreTrainedModel:()=>sn,VitPoseForPoseEstimation:()=>Go,VitPosePreTrainedModel:()=>Ro,VitsModel:()=>sl,VitsModelOutput:()=>Dd,VitsPreTrainedModel:()=>tl,Wav2Vec2BertForCTC:()=>ba,Wav2Vec2BertForSequenceClassification:()=>ka,Wav2Vec2BertModel:()=>xa,Wav2Vec2BertPreTrainedModel:()=>Ma,Wav2Vec2ForAudioFrameClassification:()=>oa,Wav2Vec2ForCTC:()=>sa,Wav2Vec2ForSequenceClassification:()=>ra,Wav2Vec2Model:()=>ta,Wav2Vec2PreTrainedModel:()=>ea,WavLMForAudioFrameClassification:()=>La,WavLMForCTC:()=>Sa,WavLMForSequenceClassification:()=>Aa,WavLMForXVector:()=>Ea,WavLMModel:()=>Ca,WavLMPreTrainedModel:()=>Fa,WeSpeakerResNetModel:()=>ca,WeSpeakerResNetPreTrainedModel:()=>la,WhisperForConditionalGeneration:()=>vs,WhisperModel:()=>ys,WhisperPreTrainedModel:()=>ks,XLMForQuestionAnswering:()=>_s,XLMForSequenceClassification:()=>ds,XLMForTokenClassification:()=>us,XLMModel:()=>ls,XLMPreTrainedModel:()=>as,XLMRobertaForMaskedLM:()=>hs,XLMRobertaForQuestionAnswering:()=>ws,XLMRobertaForSequenceClassification:()=>gs,XLMRobertaForTokenClassification:()=>fs,XLMRobertaModel:()=>ms,XLMRobertaPreTrainedModel:()=>ps,XLMWithLMHeadModel:()=>cs,XVectorOutput:()=>Sd,YolosForObjectDetection:()=>qi,YolosModel:()=>Gi,YolosObjectDetectionOutput:()=>$i,YolosPreTrainedModel:()=>Ri});var r=s("./src/configs.js"),o=s("./src/backends/onnx.js"),n=s("./src/utils/dtypes.js"),i=s("./src/utils/generic.js"),a=s("./src/utils/core.js"),l=s("./src/utils/hub.js"),c=s("./src/utils/constants.js"),d=s("./src/generation/logits_process.js"),u=s("./src/generation/configuration_utils.js"),_=s("./src/utils/tensor.js"),p=s("./src/utils/image.js"),m=s("./src/utils/maths.js"),h=s("./src/generation/stopping_criteria.js"),g=s("./src/generation/logits_sampler.js"),f=s("./src/env.js"),w=s("./src/models/whisper/generation_whisper.js"),M=s("./src/models/whisper/common_whisper.js");const x=0,b=1,k=2,y=3,v=4,T=5,P=6,F=7,C=8,S=9,A=10,E=11,L=new Map,I=new Map,z=new Map;async function j(e,t,s){return Object.fromEntries(await Promise.all(Object.keys(t).map((async i=>{const{buffer_or_path:a,session_options:c,session_config:d}=await async function(e,t,s){let i=s.config?.["transformers.js_config"]??{},a=s.device??i.device;a&&"string"!=typeof a&&(a.hasOwnProperty(t)?a=a[t]:(console.warn(`device not specified for "${t}". Using the default device.`),a=null));const c=a??(f.apis.IS_NODE_ENV?"cpu":"wasm"),d=(0,o.deviceToExecutionProviders)(c),u=i.device_config??{};u.hasOwnProperty(c)&&(i={...i,...u[c]});let _=s.dtype??i.dtype;if("string"!=typeof _&&(_&&_.hasOwnProperty(t)?_=_[t]:(_=n.DEFAULT_DEVICE_DTYPE_MAPPING[c]??n.DATA_TYPES.fp32,console.warn(`dtype not specified for "${t}". Using the default dtype (${_}) for this device (${c}).`))),_===n.DATA_TYPES.auto){let e=i.dtype;"string"!=typeof e&&(e=e?.[t]),_=e&&e!==n.DATA_TYPES.auto&&n.DATA_TYPES.hasOwnProperty(e)?e:n.DEFAULT_DEVICE_DTYPE_MAPPING[c]??n.DATA_TYPES.fp32}const p=_;if(!n.DEFAULT_DTYPE_SUFFIX_MAPPING.hasOwnProperty(p))throw new Error(`Invalid dtype: ${p}. Should be one of: ${Object.keys(n.DATA_TYPES).join(", ")}`);if(p===n.DATA_TYPES.fp16&&"webgpu"===c&&!await(0,n.isWebGpuFp16Supported)())throw new Error(`The device (${c}) does not support fp16.`);const m=i.kv_cache_dtype,h=m?"string"==typeof m?m:m[p]??"float32":void 0;if(h&&!["float32","float16"].includes(h))throw new Error(`Invalid kv_cache_dtype: ${h}. Should be one of: float32, float16`);const g={dtype:p,kv_cache_dtype:h},w=`${t}${n.DEFAULT_DTYPE_SUFFIX_MAPPING[p]}.onnx`,M=`${s.subfolder??""}/${w}`,x={...s.session_options};x.executionProviders??=d;const b=i.free_dimension_overrides;b?x.freeDimensionOverrides??=b:c.startsWith("webnn")&&!x.freeDimensionOverrides&&console.warn(`WebNN does not currently support dynamic shapes and requires 'free_dimension_overrides' to be set in config.json, preferably as a field within config["transformers.js_config"]["device_config"]["${c}"]. When 'free_dimension_overrides' is not set, you may experience significant performance degradation.`);const k=f.apis.IS_NODE_ENV&&f.env.useFSCache,y=(0,l.getModelFile)(e,M,!0,s,k),v=s.use_external_data_format??i.use_external_data_format;let T=[];if(v){let r;r="object"==typeof v?v.hasOwnProperty(w)?v[w]:!!v.hasOwnProperty(t)&&v[t]:v;const o=+r;if(o>l.MAX_EXTERNAL_DATA_CHUNKS)throw new Error(`The number of external data chunks (${o}) exceeds the maximum allowed value (${l.MAX_EXTERNAL_DATA_CHUNKS}).`);for(let t=0;t<o;++t){const r=`${w}_data${0===t?"":"_"+t}`,o=`${s.subfolder??""}/${r}`;T.push(new Promise((async(t,n)=>{const i=await(0,l.getModelFile)(e,o,!0,s,k);t(i instanceof Uint8Array?{path:r,data:i}:r)})))}}else void 0!==x.externalData&&(T=x.externalData.map((async t=>{if("string"==typeof t.data){const r=await(0,l.getModelFile)(e,t.data,!0,s);return{...t,data:r}}return t})));if(T.length>0){const e=await Promise.all(T);f.apis.IS_NODE_ENV||(x.externalData=e)}if("webgpu"===c){const e=(0,r.getKeyValueShapes)(s.config,{prefix:"present"});if(Object.keys(e).length>0&&!(0,o.isONNXProxy)()){const t={};for(const s in e)t[s]="gpu-buffer";x.preferredOutputLocation=t}}return{buffer_or_path:await y,session_options:x,session_config:g}}(e,t[i],s);return[i,await(0,o.createInferenceSession)(a,c,d)]}))))}async function D(e,t,s){return Object.fromEntries(await Promise.all(Object.keys(t).map((async r=>[r,await(0,l.getModelJSON)(e,t[r],!1,s)]))))}async function O(e,t){const s=function(e,t){const s=Object.create(null),r=[];for(const n of e.inputNames){const e=t[n];e instanceof _.Tensor?s[n]=(0,o.isONNXProxy)()?e.clone():e:r.push(n)}if(r.length>0)throw new Error(`An error occurred during model execution: "Missing the following inputs: ${r.join(", ")}.`);const n=Object.keys(t).length,i=e.inputNames.length;if(n>i){let s=Object.keys(t).filter((t=>!e.inputNames.includes(t)));console.warn(`WARNING: Too many inputs were provided (${n} > ${i}). The following inputs will be ignored: "${s.join(", ")}".`)}return s}(e,t);try{const t=Object.fromEntries(Object.entries(s).map((([e,t])=>[e,t.ort_tensor])));let r=await e.run(t);return r=V(r),r}catch(e){const t=Object.fromEntries(Object.entries(s).map((([e,{type:t,dims:s,data:r}])=>[e,{type:t,dims:s,data:r}])));throw console.error(`An error occurred during model execution: "${e}".`),console.error("Inputs given to model:",t),e}}function V(e){for(let t in e)(0,o.isONNXTensor)(e[t])?e[t]=new _.Tensor(e[t]):"object"==typeof e[t]&&V(e[t]);return e}function N(e){if(e instanceof _.Tensor)return e;if(0===e.length)throw Error("items must be non-empty");if(Array.isArray(e[0])){if(e.some((t=>t.length!==e[0].length)))throw Error("Unable to create tensor, you should probably activate truncation and/or padding with 'padding=True' and/or 'truncation=True' to have batched tensors with the same length.");return new _.Tensor("int64",BigInt64Array.from(e.flat().map((e=>BigInt(e)))),[e.length,e[0].length])}return new _.Tensor("int64",BigInt64Array.from(e.map((e=>BigInt(e)))),[1,e.length])}function B(e){return new _.Tensor("bool",[e],[1])}async function R(e,t){let{encoder_outputs:s,input_ids:r,decoder_input_ids:o,...n}=t;if(!s){const r=(0,a.pick)(t,e.sessions.model.inputNames);s=(await G(e,r)).last_hidden_state}n.input_ids=o,n.encoder_hidden_states=s,e.sessions.decoder_model_merged.inputNames.includes("encoder_attention_mask")&&(n.encoder_attention_mask=t.attention_mask);return await $(e,n,!0)}async function G(e,t){const s=e.sessions.model,r=(0,a.pick)(t,s.inputNames);if(s.inputNames.includes("inputs_embeds")&&!r.inputs_embeds){if(!t.input_ids)throw new Error("Both `input_ids` and `inputs_embeds` are missing in the model inputs.");r.inputs_embeds=await e.encode_text({input_ids:t.input_ids})}if(s.inputNames.includes("token_type_ids")&&!r.token_type_ids){if(!r.input_ids)throw new Error("Both `input_ids` and `token_type_ids` are missing in the model inputs.");r.token_type_ids=(0,_.zeros_like)(r.input_ids)}if(s.inputNames.includes("pixel_mask")&&!r.pixel_mask){if(!r.pixel_values)throw new Error("Both `pixel_values` and `pixel_mask` are missing in the model inputs.");const e=r.pixel_values.dims;r.pixel_mask=(0,_.ones)([e[0],e[2],e[3]])}return await O(s,r)}async function q(e,t){const s=await e.encode(t);return await e.decode(s)}async function $(e,t,s=!1){const r=e.sessions[s?"decoder_model_merged":"model"],{past_key_values:o,...n}=t;if(r.inputNames.includes("use_cache_branch")&&(n.use_cache_branch=B(!!o)),r.inputNames.includes("position_ids")&&n.attention_mask&&!n.position_ids){const t=["paligemma","gemma3_text","gemma3"].includes(e.config.model_type)?1:0;n.position_ids=function(e,t=null,s=0){const{input_ids:r,inputs_embeds:o,attention_mask:n}=e,{data:i,dims:a}=J(n,s);let l=new _.Tensor("int64",i,a);if(t){const e=-(r??o).dims.at(1);l=l.slice(null,[e,null])}return l}(n,o,t)}e.addPastKeyValues(n,o);const i=(0,a.pick)(n,r.inputNames);return await O(r,i)}function W({modality_token_id:e,inputs_embeds:t,modality_features:s,input_ids:r,attention_mask:o}){const n=r.tolist().map((t=>t.reduce(((t,s,r)=>(s==e&&t.push(r),t)),[]))),i=n.reduce(((e,t)=>e+t.length),0),a=s.dims[0];if(i!==a)throw new Error(`Number of tokens and features do not match: tokens: ${i}, features ${a}`);let l=0;for(let e=0;e<n.length;++e){const r=n[e],o=t[e];for(let e=0;e<r.length;++e)o[r[e]].data.set(s[l++].data)}return{inputs_embeds:t,attention_mask:o}}function U({image_token_id:e,inputs_embeds:t,image_features:s,input_ids:r,attention_mask:o}){return W({modality_token_id:e,inputs_embeds:t,modality_features:s,input_ids:r,attention_mask:o})}async function X(e,{encode_function:t,merge_function:s,modality_input_name:r,modality_output_name:o,input_ids:n=null,attention_mask:i=null,position_ids:a=null,inputs_embeds:l=null,past_key_values:c=null,generation_config:d=null,logits_processor:u=null,...p}){const m=p[r];if(!l)if(l=await e.encode_text({input_ids:n,...p}),m&&1!==n.dims[1]){const e=await t({[r]:m,...p});({inputs_embeds:l,attention_mask:i}=s({[o]:e,inputs_embeds:l,input_ids:n,attention_mask:i}))}else if(c&&m&&1===n.dims[1]){const e=n.dims[1],t=Object.values(c)[0].dims.at(-2);i=(0,_.cat)([(0,_.ones)([n.dims[0],t]),i.slice(null,[i.dims[1]-e,i.dims[1]])],1)}if(!a&&"qwen2_vl"===e.config.model_type){const{image_grid_thw:t,video_grid_thw:s}=p;[a]=e.get_rope_index(n,t,s,i)}return await $(e,{inputs_embeds:l,past_key_values:c,attention_mask:i,position_ids:a,generation_config:d,logits_processor:u},!0)}async function Q(e,t){return await X(e,{...t,modality_input_name:"audio_values",modality_output_name:"audio_features",encode_function:e.encode_audio.bind(e),merge_function:e._merge_input_ids_with_audio_features.bind(e)})}async function H(e,t){return await X(e,{...t,modality_input_name:"pixel_values",modality_output_name:"image_features",encode_function:e.encode_image.bind(e),merge_function:e._merge_input_ids_with_image_features.bind(e)})}function J(e,t=0){const[s,r]=e.dims,o=e.data,n=new BigInt64Array(o.length);for(let e=0;e<s;++e){const s=e*r;let i=BigInt(t);for(let e=0;e<r;++e){const t=s+e;0n===o[t]?n[t]=BigInt(1):(n[t]=i,i+=o[t])}}return{data:n,dims:e.dims}}function Y(e,t,s,r){if(s.past_key_values){const t=Object.values(s.past_key_values)[0].dims.at(-2),{input_ids:r,attention_mask:o}=s;if(o&&o.dims[1]>r.dims[1]);else if(t<r.dims[1])s.input_ids=r.slice(null,[t,null]);else if(null!=e.config.image_token_index&&r.data.some((t=>t==e.config.image_token_index))){const o=e.config.num_image_tokens;if(!o)throw new Error("`num_image_tokens` is missing in the model configuration.");const n=r.dims[1]-(t-o);s.input_ids=r.slice(null,[-n,null]),s.attention_mask=(0,_.ones)([1,t+n])}}return s}function K(e,t,s,r){return s.past_key_values&&(t=t.map((e=>[e.at(-1)]))),{...s,decoder_input_ids:N(t)}}function Z(e,...t){return e.config.is_encoder_decoder?K(e,...t):Y(e,...t)}function ee(e,t,s,r){const o=!!s.past_key_values;if(null!==r.guidance_scale&&r.guidance_scale>1&&(o?s.input_ids=(0,_.cat)([s.input_ids,s.input_ids],0):(s.input_ids=(0,_.cat)([s.input_ids,(0,_.full_like)(s.input_ids,BigInt(r.pad_token_id))],0),s.attention_mask=(0,_.cat)([s.attention_mask,(0,_.full_like)(s.attention_mask,0n)],0))),!o&&s.pixel_values||(s.pixel_values=(0,_.full)([0,0,3,384,384],1)),o){const e=0,t=1,r=e>0?1:0,o=1;s.images_seq_mask=new _.Tensor("bool",new Array(e+t).fill(!0).fill(!1,0,t),[o,e+t]),s.images_emb_mask=new _.Tensor("bool",new Array(e).fill(!!r),[o,1,e])}return s}class te extends i.Callable{main_input_name="input_ids";forward_params=["input_ids","attention_mask"];constructor(e,t,s){super(),this.config=e,this.sessions=t,this.configs=s;const r=z.get(this.constructor),o=L.get(r);switch(this.can_generate=!1,this._forward=null,this._prepare_inputs_for_generation=null,o){case v:this.can_generate=!0,this._forward=$,this._prepare_inputs_for_generation=Y;break;case k:case y:case F:this.can_generate=!0,this._forward=R,this._prepare_inputs_for_generation=K;break;case b:this._forward=R;break;case P:this.can_generate=!0,this._forward=H,this._prepare_inputs_for_generation=Z;break;case A:this.can_generate=!0,this._forward=Q,this._prepare_inputs_for_generation=Z;break;case S:this.can_generate=!0,this._prepare_inputs_for_generation=Z;break;case C:this.can_generate=!0,this._prepare_inputs_for_generation=ee;break;case E:this._forward=q;break;default:this._forward=G}this.can_generate&&this.forward_params.push("past_key_values"),this.custom_config=this.config["transformers.js_config"]??{}}async dispose(){const e=[];for(const t of Object.values(this.sessions))t?.handler?.dispose&&e.push(t.handler.dispose());return await Promise.all(e)}static async from_pretrained(e,{progress_callback:t=null,config:s=null,cache_dir:o=null,local_files_only:n=!1,revision:i="main",model_file_name:a=null,subfolder:l="onnx",device:d=null,dtype:u=null,use_external_data_format:_=null,session_options:p={}}={}){let m={progress_callback:t,config:s,cache_dir:o,local_files_only:n,revision:i,model_file_name:a,subfolder:l,device:d,dtype:u,use_external_data_format:_,session_options:p};const h=z.get(this),g=L.get(h);let f;if(s=m.config=await r.AutoConfig.from_pretrained(e,m),g===v)f=await Promise.all([j(e,{model:m.model_file_name??"model"},m),D(e,{generation_config:"generation_config.json"},m)]);else if(g===k||g===y)f=await Promise.all([j(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},m),D(e,{generation_config:"generation_config.json"},m)]);else if(g===T)f=await Promise.all([j(e,{model:"vision_encoder",prompt_encoder_mask_decoder:"prompt_encoder_mask_decoder"},m)]);else if(g===b)f=await Promise.all([j(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},m)]);else if(g===P){const t={embed_tokens:"embed_tokens",vision_encoder:"vision_encoder",decoder_model_merged:"decoder_model_merged"};s.is_encoder_decoder&&(t.model="encoder_model"),f=await Promise.all([j(e,t,m),D(e,{generation_config:"generation_config.json"},m)])}else if(g===A){const t={embed_tokens:"embed_tokens",audio_encoder:"audio_encoder",decoder_model_merged:"decoder_model_merged"};f=await Promise.all([j(e,t,m),D(e,{generation_config:"generation_config.json"},m)])}else if(g===F)f=await Promise.all([j(e,{model:"text_encoder",decoder_model_merged:"decoder_model_merged",encodec_decode:"encodec_decode"},m),D(e,{generation_config:"generation_config.json"},m)]);else if(g===C)f=await Promise.all([j(e,{prepare_inputs_embeds:"prepare_inputs_embeds",model:"language_model",lm_head:"lm_head",gen_head:"gen_head",gen_img_embeds:"gen_img_embeds",image_decode:"image_decode"},m),D(e,{generation_config:"generation_config.json"},m)]);else if(g===S)f=await Promise.all([j(e,{prepare_inputs_embeds:"prepare_inputs_embeds",model:"model",vision_encoder:"vision_encoder"},m),D(e,{generation_config:"generation_config.json"},m)]);else if(g===E)f=await Promise.all([j(e,{encoder_model:"encoder_model",decoder_model:"decoder_model"},m)]);else{if(g!==x){const e=h??s?.model_type;"custom"!==e&&console.warn(`Model type for '${e}' not found, assuming encoder-only architecture. Please report this at ${c.GITHUB_ISSUE_URL}.`)}f=await Promise.all([j(e,{model:m.model_file_name??"model"},m)])}return new this(s,...f)}async _call(e){return await this.forward(e)}async forward(e){return await this._forward(this,e)}get generation_config(){return this.configs?.generation_config??null}_get_logits_warper(e){const t=new d.LogitsProcessorList;return null!==e.temperature&&1!==e.temperature&&t.push(new d.TemperatureLogitsWarper(e.temperature)),null!==e.top_k&&0!==e.top_k&&t.push(new d.TopKLogitsWarper(e.top_k)),null!==e.top_p&&e.top_p<1&&t.push(new d.TopPLogitsWarper(e.top_p)),t}_get_logits_processor(e,t,s=null){const r=new d.LogitsProcessorList;if(null!==e.repetition_penalty&&1!==e.repetition_penalty&&r.push(new d.RepetitionPenaltyLogitsProcessor(e.repetition_penalty)),null!==e.no_repeat_ngram_size&&e.no_repeat_ngram_size>0&&r.push(new d.NoRepeatNGramLogitsProcessor(e.no_repeat_ngram_size)),null!==e.bad_words_ids&&r.push(new d.NoBadWordsLogitsProcessor(e.bad_words_ids,e.eos_token_id)),null!==e.min_length&&null!==e.eos_token_id&&e.min_length>0&&r.push(new d.MinLengthLogitsProcessor(e.min_length,e.eos_token_id)),null!==e.min_new_tokens&&null!==e.eos_token_id&&e.min_new_tokens>0&&r.push(new d.MinNewTokensLengthLogitsProcessor(t,e.min_new_tokens,e.eos_token_id)),null!==e.forced_bos_token_id&&r.push(new d.ForcedBOSTokenLogitsProcessor(e.forced_bos_token_id)),null!==e.forced_eos_token_id&&r.push(new d.ForcedEOSTokenLogitsProcessor(e.max_length,e.forced_eos_token_id)),null!==e.begin_suppress_tokens){const s=t>1||null===e.forced_bos_token_id?t:t+1;r.push(new d.SuppressTokensAtBeginLogitsProcessor(e.begin_suppress_tokens,s))}return null!==e.guidance_scale&&e.guidance_scale>1&&r.push(new d.ClassifierFreeGuidanceLogitsProcessor(e.guidance_scale)),null!==s&&r.extend(s),r}_prepare_generation_config(e,t,s=u.GenerationConfig){const r={...this.config};for(const e of["decoder","generator","text_config"])e in r&&Object.assign(r,r[e]);const o=new s(r);return Object.assign(o,this.generation_config??{}),e&&Object.assign(o,e),t&&Object.assign(o,(0,a.pick)(t,Object.getOwnPropertyNames(o))),o}_get_stopping_criteria(e,t=null){const s=new h.StoppingCriteriaList;return null!==e.max_length&&s.push(new h.MaxLengthCriteria(e.max_length,this.config.max_position_embeddings??null)),null!==e.eos_token_id&&s.push(new h.EosTokenCriteria(e.eos_token_id)),t&&s.extend(t),s}_validate_model_class(){if(!this.can_generate){const e=[bc,Tc,xc,hc],t=z.get(this.constructor),s=new Set,r=this.config.model_type;for(const t of e){const e=t.get(r);e&&s.add(e[0])}let o=`The current model class (${t}) is not compatible with \`.generate()\`, as it doesn't have a language model head.`;throw s.size>0&&(o+=` Please use the following class instead: ${[...s].join(", ")}`),Error(o)}}prepare_inputs_for_generation(...e){return this._prepare_inputs_for_generation(this,...e)}_update_model_kwargs_for_generation({generated_input_ids:e,outputs:t,model_inputs:s,is_encoder_decoder:r}){return s.past_key_values=this.getPastKeyValues(t,s.past_key_values),s.input_ids=new _.Tensor("int64",e.flat(),[e.length,1]),r||(s.attention_mask=(0,_.cat)([s.attention_mask,(0,_.ones)([s.attention_mask.dims[0],1])],1)),s.position_ids=null,s}_prepare_model_inputs({inputs:e,bos_token_id:t,model_kwargs:s}){const r=(0,a.pick)(s,this.forward_params),o=this.main_input_name;if(o in r){if(e)throw new Error("`inputs`: {inputs}` were passed alongside {input_name} which is not allowed. Make sure to either pass {inputs} or {input_name}=...")}else r[o]=e;return{inputs_tensor:r[o],model_inputs:r,model_input_name:o}}async _prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:e,model_inputs:t,model_input_name:s,generation_config:r}){if(this.sessions.model.inputNames.includes("inputs_embeds")&&!t.inputs_embeds&&"_prepare_inputs_embeds"in this){const{input_ids:e,pixel_values:s,attention_mask:r,...o}=t,n=await this._prepare_inputs_embeds(t);t={...o,...(0,a.pick)(n,["inputs_embeds","attention_mask"])}}let{last_hidden_state:o}=await G(this,t);if(null!==r.guidance_scale&&r.guidance_scale>1)o=(0,_.cat)([o,(0,_.full_like)(o,0)],0),"attention_mask"in t&&(t.attention_mask=(0,_.cat)([t.attention_mask,(0,_.zeros_like)(t.attention_mask)],0));else if(t.decoder_input_ids){const e=N(t.decoder_input_ids).dims[0];if(e!==o.dims[0]){if(1!==o.dims[0])throw new Error(`The encoder outputs have a different batch size (${o.dims[0]}) than the decoder inputs (${e}).`);o=(0,_.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:s,decoder_start_token_id:r,bos_token_id:o,generation_config:n}){let{decoder_input_ids:i,...a}=s;if(!(i instanceof _.Tensor)){if(i)Array.isArray(i[0])||(i=Array.from({length:e},(()=>i)));else if(r??=o,"musicgen"===this.config.model_type)i=Array.from({length:e*this.config.decoder.num_codebooks},(()=>[r]));else if(Array.isArray(r)){if(r.length!==e)throw new Error(`\`decoder_start_token_id\` expcted to have length ${e} but got ${r.length}`);i=r}else i=Array.from({length:e},(()=>[r]));i=N(i)}return s.decoder_attention_mask=(0,_.ones_like)(i),{input_ids:i,model_inputs:a}}async generate({inputs:e=null,generation_config:t=null,logits_processor:s=null,stopping_criteria:r=null,streamer:o=null,...n}){this._validate_model_class(),t=this._prepare_generation_config(t,n);let{inputs_tensor:i,model_inputs:a,model_input_name:l}=this._prepare_model_inputs({inputs:e,model_kwargs:n});const c=this.config.is_encoder_decoder;let d;c&&("encoder_outputs"in a||(a=await this._prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:i,model_inputs:a,model_input_name:l,generation_config:t}))),c?({input_ids:d,model_inputs:a}=this._prepare_decoder_input_ids_for_generation({batch_size:a[l].dims.at(0),model_input_name:l,model_kwargs:a,decoder_start_token_id:t.decoder_start_token_id,bos_token_id:t.bos_token_id,generation_config:t})):d=a[l];let u=d.dims.at(-1);null!==t.max_new_tokens&&(t.max_length=u+t.max_new_tokens);const p=this._get_logits_processor(t,u,s),m=this._get_stopping_criteria(t,r),h=a[l].dims.at(0),f=g.LogitsSampler.getSampler(t),w=new Array(h).fill(0),M=d.tolist();let x;o&&o.put(M);let b={};for(;;){if(a=this.prepare_inputs_for_generation(M,a,t),x=await this.forward(a),t.output_attentions&&t.return_dict_in_generate){const e=this.getAttentions(x);for(const t in e)t in b||(b[t]=[]),b[t].push(e[t])}const e=p(M,x.logits.slice(null,-1,null)),s=[];for(let t=0;t<e.dims.at(0);++t){const r=e[t],o=await f(r);for(const[e,r]of o){const o=BigInt(e);w[t]+=r,M[t].push(o),s.push([o]);break}}o&&o.put(s);if(m(M).every((e=>e)))break;a=this._update_model_kwargs_for_generation({generated_input_ids:s,outputs:x,model_inputs:a,is_encoder_decoder:c})}o&&o.end();const k=this.getPastKeyValues(x,a.past_key_values,!0),y=new _.Tensor("int64",M.flat(),[M.length,M[0].length]);if(t.return_dict_in_generate)return{sequences:y,past_key_values:k,...b};for(const e of Object.values(x))"gpu-buffer"===e.location&&e.dispose();return y}getPastKeyValues(e,t,s=!1){const r=Object.create(null);for(const o in e)if(o.startsWith("present")){const n=o.replace("present","past_key_values"),i=o.includes("encoder");if(r[n]=i&&t?t[n]:e[o],t&&(!i||s)){const e=t[n];"gpu-buffer"===e.location&&e.dispose()}}return r}getAttentions(e){const t={};for(const s of["cross_attentions","encoder_attentions","decoder_attentions"])for(const r in e)r.startsWith(s)&&(s in t||(t[s]=[]),t[s].push(e[r]));return t}addPastKeyValues(e,t){if(t)Object.assign(e,t);else{const t=this.sessions.decoder_model_merged??this.sessions.model,s=t?.config?.kv_cache_dtype??"float32",o="float16"===s?new _.DataTypeMap.float16:[],n=(e[this.main_input_name]??e.attention_mask)?.dims?.[0]??1,i=(0,r.getKeyValueShapes)(this.config,{batch_size:n});for(const t in i)e[t]=new _.Tensor(s,o,i[t])}}async encode_image({pixel_values:e}){const t=(await O(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 O(this.sessions.embed_tokens,{input_ids:e})).inputs_embeds}async encode_audio({audio_values:e}){return(await O(this.sessions.audio_encoder,{audio_values:e})).audio_features}}class se{}class re extends se{constructor({last_hidden_state:e,hidden_states:t=null,attentions:s=null}){super(),this.last_hidden_state=e,this.hidden_states=t,this.attentions=s}}class oe extends te{}class ne extends oe{}class ie extends oe{async _call(e){return new Ed(await super._call(e))}}class ae extends oe{async _call(e){return new Cd(await super._call(e))}}class le extends oe{async _call(e){return new Ad(await super._call(e))}}class ce extends oe{async _call(e){return new Ld(await super._call(e))}}class de extends te{}class ue extends de{}class _e extends de{async _call(e){return new Ed(await super._call(e))}}class pe extends de{async _call(e){return new Cd(await super._call(e))}}class me extends de{async _call(e){return new Ad(await super._call(e))}}class he extends te{}class ge extends he{}class fe extends te{}class we extends fe{}class Me extends fe{async _call(e){return new Ed(await super._call(e))}}class xe extends fe{async _call(e){return new Cd(await super._call(e))}}class be extends fe{async _call(e){return new Ad(await super._call(e))}}class ke extends fe{async _call(e){return new Ld(await super._call(e))}}class ye extends te{}class ve extends ye{}class Te extends ye{async _call(e){return new Ed(await super._call(e))}}class Pe extends ye{async _call(e){return new Cd(await super._call(e))}}class Fe extends ye{async _call(e){return new Ad(await super._call(e))}}class Ce extends ye{async _call(e){return new Ld(await super._call(e))}}class Se extends te{}class Ae extends Se{}class Ee extends Se{async _call(e){return new Ed(await super._call(e))}}class Le extends Se{async _call(e){return new Cd(await super._call(e))}}class Ie extends Se{async _call(e){return new Ad(await super._call(e))}}class ze extends Se{async _call(e){return new Ld(await super._call(e))}}class je extends te{}class De extends je{}class Oe extends je{async _call(e){return new Ed(await super._call(e))}}class Ve extends je{async _call(e){return new Cd(await super._call(e))}}class Ne extends je{async _call(e){return new Ad(await super._call(e))}}class Be extends je{async _call(e){return new Ld(await super._call(e))}}class Re extends te{}class Ge extends Re{}class qe extends Re{async _call(e){return new Ed(await super._call(e))}}class $e extends Re{async _call(e){return new Cd(await super._call(e))}}class We extends Re{async _call(e){return new Ad(await super._call(e))}}class Ue extends Re{async _call(e){return new Ld(await super._call(e))}}class Xe extends te{}class Qe extends Xe{}class He extends Xe{async _call(e){return new Ed(await super._call(e))}}class Je extends Xe{async _call(e){return new Cd(await super._call(e))}}class Ye extends Xe{async _call(e){return new Ad(await super._call(e))}}class Ke extends Xe{async _call(e){return new Ld(await super._call(e))}}class Ze extends te{}class et extends Ze{}class tt extends Ze{async _call(e){return new Cd(await super._call(e))}}class st extends Ze{async _call(e){return new Ad(await super._call(e))}}class rt extends Ze{async _call(e){return new Ld(await super._call(e))}}class ot extends Ze{async _call(e){return new Ed(await super._call(e))}}class nt extends te{}class it extends nt{}class at extends nt{async _call(e){return new Ed(await super._call(e))}}class lt extends nt{async _call(e){return new Cd(await super._call(e))}}class ct extends nt{async _call(e){return new Ad(await super._call(e))}}class dt extends te{}class ut extends dt{}class _t extends dt{async _call(e){return new Ed(await super._call(e))}}class pt extends dt{async _call(e){return new Cd(await super._call(e))}}class mt extends dt{async _call(e){return new Ld(await super._call(e))}}class ht extends te{}class gt extends ht{}class ft extends ht{async _call(e){return new Ed(await super._call(e))}}class wt extends ht{async _call(e){return new Cd(await super._call(e))}}class Mt extends ht{async _call(e){return new Ad(await super._call(e))}}class xt extends ht{async _call(e){return new Ld(await super._call(e))}}class bt extends te{}class kt extends bt{}class yt extends bt{async _call(e){return new Ed(await super._call(e))}}class vt extends bt{async _call(e){return new Cd(await super._call(e))}}class Tt extends bt{async _call(e){return new Ld(await super._call(e))}}class Pt extends te{}class Ft extends Pt{}class Ct extends Pt{async _call(e){return new Cd(await super._call(e))}}class St extends Pt{async _call(e){return new Ld(await super._call(e))}}class At extends Pt{async _call(e){return new Ed(await super._call(e))}}class Et extends te{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"]}class Lt extends Et{}class It extends Et{}class zt extends te{}class jt extends zt{}class Dt extends zt{}class Ot extends te{}class Vt extends Ot{}class Nt extends Ot{}class Bt extends te{}class Rt extends Bt{}class Gt extends Bt{}class qt extends Bt{async _call(e){return new Cd(await super._call(e))}}class $t extends te{}class Wt extends $t{}class Ut extends $t{}class Xt extends $t{async _call(e){return new Cd(await super._call(e))}}class Qt extends $t{}class Ht extends te{}class Jt extends Ht{}class Yt extends Ht{}class Kt extends te{}class Zt extends Kt{}class es extends Kt{}class ts extends te{}class ss extends ts{}class rs extends ts{async _call(e){return new Ed(await super._call(e))}}class os extends ts{async _call(e){return new Cd(await super._call(e))}}class ns extends ts{async _call(e){return new Ad(await super._call(e))}}class is extends ts{async _call(e){return new Ld(await super._call(e))}}class as extends te{}class ls extends as{}class cs extends as{async _call(e){return new Ed(await super._call(e))}}class ds extends as{async _call(e){return new Cd(await super._call(e))}}class us extends as{async _call(e){return new Ad(await super._call(e))}}class _s extends as{async _call(e){return new Ld(await super._call(e))}}class ps extends te{}class ms extends ps{}class hs extends ps{async _call(e){return new Ed(await super._call(e))}}class gs extends ps{async _call(e){return new Cd(await super._call(e))}}class fs extends ps{async _call(e){return new Ad(await super._call(e))}}class ws extends ps{async _call(e){return new Ld(await super._call(e))}}class Ms extends te{}class xs extends Ms{}class bs extends Ms{}class ks extends te{requires_attention_mask=!1;main_input_name="input_features";forward_params=["input_features","attention_mask","decoder_input_ids","decoder_attention_mask","past_key_values"]}class ys extends ks{}class vs extends ks{_prepare_generation_config(e,t){return super._prepare_generation_config(e,t,w.WhisperGenerationConfig)}_retrieve_init_tokens(e){const t=[e.decoder_start_token_id];let s=e.language;const r=e.task;if(e.is_multilingual){s||(console.warn("No language specified - defaulting to English (en)."),s="en");const o=`<|${(0,M.whisper_language_to_code)(s)}|>`;t.push(e.lang_to_id[o]),t.push(e.task_to_id[r??"transcribe"])}else if(s||r)throw new Error("Cannot specify `task` or `language` for an English-only model. If the model is intended to be multilingual, pass `is_multilingual=true` to generate, or update the generation config.");return!e.return_timestamps&&e.no_timestamps_token_id&&t.at(-1)!==e.no_timestamps_token_id?t.push(e.no_timestamps_token_id):e.return_timestamps&&t.at(-1)===e.no_timestamps_token_id&&(console.warn("<|notimestamps|> prompt token is removed from generation_config since `return_timestamps` is set to `true`."),t.pop()),t.filter((e=>null!=e))}async generate({inputs:e=null,generation_config:t=null,logits_processor:s=null,stopping_criteria:r=null,...o}){t=this._prepare_generation_config(t,o);const n=o.decoder_input_ids??this._retrieve_init_tokens(t);if(t.return_timestamps&&(s??=new d.LogitsProcessorList,s.push(new d.WhisperTimeStampLogitsProcessor(t,n))),t.begin_suppress_tokens&&(s??=new d.LogitsProcessorList,s.push(new d.SuppressTokensAtBeginLogitsProcessor(t.begin_suppress_tokens,n.length))),t.return_token_timestamps){if(!t.alignment_heads)throw new Error("Model generation config has no `alignment_heads`, token-level timestamps not available. See https://gist.github.com/hollance/42e32852f24243b748ae6bc1f985b13a on how to add this property to the generation config.");"translate"===t.task&&console.warn("Token-level timestamps may not be reliable for task 'translate'."),t.output_attentions=!0,t.return_dict_in_generate=!0}const i=await super.generate({inputs:e,generation_config:t,logits_processor:s,decoder_input_ids:n,...o});return t.return_token_timestamps&&(i.token_timestamps=this._extract_token_timestamps(i,t.alignment_heads,t.num_frames)),i}_extract_token_timestamps(e,t,s=null,r=.02){if(!e.cross_attentions)throw new Error("Model outputs must contain cross attentions to extract timestamps. This is most likely because the model was not exported with `output_attentions=True`.");null==s&&console.warn("`num_frames` has not been set, meaning the entire audio will be analyzed. This may lead to inaccurate token-level timestamps for short audios (< 30 seconds).");let 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 n=e.cross_attentions,i=Array.from({length:this.config.decoder_layers},((e,t)=>(0,_.cat)(n.map((e=>e[t])),2))),l=(0,_.stack)(t.map((([e,t])=>{if(e>=i.length)throw new Error(`Layer index ${e} is out of bounds for cross attentions (length ${i.length}).`);return s?i[e].slice(null,t,null,[0,s]):i[e].slice(null,t)}))).transpose(1,0,2,3),[c,d]=(0,_.std_mean)(l,-2,0,!0),u=l.clone();for(let e=0;e<u.dims[0];++e){const t=u[e];for(let s=0;s<t.dims[0];++s){const r=t[s],n=c[e][s][0].data,i=d[e][s][0].data;for(let e=0;e<r.dims[0];++e){let t=r[e].data;for(let e=0;e<t.length;++e)t[e]=(t[e]-i[e])/n[e];t.set((0,m.medianFilter)(t,o))}}}const p=[(0,_.mean)(u,1)],h=e.sequences.dims,g=new _.Tensor("float32",new Float32Array(h[0]*h[1]),h);for(let e=0;e<h[0];++e){const t=p[e].neg().squeeze_(0),[s,o]=(0,m.dynamic_time_warping)(t.tolist()),n=Array.from({length:s.length-1},((e,t)=>s[t+1]-s[t])),i=(0,a.mergeArrays)([1],n).map((e=>!!e)),l=[];for(let e=0;e<i.length;++e)i[e]&&l.push(o[e]*r);g[e].data.set(l,1)}return g}}class Ts extends vs{}class Ps extends te{requires_attention_mask=!1;main_input_name="input_values";forward_params=["input_values","decoder_input_ids","past_key_values"]}class Fs extends Ps{}class Cs extends Ps{}class Ss extends te{main_input_name="pixel_values";forward_params=["pixel_values","decoder_input_ids","encoder_hidden_states","past_key_values"]}class As extends te{forward_params=["input_ids","attention_mask","pixel_values","position_ids","past_key_values"]}class Es extends As{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:s,attention_mask:r}){const o=this.config.image_token_index,n=s.tolist().map((e=>e.findIndex((e=>e==o)))),i=n.every((e=>-1===e)),a=n.every((e=>-1!==e));if(!i&&!a)throw new Error("Every input should contain either 0 or 1 image token.");if(i)return{inputs_embeds:e,attention_mask:r};const l=[],c=[];for(let s=0;s<n.length;++s){const o=n[s],i=e[s],a=t[s],d=r[s];l.push((0,_.cat)([i.slice([0,o]),a,i.slice([o+1,i.dims[0]])],0)),c.push((0,_.cat)([d.slice([0,o]),(0,_.ones)([a.dims[0]]),d.slice([o+1,d.dims[0]])],0))}return{inputs_embeds:(0,_.stack)(l,0),attention_mask:(0,_.stack)(c,0)}}}class Ls extends Es{}class Is extends Es{}class zs extends te{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 js extends zs{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:s,attention_mask:r}){return{inputs_embeds:(0,_.cat)([t,e],1),attention_mask:(0,_.cat)([(0,_.ones)(t.dims.slice(0,2)),r],1)}}async _prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:s,attention_mask:r}){if(!e&&!t)throw new Error("Either `input_ids` or `pixel_values` should be provided.");let o,n;return e&&(o=await this.encode_text({input_ids:e})),t&&(n=await this.encode_image({pixel_values:t})),o&&n?({inputs_embeds:s,attention_mask:r}=this._merge_input_ids_with_image_features({inputs_embeds:o,image_features:n,input_ids:e,attention_mask:r})):s=o||n,{inputs_embeds:s,attention_mask:r}}async forward({input_ids:e,pixel_values:t,attention_mask:s,decoder_input_ids:r,decoder_attention_mask:o,encoder_outputs:n,past_key_values:i,inputs_embeds:a,decoder_inputs_embeds:l}){if(a||({inputs_embeds:a,attention_mask:s}=await this._prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:a,attention_mask:s})),!n){let{last_hidden_state:e}=await G(this,{inputs_embeds:a,attention_mask:s});n=e}if(!l){if(!r)throw new Error("Either `decoder_input_ids` or `decoder_inputs_embeds` should be provided.");l=await this.encode_text({input_ids:r})}const c={inputs_embeds:l,attention_mask:o,encoder_attention_mask:s,encoder_hidden_states:n,past_key_values:i};return await $(this,c,!0)}}class Ds extends te{forward_params=["input_ids","attention_mask","pixel_values","position_ids","past_key_values"]}class Os extends Ds{_merge_input_ids_with_image_features(e){const t=e.image_features.dims.at(-1),s=e.image_features.view(-1,t);return U({image_token_id:this.config.image_token_index,...e,image_features:s})}}class Vs extends te{forward_params=["input_ids","attention_mask","pixel_values","pixel_attention_mask","position_ids","past_key_values"]}class Ns extends Vs{async encode_image({pixel_values:e,pixel_attention_mask:t}){return(await O(this.sessions.vision_encoder,{pixel_values:e,pixel_attention_mask:t})).image_features}_merge_input_ids_with_image_features(e){const t=e.image_features.dims.at(-1),s=e.image_features.view(-1,t);return U({image_token_id:this.config.image_token_id,...e,image_features:s})}}class Bs extends Ns{}class Rs extends te{forward_params=["input_ids","inputs_embeds","attention_mask","position_ids","pixel_values","image_sizes","past_key_values"]}class Gs extends Rs{async forward({input_ids:e=null,attention_mask:t=null,pixel_values:s=null,image_sizes:r=null,position_ids:o=null,inputs_embeds:n=null,past_key_values:i=null,generation_config:a=null,logits_processor:l=null,...c}){if(!n){let t;if(s&&1!==e.dims[1]){if(!r)throw new Error("`image_sizes` must be provided when `pixel_values` is provided.");({image_features:t}=await O(this.sessions.vision_encoder,{pixel_values:s,image_sizes:r}))}else{const e=this.config.normalized_config.hidden_size;t=new _.Tensor("float32",[],[0,e])}({inputs_embeds:n}=await O(this.sessions.prepare_inputs_embeds,{input_ids:e,image_features:t}))}return await $(this,{inputs_embeds:n,past_key_values:i,attention_mask:t,position_ids:o,generation_config:a,logits_processor:l},!1)}}class qs extends te{}class $s extends qs{}class Ws extends qs{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"text_model"})}}class Us extends qs{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"text_model"})}}class Xs extends qs{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"vision_model"})}}class Qs extends qs{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"vision_model"})}}class Hs extends te{}class Js extends Hs{}class Ys extends Hs{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"text_model"})}}class Ks extends qs{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"vision_model"})}}class Zs extends te{}class er extends Zs{}class tr extends te{}class sr extends tr{async forward(e){const t=!e.input_ids,s=!e.pixel_values;if(t&&s)throw new Error("Either `input_ids` or `pixel_values` should be provided.");if(t&&(e.input_ids=(0,_.ones)([e.pixel_values.dims[0],1])),s){const{image_size:t}=this.config.vision_config;e.pixel_values=(0,_.full)([0,3,t,t],0)}const{text_embeddings:r,image_embeddings:o,l2norm_text_embeddings:n,l2norm_image_embeddings:i}=await super.forward(e),a={};return t||(a.text_embeddings=r,a.l2norm_text_embeddings=n),s||(a.image_embeddings=o,a.l2norm_image_embeddings=i),a}}class rr extends tr{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"text_model"})}}class or extends tr{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"vision_model"})}}class nr extends te{}class ir extends nr{}class ar extends nr{}class lr extends te{}class cr extends lr{}class dr extends lr{}class ur extends te{}class _r extends ur{}class pr extends ur{}class mr extends te{}class hr extends mr{}class gr extends mr{}class fr extends te{}class wr extends fr{}class Mr extends fr{}class xr extends te{}class br extends xr{}class kr extends xr{}class yr extends te{}class vr extends yr{}class Tr extends yr{}class Pr extends te{}class Fr extends Pr{}class Cr extends Pr{}class Sr extends te{}class Ar extends Sr{}class Er extends Sr{}class Lr extends te{}class Ir extends Lr{}class zr extends Lr{}class jr extends te{}class Dr extends jr{}class Or extends jr{}class Vr extends te{}class Nr extends Vr{}class Br extends Vr{}class Rr extends te{}class Gr extends Rr{}class qr extends Rr{}class $r extends te{}class Wr extends $r{}class Ur extends $r{}class Xr extends te{}class Qr extends Xr{}class Hr extends Xr{}class Jr extends te{}class Yr extends Jr{}class Kr extends Jr{}class Zr extends te{}class eo extends Zr{}class to extends Zr{}class so extends te{}class ro extends so{}class oo extends so{}class no extends te{}class io extends no{}class ao extends no{}class lo extends te{}class co extends lo{}class uo extends lo{}class _o extends te{}class po extends _o{}class mo extends _o{}class ho extends te{}class go extends ho{}class fo extends ho{}class wo extends te{forward_params=["input_ids","attention_mask","position_ids","past_key_values","pixel_values","image_grid_thw"]}class Mo extends wo{get_rope_index(e,t,s,r){const{vision_config:o,image_token_id:n,video_token_id:i,vision_start_token_id:a}=this.config,l=o.spatial_merge_size??2,c=[];if(t||s){let o=e.tolist();r||(r=(0,_.ones_like)(e));const d=r.tolist(),u=Array.from({length:3},(t=>Array.from({length:e.dims[0]},(t=>Array.from({length:e.dims[1]},(e=>1)))))),p=t?t.tolist():[],h=s?s.tolist():[];let g=0,f=0;for(let e=0;e<o.length;++e){const t=o[e].filter(((t,s)=>1==d[e][s])),s=t.reduce(((e,t,s)=>(t==a&&e.push(s),e)),[]).map((e=>t[e+1])),r=s.filter((e=>e==n)).length,_=s.filter((e=>e==i)).length;let w=[],M=0,x=r,b=_;for(let e=0;e<s.length;++e){const e=t.findIndex(((e,t)=>t>M&&e==n)),s=t.findIndex(((e,t)=>t>M&&e==i)),r=x>0&&-1!==e?e:t.length+1,o=b>0&&-1!==s?s:t.length+1;let a,c,d,u;r<o?([c,d,u]=p[g],++g,--x,a=r):([c,d,u]=h[f],++f,--b,a=o);const[_,k,y]=[Number(c),Math.floor(Number(d)/l),Math.floor(Number(u)/l)],v=a-M,T=w.length>0?(0,m.max)(w.at(-1))[0]+1:0;w.push(Array.from({length:3*v},((e,t)=>T+t%v)));const P=v+T,F=_*k*y,C=Array.from({length:F},((e,t)=>P+Math.floor(t/(k*y)))),S=Array.from({length:F},((e,t)=>P+Math.floor(t/y)%k)),A=Array.from({length:F},((e,t)=>P+t%y));w.push([C,S,A].flat()),M=a+F}if(M<t.length){const e=w.length>0?(0,m.max)(w.at(-1))[0]+1:0,s=t.length-M;w.push(Array.from({length:3*s},((t,r)=>e+r%s)))}const k=w.reduce(((e,t)=>e+t.length),0),y=new Array(k);let v=0;for(let e=0;e<3;++e)for(let t=0;t<w.length;++t){const s=w[t],r=s.length/3;for(let t=e*r;t<(e+1)*r;++t)y[v++]=s[t]}let T=0;const P=d[e];for(let t=0;t<P.length;++t)if(1==P[t]){for(let s=0;s<3;++s)u[s][e][t]=y[s*k/3+T];++T}const F=(0,m.max)(y)[0];c.push(F+1-o[e].length)}return[new _.Tensor("int64",u.flat(1/0),[3,e.dims[0],e.dims[1]]),new _.Tensor("int64",c,[c.length,1])]}if(r){const{data:e,dims:t}=J(r),s=BigInt64Array.from({length:3*e.length},((t,s)=>e[s%e.length])),o=Array.from({length:t[0]},((s,r)=>(0,m.max)(e.subarray(t[1]*r,t[1]*(r+1)))[0]+1n+BigInt(t[1])));return[new _.Tensor("int64",s,[3,...t]),new _.Tensor("int64",o,[o.length,1])]}{const[t,s]=e.dims,r=BigInt64Array.from({length:3*t*s},((e,r)=>BigInt(Math.floor(r%s/t))));return[new _.Tensor("int64",r,[3,...e.dims]),(0,_.zeros)([t,1])]}}async encode_image({pixel_values:e,image_grid_thw:t}){return(await O(this.sessions.vision_encoder,{pixel_values:e,grid_thw:t})).image_features}_merge_input_ids_with_image_features(e){return U({image_token_id:this.config.image_token_id,...e})}prepare_inputs_for_generation(e,t,s){if(t.attention_mask&&!t.position_ids)if(t.past_key_values){t.pixel_values=null;const e=BigInt(Object.values(t.past_key_values)[0].dims.at(-2)),s=t.rope_deltas.map((t=>e+t));t.position_ids=(0,_.stack)([s,s,s],0)}else[t.position_ids,t.rope_deltas]=this.get_rope_index(t.input_ids,t.image_grid_thw,t.video_grid_thw,t.attention_mask);return t}}class xo extends te{}class bo extends xo{}class ko extends xo{}class yo extends te{}class vo extends yo{}class To extends yo{}class Po extends te{}class Fo extends Po{}class Co extends Po{}class So extends te{}class Ao extends So{}class Eo extends So{}class Lo extends te{}class Io extends Lo{}class zo extends Lo{}class jo extends te{}class Do extends jo{}class Oo extends jo{async _call(e){return new Cd(await super._call(e))}}class Vo extends te{}class No extends Vo{}class Bo extends Vo{async _call(e){return new Cd(await super._call(e))}}class Ro extends te{}class Go extends Ro{}class qo extends te{}class $o extends qo{}class Wo extends qo{async _call(e){return new Cd(await super._call(e))}}class Uo extends te{}class Xo extends Uo{}class Qo extends te{}class Ho extends Qo{}class Jo extends Qo{async _call(e){return new Cd(await super._call(e))}}class Yo extends te{}class Ko extends Yo{}class Zo extends te{}class en extends Zo{}class tn extends Zo{async _call(e){return new Cd(await super._call(e))}}class sn extends te{}class rn extends sn{async _call(e){return new jd(await super._call(e))}}class on extends te{}class nn extends on{}class an extends on{async _call(e){return new Cd(await super._call(e))}}class ln extends te{}class cn extends ln{}class dn extends ln{async _call(e){return new Cd(await super._call(e))}}class un extends te{}class _n extends un{}class pn extends un{}class mn extends te{}class hn extends mn{}class gn extends mn{}class fn extends te{}class wn extends fn{}class Mn extends fn{async _call(e){return new Cd(await super._call(e))}}class xn extends te{}class bn extends xn{}class kn extends xn{async _call(e){return new vn(await super._call(e))}}class yn extends xn{async _call(e){return new Tn(await super._call(e))}}class vn extends se{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class Tn extends se{constructor({logits:e,pred_boxes:t,pred_masks:s}){super(),this.logits=e,this.pred_boxes=t,this.pred_masks=s}}class Pn extends te{}class Fn extends Pn{}class Cn extends Pn{async _call(e){return new Sn(await super._call(e))}}class Sn extends se{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class An extends te{}class En extends An{}class Ln extends An{async _call(e){return new In(await super._call(e))}}class In extends Sn{}class zn extends te{}class jn extends zn{}class Dn extends zn{async _call(e){return new On(await super._call(e))}}class On extends Sn{}class Vn extends te{}class Nn extends Vn{}class Bn extends Vn{async _call(e){return new Rn(await super._call(e))}}class Rn extends vn{}class Gn extends te{}class qn extends Gn{}class $n extends Gn{async _call(e){return new Cd(await super._call(e))}}class Wn extends te{}class Un extends Wn{}class Xn extends Wn{async _call(e){return new Cd(await super._call(e))}}class Qn extends te{}class Hn extends Qn{}class Jn extends Qn{async _call(e){return new Cd(await super._call(e))}}class Yn extends te{}class Kn extends Yn{}class Zn extends Yn{async _call(e){return new Cd(await super._call(e))}}class ei extends Yn{}class ti extends te{}class si extends ti{}class ri extends ti{}class oi extends te{}class ni extends oi{}class ii extends oi{}class ai extends te{}class li extends ai{}class ci extends te{}class di extends ci{}class ui extends ci{}class _i extends ci{}class pi extends te{}class mi extends pi{}class hi extends te{}class gi extends hi{}class fi extends te{}class wi extends fi{}class Mi extends te{}class xi extends Mi{}class bi extends Mi{}class ki extends te{}class yi extends ki{}class vi extends ki{}class Ti extends te{}class Pi extends Ti{}class Fi extends te{}class Ci extends Fi{}class Si extends Fi{async _call(e){return new Cd(await super._call(e))}}class Ai extends te{}class Ei extends Ai{}class Li extends Ai{async _call(e){return new Cd(await super._call(e))}}class Ii extends te{}class zi extends Ii{}class ji extends Ii{async _call(e){return new Cd(await super._call(e))}}class Di extends te{}class Oi extends Di{}class Vi extends Di{async _call(e){return new Cd(await super._call(e))}}class Ni extends te{}class Bi extends Ni{}class Ri extends te{}class Gi extends Ri{}class qi extends Ri{async _call(e){return new $i(await super._call(e))}}class $i extends se{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class Wi extends te{}class Ui extends Wi{async get_image_embeddings({pixel_values:e}){return await G(this,{pixel_values:e})}async forward(e){if(e.image_embeddings&&e.image_positional_embeddings||(e={...e,...await this.get_image_embeddings(e)}),!e.input_labels&&e.input_points){const t=e.input_points.dims.slice(0,-1),s=t.reduce(((e,t)=>e*t),1);e.input_labels=new _.Tensor("int64",new BigInt64Array(s).fill(1n),t)}const t={image_embeddings:e.image_embeddings,image_positional_embeddings:e.image_positional_embeddings};return e.input_points&&(t.input_points=e.input_points),e.input_labels&&(t.input_labels=e.input_labels),e.input_boxes&&(t.input_boxes=e.input_boxes),await O(this.sessions.prompt_encoder_mask_decoder,t)}async _call(e){return new Xi(await super._call(e))}}class Xi extends se{constructor({iou_scores:e,pred_masks:t}){super(),this.iou_scores=e,this.pred_masks=t}}class Qi extends te{}class Hi extends Qi{}class Ji extends Qi{}class Yi extends te{}class Ki extends Yi{}class Zi extends Yi{}class ea extends te{}class ta extends ea{}class sa extends ea{async _call(e){return new Id(await super._call(e))}}class ra extends ea{async _call(e){return new Cd(await super._call(e))}}class oa extends ea{async _call(e){return new Ad(await super._call(e))}}class na extends te{}class ia extends na{}class aa extends na{async _call(e){return new Ad(await super._call(e))}}class la extends te{}class ca extends la{}class da extends te{}class ua extends da{}class _a extends da{async _call(e){return new Id(await super._call(e))}}class pa extends da{async _call(e){return new Cd(await super._call(e))}}class ma extends te{}class ha extends ma{}class ga extends ma{async _call(e){return new Id(await super._call(e))}}class fa extends ma{async _call(e){return new Cd(await super._call(e))}}class wa extends ma{async _call(e){return new Ad(await super._call(e))}}class Ma extends te{}class xa extends Ma{}class ba extends Ma{async _call(e){return new Id(await super._call(e))}}class ka extends Ma{async _call(e){return new Cd(await super._call(e))}}class ya extends te{}class va extends ea{}class Ta extends ea{async _call(e){return new Id(await super._call(e))}}class Pa extends ea{async _call(e){return new Cd(await super._call(e))}}class Fa extends te{}class Ca extends Fa{}class Sa extends Fa{async _call(e){return new Id(await super._call(e))}}class Aa extends Fa{async _call(e){return new Cd(await super._call(e))}}class Ea extends Fa{async _call(e){return new Sd(await super._call(e))}}class La extends Fa{async _call(e){return new Ad(await super._call(e))}}class Ia extends te{}class za extends Ia{}class ja extends te{}class Da extends ja{}class Oa extends ja{}class Va extends ja{async generate_speech(e,t,{threshold:s=.5,minlenratio:r=0,maxlenratio:o=20,vocoder:n=null}={}){const i={input_ids:e},{encoder_outputs:a,encoder_attention_mask:l}=await G(this,i),c=a.dims[1]/this.config.reduction_factor,d=Math.floor(c*o),u=Math.floor(c*r),p=this.config.num_mel_bins;let m=[],h=null,g=null,f=0;for(;;){++f;const e=B(!!g);let r;r=g?g.output_sequence_out:new _.Tensor("float32",new Float32Array(p),[1,1,p]);let o={use_cache_branch:e,output_sequence:r,encoder_attention_mask:l,speaker_embeddings:t,encoder_hidden_states:a};this.addPastKeyValues(o,h),g=await O(this.sessions.decoder_model_merged,o),h=this.getPastKeyValues(g,h);const{prob:n,spectrum:i}=g;if(m.push(i),f>=u&&(Array.from(n.data).filter((e=>e>=s)).length>0||f>=d))break}const w=(0,_.cat)(m),{waveform:M}=await O(n.sessions.model,{spectrogram:w});return{spectrogram:w,waveform:M}}}class Na extends te{main_input_name="spectrogram"}class Ba extends te{}class Ra extends Ba{}class Ga extends te{}class qa extends Ga{}class $a extends Ga{}class Wa extends te{}class Ua extends Wa{}class Xa extends Wa{}class Qa extends te{}class Ha extends Qa{}class Ja extends Qa{}class Ya extends te{}class Ka extends Ya{}class Za extends Ya{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"text_model"})}}class el extends Ya{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"audio_model"})}}class tl extends te{}class sl extends tl{async _call(e){return new Dd(await super._call(e))}}class rl extends te{}class ol extends rl{}class nl extends rl{}class il extends rl{}class al extends te{}class ll extends al{}class cl extends al{}class dl extends te{}class ul extends dl{}class _l extends dl{async _call(e){return new Cd(await super._call(e))}}class pl extends te{}class ml extends pl{}class hl extends pl{}class gl extends te{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"];_apply_and_filter_by_delay_pattern_mask(e){const[t,s]=e.dims,r=this.config.decoder.num_codebooks,o=s-r;let n=0;for(let t=0;t<e.size;++t){if(e.data[t]===this.config.decoder.pad_token_id)continue;const i=t%s-Math.floor(t/s)%r;i>0&&i<=o&&(e.data[n++]=e.data[t])}const i=Math.floor(t/r),a=n/(i*r);return new _.Tensor(e.type,e.data.slice(0,n),[i,r,a])}prepare_inputs_for_generation(e,t,s){let r=structuredClone(e);for(let e=0;e<r.length;++e)for(let t=0;t<r[e].length;++t)e%this.config.decoder.num_codebooks>=t&&(r[e][t]=BigInt(this.config.decoder.pad_token_id));null!==s.guidance_scale&&s.guidance_scale>1&&(r=r.concat(r));return super.prepare_inputs_for_generation(r,t,s)}async generate(e){const t=await super.generate(e),s=this._apply_and_filter_by_delay_pattern_mask(t).unsqueeze_(0),{audio_values:r}=await O(this.sessions.encodec_decode,{audio_codes:s});return r}}class fl extends te{}class wl extends fl{}class Ml extends fl{async _call(e){return new Cd(await super._call(e))}}class xl extends fl{}class bl extends te{}class kl extends bl{}class yl extends bl{async _call(e){return new Cd(await super._call(e))}}class vl extends bl{}class Tl extends te{}class Pl extends Tl{}class Fl extends Tl{async _call(e){return new Cd(await super._call(e))}}class Cl extends Tl{}class Sl extends te{}class Al extends Sl{}class El extends Sl{async _call(e){return new Cd(await super._call(e))}}class Ll extends Sl{}class Il extends te{}class zl extends Il{}class jl extends te{}class Dl extends jl{forward_params=["input_ids","pixel_values","images_seq_mask","images_emb_mask","attention_mask","position_ids","past_key_values"];constructor(...e){super(...e),this._generation_mode="text"}async forward(e){const t=this._generation_mode??"text";let s;if("text"!==t&&e.past_key_values){const t=this.sessions.gen_img_embeds,r=(0,a.pick)({image_ids:e.input_ids},t.inputNames);s=await O(t,r)}else{const t=this.sessions.prepare_inputs_embeds,r=(0,a.pick)(e,t.inputNames);s=await O(t,r)}const r={...e,...s},o=await $(this,r),n=this.sessions["text"===t?"lm_head":"gen_head"];if(!n)throw new Error(`Unable to find "${n}" generation head`);const i=await O(n,(0,a.pick)(o,n.inputNames));return{...s,...o,...i}}async generate(e){return this._generation_mode="text",super.generate(e)}async generate_images(e){this._generation_mode="image";const t=(e.inputs??e[this.main_input_name]).dims[1],s=(await super.generate(e)).slice(null,[t,null]),r=this.sessions.image_decode,{decoded_image:o}=await O(r,{generated_tokens:s}),n=o.add_(1).mul_(127.5).clamp_(0,255).to("uint8"),i=[];for(const e of n){const t=p.RawImage.fromTensor(e);i.push(t)}return i}}class Ol extends se{constructor({char_logits:e,bpe_logits:t,wp_logits:s}){super(),this.char_logits=e,this.bpe_logits=t,this.wp_logits=s}get logits(){return[this.char_logits,this.bpe_logits,this.wp_logits]}}class Vl extends te{}class Nl extends Vl{async _call(e){return new Ol(await super._call(e))}}class Bl extends te{}class Rl extends Bl{}class Gl extends Bl{}class ql extends te{}class $l extends ql{}class Wl extends ql{}class Ul extends te{forward_params=["input_ids","attention_mask","position_ids","audio_values","past_key_values"]}class Xl extends Ul{_merge_input_ids_with_audio_features(e){const t=e.audio_features.dims.at(-1),s=e.audio_features.view(-1,t);return function({audio_token_id:e,inputs_embeds:t,audio_features:s,input_ids:r,attention_mask:o}){return W({modality_token_id:e,inputs_embeds:t,modality_features:s,input_ids:r,attention_mask:o})}({audio_token_id:this.config.ignore_index,...e,audio_features:s})}}class Ql extends te{main_input_name="input_values";forward_params=["input_values"]}class Hl extends se{constructor({audio_codes:e}){super(),this.audio_codes=e}}class Jl extends se{constructor({audio_values:e}){super(),this.audio_values=e}}class Yl extends Ql{async encode(e){return new Hl(await O(this.sessions.encoder_model,e))}async decode(e){return new Jl(await O(this.sessions.decoder_model,e))}}class Kl extends Ql{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"encoder_model"})}}class Zl extends Ql{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"decoder_model"})}}class ec extends te{main_input_name="input_values";forward_params=["input_values"]}class tc extends se{constructor({audio_codes:e}){super(),this.audio_codes=e}}class sc extends se{constructor({audio_values:e}){super(),this.audio_values=e}}class rc extends ec{async encode(e){return new tc(await O(this.sessions.encoder_model,e))}async decode(e){return new sc(await O(this.sessions.decoder_model,e))}}class oc extends ec{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"encoder_model"})}}class nc extends ec{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"decoder_model"})}}class ic extends te{main_input_name="input_values";forward_params=["input_values"]}class ac extends ic{async encode(e){return await O(this.sessions.encoder_model,e)}async decode(e){return await O(this.sessions.decoder_model,e)}}class lc extends ic{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"encoder_model"})}}class cc extends ic{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"decoder_model"})}}class dc{static MODEL_CLASS_MAPPINGS=null;static BASE_IF_FAIL=!1;static async from_pretrained(e,{progress_callback:t=null,config:s=null,cache_dir:o=null,local_files_only:n=!1,revision:i="main",model_file_name:a=null,subfolder:l="onnx",device:c=null,dtype:d=null,use_external_data_format:u=null,session_options:_={}}={}){const p={progress_callback:t,config:s,cache_dir:o,local_files_only:n,revision:i,model_file_name:a,subfolder:l,device:c,dtype:d,use_external_data_format:u,session_options:_};if(p.config=await r.AutoConfig.from_pretrained(e,p),!this.MODEL_CLASS_MAPPINGS)throw new Error("`MODEL_CLASS_MAPPINGS` not implemented for this type of `AutoClass`: "+this.name);const m=p.config.model_type;for(const t of this.MODEL_CLASS_MAPPINGS){let s=t.get(m);if(!s){for(const e of t.values())if(e[0]===m){s=e;break}if(!s)continue}return await s[1].from_pretrained(e,p)}if(this.BASE_IF_FAIL)return Hc.has(m)||console.warn(`Unknown model class "${m}", attempting to construct from base class.`),await te.from_pretrained(e,p);throw Error(`Unsupported model type: ${m}`)}}const uc=new Map([["bert",["BertModel",ne]],["modernbert",["ModernBertModel",ue]],["nomic_bert",["NomicBertModel",ge]],["roformer",["RoFormerModel",we]],["electra",["ElectraModel",Ae]],["esm",["EsmModel",it]],["convbert",["ConvBertModel",ve]],["camembert",["CamembertModel",De]],["deberta",["DebertaModel",Ge]],["deberta-v2",["DebertaV2Model",Qe]],["mpnet",["MPNetModel",gt]],["albert",["AlbertModel",Ft]],["distilbert",["DistilBertModel",et]],["roberta",["RobertaModel",ss]],["xlm",["XLMModel",ls]],["xlm-roberta",["XLMRobertaModel",ms]],["clap",["ClapModel",Ka]],["clip",["CLIPModel",$s]],["clipseg",["CLIPSegModel",ir]],["chinese_clip",["ChineseCLIPModel",er]],["siglip",["SiglipModel",Js]],["jina_clip",["JinaCLIPModel",sr]],["mobilebert",["MobileBertModel",ut]],["squeezebert",["SqueezeBertModel",kt]],["wav2vec2",["Wav2Vec2Model",ta]],["wav2vec2-bert",["Wav2Vec2BertModel",xa]],["unispeech",["UniSpeechModel",ua]],["unispeech-sat",["UniSpeechSatModel",ha]],["hubert",["HubertModel",va]],["wavlm",["WavLMModel",Ca]],["audio-spectrogram-transformer",["ASTModel",xs]],["vits",["VitsModel",sl]],["pyannote",["PyAnnoteModel",ia]],["wespeaker-resnet",["WeSpeakerResNetModel",ca]],["detr",["DetrModel",bn]],["rt_detr",["RTDetrModel",Fn]],["rt_detr_v2",["RTDetrV2Model",En]],["rf_detr",["RFDetrModel",jn]],["table-transformer",["TableTransformerModel",Nn]],["vit",["ViTModel",Do]],["ijepa",["IJepaModel",No]],["pvt",["PvtModel",$o]],["vit_msn",["ViTMSNModel",Ho]],["vit_mae",["ViTMAEModel",Xo]],["groupvit",["GroupViTModel",Ko]],["fastvit",["FastViTModel",en]],["mobilevit",["MobileViTModel",nn]],["mobilevitv2",["MobileViTV2Model",cn]],["owlvit",["OwlViTModel",_n]],["owlv2",["Owlv2Model",hn]],["beit",["BeitModel",wn]],["deit",["DeiTModel",qn]],["hiera",["HieraModel",Un]],["convnext",["ConvNextModel",Ci]],["convnextv2",["ConvNextV2Model",Ei]],["dinov2",["Dinov2Model",zi]],["dinov2_with_registers",["Dinov2WithRegistersModel",Oi]],["resnet",["ResNetModel",Hn]],["swin",["SwinModel",Kn]],["swin2sr",["Swin2SRModel",si]],["donut-swin",["DonutSwinModel",Pi]],["yolos",["YolosModel",Gi]],["dpt",["DPTModel",ni]],["glpn",["GLPNModel",yi]],["hifigan",["SpeechT5HifiGan",Na]],["efficientnet",["EfficientNetModel",ul]],["decision_transformer",["DecisionTransformerModel",zl]],["patchtst",["PatchTSTForPrediction",Rl]],["patchtsmixer",["PatchTSMixerForPrediction",$l]],["mobilenet_v1",["MobileNetV1Model",wl]],["mobilenet_v2",["MobileNetV2Model",kl]],["mobilenet_v3",["MobileNetV3Model",Pl]],["mobilenet_v4",["MobileNetV4Model",Al]],["maskformer",["MaskFormerModel",xi]],["mgp-str",["MgpstrForSceneTextRecognition",Nl]],["style_text_to_speech_2",["StyleTextToSpeech2Model",za]]]),_c=new Map([["t5",["T5Model",Lt]],["longt5",["LongT5Model",jt]],["mt5",["MT5Model",Vt]],["bart",["BartModel",Rt]],["mbart",["MBartModel",Wt]],["marian",["MarianModel",Hi]],["whisper",["WhisperModel",ys]],["m2m_100",["M2M100Model",Ki]],["blenderbot",["BlenderbotModel",Jt]],["blenderbot-small",["BlenderbotSmallModel",Zt]]]),pc=new Map([["mimi",["MimiModel",Yl]],["dac",["DacModel",rc]],["snac",["SnacModel",ac]]]),mc=new Map([["bloom",["BloomModel",Fo]],["jais",["JAISModel",_r]],["gpt2",["GPT2Model",cr]],["gptj",["GPTJModel",br]],["gpt_bigcode",["GPTBigCodeModel",vr]],["gpt_neo",["GPTNeoModel",hr]],["gpt_neox",["GPTNeoXModel",wr]],["codegen",["CodeGenModel",Fr]],["llama",["LlamaModel",Ar]],["exaone",["ExaoneModel",Nr]],["olmo",["OlmoModel",Wr]],["olmo2",["Olmo2Model",Qr]],["mobilellm",["MobileLLMModel",Gr]],["granite",["GraniteModel",Yr]],["cohere",["CohereModel",eo]],["gemma",["GemmaModel",ro]],["gemma2",["Gemma2Model",io]],["gemma3_text",["Gemma3Model",co]],["helium",["HeliumModel",Ir]],["glm",["GlmModel",Dr]],["openelm",["OpenELMModel",po]],["qwen2",["Qwen2Model",go]],["phi",["PhiModel",bo]],["phi3",["Phi3Model",vo]],["mpt",["MptModel",Ao]],["opt",["OPTModel",Io]],["mistral",["MistralModel",qa]],["starcoder2",["Starcoder2Model",Ua]],["falcon",["FalconModel",Ha]],["stablelm",["StableLmModel",ll]]]),hc=new Map([["speecht5",["SpeechT5ForSpeechToText",Oa]],["whisper",["WhisperForConditionalGeneration",vs]],["lite-whisper",["LiteWhisperForConditionalGeneration",Ts]],["moonshine",["MoonshineForConditionalGeneration",Cs]]]),gc=new Map([["speecht5",["SpeechT5ForTextToSpeech",Va]]]),fc=new Map([["vits",["VitsModel",sl]],["musicgen",["MusicgenForConditionalGeneration",gl]]]),wc=new Map([["bert",["BertForSequenceClassification",ae]],["modernbert",["ModernBertForSequenceClassification",pe]],["roformer",["RoFormerForSequenceClassification",xe]],["electra",["ElectraForSequenceClassification",Le]],["esm",["EsmForSequenceClassification",lt]],["convbert",["ConvBertForSequenceClassification",Pe]],["camembert",["CamembertForSequenceClassification",Ve]],["deberta",["DebertaForSequenceClassification",$e]],["deberta-v2",["DebertaV2ForSequenceClassification",Je]],["mpnet",["MPNetForSequenceClassification",wt]],["albert",["AlbertForSequenceClassification",Ct]],["distilbert",["DistilBertForSequenceClassification",tt]],["roberta",["RobertaForSequenceClassification",os]],["xlm",["XLMForSequenceClassification",ds]],["xlm-roberta",["XLMRobertaForSequenceClassification",gs]],["bart",["BartForSequenceClassification",qt]],["mbart",["MBartForSequenceClassification",Xt]],["mobilebert",["MobileBertForSequenceClassification",pt]],["squeezebert",["SqueezeBertForSequenceClassification",vt]]]),Mc=new Map([["bert",["BertForTokenClassification",le]],["modernbert",["ModernBertForTokenClassification",me]],["roformer",["RoFormerForTokenClassification",be]],["electra",["ElectraForTokenClassification",Ie]],["esm",["EsmForTokenClassification",ct]],["convbert",["ConvBertForTokenClassification",Fe]],["camembert",["CamembertForTokenClassification",Ne]],["deberta",["DebertaForTokenClassification",We]],["deberta-v2",["DebertaV2ForTokenClassification",Ye]],["mpnet",["MPNetForTokenClassification",Mt]],["distilbert",["DistilBertForTokenClassification",st]],["roberta",["RobertaForTokenClassification",ns]],["xlm",["XLMForTokenClassification",us]],["xlm-roberta",["XLMRobertaForTokenClassification",fs]]]),xc=new Map([["t5",["T5ForConditionalGeneration",It]],["longt5",["LongT5ForConditionalGeneration",Dt]],["mt5",["MT5ForConditionalGeneration",Nt]],["bart",["BartForConditionalGeneration",Gt]],["mbart",["MBartForConditionalGeneration",Ut]],["marian",["MarianMTModel",Ji]],["m2m_100",["M2M100ForConditionalGeneration",Zi]],["blenderbot",["BlenderbotForConditionalGeneration",Yt]],["blenderbot-small",["BlenderbotSmallForConditionalGeneration",es]]]),bc=new Map([["bloom",["BloomForCausalLM",Co]],["gpt2",["GPT2LMHeadModel",dr]],["jais",["JAISLMHeadModel",pr]],["gptj",["GPTJForCausalLM",kr]],["gpt_bigcode",["GPTBigCodeForCausalLM",Tr]],["gpt_neo",["GPTNeoForCausalLM",gr]],["gpt_neox",["GPTNeoXForCausalLM",Mr]],["codegen",["CodeGenForCausalLM",Cr]],["llama",["LlamaForCausalLM",Er]],["exaone",["ExaoneForCausalLM",Br]],["olmo",["OlmoForCausalLM",Ur]],["olmo2",["Olmo2ForCausalLM",Hr]],["mobilellm",["MobileLLMForCausalLM",qr]],["granite",["GraniteForCausalLM",Kr]],["cohere",["CohereForCausalLM",to]],["gemma",["GemmaForCausalLM",oo]],["gemma2",["Gemma2ForCausalLM",ao]],["gemma3_text",["Gemma3ForCausalLM",uo]],["helium",["HeliumForCausalLM",zr]],["glm",["GlmForCausalLM",Or]],["openelm",["OpenELMForCausalLM",mo]],["qwen2",["Qwen2ForCausalLM",fo]],["phi",["PhiForCausalLM",ko]],["phi3",["Phi3ForCausalLM",To]],["mpt",["MptForCausalLM",Eo]],["opt",["OPTForCausalLM",zo]],["mbart",["MBartForCausalLM",Qt]],["mistral",["MistralForCausalLM",$a]],["starcoder2",["Starcoder2ForCausalLM",Xa]],["falcon",["FalconForCausalLM",Ja]],["trocr",["TrOCRForCausalLM",Ra]],["stablelm",["StableLmForCausalLM",cl]],["phi3_v",["Phi3VForCausalLM",Gs]]]),kc=new Map([["multi_modality",["MultiModalityCausalLM",Dl]]]),yc=new Map([["bert",["BertForMaskedLM",ie]],["modernbert",["ModernBertForMaskedLM",_e]],["roformer",["RoFormerForMaskedLM",Me]],["electra",["ElectraForMaskedLM",Ee]],["esm",["EsmForMaskedLM",at]],["convbert",["ConvBertForMaskedLM",Te]],["camembert",["CamembertForMaskedLM",Oe]],["deberta",["DebertaForMaskedLM",qe]],["deberta-v2",["DebertaV2ForMaskedLM",He]],["mpnet",["MPNetForMaskedLM",ft]],["albert",["AlbertForMaskedLM",At]],["distilbert",["DistilBertForMaskedLM",ot]],["roberta",["RobertaForMaskedLM",rs]],["xlm",["XLMWithLMHeadModel",cs]],["xlm-roberta",["XLMRobertaForMaskedLM",hs]],["mobilebert",["MobileBertForMaskedLM",_t]],["squeezebert",["SqueezeBertForMaskedLM",yt]]]),vc=new Map([["bert",["BertForQuestionAnswering",ce]],["roformer",["RoFormerForQuestionAnswering",ke]],["electra",["ElectraForQuestionAnswering",ze]],["convbert",["ConvBertForQuestionAnswering",Ce]],["camembert",["CamembertForQuestionAnswering",Be]],["deberta",["DebertaForQuestionAnswering",Ue]],["deberta-v2",["DebertaV2ForQuestionAnswering",Ke]],["mpnet",["MPNetForQuestionAnswering",xt]],["albert",["AlbertForQuestionAnswering",St]],["distilbert",["DistilBertForQuestionAnswering",rt]],["roberta",["RobertaForQuestionAnswering",is]],["xlm",["XLMForQuestionAnswering",_s]],["xlm-roberta",["XLMRobertaForQuestionAnswering",ws]],["mobilebert",["MobileBertForQuestionAnswering",mt]],["squeezebert",["SqueezeBertForQuestionAnswering",Tt]]]),Tc=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",Ss]],["idefics3",["Idefics3ForConditionalGeneration",Ns]],["smolvlm",["SmolVLMForConditionalGeneration",Bs]]]),Pc=new Map([["llava",["LlavaForConditionalGeneration",Es]],["llava_onevision",["LlavaOnevisionForConditionalGeneration",Ls]],["moondream1",["Moondream1ForConditionalGeneration",Is]],["florence2",["Florence2ForConditionalGeneration",js]],["qwen2-vl",["Qwen2VLForConditionalGeneration",Mo]],["idefics3",["Idefics3ForConditionalGeneration",Ns]],["smolvlm",["SmolVLMForConditionalGeneration",Bs]],["paligemma",["PaliGemmaForConditionalGeneration",Os]]]),Fc=new Map([["ultravox",["UltravoxModel",Xl]]]),Cc=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",Ss]]]),Sc=new Map([["vit",["ViTForImageClassification",Oo]],["ijepa",["IJepaForImageClassification",Bo]],["pvt",["PvtForImageClassification",Wo]],["vit_msn",["ViTMSNForImageClassification",Jo]],["fastvit",["FastViTForImageClassification",tn]],["mobilevit",["MobileViTForImageClassification",an]],["mobilevitv2",["MobileViTV2ForImageClassification",dn]],["beit",["BeitForImageClassification",Mn]],["deit",["DeiTForImageClassification",$n]],["hiera",["HieraForImageClassification",Xn]],["convnext",["ConvNextForImageClassification",Si]],["convnextv2",["ConvNextV2ForImageClassification",Li]],["dinov2",["Dinov2ForImageClassification",ji]],["dinov2_with_registers",["Dinov2WithRegistersForImageClassification",Vi]],["resnet",["ResNetForImageClassification",Jn]],["swin",["SwinForImageClassification",Zn]],["segformer",["SegformerForImageClassification",nl]],["efficientnet",["EfficientNetForImageClassification",_l]],["mobilenet_v1",["MobileNetV1ForImageClassification",Ml]],["mobilenet_v2",["MobileNetV2ForImageClassification",yl]],["mobilenet_v3",["MobileNetV3ForImageClassification",Fl]],["mobilenet_v4",["MobileNetV4ForImageClassification",El]]]),Ac=new Map([["detr",["DetrForObjectDetection",kn]],["rt_detr",["RTDetrForObjectDetection",Cn]],["rt_detr_v2",["RTDetrV2ForObjectDetection",Ln]],["rf_detr",["RFDetrForObjectDetection",Dn]],["table-transformer",["TableTransformerForObjectDetection",Bn]],["yolos",["YolosForObjectDetection",qi]]]),Ec=new Map([["owlvit",["OwlViTForObjectDetection",pn]],["owlv2",["Owlv2ForObjectDetection",gn]],["grounding-dino",["GroundingDinoForObjectDetection",Bi]]]),Lc=new Map([["detr",["DetrForSegmentation",yn]],["clipseg",["CLIPSegForImageSegmentation",ar]]]),Ic=new Map([["segformer",["SegformerForSemanticSegmentation",il]],["sapiens",["SapiensForSemanticSegmentation",di]],["swin",["SwinForSemanticSegmentation",ei]],["mobilenet_v1",["MobileNetV1ForSemanticSegmentation",xl]],["mobilenet_v2",["MobileNetV2ForSemanticSegmentation",vl]],["mobilenet_v3",["MobileNetV3ForSemanticSegmentation",Cl]],["mobilenet_v4",["MobileNetV4ForSemanticSegmentation",Ll]]]),zc=new Map([["detr",["DetrForSegmentation",yn]],["maskformer",["MaskFormerForInstanceSegmentation",bi]]]),jc=new Map([["sam",["SamModel",Ui]]]),Dc=new Map([["wav2vec2",["Wav2Vec2ForCTC",sa]],["wav2vec2-bert",["Wav2Vec2BertForCTC",ba]],["unispeech",["UniSpeechForCTC",_a]],["unispeech-sat",["UniSpeechSatForCTC",ga]],["wavlm",["WavLMForCTC",Sa]],["hubert",["HubertForCTC",Ta]]]),Oc=new Map([["wav2vec2",["Wav2Vec2ForSequenceClassification",ra]],["wav2vec2-bert",["Wav2Vec2BertForSequenceClassification",ka]],["unispeech",["UniSpeechForSequenceClassification",pa]],["unispeech-sat",["UniSpeechSatForSequenceClassification",fa]],["wavlm",["WavLMForSequenceClassification",Aa]],["hubert",["HubertForSequenceClassification",Pa]],["audio-spectrogram-transformer",["ASTForAudioClassification",bs]]]),Vc=new Map([["wavlm",["WavLMForXVector",Ea]]]),Nc=new Map([["unispeech-sat",["UniSpeechSatForAudioFrameClassification",wa]],["wavlm",["WavLMForAudioFrameClassification",La]],["wav2vec2",["Wav2Vec2ForAudioFrameClassification",oa]],["pyannote",["PyAnnoteForAudioFrameClassification",aa]]]),Bc=new Map([["vitmatte",["VitMatteForImageMatting",rn]]]),Rc=new Map([["patchtst",["PatchTSTForPrediction",Gl]],["patchtsmixer",["PatchTSMixerForPrediction",Wl]]]),Gc=new Map([["swin2sr",["Swin2SRForImageSuperResolution",ri]]]),qc=new Map([["dpt",["DPTForDepthEstimation",ii]],["depth_anything",["DepthAnythingForDepthEstimation",li]],["glpn",["GLPNForDepthEstimation",vi]],["sapiens",["SapiensForDepthEstimation",ui]],["depth_pro",["DepthProForDepthEstimation",mi]],["metric3d",["Metric3DForDepthEstimation",gi]],["metric3dv2",["Metric3Dv2ForDepthEstimation",wi]]]),$c=new Map([["sapiens",["SapiensForNormalEstimation",_i]]]),Wc=new Map([["vitpose",["VitPoseForPoseEstimation",Go]]]),Uc=new Map([["clip",["CLIPVisionModelWithProjection",Qs]],["siglip",["SiglipVisionModel",Ks]],["jina_clip",["JinaCLIPVisionModel",or]]]),Xc=[[uc,x],[_c,b],[mc,v],[pc,E],[wc,x],[Mc,x],[xc,k],[hc,k],[bc,v],[kc,C],[yc,x],[vc,x],[Tc,y],[Pc,P],[Fc,A],[Sc,x],[Lc,x],[zc,x],[Ic,x],[Bc,x],[Rc,x],[Gc,x],[qc,x],[$c,x],[Wc,x],[Ac,x],[Ec,x],[jc,T],[Dc,x],[Oc,x],[gc,k],[fc,x],[Vc,x],[Nc,x],[Uc,x]];for(const[e,t]of Xc)for(const[s,r]of e.values())L.set(s,t),z.set(r,s),I.set(s,r);const Qc=[["MusicgenForConditionalGeneration",gl,F],["Phi3VForCausalLM",Gs,S],["CLIPTextModelWithProjection",Us,x],["SiglipTextModel",Ys,x],["JinaCLIPTextModel",rr,x],["ClapTextModelWithProjection",Za,x],["ClapAudioModelWithProjection",el,x],["DacEncoderModel",oc,x],["DacDecoderModel",nc,x],["MimiEncoderModel",Kl,x],["MimiDecoderModel",Zl,x],["SnacEncoderModel",lc,x],["SnacDecoderModel",cc,x]];for(const[e,t,s]of Qc)L.set(e,s),z.set(t,e),I.set(e,t);const Hc=new Map([["modnet",Lc],["birefnet",Lc],["isnet",Lc],["ben",Lc]]);for(const[e,t]of Hc.entries())t.set(e,["PreTrainedModel",te]),L.set(e,x),z.set(te,e),I.set(e,te);class Jc extends dc{static MODEL_CLASS_MAPPINGS=Xc.map((e=>e[0]));static BASE_IF_FAIL=!0}class Yc extends dc{static MODEL_CLASS_MAPPINGS=[wc]}class Kc extends dc{static MODEL_CLASS_MAPPINGS=[Mc]}class Zc extends dc{static MODEL_CLASS_MAPPINGS=[xc]}class ed extends dc{static MODEL_CLASS_MAPPINGS=[hc]}class td extends dc{static MODEL_CLASS_MAPPINGS=[gc]}class sd extends dc{static MODEL_CLASS_MAPPINGS=[fc]}class rd extends dc{static MODEL_CLASS_MAPPINGS=[bc]}class od extends dc{static MODEL_CLASS_MAPPINGS=[yc]}class nd extends dc{static MODEL_CLASS_MAPPINGS=[vc]}class id extends dc{static MODEL_CLASS_MAPPINGS=[Tc]}class ad extends dc{static MODEL_CLASS_MAPPINGS=[Sc]}class ld extends dc{static MODEL_CLASS_MAPPINGS=[Lc]}class cd extends dc{static MODEL_CLASS_MAPPINGS=[Ic]}class dd extends dc{static MODEL_CLASS_MAPPINGS=[zc]}class ud extends dc{static MODEL_CLASS_MAPPINGS=[Ac]}class _d extends dc{static MODEL_CLASS_MAPPINGS=[Ec]}class pd extends dc{static MODEL_CLASS_MAPPINGS=[jc]}class md extends dc{static MODEL_CLASS_MAPPINGS=[Dc]}class hd extends dc{static MODEL_CLASS_MAPPINGS=[Oc]}class gd extends dc{static MODEL_CLASS_MAPPINGS=[Vc]}class fd extends dc{static MODEL_CLASS_MAPPINGS=[Nc]}class wd extends dc{static MODEL_CLASS_MAPPINGS=[Cc]}class Md extends dc{static MODEL_CLASS_MAPPINGS=[Bc]}class xd extends dc{static MODEL_CLASS_MAPPINGS=[Gc]}class bd extends dc{static MODEL_CLASS_MAPPINGS=[qc]}class kd extends dc{static MODEL_CLASS_MAPPINGS=[$c]}class yd extends dc{static MODEL_CLASS_MAPPINGS=[Wc]}class vd extends dc{static MODEL_CLASS_MAPPINGS=[Uc]}class Td extends dc{static MODEL_CLASS_MAPPINGS=[Pc]}class Pd extends dc{static MODEL_CLASS_MAPPINGS=[Fc]}class Fd extends se{constructor({logits:e,past_key_values:t,encoder_outputs:s,decoder_attentions:r=null,cross_attentions:o=null}){super(),this.logits=e,this.past_key_values=t,this.encoder_outputs=s,this.decoder_attentions=r,this.cross_attentions=o}}class Cd extends se{constructor({logits:e,...t}){super(),this.logits=e;const s=Object.values(t);s.length>0&&(this.attentions=s)}}class Sd extends se{constructor({logits:e,embeddings:t}){super(),this.logits=e,this.embeddings=t}}class Ad extends se{constructor({logits:e}){super(),this.logits=e}}class Ed extends se{constructor({logits:e}){super(),this.logits=e}}class Ld extends se{constructor({start_logits:e,end_logits:t}){super(),this.start_logits=e,this.end_logits=t}}class Id extends se{constructor({logits:e}){super(),this.logits=e}}class zd extends se{constructor({logits:e,past_key_values:t}){super(),this.logits=e,this.past_key_values=t}}class jd extends se{constructor({alphas:e}){super(),this.alphas=e}}class Dd extends se{constructor({waveform:e,spectrogram:t}){super(),this.waveform=e,this.spectrogram=t}}},"./src/models/audio_spectrogram_transformer/feature_extraction_audio_spectrogram_transformer.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ASTFeatureExtractor:()=>n});var r=s("./src/base/feature_extraction_utils.js"),o=(s("./src/utils/tensor.js"),s("./src/utils/audio.js"));class n extends r.FeatureExtractor{constructor(e){super(e);const t=this.config.sampling_rate,s=(0,o.mel_filter_bank)(257,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);this.mel_filters=s,this.window=(0,o.window_function)(400,"hann",{periodic:!1}),this.mean=this.config.mean,this.std=this.config.std}async _extract_fbank_features(e,t){return(0,o.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,max_num_frames:t,transpose:!0})}async _call(e){(0,r.validate_audio_inputs)(e,"ASTFeatureExtractor");const t=await this._extract_fbank_features(e,this.config.max_length);if(this.config.do_normalize){const e=2*this.std,s=t.data;for(let t=0;t<s.length;++t)s[t]=(s[t]-this.mean)/e}return{input_values:t.unsqueeze_(0)}}}},"./src/models/auto/feature_extraction_auto.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{AutoFeatureExtractor:()=>i});var r=s("./src/utils/constants.js"),o=s("./src/utils/hub.js"),n=(s("./src/base/feature_extraction_utils.js"),s("./src/models/feature_extractors.js"));class i{static async from_pretrained(e,t={}){const s=await(0,o.getModelJSON)(e,r.FEATURE_EXTRACTOR_NAME,!0,t),i=s.feature_extractor_type,a=n[i];if(!a)throw new Error(`Unknown feature_extractor_type: '${i}'. Please report this at ${r.GITHUB_ISSUE_URL}.`);return new a(s)}}},"./src/models/auto/image_processing_auto.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{AutoImageProcessor:()=>a});var r=s("./src/utils/constants.js"),o=s("./src/utils/hub.js"),n=s("./src/base/image_processors_utils.js"),i=s("./src/models/image_processors.js");class a{static async from_pretrained(e,t={}){const s=await(0,o.getModelJSON)(e,r.IMAGE_PROCESSOR_NAME,!0,t),a=s.image_processor_type??s.feature_extractor_type;let l=i[a];return l||(void 0!==a&&console.warn(`Image processor type '${a}' not found, assuming base ImageProcessor. Please report this at ${r.GITHUB_ISSUE_URL}.`),l=n.ImageProcessor),new l(s)}}},"./src/models/auto/processing_auto.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{AutoProcessor:()=>c});var r=s("./src/utils/constants.js"),o=s("./src/utils/hub.js"),n=s("./src/base/processing_utils.js"),i=s("./src/models/processors.js"),a=s("./src/models/image_processors.js"),l=s("./src/models/feature_extractors.js");class c{static async from_pretrained(e,t={}){const s=await(0,o.getModelJSON)(e,r.IMAGE_PROCESSOR_NAME,!0,t),{image_processor_type:c,feature_extractor_type:d,processor_class:u}=s;if(u&&i[u])return i[u].from_pretrained(e,t);if(!c&&!d)throw new Error("No `image_processor_type` or `feature_extractor_type` found in the config.");const _={};if(c){const e=a[c];if(!e)throw new Error(`Unknown image_processor_type: '${c}'.`);_.image_processor=new e(s)}if(d){const e=a[d];if(e)_.image_processor=new e(s);else{const e=l[d];if(!e)throw new Error(`Unknown feature_extractor_type: '${d}'.`);_.feature_extractor=new e(s)}}return new n.Processor({},_)}}},"./src/models/beit/image_processing_beit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{BeitFeatureExtractor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}},"./src/models/bit/image_processing_bit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{BitImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}},"./src/models/chinese_clip/image_processing_chinese_clip.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ChineseCLIPFeatureExtractor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}},"./src/models/clap/feature_extraction_clap.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ClapFeatureExtractor:()=>n});var r=s("./src/base/feature_extraction_utils.js"),o=(s("./src/utils/tensor.js"),s("./src/utils/audio.js"));class n extends r.FeatureExtractor{constructor(e){super(e),this.mel_filters=(0,o.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,null,"htk"),this.mel_filters_slaney=(0,o.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,"slaney","slaney"),this.window=(0,o.window_function)(this.config.fft_window_size,"hann")}async _get_input_mel(e,t,s,r){let o,n=!1;const i=e.length-t;if(i>0){if("rand_trunc"!==s)throw new Error(`Truncation strategy "${s}" not implemented`);{n=!0;const s=Math.floor(Math.random()*(i+1));e=e.subarray(s,s+t),o=await this._extract_fbank_features(e,this.mel_filters_slaney,this.config.nb_max_samples)}}else{if(i<0){let s=new Float64Array(t);if(s.set(e),"repeat"===r)for(let r=e.length;r<t;r+=e.length)s.set(e.subarray(0,Math.min(e.length,t-r)),r);else if("repeatpad"===r)for(let t=e.length;t<-i;t+=e.length)s.set(e,t);e=s}if("fusion"===s)throw new Error(`Truncation strategy "${s}" not implemented`);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,s=null){return(0,o.spectrogram)(e,this.window,this.config.fft_window_size,this.config.hop_length,{power:2,mel_filters:t,log_mel:"dB",max_num_frames:s,do_pad:!1,transpose:!0})}async _call(e,{max_length:t=null}={}){(0,r.validate_audio_inputs)(e,"ClapFeatureExtractor");return{input_features:(await this._get_input_mel(e,t??this.config.nb_max_samples,this.config.truncation,this.config.padding)).unsqueeze_(0)}}}},"./src/models/clip/image_processing_clip.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{CLIPFeatureExtractor:()=>n,CLIPImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}class n extends o{}},"./src/models/convnext/image_processing_convnext.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ConvNextFeatureExtractor:()=>n,ConvNextImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{constructor(e){super(e),this.crop_pct=this.config.crop_pct??.875}async resize(e){const t=this.size?.shortest_edge;if(void 0===t)throw new Error("Size dictionary must contain 'shortest_edge' key.");if(t<384){const s=Math.floor(t/this.crop_pct),[r,o]=this.get_resize_output_image_size(e,{shortest_edge:s});e=await e.resize(r,o,{resample:this.resample}),e=await e.center_crop(t,t)}else e=await e.resize(t,t,{resample:this.resample});return e}}class n extends o{}},"./src/models/dac/feature_extraction_dac.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DacFeatureExtractor:()=>o});var r=s("./src/models/encodec/feature_extraction_encodec.js");class o extends r.EncodecFeatureExtractor{}},"./src/models/deit/image_processing_deit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DeiTFeatureExtractor:()=>n,DeiTImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}class n extends o{}},"./src/models/detr/image_processing_detr.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DetrFeatureExtractor:()=>i,DetrImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js"),o=s("./src/utils/tensor.js");class n extends r.ImageProcessor{async _call(e){const t=await super._call(e),s=[t.pixel_values.dims[0],64,64],r=(0,o.full)(s,1n);return{...t,pixel_mask:r}}post_process_object_detection(...e){return(0,r.post_process_object_detection)(...e)}post_process_panoptic_segmentation(...e){return(0,r.post_process_panoptic_segmentation)(...e)}post_process_instance_segmentation(...e){return(0,r.post_process_instance_segmentation)(...e)}}class i extends n{}},"./src/models/donut/image_processing_donut.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DonutFeatureExtractor:()=>n,DonutImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{pad_image(e,t,s,r={}){const[o,n,i]=t;let a=this.image_mean;Array.isArray(this.image_mean)||(a=new Array(i).fill(a));let l=this.image_std;Array.isArray(l)||(l=new Array(i).fill(a));const c=a.map(((e,t)=>-e/l[t]));return super.pad_image(e,t,s,{center:!0,constant_values:c,...r})}}class n extends o{}},"./src/models/dpt/image_processing_dpt.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DPTFeatureExtractor:()=>n,DPTImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}class n extends o{}},"./src/models/efficientnet/image_processing_efficientnet.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{EfficientNetImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{constructor(e){super(e),this.include_top=this.config.include_top??!0,this.include_top&&(this.image_std=this.image_std.map((e=>e*e)))}}},"./src/models/encodec/feature_extraction_encodec.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{EncodecFeatureExtractor:()=>n});var r=s("./src/base/feature_extraction_utils.js"),o=s("./src/utils/tensor.js");class n extends r.FeatureExtractor{async _call(e){(0,r.validate_audio_inputs)(e,"EncodecFeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));const t=this.config.feature_size;if(e.length%t!=0)throw new Error(`The length of the audio data must be a multiple of the number of channels (${t}).`);const s=[1,t,e.length/t];return{input_values:new o.Tensor("float32",e,s)}}}},"./src/models/feature_extractors.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ASTFeatureExtractor:()=>r.ASTFeatureExtractor,ClapFeatureExtractor:()=>n.ClapFeatureExtractor,DacFeatureExtractor:()=>i.DacFeatureExtractor,EncodecFeatureExtractor:()=>o.EncodecFeatureExtractor,ImageFeatureExtractor:()=>h.ImageProcessor,MoonshineFeatureExtractor:()=>a.MoonshineFeatureExtractor,PyAnnoteFeatureExtractor:()=>l.PyAnnoteFeatureExtractor,SeamlessM4TFeatureExtractor:()=>c.SeamlessM4TFeatureExtractor,SnacFeatureExtractor:()=>d.SnacFeatureExtractor,SpeechT5FeatureExtractor:()=>u.SpeechT5FeatureExtractor,Wav2Vec2FeatureExtractor:()=>_.Wav2Vec2FeatureExtractor,WeSpeakerFeatureExtractor:()=>p.WeSpeakerFeatureExtractor,WhisperFeatureExtractor:()=>m.WhisperFeatureExtractor});var r=s("./src/models/audio_spectrogram_transformer/feature_extraction_audio_spectrogram_transformer.js"),o=s("./src/models/encodec/feature_extraction_encodec.js"),n=s("./src/models/clap/feature_extraction_clap.js"),i=s("./src/models/dac/feature_extraction_dac.js"),a=s("./src/models/moonshine/feature_extraction_moonshine.js"),l=s("./src/models/pyannote/feature_extraction_pyannote.js"),c=s("./src/models/seamless_m4t/feature_extraction_seamless_m4t.js"),d=s("./src/models/snac/feature_extraction_snac.js"),u=s("./src/models/speecht5/feature_extraction_speecht5.js"),_=s("./src/models/wav2vec2/feature_extraction_wav2vec2.js"),p=s("./src/models/wespeaker/feature_extraction_wespeaker.js"),m=s("./src/models/whisper/feature_extraction_whisper.js"),h=s("./src/base/image_processors_utils.js")},"./src/models/florence2/processing_florence2.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Florence2Processor:()=>i});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js");class i extends r.Processor{static tokenizer_class=n.AutoTokenizer;static image_processor_class=o.AutoImageProcessor;constructor(e,t){super(e,t);const{tasks_answer_post_processing_type:s,task_prompts_without_inputs:r,task_prompts_with_input:o}=this.image_processor.config;this.tasks_answer_post_processing_type=new Map(Object.entries(s??{})),this.task_prompts_without_inputs=new Map(Object.entries(r??{})),this.task_prompts_with_input=new Map(Object.entries(o??{})),this.regexes={quad_boxes:/(.+?)<loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)>/gm,bboxes:/([^<]+)?<loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)>/gm},this.size_per_bin=1e3}construct_prompts(e){"string"==typeof e&&(e=[e]);const t=[];for(const s of e)if(this.task_prompts_without_inputs.has(s))t.push(this.task_prompts_without_inputs.get(s));else{for(const[e,r]of this.task_prompts_with_input)if(s.includes(e)){t.push(r.replaceAll("{input}",s).replaceAll(e,""));break}t.length!==e.length&&t.push(s)}return t}post_process_generation(e,t,s){const r=this.tasks_answer_post_processing_type.get(t)??"pure_text";let o;switch(e=e.replaceAll("<s>","").replaceAll("</s>",""),r){case"pure_text":o=e;break;case"description_with_bboxes":case"bboxes":case"phrase_grounding":case"ocr":const n="ocr"===r?"quad_boxes":"bboxes",i=e.matchAll(this.regexes[n]),a=[],l=[];for(const[e,t,...r]of i)a.push(t?t.trim():a.at(-1)??""),l.push(r.map(((e,t)=>(Number(e)+.5)/this.size_per_bin*s[t%2])));o={labels:a,[n]:l};break;default:throw new Error(`Task "${t}" (of type "${r}") not yet implemented.`)}return{[t]:o}}async _call(e,t=null,s={}){if(!e&&!t)throw new Error("Either text or images must be provided");return{...await this.image_processor(e,s),...t?this.tokenizer(t,s):{}}}}},"./src/models/glpn/image_processing_glpn.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{GLPNFeatureExtractor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}},"./src/models/grounding_dino/image_processing_grounding_dino.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{GroundingDinoImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js"),o=s("./src/utils/tensor.js");class n extends r.ImageProcessor{async _call(e){const t=await super._call(e),s=t.pixel_values.dims,r=(0,o.ones)([s[0],s[2],s[3]]);return{...t,pixel_mask:r}}}},"./src/models/grounding_dino/processing_grounding_dino.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{GroundingDinoProcessor:()=>l});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js"),i=s("./src/base/image_processors_utils.js");function a(e,t){const s=e.dims.at(-1)-1,r=e.tolist();r.fill(!1,0,1),r.fill(!1,s);const o=t.tolist();return r.map(((e,t)=>e?t:null)).filter((e=>null!==e)).map((e=>o[e]))}class l extends r.Processor{static tokenizer_class=n.AutoTokenizer;static image_processor_class=o.AutoImageProcessor;async _call(e,t,s={}){const r=e?await this.image_processor(e,s):{};return{...t?this.tokenizer(t,s):{},...r}}post_process_grounded_object_detection(e,t,{box_threshold:s=.25,text_threshold:r=.25,target_sizes:o=null}={}){const{logits:n,pred_boxes:l}=e,c=n.dims[0];if(null!==o&&o.length!==c)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");const d=n.dims.at(1),u=n.sigmoid(),_=u.max(-1).tolist(),p=l.tolist().map((e=>e.map((e=>(0,i.center_to_corners_format)(e))))),m=[];for(let e=0;e<c;++e){const n=null!==o?o[e]:null;null!==n&&(p[e]=p[e].map((e=>e.map(((e,t)=>e*n[(t+1)%2])))));const i=_[e],l=[],c=[],h=[];for(let o=0;o<d;++o){const n=i[o];if(n<=s)continue;const d=p[e][o],_=u[e][o];l.push(n),h.push(d);const m=a(_.gt(r),t[e]);c.push(m)}m.push({scores:l,boxes:h,labels:this.batch_decode(c)})}return m}}},"./src/models/idefics3/image_processing_idefics3.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Idefics3ImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js"),o=s("./src/utils/tensor.js");class n extends r.ImageProcessor{constructor(e){super(e),this.do_image_splitting=e.do_image_splitting??!0,this.max_image_size=e.max_image_size}get_resize_for_vision_encoder(e,t){let[s,r]=e.dims.slice(-2);const o=r/s;return r>=s?(r=Math.ceil(r/t)*t,s=Math.floor(r/o),s=Math.ceil(s/t)*t):(s=Math.ceil(s/t)*t,r=Math.floor(s*o),r=Math.ceil(r/t)*t),{height:s,width:r}}async _call(e,{do_image_splitting:t=null,return_row_col_info:s=!1}={}){let r;if(Array.isArray(e)){if(0===e.length||!e[0])throw new Error("No images provided.");r=Array.isArray(e[0])?e:[e]}else r=[[e]];let n=[],i=[],a=[];const l=[],c=[];for(const e of r){let s=await Promise.all(e.map((e=>this.preprocess(e))));l.push(...s.map((e=>e.original_size))),c.push(...s.map((e=>e.reshaped_input_size))),s.forEach((e=>e.pixel_values.unsqueeze_(0)));const{longest_edge:r}=this.max_image_size;let d;if(t??this.do_image_splitting){let e=new Array(s.length),t=new Array(s.length);d=await Promise.all(s.map((async(s,n)=>{const i=this.get_resize_for_vision_encoder(s.pixel_values,r),a=await(0,o.interpolate_4d)(s.pixel_values,{size:[i.height,i.width]}),{frames:l,num_splits_h:c,num_splits_w:d}=await this.split_image(a,this.max_image_size);return e[n]=c,t[n]=d,(0,o.cat)(l,0)}))),i.push(e),a.push(t)}else{const e=[r,r];d=await Promise.all(s.map((t=>(0,o.interpolate_4d)(t.pixel_values,{size:e})))),i.push(new Array(s.length).fill(0)),a.push(new Array(s.length).fill(0))}n.push((0,o.cat)(d,0))}const d=n.length,[u,_,p,m]=n[0].dims;let h,g;if(1===d)h=n[0].unsqueeze_(0),g=(0,o.full)([d,u,p,m],!0);else{const e=Math.max(...n.map((e=>e.dims.at(0))));g=(0,o.full)([d,e,p,m],!0);const t=g.data,s=e*p*m;for(let r=0;r<d;++r){const i=n[r].dims[0];if(i<e){n[r]=(0,o.cat)([n[r],(0,o.full)([e-i,_,p,m],0)],0);const a=r*s+i*p*m,l=(r+1)*s;t.fill(!1,a,l)}}h=(0,o.stack)(n,0)}return{pixel_values:h,pixel_attention_mask:g,original_sizes:l,reshaped_input_sizes:c,...s?{rows:i,cols:a}:{}}}async split_image(e,{longest_edge:t}){const s=t,r=t,n=[],[i,a]=e.dims.slice(-2);let l=0,c=0;if(i>s||a>r){l=Math.ceil(i/s),c=Math.ceil(a/r);const t=Math.ceil(i/l),d=Math.ceil(a/c);for(let s=0;s<l;++s)for(let r=0;r<c;++r){let u,_,p,m;s===l-1?(_=i-t,m=i):(_=s*t,m=(s+1)*t),r===c-1?(u=a-d,p=a):(u=r*d,p=(r+1)*d);const h=[_,u],g=[m,p],f=await(0,o.slice)(e,h,g,[2,3]);n.push(f)}const u=s,_=r;i===u&&a===_||(e=await(0,o.interpolate_4d)(e,{size:[u,_]}))}return n.push(e),{frames:n,num_splits_h:l,num_splits_w:c}}}},"./src/models/idefics3/processing_idefics3.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Idefics3Processor:()=>l});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js"),i=(s("./src/utils/image.js"),s("./src/utils/core.js"));function a(e,t,s,r,o,n){return 0===e&&0===t?function(e,t,s,r){return`${t}${r}`+s.repeat(e)+`${t}`}(s,r,o,n):function(e,t,s,r,o,n){let i="";for(let n=0;n<t;++n){for(let t=0;t<s;++t)i+=r+`<row_${n+1}_col_${t+1}>`+o.repeat(e);i+="\n"}return i+=`\n${r}${n}`+o.repeat(e)+`${r}`,i}(s,e,t,r,o,n)}class l extends r.Processor{static image_processor_class=o.AutoImageProcessor;static tokenizer_class=n.AutoTokenizer;static uses_processor_config=!0;fake_image_token="<fake_token_around_image>";image_token="<image>";global_img_token="<global-img>";async _call(e,t=null,s={}){let r;s.return_row_col_info??=!0,t&&(r=await this.image_processor(t,s)),Array.isArray(e)||(e=[e]);const o=r.rows??[new Array(e.length).fill(0)],n=r.cols??[new Array(e.length).fill(0)],l=this.config.image_seq_len,c=[],d=[];for(let t=0;t<e.length;++t){const s=e[t],r=o[t],u=n[t];c.push((0,i.count)(s,this.image_token));const _=r.map(((e,t)=>a(e,u[t],l,this.fake_image_token,this.image_token,this.global_img_token))),p=s.split(this.image_token);if(0===p.length)throw new Error("The image token should be present in the text.");let m=p[0];for(let e=0;e<_.length;++e)m+=_[e]+p[e+1];d.push(m)}return{...this.tokenizer(d),...r}}}},"./src/models/image_processors.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{BeitFeatureExtractor:()=>r.BeitFeatureExtractor,BitImageProcessor:()=>o.BitImageProcessor,CLIPFeatureExtractor:()=>i.CLIPFeatureExtractor,CLIPImageProcessor:()=>i.CLIPImageProcessor,ChineseCLIPFeatureExtractor:()=>n.ChineseCLIPFeatureExtractor,ConvNextFeatureExtractor:()=>a.ConvNextFeatureExtractor,ConvNextImageProcessor:()=>a.ConvNextImageProcessor,DPTFeatureExtractor:()=>u.DPTFeatureExtractor,DPTImageProcessor:()=>u.DPTImageProcessor,DeiTFeatureExtractor:()=>l.DeiTFeatureExtractor,DeiTImageProcessor:()=>l.DeiTImageProcessor,DetrFeatureExtractor:()=>c.DetrFeatureExtractor,DetrImageProcessor:()=>c.DetrImageProcessor,DonutFeatureExtractor:()=>d.DonutFeatureExtractor,DonutImageProcessor:()=>d.DonutImageProcessor,EfficientNetImageProcessor:()=>_.EfficientNetImageProcessor,GLPNFeatureExtractor:()=>p.GLPNFeatureExtractor,GroundingDinoImageProcessor:()=>m.GroundingDinoImageProcessor,Idefics3ImageProcessor:()=>h.Idefics3ImageProcessor,JinaCLIPImageProcessor:()=>f.JinaCLIPImageProcessor,LlavaOnevisionImageProcessor:()=>w.LlavaOnevisionImageProcessor,Mask2FormerImageProcessor:()=>M.Mask2FormerImageProcessor,MaskFormerFeatureExtractor:()=>x.MaskFormerFeatureExtractor,MaskFormerImageProcessor:()=>x.MaskFormerImageProcessor,MobileNetV1FeatureExtractor:()=>b.MobileNetV1FeatureExtractor,MobileNetV1ImageProcessor:()=>b.MobileNetV1ImageProcessor,MobileNetV2FeatureExtractor:()=>k.MobileNetV2FeatureExtractor,MobileNetV2ImageProcessor:()=>k.MobileNetV2ImageProcessor,MobileNetV3FeatureExtractor:()=>y.MobileNetV3FeatureExtractor,MobileNetV3ImageProcessor:()=>y.MobileNetV3ImageProcessor,MobileNetV4FeatureExtractor:()=>v.MobileNetV4FeatureExtractor,MobileNetV4ImageProcessor:()=>v.MobileNetV4ImageProcessor,MobileViTFeatureExtractor:()=>T.MobileViTFeatureExtractor,MobileViTImageProcessor:()=>T.MobileViTImageProcessor,NougatImageProcessor:()=>P.NougatImageProcessor,OwlViTFeatureExtractor:()=>C.OwlViTFeatureExtractor,OwlViTImageProcessor:()=>C.OwlViTImageProcessor,Owlv2ImageProcessor:()=>F.Owlv2ImageProcessor,Phi3VImageProcessor:()=>S.Phi3VImageProcessor,PvtImageProcessor:()=>A.PvtImageProcessor,Qwen2VLImageProcessor:()=>E.Qwen2VLImageProcessor,RTDetrImageProcessor:()=>L.RTDetrImageProcessor,SamImageProcessor:()=>I.SamImageProcessor,SegformerFeatureExtractor:()=>z.SegformerFeatureExtractor,SegformerImageProcessor:()=>z.SegformerImageProcessor,SiglipImageProcessor:()=>j.SiglipImageProcessor,SmolVLMImageProcessor:()=>D.SmolVLMImageProcessor,Swin2SRImageProcessor:()=>O.Swin2SRImageProcessor,VLMImageProcessor:()=>g.VLMImageProcessor,ViTFeatureExtractor:()=>V.ViTFeatureExtractor,ViTImageProcessor:()=>V.ViTImageProcessor,VitMatteImageProcessor:()=>N.VitMatteImageProcessor,VitPoseImageProcessor:()=>B.VitPoseImageProcessor,YolosFeatureExtractor:()=>R.YolosFeatureExtractor,YolosImageProcessor:()=>R.YolosImageProcessor});var r=s("./src/models/beit/image_processing_beit.js"),o=s("./src/models/bit/image_processing_bit.js"),n=s("./src/models/chinese_clip/image_processing_chinese_clip.js"),i=s("./src/models/clip/image_processing_clip.js"),a=s("./src/models/convnext/image_processing_convnext.js"),l=s("./src/models/deit/image_processing_deit.js"),c=s("./src/models/detr/image_processing_detr.js"),d=s("./src/models/donut/image_processing_donut.js"),u=s("./src/models/dpt/image_processing_dpt.js"),_=s("./src/models/efficientnet/image_processing_efficientnet.js"),p=s("./src/models/glpn/image_processing_glpn.js"),m=s("./src/models/grounding_dino/image_processing_grounding_dino.js"),h=s("./src/models/idefics3/image_processing_idefics3.js"),g=s("./src/models/janus/image_processing_janus.js"),f=s("./src/models/jina_clip/image_processing_jina_clip.js"),w=s("./src/models/llava_onevision/image_processing_llava_onevision.js"),M=s("./src/models/mask2former/image_processing_mask2former.js"),x=s("./src/models/maskformer/image_processing_maskformer.js"),b=s("./src/models/mobilenet_v1/image_processing_mobilenet_v1.js"),k=s("./src/models/mobilenet_v2/image_processing_mobilenet_v2.js"),y=s("./src/models/mobilenet_v3/image_processing_mobilenet_v3.js"),v=s("./src/models/mobilenet_v4/image_processing_mobilenet_v4.js"),T=s("./src/models/mobilevit/image_processing_mobilevit.js"),P=s("./src/models/nougat/image_processing_nougat.js"),F=s("./src/models/owlv2/image_processing_owlv2.js"),C=s("./src/models/owlvit/image_processing_owlvit.js"),S=s("./src/models/phi3_v/image_processing_phi3_v.js"),A=s("./src/models/pvt/image_processing_pvt.js"),E=s("./src/models/qwen2_vl/image_processing_qwen2_vl.js"),L=s("./src/models/rt_detr/image_processing_rt_detr.js"),I=s("./src/models/sam/image_processing_sam.js"),z=s("./src/models/segformer/image_processing_segformer.js"),j=s("./src/models/siglip/image_processing_siglip.js"),D=s("./src/models/smolvlm/image_processing_smolvlm.js"),O=s("./src/models/swin2sr/image_processing_swin2sr.js"),V=s("./src/models/vit/image_processing_vit.js"),N=s("./src/models/vitmatte/image_processing_vitmatte.js"),B=s("./src/models/vitpose/image_processing_vitpose.js"),R=s("./src/models/yolos/image_processing_yolos.js")},"./src/models/janus/image_processing_janus.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{VLMImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{constructor(e){super({do_pad:!0,pad_size:{width:e.image_size,height:e.image_size},...e}),this.constant_values=this.config.background_color.map((e=>e*this.rescale_factor))}pad_image(e,t,s,r){return super.pad_image(e,t,s,{constant_values:this.constant_values,center:!0,...r})}}},"./src/models/janus/processing_janus.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{VLChatProcessor:()=>c});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js"),i=s("./src/utils/core.js"),a=s("./src/utils/tensor.js"),l=s("./src/utils/image.js");class c extends r.Processor{static image_processor_class=o.AutoImageProcessor;static tokenizer_class=n.AutoTokenizer;static uses_processor_config=!0;constructor(e,t){super(e,t),this.image_tag=this.config.image_tag,this.image_start_tag=this.config.image_start_tag,this.image_end_tag=this.config.image_end_tag,this.num_image_tokens=this.config.num_image_tokens}async _call(e,{images:t=null,chat_template:s="default"}={}){t?Array.isArray(t)||(t=[t]):t=await Promise.all(e.filter((e=>e.images)).flatMap((e=>e.images)).map((e=>l.RawImage.read(e))));const r=this.tokenizer,o=e=>r.encode(e,{add_special_tokens:!1}),n=r.apply_chat_template(e,{tokenize:!1,add_generation_prompt:!0,chat_template:s}).split(this.image_tag),c=n.length-1;if(t.length!==c)throw new Error(`Number of images provided (${t.length}) does not match number of "${this.image_tag}" image tags (${c})`);const[d,u,_]=r.model.convert_tokens_to_ids([this.image_tag,this.image_start_tag,this.image_end_tag]);let p=o(n[0]),m=new Array(p.length).fill(!1);for(let e=1;e<n.length;++e){const t=new Array(this.num_image_tokens).fill(d),s=o(n[e]);p=(0,i.mergeArrays)(p,[u],t,[_],s);const r=new Array(this.num_image_tokens).fill(!0);m=(0,i.mergeArrays)(m,[!1],r,[!1],new Array(s.length).fill(!1))}const h=[1,p.length],g={input_ids:new a.Tensor("int64",p,h),attention_mask:new a.Tensor("int64",new Array(p.length).fill(1),h),images_seq_mask:new a.Tensor("bool",m,h),images_emb_mask:new a.Tensor("bool",new Array(c*this.num_image_tokens).fill(!0),[1,c,this.num_image_tokens])};if(t&&t.length>0){const e=await this.image_processor(t);return e.pixel_values.unsqueeze_(0),{...g,...e}}return g}}},"./src/models/jina_clip/image_processing_jina_clip.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{JinaCLIPImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{constructor(e){const{resize_mode:t,fill_color:s,interpolation:r,size:o,...n}=e;super({...n,size:"squash"===t?{width:o,height:o}:"shortest"===t?{shortest_edge:o}:{longest_edge:o},resample:"bicubic"===r?3:2,do_center_crop:!0,crop_size:o,do_normalize:!0})}}},"./src/models/jina_clip/processing_jina_clip.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{JinaCLIPProcessor:()=>i});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js");class i extends r.Processor{static tokenizer_class=n.AutoTokenizer;static image_processor_class=o.AutoImageProcessor;async _call(e=null,t=null,s={}){if(!e&&!t)throw new Error("Either text or images must be provided");return{...e?this.tokenizer(e,s):{},...t?await this.image_processor(t,s):{}}}}},"./src/models/llava_onevision/image_processing_llava_onevision.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{LlavaOnevisionImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}},"./src/models/mask2former/image_processing_mask2former.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Mask2FormerImageProcessor:()=>o});var r=s("./src/models/maskformer/image_processing_maskformer.js");class o extends r.MaskFormerImageProcessor{}},"./src/models/maskformer/image_processing_maskformer.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MaskFormerFeatureExtractor:()=>n,MaskFormerImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{post_process_panoptic_segmentation(...e){return(0,r.post_process_panoptic_segmentation)(...e)}post_process_instance_segmentation(...e){return(0,r.post_process_instance_segmentation)(...e)}}class n extends o{}},"./src/models/mgp_str/processing_mgp_str.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MgpstrProcessor:()=>l});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js"),i=s("./src/utils/maths.js");const a={char:["char_decode",1],bpe:["bpe_decode",2],wp:["wp_decode",102]};class l extends r.Processor{static tokenizer_class=n.AutoTokenizer;static image_processor_class=o.AutoImageProcessor;get char_tokenizer(){return this.components.char_tokenizer}get bpe_tokenizer(){return this.components.bpe_tokenizer}get wp_tokenizer(){return this.components.wp_tokenizer}_decode_helper(e,t){if(!a.hasOwnProperty(t))throw new Error(`Format ${t} is not supported.`);const[s,r]=a[t],o=this[s].bind(this),[n,l]=e.dims,c=[],d=[],u=e.tolist();for(let e=0;e<n;++e){const t=u[e],s=[],o=[];for(let e=1;e<l;++e){const[n,a]=(0,i.max)((0,i.softmax)(t[e]));if(o.push(n),a==r)break;s.push(a)}const n=o.length>0?o.reduce(((e,t)=>e*t),1):0;d.push(s),c.push(n)}return[o(d),c]}char_decode(e){return this.char_tokenizer.batch_decode(e).map((e=>e.replaceAll(" ","")))}bpe_decode(e){return this.bpe_tokenizer.batch_decode(e)}wp_decode(e){return this.wp_tokenizer.batch_decode(e).map((e=>e.replaceAll(" ","")))}batch_decode([e,t,s]){const[r,o]=this._decode_helper(e,"char"),[n,a]=this._decode_helper(t,"bpe"),[l,c]=this._decode_helper(s,"wp"),d=[],u=[];for(let e=0;e<r.length;++e){const[t,s]=(0,i.max)([o[e],a[e],c[e]]);d.push([r[e],n[e],l[e]][s]),u.push(t)}return{generated_text:d,scores:u,char_preds:r,bpe_preds:n,wp_preds:l}}static async from_pretrained(...e){const t=await super.from_pretrained(...e),s=await n.AutoTokenizer.from_pretrained("Xenova/gpt2"),r=await n.AutoTokenizer.from_pretrained("Xenova/bert-base-uncased");return t.components={image_processor:t.image_processor,char_tokenizer:t.tokenizer,bpe_tokenizer:s,wp_tokenizer:r},t}async _call(e,t=null){const s=await this.image_processor(e);return t&&(s.labels=this.tokenizer(t).input_ids),s}}},"./src/models/mobilenet_v1/image_processing_mobilenet_v1.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MobileNetV1FeatureExtractor:()=>n,MobileNetV1ImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}class n extends o{}},"./src/models/mobilenet_v2/image_processing_mobilenet_v2.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MobileNetV2FeatureExtractor:()=>n,MobileNetV2ImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}class n extends o{}},"./src/models/mobilenet_v3/image_processing_mobilenet_v3.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MobileNetV3FeatureExtractor:()=>n,MobileNetV3ImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}class n extends o{}},"./src/models/mobilenet_v4/image_processing_mobilenet_v4.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MobileNetV4FeatureExtractor:()=>n,MobileNetV4ImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}class n extends o{}},"./src/models/mobilevit/image_processing_mobilevit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MobileViTFeatureExtractor:()=>n,MobileViTImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}class n extends o{}},"./src/models/moonshine/feature_extraction_moonshine.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MoonshineFeatureExtractor:()=>n});var r=s("./src/base/feature_extraction_utils.js"),o=s("./src/utils/tensor.js");class n extends r.FeatureExtractor{async _call(e){(0,r.validate_audio_inputs)(e,"MoonshineFeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));const t=[1,e.length];return{input_values:new o.Tensor("float32",e,t)}}}},"./src/models/moonshine/processing_moonshine.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MoonshineProcessor:()=>i});var r=s("./src/models/auto/feature_extraction_auto.js"),o=s("./src/tokenizers.js"),n=s("./src/base/processing_utils.js");class i extends n.Processor{static tokenizer_class=o.AutoTokenizer;static feature_extractor_class=r.AutoFeatureExtractor;async _call(e){return await this.feature_extractor(e)}}},"./src/models/nougat/image_processing_nougat.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{NougatImageProcessor:()=>o});var r=s("./src/models/donut/image_processing_donut.js");class o extends r.DonutImageProcessor{}},"./src/models/owlv2/image_processing_owlv2.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Owlv2ImageProcessor:()=>o});var r=s("./src/models/owlvit/image_processing_owlvit.js");class o extends r.OwlViTImageProcessor{}},"./src/models/owlvit/image_processing_owlvit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{OwlViTFeatureExtractor:()=>n,OwlViTImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{post_process_object_detection(...e){return(0,r.post_process_object_detection)(...e)}}class n extends o{}},"./src/models/owlvit/processing_owlvit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{OwlViTProcessor:()=>i});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js");class i extends r.Processor{static tokenizer_class=n.AutoTokenizer;static image_processor_class=o.AutoImageProcessor}},"./src/models/paligemma/processing_paligemma.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{PaliGemmaProcessor:()=>a});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js");const i="<image>";class a extends r.Processor{static tokenizer_class=n.AutoTokenizer;static image_processor_class=o.AutoImageProcessor;static uses_processor_config=!1;async _call(e,t=null,s={}){t||(console.warn("You are using PaliGemma without a text prefix. It will perform as a picture-captioning model."),t=""),Array.isArray(e)||(e=[e]),Array.isArray(t)||(t=[t]);const r=this.tokenizer.bos_token,o=this.image_processor.config.image_seq_length;let n;t.some((e=>e.includes(i)))?n=t.map((e=>{const t=e.replaceAll(i,i.repeat(o)),s=t.lastIndexOf(i),n=-1===s?0:s+7;return t.slice(0,n)+r+t.slice(n)+"\n"})):(console.warn("You are passing both `text` and `images` to `PaliGemmaProcessor`. The processor expects special image tokens in the text, as many tokens as there are images per each text. It is recommended to add `<image>` tokens in the very beginning of your text. For this call, we will infer how many images each text has and add special tokens."),n=t.map((t=>function(e,t,s,r,o){return`${r.repeat(s*o)}${t}${e}\n`}(t,r,o,i,e.length))));const a=this.tokenizer(n,s);return{...await this.image_processor(e,s),...a}}}},"./src/models/phi3_v/image_processing_phi3_v.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Phi3VImageProcessor:()=>d});var r=s("./src/base/image_processors_utils.js"),o=s("./src/utils/tensor.js");const n=336,i=[2,3],{ceil:a,floor:l,sqrt:c}=Math;class d extends r.ImageProcessor{constructor(e){super({...e,do_normalize:!0,do_pad:!0,pad_size:"custom",do_convert_rgb:!0,do_resize:!0}),this._num_crops=e.num_crops}calc_num_image_tokens_from_image_size(e,t){const{num_img_tokens:s}=this.config;return l((l(t/n)*l(e/n)+1)*s+1+(l(t/n)+1)*c(s))}get_resize_output_image_size(e,t){const s=this._num_crops,[r,o]=e.size;let n=r/o,i=1;for(;i*Math.ceil(i/n)<=s;)i+=1;i-=1;const a=Math.floor(336*i);return[a,Math.floor(a/n)]}pad_image(e,t,s,r={}){const[o,i]=t,l=n*a(o/n),c=n*a(i/n),d=[1,1,1].map(((e,t)=>(e-this.image_mean[t])/this.image_std[t]));return super.pad_image(e,t,{width:c,height:l},{center:!0,constant_values:d,...r})}async _call(e,{num_crops:t=null}={}){if(this._num_crops=t??=this.config.num_crops,t<4||c(t)%1!=0)throw new Error("num_crops must be a square number >= 4");Array.isArray(e)||(e=[e]);const s=e.length,r=await Promise.all(e.map((e=>this.preprocess(e)))),d=r.map((e=>e.original_size)),u=r.map((e=>e.reshaped_input_size)),_=[];for(const{pixel_values:e}of r){e.unsqueeze_(0);const[s,r]=e.dims.slice(-2),a=await(0,o.interpolate_4d)(e,{size:[n,n],mode:"bicubic"});if(t>0){const d=[],u=c(t),p=l(r/u),m=l(s/u);for(let t=0;t<u;++t)for(let n=0;n<u;++n){let a,l,c,_;t===u-1?(l=s-m,_=s):(l=t*m,_=(t+1)*m),n===u-1?(a=r-p,c=r):(a=n*p,c=(n+1)*p);const h=[l,a],g=[_,c],f=await(0,o.slice)(e,h,g,i);d.push(f)}const h=await(0,o.interpolate_4d)((0,o.cat)(d,0),{size:[n,n],mode:"bicubic"});_.push((0,o.cat)([a,h],0))}else _.push(a)}const p=(0,o.stack)(_,0),m=u.map((e=>e.map((e=>n*a(e/n)))));return{pixel_values:p,original_sizes:d,reshaped_input_sizes:u,image_sizes:new o.Tensor("int64",m.flat(),[s,2]),num_img_tokens:m.map((([e,t])=>this.calc_num_image_tokens_from_image_size(t,e)))}}}},"./src/models/phi3_v/processing_phi3_v.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Phi3VProcessor:()=>l});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js");s("./src/utils/image.js");const i="<|image|>",a=/<\|image_\d+\|>/g;class l extends r.Processor{static image_processor_class=o.AutoImageProcessor;static tokenizer_class=n.AutoTokenizer;async _call(e,t=null,{padding:s=!0,truncation:r=!0,num_crops:o=null}={}){let n,l;if(Array.isArray(e)||(e=[e]),t){l=await this.image_processor(t,{num_crops:o});const{num_img_tokens:c}=l,d=e.map(((e,t)=>e.split(a).join(i.repeat(c[t]))));n=this.tokenizer(d,{padding:s,truncation:r});const u=this.tokenizer.model.convert_tokens_to_ids([i])[0];n.input_ids.map_((e=>e==u?-e:e))}else n=this.tokenizer(e);return{...n,...l}}}},"./src/models/processors.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Florence2Processor:()=>r.Florence2Processor,GroundingDinoProcessor:()=>o.GroundingDinoProcessor,Idefics3Processor:()=>n.Idefics3Processor,JinaCLIPProcessor:()=>a.JinaCLIPProcessor,MgpstrProcessor:()=>l.MgpstrProcessor,MoonshineProcessor:()=>c.MoonshineProcessor,OwlViTProcessor:()=>d.OwlViTProcessor,PaliGemmaProcessor:()=>_.PaliGemmaProcessor,Phi3VProcessor:()=>u.Phi3VProcessor,PyAnnoteProcessor:()=>p.PyAnnoteProcessor,Qwen2VLProcessor:()=>m.Qwen2VLProcessor,SamProcessor:()=>h.SamProcessor,SmolVLMProcessor:()=>g.SmolVLMProcessor,SpeechT5Processor:()=>f.SpeechT5Processor,UltravoxProcessor:()=>w.UltravoxProcessor,VLChatProcessor:()=>i.VLChatProcessor,Wav2Vec2Processor:()=>M.Wav2Vec2Processor,Wav2Vec2ProcessorWithLM:()=>x.Wav2Vec2ProcessorWithLM,WhisperProcessor:()=>b.WhisperProcessor});var r=s("./src/models/florence2/processing_florence2.js"),o=s("./src/models/grounding_dino/processing_grounding_dino.js"),n=s("./src/models/idefics3/processing_idefics3.js"),i=s("./src/models/janus/processing_janus.js"),a=s("./src/models/jina_clip/processing_jina_clip.js"),l=s("./src/models/mgp_str/processing_mgp_str.js"),c=s("./src/models/moonshine/processing_moonshine.js"),d=s("./src/models/owlvit/processing_owlvit.js"),u=s("./src/models/phi3_v/processing_phi3_v.js"),_=s("./src/models/paligemma/processing_paligemma.js"),p=s("./src/models/pyannote/processing_pyannote.js"),m=s("./src/models/qwen2_vl/processing_qwen2_vl.js"),h=s("./src/models/sam/processing_sam.js"),g=s("./src/models/smolvlm/processing_smolvlm.js"),f=s("./src/models/speecht5/processing_speecht5.js"),w=s("./src/models/ultravox/processing_ultravox.js"),M=s("./src/models/wav2vec2/processing_wav2vec2.js"),x=s("./src/models/wav2vec2_with_lm/processing_wav2vec2_with_lm.js"),b=s("./src/models/whisper/processing_whisper.js")},"./src/models/pvt/image_processing_pvt.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{PvtImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}},"./src/models/pyannote/feature_extraction_pyannote.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{PyAnnoteFeatureExtractor:()=>i});var r=s("./src/base/feature_extraction_utils.js"),o=s("./src/utils/tensor.js"),n=s("./src/utils/maths.js");class i extends r.FeatureExtractor{async _call(e){(0,r.validate_audio_inputs)(e,"PyAnnoteFeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));const t=[1,1,e.length];return{input_values:new o.Tensor("float32",e,t)}}samples_to_frames(e){return(e-this.config.offset)/this.config.step}post_process_speaker_diarization(e,t){const s=t/this.samples_to_frames(t)/this.config.sampling_rate,r=[];for(const t of e.tolist()){const e=[];let o=-1;for(let s=0;s<t.length;++s){const r=(0,n.softmax)(t[s]),[i,a]=(0,n.max)(r),[l,c]=[s,s+1];a!==o?(o=a,e.push({id:a,start:l,end:c,score:i})):(e.at(-1).end=c,e.at(-1).score+=i)}r.push(e.map((({id:e,start:t,end:r,score:o})=>({id:e,start:t*s,end:r*s,confidence:o/(r-t)}))))}return r}}},"./src/models/pyannote/processing_pyannote.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{PyAnnoteProcessor:()=>n});var r=s("./src/base/processing_utils.js"),o=s("./src/models/pyannote/feature_extraction_pyannote.js");class n extends r.Processor{static feature_extractor_class=o.PyAnnoteFeatureExtractor;async _call(e){return await this.feature_extractor(e)}post_process_speaker_diarization(...e){return this.feature_extractor.post_process_speaker_diarization(...e)}get sampling_rate(){return this.feature_extractor.config.sampling_rate}}},"./src/models/qwen2_vl/image_processing_qwen2_vl.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Qwen2VLImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js"),o=s("./src/utils/tensor.js");class n extends r.ImageProcessor{async _call(e,...t){const{pixel_values:s,original_sizes:r,reshaped_input_sizes:n}=await super._call(e,...t);let i=s;const{temporal_patch_size:a,merge_size:l,patch_size:c}=this.config;1===i.dims[0]&&(i=(0,o.cat)(Array.from({length:a},(()=>i)),0));const d=i.dims[0]/a,u=i.dims[1],_=Math.floor(i.dims[2]/c),p=Math.floor(i.dims[3]/c);return{pixel_values:i.view(d,a,u,Math.floor(_/l),l,c,Math.floor(p/l),l,c).permute(0,3,6,4,7,2,1,5,8).view(d*_*p,u*a*c*c),image_grid_thw:new o.Tensor("int64",[d,_,p],[1,3]),original_sizes:r,reshaped_input_sizes:n}}}},"./src/models/qwen2_vl/processing_qwen2_vl.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Qwen2VLProcessor:()=>i});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js");s("./src/utils/image.js");class i extends r.Processor{static image_processor_class=o.AutoImageProcessor;static tokenizer_class=n.AutoTokenizer;async _call(e,t=null,...s){let r,o;if(Array.isArray(e)||(e=[e]),t&&(r=await this.image_processor(t),o=r.image_grid_thw),o){let t=this.image_processor.config.merge_size**2,s=0;const r=o.tolist();e=e.map((e=>{for(;e.includes("<|image_pad|>");){const o=Number(r[s++].reduce(((e,t)=>e*t),1n));e=e.replace("<|image_pad|>","<|placeholder|>".repeat(Math.floor(o/t)))}return e.replaceAll("<|placeholder|>","<|image_pad|>")}))}return{...this.tokenizer(e),...r}}}},"./src/models/rt_detr/image_processing_rt_detr.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{RTDetrImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{post_process_object_detection(...e){return(0,r.post_process_object_detection)(...e)}}},"./src/models/sam/image_processing_sam.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SamImageProcessor:()=>i});var r=s("./src/base/image_processors_utils.js"),o=s("./src/utils/core.js"),n=s("./src/utils/tensor.js");class i extends r.ImageProcessor{reshape_input_points(e,t,s,r=!1){e=structuredClone(e);let i=(0,o.calculateDimensions)(e);if(3===i.length)r||(i=[1,...i]),e=[e];else if(4!==i.length)throw Error("The input_points must be a 4D tensor of shape `batch_size`, `point_batch_size`, `nb_points_per_image`, `2`.");for(let r=0;r<e.length;++r){let o=t[r],n=s[r],i=[n[0]/o[0],n[1]/o[1]];for(let t=0;t<e[r].length;++t)for(let s=0;s<e[r][t].length;++s)for(let o=0;o<e[r][t][s].length;++o)e[r][t][s][o]*=i[o%2]}return new n.Tensor("float32",Float32Array.from(e.flat(1/0)),i)}add_input_labels(e,t){let s=(0,o.calculateDimensions)(e);if(2===s.length)s=[1,...s],e=[e];else if(3!==s.length)throw Error("The input_points must be a 4D tensor of shape `batch_size`, `point_batch_size`, `nb_points_per_image`, `2`.");if(s.some(((e,s)=>e!==t.dims[s])))throw Error(`The first ${s.length} dimensions of 'input_points' and 'input_labels' must be the same.`);return new n.Tensor("int64",e.flat(1/0).map(BigInt),s)}async _call(e,{input_points:t=null,input_labels:s=null,input_boxes:r=null}={}){const o=await super._call(e);if(t&&(o.input_points=this.reshape_input_points(t,o.original_sizes,o.reshaped_input_sizes)),s){if(!o.input_points)throw Error("`input_points` must be provided if `input_labels` are provided.");o.input_labels=this.add_input_labels(s,o.input_points)}return r&&(o.input_boxes=this.reshape_input_points(r,o.original_sizes,o.reshaped_input_sizes,!0)),o}async post_process_masks(e,t,s,{mask_threshold:r=0,binarize:o=!0,pad_size:i=null}={}){const a=[],l=[(i=i??this.pad_size).height,i.width];for(let i=0;i<t.length;++i){const c=t[i],d=s[i];let u=await(0,n.interpolate_4d)(e[i],{mode:"bilinear",size:l});if(u=u.slice(null,null,[0,d[0]],[0,d[1]]),u=await(0,n.interpolate_4d)(u,{mode:"bilinear",size:c}),o){const e=u.data,t=new Uint8Array(e.length);for(let s=0;s<e.length;++s)e[s]>r&&(t[s]=1);u=new n.Tensor("bool",t,u.dims)}a.push(u)}return a}generate_crop_boxes(e,t,{crop_n_layers:s=0,overlap_ratio:r=512/1500,points_per_crop:o=32,crop_n_points_downscale_factor:n=1}={}){}}},"./src/models/sam/processing_sam.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SamProcessor:()=>n});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js");class n extends r.Processor{static image_processor_class=o.AutoImageProcessor;async _call(...e){return await this.image_processor(...e)}post_process_masks(...e){return this.image_processor.post_process_masks(...e)}reshape_input_points(...e){return this.image_processor.reshape_input_points(...e)}}},"./src/models/seamless_m4t/feature_extraction_seamless_m4t.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SeamlessM4TFeatureExtractor:()=>i});var r=s("./src/base/feature_extraction_utils.js"),o=s("./src/utils/tensor.js"),n=s("./src/utils/audio.js");class i extends r.FeatureExtractor{constructor(e){super(e);const t=this.config.sampling_rate,s=(0,n.mel_filter_bank)(257,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);this.mel_filters=s,this.window=(0,n.window_function)(400,"povey",{periodic:!1})}async _extract_fbank_features(e,t){return e=e.map((e=>32768*e)),(0,n.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,max_num_frames:t,transpose:!0})}async _call(e,{padding:t=!0,pad_to_multiple_of:s=2,do_normalize_per_mel_bins:n=!0,return_attention_mask:i=!0}={}){(0,r.validate_audio_inputs)(e,"SeamlessM4TFeatureExtractor");let a,l=await this._extract_fbank_features(e,this.config.max_length);if(n){const[e,t]=l.dims,s=l.data;for(let r=0;r<t;++r){let o=0;for(let n=0;n<e;++n)o+=s[n*t+r];const n=o/e;let i=0;for(let o=0;o<e;++o)i+=(s[o*t+r]-n)**2;i/=e-1;const a=Math.sqrt(i+1e-7);for(let o=0;o<e;++o){const e=o*t+r;s[e]=(s[e]-n)/a}}}if(t){const[e,t]=l.dims,r=l.data,n=e%s;if(n>0){const s=new Float32Array(t*(e+n));s.set(r),s.fill(this.config.padding_value,r.length);const c=e+n;l=new o.Tensor(l.type,s,[c,t]),i&&(a=new o.Tensor("int64",new BigInt64Array(c),[1,c]),a.data.fill(1n,0,e))}}const[c,d]=l.dims,u=this.config.stride;if(0!==c%u)throw new Error(`The number of frames (${c}) must be a multiple of the stride (${u}).`);const _=l.view(1,Math.floor(c/u),d*u),p={input_features:_};if(i){const e=_.dims[1],t=new BigInt64Array(e);if(a){const e=a.data;for(let s=1,r=0;s<c;s+=u,++r)t[r]=e[s]}else t.fill(1n);p.attention_mask=new o.Tensor("int64",t,[1,e])}return p}}},"./src/models/segformer/image_processing_segformer.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SegformerFeatureExtractor:()=>n,SegformerImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{post_process_semantic_segmentation(...e){return(0,r.post_process_semantic_segmentation)(...e)}}class n extends o{}},"./src/models/siglip/image_processing_siglip.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SiglipImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}},"./src/models/smolvlm/image_processing_smolvlm.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SmolVLMImageProcessor:()=>r.Idefics3ImageProcessor});var r=s("./src/models/idefics3/image_processing_idefics3.js")},"./src/models/smolvlm/processing_smolvlm.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SmolVLMProcessor:()=>r.Idefics3Processor});var r=s("./src/models/idefics3/processing_idefics3.js")},"./src/models/snac/feature_extraction_snac.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SnacFeatureExtractor:()=>o});var r=s("./src/models/dac/feature_extraction_dac.js");class o extends r.DacFeatureExtractor{}},"./src/models/speecht5/feature_extraction_speecht5.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SpeechT5FeatureExtractor:()=>o});var r=s("./src/base/feature_extraction_utils.js");class o extends r.FeatureExtractor{}},"./src/models/speecht5/processing_speecht5.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SpeechT5Processor:()=>i});var r=s("./src/base/processing_utils.js"),o=s("./src/tokenizers.js"),n=s("./src/models/auto/feature_extraction_auto.js");class i extends r.Processor{static tokenizer_class=o.AutoTokenizer;static feature_extractor_class=n.AutoFeatureExtractor;async _call(e){return await this.feature_extractor(e)}}},"./src/models/swin2sr/image_processing_swin2sr.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Swin2SRImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{pad_image(e,t,s,r={}){const[o,n,i]=t;return super.pad_image(e,t,{width:n+(s-n%s)%s,height:o+(s-o%s)%s},{mode:"symmetric",center:!1,constant_values:-1,...r})}}},"./src/models/ultravox/processing_ultravox.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{UltravoxProcessor:()=>i});var r=s("./src/models/auto/feature_extraction_auto.js"),o=s("./src/tokenizers.js"),n=s("./src/base/processing_utils.js");class i extends n.Processor{static tokenizer_class=o.AutoTokenizer;static feature_extractor_class=r.AutoFeatureExtractor;static uses_processor_config=!0;async _call(e,t=null,s={}){if(Array.isArray(e))throw new Error("Batched inputs are not supported yet.");let r={};if(t){const o=t.length,{input_features:n}=await this.feature_extractor(t,{...s,max_length:o}),i=Math.round(o/this.config.encoder_ds_factor+1e-4),a=1+Math.ceil(i/this.config.stack_factor);r.audio_token_len=[a],r.audio_values=n;const l=this.config.audio_placeholder;if(!e.includes(l))throw new Error(`The input text does not contain the image token ${l}.`);e=e.replaceAll(l,l.repeat(a))}return{...this.tokenizer(e,{add_special_tokens:!1,...s}),...r}}}},"./src/models/vit/image_processing_vit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ViTFeatureExtractor:()=>n,ViTImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}class n extends o{}},"./src/models/vitmatte/image_processing_vitmatte.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{VitMatteImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js"),o=s("./src/utils/tensor.js");class n extends r.ImageProcessor{async _call(e,t){Array.isArray(e)||(e=[e]),Array.isArray(t)||(t=[t]);const s=await Promise.all(e.map((e=>this.preprocess(e)))),r=await Promise.all(t.map((e=>this.preprocess(e,{do_normalize:!1,do_convert_rgb:!1,do_convert_grayscale:!0}))));return{pixel_values:(0,o.stack)(s.map(((e,t)=>(0,o.cat)([e.pixel_values,r[t].pixel_values],0))),0),original_sizes:s.map((e=>e.original_size)),reshaped_input_sizes:s.map((e=>e.reshaped_input_size))}}}},"./src/models/vitpose/image_processing_vitpose.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{VitPoseImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{post_process_pose_estimation(e,t,{threshold:s=null}={}){const r=e.tolist(),[o,n,i,a]=e.dims,l=[];for(let e=0;e<o;++e){const o=r[e],n=t[e],c=[];for(let e=0;e<n.length;++e){const t=n[e],r=[],l=[],d=[],u=t.at(-2)/a,_=t.at(-1)/i;for(let e=0;e<o.length;++e){let[t,n]=[0,0],i=0,a=-1/0;const c=o[e];for(let e=0;e<c.length;++e){const s=c[e];for(let r=0;r<s.length;++r){const o=s[r];i+=o,a=Math.max(a,o),t+=(r+.5)*o,n+=e*o}}if(null!=s&&a<s)continue;const p=[u*t/i,_*n/i];r.push(p),d.push(e),l.push(a)}c.push({bbox:t,scores:l,labels:d,keypoints:r})}l.push(c)}return l}}},"./src/models/wav2vec2/feature_extraction_wav2vec2.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Wav2Vec2FeatureExtractor:()=>n});var r=s("./src/base/feature_extraction_utils.js"),o=s("./src/utils/tensor.js");class n extends r.FeatureExtractor{_zero_mean_unit_var_norm(e){const t=e.reduce(((e,t)=>e+t),0)/e.length,s=e.reduce(((e,s)=>e+(s-t)**2),0)/e.length;return e.map((e=>(e-t)/Math.sqrt(s+1e-7)))}async _call(e){(0,r.validate_audio_inputs)(e,"Wav2Vec2FeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));let t=e;this.config.do_normalize&&(t=this._zero_mean_unit_var_norm(t));const s=[1,t.length];return{input_values:new o.Tensor("float32",t,s),attention_mask:new o.Tensor("int64",new BigInt64Array(t.length).fill(1n),s)}}}},"./src/models/wav2vec2/processing_wav2vec2.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Wav2Vec2Processor:()=>i});var r=s("./src/tokenizers.js"),o=s("./src/models/auto/feature_extraction_auto.js"),n=s("./src/base/processing_utils.js");class i extends n.Processor{static tokenizer_class=r.AutoTokenizer;static feature_extractor_class=o.AutoFeatureExtractor;async _call(e){return await this.feature_extractor(e)}}},"./src/models/wav2vec2_with_lm/processing_wav2vec2_with_lm.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Wav2Vec2ProcessorWithLM:()=>i});var r=s("./src/tokenizers.js"),o=s("./src/models/auto/feature_extraction_auto.js"),n=s("./src/base/processing_utils.js");class i extends n.Processor{static tokenizer_class=r.AutoTokenizer;static feature_extractor_class=o.AutoFeatureExtractor;async _call(e){return await this.feature_extractor(e)}}},"./src/models/wespeaker/feature_extraction_wespeaker.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{WeSpeakerFeatureExtractor:()=>n});var r=s("./src/base/feature_extraction_utils.js"),o=(s("./src/utils/tensor.js"),s("./src/utils/audio.js"));class n extends r.FeatureExtractor{constructor(e){super(e);const t=this.config.sampling_rate,s=(0,o.mel_filter_bank)(257,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);this.mel_filters=s,this.window=(0,o.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,o.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,transpose:!0,min_num_frames:this.min_num_frames})}async _call(e){(0,r.validate_audio_inputs)(e,"WeSpeakerFeatureExtractor");const t=(await this._extract_fbank_features(e)).unsqueeze_(0);if(null===this.config.fbank_centering_span){const e=t.mean(1).data,s=t.data,[r,o,n]=t.dims;for(let t=0;t<r;++t){const r=t*o*n,i=t*n;for(let t=0;t<o;++t){const o=r+t*n;for(let t=0;t<n;++t)s[o+t]-=e[i+t]}}}return{input_features:t}}}},"./src/models/whisper/common_whisper.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{WHISPER_LANGUAGE_MAPPING:()=>o,WHISPER_TO_LANGUAGE_CODE_MAPPING:()=>n,whisper_language_to_code:()=>i});const r=[["en","english"],["zh","chinese"],["de","german"],["es","spanish"],["ru","russian"],["ko","korean"],["fr","french"],["ja","japanese"],["pt","portuguese"],["tr","turkish"],["pl","polish"],["ca","catalan"],["nl","dutch"],["ar","arabic"],["sv","swedish"],["it","italian"],["id","indonesian"],["hi","hindi"],["fi","finnish"],["vi","vietnamese"],["he","hebrew"],["uk","ukrainian"],["el","greek"],["ms","malay"],["cs","czech"],["ro","romanian"],["da","danish"],["hu","hungarian"],["ta","tamil"],["no","norwegian"],["th","thai"],["ur","urdu"],["hr","croatian"],["bg","bulgarian"],["lt","lithuanian"],["la","latin"],["mi","maori"],["ml","malayalam"],["cy","welsh"],["sk","slovak"],["te","telugu"],["fa","persian"],["lv","latvian"],["bn","bengali"],["sr","serbian"],["az","azerbaijani"],["sl","slovenian"],["kn","kannada"],["et","estonian"],["mk","macedonian"],["br","breton"],["eu","basque"],["is","icelandic"],["hy","armenian"],["ne","nepali"],["mn","mongolian"],["bs","bosnian"],["kk","kazakh"],["sq","albanian"],["sw","swahili"],["gl","galician"],["mr","marathi"],["pa","punjabi"],["si","sinhala"],["km","khmer"],["sn","shona"],["yo","yoruba"],["so","somali"],["af","afrikaans"],["oc","occitan"],["ka","georgian"],["be","belarusian"],["tg","tajik"],["sd","sindhi"],["gu","gujarati"],["am","amharic"],["yi","yiddish"],["lo","lao"],["uz","uzbek"],["fo","faroese"],["ht","haitian creole"],["ps","pashto"],["tk","turkmen"],["nn","nynorsk"],["mt","maltese"],["sa","sanskrit"],["lb","luxembourgish"],["my","myanmar"],["bo","tibetan"],["tl","tagalog"],["mg","malagasy"],["as","assamese"],["tt","tatar"],["haw","hawaiian"],["ln","lingala"],["ha","hausa"],["ba","bashkir"],["jw","javanese"],["su","sundanese"]],o=new Map(r),n=new Map([...r.map((([e,t])=>[t,e])),["burmese","my"],["valencian","ca"],["flemish","nl"],["haitian","ht"],["letzeburgesch","lb"],["pushto","ps"],["panjabi","pa"],["moldavian","ro"],["moldovan","ro"],["sinhalese","si"],["castilian","es"]]);function i(e){e=e.toLowerCase();let t=n.get(e);if(void 0===t){const s=e.match(/^<\|([a-z]{2})\|>$/);if(s&&(e=s[1]),!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(Array.from(t))}`)}t=e}return t}},"./src/models/whisper/feature_extraction_whisper.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{WhisperFeatureExtractor:()=>i});var r=s("./src/base/feature_extraction_utils.js"),o=(s("./src/utils/tensor.js"),s("./src/utils/audio.js")),n=s("./src/utils/maths.js");class i extends r.FeatureExtractor{constructor(e){super(e),this.config.mel_filters??=(0,o.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,o.window_function)(this.config.n_fft,"hann")}async _extract_fbank_features(e){const t=await(0,o.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:Math.min(Math.floor(e.length/this.config.hop_length),this.config.nb_max_frames)}),s=t.data,r=(0,n.max)(s)[0];for(let e=0;e<s.length;++e)s[e]=(Math.max(s[e],r-8)+4)/4;return t}async _call(e,{max_length:t=null}={}){let s;(0,r.validate_audio_inputs)(e,"WhisperFeatureExtractor");const o=t??this.config.n_samples;e.length>o?(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`."),s=e.slice(0,o)):(s=new Float32Array(o),s.set(e));return{input_features:(await this._extract_fbank_features(s)).unsqueeze_(0)}}}},"./src/models/whisper/generation_whisper.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{WhisperGenerationConfig:()=>o});var r=s("./src/generation/configuration_utils.js");class o extends r.GenerationConfig{return_timestamps=null;return_token_timestamps=null;num_frames=null;alignment_heads=null;task=null;language=null;no_timestamps_token_id=null;prompt_ids=null;is_multilingual=null;lang_to_id=null;task_to_id=null;max_initial_timestamp_index=1}},"./src/models/whisper/processing_whisper.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{WhisperProcessor:()=>i});var r=s("./src/models/auto/feature_extraction_auto.js"),o=s("./src/tokenizers.js"),n=s("./src/base/processing_utils.js");class i extends n.Processor{static tokenizer_class=o.AutoTokenizer;static feature_extractor_class=r.AutoFeatureExtractor;async _call(e){return await this.feature_extractor(e)}}},"./src/models/yolos/image_processing_yolos.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{YolosFeatureExtractor:()=>n,YolosImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{post_process_object_detection(...e){return(0,r.post_process_object_detection)(...e)}}class n extends o{}},"./src/ops/registry.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{TensorOpRegistry:()=>l});var r=s("./src/backends/onnx.js"),o=s("./src/utils/tensor.js"),n=s("./src/env.js");const i=n.apis.IS_BROWSER_ENV||n.apis.IS_WEBWORKER_ENV,a=async(e,t,s)=>{const n=await(0,r.createInferenceSession)(new Uint8Array(e),t);let a=Promise.resolve();return async e=>{const t=(0,r.isONNXProxy)(),l=Object.fromEntries(Object.entries(e).map((([e,s])=>[e,(t?s.clone():s).ort_tensor]))),c=await(a=i?a.then((()=>n.run(l))):n.run(l));return Array.isArray(s)?s.map((e=>new o.Tensor(c[e]))):new o.Tensor(c[s])}};class l{static session_options={};static get nearest_interpolate_4d(){return this._nearest_interpolate_4d||(this._nearest_interpolate_4d=a([8,10,18,0,58,129,1,10,41,10,1,120,10,0,10,0,10,1,115,18,1,121,34,6,82,101,115,105,122,101,42,18,10,4,109,111,100,101,34,7,110,101,97,114,101,115,116,160,1,3,18,1,114,90,31,10,1,120,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,99,10,3,18,1,104,10,3,18,1,119,90,15,10,1,115,18,10,10,8,8,7,18,4,10,2,8,4,98,31,10,1,121,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,99,10,3,18,1,104,10,3,18,1,119,66,2,16,21],this.session_options,"y")),this._nearest_interpolate_4d}static get bilinear_interpolate_4d(){return this._bilinear_interpolate_4d||(this._bilinear_interpolate_4d=a([8,9,18,0,58,128,1,10,40,10,1,120,10,0,10,0,10,1,115,18,1,121,34,6,82,101,115,105,122,101,42,17,10,4,109,111,100,101,34,6,108,105,110,101,97,114,160,1,3,18,1,114,90,31,10,1,120,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,99,10,3,18,1,104,10,3,18,1,119,90,15,10,1,115,18,10,10,8,8,7,18,4,10,2,8,4,98,31,10,1,121,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,99,10,3,18,1,104,10,3,18,1,119,66,2,16,20],this.session_options,"y")),this._bilinear_interpolate_4d}static get bicubic_interpolate_4d(){return this._bicubic_interpolate_4d||(this._bicubic_interpolate_4d=a([8,9,18,0,58,127,10,39,10,1,120,10,0,10,0,10,1,115,18,1,121,34,6,82,101,115,105,122,101,42,16,10,4,109,111,100,101,34,5,99,117,98,105,99,160,1,3,18,1,114,90,31,10,1,120,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,99,10,3,18,1,104,10,3,18,1,119,90,15,10,1,115,18,10,10,8,8,7,18,4,10,2,8,4,98,31,10,1,121,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,99,10,3,18,1,104,10,3,18,1,119,66,2,16,20],this.session_options,"y")),this._bicubic_interpolate_4d}static get matmul(){return this._matmul||(this._matmul=a([8,9,18,0,58,55,10,17,10,1,97,10,1,98,18,1,99,34,6,77,97,116,77,117,108,18,1,114,90,9,10,1,97,18,4,10,2,8,1,90,9,10,1,98,18,4,10,2,8,1,98,9,10,1,99,18,4,10,2,8,1,66,2,16,20],this.session_options,"c")),this._matmul}static get stft(){return this._stft||(this._stft=a([8,7,18,0,58,148,1,10,38,10,1,115,10,1,106,10,1,119,10,1,108,18,1,111,34,4,83,84,70,84,42,15,10,8,111,110,101,115,105,100,101,100,24,1,160,1,2,18,1,115,90,26,10,1,115,18,21,10,19,8,1,18,15,10,3,18,1,98,10,3,18,1,115,10,3,18,1,99,90,11,10,1,106,18,6,10,4,8,7,18,0,90,16,10,1,119,18,11,10,9,8,1,18,5,10,3,18,1,119,90,11,10,1,108,18,6,10,4,8,7,18,0,98,31,10,1,111,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,102,10,3,18,1,100,10,3,18,1,99,66,2,16,17],this.session_options,"o")),this._stft}static get rfft(){return this._rfft||(this._rfft=a([8,9,18,0,58,97,10,33,10,1,120,10,0,10,1,97,18,1,121,34,3,68,70,84,42,15,10,8,111,110,101,115,105,100,101,100,24,1,160,1,2,18,1,100,90,21,10,1,120,18,16,10,14,8,1,18,10,10,3,18,1,115,10,3,18,1,99,90,11,10,1,97,18,6,10,4,8,7,18,0,98,21,10,1,121,18,16,10,14,8,1,18,10,10,3,18,1,115,10,3,18,1,99,66,2,16,20],this.session_options,"y")),this._rfft}static get top_k(){return this._top_k||(this._top_k=a([8,10,18,0,58,73,10,18,10,1,120,10,1,107,18,1,118,18,1,105,34,4,84,111,112,75,18,1,116,90,9,10,1,120,18,4,10,2,8,1,90,15,10,1,107,18,10,10,8,8,7,18,4,10,2,8,1,98,9,10,1,118,18,4,10,2,8,1,98,9,10,1,105,18,4,10,2,8,7,66,2,16,21],this.session_options,["v","i"])),this._top_k}static get slice(){return this._slice||(this._slice=a([8,7,18,0,58,96,10,25,10,1,120,10,1,115,10,1,101,10,1,97,10,1,116,18,1,121,34,5,83,108,105,99,101,18,1,114,90,9,10,1,120,18,4,10,2,8,1,90,9,10,1,115,18,4,10,2,8,7,90,9,10,1,101,18,4,10,2,8,7,90,9,10,1,97,18,4,10,2,8,7,90,9,10,1,116,18,4,10,2,8,7,98,9,10,1,121,18,4,10,2,8,1,66,2,16,13],this.session_options,"y")),this._slice}}},"./src/pipelines.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{AudioClassificationPipeline:()=>C,AutomaticSpeechRecognitionPipeline:()=>A,BackgroundRemovalPipeline:()=>z,DepthEstimationPipeline:()=>R,DocumentQuestionAnsweringPipeline:()=>V,FeatureExtractionPipeline:()=>P,FillMaskPipeline:()=>M,ImageClassificationPipeline:()=>L,ImageFeatureExtractionPipeline:()=>F,ImageSegmentationPipeline:()=>I,ImageToImagePipeline:()=>B,ImageToTextPipeline:()=>E,ObjectDetectionPipeline:()=>D,Pipeline:()=>h,QuestionAnsweringPipeline:()=>w,SummarizationPipeline:()=>b,Text2TextGenerationPipeline:()=>x,TextClassificationPipeline:()=>g,TextGenerationPipeline:()=>v,TextToAudioPipeline:()=>N,TokenClassificationPipeline:()=>f,TranslationPipeline:()=>k,ZeroShotAudioClassificationPipeline:()=>S,ZeroShotClassificationPipeline:()=>T,ZeroShotImageClassificationPipeline:()=>j,ZeroShotObjectDetectionPipeline:()=>O,pipeline:()=>$});var r=s("./src/tokenizers.js"),o=s("./src/models.js"),n=s("./src/models/auto/processing_auto.js"),i=(s("./src/base/processing_utils.js"),s("./src/utils/generic.js")),a=s("./src/utils/core.js"),l=s("./src/utils/maths.js"),c=s("./src/utils/audio.js"),d=s("./src/utils/tensor.js"),u=s("./src/utils/image.js");async function _(e){return Array.isArray(e)||(e=[e]),await Promise.all(e.map((e=>u.RawImage.read(e))))}async function p(e,t){return Array.isArray(e)||(e=[e]),await Promise.all(e.map((e=>"string"==typeof e||e instanceof URL?(0,c.read_audio)(e,t):e instanceof Float64Array?new Float32Array(e):e)))}function m(e,t){t&&(e=e.map((e=>0|e)));const[s,r,o,n]=e;return{xmin:s,ymin:r,xmax:o,ymax:n}}class h extends i.Callable{constructor({task:e,model:t,tokenizer:s=null,processor:r=null}){super(),this.task=e,this.model=t,this.tokenizer=s,this.processor=r}async dispose(){await this.model.dispose()}}class g extends h{constructor(e){super(e)}async _call(e,{top_k:t=1}={}){const s=this.tokenizer(e,{padding:!0,truncation:!0}),r=await this.model(s),o="multi_label_classification"===this.model.config.problem_type?e=>e.sigmoid():e=>new d.Tensor("float32",(0,l.softmax)(e.data),e.dims),n=this.model.config.id2label,i=[];for(const e of r.logits){const s=o(e),r=await(0,d.topk)(s,t),a=r[0].tolist(),l=r[1].tolist().map(((e,t)=>({label:n?n[e]:`LABEL_${e}`,score:a[t]})));1===t?i.push(...l):i.push(l)}return Array.isArray(e)||1===t?i:i[0]}}class f extends h{constructor(e){super(e)}async _call(e,{ignore_labels:t=["O"]}={}){const s=Array.isArray(e),r=this.tokenizer(s?e:[e],{padding:!0,truncation:!0}),o=(await this.model(r)).logits,n=this.model.config.id2label,i=[];for(let e=0;e<o.dims[0];++e){const s=r.input_ids[e],a=o[e],c=[];for(let e=0;e<a.dims[0];++e){const r=a[e],o=(0,l.max)(r.data)[1],i=n?n[o]:`LABEL_${o}`;if(t.includes(i))continue;const d=this.tokenizer.decode([s[e].item()],{skip_special_tokens:!0});if(""===d)continue;const u=(0,l.softmax)(r.data);c.push({entity:i,score:u[o],index:e,word:d})}i.push(c)}return s?i:i[0]}}class w extends h{constructor(e){super(e)}async _call(e,t,{top_k:s=1}={}){const r=this.tokenizer(e,{text_pair:t,padding:!0,truncation:!0}),{start_logits:o,end_logits:n}=await this.model(r),i=r.input_ids.tolist(),c=r.attention_mask.tolist(),d=this.tokenizer.all_special_ids,u=[];for(let e=0;e<o.dims[0];++e){const t=i[e],r=t.findIndex((e=>e==this.tokenizer.sep_token_id)),_=(c[e].map(((e,s)=>1==e&&(0===s||s>r&&-1===d.findIndex((e=>e==t[s]))))),o[e].tolist()),p=n[e].tolist();for(let s=1;s<_.length;++s)(0==c[e]||s<=r||-1!==d.findIndex((e=>e==t[s])))&&(_[s]=-1/0,p[s]=-1/0);const m=(0,l.softmax)(_).map(((e,t)=>[e,t])),h=(0,l.softmax)(p).map(((e,t)=>[e,t]));m[0][0]=0,h[0][0]=0;const g=(0,a.product)(m,h).filter((e=>e[0][1]<=e[1][1])).map((e=>[e[0][1],e[1][1],e[0][0]*e[1][0]])).sort(((e,t)=>t[2]-e[2]));for(let e=0;e<Math.min(g.length,s);++e){const[s,r,o]=g[e],n=t.slice(s,r+1),i=this.tokenizer.decode(n,{skip_special_tokens:!0});u.push({answer:i,score:o})}}return 1===s?u[0]:u}}class M extends h{constructor(e){super(e)}async _call(e,{top_k:t=5}={}){const s=this.tokenizer(e,{padding:!0,truncation:!0}),{logits:r}=await this.model(s),o=[],n=s.input_ids.tolist();for(let e=0;e<n.length;++e){const s=n[e],i=s.findIndex((e=>e==this.tokenizer.mask_token_id));if(-1===i)throw Error(`Mask token (${this.tokenizer.mask_token}) not found in text.`);const a=r[e][i],c=await(0,d.topk)(new d.Tensor("float32",(0,l.softmax)(a.data),a.dims),t),u=c[0].tolist(),_=c[1].tolist();o.push(_.map(((e,t)=>{const r=s.slice();return r[i]=e,{score:u[t],token:Number(e),token_str:this.tokenizer.decode([e]),sequence:this.tokenizer.decode(r,{skip_special_tokens:!0})}})))}return Array.isArray(e)?o:o[0]}}class x extends h{_key="generated_text";constructor(e){super(e)}async _call(e,t={}){Array.isArray(e)||(e=[e]),this.model.config.prefix&&(e=e.map((e=>this.model.config.prefix+e)));const s=this.model.config.task_specific_params;s&&s[this.task]&&s[this.task].prefix&&(e=e.map((e=>s[this.task].prefix+e)));const r=this.tokenizer,o={padding:!0,truncation:!0};let n;n=this instanceof k&&"_build_translation_inputs"in r?r._build_translation_inputs(e,o,t):r(e,o);const i=await this.model.generate({...n,...t});return r.batch_decode(i,{skip_special_tokens:!0}).map((e=>({[this._key]:e})))}}class b extends x{_key="summary_text";constructor(e){super(e)}}class k extends x{_key="translation_text";constructor(e){super(e)}}function y(e){return Array.isArray(e)&&e.every((e=>"role"in e&&"content"in e))}class v extends h{constructor(e){super(e)}async _call(e,t={}){let s,r=!1,o=!1;if("string"==typeof e)s=e=[e];else if(Array.isArray(e)&&e.every((e=>"string"==typeof e)))r=!0,s=e;else{if(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");r=!0}o=!0,s=e.map((e=>this.tokenizer.apply_chat_template(e,{tokenize:!1,add_generation_prompt:!0})))}const n=t.add_special_tokens??!1,i=!o&&(t.return_full_text??!0);this.tokenizer.padding_side="left";const a=this.tokenizer(s,{add_special_tokens:n,padding:!0,truncation:!0}),l=await this.model.generate({...a,...t}),c=this.tokenizer.batch_decode(l,{skip_special_tokens:!0});let d;!i&&a.input_ids.dims.at(-1)>0&&(d=this.tokenizer.batch_decode(a.input_ids,{skip_special_tokens:!0}).map((e=>e.length)));const u=Array.from({length:e.length},(e=>[]));for(let t=0;t<c.length;++t){const s=Math.floor(t/l.dims[0]*e.length);d&&(c[t]=c[t].slice(d[s])),u[s].push({generated_text:o?[...e[s],{role:"assistant",content:c[t]}]:c[t]})}return r||1!==u.length?u:u[0]}}class T extends h{constructor(e){super(e),this.label2id=Object.fromEntries(Object.entries(this.model.config.label2id).map((([e,t])=>[e.toLowerCase(),t]))),this.entailment_id=this.label2id.entailment,void 0===this.entailment_id&&(console.warn("Could not find 'entailment' in label2id mapping. Using 2 as entailment_id."),this.entailment_id=2),this.contradiction_id=this.label2id.contradiction??this.label2id.not_entailment,void 0===this.contradiction_id&&(console.warn("Could not find 'contradiction' in label2id mapping. Using 0 as contradiction_id."),this.contradiction_id=0)}async _call(e,t,{hypothesis_template:s="This example is {}.",multi_label:r=!1}={}){const o=Array.isArray(e);o||(e=[e]),Array.isArray(t)||(t=[t]);const n=t.map((e=>s.replace("{}",e))),i=r||1===t.length,a=[];for(const s of e){const e=[];for(const t of n){const r=this.tokenizer(s,{text_pair:t,padding:!0,truncation:!0}),o=await this.model(r);i?e.push([o.logits.data[this.contradiction_id],o.logits.data[this.entailment_id]]):e.push(o.logits.data[this.entailment_id])}const r=(i?e.map((e=>(0,l.softmax)(e)[1])):(0,l.softmax)(e)).map(((e,t)=>[e,t])).sort(((e,t)=>t[0]-e[0]));a.push({sequence:s,labels:r.map((e=>t[e[1]])),scores:r.map((e=>e[0]))})}return o?a:a[0]}}class P extends h{constructor(e){super(e)}async _call(e,{pooling:t="none",normalize:s=!1,quantize:r=!1,precision:o="binary"}={}){const n=this.tokenizer(e,{padding:!0,truncation:!0}),i=await this.model(n);let a=i.last_hidden_state??i.logits??i.token_embeddings;if("none"===t);else if("mean"===t)a=(0,d.mean_pooling)(a,n.attention_mask);else{if("cls"!==t)throw Error(`Pooling method '${t}' not supported.`);a=a.slice(null,0)}return s&&(a=a.normalize(2,-1)),r&&(a=(0,d.quantize_embeddings)(a,o)),a}}class F extends h{constructor(e){super(e)}async _call(e,{pool:t=null}={}){const s=await _(e),{pixel_values:r}=await this.processor(s),o=await this.model({pixel_values:r});let n;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.");n=o.pooler_output}else n=o.last_hidden_state??o.logits??o.image_embeds;return n}}class C extends h{constructor(e){super(e)}async _call(e,{top_k:t=5}={}){const s=this.processor.feature_extractor.config.sampling_rate,r=await p(e,s),o=this.model.config.id2label,n=[];for(const e of r){const s=await this.processor(e),r=(await this.model(s)).logits[0],i=await(0,d.topk)(new d.Tensor("float32",(0,l.softmax)(r.data),r.dims),t),a=i[0].tolist(),c=i[1].tolist().map(((e,t)=>({label:o?o[e]:`LABEL_${e}`,score:a[t]})));n.push(c)}return Array.isArray(e)?n:n[0]}}class S extends h{constructor(e){super(e)}async _call(e,t,{hypothesis_template:s="This is a sound of {}."}={}){const r=!Array.isArray(e);r&&(e=[e]);const o=t.map((e=>s.replace("{}",e))),n=this.tokenizer(o,{padding:!0,truncation:!0}),i=this.processor.feature_extractor.config.sampling_rate,a=await p(e,i),c=[];for(const e of a){const s=await this.processor(e),r=await this.model({...n,...s}),o=(0,l.softmax)(r.logits_per_audio.data);c.push([...o].map(((e,s)=>({score:e,label:t[s]}))))}return r?c[0]:c}}class A extends h{constructor(e){super(e)}async _call(e,t={}){switch(this.model.config.model_type){case"whisper":case"lite-whisper":return this._call_whisper(e,t);case"wav2vec2":case"wav2vec2-bert":case"unispeech":case"unispeech-sat":case"hubert":return this._call_wav2vec2(e,t);case"moonshine":return this._call_moonshine(e,t);default:throw new Error(`AutomaticSpeechRecognitionPipeline does not support model type '${this.model.config.model_type}'.`)}}async _call_wav2vec2(e,t){t.language&&console.warn('`language` parameter is not yet supported for `wav2vec2` models, defaulting to "English".'),t.task&&console.warn('`task` parameter is not yet supported for `wav2vec2` models, defaulting to "transcribe".');const s=!Array.isArray(e);s&&(e=[e]);const r=this.processor.feature_extractor.config.sampling_rate,o=await p(e,r),n=[];for(const e of o){const t=await this.processor(e),s=(await this.model(t)).logits[0],r=[];for(const e of s)r.push((0,l.max)(e.data)[1]);const o=this.tokenizer.decode(r);n.push({text:o})}return s?n[0]:n}async _call_whisper(e,t){const s=t.return_timestamps??!1,r=t.chunk_length_s??0,o=t.force_full_sequences??!1;let n=t.stride_length_s??null;const i={...t};"word"===s&&(i.return_token_timestamps=!0,i.return_timestamps=!1);const a=!Array.isArray(e);a&&(e=[e]);const c=this.processor.feature_extractor.config.chunk_length/this.model.config.max_source_positions,d=this.processor.feature_extractor.config.hop_length,u=this.processor.feature_extractor.config.sampling_rate,_=await p(e,u),m=[];for(const e of _){let t=[];if(r>0){if(null===n)n=r/6;else if(r<=n)throw Error("`chunk_length_s` must be larger than `stride_length_s`.");const s=u*r,o=u*n,i=s-2*o;let a=0;for(;;){const r=a+s,n=e.subarray(a,r),l=await this.processor(n),c=0===a,d=r>=e.length;if(t.push({stride:[n.length,c?0:o,d?0:o],input_features:l.input_features,is_last:d}),d)break;a+=i}}else t=[{stride:[e.length,0,0],input_features:(await this.processor(e)).input_features,is_last:!0}];for(const e of t){i.num_frames=Math.floor(e.stride[0]/d);const t=await this.model.generate({inputs:e.input_features,...i});"word"===s?(e.tokens=t.sequences.tolist()[0],e.token_timestamps=t.token_timestamps.tolist()[0].map((e=>(0,l.round)(e,2)))):e.tokens=t[0].tolist(),e.stride=e.stride.map((e=>e/u))}const[a,_]=this.tokenizer._decode_asr(t,{time_precision:c,return_timestamps:s,force_full_sequences:o});m.push({text:a,..._})}return a?m[0]:m}async _call_moonshine(e,t){const s=!Array.isArray(e);s&&(e=[e]);const r=this.processor.feature_extractor.config.sampling_rate,o=await p(e,r),n=[];for(const e of o){const s=await this.processor(e),o=6*Math.floor(e.length/r),i=await this.model.generate({max_new_tokens:o,...t,...s}),a=this.processor.batch_decode(i,{skip_special_tokens:!0})[0];n.push({text:a})}return s?n[0]:n}}class E extends h{constructor(e){super(e)}async _call(e,t={}){const s=Array.isArray(e),r=await _(e),{pixel_values:o}=await this.processor(r),n=[];for(const e of o){e.dims=[1,...e.dims];const s=await this.model.generate({inputs:e,...t}),r=this.tokenizer.batch_decode(s,{skip_special_tokens:!0}).map((e=>({generated_text:e.trim()})));n.push(r)}return s?n:n[0]}}class L extends h{constructor(e){super(e)}async _call(e,{top_k:t=5}={}){const s=await _(e),{pixel_values:r}=await this.processor(s),o=await this.model({pixel_values:r}),n=this.model.config.id2label,i=[];for(const e of o.logits){const s=await(0,d.topk)(new d.Tensor("float32",(0,l.softmax)(e.data),e.dims),t),r=s[0].tolist(),o=s[1].tolist().map(((e,t)=>({label:n?n[e]:`LABEL_${e}`,score:r[t]})));i.push(o)}return Array.isArray(e)?i:i[0]}}class I extends h{constructor(e){super(e),this.subtasks_mapping={panoptic:"post_process_panoptic_segmentation",instance:"post_process_instance_segmentation",semantic:"post_process_semantic_segmentation"}}async _call(e,{threshold:t=.5,mask_threshold:s=.5,overlap_mask_area_threshold:r=.8,label_ids_to_fuse:o=null,target_sizes:n=null,subtask:i=null}={}){if(Array.isArray(e)&&1!==e.length)throw Error("Image segmentation pipeline currently only supports a batch size of 1.");const a=await _(e),l=a.map((e=>[e.height,e.width])),c=await this.processor(a),{inputNames:d,outputNames:p}=this.model.sessions.model;if(!d.includes("pixel_values")){if(1!==d.length)throw Error(`Expected a single input name, but got ${d.length} inputs: ${d}.`);const e=d[0];if(e in c)throw Error(`Input name ${e} already exists in the inputs.`);c[e]=c.pixel_values}const m=await this.model(c);let h=null;if(null!==i)h=this.subtasks_mapping[i];else if(this.processor.image_processor)for(const[e,t]of Object.entries(this.subtasks_mapping))if(t in this.processor.image_processor){h=this.processor.image_processor[t].bind(this.processor.image_processor),i=e;break}const g=this.model.config.id2label,f=[];if(i)if("panoptic"===i||"instance"===i){const e=h(m,t,s,r,o,n??l)[0],i=e.segmentation;for(const t of e.segments_info){const e=new Uint8ClampedArray(i.data.length);for(let s=0;s<i.data.length;++s)i.data[s]===t.id&&(e[s]=255);const s=new u.RawImage(e,i.dims[1],i.dims[0],1);f.push({score:t.score,label:g[t.label_id],mask:s})}}else{if("semantic"!==i)throw Error(`Subtask ${i} not supported.`);{const{segmentation:e,labels:t}=h(m,n??l)[0];for(const s of t){const t=new Uint8ClampedArray(e.data.length);for(let r=0;r<e.data.length;++r)e.data[r]===s&&(t[r]=255);const r=new u.RawImage(t,e.dims[1],e.dims[0],1);f.push({score:null,label:g[s],mask:r})}}}else{const e=1e-5,t=m[p[0]];for(let s=0;s<l.length;++s){const r=l[s],o=t[s];o.data.some((t=>t<-e||t>1+e))&&o.sigmoid_();const n=await u.RawImage.fromTensor(o.mul_(255).to("uint8")).resize(r[1],r[0]);f.push({label:null,score:null,mask:n})}}return f}}class z extends I{constructor(e){super(e)}async _call(e,t={}){if(Array.isArray(e)&&1!==e.length)throw Error("Background removal pipeline currently only supports a batch size of 1.");const s=await _(e),r=await super._call(e,t);return s.map(((e,t)=>{const s=e.clone();return s.putAlpha(r[t].mask),s}))}}class j extends h{constructor(e){super(e)}async _call(e,t,{hypothesis_template:s="This is a photo of {}"}={}){const r=Array.isArray(e),o=await _(e),n=t.map((e=>s.replace("{}",e))),i=this.tokenizer(n,{padding:"siglip"!==this.model.config.model_type||"max_length",truncation:!0}),{pixel_values:a}=await this.processor(o),c=await this.model({...i,pixel_values:a}),d="siglip"===this.model.config.model_type?e=>e.sigmoid().data:e=>(0,l.softmax)(e.data),u=[];for(const e of c.logits_per_image){const s=[...d(e)].map(((e,s)=>({score:e,label:t[s]})));s.sort(((e,t)=>t.score-e.score)),u.push(s)}return r?u:u[0]}}class D extends h{constructor(e){super(e)}async _call(e,{threshold:t=.9,percentage:s=!1}={}){const r=Array.isArray(e);if(r&&1!==e.length)throw Error("Object detection pipeline currently only supports a batch size of 1.");const o=await _(e),n=s?null:o.map((e=>[e.height,e.width])),{pixel_values:i,pixel_mask:a}=await this.processor(o),l=await this.model({pixel_values:i,pixel_mask:a}),c=this.processor.image_processor.post_process_object_detection(l,t,n),d=this.model.config.id2label,u=c.map((e=>e.boxes.map(((t,r)=>({score:e.scores[r],label:d[e.classes[r]],box:m(t,!s)})))));return r?u:u[0]}}class O extends h{constructor(e){super(e)}async _call(e,t,{threshold:s=.1,top_k:r=null,percentage:o=!1}={}){const n=Array.isArray(e),i=await _(e),a=this.tokenizer(t,{padding:!0,truncation:!0}),l=await this.processor(i),c=[];for(let e=0;e<i.length;++e){const n=i[e],d=o?null:[[n.height,n.width]],u=l.pixel_values[e].unsqueeze_(0),_=await this.model({...a,pixel_values:u});let p;if("post_process_grounded_object_detection"in this.processor){const e=this.processor.post_process_grounded_object_detection(_,a.input_ids,{box_threshold:s,text_threshold:s,target_sizes:d})[0];p=e.boxes.map(((t,s)=>({score:e.scores[s],label:e.labels[s],box:m(t,!o)})))}else{const e=this.processor.image_processor.post_process_object_detection(_,s,d,!0)[0];p=e.boxes.map(((s,r)=>({score:e.scores[r],label:t[e.classes[r]],box:m(s,!o)})))}p.sort(((e,t)=>t.score-e.score)),null!==r&&(p=p.slice(0,r)),c.push(p)}return n?c:c[0]}}class V extends h{constructor(e){super(e)}async _call(e,t,s={}){const r=(await _(e))[0],{pixel_values:o}=await this.processor(r),n=`<s_docvqa><s_question>${t}</s_question><s_answer>`,i=this.tokenizer(n,{add_special_tokens:!1,padding:!0,truncation:!0}).input_ids,a=await this.model.generate({inputs:o,max_length:this.model.config.decoder.max_position_embeddings,decoder_input_ids:i,...s}),l=this.tokenizer.batch_decode(a)[0].match(/<s_answer>(.*?)<\/s_answer>/);let c=null;return l&&l.length>=2&&(c=l[1].trim()),[{answer:c}]}}class N extends h{DEFAULT_VOCODER_ID="Xenova/speecht5_hifigan";constructor(e){super(e),this.vocoder=e.vocoder??null}async _call(e,{speaker_embeddings:t=null}={}){return this.processor?this._call_text_to_spectrogram(e,{speaker_embeddings:t}):this._call_text_to_waveform(e)}async _call_text_to_waveform(e){const t=this.tokenizer(e,{padding:!0,truncation:!0}),{waveform:s}=await this.model(t),r=this.model.config.sampling_rate;return new c.RawAudio(s.data,r)}async _call_text_to_spectrogram(e,{speaker_embeddings:t}){if(this.vocoder||(console.log("No vocoder specified, using default HifiGan vocoder."),this.vocoder=await 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:s}=this.tokenizer(e,{padding:!0,truncation:!0}),{waveform:r}=await this.model.generate_speech(s,t,{vocoder:this.vocoder}),n=this.processor.feature_extractor.config.sampling_rate;return new c.RawAudio(r.data,n)}}class B extends h{constructor(e){super(e)}async _call(e){const t=await _(e),s=await this.processor(t),r=await this.model(s),o=[];for(const e of r.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 R extends h{constructor(e){super(e)}async _call(e){const t=await _(e),s=await this.processor(t),{predicted_depth:r}=await this.model(s),o=[];for(let e=0;e<t.length;++e){const s=r[e],[n,i]=s.dims.slice(-2),[a,l]=t[e].size,c=(await(0,d.interpolate_4d)(s.view(1,1,n,i),{size:[l,a],mode:"bilinear"})).view(l,a),_=c.min().item(),p=c.max().item(),m=c.sub(_).div_(p-_).mul_(255).to("uint8").unsqueeze(0),h=u.RawImage.fromTensor(m);o.push({predicted_depth:c,depth:h})}return o.length>1?o:o[0]}}const G=Object.freeze({"text-classification":{tokenizer:r.AutoTokenizer,pipeline:g,model:o.AutoModelForSequenceClassification,default:{model:"Xenova/distilbert-base-uncased-finetuned-sst-2-english"},type:"text"},"token-classification":{tokenizer:r.AutoTokenizer,pipeline:f,model:o.AutoModelForTokenClassification,default:{model:"Xenova/bert-base-multilingual-cased-ner-hrl"},type:"text"},"question-answering":{tokenizer:r.AutoTokenizer,pipeline:w,model:o.AutoModelForQuestionAnswering,default:{model:"Xenova/distilbert-base-cased-distilled-squad"},type:"text"},"fill-mask":{tokenizer:r.AutoTokenizer,pipeline:M,model:o.AutoModelForMaskedLM,default:{model:"Xenova/bert-base-uncased"},type:"text"},summarization:{tokenizer:r.AutoTokenizer,pipeline:b,model:o.AutoModelForSeq2SeqLM,default:{model:"Xenova/distilbart-cnn-6-6"},type:"text"},translation:{tokenizer:r.AutoTokenizer,pipeline:k,model:o.AutoModelForSeq2SeqLM,default:{model:"Xenova/t5-small"},type:"text"},"text2text-generation":{tokenizer:r.AutoTokenizer,pipeline:x,model:o.AutoModelForSeq2SeqLM,default:{model:"Xenova/flan-t5-small"},type:"text"},"text-generation":{tokenizer:r.AutoTokenizer,pipeline:v,model:o.AutoModelForCausalLM,default:{model:"Xenova/gpt2"},type:"text"},"zero-shot-classification":{tokenizer:r.AutoTokenizer,pipeline:T,model:o.AutoModelForSequenceClassification,default:{model:"Xenova/distilbert-base-uncased-mnli"},type:"text"},"audio-classification":{pipeline:C,model:o.AutoModelForAudioClassification,processor:n.AutoProcessor,default:{model:"Xenova/wav2vec2-base-superb-ks"},type:"audio"},"zero-shot-audio-classification":{tokenizer:r.AutoTokenizer,pipeline:S,model:o.AutoModel,processor:n.AutoProcessor,default:{model:"Xenova/clap-htsat-unfused"},type:"multimodal"},"automatic-speech-recognition":{tokenizer:r.AutoTokenizer,pipeline:A,model:[o.AutoModelForSpeechSeq2Seq,o.AutoModelForCTC],processor:n.AutoProcessor,default:{model:"Xenova/whisper-tiny.en"},type:"multimodal"},"text-to-audio":{tokenizer:r.AutoTokenizer,pipeline:N,model:[o.AutoModelForTextToWaveform,o.AutoModelForTextToSpectrogram],processor:[n.AutoProcessor,null],default:{model:"Xenova/speecht5_tts"},type:"text"},"image-to-text":{tokenizer:r.AutoTokenizer,pipeline:E,model:o.AutoModelForVision2Seq,processor:n.AutoProcessor,default:{model:"Xenova/vit-gpt2-image-captioning"},type:"multimodal"},"image-classification":{pipeline:L,model:o.AutoModelForImageClassification,processor:n.AutoProcessor,default:{model:"Xenova/vit-base-patch16-224"},type:"multimodal"},"image-segmentation":{pipeline:I,model:[o.AutoModelForImageSegmentation,o.AutoModelForSemanticSegmentation,o.AutoModelForUniversalSegmentation],processor:n.AutoProcessor,default:{model:"Xenova/detr-resnet-50-panoptic"},type:"multimodal"},"background-removal":{pipeline:z,model:[o.AutoModelForImageSegmentation,o.AutoModelForSemanticSegmentation,o.AutoModelForUniversalSegmentation],processor:n.AutoProcessor,default:{model:"Xenova/modnet"},type:"image"},"zero-shot-image-classification":{tokenizer:r.AutoTokenizer,pipeline:j,model:o.AutoModel,processor:n.AutoProcessor,default:{model:"Xenova/clip-vit-base-patch32"},type:"multimodal"},"object-detection":{pipeline:D,model:o.AutoModelForObjectDetection,processor:n.AutoProcessor,default:{model:"Xenova/detr-resnet-50"},type:"multimodal"},"zero-shot-object-detection":{tokenizer:r.AutoTokenizer,pipeline:O,model:o.AutoModelForZeroShotObjectDetection,processor:n.AutoProcessor,default:{model:"Xenova/owlvit-base-patch32"},type:"multimodal"},"document-question-answering":{tokenizer:r.AutoTokenizer,pipeline:V,model:o.AutoModelForDocumentQuestionAnswering,processor:n.AutoProcessor,default:{model:"Xenova/donut-base-finetuned-docvqa"},type:"multimodal"},"image-to-image":{pipeline:B,model:o.AutoModelForImageToImage,processor:n.AutoProcessor,default:{model:"Xenova/swin2SR-classical-sr-x2-64"},type:"image"},"depth-estimation":{pipeline:R,model:o.AutoModelForDepthEstimation,processor:n.AutoProcessor,default:{model:"Xenova/dpt-large"},type:"image"},"feature-extraction":{tokenizer:r.AutoTokenizer,pipeline:P,model:o.AutoModel,default:{model:"Xenova/all-MiniLM-L6-v2"},type:"text"},"image-feature-extraction":{processor:n.AutoProcessor,pipeline:F,model:[o.AutoModelForImageFeatureExtraction,o.AutoModel],default:{model:"Xenova/vit-base-patch16-224-in21k"},type:"image"}}),q=Object.freeze({"sentiment-analysis":"text-classification",ner:"token-classification",asr:"automatic-speech-recognition","text-to-speech":"text-to-audio",embeddings:"feature-extraction"});async function $(e,t=null,{progress_callback:s=null,config:r=null,cache_dir:o=null,local_files_only:n=!1,revision:i="main",device:l=null,dtype:c=null,subfolder:d="onnx",use_external_data_format:u=null,model_file_name:_=null,session_options:p={}}={}){e=q[e]??e;const m=G[e.split("_",1)[0]];if(!m)throw Error(`Unsupported pipeline: ${e}. Must be one of [${Object.keys(G)}]`);t||(t=m.default.model,console.log(`No model specified. Using default model: "${t}".`));const h={progress_callback:s,config:r,cache_dir:o,local_files_only:n,revision:i,device:l,dtype:c,subfolder:d,use_external_data_format:u,model_file_name:_,session_options:p},g=new Map([["tokenizer",m.tokenizer],["model",m.model],["processor",m.processor]]),f=await async function(e,t,s){const r=Object.create(null),o=[];for(const[n,i]of e.entries()){if(!i)continue;let e;e=Array.isArray(i)?new Promise((async(e,r)=>{let o;for(const n of i){if(null===n)return void e(null);try{return void e(await n.from_pretrained(t,s))}catch(e){if(e.message?.includes("Unsupported model type"))o=e;else{if(!e.message?.includes("Could not locate file"))return void r(e);o=e}}}r(o)})):i.from_pretrained(t,s),r[n]=e,o.push(e)}await Promise.all(o);for(const[e,t]of Object.entries(r))r[e]=await t;return r}(g,t,h);f.task=e,(0,a.dispatchCallback)(s,{status:"ready",task:e,model:t});return new(0,m.pipeline)(f)}},"./src/tokenizers.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{AlbertTokenizer:()=>ve,AutoTokenizer:()=>gt,BartTokenizer:()=>Ve,BertTokenizer:()=>ye,BlenderbotSmallTokenizer:()=>dt,BlenderbotTokenizer:()=>ct,BloomTokenizer:()=>Ge,CLIPTokenizer:()=>nt,CamembertTokenizer:()=>Ie,CodeGenTokenizer:()=>ot,CodeLlamaTokenizer:()=>We,CohereTokenizer:()=>mt,ConvBertTokenizer:()=>Ae,DebertaTokenizer:()=>Fe,DebertaV2Tokenizer:()=>Ce,DistilBertTokenizer:()=>Le,ElectraTokenizer:()=>je,EsmTokenizer:()=>Je,FalconTokenizer:()=>Qe,GPT2Tokenizer:()=>Oe,GPTNeoXTokenizer:()=>He,GemmaTokenizer:()=>Ke,Grok1Tokenizer:()=>Ze,HerbertTokenizer:()=>Se,LlamaTokenizer:()=>$e,M2M100Tokenizer:()=>st,MBart50Tokenizer:()=>Be,MBartTokenizer:()=>Ne,MPNetTokenizer:()=>Xe,MarianTokenizer:()=>at,MgpstrTokenizer:()=>ht,MobileBertTokenizer:()=>Te,NllbTokenizer:()=>tt,NougatTokenizer:()=>_t,PreTrainedTokenizer:()=>ke,Qwen2Tokenizer:()=>Ye,RoFormerTokenizer:()=>Ee,RobertaTokenizer:()=>Re,SiglipTokenizer:()=>it,SpeechT5Tokenizer:()=>ut,SqueezeBertTokenizer:()=>Pe,T5Tokenizer:()=>De,TokenizerModel:()=>y,VitsTokenizer:()=>pt,Wav2Vec2CTCTokenizer:()=>lt,WhisperTokenizer:()=>rt,XLMRobertaTokenizer:()=>Ue,XLMTokenizer:()=>ze,is_chinese_char:()=>f});var r=s("./src/utils/generic.js"),o=s("./src/utils/core.js"),n=s("./src/utils/hub.js"),i=s("./src/utils/maths.js"),a=s("./src/utils/tensor.js"),l=s("./src/utils/data-structures.js"),c=s("./node_modules/@huggingface/jinja/dist/index.js"),d=s("./src/models/whisper/common_whisper.js");async function u(e,t){const s=await Promise.all([(0,n.getModelJSON)(e,"tokenizer.json",!0,t),(0,n.getModelJSON)(e,"tokenizer_config.json",!0,t)]);return null!==t.legacy&&(s[1].legacy=t.legacy),s}function _(e,t=!0){if(void 0!==e.Regex){let t=e.Regex.replace(/\\([#&~])/g,"$1");for(const[e,s]of b)t=t.replaceAll(e,s);return new RegExp(t,"gu")}if(void 0!==e.String){const s=(0,o.escapeRegExp)(e.String);return new RegExp(t?s:`(${s})`,"gu")}return console.warn("Unknown pattern type:",e),null}function p(e){return new Map(Object.entries(e))}function m(e){const t=e.dims;switch(t.length){case 1:return e.tolist();case 2:if(1!==t[0])throw new Error("Unable to decode tensor with `batch size !== 1`. Use `tokenizer.batch_decode(...)` for batched inputs.");return e.tolist()[0];default:throw new Error(`Expected tensor to have 1-2 dimensions, got ${t.length}.`)}}function h(e){return e.replace(/ \./g,".").replace(/ \?/g,"?").replace(/ \!/g,"!").replace(/ ,/g,",").replace(/ \' /g,"'").replace(/ n\'t/g,"n't").replace(/ \'m/g,"'m").replace(/ \'s/g,"'s").replace(/ \'ve/g,"'ve").replace(/ \'re/g,"'re")}function g(e){return e.replace(/\p{M}/gu,"")}function f(e){return e>=19968&&e<=40959||e>=13312&&e<=19903||e>=131072&&e<=173791||e>=173824&&e<=177983||e>=177984&&e<=178207||e>=178208&&e<=183983||e>=63744&&e<=64255||e>=194560&&e<=195103}const w="\\p{P}\\u0021-\\u002F\\u003A-\\u0040\\u005B-\\u0060\\u007B-\\u007E",M=new RegExp(`^[${w}]+$`,"gu"),x=".,!?…。，、।۔،",b=new Map([["(?i:'s|'t|'re|'ve|'m|'ll|'d)","(?:'([sS]|[tT]|[rR][eE]|[vV][eE]|[mM]|[lL][lL]|[dD]))"],[` ?[^(\\s|[${x}])]+`,` ?[^\\s${x}]+`]]);class k{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 r.Callable{constructor(e){super(),this.config=e,this.vocab=[],this.tokens_to_ids=new Map,this.unk_token_id=void 0,this.unk_token=void 0,this.end_of_word_suffix=void 0,this.fuse_unk=this.config.fuse_unk??!1}static fromConfig(e,...t){switch(e.type){case"WordPiece":return new v(e);case"Unigram":return new T(e,...t);case"BPE":return new C(e);default:if(e.vocab)return Array.isArray(e.vocab)?new T(e,...t):"object"==typeof e.vocab&&e.continuing_subword_prefix&&e.unk_token?new v(e):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,s){const r=[];let o=0;for(;o<e.length;)if(r.push(e[o]),(t.get(e[o])??s)===s)for(;++o<e.length&&(t.get(e[o])??s)===s;)t.get(r.at(-1))!==s&&(r[r.length-1]+=e[o]);else++o;return r}(e,this.tokens_to_ids,this.unk_token_id)),e}encode(e){throw Error("encode should be implemented in subclass.")}convert_tokens_to_ids(e){return e.map((e=>this.tokens_to_ids.get(e)??this.unk_token_id))}convert_ids_to_tokens(e){return e.map((e=>this.vocab[e]??this.unk_token))}}class v extends y{constructor(e){super(e),this.tokens_to_ids=p(e.vocab),this.unk_token_id=this.tokens_to_ids.get(e.unk_token),this.unk_token=e.unk_token,this.max_input_chars_per_word=e.max_input_chars_per_word??100,this.vocab=new Array(this.tokens_to_ids.size);for(const[e,t]of this.tokens_to_ids)this.vocab[t]=e}encode(e){const t=[];for(const s of e){const e=[...s];if(e.length>this.max_input_chars_per_word){t.push(this.unk_token);continue}let r=!1,o=0;const n=[];for(;o<e.length;){let t=e.length,s=null;for(;o<t;){let r=e.slice(o,t).join("");if(o>0&&(r=this.config.continuing_subword_prefix+r),this.tokens_to_ids.has(r)){s=r;break}--t}if(null===s){r=!0;break}n.push(s),o=t}r?t.push(this.unk_token):t.push(...n)}return t}}class T extends y{constructor(e,t){super(e);const s=e.vocab.length;this.vocab=new Array(s),this.scores=new Array(s);for(let t=0;t<s;++t)[this.vocab[t],this.scores[t]]=e.vocab[t];this.unk_token_id=e.unk_id,this.unk_token=this.vocab[e.unk_id],this.tokens_to_ids=new Map(this.vocab.map(((e,t)=>[e,t]))),this.bos_token=" ",this.bos_token_id=this.tokens_to_ids.get(this.bos_token),this.eos_token=t.eos_token,this.eos_token_id=this.tokens_to_ids.get(this.eos_token),this.unk_token=this.vocab[this.unk_token_id],this.minScore=(0,i.min)(this.scores)[0],this.unk_score=this.minScore-10,this.scores[this.unk_token_id]=this.unk_score,this.trie=new l.CharTrie,this.trie.extend(this.vocab),this.fuse_unk=!0}populateNodes(e){const t=e.chars;let s=0;for(;s<t.length;){let r=!1;const n=[],i=t.slice(s).join(""),a=this.trie.commonPrefixSearch(i);for(const t of a){n.push(t);const i=this.tokens_to_ids.get(t),a=this.scores[i],l=(0,o.len)(t);e.insert(s,l,a,i),r||1!==l||(r=!0)}r||e.insert(s,1,this.unk_score,this.unk_token_id),s+=1}}tokenize(e){const t=new l.TokenLattice(e,this.bos_token_id,this.eos_token_id);return this.populateNodes(t),t.tokens()}encode(e){const t=[];for(const s of e){const e=this.tokenize(s);t.push(...e)}return t}}const P=(()=>{const e=[...Array.from({length:"~".charCodeAt(0)-"!".charCodeAt(0)+1},((e,t)=>t+"!".charCodeAt(0))),...Array.from({length:"¬".charCodeAt(0)-"¡".charCodeAt(0)+1},((e,t)=>t+"¡".charCodeAt(0))),...Array.from({length:"ÿ".charCodeAt(0)-"®".charCodeAt(0)+1},((e,t)=>t+"®".charCodeAt(0)))],t=e.slice();let s=0;for(let r=0;r<256;++r)e.includes(r)||(e.push(r),t.push(256+s),s+=1);const r=t.map((e=>String.fromCharCode(e)));return Object.fromEntries(e.map(((e,t)=>[e,r[t]])))})(),F=(0,o.reverseDictionary)(P);class C extends y{constructor(e){super(e),this.tokens_to_ids=p(e.vocab),this.unk_token_id=this.tokens_to_ids.get(e.unk_token),this.unk_token=e.unk_token,this.vocab=new Array(this.tokens_to_ids.size);for(const[e,t]of this.tokens_to_ids)this.vocab[t]=e;const t=Array.isArray(e.merges[0]);this.merges=t?e.merges:e.merges.map((e=>e.split(" ",2))),this.bpe_ranks=new Map(this.merges.map(((e,t)=>[JSON.stringify(e),t]))),this.end_of_word_suffix=e.end_of_word_suffix,this.continuing_subword_suffix=e.continuing_subword_suffix??null,this.byte_fallback=this.config.byte_fallback??!1,this.byte_fallback&&(this.text_encoder=new TextEncoder),this.ignore_merges=this.config.ignore_merges??!1,this.max_length_to_cache=256,this.cache_capacity=1e4,this.cache=new l.LRUCache(this.cache_capacity)}clear_cache(){this.cache.clear()}bpe(e){if(0===e.length)return[];const t=this.cache.get(e);if(void 0!==t)return t;const s=Array.from(e);this.end_of_word_suffix&&(s[s.length-1]+=this.end_of_word_suffix);let r=[];if(s.length>1){const e=new l.PriorityQueue(((e,t)=>e.score<t.score));let t={token:s[0],bias:0,prev:null,next:null},o=t;for(let t=1;t<s.length;++t){const r={bias:t/s.length,token:s[t],prev:o,next:null};o.next=r,this._add_node(e,o),o=r}for(;!e.isEmpty();){const s=e.pop();if(s.deleted||!s.next||s.next.deleted)continue;if(s.deleted=!0,s.next.deleted=!0,s.prev){const e={...s.prev};s.prev.deleted=!0,s.prev=e,e.prev?e.prev.next=e:t=e}const r={token:s.token+s.next.token,bias:s.bias,prev:s.prev,next:s.next.next};r.prev?(r.prev.next=r,this._add_node(e,r.prev)):t=r,r.next&&(r.next.prev=r,this._add_node(e,r))}for(let e=t;null!==e;e=e.next)r.push(e.token)}else r=s;if(this.continuing_subword_suffix)for(let e=0;e<r.length-1;++e)r[e]+=this.continuing_subword_suffix;return e.length<this.max_length_to_cache&&this.cache.put(e,r),r}_add_node(e,t){const s=this.bpe_ranks.get(JSON.stringify([t.token,t.next.token]));void 0!==s&&(t.score=s+t.bias,e.push(t))}encode(e){const t=[];for(const s of e){if(this.ignore_merges&&this.tokens_to_ids.has(s)){t.push(s);continue}const e=this.bpe(s);for(const s of e)if(this.tokens_to_ids.has(s))t.push(s);else if(this.byte_fallback){const e=Array.from(this.text_encoder.encode(s)).map((e=>`<0x${e.toString(16).toUpperCase().padStart(2,"0")}>`));e.every((e=>this.tokens_to_ids.has(e)))?t.push(...e):t.push(this.unk_token)}else t.push(this.unk_token)}return t}}class S extends y{constructor(e,t){super(e),this.tokens_to_ids=p(t.target_lang?e.vocab[t.target_lang]:e.vocab),this.bos_token=t.bos_token,this.bos_token_id=this.tokens_to_ids.get(this.bos_token),this.eos_token=t.eos_token,this.eos_token_id=this.tokens_to_ids.get(this.eos_token),this.pad_token=t.pad_token,this.pad_token_id=this.tokens_to_ids.get(this.pad_token),this.unk_token=t.unk_token,this.unk_token_id=this.tokens_to_ids.get(this.unk_token),this.vocab=new Array(this.tokens_to_ids.size);for(const[e,t]of this.tokens_to_ids)this.vocab[t]=e}encode(e){return e}}class A extends r.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"BertNormalizer":return new G(e);case"Precompiled":return new me(e);case"Sequence":return new R(e);case"Replace":return new E(e);case"NFC":return new I(e);case"NFD":return new z(e);case"NFKC":return new j(e);case"NFKD":return new D(e);case"Strip":return new O(e);case"StripAccents":return new V(e);case"Lowercase":return new N(e);case"Prepend":return new B(e);default:throw new Error(`Unknown Normalizer type: ${e.type}`)}}normalize(e){throw Error("normalize should be implemented in subclass.")}_call(e){return this.normalize(e)}}class E extends A{normalize(e){const t=_(this.config.pattern);return null===t?e:e.replaceAll(t,this.config.content)}}class L extends A{form=void 0;normalize(e){return e=e.normalize(this.form)}}class I extends L{form="NFC"}class z extends L{form="NFD"}class j extends L{form="NFKC"}class D extends L{form="NFKD"}class O extends A{normalize(e){return this.config.strip_left&&this.config.strip_right?e=e.trim():(this.config.strip_left&&(e=e.trimStart()),this.config.strip_right&&(e=e.trimEnd())),e}}class V extends A{normalize(e){return e=g(e)}}class N extends A{normalize(e){return e=e.toLowerCase()}}class B extends A{normalize(e){return e=this.config.prepend+e}}class R extends A{constructor(e){super(e),this.normalizers=e.normalizers.map((e=>A.fromConfig(e)))}normalize(e){return this.normalizers.reduce(((e,t)=>t.normalize(e)),e)}}class G extends A{_tokenize_chinese_chars(e){const t=[];for(let s=0;s<e.length;++s){const r=e[s];f(r.charCodeAt(0))?(t.push(" "),t.push(r),t.push(" ")):t.push(r)}return t.join("")}stripAccents(e){return e.normalize("NFD").replace(/\p{Mn}/gu,"")}_is_control(e){switch(e){case"\t":case"\n":case"\r":return!1;default:return/^\p{Cc}|\p{Cf}|\p{Co}|\p{Cs}$/u.test(e)}}_clean_text(e){const t=[];for(const s of e){const e=s.charCodeAt(0);0===e||65533===e||this._is_control(s)||(/^\s$/.test(s)?t.push(" "):t.push(s))}return t.join("")}normalize(e){return this.config.clean_text&&(e=this._clean_text(e)),this.config.handle_chinese_chars&&(e=this._tokenize_chinese_chars(e)),this.config.lowercase?(e=e.toLowerCase(),!1!==this.config.strip_accents&&(e=this.stripAccents(e))):this.config.strip_accents&&(e=this.stripAccents(e)),e}}class q extends r.Callable{static fromConfig(e){if(null===e)return null;switch(e.type){case"BertPreTokenizer":return new $(e);case"Sequence":return new he(e);case"Whitespace":return new ge(e);case"WhitespaceSplit":return new fe(e);case"Metaspace":return new _e(e);case"ByteLevel":return new W(e);case"Split":return new U(e);case"Punctuation":return new X(e);case"Digits":return new Q(e);case"Replace":return new we(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 $ extends q{constructor(e){super(),this.pattern=new RegExp(`[^\\s${w}]+|[${w}]`,"gu")}pre_tokenize_text(e,t){return e.trim().match(this.pattern)||[]}}class W extends q{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 U extends q{constructor(e){super(),this.config=e,this.pattern=_(this.config.pattern,this.config.invert)}pre_tokenize_text(e,t){return null===this.pattern?[]:this.config.invert?e.match(this.pattern)||[]:"removed"===this.config.behavior?.toLowerCase()?e.split(this.pattern).filter((e=>e)):function(e,t){const s=[];let r=0;for(const o of e.matchAll(t)){const t=o[0];r<o.index&&s.push(e.slice(r,o.index)),t.length>0&&s.push(t),r=o.index+t.length}return r<e.length&&s.push(e.slice(r)),s}(e,this.pattern)}}class X extends q{constructor(e){super(),this.config=e,this.pattern=new RegExp(`[^${w}]+|[${w}]+`,"gu")}pre_tokenize_text(e,t){return e.match(this.pattern)||[]}}class Q extends q{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 H extends r.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"TemplateProcessing":return new K(e);case"ByteLevel":return new Z(e);case"RobertaProcessing":return new Y(e);case"BertProcessing":return new J(e);case"Sequence":return new ee(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 J extends H{constructor(e){super(e),this.cls=e.cls[0],this.sep=e.sep[0]}post_process(e,t=null,{add_special_tokens:s=!0}={}){s&&(e=(0,o.mergeArrays)([this.cls],e,[this.sep]));let r=new Array(e.length).fill(0);if(null!==t){const n=s&&this instanceof Y?[this.sep]:[],i=s?[this.sep]:[];e=(0,o.mergeArrays)(e,n,t,i),r=(0,o.mergeArrays)(r,new Array(t.length+n.length+i.length).fill(1))}return{tokens:e,token_type_ids:r}}}class Y extends J{}class K extends H{constructor(e){super(e),this.single=e.single,this.pair=e.pair}post_process(e,t=null,{add_special_tokens:s=!0}={}){const r=null===t?this.single:this.pair;let n=[],i=[];for(const a of r)"SpecialToken"in a?s&&(n.push(a.SpecialToken.id),i.push(a.SpecialToken.type_id)):"Sequence"in a&&("A"===a.Sequence.id?(n=(0,o.mergeArrays)(n,e),i=(0,o.mergeArrays)(i,new Array(e.length).fill(a.Sequence.type_id))):"B"===a.Sequence.id&&(n=(0,o.mergeArrays)(n,t),i=(0,o.mergeArrays)(i,new Array(t.length).fill(a.Sequence.type_id))));return{tokens:n,token_type_ids:i}}}class Z extends H{post_process(e,t=null){return t&&(e=(0,o.mergeArrays)(e,t)),{tokens:e}}}class ee extends H{constructor(e){super(e),this.processors=e.processors.map((e=>H.fromConfig(e)))}post_process(e,t=null,s={}){let r;for(const o of this.processors)if(o instanceof Z){if(e=o.post_process(e).tokens,t){t=o.post_process(t).tokens}}else{const n=o.post_process(e,t,s);e=n.tokens,r=n.token_type_ids}return{tokens:e,token_type_ids:r}}}class te extends r.Callable{constructor(e){super(),this.config=e,this.added_tokens=[],this.end_of_word_suffix=null,this.trim_offsets=e.trim_offsets}static fromConfig(e){if(null===e)return null;switch(e.type){case"WordPiece":return new ie(e);case"Metaspace":return new pe(e);case"ByteLevel":return new ae(e);case"Replace":return new se(e);case"ByteFallback":return new re(e);case"Fuse":return new oe(e);case"Strip":return new ne(e);case"Sequence":return new ce(e);case"CTC":return new le(e);case"BPEDecoder":return new de(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 se extends te{decode_chain(e){const t=_(this.config.pattern);return null===t?e:e.map((e=>e.replaceAll(t,this.config.content)))}}class re extends te{constructor(e){super(e),this.text_decoder=new TextDecoder}decode_chain(e){const t=[];let s=[];for(const r of e){let e=null;if(6===r.length&&r.startsWith("<0x")&&r.endsWith(">")){const t=parseInt(r.slice(3,5),16);isNaN(t)||(e=t)}if(null!==e)s.push(e);else{if(s.length>0){const e=this.text_decoder.decode(Uint8Array.from(s));t.push(e),s=[]}t.push(r)}}if(s.length>0){const e=this.text_decoder.decode(Uint8Array.from(s));t.push(e),s=[]}return t}}class oe extends te{decode_chain(e){return[e.join("")]}}class ne extends te{constructor(e){super(e),this.content=this.config.content,this.start=this.config.start,this.stop=this.config.stop}decode_chain(e){return e.map((e=>{let t=0;for(let s=0;s<this.start&&e[s]===this.content;++s)t=s+1;let s=e.length;for(let t=0;t<this.stop;++t){const r=e.length-t-1;if(e[r]!==this.content)break;s=r}return e.slice(t,s)}))}}class ie extends te{constructor(e){super(e),this.cleanup=e.cleanup}decode_chain(e){return e.map(((e,t)=>(0!==t&&(e=e.startsWith(this.config.prefix)?e.replace(this.config.prefix,""):" "+e),this.cleanup&&(e=h(e)),e)))}}class ae extends te{constructor(e){super(e),this.byte_decoder=F,this.text_decoder=new TextDecoder("utf-8",{fatal:!1,ignoreBOM:!0}),this.end_of_word_suffix=null}convert_tokens_to_string(e){const t=e.join(""),s=new Uint8Array([...t].map((e=>this.byte_decoder[e])));return this.text_decoder.decode(s)}decode_chain(e){const t=[];let s=[];for(const r of e)void 0!==this.added_tokens.find((e=>e.content===r))?(s.length>0&&(t.push(this.convert_tokens_to_string(s)),s=[]),t.push(r)):s.push(r);return s.length>0&&t.push(this.convert_tokens_to_string(s)),t}}class le extends te{constructor(e){super(e),this.pad_token=this.config.pad_token,this.word_delimiter_token=this.config.word_delimiter_token,this.cleanup=this.config.cleanup}convert_tokens_to_string(e){if(0===e.length)return"";const t=[e[0]];for(let s=1;s<e.length;++s)e[s]!==t.at(-1)&&t.push(e[s]);let s=t.filter((e=>e!==this.pad_token)).join("");return this.cleanup&&(s=h(s).replaceAll(this.word_delimiter_token," ").trim()),s}decode_chain(e){return[this.convert_tokens_to_string(e)]}}class ce extends te{constructor(e){super(e),this.decoders=e.decoders.map((e=>te.fromConfig(e)))}decode_chain(e){return this.decoders.reduce(((e,t)=>t.decode_chain(e)),e)}}class de extends te{constructor(e){super(e),this.suffix=this.config.suffix}decode_chain(e){return e.map(((t,s)=>t.replaceAll(this.suffix,s===e.length-1?"":" ")))}}class ue extends te{decode_chain(e){let t="";for(let s=1;s<e.length;s+=2)t+=e[s];return[t]}}class _e extends q{constructor(e){super(),this.addPrefixSpace=e.add_prefix_space,this.replacement=e.replacement,this.strRep=e.str_rep||this.replacement,this.prepend_scheme=e.prepend_scheme??"always"}pre_tokenize_text(e,{section_index:t}={}){let s=e.replaceAll(" ",this.strRep);return this.addPrefixSpace&&!s.startsWith(this.replacement)&&("always"===this.prepend_scheme||"first"===this.prepend_scheme&&0===t)&&(s=this.strRep+s),[s]}}class pe extends te{constructor(e){super(e),this.addPrefixSpace=e.add_prefix_space,this.replacement=e.replacement}decode_chain(e){const t=[];for(let s=0;s<e.length;++s){let r=e[s].replaceAll(this.replacement," ");this.addPrefixSpace&&0==s&&r.startsWith(" ")&&(r=r.substring(1)),t.push(r)}return t}}class me 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 q{constructor(e){super(),this.tokenizers=e.pretokenizers.map((e=>q.fromConfig(e)))}pre_tokenize_text(e,t){return this.tokenizers.reduce(((e,s)=>s.pre_tokenize(e,t)),[e])}}class ge extends q{constructor(e){super()}pre_tokenize_text(e,t){return e.match(/\w+|[^\w\s]+/g)||[]}}class fe extends q{constructor(e){super()}pre_tokenize_text(e,t){return function(e){return e.match(/\S+/g)||[]}(e)}}class we extends q{constructor(e){super(),this.config=e,this.pattern=_(this.config.pattern),this.content=this.config.content}pre_tokenize_text(e,t){return null===this.pattern?[e]:[e.replaceAll(this.pattern,this.config.content)]}}const Me=["bos_token","eos_token","unk_token","sep_token","pad_token","cls_token","mask_token"];function xe(e,t,s,r){for(const n of Object.keys(e)){const i=t-e[n].length,a=s(n),l=new Array(i).fill(a);e[n]="right"===r?(0,o.mergeArrays)(e[n],l):(0,o.mergeArrays)(l,e[n])}}function be(e,t){for(const s of Object.keys(e))e[s].length=t}class ke extends r.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=q.fromConfig(e.pre_tokenizer),this.model=y.fromConfig(e.model,t),this.post_processor=H.fromConfig(e.post_processor),this.decoder=te.fromConfig(e.decoder),this.special_tokens=[],this.all_special_ids=[],this.added_tokens=[];for(const t of e.added_tokens){const e=new k(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_splitter=new l.DictionarySplitter(this.added_tokens.map((e=>e.content))),this.added_tokens_map=new Map(this.added_tokens.map((e=>[e.content,e]))),this.mask_token=this.getToken("mask_token"),this.mask_token_id=this.model.tokens_to_ids.get(this.mask_token),this.pad_token=this.getToken("pad_token","eos_token"),this.pad_token_id=this.model.tokens_to_ids.get(this.pad_token),this.sep_token=this.getToken("sep_token"),this.sep_token_id=this.model.tokens_to_ids.get(this.sep_token),this.unk_token=this.getToken("unk_token"),this.unk_token_id=this.model.tokens_to_ids.get(this.unk_token),this.bos_token=this.getToken("bos_token"),this.bos_token_id=this.model.tokens_to_ids.get(this.bos_token),this.eos_token=this.getToken("eos_token"),this.eos_token_id=this.model.tokens_to_ids.get(this.eos_token),this.model_max_length=t.model_max_length,this.remove_space=t.remove_space,this.clean_up_tokenization_spaces=t.clean_up_tokenization_spaces??!0,this.do_lowercase_and_remove_accent=t.do_lowercase_and_remove_accent??!1,t.padding_side&&(this.padding_side=t.padding_side),this.legacy=!1,this.chat_template=t.chat_template??null,Array.isArray(this.chat_template)){const e=Object.create(null);for(const{name:t,template:s}of this.chat_template){if("string"!=typeof t||"string"!=typeof s)throw new Error('Chat template must be a list of objects with "name" and "template" properties');e[t]=s}this.chat_template=e}this._compiled_template_cache=new Map}getToken(...e){for(const t of e){const e=this._tokenizer_config[t];if(e){if("object"==typeof e){if("AddedToken"===e.__type)return e.content;throw Error(`Unknown token: ${e}`)}return e}}return null}static async from_pretrained(e,{progress_callback:t=null,config:s=null,cache_dir:r=null,local_files_only:o=!1,revision:n="main",legacy:i=null}={}){return new this(...await u(e,{progress_callback:t,config:s,cache_dir:r,local_files_only:o,revision:n,legacy:i}))}_call(e,{text_pair:t=null,add_special_tokens:s=!0,padding:r=!1,truncation:o=null,max_length:n=null,return_tensor:l=!0,return_token_type_ids:c=null}={}){const d=Array.isArray(e);let u;if(d){if(0===e.length)throw Error("text array must be non-empty");if(null!==t){if(!Array.isArray(t))throw Error("text_pair must also be an array");if(e.length!==t.length)throw Error("text and text_pair must have the same length");u=e.map(((e,r)=>this._encode_plus(e,{text_pair:t[r],add_special_tokens:s,return_token_type_ids:c})))}else u=e.map((e=>this._encode_plus(e,{add_special_tokens:s,return_token_type_ids:c})))}else{if(null==e)throw Error("text may not be null or undefined");if(Array.isArray(t))throw Error("When specifying `text_pair`, since `text` is a string, `text_pair` must also be a string (i.e., not an array).");u=[this._encode_plus(e,{text_pair:t,add_special_tokens:s,return_token_type_ids:c})]}if(null===n?n="max_length"===r?this.model_max_length:(0,i.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."),n=Math.min(n,this.model_max_length??1/0),r||o)for(let e=0;e<u.length;++e)u[e].input_ids.length!==n&&(u[e].input_ids.length>n?o&&be(u[e],n):r&&xe(u[e],n,(e=>"input_ids"===e?this.pad_token_id:0),this.padding_side));const _={};if(l){if((!r||!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]))_[t]=new a.Tensor("int64",BigInt64Array.from(u.flatMap((e=>e[t])).map(BigInt)),e)}else{for(const e of Object.keys(u[0]))_[e]=u.map((t=>t[e]));if(!d)for(const e of Object.keys(_))_[e]=_[e][0]}return _}_encode_text(e){if(null===e)return null;const t=this.added_tokens_splitter.split(e);for(let e=0;e<t.length;++e){const s=this.added_tokens_map.get(t[e]);s&&(s.lstrip&&e>0&&(t[e-1]=t[e-1].trimEnd()),s.rstrip&&e<t.length-1&&(t[e+1]=t[e+1].trimStart()))}const s=t.flatMap(((e,t)=>{if(0===e.length)return[];if(this.added_tokens_map.has(e))return[e];if(!0===this.remove_space&&(e=e.trim().split(/\s+/).join(" ")),this.do_lowercase_and_remove_accent&&(e=function(e){return g(e.toLowerCase())}(e)),null!==this.normalizer&&(e=this.normalizer(e)),0===e.length)return[];const s=null!==this.pre_tokenizer?this.pre_tokenizer(e,{section_index:t}):[e];return this.model(s)}));return s}_encode_plus(e,{text_pair:t=null,add_special_tokens:s=!0,return_token_type_ids:r=null}={}){const{tokens:o,token_type_ids:n}=this._tokenize_helper(e,{pair:t,add_special_tokens:s}),i=this.model.convert_tokens_to_ids(o),a={input_ids:i,attention_mask:new Array(i.length).fill(1)};return(r??this.return_token_type_ids)&&n&&(a.token_type_ids=n),a}_tokenize_helper(e,{pair:t=null,add_special_tokens:s=!1}={}){const r=this._encode_text(e),n=this._encode_text(t);return this.post_processor?this.post_processor(r,n,{add_special_tokens:s}):{tokens:(0,o.mergeArrays)(r??[],n??[])}}tokenize(e,{pair:t=null,add_special_tokens:s=!1}={}){return this._tokenize_helper(e,{pair:t,add_special_tokens:s}).tokens}encode(e,{text_pair:t=null,add_special_tokens:s=!0,return_token_type_ids:r=null}={}){return this._encode_plus(e,{text_pair:t,add_special_tokens:s,return_token_type_ids:r}).input_ids}batch_decode(e,t={}){return e instanceof a.Tensor&&(e=e.tolist()),e.map((e=>this.decode(e,t)))}decode(e,t={}){if(e instanceof a.Tensor&&(e=m(e)),!Array.isArray(e)||0===e.length||!(0,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:s=null}){let r=this.model.convert_ids_to_tokens(e);t&&(r=r.filter((e=>!this.special_tokens.includes(e))));let o=this.decoder?this.decoder(r):r.join(" ");return this.decoder&&this.decoder.end_of_word_suffix&&(o=o.replaceAll(this.decoder.end_of_word_suffix," "),t&&(o=o.trim())),(s??this.clean_up_tokenization_spaces)&&(o=h(o)),o}get_chat_template({chat_template:e=null,tools:t=null}={}){if(this.chat_template&&"object"==typeof this.chat_template){const s=this.chat_template;if(null!==e&&Object.hasOwn(s,e))e=s[e];else if(null===e)if(null!==t&&"tool_use"in s)e=s.tool_use;else{if(!("default"in s))throw Error(`This model has multiple chat templates with no default specified! Please either pass a chat template or the name of the template you wish to use to the 'chat_template' argument. Available template names are ${Object.keys(s).sort()}.`);e=s.default}}else if(null===e){if(!this.chat_template)throw Error("Cannot use apply_chat_template() because tokenizer.chat_template is not set and no template argument was passed! For information about writing templates and setting the tokenizer.chat_template attribute, please see the documentation at https://huggingface.co/docs/transformers/main/en/chat_templating");e=this.chat_template}return e}apply_chat_template(e,{tools:t=null,documents:s=null,chat_template:r=null,add_generation_prompt:o=!1,tokenize:n=!0,padding:i=!1,truncation:a=!1,max_length:l=null,return_tensor:d=!0,return_dict:u=!1,tokenizer_kwargs:_={},...p}={}){if("string"!=typeof(r=this.get_chat_template({chat_template:r,tools:t})))throw Error("chat_template must be a string, but got "+typeof r);let m=this._compiled_template_cache.get(r);void 0===m&&(m=new c.Template(r),this._compiled_template_cache.set(r,m));const h=Object.create(null);for(const e of Me){const t=this.getToken(e);t&&(h[e]=t)}const g=m.render({messages:e,add_generation_prompt:o,tools:t,documents:s,...h,...p});if(n){const e=this._call(g,{add_special_tokens:!1,padding:i,truncation:a,max_length:l,return_tensor:d,..._});return u?e:e.input_ids}return g}}class ye extends ke{return_token_type_ids=!0}class ve extends ke{return_token_type_ids=!0}class Te extends ke{return_token_type_ids=!0}class Pe extends ke{return_token_type_ids=!0}class Fe extends ke{return_token_type_ids=!0}class Ce extends ke{return_token_type_ids=!0}class Se extends ke{return_token_type_ids=!0}class Ae extends ke{return_token_type_ids=!0}class Ee extends ke{return_token_type_ids=!0}class Le extends ke{}class Ie extends ke{}class ze extends ke{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 je extends ke{return_token_type_ids=!0}class De extends ke{}class Oe extends ke{}class Ve extends ke{}class Ne extends ke{constructor(e,t){super(e,t),this.languageRegex=/^[a-z]{2}_[A-Z]{2}$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))),this.lang_to_token=e=>e}_build_translation_inputs(e,t,s){return et(this,e,t,s)}}class Be extends Ne{}class Re extends ke{}class Ge extends ke{}const qe="▁";class $e extends ke{padding_side="left";constructor(e,t){super(e,t),this.legacy=t.legacy??!0,this.legacy||(this.normalizer=null,this.pre_tokenizer=new _e({replacement:qe,add_prefix_space:!0,prepend_scheme:"first"}))}_encode_text(e){if(null===e)return null;if(this.legacy||0===e.length)return super._encode_text(e);let t=super._encode_text(qe+e.replaceAll(qe," "));return t.length>1&&t[0]===qe&&this.special_tokens.includes(t[1])&&(t=t.slice(1)),t}}class We extends ke{}class Ue extends ke{}class Xe extends ke{}class Qe extends ke{}class He extends ke{}class Je extends ke{}class Ye extends ke{}class Ke extends ke{}class Ze extends ke{}function et(e,t,s,r){if(!("language_codes"in e)||!Array.isArray(e.language_codes))throw new Error("Tokenizer must have `language_codes` attribute set and it should be an array of language ids.");if(!("languageRegex"in e&&e.languageRegex instanceof RegExp))throw new Error("Tokenizer must have `languageRegex` attribute set and it should be a regular expression.");if(!("lang_to_token"in e)||"function"!=typeof e.lang_to_token)throw new Error("Tokenizer must have `lang_to_token` attribute set and it should be a function.");const o=r.src_lang,n=r.tgt_lang;if(!e.language_codes.includes(n))throw new Error(`Target language code "${n}" 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: {${e.language_codes.join(", ")}}`);for(const t of e.post_processor.config.single)if("SpecialToken"in t&&e.languageRegex.test(t.SpecialToken.id)){t.SpecialToken.id=e.lang_to_token(o);break}}return r.forced_bos_token_id=e.model.convert_tokens_to_ids([e.lang_to_token(n)])[0],e._call(t,s)}class tt extends ke{constructor(e,t){super(e,t),this.languageRegex=/^[a-z]{3}_[A-Z][a-z]{3}$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))),this.lang_to_token=e=>e}_build_translation_inputs(e,t,s){return et(this,e,t,s)}}class st extends ke{constructor(e,t){super(e,t),this.languageRegex=/^__[a-z]{2,3}__$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))).map((e=>e.slice(2,-2))),this.lang_to_token=e=>`__${e}__`}_build_translation_inputs(e,t,s){return et(this,e,t,s)}}class rt extends ke{get timestamp_begin(){return this.model.convert_tokens_to_ids(["<|notimestamps|>"])[0]+1}_decode_asr(e,{return_timestamps:t=!1,return_language:s=!1,time_precision:r=null,force_full_sequences:o=!0}={}){if(null===r)throw Error("Must specify time_precision");let n=null;const a="word"===t;function l(){return{language:n,timestamp:[null,null],text:""}}const c=[];let u=l(),_=0;const p=this.timestamp_begin,m=p+1500;let h=[],g=[],f=!1,w=null;const x=new Set(this.all_special_ids);for(const s of e){const e=s.tokens,o=a?s.token_timestamps:null;let b=null,k=p;if("stride"in s){const[t,o,n]=s.stride;if(_-=o,w=t-n,o&&(k=o/r+p),n)for(let t=e.length-1;t>=0;--t){const s=Number(e[t]);if(s>=p){if(null!==b&&(s-p)*r<w)break;b=s}}}let y=[],v=[];for(let s=0;s<e.length;++s){const w=Number(e[s]);if(x.has(w)){const e=this.decode([w]),s=d.WHISPER_LANGUAGE_MAPPING.get(e.slice(2,-2));if(void 0!==s){if(null!==n&&s!==n&&!t){h.push(y);const e=this.findLongestCommonSequence(h)[0],t=this.decode(e);u.text=t,c.push(u),h=[],y=[],u=l()}n=u.language=s}}else if(w>=p&&w<=m){const e=(w-p)*r+_,t=(0,i.round)(e,2);if(null!==b&&w>=b)f=!0;else if(f||h.length>0&&w<k)f=!1;else if(null===u.timestamp[0])u.timestamp[0]=t;else if(t===u.timestamp[0]);else{u.timestamp[1]=t,h.push(y),a&&g.push(v);const[e,s]=this.findLongestCommonSequence(h,g),r=this.decode(e);u.text=r,a&&(u.words=this.collateWordTimestamps(e,s,n)),c.push(u),h=[],y=[],g=[],v=[],u=l()}}else if(y.push(w),a){let e,t=(0,i.round)(o[s]+_,2);if(s+1<o.length){e=(0,i.round)(o[s+1]+_,2);const n=this.decode([w]);M.test(n)&&(e=(0,i.round)(Math.min(t+r,e),2))}else e=null;v.push([t,e])}}if("stride"in s){const[e,t,r]=s.stride;_+=e-r}y.length>0?(h.push(y),a&&g.push(v)):h.every((e=>0===e.length))&&(u=l(),h=[],y=[],g=[],v=[])}if(h.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,s]=this.findLongestCommonSequence(h,g),r=this.decode(e);u.text=r,a&&(u.words=this.collateWordTimestamps(e,s,n)),c.push(u)}let b=Object.create(null);const k=c.map((e=>e.text)).join("");if(t||s){for(let e=0;e<c.length;++e){const r=c[e];t||delete r.timestamp,s||delete r.language}if(a){const e=[];for(const t of c)for(const s of t.words)e.push(s);b={chunks:e}}else b={chunks:c}}return[k,b]}findLongestCommonSequence(e,t=null){let s=e[0],r=s.length,o=[];const n=Array.isArray(t)&&t.length>0;let i=n?[]:null,a=n?t[0]:null;for(let l=1;l<e.length;++l){const c=e[l];let d=0,u=[r,r,0,0];const _=c.length;for(let e=1;e<r+_;++e){const o=Math.max(0,r-e),i=Math.min(r,r+_-e),p=s.slice(o,i),m=Math.max(0,e-r),h=Math.min(_,e),g=c.slice(m,h);if(p.length!==g.length)throw new Error("There is a bug within whisper `decode_asr` function, please report it. Dropping to prevent bad inference.");let f;f=n?p.filter(((e,s)=>e===g[s]&&a[o+s]<=t[l][m+s])).length:p.filter(((e,t)=>e===g[t])).length;const w=f/e+e/1e4;f>1&&w>d&&(d=w,u=[o,i,m,h])}const[p,m,h,g]=u,f=Math.floor((m+p)/2),w=Math.floor((g+h)/2);o.push(...s.slice(0,f)),s=c.slice(w),r=s.length,n&&(i.push(...a.slice(0,f)),a=t[l].slice(w))}return o.push(...s),n?(i.push(...a),[o,i]):[o,[]]}collateWordTimestamps(e,t,s){const[r,o,n]=this.combineTokensIntoWords(e,s),i=[];for(let e=0;e<r.length;++e){const s=n[e];i.push({text:r[e],timestamp:[t[s.at(0)][0],t[s.at(-1)][1]]})}return i}combineTokensIntoWords(e,t,s="\"'“¡¿([{-",r="\"'.。,，!！?？:：”)]}、"){let o,n,i;return["chinese","japanese","thai","lao","myanmar"].includes(t=t??"english")?[o,n,i]=this.splitTokensOnUnicode(e):[o,n,i]=this.splitTokensOnSpaces(e),this.mergePunctuations(o,n,i,s,r)}decode(e,t){let s;return t?.decode_with_timestamps?(e instanceof a.Tensor&&(e=m(e)),s=this.decodeWithTimestamps(e,t)):s=super.decode(e,t),s}decodeWithTimestamps(e,t){const s=t?.time_precision??.02,r=Array.from(this.all_special_ids).at(-1)+1;let o=[[]];for(let t of e)if(t=Number(t),t>=r){const e=((t-r)*s).toFixed(2);o.push(`<|${e}|>`),o.push([])}else o[o.length-1].push(t);return o=o.map((e=>"string"==typeof e?e:super.decode(e,t))),o.join("")}splitTokensOnUnicode(e){const t=this.decode(e,{decode_with_timestamps:!0}),s=[],r=[],o=[];let n=[],i=[],a=0;for(let l=0;l<e.length;++l){const c=e[l];n.push(c),i.push(l);const d=this.decode(n,{decode_with_timestamps:!0});d.includes("�")&&"�"!==t[a+d.indexOf("�")]||(s.push(d),r.push(n),o.push(i),n=[],i=[],a+=d.length)}return[s,r,o]}splitTokensOnSpaces(e){const[t,s,r]=this.splitTokensOnUnicode(e),o=[],n=[],i=[],a=new RegExp(`^[${w}]$`,"gu");for(let e=0;e<t.length;++e){const l=t[e],c=s[e],d=r[e],u=c[0]>=this.model.tokens_to_ids.get("<|endoftext|>"),_=l.startsWith(" "),p=l.trim(),m=a.test(p);if(u||_||m||0===o.length)o.push(l),n.push(c),i.push(d);else{const e=o.length-1;o[e]+=l,n[e].push(...c),i[e].push(...d)}}return[o,n,i]}mergePunctuations(e,t,s,r,n){const i=structuredClone(e),a=structuredClone(t),l=structuredClone(s);let c=i.length-2,d=i.length-1;for(;c>=0;)i[c].startsWith(" ")&&r.includes(i[c].trim())?(i[d]=i[c]+i[d],a[d]=(0,o.mergeArrays)(a[c],a[d]),l[d]=(0,o.mergeArrays)(l[c],l[d]),i[c]="",a[c]=[],l[c]=[]):d=c,--c;for(c=0,d=1;d<i.length;)!i[c].endsWith(" ")&&n.includes(i[d])?(i[c]+=i[d],a[c]=(0,o.mergeArrays)(a[c],a[d]),l[c]=(0,o.mergeArrays)(l[c],l[d]),i[d]="",a[d]=[],l[d]=[]):c=d,++d;return[i.filter((e=>e)),a.filter((e=>e.length>0)),l.filter((e=>e.length>0))]}}class ot extends ke{}class nt extends ke{}class it extends ke{}class at extends ke{constructor(e,t){super(e,t),this.languageRegex=/^(>>\w+<<)\s*/g,this.supported_language_codes=this.model.vocab.filter((e=>this.languageRegex.test(e))),console.warn('WARNING: `MarianTokenizer` is not yet supported by Hugging Face\'s "fast" tokenizers library. Therefore, you may experience slightly inaccurate results.')}_encode_text(e){if(null===e)return null;const[t,...s]=e.trim().split(this.languageRegex);if(0===s.length)return super._encode_text(t);if(2===s.length){const[e,t]=s;return this.supported_language_codes.includes(e)||console.warn(`Unsupported language code "${e}" detected, which may lead to unexpected behavior. Should be one of: ${JSON.stringify(this.supported_language_codes)}`),(0,o.mergeArrays)([e],super._encode_text(t))}}}class lt extends ke{}class ct extends ke{}class dt extends ke{}class ut extends ke{}class _t extends ke{}class pt extends ke{constructor(e,t){super(e,t),this.decoder=new ue({})}}class mt extends ke{}class ht extends ke{}class gt{static TOKENIZER_CLASS_MAPPING={T5Tokenizer:De,DistilBertTokenizer:Le,CamembertTokenizer:Ie,DebertaTokenizer:Fe,DebertaV2Tokenizer:Ce,BertTokenizer:ye,HerbertTokenizer:Se,ConvBertTokenizer:Ae,RoFormerTokenizer:Ee,XLMTokenizer:ze,ElectraTokenizer:je,MobileBertTokenizer:Te,SqueezeBertTokenizer:Pe,AlbertTokenizer:ve,GPT2Tokenizer:Oe,BartTokenizer:Ve,MBartTokenizer:Ne,MBart50Tokenizer:Be,RobertaTokenizer:Re,WhisperTokenizer:rt,CodeGenTokenizer:ot,CLIPTokenizer:nt,SiglipTokenizer:it,MarianTokenizer:at,BloomTokenizer:Ge,NllbTokenizer:tt,M2M100Tokenizer:st,LlamaTokenizer:$e,CodeLlamaTokenizer:We,XLMRobertaTokenizer:Ue,MPNetTokenizer:Xe,FalconTokenizer:Qe,GPTNeoXTokenizer:He,EsmTokenizer:Je,Wav2Vec2CTCTokenizer:lt,BlenderbotTokenizer:ct,BlenderbotSmallTokenizer:dt,SpeechT5Tokenizer:ut,NougatTokenizer:_t,VitsTokenizer:pt,Qwen2Tokenizer:Ye,GemmaTokenizer:Ke,Grok1Tokenizer:Ze,CohereTokenizer:mt,MgpstrTokenizer:ht,PreTrainedTokenizer:ke};static async from_pretrained(e,{progress_callback:t=null,config:s=null,cache_dir:r=null,local_files_only:o=!1,revision:n="main",legacy:i=null}={}){const[a,l]=await u(e,{progress_callback:t,config:s,cache_dir:r,local_files_only:o,revision:n,legacy:i}),c=l.tokenizer_class?.replace(/Fast$/,"")??"PreTrainedTokenizer";let d=this.TOKENIZER_CLASS_MAPPING[c];return d||(console.warn(`Unknown tokenizer class "${c}", attempting to construct from base class.`),d=ke),new d(a,l)}}},"./src/utils/audio.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{RawAudio:()=>k,hamming:()=>_,hanning:()=>u,mel_filter_bank:()=>f,read_audio:()=>c,spectrogram:()=>M,window_function:()=>x});var r=s("./src/utils/hub.js"),o=s("./src/utils/maths.js"),n=s("./src/utils/core.js"),i=s("./src/env.js"),a=s("fs"),l=s("./src/utils/tensor.js");async function c(e,t){if("undefined"==typeof AudioContext)throw Error("Unable to load audio from path/URL since `AudioContext` is not available in your environment. Instead, audio data should be passed directly to the pipeline/processor. For more information and some example code, see https://huggingface.co/docs/transformers.js/guides/node-audio-processing.");const s=await(await(0,r.getFile)(e)).arrayBuffer(),o=new AudioContext({sampleRate:t});void 0===t&&console.warn(`No sampling rate provided, using default of ${o.sampleRate}Hz.`);const n=await o.decodeAudioData(s);let i;if(2===n.numberOfChannels){const e=Math.sqrt(2),t=n.getChannelData(0),s=n.getChannelData(1);i=new Float32Array(t.length);for(let r=0;r<n.length;++r)i[r]=e*(t[r]+s[r])/2}else i=n.getChannelData(0);return i}function d(e,t){if(e<1)return new Float64Array;if(1===e)return new Float64Array([1]);const s=1-t,r=2*Math.PI/(e-1),o=new Float64Array(e);for(let n=0;n<e;++n)o[n]=t-s*Math.cos(n*r);return o}function u(e){return d(e,.5)}function _(e){return d(e,.54)}const p={htk:e=>2595*Math.log10(1+e/700),kaldi:e=>1127*Math.log(1+e/700),slaney:(e,t=1e3,s=15,r=27/Math.log(6.4))=>e>=t?s+Math.log(e/t)*r:3*e/200};function m(e,t="htk"){const s=p[t];if(!s)throw new Error('mel_scale should be one of "htk", "slaney" or "kaldi".');return"number"==typeof e?s(e):e.map((e=>s(e)))}const h={htk:e=>700*(10**(e/2595)-1),kaldi:e=>700*(Math.exp(e/1127)-1),slaney:(e,t=1e3,s=15,r=Math.log(6.4)/27)=>e>=s?t*Math.exp(r*(e-s)):200*e/3};function g(e,t,s){const r=(t-e)/(s-1);return Float64Array.from({length:s},((t,s)=>e+r*s))}function f(e,t,s,r,o,n=null,i="htk",a=!1){if(null!==n&&"slaney"!==n)throw new Error('norm must be one of null or "slaney"');if(e<2)throw new Error(`Require num_frequency_bins: ${e} >= 2`);if(s>r)throw new Error(`Require min_frequency: ${s} <= max_frequency: ${r}`);const l=g(m(s,i),m(r,i),t+2);let c,d=function(e,t="htk"){const s=h[t];if(!s)throw new Error('mel_scale should be one of "htk", "slaney" or "kaldi".');return"number"==typeof e?s(e):e.map((e=>s(e)))}(l,i);if(a){const t=o/(2*(e-1));c=m(Float64Array.from({length:e},((e,s)=>s*t)),i),d=l}else c=g(0,Math.floor(o/2),e);const u=function(e,t){const s=Float64Array.from({length:t.length-1},((e,s)=>t[s+1]-t[s])),r=Array.from({length:e.length},(()=>new Array(t.length)));for(let s=0;s<e.length;++s){const o=r[s];for(let r=0;r<t.length;++r)o[r]=t[r]-e[s]}const o=t.length-2,n=Array.from({length:o},(()=>new Array(e.length)));for(let t=0;t<e.length;++t){const e=r[t];for(let r=0;r<o;++r){const o=-e[r]/s[r],i=e[r+2]/s[r+1];n[r][t]=Math.max(0,Math.min(o,i))}}return n}(c,d);if(null!==n&&"slaney"===n)for(let s=0;s<t;++s){const t=u[s],r=2/(d[s+2]-d[s]);for(let s=0;s<e;++s)t[s]*=r}return u}function w(e,t,s,r,n){if(s<=0)throw new Error("reference must be greater than zero");if(r<=0)throw new Error("min_value must be greater than zero");s=Math.max(r,s);const i=Math.log10(s);for(let s=0;s<e.length;++s)e[s]=t*Math.log10(Math.max(r,e[s])-i);if(null!==n){if(n<=0)throw new Error("db_range must be greater than zero");const t=(0,o.max)(e)[0]-n;for(let s=0;s<e.length;++s)e[s]=Math.max(e[s],t)}return e}async function M(e,t,s,r,{fft_length:i=null,power:a=1,center:c=!0,pad_mode:d="reflect",onesided:u=!0,preemphasis:_=null,mel_filters:p=null,mel_floor:m=1e-10,log_mel:h=null,reference:g=1,min_value:f=1e-10,db_range:M=null,remove_dc_offset:x=null,min_num_frames:b=null,max_num_frames:k=null,do_pad:y=!0,transpose:v=!1}={}){const T=t.length;if(null===i&&(i=s),s>i)throw Error(`frame_length (${s}) may not be larger than fft_length (${i})`);if(T!==s)throw new Error(`Length of the window (${T}) must equal frame_length (${s})`);if(r<=0)throw new Error("hop_length must be greater than zero");if(null===a&&null!==p)throw new Error("You have provided `mel_filters` but `power` is `None`. Mel spectrogram computation is not yet supported for complex-valued spectrogram. Specify `power` to fix this issue.");if(c){if("reflect"!==d)throw new Error(`pad_mode="${d}" not implemented yet.`);const t=Math.floor((i-1)/2)+1;e=function(e,t,s){const r=new e.constructor(e.length+t+s),o=e.length-1;for(let s=0;s<e.length;++s)r[t+s]=e[s];for(let s=1;s<=t;++s)r[t-s]=e[(0,n.calculateReflectOffset)(s,o)];for(let i=1;i<=s;++i)r[o+t+i]=e[(0,n.calculateReflectOffset)(o-i,o)];return r}(e,t,t)}let P=Math.floor(1+Math.floor((e.length-s)/r));null!==b&&P<b&&(P=b);const F=u?Math.floor(i/2)+1:i;let C=P,S=P;null!==k&&(k>P?y&&(S=k):S=C=k);const A=new o.FFT(i),E=new Float64Array(i),L=new Float64Array(A.outputBufferSize),I=new Float32Array(F*S);for(let o=0;o<C;++o){const n=o*r,i=Math.min(e.length-n,s);i!==s&&E.fill(0,0,s);for(let t=0;t<i;++t)E[t]=e[n+t];if(x){let e=0;for(let t=0;t<i;++t)e+=E[t];const t=e/i;for(let e=0;e<i;++e)E[e]-=t}if(null!==_){for(let e=i-1;e>=1;--e)E[e]-=_*E[e-1];E[0]*=1-_}for(let e=0;e<t.length;++e)E[e]*=t[e];A.realTransform(L,E);for(let e=0;e<F;++e){const t=e<<1;I[e*S+o]=L[t]**2+L[t+1]**2}}if(null!==a&&2!==a){const e=2/a;for(let t=0;t<I.length;++t)I[t]**=e}const z=p.length;let j=await(0,l.matmul)(new l.Tensor("float32",p.flat(),[z,F]),new l.Tensor("float32",I,[F,S]));v&&(j=j.transpose(1,0));const D=j.data;for(let e=0;e<D.length;++e)D[e]=Math.max(m,D[e]);if(null!==a&&null!==h){const e=Math.min(D.length,C*z);switch(h){case"log":for(let t=0;t<e;++t)D[t]=Math.log(D[t]);break;case"log10":for(let t=0;t<e;++t)D[t]=Math.log10(D[t]);break;case"dB":if(1===a)!function(e,t=1,s=1e-5,r=null){w(e,20,t,s,r)}(D,g,f,M);else{if(2!==a)throw new Error(`Cannot use log_mel option '${h}' with power ${a}`);!function(e,t=1,s=1e-10,r=null){w(e,10,t,s,r)}(D,g,f,M)}break;default:throw new Error(`log_mel must be one of null, 'log', 'log10' or 'dB'. Got '${h}'`)}}return j}function x(e,t,{periodic:s=!0,frame_length:r=null,center:o=!0}={}){const n=s?e+1:e;let i;switch(t){case"boxcar":i=new Float64Array(n).fill(1);break;case"hann":case"hann_window":i=u(n);break;case"hamming":i=_(n);break;case"povey":i=u(n).map((e=>Math.pow(e,.85)));break;default:throw new Error(`Unknown window type ${t}.`)}if(s&&(i=i.subarray(0,e)),null===r)return i;if(e>r)throw new Error(`Length of the window (${e}) may not be larger than frame_length (${r})`);return i}function b(e,t,s){for(let r=0;r<s.length;++r)e.setUint8(t+r,s.charCodeAt(r))}class k{constructor(e,t){this.audio=e,this.sampling_rate=t}toWav(){return function(e,t){let s=44;const r=new ArrayBuffer(s+4*e.length),o=new DataView(r);b(o,0,"RIFF"),o.setUint32(4,36+4*e.length,!0),b(o,8,"WAVE"),b(o,12,"fmt "),o.setUint32(16,16,!0),o.setUint16(20,3,!0),o.setUint16(22,1,!0),o.setUint32(24,t,!0),o.setUint32(28,4*t,!0),o.setUint16(32,4,!0),o.setUint16(34,32,!0),b(o,36,"data"),o.setUint32(40,4*e.length,!0);for(let t=0;t<e.length;++t,s+=4)o.setFloat32(s,e[t],!0);return r}(this.audio,this.sampling_rate)}toBlob(){const e=this.toWav();return new Blob([e],{type:"audio/wav"})}async save(e){let t;if(i.apis.IS_BROWSER_ENV){if(i.apis.IS_WEBWORKER_ENV)throw new Error("Unable to save a file from a Web Worker.");t=n.saveBlob}else{if(!i.apis.IS_FS_AVAILABLE)throw new Error("Unable to save because filesystem is disabled in this environment.");t=async(e,t)=>{let s=await t.arrayBuffer();a.writeFileSync(e,Buffer.from(s))}}await t(e,this.toBlob())}}},"./src/utils/constants.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{CHAT_TEMPLATE_NAME:()=>l,CONFIG_NAME:()=>o,FEATURE_EXTRACTOR_NAME:()=>n,GENERATION_CONFIG_NAME:()=>c,GITHUB_ISSUE_URL:()=>r,IMAGE_PROCESSOR_NAME:()=>i,PROCESSOR_NAME:()=>a});const r="https://github.com/huggingface/transformers.js/issues/new/choose",o="config.json",n="preprocessor_config.json",i=n,a="processor_config.json",l="chat_template.json",c="generation_config.json"},"./src/utils/core.js":(e,t,s)=>{"use strict";function r(e,t){e&&e(t)}function o(e){return Object.fromEntries(Object.entries(e).map((([e,t])=>[t,e])))}function n(e){return e.replace(/[.*+?^${}()|[\]\\]/g,"\\$&")}function i(e){return"TypedArray"===e?.prototype?.__proto__?.constructor?.name}function a(e){return Number.isInteger(e)||"bigint"==typeof e}function l(e){return null==e||-1===e}function c(e){const t=[];let s=e;for(;Array.isArray(s);)t.push(s.length),s=s[0];return t}function d(e,t,s=void 0){const r=e[t];if(void 0!==r)return delete e[t],r;if(void 0===s)throw Error(`Key ${t} does not exist in object.`);return s}function u(...e){return Array.prototype.concat.apply([],e)}function _(...e){return e.reduce(((e,t)=>e.flatMap((e=>t.map((t=>[e,t]))))))}function p(e,t){return Math.abs((e+t)%(2*t)-t)}function m(e,t){const s=URL.createObjectURL(t),r=document.createElement("a");r.href=s,r.download=e,r.click(),r.remove(),URL.revokeObjectURL(s)}function h(e,t){return Object.assign({},...t.map((t=>{if(void 0!==e[t])return{[t]:e[t]}})))}function g(e){let t=0;for(const s of e)++t;return t}function f(e,t){let s=0;for(const r of e)r===t&&++s;return s}s.r(t),s.d(t,{calculateDimensions:()=>c,calculateReflectOffset:()=>p,count:()=>f,dispatchCallback:()=>r,escapeRegExp:()=>n,isIntegralNumber:()=>a,isNullishDimension:()=>l,isTypedArray:()=>i,len:()=>g,mergeArrays:()=>u,pick:()=>h,pop:()=>d,product:()=>_,reverseDictionary:()=>o,saveBlob:()=>m})},"./src/utils/data-structures.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{CharTrie:()=>o,DictionarySplitter:()=>l,LRUCache:()=>c,PriorityQueue:()=>r,TokenLattice:()=>i});class r{constructor(e=(e,t)=>e>t,t=1/0){this._heap=[],this._comparator=e,this._maxSize=t}get size(){return this._heap.length}isEmpty(){return 0===this.size}peek(){return this._heap[0]}push(...e){return this.extend(e)}extend(e){for(const t of e)if(this.size<this._maxSize)this._heap.push(t),this._siftUp();else{const e=this._smallest();this._comparator(t,this._heap[e])&&(this._heap[e]=t,this._siftUpFrom(e))}return this.size}pop(){const e=this.peek(),t=this.size-1;return t>0&&this._swap(0,t),this._heap.pop(),this._siftDown(),e}replace(e){const t=this.peek();return this._heap[0]=e,this._siftDown(),t}_parent(e){return(e+1>>>1)-1}_left(e){return 1+(e<<1)}_right(e){return e+1<<1}_greater(e,t){return this._comparator(this._heap[e],this._heap[t])}_swap(e,t){const s=this._heap[e];this._heap[e]=this._heap[t],this._heap[t]=s}_siftUp(){this._siftUpFrom(this.size-1)}_siftUpFrom(e){for(;e>0&&this._greater(e,this._parent(e));)this._swap(e,this._parent(e)),e=this._parent(e)}_siftDown(){let e=0;for(;this._left(e)<this.size&&this._greater(this._left(e),e)||this._right(e)<this.size&&this._greater(this._right(e),e);){const t=this._right(e)<this.size&&this._greater(this._right(e),this._left(e))?this._right(e):this._left(e);this._swap(e,t),e=t}}_smallest(){return 2**Math.floor(Math.log2(this.size))-1}}class o{constructor(){this.root=n.default()}extend(e){for(const t of e)this.push(t)}push(e){let t=this.root;for(const s of e){let e=t.children.get(s);void 0===e&&(e=n.default(),t.children.set(s,e)),t=e}t.isLeaf=!0}*commonPrefixSearch(e){let t=this.root;if(void 0===t)return;let s="";for(const r of e){if(s+=r,t=t.children.get(r),void 0===t)return;t.isLeaf&&(yield s)}}}class n{constructor(e,t){this.isLeaf=e,this.children=t}static default(){return new n(!1,new Map)}}class i{constructor(e,t,s){this.chars=Array.from(e),this.len=this.chars.length,this.bosTokenId=t,this.eosTokenId=s,this.nodes=[],this.beginNodes=Array.from({length:this.len+1},(()=>[])),this.endNodes=Array.from({length:this.len+1},(()=>[]));const r=new a(this.bosTokenId,0,0,0,0),o=new a(this.eosTokenId,1,this.len,0,0);this.nodes.push(r.clone()),this.nodes.push(o.clone()),this.beginNodes[this.len].push(o),this.endNodes[0].push(r)}insert(e,t,s,r){const o=this.nodes.length,n=new a(r,o,e,t,s);this.beginNodes[e].push(n),this.endNodes[e+t].push(n),this.nodes.push(n)}viterbi(){const e=this.len;let t=0;for(;t<=e;){if(0==this.beginNodes[t].length)return[];for(let e of this.beginNodes[t]){e.prev=null;let s=0,r=null;for(let o of this.endNodes[t]){const t=o.backtraceScore+e.score;(null===r||t>s)&&(r=o.clone(),s=t)}if(null===r)return[];e.prev=r,e.backtraceScore=s}++t}const s=[],r=this.beginNodes[e][0].prev;if(null===r)return[];let o=r.clone();for(;null!==o.prev;){s.push(o.clone());const e=o.clone();o=e.prev.clone()}return s.reverse(),s}piece(e){return this.chars.slice(e.pos,e.pos+e.length).join("")}tokens(){return this.viterbi().map((e=>this.piece(e)))}tokenIds(){return this.viterbi().map((e=>e.tokenId))}}class a{constructor(e,t,s,r,o){this.tokenId=e,this.nodeId=t,this.pos=s,this.length=r,this.score=o,this.prev=null,this.backtraceScore=0}clone(){const e=new a(this.tokenId,this.nodeId,this.pos,this.length,this.score);return e.prev=this.prev,e.backtraceScore=this.backtraceScore,e}}class l{constructor(e){this.trie=this._buildTrie(e)}_buildTrie(e){const t=Object.create(null);for(const s of e){let e=t;for(let t=0;t<s.length;++t)e=e[s[t]]??=Object.create(null);e.end=s}return t}split(e){const t=[],s=e.length;let r=0,o=0;for(;o<s;){let n=this.trie,i=null,a=o;for(;a<s&&(n=n[e[a]]);)n.end&&(i=n.end),++a;i?(o>r&&t.push(e.slice(r,o)),t.push(i),o+=i.length,r=o):++o}return r<s&&t.push(e.slice(r)),t}}class c{constructor(e){this.capacity=e,this.cache=new Map}get(e){if(!this.cache.has(e))return;const t=this.cache.get(e);return this.cache.delete(e),this.cache.set(e,t),t}put(e,t){this.cache.has(e)&&this.cache.delete(e),this.cache.set(e,t),this.cache.size>this.capacity&&this.cache.delete(this.cache.keys().next().value)}clear(){this.cache.clear()}}},"./src/utils/devices.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DEVICE_TYPES:()=>r});const r=Object.freeze({auto:"auto",gpu:"gpu",cpu:"cpu",wasm:"wasm",webgpu:"webgpu",cuda:"cuda",dml:"dml",webnn:"webnn","webnn-npu":"webnn-npu","webnn-gpu":"webnn-gpu","webnn-cpu":"webnn-cpu"})},"./src/utils/dtypes.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DATA_TYPES:()=>i,DEFAULT_DEVICE_DTYPE_MAPPING:()=>a,DEFAULT_DTYPE_SUFFIX_MAPPING:()=>l,isWebGpuFp16Supported:()=>n});var r=s("./src/env.js"),o=s("./src/utils/devices.js");const n=function(){let e;return async function(){if(void 0===e)if(r.apis.IS_WEBGPU_AVAILABLE)try{const t=await navigator.gpu.requestAdapter();e=t.features.has("shader-f16")}catch(t){e=!1}else e=!1;return e}}(),i=Object.freeze({auto:"auto",fp32:"fp32",fp16:"fp16",q8:"q8",int8:"int8",uint8:"uint8",q4:"q4",bnb4:"bnb4",q4f16:"q4f16"}),a=Object.freeze({[o.DEVICE_TYPES.wasm]:i.q8}),l=Object.freeze({[i.fp32]:"",[i.fp16]:"_fp16",[i.int8]:"_int8",[i.uint8]:"_uint8",[i.q8]:"_quantized",[i.q4]:"_q4",[i.q4f16]:"_q4f16",[i.bnb4]:"_bnb4"})},"./src/utils/generic.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Callable:()=>r});const r=class{constructor(){let e=function(...t){return e._call(...t)};return Object.setPrototypeOf(e,new.target.prototype)}_call(...e){throw Error("Must implement _call method in subclass")}}},"./src/utils/hub.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MAX_EXTERNAL_DATA_CHUNKS:()=>a,getFile:()=>_,getModelFile:()=>h,getModelJSON:()=>g});var r=s("fs"),o=s("path"),n=s("./src/env.js"),i=s("./src/utils/core.js");const a=100,l={txt:"text/plain",html:"text/html",css:"text/css",js:"text/javascript",json:"application/json",png:"image/png",jpg:"image/jpeg",jpeg:"image/jpeg",gif:"image/gif"};class c{constructor(e){if(this.filePath=e,this.headers=new Headers,this.exists=r.existsSync(e),this.exists){this.status=200,this.statusText="OK";let t=r.statSync(e);this.headers.set("content-length",t.size.toString()),this.updateContentType();const s=r.createReadStream(e);this.body=new ReadableStream({start(e){s.on("data",(t=>e.enqueue(t))),s.on("end",(()=>e.close())),s.on("error",(t=>e.error(t)))},cancel(){s.destroy()}})}else this.status=404,this.statusText="Not Found",this.body=null}updateContentType(){const e=this.filePath.toString().split(".").pop().toLowerCase();this.headers.set("content-type",l[e]??"application/octet-stream")}clone(){let e=new c(this.filePath);return e.exists=this.exists,e.status=this.status,e.statusText=this.statusText,e.headers=new Headers(this.headers),e}async arrayBuffer(){return(await r.promises.readFile(this.filePath)).buffer}async blob(){const e=await r.promises.readFile(this.filePath);return new Blob([e],{type:this.headers.get("content-type")})}async text(){return await r.promises.readFile(this.filePath,"utf8")}async json(){return JSON.parse(await this.text())}}function d(e,t=null,s=null){let r;try{r=new URL(e)}catch(e){return!1}return!(t&&!t.includes(r.protocol))&&!(s&&!s.includes(r.hostname))}const u=/^(\b[\w\-.]+\b\/)?\b[\w\-.]{1,96}\b$/;async function _(e){if(n.env.useFS&&!d(e,["http:","https:","blob:"]))return new c(e.toString());if("undefined"!=typeof process&&"node"===process?.release?.name){const t=!!process.env?.TESTING_REMOTELY,s=n.env.version,r=new Headers;r.set("User-Agent",`transformers.js/${s}; is_ci/${t};`);if(d(e,["http:","https:"],["huggingface.co","hf.co"])){const e=process.env?.HF_TOKEN??process.env?.HF_ACCESS_TOKEN;e&&r.set("Authorization",`Bearer ${e}`)}return fetch(e,{headers:r})}return fetch(e)}const p={400:"Bad request error occurred while trying to load file",401:"Unauthorized access to file",403:"Forbidden access to file",404:"Could not locate file",408:"Request timeout error occurred while trying to load file",500:"Internal server error error occurred while trying to load file",502:"Bad gateway error occurred while trying to load file",503:"Service unavailable error occurred while trying to load file",504:"Gateway timeout error occurred while trying to load file"};class m{constructor(e){this.path=e}async match(e){let t=o.join(this.path,e),s=new c(t);return s.exists?s:void 0}async put(e,t,s=void 0){let n=o.join(this.path,e);try{const e=t.headers.get("Content-Length"),i=parseInt(e??"0");let a=0;await r.promises.mkdir(o.dirname(n),{recursive:!0});const l=r.createWriteStream(n),c=t.body.getReader();for(;;){const{done:e,value:t}=await c.read();if(e)break;await new Promise(((e,s)=>{l.write(t,(t=>{t?s(t):e()}))})),a+=t.length;const r=i?a/i*100:0;s?.({progress:r,loaded:a,total:i})}l.close()}catch(e){try{await r.promises.unlink(n)}catch{}throw e}}}async function h(e,t,s=!0,r={},o=!1){if(!n.env.allowLocalModels){if(r.local_files_only)throw Error("Invalid configuration detected: local models are disabled (`env.allowLocalModels=false`) but you have requested to only use local models (`local_files_only=true`).");if(!n.env.allowRemoteModels)throw Error("Invalid configuration detected: both local and remote models are disabled. Fix by setting `env.allowLocalModels` or `env.allowRemoteModels` to `true`.")}let a;if((0,i.dispatchCallback)(r.progress_callback,{status:"initiate",name:e,file:t}),!a&&n.env.useCustomCache){if(!n.env.customCache)throw Error("`env.useCustomCache=true`, but `env.customCache` is not defined.");if(!n.env.customCache.match||!n.env.customCache.put)throw new Error("`env.customCache` must be an object which implements the `match` and `put` functions of the Web Cache API. For more information, see https://developer.mozilla.org/en-US/docs/Web/API/Cache");a=n.env.customCache}if(!a&&n.env.useBrowserCache){if("undefined"==typeof caches)throw Error("Browser cache is not available in this environment.");try{a=await caches.open("transformers-cache")}catch(e){console.warn("An error occurred while opening the browser cache:",e)}}if(!a&&n.env.useFSCache){if(!n.apis.IS_FS_AVAILABLE)throw Error("File System Cache is not available in this environment.");a=new m(r.cache_dir??n.env.cacheDir)}const l=r.revision??"main",h=f(e,t),g=(w=e,!(!u.test(w)||w.includes("..")||w.includes("--")||w.endsWith(".git")||w.endsWith(".ipynb")));var w;const M=g?f(n.env.localModelPath,h):h,x=f(n.env.remoteHost,n.env.remotePathTemplate.replaceAll("{model}",e).replaceAll("{revision}",encodeURIComponent(l)),t);let b;const k=a instanceof m?"main"===l?h:f(e,l,t):x;let y,v=!1;a&&(y=await async function(e,...t){for(let s of t)try{let t=await e.match(s);if(t)return t}catch(e){continue}}(a,M,k));const T=void 0!==y;if(void 0===y){if(n.env.allowLocalModels){if(d(h,["http:","https:"])){if(r.local_files_only)throw new Error(`\`local_files_only=true\`, but attempted to load a remote file from: ${h}.`);if(!n.env.allowRemoteModels)throw new Error(`\`env.allowRemoteModels=false\`, but attempted to load a remote file from: ${h}.`)}else try{y=await _(M),b=M}catch(e){console.warn(`Unable to load from local path "${M}": "${e}"`)}}if(void 0===y||404===y.status){if(r.local_files_only||!n.env.allowRemoteModels){if(s)throw Error(`\`local_files_only=true\` or \`env.allowRemoteModels=false\` and file was not found locally at "${M}".`);return null}if(!g)throw Error(`Local file missing at "${M}" and download aborted due to invalid model ID "${e}".`);if(y=await _(x),200!==y.status)return function(e,t,s){if(!s)return null;const r=p[e]??`Error (${e}) occurred while trying to load file`;throw Error(`${r}: "${t}".`)}(y.status,x,s);b=k}v=a&&"undefined"!=typeof Response&&y instanceof Response&&200===y.status}let P;if((0,i.dispatchCallback)(r.progress_callback,{status:"download",name:e,file:t}),!n.apis.IS_NODE_ENV||!o){let s;r.progress_callback?T&&"undefined"!=typeof navigator&&/firefox/i.test(navigator.userAgent)?(s=new Uint8Array(await y.arrayBuffer()),(0,i.dispatchCallback)(r.progress_callback,{status:"progress",name:e,file:t,progress:100,loaded:s.length,total:s.length})):s=await async function(e,t){const s=e.headers.get("Content-Length");null===s&&console.warn("Unable to determine content-length from response headers. Will expand buffer when needed.");let r=parseInt(s??"0"),o=new Uint8Array(r),n=0;const i=e.body.getReader();async function a(){const{done:e,value:s}=await i.read();if(e)return;const l=n+s.length;if(l>r){r=l;const e=new Uint8Array(r);e.set(o),o=e}o.set(s,n),n=l;return t({progress:n/r*100,loaded:n,total:r}),a()}return await a(),o}(y,(s=>{(0,i.dispatchCallback)(r.progress_callback,{status:"progress",name:e,file:t,...s})})):s=new Uint8Array(await y.arrayBuffer()),P=s}if(v&&b&&void 0===await a.match(b)&&(P?await a.put(b,new Response(P,{headers:y.headers})).catch((e=>{console.warn(`Unable to add response to browser cache: ${e}.`)})):await a.put(b,y,r.progress_callback)),(0,i.dispatchCallback)(r.progress_callback,{status:"done",name:e,file:t}),P){if(!n.apis.IS_NODE_ENV&&o)throw new Error("Cannot return path in a browser environment.");return P}if(y instanceof c)return y.filePath;const F=await(a?.match(b));if(F instanceof c)return F.filePath;if(F instanceof Response)return new Uint8Array(await F.arrayBuffer());if("string"==typeof F)return F;throw new Error("Unable to get model file path or buffer.")}async function g(e,t,s=!0,r={}){const o=await h(e,t,s,r,!1);if(null===o)return{};const n=new TextDecoder("utf-8").decode(o);return JSON.parse(n)}function f(...e){return(e=e.map(((t,s)=>(s&&(t=t.replace(new RegExp("^/"),"")),s!==e.length-1&&(t=t.replace(new RegExp("/$"),"")),t)))).join("/")}},"./src/utils/image.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{RawImage:()=>m,load_image:()=>h});var r=s("./src/utils/core.js"),o=s("./src/utils/hub.js"),n=s("./src/env.js"),i=s("./src/utils/tensor.js"),a=s("sharp");let l,c,d;const u=n.apis.IS_BROWSER_ENV||n.apis.IS_WEBWORKER_ENV;if(u)l=(e,t)=>{if(!self.OffscreenCanvas)throw new Error("OffscreenCanvas not supported by this browser.");return new self.OffscreenCanvas(e,t)},d=self.createImageBitmap,c=self.ImageData;else{if(!a)throw new Error("Unable to load image processing library.");d=async e=>{const t=(await e.metadata()).channels,{data:s,info:r}=await e.rotate().raw().toBuffer({resolveWithObject:!0}),o=new m(new Uint8ClampedArray(s),r.width,r.height,r.channels);return void 0!==t&&t!==r.channels&&o.convert(t),o}}const _={0:"nearest",1:"lanczos",2:"bilinear",3:"bicubic",4:"box",5:"hamming"},p=new Map([["png","image/png"],["jpg","image/jpeg"],["jpeg","image/jpeg"],["gif","image/gif"]]);class m{constructor(e,t,s,r){this.data=e,this.width=t,this.height=s,this.channels=r}get size(){return[this.width,this.height]}static async read(e){if(e instanceof m)return e;if("string"==typeof e||e instanceof URL)return await this.fromURL(e);if(e instanceof Blob)return await this.fromBlob(e);if("undefined"!=typeof HTMLCanvasElement&&e instanceof HTMLCanvasElement||"undefined"!=typeof OffscreenCanvas&&e instanceof OffscreenCanvas)return this.fromCanvas(e);throw new Error("Unsupported input type: "+typeof e)}static fromCanvas(e){if(!u)throw new Error("fromCanvas() is only supported in browser environments.");const t=e.getContext("2d").getImageData(0,0,e.width,e.height).data;return new m(t,e.width,e.height,4)}static async fromURL(e){const t=await(0,o.getFile)(e);if(200!==t.status)throw new Error(`Unable to read image from "${e}" (${t.status} ${t.statusText})`);const s=await t.blob();return this.fromBlob(s)}static async fromBlob(e){if(u){const t=await d(e),s=l(t.width,t.height).getContext("2d");return s.drawImage(t,0,0),new this(s.getImageData(0,0,t.width,t.height).data,t.width,t.height,4)}{const t=a(await e.arrayBuffer());return await d(t)}}static fromTensor(e,t="CHW"){if(3!==e.dims.length)throw new Error(`Tensor should have 3 dimensions, but has ${e.dims.length} dimensions.`);if("CHW"===t)e=e.transpose(1,2,0);else if("HWC"!==t)throw new Error(`Unsupported channel format: ${t}`);if(!(e.data instanceof Uint8ClampedArray||e.data instanceof Uint8Array))throw new Error(`Unsupported tensor type: ${e.type}`);switch(e.dims[2]){case 1:case 2:case 3:case 4:return new m(e.data,e.dims[1],e.dims[0],e.dims[2]);default:throw new Error(`Unsupported number of channels: ${e.dims[2]}`)}}grayscale(){if(1===this.channels)return this;const e=new Uint8ClampedArray(this.width*this.height*1);switch(this.channels){case 3:case 4:for(let t=0,s=0;t<this.data.length;t+=this.channels){const r=this.data[t],o=this.data[t+1],n=this.data[t+2];e[s++]=Math.round(.2989*r+.587*o+.114*n)}break;default:throw new Error(`Conversion failed due to unsupported number of channels: ${this.channels}`)}return this._update(e,this.width,this.height,1)}rgb(){if(3===this.channels)return this;const e=new Uint8ClampedArray(this.width*this.height*3);switch(this.channels){case 1:for(let t=0,s=0;t<this.data.length;++t)e[s++]=this.data[t],e[s++]=this.data[t],e[s++]=this.data[t];break;case 4:for(let t=0,s=0;t<this.data.length;t+=4)e[s++]=this.data[t],e[s++]=this.data[t+1],e[s++]=this.data[t+2];break;default:throw new Error(`Conversion failed due to unsupported number of channels: ${this.channels}`)}return this._update(e,this.width,this.height,3)}rgba(){if(4===this.channels)return this;const e=new Uint8ClampedArray(this.width*this.height*4);switch(this.channels){case 1:for(let t=0,s=0;t<this.data.length;++t)e[s++]=this.data[t],e[s++]=this.data[t],e[s++]=this.data[t],e[s++]=255;break;case 3:for(let t=0,s=0;t<this.data.length;t+=3)e[s++]=this.data[t],e[s++]=this.data[t+1],e[s++]=this.data[t+2],e[s++]=255;break;default:throw new Error(`Conversion failed due to unsupported number of channels: ${this.channels}`)}return this._update(e,this.width,this.height,4)}putAlpha(e){if(e.width!==this.width||e.height!==this.height)throw new Error(`Expected mask size to be ${this.width}x${this.height}, but got ${e.width}x${e.height}`);if(1!==e.channels)throw new Error(`Expected mask to have 1 channel, but got ${e.channels}`);const t=this.data,s=e.data,r=this.width*this.height;if(3===this.channels){const e=new Uint8ClampedArray(4*r);for(let o=0,n=0,i=0;o<r;++o)e[i++]=t[n++],e[i++]=t[n++],e[i++]=t[n++],e[i++]=s[o];return this._update(e,this.width,this.height,4)}if(4===this.channels){for(let e=0;e<r;++e)t[4*e+3]=s[e];return this}throw new Error(`Expected image to have 3 or 4 channels, but got ${this.channels}`)}async resize(e,t,{resample:s=2}={}){if(this.width===e&&this.height===t)return this;let o=_[s]??s;const n=(0,r.isNullishDimension)(e),i=(0,r.isNullishDimension)(t);if(n&&i)return this;if(n?e=t/this.height*this.width:i&&(t=e/this.width*this.height),u){const s=this.channels,r=this.toCanvas(),o=l(e,t).getContext("2d");o.drawImage(r,0,0,e,t);return new m(o.getImageData(0,0,e,t).data,e,t,4).convert(s)}{let s=this.toSharp();switch(o){case"box":case"hamming":"box"!==o&&"hamming"!==o||(console.warn(`Resampling method ${o} is not yet supported. Using bilinear instead.`),o="bilinear");case"nearest":case"bilinear":case"bicubic":s=s.affine([e/this.width,0,0,t/this.height],{interpolator:o});break;case"lanczos":s=s.resize({width:e,height:t,fit:"fill",kernel:"lanczos3"});break;default:throw new Error(`Resampling method ${o} is not supported.`)}return await d(s)}}async pad([e,t,s,r]){if(e=Math.max(e,0),t=Math.max(t,0),s=Math.max(s,0),r=Math.max(r,0),0===e&&0===t&&0===s&&0===r)return this;if(u){const o=this.channels,n=this.toCanvas(),i=this.width+e+t,a=this.height+s+r,c=l(i,a).getContext("2d");c.drawImage(n,0,0,this.width,this.height,e,s,this.width,this.height);return new m(c.getImageData(0,0,i,a).data,i,a,4).convert(o)}{const o=this.toSharp().extend({left:e,right:t,top:s,bottom:r});return await d(o)}}async crop([e,t,s,r]){if(e=Math.max(e,0),t=Math.max(t,0),s=Math.min(s,this.width-1),r=Math.min(r,this.height-1),0===e&&0===t&&s===this.width-1&&r===this.height-1)return this;const o=s-e+1,n=r-t+1;if(u){const s=this.channels,r=this.toCanvas(),i=l(o,n).getContext("2d");i.drawImage(r,e,t,o,n,0,0,o,n);return new m(i.getImageData(0,0,o,n).data,o,n,4).convert(s)}{const s=this.toSharp().extract({left:e,top:t,width:o,height:n});return await d(s)}}async center_crop(e,t){if(this.width===e&&this.height===t)return this;const s=(this.width-e)/2,r=(this.height-t)/2;if(u){const o=this.channels,n=this.toCanvas(),i=l(e,t).getContext("2d");let a=0,c=0,d=0,u=0;s>=0?a=s:d=-s,r>=0?c=r:u=-r,i.drawImage(n,a,c,e,t,d,u,e,t);return new m(i.getImageData(0,0,e,t).data,e,t,4).convert(o)}{let o=this.toSharp();if(s>=0&&r>=0)o=o.extract({left:Math.floor(s),top:Math.floor(r),width:e,height:t});else if(s<=0&&r<=0){const n=Math.floor(-r),i=Math.floor(-s);o=o.extend({top:n,left:i,right:e-this.width-i,bottom:t-this.height-n})}else{let n=[0,0],i=0;r<0?(n[0]=Math.floor(-r),n[1]=t-this.height-n[0]):i=Math.floor(r);let a=[0,0],l=0;s<0?(a[0]=Math.floor(-s),a[1]=e-this.width-a[0]):l=Math.floor(s),o=o.extend({top:n[0],bottom:n[1],left:a[0],right:a[1]}).extract({left:l,top:i,width:e,height:t})}return await d(o)}}async toBlob(e="image/png",t=1){if(!u)throw new Error("toBlob() is only supported in browser environments.");const s=this.toCanvas();return await s.convertToBlob({type:e,quality:t})}toTensor(e="CHW"){let t=new i.Tensor("uint8",new Uint8Array(this.data),[this.height,this.width,this.channels]);if("HWC"===e);else{if("CHW"!==e)throw new Error(`Unsupported channel format: ${e}`);t=t.permute(2,0,1)}return t}toCanvas(){if(!u)throw new Error("toCanvas() is only supported in browser environments.");const e=this.clone().rgba(),t=l(e.width,e.height),s=new c(e.data,e.width,e.height);return t.getContext("2d").putImageData(s,0,0),t}split(){const{data:e,width:t,height:s,channels:r}=this,o=e.constructor,n=e.length/r,i=Array.from({length:r},(()=>new o(n)));for(let t=0;t<n;++t){const s=r*t;for(let o=0;o<r;++o)i[o][t]=e[s+o]}return i.map((e=>new m(e,t,s,1)))}_update(e,t,s,r=null){return this.data=e,this.width=t,this.height=s,null!==r&&(this.channels=r),this}clone(){return new m(this.data.slice(),this.width,this.height,this.channels)}convert(e){if(this.channels===e)return this;switch(e){case 1:this.grayscale();break;case 3:this.rgb();break;case 4:this.rgba();break;default:throw new Error(`Conversion failed due to unsupported number of channels: ${this.channels}`)}return this}async save(e){if(!u){if(n.apis.IS_FS_AVAILABLE){const t=this.toSharp();return await t.toFile(e)}throw new Error("Unable to save the image because filesystem is disabled in this environment.")}{if(n.apis.IS_WEBWORKER_ENV)throw new Error("Unable to save an image from a Web Worker.");const t=e.split(".").pop().toLowerCase(),s=p.get(t)??"image/png",o=await this.toBlob(s);(0,r.saveBlob)(e,o)}}toSharp(){if(u)throw new Error("toSharp() is only supported in server-side environments.");return a(this.data,{raw:{width:this.width,height:this.height,channels:this.channels}})}}const h=m.read.bind(m)},"./src/utils/maths.js":(e,t,s)=>{"use strict";function r(e,[t,s,r],[o,n],i="bilinear",a=!1){const l=n/r,c=o/s,d=new e.constructor(o*n*t),u=s*r,_=o*n;for(let i=0;i<o;++i)for(let o=0;o<n;++o){const a=i*n+o,p=(o+.5)/l-.5,m=(i+.5)/c-.5;let h=Math.floor(p),g=Math.floor(m);const f=Math.min(h+1,r-1),w=Math.min(g+1,s-1);h=Math.max(h,0),g=Math.max(g,0);const M=p-h,x=m-g,b=(1-M)*(1-x),k=M*(1-x),y=(1-M)*x,v=M*x,T=g*r,P=w*r,F=T+h,C=T+f,S=P+h,A=P+f;for(let s=0;s<t;++s){const t=s*u;d[s*_+a]=b*e[t+F]+k*e[t+C]+y*e[t+S]+v*e[t+A]}}return d}function o(e,t,s){const r=new Array(s.length),o=new Array(s.length);for(let e=s.length-1,n=1;e>=0;--e)o[e]=n,r[e]=t[s[e]],n*=r[e];const n=s.map(((e,t)=>o[s.indexOf(t)])),i=new e.constructor(e.length);for(let s=0;s<e.length;++s){let r=0;for(let e=t.length-1,o=s;e>=0;--e)r+=o%t[e]*n[e],o=Math.floor(o/t[e]);i[r]=e[s]}return[i,r]}function n(e){const t=u(e)[0],s=e.map((e=>Math.exp(e-t))),r=s.reduce(((e,t)=>e+t),0);return s.map((e=>e/r))}function i(e){const t=u(e)[0];let s=0;for(let r=0;r<e.length;++r)s+=Math.exp(e[r]-t);const r=Math.log(s);return e.map((e=>e-t-r))}function a(e,t){let s=0;for(let r=0;r<e.length;++r)s+=e[r]*t[r];return s}function l(e,t){return a(e,t)/(c(e)*c(t))}function c(e){return Math.sqrt(e.reduce(((e,t)=>e+t*t),0))}function d(e){if(0===e.length)throw Error("Array must not be empty");let t=e[0],s=0;for(let r=1;r<e.length;++r)e[r]<t&&(t=e[r],s=r);return[t,s]}function u(e){if(0===e.length)throw Error("Array must not be empty");let t=e[0],s=0;for(let r=1;r<e.length;++r)e[r]>t&&(t=e[r],s=r);return[t,s]}function _(e){return e>0&&!(e&e-1)}s.r(t),s.d(t,{FFT:()=>h,bankers_round:()=>w,cos_sim:()=>l,dot:()=>a,dynamic_time_warping:()=>M,interpolate_data:()=>r,log_softmax:()=>i,magnitude:()=>c,max:()=>u,medianFilter:()=>g,min:()=>d,permute_data:()=>o,round:()=>f,softmax:()=>n});class p{constructor(e){if(this.size=0|e,this.size<=1||!_(this.size))throw new Error("FFT size must be a power of two larger than 1");this._csize=e<<1,this.table=new Float64Array(2*this.size);for(let e=0;e<this.table.length;e+=2){const t=Math.PI*e/this.size;this.table[e]=Math.cos(t),this.table[e+1]=-Math.sin(t)}let t=0;for(let e=1;this.size>e;e<<=1)++t;this._width=t%2==0?t-1:t,this._bitrev=new Int32Array(1<<this._width);for(let e=0;e<this._bitrev.length;++e){this._bitrev[e]=0;for(let t=0;t<this._width;t+=2){const s=this._width-t-2;this._bitrev[e]|=(e>>>t&3)<<s}}}createComplexArray(){return new Float64Array(this._csize)}fromComplexArray(e,t){const s=t||new Array(e.length>>>1);for(let t=0;t<e.length;t+=2)s[t>>>1]=e[t];return s}toComplexArray(e,t){const s=t||this.createComplexArray();for(let t=0;t<s.length;t+=2)s[t]=e[t>>>1],s[t+1]=0;return s}transform(e,t){if(e===t)throw new Error("Input and output buffers must be different");this._transform4(e,t,1)}realTransform(e,t){if(e===t)throw new Error("Input and output buffers must be different");this._realTransform4(e,t,1)}inverseTransform(e,t){if(e===t)throw new Error("Input and output buffers must be different");this._transform4(e,t,-1);for(let t=0;t<e.length;++t)e[t]/=this.size}_transform4(e,t,s){const r=this._csize;let o,n,i=1<<this._width,a=r/i<<1;const l=this._bitrev;if(4===a)for(o=0,n=0;o<r;o+=a,++n){const s=l[n];this._singleTransform2(t,e,o,s,i)}else for(o=0,n=0;o<r;o+=a,++n){const r=l[n];this._singleTransform4(t,e,o,r,i,s)}const c=this.table;for(i>>=2;i>=2;i>>=2){a=r/i<<1;const t=a>>>2;for(o=0;o<r;o+=a){const r=o+t-1;for(let n=o,a=0;n<r;n+=2,a+=i){const r=n,o=r+t,i=o+t,l=i+t,d=e[r],u=e[r+1],_=e[o],p=e[o+1],m=e[i],h=e[i+1],g=e[l],f=e[l+1],w=c[a],M=s*c[a+1],x=_*w-p*M,b=_*M+p*w,k=c[2*a],y=s*c[2*a+1],v=m*k-h*y,T=m*y+h*k,P=c[3*a],F=s*c[3*a+1],C=g*P-f*F,S=g*F+f*P,A=d+v,E=u+T,L=d-v,I=u-T,z=x+C,j=b+S,D=s*(x-C),O=s*(b-S);e[r]=A+z,e[r+1]=E+j,e[o]=L+O,e[o+1]=I-D,e[i]=A-z,e[i+1]=E-j,e[l]=L-O,e[l+1]=I+D}}}}_singleTransform2(e,t,s,r,o){const n=e[r],i=e[r+1],a=e[r+o],l=e[r+o+1];t[s]=n+a,t[s+1]=i+l,t[s+2]=n-a,t[s+3]=i-l}_singleTransform4(e,t,s,r,o,n){const i=2*o,a=3*o,l=e[r],c=e[r+1],d=e[r+o],u=e[r+o+1],_=e[r+i],p=e[r+i+1],m=e[r+a],h=e[r+a+1],g=l+_,f=c+p,w=l-_,M=c-p,x=d+m,b=u+h,k=n*(d-m),y=n*(u-h);t[s]=g+x,t[s+1]=f+b,t[s+2]=w+y,t[s+3]=M-k,t[s+4]=g-x,t[s+5]=f-b,t[s+6]=w-y,t[s+7]=M+k}_realTransform4(e,t,s){const r=this._csize;let o,n,i=1<<this._width,a=r/i<<1;const l=this._bitrev;if(4===a)for(o=0,n=0;o<r;o+=a,++n){const s=l[n];this._singleRealTransform2(t,e,o,s>>>1,i>>>1)}else for(o=0,n=0;o<r;o+=a,++n){const r=l[n];this._singleRealTransform4(t,e,o,r>>>1,i>>>1,s)}const c=this.table;for(i>>=2;i>=2;i>>=2){a=r/i<<1;const t=a>>>1,n=t>>>1,l=n>>>1;for(o=0;o<r;o+=a)for(let r=0,a=0;r<=l;r+=2,a+=i){const i=o+r,d=i+n,u=d+n,_=u+n,p=e[i],m=e[i+1],h=e[d],g=e[d+1],f=e[u],w=e[u+1],M=e[_],x=e[_+1],b=p,k=m,y=c[a],v=s*c[a+1],T=h*y-g*v,P=h*v+g*y,F=c[2*a],C=s*c[2*a+1],S=f*F-w*C,A=f*C+w*F,E=c[3*a],L=s*c[3*a+1],I=M*E-x*L,z=M*L+x*E,j=b+S,D=k+A,O=b-S,V=k-A,N=T+I,B=P+z,R=s*(T-I),G=s*(P-z);if(e[i]=j+N,e[i+1]=D+B,e[d]=O+G,e[d+1]=V-R,0===r){e[u]=j-N,e[u+1]=D-B;continue}if(r===l)continue;const q=o+n-r,$=o+t-r;e[q]=O-s*G,e[q+1]=-V-s*R,e[$]=j-s*N,e[$+1]=s*B-D}}const d=r>>>1;for(let t=2;t<d;t+=2)e[r-t]=e[t],e[r-t+1]=-e[t+1]}_singleRealTransform2(e,t,s,r,o){const n=e[r],i=e[r+o];t[s]=n+i,t[s+1]=0,t[s+2]=n-i,t[s+3]=0}_singleRealTransform4(e,t,s,r,o,n){const i=2*o,a=3*o,l=e[r],c=e[r+o],d=e[r+i],u=e[r+a],_=l+d,p=l-d,m=c+u,h=n*(c-u);t[s]=_+m,t[s+1]=0,t[s+2]=p,t[s+3]=-h,t[s+4]=_-m,t[s+5]=0,t[s+6]=p,t[s+7]=h}}class m{constructor(e){const t=2*(e-1),s=2*(2*e-1),r=2**Math.ceil(Math.log2(s));this.bufferSize=r,this._a=t;const o=new Float64Array(s),n=new Float64Array(r);this._chirpBuffer=new Float64Array(r),this._buffer1=new Float64Array(r),this._buffer2=new Float64Array(r),this._outBuffer1=new Float64Array(r),this._outBuffer2=new Float64Array(r);const i=-2*Math.PI/e,a=Math.cos(i),l=Math.sin(i);for(let t=0;t<s>>1;++t){const s=(t+1-e)**2/2,r=Math.sqrt(a**2+l**2)**s,i=s*Math.atan2(l,a),c=2*t;o[c]=r*Math.cos(i),o[c+1]=r*Math.sin(i),n[c]=o[c],n[c+1]=-o[c+1]}this._slicedChirpBuffer=o.subarray(t,s),this._f=new p(r>>1),this._f.transform(this._chirpBuffer,n)}_transform(e,t,s){const r=this._buffer1,o=this._buffer2,n=this._outBuffer1,i=this._outBuffer2,a=this._chirpBuffer,l=this._slicedChirpBuffer,c=this._a;if(s)for(let e=0;e<l.length;e+=2){const s=e+1,o=t[e>>1];r[e]=o*l[e],r[s]=o*l[s]}else for(let e=0;e<l.length;e+=2){const s=e+1;r[e]=t[e]*l[e]-t[s]*l[s],r[s]=t[e]*l[s]+t[s]*l[e]}this._f.transform(n,r);for(let e=0;e<a.length;e+=2){const t=e+1;o[e]=n[e]*a[e]-n[t]*a[t],o[t]=n[e]*a[t]+n[t]*a[e]}this._f.inverseTransform(i,o);for(let t=0;t<i.length;t+=2){const s=i[t+c],r=i[t+c+1],o=l[t],n=l[t+1];e[t]=s*o-r*n,e[t+1]=s*n+r*o}}transform(e,t){this._transform(e,t,!1)}realTransform(e,t){this._transform(e,t,!0)}}class h{constructor(e){this.fft_length=e,this.isPowerOfTwo=_(e),this.isPowerOfTwo?(this.fft=new p(e),this.outputBufferSize=2*e):(this.fft=new m(e),this.outputBufferSize=this.fft.bufferSize)}realTransform(e,t){this.fft.realTransform(e,t)}transform(e,t){this.fft.transform(e,t)}}function g(e,t){if(t%2==0||t<=0)throw new Error("Window size must be a positive odd number");const s=new e.constructor(e.length),r=new e.constructor(t),o=Math.floor(t/2);for(let t=0;t<e.length;++t){let n=0;for(let s=-o;s<=o;++s){let o=t+s;o<0?o=Math.abs(o):o>=e.length&&(o=2*(e.length-1)-o),r[n++]=e[o]}r.sort(),s[t]=r[o]}return s}function f(e,t){const s=Math.pow(10,t);return Math.round(e*s)/s}function w(e){const t=Math.round(e);return Math.abs(e)%1==.5?t%2==0?t:t-1:t}function M(e){const t=e.length,s=e[0].length,r=[t+1,s+1],o=Array.from({length:r[0]},(()=>Array(r[1]).fill(1/0)));o[0][0]=0;const n=Array.from({length:r[0]},(()=>Array(r[1]).fill(-1)));for(let t=1;t<r[1];++t)for(let s=1;s<r[0];++s){const r=o[s-1][t-1],i=o[s-1][t],a=o[s][t-1];let l,c;r<i&&r<a?(l=r,c=0):i<r&&i<a?(l=i,c=1):(l=a,c=2),o[s][t]=e[s-1][t-1]+l,n[s][t]=c}for(let e=0;e<r[1];++e)n[0][e]=2;for(let e=0;e<r[0];++e)n[e][0]=1;let i=t,a=s,l=[],c=[];for(;i>0||a>0;)switch(l.push(i-1),c.push(a-1),n[i][a]){case 0:--i,--a;break;case 1:--i;break;case 2:--a;break;default:throw new Error(`Internal error in dynamic time warping. Unexpected trace[${i}, ${a}]. Please file a bug report.`)}return l.reverse(),c.reverse(),[l,c]}},"./src/utils/tensor.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DataTypeMap:()=>i,Tensor:()=>a,cat:()=>b,full:()=>F,full_like:()=>C,interpolate:()=>c,interpolate_4d:()=>d,layer_norm:()=>f,matmul:()=>u,mean:()=>T,mean_pooling:()=>g,ones:()=>S,ones_like:()=>A,permute:()=>l,quantize_embeddings:()=>z,rand:()=>I,rfft:()=>_,slice:()=>h,stack:()=>k,std_mean:()=>v,topk:()=>p,zeros:()=>E,zeros_like:()=>L});var r=s("./src/utils/maths.js"),o=s("./src/backends/onnx.js"),n=s("./src/ops/registry.js");const i=Object.freeze({float32:Float32Array,float16:"undefined"!=typeof Float16Array?Float16Array:Uint16Array,float64:Float64Array,string:Array,int8:Int8Array,uint8:Uint8Array,int16:Int16Array,uint16:Uint16Array,int32:Int32Array,uint32:Uint32Array,int64:BigInt64Array,uint64:BigUint64Array,bool:Uint8Array,uint4:Uint8Array,int4:Int8Array});class a{get dims(){return this.ort_tensor.dims}set dims(e){this.ort_tensor.dims=e}get type(){return this.ort_tensor.type}get data(){return this.ort_tensor.data}get size(){return this.ort_tensor.size}get location(){return this.ort_tensor.location}ort_tensor;constructor(...e){return(0,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 s=Number(t);if(Number.isInteger(s))return e._getitem(s)}return e[t]},set:(e,t,s)=>e[t]=s})}dispose(){this.ort_tensor.dispose()}*[Symbol.iterator](){const[e,...t]=this.dims;if(t.length>0){const s=t.reduce(((e,t)=>e*t));for(let r=0;r<e;++r)yield this._subarray(r,s,t)}else yield*this.data}_getitem(e){const[t,...s]=this.dims;if(e=x(e,t),s.length>0){const t=s.reduce(((e,t)=>e*t));return this._subarray(e,t,s)}return new a(this.type,[this.data[e]],s)}indexOf(e){const t=this.data;for(let s=0;s<t.length;++s)if(t[s]==e)return s;return-1}_subarray(e,t,s){const r=e*t,o=(e+1)*t,n="subarray"in this.data?this.data.subarray(r,o):this.data.slice(r,o);return new a(this.type,n,s)}item(){const e=this.data;if(1!==e.length)throw new Error(`a Tensor with ${e.length} elements cannot be converted to Scalar`);return e[0]}tolist(){return function(e,t){const s=e.length,r=t.reduce(((e,t)=>e*t));if(s!==r)throw Error(`cannot reshape array of size ${s} into shape (${t})`);let o=e;for(let e=t.length-1;e>=0;e--)o=o.reduce(((s,r)=>{let o=s[s.length-1];return o.length<t[e]?o.push(r):s.push([r]),s}),[[]]);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 s=0;s<t.length;++s)t[s]=e(t[s],s,t);return this}mul(e){return this.clone().mul_(e)}mul_(e){const t=this.data;for(let s=0;s<t.length;++s)t[s]*=e;return this}div(e){return this.clone().div_(e)}div_(e){const t=this.data;for(let s=0;s<t.length;++s)t[s]/=e;return this}add(e){return this.clone().add_(e)}add_(e){const t=this.data;for(let s=0;s<t.length;++s)t[s]+=e;return this}sub(e){return this.clone().sub_(e)}sub_(e){const t=this.data;for(let s=0;s<t.length;++s)t[s]-=e;return this}clone(){return new a(this.type,this.data.slice(),this.dims.slice())}slice(...e){const t=[],s=[];for(let r=0;r<this.dims.length;++r){let o=e[r];if(null==o)s.push([0,this.dims[r]]),t.push(this.dims[r]);else if("number"==typeof o)o=x(o,this.dims[r],r),s.push([o,o+1]);else{if(!Array.isArray(o)||2!==o.length)throw new Error(`Invalid slice: ${o}`);{let[e,n]=o;if(e=null===e?0:x(e,this.dims[r],r,!1),n=null===n?this.dims[r]:x(n,this.dims[r],r,!1),e>n)throw new Error(`Invalid slice: ${o}`);const i=[Math.max(e,0),Math.min(n,this.dims[r])];s.push(i),t.push(i[1]-i[0])}}}const r=s.map((([e,t])=>t-e)),o=r.reduce(((e,t)=>e*t)),n=this.data,i=new n.constructor(o),l=this.stride();for(let e=0;e<o;++e){let t=0;for(let o=r.length-1,n=e;o>=0;--o){const e=r[o];t+=(n%e+s[o][0])*l[o],n=Math.floor(n/e)}i[e]=n[t]}return new a(this.type,i,t)}permute(...e){return l(this,e)}transpose(...e){return this.permute(...e)}sum(e=null,t=!1){return this.norm(1,e,t)}norm(e="fro",t=null,s=!1){if("fro"===e)e=2;else if("string"==typeof e)throw Error(`Unsupported norm: ${e}`);const r=this.data,o=(t,s)=>t+s**e;if(null===t){const t=r.reduce(o,0)**(1/e);return new a(this.type,[t],[])}const[n,i,l]=y(o,this,t,s);if(1!==e)for(let t=0;t<i.length;++t)i[t]=i[t]**(1/e);return new a(n,i,l)}normalize_(e=2,t=1){t=x(t,this.dims.length);const s=this.norm(e,t,!0),r=this.data,o=s.data;for(let e=0;e<r.length;++e){let s=0;for(let r=this.dims.length-1,o=e,n=1;r>=0;--r){const e=this.dims[r];if(r!==t){s+=o%e*n,n*=this.dims[r]}o=Math.floor(o/e)}r[e]/=o[s]}return this}normalize(e=2,t=1){return this.clone().normalize_(e,t)}stride(){return function(e){const t=new Array(e.length);for(let s=e.length-1,r=1;s>=0;--s)t[s]=r,r*=e[s];return t}(this.dims)}squeeze(e=null){return new a(this.type,this.data,w(this.dims,e))}squeeze_(e=null){return this.dims=w(this.dims,e),this}unsqueeze(e=null){return new a(this.type,this.data,M(this.dims,e))}unsqueeze_(e=null){return this.dims=M(this.dims,e),this}flatten_(e=0,t=-1){t=(t+this.dims.length)%this.dims.length;let s=this.dims.slice(0,e),r=this.dims.slice(e,t+1),o=this.dims.slice(t+1);return this.dims=[...s,r.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 s=0;s<e.length;++s)if(-1===e[s]){if(-1!==t)throw new Error("Only one dimension can be inferred");t=s}const s=this.data;if(-1!==t){const r=e.reduce(((e,s,r)=>r!==t?e*s:e),1);e[t]=s.length/r}return new a(this.type,s,e)}neg_(){const e=this.data;for(let t=0;t<e.length;++t)e[t]=-e[t];return this}neg(){return this.clone().neg_()}gt(e){const t=new Uint8Array(this.data.length),s=this.data;for(let r=0;r<s.length;++r)t[r]=s[r]>e?1:0;return new a("bool",t,this.dims)}lt(e){const t=new Uint8Array(this.data.length),s=this.data;for(let r=0;r<s.length;++r)t[r]=s[r]<e?1:0;return new a("bool",t,this.dims)}clamp_(e,t){const s=this.data;for(let r=0;r<s.length;++r)s[r]=Math.min(Math.max(s[r],e),t);return this}clamp(e,t){return this.clone().clamp_(e,t)}round_(){const e=this.data;for(let t=0;t<e.length;++t)e[t]=Math.round(e[t]);return this}round(){return this.clone().round_()}mean(e=null,t=!1){return T(this,e,t)}min(e=null,t=!1){if(null===e){const e=(0,r.min)(this.data)[0];return new a(this.type,[e],[])}const[s,o,n]=y(((e,t)=>Math.min(e,t)),this,e,t,1/0);return new a(s,o,n)}max(e=null,t=!1){if(null===e){const e=(0,r.max)(this.data)[0];return new a(this.type,[e],[])}const[s,o,n]=y(((e,t)=>Math.max(e,t)),this,e,t,-1/0);return new a(s,o,n)}argmin(e=null,t=!1){if(null!==e)throw new Error("`dim !== null` not yet implemented.");const s=(0,r.min)(this.data)[1];return new a("int64",[BigInt(s)],[])}argmax(e=null,t=!1){if(null!==e)throw new Error("`dim !== null` not yet implemented.");const s=(0,r.max)(this.data)[1];return new a("int64",[BigInt(s)],[])}to(e){if(this.type===e)return this;if(!i.hasOwnProperty(e))throw new Error(`Unsupported type: ${e}`);let t;const s=["int64","uint64"].includes(this.type),r=["int64","uint64"].includes(e);return s&&!r?t=Number:!s&&r&&(t=BigInt),new a(e,i[e].from(this.data,t),this.dims)}}function l(e,t){const[s,o]=(0,r.permute_data)(e.data,e.dims,t);return new a(e.type,s,o)}function c(e,[t,s],o="bilinear",n=!1){const i=e.dims.at(-3)??1,l=e.dims.at(-2),c=e.dims.at(-1);let d=(0,r.interpolate_data)(e.data,[i,l,c],[t,s],o,n);return new a(e.type,d,[i,t,s])}async function d(e,{size:t=null,mode:s="bilinear"}={}){if(4!==e.dims.length)throw new Error("`interpolate_4d` currently only supports 4D input.");if(!t)throw new Error("`interpolate_4d` requires a `size` argument.");let r,o;if(2===t.length)r=[...e.dims.slice(0,2),...t];else if(3===t.length)r=[e.dims[0],...t];else{if(4!==t.length)throw new Error("`size` must be of length 2, 3, or 4.");r=t}if("nearest"===s)o=await n.TensorOpRegistry.nearest_interpolate_4d;else if("bilinear"===s)o=await n.TensorOpRegistry.bilinear_interpolate_4d;else{if("bicubic"!==s)throw new Error(`Unsupported mode: ${s}`);o=await n.TensorOpRegistry.bicubic_interpolate_4d}const i=new a("int64",new BigInt64Array(r.map(BigInt)),[r.length]);return await o({x:e,s:i})}async function u(e,t){const s=await n.TensorOpRegistry.matmul;return await s({a:e,b:t})}async function _(e,t){const s=await n.TensorOpRegistry.rfft;return await s({x:e,a:t})}async function p(e,t){const s=await n.TensorOpRegistry.top_k;return t=null==t?e.dims.at(-1):Math.min(t,e.dims.at(-1)),await s({x:e,k:new a("int64",[BigInt(t)],[1])})}const m=e=>new a("int64",e,[e.length]);async function h(e,t,s,r,o){const i=await n.TensorOpRegistry.slice;return await i({x:e,s:m(t),e:m(s),a:m(r),t:m(o??new Array(r.length).fill(1))})}function g(e,t){const s=e.data,r=t.data,o=[e.dims[0],e.dims[2]],n=new s.constructor(o[0]*o[1]),[i,l,c]=e.dims;let d=0;for(let e=0;e<i;++e){const t=e*c*l;for(let o=0;o<c;++o){let i=0,a=0;const u=e*l,_=t+o;for(let e=0;e<l;++e){const t=Number(r[u+e]);a+=t,i+=s[_+e*c]*t}const p=i/a;n[d++]=p}}return new a(e.type,n,o)}function f(e,t,{eps:s=1e-5}={}){if(2!==e.dims.length)throw new Error("`layer_norm` currently only supports 2D input.");const[r,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[n,i]=v(e,1,0,!0),l=n.data,c=i.data,d=e.data,u=new d.constructor(d.length);for(let e=0;e<r;++e){const t=e*o;for(let r=0;r<o;++r){const o=t+r;u[o]=(d[o]-c[e])/(l[e]+s)}}return new a(e.type,u,e.dims)}function w(e,t){return e=e.slice(),null===t?e=e.filter((e=>1!==e)):"number"==typeof t?1===e[t]&&e.splice(t,1):Array.isArray(t)&&(e=e.filter(((e,s)=>1!==e||!t.includes(s)))),e}function M(e,t){return t=x(t,e.length+1),(e=e.slice()).splice(t,0,1),e}function x(e,t,s=null,r=!0){if(e<-t||e>=t){if(r)throw new Error(`IndexError: index ${e} is out of bounds for dimension${null===s?"":" "+s} with size ${t}`);return e<-t?0:t}return e<0&&(e=(e%t+t)%t),e}function b(e,t=0){t=x(t,e[0].dims.length);const s=e[0].dims.slice();s[t]=e.reduce(((e,s)=>e+s.dims[t]),0);const r=s.reduce(((e,t)=>e*t),1),o=new e[0].data.constructor(r),n=e[0].type;if(0===t){let t=0;for(const s of e){const e=s.data;o.set(e,t),t+=e.length}}else{let r=0;for(let n=0;n<e.length;++n){const{data:i,dims:a}=e[n];for(let e=0;e<i.length;++e){let n=0;for(let o=a.length-1,i=e,l=1;o>=0;--o){const e=a[o];let c=i%e;o===t&&(c+=r),n+=c*l,l*=s[o],i=Math.floor(i/e)}o[n]=i[e]}r+=a[t]}}return new a(n,o,s)}function k(e,t=0){return b(e.map((e=>e.unsqueeze(t))),t)}function y(e,t,s=null,r=!1,o=null){const n=t.data,i=t.dims;s=x(s,i.length);const a=i.slice();a[s]=1;const l=new n.constructor(n.length/i[s]);null!==o&&l.fill(o);for(let t=0;t<n.length;++t){let r=0;for(let e=i.length-1,o=t,n=1;e>=0;--e){const t=i[e];if(e!==s){r+=o%t*n,n*=a[e]}o=Math.floor(o/t)}l[r]=e(l[r],n[t],t,r)}return r||a.splice(s,1),[t.type,l,a]}function v(e,t=null,s=1,r=!1){const o=e.data,n=e.dims;if(null===t){const t=o.reduce(((e,t)=>e+t),0)/o.length,r=Math.sqrt(o.reduce(((e,s)=>e+(s-t)**2),0)/(o.length-s)),n=new a(e.type,[t],[]);return[new a(e.type,[r],[]),n]}const i=T(e,t=x(t,n.length),r),l=i.data,[c,d,u]=y(((e,t,s,r)=>e+(t-l[r])**2),e,t,r);for(let e=0;e<d.length;++e)d[e]=Math.sqrt(d[e]/(n[t]-s));return[new a(c,d,u),i]}function T(e,t=null,s=!1){const r=e.dims,o=e.data;if(null===t){const t=o.reduce(((e,t)=>e+t),0);return new a(e.type,[t/o.length],[])}t=x(t,r.length);const[n,i,l]=y(((e,t)=>e+t),e,t,s);if(1!==r[t])for(let e=0;e<i.length;++e)i[e]/=r[t];return new a(n,i,l)}function P(e,t,s,r){const o=e.reduce(((e,t)=>e*t),1);return new a(s,new r(o).fill(t),e)}function F(e,t){let s,r;if("number"==typeof t)s="float32",r=Float32Array;else if("bigint"==typeof t)s="int64",r=BigInt64Array;else{if("boolean"!=typeof t)throw new Error("Unsupported data type: "+typeof t);s="bool",r=Uint8Array}return P(e,t,s,r)}function C(e,t){return F(e.dims,t)}function S(e){return P(e,1n,"int64",BigInt64Array)}function A(e){return S(e.dims)}function E(e){return P(e,0n,"int64",BigInt64Array)}function L(e){return E(e.dims)}function I(e){const t=e.reduce(((e,t)=>e*t),1);return new a("float32",Float32Array.from({length:t},(()=>Math.random())),e)}function z(e,t){if(2!==e.dims.length)throw new Error("The tensor must have 2 dimensions");if(e.dims.at(-1)%8!=0)throw new Error("The last dimension of the tensor must be a multiple of 8");if(!["binary","ubinary"].includes(t))throw new Error("The precision must be either 'binary' or 'ubinary'");const s="binary"===t,r=s?"int8":"uint8",o=s?Int8Array:Uint8Array,n=e.data,i=new o(n.length/8);for(let e=0;e<n.length;++e){const t=n[e]>0?1:0,r=Math.floor(e/8),o=e%8;i[r]|=t<<7-o,s&&0===o&&(i[r]-=128)}return new a(r,i,[e.dims[0],e.dims[1]/8])}},"./src/utils/video.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{RawVideo:()=>i,RawVideoFrame:()=>n,load_video:()=>a});var r=s("./src/utils/image.js"),o=s("./src/env.js");class n{constructor(e,t){this.image=e,this.timestamp=t}}class i{constructor(e,t){e.length>0&&e[0]instanceof r.RawImage&&(e=e.map(((s,r)=>new n(s,(r+1)/(e.length+1)*t)))),this.frames=e,this.duration=t}get width(){return this.frames[0].image.width}get height(){return this.frames[0].image.height}get fps(){return this.frames.length/this.duration}}async function a(e,{num_frames:t=null,fps:s=null}={}){if(!o.apis.IS_BROWSER_ENV)throw new Error("`load_video` is currently only supported in browser environments.");if(null==t&&null==s)throw new Error("Either num_frames or fps must be provided.");const a=[],l=document.createElement("video");if(l.crossOrigin="anonymous",l.muted=!0,"string"==typeof e)l.src=e;else if(e instanceof Blob)l.src=URL.createObjectURL(e);else{if(!(e instanceof HTMLVideoElement))throw new Error("Invalid URL or video element provided.");l.src=e.src}if(await new Promise((e=>l.onloadedmetadata=e)),l.seekable.start(0)===l.seekable.end(0)){const e=await fetch(l.src),t=await e.blob();l.src=URL.createObjectURL(t),await new Promise((e=>l.onloadedmetadata=e))}const c=l.duration;let d,u;null!=t?(d=t,u=1===t?0:c/(t-1)):(u=1/s,d=Math.floor(c/u));let _=[];for(let e=0;e<d;++e)_.push(1===t?c/2:e*u);const p=document.createElement("canvas");p.width=l.videoWidth,p.height=l.videoHeight;const m=p.getContext("2d",{willReadFrequently:!0});for(const e of _){l.currentTime=e,await new Promise((e=>{l.onseeked=e})),m.drawImage(l,0,0,p.width,p.height);const t=m.getImageData(0,0,p.width,p.height),s=new r.RawImage(t.data,p.width,p.height,4),o=new n(s,e);a.push(o)}return l.remove(),new i(a,c)}}},r={};function o(e){var t=r[e];if(void 0!==t)return t.exports;var n=r[e]={exports:{}};return s[e](n,n.exports,o),n.exports}t=Object.getPrototypeOf?e=>Object.getPrototypeOf(e):e=>e.__proto__,o.t=function(s,r){if(1&r&&(s=this(s)),8&r)return s;if("object"==typeof s&&s){if(4&r&&s.__esModule)return s;if(16&r&&"function"==typeof s.then)return s}var n=Object.create(null);o.r(n);var i={};e=e||[null,t({}),t([]),t(t)];for(var a=2&r&&s;"object"==typeof a&&!~e.indexOf(a);a=t(a))Object.getOwnPropertyNames(a).forEach((e=>i[e]=()=>s[e]));return i.default=()=>s,o.d(n,i),n},o.d=(e,t)=>{for(var s in t)o.o(t,s)&&!o.o(e,s)&&Object.defineProperty(e,s,{enumerable:!0,get:t[s]})},o.o=(e,t)=>Object.prototype.hasOwnProperty.call(e,t),o.r=e=>{"undefined"!=typeof Symbol&&Symbol.toStringTag&&Object.defineProperty(e,Symbol.toStringTag,{value:"Module"}),Object.defineProperty(e,"__esModule",{value:!0})};var n={};(()=>{"use strict";o.r(n),o.d(n,{ASTFeatureExtractor:()=>p.ASTFeatureExtractor,ASTForAudioClassification:()=>s.ASTForAudioClassification,ASTModel:()=>s.ASTModel,ASTPreTrainedModel:()=>s.ASTPreTrainedModel,AlbertForMaskedLM:()=>s.AlbertForMaskedLM,AlbertForQuestionAnswering:()=>s.AlbertForQuestionAnswering,AlbertForSequenceClassification:()=>s.AlbertForSequenceClassification,AlbertModel:()=>s.AlbertModel,AlbertPreTrainedModel:()=>s.AlbertPreTrainedModel,AlbertTokenizer:()=>r.AlbertTokenizer,AudioClassificationPipeline:()=>t.AudioClassificationPipeline,AutoConfig:()=>i.AutoConfig,AutoFeatureExtractor:()=>m.AutoFeatureExtractor,AutoImageProcessor:()=>f.AutoImageProcessor,AutoModel:()=>s.AutoModel,AutoModelForAudioClassification:()=>s.AutoModelForAudioClassification,AutoModelForAudioFrameClassification:()=>s.AutoModelForAudioFrameClassification,AutoModelForAudioTextToText:()=>s.AutoModelForAudioTextToText,AutoModelForCTC:()=>s.AutoModelForCTC,AutoModelForCausalLM:()=>s.AutoModelForCausalLM,AutoModelForDepthEstimation:()=>s.AutoModelForDepthEstimation,AutoModelForDocumentQuestionAnswering:()=>s.AutoModelForDocumentQuestionAnswering,AutoModelForImageClassification:()=>s.AutoModelForImageClassification,AutoModelForImageFeatureExtraction:()=>s.AutoModelForImageFeatureExtraction,AutoModelForImageMatting:()=>s.AutoModelForImageMatting,AutoModelForImageSegmentation:()=>s.AutoModelForImageSegmentation,AutoModelForImageTextToText:()=>s.AutoModelForImageTextToText,AutoModelForImageToImage:()=>s.AutoModelForImageToImage,AutoModelForMaskGeneration:()=>s.AutoModelForMaskGeneration,AutoModelForMaskedLM:()=>s.AutoModelForMaskedLM,AutoModelForNormalEstimation:()=>s.AutoModelForNormalEstimation,AutoModelForObjectDetection:()=>s.AutoModelForObjectDetection,AutoModelForPoseEstimation:()=>s.AutoModelForPoseEstimation,AutoModelForQuestionAnswering:()=>s.AutoModelForQuestionAnswering,AutoModelForSemanticSegmentation:()=>s.AutoModelForSemanticSegmentation,AutoModelForSeq2SeqLM:()=>s.AutoModelForSeq2SeqLM,AutoModelForSequenceClassification:()=>s.AutoModelForSequenceClassification,AutoModelForSpeechSeq2Seq:()=>s.AutoModelForSpeechSeq2Seq,AutoModelForTextToSpectrogram:()=>s.AutoModelForTextToSpectrogram,AutoModelForTextToWaveform:()=>s.AutoModelForTextToWaveform,AutoModelForTokenClassification:()=>s.AutoModelForTokenClassification,AutoModelForUniversalSegmentation:()=>s.AutoModelForUniversalSegmentation,AutoModelForVision2Seq:()=>s.AutoModelForVision2Seq,AutoModelForXVector:()=>s.AutoModelForXVector,AutoModelForZeroShotObjectDetection:()=>s.AutoModelForZeroShotObjectDetection,AutoProcessor:()=>x.AutoProcessor,AutoTokenizer:()=>r.AutoTokenizer,AutomaticSpeechRecognitionPipeline:()=>t.AutomaticSpeechRecognitionPipeline,BackgroundRemovalPipeline:()=>t.BackgroundRemovalPipeline,BartForConditionalGeneration:()=>s.BartForConditionalGeneration,BartForSequenceClassification:()=>s.BartForSequenceClassification,BartModel:()=>s.BartModel,BartPretrainedModel:()=>s.BartPretrainedModel,BartTokenizer:()=>r.BartTokenizer,BaseModelOutput:()=>s.BaseModelOutput,BaseStreamer:()=>b.BaseStreamer,BeitFeatureExtractor:()=>g.BeitFeatureExtractor,BeitForImageClassification:()=>s.BeitForImageClassification,BeitModel:()=>s.BeitModel,BeitPreTrainedModel:()=>s.BeitPreTrainedModel,BertForMaskedLM:()=>s.BertForMaskedLM,BertForQuestionAnswering:()=>s.BertForQuestionAnswering,BertForSequenceClassification:()=>s.BertForSequenceClassification,BertForTokenClassification:()=>s.BertForTokenClassification,BertModel:()=>s.BertModel,BertPreTrainedModel:()=>s.BertPreTrainedModel,BertTokenizer:()=>r.BertTokenizer,BitImageProcessor:()=>g.BitImageProcessor,BlenderbotForConditionalGeneration:()=>s.BlenderbotForConditionalGeneration,BlenderbotModel:()=>s.BlenderbotModel,BlenderbotPreTrainedModel:()=>s.BlenderbotPreTrainedModel,BlenderbotSmallForConditionalGeneration:()=>s.BlenderbotSmallForConditionalGeneration,BlenderbotSmallModel:()=>s.BlenderbotSmallModel,BlenderbotSmallPreTrainedModel:()=>s.BlenderbotSmallPreTrainedModel,BlenderbotSmallTokenizer:()=>r.BlenderbotSmallTokenizer,BlenderbotTokenizer:()=>r.BlenderbotTokenizer,BloomForCausalLM:()=>s.BloomForCausalLM,BloomModel:()=>s.BloomModel,BloomPreTrainedModel:()=>s.BloomPreTrainedModel,BloomTokenizer:()=>r.BloomTokenizer,CLIPFeatureExtractor:()=>g.CLIPFeatureExtractor,CLIPImageProcessor:()=>g.CLIPImageProcessor,CLIPModel:()=>s.CLIPModel,CLIPPreTrainedModel:()=>s.CLIPPreTrainedModel,CLIPSegForImageSegmentation:()=>s.CLIPSegForImageSegmentation,CLIPSegModel:()=>s.CLIPSegModel,CLIPSegPreTrainedModel:()=>s.CLIPSegPreTrainedModel,CLIPTextModel:()=>s.CLIPTextModel,CLIPTextModelWithProjection:()=>s.CLIPTextModelWithProjection,CLIPTokenizer:()=>r.CLIPTokenizer,CLIPVisionModel:()=>s.CLIPVisionModel,CLIPVisionModelWithProjection:()=>s.CLIPVisionModelWithProjection,CamembertForMaskedLM:()=>s.CamembertForMaskedLM,CamembertForQuestionAnswering:()=>s.CamembertForQuestionAnswering,CamembertForSequenceClassification:()=>s.CamembertForSequenceClassification,CamembertForTokenClassification:()=>s.CamembertForTokenClassification,CamembertModel:()=>s.CamembertModel,CamembertPreTrainedModel:()=>s.CamembertPreTrainedModel,CamembertTokenizer:()=>r.CamembertTokenizer,CausalLMOutput:()=>s.CausalLMOutput,CausalLMOutputWithPast:()=>s.CausalLMOutputWithPast,ChineseCLIPFeatureExtractor:()=>g.ChineseCLIPFeatureExtractor,ChineseCLIPModel:()=>s.ChineseCLIPModel,ChineseCLIPPreTrainedModel:()=>s.ChineseCLIPPreTrainedModel,ClapAudioModelWithProjection:()=>s.ClapAudioModelWithProjection,ClapFeatureExtractor:()=>p.ClapFeatureExtractor,ClapModel:()=>s.ClapModel,ClapPreTrainedModel:()=>s.ClapPreTrainedModel,ClapTextModelWithProjection:()=>s.ClapTextModelWithProjection,ClassifierFreeGuidanceLogitsProcessor:()=>y.ClassifierFreeGuidanceLogitsProcessor,CodeGenForCausalLM:()=>s.CodeGenForCausalLM,CodeGenModel:()=>s.CodeGenModel,CodeGenPreTrainedModel:()=>s.CodeGenPreTrainedModel,CodeGenTokenizer:()=>r.CodeGenTokenizer,CodeLlamaTokenizer:()=>r.CodeLlamaTokenizer,CohereForCausalLM:()=>s.CohereForCausalLM,CohereModel:()=>s.CohereModel,CoherePreTrainedModel:()=>s.CoherePreTrainedModel,CohereTokenizer:()=>r.CohereTokenizer,ConvBertForMaskedLM:()=>s.ConvBertForMaskedLM,ConvBertForQuestionAnswering:()=>s.ConvBertForQuestionAnswering,ConvBertForSequenceClassification:()=>s.ConvBertForSequenceClassification,ConvBertForTokenClassification:()=>s.ConvBertForTokenClassification,ConvBertModel:()=>s.ConvBertModel,ConvBertPreTrainedModel:()=>s.ConvBertPreTrainedModel,ConvBertTokenizer:()=>r.ConvBertTokenizer,ConvNextFeatureExtractor:()=>g.ConvNextFeatureExtractor,ConvNextForImageClassification:()=>s.ConvNextForImageClassification,ConvNextImageProcessor:()=>g.ConvNextImageProcessor,ConvNextModel:()=>s.ConvNextModel,ConvNextPreTrainedModel:()=>s.ConvNextPreTrainedModel,ConvNextV2ForImageClassification:()=>s.ConvNextV2ForImageClassification,ConvNextV2Model:()=>s.ConvNextV2Model,ConvNextV2PreTrainedModel:()=>s.ConvNextV2PreTrainedModel,DPTFeatureExtractor:()=>g.DPTFeatureExtractor,DPTForDepthEstimation:()=>s.DPTForDepthEstimation,DPTImageProcessor:()=>g.DPTImageProcessor,DPTModel:()=>s.DPTModel,DPTPreTrainedModel:()=>s.DPTPreTrainedModel,DacDecoderModel:()=>s.DacDecoderModel,DacDecoderOutput:()=>s.DacDecoderOutput,DacEncoderModel:()=>s.DacEncoderModel,DacEncoderOutput:()=>s.DacEncoderOutput,DacFeatureExtractor:()=>p.DacFeatureExtractor,DacModel:()=>s.DacModel,DacPreTrainedModel:()=>s.DacPreTrainedModel,DataTypeMap:()=>d.DataTypeMap,DebertaForMaskedLM:()=>s.DebertaForMaskedLM,DebertaForQuestionAnswering:()=>s.DebertaForQuestionAnswering,DebertaForSequenceClassification:()=>s.DebertaForSequenceClassification,DebertaForTokenClassification:()=>s.DebertaForTokenClassification,DebertaModel:()=>s.DebertaModel,DebertaPreTrainedModel:()=>s.DebertaPreTrainedModel,DebertaTokenizer:()=>r.DebertaTokenizer,DebertaV2ForMaskedLM:()=>s.DebertaV2ForMaskedLM,DebertaV2ForQuestionAnswering:()=>s.DebertaV2ForQuestionAnswering,DebertaV2ForSequenceClassification:()=>s.DebertaV2ForSequenceClassification,DebertaV2ForTokenClassification:()=>s.DebertaV2ForTokenClassification,DebertaV2Model:()=>s.DebertaV2Model,DebertaV2PreTrainedModel:()=>s.DebertaV2PreTrainedModel,DebertaV2Tokenizer:()=>r.DebertaV2Tokenizer,DecisionTransformerModel:()=>s.DecisionTransformerModel,DecisionTransformerPreTrainedModel:()=>s.DecisionTransformerPreTrainedModel,DeiTFeatureExtractor:()=>g.DeiTFeatureExtractor,DeiTForImageClassification:()=>s.DeiTForImageClassification,DeiTImageProcessor:()=>g.DeiTImageProcessor,DeiTModel:()=>s.DeiTModel,DeiTPreTrainedModel:()=>s.DeiTPreTrainedModel,DepthAnythingForDepthEstimation:()=>s.DepthAnythingForDepthEstimation,DepthAnythingPreTrainedModel:()=>s.DepthAnythingPreTrainedModel,DepthEstimationPipeline:()=>t.DepthEstimationPipeline,DepthProForDepthEstimation:()=>s.DepthProForDepthEstimation,DepthProPreTrainedModel:()=>s.DepthProPreTrainedModel,DetrFeatureExtractor:()=>g.DetrFeatureExtractor,DetrForObjectDetection:()=>s.DetrForObjectDetection,DetrForSegmentation:()=>s.DetrForSegmentation,DetrImageProcessor:()=>g.DetrImageProcessor,DetrModel:()=>s.DetrModel,DetrObjectDetectionOutput:()=>s.DetrObjectDetectionOutput,DetrPreTrainedModel:()=>s.DetrPreTrainedModel,DetrSegmentationOutput:()=>s.DetrSegmentationOutput,Dinov2ForImageClassification:()=>s.Dinov2ForImageClassification,Dinov2Model:()=>s.Dinov2Model,Dinov2PreTrainedModel:()=>s.Dinov2PreTrainedModel,Dinov2WithRegistersForImageClassification:()=>s.Dinov2WithRegistersForImageClassification,Dinov2WithRegistersModel:()=>s.Dinov2WithRegistersModel,Dinov2WithRegistersPreTrainedModel:()=>s.Dinov2WithRegistersPreTrainedModel,DistilBertForMaskedLM:()=>s.DistilBertForMaskedLM,DistilBertForQuestionAnswering:()=>s.DistilBertForQuestionAnswering,DistilBertForSequenceClassification:()=>s.DistilBertForSequenceClassification,DistilBertForTokenClassification:()=>s.DistilBertForTokenClassification,DistilBertModel:()=>s.DistilBertModel,DistilBertPreTrainedModel:()=>s.DistilBertPreTrainedModel,DistilBertTokenizer:()=>r.DistilBertTokenizer,DocumentQuestionAnsweringPipeline:()=>t.DocumentQuestionAnsweringPipeline,DonutFeatureExtractor:()=>g.DonutFeatureExtractor,DonutImageProcessor:()=>g.DonutImageProcessor,DonutSwinModel:()=>s.DonutSwinModel,DonutSwinPreTrainedModel:()=>s.DonutSwinPreTrainedModel,EfficientNetForImageClassification:()=>s.EfficientNetForImageClassification,EfficientNetImageProcessor:()=>g.EfficientNetImageProcessor,EfficientNetModel:()=>s.EfficientNetModel,EfficientNetPreTrainedModel:()=>s.EfficientNetPreTrainedModel,ElectraForMaskedLM:()=>s.ElectraForMaskedLM,ElectraForQuestionAnswering:()=>s.ElectraForQuestionAnswering,ElectraForSequenceClassification:()=>s.ElectraForSequenceClassification,ElectraForTokenClassification:()=>s.ElectraForTokenClassification,ElectraModel:()=>s.ElectraModel,ElectraPreTrainedModel:()=>s.ElectraPreTrainedModel,ElectraTokenizer:()=>r.ElectraTokenizer,EncodecFeatureExtractor:()=>p.EncodecFeatureExtractor,EosTokenCriteria:()=>k.EosTokenCriteria,EsmForMaskedLM:()=>s.EsmForMaskedLM,EsmForSequenceClassification:()=>s.EsmForSequenceClassification,EsmForTokenClassification:()=>s.EsmForTokenClassification,EsmModel:()=>s.EsmModel,EsmPreTrainedModel:()=>s.EsmPreTrainedModel,EsmTokenizer:()=>r.EsmTokenizer,ExaoneForCausalLM:()=>s.ExaoneForCausalLM,ExaoneModel:()=>s.ExaoneModel,ExaonePreTrainedModel:()=>s.ExaonePreTrainedModel,FFT:()=>u.FFT,FalconForCausalLM:()=>s.FalconForCausalLM,FalconModel:()=>s.FalconModel,FalconPreTrainedModel:()=>s.FalconPreTrainedModel,FalconTokenizer:()=>r.FalconTokenizer,FastViTForImageClassification:()=>s.FastViTForImageClassification,FastViTModel:()=>s.FastViTModel,FastViTPreTrainedModel:()=>s.FastViTPreTrainedModel,FeatureExtractionPipeline:()=>t.FeatureExtractionPipeline,FeatureExtractor:()=>_.FeatureExtractor,FillMaskPipeline:()=>t.FillMaskPipeline,Florence2ForConditionalGeneration:()=>s.Florence2ForConditionalGeneration,Florence2PreTrainedModel:()=>s.Florence2PreTrainedModel,Florence2Processor:()=>M.Florence2Processor,ForcedBOSTokenLogitsProcessor:()=>y.ForcedBOSTokenLogitsProcessor,ForcedEOSTokenLogitsProcessor:()=>y.ForcedEOSTokenLogitsProcessor,GLPNFeatureExtractor:()=>g.GLPNFeatureExtractor,GLPNForDepthEstimation:()=>s.GLPNForDepthEstimation,GLPNModel:()=>s.GLPNModel,GLPNPreTrainedModel:()=>s.GLPNPreTrainedModel,GPT2LMHeadModel:()=>s.GPT2LMHeadModel,GPT2Model:()=>s.GPT2Model,GPT2PreTrainedModel:()=>s.GPT2PreTrainedModel,GPT2Tokenizer:()=>r.GPT2Tokenizer,GPTBigCodeForCausalLM:()=>s.GPTBigCodeForCausalLM,GPTBigCodeModel:()=>s.GPTBigCodeModel,GPTBigCodePreTrainedModel:()=>s.GPTBigCodePreTrainedModel,GPTJForCausalLM:()=>s.GPTJForCausalLM,GPTJModel:()=>s.GPTJModel,GPTJPreTrainedModel:()=>s.GPTJPreTrainedModel,GPTNeoForCausalLM:()=>s.GPTNeoForCausalLM,GPTNeoModel:()=>s.GPTNeoModel,GPTNeoPreTrainedModel:()=>s.GPTNeoPreTrainedModel,GPTNeoXForCausalLM:()=>s.GPTNeoXForCausalLM,GPTNeoXModel:()=>s.GPTNeoXModel,GPTNeoXPreTrainedModel:()=>s.GPTNeoXPreTrainedModel,GPTNeoXTokenizer:()=>r.GPTNeoXTokenizer,Gemma2ForCausalLM:()=>s.Gemma2ForCausalLM,Gemma2Model:()=>s.Gemma2Model,Gemma2PreTrainedModel:()=>s.Gemma2PreTrainedModel,Gemma3ForCausalLM:()=>s.Gemma3ForCausalLM,Gemma3Model:()=>s.Gemma3Model,Gemma3PreTrainedModel:()=>s.Gemma3PreTrainedModel,GemmaForCausalLM:()=>s.GemmaForCausalLM,GemmaModel:()=>s.GemmaModel,GemmaPreTrainedModel:()=>s.GemmaPreTrainedModel,GemmaTokenizer:()=>r.GemmaTokenizer,GlmForCausalLM:()=>s.GlmForCausalLM,GlmModel:()=>s.GlmModel,GlmPreTrainedModel:()=>s.GlmPreTrainedModel,GraniteForCausalLM:()=>s.GraniteForCausalLM,GraniteModel:()=>s.GraniteModel,GranitePreTrainedModel:()=>s.GranitePreTrainedModel,Grok1Tokenizer:()=>r.Grok1Tokenizer,GroundingDinoForObjectDetection:()=>s.GroundingDinoForObjectDetection,GroundingDinoImageProcessor:()=>g.GroundingDinoImageProcessor,GroundingDinoPreTrainedModel:()=>s.GroundingDinoPreTrainedModel,GroundingDinoProcessor:()=>M.GroundingDinoProcessor,GroupViTModel:()=>s.GroupViTModel,GroupViTPreTrainedModel:()=>s.GroupViTPreTrainedModel,HeliumForCausalLM:()=>s.HeliumForCausalLM,HeliumModel:()=>s.HeliumModel,HeliumPreTrainedModel:()=>s.HeliumPreTrainedModel,HerbertTokenizer:()=>r.HerbertTokenizer,HieraForImageClassification:()=>s.HieraForImageClassification,HieraModel:()=>s.HieraModel,HieraPreTrainedModel:()=>s.HieraPreTrainedModel,HubertForCTC:()=>s.HubertForCTC,HubertForSequenceClassification:()=>s.HubertForSequenceClassification,HubertModel:()=>s.HubertModel,HubertPreTrainedModel:()=>s.HubertPreTrainedModel,IJepaForImageClassification:()=>s.IJepaForImageClassification,IJepaModel:()=>s.IJepaModel,IJepaPreTrainedModel:()=>s.IJepaPreTrainedModel,Idefics3ForConditionalGeneration:()=>s.Idefics3ForConditionalGeneration,Idefics3ImageProcessor:()=>g.Idefics3ImageProcessor,Idefics3PreTrainedModel:()=>s.Idefics3PreTrainedModel,Idefics3Processor:()=>M.Idefics3Processor,ImageClassificationPipeline:()=>t.ImageClassificationPipeline,ImageFeatureExtractionPipeline:()=>t.ImageFeatureExtractionPipeline,ImageFeatureExtractor:()=>p.ImageFeatureExtractor,ImageMattingOutput:()=>s.ImageMattingOutput,ImageProcessor:()=>h.ImageProcessor,ImageSegmentationPipeline:()=>t.ImageSegmentationPipeline,ImageToImagePipeline:()=>t.ImageToImagePipeline,ImageToTextPipeline:()=>t.ImageToTextPipeline,InterruptableStoppingCriteria:()=>k.InterruptableStoppingCriteria,JAISLMHeadModel:()=>s.JAISLMHeadModel,JAISModel:()=>s.JAISModel,JAISPreTrainedModel:()=>s.JAISPreTrainedModel,JinaCLIPImageProcessor:()=>g.JinaCLIPImageProcessor,JinaCLIPModel:()=>s.JinaCLIPModel,JinaCLIPPreTrainedModel:()=>s.JinaCLIPPreTrainedModel,JinaCLIPProcessor:()=>M.JinaCLIPProcessor,JinaCLIPTextModel:()=>s.JinaCLIPTextModel,JinaCLIPVisionModel:()=>s.JinaCLIPVisionModel,LiteWhisperForConditionalGeneration:()=>s.LiteWhisperForConditionalGeneration,LlamaForCausalLM:()=>s.LlamaForCausalLM,LlamaModel:()=>s.LlamaModel,LlamaPreTrainedModel:()=>s.LlamaPreTrainedModel,LlamaTokenizer:()=>r.LlamaTokenizer,LlavaForConditionalGeneration:()=>s.LlavaForConditionalGeneration,LlavaOnevisionForConditionalGeneration:()=>s.LlavaOnevisionForConditionalGeneration,LlavaOnevisionImageProcessor:()=>g.LlavaOnevisionImageProcessor,LlavaPreTrainedModel:()=>s.LlavaPreTrainedModel,LogitsProcessor:()=>y.LogitsProcessor,LogitsProcessorList:()=>y.LogitsProcessorList,LogitsWarper:()=>y.LogitsWarper,LongT5ForConditionalGeneration:()=>s.LongT5ForConditionalGeneration,LongT5Model:()=>s.LongT5Model,LongT5PreTrainedModel:()=>s.LongT5PreTrainedModel,M2M100ForConditionalGeneration:()=>s.M2M100ForConditionalGeneration,M2M100Model:()=>s.M2M100Model,M2M100PreTrainedModel:()=>s.M2M100PreTrainedModel,M2M100Tokenizer:()=>r.M2M100Tokenizer,MBart50Tokenizer:()=>r.MBart50Tokenizer,MBartForCausalLM:()=>s.MBartForCausalLM,MBartForConditionalGeneration:()=>s.MBartForConditionalGeneration,MBartForSequenceClassification:()=>s.MBartForSequenceClassification,MBartModel:()=>s.MBartModel,MBartPreTrainedModel:()=>s.MBartPreTrainedModel,MBartTokenizer:()=>r.MBartTokenizer,MPNetForMaskedLM:()=>s.MPNetForMaskedLM,MPNetForQuestionAnswering:()=>s.MPNetForQuestionAnswering,MPNetForSequenceClassification:()=>s.MPNetForSequenceClassification,MPNetForTokenClassification:()=>s.MPNetForTokenClassification,MPNetModel:()=>s.MPNetModel,MPNetPreTrainedModel:()=>s.MPNetPreTrainedModel,MPNetTokenizer:()=>r.MPNetTokenizer,MT5ForConditionalGeneration:()=>s.MT5ForConditionalGeneration,MT5Model:()=>s.MT5Model,MT5PreTrainedModel:()=>s.MT5PreTrainedModel,MarianMTModel:()=>s.MarianMTModel,MarianModel:()=>s.MarianModel,MarianPreTrainedModel:()=>s.MarianPreTrainedModel,MarianTokenizer:()=>r.MarianTokenizer,Mask2FormerImageProcessor:()=>g.Mask2FormerImageProcessor,MaskFormerFeatureExtractor:()=>g.MaskFormerFeatureExtractor,MaskFormerForInstanceSegmentation:()=>s.MaskFormerForInstanceSegmentation,MaskFormerImageProcessor:()=>g.MaskFormerImageProcessor,MaskFormerModel:()=>s.MaskFormerModel,MaskFormerPreTrainedModel:()=>s.MaskFormerPreTrainedModel,MaskedLMOutput:()=>s.MaskedLMOutput,MaxLengthCriteria:()=>k.MaxLengthCriteria,Metric3DForDepthEstimation:()=>s.Metric3DForDepthEstimation,Metric3DPreTrainedModel:()=>s.Metric3DPreTrainedModel,Metric3Dv2ForDepthEstimation:()=>s.Metric3Dv2ForDepthEstimation,Metric3Dv2PreTrainedModel:()=>s.Metric3Dv2PreTrainedModel,MgpstrForSceneTextRecognition:()=>s.MgpstrForSceneTextRecognition,MgpstrModelOutput:()=>s.MgpstrModelOutput,MgpstrPreTrainedModel:()=>s.MgpstrPreTrainedModel,MgpstrProcessor:()=>M.MgpstrProcessor,MgpstrTokenizer:()=>r.MgpstrTokenizer,MimiDecoderModel:()=>s.MimiDecoderModel,MimiDecoderOutput:()=>s.MimiDecoderOutput,MimiEncoderModel:()=>s.MimiEncoderModel,MimiEncoderOutput:()=>s.MimiEncoderOutput,MimiModel:()=>s.MimiModel,MimiPreTrainedModel:()=>s.MimiPreTrainedModel,MinLengthLogitsProcessor:()=>y.MinLengthLogitsProcessor,MinNewTokensLengthLogitsProcessor:()=>y.MinNewTokensLengthLogitsProcessor,MistralForCausalLM:()=>s.MistralForCausalLM,MistralModel:()=>s.MistralModel,MistralPreTrainedModel:()=>s.MistralPreTrainedModel,MobileBertForMaskedLM:()=>s.MobileBertForMaskedLM,MobileBertForQuestionAnswering:()=>s.MobileBertForQuestionAnswering,MobileBertForSequenceClassification:()=>s.MobileBertForSequenceClassification,MobileBertModel:()=>s.MobileBertModel,MobileBertPreTrainedModel:()=>s.MobileBertPreTrainedModel,MobileBertTokenizer:()=>r.MobileBertTokenizer,MobileLLMForCausalLM:()=>s.MobileLLMForCausalLM,MobileLLMModel:()=>s.MobileLLMModel,MobileLLMPreTrainedModel:()=>s.MobileLLMPreTrainedModel,MobileNetV1FeatureExtractor:()=>g.MobileNetV1FeatureExtractor,MobileNetV1ForImageClassification:()=>s.MobileNetV1ForImageClassification,MobileNetV1ForSemanticSegmentation:()=>s.MobileNetV1ForSemanticSegmentation,MobileNetV1ImageProcessor:()=>g.MobileNetV1ImageProcessor,MobileNetV1Model:()=>s.MobileNetV1Model,MobileNetV1PreTrainedModel:()=>s.MobileNetV1PreTrainedModel,MobileNetV2FeatureExtractor:()=>g.MobileNetV2FeatureExtractor,MobileNetV2ForImageClassification:()=>s.MobileNetV2ForImageClassification,MobileNetV2ForSemanticSegmentation:()=>s.MobileNetV2ForSemanticSegmentation,MobileNetV2ImageProcessor:()=>g.MobileNetV2ImageProcessor,MobileNetV2Model:()=>s.MobileNetV2Model,MobileNetV2PreTrainedModel:()=>s.MobileNetV2PreTrainedModel,MobileNetV3FeatureExtractor:()=>g.MobileNetV3FeatureExtractor,MobileNetV3ForImageClassification:()=>s.MobileNetV3ForImageClassification,MobileNetV3ForSemanticSegmentation:()=>s.MobileNetV3ForSemanticSegmentation,MobileNetV3ImageProcessor:()=>g.MobileNetV3ImageProcessor,MobileNetV3Model:()=>s.MobileNetV3Model,MobileNetV3PreTrainedModel:()=>s.MobileNetV3PreTrainedModel,MobileNetV4FeatureExtractor:()=>g.MobileNetV4FeatureExtractor,MobileNetV4ForImageClassification:()=>s.MobileNetV4ForImageClassification,MobileNetV4ForSemanticSegmentation:()=>s.MobileNetV4ForSemanticSegmentation,MobileNetV4ImageProcessor:()=>g.MobileNetV4ImageProcessor,MobileNetV4Model:()=>s.MobileNetV4Model,MobileNetV4PreTrainedModel:()=>s.MobileNetV4PreTrainedModel,MobileViTFeatureExtractor:()=>g.MobileViTFeatureExtractor,MobileViTForImageClassification:()=>s.MobileViTForImageClassification,MobileViTImageProcessor:()=>g.MobileViTImageProcessor,MobileViTModel:()=>s.MobileViTModel,MobileViTPreTrainedModel:()=>s.MobileViTPreTrainedModel,MobileViTV2ForImageClassification:()=>s.MobileViTV2ForImageClassification,MobileViTV2Model:()=>s.MobileViTV2Model,MobileViTV2PreTrainedModel:()=>s.MobileViTV2PreTrainedModel,ModelOutput:()=>s.ModelOutput,ModernBertForMaskedLM:()=>s.ModernBertForMaskedLM,ModernBertForSequenceClassification:()=>s.ModernBertForSequenceClassification,ModernBertForTokenClassification:()=>s.ModernBertForTokenClassification,ModernBertModel:()=>s.ModernBertModel,ModernBertPreTrainedModel:()=>s.ModernBertPreTrainedModel,Moondream1ForConditionalGeneration:()=>s.Moondream1ForConditionalGeneration,MoonshineFeatureExtractor:()=>p.MoonshineFeatureExtractor,MoonshineForConditionalGeneration:()=>s.MoonshineForConditionalGeneration,MoonshineModel:()=>s.MoonshineModel,MoonshinePreTrainedModel:()=>s.MoonshinePreTrainedModel,MoonshineProcessor:()=>M.MoonshineProcessor,MptForCausalLM:()=>s.MptForCausalLM,MptModel:()=>s.MptModel,MptPreTrainedModel:()=>s.MptPreTrainedModel,MultiModalityCausalLM:()=>s.MultiModalityCausalLM,MultiModalityPreTrainedModel:()=>s.MultiModalityPreTrainedModel,MusicgenForCausalLM:()=>s.MusicgenForCausalLM,MusicgenForConditionalGeneration:()=>s.MusicgenForConditionalGeneration,MusicgenModel:()=>s.MusicgenModel,MusicgenPreTrainedModel:()=>s.MusicgenPreTrainedModel,NllbTokenizer:()=>r.NllbTokenizer,NoBadWordsLogitsProcessor:()=>y.NoBadWordsLogitsProcessor,NoRepeatNGramLogitsProcessor:()=>y.NoRepeatNGramLogitsProcessor,NomicBertModel:()=>s.NomicBertModel,NomicBertPreTrainedModel:()=>s.NomicBertPreTrainedModel,NougatImageProcessor:()=>g.NougatImageProcessor,NougatTokenizer:()=>r.NougatTokenizer,OPTForCausalLM:()=>s.OPTForCausalLM,OPTModel:()=>s.OPTModel,OPTPreTrainedModel:()=>s.OPTPreTrainedModel,ObjectDetectionPipeline:()=>t.ObjectDetectionPipeline,Olmo2ForCausalLM:()=>s.Olmo2ForCausalLM,Olmo2Model:()=>s.Olmo2Model,Olmo2PreTrainedModel:()=>s.Olmo2PreTrainedModel,OlmoForCausalLM:()=>s.OlmoForCausalLM,OlmoModel:()=>s.OlmoModel,OlmoPreTrainedModel:()=>s.OlmoPreTrainedModel,OpenELMForCausalLM:()=>s.OpenELMForCausalLM,OpenELMModel:()=>s.OpenELMModel,OpenELMPreTrainedModel:()=>s.OpenELMPreTrainedModel,OwlViTFeatureExtractor:()=>g.OwlViTFeatureExtractor,OwlViTForObjectDetection:()=>s.OwlViTForObjectDetection,OwlViTImageProcessor:()=>g.OwlViTImageProcessor,OwlViTModel:()=>s.OwlViTModel,OwlViTPreTrainedModel:()=>s.OwlViTPreTrainedModel,OwlViTProcessor:()=>M.OwlViTProcessor,Owlv2ForObjectDetection:()=>s.Owlv2ForObjectDetection,Owlv2ImageProcessor:()=>g.Owlv2ImageProcessor,Owlv2Model:()=>s.Owlv2Model,Owlv2PreTrainedModel:()=>s.Owlv2PreTrainedModel,PaliGemmaForConditionalGeneration:()=>s.PaliGemmaForConditionalGeneration,PaliGemmaPreTrainedModel:()=>s.PaliGemmaPreTrainedModel,PaliGemmaProcessor:()=>M.PaliGemmaProcessor,PatchTSMixerForPrediction:()=>s.PatchTSMixerForPrediction,PatchTSMixerModel:()=>s.PatchTSMixerModel,PatchTSMixerPreTrainedModel:()=>s.PatchTSMixerPreTrainedModel,PatchTSTForPrediction:()=>s.PatchTSTForPrediction,PatchTSTModel:()=>s.PatchTSTModel,PatchTSTPreTrainedModel:()=>s.PatchTSTPreTrainedModel,Phi3ForCausalLM:()=>s.Phi3ForCausalLM,Phi3Model:()=>s.Phi3Model,Phi3PreTrainedModel:()=>s.Phi3PreTrainedModel,Phi3VForCausalLM:()=>s.Phi3VForCausalLM,Phi3VImageProcessor:()=>g.Phi3VImageProcessor,Phi3VPreTrainedModel:()=>s.Phi3VPreTrainedModel,Phi3VProcessor:()=>M.Phi3VProcessor,PhiForCausalLM:()=>s.PhiForCausalLM,PhiModel:()=>s.PhiModel,PhiPreTrainedModel:()=>s.PhiPreTrainedModel,Pipeline:()=>t.Pipeline,PreTrainedModel:()=>s.PreTrainedModel,PreTrainedTokenizer:()=>r.PreTrainedTokenizer,PretrainedConfig:()=>i.PretrainedConfig,PretrainedMixin:()=>s.PretrainedMixin,Processor:()=>w.Processor,PvtForImageClassification:()=>s.PvtForImageClassification,PvtImageProcessor:()=>g.PvtImageProcessor,PvtModel:()=>s.PvtModel,PvtPreTrainedModel:()=>s.PvtPreTrainedModel,PyAnnoteFeatureExtractor:()=>p.PyAnnoteFeatureExtractor,PyAnnoteForAudioFrameClassification:()=>s.PyAnnoteForAudioFrameClassification,PyAnnoteModel:()=>s.PyAnnoteModel,PyAnnotePreTrainedModel:()=>s.PyAnnotePreTrainedModel,PyAnnoteProcessor:()=>M.PyAnnoteProcessor,QuestionAnsweringModelOutput:()=>s.QuestionAnsweringModelOutput,QuestionAnsweringPipeline:()=>t.QuestionAnsweringPipeline,Qwen2ForCausalLM:()=>s.Qwen2ForCausalLM,Qwen2Model:()=>s.Qwen2Model,Qwen2PreTrainedModel:()=>s.Qwen2PreTrainedModel,Qwen2Tokenizer:()=>r.Qwen2Tokenizer,Qwen2VLForConditionalGeneration:()=>s.Qwen2VLForConditionalGeneration,Qwen2VLImageProcessor:()=>g.Qwen2VLImageProcessor,Qwen2VLPreTrainedModel:()=>s.Qwen2VLPreTrainedModel,Qwen2VLProcessor:()=>M.Qwen2VLProcessor,RFDetrForObjectDetection:()=>s.RFDetrForObjectDetection,RFDetrModel:()=>s.RFDetrModel,RFDetrObjectDetectionOutput:()=>s.RFDetrObjectDetectionOutput,RFDetrPreTrainedModel:()=>s.RFDetrPreTrainedModel,RTDetrForObjectDetection:()=>s.RTDetrForObjectDetection,RTDetrImageProcessor:()=>g.RTDetrImageProcessor,RTDetrModel:()=>s.RTDetrModel,RTDetrObjectDetectionOutput:()=>s.RTDetrObjectDetectionOutput,RTDetrPreTrainedModel:()=>s.RTDetrPreTrainedModel,RTDetrV2ForObjectDetection:()=>s.RTDetrV2ForObjectDetection,RTDetrV2Model:()=>s.RTDetrV2Model,RTDetrV2ObjectDetectionOutput:()=>s.RTDetrV2ObjectDetectionOutput,RTDetrV2PreTrainedModel:()=>s.RTDetrV2PreTrainedModel,RawAudio:()=>a.RawAudio,RawImage:()=>l.RawImage,RawVideo:()=>c.RawVideo,RawVideoFrame:()=>c.RawVideoFrame,RepetitionPenaltyLogitsProcessor:()=>y.RepetitionPenaltyLogitsProcessor,ResNetForImageClassification:()=>s.ResNetForImageClassification,ResNetModel:()=>s.ResNetModel,ResNetPreTrainedModel:()=>s.ResNetPreTrainedModel,RoFormerForMaskedLM:()=>s.RoFormerForMaskedLM,RoFormerForQuestionAnswering:()=>s.RoFormerForQuestionAnswering,RoFormerForSequenceClassification:()=>s.RoFormerForSequenceClassification,RoFormerForTokenClassification:()=>s.RoFormerForTokenClassification,RoFormerModel:()=>s.RoFormerModel,RoFormerPreTrainedModel:()=>s.RoFormerPreTrainedModel,RoFormerTokenizer:()=>r.RoFormerTokenizer,RobertaForMaskedLM:()=>s.RobertaForMaskedLM,RobertaForQuestionAnswering:()=>s.RobertaForQuestionAnswering,RobertaForSequenceClassification:()=>s.RobertaForSequenceClassification,RobertaForTokenClassification:()=>s.RobertaForTokenClassification,RobertaModel:()=>s.RobertaModel,RobertaPreTrainedModel:()=>s.RobertaPreTrainedModel,RobertaTokenizer:()=>r.RobertaTokenizer,SamImageProcessor:()=>g.SamImageProcessor,SamImageSegmentationOutput:()=>s.SamImageSegmentationOutput,SamModel:()=>s.SamModel,SamPreTrainedModel:()=>s.SamPreTrainedModel,SamProcessor:()=>M.SamProcessor,SapiensForDepthEstimation:()=>s.SapiensForDepthEstimation,SapiensForNormalEstimation:()=>s.SapiensForNormalEstimation,SapiensForSemanticSegmentation:()=>s.SapiensForSemanticSegmentation,SapiensPreTrainedModel:()=>s.SapiensPreTrainedModel,SeamlessM4TFeatureExtractor:()=>p.SeamlessM4TFeatureExtractor,SegformerFeatureExtractor:()=>g.SegformerFeatureExtractor,SegformerForImageClassification:()=>s.SegformerForImageClassification,SegformerForSemanticSegmentation:()=>s.SegformerForSemanticSegmentation,SegformerImageProcessor:()=>g.SegformerImageProcessor,SegformerModel:()=>s.SegformerModel,SegformerPreTrainedModel:()=>s.SegformerPreTrainedModel,Seq2SeqLMOutput:()=>s.Seq2SeqLMOutput,SequenceClassifierOutput:()=>s.SequenceClassifierOutput,SiglipImageProcessor:()=>g.SiglipImageProcessor,SiglipModel:()=>s.SiglipModel,SiglipPreTrainedModel:()=>s.SiglipPreTrainedModel,SiglipTextModel:()=>s.SiglipTextModel,SiglipTokenizer:()=>r.SiglipTokenizer,SiglipVisionModel:()=>s.SiglipVisionModel,SmolVLMForConditionalGeneration:()=>s.SmolVLMForConditionalGeneration,SmolVLMImageProcessor:()=>g.SmolVLMImageProcessor,SmolVLMProcessor:()=>M.SmolVLMProcessor,SnacDecoderModel:()=>s.SnacDecoderModel,SnacEncoderModel:()=>s.SnacEncoderModel,SnacFeatureExtractor:()=>p.SnacFeatureExtractor,SnacModel:()=>s.SnacModel,SnacPreTrainedModel:()=>s.SnacPreTrainedModel,SpeechT5FeatureExtractor:()=>p.SpeechT5FeatureExtractor,SpeechT5ForSpeechToText:()=>s.SpeechT5ForSpeechToText,SpeechT5ForTextToSpeech:()=>s.SpeechT5ForTextToSpeech,SpeechT5HifiGan:()=>s.SpeechT5HifiGan,SpeechT5Model:()=>s.SpeechT5Model,SpeechT5PreTrainedModel:()=>s.SpeechT5PreTrainedModel,SpeechT5Processor:()=>M.SpeechT5Processor,SpeechT5Tokenizer:()=>r.SpeechT5Tokenizer,SqueezeBertForMaskedLM:()=>s.SqueezeBertForMaskedLM,SqueezeBertForQuestionAnswering:()=>s.SqueezeBertForQuestionAnswering,SqueezeBertForSequenceClassification:()=>s.SqueezeBertForSequenceClassification,SqueezeBertModel:()=>s.SqueezeBertModel,SqueezeBertPreTrainedModel:()=>s.SqueezeBertPreTrainedModel,SqueezeBertTokenizer:()=>r.SqueezeBertTokenizer,StableLmForCausalLM:()=>s.StableLmForCausalLM,StableLmModel:()=>s.StableLmModel,StableLmPreTrainedModel:()=>s.StableLmPreTrainedModel,Starcoder2ForCausalLM:()=>s.Starcoder2ForCausalLM,Starcoder2Model:()=>s.Starcoder2Model,Starcoder2PreTrainedModel:()=>s.Starcoder2PreTrainedModel,StoppingCriteria:()=>k.StoppingCriteria,StoppingCriteriaList:()=>k.StoppingCriteriaList,StyleTextToSpeech2Model:()=>s.StyleTextToSpeech2Model,StyleTextToSpeech2PreTrainedModel:()=>s.StyleTextToSpeech2PreTrainedModel,SummarizationPipeline:()=>t.SummarizationPipeline,SuppressTokensAtBeginLogitsProcessor:()=>y.SuppressTokensAtBeginLogitsProcessor,Swin2SRForImageSuperResolution:()=>s.Swin2SRForImageSuperResolution,Swin2SRImageProcessor:()=>g.Swin2SRImageProcessor,Swin2SRModel:()=>s.Swin2SRModel,Swin2SRPreTrainedModel:()=>s.Swin2SRPreTrainedModel,SwinForImageClassification:()=>s.SwinForImageClassification,SwinForSemanticSegmentation:()=>s.SwinForSemanticSegmentation,SwinModel:()=>s.SwinModel,SwinPreTrainedModel:()=>s.SwinPreTrainedModel,T5ForConditionalGeneration:()=>s.T5ForConditionalGeneration,T5Model:()=>s.T5Model,T5PreTrainedModel:()=>s.T5PreTrainedModel,T5Tokenizer:()=>r.T5Tokenizer,TableTransformerForObjectDetection:()=>s.TableTransformerForObjectDetection,TableTransformerModel:()=>s.TableTransformerModel,TableTransformerObjectDetectionOutput:()=>s.TableTransformerObjectDetectionOutput,TableTransformerPreTrainedModel:()=>s.TableTransformerPreTrainedModel,TemperatureLogitsWarper:()=>y.TemperatureLogitsWarper,Tensor:()=>d.Tensor,Text2TextGenerationPipeline:()=>t.Text2TextGenerationPipeline,TextClassificationPipeline:()=>t.TextClassificationPipeline,TextGenerationPipeline:()=>t.TextGenerationPipeline,TextStreamer:()=>b.TextStreamer,TextToAudioPipeline:()=>t.TextToAudioPipeline,TokenClassificationPipeline:()=>t.TokenClassificationPipeline,TokenClassifierOutput:()=>s.TokenClassifierOutput,TokenizerModel:()=>r.TokenizerModel,TopKLogitsWarper:()=>y.TopKLogitsWarper,TopPLogitsWarper:()=>y.TopPLogitsWarper,TrOCRForCausalLM:()=>s.TrOCRForCausalLM,TrOCRPreTrainedModel:()=>s.TrOCRPreTrainedModel,TranslationPipeline:()=>t.TranslationPipeline,UltravoxModel:()=>s.UltravoxModel,UltravoxPreTrainedModel:()=>s.UltravoxPreTrainedModel,UltravoxProcessor:()=>M.UltravoxProcessor,UniSpeechForCTC:()=>s.UniSpeechForCTC,UniSpeechForSequenceClassification:()=>s.UniSpeechForSequenceClassification,UniSpeechModel:()=>s.UniSpeechModel,UniSpeechPreTrainedModel:()=>s.UniSpeechPreTrainedModel,UniSpeechSatForAudioFrameClassification:()=>s.UniSpeechSatForAudioFrameClassification,UniSpeechSatForCTC:()=>s.UniSpeechSatForCTC,UniSpeechSatForSequenceClassification:()=>s.UniSpeechSatForSequenceClassification,UniSpeechSatModel:()=>s.UniSpeechSatModel,UniSpeechSatPreTrainedModel:()=>s.UniSpeechSatPreTrainedModel,VLChatProcessor:()=>M.VLChatProcessor,VLMImageProcessor:()=>g.VLMImageProcessor,ViTFeatureExtractor:()=>g.ViTFeatureExtractor,ViTForImageClassification:()=>s.ViTForImageClassification,ViTImageProcessor:()=>g.ViTImageProcessor,ViTMAEModel:()=>s.ViTMAEModel,ViTMAEPreTrainedModel:()=>s.ViTMAEPreTrainedModel,ViTMSNForImageClassification:()=>s.ViTMSNForImageClassification,ViTMSNModel:()=>s.ViTMSNModel,ViTMSNPreTrainedModel:()=>s.ViTMSNPreTrainedModel,ViTModel:()=>s.ViTModel,ViTPreTrainedModel:()=>s.ViTPreTrainedModel,VisionEncoderDecoderModel:()=>s.VisionEncoderDecoderModel,VitMatteForImageMatting:()=>s.VitMatteForImageMatting,VitMatteImageProcessor:()=>g.VitMatteImageProcessor,VitMattePreTrainedModel:()=>s.VitMattePreTrainedModel,VitPoseForPoseEstimation:()=>s.VitPoseForPoseEstimation,VitPoseImageProcessor:()=>g.VitPoseImageProcessor,VitPosePreTrainedModel:()=>s.VitPosePreTrainedModel,VitsModel:()=>s.VitsModel,VitsModelOutput:()=>s.VitsModelOutput,VitsPreTrainedModel:()=>s.VitsPreTrainedModel,VitsTokenizer:()=>r.VitsTokenizer,Wav2Vec2BertForCTC:()=>s.Wav2Vec2BertForCTC,Wav2Vec2BertForSequenceClassification:()=>s.Wav2Vec2BertForSequenceClassification,Wav2Vec2BertModel:()=>s.Wav2Vec2BertModel,Wav2Vec2BertPreTrainedModel:()=>s.Wav2Vec2BertPreTrainedModel,Wav2Vec2CTCTokenizer:()=>r.Wav2Vec2CTCTokenizer,Wav2Vec2FeatureExtractor:()=>p.Wav2Vec2FeatureExtractor,Wav2Vec2ForAudioFrameClassification:()=>s.Wav2Vec2ForAudioFrameClassification,Wav2Vec2ForCTC:()=>s.Wav2Vec2ForCTC,Wav2Vec2ForSequenceClassification:()=>s.Wav2Vec2ForSequenceClassification,Wav2Vec2Model:()=>s.Wav2Vec2Model,Wav2Vec2PreTrainedModel:()=>s.Wav2Vec2PreTrainedModel,Wav2Vec2Processor:()=>M.Wav2Vec2Processor,Wav2Vec2ProcessorWithLM:()=>M.Wav2Vec2ProcessorWithLM,WavLMForAudioFrameClassification:()=>s.WavLMForAudioFrameClassification,WavLMForCTC:()=>s.WavLMForCTC,WavLMForSequenceClassification:()=>s.WavLMForSequenceClassification,WavLMForXVector:()=>s.WavLMForXVector,WavLMModel:()=>s.WavLMModel,WavLMPreTrainedModel:()=>s.WavLMPreTrainedModel,WeSpeakerFeatureExtractor:()=>p.WeSpeakerFeatureExtractor,WeSpeakerResNetModel:()=>s.WeSpeakerResNetModel,WeSpeakerResNetPreTrainedModel:()=>s.WeSpeakerResNetPreTrainedModel,WhisperFeatureExtractor:()=>p.WhisperFeatureExtractor,WhisperForConditionalGeneration:()=>s.WhisperForConditionalGeneration,WhisperModel:()=>s.WhisperModel,WhisperPreTrainedModel:()=>s.WhisperPreTrainedModel,WhisperProcessor:()=>M.WhisperProcessor,WhisperTextStreamer:()=>b.WhisperTextStreamer,WhisperTimeStampLogitsProcessor:()=>y.WhisperTimeStampLogitsProcessor,WhisperTokenizer:()=>r.WhisperTokenizer,XLMForQuestionAnswering:()=>s.XLMForQuestionAnswering,XLMForSequenceClassification:()=>s.XLMForSequenceClassification,XLMForTokenClassification:()=>s.XLMForTokenClassification,XLMModel:()=>s.XLMModel,XLMPreTrainedModel:()=>s.XLMPreTrainedModel,XLMRobertaForMaskedLM:()=>s.XLMRobertaForMaskedLM,XLMRobertaForQuestionAnswering:()=>s.XLMRobertaForQuestionAnswering,XLMRobertaForSequenceClassification:()=>s.XLMRobertaForSequenceClassification,XLMRobertaForTokenClassification:()=>s.XLMRobertaForTokenClassification,XLMRobertaModel:()=>s.XLMRobertaModel,XLMRobertaPreTrainedModel:()=>s.XLMRobertaPreTrainedModel,XLMRobertaTokenizer:()=>r.XLMRobertaTokenizer,XLMTokenizer:()=>r.XLMTokenizer,XLMWithLMHeadModel:()=>s.XLMWithLMHeadModel,XVectorOutput:()=>s.XVectorOutput,YolosFeatureExtractor:()=>g.YolosFeatureExtractor,YolosForObjectDetection:()=>s.YolosForObjectDetection,YolosImageProcessor:()=>g.YolosImageProcessor,YolosModel:()=>s.YolosModel,YolosObjectDetectionOutput:()=>s.YolosObjectDetectionOutput,YolosPreTrainedModel:()=>s.YolosPreTrainedModel,ZeroShotAudioClassificationPipeline:()=>t.ZeroShotAudioClassificationPipeline,ZeroShotClassificationPipeline:()=>t.ZeroShotClassificationPipeline,ZeroShotImageClassificationPipeline:()=>t.ZeroShotImageClassificationPipeline,ZeroShotObjectDetectionPipeline:()=>t.ZeroShotObjectDetectionPipeline,bankers_round:()=>u.bankers_round,cat:()=>d.cat,cos_sim:()=>u.cos_sim,dot:()=>u.dot,dynamic_time_warping:()=>u.dynamic_time_warping,env:()=>e.env,full:()=>d.full,full_like:()=>d.full_like,getKeyValueShapes:()=>i.getKeyValueShapes,hamming:()=>a.hamming,hanning:()=>a.hanning,interpolate:()=>d.interpolate,interpolate_4d:()=>d.interpolate_4d,interpolate_data:()=>u.interpolate_data,is_chinese_char:()=>r.is_chinese_char,layer_norm:()=>d.layer_norm,load_image:()=>l.load_image,load_video:()=>c.load_video,log_softmax:()=>u.log_softmax,magnitude:()=>u.magnitude,matmul:()=>d.matmul,max:()=>u.max,mean:()=>d.mean,mean_pooling:()=>d.mean_pooling,medianFilter:()=>u.medianFilter,mel_filter_bank:()=>a.mel_filter_bank,min:()=>u.min,ones:()=>d.ones,ones_like:()=>d.ones_like,permute:()=>d.permute,permute_data:()=>u.permute_data,pipeline:()=>t.pipeline,quantize_embeddings:()=>d.quantize_embeddings,rand:()=>d.rand,read_audio:()=>a.read_audio,rfft:()=>d.rfft,round:()=>u.round,slice:()=>d.slice,softmax:()=>u.softmax,spectrogram:()=>a.spectrogram,stack:()=>d.stack,std_mean:()=>d.std_mean,topk:()=>d.topk,window_function:()=>a.window_function,zeros:()=>d.zeros,zeros_like:()=>d.zeros_like});var e=o("./src/env.js"),t=o("./src/pipelines.js"),s=o("./src/models.js"),r=o("./src/tokenizers.js"),i=o("./src/configs.js"),a=o("./src/utils/audio.js"),l=o("./src/utils/image.js"),c=o("./src/utils/video.js"),d=o("./src/utils/tensor.js"),u=o("./src/utils/maths.js"),_=o("./src/base/feature_extraction_utils.js"),p=o("./src/models/feature_extractors.js"),m=o("./src/models/auto/feature_extraction_auto.js"),h=o("./src/base/image_processors_utils.js"),g=o("./src/models/image_processors.js"),f=o("./src/models/auto/image_processing_auto.js"),w=o("./src/base/processing_utils.js"),M=o("./src/models/processors.js"),x=o("./src/models/auto/processing_auto.js"),b=o("./src/generation/streamers.js"),k=o("./src/generation/stopping_criteria.js"),y=o("./src/generation/logits_process.js")})();var i=exports;for(var a in n)i[a]=n[a];n.__esModule&&Object.defineProperty(i,"__esModule",{value:!0})})();
+(()=>{var e,t,s={fs:e=>{"use strict";e.exports=require("fs")},"onnxruntime-common":e=>{"use strict";e.exports=require("onnxruntime-common")},"onnxruntime-node":e=>{"use strict";e.exports=require("onnxruntime-node")},path:e=>{"use strict";e.exports=require("path")},sharp:e=>{"use strict";e.exports=require("sharp")},url:e=>{"use strict";e.exports=require("url")},"?8b6b":()=>{},"./node_modules/@huggingface/jinja/dist/index.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Environment:()=>te,Interpreter:()=>se,Template:()=>_e,parse:()=>V,tokenize:()=>d});var r=Object.freeze({Text:"Text",NumericLiteral:"NumericLiteral",BooleanLiteral:"BooleanLiteral",NullLiteral:"NullLiteral",StringLiteral:"StringLiteral",Identifier:"Identifier",Equals:"Equals",OpenParen:"OpenParen",CloseParen:"CloseParen",OpenStatement:"OpenStatement",CloseStatement:"CloseStatement",OpenExpression:"OpenExpression",CloseExpression:"CloseExpression",OpenSquareBracket:"OpenSquareBracket",CloseSquareBracket:"CloseSquareBracket",OpenCurlyBracket:"OpenCurlyBracket",CloseCurlyBracket:"CloseCurlyBracket",Comma:"Comma",Dot:"Dot",Colon:"Colon",Pipe:"Pipe",CallOperator:"CallOperator",AdditiveBinaryOperator:"AdditiveBinaryOperator",MultiplicativeBinaryOperator:"MultiplicativeBinaryOperator",ComparisonBinaryOperator:"ComparisonBinaryOperator",UnaryOperator:"UnaryOperator",Set:"Set",If:"If",For:"For",In:"In",Is:"Is",NotIn:"NotIn",Else:"Else",EndSet:"EndSet",EndIf:"EndIf",ElseIf:"ElseIf",EndFor:"EndFor",And:"And",Or:"Or",Not:"UnaryOperator",Macro:"Macro",EndMacro:"EndMacro",Break:"Break",Continue:"Continue"}),o=Object.freeze({set:r.Set,for:r.For,in:r.In,is:r.Is,if:r.If,else:r.Else,endset:r.EndSet,endif:r.EndIf,elif:r.ElseIf,endfor:r.EndFor,and:r.And,or:r.Or,not:r.Not,"not in":r.NotIn,macro:r.Macro,endmacro:r.EndMacro,break:r.Break,continue:r.Continue,true:r.BooleanLiteral,false:r.BooleanLiteral,none:r.NullLiteral,True:r.BooleanLiteral,False:r.BooleanLiteral,None:r.NullLiteral}),n=class{constructor(e,t){this.value=e,this.type=t}};function i(e){return/\w/.test(e)}function a(e){return/[0-9]/.test(e)}var l=[["{%",r.OpenStatement],["%}",r.CloseStatement],["{{",r.OpenExpression],["}}",r.CloseExpression],["(",r.OpenParen],[")",r.CloseParen],["{",r.OpenCurlyBracket],["}",r.CloseCurlyBracket],["[",r.OpenSquareBracket],["]",r.CloseSquareBracket],[",",r.Comma],[".",r.Dot],[":",r.Colon],["|",r.Pipe],["<=",r.ComparisonBinaryOperator],[">=",r.ComparisonBinaryOperator],["==",r.ComparisonBinaryOperator],["!=",r.ComparisonBinaryOperator],["<",r.ComparisonBinaryOperator],[">",r.ComparisonBinaryOperator],["+",r.AdditiveBinaryOperator],["-",r.AdditiveBinaryOperator],["*",r.MultiplicativeBinaryOperator],["/",r.MultiplicativeBinaryOperator],["%",r.MultiplicativeBinaryOperator],["=",r.Equals]],c=new Map([["n","\n"],["t","\t"],["r","\r"],["b","\b"],["f","\f"],["v","\v"],["'","'"],['"','"'],["\\","\\"]]);function d(e,t={}){const s=[],d=function(e,t={}){return e.endsWith("\n")&&(e=e.slice(0,-1)),e=e.replace(/{#.*?#}/gs,"{##}"),t.lstrip_blocks&&(e=e.replace(/^[ \t]*({[#%])/gm,"$1")),t.trim_blocks&&(e=e.replace(/([#%]})\n/g,"$1")),e.replace(/{##}/g,"").replace(/-%}\s*/g,"%}").replace(/\s*{%-/g,"{%").replace(/-}}\s*/g,"}}").replace(/\s*{{-/g,"{{")}(e,t);let u=0;const _=e=>{let t="";for(;e(d[u]);)if("\\"!==d[u]){if(t+=d[u++],u>=d.length)throw new SyntaxError("Unexpected end of input")}else{if(++u,u>=d.length)throw new SyntaxError("Unexpected end of input");const e=d[u++],s=c.get(e);if(void 0===s)throw new SyntaxError(`Unexpected escaped character: ${e}`);t+=s}return t};e:for(;u<d.length;){const e=s.at(-1)?.type;if(void 0===e||e===r.CloseStatement||e===r.CloseExpression){let e="";for(;u<d.length&&("{"!==d[u]||"%"!==d[u+1]&&"{"!==d[u+1]);)e+=d[u++];if(e.length>0){s.push(new n(e,r.Text));continue}}_((e=>/\s/.test(e)));const t=d[u];if("-"===t||"+"===t){const e=s.at(-1)?.type;if(e===r.Text||void 0===e)throw new SyntaxError(`Unexpected character: ${t}`);switch(e){case r.Identifier:case r.NumericLiteral:case r.BooleanLiteral:case r.NullLiteral:case r.StringLiteral:case r.CloseParen:case r.CloseSquareBracket:break;default:{++u;const e=_(a);s.push(new n(`${t}${e}`,e.length>0?r.NumericLiteral:r.UnaryOperator));continue}}}for(const[e,t]of l){if(d.slice(u,u+e.length)===e){s.push(new n(e,t)),u+=e.length;continue e}}if("'"!==t&&'"'!==t)if(a(t)){const e=_(a);s.push(new n(e,r.NumericLiteral))}else{if(!i(t))throw new SyntaxError(`Unexpected character: ${t}`);{const e=_(i),t=Object.hasOwn(o,e)?o[e]:r.Identifier;t===r.In&&s.at(-1)?.type===r.Not?(s.pop(),s.push(new n("not in",r.NotIn))):s.push(new n(e,t))}}else{++u;const e=_((e=>e!==t));s.push(new n(e,r.StringLiteral)),++u}}return s}var u=class{type="Statement"},_=class extends u{constructor(e){super(),this.body=e}type="Program"},p=class extends u{constructor(e,t,s){super(),this.test=e,this.body=t,this.alternate=s}type="If"},m=class extends u{constructor(e,t,s,r){super(),this.loopvar=e,this.iterable=t,this.body=s,this.defaultBlock=r}type="For"},h=class extends u{type="Break"},g=class extends u{type="Continue"},f=class extends u{constructor(e,t,s){super(),this.assignee=e,this.value=t,this.body=s}type="Set"},w=class extends u{constructor(e,t,s){super(),this.name=e,this.args=t,this.body=s}type="Macro"},M=class extends u{type="Expression"},x=class extends M{constructor(e,t,s){super(),this.object=e,this.property=t,this.computed=s}type="MemberExpression"},b=class extends M{constructor(e,t){super(),this.callee=e,this.args=t}type="CallExpression"},k=class extends M{constructor(e){super(),this.value=e}type="Identifier"},y=class extends M{constructor(e){super(),this.value=e}type="Literal"},v=class extends y{type="NumericLiteral"},T=class extends y{type="StringLiteral"},P=class extends y{type="BooleanLiteral"},F=class extends y{type="NullLiteral"},C=class extends y{type="ArrayLiteral"},S=class extends y{type="TupleLiteral"},A=class extends y{type="ObjectLiteral"},E=class extends M{constructor(e,t,s){super(),this.operator=e,this.left=t,this.right=s}type="BinaryExpression"},L=class extends M{constructor(e,t){super(),this.operand=e,this.filter=t}type="FilterExpression"},I=class extends M{constructor(e,t){super(),this.iterable=e,this.test=t}type="SelectExpression"},z=class extends M{constructor(e,t,s){super(),this.operand=e,this.negate=t,this.test=s}type="TestExpression"},j=class extends M{constructor(e,t){super(),this.operator=e,this.argument=t}type="UnaryExpression"},O=class extends M{constructor(e=void 0,t=void 0,s=void 0){super(),this.start=e,this.stop=t,this.step=s}type="SliceExpression"},D=class extends M{constructor(e,t){super(),this.key=e,this.value=t}type="KeywordArgumentExpression"};function V(e){const t=new _([]);let s=0;function o(t,r){const o=e[s++];if(!o||o.type!==t)throw new Error(`Parser Error: ${r}. ${o.type} !== ${t}.`);return o}function n(){switch(e[s].type){case r.Text:return new T(o(r.Text,"Expected text token").value);case r.OpenStatement:return function(){let t;switch(o(r.OpenStatement,"Expected opening statement token"),e[s].type){case r.Set:++s,t=function(){const t=d();if(a(r.Equals)){++s;const e=d();return new f(t,e,[])}{const i=[];for(o(r.CloseStatement,"Expected %} token");e[s]?.type!==r.OpenStatement||e[s+1]?.type!==r.EndSet;){const e=n();i.push(e)}return o(r.OpenStatement,"Expected {% token"),o(r.EndSet,"Expected endset token"),new f(t,null,i)}}(),o(r.CloseStatement,"Expected closing statement token");break;case r.If:++s,t=l(),o(r.OpenStatement,"Expected {% token"),o(r.EndIf,"Expected endif token"),o(r.CloseStatement,"Expected %} token");break;case r.Macro:++s,t=function(){const e=W();if("Identifier"!==e.type)throw new SyntaxError("Expected identifier following macro statement");const t=B();o(r.CloseStatement,"Expected closing statement token");const s=[];for(;i(r.OpenStatement,r.EndMacro);)s.push(n());return new w(e,t,s)}(),o(r.OpenStatement,"Expected {% token"),o(r.EndMacro,"Expected endmacro token"),o(r.CloseStatement,"Expected %} token");break;case r.For:++s,t=function(){const e=c(!0);if(!(e instanceof k||e instanceof S))throw new SyntaxError(`Expected identifier/tuple for the loop variable, got ${e.type} instead`);o(r.In,"Expected `in` keyword following loop variable");const t=d();o(r.CloseStatement,"Expected closing statement token");const l=[];for(;i(r.OpenStatement,r.EndFor)&&i(r.OpenStatement,r.Else);)l.push(n());const u=[];if(a(r.OpenStatement,r.Else))for(++s,++s,o(r.CloseStatement,"Expected closing statement token");i(r.OpenStatement,r.EndFor);)u.push(n());return new m(e,t,l,u)}(),o(r.OpenStatement,"Expected {% token"),o(r.EndFor,"Expected endfor token"),o(r.CloseStatement,"Expected %} token");break;case r.Break:++s,o(r.CloseStatement,"Expected closing statement token"),t=new h;break;case r.Continue:++s,o(r.CloseStatement,"Expected closing statement token"),t=new g;break;default:throw new SyntaxError(`Unknown statement type: ${e[s].type}`)}return t}();case r.OpenExpression:return function(){o(r.OpenExpression,"Expected opening expression token");const e=d();return o(r.CloseExpression,"Expected closing expression token"),e}();default:throw new SyntaxError(`Unexpected token type: ${e[s].type}`)}}function i(...t){return s+t.length<=e.length&&t.some(((t,r)=>t!==e[s+r].type))}function a(...t){return s+t.length<=e.length&&t.every(((t,r)=>t===e[s+r].type))}function l(){const t=d();o(r.CloseStatement,"Expected closing statement token");const i=[],c=[];for(;e[s]?.type!==r.OpenStatement||e[s+1]?.type!==r.ElseIf&&e[s+1]?.type!==r.Else&&e[s+1]?.type!==r.EndIf;)i.push(n());if(e[s]?.type===r.OpenStatement&&e[s+1]?.type!==r.EndIf)if(++s,a(r.ElseIf))o(r.ElseIf,"Expected elseif token"),c.push(l());else for(o(r.Else,"Expected else token"),o(r.CloseStatement,"Expected closing statement token");e[s]?.type!==r.OpenStatement||e[s+1]?.type!==r.EndIf;)c.push(n());return new p(t,i,c)}function c(e=!1){const t=e?W:d,o=[t()],n=a(r.Comma);for(;n&&(++s,o.push(t()),a(r.Comma)););return n?new S(o):o[0]}function d(){return function(){const e=u();if(a(r.If)){++s;const t=u();if(a(r.Else)){++s;const r=u();return new p(t,[e],[r])}return new I(e,t)}return e}()}function u(){let t=M();for(;a(r.Or);){const r=e[s];++s;const o=M();t=new E(r,t,o)}return t}function M(){let t=y();for(;a(r.And);){const r=e[s];++s;const o=y();t=new E(r,t,o)}return t}function y(){let t;for(;a(r.Not);){const r=e[s];++s;const o=y();t=new j(r,o)}return t??function(){let t=V();for(;a(r.ComparisonBinaryOperator)||a(r.In)||a(r.NotIn);){const r=e[s];++s;const o=V();t=new E(r,t,o)}return t}()}function V(){let t=$();for(;a(r.AdditiveBinaryOperator);){const r=e[s];++s;const o=$();t=new E(r,t,o)}return t}function N(e){let t=new b(e,B());return t=G(t),a(r.OpenParen)&&(t=N(t)),t}function B(){o(r.OpenParen,"Expected opening parenthesis for arguments list");const e=function(){const e=[];for(;!a(r.CloseParen);){let t=d();if(a(r.Equals)){if(++s,!(t instanceof k))throw new SyntaxError("Expected identifier for keyword argument");const e=d();t=new D(t,e)}e.push(t),a(r.Comma)&&++s}return e}();return o(r.CloseParen,"Expected closing parenthesis for arguments list"),e}function R(){const e=[];let t=!1;for(;!a(r.CloseSquareBracket);)a(r.Colon)?(e.push(void 0),++s,t=!0):(e.push(d()),a(r.Colon)&&(++s,t=!0));if(0===e.length)throw new SyntaxError("Expected at least one argument for member/slice expression");if(t){if(e.length>3)throw new SyntaxError("Expected 0-3 arguments for slice expression");return new O(...e)}return e[0]}function G(t){for(;a(r.Dot)||a(r.OpenSquareBracket);){const n=e[s];let i;++s;const a=n.type!==r.Dot;if(a)i=R(),o(r.CloseSquareBracket,"Expected closing square bracket");else if(i=W(),"Identifier"!==i.type)throw new SyntaxError("Expected identifier following dot operator");t=new x(t,i,a)}return t}function $(){let t=q();for(;a(r.MultiplicativeBinaryOperator);){const r=e[s];++s;const o=q();t=new E(r,t,o)}return t}function q(){let e=function(){let e=function(){const e=G(W());return a(r.OpenParen)?N(e):e}();for(;a(r.Pipe);){++s;let t=W();if(!(t instanceof k))throw new SyntaxError("Expected identifier for the filter");a(r.OpenParen)&&(t=N(t)),e=new L(e,t)}return e}();for(;a(r.Is);){++s;const t=a(r.Not);t&&++s;let o=W();if(o instanceof P?o=new k(o.value.toString()):o instanceof F&&(o=new k("none")),!(o instanceof k))throw new SyntaxError("Expected identifier for the test");e=new z(e,t,o)}return e}function W(){const t=e[s];switch(t.type){case r.NumericLiteral:return++s,new v(Number(t.value));case r.StringLiteral:return++s,new T(t.value);case r.BooleanLiteral:return++s,new P("true"===t.value.toLowerCase());case r.NullLiteral:return++s,new F(null);case r.Identifier:return++s,new k(t.value);case r.OpenParen:{++s;const t=c();if(e[s].type!==r.CloseParen)throw new SyntaxError(`Expected closing parenthesis, got ${e[s].type} instead`);return++s,t}case r.OpenSquareBracket:{++s;const e=[];for(;!a(r.CloseSquareBracket);)e.push(d()),a(r.Comma)&&++s;return++s,new C(e)}case r.OpenCurlyBracket:{++s;const e=new Map;for(;!a(r.CloseCurlyBracket);){const t=d();o(r.Colon,"Expected colon between key and value in object literal");const n=d();e.set(t,n),a(r.Comma)&&++s}return++s,new A(e)}default:throw new SyntaxError(`Unexpected token: ${t.type}`)}}for(;s<e.length;)t.body.push(n());return t}function N(e,t,s=1){void 0===t&&(t=e,e=0);const r=[];for(let o=e;o<t;o+=s)r.push(o);return r}function B(e,t,s,r=1){const o=Math.sign(r);o>=0?(t=(t??=0)<0?Math.max(e.length+t,0):Math.min(t,e.length),s=(s??=e.length)<0?Math.max(e.length+s,0):Math.min(s,e.length)):(t=(t??=e.length-1)<0?Math.max(e.length+t,-1):Math.min(t,e.length-1),s=(s??=-1)<-1?Math.max(e.length+s,-1):Math.min(s,e.length-1));const n=[];for(let i=t;o*i<o*s;i+=r)n.push(e[i]);return n}function R(e){return e.replace(/\b\w/g,(e=>e.toUpperCase()))}var G=class extends Error{},$=class extends Error{},q=class{type="RuntimeValue";value;builtins=new Map;constructor(e=void 0){this.value=e}__bool__(){return new X(!!this.value)}},W=class extends q{type="NumericValue"},U=class extends q{type="StringValue";builtins=new Map([["upper",new K((()=>new U(this.value.toUpperCase())))],["lower",new K((()=>new U(this.value.toLowerCase())))],["strip",new K((()=>new U(this.value.trim())))],["title",new K((()=>new U(R(this.value))))],["length",new W(this.value.length)],["rstrip",new K((()=>new U(this.value.trimEnd())))],["lstrip",new K((()=>new U(this.value.trimStart())))],["startswith",new K((e=>{if(0===e.length)throw new Error("startswith() requires at least one argument");const t=e[0];if(!(t instanceof U))throw new Error("startswith() argument must be a string");return new X(this.value.startsWith(t.value))}))],["endswith",new K((e=>{if(0===e.length)throw new Error("endswith() requires at least one argument");const t=e[0];if(!(t instanceof U))throw new Error("endswith() argument must be a string");return new X(this.value.endsWith(t.value))}))],["split",new K((e=>{const t=e[0]??new Z;if(!(t instanceof U||t instanceof Z))throw new Error("sep argument must be a string or null");const s=e[1]??new W(-1);if(!(s instanceof W))throw new Error("maxsplit argument must be a number");let r=[];if(t instanceof Z){const e=this.value.trimStart();for(const{0:t,index:o}of e.matchAll(/\S+/g)){if(-1!==s.value&&r.length>=s.value&&void 0!==o){r.push(t+e.slice(o+t.length));break}r.push(t)}}else{if(""===t.value)throw new Error("empty separator");r=this.value.split(t.value),-1!==s.value&&r.length>s.value&&r.push(r.splice(s.value).join(t.value))}return new J(r.map((e=>new U(e))))}))]])},X=class extends q{type="BooleanValue"},Q=class extends q{type="ObjectValue";__bool__(){return new X(this.value.size>0)}builtins=new Map([["get",new K((([e,t])=>{if(!(e instanceof U))throw new Error(`Object key must be a string: got ${e.type}`);return this.value.get(e.value)??t??new Z}))],["items",new K((()=>new J(Array.from(this.value.entries()).map((([e,t])=>new J([new U(e),t]))))))]])},H=class extends Q{type="KeywordArgumentsValue"},J=class extends q{type="ArrayValue";builtins=new Map([["length",new W(this.value.length)]]);__bool__(){return new X(this.value.length>0)}},Y=class extends J{type="TupleValue"},K=class extends q{type="FunctionValue"},Z=class extends q{type="NullValue"},ee=class extends q{type="UndefinedValue"},te=class{constructor(e){this.parent=e}variables=new Map([["namespace",new K((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 K],["odd",e=>{if("NumericValue"!==e.type)throw new Error(`Cannot apply test "odd" to type: ${e.type}`);return e.value%2!=0}],["even",e=>{if("NumericValue"!==e.type)throw new Error(`Cannot apply test "even" to type: ${e.type}`);return e.value%2==0}],["false",e=>"BooleanValue"===e.type&&!e.value],["true",e=>"BooleanValue"===e.type&&e.value],["none",e=>"NullValue"===e.type],["string",e=>"StringValue"===e.type],["number",e=>"NumericValue"===e.type],["integer",e=>"NumericValue"===e.type&&Number.isInteger(e.value)],["iterable",e=>"ArrayValue"===e.type||"StringValue"===e.type],["mapping",e=>"ObjectValue"===e.type],["lower",e=>{const t=e.value;return"StringValue"===e.type&&t===t.toLowerCase()}],["upper",e=>{const t=e.value;return"StringValue"===e.type&&t===t.toUpperCase()}],["none",e=>"NullValue"===e.type],["defined",e=>"UndefinedValue"!==e.type],["undefined",e=>"UndefinedValue"===e.type],["equalto",(e,t)=>e.value===t.value],["eq",(e,t)=>e.value===t.value]]);set(e,t){return this.declareVariable(e,re(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: ${e}`)}lookupVariable(e){try{return this.resolve(e).variables.get(e)??new ee}catch{return new ee}}},se=class{global;constructor(e){this.global=e??new te}run(e){return this.evaluate(e,this.global)}evaluateBinaryExpression(e,t){const s=this.evaluate(e.left,t);switch(e.operator.value){case"and":return s.__bool__().value?this.evaluate(e.right,t):s;case"or":return s.__bool__().value?s:this.evaluate(e.right,t)}const r=this.evaluate(e.right,t);switch(e.operator.value){case"==":return new X(s.value==r.value);case"!=":return new X(s.value!=r.value)}if(s instanceof ee||r instanceof ee)throw new Error("Cannot perform operation on undefined values");if(s instanceof Z||r instanceof Z)throw new Error("Cannot perform operation on null values");if(s instanceof W&&r instanceof W)switch(e.operator.value){case"+":return new W(s.value+r.value);case"-":return new W(s.value-r.value);case"*":return new W(s.value*r.value);case"/":return new W(s.value/r.value);case"%":return new W(s.value%r.value);case"<":return new X(s.value<r.value);case">":return new X(s.value>r.value);case">=":return new X(s.value>=r.value);case"<=":return new X(s.value<=r.value)}else if(s instanceof J&&r instanceof J){if("+"===e.operator.value)return new J(s.value.concat(r.value))}else if(r instanceof J){const t=void 0!==r.value.find((e=>e.value===s.value));switch(e.operator.value){case"in":return new X(t);case"not in":return new X(!t)}}if((s instanceof U||r instanceof U)&&"+"===e.operator.value)return new U(s.value.toString()+r.value.toString());if(s instanceof U&&r instanceof U)switch(e.operator.value){case"in":return new X(r.value.includes(s.value));case"not in":return new X(!r.value.includes(s.value))}if(s instanceof U&&r instanceof Q)switch(e.operator.value){case"in":return new X(r.value.has(s.value));case"not in":return new X(!r.value.has(s.value))}throw new SyntaxError(`Unknown operator "${e.operator.value}" between ${s.type} and ${r.type}`)}evaluateArguments(e,t){const s=[],r=new Map;for(const o of e)if("KeywordArgumentExpression"===o.type){const e=o;r.set(e.key.value,this.evaluate(e.value,t))}else{if(r.size>0)throw new Error("Positional arguments must come before keyword arguments");s.push(this.evaluate(o,t))}return[s,r]}evaluateFilterExpression(e,t){const s=this.evaluate(e.operand,t);if("Identifier"===e.filter.type){const t=e.filter;if("tojson"===t.value)return new U(oe(s));if(s instanceof J)switch(t.value){case"list":return s;case"first":return s.value[0];case"last":return s.value[s.value.length-1];case"length":return new W(s.value.length);case"reverse":return new J(s.value.reverse());case"sort":return new J(s.value.sort(((e,t)=>{if(e.type!==t.type)throw new Error(`Cannot compare different types: ${e.type} and ${t.type}`);switch(e.type){case"NumericValue":return e.value-t.value;case"StringValue":return e.value.localeCompare(t.value);default:throw new Error(`Cannot compare type: ${e.type}`)}})));case"join":return new U(s.value.map((e=>e.value)).join(""));case"string":return new U(oe(s));default:throw new Error(`Unknown ArrayValue filter: ${t.value}`)}else if(s instanceof U)switch(t.value){case"length":return new W(s.value.length);case"upper":return new U(s.value.toUpperCase());case"lower":return new U(s.value.toLowerCase());case"title":return new U(R(s.value));case"capitalize":return new U(s.value.charAt(0).toUpperCase()+s.value.slice(1));case"trim":return new U(s.value.trim());case"indent":return new U(s.value.split("\n").map(((e,t)=>0===t||0===e.length?e:"    "+e)).join("\n"));case"join":case"string":return s;default:throw new Error(`Unknown StringValue filter: ${t.value}`)}else{if(s instanceof W){if("abs"===t.value)return new W(Math.abs(s.value));throw new Error(`Unknown NumericValue filter: ${t.value}`)}if(s instanceof Q)switch(t.value){case"items":return new J(Array.from(s.value.entries()).map((([e,t])=>new J([new U(e),t]))));case"length":return new W(s.value.size);default:throw new Error(`Unknown ObjectValue filter: ${t.value}`)}}throw new Error(`Cannot apply filter "${t.value}" to type: ${s.type}`)}if("CallExpression"===e.filter.type){const r=e.filter;if("Identifier"!==r.callee.type)throw new Error(`Unknown filter: ${r.callee.type}`);const o=r.callee.value;if("tojson"===o){const[,e]=this.evaluateArguments(r.args,t),o=e.get("indent")??new Z;if(!(o instanceof W||o instanceof Z))throw new Error("If set, indent must be a number");return new U(oe(s,o.value))}if("join"===o){let e;if(s instanceof U)e=Array.from(s.value);else{if(!(s instanceof J))throw new Error(`Cannot apply filter "${o}" to type: ${s.type}`);e=s.value.map((e=>e.value))}const[n,i]=this.evaluateArguments(r.args,t),a=n.at(0)??i.get("separator")??new U("");if(!(a instanceof U))throw new Error("separator must be a string");return new U(e.join(a.value))}if(s instanceof J){switch(o){case"selectattr":case"rejectattr":{const e="selectattr"===o;if(s.value.some((e=>!(e instanceof Q))))throw new Error(`\`${o}\` can only be applied to array of objects`);if(r.args.some((e=>"StringLiteral"!==e.type)))throw new Error(`arguments of \`${o}\` must be strings`);const[n,i,a]=r.args.map((e=>this.evaluate(e,t)));let l;if(i){const e=t.tests.get(i.value);if(!e)throw new Error(`Unknown test: ${i.value}`);l=e}else l=(...e)=>e[0].__bool__().value;const c=s.value.filter((t=>{const s=t.value.get(n.value),r=!!s&&l(s,a);return e?r:!r}));return new J(c)}case"map":{const[,e]=this.evaluateArguments(r.args,t);if(e.has("attribute")){const t=e.get("attribute");if(!(t instanceof U))throw new Error("attribute must be a string");const r=e.get("default"),o=s.value.map((e=>{if(!(e instanceof Q))throw new Error("items in map must be an object");return e.value.get(t.value)??r??new ee}));return new J(o)}throw new Error("`map` expressions without `attribute` set are not currently supported.")}}throw new Error(`Unknown ArrayValue filter: ${o}`)}if(s instanceof U){if("indent"===o){const[e,o]=this.evaluateArguments(r.args,t),n=e.at(0)??o.get("width")??new W(4);if(!(n instanceof W))throw new Error("width must be a number");const i=e.at(1)??o.get("first")??new X(!1),a=e.at(2)??o.get("blank")??new X(!1),l=s.value.split("\n"),c=" ".repeat(n.value),d=l.map(((e,t)=>!i.value&&0===t||!a.value&&0===e.length?e:c+e));return new U(d.join("\n"))}throw new Error(`Unknown StringValue filter: ${o}`)}throw new Error(`Cannot apply filter "${o}" to type: ${s.type}`)}throw new Error(`Unknown filter: ${e.filter.type}`)}evaluateTestExpression(e,t){const s=this.evaluate(e.operand,t),r=t.tests.get(e.test.value);if(!r)throw new Error(`Unknown test: ${e.test.value}`);const o=r(s);return new X(e.negate?!o:o)}evaluateUnaryExpression(e,t){const s=this.evaluate(e.argument,t);if("not"===e.operator.value)return new X(!s.value);throw new SyntaxError(`Unknown operator: ${e.operator.value}`)}evalProgram(e,t){return this.evaluateBlock(e.body,t)}evaluateBlock(e,t){let s="";for(const r of e){const e=this.evaluate(r,t);"NullValue"!==e.type&&"UndefinedValue"!==e.type&&(s+=e.value)}return new U(s)}evaluateIdentifier(e,t){return t.lookupVariable(e.value)}evaluateCallExpression(e,t){const[s,r]=this.evaluateArguments(e.args,t);r.size>0&&s.push(new H(r));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(s,t)}evaluateSliceExpression(e,t,s){if(!(e instanceof J||e instanceof U))throw new Error("Slice object must be an array or string");const r=this.evaluate(t.start,s),o=this.evaluate(t.stop,s),n=this.evaluate(t.step,s);if(!(r instanceof W||r instanceof ee))throw new Error("Slice start must be numeric or undefined");if(!(o instanceof W||o instanceof ee))throw new Error("Slice stop must be numeric or undefined");if(!(n instanceof W||n instanceof ee))throw new Error("Slice step must be numeric or undefined");return e instanceof J?new J(B(e.value,r.value,o.value,n.value)):new U(B(Array.from(e.value),r.value,o.value,n.value).join(""))}evaluateMemberExpression(e,t){const s=this.evaluate(e.object,t);let r,o;if(e.computed){if("SliceExpression"===e.property.type)return this.evaluateSliceExpression(s,e.property,t);r=this.evaluate(e.property,t)}else r=new U(e.property.value);if(s instanceof Q){if(!(r instanceof U))throw new Error(`Cannot access property with non-string: got ${r.type}`);o=s.value.get(r.value)??s.builtins.get(r.value)}else if(s instanceof J||s instanceof U)if(r instanceof W)o=s.value.at(r.value),s instanceof U&&(o=new U(s.value.at(r.value)));else{if(!(r instanceof U))throw new Error(`Cannot access property with non-string/non-number: got ${r.type}`);o=s.builtins.get(r.value)}else{if(!(r instanceof U))throw new Error(`Cannot access property with non-string: got ${r.type}`);o=s.builtins.get(r.value)}return o instanceof q?o:new ee}evaluateSet(e,t){const s=e.value?this.evaluate(e.value,t):this.evaluateBlock(e.body,t);if("Identifier"===e.assignee.type){const r=e.assignee.value;t.setVariable(r,s)}else{if("MemberExpression"!==e.assignee.type)throw new Error(`Invalid LHS inside assignment expression: ${JSON.stringify(e.assignee)}`);{const r=e.assignee,o=this.evaluate(r.object,t);if(!(o instanceof Q))throw new Error("Cannot assign to member of non-object");if("Identifier"!==r.property.type)throw new Error("Cannot assign to member with non-identifier property");o.value.set(r.property.value,s)}}return new Z}evaluateIf(e,t){const s=this.evaluate(e.test,t);return this.evaluateBlock(s.__bool__().value?e.body:e.alternate,t)}evaluateFor(e,t){const s=new te(t);let r,o;if("SelectExpression"===e.iterable.type){const t=e.iterable;o=this.evaluate(t.iterable,s),r=t.test}else o=this.evaluate(e.iterable,s);if(!(o instanceof J))throw new Error(`Expected iterable type in for loop: got ${o.type}`);const n=[],i=[];for(let t=0;t<o.value.length;++t){const a=new te(s),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 s=l;if(t.value.length!==s.value.length)throw new Error(`Too ${t.value.length>s.value.length?"few":"many"} items to unpack`);c=e=>{for(let r=0;r<t.value.length;++r){if("Identifier"!==t.value[r].type)throw new Error(`Cannot unpack non-identifier type: ${t.value[r].type}`);e.setVariable(t.value[r].value,s.value[r])}}}}if(r){c(a);if(!this.evaluate(r,a).__bool__().value)continue}n.push(l),i.push(c)}let a="",l=!0;for(let t=0;t<n.length;++t){const r=new Map([["index",new W(t+1)],["index0",new W(t)],["revindex",new W(n.length-t)],["revindex0",new W(n.length-t-1)],["first",new X(0===t)],["last",new X(t===n.length-1)],["length",new W(n.length)],["previtem",t>0?n[t-1]:new ee],["nextitem",t<n.length-1?n[t+1]:new ee]]);s.setVariable("loop",new Q(r)),i[t](s);try{a+=this.evaluateBlock(e.body,s).value}catch(e){if(e instanceof $)continue;if(e instanceof G)break;throw e}l=!1}if(l){a+=this.evaluateBlock(e.defaultBlock,s).value}return new U(a)}evaluateMacro(e,t){return t.setVariable(e.name.value,new K(((t,s)=>{const r=new te(s);let o;t=t.slice(),"KeywordArgumentsValue"===t.at(-1)?.type&&(o=t.pop());for(let s=0;s<e.args.length;++s){const n=e.args[s],i=t[s];if("Identifier"===n.type){const e=n;if(!i)throw new Error(`Missing positional argument: ${e.value}`);r.setVariable(e.value,i)}else{if("KeywordArgumentExpression"!==n.type)throw new Error(`Unknown argument type: ${n.type}`);{const e=n,t=i??o?.value.get(e.key.value)??this.evaluate(e.value,r);r.setVariable(e.key.value,t)}}}return this.evaluateBlock(e.body,r)}))),new Z}evaluate(e,t){if(void 0===e)return new ee;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"Break":throw new G;case"Continue":throw new $;case"NumericLiteral":return new W(Number(e.value));case"StringLiteral":return new U(e.value);case"BooleanLiteral":return new X(e.value);case"NullLiteral":return new Z(e.value);case"ArrayLiteral":return new J(e.value.map((e=>this.evaluate(e,t))));case"TupleLiteral":return new Y(e.value.map((e=>this.evaluate(e,t))));case"ObjectLiteral":{const s=new Map;for(const[r,o]of e.value){const e=this.evaluate(r,t);if(!(e instanceof U))throw new Error(`Object keys must be strings: got ${e.type}`);s.set(e.value,this.evaluate(o,t))}return new Q(s)}case"Identifier":return this.evaluateIdentifier(e,t);case"CallExpression":return this.evaluateCallExpression(e,t);case"MemberExpression":return this.evaluateMemberExpression(e,t);case"UnaryExpression":return this.evaluateUnaryExpression(e,t);case"BinaryExpression":return this.evaluateBinaryExpression(e,t);case"FilterExpression":return this.evaluateFilterExpression(e,t);case"TestExpression":return this.evaluateTestExpression(e,t);default:throw new SyntaxError(`Unknown node type: ${e.type}`)}}};function re(e){switch(typeof e){case"number":return new W(e);case"string":return new U(e);case"boolean":return new X(e);case"undefined":return new ee;case"object":return null===e?new Z:Array.isArray(e)?new J(e.map(re)):new Q(new Map(Object.entries(e).map((([e,t])=>[e,re(t)]))));case"function":return new K(((t,s)=>re(e(...t.map((e=>e.value)))??null)));default:throw new Error(`Cannot convert to runtime value: ${e}`)}}function oe(e,t,s){const r=s??0;switch(e.type){case"NullValue":case"UndefinedValue":return"null";case"NumericValue":case"StringValue":case"BooleanValue":return JSON.stringify(e.value);case"ArrayValue":case"ObjectValue":{const s=t?" ".repeat(t):"",o="\n"+s.repeat(r),n=o+s;if("ArrayValue"===e.type){const s=e.value.map((e=>oe(e,t,r+1)));return t?`[${n}${s.join(`,${n}`)}${o}]`:`[${s.join(", ")}]`}{const s=Array.from(e.value.entries()).map((([e,s])=>{const o=`"${e}": ${oe(s,t,r+1)}`;return t?`${n}${o}`:o}));return t?`{${s.join(",")}${o}}`:`{${s.join(", ")}}`}}default:throw new Error(`Cannot convert to JSON: ${e.type}`)}}var ne="\n",ie="{%- ",ae=" -%}",le={MultiplicativeBinaryOperator:2,AdditiveBinaryOperator:1,ComparisonBinaryOperator:0};function ce(...e){return ie+e.join(" ")+ae}function de(e,t,s){return e.map((e=>function(e,t,s){const r=s.repeat(t);switch(e.type){case"Program":return de(e.body,t,s);case"If":return function(e,t,s){const r=s.repeat(t),o=[];let n=e;for(;n&&(o.push({test:n.test,body:n.body}),1===n.alternate.length&&"If"===n.alternate[0].type);)n=n.alternate[0];let i=r+ce("if",ue(o[0].test))+ne+de(o[0].body,t+1,s);for(let e=1;e<o.length;e++)i+=ne+r+ce("elif",ue(o[e].test))+ne+de(o[e].body,t+1,s);n&&n.alternate.length>0&&(i+=ne+r+ce("else")+ne+de(n.alternate,t+1,s));return i+=ne+r+ce("endif"),i}(e,t,s);case"For":return function(e,t,s){const r=s.repeat(t);let o="";if("SelectExpression"===e.iterable.type){const t=e.iterable;o=`${ue(t.iterable)} if ${ue(t.test)}`}else o=ue(e.iterable);let n=r+ce("for",ue(e.loopvar),"in",o)+ne+de(e.body,t+1,s);e.defaultBlock.length>0&&(n+=ne+r+ce("else")+ne+de(e.defaultBlock,t+1,s));return n+=ne+r+ce("endfor"),n}(e,t,s);case"Set":return function(e,t,s){const r=s.repeat(t),o=ue(e.assignee),n=e.value?ue(e.value):"",i=r+ce("set",`${o}${e.value?" = "+n:""}`);if(0===e.body.length)return i;return i+ne+de(e.body,t+1,s)+ne+r+ce("endset")}(e,t,s);case"Macro":return function(e,t,s){const r=s.repeat(t),o=e.args.map(ue).join(", ");return r+ce("macro",`${e.name.value}(${o})`)+ne+de(e.body,t+1,s)+ne+r+ce("endmacro")}(e,t,s);case"Break":return r+ce("break");case"Continue":return r+ce("continue");default:return r+"{{- "+ue(e)+" -}}"}}(e,t,s))).join(ne)}function ue(e,t=-1){switch(e.type){case"Identifier":return e.value;case"NullLiteral":return"none";case"NumericLiteral":case"BooleanLiteral":return`${e.value}`;case"StringLiteral":return JSON.stringify(e.value);case"BinaryExpression":{const s=e,r=le[s.operator.type]??0,o=ue(s.left,r),n=ue(s.right,r+1),i=`${o} ${s.operator.value} ${n}`;return r<t?`(${i})`:i}case"UnaryExpression":{const t=e;return t.operator.value+("not"===t.operator.value?" ":"")+ue(t.argument,1/0)}case"LogicalNegationExpression":return`not ${ue(e.argument,1/0)}`;case"CallExpression":{const t=e,s=t.args.map((e=>ue(e,-1))).join(", ");return`${ue(t.callee,-1)}(${s})`}case"MemberExpression":{const t=e;let s=ue(t.object,-1);"Identifier"!==t.object.type&&(s=`(${s})`);let r=ue(t.property,-1);return t.computed||"Identifier"===t.property.type||(r=`(${r})`),t.computed?`${s}[${r}]`:`${s}.${r}`}case"FilterExpression":{const t=e,s=ue(t.operand,1/0);return"CallExpression"===t.filter.type?`${s} | ${ue(t.filter,-1)}`:`${s} | ${t.filter.value}`}case"SelectExpression":{const t=e;return`${ue(t.iterable,-1)} | select(${ue(t.test,-1)})`}case"TestExpression":{const t=e;return`${ue(t.operand,-1)} is${t.negate?" not":""} ${t.test.value}`}case"ArrayLiteral":case"TupleLiteral":{const t=e.value.map((e=>ue(e,-1))),s="ArrayLiteral"===e.type?"[]":"()";return`${s[0]}${t.join(", ")}${s[1]}`}case"ObjectLiteral":return`{ ${Array.from(e.value.entries()).map((([e,t])=>`${ue(e,-1)}: ${ue(t,-1)}`)).join(", ")} }`;case"SliceExpression":{const t=e;return`${t.start?ue(t.start,-1):""}:${t.stop?ue(t.stop,-1):""}${t.step?`:${ue(t.step,-1)}`:""}`}case"KeywordArgumentExpression":{const t=e;return`${t.key.value}=${ue(t.value,-1)}`}case"If":{const t=e,s=ue(t.test,-1);return`${ue(t.body[0],0)} if ${s} else ${ue(t.alternate[0],-1)}`}default:throw new Error(`Unknown expression type: ${e.type}`)}}var _e=class{parsed;constructor(e){const t=d(e,{lstrip_blocks:!0,trim_blocks:!0});this.parsed=V(t)}render(e){const t=new te;if(t.set("false",!1),t.set("true",!0),t.set("raise_exception",(e=>{throw new Error(e)})),t.set("range",N),e)for(const[s,r]of Object.entries(e))t.set(s,r);return new se(t).run(this.parsed).value}format(e){return function(e,t="\t"){const s="number"==typeof t?" ".repeat(t):t;return de(e.body,0,s).replace(/\n$/,"")}(this.parsed,e?.indent||"\t")}}},"./src/backends/onnx.js":(e,t,s)=>{"use strict";var r,o;s.r(t),s.d(t,{Tensor:()=>l.Tensor,createInferenceSession:()=>f,deviceToExecutionProviders:()=>h,isONNXProxy:()=>x,isONNXTensor:()=>w});var n=s("./src/env.js"),i=s("onnxruntime-node"),a=s("?8b6b"),l=s("onnxruntime-common");const c=Object.freeze({auto:null,gpu:null,cpu:"cpu",wasm:"wasm",webgpu:"webgpu",cuda:"cuda",dml:"dml",webnn:{name:"webnn",deviceType:"cpu"},"webnn-npu":{name:"webnn",deviceType:"npu"},"webnn-gpu":{name:"webnn",deviceType:"gpu"},"webnn-cpu":{name:"webnn",deviceType:"cpu"}}),d=[];let u,_;const p=Symbol.for("onnxruntime");if(p in globalThis)_=globalThis[p];else if(n.apis.IS_NODE_ENV){switch(_=i??(r||(r=s.t(i,2))),process.platform){case"win32":d.push("dml");break;case"linux":"x64"===process.arch&&d.push("cuda")}d.push("cpu"),u=["cpu"]}else _=o||(o=s.t(a,2)),n.apis.IS_WEBNN_AVAILABLE&&d.push("webnn-npu","webnn-gpu","webnn-cpu","webnn"),n.apis.IS_WEBGPU_AVAILABLE&&d.push("webgpu"),d.push("wasm"),u=["wasm"];const m=_.InferenceSession;function h(e=null){if(!e)return u;switch(e){case"auto":return d;case"gpu":return d.filter((e=>["webgpu","cuda","dml","webnn-gpu"].includes(e)))}if(d.includes(e))return[c[e]??e];throw new Error(`Unsupported device: "${e}". Should be one of: ${d.join(", ")}.`)}let g=null;async function f(e,t,s){g&&await g;const r=m.create(e,t);g??=r;const o=await r;return o.config=s,o}function w(e){return e instanceof _.Tensor}const M=_?.env;function x(){return M?.wasm?.proxy}M?.wasm&&("undefined"!=typeof ServiceWorkerGlobalScope&&self instanceof ServiceWorkerGlobalScope||M.wasm.wasmPaths||(M.wasm.wasmPaths=`https://cdn.jsdelivr.net/npm/@huggingface/transformers@${n.env.version}/dist/`),M.wasm.proxy=!1),M?.webgpu&&(M.webgpu.powerPreference="high-performance"),n.env.backends.onnx=M},"./src/base/feature_extraction_utils.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{FeatureExtractor:()=>i,validate_audio_inputs:()=>a});var r=s("./src/utils/constants.js"),o=s("./src/utils/generic.js"),n=s("./src/utils/hub.js");class i extends o.Callable{constructor(e){super(),this.config=e}static async from_pretrained(e,t){return new this(await(0,n.getModelJSON)(e,r.FEATURE_EXTRACTOR_NAME,!0,t))}}function a(e,t){if(!(e instanceof Float32Array||e instanceof Float64Array))throw new Error(`${t} expects input to be a Float32Array or a Float64Array, but got ${e?.constructor?.name??typeof e} instead. If using the feature extractor directly, remember to use \`read_audio(url, sampling_rate)\` to obtain the raw audio data of the file/url.`)}},"./src/base/image_processors_utils.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ImageProcessor:()=>M,center_to_corners_format:()=>u,post_process_instance_segmentation:()=>w,post_process_object_detection:()=>_,post_process_panoptic_segmentation:()=>f,post_process_semantic_segmentation:()=>p});var r=s("./src/utils/generic.js"),o=s("./src/utils/tensor.js"),n=s("./src/utils/maths.js"),i=(s("./src/utils/image.js"),s("./src/utils/core.js")),a=s("./src/utils/hub.js"),l=s("./src/utils/constants.js");function c(e,t,s=0,r=null){const o=e/t;let i=(0,n.bankers_round)(o)*t;return null!==r&&i>r&&(i=Math.floor(o)*t),i<s&&(i=Math.ceil(o)*t),i}function d([e,t],s){return[Math.max(Math.floor(e/s),1)*s,Math.max(Math.floor(t/s),1)*s]}function u([e,t,s,r]){return[e-s/2,t-r/2,e+s/2,t+r/2]}function _(e,t=.5,s=null,r=!1){const o=e.logits,i=e.pred_boxes,[a,l,c]=o.dims;if(null!==s&&s.length!==a)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");let d=[];for(let e=0;e<a;++e){let a=null!==s?s[e]:null,_={boxes:[],classes:[],scores:[]},p=o[e],m=i[e];for(let e=0;e<l;++e){let s,o=p[e],i=[];if(r){s=o.sigmoid().data;for(let e=0;e<s.length;++e)s[e]>t&&i.push(e)}else{let e=(0,n.max)(o.data)[1];if(e===c-1)continue;if(s=(0,n.softmax)(o.data),s[e]<t)continue;i.push(e)}for(const t of i){let r=m[e].data;r=u(r),null!==a&&(r=r.map(((e,t)=>e*a[(t+1)%2]))),_.boxes.push(r),_.classes.push(t),_.scores.push(s[t])}}d.push(_)}return d}function p(e,t=null){const s=e.logits,r=s.dims[0];if(null!==t&&t.length!==r)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");const n=[];for(let e=0;e<r;++e){const r=null!==t?t[e]:null;let i=s[e];null!==r&&(i=(0,o.interpolate)(i,r,"bilinear",!1));const[a,l]=r??i.dims.slice(-2),c=new o.Tensor("int32",new Int32Array(a*l),[a,l]),d=i[0].data,u=c.data;for(let e=1;e<i.dims[0];++e){const t=i[e].data;for(let s=0;s<t.length;++s)t[s]>d[s]&&(d[s]=t[s],u[s]=e)}const _=new Array(i.dims[0]);for(let e=0;e<u.length;++e){const t=u[e];_[t]=t}const p=_.filter((e=>void 0!==e));n.push({segmentation:c,labels:p})}return n}function m(e,t,s,r){const o=[],i=[],a=[];for(let l=0;l<e.dims[0];++l){const c=e[l],d=t[l],u=(0,n.max)(c.data)[1];if(u===r)continue;const _=(0,n.softmax)(c.data)[u];_>s&&(o.push(d),i.push(_),a.push(u))}return[o,i,a]}function h(e,t,s,r=.5,o=.8){const n=[];let i=0,a=0;const l=t[s].data;for(let t=0;t<e.length;++t)e[t]===s&&(n.push(t),++i),l[t]>=r&&++a;let c=i>0&&a>0;if(c){c=i/a>o}return[c,n]}function g(e,t,s,r,n,i=null,a=null){const[l,c]=a??e[0].dims,d=new o.Tensor("int32",new Int32Array(l*c),[l,c]),u=[];if(null!==a)for(let t=0;t<e.length;++t)e[t]=(0,o.interpolate)(e[t],a,"bilinear",!1);const _=new Int32Array(e[0].data.length),p=new Float32Array(e[0].data.length);for(let s=0;s<e.length;++s){let r=t[s];const o=e[s].data;for(let e=0;e<o.length;++e)o[e]*=r,o[e]>p[e]&&(_[e]=s,p[e]=o[e])}let m=0;const g=d.data;for(let o=0;o<s.length;++o){const i=s[o],[a,l]=h(_,e,o,r,n);if(a){++m;for(const e of l)g[e]=m;u.push({id:m,label_id:i,score:t[o]})}}return[d,u]}function f(e,t=.5,s=.5,r=.8,n=null,i=null){null===n&&(console.warn("`label_ids_to_fuse` unset. No instance will be fused."),n=new Set);const a=e.class_queries_logits??e.logits,l=(e.masks_queries_logits??e.pred_masks).sigmoid();let[c,d,u]=a.dims;if(u-=1,null!==i&&i.length!==c)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");let _=[];for(let e=0;e<c;++e){let c=null!==i?i[e]:null,d=a[e],p=l[e],[h,f,w]=m(d,p,t,u);if(0===w.length){let[e,t]=c??p.dims.slice(-2),s=new o.Tensor("int32",new Int32Array(e*t).fill(-1),[e,t]);_.push({segmentation:s,segments_info:[]});continue}let[M,x]=g(h,f,w,s,r,n,c);_.push({segmentation:M,segments_info:x})}return _}function w(e,t=.5,s=null){throw new Error("`post_process_instance_segmentation` is not yet implemented.")}class M extends r.Callable{constructor(e){super(),this.image_mean=e.image_mean??e.mean,this.image_std=e.image_std??e.std,this.resample=e.resample??2,this.do_rescale=e.do_rescale??!0,this.rescale_factor=e.rescale_factor??1/255,this.do_normalize=e.do_normalize,this.do_thumbnail=e.do_thumbnail,this.size=e.size??e.image_size,this.do_resize=e.do_resize??void 0!==this.size,this.size_divisibility=e.size_divisibility??e.size_divisor,this.do_center_crop=e.do_center_crop,this.crop_size=e.crop_size,this.do_convert_rgb=e.do_convert_rgb??!0,this.do_crop_margin=e.do_crop_margin,this.pad_size=e.pad_size,this.do_pad=e.do_pad,this.min_pixels=e.min_pixels,this.max_pixels=e.max_pixels,this.do_pad&&!this.pad_size&&this.size&&void 0!==this.size.width&&void 0!==this.size.height&&(this.pad_size=this.size),this.do_flip_channel_order=e.do_flip_channel_order??!1,this.config=e}async thumbnail(e,t,s=2){const r=e.height,o=e.width,n=t.height,i=t.width;let a=Math.min(r,n),l=Math.min(o,i);return a===r&&l===o?e:(r>o?l=Math.floor(o*a/r):o>r&&(a=Math.floor(r*l/o)),await e.resize(l,a,{resample:s}))}async crop_margin(e,t=200){const s=e.clone().grayscale(),r=(0,n.min)(s.data)[0],o=(0,n.max)(s.data)[0]-r;if(0===o)return e;const i=t/255;let a=s.width,l=s.height,c=0,d=0;const u=s.data;for(let e=0;e<s.height;++e){const t=e*s.width;for(let n=0;n<s.width;++n)(u[t+n]-r)/o<i&&(a=Math.min(a,n),l=Math.min(l,e),c=Math.max(c,n),d=Math.max(d,e))}return e=await e.crop([a,l,c,d])}pad_image(e,t,s,{mode:r="constant",center:o=!1,constant_values:n=0}={}){const[a,l,c]=t;let d,u;if("number"==typeof s?(d=s,u=s):"square"===s?d=u=Math.max(a,l):(d=s.width,u=s.height),d!==l||u!==a){const s=new Float32Array(d*u*c);if(Array.isArray(n))for(let e=0;e<s.length;++e)s[e]=n[e%c];else 0!==n&&s.fill(n);const[_,p]=o?[Math.floor((d-l)/2),Math.floor((u-a)/2)]:[0,0];for(let t=0;t<a;++t){const r=(t+p)*d,o=t*l;for(let t=0;t<l;++t){const n=(r+t+_)*c,i=(o+t)*c;for(let t=0;t<c;++t)s[n+t]=e[i+t]}}if("symmetric"===r){if(o)throw new Error("`center` padding is not supported when `mode` is set to `symmetric`.");const t=a-1,r=l-1;for(let o=0;o<u;++o){const n=o*d,u=(0,i.calculateReflectOffset)(o,t)*l;for(let t=0;t<d;++t){if(o<a&&t<l)continue;const d=(n+t)*c,_=(u+(0,i.calculateReflectOffset)(t,r))*c;for(let t=0;t<c;++t)s[d+t]=e[_+t]}}}e=s,t=[u,d,c]}return[e,t]}rescale(e){for(let t=0;t<e.length;++t)e[t]=this.rescale_factor*e[t]}get_resize_output_image_size(e,t){const[s,r]=e.size;let o,n;if(this.do_thumbnail){const{height:e,width:s}=t;o=Math.min(e,s)}else Number.isInteger(t)?(o=t,n=this.config.max_size??o):void 0!==t&&(o=t.shortest_edge,n=t.longest_edge);if(void 0!==o||void 0!==n){const e=void 0===o?1:Math.max(o/s,o/r),t=s*e,i=r*e,a=void 0===n?1:Math.min(n/t,n/i);let l=Math.floor(Number((t*a).toFixed(2))),c=Math.floor(Number((i*a).toFixed(2)));return void 0!==this.size_divisibility&&([l,c]=d([l,c],this.size_divisibility)),[l,c]}if(void 0!==t&&void 0!==t.width&&void 0!==t.height){let e=t.width,o=t.height;if(this.config.keep_aspect_ratio&&this.config.ensure_multiple_of){let t=o/r,n=e/s;Math.abs(1-n)<Math.abs(1-t)?t=n:n=t,o=c(t*r,this.config.ensure_multiple_of),e=c(n*s,this.config.ensure_multiple_of)}return[e,o]}if(void 0!==this.size_divisibility)return d([s,r],this.size_divisibility);if(void 0!==this.min_pixels&&void 0!==this.max_pixels){return function(e,t,s=28,r=3136,o=1003520){if(e<s||t<s)throw new Error(`height:${e} or width:${t} must be larger than factor:${s}`);if(Math.max(e,t)/Math.min(e,t)>200)throw new Error("absolute aspect ratio must be smaller than 200, got "+Math.max(e,t)/Math.min(e,t));let n=Math.round(e/s)*s,i=Math.round(t/s)*s;if(n*i>o){const r=Math.sqrt(e*t/o);n=Math.floor(e/r/s)*s,i=Math.floor(t/r/s)*s}else if(n*i<r){const o=Math.sqrt(r/(e*t));n=Math.ceil(e*o/s)*s,i=Math.ceil(t*o/s)*s}return[n,i]}(r,s,this.config.patch_size*this.config.merge_size,this.min_pixels,this.max_pixels)}throw new Error(`Could not resize image due to unsupported \`this.size\` option in config: ${JSON.stringify(t)}`)}async resize(e){const[t,s]=this.get_resize_output_image_size(e,this.size);return await e.resize(t,s,{resample:this.resample})}async preprocess(e,{do_normalize:t=null,do_pad:s=null,do_convert_rgb:r=null,do_convert_grayscale:n=null,do_flip_channel_order:i=null}={}){this.do_crop_margin&&(e=await this.crop_margin(e));const[a,l]=e.size;if(r??this.do_convert_rgb?e=e.rgb():n&&(e=e.grayscale()),this.do_resize&&(e=await this.resize(e)),this.do_thumbnail&&(e=await this.thumbnail(e,this.size,this.resample)),this.do_center_crop){let t,s;Number.isInteger(this.crop_size)?(t=this.crop_size,s=this.crop_size):(t=this.crop_size.width,s=this.crop_size.height),e=await e.center_crop(t,s)}const c=[e.height,e.width];let u=Float32Array.from(e.data),_=[e.height,e.width,e.channels];if(this.do_rescale&&this.rescale(u),t??this.do_normalize){let t=this.image_mean;Array.isArray(this.image_mean)||(t=new Array(e.channels).fill(t));let s=this.image_std;if(Array.isArray(this.image_std)||(s=new Array(e.channels).fill(t)),t.length!==e.channels||s.length!==e.channels)throw new Error(`When set to arrays, the length of \`image_mean\` (${t.length}) and \`image_std\` (${s.length}) must match the number of channels in the image (${e.channels}).`);for(let r=0;r<u.length;r+=e.channels)for(let o=0;o<e.channels;++o)u[r+o]=(u[r+o]-t[o])/s[o]}if(s??this.do_pad)if(this.pad_size){const t=this.pad_image(u,[e.height,e.width,e.channels],this.pad_size);[u,_]=t}else if(this.size_divisibility){const[e,t]=d([_[1],_[0]],this.size_divisibility);[u,_]=this.pad_image(u,_,{width:e,height:t})}if(i??this.do_flip_channel_order){if(3!==_[2])throw new Error("Flipping channel order is only supported for RGB images.");for(let e=0;e<u.length;e+=3){const t=u[e];u[e]=u[e+2],u[e+2]=t}}return{original_size:[l,a],reshaped_input_size:c,pixel_values:new o.Tensor("float32",u,_).permute(2,0,1)}}async _call(e,...t){Array.isArray(e)||(e=[e]);const s=await Promise.all(e.map((e=>this.preprocess(e))));return{pixel_values:(0,o.stack)(s.map((e=>e.pixel_values)),0),original_sizes:s.map((e=>e.original_size)),reshaped_input_sizes:s.map((e=>e.reshaped_input_size))}}static async from_pretrained(e,t){return new this(await(0,a.getModelJSON)(e,l.IMAGE_PROCESSOR_NAME,!0,t))}}},"./src/base/processing_utils.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Processor:()=>i});var r=s("./src/utils/constants.js"),o=s("./src/utils/generic.js"),n=s("./src/utils/hub.js");class i extends o.Callable{static classes=["image_processor_class","tokenizer_class","feature_extractor_class"];static uses_processor_config=!1;constructor(e,t){super(),this.config=e,this.components=t}get image_processor(){return this.components.image_processor}get tokenizer(){return this.components.tokenizer}get feature_extractor(){return this.components.feature_extractor}apply_chat_template(e,t={}){if(!this.tokenizer)throw new Error("Unable to apply chat template without a tokenizer.");return this.tokenizer.apply_chat_template(e,{tokenize:!1,...t})}batch_decode(...e){if(!this.tokenizer)throw new Error("Unable to decode without a tokenizer.");return this.tokenizer.batch_decode(...e)}decode(...e){if(!this.tokenizer)throw new Error("Unable to decode without a tokenizer.");return this.tokenizer.decode(...e)}async _call(e,...t){for(const s of[this.image_processor,this.feature_extractor,this.tokenizer])if(s)return s(e,...t);throw new Error("No image processor, feature extractor, or tokenizer found.")}static async from_pretrained(e,t){const[s,o]=await Promise.all([this.uses_processor_config?(0,n.getModelJSON)(e,r.PROCESSOR_NAME,!0,t):{},Promise.all(this.classes.filter((e=>e in this)).map((async s=>{const r=await this[s].from_pretrained(e,t);return[s.replace(/_class$/,""),r]}))).then(Object.fromEntries)]);return new this(s,o)}}},"./src/configs.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{AutoConfig:()=>l,PretrainedConfig:()=>a,getKeyValueShapes:()=>i});var r=s("./src/utils/core.js"),o=s("./src/utils/hub.js");function n(e){const t={};let s={};switch(e.model_type){case"llava":case"paligemma":case"gemma3":case"florence2":case"llava_onevision":case"idefics3":case"ultravox":case"smolvlm":s=n(e.text_config);break;case"moondream1":s=n(e.phi_config);break;case"musicgen":s=n(e.decoder);break;case"multi_modality":s=n(e.language_config);break;case"gpt2":case"gptj":case"jais":case"codegen":case"gpt_bigcode":t.num_heads="n_head",t.num_layers="n_layer",t.hidden_size="n_embd";break;case"gpt_neox":case"stablelm":case"opt":case"falcon":t.num_heads="num_attention_heads",t.num_layers="num_hidden_layers",t.hidden_size="hidden_size";break;case"llama":case"olmo":case"olmo2":case"mobilellm":case"granite":case"cohere":case"mistral":case"starcoder2":case"qwen2":case"qwen2_vl":case"phi":case"phi3":case"phi3_v":t.num_heads="num_key_value_heads",t.num_layers="num_hidden_layers",t.hidden_size="hidden_size",t.num_attention_heads="num_attention_heads";break;case"qwen3":case"gemma":case"gemma2":case"gemma3_text":case"glm":case"helium":t.num_heads="num_key_value_heads",t.num_layers="num_hidden_layers",t.dim_kv="head_dim";break;case"openelm":t.num_heads="num_kv_heads",t.num_layers="num_transformer_layers",t.dim_kv="head_dim";break;case"gpt_neo":case"donut-swin":t.num_heads="num_heads",t.num_layers="num_layers",t.hidden_size="hidden_size";break;case"bloom":t.num_heads="n_head",t.num_layers="n_layer",t.hidden_size="hidden_size";break;case"mpt":t.num_heads="n_heads",t.num_layers="n_layers",t.hidden_size="d_model";break;case"exaone":t.num_heads="num_key_value_heads",t.num_layers="num_layers",t.dim_kv="head_dim",t.num_attention_heads="num_attention_heads";break;case"t5":case"mt5":case"longt5":t.num_decoder_layers="num_decoder_layers",t.num_decoder_heads="num_heads",t.decoder_dim_kv="d_kv",t.num_encoder_layers="num_layers",t.num_encoder_heads="num_heads",t.encoder_dim_kv="d_kv";break;case"bart":case"mbart":case"marian":case"whisper":case"lite-whisper":case"m2m_100":case"blenderbot":case"blenderbot-small":case"florence2_language":t.num_decoder_layers="decoder_layers",t.num_decoder_heads="decoder_attention_heads",t.decoder_hidden_size="d_model",t.num_encoder_layers="encoder_layers",t.num_encoder_heads="encoder_attention_heads",t.encoder_hidden_size="d_model";break;case"speecht5":t.num_decoder_layers="decoder_layers",t.num_decoder_heads="decoder_attention_heads",t.decoder_hidden_size="hidden_size",t.num_encoder_layers="encoder_layers",t.num_encoder_heads="encoder_attention_heads",t.encoder_hidden_size="hidden_size";break;case"trocr":t.num_encoder_layers=t.num_decoder_layers="decoder_layers",t.num_encoder_heads=t.num_decoder_heads="decoder_attention_heads",t.encoder_hidden_size=t.decoder_hidden_size="d_model";break;case"musicgen_decoder":t.num_encoder_layers=t.num_decoder_layers="num_hidden_layers",t.num_encoder_heads=t.num_decoder_heads="num_attention_heads",t.encoder_hidden_size=t.decoder_hidden_size="hidden_size";break;case"moonshine":t.num_decoder_layers="decoder_num_hidden_layers",t.num_decoder_heads="decoder_num_key_value_heads",t.num_encoder_layers="encoder_num_hidden_layers",t.num_encoder_heads="encoder_num_key_value_heads",t.encoder_hidden_size=t.decoder_hidden_size="hidden_size";break;case"vision-encoder-decoder":const o=n(e.decoder),i="num_decoder_layers"in o,a=(0,r.pick)(e,["model_type","is_encoder_decoder"]);return i?(a.num_decoder_layers=o.num_decoder_layers,a.num_decoder_heads=o.num_decoder_heads,a.decoder_hidden_size=o.decoder_hidden_size,a.num_encoder_layers=o.num_encoder_layers,a.num_encoder_heads=o.num_encoder_heads,a.encoder_hidden_size=o.encoder_hidden_size):(a.num_layers=o.num_layers,a.num_heads=o.num_heads,a.hidden_size=o.hidden_size),a}const o={...s,...(0,r.pick)(e,["model_type","multi_query","is_encoder_decoder"])};for(const s in t)o[s]=e[t[s]];return o}function i(e,{prefix:t="past_key_values",batch_size:s=1}={}){const r={},o=e.normalized_config;if(o.is_encoder_decoder&&"num_encoder_heads"in o&&"num_decoder_heads"in o){const e=o.encoder_dim_kv??o.encoder_hidden_size/o.num_encoder_heads,n=o.decoder_dim_kv??o.decoder_hidden_size/o.num_decoder_heads,i=[s,o.num_encoder_heads,0,e],a=[s,o.num_decoder_heads,0,n];for(let e=0;e<o.num_decoder_layers;++e)r[`${t}.${e}.encoder.key`]=i,r[`${t}.${e}.encoder.value`]=i,r[`${t}.${e}.decoder.key`]=a,r[`${t}.${e}.decoder.value`]=a}else{const e=o.num_heads,n=o.num_layers,i=o.dim_kv??o.hidden_size/(o.num_attention_heads??e);if("falcon"===o.model_type){const o=[s*e,0,i];for(let e=0;e<n;++e)r[`${t}.${e}.key`]=o,r[`${t}.${e}.value`]=o}else if(o.multi_query){const o=[s*e,0,2*i];for(let e=0;e<n;++e)r[`${t}.${e}.key_value`]=o}else if("bloom"===o.model_type){const o=[s*e,i,0],a=[s*e,0,i];for(let e=0;e<n;++e)r[`${t}.${e}.key`]=o,r[`${t}.${e}.value`]=a}else if("openelm"===o.model_type)for(let o=0;o<n;++o){const n=[s,e[o],0,i];r[`${t}.${o}.key`]=n,r[`${t}.${o}.value`]=n}else{const o=[s,e,0,i];for(let e=0;e<n;++e)r[`${t}.${e}.key`]=o,r[`${t}.${e}.value`]=o}}return r}class a{model_type=null;is_encoder_decoder=!1;max_position_embeddings;"transformers.js_config";constructor(e){Object.assign(this,e),this.normalized_config=n(this)}static async from_pretrained(e,{progress_callback:t=null,config:s=null,cache_dir:r=null,local_files_only:n=!1,revision:i="main"}={}){!s||s instanceof a||(s=new a(s));const l=s??await async function(e,t){return await(0,o.getModelJSON)(e,"config.json",!0,t)}(e,{progress_callback:t,config:s,cache_dir:r,local_files_only:n,revision:i});return new this(l)}}class l{static async from_pretrained(...e){return a.from_pretrained(...e)}}},"./src/env.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{apis:()=>h,env:()=>x});var r=s("fs"),o=s("path"),n=s("url");const i="undefined"!=typeof window&&void 0!==window.document,a="undefined"!=typeof self&&"DedicatedWorkerGlobalScope"===self.constructor?.name,l="undefined"!=typeof self&&"caches"in self,c="undefined"!=typeof navigator&&"gpu"in navigator,d="undefined"!=typeof navigator&&"ml"in navigator,u="undefined"!=typeof process,_=u&&"node"===process?.release?.name,p=!b(r),m=!b(o),h=Object.freeze({IS_BROWSER_ENV:i,IS_WEBWORKER_ENV:a,IS_WEB_CACHE_AVAILABLE:l,IS_WEBGPU_AVAILABLE:c,IS_WEBNN_AVAILABLE:d,IS_PROCESS_AVAILABLE:u,IS_NODE_ENV:_,IS_FS_AVAILABLE:p,IS_PATH_AVAILABLE:m}),g=p&&m;let f="./";if(g){const e=Object({}).url;e?f=o.dirname(o.dirname(n.fileURLToPath(e))):"undefined"!=typeof __dirname&&(f=o.dirname(__dirname))}const w=g?o.join(f,"/.cache/"):null,M="/models/",x={version:"3.5.2",backends:{onnx:{}},allowRemoteModels:!0,remoteHost:"https://huggingface.co/",remotePathTemplate:"{model}/resolve/{revision}/",allowLocalModels:!(i||a),localModelPath:g?o.join(f,M):M,useFS:p,useBrowserCache:l,useFSCache:p,cacheDir:w,useCustomCache:!1,customCache:null};function b(e){return 0===Object.keys(e).length}},"./src/generation/configuration_utils.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{GenerationConfig:()=>o});var r=s("./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;streamer=null;begin_suppress_tokens=null;forced_decoder_ids=null;guidance_scale=null;num_return_sequences=1;output_attentions=!1;output_hidden_states=!1;output_scores=!1;return_dict_in_generate=!1;pad_token_id=null;bos_token_id=null;eos_token_id=null;encoder_no_repeat_ngram_size=0;decoder_start_token_id=null;generation_kwargs={};constructor(e){Object.assign(this,(0,r.pick)(e,Object.getOwnPropertyNames(this)))}}},"./src/generation/logits_process.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ClassifierFreeGuidanceLogitsProcessor:()=>f,ForcedBOSTokenLogitsProcessor:()=>l,ForcedEOSTokenLogitsProcessor:()=>c,LogitsProcessor:()=>n,LogitsProcessorList:()=>a,LogitsWarper:()=>i,MinLengthLogitsProcessor:()=>m,MinNewTokensLengthLogitsProcessor:()=>h,NoBadWordsLogitsProcessor:()=>g,NoRepeatNGramLogitsProcessor:()=>_,RepetitionPenaltyLogitsProcessor:()=>p,SuppressTokensAtBeginLogitsProcessor:()=>d,TemperatureLogitsWarper:()=>w,TopKLogitsWarper:()=>x,TopPLogitsWarper:()=>M,WhisperTimeStampLogitsProcessor:()=>u});var r=s("./src/utils/generic.js"),o=(s("./src/utils/tensor.js"),s("./src/utils/maths.js"));class n extends r.Callable{_call(e,t){throw Error("`_call` should be implemented in a subclass")}}class i extends r.Callable{_call(e,t){throw Error("`_call` should be implemented in a subclass")}}class a extends r.Callable{constructor(){super(),this.processors=[]}push(e){this.processors.push(e)}extend(e){this.processors.push(...e)}_call(e,t){let s=t;for(const t of this.processors)s=t(e,s);return s}[Symbol.iterator](){return this.processors.values()}}class l extends n{constructor(e){super(),this.bos_token_id=e}_call(e,t){for(let s=0;s<e.length;++s)if(1===e[s].length){const e=t[s].data;e.fill(-1/0),e[this.bos_token_id]=0}return t}}class c extends n{constructor(e,t){super(),this.max_length=e,this.eos_token_id=Array.isArray(t)?t:[t]}_call(e,t){for(let s=0;s<e.length;++s)if(e[s].length===this.max_length-1){const e=t[s].data;e.fill(-1/0);for(const t of this.eos_token_id)e[t]=0}return t}}class d extends n{constructor(e,t){super(),this.begin_suppress_tokens=e,this.begin_index=t}_call(e,t){for(let s=0;s<e.length;++s)if(e[s].length===this.begin_index){const e=t[s].data;for(const t of this.begin_suppress_tokens)e[t]=-1/0}return t}}class u extends n{constructor(e,t){super(),this.eos_token_id=Array.isArray(e.eos_token_id)?e.eos_token_id[0]:e.eos_token_id,this.no_timestamps_token_id=e.no_timestamps_token_id,this.timestamp_begin=this.no_timestamps_token_id+1,this.begin_index=t.length,t.at(-1)===this.no_timestamps_token_id&&(this.begin_index-=1),this.max_initial_timestamp_index=e.max_initial_timestamp_index}_call(e,t){for(let s=0;s<e.length;++s){const r=t[s].data;if(r[this.no_timestamps_token_id]=-1/0,e[s].length===this.begin_index-1){r.fill(-1/0),r[this.timestamp_begin]=0;continue}const n=e[s].slice(this.begin_index),i=n.length>=1&&n[n.length-1]>=this.timestamp_begin,a=n.length<2||n[n.length-2]>=this.timestamp_begin;if(i&&(a?r.subarray(this.timestamp_begin).fill(-1/0):r.subarray(0,this.eos_token_id).fill(-1/0)),e[s].length===this.begin_index&&null!==this.max_initial_timestamp_index){const e=this.timestamp_begin+this.max_initial_timestamp_index;r.subarray(e+1).fill(-1/0)}const l=(0,o.log_softmax)(r);Math.log(l.subarray(this.timestamp_begin).map(Math.exp).reduce(((e,t)=>e+t)))>(0,o.max)(l.subarray(0,this.timestamp_begin))[0]&&r.subarray(0,this.timestamp_begin).fill(-1/0)}return t}}class _ extends n{constructor(e){super(),this.no_repeat_ngram_size=e}getNgrams(e){const t=e.length,s=[];for(let r=0;r<t+1-this.no_repeat_ngram_size;++r){const t=[];for(let s=0;s<this.no_repeat_ngram_size;++s)t.push(e[r+s]);s.push(t.map(Number))}const r=new Map;for(const e of s){const t=e.slice(0,e.length-1),s=JSON.stringify(t),o=r.get(s)??[];o.push(e[e.length-1]),r.set(s,o)}return r}getGeneratedNgrams(e,t){const s=t.slice(t.length+1-this.no_repeat_ngram_size,t.length);return e.get(JSON.stringify(s.map(Number)))??[]}calcBannedNgramTokens(e){const t=[];if(e.length+1<this.no_repeat_ngram_size)return t;{const t=this.getNgrams(e);return this.getGeneratedNgrams(t,e)}}_call(e,t){for(let s=0;s<e.length;++s){const r=t[s].data,o=this.calcBannedNgramTokens(e[s]);for(const e of o)r[e]=-1/0}return t}}class p extends n{constructor(e){super(),this.penalty=e}_call(e,t){for(let s=0;s<e.length;++s){const r=t[s].data;for(const t of new Set(e[s])){const e=Number(t);r[e]<0?r[e]*=this.penalty:r[e]/=this.penalty}}return t}}class m extends n{constructor(e,t){super(),this.min_length=e,this.eos_token_id=Array.isArray(t)?t:[t]}_call(e,t){for(let s=0;s<e.length;++s)if(e[s].length<this.min_length){const e=t[s].data;for(const t of this.eos_token_id)e[t]=-1/0}return t}}class h extends n{constructor(e,t,s){super(),this.prompt_length_to_skip=e,this.min_new_tokens=t,this.eos_token_id=Array.isArray(s)?s:[s]}_call(e,t){for(let s=0;s<e.length;++s){if(e[s].length-this.prompt_length_to_skip<this.min_new_tokens){const e=t[s].data;for(const t of this.eos_token_id)e[t]=-1/0}}return t}}class g extends n{constructor(e,t){super(),this.bad_words_ids=e,this.eos_token_id=Array.isArray(t)?t:[t]}_call(e,t){for(let s=0;s<e.length;++s){const r=t[s].data,o=e[s];for(const e of this.bad_words_ids){if(o.length<e.length-1)continue;let t=!0;for(let s=1;s<=e.length-1;++s)if(e.at(-s-1)!=o.at(-s)){t=!1;break}t&&(r[e.at(-1)]=-1/0)}}return t}}class f extends n{constructor(e){if(super(),e<=1)throw new Error(`Require guidance scale >1 to use the classifier free guidance processor, got guidance scale ${e}.`);this.guidance_scale=e}_call(e,t){if(t.dims[0]!==2*e.length)throw new Error(`Logits should have twice the batch size of the input ids, the first half of batches corresponding to the conditional inputs, and the second half of batches corresponding to the unconditional inputs. Got batch size ${t.dims[0]} for the logits and ${e.length} for the input ids.`);const s=e.length,r=t.slice([0,s],null),o=t.slice([s,t.dims[0]],null);for(let e=0;e<o.data.length;++e)o.data[e]+=(r.data[e]-o.data[e])*this.guidance_scale;return o}}class w extends i{constructor(e){if(super(),"number"!=typeof e||e<=0){let t=`\`temperature\` (=${e}) must be a strictly positive float, otherwise your next token scores will be invalid.`;0===e&&(t+=" If you're looking for greedy decoding strategies, set `do_sample=false`.")}this.temperature=e}_call(e,t){const s=t.data;for(let e=0;e<s.length;++e)s[e]/=this.temperature;return t}}class M extends i{constructor(e,{filter_value:t=-1/0,min_tokens_to_keep:s=1}={}){if(super(),e<0||e>1)throw new Error(`\`top_p\` must be a float > 0 and < 1, but is ${e}`);if(!Number.isInteger(s)||s<1)throw new Error(`\`min_tokens_to_keep\` must be a positive integer, but is ${s}`);this.top_p=e,this.filter_value=t,this.min_tokens_to_keep=s}}class x extends i{constructor(e,{filter_value:t=-1/0,min_tokens_to_keep:s=1}={}){if(super(),!Number.isInteger(e)||e<0)throw new Error(`\`top_k\` must be a positive integer, but is ${e}`);this.top_k=Math.max(e,s),this.filter_value=t}}},"./src/generation/logits_sampler.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{LogitsSampler:()=>i});var r=s("./src/utils/generic.js"),o=s("./src/utils/tensor.js"),n=s("./src/utils/maths.js");s("./src/generation/configuration_utils.js");class i extends r.Callable{constructor(e){super(),this.generation_config=e}async _call(e){return this.sample(e)}async sample(e){throw Error("sample should be implemented in subclasses.")}getLogits(e,t){let s=e.dims.at(-1),r=e.data;if(-1===t)r=r.slice(-s);else{let e=t*s;r=r.slice(e,e+s)}return r}randomSelect(e){let t=0;for(let s=0;s<e.length;++s)t+=e[s];let s=Math.random()*t;for(let t=0;t<e.length;++t)if(s-=e[t],s<=0)return t;return 0}static getSampler(e){if(e.do_sample)return new l(e);if(e.num_beams>1)return new c(e);if(e.num_return_sequences>1)throw Error(`num_return_sequences has to be 1 when doing greedy search, but is ${e.num_return_sequences}.`);return new a(e)}}class a extends i{async sample(e){const t=(0,n.max)(e.data)[1];return[[BigInt(t),0]]}}class l extends i{async sample(e){let t=e.dims.at(-1);this.generation_config.top_k>0&&(t=Math.min(this.generation_config.top_k,t));const[s,r]=await(0,o.topk)(e,t),i=(0,n.softmax)(s.data);return Array.from({length:this.generation_config.num_beams},(()=>{const e=this.randomSelect(i);return[r.data[e],Math.log(i[e])]}))}}class c extends i{async sample(e){let t=e.dims.at(-1);this.generation_config.top_k>0&&(t=Math.min(this.generation_config.top_k,t));const[s,r]=await(0,o.topk)(e,t),i=(0,n.softmax)(s.data);return Array.from({length:this.generation_config.num_beams},((e,t)=>[r.data[t],Math.log(i[t])]))}}},"./src/generation/stopping_criteria.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{EosTokenCriteria:()=>a,InterruptableStoppingCriteria:()=>l,MaxLengthCriteria:()=>i,StoppingCriteria:()=>o,StoppingCriteriaList:()=>n});var r=s("./src/utils/generic.js");class o extends r.Callable{_call(e,t){throw Error("StoppingCriteria needs to be subclassed")}}class n extends r.Callable{constructor(){super(),this.criteria=[]}push(e){this.criteria.push(e)}extend(e){e instanceof n?e=e.criteria:e instanceof o&&(e=[e]),this.criteria.push(...e)}_call(e,t){const s=new Array(e.length).fill(!1);for(const r of this.criteria){const o=r(e,t);for(let e=0;e<s.length;++e)s[e]||=o[e]}return s}[Symbol.iterator](){return this.criteria.values()}}class i extends o{constructor(e,t=null){super(),this.max_length=e,this.max_position_embeddings=t}_call(e){return e.map((e=>e.length>=this.max_length))}}class a extends o{constructor(e){super(),Array.isArray(e)||(e=[e]),this.eos_token_id=e}_call(e,t){return e.map((e=>{const t=e.at(-1);return this.eos_token_id.some((e=>t==e))}))}}class l extends o{constructor(){super(),this.interrupted=!1}interrupt(){this.interrupted=!0}reset(){this.interrupted=!1}_call(e,t){return new Array(e.length).fill(this.interrupted)}}},"./src/generation/streamers.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{BaseStreamer:()=>i,TextStreamer:()=>l,WhisperTextStreamer:()=>c});var r=s("./src/utils/core.js"),o=s("./src/tokenizers.js"),n=s("./src/env.js");class i{put(e){throw Error("Not implemented")}end(){throw Error("Not implemented")}}const a=n.apis.IS_PROCESS_AVAILABLE?e=>process.stdout.write(e):e=>console.log(e);class l extends i{constructor(e,{skip_prompt:t=!1,callback_function:s=null,token_callback_function:r=null,skip_special_tokens:o=!0,decode_kwargs:n={},...i}={}){super(),this.tokenizer=e,this.skip_prompt=t,this.callback_function=s??a,this.token_callback_function=r,this.decode_kwargs={skip_special_tokens:o,...n,...i},this.token_cache=[],this.print_len=0,this.next_tokens_are_prompt=!0}put(e){if(e.length>1)throw Error("TextStreamer only supports batch size of 1");const t=this.next_tokens_are_prompt;if(t&&(this.next_tokens_are_prompt=!1,this.skip_prompt))return;const s=e[0];this.token_callback_function?.(s),this.token_cache=(0,r.mergeArrays)(this.token_cache,s);const n=this.tokenizer.decode(this.token_cache,this.decode_kwargs);let i;t||n.endsWith("\n")?(i=n.slice(this.print_len),this.token_cache=[],this.print_len=0):n.length>0&&(0,o.is_chinese_char)(n.charCodeAt(n.length-1))?(i=n.slice(this.print_len),this.print_len+=i.length):(i=n.slice(this.print_len,n.lastIndexOf(" ")+1),this.print_len+=i.length),this.on_finalized_text(i,!1)}end(){let e;if(this.token_cache.length>0){e=this.tokenizer.decode(this.token_cache,this.decode_kwargs).slice(this.print_len),this.token_cache=[],this.print_len=0}else e="";this.next_tokens_are_prompt=!0,this.on_finalized_text(e,!0)}on_finalized_text(e,t){e.length>0&&this.callback_function?.(e),t&&this.callback_function===a&&n.apis.IS_PROCESS_AVAILABLE&&this.callback_function?.("\n")}}class c extends l{constructor(e,{skip_prompt:t=!1,callback_function:s=null,token_callback_function:r=null,on_chunk_start:o=null,on_chunk_end:n=null,on_finalize:i=null,time_precision:a=.02,skip_special_tokens:l=!0,decode_kwargs:c={}}={}){super(e,{skip_prompt:t,skip_special_tokens:l,callback_function:s,token_callback_function:r,decode_kwargs:c}),this.timestamp_begin=e.timestamp_begin,this.on_chunk_start=o,this.on_chunk_end=n,this.on_finalize=i,this.time_precision=a,this.waiting_for_timestamp=!1}put(e){if(e.length>1)throw Error("WhisperTextStreamer only supports batch size of 1");const t=e[0];if(1===t.length){const e=Number(t[0])-this.timestamp_begin;if(e>=0){const s=e*this.time_precision;return this.waiting_for_timestamp?this.on_chunk_end?.(s):this.on_chunk_start?.(s),this.waiting_for_timestamp=!this.waiting_for_timestamp,void this.token_callback_function?.(t)}}return super.put(e)}end(){super.end(),this.on_finalize?.()}}},"./src/models.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ASTForAudioClassification:()=>ks,ASTModel:()=>bs,ASTPreTrainedModel:()=>xs,AlbertForMaskedLM:()=>Et,AlbertForQuestionAnswering:()=>At,AlbertForSequenceClassification:()=>St,AlbertModel:()=>Ct,AlbertPreTrainedModel:()=>Ft,AutoModel:()=>rd,AutoModelForAudioClassification:()=>kd,AutoModelForAudioFrameClassification:()=>vd,AutoModelForAudioTextToText:()=>Id,AutoModelForCTC:()=>bd,AutoModelForCausalLM:()=>dd,AutoModelForDepthEstimation:()=>Cd,AutoModelForDocumentQuestionAnswering:()=>Td,AutoModelForImageClassification:()=>md,AutoModelForImageFeatureExtraction:()=>Ed,AutoModelForImageMatting:()=>Pd,AutoModelForImageSegmentation:()=>hd,AutoModelForImageTextToText:()=>Ld,AutoModelForImageToImage:()=>Fd,AutoModelForMaskGeneration:()=>xd,AutoModelForMaskedLM:()=>ud,AutoModelForNormalEstimation:()=>Sd,AutoModelForObjectDetection:()=>wd,AutoModelForPoseEstimation:()=>Ad,AutoModelForQuestionAnswering:()=>_d,AutoModelForSemanticSegmentation:()=>gd,AutoModelForSeq2SeqLM:()=>id,AutoModelForSequenceClassification:()=>od,AutoModelForSpeechSeq2Seq:()=>ad,AutoModelForTextToSpectrogram:()=>ld,AutoModelForTextToWaveform:()=>cd,AutoModelForTokenClassification:()=>nd,AutoModelForUniversalSegmentation:()=>fd,AutoModelForVision2Seq:()=>pd,AutoModelForXVector:()=>yd,AutoModelForZeroShotObjectDetection:()=>Md,BartForConditionalGeneration:()=>$t,BartForSequenceClassification:()=>qt,BartModel:()=>Gt,BartPretrainedModel:()=>Rt,BaseModelOutput:()=>oe,BeitForImageClassification:()=>yn,BeitModel:()=>kn,BeitPreTrainedModel:()=>bn,BertForMaskedLM:()=>ae,BertForQuestionAnswering:()=>de,BertForSequenceClassification:()=>le,BertForTokenClassification:()=>ce,BertModel:()=>ie,BertPreTrainedModel:()=>ne,BlenderbotForConditionalGeneration:()=>Kt,BlenderbotModel:()=>Yt,BlenderbotPreTrainedModel:()=>Jt,BlenderbotSmallForConditionalGeneration:()=>ts,BlenderbotSmallModel:()=>es,BlenderbotSmallPreTrainedModel:()=>Zt,BloomForCausalLM:()=>Lo,BloomModel:()=>Eo,BloomPreTrainedModel:()=>Ao,CLIPModel:()=>Ws,CLIPPreTrainedModel:()=>qs,CLIPSegForImageSegmentation:()=>lr,CLIPSegModel:()=>ar,CLIPSegPreTrainedModel:()=>ir,CLIPTextModel:()=>Us,CLIPTextModelWithProjection:()=>Xs,CLIPVisionModel:()=>Qs,CLIPVisionModelWithProjection:()=>Hs,CamembertForMaskedLM:()=>Ve,CamembertForQuestionAnswering:()=>Re,CamembertForSequenceClassification:()=>Ne,CamembertForTokenClassification:()=>Be,CamembertModel:()=>De,CamembertPreTrainedModel:()=>Oe,CausalLMOutput:()=>Bd,CausalLMOutputWithPast:()=>Rd,ChineseCLIPModel:()=>tr,ChineseCLIPPreTrainedModel:()=>er,ClapAudioModelWithProjection:()=>al,ClapModel:()=>nl,ClapPreTrainedModel:()=>ol,ClapTextModelWithProjection:()=>il,CodeGenForCausalLM:()=>Sr,CodeGenModel:()=>Cr,CodeGenPreTrainedModel:()=>Fr,CohereForCausalLM:()=>so,CohereModel:()=>to,CoherePreTrainedModel:()=>eo,ConvBertForMaskedLM:()=>Pe,ConvBertForQuestionAnswering:()=>Se,ConvBertForSequenceClassification:()=>Fe,ConvBertForTokenClassification:()=>Ce,ConvBertModel:()=>Te,ConvBertPreTrainedModel:()=>ve,ConvNextForImageClassification:()=>Oi,ConvNextModel:()=>ji,ConvNextPreTrainedModel:()=>zi,ConvNextV2ForImageClassification:()=>Ni,ConvNextV2Model:()=>Vi,ConvNextV2PreTrainedModel:()=>Di,DFineForObjectDetection:()=>qn,DFineModel:()=>$n,DFinePreTrainedModel:()=>Gn,DPTForDepthEstimation:()=>pi,DPTModel:()=>_i,DPTPreTrainedModel:()=>ui,DacDecoderModel:()=>_c,DacDecoderOutput:()=>cc,DacEncoderModel:()=>uc,DacEncoderOutput:()=>lc,DacModel:()=>dc,DacPreTrainedModel:()=>ac,DebertaForMaskedLM:()=>qe,DebertaForQuestionAnswering:()=>Xe,DebertaForSequenceClassification:()=>We,DebertaForTokenClassification:()=>Ue,DebertaModel:()=>$e,DebertaPreTrainedModel:()=>Ge,DebertaV2ForMaskedLM:()=>Je,DebertaV2ForQuestionAnswering:()=>Ze,DebertaV2ForSequenceClassification:()=>Ye,DebertaV2ForTokenClassification:()=>Ke,DebertaV2Model:()=>He,DebertaV2PreTrainedModel:()=>Qe,DecisionTransformerModel:()=>Rl,DecisionTransformerPreTrainedModel:()=>Bl,DeiTForImageClassification:()=>Yn,DeiTModel:()=>Jn,DeiTPreTrainedModel:()=>Hn,DepthAnythingForDepthEstimation:()=>hi,DepthAnythingPreTrainedModel:()=>mi,DepthProForDepthEstimation:()=>bi,DepthProPreTrainedModel:()=>xi,DetrForObjectDetection:()=>Pn,DetrForSegmentation:()=>Fn,DetrModel:()=>Tn,DetrObjectDetectionOutput:()=>Cn,DetrPreTrainedModel:()=>vn,DetrSegmentationOutput:()=>Sn,Dinov2ForImageClassification:()=>Gi,Dinov2Model:()=>Ri,Dinov2PreTrainedModel:()=>Bi,Dinov2WithRegistersForImageClassification:()=>Wi,Dinov2WithRegistersModel:()=>qi,Dinov2WithRegistersPreTrainedModel:()=>$i,DistilBertForMaskedLM:()=>nt,DistilBertForQuestionAnswering:()=>ot,DistilBertForSequenceClassification:()=>st,DistilBertForTokenClassification:()=>rt,DistilBertModel:()=>tt,DistilBertPreTrainedModel:()=>et,DonutSwinModel:()=>Ii,DonutSwinPreTrainedModel:()=>Li,EfficientNetForImageClassification:()=>Ml,EfficientNetModel:()=>wl,EfficientNetPreTrainedModel:()=>fl,ElectraForMaskedLM:()=>Le,ElectraForQuestionAnswering:()=>je,ElectraForSequenceClassification:()=>Ie,ElectraForTokenClassification:()=>ze,ElectraModel:()=>Ee,ElectraPreTrainedModel:()=>Ae,EsmForMaskedLM:()=>lt,EsmForSequenceClassification:()=>ct,EsmForTokenClassification:()=>dt,EsmModel:()=>at,EsmPreTrainedModel:()=>it,ExaoneForCausalLM:()=>Rr,ExaoneModel:()=>Br,ExaonePreTrainedModel:()=>Nr,FalconForCausalLM:()=>rl,FalconModel:()=>sl,FalconPreTrainedModel:()=>tl,FastViTForImageClassification:()=>nn,FastViTModel:()=>on,FastViTPreTrainedModel:()=>rn,Florence2ForConditionalGeneration:()=>Os,Florence2PreTrainedModel:()=>js,GLPNForDepthEstimation:()=>Ei,GLPNModel:()=>Ai,GLPNPreTrainedModel:()=>Si,GPT2LMHeadModel:()=>ur,GPT2Model:()=>dr,GPT2PreTrainedModel:()=>cr,GPTBigCodeForCausalLM:()=>Pr,GPTBigCodeModel:()=>Tr,GPTBigCodePreTrainedModel:()=>vr,GPTJForCausalLM:()=>yr,GPTJModel:()=>kr,GPTJPreTrainedModel:()=>br,GPTNeoForCausalLM:()=>fr,GPTNeoModel:()=>gr,GPTNeoPreTrainedModel:()=>hr,GPTNeoXForCausalLM:()=>xr,GPTNeoXModel:()=>Mr,GPTNeoXPreTrainedModel:()=>wr,Gemma2ForCausalLM:()=>lo,Gemma2Model:()=>ao,Gemma2PreTrainedModel:()=>io,Gemma3ForCausalLM:()=>_o,Gemma3Model:()=>uo,Gemma3PreTrainedModel:()=>co,GemmaForCausalLM:()=>no,GemmaModel:()=>oo,GemmaPreTrainedModel:()=>ro,GlmForCausalLM:()=>Vr,GlmModel:()=>Dr,GlmPreTrainedModel:()=>Or,GraniteForCausalLM:()=>Zr,GraniteModel:()=>Kr,GranitePreTrainedModel:()=>Yr,GroundingDinoForObjectDetection:()=>Xi,GroundingDinoPreTrainedModel:()=>Ui,GroupViTModel:()=>sn,GroupViTPreTrainedModel:()=>tn,HeliumForCausalLM:()=>jr,HeliumModel:()=>zr,HeliumPreTrainedModel:()=>Ir,HieraForImageClassification:()=>ei,HieraModel:()=>Zn,HieraPreTrainedModel:()=>Kn,HubertForCTC:()=>La,HubertForSequenceClassification:()=>Ia,HubertModel:()=>Ea,HubertPreTrainedModel:()=>Aa,IJepaForImageClassification:()=>qo,IJepaModel:()=>$o,IJepaPreTrainedModel:()=>Go,Idefics3ForConditionalGeneration:()=>Bs,Idefics3PreTrainedModel:()=>Ns,ImageMattingOutput:()=>Gd,JAISLMHeadModel:()=>mr,JAISModel:()=>pr,JAISPreTrainedModel:()=>_r,JinaCLIPModel:()=>rr,JinaCLIPPreTrainedModel:()=>sr,JinaCLIPTextModel:()=>or,JinaCLIPVisionModel:()=>nr,LiteWhisperForConditionalGeneration:()=>Ps,LlamaForCausalLM:()=>Lr,LlamaModel:()=>Er,LlamaPreTrainedModel:()=>Ar,LlavaForConditionalGeneration:()=>Ls,LlavaOnevisionForConditionalGeneration:()=>Is,LlavaPreTrainedModel:()=>Es,LongT5ForConditionalGeneration:()=>Dt,LongT5Model:()=>Ot,LongT5PreTrainedModel:()=>jt,M2M100ForConditionalGeneration:()=>ia,M2M100Model:()=>na,M2M100PreTrainedModel:()=>oa,MBartForCausalLM:()=>Ht,MBartForConditionalGeneration:()=>Xt,MBartForSequenceClassification:()=>Qt,MBartModel:()=>Ut,MBartPreTrainedModel:()=>Wt,MPNetForMaskedLM:()=>wt,MPNetForQuestionAnswering:()=>bt,MPNetForSequenceClassification:()=>Mt,MPNetForTokenClassification:()=>xt,MPNetModel:()=>ft,MPNetPreTrainedModel:()=>gt,MT5ForConditionalGeneration:()=>Bt,MT5Model:()=>Nt,MT5PreTrainedModel:()=>Vt,MarianMTModel:()=>ra,MarianModel:()=>sa,MarianPreTrainedModel:()=>ta,MaskFormerForInstanceSegmentation:()=>Ci,MaskFormerModel:()=>Fi,MaskFormerPreTrainedModel:()=>Pi,MaskedLMOutput:()=>Vd,Metric3DForDepthEstimation:()=>yi,Metric3DPreTrainedModel:()=>ki,Metric3Dv2ForDepthEstimation:()=>Ti,Metric3Dv2PreTrainedModel:()=>vi,MgpstrForSceneTextRecognition:()=>Ul,MgpstrModelOutput:()=>ql,MgpstrPreTrainedModel:()=>Wl,MimiDecoderModel:()=>ic,MimiDecoderOutput:()=>rc,MimiEncoderModel:()=>nc,MimiEncoderOutput:()=>sc,MimiModel:()=>oc,MimiPreTrainedModel:()=>tc,MistralForCausalLM:()=>Ya,MistralModel:()=>Ja,MistralPreTrainedModel:()=>Ha,MobileBertForMaskedLM:()=>pt,MobileBertForQuestionAnswering:()=>ht,MobileBertForSequenceClassification:()=>mt,MobileBertModel:()=>_t,MobileBertPreTrainedModel:()=>ut,MobileLLMForCausalLM:()=>qr,MobileLLMModel:()=>$r,MobileLLMPreTrainedModel:()=>Gr,MobileNetV1ForImageClassification:()=>Pl,MobileNetV1ForSemanticSegmentation:()=>Fl,MobileNetV1Model:()=>Tl,MobileNetV1PreTrainedModel:()=>vl,MobileNetV2ForImageClassification:()=>Al,MobileNetV2ForSemanticSegmentation:()=>El,MobileNetV2Model:()=>Sl,MobileNetV2PreTrainedModel:()=>Cl,MobileNetV3ForImageClassification:()=>zl,MobileNetV3ForSemanticSegmentation:()=>jl,MobileNetV3Model:()=>Il,MobileNetV3PreTrainedModel:()=>Ll,MobileNetV4ForImageClassification:()=>Vl,MobileNetV4ForSemanticSegmentation:()=>Nl,MobileNetV4Model:()=>Dl,MobileNetV4PreTrainedModel:()=>Ol,MobileViTForImageClassification:()=>un,MobileViTModel:()=>dn,MobileViTPreTrainedModel:()=>cn,MobileViTV2ForImageClassification:()=>mn,MobileViTV2Model:()=>pn,MobileViTV2PreTrainedModel:()=>_n,ModelOutput:()=>re,ModernBertForMaskedLM:()=>pe,ModernBertForSequenceClassification:()=>me,ModernBertForTokenClassification:()=>he,ModernBertModel:()=>_e,ModernBertPreTrainedModel:()=>ue,Moondream1ForConditionalGeneration:()=>zs,MoonshineForConditionalGeneration:()=>Ss,MoonshineModel:()=>Cs,MoonshinePreTrainedModel:()=>Fs,MptForCausalLM:()=>jo,MptModel:()=>zo,MptPreTrainedModel:()=>Io,MultiModalityCausalLM:()=>$l,MultiModalityPreTrainedModel:()=>Gl,MusicgenForCausalLM:()=>kl,MusicgenForConditionalGeneration:()=>yl,MusicgenModel:()=>bl,MusicgenPreTrainedModel:()=>xl,NomicBertModel:()=>fe,NomicBertPreTrainedModel:()=>ge,OPTForCausalLM:()=>Vo,OPTModel:()=>Do,OPTPreTrainedModel:()=>Oo,Olmo2ForCausalLM:()=>Jr,Olmo2Model:()=>Hr,Olmo2PreTrainedModel:()=>Qr,OlmoForCausalLM:()=>Xr,OlmoModel:()=>Ur,OlmoPreTrainedModel:()=>Wr,OpenELMForCausalLM:()=>ho,OpenELMModel:()=>mo,OpenELMPreTrainedModel:()=>po,OwlViTForObjectDetection:()=>fn,OwlViTModel:()=>gn,OwlViTPreTrainedModel:()=>hn,Owlv2ForObjectDetection:()=>xn,Owlv2Model:()=>Mn,Owlv2PreTrainedModel:()=>wn,PaliGemmaForConditionalGeneration:()=>Vs,PaliGemmaPreTrainedModel:()=>Ds,PatchTSMixerForPrediction:()=>Kl,PatchTSMixerModel:()=>Yl,PatchTSMixerPreTrainedModel:()=>Jl,PatchTSTForPrediction:()=>Hl,PatchTSTModel:()=>Ql,PatchTSTPreTrainedModel:()=>Xl,Phi3ForCausalLM:()=>So,Phi3Model:()=>Co,Phi3PreTrainedModel:()=>Fo,Phi3VForCausalLM:()=>$s,Phi3VPreTrainedModel:()=>Gs,PhiForCausalLM:()=>Po,PhiModel:()=>To,PhiPreTrainedModel:()=>vo,PreTrainedModel:()=>se,PretrainedMixin:()=>fc,PvtForImageClassification:()=>Ho,PvtModel:()=>Qo,PvtPreTrainedModel:()=>Xo,PyAnnoteForAudioFrameClassification:()=>ma,PyAnnoteModel:()=>pa,PyAnnotePreTrainedModel:()=>_a,QuestionAnsweringModelOutput:()=>Nd,Qwen2ForCausalLM:()=>wo,Qwen2Model:()=>fo,Qwen2PreTrainedModel:()=>go,Qwen2VLForConditionalGeneration:()=>yo,Qwen2VLPreTrainedModel:()=>ko,Qwen3ForCausalLM:()=>bo,Qwen3Model:()=>xo,Qwen3PreTrainedModel:()=>Mo,RFDetrForObjectDetection:()=>Bn,RFDetrModel:()=>Nn,RFDetrObjectDetectionOutput:()=>Rn,RFDetrPreTrainedModel:()=>Vn,RTDetrForObjectDetection:()=>Ln,RTDetrModel:()=>En,RTDetrObjectDetectionOutput:()=>In,RTDetrPreTrainedModel:()=>An,RTDetrV2ForObjectDetection:()=>On,RTDetrV2Model:()=>jn,RTDetrV2ObjectDetectionOutput:()=>Dn,RTDetrV2PreTrainedModel:()=>zn,ResNetForImageClassification:()=>ri,ResNetModel:()=>si,ResNetPreTrainedModel:()=>ti,RoFormerForMaskedLM:()=>xe,RoFormerForQuestionAnswering:()=>ye,RoFormerForSequenceClassification:()=>be,RoFormerForTokenClassification:()=>ke,RoFormerModel:()=>Me,RoFormerPreTrainedModel:()=>we,RobertaForMaskedLM:()=>os,RobertaForQuestionAnswering:()=>as,RobertaForSequenceClassification:()=>ns,RobertaForTokenClassification:()=>is,RobertaModel:()=>rs,RobertaPreTrainedModel:()=>ss,SamImageSegmentationOutput:()=>ea,SamModel:()=>Zi,SamPreTrainedModel:()=>Ki,SapiensForDepthEstimation:()=>wi,SapiensForNormalEstimation:()=>Mi,SapiensForSemanticSegmentation:()=>fi,SapiensPreTrainedModel:()=>gi,SegformerForImageClassification:()=>_l,SegformerForSemanticSegmentation:()=>pl,SegformerModel:()=>ul,SegformerPreTrainedModel:()=>dl,Seq2SeqLMOutput:()=>zd,SequenceClassifierOutput:()=>jd,SiglipModel:()=>Ys,SiglipPreTrainedModel:()=>Js,SiglipTextModel:()=>Ks,SiglipVisionModel:()=>Zs,SmolVLMForConditionalGeneration:()=>Rs,SnacDecoderModel:()=>gc,SnacEncoderModel:()=>hc,SnacModel:()=>mc,SnacPreTrainedModel:()=>pc,SpeechT5ForSpeechToText:()=>qa,SpeechT5ForTextToSpeech:()=>Wa,SpeechT5HifiGan:()=>Ua,SpeechT5Model:()=>$a,SpeechT5PreTrainedModel:()=>Ga,SqueezeBertForMaskedLM:()=>vt,SqueezeBertForQuestionAnswering:()=>Pt,SqueezeBertForSequenceClassification:()=>Tt,SqueezeBertModel:()=>yt,SqueezeBertPreTrainedModel:()=>kt,StableLmForCausalLM:()=>gl,StableLmModel:()=>hl,StableLmPreTrainedModel:()=>ml,Starcoder2ForCausalLM:()=>el,Starcoder2Model:()=>Za,Starcoder2PreTrainedModel:()=>Ka,StyleTextToSpeech2Model:()=>Ra,StyleTextToSpeech2PreTrainedModel:()=>Ba,Swin2SRForImageSuperResolution:()=>di,Swin2SRModel:()=>ci,Swin2SRPreTrainedModel:()=>li,SwinForImageClassification:()=>ii,SwinForSemanticSegmentation:()=>ai,SwinModel:()=>ni,SwinPreTrainedModel:()=>oi,T5ForConditionalGeneration:()=>zt,T5Model:()=>It,T5PreTrainedModel:()=>Lt,TableTransformerForObjectDetection:()=>Xn,TableTransformerModel:()=>Un,TableTransformerObjectDetectionOutput:()=>Qn,TableTransformerPreTrainedModel:()=>Wn,TokenClassifierOutput:()=>Dd,TrOCRForCausalLM:()=>Qa,TrOCRPreTrainedModel:()=>Xa,UltravoxModel:()=>ec,UltravoxPreTrainedModel:()=>Zl,UniSpeechForCTC:()=>Ma,UniSpeechForSequenceClassification:()=>xa,UniSpeechModel:()=>wa,UniSpeechPreTrainedModel:()=>fa,UniSpeechSatForAudioFrameClassification:()=>Ta,UniSpeechSatForCTC:()=>ya,UniSpeechSatForSequenceClassification:()=>va,UniSpeechSatModel:()=>ka,UniSpeechSatPreTrainedModel:()=>ba,ViTForImageClassification:()=>Ro,ViTMAEModel:()=>Yo,ViTMAEPreTrainedModel:()=>Jo,ViTMSNForImageClassification:()=>en,ViTMSNModel:()=>Zo,ViTMSNPreTrainedModel:()=>Ko,ViTModel:()=>Bo,ViTPreTrainedModel:()=>No,VisionEncoderDecoderModel:()=>As,VitMatteForImageMatting:()=>ln,VitMattePreTrainedModel:()=>an,VitPoseForPoseEstimation:()=>Uo,VitPosePreTrainedModel:()=>Wo,VitsModel:()=>cl,VitsModelOutput:()=>$d,VitsPreTrainedModel:()=>ll,Wav2Vec2BertForCTC:()=>Ca,Wav2Vec2BertForSequenceClassification:()=>Sa,Wav2Vec2BertModel:()=>Fa,Wav2Vec2BertPreTrainedModel:()=>Pa,Wav2Vec2ForAudioFrameClassification:()=>ua,Wav2Vec2ForCTC:()=>ca,Wav2Vec2ForSequenceClassification:()=>da,Wav2Vec2Model:()=>la,Wav2Vec2PreTrainedModel:()=>aa,WavLMForAudioFrameClassification:()=>Na,WavLMForCTC:()=>Oa,WavLMForSequenceClassification:()=>Da,WavLMForXVector:()=>Va,WavLMModel:()=>ja,WavLMPreTrainedModel:()=>za,WeSpeakerResNetModel:()=>ga,WeSpeakerResNetPreTrainedModel:()=>ha,WhisperForConditionalGeneration:()=>Ts,WhisperModel:()=>vs,WhisperPreTrainedModel:()=>ys,XLMForQuestionAnswering:()=>ps,XLMForSequenceClassification:()=>us,XLMForTokenClassification:()=>_s,XLMModel:()=>cs,XLMPreTrainedModel:()=>ls,XLMRobertaForMaskedLM:()=>gs,XLMRobertaForQuestionAnswering:()=>Ms,XLMRobertaForSequenceClassification:()=>fs,XLMRobertaForTokenClassification:()=>ws,XLMRobertaModel:()=>hs,XLMRobertaPreTrainedModel:()=>ms,XLMWithLMHeadModel:()=>ds,XVectorOutput:()=>Od,YolosForObjectDetection:()=>Ji,YolosModel:()=>Hi,YolosObjectDetectionOutput:()=>Yi,YolosPreTrainedModel:()=>Qi});var r=s("./src/configs.js"),o=s("./src/backends/onnx.js"),n=s("./src/utils/dtypes.js"),i=s("./src/utils/generic.js"),a=s("./src/utils/core.js"),l=s("./src/utils/hub.js"),c=s("./src/utils/constants.js"),d=s("./src/generation/logits_process.js"),u=s("./src/generation/configuration_utils.js"),_=s("./src/utils/tensor.js"),p=s("./src/utils/image.js"),m=s("./src/utils/maths.js"),h=s("./src/generation/stopping_criteria.js"),g=s("./src/generation/logits_sampler.js"),f=s("./src/env.js"),w=s("./src/models/whisper/generation_whisper.js"),M=s("./src/models/whisper/common_whisper.js");const x=0,b=1,k=2,y=3,v=4,T=5,P=6,F=7,C=8,S=9,A=10,E=11,L=new Map,I=new Map,z=new Map;async function j(e,t,s){return Object.fromEntries(await Promise.all(Object.keys(t).map((async i=>{const{buffer_or_path:a,session_options:c,session_config:d}=await async function(e,t,s){let i=s.config?.["transformers.js_config"]??{},a=s.device??i.device;a&&"string"!=typeof a&&(a.hasOwnProperty(t)?a=a[t]:(console.warn(`device not specified for "${t}". Using the default device.`),a=null));const c=a??(f.apis.IS_NODE_ENV?"cpu":"wasm"),d=(0,o.deviceToExecutionProviders)(c),u=i.device_config??{};u.hasOwnProperty(c)&&(i={...i,...u[c]});let _=s.dtype??i.dtype;if("string"!=typeof _&&(_&&_.hasOwnProperty(t)?_=_[t]:(_=n.DEFAULT_DEVICE_DTYPE_MAPPING[c]??n.DATA_TYPES.fp32,console.warn(`dtype not specified for "${t}". Using the default dtype (${_}) for this device (${c}).`))),_===n.DATA_TYPES.auto){let e=i.dtype;"string"!=typeof e&&(e=e?.[t]),_=e&&e!==n.DATA_TYPES.auto&&n.DATA_TYPES.hasOwnProperty(e)?e:n.DEFAULT_DEVICE_DTYPE_MAPPING[c]??n.DATA_TYPES.fp32}const p=_;if(!n.DEFAULT_DTYPE_SUFFIX_MAPPING.hasOwnProperty(p))throw new Error(`Invalid dtype: ${p}. Should be one of: ${Object.keys(n.DATA_TYPES).join(", ")}`);if(p===n.DATA_TYPES.fp16&&"webgpu"===c&&!await(0,n.isWebGpuFp16Supported)())throw new Error(`The device (${c}) does not support fp16.`);const m=i.kv_cache_dtype,h=m?"string"==typeof m?m:m[p]??"float32":void 0;if(h&&!["float32","float16"].includes(h))throw new Error(`Invalid kv_cache_dtype: ${h}. Should be one of: float32, float16`);const g={dtype:p,kv_cache_dtype:h,device:c},w=`${t}${n.DEFAULT_DTYPE_SUFFIX_MAPPING[p]}.onnx`,M=`${s.subfolder??""}/${w}`,x={...s.session_options};x.executionProviders??=d;const b=i.free_dimension_overrides;b?x.freeDimensionOverrides??=b:c.startsWith("webnn")&&!x.freeDimensionOverrides&&console.warn(`WebNN does not currently support dynamic shapes and requires 'free_dimension_overrides' to be set in config.json, preferably as a field within config["transformers.js_config"]["device_config"]["${c}"]. When 'free_dimension_overrides' is not set, you may experience significant performance degradation.`);const k=f.apis.IS_NODE_ENV&&f.env.useFSCache,y=(0,l.getModelFile)(e,M,!0,s,k),v=s.use_external_data_format??i.use_external_data_format;let T=[];if(v){let r;r="object"==typeof v?v.hasOwnProperty(w)?v[w]:!!v.hasOwnProperty(t)&&v[t]:v;const o=+r;if(o>l.MAX_EXTERNAL_DATA_CHUNKS)throw new Error(`The number of external data chunks (${o}) exceeds the maximum allowed value (${l.MAX_EXTERNAL_DATA_CHUNKS}).`);for(let t=0;t<o;++t){const r=`${w}_data${0===t?"":"_"+t}`,o=`${s.subfolder??""}/${r}`;T.push(new Promise((async(t,n)=>{const i=await(0,l.getModelFile)(e,o,!0,s,k);t(i instanceof Uint8Array?{path:r,data:i}:r)})))}}else void 0!==x.externalData&&(T=x.externalData.map((async t=>{if("string"==typeof t.data){const r=await(0,l.getModelFile)(e,t.data,!0,s);return{...t,data:r}}return t})));if(T.length>0){const e=await Promise.all(T);f.apis.IS_NODE_ENV||(x.externalData=e)}if("webgpu"===c){const e=(0,r.getKeyValueShapes)(s.config,{prefix:"present"});if(Object.keys(e).length>0&&!(0,o.isONNXProxy)()){const t={};for(const s in e)t[s]="gpu-buffer";x.preferredOutputLocation=t}}return{buffer_or_path:await y,session_options:x,session_config:g}}(e,t[i],s);return[i,await(0,o.createInferenceSession)(a,c,d)]}))))}async function O(e,t,s){return Object.fromEntries(await Promise.all(Object.keys(t).map((async r=>[r,await(0,l.getModelJSON)(e,t[r],!1,s)]))))}let D=Promise.resolve();async function V(e,t){const s=function(e,t){const s=Object.create(null),r=[];for(const n of e.inputNames){const e=t[n];e instanceof _.Tensor?s[n]=(0,o.isONNXProxy)()?e.clone():e:r.push(n)}if(r.length>0)throw new Error(`An error occurred during model execution: "Missing the following inputs: ${r.join(", ")}.`);const n=Object.keys(t).length,i=e.inputNames.length;if(n>i){let s=Object.keys(t).filter((t=>!e.inputNames.includes(t)));console.warn(`WARNING: Too many inputs were provided (${n} > ${i}). The following inputs will be ignored: "${s.join(", ")}".`)}return s}(e,t);try{const t=Object.fromEntries(Object.entries(s).map((([e,t])=>[e,t.ort_tensor]))),r=()=>e.run(t);return N(await(f.apis.IS_BROWSER_ENV||f.apis.IS_WEBWORKER_ENV?D=D.then(r):r()))}catch(e){const t=Object.fromEntries(Object.entries(s).map((([e,t])=>{const s={type:t.type,dims:t.dims,location:t.location};return"gpu-buffer"!==s.location&&(s.data=t.data),[e,s]})));throw console.error(`An error occurred during model execution: "${e}".`),console.error("Inputs given to model:",t),e}}function N(e){for(let t in e)(0,o.isONNXTensor)(e[t])?e[t]=new _.Tensor(e[t]):"object"==typeof e[t]&&N(e[t]);return e}function B(e){if(e instanceof _.Tensor)return e;if(0===e.length)throw Error("items must be non-empty");if(Array.isArray(e[0])){if(e.some((t=>t.length!==e[0].length)))throw Error("Unable to create tensor, you should probably activate truncation and/or padding with 'padding=True' and/or 'truncation=True' to have batched tensors with the same length.");return new _.Tensor("int64",BigInt64Array.from(e.flat().map((e=>BigInt(e)))),[e.length,e[0].length])}return new _.Tensor("int64",BigInt64Array.from(e.map((e=>BigInt(e)))),[1,e.length])}function R(e){return new _.Tensor("bool",[e],[1])}async function G(e,t){let{encoder_outputs:s,input_ids:r,decoder_input_ids:o,...n}=t;if(!s){const r=(0,a.pick)(t,e.sessions.model.inputNames);s=(await $(e,r)).last_hidden_state}n.input_ids=o,n.encoder_hidden_states=s,e.sessions.decoder_model_merged.inputNames.includes("encoder_attention_mask")&&(n.encoder_attention_mask=t.attention_mask);return await W(e,n,!0)}async function $(e,t){const s=e.sessions.model,r=(0,a.pick)(t,s.inputNames);if(s.inputNames.includes("inputs_embeds")&&!r.inputs_embeds){if(!t.input_ids)throw new Error("Both `input_ids` and `inputs_embeds` are missing in the model inputs.");r.inputs_embeds=await e.encode_text({input_ids:t.input_ids})}if(s.inputNames.includes("token_type_ids")&&!r.token_type_ids){if(!r.input_ids)throw new Error("Both `input_ids` and `token_type_ids` are missing in the model inputs.");r.token_type_ids=(0,_.zeros_like)(r.input_ids)}if(s.inputNames.includes("pixel_mask")&&!r.pixel_mask){if(!r.pixel_values)throw new Error("Both `pixel_values` and `pixel_mask` are missing in the model inputs.");const e=r.pixel_values.dims;r.pixel_mask=(0,_.ones)([e[0],e[2],e[3]])}return await V(s,r)}async function q(e,t){const s=await e.encode(t);return await e.decode(s)}async function W(e,t,s=!1){const r=e.sessions[s?"decoder_model_merged":"model"],{past_key_values:o,...n}=t;if(r.inputNames.includes("use_cache_branch")&&(n.use_cache_branch=R(!!o)),r.inputNames.includes("position_ids")&&n.attention_mask&&!n.position_ids){const t=["paligemma","gemma3_text","gemma3"].includes(e.config.model_type)?1:0;n.position_ids=function(e,t=null,s=0){const{input_ids:r,inputs_embeds:o,attention_mask:n}=e,{data:i,dims:a}=Y(n,s);let l=new _.Tensor("int64",i,a);if(t){const e=-(r??o).dims.at(1);l=l.slice(null,[e,null])}return l}(n,o,t)}e.addPastKeyValues(n,o);const i=(0,a.pick)(n,r.inputNames);return await V(r,i)}function U({modality_token_id:e,inputs_embeds:t,modality_features:s,input_ids:r,attention_mask:o}){const n=r.tolist().map((t=>t.reduce(((t,s,r)=>(s==e&&t.push(r),t)),[]))),i=n.reduce(((e,t)=>e+t.length),0),a=s.dims[0];if(i!==a)throw new Error(`Number of tokens and features do not match: tokens: ${i}, features ${a}`);let l=0;for(let e=0;e<n.length;++e){const r=n[e],o=t[e];for(let e=0;e<r.length;++e)o[r[e]].data.set(s[l++].data)}return{inputs_embeds:t,attention_mask:o}}function X({image_token_id:e,inputs_embeds:t,image_features:s,input_ids:r,attention_mask:o}){return U({modality_token_id:e,inputs_embeds:t,modality_features:s,input_ids:r,attention_mask:o})}async function Q(e,{encode_function:t,merge_function:s,modality_input_name:r,modality_output_name:o,input_ids:n=null,attention_mask:i=null,position_ids:a=null,inputs_embeds:l=null,past_key_values:c=null,generation_config:d=null,logits_processor:u=null,...p}){const m=p[r];if(!l)if(l=await e.encode_text({input_ids:n,...p}),m&&1!==n.dims[1]){const e=await t({[r]:m,...p});({inputs_embeds:l,attention_mask:i}=s({[o]:e,inputs_embeds:l,input_ids:n,attention_mask:i}))}else if(c&&m&&1===n.dims[1]){const e=n.dims[1],t=Object.values(c)[0].dims.at(-2);i=(0,_.cat)([(0,_.ones)([n.dims[0],t]),i.slice(null,[i.dims[1]-e,i.dims[1]])],1)}if(!a&&"qwen2_vl"===e.config.model_type){const{image_grid_thw:t,video_grid_thw:s}=p;[a]=e.get_rope_index(n,t,s,i)}return await W(e,{inputs_embeds:l,past_key_values:c,attention_mask:i,position_ids:a,generation_config:d,logits_processor:u},!0)}async function H(e,t){return await Q(e,{...t,modality_input_name:"audio_values",modality_output_name:"audio_features",encode_function:e.encode_audio.bind(e),merge_function:e._merge_input_ids_with_audio_features.bind(e)})}async function J(e,t){return await Q(e,{...t,modality_input_name:"pixel_values",modality_output_name:"image_features",encode_function:e.encode_image.bind(e),merge_function:e._merge_input_ids_with_image_features.bind(e)})}function Y(e,t=0){const[s,r]=e.dims,o=e.data,n=new BigInt64Array(o.length);for(let e=0;e<s;++e){const s=e*r;let i=BigInt(t);for(let e=0;e<r;++e){const t=s+e;0n===o[t]?n[t]=BigInt(1):(n[t]=i,i+=o[t])}}return{data:n,dims:e.dims}}function K(e,t,s,r){if(s.past_key_values){const t=Object.values(s.past_key_values)[0].dims.at(-2),{input_ids:r,attention_mask:o}=s;if(o&&o.dims[1]>r.dims[1]);else if(t<r.dims[1])s.input_ids=r.slice(null,[t,null]);else if(null!=e.config.image_token_index&&r.data.some((t=>t==e.config.image_token_index))){const o=e.config.num_image_tokens;if(!o)throw new Error("`num_image_tokens` is missing in the model configuration.");const n=r.dims[1]-(t-o);s.input_ids=r.slice(null,[-n,null]),s.attention_mask=(0,_.ones)([1,t+n])}}return s}function Z(e,t,s,r){return s.past_key_values&&(t=t.map((e=>[e.at(-1)]))),{...s,decoder_input_ids:B(t)}}function ee(e,...t){return e.config.is_encoder_decoder?Z(e,...t):K(e,...t)}function te(e,t,s,r){const o=!!s.past_key_values;if(null!==r.guidance_scale&&r.guidance_scale>1&&(o?s.input_ids=(0,_.cat)([s.input_ids,s.input_ids],0):(s.input_ids=(0,_.cat)([s.input_ids,(0,_.full_like)(s.input_ids,BigInt(r.pad_token_id))],0),s.attention_mask=(0,_.cat)([s.attention_mask,(0,_.full_like)(s.attention_mask,0n)],0))),!o&&s.pixel_values||(s.pixel_values=(0,_.full)([0,0,3,384,384],1)),o){const e=0,t=1,r=e>0?1:0,o=1;s.images_seq_mask=new _.Tensor("bool",new Array(e+t).fill(!0).fill(!1,0,t),[o,e+t]),s.images_emb_mask=new _.Tensor("bool",new Array(e).fill(!!r),[o,1,e])}return s}class se extends i.Callable{main_input_name="input_ids";forward_params=["input_ids","attention_mask"];constructor(e,t,s){super(),this.config=e,this.sessions=t,this.configs=s;const r=z.get(this.constructor),o=L.get(r);switch(this.can_generate=!1,this._forward=null,this._prepare_inputs_for_generation=null,o){case v:this.can_generate=!0,this._forward=W,this._prepare_inputs_for_generation=K;break;case k:case y:case F:this.can_generate=!0,this._forward=G,this._prepare_inputs_for_generation=Z;break;case b:this._forward=G;break;case P:this.can_generate=!0,this._forward=J,this._prepare_inputs_for_generation=ee;break;case A:this.can_generate=!0,this._forward=H,this._prepare_inputs_for_generation=ee;break;case S:this.can_generate=!0,this._prepare_inputs_for_generation=ee;break;case C:this.can_generate=!0,this._prepare_inputs_for_generation=te;break;case E:this._forward=q;break;default:this._forward=$}this.can_generate&&this.forward_params.push("past_key_values"),this.custom_config=this.config["transformers.js_config"]??{}}async dispose(){const e=[];for(const t of Object.values(this.sessions))t?.handler?.dispose&&e.push(t.handler.dispose());return await Promise.all(e)}static async from_pretrained(e,{progress_callback:t=null,config:s=null,cache_dir:o=null,local_files_only:n=!1,revision:i="main",model_file_name:a=null,subfolder:l="onnx",device:d=null,dtype:u=null,use_external_data_format:_=null,session_options:p={}}={}){let m={progress_callback:t,config:s,cache_dir:o,local_files_only:n,revision:i,model_file_name:a,subfolder:l,device:d,dtype:u,use_external_data_format:_,session_options:p};const h=z.get(this),g=L.get(h);let f;if(s=m.config=await r.AutoConfig.from_pretrained(e,m),g===v)f=await Promise.all([j(e,{model:m.model_file_name??"model"},m),O(e,{generation_config:"generation_config.json"},m)]);else if(g===k||g===y)f=await Promise.all([j(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},m),O(e,{generation_config:"generation_config.json"},m)]);else if(g===T)f=await Promise.all([j(e,{model:"vision_encoder",prompt_encoder_mask_decoder:"prompt_encoder_mask_decoder"},m)]);else if(g===b)f=await Promise.all([j(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},m)]);else if(g===P){const t={embed_tokens:"embed_tokens",vision_encoder:"vision_encoder",decoder_model_merged:"decoder_model_merged"};s.is_encoder_decoder&&(t.model="encoder_model"),f=await Promise.all([j(e,t,m),O(e,{generation_config:"generation_config.json"},m)])}else if(g===A){const t={embed_tokens:"embed_tokens",audio_encoder:"audio_encoder",decoder_model_merged:"decoder_model_merged"};f=await Promise.all([j(e,t,m),O(e,{generation_config:"generation_config.json"},m)])}else if(g===F)f=await Promise.all([j(e,{model:"text_encoder",decoder_model_merged:"decoder_model_merged",encodec_decode:"encodec_decode"},m),O(e,{generation_config:"generation_config.json"},m)]);else if(g===C)f=await Promise.all([j(e,{prepare_inputs_embeds:"prepare_inputs_embeds",model:"language_model",lm_head:"lm_head",gen_head:"gen_head",gen_img_embeds:"gen_img_embeds",image_decode:"image_decode"},m),O(e,{generation_config:"generation_config.json"},m)]);else if(g===S)f=await Promise.all([j(e,{prepare_inputs_embeds:"prepare_inputs_embeds",model:"model",vision_encoder:"vision_encoder"},m),O(e,{generation_config:"generation_config.json"},m)]);else if(g===E)f=await Promise.all([j(e,{encoder_model:"encoder_model",decoder_model:"decoder_model"},m)]);else{if(g!==x){const e=h??s?.model_type;"custom"!==e&&console.warn(`Model type for '${e}' not found, assuming encoder-only architecture. Please report this at ${c.GITHUB_ISSUE_URL}.`)}f=await Promise.all([j(e,{model:m.model_file_name??"model"},m)])}return new this(s,...f)}async _call(e){return await this.forward(e)}async forward(e){return await this._forward(this,e)}get generation_config(){return this.configs?.generation_config??null}_get_logits_warper(e){const t=new d.LogitsProcessorList;return null!==e.temperature&&1!==e.temperature&&t.push(new d.TemperatureLogitsWarper(e.temperature)),null!==e.top_k&&0!==e.top_k&&t.push(new d.TopKLogitsWarper(e.top_k)),null!==e.top_p&&e.top_p<1&&t.push(new d.TopPLogitsWarper(e.top_p)),t}_get_logits_processor(e,t,s=null){const r=new d.LogitsProcessorList;if(null!==e.repetition_penalty&&1!==e.repetition_penalty&&r.push(new d.RepetitionPenaltyLogitsProcessor(e.repetition_penalty)),null!==e.no_repeat_ngram_size&&e.no_repeat_ngram_size>0&&r.push(new d.NoRepeatNGramLogitsProcessor(e.no_repeat_ngram_size)),null!==e.bad_words_ids&&r.push(new d.NoBadWordsLogitsProcessor(e.bad_words_ids,e.eos_token_id)),null!==e.min_length&&null!==e.eos_token_id&&e.min_length>0&&r.push(new d.MinLengthLogitsProcessor(e.min_length,e.eos_token_id)),null!==e.min_new_tokens&&null!==e.eos_token_id&&e.min_new_tokens>0&&r.push(new d.MinNewTokensLengthLogitsProcessor(t,e.min_new_tokens,e.eos_token_id)),null!==e.forced_bos_token_id&&r.push(new d.ForcedBOSTokenLogitsProcessor(e.forced_bos_token_id)),null!==e.forced_eos_token_id&&r.push(new d.ForcedEOSTokenLogitsProcessor(e.max_length,e.forced_eos_token_id)),null!==e.begin_suppress_tokens){const s=t>1||null===e.forced_bos_token_id?t:t+1;r.push(new d.SuppressTokensAtBeginLogitsProcessor(e.begin_suppress_tokens,s))}return null!==e.guidance_scale&&e.guidance_scale>1&&r.push(new d.ClassifierFreeGuidanceLogitsProcessor(e.guidance_scale)),null!==s&&r.extend(s),r}_prepare_generation_config(e,t,s=u.GenerationConfig){const r={...this.config};for(const e of["decoder","generator","text_config"])e in r&&Object.assign(r,r[e]);const o=new s(r);return Object.assign(o,this.generation_config??{}),e&&Object.assign(o,e),t&&Object.assign(o,(0,a.pick)(t,Object.getOwnPropertyNames(o))),o}_get_stopping_criteria(e,t=null){const s=new h.StoppingCriteriaList;return null!==e.max_length&&s.push(new h.MaxLengthCriteria(e.max_length,this.config.max_position_embeddings??null)),null!==e.eos_token_id&&s.push(new h.EosTokenCriteria(e.eos_token_id)),t&&s.extend(t),s}_validate_model_class(){if(!this.can_generate){const e=[Cc,Lc,Fc,kc],t=z.get(this.constructor),s=new Set,r=this.config.model_type;for(const t of e){const e=t.get(r);e&&s.add(e[0])}let o=`The current model class (${t}) is not compatible with \`.generate()\`, as it doesn't have a language model head.`;throw s.size>0&&(o+=` Please use the following class instead: ${[...s].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:s,is_encoder_decoder:r}){return s.past_key_values=this.getPastKeyValues(t,s.past_key_values),s.input_ids=new _.Tensor("int64",e.flat(),[e.length,1]),r||(s.attention_mask=(0,_.cat)([s.attention_mask,(0,_.ones)([s.attention_mask.dims[0],1])],1)),s.position_ids=null,s}_prepare_model_inputs({inputs:e,bos_token_id:t,model_kwargs:s}){const r=(0,a.pick)(s,this.forward_params),o=this.main_input_name;if(o in r){if(e)throw new Error("`inputs`: {inputs}` were passed alongside {input_name} which is not allowed. Make sure to either pass {inputs} or {input_name}=...")}else r[o]=e;return{inputs_tensor:r[o],model_inputs:r,model_input_name:o}}async _prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:e,model_inputs:t,model_input_name:s,generation_config:r}){if(this.sessions.model.inputNames.includes("inputs_embeds")&&!t.inputs_embeds&&"_prepare_inputs_embeds"in this){const{input_ids:e,pixel_values:s,attention_mask:r,...o}=t,n=await this._prepare_inputs_embeds(t);t={...o,...(0,a.pick)(n,["inputs_embeds","attention_mask"])}}let{last_hidden_state:o}=await $(this,t);if(null!==r.guidance_scale&&r.guidance_scale>1)o=(0,_.cat)([o,(0,_.full_like)(o,0)],0),"attention_mask"in t&&(t.attention_mask=(0,_.cat)([t.attention_mask,(0,_.zeros_like)(t.attention_mask)],0));else if(t.decoder_input_ids){const e=B(t.decoder_input_ids).dims[0];if(e!==o.dims[0]){if(1!==o.dims[0])throw new Error(`The encoder outputs have a different batch size (${o.dims[0]}) than the decoder inputs (${e}).`);o=(0,_.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:s,decoder_start_token_id:r,bos_token_id:o,generation_config:n}){let{decoder_input_ids:i,...a}=s;if(!(i instanceof _.Tensor)){if(i)Array.isArray(i[0])||(i=Array.from({length:e},(()=>i)));else if(r??=o,"musicgen"===this.config.model_type)i=Array.from({length:e*this.config.decoder.num_codebooks},(()=>[r]));else if(Array.isArray(r)){if(r.length!==e)throw new Error(`\`decoder_start_token_id\` expcted to have length ${e} but got ${r.length}`);i=r}else i=Array.from({length:e},(()=>[r]));i=B(i)}return s.decoder_attention_mask=(0,_.ones_like)(i),{input_ids:i,model_inputs:a}}async generate({inputs:e=null,generation_config:t=null,logits_processor:s=null,stopping_criteria:r=null,streamer:o=null,...n}){this._validate_model_class(),t=this._prepare_generation_config(t,n);let{inputs_tensor:i,model_inputs:a,model_input_name:l}=this._prepare_model_inputs({inputs:e,model_kwargs:n});const c=this.config.is_encoder_decoder;let d;c&&("encoder_outputs"in a||(a=await this._prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:i,model_inputs:a,model_input_name:l,generation_config:t}))),c?({input_ids:d,model_inputs:a}=this._prepare_decoder_input_ids_for_generation({batch_size:a[l].dims.at(0),model_input_name:l,model_kwargs:a,decoder_start_token_id:t.decoder_start_token_id,bos_token_id:t.bos_token_id,generation_config:t})):d=a[l];let u=d.dims.at(-1);null!==t.max_new_tokens&&(t.max_length=u+t.max_new_tokens);const p=this._get_logits_processor(t,u,s),m=this._get_stopping_criteria(t,r),h=a[l].dims.at(0),f=g.LogitsSampler.getSampler(t),w=new Array(h).fill(0),M=d.tolist();let x;o&&o.put(M);let b={};for(;;){if(a=this.prepare_inputs_for_generation(M,a,t),x=await this.forward(a),t.output_attentions&&t.return_dict_in_generate){const e=this.getAttentions(x);for(const t in e)t in b||(b[t]=[]),b[t].push(e[t])}const e=p(M,x.logits.slice(null,-1,null)),s=[];for(let t=0;t<e.dims.at(0);++t){const r=e[t],o=await f(r);for(const[e,r]of o){const o=BigInt(e);w[t]+=r,M[t].push(o),s.push([o]);break}}o&&o.put(s);if(m(M).every((e=>e)))break;a=this._update_model_kwargs_for_generation({generated_input_ids:s,outputs:x,model_inputs:a,is_encoder_decoder:c})}o&&o.end();const k=this.getPastKeyValues(x,a.past_key_values,!0),y=new _.Tensor("int64",M.flat(),[M.length,M[0].length]);if(t.return_dict_in_generate)return{sequences:y,past_key_values:k,...b};for(const e of Object.values(x))"gpu-buffer"===e.location&&e.dispose();return y}getPastKeyValues(e,t,s=!1){const r=Object.create(null);for(const o in e)if(o.startsWith("present")){const n=o.replace("present","past_key_values"),i=o.includes("encoder");if(r[n]=i&&t?t[n]:e[o],t&&(!i||s)){const e=t[n];"gpu-buffer"===e.location&&e.dispose()}}return r}getAttentions(e){const t={};for(const s of["cross_attentions","encoder_attentions","decoder_attentions"])for(const r in e)r.startsWith(s)&&(s in t||(t[s]=[]),t[s].push(e[r]));return t}addPastKeyValues(e,t){if(t)Object.assign(e,t);else{const t=this.sessions.decoder_model_merged??this.sessions.model,s=t?.config?.kv_cache_dtype??"float32",o="float16"===s?new _.DataTypeMap.float16:[],n=(e[this.main_input_name]??e.attention_mask)?.dims?.[0]??1,i=(0,r.getKeyValueShapes)(this.config,{batch_size:n});for(const t in i)e[t]=new _.Tensor(s,o,i[t])}}async encode_image({pixel_values:e}){const t=(await V(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 V(this.sessions.embed_tokens,{input_ids:e})).inputs_embeds}async encode_audio({audio_values:e}){return(await V(this.sessions.audio_encoder,{audio_values:e})).audio_features}}class re{}class oe extends re{constructor({last_hidden_state:e,hidden_states:t=null,attentions:s=null}){super(),this.last_hidden_state=e,this.hidden_states=t,this.attentions=s}}class ne extends se{}class ie extends ne{}class ae extends ne{async _call(e){return new Vd(await super._call(e))}}class le extends ne{async _call(e){return new jd(await super._call(e))}}class ce extends ne{async _call(e){return new Dd(await super._call(e))}}class de extends ne{async _call(e){return new Nd(await super._call(e))}}class ue extends se{}class _e extends ue{}class pe extends ue{async _call(e){return new Vd(await super._call(e))}}class me extends ue{async _call(e){return new jd(await super._call(e))}}class he extends ue{async _call(e){return new Dd(await super._call(e))}}class ge extends se{}class fe extends ge{}class we extends se{}class Me extends we{}class xe extends we{async _call(e){return new Vd(await super._call(e))}}class be extends we{async _call(e){return new jd(await super._call(e))}}class ke extends we{async _call(e){return new Dd(await super._call(e))}}class ye extends we{async _call(e){return new Nd(await super._call(e))}}class ve extends se{}class Te extends ve{}class Pe extends ve{async _call(e){return new Vd(await super._call(e))}}class Fe extends ve{async _call(e){return new jd(await super._call(e))}}class Ce extends ve{async _call(e){return new Dd(await super._call(e))}}class Se extends ve{async _call(e){return new Nd(await super._call(e))}}class Ae extends se{}class Ee extends Ae{}class Le extends Ae{async _call(e){return new Vd(await super._call(e))}}class Ie extends Ae{async _call(e){return new jd(await super._call(e))}}class ze extends Ae{async _call(e){return new Dd(await super._call(e))}}class je extends Ae{async _call(e){return new Nd(await super._call(e))}}class Oe extends se{}class De extends Oe{}class Ve extends Oe{async _call(e){return new Vd(await super._call(e))}}class Ne extends Oe{async _call(e){return new jd(await super._call(e))}}class Be extends Oe{async _call(e){return new Dd(await super._call(e))}}class Re extends Oe{async _call(e){return new Nd(await super._call(e))}}class Ge extends se{}class $e extends Ge{}class qe extends Ge{async _call(e){return new Vd(await super._call(e))}}class We extends Ge{async _call(e){return new jd(await super._call(e))}}class Ue extends Ge{async _call(e){return new Dd(await super._call(e))}}class Xe extends Ge{async _call(e){return new Nd(await super._call(e))}}class Qe extends se{}class He extends Qe{}class Je extends Qe{async _call(e){return new Vd(await super._call(e))}}class Ye extends Qe{async _call(e){return new jd(await super._call(e))}}class Ke extends Qe{async _call(e){return new Dd(await super._call(e))}}class Ze extends Qe{async _call(e){return new Nd(await super._call(e))}}class et extends se{}class tt extends et{}class st extends et{async _call(e){return new jd(await super._call(e))}}class rt extends et{async _call(e){return new Dd(await super._call(e))}}class ot extends et{async _call(e){return new Nd(await super._call(e))}}class nt extends et{async _call(e){return new Vd(await super._call(e))}}class it extends se{}class at extends it{}class lt extends it{async _call(e){return new Vd(await super._call(e))}}class ct extends it{async _call(e){return new jd(await super._call(e))}}class dt extends it{async _call(e){return new Dd(await super._call(e))}}class ut extends se{}class _t extends ut{}class pt extends ut{async _call(e){return new Vd(await super._call(e))}}class mt extends ut{async _call(e){return new jd(await super._call(e))}}class ht extends ut{async _call(e){return new Nd(await super._call(e))}}class gt extends se{}class ft extends gt{}class wt extends gt{async _call(e){return new Vd(await super._call(e))}}class Mt extends gt{async _call(e){return new jd(await super._call(e))}}class xt extends gt{async _call(e){return new Dd(await super._call(e))}}class bt extends gt{async _call(e){return new Nd(await super._call(e))}}class kt extends se{}class yt extends kt{}class vt extends kt{async _call(e){return new Vd(await super._call(e))}}class Tt extends kt{async _call(e){return new jd(await super._call(e))}}class Pt extends kt{async _call(e){return new Nd(await super._call(e))}}class Ft extends se{}class Ct extends Ft{}class St extends Ft{async _call(e){return new jd(await super._call(e))}}class At extends Ft{async _call(e){return new Nd(await super._call(e))}}class Et extends Ft{async _call(e){return new Vd(await super._call(e))}}class Lt extends se{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"]}class It extends Lt{}class zt extends Lt{}class jt extends se{}class Ot extends jt{}class Dt extends jt{}class Vt extends se{}class Nt extends Vt{}class Bt extends Vt{}class Rt extends se{}class Gt extends Rt{}class $t extends Rt{}class qt extends Rt{async _call(e){return new jd(await super._call(e))}}class Wt extends se{}class Ut extends Wt{}class Xt extends Wt{}class Qt extends Wt{async _call(e){return new jd(await super._call(e))}}class Ht extends Wt{}class Jt extends se{}class Yt extends Jt{}class Kt extends Jt{}class Zt extends se{}class es extends Zt{}class ts extends Zt{}class ss extends se{}class rs extends ss{}class os extends ss{async _call(e){return new Vd(await super._call(e))}}class ns extends ss{async _call(e){return new jd(await super._call(e))}}class is extends ss{async _call(e){return new Dd(await super._call(e))}}class as extends ss{async _call(e){return new Nd(await super._call(e))}}class ls extends se{}class cs extends ls{}class ds extends ls{async _call(e){return new Vd(await super._call(e))}}class us extends ls{async _call(e){return new jd(await super._call(e))}}class _s extends ls{async _call(e){return new Dd(await super._call(e))}}class ps extends ls{async _call(e){return new Nd(await super._call(e))}}class ms extends se{}class hs extends ms{}class gs extends ms{async _call(e){return new Vd(await super._call(e))}}class fs extends ms{async _call(e){return new jd(await super._call(e))}}class ws extends ms{async _call(e){return new Dd(await super._call(e))}}class Ms extends ms{async _call(e){return new Nd(await super._call(e))}}class xs extends se{}class bs extends xs{}class ks extends xs{}class ys extends se{requires_attention_mask=!1;main_input_name="input_features";forward_params=["input_features","attention_mask","decoder_input_ids","decoder_attention_mask","past_key_values"]}class vs extends ys{}class Ts extends ys{_prepare_generation_config(e,t){return super._prepare_generation_config(e,t,w.WhisperGenerationConfig)}_retrieve_init_tokens(e){const t=[e.decoder_start_token_id];let s=e.language;const r=e.task;if(e.is_multilingual){s||(console.warn("No language specified - defaulting to English (en)."),s="en");const o=`<|${(0,M.whisper_language_to_code)(s)}|>`;t.push(e.lang_to_id[o]),t.push(e.task_to_id[r??"transcribe"])}else if(s||r)throw new Error("Cannot specify `task` or `language` for an English-only model. If the model is intended to be multilingual, pass `is_multilingual=true` to generate, or update the generation config.");return!e.return_timestamps&&e.no_timestamps_token_id&&t.at(-1)!==e.no_timestamps_token_id?t.push(e.no_timestamps_token_id):e.return_timestamps&&t.at(-1)===e.no_timestamps_token_id&&(console.warn("<|notimestamps|> prompt token is removed from generation_config since `return_timestamps` is set to `true`."),t.pop()),t.filter((e=>null!=e))}async generate({inputs:e=null,generation_config:t=null,logits_processor:s=null,stopping_criteria:r=null,...o}){t=this._prepare_generation_config(t,o);const n=o.decoder_input_ids??this._retrieve_init_tokens(t);if(t.return_timestamps&&(s??=new d.LogitsProcessorList,s.push(new d.WhisperTimeStampLogitsProcessor(t,n))),t.begin_suppress_tokens&&(s??=new d.LogitsProcessorList,s.push(new d.SuppressTokensAtBeginLogitsProcessor(t.begin_suppress_tokens,n.length))),t.return_token_timestamps){if(!t.alignment_heads)throw new Error("Model generation config has no `alignment_heads`, token-level timestamps not available. See https://gist.github.com/hollance/42e32852f24243b748ae6bc1f985b13a on how to add this property to the generation config.");"translate"===t.task&&console.warn("Token-level timestamps may not be reliable for task 'translate'."),t.output_attentions=!0,t.return_dict_in_generate=!0}const i=await super.generate({inputs:e,generation_config:t,logits_processor:s,decoder_input_ids:n,...o});return t.return_token_timestamps&&(i.token_timestamps=this._extract_token_timestamps(i,t.alignment_heads,t.num_frames)),i}_extract_token_timestamps(e,t,s=null,r=.02){if(!e.cross_attentions)throw new Error("Model outputs must contain cross attentions to extract timestamps. This is most likely because the model was not exported with `output_attentions=True`.");null==s&&console.warn("`num_frames` has not been set, meaning the entire audio will be analyzed. This may lead to inaccurate token-level timestamps for short audios (< 30 seconds).");let 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 n=e.cross_attentions,i=Array.from({length:this.config.decoder_layers},((e,t)=>(0,_.cat)(n.map((e=>e[t])),2))),l=(0,_.stack)(t.map((([e,t])=>{if(e>=i.length)throw new Error(`Layer index ${e} is out of bounds for cross attentions (length ${i.length}).`);return s?i[e].slice(null,t,null,[0,s]):i[e].slice(null,t)}))).transpose(1,0,2,3),[c,d]=(0,_.std_mean)(l,-2,0,!0),u=l.clone();for(let e=0;e<u.dims[0];++e){const t=u[e];for(let s=0;s<t.dims[0];++s){const r=t[s],n=c[e][s][0].data,i=d[e][s][0].data;for(let e=0;e<r.dims[0];++e){let t=r[e].data;for(let e=0;e<t.length;++e)t[e]=(t[e]-i[e])/n[e];t.set((0,m.medianFilter)(t,o))}}}const p=[(0,_.mean)(u,1)],h=e.sequences.dims,g=new _.Tensor("float32",new Float32Array(h[0]*h[1]),h);for(let e=0;e<h[0];++e){const t=p[e].neg().squeeze_(0),[s,o]=(0,m.dynamic_time_warping)(t.tolist()),n=Array.from({length:s.length-1},((e,t)=>s[t+1]-s[t])),i=(0,a.mergeArrays)([1],n).map((e=>!!e)),l=[];for(let e=0;e<i.length;++e)i[e]&&l.push(o[e]*r);g[e].data.set(l,1)}return g}}class Ps extends Ts{}class Fs extends se{requires_attention_mask=!1;main_input_name="input_values";forward_params=["input_values","decoder_input_ids","past_key_values"]}class Cs extends Fs{}class Ss extends Fs{}class As extends se{main_input_name="pixel_values";forward_params=["pixel_values","decoder_input_ids","encoder_hidden_states","past_key_values"]}class Es extends se{forward_params=["input_ids","attention_mask","pixel_values","position_ids","past_key_values"]}class Ls extends Es{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:s,attention_mask:r}){const o=this.config.image_token_index,n=s.tolist().map((e=>e.findIndex((e=>e==o)))),i=n.every((e=>-1===e)),a=n.every((e=>-1!==e));if(!i&&!a)throw new Error("Every input should contain either 0 or 1 image token.");if(i)return{inputs_embeds:e,attention_mask:r};const l=[],c=[];for(let s=0;s<n.length;++s){const o=n[s],i=e[s],a=t[s],d=r[s];l.push((0,_.cat)([i.slice([0,o]),a,i.slice([o+1,i.dims[0]])],0)),c.push((0,_.cat)([d.slice([0,o]),(0,_.ones)([a.dims[0]]),d.slice([o+1,d.dims[0]])],0))}return{inputs_embeds:(0,_.stack)(l,0),attention_mask:(0,_.stack)(c,0)}}}class Is extends Ls{}class zs extends Ls{}class js extends se{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 Os extends js{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:s,attention_mask:r}){return{inputs_embeds:(0,_.cat)([t,e],1),attention_mask:(0,_.cat)([(0,_.ones)(t.dims.slice(0,2)),r],1)}}async _prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:s,attention_mask:r}){if(!e&&!t)throw new Error("Either `input_ids` or `pixel_values` should be provided.");let o,n;return e&&(o=await this.encode_text({input_ids:e})),t&&(n=await this.encode_image({pixel_values:t})),o&&n?({inputs_embeds:s,attention_mask:r}=this._merge_input_ids_with_image_features({inputs_embeds:o,image_features:n,input_ids:e,attention_mask:r})):s=o||n,{inputs_embeds:s,attention_mask:r}}async forward({input_ids:e,pixel_values:t,attention_mask:s,decoder_input_ids:r,decoder_attention_mask:o,encoder_outputs:n,past_key_values:i,inputs_embeds:a,decoder_inputs_embeds:l}){if(a||({inputs_embeds:a,attention_mask:s}=await this._prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:a,attention_mask:s})),!n){let{last_hidden_state:e}=await $(this,{inputs_embeds:a,attention_mask:s});n=e}if(!l){if(!r)throw new Error("Either `decoder_input_ids` or `decoder_inputs_embeds` should be provided.");l=await this.encode_text({input_ids:r})}const c={inputs_embeds:l,attention_mask:o,encoder_attention_mask:s,encoder_hidden_states:n,past_key_values:i};return await W(this,c,!0)}}class Ds extends se{forward_params=["input_ids","attention_mask","pixel_values","position_ids","past_key_values"]}class Vs extends Ds{_merge_input_ids_with_image_features(e){const t=e.image_features.dims.at(-1),s=e.image_features.view(-1,t);return X({image_token_id:this.config.image_token_index,...e,image_features:s})}}class Ns extends se{forward_params=["input_ids","attention_mask","pixel_values","pixel_attention_mask","position_ids","past_key_values"]}class Bs extends Ns{async encode_image({pixel_values:e,pixel_attention_mask:t}){return(await V(this.sessions.vision_encoder,{pixel_values:e,pixel_attention_mask:t})).image_features}_merge_input_ids_with_image_features(e){const t=e.image_features.dims.at(-1),s=e.image_features.view(-1,t);return X({image_token_id:this.config.image_token_id,...e,image_features:s})}}class Rs extends Bs{}class Gs extends se{forward_params=["input_ids","inputs_embeds","attention_mask","position_ids","pixel_values","image_sizes","past_key_values"]}class $s extends Gs{async forward({input_ids:e=null,attention_mask:t=null,pixel_values:s=null,image_sizes:r=null,position_ids:o=null,inputs_embeds:n=null,past_key_values:i=null,generation_config:a=null,logits_processor:l=null,...c}){if(!n){let t;if(s&&1!==e.dims[1]){if(!r)throw new Error("`image_sizes` must be provided when `pixel_values` is provided.");({image_features:t}=await V(this.sessions.vision_encoder,{pixel_values:s,image_sizes:r}))}else{const e=this.config.normalized_config.hidden_size;t=new _.Tensor("float32",[],[0,e])}({inputs_embeds:n}=await V(this.sessions.prepare_inputs_embeds,{input_ids:e,image_features:t}))}return await W(this,{inputs_embeds:n,past_key_values:i,attention_mask:t,position_ids:o,generation_config:a,logits_processor:l},!1)}}class qs extends se{}class Ws extends qs{}class Us extends qs{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"text_model"})}}class Xs extends qs{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"text_model"})}}class Qs extends qs{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"vision_model"})}}class Hs extends qs{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"vision_model"})}}class Js extends se{}class Ys extends Js{}class Ks extends Js{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"text_model"})}}class Zs extends qs{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"vision_model"})}}class er extends se{}class tr extends er{}class sr extends se{}class rr extends sr{async forward(e){const t=!e.input_ids,s=!e.pixel_values;if(t&&s)throw new Error("Either `input_ids` or `pixel_values` should be provided.");if(t&&(e.input_ids=(0,_.ones)([e.pixel_values.dims[0],1])),s){const{image_size:t}=this.config.vision_config;e.pixel_values=(0,_.full)([0,3,t,t],0)}const{text_embeddings:r,image_embeddings:o,l2norm_text_embeddings:n,l2norm_image_embeddings:i}=await super.forward(e),a={};return t||(a.text_embeddings=r,a.l2norm_text_embeddings=n),s||(a.image_embeddings=o,a.l2norm_image_embeddings=i),a}}class or extends sr{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"text_model"})}}class nr extends sr{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"vision_model"})}}class ir extends se{}class ar extends ir{}class lr extends ir{}class cr extends se{}class dr extends cr{}class ur extends cr{}class _r extends se{}class pr extends _r{}class mr extends _r{}class hr extends se{}class gr extends hr{}class fr extends hr{}class wr extends se{}class Mr extends wr{}class xr extends wr{}class br extends se{}class kr extends br{}class yr extends br{}class vr extends se{}class Tr extends vr{}class Pr extends vr{}class Fr extends se{}class Cr extends Fr{}class Sr extends Fr{}class Ar extends se{}class Er extends Ar{}class Lr extends Ar{}class Ir extends se{}class zr extends Ir{}class jr extends Ir{}class Or extends se{}class Dr extends Or{}class Vr extends Or{}class Nr extends se{}class Br extends Nr{}class Rr extends Nr{}class Gr extends se{}class $r extends Gr{}class qr extends Gr{}class Wr extends se{}class Ur extends Wr{}class Xr extends Wr{}class Qr extends se{}class Hr extends Qr{}class Jr extends Qr{}class Yr extends se{}class Kr extends Yr{}class Zr extends Yr{}class eo extends se{}class to extends eo{}class so extends eo{}class ro extends se{}class oo extends ro{}class no extends ro{}class io extends se{}class ao extends io{}class lo extends io{}class co extends se{}class uo extends co{}class _o extends co{}class po extends se{}class mo extends po{}class ho extends po{}class go extends se{}class fo extends go{}class wo extends go{}class Mo extends se{}class xo extends Mo{}class bo extends Mo{}class ko extends se{forward_params=["input_ids","attention_mask","position_ids","past_key_values","pixel_values","image_grid_thw"]}class yo extends ko{get_rope_index(e,t,s,r){const{vision_config:o,image_token_id:n,video_token_id:i,vision_start_token_id:a}=this.config,l=o.spatial_merge_size??2,c=[];if(t||s){let o=e.tolist();r||(r=(0,_.ones_like)(e));const d=r.tolist(),u=Array.from({length:3},(t=>Array.from({length:e.dims[0]},(t=>Array.from({length:e.dims[1]},(e=>1)))))),p=t?t.tolist():[],h=s?s.tolist():[];let g=0,f=0;for(let e=0;e<o.length;++e){const t=o[e].filter(((t,s)=>1==d[e][s])),s=t.reduce(((e,t,s)=>(t==a&&e.push(s),e)),[]).map((e=>t[e+1])),r=s.filter((e=>e==n)).length,_=s.filter((e=>e==i)).length;let w=[],M=0,x=r,b=_;for(let e=0;e<s.length;++e){const e=t.findIndex(((e,t)=>t>M&&e==n)),s=t.findIndex(((e,t)=>t>M&&e==i)),r=x>0&&-1!==e?e:t.length+1,o=b>0&&-1!==s?s:t.length+1;let a,c,d,u;r<o?([c,d,u]=p[g],++g,--x,a=r):([c,d,u]=h[f],++f,--b,a=o);const[_,k,y]=[Number(c),Math.floor(Number(d)/l),Math.floor(Number(u)/l)],v=a-M,T=w.length>0?(0,m.max)(w.at(-1))[0]+1:0;w.push(Array.from({length:3*v},((e,t)=>T+t%v)));const P=v+T,F=_*k*y,C=Array.from({length:F},((e,t)=>P+Math.floor(t/(k*y)))),S=Array.from({length:F},((e,t)=>P+Math.floor(t/y)%k)),A=Array.from({length:F},((e,t)=>P+t%y));w.push([C,S,A].flat()),M=a+F}if(M<t.length){const e=w.length>0?(0,m.max)(w.at(-1))[0]+1:0,s=t.length-M;w.push(Array.from({length:3*s},((t,r)=>e+r%s)))}const k=w.reduce(((e,t)=>e+t.length),0),y=new Array(k);let v=0;for(let e=0;e<3;++e)for(let t=0;t<w.length;++t){const s=w[t],r=s.length/3;for(let t=e*r;t<(e+1)*r;++t)y[v++]=s[t]}let T=0;const P=d[e];for(let t=0;t<P.length;++t)if(1==P[t]){for(let s=0;s<3;++s)u[s][e][t]=y[s*k/3+T];++T}const F=(0,m.max)(y)[0];c.push(F+1-o[e].length)}return[new _.Tensor("int64",u.flat(1/0),[3,e.dims[0],e.dims[1]]),new _.Tensor("int64",c,[c.length,1])]}if(r){const{data:e,dims:t}=Y(r),s=BigInt64Array.from({length:3*e.length},((t,s)=>e[s%e.length])),o=Array.from({length:t[0]},((s,r)=>(0,m.max)(e.subarray(t[1]*r,t[1]*(r+1)))[0]+1n+BigInt(t[1])));return[new _.Tensor("int64",s,[3,...t]),new _.Tensor("int64",o,[o.length,1])]}{const[t,s]=e.dims,r=BigInt64Array.from({length:3*t*s},((e,r)=>BigInt(Math.floor(r%s/t))));return[new _.Tensor("int64",r,[3,...e.dims]),(0,_.zeros)([t,1])]}}async encode_image({pixel_values:e,image_grid_thw:t}){return(await V(this.sessions.vision_encoder,{pixel_values:e,grid_thw:t})).image_features}_merge_input_ids_with_image_features(e){return X({image_token_id:this.config.image_token_id,...e})}prepare_inputs_for_generation(e,t,s){if(t.attention_mask&&!t.position_ids)if(t.past_key_values){t.pixel_values=null;const e=BigInt(Object.values(t.past_key_values)[0].dims.at(-2)),s=t.rope_deltas.map((t=>e+t));t.position_ids=(0,_.stack)([s,s,s],0)}else[t.position_ids,t.rope_deltas]=this.get_rope_index(t.input_ids,t.image_grid_thw,t.video_grid_thw,t.attention_mask);return t}}class vo extends se{}class To extends vo{}class Po extends vo{}class Fo extends se{}class Co extends Fo{}class So extends Fo{}class Ao extends se{}class Eo extends Ao{}class Lo extends Ao{}class Io extends se{}class zo extends Io{}class jo extends Io{}class Oo extends se{}class Do extends Oo{}class Vo extends Oo{}class No extends se{}class Bo extends No{}class Ro extends No{async _call(e){return new jd(await super._call(e))}}class Go extends se{}class $o extends Go{}class qo extends Go{async _call(e){return new jd(await super._call(e))}}class Wo extends se{}class Uo extends Wo{}class Xo extends se{}class Qo extends Xo{}class Ho extends Xo{async _call(e){return new jd(await super._call(e))}}class Jo extends se{}class Yo extends Jo{}class Ko extends se{}class Zo extends Ko{}class en extends Ko{async _call(e){return new jd(await super._call(e))}}class tn extends se{}class sn extends tn{}class rn extends se{}class on extends rn{}class nn extends rn{async _call(e){return new jd(await super._call(e))}}class an extends se{}class ln extends an{async _call(e){return new Gd(await super._call(e))}}class cn extends se{}class dn extends cn{}class un extends cn{async _call(e){return new jd(await super._call(e))}}class _n extends se{}class pn extends _n{}class mn extends _n{async _call(e){return new jd(await super._call(e))}}class hn extends se{}class gn extends hn{}class fn extends hn{}class wn extends se{}class Mn extends wn{}class xn extends wn{}class bn extends se{}class kn extends bn{}class yn extends bn{async _call(e){return new jd(await super._call(e))}}class vn extends se{}class Tn extends vn{}class Pn extends vn{async _call(e){return new Cn(await super._call(e))}}class Fn extends vn{async _call(e){return new Sn(await super._call(e))}}class Cn extends re{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class Sn extends re{constructor({logits:e,pred_boxes:t,pred_masks:s}){super(),this.logits=e,this.pred_boxes=t,this.pred_masks=s}}class An extends se{}class En extends An{}class Ln extends An{async _call(e){return new In(await super._call(e))}}class In extends re{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class zn extends se{}class jn extends zn{}class On extends zn{async _call(e){return new Dn(await super._call(e))}}class Dn extends In{}class Vn extends se{}class Nn extends Vn{}class Bn extends Vn{async _call(e){return new Rn(await super._call(e))}}class Rn extends In{}class Gn extends se{}class $n extends Gn{}class qn extends Gn{async _call(e){return new In(await super._call(e))}}class Wn extends se{}class Un extends Wn{}class Xn extends Wn{async _call(e){return new Qn(await super._call(e))}}class Qn extends Cn{}class Hn extends se{}class Jn extends Hn{}class Yn extends Hn{async _call(e){return new jd(await super._call(e))}}class Kn extends se{}class Zn extends Kn{}class ei extends Kn{async _call(e){return new jd(await super._call(e))}}class ti extends se{}class si extends ti{}class ri extends ti{async _call(e){return new jd(await super._call(e))}}class oi extends se{}class ni extends oi{}class ii extends oi{async _call(e){return new jd(await super._call(e))}}class ai extends oi{}class li extends se{}class ci extends li{}class di extends li{}class ui extends se{}class _i extends ui{}class pi extends ui{}class mi extends se{}class hi extends mi{}class gi extends se{}class fi extends gi{}class wi extends gi{}class Mi extends gi{}class xi extends se{}class bi extends xi{}class ki extends se{}class yi extends ki{}class vi extends se{}class Ti extends vi{}class Pi extends se{}class Fi extends Pi{}class Ci extends Pi{}class Si extends se{}class Ai extends Si{}class Ei extends Si{}class Li extends se{}class Ii extends Li{}class zi extends se{}class ji extends zi{}class Oi extends zi{async _call(e){return new jd(await super._call(e))}}class Di extends se{}class Vi extends Di{}class Ni extends Di{async _call(e){return new jd(await super._call(e))}}class Bi extends se{}class Ri extends Bi{}class Gi extends Bi{async _call(e){return new jd(await super._call(e))}}class $i extends se{}class qi extends $i{}class Wi extends $i{async _call(e){return new jd(await super._call(e))}}class Ui extends se{}class Xi extends Ui{}class Qi extends se{}class Hi extends Qi{}class Ji extends Qi{async _call(e){return new Yi(await super._call(e))}}class Yi extends re{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class Ki extends se{}class Zi extends Ki{async get_image_embeddings({pixel_values:e}){return await $(this,{pixel_values:e})}async forward(e){if(e.image_embeddings&&e.image_positional_embeddings||(e={...e,...await this.get_image_embeddings(e)}),!e.input_labels&&e.input_points){const t=e.input_points.dims.slice(0,-1),s=t.reduce(((e,t)=>e*t),1);e.input_labels=new _.Tensor("int64",new BigInt64Array(s).fill(1n),t)}const t={image_embeddings:e.image_embeddings,image_positional_embeddings:e.image_positional_embeddings};return e.input_points&&(t.input_points=e.input_points),e.input_labels&&(t.input_labels=e.input_labels),e.input_boxes&&(t.input_boxes=e.input_boxes),await V(this.sessions.prompt_encoder_mask_decoder,t)}async _call(e){return new ea(await super._call(e))}}class ea extends re{constructor({iou_scores:e,pred_masks:t}){super(),this.iou_scores=e,this.pred_masks=t}}class ta extends se{}class sa extends ta{}class ra extends ta{}class oa extends se{}class na extends oa{}class ia extends oa{}class aa extends se{}class la extends aa{}class ca extends aa{async _call(e){return new Bd(await super._call(e))}}class da extends aa{async _call(e){return new jd(await super._call(e))}}class ua extends aa{async _call(e){return new Dd(await super._call(e))}}class _a extends se{}class pa extends _a{}class ma extends _a{async _call(e){return new Dd(await super._call(e))}}class ha extends se{}class ga extends ha{}class fa extends se{}class wa extends fa{}class Ma extends fa{async _call(e){return new Bd(await super._call(e))}}class xa extends fa{async _call(e){return new jd(await super._call(e))}}class ba extends se{}class ka extends ba{}class ya extends ba{async _call(e){return new Bd(await super._call(e))}}class va extends ba{async _call(e){return new jd(await super._call(e))}}class Ta extends ba{async _call(e){return new Dd(await super._call(e))}}class Pa extends se{}class Fa extends Pa{}class Ca extends Pa{async _call(e){return new Bd(await super._call(e))}}class Sa extends Pa{async _call(e){return new jd(await super._call(e))}}class Aa extends se{}class Ea extends aa{}class La extends aa{async _call(e){return new Bd(await super._call(e))}}class Ia extends aa{async _call(e){return new jd(await super._call(e))}}class za extends se{}class ja extends za{}class Oa extends za{async _call(e){return new Bd(await super._call(e))}}class Da extends za{async _call(e){return new jd(await super._call(e))}}class Va extends za{async _call(e){return new Od(await super._call(e))}}class Na extends za{async _call(e){return new Dd(await super._call(e))}}class Ba extends se{}class Ra extends Ba{}class Ga extends se{}class $a extends Ga{}class qa extends Ga{}class Wa extends Ga{async generate_speech(e,t,{threshold:s=.5,minlenratio:r=0,maxlenratio:o=20,vocoder:n=null}={}){const i={input_ids:e},{encoder_outputs:a,encoder_attention_mask:l}=await $(this,i),c=a.dims[1]/this.config.reduction_factor,d=Math.floor(c*o),u=Math.floor(c*r),p=this.config.num_mel_bins;let m=[],h=null,g=null,f=0;for(;;){++f;const e=R(!!g);let r;r=g?g.output_sequence_out:new _.Tensor("float32",new Float32Array(p),[1,1,p]);let o={use_cache_branch:e,output_sequence:r,encoder_attention_mask:l,speaker_embeddings:t,encoder_hidden_states:a};this.addPastKeyValues(o,h),g=await V(this.sessions.decoder_model_merged,o),h=this.getPastKeyValues(g,h);const{prob:n,spectrum:i}=g;if(m.push(i),f>=u&&(Array.from(n.data).filter((e=>e>=s)).length>0||f>=d))break}const w=(0,_.cat)(m),{waveform:M}=await V(n.sessions.model,{spectrogram:w});return{spectrogram:w,waveform:M}}}class Ua extends se{main_input_name="spectrogram"}class Xa extends se{}class Qa extends Xa{}class Ha extends se{}class Ja extends Ha{}class Ya extends Ha{}class Ka extends se{}class Za extends Ka{}class el extends Ka{}class tl extends se{}class sl extends tl{}class rl extends tl{}class ol extends se{}class nl extends ol{}class il extends ol{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"text_model"})}}class al extends ol{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"audio_model"})}}class ll extends se{}class cl extends ll{async _call(e){return new $d(await super._call(e))}}class dl extends se{}class ul extends dl{}class _l extends dl{}class pl extends dl{}class ml extends se{}class hl extends ml{}class gl extends ml{}class fl extends se{}class wl extends fl{}class Ml extends fl{async _call(e){return new jd(await super._call(e))}}class xl extends se{}class bl extends xl{}class kl extends xl{}class yl extends se{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"];_apply_and_filter_by_delay_pattern_mask(e){const[t,s]=e.dims,r=this.config.decoder.num_codebooks,o=s-r;let n=0;for(let t=0;t<e.size;++t){if(e.data[t]===this.config.decoder.pad_token_id)continue;const i=t%s-Math.floor(t/s)%r;i>0&&i<=o&&(e.data[n++]=e.data[t])}const i=Math.floor(t/r),a=n/(i*r);return new _.Tensor(e.type,e.data.slice(0,n),[i,r,a])}prepare_inputs_for_generation(e,t,s){let r=structuredClone(e);for(let e=0;e<r.length;++e)for(let t=0;t<r[e].length;++t)e%this.config.decoder.num_codebooks>=t&&(r[e][t]=BigInt(this.config.decoder.pad_token_id));null!==s.guidance_scale&&s.guidance_scale>1&&(r=r.concat(r));return super.prepare_inputs_for_generation(r,t,s)}async generate(e){const t=await super.generate(e),s=this._apply_and_filter_by_delay_pattern_mask(t).unsqueeze_(0),{audio_values:r}=await V(this.sessions.encodec_decode,{audio_codes:s});return r}}class vl extends se{}class Tl extends vl{}class Pl extends vl{async _call(e){return new jd(await super._call(e))}}class Fl extends vl{}class Cl extends se{}class Sl extends Cl{}class Al extends Cl{async _call(e){return new jd(await super._call(e))}}class El extends Cl{}class Ll extends se{}class Il extends Ll{}class zl extends Ll{async _call(e){return new jd(await super._call(e))}}class jl extends Ll{}class Ol extends se{}class Dl extends Ol{}class Vl extends Ol{async _call(e){return new jd(await super._call(e))}}class Nl extends Ol{}class Bl extends se{}class Rl extends Bl{}class Gl extends se{}class $l extends Gl{forward_params=["input_ids","pixel_values","images_seq_mask","images_emb_mask","attention_mask","position_ids","past_key_values"];constructor(...e){super(...e),this._generation_mode="text"}async forward(e){const t=this._generation_mode??"text";let s;if("text"!==t&&e.past_key_values){const t=this.sessions.gen_img_embeds,r=(0,a.pick)({image_ids:e.input_ids},t.inputNames);s=await V(t,r)}else{const t=this.sessions.prepare_inputs_embeds,r=(0,a.pick)(e,t.inputNames);s=await V(t,r)}const r={...e,...s},o=await W(this,r),n=this.sessions["text"===t?"lm_head":"gen_head"];if(!n)throw new Error(`Unable to find "${n}" generation head`);const i=await V(n,(0,a.pick)(o,n.inputNames));return{...s,...o,...i}}async generate(e){return this._generation_mode="text",super.generate(e)}async generate_images(e){this._generation_mode="image";const t=(e.inputs??e[this.main_input_name]).dims[1],s=(await super.generate(e)).slice(null,[t,null]),r=this.sessions.image_decode,{decoded_image:o}=await V(r,{generated_tokens:s}),n=o.add_(1).mul_(127.5).clamp_(0,255).to("uint8"),i=[];for(const e of n){const t=p.RawImage.fromTensor(e);i.push(t)}return i}}class ql extends re{constructor({char_logits:e,bpe_logits:t,wp_logits:s}){super(),this.char_logits=e,this.bpe_logits=t,this.wp_logits=s}get logits(){return[this.char_logits,this.bpe_logits,this.wp_logits]}}class Wl extends se{}class Ul extends Wl{async _call(e){return new ql(await super._call(e))}}class Xl extends se{}class Ql extends Xl{}class Hl extends Xl{}class Jl extends se{}class Yl extends Jl{}class Kl extends Jl{}class Zl extends se{forward_params=["input_ids","attention_mask","position_ids","audio_values","past_key_values"]}class ec extends Zl{_merge_input_ids_with_audio_features(e){const t=e.audio_features.dims.at(-1),s=e.audio_features.view(-1,t);return function({audio_token_id:e,inputs_embeds:t,audio_features:s,input_ids:r,attention_mask:o}){return U({modality_token_id:e,inputs_embeds:t,modality_features:s,input_ids:r,attention_mask:o})}({audio_token_id:this.config.ignore_index,...e,audio_features:s})}}class tc extends se{main_input_name="input_values";forward_params=["input_values"]}class sc extends re{constructor({audio_codes:e}){super(),this.audio_codes=e}}class rc extends re{constructor({audio_values:e}){super(),this.audio_values=e}}class oc extends tc{async encode(e){return new sc(await V(this.sessions.encoder_model,e))}async decode(e){return new rc(await V(this.sessions.decoder_model,e))}}class nc extends tc{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"encoder_model"})}}class ic extends tc{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"decoder_model"})}}class ac extends se{main_input_name="input_values";forward_params=["input_values"]}class lc extends re{constructor({audio_codes:e}){super(),this.audio_codes=e}}class cc extends re{constructor({audio_values:e}){super(),this.audio_values=e}}class dc extends ac{async encode(e){return new lc(await V(this.sessions.encoder_model,e))}async decode(e){return new cc(await V(this.sessions.decoder_model,e))}}class uc extends ac{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"encoder_model"})}}class _c extends ac{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"decoder_model"})}}class pc extends se{main_input_name="input_values";forward_params=["input_values"]}class mc extends pc{async encode(e){return await V(this.sessions.encoder_model,e)}async decode(e){return await V(this.sessions.decoder_model,e)}}class hc extends pc{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"encoder_model"})}}class gc extends pc{static async from_pretrained(e,t={}){return super.from_pretrained(e,{...t,model_file_name:t.model_file_name??"decoder_model"})}}class fc{static MODEL_CLASS_MAPPINGS=null;static BASE_IF_FAIL=!1;static async from_pretrained(e,{progress_callback:t=null,config:s=null,cache_dir:o=null,local_files_only:n=!1,revision:i="main",model_file_name:a=null,subfolder:l="onnx",device:c=null,dtype:d=null,use_external_data_format:u=null,session_options:_={}}={}){const p={progress_callback:t,config:s,cache_dir:o,local_files_only:n,revision:i,model_file_name:a,subfolder:l,device:c,dtype:d,use_external_data_format:u,session_options:_};if(p.config=await r.AutoConfig.from_pretrained(e,p),!this.MODEL_CLASS_MAPPINGS)throw new Error("`MODEL_CLASS_MAPPINGS` not implemented for this type of `AutoClass`: "+this.name);const m=p.config.model_type;for(const t of this.MODEL_CLASS_MAPPINGS){let s=t.get(m);if(!s){for(const e of t.values())if(e[0]===m){s=e;break}if(!s)continue}return await s[1].from_pretrained(e,p)}if(this.BASE_IF_FAIL)return sd.has(m)||console.warn(`Unknown model class "${m}", attempting to construct from base class.`),await se.from_pretrained(e,p);throw Error(`Unsupported model type: ${m}`)}}const wc=new Map([["bert",["BertModel",ie]],["modernbert",["ModernBertModel",_e]],["nomic_bert",["NomicBertModel",fe]],["roformer",["RoFormerModel",Me]],["electra",["ElectraModel",Ee]],["esm",["EsmModel",at]],["convbert",["ConvBertModel",Te]],["camembert",["CamembertModel",De]],["deberta",["DebertaModel",$e]],["deberta-v2",["DebertaV2Model",He]],["mpnet",["MPNetModel",ft]],["albert",["AlbertModel",Ct]],["distilbert",["DistilBertModel",tt]],["roberta",["RobertaModel",rs]],["xlm",["XLMModel",cs]],["xlm-roberta",["XLMRobertaModel",hs]],["clap",["ClapModel",nl]],["clip",["CLIPModel",Ws]],["clipseg",["CLIPSegModel",ar]],["chinese_clip",["ChineseCLIPModel",tr]],["siglip",["SiglipModel",Ys]],["jina_clip",["JinaCLIPModel",rr]],["mobilebert",["MobileBertModel",_t]],["squeezebert",["SqueezeBertModel",yt]],["wav2vec2",["Wav2Vec2Model",la]],["wav2vec2-bert",["Wav2Vec2BertModel",Fa]],["unispeech",["UniSpeechModel",wa]],["unispeech-sat",["UniSpeechSatModel",ka]],["hubert",["HubertModel",Ea]],["wavlm",["WavLMModel",ja]],["audio-spectrogram-transformer",["ASTModel",bs]],["vits",["VitsModel",cl]],["pyannote",["PyAnnoteModel",pa]],["wespeaker-resnet",["WeSpeakerResNetModel",ga]],["detr",["DetrModel",Tn]],["rt_detr",["RTDetrModel",En]],["rt_detr_v2",["RTDetrV2Model",jn]],["rf_detr",["RFDetrModel",Nn]],["d_fine",["DFineModel",$n]],["table-transformer",["TableTransformerModel",Un]],["vit",["ViTModel",Bo]],["ijepa",["IJepaModel",$o]],["pvt",["PvtModel",Qo]],["vit_msn",["ViTMSNModel",Zo]],["vit_mae",["ViTMAEModel",Yo]],["groupvit",["GroupViTModel",sn]],["fastvit",["FastViTModel",on]],["mobilevit",["MobileViTModel",dn]],["mobilevitv2",["MobileViTV2Model",pn]],["owlvit",["OwlViTModel",gn]],["owlv2",["Owlv2Model",Mn]],["beit",["BeitModel",kn]],["deit",["DeiTModel",Jn]],["hiera",["HieraModel",Zn]],["convnext",["ConvNextModel",ji]],["convnextv2",["ConvNextV2Model",Vi]],["dinov2",["Dinov2Model",Ri]],["dinov2_with_registers",["Dinov2WithRegistersModel",qi]],["resnet",["ResNetModel",si]],["swin",["SwinModel",ni]],["swin2sr",["Swin2SRModel",ci]],["donut-swin",["DonutSwinModel",Ii]],["yolos",["YolosModel",Hi]],["dpt",["DPTModel",_i]],["glpn",["GLPNModel",Ai]],["hifigan",["SpeechT5HifiGan",Ua]],["efficientnet",["EfficientNetModel",wl]],["decision_transformer",["DecisionTransformerModel",Rl]],["patchtst",["PatchTSTForPrediction",Ql]],["patchtsmixer",["PatchTSMixerForPrediction",Yl]],["mobilenet_v1",["MobileNetV1Model",Tl]],["mobilenet_v2",["MobileNetV2Model",Sl]],["mobilenet_v3",["MobileNetV3Model",Il]],["mobilenet_v4",["MobileNetV4Model",Dl]],["maskformer",["MaskFormerModel",Fi]],["mgp-str",["MgpstrForSceneTextRecognition",Ul]],["style_text_to_speech_2",["StyleTextToSpeech2Model",Ra]]]),Mc=new Map([["t5",["T5Model",It]],["longt5",["LongT5Model",Ot]],["mt5",["MT5Model",Nt]],["bart",["BartModel",Gt]],["mbart",["MBartModel",Ut]],["marian",["MarianModel",sa]],["whisper",["WhisperModel",vs]],["m2m_100",["M2M100Model",na]],["blenderbot",["BlenderbotModel",Yt]],["blenderbot-small",["BlenderbotSmallModel",es]]]),xc=new Map([["mimi",["MimiModel",oc]],["dac",["DacModel",dc]],["snac",["SnacModel",mc]]]),bc=new Map([["bloom",["BloomModel",Eo]],["jais",["JAISModel",pr]],["gpt2",["GPT2Model",dr]],["gptj",["GPTJModel",kr]],["gpt_bigcode",["GPTBigCodeModel",Tr]],["gpt_neo",["GPTNeoModel",gr]],["gpt_neox",["GPTNeoXModel",Mr]],["codegen",["CodeGenModel",Cr]],["llama",["LlamaModel",Er]],["exaone",["ExaoneModel",Br]],["olmo",["OlmoModel",Ur]],["olmo2",["Olmo2Model",Hr]],["mobilellm",["MobileLLMModel",$r]],["granite",["GraniteModel",Kr]],["cohere",["CohereModel",to]],["gemma",["GemmaModel",oo]],["gemma2",["Gemma2Model",ao]],["gemma3_text",["Gemma3Model",uo]],["helium",["HeliumModel",zr]],["glm",["GlmModel",Dr]],["openelm",["OpenELMModel",mo]],["qwen2",["Qwen2Model",fo]],["qwen3",["Qwen3Model",xo]],["phi",["PhiModel",To]],["phi3",["Phi3Model",Co]],["mpt",["MptModel",zo]],["opt",["OPTModel",Do]],["mistral",["MistralModel",Ja]],["starcoder2",["Starcoder2Model",Za]],["falcon",["FalconModel",sl]],["stablelm",["StableLmModel",hl]]]),kc=new Map([["speecht5",["SpeechT5ForSpeechToText",qa]],["whisper",["WhisperForConditionalGeneration",Ts]],["lite-whisper",["LiteWhisperForConditionalGeneration",Ps]],["moonshine",["MoonshineForConditionalGeneration",Ss]]]),yc=new Map([["speecht5",["SpeechT5ForTextToSpeech",Wa]]]),vc=new Map([["vits",["VitsModel",cl]],["musicgen",["MusicgenForConditionalGeneration",yl]]]),Tc=new Map([["bert",["BertForSequenceClassification",le]],["modernbert",["ModernBertForSequenceClassification",me]],["roformer",["RoFormerForSequenceClassification",be]],["electra",["ElectraForSequenceClassification",Ie]],["esm",["EsmForSequenceClassification",ct]],["convbert",["ConvBertForSequenceClassification",Fe]],["camembert",["CamembertForSequenceClassification",Ne]],["deberta",["DebertaForSequenceClassification",We]],["deberta-v2",["DebertaV2ForSequenceClassification",Ye]],["mpnet",["MPNetForSequenceClassification",Mt]],["albert",["AlbertForSequenceClassification",St]],["distilbert",["DistilBertForSequenceClassification",st]],["roberta",["RobertaForSequenceClassification",ns]],["xlm",["XLMForSequenceClassification",us]],["xlm-roberta",["XLMRobertaForSequenceClassification",fs]],["bart",["BartForSequenceClassification",qt]],["mbart",["MBartForSequenceClassification",Qt]],["mobilebert",["MobileBertForSequenceClassification",mt]],["squeezebert",["SqueezeBertForSequenceClassification",Tt]]]),Pc=new Map([["bert",["BertForTokenClassification",ce]],["modernbert",["ModernBertForTokenClassification",he]],["roformer",["RoFormerForTokenClassification",ke]],["electra",["ElectraForTokenClassification",ze]],["esm",["EsmForTokenClassification",dt]],["convbert",["ConvBertForTokenClassification",Ce]],["camembert",["CamembertForTokenClassification",Be]],["deberta",["DebertaForTokenClassification",Ue]],["deberta-v2",["DebertaV2ForTokenClassification",Ke]],["mpnet",["MPNetForTokenClassification",xt]],["distilbert",["DistilBertForTokenClassification",rt]],["roberta",["RobertaForTokenClassification",is]],["xlm",["XLMForTokenClassification",_s]],["xlm-roberta",["XLMRobertaForTokenClassification",ws]]]),Fc=new Map([["t5",["T5ForConditionalGeneration",zt]],["longt5",["LongT5ForConditionalGeneration",Dt]],["mt5",["MT5ForConditionalGeneration",Bt]],["bart",["BartForConditionalGeneration",$t]],["mbart",["MBartForConditionalGeneration",Xt]],["marian",["MarianMTModel",ra]],["m2m_100",["M2M100ForConditionalGeneration",ia]],["blenderbot",["BlenderbotForConditionalGeneration",Kt]],["blenderbot-small",["BlenderbotSmallForConditionalGeneration",ts]]]),Cc=new Map([["bloom",["BloomForCausalLM",Lo]],["gpt2",["GPT2LMHeadModel",ur]],["jais",["JAISLMHeadModel",mr]],["gptj",["GPTJForCausalLM",yr]],["gpt_bigcode",["GPTBigCodeForCausalLM",Pr]],["gpt_neo",["GPTNeoForCausalLM",fr]],["gpt_neox",["GPTNeoXForCausalLM",xr]],["codegen",["CodeGenForCausalLM",Sr]],["llama",["LlamaForCausalLM",Lr]],["exaone",["ExaoneForCausalLM",Rr]],["olmo",["OlmoForCausalLM",Xr]],["olmo2",["Olmo2ForCausalLM",Jr]],["mobilellm",["MobileLLMForCausalLM",qr]],["granite",["GraniteForCausalLM",Zr]],["cohere",["CohereForCausalLM",so]],["gemma",["GemmaForCausalLM",no]],["gemma2",["Gemma2ForCausalLM",lo]],["gemma3_text",["Gemma3ForCausalLM",_o]],["helium",["HeliumForCausalLM",jr]],["glm",["GlmForCausalLM",Vr]],["openelm",["OpenELMForCausalLM",ho]],["qwen2",["Qwen2ForCausalLM",wo]],["qwen3",["Qwen3ForCausalLM",bo]],["phi",["PhiForCausalLM",Po]],["phi3",["Phi3ForCausalLM",So]],["mpt",["MptForCausalLM",jo]],["opt",["OPTForCausalLM",Vo]],["mbart",["MBartForCausalLM",Ht]],["mistral",["MistralForCausalLM",Ya]],["starcoder2",["Starcoder2ForCausalLM",el]],["falcon",["FalconForCausalLM",rl]],["trocr",["TrOCRForCausalLM",Qa]],["stablelm",["StableLmForCausalLM",gl]],["phi3_v",["Phi3VForCausalLM",$s]]]),Sc=new Map([["multi_modality",["MultiModalityCausalLM",$l]]]),Ac=new Map([["bert",["BertForMaskedLM",ae]],["modernbert",["ModernBertForMaskedLM",pe]],["roformer",["RoFormerForMaskedLM",xe]],["electra",["ElectraForMaskedLM",Le]],["esm",["EsmForMaskedLM",lt]],["convbert",["ConvBertForMaskedLM",Pe]],["camembert",["CamembertForMaskedLM",Ve]],["deberta",["DebertaForMaskedLM",qe]],["deberta-v2",["DebertaV2ForMaskedLM",Je]],["mpnet",["MPNetForMaskedLM",wt]],["albert",["AlbertForMaskedLM",Et]],["distilbert",["DistilBertForMaskedLM",nt]],["roberta",["RobertaForMaskedLM",os]],["xlm",["XLMWithLMHeadModel",ds]],["xlm-roberta",["XLMRobertaForMaskedLM",gs]],["mobilebert",["MobileBertForMaskedLM",pt]],["squeezebert",["SqueezeBertForMaskedLM",vt]]]),Ec=new Map([["bert",["BertForQuestionAnswering",de]],["roformer",["RoFormerForQuestionAnswering",ye]],["electra",["ElectraForQuestionAnswering",je]],["convbert",["ConvBertForQuestionAnswering",Se]],["camembert",["CamembertForQuestionAnswering",Re]],["deberta",["DebertaForQuestionAnswering",Xe]],["deberta-v2",["DebertaV2ForQuestionAnswering",Ze]],["mpnet",["MPNetForQuestionAnswering",bt]],["albert",["AlbertForQuestionAnswering",At]],["distilbert",["DistilBertForQuestionAnswering",ot]],["roberta",["RobertaForQuestionAnswering",as]],["xlm",["XLMForQuestionAnswering",ps]],["xlm-roberta",["XLMRobertaForQuestionAnswering",Ms]],["mobilebert",["MobileBertForQuestionAnswering",ht]],["squeezebert",["SqueezeBertForQuestionAnswering",Pt]]]),Lc=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",As]],["idefics3",["Idefics3ForConditionalGeneration",Bs]],["smolvlm",["SmolVLMForConditionalGeneration",Rs]]]),Ic=new Map([["llava",["LlavaForConditionalGeneration",Ls]],["llava_onevision",["LlavaOnevisionForConditionalGeneration",Is]],["moondream1",["Moondream1ForConditionalGeneration",zs]],["florence2",["Florence2ForConditionalGeneration",Os]],["qwen2-vl",["Qwen2VLForConditionalGeneration",yo]],["idefics3",["Idefics3ForConditionalGeneration",Bs]],["smolvlm",["SmolVLMForConditionalGeneration",Rs]],["paligemma",["PaliGemmaForConditionalGeneration",Vs]]]),zc=new Map([["ultravox",["UltravoxModel",ec]]]),jc=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",As]]]),Oc=new Map([["vit",["ViTForImageClassification",Ro]],["ijepa",["IJepaForImageClassification",qo]],["pvt",["PvtForImageClassification",Ho]],["vit_msn",["ViTMSNForImageClassification",en]],["fastvit",["FastViTForImageClassification",nn]],["mobilevit",["MobileViTForImageClassification",un]],["mobilevitv2",["MobileViTV2ForImageClassification",mn]],["beit",["BeitForImageClassification",yn]],["deit",["DeiTForImageClassification",Yn]],["hiera",["HieraForImageClassification",ei]],["convnext",["ConvNextForImageClassification",Oi]],["convnextv2",["ConvNextV2ForImageClassification",Ni]],["dinov2",["Dinov2ForImageClassification",Gi]],["dinov2_with_registers",["Dinov2WithRegistersForImageClassification",Wi]],["resnet",["ResNetForImageClassification",ri]],["swin",["SwinForImageClassification",ii]],["segformer",["SegformerForImageClassification",_l]],["efficientnet",["EfficientNetForImageClassification",Ml]],["mobilenet_v1",["MobileNetV1ForImageClassification",Pl]],["mobilenet_v2",["MobileNetV2ForImageClassification",Al]],["mobilenet_v3",["MobileNetV3ForImageClassification",zl]],["mobilenet_v4",["MobileNetV4ForImageClassification",Vl]]]),Dc=new Map([["detr",["DetrForObjectDetection",Pn]],["rt_detr",["RTDetrForObjectDetection",Ln]],["rt_detr_v2",["RTDetrV2ForObjectDetection",On]],["rf_detr",["RFDetrForObjectDetection",Bn]],["d_fine",["DFineForObjectDetection",qn]],["table-transformer",["TableTransformerForObjectDetection",Xn]],["yolos",["YolosForObjectDetection",Ji]]]),Vc=new Map([["owlvit",["OwlViTForObjectDetection",fn]],["owlv2",["Owlv2ForObjectDetection",xn]],["grounding-dino",["GroundingDinoForObjectDetection",Xi]]]),Nc=new Map([["detr",["DetrForSegmentation",Fn]],["clipseg",["CLIPSegForImageSegmentation",lr]]]),Bc=new Map([["segformer",["SegformerForSemanticSegmentation",pl]],["sapiens",["SapiensForSemanticSegmentation",fi]],["swin",["SwinForSemanticSegmentation",ai]],["mobilenet_v1",["MobileNetV1ForSemanticSegmentation",Fl]],["mobilenet_v2",["MobileNetV2ForSemanticSegmentation",El]],["mobilenet_v3",["MobileNetV3ForSemanticSegmentation",jl]],["mobilenet_v4",["MobileNetV4ForSemanticSegmentation",Nl]]]),Rc=new Map([["detr",["DetrForSegmentation",Fn]],["maskformer",["MaskFormerForInstanceSegmentation",Ci]]]),Gc=new Map([["sam",["SamModel",Zi]]]),$c=new Map([["wav2vec2",["Wav2Vec2ForCTC",ca]],["wav2vec2-bert",["Wav2Vec2BertForCTC",Ca]],["unispeech",["UniSpeechForCTC",Ma]],["unispeech-sat",["UniSpeechSatForCTC",ya]],["wavlm",["WavLMForCTC",Oa]],["hubert",["HubertForCTC",La]]]),qc=new Map([["wav2vec2",["Wav2Vec2ForSequenceClassification",da]],["wav2vec2-bert",["Wav2Vec2BertForSequenceClassification",Sa]],["unispeech",["UniSpeechForSequenceClassification",xa]],["unispeech-sat",["UniSpeechSatForSequenceClassification",va]],["wavlm",["WavLMForSequenceClassification",Da]],["hubert",["HubertForSequenceClassification",Ia]],["audio-spectrogram-transformer",["ASTForAudioClassification",ks]]]),Wc=new Map([["wavlm",["WavLMForXVector",Va]]]),Uc=new Map([["unispeech-sat",["UniSpeechSatForAudioFrameClassification",Ta]],["wavlm",["WavLMForAudioFrameClassification",Na]],["wav2vec2",["Wav2Vec2ForAudioFrameClassification",ua]],["pyannote",["PyAnnoteForAudioFrameClassification",ma]]]),Xc=new Map([["vitmatte",["VitMatteForImageMatting",ln]]]),Qc=new Map([["patchtst",["PatchTSTForPrediction",Hl]],["patchtsmixer",["PatchTSMixerForPrediction",Kl]]]),Hc=new Map([["swin2sr",["Swin2SRForImageSuperResolution",di]]]),Jc=new Map([["dpt",["DPTForDepthEstimation",pi]],["depth_anything",["DepthAnythingForDepthEstimation",hi]],["glpn",["GLPNForDepthEstimation",Ei]],["sapiens",["SapiensForDepthEstimation",wi]],["depth_pro",["DepthProForDepthEstimation",bi]],["metric3d",["Metric3DForDepthEstimation",yi]],["metric3dv2",["Metric3Dv2ForDepthEstimation",Ti]]]),Yc=new Map([["sapiens",["SapiensForNormalEstimation",Mi]]]),Kc=new Map([["vitpose",["VitPoseForPoseEstimation",Uo]]]),Zc=new Map([["clip",["CLIPVisionModelWithProjection",Hs]],["siglip",["SiglipVisionModel",Zs]],["jina_clip",["JinaCLIPVisionModel",nr]]]),ed=[[wc,x],[Mc,b],[bc,v],[xc,E],[Tc,x],[Pc,x],[Fc,k],[kc,k],[Cc,v],[Sc,C],[Ac,x],[Ec,x],[Lc,y],[Ic,P],[zc,A],[Oc,x],[Nc,x],[Rc,x],[Bc,x],[Xc,x],[Qc,x],[Hc,x],[Jc,x],[Yc,x],[Kc,x],[Dc,x],[Vc,x],[Gc,T],[$c,x],[qc,x],[yc,k],[vc,x],[Wc,x],[Uc,x],[Zc,x]];for(const[e,t]of ed)for(const[s,r]of e.values())L.set(s,t),z.set(r,s),I.set(s,r);const td=[["MusicgenForConditionalGeneration",yl,F],["Phi3VForCausalLM",$s,S],["CLIPTextModelWithProjection",Xs,x],["SiglipTextModel",Ks,x],["JinaCLIPTextModel",or,x],["ClapTextModelWithProjection",il,x],["ClapAudioModelWithProjection",al,x],["DacEncoderModel",uc,x],["DacDecoderModel",_c,x],["MimiEncoderModel",nc,x],["MimiDecoderModel",ic,x],["SnacEncoderModel",hc,x],["SnacDecoderModel",gc,x]];for(const[e,t,s]of td)L.set(e,s),z.set(t,e),I.set(e,t);const sd=new Map([["modnet",Nc],["birefnet",Nc],["isnet",Nc],["ben",Nc]]);for(const[e,t]of sd.entries())t.set(e,["PreTrainedModel",se]),L.set(e,x),z.set(se,e),I.set(e,se);class rd extends fc{static MODEL_CLASS_MAPPINGS=ed.map((e=>e[0]));static BASE_IF_FAIL=!0}class od extends fc{static MODEL_CLASS_MAPPINGS=[Tc]}class nd extends fc{static MODEL_CLASS_MAPPINGS=[Pc]}class id extends fc{static MODEL_CLASS_MAPPINGS=[Fc]}class ad extends fc{static MODEL_CLASS_MAPPINGS=[kc]}class ld extends fc{static MODEL_CLASS_MAPPINGS=[yc]}class cd extends fc{static MODEL_CLASS_MAPPINGS=[vc]}class dd extends fc{static MODEL_CLASS_MAPPINGS=[Cc]}class ud extends fc{static MODEL_CLASS_MAPPINGS=[Ac]}class _d extends fc{static MODEL_CLASS_MAPPINGS=[Ec]}class pd extends fc{static MODEL_CLASS_MAPPINGS=[Lc]}class md extends fc{static MODEL_CLASS_MAPPINGS=[Oc]}class hd extends fc{static MODEL_CLASS_MAPPINGS=[Nc]}class gd extends fc{static MODEL_CLASS_MAPPINGS=[Bc]}class fd extends fc{static MODEL_CLASS_MAPPINGS=[Rc]}class wd extends fc{static MODEL_CLASS_MAPPINGS=[Dc]}class Md extends fc{static MODEL_CLASS_MAPPINGS=[Vc]}class xd extends fc{static MODEL_CLASS_MAPPINGS=[Gc]}class bd extends fc{static MODEL_CLASS_MAPPINGS=[$c]}class kd extends fc{static MODEL_CLASS_MAPPINGS=[qc]}class yd extends fc{static MODEL_CLASS_MAPPINGS=[Wc]}class vd extends fc{static MODEL_CLASS_MAPPINGS=[Uc]}class Td extends fc{static MODEL_CLASS_MAPPINGS=[jc]}class Pd extends fc{static MODEL_CLASS_MAPPINGS=[Xc]}class Fd extends fc{static MODEL_CLASS_MAPPINGS=[Hc]}class Cd extends fc{static MODEL_CLASS_MAPPINGS=[Jc]}class Sd extends fc{static MODEL_CLASS_MAPPINGS=[Yc]}class Ad extends fc{static MODEL_CLASS_MAPPINGS=[Kc]}class Ed extends fc{static MODEL_CLASS_MAPPINGS=[Zc]}class Ld extends fc{static MODEL_CLASS_MAPPINGS=[Ic]}class Id extends fc{static MODEL_CLASS_MAPPINGS=[zc]}class zd extends re{constructor({logits:e,past_key_values:t,encoder_outputs:s,decoder_attentions:r=null,cross_attentions:o=null}){super(),this.logits=e,this.past_key_values=t,this.encoder_outputs=s,this.decoder_attentions=r,this.cross_attentions=o}}class jd extends re{constructor({logits:e,...t}){super(),this.logits=e;const s=Object.values(t);s.length>0&&(this.attentions=s)}}class Od extends re{constructor({logits:e,embeddings:t}){super(),this.logits=e,this.embeddings=t}}class Dd extends re{constructor({logits:e}){super(),this.logits=e}}class Vd extends re{constructor({logits:e}){super(),this.logits=e}}class Nd extends re{constructor({start_logits:e,end_logits:t}){super(),this.start_logits=e,this.end_logits=t}}class Bd extends re{constructor({logits:e}){super(),this.logits=e}}class Rd extends re{constructor({logits:e,past_key_values:t}){super(),this.logits=e,this.past_key_values=t}}class Gd extends re{constructor({alphas:e}){super(),this.alphas=e}}class $d extends re{constructor({waveform:e,spectrogram:t}){super(),this.waveform=e,this.spectrogram=t}}},"./src/models/audio_spectrogram_transformer/feature_extraction_audio_spectrogram_transformer.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ASTFeatureExtractor:()=>n});var r=s("./src/base/feature_extraction_utils.js"),o=(s("./src/utils/tensor.js"),s("./src/utils/audio.js"));class n extends r.FeatureExtractor{constructor(e){super(e);const t=this.config.sampling_rate,s=(0,o.mel_filter_bank)(257,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);this.mel_filters=s,this.window=(0,o.window_function)(400,"hann",{periodic:!1}),this.mean=this.config.mean,this.std=this.config.std}async _extract_fbank_features(e,t){return(0,o.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,max_num_frames:t,transpose:!0})}async _call(e){(0,r.validate_audio_inputs)(e,"ASTFeatureExtractor");const t=await this._extract_fbank_features(e,this.config.max_length);if(this.config.do_normalize){const e=2*this.std,s=t.data;for(let t=0;t<s.length;++t)s[t]=(s[t]-this.mean)/e}return{input_values:t.unsqueeze_(0)}}}},"./src/models/auto/feature_extraction_auto.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{AutoFeatureExtractor:()=>i});var r=s("./src/utils/constants.js"),o=s("./src/utils/hub.js"),n=(s("./src/base/feature_extraction_utils.js"),s("./src/models/feature_extractors.js"));class i{static async from_pretrained(e,t={}){const s=await(0,o.getModelJSON)(e,r.FEATURE_EXTRACTOR_NAME,!0,t),i=s.feature_extractor_type,a=n[i];if(!a)throw new Error(`Unknown feature_extractor_type: '${i}'. Please report this at ${r.GITHUB_ISSUE_URL}.`);return new a(s)}}},"./src/models/auto/image_processing_auto.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{AutoImageProcessor:()=>a});var r=s("./src/utils/constants.js"),o=s("./src/utils/hub.js"),n=s("./src/base/image_processors_utils.js"),i=s("./src/models/image_processors.js");class a{static async from_pretrained(e,t={}){const s=await(0,o.getModelJSON)(e,r.IMAGE_PROCESSOR_NAME,!0,t),a=s.image_processor_type??s.feature_extractor_type;let l=i[a];return l||(void 0!==a&&console.warn(`Image processor type '${a}' not found, assuming base ImageProcessor. Please report this at ${r.GITHUB_ISSUE_URL}.`),l=n.ImageProcessor),new l(s)}}},"./src/models/auto/processing_auto.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{AutoProcessor:()=>c});var r=s("./src/utils/constants.js"),o=s("./src/utils/hub.js"),n=s("./src/base/processing_utils.js"),i=s("./src/models/processors.js"),a=s("./src/models/image_processors.js"),l=s("./src/models/feature_extractors.js");class c{static async from_pretrained(e,t={}){const s=await(0,o.getModelJSON)(e,r.IMAGE_PROCESSOR_NAME,!0,t),{image_processor_type:c,feature_extractor_type:d,processor_class:u}=s;if(u&&i[u])return i[u].from_pretrained(e,t);if(!c&&!d)throw new Error("No `image_processor_type` or `feature_extractor_type` found in the config.");const _={};if(c){const e=a[c];if(!e)throw new Error(`Unknown image_processor_type: '${c}'.`);_.image_processor=new e(s)}if(d){const e=a[d];if(e)_.image_processor=new e(s);else{const e=l[d];if(!e)throw new Error(`Unknown feature_extractor_type: '${d}'.`);_.feature_extractor=new e(s)}}return new n.Processor({},_)}}},"./src/models/beit/image_processing_beit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{BeitFeatureExtractor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}},"./src/models/bit/image_processing_bit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{BitImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}},"./src/models/chinese_clip/image_processing_chinese_clip.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ChineseCLIPFeatureExtractor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}},"./src/models/clap/feature_extraction_clap.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ClapFeatureExtractor:()=>n});var r=s("./src/base/feature_extraction_utils.js"),o=(s("./src/utils/tensor.js"),s("./src/utils/audio.js"));class n extends r.FeatureExtractor{constructor(e){super(e),this.mel_filters=(0,o.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,null,"htk"),this.mel_filters_slaney=(0,o.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,"slaney","slaney"),this.window=(0,o.window_function)(this.config.fft_window_size,"hann")}async _get_input_mel(e,t,s,r){let o,n=!1;const i=e.length-t;if(i>0){if("rand_trunc"!==s)throw new Error(`Truncation strategy "${s}" not implemented`);{n=!0;const s=Math.floor(Math.random()*(i+1));e=e.subarray(s,s+t),o=await this._extract_fbank_features(e,this.mel_filters_slaney,this.config.nb_max_samples)}}else{if(i<0){let s=new Float64Array(t);if(s.set(e),"repeat"===r)for(let r=e.length;r<t;r+=e.length)s.set(e.subarray(0,Math.min(e.length,t-r)),r);else if("repeatpad"===r)for(let t=e.length;t<-i;t+=e.length)s.set(e,t);e=s}if("fusion"===s)throw new Error(`Truncation strategy "${s}" not implemented`);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,s=null){return(0,o.spectrogram)(e,this.window,this.config.fft_window_size,this.config.hop_length,{power:2,mel_filters:t,log_mel:"dB",max_num_frames:s,do_pad:!1,transpose:!0})}async _call(e,{max_length:t=null}={}){(0,r.validate_audio_inputs)(e,"ClapFeatureExtractor");return{input_features:(await this._get_input_mel(e,t??this.config.nb_max_samples,this.config.truncation,this.config.padding)).unsqueeze_(0)}}}},"./src/models/clip/image_processing_clip.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{CLIPFeatureExtractor:()=>n,CLIPImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}class n extends o{}},"./src/models/convnext/image_processing_convnext.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ConvNextFeatureExtractor:()=>n,ConvNextImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{constructor(e){super(e),this.crop_pct=this.config.crop_pct??.875}async resize(e){const t=this.size?.shortest_edge;if(void 0===t)throw new Error("Size dictionary must contain 'shortest_edge' key.");if(t<384){const s=Math.floor(t/this.crop_pct),[r,o]=this.get_resize_output_image_size(e,{shortest_edge:s});e=await e.resize(r,o,{resample:this.resample}),e=await e.center_crop(t,t)}else e=await e.resize(t,t,{resample:this.resample});return e}}class n extends o{}},"./src/models/dac/feature_extraction_dac.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DacFeatureExtractor:()=>o});var r=s("./src/models/encodec/feature_extraction_encodec.js");class o extends r.EncodecFeatureExtractor{}},"./src/models/deit/image_processing_deit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DeiTFeatureExtractor:()=>n,DeiTImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}class n extends o{}},"./src/models/detr/image_processing_detr.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DetrFeatureExtractor:()=>i,DetrImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js"),o=s("./src/utils/tensor.js");class n extends r.ImageProcessor{async _call(e){const t=await super._call(e),s=[t.pixel_values.dims[0],64,64],r=(0,o.full)(s,1n);return{...t,pixel_mask:r}}post_process_object_detection(...e){return(0,r.post_process_object_detection)(...e)}post_process_panoptic_segmentation(...e){return(0,r.post_process_panoptic_segmentation)(...e)}post_process_instance_segmentation(...e){return(0,r.post_process_instance_segmentation)(...e)}}class i extends n{}},"./src/models/donut/image_processing_donut.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DonutFeatureExtractor:()=>n,DonutImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{pad_image(e,t,s,r={}){const[o,n,i]=t;let a=this.image_mean;Array.isArray(this.image_mean)||(a=new Array(i).fill(a));let l=this.image_std;Array.isArray(l)||(l=new Array(i).fill(a));const c=a.map(((e,t)=>-e/l[t]));return super.pad_image(e,t,s,{center:!0,constant_values:c,...r})}}class n extends o{}},"./src/models/dpt/image_processing_dpt.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DPTFeatureExtractor:()=>n,DPTImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}class n extends o{}},"./src/models/efficientnet/image_processing_efficientnet.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{EfficientNetImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{constructor(e){super(e),this.include_top=this.config.include_top??!0,this.include_top&&(this.image_std=this.image_std.map((e=>e*e)))}}},"./src/models/encodec/feature_extraction_encodec.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{EncodecFeatureExtractor:()=>n});var r=s("./src/base/feature_extraction_utils.js"),o=s("./src/utils/tensor.js");class n extends r.FeatureExtractor{async _call(e){(0,r.validate_audio_inputs)(e,"EncodecFeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));const t=this.config.feature_size;if(e.length%t!=0)throw new Error(`The length of the audio data must be a multiple of the number of channels (${t}).`);const s=[1,t,e.length/t];return{input_values:new o.Tensor("float32",e,s)}}}},"./src/models/feature_extractors.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ASTFeatureExtractor:()=>r.ASTFeatureExtractor,ClapFeatureExtractor:()=>n.ClapFeatureExtractor,DacFeatureExtractor:()=>i.DacFeatureExtractor,EncodecFeatureExtractor:()=>o.EncodecFeatureExtractor,ImageFeatureExtractor:()=>h.ImageProcessor,MoonshineFeatureExtractor:()=>a.MoonshineFeatureExtractor,PyAnnoteFeatureExtractor:()=>l.PyAnnoteFeatureExtractor,SeamlessM4TFeatureExtractor:()=>c.SeamlessM4TFeatureExtractor,SnacFeatureExtractor:()=>d.SnacFeatureExtractor,SpeechT5FeatureExtractor:()=>u.SpeechT5FeatureExtractor,Wav2Vec2FeatureExtractor:()=>_.Wav2Vec2FeatureExtractor,WeSpeakerFeatureExtractor:()=>p.WeSpeakerFeatureExtractor,WhisperFeatureExtractor:()=>m.WhisperFeatureExtractor});var r=s("./src/models/audio_spectrogram_transformer/feature_extraction_audio_spectrogram_transformer.js"),o=s("./src/models/encodec/feature_extraction_encodec.js"),n=s("./src/models/clap/feature_extraction_clap.js"),i=s("./src/models/dac/feature_extraction_dac.js"),a=s("./src/models/moonshine/feature_extraction_moonshine.js"),l=s("./src/models/pyannote/feature_extraction_pyannote.js"),c=s("./src/models/seamless_m4t/feature_extraction_seamless_m4t.js"),d=s("./src/models/snac/feature_extraction_snac.js"),u=s("./src/models/speecht5/feature_extraction_speecht5.js"),_=s("./src/models/wav2vec2/feature_extraction_wav2vec2.js"),p=s("./src/models/wespeaker/feature_extraction_wespeaker.js"),m=s("./src/models/whisper/feature_extraction_whisper.js"),h=s("./src/base/image_processors_utils.js")},"./src/models/florence2/processing_florence2.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Florence2Processor:()=>i});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js");class i extends r.Processor{static tokenizer_class=n.AutoTokenizer;static image_processor_class=o.AutoImageProcessor;constructor(e,t){super(e,t);const{tasks_answer_post_processing_type:s,task_prompts_without_inputs:r,task_prompts_with_input:o}=this.image_processor.config;this.tasks_answer_post_processing_type=new Map(Object.entries(s??{})),this.task_prompts_without_inputs=new Map(Object.entries(r??{})),this.task_prompts_with_input=new Map(Object.entries(o??{})),this.regexes={quad_boxes:/(.+?)<loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)>/gm,bboxes:/([^<]+)?<loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)>/gm},this.size_per_bin=1e3}construct_prompts(e){"string"==typeof e&&(e=[e]);const t=[];for(const s of e)if(this.task_prompts_without_inputs.has(s))t.push(this.task_prompts_without_inputs.get(s));else{for(const[e,r]of this.task_prompts_with_input)if(s.includes(e)){t.push(r.replaceAll("{input}",s).replaceAll(e,""));break}t.length!==e.length&&t.push(s)}return t}post_process_generation(e,t,s){const r=this.tasks_answer_post_processing_type.get(t)??"pure_text";let o;switch(e=e.replaceAll("<s>","").replaceAll("</s>",""),r){case"pure_text":o=e;break;case"description_with_bboxes":case"bboxes":case"phrase_grounding":case"ocr":const n="ocr"===r?"quad_boxes":"bboxes",i=e.matchAll(this.regexes[n]),a=[],l=[];for(const[e,t,...r]of i)a.push(t?t.trim():a.at(-1)??""),l.push(r.map(((e,t)=>(Number(e)+.5)/this.size_per_bin*s[t%2])));o={labels:a,[n]:l};break;default:throw new Error(`Task "${t}" (of type "${r}") not yet implemented.`)}return{[t]:o}}async _call(e,t=null,s={}){if(!e&&!t)throw new Error("Either text or images must be provided");return{...await this.image_processor(e,s),...t?this.tokenizer(t,s):{}}}}},"./src/models/glpn/image_processing_glpn.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{GLPNFeatureExtractor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}},"./src/models/grounding_dino/image_processing_grounding_dino.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{GroundingDinoImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js"),o=s("./src/utils/tensor.js");class n extends r.ImageProcessor{async _call(e){const t=await super._call(e),s=t.pixel_values.dims,r=(0,o.ones)([s[0],s[2],s[3]]);return{...t,pixel_mask:r}}}},"./src/models/grounding_dino/processing_grounding_dino.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{GroundingDinoProcessor:()=>l});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js"),i=s("./src/base/image_processors_utils.js");function a(e,t){const s=e.dims.at(-1)-1,r=e.tolist();r.fill(!1,0,1),r.fill(!1,s);const o=t.tolist();return r.map(((e,t)=>e?t:null)).filter((e=>null!==e)).map((e=>o[e]))}class l extends r.Processor{static tokenizer_class=n.AutoTokenizer;static image_processor_class=o.AutoImageProcessor;async _call(e,t,s={}){const r=e?await this.image_processor(e,s):{};return{...t?this.tokenizer(t,s):{},...r}}post_process_grounded_object_detection(e,t,{box_threshold:s=.25,text_threshold:r=.25,target_sizes:o=null}={}){const{logits:n,pred_boxes:l}=e,c=n.dims[0];if(null!==o&&o.length!==c)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");const d=n.dims.at(1),u=n.sigmoid(),_=u.max(-1).tolist(),p=l.tolist().map((e=>e.map((e=>(0,i.center_to_corners_format)(e))))),m=[];for(let e=0;e<c;++e){const n=null!==o?o[e]:null;null!==n&&(p[e]=p[e].map((e=>e.map(((e,t)=>e*n[(t+1)%2])))));const i=_[e],l=[],c=[],h=[];for(let o=0;o<d;++o){const n=i[o];if(n<=s)continue;const d=p[e][o],_=u[e][o];l.push(n),h.push(d);const m=a(_.gt(r),t[e]);c.push(m)}m.push({scores:l,boxes:h,labels:this.batch_decode(c)})}return m}}},"./src/models/idefics3/image_processing_idefics3.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Idefics3ImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js"),o=s("./src/utils/tensor.js");class n extends r.ImageProcessor{constructor(e){super(e),this.do_image_splitting=e.do_image_splitting??!0,this.max_image_size=e.max_image_size}get_resize_for_vision_encoder(e,t){let[s,r]=e.dims.slice(-2);const o=r/s;return r>=s?(r=Math.ceil(r/t)*t,s=Math.floor(r/o),s=Math.ceil(s/t)*t):(s=Math.ceil(s/t)*t,r=Math.floor(s*o),r=Math.ceil(r/t)*t),{height:s,width:r}}async _call(e,{do_image_splitting:t=null,return_row_col_info:s=!1}={}){let r;if(Array.isArray(e)){if(0===e.length||!e[0])throw new Error("No images provided.");r=Array.isArray(e[0])?e:[e]}else r=[[e]];let n=[],i=[],a=[];const l=[],c=[];for(const e of r){let s=await Promise.all(e.map((e=>this.preprocess(e))));l.push(...s.map((e=>e.original_size))),c.push(...s.map((e=>e.reshaped_input_size))),s.forEach((e=>e.pixel_values.unsqueeze_(0)));const{longest_edge:r}=this.max_image_size;let d;if(t??this.do_image_splitting){let e=new Array(s.length),t=new Array(s.length);d=await Promise.all(s.map((async(s,n)=>{const i=this.get_resize_for_vision_encoder(s.pixel_values,r),a=await(0,o.interpolate_4d)(s.pixel_values,{size:[i.height,i.width]}),{frames:l,num_splits_h:c,num_splits_w:d}=await this.split_image(a,this.max_image_size);return e[n]=c,t[n]=d,(0,o.cat)(l,0)}))),i.push(e),a.push(t)}else{const e=[r,r];d=await Promise.all(s.map((t=>(0,o.interpolate_4d)(t.pixel_values,{size:e})))),i.push(new Array(s.length).fill(0)),a.push(new Array(s.length).fill(0))}n.push((0,o.cat)(d,0))}const d=n.length,[u,_,p,m]=n[0].dims;let h,g;if(1===d)h=n[0].unsqueeze_(0),g=(0,o.full)([d,u,p,m],!0);else{const e=Math.max(...n.map((e=>e.dims.at(0))));g=(0,o.full)([d,e,p,m],!0);const t=g.data,s=e*p*m;for(let r=0;r<d;++r){const i=n[r].dims[0];if(i<e){n[r]=(0,o.cat)([n[r],(0,o.full)([e-i,_,p,m],0)],0);const a=r*s+i*p*m,l=(r+1)*s;t.fill(!1,a,l)}}h=(0,o.stack)(n,0)}return{pixel_values:h,pixel_attention_mask:g,original_sizes:l,reshaped_input_sizes:c,...s?{rows:i,cols:a}:{}}}async split_image(e,{longest_edge:t}){const s=t,r=t,n=[],[i,a]=e.dims.slice(-2);let l=0,c=0;if(i>s||a>r){l=Math.ceil(i/s),c=Math.ceil(a/r);const t=Math.ceil(i/l),d=Math.ceil(a/c);for(let s=0;s<l;++s)for(let r=0;r<c;++r){let u,_,p,m;s===l-1?(_=i-t,m=i):(_=s*t,m=(s+1)*t),r===c-1?(u=a-d,p=a):(u=r*d,p=(r+1)*d);const h=[_,u],g=[m,p],f=await(0,o.slice)(e,h,g,[2,3]);n.push(f)}const u=s,_=r;i===u&&a===_||(e=await(0,o.interpolate_4d)(e,{size:[u,_]}))}return n.push(e),{frames:n,num_splits_h:l,num_splits_w:c}}}},"./src/models/idefics3/processing_idefics3.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Idefics3Processor:()=>l});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js"),i=(s("./src/utils/image.js"),s("./src/utils/core.js"));function a(e,t,s,r,o,n){return 0===e&&0===t?function(e,t,s,r){return`${t}${r}`+s.repeat(e)+`${t}`}(s,r,o,n):function(e,t,s,r,o,n){let i="";for(let n=0;n<t;++n){for(let t=0;t<s;++t)i+=r+`<row_${n+1}_col_${t+1}>`+o.repeat(e);i+="\n"}return i+=`\n${r}${n}`+o.repeat(e)+`${r}`,i}(s,e,t,r,o,n)}class l extends r.Processor{static image_processor_class=o.AutoImageProcessor;static tokenizer_class=n.AutoTokenizer;static uses_processor_config=!0;fake_image_token="<fake_token_around_image>";image_token="<image>";global_img_token="<global-img>";async _call(e,t=null,s={}){let r;s.return_row_col_info??=!0,t&&(r=await this.image_processor(t,s)),Array.isArray(e)||(e=[e]);const o=r.rows??[new Array(e.length).fill(0)],n=r.cols??[new Array(e.length).fill(0)],l=this.config.image_seq_len,c=[],d=[];for(let t=0;t<e.length;++t){const s=e[t],r=o[t],u=n[t];c.push((0,i.count)(s,this.image_token));const _=r.map(((e,t)=>a(e,u[t],l,this.fake_image_token,this.image_token,this.global_img_token))),p=s.split(this.image_token);if(0===p.length)throw new Error("The image token should be present in the text.");let m=p[0];for(let e=0;e<_.length;++e)m+=_[e]+p[e+1];d.push(m)}return{...this.tokenizer(d),...r}}}},"./src/models/image_processors.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{BeitFeatureExtractor:()=>r.BeitFeatureExtractor,BitImageProcessor:()=>o.BitImageProcessor,CLIPFeatureExtractor:()=>i.CLIPFeatureExtractor,CLIPImageProcessor:()=>i.CLIPImageProcessor,ChineseCLIPFeatureExtractor:()=>n.ChineseCLIPFeatureExtractor,ConvNextFeatureExtractor:()=>a.ConvNextFeatureExtractor,ConvNextImageProcessor:()=>a.ConvNextImageProcessor,DPTFeatureExtractor:()=>u.DPTFeatureExtractor,DPTImageProcessor:()=>u.DPTImageProcessor,DeiTFeatureExtractor:()=>l.DeiTFeatureExtractor,DeiTImageProcessor:()=>l.DeiTImageProcessor,DetrFeatureExtractor:()=>c.DetrFeatureExtractor,DetrImageProcessor:()=>c.DetrImageProcessor,DonutFeatureExtractor:()=>d.DonutFeatureExtractor,DonutImageProcessor:()=>d.DonutImageProcessor,EfficientNetImageProcessor:()=>_.EfficientNetImageProcessor,GLPNFeatureExtractor:()=>p.GLPNFeatureExtractor,GroundingDinoImageProcessor:()=>m.GroundingDinoImageProcessor,Idefics3ImageProcessor:()=>h.Idefics3ImageProcessor,JinaCLIPImageProcessor:()=>f.JinaCLIPImageProcessor,LlavaOnevisionImageProcessor:()=>w.LlavaOnevisionImageProcessor,Mask2FormerImageProcessor:()=>M.Mask2FormerImageProcessor,MaskFormerFeatureExtractor:()=>x.MaskFormerFeatureExtractor,MaskFormerImageProcessor:()=>x.MaskFormerImageProcessor,MobileNetV1FeatureExtractor:()=>b.MobileNetV1FeatureExtractor,MobileNetV1ImageProcessor:()=>b.MobileNetV1ImageProcessor,MobileNetV2FeatureExtractor:()=>k.MobileNetV2FeatureExtractor,MobileNetV2ImageProcessor:()=>k.MobileNetV2ImageProcessor,MobileNetV3FeatureExtractor:()=>y.MobileNetV3FeatureExtractor,MobileNetV3ImageProcessor:()=>y.MobileNetV3ImageProcessor,MobileNetV4FeatureExtractor:()=>v.MobileNetV4FeatureExtractor,MobileNetV4ImageProcessor:()=>v.MobileNetV4ImageProcessor,MobileViTFeatureExtractor:()=>T.MobileViTFeatureExtractor,MobileViTImageProcessor:()=>T.MobileViTImageProcessor,NougatImageProcessor:()=>P.NougatImageProcessor,OwlViTFeatureExtractor:()=>C.OwlViTFeatureExtractor,OwlViTImageProcessor:()=>C.OwlViTImageProcessor,Owlv2ImageProcessor:()=>F.Owlv2ImageProcessor,Phi3VImageProcessor:()=>S.Phi3VImageProcessor,PvtImageProcessor:()=>A.PvtImageProcessor,Qwen2VLImageProcessor:()=>E.Qwen2VLImageProcessor,RTDetrImageProcessor:()=>L.RTDetrImageProcessor,SamImageProcessor:()=>I.SamImageProcessor,SegformerFeatureExtractor:()=>z.SegformerFeatureExtractor,SegformerImageProcessor:()=>z.SegformerImageProcessor,SiglipImageProcessor:()=>j.SiglipImageProcessor,SmolVLMImageProcessor:()=>O.SmolVLMImageProcessor,Swin2SRImageProcessor:()=>D.Swin2SRImageProcessor,VLMImageProcessor:()=>g.VLMImageProcessor,ViTFeatureExtractor:()=>V.ViTFeatureExtractor,ViTImageProcessor:()=>V.ViTImageProcessor,VitMatteImageProcessor:()=>N.VitMatteImageProcessor,VitPoseImageProcessor:()=>B.VitPoseImageProcessor,YolosFeatureExtractor:()=>R.YolosFeatureExtractor,YolosImageProcessor:()=>R.YolosImageProcessor});var r=s("./src/models/beit/image_processing_beit.js"),o=s("./src/models/bit/image_processing_bit.js"),n=s("./src/models/chinese_clip/image_processing_chinese_clip.js"),i=s("./src/models/clip/image_processing_clip.js"),a=s("./src/models/convnext/image_processing_convnext.js"),l=s("./src/models/deit/image_processing_deit.js"),c=s("./src/models/detr/image_processing_detr.js"),d=s("./src/models/donut/image_processing_donut.js"),u=s("./src/models/dpt/image_processing_dpt.js"),_=s("./src/models/efficientnet/image_processing_efficientnet.js"),p=s("./src/models/glpn/image_processing_glpn.js"),m=s("./src/models/grounding_dino/image_processing_grounding_dino.js"),h=s("./src/models/idefics3/image_processing_idefics3.js"),g=s("./src/models/janus/image_processing_janus.js"),f=s("./src/models/jina_clip/image_processing_jina_clip.js"),w=s("./src/models/llava_onevision/image_processing_llava_onevision.js"),M=s("./src/models/mask2former/image_processing_mask2former.js"),x=s("./src/models/maskformer/image_processing_maskformer.js"),b=s("./src/models/mobilenet_v1/image_processing_mobilenet_v1.js"),k=s("./src/models/mobilenet_v2/image_processing_mobilenet_v2.js"),y=s("./src/models/mobilenet_v3/image_processing_mobilenet_v3.js"),v=s("./src/models/mobilenet_v4/image_processing_mobilenet_v4.js"),T=s("./src/models/mobilevit/image_processing_mobilevit.js"),P=s("./src/models/nougat/image_processing_nougat.js"),F=s("./src/models/owlv2/image_processing_owlv2.js"),C=s("./src/models/owlvit/image_processing_owlvit.js"),S=s("./src/models/phi3_v/image_processing_phi3_v.js"),A=s("./src/models/pvt/image_processing_pvt.js"),E=s("./src/models/qwen2_vl/image_processing_qwen2_vl.js"),L=s("./src/models/rt_detr/image_processing_rt_detr.js"),I=s("./src/models/sam/image_processing_sam.js"),z=s("./src/models/segformer/image_processing_segformer.js"),j=s("./src/models/siglip/image_processing_siglip.js"),O=s("./src/models/smolvlm/image_processing_smolvlm.js"),D=s("./src/models/swin2sr/image_processing_swin2sr.js"),V=s("./src/models/vit/image_processing_vit.js"),N=s("./src/models/vitmatte/image_processing_vitmatte.js"),B=s("./src/models/vitpose/image_processing_vitpose.js"),R=s("./src/models/yolos/image_processing_yolos.js")},"./src/models/janus/image_processing_janus.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{VLMImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{constructor(e){super({do_pad:!0,pad_size:{width:e.image_size,height:e.image_size},...e}),this.constant_values=this.config.background_color.map((e=>e*this.rescale_factor))}pad_image(e,t,s,r){return super.pad_image(e,t,s,{constant_values:this.constant_values,center:!0,...r})}}},"./src/models/janus/processing_janus.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{VLChatProcessor:()=>c});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js"),i=s("./src/utils/core.js"),a=s("./src/utils/tensor.js"),l=s("./src/utils/image.js");class c extends r.Processor{static image_processor_class=o.AutoImageProcessor;static tokenizer_class=n.AutoTokenizer;static uses_processor_config=!0;constructor(e,t){super(e,t),this.image_tag=this.config.image_tag,this.image_start_tag=this.config.image_start_tag,this.image_end_tag=this.config.image_end_tag,this.num_image_tokens=this.config.num_image_tokens}async _call(e,{images:t=null,chat_template:s="default"}={}){t?Array.isArray(t)||(t=[t]):t=await Promise.all(e.filter((e=>e.images)).flatMap((e=>e.images)).map((e=>l.RawImage.read(e))));const r=this.tokenizer,o=e=>r.encode(e,{add_special_tokens:!1}),n=r.apply_chat_template(e,{tokenize:!1,add_generation_prompt:!0,chat_template:s}).split(this.image_tag),c=n.length-1;if(t.length!==c)throw new Error(`Number of images provided (${t.length}) does not match number of "${this.image_tag}" image tags (${c})`);const[d,u,_]=r.model.convert_tokens_to_ids([this.image_tag,this.image_start_tag,this.image_end_tag]);let p=o(n[0]),m=new Array(p.length).fill(!1);for(let e=1;e<n.length;++e){const t=new Array(this.num_image_tokens).fill(d),s=o(n[e]);p=(0,i.mergeArrays)(p,[u],t,[_],s);const r=new Array(this.num_image_tokens).fill(!0);m=(0,i.mergeArrays)(m,[!1],r,[!1],new Array(s.length).fill(!1))}const h=[1,p.length],g={input_ids:new a.Tensor("int64",p,h),attention_mask:new a.Tensor("int64",new Array(p.length).fill(1),h),images_seq_mask:new a.Tensor("bool",m,h),images_emb_mask:new a.Tensor("bool",new Array(c*this.num_image_tokens).fill(!0),[1,c,this.num_image_tokens])};if(t&&t.length>0){const e=await this.image_processor(t);return e.pixel_values.unsqueeze_(0),{...g,...e}}return g}}},"./src/models/jina_clip/image_processing_jina_clip.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{JinaCLIPImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{constructor(e){const{resize_mode:t,fill_color:s,interpolation:r,size:o,...n}=e;super({...n,size:"squash"===t?{width:o,height:o}:"shortest"===t?{shortest_edge:o}:{longest_edge:o},resample:"bicubic"===r?3:2,do_center_crop:!0,crop_size:o,do_normalize:!0})}}},"./src/models/jina_clip/processing_jina_clip.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{JinaCLIPProcessor:()=>i});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js");class i extends r.Processor{static tokenizer_class=n.AutoTokenizer;static image_processor_class=o.AutoImageProcessor;async _call(e=null,t=null,s={}){if(!e&&!t)throw new Error("Either text or images must be provided");return{...e?this.tokenizer(e,s):{},...t?await this.image_processor(t,s):{}}}}},"./src/models/llava_onevision/image_processing_llava_onevision.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{LlavaOnevisionImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}},"./src/models/mask2former/image_processing_mask2former.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Mask2FormerImageProcessor:()=>o});var r=s("./src/models/maskformer/image_processing_maskformer.js");class o extends r.MaskFormerImageProcessor{}},"./src/models/maskformer/image_processing_maskformer.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MaskFormerFeatureExtractor:()=>n,MaskFormerImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{post_process_panoptic_segmentation(...e){return(0,r.post_process_panoptic_segmentation)(...e)}post_process_instance_segmentation(...e){return(0,r.post_process_instance_segmentation)(...e)}}class n extends o{}},"./src/models/mgp_str/processing_mgp_str.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MgpstrProcessor:()=>l});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js"),i=s("./src/utils/maths.js");const a={char:["char_decode",1],bpe:["bpe_decode",2],wp:["wp_decode",102]};class l extends r.Processor{static tokenizer_class=n.AutoTokenizer;static image_processor_class=o.AutoImageProcessor;get char_tokenizer(){return this.components.char_tokenizer}get bpe_tokenizer(){return this.components.bpe_tokenizer}get wp_tokenizer(){return this.components.wp_tokenizer}_decode_helper(e,t){if(!a.hasOwnProperty(t))throw new Error(`Format ${t} is not supported.`);const[s,r]=a[t],o=this[s].bind(this),[n,l]=e.dims,c=[],d=[],u=e.tolist();for(let e=0;e<n;++e){const t=u[e],s=[],o=[];for(let e=1;e<l;++e){const[n,a]=(0,i.max)((0,i.softmax)(t[e]));if(o.push(n),a==r)break;s.push(a)}const n=o.length>0?o.reduce(((e,t)=>e*t),1):0;d.push(s),c.push(n)}return[o(d),c]}char_decode(e){return this.char_tokenizer.batch_decode(e).map((e=>e.replaceAll(" ","")))}bpe_decode(e){return this.bpe_tokenizer.batch_decode(e)}wp_decode(e){return this.wp_tokenizer.batch_decode(e).map((e=>e.replaceAll(" ","")))}batch_decode([e,t,s]){const[r,o]=this._decode_helper(e,"char"),[n,a]=this._decode_helper(t,"bpe"),[l,c]=this._decode_helper(s,"wp"),d=[],u=[];for(let e=0;e<r.length;++e){const[t,s]=(0,i.max)([o[e],a[e],c[e]]);d.push([r[e],n[e],l[e]][s]),u.push(t)}return{generated_text:d,scores:u,char_preds:r,bpe_preds:n,wp_preds:l}}static async from_pretrained(...e){const t=await super.from_pretrained(...e),s=await n.AutoTokenizer.from_pretrained("Xenova/gpt2"),r=await n.AutoTokenizer.from_pretrained("Xenova/bert-base-uncased");return t.components={image_processor:t.image_processor,char_tokenizer:t.tokenizer,bpe_tokenizer:s,wp_tokenizer:r},t}async _call(e,t=null){const s=await this.image_processor(e);return t&&(s.labels=this.tokenizer(t).input_ids),s}}},"./src/models/mobilenet_v1/image_processing_mobilenet_v1.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MobileNetV1FeatureExtractor:()=>n,MobileNetV1ImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}class n extends o{}},"./src/models/mobilenet_v2/image_processing_mobilenet_v2.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MobileNetV2FeatureExtractor:()=>n,MobileNetV2ImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}class n extends o{}},"./src/models/mobilenet_v3/image_processing_mobilenet_v3.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MobileNetV3FeatureExtractor:()=>n,MobileNetV3ImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}class n extends o{}},"./src/models/mobilenet_v4/image_processing_mobilenet_v4.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MobileNetV4FeatureExtractor:()=>n,MobileNetV4ImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}class n extends o{}},"./src/models/mobilevit/image_processing_mobilevit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MobileViTFeatureExtractor:()=>n,MobileViTImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}class n extends o{}},"./src/models/moonshine/feature_extraction_moonshine.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MoonshineFeatureExtractor:()=>n});var r=s("./src/base/feature_extraction_utils.js"),o=s("./src/utils/tensor.js");class n extends r.FeatureExtractor{async _call(e){(0,r.validate_audio_inputs)(e,"MoonshineFeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));const t=[1,e.length];return{input_values:new o.Tensor("float32",e,t)}}}},"./src/models/moonshine/processing_moonshine.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MoonshineProcessor:()=>i});var r=s("./src/models/auto/feature_extraction_auto.js"),o=s("./src/tokenizers.js"),n=s("./src/base/processing_utils.js");class i extends n.Processor{static tokenizer_class=o.AutoTokenizer;static feature_extractor_class=r.AutoFeatureExtractor;async _call(e){return await this.feature_extractor(e)}}},"./src/models/nougat/image_processing_nougat.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{NougatImageProcessor:()=>o});var r=s("./src/models/donut/image_processing_donut.js");class o extends r.DonutImageProcessor{}},"./src/models/owlv2/image_processing_owlv2.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Owlv2ImageProcessor:()=>o});var r=s("./src/models/owlvit/image_processing_owlvit.js");class o extends r.OwlViTImageProcessor{}},"./src/models/owlvit/image_processing_owlvit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{OwlViTFeatureExtractor:()=>n,OwlViTImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{post_process_object_detection(...e){return(0,r.post_process_object_detection)(...e)}}class n extends o{}},"./src/models/owlvit/processing_owlvit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{OwlViTProcessor:()=>i});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js");class i extends r.Processor{static tokenizer_class=n.AutoTokenizer;static image_processor_class=o.AutoImageProcessor}},"./src/models/paligemma/processing_paligemma.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{PaliGemmaProcessor:()=>a});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js");const i="<image>";class a extends r.Processor{static tokenizer_class=n.AutoTokenizer;static image_processor_class=o.AutoImageProcessor;static uses_processor_config=!1;async _call(e,t=null,s={}){t||(console.warn("You are using PaliGemma without a text prefix. It will perform as a picture-captioning model."),t=""),Array.isArray(e)||(e=[e]),Array.isArray(t)||(t=[t]);const r=this.tokenizer.bos_token,o=this.image_processor.config.image_seq_length;let n;t.some((e=>e.includes(i)))?n=t.map((e=>{const t=e.replaceAll(i,i.repeat(o)),s=t.lastIndexOf(i),n=-1===s?0:s+7;return t.slice(0,n)+r+t.slice(n)+"\n"})):(console.warn("You are passing both `text` and `images` to `PaliGemmaProcessor`. The processor expects special image tokens in the text, as many tokens as there are images per each text. It is recommended to add `<image>` tokens in the very beginning of your text. For this call, we will infer how many images each text has and add special tokens."),n=t.map((t=>function(e,t,s,r,o){return`${r.repeat(s*o)}${t}${e}\n`}(t,r,o,i,e.length))));const a=this.tokenizer(n,s);return{...await this.image_processor(e,s),...a}}}},"./src/models/phi3_v/image_processing_phi3_v.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Phi3VImageProcessor:()=>d});var r=s("./src/base/image_processors_utils.js"),o=s("./src/utils/tensor.js");const n=336,i=[2,3],{ceil:a,floor:l,sqrt:c}=Math;class d extends r.ImageProcessor{constructor(e){super({...e,do_normalize:!0,do_pad:!0,pad_size:"custom",do_convert_rgb:!0,do_resize:!0}),this._num_crops=e.num_crops}calc_num_image_tokens_from_image_size(e,t){const{num_img_tokens:s}=this.config;return l((l(t/n)*l(e/n)+1)*s+1+(l(t/n)+1)*c(s))}get_resize_output_image_size(e,t){const s=this._num_crops,[r,o]=e.size;let n=r/o,i=1;for(;i*Math.ceil(i/n)<=s;)i+=1;i-=1;const a=Math.floor(336*i);return[a,Math.floor(a/n)]}pad_image(e,t,s,r={}){const[o,i]=t,l=n*a(o/n),c=n*a(i/n),d=[1,1,1].map(((e,t)=>(e-this.image_mean[t])/this.image_std[t]));return super.pad_image(e,t,{width:c,height:l},{center:!0,constant_values:d,...r})}async _call(e,{num_crops:t=null}={}){if(this._num_crops=t??=this.config.num_crops,t<4||c(t)%1!=0)throw new Error("num_crops must be a square number >= 4");Array.isArray(e)||(e=[e]);const s=e.length,r=await Promise.all(e.map((e=>this.preprocess(e)))),d=r.map((e=>e.original_size)),u=r.map((e=>e.reshaped_input_size)),_=[];for(const{pixel_values:e}of r){e.unsqueeze_(0);const[s,r]=e.dims.slice(-2),a=await(0,o.interpolate_4d)(e,{size:[n,n],mode:"bicubic"});if(t>0){const d=[],u=c(t),p=l(r/u),m=l(s/u);for(let t=0;t<u;++t)for(let n=0;n<u;++n){let a,l,c,_;t===u-1?(l=s-m,_=s):(l=t*m,_=(t+1)*m),n===u-1?(a=r-p,c=r):(a=n*p,c=(n+1)*p);const h=[l,a],g=[_,c],f=await(0,o.slice)(e,h,g,i);d.push(f)}const h=await(0,o.interpolate_4d)((0,o.cat)(d,0),{size:[n,n],mode:"bicubic"});_.push((0,o.cat)([a,h],0))}else _.push(a)}const p=(0,o.stack)(_,0),m=u.map((e=>e.map((e=>n*a(e/n)))));return{pixel_values:p,original_sizes:d,reshaped_input_sizes:u,image_sizes:new o.Tensor("int64",m.flat(),[s,2]),num_img_tokens:m.map((([e,t])=>this.calc_num_image_tokens_from_image_size(t,e)))}}}},"./src/models/phi3_v/processing_phi3_v.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Phi3VProcessor:()=>l});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js");s("./src/utils/image.js");const i="<|image|>",a=/<\|image_\d+\|>/g;class l extends r.Processor{static image_processor_class=o.AutoImageProcessor;static tokenizer_class=n.AutoTokenizer;async _call(e,t=null,{padding:s=!0,truncation:r=!0,num_crops:o=null}={}){let n,l;if(Array.isArray(e)||(e=[e]),t){l=await this.image_processor(t,{num_crops:o});const{num_img_tokens:c}=l,d=e.map(((e,t)=>e.split(a).join(i.repeat(c[t]))));n=this.tokenizer(d,{padding:s,truncation:r});const u=this.tokenizer.model.convert_tokens_to_ids([i])[0];n.input_ids.map_((e=>e==u?-e:e))}else n=this.tokenizer(e);return{...n,...l}}}},"./src/models/processors.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Florence2Processor:()=>r.Florence2Processor,GroundingDinoProcessor:()=>o.GroundingDinoProcessor,Idefics3Processor:()=>n.Idefics3Processor,JinaCLIPProcessor:()=>a.JinaCLIPProcessor,MgpstrProcessor:()=>l.MgpstrProcessor,MoonshineProcessor:()=>c.MoonshineProcessor,OwlViTProcessor:()=>d.OwlViTProcessor,PaliGemmaProcessor:()=>_.PaliGemmaProcessor,Phi3VProcessor:()=>u.Phi3VProcessor,PyAnnoteProcessor:()=>p.PyAnnoteProcessor,Qwen2VLProcessor:()=>m.Qwen2VLProcessor,SamProcessor:()=>h.SamProcessor,SmolVLMProcessor:()=>g.SmolVLMProcessor,SpeechT5Processor:()=>f.SpeechT5Processor,UltravoxProcessor:()=>w.UltravoxProcessor,VLChatProcessor:()=>i.VLChatProcessor,Wav2Vec2Processor:()=>M.Wav2Vec2Processor,Wav2Vec2ProcessorWithLM:()=>x.Wav2Vec2ProcessorWithLM,WhisperProcessor:()=>b.WhisperProcessor});var r=s("./src/models/florence2/processing_florence2.js"),o=s("./src/models/grounding_dino/processing_grounding_dino.js"),n=s("./src/models/idefics3/processing_idefics3.js"),i=s("./src/models/janus/processing_janus.js"),a=s("./src/models/jina_clip/processing_jina_clip.js"),l=s("./src/models/mgp_str/processing_mgp_str.js"),c=s("./src/models/moonshine/processing_moonshine.js"),d=s("./src/models/owlvit/processing_owlvit.js"),u=s("./src/models/phi3_v/processing_phi3_v.js"),_=s("./src/models/paligemma/processing_paligemma.js"),p=s("./src/models/pyannote/processing_pyannote.js"),m=s("./src/models/qwen2_vl/processing_qwen2_vl.js"),h=s("./src/models/sam/processing_sam.js"),g=s("./src/models/smolvlm/processing_smolvlm.js"),f=s("./src/models/speecht5/processing_speecht5.js"),w=s("./src/models/ultravox/processing_ultravox.js"),M=s("./src/models/wav2vec2/processing_wav2vec2.js"),x=s("./src/models/wav2vec2_with_lm/processing_wav2vec2_with_lm.js"),b=s("./src/models/whisper/processing_whisper.js")},"./src/models/pvt/image_processing_pvt.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{PvtImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}},"./src/models/pyannote/feature_extraction_pyannote.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{PyAnnoteFeatureExtractor:()=>i});var r=s("./src/base/feature_extraction_utils.js"),o=s("./src/utils/tensor.js"),n=s("./src/utils/maths.js");class i extends r.FeatureExtractor{async _call(e){(0,r.validate_audio_inputs)(e,"PyAnnoteFeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));const t=[1,1,e.length];return{input_values:new o.Tensor("float32",e,t)}}samples_to_frames(e){return(e-this.config.offset)/this.config.step}post_process_speaker_diarization(e,t){const s=t/this.samples_to_frames(t)/this.config.sampling_rate,r=[];for(const t of e.tolist()){const e=[];let o=-1;for(let s=0;s<t.length;++s){const r=(0,n.softmax)(t[s]),[i,a]=(0,n.max)(r),[l,c]=[s,s+1];a!==o?(o=a,e.push({id:a,start:l,end:c,score:i})):(e.at(-1).end=c,e.at(-1).score+=i)}r.push(e.map((({id:e,start:t,end:r,score:o})=>({id:e,start:t*s,end:r*s,confidence:o/(r-t)}))))}return r}}},"./src/models/pyannote/processing_pyannote.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{PyAnnoteProcessor:()=>n});var r=s("./src/base/processing_utils.js"),o=s("./src/models/pyannote/feature_extraction_pyannote.js");class n extends r.Processor{static feature_extractor_class=o.PyAnnoteFeatureExtractor;async _call(e){return await this.feature_extractor(e)}post_process_speaker_diarization(...e){return this.feature_extractor.post_process_speaker_diarization(...e)}get sampling_rate(){return this.feature_extractor.config.sampling_rate}}},"./src/models/qwen2_vl/image_processing_qwen2_vl.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Qwen2VLImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js"),o=s("./src/utils/tensor.js");class n extends r.ImageProcessor{async _call(e,...t){const{pixel_values:s,original_sizes:r,reshaped_input_sizes:n}=await super._call(e,...t);let i=s;const{temporal_patch_size:a,merge_size:l,patch_size:c}=this.config;1===i.dims[0]&&(i=(0,o.cat)(Array.from({length:a},(()=>i)),0));const d=i.dims[0]/a,u=i.dims[1],_=Math.floor(i.dims[2]/c),p=Math.floor(i.dims[3]/c);return{pixel_values:i.view(d,a,u,Math.floor(_/l),l,c,Math.floor(p/l),l,c).permute(0,3,6,4,7,2,1,5,8).view(d*_*p,u*a*c*c),image_grid_thw:new o.Tensor("int64",[d,_,p],[1,3]),original_sizes:r,reshaped_input_sizes:n}}}},"./src/models/qwen2_vl/processing_qwen2_vl.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Qwen2VLProcessor:()=>i});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js"),n=s("./src/tokenizers.js");s("./src/utils/image.js");class i extends r.Processor{static image_processor_class=o.AutoImageProcessor;static tokenizer_class=n.AutoTokenizer;async _call(e,t=null,...s){let r,o;if(Array.isArray(e)||(e=[e]),t&&(r=await this.image_processor(t),o=r.image_grid_thw),o){let t=this.image_processor.config.merge_size**2,s=0;const r=o.tolist();e=e.map((e=>{for(;e.includes("<|image_pad|>");){const o=Number(r[s++].reduce(((e,t)=>e*t),1n));e=e.replace("<|image_pad|>","<|placeholder|>".repeat(Math.floor(o/t)))}return e.replaceAll("<|placeholder|>","<|image_pad|>")}))}return{...this.tokenizer(e),...r}}}},"./src/models/rt_detr/image_processing_rt_detr.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{RTDetrImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{post_process_object_detection(...e){return(0,r.post_process_object_detection)(...e)}}},"./src/models/sam/image_processing_sam.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SamImageProcessor:()=>i});var r=s("./src/base/image_processors_utils.js"),o=s("./src/utils/core.js"),n=s("./src/utils/tensor.js");class i extends r.ImageProcessor{reshape_input_points(e,t,s,r=!1){e=structuredClone(e);let i=(0,o.calculateDimensions)(e);if(3===i.length)r||(i=[1,...i]),e=[e];else if(4!==i.length)throw Error("The input_points must be a 4D tensor of shape `batch_size`, `point_batch_size`, `nb_points_per_image`, `2`.");for(let r=0;r<e.length;++r){let o=t[r],n=s[r],i=[n[0]/o[0],n[1]/o[1]];for(let t=0;t<e[r].length;++t)for(let s=0;s<e[r][t].length;++s)for(let o=0;o<e[r][t][s].length;++o)e[r][t][s][o]*=i[o%2]}return new n.Tensor("float32",Float32Array.from(e.flat(1/0)),i)}add_input_labels(e,t){let s=(0,o.calculateDimensions)(e);if(2===s.length)s=[1,...s],e=[e];else if(3!==s.length)throw Error("The input_points must be a 4D tensor of shape `batch_size`, `point_batch_size`, `nb_points_per_image`, `2`.");if(s.some(((e,s)=>e!==t.dims[s])))throw Error(`The first ${s.length} dimensions of 'input_points' and 'input_labels' must be the same.`);return new n.Tensor("int64",e.flat(1/0).map(BigInt),s)}async _call(e,{input_points:t=null,input_labels:s=null,input_boxes:r=null}={}){const o=await super._call(e);if(t&&(o.input_points=this.reshape_input_points(t,o.original_sizes,o.reshaped_input_sizes)),s){if(!o.input_points)throw Error("`input_points` must be provided if `input_labels` are provided.");o.input_labels=this.add_input_labels(s,o.input_points)}return r&&(o.input_boxes=this.reshape_input_points(r,o.original_sizes,o.reshaped_input_sizes,!0)),o}async post_process_masks(e,t,s,{mask_threshold:r=0,binarize:o=!0,pad_size:i=null}={}){const a=[],l=[(i=i??this.pad_size).height,i.width];for(let i=0;i<t.length;++i){const c=t[i],d=s[i];let u=await(0,n.interpolate_4d)(e[i],{mode:"bilinear",size:l});if(u=u.slice(null,null,[0,d[0]],[0,d[1]]),u=await(0,n.interpolate_4d)(u,{mode:"bilinear",size:c}),o){const e=u.data,t=new Uint8Array(e.length);for(let s=0;s<e.length;++s)e[s]>r&&(t[s]=1);u=new n.Tensor("bool",t,u.dims)}a.push(u)}return a}generate_crop_boxes(e,t,{crop_n_layers:s=0,overlap_ratio:r=512/1500,points_per_crop:o=32,crop_n_points_downscale_factor:n=1}={}){}}},"./src/models/sam/processing_sam.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SamProcessor:()=>n});var r=s("./src/base/processing_utils.js"),o=s("./src/models/auto/image_processing_auto.js");class n extends r.Processor{static image_processor_class=o.AutoImageProcessor;async _call(...e){return await this.image_processor(...e)}post_process_masks(...e){return this.image_processor.post_process_masks(...e)}reshape_input_points(...e){return this.image_processor.reshape_input_points(...e)}}},"./src/models/seamless_m4t/feature_extraction_seamless_m4t.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SeamlessM4TFeatureExtractor:()=>i});var r=s("./src/base/feature_extraction_utils.js"),o=s("./src/utils/tensor.js"),n=s("./src/utils/audio.js");class i extends r.FeatureExtractor{constructor(e){super(e);const t=this.config.sampling_rate,s=(0,n.mel_filter_bank)(257,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);this.mel_filters=s,this.window=(0,n.window_function)(400,"povey",{periodic:!1})}async _extract_fbank_features(e,t){return e=e.map((e=>32768*e)),(0,n.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,max_num_frames:t,transpose:!0})}async _call(e,{padding:t=!0,pad_to_multiple_of:s=2,do_normalize_per_mel_bins:n=!0,return_attention_mask:i=!0}={}){(0,r.validate_audio_inputs)(e,"SeamlessM4TFeatureExtractor");let a,l=await this._extract_fbank_features(e,this.config.max_length);if(n){const[e,t]=l.dims,s=l.data;for(let r=0;r<t;++r){let o=0;for(let n=0;n<e;++n)o+=s[n*t+r];const n=o/e;let i=0;for(let o=0;o<e;++o)i+=(s[o*t+r]-n)**2;i/=e-1;const a=Math.sqrt(i+1e-7);for(let o=0;o<e;++o){const e=o*t+r;s[e]=(s[e]-n)/a}}}if(t){const[e,t]=l.dims,r=l.data,n=e%s;if(n>0){const s=new Float32Array(t*(e+n));s.set(r),s.fill(this.config.padding_value,r.length);const c=e+n;l=new o.Tensor(l.type,s,[c,t]),i&&(a=new o.Tensor("int64",new BigInt64Array(c),[1,c]),a.data.fill(1n,0,e))}}const[c,d]=l.dims,u=this.config.stride;if(0!==c%u)throw new Error(`The number of frames (${c}) must be a multiple of the stride (${u}).`);const _=l.view(1,Math.floor(c/u),d*u),p={input_features:_};if(i){const e=_.dims[1],t=new BigInt64Array(e);if(a){const e=a.data;for(let s=1,r=0;s<c;s+=u,++r)t[r]=e[s]}else t.fill(1n);p.attention_mask=new o.Tensor("int64",t,[1,e])}return p}}},"./src/models/segformer/image_processing_segformer.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SegformerFeatureExtractor:()=>n,SegformerImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{post_process_semantic_segmentation(...e){return(0,r.post_process_semantic_segmentation)(...e)}}class n extends o{}},"./src/models/siglip/image_processing_siglip.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SiglipImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}},"./src/models/smolvlm/image_processing_smolvlm.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SmolVLMImageProcessor:()=>r.Idefics3ImageProcessor});var r=s("./src/models/idefics3/image_processing_idefics3.js")},"./src/models/smolvlm/processing_smolvlm.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SmolVLMProcessor:()=>r.Idefics3Processor});var r=s("./src/models/idefics3/processing_idefics3.js")},"./src/models/snac/feature_extraction_snac.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SnacFeatureExtractor:()=>o});var r=s("./src/models/dac/feature_extraction_dac.js");class o extends r.DacFeatureExtractor{}},"./src/models/speecht5/feature_extraction_speecht5.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SpeechT5FeatureExtractor:()=>o});var r=s("./src/base/feature_extraction_utils.js");class o extends r.FeatureExtractor{}},"./src/models/speecht5/processing_speecht5.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{SpeechT5Processor:()=>i});var r=s("./src/base/processing_utils.js"),o=s("./src/tokenizers.js"),n=s("./src/models/auto/feature_extraction_auto.js");class i extends r.Processor{static tokenizer_class=o.AutoTokenizer;static feature_extractor_class=n.AutoFeatureExtractor;async _call(e){return await this.feature_extractor(e)}}},"./src/models/swin2sr/image_processing_swin2sr.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Swin2SRImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{pad_image(e,t,s,r={}){const[o,n,i]=t;return super.pad_image(e,t,{width:n+(s-n%s)%s,height:o+(s-o%s)%s},{mode:"symmetric",center:!1,constant_values:-1,...r})}}},"./src/models/ultravox/processing_ultravox.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{UltravoxProcessor:()=>i});var r=s("./src/models/auto/feature_extraction_auto.js"),o=s("./src/tokenizers.js"),n=s("./src/base/processing_utils.js");class i extends n.Processor{static tokenizer_class=o.AutoTokenizer;static feature_extractor_class=r.AutoFeatureExtractor;static uses_processor_config=!0;async _call(e,t=null,s={}){if(Array.isArray(e))throw new Error("Batched inputs are not supported yet.");let r={};if(t){const o=t.length,{input_features:n}=await this.feature_extractor(t,{...s,max_length:o}),i=Math.round(o/this.config.encoder_ds_factor+1e-4),a=1+Math.ceil(i/this.config.stack_factor);r.audio_token_len=[a],r.audio_values=n;const l=this.config.audio_placeholder;if(!e.includes(l))throw new Error(`The input text does not contain the image token ${l}.`);e=e.replaceAll(l,l.repeat(a))}return{...this.tokenizer(e,{add_special_tokens:!1,...s}),...r}}}},"./src/models/vit/image_processing_vit.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{ViTFeatureExtractor:()=>n,ViTImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{}class n extends o{}},"./src/models/vitmatte/image_processing_vitmatte.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{VitMatteImageProcessor:()=>n});var r=s("./src/base/image_processors_utils.js"),o=s("./src/utils/tensor.js");class n extends r.ImageProcessor{async _call(e,t){Array.isArray(e)||(e=[e]),Array.isArray(t)||(t=[t]);const s=await Promise.all(e.map((e=>this.preprocess(e)))),r=await Promise.all(t.map((e=>this.preprocess(e,{do_normalize:!1,do_convert_rgb:!1,do_convert_grayscale:!0}))));return{pixel_values:(0,o.stack)(s.map(((e,t)=>(0,o.cat)([e.pixel_values,r[t].pixel_values],0))),0),original_sizes:s.map((e=>e.original_size)),reshaped_input_sizes:s.map((e=>e.reshaped_input_size))}}}},"./src/models/vitpose/image_processing_vitpose.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{VitPoseImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{post_process_pose_estimation(e,t,{threshold:s=null}={}){const r=e.tolist(),[o,n,i,a]=e.dims,l=[];for(let e=0;e<o;++e){const o=r[e],n=t[e],c=[];for(let e=0;e<n.length;++e){const t=n[e],r=[],l=[],d=[],u=t.at(-2)/a,_=t.at(-1)/i;for(let e=0;e<o.length;++e){let[t,n]=[0,0],i=0,a=-1/0;const c=o[e];for(let e=0;e<c.length;++e){const s=c[e];for(let r=0;r<s.length;++r){const o=s[r];i+=o,a=Math.max(a,o),t+=(r+.5)*o,n+=e*o}}if(null!=s&&a<s)continue;const p=[u*t/i,_*n/i];r.push(p),d.push(e),l.push(a)}c.push({bbox:t,scores:l,labels:d,keypoints:r})}l.push(c)}return l}}},"./src/models/wav2vec2/feature_extraction_wav2vec2.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Wav2Vec2FeatureExtractor:()=>n});var r=s("./src/base/feature_extraction_utils.js"),o=s("./src/utils/tensor.js");class n extends r.FeatureExtractor{_zero_mean_unit_var_norm(e){const t=e.reduce(((e,t)=>e+t),0)/e.length,s=e.reduce(((e,s)=>e+(s-t)**2),0)/e.length;return e.map((e=>(e-t)/Math.sqrt(s+1e-7)))}async _call(e){(0,r.validate_audio_inputs)(e,"Wav2Vec2FeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));let t=e;this.config.do_normalize&&(t=this._zero_mean_unit_var_norm(t));const s=[1,t.length];return{input_values:new o.Tensor("float32",t,s),attention_mask:new o.Tensor("int64",new BigInt64Array(t.length).fill(1n),s)}}}},"./src/models/wav2vec2/processing_wav2vec2.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Wav2Vec2Processor:()=>i});var r=s("./src/tokenizers.js"),o=s("./src/models/auto/feature_extraction_auto.js"),n=s("./src/base/processing_utils.js");class i extends n.Processor{static tokenizer_class=r.AutoTokenizer;static feature_extractor_class=o.AutoFeatureExtractor;async _call(e){return await this.feature_extractor(e)}}},"./src/models/wav2vec2_with_lm/processing_wav2vec2_with_lm.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Wav2Vec2ProcessorWithLM:()=>i});var r=s("./src/tokenizers.js"),o=s("./src/models/auto/feature_extraction_auto.js"),n=s("./src/base/processing_utils.js");class i extends n.Processor{static tokenizer_class=r.AutoTokenizer;static feature_extractor_class=o.AutoFeatureExtractor;async _call(e){return await this.feature_extractor(e)}}},"./src/models/wespeaker/feature_extraction_wespeaker.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{WeSpeakerFeatureExtractor:()=>n});var r=s("./src/base/feature_extraction_utils.js"),o=(s("./src/utils/tensor.js"),s("./src/utils/audio.js"));class n extends r.FeatureExtractor{constructor(e){super(e);const t=this.config.sampling_rate,s=(0,o.mel_filter_bank)(257,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);this.mel_filters=s,this.window=(0,o.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,o.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,transpose:!0,min_num_frames:this.min_num_frames})}async _call(e){(0,r.validate_audio_inputs)(e,"WeSpeakerFeatureExtractor");const t=(await this._extract_fbank_features(e)).unsqueeze_(0);if(null===this.config.fbank_centering_span){const e=t.mean(1).data,s=t.data,[r,o,n]=t.dims;for(let t=0;t<r;++t){const r=t*o*n,i=t*n;for(let t=0;t<o;++t){const o=r+t*n;for(let t=0;t<n;++t)s[o+t]-=e[i+t]}}}return{input_features:t}}}},"./src/models/whisper/common_whisper.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{WHISPER_LANGUAGE_MAPPING:()=>o,WHISPER_TO_LANGUAGE_CODE_MAPPING:()=>n,whisper_language_to_code:()=>i});const r=[["en","english"],["zh","chinese"],["de","german"],["es","spanish"],["ru","russian"],["ko","korean"],["fr","french"],["ja","japanese"],["pt","portuguese"],["tr","turkish"],["pl","polish"],["ca","catalan"],["nl","dutch"],["ar","arabic"],["sv","swedish"],["it","italian"],["id","indonesian"],["hi","hindi"],["fi","finnish"],["vi","vietnamese"],["he","hebrew"],["uk","ukrainian"],["el","greek"],["ms","malay"],["cs","czech"],["ro","romanian"],["da","danish"],["hu","hungarian"],["ta","tamil"],["no","norwegian"],["th","thai"],["ur","urdu"],["hr","croatian"],["bg","bulgarian"],["lt","lithuanian"],["la","latin"],["mi","maori"],["ml","malayalam"],["cy","welsh"],["sk","slovak"],["te","telugu"],["fa","persian"],["lv","latvian"],["bn","bengali"],["sr","serbian"],["az","azerbaijani"],["sl","slovenian"],["kn","kannada"],["et","estonian"],["mk","macedonian"],["br","breton"],["eu","basque"],["is","icelandic"],["hy","armenian"],["ne","nepali"],["mn","mongolian"],["bs","bosnian"],["kk","kazakh"],["sq","albanian"],["sw","swahili"],["gl","galician"],["mr","marathi"],["pa","punjabi"],["si","sinhala"],["km","khmer"],["sn","shona"],["yo","yoruba"],["so","somali"],["af","afrikaans"],["oc","occitan"],["ka","georgian"],["be","belarusian"],["tg","tajik"],["sd","sindhi"],["gu","gujarati"],["am","amharic"],["yi","yiddish"],["lo","lao"],["uz","uzbek"],["fo","faroese"],["ht","haitian creole"],["ps","pashto"],["tk","turkmen"],["nn","nynorsk"],["mt","maltese"],["sa","sanskrit"],["lb","luxembourgish"],["my","myanmar"],["bo","tibetan"],["tl","tagalog"],["mg","malagasy"],["as","assamese"],["tt","tatar"],["haw","hawaiian"],["ln","lingala"],["ha","hausa"],["ba","bashkir"],["jw","javanese"],["su","sundanese"]],o=new Map(r),n=new Map([...r.map((([e,t])=>[t,e])),["burmese","my"],["valencian","ca"],["flemish","nl"],["haitian","ht"],["letzeburgesch","lb"],["pushto","ps"],["panjabi","pa"],["moldavian","ro"],["moldovan","ro"],["sinhalese","si"],["castilian","es"]]);function i(e){e=e.toLowerCase();let t=n.get(e);if(void 0===t){const s=e.match(/^<\|([a-z]{2})\|>$/);if(s&&(e=s[1]),!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(Array.from(t))}`)}t=e}return t}},"./src/models/whisper/feature_extraction_whisper.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{WhisperFeatureExtractor:()=>i});var r=s("./src/base/feature_extraction_utils.js"),o=(s("./src/utils/tensor.js"),s("./src/utils/audio.js")),n=s("./src/utils/maths.js");class i extends r.FeatureExtractor{constructor(e){super(e),this.config.mel_filters??=(0,o.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,o.window_function)(this.config.n_fft,"hann")}async _extract_fbank_features(e){const t=await(0,o.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:Math.min(Math.floor(e.length/this.config.hop_length),this.config.nb_max_frames)}),s=t.data,r=(0,n.max)(s)[0];for(let e=0;e<s.length;++e)s[e]=(Math.max(s[e],r-8)+4)/4;return t}async _call(e,{max_length:t=null}={}){let s;(0,r.validate_audio_inputs)(e,"WhisperFeatureExtractor");const o=t??this.config.n_samples;e.length>o?(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`."),s=e.slice(0,o)):(s=new Float32Array(o),s.set(e));return{input_features:(await this._extract_fbank_features(s)).unsqueeze_(0)}}}},"./src/models/whisper/generation_whisper.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{WhisperGenerationConfig:()=>o});var r=s("./src/generation/configuration_utils.js");class o extends r.GenerationConfig{return_timestamps=null;return_token_timestamps=null;num_frames=null;alignment_heads=null;task=null;language=null;no_timestamps_token_id=null;prompt_ids=null;is_multilingual=null;lang_to_id=null;task_to_id=null;max_initial_timestamp_index=1}},"./src/models/whisper/processing_whisper.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{WhisperProcessor:()=>i});var r=s("./src/models/auto/feature_extraction_auto.js"),o=s("./src/tokenizers.js"),n=s("./src/base/processing_utils.js");class i extends n.Processor{static tokenizer_class=o.AutoTokenizer;static feature_extractor_class=r.AutoFeatureExtractor;async _call(e){return await this.feature_extractor(e)}}},"./src/models/yolos/image_processing_yolos.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{YolosFeatureExtractor:()=>n,YolosImageProcessor:()=>o});var r=s("./src/base/image_processors_utils.js");class o extends r.ImageProcessor{post_process_object_detection(...e){return(0,r.post_process_object_detection)(...e)}}class n extends o{}},"./src/ops/registry.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{TensorOpRegistry:()=>l});var r=s("./src/backends/onnx.js"),o=s("./src/utils/tensor.js"),n=s("./src/env.js");const i=n.apis.IS_BROWSER_ENV||n.apis.IS_WEBWORKER_ENV,a=async(e,t,s)=>{const n=await(0,r.createInferenceSession)(new Uint8Array(e),t);let a=Promise.resolve();return async e=>{const t=(0,r.isONNXProxy)(),l=Object.fromEntries(Object.entries(e).map((([e,s])=>[e,(t?s.clone():s).ort_tensor]))),c=await(a=i?a.then((()=>n.run(l))):n.run(l));return Array.isArray(s)?s.map((e=>new o.Tensor(c[e]))):new o.Tensor(c[s])}};class l{static session_options={};static get nearest_interpolate_4d(){return this._nearest_interpolate_4d||(this._nearest_interpolate_4d=a([8,10,18,0,58,129,1,10,41,10,1,120,10,0,10,0,10,1,115,18,1,121,34,6,82,101,115,105,122,101,42,18,10,4,109,111,100,101,34,7,110,101,97,114,101,115,116,160,1,3,18,1,114,90,31,10,1,120,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,99,10,3,18,1,104,10,3,18,1,119,90,15,10,1,115,18,10,10,8,8,7,18,4,10,2,8,4,98,31,10,1,121,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,99,10,3,18,1,104,10,3,18,1,119,66,2,16,21],this.session_options,"y")),this._nearest_interpolate_4d}static get bilinear_interpolate_4d(){return this._bilinear_interpolate_4d||(this._bilinear_interpolate_4d=a([8,9,18,0,58,128,1,10,40,10,1,120,10,0,10,0,10,1,115,18,1,121,34,6,82,101,115,105,122,101,42,17,10,4,109,111,100,101,34,6,108,105,110,101,97,114,160,1,3,18,1,114,90,31,10,1,120,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,99,10,3,18,1,104,10,3,18,1,119,90,15,10,1,115,18,10,10,8,8,7,18,4,10,2,8,4,98,31,10,1,121,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,99,10,3,18,1,104,10,3,18,1,119,66,2,16,20],this.session_options,"y")),this._bilinear_interpolate_4d}static get bicubic_interpolate_4d(){return this._bicubic_interpolate_4d||(this._bicubic_interpolate_4d=a([8,9,18,0,58,127,10,39,10,1,120,10,0,10,0,10,1,115,18,1,121,34,6,82,101,115,105,122,101,42,16,10,4,109,111,100,101,34,5,99,117,98,105,99,160,1,3,18,1,114,90,31,10,1,120,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,99,10,3,18,1,104,10,3,18,1,119,90,15,10,1,115,18,10,10,8,8,7,18,4,10,2,8,4,98,31,10,1,121,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,99,10,3,18,1,104,10,3,18,1,119,66,2,16,20],this.session_options,"y")),this._bicubic_interpolate_4d}static get matmul(){return this._matmul||(this._matmul=a([8,9,18,0,58,55,10,17,10,1,97,10,1,98,18,1,99,34,6,77,97,116,77,117,108,18,1,114,90,9,10,1,97,18,4,10,2,8,1,90,9,10,1,98,18,4,10,2,8,1,98,9,10,1,99,18,4,10,2,8,1,66,2,16,20],this.session_options,"c")),this._matmul}static get stft(){return this._stft||(this._stft=a([8,7,18,0,58,148,1,10,38,10,1,115,10,1,106,10,1,119,10,1,108,18,1,111,34,4,83,84,70,84,42,15,10,8,111,110,101,115,105,100,101,100,24,1,160,1,2,18,1,115,90,26,10,1,115,18,21,10,19,8,1,18,15,10,3,18,1,98,10,3,18,1,115,10,3,18,1,99,90,11,10,1,106,18,6,10,4,8,7,18,0,90,16,10,1,119,18,11,10,9,8,1,18,5,10,3,18,1,119,90,11,10,1,108,18,6,10,4,8,7,18,0,98,31,10,1,111,18,26,10,24,8,1,18,20,10,3,18,1,98,10,3,18,1,102,10,3,18,1,100,10,3,18,1,99,66,2,16,17],this.session_options,"o")),this._stft}static get rfft(){return this._rfft||(this._rfft=a([8,9,18,0,58,97,10,33,10,1,120,10,0,10,1,97,18,1,121,34,3,68,70,84,42,15,10,8,111,110,101,115,105,100,101,100,24,1,160,1,2,18,1,100,90,21,10,1,120,18,16,10,14,8,1,18,10,10,3,18,1,115,10,3,18,1,99,90,11,10,1,97,18,6,10,4,8,7,18,0,98,21,10,1,121,18,16,10,14,8,1,18,10,10,3,18,1,115,10,3,18,1,99,66,2,16,20],this.session_options,"y")),this._rfft}static get top_k(){return this._top_k||(this._top_k=a([8,10,18,0,58,73,10,18,10,1,120,10,1,107,18,1,118,18,1,105,34,4,84,111,112,75,18,1,116,90,9,10,1,120,18,4,10,2,8,1,90,15,10,1,107,18,10,10,8,8,7,18,4,10,2,8,1,98,9,10,1,118,18,4,10,2,8,1,98,9,10,1,105,18,4,10,2,8,7,66,2,16,21],this.session_options,["v","i"])),this._top_k}static get slice(){return this._slice||(this._slice=a([8,7,18,0,58,96,10,25,10,1,120,10,1,115,10,1,101,10,1,97,10,1,116,18,1,121,34,5,83,108,105,99,101,18,1,114,90,9,10,1,120,18,4,10,2,8,1,90,9,10,1,115,18,4,10,2,8,7,90,9,10,1,101,18,4,10,2,8,7,90,9,10,1,97,18,4,10,2,8,7,90,9,10,1,116,18,4,10,2,8,7,98,9,10,1,121,18,4,10,2,8,1,66,2,16,13],this.session_options,"y")),this._slice}}},"./src/pipelines.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{AudioClassificationPipeline:()=>C,AutomaticSpeechRecognitionPipeline:()=>A,BackgroundRemovalPipeline:()=>z,DepthEstimationPipeline:()=>R,DocumentQuestionAnsweringPipeline:()=>V,FeatureExtractionPipeline:()=>P,FillMaskPipeline:()=>M,ImageClassificationPipeline:()=>L,ImageFeatureExtractionPipeline:()=>F,ImageSegmentationPipeline:()=>I,ImageToImagePipeline:()=>B,ImageToTextPipeline:()=>E,ObjectDetectionPipeline:()=>O,Pipeline:()=>h,QuestionAnsweringPipeline:()=>w,SummarizationPipeline:()=>b,Text2TextGenerationPipeline:()=>x,TextClassificationPipeline:()=>g,TextGenerationPipeline:()=>v,TextToAudioPipeline:()=>N,TokenClassificationPipeline:()=>f,TranslationPipeline:()=>k,ZeroShotAudioClassificationPipeline:()=>S,ZeroShotClassificationPipeline:()=>T,ZeroShotImageClassificationPipeline:()=>j,ZeroShotObjectDetectionPipeline:()=>D,pipeline:()=>q});var r=s("./src/tokenizers.js"),o=s("./src/models.js"),n=s("./src/models/auto/processing_auto.js"),i=(s("./src/base/processing_utils.js"),s("./src/utils/generic.js")),a=s("./src/utils/core.js"),l=s("./src/utils/maths.js"),c=s("./src/utils/audio.js"),d=s("./src/utils/tensor.js"),u=s("./src/utils/image.js");async function _(e){return Array.isArray(e)||(e=[e]),await Promise.all(e.map((e=>u.RawImage.read(e))))}async function p(e,t){return Array.isArray(e)||(e=[e]),await Promise.all(e.map((e=>"string"==typeof e||e instanceof URL?(0,c.read_audio)(e,t):e instanceof Float64Array?new Float32Array(e):e)))}function m(e,t){t&&(e=e.map((e=>0|e)));const[s,r,o,n]=e;return{xmin:s,ymin:r,xmax:o,ymax:n}}class h extends i.Callable{constructor({task:e,model:t,tokenizer:s=null,processor:r=null}){super(),this.task=e,this.model=t,this.tokenizer=s,this.processor=r}async dispose(){await this.model.dispose()}}class g extends h{constructor(e){super(e)}async _call(e,{top_k:t=1}={}){const s=this.tokenizer(e,{padding:!0,truncation:!0}),r=await this.model(s),o="multi_label_classification"===this.model.config.problem_type?e=>e.sigmoid():e=>new d.Tensor("float32",(0,l.softmax)(e.data),e.dims),n=this.model.config.id2label,i=[];for(const e of r.logits){const s=o(e),r=await(0,d.topk)(s,t),a=r[0].tolist(),l=r[1].tolist().map(((e,t)=>({label:n?n[e]:`LABEL_${e}`,score:a[t]})));1===t?i.push(...l):i.push(l)}return Array.isArray(e)||1===t?i:i[0]}}class f extends h{constructor(e){super(e)}async _call(e,{ignore_labels:t=["O"]}={}){const s=Array.isArray(e),r=this.tokenizer(s?e:[e],{padding:!0,truncation:!0}),o=(await this.model(r)).logits,n=this.model.config.id2label,i=[];for(let e=0;e<o.dims[0];++e){const s=r.input_ids[e],a=o[e],c=[];for(let e=0;e<a.dims[0];++e){const r=a[e],o=(0,l.max)(r.data)[1],i=n?n[o]:`LABEL_${o}`;if(t.includes(i))continue;const d=this.tokenizer.decode([s[e].item()],{skip_special_tokens:!0});if(""===d)continue;const u=(0,l.softmax)(r.data);c.push({entity:i,score:u[o],index:e,word:d})}i.push(c)}return s?i:i[0]}}class w extends h{constructor(e){super(e)}async _call(e,t,{top_k:s=1}={}){const r=this.tokenizer(e,{text_pair:t,padding:!0,truncation:!0}),{start_logits:o,end_logits:n}=await this.model(r),i=r.input_ids.tolist(),c=r.attention_mask.tolist(),d=this.tokenizer.all_special_ids,u=[];for(let e=0;e<o.dims[0];++e){const t=i[e],r=t.findIndex((e=>e==this.tokenizer.sep_token_id)),_=(c[e].map(((e,s)=>1==e&&(0===s||s>r&&-1===d.findIndex((e=>e==t[s]))))),o[e].tolist()),p=n[e].tolist();for(let s=1;s<_.length;++s)(0==c[e]||s<=r||-1!==d.findIndex((e=>e==t[s])))&&(_[s]=-1/0,p[s]=-1/0);const m=(0,l.softmax)(_).map(((e,t)=>[e,t])),h=(0,l.softmax)(p).map(((e,t)=>[e,t]));m[0][0]=0,h[0][0]=0;const g=(0,a.product)(m,h).filter((e=>e[0][1]<=e[1][1])).map((e=>[e[0][1],e[1][1],e[0][0]*e[1][0]])).sort(((e,t)=>t[2]-e[2]));for(let e=0;e<Math.min(g.length,s);++e){const[s,r,o]=g[e],n=t.slice(s,r+1),i=this.tokenizer.decode(n,{skip_special_tokens:!0});u.push({answer:i,score:o})}}return 1===s?u[0]:u}}class M extends h{constructor(e){super(e)}async _call(e,{top_k:t=5}={}){const s=this.tokenizer(e,{padding:!0,truncation:!0}),{logits:r}=await this.model(s),o=[],n=s.input_ids.tolist();for(let e=0;e<n.length;++e){const s=n[e],i=s.findIndex((e=>e==this.tokenizer.mask_token_id));if(-1===i)throw Error(`Mask token (${this.tokenizer.mask_token}) not found in text.`);const a=r[e][i],c=await(0,d.topk)(new d.Tensor("float32",(0,l.softmax)(a.data),a.dims),t),u=c[0].tolist(),_=c[1].tolist();o.push(_.map(((e,t)=>{const r=s.slice();return r[i]=e,{score:u[t],token:Number(e),token_str:this.tokenizer.decode([e]),sequence:this.tokenizer.decode(r,{skip_special_tokens:!0})}})))}return Array.isArray(e)?o:o[0]}}class x extends h{_key="generated_text";constructor(e){super(e)}async _call(e,t={}){Array.isArray(e)||(e=[e]),this.model.config.prefix&&(e=e.map((e=>this.model.config.prefix+e)));const s=this.model.config.task_specific_params;s&&s[this.task]&&s[this.task].prefix&&(e=e.map((e=>s[this.task].prefix+e)));const r=this.tokenizer,o={padding:!0,truncation:!0};let n;n=this instanceof k&&"_build_translation_inputs"in r?r._build_translation_inputs(e,o,t):r(e,o);const i=await this.model.generate({...n,...t});return r.batch_decode(i,{skip_special_tokens:!0}).map((e=>({[this._key]:e})))}}class b extends x{_key="summary_text";constructor(e){super(e)}}class k extends x{_key="translation_text";constructor(e){super(e)}}function y(e){return Array.isArray(e)&&e.every((e=>"role"in e&&"content"in e))}class v extends h{constructor(e){super(e)}async _call(e,t={}){let s,r=!1,o=!1;if("string"==typeof e)s=e=[e];else if(Array.isArray(e)&&e.every((e=>"string"==typeof e)))r=!0,s=e;else{if(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");r=!0}o=!0,s=e.map((e=>this.tokenizer.apply_chat_template(e,{tokenize:!1,add_generation_prompt:!0})))}const n=t.add_special_tokens??!1,i=!o&&(t.return_full_text??!0);this.tokenizer.padding_side="left";const a=this.tokenizer(s,{add_special_tokens:n,padding:!0,truncation:!0}),l=await this.model.generate({...a,...t}),c=this.tokenizer.batch_decode(l,{skip_special_tokens:!0});let d;!i&&a.input_ids.dims.at(-1)>0&&(d=this.tokenizer.batch_decode(a.input_ids,{skip_special_tokens:!0}).map((e=>e.length)));const u=Array.from({length:e.length},(e=>[]));for(let t=0;t<c.length;++t){const s=Math.floor(t/l.dims[0]*e.length);d&&(c[t]=c[t].slice(d[s])),u[s].push({generated_text:o?[...e[s],{role:"assistant",content:c[t]}]:c[t]})}return r||1!==u.length?u:u[0]}}class T extends h{constructor(e){super(e),this.label2id=Object.fromEntries(Object.entries(this.model.config.label2id).map((([e,t])=>[e.toLowerCase(),t]))),this.entailment_id=this.label2id.entailment,void 0===this.entailment_id&&(console.warn("Could not find 'entailment' in label2id mapping. Using 2 as entailment_id."),this.entailment_id=2),this.contradiction_id=this.label2id.contradiction??this.label2id.not_entailment,void 0===this.contradiction_id&&(console.warn("Could not find 'contradiction' in label2id mapping. Using 0 as contradiction_id."),this.contradiction_id=0)}async _call(e,t,{hypothesis_template:s="This example is {}.",multi_label:r=!1}={}){const o=Array.isArray(e);o||(e=[e]),Array.isArray(t)||(t=[t]);const n=t.map((e=>s.replace("{}",e))),i=r||1===t.length,a=[];for(const s of e){const e=[];for(const t of n){const r=this.tokenizer(s,{text_pair:t,padding:!0,truncation:!0}),o=await this.model(r);i?e.push([o.logits.data[this.contradiction_id],o.logits.data[this.entailment_id]]):e.push(o.logits.data[this.entailment_id])}const r=(i?e.map((e=>(0,l.softmax)(e)[1])):(0,l.softmax)(e)).map(((e,t)=>[e,t])).sort(((e,t)=>t[0]-e[0]));a.push({sequence:s,labels:r.map((e=>t[e[1]])),scores:r.map((e=>e[0]))})}return o?a:a[0]}}class P extends h{constructor(e){super(e)}async _call(e,{pooling:t="none",normalize:s=!1,quantize:r=!1,precision:o="binary"}={}){const n=this.tokenizer(e,{padding:!0,truncation:!0}),i=await this.model(n);let a=i.last_hidden_state??i.logits??i.token_embeddings;if("none"===t);else if("mean"===t)a=(0,d.mean_pooling)(a,n.attention_mask);else{if("cls"!==t)throw Error(`Pooling method '${t}' not supported.`);a=a.slice(null,0)}return s&&(a=a.normalize(2,-1)),r&&(a=(0,d.quantize_embeddings)(a,o)),a}}class F extends h{constructor(e){super(e)}async _call(e,{pool:t=null}={}){const s=await _(e),{pixel_values:r}=await this.processor(s),o=await this.model({pixel_values:r});let n;if(t){if(!("pooler_output"in o))throw Error("No pooled output was returned. Make sure the model has a 'pooler' layer when using the 'pool' option.");n=o.pooler_output}else n=o.last_hidden_state??o.logits??o.image_embeds;return n}}class C extends h{constructor(e){super(e)}async _call(e,{top_k:t=5}={}){const s=this.processor.feature_extractor.config.sampling_rate,r=await p(e,s),o=this.model.config.id2label,n=[];for(const e of r){const s=await this.processor(e),r=(await this.model(s)).logits[0],i=await(0,d.topk)(new d.Tensor("float32",(0,l.softmax)(r.data),r.dims),t),a=i[0].tolist(),c=i[1].tolist().map(((e,t)=>({label:o?o[e]:`LABEL_${e}`,score:a[t]})));n.push(c)}return Array.isArray(e)?n:n[0]}}class S extends h{constructor(e){super(e)}async _call(e,t,{hypothesis_template:s="This is a sound of {}."}={}){const r=!Array.isArray(e);r&&(e=[e]);const o=t.map((e=>s.replace("{}",e))),n=this.tokenizer(o,{padding:!0,truncation:!0}),i=this.processor.feature_extractor.config.sampling_rate,a=await p(e,i),c=[];for(const e of a){const s=await this.processor(e),r=await this.model({...n,...s}),o=(0,l.softmax)(r.logits_per_audio.data);c.push([...o].map(((e,s)=>({score:e,label:t[s]}))))}return r?c[0]:c}}class A extends h{constructor(e){super(e)}async _call(e,t={}){switch(this.model.config.model_type){case"whisper":case"lite-whisper":return this._call_whisper(e,t);case"wav2vec2":case"wav2vec2-bert":case"unispeech":case"unispeech-sat":case"hubert":return this._call_wav2vec2(e,t);case"moonshine":return this._call_moonshine(e,t);default:throw new Error(`AutomaticSpeechRecognitionPipeline does not support model type '${this.model.config.model_type}'.`)}}async _call_wav2vec2(e,t){t.language&&console.warn('`language` parameter is not yet supported for `wav2vec2` models, defaulting to "English".'),t.task&&console.warn('`task` parameter is not yet supported for `wav2vec2` models, defaulting to "transcribe".');const s=!Array.isArray(e);s&&(e=[e]);const r=this.processor.feature_extractor.config.sampling_rate,o=await p(e,r),n=[];for(const e of o){const t=await this.processor(e),s=(await this.model(t)).logits[0],r=[];for(const e of s)r.push((0,l.max)(e.data)[1]);const o=this.tokenizer.decode(r);n.push({text:o})}return s?n[0]:n}async _call_whisper(e,t){const s=t.return_timestamps??!1,r=t.chunk_length_s??0,o=t.force_full_sequences??!1;let n=t.stride_length_s??null;const i={...t};"word"===s&&(i.return_token_timestamps=!0,i.return_timestamps=!1);const a=!Array.isArray(e);a&&(e=[e]);const c=this.processor.feature_extractor.config.chunk_length/this.model.config.max_source_positions,d=this.processor.feature_extractor.config.hop_length,u=this.processor.feature_extractor.config.sampling_rate,_=await p(e,u),m=[];for(const e of _){let t=[];if(r>0){if(null===n)n=r/6;else if(r<=n)throw Error("`chunk_length_s` must be larger than `stride_length_s`.");const s=u*r,o=u*n,i=s-2*o;let a=0;for(;;){const r=a+s,n=e.subarray(a,r),l=await this.processor(n),c=0===a,d=r>=e.length;if(t.push({stride:[n.length,c?0:o,d?0:o],input_features:l.input_features,is_last:d}),d)break;a+=i}}else t=[{stride:[e.length,0,0],input_features:(await this.processor(e)).input_features,is_last:!0}];for(const e of t){i.num_frames=Math.floor(e.stride[0]/d);const t=await this.model.generate({inputs:e.input_features,...i});"word"===s?(e.tokens=t.sequences.tolist()[0],e.token_timestamps=t.token_timestamps.tolist()[0].map((e=>(0,l.round)(e,2)))):e.tokens=t[0].tolist(),e.stride=e.stride.map((e=>e/u))}const[a,_]=this.tokenizer._decode_asr(t,{time_precision:c,return_timestamps:s,force_full_sequences:o});m.push({text:a,..._})}return a?m[0]:m}async _call_moonshine(e,t){const s=!Array.isArray(e);s&&(e=[e]);const r=this.processor.feature_extractor.config.sampling_rate,o=await p(e,r),n=[];for(const e of o){const s=await this.processor(e),o=6*Math.floor(e.length/r),i=await this.model.generate({max_new_tokens:o,...t,...s}),a=this.processor.batch_decode(i,{skip_special_tokens:!0})[0];n.push({text:a})}return s?n[0]:n}}class E extends h{constructor(e){super(e)}async _call(e,t={}){const s=Array.isArray(e),r=await _(e),{pixel_values:o}=await this.processor(r),n=[];for(const e of o){e.dims=[1,...e.dims];const s=await this.model.generate({inputs:e,...t}),r=this.tokenizer.batch_decode(s,{skip_special_tokens:!0}).map((e=>({generated_text:e.trim()})));n.push(r)}return s?n:n[0]}}class L extends h{constructor(e){super(e)}async _call(e,{top_k:t=5}={}){const s=await _(e),{pixel_values:r}=await this.processor(s),o=await this.model({pixel_values:r}),n=this.model.config.id2label,i=[];for(const e of o.logits){const s=await(0,d.topk)(new d.Tensor("float32",(0,l.softmax)(e.data),e.dims),t),r=s[0].tolist(),o=s[1].tolist().map(((e,t)=>({label:n?n[e]:`LABEL_${e}`,score:r[t]})));i.push(o)}return Array.isArray(e)?i:i[0]}}class I extends h{constructor(e){super(e),this.subtasks_mapping={panoptic:"post_process_panoptic_segmentation",instance:"post_process_instance_segmentation",semantic:"post_process_semantic_segmentation"}}async _call(e,{threshold:t=.5,mask_threshold:s=.5,overlap_mask_area_threshold:r=.8,label_ids_to_fuse:o=null,target_sizes:n=null,subtask:i=null}={}){if(Array.isArray(e)&&1!==e.length)throw Error("Image segmentation pipeline currently only supports a batch size of 1.");const a=await _(e),l=a.map((e=>[e.height,e.width])),c=await this.processor(a),{inputNames:d,outputNames:p}=this.model.sessions.model;if(!d.includes("pixel_values")){if(1!==d.length)throw Error(`Expected a single input name, but got ${d.length} inputs: ${d}.`);const e=d[0];if(e in c)throw Error(`Input name ${e} already exists in the inputs.`);c[e]=c.pixel_values}const m=await this.model(c);let h=null;if(null!==i)h=this.subtasks_mapping[i];else if(this.processor.image_processor)for(const[e,t]of Object.entries(this.subtasks_mapping))if(t in this.processor.image_processor){h=this.processor.image_processor[t].bind(this.processor.image_processor),i=e;break}const g=this.model.config.id2label,f=[];if(i)if("panoptic"===i||"instance"===i){const e=h(m,t,s,r,o,n??l)[0],i=e.segmentation;for(const t of e.segments_info){const e=new Uint8ClampedArray(i.data.length);for(let s=0;s<i.data.length;++s)i.data[s]===t.id&&(e[s]=255);const s=new u.RawImage(e,i.dims[1],i.dims[0],1);f.push({score:t.score,label:g[t.label_id],mask:s})}}else{if("semantic"!==i)throw Error(`Subtask ${i} not supported.`);{const{segmentation:e,labels:t}=h(m,n??l)[0];for(const s of t){const t=new Uint8ClampedArray(e.data.length);for(let r=0;r<e.data.length;++r)e.data[r]===s&&(t[r]=255);const r=new u.RawImage(t,e.dims[1],e.dims[0],1);f.push({score:null,label:g[s],mask:r})}}}else{const e=1e-5,t=m[p[0]];for(let s=0;s<l.length;++s){const r=l[s],o=t[s];o.data.some((t=>t<-e||t>1+e))&&o.sigmoid_();const n=await u.RawImage.fromTensor(o.mul_(255).to("uint8")).resize(r[1],r[0]);f.push({label:null,score:null,mask:n})}}return f}}class z extends I{constructor(e){super(e)}async _call(e,t={}){if(Array.isArray(e)&&1!==e.length)throw Error("Background removal pipeline currently only supports a batch size of 1.");const s=await _(e),r=await super._call(e,t);return s.map(((e,t)=>{const s=e.clone();return s.putAlpha(r[t].mask),s}))}}class j extends h{constructor(e){super(e)}async _call(e,t,{hypothesis_template:s="This is a photo of {}"}={}){const r=Array.isArray(e),o=await _(e),n=t.map((e=>s.replace("{}",e))),i=this.tokenizer(n,{padding:"siglip"!==this.model.config.model_type||"max_length",truncation:!0}),{pixel_values:a}=await this.processor(o),c=await this.model({...i,pixel_values:a}),d="siglip"===this.model.config.model_type?e=>e.sigmoid().data:e=>(0,l.softmax)(e.data),u=[];for(const e of c.logits_per_image){const s=[...d(e)].map(((e,s)=>({score:e,label:t[s]})));s.sort(((e,t)=>t.score-e.score)),u.push(s)}return r?u:u[0]}}class O extends h{constructor(e){super(e)}async _call(e,{threshold:t=.9,percentage:s=!1}={}){const r=Array.isArray(e);if(r&&1!==e.length)throw Error("Object detection pipeline currently only supports a batch size of 1.");const o=await _(e),n=s?null:o.map((e=>[e.height,e.width])),{pixel_values:i,pixel_mask:a}=await this.processor(o),l=await this.model({pixel_values:i,pixel_mask:a}),c=this.processor.image_processor.post_process_object_detection(l,t,n),d=this.model.config.id2label,u=c.map((e=>e.boxes.map(((t,r)=>({score:e.scores[r],label:d[e.classes[r]],box:m(t,!s)})))));return r?u:u[0]}}class D extends h{constructor(e){super(e)}async _call(e,t,{threshold:s=.1,top_k:r=null,percentage:o=!1}={}){const n=Array.isArray(e),i=await _(e),a=this.tokenizer(t,{padding:!0,truncation:!0}),l=await this.processor(i),c=[];for(let e=0;e<i.length;++e){const n=i[e],d=o?null:[[n.height,n.width]],u=l.pixel_values[e].unsqueeze_(0),_=await this.model({...a,pixel_values:u});let p;if("post_process_grounded_object_detection"in this.processor){const e=this.processor.post_process_grounded_object_detection(_,a.input_ids,{box_threshold:s,text_threshold:s,target_sizes:d})[0];p=e.boxes.map(((t,s)=>({score:e.scores[s],label:e.labels[s],box:m(t,!o)})))}else{const e=this.processor.image_processor.post_process_object_detection(_,s,d,!0)[0];p=e.boxes.map(((s,r)=>({score:e.scores[r],label:t[e.classes[r]],box:m(s,!o)})))}p.sort(((e,t)=>t.score-e.score)),null!==r&&(p=p.slice(0,r)),c.push(p)}return n?c:c[0]}}class V extends h{constructor(e){super(e)}async _call(e,t,s={}){const r=(await _(e))[0],{pixel_values:o}=await this.processor(r),n=`<s_docvqa><s_question>${t}</s_question><s_answer>`,i=this.tokenizer(n,{add_special_tokens:!1,padding:!0,truncation:!0}).input_ids,a=await this.model.generate({inputs:o,max_length:this.model.config.decoder.max_position_embeddings,decoder_input_ids:i,...s}),l=this.tokenizer.batch_decode(a)[0].match(/<s_answer>(.*?)<\/s_answer>/);let c=null;return l&&l.length>=2&&(c=l[1].trim()),[{answer:c}]}}class N extends h{DEFAULT_VOCODER_ID="Xenova/speecht5_hifigan";constructor(e){super(e),this.vocoder=e.vocoder??null}async _call(e,{speaker_embeddings:t=null}={}){return this.processor?this._call_text_to_spectrogram(e,{speaker_embeddings:t}):this._call_text_to_waveform(e)}async _call_text_to_waveform(e){const t=this.tokenizer(e,{padding:!0,truncation:!0}),{waveform:s}=await this.model(t),r=this.model.config.sampling_rate;return new c.RawAudio(s.data,r)}async _call_text_to_spectrogram(e,{speaker_embeddings:t}){if(this.vocoder||(console.log("No vocoder specified, using default HifiGan vocoder."),this.vocoder=await 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:s}=this.tokenizer(e,{padding:!0,truncation:!0}),{waveform:r}=await this.model.generate_speech(s,t,{vocoder:this.vocoder}),n=this.processor.feature_extractor.config.sampling_rate;return new c.RawAudio(r.data,n)}}class B extends h{constructor(e){super(e)}async _call(e){const t=await _(e),s=await this.processor(t),r=await this.model(s),o=[];for(const e of r.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 R extends h{constructor(e){super(e)}async _call(e){const t=await _(e),s=await this.processor(t),{predicted_depth:r}=await this.model(s),o=[];for(let e=0;e<t.length;++e){const s=r[e],[n,i]=s.dims.slice(-2),[a,l]=t[e].size,c=(await(0,d.interpolate_4d)(s.view(1,1,n,i),{size:[l,a],mode:"bilinear"})).view(l,a),_=c.min().item(),p=c.max().item(),m=c.sub(_).div_(p-_).mul_(255).to("uint8").unsqueeze(0),h=u.RawImage.fromTensor(m);o.push({predicted_depth:c,depth:h})}return o.length>1?o:o[0]}}const G=Object.freeze({"text-classification":{tokenizer:r.AutoTokenizer,pipeline:g,model:o.AutoModelForSequenceClassification,default:{model:"Xenova/distilbert-base-uncased-finetuned-sst-2-english"},type:"text"},"token-classification":{tokenizer:r.AutoTokenizer,pipeline:f,model:o.AutoModelForTokenClassification,default:{model:"Xenova/bert-base-multilingual-cased-ner-hrl"},type:"text"},"question-answering":{tokenizer:r.AutoTokenizer,pipeline:w,model:o.AutoModelForQuestionAnswering,default:{model:"Xenova/distilbert-base-cased-distilled-squad"},type:"text"},"fill-mask":{tokenizer:r.AutoTokenizer,pipeline:M,model:o.AutoModelForMaskedLM,default:{model:"Xenova/bert-base-uncased"},type:"text"},summarization:{tokenizer:r.AutoTokenizer,pipeline:b,model:o.AutoModelForSeq2SeqLM,default:{model:"Xenova/distilbart-cnn-6-6"},type:"text"},translation:{tokenizer:r.AutoTokenizer,pipeline:k,model:o.AutoModelForSeq2SeqLM,default:{model:"Xenova/t5-small"},type:"text"},"text2text-generation":{tokenizer:r.AutoTokenizer,pipeline:x,model:o.AutoModelForSeq2SeqLM,default:{model:"Xenova/flan-t5-small"},type:"text"},"text-generation":{tokenizer:r.AutoTokenizer,pipeline:v,model:o.AutoModelForCausalLM,default:{model:"Xenova/gpt2"},type:"text"},"zero-shot-classification":{tokenizer:r.AutoTokenizer,pipeline:T,model:o.AutoModelForSequenceClassification,default:{model:"Xenova/distilbert-base-uncased-mnli"},type:"text"},"audio-classification":{pipeline:C,model:o.AutoModelForAudioClassification,processor:n.AutoProcessor,default:{model:"Xenova/wav2vec2-base-superb-ks"},type:"audio"},"zero-shot-audio-classification":{tokenizer:r.AutoTokenizer,pipeline:S,model:o.AutoModel,processor:n.AutoProcessor,default:{model:"Xenova/clap-htsat-unfused"},type:"multimodal"},"automatic-speech-recognition":{tokenizer:r.AutoTokenizer,pipeline:A,model:[o.AutoModelForSpeechSeq2Seq,o.AutoModelForCTC],processor:n.AutoProcessor,default:{model:"Xenova/whisper-tiny.en"},type:"multimodal"},"text-to-audio":{tokenizer:r.AutoTokenizer,pipeline:N,model:[o.AutoModelForTextToWaveform,o.AutoModelForTextToSpectrogram],processor:[n.AutoProcessor,null],default:{model:"Xenova/speecht5_tts"},type:"text"},"image-to-text":{tokenizer:r.AutoTokenizer,pipeline:E,model:o.AutoModelForVision2Seq,processor:n.AutoProcessor,default:{model:"Xenova/vit-gpt2-image-captioning"},type:"multimodal"},"image-classification":{pipeline:L,model:o.AutoModelForImageClassification,processor:n.AutoProcessor,default:{model:"Xenova/vit-base-patch16-224"},type:"multimodal"},"image-segmentation":{pipeline:I,model:[o.AutoModelForImageSegmentation,o.AutoModelForSemanticSegmentation,o.AutoModelForUniversalSegmentation],processor:n.AutoProcessor,default:{model:"Xenova/detr-resnet-50-panoptic"},type:"multimodal"},"background-removal":{pipeline:z,model:[o.AutoModelForImageSegmentation,o.AutoModelForSemanticSegmentation,o.AutoModelForUniversalSegmentation],processor:n.AutoProcessor,default:{model:"Xenova/modnet"},type:"image"},"zero-shot-image-classification":{tokenizer:r.AutoTokenizer,pipeline:j,model:o.AutoModel,processor:n.AutoProcessor,default:{model:"Xenova/clip-vit-base-patch32"},type:"multimodal"},"object-detection":{pipeline:O,model:o.AutoModelForObjectDetection,processor:n.AutoProcessor,default:{model:"Xenova/detr-resnet-50"},type:"multimodal"},"zero-shot-object-detection":{tokenizer:r.AutoTokenizer,pipeline:D,model:o.AutoModelForZeroShotObjectDetection,processor:n.AutoProcessor,default:{model:"Xenova/owlvit-base-patch32"},type:"multimodal"},"document-question-answering":{tokenizer:r.AutoTokenizer,pipeline:V,model:o.AutoModelForDocumentQuestionAnswering,processor:n.AutoProcessor,default:{model:"Xenova/donut-base-finetuned-docvqa"},type:"multimodal"},"image-to-image":{pipeline:B,model:o.AutoModelForImageToImage,processor:n.AutoProcessor,default:{model:"Xenova/swin2SR-classical-sr-x2-64"},type:"image"},"depth-estimation":{pipeline:R,model:o.AutoModelForDepthEstimation,processor:n.AutoProcessor,default:{model:"Xenova/dpt-large"},type:"image"},"feature-extraction":{tokenizer:r.AutoTokenizer,pipeline:P,model:o.AutoModel,default:{model:"Xenova/all-MiniLM-L6-v2"},type:"text"},"image-feature-extraction":{processor:n.AutoProcessor,pipeline:F,model:[o.AutoModelForImageFeatureExtraction,o.AutoModel],default:{model:"Xenova/vit-base-patch16-224-in21k"},type:"image"}}),$=Object.freeze({"sentiment-analysis":"text-classification",ner:"token-classification",asr:"automatic-speech-recognition","text-to-speech":"text-to-audio",embeddings:"feature-extraction"});async function q(e,t=null,{progress_callback:s=null,config:r=null,cache_dir:o=null,local_files_only:n=!1,revision:i="main",device:l=null,dtype:c=null,subfolder:d="onnx",use_external_data_format:u=null,model_file_name:_=null,session_options:p={}}={}){e=$[e]??e;const m=G[e.split("_",1)[0]];if(!m)throw Error(`Unsupported pipeline: ${e}. Must be one of [${Object.keys(G)}]`);t||(t=m.default.model,console.log(`No model specified. Using default model: "${t}".`));const h={progress_callback:s,config:r,cache_dir:o,local_files_only:n,revision:i,device:l,dtype:c,subfolder:d,use_external_data_format:u,model_file_name:_,session_options:p},g=new Map([["tokenizer",m.tokenizer],["model",m.model],["processor",m.processor]]),f=await async function(e,t,s){const r=Object.create(null),o=[];for(const[n,i]of e.entries()){if(!i)continue;let e;e=Array.isArray(i)?new Promise((async(e,r)=>{let o;for(const n of i){if(null===n)return void e(null);try{return void e(await n.from_pretrained(t,s))}catch(e){if(e.message?.includes("Unsupported model type"))o=e;else{if(!e.message?.includes("Could not locate file"))return void r(e);o=e}}}r(o)})):i.from_pretrained(t,s),r[n]=e,o.push(e)}await Promise.all(o);for(const[e,t]of Object.entries(r))r[e]=await t;return r}(g,t,h);f.task=e,(0,a.dispatchCallback)(s,{status:"ready",task:e,model:t});return new(0,m.pipeline)(f)}},"./src/tokenizers.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{AlbertTokenizer:()=>ve,AutoTokenizer:()=>gt,BartTokenizer:()=>Ve,BertTokenizer:()=>ye,BlenderbotSmallTokenizer:()=>dt,BlenderbotTokenizer:()=>ct,BloomTokenizer:()=>Ge,CLIPTokenizer:()=>nt,CamembertTokenizer:()=>Ie,CodeGenTokenizer:()=>ot,CodeLlamaTokenizer:()=>We,CohereTokenizer:()=>mt,ConvBertTokenizer:()=>Ae,DebertaTokenizer:()=>Fe,DebertaV2Tokenizer:()=>Ce,DistilBertTokenizer:()=>Le,ElectraTokenizer:()=>je,EsmTokenizer:()=>Je,FalconTokenizer:()=>Qe,GPT2Tokenizer:()=>De,GPTNeoXTokenizer:()=>He,GemmaTokenizer:()=>Ke,Grok1Tokenizer:()=>Ze,HerbertTokenizer:()=>Se,LlamaTokenizer:()=>qe,M2M100Tokenizer:()=>st,MBart50Tokenizer:()=>Be,MBartTokenizer:()=>Ne,MPNetTokenizer:()=>Xe,MarianTokenizer:()=>at,MgpstrTokenizer:()=>ht,MobileBertTokenizer:()=>Te,NllbTokenizer:()=>tt,NougatTokenizer:()=>_t,PreTrainedTokenizer:()=>ke,Qwen2Tokenizer:()=>Ye,RoFormerTokenizer:()=>Ee,RobertaTokenizer:()=>Re,SiglipTokenizer:()=>it,SpeechT5Tokenizer:()=>ut,SqueezeBertTokenizer:()=>Pe,T5Tokenizer:()=>Oe,TokenizerModel:()=>y,VitsTokenizer:()=>pt,Wav2Vec2CTCTokenizer:()=>lt,WhisperTokenizer:()=>rt,XLMRobertaTokenizer:()=>Ue,XLMTokenizer:()=>ze,is_chinese_char:()=>f});var r=s("./src/utils/generic.js"),o=s("./src/utils/core.js"),n=s("./src/utils/hub.js"),i=s("./src/utils/maths.js"),a=s("./src/utils/tensor.js"),l=s("./src/utils/data-structures.js"),c=s("./node_modules/@huggingface/jinja/dist/index.js"),d=s("./src/models/whisper/common_whisper.js");async function u(e,t){const s=await Promise.all([(0,n.getModelJSON)(e,"tokenizer.json",!0,t),(0,n.getModelJSON)(e,"tokenizer_config.json",!0,t)]);return null!==t.legacy&&(s[1].legacy=t.legacy),s}function _(e,t=!0){if(void 0!==e.Regex){let t=e.Regex.replace(/\\([#&~])/g,"$1");for(const[e,s]of b)t=t.replaceAll(e,s);return new RegExp(t,"gu")}if(void 0!==e.String){const s=(0,o.escapeRegExp)(e.String);return new RegExp(t?s:`(${s})`,"gu")}return console.warn("Unknown pattern type:",e),null}function p(e){return new Map(Object.entries(e))}function m(e){const t=e.dims;switch(t.length){case 1:return e.tolist();case 2:if(1!==t[0])throw new Error("Unable to decode tensor with `batch size !== 1`. Use `tokenizer.batch_decode(...)` for batched inputs.");return e.tolist()[0];default:throw new Error(`Expected tensor to have 1-2 dimensions, got ${t.length}.`)}}function h(e){return e.replace(/ \./g,".").replace(/ \?/g,"?").replace(/ \!/g,"!").replace(/ ,/g,",").replace(/ \' /g,"'").replace(/ n\'t/g,"n't").replace(/ \'m/g,"'m").replace(/ \'s/g,"'s").replace(/ \'ve/g,"'ve").replace(/ \'re/g,"'re")}function g(e){return e.replace(/\p{M}/gu,"")}function f(e){return e>=19968&&e<=40959||e>=13312&&e<=19903||e>=131072&&e<=173791||e>=173824&&e<=177983||e>=177984&&e<=178207||e>=178208&&e<=183983||e>=63744&&e<=64255||e>=194560&&e<=195103}const w="\\p{P}\\u0021-\\u002F\\u003A-\\u0040\\u005B-\\u0060\\u007B-\\u007E",M=new RegExp(`^[${w}]+$`,"gu"),x=".,!?…。，、।۔،",b=new Map([["(?i:'s|'t|'re|'ve|'m|'ll|'d)","(?:'([sS]|[tT]|[rR][eE]|[vV][eE]|[mM]|[lL][lL]|[dD]))"],[` ?[^(\\s|[${x}])]+`,` ?[^\\s${x}]+`]]);class k{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 r.Callable{constructor(e){super(),this.config=e,this.vocab=[],this.tokens_to_ids=new Map,this.unk_token_id=void 0,this.unk_token=void 0,this.end_of_word_suffix=void 0,this.fuse_unk=this.config.fuse_unk??!1}static fromConfig(e,...t){switch(e.type){case"WordPiece":return new v(e);case"Unigram":return new T(e,...t);case"BPE":return new C(e);default:if(e.vocab)return Array.isArray(e.vocab)?new T(e,...t):Object.hasOwn(e,"continuing_subword_prefix")&&Object.hasOwn(e,"unk_token")?Object.hasOwn(e,"merges")?new C(e):new v(e):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,s){const r=[];let o=0;for(;o<e.length;)if(r.push(e[o]),(t.get(e[o])??s)===s)for(;++o<e.length&&(t.get(e[o])??s)===s;)t.get(r.at(-1))!==s&&(r[r.length-1]+=e[o]);else++o;return r}(e,this.tokens_to_ids,this.unk_token_id)),e}encode(e){throw Error("encode should be implemented in subclass.")}convert_tokens_to_ids(e){return e.map((e=>this.tokens_to_ids.get(e)??this.unk_token_id))}convert_ids_to_tokens(e){return e.map((e=>this.vocab[e]??this.unk_token))}}class v extends y{constructor(e){super(e),this.tokens_to_ids=p(e.vocab),this.unk_token_id=this.tokens_to_ids.get(e.unk_token),this.unk_token=e.unk_token,this.max_input_chars_per_word=e.max_input_chars_per_word??100,this.vocab=new Array(this.tokens_to_ids.size);for(const[e,t]of this.tokens_to_ids)this.vocab[t]=e}encode(e){const t=[];for(const s of e){const e=[...s];if(e.length>this.max_input_chars_per_word){t.push(this.unk_token);continue}let r=!1,o=0;const n=[];for(;o<e.length;){let t=e.length,s=null;for(;o<t;){let r=e.slice(o,t).join("");if(o>0&&(r=this.config.continuing_subword_prefix+r),this.tokens_to_ids.has(r)){s=r;break}--t}if(null===s){r=!0;break}n.push(s),o=t}r?t.push(this.unk_token):t.push(...n)}return t}}class T extends y{constructor(e,t){super(e);const s=e.vocab.length;this.vocab=new Array(s),this.scores=new Array(s);for(let t=0;t<s;++t)[this.vocab[t],this.scores[t]]=e.vocab[t];this.unk_token_id=e.unk_id,this.unk_token=this.vocab[e.unk_id],this.tokens_to_ids=new Map(this.vocab.map(((e,t)=>[e,t]))),this.bos_token=" ",this.bos_token_id=this.tokens_to_ids.get(this.bos_token),this.eos_token=t.eos_token,this.eos_token_id=this.tokens_to_ids.get(this.eos_token),this.unk_token=this.vocab[this.unk_token_id],this.minScore=(0,i.min)(this.scores)[0],this.unk_score=this.minScore-10,this.scores[this.unk_token_id]=this.unk_score,this.trie=new l.CharTrie,this.trie.extend(this.vocab),this.fuse_unk=!0}populateNodes(e){const t=e.chars;let s=0;for(;s<t.length;){let r=!1;const n=[],i=t.slice(s).join(""),a=this.trie.commonPrefixSearch(i);for(const t of a){n.push(t);const i=this.tokens_to_ids.get(t),a=this.scores[i],l=(0,o.len)(t);e.insert(s,l,a,i),r||1!==l||(r=!0)}r||e.insert(s,1,this.unk_score,this.unk_token_id),s+=1}}tokenize(e){const t=new l.TokenLattice(e,this.bos_token_id,this.eos_token_id);return this.populateNodes(t),t.tokens()}encode(e){const t=[];for(const s of e){const e=this.tokenize(s);t.push(...e)}return t}}const P=(()=>{const e=[...Array.from({length:"~".charCodeAt(0)-"!".charCodeAt(0)+1},((e,t)=>t+"!".charCodeAt(0))),...Array.from({length:"¬".charCodeAt(0)-"¡".charCodeAt(0)+1},((e,t)=>t+"¡".charCodeAt(0))),...Array.from({length:"ÿ".charCodeAt(0)-"®".charCodeAt(0)+1},((e,t)=>t+"®".charCodeAt(0)))],t=e.slice();let s=0;for(let r=0;r<256;++r)e.includes(r)||(e.push(r),t.push(256+s),s+=1);const r=t.map((e=>String.fromCharCode(e)));return Object.fromEntries(e.map(((e,t)=>[e,r[t]])))})(),F=(0,o.reverseDictionary)(P);class C extends y{constructor(e){super(e),this.tokens_to_ids=p(e.vocab),this.unk_token_id=this.tokens_to_ids.get(e.unk_token),this.unk_token=e.unk_token,this.vocab=new Array(this.tokens_to_ids.size);for(const[e,t]of this.tokens_to_ids)this.vocab[t]=e;const t=Array.isArray(e.merges[0]);this.merges=t?e.merges:e.merges.map((e=>e.split(" ",2))),this.bpe_ranks=new Map(this.merges.map(((e,t)=>[JSON.stringify(e),t]))),this.end_of_word_suffix=e.end_of_word_suffix,this.continuing_subword_suffix=e.continuing_subword_suffix??null,this.byte_fallback=this.config.byte_fallback??!1,this.byte_fallback&&(this.text_encoder=new TextEncoder),this.ignore_merges=this.config.ignore_merges??!1,this.max_length_to_cache=256,this.cache_capacity=1e4,this.cache=new l.LRUCache(this.cache_capacity)}clear_cache(){this.cache.clear()}bpe(e){if(0===e.length)return[];const t=this.cache.get(e);if(void 0!==t)return t;const s=Array.from(e);this.end_of_word_suffix&&(s[s.length-1]+=this.end_of_word_suffix);let r=[];if(s.length>1){const e=new l.PriorityQueue(((e,t)=>e.score<t.score));let t={token:s[0],bias:0,prev:null,next:null},o=t;for(let t=1;t<s.length;++t){const r={bias:t/s.length,token:s[t],prev:o,next:null};o.next=r,this._add_node(e,o),o=r}for(;!e.isEmpty();){const s=e.pop();if(s.deleted||!s.next||s.next.deleted)continue;if(s.deleted=!0,s.next.deleted=!0,s.prev){const e={...s.prev};s.prev.deleted=!0,s.prev=e,e.prev?e.prev.next=e:t=e}const r={token:s.token+s.next.token,bias:s.bias,prev:s.prev,next:s.next.next};r.prev?(r.prev.next=r,this._add_node(e,r.prev)):t=r,r.next&&(r.next.prev=r,this._add_node(e,r))}for(let e=t;null!==e;e=e.next)r.push(e.token)}else r=s;if(this.continuing_subword_suffix)for(let e=0;e<r.length-1;++e)r[e]+=this.continuing_subword_suffix;return e.length<this.max_length_to_cache&&this.cache.put(e,r),r}_add_node(e,t){const s=this.bpe_ranks.get(JSON.stringify([t.token,t.next.token]));void 0!==s&&(t.score=s+t.bias,e.push(t))}encode(e){const t=[];for(const s of e){if(this.ignore_merges&&this.tokens_to_ids.has(s)){t.push(s);continue}const e=this.bpe(s);for(const s of e)if(this.tokens_to_ids.has(s))t.push(s);else if(this.byte_fallback){const e=Array.from(this.text_encoder.encode(s)).map((e=>`<0x${e.toString(16).toUpperCase().padStart(2,"0")}>`));e.every((e=>this.tokens_to_ids.has(e)))?t.push(...e):t.push(this.unk_token)}else t.push(this.unk_token)}return t}}class S extends y{constructor(e,t){super(e),this.tokens_to_ids=p(t.target_lang?e.vocab[t.target_lang]:e.vocab),this.bos_token=t.bos_token,this.bos_token_id=this.tokens_to_ids.get(this.bos_token),this.eos_token=t.eos_token,this.eos_token_id=this.tokens_to_ids.get(this.eos_token),this.pad_token=t.pad_token,this.pad_token_id=this.tokens_to_ids.get(this.pad_token),this.unk_token=t.unk_token,this.unk_token_id=this.tokens_to_ids.get(this.unk_token),this.vocab=new Array(this.tokens_to_ids.size);for(const[e,t]of this.tokens_to_ids)this.vocab[t]=e}encode(e){return e}}class A extends r.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"BertNormalizer":return new G(e);case"Precompiled":return new me(e);case"Sequence":return new R(e);case"Replace":return new E(e);case"NFC":return new I(e);case"NFD":return new z(e);case"NFKC":return new j(e);case"NFKD":return new O(e);case"Strip":return new D(e);case"StripAccents":return new V(e);case"Lowercase":return new N(e);case"Prepend":return new B(e);default:throw new Error(`Unknown Normalizer type: ${e.type}`)}}normalize(e){throw Error("normalize should be implemented in subclass.")}_call(e){return this.normalize(e)}}class E extends A{normalize(e){const t=_(this.config.pattern);return null===t?e:e.replaceAll(t,this.config.content)}}class L extends A{form=void 0;normalize(e){return e=e.normalize(this.form)}}class I extends L{form="NFC"}class z extends L{form="NFD"}class j extends L{form="NFKC"}class O extends L{form="NFKD"}class D extends A{normalize(e){return this.config.strip_left&&this.config.strip_right?e=e.trim():(this.config.strip_left&&(e=e.trimStart()),this.config.strip_right&&(e=e.trimEnd())),e}}class V extends A{normalize(e){return e=g(e)}}class N extends A{normalize(e){return e=e.toLowerCase()}}class B extends A{normalize(e){return e=this.config.prepend+e}}class R extends A{constructor(e){super(e),this.normalizers=e.normalizers.map((e=>A.fromConfig(e)))}normalize(e){return this.normalizers.reduce(((e,t)=>t.normalize(e)),e)}}class G extends A{_tokenize_chinese_chars(e){const t=[];for(let s=0;s<e.length;++s){const r=e[s];f(r.charCodeAt(0))?(t.push(" "),t.push(r),t.push(" ")):t.push(r)}return t.join("")}stripAccents(e){return e.normalize("NFD").replace(/\p{Mn}/gu,"")}_is_control(e){switch(e){case"\t":case"\n":case"\r":return!1;default:return/^\p{Cc}|\p{Cf}|\p{Co}|\p{Cs}$/u.test(e)}}_clean_text(e){const t=[];for(const s of e){const e=s.charCodeAt(0);0===e||65533===e||this._is_control(s)||(/^\s$/.test(s)?t.push(" "):t.push(s))}return t.join("")}normalize(e){return this.config.clean_text&&(e=this._clean_text(e)),this.config.handle_chinese_chars&&(e=this._tokenize_chinese_chars(e)),this.config.lowercase?(e=e.toLowerCase(),!1!==this.config.strip_accents&&(e=this.stripAccents(e))):this.config.strip_accents&&(e=this.stripAccents(e)),e}}class $ extends r.Callable{static fromConfig(e){if(null===e)return null;switch(e.type){case"BertPreTokenizer":return new q(e);case"Sequence":return new he(e);case"Whitespace":return new ge(e);case"WhitespaceSplit":return new fe(e);case"Metaspace":return new _e(e);case"ByteLevel":return new W(e);case"Split":return new U(e);case"Punctuation":return new X(e);case"Digits":return new Q(e);case"Replace":return new we(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 q extends ${constructor(e){super(),this.pattern=new RegExp(`[^\\s${w}]+|[${w}]`,"gu")}pre_tokenize_text(e,t){return e.trim().match(this.pattern)||[]}}class W extends ${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 U extends ${constructor(e){super(),this.config=e,this.pattern=_(this.config.pattern,this.config.invert)}pre_tokenize_text(e,t){return null===this.pattern?[]:this.config.invert?e.match(this.pattern)||[]:"removed"===this.config.behavior?.toLowerCase()?e.split(this.pattern).filter((e=>e)):function(e,t){const s=[];let r=0;for(const o of e.matchAll(t)){const t=o[0];r<o.index&&s.push(e.slice(r,o.index)),t.length>0&&s.push(t),r=o.index+t.length}return r<e.length&&s.push(e.slice(r)),s}(e,this.pattern)}}class X extends ${constructor(e){super(),this.config=e,this.pattern=new RegExp(`[^${w}]+|[${w}]+`,"gu")}pre_tokenize_text(e,t){return e.match(this.pattern)||[]}}class Q extends ${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 H extends r.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"TemplateProcessing":return new K(e);case"ByteLevel":return new Z(e);case"RobertaProcessing":return new Y(e);case"BertProcessing":return new J(e);case"Sequence":return new ee(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 J extends H{constructor(e){super(e),this.cls=e.cls[0],this.sep=e.sep[0]}post_process(e,t=null,{add_special_tokens:s=!0}={}){s&&(e=(0,o.mergeArrays)([this.cls],e,[this.sep]));let r=new Array(e.length).fill(0);if(null!==t){const n=s&&this instanceof Y?[this.sep]:[],i=s?[this.sep]:[];e=(0,o.mergeArrays)(e,n,t,i),r=(0,o.mergeArrays)(r,new Array(t.length+n.length+i.length).fill(1))}return{tokens:e,token_type_ids:r}}}class Y extends J{}class K extends H{constructor(e){super(e),this.single=e.single,this.pair=e.pair}post_process(e,t=null,{add_special_tokens:s=!0}={}){const r=null===t?this.single:this.pair;let n=[],i=[];for(const a of r)"SpecialToken"in a?s&&(n.push(a.SpecialToken.id),i.push(a.SpecialToken.type_id)):"Sequence"in a&&("A"===a.Sequence.id?(n=(0,o.mergeArrays)(n,e),i=(0,o.mergeArrays)(i,new Array(e.length).fill(a.Sequence.type_id))):"B"===a.Sequence.id&&(n=(0,o.mergeArrays)(n,t),i=(0,o.mergeArrays)(i,new Array(t.length).fill(a.Sequence.type_id))));return{tokens:n,token_type_ids:i}}}class Z extends H{post_process(e,t=null){return t&&(e=(0,o.mergeArrays)(e,t)),{tokens:e}}}class ee extends H{constructor(e){super(e),this.processors=e.processors.map((e=>H.fromConfig(e)))}post_process(e,t=null,s={}){let r;for(const o of this.processors)if(o instanceof Z){if(e=o.post_process(e).tokens,t){t=o.post_process(t).tokens}}else{const n=o.post_process(e,t,s);e=n.tokens,r=n.token_type_ids}return{tokens:e,token_type_ids:r}}}class te extends r.Callable{constructor(e){super(),this.config=e,this.added_tokens=[],this.end_of_word_suffix=null,this.trim_offsets=e.trim_offsets}static fromConfig(e){if(null===e)return null;switch(e.type){case"WordPiece":return new ie(e);case"Metaspace":return new pe(e);case"ByteLevel":return new ae(e);case"Replace":return new se(e);case"ByteFallback":return new re(e);case"Fuse":return new oe(e);case"Strip":return new ne(e);case"Sequence":return new ce(e);case"CTC":return new le(e);case"BPEDecoder":return new de(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 se extends te{decode_chain(e){const t=_(this.config.pattern);return null===t?e:e.map((e=>e.replaceAll(t,this.config.content)))}}class re extends te{constructor(e){super(e),this.text_decoder=new TextDecoder}decode_chain(e){const t=[];let s=[];for(const r of e){let e=null;if(6===r.length&&r.startsWith("<0x")&&r.endsWith(">")){const t=parseInt(r.slice(3,5),16);isNaN(t)||(e=t)}if(null!==e)s.push(e);else{if(s.length>0){const e=this.text_decoder.decode(Uint8Array.from(s));t.push(e),s=[]}t.push(r)}}if(s.length>0){const e=this.text_decoder.decode(Uint8Array.from(s));t.push(e),s=[]}return t}}class oe extends te{decode_chain(e){return[e.join("")]}}class ne extends te{constructor(e){super(e),this.content=this.config.content,this.start=this.config.start,this.stop=this.config.stop}decode_chain(e){return e.map((e=>{let t=0;for(let s=0;s<this.start&&e[s]===this.content;++s)t=s+1;let s=e.length;for(let t=0;t<this.stop;++t){const r=e.length-t-1;if(e[r]!==this.content)break;s=r}return e.slice(t,s)}))}}class ie extends te{constructor(e){super(e),this.cleanup=e.cleanup}decode_chain(e){return e.map(((e,t)=>(0!==t&&(e=e.startsWith(this.config.prefix)?e.replace(this.config.prefix,""):" "+e),this.cleanup&&(e=h(e)),e)))}}class ae extends te{constructor(e){super(e),this.byte_decoder=F,this.text_decoder=new TextDecoder("utf-8",{fatal:!1,ignoreBOM:!0}),this.end_of_word_suffix=null}convert_tokens_to_string(e){const t=e.join(""),s=new Uint8Array([...t].map((e=>this.byte_decoder[e])));return this.text_decoder.decode(s)}decode_chain(e){const t=[];let s=[];for(const r of e)void 0!==this.added_tokens.find((e=>e.content===r))?(s.length>0&&(t.push(this.convert_tokens_to_string(s)),s=[]),t.push(r)):s.push(r);return s.length>0&&t.push(this.convert_tokens_to_string(s)),t}}class le extends te{constructor(e){super(e),this.pad_token=this.config.pad_token,this.word_delimiter_token=this.config.word_delimiter_token,this.cleanup=this.config.cleanup}convert_tokens_to_string(e){if(0===e.length)return"";const t=[e[0]];for(let s=1;s<e.length;++s)e[s]!==t.at(-1)&&t.push(e[s]);let s=t.filter((e=>e!==this.pad_token)).join("");return this.cleanup&&(s=h(s).replaceAll(this.word_delimiter_token," ").trim()),s}decode_chain(e){return[this.convert_tokens_to_string(e)]}}class ce extends te{constructor(e){super(e),this.decoders=e.decoders.map((e=>te.fromConfig(e)))}decode_chain(e){return this.decoders.reduce(((e,t)=>t.decode_chain(e)),e)}}class de extends te{constructor(e){super(e),this.suffix=this.config.suffix}decode_chain(e){return e.map(((t,s)=>t.replaceAll(this.suffix,s===e.length-1?"":" ")))}}class ue extends te{decode_chain(e){let t="";for(let s=1;s<e.length;s+=2)t+=e[s];return[t]}}class _e extends ${constructor(e){super(),this.addPrefixSpace=e.add_prefix_space,this.replacement=e.replacement,this.strRep=e.str_rep||this.replacement,this.prepend_scheme=e.prepend_scheme??"always"}pre_tokenize_text(e,{section_index:t}={}){let s=e.replaceAll(" ",this.strRep);return this.addPrefixSpace&&!s.startsWith(this.replacement)&&("always"===this.prepend_scheme||"first"===this.prepend_scheme&&0===t)&&(s=this.strRep+s),[s]}}class pe extends te{constructor(e){super(e),this.addPrefixSpace=e.add_prefix_space,this.replacement=e.replacement}decode_chain(e){const t=[];for(let s=0;s<e.length;++s){let r=e[s].replaceAll(this.replacement," ");this.addPrefixSpace&&0==s&&r.startsWith(" ")&&(r=r.substring(1)),t.push(r)}return t}}class me 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 ${constructor(e){super(),this.tokenizers=e.pretokenizers.map((e=>$.fromConfig(e)))}pre_tokenize_text(e,t){return this.tokenizers.reduce(((e,s)=>s.pre_tokenize(e,t)),[e])}}class ge extends ${constructor(e){super()}pre_tokenize_text(e,t){return e.match(/\w+|[^\w\s]+/g)||[]}}class fe extends ${constructor(e){super()}pre_tokenize_text(e,t){return function(e){return e.match(/\S+/g)||[]}(e)}}class we extends ${constructor(e){super(),this.config=e,this.pattern=_(this.config.pattern),this.content=this.config.content}pre_tokenize_text(e,t){return null===this.pattern?[e]:[e.replaceAll(this.pattern,this.config.content)]}}const Me=["bos_token","eos_token","unk_token","sep_token","pad_token","cls_token","mask_token"];function xe(e,t,s,r){for(const n of Object.keys(e)){const i=t-e[n].length,a=s(n),l=new Array(i).fill(a);e[n]="right"===r?(0,o.mergeArrays)(e[n],l):(0,o.mergeArrays)(l,e[n])}}function be(e,t){for(const s of Object.keys(e))e[s].length=t}class ke extends r.Callable{return_token_type_ids=!1;padding_side="right";constructor(e,t){super(),this._tokenizer_config=t,this.normalizer=A.fromConfig(e.normalizer),this.pre_tokenizer=$.fromConfig(e.pre_tokenizer),this.model=y.fromConfig(e.model,t),this.post_processor=H.fromConfig(e.post_processor),this.decoder=te.fromConfig(e.decoder),this.special_tokens=[],this.all_special_ids=[],this.added_tokens=[];for(const t of e.added_tokens){const e=new k(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_splitter=new l.DictionarySplitter(this.added_tokens.map((e=>e.content))),this.added_tokens_map=new Map(this.added_tokens.map((e=>[e.content,e]))),this.mask_token=this.getToken("mask_token"),this.mask_token_id=this.model.tokens_to_ids.get(this.mask_token),this.pad_token=this.getToken("pad_token","eos_token"),this.pad_token_id=this.model.tokens_to_ids.get(this.pad_token),this.sep_token=this.getToken("sep_token"),this.sep_token_id=this.model.tokens_to_ids.get(this.sep_token),this.unk_token=this.getToken("unk_token"),this.unk_token_id=this.model.tokens_to_ids.get(this.unk_token),this.bos_token=this.getToken("bos_token"),this.bos_token_id=this.model.tokens_to_ids.get(this.bos_token),this.eos_token=this.getToken("eos_token"),this.eos_token_id=this.model.tokens_to_ids.get(this.eos_token),this.model_max_length=t.model_max_length,this.remove_space=t.remove_space,this.clean_up_tokenization_spaces=t.clean_up_tokenization_spaces??!0,this.do_lowercase_and_remove_accent=t.do_lowercase_and_remove_accent??!1,t.padding_side&&(this.padding_side=t.padding_side),this.legacy=!1,this.chat_template=t.chat_template??null,Array.isArray(this.chat_template)){const e=Object.create(null);for(const{name:t,template:s}of this.chat_template){if("string"!=typeof t||"string"!=typeof s)throw new Error('Chat template must be a list of objects with "name" and "template" properties');e[t]=s}this.chat_template=e}this._compiled_template_cache=new Map}getToken(...e){for(const t of e){const e=this._tokenizer_config[t];if(e){if("object"==typeof e){if("AddedToken"===e.__type)return e.content;throw Error(`Unknown token: ${e}`)}return e}}return null}static async from_pretrained(e,{progress_callback:t=null,config:s=null,cache_dir:r=null,local_files_only:o=!1,revision:n="main",legacy:i=null}={}){return new this(...await u(e,{progress_callback:t,config:s,cache_dir:r,local_files_only:o,revision:n,legacy:i}))}_call(e,{text_pair:t=null,add_special_tokens:s=!0,padding:r=!1,truncation:o=null,max_length:n=null,return_tensor:l=!0,return_token_type_ids:c=null}={}){const d=Array.isArray(e);let u;if(d){if(0===e.length)throw Error("text array must be non-empty");if(null!==t){if(!Array.isArray(t))throw Error("text_pair must also be an array");if(e.length!==t.length)throw Error("text and text_pair must have the same length");u=e.map(((e,r)=>this._encode_plus(e,{text_pair:t[r],add_special_tokens:s,return_token_type_ids:c})))}else u=e.map((e=>this._encode_plus(e,{add_special_tokens:s,return_token_type_ids:c})))}else{if(null==e)throw Error("text may not be null or undefined");if(Array.isArray(t))throw Error("When specifying `text_pair`, since `text` is a string, `text_pair` must also be a string (i.e., not an array).");u=[this._encode_plus(e,{text_pair:t,add_special_tokens:s,return_token_type_ids:c})]}if(null===n?n=this.model_max_length:null===o&&(!0===r?(console.warn("`max_length` is ignored when `padding: true` and there is no truncation strategy. To pad to max length, use `padding: 'max_length'`."),n=this.model_max_length):!1===r&&(console.warn("Truncation was not explicitly activated but `max_length` is provided a specific value, please use `truncation: true` to explicitly truncate examples to max length."),o=!0)),!0===r&&(n=Math.min((0,i.max)(u.map((e=>e.input_ids.length)))[0],n??1/0)),n=Math.min(n,this.model_max_length??1/0),r||o)for(let e=0;e<u.length;++e)u[e].input_ids.length!==n&&(u[e].input_ids.length>n?o&&be(u[e],n):r&&xe(u[e],n,(e=>"input_ids"===e?this.pad_token_id:0),this.padding_side));const _={};if(l){if((!r||!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]))_[t]=new a.Tensor("int64",BigInt64Array.from(u.flatMap((e=>e[t])).map(BigInt)),e)}else{for(const e of Object.keys(u[0]))_[e]=u.map((t=>t[e]));if(!d)for(const e of Object.keys(_))_[e]=_[e][0]}return _}_encode_text(e){if(null===e)return null;const t=this.added_tokens_splitter.split(e);for(let e=0;e<t.length;++e){const s=this.added_tokens_map.get(t[e]);s&&(s.lstrip&&e>0&&(t[e-1]=t[e-1].trimEnd()),s.rstrip&&e<t.length-1&&(t[e+1]=t[e+1].trimStart()))}const s=t.flatMap(((e,t)=>{if(0===e.length)return[];if(this.added_tokens_map.has(e))return[e];if(!0===this.remove_space&&(e=e.trim().split(/\s+/).join(" ")),this.do_lowercase_and_remove_accent&&(e=function(e){return g(e.toLowerCase())}(e)),null!==this.normalizer&&(e=this.normalizer(e)),0===e.length)return[];const s=null!==this.pre_tokenizer?this.pre_tokenizer(e,{section_index:t}):[e];return this.model(s)}));return s}_encode_plus(e,{text_pair:t=null,add_special_tokens:s=!0,return_token_type_ids:r=null}={}){const{tokens:o,token_type_ids:n}=this._tokenize_helper(e,{pair:t,add_special_tokens:s}),i=this.model.convert_tokens_to_ids(o),a={input_ids:i,attention_mask:new Array(i.length).fill(1)};return(r??this.return_token_type_ids)&&n&&(a.token_type_ids=n),a}_tokenize_helper(e,{pair:t=null,add_special_tokens:s=!1}={}){const r=this._encode_text(e),n=this._encode_text(t);return this.post_processor?this.post_processor(r,n,{add_special_tokens:s}):{tokens:(0,o.mergeArrays)(r??[],n??[])}}tokenize(e,{pair:t=null,add_special_tokens:s=!1}={}){return this._tokenize_helper(e,{pair:t,add_special_tokens:s}).tokens}encode(e,{text_pair:t=null,add_special_tokens:s=!0,return_token_type_ids:r=null}={}){return this._encode_plus(e,{text_pair:t,add_special_tokens:s,return_token_type_ids:r}).input_ids}batch_decode(e,t={}){return e instanceof a.Tensor&&(e=e.tolist()),e.map((e=>this.decode(e,t)))}decode(e,t={}){if(e instanceof a.Tensor&&(e=m(e)),!Array.isArray(e)||0===e.length||!(0,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:s=null}){let r=this.model.convert_ids_to_tokens(e);t&&(r=r.filter((e=>!this.special_tokens.includes(e))));let o=this.decoder?this.decoder(r):r.join(" ");return this.decoder&&this.decoder.end_of_word_suffix&&(o=o.replaceAll(this.decoder.end_of_word_suffix," "),t&&(o=o.trim())),(s??this.clean_up_tokenization_spaces)&&(o=h(o)),o}get_chat_template({chat_template:e=null,tools:t=null}={}){if(this.chat_template&&"object"==typeof this.chat_template){const s=this.chat_template;if(null!==e&&Object.hasOwn(s,e))e=s[e];else if(null===e)if(null!==t&&"tool_use"in s)e=s.tool_use;else{if(!("default"in s))throw Error(`This model has multiple chat templates with no default specified! Please either pass a chat template or the name of the template you wish to use to the 'chat_template' argument. Available template names are ${Object.keys(s).sort()}.`);e=s.default}}else if(null===e){if(!this.chat_template)throw Error("Cannot use apply_chat_template() because tokenizer.chat_template is not set and no template argument was passed! For information about writing templates and setting the tokenizer.chat_template attribute, please see the documentation at https://huggingface.co/docs/transformers/main/en/chat_templating");e=this.chat_template}return e}apply_chat_template(e,{tools:t=null,documents:s=null,chat_template:r=null,add_generation_prompt:o=!1,tokenize:n=!0,padding:i=!1,truncation:a=!1,max_length:l=null,return_tensor:d=!0,return_dict:u=!1,tokenizer_kwargs:_={},...p}={}){if("string"!=typeof(r=this.get_chat_template({chat_template:r,tools:t})))throw Error("chat_template must be a string, but got "+typeof r);let m=this._compiled_template_cache.get(r);void 0===m&&(m=new c.Template(r),this._compiled_template_cache.set(r,m));const h=Object.create(null);for(const e of Me){const t=this.getToken(e);t&&(h[e]=t)}const g=m.render({messages:e,add_generation_prompt:o,tools:t,documents:s,...h,...p});if(n){const e=this._call(g,{add_special_tokens:!1,padding:i,truncation:a,max_length:l,return_tensor:d,..._});return u?e:e.input_ids}return g}}class ye extends ke{return_token_type_ids=!0}class ve extends ke{return_token_type_ids=!0}class Te extends ke{return_token_type_ids=!0}class Pe extends ke{return_token_type_ids=!0}class Fe extends ke{return_token_type_ids=!0}class Ce extends ke{return_token_type_ids=!0}class Se extends ke{return_token_type_ids=!0}class Ae extends ke{return_token_type_ids=!0}class Ee extends ke{return_token_type_ids=!0}class Le extends ke{}class Ie extends ke{}class ze extends ke{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 je extends ke{return_token_type_ids=!0}class Oe extends ke{}class De extends ke{}class Ve extends ke{}class Ne extends ke{constructor(e,t){super(e,t),this.languageRegex=/^[a-z]{2}_[A-Z]{2}$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))),this.lang_to_token=e=>e}_build_translation_inputs(e,t,s){return et(this,e,t,s)}}class Be extends Ne{}class Re extends ke{}class Ge extends ke{}const $e="▁";class qe extends ke{padding_side="left";constructor(e,t){super(e,t),this.legacy=t.legacy??!0,this.legacy||(this.normalizer=null,this.pre_tokenizer=new _e({replacement:$e,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($e+e.replaceAll($e," "));return t.length>1&&t[0]===$e&&this.special_tokens.includes(t[1])&&(t=t.slice(1)),t}}class We extends ke{}class Ue extends ke{}class Xe extends ke{}class Qe extends ke{}class He extends ke{}class Je extends ke{}class Ye extends ke{}class Ke extends ke{}class Ze extends ke{}function et(e,t,s,r){if(!("language_codes"in e)||!Array.isArray(e.language_codes))throw new Error("Tokenizer must have `language_codes` attribute set and it should be an array of language ids.");if(!("languageRegex"in e&&e.languageRegex instanceof RegExp))throw new Error("Tokenizer must have `languageRegex` attribute set and it should be a regular expression.");if(!("lang_to_token"in e)||"function"!=typeof e.lang_to_token)throw new Error("Tokenizer must have `lang_to_token` attribute set and it should be a function.");const o=r.src_lang,n=r.tgt_lang;if(!e.language_codes.includes(n))throw new Error(`Target language code "${n}" 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: {${e.language_codes.join(", ")}}`);for(const t of e.post_processor.config.single)if("SpecialToken"in t&&e.languageRegex.test(t.SpecialToken.id)){t.SpecialToken.id=e.lang_to_token(o);break}}return r.forced_bos_token_id=e.model.convert_tokens_to_ids([e.lang_to_token(n)])[0],e._call(t,s)}class tt extends ke{constructor(e,t){super(e,t),this.languageRegex=/^[a-z]{3}_[A-Z][a-z]{3}$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))),this.lang_to_token=e=>e}_build_translation_inputs(e,t,s){return et(this,e,t,s)}}class st extends ke{constructor(e,t){super(e,t),this.languageRegex=/^__[a-z]{2,3}__$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))).map((e=>e.slice(2,-2))),this.lang_to_token=e=>`__${e}__`}_build_translation_inputs(e,t,s){return et(this,e,t,s)}}class rt extends ke{get timestamp_begin(){return this.model.convert_tokens_to_ids(["<|notimestamps|>"])[0]+1}_decode_asr(e,{return_timestamps:t=!1,return_language:s=!1,time_precision:r=null,force_full_sequences:o=!0}={}){if(null===r)throw Error("Must specify time_precision");let n=null;const a="word"===t;function l(){return{language:n,timestamp:[null,null],text:""}}const c=[];let u=l(),_=0;const p=this.timestamp_begin,m=p+1500;let h=[],g=[],f=!1,w=null;const x=new Set(this.all_special_ids);for(const s of e){const e=s.tokens,o=a?s.token_timestamps:null;let b=null,k=p;if("stride"in s){const[t,o,n]=s.stride;if(_-=o,w=t-n,o&&(k=o/r+p),n)for(let t=e.length-1;t>=0;--t){const s=Number(e[t]);if(s>=p){if(null!==b&&(s-p)*r<w)break;b=s}}}let y=[],v=[];for(let s=0;s<e.length;++s){const w=Number(e[s]);if(x.has(w)){const e=this.decode([w]),s=d.WHISPER_LANGUAGE_MAPPING.get(e.slice(2,-2));if(void 0!==s){if(null!==n&&s!==n&&!t){h.push(y);const e=this.findLongestCommonSequence(h)[0],t=this.decode(e);u.text=t,c.push(u),h=[],y=[],u=l()}n=u.language=s}}else if(w>=p&&w<=m){const e=(w-p)*r+_,t=(0,i.round)(e,2);if(null!==b&&w>=b)f=!0;else if(f||h.length>0&&w<k)f=!1;else if(null===u.timestamp[0])u.timestamp[0]=t;else if(t===u.timestamp[0]);else{u.timestamp[1]=t,h.push(y),a&&g.push(v);const[e,s]=this.findLongestCommonSequence(h,g),r=this.decode(e);u.text=r,a&&(u.words=this.collateWordTimestamps(e,s,n)),c.push(u),h=[],y=[],g=[],v=[],u=l()}}else if(y.push(w),a){let e,t=(0,i.round)(o[s]+_,2);if(s+1<o.length){e=(0,i.round)(o[s+1]+_,2);const n=this.decode([w]);M.test(n)&&(e=(0,i.round)(Math.min(t+r,e),2))}else e=null;v.push([t,e])}}if("stride"in s){const[e,t,r]=s.stride;_+=e-r}y.length>0?(h.push(y),a&&g.push(v)):h.every((e=>0===e.length))&&(u=l(),h=[],y=[],g=[],v=[])}if(h.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,s]=this.findLongestCommonSequence(h,g),r=this.decode(e);u.text=r,a&&(u.words=this.collateWordTimestamps(e,s,n)),c.push(u)}let b=Object.create(null);const k=c.map((e=>e.text)).join("");if(t||s){for(let e=0;e<c.length;++e){const r=c[e];t||delete r.timestamp,s||delete r.language}if(a){const e=[];for(const t of c)for(const s of t.words)e.push(s);b={chunks:e}}else b={chunks:c}}return[k,b]}findLongestCommonSequence(e,t=null){let s=e[0],r=s.length,o=[];const n=Array.isArray(t)&&t.length>0;let i=n?[]:null,a=n?t[0]:null;for(let l=1;l<e.length;++l){const c=e[l];let d=0,u=[r,r,0,0];const _=c.length;for(let e=1;e<r+_;++e){const o=Math.max(0,r-e),i=Math.min(r,r+_-e),p=s.slice(o,i),m=Math.max(0,e-r),h=Math.min(_,e),g=c.slice(m,h);if(p.length!==g.length)throw new Error("There is a bug within whisper `decode_asr` function, please report it. Dropping to prevent bad inference.");let f;f=n?p.filter(((e,s)=>e===g[s]&&a[o+s]<=t[l][m+s])).length:p.filter(((e,t)=>e===g[t])).length;const w=f/e+e/1e4;f>1&&w>d&&(d=w,u=[o,i,m,h])}const[p,m,h,g]=u,f=Math.floor((m+p)/2),w=Math.floor((g+h)/2);o.push(...s.slice(0,f)),s=c.slice(w),r=s.length,n&&(i.push(...a.slice(0,f)),a=t[l].slice(w))}return o.push(...s),n?(i.push(...a),[o,i]):[o,[]]}collateWordTimestamps(e,t,s){const[r,o,n]=this.combineTokensIntoWords(e,s),i=[];for(let e=0;e<r.length;++e){const s=n[e];i.push({text:r[e],timestamp:[t[s.at(0)][0],t[s.at(-1)][1]]})}return i}combineTokensIntoWords(e,t,s="\"'“¡¿([{-",r="\"'.。,，!！?？:：”)]}、"){let o,n,i;return["chinese","japanese","thai","lao","myanmar"].includes(t=t??"english")?[o,n,i]=this.splitTokensOnUnicode(e):[o,n,i]=this.splitTokensOnSpaces(e),this.mergePunctuations(o,n,i,s,r)}decode(e,t){let s;return t?.decode_with_timestamps?(e instanceof a.Tensor&&(e=m(e)),s=this.decodeWithTimestamps(e,t)):s=super.decode(e,t),s}decodeWithTimestamps(e,t){const s=t?.time_precision??.02,r=Array.from(this.all_special_ids).at(-1)+1;let o=[[]];for(let t of e)if(t=Number(t),t>=r){const e=((t-r)*s).toFixed(2);o.push(`<|${e}|>`),o.push([])}else o[o.length-1].push(t);return o=o.map((e=>"string"==typeof e?e:super.decode(e,t))),o.join("")}splitTokensOnUnicode(e){const t=this.decode(e,{decode_with_timestamps:!0}),s=[],r=[],o=[];let n=[],i=[],a=0;for(let l=0;l<e.length;++l){const c=e[l];n.push(c),i.push(l);const d=this.decode(n,{decode_with_timestamps:!0});d.includes("�")&&"�"!==t[a+d.indexOf("�")]||(s.push(d),r.push(n),o.push(i),n=[],i=[],a+=d.length)}return[s,r,o]}splitTokensOnSpaces(e){const[t,s,r]=this.splitTokensOnUnicode(e),o=[],n=[],i=[],a=new RegExp(`^[${w}]$`,"gu");for(let e=0;e<t.length;++e){const l=t[e],c=s[e],d=r[e],u=c[0]>=this.model.tokens_to_ids.get("<|endoftext|>"),_=l.startsWith(" "),p=l.trim(),m=a.test(p);if(u||_||m||0===o.length)o.push(l),n.push(c),i.push(d);else{const e=o.length-1;o[e]+=l,n[e].push(...c),i[e].push(...d)}}return[o,n,i]}mergePunctuations(e,t,s,r,n){const i=structuredClone(e),a=structuredClone(t),l=structuredClone(s);let c=i.length-2,d=i.length-1;for(;c>=0;)i[c].startsWith(" ")&&r.includes(i[c].trim())?(i[d]=i[c]+i[d],a[d]=(0,o.mergeArrays)(a[c],a[d]),l[d]=(0,o.mergeArrays)(l[c],l[d]),i[c]="",a[c]=[],l[c]=[]):d=c,--c;for(c=0,d=1;d<i.length;)!i[c].endsWith(" ")&&n.includes(i[d])?(i[c]+=i[d],a[c]=(0,o.mergeArrays)(a[c],a[d]),l[c]=(0,o.mergeArrays)(l[c],l[d]),i[d]="",a[d]=[],l[d]=[]):c=d,++d;return[i.filter((e=>e)),a.filter((e=>e.length>0)),l.filter((e=>e.length>0))]}}class ot extends ke{}class nt extends ke{}class it extends ke{}class at extends ke{constructor(e,t){super(e,t),this.languageRegex=/^(>>\w+<<)\s*/g,this.supported_language_codes=this.model.vocab.filter((e=>this.languageRegex.test(e))),console.warn('WARNING: `MarianTokenizer` is not yet supported by Hugging Face\'s "fast" tokenizers library. Therefore, you may experience slightly inaccurate results.')}_encode_text(e){if(null===e)return null;const[t,...s]=e.trim().split(this.languageRegex);if(0===s.length)return super._encode_text(t);if(2===s.length){const[e,t]=s;return this.supported_language_codes.includes(e)||console.warn(`Unsupported language code "${e}" detected, which may lead to unexpected behavior. Should be one of: ${JSON.stringify(this.supported_language_codes)}`),(0,o.mergeArrays)([e],super._encode_text(t))}}}class lt extends ke{}class ct extends ke{}class dt extends ke{}class ut extends ke{}class _t extends ke{}class pt extends ke{constructor(e,t){super(e,t),this.decoder=new ue({})}}class mt extends ke{}class ht extends ke{}class gt{static TOKENIZER_CLASS_MAPPING={T5Tokenizer:Oe,DistilBertTokenizer:Le,CamembertTokenizer:Ie,DebertaTokenizer:Fe,DebertaV2Tokenizer:Ce,BertTokenizer:ye,HerbertTokenizer:Se,ConvBertTokenizer:Ae,RoFormerTokenizer:Ee,XLMTokenizer:ze,ElectraTokenizer:je,MobileBertTokenizer:Te,SqueezeBertTokenizer:Pe,AlbertTokenizer:ve,GPT2Tokenizer:De,BartTokenizer:Ve,MBartTokenizer:Ne,MBart50Tokenizer:Be,RobertaTokenizer:Re,WhisperTokenizer:rt,CodeGenTokenizer:ot,CLIPTokenizer:nt,SiglipTokenizer:it,MarianTokenizer:at,BloomTokenizer:Ge,NllbTokenizer:tt,M2M100Tokenizer:st,LlamaTokenizer:qe,CodeLlamaTokenizer:We,XLMRobertaTokenizer:Ue,MPNetTokenizer:Xe,FalconTokenizer:Qe,GPTNeoXTokenizer:He,EsmTokenizer:Je,Wav2Vec2CTCTokenizer:lt,BlenderbotTokenizer:ct,BlenderbotSmallTokenizer:dt,SpeechT5Tokenizer:ut,NougatTokenizer:_t,VitsTokenizer:pt,Qwen2Tokenizer:Ye,GemmaTokenizer:Ke,Grok1Tokenizer:Ze,CohereTokenizer:mt,MgpstrTokenizer:ht,PreTrainedTokenizer:ke};static async from_pretrained(e,{progress_callback:t=null,config:s=null,cache_dir:r=null,local_files_only:o=!1,revision:n="main",legacy:i=null}={}){const[a,l]=await u(e,{progress_callback:t,config:s,cache_dir:r,local_files_only:o,revision:n,legacy:i}),c=l.tokenizer_class?.replace(/Fast$/,"")??"PreTrainedTokenizer";let d=this.TOKENIZER_CLASS_MAPPING[c];return d||(console.warn(`Unknown tokenizer class "${c}", attempting to construct from base class.`),d=ke),new d(a,l)}}},"./src/utils/audio.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{RawAudio:()=>k,hamming:()=>_,hanning:()=>u,mel_filter_bank:()=>f,read_audio:()=>c,spectrogram:()=>M,window_function:()=>x});var r=s("./src/utils/hub.js"),o=s("./src/utils/maths.js"),n=s("./src/utils/core.js"),i=s("./src/env.js"),a=s("fs"),l=s("./src/utils/tensor.js");async function c(e,t){if("undefined"==typeof AudioContext)throw Error("Unable to load audio from path/URL since `AudioContext` is not available in your environment. Instead, audio data should be passed directly to the pipeline/processor. For more information and some example code, see https://huggingface.co/docs/transformers.js/guides/node-audio-processing.");const s=await(await(0,r.getFile)(e)).arrayBuffer(),o=new AudioContext({sampleRate:t});void 0===t&&console.warn(`No sampling rate provided, using default of ${o.sampleRate}Hz.`);const n=await o.decodeAudioData(s);let i;if(2===n.numberOfChannels){const e=Math.sqrt(2),t=n.getChannelData(0),s=n.getChannelData(1);i=new Float32Array(t.length);for(let r=0;r<n.length;++r)i[r]=e*(t[r]+s[r])/2}else i=n.getChannelData(0);return i}function d(e,t){if(e<1)return new Float64Array;if(1===e)return new Float64Array([1]);const s=1-t,r=2*Math.PI/(e-1),o=new Float64Array(e);for(let n=0;n<e;++n)o[n]=t-s*Math.cos(n*r);return o}function u(e){return d(e,.5)}function _(e){return d(e,.54)}const p={htk:e=>2595*Math.log10(1+e/700),kaldi:e=>1127*Math.log(1+e/700),slaney:(e,t=1e3,s=15,r=27/Math.log(6.4))=>e>=t?s+Math.log(e/t)*r:3*e/200};function m(e,t="htk"){const s=p[t];if(!s)throw new Error('mel_scale should be one of "htk", "slaney" or "kaldi".');return"number"==typeof e?s(e):e.map((e=>s(e)))}const h={htk:e=>700*(10**(e/2595)-1),kaldi:e=>700*(Math.exp(e/1127)-1),slaney:(e,t=1e3,s=15,r=Math.log(6.4)/27)=>e>=s?t*Math.exp(r*(e-s)):200*e/3};function g(e,t,s){const r=(t-e)/(s-1);return Float64Array.from({length:s},((t,s)=>e+r*s))}function f(e,t,s,r,o,n=null,i="htk",a=!1){if(null!==n&&"slaney"!==n)throw new Error('norm must be one of null or "slaney"');if(e<2)throw new Error(`Require num_frequency_bins: ${e} >= 2`);if(s>r)throw new Error(`Require min_frequency: ${s} <= max_frequency: ${r}`);const l=g(m(s,i),m(r,i),t+2);let c,d=function(e,t="htk"){const s=h[t];if(!s)throw new Error('mel_scale should be one of "htk", "slaney" or "kaldi".');return"number"==typeof e?s(e):e.map((e=>s(e)))}(l,i);if(a){const t=o/(2*(e-1));c=m(Float64Array.from({length:e},((e,s)=>s*t)),i),d=l}else c=g(0,Math.floor(o/2),e);const u=function(e,t){const s=Float64Array.from({length:t.length-1},((e,s)=>t[s+1]-t[s])),r=Array.from({length:e.length},(()=>new Array(t.length)));for(let s=0;s<e.length;++s){const o=r[s];for(let r=0;r<t.length;++r)o[r]=t[r]-e[s]}const o=t.length-2,n=Array.from({length:o},(()=>new Array(e.length)));for(let t=0;t<e.length;++t){const e=r[t];for(let r=0;r<o;++r){const o=-e[r]/s[r],i=e[r+2]/s[r+1];n[r][t]=Math.max(0,Math.min(o,i))}}return n}(c,d);if(null!==n&&"slaney"===n)for(let s=0;s<t;++s){const t=u[s],r=2/(d[s+2]-d[s]);for(let s=0;s<e;++s)t[s]*=r}return u}function w(e,t,s,r,n){if(s<=0)throw new Error("reference must be greater than zero");if(r<=0)throw new Error("min_value must be greater than zero");s=Math.max(r,s);const i=Math.log10(s);for(let s=0;s<e.length;++s)e[s]=t*Math.log10(Math.max(r,e[s])-i);if(null!==n){if(n<=0)throw new Error("db_range must be greater than zero");const t=(0,o.max)(e)[0]-n;for(let s=0;s<e.length;++s)e[s]=Math.max(e[s],t)}return e}async function M(e,t,s,r,{fft_length:i=null,power:a=1,center:c=!0,pad_mode:d="reflect",onesided:u=!0,preemphasis:_=null,mel_filters:p=null,mel_floor:m=1e-10,log_mel:h=null,reference:g=1,min_value:f=1e-10,db_range:M=null,remove_dc_offset:x=null,min_num_frames:b=null,max_num_frames:k=null,do_pad:y=!0,transpose:v=!1}={}){const T=t.length;if(null===i&&(i=s),s>i)throw Error(`frame_length (${s}) may not be larger than fft_length (${i})`);if(T!==s)throw new Error(`Length of the window (${T}) must equal frame_length (${s})`);if(r<=0)throw new Error("hop_length must be greater than zero");if(null===a&&null!==p)throw new Error("You have provided `mel_filters` but `power` is `None`. Mel spectrogram computation is not yet supported for complex-valued spectrogram. Specify `power` to fix this issue.");if(c){if("reflect"!==d)throw new Error(`pad_mode="${d}" not implemented yet.`);const t=Math.floor((i-1)/2)+1;e=function(e,t,s){const r=new e.constructor(e.length+t+s),o=e.length-1;for(let s=0;s<e.length;++s)r[t+s]=e[s];for(let s=1;s<=t;++s)r[t-s]=e[(0,n.calculateReflectOffset)(s,o)];for(let i=1;i<=s;++i)r[o+t+i]=e[(0,n.calculateReflectOffset)(o-i,o)];return r}(e,t,t)}let P=Math.floor(1+Math.floor((e.length-s)/r));null!==b&&P<b&&(P=b);const F=u?Math.floor(i/2)+1:i;let C=P,S=P;null!==k&&(k>P?y&&(S=k):S=C=k);const A=new o.FFT(i),E=new Float64Array(i),L=new Float64Array(A.outputBufferSize),I=new Float32Array(F*S);for(let o=0;o<C;++o){const n=o*r,i=Math.min(e.length-n,s);i!==s&&E.fill(0,0,s);for(let t=0;t<i;++t)E[t]=e[n+t];if(x){let e=0;for(let t=0;t<i;++t)e+=E[t];const t=e/i;for(let e=0;e<i;++e)E[e]-=t}if(null!==_){for(let e=i-1;e>=1;--e)E[e]-=_*E[e-1];E[0]*=1-_}for(let e=0;e<t.length;++e)E[e]*=t[e];A.realTransform(L,E);for(let e=0;e<F;++e){const t=e<<1;I[e*S+o]=L[t]**2+L[t+1]**2}}if(null!==a&&2!==a){const e=2/a;for(let t=0;t<I.length;++t)I[t]**=e}const z=p.length;let j=await(0,l.matmul)(new l.Tensor("float32",p.flat(),[z,F]),new l.Tensor("float32",I,[F,S]));v&&(j=j.transpose(1,0));const O=j.data;for(let e=0;e<O.length;++e)O[e]=Math.max(m,O[e]);if(null!==a&&null!==h){const e=Math.min(O.length,C*z);switch(h){case"log":for(let t=0;t<e;++t)O[t]=Math.log(O[t]);break;case"log10":for(let t=0;t<e;++t)O[t]=Math.log10(O[t]);break;case"dB":if(1===a)!function(e,t=1,s=1e-5,r=null){w(e,20,t,s,r)}(O,g,f,M);else{if(2!==a)throw new Error(`Cannot use log_mel option '${h}' with power ${a}`);!function(e,t=1,s=1e-10,r=null){w(e,10,t,s,r)}(O,g,f,M)}break;default:throw new Error(`log_mel must be one of null, 'log', 'log10' or 'dB'. Got '${h}'`)}}return j}function x(e,t,{periodic:s=!0,frame_length:r=null,center:o=!0}={}){const n=s?e+1:e;let i;switch(t){case"boxcar":i=new Float64Array(n).fill(1);break;case"hann":case"hann_window":i=u(n);break;case"hamming":i=_(n);break;case"povey":i=u(n).map((e=>Math.pow(e,.85)));break;default:throw new Error(`Unknown window type ${t}.`)}if(s&&(i=i.subarray(0,e)),null===r)return i;if(e>r)throw new Error(`Length of the window (${e}) may not be larger than frame_length (${r})`);return i}function b(e,t,s){for(let r=0;r<s.length;++r)e.setUint8(t+r,s.charCodeAt(r))}class k{constructor(e,t){this.audio=e,this.sampling_rate=t}toWav(){return function(e,t){let s=44;const r=new ArrayBuffer(s+4*e.length),o=new DataView(r);b(o,0,"RIFF"),o.setUint32(4,36+4*e.length,!0),b(o,8,"WAVE"),b(o,12,"fmt "),o.setUint32(16,16,!0),o.setUint16(20,3,!0),o.setUint16(22,1,!0),o.setUint32(24,t,!0),o.setUint32(28,4*t,!0),o.setUint16(32,4,!0),o.setUint16(34,32,!0),b(o,36,"data"),o.setUint32(40,4*e.length,!0);for(let t=0;t<e.length;++t,s+=4)o.setFloat32(s,e[t],!0);return r}(this.audio,this.sampling_rate)}toBlob(){const e=this.toWav();return new Blob([e],{type:"audio/wav"})}async save(e){let t;if(i.apis.IS_BROWSER_ENV){if(i.apis.IS_WEBWORKER_ENV)throw new Error("Unable to save a file from a Web Worker.");t=n.saveBlob}else{if(!i.apis.IS_FS_AVAILABLE)throw new Error("Unable to save because filesystem is disabled in this environment.");t=async(e,t)=>{let s=await t.arrayBuffer();a.writeFileSync(e,Buffer.from(s))}}await t(e,this.toBlob())}}},"./src/utils/constants.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{CHAT_TEMPLATE_NAME:()=>l,CONFIG_NAME:()=>o,FEATURE_EXTRACTOR_NAME:()=>n,GENERATION_CONFIG_NAME:()=>c,GITHUB_ISSUE_URL:()=>r,IMAGE_PROCESSOR_NAME:()=>i,PROCESSOR_NAME:()=>a});const r="https://github.com/huggingface/transformers.js/issues/new/choose",o="config.json",n="preprocessor_config.json",i=n,a="processor_config.json",l="chat_template.json",c="generation_config.json"},"./src/utils/core.js":(e,t,s)=>{"use strict";function r(e,t){e&&e(t)}function o(e){return Object.fromEntries(Object.entries(e).map((([e,t])=>[t,e])))}function n(e){return e.replace(/[.*+?^${}()|[\]\\]/g,"\\$&")}function i(e){return"TypedArray"===e?.prototype?.__proto__?.constructor?.name}function a(e){return Number.isInteger(e)||"bigint"==typeof e}function l(e){return null==e||-1===e}function c(e){const t=[];let s=e;for(;Array.isArray(s);)t.push(s.length),s=s[0];return t}function d(e,t,s=void 0){const r=e[t];if(void 0!==r)return delete e[t],r;if(void 0===s)throw Error(`Key ${t} does not exist in object.`);return s}function u(...e){return Array.prototype.concat.apply([],e)}function _(...e){return e.reduce(((e,t)=>e.flatMap((e=>t.map((t=>[e,t]))))))}function p(e,t){return Math.abs((e+t)%(2*t)-t)}function m(e,t){const s=URL.createObjectURL(t),r=document.createElement("a");r.href=s,r.download=e,r.click(),r.remove(),URL.revokeObjectURL(s)}function h(e,t){return Object.assign({},...t.map((t=>{if(void 0!==e[t])return{[t]:e[t]}})))}function g(e){let t=0;for(const s of e)++t;return t}function f(e,t){let s=0;for(const r of e)r===t&&++s;return s}s.r(t),s.d(t,{calculateDimensions:()=>c,calculateReflectOffset:()=>p,count:()=>f,dispatchCallback:()=>r,escapeRegExp:()=>n,isIntegralNumber:()=>a,isNullishDimension:()=>l,isTypedArray:()=>i,len:()=>g,mergeArrays:()=>u,pick:()=>h,pop:()=>d,product:()=>_,reverseDictionary:()=>o,saveBlob:()=>m})},"./src/utils/data-structures.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{CharTrie:()=>o,DictionarySplitter:()=>l,LRUCache:()=>c,PriorityQueue:()=>r,TokenLattice:()=>i});class r{constructor(e=(e,t)=>e>t,t=1/0){this._heap=[],this._comparator=e,this._maxSize=t}get size(){return this._heap.length}isEmpty(){return 0===this.size}peek(){return this._heap[0]}push(...e){return this.extend(e)}extend(e){for(const t of e)if(this.size<this._maxSize)this._heap.push(t),this._siftUp();else{const e=this._smallest();this._comparator(t,this._heap[e])&&(this._heap[e]=t,this._siftUpFrom(e))}return this.size}pop(){const e=this.peek(),t=this.size-1;return t>0&&this._swap(0,t),this._heap.pop(),this._siftDown(),e}replace(e){const t=this.peek();return this._heap[0]=e,this._siftDown(),t}_parent(e){return(e+1>>>1)-1}_left(e){return 1+(e<<1)}_right(e){return e+1<<1}_greater(e,t){return this._comparator(this._heap[e],this._heap[t])}_swap(e,t){const s=this._heap[e];this._heap[e]=this._heap[t],this._heap[t]=s}_siftUp(){this._siftUpFrom(this.size-1)}_siftUpFrom(e){for(;e>0&&this._greater(e,this._parent(e));)this._swap(e,this._parent(e)),e=this._parent(e)}_siftDown(){let e=0;for(;this._left(e)<this.size&&this._greater(this._left(e),e)||this._right(e)<this.size&&this._greater(this._right(e),e);){const t=this._right(e)<this.size&&this._greater(this._right(e),this._left(e))?this._right(e):this._left(e);this._swap(e,t),e=t}}_smallest(){return 2**Math.floor(Math.log2(this.size))-1}}class o{constructor(){this.root=n.default()}extend(e){for(const t of e)this.push(t)}push(e){let t=this.root;for(const s of e){let e=t.children.get(s);void 0===e&&(e=n.default(),t.children.set(s,e)),t=e}t.isLeaf=!0}*commonPrefixSearch(e){let t=this.root;if(void 0===t)return;let s="";for(const r of e){if(s+=r,t=t.children.get(r),void 0===t)return;t.isLeaf&&(yield s)}}}class n{constructor(e,t){this.isLeaf=e,this.children=t}static default(){return new n(!1,new Map)}}class i{constructor(e,t,s){this.chars=Array.from(e),this.len=this.chars.length,this.bosTokenId=t,this.eosTokenId=s,this.nodes=[],this.beginNodes=Array.from({length:this.len+1},(()=>[])),this.endNodes=Array.from({length:this.len+1},(()=>[]));const r=new a(this.bosTokenId,0,0,0,0),o=new a(this.eosTokenId,1,this.len,0,0);this.nodes.push(r.clone()),this.nodes.push(o.clone()),this.beginNodes[this.len].push(o),this.endNodes[0].push(r)}insert(e,t,s,r){const o=this.nodes.length,n=new a(r,o,e,t,s);this.beginNodes[e].push(n),this.endNodes[e+t].push(n),this.nodes.push(n)}viterbi(){const e=this.len;let t=0;for(;t<=e;){if(0==this.beginNodes[t].length)return[];for(let e of this.beginNodes[t]){e.prev=null;let s=0,r=null;for(let o of this.endNodes[t]){const t=o.backtraceScore+e.score;(null===r||t>s)&&(r=o.clone(),s=t)}if(null===r)return[];e.prev=r,e.backtraceScore=s}++t}const s=[],r=this.beginNodes[e][0].prev;if(null===r)return[];let o=r.clone();for(;null!==o.prev;){s.push(o.clone());const e=o.clone();o=e.prev.clone()}return s.reverse(),s}piece(e){return this.chars.slice(e.pos,e.pos+e.length).join("")}tokens(){return this.viterbi().map((e=>this.piece(e)))}tokenIds(){return this.viterbi().map((e=>e.tokenId))}}class a{constructor(e,t,s,r,o){this.tokenId=e,this.nodeId=t,this.pos=s,this.length=r,this.score=o,this.prev=null,this.backtraceScore=0}clone(){const e=new a(this.tokenId,this.nodeId,this.pos,this.length,this.score);return e.prev=this.prev,e.backtraceScore=this.backtraceScore,e}}class l{constructor(e){this.trie=this._buildTrie(e)}_buildTrie(e){const t=Object.create(null);for(const s of e){let e=t;for(let t=0;t<s.length;++t)e=e[s[t]]??=Object.create(null);e.end=s}return t}split(e){const t=[],s=e.length;let r=0,o=0;for(;o<s;){let n=this.trie,i=null,a=o;for(;a<s&&(n=n[e[a]]);)n.end&&(i=n.end),++a;i?(o>r&&t.push(e.slice(r,o)),t.push(i),o+=i.length,r=o):++o}return r<s&&t.push(e.slice(r)),t}}class c{constructor(e){this.capacity=e,this.cache=new Map}get(e){if(!this.cache.has(e))return;const t=this.cache.get(e);return this.cache.delete(e),this.cache.set(e,t),t}put(e,t){this.cache.has(e)&&this.cache.delete(e),this.cache.set(e,t),this.cache.size>this.capacity&&this.cache.delete(this.cache.keys().next().value)}clear(){this.cache.clear()}}},"./src/utils/devices.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DEVICE_TYPES:()=>r});const r=Object.freeze({auto:"auto",gpu:"gpu",cpu:"cpu",wasm:"wasm",webgpu:"webgpu",cuda:"cuda",dml:"dml",webnn:"webnn","webnn-npu":"webnn-npu","webnn-gpu":"webnn-gpu","webnn-cpu":"webnn-cpu"})},"./src/utils/dtypes.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DATA_TYPES:()=>i,DEFAULT_DEVICE_DTYPE_MAPPING:()=>a,DEFAULT_DTYPE_SUFFIX_MAPPING:()=>l,isWebGpuFp16Supported:()=>n});var r=s("./src/env.js"),o=s("./src/utils/devices.js");const n=function(){let e;return async function(){if(void 0===e)if(r.apis.IS_WEBGPU_AVAILABLE)try{const t=await navigator.gpu.requestAdapter();e=t.features.has("shader-f16")}catch(t){e=!1}else e=!1;return e}}(),i=Object.freeze({auto:"auto",fp32:"fp32",fp16:"fp16",q8:"q8",int8:"int8",uint8:"uint8",q4:"q4",bnb4:"bnb4",q4f16:"q4f16"}),a=Object.freeze({[o.DEVICE_TYPES.wasm]:i.q8}),l=Object.freeze({[i.fp32]:"",[i.fp16]:"_fp16",[i.int8]:"_int8",[i.uint8]:"_uint8",[i.q8]:"_quantized",[i.q4]:"_q4",[i.q4f16]:"_q4f16",[i.bnb4]:"_bnb4"})},"./src/utils/generic.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{Callable:()=>r});const r=class{constructor(){let e=function(...t){return e._call(...t)};return Object.setPrototypeOf(e,new.target.prototype)}_call(...e){throw Error("Must implement _call method in subclass")}}},"./src/utils/hub.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{MAX_EXTERNAL_DATA_CHUNKS:()=>a,getFile:()=>_,getModelFile:()=>h,getModelJSON:()=>g});var r=s("fs"),o=s("path"),n=s("./src/env.js"),i=s("./src/utils/core.js");const a=100,l={txt:"text/plain",html:"text/html",css:"text/css",js:"text/javascript",json:"application/json",png:"image/png",jpg:"image/jpeg",jpeg:"image/jpeg",gif:"image/gif"};class c{constructor(e){if(this.filePath=e,this.headers=new Headers,this.exists=r.existsSync(e),this.exists){this.status=200,this.statusText="OK";let t=r.statSync(e);this.headers.set("content-length",t.size.toString()),this.updateContentType();const s=r.createReadStream(e);this.body=new ReadableStream({start(e){s.on("data",(t=>e.enqueue(t))),s.on("end",(()=>e.close())),s.on("error",(t=>e.error(t)))},cancel(){s.destroy()}})}else this.status=404,this.statusText="Not Found",this.body=null}updateContentType(){const e=this.filePath.toString().split(".").pop().toLowerCase();this.headers.set("content-type",l[e]??"application/octet-stream")}clone(){let e=new c(this.filePath);return e.exists=this.exists,e.status=this.status,e.statusText=this.statusText,e.headers=new Headers(this.headers),e}async arrayBuffer(){return(await r.promises.readFile(this.filePath)).buffer}async blob(){const e=await r.promises.readFile(this.filePath);return new Blob([e],{type:this.headers.get("content-type")})}async text(){return await r.promises.readFile(this.filePath,"utf8")}async json(){return JSON.parse(await this.text())}}function d(e,t=null,s=null){let r;try{r=new URL(e)}catch(e){return!1}return!(t&&!t.includes(r.protocol))&&!(s&&!s.includes(r.hostname))}const u=/^(\b[\w\-.]+\b\/)?\b[\w\-.]{1,96}\b$/;async function _(e){if(n.env.useFS&&!d(e,["http:","https:","blob:"]))return new c(e instanceof URL?"file:"===e.protocol?e.pathname:e.toString():e);if("undefined"!=typeof process&&"node"===process?.release?.name){const t=!!process.env?.TESTING_REMOTELY,s=n.env.version,r=new Headers;r.set("User-Agent",`transformers.js/${s}; is_ci/${t};`);if(d(e,["http:","https:"],["huggingface.co","hf.co"])){const e=process.env?.HF_TOKEN??process.env?.HF_ACCESS_TOKEN;e&&r.set("Authorization",`Bearer ${e}`)}return fetch(e,{headers:r})}return fetch(e)}const p={400:"Bad request error occurred while trying to load file",401:"Unauthorized access to file",403:"Forbidden access to file",404:"Could not locate file",408:"Request timeout error occurred while trying to load file",500:"Internal server error error occurred while trying to load file",502:"Bad gateway error occurred while trying to load file",503:"Service unavailable error occurred while trying to load file",504:"Gateway timeout error occurred while trying to load file"};class m{constructor(e){this.path=e}async match(e){let t=o.join(this.path,e),s=new c(t);return s.exists?s:void 0}async put(e,t,s=void 0){let n=o.join(this.path,e);try{const e=t.headers.get("Content-Length"),i=parseInt(e??"0");let a=0;await r.promises.mkdir(o.dirname(n),{recursive:!0});const l=r.createWriteStream(n),c=t.body.getReader();for(;;){const{done:e,value:t}=await c.read();if(e)break;await new Promise(((e,s)=>{l.write(t,(t=>{t?s(t):e()}))})),a+=t.length;const r=i?a/i*100:0;s?.({progress:r,loaded:a,total:i})}l.close()}catch(e){try{await r.promises.unlink(n)}catch{}throw e}}}async function h(e,t,s=!0,r={},o=!1){if(!n.env.allowLocalModels){if(r.local_files_only)throw Error("Invalid configuration detected: local models are disabled (`env.allowLocalModels=false`) but you have requested to only use local models (`local_files_only=true`).");if(!n.env.allowRemoteModels)throw Error("Invalid configuration detected: both local and remote models are disabled. Fix by setting `env.allowLocalModels` or `env.allowRemoteModels` to `true`.")}let a;if((0,i.dispatchCallback)(r.progress_callback,{status:"initiate",name:e,file:t}),!a&&n.env.useCustomCache){if(!n.env.customCache)throw Error("`env.useCustomCache=true`, but `env.customCache` is not defined.");if(!n.env.customCache.match||!n.env.customCache.put)throw new Error("`env.customCache` must be an object which implements the `match` and `put` functions of the Web Cache API. For more information, see https://developer.mozilla.org/en-US/docs/Web/API/Cache");a=n.env.customCache}if(!a&&n.env.useBrowserCache){if("undefined"==typeof caches)throw Error("Browser cache is not available in this environment.");try{a=await caches.open("transformers-cache")}catch(e){console.warn("An error occurred while opening the browser cache:",e)}}if(!a&&n.env.useFSCache){if(!n.apis.IS_FS_AVAILABLE)throw Error("File System Cache is not available in this environment.");a=new m(r.cache_dir??n.env.cacheDir)}const l=r.revision??"main",h=f(e,t),g=(w=e,!(!u.test(w)||w.includes("..")||w.includes("--")||w.endsWith(".git")||w.endsWith(".ipynb")));var w;const M=g?f(n.env.localModelPath,h):h,x=f(n.env.remoteHost,n.env.remotePathTemplate.replaceAll("{model}",e).replaceAll("{revision}",encodeURIComponent(l)),t);let b;const k=a instanceof m?"main"===l?h:f(e,l,t):x;let y,v=!1;a&&(y=await async function(e,...t){for(let s of t)try{let t=await e.match(s);if(t)return t}catch(e){continue}}(a,M,k));const T=void 0!==y;if(void 0===y){if(n.env.allowLocalModels){if(d(h,["http:","https:"])){if(r.local_files_only)throw new Error(`\`local_files_only=true\`, but attempted to load a remote file from: ${h}.`);if(!n.env.allowRemoteModels)throw new Error(`\`env.allowRemoteModels=false\`, but attempted to load a remote file from: ${h}.`)}else try{y=await _(M),b=M}catch(e){console.warn(`Unable to load from local path "${M}": "${e}"`)}}if(void 0===y||404===y.status){if(r.local_files_only||!n.env.allowRemoteModels){if(s)throw Error(`\`local_files_only=true\` or \`env.allowRemoteModels=false\` and file was not found locally at "${M}".`);return null}if(!g)throw Error(`Local file missing at "${M}" and download aborted due to invalid model ID "${e}".`);if(y=await _(x),200!==y.status)return function(e,t,s){if(!s)return null;const r=p[e]??`Error (${e}) occurred while trying to load file`;throw Error(`${r}: "${t}".`)}(y.status,x,s);b=k}v=a&&"undefined"!=typeof Response&&y instanceof Response&&200===y.status}let P;if((0,i.dispatchCallback)(r.progress_callback,{status:"download",name:e,file:t}),!n.apis.IS_NODE_ENV||!o){let s;r.progress_callback?T&&"undefined"!=typeof navigator&&/firefox/i.test(navigator.userAgent)?(s=new Uint8Array(await y.arrayBuffer()),(0,i.dispatchCallback)(r.progress_callback,{status:"progress",name:e,file:t,progress:100,loaded:s.length,total:s.length})):s=await async function(e,t){const s=e.headers.get("Content-Length");null===s&&console.warn("Unable to determine content-length from response headers. Will expand buffer when needed.");let r=parseInt(s??"0"),o=new Uint8Array(r),n=0;const i=e.body.getReader();async function a(){const{done:e,value:s}=await i.read();if(e)return;const l=n+s.length;if(l>r){r=l;const e=new Uint8Array(r);e.set(o),o=e}o.set(s,n),n=l;return t({progress:n/r*100,loaded:n,total:r}),a()}return await a(),o}(y,(s=>{(0,i.dispatchCallback)(r.progress_callback,{status:"progress",name:e,file:t,...s})})):s=new Uint8Array(await y.arrayBuffer()),P=s}if(v&&b&&void 0===await a.match(b)&&(P?await a.put(b,new Response(P,{headers:y.headers})).catch((e=>{console.warn(`Unable to add response to browser cache: ${e}.`)})):await a.put(b,y,r.progress_callback)),(0,i.dispatchCallback)(r.progress_callback,{status:"done",name:e,file:t}),P){if(!n.apis.IS_NODE_ENV&&o)throw new Error("Cannot return path in a browser environment.");return P}if(y instanceof c)return y.filePath;const F=await(a?.match(b));if(F instanceof c)return F.filePath;if(F instanceof Response)return new Uint8Array(await F.arrayBuffer());if("string"==typeof F)return F;throw new Error("Unable to get model file path or buffer.")}async function g(e,t,s=!0,r={}){const o=await h(e,t,s,r,!1);if(null===o)return{};const n=new TextDecoder("utf-8").decode(o);return JSON.parse(n)}function f(...e){return(e=e.map(((t,s)=>(s&&(t=t.replace(new RegExp("^/"),"")),s!==e.length-1&&(t=t.replace(new RegExp("/$"),"")),t)))).join("/")}},"./src/utils/image.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{RawImage:()=>m,load_image:()=>h});var r=s("./src/utils/core.js"),o=s("./src/utils/hub.js"),n=s("./src/env.js"),i=s("./src/utils/tensor.js"),a=s("sharp");let l,c,d;const u=n.apis.IS_BROWSER_ENV||n.apis.IS_WEBWORKER_ENV;if(u)l=(e,t)=>{if(!self.OffscreenCanvas)throw new Error("OffscreenCanvas not supported by this browser.");return new self.OffscreenCanvas(e,t)},d=self.createImageBitmap,c=self.ImageData;else{if(!a)throw new Error("Unable to load image processing library.");d=async e=>{const t=(await e.metadata()).channels,{data:s,info:r}=await e.rotate().raw().toBuffer({resolveWithObject:!0}),o=new m(new Uint8ClampedArray(s),r.width,r.height,r.channels);return void 0!==t&&t!==r.channels&&o.convert(t),o}}const _={0:"nearest",1:"lanczos",2:"bilinear",3:"bicubic",4:"box",5:"hamming"},p=new Map([["png","image/png"],["jpg","image/jpeg"],["jpeg","image/jpeg"],["gif","image/gif"]]);class m{constructor(e,t,s,r){this.data=e,this.width=t,this.height=s,this.channels=r}get size(){return[this.width,this.height]}static async read(e){if(e instanceof m)return e;if("string"==typeof e||e instanceof URL)return await this.fromURL(e);if(e instanceof Blob)return await this.fromBlob(e);if("undefined"!=typeof HTMLCanvasElement&&e instanceof HTMLCanvasElement||"undefined"!=typeof OffscreenCanvas&&e instanceof OffscreenCanvas)return this.fromCanvas(e);throw new Error("Unsupported input type: "+typeof e)}static fromCanvas(e){if(!u)throw new Error("fromCanvas() is only supported in browser environments.");const t=e.getContext("2d").getImageData(0,0,e.width,e.height).data;return new m(t,e.width,e.height,4)}static async fromURL(e){const t=await(0,o.getFile)(e);if(200!==t.status)throw new Error(`Unable to read image from "${e}" (${t.status} ${t.statusText})`);const s=await t.blob();return this.fromBlob(s)}static async fromBlob(e){if(u){const t=await d(e),s=l(t.width,t.height).getContext("2d");return s.drawImage(t,0,0),new this(s.getImageData(0,0,t.width,t.height).data,t.width,t.height,4)}{const t=a(await e.arrayBuffer());return await d(t)}}static fromTensor(e,t="CHW"){if(3!==e.dims.length)throw new Error(`Tensor should have 3 dimensions, but has ${e.dims.length} dimensions.`);if("CHW"===t)e=e.transpose(1,2,0);else if("HWC"!==t)throw new Error(`Unsupported channel format: ${t}`);if(!(e.data instanceof Uint8ClampedArray||e.data instanceof Uint8Array))throw new Error(`Unsupported tensor type: ${e.type}`);switch(e.dims[2]){case 1:case 2:case 3:case 4:return new m(e.data,e.dims[1],e.dims[0],e.dims[2]);default:throw new Error(`Unsupported number of channels: ${e.dims[2]}`)}}grayscale(){if(1===this.channels)return this;const e=new Uint8ClampedArray(this.width*this.height*1);switch(this.channels){case 3:case 4:for(let t=0,s=0;t<this.data.length;t+=this.channels){const r=this.data[t],o=this.data[t+1],n=this.data[t+2];e[s++]=Math.round(.2989*r+.587*o+.114*n)}break;default:throw new Error(`Conversion failed due to unsupported number of channels: ${this.channels}`)}return this._update(e,this.width,this.height,1)}rgb(){if(3===this.channels)return this;const e=new Uint8ClampedArray(this.width*this.height*3);switch(this.channels){case 1:for(let t=0,s=0;t<this.data.length;++t)e[s++]=this.data[t],e[s++]=this.data[t],e[s++]=this.data[t];break;case 4:for(let t=0,s=0;t<this.data.length;t+=4)e[s++]=this.data[t],e[s++]=this.data[t+1],e[s++]=this.data[t+2];break;default:throw new Error(`Conversion failed due to unsupported number of channels: ${this.channels}`)}return this._update(e,this.width,this.height,3)}rgba(){if(4===this.channels)return this;const e=new Uint8ClampedArray(this.width*this.height*4);switch(this.channels){case 1:for(let t=0,s=0;t<this.data.length;++t)e[s++]=this.data[t],e[s++]=this.data[t],e[s++]=this.data[t],e[s++]=255;break;case 3:for(let t=0,s=0;t<this.data.length;t+=3)e[s++]=this.data[t],e[s++]=this.data[t+1],e[s++]=this.data[t+2],e[s++]=255;break;default:throw new Error(`Conversion failed due to unsupported number of channels: ${this.channels}`)}return this._update(e,this.width,this.height,4)}putAlpha(e){if(e.width!==this.width||e.height!==this.height)throw new Error(`Expected mask size to be ${this.width}x${this.height}, but got ${e.width}x${e.height}`);if(1!==e.channels)throw new Error(`Expected mask to have 1 channel, but got ${e.channels}`);const t=this.data,s=e.data,r=this.width*this.height;if(3===this.channels){const e=new Uint8ClampedArray(4*r);for(let o=0,n=0,i=0;o<r;++o)e[i++]=t[n++],e[i++]=t[n++],e[i++]=t[n++],e[i++]=s[o];return this._update(e,this.width,this.height,4)}if(4===this.channels){for(let e=0;e<r;++e)t[4*e+3]=s[e];return this}throw new Error(`Expected image to have 3 or 4 channels, but got ${this.channels}`)}async resize(e,t,{resample:s=2}={}){if(this.width===e&&this.height===t)return this;let o=_[s]??s;const n=(0,r.isNullishDimension)(e),i=(0,r.isNullishDimension)(t);if(n&&i)return this;if(n?e=t/this.height*this.width:i&&(t=e/this.width*this.height),u){const s=this.channels,r=this.toCanvas(),o=l(e,t).getContext("2d");o.drawImage(r,0,0,e,t);return new m(o.getImageData(0,0,e,t).data,e,t,4).convert(s)}{let s=this.toSharp();switch(o){case"box":case"hamming":"box"!==o&&"hamming"!==o||(console.warn(`Resampling method ${o} is not yet supported. Using bilinear instead.`),o="bilinear");case"nearest":case"bilinear":case"bicubic":s=s.affine([e/this.width,0,0,t/this.height],{interpolator:o});break;case"lanczos":s=s.resize({width:e,height:t,fit:"fill",kernel:"lanczos3"});break;default:throw new Error(`Resampling method ${o} is not supported.`)}return await d(s)}}async pad([e,t,s,r]){if(e=Math.max(e,0),t=Math.max(t,0),s=Math.max(s,0),r=Math.max(r,0),0===e&&0===t&&0===s&&0===r)return this;if(u){const o=this.channels,n=this.toCanvas(),i=this.width+e+t,a=this.height+s+r,c=l(i,a).getContext("2d");c.drawImage(n,0,0,this.width,this.height,e,s,this.width,this.height);return new m(c.getImageData(0,0,i,a).data,i,a,4).convert(o)}{const o=this.toSharp().extend({left:e,right:t,top:s,bottom:r});return await d(o)}}async crop([e,t,s,r]){if(e=Math.max(e,0),t=Math.max(t,0),s=Math.min(s,this.width-1),r=Math.min(r,this.height-1),0===e&&0===t&&s===this.width-1&&r===this.height-1)return this;const o=s-e+1,n=r-t+1;if(u){const s=this.channels,r=this.toCanvas(),i=l(o,n).getContext("2d");i.drawImage(r,e,t,o,n,0,0,o,n);return new m(i.getImageData(0,0,o,n).data,o,n,4).convert(s)}{const s=this.toSharp().extract({left:e,top:t,width:o,height:n});return await d(s)}}async center_crop(e,t){if(this.width===e&&this.height===t)return this;const s=(this.width-e)/2,r=(this.height-t)/2;if(u){const o=this.channels,n=this.toCanvas(),i=l(e,t).getContext("2d");let a=0,c=0,d=0,u=0;s>=0?a=s:d=-s,r>=0?c=r:u=-r,i.drawImage(n,a,c,e,t,d,u,e,t);return new m(i.getImageData(0,0,e,t).data,e,t,4).convert(o)}{let o=this.toSharp();if(s>=0&&r>=0)o=o.extract({left:Math.floor(s),top:Math.floor(r),width:e,height:t});else if(s<=0&&r<=0){const n=Math.floor(-r),i=Math.floor(-s);o=o.extend({top:n,left:i,right:e-this.width-i,bottom:t-this.height-n})}else{let n=[0,0],i=0;r<0?(n[0]=Math.floor(-r),n[1]=t-this.height-n[0]):i=Math.floor(r);let a=[0,0],l=0;s<0?(a[0]=Math.floor(-s),a[1]=e-this.width-a[0]):l=Math.floor(s),o=o.extend({top:n[0],bottom:n[1],left:a[0],right:a[1]}).extract({left:l,top:i,width:e,height:t})}return await d(o)}}async toBlob(e="image/png",t=1){if(!u)throw new Error("toBlob() is only supported in browser environments.");const s=this.toCanvas();return await s.convertToBlob({type:e,quality:t})}toTensor(e="CHW"){let t=new i.Tensor("uint8",new Uint8Array(this.data),[this.height,this.width,this.channels]);if("HWC"===e);else{if("CHW"!==e)throw new Error(`Unsupported channel format: ${e}`);t=t.permute(2,0,1)}return t}toCanvas(){if(!u)throw new Error("toCanvas() is only supported in browser environments.");const e=this.clone().rgba(),t=l(e.width,e.height),s=new c(e.data,e.width,e.height);return t.getContext("2d").putImageData(s,0,0),t}split(){const{data:e,width:t,height:s,channels:r}=this,o=e.constructor,n=e.length/r,i=Array.from({length:r},(()=>new o(n)));for(let t=0;t<n;++t){const s=r*t;for(let o=0;o<r;++o)i[o][t]=e[s+o]}return i.map((e=>new m(e,t,s,1)))}_update(e,t,s,r=null){return this.data=e,this.width=t,this.height=s,null!==r&&(this.channels=r),this}clone(){return new m(this.data.slice(),this.width,this.height,this.channels)}convert(e){if(this.channels===e)return this;switch(e){case 1:this.grayscale();break;case 3:this.rgb();break;case 4:this.rgba();break;default:throw new Error(`Conversion failed due to unsupported number of channels: ${this.channels}`)}return this}async save(e){if(!u){if(n.apis.IS_FS_AVAILABLE){const t=this.toSharp();return await t.toFile(e)}throw new Error("Unable to save the image because filesystem is disabled in this environment.")}{if(n.apis.IS_WEBWORKER_ENV)throw new Error("Unable to save an image from a Web Worker.");const t=e.split(".").pop().toLowerCase(),s=p.get(t)??"image/png",o=await this.toBlob(s);(0,r.saveBlob)(e,o)}}toSharp(){if(u)throw new Error("toSharp() is only supported in server-side environments.");return a(this.data,{raw:{width:this.width,height:this.height,channels:this.channels}})}}const h=m.read.bind(m)},"./src/utils/maths.js":(e,t,s)=>{"use strict";function r(e,[t,s,r],[o,n],i="bilinear",a=!1){const l=n/r,c=o/s,d=new e.constructor(o*n*t),u=s*r,_=o*n;for(let i=0;i<o;++i)for(let o=0;o<n;++o){const a=i*n+o,p=(o+.5)/l-.5,m=(i+.5)/c-.5;let h=Math.floor(p),g=Math.floor(m);const f=Math.min(h+1,r-1),w=Math.min(g+1,s-1);h=Math.max(h,0),g=Math.max(g,0);const M=p-h,x=m-g,b=(1-M)*(1-x),k=M*(1-x),y=(1-M)*x,v=M*x,T=g*r,P=w*r,F=T+h,C=T+f,S=P+h,A=P+f;for(let s=0;s<t;++s){const t=s*u;d[s*_+a]=b*e[t+F]+k*e[t+C]+y*e[t+S]+v*e[t+A]}}return d}function o(e,t,s){const r=new Array(s.length),o=new Array(s.length);for(let e=s.length-1,n=1;e>=0;--e)o[e]=n,r[e]=t[s[e]],n*=r[e];const n=s.map(((e,t)=>o[s.indexOf(t)])),i=new e.constructor(e.length);for(let s=0;s<e.length;++s){let r=0;for(let e=t.length-1,o=s;e>=0;--e)r+=o%t[e]*n[e],o=Math.floor(o/t[e]);i[r]=e[s]}return[i,r]}function n(e){const t=u(e)[0],s=e.map((e=>Math.exp(e-t))),r=s.reduce(((e,t)=>e+t),0);return s.map((e=>e/r))}function i(e){const t=u(e)[0];let s=0;for(let r=0;r<e.length;++r)s+=Math.exp(e[r]-t);const r=Math.log(s);return e.map((e=>e-t-r))}function a(e,t){let s=0;for(let r=0;r<e.length;++r)s+=e[r]*t[r];return s}function l(e,t){return a(e,t)/(c(e)*c(t))}function c(e){return Math.sqrt(e.reduce(((e,t)=>e+t*t),0))}function d(e){if(0===e.length)throw Error("Array must not be empty");let t=e[0],s=0;for(let r=1;r<e.length;++r)e[r]<t&&(t=e[r],s=r);return[t,s]}function u(e){if(0===e.length)throw Error("Array must not be empty");let t=e[0],s=0;for(let r=1;r<e.length;++r)e[r]>t&&(t=e[r],s=r);return[t,s]}function _(e){return e>0&&!(e&e-1)}s.r(t),s.d(t,{FFT:()=>h,bankers_round:()=>w,cos_sim:()=>l,dot:()=>a,dynamic_time_warping:()=>M,interpolate_data:()=>r,log_softmax:()=>i,magnitude:()=>c,max:()=>u,medianFilter:()=>g,min:()=>d,permute_data:()=>o,round:()=>f,softmax:()=>n});class p{constructor(e){if(this.size=0|e,this.size<=1||!_(this.size))throw new Error("FFT size must be a power of two larger than 1");this._csize=e<<1,this.table=new Float64Array(2*this.size);for(let e=0;e<this.table.length;e+=2){const t=Math.PI*e/this.size;this.table[e]=Math.cos(t),this.table[e+1]=-Math.sin(t)}let t=0;for(let e=1;this.size>e;e<<=1)++t;this._width=t%2==0?t-1:t,this._bitrev=new Int32Array(1<<this._width);for(let e=0;e<this._bitrev.length;++e){this._bitrev[e]=0;for(let t=0;t<this._width;t+=2){const s=this._width-t-2;this._bitrev[e]|=(e>>>t&3)<<s}}}createComplexArray(){return new Float64Array(this._csize)}fromComplexArray(e,t){const s=t||new Array(e.length>>>1);for(let t=0;t<e.length;t+=2)s[t>>>1]=e[t];return s}toComplexArray(e,t){const s=t||this.createComplexArray();for(let t=0;t<s.length;t+=2)s[t]=e[t>>>1],s[t+1]=0;return s}transform(e,t){if(e===t)throw new Error("Input and output buffers must be different");this._transform4(e,t,1)}realTransform(e,t){if(e===t)throw new Error("Input and output buffers must be different");this._realTransform4(e,t,1)}inverseTransform(e,t){if(e===t)throw new Error("Input and output buffers must be different");this._transform4(e,t,-1);for(let t=0;t<e.length;++t)e[t]/=this.size}_transform4(e,t,s){const r=this._csize;let o,n,i=1<<this._width,a=r/i<<1;const l=this._bitrev;if(4===a)for(o=0,n=0;o<r;o+=a,++n){const s=l[n];this._singleTransform2(t,e,o,s,i)}else for(o=0,n=0;o<r;o+=a,++n){const r=l[n];this._singleTransform4(t,e,o,r,i,s)}const c=this.table;for(i>>=2;i>=2;i>>=2){a=r/i<<1;const t=a>>>2;for(o=0;o<r;o+=a){const r=o+t-1;for(let n=o,a=0;n<r;n+=2,a+=i){const r=n,o=r+t,i=o+t,l=i+t,d=e[r],u=e[r+1],_=e[o],p=e[o+1],m=e[i],h=e[i+1],g=e[l],f=e[l+1],w=c[a],M=s*c[a+1],x=_*w-p*M,b=_*M+p*w,k=c[2*a],y=s*c[2*a+1],v=m*k-h*y,T=m*y+h*k,P=c[3*a],F=s*c[3*a+1],C=g*P-f*F,S=g*F+f*P,A=d+v,E=u+T,L=d-v,I=u-T,z=x+C,j=b+S,O=s*(x-C),D=s*(b-S);e[r]=A+z,e[r+1]=E+j,e[o]=L+D,e[o+1]=I-O,e[i]=A-z,e[i+1]=E-j,e[l]=L-D,e[l+1]=I+O}}}}_singleTransform2(e,t,s,r,o){const n=e[r],i=e[r+1],a=e[r+o],l=e[r+o+1];t[s]=n+a,t[s+1]=i+l,t[s+2]=n-a,t[s+3]=i-l}_singleTransform4(e,t,s,r,o,n){const i=2*o,a=3*o,l=e[r],c=e[r+1],d=e[r+o],u=e[r+o+1],_=e[r+i],p=e[r+i+1],m=e[r+a],h=e[r+a+1],g=l+_,f=c+p,w=l-_,M=c-p,x=d+m,b=u+h,k=n*(d-m),y=n*(u-h);t[s]=g+x,t[s+1]=f+b,t[s+2]=w+y,t[s+3]=M-k,t[s+4]=g-x,t[s+5]=f-b,t[s+6]=w-y,t[s+7]=M+k}_realTransform4(e,t,s){const r=this._csize;let o,n,i=1<<this._width,a=r/i<<1;const l=this._bitrev;if(4===a)for(o=0,n=0;o<r;o+=a,++n){const s=l[n];this._singleRealTransform2(t,e,o,s>>>1,i>>>1)}else for(o=0,n=0;o<r;o+=a,++n){const r=l[n];this._singleRealTransform4(t,e,o,r>>>1,i>>>1,s)}const c=this.table;for(i>>=2;i>=2;i>>=2){a=r/i<<1;const t=a>>>1,n=t>>>1,l=n>>>1;for(o=0;o<r;o+=a)for(let r=0,a=0;r<=l;r+=2,a+=i){const i=o+r,d=i+n,u=d+n,_=u+n,p=e[i],m=e[i+1],h=e[d],g=e[d+1],f=e[u],w=e[u+1],M=e[_],x=e[_+1],b=p,k=m,y=c[a],v=s*c[a+1],T=h*y-g*v,P=h*v+g*y,F=c[2*a],C=s*c[2*a+1],S=f*F-w*C,A=f*C+w*F,E=c[3*a],L=s*c[3*a+1],I=M*E-x*L,z=M*L+x*E,j=b+S,O=k+A,D=b-S,V=k-A,N=T+I,B=P+z,R=s*(T-I),G=s*(P-z);if(e[i]=j+N,e[i+1]=O+B,e[d]=D+G,e[d+1]=V-R,0===r){e[u]=j-N,e[u+1]=O-B;continue}if(r===l)continue;const $=o+n-r,q=o+t-r;e[$]=D-s*G,e[$+1]=-V-s*R,e[q]=j-s*N,e[q+1]=s*B-O}}const d=r>>>1;for(let t=2;t<d;t+=2)e[r-t]=e[t],e[r-t+1]=-e[t+1]}_singleRealTransform2(e,t,s,r,o){const n=e[r],i=e[r+o];t[s]=n+i,t[s+1]=0,t[s+2]=n-i,t[s+3]=0}_singleRealTransform4(e,t,s,r,o,n){const i=2*o,a=3*o,l=e[r],c=e[r+o],d=e[r+i],u=e[r+a],_=l+d,p=l-d,m=c+u,h=n*(c-u);t[s]=_+m,t[s+1]=0,t[s+2]=p,t[s+3]=-h,t[s+4]=_-m,t[s+5]=0,t[s+6]=p,t[s+7]=h}}class m{constructor(e){const t=2*(e-1),s=2*(2*e-1),r=2**Math.ceil(Math.log2(s));this.bufferSize=r,this._a=t;const o=new Float64Array(s),n=new Float64Array(r);this._chirpBuffer=new Float64Array(r),this._buffer1=new Float64Array(r),this._buffer2=new Float64Array(r),this._outBuffer1=new Float64Array(r),this._outBuffer2=new Float64Array(r);const i=-2*Math.PI/e,a=Math.cos(i),l=Math.sin(i);for(let t=0;t<s>>1;++t){const s=(t+1-e)**2/2,r=Math.sqrt(a**2+l**2)**s,i=s*Math.atan2(l,a),c=2*t;o[c]=r*Math.cos(i),o[c+1]=r*Math.sin(i),n[c]=o[c],n[c+1]=-o[c+1]}this._slicedChirpBuffer=o.subarray(t,s),this._f=new p(r>>1),this._f.transform(this._chirpBuffer,n)}_transform(e,t,s){const r=this._buffer1,o=this._buffer2,n=this._outBuffer1,i=this._outBuffer2,a=this._chirpBuffer,l=this._slicedChirpBuffer,c=this._a;if(s)for(let e=0;e<l.length;e+=2){const s=e+1,o=t[e>>1];r[e]=o*l[e],r[s]=o*l[s]}else for(let e=0;e<l.length;e+=2){const s=e+1;r[e]=t[e]*l[e]-t[s]*l[s],r[s]=t[e]*l[s]+t[s]*l[e]}this._f.transform(n,r);for(let e=0;e<a.length;e+=2){const t=e+1;o[e]=n[e]*a[e]-n[t]*a[t],o[t]=n[e]*a[t]+n[t]*a[e]}this._f.inverseTransform(i,o);for(let t=0;t<i.length;t+=2){const s=i[t+c],r=i[t+c+1],o=l[t],n=l[t+1];e[t]=s*o-r*n,e[t+1]=s*n+r*o}}transform(e,t){this._transform(e,t,!1)}realTransform(e,t){this._transform(e,t,!0)}}class h{constructor(e){this.fft_length=e,this.isPowerOfTwo=_(e),this.isPowerOfTwo?(this.fft=new p(e),this.outputBufferSize=2*e):(this.fft=new m(e),this.outputBufferSize=this.fft.bufferSize)}realTransform(e,t){this.fft.realTransform(e,t)}transform(e,t){this.fft.transform(e,t)}}function g(e,t){if(t%2==0||t<=0)throw new Error("Window size must be a positive odd number");const s=new e.constructor(e.length),r=new e.constructor(t),o=Math.floor(t/2);for(let t=0;t<e.length;++t){let n=0;for(let s=-o;s<=o;++s){let o=t+s;o<0?o=Math.abs(o):o>=e.length&&(o=2*(e.length-1)-o),r[n++]=e[o]}r.sort(),s[t]=r[o]}return s}function f(e,t){const s=Math.pow(10,t);return Math.round(e*s)/s}function w(e){const t=Math.round(e);return Math.abs(e)%1==.5?t%2==0?t:t-1:t}function M(e){const t=e.length,s=e[0].length,r=[t+1,s+1],o=Array.from({length:r[0]},(()=>Array(r[1]).fill(1/0)));o[0][0]=0;const n=Array.from({length:r[0]},(()=>Array(r[1]).fill(-1)));for(let t=1;t<r[1];++t)for(let s=1;s<r[0];++s){const r=o[s-1][t-1],i=o[s-1][t],a=o[s][t-1];let l,c;r<i&&r<a?(l=r,c=0):i<r&&i<a?(l=i,c=1):(l=a,c=2),o[s][t]=e[s-1][t-1]+l,n[s][t]=c}for(let e=0;e<r[1];++e)n[0][e]=2;for(let e=0;e<r[0];++e)n[e][0]=1;let i=t,a=s,l=[],c=[];for(;i>0||a>0;)switch(l.push(i-1),c.push(a-1),n[i][a]){case 0:--i,--a;break;case 1:--i;break;case 2:--a;break;default:throw new Error(`Internal error in dynamic time warping. Unexpected trace[${i}, ${a}]. Please file a bug report.`)}return l.reverse(),c.reverse(),[l,c]}},"./src/utils/tensor.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{DataTypeMap:()=>i,Tensor:()=>a,cat:()=>b,full:()=>F,full_like:()=>C,interpolate:()=>c,interpolate_4d:()=>d,layer_norm:()=>f,matmul:()=>u,mean:()=>T,mean_pooling:()=>g,ones:()=>S,ones_like:()=>A,permute:()=>l,quantize_embeddings:()=>z,rand:()=>I,rfft:()=>_,slice:()=>h,stack:()=>k,std_mean:()=>v,topk:()=>p,zeros:()=>E,zeros_like:()=>L});var r=s("./src/utils/maths.js"),o=s("./src/backends/onnx.js"),n=s("./src/ops/registry.js");const i=Object.freeze({float32:Float32Array,float16:"undefined"!=typeof Float16Array?Float16Array:Uint16Array,float64:Float64Array,string:Array,int8:Int8Array,uint8:Uint8Array,int16:Int16Array,uint16:Uint16Array,int32:Int32Array,uint32:Uint32Array,int64:BigInt64Array,uint64:BigUint64Array,bool:Uint8Array,uint4:Uint8Array,int4:Int8Array});class a{get dims(){return this.ort_tensor.dims}set dims(e){this.ort_tensor.dims=e}get type(){return this.ort_tensor.type}get data(){return this.ort_tensor.data}get size(){return this.ort_tensor.size}get location(){return this.ort_tensor.location}ort_tensor;constructor(...e){return(0,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 s=Number(t);if(Number.isInteger(s))return e._getitem(s)}return e[t]},set:(e,t,s)=>e[t]=s})}dispose(){this.ort_tensor.dispose()}*[Symbol.iterator](){const[e,...t]=this.dims;if(t.length>0){const s=t.reduce(((e,t)=>e*t));for(let r=0;r<e;++r)yield this._subarray(r,s,t)}else yield*this.data}_getitem(e){const[t,...s]=this.dims;if(e=x(e,t),s.length>0){const t=s.reduce(((e,t)=>e*t));return this._subarray(e,t,s)}return new a(this.type,[this.data[e]],s)}indexOf(e){const t=this.data;for(let s=0;s<t.length;++s)if(t[s]==e)return s;return-1}_subarray(e,t,s){const r=e*t,o=(e+1)*t,n="subarray"in this.data?this.data.subarray(r,o):this.data.slice(r,o);return new a(this.type,n,s)}item(){const e=this.data;if(1!==e.length)throw new Error(`a Tensor with ${e.length} elements cannot be converted to Scalar`);return e[0]}tolist(){return function(e,t){const s=e.length,r=t.reduce(((e,t)=>e*t));if(s!==r)throw Error(`cannot reshape array of size ${s} into shape (${t})`);let o=e;for(let e=t.length-1;e>=0;e--)o=o.reduce(((s,r)=>{let o=s[s.length-1];return o.length<t[e]?o.push(r):s.push([r]),s}),[[]]);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 s=0;s<t.length;++s)t[s]=e(t[s],s,t);return this}mul(e){return this.clone().mul_(e)}mul_(e){const t=this.data;for(let s=0;s<t.length;++s)t[s]*=e;return this}div(e){return this.clone().div_(e)}div_(e){const t=this.data;for(let s=0;s<t.length;++s)t[s]/=e;return this}add(e){return this.clone().add_(e)}add_(e){const t=this.data;for(let s=0;s<t.length;++s)t[s]+=e;return this}sub(e){return this.clone().sub_(e)}sub_(e){const t=this.data;for(let s=0;s<t.length;++s)t[s]-=e;return this}clone(){return new a(this.type,this.data.slice(),this.dims.slice())}slice(...e){const t=[],s=[];for(let r=0;r<this.dims.length;++r){let o=e[r];if(null==o)s.push([0,this.dims[r]]),t.push(this.dims[r]);else if("number"==typeof o)o=x(o,this.dims[r],r),s.push([o,o+1]);else{if(!Array.isArray(o)||2!==o.length)throw new Error(`Invalid slice: ${o}`);{let[e,n]=o;if(e=null===e?0:x(e,this.dims[r],r,!1),n=null===n?this.dims[r]:x(n,this.dims[r],r,!1),e>n)throw new Error(`Invalid slice: ${o}`);const i=[Math.max(e,0),Math.min(n,this.dims[r])];s.push(i),t.push(i[1]-i[0])}}}const r=s.map((([e,t])=>t-e)),o=r.reduce(((e,t)=>e*t)),n=this.data,i=new n.constructor(o),l=this.stride();for(let e=0;e<o;++e){let t=0;for(let o=r.length-1,n=e;o>=0;--o){const e=r[o];t+=(n%e+s[o][0])*l[o],n=Math.floor(n/e)}i[e]=n[t]}return new a(this.type,i,t)}permute(...e){return l(this,e)}transpose(...e){return this.permute(...e)}sum(e=null,t=!1){return this.norm(1,e,t)}norm(e="fro",t=null,s=!1){if("fro"===e)e=2;else if("string"==typeof e)throw Error(`Unsupported norm: ${e}`);const r=this.data,o=(t,s)=>t+s**e;if(null===t){const t=r.reduce(o,0)**(1/e);return new a(this.type,[t],[])}const[n,i,l]=y(o,this,t,s);if(1!==e)for(let t=0;t<i.length;++t)i[t]=i[t]**(1/e);return new a(n,i,l)}normalize_(e=2,t=1){t=x(t,this.dims.length);const s=this.norm(e,t,!0),r=this.data,o=s.data;for(let e=0;e<r.length;++e){let s=0;for(let r=this.dims.length-1,o=e,n=1;r>=0;--r){const e=this.dims[r];if(r!==t){s+=o%e*n,n*=this.dims[r]}o=Math.floor(o/e)}r[e]/=o[s]}return this}normalize(e=2,t=1){return this.clone().normalize_(e,t)}stride(){return function(e){const t=new Array(e.length);for(let s=e.length-1,r=1;s>=0;--s)t[s]=r,r*=e[s];return t}(this.dims)}squeeze(e=null){return new a(this.type,this.data,w(this.dims,e))}squeeze_(e=null){return this.dims=w(this.dims,e),this}unsqueeze(e=null){return new a(this.type,this.data,M(this.dims,e))}unsqueeze_(e=null){return this.dims=M(this.dims,e),this}flatten_(e=0,t=-1){t=(t+this.dims.length)%this.dims.length;let s=this.dims.slice(0,e),r=this.dims.slice(e,t+1),o=this.dims.slice(t+1);return this.dims=[...s,r.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 s=0;s<e.length;++s)if(-1===e[s]){if(-1!==t)throw new Error("Only one dimension can be inferred");t=s}const s=this.data;if(-1!==t){const r=e.reduce(((e,s,r)=>r!==t?e*s:e),1);e[t]=s.length/r}return new a(this.type,s,e)}neg_(){const e=this.data;for(let t=0;t<e.length;++t)e[t]=-e[t];return this}neg(){return this.clone().neg_()}gt(e){const t=new Uint8Array(this.data.length),s=this.data;for(let r=0;r<s.length;++r)t[r]=s[r]>e?1:0;return new a("bool",t,this.dims)}lt(e){const t=new Uint8Array(this.data.length),s=this.data;for(let r=0;r<s.length;++r)t[r]=s[r]<e?1:0;return new a("bool",t,this.dims)}clamp_(e,t){const s=this.data;for(let r=0;r<s.length;++r)s[r]=Math.min(Math.max(s[r],e),t);return this}clamp(e,t){return this.clone().clamp_(e,t)}round_(){const e=this.data;for(let t=0;t<e.length;++t)e[t]=Math.round(e[t]);return this}round(){return this.clone().round_()}mean(e=null,t=!1){return T(this,e,t)}min(e=null,t=!1){if(null===e){const e=(0,r.min)(this.data)[0];return new a(this.type,[e],[])}const[s,o,n]=y(((e,t)=>Math.min(e,t)),this,e,t,1/0);return new a(s,o,n)}max(e=null,t=!1){if(null===e){const e=(0,r.max)(this.data)[0];return new a(this.type,[e],[])}const[s,o,n]=y(((e,t)=>Math.max(e,t)),this,e,t,-1/0);return new a(s,o,n)}argmin(e=null,t=!1){if(null!==e)throw new Error("`dim !== null` not yet implemented.");const s=(0,r.min)(this.data)[1];return new a("int64",[BigInt(s)],[])}argmax(e=null,t=!1){if(null!==e)throw new Error("`dim !== null` not yet implemented.");const s=(0,r.max)(this.data)[1];return new a("int64",[BigInt(s)],[])}to(e){if(this.type===e)return this;if(!i.hasOwnProperty(e))throw new Error(`Unsupported type: ${e}`);let t;const s=["int64","uint64"].includes(this.type),r=["int64","uint64"].includes(e);return s&&!r?t=Number:!s&&r&&(t=BigInt),new a(e,i[e].from(this.data,t),this.dims)}}function l(e,t){const[s,o]=(0,r.permute_data)(e.data,e.dims,t);return new a(e.type,s,o)}function c(e,[t,s],o="bilinear",n=!1){const i=e.dims.at(-3)??1,l=e.dims.at(-2),c=e.dims.at(-1);let d=(0,r.interpolate_data)(e.data,[i,l,c],[t,s],o,n);return new a(e.type,d,[i,t,s])}async function d(e,{size:t=null,mode:s="bilinear"}={}){if(4!==e.dims.length)throw new Error("`interpolate_4d` currently only supports 4D input.");if(!t)throw new Error("`interpolate_4d` requires a `size` argument.");let r,o;if(2===t.length)r=[...e.dims.slice(0,2),...t];else if(3===t.length)r=[e.dims[0],...t];else{if(4!==t.length)throw new Error("`size` must be of length 2, 3, or 4.");r=t}if("nearest"===s)o=await n.TensorOpRegistry.nearest_interpolate_4d;else if("bilinear"===s)o=await n.TensorOpRegistry.bilinear_interpolate_4d;else{if("bicubic"!==s)throw new Error(`Unsupported mode: ${s}`);o=await n.TensorOpRegistry.bicubic_interpolate_4d}const i=new a("int64",new BigInt64Array(r.map(BigInt)),[r.length]);return await o({x:e,s:i})}async function u(e,t){const s=await n.TensorOpRegistry.matmul;return await s({a:e,b:t})}async function _(e,t){const s=await n.TensorOpRegistry.rfft;return await s({x:e,a:t})}async function p(e,t){const s=await n.TensorOpRegistry.top_k;return t=null==t?e.dims.at(-1):Math.min(t,e.dims.at(-1)),await s({x:e,k:new a("int64",[BigInt(t)],[1])})}const m=e=>new a("int64",e,[e.length]);async function h(e,t,s,r,o){const i=await n.TensorOpRegistry.slice;return await i({x:e,s:m(t),e:m(s),a:m(r),t:m(o??new Array(r.length).fill(1))})}function g(e,t){const s=e.data,r=t.data,o=[e.dims[0],e.dims[2]],n=new s.constructor(o[0]*o[1]),[i,l,c]=e.dims;let d=0;for(let e=0;e<i;++e){const t=e*c*l;for(let o=0;o<c;++o){let i=0,a=0;const u=e*l,_=t+o;for(let e=0;e<l;++e){const t=Number(r[u+e]);a+=t,i+=s[_+e*c]*t}const p=i/a;n[d++]=p}}return new a(e.type,n,o)}function f(e,t,{eps:s=1e-5}={}){if(2!==e.dims.length)throw new Error("`layer_norm` currently only supports 2D input.");const[r,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[n,i]=v(e,1,0,!0),l=n.data,c=i.data,d=e.data,u=new d.constructor(d.length);for(let e=0;e<r;++e){const t=e*o;for(let r=0;r<o;++r){const o=t+r;u[o]=(d[o]-c[e])/(l[e]+s)}}return new a(e.type,u,e.dims)}function w(e,t){return e=e.slice(),null===t?e=e.filter((e=>1!==e)):"number"==typeof t?1===e[t]&&e.splice(t,1):Array.isArray(t)&&(e=e.filter(((e,s)=>1!==e||!t.includes(s)))),e}function M(e,t){return t=x(t,e.length+1),(e=e.slice()).splice(t,0,1),e}function x(e,t,s=null,r=!0){if(e<-t||e>=t){if(r)throw new Error(`IndexError: index ${e} is out of bounds for dimension${null===s?"":" "+s} with size ${t}`);return e<-t?0:t}return e<0&&(e=(e%t+t)%t),e}function b(e,t=0){t=x(t,e[0].dims.length);const s=e[0].dims.slice();s[t]=e.reduce(((e,s)=>e+s.dims[t]),0);const r=s.reduce(((e,t)=>e*t),1),o=new e[0].data.constructor(r),n=e[0].type;if(0===t){let t=0;for(const s of e){const e=s.data;o.set(e,t),t+=e.length}}else{let r=0;for(let n=0;n<e.length;++n){const{data:i,dims:a}=e[n];for(let e=0;e<i.length;++e){let n=0;for(let o=a.length-1,i=e,l=1;o>=0;--o){const e=a[o];let c=i%e;o===t&&(c+=r),n+=c*l,l*=s[o],i=Math.floor(i/e)}o[n]=i[e]}r+=a[t]}}return new a(n,o,s)}function k(e,t=0){return b(e.map((e=>e.unsqueeze(t))),t)}function y(e,t,s=null,r=!1,o=null){const n=t.data,i=t.dims;s=x(s,i.length);const a=i.slice();a[s]=1;const l=new n.constructor(n.length/i[s]);null!==o&&l.fill(o);for(let t=0;t<n.length;++t){let r=0;for(let e=i.length-1,o=t,n=1;e>=0;--e){const t=i[e];if(e!==s){r+=o%t*n,n*=a[e]}o=Math.floor(o/t)}l[r]=e(l[r],n[t],t,r)}return r||a.splice(s,1),[t.type,l,a]}function v(e,t=null,s=1,r=!1){const o=e.data,n=e.dims;if(null===t){const t=o.reduce(((e,t)=>e+t),0)/o.length,r=Math.sqrt(o.reduce(((e,s)=>e+(s-t)**2),0)/(o.length-s)),n=new a(e.type,[t],[]);return[new a(e.type,[r],[]),n]}const i=T(e,t=x(t,n.length),r),l=i.data,[c,d,u]=y(((e,t,s,r)=>e+(t-l[r])**2),e,t,r);for(let e=0;e<d.length;++e)d[e]=Math.sqrt(d[e]/(n[t]-s));return[new a(c,d,u),i]}function T(e,t=null,s=!1){const r=e.dims,o=e.data;if(null===t){const t=o.reduce(((e,t)=>e+t),0);return new a(e.type,[t/o.length],[])}t=x(t,r.length);const[n,i,l]=y(((e,t)=>e+t),e,t,s);if(1!==r[t])for(let e=0;e<i.length;++e)i[e]/=r[t];return new a(n,i,l)}function P(e,t,s,r){const o=e.reduce(((e,t)=>e*t),1);return new a(s,new r(o).fill(t),e)}function F(e,t){let s,r;if("number"==typeof t)s="float32",r=Float32Array;else if("bigint"==typeof t)s="int64",r=BigInt64Array;else{if("boolean"!=typeof t)throw new Error("Unsupported data type: "+typeof t);s="bool",r=Uint8Array}return P(e,t,s,r)}function C(e,t){return F(e.dims,t)}function S(e){return P(e,1n,"int64",BigInt64Array)}function A(e){return S(e.dims)}function E(e){return P(e,0n,"int64",BigInt64Array)}function L(e){return E(e.dims)}function I(e){const t=e.reduce(((e,t)=>e*t),1);return new a("float32",Float32Array.from({length:t},(()=>Math.random())),e)}function z(e,t){if(2!==e.dims.length)throw new Error("The tensor must have 2 dimensions");if(e.dims.at(-1)%8!=0)throw new Error("The last dimension of the tensor must be a multiple of 8");if(!["binary","ubinary"].includes(t))throw new Error("The precision must be either 'binary' or 'ubinary'");const s="binary"===t,r=s?"int8":"uint8",o=s?Int8Array:Uint8Array,n=e.data,i=new o(n.length/8);for(let e=0;e<n.length;++e){const t=n[e]>0?1:0,r=Math.floor(e/8),o=e%8;i[r]|=t<<7-o,s&&0===o&&(i[r]-=128)}return new a(r,i,[e.dims[0],e.dims[1]/8])}},"./src/utils/video.js":(e,t,s)=>{"use strict";s.r(t),s.d(t,{RawVideo:()=>i,RawVideoFrame:()=>n,load_video:()=>a});var r=s("./src/utils/image.js"),o=s("./src/env.js");class n{constructor(e,t){this.image=e,this.timestamp=t}}class i{constructor(e,t){e.length>0&&e[0]instanceof r.RawImage&&(e=e.map(((s,r)=>new n(s,(r+1)/(e.length+1)*t)))),this.frames=e,this.duration=t}get width(){return this.frames[0].image.width}get height(){return this.frames[0].image.height}get fps(){return this.frames.length/this.duration}}async function a(e,{num_frames:t=null,fps:s=null}={}){if(!o.apis.IS_BROWSER_ENV)throw new Error("`load_video` is currently only supported in browser environments.");if(null==t&&null==s)throw new Error("Either num_frames or fps must be provided.");const a=[],l=document.createElement("video");if(l.crossOrigin="anonymous",l.muted=!0,"string"==typeof e)l.src=e;else if(e instanceof Blob)l.src=URL.createObjectURL(e);else{if(!(e instanceof HTMLVideoElement))throw new Error("Invalid URL or video element provided.");l.src=e.src}if(await new Promise((e=>l.onloadedmetadata=e)),l.seekable.start(0)===l.seekable.end(0)){const e=await fetch(l.src),t=await e.blob();l.src=URL.createObjectURL(t),await new Promise((e=>l.onloadedmetadata=e))}const c=l.duration;let d,u;null!=t?(d=t,u=1===t?0:c/(t-1)):(u=1/s,d=Math.floor(c/u));let _=[];for(let e=0;e<d;++e)_.push(1===t?c/2:e*u);const p=document.createElement("canvas");p.width=l.videoWidth,p.height=l.videoHeight;const m=p.getContext("2d",{willReadFrequently:!0});for(const e of _){l.currentTime=e,await new Promise((e=>{l.onseeked=e})),m.drawImage(l,0,0,p.width,p.height);const t=m.getImageData(0,0,p.width,p.height),s=new r.RawImage(t.data,p.width,p.height,4),o=new n(s,e);a.push(o)}return l.remove(),new i(a,c)}}},r={};function o(e){var t=r[e];if(void 0!==t)return t.exports;var n=r[e]={exports:{}};return s[e](n,n.exports,o),n.exports}t=Object.getPrototypeOf?e=>Object.getPrototypeOf(e):e=>e.__proto__,o.t=function(s,r){if(1&r&&(s=this(s)),8&r)return s;if("object"==typeof s&&s){if(4&r&&s.__esModule)return s;if(16&r&&"function"==typeof s.then)return s}var n=Object.create(null);o.r(n);var i={};e=e||[null,t({}),t([]),t(t)];for(var a=2&r&&s;"object"==typeof a&&!~e.indexOf(a);a=t(a))Object.getOwnPropertyNames(a).forEach((e=>i[e]=()=>s[e]));return i.default=()=>s,o.d(n,i),n},o.d=(e,t)=>{for(var s in t)o.o(t,s)&&!o.o(e,s)&&Object.defineProperty(e,s,{enumerable:!0,get:t[s]})},o.o=(e,t)=>Object.prototype.hasOwnProperty.call(e,t),o.r=e=>{"undefined"!=typeof Symbol&&Symbol.toStringTag&&Object.defineProperty(e,Symbol.toStringTag,{value:"Module"}),Object.defineProperty(e,"__esModule",{value:!0})};var n={};(()=>{"use strict";o.r(n),o.d(n,{ASTFeatureExtractor:()=>p.ASTFeatureExtractor,ASTForAudioClassification:()=>s.ASTForAudioClassification,ASTModel:()=>s.ASTModel,ASTPreTrainedModel:()=>s.ASTPreTrainedModel,AlbertForMaskedLM:()=>s.AlbertForMaskedLM,AlbertForQuestionAnswering:()=>s.AlbertForQuestionAnswering,AlbertForSequenceClassification:()=>s.AlbertForSequenceClassification,AlbertModel:()=>s.AlbertModel,AlbertPreTrainedModel:()=>s.AlbertPreTrainedModel,AlbertTokenizer:()=>r.AlbertTokenizer,AudioClassificationPipeline:()=>t.AudioClassificationPipeline,AutoConfig:()=>i.AutoConfig,AutoFeatureExtractor:()=>m.AutoFeatureExtractor,AutoImageProcessor:()=>f.AutoImageProcessor,AutoModel:()=>s.AutoModel,AutoModelForAudioClassification:()=>s.AutoModelForAudioClassification,AutoModelForAudioFrameClassification:()=>s.AutoModelForAudioFrameClassification,AutoModelForAudioTextToText:()=>s.AutoModelForAudioTextToText,AutoModelForCTC:()=>s.AutoModelForCTC,AutoModelForCausalLM:()=>s.AutoModelForCausalLM,AutoModelForDepthEstimation:()=>s.AutoModelForDepthEstimation,AutoModelForDocumentQuestionAnswering:()=>s.AutoModelForDocumentQuestionAnswering,AutoModelForImageClassification:()=>s.AutoModelForImageClassification,AutoModelForImageFeatureExtraction:()=>s.AutoModelForImageFeatureExtraction,AutoModelForImageMatting:()=>s.AutoModelForImageMatting,AutoModelForImageSegmentation:()=>s.AutoModelForImageSegmentation,AutoModelForImageTextToText:()=>s.AutoModelForImageTextToText,AutoModelForImageToImage:()=>s.AutoModelForImageToImage,AutoModelForMaskGeneration:()=>s.AutoModelForMaskGeneration,AutoModelForMaskedLM:()=>s.AutoModelForMaskedLM,AutoModelForNormalEstimation:()=>s.AutoModelForNormalEstimation,AutoModelForObjectDetection:()=>s.AutoModelForObjectDetection,AutoModelForPoseEstimation:()=>s.AutoModelForPoseEstimation,AutoModelForQuestionAnswering:()=>s.AutoModelForQuestionAnswering,AutoModelForSemanticSegmentation:()=>s.AutoModelForSemanticSegmentation,AutoModelForSeq2SeqLM:()=>s.AutoModelForSeq2SeqLM,AutoModelForSequenceClassification:()=>s.AutoModelForSequenceClassification,AutoModelForSpeechSeq2Seq:()=>s.AutoModelForSpeechSeq2Seq,AutoModelForTextToSpectrogram:()=>s.AutoModelForTextToSpectrogram,AutoModelForTextToWaveform:()=>s.AutoModelForTextToWaveform,AutoModelForTokenClassification:()=>s.AutoModelForTokenClassification,AutoModelForUniversalSegmentation:()=>s.AutoModelForUniversalSegmentation,AutoModelForVision2Seq:()=>s.AutoModelForVision2Seq,AutoModelForXVector:()=>s.AutoModelForXVector,AutoModelForZeroShotObjectDetection:()=>s.AutoModelForZeroShotObjectDetection,AutoProcessor:()=>x.AutoProcessor,AutoTokenizer:()=>r.AutoTokenizer,AutomaticSpeechRecognitionPipeline:()=>t.AutomaticSpeechRecognitionPipeline,BackgroundRemovalPipeline:()=>t.BackgroundRemovalPipeline,BartForConditionalGeneration:()=>s.BartForConditionalGeneration,BartForSequenceClassification:()=>s.BartForSequenceClassification,BartModel:()=>s.BartModel,BartPretrainedModel:()=>s.BartPretrainedModel,BartTokenizer:()=>r.BartTokenizer,BaseModelOutput:()=>s.BaseModelOutput,BaseStreamer:()=>b.BaseStreamer,BeitFeatureExtractor:()=>g.BeitFeatureExtractor,BeitForImageClassification:()=>s.BeitForImageClassification,BeitModel:()=>s.BeitModel,BeitPreTrainedModel:()=>s.BeitPreTrainedModel,BertForMaskedLM:()=>s.BertForMaskedLM,BertForQuestionAnswering:()=>s.BertForQuestionAnswering,BertForSequenceClassification:()=>s.BertForSequenceClassification,BertForTokenClassification:()=>s.BertForTokenClassification,BertModel:()=>s.BertModel,BertPreTrainedModel:()=>s.BertPreTrainedModel,BertTokenizer:()=>r.BertTokenizer,BitImageProcessor:()=>g.BitImageProcessor,BlenderbotForConditionalGeneration:()=>s.BlenderbotForConditionalGeneration,BlenderbotModel:()=>s.BlenderbotModel,BlenderbotPreTrainedModel:()=>s.BlenderbotPreTrainedModel,BlenderbotSmallForConditionalGeneration:()=>s.BlenderbotSmallForConditionalGeneration,BlenderbotSmallModel:()=>s.BlenderbotSmallModel,BlenderbotSmallPreTrainedModel:()=>s.BlenderbotSmallPreTrainedModel,BlenderbotSmallTokenizer:()=>r.BlenderbotSmallTokenizer,BlenderbotTokenizer:()=>r.BlenderbotTokenizer,BloomForCausalLM:()=>s.BloomForCausalLM,BloomModel:()=>s.BloomModel,BloomPreTrainedModel:()=>s.BloomPreTrainedModel,BloomTokenizer:()=>r.BloomTokenizer,CLIPFeatureExtractor:()=>g.CLIPFeatureExtractor,CLIPImageProcessor:()=>g.CLIPImageProcessor,CLIPModel:()=>s.CLIPModel,CLIPPreTrainedModel:()=>s.CLIPPreTrainedModel,CLIPSegForImageSegmentation:()=>s.CLIPSegForImageSegmentation,CLIPSegModel:()=>s.CLIPSegModel,CLIPSegPreTrainedModel:()=>s.CLIPSegPreTrainedModel,CLIPTextModel:()=>s.CLIPTextModel,CLIPTextModelWithProjection:()=>s.CLIPTextModelWithProjection,CLIPTokenizer:()=>r.CLIPTokenizer,CLIPVisionModel:()=>s.CLIPVisionModel,CLIPVisionModelWithProjection:()=>s.CLIPVisionModelWithProjection,CamembertForMaskedLM:()=>s.CamembertForMaskedLM,CamembertForQuestionAnswering:()=>s.CamembertForQuestionAnswering,CamembertForSequenceClassification:()=>s.CamembertForSequenceClassification,CamembertForTokenClassification:()=>s.CamembertForTokenClassification,CamembertModel:()=>s.CamembertModel,CamembertPreTrainedModel:()=>s.CamembertPreTrainedModel,CamembertTokenizer:()=>r.CamembertTokenizer,CausalLMOutput:()=>s.CausalLMOutput,CausalLMOutputWithPast:()=>s.CausalLMOutputWithPast,ChineseCLIPFeatureExtractor:()=>g.ChineseCLIPFeatureExtractor,ChineseCLIPModel:()=>s.ChineseCLIPModel,ChineseCLIPPreTrainedModel:()=>s.ChineseCLIPPreTrainedModel,ClapAudioModelWithProjection:()=>s.ClapAudioModelWithProjection,ClapFeatureExtractor:()=>p.ClapFeatureExtractor,ClapModel:()=>s.ClapModel,ClapPreTrainedModel:()=>s.ClapPreTrainedModel,ClapTextModelWithProjection:()=>s.ClapTextModelWithProjection,ClassifierFreeGuidanceLogitsProcessor:()=>y.ClassifierFreeGuidanceLogitsProcessor,CodeGenForCausalLM:()=>s.CodeGenForCausalLM,CodeGenModel:()=>s.CodeGenModel,CodeGenPreTrainedModel:()=>s.CodeGenPreTrainedModel,CodeGenTokenizer:()=>r.CodeGenTokenizer,CodeLlamaTokenizer:()=>r.CodeLlamaTokenizer,CohereForCausalLM:()=>s.CohereForCausalLM,CohereModel:()=>s.CohereModel,CoherePreTrainedModel:()=>s.CoherePreTrainedModel,CohereTokenizer:()=>r.CohereTokenizer,ConvBertForMaskedLM:()=>s.ConvBertForMaskedLM,ConvBertForQuestionAnswering:()=>s.ConvBertForQuestionAnswering,ConvBertForSequenceClassification:()=>s.ConvBertForSequenceClassification,ConvBertForTokenClassification:()=>s.ConvBertForTokenClassification,ConvBertModel:()=>s.ConvBertModel,ConvBertPreTrainedModel:()=>s.ConvBertPreTrainedModel,ConvBertTokenizer:()=>r.ConvBertTokenizer,ConvNextFeatureExtractor:()=>g.ConvNextFeatureExtractor,ConvNextForImageClassification:()=>s.ConvNextForImageClassification,ConvNextImageProcessor:()=>g.ConvNextImageProcessor,ConvNextModel:()=>s.ConvNextModel,ConvNextPreTrainedModel:()=>s.ConvNextPreTrainedModel,ConvNextV2ForImageClassification:()=>s.ConvNextV2ForImageClassification,ConvNextV2Model:()=>s.ConvNextV2Model,ConvNextV2PreTrainedModel:()=>s.ConvNextV2PreTrainedModel,DFineForObjectDetection:()=>s.DFineForObjectDetection,DFineModel:()=>s.DFineModel,DFinePreTrainedModel:()=>s.DFinePreTrainedModel,DPTFeatureExtractor:()=>g.DPTFeatureExtractor,DPTForDepthEstimation:()=>s.DPTForDepthEstimation,DPTImageProcessor:()=>g.DPTImageProcessor,DPTModel:()=>s.DPTModel,DPTPreTrainedModel:()=>s.DPTPreTrainedModel,DacDecoderModel:()=>s.DacDecoderModel,DacDecoderOutput:()=>s.DacDecoderOutput,DacEncoderModel:()=>s.DacEncoderModel,DacEncoderOutput:()=>s.DacEncoderOutput,DacFeatureExtractor:()=>p.DacFeatureExtractor,DacModel:()=>s.DacModel,DacPreTrainedModel:()=>s.DacPreTrainedModel,DataTypeMap:()=>d.DataTypeMap,DebertaForMaskedLM:()=>s.DebertaForMaskedLM,DebertaForQuestionAnswering:()=>s.DebertaForQuestionAnswering,DebertaForSequenceClassification:()=>s.DebertaForSequenceClassification,DebertaForTokenClassification:()=>s.DebertaForTokenClassification,DebertaModel:()=>s.DebertaModel,DebertaPreTrainedModel:()=>s.DebertaPreTrainedModel,DebertaTokenizer:()=>r.DebertaTokenizer,DebertaV2ForMaskedLM:()=>s.DebertaV2ForMaskedLM,DebertaV2ForQuestionAnswering:()=>s.DebertaV2ForQuestionAnswering,DebertaV2ForSequenceClassification:()=>s.DebertaV2ForSequenceClassification,DebertaV2ForTokenClassification:()=>s.DebertaV2ForTokenClassification,DebertaV2Model:()=>s.DebertaV2Model,DebertaV2PreTrainedModel:()=>s.DebertaV2PreTrainedModel,DebertaV2Tokenizer:()=>r.DebertaV2Tokenizer,DecisionTransformerModel:()=>s.DecisionTransformerModel,DecisionTransformerPreTrainedModel:()=>s.DecisionTransformerPreTrainedModel,DeiTFeatureExtractor:()=>g.DeiTFeatureExtractor,DeiTForImageClassification:()=>s.DeiTForImageClassification,DeiTImageProcessor:()=>g.DeiTImageProcessor,DeiTModel:()=>s.DeiTModel,DeiTPreTrainedModel:()=>s.DeiTPreTrainedModel,DepthAnythingForDepthEstimation:()=>s.DepthAnythingForDepthEstimation,DepthAnythingPreTrainedModel:()=>s.DepthAnythingPreTrainedModel,DepthEstimationPipeline:()=>t.DepthEstimationPipeline,DepthProForDepthEstimation:()=>s.DepthProForDepthEstimation,DepthProPreTrainedModel:()=>s.DepthProPreTrainedModel,DetrFeatureExtractor:()=>g.DetrFeatureExtractor,DetrForObjectDetection:()=>s.DetrForObjectDetection,DetrForSegmentation:()=>s.DetrForSegmentation,DetrImageProcessor:()=>g.DetrImageProcessor,DetrModel:()=>s.DetrModel,DetrObjectDetectionOutput:()=>s.DetrObjectDetectionOutput,DetrPreTrainedModel:()=>s.DetrPreTrainedModel,DetrSegmentationOutput:()=>s.DetrSegmentationOutput,Dinov2ForImageClassification:()=>s.Dinov2ForImageClassification,Dinov2Model:()=>s.Dinov2Model,Dinov2PreTrainedModel:()=>s.Dinov2PreTrainedModel,Dinov2WithRegistersForImageClassification:()=>s.Dinov2WithRegistersForImageClassification,Dinov2WithRegistersModel:()=>s.Dinov2WithRegistersModel,Dinov2WithRegistersPreTrainedModel:()=>s.Dinov2WithRegistersPreTrainedModel,DistilBertForMaskedLM:()=>s.DistilBertForMaskedLM,DistilBertForQuestionAnswering:()=>s.DistilBertForQuestionAnswering,DistilBertForSequenceClassification:()=>s.DistilBertForSequenceClassification,DistilBertForTokenClassification:()=>s.DistilBertForTokenClassification,DistilBertModel:()=>s.DistilBertModel,DistilBertPreTrainedModel:()=>s.DistilBertPreTrainedModel,DistilBertTokenizer:()=>r.DistilBertTokenizer,DocumentQuestionAnsweringPipeline:()=>t.DocumentQuestionAnsweringPipeline,DonutFeatureExtractor:()=>g.DonutFeatureExtractor,DonutImageProcessor:()=>g.DonutImageProcessor,DonutSwinModel:()=>s.DonutSwinModel,DonutSwinPreTrainedModel:()=>s.DonutSwinPreTrainedModel,EfficientNetForImageClassification:()=>s.EfficientNetForImageClassification,EfficientNetImageProcessor:()=>g.EfficientNetImageProcessor,EfficientNetModel:()=>s.EfficientNetModel,EfficientNetPreTrainedModel:()=>s.EfficientNetPreTrainedModel,ElectraForMaskedLM:()=>s.ElectraForMaskedLM,ElectraForQuestionAnswering:()=>s.ElectraForQuestionAnswering,ElectraForSequenceClassification:()=>s.ElectraForSequenceClassification,ElectraForTokenClassification:()=>s.ElectraForTokenClassification,ElectraModel:()=>s.ElectraModel,ElectraPreTrainedModel:()=>s.ElectraPreTrainedModel,ElectraTokenizer:()=>r.ElectraTokenizer,EncodecFeatureExtractor:()=>p.EncodecFeatureExtractor,EosTokenCriteria:()=>k.EosTokenCriteria,EsmForMaskedLM:()=>s.EsmForMaskedLM,EsmForSequenceClassification:()=>s.EsmForSequenceClassification,EsmForTokenClassification:()=>s.EsmForTokenClassification,EsmModel:()=>s.EsmModel,EsmPreTrainedModel:()=>s.EsmPreTrainedModel,EsmTokenizer:()=>r.EsmTokenizer,ExaoneForCausalLM:()=>s.ExaoneForCausalLM,ExaoneModel:()=>s.ExaoneModel,ExaonePreTrainedModel:()=>s.ExaonePreTrainedModel,FFT:()=>u.FFT,FalconForCausalLM:()=>s.FalconForCausalLM,FalconModel:()=>s.FalconModel,FalconPreTrainedModel:()=>s.FalconPreTrainedModel,FalconTokenizer:()=>r.FalconTokenizer,FastViTForImageClassification:()=>s.FastViTForImageClassification,FastViTModel:()=>s.FastViTModel,FastViTPreTrainedModel:()=>s.FastViTPreTrainedModel,FeatureExtractionPipeline:()=>t.FeatureExtractionPipeline,FeatureExtractor:()=>_.FeatureExtractor,FillMaskPipeline:()=>t.FillMaskPipeline,Florence2ForConditionalGeneration:()=>s.Florence2ForConditionalGeneration,Florence2PreTrainedModel:()=>s.Florence2PreTrainedModel,Florence2Processor:()=>M.Florence2Processor,ForcedBOSTokenLogitsProcessor:()=>y.ForcedBOSTokenLogitsProcessor,ForcedEOSTokenLogitsProcessor:()=>y.ForcedEOSTokenLogitsProcessor,GLPNFeatureExtractor:()=>g.GLPNFeatureExtractor,GLPNForDepthEstimation:()=>s.GLPNForDepthEstimation,GLPNModel:()=>s.GLPNModel,GLPNPreTrainedModel:()=>s.GLPNPreTrainedModel,GPT2LMHeadModel:()=>s.GPT2LMHeadModel,GPT2Model:()=>s.GPT2Model,GPT2PreTrainedModel:()=>s.GPT2PreTrainedModel,GPT2Tokenizer:()=>r.GPT2Tokenizer,GPTBigCodeForCausalLM:()=>s.GPTBigCodeForCausalLM,GPTBigCodeModel:()=>s.GPTBigCodeModel,GPTBigCodePreTrainedModel:()=>s.GPTBigCodePreTrainedModel,GPTJForCausalLM:()=>s.GPTJForCausalLM,GPTJModel:()=>s.GPTJModel,GPTJPreTrainedModel:()=>s.GPTJPreTrainedModel,GPTNeoForCausalLM:()=>s.GPTNeoForCausalLM,GPTNeoModel:()=>s.GPTNeoModel,GPTNeoPreTrainedModel:()=>s.GPTNeoPreTrainedModel,GPTNeoXForCausalLM:()=>s.GPTNeoXForCausalLM,GPTNeoXModel:()=>s.GPTNeoXModel,GPTNeoXPreTrainedModel:()=>s.GPTNeoXPreTrainedModel,GPTNeoXTokenizer:()=>r.GPTNeoXTokenizer,Gemma2ForCausalLM:()=>s.Gemma2ForCausalLM,Gemma2Model:()=>s.Gemma2Model,Gemma2PreTrainedModel:()=>s.Gemma2PreTrainedModel,Gemma3ForCausalLM:()=>s.Gemma3ForCausalLM,Gemma3Model:()=>s.Gemma3Model,Gemma3PreTrainedModel:()=>s.Gemma3PreTrainedModel,GemmaForCausalLM:()=>s.GemmaForCausalLM,GemmaModel:()=>s.GemmaModel,GemmaPreTrainedModel:()=>s.GemmaPreTrainedModel,GemmaTokenizer:()=>r.GemmaTokenizer,GlmForCausalLM:()=>s.GlmForCausalLM,GlmModel:()=>s.GlmModel,GlmPreTrainedModel:()=>s.GlmPreTrainedModel,GraniteForCausalLM:()=>s.GraniteForCausalLM,GraniteModel:()=>s.GraniteModel,GranitePreTrainedModel:()=>s.GranitePreTrainedModel,Grok1Tokenizer:()=>r.Grok1Tokenizer,GroundingDinoForObjectDetection:()=>s.GroundingDinoForObjectDetection,GroundingDinoImageProcessor:()=>g.GroundingDinoImageProcessor,GroundingDinoPreTrainedModel:()=>s.GroundingDinoPreTrainedModel,GroundingDinoProcessor:()=>M.GroundingDinoProcessor,GroupViTModel:()=>s.GroupViTModel,GroupViTPreTrainedModel:()=>s.GroupViTPreTrainedModel,HeliumForCausalLM:()=>s.HeliumForCausalLM,HeliumModel:()=>s.HeliumModel,HeliumPreTrainedModel:()=>s.HeliumPreTrainedModel,HerbertTokenizer:()=>r.HerbertTokenizer,HieraForImageClassification:()=>s.HieraForImageClassification,HieraModel:()=>s.HieraModel,HieraPreTrainedModel:()=>s.HieraPreTrainedModel,HubertForCTC:()=>s.HubertForCTC,HubertForSequenceClassification:()=>s.HubertForSequenceClassification,HubertModel:()=>s.HubertModel,HubertPreTrainedModel:()=>s.HubertPreTrainedModel,IJepaForImageClassification:()=>s.IJepaForImageClassification,IJepaModel:()=>s.IJepaModel,IJepaPreTrainedModel:()=>s.IJepaPreTrainedModel,Idefics3ForConditionalGeneration:()=>s.Idefics3ForConditionalGeneration,Idefics3ImageProcessor:()=>g.Idefics3ImageProcessor,Idefics3PreTrainedModel:()=>s.Idefics3PreTrainedModel,Idefics3Processor:()=>M.Idefics3Processor,ImageClassificationPipeline:()=>t.ImageClassificationPipeline,ImageFeatureExtractionPipeline:()=>t.ImageFeatureExtractionPipeline,ImageFeatureExtractor:()=>p.ImageFeatureExtractor,ImageMattingOutput:()=>s.ImageMattingOutput,ImageProcessor:()=>h.ImageProcessor,ImageSegmentationPipeline:()=>t.ImageSegmentationPipeline,ImageToImagePipeline:()=>t.ImageToImagePipeline,ImageToTextPipeline:()=>t.ImageToTextPipeline,InterruptableStoppingCriteria:()=>k.InterruptableStoppingCriteria,JAISLMHeadModel:()=>s.JAISLMHeadModel,JAISModel:()=>s.JAISModel,JAISPreTrainedModel:()=>s.JAISPreTrainedModel,JinaCLIPImageProcessor:()=>g.JinaCLIPImageProcessor,JinaCLIPModel:()=>s.JinaCLIPModel,JinaCLIPPreTrainedModel:()=>s.JinaCLIPPreTrainedModel,JinaCLIPProcessor:()=>M.JinaCLIPProcessor,JinaCLIPTextModel:()=>s.JinaCLIPTextModel,JinaCLIPVisionModel:()=>s.JinaCLIPVisionModel,LiteWhisperForConditionalGeneration:()=>s.LiteWhisperForConditionalGeneration,LlamaForCausalLM:()=>s.LlamaForCausalLM,LlamaModel:()=>s.LlamaModel,LlamaPreTrainedModel:()=>s.LlamaPreTrainedModel,LlamaTokenizer:()=>r.LlamaTokenizer,LlavaForConditionalGeneration:()=>s.LlavaForConditionalGeneration,LlavaOnevisionForConditionalGeneration:()=>s.LlavaOnevisionForConditionalGeneration,LlavaOnevisionImageProcessor:()=>g.LlavaOnevisionImageProcessor,LlavaPreTrainedModel:()=>s.LlavaPreTrainedModel,LogitsProcessor:()=>y.LogitsProcessor,LogitsProcessorList:()=>y.LogitsProcessorList,LogitsWarper:()=>y.LogitsWarper,LongT5ForConditionalGeneration:()=>s.LongT5ForConditionalGeneration,LongT5Model:()=>s.LongT5Model,LongT5PreTrainedModel:()=>s.LongT5PreTrainedModel,M2M100ForConditionalGeneration:()=>s.M2M100ForConditionalGeneration,M2M100Model:()=>s.M2M100Model,M2M100PreTrainedModel:()=>s.M2M100PreTrainedModel,M2M100Tokenizer:()=>r.M2M100Tokenizer,MBart50Tokenizer:()=>r.MBart50Tokenizer,MBartForCausalLM:()=>s.MBartForCausalLM,MBartForConditionalGeneration:()=>s.MBartForConditionalGeneration,MBartForSequenceClassification:()=>s.MBartForSequenceClassification,MBartModel:()=>s.MBartModel,MBartPreTrainedModel:()=>s.MBartPreTrainedModel,MBartTokenizer:()=>r.MBartTokenizer,MPNetForMaskedLM:()=>s.MPNetForMaskedLM,MPNetForQuestionAnswering:()=>s.MPNetForQuestionAnswering,MPNetForSequenceClassification:()=>s.MPNetForSequenceClassification,MPNetForTokenClassification:()=>s.MPNetForTokenClassification,MPNetModel:()=>s.MPNetModel,MPNetPreTrainedModel:()=>s.MPNetPreTrainedModel,MPNetTokenizer:()=>r.MPNetTokenizer,MT5ForConditionalGeneration:()=>s.MT5ForConditionalGeneration,MT5Model:()=>s.MT5Model,MT5PreTrainedModel:()=>s.MT5PreTrainedModel,MarianMTModel:()=>s.MarianMTModel,MarianModel:()=>s.MarianModel,MarianPreTrainedModel:()=>s.MarianPreTrainedModel,MarianTokenizer:()=>r.MarianTokenizer,Mask2FormerImageProcessor:()=>g.Mask2FormerImageProcessor,MaskFormerFeatureExtractor:()=>g.MaskFormerFeatureExtractor,MaskFormerForInstanceSegmentation:()=>s.MaskFormerForInstanceSegmentation,MaskFormerImageProcessor:()=>g.MaskFormerImageProcessor,MaskFormerModel:()=>s.MaskFormerModel,MaskFormerPreTrainedModel:()=>s.MaskFormerPreTrainedModel,MaskedLMOutput:()=>s.MaskedLMOutput,MaxLengthCriteria:()=>k.MaxLengthCriteria,Metric3DForDepthEstimation:()=>s.Metric3DForDepthEstimation,Metric3DPreTrainedModel:()=>s.Metric3DPreTrainedModel,Metric3Dv2ForDepthEstimation:()=>s.Metric3Dv2ForDepthEstimation,Metric3Dv2PreTrainedModel:()=>s.Metric3Dv2PreTrainedModel,MgpstrForSceneTextRecognition:()=>s.MgpstrForSceneTextRecognition,MgpstrModelOutput:()=>s.MgpstrModelOutput,MgpstrPreTrainedModel:()=>s.MgpstrPreTrainedModel,MgpstrProcessor:()=>M.MgpstrProcessor,MgpstrTokenizer:()=>r.MgpstrTokenizer,MimiDecoderModel:()=>s.MimiDecoderModel,MimiDecoderOutput:()=>s.MimiDecoderOutput,MimiEncoderModel:()=>s.MimiEncoderModel,MimiEncoderOutput:()=>s.MimiEncoderOutput,MimiModel:()=>s.MimiModel,MimiPreTrainedModel:()=>s.MimiPreTrainedModel,MinLengthLogitsProcessor:()=>y.MinLengthLogitsProcessor,MinNewTokensLengthLogitsProcessor:()=>y.MinNewTokensLengthLogitsProcessor,MistralForCausalLM:()=>s.MistralForCausalLM,MistralModel:()=>s.MistralModel,MistralPreTrainedModel:()=>s.MistralPreTrainedModel,MobileBertForMaskedLM:()=>s.MobileBertForMaskedLM,MobileBertForQuestionAnswering:()=>s.MobileBertForQuestionAnswering,MobileBertForSequenceClassification:()=>s.MobileBertForSequenceClassification,MobileBertModel:()=>s.MobileBertModel,MobileBertPreTrainedModel:()=>s.MobileBertPreTrainedModel,MobileBertTokenizer:()=>r.MobileBertTokenizer,MobileLLMForCausalLM:()=>s.MobileLLMForCausalLM,MobileLLMModel:()=>s.MobileLLMModel,MobileLLMPreTrainedModel:()=>s.MobileLLMPreTrainedModel,MobileNetV1FeatureExtractor:()=>g.MobileNetV1FeatureExtractor,MobileNetV1ForImageClassification:()=>s.MobileNetV1ForImageClassification,MobileNetV1ForSemanticSegmentation:()=>s.MobileNetV1ForSemanticSegmentation,MobileNetV1ImageProcessor:()=>g.MobileNetV1ImageProcessor,MobileNetV1Model:()=>s.MobileNetV1Model,MobileNetV1PreTrainedModel:()=>s.MobileNetV1PreTrainedModel,MobileNetV2FeatureExtractor:()=>g.MobileNetV2FeatureExtractor,MobileNetV2ForImageClassification:()=>s.MobileNetV2ForImageClassification,MobileNetV2ForSemanticSegmentation:()=>s.MobileNetV2ForSemanticSegmentation,MobileNetV2ImageProcessor:()=>g.MobileNetV2ImageProcessor,MobileNetV2Model:()=>s.MobileNetV2Model,MobileNetV2PreTrainedModel:()=>s.MobileNetV2PreTrainedModel,MobileNetV3FeatureExtractor:()=>g.MobileNetV3FeatureExtractor,MobileNetV3ForImageClassification:()=>s.MobileNetV3ForImageClassification,MobileNetV3ForSemanticSegmentation:()=>s.MobileNetV3ForSemanticSegmentation,MobileNetV3ImageProcessor:()=>g.MobileNetV3ImageProcessor,MobileNetV3Model:()=>s.MobileNetV3Model,MobileNetV3PreTrainedModel:()=>s.MobileNetV3PreTrainedModel,MobileNetV4FeatureExtractor:()=>g.MobileNetV4FeatureExtractor,MobileNetV4ForImageClassification:()=>s.MobileNetV4ForImageClassification,MobileNetV4ForSemanticSegmentation:()=>s.MobileNetV4ForSemanticSegmentation,MobileNetV4ImageProcessor:()=>g.MobileNetV4ImageProcessor,MobileNetV4Model:()=>s.MobileNetV4Model,MobileNetV4PreTrainedModel:()=>s.MobileNetV4PreTrainedModel,MobileViTFeatureExtractor:()=>g.MobileViTFeatureExtractor,MobileViTForImageClassification:()=>s.MobileViTForImageClassification,MobileViTImageProcessor:()=>g.MobileViTImageProcessor,MobileViTModel:()=>s.MobileViTModel,MobileViTPreTrainedModel:()=>s.MobileViTPreTrainedModel,MobileViTV2ForImageClassification:()=>s.MobileViTV2ForImageClassification,MobileViTV2Model:()=>s.MobileViTV2Model,MobileViTV2PreTrainedModel:()=>s.MobileViTV2PreTrainedModel,ModelOutput:()=>s.ModelOutput,ModernBertForMaskedLM:()=>s.ModernBertForMaskedLM,ModernBertForSequenceClassification:()=>s.ModernBertForSequenceClassification,ModernBertForTokenClassification:()=>s.ModernBertForTokenClassification,ModernBertModel:()=>s.ModernBertModel,ModernBertPreTrainedModel:()=>s.ModernBertPreTrainedModel,Moondream1ForConditionalGeneration:()=>s.Moondream1ForConditionalGeneration,MoonshineFeatureExtractor:()=>p.MoonshineFeatureExtractor,MoonshineForConditionalGeneration:()=>s.MoonshineForConditionalGeneration,MoonshineModel:()=>s.MoonshineModel,MoonshinePreTrainedModel:()=>s.MoonshinePreTrainedModel,MoonshineProcessor:()=>M.MoonshineProcessor,MptForCausalLM:()=>s.MptForCausalLM,MptModel:()=>s.MptModel,MptPreTrainedModel:()=>s.MptPreTrainedModel,MultiModalityCausalLM:()=>s.MultiModalityCausalLM,MultiModalityPreTrainedModel:()=>s.MultiModalityPreTrainedModel,MusicgenForCausalLM:()=>s.MusicgenForCausalLM,MusicgenForConditionalGeneration:()=>s.MusicgenForConditionalGeneration,MusicgenModel:()=>s.MusicgenModel,MusicgenPreTrainedModel:()=>s.MusicgenPreTrainedModel,NllbTokenizer:()=>r.NllbTokenizer,NoBadWordsLogitsProcessor:()=>y.NoBadWordsLogitsProcessor,NoRepeatNGramLogitsProcessor:()=>y.NoRepeatNGramLogitsProcessor,NomicBertModel:()=>s.NomicBertModel,NomicBertPreTrainedModel:()=>s.NomicBertPreTrainedModel,NougatImageProcessor:()=>g.NougatImageProcessor,NougatTokenizer:()=>r.NougatTokenizer,OPTForCausalLM:()=>s.OPTForCausalLM,OPTModel:()=>s.OPTModel,OPTPreTrainedModel:()=>s.OPTPreTrainedModel,ObjectDetectionPipeline:()=>t.ObjectDetectionPipeline,Olmo2ForCausalLM:()=>s.Olmo2ForCausalLM,Olmo2Model:()=>s.Olmo2Model,Olmo2PreTrainedModel:()=>s.Olmo2PreTrainedModel,OlmoForCausalLM:()=>s.OlmoForCausalLM,OlmoModel:()=>s.OlmoModel,OlmoPreTrainedModel:()=>s.OlmoPreTrainedModel,OpenELMForCausalLM:()=>s.OpenELMForCausalLM,OpenELMModel:()=>s.OpenELMModel,OpenELMPreTrainedModel:()=>s.OpenELMPreTrainedModel,OwlViTFeatureExtractor:()=>g.OwlViTFeatureExtractor,OwlViTForObjectDetection:()=>s.OwlViTForObjectDetection,OwlViTImageProcessor:()=>g.OwlViTImageProcessor,OwlViTModel:()=>s.OwlViTModel,OwlViTPreTrainedModel:()=>s.OwlViTPreTrainedModel,OwlViTProcessor:()=>M.OwlViTProcessor,Owlv2ForObjectDetection:()=>s.Owlv2ForObjectDetection,Owlv2ImageProcessor:()=>g.Owlv2ImageProcessor,Owlv2Model:()=>s.Owlv2Model,Owlv2PreTrainedModel:()=>s.Owlv2PreTrainedModel,PaliGemmaForConditionalGeneration:()=>s.PaliGemmaForConditionalGeneration,PaliGemmaPreTrainedModel:()=>s.PaliGemmaPreTrainedModel,PaliGemmaProcessor:()=>M.PaliGemmaProcessor,PatchTSMixerForPrediction:()=>s.PatchTSMixerForPrediction,PatchTSMixerModel:()=>s.PatchTSMixerModel,PatchTSMixerPreTrainedModel:()=>s.PatchTSMixerPreTrainedModel,PatchTSTForPrediction:()=>s.PatchTSTForPrediction,PatchTSTModel:()=>s.PatchTSTModel,PatchTSTPreTrainedModel:()=>s.PatchTSTPreTrainedModel,Phi3ForCausalLM:()=>s.Phi3ForCausalLM,Phi3Model:()=>s.Phi3Model,Phi3PreTrainedModel:()=>s.Phi3PreTrainedModel,Phi3VForCausalLM:()=>s.Phi3VForCausalLM,Phi3VImageProcessor:()=>g.Phi3VImageProcessor,Phi3VPreTrainedModel:()=>s.Phi3VPreTrainedModel,Phi3VProcessor:()=>M.Phi3VProcessor,PhiForCausalLM:()=>s.PhiForCausalLM,PhiModel:()=>s.PhiModel,PhiPreTrainedModel:()=>s.PhiPreTrainedModel,Pipeline:()=>t.Pipeline,PreTrainedModel:()=>s.PreTrainedModel,PreTrainedTokenizer:()=>r.PreTrainedTokenizer,PretrainedConfig:()=>i.PretrainedConfig,PretrainedMixin:()=>s.PretrainedMixin,Processor:()=>w.Processor,PvtForImageClassification:()=>s.PvtForImageClassification,PvtImageProcessor:()=>g.PvtImageProcessor,PvtModel:()=>s.PvtModel,PvtPreTrainedModel:()=>s.PvtPreTrainedModel,PyAnnoteFeatureExtractor:()=>p.PyAnnoteFeatureExtractor,PyAnnoteForAudioFrameClassification:()=>s.PyAnnoteForAudioFrameClassification,PyAnnoteModel:()=>s.PyAnnoteModel,PyAnnotePreTrainedModel:()=>s.PyAnnotePreTrainedModel,PyAnnoteProcessor:()=>M.PyAnnoteProcessor,QuestionAnsweringModelOutput:()=>s.QuestionAnsweringModelOutput,QuestionAnsweringPipeline:()=>t.QuestionAnsweringPipeline,Qwen2ForCausalLM:()=>s.Qwen2ForCausalLM,Qwen2Model:()=>s.Qwen2Model,Qwen2PreTrainedModel:()=>s.Qwen2PreTrainedModel,Qwen2Tokenizer:()=>r.Qwen2Tokenizer,Qwen2VLForConditionalGeneration:()=>s.Qwen2VLForConditionalGeneration,Qwen2VLImageProcessor:()=>g.Qwen2VLImageProcessor,Qwen2VLPreTrainedModel:()=>s.Qwen2VLPreTrainedModel,Qwen2VLProcessor:()=>M.Qwen2VLProcessor,Qwen3ForCausalLM:()=>s.Qwen3ForCausalLM,Qwen3Model:()=>s.Qwen3Model,Qwen3PreTrainedModel:()=>s.Qwen3PreTrainedModel,RFDetrForObjectDetection:()=>s.RFDetrForObjectDetection,RFDetrModel:()=>s.RFDetrModel,RFDetrObjectDetectionOutput:()=>s.RFDetrObjectDetectionOutput,RFDetrPreTrainedModel:()=>s.RFDetrPreTrainedModel,RTDetrForObjectDetection:()=>s.RTDetrForObjectDetection,RTDetrImageProcessor:()=>g.RTDetrImageProcessor,RTDetrModel:()=>s.RTDetrModel,RTDetrObjectDetectionOutput:()=>s.RTDetrObjectDetectionOutput,RTDetrPreTrainedModel:()=>s.RTDetrPreTrainedModel,RTDetrV2ForObjectDetection:()=>s.RTDetrV2ForObjectDetection,RTDetrV2Model:()=>s.RTDetrV2Model,RTDetrV2ObjectDetectionOutput:()=>s.RTDetrV2ObjectDetectionOutput,RTDetrV2PreTrainedModel:()=>s.RTDetrV2PreTrainedModel,RawAudio:()=>a.RawAudio,RawImage:()=>l.RawImage,RawVideo:()=>c.RawVideo,RawVideoFrame:()=>c.RawVideoFrame,RepetitionPenaltyLogitsProcessor:()=>y.RepetitionPenaltyLogitsProcessor,ResNetForImageClassification:()=>s.ResNetForImageClassification,ResNetModel:()=>s.ResNetModel,ResNetPreTrainedModel:()=>s.ResNetPreTrainedModel,RoFormerForMaskedLM:()=>s.RoFormerForMaskedLM,RoFormerForQuestionAnswering:()=>s.RoFormerForQuestionAnswering,RoFormerForSequenceClassification:()=>s.RoFormerForSequenceClassification,RoFormerForTokenClassification:()=>s.RoFormerForTokenClassification,RoFormerModel:()=>s.RoFormerModel,RoFormerPreTrainedModel:()=>s.RoFormerPreTrainedModel,RoFormerTokenizer:()=>r.RoFormerTokenizer,RobertaForMaskedLM:()=>s.RobertaForMaskedLM,RobertaForQuestionAnswering:()=>s.RobertaForQuestionAnswering,RobertaForSequenceClassification:()=>s.RobertaForSequenceClassification,RobertaForTokenClassification:()=>s.RobertaForTokenClassification,RobertaModel:()=>s.RobertaModel,RobertaPreTrainedModel:()=>s.RobertaPreTrainedModel,RobertaTokenizer:()=>r.RobertaTokenizer,SamImageProcessor:()=>g.SamImageProcessor,SamImageSegmentationOutput:()=>s.SamImageSegmentationOutput,SamModel:()=>s.SamModel,SamPreTrainedModel:()=>s.SamPreTrainedModel,SamProcessor:()=>M.SamProcessor,SapiensForDepthEstimation:()=>s.SapiensForDepthEstimation,SapiensForNormalEstimation:()=>s.SapiensForNormalEstimation,SapiensForSemanticSegmentation:()=>s.SapiensForSemanticSegmentation,SapiensPreTrainedModel:()=>s.SapiensPreTrainedModel,SeamlessM4TFeatureExtractor:()=>p.SeamlessM4TFeatureExtractor,SegformerFeatureExtractor:()=>g.SegformerFeatureExtractor,SegformerForImageClassification:()=>s.SegformerForImageClassification,SegformerForSemanticSegmentation:()=>s.SegformerForSemanticSegmentation,SegformerImageProcessor:()=>g.SegformerImageProcessor,SegformerModel:()=>s.SegformerModel,SegformerPreTrainedModel:()=>s.SegformerPreTrainedModel,Seq2SeqLMOutput:()=>s.Seq2SeqLMOutput,SequenceClassifierOutput:()=>s.SequenceClassifierOutput,SiglipImageProcessor:()=>g.SiglipImageProcessor,SiglipModel:()=>s.SiglipModel,SiglipPreTrainedModel:()=>s.SiglipPreTrainedModel,SiglipTextModel:()=>s.SiglipTextModel,SiglipTokenizer:()=>r.SiglipTokenizer,SiglipVisionModel:()=>s.SiglipVisionModel,SmolVLMForConditionalGeneration:()=>s.SmolVLMForConditionalGeneration,SmolVLMImageProcessor:()=>g.SmolVLMImageProcessor,SmolVLMProcessor:()=>M.SmolVLMProcessor,SnacDecoderModel:()=>s.SnacDecoderModel,SnacEncoderModel:()=>s.SnacEncoderModel,SnacFeatureExtractor:()=>p.SnacFeatureExtractor,SnacModel:()=>s.SnacModel,SnacPreTrainedModel:()=>s.SnacPreTrainedModel,SpeechT5FeatureExtractor:()=>p.SpeechT5FeatureExtractor,SpeechT5ForSpeechToText:()=>s.SpeechT5ForSpeechToText,SpeechT5ForTextToSpeech:()=>s.SpeechT5ForTextToSpeech,SpeechT5HifiGan:()=>s.SpeechT5HifiGan,SpeechT5Model:()=>s.SpeechT5Model,SpeechT5PreTrainedModel:()=>s.SpeechT5PreTrainedModel,SpeechT5Processor:()=>M.SpeechT5Processor,SpeechT5Tokenizer:()=>r.SpeechT5Tokenizer,SqueezeBertForMaskedLM:()=>s.SqueezeBertForMaskedLM,SqueezeBertForQuestionAnswering:()=>s.SqueezeBertForQuestionAnswering,SqueezeBertForSequenceClassification:()=>s.SqueezeBertForSequenceClassification,SqueezeBertModel:()=>s.SqueezeBertModel,SqueezeBertPreTrainedModel:()=>s.SqueezeBertPreTrainedModel,SqueezeBertTokenizer:()=>r.SqueezeBertTokenizer,StableLmForCausalLM:()=>s.StableLmForCausalLM,StableLmModel:()=>s.StableLmModel,StableLmPreTrainedModel:()=>s.StableLmPreTrainedModel,Starcoder2ForCausalLM:()=>s.Starcoder2ForCausalLM,Starcoder2Model:()=>s.Starcoder2Model,Starcoder2PreTrainedModel:()=>s.Starcoder2PreTrainedModel,StoppingCriteria:()=>k.StoppingCriteria,StoppingCriteriaList:()=>k.StoppingCriteriaList,StyleTextToSpeech2Model:()=>s.StyleTextToSpeech2Model,StyleTextToSpeech2PreTrainedModel:()=>s.StyleTextToSpeech2PreTrainedModel,SummarizationPipeline:()=>t.SummarizationPipeline,SuppressTokensAtBeginLogitsProcessor:()=>y.SuppressTokensAtBeginLogitsProcessor,Swin2SRForImageSuperResolution:()=>s.Swin2SRForImageSuperResolution,Swin2SRImageProcessor:()=>g.Swin2SRImageProcessor,Swin2SRModel:()=>s.Swin2SRModel,Swin2SRPreTrainedModel:()=>s.Swin2SRPreTrainedModel,SwinForImageClassification:()=>s.SwinForImageClassification,SwinForSemanticSegmentation:()=>s.SwinForSemanticSegmentation,SwinModel:()=>s.SwinModel,SwinPreTrainedModel:()=>s.SwinPreTrainedModel,T5ForConditionalGeneration:()=>s.T5ForConditionalGeneration,T5Model:()=>s.T5Model,T5PreTrainedModel:()=>s.T5PreTrainedModel,T5Tokenizer:()=>r.T5Tokenizer,TableTransformerForObjectDetection:()=>s.TableTransformerForObjectDetection,TableTransformerModel:()=>s.TableTransformerModel,TableTransformerObjectDetectionOutput:()=>s.TableTransformerObjectDetectionOutput,TableTransformerPreTrainedModel:()=>s.TableTransformerPreTrainedModel,TemperatureLogitsWarper:()=>y.TemperatureLogitsWarper,Tensor:()=>d.Tensor,Text2TextGenerationPipeline:()=>t.Text2TextGenerationPipeline,TextClassificationPipeline:()=>t.TextClassificationPipeline,TextGenerationPipeline:()=>t.TextGenerationPipeline,TextStreamer:()=>b.TextStreamer,TextToAudioPipeline:()=>t.TextToAudioPipeline,TokenClassificationPipeline:()=>t.TokenClassificationPipeline,TokenClassifierOutput:()=>s.TokenClassifierOutput,TokenizerModel:()=>r.TokenizerModel,TopKLogitsWarper:()=>y.TopKLogitsWarper,TopPLogitsWarper:()=>y.TopPLogitsWarper,TrOCRForCausalLM:()=>s.TrOCRForCausalLM,TrOCRPreTrainedModel:()=>s.TrOCRPreTrainedModel,TranslationPipeline:()=>t.TranslationPipeline,UltravoxModel:()=>s.UltravoxModel,UltravoxPreTrainedModel:()=>s.UltravoxPreTrainedModel,UltravoxProcessor:()=>M.UltravoxProcessor,UniSpeechForCTC:()=>s.UniSpeechForCTC,UniSpeechForSequenceClassification:()=>s.UniSpeechForSequenceClassification,UniSpeechModel:()=>s.UniSpeechModel,UniSpeechPreTrainedModel:()=>s.UniSpeechPreTrainedModel,UniSpeechSatForAudioFrameClassification:()=>s.UniSpeechSatForAudioFrameClassification,UniSpeechSatForCTC:()=>s.UniSpeechSatForCTC,UniSpeechSatForSequenceClassification:()=>s.UniSpeechSatForSequenceClassification,UniSpeechSatModel:()=>s.UniSpeechSatModel,UniSpeechSatPreTrainedModel:()=>s.UniSpeechSatPreTrainedModel,VLChatProcessor:()=>M.VLChatProcessor,VLMImageProcessor:()=>g.VLMImageProcessor,ViTFeatureExtractor:()=>g.ViTFeatureExtractor,ViTForImageClassification:()=>s.ViTForImageClassification,ViTImageProcessor:()=>g.ViTImageProcessor,ViTMAEModel:()=>s.ViTMAEModel,ViTMAEPreTrainedModel:()=>s.ViTMAEPreTrainedModel,ViTMSNForImageClassification:()=>s.ViTMSNForImageClassification,ViTMSNModel:()=>s.ViTMSNModel,ViTMSNPreTrainedModel:()=>s.ViTMSNPreTrainedModel,ViTModel:()=>s.ViTModel,ViTPreTrainedModel:()=>s.ViTPreTrainedModel,VisionEncoderDecoderModel:()=>s.VisionEncoderDecoderModel,VitMatteForImageMatting:()=>s.VitMatteForImageMatting,VitMatteImageProcessor:()=>g.VitMatteImageProcessor,VitMattePreTrainedModel:()=>s.VitMattePreTrainedModel,VitPoseForPoseEstimation:()=>s.VitPoseForPoseEstimation,VitPoseImageProcessor:()=>g.VitPoseImageProcessor,VitPosePreTrainedModel:()=>s.VitPosePreTrainedModel,VitsModel:()=>s.VitsModel,VitsModelOutput:()=>s.VitsModelOutput,VitsPreTrainedModel:()=>s.VitsPreTrainedModel,VitsTokenizer:()=>r.VitsTokenizer,Wav2Vec2BertForCTC:()=>s.Wav2Vec2BertForCTC,Wav2Vec2BertForSequenceClassification:()=>s.Wav2Vec2BertForSequenceClassification,Wav2Vec2BertModel:()=>s.Wav2Vec2BertModel,Wav2Vec2BertPreTrainedModel:()=>s.Wav2Vec2BertPreTrainedModel,Wav2Vec2CTCTokenizer:()=>r.Wav2Vec2CTCTokenizer,Wav2Vec2FeatureExtractor:()=>p.Wav2Vec2FeatureExtractor,Wav2Vec2ForAudioFrameClassification:()=>s.Wav2Vec2ForAudioFrameClassification,Wav2Vec2ForCTC:()=>s.Wav2Vec2ForCTC,Wav2Vec2ForSequenceClassification:()=>s.Wav2Vec2ForSequenceClassification,Wav2Vec2Model:()=>s.Wav2Vec2Model,Wav2Vec2PreTrainedModel:()=>s.Wav2Vec2PreTrainedModel,Wav2Vec2Processor:()=>M.Wav2Vec2Processor,Wav2Vec2ProcessorWithLM:()=>M.Wav2Vec2ProcessorWithLM,WavLMForAudioFrameClassification:()=>s.WavLMForAudioFrameClassification,WavLMForCTC:()=>s.WavLMForCTC,WavLMForSequenceClassification:()=>s.WavLMForSequenceClassification,WavLMForXVector:()=>s.WavLMForXVector,WavLMModel:()=>s.WavLMModel,WavLMPreTrainedModel:()=>s.WavLMPreTrainedModel,WeSpeakerFeatureExtractor:()=>p.WeSpeakerFeatureExtractor,WeSpeakerResNetModel:()=>s.WeSpeakerResNetModel,WeSpeakerResNetPreTrainedModel:()=>s.WeSpeakerResNetPreTrainedModel,WhisperFeatureExtractor:()=>p.WhisperFeatureExtractor,WhisperForConditionalGeneration:()=>s.WhisperForConditionalGeneration,WhisperModel:()=>s.WhisperModel,WhisperPreTrainedModel:()=>s.WhisperPreTrainedModel,WhisperProcessor:()=>M.WhisperProcessor,WhisperTextStreamer:()=>b.WhisperTextStreamer,WhisperTimeStampLogitsProcessor:()=>y.WhisperTimeStampLogitsProcessor,WhisperTokenizer:()=>r.WhisperTokenizer,XLMForQuestionAnswering:()=>s.XLMForQuestionAnswering,XLMForSequenceClassification:()=>s.XLMForSequenceClassification,XLMForTokenClassification:()=>s.XLMForTokenClassification,XLMModel:()=>s.XLMModel,XLMPreTrainedModel:()=>s.XLMPreTrainedModel,XLMRobertaForMaskedLM:()=>s.XLMRobertaForMaskedLM,XLMRobertaForQuestionAnswering:()=>s.XLMRobertaForQuestionAnswering,XLMRobertaForSequenceClassification:()=>s.XLMRobertaForSequenceClassification,XLMRobertaForTokenClassification:()=>s.XLMRobertaForTokenClassification,XLMRobertaModel:()=>s.XLMRobertaModel,XLMRobertaPreTrainedModel:()=>s.XLMRobertaPreTrainedModel,XLMRobertaTokenizer:()=>r.XLMRobertaTokenizer,XLMTokenizer:()=>r.XLMTokenizer,XLMWithLMHeadModel:()=>s.XLMWithLMHeadModel,XVectorOutput:()=>s.XVectorOutput,YolosFeatureExtractor:()=>g.YolosFeatureExtractor,YolosForObjectDetection:()=>s.YolosForObjectDetection,YolosImageProcessor:()=>g.YolosImageProcessor,YolosModel:()=>s.YolosModel,YolosObjectDetectionOutput:()=>s.YolosObjectDetectionOutput,YolosPreTrainedModel:()=>s.YolosPreTrainedModel,ZeroShotAudioClassificationPipeline:()=>t.ZeroShotAudioClassificationPipeline,ZeroShotClassificationPipeline:()=>t.ZeroShotClassificationPipeline,ZeroShotImageClassificationPipeline:()=>t.ZeroShotImageClassificationPipeline,ZeroShotObjectDetectionPipeline:()=>t.ZeroShotObjectDetectionPipeline,bankers_round:()=>u.bankers_round,cat:()=>d.cat,cos_sim:()=>u.cos_sim,dot:()=>u.dot,dynamic_time_warping:()=>u.dynamic_time_warping,env:()=>e.env,full:()=>d.full,full_like:()=>d.full_like,getKeyValueShapes:()=>i.getKeyValueShapes,hamming:()=>a.hamming,hanning:()=>a.hanning,interpolate:()=>d.interpolate,interpolate_4d:()=>d.interpolate_4d,interpolate_data:()=>u.interpolate_data,is_chinese_char:()=>r.is_chinese_char,layer_norm:()=>d.layer_norm,load_image:()=>l.load_image,load_video:()=>c.load_video,log_softmax:()=>u.log_softmax,magnitude:()=>u.magnitude,matmul:()=>d.matmul,max:()=>u.max,mean:()=>d.mean,mean_pooling:()=>d.mean_pooling,medianFilter:()=>u.medianFilter,mel_filter_bank:()=>a.mel_filter_bank,min:()=>u.min,ones:()=>d.ones,ones_like:()=>d.ones_like,permute:()=>d.permute,permute_data:()=>u.permute_data,pipeline:()=>t.pipeline,quantize_embeddings:()=>d.quantize_embeddings,rand:()=>d.rand,read_audio:()=>a.read_audio,rfft:()=>d.rfft,round:()=>u.round,slice:()=>d.slice,softmax:()=>u.softmax,spectrogram:()=>a.spectrogram,stack:()=>d.stack,std_mean:()=>d.std_mean,topk:()=>d.topk,window_function:()=>a.window_function,zeros:()=>d.zeros,zeros_like:()=>d.zeros_like});var e=o("./src/env.js"),t=o("./src/pipelines.js"),s=o("./src/models.js"),r=o("./src/tokenizers.js"),i=o("./src/configs.js"),a=o("./src/utils/audio.js"),l=o("./src/utils/image.js"),c=o("./src/utils/video.js"),d=o("./src/utils/tensor.js"),u=o("./src/utils/maths.js"),_=o("./src/base/feature_extraction_utils.js"),p=o("./src/models/feature_extractors.js"),m=o("./src/models/auto/feature_extraction_auto.js"),h=o("./src/base/image_processors_utils.js"),g=o("./src/models/image_processors.js"),f=o("./src/models/auto/image_processing_auto.js"),w=o("./src/base/processing_utils.js"),M=o("./src/models/processors.js"),x=o("./src/models/auto/processing_auto.js"),b=o("./src/generation/streamers.js"),k=o("./src/generation/stopping_criteria.js"),y=o("./src/generation/logits_process.js")})();var i=exports;for(var a in n)i[a]=n[a];n.__esModule&&Object.defineProperty(i,"__esModule",{value:!0})})();
 //# sourceMappingURL=transformers.node.min.cjs.map