diff --git "a/assets/stt-worker-DZFESI5a.js" "b/assets/stt-worker-DZFESI5a.js" new file mode 100644--- /dev/null +++ "b/assets/stt-worker-DZFESI5a.js" @@ -0,0 +1,2856 @@ +var d1=Object.defineProperty;var p1=(e,r,t)=>r in e?d1(e,r,{enumerable:!0,configurable:!0,writable:!0,value:t}):e[r]=t;var de=(e,r,t)=>p1(e,typeof r!="symbol"?r+"":r,t);const Hh=new Map,ui=[],h1=(e,r,t)=>{if(r&&typeof r.init=="function"&&typeof r.createInferenceSessionHandler=="function"){const i=Hh.get(e);if(i===void 0)Hh.set(e,{backend:r,priority:t});else{if(i.priority>t)return;if(i.priority===t&&i.backend!==r)throw new Error(`cannot register backend "${e}" using priority ${t}`)}if(t>=0){const o=ui.indexOf(e);o!==-1&&ui.splice(o,1);for(let n=0;n{const r=Hh.get(e);if(!r)return"backend not found.";if(r.initialized)return r.backend;if(r.aborted)return r.error;{const t=!!r.initPromise;try{return t||(r.initPromise=r.backend.init(e)),await r.initPromise,r.initialized=!0,r.backend}catch(i){return t||(r.error=`${i}`,r.aborted=!0),r.error}finally{delete r.initPromise}}},m1=async e=>{const r=e.executionProviders||[],t=r.map(h=>typeof h=="string"?h:h.name),i=t.length===0?ui:t;let o;const n=[],u=new Set;for(const h of i){const f=await f1(h);typeof f=="string"?n.push({name:h,err:f}):(o||(o=f),o===f&&u.add(h))}if(!o)throw new Error(`no available backend found. ERR: ${n.map(h=>`[${h.name}] ${h.err}`).join(", ")}`);for(const{name:h,err:f}of n)t.includes(h)&&console.warn(`removing requested execution provider "${h}" from session options because it is not available: ${f}`);const d=r.filter(h=>u.has(typeof h=="string"?h:h.name));return[o,new Proxy(e,{get:(h,f)=>f==="executionProviders"?d:Reflect.get(h,f)})]},_1="1.23.0";let M_="warning";const Rs={wasm:{},webgl:{},webgpu:{},versions:{common:_1},set logLevel(e){if(e!==void 0){if(typeof e!="string"||["verbose","info","warning","error","fatal"].indexOf(e)===-1)throw new Error(`Unsupported logging level: ${e}`);M_=e}},get logLevel(){return M_}};Object.defineProperty(Rs,"logLevel",{enumerable:!0});const g1=Rs,y1=(e,r)=>{const t=typeof document<"u"?document.createElement("canvas"):new OffscreenCanvas(1,1);t.width=e.dims[3],t.height=e.dims[2];const i=t.getContext("2d");if(i!=null){let o,n;(r==null?void 0:r.tensorLayout)!==void 0&&r.tensorLayout==="NHWC"?(o=e.dims[2],n=e.dims[3]):(o=e.dims[3],n=e.dims[2]);const u=(r==null?void 0:r.format)!==void 0?r.format:"RGB",d=r==null?void 0:r.norm;let h,f;d===void 0||d.mean===void 0?h=[255,255,255,255]:typeof d.mean=="number"?h=[d.mean,d.mean,d.mean,d.mean]:(h=[d.mean[0],d.mean[1],d.mean[2],0],d.mean[3]!==void 0&&(h[3]=d.mean[3])),d===void 0||d.bias===void 0?f=[0,0,0,0]:typeof d.bias=="number"?f=[d.bias,d.bias,d.bias,d.bias]:(f=[d.bias[0],d.bias[1],d.bias[2],0],d.bias[3]!==void 0&&(f[3]=d.bias[3]));const y=n*o;let m=0,g=y,T=y*2,M=-1;u==="RGBA"?(m=0,g=y,T=y*2,M=y*3):u==="RGB"?(m=0,g=y,T=y*2):u==="RBG"&&(m=0,T=y,g=y*2);for(let I=0;I{const t=typeof document<"u"?document.createElement("canvas").getContext("2d"):new OffscreenCanvas(1,1).getContext("2d");let i;if(t!=null){let o,n,u;(r==null?void 0:r.tensorLayout)!==void 0&&r.tensorLayout==="NHWC"?(o=e.dims[2],n=e.dims[1],u=e.dims[3]):(o=e.dims[3],n=e.dims[2],u=e.dims[1]);const d=r!==void 0&&r.format!==void 0?r.format:"RGB",h=r==null?void 0:r.norm;let f,y;h===void 0||h.mean===void 0?f=[255,255,255,255]:typeof h.mean=="number"?f=[h.mean,h.mean,h.mean,h.mean]:(f=[h.mean[0],h.mean[1],h.mean[2],255],h.mean[3]!==void 0&&(f[3]=h.mean[3])),h===void 0||h.bias===void 0?y=[0,0,0,0]:typeof h.bias=="number"?y=[h.bias,h.bias,h.bias,h.bias]:(y=[h.bias[0],h.bias[1],h.bias[2],0],h.bias[3]!==void 0&&(y[3]=h.bias[3]));const m=n*o;if(r!==void 0&&(r.format!==void 0&&u===4&&r.format!=="RGBA"||u===3&&r.format!=="RGB"&&r.format!=="BGR"))throw new Error("Tensor format doesn't match input tensor dims");const g=4;let T=0,M=1,I=2,z=3,E=0,C=m,D=m*2,A=-1;d==="RGBA"?(E=0,C=m,D=m*2,A=m*3):d==="RGB"?(E=0,C=m,D=m*2):d==="RBG"&&(E=0,D=m,C=m*2),i=t.createImageData(o,n);for(let $=0;${if(e===void 0)throw new Error("Image buffer must be defined");if(r.height===void 0||r.width===void 0)throw new Error("Image height and width must be defined");if(r.tensorLayout==="NHWC")throw new Error("NHWC Tensor layout is not supported yet");const{height:t,width:i}=r,o=r.norm??{mean:255,bias:0};let n,u;typeof o.mean=="number"?n=[o.mean,o.mean,o.mean,o.mean]:n=[o.mean[0],o.mean[1],o.mean[2],o.mean[3]??255],typeof o.bias=="number"?u=[o.bias,o.bias,o.bias,o.bias]:u=[o.bias[0],o.bias[1],o.bias[2],o.bias[3]??0];const d=r.format!==void 0?r.format:"RGBA",h=r.tensorFormat!==void 0&&r.tensorFormat!==void 0?r.tensorFormat:"RGB",f=t*i,y=h==="RGBA"?new Float32Array(f*4):new Float32Array(f*3);let m=4,g=0,T=1,M=2,I=3,z=0,E=f,C=f*2,D=-1;d==="RGB"&&(m=3,g=0,T=1,M=2,I=-1),h==="RGBA"?D=f*3:h==="RBG"?(z=0,C=f,E=f*2):h==="BGR"&&(C=0,E=f,z=f*2);for(let $=0;${const t=typeof HTMLImageElement<"u"&&e instanceof HTMLImageElement,i=typeof ImageData<"u"&&e instanceof ImageData,o=typeof ImageBitmap<"u"&&e instanceof ImageBitmap,n=typeof e=="string";let u,d=r??{};const h=()=>{if(typeof document<"u")return document.createElement("canvas");if(typeof OffscreenCanvas<"u")return new OffscreenCanvas(1,1);throw new Error("Canvas is not supported")},f=y=>typeof HTMLCanvasElement<"u"&&y instanceof HTMLCanvasElement||y instanceof OffscreenCanvas?y.getContext("2d"):null;if(t){const y=h();y.width=e.width,y.height=e.height;const m=f(y);if(m!=null){let g=e.height,T=e.width;if(r!==void 0&&r.resizedHeight!==void 0&&r.resizedWidth!==void 0&&(g=r.resizedHeight,T=r.resizedWidth),r!==void 0){if(d=r,r.tensorFormat!==void 0)throw new Error("Image input config format must be RGBA for HTMLImageElement");d.tensorFormat="RGBA",d.height=g,d.width=T}else d.tensorFormat="RGBA",d.height=g,d.width=T;m.drawImage(e,0,0),u=m.getImageData(0,0,T,g).data}else throw new Error("Can not access image data")}else if(i){let y,m;if(r!==void 0&&r.resizedWidth!==void 0&&r.resizedHeight!==void 0?(y=r.resizedHeight,m=r.resizedWidth):(y=e.height,m=e.width),r!==void 0&&(d=r),d.format="RGBA",d.height=y,d.width=m,r!==void 0){const g=h();g.width=m,g.height=y;const T=f(g);if(T!=null)T.putImageData(e,0,0),u=T.getImageData(0,0,m,y).data;else throw new Error("Can not access image data")}else u=e.data}else if(o){if(r===void 0)throw new Error("Please provide image config with format for Imagebitmap");const y=h();y.width=e.width,y.height=e.height;const m=f(y);if(m!=null){const g=e.height,T=e.width;return m.drawImage(e,0,0,T,g),u=m.getImageData(0,0,T,g).data,d.height=g,d.width=T,Ef(u,d)}else throw new Error("Can not access image data")}else{if(n)return new Promise((y,m)=>{const g=h(),T=f(g);if(!e||!T)return m();const M=new Image;M.crossOrigin="Anonymous",M.src=e,M.onload=()=>{g.width=M.width,g.height=M.height,T.drawImage(M,0,0,g.width,g.height);const I=T.getImageData(0,0,g.width,g.height);d.height=g.height,d.width=g.width,y(Ef(I.data,d))}});throw new Error("Input data provided is not supported - aborted tensor creation")}if(u!==void 0)return Ef(u,d);throw new Error("Input data provided is not supported - aborted tensor creation")},M1=(e,r)=>{const{width:t,height:i,download:o,dispose:n}=r,u=[1,i,t,4];return new Ms({location:"texture",type:"float32",texture:e,dims:u,download:o,dispose:n})},v1=(e,r)=>{const{dataType:t,dims:i,download:o,dispose:n}=r;return new Ms({location:"gpu-buffer",type:t??"float32",gpuBuffer:e,dims:i,download:o,dispose:n})},x1=(e,r)=>{const{dataType:t,dims:i,download:o,dispose:n}=r;return new Ms({location:"ml-tensor",type:t??"float32",mlTensor:e,dims:i,download:o,dispose:n})},T1=(e,r,t)=>new Ms({location:"cpu-pinned",type:e,data:r,dims:t??[r.length]}),za=new Map([["float32",Float32Array],["uint8",Uint8Array],["int8",Int8Array],["uint16",Uint16Array],["int16",Int16Array],["int32",Int32Array],["bool",Uint8Array],["float64",Float64Array],["uint32",Uint32Array],["int4",Uint8Array],["uint4",Uint8Array]]),qh=new Map([[Float32Array,"float32"],[Uint8Array,"uint8"],[Int8Array,"int8"],[Uint16Array,"uint16"],[Int16Array,"int16"],[Int32Array,"int32"],[Float64Array,"float64"],[Uint32Array,"uint32"]]);let v_=!1;const C1=()=>{if(!v_){v_=!0;const e=typeof BigInt64Array<"u"&&BigInt64Array.from,r=typeof BigUint64Array<"u"&&BigUint64Array.from,t=globalThis.Float16Array,i=typeof t<"u"&&t.from;e&&(za.set("int64",BigInt64Array),qh.set(BigInt64Array,"int64")),r&&(za.set("uint64",BigUint64Array),qh.set(BigUint64Array,"uint64")),i?(za.set("float16",t),qh.set(t,"float16")):za.set("float16",Uint16Array)}},E1=e=>{let r=1;for(let t=0;t{switch(e.location){case"cpu":return new Ms(e.type,e.data,r);case"cpu-pinned":return new Ms({location:"cpu-pinned",data:e.data,type:e.type,dims:r});case"texture":return new Ms({location:"texture",texture:e.texture,type:e.type,dims:r});case"gpu-buffer":return new Ms({location:"gpu-buffer",gpuBuffer:e.gpuBuffer,type:e.type,dims:r});case"ml-tensor":return new Ms({location:"ml-tensor",mlTensor:e.mlTensor,type:e.type,dims:r});default:throw new Error(`tensorReshape: tensor location ${e.location} is not supported`)}};let Ms=class{constructor(r,t,i){C1();let o,n;if(typeof r=="object"&&"location"in r)switch(this.dataLocation=r.location,o=r.type,n=r.dims,r.location){case"cpu-pinned":{const d=za.get(o);if(!d)throw new TypeError(`unsupported type "${o}" to create tensor from pinned buffer`);if(!(r.data instanceof d))throw new TypeError(`buffer should be of type ${d.name}`);this.cpuData=r.data;break}case"texture":{if(o!=="float32")throw new TypeError(`unsupported type "${o}" to create tensor from texture`);this.gpuTextureData=r.texture,this.downloader=r.download,this.disposer=r.dispose;break}case"gpu-buffer":{if(o!=="float32"&&o!=="float16"&&o!=="int32"&&o!=="int64"&&o!=="uint32"&&o!=="uint8"&&o!=="bool"&&o!=="uint4"&&o!=="int4")throw new TypeError(`unsupported type "${o}" to create tensor from gpu buffer`);this.gpuBufferData=r.gpuBuffer,this.downloader=r.download,this.disposer=r.dispose;break}case"ml-tensor":{if(o!=="float32"&&o!=="float16"&&o!=="int32"&&o!=="int64"&&o!=="uint32"&&o!=="uint64"&&o!=="int8"&&o!=="uint8"&&o!=="bool"&&o!=="uint4"&&o!=="int4")throw new TypeError(`unsupported type "${o}" to create tensor from MLTensor`);this.mlTensorData=r.mlTensor,this.downloader=r.download,this.disposer=r.dispose;break}default:throw new Error(`Tensor constructor: unsupported location '${this.dataLocation}'`)}else{let d,h;if(typeof r=="string")if(o=r,h=i,r==="string"){if(!Array.isArray(t))throw new TypeError("A string tensor's data must be a string array.");d=t}else{const f=za.get(r);if(f===void 0)throw new TypeError(`Unsupported tensor type: ${r}.`);if(Array.isArray(t)){if(r==="float16"&&f===Uint16Array||r==="uint4"||r==="int4")throw new TypeError(`Creating a ${r} tensor from number array is not supported. Please use ${f.name} as data.`);r==="uint64"||r==="int64"?d=f.from(t,BigInt):d=f.from(t)}else if(t instanceof f)d=t;else if(t instanceof Uint8ClampedArray)if(r==="uint8")d=Uint8Array.from(t);else throw new TypeError("A Uint8ClampedArray tensor's data must be type of uint8");else if(r==="float16"&&t instanceof Uint16Array&&f!==Uint16Array)d=new globalThis.Float16Array(t.buffer,t.byteOffset,t.length);else throw new TypeError(`A ${o} tensor's data must be type of ${f}`)}else if(h=t,Array.isArray(r)){if(r.length===0)throw new TypeError("Tensor type cannot be inferred from an empty array.");const f=typeof r[0];if(f==="string")o="string",d=r;else if(f==="boolean")o="bool",d=Uint8Array.from(r);else throw new TypeError(`Invalid element type of data array: ${f}.`)}else if(r instanceof Uint8ClampedArray)o="uint8",d=Uint8Array.from(r);else{const f=qh.get(r.constructor);if(f===void 0)throw new TypeError(`Unsupported type for tensor data: ${r.constructor}.`);o=f,d=r}if(h===void 0)h=[d.length];else if(!Array.isArray(h))throw new TypeError("A tensor's dims must be a number array");n=h,this.cpuData=d,this.dataLocation="cpu"}const u=E1(n);if(this.cpuData&&u!==this.cpuData.length&&!((o==="uint4"||o==="int4")&&Math.ceil(u/2)===this.cpuData.length))throw new Error(`Tensor's size(${u}) does not match data length(${this.cpuData.length}).`);this.type=o,this.dims=n,this.size=u}static async fromImage(r,t){return b1(r,t)}static fromTexture(r,t){return M1(r,t)}static fromGpuBuffer(r,t){return v1(r,t)}static fromMLTensor(r,t){return x1(r,t)}static fromPinnedBuffer(r,t,i){return T1(r,t,i)}toDataURL(r){return y1(this,r)}toImageData(r){return w1(this,r)}get data(){if(this.ensureValid(),!this.cpuData)throw new Error("The data is not on CPU. Use `getData()` to download GPU data to CPU, or use `texture` or `gpuBuffer` property to access the GPU data directly.");return this.cpuData}get location(){return this.dataLocation}get texture(){if(this.ensureValid(),!this.gpuTextureData)throw new Error("The data is not stored as a WebGL texture.");return this.gpuTextureData}get gpuBuffer(){if(this.ensureValid(),!this.gpuBufferData)throw new Error("The data is not stored as a WebGPU buffer.");return this.gpuBufferData}get mlTensor(){if(this.ensureValid(),!this.mlTensorData)throw new Error("The data is not stored as a WebNN MLTensor.");return this.mlTensorData}async getData(r){switch(this.ensureValid(),this.dataLocation){case"cpu":case"cpu-pinned":return this.data;case"texture":case"gpu-buffer":case"ml-tensor":{if(!this.downloader)throw new Error("The current tensor is not created with a specified data downloader.");if(this.isDownloading)throw new Error("The current tensor is being downloaded.");try{this.isDownloading=!0;const t=await this.downloader();return this.downloader=void 0,this.dataLocation="cpu",this.cpuData=t,r&&this.disposer&&(this.disposer(),this.disposer=void 0),t}finally{this.isDownloading=!1}}default:throw new Error(`cannot get data from location: ${this.dataLocation}`)}}dispose(){if(this.isDownloading)throw new Error("The current tensor is being downloaded.");this.disposer&&(this.disposer(),this.disposer=void 0),this.cpuData=void 0,this.gpuTextureData=void 0,this.gpuBufferData=void 0,this.mlTensorData=void 0,this.downloader=void 0,this.isDownloading=void 0,this.dataLocation="none"}ensureValid(){if(this.dataLocation==="none")throw new Error("The tensor is disposed.")}reshape(r){if(this.ensureValid(),this.downloader||this.disposer)throw new Error("Cannot reshape a tensor that owns GPU resource.");return P1(this,r)}};const La=Ms,n0=(e,r)=>{(typeof Rs.trace>"u"?!Rs.wasm.trace:!Rs.trace)||console.timeStamp(`${e}::ORT::${r}`)},i0=(e,r)=>{var o;const t=((o=new Error().stack)==null?void 0:o.split(/\r\n|\r|\n/g))||[];let i=!1;for(let n=0;n{(typeof Rs.trace>"u"?!Rs.wasm.trace:!Rs.trace)||i0("BEGIN",e)},Em=e=>{(typeof Rs.trace>"u"?!Rs.wasm.trace:!Rs.trace)||i0("END",e)};let S1=class a0{constructor(r){this.handler=r}async run(r,t,i){Cm();const o={};let n={};if(typeof r!="object"||r===null||r instanceof La||Array.isArray(r))throw new TypeError("'feeds' must be an object that use input names as keys and OnnxValue as corresponding values.");let u=!0;if(typeof t=="object"){if(t===null)throw new TypeError("Unexpected argument[1]: cannot be null.");if(t instanceof La)throw new TypeError("'fetches' cannot be a Tensor");if(Array.isArray(t)){if(t.length===0)throw new TypeError("'fetches' cannot be an empty array.");u=!1;for(const f of t){if(typeof f!="string")throw new TypeError("'fetches' must be a string array or an object.");if(this.outputNames.indexOf(f)===-1)throw new RangeError(`'fetches' contains invalid output name: ${f}.`);o[f]=null}if(typeof i=="object"&&i!==null)n=i;else if(typeof i<"u")throw new TypeError("'options' must be an object.")}else{let f=!1;const y=Object.getOwnPropertyNames(t);for(const m of this.outputNames)if(y.indexOf(m)!==-1){const g=t[m];(g===null||g instanceof La)&&(f=!0,u=!1,o[m]=g)}if(f){if(typeof i=="object"&&i!==null)n=i;else if(typeof i<"u")throw new TypeError("'options' must be an object.")}else n=t}}else if(typeof t<"u")throw new TypeError("Unexpected argument[1]: must be 'fetches' or 'options'.");for(const f of this.inputNames)if(typeof r[f]>"u")throw new Error(`input '${f}' is missing in 'feeds'.`);if(u)for(const f of this.outputNames)o[f]=null;const d=await this.handler.run(r,o,n),h={};for(const f in d)if(Object.hasOwnProperty.call(d,f)){const y=d[f];y instanceof La?h[f]=y:h[f]=new La(y.type,y.data,y.dims)}return Em(),h}async release(){return this.handler.dispose()}static async create(r,t,i,o){Cm();let n,u={};if(typeof r=="string"){if(n=r,typeof t=="object"&&t!==null)u=t;else if(typeof t<"u")throw new TypeError("'options' must be an object.")}else if(r instanceof Uint8Array){if(n=r,typeof t=="object"&&t!==null)u=t;else if(typeof t<"u")throw new TypeError("'options' must be an object.")}else if(r instanceof ArrayBuffer||typeof SharedArrayBuffer<"u"&&r instanceof SharedArrayBuffer){const y=r;let m=0,g=r.byteLength;if(typeof t=="object"&&t!==null)u=t;else if(typeof t=="number"){if(m=t,!Number.isSafeInteger(m))throw new RangeError("'byteOffset' must be an integer.");if(m<0||m>=y.byteLength)throw new RangeError(`'byteOffset' is out of range [0, ${y.byteLength}).`);if(g=r.byteLength-m,typeof i=="number"){if(g=i,!Number.isSafeInteger(g))throw new RangeError("'byteLength' must be an integer.");if(g<=0||m+g>y.byteLength)throw new RangeError(`'byteLength' is out of range (0, ${y.byteLength-m}].`);if(typeof o=="object"&&o!==null)u=o;else if(typeof o<"u")throw new TypeError("'options' must be an object.")}else if(typeof i<"u")throw new TypeError("'byteLength' must be a number.")}else if(typeof t<"u")throw new TypeError("'options' must be an object.");n=new Uint8Array(y,m,g)}else throw new TypeError("Unexpected argument[0]: must be 'path' or 'buffer'.");const[d,h]=await m1(u),f=await d.createInferenceSessionHandler(n,h);return Em(),new a0(f)}startProfiling(){this.handler.startProfiling()}endProfiling(){this.handler.endProfiling()}get inputNames(){return this.handler.inputNames}get outputNames(){return this.handler.outputNames}get inputMetadata(){return this.handler.inputMetadata}get outputMetadata(){return this.handler.outputMetadata}};const k1=S1;var $1=Object.freeze({__proto__:null,InferenceSession:k1,TRACE:n0,TRACE_FUNC_BEGIN:Cm,TRACE_FUNC_END:Em,Tensor:La,env:g1,registerBackend:h1});/*! + * ONNX Runtime Web v1.23.0 + * Copyright (c) Microsoft Corporation. All rights reserved. + * Licensed under the MIT License. + */var Wm=Object.defineProperty,I1=Object.getOwnPropertyDescriptor,A1=Object.getOwnPropertyNames,O1=Object.prototype.hasOwnProperty,F1=(e=>typeof require<"u"?require:typeof Proxy<"u"?new Proxy(e,{get:(r,t)=>(typeof require<"u"?require:r)[t]}):e)(function(e){if(typeof require<"u")return require.apply(this,arguments);throw Error('Dynamic require of "'+e+'" is not supported')}),Ye=(e,r)=>()=>(e&&(r=e(e=0)),r),Wa=(e,r)=>{for(var t in r)Wm(e,t,{get:r[t],enumerable:!0})},D1=(e,r,t,i)=>{if(r&&typeof r=="object"||typeof r=="function")for(let o of A1(r))!O1.call(e,o)&&o!==t&&Wm(e,o,{get:()=>r[o],enumerable:!(i=I1(r,o))||i.enumerable});return e},wd=e=>D1(Wm({},"__esModule",{value:!0}),e),rd,Sn,fi,x_,o0,l0=Ye(()=>{rd=new Map,Sn=[],fi=(e,r,t)=>{if(r&&typeof r.init=="function"&&typeof r.createInferenceSessionHandler=="function"){let i=rd.get(e);if(i===void 0)rd.set(e,{backend:r,priority:t});else{if(i.priority>t)return;if(i.priority===t&&i.backend!==r)throw new Error(`cannot register backend "${e}" using priority ${t}`)}if(t>=0){let o=Sn.indexOf(e);o!==-1&&Sn.splice(o,1);for(let n=0;n{let r=rd.get(e);if(!r)return"backend not found.";if(r.initialized)return r.backend;if(r.aborted)return r.error;{let t=!!r.initPromise;try{return t||(r.initPromise=r.backend.init(e)),await r.initPromise,r.initialized=!0,r.backend}catch(i){return t||(r.error=`${i}`,r.aborted=!0),r.error}finally{delete r.initPromise}}},o0=async e=>{let r=e.executionProviders||[],t=r.map(h=>typeof h=="string"?h:h.name),i=t.length===0?Sn:t,o,n=[],u=new Set;for(let h of i){let f=await x_(h);typeof f=="string"?n.push({name:h,err:f}):(o||(o=f),o===f&&u.add(h))}if(!o)throw new Error(`no available backend found. ERR: ${n.map(h=>`[${h.name}] ${h.err}`).join(", ")}`);for(let{name:h,err:f}of n)t.includes(h)&&console.warn(`removing requested execution provider "${h}" from session options because it is not available: ${f}`);let d=r.filter(h=>u.has(typeof h=="string"?h:h.name));return[o,new Proxy(e,{get:(h,f)=>f==="executionProviders"?d:Reflect.get(h,f)})]}}),j1=Ye(()=>{l0()}),c0,L1=Ye(()=>{c0="1.23.0"}),Pf,bs,u0=Ye(()=>{L1(),Pf="warning",bs={wasm:{},webgl:{},webgpu:{},versions:{common:c0},set logLevel(e){if(e!==void 0){if(typeof e!="string"||["verbose","info","warning","error","fatal"].indexOf(e)===-1)throw new Error(`Unsupported logging level: ${e}`);Pf=e}},get logLevel(){return Pf}},Object.defineProperty(bs,"logLevel",{enumerable:!0})}),dr,z1=Ye(()=>{u0(),dr=bs}),d0,p0,B1=Ye(()=>{d0=(e,r)=>{let t=typeof document<"u"?document.createElement("canvas"):new OffscreenCanvas(1,1);t.width=e.dims[3],t.height=e.dims[2];let i=t.getContext("2d");if(i!=null){let o,n;(r==null?void 0:r.tensorLayout)!==void 0&&r.tensorLayout==="NHWC"?(o=e.dims[2],n=e.dims[3]):(o=e.dims[3],n=e.dims[2]);let u=(r==null?void 0:r.format)!==void 0?r.format:"RGB",d=r==null?void 0:r.norm,h,f;d===void 0||d.mean===void 0?h=[255,255,255,255]:typeof d.mean=="number"?h=[d.mean,d.mean,d.mean,d.mean]:(h=[d.mean[0],d.mean[1],d.mean[2],0],d.mean[3]!==void 0&&(h[3]=d.mean[3])),d===void 0||d.bias===void 0?f=[0,0,0,0]:typeof d.bias=="number"?f=[d.bias,d.bias,d.bias,d.bias]:(f=[d.bias[0],d.bias[1],d.bias[2],0],d.bias[3]!==void 0&&(f[3]=d.bias[3]));let y=n*o,m=0,g=y,T=y*2,M=-1;u==="RGBA"?(m=0,g=y,T=y*2,M=y*3):u==="RGB"?(m=0,g=y,T=y*2):u==="RBG"&&(m=0,T=y,g=y*2);for(let I=0;I{let t=typeof document<"u"?document.createElement("canvas").getContext("2d"):new OffscreenCanvas(1,1).getContext("2d"),i;if(t!=null){let o,n,u;(r==null?void 0:r.tensorLayout)!==void 0&&r.tensorLayout==="NHWC"?(o=e.dims[2],n=e.dims[1],u=e.dims[3]):(o=e.dims[3],n=e.dims[2],u=e.dims[1]);let d=r!==void 0&&r.format!==void 0?r.format:"RGB",h=r==null?void 0:r.norm,f,y;h===void 0||h.mean===void 0?f=[255,255,255,255]:typeof h.mean=="number"?f=[h.mean,h.mean,h.mean,h.mean]:(f=[h.mean[0],h.mean[1],h.mean[2],255],h.mean[3]!==void 0&&(f[3]=h.mean[3])),h===void 0||h.bias===void 0?y=[0,0,0,0]:typeof h.bias=="number"?y=[h.bias,h.bias,h.bias,h.bias]:(y=[h.bias[0],h.bias[1],h.bias[2],0],h.bias[3]!==void 0&&(y[3]=h.bias[3]));let m=n*o;if(r!==void 0&&(r.format!==void 0&&u===4&&r.format!=="RGBA"||u===3&&r.format!=="RGB"&&r.format!=="BGR"))throw new Error("Tensor format doesn't match input tensor dims");let g=4,T=0,M=1,I=2,z=3,E=0,C=m,D=m*2,A=-1;d==="RGBA"?(E=0,C=m,D=m*2,A=m*3):d==="RGB"?(E=0,C=m,D=m*2):d==="RBG"&&(E=0,D=m,C=m*2),i=t.createImageData(o,n);for(let $=0;${Um(),Fh=(e,r)=>{if(e===void 0)throw new Error("Image buffer must be defined");if(r.height===void 0||r.width===void 0)throw new Error("Image height and width must be defined");if(r.tensorLayout==="NHWC")throw new Error("NHWC Tensor layout is not supported yet");let{height:t,width:i}=r,o=r.norm??{mean:255,bias:0},n,u;typeof o.mean=="number"?n=[o.mean,o.mean,o.mean,o.mean]:n=[o.mean[0],o.mean[1],o.mean[2],o.mean[3]??255],typeof o.bias=="number"?u=[o.bias,o.bias,o.bias,o.bias]:u=[o.bias[0],o.bias[1],o.bias[2],o.bias[3]??0];let d=r.format!==void 0?r.format:"RGBA",h=r.tensorFormat!==void 0&&r.tensorFormat!==void 0?r.tensorFormat:"RGB",f=t*i,y=h==="RGBA"?new Float32Array(f*4):new Float32Array(f*3),m=4,g=0,T=1,M=2,I=3,z=0,E=f,C=f*2,D=-1;d==="RGB"&&(m=3,g=0,T=1,M=2,I=-1),h==="RGBA"?D=f*3:h==="RBG"?(z=0,C=f,E=f*2):h==="BGR"&&(C=0,E=f,z=f*2);for(let A=0;A{let t=typeof HTMLImageElement<"u"&&e instanceof HTMLImageElement,i=typeof ImageData<"u"&&e instanceof ImageData,o=typeof ImageBitmap<"u"&&e instanceof ImageBitmap,n=typeof e=="string",u,d=r??{},h=()=>{if(typeof document<"u")return document.createElement("canvas");if(typeof OffscreenCanvas<"u")return new OffscreenCanvas(1,1);throw new Error("Canvas is not supported")},f=y=>typeof HTMLCanvasElement<"u"&&y instanceof HTMLCanvasElement||y instanceof OffscreenCanvas?y.getContext("2d"):null;if(t){let y=h();y.width=e.width,y.height=e.height;let m=f(y);if(m!=null){let g=e.height,T=e.width;if(r!==void 0&&r.resizedHeight!==void 0&&r.resizedWidth!==void 0&&(g=r.resizedHeight,T=r.resizedWidth),r!==void 0){if(d=r,r.tensorFormat!==void 0)throw new Error("Image input config format must be RGBA for HTMLImageElement");d.tensorFormat="RGBA",d.height=g,d.width=T}else d.tensorFormat="RGBA",d.height=g,d.width=T;m.drawImage(e,0,0),u=m.getImageData(0,0,T,g).data}else throw new Error("Can not access image data")}else if(i){let y,m;if(r!==void 0&&r.resizedWidth!==void 0&&r.resizedHeight!==void 0?(y=r.resizedHeight,m=r.resizedWidth):(y=e.height,m=e.width),r!==void 0&&(d=r),d.format="RGBA",d.height=y,d.width=m,r!==void 0){let g=h();g.width=m,g.height=y;let T=f(g);if(T!=null)T.putImageData(e,0,0),u=T.getImageData(0,0,m,y).data;else throw new Error("Can not access image data")}else u=e.data}else if(o){if(r===void 0)throw new Error("Please provide image config with format for Imagebitmap");let y=h();y.width=e.width,y.height=e.height;let m=f(y);if(m!=null){let g=e.height,T=e.width;return m.drawImage(e,0,0,T,g),u=m.getImageData(0,0,T,g).data,d.height=g,d.width=T,Fh(u,d)}else throw new Error("Can not access image data")}else{if(n)return new Promise((y,m)=>{let g=h(),T=f(g);if(!e||!T)return m();let M=new Image;M.crossOrigin="Anonymous",M.src=e,M.onload=()=>{g.width=M.width,g.height=M.height,T.drawImage(M,0,0,g.width,g.height);let I=T.getImageData(0,0,g.width,g.height);d.height=g.height,d.width=g.width,y(Fh(I.data,d))}});throw new Error("Input data provided is not supported - aborted tensor creation")}if(u!==void 0)return Fh(u,d);throw new Error("Input data provided is not supported - aborted tensor creation")},f0=(e,r)=>{let{width:t,height:i,download:o,dispose:n}=r,u=[1,i,t,4];return new ms({location:"texture",type:"float32",texture:e,dims:u,download:o,dispose:n})},m0=(e,r)=>{let{dataType:t,dims:i,download:o,dispose:n}=r;return new ms({location:"gpu-buffer",type:t??"float32",gpuBuffer:e,dims:i,download:o,dispose:n})},_0=(e,r)=>{let{dataType:t,dims:i,download:o,dispose:n}=r;return new ms({location:"ml-tensor",type:t??"float32",mlTensor:e,dims:i,download:o,dispose:n})},g0=(e,r,t)=>new ms({location:"cpu-pinned",type:e,data:r,dims:t??[r.length]})}),di,hd,Sf,y0,N1=Ye(()=>{di=new Map([["float32",Float32Array],["uint8",Uint8Array],["int8",Int8Array],["uint16",Uint16Array],["int16",Int16Array],["int32",Int32Array],["bool",Uint8Array],["float64",Float64Array],["uint32",Uint32Array],["int4",Uint8Array],["uint4",Uint8Array]]),hd=new Map([[Float32Array,"float32"],[Uint8Array,"uint8"],[Int8Array,"int8"],[Uint16Array,"uint16"],[Int16Array,"int16"],[Int32Array,"int32"],[Float64Array,"float64"],[Uint32Array,"uint32"]]),Sf=!1,y0=()=>{if(!Sf){Sf=!0;let e=typeof BigInt64Array<"u"&&BigInt64Array.from,r=typeof BigUint64Array<"u"&&BigUint64Array.from,t=globalThis.Float16Array,i=typeof t<"u"&&t.from;e&&(di.set("int64",BigInt64Array),hd.set(BigInt64Array,"int64")),r&&(di.set("uint64",BigUint64Array),hd.set(BigUint64Array,"uint64")),i?(di.set("float16",t),hd.set(t,"float16")):di.set("float16",Uint16Array)}}}),w0,b0,V1=Ye(()=>{Um(),w0=e=>{let r=1;for(let t=0;t{switch(e.location){case"cpu":return new ms(e.type,e.data,r);case"cpu-pinned":return new ms({location:"cpu-pinned",data:e.data,type:e.type,dims:r});case"texture":return new ms({location:"texture",texture:e.texture,type:e.type,dims:r});case"gpu-buffer":return new ms({location:"gpu-buffer",gpuBuffer:e.gpuBuffer,type:e.type,dims:r});case"ml-tensor":return new ms({location:"ml-tensor",mlTensor:e.mlTensor,type:e.type,dims:r});default:throw new Error(`tensorReshape: tensor location ${e.location} is not supported`)}}}),ms,Um=Ye(()=>{B1(),R1(),N1(),V1(),ms=class{constructor(e,r,t){y0();let i,o;if(typeof e=="object"&&"location"in e)switch(this.dataLocation=e.location,i=e.type,o=e.dims,e.location){case"cpu-pinned":{let u=di.get(i);if(!u)throw new TypeError(`unsupported type "${i}" to create tensor from pinned buffer`);if(!(e.data instanceof u))throw new TypeError(`buffer should be of type ${u.name}`);this.cpuData=e.data;break}case"texture":{if(i!=="float32")throw new TypeError(`unsupported type "${i}" to create tensor from texture`);this.gpuTextureData=e.texture,this.downloader=e.download,this.disposer=e.dispose;break}case"gpu-buffer":{if(i!=="float32"&&i!=="float16"&&i!=="int32"&&i!=="int64"&&i!=="uint32"&&i!=="uint8"&&i!=="bool"&&i!=="uint4"&&i!=="int4")throw new TypeError(`unsupported type "${i}" to create tensor from gpu buffer`);this.gpuBufferData=e.gpuBuffer,this.downloader=e.download,this.disposer=e.dispose;break}case"ml-tensor":{if(i!=="float32"&&i!=="float16"&&i!=="int32"&&i!=="int64"&&i!=="uint32"&&i!=="uint64"&&i!=="int8"&&i!=="uint8"&&i!=="bool"&&i!=="uint4"&&i!=="int4")throw new TypeError(`unsupported type "${i}" to create tensor from MLTensor`);this.mlTensorData=e.mlTensor,this.downloader=e.download,this.disposer=e.dispose;break}default:throw new Error(`Tensor constructor: unsupported location '${this.dataLocation}'`)}else{let u,d;if(typeof e=="string")if(i=e,d=t,e==="string"){if(!Array.isArray(r))throw new TypeError("A string tensor's data must be a string array.");u=r}else{let h=di.get(e);if(h===void 0)throw new TypeError(`Unsupported tensor type: ${e}.`);if(Array.isArray(r)){if(e==="float16"&&h===Uint16Array||e==="uint4"||e==="int4")throw new TypeError(`Creating a ${e} tensor from number array is not supported. Please use ${h.name} as data.`);e==="uint64"||e==="int64"?u=h.from(r,BigInt):u=h.from(r)}else if(r instanceof h)u=r;else if(r instanceof Uint8ClampedArray)if(e==="uint8")u=Uint8Array.from(r);else throw new TypeError("A Uint8ClampedArray tensor's data must be type of uint8");else if(e==="float16"&&r instanceof Uint16Array&&h!==Uint16Array)u=new globalThis.Float16Array(r.buffer,r.byteOffset,r.length);else throw new TypeError(`A ${i} tensor's data must be type of ${h}`)}else if(d=r,Array.isArray(e)){if(e.length===0)throw new TypeError("Tensor type cannot be inferred from an empty array.");let h=typeof e[0];if(h==="string")i="string",u=e;else if(h==="boolean")i="bool",u=Uint8Array.from(e);else throw new TypeError(`Invalid element type of data array: ${h}.`)}else if(e instanceof Uint8ClampedArray)i="uint8",u=Uint8Array.from(e);else{let h=hd.get(e.constructor);if(h===void 0)throw new TypeError(`Unsupported type for tensor data: ${e.constructor}.`);i=h,u=e}if(d===void 0)d=[u.length];else if(!Array.isArray(d))throw new TypeError("A tensor's dims must be a number array");o=d,this.cpuData=u,this.dataLocation="cpu"}let n=w0(o);if(this.cpuData&&n!==this.cpuData.length&&!((i==="uint4"||i==="int4")&&Math.ceil(n/2)===this.cpuData.length))throw new Error(`Tensor's size(${n}) does not match data length(${this.cpuData.length}).`);this.type=i,this.dims=o,this.size=n}static async fromImage(e,r){return h0(e,r)}static fromTexture(e,r){return f0(e,r)}static fromGpuBuffer(e,r){return m0(e,r)}static fromMLTensor(e,r){return _0(e,r)}static fromPinnedBuffer(e,r,t){return g0(e,r,t)}toDataURL(e){return d0(this,e)}toImageData(e){return p0(this,e)}get data(){if(this.ensureValid(),!this.cpuData)throw new Error("The data is not on CPU. Use `getData()` to download GPU data to CPU, or use `texture` or `gpuBuffer` property to access the GPU data directly.");return this.cpuData}get location(){return this.dataLocation}get texture(){if(this.ensureValid(),!this.gpuTextureData)throw new Error("The data is not stored as a WebGL texture.");return this.gpuTextureData}get gpuBuffer(){if(this.ensureValid(),!this.gpuBufferData)throw new Error("The data is not stored as a WebGPU buffer.");return this.gpuBufferData}get mlTensor(){if(this.ensureValid(),!this.mlTensorData)throw new Error("The data is not stored as a WebNN MLTensor.");return this.mlTensorData}async getData(e){switch(this.ensureValid(),this.dataLocation){case"cpu":case"cpu-pinned":return this.data;case"texture":case"gpu-buffer":case"ml-tensor":{if(!this.downloader)throw new Error("The current tensor is not created with a specified data downloader.");if(this.isDownloading)throw new Error("The current tensor is being downloaded.");try{this.isDownloading=!0;let r=await this.downloader();return this.downloader=void 0,this.dataLocation="cpu",this.cpuData=r,e&&this.disposer&&(this.disposer(),this.disposer=void 0),r}finally{this.isDownloading=!1}}default:throw new Error(`cannot get data from location: ${this.dataLocation}`)}}dispose(){if(this.isDownloading)throw new Error("The current tensor is being downloaded.");this.disposer&&(this.disposer(),this.disposer=void 0),this.cpuData=void 0,this.gpuTextureData=void 0,this.gpuBufferData=void 0,this.mlTensorData=void 0,this.downloader=void 0,this.isDownloading=void 0,this.dataLocation="none"}ensureValid(){if(this.dataLocation==="none")throw new Error("The tensor is disposed.")}reshape(e){if(this.ensureValid(),this.downloader||this.disposer)throw new Error("Cannot reshape a tensor that owns GPU resource.");return b0(this,e)}}}),Bs,M0=Ye(()=>{Um(),Bs=ms}),bd,kf,Ns,vs,v0=Ye(()=>{u0(),bd=(e,r)=>{(typeof bs.trace>"u"?!bs.wasm.trace:!bs.trace)||console.timeStamp(`${e}::ORT::${r}`)},kf=(e,r)=>{var o;let t=((o=new Error().stack)==null?void 0:o.split(/\r\n|\r|\n/g))||[],i=!1;for(let n=0;n{(typeof bs.trace>"u"?!bs.wasm.trace:!bs.trace)||kf("BEGIN",e)},vs=e=>{(typeof bs.trace>"u"?!bs.wasm.trace:!bs.trace)||kf("END",e)}}),x0,W1=Ye(()=>{l0(),M0(),v0(),x0=class T0{constructor(r){this.handler=r}async run(r,t,i){Ns();let o={},n={};if(typeof r!="object"||r===null||r instanceof Bs||Array.isArray(r))throw new TypeError("'feeds' must be an object that use input names as keys and OnnxValue as corresponding values.");let u=!0;if(typeof t=="object"){if(t===null)throw new TypeError("Unexpected argument[1]: cannot be null.");if(t instanceof Bs)throw new TypeError("'fetches' cannot be a Tensor");if(Array.isArray(t)){if(t.length===0)throw new TypeError("'fetches' cannot be an empty array.");u=!1;for(let f of t){if(typeof f!="string")throw new TypeError("'fetches' must be a string array or an object.");if(this.outputNames.indexOf(f)===-1)throw new RangeError(`'fetches' contains invalid output name: ${f}.`);o[f]=null}if(typeof i=="object"&&i!==null)n=i;else if(typeof i<"u")throw new TypeError("'options' must be an object.")}else{let f=!1,y=Object.getOwnPropertyNames(t);for(let m of this.outputNames)if(y.indexOf(m)!==-1){let g=t[m];(g===null||g instanceof Bs)&&(f=!0,u=!1,o[m]=g)}if(f){if(typeof i=="object"&&i!==null)n=i;else if(typeof i<"u")throw new TypeError("'options' must be an object.")}else n=t}}else if(typeof t<"u")throw new TypeError("Unexpected argument[1]: must be 'fetches' or 'options'.");for(let f of this.inputNames)if(typeof r[f]>"u")throw new Error(`input '${f}' is missing in 'feeds'.`);if(u)for(let f of this.outputNames)o[f]=null;let d=await this.handler.run(r,o,n),h={};for(let f in d)if(Object.hasOwnProperty.call(d,f)){let y=d[f];y instanceof Bs?h[f]=y:h[f]=new Bs(y.type,y.data,y.dims)}return vs(),h}async release(){return this.handler.dispose()}static async create(r,t,i,o){Ns();let n,u={};if(typeof r=="string"){if(n=r,typeof t=="object"&&t!==null)u=t;else if(typeof t<"u")throw new TypeError("'options' must be an object.")}else if(r instanceof Uint8Array){if(n=r,typeof t=="object"&&t!==null)u=t;else if(typeof t<"u")throw new TypeError("'options' must be an object.")}else if(r instanceof ArrayBuffer||typeof SharedArrayBuffer<"u"&&r instanceof SharedArrayBuffer){let y=r,m=0,g=r.byteLength;if(typeof t=="object"&&t!==null)u=t;else if(typeof t=="number"){if(m=t,!Number.isSafeInteger(m))throw new RangeError("'byteOffset' must be an integer.");if(m<0||m>=y.byteLength)throw new RangeError(`'byteOffset' is out of range [0, ${y.byteLength}).`);if(g=r.byteLength-m,typeof i=="number"){if(g=i,!Number.isSafeInteger(g))throw new RangeError("'byteLength' must be an integer.");if(g<=0||m+g>y.byteLength)throw new RangeError(`'byteLength' is out of range (0, ${y.byteLength-m}].`);if(typeof o=="object"&&o!==null)u=o;else if(typeof o<"u")throw new TypeError("'options' must be an object.")}else if(typeof i<"u")throw new TypeError("'byteLength' must be a number.")}else if(typeof t<"u")throw new TypeError("'options' must be an object.");n=new Uint8Array(y,m,g)}else throw new TypeError("Unexpected argument[0]: must be 'path' or 'buffer'.");let[d,h]=await o0(u),f=await d.createInferenceSessionHandler(n,h);return vs(),new T0(f)}startProfiling(){this.handler.startProfiling()}endProfiling(){this.handler.endProfiling()}get inputNames(){return this.handler.inputNames}get outputNames(){return this.handler.outputNames}get inputMetadata(){return this.handler.inputMetadata}get outputMetadata(){return this.handler.outputMetadata}}}),Gm,U1=Ye(()=>{W1(),Gm=x0}),G1=Ye(()=>{}),K1=Ye(()=>{}),H1=Ye(()=>{}),q1=Ye(()=>{}),C0={};Wa(C0,{InferenceSession:()=>Gm,TRACE:()=>bd,TRACE_FUNC_BEGIN:()=>Ns,TRACE_FUNC_END:()=>vs,Tensor:()=>Bs,env:()=>dr,registerBackend:()=>fi});var Vs=Ye(()=>{j1(),z1(),U1(),M0(),G1(),K1(),v0(),H1(),q1()}),Km=Ye(()=>{}),E0={};Wa(E0,{default:()=>P0});var $f,If,P0,Q1=Ye(()=>{var e;Av(),yi(),Hm(),$f="ort-wasm-proxy-worker",If=((e=globalThis.self)==null?void 0:e.name)===$f,If&&(self.onmessage=r=>{let{type:t,in:i}=r.data;try{switch(t){case"init-wasm":qm(i.wasm).then(()=>{d_(i).then(()=>{postMessage({type:t})},o=>{postMessage({type:t,err:o})})},o=>{postMessage({type:t,err:o})});break;case"init-ep":{let{epName:o,env:n}=i;p_(n,o).then(()=>{postMessage({type:t})},u=>{postMessage({type:t,err:u})});break}case"copy-from":{let{buffer:o}=i,n=nf(o);postMessage({type:t,out:n});break}case"create":{let{model:o,options:n}=i;h_(o,n).then(u=>{postMessage({type:t,out:u})},u=>{postMessage({type:t,err:u})});break}case"release":f_(i),postMessage({type:t});break;case"run":{let{sessionId:o,inputIndices:n,inputs:u,outputIndices:d,options:h}=i;m_(o,n,u,d,new Array(d.length).fill(null),h).then(f=>{f.some(y=>y[3]!=="cpu")?postMessage({type:t,err:"Proxy does not support non-cpu tensor location."}):postMessage({type:t,out:f},g_([...u,...f]))},f=>{postMessage({type:t,err:f})});break}case"end-profiling":__(i),postMessage({type:t});break;default:}}catch(o){postMessage({type:t,err:o})}}),P0=If?null:r=>new Worker(r??fs,{type:"module",name:$f})}),S0={};Wa(S0,{default:()=>k0});var Af,Of,k0,T_,X1=Ye(()=>{var e,r;Of=(Af=import.meta.url,async function(t={}){var b_;var i,o,n=t,u=new Promise((s,a)=>{i=s,o=a}),d=typeof window=="object",h=typeof WorkerGlobalScope<"u",f=h&&((b_=self.name)==null?void 0:b_.startsWith("em-pthread"));n.mountExternalData=(s,a)=>{s.startsWith("./")&&(s=s.substring(2)),(n.MountedFiles||(n.MountedFiles=new Map)).set(s,a)},n.unmountExternalData=()=>{delete n.MountedFiles};var y=globalThis.SharedArrayBuffer??new WebAssembly.Memory({initial:0,maximum:0,shared:!0}).buffer.constructor;let m=s=>async(...a)=>{var l;try{if(n.jsepSessionState)throw new Error("Session already started");let c=n.jsepSessionState={sessionHandle:a[0],errors:[]},p=await s(...a);if(n.jsepSessionState!==c)throw new Error("Session mismatch");(l=n.jsepBackend)==null||l.flush();let _=c.errors;if(_.length>0){let w=await Promise.all(_);if(w=w.filter(v=>v),w.length>0)throw new Error(w.join(` +`))}return p}finally{n.jsepSessionState=null}};n.jsepInit=(s,a)=>{if(s==="webgpu"){[n.jsepBackend,n.jsepAlloc,n.jsepFree,n.jsepCopy,n.jsepCopyAsync,n.jsepCreateKernel,n.jsepReleaseKernel,n.jsepRunKernel,n.jsepCaptureBegin,n.jsepCaptureEnd,n.jsepReplay]=a;let l=n.jsepBackend;n.jsepRegisterBuffer=(c,p,_,w)=>l.registerBuffer(c,p,_,w),n.jsepGetBuffer=c=>l.getBuffer(c),n.jsepCreateDownloader=(c,p,_)=>l.createDownloader(c,p,_),n.jsepOnCreateSession=c=>{l.onCreateSession(c)},n.jsepOnReleaseSession=c=>{l.onReleaseSession(c)},n.jsepOnRunStart=c=>l.onRunStart(c),n.jsepUploadExternalBuffer=(c,p)=>{l.upload(c,p)}}else if(s==="webnn"){let l=a[0];[n.webnnReserveTensorId,n.webnnReleaseTensorId,n.webnnEnsureTensor,n.webnnUploadTensor,n.webnnDownloadTensor]=a.slice(1),n.webnnReleaseTensorId=n.webnnReleaseTensorId,n.webnnUploadTensor=n.webnnUploadTensor,n.webnnOnRunStart=c=>l.onRunStart(c),n.webnnOnRunEnd=l.onRunEnd.bind(l),n.webnnRegisterMLContext=(c,p)=>{l.registerMLContext(c,p)},n.webnnOnReleaseSession=c=>{l.onReleaseSession(c)},n.webnnCreateMLTensorDownloader=(c,p)=>l.createMLTensorDownloader(c,p),n.webnnRegisterMLTensor=(c,p,_,w)=>l.registerMLTensor(c,p,_,w),n.webnnCreateMLContext=c=>l.createMLContext(c),n.webnnRegisterMLConstant=(c,p,_,w,v,x)=>l.registerMLConstant(c,p,_,w,v,n.MountedFiles,x),n.webnnRegisterGraphInput=l.registerGraphInput.bind(l),n.webnnIsGraphInput=l.isGraphInput.bind(l),n.webnnRegisterGraphOutput=l.registerGraphOutput.bind(l),n.webnnIsGraphOutput=l.isGraphOutput.bind(l),n.webnnCreateTemporaryTensor=l.createTemporaryTensor.bind(l),n.webnnIsGraphInputOutputTypeSupported=l.isGraphInputOutputTypeSupported.bind(l)}};let g=()=>{let s=(a,l,c)=>(...p)=>{let _=Tt.currData,w=l==null?void 0:l(),v=a(...p),x=l==null?void 0:l();return w!==x&&(a=x,c(w),c=null,l=null),Tt.currData!=_?Tt.whenDone():v};(()=>{for(let a of["_OrtAppendExecutionProvider","_OrtCreateSession","_OrtRun","_OrtRunWithBinding","_OrtBindInput"])n[a]=s(n[a],()=>n[a],l=>n[a]=l)})(),m!==void 0&&(n._OrtRun=m(n._OrtRun),n._OrtRunWithBinding=m(n._OrtRunWithBinding)),g=void 0};n.asyncInit=()=>{g==null||g()};var T,M,I=Object.assign({},n),z=(s,a)=>{throw a},E="";(d||h)&&(h?E=self.location.href:typeof document<"u"&&document.currentScript&&(E=document.currentScript.src),Af&&(E=Af),E=E.startsWith("blob:")?"":E.slice(0,E.replace(/[?#].*/,"").lastIndexOf("/")+1),h&&(M=s=>{var a=new XMLHttpRequest;return a.open("GET",s,!1),a.responseType="arraybuffer",a.send(null),new Uint8Array(a.response)}),T=async s=>{if(q(s))return new Promise((l,c)=>{var p=new XMLHttpRequest;p.open("GET",s,!0),p.responseType="arraybuffer",p.onload=()=>{p.status==200||p.status==0&&p.response?l(p.response):c(p.status)},p.onerror=c,p.send(null)});var a=await fetch(s,{credentials:"same-origin"});if(a.ok)return a.arrayBuffer();throw new Error(a.status+" : "+a.url)});var C=console.log.bind(console),D=console.error.bind(console),A=C,$=D;Object.assign(n,I),I=null;var P,k,O,R,U,te,se,K,pe,re,oe,ge,le,Se=n.wasmBinary,Ce=!1,q=s=>s.startsWith("file://");function N(){return P.buffer!=R.buffer&&me(),R}function Q(){return P.buffer!=R.buffer&&me(),U}function ue(){return P.buffer!=R.buffer&&me(),te}function Ae(){return P.buffer!=R.buffer&&me(),se}function we(){return P.buffer!=R.buffer&&me(),K}function ae(){return P.buffer!=R.buffer&&me(),pe}function Ze(){return P.buffer!=R.buffer&&me(),re}function et(){return P.buffer!=R.buffer&&me(),le}if(f){let s=function(...c){var p=c.join(" ");console.error(p)},a=function(...c){var p=c.join(" ");postMessage({cmd:"alert",text:p,threadId:vn()})},l=function(c){try{var p=c.data,_=p.cmd;if(_==="load"){let w=[];self.onmessage=v=>w.push(v),self.startWorker=v=>{postMessage({cmd:"loaded"});for(let x of w)l(x);self.onmessage=l};for(let v of p.handlers)n[v]&&!n[v].proxy||(n[v]=(...x)=>{postMessage({cmd:"callHandler",handler:v,args:x})},v=="print"&&(A=n[v]),v=="printErr"&&($=n[v]));P=p.wasmMemory,me(),Ee(p.wasmModule)}else if(_==="run"){$e(p.pthread_ptr),Ui(p.pthread_ptr,0,0,1,0,0),ne.threadInitTLS(),Yr(p.pthread_ptr),Z||(Oo(),Z=!0);try{Xe(p.start_routine,p.arg)}catch(w){if(w!="unwind")throw w}}else p.target==="setimmediate"||(_==="checkMailbox"?Z&&ss():_&&($(`worker: received unknown command ${_}`),$(p)))}catch(w){throw Ko(),w}};var Ee,Z=!1;$=s,self.alert=a,self.onunhandledrejection=c=>{throw c.reason||c},self.onmessage=l}function me(){var s=P.buffer;n.HEAP8=R=new Int8Array(s),n.HEAP16=te=new Int16Array(s),n.HEAPU8=U=new Uint8Array(s),n.HEAPU16=se=new Uint16Array(s),n.HEAP32=K=new Int32Array(s),n.HEAPU32=pe=new Uint32Array(s),n.HEAPF32=re=new Float32Array(s),n.HEAPF64=le=new Float64Array(s),n.HEAP64=oe=new BigInt64Array(s),n.HEAPU64=ge=new BigUint64Array(s)}function X(){if(f)return startWorker(n);Y.qf()}f||(P=new WebAssembly.Memory({initial:256,maximum:65536,shared:!0}),me());var Oe,Fe=0,Ve=null;function Re(s){Fe++}function Ue(s){if(--Fe==0&&Ve){var a=Ve;Ve=null,a()}}function Ge(s){$(s="Aborted("+s+")"),Ce=!0,s+=". Build with -sASSERTIONS for more info.";var a=new WebAssembly.RuntimeError(s);throw o(a),a}async function Je(s,a,l){if(!s&&typeof WebAssembly.instantiateStreaming=="function"&&!q(a))try{var c=fetch(a,{credentials:"same-origin"});return await WebAssembly.instantiateStreaming(c,l)}catch(p){$(`wasm streaming compile failed: ${p}`),$("falling back to ArrayBuffer instantiation")}return async function(p,_){try{var w=await async function(v){if(!Se)try{var x=await T(v);return new Uint8Array(x)}catch{}return function(j){if(j==Oe&&Se)return new Uint8Array(Se);if(M)return M(j);throw"both async and sync fetching of the wasm failed"}(v)}(p);return await WebAssembly.instantiate(w,_)}catch(v){$(`failed to asynchronously prepare wasm: ${v}`),Ge(v)}}(a,l)}function at(){return{a:{rd:st,Md:Ke,v:Qt,_:Nt,b:pr,o:Kr,A:xs,F:Ws,r:Ts,uc:en,u:rs,Bb:tn,pf:Gs,g:tr,tc:Cs,of:Hs,nf:ut,mf:bt,lf:cr,kf:ls,jf:Es,hf:Xr,qd:Ps,gf:Ss,ff:ks,ef:$s,df:gs,Ze:Fn,nd:qs,Ye:L,Xe:ct,md:dt,Ma:At,na:qt,We:Ht,rc:Qs,Ve:ur,Ue:Ir,Te:Br,Se:Ar,Re:yr,ld:Mr,Qe:Yr,Pe:vr,kd:bi,Oe:xd,Ne:Ua,Ab:Td,sa:Ga,Me:ve,Le:Ed,Ke:Ha,ma:Xa,zb:uf,Je:Sd,Z:Ja,qc:kd,Ie:Mi,ha:$d,lb:Id,He:Od,Ge:Fd,fb:Dd,Fe:jd,Ee:eo,De:sn,Ce:to,Be:Ld,Ae:so,ze:no,cf:Bd,C:Ti,pc:Rd,jd:Nd,ye:io,Tb:oo,xe:Vd,we:co,ba:ro,ve:Wd,Sb:Ud,id:uo,ue:po,te:Hd,se:fo,re:mo,hd:qd,gd:Qd,fd:_o,qe:Xd,oc:Jd,wa:go,pe:Yd,oe:Zd,ne:yo,me:ep,le:tp,ke:Ei,bf:Si,af:wo,je:G,sc:bo,pd:ki,$e:Mo,od:xo,Rb:Qp,Qb:cx,qa:Xx,S:Yx,R:Jx,Ca:Da,Ga:Hp,ed:rx,nc:lx,ie:qx,Y:$a,y:Fp,c:ff,dd:mh,va:Bp,he:Bx,cd:Nx,h:pf,eb:Hx,Ja:i1,d:df,ja:S,k:Ap,ge:c1,j:Ip,Pb:Dx,s:kp,t:th,q:gf,yb:mr,O:Up,Ba:Hv,oa:oh,mc:Gr,kb:cn,lc:Px,xb:hx,bd:F,ad:eh,$c:ke,jb:Nr,_c:_e,kc:l1,ra:dn,Ob:Wp,wb:Ih,da:Kx,la:Ah,Zc:Sr,N:ar,Yc:V,Ya:Aa,H:Dp,Xc:Zx,vb:Mh,fe:Ie,Aa:sx,La:Th,B:Rp,jc:It,Wc:Be,Vc:wf,Uc:kh,ic:Sh,ee:rt,Xa:wh,aa:Zv,ib:r1,hc:n1,hb:a1,db:e1,Tc:Dt,Sc:o1,gc:vf,fa:qp,K:sh,D:jp,Ia:xh,U:Fa,Rc:fx,Qc:lt,Pc:$h,X:Pt,e:Sp,cb:kr,l:Gu,Oc:dh,Wa:Vx,pa:Or,Nb:Ct,i:hf,Nc:uh,V:bf,ua:xr,Mb:mx,ub:Bt,f:Ku,de:Ox,ce:$x,m:mf,Mc:Rt,n:qu,be:Ax,Lc:kt,ae:kx,Kc:es,Jc:Rx,$d:gx,p:Op,_d:zx,Va:un,Lb:Ys,Ua:Pn,fc:dx,E:Vp,G:Zp,J:Kp,bb:Fx,Zd:Ix,Kb:wr,$:lh,ga:fh,Ha:xx,Fa:ns,Ta:nx,Yd:jt,ab:vh,Ea:ch,Ic:Tx,gb:vx,Jb:Mx,Sa:_x,Xd:ln,za:ax,Wd:Ex,Vd:Yv,Ud:or,ec:xf,Ra:ix,W:nh,tb:ox,Hc:hs,Ib:Tf,ta:ex,Qa:_h,Gc:Ph,Hb:yf,sb:Qe,dc:Cx,P:ph,Gb:Jv,ka:Jp,Td:gh,ea:Qv,rb:ah,cc:Sx,Fc:bx,bc:Ux,Ec:Wx,Dc:Lx,$a:on,Cc:t1,qb:yh,z:En,T:rh,ya:_f,Ka:Yp,Sd:bh,ac:_t,Fb:Lp,L:Np,pb:px,Da:Xv,Bc:s1,Ac:ux,Rd:ih,x:zp,I:Qu,ca:$p,Qd:tx,$b:Mf,zc:wx,Pa:jx,_a:Xp,_b:Zu,Za:hh,Zb:qv,Pd:Hu,Od:Eh,Yb:yx,Xb:Ia,ob:Gx,yc:Gp,Wb:nr,ia:ed,xc:Wt,M:hr,nb:Ch,Nd:Qx,xa:td,Q:Oh,Eb:Ju,Oa:Xu,Na:Tr,Db:Yu,Cb:Oa,w:sp,a:P,_e:as,Ld:To,Kd:np,mb:ip,wc:Co,Jd:ap,Id:op,Hd:Eo,Gd:lp,Fd:cp,Vb:Po,Ed:up,Dd:dp,Cd:So,Bd:pp,Ad:ko,Ub:fp,zd:mp,yd:$o,xd:gp,wd:Io,vd:yp,ud:wp,vc:Ao,td:bp,sd:Mp}}}var ot={1635076:(s,a,l,c,p)=>{if(n===void 0||!n.MountedFiles)return 1;let _=Jt(Number(s>>>0));_.startsWith("./")&&(_=_.substring(2));let w=n.MountedFiles.get(_);if(!w)return 2;let v=Number(a>>>0),x=Number(l>>>0),j=Number(c>>>0),B=p;if(v+x>w.byteLength)return 3;try{let H=w.subarray(v,v+x);switch(B){case 0:Q().set(H,j>>>0);break;case 1:n.webgpuUploadExternalBuffer?n.webgpuUploadExternalBuffer(j,H):n.jsepUploadExternalBuffer(j,H);break;default:return 4}return 0}catch{return 4}},1635900:(s,a,l)=>{let c=Number(s),p=Number(a),_=Jt(l),w=new Uint8Array(p);w.set(Q().subarray(c>>>0,c+p>>>0));{let v=new File([w],_,{type:"application/octet-stream"}),x=URL.createObjectURL(v);window.open(x,"_blank")}},1636424:(s,a,l)=>{n.webnnUploadTensor(s,Q().subarray(a>>>0,a+l>>>0))},1636488:()=>n.webnnReserveTensorId(),1636530:s=>{n.webnnReleaseTensorId(s)},1636567:()=>{n.jsepCaptureBegin()},1636598:()=>{n.jsepCaptureEnd()},1636627:()=>{n.jsepReplay()},1636652:s=>n.jsepAlloc(s),1636685:s=>n.jsepFree(s),1636717:(s,a,l)=>{n.jsepCopy(Number(s),Number(a),Number(l),!0)},1636780:(s,a,l)=>{n.jsepCopy(Number(s),Number(a),Number(l))},1636837:()=>typeof wasmOffsetConverter<"u",1636894:s=>{n.jsepCreateKernel("Abs",s,void 0)},1636945:s=>{n.jsepCreateKernel("Neg",s,void 0)},1636996:s=>{n.jsepCreateKernel("Floor",s,void 0)},1637049:s=>{n.jsepCreateKernel("Ceil",s,void 0)},1637101:s=>{n.jsepCreateKernel("Reciprocal",s,void 0)},1637159:s=>{n.jsepCreateKernel("Sqrt",s,void 0)},1637211:s=>{n.jsepCreateKernel("Exp",s,void 0)},1637262:s=>{n.jsepCreateKernel("Erf",s,void 0)},1637313:s=>{n.jsepCreateKernel("Sigmoid",s,void 0)},1637368:(s,a,l)=>{n.jsepCreateKernel("HardSigmoid",s,{alpha:a,beta:l})},1637447:s=>{n.jsepCreateKernel("Log",s,void 0)},1637498:s=>{n.jsepCreateKernel("Sin",s,void 0)},1637549:s=>{n.jsepCreateKernel("Cos",s,void 0)},1637600:s=>{n.jsepCreateKernel("Tan",s,void 0)},1637651:s=>{n.jsepCreateKernel("Asin",s,void 0)},1637703:s=>{n.jsepCreateKernel("Acos",s,void 0)},1637755:s=>{n.jsepCreateKernel("Atan",s,void 0)},1637807:s=>{n.jsepCreateKernel("Sinh",s,void 0)},1637859:s=>{n.jsepCreateKernel("Cosh",s,void 0)},1637911:s=>{n.jsepCreateKernel("Asinh",s,void 0)},1637964:s=>{n.jsepCreateKernel("Acosh",s,void 0)},1638017:s=>{n.jsepCreateKernel("Atanh",s,void 0)},1638070:s=>{n.jsepCreateKernel("Tanh",s,void 0)},1638122:s=>{n.jsepCreateKernel("Not",s,void 0)},1638173:(s,a,l)=>{n.jsepCreateKernel("Clip",s,{min:a,max:l})},1638242:s=>{n.jsepCreateKernel("Clip",s,void 0)},1638294:(s,a)=>{n.jsepCreateKernel("Elu",s,{alpha:a})},1638352:s=>{n.jsepCreateKernel("Gelu",s,void 0)},1638404:s=>{n.jsepCreateKernel("Relu",s,void 0)},1638456:(s,a)=>{n.jsepCreateKernel("LeakyRelu",s,{alpha:a})},1638520:(s,a)=>{n.jsepCreateKernel("ThresholdedRelu",s,{alpha:a})},1638590:(s,a)=>{n.jsepCreateKernel("Cast",s,{to:a})},1638648:s=>{n.jsepCreateKernel("Add",s,void 0)},1638699:s=>{n.jsepCreateKernel("Sub",s,void 0)},1638750:s=>{n.jsepCreateKernel("Mul",s,void 0)},1638801:s=>{n.jsepCreateKernel("Div",s,void 0)},1638852:s=>{n.jsepCreateKernel("Pow",s,void 0)},1638903:s=>{n.jsepCreateKernel("Equal",s,void 0)},1638956:s=>{n.jsepCreateKernel("Greater",s,void 0)},1639011:s=>{n.jsepCreateKernel("GreaterOrEqual",s,void 0)},1639073:s=>{n.jsepCreateKernel("Less",s,void 0)},1639125:s=>{n.jsepCreateKernel("LessOrEqual",s,void 0)},1639184:(s,a,l,c,p)=>{n.jsepCreateKernel("ReduceMean",s,{keepDims:!!a,noopWithEmptyAxes:!!l,axes:c?Array.from(we().subarray(Number(c)>>>0,Number(p)>>>0)):[]})},1639359:(s,a,l,c,p)=>{n.jsepCreateKernel("ReduceMax",s,{keepDims:!!a,noopWithEmptyAxes:!!l,axes:c?Array.from(we().subarray(Number(c)>>>0,Number(p)>>>0)):[]})},1639533:(s,a,l,c,p)=>{n.jsepCreateKernel("ReduceMin",s,{keepDims:!!a,noopWithEmptyAxes:!!l,axes:c?Array.from(we().subarray(Number(c)>>>0,Number(p)>>>0)):[]})},1639707:(s,a,l,c,p)=>{n.jsepCreateKernel("ReduceProd",s,{keepDims:!!a,noopWithEmptyAxes:!!l,axes:c?Array.from(we().subarray(Number(c)>>>0,Number(p)>>>0)):[]})},1639882:(s,a,l,c,p)=>{n.jsepCreateKernel("ReduceSum",s,{keepDims:!!a,noopWithEmptyAxes:!!l,axes:c?Array.from(we().subarray(Number(c)>>>0,Number(p)>>>0)):[]})},1640056:(s,a,l,c,p)=>{n.jsepCreateKernel("ReduceL1",s,{keepDims:!!a,noopWithEmptyAxes:!!l,axes:c?Array.from(we().subarray(Number(c)>>>0,Number(p)>>>0)):[]})},1640229:(s,a,l,c,p)=>{n.jsepCreateKernel("ReduceL2",s,{keepDims:!!a,noopWithEmptyAxes:!!l,axes:c?Array.from(we().subarray(Number(c)>>>0,Number(p)>>>0)):[]})},1640402:(s,a,l,c,p)=>{n.jsepCreateKernel("ReduceLogSum",s,{keepDims:!!a,noopWithEmptyAxes:!!l,axes:c?Array.from(we().subarray(Number(c)>>>0,Number(p)>>>0)):[]})},1640579:(s,a,l,c,p)=>{n.jsepCreateKernel("ReduceSumSquare",s,{keepDims:!!a,noopWithEmptyAxes:!!l,axes:c?Array.from(we().subarray(Number(c)>>>0,Number(p)>>>0)):[]})},1640759:(s,a,l,c,p)=>{n.jsepCreateKernel("ReduceLogSumExp",s,{keepDims:!!a,noopWithEmptyAxes:!!l,axes:c?Array.from(we().subarray(Number(c)>>>0,Number(p)>>>0)):[]})},1640939:s=>{n.jsepCreateKernel("Where",s,void 0)},1640992:(s,a,l)=>{n.jsepCreateKernel("Transpose",s,{perm:a?Array.from(we().subarray(Number(a)>>>0,Number(l)>>>0)):[]})},1641116:(s,a,l,c)=>{n.jsepCreateKernel("DepthToSpace",s,{blocksize:a,mode:Jt(l),format:c?"NHWC":"NCHW"})},1641249:(s,a,l,c)=>{n.jsepCreateKernel("DepthToSpace",s,{blocksize:a,mode:Jt(l),format:c?"NHWC":"NCHW"})},1641382:(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe)=>{n.jsepCreateKernel("ConvTranspose",s,{format:x?"NHWC":"NCHW",autoPad:a,dilations:[l],group:c,kernelShape:[p],pads:[_,w],strides:[v],wIsConst:()=>!!N()[j>>>0],outputPadding:B?Array.from(we().subarray(Number(B)>>>0,Number(H)>>>0)):[],outputShape:ee?Array.from(we().subarray(Number(ee)>>>0,Number(fe)>>>0)):[],activation:Jt(xe)})},1641815:(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe)=>{n.jsepCreateKernel("ConvTranspose",s,{format:v?"NHWC":"NCHW",autoPad:a,dilations:Array.from(we().subarray(Number(l)>>>0,2+(Number(l)>>>0)>>>0)),group:c,kernelShape:Array.from(we().subarray(Number(p)>>>0,2+(Number(p)>>>0)>>>0)),pads:Array.from(we().subarray(Number(_)>>>0,4+(Number(_)>>>0)>>>0)),strides:Array.from(we().subarray(Number(w)>>>0,2+(Number(w)>>>0)>>>0)),wIsConst:()=>!!N()[x>>>0],outputPadding:j?Array.from(we().subarray(Number(j)>>>0,Number(B)>>>0)):[],outputShape:H?Array.from(we().subarray(Number(H)>>>0,Number(ee)>>>0)):[],activation:Jt(fe)})},1642476:(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe)=>{n.jsepCreateKernel("ConvTranspose",s,{format:x?"NHWC":"NCHW",autoPad:a,dilations:[l],group:c,kernelShape:[p],pads:[_,w],strides:[v],wIsConst:()=>!!N()[j>>>0],outputPadding:B?Array.from(we().subarray(Number(B)>>>0,Number(H)>>>0)):[],outputShape:ee?Array.from(we().subarray(Number(ee)>>>0,Number(fe)>>>0)):[],activation:Jt(xe)})},1642909:(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe)=>{n.jsepCreateKernel("ConvTranspose",s,{format:v?"NHWC":"NCHW",autoPad:a,dilations:Array.from(we().subarray(Number(l)>>>0,2+(Number(l)>>>0)>>>0)),group:c,kernelShape:Array.from(we().subarray(Number(p)>>>0,2+(Number(p)>>>0)>>>0)),pads:Array.from(we().subarray(Number(_)>>>0,4+(Number(_)>>>0)>>>0)),strides:Array.from(we().subarray(Number(w)>>>0,2+(Number(w)>>>0)>>>0)),wIsConst:()=>!!N()[x>>>0],outputPadding:j?Array.from(we().subarray(Number(j)>>>0,Number(B)>>>0)):[],outputShape:H?Array.from(we().subarray(Number(H)>>>0,Number(ee)>>>0)):[],activation:Jt(fe)})},1643570:(s,a)=>{n.jsepCreateKernel("GlobalAveragePool",s,{format:a?"NHWC":"NCHW"})},1643661:(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe)=>{n.jsepCreateKernel("AveragePool",s,{format:fe?"NHWC":"NCHW",auto_pad:a,ceil_mode:l,count_include_pad:c,storage_order:p,dilations:_?Array.from(we().subarray(Number(_)>>>0,Number(w)>>>0)):[],kernel_shape:v?Array.from(we().subarray(Number(v)>>>0,Number(x)>>>0)):[],pads:j?Array.from(we().subarray(Number(j)>>>0,Number(B)>>>0)):[],strides:H?Array.from(we().subarray(Number(H)>>>0,Number(ee)>>>0)):[]})},1644140:(s,a)=>{n.jsepCreateKernel("GlobalAveragePool",s,{format:a?"NHWC":"NCHW"})},1644231:(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe)=>{n.jsepCreateKernel("AveragePool",s,{format:fe?"NHWC":"NCHW",auto_pad:a,ceil_mode:l,count_include_pad:c,storage_order:p,dilations:_?Array.from(we().subarray(Number(_)>>>0,Number(w)>>>0)):[],kernel_shape:v?Array.from(we().subarray(Number(v)>>>0,Number(x)>>>0)):[],pads:j?Array.from(we().subarray(Number(j)>>>0,Number(B)>>>0)):[],strides:H?Array.from(we().subarray(Number(H)>>>0,Number(ee)>>>0)):[]})},1644710:(s,a)=>{n.jsepCreateKernel("GlobalMaxPool",s,{format:a?"NHWC":"NCHW"})},1644797:(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe)=>{n.jsepCreateKernel("MaxPool",s,{format:fe?"NHWC":"NCHW",auto_pad:a,ceil_mode:l,count_include_pad:c,storage_order:p,dilations:_?Array.from(we().subarray(Number(_)>>>0,Number(w)>>>0)):[],kernel_shape:v?Array.from(we().subarray(Number(v)>>>0,Number(x)>>>0)):[],pads:j?Array.from(we().subarray(Number(j)>>>0,Number(B)>>>0)):[],strides:H?Array.from(we().subarray(Number(H)>>>0,Number(ee)>>>0)):[]})},1645272:(s,a)=>{n.jsepCreateKernel("GlobalMaxPool",s,{format:a?"NHWC":"NCHW"})},1645359:(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe)=>{n.jsepCreateKernel("MaxPool",s,{format:fe?"NHWC":"NCHW",auto_pad:a,ceil_mode:l,count_include_pad:c,storage_order:p,dilations:_?Array.from(we().subarray(Number(_)>>>0,Number(w)>>>0)):[],kernel_shape:v?Array.from(we().subarray(Number(v)>>>0,Number(x)>>>0)):[],pads:j?Array.from(we().subarray(Number(j)>>>0,Number(B)>>>0)):[],strides:H?Array.from(we().subarray(Number(H)>>>0,Number(ee)>>>0)):[]})},1645834:(s,a,l,c,p)=>{n.jsepCreateKernel("Gemm",s,{alpha:a,beta:l,transA:c,transB:p})},1645938:s=>{n.jsepCreateKernel("MatMul",s,void 0)},1645992:(s,a,l,c)=>{n.jsepCreateKernel("ArgMax",s,{keepDims:!!a,selectLastIndex:!!l,axis:c})},1646100:(s,a,l,c)=>{n.jsepCreateKernel("ArgMin",s,{keepDims:!!a,selectLastIndex:!!l,axis:c})},1646208:(s,a)=>{n.jsepCreateKernel("Softmax",s,{axis:a})},1646271:(s,a)=>{n.jsepCreateKernel("Concat",s,{axis:a})},1646331:(s,a,l,c,p)=>{n.jsepCreateKernel("Split",s,{axis:a,numOutputs:l,splitSizes:c?Array.from(we().subarray(Number(c)>>>0,Number(p)>>>0)):[]})},1646487:s=>{n.jsepCreateKernel("Expand",s,void 0)},1646541:(s,a)=>{n.jsepCreateKernel("Gather",s,{axis:Number(a)})},1646612:(s,a)=>{n.jsepCreateKernel("GatherElements",s,{axis:Number(a)})},1646691:(s,a)=>{n.jsepCreateKernel("GatherND",s,{batch_dims:Number(a)})},1646770:(s,a,l,c,p,_,w,v,x,j,B)=>{n.jsepCreateKernel("Resize",s,{antialias:a,axes:l?Array.from(we().subarray(Number(l)>>>0,Number(c)>>>0)):[],coordinateTransformMode:Jt(p),cubicCoeffA:_,excludeOutside:w,extrapolationValue:v,keepAspectRatioPolicy:Jt(x),mode:Jt(j),nearestMode:Jt(B)})},1647132:(s,a,l,c,p,_,w)=>{n.jsepCreateKernel("Slice",s,{starts:a?Array.from(we().subarray(Number(a)>>>0,Number(l)>>>0)):[],ends:c?Array.from(we().subarray(Number(c)>>>0,Number(p)>>>0)):[],axes:_?Array.from(we().subarray(Number(_)>>>0,Number(w)>>>0)):[]})},1647396:s=>{n.jsepCreateKernel("Tile",s,void 0)},1647448:(s,a,l)=>{n.jsepCreateKernel("InstanceNormalization",s,{epsilon:a,format:l?"NHWC":"NCHW"})},1647562:(s,a,l)=>{n.jsepCreateKernel("InstanceNormalization",s,{epsilon:a,format:l?"NHWC":"NCHW"})},1647676:s=>{n.jsepCreateKernel("Range",s,void 0)},1647729:(s,a)=>{n.jsepCreateKernel("Einsum",s,{equation:Jt(a)})},1647810:(s,a,l,c,p)=>{n.jsepCreateKernel("Pad",s,{mode:a,value:l,pads:c?Array.from(we().subarray(Number(c)>>>0,Number(p)>>>0)):[]})},1647953:(s,a,l,c,p,_)=>{n.jsepCreateKernel("BatchNormalization",s,{epsilon:a,momentum:l,spatial:!!p,trainingMode:!!c,format:_?"NHWC":"NCHW"})},1648122:(s,a,l,c,p,_)=>{n.jsepCreateKernel("BatchNormalization",s,{epsilon:a,momentum:l,spatial:!!p,trainingMode:!!c,format:_?"NHWC":"NCHW"})},1648291:(s,a,l)=>{n.jsepCreateKernel("CumSum",s,{exclusive:Number(a),reverse:Number(l)})},1648388:(s,a,l)=>{n.jsepCreateKernel("DequantizeLinear",s,{axis:a,blockSize:l})},1648478:(s,a,l,c,p)=>{n.jsepCreateKernel("GridSample",s,{align_corners:a,mode:Jt(l),padding_mode:Jt(c),format:p?"NHWC":"NCHW"})},1648648:(s,a,l,c,p)=>{n.jsepCreateKernel("GridSample",s,{align_corners:a,mode:Jt(l),padding_mode:Jt(c),format:p?"NHWC":"NCHW"})},1648818:(s,a)=>{n.jsepCreateKernel("ScatterND",s,{reduction:Jt(a)})},1648903:(s,a,l,c,p,_,w,v,x)=>{n.jsepCreateKernel("Attention",s,{numHeads:a,isUnidirectional:l,maskFilterValue:c,scale:p,doRotary:_,qkvHiddenSizes:w?Array.from(we().subarray(Number(v)>>>0,Number(v)+w>>>0)):[],pastPresentShareBuffer:!!x})},1649175:s=>{n.jsepCreateKernel("BiasAdd",s,void 0)},1649230:s=>{n.jsepCreateKernel("BiasSplitGelu",s,void 0)},1649291:s=>{n.jsepCreateKernel("FastGelu",s,void 0)},1649347:(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De)=>{n.jsepCreateKernel("Conv",s,{format:H?"NHWC":"NCHW",auto_pad:a,dilations:l?Array.from(we().subarray(Number(l)>>>0,Number(c)>>>0)):[],group:p,kernel_shape:_?Array.from(we().subarray(Number(_)>>>0,Number(w)>>>0)):[],pads:v?Array.from(we().subarray(Number(v)>>>0,Number(x)>>>0)):[],strides:j?Array.from(we().subarray(Number(j)>>>0,Number(B)>>>0)):[],w_is_const:()=>!!N()[Number(ee)>>>0],activation:Jt(fe),activation_params:xe?Array.from(Ze().subarray(Number(xe)>>>0,Number(De)>>>0)):[]})},1649931:s=>{n.jsepCreateKernel("Gelu",s,void 0)},1649983:(s,a,l,c,p,_,w,v,x)=>{n.jsepCreateKernel("GroupQueryAttention",s,{numHeads:a,kvNumHeads:l,scale:c,softcap:p,doRotary:_,rotaryInterleaved:w,smoothSoftmax:v,localWindowSize:x})},1650200:(s,a,l,c)=>{n.jsepCreateKernel("LayerNormalization",s,{axis:a,epsilon:l,simplified:!!c})},1650311:(s,a,l,c)=>{n.jsepCreateKernel("LayerNormalization",s,{axis:a,epsilon:l,simplified:!!c})},1650422:(s,a,l,c,p,_)=>{n.jsepCreateKernel("MatMulNBits",s,{k:a,n:l,accuracyLevel:c,bits:p,blockSize:_})},1650549:(s,a,l,c,p,_)=>{n.jsepCreateKernel("MultiHeadAttention",s,{numHeads:a,isUnidirectional:l,maskFilterValue:c,scale:p,doRotary:_})},1650708:(s,a)=>{n.jsepCreateKernel("QuickGelu",s,{alpha:a})},1650772:(s,a,l,c,p)=>{n.jsepCreateKernel("RotaryEmbedding",s,{interleaved:!!a,numHeads:l,rotaryEmbeddingDim:c,scale:p})},1650911:(s,a,l)=>{n.jsepCreateKernel("SkipLayerNormalization",s,{epsilon:a,simplified:!!l})},1651013:(s,a,l)=>{n.jsepCreateKernel("SkipLayerNormalization",s,{epsilon:a,simplified:!!l})},1651115:(s,a,l,c)=>{n.jsepCreateKernel("GatherBlockQuantized",s,{gatherAxis:a,quantizeAxis:l,blockSize:c})},1651236:s=>{n.jsepReleaseKernel(s)},1651270:(s,a)=>n.jsepRunKernel(Number(s),Number(a),n.jsepSessionState.sessionHandle,n.jsepSessionState.errors)};function Ke(s,a,l){return Tt.handleAsync(async()=>{await n.jsepCopyAsync(Number(s),Number(a),Number(l))})}function st(){return typeof wasmOffsetConverter<"u"}class ft{constructor(a){de(this,"name","ExitStatus");this.message=`Program terminated with exit(${a})`,this.status=a}}var yt=s=>{s.terminate(),s.onmessage=a=>{}},St=s=>{var a=ne.pthreads[s];ne.returnWorkerToPool(a)},vt=[],He=s=>{var a=ne.getNewWorker();if(!a)return 6;ne.runningWorkers.push(a),ne.pthreads[s.pthread_ptr]=a,a.pthread_ptr=s.pthread_ptr;var l={cmd:"run",start_routine:s.startRoutine,arg:s.arg,pthread_ptr:s.pthread_ptr};return a.postMessage(l,s.transferList),0},xt=0,Vt=()=>xt>0,he=()=>Jo(),ce=s=>Xo(s),Cr=s=>Xi(s),er=s=>s<-9007199254740992||s>9007199254740992?NaN:Number(s),ht=(s,a,l,...c)=>{for(var p=2*c.length,_=he(),w=Cr(8*p),v=w>>>3,x=0;x>>0]=j)}var B=Ho(s,a,p,w,l);return ce(_),B};function as(s){if(f)return ht(0,0,1,s);O=s,Vt()||(ne.terminateAllThreads(),Ce=!0),z(0,new ft(s))}var W=s=>{if(s instanceof ft||s=="unwind")return O;z(0,s)};function ye(s){if(f)return ht(1,0,0,s);G(s)}var G=(s,a)=>{if(O=s,f)throw ye(s),"unwind";as(s)},ne={unusedWorkers:[],runningWorkers:[],tlsInitFunctions:[],pthreads:{},init(){f||ne.initMainThread()},initMainThread(){for(var s,a=n.numThreads-1;a--;)ne.allocateUnusedWorker();s=()=>{Re(),ne.loadWasmModuleToAllWorkers(()=>Ue())},vt.unshift(s)},terminateAllThreads:()=>{for(var s of ne.runningWorkers)yt(s);for(var s of ne.unusedWorkers)yt(s);ne.unusedWorkers=[],ne.runningWorkers=[],ne.pthreads={}},returnWorkerToPool:s=>{var a=s.pthread_ptr;delete ne.pthreads[a],ne.unusedWorkers.push(s),ne.runningWorkers.splice(ne.runningWorkers.indexOf(s),1),s.pthread_ptr=0,Gi(a)},threadInitTLS(){ne.tlsInitFunctions.forEach(s=>s())},loadWasmModuleToWorker:s=>new Promise(a=>{s.onmessage=p=>{var _=p.data,w=_.cmd;if(_.targetThread&&_.targetThread!=vn()){var v=ne.pthreads[_.targetThread];v?v.postMessage(_,_.transferList):$(`Internal error! Worker sent a message "${w}" to target pthread ${_.targetThread}, but that thread no longer exists!`)}else w==="checkMailbox"?ss():w==="spawnThread"?He(_):w==="cleanupThread"?St(_.thread):w==="loaded"?(s.loaded=!0,a(s)):w==="alert"?alert(`Thread ${_.threadId}: ${_.text}`):_.target==="setimmediate"?s.postMessage(_):w==="callHandler"?n[_.handler](..._.args):w&&$(`worker sent an unknown command ${w}`)},s.onerror=p=>{throw $(`worker sent an error! ${p.filename}:${p.lineno}: ${p.message}`),p};var l=[];for(var c of[])n.propertyIsEnumerable(c)&&l.push(c);s.postMessage({cmd:"load",handlers:l,wasmMemory:P,wasmModule:k})}),loadWasmModuleToAllWorkers(s){if(f)return s();Promise.all(ne.unusedWorkers.map(ne.loadWasmModuleToWorker)).then(s)},allocateUnusedWorker(){var s;s=new Worker((()=>{let a=URL;return import.meta.url>"file:"&&import.meta.url<"file;"?new a("ort.bundle.min.mjs",import.meta.url):new URL(import.meta.url)})(),{type:"module",workerData:"em-pthread",name:"em-pthread"}),ne.unusedWorkers.push(s)},getNewWorker:()=>(ne.unusedWorkers.length==0&&(ne.allocateUnusedWorker(),ne.loadWasmModuleToWorker(ne.unusedWorkers[0])),ne.unusedWorkers.pop())},$e=s=>{me();var a=ae()[s+52>>>2>>>0],l=ae()[s+56>>>2>>>0];Qo(a,a-l),ce(a)},Xe=(s,a)=>{var l;xt=0,l=Gn(s,a),Vt()?O=l:Ki(l)},Le=[],Ft=0;function Qt(s){var a=new mt(s>>>=0);return a.get_caught()||(a.set_caught(!0),Ft--),a.set_rethrown(!1),Le.push(a),Ji(s),el(s)}var wt=0,Nt=()=>{Me(0,0);var s=Le.pop();Yo(s.excPtr),wt=0};class mt{constructor(a){this.excPtr=a,this.ptr=a-24}set_type(a){ae()[this.ptr+4>>>2>>>0]=a}get_type(){return ae()[this.ptr+4>>>2>>>0]}set_destructor(a){ae()[this.ptr+8>>>2>>>0]=a}get_destructor(){return ae()[this.ptr+8>>>2>>>0]}set_caught(a){a=a?1:0,N()[this.ptr+12>>>0]=a}get_caught(){return N()[this.ptr+12>>>0]!=0}set_rethrown(a){a=a?1:0,N()[this.ptr+13>>>0]=a}get_rethrown(){return N()[this.ptr+13>>>0]!=0}init(a,l){this.set_adjusted_ptr(0),this.set_type(a),this.set_destructor(l)}set_adjusted_ptr(a){ae()[this.ptr+16>>>2>>>0]=a}get_adjusted_ptr(){return ae()[this.ptr+16>>>2>>>0]}}function tr(s){throw wt||(wt=s>>>=0),wt}var rr=s=>Qi(s),Fr=s=>{var a=wt;if(!a)return rr(0),0;var l=new mt(a);l.set_adjusted_ptr(a);var c=l.get_type();if(!c)return rr(0),a;for(var p of s){if(p===0||p===c)break;var _=l.ptr+16;if(Zo(p,c,_))return rr(p),a}return rr(c),a};function pr(){return Fr([])}function Kr(s){return Fr([s>>>=0])}function xs(s,a){return Fr([s>>>=0,a>>>=0])}function Ws(s,a,l){return Fr([s>>>=0,a>>>=0,l>>>=0])}function Ts(s,a,l,c){return Fr([s>>>=0,a>>>=0,l>>>=0,c>>>=0])}var en=()=>{var s=Le.pop();s||Ge("no exception to throw");var a=s.excPtr;throw s.get_rethrown()||(Le.push(s),s.set_rethrown(!0),s.set_caught(!1),Ft++),wt=a};function rs(s,a,l){throw a>>>=0,l>>>=0,new mt(s>>>=0).init(a,l),Ft++,wt=s}var tn=()=>Ft;function os(s,a,l,c){return f?ht(2,0,1,s,a,l,c):Gs(s,a,l,c)}var Us=()=>y!==void 0;function Gs(s,a,l,c){if(s>>>=0,a>>>=0,l>>>=0,c>>>=0,!Us())return 6;var p=[];if(f&&p.length===0)return os(s,a,l,c);var _={startRoutine:l,pthread_ptr:s,arg:c,transferList:p};return f?(_.cmd="spawnThread",postMessage(_,p),0):He(_)}var Ks=typeof TextDecoder<"u"?new TextDecoder:void 0,Wr=(s,a=0,l=NaN)=>{for(var c=(a>>>=0)+l,p=a;s[p]&&!(p>=c);)++p;if(p-a>16&&s.buffer&&Ks)return Ks.decode(s.buffer instanceof ArrayBuffer?s.subarray(a,p):s.slice(a,p));for(var _="";a>10,56320|1023&j)}}else _+=String.fromCharCode((31&w)<<6|v)}else _+=String.fromCharCode(w)}return _},Jt=(s,a)=>(s>>>=0)?Wr(Q(),s,a):"";function Cs(s,a,l){return f?ht(3,0,1,s,a,l):(l>>>=0,0)}function Hs(s,a){if(f)return ht(4,0,1,s,a);a>>>=0}var Er=s=>{for(var a=0,l=0;l=55296&&c<=57343?(a+=4,++l):a+=3}return a},qe=(s,a,l)=>((c,p,_,w)=>{if(!(w>0))return 0;for(var v=_>>>=0,x=_+w-1,j=0;j=55296&&B<=57343&&(B=65536+((1023&B)<<10)|1023&c.charCodeAt(++j)),B<=127){if(_>=x)break;p[_++>>>0]=B}else if(B<=2047){if(_+1>=x)break;p[_++>>>0]=192|B>>6,p[_++>>>0]=128|63&B}else if(B<=65535){if(_+2>=x)break;p[_++>>>0]=224|B>>12,p[_++>>>0]=128|B>>6&63,p[_++>>>0]=128|63&B}else{if(_+3>=x)break;p[_++>>>0]=240|B>>18,p[_++>>>0]=128|B>>12&63,p[_++>>>0]=128|B>>6&63,p[_++>>>0]=128|63&B}}return p[_>>>0]=0,_-v})(s,Q(),a,l);function ut(s,a){if(f)return ht(5,0,1,s,a);s>>>=0,a>>>=0}function bt(s,a,l){if(f)return ht(6,0,1,s,a,l);a>>>=0,l>>>=0}function cr(s,a,l){return f?ht(7,0,1,s,a,l):(l>>>=0,0)}function ls(s,a){if(f)return ht(8,0,1,s,a);s>>>=0,a>>>=0}function Es(s,a,l){if(f)return ht(9,0,1,s,a,l);a>>>=0}function Xr(s,a,l,c){if(f)return ht(10,0,1,s,a,l,c);a>>>=0,l>>>=0}function Ps(s,a,l,c){if(f)return ht(11,0,1,s,a,l,c);a>>>=0,c>>>=0}function Ss(s,a,l,c){if(f)return ht(12,0,1,s,a,l,c);a>>>=0,l>>>=0,c>>>=0}function ks(s){if(f)return ht(13,0,1,s);s>>>=0}function $s(s,a){if(f)return ht(14,0,1,s,a);s>>>=0,a>>>=0}function gs(s,a,l){if(f)return ht(15,0,1,s,a,l);a>>>=0}var _n,gn,Fn=()=>Ge(""),Dn=s=>{if(s===null)return"null";var a=typeof s;return a==="object"||a==="array"||a==="function"?s.toString():""+s},Hr=s=>{for(var a="",l=s;Q()[l>>>0];)a+=_n[Q()[l++>>>0]];return a},ys={},rn={},cs=s=>{throw new gn(s)};function Jr(s,a,l={}){return function(c,p,_={}){var w=p.name;if(c||cs(`type "${w}" must have a positive integer typeid pointer`),rn.hasOwnProperty(c)){if(_.ignoreDuplicateRegistrations)return;cs(`Cannot register type '${w}' twice`)}if(rn[c]=p,ys.hasOwnProperty(c)){var v=ys[c];delete ys[c],v.forEach(x=>x())}}(s,a,l)}var yn=(s,a,l)=>{switch(a){case 1:return l?c=>N()[c>>>0]:c=>Q()[c>>>0];case 2:return l?c=>ue()[c>>>1>>>0]:c=>Ae()[c>>>1>>>0];case 4:return l?c=>we()[c>>>2>>>0]:c=>ae()[c>>>2>>>0];case 8:return l?c=>oe[c>>>3]:c=>ge[c>>>3];default:throw new TypeError(`invalid integer width (${a}): ${s}`)}};function qs(s,a,l,c,p){s>>>=0,l>>>=0;var _=(a=Hr(a>>>=0)).indexOf("u")!=-1;Jr(s,{name:a,fromWireType:w=>w,toWireType:function(w,v){if(typeof v!="bigint"&&typeof v!="number")throw new TypeError(`Cannot convert "${Dn(v)}" to ${this.name}`);return typeof v=="number"&&(v=BigInt(v)),v},argPackAdvance:Te,readValueFromPointer:yn(a,l,!_),destructorFunction:null})}var Te=8;function L(s,a,l,c){Jr(s>>>=0,{name:a=Hr(a>>>=0),fromWireType:function(p){return!!p},toWireType:function(p,_){return _?l:c},argPackAdvance:Te,readValueFromPointer:function(p){return this.fromWireType(Q()[p>>>0])},destructorFunction:null})}var J=[],ie=[];function ve(s){(s>>>=0)>9&&--ie[s+1]==0&&(ie[s]=void 0,J.push(s))}var be={toValue:s=>(s||cs("Cannot use deleted val. handle = "+s),ie[s]),toHandle:s=>{switch(s){case void 0:return 2;case null:return 4;case!0:return 6;case!1:return 8;default:{let a=J.pop()||ie.length;return ie[a]=s,ie[a+1]=1,a}}}};function ze(s){return this.fromWireType(ae()[s>>>2>>>0])}var tt={name:"emscripten::val",fromWireType:s=>{var a=be.toValue(s);return ve(s),a},toWireType:(s,a)=>be.toHandle(a),argPackAdvance:Te,readValueFromPointer:ze,destructorFunction:null};function ct(s){return Jr(s>>>=0,tt)}var it=(s,a)=>{switch(a){case 4:return function(l){return this.fromWireType(Ze()[l>>>2>>>0])};case 8:return function(l){return this.fromWireType(et()[l>>>3>>>0])};default:throw new TypeError(`invalid float width (${a}): ${s}`)}},dt=function(s,a,l){l>>>=0,Jr(s>>>=0,{name:a=Hr(a>>>=0),fromWireType:c=>c,toWireType:(c,p)=>p,argPackAdvance:Te,readValueFromPointer:it(a,l),destructorFunction:null})};function At(s,a,l,c,p){s>>>=0,l>>>=0,a=Hr(a>>>=0);var _=x=>x;if(c===0){var w=32-8*l;_=x=>x<>>w}var v=a.includes("unsigned");Jr(s,{name:a,fromWireType:_,toWireType:v?function(x,j){return this.name,j>>>0}:function(x,j){return this.name,j},argPackAdvance:Te,readValueFromPointer:yn(a,l,c!==0),destructorFunction:null})}function qt(s,a,l){l>>>=0;var c=[Int8Array,Uint8Array,Int16Array,Uint16Array,Int32Array,Uint32Array,Float32Array,Float64Array,BigInt64Array,BigUint64Array][a];function p(_){var w=ae()[_>>>2>>>0],v=ae()[_+4>>>2>>>0];return new c(N().buffer,v,w)}Jr(s>>>=0,{name:l=Hr(l),fromWireType:p,argPackAdvance:Te,readValueFromPointer:p},{ignoreDuplicateRegistrations:!0})}function Ht(s,a){Jr(s>>>=0,{name:a=Hr(a>>>=0),fromWireType(l){for(var c,p=ae()[l>>>2>>>0],_=l+4,w=_,v=0;v<=p;++v){var x=_+v;if(v==p||Q()[x>>>0]==0){var j=Jt(w,x-w);c===void 0?c=j:(c+="\0",c+=j),w=x+1}}return Zr(l),c},toWireType(l,c){var p;c instanceof ArrayBuffer&&(c=new Uint8Array(c));var _=typeof c=="string";_||c instanceof Uint8Array||c instanceof Uint8ClampedArray||c instanceof Int8Array||cs("Cannot pass non-string to std::string"),p=_?Er(c):c.length;var w=nn(4+p+1),v=w+4;if(ae()[w>>>2>>>0]=p,_)qe(c,v,p+1);else if(_)for(var x=0;x255&&(Zr(w),cs("String has UTF-16 code units that do not fit in 8 bits")),Q()[v+x>>>0]=j}else for(x=0;x>>0]=c[x];return l!==null&&l.push(Zr,w),w},argPackAdvance:Te,readValueFromPointer:ze,destructorFunction(l){Zr(l)}})}var fr=typeof TextDecoder<"u"?new TextDecoder("utf-16le"):void 0,sr=(s,a)=>{for(var l=s,c=l>>1,p=c+a/2;!(c>=p)&&Ae()[c>>>0];)++c;if((l=c<<1)-s>32&&fr)return fr.decode(Q().slice(s,l));for(var _="",w=0;!(w>=a/2);++w){var v=ue()[s+2*w>>>1>>>0];if(v==0)break;_+=String.fromCharCode(v)}return _},Pr=(s,a,l)=>{if(l??(l=2147483647),l<2)return 0;for(var c=a,p=(l-=2)<2*s.length?l/2:s.length,_=0;_>>1>>>0]=w,a+=2}return ue()[a>>>1>>>0]=0,a-c},$r=s=>2*s.length,qr=(s,a)=>{for(var l=0,c="";!(l>=a/4);){var p=we()[s+4*l>>>2>>>0];if(p==0)break;if(++l,p>=65536){var _=p-65536;c+=String.fromCharCode(55296|_>>10,56320|1023&_)}else c+=String.fromCharCode(p)}return c},ws=(s,a,l)=>{if(a>>>=0,l??(l=2147483647),l<4)return 0;for(var c=a,p=c+l-4,_=0;_=55296&&w<=57343&&(w=65536+((1023&w)<<10)|1023&s.charCodeAt(++_)),we()[a>>>2>>>0]=w,(a+=4)+4>p)break}return we()[a>>>2>>>0]=0,a-c},Ur=s=>{for(var a=0,l=0;l=55296&&c<=57343&&++l,a+=4}return a},Qs=function(s,a,l){var c,p,_,w;s>>>=0,a>>>=0,l=Hr(l>>>=0),a===2?(c=sr,p=Pr,w=$r,_=v=>Ae()[v>>>1>>>0]):a===4&&(c=qr,p=ws,w=Ur,_=v=>ae()[v>>>2>>>0]),Jr(s,{name:l,fromWireType:v=>{for(var x,j=ae()[v>>>2>>>0],B=v+4,H=0;H<=j;++H){var ee=v+4+H*a;if(H==j||_(ee)==0){var fe=c(B,ee-B);x===void 0?x=fe:(x+="\0",x+=fe),B=ee+a}}return Zr(v),x},toWireType:(v,x)=>{typeof x!="string"&&cs(`Cannot pass non-string to C++ string type ${l}`);var j=w(x),B=nn(4+j+a);return ae()[B>>>2>>>0]=j/a,p(x,B+4,j+a),v!==null&&v.push(Zr,B),B},argPackAdvance:Te,readValueFromPointer:ze,destructorFunction(v){Zr(v)}})},ur=function(s,a){Jr(s>>>=0,{isVoid:!0,name:a=Hr(a>>>=0),argPackAdvance:0,fromWireType:()=>{},toWireType:(l,c)=>{}})};function Ir(s){Ui(s>>>=0,!h,1,!d,131072,!1),ne.threadInitTLS()}var zr=s=>{if(!Ce)try{s(),(()=>{if(!Vt())try{f?Ki(O):G(O)}catch(a){W(a)}})()}catch(a){W(a)}};function Yr(s){if(s>>>=0,typeof Atomics.waitAsync=="function"){Atomics.waitAsync(we(),s>>>2,s).value.then(ss);var a=s+128;Atomics.store(we(),a>>>2,1)}}var ss=()=>{var s=vn();s&&(Yr(s),zr(Hi))};function Br(s,a){if((s>>>=0)==(a>>>=0))setTimeout(ss);else if(f)postMessage({targetThread:s,cmd:"checkMailbox"});else{var l=ne.pthreads[s];if(!l)return;l.postMessage({cmd:"checkMailbox"})}}var Dr=[];function Ar(s,a,l,c,p){a>>>=0,l>>>=0,p>>>=0,c/=2,Dr.length=c;for(var _=p>>>3,w=0;w>>0];var v=a?ot[a]:vp[s];ne.currentProxiedOperationCallerThread=l;var x=v(...Dr);return ne.currentProxiedOperationCallerThread=0,x}var yr=()=>{xt=0};function Mr(s){s>>>=0,f?postMessage({cmd:"cleanupThread",thread:s}):St(s)}function vr(s){}var Rr=(s,a)=>{var l,c,p,_=rn[s];return _===void 0&&cs(`${a} has unknown type ${l=s,c=$i(l),p=Hr(c),Zr(c),p}`),_},us=(s,a,l)=>{var c=[],p=s.toWireType(c,l);return c.length&&(ae()[a>>>2>>>0]=be.toHandle(c)),p};function bi(s,a,l){return s>>>=0,a>>>=0,l>>>=0,s=be.toValue(s),a=Rr(a,"emval::as"),us(a,l,s)}function xd(s,a){return s>>>=0,a>>>=0,s=be.toValue(s),(a=Rr(a,"emval::as")).toWireType(null,s)}var jn=s=>{try{return s()}catch(a){Ge(a)}},Xs=()=>{xt+=1},ds=()=>{xt-=1},Tt={instrumentWasmImports(s){var a=/^(invoke_.*|__asyncjs__.*)$/;for(let[l,c]of Object.entries(s))typeof c=="function"&&(c.isAsync||a.test(l))},instrumentWasmExports(s){var a={};for(let[l,c]of Object.entries(s))a[l]=typeof c=="function"?(...p)=>{Tt.exportCallStack.push(l);try{return c(...p)}finally{Ce||(Tt.exportCallStack.pop(),Tt.maybeStopUnwind())}}:c;return a},State:{Normal:0,Unwinding:1,Rewinding:2,Disabled:3},state:0,StackSize:65536,currData:null,handleSleepReturnValue:0,exportCallStack:[],callStackNameToId:{},callStackIdToName:{},callStackId:0,asyncPromiseHandlers:null,sleepCallbacks:[],getCallStackId(s){var a=Tt.callStackNameToId[s];return a===void 0&&(a=Tt.callStackId++,Tt.callStackNameToId[s]=a,Tt.callStackIdToName[a]=s),a},maybeStopUnwind(){Tt.currData&&Tt.state===Tt.State.Unwinding&&Tt.exportCallStack.length===0&&(Tt.state=Tt.State.Normal,Xs(),jn(Wu),typeof Fibers<"u"&&Fibers.trampoline())},whenDone:()=>new Promise((s,a)=>{Tt.asyncPromiseHandlers={resolve:s,reject:a}}),allocateData(){var s=nn(12+Tt.StackSize);return Tt.setDataHeader(s,s+12,Tt.StackSize),Tt.setDataRewindFunc(s),s},setDataHeader(s,a,l){ae()[s>>>2>>>0]=a,ae()[s+4>>>2>>>0]=a+l},setDataRewindFunc(s){var a=Tt.exportCallStack[0],l=Tt.getCallStackId(a);we()[s+8>>>2>>>0]=l},getDataRewindFuncName(s){var a=we()[s+8>>>2>>>0];return Tt.callStackIdToName[a]},getDataRewindFunc:s=>Y[s],doRewind(s){var a=Tt.getDataRewindFuncName(s),l=Tt.getDataRewindFunc(a);return ds(),l()},handleSleep(s){if(!Ce){if(Tt.state===Tt.State.Normal){var a=!1,l=!1;s((c=0)=>{if(!Ce&&(Tt.handleSleepReturnValue=c,a=!0,l)){Tt.state=Tt.State.Rewinding,jn(()=>Uu(Tt.currData)),typeof MainLoop<"u"&&MainLoop.func&&MainLoop.resume();var p,_=!1;try{p=Tt.doRewind(Tt.currData)}catch(x){p=x,_=!0}var w=!1;if(!Tt.currData){var v=Tt.asyncPromiseHandlers;v&&(Tt.asyncPromiseHandlers=null,(_?v.reject:v.resolve)(p),w=!0)}if(_&&!w)throw p}}),l=!0,a||(Tt.state=Tt.State.Unwinding,Tt.currData=Tt.allocateData(),typeof MainLoop<"u"&&MainLoop.func&&MainLoop.pause(),jn(()=>Vu(Tt.currData)))}else Tt.state===Tt.State.Rewinding?(Tt.state=Tt.State.Normal,jn(Ut),Zr(Tt.currData),Tt.currData=null,Tt.sleepCallbacks.forEach(zr)):Ge(`invalid state: ${Tt.state}`);return Tt.handleSleepReturnValue}},handleAsync:s=>Tt.handleSleep(a=>{s().then(a)})},Ua=function(s){return s>>>=0,Tt.handleAsync(async()=>{var a=await be.toValue(s);return be.toHandle(a)})};Ua.isAsync=!0;var Ln=[];function Td(s,a,l,c){return a>>>=0,l>>>=0,c>>>=0,(s=Ln[s>>>=0])(null,a=be.toValue(a),l,c)}var Cd={},zn=s=>{var a=Cd[s];return a===void 0?Hr(s):a};function Ga(s,a,l,c,p){return a>>>=0,l>>>=0,c>>>=0,p>>>=0,(s=Ln[s>>>=0])(a=be.toValue(a),a[l=zn(l)],c,p)}function Ed(s,a){return s>>>=0,a>>>=0,(s=be.toValue(s))==be.toValue(a)}var Ka=()=>typeof globalThis=="object"?globalThis:Function("return this")();function Ha(s){return(s>>>=0)==0?be.toHandle(Ka()):(s=zn(s),be.toHandle(Ka()[s]))}var Pd=s=>{var a=Ln.length;return Ln.push(s),a},qa=(s,a)=>{for(var l=new Array(s),c=0;c>>2>>>0],"parameter "+c);return l},Qa=(s,a)=>Object.defineProperty(a,"name",{value:s});function Xa(s,a,l){var c=qa(s,a>>>=0),p=c.shift();s--;var _=`return function (obj, func, destructorsRef, args) { +`,w=0,v=[];l===0&&v.push("obj");for(var x=["retType"],j=[p],B=0;Bfe.name).join(", ")}) => ${p.name}>`;return Pd(Qa(ee,H))}function uf(s){return s=zn(s>>>=0),be.toHandle(n[s])}function Sd(s,a){return s>>>=0,a>>>=0,s=be.toValue(s),a=be.toValue(a),be.toHandle(s[a])}function Ja(s){(s>>>=0)>9&&(ie[s+1]+=1)}function kd(){return be.toHandle([])}function Mi(s){s>>>=0,s=be.toValue(s);for(var a=new Array(s.length),l=0;l>>=0,be.toHandle(zn(s))}function Id(){return be.toHandle({})}var Ad=s=>{for(;s.length;){var a=s.pop();s.pop()(a)}};function Od(s){s>>>=0;var a=be.toValue(s);Ad(a),ve(s)}function Fd(s,a,l){s>>>=0,a>>>=0,l>>>=0,s=be.toValue(s),a=be.toValue(a),l=be.toValue(l),s[a]=l}function Dd(s,a){a>>>=0;var l=(s=Rr(s>>>=0,"_emval_take_value")).readValueFromPointer(a);return be.toHandle(l)}function jd(s,a){s=er(s),a>>>=0;var l=new Date(1e3*s);we()[a>>>2>>>0]=l.getUTCSeconds(),we()[a+4>>>2>>>0]=l.getUTCMinutes(),we()[a+8>>>2>>>0]=l.getUTCHours(),we()[a+12>>>2>>>0]=l.getUTCDate(),we()[a+16>>>2>>>0]=l.getUTCMonth(),we()[a+20>>>2>>>0]=l.getUTCFullYear()-1900,we()[a+24>>>2>>>0]=l.getUTCDay();var c=Date.UTC(l.getUTCFullYear(),0,1,0,0,0,0),p=(l.getTime()-c)/864e5|0;we()[a+28>>>2>>>0]=p}var vi=[0,31,60,91,121,152,182,213,244,274,305,335],Ya=[0,31,59,90,120,151,181,212,243,273,304,334],Za=s=>{var a;return((a=s.getFullYear())%4!=0||a%100==0&&a%400!=0?Ya:vi)[s.getMonth()]+s.getDate()-1};function eo(s,a){s=er(s),a>>>=0;var l=new Date(1e3*s);we()[a>>>2>>>0]=l.getSeconds(),we()[a+4>>>2>>>0]=l.getMinutes(),we()[a+8>>>2>>>0]=l.getHours(),we()[a+12>>>2>>>0]=l.getDate(),we()[a+16>>>2>>>0]=l.getMonth(),we()[a+20>>>2>>>0]=l.getFullYear()-1900,we()[a+24>>>2>>>0]=l.getDay();var c=0|Za(l);we()[a+28>>>2>>>0]=c,we()[a+36>>>2>>>0]=-60*l.getTimezoneOffset();var p=new Date(l.getFullYear(),0,1),_=new Date(l.getFullYear(),6,1).getTimezoneOffset(),w=p.getTimezoneOffset(),v=0|(_!=w&&l.getTimezoneOffset()==Math.min(w,_));we()[a+32>>>2>>>0]=v}var sn=function(s){s>>>=0;var a=(()=>{var l=new Date(we()[s+20>>>2>>>0]+1900,we()[s+16>>>2>>>0],we()[s+12>>>2>>>0],we()[s+8>>>2>>>0],we()[s+4>>>2>>>0],we()[s>>>2>>>0],0),c=we()[s+32>>>2>>>0],p=l.getTimezoneOffset(),_=new Date(l.getFullYear(),0,1),w=new Date(l.getFullYear(),6,1).getTimezoneOffset(),v=_.getTimezoneOffset(),x=Math.min(v,w);if(c<0)we()[s+32>>>2>>>0]=+(w!=v&&x==p);else if(c>0!=(x==p)){var j=Math.max(v,w),B=c>0?x:j;l.setTime(l.getTime()+6e4*(B-p))}we()[s+24>>>2>>>0]=l.getDay();var H=0|Za(l);we()[s+28>>>2>>>0]=H,we()[s>>>2>>>0]=l.getSeconds(),we()[s+4>>>2>>>0]=l.getMinutes(),we()[s+8>>>2>>>0]=l.getHours(),we()[s+12>>>2>>>0]=l.getDate(),we()[s+16>>>2>>>0]=l.getMonth(),we()[s+20>>>2>>>0]=l.getYear();var ee=l.getTime();return isNaN(ee)?-1:ee/1e3})();return BigInt(a)};function to(s,a,l,c,p,_,w){return f?ht(16,0,1,s,a,l,c,p,_,w):(s>>>=0,p=er(p),_>>>=0,w>>>=0,-52)}function Ld(s,a,l,c,p,_){if(f)return ht(17,0,1,s,a,l,c,p,_);s>>>=0,a>>>=0,_=er(_)}var wn={},ro=()=>performance.timeOrigin+performance.now();function so(s,a){if(f)return ht(18,0,1,s,a);if(wn[s]&&(clearTimeout(wn[s].id),delete wn[s]),!a)return 0;var l=setTimeout(()=>{delete wn[s],zr(()=>qo(s,ro()))},a);return wn[s]={id:l,timeout_ms:a},0}var no=function(s,a,l,c){s>>>=0,a>>>=0,l>>>=0,c>>>=0;var p=new Date().getFullYear(),_=new Date(p,0,1),w=new Date(p,6,1),v=_.getTimezoneOffset(),x=w.getTimezoneOffset(),j=Math.max(v,x);ae()[s>>>2>>>0]=60*j,we()[a>>>2>>>0]=+(v!=x);var B=fe=>{var xe=fe>=0?"-":"+",De=Math.abs(fe);return`UTC${xe}${String(Math.floor(De/60)).padStart(2,"0")}${String(De%60).padStart(2,"0")}`},H=B(v),ee=B(x);xDate.now(),zd=s=>s>=0&&s<=3;function Bd(s,a,l){if(l>>>=0,!zd(s))return 28;var c;s===0?c=io():c=ro();var p=Math.round(1e3*c*1e3);return oe[l>>>3]=BigInt(p),0}var xi=[],ao=(s,a,l)=>{var c=((p,_)=>{var w;for(xi.length=0;w=Q()[p++>>>0];){var v=w!=105;_+=(v&=w!=112)&&_%8?4:0,xi.push(w==112?ae()[_>>>2>>>0]:w==106?oe[_>>>3]:w==105?we()[_>>>2>>>0]:et()[_>>>3>>>0]),_+=v?8:4}return xi})(a,l);return ot[s](...c)};function Ti(s,a,l){return ao(s>>>=0,a>>>=0,l>>>=0)}function Rd(s,a,l){return ao(s>>>=0,a>>>=0,l>>>=0)}var Js=s=>{Js.shown||(Js.shown={}),Js.shown[s]||(Js.shown[s]=1,$(s))},Nd=()=>{};function oo(s,a){return $(Jt(s>>>=0,a>>>=0))}var Vd=()=>{throw Xs(),"unwind"},lo=()=>4294901760;function co(){return lo()}var Wd=()=>1,Ud=()=>navigator.hardwareConcurrency;function uo(s){return Ge("Cannot use emscripten_pc_get_function without -sUSE_OFFSET_CONVERTER"),0}var Gd=(s,a)=>Math.ceil(s/a)*a,Kd=s=>{var a=(s-P.buffer.byteLength+65535)/65536|0;try{return P.grow(a),me(),1}catch{}};function po(s){s>>>=0;var a=Q().length;if(s<=a)return!1;var l=lo();if(s>l)return!1;for(var c=1;c<=4;c*=2){var p=a*(1+.2/c);p=Math.min(p,s+100663296);var _=Math.min(l,Gd(Math.max(s,p),65536));if(Kd(_))return!0}return!1}var Bn=s=>(Ge("Cannot use convertFrameToPC (needed by __builtin_return_address) without -sUSE_OFFSET_CONVERTER"),0),bn={},Ci=s=>{s.forEach(a=>{Bn()})},ho=()=>new Error().stack.toString();function Hd(){var s=ho().split(` +`);return s[0]=="Error"&&s.shift(),Ci(s),bn.last_addr=Bn(s[3]),bn.last_stack=s,bn.last_addr}function fo(s,a,l){var c;s>>>=0,a>>>=0,bn.last_addr==s?c=bn.last_stack:((c=ho().split(` +`))[0]=="Error"&&c.shift(),Ci(c));for(var p=3;c[p]&&Bn(c[p])!=s;)++p;for(var _=0;_>>2>>>0]=Bn(c[_+p]);return _}var ps=s=>{var a=Er(s)+1,l=Cr(a);return qe(s,l,a),l},Pe={Internals:{jsObjects:[],jsObjectInsert:(s,a)=>{Pe.Internals.jsObjects[s]=a},bufferOnUnmaps:[],futures:[],futureInsert:(s,a)=>{Pe.Internals.futures[s]=new Promise(l=>a.finally(()=>l(s)))}},getJsObject:s=>{if(s)return Pe.Internals.jsObjects[s]},importJsAdapter:(s,a=0)=>{var l=Vo(a);return Pe.Internals.jsObjects[l]=s,l},importJsBindGroup:(s,a=0)=>{var l=Oi(a);return Pe.Internals.jsObjects[l]=s,l},importJsBindGroupLayout:(s,a=0)=>{var l=Rn(a);return Pe.Internals.jsObjects[l]=s,l},importJsBuffer:(s,a=0)=>{s.mapState!="pending"||Ge(void 0);var l=s.mapState=="mapped"?3:1,c=Wo(a,l);return Pe.Internals.jsObjectInsert(c,s),s.mapState=="mapped"&&(Pe.Internals.bufferOnUnmaps[c]=[]),c},importJsCommandBuffer:(s,a=0)=>{var l=Fi(a);return Pe.Internals.jsObjects[l]=s,l},importJsCommandEncoder:(s,a=0)=>{var l=Di(a);return Pe.Internals.jsObjects[l]=s,l},importJsComputePassEncoder:(s,a=0)=>{var l=Nn(a);return Pe.Internals.jsObjects[l]=s,l},importJsComputePipeline:(s,a=0)=>{var l=ji(a);return Pe.Internals.jsObjects[l]=s,l},importJsDevice:(s,a=0)=>{var l=Bi(a),c=Uo(a,l);return Pe.Internals.jsObjectInsert(l,s.queue),Pe.Internals.jsObjectInsert(c,s),c},importJsPipelineLayout:(s,a=0)=>{var l=Do(a);return Pe.Internals.jsObjects[l]=s,l},importJsQuerySet:(s,a=0)=>{var l=Vn(a);return Pe.Internals.jsObjects[l]=s,l},importJsQueue:(s,a=0)=>{var l=Bi(a);return Pe.Internals.jsObjects[l]=s,l},importJsRenderBundle:(s,a=0)=>{var l=jo(a);return Pe.Internals.jsObjects[l]=s,l},importJsRenderBundleEncoder:(s,a=0)=>{var l=Lo(a);return Pe.Internals.jsObjects[l]=s,l},importJsRenderPassEncoder:(s,a=0)=>{var l=Li(a);return Pe.Internals.jsObjects[l]=s,l},importJsRenderPipeline:(s,a=0)=>{var l=zo(a);return Pe.Internals.jsObjects[l]=s,l},importJsSampler:(s,a=0)=>{var l=Bo(a);return Pe.Internals.jsObjects[l]=s,l},importJsShaderModule:(s,a=0)=>{var l=Ri(a);return Pe.Internals.jsObjects[l]=s,l},importJsSurface:(s,a=0)=>{var l=Ro(a);return Pe.Internals.jsObjects[l]=s,l},importJsTexture:(s,a=0)=>{var l=No(a);return Pe.Internals.jsObjects[l]=s,l},importJsTextureView:(s,a=0)=>{var l=zi(a);return Pe.Internals.jsObjects[l]=s,l},errorCallback:(s,a,l,c)=>{var p=he(),_=ps(l);ea(s,a,_,c),ce(p)},setStringView:(s,a,l)=>{ae()[s>>>2>>>0]=a,ae()[s+4>>>2>>>0]=l},makeStringFromStringView:s=>{var a=ae()[s>>>2>>>0],l=ae()[s+4>>>2>>>0];return Jt(a,l)},makeStringFromOptionalStringView:s=>{var a=ae()[s>>>2>>>0],l=ae()[s+4>>>2>>>0];return a?Jt(a,l):l===0?"":void 0},makeColor:s=>({r:et()[s>>>3>>>0],g:et()[s+8>>>3>>>0],b:et()[s+16>>>3>>>0],a:et()[s+24>>>3>>>0]}),makeExtent3D:s=>({width:ae()[s>>>2>>>0],height:ae()[s+4>>>2>>>0],depthOrArrayLayers:ae()[s+8>>>2>>>0]}),makeOrigin3D:s=>({x:ae()[s>>>2>>>0],y:ae()[s+4>>>2>>>0],z:ae()[s+8>>>2>>>0]}),makeTexelCopyTextureInfo:s=>({texture:Pe.getJsObject(ae()[s>>>2>>>0]),mipLevel:ae()[s+4>>>2>>>0],origin:Pe.makeOrigin3D(s+8),aspect:Pe.TextureAspect[ae()[s+20>>>2>>>0]]}),makeTexelCopyBufferLayout:s=>{var a=ae()[s+8>>>2>>>0],l=ae()[s+12>>>2>>>0];return{offset:4294967296*ae()[s+4>>>2>>>0]+ae()[s>>>2>>>0],bytesPerRow:a===4294967295?void 0:a,rowsPerImage:l===4294967295?void 0:l}},makeTexelCopyBufferInfo:s=>{var a=s+0,l=Pe.makeTexelCopyBufferLayout(a);return l.buffer=Pe.getJsObject(ae()[s+16>>>2>>>0]),l},makePassTimestampWrites:s=>{if(s!==0)return{querySet:Pe.getJsObject(ae()[s+4>>>2>>>0]),beginningOfPassWriteIndex:ae()[s+8>>>2>>>0],endOfPassWriteIndex:ae()[s+12>>>2>>>0]}},makePipelineConstants:(s,a)=>{if(s){for(var l={},c=0;c>>3>>>0]}return l}},makePipelineLayout:s=>s?Pe.getJsObject(s):"auto",makeComputeState:s=>{if(s)return{module:Pe.getJsObject(ae()[s+4>>>2>>>0]),constants:Pe.makePipelineConstants(ae()[s+16>>>2>>>0],ae()[s+20>>>2>>>0]),entryPoint:Pe.makeStringFromOptionalStringView(s+8)}},makeComputePipelineDesc:s=>({label:Pe.makeStringFromOptionalStringView(s+4),layout:Pe.makePipelineLayout(ae()[s+12>>>2>>>0]),compute:Pe.makeComputeState(s+16)}),makeRenderPipelineDesc:s=>{function a(x){if(x)return{operation:Pe.BlendOperation[ae()[x>>>2>>>0]],srcFactor:Pe.BlendFactor[ae()[x+4>>>2>>>0]],dstFactor:Pe.BlendFactor[ae()[x+8>>>2>>>0]]}}function l(x){if(x)return{alpha:a(x+12),color:a(x+0)}}function c(x,j){for(var B,H,ee=[],fe=0;fe>>2>>>0])===0?void 0:{format:Pe.TextureFormat[H],blend:l(ae()[B+8>>>2>>>0]),writeMask:ae()[B+16>>>2>>>0]}));return ee}function p(x){return{compare:Pe.CompareFunction[ae()[x>>>2>>>0]],failOp:Pe.StencilOperation[ae()[x+4>>>2>>>0]],depthFailOp:Pe.StencilOperation[ae()[x+8>>>2>>>0]],passOp:Pe.StencilOperation[ae()[x+12>>>2>>>0]]}}function _(x,j){for(var B,H=[],ee=0;ee>>2>>>0]],offset:4294967296*ae()[B+4+8>>>2>>>0]+ae()[B+8>>>2>>>0],shaderLocation:ae()[B+16>>>2>>>0]}));return H}function w(x){if(x){var j=ae()[x+4>>>2>>>0],B=ae()[x+16>>>2>>>0];return j===0&&B===0?null:{arrayStride:4294967296*ae()[x+4+8>>>2>>>0]+ae()[x+8>>>2>>>0],stepMode:Pe.VertexStepMode[j],attributes:_(B,ae()[x+20>>>2>>>0])}}}function v(x,j){if(x){for(var B=[],H=0;H>>2>>>0]),vertex:function(x){if(x)return{module:Pe.getJsObject(ae()[x+4>>>2>>>0]),constants:Pe.makePipelineConstants(ae()[x+16>>>2>>>0],ae()[x+20>>>2>>>0]),buffers:v(ae()[x+24>>>2>>>0],ae()[x+28>>>2>>>0]),entryPoint:Pe.makeStringFromOptionalStringView(x+8)}}(s+16),primitive:function(x){if(x)return{topology:Pe.PrimitiveTopology[ae()[x+4>>>2>>>0]],stripIndexFormat:Pe.IndexFormat[ae()[x+8>>>2>>>0]],frontFace:Pe.FrontFace[ae()[x+12>>>2>>>0]],cullMode:Pe.CullMode[ae()[x+16>>>2>>>0]],unclippedDepth:!!ae()[x+20>>>2>>>0]}}(s+48),depthStencil:function(x){if(x)return{format:Pe.TextureFormat[ae()[x+4>>>2>>>0]],depthWriteEnabled:!!ae()[x+8>>>2>>>0],depthCompare:Pe.CompareFunction[ae()[x+12>>>2>>>0]],stencilFront:p(x+16),stencilBack:p(x+32),stencilReadMask:ae()[x+48>>>2>>>0],stencilWriteMask:ae()[x+52>>>2>>>0],depthBias:we()[x+56>>>2>>>0],depthBiasSlopeScale:Ze()[x+60>>>2>>>0],depthBiasClamp:Ze()[x+64>>>2>>>0]}}(ae()[s+72>>>2>>>0]),multisample:function(x){if(x)return{count:ae()[x+4>>>2>>>0],mask:ae()[x+8>>>2>>>0],alphaToCoverageEnabled:!!ae()[x+12>>>2>>>0]}}(s+76),fragment:function(x){if(x)return{module:Pe.getJsObject(ae()[x+4>>>2>>>0]),constants:Pe.makePipelineConstants(ae()[x+16>>>2>>>0],ae()[x+20>>>2>>>0]),targets:c(ae()[x+24>>>2>>>0],ae()[x+28>>>2>>>0]),entryPoint:Pe.makeStringFromOptionalStringView(x+8)}}(ae()[s+92>>>2>>>0])}},fillLimitStruct:(s,a)=>{function l(p,_){var w=s[p];we()[a+_>>>2>>>0]=w}function c(p,_){var w=s[p];oe[a+_>>>3]=BigInt(w)}l("maxTextureDimension1D",4),l("maxTextureDimension2D",8),l("maxTextureDimension3D",12),l("maxTextureArrayLayers",16),l("maxBindGroups",20),l("maxBindGroupsPlusVertexBuffers",24),l("maxBindingsPerBindGroup",28),l("maxDynamicUniformBuffersPerPipelineLayout",32),l("maxDynamicStorageBuffersPerPipelineLayout",36),l("maxSampledTexturesPerShaderStage",40),l("maxSamplersPerShaderStage",44),l("maxStorageBuffersPerShaderStage",48),l("maxStorageTexturesPerShaderStage",52),l("maxUniformBuffersPerShaderStage",56),l("minUniformBufferOffsetAlignment",80),l("minStorageBufferOffsetAlignment",84),c("maxUniformBufferBindingSize",64),c("maxStorageBufferBindingSize",72),l("maxVertexBuffers",88),c("maxBufferSize",96),l("maxVertexAttributes",104),l("maxVertexBufferArrayStride",108),l("maxInterStageShaderVariables",112),l("maxColorAttachments",116),l("maxColorAttachmentBytesPerSample",120),l("maxComputeWorkgroupStorageSize",124),l("maxComputeInvocationsPerWorkgroup",128),l("maxComputeWorkgroupSizeX",132),l("maxComputeWorkgroupSizeY",136),l("maxComputeWorkgroupSizeZ",140),l("maxComputeWorkgroupsPerDimension",144)},Int_BufferMapState:{unmapped:1,pending:2,mapped:3},Int_CompilationMessageType:{error:1,warning:2,info:3},Int_DeviceLostReason:{undefined:1,unknown:1,destroyed:2},Int_PreferredFormat:{rgba8unorm:18,bgra8unorm:23},WGSLLanguageFeatureName:{1:"readonly_and_readwrite_storage_textures",2:"packed_4x8_integer_dot_product",3:"unrestricted_pointer_parameters",4:"pointer_composite_access",5:"sized_binding_array"},AddressMode:[,"clamp-to-edge","repeat","mirror-repeat"],BlendFactor:[,"zero","one","src","one-minus-src","src-alpha","one-minus-src-alpha","dst","one-minus-dst","dst-alpha","one-minus-dst-alpha","src-alpha-saturated","constant","one-minus-constant","src1","one-minus-src1","src1alpha","one-minus-src1alpha"],BlendOperation:[,"add","subtract","reverse-subtract","min","max"],BufferBindingType:["binding-not-used",,"uniform","storage","read-only-storage"],BufferMapState:{1:"unmapped",2:"pending",3:"mapped"},CompareFunction:[,"never","less","equal","less-equal","greater","not-equal","greater-equal","always"],CompilationInfoRequestStatus:{1:"success",2:"callback-cancelled"},CompositeAlphaMode:[,"opaque","premultiplied","unpremultiplied","inherit"],CullMode:[,"none","front","back"],ErrorFilter:{1:"validation",2:"out-of-memory",3:"internal"},FeatureLevel:[,"compatibility","core"],FeatureName:{1:"depth-clip-control",2:"depth32float-stencil8",3:"timestamp-query",4:"texture-compression-bc",5:"texture-compression-bc-sliced-3d",6:"texture-compression-etc2",7:"texture-compression-astc",8:"texture-compression-astc-sliced-3d",9:"indirect-first-instance",10:"shader-f16",11:"rg11b10ufloat-renderable",12:"bgra8unorm-storage",13:"float32-filterable",14:"float32-blendable",15:"clip-distances",16:"dual-source-blending",17:"subgroups",18:"core-features-and-limits",327688:"subgroups-f16",327693:"chromium-experimental-unorm16-texture-formats",327694:"chromium-experimental-snorm16-texture-formats",327733:"chromium-experimental-multi-draw-indirect"},FilterMode:[,"nearest","linear"],FrontFace:[,"ccw","cw"],IndexFormat:[,"uint16","uint32"],LoadOp:[,"load","clear"],MipmapFilterMode:[,"nearest","linear"],OptionalBool:["false","true"],PowerPreference:[,"low-power","high-performance"],PredefinedColorSpace:{1:"srgb",2:"display-p3"},PrimitiveTopology:[,"point-list","line-list","line-strip","triangle-list","triangle-strip"],QueryType:{1:"occlusion",2:"timestamp"},SamplerBindingType:["binding-not-used",,"filtering","non-filtering","comparison"],Status:{1:"success",2:"error"},StencilOperation:[,"keep","zero","replace","invert","increment-clamp","decrement-clamp","increment-wrap","decrement-wrap"],StorageTextureAccess:["binding-not-used",,"write-only","read-only","read-write"],StoreOp:[,"store","discard"],SurfaceGetCurrentTextureStatus:{1:"success-optimal",2:"success-suboptimal",3:"timeout",4:"outdated",5:"lost",6:"error"},TextureAspect:[,"all","stencil-only","depth-only"],TextureDimension:[,"1d","2d","3d"],TextureFormat:[,"r8unorm","r8snorm","r8uint","r8sint","r16uint","r16sint","r16float","rg8unorm","rg8snorm","rg8uint","rg8sint","r32float","r32uint","r32sint","rg16uint","rg16sint","rg16float","rgba8unorm","rgba8unorm-srgb","rgba8snorm","rgba8uint","rgba8sint","bgra8unorm","bgra8unorm-srgb","rgb10a2uint","rgb10a2unorm","rg11b10ufloat","rgb9e5ufloat","rg32float","rg32uint","rg32sint","rgba16uint","rgba16sint","rgba16float","rgba32float","rgba32uint","rgba32sint","stencil8","depth16unorm","depth24plus","depth24plus-stencil8","depth32float","depth32float-stencil8","bc1-rgba-unorm","bc1-rgba-unorm-srgb","bc2-rgba-unorm","bc2-rgba-unorm-srgb","bc3-rgba-unorm","bc3-rgba-unorm-srgb","bc4-r-unorm","bc4-r-snorm","bc5-rg-unorm","bc5-rg-snorm","bc6h-rgb-ufloat","bc6h-rgb-float","bc7-rgba-unorm","bc7-rgba-unorm-srgb","etc2-rgb8unorm","etc2-rgb8unorm-srgb","etc2-rgb8a1unorm","etc2-rgb8a1unorm-srgb","etc2-rgba8unorm","etc2-rgba8unorm-srgb","eac-r11unorm","eac-r11snorm","eac-rg11unorm","eac-rg11snorm","astc-4x4-unorm","astc-4x4-unorm-srgb","astc-5x4-unorm","astc-5x4-unorm-srgb","astc-5x5-unorm","astc-5x5-unorm-srgb","astc-6x5-unorm","astc-6x5-unorm-srgb","astc-6x6-unorm","astc-6x6-unorm-srgb","astc-8x5-unorm","astc-8x5-unorm-srgb","astc-8x6-unorm","astc-8x6-unorm-srgb","astc-8x8-unorm","astc-8x8-unorm-srgb","astc-10x5-unorm","astc-10x5-unorm-srgb","astc-10x6-unorm","astc-10x6-unorm-srgb","astc-10x8-unorm","astc-10x8-unorm-srgb","astc-10x10-unorm","astc-10x10-unorm-srgb","astc-12x10-unorm","astc-12x10-unorm-srgb","astc-12x12-unorm","astc-12x12-unorm-srgb"],TextureSampleType:["binding-not-used",,"float","unfilterable-float","depth","sint","uint"],TextureViewDimension:[,"1d","2d","2d-array","cube","cube-array","3d"],ToneMappingMode:{1:"standard",2:"extended"},VertexFormat:{1:"uint8",2:"uint8x2",3:"uint8x4",4:"sint8",5:"sint8x2",6:"sint8x4",7:"unorm8",8:"unorm8x2",9:"unorm8x4",10:"snorm8",11:"snorm8x2",12:"snorm8x4",13:"uint16",14:"uint16x2",15:"uint16x4",16:"sint16",17:"sint16x2",18:"sint16x4",19:"unorm16",20:"unorm16x2",21:"unorm16x4",22:"snorm16",23:"snorm16x2",24:"snorm16x4",25:"float16",26:"float16x2",27:"float16x4",28:"float32",29:"float32x2",30:"float32x3",31:"float32x4",32:"uint32",33:"uint32x2",34:"uint32x3",35:"uint32x4",36:"sint32",37:"sint32x2",38:"sint32x3",39:"sint32x4",40:"unorm10-10-10-2",41:"unorm8x4-bgra"},VertexStepMode:[,"vertex","instance"],FeatureNameString2Enum:{"depth-clip-control":"1","depth32float-stencil8":"2","timestamp-query":"3","texture-compression-bc":"4","texture-compression-bc-sliced-3d":"5","texture-compression-etc2":"6","texture-compression-astc":"7","texture-compression-astc-sliced-3d":"8","indirect-first-instance":"9","shader-f16":"10","rg11b10ufloat-renderable":"11","bgra8unorm-storage":"12","float32-filterable":"13","float32-blendable":"14","clip-distances":"15","dual-source-blending":"16",subgroups:"17","core-features-and-limits":"18","subgroups-f16":"327688","chromium-experimental-unorm16-texture-formats":"327693","chromium-experimental-snorm16-texture-formats":"327694","chromium-experimental-multi-draw-indirect":"327733"},WGSLLanguageFeatureNameString2Enum:{readonly_and_readwrite_storage_textures:"1",packed_4x8_integer_dot_product:"2",unrestricted_pointer_parameters:"3",pointer_composite_access:"4",sized_binding_array:"5"}};function mo(s,a,l,c,p,_){s>>>=0,a=er(a),l=er(l),c>>>=0,p>>>=0,_>>>=0;var w=Pe.getJsObject(s),v={};if(_){var x=ae()[_+12>>>2>>>0];if(x){var j=ae()[_+16>>>2>>>0];v.requiredFeatures=Array.from(ae().subarray(j>>>2>>>0,j+4*x>>>2>>>0),xe=>Pe.FeatureName[xe])}var B=ae()[_+20>>>2>>>0];if(B){let xe=function(We,nt){var pt=B+nt,Ot=ae()[pt>>>2>>>0];Ot!=4294967295&&(H[We]=Ot)},De=function(We,nt){var pt=B+nt,Ot=ae()[pt>>>2>>>0],Gt=ae()[pt+4>>>2>>>0];Ot==4294967295&&Gt==4294967295||(H[We]=4294967296*ae()[pt+4>>>2>>>0]+ae()[pt>>>2>>>0])};var H={};xe("maxTextureDimension1D",4),xe("maxTextureDimension2D",8),xe("maxTextureDimension3D",12),xe("maxTextureArrayLayers",16),xe("maxBindGroups",20),xe("maxBindGroupsPlusVertexBuffers",24),xe("maxDynamicUniformBuffersPerPipelineLayout",32),xe("maxDynamicStorageBuffersPerPipelineLayout",36),xe("maxSampledTexturesPerShaderStage",40),xe("maxSamplersPerShaderStage",44),xe("maxStorageBuffersPerShaderStage",48),xe("maxStorageTexturesPerShaderStage",52),xe("maxUniformBuffersPerShaderStage",56),xe("minUniformBufferOffsetAlignment",80),xe("minStorageBufferOffsetAlignment",84),De("maxUniformBufferBindingSize",64),De("maxStorageBufferBindingSize",72),xe("maxVertexBuffers",88),De("maxBufferSize",96),xe("maxVertexAttributes",104),xe("maxVertexBufferArrayStride",108),xe("maxInterStageShaderVariables",112),xe("maxColorAttachments",116),xe("maxColorAttachmentBytesPerSample",120),xe("maxComputeWorkgroupStorageSize",124),xe("maxComputeInvocationsPerWorkgroup",128),xe("maxComputeWorkgroupSizeX",132),xe("maxComputeWorkgroupSizeY",136),xe("maxComputeWorkgroupSizeZ",140),xe("maxComputeWorkgroupsPerDimension",144),v.requiredLimits=H}var ee=ae()[_+24>>>2>>>0];if(ee){var fe={label:Pe.makeStringFromOptionalStringView(ee+4)};v.defaultQueue=fe}v.label=Pe.makeStringFromOptionalStringView(_+4)}Xs(),Pe.Internals.futureInsert(a,w.requestDevice(v).then(xe=>{ds(),Pe.Internals.jsObjectInsert(p,xe.queue),Pe.Internals.jsObjectInsert(c,xe),l&&(Xs(),Pe.Internals.futureInsert(l,xe.lost.then(De=>{ds(),xe.onuncapturederror=pt=>{};var We=he(),nt=ps(De.message);Ni(l,Pe.Int_DeviceLostReason[De.reason],nt),ce(We)}))),xe.onuncapturederror=De=>{var We=5;De.error instanceof GPUValidationError?We=2:De.error instanceof GPUOutOfMemoryError?We=3:De.error instanceof GPUInternalError&&(We=4);var nt=he(),pt=ps(De.error.message);Go(c,We,pt),ce(nt)},Wn(a,1,c,0)},xe=>{ds();var De=he(),We=ps(xe.message);Wn(a,3,c,We),l&&Ni(l,4,We),ce(De)}))}function qd(s){s>>>=0;var a=Pe.getJsObject(s),l=Pe.Internals.bufferOnUnmaps[s];if(l){for(var c=0;c>>=0,a>>>=0,l>>>=0;var c,p=Pe.getJsObject(s);l===0&&Js("getMappedRange size=0 no longer means WGPU_WHOLE_MAP_SIZE"),l==4294967295&&(l=void 0);try{c=p.getMappedRange(a,l)}catch{return 0}var _=qi(16,c.byteLength);return Q().set(new Uint8Array(c),_>>>0),Pe.Internals.bufferOnUnmaps[s].push(()=>Zr(_)),_}function _o(s,a,l){s>>>=0,a>>>=0,l>>>=0;var c,p=Pe.getJsObject(s);l===0&&Js("getMappedRange size=0 no longer means WGPU_WHOLE_MAP_SIZE"),l==4294967295&&(l=void 0);try{c=p.getMappedRange(a,l)}catch{return 0}var _=qi(16,c.byteLength);return Q().fill(0,_,c.byteLength),Pe.Internals.bufferOnUnmaps[s].push(()=>{new Uint8Array(c).set(Q().subarray(_>>>0,_+c.byteLength>>>0)),Zr(_)}),_}var Xd=function(s,a,l,c,p){s>>>=0,a=er(a),l=er(l),c>>>=0,p>>>=0;var _=Pe.getJsObject(s);Pe.Internals.bufferOnUnmaps[s]=[],p==4294967295&&(p=void 0),Xs(),Pe.Internals.futureInsert(a,_.mapAsync(l,c,p).then(()=>{ds(),Vi(a,1,0)},w=>{ds(),he();var v=ps(w.message),x=w.name==="AbortError"?4:w.name==="OperationError"?3:0;Vi(a,x,v),delete Pe.Internals.bufferOnUnmaps[s]}))};function Jd(s){s>>>=0;var a=Pe.getJsObject(s),l=Pe.Internals.bufferOnUnmaps[s];if(l){for(var c=0;c>>=0,delete Pe.Internals.jsObjects[s]}function Yd(s,a,l){s>>>=0,a>>>=0,l>>>=0;var c,p=!!ae()[a+32>>>2>>>0],_={label:Pe.makeStringFromOptionalStringView(a+4),usage:ae()[a+16>>>2>>>0],size:4294967296*ae()[a+4+24>>>2>>>0]+ae()[a+24>>>2>>>0],mappedAtCreation:p},w=Pe.getJsObject(s);try{c=w.createBuffer(_)}catch{return!1}return Pe.Internals.jsObjectInsert(l,c),p&&(Pe.Internals.bufferOnUnmaps[l]=[]),!0}function Zd(s,a,l){s>>>=0,a>>>=0,l>>>=0;var c=ae()[a>>>2>>>0],p=ae()[c+4>>>2>>>0],_={label:Pe.makeStringFromOptionalStringView(a+4),code:""};p===2&&(_.code=Pe.makeStringFromStringView(c+8));var w=Pe.getJsObject(s);Pe.Internals.jsObjectInsert(l,w.createShaderModule(_))}var yo=s=>{Pe.getJsObject(s).destroy()},ep=function(s,a){s>>>=0,a=er(a);var l=Pe.getJsObject(s);Xs(),Pe.Internals.futureInsert(a,l.popErrorScope().then(c=>{ds();var p=5;c?c instanceof GPUValidationError?p=2:c instanceof GPUOutOfMemoryError?p=3:c instanceof GPUInternalError&&(p=4):p=1;var _=he(),w=c?ps(c.message):0;Wi(a,1,p,w),ce(_)},c=>{ds();var p=he(),_=ps(c.message);Wi(a,1,5,_),ce(p)}))};function tp(s,a,l,c){var p;if(a=er(a),c>>>=0,l>>>=0){var _=ae()[l+4>>>2>>>0];p={featureLevel:Pe.FeatureLevel[_],powerPreference:Pe.PowerPreference[ae()[l+8>>>2>>>0]],forceFallbackAdapter:!!ae()[l+12>>>2>>>0]};var w=ae()[l>>>2>>>0];if(w!==0){ae()[w+4>>>2>>>0];var v=w;p.xrCompatible=!!ae()[v+8>>>2>>>0]}}if(!("gpu"in navigator)){var x=he(),j=ps("WebGPU not available on this browser (navigator.gpu is not available)");return xn(a,3,c,j),void ce(x)}Xs(),Pe.Internals.futureInsert(a,navigator.gpu.requestAdapter(p).then(B=>{if(ds(),B)Pe.Internals.jsObjectInsert(c,B),xn(a,1,c,0);else{var H=he(),ee=ps("WebGPU not available on this browser (requestAdapter returned null)");xn(a,3,c,ee),ce(H)}},B=>{ds();var H=he(),ee=ps(B.message);xn(a,4,c,ee),ce(H)}))}var Ei=function(s,a,l){return s>>>=0,a>>>=0,l>>>=0,Tt.handleAsync(async()=>{var c=[];if(l){var p=(4294967296*ae()[l+4>>>2>>>0]+ae()[l>>>2>>>0])/1e6;c.length=a+1,c[a]=new Promise(x=>setTimeout(x,p,0))}else c.length=a;for(var _=0;_>>2>>>0]+ae()[s+8*_>>>2>>>0];if(!(w in Pe.Internals.futures))return w;c[_]=Pe.Internals.futures[w]}let v=await Promise.race(c);return delete Pe.Internals.futures[v],v})};Ei.isAsync=!0;var Pi={},Mn=()=>{if(!Mn.strings){var s={USER:"web_user",LOGNAME:"web_user",PATH:"/",PWD:"/",HOME:"/home/web_user",LANG:(typeof navigator=="object"&&navigator.languages&&navigator.languages[0]||"C").replace("-","_")+".UTF-8",_:"./this.program"};for(var a in Pi)Pi[a]===void 0?delete s[a]:s[a]=Pi[a];var l=[];for(var a in s)l.push(`${a}=${s[a]}`);Mn.strings=l}return Mn.strings},Si=function(s,a){if(f)return ht(19,0,1,s,a);s>>>=0,a>>>=0;var l=0;return Mn().forEach((c,p)=>{var _=a+l;ae()[s+4*p>>>2>>>0]=_,((w,v)=>{for(var x=0;x>>0]=w.charCodeAt(x);N()[v>>>0]=0})(c,_),l+=c.length+1}),0},wo=function(s,a){if(f)return ht(20,0,1,s,a);s>>>=0,a>>>=0;var l=Mn();ae()[s>>>2>>>0]=l.length;var c=0;return l.forEach(p=>c+=p.length+1),ae()[a>>>2>>>0]=c,0};function bo(s){return f?ht(21,0,1,s):52}function ki(s,a,l,c){return f?ht(22,0,1,s,a,l,c):(a>>>=0,l>>>=0,c>>>=0,52)}function Mo(s,a,l,c){return f?ht(23,0,1,s,a,l,c):(a=er(a),c>>>=0,70)}var rp=[null,[],[]],vo=(s,a)=>{var l=rp[s];a===0||a===10?((s===1?A:$)(Wr(l)),l.length=0):l.push(a)};function xo(s,a,l,c){if(f)return ht(24,0,1,s,a,l,c);a>>>=0,l>>>=0,c>>>=0;for(var p=0,_=0;_>>2>>>0],v=ae()[a+4>>>2>>>0];a+=8;for(var x=0;x>>0]);p+=v}return ae()[c>>>2>>>0]=p,0}function sp(s){return s>>>0}function To(s,a){s>>>=0,a>>>=0;var l=Pe.getJsObject(s);return Pe.fillLimitStruct(l.limits,a),1}function np(s,a){return s>>>=0,Pe.getJsObject(s).features.has(Pe.FeatureName[a])}var ip=function(s){s>>>=0;var a=Pe.getJsObject(s).size;return BigInt(a)},Co=function(s){s>>>=0;var a=Pe.getJsObject(s).usage;return BigInt(a)};function ap(s,a){var l;s>>>=0,(a>>>=0)&&(l={label:Pe.makeStringFromOptionalStringView(a+4),timestampWrites:Pe.makePassTimestampWrites(ae()[a+12>>>2>>>0])});var c=Pe.getJsObject(s),p=Nn(0);return Pe.Internals.jsObjectInsert(p,c.beginComputePass(l)),p}function op(s,a,l,c,p,_){s>>>=0,a>>>=0,l=er(l),c>>>=0,p=er(p),_=er(_);var w=Pe.getJsObject(s),v=Pe.getJsObject(a),x=Pe.getJsObject(c);w.copyBufferToBuffer(v,l,x,p,_)}function Eo(s,a){s>>>=0;var l=Pe.getJsObject(s),c=Fi(0);return Pe.Internals.jsObjectInsert(c,l.finish()),c}function lp(s,a,l,c,p,_){s>>>=0,a>>>=0,p>>>=0,_=er(_);var w=Pe.getJsObject(s),v=Pe.getJsObject(a),x=Pe.getJsObject(p);w.resolveQuerySet(v,l,c,x,_)}function cp(s,a,l,c){s>>>=0,Pe.getJsObject(s).dispatchWorkgroups(a,l,c)}function Po(s){s>>>=0,Pe.getJsObject(s).end()}function up(s,a,l,c,p){s>>>=0,l>>>=0,c>>>=0,p>>>=0;var _=Pe.getJsObject(s),w=Pe.getJsObject(l);if(c==0)_.setBindGroup(a,w);else{for(var v=[],x=0;x>>2>>>0]);_.setBindGroup(a,w,v)}}function dp(s,a){s>>>=0,a>>>=0;var l=Pe.getJsObject(s),c=Pe.getJsObject(a);l.setPipeline(c)}function So(s,a,l){s>>>=0,a>>>=0;var c=Pe.getJsObject(s),p=Pe.getJsObject(a);c.writeTimestamp(p,l)}function pp(s,a){s>>>=0;var l=Pe.getJsObject(s),c=Rn(0);return Pe.Internals.jsObjectInsert(c,l.getBindGroupLayout(a)),c}var hp=s=>ae()[s>>>2>>>0]+4294967296*we()[s+4>>>2>>>0];function ko(s,a){function l(w){var v=ae()[w+8>>>2>>>0],x=ae()[w+32>>>2>>>0],j=ae()[w+36>>>2>>>0],B=ae()[w+4>>>2>>>0];if(v){var H=hp(w+24);return H==-1&&(H=void 0),{binding:B,resource:{buffer:Pe.getJsObject(v),offset:4294967296*ae()[w+4+16>>>2>>>0]+ae()[w+16>>>2>>>0],size:H}}}return x?{binding:B,resource:Pe.getJsObject(x)}:{binding:B,resource:Pe.getJsObject(j)}}s>>>=0,a>>>=0;var c={label:Pe.makeStringFromOptionalStringView(a+4),layout:Pe.getJsObject(ae()[a+12>>>2>>>0]),entries:function(w,v){for(var x=[],j=0;j>>2>>>0],ae()[a+20>>>2>>>0])},p=Pe.getJsObject(s),_=Oi(0);return Pe.Internals.jsObjectInsert(_,p.createBindGroup(c)),_}function fp(s,a){var l;s>>>=0,(a>>>=0)&&(l={label:Pe.makeStringFromOptionalStringView(a+4)});var c=Pe.getJsObject(s),p=Di(0);return Pe.Internals.jsObjectInsert(p,c.createCommandEncoder(l)),p}function mp(s,a){s>>>=0,a>>>=0;var l=Pe.makeComputePipelineDesc(a),c=Pe.getJsObject(s),p=ji(0);return Pe.Internals.jsObjectInsert(p,c.createComputePipeline(l)),p}function $o(s,a){s>>>=0,a>>>=0;var l={type:Pe.QueryType[ae()[a+12>>>2>>>0]],count:ae()[a+16>>>2>>>0]},c=Pe.getJsObject(s),p=Vn(0);return Pe.Internals.jsObjectInsert(p,c.createQuerySet(l)),p}var _p=s=>{var a=Er(s)+1,l=nn(a);return l&&qe(s,l,a),l};function gp(s,a){s>>>=0,a>>>=0;var l=Pe.getJsObject(s),c=l.adapterInfo.vendor+l.adapterInfo.architecture+l.adapterInfo.device+l.adapterInfo.description,p=_p(c),_=Er(l.adapterInfo.vendor);Pe.setStringView(a+4,p,_),p+=_;var w=Er(l.adapterInfo.architecture);Pe.setStringView(a+12,p,w),p+=w;var v=Er(l.adapterInfo.device);Pe.setStringView(a+20,p,v),p+=v;var x=Er(l.adapterInfo.description);Pe.setStringView(a+28,p,x),p+=x,we()[a+36>>>2>>>0]=2;var j=l.adapterInfo.isFallbackAdapter?3:4;return we()[a+40>>>2>>>0]=j,we()[a+44>>>2>>>0]=0,we()[a+48>>>2>>>0]=0,1}var Io=function(s,a){s>>>=0,a>>>=0;var l=Pe.getJsObject(s),c=nn(4*l.features.size),p=0,_=0;l.features.forEach(w=>{var v=Pe.FeatureNameString2Enum[w];v!==void 0&&(we()[c+p>>>2>>>0]=v,p+=4,_++)}),ae()[a+4>>>2>>>0]=c,ae()[a>>>2>>>0]=_};function yp(s,a){s>>>=0,a>>>=0;var l=Pe.getJsObject(s);return Pe.fillLimitStruct(l.limits,a),1}function wp(s,a){return s>>>=0,Pe.getJsObject(s).features.has(Pe.FeatureName[a])}function Ao(s,a){s>>>=0,Pe.getJsObject(s).pushErrorScope(Pe.ErrorFilter[a])}var bp=function(s,a,l){s>>>=0,a>>>=0,l>>>=0;var c=Pe.getJsObject(s),p=Array.from(we().subarray(l>>>2>>>0,l+4*a>>>2>>>0),_=>Pe.getJsObject(_));c.submit(p)};function Mp(s,a,l,c,p){s>>>=0,a>>>=0,l=er(l),c>>>=0,p>>>=0;var _=Pe.getJsObject(s),w=Pe.getJsObject(a),v=Q().subarray(c>>>0,c+p>>>0);_.writeBuffer(w,l,v,0,p)}ne.init(),(()=>{for(var s=new Array(256),a=0;a<256;++a)s[a]=String.fromCharCode(a);_n=s})(),gn=n.BindingError=class extends Error{constructor(s){super(s),this.name="BindingError"}},n.InternalError=class extends Error{constructor(s){super(s),this.name="InternalError"}},ie.push(0,1,void 0,1,null,1,!0,1,!1,1),n.count_emval_handles=()=>ie.length/2-5-J.length;var vp=[as,ye,os,Cs,Hs,ut,bt,cr,ls,Es,Xr,Ps,Ss,ks,$s,gs,to,Ld,so,Si,wo,bo,ki,Mo,xo],Y=await async function(){function s(p,_){return Y=p.exports,Y=function(v){var x,j=H=>ee=>H(ee)>>>0,B=H=>()=>H()>>>0;return(v=Object.assign({},v)).rf=j(v.rf),v.Wf=B(v.Wf),v.Yf=j(v.Yf),v.Jg=(x=v.Jg,(H,ee)=>x(H,ee)>>>0),v.Og=j(v.Og),v.Pg=B(v.Pg),v.Tg=j(v.Tg),v}(Y=Tt.instrumentWasmExports(Y)),w=Y.Ag,ne.tlsInitFunctions.push(w),Y.Bg,k=_,Ue(),Y;var w}Re();var a,l,c=at();if(n.instantiateWasm)return new Promise((p,_)=>{n.instantiateWasm(c,(w,v)=>{s(w,v),p(w.exports)})});if(f)return new Promise(p=>{Ee=_=>{var w=new WebAssembly.Instance(_,at());p(s(w,_))}});Oe??(Oe=n.locateFile?(a="ort-wasm-simd-threaded.jsep.wasm",n.locateFile?n.locateFile(a,E):E+a):new URL("/assets/ort-wasm-simd-threaded.jsep-CSH94wPT.wasm",import.meta.url).href);try{return s((l=await Je(Se,Oe,c)).instance,l.module)}catch(p){return o(p),Promise.reject(p)}}(),$i=s=>($i=Y.rf)(s),Oo=()=>(Oo=Y.sf)(),Ii=(n._OrtInit=(s,a)=>(n._OrtInit=Y.tf)(s,a),n._OrtGetLastError=(s,a)=>(n._OrtGetLastError=Y.uf)(s,a),n._OrtCreateSessionOptions=(s,a,l,c,p,_,w,v,x,j)=>(n._OrtCreateSessionOptions=Y.vf)(s,a,l,c,p,_,w,v,x,j),n._OrtAppendExecutionProvider=(s,a,l,c,p)=>(n._OrtAppendExecutionProvider=Y.wf)(s,a,l,c,p),n._OrtAddFreeDimensionOverride=(s,a,l)=>(n._OrtAddFreeDimensionOverride=Y.xf)(s,a,l),n._OrtAddSessionConfigEntry=(s,a,l)=>(n._OrtAddSessionConfigEntry=Y.yf)(s,a,l),n._OrtReleaseSessionOptions=s=>(n._OrtReleaseSessionOptions=Y.zf)(s),n._OrtCreateSession=(s,a,l)=>(n._OrtCreateSession=Y.Af)(s,a,l),n._OrtReleaseSession=s=>(n._OrtReleaseSession=Y.Bf)(s),n._OrtGetInputOutputCount=(s,a,l)=>(n._OrtGetInputOutputCount=Y.Cf)(s,a,l),n._OrtGetInputOutputMetadata=(s,a,l,c)=>(n._OrtGetInputOutputMetadata=Y.Df)(s,a,l,c),n._OrtFree=s=>(n._OrtFree=Y.Ef)(s),n._OrtCreateTensor=(s,a,l,c,p,_)=>(n._OrtCreateTensor=Y.Ff)(s,a,l,c,p,_),n._OrtGetTensorData=(s,a,l,c,p)=>(n._OrtGetTensorData=Y.Gf)(s,a,l,c,p),n._OrtReleaseTensor=s=>(n._OrtReleaseTensor=Y.Hf)(s),n._OrtCreateRunOptions=(s,a,l,c)=>(n._OrtCreateRunOptions=Y.If)(s,a,l,c),n._OrtAddRunConfigEntry=(s,a,l)=>(n._OrtAddRunConfigEntry=Y.Jf)(s,a,l),n._OrtReleaseRunOptions=s=>(n._OrtReleaseRunOptions=Y.Kf)(s),n._OrtCreateBinding=s=>(n._OrtCreateBinding=Y.Lf)(s),n._OrtBindInput=(s,a,l)=>(n._OrtBindInput=Y.Mf)(s,a,l),n._OrtBindOutput=(s,a,l,c)=>(n._OrtBindOutput=Y.Nf)(s,a,l,c),n._OrtClearBoundOutputs=s=>(n._OrtClearBoundOutputs=Y.Of)(s),n._OrtReleaseBinding=s=>(n._OrtReleaseBinding=Y.Pf)(s),n._OrtRunWithBinding=(s,a,l,c,p)=>(n._OrtRunWithBinding=Y.Qf)(s,a,l,c,p),n._OrtRun=(s,a,l,c,p,_,w,v)=>(n._OrtRun=Y.Rf)(s,a,l,c,p,_,w,v),n._OrtEndProfiling=s=>(n._OrtEndProfiling=Y.Sf)(s),n._OrtGetWebGpuDevice=s=>(Ii=n._OrtGetWebGpuDevice=Y.Tf)(s)),vn=(n._JsepOutput=(s,a,l)=>(n._JsepOutput=Y.Uf)(s,a,l),n._JsepGetNodeName=s=>(n._JsepGetNodeName=Y.Vf)(s),()=>(vn=Y.Wf)()),Zr=n._free=s=>(Zr=n._free=Y.Xf)(s),nn=n._malloc=s=>(nn=n._malloc=Y.Yf)(s),Ai=n._wgpuBufferRelease=s=>(Ai=n._wgpuBufferRelease=Y.Zf)(s),Fo=n._wgpuCreateInstance=s=>(Fo=n._wgpuCreateInstance=Y._f)(s),Oi=s=>(Oi=Y.$f)(s),Rn=s=>(Rn=Y.ag)(s),Fi=s=>(Fi=Y.bg)(s),Di=s=>(Di=Y.cg)(s),Nn=s=>(Nn=Y.dg)(s),ji=s=>(ji=Y.eg)(s),Do=s=>(Do=Y.fg)(s),Vn=s=>(Vn=Y.gg)(s),jo=s=>(jo=Y.hg)(s),Lo=s=>(Lo=Y.ig)(s),Li=s=>(Li=Y.jg)(s),zo=s=>(zo=Y.kg)(s),Bo=s=>(Bo=Y.lg)(s),Ro=s=>(Ro=Y.mg)(s),No=s=>(No=Y.ng)(s),zi=s=>(zi=Y.og)(s),Vo=s=>(Vo=Y.pg)(s),Wo=(s,a)=>(Wo=Y.qg)(s,a),Uo=(s,a)=>(Uo=Y.rg)(s,a),Bi=s=>(Bi=Y.sg)(s),Ri=s=>(Ri=Y.tg)(s),Ni=(s,a,l)=>(Ni=Y.ug)(s,a,l),Vi=(s,a,l)=>(Vi=Y.vg)(s,a,l),Wi=(s,a,l,c)=>(Wi=Y.wg)(s,a,l,c),xn=(s,a,l,c)=>(xn=Y.xg)(s,a,l,c),Wn=(s,a,l,c)=>(Wn=Y.yg)(s,a,l,c),Go=(s,a,l)=>(Go=Y.zg)(s,a,l),Ui=(s,a,l,c,p,_)=>(Ui=Y.Cg)(s,a,l,c,p,_),Ko=()=>(Ko=Y.Dg)(),Ho=(s,a,l,c,p)=>(Ho=Y.Eg)(s,a,l,c,p),Gi=s=>(Gi=Y.Fg)(s),Ki=s=>(Ki=Y.Gg)(s),qo=(s,a)=>(qo=Y.Hg)(s,a),Hi=()=>(Hi=Y.Ig)(),qi=(s,a)=>(qi=Y.Jg)(s,a),Me=(s,a)=>(Me=Y.Kg)(s,a),Qi=s=>(Qi=Y.Lg)(s),Qo=(s,a)=>(Qo=Y.Mg)(s,a),Xo=s=>(Xo=Y.Ng)(s),Xi=s=>(Xi=Y.Og)(s),Jo=()=>(Jo=Y.Pg)(),Yo=s=>(Yo=Y.Qg)(s),Ji=s=>(Ji=Y.Rg)(s),Zo=(s,a,l)=>(Zo=Y.Sg)(s,a,l),el=s=>(el=Y.Tg)(s),Un=n.dynCall_vii=(s,a,l)=>(Un=n.dynCall_vii=Y.Ug)(s,a,l),tl=n.dynCall_iiii=(s,a,l,c)=>(tl=n.dynCall_iiii=Y.Vg)(s,a,l,c),rl=n.dynCall_iii=(s,a,l)=>(rl=n.dynCall_iii=Y.Wg)(s,a,l),Gn=n.dynCall_ii=(s,a)=>(Gn=n.dynCall_ii=Y.Xg)(s,a),Yi=n.dynCall_iiiiiii=(s,a,l,c,p,_,w)=>(Yi=n.dynCall_iiiiiii=Y.Yg)(s,a,l,c,p,_,w),sl=n.dynCall_vi=(s,a)=>(sl=n.dynCall_vi=Y.Zg)(s,a),Zi=n.dynCall_v=s=>(Zi=n.dynCall_v=Y._g)(s),nl=n.dynCall_iiiiii=(s,a,l,c,p,_)=>(nl=n.dynCall_iiiiii=Y.$g)(s,a,l,c,p,_),ea=n.dynCall_viii=(s,a,l,c)=>(ea=n.dynCall_viii=Y.ah)(s,a,l,c),Tn=n.dynCall_i=s=>(Tn=n.dynCall_i=Y.bh)(s),ta=n.dynCall_iiiii=(s,a,l,c,p)=>(ta=n.dynCall_iiiii=Y.ch)(s,a,l,c,p),il=n.dynCall_viiii=(s,a,l,c,p)=>(il=n.dynCall_viiii=Y.dh)(s,a,l,c,p),al=n.dynCall_viiiii=(s,a,l,c,p,_)=>(al=n.dynCall_viiiii=Y.eh)(s,a,l,c,p,_),ol=n.dynCall_vijii=(s,a,l,c,p)=>(ol=n.dynCall_vijii=Y.fh)(s,a,l,c,p),ra=n.dynCall_vijjiii=(s,a,l,c,p,_,w)=>(ra=n.dynCall_vijjiii=Y.gh)(s,a,l,c,p,_,w),ll=n.dynCall_vijj=(s,a,l,c)=>(ll=n.dynCall_vijj=Y.hh)(s,a,l,c),cl=n.dynCall_vijji=(s,a,l,c,p)=>(cl=n.dynCall_vijji=Y.ih)(s,a,l,c,p),ul=n.dynCall_diii=(s,a,l,c)=>(ul=n.dynCall_diii=Y.jh)(s,a,l,c),sa=n.dynCall_viiij=(s,a,l,c,p)=>(sa=n.dynCall_viiij=Y.kh)(s,a,l,c,p),dl=n.dynCall_iiij=(s,a,l,c)=>(dl=n.dynCall_iiij=Y.lh)(s,a,l,c),pl=n.dynCall_viijii=(s,a,l,c,p,_)=>(pl=n.dynCall_viijii=Y.mh)(s,a,l,c,p,_),na=n.dynCall_iif=(s,a,l)=>(na=n.dynCall_iif=Y.nh)(s,a,l),hl=n.dynCall_jiii=(s,a,l,c)=>(hl=n.dynCall_jiii=Y.oh)(s,a,l,c),fl=n.dynCall_viijijj=(s,a,l,c,p,_,w)=>(fl=n.dynCall_viijijj=Y.ph)(s,a,l,c,p,_,w),ml=n.dynCall_vij=(s,a,l)=>(ml=n.dynCall_vij=Y.qh)(s,a,l),ia=n.dynCall_viij=(s,a,l,c)=>(ia=n.dynCall_viij=Y.rh)(s,a,l,c),_l=n.dynCall_iiiiid=(s,a,l,c,p,_)=>(_l=n.dynCall_iiiiid=Y.sh)(s,a,l,c,p,_),gl=n.dynCall_viijj=(s,a,l,c,p)=>(gl=n.dynCall_viijj=Y.th)(s,a,l,c,p),aa=n.dynCall_iij=(s,a,l)=>(aa=n.dynCall_iij=Y.uh)(s,a,l),yl=n.dynCall_iiiiij=(s,a,l,c,p,_)=>(yl=n.dynCall_iiiiij=Y.vh)(s,a,l,c,p,_),wl=n.dynCall_j=s=>(wl=n.dynCall_j=Y.wh)(s),oa=n.dynCall_viiiiiii=(s,a,l,c,p,_,w,v)=>(oa=n.dynCall_viiiiiii=Y.xh)(s,a,l,c,p,_,w,v),bl=n.dynCall_iiiiiiiii=(s,a,l,c,p,_,w,v,x)=>(bl=n.dynCall_iiiiiiiii=Y.yh)(s,a,l,c,p,_,w,v,x),Ml=n.dynCall_iiiiijiiiii=(s,a,l,c,p,_,w,v,x,j,B)=>(Ml=n.dynCall_iiiiijiiiii=Y.zh)(s,a,l,c,p,_,w,v,x,j,B),Kn=n.dynCall_iiiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B)=>(Kn=n.dynCall_iiiiiiiiiii=Y.Ah)(s,a,l,c,p,_,w,v,x,j,B),vl=n.dynCall_vijjjiiiiij=(s,a,l,c,p,_,w,v,x,j,B)=>(vl=n.dynCall_vijjjiiiiij=Y.Bh)(s,a,l,c,p,_,w,v,x,j,B),xl=n.dynCall_viiiiii=(s,a,l,c,p,_,w)=>(xl=n.dynCall_viiiiii=Y.Ch)(s,a,l,c,p,_,w),Tl=n.dynCall_viji=(s,a,l,c)=>(Tl=n.dynCall_viji=Y.Dh)(s,a,l,c),la=n.dynCall_viiiiiiiii=(s,a,l,c,p,_,w,v,x,j)=>(la=n.dynCall_viiiiiiiii=Y.Eh)(s,a,l,c,p,_,w,v,x,j),Cl=n.dynCall_fi=(s,a)=>(Cl=n.dynCall_fi=Y.Fh)(s,a),El=n.dynCall_fii=(s,a,l)=>(El=n.dynCall_fii=Y.Gh)(s,a,l),ca=n.dynCall_ji=(s,a)=>(ca=n.dynCall_ji=Y.Hh)(s,a),Pl=n.dynCall_di=(s,a)=>(Pl=n.dynCall_di=Y.Ih)(s,a),Sl=n.dynCall_jii=(s,a,l)=>(Sl=n.dynCall_jii=Y.Jh)(s,a,l),kl=n.dynCall_dii=(s,a,l)=>(kl=n.dynCall_dii=Y.Kh)(s,a,l),$l=n.dynCall_viijiii=(s,a,l,c,p,_,w)=>($l=n.dynCall_viijiii=Y.Lh)(s,a,l,c,p,_,w),Hn=n.dynCall_viiiiiiii=(s,a,l,c,p,_,w,v,x)=>(Hn=n.dynCall_viiiiiiii=Y.Mh)(s,a,l,c,p,_,w,v,x),Il=n.dynCall_iiiiiiiij=(s,a,l,c,p,_,w,v,x)=>(Il=n.dynCall_iiiiiiiij=Y.Nh)(s,a,l,c,p,_,w,v,x),Al=n.dynCall_iiiiiiii=(s,a,l,c,p,_,w,v)=>(Al=n.dynCall_iiiiiiii=Y.Oh)(s,a,l,c,p,_,w,v),Ol=n.dynCall_viiji=(s,a,l,c,p)=>(Ol=n.dynCall_viiji=Y.Ph)(s,a,l,c,p),Fl=n.dynCall_viiiiij=(s,a,l,c,p,_,w)=>(Fl=n.dynCall_viiiiij=Y.Qh)(s,a,l,c,p,_,w),Dl=n.dynCall_viijjjiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H)=>(Dl=n.dynCall_viijjjiiiiii=Y.Rh)(s,a,l,c,p,_,w,v,x,j,B,H),jl=n.dynCall_viiijiiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H)=>(jl=n.dynCall_viiijiiiiiii=Y.Sh)(s,a,l,c,p,_,w,v,x,j,B,H),Ll=n.dynCall_iiiiiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee)=>(Ll=n.dynCall_iiiiiiiiiiiii=Y.Th)(s,a,l,c,p,_,w,v,x,j,B,H,ee),zl=n.dynCall_viiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B)=>(zl=n.dynCall_viiiiiiiiii=Y.Uh)(s,a,l,c,p,_,w,v,x,j,B),Bl=n.dynCall_viiiiiiiiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We)=>(Bl=n.dynCall_viiiiiiiiiiiiiiii=Y.Vh)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We),ua=n.dynCall_viid=(s,a,l,c)=>(ua=n.dynCall_viid=Y.Wh)(s,a,l,c),Rl=n.dynCall_vid=(s,a,l)=>(Rl=n.dynCall_vid=Y.Xh)(s,a,l),Nl=n.dynCall_viiiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H)=>(Nl=n.dynCall_viiiiiiiiiii=Y.Yh)(s,a,l,c,p,_,w,v,x,j,B,H),da=n.dynCall_viiijjjii=(s,a,l,c,p,_,w,v,x)=>(da=n.dynCall_viiijjjii=Y.Zh)(s,a,l,c,p,_,w,v,x),Vl=n.dynCall_iid=(s,a,l)=>(Vl=n.dynCall_iid=Y._h)(s,a,l),Wl=n.dynCall_viiiij=(s,a,l,c,p,_)=>(Wl=n.dynCall_viiiij=Y.$h)(s,a,l,c,p,_),Ul=n.dynCall_viiijiiiii=(s,a,l,c,p,_,w,v,x,j)=>(Ul=n.dynCall_viiijiiiii=Y.ai)(s,a,l,c,p,_,w,v,x,j),Gl=n.dynCall_jj=(s,a)=>(Gl=n.dynCall_jj=Y.bi)(s,a),pa=n.dynCall_iiiijii=(s,a,l,c,p,_,w)=>(pa=n.dynCall_iiiijii=Y.ci)(s,a,l,c,p,_,w),Kl=n.dynCall_iiijii=(s,a,l,c,p,_)=>(Kl=n.dynCall_iiijii=Y.di)(s,a,l,c,p,_),Hl=n.dynCall_viiiiiiiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De)=>(Hl=n.dynCall_viiiiiiiiiiiiiii=Y.ei)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De),ha=n.dynCall_iiijjj=(s,a,l,c,p,_)=>(ha=n.dynCall_iiijjj=Y.fi)(s,a,l,c,p,_),ql=n.dynCall_ij=(s,a)=>(ql=n.dynCall_ij=Y.gi)(s,a),Ql=n.dynCall_viiiiji=(s,a,l,c,p,_,w)=>(Ql=n.dynCall_viiiiji=Y.hi)(s,a,l,c,p,_,w),fa=n.dynCall_iijjji=(s,a,l,c,p,_)=>(fa=n.dynCall_iijjji=Y.ii)(s,a,l,c,p,_),Xl=n.dynCall_vjiiiiii=(s,a,l,c,p,_,w,v)=>(Xl=n.dynCall_vjiiiiii=Y.ji)(s,a,l,c,p,_,w,v),Jl=n.dynCall_vijjiiiii=(s,a,l,c,p,_,w,v,x)=>(Jl=n.dynCall_vijjiiiii=Y.ki)(s,a,l,c,p,_,w,v,x),ma=n.dynCall_jiij=(s,a,l,c)=>(ma=n.dynCall_jiij=Y.li)(s,a,l,c),Yl=n.dynCall_iijijjijiji=(s,a,l,c,p,_,w,v,x,j,B)=>(Yl=n.dynCall_iijijjijiji=Y.mi)(s,a,l,c,p,_,w,v,x,j,B),Zl=n.dynCall_iijijji=(s,a,l,c,p,_,w)=>(Zl=n.dynCall_iijijji=Y.ni)(s,a,l,c,p,_,w),ec=n.dynCall_ijijji=(s,a,l,c,p,_)=>(ec=n.dynCall_ijijji=Y.oi)(s,a,l,c,p,_),tc=n.dynCall_iiiiiiij=(s,a,l,c,p,_,w,v)=>(tc=n.dynCall_iiiiiiij=Y.pi)(s,a,l,c,p,_,w,v),_a=n.dynCall_viiijjiii=(s,a,l,c,p,_,w,v,x)=>(_a=n.dynCall_viiijjiii=Y.qi)(s,a,l,c,p,_,w,v,x),rc=n.dynCall_vif=(s,a,l)=>(rc=n.dynCall_vif=Y.ri)(s,a,l),sc=n.dynCall_viif=(s,a,l,c)=>(sc=n.dynCall_viif=Y.si)(s,a,l,c),nc=n.dynCall_iiiiijji=(s,a,l,c,p,_,w,v)=>(nc=n.dynCall_iiiiijji=Y.ti)(s,a,l,c,p,_,w,v),ic=n.dynCall_iiiiji=(s,a,l,c,p,_)=>(ic=n.dynCall_iiiiji=Y.ui)(s,a,l,c,p,_),ac=n.dynCall_iiiifi=(s,a,l,c,p,_)=>(ac=n.dynCall_iiiifi=Y.vi)(s,a,l,c,p,_),oc=n.dynCall_iiiiiiiiijii=(s,a,l,c,p,_,w,v,x,j,B,H)=>(oc=n.dynCall_iiiiiiiiijii=Y.wi)(s,a,l,c,p,_,w,v,x,j,B,H),ga=n.dynCall_iiiijjii=(s,a,l,c,p,_,w,v)=>(ga=n.dynCall_iiiijjii=Y.xi)(s,a,l,c,p,_,w,v),lc=n.dynCall_iiiiiijjjii=(s,a,l,c,p,_,w,v,x,j,B)=>(lc=n.dynCall_iiiiiijjjii=Y.yi)(s,a,l,c,p,_,w,v,x,j,B),cc=n.dynCall_iiijiii=(s,a,l,c,p,_,w)=>(cc=n.dynCall_iiijiii=Y.zi)(s,a,l,c,p,_,w),ya=n.dynCall_iiiiiiiijjjfi=(s,a,l,c,p,_,w,v,x,j,B,H,ee)=>(ya=n.dynCall_iiiiiiiijjjfi=Y.Ai)(s,a,l,c,p,_,w,v,x,j,B,H,ee),uc=n.dynCall_iijiiii=(s,a,l,c,p,_,w)=>(uc=n.dynCall_iijiiii=Y.Bi)(s,a,l,c,p,_,w),dc=n.dynCall_viiiijj=(s,a,l,c,p,_,w)=>(dc=n.dynCall_viiiijj=Y.Ci)(s,a,l,c,p,_,w),Is=n.dynCall_iijjjii=(s,a,l,c,p,_,w)=>(Is=n.dynCall_iijjjii=Y.Di)(s,a,l,c,p,_,w),pc=n.dynCall_jij=(s,a,l)=>(pc=n.dynCall_jij=Y.Ei)(s,a,l),hc=n.dynCall_jjj=(s,a,l)=>(hc=n.dynCall_jjj=Y.Fi)(s,a,l),fc=n.dynCall_iiji=(s,a,l,c)=>(fc=n.dynCall_iiji=Y.Gi)(s,a,l,c),mc=n.dynCall_viffiii=(s,a,l,c,p,_,w)=>(mc=n.dynCall_viffiii=Y.Hi)(s,a,l,c,p,_,w),wa=n.dynCall_viifiii=(s,a,l,c,p,_,w)=>(wa=n.dynCall_viifiii=Y.Ii)(s,a,l,c,p,_,w),_c=n.dynCall_viiiiidiidi=(s,a,l,c,p,_,w,v,x,j,B)=>(_c=n.dynCall_viiiiidiidi=Y.Ji)(s,a,l,c,p,_,w,v,x,j,B),gc=n.dynCall_viiiiiiiiidi=(s,a,l,c,p,_,w,v,x,j,B,H)=>(gc=n.dynCall_viiiiiiiiidi=Y.Ki)(s,a,l,c,p,_,w,v,x,j,B,H),yc=n.dynCall_viiiiiiiiiiiiiifi=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We)=>(yc=n.dynCall_viiiiiiiiiiiiiifi=Y.Li)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We),wc=n.dynCall_ijii=(s,a,l,c)=>(wc=n.dynCall_ijii=Y.Mi)(s,a,l,c),qn=n.dynCall_viijiiiijiii=(s,a,l,c,p,_,w,v,x,j,B,H)=>(qn=n.dynCall_viijiiiijiii=Y.Ni)(s,a,l,c,p,_,w,v,x,j,B,H),bc=n.dynCall_vijjjjjjjjjjjjji=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De)=>(bc=n.dynCall_vijjjjjjjjjjjjji=Y.Oi)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De),Mc=n.dynCall_viiijii=(s,a,l,c,p,_,w)=>(Mc=n.dynCall_viiijii=Y.Pi)(s,a,l,c,p,_,w),vc=n.dynCall_vijjjiiji=(s,a,l,c,p,_,w,v,x)=>(vc=n.dynCall_vijjjiiji=Y.Qi)(s,a,l,c,p,_,w,v,x),Cn=n.dynCall_iiiijiiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe)=>(Cn=n.dynCall_iiiijiiiiiiiiii=Y.Ri)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe),xc=n.dynCall_iiiiiiiiii=(s,a,l,c,p,_,w,v,x,j)=>(xc=n.dynCall_iiiiiiiiii=Y.Si)(s,a,l,c,p,_,w,v,x,j),Tc=n.dynCall_vj=(s,a)=>(Tc=n.dynCall_vj=Y.Ti)(s,a),Cc=n.dynCall_vfiii=(s,a,l,c,p)=>(Cc=n.dynCall_vfiii=Y.Ui)(s,a,l,c,p),Ec=n.dynCall_viiiiff=(s,a,l,c,p,_,w)=>(Ec=n.dynCall_viiiiff=Y.Vi)(s,a,l,c,p,_,w),Qn=n.dynCall_viiiiiff=(s,a,l,c,p,_,w,v)=>(Qn=n.dynCall_viiiiiff=Y.Wi)(s,a,l,c,p,_,w,v),Pc=n.dynCall_viiff=(s,a,l,c,p)=>(Pc=n.dynCall_viiff=Y.Xi)(s,a,l,c,p),Sc=n.dynCall_viiiiiiiiifiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe)=>(Sc=n.dynCall_viiiiiiiiifiiii=Y.Yi)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe),kc=n.dynCall_viiiiiiiijj=(s,a,l,c,p,_,w,v,x,j,B)=>(kc=n.dynCall_viiiiiiiijj=Y.Zi)(s,a,l,c,p,_,w,v,x,j,B),xp=n.dynCall_iiiiiiiiiiiiiifii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We)=>(xp=n.dynCall_iiiiiiiiiiiiiifii=Y._i)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We),$c=n.dynCall_viiiiiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe)=>($c=n.dynCall_viiiiiiiiiiiii=Y.$i)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe),Ic=n.dynCall_iiiiiiiiiiiiiiiiiiifii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt,_r)=>(Ic=n.dynCall_iiiiiiiiiiiiiiiiiiifii=Y.aj)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt,_r),Ac=n.dynCall_vijjiiiiiii=(s,a,l,c,p,_,w,v,x,j,B)=>(Ac=n.dynCall_vijjiiiiiii=Y.bj)(s,a,l,c,p,_,w,v,x,j,B),an=n.dynCall_iiiijjj=(s,a,l,c,p,_,w)=>(an=n.dynCall_iiiijjj=Y.cj)(s,a,l,c,p,_,w),Oc=n.dynCall_fffffff=(s,a,l,c,p,_,w)=>(Oc=n.dynCall_fffffff=Y.dj)(s,a,l,c,p,_,w),Fc=n.dynCall_viiiiiijiifiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe)=>(Fc=n.dynCall_viiiiiijiifiii=Y.ej)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe),Dc=n.dynCall_vjjjjjjffjifiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt)=>(Dc=n.dynCall_vjjjjjjffjifiiiiii=Y.fj)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt),jc=n.dynCall_viiiiiiffjifiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We)=>(jc=n.dynCall_viiiiiiffjifiiiii=Y.gj)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We),Lc=n.dynCall_viiiiiiffjfiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De)=>(Lc=n.dynCall_viiiiiiffjfiiiii=Y.hj)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De),zc=n.dynCall_viiiiiiffjiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe)=>(zc=n.dynCall_viiiiiiffjiiiii=Y.ij)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe),Bc=n.dynCall_vjjjjjjjjfffjifiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt)=>(Bc=n.dynCall_vjjjjjjjjfffjifiiiiii=Y.jj)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt),Xn=n.dynCall_vjjjjjjfffifijiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt)=>(Xn=n.dynCall_vjjjjjjfffifijiiiii=Y.kj)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt),Tp=n.dynCall_vjjjjjjfffifiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt)=>(Tp=n.dynCall_vjjjjjjfffifiiiiii=Y.lj)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt),Rc=n.dynCall_vjjjjjjjjfffiiifiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt)=>(Rc=n.dynCall_vjjjjjjjjfffiiifiiiii=Y.mj)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt),Nc=n.dynCall_vijiiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee)=>(Nc=n.dynCall_vijiiiiiiiiii=Y.nj)(s,a,l,c,p,_,w,v,x,j,B,H,ee),Vc=n.dynCall_vijjfffiii=(s,a,l,c,p,_,w,v,x,j)=>(Vc=n.dynCall_vijjfffiii=Y.oj)(s,a,l,c,p,_,w,v,x,j),Wc=n.dynCall_jiijjiif=(s,a,l,c,p,_,w,v)=>(Wc=n.dynCall_jiijjiif=Y.pj)(s,a,l,c,p,_,w,v),Uc=n.dynCall_vijjjjjjifiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe)=>(Uc=n.dynCall_vijjjjjjifiiiii=Y.qj)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe),ba=n.dynCall_vjjjjjiiii=(s,a,l,c,p,_,w,v,x,j)=>(ba=n.dynCall_vjjjjjiiii=Y.rj)(s,a,l,c,p,_,w,v,x,j),Gc=n.dynCall_vjjjjfiii=(s,a,l,c,p,_,w,v,x)=>(Gc=n.dynCall_vjjjjfiii=Y.sj)(s,a,l,c,p,_,w,v,x),Kc=n.dynCall_viiiiiijiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe)=>(Kc=n.dynCall_viiiiiijiiiiii=Y.tj)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe),Ma=n.dynCall_vijjii=(s,a,l,c,p,_)=>(Ma=n.dynCall_vijjii=Y.uj)(s,a,l,c,p,_),Hc=n.dynCall_viiiiijjiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee)=>(Hc=n.dynCall_viiiiijjiiiii=Y.vj)(s,a,l,c,p,_,w,v,x,j,B,H,ee),qc=n.dynCall_iiiiiji=(s,a,l,c,p,_,w)=>(qc=n.dynCall_iiiiiji=Y.wj)(s,a,l,c,p,_,w),va=n.dynCall_viiiiijiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee)=>(va=n.dynCall_viiiiijiiiiii=Y.xj)(s,a,l,c,p,_,w,v,x,j,B,H,ee),Qc=n.dynCall_iiiiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H)=>(Qc=n.dynCall_iiiiiiiiiiii=Y.yj)(s,a,l,c,p,_,w,v,x,j,B,H),Xc=n.dynCall_viiijiiiiii=(s,a,l,c,p,_,w,v,x,j,B)=>(Xc=n.dynCall_viiijiiiiii=Y.zj)(s,a,l,c,p,_,w,v,x,j,B),Jn=n.dynCall_viiiijii=(s,a,l,c,p,_,w,v)=>(Jn=n.dynCall_viiiijii=Y.Aj)(s,a,l,c,p,_,w,v),Jc=n.dynCall_viijjiii=(s,a,l,c,p,_,w,v)=>(Jc=n.dynCall_viijjiii=Y.Bj)(s,a,l,c,p,_,w,v),Yc=n.dynCall_viiiji=(s,a,l,c,p,_)=>(Yc=n.dynCall_viiiji=Y.Cj)(s,a,l,c,p,_),Zc=n.dynCall_viiiiiijii=(s,a,l,c,p,_,w,v,x,j)=>(Zc=n.dynCall_viiiiiijii=Y.Dj)(s,a,l,c,p,_,w,v,x,j),eu=n.dynCall_viiiiijjji=(s,a,l,c,p,_,w,v,x,j)=>(eu=n.dynCall_viiiiijjji=Y.Ej)(s,a,l,c,p,_,w,v,x,j),xa=n.dynCall_vijiii=(s,a,l,c,p,_)=>(xa=n.dynCall_vijiii=Y.Fj)(s,a,l,c,p,_),Yn=n.dynCall_iiijiiii=(s,a,l,c,p,_,w,v)=>(Yn=n.dynCall_iiijiiii=Y.Gj)(s,a,l,c,p,_,w,v),Cp=n.dynCall_viiiiiijjiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe)=>(Cp=n.dynCall_viiiiiijjiiiii=Y.Hj)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe),tu=n.dynCall_viiiiiiijiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe)=>(tu=n.dynCall_viiiiiiijiiiiii=Y.Ij)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe),ru=n.dynCall_viiiiiji=(s,a,l,c,p,_,w,v)=>(ru=n.dynCall_viiiiiji=Y.Jj)(s,a,l,c,p,_,w,v),Ta=n.dynCall_fiif=(s,a,l,c)=>(Ta=n.dynCall_fiif=Y.Kj)(s,a,l,c),su=n.dynCall_viijjjiii=(s,a,l,c,p,_,w,v,x)=>(su=n.dynCall_viijjjiii=Y.Lj)(s,a,l,c,p,_,w,v,x),nu=n.dynCall_viiiiiifiii=(s,a,l,c,p,_,w,v,x,j,B)=>(nu=n.dynCall_viiiiiifiii=Y.Mj)(s,a,l,c,p,_,w,v,x,j,B),Ca=n.dynCall_viijji=(s,a,l,c,p,_)=>(Ca=n.dynCall_viijji=Y.Nj)(s,a,l,c,p,_),iu=n.dynCall_iiiiiiiiiiijijji=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De)=>(iu=n.dynCall_iiiiiiiiiiijijji=Y.Oj)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De),au=n.dynCall_jiijjiii=(s,a,l,c,p,_,w,v)=>(au=n.dynCall_jiijjiii=Y.Pj)(s,a,l,c,p,_,w,v),Ea=n.dynCall_viifiifijjjii=(s,a,l,c,p,_,w,v,x,j,B,H,ee)=>(Ea=n.dynCall_viifiifijjjii=Y.Qj)(s,a,l,c,p,_,w,v,x,j,B,H,ee),Ep=n.dynCall_viiiiiiiiiiiiiiiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt,_r,jr,As)=>(Ep=n.dynCall_viiiiiiiiiiiiiiiiiiiiiii=Y.Rj)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt,_r,jr,As),Pp=n.dynCall_viiiiifiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee)=>(Pp=n.dynCall_viiiiifiiiiii=Y.Sj)(s,a,l,c,p,_,w,v,x,j,B,H,ee),Pa=n.dynCall_vijjiiiiii=(s,a,l,c,p,_,w,v,x,j)=>(Pa=n.dynCall_vijjiiiiii=Y.Tj)(s,a,l,c,p,_,w,v,x,j),Zn=n.dynCall_vijiiiiiiijjii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe)=>(Zn=n.dynCall_vijiiiiiiijjii=Y.Uj)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe),ou=n.dynCall_viiiiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee)=>(ou=n.dynCall_viiiiiiiiiiii=Y.Vj)(s,a,l,c,p,_,w,v,x,j,B,H,ee),lu=n.dynCall_viiiiiiiiiiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt)=>(lu=n.dynCall_viiiiiiiiiiiiiiiiii=Y.Wj)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt),cu=n.dynCall_viiiiiiiiiiiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot)=>(cu=n.dynCall_viiiiiiiiiiiiiiiiiii=Y.Xj)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot),ei=n.dynCall_viiijiiiiiiiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot)=>(ei=n.dynCall_viiijiiiiiiiiiiiiiii=Y.Yj)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot),uu=n.dynCall_viiiijjj=(s,a,l,c,p,_,w,v)=>(uu=n.dynCall_viiiijjj=Y.Zj)(s,a,l,c,p,_,w,v),du=n.dynCall_viiiiiiijjj=(s,a,l,c,p,_,w,v,x,j,B)=>(du=n.dynCall_viiiiiiijjj=Y._j)(s,a,l,c,p,_,w,v,x,j,B),pu=n.dynCall_iiiiiiiiiiiiiiiiiiiifi=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt,_r)=>(pu=n.dynCall_iiiiiiiiiiiiiiiiiiiifi=Y.$j)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt,_r),ti=n.dynCall_viiijiiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe)=>(ti=n.dynCall_viiijiiiiiiiiii=Y.ak)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe),hu=n.dynCall_viiiiif=(s,a,l,c,p,_,w)=>(hu=n.dynCall_viiiiif=Y.bk)(s,a,l,c,p,_,w),fu=n.dynCall_viiif=(s,a,l,c,p)=>(fu=n.dynCall_viiif=Y.ck)(s,a,l,c,p),mu=n.dynCall_viiiiiiiiifi=(s,a,l,c,p,_,w,v,x,j,B,H)=>(mu=n.dynCall_viiiiiiiiifi=Y.dk)(s,a,l,c,p,_,w,v,x,j,B,H),ri=n.dynCall_viiiiid=(s,a,l,c,p,_,w)=>(ri=n.dynCall_viiiiid=Y.ek)(s,a,l,c,p,_,w),_u=n.dynCall_viiid=(s,a,l,c,p)=>(_u=n.dynCall_viiid=Y.fk)(s,a,l,c,p),gu=n.dynCall_iiif=(s,a,l,c)=>(gu=n.dynCall_iiif=Y.gk)(s,a,l,c),yu=n.dynCall_iiiij=(s,a,l,c,p)=>(yu=n.dynCall_iiiij=Y.hk)(s,a,l,c,p),wu=n.dynCall_iiiiiif=(s,a,l,c,p,_,w)=>(wu=n.dynCall_iiiiiif=Y.ik)(s,a,l,c,p,_,w),bu=n.dynCall_vijij=(s,a,l,c,p)=>(bu=n.dynCall_vijij=Y.jk)(s,a,l,c,p),Mu=n.dynCall_viiijji=(s,a,l,c,p,_,w)=>(Mu=n.dynCall_viiijji=Y.kk)(s,a,l,c,p,_,w),vu=n.dynCall_viiiiiiffii=(s,a,l,c,p,_,w,v,x,j,B)=>(vu=n.dynCall_viiiiiiffii=Y.lk)(s,a,l,c,p,_,w,v,x,j,B),xu=n.dynCall_iifii=(s,a,l,c,p)=>(xu=n.dynCall_iifii=Y.mk)(s,a,l,c,p),Tu=n.dynCall_viiiiifi=(s,a,l,c,p,_,w,v)=>(Tu=n.dynCall_viiiiifi=Y.nk)(s,a,l,c,p,_,w,v),Cu=n.dynCall_iifjii=(s,a,l,c,p,_)=>(Cu=n.dynCall_iifjii=Y.ok)(s,a,l,c,p,_),Sa=n.dynCall_vidi=(s,a,l,c)=>(Sa=n.dynCall_vidi=Y.pk)(s,a,l,c),Eu=n.dynCall_viiijiji=(s,a,l,c,p,_,w,v)=>(Eu=n.dynCall_viiijiji=Y.qk)(s,a,l,c,p,_,w,v),Pu=n.dynCall_viiijij=(s,a,l,c,p,_,w)=>(Pu=n.dynCall_viiijij=Y.rk)(s,a,l,c,p,_,w),ka=n.dynCall_vijjj=(s,a,l,c,p)=>(ka=n.dynCall_vijjj=Y.sk)(s,a,l,c,p),Su=n.dynCall_vjiij=(s,a,l,c,p)=>(Su=n.dynCall_vjiij=Y.tk)(s,a,l,c,p),ku=n.dynCall_diiiii=(s,a,l,c,p,_)=>(ku=n.dynCall_diiiii=Y.uk)(s,a,l,c,p,_),$u=n.dynCall_diiii=(s,a,l,c,p)=>($u=n.dynCall_diiii=Y.vk)(s,a,l,c,p),Iu=n.dynCall_iiiji=(s,a,l,c,p)=>(Iu=n.dynCall_iiiji=Y.wk)(s,a,l,c,p),si=n.dynCall_ijiijji=(s,a,l,c,p,_,w)=>(si=n.dynCall_ijiijji=Y.xk)(s,a,l,c,p,_,w),Au=n.dynCall_viiijjiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H)=>(Au=n.dynCall_viiijjiiiiii=Y.yk)(s,a,l,c,p,_,w,v,x,j,B,H),Ou=n.dynCall_viijjijjjjiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee)=>(Ou=n.dynCall_viijjijjjjiii=Y.zk)(s,a,l,c,p,_,w,v,x,j,B,H,ee),Fu=n.dynCall_ijiii=(s,a,l,c,p)=>(Fu=n.dynCall_ijiii=Y.Ak)(s,a,l,c,p),Du=n.dynCall_ijiiiiji=(s,a,l,c,p,_,w,v)=>(Du=n.dynCall_ijiiiiji=Y.Bk)(s,a,l,c,p,_,w,v),ju=n.dynCall_iiifii=(s,a,l,c,p,_)=>(ju=n.dynCall_iiifii=Y.Ck)(s,a,l,c,p,_),ni=n.dynCall_ijiij=(s,a,l,c,p)=>(ni=n.dynCall_ijiij=Y.Dk)(s,a,l,c,p),Lu=n.dynCall_viijiiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe)=>(Lu=n.dynCall_viijiiiiiiiiii=Y.Ek)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe),zu=n.dynCall_fiiii=(s,a,l,c,p)=>(zu=n.dynCall_fiiii=Y.Fk)(s,a,l,c,p),Bu=n.dynCall_jfi=(s,a,l)=>(Bu=n.dynCall_jfi=Y.Gk)(s,a,l),Ru=n.dynCall_viiiiiiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe)=>(Ru=n.dynCall_viiiiiiiiiiiiii=Y.Hk)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe),Nu=n.dynCall_viiiiiiiiiiiiiiiiiiii=(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt)=>(Nu=n.dynCall_viiiiiiiiiiiiiiiiiiii=Y.Ik)(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt),ii=n.dynCall_fiii=(s,a,l,c)=>(ii=n.dynCall_fiii=Y.Jk)(s,a,l,c),Vu=s=>(Vu=Y.Kk)(s),Wu=()=>(Wu=Y.Lk)(),Uu=s=>(Uu=Y.Mk)(s),Ut=()=>(Ut=Y.Nk)();function df(s,a,l,c){var p=he();try{return tl(s,a,l,c)}catch(_){if(ce(p),_!==_+0)throw _;Me(1,0)}}function pf(s,a,l){var c=he();try{return rl(s,a,l)}catch(p){if(ce(c),p!==p+0)throw p;Me(1,0)}}function hf(s,a,l){var c=he();try{Un(s,a,l)}catch(p){if(ce(c),p!==p+0)throw p;Me(1,0)}}function ff(s,a){var l=he();try{return Gn(s,a)}catch(c){if(ce(l),c!==c+0)throw c;Me(1,0)}}function Gu(s,a){var l=he();try{sl(s,a)}catch(c){if(ce(l),c!==c+0)throw c;Me(1,0)}}function Sp(s){var a=he();try{Zi(s)}catch(l){if(ce(a),l!==l+0)throw l;Me(1,0)}}function kp(s,a,l,c,p,_,w){var v=he();try{return Yi(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function $p(s,a,l,c,p){var _=he();try{ol(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function Ip(s,a,l,c,p,_){var w=he();try{return nl(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}function Ku(s,a,l,c){var p=he();try{ea(s,a,l,c)}catch(_){if(ce(p),_!==_+0)throw _;Me(1,0)}}function Hu(s,a,l,c,p,_,w){var v=he();try{ra(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function mf(s,a,l,c,p){var _=he();try{il(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function Ap(s,a,l,c,p){var _=he();try{return ta(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function Op(s,a,l,c,p,_,w){var v=he();try{xl(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function qu(s,a,l,c,p,_){var w=he();try{al(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}function Fp(s){var a=he();try{return Tn(s)}catch(l){if(ce(a),l!==l+0)throw l;Me(1,0)}}function Dp(s,a,l,c){var p=he();try{return dl(s,a,l,c)}catch(_){if(ce(p),_!==_+0)throw _;Me(1,0)}}function _f(s,a,l,c,p,_){var w=he();try{pl(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}function jp(s,a,l,c){var p=he();try{return hl(s,a,l,c)}catch(_){if(ce(p),_!==_+0)throw _;return Me(1,0),0n}}function Lp(s,a,l,c,p,_,w){var v=he();try{fl(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function zp(s,a,l){var c=he();try{ml(s,a,l)}catch(p){if(ce(c),p!==p+0)throw p;Me(1,0)}}function En(s,a,l,c){var p=he();try{ia(s,a,l,c)}catch(_){if(ce(p),_!==_+0)throw _;Me(1,0)}}function Bp(s,a,l){var c=he();try{return na(s,a,l)}catch(p){if(ce(c),p!==p+0)throw p;Me(1,0)}}function Rp(s,a,l){var c=he();try{return aa(s,a,l)}catch(p){if(ce(c),p!==p+0)throw p;Me(1,0)}}function Np(s,a,l,c,p){var _=he();try{gl(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function Vp(s,a,l,c,p,_,w,v){var x=he();try{oa(s,a,l,c,p,_,w,v)}catch(j){if(ce(x),j!==j+0)throw j;Me(1,0)}}function Wp(s,a,l,c,p,_,w,v,x,j,B){var H=he();try{return Ml(s,a,l,c,p,_,w,v,x,j,B)}catch(ee){if(ce(H),ee!==ee+0)throw ee;Me(1,0)}}function Up(s,a,l,c,p,_,w,v,x,j,B){var H=he();try{return Kn(s,a,l,c,p,_,w,v,x,j,B)}catch(ee){if(ce(H),ee!==ee+0)throw ee;Me(1,0)}}function Gp(s,a,l,c,p,_,w,v,x,j,B){var H=he();try{vl(s,a,l,c,p,_,w,v,x,j,B)}catch(ee){if(ce(H),ee!==ee+0)throw ee;Me(1,0)}}function Kp(s,a,l,c,p,_,w,v,x,j){var B=he();try{la(s,a,l,c,p,_,w,v,x,j)}catch(H){if(ce(B),H!==H+0)throw H;Me(1,0)}}function gf(s,a,l,c,p,_,w,v,x){var j=he();try{return bl(s,a,l,c,p,_,w,v,x)}catch(B){if(ce(j),B!==B+0)throw B;Me(1,0)}}function Hp(s,a){var l=he();try{return Cl(s,a)}catch(c){if(ce(l),c!==c+0)throw c;Me(1,0)}}function qp(s,a){var l=he();try{return ca(s,a)}catch(c){if(ce(l),c!==c+0)throw c;return Me(1,0),0n}}function Qp(s,a){var l=he();try{return Pl(s,a)}catch(c){if(ce(l),c!==c+0)throw c;Me(1,0)}}function Xp(s,a,l,c){var p=he();try{ll(s,a,l,c)}catch(_){if(ce(p),_!==_+0)throw _;Me(1,0)}}function Jp(s,a,l,c,p,_,w){var v=he();try{Mc(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function Yp(s,a,l,c,p,_,w){var v=he();try{$l(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function yf(s,a,l,c,p,_,w,v){var x=he();try{Jn(s,a,l,c,p,_,w,v)}catch(j){if(ce(x),j!==j+0)throw j;Me(1,0)}}function Zp(s,a,l,c,p,_,w,v,x){var j=he();try{Hn(s,a,l,c,p,_,w,v,x)}catch(B){if(ce(j),B!==B+0)throw B;Me(1,0)}}function Qu(s,a,l,c){var p=he();try{Tl(s,a,l,c)}catch(_){if(ce(p),_!==_+0)throw _;Me(1,0)}}function eh(s,a,l,c,p,_,w,v,x){var j=he();try{return Il(s,a,l,c,p,_,w,v,x)}catch(B){if(ce(j),B!==B+0)throw B;Me(1,0)}}function th(s,a,l,c,p,_,w,v){var x=he();try{return Al(s,a,l,c,p,_,w,v)}catch(j){if(ce(x),j!==j+0)throw j;Me(1,0)}}function rh(s,a,l,c,p){var _=he();try{Ol(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function sh(s,a,l){var c=he();try{return Sl(s,a,l)}catch(p){if(ce(c),p!==p+0)throw p;return Me(1,0),0n}}function nh(s,a,l,c,p,_,w){var v=he();try{Fl(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function ih(s,a,l,c,p,_,w,v,x,j,B,H){var ee=he();try{Dl(s,a,l,c,p,_,w,v,x,j,B,H)}catch(fe){if(ce(ee),fe!==fe+0)throw fe;Me(1,0)}}function ah(s,a,l,c,p,_,w,v,x,j,B,H){var ee=he();try{jl(s,a,l,c,p,_,w,v,x,j,B,H)}catch(fe){if(ce(ee),fe!==fe+0)throw fe;Me(1,0)}}function oh(s,a,l,c,p,_,w,v,x,j,B,H,ee){var fe=he();try{return Ll(s,a,l,c,p,_,w,v,x,j,B,H,ee)}catch(xe){if(ce(fe),xe!==xe+0)throw xe;Me(1,0)}}function lh(s,a,l,c,p,_,w,v,x,j,B){var H=he();try{zl(s,a,l,c,p,_,w,v,x,j,B)}catch(ee){if(ce(H),ee!==ee+0)throw ee;Me(1,0)}}function ch(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We){var nt=he();try{Bl(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We)}catch(pt){if(ce(nt),pt!==pt+0)throw pt;Me(1,0)}}function uh(s,a,l,c){var p=he();try{ua(s,a,l,c)}catch(_){if(ce(p),_!==_+0)throw _;Me(1,0)}}function dh(s,a,l){var c=he();try{Rl(s,a,l)}catch(p){if(ce(c),p!==p+0)throw p;Me(1,0)}}function ph(s,a,l,c,p){var _=he();try{sa(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function hh(s,a,l,c,p){var _=he();try{cl(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function fh(s,a,l,c,p,_,w,v,x,j,B,H){var ee=he();try{Nl(s,a,l,c,p,_,w,v,x,j,B,H)}catch(fe){if(ce(ee),fe!==fe+0)throw fe;Me(1,0)}}function mh(s,a,l){var c=he();try{return Vl(s,a,l)}catch(p){if(ce(c),p!==p+0)throw p;Me(1,0)}}function _h(s,a,l,c,p,_){var w=he();try{Wl(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}function gh(s,a,l,c,p,_,w,v,x,j){var B=he();try{Ul(s,a,l,c,p,_,w,v,x,j)}catch(H){if(ce(B),H!==H+0)throw H;Me(1,0)}}function yh(s,a,l,c,p,_,w,v,x){var j=he();try{da(s,a,l,c,p,_,w,v,x)}catch(B){if(ce(j),B!==B+0)throw B;Me(1,0)}}function wh(s,a){var l=he();try{return ql(s,a)}catch(c){if(ce(l),c!==c+0)throw c;Me(1,0)}}function bh(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe){var xe=he();try{Lu(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe)}catch(De){if(ce(xe),De!==De+0)throw De;Me(1,0)}}function Mh(s,a,l,c,p,_){var w=he();try{return Kl(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}function vh(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De){var We=he();try{Hl(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De)}catch(nt){if(ce(We),nt!==nt+0)throw nt;Me(1,0)}}function xh(s,a,l,c){var p=he();try{return ma(s,a,l,c)}catch(_){if(ce(p),_!==_+0)throw _;return Me(1,0),0n}}function Th(s,a,l,c,p,_){var w=he();try{return ha(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}function Ch(s,a,l,c,p,_,w,v){var x=he();try{Xl(s,a,l,c,p,_,w,v)}catch(j){if(ce(x),j!==j+0)throw j;Me(1,0)}}function Eh(s,a,l,c,p,_,w,v,x){var j=he();try{Jl(s,a,l,c,p,_,w,v,x)}catch(B){if(ce(j),B!==B+0)throw B;Me(1,0)}}function Ph(s,a,l,c,p,_,w){var v=he();try{Ql(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function Sh(s,a,l,c,p,_){var w=he();try{return fa(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}function kh(s,a,l,c,p,_,w,v,x,j,B){var H=he();try{return Yl(s,a,l,c,p,_,w,v,x,j,B)}catch(ee){if(ce(H),ee!==ee+0)throw ee;Me(1,0)}}function wf(s,a,l,c,p,_,w){var v=he();try{return Zl(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function Dt(s,a,l,c,p,_){var w=he();try{return ec(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}function $h(s,a){var l=he();try{return Gl(s,a)}catch(c){if(ce(l),c!==c+0)throw c;return Me(1,0),0n}}function Sr(s,a,l,c,p,_,w){var v=he();try{return pa(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function Or(s,a,l){var c=he();try{rc(s,a,l)}catch(p){if(ce(c),p!==p+0)throw p;Me(1,0)}}function Nr(s,a,l,c,p,_,w,v){var x=he();try{return tc(s,a,l,c,p,_,w,v)}catch(j){if(ce(x),j!==j+0)throw j;Me(1,0)}}function on(s,a,l,c,p,_,w,v,x){var j=he();try{_a(s,a,l,c,p,_,w,v,x)}catch(B){if(ce(j),B!==B+0)throw B;Me(1,0)}}function bf(s,a,l,c){var p=he();try{sc(s,a,l,c)}catch(_){if(ce(p),_!==_+0)throw _;Me(1,0)}}function Ih(s,a,l,c,p,_,w,v){var x=he();try{return nc(s,a,l,c,p,_,w,v)}catch(j){if(ce(x),j!==j+0)throw j;Me(1,0)}}function Ah(s,a,l,c,p,_){var w=he();try{return ic(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}function S(s,a,l,c,p,_){var w=he();try{return ac(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}function F(s,a,l,c,p,_,w,v,x,j,B,H){var ee=he();try{return oc(s,a,l,c,p,_,w,v,x,j,B,H)}catch(fe){if(ce(ee),fe!==fe+0)throw fe;Me(1,0)}}function V(s,a,l,c,p,_,w,v){var x=he();try{return ga(s,a,l,c,p,_,w,v)}catch(j){if(ce(x),j!==j+0)throw j;Me(1,0)}}function _e(s,a,l,c,p,_,w,v,x,j,B){var H=he();try{return lc(s,a,l,c,p,_,w,v,x,j,B)}catch(ee){if(ce(H),ee!==ee+0)throw ee;Me(1,0)}}function Ie(s,a,l,c,p,_,w){var v=he();try{return cc(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function ke(s,a,l,c,p,_,w,v,x,j,B,H,ee){var fe=he();try{return ya(s,a,l,c,p,_,w,v,x,j,B,H,ee)}catch(xe){if(ce(fe),xe!==xe+0)throw xe;Me(1,0)}}function Be(s,a,l,c,p,_,w){var v=he();try{return uc(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function Qe(s,a,l,c,p,_,w){var v=he();try{dc(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function rt(s,a,l,c,p,_,w){var v=he();try{return Is(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function lt(s,a,l){var c=he();try{return pc(s,a,l)}catch(p){if(ce(c),p!==p+0)throw p;return Me(1,0),0n}}function Pt(s,a,l){var c=he();try{return hc(s,a,l)}catch(p){if(ce(c),p!==p+0)throw p;return Me(1,0),0n}}function It(s,a,l,c){var p=he();try{return fc(s,a,l,c)}catch(_){if(ce(p),_!==_+0)throw _;Me(1,0)}}function Ct(s,a,l,c,p,_,w){var v=he();try{mc(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function Bt(s,a,l,c,p,_,w){var v=he();try{wa(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function kt(s,a,l,c,p,_,w,v,x,j,B){var H=he();try{_c(s,a,l,c,p,_,w,v,x,j,B)}catch(ee){if(ce(H),ee!==ee+0)throw ee;Me(1,0)}}function jt(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We){var nt=he();try{yc(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We)}catch(pt){if(ce(nt),pt!==pt+0)throw pt;Me(1,0)}}function _t(s,a,l,c,p,_,w,v,x,j,B,H){var ee=he();try{qn(s,a,l,c,p,_,w,v,x,j,B,H)}catch(fe){if(ce(ee),fe!==fe+0)throw fe;Me(1,0)}}function Wt(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De){var We=he();try{bc(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De)}catch(nt){if(ce(We),nt!==nt+0)throw nt;Me(1,0)}}function nr(s,a,l,c,p,_,w,v,x){var j=he();try{vc(s,a,l,c,p,_,w,v,x)}catch(B){if(ce(j),B!==B+0)throw B;Me(1,0)}}function ar(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe){var De=he();try{return Cn(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe)}catch(We){if(ce(De),We!==We+0)throw We;Me(1,0)}}function mr(s,a,l,c,p,_,w,v,x,j){var B=he();try{return xc(s,a,l,c,p,_,w,v,x,j)}catch(H){if(ce(B),H!==H+0)throw H;Me(1,0)}}function hr(s,a){var l=he();try{Tc(s,a)}catch(c){if(ce(l),c!==c+0)throw c;Me(1,0)}}function wr(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe){var De=he();try{Sc(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe)}catch(We){if(ce(De),We!==We+0)throw We;Me(1,0)}}function kr(s,a,l,c,p){var _=he();try{Cc(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function Rt(s,a,l,c,p,_,w){var v=he();try{Ec(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function xr(s,a,l,c,p){var _=he();try{Pc(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function es(s,a,l,c,p,_,w,v){var x=he();try{Qn(s,a,l,c,p,_,w,v)}catch(j){if(ce(x),j!==j+0)throw j;Me(1,0)}}function ln(s,a,l,c,p,_,w,v,x,j,B){var H=he();try{kc(s,a,l,c,p,_,w,v,x,j,B)}catch(ee){if(ce(H),ee!==ee+0)throw ee;Me(1,0)}}function Gr(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We){var nt=he();try{return xp(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We)}catch(pt){if(ce(nt),pt!==pt+0)throw pt;Me(1,0)}}function ns(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe){var xe=he();try{$c(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe)}catch(De){if(ce(xe),De!==De+0)throw De;Me(1,0)}}function $a(s,a,l,c,p){var _=he();try{return zu(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function cn(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt,_r){var jr=he();try{return Ic(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt,_r)}catch(As){if(ce(jr),As!==As+0)throw As;Me(1,0)}}function Ia(s,a,l,c,p,_,w,v,x,j,B){var H=he();try{Ac(s,a,l,c,p,_,w,v,x,j,B)}catch(ee){if(ce(H),ee!==ee+0)throw ee;Me(1,0)}}function Aa(s,a,l,c,p,_,w){var v=he();try{return an(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function or(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe){var xe=he();try{Fc(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe)}catch(De){if(ce(xe),De!==De+0)throw De;Me(1,0)}}function Tr(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt){var pt=he();try{Dc(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt)}catch(Ot){if(ce(pt),Ot!==Ot+0)throw Ot;Me(1,0)}}function Ys(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We){var nt=he();try{jc(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We)}catch(pt){if(ce(nt),pt!==pt+0)throw pt;Me(1,0)}}function un(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De){var We=he();try{Lc(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De)}catch(nt){if(ce(We),nt!==nt+0)throw nt;Me(1,0)}}function Pn(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe){var De=he();try{zc(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe)}catch(We){if(ce(De),We!==We+0)throw We;Me(1,0)}}function Oa(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt){var _r=he();try{Bc(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt)}catch(jr){if(ce(_r),jr!==jr+0)throw jr;Me(1,0)}}function Xu(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt){var Ot=he();try{Xn(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt)}catch(Gt){if(ce(Ot),Gt!==Gt+0)throw Gt;Me(1,0)}}function Ju(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt){var pt=he();try{Tp(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt)}catch(Ot){if(ce(pt),Ot!==Ot+0)throw Ot;Me(1,0)}}function Yu(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt){var _r=he();try{Rc(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt)}catch(jr){if(ce(_r),jr!==jr+0)throw jr;Me(1,0)}}function Mf(s,a,l,c,p,_,w,v,x,j,B,H,ee){var fe=he();try{Nc(s,a,l,c,p,_,w,v,x,j,B,H,ee)}catch(xe){if(ce(fe),xe!==xe+0)throw xe;Me(1,0)}}function Zu(s,a,l,c,p,_,w,v,x,j){var B=he();try{Vc(s,a,l,c,p,_,w,v,x,j)}catch(H){if(ce(B),H!==H+0)throw H;Me(1,0)}}function Fa(s,a,l,c,p,_,w,v){var x=he();try{return Wc(s,a,l,c,p,_,w,v)}catch(j){if(ce(x),j!==j+0)throw j;return Me(1,0),0n}}function ed(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe){var De=he();try{Uc(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe)}catch(We){if(ce(De),We!==We+0)throw We;Me(1,0)}}function Oh(s,a,l,c,p,_,w,v,x,j){var B=he();try{ba(s,a,l,c,p,_,w,v,x,j)}catch(H){if(ce(B),H!==H+0)throw H;Me(1,0)}}function td(s,a,l,c,p,_,w,v,x){var j=he();try{Gc(s,a,l,c,p,_,w,v,x)}catch(B){if(ce(j),B!==B+0)throw B;Me(1,0)}}function Da(s,a,l,c,p,_,w){var v=he();try{return Oc(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function vf(s,a,l){var c=he();try{return Bu(s,a,l)}catch(p){if(ce(c),p!==p+0)throw p;return Me(1,0),0n}}function xf(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe){var xe=he();try{Kc(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe)}catch(De){if(ce(xe),De!==De+0)throw De;Me(1,0)}}function Tf(s,a,l,c,p,_,w,v,x,j,B,H,ee){var fe=he();try{Hc(s,a,l,c,p,_,w,v,x,j,B,H,ee)}catch(xe){if(ce(fe),xe!==xe+0)throw xe;Me(1,0)}}function dn(s,a,l,c,p,_,w){var v=he();try{return qc(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function hs(s,a,l,c,p,_,w,v,x,j,B,H,ee){var fe=he();try{va(s,a,l,c,p,_,w,v,x,j,B,H,ee)}catch(xe){if(ce(fe),xe!==xe+0)throw xe;Me(1,0)}}function Hv(s,a,l,c,p,_,w,v,x,j,B,H){var ee=he();try{return Qc(s,a,l,c,p,_,w,v,x,j,B,H)}catch(fe){if(ce(ee),fe!==fe+0)throw fe;Me(1,0)}}function qv(s,a,l,c,p,_){var w=he();try{Ma(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}function Qv(s,a,l,c,p,_,w,v,x,j,B){var H=he();try{Xc(s,a,l,c,p,_,w,v,x,j,B)}catch(ee){if(ce(H),ee!==ee+0)throw ee;Me(1,0)}}function Xv(s,a,l,c,p,_,w,v){var x=he();try{Jc(s,a,l,c,p,_,w,v)}catch(j){if(ce(x),j!==j+0)throw j;Me(1,0)}}function Jv(s,a,l,c,p,_){var w=he();try{Yc(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}function Yv(s,a,l,c,p,_,w,v,x,j){var B=he();try{Zc(s,a,l,c,p,_,w,v,x,j)}catch(H){if(ce(B),H!==H+0)throw H;Me(1,0)}}function Zv(s,a,l,c){var p=he();try{return wc(s,a,l,c)}catch(_){if(ce(p),_!==_+0)throw _;Me(1,0)}}function ex(s,a,l,c,p,_,w,v,x,j){var B=he();try{eu(s,a,l,c,p,_,w,v,x,j)}catch(H){if(ce(B),H!==H+0)throw H;Me(1,0)}}function tx(s,a,l,c,p,_){var w=he();try{xa(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}function rx(s,a,l){var c=he();try{return El(s,a,l)}catch(p){if(ce(c),p!==p+0)throw p;Me(1,0)}}function sx(s,a,l,c,p,_,w,v){var x=he();try{return Yn(s,a,l,c,p,_,w,v)}catch(j){if(ce(x),j!==j+0)throw j;Me(1,0)}}function nx(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe){var De=he();try{Ru(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe)}catch(We){if(ce(De),We!==We+0)throw We;Me(1,0)}}function ix(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe){var xe=he();try{Cp(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe)}catch(De){if(ce(xe),De!==De+0)throw De;Me(1,0)}}function ax(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe){var De=he();try{tu(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe)}catch(We){if(ce(De),We!==We+0)throw We;Me(1,0)}}function ox(s,a,l,c,p,_,w,v){var x=he();try{ru(s,a,l,c,p,_,w,v)}catch(j){if(ce(x),j!==j+0)throw j;Me(1,0)}}function lx(s,a,l,c){var p=he();try{return Ta(s,a,l,c)}catch(_){if(ce(p),_!==_+0)throw _;Me(1,0)}}function cx(s,a,l){var c=he();try{return kl(s,a,l)}catch(p){if(ce(c),p!==p+0)throw p;Me(1,0)}}function ux(s,a,l,c,p,_,w,v,x){var j=he();try{su(s,a,l,c,p,_,w,v,x)}catch(B){if(ce(j),B!==B+0)throw B;Me(1,0)}}function dx(s,a,l,c,p,_,w,v,x,j,B){var H=he();try{nu(s,a,l,c,p,_,w,v,x,j,B)}catch(ee){if(ce(H),ee!==ee+0)throw ee;Me(1,0)}}function px(s,a,l,c,p,_){var w=he();try{Ca(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}function hx(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De){var We=he();try{return iu(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De)}catch(nt){if(ce(We),nt!==nt+0)throw nt;Me(1,0)}}function fx(s,a,l,c,p,_,w,v){var x=he();try{return au(s,a,l,c,p,_,w,v)}catch(j){if(ce(x),j!==j+0)throw j;return Me(1,0),0n}}function mx(s,a,l,c,p,_,w,v,x,j,B,H,ee){var fe=he();try{Ea(s,a,l,c,p,_,w,v,x,j,B,H,ee)}catch(xe){if(ce(fe),xe!==xe+0)throw xe;Me(1,0)}}function _x(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt,_r,jr,As){var u1=he();try{Ep(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt,_r,jr,As)}catch(Cf){if(ce(u1),Cf!==Cf+0)throw Cf;Me(1,0)}}function gx(s,a,l,c,p,_,w,v,x,j,B,H,ee){var fe=he();try{Pp(s,a,l,c,p,_,w,v,x,j,B,H,ee)}catch(xe){if(ce(fe),xe!==xe+0)throw xe;Me(1,0)}}function yx(s,a,l,c,p,_,w,v,x,j){var B=he();try{Pa(s,a,l,c,p,_,w,v,x,j)}catch(H){if(ce(B),H!==H+0)throw H;Me(1,0)}}function wx(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe){var xe=he();try{Zn(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe)}catch(De){if(ce(xe),De!==De+0)throw De;Me(1,0)}}function bx(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot){var Gt=he();try{ei(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot)}catch(_r){if(ce(Gt),_r!==_r+0)throw _r;Me(1,0)}}function Mx(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt){var _r=he();try{Nu(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt)}catch(jr){if(ce(_r),jr!==jr+0)throw jr;Me(1,0)}}function vx(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot){var Gt=he();try{cu(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot)}catch(_r){if(ce(Gt),_r!==_r+0)throw _r;Me(1,0)}}function xx(s,a,l,c,p,_,w,v,x,j,B,H,ee){var fe=he();try{ou(s,a,l,c,p,_,w,v,x,j,B,H,ee)}catch(xe){if(ce(fe),xe!==xe+0)throw xe;Me(1,0)}}function Tx(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt){var Ot=he();try{lu(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt)}catch(Gt){if(ce(Ot),Gt!==Gt+0)throw Gt;Me(1,0)}}function Cx(s,a,l,c,p,_,w,v){var x=he();try{uu(s,a,l,c,p,_,w,v)}catch(j){if(ce(x),j!==j+0)throw j;Me(1,0)}}function Ex(s,a,l,c,p,_,w,v,x,j,B){var H=he();try{du(s,a,l,c,p,_,w,v,x,j,B)}catch(ee){if(ce(H),ee!==ee+0)throw ee;Me(1,0)}}function Px(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt,_r){var jr=he();try{return pu(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe,De,We,nt,pt,Ot,Gt,_r)}catch(As){if(ce(jr),As!==As+0)throw As;Me(1,0)}}function Sx(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe){var De=he();try{ti(s,a,l,c,p,_,w,v,x,j,B,H,ee,fe,xe)}catch(We){if(ce(De),We!==We+0)throw We;Me(1,0)}}function kx(s,a,l,c,p,_,w){var v=he();try{hu(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function $x(s,a,l,c,p){var _=he();try{fu(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function Ix(s,a,l,c,p,_,w,v,x,j,B,H){var ee=he();try{mu(s,a,l,c,p,_,w,v,x,j,B,H)}catch(fe){if(ce(ee),fe!==fe+0)throw fe;Me(1,0)}}function Ax(s,a,l,c,p,_,w){var v=he();try{ri(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function Ox(s,a,l,c,p){var _=he();try{_u(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function Fx(s,a,l,c,p,_,w,v,x,j,B,H){var ee=he();try{gc(s,a,l,c,p,_,w,v,x,j,B,H)}catch(fe){if(ce(ee),fe!==fe+0)throw fe;Me(1,0)}}function Dx(s,a,l,c,p,_,w){var v=he();try{return wu(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function jx(s,a,l,c,p){var _=he();try{bu(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function Lx(s,a,l,c,p,_,w){var v=he();try{Mu(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function zx(s,a,l,c,p,_,w,v,x,j,B){var H=he();try{vu(s,a,l,c,p,_,w,v,x,j,B)}catch(ee){if(ce(H),ee!==ee+0)throw ee;Me(1,0)}}function Bx(s,a,l,c,p){var _=he();try{return xu(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function Rx(s,a,l,c,p,_,w,v){var x=he();try{Tu(s,a,l,c,p,_,w,v)}catch(j){if(ce(x),j!==j+0)throw j;Me(1,0)}}function Nx(s,a,l,c,p,_){var w=he();try{return Cu(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}function Vx(s,a,l,c){var p=he();try{Sa(s,a,l,c)}catch(_){if(ce(p),_!==_+0)throw _;Me(1,0)}}function Wx(s,a,l,c,p,_,w,v){var x=he();try{Eu(s,a,l,c,p,_,w,v)}catch(j){if(ce(x),j!==j+0)throw j;Me(1,0)}}function Ux(s,a,l,c,p,_,w){var v=he();try{Pu(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function Gx(s,a,l,c,p){var _=he();try{ka(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function Kx(s,a,l,c,p){var _=he();try{return yu(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function Hx(s,a,l,c){var p=he();try{return gu(s,a,l,c)}catch(_){if(ce(p),_!==_+0)throw _;Me(1,0)}}function qx(s,a,l,c){var p=he();try{return ii(s,a,l,c)}catch(_){if(ce(p),_!==_+0)throw _;Me(1,0)}}function Qx(s,a,l,c,p){var _=he();try{Su(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function Xx(s,a,l,c){var p=he();try{return ul(s,a,l,c)}catch(_){if(ce(p),_!==_+0)throw _;Me(1,0)}}function Jx(s,a,l,c,p,_){var w=he();try{return ku(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}function Yx(s,a,l,c,p){var _=he();try{return $u(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function Zx(s,a,l,c,p){var _=he();try{return Iu(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function e1(s,a,l,c,p,_,w){var v=he();try{return si(s,a,l,c,p,_,w)}catch(x){if(ce(v),x!==x+0)throw x;Me(1,0)}}function t1(s,a,l,c,p,_,w,v,x,j,B,H){var ee=he();try{Au(s,a,l,c,p,_,w,v,x,j,B,H)}catch(fe){if(ce(ee),fe!==fe+0)throw fe;Me(1,0)}}function r1(s,a,l,c,p){var _=he();try{return Fu(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function s1(s,a,l,c,p,_,w,v,x,j,B,H,ee){var fe=he();try{Ou(s,a,l,c,p,_,w,v,x,j,B,H,ee)}catch(xe){if(ce(fe),xe!==xe+0)throw xe;Me(1,0)}}function n1(s,a,l,c,p,_,w,v){var x=he();try{return Du(s,a,l,c,p,_,w,v)}catch(j){if(ce(x),j!==j+0)throw j;Me(1,0)}}function i1(s,a,l,c,p,_){var w=he();try{return ju(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}function a1(s,a,l,c,p){var _=he();try{return ni(s,a,l,c,p)}catch(w){if(ce(_),w!==w+0)throw w;Me(1,0)}}function o1(s){var a=he();try{return wl(s)}catch(l){if(ce(a),l!==l+0)throw l;return Me(1,0),0n}}function l1(s,a,l,c,p,_){var w=he();try{return yl(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}function c1(s,a,l,c,p,_){var w=he();try{return _l(s,a,l,c,p,_)}catch(v){if(ce(w),v!==v+0)throw v;Me(1,0)}}return n.stackSave=he,n.stackRestore=ce,n.stackAlloc=Cr,n.setValue=function(s,a,l="i8"){switch(l.endsWith("*")&&(l="*"),l){case"i1":case"i8":N()[s>>>0]=a;break;case"i16":ue()[s>>>1>>>0]=a;break;case"i32":we()[s>>>2>>>0]=a;break;case"i64":oe[s>>>3]=BigInt(a);break;case"float":Ze()[s>>>2>>>0]=a;break;case"double":et()[s>>>3>>>0]=a;break;case"*":ae()[s>>>2>>>0]=a;break;default:Ge(`invalid type for setValue: ${l}`)}},n.getValue=function(s,a="i8"){switch(a.endsWith("*")&&(a="*"),a){case"i1":case"i8":return N()[s>>>0];case"i16":return ue()[s>>>1>>>0];case"i32":return we()[s>>>2>>>0];case"i64":return oe[s>>>3];case"float":return Ze()[s>>>2>>>0];case"double":return et()[s>>>3>>>0];case"*":return ae()[s>>>2>>>0];default:Ge(`invalid type for getValue: ${a}`)}},n.UTF8ToString=Jt,n.stringToUTF8=qe,n.lengthBytesUTF8=Er,function s(){if(Fe>0)Ve=s;else{if(f)return i(n),void X();(a=>{for(;a.length>0;)a.shift()(n)})(vt),Fe>0?Ve=s:(n.calledRun=!0,Ce||(X(),i(n)))}}(),n.PTR_SIZE=4,n.webgpuInit=s=>{let a=new WeakMap,l=1,c=s,p,_;n.webgpuRegisterDevice=x=>{if(_!==void 0)throw new Error("another WebGPU EP inference session is being created.");if(x){let j=a.get(x);if(!j){let B=Fo(0),H=Pe.importJsDevice(x,B);j=[l++,B,H],a.set(x,j)}return p=x,_=j[0],j}return p=void 0,void(_=0)};let w=new Map;n.webgpuOnCreateSession=x=>{if(_===void 0)return;let j=_;if(_=void 0,x){let B=Ii(j);if(w.set(x,B),j===0){let H=p??Pe.getJsObject(B);c(H)}}p=void 0},n.webgpuOnReleaseSession=x=>{w.delete(x)};let v=Symbol("gpuBufferMetadata");n.webgpuRegisterBuffer=(x,j,B)=>{if(B)return x[v]=[B,NaN],B;{let H=x[v];if(H)return H[1]++,H[0];let ee=w.get(j);if(ee===void 0)throw new Error("Invalid session handle passed to webgpuRegisterBuffer");let fe=Pe.importJsBuffer(x,ee);return x[v]=[fe,1],fe}},n.webgpuUnregisterBuffer=x=>{let j=x[v];if(!j)throw new Error("Buffer is not registered");j[1]--,j[1]===0&&(Ai(j[0]),delete x[v])},n.webgpuGetBuffer=x=>Pe.getJsObject(x),n.webgpuCreateDownloader=(x,j,B)=>{let H=w.get(B);if(H===void 0)throw new Error("Invalid session handle passed to webgpuRegisterBuffer");let ee=x,fe=Pe.getJsObject(H),xe=j,De=16*Math.ceil(Number(xe)/16);return async()=>{let We={size:De,usage:9},nt=fe.createBuffer(We);try{let pt=fe.createCommandEncoder();return pt.copyBufferToBuffer(ee,0,nt,0,De),fe.queue.submit([pt.finish()]),await nt.mapAsync(GPUMapMode.READ),nt.getMappedRange().slice(0,xe)}finally{nt.destroy()}}},n.webgpuUploadExternalBuffer=(x,j)=>{let B=j.buffer,H=j.byteOffset,ee=j.byteLength,fe=16*Math.ceil(Number(ee)/16),xe=Pe.getJsObject(x);if(!p){let Ot=Ii(_);p=Pe.getJsObject(Ot)}let De={mappedAtCreation:!0,size:fe,usage:6},We=p.createBuffer(De),nt=We.getMappedRange();new Uint8Array(nt).set(new Uint8Array(B,H,ee)),We.unmap();let pt=p.createCommandEncoder();pt.copyBufferToBuffer(We,0,xe,0,fe),p.queue.submit([pt.finish()]),We.destroy()}},u}),k0=Of,T_=(r=(e=globalThis.self)==null?void 0:e.name)==null?void 0:r.startsWith("em-pthread"),T_&&Of()}),Ff,Pm,C_,fs,$0,Dh,E_,P_,Df,S_,jf,I0,Lf,A0,Hm=Ye(()=>{Km(),Ff=typeof location>"u"?void 0:location.origin,Pm=import.meta.url>"file:"&&import.meta.url<"file;",C_=()=>{{if(Pm){let e=URL;return new URL(new e("ort.bundle.min.mjs",import.meta.url).href,Ff).href}return import.meta.url}},fs=C_(),$0=()=>{if(fs&&!fs.startsWith("blob:"))return fs.substring(0,fs.lastIndexOf("/")+1)},Dh=(e,r)=>{try{let t=r??fs;return(t?new URL(e,t):new URL(e)).origin===Ff}catch{return!1}},E_=(e,r)=>{let t=r??fs;try{return(t?new URL(e,t):new URL(e)).href}catch{return}},P_=(e,r)=>`${r??"./"}${e}`,Df=async e=>{let r=await(await fetch(e,{credentials:"same-origin"})).blob();return URL.createObjectURL(r)},S_=async e=>(await import(e)).default,jf=(Q1(),wd(E0)).default,I0=async()=>{if(!fs)throw new Error("Failed to load proxy worker: cannot determine the script source URL.");if(Dh(fs))return[void 0,jf()];let e=await Df(fs);return[e,jf(e)]},Lf=(X1(),wd(S0)).default,A0=async(e,r,t)=>{if(!e&&!r&&Lf&&fs&&Dh(fs))return[void 0,Lf];{let i="ort-wasm-simd-threaded.jsep.mjs",o=e??E_(i,r),n=t&&o&&!Dh(o,r),u=n?await Df(o):o??P_(i,r);return[n?u:void 0,await S_(u)]}}}),zf,jh,sd,Bf,k_,$_,I_,qm,lr,yi=Ye(()=>{Hm(),jh=!1,sd=!1,Bf=!1,k_=()=>{if(typeof SharedArrayBuffer>"u")return!1;try{return typeof MessageChannel<"u"&&new MessageChannel().port1.postMessage(new SharedArrayBuffer(1)),WebAssembly.validate(new Uint8Array([0,97,115,109,1,0,0,0,1,4,1,96,0,0,3,2,1,0,5,4,1,3,1,1,10,11,1,9,0,65,0,254,16,2,0,26,11]))}catch{return!1}},$_=()=>{try{return WebAssembly.validate(new Uint8Array([0,97,115,109,1,0,0,0,1,4,1,96,0,0,3,2,1,0,10,30,1,28,0,65,0,253,15,253,12,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,253,186,1,26,11]))}catch{return!1}},I_=()=>{try{return WebAssembly.validate(new Uint8Array([0,97,115,109,1,0,0,0,1,5,1,96,0,1,123,3,2,1,0,10,19,1,17,0,65,1,253,15,65,2,253,15,65,3,253,15,253,147,2,11]))}catch{return!1}},qm=async e=>{if(jh)return Promise.resolve();if(sd)throw new Error("multiple calls to 'initializeWebAssembly()' detected.");if(Bf)throw new Error("previous call to 'initializeWebAssembly()' failed.");sd=!0;let r=e.initTimeout,t=e.numThreads;if(e.simd!==!1){if(e.simd==="relaxed"){if(!I_())throw new Error("Relaxed WebAssembly SIMD is not supported in the current environment.")}else if(!$_())throw new Error("WebAssembly SIMD is not supported in the current environment.")}let i=k_();t>1&&!i&&(typeof self<"u"&&!self.crossOriginIsolated&&console.warn("env.wasm.numThreads is set to "+t+", but this will not work unless you enable crossOriginIsolated mode. See https://web.dev/cross-origin-isolation-guide/ for more info."),console.warn("WebAssembly multi-threading is not supported in the current environment. Falling back to single-threading."),e.numThreads=t=1);let o=e.wasmPaths,n=typeof o=="string"?o:void 0,u=o==null?void 0:o.mjs,d=(u==null?void 0:u.href)??u,h=o==null?void 0:o.wasm,f=(h==null?void 0:h.href)??h,y=e.wasmBinary,[m,g]=await A0(d,n,t>1),T=!1,M=[];if(r>0&&M.push(new Promise(I=>{setTimeout(()=>{T=!0,I()},r)})),M.push(new Promise((I,z)=>{let E={numThreads:t};if(y)E.wasmBinary=y;else if(f||n)E.locateFile=C=>f??n+C;else if(d&&d.indexOf("blob:")!==0)E.locateFile=C=>new URL(C,d).href;else if(m){let C=$0();C&&(E.locateFile=D=>C+D)}g(E).then(C=>{sd=!1,jh=!0,zf=C,I(),m&&URL.revokeObjectURL(m)},C=>{sd=!1,Bf=!0,z(C)})})),await Promise.race(M),T)throw new Error(`WebAssembly backend initializing failed due to timeout: ${r}ms`)},lr=()=>{if(jh&&zf)return zf;throw new Error("WebAssembly is not initialized yet.")}}),zs,Yh,ir,Qm=Ye(()=>{yi(),zs=(e,r)=>{let t=lr(),i=t.lengthBytesUTF8(e)+1,o=t._malloc(i);return t.stringToUTF8(e,o,i),r.push(o),o},Yh=(e,r,t,i)=>{if(typeof e=="object"&&e!==null){if(t.has(e))throw new Error("Circular reference in options");t.add(e)}Object.entries(e).forEach(([o,n])=>{let u=r?r+o:o;if(typeof n=="object")Yh(n,u+".",t,i);else if(typeof n=="string"||typeof n=="number")i(u,n.toString());else if(typeof n=="boolean")i(u,n?"1":"0");else throw new Error(`Can't handle extra config type: ${typeof n}`)})},ir=e=>{let r=lr(),t=r.stackSave();try{let i=r.PTR_SIZE,o=r.stackAlloc(2*i);r._OrtGetLastError(o,o+i);let n=Number(r.getValue(o,i===4?"i32":"i64")),u=r.getValue(o+i,"*"),d=u?r.UTF8ToString(u):"";throw new Error(`${e} ERROR_CODE: ${n}, ERROR_MESSAGE: ${d}`)}finally{r.stackRestore(t)}}}),O0,J1=Ye(()=>{yi(),Qm(),O0=e=>{let r=lr(),t=0,i=[],o=e||{};try{if((e==null?void 0:e.logSeverityLevel)===void 0)o.logSeverityLevel=2;else if(typeof e.logSeverityLevel!="number"||!Number.isInteger(e.logSeverityLevel)||e.logSeverityLevel<0||e.logSeverityLevel>4)throw new Error(`log serverity level is not valid: ${e.logSeverityLevel}`);if((e==null?void 0:e.logVerbosityLevel)===void 0)o.logVerbosityLevel=0;else if(typeof e.logVerbosityLevel!="number"||!Number.isInteger(e.logVerbosityLevel))throw new Error(`log verbosity level is not valid: ${e.logVerbosityLevel}`);(e==null?void 0:e.terminate)===void 0&&(o.terminate=!1);let n=0;return(e==null?void 0:e.tag)!==void 0&&(n=zs(e.tag,i)),t=r._OrtCreateRunOptions(o.logSeverityLevel,o.logVerbosityLevel,!!o.terminate,n),t===0&&ir("Can't create run options."),(e==null?void 0:e.extra)!==void 0&&Yh(e.extra,"",new WeakSet,(u,d)=>{let h=zs(u,i),f=zs(d,i);r._OrtAddRunConfigEntry(t,h,f)!==0&&ir(`Can't set a run config entry: ${u} - ${d}.`)}),[t,i]}catch(n){throw t!==0&&r._OrtReleaseRunOptions(t),i.forEach(u=>r._free(u)),n}}}),A_,O_,F_,nd,D_,F0,Y1=Ye(()=>{yi(),Qm(),A_=e=>{switch(e){case"disabled":return 0;case"basic":return 1;case"extended":return 2;case"all":return 99;default:throw new Error(`unsupported graph optimization level: ${e}`)}},O_=e=>{switch(e){case"sequential":return 0;case"parallel":return 1;default:throw new Error(`unsupported execution mode: ${e}`)}},F_=e=>{e.extra||(e.extra={}),e.extra.session||(e.extra.session={});let r=e.extra.session;r.use_ort_model_bytes_directly||(r.use_ort_model_bytes_directly="1"),e.executionProviders&&e.executionProviders.some(t=>(typeof t=="string"?t:t.name)==="webgpu")&&(e.enableMemPattern=!1)},nd=(e,r,t,i)=>{let o=zs(r,i),n=zs(t,i);lr()._OrtAddSessionConfigEntry(e,o,n)!==0&&ir(`Can't set a session config entry: ${r} - ${t}.`)},D_=async(e,r,t)=>{for(let i of r){let o=typeof i=="string"?i:i.name,n=[];switch(o){case"webnn":if(o="WEBNN",typeof i!="string"){let y=i==null?void 0:i.deviceType;y&&nd(e,"deviceType",y,t)}break;case"webgpu":if(o="JS",typeof i!="string"){let y=i;if(y!=null&&y.preferredLayout){if(y.preferredLayout!=="NCHW"&&y.preferredLayout!=="NHWC")throw new Error(`preferredLayout must be either 'NCHW' or 'NHWC': ${y.preferredLayout}`);nd(e,"preferredLayout",y.preferredLayout,t)}}break;case"wasm":case"cpu":continue;default:throw new Error(`not supported execution provider: ${o}`)}let u=zs(o,t),d=n.length,h=0,f=0;if(d>0){h=lr()._malloc(d*lr().PTR_SIZE),t.push(h),f=lr()._malloc(d*lr().PTR_SIZE),t.push(f);for(let y=0;y{let r=lr(),t=0,i=[],o=e||{};F_(o);try{let n=A_(o.graphOptimizationLevel??"all"),u=O_(o.executionMode??"sequential"),d=typeof o.logId=="string"?zs(o.logId,i):0,h=o.logSeverityLevel??2;if(!Number.isInteger(h)||h<0||h>4)throw new Error(`log serverity level is not valid: ${h}`);let f=o.logVerbosityLevel??0;if(!Number.isInteger(f)||f<0||f>4)throw new Error(`log verbosity level is not valid: ${f}`);let y=typeof o.optimizedModelFilePath=="string"?zs(o.optimizedModelFilePath,i):0;if(t=r._OrtCreateSessionOptions(n,!!o.enableCpuMemArena,!!o.enableMemPattern,u,!!o.enableProfiling,0,d,h,f,y),t===0&&ir("Can't create session options."),o.executionProviders&&await D_(t,o.executionProviders,i),o.enableGraphCapture!==void 0){if(typeof o.enableGraphCapture!="boolean")throw new Error(`enableGraphCapture must be a boolean value: ${o.enableGraphCapture}`);nd(t,"enableGraphCapture",o.enableGraphCapture.toString(),i)}if(o.freeDimensionOverrides)for(let[m,g]of Object.entries(o.freeDimensionOverrides)){if(typeof m!="string")throw new Error(`free dimension override name must be a string: ${m}`);if(typeof g!="number"||!Number.isInteger(g)||g<0)throw new Error(`free dimension override value must be a non-negative integer: ${g}`);let T=zs(m,i);r._OrtAddFreeDimensionOverride(t,T,g)!==0&&ir(`Can't set a free dimension override: ${m} - ${g}.`)}return o.extra!==void 0&&Yh(o.extra,"",new WeakSet,(m,g)=>{nd(t,m,g,i)}),[t,i]}catch(n){throw t!==0&&r._OrtReleaseSessionOptions(t)!==0&&ir("Can't release session options."),i.forEach(u=>r._free(u)),n}}}),pi,fn,hi,of,Zh,Xm,Jm,Sm,$t=Ye(()=>{pi=e=>{switch(e){case"int8":return 3;case"uint8":return 2;case"bool":return 9;case"int16":return 5;case"uint16":return 4;case"int32":return 6;case"uint32":return 12;case"float16":return 10;case"float32":return 1;case"float64":return 11;case"string":return 8;case"int64":return 7;case"uint64":return 13;case"int4":return 22;case"uint4":return 21;default:throw new Error(`unsupported data type: ${e}`)}},fn=e=>{switch(e){case 3:return"int8";case 2:return"uint8";case 9:return"bool";case 5:return"int16";case 4:return"uint16";case 6:return"int32";case 12:return"uint32";case 10:return"float16";case 1:return"float32";case 11:return"float64";case 8:return"string";case 7:return"int64";case 13:return"uint64";case 22:return"int4";case 21:return"uint4";default:throw new Error(`unsupported data type: ${e}`)}},hi=(e,r)=>{let t=[-1,4,1,1,2,2,4,8,-1,1,2,8,4,8,-1,-1,-1,-1,-1,-1,-1,.5,.5][e],i=typeof r=="number"?r:r.reduce((o,n)=>o*n,1);return t>0?Math.ceil(i*t):void 0},of=e=>{switch(e){case"float16":return typeof Float16Array<"u"&&Float16Array.from?Float16Array:Uint16Array;case"float32":return Float32Array;case"uint8":return Uint8Array;case"int8":return Int8Array;case"uint16":return Uint16Array;case"int16":return Int16Array;case"int32":return Int32Array;case"bool":return Uint8Array;case"float64":return Float64Array;case"uint32":return Uint32Array;case"int64":return BigInt64Array;case"uint64":return BigUint64Array;default:throw new Error(`unsupported type: ${e}`)}},Zh=e=>{switch(e){case"verbose":return 0;case"info":return 1;case"warning":return 2;case"error":return 3;case"fatal":return 4;default:throw new Error(`unsupported logging level: ${e}`)}},Xm=e=>e==="float32"||e==="float16"||e==="int32"||e==="int64"||e==="uint32"||e==="uint8"||e==="bool"||e==="uint4"||e==="int4",Jm=e=>e==="float32"||e==="float16"||e==="int32"||e==="int64"||e==="uint32"||e==="uint64"||e==="int8"||e==="uint8"||e==="bool"||e==="uint4"||e==="int4",Sm=e=>{switch(e){case"none":return 0;case"cpu":return 1;case"cpu-pinned":return 2;case"texture":return 3;case"gpu-buffer":return 4;case"ml-tensor":return 5;default:throw new Error(`unsupported data location: ${e}`)}}}),Ym,D0=Ye(()=>{Km(),Ym=async e=>{if(typeof e=="string"){let r=await fetch(e);if(!r.ok)throw new Error(`failed to load external data file: ${e}`);let t=r.headers.get("Content-Length"),i=t?parseInt(t,10):0;if(i<1073741824)return new Uint8Array(await r.arrayBuffer());{if(!r.body)throw new Error(`failed to load external data file: ${e}, no response body.`);let o=r.body.getReader(),n;try{n=new ArrayBuffer(i)}catch(d){if(d instanceof RangeError){let h=Math.ceil(i/65536);n=new WebAssembly.Memory({initial:h,maximum:h}).buffer}else throw d}let u=0;for(;;){let{done:d,value:h}=await o.read();if(d)break;let f=h.byteLength;new Uint8Array(n,u,f).set(h),u+=f}return new Uint8Array(n,0,i)}}else return e instanceof Blob?new Uint8Array(await e.arrayBuffer()):e instanceof Uint8Array?e:new Uint8Array(e)}}),j_,L_,z_,B_,Zm,R_,Kt,mn=Ye(()=>{$t(),j_=["V","I","W","E","F"],L_=(e,r)=>{console.log(`[${j_[e]},${new Date().toISOString()}]${r}`)},Zm=(e,r)=>{z_=e,B_=r},R_=(e,r)=>{let t=Zh(e),i=Zh(z_);t>=i&&L_(t,typeof r=="function"?r():r)},Kt=(...e)=>{B_&&R_(...e)}}),N_,Na,je,ef,j0,L0,z0,Lt=Ye(()=>{N_=class{static calcMatMulShape(e,r){return e[1]!==r[0]?void 0:[e[0],r[1]]}},Na=class{static calcShape(e,r,t=!1){let i=e.length,o=r.length;if(i===0)return r;if(o===0)return e;let n=Math.max(e.length,r.length),u=new Array(n);if(t){if(i<2||o<2)return;let d=N_.calcMatMulShape([e[i-2],e[i-1]],[r[o-2],r[o-1]]);if(d===void 0)return;[u[n-2],u[n-1]]=d}for(let d=t?3:1;d<=n;d++){let h=i-d<0?1:e[i-d],f=o-d<0?1:r[o-d];if(h!==f&&h>1&&f>1)return;let y=Math.max(h,f);if(h&&f)u[n-d]=Math.max(h,f);else{if(y>1)return;u[n-d]=0}}return u}static isValidBroadcast(e,r){let t=e.length,i=r.length;if(t>i)return!1;for(let o=1;o<=t;o++)if(e[t-o]!==1&&e[t-o]!==r[i-o])return!1;return!0}},je=class Qh{static size(r){return Qh.getSizeFromDimensionRange(r,0,r.length)}static convertShape(r,t=4){let i=r.length;if(i===0)return[];let o=new Array(i),n=i-1;for(;n>=0;){if(r[n]%t===0){o[n]=r[n]/t;break}if(t%r[n]!==0)throw new Error("cannot convert shape");o[n]=1,t/=r[n],n--}for(n--;n>=0;n--)o[n]=r[n];return o}static sizeFromDimension(r,t){if(t<0||t>r.length)throw new Error(`invalid dimension of ${t} for sizeFromDimension as Tensor has ${r.length} dimensions.`);return Qh.getSizeFromDimensionRange(r,t,r.length)}static sizeToDimension(r,t){if(t<0||t>r.length)throw new Error(`invalid dimension of ${t} for sizeToDimension as Tensor has ${r.length} dimensions.`);return Qh.getSizeFromDimensionRange(r,0,t)}static getSizeFromDimensionRange(r,t,i){let o=1;for(let n=t;n=0;--o)i[o]=i[o+1]*r[o+1];return i}static normalizeAxis(r,t){if(r<-t&&r>=t)throw new Error("unsupported axis for this operation.");return r<0?r+t:r}static normalizeAxes(r,t){return r.map(i=>this.normalizeAxis(i,t??r.length))}static sortBasedOnPerm(r,t){return t?t.map(i=>r[i]):r.slice().reverse()}static padShape(r,t){let i=r.length;return r.map((o,n)=>o+t[n]+t[n+i])}static areEqual(r,t){return r.length!==t.length?!1:r.every((i,o)=>i===t[o])}},ef=class fd{static adjustPoolAttributes(r,t,i,o,n,u){if(!r&&i.length!==t.length-2)throw new Error("length of specified kernel shapes should be 2 less than length of input dimensions");if(r)for(let d=0;d=i.length?i.push(t[d+2]):i[d]=t[d+2];for(let d=0;d=i[d]||u[d+i.length]>=i[d])throw new Error("pads should be smaller than kernel")}}static adjustPadsBasedOnAutoPad(r,t,i,o,n,u,d){if(d){if(n.length!==2*(r.length-2))throw new Error("length of pads should be twice the length of data dimensions");if(t.length!==r.length-2)throw new Error("length of strides should be the length of data dimensions");if(o.length!==r.length-2)throw new Error("length of kernel shapes should be the length of data dimensions");for(let h=0;h{$t(),e_=(e,r)=>new(of(r))(e)}),Rf,km,Nf,V_,Vf,W_,Wf,Uf,Gf,U_,R0,Z1=Ye(()=>{$t(),mn(),Rf=new Map([["float32",32],["float16",16],["int32",32],["uint32",32],["int64",64],["uint64",64],["int8",8],["uint8",8],["int4",4],["uint4",4]]),km=(e,r)=>{if(r==="int32")return e;let t=Rf.get(r);if(!t)throw new Error(`WebNN backend does not support data type: ${r}`);let i=t/8;if(e.byteLength%i!==0)throw new Error(`Invalid Uint8Array length - must be a multiple of ${i}.`);let o=e.byteLength/i,n=new(of(r))(e.buffer,e.byteOffset,o);switch(r){case"int64":case"uint64":{let u=new Int32Array(o);for(let d=0;d2147483647n||h<-2147483648n)throw new Error("Can not convert int64 data to int32 - value out of range.");u[d]=Number(h)}return new Uint8Array(u.buffer)}case"int8":case"uint8":case"uint32":{if(r==="uint32"&&n.some(d=>d>2147483647))throw new Error("Can not convert uint32 data to int32 - value out of range.");let u=Int32Array.from(n,Number);return new Uint8Array(u.buffer)}default:throw new Error(`Unsupported data conversion from ${r} to 'int32'`)}},Nf=(e,r)=>{if(r==="int32")return e;if(e.byteLength%4!==0)throw new Error("Invalid Uint8Array length - must be a multiple of 4 (int32).");let t=e.byteLength/4,i=new Int32Array(e.buffer,e.byteOffset,t);switch(r){case"int64":{let o=BigInt64Array.from(i,BigInt);return new Uint8Array(o.buffer)}case"uint64":{if(i.some(n=>n<0))throw new Error("Can not convert int32 data to uin64 - negative value found.");let o=BigUint64Array.from(i,BigInt);return new Uint8Array(o.buffer)}case"int8":{if(i.some(n=>n<-128||n>127))throw new Error("Can not convert int32 data to int8 - value out of range.");let o=Int8Array.from(i,Number);return new Uint8Array(o.buffer)}case"uint8":{if(i.some(o=>o<0||o>255))throw new Error("Can not convert int32 data to uint8 - value out of range.");return Uint8Array.from(i,Number)}case"uint32":{if(i.some(n=>n<0))throw new Error("Can not convert int32 data to uint32 - negative value found.");let o=Uint32Array.from(i,Number);return new Uint8Array(o.buffer)}default:throw new Error(`Unsupported data conversion from 'int32' to ${r}`)}},V_=1,Vf=()=>V_++,W_=new Map([["int8","int32"],["uint8","int32"],["uint32","int32"],["int64","int32"]]),Wf=(e,r)=>{let t=Rf.get(e);if(!t)throw new Error(`WebNN backend does not support data type: ${e}`);return r.length>0?Math.ceil(r.reduce((i,o)=>i*o)*t/8):0},Uf=class{constructor(e){this.isDataConverted=!1;let{sessionId:r,context:t,tensor:i,dataType:o,shape:n,fallbackDataType:u}=e;this.sessionId=r,this.mlContext=t,this.mlTensor=i,this.dataType=o,this.tensorShape=n,this.fallbackDataType=u}get tensor(){return this.mlTensor}get type(){return this.dataType}get fallbackType(){return this.fallbackDataType}get shape(){return this.tensorShape}get byteLength(){return Wf(this.dataType,this.tensorShape)}destroy(){Kt("verbose",()=>"[WebNN] TensorWrapper.destroy"),this.mlTensor.destroy()}write(e){this.mlContext.writeTensor(this.mlTensor,e)}async read(e){if(this.fallbackDataType){let r=await this.mlContext.readTensor(this.mlTensor),t=Nf(new Uint8Array(r),this.dataType);if(e){(e instanceof ArrayBuffer?new Uint8Array(e):new Uint8Array(e.buffer,e.byteOffset,e.byteLength)).set(t);return}else return t.buffer}else return e?this.mlContext.readTensor(this.mlTensor,e):this.mlContext.readTensor(this.mlTensor)}canReuseTensor(e,r,t){return this.mlContext===e&&this.dataType===r&&this.tensorShape.length===t.length&&this.tensorShape.every((i,o)=>i===t[o])}setIsDataConverted(e){this.isDataConverted=e}},Gf=class{constructor(e,r){this.tensorManager=e,this.wrapper=r}get tensorWrapper(){return this.wrapper}releaseTensor(){this.tensorWrapper&&(this.tensorManager.releaseTensor(this.tensorWrapper),this.wrapper=void 0)}async ensureTensor(e,r,t,i){let o=this.tensorManager.getMLContext(e),n;if(!o.opSupportLimits().input.dataTypes.includes(r)){if(n=W_.get(r),!n||!o.opSupportLimits().input.dataTypes.includes(n))throw new Error(`WebNN backend does not support data type: ${r}`);Kt("verbose",()=>`[WebNN] TensorIdTracker.ensureTensor: fallback dataType from ${r} to ${n}`)}if(this.wrapper){if(this.wrapper.canReuseTensor(o,r,t))return this.wrapper.tensor;if(i){if(this.wrapper.byteLength!==Wf(r,t))throw new Error("Unable to copy data to tensor with different size.");this.activeUpload=new Uint8Array(await this.wrapper.read())}this.tensorManager.releaseTensor(this.wrapper)}let u=typeof MLTensorUsage>"u"?void 0:MLTensorUsage.READ|MLTensorUsage.WRITE;return this.wrapper=await this.tensorManager.getCachedTensor(e,r,t,u,!0,!0,n),i&&this.activeUpload&&(this.wrapper.write(this.activeUpload),this.activeUpload=void 0),this.wrapper.tensor}upload(e){let r=e;if(this.wrapper){if(this.wrapper.fallbackType)if(this.wrapper.fallbackType==="int32")r=km(e,this.wrapper.type),this.wrapper.setIsDataConverted(!0);else throw new Error(`Unsupported fallback data type: ${this.wrapper.fallbackType}`);if(e.byteLength===this.wrapper.byteLength){this.wrapper.write(r);return}else Kt("verbose",()=>"Data size does not match tensor size. Releasing tensor."),this.releaseTensor()}this.activeUpload?this.activeUpload.set(r):this.activeUpload=new Uint8Array(r)}async download(e){var r,t;if(this.activeUpload){let i=(r=this.wrapper)!=null&&r.isDataConverted?Nf(this.activeUpload,(t=this.wrapper)==null?void 0:t.type):this.activeUpload;if(e){e instanceof ArrayBuffer?new Uint8Array(e).set(i):new Uint8Array(e.buffer,e.byteOffset,e.byteLength).set(i);return}else return i.buffer}if(!this.wrapper)throw new Error("Tensor has not been created.");return e?this.wrapper.read(e):this.wrapper.read()}},U_=class{constructor(e){this.backend=e,this.tensorTrackersById=new Map,this.freeTensors=[],this.externalTensors=new Set}getMLContext(e){let r=this.backend.getMLContext(e);if(!r)throw new Error("MLContext not found for session.");return r}reserveTensorId(){let e=Vf();return this.tensorTrackersById.set(e,new Gf(this)),e}releaseTensorId(e){let r=this.tensorTrackersById.get(e);r&&(this.tensorTrackersById.delete(e),r.tensorWrapper&&this.releaseTensor(r.tensorWrapper))}async ensureTensor(e,r,t,i,o){Kt("verbose",()=>`[WebNN] TensorManager.ensureTensor {tensorId: ${r}, dataType: ${t}, shape: ${i}, copyOld: ${o}}`);let n=this.tensorTrackersById.get(r);if(!n)throw new Error("Tensor not found.");return n.ensureTensor(e,t,i,o)}upload(e,r){let t=this.tensorTrackersById.get(e);if(!t)throw new Error("Tensor not found.");t.upload(r)}async download(e,r){Kt("verbose",()=>`[WebNN] TensorManager.download {tensorId: ${e}, dstBuffer: ${r==null?void 0:r.byteLength}}`);let t=this.tensorTrackersById.get(e);if(!t)throw new Error("Tensor not found.");return t.download(r)}releaseTensorsForSession(e){for(let r of this.freeTensors)r.sessionId===e&&r.destroy();this.freeTensors=this.freeTensors.filter(r=>r.sessionId!==e)}registerTensor(e,r,t,i){let o=this.getMLContext(e),n=Vf(),u=new Uf({sessionId:e,context:o,tensor:r,dataType:t,shape:i});return this.tensorTrackersById.set(n,new Gf(this,u)),this.externalTensors.add(u),n}async getCachedTensor(e,r,t,i,o,n,u){let d=this.getMLContext(e);for(let[f,y]of this.freeTensors.entries())if(y.canReuseTensor(d,r,t)){Kt("verbose",()=>`[WebNN] Reusing tensor {dataType: ${r}, ${u?`fallbackDataType: ${u},`:""} shape: ${t}`);let m=this.freeTensors.splice(f,1)[0];return m.sessionId=e,m}Kt("verbose",()=>`[WebNN] MLContext.createTensor {dataType: ${r}, ${u?`fallbackDataType: ${u},`:""} shape: ${t}}`);let h=await d.createTensor({dataType:u??r,shape:t,dimensions:t,usage:i,writable:o,readable:n});return new Uf({sessionId:e,context:d,tensor:h,dataType:r,shape:t,fallbackDataType:u})}releaseTensor(e){this.externalTensors.has(e)&&this.externalTensors.delete(e),this.freeTensors.push(e)}},R0=(...e)=>new U_(...e)}),id,G_,N0,eT=Ye(()=>{$t(),yi(),B0(),Z1(),mn(),id=new Map([[1,"float32"],[10,"float16"],[6,"int32"],[12,"uint32"],[7,"int64"],[13,"uint64"],[22,"int4"],[21,"uint4"],[3,"int8"],[2,"uint8"],[9,"uint8"]]),G_=(e,r)=>{if(e===r)return!0;if(e===void 0||r===void 0)return!1;let t=Object.keys(e).sort(),i=Object.keys(r).sort();return t.length===i.length&&t.every((o,n)=>o===i[n]&&e[o]===r[o])},N0=class{constructor(e){this.tensorManager=R0(this),this.mlContextBySessionId=new Map,this.sessionIdsByMLContext=new Map,this.mlContextCache=[],this.sessionGraphInputs=new Map,this.sessionGraphOutputs=new Map,this.temporaryGraphInputs=[],this.temporaryGraphOutputs=[],this.temporarySessionTensorIds=new Map,Zm(e.logLevel,!!e.debug)}get currentSessionId(){if(this.activeSessionId===void 0)throw new Error("No active session");return this.activeSessionId}onRunStart(e){Kt("verbose",()=>`[WebNN] onRunStart {sessionId: ${e}}`),this.activeSessionId=e}onRunEnd(e){Kt("verbose",()=>`[WebNN] onRunEnd {sessionId: ${e}}`);let r=this.temporarySessionTensorIds.get(e);if(r){for(let t of r)Kt("verbose",()=>`[WebNN] releasing temporary tensor {tensorId: ${t}}`),this.tensorManager.releaseTensorId(t);this.temporarySessionTensorIds.delete(e),this.activeSessionId=void 0}}async createMLContext(e){if(e instanceof GPUDevice){let t=this.mlContextCache.findIndex(i=>i.gpuDevice===e);if(t!==-1)return this.mlContextCache[t].mlContext;{let i=await navigator.ml.createContext(e);return this.mlContextCache.push({gpuDevice:e,mlContext:i}),i}}else if(e===void 0){let t=this.mlContextCache.findIndex(i=>i.options===void 0&&i.gpuDevice===void 0);if(t!==-1)return this.mlContextCache[t].mlContext;{let i=await navigator.ml.createContext();return this.mlContextCache.push({mlContext:i}),i}}let r=this.mlContextCache.findIndex(t=>G_(t.options,e));if(r!==-1)return this.mlContextCache[r].mlContext;{let t=await navigator.ml.createContext(e);return this.mlContextCache.push({options:e,mlContext:t}),t}}registerMLContext(e,r){this.mlContextBySessionId.set(e,r);let t=this.sessionIdsByMLContext.get(r);t||(t=new Set,this.sessionIdsByMLContext.set(r,t)),t.add(e),this.temporaryGraphInputs.length>0&&(this.sessionGraphInputs.set(e,this.temporaryGraphInputs),this.temporaryGraphInputs=[]),this.temporaryGraphOutputs.length>0&&(this.sessionGraphOutputs.set(e,this.temporaryGraphOutputs),this.temporaryGraphOutputs=[])}onReleaseSession(e){this.sessionGraphInputs.delete(e),this.sessionGraphOutputs.delete(e);let r=this.mlContextBySessionId.get(e);if(!r)return;this.tensorManager.releaseTensorsForSession(e),this.mlContextBySessionId.delete(e);let t=this.sessionIdsByMLContext.get(r);if(t.delete(e),t.size===0){this.sessionIdsByMLContext.delete(r);let i=this.mlContextCache.findIndex(o=>o.mlContext===r);i!==-1&&this.mlContextCache.splice(i,1)}}getMLContext(e){return this.mlContextBySessionId.get(e)}reserveTensorId(){return this.tensorManager.reserveTensorId()}releaseTensorId(e){Kt("verbose",()=>`[WebNN] releaseTensorId {tensorId: ${e}}`),this.tensorManager.releaseTensorId(e)}async ensureTensor(e,r,t,i,o){let n=id.get(t);if(!n)throw new Error(`Unsupported ONNX data type: ${t}`);return this.tensorManager.ensureTensor(e??this.currentSessionId,r,n,i,o)}async createTemporaryTensor(e,r,t){Kt("verbose",()=>`[WebNN] createTemporaryTensor {onnxDataType: ${r}, shape: ${t}}`);let i=id.get(r);if(!i)throw new Error(`Unsupported ONNX data type: ${r}`);let o=this.tensorManager.reserveTensorId();await this.tensorManager.ensureTensor(e,o,i,t,!1);let n=this.temporarySessionTensorIds.get(e);return n?n.push(o):this.temporarySessionTensorIds.set(e,[o]),o}uploadTensor(e,r){if(!lr().shouldTransferToMLTensor)throw new Error("Trying to upload to a MLTensor while shouldTransferToMLTensor is false");Kt("verbose",()=>`[WebNN] uploadTensor {tensorId: ${e}, data: ${r.byteLength}}`),this.tensorManager.upload(e,r)}async downloadTensor(e,r){return this.tensorManager.download(e,r)}createMLTensorDownloader(e,r){return async()=>{let t=await this.tensorManager.download(e);return e_(t,r)}}registerMLTensor(e,r,t,i){let o=id.get(t);if(!o)throw new Error(`Unsupported ONNX data type: ${t}`);let n=this.tensorManager.registerTensor(e,r,o,i);return Kt("verbose",()=>`[WebNN] registerMLTensor {tensor: ${r}, dataType: ${o}, dimensions: ${i}} -> {tensorId: ${n}}`),n}registerMLConstant(e,r,t,i,o,n,u=!1){if(!n)throw new Error("External mounted files are not available.");let d=e;e.startsWith("./")&&(d=e.substring(2));let h=n.get(d);if(!h)throw new Error(`File with name ${d} not found in preloaded files.`);if(r+t>h.byteLength)throw new Error("Out of bounds: data offset and length exceed the external file data size.");let f=h.slice(r,r+t).buffer,y;switch(o.dataType){case"float32":y=new Float32Array(f);break;case"float16":y=typeof Float16Array<"u"&&Float16Array.from?new Float16Array(f):new Uint16Array(f);break;case"int32":y=new Int32Array(f);break;case"uint32":y=new Uint32Array(f);break;case"int64":if(u){let m=km(new Uint8Array(f),"int64");y=new Int32Array(m.buffer),o.dataType="int32"}else y=new BigInt64Array(f);break;case"uint64":y=new BigUint64Array(f);break;case"int8":y=new Int8Array(f);break;case"int4":case"uint4":case"uint8":y=new Uint8Array(f);break;default:throw new Error(`Unsupported data type: ${o.dataType} in creating WebNN Constant from external data.`)}return Kt("verbose",()=>`[WebNN] registerMLConstant {dataType: ${o.dataType}, shape: ${o.shape}}} ${u?"(Note: it was int64 data type and registered to int32 as workaround)":""}`),i.constant(o,y)}registerGraphInput(e){this.temporaryGraphInputs.push(e)}registerGraphOutput(e){this.temporaryGraphOutputs.push(e)}isGraphInput(e,r){let t=this.sessionGraphInputs.get(e);return t?t.includes(r):!1}isGraphOutput(e,r){let t=this.sessionGraphOutputs.get(e);return t?t.includes(r):!1}isGraphInputOutputTypeSupported(e,r,t=!0){let i=this.mlContextBySessionId.get(e),o=id.get(pi(r));return typeof o>"u"?!1:t?!!(i!=null&&i.opSupportLimits().input.dataTypes.includes(o)):!!(i!=null&&i.opSupportLimits().output.dataTypes.includes(o))}flush(){}}}),t_=Ye(()=>{}),Kf,Lh,zh,K_,H_,Hf,$m,q_,V0,tT=Ye(()=>{mn(),t_(),Kf=new Map([[64,250],[128,200],[256,200],[512,200],[2048,230],[4096,200],[8192,50],[16384,50],[32768,50],[65536,50],[131072,50],[262144,50],[524288,50],[1048576,50],[2097152,30],[4194304,20],[8388608,10],[12582912,10],[16777216,10],[26214400,15],[33554432,22],[44236800,2],[58982400,6],[67108864,6],[134217728,6],[167772160,6]]),Lh=[],zh=e=>Math.ceil(Number(e)/16)*16,K_=e=>{for(let r=0;rH_++,$m=async(e,r,t,i)=>{let o=zh(t),n=e.device.createBuffer({size:o,usage:GPUBufferUsage.COPY_DST|GPUBufferUsage.MAP_READ});try{let u=e.getCommandEncoder();e.endComputePass(),u.copyBufferToBuffer(r,0,n,0,o),e.flush(),await n.mapAsync(GPUMapMode.READ);let d=n.getMappedRange();if(i){let h=i();return h.set(new Uint8Array(d,0,t)),h}else return new Uint8Array(d.slice(0,t))}finally{n.destroy()}},q_=class{constructor(e){this.backend=e,this.storageCache=new Map,this.freeBuffers=new Map,this.freeUniformBuffers=new Map,this.buffersPending=[],this.capturedPendingBuffers=new Map;for(let[r]of Kf)Lh.push(r),this.freeBuffers.set(r,[]),this.freeUniformBuffers.set(r,[]);this.sessionCount=0}upload(e,r){let t=r.buffer,i=r.byteOffset,o=r.byteLength,n=zh(o),u=this.storageCache.get(e);if(!u)throw new Error("gpu data for uploading does not exist");if(Number(u.originalSize)!==o)throw new Error(`inconsistent data size. gpu data size=${u.originalSize}, data size=${o}`);let d=this.backend.device.createBuffer({mappedAtCreation:!0,size:n,usage:GPUBufferUsage.MAP_WRITE|GPUBufferUsage.COPY_SRC}),h=d.getMappedRange();new Uint8Array(h).set(new Uint8Array(t,i,o)),d.unmap();let f=this.backend.device.createCommandEncoder();f.copyBufferToBuffer(d,0,u.gpuData.buffer,0,n),this.backend.device.queue.submit([f.finish()]),d.destroy(),Kt("verbose",()=>`[WebGPU] GpuDataManager.upload(id=${e})`)}memcpy(e,r){let t=this.storageCache.get(e);if(!t)throw new Error("source gpu data for memcpy does not exist");let i=this.storageCache.get(r);if(!i)throw new Error("destination gpu data for memcpy does not exist");if(t.originalSize!==i.originalSize)throw new Error("inconsistent source and destination gpu data size");let o=zh(t.originalSize),n=this.backend.getCommandEncoder();this.backend.endComputePass(),n.copyBufferToBuffer(t.gpuData.buffer,0,i.gpuData.buffer,0,o)}registerExternalBuffer(e,r,t){let i;if(t){if(i=t[0],e===t[1])return Kt("verbose",()=>`[WebGPU] GpuDataManager.registerExternalBuffer(size=${r}) => id=${i}, buffer is the same, skip.`),i;if(this.backend.capturedCommandList.has(this.backend.currentSessionId))throw new Error(`Registering a different external buffer under graph capture mode is not supported yet. + Please use the previous external buffer!`)}else i=Hf();return this.storageCache.set(i,{gpuData:{id:i,type:0,buffer:e},originalSize:r}),Kt("verbose",()=>`[WebGPU] GpuDataManager.registerExternalBuffer(size=${r}) => id=${i}, registered.`),i}unregisterExternalBuffer(e){e!==void 0&&(this.storageCache.delete(e),Kt("verbose",()=>`[WebGPU] GpuDataManager.unregisterExternalBuffer() => id=${e}`))}create(e,r=GPUBufferUsage.STORAGE|GPUBufferUsage.COPY_SRC|GPUBufferUsage.COPY_DST){let t=K_(e),i,o=(r&GPUBufferUsage.STORAGE)===GPUBufferUsage.STORAGE,n=(r&GPUBufferUsage.UNIFORM)===GPUBufferUsage.UNIFORM;if(o||n){let d=(o?this.freeBuffers:this.freeUniformBuffers).get(t);d?d.length>0?i=d.pop():i=this.backend.device.createBuffer({size:t,usage:r}):i=this.backend.device.createBuffer({size:t,usage:r})}else i=this.backend.device.createBuffer({size:t,usage:r});let u={id:Hf(),type:0,buffer:i};return this.storageCache.set(u.id,{gpuData:u,originalSize:Number(e)}),Kt("verbose",()=>`[WebGPU] GpuDataManager.create(size=${e}) => id=${u.id}`),u}get(e){var r;return(r=this.storageCache.get(e))==null?void 0:r.gpuData}release(e){let r=typeof e=="bigint"?Number(e):e,t=this.storageCache.get(r);if(!t){if(this.storageCache.size===0)return 0;throw new Error("releasing data does not exist")}return Kt("verbose",()=>`[WebGPU] GpuDataManager.release(id=${r}), gpuDataId=${t.gpuData.id}`),this.storageCache.delete(r),this.buffersPending.push(t.gpuData.buffer),t.originalSize}async download(e,r){let t=this.storageCache.get(Number(e));if(!t)throw new Error("data does not exist");await $m(this.backend,t.gpuData.buffer,t.originalSize,r)}refreshPendingBuffers(){if(this.buffersPending.length!==0)if(this.backend.sessionStatus==="default"){for(let e of this.buffersPending){let r=Kf.get(e.size);if((e.usage&GPUBufferUsage.STORAGE)===GPUBufferUsage.STORAGE){let t=this.freeBuffers.get(e.size)||[];r===void 0||t.length>=r?e.destroy():t.push(e)}else if((e.usage&GPUBufferUsage.UNIFORM)===GPUBufferUsage.UNIFORM){let t=this.freeUniformBuffers.get(e.size)||[];r===void 0||t.length>=r?e.destroy():t.push(e)}else e.destroy()}this.buffersPending=[]}else{let e=this.capturedPendingBuffers.get(this.backend.currentSessionId);e||(e=[],this.capturedPendingBuffers.set(this.backend.currentSessionId,e));for(let r of this.buffersPending)e.push(r);this.buffersPending=[]}}dispose(){this.freeBuffers.forEach(e=>{e.forEach(r=>{r.destroy()})}),this.freeUniformBuffers.forEach(e=>{e.forEach(r=>{r.destroy()})}),this.storageCache.forEach(e=>{e.gpuData.buffer.destroy()}),this.capturedPendingBuffers.forEach(e=>{e.forEach(r=>{r.destroy()})}),this.storageCache=new Map,this.freeBuffers=new Map,this.freeUniformBuffers=new Map,this.capturedPendingBuffers=new Map}onCreateSession(){this.sessionCount+=1}onReleaseSession(e){let r=this.capturedPendingBuffers.get(e);r&&(r.forEach(t=>{t.destroy()}),this.capturedPendingBuffers.delete(e)),this.sessionCount-=1,this.sessionCount===0&&(Kt("warning",()=>"[WebGPU] Clearing webgpu buffer cache"),this.storageCache.forEach(t=>{t.gpuData.buffer.destroy()}),this.storageCache=new Map)}},V0=(...e)=>new q_(...e)}),Q_,Zt,br=Ye(()=>{Q_=class{constructor(e){Object.assign(this,e)}get cacheKey(){return this.key||(this.key=Object.getOwnPropertyNames(this).sort().map(e=>`${this[e]}`).join(";")),this.key}},Zt=e=>new Q_(e)}),Va,Bh,Lr,Qr,Et,gr,Im,Ba,An,Mt,ad,Ne,gt,W0,r_,X_,U0,zt=Ye(()=>{$t(),Lt(),Va=64,Bh=(e,r)=>{if(r===3)throw new Error("vec3 has same alignment as vec4, use vec4 instead");switch(Number(e)){case 10:return r>1?`vec${r}`:"f16";case 1:return r>1?`vec${r}`:"f32";case 6:return r>1?`vec${r}`:"i32";case 12:return r>1?`vec${r}`:"u32";case 7:if(r>1)throw new Error("currently not supported vecX of uint64 yet");return["vec2","i32"];case 13:if(r>1)throw new Error("currently not supported vecX of uint64 yet");return["vec2","u32"];case 9:if(r!==4)throw new Error("bool must be vec4");return["u32","vec4"];case 22:return"i32";case 21:return"u32";default:throw new Error(`Unknown data type: ${e}`)}},Lr=(e,r=1)=>{let t=Bh(e,r);return typeof t=="string"?t:t[0]},Qr=(e,r=1)=>{let t=Bh(e,r);return typeof t=="string"?t:t[1]},Et=(...e)=>{let r=[];return e.forEach(t=>{t.length!==0&&r.push({type:12,data:t},{type:12,data:je.computeStrides(t)})}),r},gr=e=>e%4===0?4:e%2===0?2:1,Im=(e="f32",r,t="0")=>!r||r===1?`${e}(${t})`:`vec${r}<${e}>(${t})`,Ba=(e,r,t)=>e==="f32"?t:r===1?`f32(${t})`:`vec${r}(${t})`,An=(e,r)=>r===4?`(${e}.x + ${e}.y + ${e}.z + ${e}.w)`:r===2?`(${e}.x + ${e}.y)`:r===3?`(${e}.x + ${e}.y + ${e}.z)`:e,Mt=(e,r,t,i)=>e.startsWith("uniforms.")&&t>4?typeof r=="string"?i==="f16"?`${e}[(${r}) / 8][(${r}) % 8 / 4][(${r}) % 8 % 4]`:`${e}[(${r}) / 4][(${r}) % 4]`:i==="f16"?`${e}[${Math.floor(r/8)}][${Math.floor(r%8/4)}][${r%8%4}]`:`${e}[${Math.floor(r/4)}][${r%4}]`:t>1?`${e}[${r}]`:e,ad=(e,r,t,i,o)=>{let n=typeof t=="number",u=n?t:t.length,d=[...new Array(u).keys()],h=u<2?"u32":u<=4?`vec${u}`:`array`,f=Bh(r,o),y=typeof f=="string"?f:f[1],m=typeof f=="string"?f:f[0],g={indices:h,value:y,storage:m,tensor:r},T=q=>typeof q=="string"?q:`${q}u`,M={offsetToIndices:!1,indicesToOffset:!1,broadcastedIndicesToOffset:!1,set:!1,setByIndices:!1,get:!1,getByIndices:!1},I=n?"uniforms.":"",z=`${I}${e}_shape`,E=`${I}${e}_strides`,C="";for(let q=0;q ${g.indices} { + var indices: ${g.indices}; + var current = offset; + ${C} + return indices; + }`,A=q=>(M.offsetToIndices=!0,u<2?q:`o2i_${e}(${q})`),$=[];if(u>=2)for(let q=u-1;q>=0;q--)$.push(`${Mt(E,q,u)} * (indices[${q}])`);let P=u<2?"":` + fn i2o_${e}(indices: ${g.indices}) -> u32 { + return ${$.join("+")}; + }`,k=q=>(M.indicesToOffset=!0,u<2?q:`i2o_${e}(${q})`),O=(...q)=>u===0?"0u":`${g.indices}(${q.map(T).join(",")})`,R=(q,N)=>u<2?`${q}`:`${Mt(q,N,u)}`,U=(q,N,Q)=>u<2?`${q}=${Q};`:`${Mt(q,N,u)}=${Q};`,te={},se=(q,N)=>{M.broadcastedIndicesToOffset=!0;let Q=`${N.name}broadcastedIndicesTo${e}Offset`;if(Q in te)return`${Q}(${q})`;let ue=[];for(let Ae=u-1;Ae>=0;Ae--){let we=N.indicesGet("outputIndices",Ae+N.rank-u);ue.push(`${R(E,Ae)} * (${we} % ${R(z,Ae)})`)}return te[Q]=`fn ${Q}(outputIndices: ${N.type.indices}) -> u32 { + return ${ue.length>0?ue.join("+"):"0u"}; + }`,`${Q}(${q})`},K=(q,N)=>(()=>{if(g.storage===g.value)return`${e}[${q}]=${N};`;if(g.storage==="vec2"&&g.value==="i32")return`${e}[${q}]=vec2(u32(${N}), select(0u, 0xFFFFFFFFu, ${N} < 0));`;if(g.storage==="vec2"&&g.value==="u32")return`${e}[${q}]=vec2(u32(${N}), 0u);`;if(g.storage==="u32"&&g.value==="vec4")return`${e}[${q}]=dot(vec4(0x1, 0x100, 0x10000, 0x1000000), vec4(${N}));`;throw new Error(`not supported combination of storage type ${g.storage} and value type ${g.value} yet`)})(),pe=q=>(()=>{if(g.storage===g.value)return`${e}[${q}]`;if(g.storage==="vec2"&&g.value==="i32")return`i32(${e}[${q}].x)`;if(g.storage==="vec2"&&g.value==="u32")return`u32(${e}[${q}].x)`;if(g.storage==="u32"&&g.value==="vec4")return`vec4(bool(${e}[${q}] & 0xFFu), bool(${e}[${q}] & 0xFF00u), bool(${e}[${q}] & 0xFF0000u), bool(${e}[${q}] & 0xFF000000u))`;throw new Error(`not supported combination of storage type ${g.storage} and value type ${g.value} yet`)})(),re=u<2?"":` + fn get_${e}ByIndices(indices: ${g.indices}) -> ${y} { + return ${pe(`i2o_${e}(indices)`)}; + }`,oe=u<2?"":(()=>{let q=d.map(Q=>`d${Q}: u32`).join(", "),N=d.map(Q=>`d${Q}`).join(", ");return` + fn get_${e}(${q}) -> ${y} { + return get_${e}ByIndices(${O(N)}); + }`})(),ge=(...q)=>{if(q.length!==u)throw new Error(`indices length must be ${u}`);let N=q.map(T).join(",");return u===0?pe("0u"):u===1?pe(N[0]):(M.get=!0,M.getByIndices=!0,M.indicesToOffset=!0,`get_${e}(${N})`)},le=q=>u<2?pe(q):(M.getByIndices=!0,M.indicesToOffset=!0,`get_${e}ByIndices(${q})`),Se=u<2?"":` + fn set_${e}ByIndices(indices: ${g.indices}, value: ${y}) { + ${K(`i2o_${e}(indices)`,"value")} + }`,Ce=u<2?"":(()=>{let q=d.map(Q=>`d${Q}: u32`).join(", "),N=d.map(Q=>`d${Q}`).join(", ");return` + fn set_${e}(${q}, value: ${y}) { + set_${e}ByIndices(${O(N)}, value); + }`})();return{impl:()=>{let q=[],N=!1;return M.offsetToIndices&&(q.push(D),N=!0),M.indicesToOffset&&(q.push(P),N=!0),M.broadcastedIndicesToOffset&&(Object.values(te).forEach(Q=>q.push(Q)),N=!0),M.set&&(q.push(Ce),N=!0),M.setByIndices&&(q.push(Se),N=!0),M.get&&(q.push(oe),N=!0),M.getByIndices&&(q.push(re),N=!0),!n&&N&&q.unshift(`const ${z} = ${g.indices}(${t.join(",")});`,`const ${E} = ${g.indices}(${je.computeStrides(t).join(",")});`),q.join(` +`)},type:g,offsetToIndices:A,indicesToOffset:k,broadcastedIndicesToOffset:se,indices:O,indicesGet:R,indicesSet:U,set:(...q)=>{if(q.length!==u+1)throw new Error(`indices length must be ${u}`);let N=q[u];if(typeof N!="string")throw new Error("value must be string");let Q=q.slice(0,u).map(T).join(",");return u===0?K("0u",N):u===1?K(Q[0],N):(M.set=!0,M.setByIndices=!0,M.indicesToOffset=!0,`set_${e}(${Q}, ${N})`)},setByOffset:K,setByIndices:(q,N)=>u<2?K(q,N):(M.setByIndices=!0,M.indicesToOffset=!0,`set_${e}ByIndices(${q}, ${N});`),get:ge,getByOffset:pe,getByIndices:le,usage:i,name:e,strides:E,shape:z,rank:u}},Ne=(e,r,t,i=1)=>ad(e,r,t,"input",i),gt=(e,r,t,i=1)=>ad(e,r,t,"output",i),W0=(e,r,t)=>ad(e,r,t,"atomicOutput",1),r_=(e,r,t,i=1)=>ad(e,r,t,"internal",i),X_=class{constructor(e,r){this.normalizedDispatchGroup=e,this.limits=r,this.internalVariables=[],this.variables=[],this.uniforms=[],this.variableIndex=0}guardAgainstOutOfBoundsWorkgroupSizes(e){return`if (global_idx >= ${typeof e=="number"?`${e}u`:e}) { return; }`}mainStart(e=Va){let r=typeof e=="number"?e:e[0],t=typeof e=="number"?1:e[1],i=typeof e=="number"?1:e[2];if(r>this.limits.maxComputeWorkgroupSizeX||t>this.limits.maxComputeWorkgroupSizeY||i>this.limits.maxComputeWorkgroupSizeZ)throw new Error(`workgroup size [${r}, ${t}, ${i}] exceeds the maximum workgroup size [${this.limits.maxComputeWorkgroupSizeX}, ${this.limits.maxComputeWorkgroupSizeY}, ${this.limits.maxComputeWorkgroupSizeZ}].`);if(r*t*i>this.limits.maxComputeInvocationsPerWorkgroup)throw new Error(`workgroup size [${r}, ${t}, ${i}] exceeds the maximum workgroup invocations ${this.limits.maxComputeInvocationsPerWorkgroup}.`);let o=this.normalizedDispatchGroup[1]===1&&this.normalizedDispatchGroup[2]===1,n=o?`@builtin(global_invocation_id) global_id : vec3, + @builtin(workgroup_id) workgroup_id : vec3, + @builtin(local_invocation_index) local_idx : u32, + @builtin(local_invocation_id) local_id : vec3`:`@builtin(global_invocation_id) global_id : vec3, + @builtin(local_invocation_id) local_id : vec3, + @builtin(local_invocation_index) local_idx : u32, + @builtin(workgroup_id) workgroup_id : vec3, + @builtin(num_workgroups) num_workgroups : vec3`,u=o?`let global_idx = global_id.x; + let workgroup_index = workgroup_id.x;`:`let workgroup_index = workgroup_id.z * num_workgroups[0] * num_workgroups[1] + + workgroup_id.y * num_workgroups[0] + workgroup_id.x; + let global_idx = workgroup_index * ${r*t*i}u + local_idx;`;return`@compute @workgroup_size(${r}, ${t}, ${i}) + fn main(${n}) { + ${u} + `}appendVariableUniforms(e){e.rank!==0&&(e.shape.startsWith("uniforms.")&&this.uniforms.push({name:e.shape.replace("uniforms.",""),type:"u32",length:e.rank}),e.strides.startsWith("uniforms.")&&this.uniforms.push({name:e.strides.replace("uniforms.",""),type:"u32",length:e.rank}))}declareVariable(e,r){if(e.usage==="internal")throw new Error("cannot use internal variable with declareVariable(). use registerInternalVariables() instead.");this.variables.push(e),this.appendVariableUniforms(e);let t=e.usage==="input"?"read":"read_write",i=e.usage==="atomicOutput"?"atomic":e.type.storage;return`@group(0) @binding(${r}) var ${e.name}: array<${i}>;`}declareVariables(...e){return e.map(r=>this.declareVariable(r,this.variableIndex++)).join(` +`)}registerInternalVariable(e){if(e.usage!=="internal")throw new Error("cannot use input or output variable with registerInternalVariable(). use declareVariables() instead.");this.internalVariables.push(e),this.appendVariableUniforms(e)}registerInternalVariables(...e){return e.forEach(r=>this.registerInternalVariable(r)),this}registerUniform(e,r,t=1){return this.uniforms.push({name:e,type:r,length:t}),this}registerUniforms(e){return this.uniforms=this.uniforms.concat(e),this}uniformDeclaration(){if(this.uniforms.length===0)return"";let e=[];for(let{name:r,type:t,length:i}of this.uniforms)if(i&&i>4)t==="f16"?e.push(`@align(16) ${r}:array, ${Math.ceil(i/8)}>`):e.push(`${r}:array, ${Math.ceil(i/4)}>`);else{let o=i==null||i===1?t:`vec${i}<${t}>`;e.push(`${r}:${o}`)}return` + struct Uniforms { ${e.join(", ")} }; + @group(0) @binding(${this.variableIndex}) var uniforms: Uniforms;`}get additionalImplementations(){return this.uniformDeclaration()+this.variables.map(e=>e.impl()).join(` +`)+this.internalVariables.map(e=>e.impl()).join(` +`)}get variablesInfo(){if(this.uniforms.length===0)return;let e=r=>[12,10,1,6][["u32","f16","f32","i32"].indexOf(r)];return this.uniforms.map(r=>[e(r.type),r.length??1])}},U0=(e,r)=>new X_(e,r)}),J_,qf,Y_,Z_,eg,tg,_s,G0,K0,On=Ye(()=>{$t(),Lt(),br(),zt(),J_=(e,r)=>{if(!e||e.length!==1)throw new Error("Transpose requires 1 input.");if(r.length!==0&&r.length!==e[0].dims.length)throw new Error(`perm size ${r.length} does not match input rank ${e[0].dims.length}`)},qf=(e,r)=>r.length!==0?r:[...new Array(e).keys()].reverse(),Y_=(e,r)=>je.sortBasedOnPerm(e,qf(e.length,r)),Z_=(e,r,t,i)=>{let o=`fn perm(i: ${i.type.indices}) -> ${t.type.indices} { + var a: ${t.type.indices};`;for(let n=0;n{let t=[],i=[];for(let o=0;o{let t=0;for(let i=0;i{let t=e.dataType,i=e.dims.length,o=qf(i,r),n=Y_(e.dims,o),u=e.dims,d=n,h=i<2||tg(o,e.dims),f;if(h)return f=M=>{let I=Ne("input",t,u,4),z=gt("output",t,d,4);return` + ${M.registerUniform("output_size","u32").declareVariables(I,z)} + ${M.mainStart()} + ${M.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + output[global_idx] = input[global_idx]; + }`},{name:"TransposeCopy",shaderCache:{inputDependencies:["type"]},getRunData:()=>{let M=je.size(n);return{outputs:[{dims:n,dataType:e.dataType}],dispatchGroup:{x:Math.ceil(M/64/4)},programUniforms:[{type:12,data:Math.ceil(M/4)}]}},getShaderSource:f};let{newShape:y,newPerm:m}=eg(e.dims,o),g=je.areEqual(m,[2,3,1]),T=je.areEqual(m,[3,1,2]);if(y.length===2||g||T){u=g?[y[0],y[1]*y[2]]:T?[y[0]*y[1],y[2]]:y,d=[u[1],u[0]];let M=16;return f=I=>{let z=Ne("a",t,u.length),E=gt("output",t,d.length);return` + ${I.registerUniform("output_size","u32").declareVariables(z,E)} + var tile : array, ${M}>; + ${I.mainStart([M,M,1])} + let stride = (uniforms.output_shape[1] - 1) / ${M} + 1; + let workgroup_id_x = workgroup_index % stride; + let workgroup_id_y = workgroup_index / stride; + let input_col = workgroup_id_y * ${M}u + local_id.x; + let input_row = workgroup_id_x * ${M}u + local_id.y; + if (input_row < uniforms.a_shape[0] && input_col < uniforms.a_shape[1]) { + tile[local_id.y][local_id.x] = ${z.getByIndices(`${z.type.indices}(input_row, input_col)`)}; + } + workgroupBarrier(); + + let output_col = workgroup_id_x * ${M}u + local_id.x; + let output_row = workgroup_id_y * ${M}u + local_id.y; + if (output_row < uniforms.output_shape[0] && output_col < uniforms.output_shape[1]) { + ${E.setByIndices(`${E.type.indices}(output_row, output_col)`,"tile[local_id.x][local_id.y]")} + } + }`},{name:"TransposeShared",shaderCache:{inputDependencies:["type"]},getRunData:()=>{let I=je.size(n);return{outputs:[{dims:n,dataType:e.dataType}],dispatchGroup:{x:Math.ceil(d[1]/M),y:Math.ceil(d[0]/M)},programUniforms:[{type:12,data:I},...Et(u,d)]}},getShaderSource:f}}return f=M=>{let I=Ne("a",t,u.length),z=gt("output",t,d.length);return` + ${M.registerUniform("output_size","u32").declareVariables(I,z)} + + ${Z_(o,i,I,z)} + + ${M.mainStart()} + ${M.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + + let indices = ${z.offsetToIndices("global_idx")}; + let aIndices = perm(indices); + + ${z.setByOffset("global_idx",I.getByIndices("aIndices"))} + }`},{name:"Transpose",shaderCache:{hint:`${r}`,inputDependencies:["rank"]},getRunData:()=>{let M=je.size(n);return{outputs:[{dims:n,dataType:e.dataType}],dispatchGroup:{x:Math.ceil(M/64)},programUniforms:[{type:12,data:M},...Et(u,d)]}},getShaderSource:f}},G0=(e,r)=>{J_(e.inputs,r.perm),e.compute(_s(e.inputs[0],r.perm))},K0=e=>Zt({perm:e.perm})}),rg,sg,ng,ig,ag,og,lg,cg,ug,dg,Os,H0,q0,Q0,X0,J0,Y0,Z0,eb,tb,rb,rT=Ye(()=>{$t(),Lt(),zt(),s_(),On(),rg={max:"select(bestValue, candidate, candidate > bestValue)",min:"select(bestValue, candidate, candidate < bestValue)",mean:"bestValue + candidate",sum:"bestValue + candidate",prod:"bestValue * candidate",sumSquare:"bestValue + candidate * candidate",logSumExp:"bestValue + exp(candidate)",l1:"bestValue + abs(candidate)",l2:"bestValue + candidate * candidate",logSum:"bestValue + candidate"},sg={max:"select(bestValue, candidate, candidate > bestValue)",min:"select(bestValue, candidate, candidate < bestValue)",mean:"bestValue + candidate",sum:"bestValue + candidate",prod:"bestValue * candidate",sumSquare:"bestValue + candidate",logSumExp:"bestValue + candidate",l1:"bestValue + candidate",l2:"bestValue + candidate",logSum:"bestValue + candidate"},ng={max:"_A[offset]",min:"_A[offset]",mean:"0",sum:"0",prod:"1",sumSquare:"0",logSumExp:"0",l1:"0",l2:"0",logSum:"0"},ig={max:"bestValue",min:"bestValue",sum:"bestValue",prod:"bestValue",sumSquare:"bestValue",logSumExp:"log(bestValue)",l1:"bestValue",l2:"sqrt(bestValue)",logSum:"log(bestValue)"},ag=(e,r)=>{let t=[];for(let i=r-e;i{let t=[],i=e.length;for(let n=0;ne[n]);return[t,o]},lg=(e,r)=>{let t=e.length+r.length,i=[],o=0;for(let n=0;n{for(let t=0;t{let t=[];if(!cg(e,r)){for(let i=0;it.push(i))}return t},dg=(e,r,t,i,o,n,u)=>{let d=t[0].dims,h=je.size(n),f=je.size(u),y=Ne("_A",t[0].dataType,d),m=gt("output",o,n),g=64;h===1&&(g=256);let T=` + var aBestValues : array; + `,M=I=>` + ${I.registerUniform("reduceSize","u32").declareVariables(y,m)} + ${T} + fn DIV_CEIL(a : u32, b : u32) -> u32 { + return ((a - 1u) / b + 1u); + } + ${I.mainStart(g)} + + let outputIndex = global_idx / ${g}; + let offset = outputIndex * uniforms.reduceSize; + + var bestValue = f32(${ng[i]}); + let Length = uniforms.reduceSize; + for (var k = local_idx; k < Length; k = k + ${g}) { + let candidate = f32(${y.getByOffset("offset + k")}); + bestValue = ${rg[i]}; + } + aBestValues[local_idx] = bestValue; + workgroupBarrier(); + + var reduceSize = min(Length, ${g}u); + for (var currentSize = reduceSize / 2u; reduceSize > 1u; + currentSize = reduceSize / 2u) { + let interval = DIV_CEIL(reduceSize, 2u); + if (local_idx < currentSize) { + let candidate = aBestValues[local_idx + interval]; + bestValue = ${sg[i]}; + aBestValues[local_idx] = bestValue; + } + reduceSize = interval; + workgroupBarrier(); + } + + if (local_idx == 0u) { + ${m.setByOffset("outputIndex",`${i==="mean"?`${m.type.storage}(bestValue / f32(uniforms.reduceSize))`:`${m.type.storage}(${ig[i]})`}`)}; + } + }`;return{name:e,shaderCache:{hint:`${r};${g}`,inputDependencies:["type"]},getShaderSource:M,getRunData:()=>({outputs:[{dims:n,dataType:o}],dispatchGroup:{x:h},programUniforms:[{type:12,data:f}]})}},Os=(e,r,t,i)=>{let o=e.inputs.length===1?t:Am(e.inputs,t),n=o.axes;n.length===0&&!o.noopWithEmptyAxes&&(n=e.inputs[0].dims.map((T,M)=>M));let u=je.normalizeAxes(n,e.inputs[0].dims.length),d=u,h=e.inputs[0],f=ug(d,e.inputs[0].dims.length);f.length>0&&(h=e.compute(_s(e.inputs[0],f),{inputs:[0],outputs:[-1]})[0],d=ag(d.length,h.dims.length));let[y,m]=og(h.dims,d),g=y;o.keepDims&&(g=lg(y,u)),e.compute(dg(r,o.cacheKey,[h],i,e.inputs[0].dataType,g,m),{inputs:[h]})},H0=(e,r)=>{Os(e,"ReduceMeanShared",r,"mean")},q0=(e,r)=>{Os(e,"ReduceL1Shared",r,"l1")},Q0=(e,r)=>{Os(e,"ReduceL2Shared",r,"l2")},X0=(e,r)=>{Os(e,"ReduceLogSumExpShared",r,"logSumExp")},J0=(e,r)=>{Os(e,"ReduceMaxShared",r,"max")},Y0=(e,r)=>{Os(e,"ReduceMinShared",r,"min")},Z0=(e,r)=>{Os(e,"ReduceProdShared",r,"prod")},eb=(e,r)=>{Os(e,"ReduceSumShared",r,"sum")},tb=(e,r)=>{Os(e,"ReduceSumSquareShared",r,"sumSquare")},rb=(e,r)=>{Os(e,"ReduceLogSumShared",r,"logSum")}}),Fs,pg,tf,Am,Ds,hg,fg,mg,_g,gg,yg,wg,bg,Mg,vg,js,sb,nb,ib,ab,ob,lb,cb,ub,db,pb,s_=Ye(()=>{$t(),Lt(),br(),zt(),rT(),Fs=e=>{if(!e||e.length===0||e.length>2)throw new Error("Reduce op requires 1 or 2 inputs.");if(e.length===2&&e[1].dims.length!==1)throw new Error("Invalid axes input dims.")},pg=e=>["","",`var value = ${e.getByIndices("input_indices")};`,""],tf=(e,r,t,i,o,n,u=!1,d=!1)=>{let h=[],f=t[0].dims,y=f.length,m=je.normalizeAxes(o,y),g=!d&&m.length===0;f.forEach((I,z)=>{g||m.indexOf(z)>=0?u&&h.push(1):h.push(I)});let T=h.length,M=je.size(h);return{name:e,shaderCache:r,getShaderSource:I=>{let z=[],E=Ne("_A",t[0].dataType,y),C=gt("output",n,T),D=i(E,C,m),A=D[2];for(let $=0,P=0;$=0?(u&&P++,A=`for(var j${$}: u32 = 0; j${$} < ${f[$]}; j${$}++) { + ${D[2].includes("last_index")?`let last_index = j${$};`:""} + ${E.indicesSet("input_indices",$,`j${$}`)} + ${A} + }`):(z.push(`${E.indicesSet("input_indices",$,C.indicesGet("output_indices",P))};`),P++);return` + + ${I.registerUniform("output_size","u32").declareVariables(E,C)} + + ${I.mainStart()} + ${I.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + var input_indices: ${E.type.indices}; + let output_indices = ${C.offsetToIndices("global_idx")}; + + ${z.join(` +`)} + ${D[0]} // init ops for reduce max/min + ${D[1]} + ${A} + ${D[3]} + ${D.length===4?C.setByOffset("global_idx","value"):D.slice(4).join(` +`)} + }`},getRunData:()=>({outputs:[{dims:h,dataType:n}],dispatchGroup:{x:Math.ceil(M/64)},programUniforms:[{type:12,data:M},...Et(f,h)]})}},Am=(e,r)=>{let t=[];return e[1].dims[0]>0&&e[1].getBigInt64Array().forEach(i=>t.push(Number(i))),Zt({axes:t,keepDims:r.keepDims,noopWithEmptyAxes:r.noopWithEmptyAxes})},Ds=(e,r,t,i)=>{let o=e.inputs,n=o.length===1?t:Am(o,t);e.compute(tf(r,{hint:n.cacheKey,inputDependencies:["rank"]},[o[0]],n.noopWithEmptyAxes&&n.axes.length===0?pg:i,n.axes,o[0].dataType,n.keepDims,n.noopWithEmptyAxes),{inputs:[0]})},hg=(e,r)=>{Fs(e.inputs),Ds(e,"ReduceLogSum",r,(t,i)=>[`var value = ${i.type.storage}(0);`,"",`value += ${t.getByIndices("input_indices")};`,"value = log(value);"])},fg=(e,r)=>{Fs(e.inputs),Ds(e,"ReduceL1",r,(t,i)=>[`var value = ${i.type.storage}(0);`,"",`value += abs(${t.getByIndices("input_indices")});`,""])},mg=(e,r)=>{Fs(e.inputs),Ds(e,"ReduceL2",r,(t,i)=>[`var t = ${i.type.value}(0); var value = ${i.type.value}(0);`,"",`t = ${t.getByIndices("input_indices")}; value += (t * t);`,"value = sqrt(value);"])},_g=(e,r)=>{Fs(e.inputs),Ds(e,"ReduceLogSumExp",r,(t,i)=>[`var value = ${i.type.storage}(0);`,"",`value += exp(${t.getByIndices("input_indices")});`,"value = log(value);"])},gg=(e,r)=>{Fs(e.inputs),Ds(e,"ReduceMax",r,(t,i,o)=>{let n=[];for(let u=0;u=0||o.length===0)&&n.push(t.indicesSet("input_indices",u,0));return[`${n.join(` +`)}`,`var value = ${t.getByIndices("input_indices")};`,`value = max(value, ${t.getByIndices("input_indices")});`,""]})},yg=(e,r)=>{Fs(e.inputs),Ds(e,"ReduceMean",r,(t,i,o)=>{let n=1;for(let u=0;u=0||o.length===0)&&(n*=e.inputs[0].dims[u]);return["var sum = f32(0);","",`sum += f32(${t.getByIndices("input_indices")});`,`let value = ${i.type.value}(sum / ${n});`]})},wg=(e,r)=>{Fs(e.inputs),Ds(e,"ReduceMin",r,(t,i,o)=>{let n=[];for(let u=0;u=0||o.length===0)&&n.push(`input_indices[${u}] = 0;`);return[`${n.join(` +`)}`,`var value = ${t.getByIndices("input_indices")};`,`value = min(value, ${t.getByIndices("input_indices")});`,""]})},bg=(e,r)=>{Fs(e.inputs),Ds(e,"ReduceProd",r,(t,i)=>[`var value = ${i.type.storage}(1);`,"",`value *= ${t.getByIndices("input_indices")};`,""])},Mg=(e,r)=>{Fs(e.inputs),Ds(e,"ReduceSum",r,(t,i)=>[`var value = ${i.type.storage}(0);`,"",`value += ${t.getByIndices("input_indices")};`,""])},vg=(e,r)=>{Fs(e.inputs),Ds(e,"ReduceSumSquare",r,(t,i)=>[`var t = ${i.type.value}(0); var value = ${i.type.value}(0);`,"",`t = ${t.getByIndices("input_indices")}; value += t * t;`,""])},js=(e,r,t)=>{if(r.length===0)return t;let i=1,o=1;for(let n=0;n1024},sb=(e,r)=>{js(e.inputs[0].dims,r.axes,r.noopWithEmptyAxes)?yg(e,r):H0(e,r)},nb=(e,r)=>{js(e.inputs[0].dims,r.axes,r.noopWithEmptyAxes)?fg(e,r):q0(e,r)},ib=(e,r)=>{js(e.inputs[0].dims,r.axes,r.noopWithEmptyAxes)?mg(e,r):Q0(e,r)},ab=(e,r)=>{js(e.inputs[0].dims,r.axes,r.noopWithEmptyAxes)?_g(e,r):X0(e,r)},ob=(e,r)=>{js(e.inputs[0].dims,r.axes,r.noopWithEmptyAxes)?gg(e,r):J0(e,r)},lb=(e,r)=>{js(e.inputs[0].dims,r.axes,r.noopWithEmptyAxes)?wg(e,r):Y0(e,r)},cb=(e,r)=>{js(e.inputs[0].dims,r.axes,r.noopWithEmptyAxes)?bg(e,r):Z0(e,r)},ub=(e,r)=>{js(e.inputs[0].dims,r.axes,r.noopWithEmptyAxes)?Mg(e,r):eb(e,r)},db=(e,r)=>{js(e.inputs[0].dims,r.axes,r.noopWithEmptyAxes)?vg(e,r):tb(e,r)},pb=(e,r)=>{js(e.inputs[0].dims,r.axes,r.noopWithEmptyAxes)?hg(e,r):rb(e,r)}}),Qf,hb,fb,Om,sT=Ye(()=>{$t(),br(),s_(),Qf=e=>{if(!e||e.length===0||e.length>2)throw new Error("ArgMinMaxOp op requires 1 or 2 inputs.");if(e[0].dataType!==1)throw new Error("Invalid input type.")},hb=(e,r)=>{Qf(e.inputs);let t=(i,o,n)=>{let u=[];for(let d=0;d=0||n.length===0)&&u.push(`input_indices[${d}] = 0;`);return[`${u.join(` +`)}`,`var value = ${i.getByIndices("input_indices")}; +var best_index : i32 = 0;`,`if (${i.getByIndices("input_indices")} ${r.selectLastIndex>0?"<=":"<"} value) { + value = ${i.getByIndices("input_indices")}; + best_index = i32(last_index); + }`,"",o.setByOffset("global_idx","best_index")]};e.compute(tf("ArgMin",{hint:r.cacheKey,inputDependencies:["rank"]},[e.inputs[0]],t,[r.axis],7,r.keepDims),{inputs:[0]})},fb=(e,r)=>{Qf(e.inputs);let t=(i,o,n)=>{let u=[];for(let d=0;d=0||n.length===0)&&u.push(`input_indices[${d}] = 0;`);return[`${u.join(` +`)}`,`var value = ${i.getByIndices("input_indices")}; +var best_index : i32 = 0;`,`if (${i.getByIndices("input_indices")} ${r.selectLastIndex>0?">=":">"} value) { + value = ${i.getByIndices("input_indices")}; + best_index = i32(last_index); + }`,"",o.setByOffset("global_idx","best_index")]};e.compute(tf("argMax",{hint:r.cacheKey,inputDependencies:["rank"]},[e.inputs[0]],t,[r.axis],7,r.keepDims),{inputs:[0]})},Om=e=>Zt(e)}),xg,Rh,Tg,Cg,Eg,Md,Pg,mb,n_=Ye(()=>{$t(),Lt(),t_(),zt(),xg=(e,r)=>{let t=e[0],i=e[1],o=e[2],n=e[3],u=e[4],d=e[5];if(u&&d)throw new Error("Attention cannot have both past and attention_bias");if(t.dims.length!==3)throw new Error('Input "input" must have 3 dimensions');let h=t.dims[0],f=t.dims[1],y=t.dims[2];if(o.dims.length!==1)throw new Error('Input "bias" is expected to have 1 dimensions');if(i.dims.length!==2)throw new Error('Input "weights" is expected to have 2 dimensions');if(i.dims[0]!==y)throw new Error("Input 1 dimension 0 should have same length as dimension 2 of input 0");if(o.dims[0]!==i.dims[1])throw new Error('Input "bias" dimension 0 should have same length as dimension 1 of input "weights"');let m=o.dims[0]/3,g=m,T=g;if(r.qkvHiddenSizes.length>0){if(r.qkvHiddenSizes.length!==3)throw new Error("qkv_hidden_sizes attribute should have 3 elements");for(let D of r.qkvHiddenSizes)if(D%r.numHeads!==0)throw new Error("qkv_hidden_sizes should be divisible by num_heads");m=r.qkvHiddenSizes[0],g=r.qkvHiddenSizes[1],T=r.qkvHiddenSizes[2]}let M=f;if(m!==g)throw new Error("qkv_hidden_sizes first element should be same as the second");if(o.dims[0]!==m+g+T)throw new Error('Input "bias" dimension 0 should have same length as sum of Q/K/V hidden sizes');let I=0;if(u){if(g!==T)throw new Error('Input "past" expect k_hidden_size == v_hidden_size');if(u.dims.length!==5)throw new Error('Input "past" must have 5 dimensions');if(u.dims[0]!==2)throw new Error('Input "past" first dimension must be 2');if(u.dims[1]!==h)throw new Error('Input "past" second dimension must be batch_size');if(u.dims[2]!==r.numHeads)throw new Error('Input "past" third dimension must be num_heads');if(u.dims[4]!==g/r.numHeads)throw new Error('Input "past" fifth dimension must be k_hidden_size / num_heads');r.pastPresentShareBuffer||(I=u.dims[3])}let z=M+I,E=-1,C=0;if(n)throw new Error("Mask not supported");if(u)throw new Error("past is not supported");if(d){if(d.dims.length!==4)throw new Error('Input "attention_bias" must have 4 dimensions');if(d.dims[0]!==h||d.dims[1]!==r.numHeads||d.dims[2]!==f||d.dims[3]!==z)throw new Error('Expect "attention_bias" shape (batch_size, num_heads, sequence_length, total_sequence_length)')}return{batchSize:h,sequenceLength:f,pastSequenceLength:I,kvSequenceLength:M,totalSequenceLength:z,maxSequenceLength:E,inputHiddenSize:y,hiddenSize:m,vHiddenSize:T,headSize:Math.floor(m/r.numHeads),vHeadSize:Math.floor(T/r.numHeads),numHeads:r.numHeads,isUnidirectional:!1,pastPresentShareBuffer:!1,maskFilterValue:r.maskFilterValue,maskType:C,scale:r.scale,broadcastResPosBias:!1,passPastInKv:!1,qkvFormat:1}},Rh=(e,r,t)=>r&&e?` + let total_sequence_length_input = u32(${r.getByOffset("0")}); + let present_sequence_length = max(total_sequence_length_input, uniforms.past_sequence_length); + let is_subsequent_prompt: bool = sequence_length > 1 && sequence_length != total_sequence_length_input; + let is_first_prompt: bool = is_subsequent_prompt == false && sequence_length == total_sequence_length_input; + total_sequence_length = u32(${e==null?void 0:e.getByOffset("batchIdx")}) + 1; + var past_sequence_length: u32 = 0; + if (is_first_prompt == false) { + past_sequence_length = total_sequence_length - sequence_length; + } + `:` + ${t?"let past_sequence_length = uniforms.past_sequence_length":""}; + let present_sequence_length = total_sequence_length; + `,Tg=(e,r,t,i,o,n,u,d)=>{let h=gr(u?1:n),f=64,y=n/h;y{let C=gt("x",e.dataType,e.dims,h),D=[C],A=u?Ne("seq_lens",u.dataType,u.dims):void 0;A&&D.push(A);let $=d?Ne("total_sequence_length_input",d.dataType,d.dims):void 0;$&&D.push($);let P=Qr(e.dataType),k=[{name:"batch_size",type:"u32"},{name:"num_heads",type:"u32"},{name:"past_sequence_length",type:"u32"},{name:"sequence_length",type:"u32"},{name:"total_sequence_length",type:"u32"},{name:"elements_per_thread",type:"u32"}];return` + var thread_max: array; + var thread_sum: array; + ${E.registerUniforms(k).declareVariables(...D)} + ${E.mainStart([f,1,1])} + let batchIdx = workgroup_id.z / uniforms.num_heads; + let headIdx = workgroup_id.z % uniforms.num_heads; + let sequence_length = uniforms.sequence_length; + var total_sequence_length = uniforms.total_sequence_length; + ${Rh(A,$,!1)} + let local_offset = local_idx * uniforms.elements_per_thread; + let offset = (global_idx / ${f}) * uniforms.total_sequence_length + local_offset; + let seq_causal_length = ${u?"u32(past_sequence_length + workgroup_id.y + 1)":"total_sequence_length"}; + var thread_max_vector = ${M}(-3.402823e+38f); + for (var i: u32 = 0; i < uniforms.elements_per_thread && i + local_offset < seq_causal_length; i++) { + thread_max_vector = max(${M}(x[offset + i]), thread_max_vector); + } + thread_max[local_idx] = ${(()=>{switch(h){case 1:return"thread_max_vector";case 2:return"max(thread_max_vector.x, thread_max_vector.y)";case 4:return"max(max(thread_max_vector.x, thread_max_vector.y), max(thread_max_vector.z, thread_max_vector.w))";default:throw new Error(`Unsupported components: ${h}`)}})()}; + workgroupBarrier(); + + var max_value = f32(-3.402823e+38f); + for (var i = 0u; i < ${f}; i++) { + max_value = max(thread_max[i], max_value); + } + + var sum_vector = ${M}(0); + for (var i: u32 = 0; i < uniforms.elements_per_thread && i + local_offset < seq_causal_length; i++) { + sum_vector += exp(${M}(x[offset + i]) - max_value); + } + thread_sum[local_idx] = ${(()=>{switch(h){case 1:return"sum_vector";case 2:return"sum_vector.x + sum_vector.y";case 4:return"sum_vector.x + sum_vector.y + sum_vector.z + sum_vector.w";default:throw new Error(`Unsupported components: ${h}`)}})()}; + workgroupBarrier(); + + var sum: f32 = 0; + for (var i = 0u; i < ${f}; i++) { + sum += thread_sum[i]; + } + + if (sum == 0) { + for (var i: u32 = 0; i < uniforms.elements_per_thread && i + local_offset < seq_causal_length; i++) { + x[offset + i] = ${C.type.value}(${P}(1.0) / ${P}(seq_causal_length)); + } + } else { + for (var i: u32 = 0; i < uniforms.elements_per_thread && i + local_offset < seq_causal_length; i++) { + var f32input = ${M}(x[offset + i]); + x[offset + i] = ${C.type.value}(exp(f32input - max_value) / sum); + } + } + ${u?` + for (var total_seq_id: u32 = seq_causal_length; total_seq_id + local_offset < uniforms.total_sequence_length; total_seq_id++) { + x[offset + total_seq_id] = ${C.type.value}(${P}(0)); + }`:""}; + }`};return{name:"AttentionProbsSoftmax",shaderCache:{hint:`${f};${T};${h}`,inputDependencies:I},getShaderSource:z,getRunData:()=>({outputs:[],dispatchGroup:{x:1,y:o,z:r*t},programUniforms:g})}},Cg=(e,r,t,i,o,n,u,d,h)=>{let f=u+n.kvSequenceLength,y=[n.batchSize,n.numHeads,n.sequenceLength,f],m=e>1&&i,g=n.kvNumHeads?n.kvNumHeads:n.numHeads,T=m?[n.batchSize,g,f,n.headSize]:void 0,M=n.nReps?n.nReps:1,I=n.scale===0?1/Math.sqrt(n.headSize):n.scale,z=gr(n.headSize),E=n.headSize/z,C=12,D={x:Math.ceil(f/C),y:Math.ceil(n.sequenceLength/C),z:n.batchSize*n.numHeads},A=[{type:12,data:n.sequenceLength},{type:12,data:E},{type:12,data:f},{type:12,data:n.numHeads},{type:12,data:n.headSize},{type:1,data:I},{type:12,data:u},{type:12,data:n.kvSequenceLength},{type:12,data:M}],$=m&&i&&je.size(i.dims)>0,P=["type","type"];$&&P.push("type"),o&&P.push("type"),d&&P.push("type"),h&&P.push("type");let k=[{dims:y,dataType:r.dataType,gpuDataType:0}];m&&k.push({dims:T,dataType:r.dataType,gpuDataType:0});let O=R=>{let U=Ne("q",r.dataType,r.dims,z),te=Ne("key",t.dataType,t.dims,z),se=[U,te];if($){let Se=Ne("past_key",i.dataType,i.dims,z);se.push(Se)}o&&se.push(Ne("attention_bias",o.dataType,o.dims));let K=d?Ne("seq_lens",d.dataType,d.dims):void 0;K&&se.push(K);let pe=h?Ne("total_sequence_length_input",h.dataType,h.dims):void 0;pe&&se.push(pe);let re=gt("output",r.dataType,y),oe=[re];m&&oe.push(gt("present_key",r.dataType,T,z));let ge=Qr(1,z),le=[{name:"M",type:"u32"},{name:"K",type:"u32"},{name:"N",type:"u32"},{name:"num_heads",type:"u32"},{name:"head_size",type:"u32"},{name:"alpha",type:"f32"},{name:"past_sequence_length",type:"u32"},{name:"kv_sequence_length",type:"u32"},{name:"n_reps",type:"u32"}];return` + const TILE_SIZE = ${C}u; + + var tileQ: array<${U.type.storage}, ${C*C}>; + var tileK: array<${U.type.storage}, ${C*C}>; + ${R.registerUniforms(le).declareVariables(...se,...oe)} + ${R.mainStart([C,C,1])} + // x holds the N and y holds the M + let headIdx = workgroup_id.z % uniforms.num_heads; + let kvHeadIdx = ${M===1?"headIdx":"headIdx / uniforms.n_reps"}; + let kv_num_heads = ${M===1?"uniforms.num_heads":"uniforms.num_heads / uniforms.n_reps"}; + let batchIdx = workgroup_id.z / uniforms.num_heads; + let m = workgroup_id.y * TILE_SIZE; + let n = workgroup_id.x * TILE_SIZE; + let sequence_length = uniforms.M; + var total_sequence_length = uniforms.N; + ${Rh(K,pe,!0)} + let absKvHeadIdx = batchIdx * kv_num_heads + kvHeadIdx; + let qOffset = workgroup_id.z * uniforms.M * uniforms.K + m * uniforms.K; + ${$&&m?"let pastKeyOffset = absKvHeadIdx * uniforms.past_sequence_length * uniforms.K;":""}; + let kOffset = absKvHeadIdx * uniforms.kv_sequence_length * uniforms.K; + ${m?"let presentKeyOffset = absKvHeadIdx * uniforms.N * uniforms.K;":""} + var value = ${ge}(0); + for (var w: u32 = 0u; w < uniforms.K; w += TILE_SIZE) { + if (global_id.y < uniforms.M && w + local_id.x < uniforms.K) { + tileQ[TILE_SIZE * local_id.y + local_id.x] = q[qOffset + local_id.y * uniforms.K + w + local_id.x]; + } + if (n + local_id.y < uniforms.N && w + local_id.x < uniforms.K) { + var idx = TILE_SIZE * local_id.y + local_id.x; + ${$&&m?` + if (n + local_id.y < past_sequence_length) { + tileK[idx] = past_key[pastKeyOffset + (n + local_id.y) * uniforms.K + w + local_id.x]; + } else if (n + local_id.y - past_sequence_length < uniforms.kv_sequence_length) { + tileK[idx] = key[kOffset + (n + local_id.y - past_sequence_length) * uniforms.K + w + local_id.x]; + }`:` + if (n + local_id.y < uniforms.kv_sequence_length) { + tileK[idx] = key[kOffset + (n + local_id.y) * uniforms.K + w + local_id.x]; + }`} + ${m?`if (n + local_id.y < present_sequence_length) { + present_key[presentKeyOffset + (n + local_id.y) * uniforms.K + w + local_id.x] = tileK[idx]; + }`:""} + } + workgroupBarrier(); + + for (var k: u32 = 0u; k < TILE_SIZE && w+k < uniforms.K; k++) { + value += ${ge}(tileQ[TILE_SIZE * local_id.y + k] * tileK[TILE_SIZE * local_id.x + k]); + } + + workgroupBarrier(); + } + + if (global_id.y < uniforms.M && global_id.x < total_sequence_length) { + let headOffset = workgroup_id.z * uniforms.M * uniforms.N; + let outputIdx = headOffset + global_id.y * uniforms.N + global_id.x; + var sum: f32 = ${(()=>{switch(z){case 1:return"value";case 2:return"value.x + value.y";case 4:return"value.x + value.y + value.z + value.w";default:throw new Error(`Unsupported components: ${z}`)}})()}; + output[outputIdx] = ${re.type.value} (sum * uniforms.alpha) + ${o?"attention_bias[outputIdx]":"0.0"}; + } + }`};return{name:"AttentionProbs",shaderCache:{hint:`${z};${o!==void 0};${i!==void 0};${e}`,inputDependencies:P},getRunData:()=>({outputs:k,dispatchGroup:D,programUniforms:A}),getShaderSource:O}},Eg=(e,r,t,i,o,n,u=void 0,d=void 0)=>{let h=n+o.kvSequenceLength,f=o.nReps?o.nReps:1,y=o.vHiddenSize*f,m=e>1&&i,g=o.kvNumHeads?o.kvNumHeads:o.numHeads,T=m?[o.batchSize,g,h,o.headSize]:void 0,M=[o.batchSize,o.sequenceLength,y],I=12,z={x:Math.ceil(o.vHeadSize/I),y:Math.ceil(o.sequenceLength/I),z:o.batchSize*o.numHeads},E=[{type:12,data:o.sequenceLength},{type:12,data:h},{type:12,data:o.vHeadSize},{type:12,data:o.numHeads},{type:12,data:o.headSize},{type:12,data:y},{type:12,data:n},{type:12,data:o.kvSequenceLength},{type:12,data:f}],C=m&&i&&je.size(i.dims)>0,D=["type","type"];C&&D.push("type"),u&&D.push("type"),d&&D.push("type");let A=[{dims:M,dataType:r.dataType,gpuDataType:0}];m&&A.push({dims:T,dataType:r.dataType,gpuDataType:0});let $=P=>{let k=Ne("probs",r.dataType,r.dims),O=Ne("v",t.dataType,t.dims),R=[k,O];C&&R.push(Ne("past_value",i.dataType,i.dims));let U=u?Ne("seq_lens",u.dataType,u.dims):void 0;u&&R.push(U);let te=d?Ne("total_sequence_length_input",d.dataType,d.dims):void 0;d&&R.push(te);let se=[gt("output",r.dataType,M)];m&&se.push(gt("present_value",r.dataType,T));let K=[{name:"M",type:"u32"},{name:"K",type:"u32"},{name:"N",type:"u32"},{name:"num_heads",type:"u32"},{name:"head_size",type:"u32"},{name:"v_hidden_size",type:"u32"},{name:"past_sequence_length",type:"u32"},{name:"kv_sequence_length",type:"u32"},{name:"n_reps",type:"u32"}];return` + const TILE_SIZE = ${I}u; + var tileQ: array<${k.type.value}, ${I*I}>; + var tileV: array<${k.type.value}, ${I*I}>; + ${P.registerUniforms(K).declareVariables(...R,...se)} + ${P.mainStart([I,I,1])} + let headIdx = workgroup_id.z % uniforms.num_heads; + let batchIdx = workgroup_id.z / uniforms.num_heads; + let kvHeadIdx = ${f===1?"headIdx":"headIdx / uniforms.n_reps"}; + let kv_num_heads = ${f===1?"uniforms.num_heads":"uniforms.num_heads / uniforms.n_reps"}; + let m = global_id.y; + let n = global_id.x; + let sequence_length = uniforms.M; + var total_sequence_length = uniforms.K; + ${Rh(U,te,!0)} + let offsetA = workgroup_id.z * uniforms.M * uniforms.K + m * uniforms.K; + let absKvHeadIdx = batchIdx * kv_num_heads + kvHeadIdx; // kvHeadIdx is relative to the batch + ${C&&m?"let pastValueOffset = absKvHeadIdx * uniforms.N * uniforms.past_sequence_length + n;":""}; + let vOffset = absKvHeadIdx * uniforms.N * uniforms.kv_sequence_length + n; + ${m?"let presentValueOffset = absKvHeadIdx * uniforms.N * uniforms.K + n;":""} + var value = ${k.type.storage}(0); + for (var w: u32 = 0u; w < uniforms.K; w += TILE_SIZE) { + if (m < uniforms.M && w + local_id.x < uniforms.K) { + tileQ[TILE_SIZE * local_id.y + local_id.x] = probs[offsetA + w + local_id.x]; + } + if (n < uniforms.N && w + local_id.y < uniforms.K) { + var idx = TILE_SIZE * local_id.y + local_id.x; + ${C&&m?` + if (w + local_id.y < past_sequence_length) { + tileV[idx] = past_value[pastValueOffset + (w + local_id.y) * uniforms.N]; + } else if (w + local_id.y - past_sequence_length < uniforms.kv_sequence_length) { + tileV[idx] = v[vOffset + (w + local_id.y - past_sequence_length) * uniforms.N]; + } + `:` + if (w + local_id.y < uniforms.kv_sequence_length) { + tileV[idx] = v[vOffset + (w + local_id.y) * uniforms.N]; + }`} + ${m?` + if (w + local_id.y < present_sequence_length) { + present_value[presentValueOffset + (w + local_id.y) * uniforms.N] = tileV[idx]; + }`:""} + } + workgroupBarrier(); + for (var k: u32 = 0u; k < TILE_SIZE && w+k < total_sequence_length; k++) { + value += tileQ[TILE_SIZE * local_id.y + k] * tileV[TILE_SIZE * k + local_id.x]; + } + workgroupBarrier(); + } + + // we need to transpose output from BNSH_v to BSND_v + if (m < uniforms.M && n < uniforms.N) { + let outputIdx = batchIdx * uniforms.M * uniforms.v_hidden_size + m * uniforms.v_hidden_size + + headIdx * uniforms.N + n; + output[outputIdx] = value; + } + }`};return{name:"AttentionScore",shaderCache:{hint:`${i!==void 0};${e}`,inputDependencies:D},getRunData:()=>({outputs:A,dispatchGroup:z,programUniforms:E}),getShaderSource:$}},Md=(e,r,t,i,o,n,u,d,h,f,y=void 0,m=void 0)=>{let g=Math.min(e.outputCount,1+(u?1:0)+(d?1:0)),T=g>1?f.pastSequenceLength:0,M=T+f.kvSequenceLength,I=h&&je.size(h.dims)>0?h:void 0,z=[r,t];g>1&&u&&je.size(u.dims)>0&&z.push(u),I&&z.push(I),y&&z.push(y),m&&z.push(m);let E=e.compute(Cg(g,r,t,u,I,f,T,y,m),{inputs:z,outputs:g>1?[-1,1]:[-1]})[0];e.compute(Tg(E,f.batchSize,f.numHeads,T,f.sequenceLength,M,y,m),{inputs:y&&m?[E,y,m]:[E],outputs:[]});let C=[E,i];g>1&&d&&je.size(d.dims)>0&&C.push(d),y&&C.push(y),m&&C.push(m),e.compute(Eg(g,E,i,d,f,T,y,m),{inputs:C,outputs:g>1?[0,2]:[0]})},Pg=(e,r)=>{let t=[r.batchSize,r.numHeads,r.sequenceLength,r.headSize],i=r.sequenceLength,o=r.inputHiddenSize,n=r.headSize,u=12,d={x:Math.ceil(r.headSize/u),y:Math.ceil(r.sequenceLength/u),z:r.batchSize*r.numHeads},h=[e.inputs[0],e.inputs[1],e.inputs[2]],f=[{type:12,data:i},{type:12,data:o},{type:12,data:n},{type:12,data:r.numHeads},{type:12,data:r.headSize},{type:12,data:r.hiddenSize},{type:12,data:r.hiddenSize+r.hiddenSize+r.vHiddenSize}],y=m=>{let g=gt("output_q",h[0].dataType,t),T=gt("output_k",h[0].dataType,t),M=gt("output_v",h[0].dataType,t),I=Ne("input",h[0].dataType,h[0].dims),z=Ne("weight",h[1].dataType,h[1].dims),E=Ne("bias",h[2].dataType,h[2].dims),C=I.type.storage,D=[{name:"M",type:"u32"},{name:"K",type:"u32"},{name:"N",type:"u32"},{name:"num_heads",type:"u32"},{name:"head_size",type:"u32"},{name:"hidden_size",type:"u32"},{name:"ldb",type:"u32"}];return` + const TILE_SIZE = ${u}u; + var tileInput: array<${C}, ${u*u}>; + var tileWeightQ: array<${C}, ${u*u}>; + var tileWeightK: array<${C}, ${u*u}>; + var tileWeightV: array<${C}, ${u*u}>; + ${m.registerUniforms(D).declareVariables(I,z,E,g,T,M)} + ${m.mainStart([u,u,1])} + let batchIndex = workgroup_id.z / uniforms.num_heads; + let headNumber = workgroup_id.z % uniforms.num_heads; + let m = global_id.y; + let n = global_id.x; + + let inputOffset = batchIndex * (uniforms.M * uniforms.K) + m * uniforms.K; + let biasOffsetQ = headNumber * uniforms.head_size; + let biasOffsetK = uniforms.hidden_size + biasOffsetQ; + let biasOffsetV = uniforms.hidden_size + biasOffsetK; + + var valueQ = ${C}(0); + var valueK = ${C}(0); + var valueV = ${C}(0); + for (var w: u32 = 0u; w < uniforms.K; w += TILE_SIZE) { + if (m < uniforms.M && w + local_id.x < uniforms.K) { + tileInput[TILE_SIZE * local_id.y + local_id.x] = input[inputOffset + w + local_id.x]; + } + if (n < uniforms.N && w + local_id.y < uniforms.K) { + let offset = n + (w + local_id.y) * uniforms.ldb; + tileWeightQ[TILE_SIZE * local_id.y + local_id.x] = weight[biasOffsetQ + offset]; + tileWeightK[TILE_SIZE * local_id.y + local_id.x] = weight[biasOffsetK + offset]; + tileWeightV[TILE_SIZE * local_id.y + local_id.x] = weight[biasOffsetV + offset]; + } + workgroupBarrier(); + for (var k: u32 = 0u; k({outputs:[{dims:t,dataType:e.inputs[0].dataType,gpuDataType:0},{dims:t,dataType:e.inputs[0].dataType,gpuDataType:0},{dims:t,dataType:e.inputs[0].dataType,gpuDataType:0}],dispatchGroup:d,programUniforms:f}),getShaderSource:y},{inputs:h,outputs:[-1,-1,-1]})},mb=(e,r)=>{let t=xg(e.inputs,r),[i,o,n]=Pg(e,t);return Md(e,i,o,n,e.inputs[4],void 0,void 0,void 0,e.inputs[5],t)}}),Sg,kg,$g,_b,nT=Ye(()=>{Vs(),$t(),Lt(),br(),zt(),Sg=(e,r)=>{if(!e||e.length!==5)throw new Error("BatchNormalization requires 5 inputs");let t=(i,o,n)=>{let u=o.length;if(u!==i.length)throw new Error(`${n}: num dimensions != ${u}`);o.forEach((d,h)=>{if(d!==i[h])throw new Error(`${n}: dim[${h}] do not match`)})};if(e[0].dims.length>1){let i=r.format==="NHWC"?r.spatial?e[0].dims.slice(-1):e[0].dims.slice(-1).concat(e[0].dims.slice(1,e[0].dims.length-1)):e[0].dims.slice(1,r.spatial?2:void 0);t(e[1].dims,i,"Invalid input scale"),t(e[2].dims,i,"Invalid input B"),t(e[3].dims,i,"Invalid input mean"),t(e[4].dims,i,"Invalid input var")}else t(e[1].dims,[1],"Invalid input scale"),t(e[2].dims,[1],"Invalid input B"),t(e[3].dims,[1],"Invalid input mean"),t(e[4].dims,[1],"Invalid input var")},kg=(e,r)=>{let{epsilon:t,spatial:i,format:o}=r,n=e[0].dims,u=i?gr(n[n.length-1]):1,d=o==="NHWC"&&n.length>1?u:1,h=je.size(n)/u,f=i,y=f?n.length:n,m=Ne("x",e[0].dataType,e[0].dims,u),g=Ne("scale",e[1].dataType,e[1].dims,d),T=Ne("bias",e[2].dataType,e[2].dims,d),M=Ne("inputMean",e[3].dataType,e[3].dims,d),I=Ne("inputVar",e[4].dataType,e[4].dims,d),z=gt("y",e[0].dataType,y,u),E=()=>{let D="";if(i)D=`let cOffset = ${n.length===1?"0u":o==="NHWC"?`outputIndices[${n.length-1}] / ${u}`:"outputIndices[1]"};`;else if(o==="NCHW")D=` + ${z.indicesSet("outputIndices","0","0")} + let cOffset = ${z.indicesToOffset("outputIndices")};`;else{D=`var cIndices = ${g.type.indices}(0); + cIndices[0] = outputIndices[${n.length-1}];`;for(let A=1;A` + const epsilon = ${t}; + ${D.registerUniform("outputSize","u32").declareVariables(m,g,T,M,I,z)} + ${D.mainStart()} + ${D.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")} + var outputIndices = ${z.offsetToIndices(`global_idx * ${u}`)}; + ${E()} + let scale = ${g.getByOffset("cOffset")}; + let bias = ${T.getByOffset("cOffset")}; + let inputMean = ${M.getByOffset("cOffset")}; + let inputVar = ${I.getByOffset("cOffset")}; + let x = ${m.getByOffset("global_idx")}; + let value = (x - inputMean) * inverseSqrt(inputVar + epsilon) * scale + bias; + ${z.setByOffset("global_idx","value")} + }`;return{name:"BatchNormalization",shaderCache:{hint:`${r.epsilon}_${r.format}_${i}_${u}`,inputDependencies:f?["rank","type","type","type","type"]:void 0},getShaderSource:C,getRunData:()=>({outputs:[{dims:e[0].dims,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(h/64)},programUniforms:f?[{type:12,data:h},...Et(n)]:[{type:12,data:h}]})}},$g=e=>Zt(e),_b=(e,r)=>{let{inputs:t,outputCount:i}=e,o=$g({...r,outputCount:i});if(dr.webgpu.validateInputContent&&Sg(t,o),r.trainingMode)throw new Error("BatchNormalization trainingMode is not supported yet.");e.compute(kg(t,o))}}),Ig,Ag,gb,iT=Ye(()=>{Lt(),zt(),Ig=e=>{if(e[0].dims.length!==3)throw new Error("input should have 3 dimensions");if(![320,640,1280].includes(e[0].dims[2]))throw new Error("number of channels should be 320, 640 or 1280");if(e[1].dims.length!==1)throw new Error("bias is expected to have 1 dimensions");if(e[0].dims[2]!==e[1].dims[0])throw new Error("last dimension of input and bias are not the same")},Ag=e=>{let r=e[0].dims,t=e[0].dims[2],i=je.size(r)/4,o=e[0].dataType,n=Ne("input",o,r,4),u=Ne("bias",o,[t],4),d=Ne("residual",o,r,4),h=gt("output",o,r,4);return{name:"BiasAdd",getRunData:()=>({outputs:[{dims:r,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(i/64)}}),getShaderSource:f=>` + const channels = ${t}u / 4; + ${f.declareVariables(n,u,d,h)} + + ${f.mainStart()} + ${f.guardAgainstOutOfBoundsWorkgroupSizes(i)} + let value = ${n.getByOffset("global_idx")} + + ${u.getByOffset("global_idx % channels")} + ${d.getByOffset("global_idx")}; + ${h.setByOffset("global_idx","value")} + }`}},gb=e=>{Ig(e.inputs),e.compute(Ag(e.inputs))}}),Og,Xt,yb,wb,bb,Mb,vb,xb,Tb,Cb,Eb,Fg,Pb,Sb,kb,$b,md,Ib,Xh,Ab,Ob,Fb,Db,jb,Lb,zb,Bb,Rb,Nb,Vb,Wb,Ub,Gb,Kb,Hb,Xf,qb,Fm,Dm,Qb,Xb,Jb,Dg,jg,Yb,i_=Ye(()=>{$t(),Lt(),br(),zt(),Og=(e,r,t,i,o,n,u)=>{let d=Math.ceil(r/4),h="";typeof o=="string"?h=`${o}(a)`:h=o("a");let f=Ne("inputData",t,[d],4),y=gt("outputData",i,[d],4),m=[{name:"vec_size",type:"u32"}];return u&&m.push(...u),` + ${e.registerUniforms(m).declareVariables(f,y)} + + ${n??""} + + ${e.mainStart()} + ${e.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.vec_size")} + + let a = ${f.getByOffset("global_idx")}; + ${y.setByOffset("global_idx",h)} + }`},Xt=(e,r,t,i,o,n=e.dataType,u,d)=>{let h=[{type:12,data:Math.ceil(je.size(e.dims)/4)}];return u&&h.push(...u),{name:r,shaderCache:{hint:o,inputDependencies:["type"]},getShaderSource:f=>Og(f,je.size(e.dims),e.dataType,n,t,i,d),getRunData:f=>({outputs:[{dims:e.dims,dataType:n}],dispatchGroup:{x:Math.ceil(je.size(f[0].dims)/64/4)},programUniforms:h})}},yb=e=>{e.compute(Xt(e.inputs[0],"Abs","abs"))},wb=e=>{e.compute(Xt(e.inputs[0],"Acos","acos"))},bb=e=>{e.compute(Xt(e.inputs[0],"Acosh","acosh"))},Mb=e=>{e.compute(Xt(e.inputs[0],"Asin","asin"))},vb=e=>{e.compute(Xt(e.inputs[0],"Asinh","asinh"))},xb=e=>{e.compute(Xt(e.inputs[0],"Atan","atan"))},Tb=e=>{e.compute(Xt(e.inputs[0],"Atanh","atanh"))},Cb=e=>Zt(e),Eb=(e,r)=>{let t;switch(r.to){case 10:t="vec4";break;case 1:t="vec4";break;case 12:t="vec4";break;case 6:t="vec4";break;case 9:t="vec4";break;default:throw new RangeError(`not supported type (specified in attribute 'to' from 'Cast' operator): ${r.to}`)}e.compute(Xt(e.inputs[0],"Cast",t,void 0,r.cacheKey,r.to))},Fg=e=>{let r,t,i=e.length>=2&&e[1].data!==0,o=e.length>=3&&e[2].data!==0;switch(e[0].dataType){case 1:r=i?e[1].getFloat32Array()[0]:-34028234663852886e22,t=o?e[2].getFloat32Array()[0]:34028234663852886e22;break;case 10:r=i?e[1].getUint16Array()[0]:64511,t=o?e[2].getUint16Array()[0]:31743;break;default:throw new Error("Unsupport data type")}return Zt({min:r,max:t})},Pb=(e,r)=>{let t=r||Fg(e.inputs),i=Qr(e.inputs[0].dataType);e.compute(Xt(e.inputs[0],"Clip",o=>`clamp(${o}, vec4<${i}>(uniforms.min), vec4<${i}>(uniforms.max))`,void 0,t.cacheKey,void 0,[{type:e.inputs[0].dataType,data:t.min},{type:e.inputs[0].dataType,data:t.max}],[{name:"min",type:i},{name:"max",type:i}]),{inputs:[0]})},Sb=e=>{e.compute(Xt(e.inputs[0],"Ceil","ceil"))},kb=e=>{e.compute(Xt(e.inputs[0],"Cos","cos"))},$b=e=>{e.compute(Xt(e.inputs[0],"Cosh","cosh"))},md=e=>Zt(e),Ib=(e,r)=>{let t=Qr(e.inputs[0].dataType);e.compute(Xt(e.inputs[0],"Elu",i=>`elu_vf32(${i})`,` + const elu_alpha_ = ${t}(${r.alpha}); + + fn elu_f32(a: ${t}) -> ${t} { + return select((exp(a) - 1.0) * elu_alpha_, a, a >= 0.0); + } + + fn elu_vf32(v: vec4<${t}>) -> vec4<${t}> { + return vec4(elu_f32(v.x), elu_f32(v.y), elu_f32(v.z), elu_f32(v.w)); + }`,r.cacheKey))},Xh=(e="f32")=>` +const r0: ${e} = 0.3275911; +const r1: ${e} = 0.254829592; +const r2: ${e} = -0.284496736; +const r3: ${e} = 1.421413741; +const r4: ${e} = -1.453152027; +const r5: ${e} = 1.061405429; + +fn erf_vf32(v: vec4<${e}>) -> vec4<${e}> { + let absv = abs(v); + let x = 1.0 / (1.0 + r0 * absv); + return sign(v) * (1.0 - ((((r5 * x + r4) * x + r3) * x + r2) * x + r1) * x * exp(-absv * absv)); +}`,Ab=e=>{let r=Qr(e.inputs[0].dataType);e.compute(Xt(e.inputs[0],"Erf",t=>`erf_vf32(${t})`,Xh(r)))},Ob=e=>{e.compute(Xt(e.inputs[0],"Exp","exp"))},Fb=e=>{e.compute(Xt(e.inputs[0],"Floor","floor"))},Db=e=>{let r=Qr(e.inputs[0].dataType);e.compute(Xt(e.inputs[0],"Gelu",t=>`0.5 * ${t} * (1.0 + erf_vf32(${t} * 0.7071067811865475))`,Xh(r)))},jb=(e,r)=>{let t=Qr(e.inputs[0].dataType);e.compute(Xt(e.inputs[0],"LeakyRelu",i=>`select(leaky_relu_alpha_ * ${i}, ${i}, ${i} >= vec4<${t}>(0.0))`,`const leaky_relu_alpha_ = ${t}(${r.alpha});`,r.cacheKey))},Lb=e=>{e.compute(Xt(e.inputs[0],"Not",r=>`!${r}`))},zb=e=>{e.compute(Xt(e.inputs[0],"Neg",r=>`-${r}`))},Bb=e=>{e.compute(Xt(e.inputs[0],"Reciprocal",r=>`1.0/${r}`))},Rb=e=>{let r=Qr(e.inputs[0].dataType);e.compute(Xt(e.inputs[0],"Relu",t=>`select(vec4<${r}>(0.0), ${t}, ${t} > vec4<${r}>(0.0))`))},Nb=e=>{e.compute(Xt(e.inputs[0],"Sigmoid",r=>`(1.0 / (1.0 + exp(-${r})))`))},Vb=e=>Zt(e),Wb=(e,r)=>{let t=Qr(e.inputs[0].dataType);e.compute(Xt(e.inputs[0],"HardSigmoid",i=>`max(vec4<${t}>(0.0), min(vec4<${t}>(1.0), ${r.alpha} * ${i} + vec4<${t}>(${r.beta})))`,void 0,r.cacheKey))},Ub=e=>{e.compute(Xt(e.inputs[0],"Sin","sin"))},Gb=e=>{e.compute(Xt(e.inputs[0],"Sinh","sinh"))},Kb=e=>{e.compute(Xt(e.inputs[0],"Sqrt","sqrt"))},Hb=e=>{e.compute(Xt(e.inputs[0],"Tan","tan"))},Xf=e=>`sign(${e}) * (1 - exp(-2 * abs(${e}))) / (1 + exp(-2 * abs(${e})))`,qb=e=>{e.compute(Xt(e.inputs[0],"Tanh",Xf))},Fm=(e="f32")=>` +const fast_gelu_a: ${e} = 0.5; +const fast_gelu_b: ${e} = 0.7978845608028654; +const fast_gelu_c: ${e} = 0.035677408136300125; + +fn tanh_v(v: vec4<${e}>) -> vec4<${e}> { + return ${Xf("v")}; +} +`,Dm=e=>`(fast_gelu_a + fast_gelu_a * tanh_v(${e} * (fast_gelu_c * ${e} * ${e} + fast_gelu_b))) * ${e}`,Qb=e=>{let r=Qr(e.inputs[0].dataType);e.compute(Xt(e.inputs[0],"FastGelu",Dm,Fm(r),void 0,e.inputs[0].dataType))},Xb=(e,r)=>{let t=Qr(e.inputs[0].dataType);return e.compute(Xt(e.inputs[0],"ThresholdedRelu",i=>`select(vec4<${t}>(0.0), ${i}, ${i} > thresholded_relu_alpha_)`,`const thresholded_relu_alpha_ = vec4<${t}>(${r.alpha});`,r.cacheKey)),0},Jb=e=>{e.compute(Xt(e.inputs[0],"Log","log"))},Dg=(e,r)=>` +const alpha = vec4<${e}>(${r}); +const one = ${e}(1.0); +const zero = ${e}(0.0); + +fn quick_gelu_impl(x: vec4<${e}>) -> vec4<${e}> { + let v = x *alpha; + var x1 : vec4<${e}>; + for (var i = 0; i < 4; i = i + 1) { + if (v[i] >= zero) { + x1[i] = one / (one + exp(-v[i])); + } else { + x1[i] = one - one / (one + exp(v[i])); + } + } + return x * x1; +} +`,jg=e=>`quick_gelu_impl(${e})`,Yb=(e,r)=>{let t=Qr(e.inputs[0].dataType);e.compute(Xt(e.inputs[0],"QuickGelu",jg,Dg(t,r.alpha),r.cacheKey,e.inputs[0].dataType))}}),Lg,zg,Zb,aT=Ye(()=>{Lt(),zt(),i_(),Lg=e=>{if(e[0].dims.length!==3)throw new Error("input should have 3 dimensions");if(![2560,5120,10240].includes(e[0].dims[2]))throw new Error("hidden state should be 2560, 5120 or 10240");if(e[1].dims.length!==1)throw new Error("bias is expected to have 1 dimensions");if(e[0].dims[2]!==e[1].dims[0])throw new Error("last dimension of input and bias are not the same")},zg=e=>{let r=e[0].dims.slice();r[2]=r[2]/2;let t=Ne("input",e[0].dataType,e[0].dims,4),i=Ne("bias",e[0].dataType,[e[0].dims[2]],4),o=gt("output",e[0].dataType,r,4),n=je.size(r)/4,u=Lr(e[0].dataType);return{name:"BiasSplitGelu",getRunData:()=>({outputs:[{dims:r,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(n/64)}}),getShaderSource:d=>` + const M_SQRT2 = sqrt(2.0); + const halfChannels = ${e[0].dims[2]/4/2}u; + + ${d.declareVariables(t,i,o)} + + ${Xh(u)} + + ${d.mainStart()} + ${d.guardAgainstOutOfBoundsWorkgroupSizes(n)} + let biasIdx = global_idx % halfChannels; + let batchIndex = global_idx / halfChannels; + let inputOffset = biasIdx + batchIndex * halfChannels * 2; + let valueLeft = input[inputOffset] + bias[biasIdx]; + let valueRight = input[inputOffset + halfChannels] + bias[biasIdx + halfChannels]; + let geluRight = valueRight * 0.5 * (erf_vf32(valueRight / M_SQRT2) + 1); + + ${o.setByOffset("global_idx","valueLeft * geluRight")} + }`}},Zb=e=>{Lg(e.inputs),e.compute(zg(e.inputs))}}),Bg,Rg,Ls,eM,tM,rM,sM,nM,iM,aM,oM,lM,cM,oT=Ye(()=>{$t(),Lt(),zt(),Bg=(e,r,t,i,o,n,u,d,h,f,y,m)=>{let g,T;typeof d=="string"?g=T=(C,D)=>`${d}((${C}),(${D}))`:typeof d=="function"?g=T=d:(g=d.scalar,T=d.vector);let M=gt("outputData",y,i.length,4),I=Ne("aData",h,r.length,4),z=Ne("bData",f,t.length,4),E;if(o)if(n){let C=je.size(r)===1,D=je.size(t)===1,A=r.length>0&&r[r.length-1]%4===0,$=t.length>0&&t[t.length-1]%4===0;C||D?E=M.setByOffset("global_idx",T(C?`${I.type.value}(${I.getByOffset("0")}.x)`:I.getByOffset("global_idx"),D?`${z.type.value}(${z.getByOffset("0")}.x)`:z.getByOffset("global_idx"))):E=` + let outputIndices = ${M.offsetToIndices("global_idx * 4u")}; + let offsetA = ${I.broadcastedIndicesToOffset("outputIndices",M)}; + let offsetB = ${z.broadcastedIndicesToOffset("outputIndices",M)}; + ${M.setByOffset("global_idx",T(u||A?I.getByOffset("offsetA / 4u"):`${I.type.value}(${I.getByOffset("offsetA / 4u")}[offsetA % 4u])`,u||$?z.getByOffset("offsetB / 4u"):`${z.type.value}(${z.getByOffset("offsetB / 4u")}[offsetB % 4u])`))} + `}else E=M.setByOffset("global_idx",T(I.getByOffset("global_idx"),z.getByOffset("global_idx")));else{if(!n)throw new Error("no necessary to use scalar implementation for element-wise binary op implementation.");let C=(D,A,$="")=>{let P=`aData[indexA${A}][componentA${A}]`,k=`bData[indexB${A}][componentB${A}]`;return` + let outputIndices${A} = ${M.offsetToIndices(`global_idx * 4u + ${A}u`)}; + let offsetA${A} = ${I.broadcastedIndicesToOffset(`outputIndices${A}`,M)}; + let offsetB${A} = ${z.broadcastedIndicesToOffset(`outputIndices${A}`,M)}; + let indexA${A} = offsetA${A} / 4u; + let indexB${A} = offsetB${A} / 4u; + let componentA${A} = offsetA${A} % 4u; + let componentB${A} = offsetB${A} % 4u; + ${D}[${A}] = ${$}(${g(P,k)}); + `};y===9?E=` + var data = vec4(0); + ${C("data",0,"u32")} + ${C("data",1,"u32")} + ${C("data",2,"u32")} + ${C("data",3,"u32")} + outputData[global_idx] = dot(vec4(0x1, 0x100, 0x10000, 0x1000000), vec4(data));`:E=` + ${C("outputData[global_idx]",0)} + ${C("outputData[global_idx]",1)} + ${C("outputData[global_idx]",2)} + ${C("outputData[global_idx]",3)} + `}return` + ${e.registerUniform("vec_size","u32").declareVariables(I,z,M)} + + ${m??""} + + ${e.mainStart()} + ${e.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.vec_size")} + ${E} + }`},Rg=(e,r,t,i,o,n,u=t.dataType)=>{let d=t.dims.map(I=>Number(I)??1),h=i.dims.map(I=>Number(I)??1),f=!je.areEqual(d,h),y=d,m=je.size(d),g=!1,T=!1,M=[f];if(f){let I=Na.calcShape(d,h,!1);if(!I)throw new Error("Can't perform binary op on the given tensors");y=I.slice(),m=je.size(y);let z=je.size(d)===1,E=je.size(h)===1,C=d.length>0&&d[d.length-1]%4===0,D=h.length>0&&h[h.length-1]%4===0;M.push(z),M.push(E),M.push(C),M.push(D);let A=1;for(let $=1;$I.toString()).join("_"),inputDependencies:["rank","rank"]},getShaderSource:I=>Bg(I,d,h,y,g,f,T,o,t.dataType,i.dataType,u,n),getRunData:()=>({outputs:[{dims:y,dataType:u}],dispatchGroup:{x:Math.ceil(m/64/4)},programUniforms:[{type:12,data:Math.ceil(je.size(y)/4)},...Et(d,h,y)]})}},Ls=(e,r,t,i,o,n)=>{e.compute(Rg(r,o??"",e.inputs[0],e.inputs[1],t,i,n))},eM=e=>{Ls(e,"Add",(r,t)=>`${r}+${t}`)},tM=e=>{Ls(e,"Div",(r,t)=>`${r}/${t}`)},rM=e=>{Ls(e,"Equal",{scalar:(r,t)=>`u32(${r}==${t})`,vector:(r,t)=>`vec4(${r}==${t})`},void 0,void 0,9)},sM=e=>{Ls(e,"Mul",(r,t)=>`${r}*${t}`)},nM=e=>{let r=Ne("input",e.inputs[0].dataType,e.inputs[0].dims).type.value;Ls(e,"Pow",{scalar:(t,i)=>`pow_custom(${t},${i})`,vector:(t,i)=>`pow_vector_custom(${t},${i})`},` + fn pow_custom(a : ${r}, b : ${r}) -> ${r} { + if (b == ${r}(0.0)) { + return ${r}(1.0); + } else if (a < ${r}(0.0) && f32(b) != floor(f32(b))) { + return ${r}(pow(f32(a), f32(b))); // NaN + } + return select(sign(a), ${r}(1.0), round(f32(abs(b) % ${r}(2.0))) != 1.0) * ${r}(${r==="i32"?"round":""}(pow(f32(abs(a)), f32(b)))); + } + fn pow_vector_custom(a : vec4<${r}>, b : vec4<${r}>) -> vec4<${r}> { + // TODO: implement vectorized pow + return vec4<${r}>(pow_custom(a.x, b.x), pow_custom(a.y, b.y), pow_custom(a.z, b.z), pow_custom(a.w, b.w)); + } + `)},iM=e=>{Ls(e,"Sub",(r,t)=>`${r}-${t}`)},aM=e=>{Ls(e,"Greater",{scalar:(r,t)=>`u32(${r}>${t})`,vector:(r,t)=>`vec4(${r}>${t})`},void 0,void 0,9)},oM=e=>{Ls(e,"Less",{scalar:(r,t)=>`u32(${r}<${t})`,vector:(r,t)=>`vec4(${r}<${t})`},void 0,void 0,9)},lM=e=>{Ls(e,"GreaterOrEqual",{scalar:(r,t)=>`u32(${r}>=${t})`,vector:(r,t)=>`vec4(${r}>=${t})`},void 0,void 0,9)},cM=e=>{Ls(e,"LessOrEqual",{scalar:(r,t)=>`u32(${r}<=${t})`,vector:(r,t)=>`vec4(${r}<=${t})`},void 0,void 0,9)}}),Ng,Vg,Wg,Ug,uM,dM,lT=Ye(()=>{$t(),Lt(),br(),zt(),Ng=(e,r)=>{if(!e||e.length<1)throw new Error("too few inputs");let t=0,i=e[t],o=i.dataType,n=i.dims.length;e.forEach((u,d)=>{if(d!==t){if(u.dataType!==o)throw new Error("input tensors should be one type");if(u.dims.length!==n)throw new Error("input tensors should have the same shape");u.dims.forEach((h,f)=>{if(f!==r&&h!==i.dims[f])throw new Error("non concat dimensions must match")})}})},Vg=(e,r)=>` + fn calculateInputIndex(index: u32) -> u32 { + let sizeInConcatAxis = array(${r}); + for (var i: u32 = 0u; i < ${e}; i += 1u ) { + if (index < sizeInConcatAxis[i]) { + return i; + } + } + return ${e}u; + }`,Wg=(e,r)=>{let t=e.length,i=[];for(let o=0;o{let o=je.size(t),n=new Array(e.length),u=new Array(e.length),d=0,h=[],f=[],y=[{type:12,data:o}];for(let I=0;I`uniforms.sizeInConcatAxis${I}`).join(","),M=I=>` + + ${(()=>{I.registerUniform("outputSize","u32");for(let z=0;z(${T}); + ${g} -= sizeInConcatAxis[inputIndex - 1u]; + } + + ${Wg(u,m)} + }`;return{name:"Concat",shaderCache:{hint:`${r}`,inputDependencies:h},getRunData:()=>({outputs:[{dims:t,dataType:i}],dispatchGroup:{x:Math.ceil(o/64)},programUniforms:y}),getShaderSource:M}},uM=(e,r)=>{let t=e.inputs,i=t[0].dims,o=je.normalizeAxis(r.axis,i.length);Ng(t,o);let n=i.slice();n[o]=t.reduce((d,h)=>d+(h.dims.length>o?h.dims[o]:0),0);let u=t.filter(d=>je.size(d.dims)>0);e.compute(Ug(u,o,n,t[0].dataType),{inputs:u})},dM=e=>Zt({axis:e.axis})}),mi,_i,gi,a_,wi=Ye(()=>{$t(),Lt(),mi=(e,r,t="f32")=>{switch(e.activation){case"Relu":return`value = max(value, ${r}(0.0));`;case"Sigmoid":return`value = (${r}(1.0) / (${r}(1.0) + exp(-value)));`;case"Clip":return`value = clamp(value, ${r}(${t}(uniforms.clip_min)), ${r}(${t}(uniforms.clip_max)));`;case"HardSigmoid":return`value = max(${r}(0.0), min(${r}(1.0), ${t}(uniforms.alpha) * value + ${t}(uniforms.beta)));`;case"LeakyRelu":return`value = select(${t}(uniforms.alpha) * value, value, value >= ${r}(0.0));`;case"Tanh":return`let e2x = exp(-2.0 * abs(value)); + value = sign(value) * (1.0 - e2x) / (1.0 + e2x); + `;case"":return"";default:throw new Error(`Unsupported activation ${e.activation}`)}},_i=(e,r)=>{e.activation==="Clip"?r.push({type:1,data:e.clipMax},{type:1,data:e.clipMin}):e.activation==="HardSigmoid"?r.push({type:1,data:e.alpha},{type:1,data:e.beta}):e.activation==="LeakyRelu"&&r.push({type:1,data:e.alpha})},gi=(e,r)=>{e.activation==="Clip"?r.push({name:"clip_max",type:"f32"},{name:"clip_min",type:"f32"}):e.activation==="HardSigmoid"?r.push({name:"alpha",type:"f32"},{name:"beta",type:"f32"}):e.activation==="LeakyRelu"&&r.push({name:"alpha",type:"f32"})},a_=e=>{let r=(e==null?void 0:e.activation)||"";if(r==="HardSigmoid"){let[t,i]=(e==null?void 0:e.activation_params)||[.2,.5];return{activation:r,alpha:t,beta:i}}else if(r==="Clip"){let[t,i]=(e==null?void 0:e.activation_params)||[L0,z0];return{activation:r,clipMax:i,clipMin:t}}else if(r==="LeakyRelu"){let[t]=(e==null?void 0:e.activation_params)||[.01];return{activation:r,alpha:t}}return{activation:r}}}),Vr,pM,o_=Ye(()=>{Vr=(e,r)=>{switch(e){case 1:return r;case 2:return`vec2<${r}>`;case 3:return`vec3<${r}>`;case 4:return`vec4<${r}>`;default:throw new Error(`${e}-component is not supported.`)}},pM=e=>` + ${e?"value = value + getBiasByOutputCoords(coords);":""} + `}),hM,cT=Ye(()=>{hM=e=>` +fn getIndexFromCoords4D(coords : vec4, shape : vec4) -> i32 { + return dot(coords, vec4( + shape.y * shape.z * shape.w, shape.z * shape.w, shape.w, 1)); +} +fn getOutputIndexFromCoords(coords : vec4) -> i32 { + return dot(coords, vec4( + i32(${e}.x), i32(${e}.y), i32(${e}.z), 1)); +} +`}),gd,l_,c_=Ye(()=>{$t(),Lt(),zt(),wi(),gd=(e,r,t,i,o)=>{let n=i-t;return` + ${Array.from({length:t}).map((u,d)=>` + if (${Mt(r.shape,d,r.rank)} != 1) { + ${r.indicesSet(e,d,Mt(o,d+n,i))} + } else { + ${r.indicesSet(e,d,0)} + }`).join("")} +`},l_=(e,r,t,i,o=!1,n)=>{let u=e[0].dims,d=e[1].dims,h=u[u.length-2],f=d[d.length-1],y=u[u.length-1],m=gr(f),g=gr(y),T=gr(h),M=je.size(t)/m/T,I=e.length>2,z=i?i.slice(0,-2):t.slice(0,-2),E=[je.size(z),h,f],C=[{type:12,data:M},{type:12,data:h},{type:12,data:f},{type:12,data:y}];_i(r,C),C.push(...Et(z,u,d)),I&&C.push(...Et(e[2].dims)),C.push(...Et(E));let D=A=>{let $=r_("batch_dims",e[0].dataType,z.length),P=Ne("a",e[0].dataType,u.length,g),k=Ne("b",e[1].dataType,d.length,m),O=gt("output",e[0].dataType,E.length,m),R=Lr(O.type.tensor),U=mi(r,O.type.value,R),te=[P,k],se="";if(I){let re=o?m:1;te.push(Ne("bias",e[2].dataType,e[2].dims.length,re)),se=`${o?`value += bias[col / ${re}];`:`value += ${O.type.value}(bias[row + i]);`}`}let K=[{name:"output_size",type:"u32"},{name:"M",type:"u32"},{name:"N",type:"u32"},{name:"K",type:"u32"}];gi(r,K);let pe=()=>{let re=`var a_data: ${P.type.value};`;for(let oe=0;oe; + for (var k: u32 = 0u; k < uniforms.K; k = k + ${g}) { + ${pe()} + } + for (var i = 0u; i < ${T}u; i++) { + var value = values[i]; + ${se} + ${U} + let cur_indices = ${O.type.indices}(batch, row + i, col); + let offset = ${O.indicesToOffset("cur_indices")}; + ${O.setByOffset(`offset / ${m}`,"value")}; + } + } + `};return{name:"MatMulNaive",shaderCache:{hint:`${r.activation};${m};${g};${T};${o}`,inputDependencies:I?["rank","rank","rank"]:["rank","rank"]},getRunData:()=>({outputs:[{dims:n?n(t):t,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(M/64)},programUniforms:C}),getShaderSource:D}}}),Gg,Kg,jm,Jf,Hg,Lm,qg,rf,u_=Ye(()=>{$t(),Lt(),zt(),wi(),c_(),o_(),Gg=(e,r)=>e?` + mm_Asub[inputRow][inputCol] = mm_readA(batch, + kStart + inputRow, + globalRowStart / innerElementSize + inputCol${r?", batchIndices":""}); + `:` + mm_Asub[inputRow][inputCol] = mm_readA(batch, + globalRow + innerRow, + kStart / innerElementSize + inputCol${r?", batchIndices":""}); + `,Kg=(e,r)=>e?` + let ACached0 = mm_Asub[k * innerElementSize][localRow]; + let ACached1 = mm_Asub[k * innerElementSize + 1][localRow]; + let ACached2 = mm_Asub[k * innerElementSize + 2][localRow]; + ${r===3?"":"let ACached3 = mm_Asub[k * innerElementSize + 3][localRow];"} + for (var i = 0; i < rowPerThread; i = i + 1) { + acc[i] = BCached0 * ACached0[i] + acc[i]; + acc[i] = BCached1 * ACached1[i] + acc[i]; + acc[i] = BCached2 * ACached2[i] + acc[i]; + ${r===3?"":"acc[i] = BCached3 * ACached3[i] + acc[i];"} + }`:` + for (var i = 0; i < rowPerThread; i = i + 1) { + let ACached = mm_Asub[tileRow + i][k]; + acc[i] = BCached0 * ACached.x + acc[i]; + acc[i] = BCached1 * ACached.y + acc[i]; + acc[i] = BCached2 * ACached.z + acc[i]; + ${r===3?"":"acc[i] = BCached3 * ACached.w + acc[i];"} + }`,jm=(e,r,t="f32",i,o=!1,n=32,u=!1,d=32)=>{let h=r[1]*e[1],f=r[0]*e[0],y=o?h:n,m=o?n:h,g=y/r[0],T=n/r[1];if(!((o&&g===4&&e[1]===4||!o&&(g===3||g===4))&&y%r[0]===0&&n%r[1]===0&&e[0]===4))throw new Error(`If transposeA ${o} is true, innerElementSize ${g} and workPerThread[1] ${e[1]} must be 4. + Otherwise, innerElementSize ${g} must be 3 or 4. + tileAWidth ${y} must be divisible by workgroupSize[0]${r[0]}. tileInner ${n} must be divisible by workgroupSize[1] ${r[1]}. colPerThread ${e[0]} must be 4.`);return` +var mm_Asub: array, ${y/g}>, ${m}>; +var mm_Bsub: array, ${f/e[0]}>, ${n}>; + +const rowPerThread = ${e[1]}; +const colPerThread = ${e[0]}; +const innerElementSize = ${g}; +const tileInner = ${n}; + +@compute @workgroup_size(${r[0]}, ${r[1]}, ${r[2]}) +fn main(@builtin(local_invocation_id) localId : vec3, + @builtin(global_invocation_id) globalId : vec3, + @builtin(workgroup_id) workgroupId : vec3) { + let localRow = i32(localId.y); + let tileRow = localRow * rowPerThread; + let tileCol = i32(localId.x); + + let globalRow =i32(globalId.y) * rowPerThread; + let globalCol = i32(globalId.x); + let batch = ${u?"0":"i32(globalId.z)"}; + ${i?`let batchIndices = ${i.offsetToIndices("u32(batch)")};`:""} + let globalRowStart = i32(workgroupId.y) * ${h}; + + let num_tiles = ${u?`${Math.ceil(d/n)}`:"(uniforms.dim_inner - 1) / tileInner + 1"}; + var kStart = ${u?`i32(globalId.z) * ${d}`:"0"}; + + var acc: array, rowPerThread>; + + // Loop over shared dimension. + let tileRowB = localRow * ${T}; + for (var t = 0; t < num_tiles; t = t + 1) { + // Load one tile of A into local memory. + for (var innerRow = 0; innerRow < rowPerThread; innerRow = innerRow + 1) { + let inputRow = tileRow + innerRow; + let inputCol = tileCol; + ${Gg(o,i)} + } + + // Load one tile of B into local memory. + for (var innerRow = 0; innerRow < ${T}; innerRow = innerRow + 1) { + let inputRow = tileRowB + innerRow; + let inputCol = tileCol; + mm_Bsub[inputRow][inputCol] = mm_readB(batch, kStart + inputRow, globalCol${i?", batchIndices":""}); + } + kStart = kStart + tileInner; + workgroupBarrier(); + + // Compute acc values for a single thread. + for (var k = 0; k < tileInner / innerElementSize; k = k + 1) { + let BCached0 = mm_Bsub[k * innerElementSize][tileCol]; + let BCached1 = mm_Bsub[k * innerElementSize + 1][tileCol]; + let BCached2 = mm_Bsub[k * innerElementSize + 2][tileCol]; + ${g===3?"":"let BCached3 = mm_Bsub[k * innerElementSize + 3][tileCol];"} + + ${Kg(o,g)} + } + + workgroupBarrier(); + } + + for (var innerRow = 0; innerRow < rowPerThread; innerRow = innerRow + 1) { + mm_write(batch, globalRow + innerRow, globalCol, acc[innerRow]); + } +}`},Jf=(e,r)=>e?` + mm_Asub[inputRow][inputCol] = mm_readA(batch, + kStart + inputRow, + globalRowStart + inputCol${r?", batchIndices":""}); + `:` + mm_Asub[inputRow][inputCol] = mm_readA(batch, + globalRowStart + inputRow, + kStart + inputCol${r?", batchIndices":""}); + `,Hg=e=>e?"let ACached = mm_Asub[k][tileRow + innerRow];":"let ACached = mm_Asub[tileRow + innerRow][k];",Lm=(e,r,t="f32",i,o=!1,n=32,u=!1,d=32,h=!1)=>{let f=e[1]*r[1],y=e[0]*r[0],m=o?f:n,g=o?n:f;if(!(g%r[1]===0&&m%r[0]===0&&n%r[1]===0))throw new Error(`tileAHight ${g} must be divisible by workgroupSize[1]${r[1]}, tileAWidth ${m} must be divisible by workgroupSize[0]${r[0]}, tileInner ${n} must be divisible by workgroupSize[1]${r[1]}`);let T=g/r[1],M=m/r[0],I=n/r[1],z=h?` + let localRow = i32(localId.y); + let localCol = i32(localId.x); + let globalRowStart = i32(workgroupId.y) * ${f}; + let globalColStart = i32(workgroupId.x) * ${y}; + + // Loop over shared dimension. + for (var t = 0; t < num_tiles; t = t + 1) { + // Load one tile of A into local memory. + for (var inputRow = localRow; inputRow < ${g}; inputRow = inputRow + ${r[1]}) { + for (var inputCol = localCol; inputCol < ${m}; inputCol = inputCol + ${r[0]}) { + ${Jf(o,i)} + } + } + // Load one tile of B into local memory. + for (var inputRow = localRow; inputRow < ${n}; inputRow = inputRow + ${r[1]}) { + for (var inputCol = localCol; inputCol < ${y}; inputCol = inputCol + ${r[0]}) { + mm_Bsub[inputRow][inputCol] = mm_readB(batch, + kStart + inputRow, + globalColStart + inputCol${i?", batchIndices":""}); + } + } + kStart = kStart + tileInner; + workgroupBarrier(); + + // Compute acc values for a single thread. + var BCached : array<${t}, colPerThread>; + for (var k = 0; k < tileInner; k = k + 1) { + for (var inner = 0; inner < colPerThread; inner = inner + 1) { + BCached[inner] = mm_Bsub[k][localCol + inner * ${r[0]}]; + } + for (var innerRow = 0; innerRow < rowPerThread; innerRow = innerRow + 1) { + let ACached = ${o?`mm_Asub[k][localRow + innerRow * ${r[1]}];`:`mm_Asub[localRow + innerRow * ${r[1]}][k];`} + for (var innerCol = 0; innerCol < colPerThread; innerCol = innerCol + 1) { + acc[innerRow][innerCol] = acc[innerRow][innerCol] + + ACached * BCached[innerCol]; + } + } + } + workgroupBarrier(); + } + for (var innerRow = 0; innerRow < rowPerThread; innerRow = innerRow + 1) { + let gRow = globalRowStart + localRow + innerRow * ${r[1]}; + for (var innerCol = 0; innerCol < colPerThread; innerCol = innerCol + 1) { + let gCol = globalColStart + localCol + innerCol * ${r[0]}; + mm_write(batch, gRow, gCol, acc[innerRow][innerCol]); + } + } + `:` +let tileRow = i32(localId.y) * rowPerThread; +let tileCol = i32(localId.x) * colPerThread; + +let globalRow = i32(globalId.y) * rowPerThread; +let globalCol = i32(globalId.x) * colPerThread; +let globalRowStart = i32(workgroupId.y) * ${f}; + +let tileRowA = i32(localId.y) * ${T}; +let tileColA = i32(localId.x) * ${M}; +let tileRowB = i32(localId.y) * ${I}; +// Loop over shared dimension. +for (var t = 0; t < num_tiles; t = t + 1) { + // Load one tile of A into local memory. + for (var innerRow = 0; innerRow < ${T}; innerRow = innerRow + 1) { + for (var innerCol = 0; innerCol < ${M}; innerCol = innerCol + 1) { + let inputRow = tileRowA + innerRow; + let inputCol = tileColA + innerCol; + ${Jf(o,i)} + } + } + + // Load one tile of B into local memory. + for (var innerRow = 0; innerRow < ${I}; innerRow = innerRow + 1) { + for (var innerCol = 0; innerCol < colPerThread; innerCol = innerCol + 1) { + let inputRow = tileRowB + innerRow; + let inputCol = tileCol + innerCol; + mm_Bsub[inputRow][inputCol] = mm_readB(batch, + kStart + inputRow, + globalCol + innerCol${i?", batchIndices":""}); + } + } + kStart = kStart + tileInner; + workgroupBarrier(); + + // Compute acc values for a single thread. + var BCached : array<${t}, colPerThread>; + for (var k = 0; k < tileInner; k = k + 1) { + for (var inner = 0; inner < colPerThread; inner = inner + 1) { + BCached[inner] = mm_Bsub[k][tileCol + inner]; + } + + for (var innerRow = 0; innerRow < rowPerThread; innerRow = innerRow + 1) { + ${Hg(o)} + for (var innerCol = 0; innerCol < colPerThread; innerCol = innerCol + 1) { + acc[innerRow][innerCol] = acc[innerRow][innerCol] + ACached * BCached[innerCol]; + } + } + } + + workgroupBarrier(); +} + +for (var innerRow = 0; innerRow < rowPerThread; innerRow = innerRow + 1) { + for (var innerCol = 0; innerCol < colPerThread; innerCol = innerCol + 1) { + mm_write(batch, globalRow + innerRow, globalCol + innerCol, + acc[innerRow][innerCol]); + } +} +`;return` + var mm_Asub : array, ${g}>; + var mm_Bsub : array, ${n}>; + const rowPerThread = ${e[1]}; + const colPerThread = ${e[0]}; + const tileInner = ${n}; + +@compute @workgroup_size(${r[0]}, ${r[1]}, ${r[2]}) +fn main(@builtin(local_invocation_id) localId : vec3, + @builtin(global_invocation_id) globalId : vec3, + @builtin(workgroup_id) workgroupId : vec3) { + let batch = ${u?"0":"i32(globalId.z)"}; + ${i?`let batchIndices = ${i.offsetToIndices("u32(batch)")};`:""} + let num_tiles = ${u?`${Math.ceil(d/n)}`:"(uniforms.dim_inner - 1) / tileInner + 1"}; + var kStart = ${u?`i32(globalId.z) * ${d}`:"0"}; + + var acc : array, rowPerThread>; + ${z} + } +`},qg=(e,r,t,i,o=!1)=>{let[n,u,d,h]=i,f=Lr(i[0].type.tensor);return` + fn mm_readA(batch: i32, row: i32, colIn: i32, batchIndices: ${n.type.indices}) -> ${Vr(e,f)} { + var value = ${Vr(e,f)}(0.0); + let col = colIn * ${e}; + if(row < uniforms.dim_a_outer && col < uniforms.dim_inner) + { + var aIndices: ${u.type.indices}; + ${gd("aIndices",u,u.rank-2,n.rank,"batchIndices")} + ${u.indicesSet("aIndices",u.rank-2,"u32(row)")} + ${u.indicesSet("aIndices",u.rank-1,"u32(colIn)")} + value = ${u.getByIndices("aIndices")}; + } + return value; + } + + fn mm_readB(batch: i32, row: i32, colIn: i32, batchIndices: ${n.type.indices}) -> ${Vr(e,f)} { + var value = ${Vr(e,f)}(0.0); + let col = colIn * ${e}; + if(row < uniforms.dim_inner && col < uniforms.dim_b_outer) + { + var bIndices: ${d.type.indices}; + ${gd("bIndices",d,d.rank-2,n.rank,"batchIndices")} + ${d.indicesSet("bIndices",d.rank-2,"u32(row)")} + ${d.indicesSet("bIndices",d.rank-1,"u32(colIn)")} + value = ${d.getByIndices("bIndices")}; + } + return value; + } + + fn mm_write(batch: i32, row: i32, colIn: i32, valueIn: ${Vr(e,f)}) { + let col = colIn * ${e}; + if (row < uniforms.dim_a_outer && col < uniforms.dim_b_outer) { + var value = valueIn; + let coords = vec3(batch, row, colIn); + ${r?`value = value + ${o?"bias[colIn]":`${Vr(e,f)}(bias[row])`};`:""} + ${t} + ${h.setByIndices("vec3(coords)","value")} + } + } + `},rf=(e,r,t,i,o=!1,n)=>{let u=e[0].dims,d=e[1].dims,h=u.slice(0,-2),f=d.slice(0,-2),y=i?i.slice(0,-2):t.slice(0,-2),m=je.size(y),g=u[u.length-2],T=u[u.length-1],M=d[d.length-1],I=T%4===0&&M%4===0,z=g<=8?[4,1,1]:[4,4,1],E=[8,8,1],C=[Math.ceil(M/E[0]/z[0]),Math.ceil(g/E[1]/z[1]),Math.ceil(m/E[2]/z[2])],D=I?4:1,A=[...h,g,T/D],$=A.length,P=[...f,T,M/D],k=P.length,O=[m,g,M/D],R=[{type:6,data:g},{type:6,data:M},{type:6,data:T}];_i(r,R),R.push(...Et(y,A,P));let U=["rank","rank"],te=e.length>2;te&&(R.push(...Et(e[2].dims)),U.push("rank")),R.push(...Et(O));let se=K=>{let pe=y.length,re=r_("batchDims",e[0].dataType,pe,1),oe=Lr(e[0].dataType),ge=Ne("a",e[0].dataType,$,D),le=Ne("b",e[1].dataType,k,D),Se=gt("result",e[0].dataType,O.length,D),Ce=[ge,le];if(te){let Ae=o?D:1;Ce.push(Ne("bias",e[2].dataType,e[2].dims.length,Ae))}let q=[{name:"dim_a_outer",type:"i32"},{name:"dim_b_outer",type:"i32"},{name:"dim_inner",type:"i32"}];gi(r,q);let N=Lr(Se.type.tensor),Q=mi(r,Se.type.value,N),ue=qg(D,te,Q,[re,ge,le,Se],o);return` + ${K.registerUniforms(q).registerInternalVariables(re).declareVariables(...Ce,Se)} + ${ue} + ${I?jm(z,E,oe,re):Lm(z,E,oe,re)} + `};return{name:"MatMul",shaderCache:{hint:`${z};${r.activation};${I};${o}`,inputDependencies:U},getRunData:()=>({outputs:[{dims:n?n(t):t,dataType:e[0].dataType}],dispatchGroup:{x:C[0],y:C[1],z:C[2]},programUniforms:R}),getShaderSource:se}}}),Qg,fM,uT=Ye(()=>{$t(),mn(),zt(),wi(),o_(),cT(),u_(),Qg=(e,r,t,i,o=!1,n,u=4,d=4,h=4,f="f32")=>{let y=R=>{switch(R){case 1:return"resData = x[xIndex];";case 3:return`resData = vec3<${f}>(x[xIndex], x[xIndex + 1], x[xIndex + 2]);`;case 4:return"resData = x[xIndex / 4];";default:throw new Error(`innerElementSize ${R} is not supported.`)}},m=R=>{switch(R){case 1:return"return w[row * i32(uniforms.w_shape[3]) + colIn];";case 4:return"return w[row * i32(uniforms.w_shape[3]) / 4 + colIn];";default:throw new Error(`innerElementSize ${R} is not supported.`)}},g=e?` + let coord = vec4(batch, xRow, xCol, xCh); + `:` + let coord = vec4(batch, xCh, xRow, xCol); + `,T=e?` + let coords = vec4( + batch, + row / outWidth, + row % outWidth, + col); + `:` + let coords = vec4( + batch, + row, + col / outWidth, + col % outWidth); + `,M=e?"i32(uniforms.x_shape[1])":"i32(uniforms.x_shape[2])",I=e?"i32(uniforms.x_shape[2])":"i32(uniforms.x_shape[3])",z=e?"row":"col",E=e?"col":"row",C=` + let inChannels = i32(uniforms.w_shape[2]); + let outWidth = ${e?"i32(uniforms.result_shape[2])":"i32(uniforms.result_shape[3])"}; + let outRow = ${z} / outWidth; + let outCol = ${z} % outWidth; + + let WRow = ${E} / (i32(uniforms.w_shape[1]) * inChannels); + let WCol = ${E} / inChannels % i32(uniforms.w_shape[1]); + let xRow = outRow * uniforms.stride[0] + uniforms.dilation[0] * WRow - uniforms.pad[0]; + let xCol = outCol * uniforms.stride[1] + uniforms.dilation[1] * WCol - uniforms.pad[1]; + let xCh = ${E} % inChannels; + var resData = ${Vr(u,f)}(0.0); + // The bounds checking is always needed since we use it to pad zero for + // the 'same' padding type. + if (xRow >= 0 && xRow < ${M} && xCol >= 0 && xCol < ${I}) { + ${g} + let xIndex = getIndexFromCoords4D(coord, vec4(uniforms.x_shape)); + ${y(u)} + } + return resData;`,D=e?r&&i?` + let col = colIn * ${u}; + ${C}`:` + let col = colIn * ${u}; + if (row < uniforms.dim_a_outer && col < uniforms.dim_inner) { + ${C} + } + return ${Vr(u,f)}(0.0);`:i&&t?` + let col = colIn * ${u}; + ${C}`:` + let col = colIn * ${u}; + if (row < uniforms.dim_inner && col < uniforms.dim_b_outer) { + ${C} + } + return ${Vr(u,f)}(0.0);`,A=e?i&&t?m(d):` + let col = colIn * ${d}; + if (row < uniforms.dim_inner && col < uniforms.dim_b_outer) { + ${m(d)} + } + return ${Vr(d,f)}(0.0);`:` + let col = colIn * ${d}; + if (row < uniforms.dim_inner && col < uniforms.dim_a_outer) { + ${m(d)} + } + return ${Vr(d,f)}(0.0);`,$=Vr(h,f),P=Vr(e?u:d,f),k=Vr(e?d:u,f),O=mi(n,$,f);return` + fn mm_readA(batch: i32, row : i32, colIn : i32) -> ${P} { + ${e?D:A} + } + + fn mm_readB(batch: i32, row : i32, colIn : i32) -> ${k} { + ${e?A:D} + } + + fn mm_write(batch: i32, row : i32, colIn : i32, valueIn : ${$}) { + let col = colIn * ${h}; + if (row < uniforms.dim_a_outer && col < uniforms.dim_b_outer) + { + var value = valueIn; + let outWidth = ${e?"i32(uniforms.result_shape[2])":"i32(uniforms.result_shape[3])"}; + ${T} + ${pM(o)} + ${O} + setOutputAtCoords(coords[0], coords[1], coords[2], coords[3], value); + } + }`},fM=(e,r,t,i,o,n,u,d,h)=>{let f=r.format==="NHWC",y=f?e[0].dims[3]:e[0].dims[1],m=t[0],g=f?t[2]:t[3],T=f?t[1]:t[2],M=f?t[3]:t[1],I=f&&(y%4===0||y%3===0)&&M%4===0,z=f?M:g*T,E=f?g*T:M,C=[8,8,1],D=i<=8?[4,1,1]:[4,4,1],A=[Math.ceil(z/C[0]/D[0]),Math.ceil(E/C[1]/D[1]),Math.ceil(m/C[2]/D[2])];Kt("verbose",()=>`[conv2d_mm_webgpu] dispatch = ${A}`);let $=I?f&&y%4!==0?3:4:1,P=C[1]*D[1],k=C[0]*D[0],O=Math.max(C[0]*$,C[1]),R=i%P===0,U=o%k===0,te=n%O===0,se=I?[$,4,4]:[1,1,1],K=[{type:6,data:i},{type:6,data:o},{type:6,data:n},{type:6,data:[r.pads[0],r.pads[1]]},{type:6,data:r.strides},{type:6,data:r.dilations}];_i(r,K),K.push(...Et(e[0].dims,e[1].dims));let pe=["rank","rank"];u&&(K.push(...Et(e[2].dims)),pe.push("rank")),K.push(...Et(t));let re=oe=>{let ge=[{name:"dim_a_outer",type:"i32"},{name:"dim_b_outer",type:"i32"},{name:"dim_inner",type:"i32"},{name:"pad",type:"i32",length:2},{name:"stride",type:"i32",length:2},{name:"dilation",type:"i32",length:2}];gi(r,ge);let le=I?4:1,Se=Lr(e[0].dataType),Ce=` + fn setOutputAtIndex(flatIndex : i32, value : ${I?`vec4<${Se}>`:Se}) { + result[flatIndex] = ${I?`vec4<${Se}>`:Se}(value); + } + fn setOutputAtCoords(d0 : i32, d1 : i32, d2 : i32, d3 : i32, value : ${I?`vec4<${Se}>`:Se}) { + let flatIndex = getOutputIndexFromCoords(vec4(d0, d1, d2, d3)); + setOutputAtIndex(flatIndex ${I?"/ 4":""}, value); + }`,q=Ne("x",e[0].dataType,e[0].dims.length,$===3?1:$),N=Ne("w",e[1].dataType,e[1].dims.length,le),Q=[q,N],ue=gt("result",e[0].dataType,t.length,le);if(u){let Ae=Ne("bias",e[2].dataType,e[2].dims.length,le);Q.push(Ae),Ce+=` + fn getBiasByOutputCoords(coords : vec4) -> ${I?`vec4<${Se}>`:Se} { + return bias[coords.${f?"w":"y"}${I?"/ 4":""}]; + }`}return` + ${hM("uniforms.result_strides")} + //struct Uniforms { xShape : vec4, wShape : vec4, outShape : vec4, + // outShapeStrides: vec3, filterDims : vec2, pad : vec2, stride : vec2, + // dilation : vec2, dimAOuter : i32, dimBOuter : i32, dimInner : i32 }; + ${oe.registerUniforms(ge).declareVariables(...Q,ue)} + ${Ce} + ${Qg(f,R,U,te,u,r,se[0],se[1],se[2],Se)} + ${I?jm(D,C,Se,void 0,!f,O):Lm(D,C,Se,void 0,!f,O,!1,void 0,d)}`};return{name:"Conv2DMatMul",shaderCache:{hint:`${r.cacheKey};${$};${I};${R};${U};${te};${P};${k};${O}`,inputDependencies:pe},getRunData:()=>({outputs:[{dims:h?h(t):t,dataType:e[0].dataType}],dispatchGroup:{x:A[0],y:A[1],z:A[2]},programUniforms:K}),getShaderSource:re}}}),Xg,Yf,od,Jg,Zf,Yg,mM,_M,dT=Ye(()=>{$t(),mn(),Lt(),zt(),wi(),o_(),Xg=e=>{let r=1;for(let t=0;ttypeof e=="number"?[e,e,e]:e,od=(e,r)=>r<=1?e:e+(e-1)*(r-1),Jg=(e,r,t,i=1)=>{let o=od(r,i);return Math.floor((e[0]*(t-1)-t+o)/2)},Zf=(e,r,t,i,o)=>{o==null&&(o=Jg(e,r[0],i[0]));let n=[0,0,0,t];for(let u=0;u<3;u++)e[u]+2*o>=r[u]&&(n[u]=Math.trunc((e[u]-r[u]+2*o)/i[u]+1));return n},Yg=(e,r,t,i,o,n,u,d,h,f)=>{let y,m,g,T;if(e==="VALID"&&(e=0),typeof e=="number"){y={top:e,bottom:e,left:e,right:e,front:e,back:e};let M=Zf([r,t,i,1],[d,h,f],1,[o,n,u],e);m=M[0],g=M[1],T=M[2]}else if(Array.isArray(e)){if(!e.every((I,z,E)=>I===E[0]))throw Error(`Unsupported padding parameter: ${e}`);y={top:e[0],bottom:e[1],left:e[2],right:e[3],front:e[4],back:e[5]};let M=Zf([r,t,i,1],[d,h,f],1,[o,n,u],e[0]);m=M[0],g=M[1],T=M[2]}else if(e==="SAME_UPPER"){m=Math.ceil(r/o),g=Math.ceil(t/n),T=Math.ceil(i/u);let M=(m-1)*o+d-r,I=(g-1)*n+h-t,z=(T-1)*u+f-i,E=Math.floor(M/2),C=M-E,D=Math.floor(I/2),A=I-D,$=Math.floor(z/2),P=z-$;y={top:D,bottom:A,left:$,right:P,front:E,back:C}}else throw Error(`Unknown padding parameter: ${e}`);return{padInfo:y,outDepth:m,outHeight:g,outWidth:T}},mM=(e,r,t,i,o,n=!1,u="channelsLast")=>{let d,h,f,y,m;if(u==="channelsLast")[d,h,f,y,m]=e;else if(u==="channelsFirst")[d,m,h,f,y]=e;else throw new Error(`Unknown dataFormat ${u}`);let[g,,T,M,I]=r,[z,E,C]=Yf(t),[D,A,$]=Yf(i),P=od(T,D),k=od(M,A),O=od(I,$),{padInfo:R,outDepth:U,outHeight:te,outWidth:se}=Yg(o,h,f,y,z,E,C,P,k,O),K=n?g*m:g,pe=[0,0,0,0,0];return u==="channelsFirst"?pe=[d,K,U,te,se]:u==="channelsLast"&&(pe=[d,U,te,se,K]),{batchSize:d,dataFormat:u,inDepth:h,inHeight:f,inWidth:y,inChannels:m,outDepth:U,outHeight:te,outWidth:se,outChannels:K,padInfo:R,strideDepth:z,strideHeight:E,strideWidth:C,filterDepth:T,filterHeight:M,filterWidth:I,effectiveFilterDepth:P,effectiveFilterHeight:k,effectiveFilterWidth:O,dilationDepth:D,dilationHeight:A,dilationWidth:$,inShape:e,outShape:pe,filterShape:r}},_M=(e,r,t,i,o,n)=>{let u=n==="channelsLast";u?e[0].dims[3]:e[0].dims[1];let d=[64,1,1],h={x:t.map((z,E)=>E)},f=[Math.ceil(Xg(h.x.map(z=>t[z]))/d[0]),1,1];Kt("verbose",()=>`[conv3d_naive_webgpu] dispatch = ${f}`);let y=1,m=je.size(t),g=[{type:12,data:m},{type:12,data:i},{type:12,data:o},{type:12,data:r.strides},{type:12,data:r.dilations}];_i(r,g),g.push(...Et(e[0].dims,e[1].dims));let T=["rank","rank"],M=e.length===3;M&&(g.push(...Et(e[2].dims)),T.push("rank")),g.push(...Et(t));let I=z=>{let E=[{name:"output_size",type:"u32"},{name:"filter_dims",type:"u32",length:i.length},{name:"pads",type:"u32",length:o.length},{name:"strides",type:"u32",length:r.strides.length},{name:"dilations",type:"u32",length:r.dilations.length}];gi(r,E);let C=1,D=Lr(e[0].dataType),A=Ne("x",e[0].dataType,e[0].dims.length,y),$=Ne("W",e[1].dataType,e[1].dims.length,C),P=[A,$],k=gt("result",e[0].dataType,t.length,C),O="";if(M){let te=Ne("bias",e[2].dataType,e[2].dims.length,C);P.push(te),O+=` + fn getBiasByOutputCoords(coords : array) -> ${D} { + return bias[${u?Mt("coords",4,5):Mt("coords",1,5)}]; + }`}let R=Vr(y,D),U=mi(r,R,D);return` + ${O} + fn getX(d0 : u32, d1 : u32, d2 : u32, d3 : u32, d4 : u32) -> f32 { + let aIndices = array(d0, d1, d2, d3, d4); + return ${A.getByIndices("aIndices")}; + } + fn getW(d0 : u32, d1 : u32, d2 : u32, d3 : u32, d4 : u32) -> f32 { + let aIndices = array(d0, d1, d2, d3, d4); + return ${$.getByIndices("aIndices")}; + } + ${z.registerUniforms(E).declareVariables(...P,k)} + ${z.mainStart()} + ${z.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let coords = ${k.offsetToIndices("global_idx")}; + let batch = ${Mt("coords",0,A.rank)}; + let d2 = ${u?Mt("coords",A.rank-1,A.rank):Mt("coords",1,A.rank)}; + let xFRCCorner = vec3(${u?Mt("coords",1,A.rank):Mt("coords",2,A.rank)}, + ${u?Mt("coords",2,A.rank):Mt("coords",3,A.rank)}, + ${u?Mt("coords",3,A.rank):Mt("coords",4,A.rank)}) * uniforms.strides - uniforms.pads; + let xFCorner = xFRCCorner.x; + let xRCorner = xFRCCorner.y; + let xCCorner = xFRCCorner.z; + let xShapeY = ${u?Mt("uniforms.x_shape",1,A.rank):Mt("uniforms.x_shape",2,A.rank)}; + let xShapeZ = ${u?Mt("uniforms.x_shape",2,A.rank):Mt("uniforms.x_shape",3,A.rank)}; + let xShapeW = ${u?Mt("uniforms.x_shape",3,A.rank):Mt("uniforms.x_shape",4,A.rank)}; + let xShapeU = ${u?Mt("uniforms.x_shape",4,A.rank):Mt("uniforms.x_shape",1,A.rank)}; + let inputDepthNearestVec4 = (xShapeU / 4) * 4; + let inputDepthVec4Remainder = xShapeU % 4; + + var value = 0.0; + for (var wF = 0u; wF < uniforms.filter_dims[0]; wF++) { + let xF = xFCorner + wF * uniforms.dilations[0]; + if (xF < 0 || xF >= xShapeY) { + continue; + } + + for (var wR = 0u; wR < uniforms.filter_dims[1]; wR++) { + let xR = xRCorner + wR * uniforms.dilations[1]; + if (xR < 0 || xR >= xShapeZ) { + continue; + } + + for (var wC = 0u; wC < uniforms.filter_dims[2]; wC++) { + let xC = xCCorner + wC * uniforms.dilations[2]; + if (xC < 0 || xC >= xShapeW) { + continue; + } + + for (var d1 = 0u; d1 < inputDepthNearestVec4; d1 += 4) { + ${u?`let xValues = vec4( + getX(batch, xF, xR, xC, d1), + getX(batch, xF, xR, xC, d1 + 1), + getX(batch, xF, xR, xC, d1 + 2), + getX(batch, xF, xR, xC, d1 + 3)); + `:`let xValues = vec4( + getX(batch, d1, xF, xR, xC), + getX(batch, d1 + 1, xF, xR, xC), + getX(batch, d1 + 2, xF, xR, xC), + getX(batch, d1 + 3, xF, xR, xC)); + `} + let wValues = vec4( + getW(d2, d1, wF, wR, wC), + getW(d2, d1 + 1, wF, wR, wC), + getW(d2, d1 + 2, wF, wR, wC), + getW(d2, d1 + 3, wF, wR, wC)); + value += dot(xValues, wValues); + } + if (inputDepthVec4Remainder == 1) { + ${u?`value += getX(batch, xF, xR, xC, inputDepthNearestVec4) + * getW(d2, inputDepthNearestVec4, wF, wR, wC);`:`value += getX(batch, inputDepthNearestVec4, xF, xR, xC) + * getW(d2, inputDepthNearestVec4, wF, wR, wC);`} + } else if (inputDepthVec4Remainder == 2) { + ${u?`let xValues = vec2( + getX(batch, xF, xR, xC, inputDepthNearestVec4), + getX(batch, xF, xR, xC, inputDepthNearestVec4 + 1)); + `:`let xValues = vec2( + getX(batch, inputDepthNearestVec4, xF, xR, xC), + getX(batch, inputDepthNearestVec4 + 1, xF, xR, xC)); + `} + let wValues = vec2( + getW(d2, inputDepthNearestVec4, wF, wR, wC), + getW(d2, inputDepthNearestVec4 + 1, wF, wR, wC)); + value += dot(xValues, wValues); + } else if (inputDepthVec4Remainder == 3) { + ${u?`let xValues = vec3( + getX(batch, xF, xR, xC, inputDepthNearestVec4), + getX(batch, xF, xR, xC, inputDepthNearestVec4 + 1), + getX(batch, xF, xR, xC, inputDepthNearestVec4 + 2)); + `:`let xValues = vec3( + getX(batch, inputDepthNearestVec4, xF, xR, xC), + getX(batch, inputDepthNearestVec4 + 1, xF, xR, xC), + getX(batch, inputDepthNearestVec4 + 2, xF, xR, xC)); + `} + let wValues = vec3( + getW(d2, inputDepthNearestVec4, wF, wR, wC), + getW(d2, inputDepthNearestVec4 + 1, wF, wR, wC), + getW(d2, inputDepthNearestVec4 + 2, wF, wR, wC)); + value += dot(xValues, wValues); + } + } + } + } + ${M?"value = value + getBiasByOutputCoords(coords)":""}; + ${U} + result[global_idx] = f32(value); + }`};return{name:"Conv3DNaive",shaderCache:{hint:`${r.cacheKey};${u};${y};${M}`,inputDependencies:T},getRunData:()=>({outputs:[{dims:t,dataType:e[0].dataType}],dispatchGroup:{x:f[0],y:f[1],z:f[2]},programUniforms:g}),getShaderSource:I}}}),gM,yM,pT=Ye(()=>{$t(),Lt(),zt(),wi(),gM=(e,r,t,i)=>{let o=e.length>2,n=o?"value += b[output_channel];":"",u=e[0].dims,d=e[1].dims,h=r.format==="NHWC",f=h?t[3]:t[1],y=f/r.group,m=h&&y>=4?gr(f):1,g=je.size(t)/m,T=[{type:12,data:g},{type:12,data:r.dilations},{type:12,data:[r.strides[0],r.strides[1]]},{type:12,data:[r.pads[0],r.pads[1]]},{type:12,data:y}];_i(r,T),T.push(...Et(u,[d[0],d[1],d[2],d[3]/m]));let M=o?["rank","rank","rank"]:["rank","rank"];T.push(...Et([t[0],t[1],t[2],t[3]/m]));let I=z=>{let E=gt("output",e[0].dataType,t.length,m),C=Lr(E.type.tensor),D=mi(r,E.type.value,C),A=Ne("x",e[0].dataType,u.length),$=Ne("w",e[1].dataType,d.length,m),P=[A,$];o&&P.push(Ne("b",e[2].dataType,e[2].dims,m));let k=[{name:"output_size",type:"u32"},{name:"dilations",type:"u32",length:r.dilations.length},{name:"strides",type:"u32",length:2},{name:"pads",type:"u32",length:2},{name:"output_channels_per_group",type:"u32"}];gi(r,k);let O=h?` + for (var wHeight: u32 = 0u; wHeight < uniforms.w_shape[0]; wHeight++) { + let xHeight = xRCCorner.x + wHeight * uniforms.dilations[0]; + + if (xHeight < 0u || xHeight >= uniforms.x_shape[1]) { + continue; + } + + for (var wWidth: u32 = 0u; wWidth < uniforms.w_shape[1]; wWidth++) { + let xWidth = xRCCorner.y + wWidth * uniforms.dilations[1]; + if (xWidth < 0u || xWidth >= uniforms.x_shape[2]) { + continue; + } + + for (var wInChannel: u32 = 0u; wInChannel < uniforms.w_shape[2]; wInChannel++) { + let input_channel = in_channel_offset + wInChannel; + let xVal = ${A.get("batch","xHeight","xWidth","input_channel")}; + let wVal = ${$.get("wHeight","wWidth","wInChannel","output_channel")}; + value += xVal * wVal; + } + } + } + `:` + for (var wInChannel: u32 = 0u; wInChannel < uniforms.w_shape[1]; wInChannel++) { + let input_channel = in_channel_offset + wInChannel; + for (var wHeight: u32 = 0u; wHeight < uniforms.w_shape[2]; wHeight++) { + let xHeight = xRCCorner.x + wHeight * uniforms.dilations[0]; + + if (xHeight < 0u || xHeight >= uniforms.x_shape[2]) { + continue; + } + + for (var wWidth: u32 = 0u; wWidth < uniforms.w_shape[3]; wWidth++) { + let xWidth = xRCCorner.y + wWidth * uniforms.dilations[1]; + if (xWidth < 0u || xWidth >= uniforms.x_shape[3]) { + continue; + } + + let xVal = ${A.get("batch","input_channel","xHeight","xWidth")}; + let wVal = ${$.get("output_channel","wInChannel","wHeight","wWidth")}; + value += xVal * wVal; + } + } + } + `;return` + ${z.registerUniforms(k).declareVariables(...P,E)} + + ${z.mainStart()} + ${z.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + + let outputIndices = ${E.offsetToIndices("global_idx")}; + let batch: u32 = outputIndices[0]; + let output_channel: u32 = outputIndices[${h?3:1}]; + let xRCCorner: vec2 = vec2(outputIndices[${h?1:2}], outputIndices[${h?2:3}]) * uniforms.strides - uniforms.pads; + let group_id: u32 = output_channel * ${m} / uniforms.output_channels_per_group; + var in_channel_offset = group_id * uniforms.w_shape[${h?2:1}]; + + var value: ${E.type.value} = ${E.type.value}(0); + ${O} + ${n} + ${D} + ${E.setByOffset("global_idx","value")} + }`};return{name:"GroupedConv",shaderCache:{hint:`${r.cacheKey}_${m}`,inputDependencies:M},getRunData:()=>({outputs:[{dims:i?i(t):t,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(g/64)},programUniforms:T}),getShaderSource:I}},yM=(e,r,t,i)=>{let o=e.length>2,n=gr(t[3]),u=gr(t[2]),d=je.size(t)/n/u,h=[e[0].dims[0],e[0].dims[1],e[0].dims[2],e[0].dims[3]/n],f=[e[1].dims[0],e[1].dims[1],e[1].dims[2],e[1].dims[3]/n],y=[t[0],t[1],t[2],t[3]/n],m=[{type:12,data:d},{type:6,data:[r.strides[0],r.strides[1]]},{type:6,data:[r.pads[0],r.pads[1]]}];_i(r,m),m.push(...Et(h,f,y));let g=(u-1)*r.strides[1]+f[1],T=M=>{let I=gt("output",e[0].dataType,y.length,n),z=Lr(I.type.tensor),E=mi(r,I.type.value,z),C=Ne("x",e[0].dataType,h.length,n),D=Ne("w",e[1].dataType,f.length,n),A=[C,D];o&&A.push(Ne("b",e[2].dataType,e[2].dims,n));let $=o?"value += b[output_channel];":"",P=[{name:"output_size",type:"u32"},{name:"strides",type:"i32",length:2},{name:"pads",type:"i32",length:2}];return gi(r,P),` + ${M.registerUniforms(P).declareVariables(...A,I)} + ${M.mainStart()} + ${M.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let width0 = uniforms.output_shape[3]; + let output_channel = global_idx % width0; + var index1 = global_idx / width0; + let width1 = uniforms.output_shape[2] / ${u}u; + let col = (index1 % width1) * ${u}u; + index1 = index1 / width1; + let row = index1 % uniforms.output_shape[1]; + let batch = index1 / uniforms.output_shape[1]; + + let x_corner = vec2(i32(row), i32(col)) * uniforms.strides - uniforms.pads; + + var x_vals: array<${C.type.value}, ${g}>; + var values: array<${I.type.value}, ${u}>; + let input_channel = output_channel; + // Use constant instead of uniform can give better performance for w's height/width. + for (var w_height: u32 = 0u; w_height < ${f[0]}; w_height++) { + let x_height = x_corner.x + i32(w_height); + if (x_height >= 0 && u32(x_height) < uniforms.x_shape[1]) { + for (var i = 0; i < ${g}; i++) { + let x_width = x_corner.y + i; + if (x_width >= 0 && u32(x_width) < uniforms.x_shape[2]) { + x_vals[i] = ${C.get("batch","u32(x_height)","u32(x_width)","input_channel")}; + } else { + x_vals[i] = ${C.type.value}(0); + } + } + for (var w_width: u32 = 0u; w_width < ${f[1]}; w_width++) { + let w_val = ${D.get("w_height","w_width","0","output_channel")}; + for (var i = 0u; i < ${u}u; i++) { + values[i] = fma(x_vals[i * u32(uniforms.strides[1]) + w_width], w_val, values[i]); + } + } + } + } + + for (var i = 0u; i < ${u}u; i++) { + var value = values[i]; + ${$} + ${E} + ${I.set("batch","row","col + i","output_channel","value")}; + } + }`};return{name:"GroupedConv-Vectorize",shaderCache:{hint:`${r.cacheKey};${n};${u};${g};${f[0]};${f[1]}`,inputDependencies:o?["rank","rank","type"]:["rank","rank"]},getRunData:()=>({outputs:[{dims:i?i(t):t,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(d/64)},programUniforms:m}),getShaderSource:T}}}),Zg,Nh,ey,Vh,zm,em,ty,ry,Bm,hT=Ye(()=>{Lt(),uT(),dT(),u_(),pT(),wi(),c_(),On(),Zg=(e,r,t,i,o,n)=>{let u=e[0],d=e.slice(n?1:2,n?3:4),h=d.length,f=r[0],y=r.slice(2).map((g,T)=>g+(g-1)*(t[T]-1)),m=d.map((g,T)=>g+i[T]+i[T+h]).map((g,T)=>Math.floor((g-y[T]+o[T])/o[T]));return m.splice(0,0,u),m.splice(n?3:1,0,f),m},Nh=[2,3,1,0],ey=(e,r)=>{if(!e||e.length!==2&&e.length!==3)throw new Error("Conv requires 2 or 3 inputs");if(e[0].dims.length>5)throw new Error("greater than 5D is not supported");if(e[0].dims.length!==e[1].dims.length)throw new Error("filter does not have same dimension as input");let t=e[0].dims[r.format==="NHWC"?e[0].dims.length-1:1],i=e[1].dims[1]*r.group;if(t!==i)throw new Error("FILTER_IN_CHANNEL should be equal to DATA_CHANNEL");if(e.length===3&&(e[2].dims.length!==1||e[1].dims[0]!==e[2].dims[0]))throw new Error("invalid bias");let o=e[0].dims.length-2;if(r.dilations.length!==o)throw new Error(`dilations should be ${o}D`);if(r.strides.length!==o)throw new Error(`strides should be ${o}D`);if(r.pads.length!==o*2)throw new Error(`pads should be ${o*2}D`);if(r.kernelShape.length!==0&&r.kernelShape.length!==e[1].dims.length-2)throw new Error("invalid kernel shape")},Vh=(e,r)=>{let t=e.kernelShape.slice();t.length{let r=a_(e),t=e.format,i=["NOTSET","VALID","SAME_UPPER","SAME_LOWER"][e.auto_pad],o=e.dilations,n=e.group,u=e.kernel_shape,d=e.pads,h=e.strides,f=e.w_is_const();return{autoPad:i,format:t,dilations:o,group:n,kernelShape:u,pads:d,strides:h,wIsConst:f,...r,cacheKey:`${e.format};${r.activation};`}},em=(e,r,t,i)=>{let o=t.format==="NHWC",n=Zg(r[0].dims,r[1].dims,t.dilations,t.pads,t.strides,o);if(t.group!==1){let P=[r[0]];if(o){let k=e.kernelCustomData.wT??e.compute(_s(r[1],Nh),{inputs:[1],outputs:[t.wIsConst?-2:-1]})[0];t.wIsConst&&!e.kernelCustomData.wT&&(e.kernelCustomData.wT=k),P.push(k)}else P.push(r[1]);r.length===3&&P.push(r[2]),!e.adapterInfo.isArchitecture("ampere")&&o&&r[1].dims[0]===t.group&&r[1].dims[1]===1&&t.dilations[0]===1&&t.dilations[1]===1?e.compute(yM(P,t,n,i),{inputs:P}):e.compute(gM(P,t,n,i),{inputs:P});return}let u=r.length===3,d=r[0].dims[o?1:2],h=r[0].dims[o?2:3],f=r[0].dims[o?3:1],y=r[1].dims[2],m=r[1].dims[3],g=n[o?1:2],T=n[o?2:3],M=n[o?3:1],I=o&&y===d&&m===h&&t.pads[0]===0&&t.pads[1]===0;if(I||y===1&&m===1&&t.dilations[0]===1&&t.dilations[1]===1&&t.strides[0]===1&&t.strides[1]===1&&t.pads[0]===0&&t.pads[1]===0){let P=n[0],k,O,R,U=[];if(o){let K=e.kernelCustomData.wT??e.compute(_s(r[1],Nh),{inputs:[1],outputs:[t.wIsConst?-2:-1]})[0];if(t.wIsConst&&!e.kernelCustomData.wT&&(e.kernelCustomData.wT=K),I){let pe=d*h*f;k=r[0].reshape([1,P,pe]),O=K.reshape([1,pe,M]),R=[1,P,M]}else k=r[0].reshape([P,d*h,f]),O=K.reshape([1,f,M]),R=[P,g*T,M];U.push(k),U.push(O)}else k=r[0].reshape([P,f,d*h]),O=r[1].reshape([1,M,f]),R=[P,M,g*T],U.push(O),U.push(k);u&&U.push(r[2]);let te=R[2],se=U[0].dims[U[0].dims.length-1];te<8&&se<8?e.compute(l_(U,t,n,R,o,i),{inputs:U}):e.compute(rf(U,t,n,R,o,i),{inputs:U});return}let z=!0,E=e.kernelCustomData.wT??e.compute(_s(r[1],Nh),{inputs:[1],outputs:[t.wIsConst?-2:-1]})[0];t.wIsConst&&!e.kernelCustomData.wT&&(e.kernelCustomData.wT=E);let C=[r[0],E];u&&C.push(r[2]);let D=o?g*T:M,A=o?M:g*T,$=y*m*f;e.compute(fM(C,t,n,D,A,$,u,z,i),{inputs:C})},ty=(e,r)=>{let t=r.format==="NHWC",i=[e.inputs[0].reshape(t?[e.inputs[0].dims[0],1,e.inputs[0].dims[1],e.inputs[0].dims[2]]:[e.inputs[0].dims[0],e.inputs[0].dims[1],1,e.inputs[0].dims[2]]),e.inputs[1].reshape([e.inputs[1].dims[0],e.inputs[1].dims[1],1,e.inputs[1].dims[2]])];e.inputs.length===3&&i.push(e.inputs[2]);let o=[0,r.pads[0],0,r.pads[1]],n=[1].concat(r.strides),u=[1].concat(r.dilations),d=[1].concat(r.kernelShape),h=Vh({...r,pads:o,strides:n,dilations:u,kernelShape:d},i);em(e,i,h,f=>t?[f[0],f[2],f[3]]:[f[0],f[1],f[3]])},ry=(e,r,t)=>{let i=t.format==="NHWC"?"channelsLast":"channelsFirst",o=Vh(t,r),n=t.autoPad==="NOTSET"?t.pads:t.autoPad,u=mM(r[0].dims,r[1].dims,t.strides,t.dilations,n,!1,i);e.compute(_M(r,o,u.outShape,[u.filterDepth,u.filterHeight,u.filterWidth],[u.padInfo.front,u.padInfo.top,u.padInfo.left],i))},Bm=(e,r)=>{if(ey(e.inputs,r),e.inputs[0].dims.length===3)ty(e,r);else if(e.inputs[0].dims.length===5)ry(e,e.inputs,r);else{let t=Vh(r,e.inputs);em(e,e.inputs,t)}}}),wM,fT=Ye(()=>{$t(),mn(),Lt(),zt(),wM=(e,r,t)=>{let i=e.length>2,o=r.outputShape,n=r.format==="NHWC",u=r.group,d=e[1].dims,h=d[2]/u,f=d[3],y=n?gr(h):1,m=n&&f===1&&h>=4,g=m?Math.floor(h/4)*4:Math.floor(h/y)*y,T=h-g,M=n?gr(f):1,I=n?f===1?y:M:1,z=je.size(o)/M,E=[Math.ceil(z/64),1,1];Kt("verbose",()=>`[conv2d_backprop_webgpu] dispatch = ${E}`);let C=["rank","rank"],D=[r.strides[0],r.strides[1]],A=[r.kernelShape[n?1:2],r.kernelShape[n?2:3]],$=[r.dilations[0],r.dilations[1]],P=[A[0]+(r.dilations[0]<=1?0:(r.kernelShape[n?1:2]-1)*(r.dilations[0]-1)),A[1]+(r.dilations[1]<=1?0:(r.kernelShape[n?2:3]-1)*(r.dilations[1]-1))],k=[P[0]-1-Math.floor((r.pads[0]+r.pads[2])/2),P[1]-1-Math.floor((r.pads[1]+r.pads[3])/2)],O=[{type:12,data:z},{type:12,data:D},{type:12,data:A},{type:12,data:$},{type:12,data:P},{type:6,data:k},{type:12,data:g},{type:12,data:h},{type:12,data:f},...Et(e[0].dims,e[1].dims)];i&&(O.push(...Et(e[2].dims)),C.push("rank")),O.push(...Et(o));let R=U=>{let te=[{name:"output_size",type:"u32"},{name:"strides",type:"u32",length:D.length},{name:"filter_dims",type:"u32",length:A.length},{name:"dilations",type:"u32",length:A.length},{name:"effective_filter_dims",type:"u32",length:P.length},{name:"pads",type:"i32",length:k.length},{name:"input_channels_per_group_int",type:"u32"},{name:"input_channels_per_group",type:"u32"},{name:"output_channels_per_group",type:"u32"}],se=Lr(e[0].dataType),K=n?1:2,pe=n?2:3,re=n?3:1,oe=Ne("W",e[1].dataType,e[1].dims.length,I),ge=Ne("Dy",e[0].dataType,e[0].dims.length,y),le=[ge,oe];i&&le.push(Ne("bias",e[2].dataType,[o[re]].length,M));let Se=gt("result",e[0].dataType,o.length,M),Ce=()=>{let Q="";if(m)y===4?Q+=` + let xValue = ${ge.getByOffset("x_offset")}; + let wValue = ${oe.getByOffset("w_offset")}; + dotProd = dotProd + dot(xValue, wValue); + x_offset += 1u; + w_offset += 1u;`:y===2?Q+=` + dotProd = dotProd + dot(vec4<${se}>(${ge.getByOffset("x_offset")}, ${ge.getByOffset("x_offset + 1u")}), vec4<${se}>(${oe.getByOffset("w_offset")}, ${oe.getByOffset("w_offset + 1u")})); + x_offset += 2u; + w_offset += 2u;`:y===1&&(Q+=` + dotProd = dotProd + dot(vec4<${se}>(${ge.getByOffset("x_offset")}, ${ge.getByOffset("x_offset + 1u")}, ${ge.getByOffset("x_offset + 2u")}, ${ge.getByOffset("x_offset + 3u")}), vec4<${se}>(${oe.getByOffset("w_offset")}, ${oe.getByOffset("w_offset + 1u")}, ${oe.getByOffset("w_offset + 2u")}, ${oe.getByOffset("w_offset + 3u")})); + x_offset += 4u; + w_offset += 4u;`);else if(Q+=` + let xValue = ${n?ge.getByOffset(`${ge.indicesToOffset(`${ge.type.indices}(batch, idyR, idyC, inputChannel)`)} / ${y}`):ge.get("batch","inputChannel","idyR","idyC")}; + `,y===1)Q+=` + let w_offset = ${oe.indicesToOffset(`${oe.type.indices}(u32(wRPerm), u32(wCPerm), inputChannel, wOutChannel)`)}; + let wValue = ${oe.getByOffset(`w_offset / ${I}`)}; + dotProd = dotProd + xValue * wValue;`;else for(let ue=0;ue{if(T===0)return"";if(!m)throw new Error(`packInputAs4 ${m} is not true.`);let Q="";if(y===1){Q+="dotProd = dotProd";for(let ue=0;ue(i32(r), i32(c)) - uniforms.pads; + let dyRCorner = dyCorner.x; + let dyCCorner = dyCorner.y; + let groupId = d1 / uniforms.output_channels_per_group; + let wOutChannel = d1 - groupId * uniforms.output_channels_per_group; + // Convolve dy(?, ?, d2) with w(:, :, d1, d2) to compute dx(xR, xC, d1). + // ? = to be determined. : = across all values in that axis. + var dotProd = ${Se.type.value}(0.0); + var wR: u32 = 0; + if (uniforms.dilations.x == 1) { + // Minimum wR >= 0 that satisfies (dyRCorner + wR) % (uniforms.strides.x) == 0 + wR = u32(((dyRCorner + i32(uniforms.strides.x) - 1) / i32(uniforms.strides.x)) * i32(uniforms.strides.x) - dyRCorner); + } + for (; wR < uniforms.effective_filter_dims.x; wR = wR + 1) { + if (wR % uniforms.dilations.x != 0) { + continue; + } + let dyR = (${se}(dyRCorner) + ${se}(wR)) / ${se}(uniforms.strides[0]); + let wRPerm = uniforms.filter_dims.x - 1 - wR / uniforms.dilations.x; + if (dyR < 0.0 || dyR >= ${se}(uniforms.Dy_shape[${K}]) || fract(dyR) > 0.0 || + wRPerm < 0) { + continue; + } + let idyR: u32 = u32(dyR); + var wC: u32 = 0; + if (uniforms.dilations.y == 1) { + // Minimum wC >= 0 that satisfies (dyCCorner + wC) % (uniforms.strides.y) == 0 + wC = u32(((dyCCorner + i32(uniforms.strides.y) - 1) / i32(uniforms.strides.y)) * i32(uniforms.strides.y) - dyCCorner); + } + for (; wC < uniforms.effective_filter_dims.y; wC = wC + 1) { + if (wC % uniforms.dilations.y != 0) { + continue; + } + let dyC = (${se}(dyCCorner) + ${se}(wC)) / ${se}(uniforms.strides.y); + let wCPerm = uniforms.filter_dims.y - 1 - wC / uniforms.dilations.y; + if (dyC < 0.0 || dyC >= ${se}(uniforms.Dy_shape[${pe}]) || + fract(dyC) > 0.0 || wCPerm < 0) { + continue; + } + let idyC: u32 = u32(dyC); + var inputChannel = groupId * uniforms.input_channels_per_group; + ${m?` + var x_offset = ${ge.indicesToOffset(`${ge.type.indices}(batch, idyR, idyC, inputChannel)`)} / ${y}; + var w_offset = ${oe.indicesToOffset(`${oe.type.indices}(wRPerm, wCPerm, inputChannel, wOutChannel)`)} / ${I}; + `:""} + for (var d2: u32 = 0; d2 < uniforms.input_channels_per_group_int; d2 = d2 + ${m?4:y}) { + ${Ce()} + inputChannel = inputChannel + ${m?4:y}; + } + ${q()} + wC = wC + uniforms.strides.y - 1; + } + wR = wR + uniforms.strides[0] - 1; + } + let value = dotProd${i?` + bias[d1 / ${M}]`:""}; + ${Se.setByOffset("global_idx","value")}; + `;return` + ${U.registerUniforms(te).declareVariables(...le,Se)} + ${U.mainStart()} + ${U.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")}; + ${N}}`};return{name:"ConvTranspose2D",shaderCache:{hint:`${r.cacheKey};${y}${I}${M}${m}${T}`,inputDependencies:C},getRunData:()=>({dispatchGroup:{x:E[0],y:E[1],z:E[2]},outputs:[{dims:t?t(o):o,dataType:e[0].dataType}],programUniforms:O}),getShaderSource:R}}}),sy,ny,iy,tm,bM,ay,rm,oy,MM,mT=Ye(()=>{fT(),wi(),On(),sy=(e,r,t,i,o,n)=>(e-1)*r+t+(i-1)*o+1-n,ny=(e,r,t,i,o)=>{let n=Math.floor(e/2);r==="SAME_UPPER"?(t[i]=n,t[o]=e-n):r==="SAME_LOWER"&&(t[i]=e-n,t[o]=n)},iy=(e,r,t,i,o,n,u,d,h,f)=>{let y=e.length-2,m=f.length===0;h.length{let t=e.kernelShape.slice();if(e.kernelShape.length===0||e.kernelShape.reduce((m,g)=>m*g,1)===0){t.length=0;for(let m=2;mm+g,0)===0){let m=r[0].dims.length-2;h=new Array(m).fill(1)}let f=e.strides.slice();if(f.reduce((m,g)=>m+g,0)===0){let m=r[0].dims.length-2;f=new Array(m).fill(1)}iy(d,t,h,e.autoPad,e.group,o,f,i,u,n);let y=Object.assign({},e);return Object.assign(y,{kernelShape:t,pads:o,outputPadding:u,outputShape:n,dilations:h,strides:f}),y},bM=e=>{let r=a_(e),t=e.format,i=["NOTSET","VALID","SAME_UPPER","SAME_LOWER"][typeof e.autoPad>"u"?0:e.autoPad],o=e.dilations,n=e.group,u=e.kernelShape,d=e.pads,h=e.strides,f=e.wIsConst(),y=e.outputPadding,m=e.outputShape;return{autoPad:i,format:t,dilations:o,group:n,kernelShape:u,outputPadding:y,outputShape:m,pads:d,strides:h,wIsConst:f,...r,cacheKey:`${e.format};${r.activation};`}},ay=(e,r)=>{if(!e||e.length!==2&&e.length!==3)throw new Error("Conv requires 2 or 3 inputs");if(e[0].dims.length!==4&&e[0].dims.length!==3)throw new Error("currently only support 2-dimensional conv");if(e[0].dims.length!==e[1].dims.length)throw new Error("filter does not have same dimension as input");let t=e[0].dims[r.format==="NHWC"?e[0].dims.length-1:1],i=e[1].dims[0];if(t!==i)throw new Error("FILTER_IN_CHANNEL should be equal to DATA_CHANNEL");let o=e[1].dims[1]*r.group;if(e.length===3&&(e[2].dims.length!==1||e[2].dims[0]!==o))throw new Error("invalid bias");let n=e[0].dims.length-2;if(r.dilations.reduce((u,d)=>u+d,0)>0&&r.dilations.length!==n)throw new Error(`dilations should be ${n}D`);if(r.strides.reduce((u,d)=>u+d,0)>0&&r.strides.length!==n)throw new Error(`strides should be ${n}D`);if(r.pads.reduce((u,d)=>u+d,0)>0&&r.pads.length!==n*2)throw new Error(`pads should be ${n*2}D`);if(r.outputPadding.length!==n&&r.outputPadding.length!==0)throw new Error(`output_padding should be ${n}D`);if(r.kernelShape.reduce((u,d)=>u+d,0)>0&&r.kernelShape.length!==0&&r.kernelShape.length!==e[1].dims.length-2)throw new Error("invalid kernel shape");if(r.outputShape.length!==0&&r.outputShape.length!==e[0].dims.length-2)throw new Error("invalid output shape")},rm=(e,r,t,i)=>{let o=e.kernelCustomData.wT??e.compute(_s(r[1],[2,3,0,1]),{inputs:[1],outputs:[t.wIsConst?-2:-1]})[0];t.wIsConst&&!e.kernelCustomData.wT&&(e.kernelCustomData.wT=o);let n=[r[0],o];r.length===3&&n.push(r[2]),e.compute(wM(n,t,i),{inputs:n})},oy=(e,r)=>{let t=r.format==="NHWC",i=[e.inputs[0].reshape(t?[e.inputs[0].dims[0],1,e.inputs[0].dims[1],e.inputs[0].dims[2]]:[e.inputs[0].dims[0],e.inputs[0].dims[1],1,e.inputs[0].dims[2]]),e.inputs[1].reshape([e.inputs[1].dims[0],e.inputs[1].dims[1],1,e.inputs[1].dims[2]])];e.inputs.length===3&&i.push(e.inputs[2]);let o=r.kernelShape;(o.length===0||o[0]===0)&&(o=[e.inputs[1].dims[2]]);let n=r.dilations;(n.length===0||n[0]===0)&&(n=[1]);let u=r.strides;(u.length===0||u[0]===0)&&(u=[1]);let d=r.pads;d.length===0&&(d=[0,0]),d=[0,d[0],0,d[1]],u=[1].concat(u),n=[1].concat(n),o=[1].concat(o);let h=r.outputPadding;h=[0].concat(h);let f=tm({...r,pads:d,strides:u,dilations:n,kernelShape:o,outputPadding:h},i);rm(e,i,f,y=>t?[y[0],y[2],y[3]]:[y[0],y[1],y[3]])},MM=(e,r)=>{if(ay(e.inputs,r),e.inputs[0].dims.length===3)oy(e,r);else{let t=tm(r,e.inputs);rm(e,e.inputs,t)}}}),ly,vM,xM,_T=Ye(()=>{$t(),Lt(),br(),zt(),ly=(e,r,t,i)=>{let o=je.size(r),n=r.length,u=Ne("input",e,n),d=gt("output",e,n),h=t.dataType===6?t.getInt32Array()[0]:Number(t.getBigInt64Array()[0]),f=je.normalizeAxis(h,n),y=m=>{let g=` i32(${u.indicesGet("inputIndices","uniforms.axis")}) `,T=Mt("uniforms.input_shape","uniforms.axis",n),M=i.reverse?g+(i.exclusive?" + 1":""):"0",I=i.reverse?T:g+(i.exclusive?"":" + 1");return` + ${m.registerUniform("outputSize","u32").registerUniform("axis","u32").declareVariables(u,d)} + ${m.mainStart()} + ${m.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")} + var inputIndices = ${d.offsetToIndices("global_idx")}; + var sum = ${d.type.value}(0); + let first : i32 = ${M}; + let last : i32 = ${I}; + for (var i : i32 = first; i < last; i++) { + ${u.indicesSet("inputIndices","uniforms.axis","u32(i)")}; + sum = sum + ${u.getByIndices("inputIndices")}; + } + ${d.setByOffset("global_idx","sum")}; + }`};return{name:"CumSum",shaderCache:{hint:i.cacheKey,inputDependencies:["rank"]},getRunData:()=>({outputs:[{dims:r,dataType:e}],dispatchGroup:{x:Math.ceil(o/64)},programUniforms:[{type:12,data:o},{type:12,data:f},...Et(r,r)]}),getShaderSource:y}},vM=(e,r)=>{let t=e.inputs[0].dims,i=e.inputs[0].dataType,o=e.inputs[1];e.compute(ly(i,t,o,r),{inputs:[0]})},xM=e=>{let r=e.exclusive===1,t=e.reverse===1;return Zt({exclusive:r,reverse:t})}}),cy,uy,dy,TM,CM,gT=Ye(()=>{$t(),Lt(),br(),zt(),cy=e=>{if(!e||e.length!==1)throw new Error("DepthToSpace requires 1 input.");if(e[0].dims.length!==4)throw new Error("DepthToSpace requires 4D input.")},uy=(e,r,t,i)=>{let o=[];o.push(`fn perm(i: ${i.type.indices}) -> ${t.type.indices} { + var a: ${t.type.indices};`);for(let n=0;n{let t,i,o,n,u,d,h=r.format==="NHWC",f=r.blocksize,y=r.mode==="DCR";h?([t,i,o,n]=e.dims,u=y?[t,i,o,f,f,n/f**2]:[t,i,o,n/f**2,f,f],d=y?[0,1,3,2,4,5]:[0,1,4,2,5,3]):([t,i,o,n]=[e.dims[0],e.dims[2],e.dims[3],e.dims[1]],u=y?[t,f,f,n/f**2,i,o]:[t,n/f**2,f,f,i,o],d=y?[0,3,4,1,5,2]:[0,1,4,2,5,3]);let m=e.reshape(u),g=m.dims.length,T=e.dataType,M=Ne("a",T,g),I=gt("output",T,g),z=E=>` + ${E.registerUniform("output_size","u32").declareVariables(M,I)} + + ${uy(d,g,M,I)} + + ${E.mainStart()} + ${E.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + + let indices = ${I.offsetToIndices("global_idx")}; + let aIndices = perm(indices); + + ${I.setByOffset("global_idx",M.getByIndices("aIndices"))} + }`;return{name:"DepthToSpace",shaderCache:{hint:`${e.dims};${r.blocksize};${r.mode}`,inputDependencies:["rank"]},getRunData:E=>{let C=h?[t,i*f,o*f,n/f**2]:[t,n/f**2,i*f,o*f],D=je.size(C),A=m.dims,$=je.sortBasedOnPerm(A,d);return{outputs:[{dims:C,dataType:E[0].dataType}],dispatchGroup:{x:Math.ceil(D/64)},programUniforms:[{type:12,data:D},...Et(A,$)]}},getShaderSource:z}},TM=(e,r)=>{cy(e.inputs),e.compute(dy(e.inputs[0],r))},CM=e=>Zt({blocksize:e.blocksize,mode:e.mode,format:e.format})}),Wh,ld,sm,py,hy,fy,my,nm,_y,EM,PM,yT=Ye(()=>{$t(),Lt(),br(),zt(),Wh="[a-zA-Z]|\\.\\.\\.",ld="("+Wh+")+",sm="^"+ld+"$",py="("+ld+",)*"+ld,hy="^"+py+"$",fy=class{constructor(e=-1){this.symbolToIndices=new Map,this.inputIndex=e}addSymbol(e,r){let t=this.symbolToIndices.get(e);t===void 0?t=[r]:t.push(r),this.symbolToIndices.set(e,t)}},my=class{constructor(e,r){var o;this.equation=r,this.hasEllipsis=!1,this.symbolToInfo=new Map,this.lhs=new Array,this.outputDims=[];let[t,i]=r.includes("->")?r.split("->",2):[r,""];if(!t.match(RegExp(hy)))throw new Error("Invalid LHS term");if(t.split(",").forEach((n,u)=>{let d=e[u].dims.slice();if(!n.match(RegExp(sm)))throw new Error("Invalid LHS term");let h=this.processTerm(n,!0,d,u);this.lhs.push(h)}),i==="")i+=[...this.symbolToInfo.entries()].filter(([n,u])=>u.count===1||n==="...").map(([n])=>n).join("");else if(!i.match(RegExp(ld)))throw new Error("Invalid RHS");(o=i.match(RegExp(Wh,"g")))==null||o.forEach(n=>{if(n==="...")this.outputDims=this.outputDims.concat(this.ellipsisDims);else{let u=this.symbolToInfo.get(n);if(u===void 0)throw new Error("Invalid RHS symbol");this.outputDims.push(u.dimValue)}}),this.rhs=this.processTerm(i,!1,this.outputDims)}addSymbol(e,r,t){let i=this.symbolToInfo.get(e);if(i!==void 0){if(i.dimValue!==r&&i.count!==1)throw new Error("Dimension mismatch");i.count++,i.inputIndices.push(t)}else i={count:1,dimValue:r,inputIndices:[t]};this.symbolToInfo.set(e,i)}processTerm(e,r,t,i=-1){let o=t.length,n=!1,u=[],d=0;if(!e.match(RegExp(sm))&&!r&&e!=="")throw new Error("Invalid LHS term");let h=e.match(RegExp(Wh,"g")),f=new fy(i);return h==null||h.forEach((y,m)=>{if(y==="..."){if(n)throw new Error("Only one ellipsis is allowed per input term");n=!0;let g=o-h.length+1;if(g<0)throw new Error("Ellipsis out of bounds");if(u=t.slice(d,d+g),this.hasEllipsis){if(this.ellipsisDims.length!==u.length||this.ellipsisDims.toString()!==u.toString())throw new Error("Ellipsis dimensions mismatch")}else if(r)this.hasEllipsis=!0,this.ellipsisDims=u;else throw new Error("Ellipsis must be specified in the LHS");for(let T=0;Te+"_max",_y=(e,r,t,i)=>{let o=e.map(f=>f.length).map((f,y)=>Ne(`input${y}`,r,f)),n=je.size(i),u=gt("output",r,i.length),d=[...t.symbolToInfo.keys()].filter(f=>!t.rhs.symbolToIndices.has(f)),h=f=>{let y=[],m="var prod = 1.0;",g="var sum = 0.0;",T="sum += prod;",M=[],I=[],z=[],E=[],C=t.symbolToInfo.size===t.rhs.symbolToIndices.size;t.symbolToInfo.forEach((A,$)=>{var P;if(t.rhs.symbolToIndices.has($)){let k=(P=t.rhs.symbolToIndices.get($))==null?void 0:P[0];k!==void 0&&t.lhs.forEach((O,R)=>{if(A.inputIndices.includes(R)){let U=O.symbolToIndices.get($);if(U===void 0)throw new Error("Invalid symbol error");U.forEach(te=>{y.push(`${o[R].indicesSet(`input${R}Indices`,te,u.indicesGet("outputIndices",k))}`)})}})}else t.lhs.forEach((k,O)=>{if(A.inputIndices.includes(O)){let R=k.symbolToIndices.get($);if(R===void 0)throw new Error("Invalid symbol error");R.forEach(U=>{M.push(`${o[O].indicesSet(`input${O}Indices`,U,`${$}`)}`)}),E.push(`prod *= ${o[O].getByIndices(`input${O}Indices`)};`)}}),I.push(`for(var ${$}: u32 = 0; ${$} < uniforms.${nm($)}; ${$}++) {`),z.push("}")});let D=C?[...y,`let sum = ${o.map((A,$)=>A.getByIndices(`input${$}Indices`)).join(" * ")};`]:[...y,g,...I,...M,m,...E,T,...z];return` + ${f.registerUniforms(d.map(A=>({name:`${nm(A)}`,type:"u32"}))).registerUniform("outputSize","u32").declareVariables(...o,u)} + + ${f.mainStart()} + ${f.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")} + var outputIndices = ${u.offsetToIndices("global_idx")}; + ${o.map((A,$)=>`var input${$}Indices: ${o[$].type.indices};`).join(` +`)} + ${D.join(` +`)}; + ${u.setByOffset("global_idx","sum")}; + }`};return{name:"Einsum",shaderCache:{hint:t.equation,inputDependencies:e.map(()=>"rank")},getRunData:()=>{let f=d.filter(m=>t.symbolToInfo.has(m)).map(m=>{var g;return{type:12,data:((g=t.symbolToInfo.get(m))==null?void 0:g.dimValue)||0}});f.push({type:12,data:n});let y=e.map((m,g)=>[...Et(m)]).reduce((m,g)=>m.concat(g),f);return y.push(...Et(i)),{outputs:[{dims:i,dataType:r}],dispatchGroup:{x:Math.ceil(n/64)},programUniforms:y}},getShaderSource:h}},EM=(e,r)=>{let t=new my(e.inputs,r.equation),i=t.outputDims,o=e.inputs.map((n,u)=>n.dims);e.compute(_y(o,e.inputs[0].dataType,t,i))},PM=e=>{let r=e.equation.replace(/\s+/g,"");return Zt({equation:r})}}),gy,im,yy,wy,SM,wT=Ye(()=>{$t(),Lt(),zt(),gy=e=>{if(!e||e.length!==2)throw new Error("Expand requires 2 input.");let r=e[0].dims,t=Array.from(e[1].getBigInt64Array(),Number),i=t.length{let t=e.length-r.length,i=[];for(let o=0;oe.length>r.length?im(e,r):im(r,e),wy=e=>{let r=e[0].dims,t=Array.from(e[1].getBigInt64Array(),Number),i=yy(r,t),o=e[0].dataType,n=o===9||je.size(r)===1,u=o===9||r.length>0&&r[r.length-1]%4===0?4:1,d=n||i.length>0&&i[i.length-1]%4===0?4:1,h=Math.ceil(je.size(i)/d),f=m=>{let g=Ne("input",o,r.length,u),T=gt("output",o,i.length,d),M;if(o===9){let I=(z,E,C="")=>` + let outputIndices${E} = ${T.offsetToIndices(`outputOffset + ${E}u`)}; + let offset${E} = ${g.broadcastedIndicesToOffset(`outputIndices${E}`,T)}; + let index${E} = offset${E} / 4u; + let component${E} = offset${E} % 4u; + ${z}[${E}] = ${C}(${g.getByOffset(`index${E}`)}[component${E}]); + `;M=` + let outputOffset = global_idx * ${d}; + var data = vec4(0); + ${I("data",0,"u32")} + ${I("data",1,"u32")} + ${I("data",2,"u32")} + ${I("data",3,"u32")} + ${T.setByOffset("global_idx","data")} + }`}else M=` + let outputIndices = ${T.offsetToIndices(`global_idx * ${d}`)}; + let inputOffset = ${g.broadcastedIndicesToOffset("outputIndices",T)}; + let data = ${T.type.value}(${g.getByOffset(`inputOffset / ${u}`)}); + ${T.setByOffset("global_idx","data")} + }`;return` + ${m.registerUniform("vec_size","u32").declareVariables(g,T)} + ${m.mainStart()} + ${m.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.vec_size")} + ${M}`},y=[{type:12,data:h},...Et(r,i)];return{name:"Expand",shaderCache:{hint:`${i.length};${u}${d}`,inputDependencies:["rank"]},getShaderSource:f,getRunData:()=>({outputs:[{dims:i,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(h/64)},programUniforms:y})}},SM=e=>{gy(e.inputs),e.compute(wy(e.inputs),{inputs:[0]})}}),by,kM,bT=Ye(()=>{$t(),Lt(),zt(),i_(),by=e=>{let r=e[0].dataType,t=je.size(e[0].dims),i=je.size(e[1].dims),o=i%4===0,n=u=>{let d=Ne("x",r,[1],4),h=Ne("bias",r,[1],4),f=gt("y",r,[1],4),y=[{name:"output_vec_size",type:"u32"},{name:"bias_size",type:"u32"}],m=T=>` + let bias${T}_offset: u32 = (global_idx * 4 + ${T}) % uniforms.bias_size; + let bias${T} = ${h.getByOffset(`bias${T}_offset / 4`)}[bias${T}_offset % 4];`,g=o?` + let bias = ${h.getByOffset("global_idx % (uniforms.bias_size / 4)")};`:`${m(0)}${m(1)}${m(2)}${m(3)} + let bias = ${d.type.value}(bias0, bias1, bias2, bias3);`;return`${u.registerUniforms(y).declareVariables(d,h,f)} + + ${Fm(Qr(r))} + + ${u.mainStart(Va)} + ${u.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_vec_size")} + + let x = ${d.getByOffset("global_idx")}; + ${g} + let x_in = x + bias; + ${f.setByOffset("global_idx",Dm("x_in"))} + }`};return{name:"FastGeluWithBias",shaderCache:{hint:`${o}`,inputDependencies:["type","type"]},getShaderSource:n,getRunData:u=>({outputs:[{dims:u[0].dims,dataType:u[0].dataType}],programUniforms:[{type:12,data:Math.ceil(t/4)},{type:12,data:i}],dispatchGroup:{x:Math.ceil(t/Va/4)}})}},kM=e=>{e.inputs.length<2||je.size(e.inputs[1].dims)===0?Qb(e):e.compute(by(e.inputs))}}),My,vy,$M,IM,MT=Ye(()=>{$t(),Lt(),br(),zt(),My=e=>{if(!e||e.length!==2)throw new Error("Gather requires 2 inputs.")},vy=(e,r)=>{let t=e[0].dims,i=e[1].dims,o=t.length,n=je.normalizeAxis(r.axis,o),u=t.slice(0);u.splice(n,1,...i);let d=t[n],h=e[0].dataType===9?4:1,f=Math.ceil(je.size(u)/h),y=[{type:12,data:f},{type:6,data:d},{type:12,data:n},...Et(e[0].dims,e[1].dims,u)],m=g=>{let T=Ne("data",e[0].dataType,e[0].dims.length,h),M=Ne("inputIndices",e[1].dataType,e[1].dims.length),I=gt("output",e[0].dataType,u.length,h),z=C=>{let D=i.length,A=`var indicesIndices${C} = ${M.type.indices}(0);`;for(let $=0;$1?`indicesIndices${C}[${$}]`:`indicesIndices${C}`} = ${u.length>1?`outputIndices${C}[uniforms.axis + ${$}]`:`outputIndices${C}`};`;A+=` + var idx${C} = ${M.getByIndices(`indicesIndices${C}`)}; + if (idx${C} < 0) { + idx${C} = idx${C} + uniforms.axisDimLimit; + } + var dataIndices${C} : ${T.type.indices}; + `;for(let $=0,P=0;$1?`dataIndices${C}[${$}]`:`dataIndices${C}`} = u32(idx${C});`,P+=D):(A+=`${o>1?`dataIndices${C}[${$}]`:`dataIndices${C}`} = ${u.length>1?`outputIndices${C}[${P}]`:`outputIndices${C}`};`,P++);return A},E;if(e[0].dataType===9){let C=(D,A,$="")=>` + let outputIndices${A} = ${I.offsetToIndices(`outputOffset + ${A}u`)}; + ${z(A)}; + let offset${A} = ${T.indicesToOffset(`dataIndices${A}`)}; + let index${A} = offset${A} / 4u; + let component${A} = offset${A} % 4u; + ${D}[${A}] = ${$}(${T.getByOffset(`index${A}`)}[component${A}]); + `;E=` + let outputOffset = global_idx * ${h}; + var value = vec4(0); + ${C("value",0,"u32")} + ${C("value",1,"u32")} + ${C("value",2,"u32")} + ${C("value",3,"u32")} + ${I.setByOffset("global_idx","value")} + `}else E=` + let outputIndices = ${I.offsetToIndices("global_idx")}; + ${z("")}; + let value = ${T.getByIndices("dataIndices")}; + ${I.setByOffset("global_idx","value")}; + `;return` + ${g.registerUniform("outputSize","u32").registerUniform("axisDimLimit","i32").registerUniform("axis","u32").declareVariables(T,M,I)} + ${g.mainStart()} + ${g.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")} + ${E} + }`};return{name:"Gather",shaderCache:{hint:r.cacheKey,inputDependencies:["rank","rank"]},getRunData:()=>({outputs:[{dims:u,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(f/64)},programUniforms:y}),getShaderSource:m}},$M=e=>Zt({axis:e.axis}),IM=(e,r)=>{let t=e.inputs;My(t),e.compute(vy(e.inputs,r))}}),xy,AM,OM,vT=Ye(()=>{$t(),Lt(),zt(),xy=(e,r,t,i,o,n,u,d,h)=>{let f=[{type:12,data:n},{type:12,data:i},{type:12,data:o},{type:12,data:t},{type:12,data:u},{type:12,data:d},{type:12,data:h}],y=[n];f.push(...Et(r.dims,y));let m=g=>{let T=Ne("indices_data",r.dataType,r.dims.length),M=gt("input_slice_offsets_data",12,1,1),I=[T,M],z=[{name:"output_size",type:"u32"},{name:"batch_dims",type:"u32"},{name:"input_dims",type:"u32",length:o.length},{name:"sizes_from_slice_dims_data",type:"u32",length:t.length},{name:"num_slices_per_batch",type:"u32"},{name:"input_batch_stride",type:"u32"},{name:"num_slice_dims",type:"u32"}];return` + ${g.registerUniforms(z).declareVariables(...I)} + ${g.mainStart()} + ${g.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let batch_idx = global_idx / uniforms.num_slices_per_batch; + let base_offset = batch_idx * uniforms.input_batch_stride; + + let slice_indices_base_offset = global_idx * uniforms.num_slice_dims; + var relative_slice_offset = 0; + for (var dim_idx = 0u; dim_idx < uniforms.num_slice_dims; dim_idx ++) { + var index = i32(indices_data[dim_idx + slice_indices_base_offset].x); + let input_dim_idx = uniforms.batch_dims + dim_idx; + if (index < 0) { + ${o.length===1?"index += i32(uniforms.input_dims);":"index += i32(uniforms.input_dims[input_dim_idx]);"} + } + ${t.length===1?"relative_slice_offset += index * i32(uniforms.sizes_from_slice_dims_data);":"relative_slice_offset += index * i32(uniforms.sizes_from_slice_dims_data[dim_idx]);"} + } + + input_slice_offsets_data[global_idx] = base_offset + u32(relative_slice_offset); + }`};return e.compute({name:"computeSliceOffsets",shaderCache:{hint:`${o.length}_${t.length}`,inputDependencies:["rank"]},getRunData:()=>({outputs:[{dims:y,dataType:e.inputs[1].dataType}],dispatchGroup:{x:Math.ceil(n/64)},programUniforms:f}),getShaderSource:m},{inputs:[r],outputs:[-1]})[0]},AM=(e,r)=>{let t=e.inputs,i=t[0].dims,o=t[0].dataType,n=t[1].dims,u=n[n.length-1],d=je.sizeToDimension(n,n.length-1),h=je.sizeFromDimension(i,r.batchDims+u),f=je.sizeToDimension(i,r.batchDims),y=je.sizeFromDimension(i,r.batchDims),m=d/f,g=new Array(u),T=h;for(let A=0;Ai.length)throw new Error("last dimension of indices must not be larger than rank of input tensor");let z=n.slice(0,-1).concat(i.slice(I)),E=je.size(z),C=[{type:12,data:E},{type:12,data:h},...Et(t[0].dims,M.dims,z)],D=A=>{let $=Ne("data",t[0].dataType,t[0].dims.length),P=Ne("slice_offsets",12,M.dims.length),k=gt("output",t[0].dataType,z.length);return` + ${A.registerUniform("output_size","u32").registerUniform("slice_size","u32").declareVariables($,P,k)} + ${A.mainStart()} + ${A.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let slice_offset = slice_offsets[global_idx / uniforms.slice_size]; + output[global_idx] = data[u32(slice_offset) + global_idx % uniforms.slice_size]; + }`};e.compute({name:"GatherND",shaderCache:{hint:r.cacheKey,inputDependencies:["rank","rank"]},getRunData:()=>({outputs:[{dims:z,dataType:o}],dispatchGroup:{x:Math.ceil(E/64)},programUniforms:C}),getShaderSource:D},{inputs:[t[0],M]})},OM=e=>({batchDims:e.batch_dims,cacheKey:""})}),Ty,Cy,FM,DM,xT=Ye(()=>{$t(),Lt(),br(),zt(),Ty=(e,r)=>{if(e.length<3||e.length>4)throw new Error("GatherBlockQuantized requires 3 or 4 inputs.");let t=je.normalizeAxis(r.quantizeAxis,e[0].dims.length),i=r.blockSize,o=e[0],n=e[2],u=e.length===4?e[3]:void 0;if(n.dims.length!==o.dims.length||!o.dims.map((d,h)=>h===t?Math.ceil(d/i)===n.dims[h]:d===n.dims[h]).reduce((d,h)=>d&&h,!0))throw new Error("Scales must have the same rank as the input tensor and the dims should match except on gatherAxis.");if(u){if(u.dataType!==o.dataType)throw new Error("Zero point must have the same data type as the input tensor.");if(u.dims.length!==n.dims.length||!u.dims.map((d,h)=>d===n.dims[h]).reduce((d,h)=>d&&h,!0))throw new Error("Zero point must have the same rank as the input tensor and the dims should match except on quantizeAxis.")}},Cy=(e,r)=>{let t=e[0].dims,i=e[1].dims,o=t.length,n=je.normalizeAxis(r.gatherAxis,o),u=je.normalizeAxis(r.quantizeAxis,o),d=t.slice(0);d.splice(n,1,...i);let h=je.size(d),f=e[2].dataType,y=e[0].dataType===22,m=[{type:12,data:h},{type:12,data:u},{type:12,data:n},{type:12,data:r.blockSize},...Et(...e.map((T,M)=>T.dims),d)],g=T=>{let M=Ne("data",e[0].dataType,e[0].dims.length),I=Ne("inputIndices",e[1].dataType,e[1].dims.length),z=Ne("scales",e[2].dataType,e[2].dims.length),E=e.length>3?Ne("zeroPoint",e[3].dataType,e[3].dims.length):void 0,C=gt("output",f,d.length),D=[M,I,z];E&&D.push(E);let A=[{name:"output_size",type:"u32"},{name:"quantize_axis",type:"u32"},{name:"gather_axis",type:"u32"},{name:"block_size",type:"u32"}];return` + ${T.registerUniforms(A).declareVariables(...D,C)} + ${T.mainStart()} + let output_indices = ${C.offsetToIndices("global_idx")}; + var indices_indices = ${I.type.indices}(0); + ${i.length>1?` + for (var i: u32 = 0; i < ${i.length}; i++) { + let index = ${C.indicesGet("output_indices","uniforms.gather_axis + i")}; + ${I.indicesSet("indices_indices","i","index")}; + }`:`indices_indices = ${C.indicesGet("output_indices","uniforms.gather_axis")};`}; + var data_indices = ${M.type.indices}(0); + for (var i: u32 = 0; i < uniforms.gather_axis; i++) { + let index = ${C.indicesGet("output_indices","i")}; + ${M.indicesSet("data_indices","i","index")}; + } + var index_from_indices = ${I.getByIndices("indices_indices")}; + if (index_from_indices < 0) { + index_from_indices += ${t[n]}; + } + ${M.indicesSet("data_indices","uniforms.gather_axis","u32(index_from_indices)")}; + for (var i = uniforms.gather_axis + 1; i < ${d.length}; i++) { + let index = ${C.indicesGet("output_indices",`i + ${i.length} - 1`)}; + ${M.indicesSet("data_indices","i","index")}; + } + let data_offset = ${M.indicesToOffset("data_indices")}; + let data_index = data_offset % 8; + // Convert 4-bit packed data to 8-bit packed data. + let packed_4bit_quantized_data = ${M.getByOffset("data_offset / 8")}; + let packed_8bit_quantized_data = (packed_4bit_quantized_data >> (4 * (data_index % 2))) & 0x0f0f0f0f; + let quantized_data_vec = ${y?"unpack4xI8":"unpack4xU8"}(u32(packed_8bit_quantized_data)); + let quantized_data = quantized_data_vec[data_index / 2]; + var scale_indices = data_indices; + let quantize_axis_index = ${z.indicesGet("data_indices","uniforms.quantize_axis")} / uniforms.block_size; + ${z.indicesSet("scale_indices","uniforms.quantize_axis","quantize_axis_index")}; + var scale = ${z.getByIndices("scale_indices")}; + ${E?` + let zero_point_indices = scale_indices; + let zero_point_offset = ${E.indicesToOffset("zero_point_indices")}; + let zero_point_index = zero_point_offset % 8; + let packed_4bit_zero_points = ${E.getByOffset("zero_point_offset / 8")}; + let packed_8bit_zero_points = (packed_4bit_zero_points >> (4 * (zero_point_index % 2))) & 0x0f0f0f0f; + let zero_point_vec = ${y?"unpack4xI8":"unpack4xU8"}(u32(packed_8bit_zero_points)); + let zero_point = zero_point_vec[zero_point_index / 2];`:"var zero_point = 0"}; + let dequantized_data = ${Qr(f)}(quantized_data - zero_point) * scale; + ${C.setByOffset("global_idx","dequantized_data")}; + }`};return{name:"GatherBlockQuantized",shaderCache:{hint:`${r.cacheKey};${e.filter((T,M)=>M!==1).map(T=>T.dims.join("_")).join(";")}`,inputDependencies:Array.from({length:e.length},(T,M)=>"rank")},getRunData:()=>({outputs:[{dims:d,dataType:f}],dispatchGroup:{x:Math.ceil(h/64)},programUniforms:m}),getShaderSource:g}},FM=(e,r)=>{let t=e.inputs;Ty(t,r),e.compute(Cy(e.inputs,r))},DM=e=>Zt({blockSize:e.blockSize,gatherAxis:e.gatherAxis,quantizeAxis:e.quantizeAxis})}),Ey,Py,jM,LM,TT=Ye(()=>{$t(),Lt(),br(),zt(),Ey=e=>{if(!e||e.length!==2)throw new Error("GatherElements requires 2 inputs.");if(e[0].dims.length<1)throw new Error("GatherElements requires that the data input be rank >= 1.");if(e[0].dims.length!==e[1].dims.length)throw new Error(`GatherElements requires that the data input and + indices input tensors be of same rank.`)},Py=(e,r)=>{let t=e[0].dims,i=e[0].dataType,o=t.length,n=e[1].dims,u=e[1].dataType,d=je.normalizeAxis(r.axis,o),h=t[d],f=n.slice(0),y=je.size(f),m=Ne("input",i,o),g=Ne("indicesInput",u,n.length),T=gt("output",i,f.length),M=[{type:12,data:y},{type:6,data:h},{type:12,data:d}];return M.push(...Et(t,n,f)),{name:"GatherElements",shaderCache:{inputDependencies:["rank","rank"]},getRunData:()=>({outputs:[{dims:f,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(y/64)},programUniforms:M}),getShaderSource:I=>` + ${I.registerUniform("outputSize","u32").registerUniform("axisDimLimit","i32").registerUniform("axis","u32").declareVariables(m,g,T)} + ${I.mainStart()} + ${I.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")} + + let outputIndices = ${T.offsetToIndices("global_idx")}; + + var idx = ${g.getByOffset("global_idx")}; + if (idx < 0) { + idx = idx + uniforms.axisDimLimit; + } + var inputIndices = ${m.type.indices}(outputIndices); + ${m.indicesSet("inputIndices","uniforms.axis","u32(idx)")}; + let value = ${m.getByIndices("inputIndices")}; + + ${T.setByOffset("global_idx","value")}; + }`}},jM=e=>Zt({axis:e.axis}),LM=(e,r)=>{let t=e.inputs;Ey(t),e.compute(Py(e.inputs,r))}}),Sy,ky,zM,BM,CT=Ye(()=>{$t(),Lt(),zt(),Sy=e=>{if(!e)throw new Error("Input is missing");if(e.length<2||e.length>3)throw new Error("Invaid input number.");if(e.length===3&&e[2].dims.length>2)throw new Error("Invalid input shape of C");if(e[0].dataType!==e[1].dataType||e.length===3&&e[0].dataType!==e[2].dataType)throw new Error("Input types are mismatched")},ky=(e,r)=>{let t=e[0].dims.slice(),i=e[1].dims.slice(),[o,n,u]=j0.getShapeOfGemmResult(t,r.transA,i,r.transB,e.length===3?e[2].dims:void 0),d=[o,n];if(!d)throw new Error("Can't use gemm on the given tensors");let h=16,f=Math.ceil(n/h),y=Math.ceil(o/h),m=!0,g=je.size(d),T=[{type:12,data:m?f:g},{type:12,data:o},{type:12,data:n},{type:12,data:u},{type:1,data:r.alpha},{type:1,data:r.beta}],M=["type","type"];e.length===3&&(T.push(...Et(e[2].dims)),M.push("rank")),T.push(...Et(d));let I=E=>{let C="";r.transA&&r.transB?C="value += a[k * uniforms.M + m] * b[n * uniforms.K + k];":r.transA&&!r.transB?C="value += a[k * uniforms.M + m] * b[k * uniforms.N + n];":!r.transA&&r.transB?C="value += a[m * uniforms.K + k] * b[n * uniforms.K + k];":!r.transA&&!r.transB&&(C="value += a[m * uniforms.K + k] * b[k * uniforms.N + n];");let D=r.alpha===1?"":"value *= uniforms.alpha;",A=Ne("a",e[0].dataType,e[0].dims),$=Ne("b",e[1].dataType,e[1].dims),P=A.type.value,k=null,O=[A,$];e.length===3&&(k=Ne("c",e[2].dataType,e[2].dims.length),O.push(k));let R=gt("output",e[0].dataType,d.length);O.push(R);let U=[{name:"output_size",type:"u32"},{name:"M",type:"u32"},{name:"N",type:"u32"},{name:"K",type:"u32"},{name:"alpha",type:"f32"},{name:"beta",type:"f32"}];return` + ${E.registerUniforms(U).declareVariables(...O)} + + ${E.mainStart()} + ${E.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + + let m = global_idx / uniforms.N; + let n = global_idx % uniforms.N; + + var value = ${P}(0); + for (var k: u32 = 0u; k < uniforms.K; k++) { + ${C} + } + + ${D} + ${k!=null?`let cOffset = ${k.broadcastedIndicesToOffset("vec2(m, n)",R)}; value += ${P}(uniforms.beta) * ${k.getByOffset("cOffset")};`:""} + output[global_idx] = value; + }`},z=E=>{let C=Ne("a",e[0].dataType,e[0].dims),D=Ne("b",e[1].dataType,e[1].dims),A=null,$=[C,D];e.length===3&&(A=Ne("c",e[2].dataType,e[2].dims.length),$.push(A));let P=gt("output",e[0].dataType,d.length);$.push(P);let k=[{name:"num_tile_n",type:"u32"},{name:"M",type:"u32"},{name:"N",type:"u32"},{name:"K",type:"u32"},{name:"alpha",type:"f32"},{name:"beta",type:"f32"}],O="",R="";r.transA&&r.transB?(R=` + var col = tile_row_start + local_id.x; + var row = k_start + local_id.y; + if (col < uniforms.M && row < uniforms.K) { + tile_a[local_id.y][local_id.x] = a[row * uniforms.M + col]; + } else { + tile_a[local_id.y][local_id.x] = ${C.type.value}(0); + } + + col = k_start + local_id.x; + row = tile_col_start + local_id.y; + if (col < uniforms.K && row < uniforms.N) { + tile_b[local_id.y][local_id.x] = b[row * uniforms.K + col]; + } else { + tile_b[local_id.y][local_id.x] = ${D.type.value}(0); + } + `,O="value += tile_a[k][local_id.y] * tile_b[local_id.x][k];"):r.transA&&!r.transB?(R=` + var col = tile_row_start + local_id.x; + var row = k_start + local_id.y; + if (col < uniforms.M && row < uniforms.K) { + tile_a[local_id.y][local_id.x] = a[row * uniforms.M + col]; + } else { + tile_a[local_id.y][local_id.x] = ${C.type.value}(0); + } + + col = tile_col_start + local_id.x; + row = k_start + local_id.y; + if (col < uniforms.N && row < uniforms.K) { + tile_b[local_id.y][local_id.x] = b[row * uniforms.N + col]; + } else { + tile_b[local_id.y][local_id.x] = ${D.type.value}(0); + } + `,O="value += tile_a[k][local_id.y] * tile_b[k][local_id.x];"):!r.transA&&r.transB?(R=` + var col = k_start + local_id.x; + var row = tile_row_start + local_id.y; + if (col < uniforms.K && row < uniforms.M) { + tile_a[local_id.y][local_id.x] = a[row * uniforms.K + col]; + } else { + tile_a[local_id.y][local_id.x] = ${C.type.value}(0); + } + + col = k_start + local_id.x; + row = tile_col_start + local_id.y; + if (col < uniforms.K && row < uniforms.N) { + tile_b[local_id.y][local_id.x] = b[row * uniforms.K + col]; + } else { + tile_b[local_id.y][local_id.x] = ${D.type.value}(0); + } + `,O="value += tile_a[local_id.y][k] * tile_b[local_id.x][k];"):!r.transA&&!r.transB&&(R=` + var col = k_start + local_id.x; + var row = tile_row_start + local_id.y; + if (col < uniforms.K && row < uniforms.M) { + tile_a[local_id.y][local_id.x] = a[row * uniforms.K + col]; + } else { + tile_a[local_id.y][local_id.x] = ${C.type.value}(0); + } + + col = tile_col_start + local_id.x; + row = k_start + local_id.y; + if (col < uniforms.N && row < uniforms.K) { + tile_b[local_id.y][local_id.x] = b[row * uniforms.N + col]; + } else { + tile_b[local_id.y][local_id.x] = ${D.type.value}(0); + } + `,O="value += tile_a[local_id.y][k] * tile_b[k][local_id.x];");let U=r.alpha===1?"":"value *= uniforms.alpha;";return` + ${E.registerUniforms(k).declareVariables(...$)} + var tile_a: array, ${h}>; + var tile_b: array, ${h}>; + ${E.mainStart([h,h,1])} + let tile_col_start = (workgroup_index % uniforms.num_tile_n) * ${h}; + let tile_row_start = (workgroup_index / uniforms.num_tile_n) * ${h}; + let num_tiles = (uniforms.K - 1) / ${h} + 1; + var k_start = 0u; + var value = ${P.type.value}(0); + for (var t: u32 = 0u; t < num_tiles; t++) { + ${R} + k_start = k_start + ${h}; + workgroupBarrier(); + + for (var k: u32 = 0u; k < ${h}; k++) { + ${O} + } + workgroupBarrier(); + } + + ${U} + let m = tile_row_start + local_id.y; + let n = tile_col_start + local_id.x; + ${A!=null?`let cOffset = ${A.broadcastedIndicesToOffset("vec2(m, n)",P)}; value += ${P.type.value}(uniforms.beta) * ${A.getByOffset("cOffset")};`:""} + if (m < uniforms.M && n < uniforms.N) { + output[m * uniforms.N + n] = value; + } + }`};return m?{name:"GemmShared",shaderCache:{hint:`${r.cacheKey}`,inputDependencies:M},getRunData:()=>({outputs:[{dims:d,dataType:e[0].dataType}],dispatchGroup:{x:f*y},programUniforms:T}),getShaderSource:z}:{name:"Gemm",shaderCache:{hint:`${r.cacheKey}`,inputDependencies:M},getRunData:()=>({outputs:[{dims:d,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(g/64)},programUniforms:T}),getShaderSource:I}},zM=e=>{let r=e.transA,t=e.transB,i=e.alpha,o=e.beta;return{transA:r,transB:t,alpha:i,beta:o,cacheKey:`${e.transA};${e.transB};${e.alpha===1}`}},BM=(e,r)=>{Sy(e.inputs),e.compute(ky(e.inputs,r))}}),Zs,pn,ai,oi,$y,Iy,Ay,Oy,Fy,Dy,jy,Ly,RM,NM,ET=Ye(()=>{$t(),Lt(),br(),zt(),[Zs,pn,ai,oi]=[0,1,2,3],$y=e=>{if(e[0].dims.length!==4)throw new Error("only 4-D tensor is supported.");if(e[0].dims.length!==e[1].dims.length)throw new Error("input dimensions must be equal to grid dimensions");if(e[0].dims.length-2!==e[1].dims[e[1].dims.length-1])throw new Error(`last dimension of grid must be equal to ${e[0].dims.length-2}`);if(e[0].dims[0]!==e[1].dims[0])throw new Error("grid batch size must match input batch size")},Iy=` + fn gs_get_cubic_coeffs(x: f32) -> vec4 { + let cubic_alpha = -0.75f; + let x_abs = abs(x); + var coeffs: vec4; + coeffs[0] = (((cubic_alpha * (x_abs + 1) - 5 * cubic_alpha) * (x_abs + 1) + 8 * cubic_alpha) * (x_abs + 1) - 4 * cubic_alpha); + coeffs[1] = (((cubic_alpha + 2) * x_abs - (cubic_alpha + 3)) * x_abs * x_abs + 1); + coeffs[2] = (((cubic_alpha + 2) * (1 - x_abs) - (cubic_alpha + 3)) * (1 - x_abs) * (1 - x_abs) + 1); + coeffs[3] = (((cubic_alpha * (2 - x_abs) - 5 * cubic_alpha) * (2 - x_abs) + 8 * cubic_alpha) * (2 - x_abs) - 4 * cubic_alpha); + return coeffs; + } +`,Ay=e=>` + fn gs_bicubic_interpolate(p: mat4x4<${e}>, x: f32, y: f32) -> ${e} { + var v: vec4; + var coeffs = gs_get_cubic_coeffs(x); + for (var i = 0; i < 4; i++) { + v[i] = coeffs[0] * p[i][0] + coeffs[1] * p[i][1] + coeffs[2] * p[i][2] + coeffs[3] * p[i][3]; + } + coeffs = gs_get_cubic_coeffs(y); + let pixel = ${e}(coeffs[0] * v[0] + coeffs[1] * v[1] + coeffs[2] * v[2] + coeffs[3] * v[3]); + return pixel; + } +`,Oy=e=>` + fn gs_denormalize(n: f32, length: i32) -> f32 { + ${e.alignCorners===0?` + // alignCorners: false => [-1, 1] to [-0.5, length - 0.5] + return ((n + 1.0) * f32(length) - 1.0) / 2.0; + `:` + // alignCorners: true => [-1, 1] to [0, length - 1] + return (n + 1.0) / 2.0 * (f32(length - 1)); + `} + } +`,Fy=e=>` + ${e.paddingMode==="reflection"?` + fn gs_reflect(x: i32, x_min: f32, x_max: f32) -> u32 { + var dx = 0.0; + var fx = f32(x); + let range = x_max - x_min; + if (fx < x_min) { + dx = x_min - fx; + let n = u32(dx / range); + let r = dx - f32(n) * range; + if (n % 2 == 0) { + fx = x_min + r; + } else { + fx = x_max - r; + } + } else if (fx > x_max) { + dx = fx - x_max; + let n = u32(dx / range); + let r = dx - f32(n) * range; + if (n % 2 == 0) { + fx = x_max - r; + } else { + fx = x_min + r; + } + } + return u32(fx); + }`:""} +`,Dy=(e,r,t)=>` + fn pixel_at_grid(r: i32, c: i32, H: i32, W: i32, batch: u32, channel: u32, border: vec4) -> ${r} { + var pixel = ${r}(0); + var indices = vec4(0); + indices[${Zs}] = batch; + indices[${pn}] = channel;`+(()=>{switch(t.paddingMode){case"zeros":return` + if (r >= 0 && r < H && c >=0 && c < W) { + indices[${ai}] = u32(r); + indices[${oi}] = u32(c); + } else { + return ${r}(0); + } + `;case"border":return` + indices[${ai}] = u32(clamp(r, 0, H - 1)); + indices[${oi}] = u32(clamp(c, 0, W - 1)); + `;case"reflection":return` + indices[${ai}] = gs_reflect(r, border[1], border[3]); + indices[${oi}] = gs_reflect(c, border[0], border[2]); + `;default:throw new Error(`padding mode ${t.paddingMode} is not supported`)}})()+` + return ${e.getByIndices("indices")}; + } +`,jy=(e,r,t)=>(()=>{switch(t.mode){case"nearest":return` + let result = pixel_at_grid(i32(round(y)), i32(round(x)), H_in, W_in, indices[${Zs}], indices[${pn}], border); + `;case"bilinear":return` + let x1 = i32(floor(x)); + let y1 = i32(floor(y)); + let x2 = x1 + 1; + let y2 = y1 + 1; + + let p11 = pixel_at_grid(y1, x1, H_in, W_in, indices[${Zs}], indices[${pn}], border); + let p12 = pixel_at_grid(y1, x2, H_in, W_in, indices[${Zs}], indices[${pn}], border); + let p21 = pixel_at_grid(y2, x1, H_in, W_in, indices[${Zs}], indices[${pn}], border); + let p22 = pixel_at_grid(y2, x2, H_in, W_in, indices[${Zs}], indices[${pn}], border); + + let dx2 = ${r}(f32(x2) - x); + let dx1 = ${r}(x - f32(x1)); + let dy2 = ${r}(f32(y2) - y); + let dy1 = ${r}(y - f32(y1)); + let result = dy2 * (dx2 * p11 + dx1 * p12) + dy1 * (dx2 * p21 + dx1 * p22); + `;case"bicubic":return` + let x0 = i32(floor(x)) - 1; + let y0 = i32(floor(y)) - 1; + var p: mat4x4<${r}>; + for (var h = 0; h < 4; h++) { + for (var w = 0; w < 4; w++) { + p[h][w] = pixel_at_grid(h + y0, w + x0, H_in, W_in, indices[${Zs}], indices[${pn}], border); + } + } + + let dx = x - f32(x0 + 1); + let dy = y - f32(y0 + 1); + let result = gs_bicubic_interpolate(p, dx, dy); + `;default:throw new Error(`mode ${t.mode} is not supported`)}})()+`${e.setByOffset("global_idx","result")}`,Ly=(e,r)=>{let t=Ne("x",e[0].dataType,e[0].dims.length),i=[e[1].dims[0],e[1].dims[1],e[1].dims[2]],o=Ne("grid",e[1].dataType,i.length,2),n=[e[0].dims[0],e[0].dims[1],e[1].dims[1],e[1].dims[2]];r.format==="NHWC"&&(n=[e[0].dims[0],e[1].dims[1],e[1].dims[2],e[0].dims[3]],[Zs,pn,ai,oi]=[0,3,1,2]);let u=gt("output",e[0].dataType,n.length),d=t.type.value,h=je.size(n),f=[{type:12,data:h},...Et(e[0].dims,i,n)],y=m=>` + ${m.registerUniform("output_size","u32").declareVariables(t,o,u)} + ${Iy} + ${Ay(d)} + ${Oy(r)} + ${Fy(r)} + ${Dy(t,d,r)} + + ${m.mainStart()} + ${m.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let H_in = i32(uniforms.x_shape[${ai}]); + let W_in = i32(uniforms.x_shape[${oi}]); + + ${r.alignCorners===0?` + let x_min = -0.5; + let x_max = f32(W_in) - 0.5; + let y_min = -0.5; + let y_max = f32(H_in) - 0.5; + `:` + let x_min = 0.0; + let x_max = f32(W_in) - 1.0; + let y_min = 0.0; + let y_max = f32(H_in) - 1.0; + `}; + let border = vec4(x_min, y_min, x_max, y_max); + + let indices = ${u.offsetToIndices("global_idx")}; + var grid_indices = vec3(indices[${Zs}], indices[${ai}], indices[${oi}]); + let nxy = ${o.getByIndices("grid_indices")}; + var x = gs_denormalize(f32(nxy[0]), W_in); + var y = gs_denormalize(f32(nxy[1]), H_in); + + ${jy(u,d,r)} + }`;return{name:"GridSample",shaderCache:{hint:`${r.cacheKey}`,inputDependencies:["type","type"]},getRunData:m=>{let g=je.size(n);return{outputs:[{dims:n,dataType:m[0].dataType}],dispatchGroup:{x:Math.ceil(g/64)},programUniforms:f}},getShaderSource:y}},RM=(e,r)=>{$y(e.inputs),e.compute(Ly(e.inputs,r))},NM=e=>Zt({alignCorners:e.align_corners,mode:e.mode,paddingMode:e.padding_mode,format:e.format})}),ts,zy,VM,am,By,_d,WM,UM=Ye(()=>{$t(),Lt(),br(),t_(),n_(),zt(),On(),ts=(e,r)=>e.length>r&&e[r].dims.length>0?e[r]:void 0,zy=(e,r)=>{let t=e[0],i=ts(e,1),o=ts(e,2),n=ts(e,3),u=ts(e,4),d=ts(e,5),h=ts(e,6),f=ts(e,7);if(t.dims.length!==3&&t.dims.length!==5)throw new Error("Input query is expected to have 3 or 5 dimensions");let y=t.dims[0],m=t.dims[1],g=t.dims.length===3?t.dims[2]:r.numHeads*t.dims[4],T=m,M=0,I=0,z=Math.floor(g/r.numHeads);if(h&&f&&je.size(h.dims)&&je.size(f.dims)){if(h.dims.length!==4)throw new Error('Input "past_key" is expected to have 4 dimensions');if(h.dims[0]!==y||h.dims[1]!==r.numHeads||h.dims[3]!==z)throw new Error('Input "past_key" shape (batch_size, num_heads, past_sequence_length, head_size)');if(f.dims[0]!==y||f.dims[1]!==r.numHeads||f.dims[3]!==z)throw new Error('Input "past_value" shape (batch_size, num_heads, past_sequence_length, head_size)');if(h.dims[2]!==f.dims[2])throw new Error('Input "past_key" and "past_value" shall have same dim 2 (past_sequence_length)');if(f.dims.length!==4)throw new Error('Input "past_value" is expected to have 4 dimensions');M=h.dims[2],I=h.dims[2]}else if(h&&je.size(h.dims)||f&&je.size(f.dims))throw new Error('Input "past_key" and "past_value" shall be both present or both absent');let E;if(i&&je.size(i.dims)>0){if(t.dims.length!==3)throw new Error('Input "query" is expected to have 3 dimensions when key is given');if(i.dims.length<3||i.dims.length>5)throw new Error('Input "key" is expected to have 3, 4, or 5 dimensions');if(t.dims[0]!==i.dims[0])throw new Error('Input "query" and "key" shall have same dim 0 (batch size)');if(i.dims.length===3){if(i.dims[2]!==t.dims[2])throw new Error('Input "query" and "key" shall have same dim 2 (hidden_size)');E=2,T=i.dims[1]}else if(i.dims.length===5){if(i.dims[2]!==r.numHeads||i.dims[3]!==2||i.dims[4]!==z)throw new Error('Expect "key" shape (batch_size, kv_sequence_length, num_heads, 2, head_size) for packed kv');if(o)throw new Error('Expect "value" be none when "key" has packed kv format.');E=5,T=i.dims[1]}else{if(i.dims[1]!==r.numHeads||i.dims[3]!==z)throw new Error('Expect "key" shape (batch_size, num_heads, kv_sequence_length, head_size) for past_key');E=0,T=i.dims[2]}}else{if(t.dims.length!==5)throw new Error('Input "query" is expected to have 5 dimensions when key is empty');if(t.dims[2]!==r.numHeads||t.dims[3]!==3)throw new Error('Expect "query" shape (batch_size, kv_sequence_length, num_heads, 3, head_size) for packed kv');E=3}if(n&&je.size(n.dims)>0){if(n.dims.length!==1)throw new Error('Input "bias" is expected to have 1 dimension');if(i&&i.dims.length===5&&i.dims[3]===2)throw new Error("bias is not allowed for packed kv.")}let C=M+T,D=0;if(u&&je.size(u.dims)>0){D=8;let k=u.dims;throw k.length===1?k[0]===y?D=1:k[0]===3*y+2&&(D=3):k.length===2&&k[0]===y&&k[1]===C&&(D=5),D===8?new Error('Input "key_padding_mask" shape shall be (batch_size) or (batch_size, total_sequence_length)'):new Error("Mask not supported")}let A=!1,$=g;if(o&&je.size(o.dims)>0){if(o.dims.length!==3&&o.dims.length!==4)throw new Error('Input "value" is expected to have 3 or 4 dimensions');if(t.dims[0]!==o.dims[0])throw new Error('Input "query" and "value" shall have same dim 0 (batch_size)');if(o.dims.length===3){if(T!==o.dims[1])throw new Error('Input "key" and "value" shall have the same dim 1 (kv_sequence_length)');$=o.dims[2]}else{if(T!==o.dims[2])throw new Error('Input "key" and "value" shall have the same dim 2 (kv_sequence_length)');$=o.dims[1]*o.dims[3],A=!0}}let P=!1;if(u&&je.size(u.dims)>0)throw new Error("Key padding mask is not supported");if(d&&je.size(d.dims)>0){if(d.dims.length!==4)throw new Error('Input "attention_bias" is expected to have 4 dimensions');if(d.dims[0]!==y||d.dims[1]!==r.numHeads||d.dims[2]!==m||d.dims[3]!==C)throw new Error('Expect "attention_bias" shape (batch_size, num_heads, sequence_length, total_sequence_length)')}return{batchSize:y,sequenceLength:m,pastSequenceLength:M,kvSequenceLength:T,totalSequenceLength:C,maxSequenceLength:I,inputHiddenSize:0,hiddenSize:g,vHiddenSize:$,headSize:z,vHeadSize:Math.floor($/r.numHeads),numHeads:r.numHeads,isUnidirectional:!1,pastPresentShareBuffer:!1,maskFilterValue:r.maskFilterValue,maskType:D,scale:r.scale,broadcastResPosBias:P,passPastInKv:A,qkvFormat:E}},VM=e=>Zt({...e}),am=Zt({perm:[0,2,1,3]}),By=(e,r,t,i,o,n,u)=>{let d=[i,o,n],h=je.size(d),f=[{type:12,data:h},{type:12,data:u},{type:12,data:n}],y=m=>{let g=gt("qkv_with_bias",r.dataType,d),T=Ne("qkv",r.dataType,d),M=Ne("bias",t.dataType,d),I=[{name:"output_size",type:"u32"},{name:"bias_offset",type:"u32"},{name:"hidden_size",type:"u32"}];return` + ${m.registerUniforms(I).declareVariables(T,M,g)} + ${m.mainStart()} + ${m.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let bias_offset_idx = (global_idx % uniforms.hidden_size) + uniforms.bias_offset; + + qkv_with_bias[global_idx] = qkv[global_idx] + bias[bias_offset_idx]; + }`};return e.compute({name:"MultiHeadAttentionAddBias",shaderCache:{inputDependencies:["type","type"]},getRunData:()=>({outputs:[{dims:d,dataType:r.dataType,gpuDataType:0}],dispatchGroup:{x:Math.ceil(h/64)},programUniforms:f}),getShaderSource:y},{inputs:[r,t],outputs:[-1]})[0]},_d=(e,r,t,i,o,n,u,d)=>{let h=n;if(u&&je.size(u.dims)>0){if(i===1)throw new Error("AddBiasReshape is not implemented. Please export your model with packed QKV or KV");return h=By(e,n,u,r,i,t*o,d),h=h.reshape([r,i,t,o]),t===1||i===1?h:e.compute(_s(h,am.perm),{inputs:[h],outputs:[-1]})[0]}else return n.dims.length===3&&(h=n.reshape([r,i,t,o])),t===1||i===1?h:e.compute(_s(h,am.perm),{inputs:[h],outputs:[-1]})[0]},WM=(e,r)=>{let t=zy(e.inputs,r),i=e.inputs[0],o=ts(e.inputs,1),n=ts(e.inputs,2),u=ts(e.inputs,3),d=ts(e.inputs,4),h=ts(e.inputs,5),f=ts(e.inputs,6),y=ts(e.inputs,7);if(i.dims.length===5)throw new Error("Packed QKV is not implemented");if((o==null?void 0:o.dims.length)===5)throw new Error("Packed KV is not implemented");let m=o&&n&&o.dims.length===4&&n.dims.length===4,g=_d(e,t.batchSize,t.numHeads,t.sequenceLength,t.headSize,i,u,0);if(m)return Md(e,g,o,n,d,void 0,f,y,h,t);if(!o||!n)throw new Error("key and value must be provided");let T=_d(e,t.batchSize,t.numHeads,t.kvSequenceLength,t.headSize,o,u,t.hiddenSize),M=_d(e,t.batchSize,t.numHeads,t.kvSequenceLength,t.vHeadSize,n,u,2*t.hiddenSize);Md(e,g,T,M,d,void 0,f,y,h,t)}}),Ry,Ny,Vy,Wy,Rm,GM,KM,HM=Ye(()=>{$t(),Lt(),br(),zt(),Ry=e=>{if(!e||e.length<1)throw new Error("too few inputs")},Ny=(e,r)=>{let t=[],i=r.numOutputs;return e[1].dims[0]>0&&(e[1].getBigInt64Array().forEach(o=>t.push(Number(o))),i=t.length),Zt({numOutputs:i,axis:r.axis,splitSizes:t})},Vy=e=>` +fn calculateOutputIndex(index: u32) -> u32 { + for (var i: u32 = 0u; i < ${e}u; i += 1u ) { + if (index < ${Mt("uniforms.size_in_split_axis","i",e)}) { + return i; + } + } + return ${e}u; +}`,Wy=e=>{let r=e.length,t=[];for(let i=0;i{let t=e[0].dims,i=je.size(t),o=e[0].dataType,n=je.normalizeAxis(r.axis,t.length),u=new Array(r.numOutputs),d=Ne("input",o,t.length),h=new Array(r.numOutputs),f=[],y=[],m=0,g=[{type:12,data:i}];for(let M=0;M` + ${M.registerUniform("input_size","u32").registerUniform("size_in_split_axis","u32",h.length).declareVariables(d,...u)} + ${Vy(h.length)} + ${Wy(u)} + + ${M.mainStart()} + ${M.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.input_size")} + + var indices = ${d.offsetToIndices("global_idx")}; + var index = ${d.indicesGet("indices",n)}; + let output_number = calculateOutputIndex(index); + if (output_number != 0) { + index -= ${Mt("uniforms.size_in_split_axis","output_number - 1u",h.length)}; + ${d.indicesSet("indices",n,"index")}; + } + writeBufferData(output_number, indices, global_idx); + }`;return{name:"Split",shaderCache:{hint:r.cacheKey,inputDependencies:["rank"]},getShaderSource:T,getRunData:()=>({outputs:f,dispatchGroup:{x:Math.ceil(i/64)},programUniforms:g})}},GM=(e,r)=>{Ry(e.inputs);let t=e.inputs.length===1?r:Ny(e.inputs,r);e.compute(Rm(e.inputs,t),{inputs:[0]})},KM=e=>{let r=e.axis,t=e.splitSizes,i=e.numOutputs<0?t.length:e.numOutputs;if(i!==t.length)throw new Error("numOutputs and splitSizes lengh must be equal");return Zt({axis:r,numOutputs:i,splitSizes:t})}}),Uy,sf,qM,QM=Ye(()=>{$t(),Lt(),br(),zt(),Uy=(e,r)=>{let[t,i,o,n]=e,{numHeads:u,rotaryEmbeddingDim:d}=r;if(t.dims.length!==3&&t.dims.length!==4)throw new Error(`Input 'x' is expected to have 3 or 4 dimensions, got ${t.dims.length}`);if(!je.areEqual(i.dims,[])&&!je.areEqual(i.dims,[1])&&i.dims.length!==2)throw new Error(`Input 'position_ids' is expected to have 0, 1, or 2 dimensions, got ${i.dims.length}`);if(o.dims.length!==2)throw new Error(`Input 'cos_cache' is expected to have 2 dimensions, got ${o.dims.length}`);if(n.dims.length!==2)throw new Error(`Input 'sin_cache' is expected to have 2 dimensions, got ${n.dims.length}`);if(!je.areEqual(o.dims,n.dims))throw new Error("Inputs 'cos_cache' and 'sin_cache' are expected to have the same shape");if(d>0&&u===0)throw new Error("num_heads must be provided if rotary_embedding_dim is specified");let h=t.dims[0],f=t.dims[t.dims.length-2],y=o.dims[0],m=je.sizeFromDimension(t.dims,1)/f,g=d===0?o.dims[1]*2:m/u;if(d>g)throw new Error("rotary_embedding_dim must be less than or equal to head_size");if(i.dims.length===2){if(h!==i.dims[0])throw new Error(`Input 'position_ids' dimension 0 should be of size batch_size, got ${i.dims[0]}`);if(f!==i.dims[1])throw new Error(`Input 'position_ids' dimension 1 should be of size sequence_length, got ${i.dims[1]}`)}if(g/2!==o.dims[1]&&d/2!==o.dims[1])throw new Error(`Input 'cos_cache' dimension 1 should be same as head_size / 2 or rotary_embedding_dim / 2, got ${o.dims[1]}`);if(f>y)throw new Error("Updating cos_cache and sin_cache in RotaryEmbedding is not currently supported")},sf=(e,r)=>{let{interleaved:t,numHeads:i,rotaryEmbeddingDim:o,scale:n}=r,u=e[0].dims[0],d=je.sizeFromDimension(e[0].dims,1),h=e[0].dims[e[0].dims.length-2],f=d/h,y=e[2].dims[1],m=o===0?y*2:f/i,g=new Array(u,h,f/m,m-y),T=je.computeStrides(g),M=[{type:1,data:n},{type:12,data:g},{type:12,data:T},...e[0].dims.length===3?new Array({type:12,data:[d,f,m,1]}):[],...e[0].dims.length===4?new Array({type:12,data:[d,m,h*m,1]}):[],...Et(e[0].dims,e[1].dims,e[2].dims,e[3].dims,e[0].dims)],I=z=>{let E=Ne("input",e[0].dataType,e[0].dims.length),C=Ne("position_ids",e[1].dataType,e[1].dims.length),D=Ne("cos_cache",e[2].dataType,e[2].dims.length),A=Ne("sin_cache",e[3].dataType,e[3].dims.length),$=gt("output",e[0].dataType,e[0].dims.length);return z.registerUniforms([{name:"scale",type:"f32"},{name:"global_shape",type:"u32",length:g.length},{name:"global_strides",type:"u32",length:T.length},{name:"input_output_strides",type:"u32",length:T.length}]),` + ${z.declareVariables(E,C,D,A,$)} + + ${z.mainStart(Va)} + let half_rotary_emb_dim = uniforms.${D.name}_shape[1]; + let bsnh = global_idx / uniforms.global_strides % uniforms.global_shape; + let size = uniforms.global_shape[0] * uniforms.global_strides[0]; + ${z.guardAgainstOutOfBoundsWorkgroupSizes("size")} + + if (bsnh[3] < half_rotary_emb_dim) { + let position_ids_idx = + ${C.broadcastedIndicesToOffset("bsnh.xy",gt("",C.type.tensor,2))}; + let position_id = + u32(${C.getByOffset("position_ids_idx")}) + select(0, bsnh[1], position_ids_idx == 0); + let i = dot(bsnh, uniforms.input_output_strides) + select(0, bsnh[3], ${t}); + let j = i + select(half_rotary_emb_dim, 1, ${t}); + let re = ${E.getByOffset("i")} * ${D.get("position_id","bsnh[3]")} - + ${E.getByOffset("j")} * ${A.get("position_id","bsnh[3]")}; + ${$.setByOffset("i","re")} + let im = ${E.getByOffset("i")} * ${A.get("position_id","bsnh[3]")} + + ${E.getByOffset("j")} * ${D.get("position_id","bsnh[3]")}; + ${$.setByOffset("j","im")} + } else { + let k = dot(bsnh, uniforms.input_output_strides) + half_rotary_emb_dim; + ${$.setByOffset("k",E.getByOffset("k"))} + } + }`};return{name:"RotaryEmbedding",shaderCache:{hint:Zt({interleaved:t}).cacheKey,inputDependencies:["rank","rank","rank","rank"]},getShaderSource:I,getRunData:()=>({outputs:[{dims:e[0].dims,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(je.size(g)/Va)},programUniforms:M})}},qM=(e,r)=>{Uy(e.inputs,r),e.compute(sf(e.inputs,r))}}),Gy,Ky,om,Hy,XM,PT=Ye(()=>{br(),$t(),n_(),UM(),HM(),On(),QM(),zt(),Gy=(e,r)=>{if(r.doRotary&&e.length<=7)throw new Error("cos_cache and sin_cache inputs are required if do_rotary is specified");let t=e[0],i=e[1],o=e[2],n=e[3],u=e[4];if(r.doRotary!==0&&e.length<=7)throw new Error("cos_cast and sin_cache are expected if do_rotary attribute is non-zero");if(r.localWindowSize!==-1)throw new Error("Local attention is not supported");if(r.softcap!==0)throw new Error("Softcap is not supported");if(r.rotaryInterleaved!==0)throw new Error("Rotary interleaved is not supported");if(r.smoothSoftmax)throw new Error("Smooth softmax is not supported");if(t.dims.length!==3&&t.dims.length!==5)throw new Error("Input query is expected to have 3 or 5 dimensions");let d=!1,h=t.dims[0],f=t.dims[1],y=t.dims.length===3?d?t.dims[2]/3:t.dims[2]:r.numHeads*t.dims[4],m=f,g=0,T=!i||i.dims.length===0,M=Math.floor(T?y/(r.numHeads+2*r.kvNumHeads):y/r.numHeads);T&&(y=M*r.numHeads);let I=n&&n.dims.length!==0,z=u&&u.dims.length!==0;if(I&&n.dims.length===4&&n.dims[0]===h&&n.dims[1]!==r.kvNumHeads&&n.dims[2]===r.kvNumHeads&&n.dims[3]===M)throw new Error("BSNH pastKey/pastValue is not supported");if(I&&z){if(n.dims.length!==4)throw new Error('Input "past_key" is expected to have 4 dimensions');if(u.dims.length!==4)throw new Error('Input "past_value" is expected to have 4 dimensions');g=n.dims[2]}else if(I||z)throw new Error('Input "past_key" and "past_value" shall be both present or both absent');let E=1;if(i&&i.dims.length>0){if(t.dims.length!==3)throw new Error('Input "query" is expected to have 3 dimensions when key is given');if(i.dims.length<3||i.dims.length>5)throw new Error('Input "key" is expected to have 3, 4, or 5 dimensions');if(t.dims[0]!==i.dims[0])throw new Error('Input "query" and "key" shall have same dim 0 (batch size)');if(i.dims.length===3){if(t.dims[2]%i.dims[2]!==0)throw new Error('Dimension 2 of "query" should be a multiple of "key"');m=i.dims[1]}else if(i.dims.length===5){if(i.dims[2]!==r.numHeads||i.dims[3]!==2||i.dims[4]!==M)throw new Error('Expect "key" shape (batch_size, kv_sequence_length, num_heads, 2, head_size) for packed kv');if(o)throw new Error('Expect "value" be none when "key" has packed kv format.');m=i.dims[1]}else{if(i.dims[1]!==r.numHeads||i.dims[3]!==M)throw new Error('Expect "key" shape (batch_size, num_heads, kv_sequence_length, head_size) for past_key');m=i.dims[2]}}else{if(t.dims.length!==3&&t.dims.length!==5)throw new Error('Input "query" is expected to have 3 or 5 dimensions when key is empty');if(t.dims.length===5&&(t.dims[2]!==r.numHeads||t.dims[3]!==3))throw new Error('Expect "query" shape (batch_size, kv_sequence_length, num_heads, 3, head_size) for packed kv');E=3}let C=0,D=!1,A=r.kvNumHeads?M*r.kvNumHeads:y;if(o&&o.dims.length>0){if(o.dims.length!==3&&o.dims.length!==4)throw new Error('Input "value" is expected to have 3 or 4 dimensions');if(t.dims[0]!==o.dims[0])throw new Error('Input "query" and "value" shall have same dim 0 (batch_size)');if(o.dims.length===3){if(m!==o.dims[1])throw new Error('Input "key" and "value" shall have the same dim 1 (kv_sequence_length)');A=o.dims[2]}else{if(m!==o.dims[2])throw new Error('Input "past_key" and "past_value" shall have the same dim 2 (kv_sequence_length)');A=o.dims[1]*o.dims[3],D=!0}}let $=e.length>4?e[5]:void 0;if($&&$.dims.length!==1&&$.dims[0]!==h)throw new Error('Input "seqlens" is expected to have 1 dimension and the same dim 0 as batch_size');return{batchSize:h,sequenceLength:f,pastSequenceLength:g,kvSequenceLength:m,totalSequenceLength:-1,maxSequenceLength:-1,inputHiddenSize:0,hiddenSize:y,vHiddenSize:A,headSize:M,vHeadSize:Math.floor(A/r.kvNumHeads),numHeads:r.numHeads,kvNumHeads:r.kvNumHeads,nReps:r.numHeads/r.kvNumHeads,pastPresentShareBuffer:!1,maskType:C,scale:r.scale,broadcastResPosBias:!1,passPastInKv:D,qkvFormat:E}},Ky=Zt({perm:[0,2,1,3]}),om=(e,r,t)=>{let i=r,o=t.kvNumHeads;return r.dims.length===3&&t.kvSequenceLength!==0&&(i=r.reshape([t.batchSize,t.kvSequenceLength,o,t.headSize]),i=e.compute(_s(i,Ky.perm),{inputs:[i],outputs:[-1]})[0]),i},Hy=(e,r,t,i)=>{let o=7,n=["type","type"],u=[e*r],d=e*r,h=[{type:12,data:d},{type:12,data:r},{type:12,data:e}],f=y=>{let m=Ne("seq_lens",t.dataType,t.dims),g=Ne("total_seq_lens",i.dataType,i.dims),T=gt("pos_ids",o,u),M=[{name:"output_size",type:"u32"},{name:"sequence_length",type:"u32"},{name:"batch_size",type:"u32"}];return` + ${y.registerUniforms(M).declareVariables(m,g,T)} + ${y.mainStart()} + ${y.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let total_sequence_length = u32(${g.getByOffset("0")}); + let is_subsequent_prompt = uniforms.sequence_length > 1 && uniforms.sequence_length != total_sequence_length; + let is_first_prompt = !is_subsequent_prompt && uniforms.sequence_length == total_sequence_length; + let batch_idx = global_idx / uniforms.sequence_length; + let sequence_idx = i32(global_idx % uniforms.sequence_length); + var pos_id: i32 = 0; + let seqlen = ${m.getByOffset("batch_idx")}; + let total_seqlen = seqlen + 1; + if (is_first_prompt) { + if (sequence_idx < total_seqlen) { + pos_id = sequence_idx; + } else { + pos_id = 1; + } + ${T.setByOffset("global_idx","pos_id")} + } else if (is_subsequent_prompt) { + let past_seqlen = total_seqlen - i32(uniforms.sequence_length); + if (past_seqlen + sequence_idx < total_seqlen) { + pos_id = past_seqlen + sequence_idx; + } else { + pos_id = 1; + } + ${T.setByOffset("global_idx","pos_id")} + } else if (global_idx < uniforms.batch_size) { + ${T.setByOffset("global_idx","seqlen")} + }; + } + `};return{name:"GeneratePositionIds",shaderCache:{hint:`${e};${r}`,inputDependencies:n},getRunData:()=>({outputs:[{dims:u,dataType:o}],dispatchGroup:{x:Math.ceil(d/64)},programUniforms:h}),getShaderSource:f}},XM=(e,r)=>{var A;let t=Gy(e.inputs,r);if(e.inputs[0].dims.length===5)throw new Error("Packed QKV is not implemented");if(((A=e.inputs[1])==null?void 0:A.dims.length)===5)throw new Error("Packed KV is not implemented");let i=e.inputs[0],o=e.inputs[1]&&e.inputs[1].dims.length>0?e.inputs[1]:void 0,n=e.inputs[2]&&e.inputs[2].dims.length>0?e.inputs[2]:void 0,u=e.inputs[3]&&e.inputs[3].dims.length!==0?e.inputs[3]:void 0,d=e.inputs[4]&&e.inputs[4].dims.length!==0?e.inputs[4]:void 0,h=e.inputs.length>4?e.inputs[5]:void 0,f=e.inputs.length>5?e.inputs[6]:void 0,y=t.kvNumHeads?t.kvNumHeads:t.numHeads,m=Zt({axis:2,numOutputs:3,splitSizes:[t.numHeads*t.headSize,y*t.headSize,y*t.headSize]}),[g,T,M]=!o&&!n?e.compute(Rm([i],m),{inputs:[i],outputs:[-1,-1,-1]}):[i,o,n],I,z;if(r.doRotary){let $=e.compute(Hy(t.batchSize,t.sequenceLength,h,f),{inputs:[h,f],outputs:[-1]})[0],P=e.inputs[7],k=e.inputs[8],O=Zt({interleaved:r.rotaryInterleaved!==0,numHeads:t.numHeads,rotaryEmbeddingDim:0,scale:r.scale}),R=[g,$,P,k],U=[-1];I=e.compute(sf(R,O),{inputs:R,outputs:U})[0],R.splice(0,1,T);let te=Zt({interleaved:r.rotaryInterleaved!==0,numHeads:t.kvNumHeads,rotaryEmbeddingDim:0,scale:r.scale});z=e.compute(sf(R,te),{inputs:R,outputs:U})[0]}let E=_d(e,t.batchSize,t.numHeads,t.sequenceLength,t.headSize,r.doRotary?I:g,void 0,0),C=om(e,r.doRotary?z:T,t),D=om(e,M,t);Md(e,E,C,D,void 0,void 0,u,d,void 0,t,h,f)}}),lm,qy,Qy,JM,ST=Ye(()=>{$t(),Lt(),On(),zt(),lm=(e,r,t,i,o,n,u,d)=>{let h=gr(n),f=h===1?"f32":`vec${h}f`,y=h===1?"vec2f":`mat2x${h}f`,m=o*u,g=64;m===1&&(g=256);let T=[o,u,n/h],M=[o,u,2],I=["rank","type","type"],z=[];z.push(...Et(T,M));let E=C=>{let D=Ne("x",r.dataType,3,h),A=Ne("scale",t.dataType,t.dims),$=Ne("bias",i.dataType,i.dims),P=gt("output",1,3,2),k=[D,A,$,P];return` + var workgroup_shared : array<${y}, ${g}>; + const workgroup_size = ${g}u; + ${C.declareVariables(...k)} + ${C.mainStart(g)} + let batch = workgroup_index / uniforms.x_shape[1]; + let channel = workgroup_index % uniforms.x_shape[1]; + let hight = uniforms.x_shape[2]; + // initialize workgroup memory + var sum = ${f}(0); + var squared_sum = ${f}(0); + for (var h = local_idx; h < hight; h += workgroup_size) { + let value = ${f}(${D.get("batch","channel","h")}); + sum += value; + squared_sum += value * value; + } + workgroup_shared[local_idx] = ${y}(sum, squared_sum); + workgroupBarrier(); + + for (var currSize = workgroup_size >> 1; currSize > 0; currSize = currSize >> 1) { + if (local_idx < currSize) { + workgroup_shared[local_idx] = workgroup_shared[local_idx] + workgroup_shared[local_idx + currSize]; + } + workgroupBarrier(); + } + if (local_idx == 0) { + let sum_final = ${An("workgroup_shared[0][0]",h)} / f32(hight * ${h}); + let squared_sum_final = ${An("workgroup_shared[0][1]",h)} / f32(hight * ${h}); + + let inv_std_dev = inverseSqrt(squared_sum_final - sum_final * sum_final + f32(${d})); + let channel_scale = inv_std_dev * f32(scale[channel]); + let channel_shift = f32(bias[channel]) - sum_final * channel_scale; + output[workgroup_index] = vec2f(channel_scale, channel_shift); + } + }`};return e.compute({name:"InstanceNormComputeChannelScaleShift",shaderCache:{hint:`${h};${d};${g}`,inputDependencies:I},getRunData:()=>({outputs:[{dims:M,dataType:1}],dispatchGroup:{x:m},programUniforms:z}),getShaderSource:E},{inputs:[r,t,i],outputs:[-1]})[0]},qy=(e,r,t)=>{let i=r[0].dims,o=i,n=2,u=i[0],d=i[1],h=je.sizeFromDimension(i,n),f=gr(h),y=je.size(o)/f,m=lm(e,r[0],r[1],r[2],u,h,d,t.epsilon),g=[u,d,h/f],T=[u,d],M=["type","none"],I=z=>{let E=Ne("x",r[0].dataType,g.length,f),C=Ne("scale_shift",1,T.length,2),D=gt("output",r[0].dataType,g.length,f),A=[E,C,D];return` + ${z.registerUniform("output_size","u32").declareVariables(...A)} + ${z.mainStart()} + ${z.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let outputIndices = ${D.offsetToIndices("global_idx")}; + let batch = outputIndices[0]; + let channel = outputIndices[1]; + let scale_shift = ${C.getByIndices("vec2(batch, channel)")}; + let value = ${E.getByOffset("global_idx")} * ${D.type.value}(scale_shift.x) + ${D.type.value}(scale_shift.y); + ${D.setByOffset("global_idx","value")}; + }`};e.compute({name:"InstanceNormalization",shaderCache:{hint:`${f}`,inputDependencies:M},getRunData:()=>({outputs:[{dims:o,dataType:r[0].dataType}],dispatchGroup:{x:Math.ceil(y/64)},programUniforms:[{type:12,data:y},...Et(g,T,g)]}),getShaderSource:I},{inputs:[r[0],m]})},Qy=(e,r,t)=>{let i=r[0].dims,o=i,n=i[0],u=i[i.length-1],d=je.sizeFromDimension(i,1)/u,h=gr(u),f=je.size(o)/h,y=[{type:12,data:d},{type:12,data:Math.floor(u/h)}],m=["type","type"],g=!1,T=[0,i.length-1];for(let E=0;Ei[T[C]])),I=lm(e,M,r[1],r[2],n,d,u,t.epsilon),z=E=>{let C=Lr(r[0].dataType),D=h===1?"vec2f":`mat${h}x2f`,A=k=>{let O=k===0?"x":"y",R=h===1?"f32":`vec${h}f`;switch(h){case 1:return`${C}(${R}(scale.${O}))`;case 2:return`vec2<${C}>(${R}(scale[0].${O}, scale[1].${O}))`;case 4:return`vec4<${C}>(${R}(scale[0].${O}, scale[1].${O}, scale[2].${O}, scale[3].${O}))`;default:throw new Error(`Not supported compoents ${h}`)}},$=Ne("input",r[0].dataType,r[0].dims,h),P=gt("output",r[0].dataType,o,h);return` + @group(0) @binding(0) var input : array<${$.type.storage}>; + @group(0) @binding(1) var scale_input : array<${D}>; + @group(0) @binding(2) var output : array<${P.type.storage}>; + struct Uniforms {H: u32, C : u32}; + @group(0) @binding(3) var uniforms: Uniforms; + + ${E.mainStart()} + let current_image_number = global_idx / (uniforms.C * uniforms.H); + let current_channel_number = global_idx % uniforms.C; + + let scale_offset = current_image_number * uniforms.C + current_channel_number; + let scale = scale_input[scale_offset]; + output[global_idx] = fma(input[global_idx], ${A(0)}, ${A(1)}); + }`};e.compute({name:"InstanceNormalizationNHWC",shaderCache:{hint:`${h}`,inputDependencies:m},getRunData:()=>({outputs:[{dims:o,dataType:r[0].dataType}],dispatchGroup:{x:Math.ceil(f/64)},programUniforms:y}),getShaderSource:z},{inputs:[r[0],I]})},JM=(e,r)=>{r.format==="NHWC"?Qy(e,e.inputs,r):qy(e,e.inputs,r)}}),Xy,Jy,YM,kT=Ye(()=>{$t(),Lt(),zt(),Xy=e=>{if(!e||e.length<2)throw new Error("layerNorm requires at least 2 inputs.")},Jy=(e,r,t)=>{let i=r.simplified,o=e[0].dims,n=e[1],u=!i&&e[2],d=o,h=je.normalizeAxis(r.axis,o.length),f=je.sizeToDimension(o,h),y=je.sizeFromDimension(o,h),m=je.size(n.dims),g=u?je.size(u.dims):0;if(m!==y||u&&g!==y)throw new Error(`Size of X.shape()[axis:] == ${y}. + Size of scale and bias (if provided) must match this. + Got scale size of ${m} and bias size of ${g}`);let T=[];for(let $=0;$1,C=t>2,D=$=>{let P=Lr(e[0].dataType),k=[Ne("x",e[0].dataType,e[0].dims,M),Ne("scale",n.dataType,n.dims,M)];u&&k.push(Ne("bias",u.dataType,u.dims,M)),k.push(gt("output",e[0].dataType,d,M)),E&&k.push(gt("mean_data_output",1,T)),C&&k.push(gt("inv_std_output",1,T));let O=[{name:"norm_count",type:"u32"},{name:"norm_size",type:"f32"},{name:"norm_size_vectorized",type:"u32"},{name:"epsilon",type:"f32"}];return` + ${$.registerUniforms(O).declareVariables(...k)} + ${$.mainStart()} + ${$.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.norm_count")} + let offset = global_idx * uniforms.norm_size_vectorized; + var mean_vector = ${Im("f32",M)}; + var mean_square_vector = ${Im("f32",M)}; + + for (var h: u32 = 0u; h < uniforms.norm_size_vectorized; h++) { + let value = ${Ba(P,M,"x[h + offset]")}; + mean_vector += value; + mean_square_vector += value * value; + } + let mean = ${An("mean_vector",M)} / uniforms.norm_size; + let inv_std_dev = inverseSqrt(${An("mean_square_vector",M)} / uniforms.norm_size ${i?"":"- mean * mean"} + uniforms.epsilon); + + for (var j: u32 = 0; j < uniforms.norm_size_vectorized; j++) { + let f32input = ${Ba(P,M,"x[j + offset]")}; + let f32scale = ${Ba(P,M,"scale[j]")}; + output[j + offset] = ${k[0].type.value}((f32input ${i?"":"- mean"}) * inv_std_dev * f32scale + ${u?`+ ${Ba(P,M,"bias[j]")}`:""} + ); + } + + ${E?"mean_data_output[global_idx] = mean":""}; + ${C?"inv_std_output[global_idx] = inv_std_dev":""}; + }`},A=[{dims:d,dataType:e[0].dataType}];return E&&A.push({dims:T,dataType:1}),C&&A.push({dims:T,dataType:1}),{name:"LayerNormalization",shaderCache:{hint:`${M};${t};${i}`,inputDependencies:I},getRunData:()=>({outputs:A,dispatchGroup:{x:Math.ceil(f/64)},programUniforms:z}),getShaderSource:D}},YM=(e,r)=>{Xy(e.inputs),e.compute(Jy(e.inputs,r,e.outputCount))}}),Yy,ZM,$T=Ye(()=>{Lt(),c_(),u_(),Yy=e=>{if(!e||e.length!==2)throw new Error("MatMul requires 2 inputs.");if(e[0].dims[e[0].dims.length-1]!==e[1].dims[e[1].dims.length-2])throw new Error("shared dimension does not match.")},ZM=e=>{Yy(e.inputs);let r=Na.calcShape(e.inputs[0].dims,e.inputs[1].dims,!0);if(!r)throw new Error("Can't use matmul on the given tensors");let t=r[r.length-1],i=e.inputs[0].dims[e.inputs[0].dims.length-1];if(t<8&&i<8)e.compute(l_(e.inputs,{activation:""},r));else{let o=r[r.length-2],n=je.size(e.inputs[0].dims.slice(0,-2)),u=je.size(e.inputs[1].dims.slice(0,-2));if(n!==1&&o===1&&u===1){let d=e.inputs[0].reshape([1,n,i]),h=e.inputs[1].reshape([1,i,t]),f=[1,n,t],y=[d,h];e.compute(rf(y,{activation:""},r,f),{inputs:y})}else e.compute(rf(e.inputs,{activation:""},r))}}}),Zy,ew,tw,ev,tv,IT=Ye(()=>{$t(),Lt(),br(),zt(),Zy=(e,r)=>{if(e.length<3||e.length>4)throw new Error("MatMulNBits requires 3 or 4 inputs");let t=e[0],i=t.dims.length;if(t.dims[i-1]!==r.k)throw new Error("The last dim of input shape does not match the k value");let o=Math.floor((r.k+r.blockSize-1)/r.blockSize),n=r.blockSize/8*r.bits,u=e[1];if(!je.areEqual(u.dims,[r.n,o,n]))throw new Error("The second inputs must be 3D tensor with shape N X nBlocksPerCol X blobSize");let d=e[2].dims;if(je.size(d)!==r.n*o)throw new Error("scales input size error.");if(e.length===4){let h=e[3].dims,f=r.bits>4?r.n*o:r.n*Math.floor((o+1)/2);if(je.size(h)!==f)throw new Error("zeroPoints input size error.")}},ew=(e,r)=>{let t=e[0].dims,i=t.length,o=t[i-2],n=r.k,u=r.n,d=t.slice(0,i-2),h=je.size(d),f=e[1].dims[2]/4,y=e[0].dataType,m=gr(r.k),g=gr(f),T=gr(u),M=d.concat([o,u]),I=o>1&&u/T%2===0?2:1,z=je.size(M)/T/I,E=64,C=[],D=[h,o,n/m],A=je.convertShape(e[1].dims).slice();A.splice(-1,1,f/g),C.push(...Et(D)),C.push(...Et(A)),C.push(...Et(e[2].dims)),e.length===4&&C.push(...Et(je.convertShape(e[3].dims)));let $=[h,o,u/T];C.push(...Et($));let P=k=>{let O=D.length,R=Ne("a",e[0].dataType,O,m),U=Ne("b",12,A.length,g),te=Ne("scales",e[2].dataType,e[2].dims.length),se=[R,U,te],K=e.length===4?Ne("zero_points",12,e[3].dims.length):void 0;K&&se.push(K);let pe=$.length,re=gt("output",e[0].dataType,pe,T),oe=Lr(e[0].dataType),ge=(()=>{switch(m){case 1:return`array<${oe}, 8>`;case 2:return`mat4x2<${oe}>`;case 4:return`mat2x4<${oe}>`;default:throw new Error(`${m}-component is not supported.`)}})(),le=()=>{let q=` + // reuse a data + var input_offset = ${R.indicesToOffset(`${R.type.indices}(batch, row, word_offset)`)}; + var a_data: ${ge}; + for (var j: u32 = 0; j < ${8/m}; j++) { + a_data[j] = ${R.getByOffset("input_offset")}; + input_offset++; + } + `;for(let N=0;N> 4) & b_mask); + b_quantized_values = ${ge}(${Array.from({length:4},(Q,ue)=>`${oe}(b_value_lower[${ue}]), ${oe}(b_value_upper[${ue}])`).join(", ")}); + b_dequantized_values = ${m===1?`${ge}(${Array.from({length:8},(Q,ue)=>`(b_quantized_values[${ue}] - ${K?`zero_point${N}`:"zero_point"}) * scale${N}`).join(", ")});`:`(b_quantized_values - ${ge}(${Array(8).fill(`${K?`zero_point${N}`:"zero_point"}`).join(",")})) * scale${N};`}; + workgroup_shared[local_id.x * ${I} + ${Math.floor(N/T)}]${T>1?`[${N%T}]`:""} += ${Array.from({length:8/m},(Q,ue)=>`${m===1?`a_data[${ue}] * b_dequantized_values[${ue}]`:`dot(a_data[${ue}], b_dequantized_values[${ue}])`}`).join(" + ")}; + `;return q},Se=()=>{let q=` + var col_index = col * ${T}; + ${K?` + let zero_point_bytes_per_col = (nBlocksPerCol + 1) / 2; + var zero_point_byte_count: u32; + var zero_point_word_index: u32; + var zero_point_byte_offset: u32; + let zero_point_nibble_offset: u32 = block & 0x1u; + var zero_point_bits_offset: u32; + var zero_point_word: u32;`:` + // The default zero point is 8 for unsigned 4-bit quantization. + let zero_point = ${oe}(8);`} + `;for(let N=0;N> 0x1u); + zero_point_word_index = zero_point_byte_count >> 0x2u; + zero_point_byte_offset = zero_point_byte_count & 0x3u; + zero_point_bits_offset = (zero_point_byte_offset << 3) + (zero_point_nibble_offset << 2); + zero_point_word = ${K.getByOffset("zero_point_word_index")} >> zero_point_bits_offset; + let zero_point${N} = ${oe}((zero_point_word) & 0xFu);`:""} + col_index += 1;`;return q},Ce=()=>{let q=`col_index = col * ${T};`;for(let N=0;N; + var b_value_upper: vec4; + var b_quantized_values: ${ge}; + var b_dequantized_values: ${ge};`,q};return` + var workgroup_shared: array<${re.type.value}, ${I*E}>; + ${k.declareVariables(...se,re)} + ${k.mainStart([E,1,1])} + let output_indices = ${re.offsetToIndices(`(global_idx / ${E}) * ${I}`)}; + let col = output_indices[2]; + let row = output_indices[1]; + let batch = output_indices[0]; + let nBlocksPerCol = uniforms.b_shape[1]; + + for (var block = local_id.x; block < nBlocksPerCol; block += ${E}) { + //process one block + var word_offset: u32 = block * ${r.blockSize/m}; + ${Se()} + for (var word: u32 = 0; word < ${f}; word += ${g}) { + ${Ce()} + for (var i: u32 = 0; i < ${g}; i++) { + ${le()} + word_offset += ${8/m}; + } + } + } + workgroupBarrier(); + + if (local_id.x < ${I}) { + var output_value: ${re.type.value} = ${re.type.value}(0); + var workgroup_shared_offset: u32 = local_id.x; + for (var b: u32 = 0u; b < ${E}u; b++) { + output_value += workgroup_shared[workgroup_shared_offset]; + workgroup_shared_offset += ${I}; + } + ${re.setByIndices(`${re.type.indices}(batch, row, col + local_id.x)`,"output_value")}; + } + }`};return{name:"MatMulNBits",shaderCache:{hint:`${r.blockSize};${r.bits};${m};${g};${T};${I};${E}`,inputDependencies:Array(e.length).fill("rank")},getRunData:()=>({outputs:[{dims:M,dataType:y}],dispatchGroup:{x:z},programUniforms:C}),getShaderSource:P}},tw=(e,r)=>{let t=e[0].dims,i=t.length,o=t[i-2],n=r.k,u=r.n,d=t.slice(0,i-2),h=je.size(d),f=e[1].dims[2]/4,y=e[0].dataType,m=gr(r.k),g=gr(f),T=d.concat([o,u]),M=128,I=u%8===0?8:u%4===0?4:1,z=M/I,E=z*g*8,C=E/m,D=E/r.blockSize,A=je.size(T)/I,$=[],P=[h,o,n/m],k=je.convertShape(e[1].dims).slice();k.splice(-1,1,f/g),$.push(...Et(P)),$.push(...Et(k)),$.push(...Et(e[2].dims)),e.length===4&&$.push(...Et(je.convertShape(e[3].dims)));let O=[h,o,u];$.push(...Et(O));let R=U=>{let te=P.length,se=Ne("a",e[0].dataType,te,m),K=Ne("b",12,k.length,g),pe=Ne("scales",e[2].dataType,e[2].dims.length),re=[se,K,pe],oe=e.length===4?Ne("zero_points",12,e[3].dims.length):void 0;oe&&re.push(oe);let ge=O.length,le=gt("output",e[0].dataType,ge),Se=Lr(e[0].dataType),Ce=()=>{switch(m){case 1:return` + let a_data0 = vec4<${Se}>(sub_a[word_offset], sub_a[word_offset + 1], sub_a[word_offset + 2], sub_a[word_offset + 3]); + let a_data1 = vec4<${Se}>(sub_a[word_offset + 4], sub_a[word_offset + 5], sub_a[word_offset + 6], sub_a[word_offset + 7]);`;case 2:return` + let a_data0 = vec4<${Se}>(sub_a[word_offset], sub_a[word_offset + 1]); + let a_data1 = vec4<${Se}>(sub_a[word_offset + 2], sub_a[word_offset + 3]);`;case 4:return` + let a_data0 = sub_a[word_offset]; + let a_data1 = sub_a[word_offset + 1];`;default:throw new Error(`${m}-component is not supported.`)}};return` + var sub_a: array<${se.type.value}, ${C}>; + var inter_results: array, ${I}>; + ${U.declareVariables(...re,le)} + ${U.mainStart([z,I,1])} + let output_indices = ${le.offsetToIndices(`workgroup_index * ${I}`)}; + let col = output_indices[2]; + let row = output_indices[1]; + let batch = output_indices[0]; + let n_blocks_per_col = uniforms.b_shape[1]; + let num_tiles = (n_blocks_per_col - 1) / ${D} + 1; + + // Loop over shared dimension. + for (var tile: u32 = 0; tile < num_tiles; tile += 1) { + let a_col_start = tile * ${C}; + // load one tile A data into shared memory. + for (var a_offset = local_idx; a_offset < ${C}; a_offset += ${M}) + { + let a_col = a_col_start + a_offset; + if (a_col < uniforms.a_shape[2]) + { + sub_a[a_offset] = ${se.getByIndices(`${se.type.indices}(batch, row, a_col)`)}; + } else { + sub_a[a_offset] = ${se.type.value}(0); + } + } + workgroupBarrier(); + + // each thread process one block + let b_row = col + local_id.y; + let block = tile * ${D} + local_id.x; + ${oe?` + let zero_point_bytes_per_col = (n_blocks_per_col + 1) / 2; + let zero_point_byte_count = b_row * zero_point_bytes_per_col + (block >> 0x1u); + let zero_point_word_index = zero_point_byte_count >> 0x2u; + let zero_point_byte_offset = zero_point_byte_count & 0x3u; + let zero_point_nibble_offset: u32 = block & 0x1u; + let zero_point_bits_offset = (zero_point_byte_offset << 3) + (zero_point_nibble_offset << 2); + let zero_point_word = ${oe.getByOffset("zero_point_word_index")} >> zero_point_bits_offset; + let zero_point = ${Se}((zero_point_word) & 0xFu);`:` + // The default zero point is 8 for unsigned 4-bit quantization. + let zero_point = ${Se}(8);`} + let scale = ${pe.getByOffset("b_row * n_blocks_per_col + block")}; + let b_data = ${K.getByIndices(`${K.type.indices}(b_row, block, 0)`)}; + var word_offset = local_id.x * ${r.blockSize/m}; + for (var i: u32 = 0; i < ${g}; i++) { + ${Ce()} + let b_value = ${g===1?"b_data":"b_data[i]"}; + let b_value_lower = unpack4xU8(b_value & 0x0F0F0F0Fu); + let b_value_upper = unpack4xU8((b_value >> 4) & 0x0F0F0F0Fu); + let b_quantized_values = mat2x4<${Se}>(${Array.from({length:4},(q,N)=>`${Se}(b_value_lower[${N}]), ${Se}(b_value_upper[${N}])`).join(", ")}); + let b_dequantized_values = (b_quantized_values - mat2x4<${Se}>(${Array(8).fill("zero_point").join(",")})) * scale; + inter_results[local_id.y][local_id.x] += ${Array.from({length:2},(q,N)=>`${`dot(a_data${N}, b_dequantized_values[${N}])`}`).join(" + ")}; + word_offset += ${8/m}; + } + workgroupBarrier(); + } + + if (local_idx < ${I}) { + var output_value: ${le.type.value} = ${le.type.value}(0); + for (var b = 0u; b < ${z}; b++) { + output_value += inter_results[local_idx][b]; + } + if (col + local_idx < uniforms.output_shape[2]) + { + ${le.setByIndices(`${le.type.indices}(batch, row, col + local_idx)`,"output_value")} + } + } + }`};return{name:"BlockwiseMatMulNBits32",shaderCache:{hint:`${r.blockSize};${m};${g};${z};${I}`,inputDependencies:Array(e.length).fill("rank")},getRunData:()=>({outputs:[{dims:T,dataType:y}],dispatchGroup:{x:A},programUniforms:$}),getShaderSource:R}},ev=(e,r)=>{Zy(e.inputs,r),r.blockSize===32&&e.adapterInfo.isVendor("intel")&&e.adapterInfo.isArchitecture("gen-12lp")?e.compute(tw(e.inputs,r)):e.compute(ew(e.inputs,r))},tv=e=>Zt(e)}),rw,sw,nw,iw,aw,ow,lw,cw,rv,AT=Ye(()=>{$t(),Lt(),zt(),rw=e=>{if(!e||e.length<1)throw new Error("Too few inputs");if(e[0].dataType!==1&&e[0].dataType!==10)throw new Error("Input type must be float or float16.");if(e.length>=2){let r=e[0].dims.length*2===e[1].dims[0];if(e.length===4&&(r=e[3].dims[0]*2===e[1].dims[0]),!r)throw new Error("The pads should be a 1D tensor of shape [2 * input_rank] or [2 * num_axes].")}},sw=(e,r,t)=>{let i="";for(let o=r-1;o>=0;--o)i+=` + k = i32(${e.indicesGet("indices",o)}) - ${Mt("uniforms.pads",o,t)}; + if (k < 0) { + break; + } + if (k >= i32(${Mt("uniforms.x_shape",o,r)})) { + break; + } + offset += k * i32(${Mt("uniforms.x_strides",o,r)}); + `;return` + value = ${e.type.value}(uniforms.constant_value); + for (var i = 0; i < 1; i++) { + var offset = 0; + var k = 0; + ${i} + value = x[offset]; + } + `},nw=(e,r,t)=>{let i="";for(let o=r-1;o>=0;--o)i+=` + k = i32(${e.indicesGet("indices",o)}) - ${Mt("uniforms.pads",o,t)}; + if (k < 0) { + k = -k; + } + { + let _2n_1 = 2 * (i32(${Mt("uniforms.x_shape",o,r)}) - 1); + k = k % _2n_1; + if(k >= i32(${Mt("uniforms.x_shape",o,r)})) { + k = _2n_1 - k; + } + } + offset += k * i32(${Mt("uniforms.x_strides",o,r)}); + `;return` + var offset = 0; + var k = 0; + ${i} + value = x[offset]; + `},iw=(e,r,t)=>{let i="";for(let o=r-1;o>=0;--o)i+=` + k = i32(${e.indicesGet("indices",o)}) - ${Mt("uniforms.pads",o,t)}; + if (k < 0) { + k = 0; + } + if (k >= i32(${Mt("uniforms.x_shape",o,r)})) { + k = i32(${Mt("uniforms.x_shape",o,r)}) - 1; + } + offset += k * i32(${Mt("uniforms.x_strides",o,r)}); + `;return` + var offset = 0; + var k = 0; + ${i} + value = x[offset]; + `},aw=(e,r,t)=>{let i="";for(let o=r-1;o>=0;--o)i+=` + k = i32(${e.indicesGet("indices",o)}) - ${Mt("uniforms.pads",o,t)}; + if (k < 0) { + k += i32(${Mt("uniforms.x_shape",o,r)}]); + } + if (k >= i32(${Mt("uniforms.x_shape",o,r)})) { + k -= i32(${Mt("uniforms.x_shape",o,r)}); + } + offset += k * i32(${Mt("uniforms.x_strides",o,r)}); + `;return` + var offset = 0; + var k = 0; + ${i} + value = x[offset]; + `},ow=(e,r,t)=>{switch(t.mode){case 0:return sw(e,r,t.pads.length);case 1:return nw(e,r,t.pads.length);case 2:return iw(e,r,t.pads.length);case 3:return aw(e,r,t.pads.length);default:throw new Error("Invalid mode")}},lw=(e,r)=>{let t=je.padShape(e[0].dims.slice(),r.pads),i=e[0].dims,o=je.size(t),n=[{type:12,data:o},{type:6,data:r.pads}],u=e.length>=3&&e[2].data;r.mode===0&&n.push({type:u?e[2].dataType:1,data:r.value}),n.push(...Et(e[0].dims,t));let d=["rank"],h=f=>{let y=gt("output",e[0].dataType,t.length),m=Ne("x",e[0].dataType,i.length),g=m.type.value,T=ow(y,i.length,r),M=[{name:"output_size",type:"u32"},{name:"pads",type:"i32",length:r.pads.length}];return r.mode===0&&M.push({name:"constant_value",type:u?g:"f32"}),` + ${f.registerUniforms(M).declareVariables(m,y)} + ${f.mainStart()} + ${f.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + + let indices = ${y.offsetToIndices("global_idx")}; + + var value = ${g}(0); + ${T} + output[global_idx] = value; + }`};return{name:"Pad",shaderCache:{hint:`${r.mode}${u}`,inputDependencies:d},getRunData:()=>({outputs:[{dims:t,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(je.size(t)/64)},programUniforms:n}),getShaderSource:h}},cw=(e,r)=>{if(e.length>1){let t=e[1].getBigInt64Array(),i=e.length>=3&&e[2].data?e[2].dataType===10?e[2].getUint16Array()[0]:e[2].getFloat32Array()[0]:0,o=e[0].dims.length,n=new Int32Array(2*o).fill(0);if(e.length>=4){let d=e[3].getBigInt64Array();for(let h=0;hn[Number(h)]=Number(d));let u=[];return n.forEach(d=>u.push(d)),{mode:r.mode,value:i,pads:u}}else return r},rv=(e,r)=>{rw(e.inputs);let t=cw(e.inputs,r);e.compute(lw(e.inputs,t),{inputs:[0]})}}),cd,cm,um,dm,pm,uw,dw,hm,fm,sv,nv,mm,iv,av,_m,ov,lv,cv,uv,OT=Ye(()=>{Vs(),$t(),Lt(),zt(),cd=e=>{if(dr.webgpu.validateInputContent&&(!e||e.length!==1))throw new Error("Pool ops requires 1 input.")},cm=(e,r,t)=>{let i=r.format==="NHWC",o=e.dims.slice();i&&o.splice(1,0,o.pop());let n=Object.hasOwnProperty.call(r,"dilations"),u=r.kernelShape.slice(),d=r.strides.slice(),h=n?r.dilations.slice():[],f=r.pads.slice();ef.adjustPoolAttributes(t,o,u,d,h,f);let y=ef.computePoolOutputShape(t,o,d,h,u,f,r.autoPad),m=Object.assign({},r);n?Object.assign(m,{kernelShape:u,strides:d,pads:f,dilations:h,cacheKey:r.cacheKey}):Object.assign(m,{kernelShape:u,strides:d,pads:f,cacheKey:r.cacheKey});let g=y.slice();return g.push(g.splice(1,1)[0]),[m,i?g:y]},um=(e,r)=>{let t=r.format==="NHWC",i=je.size(e),o=je.size(r.kernelShape),n=[{type:12,data:i},{type:12,data:o}],u=[{name:"outputSize",type:"u32"},{name:"kernelSize",type:"u32"}];if(r.kernelShape.length<=2){let d=r.kernelShape[r.kernelShape.length-1],h=r.strides[r.strides.length-1],f=r.pads[r.pads.length/2-1],y=r.pads[r.pads.length-1],m=!!(f+y);n.push({type:12,data:d},{type:12,data:h},{type:12,data:f},{type:12,data:y}),u.push({name:"kw",type:"u32"},{name:"sw",type:"u32"},{name:"pwStart",type:"u32"},{name:"pwEnd",type:"u32"});let g=!1;if(r.kernelShape.length===2){let T=r.kernelShape[r.kernelShape.length-2],M=r.strides[r.strides.length-2],I=r.pads[r.pads.length/2-2],z=r.pads[r.pads.length-2];g=!!(I+z),n.push({type:12,data:T},{type:12,data:M},{type:12,data:I},{type:12,data:z}),u.push({name:"kh",type:"u32"},{name:"sh",type:"u32"},{name:"phStart",type:"u32"},{name:"phEnd",type:"u32"})}return[n,u,!0,m,g]}else{if(t)throw new Error("Pooling with kernelShape.length > 2 is not supported for NHWC format.");let d=je.computeStrides(r.kernelShape);n.push({type:12,data:d},{type:12,data:r.pads},{type:12,data:r.strides}),u.push({name:"kernelStrides",type:"u32",length:d.length},{name:"pads",type:"u32",length:r.pads.length},{name:"strides",type:"u32",length:r.strides.length});let h=r.pads.reduce((f,y)=>f+y);return[n,u,!!h,!1,!1]}},dm=(e,r,t,i,o,n,u,d,h,f,y,m)=>{let g=o.format==="NHWC",T=r.type.value,M=gt("output",r.type.tensor,i);if(o.kernelShape.length<=2){let I="",z="",E="",C=t-(g?2:1);if(y?I=` + for (var i: u32 = 0u; i < uniforms.kw; i++) { + xIndices[${C}] = indices[${C}] * uniforms.sw - uniforms.pwStart + i; + if (xIndices[${C}] < 0 || xIndices[${C}] + >= uniforms.x_shape[${C}]) { + pad++; + continue; + } + let x_val = x[${r.indicesToOffset("xIndices")}]; + ${n} + }`:I=` + for (var i: u32 = 0u; i < uniforms.kw; i++) { + xIndices[${C}] = indices[${C}] * uniforms.sw - uniforms.pwStart + i; + let x_val = x[${r.indicesToOffset("xIndices")}]; + ${n} + }`,o.kernelShape.length===2){let D=t-(g?3:2);m?z=` + for (var j: u32 = 0u; j < uniforms.kh; j++) { + xIndices[${D}] = indices[${D}] * uniforms.sh - uniforms.phStart + j; + if (xIndices[${D}] < 0 || xIndices[${D}] >= uniforms.x_shape[${D}]) { + pad += i32(uniforms.kw); + continue; + } + `:z=` + for (var j: u32 = 0u; j < uniforms.kh; j++) { + xIndices[${D}] = indices[${D}] * uniforms.sh - uniforms.phStart + j; + `,E=` + } + `}return` + ${e.registerUniforms(h).declareVariables(r,M)} + + ${e.mainStart()} + ${e.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")} + + let indices = ${M.offsetToIndices("global_idx")}; + var xIndices = ${M.offsetToIndices("global_idx")}; + + var value = ${T}(${d}); + var pad = 0; + ${z} + ${I} + ${E} + ${u} + + output[global_idx] = value; + }`}else{if(g)throw new Error("Pooling with kernelShape.length > 2 is not supported for NHWC format.");let I=o.kernelShape.length,z=o.pads.length,E="";return f?E=` + if (xIndices[j] >= uniforms.x_shape[j]) { + pad++; + isPad = true; + break; + } + } + if (!isPad) { + let x_val = x[${r.indicesToOffset("xIndices")}]; + ${n} + }`:E=` + } + let x_val = x[${r.indicesToOffset("xIndices")}]; + ${n} + `,` + ${e.registerUniforms(h).declareVariables(r,M)} + + ${e.mainStart()} + ${e.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")} + let indices = ${M.offsetToIndices("global_idx")}; + var xIndices = ${M.offsetToIndices("global_idx")}; + + var offsets: array; + + var value = ${T}(${d}); + var pad = 0; + var isPad = false; + + for (var i: u32 = 0u; i < uniforms.kernelSize; i++) { + var offset = i; + for (var j = 0u; j < ${I-1}u; j++) { + offsets[j] = offset / ${Mt("uniforms.kernelStrides","j",I)}; + offset -= offsets[j] * ${Mt("uniforms.kernelStrides","j",I)}; + } + offsets[${I-1}] = offset; + + isPad = false; + for (var j = ${t-I}u; j < ${t}u; j++) { + xIndices[j] = indices[j] * ${Mt("uniforms.strides",`j - ${t-I}u`,I)} + + offsets[j - ${t-I}u] - ${Mt("uniforms.pads","j - 2u",z)}; + ${E} + } + ${u} + + output[global_idx] = value; + }`}},pm=e=>`${e.format};${e.ceilMode};${e.autoPad};${e.kernelShape.length}`,uw=e=>`${pm(e)};${e.countIncludePad}`,dw=e=>`${pm(e)};${e.storageOrder};${e.dilations}`,hm=e=>({format:e.format,autoPad:["NOTSET","VALID","SAME_UPPER","SAME_LOWER"][e.auto_pad],ceilMode:e.ceil_mode,kernelShape:e.kernel_shape,strides:e.strides,pads:e.pads}),fm=(e,r,t,i)=>{let[o,n]=cm(r,i,t),u=Ne("x",r.dataType,r.dims.length),d=u.type.value,h="value += x_val;",f="";o.countIncludePad?f+=`value /= ${d}(uniforms.kernelSize);`:f+=`value /= ${d}(i32(uniforms.kernelSize) - pad);`;let[y,m,g,T,M]=um(n,o);y.push(...Et(r.dims,n));let I=["rank"];return{name:e,shaderCache:{hint:`${i.cacheKey};${g};${T};${M}`,inputDependencies:I},getRunData:()=>({outputs:[{dims:n,dataType:r.dataType}],dispatchGroup:{x:Math.ceil(je.size(n)/64)},programUniforms:y}),getShaderSource:z=>dm(z,u,r.dims.length,n.length,o,h,f,0,m,g,T,M)}},sv=e=>{let r=e.count_include_pad!==0,t=hm(e);if(t.ceilMode!==0)throw new Error("using ceil() in shape computation is not yet supported for AveragePool");let i={countIncludePad:r,...t,cacheKey:""};return{...i,cacheKey:uw(i)}},nv=(e,r)=>{cd(e.inputs),e.compute(fm("AveragePool",e.inputs[0],!1,r))},mm={autoPad:"",ceilMode:0,countIncludePad:!1,kernelShape:[],strides:[],pads:[],storageOrder:0,dilations:[]},iv=e=>{let r=e.format;return{format:r,...mm,cacheKey:r}},av=(e,r)=>{cd(e.inputs),e.compute(fm("GlobalAveragePool",e.inputs[0],!0,r))},_m=(e,r,t,i)=>{let[o,n]=cm(r,i,t),u=` + value = max(x_val, value); + `,d="",h=Ne("x",r.dataType,r.dims.length),f=["rank"],[y,m,g,T,M]=um(n,o);return y.push(...Et(r.dims,n)),{name:e,shaderCache:{hint:`${i.cacheKey};${g};${T};${M}`,inputDependencies:f},getRunData:()=>({outputs:[{dims:n,dataType:r.dataType}],dispatchGroup:{x:Math.ceil(je.size(n)/64)},programUniforms:y}),getShaderSource:I=>dm(I,h,r.dims.length,n.length,o,u,d,r.dataType===10?-65504:-1e5,m,g,T,M)}},ov=(e,r)=>{cd(e.inputs),e.compute(_m("MaxPool",e.inputs[0],!1,r))},lv=e=>{let r=e.storage_order,t=e.dilations,i=hm(e);if(r!==0)throw new Error("column major storage order is not yet supported for MaxPool");if(i.ceilMode!==0)throw new Error("using ceil() in shape computation is not yet supported for MaxPool");let o={storageOrder:r,dilations:t,...i,cacheKey:""};return{...o,cacheKey:dw(o)}},cv=e=>{let r=e.format;return{format:r,...mm,cacheKey:r}},uv=(e,r)=>{cd(e.inputs),e.compute(_m("GlobalMaxPool",e.inputs[0],!0,r))}}),pw,hw,dv,pv,FT=Ye(()=>{$t(),Lt(),br(),zt(),pw=(e,r)=>{if(e.length<2||e.length>3)throw new Error("DequantizeLinear requires 2 or 3 inputs.");if(e.length===3&&e[1].dims===e[2].dims)throw new Error("x-scale and x-zero-point must have the same shape.");if(e.length===3&&e[0].dataType!==e[2].dataType)throw new Error("x and x-zero-point must have the same data type.");if(e[0].dataType===6&&e.length>2)throw new Error("In the case of dequantizing int32 there is no zero point.");if(e[1].dims.length!==0&&e[1].dims.length!==1&&e[1].dims.length!==e[0].dims.length)throw new Error("scale input must be a scalar, a 1D tensor, or have the same rank as the input tensor.");if(e.length>2){if(e[0].dataType!==e[2].dataType)throw new Error("x and x-zero-point must have the same data type.");if(e[1].dims.length!==e[2].dims.length)throw new Error("scale and zero-point inputs must have the same rank.");if(!e[1].dims.map((t,i)=>t===e[2].dims[i]).reduce((t,i)=>t&&i,!0))throw new Error("scale and zero-point inputs must have the same shape.")}if(r.blockSize>0){if(e[1].dims.length===0||e[1].dims.length===1&&e[1].dims[0]===1)throw new Error("blockSize must be set only for block quantization.");if(!e[1].dims.map((o,n)=>n===r.axis||o===e[0].dims[n]).reduce((o,n)=>o&&n,!0))throw new Error("For block qunatization, scale input shape to match the input shape except for the axis");if(e[1].dims.length!==e[0].dims.length)throw new Error("For block qunatization the scale input rank must be the same as the x rank.");let t=e[0].dims[r.axis],i=e[1].dims[r.axis];if(r.blockSizeMath.ceil(t/(i-1)-1))throw new Error("blockSize must be with in the range [ceil(dI / Si), ceil(dI / (Si - 1) - 1)].")}},hw=(e,r)=>{let t=je.normalizeAxis(r.axis,e[0].dims.length),i=e[0].dataType,o=i===3,n=e[0].dims,u=e[1].dataType,d=je.size(n),h=i===3||i===2,f=h?[Math.ceil(je.size(e[0].dims)/4)]:e[0].dims,y=e[1].dims,m=e.length>2?e[2]:void 0,g=m?h?[Math.ceil(je.size(m.dims)/4)]:m.dims:void 0,T=y.length===0||y.length===1&&y[0]===1,M=T===!1&&y.length===1,I=gr(d),z=T&&(!h||I===4),E=z?I:1,C=z&&!h?I:1,D=Ne("input",h?12:i,f.length,C),A=Ne("scale",u,y.length),$=m?Ne("zero_point",h?12:i,g.length):void 0,P=gt("output",u,n.length,E),k=[D,A];$&&k.push($);let O=[f,y];m&&O.push(g);let R=[{type:12,data:d/E},{type:12,data:t},{type:12,data:r.blockSize},...Et(...O,n)],U=te=>{let se=[{name:"output_size",type:"u32"},{name:"axis",type:"u32"},{name:"block_size",type:"u32"}];return` + ${te.registerUniforms(se).declareVariables(...k,P)} + ${te.mainStart()} + ${te.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let output_indices = ${P.offsetToIndices("global_idx")}; + + // Set input x + ${h?` + let input = ${D.getByOffset("global_idx / 4")}; + let x_vec = ${o?"unpack4xI8(input)":"unpack4xU8(input)"}; + let x_value = ${E===1?"x_vec[global_idx % 4]":"x_vec"};`:`let x_value = ${D.getByOffset("global_idx")};`}; + + // Set scale input + ${T?`let scale_value= ${A.getByOffset("0")}`:M?` + let scale_index = ${P.indicesGet("output_indices","uniforms.axis")}; + let scale_value= ${A.getByOffset("scale_index")};`:` + var scale_indices: ${A.type.indices} = output_indices; + let index = ${A.indicesGet("scale_indices","uniforms.axis")} / uniforms.block_size; + ${A.indicesSet("scale_indices","uniforms.axis","index")}; + let scale_value= ${A.getByIndices("scale_indices")};`}; + + // Set zero-point input + ${$?T?h?` + let zero_point_input = ${$.getByOffset("0")}; + let zero_point_vec = ${o?"unpack4xI8(zero_point_input)":"unpack4xU8(zero_point_input)"}; + let zero_point_value= zero_point_vec[0]`:`let zero_point_value = ${$.getByOffset("0")}`:M?h?` + let zero_point_index = ${P.indicesGet("output_indices","uniforms.axis")}; + let zero_point_input = ${$.getByOffset("zero_point_index / 4")}; + let zero_point_vec = ${o?"unpack4xI8(zero_point_input)":"unpack4xU8(zero_point_input)"}; + let zero_point_value = zero_point_vec[zero_point_index % 4]`:` + let zero_point_index = ${P.indicesGet("output_indices","uniforms.axis")}; + let zero_point_value = ${$.getByOffset("zero_point_index")};`:h?` + let zero_point_offset = ${A.indicesToOffset("scale_indices")}; + let zero_point_input = ${$.getByOffset("zero_point_offset / 4")}; + let zero_point_vec = ${o?"unpack4xI8(zero_point_input)":"unpack4xU8(zero_point_input)"}; + let zero_point_value = zero_point_vec[zero_point_offset % 4];`:`let zero_point_value = ${$.getByIndices("scale_indices")};`:`let zero_point_value = ${h?o?"i32":"u32":D.type.value}(0);`}; + // Compute and write output + ${P.setByOffset("global_idx",`${P.type.value}(x_value - zero_point_value) * scale_value`)}; + }`};return{name:"DequantizeLinear",shaderCache:{hint:r.cacheKey,inputDependencies:$?["rank","rank","rank"]:["rank","rank"]},getShaderSource:U,getRunData:()=>({outputs:[{dims:n,dataType:u}],dispatchGroup:{x:Math.ceil(d/E/64),y:1,z:1},programUniforms:R})}},dv=(e,r)=>{pw(e.inputs,r),e.compute(hw(e.inputs,r))},pv=e=>Zt({axis:e.axis,blockSize:e.blockSize})}),fw,mw,hv,DT=Ye(()=>{Vs(),$t(),zt(),fw=(e,r,t)=>{let i=e===r,o=er&&t>0;if(i||o||n)throw new Error("Range these inputs' contents are invalid.")},mw=(e,r,t,i)=>{let o=Math.abs(Math.ceil((r-e)/t)),n=[o],u=o,d=[{type:12,data:u},{type:i,data:e},{type:i,data:t},...Et(n)],h=f=>{let y=gt("output",i,n.length),m=y.type.value,g=[{name:"outputSize",type:"u32"},{name:"start",type:m},{name:"delta",type:m}];return` + ${f.registerUniforms(g).declareVariables(y)} + ${f.mainStart()} + ${f.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")} + output[global_idx] = uniforms.start + ${m}(global_idx) * uniforms.delta; + }`};return{name:"Range",shaderCache:{hint:`${i}`},getShaderSource:h,getRunData:()=>({outputs:[{dims:n,dataType:i}],dispatchGroup:{x:Math.ceil(u/64)},programUniforms:d})}},hv=e=>{let r=0,t=0,i=0;e.inputs[0].dataType===6?(r=e.inputs[0].getInt32Array()[0],t=e.inputs[1].getInt32Array()[0],i=e.inputs[2].getInt32Array()[0]):e.inputs[0].dataType===1&&(r=e.inputs[0].getFloat32Array()[0],t=e.inputs[1].getFloat32Array()[0],i=e.inputs[2].getFloat32Array()[0]),dr.webgpu.validateInputContent&&fw(r,t,i),e.compute(mw(r,t,i,e.inputs[0].dataType),{inputs:[]})}}),_w,gw,fv,mv,jT=Ye(()=>{$t(),Lt(),br(),zt(),_w=(e,r,t,i)=>{if(e!=="none"&&i!=="i32"&&i!=="u32"&&i!=="f32")throw new Error(`Input ${i} is not supported with reduction ${e}.`);let o=`{ + var oldValue = 0; + loop { + let newValueF32 =`,n=`; + let newValue = bitcast(newValueF32); + let res = atomicCompareExchangeWeak(&${r}, oldValue, newValue); + if res.exchanged { + break; + } + oldValue = res.old_value; + } + }`;switch(e){case"none":return`${r}=${t};`;case"add":return i==="i32"||i==="u32"?`atomicAdd(&${r}, bitcast<${i}>(${t}));`:` + ${o}bitcast<${i}>(oldValue) + (${t})${n}`;case"max":return i==="i32"||i==="u32"?`atomicMax(&${r}, bitcast<${i}>(${t}));`:` + ${o}max(bitcast(oldValue), (${t}))${n}`;case"min":return i==="i32"||i==="u32"?`atomicMin(&${r}, bitcast<${i}>(${t}));`:`${o}min(bitcast<${i}>(oldValue), (${t}))${n}`;case"mul":return`${o}(bitcast<${i}>(oldValue) * (${t}))${n}`;default:throw new Error(`Reduction ${e} is not supported.`)}},gw=(e,r)=>{let t=e[0].dims,i=e[1].dims,o=t,n=1,u=Math.ceil(je.sizeToDimension(i,i.length-1)/n),d=i[i.length-1],h=je.sizeFromDimension(t,d),f=[{type:12,data:u},{type:12,data:d},{type:12,data:h},...Et(e[1].dims,e[2].dims,o)],y=m=>{let g=Ne("indices",e[1].dataType,e[1].dims.length),T=Ne("updates",e[2].dataType,e[2].dims.length,n),M=r.reduction!=="none"&&r.reduction!==""?W0("output",e[0].dataType,o.length):gt("output",e[0].dataType,o.length,n);return` + ${m.registerUniform("output_size","u32").registerUniform("last_index_dimension","u32").registerUniform("num_updates_elements","u32").declareVariables(g,T,M)} + ${m.mainStart()} + ${m.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + var data_offset = 0u; + let indices_start = uniforms.last_index_dimension * global_idx; + let indices_end = indices_start + uniforms.last_index_dimension; + for (var i = indices_start; i < indices_end; i++) { + var index = i32(indices[i].x); + ${e[0].dims.length===1?` + let element_count_dim = uniforms.output_strides; + let dim_value = uniforms.output_shape;`:` + let element_count_dim = uniforms.output_strides[i - indices_start]; + let dim_value = uniforms.output_shape[i - indices_start];`} + if (index >= 0) { + if (index >= i32(dim_value)) { + index = i32(dim_value - 1); + } + } else { + if (index < -i32(dim_value)) { + index = 0; + } else { + index += i32(dim_value); + } + } + data_offset += u32((u32(index) * element_count_dim)); + } + + for (var i = 0u; i < uniforms.num_updates_elements; i++) { + let value = updates[uniforms.num_updates_elements * global_idx + i]; + ${_w(r.reduction,"output[data_offset + i]","value",M.type.value)} + } + + }`};return{name:"ScatterND",shaderCache:{hint:`${r.cacheKey}_${r.reduction}`,inputDependencies:["rank","rank"]},getRunData:()=>({outputs:[{dims:o,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(u/64)},programUniforms:f}),getShaderSource:y}},fv=e=>Zt({reduction:e.reduction}),mv=(e,r)=>{e.compute(gw(e.inputs,r),{inputs:[e.inputs[1],e.inputs[2]],outputs:[]})}}),yw,ww,bw,gm,Mw,vw,xw,Tw,Cw,Ew,Pw,Sw,ym,kw,$w,Iw,Aw,Ow,_v,gv,LT=Ye(()=>{$t(),Lt(),br(),zt(),yw=(e,r)=>{if(e.every(t=>t>0||(()=>{throw new Error("Resize requires scales input values to be positive")})),e.length>0){if(r.mode==="linear"){if(!(e.length===2||e.length===3||e.length===4&&e[0]===1&&e[1]===1||e.length===4&&e[0]===1&&e[3]===1||e.length===5&&e[0]===1&&e[1]===1))throw new Error(`For linear mode, Resize requires scales to be 2D, 3D, 4D with either two outermost or one innermost and + one outermost scale values equal to 1, or 5D with two outermost scale values equal to 1`)}else if(r.mode==="cubic"&&!(e.length===2||e.length===4&&e[0]===1&&e[1]===1||e.length===4&&e[0]===1&&e[3]===1))throw new Error("Resize requires scales input size to be 2 or 4 for cubic mode")}},ww=(e,r,t)=>{r.every(o=>o>=0&&o{throw new Error("Resize requires axes input values to be positive and less than rank")}));let i=new Array(t).fill(1);return r.forEach((o,n)=>i[o]=e[n]),i},bw=(e,r,t,i,o,n)=>{let[u,d,h]=t>10?[1,2,3]:[-1,e.length>1?1:-1,-1],f=e[0].dims.length;if(u>0&&e.length>u&&e[u].dims.length>0)e[u].getFloat32Array().forEach(y=>n.push(y));else if(r.coordinateTransformMode==="tf_crop_and_resize")throw new Error("Resize requires RoI input to be specified when coordinateTransformMode is tfCropAndResize");if(d>0&&e.length>d&&e[d].dims.length===1&&e[d].dims[0]>0){if(e[d].getFloat32Array().forEach(y=>i.push(y)),i.length!==0&&i.length!==f&&t>=18&&i.length!==r.axes.length)throw new Error("Resize requires scales input size to be same as input rank or axes size for opset 18 and up");yw(i,r),r.axes.length>0&&ww(i,r.axes,f).forEach((y,m)=>i[m]=y)}if(h>0&&e.length>h&&e[h].dims.length===1&&e[h].dims[0]>0&&(e[h].getBigInt64Array().forEach(y=>o.push(Number(y))),o.length!==0&&o.length!==f&&t>=18&&o.length!==r.axes.length))throw new Error("Resize requires sizes input size to be same as input rank or axes size for opset 18 and up");if(r.axes.length>0){if(i.length!==0&&i.length!==r.axes.length)throw new Error('Resize requires "scales" input size to be of axes rank when axes attributes is specified');if(o.length!==0&&o.length!==r.axes.length)throw new Error('Resize requires "sizes" input size to be of rank axes rank when axes attributes is specified')}if(typeof i<"u"&&typeof o<"u"&&i.length>0&&o.length>f)throw new Error("Resize requires only of scales or sizes to be specified")},gm=(e,r,t,i)=>` + // The whole part and the fractional part are calculated separately due to inaccuracy of floating + // point division. As an example, f32(21) / f32(7) may evaluate to 2.99... instead of 3, causing an + // offset-by-one error later in floor(). + let big = (${e}) * (${r}); + let whole = ${i}(big / (${t})); + let fract = ${i}(big % (${t})) / ${i}(${t}); + return whole + fract; +`,Mw=(e,r)=>`fn getOriginalCoordinateFromResizedCoordinate(xResized: u32, xScale: f32, lengthResized: u32, + lengthOriginal: u32, roiStart: f32, roiEnd: f32) -> ${r} { `+(()=>{switch(e){case"asymmetric":return` + if (xScale < 1.0 || floor(xScale) != xScale) { + return ${r}(xResized) / ${r}(xScale); + } else { + ${gm("xResized","lengthOriginal","lengthResized",r)} + } + `;case"pytorch_half_pixel":return`if (lengthResized > 1) { + return (${r}(xResized) + 0.5) / ${r}(xScale) - 0.5; + } else { + return 0.0; + }`;case"tf_half_pixel_for_nn":return`return (${r}(xResized) + 0.5) / ${r}(xScale);`;case"align_corners":return`if (lengthResized == 1) { + return 0.0; + } else { + ${gm("xResized","lengthOriginal - 1","lengthResized - 1",r)} + }`;case"tf_crop_and_resize":return`if (lengthResized > 1) { + return ${r}(roiStart) * ${r}(lengthOriginal - 1) + + (${r}(xResized) * ${r}(roiEnd - roiStart) * ${r}(lengthOriginal - 1)) / + ${r}(lengthResized - 1); + } else { + return 0.5 * ${r}(roiStart + roiEnd) * ${r}(lengthOriginal - 1); + }`;case"half_pixel_symmetric":return`const outputWidth = ${r}xScale * ${r}(lengthResized); + const adjustment = ${r}(lengthResized) / outputWidth; + const center = ${r}(lengthOriginal) / 2; + const offset = center * (1 - adjustment); + return offset + ((${r}(xResized) + 0.5) / ${r}(xScale)) - 0.5;`;case"half_pixel":return`return ((${r}(xResized) + 0.5) / ${r}(xScale)) - 0.5;`;default:throw new Error(`Coordinate transform mode ${e} is not supported`)}})()+"}",vw=(e,r,t)=>`fn getNearestPixelFromOriginal(xOriginal: ${t}, isDownSample: bool) -> ${t} {`+(()=>{switch(e){case"round_prefer_ceil":return"if (fract(xOriginal) == 0.5) { return ceil(xOriginal); } else { return round(xOriginal); }";case"floor":return"return floor(xOriginal);";case"ceil":return"return ceil(xOriginal);";case"round_prefer_floor":return"if (fract(xOriginal) == 0.5) { return floor(xOriginal); } else { return round(xOriginal); }";case"simple":default:if(r<11)return"if (isDownSample) { return ceil(xOriginal); } else { return xOriginal; }";throw new Error(`Nearest mode ${e} is not supported`)}})()+"}",xw=(e,r,t)=>{let i=new Array(t).fill(0).concat(new Array(t).fill(1)),o=e.length===0?i:e.slice();return r.length>0?(r.forEach((n,u)=>{i[n]=o[u],i[u+t]=o[r.length+u]}),i):o},Tw=(e,r,t,i)=>{let o=[];if(t.length>0)if(i.length>0){if(e.forEach(n=>o.push(n)),Math.max(...i)>e.length)throw new Error("axes is out of bound");i.forEach((n,u)=>o[n]=t[u])}else t.forEach(n=>o.push(n));else{if(r.length===0)throw new Error("Resize requires either scales or sizes.");o=e.map((n,u)=>Math.round(n*r[u]))}return o},Cw=(e,r,t)=>{let i=(()=>{switch(t.keepAspectRatioPolicy){case"not_larger":return t.axes.length>0?Math.min(...t.axes.map(n=>r[n]),Number.MAX_VALUE):Math.min(...r,Number.MAX_VALUE);case"not_smaller":return t.axes.length>0?Math.max(...t.axes.map(n=>r[n]),Number.MIN_VALUE):Math.max(...r,Number.MIN_VALUE);default:throw new Error(`Keep aspect ratio policy ${t.keepAspectRatioPolicy} is not supported`)}})();r.fill(1,0,r.length);let o=e.slice();return t.axes.length>0?(t.axes.forEach(n=>r[n]=i),t.axes.forEach(n=>o[n]=Math.round(e[n]*r[n]))):(r.fill(i,0,r.length),o.forEach((n,u)=>o[u]=Math.round(n*r[u]))),o},Ew=(e,r,t,i,o)=>` + fn calculateOriginalIndicesFromOutputIndices(output_indices: ${e.type.indices}) -> array<${e.type.value}, ${t.length}> { + var original_indices: array<${e.type.value}, ${t.length}>; + for (var i:u32 = 0; i < ${t.length}; i++) { + var output_index = ${e.indicesGet("output_indices","i")}; + var scale = ${Mt("uniforms.scales","i",i)}; + var roi_low = ${Mt("uniforms.roi","i",o)}; + var roi_hi = ${Mt("uniforms.roi",`i + ${r.length}`,o)}; + if (scale == 1.0) { + original_indices[i] = ${e.type.value}(output_index); + } else { + var input_shape_i = ${Mt("uniforms.input_shape","i",r.length)}; + var output_shape_i = ${Mt("uniforms.output_shape","i",t.length)}; + original_indices[i] = getOriginalCoordinateFromResizedCoordinate(output_index, scale, output_shape_i, + input_shape_i, roi_low, roi_hi); + } + } + return original_indices; + }`,Pw=(e,r,t,i,o,n,u)=>` + fn calculateInputIndicesFromOutputIndices(output_indices: ${r.type.indices}) -> ${e.type.indices} { + var input_indices: ${e.type.indices}; + for (var i:u32 = 0; i < ${i.length}; i++) { + var output_index = ${r.indicesGet("output_indices","i")}; + var input_index: u32; + var scale = ${Mt("uniforms.scales","i",o)}; + if (scale == 1.0) { + input_index = output_index; + } else { + var roi_low = ${Mt("uniforms.roi","i",n)}; + var roi_hi = ${Mt("uniforms.roi",`i + ${t.length}`,n)}; + var input_shape_i = ${Mt("uniforms.input_shape","i",t.length)}; + var output_shape_i = ${Mt("uniforms.output_shape","i",i.length)}; + var original_idx = getOriginalCoordinateFromResizedCoordinate(output_index, scale, output_shape_i, + input_shape_i, roi_low, roi_hi); + if (!${u} || (original_idx >= 0 && original_idx < ${r.type.value}(input_shape_i))) { + if (original_idx < 0) { + input_index = 0; + } else if (original_idx > ${r.type.value}(input_shape_i - 1)) { + input_index = input_shape_i - 1; + } else { + input_index = u32(getNearestPixelFromOriginal(original_idx, scale < 1)); + } + } else { + input_index = u32(original_idx); + } + } + ${e.indicesSet("input_indices","i","input_index")} + } + return input_indices; + }`,Sw=(e,r)=>` + fn checkInputIndices(input_indices: ${e.type.indices}) -> bool { + for (var i:u32 = 0; i < ${r.length}; i++) { + var input_index = ${e.indicesGet("input_indices","i")}; + if (input_index < 0 || input_index >= ${Mt("uniforms.input_shape","i",r.length)}) { + return false; + } + } + return true; + }`,ym=(e,r,t,i)=>e.rank>i?` + ${e.indicesSet("input_indices",r,"channel")}; + ${e.indicesSet("input_indices",t,"batch")}; +`:"",kw=(e,r,t,i,o)=>{let[n,u,d,h]=t.length===2?[-1,0,1,-1]:[0,2,3,1],f=e.type.value;return` + fn getInputValue(batch: u32, channel: u32, row: u32, col: u32) -> ${f} { + var input_indices: ${e.type.indices}; + ${e.indicesSet("input_indices",u,`max(0, min(row, ${t[u]} - 1))`)}; + ${e.indicesSet("input_indices",d,`max(0, min(col, ${t[d]} - 1))`)}; + ${ym(e,h,n,2)} + return ${e.getByIndices("input_indices")}; + } + + fn bilinearInterpolation(output_indices: ${r.type.indices}) -> ${f} { + var originalIndices = calculateOriginalIndicesFromOutputIndices(output_indices); + var row:${f} = originalIndices[${u}]; + var col:${f} = originalIndices[${d}]; + ${i?`if (row < 0 || row > (${t[u]} - 1) || col < 0 || col > (${t[d]} - 1)) { + return ${o}; + }`:""}; + row = max(0, min(row, ${t[u]} - 1)); + col = max(0, min(col, ${t[d]} - 1)); + var row1: u32 = u32(row); + var col1: u32 = u32(col); + var row2: u32 = u32(row + 1); + var col2: u32 = u32(col + 1); + var channel: u32 = ${t.length>2?`u32(originalIndices[${h}])`:"0"}; + var batch: u32 = ${t.length>2?`u32(originalIndices[${n}])`:"0"}; + var x11: ${f} = getInputValue(batch, channel, row1, col1); + var x12: ${f} = getInputValue(batch, channel, row1, col2); + var x21: ${f} = getInputValue(batch, channel, row2, col1); + var x22: ${f} = getInputValue(batch, channel, row2, col2); + var dx1: ${f} = abs(row - ${f}(row1)); + var dx2: ${f} = abs(${f}(row2) - row); + var dy1: ${f} = abs(col - ${f}(col1)); + var dy2: ${f} = abs(${f}(col2) - col); + if (row1 == row2) { + dx1 = 0.5; + dx2 = 0.5; + } + if (col1 == col2) { + dy1 = 0.5; + dy2 = 0.5; + } + return (x11 * dx2 * dy2 + x12 * dx2 * dy1 + x21 * dx1 * dy2 + x22 * dx1 * dy1); + }`},$w=(e,r,t,i,o,n,u,d,h,f)=>{let y=t.length===2,[m,g]=y?[0,1]:[2,3],T=e.type.value,M=I=>{let z=I===m?"row":"col";return` + fn ${z}CubicInterpolation(input_indices: ${e.type.indices}, output_indices: ${r.type.indices}) -> ${T} { + var output_index = ${r.indicesGet("output_indices",I)}; + var originalIdx: ${T} = getOriginalCoordinateFromResizedCoordinate(output_index, ${o[I]}, + ${i[I]}, ${t[I]}, ${n[I]}, ${n[I]} + ${t.length}); + var fractOriginalIdx: ${T} = originalIdx - floor(originalIdx); + var coefs = getCubicInterpolationCoefs(fractOriginalIdx); + + if (${d} && (originalIdx < 0 || originalIdx > (${t[I]} - 1))) { + return ${h}; + } + var data: array<${T}, 4> = array<${T}, 4>(0.0, 0.0, 0.0, 0.0); + for (var i: i32 = -1; i < 3; i++) { + var ${z}: ${T} = originalIdx + ${T}(i); + if (${z} < 0 || ${z} >= ${t[I]}) { + ${f?`coefs[i + 1] = 0.0; + continue;`:d?`return ${h};`:`${z} = max(0, min(${z}, ${t[I]} - 1));`}; + } + var input_indices_copy: ${e.type.indices} = input_indices; + ${e.indicesSet("input_indices_copy",I,`u32(${z})`)}; + data[i + 1] = ${I===m?e.getByIndices("input_indices_copy"):"rowCubicInterpolation(input_indices_copy, output_indices)"}; + } + return cubicInterpolation1D(data, coefs); + }`};return` + ${M(m)}; + ${M(g)}; + fn getCubicInterpolationCoefs(s: ${T}) -> array<${T}, 4> { + var absS = abs(s); + var coeffs: array<${T}, 4> = array<${T}, 4>(0.0, 0.0, 0.0, 0.0); + var oneMinusAbsS: ${T} = 1.0 - absS; + var twoMinusAbsS: ${T} = 2.0 - absS; + var onePlusAbsS: ${T} = 1.0 + absS; + coeffs[0] = ((${u} * onePlusAbsS - 5 * ${u}) * onePlusAbsS + 8 * ${u}) * onePlusAbsS - 4 * ${u}; + coeffs[1] = ((${u} + 2) * absS - (${u} + 3)) * absS * absS + 1; + coeffs[2] = ((${u} + 2) * oneMinusAbsS - (${u} + 3)) * oneMinusAbsS * oneMinusAbsS + 1; + coeffs[3] = ((${u} * twoMinusAbsS - 5 * ${u}) * twoMinusAbsS + 8 * ${u}) * twoMinusAbsS - 4 * ${u}; + return coeffs; + } + + fn cubicInterpolation1D(x: array<${T}, 4>, coefs: array<${T}, 4>) -> ${T} { + var coefsSum: ${T} = coefs[0] + coefs[1] + coefs[2] + coefs[3]; + return (x[0] * coefs[0] + x[1] * coefs[1]+ x[2] * coefs[2]+ x[3] * coefs[3]) / coefsSum; + } + + fn bicubicInterpolation(output_indices: ${r.type.indices}) -> ${T} { + var input_indices: ${e.type.indices} = output_indices; + return colCubicInterpolation(input_indices, output_indices); + } + `},Iw=(e,r,t,i,o)=>{let[n,u,d,h,f]=t.length===3?[-1,0,1,2,-1]:[0,2,3,4,1],y=e.type.value;return` + fn getInputValue(batch: u32, channel: u32, depth:u32, height: u32, width: u32) -> ${y} { + var input_indices: ${e.type.indices}; + ${e.indicesSet("input_indices",u,`max(0, min(depth, ${t[u]} - 1))`)}; + ${e.indicesSet("input_indices",d,`max(0, min(height, ${t[d]} - 1))`)}; + ${e.indicesSet("input_indices",h,`max(0, min(width, ${t[h]} - 1))`)}; + ${ym(e,f,n,3)} + return ${e.getByIndices("input_indices")}; + } + + fn trilinearInterpolation(output_indices: ${r.type.indices}) -> ${y} { + var originalIndices = calculateOriginalIndicesFromOutputIndices(output_indices); + var depth:${y} = originalIndices[${u}]; + var height:${y} = originalIndices[${d}]; + var width:${y} = originalIndices[${h}]; + ${i?`if (depth < 0 || depth > (${t[u]} - 1) || height < 0 || height > (${t[d]} - 1) || width < 0 || (width > ${t[h]} - 1)) { + return ${o}; + }`:""}; + + depth = max(0, min(depth, ${t[u]} - 1)); + height = max(0, min(height, ${t[d]} - 1)); + width = max(0, min(width, ${t[h]} - 1)); + var depth1: u32 = u32(depth); + var height1: u32 = u32(height); + var width1: u32 = u32(width); + var depth2: u32 = u32(depth + 1); + var height2: u32 = u32(height + 1); + var width2: u32 = u32(width + 1); + var channel: u32 = ${t.length>3?`u32(originalIndices[${f}])`:"0"}; + var batch: u32 = ${t.length>3?`u32(originalIndices[${n}])`:"0"}; + + var x111: ${y} = getInputValue(batch, channel, depth1, height1, width1); + var x112: ${y} = getInputValue(batch, channel, depth1, height1, width2); + var x121: ${y} = getInputValue(batch, channel, depth1, height2, width1); + var x122: ${y} = getInputValue(batch, channel, depth1, height2, width2); + var x211: ${y} = getInputValue(batch, channel, depth2, height1, width1); + var x212: ${y} = getInputValue(batch, channel, depth2, height1, width2); + var x221: ${y} = getInputValue(batch, channel, depth2, height2, width1); + var x222: ${y} = getInputValue(batch, channel, depth2, height2, width2); + var dx1: ${y} = abs(depth - ${y}(depth1)); + var dx2: ${y} = abs(${y}(depth2) - depth); + var dy1: ${y} = abs(height - ${y}(height1)); + var dy2: ${y} = abs(${y}(height2) - height); + var dz1: ${y} = abs(width - ${y}(width1)); + var dz2: ${y} = abs(${y}(width2) - width); + if (depth1 == depth2) { + dx1 = 0.5; + dx2 = 0.5; + } + if (height1 == height2) { + dy1 = 0.5; + dy2 = 0.5; + } + if (width1 == width2) { + dz1 = 0.5; + dz2 = 0.5; + } + return (x111 * dx2 * dy2 * dz2 + x112 * dx2 * dy2 * dz1 + x121 * dx2 * dy1 *dz2 + x122 * dx2 * dy1 * dz1 + + x211 * dx1 * dy2 * dz2 + x212 * dx1 * dy2 * dz1 + x221 * dx1 * dy1 *dz2 + x222 * dx1 * dy1 * dz1); + }`},Aw=(e,r,t,i,o,n)=>{let u=e.dims,d=xw(n,r.axes,u.length),h=Tw(u,i,o,r.axes),f=i.slice();i.length===0&&(f=u.map((C,D)=>C===0?1:h[D]/C),r.keepAspectRatioPolicy!=="stretch"&&(h=Cw(u,f,r)));let y=gt("output",e.dataType,h.length),m=Ne("input",e.dataType,u.length),g=je.size(h),T=u.length===h.length&&u.every((C,D)=>C===h[D]),M=r.coordinateTransformMode==="tf_crop_and_resize",I=r.extrapolationValue,z=m.type.value,E=C=>` + ${T?"":` + ${Mw(r.coordinateTransformMode,z)}; + ${(()=>{switch(r.mode){case"nearest":return` + ${Sw(m,u)}; + ${vw(r.nearestMode,t,z)}; + ${Pw(m,y,u,h,f.length,d.length,M)}; + `;case"linear":return` + ${Ew(y,u,h,f.length,d.length)}; + ${(()=>{if(u.length===2||u.length===4)return`${kw(m,y,u,M,I)}`;if(u.length===3||u.length===5)return`${Iw(m,y,u,M,I)}`;throw Error("Linear mode only supports input dims 2, 3, 4 and 5 are supported in linear mode.")})()}; + `;case"cubic":return` + ${(()=>{if(u.length===2||u.length===4)return`${$w(m,y,u,h,f,d,r.cubicCoeffA,M,r.extrapolationValue,r.excludeOutside)}`;throw Error("Cubic mode only supports input dims 2 and 4 are supported in linear mode.")})()}; + `;default:throw Error("Invalid resize mode")}})()}; + `} + ${C.registerUniform("output_size","u32").registerUniform("scales","f32",f.length).registerUniform("roi","f32",d.length).declareVariables(m,y)} + ${C.mainStart()} + ${C.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + ${T?"output[global_idx] = input[global_idx];":` + let output_indices = ${y.offsetToIndices("global_idx")}; + var input_indices: ${m.type.indices}; + ${(()=>{switch(r.mode){case"nearest":return`input_indices = calculateInputIndicesFromOutputIndices(output_indices); + if (checkInputIndices(input_indices)) { + output[global_idx] = ${m.getByIndices("input_indices")}; + } else { + output[global_idx] = ${r.extrapolationValue}; + }`;case"linear":return`output[global_idx] = ${u.length===2||u.length===4?"bilinearInterpolation":"trilinearInterpolation"}(output_indices);`;case"cubic":return"output[global_idx] = bicubicInterpolation(output_indices);";default:throw Error(`Unsupported resize mode: ${r.mode}`)}})()}; +`} + }`;return{name:"Resize",shaderCache:{hint:`${r.cacheKey}|${t}|${f.length>0?r.mode==="cubic"?f:f.length:""}|${o.length>0?o:""}|${d.length>0?d:""}|${T}|${r.mode==="nearest"?u.length:u}`,inputDependencies:["rank"]},getShaderSource:E,getRunData:()=>({outputs:[{dims:h,dataType:e.dataType}],dispatchGroup:{x:Math.ceil(g/64)},programUniforms:[{type:12,data:g},{type:1,data:f},{type:1,data:d},...Et(u,h)]})}},Ow=e=>{let r=e.customDataBuffer;return new Uint32Array(r,r.byteOffset,1)[0]},_v=(e,r)=>{let t=[],i=[],o=[],n=Ow(e);if(r.antialias!==0)throw Error("Only default value (0) for Antialias attribute is supported");bw(e.inputs,r,n,t,i,o),e.compute(Aw(e.inputs[0],r,n,t,i,o),{inputs:[0]})},gv=e=>{let r=e.antialias,t=e.axes,i=e.coordinateTransformMode,o=e.cubicCoeffA,n=e.excludeOutside!==0,u=e.extrapolationValue,d=e.keepAspectRatioPolicy,h=e.mode,f=e.nearestMode===""?"simple":e.nearestMode;return Zt({antialias:r,axes:t,coordinateTransformMode:i,cubicCoeffA:o,excludeOutside:n,extrapolationValue:u,keepAspectRatioPolicy:d,mode:h,nearestMode:f})}}),Fw,Dw,yv,zT=Ye(()=>{$t(),Lt(),zt(),Fw=e=>{if(!e||e.length<3)throw new Error("layerNorm requires at least 3 inputs.");let r=e[0],t=e[1],i=e[2];if(r.dataType!==t.dataType||r.dataType!==i.dataType)throw new Error("All inputs must have the same data type");if(r.dims.length!==3&&r.dims.length!==2)throw new Error("Input must be 2D or 3D");if(t.dims.length!==3&&t.dims.length!==2)throw new Error("Skip must be 2D or 3D");let o=r.dims[r.dims.length-1],n=r.dims[r.dims.length-2];if(t.dims[t.dims.length-1]!==o)throw new Error("Skip must have the same hidden size as input");if(t.dims[t.dims.length-2]!==n)throw new Error("Skip must have the same sequence length as input");if(i.dims.length!==1)throw new Error("Gamma must be 1D");if(i.dims[i.dims.length-1]!==o)throw new Error("Gamma must have the same hidden size as input");if(e.length>3){let u=e[3];if(u.dims.length!==1)throw new Error("Beta must be 1D");if(u.dims[u.dims.length-1]!==o)throw new Error("Beta must have the same hidden size as input")}if(e.length>4){let u=e[4];if(u.dims.length!==1)throw new Error("Bias must be 1D");if(u.dims[u.dims.length-1]!==o)throw new Error("Bias must have the same hidden size as input")}},Dw=(e,r,t,i)=>{let o=r.simplified,n=e[0].dims,u=je.size(n),d=n,h=u,f=n.slice(-1)[0],y=i?n.slice(0,-1).concat(1):[],m=!o&&e.length>3,g=e.length>4,T=i&&t>1,M=i&&t>2,I=t>3,z=64,E=gr(f),C=[{type:12,data:h},{type:12,data:E},{type:12,data:f},{type:1,data:r.epsilon}],D=$=>{let P=[{name:"output_size",type:"u32"},{name:"components",type:"u32"},{name:"hidden_size",type:"u32"},{name:"epsilon",type:"f32"}],k=[Ne("x",e[0].dataType,e[0].dims,E),Ne("skip",e[1].dataType,e[1].dims,E),Ne("gamma",e[2].dataType,e[2].dims,E)];m&&k.push(Ne("beta",e[3].dataType,e[3].dims,E)),g&&k.push(Ne("bias",e[4].dataType,e[4].dims,E)),k.push(gt("output",e[0].dataType,d,E)),T&&k.push(gt("mean_output",1,y)),M&&k.push(gt("inv_std_output",1,y)),I&&k.push(gt("input_skip_bias_sum",e[0].dataType,d,E));let O=Lr(e[0].dataType),R=Lr(1,E);return` + + ${$.registerUniforms(P).declareVariables(...k)} + var sum_shared : array<${R}, ${z}>; + var sum_squared_shared : array<${R}, ${z}>; + + ${$.mainStart([z,1,1])} + let ix = local_id.x; + let iy = global_id.x / ${z}; + + let hidden_size_vectorized: u32 = uniforms.hidden_size / uniforms.components; + var stride = hidden_size_vectorized / ${z}; + let offset = ix * stride + iy * hidden_size_vectorized; + let offset1d = stride * ix; + if (ix == ${z-1}) { + stride = hidden_size_vectorized - stride * ix; + } + for (var i: u32 = 0; i < stride; i++) { + let skip_value = skip[offset + i]; + let bias_value = ${g?"bias[offset1d + i]":O+"(0.0)"}; + let input_value = x[offset + i]; + let value = input_value + skip_value + bias_value; + ${I?"input_skip_bias_sum[offset + i] = value;":""} + output[offset + i] = value; + let f32_value = ${Ba(O,E,"value")}; + sum_shared[ix] += f32_value; + sum_squared_shared[ix] += f32_value * f32_value; + } + workgroupBarrier(); + + var reduce_size : u32 = ${z}; + for (var curr_size = reduce_size >> 1; curr_size > 0; curr_size = reduce_size >> 1) { + reduce_size = curr_size + (reduce_size & 1); + if (ix < curr_size) { + sum_shared[ix] += sum_shared[ix + reduce_size]; + sum_squared_shared[ix] += sum_squared_shared[ix + reduce_size]; + } + workgroupBarrier(); + } + + let sum = sum_shared[0]; + let square_sum = sum_squared_shared[0]; + let mean = ${An("sum",E)} / f32(uniforms.hidden_size); + let inv_std_dev = inverseSqrt(${An("square_sum",E)} / f32(uniforms.hidden_size) ${o?"":"- mean * mean"} + uniforms.epsilon); + ${T?"mean_output[global_idx] = mean;":""} + ${M?"inv_std_output[global_idx] = inv_std_dev;":""} + + for (var i: u32 = 0; i < stride; i++) { + output[offset + i] = (output[offset + i] ${o?"":`- ${O}(mean)`}) * + ${O}(inv_std_dev) * gamma[offset1d + i] + ${m?"+ beta[offset1d + i]":""}; + } + }`},A=[{dims:d,dataType:e[0].dataType}];return t>1&&A.push({dims:y,dataType:1}),t>2&&A.push({dims:y,dataType:1}),t>3&&A.push({dims:n,dataType:e[0].dataType}),{name:"SkipLayerNormalization",shaderCache:{hint:`${E};${T};${M};${I}`,inputDependencies:e.map(($,P)=>"type")},getShaderSource:D,getRunData:()=>({outputs:A,dispatchGroup:{x:Math.ceil(h/f)},programUniforms:C})}},yv=(e,r)=>{Fw(e.inputs);let t=[0];e.outputCount>1&&t.push(-3),e.outputCount>2&&t.push(-3),e.outputCount>3&&t.push(3),e.compute(Dw(e.inputs,r,e.outputCount,!1),{outputs:t})}}),jw,ud,Lw,wm,zw,Bw,wv,bv,BT=Ye(()=>{$t(),Lt(),br(),zt(),jw=(e,r)=>{if(!e||e.length<1)throw new Error("too few inputs");if(r.axes.length!==0){if(r.axes.length!==r.starts.length||r.axes.length!==r.ends.length)throw new Error("axes, starts and ends must have the same length")}else if(r.starts.length!==r.ends.length)throw new Error("starts and ends must have the same length");e.slice(1).forEach((t,i)=>{if(e[i+1].dataType!==6&&e[i+1].dataType!==7)throw new Error(`Input ${i} must be an array of int32 or int64`)})},ud=(e,r)=>{let t=[];if(e.length>r)if(e[r].dataType===7)e[r].getBigInt64Array().forEach(i=>t.push(Number(i)));else if(e[r].dataType===6)e[r].getInt32Array().forEach(i=>t.push(Number(i)));else throw new Error(`Input ${r} must be an array of int32 or int64`);return t},Lw=(e,r)=>{if(e.length>1){let t=ud(e,1),i=ud(e,2),o=ud(e,3);return o.length===0&&(o=[...Array(e[0].dims.length).keys()]),Zt({starts:t,ends:i,axes:o})}else return r},wm=(e,r,t,i,o)=>{let n=e;return e<0&&(n+=t[i[r]]),o[r]<0?Math.max(0,Math.min(n,t[i[r]]-1)):Math.max(0,Math.min(n,t[i[r]]))},zw=(e,r,t)=>`fn calculateInputIndices(output_indices: ${r.type.indices}) -> ${e.type.indices} { + var input_indices: ${e.type.indices}; + var carry = 0u; + for (var i = ${t.length}; i >= 0; i--) { + let input_shape_i = ${Mt("uniforms.input_shape","i",t.length)}; + let steps_i = ${Mt("uniforms.steps","i",t.length)}; + let signs_i = ${Mt("uniforms.signs","i",t.length)}; + let starts_i = ${Mt("uniforms.starts","i",t.length)}; + var output_index = ${r.indicesGet("output_indices","i")}; + var input_index = output_index * steps_i + starts_i + carry; + carry = input_index / input_shape_i; + input_index = input_index % input_shape_i; + if (signs_i < 0) { + input_index = input_shape_i - input_index - 1u + starts_i; + } + ${e.indicesSet("input_indices","i","input_index")}; + } + return input_indices; + }`,Bw=(e,r)=>{let t=e[0].dims,i=je.size(t),o=r.axes.length>0?je.normalizeAxes(r.axes,t.length):[...Array(t.length).keys()],n=ud(e,4);n.forEach(E=>E!==0||(()=>{throw new Error("step cannot be 0")})),n.length===0&&(n=Array(o.length).fill(1));let u=r.starts.map((E,C)=>wm(E,C,t,o,n)),d=r.ends.map((E,C)=>wm(E,C,t,o,n));if(o.length!==u.length||o.length!==d.length)throw new Error("start, ends and axes should have the same number of elements");if(o.length!==t.length)for(let E=0;EMath.sign(E));n.forEach((E,C,D)=>{if(E<0){let A=(d[C]-u[C])/E,$=u[C],P=$+A*n[C];u[C]=P,d[C]=$,D[C]=-E}});let f=t.slice(0);o.forEach((E,C)=>{f[E]=Math.ceil((d[E]-u[E])/n[E])});let y={dims:f,dataType:e[0].dataType},m=gt("output",e[0].dataType,f.length),g=Ne("input",e[0].dataType,e[0].dims.length),T=je.size(f),M=[{name:"outputSize",type:"u32"},{name:"starts",type:"u32",length:u.length},{name:"signs",type:"i32",length:h.length},{name:"steps",type:"u32",length:n.length}],I=[{type:12,data:T},{type:12,data:u},{type:6,data:h},{type:12,data:n},...Et(e[0].dims,f)],z=E=>` + ${E.registerUniforms(M).declareVariables(g,m)} + ${zw(g,m,t)} + ${E.mainStart()} + ${E.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")} + let output_indices = ${m.offsetToIndices("global_idx")}; + let input_indices = calculateInputIndices(output_indices); + ${m.setByOffset("global_idx",g.getByIndices("input_indices"))} + }`;return{name:"Slice",shaderCache:{hint:`${h.length}_${u.length}_${n.length}`,inputDependencies:["rank"]},getShaderSource:z,getRunData:()=>({outputs:[y],dispatchGroup:{x:Math.ceil(i/64)},programUniforms:I})}},wv=(e,r)=>{jw(e.inputs,r);let t=Lw(e.inputs,r);e.compute(Bw(e.inputs,t),{inputs:[0]})},bv=e=>{let r=e.starts,t=e.ends,i=e.axes;return Zt({starts:r,ends:t,axes:i})}}),Rw,Nw,Mv,vv,RT=Ye(()=>{$t(),Lt(),br(),On(),zt(),Rw=e=>{if(!e||e.length!==1)throw new Error("Softmax op requires 1 input.")},Nw=(e,r)=>{let t=e.inputs[0],i=t.dims,o=je.size(i),n=i.length,u=je.normalizeAxis(r.axis,n),d=uO),f[u]=n-1,f[n-1]=u,h=e.compute(_s(t,f),{inputs:[t],outputs:[-1]})[0]):h=t;let y=h.dims,m=y[n-1],g=o/m,T=gr(m),M=m/T,I=64;g===1&&(I=256);let z=(k,O)=>O===4?`max(max(${k}.x, ${k}.y), max(${k}.z, ${k}.w))`:O===2?`max(${k}.x, ${k}.y)`:O===3?`max(max(${k}.x, ${k}.y), ${k}.z)`:k,E=Ne("x",h.dataType,h.dims,T),C=gt("result",h.dataType,h.dims,T),D=E.type.value,A=Lr(h.dataType)==="f32"?`var threadMax = ${D}(-3.402823e+38f);`:`var threadMax = ${D}(-65504.0h);`,$=k=>` + var rowMaxShared : ${D}; + var rowSumShared : ${D}; + var threadShared : array<${D}, ${I}>; + + fn getValue(row: i32, col: i32, row_stride: i32) -> ${D} { + let index = row * row_stride + col; + return x[index]; + } + + fn setValue(row: i32, col: i32, row_stride: i32, value: ${D}) { + let index = row * row_stride + col; + result[index] = value; + } + ${k.registerUniform("packedCols","i32").declareVariables(E,C)} + ${k.mainStart(I)} + let gindex = i32(global_idx); + let lindex = i32(local_idx); + const wg = ${I}; + let row = gindex / wg; + let cols = uniforms.packedCols; + let row_stride : i32 = uniforms.packedCols; + + // find the rows max + ${A} + for (var col = lindex; col < cols; col += wg) { + let value = getValue(row, col, row_stride); + threadMax = max(threadMax, value); + } + if (lindex < cols) { + threadShared[lindex] = threadMax; + } + workgroupBarrier(); + + var reduceSize = min(cols, wg); + for (var currSize = reduceSize >> 1; currSize > 0; currSize = reduceSize >> 1) { + reduceSize = currSize + (reduceSize & 1); + if (lindex < currSize) { + threadShared[lindex] = max(threadShared[lindex], threadShared[lindex + reduceSize]); + } + workgroupBarrier(); + } + if (lindex == 0) { + rowMaxShared = ${D}(${z("threadShared[0]",T)}); + } + workgroupBarrier(); + + // find the rows sum + var threadSum = ${D}(0.0); + for (var col = lindex; col < cols; col += wg) { + let subExp = exp(getValue(row, col, row_stride) - rowMaxShared); + threadSum += subExp; + } + threadShared[lindex] = threadSum; + workgroupBarrier(); + + for (var currSize = wg >> 1; currSize > 0; currSize = currSize >> 1) { + if (lindex < currSize) { + threadShared[lindex] = threadShared[lindex] + threadShared[lindex + currSize]; + } + workgroupBarrier(); + } + if (lindex == 0) { + rowSumShared = ${D}(${An("threadShared[0]",T)}); + } + workgroupBarrier(); + + // calculate final value for each element in the row + for (var col = lindex; col < cols; col += wg) { + let value = exp(getValue(row, col, row_stride) - rowMaxShared) / rowSumShared; + setValue(row, col, row_stride, value); + } + }`,P=e.compute({name:"Softmax",shaderCache:{hint:`${T};${I}`,inputDependencies:["type"]},getRunData:()=>({outputs:[{dims:y,dataType:h.dataType}],dispatchGroup:{x:g},programUniforms:[{type:6,data:M}]}),getShaderSource:$},{inputs:[h],outputs:[d?-1:0]})[0];d&&e.compute(_s(P,f),{inputs:[P]})},Mv=(e,r)=>{Rw(e.inputs),Nw(e,r)},vv=e=>Zt({axis:e.axis})}),bm,Vw,Ww,Uw,xv,NT=Ye(()=>{$t(),Lt(),zt(),bm=e=>Array.from(e.getBigInt64Array(),Number),Vw=e=>{if(!e||e.length!==2)throw new Error("Tile requires 2 inputs.");if(e[0].dataType!==1&&e[0].dataType!==10&&e[0].dataType!==6&&e[0].dataType!==12)throw new Error("Tile only support float, float16, int32, and uint32 data types");if(e[1].dataType!==7)throw new Error("Tile `repeats` input should be of int64 data type");if(e[1].dims.length!==1)throw new Error("Tile `repeats` input should be 1-D");if(bm(e[1]).length!==e[0].dims.length)throw new Error("Tile `repeats` input should have same number of elements as rank of input data tensor")},Ww=(e,r)=>{let t=[];for(let i=0;i{let t=e[0].dims,i=r??bm(e[1]),o=Ww(t,i),n=je.size(o),u=e[0].dataType,d=Ne("input",u,t.length),h=gt("output",u,o.length),f=y=>` + const inputShape = ${d.indices(...t)}; + ${y.registerUniform("output_size","u32").declareVariables(d,h)} + ${y.mainStart()} + ${y.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")} + let output_indices = ${h.offsetToIndices("global_idx")}; + var input_indices: ${d.type.indices}; + for (var i = 0; i < ${t.length}; i++) { + let input_dim_i = ${d.indicesGet("uniforms.input_shape","i")}; + let input_dim_value = ${h.indicesGet("output_indices","i")} % input_dim_i; + + ${d.indicesSet("input_indices","i","input_dim_value")} + } + ${h.setByOffset("global_idx",d.getByIndices("input_indices"))} + }`;return{name:"Tile",shaderCache:{hint:`${i}`,inputDependencies:["rank"]},getRunData:()=>({outputs:[{dims:o,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(n/64)},programUniforms:[{type:12,data:n},...Et(e[0].dims,o)]}),getShaderSource:f}},xv=e=>{Vw(e.inputs),e.compute(Uw(e.inputs),{inputs:[0]})}}),Gw,Kw,Tv,VT=Ye(()=>{$t(),Lt(),zt(),Gw=(e,r,t,i,o)=>{let n=gt("output_data",o,t.length,4),u=Ne("a_data",r[1].dataType,r[1].dims.length,4),d=Ne("b_data",r[2].dataType,r[2].dims.length,4),h=Ne("c_data",r[0].dataType,r[0].dims.length,4),f,y=(m,g,T)=>`select(${g}, ${m}, ${T})`;if(!i)f=n.setByOffset("global_idx",y(u.getByOffset("global_idx"),d.getByOffset("global_idx"),h.getByOffset("global_idx")));else{let m=(g,T,M="")=>{let I=`a_data[index_a${T}][component_a${T}]`,z=`b_data[index_b${T}][component_b${T}]`,E=`bool(c_data[index_c${T}] & (0xffu << (component_c${T} * 8)))`;return` + let output_indices${T} = ${n.offsetToIndices(`global_idx * 4u + ${T}u`)}; + let offset_a${T} = ${u.broadcastedIndicesToOffset(`output_indices${T}`,n)}; + let offset_b${T} = ${d.broadcastedIndicesToOffset(`output_indices${T}`,n)}; + let offset_c${T} = ${h.broadcastedIndicesToOffset(`output_indices${T}`,n)}; + let index_a${T} = offset_a${T} / 4u; + let index_b${T} = offset_b${T} / 4u; + let index_c${T} = offset_c${T} / 4u; + let component_a${T} = offset_a${T} % 4u; + let component_b${T} = offset_b${T} % 4u; + let component_c${T} = offset_c${T} % 4u; + ${g}[${T}] = ${M}(${y(I,z,E)}); + `};o===9?f=` + var data = vec4(0); + ${m("data",0,"u32")} + ${m("data",1,"u32")} + ${m("data",2,"u32")} + ${m("data",3,"u32")} + output_data[global_idx] = dot(vec4(0x1, 0x100, 0x10000, 0x1000000), vec4(data));`:f=` + ${m("output_data[global_idx]",0)} + ${m("output_data[global_idx]",1)} + ${m("output_data[global_idx]",2)} + ${m("output_data[global_idx]",3)} + `}return` + ${e.registerUniform("vec_size","u32").declareVariables(h,u,d,n)} + ${e.mainStart()} + ${e.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.vec_size")} + ${f} + }`},Kw=e=>{let r=e[1].dims,t=e[2].dims,i=e[0].dims,o=e[1].dataType,n=!(je.areEqual(r,t)&&je.areEqual(t,i)),u=r,d=je.size(r);if(n){let f=Na.calcShape(Na.calcShape(r,t,!1),i,!1);if(!f)throw new Error("Can't perform where op on the given tensors");u=f,d=je.size(u)}let h=Math.ceil(d/4);return{name:"Where",shaderCache:{inputDependencies:["rank","rank","rank"]},getShaderSource:f=>Gw(f,e,u,n,o),getRunData:()=>({outputs:[{dims:u,dataType:o}],dispatchGroup:{x:Math.ceil(d/64/4)},programUniforms:[{type:12,data:h},...Et(i,r,t,u)]})}},Tv=e=>{e.compute(Kw(e.inputs))}}),Cv,WT=Ye(()=>{sT(),n_(),nT(),iT(),aT(),oT(),lT(),hT(),mT(),_T(),gT(),yT(),wT(),bT(),MT(),vT(),xT(),TT(),CT(),ET(),PT(),ST(),kT(),$T(),IT(),UM(),AT(),OT(),FT(),DT(),jT(),s_(),LT(),QM(),zT(),BT(),RT(),HM(),NT(),On(),i_(),VT(),Cv=new Map([["Abs",[yb]],["Acos",[wb]],["Acosh",[bb]],["Add",[eM]],["ArgMax",[fb,Om]],["ArgMin",[hb,Om]],["Asin",[Mb]],["Asinh",[vb]],["Atan",[xb]],["Atanh",[Tb]],["Attention",[mb]],["AveragePool",[nv,sv]],["BatchNormalization",[_b]],["BiasAdd",[gb]],["BiasSplitGelu",[Zb]],["Cast",[Eb,Cb]],["Ceil",[Sb]],["Clip",[Pb]],["Concat",[uM,dM]],["Conv",[Bm,zm]],["ConvTranspose",[MM,bM]],["Cos",[kb]],["Cosh",[$b]],["CumSum",[vM,xM]],["DepthToSpace",[TM,CM]],["DequantizeLinear",[dv,pv]],["Div",[tM]],["Einsum",[EM,PM]],["Elu",[Ib,md]],["Equal",[rM]],["Erf",[Ab]],["Exp",[Ob]],["Expand",[SM]],["FastGelu",[kM]],["Floor",[Fb]],["FusedConv",[Bm,zm]],["Gather",[IM,$M]],["GatherElements",[LM,jM]],["GatherBlockQuantized",[FM,DM]],["GatherND",[AM,OM]],["Gelu",[Db]],["Gemm",[BM,zM]],["GlobalAveragePool",[av,iv]],["GlobalMaxPool",[uv,cv]],["Greater",[aM]],["GreaterOrEqual",[lM]],["GridSample",[RM,NM]],["GroupQueryAttention",[XM]],["HardSigmoid",[Wb,Vb]],["InstanceNormalization",[JM]],["LayerNormalization",[YM]],["LeakyRelu",[jb,md]],["Less",[oM]],["LessOrEqual",[cM]],["Log",[Jb]],["MatMul",[ZM]],["MatMulNBits",[ev,tv]],["MaxPool",[ov,lv]],["Mul",[sM]],["MultiHeadAttention",[WM,VM]],["Neg",[zb]],["Not",[Lb]],["Pad",[rv]],["Pow",[nM]],["QuickGelu",[Yb,md]],["Range",[hv]],["Reciprocal",[Bb]],["ReduceMin",[lb]],["ReduceMean",[sb]],["ReduceMax",[ob]],["ReduceSum",[ub]],["ReduceProd",[cb]],["ReduceL1",[nb]],["ReduceL2",[ib]],["ReduceLogSum",[pb]],["ReduceLogSumExp",[ab]],["ReduceSumSquare",[db]],["Relu",[Rb]],["Resize",[_v,gv]],["RotaryEmbedding",[qM]],["ScatterND",[mv,fv]],["Sigmoid",[Nb]],["Sin",[Ub]],["Sinh",[Gb]],["Slice",[wv,bv]],["SkipLayerNormalization",[yv]],["Split",[GM,KM]],["Sqrt",[Kb]],["Softmax",[Mv,vv]],["Sub",[iM]],["Tan",[Hb]],["Tanh",[qb]],["ThresholdedRelu",[Xb,md]],["Tile",[xv]],["Transpose",[G0,K0]],["Where",[Tv]]])}),Ev,UT=Ye(()=>{Vs(),mn(),zt(),Ev=class{constructor(e){this.backend=e,this.repo=new Map,this.attributesBound=!1}getArtifact(e){return this.repo.get(e)}setArtifact(e,r){this.repo.set(e,r)}run(e,r,t,i,o){Ns(e.programInfo.name);let n=this.backend.device,u=this.backend.getComputePassEncoder();this.backend.writeTimestamp(this.backend.pendingDispatchNumber*2);let d=[];for(let f of r)d.push({binding:d.length,resource:{buffer:f.buffer}});for(let f of t)d.push({binding:d.length,resource:{buffer:f.buffer}});o&&d.push({binding:d.length,resource:o});let h=n.createBindGroup({layout:e.computePipeline.getBindGroupLayout(0),entries:d,label:e.programInfo.name});if(this.backend.sessionStatus==="capturing"){let f={kernelId:this.backend.currentKernelId,computePipeline:e.computePipeline,bindGroup:h,dispatchGroup:i};this.backend.capturedCommandList.get(this.backend.currentSessionId).push(f)}u.setPipeline(e.computePipeline),u.setBindGroup(0,h),u.dispatchWorkgroups(...i),this.backend.writeTimestamp(this.backend.pendingDispatchNumber*2+1),this.backend.pendingDispatchNumber++,(this.backend.pendingDispatchNumber>=this.backend.maxDispatchNumber||this.backend.queryType==="at-passes")&&this.backend.endComputePass(),this.backend.pendingDispatchNumber>=this.backend.maxDispatchNumber&&this.backend.flush(),vs(e.programInfo.name)}dispose(){}build(e,r){Ns(e.name);let t=this.backend.device,i=[];[{feature:"shader-f16",extension:"f16"},{feature:"subgroups",extension:"subgroups"}].forEach(f=>{t.features.has(f.feature)&&i.push(`enable ${f.extension};`)});let o=U0(r,this.backend.device.limits),n=e.getShaderSource(o),u=`${i.join(` +`)} +${o.additionalImplementations} +${n}`,d=t.createShaderModule({code:u,label:e.name});Kt("verbose",()=>`[WebGPU] ${e.name} shader code: ${u}`);let h=t.createComputePipeline({compute:{module:d,entryPoint:"main"},layout:"auto",label:e.name});return vs(e.name),{programInfo:e,computePipeline:h,uniformVariablesInfo:o.variablesInfo}}normalizeDispatchGroupSize(e){let r=typeof e=="number"?e:e.x,t=typeof e=="number"?1:e.y||1,i=typeof e=="number"?1:e.z||1,o=this.backend.device.limits.maxComputeWorkgroupsPerDimension;if(r<=o&&t<=o&&i<=o)return[r,t,i];let n=r*t*i,u=Math.ceil(Math.sqrt(n));if(u>o){if(u=Math.ceil(Math.cbrt(n)),u>o)throw new Error("Total dispatch size exceeds WebGPU maximum.");return[u,u,u]}else return[u,u,1]}}}),Pv={};Wa(Pv,{WebGpuBackend:()=>Sv});var Hw,qw,Qw,Sv,GT=Ye(()=>{Vs(),$t(),mn(),B0(),tT(),WT(),UT(),Hw=(e,r)=>{if(r.length!==e.length)throw new Error(`inputDependencies length ${r.length} is not equal to inputTensors length ${e.length}.`);let t=[];for(let i=0;i{var o,n;let i=e.name;return(o=e.shaderCache)!=null&&o.hint&&(i+="["+e.shaderCache.hint+"]"),i+=":"+t+`:${Hw(r,((n=e.shaderCache)==null?void 0:n.inputDependencies)??new Array(r.length).fill("dims"))}`,i},Qw=class{constructor(e){e&&(this.architecture=e.architecture,this.vendor=e.vendor)}isArchitecture(e){return this.architecture===e}isVendor(e){return this.vendor===e}},Sv=class{constructor(){this.currentSessionId=null,this.currentKernelId=null,this.commandEncoder=null,this.computePassEncoder=null,this.maxDispatchNumber=16,this.pendingDispatchNumber=0,this.pendingKernels=[],this.pendingQueries=new Map,this.sessionStatus="default",this.capturedCommandList=new Map,this.capturedPendingKernels=new Map,this.sessionExternalDataMapping=new Map}get currentKernelCustomData(){if(this.currentKernelId===null)throw new Error("currentKernelCustomData(): currentKernelId is null. (should not happen)");let e=this.kernelCustomData.get(this.currentKernelId);return e||(e={},this.kernelCustomData.set(this.currentKernelId,e)),e}async initialize(e,r){this.env=e;let t=[],i={requiredLimits:{maxComputeWorkgroupStorageSize:r.limits.maxComputeWorkgroupStorageSize,maxComputeWorkgroupsPerDimension:r.limits.maxComputeWorkgroupsPerDimension,maxStorageBufferBindingSize:r.limits.maxStorageBufferBindingSize,maxBufferSize:r.limits.maxBufferSize,maxComputeInvocationsPerWorkgroup:r.limits.maxComputeInvocationsPerWorkgroup,maxComputeWorkgroupSizeX:r.limits.maxComputeWorkgroupSizeX,maxComputeWorkgroupSizeY:r.limits.maxComputeWorkgroupSizeY,maxComputeWorkgroupSizeZ:r.limits.maxComputeWorkgroupSizeZ},requiredFeatures:t},o=n=>r.features.has(n)&&t.push(n)&&!0;o("chromium-experimental-timestamp-query-inside-passes")||o("timestamp-query"),o("shader-f16"),o("subgroups"),this.device=await r.requestDevice(i),this.adapterInfo=new Qw(r.info||await r.requestAdapterInfo()),this.gpuDataManager=V0(this),this.programManager=new Ev(this),this.kernels=new Map,this.kernelPersistentData=new Map,this.kernelCustomData=new Map,Zm(e.logLevel,!!e.debug),this.device.onuncapturederror=n=>{n.error instanceof GPUValidationError&&console.error(`An uncaught WebGPU validation error was raised: ${n.error.message}`)},Object.defineProperty(this.env.webgpu,"device",{value:this.device,writable:!1,enumerable:!0,configurable:!1}),Object.defineProperty(this.env.webgpu,"adapter",{value:r,writable:!1,enumerable:!0,configurable:!1}),this.setQueryType()}dispose(){typeof this.querySet<"u"&&this.querySet.destroy(),this.gpuDataManager.dispose()}getCommandEncoder(){return this.commandEncoder||(this.commandEncoder=this.device.createCommandEncoder()),this.commandEncoder}getComputePassEncoder(){if(!this.computePassEncoder){let e=this.getCommandEncoder(),r={};this.queryType==="at-passes"&&(r.timestampWrites={querySet:this.querySet,beginningOfPassWriteIndex:this.pendingDispatchNumber*2,endOfPassWriteIndex:this.pendingDispatchNumber*2+1}),this.computePassEncoder=e.beginComputePass(r)}return this.computePassEncoder}endComputePass(){this.computePassEncoder&&(this.computePassEncoder.end(),this.computePassEncoder=null)}flush(){if(!this.commandEncoder)return;Ns(),this.endComputePass();let e;this.queryType!=="none"&&(this.commandEncoder.resolveQuerySet(this.querySet,0,this.pendingDispatchNumber*2,this.queryResolveBuffer,0),e=this.device.createBuffer({size:this.pendingDispatchNumber*2*8,usage:GPUBufferUsage.MAP_READ|GPUBufferUsage.COPY_DST}),this.pendingQueries.set(e,this.pendingKernels),this.pendingKernels=[],this.commandEncoder.copyBufferToBuffer(this.queryResolveBuffer,0,e,0,this.pendingDispatchNumber*2*8)),this.device.queue.submit([this.commandEncoder.finish()]),this.gpuDataManager.refreshPendingBuffers(),this.commandEncoder=null,this.pendingDispatchNumber=0,this.queryType!=="none"&&e.mapAsync(GPUMapMode.READ).then(()=>{var i;let r=new BigUint64Array(e.getMappedRange()),t=this.pendingQueries.get(e);for(let o=0;o"u"&&(this.queryTimeBase=T);let I=Number(T-this.queryTimeBase),z=Number(M-this.queryTimeBase);if(!Number.isSafeInteger(I)||!Number.isSafeInteger(z))throw new RangeError("incorrect timestamp range");if((i=this.env.webgpu.profiling)!=null&&i.ondata)this.env.webgpu.profiling.ondata({version:1,inputsMetadata:m.map(E=>({dims:E.dims,dataType:fn(E.dataType)})),outputsMetadata:g.map(E=>({dims:E.dims,dataType:fn(E.dataType)})),kernelId:u,kernelType:h,kernelName:f,programName:y,startTime:I,endTime:z});else{let E="";m.forEach((D,A)=>{E+=`input[${A}]: [${D.dims}] | ${fn(D.dataType)}, `});let C="";g.forEach((D,A)=>{C+=`output[${A}]: [${D.dims}] | ${fn(D.dataType)}, `}),console.log(`[profiling] kernel "${u}|${h}|${f}|${y}" ${E}${C}execution time: ${z-I} ns`)}bd("GPU",`${y}::${T}::${M}`)}e.unmap(),this.pendingQueries.delete(e)}),vs()}run(e,r,t,i,o,n){Ns(e.name);let u=[];for(let C=0;CD):t;if(y.length!==d.length)throw new Error(`Output size ${y.length} must be equal to ${d.length}.`);let m=[],g=[];for(let C=0;C=n)throw new Error(`Invalid output index: ${y[C]}`);if(y[C]===-3)continue;let D=y[C]===-1,A=y[C]===-2,$=D||A?o(d[C].dataType,d[C].dims):i(y[C],d[C].dataType,d[C].dims);if(m.push($),$.data===0)continue;let P=this.gpuDataManager.get($.data);if(!P)throw new Error(`no GPU data for output: ${$.data}`);if(D&&this.temporaryData.push(P),A){let k=this.kernelPersistentData.get(this.currentKernelId);k||(k=[],this.kernelPersistentData.set(this.currentKernelId,k)),k.push(P)}g.push(P)}if(u.length!==r.length||g.length!==m.length){if(g.length===0)return vs(e.name),m;throw new Error(`Program ${e.name} has zero-sized tensor(s) in inputs or outputs. This is not supported now.`)}let T;if(f){let C=0,D=[];f.forEach(k=>{let O=typeof k.data=="number"?[k.data]:k.data;if(O.length===0)return;let R=k.type===10?2:4,U,te;k.type===10?(te=O.length>4?16:O.length>2?8:O.length*R,U=O.length>4?16:R*O.length):(te=O.length<=2?O.length*R:16,U=16),C=Math.ceil(C/te)*te,D.push(C);let se=k.type===10?8:4;C+=O.length>4?Math.ceil(O.length/se)*U:O.length*R});let A=16;C=Math.ceil(C/A)*A;let $=new ArrayBuffer(C);f.forEach((k,O)=>{let R=D[O],U=typeof k.data=="number"?[k.data]:k.data;if(k.type===6)new Int32Array($,R,U.length).set(U);else if(k.type===12)new Uint32Array($,R,U.length).set(U);else if(k.type===10)new Uint16Array($,R,U.length).set(U);else if(k.type===1)new Float32Array($,R,U.length).set(U);else throw new Error(`Unsupported uniform type: ${fn(k.type)}`)});let P=this.gpuDataManager.create(C,GPUBufferUsage.COPY_DST|GPUBufferUsage.UNIFORM);this.device.queue.writeBuffer(P.buffer,0,$,0,C),this.gpuDataManager.release(P.id),T={offset:0,size:C,buffer:P.buffer}}let M=this.programManager.normalizeDispatchGroupSize(h),I=M[1]===1&&M[2]===1,z=qw(e,r,I),E=this.programManager.getArtifact(z);if(E||(E=this.programManager.build(e,M),this.programManager.setArtifact(z,E),Kt("info",()=>`[artifact] key: ${z}, programName: ${e.name}`)),f&&E.uniformVariablesInfo){if(f.length!==E.uniformVariablesInfo.length)throw new Error(`Uniform variables count mismatch: expect ${E.uniformVariablesInfo.length}, got ${f.length} in program "${E.programInfo.name}".`);for(let C=0;C`[ProgramManager] run "${e.name}" (key=${z}) with ${M[0]}x${M[1]}x${M[2]}`),this.queryType!=="none"||this.sessionStatus==="capturing"){let C={kernelId:this.currentKernelId,programName:E.programInfo.name,inputTensorViews:r,outputTensorViews:m};this.pendingKernels.push(C),this.sessionStatus==="capturing"&&this.capturedPendingKernels.get(this.currentSessionId).push(C)}return this.programManager.run(E,u,g,M,T),vs(e.name),m}upload(e,r){this.gpuDataManager.upload(e,r)}memcpy(e,r){this.gpuDataManager.memcpy(e,r)}async download(e,r){await this.gpuDataManager.download(e,r)}alloc(e){return this.gpuDataManager.create(e).id}free(e){return this.gpuDataManager.release(e)}createKernel(e,r,t,i){let o=Cv.get(e);if(!o)throw new Error(`kernel not implemented: ${e}`);let n={kernelType:e,kernelName:i,kernelEntry:o[0],attributes:[o[1],t]};this.kernels.set(r,n)}releaseKernel(e){let r=this.kernelPersistentData.get(e);if(r){for(let t of r)this.gpuDataManager.release(t.id);this.kernelPersistentData.delete(e)}this.kernelCustomData.delete(e),this.kernels.delete(e)}computeKernel(e,r,t){let i=this.kernels.get(e);if(!i)throw new Error(`kernel not created: ${e}`);let o=i.kernelType,n=i.kernelName,u=i.kernelEntry,d=i.attributes;if(this.currentKernelId!==null)throw new Error(`kernel "[${o}] ${n}" is not allowed to be called recursively`);this.currentKernelId=e,d[0]&&(d[1]=d[0](d[1]),d[0]=void 0),Kt("info",()=>`[WebGPU] Start to run kernel "[${o}] ${n}"...`);let h=this.env.debug;this.temporaryData=[];try{return h&&this.device.pushErrorScope("validation"),u(r,d[1]),0}catch(f){return t.push(Promise.resolve(`[WebGPU] Kernel "[${o}] ${n}" failed. ${f}`)),1}finally{h&&t.push(this.device.popErrorScope().then(f=>f?`GPU validation error for kernel "[${o}] ${n}": ${f.message}`:null));for(let f of this.temporaryData)this.gpuDataManager.release(f.id);this.temporaryData=[],this.currentKernelId=null}}registerBuffer(e,r,t,i){let o=this.sessionExternalDataMapping.get(e);o||(o=new Map,this.sessionExternalDataMapping.set(e,o));let n=o.get(r),u=this.gpuDataManager.registerExternalBuffer(t,i,n);return o.set(r,[u,t]),u}unregisterBuffers(e){let r=this.sessionExternalDataMapping.get(e);r&&(r.forEach(t=>this.gpuDataManager.unregisterExternalBuffer(t[0])),this.sessionExternalDataMapping.delete(e))}getBuffer(e){let r=this.gpuDataManager.get(e);if(!r)throw new Error(`no GPU data for buffer: ${e}`);return r.buffer}createDownloader(e,r,t){return async()=>{let i=await $m(this,e,r);return e_(i.buffer,t)}}writeTimestamp(e){this.queryType==="inside-passes"&&this.computePassEncoder.writeTimestamp(this.querySet,e)}setQueryType(){var e;this.queryType="none",(((e=this.env.webgpu.profiling)==null?void 0:e.mode)==="default"||(typeof this.env.trace>"u"?this.env.wasm.trace:this.env.trace))&&(this.device.features.has("chromium-experimental-timestamp-query-inside-passes")?this.queryType="inside-passes":this.device.features.has("timestamp-query")&&(this.queryType="at-passes"),this.queryType!=="none"&&typeof this.querySet>"u"&&(this.querySet=this.device.createQuerySet({type:"timestamp",count:this.maxDispatchNumber*2}),this.queryResolveBuffer=this.device.createBuffer({size:this.maxDispatchNumber*2*8,usage:GPUBufferUsage.COPY_SRC|GPUBufferUsage.QUERY_RESOLVE})))}captureBegin(){Kt("info","captureBegin"),this.capturedCommandList.get(this.currentSessionId)||this.capturedCommandList.set(this.currentSessionId,[]),this.capturedPendingKernels.get(this.currentSessionId)||this.capturedPendingKernels.set(this.currentSessionId,[]),this.flush(),this.sessionStatus="capturing"}captureEnd(){Kt("info","captureEnd"),this.flush(),this.sessionStatus="default"}replay(){Kt("info","replay"),this.sessionStatus="replaying";let e=this.capturedCommandList.get(this.currentSessionId),r=this.capturedPendingKernels.get(this.currentSessionId),t=e.length;this.pendingKernels=[];for(let i=0;i=this.maxDispatchNumber||this.queryType==="at-passes")&&this.endComputePass(),this.pendingDispatchNumber>=this.maxDispatchNumber&&this.flush()}this.flush(),this.sessionStatus="default"}onCreateSession(){this.gpuDataManager.onCreateSession()}onReleaseSession(e){this.unregisterBuffers(e),this.capturedCommandList.has(e)&&this.capturedCommandList.delete(e),this.capturedPendingKernels.has(e)&&this.capturedPendingKernels.delete(e),this.gpuDataManager.onReleaseSession(e)}onRunStart(e){this.currentSessionId=e,this.setQueryType()}}}),kv={};Wa(kv,{init:()=>$v});var Uh,Xw,$v,KT=Ye(()=>{$t(),mn(),Lt(),eT(),Uh=class Iv{constructor(r,t,i,o){this.module=r,this.dataType=t,this.data=i,this.dims=o}getFloat32Array(){if(this.dataType!==1)throw new Error("Invalid data type");let r=je.size(this.dims);return r===0?new Float32Array:new Float32Array(this.module.HEAP8.buffer,this.data,r)}getBigInt64Array(){if(this.dataType!==7)throw new Error("Invalid data type");let r=je.size(this.dims);return r===0?new BigInt64Array:new BigInt64Array(this.module.HEAP8.buffer,this.data,r)}getInt32Array(){if(this.dataType!==6)throw new Error("Invalid data type");let r=je.size(this.dims);return r===0?new Int32Array:new Int32Array(this.module.HEAP8.buffer,this.data,r)}getUint16Array(){if(this.dataType!==10&&this.dataType!==4)throw new Error("Invalid data type");let r=je.size(this.dims);return r===0?new Uint16Array:new Uint16Array(this.module.HEAP8.buffer,this.data,r)}reshape(r){if(je.size(r)!==je.size(this.dims))throw new Error("Invalid new shape");return new Iv(this.module,this.dataType,this.data,r)}},Xw=class{constructor(e,r,t){this.module=e,this.backend=r,this.customDataOffset=0,this.customDataSize=0,this.adapterInfo=r.adapterInfo;let i=e.PTR_SIZE,o=t/e.PTR_SIZE,n=i===4?"i32":"i64";this.opKernelContext=Number(e.getValue(i*o++,n));let u=Number(e.getValue(i*o++,n));this.outputCount=Number(e.getValue(i*o++,n)),this.customDataOffset=Number(e.getValue(i*o++,"*")),this.customDataSize=Number(e.getValue(i*o++,n));let d=[];for(let h=0;htypeof d=="number"?this.inputs[d]:d))??this.inputs,i=(r==null?void 0:r.outputs)??[],o=(d,h,f)=>new Uh(this.module,h,this.output(d,f),f),n=(d,h)=>{let f=hi(d,h);if(!f)throw new Error(`Unsupported data type: ${d}`);let y=f>0?this.backend.gpuDataManager.create(f).id:0;return new Uh(this.module,d,y,h)};return this.backend.run(e,t,i,o,n,this.outputCount)}output(e,r){let t=this.module.stackSave();try{let i=this.module.PTR_SIZE,o=i===4?"i32":"i64",n=this.module.stackAlloc((1+r.length)*i);this.module.setValue(n,r.length,o);for(let u=0;u{let o=r.jsepInit;if(!o)throw new Error("Failed to initialize JSEP. The WebAssembly module is not built with JSEP support.");if(e==="webgpu"){let n=(GT(),wd(Pv)).WebGpuBackend,u=new n;await u.initialize(t,i),o("webgpu",[u,d=>u.alloc(Number(d)),d=>u.free(d),(d,h,f,y=!1)=>{if(y)Kt("verbose",()=>`[WebGPU] jsepCopyGpuToGpu: src=${Number(d)}, dst=${Number(h)}, size=${Number(f)}`),u.memcpy(Number(d),Number(h));else{Kt("verbose",()=>`[WebGPU] jsepCopyCpuToGpu: dataOffset=${Number(d)}, gpuDataId=${Number(h)}, size=${Number(f)}`);let m=r.HEAPU8.subarray(Number(d>>>0),Number(d>>>0)+Number(f));u.upload(Number(h),m)}},async(d,h,f)=>{Kt("verbose",()=>`[WebGPU] jsepCopyGpuToCpu: gpuDataId=${d}, dataOffset=${h}, size=${f}`),await u.download(Number(d),()=>r.HEAPU8.subarray(Number(h)>>>0,Number(h+f)>>>0))},(d,h,f)=>u.createKernel(d,Number(h),f,r.UTF8ToString(r._JsepGetNodeName(Number(h)))),d=>u.releaseKernel(d),(d,h,f,y)=>{Kt("verbose",()=>`[WebGPU] jsepRun: sessionHandle=${f}, kernel=${d}, contextDataOffset=${h}`);let m=new Xw(r,u,Number(h));return u.computeKernel(Number(d),m,y)},()=>u.captureBegin(),()=>u.captureEnd(),()=>u.replay()])}else{let n=new N0(t);o("webnn",[n,()=>n.reserveTensorId(),u=>n.releaseTensorId(u),async(u,d,h,f,y)=>n.ensureTensor(u,d,h,f,y),(u,d)=>{n.uploadTensor(u,d)},async(u,d)=>n.downloadTensor(u,d)])}}}),Jw,d_,p_,kn,Yw,Mm,nf,h_,f_,vm,m_,__,g_,Av=Ye(()=>{J1(),Y1(),$t(),yi(),Qm(),D0(),Jw=(e,r)=>{lr()._OrtInit(e,r)!==0&&ir("Can't initialize onnxruntime.")},d_=async e=>{Jw(e.wasm.numThreads,Zh(e.logLevel))},p_=async(e,r)=>{var t,i;(i=(t=lr()).asyncInit)==null||i.call(t);{let o=(KT(),wd(kv)).init;if(r==="webgpu"){if(typeof navigator>"u"||!navigator.gpu)throw new Error("WebGPU is not supported in current environment");let n=e.webgpu.adapter;if(n){if(typeof n.limits!="object"||typeof n.features!="object"||typeof n.requestDevice!="function")throw new Error("Invalid GPU adapter set in `env.webgpu.adapter`. It must be a GPUAdapter object.")}else{let u=e.webgpu.powerPreference;if(u!==void 0&&u!=="low-power"&&u!=="high-performance")throw new Error(`Invalid powerPreference setting: "${u}"`);let d=e.webgpu.forceFallbackAdapter;if(d!==void 0&&typeof d!="boolean")throw new Error(`Invalid forceFallbackAdapter setting: "${d}"`);if(n=await navigator.gpu.requestAdapter({powerPreference:u,forceFallbackAdapter:d}),!n)throw new Error('Failed to get GPU adapter. You may need to enable flag "--enable-unsafe-webgpu" if you are using Chrome.')}await o("webgpu",lr(),e,n)}if(r==="webnn"){if(typeof navigator>"u"||!navigator.ml)throw new Error("WebNN is not supported in current environment");await o("webnn",lr(),e)}}},kn=new Map,Yw=e=>{let r=lr(),t=r.stackSave();try{let i=r.PTR_SIZE,o=r.stackAlloc(2*i);r._OrtGetInputOutputCount(e,o,o+i)!==0&&ir("Can't get session input/output count.");let n=i===4?"i32":"i64";return[Number(r.getValue(o,n)),Number(r.getValue(o+i,n))]}finally{r.stackRestore(t)}},Mm=(e,r)=>{let t=lr(),i=t.stackSave(),o=0;try{let n=t.PTR_SIZE,u=t.stackAlloc(2*n);t._OrtGetInputOutputMetadata(e,r,u,u+n)!==0&&ir("Can't get session input/output metadata.");let d=Number(t.getValue(u,"*"));o=Number(t.getValue(u+n,"*"));let h=t.HEAP32[o/4];if(h===0)return[d,0];let f=t.HEAPU32[o/4+1],y=[];for(let m=0;m{let r=lr(),t=r._malloc(e.byteLength);if(t===0)throw new Error(`Can't create a session. failed to allocate a buffer of size ${e.byteLength}.`);return r.HEAPU8.set(e,t),[t,e.byteLength]},h_=async(e,r)=>{var m,g,T,M;let t,i,o=lr();Array.isArray(e)?[t,i]=e:e.buffer===o.HEAPU8.buffer?[t,i]=[e.byteOffset,e.byteLength]:[t,i]=nf(e);let n=0,u=0,d=0,h=[],f=[],y=[];try{if([u,h]=await F0(r),(r==null?void 0:r.externalData)&&o.mountExternalData){let O=[];for(let R of r.externalData){let U=typeof R=="string"?R:R.path;O.push(Ym(typeof R=="string"?R:R.data).then(te=>{o.mountExternalData(U,te)}))}await Promise.all(O)}for(let O of(r==null?void 0:r.executionProviders)??[])if((typeof O=="string"?O:O.name)==="webnn"){if(o.shouldTransferToMLTensor=!1,typeof O!="string"){let R=O,U=R==null?void 0:R.context,te=R==null?void 0:R.gpuDevice,se=R==null?void 0:R.deviceType,K=R==null?void 0:R.powerPreference;U?o.currentContext=U:te?o.currentContext=await o.webnnCreateMLContext(te):o.currentContext=await o.webnnCreateMLContext({deviceType:se,powerPreference:K})}else o.currentContext=await o.webnnCreateMLContext();break}n=await o._OrtCreateSession(t,i,u),(m=o.webgpuOnCreateSession)==null||m.call(o,n),n===0&&ir("Can't create a session."),(g=o.jsepOnCreateSession)==null||g.call(o),o.currentContext&&(o.webnnRegisterMLContext(n,o.currentContext),o.currentContext=void 0,o.shouldTransferToMLTensor=!0);let[I,z]=Yw(n),E=!!(r!=null&&r.enableGraphCapture),C=[],D=[],A=[],$=[],P=[];for(let O=0;OO==="gpu-buffer"||O==="ml-tensor"||O==="ml-tensor-cpu-output")&&(d=o._OrtCreateBinding(n),d===0&&ir("Can't create IO binding."),k={handle:d,outputPreferredLocations:P,outputPreferredLocationsEncoded:P.map(O=>O==="ml-tensor-cpu-output"?"ml-tensor":O).map(O=>Sm(O))}),kn.set(n,[n,f,y,k,E,!1]),[n,C,D,A,$]}catch(I){throw f.forEach(z=>o._OrtFree(z)),y.forEach(z=>o._OrtFree(z)),d!==0&&o._OrtReleaseBinding(d)!==0&&ir("Can't release IO binding."),n!==0&&o._OrtReleaseSession(n)!==0&&ir("Can't release session."),I}finally{o._free(t),u!==0&&o._OrtReleaseSessionOptions(u)!==0&&ir("Can't release session options."),h.forEach(I=>o._free(I)),(M=o.unmountExternalData)==null||M.call(o)}},f_=e=>{var h,f,y;let r=lr(),t=kn.get(e);if(!t)throw new Error(`cannot release session. invalid session id: ${e}`);let[i,o,n,u,d]=t;u&&(d&&r._OrtClearBoundOutputs(u.handle)!==0&&ir("Can't clear bound outputs."),r._OrtReleaseBinding(u.handle)!==0&&ir("Can't release IO binding.")),(h=r.jsepOnReleaseSession)==null||h.call(r,e),(f=r.webnnOnReleaseSession)==null||f.call(r,e),(y=r.webgpuOnReleaseSession)==null||y.call(r,e),o.forEach(m=>r._OrtFree(m)),n.forEach(m=>r._OrtFree(m)),r._OrtReleaseSession(i)!==0&&ir("Can't release session."),kn.delete(e)},vm=async(e,r,t,i,o,n,u=!1)=>{if(!e){r.push(0);return}let d=lr(),h=d.PTR_SIZE,f=e[0],y=e[1],m=e[3],g=m,T,M;if(f==="string"&&(m==="gpu-buffer"||m==="ml-tensor"))throw new Error("String tensor is not supported on GPU.");if(u&&m!=="gpu-buffer")throw new Error(`External buffer must be provided for input/output index ${n} when enableGraphCapture is true.`);if(m==="gpu-buffer"){let E=e[2].gpuBuffer;M=hi(pi(f),y);{let C=d.jsepRegisterBuffer;if(!C)throw new Error('Tensor location "gpu-buffer" is not supported without using WebGPU.');T=C(i,n,E,M)}}else if(m==="ml-tensor"){let E=e[2].mlTensor;M=hi(pi(f),y);let C=d.webnnRegisterMLTensor;if(!C)throw new Error('Tensor location "ml-tensor" is not supported without using WebNN.');T=C(i,E,pi(f),y)}else{let E=e[2];if(Array.isArray(E)){M=h*E.length,T=d._malloc(M),t.push(T);for(let C=0;Cd.setValue(z+D*h,C,h===4?"i32":"i64"));let E=d._OrtCreateTensor(pi(f),T,M,z,y.length,Sm(g));E===0&&ir(`Can't create tensor for input/output. session=${i}, index=${n}.`),r.push(E)}finally{d.stackRestore(I)}},m_=async(e,r,t,i,o,n)=>{var te,se,K,pe;let u=lr(),d=u.PTR_SIZE,h=kn.get(e);if(!h)throw new Error(`cannot run inference. invalid session id: ${e}`);let f=h[0],y=h[1],m=h[2],g=h[3],T=h[4],M=h[5],I=r.length,z=i.length,E=0,C=[],D=[],A=[],$=[],P=u.stackSave(),k=u.stackAlloc(I*d),O=u.stackAlloc(I*d),R=u.stackAlloc(z*d),U=u.stackAlloc(z*d);try{[E,C]=O0(n);for(let le=0;leme*X,1);Q=fn(we);let Z=g==null?void 0:g.outputPreferredLocations[i[le]];if(Q==="string"){if(Z==="gpu-buffer"||Z==="ml-tensor")throw new Error("String tensor is not supported on GPU.");let me=[];for(let X=0;X0){let me=u.jsepGetBuffer;if(!me)throw new Error('preferredLocation "gpu-buffer" is not supported without using WebGPU.');let X=me(ue),Oe=hi(we,Ee);if(Oe===void 0||!Xm(Q))throw new Error(`Unsupported data type: ${Q}`);N=!0,oe.push([Q,et,{gpuBuffer:X,download:u.jsepCreateDownloader(X,Oe,Q),dispose:()=>{u._OrtReleaseTensor(Se)!==0&&ir("Can't release tensor.")}},"gpu-buffer"])}else if(Z==="ml-tensor"&&Ee>0){let me=u.webnnEnsureTensor,X=u.webnnIsGraphInputOutputTypeSupported;if(!me||!X)throw new Error('preferredLocation "ml-tensor" is not supported without using WebNN.');if(hi(we,Ee)===void 0||!Jm(Q))throw new Error(`Unsupported data type: ${Q}`);if(!X(e,Q,!1))throw new Error(`preferredLocation "ml-tensor" for ${Q} output is not supported by current WebNN Context.`);let Oe=await me(e,ue,we,et,!1);N=!0,oe.push([Q,et,{mlTensor:Oe,download:u.webnnCreateMLTensorDownloader(ue,Q),dispose:()=>{u.webnnReleaseTensorId(ue),u._OrtReleaseTensor(Se)}},"ml-tensor"])}else if(Z==="ml-tensor-cpu-output"&&Ee>0){let me=u.webnnCreateMLTensorDownloader(ue,Q)(),X=oe.length;N=!0,ge.push((async()=>{let Oe=[X,await me];return u.webnnReleaseTensorId(ue),u._OrtReleaseTensor(Se),Oe})()),oe.push([Q,et,[],"cpu"])}else{let me=of(Q),X=new me(Ee);new Uint8Array(X.buffer,X.byteOffset,X.byteLength).set(u.HEAPU8.subarray(ue,ue+X.byteLength)),oe.push([Q,et,X,"cpu"])}}finally{u.stackRestore(Ce),Q==="string"&&ue&&u._free(ue),N||u._OrtReleaseTensor(Se)}}g&&!T&&(u._OrtClearBoundOutputs(g.handle)!==0&&ir("Can't clear bound outputs."),kn.set(e,[f,y,m,g,T,!1]));for(let[le,Se]of await Promise.all(ge))oe[le][2]=Se;return oe}finally{(pe=u.webnnOnRunEnd)==null||pe.call(u,f),u.stackRestore(P),D.forEach(re=>u._OrtReleaseTensor(re)),A.forEach(re=>u._OrtReleaseTensor(re)),$.forEach(re=>u._free(re)),E!==0&&u._OrtReleaseRunOptions(E),C.forEach(re=>u._free(re))}},__=e=>{let r=lr(),t=kn.get(e);if(!t)throw new Error("invalid session id");let i=t[0],o=r._OrtEndProfiling(i);o===0&&ir("Can't get an profile file name."),r._OrtFree(o)},g_=e=>{let r=[];for(let t of e){let i=t[2];!Array.isArray(i)&&"buffer"in i&&r.push(i.buffer)}return r}}),$n,is,ja,dd,pd,Gh,xm,Kh,li,ci,Zw,Ov,Fv,Dv,jv,Lv,zv,Bv,Rv=Ye(()=>{Vs(),Av(),yi(),Hm(),$n=()=>!!dr.wasm.proxy&&typeof document<"u",ja=!1,dd=!1,pd=!1,Kh=new Map,li=(e,r)=>{let t=Kh.get(e);t?t.push(r):Kh.set(e,[r])},ci=()=>{if(ja||!dd||pd||!is)throw new Error("worker not ready")},Zw=e=>{switch(e.data.type){case"init-wasm":ja=!1,e.data.err?(pd=!0,xm[1](e.data.err)):(dd=!0,xm[0]()),Gh&&(URL.revokeObjectURL(Gh),Gh=void 0);break;case"init-ep":case"copy-from":case"create":case"release":case"run":case"end-profiling":{let r=Kh.get(e.data.type);e.data.err?r.shift()[1](e.data.err):r.shift()[0](e.data.out);break}}},Ov=async()=>{if(!dd){if(ja)throw new Error("multiple calls to 'initWasm()' detected.");if(pd)throw new Error("previous call to 'initWasm()' failed.");if(ja=!0,$n())return new Promise((e,r)=>{is==null||is.terminate(),I0().then(([t,i])=>{try{is=i,is.onerror=n=>r(n),is.onmessage=Zw,xm=[e,r];let o={type:"init-wasm",in:dr};!o.in.wasm.wasmPaths&&(t||Pm)&&(o.in.wasm.wasmPaths={wasm:new URL("/assets/ort-wasm-simd-threaded.jsep-CSH94wPT.wasm",import.meta.url).href}),is.postMessage(o),Gh=t}catch(o){r(o)}},r)});try{await qm(dr.wasm),await d_(dr),dd=!0}catch(e){throw pd=!0,e}finally{ja=!1}}},Fv=async e=>{if($n())return ci(),new Promise((r,t)=>{li("init-ep",[r,t]);let i={type:"init-ep",in:{epName:e,env:dr}};is.postMessage(i)});await p_(dr,e)},Dv=async e=>$n()?(ci(),new Promise((r,t)=>{li("copy-from",[r,t]);let i={type:"copy-from",in:{buffer:e}};is.postMessage(i,[e.buffer])})):nf(e),jv=async(e,r)=>{if($n()){if(r!=null&&r.preferredOutputLocation)throw new Error('session option "preferredOutputLocation" is not supported for proxy.');return ci(),new Promise((t,i)=>{li("create",[t,i]);let o={type:"create",in:{model:e,options:{...r}}},n=[];e instanceof Uint8Array&&n.push(e.buffer),is.postMessage(o,n)})}else return h_(e,r)},Lv=async e=>{if($n())return ci(),new Promise((r,t)=>{li("release",[r,t]);let i={type:"release",in:e};is.postMessage(i)});f_(e)},zv=async(e,r,t,i,o,n)=>{if($n()){if(t.some(u=>u[3]!=="cpu"))throw new Error("input tensor on GPU is not supported for proxy.");if(o.some(u=>u))throw new Error("pre-allocated output tensor is not supported for proxy.");return ci(),new Promise((u,d)=>{li("run",[u,d]);let h=t,f={type:"run",in:{sessionId:e,inputIndices:r,inputs:h,outputIndices:i,options:n}};is.postMessage(f,g_(h))})}else return m_(e,r,t,i,o,n)},Bv=async e=>{if($n())return ci(),new Promise((r,t)=>{li("end-profiling",[r,t]);let i={type:"end-profiling",in:e};is.postMessage(i)});__(e)}}),Tm,e0,Nv,HT=Ye(()=>{Vs(),Rv(),$t(),Km(),D0(),Tm=(e,r)=>{switch(e.location){case"cpu":return[e.type,e.dims,e.data,"cpu"];case"gpu-buffer":return[e.type,e.dims,{gpuBuffer:e.gpuBuffer},"gpu-buffer"];case"ml-tensor":return[e.type,e.dims,{mlTensor:e.mlTensor},"ml-tensor"];default:throw new Error(`invalid data location: ${e.location} for ${r()}`)}},e0=e=>{switch(e[3]){case"cpu":return new Bs(e[0],e[2],e[1]);case"gpu-buffer":{let r=e[0];if(!Xm(r))throw new Error(`not supported data type: ${r} for deserializing GPU tensor`);let{gpuBuffer:t,download:i,dispose:o}=e[2];return Bs.fromGpuBuffer(t,{dataType:r,dims:e[1],download:i,dispose:o})}case"ml-tensor":{let r=e[0];if(!Jm(r))throw new Error(`not supported data type: ${r} for deserializing MLTensor tensor`);let{mlTensor:t,download:i,dispose:o}=e[2];return Bs.fromMLTensor(t,{dataType:r,dims:e[1],download:i,dispose:o})}default:throw new Error(`invalid data location: ${e[3]}`)}},Nv=class{async fetchModelAndCopyToWasmMemory(e){return Dv(await Ym(e))}async loadModel(e,r){Ns();let t;typeof e=="string"?t=await this.fetchModelAndCopyToWasmMemory(e):t=e,[this.sessionId,this.inputNames,this.outputNames,this.inputMetadata,this.outputMetadata]=await jv(t,r),vs()}async dispose(){return Lv(this.sessionId)}async run(e,r,t){Ns();let i=[],o=[];Object.entries(e).forEach(m=>{let g=m[0],T=m[1],M=this.inputNames.indexOf(g);if(M===-1)throw new Error(`invalid input '${g}'`);i.push(T),o.push(M)});let n=[],u=[];Object.entries(r).forEach(m=>{let g=m[0],T=m[1],M=this.outputNames.indexOf(g);if(M===-1)throw new Error(`invalid output '${g}'`);n.push(T),u.push(M)});let d=i.map((m,g)=>Tm(m,()=>`input "${this.inputNames[o[g]]}"`)),h=n.map((m,g)=>m?Tm(m,()=>`output "${this.outputNames[u[g]]}"`):null),f=await zv(this.sessionId,o,d,u,h,t),y={};for(let m=0;mVm,initializeFlags:()=>Nm,wasmBackend:()=>Wv});var Nm,Vm,Wv,qT=Ye(()=>{Vs(),Rv(),HT(),Nm=()=>{(typeof dr.wasm.initTimeout!="number"||dr.wasm.initTimeout<0)&&(dr.wasm.initTimeout=0);let e=dr.wasm.simd;if(typeof e!="boolean"&&e!==void 0&&e!=="fixed"&&e!=="relaxed"&&(console.warn(`Property "env.wasm.simd" is set to unknown value "${e}". Reset it to \`false\` and ignore SIMD feature checking.`),dr.wasm.simd=!1),typeof dr.wasm.proxy!="boolean"&&(dr.wasm.proxy=!1),typeof dr.wasm.trace!="boolean"&&(dr.wasm.trace=!1),typeof dr.wasm.numThreads!="number"||!Number.isInteger(dr.wasm.numThreads)||dr.wasm.numThreads<=0)if(typeof self<"u"&&!self.crossOriginIsolated)dr.wasm.numThreads=1;else{let r=typeof navigator>"u"?F1("node:os").cpus().length:navigator.hardwareConcurrency;dr.wasm.numThreads=Math.min(4,Math.ceil((r||1)/2))}},Vm=class{async init(e){Nm(),await Ov(),await Fv(e)}async createInferenceSessionHandler(e,r){let t=new Nv;return await t.loadModel(e,r),t}},Wv=new Vm});Vs();Vs();Vs();var QT="1.23.0",XT=C0;{let e=(qT(),wd(Vv)).wasmBackend;fi("webgpu",e,5),fi("webnn",e,5),fi("cpu",e,10),fi("wasm",e,10)}Object.defineProperty(dr.versions,"web",{value:QT,enumerable:!0});/** +* @license +* Copyright 2021 Google LLC. All Rights Reserved. +* Licensed under the Apache License, Version 2.0 (the "License"); +* you may not use this file except in compliance with the License. +* You may obtain a copy of the License at +* +* http://www.apache.org/licenses/LICENSE-2.0 +* +* Unless required by applicable law or agreed to in writing, software +* distributed under the License is distributed on an "AS IS" BASIS, +* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +* See the License for the specific language governing permissions and +* limitations under the License. +* ============================================================================= +*//** + * @license + * Copyright 2020 Google LLC. All Rights Reserved. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * ============================================================================= + *//** + * @license + * Copyright 2019 Google LLC. All Rights Reserved. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + * ============================================================================= + */var JT=Object.freeze({__proto__:null,get InferenceSession(){return Gm},get TRACE(){return bd},get TRACE_FUNC_BEGIN(){return Ns},get TRACE_FUNC_END(){return vs},get Tensor(){return Bs},default:XT,get env(){return dr},get registerBackend(){return fi}}),hn={},YT={"onnxruntime-common":e=>{e.exports=$1},"onnxruntime-web":e=>{e.exports=JT},"?2ce3":()=>{},"?7a2c":()=>{},"?a42a":()=>{},"?2b25":()=>{},"?569f":()=>{},"?3f59":()=>{},"?154a":()=>{},"./node_modules/@huggingface/jinja/dist/index.js":(e,r,t)=>{t.r(r),t.d(r,{Environment:()=>Je,Interpreter:()=>at,Template:()=>as,parse:()=>Q,tokenize:()=>m});var i=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:i.Set,for:i.For,in:i.In,is:i.Is,if:i.If,else:i.Else,endset:i.EndSet,endif:i.EndIf,elif:i.ElseIf,endfor:i.EndFor,and:i.And,or:i.Or,not:i.Not,"not in":i.NotIn,macro:i.Macro,endmacro:i.EndMacro,break:i.Break,continue:i.Continue,true:i.BooleanLiteral,false:i.BooleanLiteral,none:i.NullLiteral,True:i.BooleanLiteral,False:i.BooleanLiteral,None:i.NullLiteral}),n=class{constructor(W,ye){this.value=W,this.type=ye}};function u(W){return/\w/.test(W)}function d(W){return/[0-9]/.test(W)}var h=[["{%",i.OpenStatement],["%}",i.CloseStatement],["{{",i.OpenExpression],["}}",i.CloseExpression],["(",i.OpenParen],[")",i.CloseParen],["{",i.OpenCurlyBracket],["}",i.CloseCurlyBracket],["[",i.OpenSquareBracket],["]",i.CloseSquareBracket],[",",i.Comma],[".",i.Dot],[":",i.Colon],["|",i.Pipe],["<=",i.ComparisonBinaryOperator],[">=",i.ComparisonBinaryOperator],["==",i.ComparisonBinaryOperator],["!=",i.ComparisonBinaryOperator],["<",i.ComparisonBinaryOperator],[">",i.ComparisonBinaryOperator],["+",i.AdditiveBinaryOperator],["-",i.AdditiveBinaryOperator],["*",i.MultiplicativeBinaryOperator],["/",i.MultiplicativeBinaryOperator],["%",i.MultiplicativeBinaryOperator],["=",i.Equals]],f=new Map([["n",` +`],["t"," "],["r","\r"],["b","\b"],["f","\f"],["v","\v"],["'","'"],['"','"'],["\\","\\"]]);function y(W,ye={}){return W.endsWith(` +`)&&(W=W.slice(0,-1)),W=W.replace(/{#.*?#}/gs,"{##}"),ye.lstrip_blocks&&(W=W.replace(/^[ \t]*({[#%])/gm,"$1")),ye.trim_blocks&&(W=W.replace(/([#%]})\n/g,"$1")),W.replace(/{##}/g,"").replace(/-%}\s*/g,"%}").replace(/\s*{%-/g,"{%").replace(/-}}\s*/g,"}}").replace(/\s*{{-/g,"{{")}function m(W,ye={}){var Le,Ft,Qt;const G=[],ne=y(W,ye);let $e=0;const Xe=wt=>{let Nt="";for(;wt(ne[$e]);){if(ne[$e]==="\\"){if(++$e,$e>=ne.length)throw new SyntaxError("Unexpected end of input");const mt=ne[$e++],tr=f.get(mt);if(tr===void 0)throw new SyntaxError(`Unexpected escaped character: ${mt}`);Nt+=tr;continue}if(Nt+=ne[$e++],$e>=ne.length)throw new SyntaxError("Unexpected end of input")}return Nt};e:for(;$e0){G.push(new n(mt,i.Text));continue}}Xe(mt=>/\s/.test(mt));const Nt=ne[$e];if(Nt==="-"||Nt==="+"){const mt=(Ft=G.at(-1))==null?void 0:Ft.type;if(mt===i.Text||mt===void 0)throw new SyntaxError(`Unexpected character: ${Nt}`);switch(mt){case i.Identifier:case i.NumericLiteral:case i.BooleanLiteral:case i.NullLiteral:case i.StringLiteral:case i.CloseParen:case i.CloseSquareBracket:break;default:{++$e;const tr=Xe(d);G.push(new n(`${Nt}${tr}`,tr.length>0?i.NumericLiteral:i.UnaryOperator));continue}}}for(const[mt,tr]of h)if(ne.slice($e,$e+mt.length)===mt){G.push(new n(mt,tr)),$e+=mt.length;continue e}if(Nt==="'"||Nt==='"'){++$e;const mt=Xe(tr=>tr!==Nt);G.push(new n(mt,i.StringLiteral)),++$e;continue}if(d(Nt)){const mt=Xe(d);G.push(new n(mt,i.NumericLiteral));continue}if(u(Nt)){const mt=Xe(u),tr=Object.hasOwn(o,mt)?o[mt]:i.Identifier;tr===i.In&&((Qt=G.at(-1))==null?void 0:Qt.type)===i.Not?(G.pop(),G.push(new n("not in",i.NotIn))):G.push(new n(mt,tr));continue}throw new SyntaxError(`Unexpected character: ${Nt}`)}return G}var g=class{constructor(){de(this,"type","Statement")}},T=class extends g{constructor(ye){super();de(this,"type","Program");this.body=ye}},M=class extends g{constructor(ye,G,ne){super();de(this,"type","If");this.test=ye,this.body=G,this.alternate=ne}},I=class extends g{constructor(ye,G,ne,$e){super();de(this,"type","For");this.loopvar=ye,this.iterable=G,this.body=ne,this.defaultBlock=$e}},z=class extends g{constructor(){super(...arguments);de(this,"type","Break")}},E=class extends g{constructor(){super(...arguments);de(this,"type","Continue")}},C=class extends g{constructor(ye,G,ne){super();de(this,"type","Set");this.assignee=ye,this.value=G,this.body=ne}},D=class extends g{constructor(ye,G,ne){super();de(this,"type","Macro");this.name=ye,this.args=G,this.body=ne}},A=class extends g{constructor(){super(...arguments);de(this,"type","Expression")}},$=class extends A{constructor(ye,G,ne){super();de(this,"type","MemberExpression");this.object=ye,this.property=G,this.computed=ne}},P=class extends A{constructor(ye,G){super();de(this,"type","CallExpression");this.callee=ye,this.args=G}},k=class extends A{constructor(ye){super();de(this,"type","Identifier");this.value=ye}},O=class extends A{constructor(ye){super();de(this,"type","Literal");this.value=ye}},R=class extends O{constructor(){super(...arguments);de(this,"type","NumericLiteral")}},U=class extends O{constructor(){super(...arguments);de(this,"type","StringLiteral")}},te=class extends O{constructor(){super(...arguments);de(this,"type","BooleanLiteral")}},se=class extends O{constructor(){super(...arguments);de(this,"type","NullLiteral")}},K=class extends O{constructor(){super(...arguments);de(this,"type","ArrayLiteral")}},pe=class extends O{constructor(){super(...arguments);de(this,"type","TupleLiteral")}},re=class extends O{constructor(){super(...arguments);de(this,"type","ObjectLiteral")}},oe=class extends A{constructor(ye,G,ne){super();de(this,"type","BinaryExpression");this.operator=ye,this.left=G,this.right=ne}},ge=class extends A{constructor(ye,G){super();de(this,"type","FilterExpression");this.operand=ye,this.filter=G}},le=class extends A{constructor(ye,G){super();de(this,"type","SelectExpression");this.iterable=ye,this.test=G}},Se=class extends A{constructor(ye,G,ne){super();de(this,"type","TestExpression");this.operand=ye,this.negate=G,this.test=ne}},Ce=class extends A{constructor(ye,G){super();de(this,"type","UnaryExpression");this.operator=ye,this.argument=G}},q=class extends A{constructor(ye=void 0,G=void 0,ne=void 0){super();de(this,"type","SliceExpression");this.start=ye,this.stop=G,this.step=ne}},N=class extends A{constructor(ye,G){super();de(this,"type","KeywordArgumentExpression");this.key=ye,this.value=G}};function Q(W){const ye=new T([]);let G=0;function ne(qe,ut){const bt=W[G++];if(!bt||bt.type!==qe)throw new Error(`Parser Error: ${ut}. ${bt.type} !== ${qe}.`);return bt}function $e(){switch(W[G].type){case i.Text:return Ft();case i.OpenStatement:return Qt();case i.OpenExpression:return wt();default:throw new SyntaxError(`Unexpected token type: ${W[G].type}`)}}function Xe(...qe){return G+qe.length<=W.length&&qe.some((ut,bt)=>ut!==W[G+bt].type)}function Le(...qe){return G+qe.length<=W.length&&qe.every((ut,bt)=>ut===W[G+bt].type)}function Ft(){return new U(ne(i.Text,"Expected text token").value)}function Qt(){ne(i.OpenStatement,"Expected opening statement token");let qe;switch(W[G].type){case i.Set:++G,qe=Nt(),ne(i.CloseStatement,"Expected closing statement token");break;case i.If:++G,qe=mt(),ne(i.OpenStatement,"Expected {% token"),ne(i.EndIf,"Expected endif token"),ne(i.CloseStatement,"Expected %} token");break;case i.Macro:++G,qe=tr(),ne(i.OpenStatement,"Expected {% token"),ne(i.EndMacro,"Expected endmacro token"),ne(i.CloseStatement,"Expected %} token");break;case i.For:++G,qe=Fr(),ne(i.OpenStatement,"Expected {% token"),ne(i.EndFor,"Expected endfor token"),ne(i.CloseStatement,"Expected %} token");break;case i.Break:++G,ne(i.CloseStatement,"Expected closing statement token"),qe=new z;break;case i.Continue:++G,ne(i.CloseStatement,"Expected closing statement token"),qe=new E;break;default:throw new SyntaxError(`Unknown statement type: ${W[G].type}`)}return qe}function wt(){ne(i.OpenExpression,"Expected opening expression token");const qe=pr();return ne(i.CloseExpression,"Expected closing expression token"),qe}function Nt(){var ut,bt;const qe=pr();if(Le(i.Equals)){++G;const cr=pr();return new C(qe,cr,[])}else{const cr=[];for(ne(i.CloseStatement,"Expected %} token");!(((ut=W[G])==null?void 0:ut.type)===i.OpenStatement&&((bt=W[G+1])==null?void 0:bt.type)===i.EndSet);){const ls=$e();cr.push(ls)}return ne(i.OpenStatement,"Expected {% token"),ne(i.EndSet,"Expected endset token"),new C(qe,null,cr)}}function mt(){var cr,ls,Es,Xr,Ps,Ss,ks,$s;const qe=pr();ne(i.CloseStatement,"Expected closing statement token");const ut=[],bt=[];for(;!(((cr=W[G])==null?void 0:cr.type)===i.OpenStatement&&(((ls=W[G+1])==null?void 0:ls.type)===i.ElseIf||((Es=W[G+1])==null?void 0:Es.type)===i.Else||((Xr=W[G+1])==null?void 0:Xr.type)===i.EndIf));)ut.push($e());if(((Ps=W[G])==null?void 0:Ps.type)===i.OpenStatement&&((Ss=W[G+1])==null?void 0:Ss.type)!==i.EndIf)if(++G,Le(i.ElseIf))ne(i.ElseIf,"Expected elseif token"),bt.push(mt());else for(ne(i.Else,"Expected else token"),ne(i.CloseStatement,"Expected closing statement token");!(((ks=W[G])==null?void 0:ks.type)===i.OpenStatement&&(($s=W[G+1])==null?void 0:$s.type)===i.EndIf);)bt.push($e());return new M(qe,ut,bt)}function tr(){const qe=Er();if(qe.type!=="Identifier")throw new SyntaxError("Expected identifier following macro statement");const ut=Us();ne(i.CloseStatement,"Expected closing statement token");const bt=[];for(;Xe(i.OpenStatement,i.EndMacro);)bt.push($e());return new D(qe,ut,bt)}function rr(qe=!1){const ut=qe?Er:pr,bt=[ut()],cr=Le(i.Comma);for(;cr&&(++G,bt.push(ut()),!!Le(i.Comma)););return cr?new pe(bt):bt[0]}function Fr(){const qe=rr(!0);if(!(qe instanceof k||qe instanceof pe))throw new SyntaxError(`Expected identifier/tuple for the loop variable, got ${qe.type} instead`);ne(i.In,"Expected `in` keyword following loop variable");const ut=pr();ne(i.CloseStatement,"Expected closing statement token");const bt=[];for(;Xe(i.OpenStatement,i.EndFor)&&Xe(i.OpenStatement,i.Else);)bt.push($e());const cr=[];if(Le(i.OpenStatement,i.Else))for(++G,++G,ne(i.CloseStatement,"Expected closing statement token");Xe(i.OpenStatement,i.EndFor);)cr.push($e());return new I(qe,ut,bt,cr)}function pr(){return Kr()}function Kr(){const qe=xs();if(Le(i.If)){++G;const ut=xs();if(Le(i.Else)){++G;const bt=xs();return new M(ut,[qe],[bt])}else return new le(qe,ut)}return qe}function xs(){let qe=Ws();for(;Le(i.Or);){const ut=W[G];++G;const bt=Ws();qe=new oe(ut,qe,bt)}return qe}function Ws(){let qe=Ts();for(;Le(i.And);){const ut=W[G];++G;const bt=Ts();qe=new oe(ut,qe,bt)}return qe}function Ts(){let qe;for(;Le(i.Not);){const ut=W[G];++G;const bt=Ts();qe=new Ce(ut,bt)}return qe??en()}function en(){let qe=rs();for(;Le(i.ComparisonBinaryOperator)||Le(i.In)||Le(i.NotIn);){const ut=W[G];++G;const bt=rs();qe=new oe(ut,qe,bt)}return qe}function rs(){let qe=Jt();for(;Le(i.AdditiveBinaryOperator);){const ut=W[G];++G;const bt=Jt();qe=new oe(ut,qe,bt)}return qe}function tn(){const qe=Wr(Er());return Le(i.OpenParen)?os(qe):qe}function os(qe){let ut=new P(qe,Us());return ut=Wr(ut),Le(i.OpenParen)&&(ut=os(ut)),ut}function Us(){ne(i.OpenParen,"Expected opening parenthesis for arguments list");const qe=Gs();return ne(i.CloseParen,"Expected closing parenthesis for arguments list"),qe}function Gs(){const qe=[];for(;!Le(i.CloseParen);){let ut=pr();if(Le(i.Equals)){if(++G,!(ut instanceof k))throw new SyntaxError("Expected identifier for keyword argument");const bt=pr();ut=new N(ut,bt)}qe.push(ut),Le(i.Comma)&&++G}return qe}function Ks(){const qe=[];let ut=!1;for(;!Le(i.CloseSquareBracket);)Le(i.Colon)?(qe.push(void 0),++G,ut=!0):(qe.push(pr()),Le(i.Colon)&&(++G,ut=!0));if(qe.length===0)throw new SyntaxError("Expected at least one argument for member/slice expression");if(ut){if(qe.length>3)throw new SyntaxError("Expected 0-3 arguments for slice expression");return new q(...qe)}return qe[0]}function Wr(qe){for(;Le(i.Dot)||Le(i.OpenSquareBracket);){const ut=W[G];++G;let bt;const cr=ut.type!==i.Dot;if(cr)bt=Ks(),ne(i.CloseSquareBracket,"Expected closing square bracket");else if(bt=Er(),bt.type!=="Identifier")throw new SyntaxError("Expected identifier following dot operator");qe=new $(qe,bt,cr)}return qe}function Jt(){let qe=Cs();for(;Le(i.MultiplicativeBinaryOperator);){const ut=W[G];++G;const bt=Cs();qe=new oe(ut,qe,bt)}return qe}function Cs(){let qe=Hs();for(;Le(i.Is);){++G;const ut=Le(i.Not);ut&&++G;let bt=Er();if(bt instanceof te?bt=new k(bt.value.toString()):bt instanceof se&&(bt=new k("none")),!(bt instanceof k))throw new SyntaxError("Expected identifier for the test");qe=new Se(qe,ut,bt)}return qe}function Hs(){let qe=tn();for(;Le(i.Pipe);){++G;let ut=Er();if(!(ut instanceof k))throw new SyntaxError("Expected identifier for the filter");Le(i.OpenParen)&&(ut=os(ut)),qe=new ge(qe,ut)}return qe}function Er(){const qe=W[G];switch(qe.type){case i.NumericLiteral:return++G,new R(Number(qe.value));case i.StringLiteral:return++G,new U(qe.value);case i.BooleanLiteral:return++G,new te(qe.value.toLowerCase()==="true");case i.NullLiteral:return++G,new se(null);case i.Identifier:return++G,new k(qe.value);case i.OpenParen:{++G;const ut=rr();if(W[G].type!==i.CloseParen)throw new SyntaxError(`Expected closing parenthesis, got ${W[G].type} instead`);return++G,ut}case i.OpenSquareBracket:{++G;const ut=[];for(;!Le(i.CloseSquareBracket);)ut.push(pr()),Le(i.Comma)&&++G;return++G,new K(ut)}case i.OpenCurlyBracket:{++G;const ut=new Map;for(;!Le(i.CloseCurlyBracket);){const bt=pr();ne(i.Colon,"Expected colon between key and value in object literal");const cr=pr();ut.set(bt,cr),Le(i.Comma)&&++G}return++G,new re(ut)}default:throw new SyntaxError(`Unexpected token: ${qe.type}`)}}for(;G=0?(ye=(ye??(ye=0))<0?Math.max(W.length+ye,0):Math.min(ye,W.length),G=(G??(G=W.length))<0?Math.max(W.length+G,0):Math.min(G,W.length)):(ye=(ye??(ye=W.length-1))<0?Math.max(W.length+ye,-1):Math.min(ye,W.length-1),G=(G??(G=-1))<-1?Math.max(W.length+G,-1):Math.min(G,W.length-1));const Xe=[];for(let Le=ye;$e*Le<$e*G;Le+=ne)Xe.push(W[Le]);return Xe}function we(W){return W.replace(/\b\w/g,ye=>ye.toUpperCase())}var ae=class extends Error{},Ze=class extends Error{},et=class{constructor(W=void 0){de(this,"type","RuntimeValue");de(this,"value");de(this,"builtins",new Map);this.value=W}__bool__(){return new me(!!this.value)}},Ee=class extends et{constructor(){super(...arguments);de(this,"type","NumericValue")}},Z=class extends et{constructor(){super(...arguments);de(this,"type","StringValue");de(this,"builtins",new Map([["upper",new Re(()=>new Z(this.value.toUpperCase()))],["lower",new Re(()=>new Z(this.value.toLowerCase()))],["strip",new Re(()=>new Z(this.value.trim()))],["title",new Re(()=>new Z(we(this.value)))],["length",new Ee(this.value.length)],["rstrip",new Re(()=>new Z(this.value.trimEnd()))],["lstrip",new Re(()=>new Z(this.value.trimStart()))],["startswith",new Re(ye=>{if(ye.length===0)throw new Error("startswith() requires at least one argument");const G=ye[0];if(!(G instanceof Z))throw new Error("startswith() argument must be a string");return new me(this.value.startsWith(G.value))})],["endswith",new Re(ye=>{if(ye.length===0)throw new Error("endswith() requires at least one argument");const G=ye[0];if(!(G instanceof Z))throw new Error("endswith() argument must be a string");return new me(this.value.endsWith(G.value))})],["split",new Re(ye=>{const G=ye[0]??new Ue;if(!(G instanceof Z||G instanceof Ue))throw new Error("sep argument must be a string or null");const ne=ye[1]??new Ee(-1);if(!(ne instanceof Ee))throw new Error("maxsplit argument must be a number");let $e=[];if(G instanceof Ue){const Xe=this.value.trimStart();for(const{0:Le,index:Ft}of Xe.matchAll(/\S+/g)){if(ne.value!==-1&&$e.length>=ne.value&&Ft!==void 0){$e.push(Le+Xe.slice(Ft+Le.length));break}$e.push(Le)}}else{if(G.value==="")throw new Error("empty separator");$e=this.value.split(G.value),ne.value!==-1&&$e.length>ne.value&&$e.push($e.splice(ne.value).join(G.value))}return new Fe($e.map(Xe=>new Z(Xe)))})]]))}},me=class extends et{constructor(){super(...arguments);de(this,"type","BooleanValue")}},X=class extends et{constructor(){super(...arguments);de(this,"type","ObjectValue");de(this,"builtins",new Map([["get",new Re(([ye,G])=>{if(!(ye instanceof Z))throw new Error(`Object key must be a string: got ${ye.type}`);return this.value.get(ye.value)??G??new Ue})],["items",new Re(()=>new Fe(Array.from(this.value.entries()).map(([ye,G])=>new Fe([new Z(ye),G]))))]]))}__bool__(){return new me(this.value.size>0)}},Oe=class extends X{constructor(){super(...arguments);de(this,"type","KeywordArgumentsValue")}},Fe=class extends et{constructor(){super(...arguments);de(this,"type","ArrayValue");de(this,"builtins",new Map([["length",new Ee(this.value.length)]]))}__bool__(){return new me(this.value.length>0)}},Ve=class extends Fe{constructor(){super(...arguments);de(this,"type","TupleValue")}},Re=class extends et{constructor(){super(...arguments);de(this,"type","FunctionValue")}},Ue=class extends et{constructor(){super(...arguments);de(this,"type","NullValue")}},Ge=class extends et{constructor(){super(...arguments);de(this,"type","UndefinedValue")}},Je=class{constructor(W){de(this,"variables",new Map([["namespace",new Re(W=>{if(W.length===0)return new X(new Map);if(W.length!==1||!(W[0]instanceof X))throw new Error("`namespace` expects either zero arguments or a single object argument");return W[0]})]]));de(this,"tests",new Map([["boolean",W=>W.type==="BooleanValue"],["callable",W=>W instanceof Re],["odd",W=>{if(W.type!=="NumericValue")throw new Error(`Cannot apply test "odd" to type: ${W.type}`);return W.value%2!==0}],["even",W=>{if(W.type!=="NumericValue")throw new Error(`Cannot apply test "even" to type: ${W.type}`);return W.value%2===0}],["false",W=>W.type==="BooleanValue"&&!W.value],["true",W=>W.type==="BooleanValue"&&W.value],["none",W=>W.type==="NullValue"],["string",W=>W.type==="StringValue"],["number",W=>W.type==="NumericValue"],["integer",W=>W.type==="NumericValue"&&Number.isInteger(W.value)],["iterable",W=>W.type==="ArrayValue"||W.type==="StringValue"],["mapping",W=>W.type==="ObjectValue"],["lower",W=>{const ye=W.value;return W.type==="StringValue"&&ye===ye.toLowerCase()}],["upper",W=>{const ye=W.value;return W.type==="StringValue"&&ye===ye.toUpperCase()}],["none",W=>W.type==="NullValue"],["defined",W=>W.type!=="UndefinedValue"],["undefined",W=>W.type==="UndefinedValue"],["equalto",(W,ye)=>W.value===ye.value],["eq",(W,ye)=>W.value===ye.value]]));this.parent=W}set(W,ye){return this.declareVariable(W,ot(ye))}declareVariable(W,ye){if(this.variables.has(W))throw new SyntaxError(`Variable already declared: ${W}`);return this.variables.set(W,ye),ye}setVariable(W,ye){return this.variables.set(W,ye),ye}resolve(W){if(this.variables.has(W))return this;if(this.parent)return this.parent.resolve(W);throw new Error(`Unknown variable: ${W}`)}lookupVariable(W){try{return this.resolve(W).variables.get(W)??new Ge}catch{return new Ge}}},at=class{constructor(W){de(this,"global");this.global=W??new Je}run(W){return this.evaluate(W,this.global)}evaluateBinaryExpression(W,ye){const G=this.evaluate(W.left,ye);switch(W.operator.value){case"and":return G.__bool__().value?this.evaluate(W.right,ye):G;case"or":return G.__bool__().value?G:this.evaluate(W.right,ye)}const ne=this.evaluate(W.right,ye);switch(W.operator.value){case"==":return new me(G.value==ne.value);case"!=":return new me(G.value!=ne.value)}if(G instanceof Ge||ne instanceof Ge)throw new Error("Cannot perform operation on undefined values");if(G instanceof Ue||ne instanceof Ue)throw new Error("Cannot perform operation on null values");if(G instanceof Ee&&ne instanceof Ee)switch(W.operator.value){case"+":return new Ee(G.value+ne.value);case"-":return new Ee(G.value-ne.value);case"*":return new Ee(G.value*ne.value);case"/":return new Ee(G.value/ne.value);case"%":return new Ee(G.value%ne.value);case"<":return new me(G.value":return new me(G.value>ne.value);case">=":return new me(G.value>=ne.value);case"<=":return new me(G.value<=ne.value)}else if(G instanceof Fe&&ne instanceof Fe)switch(W.operator.value){case"+":return new Fe(G.value.concat(ne.value))}else if(ne instanceof Fe){const $e=ne.value.find(Xe=>Xe.value===G.value)!==void 0;switch(W.operator.value){case"in":return new me($e);case"not in":return new me(!$e)}}if(G instanceof Z||ne instanceof Z)switch(W.operator.value){case"+":return new Z(G.value.toString()+ne.value.toString())}if(G instanceof Z&&ne instanceof Z)switch(W.operator.value){case"in":return new me(ne.value.includes(G.value));case"not in":return new me(!ne.value.includes(G.value))}if(G instanceof Z&&ne instanceof X)switch(W.operator.value){case"in":return new me(ne.value.has(G.value));case"not in":return new me(!ne.value.has(G.value))}throw new SyntaxError(`Unknown operator "${W.operator.value}" between ${G.type} and ${ne.type}`)}evaluateArguments(W,ye){const G=[],ne=new Map;for(const $e of W)if($e.type==="KeywordArgumentExpression"){const Xe=$e;ne.set(Xe.key.value,this.evaluate(Xe.value,ye))}else{if(ne.size>0)throw new Error("Positional arguments must come before keyword arguments");G.push(this.evaluate($e,ye))}return[G,ne]}evaluateFilterExpression(W,ye){const G=this.evaluate(W.operand,ye);if(W.filter.type==="Identifier"){const ne=W.filter;if(ne.value==="tojson")return new Z(Ke(G));if(G instanceof Fe)switch(ne.value){case"list":return G;case"first":return G.value[0];case"last":return G.value[G.value.length-1];case"length":return new Ee(G.value.length);case"reverse":return new Fe(G.value.reverse());case"sort":return new Fe(G.value.sort(($e,Xe)=>{if($e.type!==Xe.type)throw new Error(`Cannot compare different types: ${$e.type} and ${Xe.type}`);switch($e.type){case"NumericValue":return $e.value-Xe.value;case"StringValue":return $e.value.localeCompare(Xe.value);default:throw new Error(`Cannot compare type: ${$e.type}`)}}));case"join":return new Z(G.value.map($e=>$e.value).join(""));case"string":return new Z(Ke(G));default:throw new Error(`Unknown ArrayValue filter: ${ne.value}`)}else if(G instanceof Z)switch(ne.value){case"length":return new Ee(G.value.length);case"upper":return new Z(G.value.toUpperCase());case"lower":return new Z(G.value.toLowerCase());case"title":return new Z(we(G.value));case"capitalize":return new Z(G.value.charAt(0).toUpperCase()+G.value.slice(1));case"trim":return new Z(G.value.trim());case"indent":return new Z(G.value.split(` +`).map(($e,Xe)=>Xe===0||$e.length===0?$e:" "+$e).join(` +`));case"join":case"string":return G;default:throw new Error(`Unknown StringValue filter: ${ne.value}`)}else if(G instanceof Ee)switch(ne.value){case"abs":return new Ee(Math.abs(G.value));default:throw new Error(`Unknown NumericValue filter: ${ne.value}`)}else if(G instanceof X)switch(ne.value){case"items":return new Fe(Array.from(G.value.entries()).map(([$e,Xe])=>new Fe([new Z($e),Xe])));case"length":return new Ee(G.value.size);default:throw new Error(`Unknown ObjectValue filter: ${ne.value}`)}throw new Error(`Cannot apply filter "${ne.value}" to type: ${G.type}`)}else if(W.filter.type==="CallExpression"){const ne=W.filter;if(ne.callee.type!=="Identifier")throw new Error(`Unknown filter: ${ne.callee.type}`);const $e=ne.callee.value;if($e==="tojson"){const[,Xe]=this.evaluateArguments(ne.args,ye),Le=Xe.get("indent")??new Ue;if(!(Le instanceof Ee||Le instanceof Ue))throw new Error("If set, indent must be a number");return new Z(Ke(G,Le.value))}else if($e==="join"){let Xe;if(G instanceof Z)Xe=Array.from(G.value);else if(G instanceof Fe)Xe=G.value.map(wt=>wt.value);else throw new Error(`Cannot apply filter "${$e}" to type: ${G.type}`);const[Le,Ft]=this.evaluateArguments(ne.args,ye),Qt=Le.at(0)??Ft.get("separator")??new Z("");if(!(Qt instanceof Z))throw new Error("separator must be a string");return new Z(Xe.join(Qt.value))}if(G instanceof Fe){switch($e){case"selectattr":case"rejectattr":{const Xe=$e==="selectattr";if(G.value.some(mt=>!(mt instanceof X)))throw new Error(`\`${$e}\` can only be applied to array of objects`);if(ne.args.some(mt=>mt.type!=="StringLiteral"))throw new Error(`arguments of \`${$e}\` must be strings`);const[Le,Ft,Qt]=ne.args.map(mt=>this.evaluate(mt,ye));let wt;if(Ft){const mt=ye.tests.get(Ft.value);if(!mt)throw new Error(`Unknown test: ${Ft.value}`);wt=mt}else wt=(...mt)=>mt[0].__bool__().value;const Nt=G.value.filter(mt=>{const tr=mt.value.get(Le.value),rr=tr?wt(tr,Qt):!1;return Xe?rr:!rr});return new Fe(Nt)}case"map":{const[,Xe]=this.evaluateArguments(ne.args,ye);if(Xe.has("attribute")){const Le=Xe.get("attribute");if(!(Le instanceof Z))throw new Error("attribute must be a string");const Ft=Xe.get("default"),Qt=G.value.map(wt=>{if(!(wt instanceof X))throw new Error("items in map must be an object");return wt.value.get(Le.value)??Ft??new Ge});return new Fe(Qt)}else throw new Error("`map` expressions without `attribute` set are not currently supported.")}}throw new Error(`Unknown ArrayValue filter: ${$e}`)}else if(G instanceof Z){switch($e){case"indent":{const[Xe,Le]=this.evaluateArguments(ne.args,ye),Ft=Xe.at(0)??Le.get("width")??new Ee(4);if(!(Ft instanceof Ee))throw new Error("width must be a number");const Qt=Xe.at(1)??Le.get("first")??new me(!1),wt=Xe.at(2)??Le.get("blank")??new me(!1),Nt=G.value.split(` +`),mt=" ".repeat(Ft.value),tr=Nt.map((rr,Fr)=>!Qt.value&&Fr===0||!wt.value&&rr.length===0?rr:mt+rr);return new Z(tr.join(` +`))}}throw new Error(`Unknown StringValue filter: ${$e}`)}else throw new Error(`Cannot apply filter "${$e}" to type: ${G.type}`)}throw new Error(`Unknown filter: ${W.filter.type}`)}evaluateTestExpression(W,ye){const G=this.evaluate(W.operand,ye),ne=ye.tests.get(W.test.value);if(!ne)throw new Error(`Unknown test: ${W.test.value}`);const $e=ne(G);return new me(W.negate?!$e:$e)}evaluateUnaryExpression(W,ye){const G=this.evaluate(W.argument,ye);switch(W.operator.value){case"not":return new me(!G.value);default:throw new SyntaxError(`Unknown operator: ${W.operator.value}`)}}evalProgram(W,ye){return this.evaluateBlock(W.body,ye)}evaluateBlock(W,ye){let G="";for(const ne of W){const $e=this.evaluate(ne,ye);$e.type!=="NullValue"&&$e.type!=="UndefinedValue"&&(G+=$e.value)}return new Z(G)}evaluateIdentifier(W,ye){return ye.lookupVariable(W.value)}evaluateCallExpression(W,ye){const[G,ne]=this.evaluateArguments(W.args,ye);ne.size>0&&G.push(new Oe(ne));const $e=this.evaluate(W.callee,ye);if($e.type!=="FunctionValue")throw new Error(`Cannot call something that is not a function: got ${$e.type}`);return $e.value(G,ye)}evaluateSliceExpression(W,ye,G){if(!(W instanceof Fe||W instanceof Z))throw new Error("Slice object must be an array or string");const ne=this.evaluate(ye.start,G),$e=this.evaluate(ye.stop,G),Xe=this.evaluate(ye.step,G);if(!(ne instanceof Ee||ne instanceof Ge))throw new Error("Slice start must be numeric or undefined");if(!($e instanceof Ee||$e instanceof Ge))throw new Error("Slice stop must be numeric or undefined");if(!(Xe instanceof Ee||Xe instanceof Ge))throw new Error("Slice step must be numeric or undefined");return W instanceof Fe?new Fe(Ae(W.value,ne.value,$e.value,Xe.value)):new Z(Ae(Array.from(W.value),ne.value,$e.value,Xe.value).join(""))}evaluateMemberExpression(W,ye){const G=this.evaluate(W.object,ye);let ne;if(W.computed){if(W.property.type==="SliceExpression")return this.evaluateSliceExpression(G,W.property,ye);ne=this.evaluate(W.property,ye)}else ne=new Z(W.property.value);let $e;if(G instanceof X){if(!(ne instanceof Z))throw new Error(`Cannot access property with non-string: got ${ne.type}`);$e=G.value.get(ne.value)??G.builtins.get(ne.value)}else if(G instanceof Fe||G instanceof Z)if(ne instanceof Ee)$e=G.value.at(ne.value),G instanceof Z&&($e=new Z(G.value.at(ne.value)));else if(ne instanceof Z)$e=G.builtins.get(ne.value);else throw new Error(`Cannot access property with non-string/non-number: got ${ne.type}`);else{if(!(ne instanceof Z))throw new Error(`Cannot access property with non-string: got ${ne.type}`);$e=G.builtins.get(ne.value)}return $e instanceof et?$e:new Ge}evaluateSet(W,ye){const G=W.value?this.evaluate(W.value,ye):this.evaluateBlock(W.body,ye);if(W.assignee.type==="Identifier"){const ne=W.assignee.value;ye.setVariable(ne,G)}else if(W.assignee.type==="MemberExpression"){const ne=W.assignee,$e=this.evaluate(ne.object,ye);if(!($e instanceof X))throw new Error("Cannot assign to member of non-object");if(ne.property.type!=="Identifier")throw new Error("Cannot assign to member with non-identifier property");$e.value.set(ne.property.value,G)}else throw new Error(`Invalid LHS inside assignment expression: ${JSON.stringify(W.assignee)}`);return new Ue}evaluateIf(W,ye){const G=this.evaluate(W.test,ye);return this.evaluateBlock(G.__bool__().value?W.body:W.alternate,ye)}evaluateFor(W,ye){const G=new Je(ye);let ne,$e;if(W.iterable.type==="SelectExpression"){const wt=W.iterable;$e=this.evaluate(wt.iterable,G),ne=wt.test}else $e=this.evaluate(W.iterable,G);if(!($e instanceof Fe))throw new Error(`Expected iterable type in for loop: got ${$e.type}`);const Xe=[],Le=[];for(let wt=0;wt<$e.value.length;++wt){const Nt=new Je(G),mt=$e.value[wt];let tr;if(W.loopvar.type==="Identifier")tr=rr=>rr.setVariable(W.loopvar.value,mt);else if(W.loopvar.type==="TupleLiteral"){const rr=W.loopvar;if(mt.type!=="ArrayValue")throw new Error(`Cannot unpack non-iterable type: ${mt.type}`);const Fr=mt;if(rr.value.length!==Fr.value.length)throw new Error(`Too ${rr.value.length>Fr.value.length?"few":"many"} items to unpack`);tr=pr=>{for(let Kr=0;Kr0?Xe[wt-1]:new Ge],["nextitem",wt{var Le;const $e=new Je(ne);G=G.slice();let Xe;((Le=G.at(-1))==null?void 0:Le.type)==="KeywordArgumentsValue"&&(Xe=G.pop());for(let Ft=0;Ftthis.evaluate(G,ye)));case"TupleLiteral":return new Ve(W.value.map(G=>this.evaluate(G,ye)));case"ObjectLiteral":{const G=new Map;for(const[ne,$e]of W.value){const Xe=this.evaluate(ne,ye);if(!(Xe instanceof Z))throw new Error(`Object keys must be strings: got ${Xe.type}`);G.set(Xe.value,this.evaluate($e,ye))}return new X(G)}case"Identifier":return this.evaluateIdentifier(W,ye);case"CallExpression":return this.evaluateCallExpression(W,ye);case"MemberExpression":return this.evaluateMemberExpression(W,ye);case"UnaryExpression":return this.evaluateUnaryExpression(W,ye);case"BinaryExpression":return this.evaluateBinaryExpression(W,ye);case"FilterExpression":return this.evaluateFilterExpression(W,ye);case"TestExpression":return this.evaluateTestExpression(W,ye);default:throw new SyntaxError(`Unknown node type: ${W.type}`)}}};function ot(W){switch(typeof W){case"number":return new Ee(W);case"string":return new Z(W);case"boolean":return new me(W);case"undefined":return new Ge;case"object":return W===null?new Ue:Array.isArray(W)?new Fe(W.map(ot)):new X(new Map(Object.entries(W).map(([ye,G])=>[ye,ot(G)])));case"function":return new Re((ye,G)=>{const ne=W(...ye.map($e=>$e.value))??null;return ot(ne)});default:throw new Error(`Cannot convert to runtime value: ${W}`)}}function Ke(W,ye,G){const ne=G??0;switch(W.type){case"NullValue":case"UndefinedValue":return"null";case"NumericValue":case"StringValue":case"BooleanValue":return JSON.stringify(W.value);case"ArrayValue":case"ObjectValue":{const $e=ye?" ".repeat(ye):"",Xe=` +`+$e.repeat(ne),Le=Xe+$e;if(W.type==="ArrayValue"){const Ft=W.value.map(Qt=>Ke(Qt,ye,ne+1));return ye?`[${Le}${Ft.join(`,${Le}`)}${Xe}]`:`[${Ft.join(", ")}]`}else{const Ft=Array.from(W.value.entries()).map(([Qt,wt])=>{const Nt=`"${Qt}": ${Ke(wt,ye,ne+1)}`;return ye?`${Le}${Nt}`:Nt});return ye?`{${Ft.join(",")}${Xe}}`:`{${Ft.join(", ")}}`}}default:throw new Error(`Cannot convert to JSON: ${W.type}`)}}var st=` +`,ft="{%- ",yt=" -%}",St={MultiplicativeBinaryOperator:2,AdditiveBinaryOperator:1,ComparisonBinaryOperator:0};function vt(W,ye=" "){const G=typeof ye=="number"?" ".repeat(ye):ye;return xt(W.body,0,G).replace(/\n$/,"")}function He(...W){return ft+W.join(" ")+yt}function xt(W,ye,G){return W.map(ne=>Vt(ne,ye,G)).join(st)}function Vt(W,ye,G){const ne=G.repeat(ye);switch(W.type){case"Program":return xt(W.body,ye,G);case"If":return he(W,ye,G);case"For":return ce(W,ye,G);case"Set":return Cr(W,ye,G);case"Macro":return er(W,ye,G);case"Break":return ne+He("break");case"Continue":return ne+He("continue");default:return ne+"{{- "+ht(W)+" -}}"}}function he(W,ye,G){const ne=G.repeat(ye),$e=[];let Xe=W;for(;Xe&&($e.push({test:Xe.test,body:Xe.body}),Xe.alternate.length===1&&Xe.alternate[0].type==="If");)Xe=Xe.alternate[0];let Le=ne+He("if",ht($e[0].test))+st+xt($e[0].body,ye+1,G);for(let Ft=1;Ft<$e.length;Ft++)Le+=st+ne+He("elif",ht($e[Ft].test))+st+xt($e[Ft].body,ye+1,G);return Xe&&Xe.alternate.length>0&&(Le+=st+ne+He("else")+st+xt(Xe.alternate,ye+1,G)),Le+=st+ne+He("endif"),Le}function ce(W,ye,G){const ne=G.repeat(ye);let $e="";if(W.iterable.type==="SelectExpression"){const Le=W.iterable;$e=`${ht(Le.iterable)} if ${ht(Le.test)}`}else $e=ht(W.iterable);let Xe=ne+He("for",ht(W.loopvar),"in",$e)+st+xt(W.body,ye+1,G);return W.defaultBlock.length>0&&(Xe+=st+ne+He("else")+st+xt(W.defaultBlock,ye+1,G)),Xe+=st+ne+He("endfor"),Xe}function Cr(W,ye,G){const ne=G.repeat(ye),$e=ht(W.assignee),Xe=W.value?ht(W.value):"",Le=ne+He("set",`${$e}${W.value?" = "+Xe:""}`);return W.body.length===0?Le:Le+st+xt(W.body,ye+1,G)+st+ne+He("endset")}function er(W,ye,G){const ne=G.repeat(ye),$e=W.args.map(ht).join(", ");return ne+He("macro",`${W.name.value}(${$e})`)+st+xt(W.body,ye+1,G)+st+ne+He("endmacro")}function ht(W,ye=-1){switch(W.type){case"Identifier":return W.value;case"NullLiteral":return"none";case"NumericLiteral":case"BooleanLiteral":return`${W.value}`;case"StringLiteral":return JSON.stringify(W.value);case"BinaryExpression":{const G=W,ne=St[G.operator.type]??0,$e=ht(G.left,ne),Xe=ht(G.right,ne+1),Le=`${$e} ${G.operator.value} ${Xe}`;return neht($e,-1)).join(", ");return`${ht(G.callee,-1)}(${ne})`}case"MemberExpression":{const G=W;let ne=ht(G.object,-1);G.object.type!=="Identifier"&&(ne=`(${ne})`);let $e=ht(G.property,-1);return!G.computed&&G.property.type!=="Identifier"&&($e=`(${$e})`),G.computed?`${ne}[${$e}]`:`${ne}.${$e}`}case"FilterExpression":{const G=W,ne=ht(G.operand,1/0);return G.filter.type==="CallExpression"?`${ne} | ${ht(G.filter,-1)}`:`${ne} | ${G.filter.value}`}case"SelectExpression":{const G=W;return`${ht(G.iterable,-1)} | select(${ht(G.test,-1)})`}case"TestExpression":{const G=W;return`${ht(G.operand,-1)} is${G.negate?" not":""} ${G.test.value}`}case"ArrayLiteral":case"TupleLiteral":{const G=W.value.map($e=>ht($e,-1)),ne=W.type==="ArrayLiteral"?"[]":"()";return`${ne[0]}${G.join(", ")}${ne[1]}`}case"ObjectLiteral":return`{ ${Array.from(W.value.entries()).map(([ne,$e])=>`${ht(ne,-1)}: ${ht($e,-1)}`).join(", ")} }`;case"SliceExpression":{const G=W,ne=G.start?ht(G.start,-1):"",$e=G.stop?ht(G.stop,-1):"",Xe=G.step?`:${ht(G.step,-1)}`:"";return`${ne}:${$e}${Xe}`}case"KeywordArgumentExpression":{const G=W;return`${G.key.value}=${ht(G.value,-1)}`}case"If":{const G=W,ne=ht(G.test,-1),$e=ht(G.body[0],0),Xe=ht(G.alternate[0],-1);return`${$e} if ${ne} else ${Xe}`}default:throw new Error(`Unknown expression type: ${W.type}`)}}var as=class{constructor(W){de(this,"parsed");const ye=m(W,{lstrip_blocks:!0,trim_blocks:!0});this.parsed=Q(ye)}render(W){const ye=new Je;if(ye.set("false",!1),ye.set("true",!0),ye.set("raise_exception",$e=>{throw new Error($e)}),ye.set("range",ue),W)for(const[$e,Xe]of Object.entries(W))ye.set($e,Xe);return new at(ye).run(this.parsed).value}format(W){return vt(this.parsed,(W==null?void 0:W.indent)||" ")}}},"./src/backends/onnx.js":(e,r,t)=>{var i;t.r(r),t.d(r,{Tensor:()=>d.Tensor,createInferenceSession:()=>z,deviceToExecutionProviders:()=>M,isONNXProxy:()=>D,isONNXTensor:()=>E});var o=t("./src/env.js"),n=t("?2ce3"),u=t("onnxruntime-web"),d=t("onnxruntime-common");const h=Object.freeze({auto:null,gpu:null,cpu:"cpu",wasm:"wasm",webgpu:"webgpu",cuda:"cuda",dml:"dml",webnn:{name:"webnn",deviceType:"cpu"},"webnn-npu":{name:"webnn",deviceType:"npu"},"webnn-gpu":{name:"webnn",deviceType:"gpu"},"webnn-cpu":{name:"webnn",deviceType:"cpu"}}),f=[];let y,m;const g=Symbol.for("onnxruntime");if(g in globalThis)m=globalThis[g];else if(o.apis.IS_NODE_ENV){switch(m=n??(i||(i=t.t(n,2))),process.platform){case"win32":f.push("dml");break;case"linux":process.arch==="x64"&&f.push("cuda");break}f.push("cpu"),y=["cpu"]}else m=u,o.apis.IS_WEBNN_AVAILABLE&&f.push("webnn-npu","webnn-gpu","webnn-cpu","webnn"),o.apis.IS_WEBGPU_AVAILABLE&&f.push("webgpu"),f.push("wasm"),y=["wasm"];const T=m.InferenceSession;function M(A=null){if(!A)return y;switch(A){case"auto":return f;case"gpu":return f.filter($=>["webgpu","cuda","dml","webnn-gpu"].includes($))}if(f.includes(A))return[h[A]??A];throw new Error(`Unsupported device: "${A}". Should be one of: ${f.join(", ")}.`)}let I=null;async function z(A,$,P){I&&await I;const k=T.create(A,$);I??(I=k);const O=await k;return O.config=P,O}function E(A){return A instanceof m.Tensor}const C=m==null?void 0:m.env;C!=null&&C.wasm&&(!(typeof ServiceWorkerGlobalScope<"u"&&self instanceof ServiceWorkerGlobalScope)&&!C.wasm.wasmPaths&&(C.wasm.wasmPaths=`https://cdn.jsdelivr.net/npm/@huggingface/transformers@${o.env.version}/dist/`),C.wasm.proxy=!1),C!=null&&C.webgpu&&(C.webgpu.powerPreference="high-performance");function D(){var A;return(A=C==null?void 0:C.wasm)==null?void 0:A.proxy}o.env.backends.onnx=C},"./src/base/feature_extraction_utils.js":(e,r,t)=>{t.r(r),t.d(r,{FeatureExtractor:()=>u,validate_audio_inputs:()=>d});var i=t("./src/utils/constants.js"),o=t("./src/utils/generic.js"),n=t("./src/utils/hub.js");class u extends o.Callable{constructor(f){super(),this.config=f}static async from_pretrained(f,y){const m=await(0,n.getModelJSON)(f,i.FEATURE_EXTRACTOR_NAME,!0,y);return new this(m)}}function d(h,f){var y;if(!(h instanceof Float32Array||h instanceof Float64Array))throw new Error(`${f} expects input to be a Float32Array or a Float64Array, but got ${((y=h==null?void 0:h.constructor)==null?void 0:y.name)??typeof h} instead. If using the feature extractor directly, remember to use \`read_audio(url, sampling_rate)\` to obtain the raw audio data of the file/url.`)}},"./src/base/image_processors_utils.js":(e,r,t)=>{t.r(r),t.d(r,{ImageProcessor:()=>A,center_to_corners_format:()=>m,post_process_instance_segmentation:()=>D,post_process_object_detection:()=>g,post_process_panoptic_segmentation:()=>C,post_process_semantic_segmentation:()=>T});var i=t("./src/utils/generic.js"),o=t("./src/utils/tensor.js"),n=t("./src/utils/maths.js");t("./src/utils/image.js");var u=t("./src/utils/core.js"),d=t("./src/utils/hub.js"),h=t("./src/utils/constants.js");function f($,P,k=0,O=null){const R=$/P;let U=(0,n.bankers_round)(R)*P;return O!==null&&U>O&&(U=Math.floor(R)*P),UP&&N.push(ue)}else{let ue=(0,n.max)(q.data)[1];if(ue===K-1||(Q=(0,n.softmax)(q.data),Q[ue]we*oe[(ae+1)%2])),ge.boxes.push(Ae),ge.classes.push(ue),ge.scores.push(Q[ue])}}pe.push(ge)}return pe}function T($,P=null){const k=$.logits,O=k.dims[0];if(P!==null&&P.length!==O)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");const R=[];for(let U=0;Uoe[N]&&(oe[N]=q[N],ge[N]=Ce)}const le=new Array(se.dims[0]);for(let Ce=0;CeCe!==void 0);R.push({segmentation:re,labels:Se})}return R}function M($,P,k,O){const R=[],U=[],te=[];for(let se=0;se<$.dims[0];++se){const K=$[se],pe=P[se],re=(0,n.max)(K.data)[1];if(re===O)continue;const ge=(0,n.softmax)(K.data)[re];ge>k&&(R.push(pe),U.push(ge),te.push(re))}return[R,U,te]}function I($,P,k,O=.5,R=.8){const U=[];let te=0,se=0;const K=P[k].data;for(let re=0;re<$.length;++re)$[re]===k&&(U.push(re),++te),K[re]>=O&&++se;let pe=te>0&&se>0;return pe&&(pe=te/se>R),[pe,U]}function z($,P,k,O,R,U=null,te=null){const[se,K]=te??$[0].dims,pe=new o.Tensor("int32",new Int32Array(se*K),[se,K]),re=[];if(te!==null)for(let Ce=0;Ce<$.length;++Ce)$[Ce]=(0,o.interpolate)($[Ce],te,"bilinear",!1);const oe=new Int32Array($[0].data.length),ge=new Float32Array($[0].data.length);for(let Ce=0;Ce<$.length;++Ce){let q=P[Ce];const N=$[Ce].data;for(let Q=0;Qge[Q]&&(oe[Q]=Ce,ge[Q]=N[Q])}let le=0;const Se=pe.data;for(let Ce=0;Ce200)throw new Error(`absolute aspect ratio must be smaller than 200, got ${Math.max($,P)/Math.min($,P)}`);let U=Math.round($/k)*k,te=Math.round(P/k)*k;if(U*te>R){const se=Math.sqrt($*P/R);U=Math.floor($/se/k)*k,te=Math.floor(P/se/k)*k}else if(U*teU?pe=Math.floor(U*K/R):U>R&&(K=Math.floor(R*pe/U)),await P.resize(pe,K,{resample:O}))}async crop_margin(P,k=200){const O=P.clone().grayscale(),R=(0,n.min)(O.data)[0],te=(0,n.max)(O.data)[0]-R;if(te===0)return P;const se=k/255;let K=O.width,pe=O.height,re=0,oe=0;const ge=O.data;for(let le=0;lethis.preprocess(U)));return{pixel_values:(0,o.stack)(O.map(U=>U.pixel_values),0),original_sizes:O.map(U=>U.original_size),reshaped_input_sizes:O.map(U=>U.reshaped_input_size)}}static async from_pretrained(P,k){const O=await(0,d.getModelJSON)(P,h.IMAGE_PROCESSOR_NAME,!0,k);return new this(O)}}},"./src/base/processing_utils.js":(e,r,t)=>{t.r(r),t.d(r,{Processor:()=>u});var i=t("./src/utils/constants.js"),o=t("./src/utils/generic.js"),n=t("./src/utils/hub.js");class u extends o.Callable{constructor(h,f){super(),this.config=h,this.components=f}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(h,f={}){if(!this.tokenizer)throw new Error("Unable to apply chat template without a tokenizer.");return this.tokenizer.apply_chat_template(h,{tokenize:!1,...f})}batch_decode(...h){if(!this.tokenizer)throw new Error("Unable to decode without a tokenizer.");return this.tokenizer.batch_decode(...h)}decode(...h){if(!this.tokenizer)throw new Error("Unable to decode without a tokenizer.");return this.tokenizer.decode(...h)}async _call(h,...f){for(const y of[this.image_processor,this.feature_extractor,this.tokenizer])if(y)return y(h,...f);throw new Error("No image processor, feature extractor, or tokenizer found.")}static async from_pretrained(h,f){const[y,m]=await Promise.all([this.uses_processor_config?(0,n.getModelJSON)(h,i.PROCESSOR_NAME,!0,f):{},Promise.all(this.classes.filter(g=>g in this).map(async g=>{const T=await this[g].from_pretrained(h,f);return[g.replace(/_class$/,""),T]})).then(Object.fromEntries)]);return new this(y,m)}}de(u,"classes",["image_processor_class","tokenizer_class","feature_extractor_class"]),de(u,"uses_processor_config",!1)},"./src/configs.js":(e,r,t)=>{t.r(r),t.d(r,{AutoConfig:()=>f,PretrainedConfig:()=>h,getKeyValueShapes:()=>d});var i=t("./src/utils/core.js"),o=t("./src/utils/hub.js");async function n(y,m){return await(0,o.getModelJSON)(y,"config.json",!0,m)}function u(y){const m={};let g={};switch(y.model_type){case"llava":case"paligemma":case"gemma3":case"florence2":case"llava_onevision":case"idefics3":case"ultravox":case"smolvlm":g=u(y.text_config);break;case"moondream1":g=u(y.phi_config);break;case"musicgen":g=u(y.decoder);break;case"multi_modality":g=u(y.language_config);break;case"gpt2":case"gptj":case"jais":case"codegen":case"gpt_bigcode":m.num_heads="n_head",m.num_layers="n_layer",m.hidden_size="n_embd";break;case"gpt_neox":case"stablelm":case"opt":case"falcon":m.num_heads="num_attention_heads",m.num_layers="num_hidden_layers",m.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":m.num_heads="num_key_value_heads",m.num_layers="num_hidden_layers",m.hidden_size="hidden_size",m.num_attention_heads="num_attention_heads";break;case"qwen3":case"gemma":case"gemma2":case"gemma3_text":case"glm":case"helium":m.num_heads="num_key_value_heads",m.num_layers="num_hidden_layers",m.dim_kv="head_dim";break;case"openelm":m.num_heads="num_kv_heads",m.num_layers="num_transformer_layers",m.dim_kv="head_dim";break;case"gpt_neo":case"donut-swin":m.num_heads="num_heads",m.num_layers="num_layers",m.hidden_size="hidden_size";break;case"bloom":m.num_heads="n_head",m.num_layers="n_layer",m.hidden_size="hidden_size";break;case"mpt":m.num_heads="n_heads",m.num_layers="n_layers",m.hidden_size="d_model";break;case"exaone":m.num_heads="num_key_value_heads",m.num_layers="num_layers",m.dim_kv="head_dim",m.num_attention_heads="num_attention_heads";break;case"t5":case"mt5":case"longt5":m.num_decoder_layers="num_decoder_layers",m.num_decoder_heads="num_heads",m.decoder_dim_kv="d_kv",m.num_encoder_layers="num_layers",m.num_encoder_heads="num_heads",m.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":m.num_decoder_layers="decoder_layers",m.num_decoder_heads="decoder_attention_heads",m.decoder_hidden_size="d_model",m.num_encoder_layers="encoder_layers",m.num_encoder_heads="encoder_attention_heads",m.encoder_hidden_size="d_model";break;case"speecht5":m.num_decoder_layers="decoder_layers",m.num_decoder_heads="decoder_attention_heads",m.decoder_hidden_size="hidden_size",m.num_encoder_layers="encoder_layers",m.num_encoder_heads="encoder_attention_heads",m.encoder_hidden_size="hidden_size";break;case"trocr":m.num_encoder_layers=m.num_decoder_layers="decoder_layers",m.num_encoder_heads=m.num_decoder_heads="decoder_attention_heads",m.encoder_hidden_size=m.decoder_hidden_size="d_model";break;case"musicgen_decoder":m.num_encoder_layers=m.num_decoder_layers="num_hidden_layers",m.num_encoder_heads=m.num_decoder_heads="num_attention_heads",m.encoder_hidden_size=m.decoder_hidden_size="hidden_size";break;case"moonshine":m.num_decoder_layers="decoder_num_hidden_layers",m.num_decoder_heads="decoder_num_key_value_heads",m.num_encoder_layers="encoder_num_hidden_layers",m.num_encoder_heads="encoder_num_key_value_heads",m.encoder_hidden_size=m.decoder_hidden_size="hidden_size";break;case"vision-encoder-decoder":const M=u(y.decoder),I="num_decoder_layers"in M,z=(0,i.pick)(y,["model_type","is_encoder_decoder"]);return I?(z.num_decoder_layers=M.num_decoder_layers,z.num_decoder_heads=M.num_decoder_heads,z.decoder_hidden_size=M.decoder_hidden_size,z.num_encoder_layers=M.num_encoder_layers,z.num_encoder_heads=M.num_encoder_heads,z.encoder_hidden_size=M.encoder_hidden_size):(z.num_layers=M.num_layers,z.num_heads=M.num_heads,z.hidden_size=M.hidden_size),z}const T={...g,...(0,i.pick)(y,["model_type","multi_query","is_encoder_decoder"])};for(const M in m)T[M]=y[m[M]];return T}function d(y,{prefix:m="past_key_values",batch_size:g=1}={}){const T={},M=y.normalized_config;if(M.is_encoder_decoder&&"num_encoder_heads"in M&&"num_decoder_heads"in M){const I=M.encoder_dim_kv??M.encoder_hidden_size/M.num_encoder_heads,z=M.decoder_dim_kv??M.decoder_hidden_size/M.num_decoder_heads,E=[g,M.num_encoder_heads,0,I],C=[g,M.num_decoder_heads,0,z];for(let D=0;D{var O,R;t.r(r),t.d(r,{apis:()=>z,env:()=>P});var i=t("?569f"),o=t("?3f59"),n=t("?154a");const u="3.5.1",d=typeof window<"u"&&typeof window.document<"u",h=typeof self<"u"&&((O=self.constructor)==null?void 0:O.name)==="DedicatedWorkerGlobalScope",f=typeof self<"u"&&"caches"in self,y=typeof navigator<"u"&&"gpu"in navigator,m=typeof navigator<"u"&&"ml"in navigator,g=typeof process<"u",T=g&&((R=process==null?void 0:process.release)==null?void 0:R.name)==="node",M=!k(i),I=!k(o),z=Object.freeze({IS_BROWSER_ENV:d,IS_WEBWORKER_ENV:h,IS_WEB_CACHE_AVAILABLE:f,IS_WEBGPU_AVAILABLE:y,IS_WEBNN_AVAILABLE:m,IS_PROCESS_AVAILABLE:g,IS_NODE_ENV:T,IS_FS_AVAILABLE:M,IS_PATH_AVAILABLE:I}),E=M&&I;let C="./";if(E){const U=Object(import.meta).url;U?C=o.dirname(o.dirname(n.fileURLToPath(U))):typeof __dirname<"u"&&(C=o.dirname(__dirname))}const D=E?o.join(C,"/.cache/"):null,A="/models/",$=E?o.join(C,A):A,P={version:u,backends:{onnx:{}},allowRemoteModels:!0,remoteHost:"https://huggingface.co/",remotePathTemplate:"{model}/resolve/{revision}/",allowLocalModels:!(d||h),localModelPath:$,useFS:M,useBrowserCache:f,useFSCache:M,cacheDir:D,useCustomCache:!1,customCache:null};function k(U){return Object.keys(U).length===0}},"./src/generation/configuration_utils.js":(e,r,t)=>{t.r(r),t.d(r,{GenerationConfig:()=>o});var i=t("./src/utils/core.js");class o{constructor(u){de(this,"max_length",20);de(this,"max_new_tokens",null);de(this,"min_length",0);de(this,"min_new_tokens",null);de(this,"early_stopping",!1);de(this,"max_time",null);de(this,"do_sample",!1);de(this,"num_beams",1);de(this,"num_beam_groups",1);de(this,"penalty_alpha",null);de(this,"use_cache",!0);de(this,"temperature",1);de(this,"top_k",50);de(this,"top_p",1);de(this,"typical_p",1);de(this,"epsilon_cutoff",0);de(this,"eta_cutoff",0);de(this,"diversity_penalty",0);de(this,"repetition_penalty",1);de(this,"encoder_repetition_penalty",1);de(this,"length_penalty",1);de(this,"no_repeat_ngram_size",0);de(this,"bad_words_ids",null);de(this,"force_words_ids",null);de(this,"renormalize_logits",!1);de(this,"constraints",null);de(this,"forced_bos_token_id",null);de(this,"forced_eos_token_id",null);de(this,"remove_invalid_values",!1);de(this,"exponential_decay_length_penalty",null);de(this,"suppress_tokens",null);de(this,"streamer",null);de(this,"begin_suppress_tokens",null);de(this,"forced_decoder_ids",null);de(this,"guidance_scale",null);de(this,"num_return_sequences",1);de(this,"output_attentions",!1);de(this,"output_hidden_states",!1);de(this,"output_scores",!1);de(this,"return_dict_in_generate",!1);de(this,"pad_token_id",null);de(this,"bos_token_id",null);de(this,"eos_token_id",null);de(this,"encoder_no_repeat_ngram_size",0);de(this,"decoder_start_token_id",null);de(this,"generation_kwargs",{});Object.assign(this,(0,i.pick)(u,Object.getOwnPropertyNames(this)))}}},"./src/generation/logits_process.js":(e,r,t)=>{t.r(r),t.d(r,{ClassifierFreeGuidanceLogitsProcessor:()=>E,ForcedBOSTokenLogitsProcessor:()=>h,ForcedEOSTokenLogitsProcessor:()=>f,LogitsProcessor:()=>n,LogitsProcessorList:()=>d,LogitsWarper:()=>u,MinLengthLogitsProcessor:()=>M,MinNewTokensLengthLogitsProcessor:()=>I,NoBadWordsLogitsProcessor:()=>z,NoRepeatNGramLogitsProcessor:()=>g,RepetitionPenaltyLogitsProcessor:()=>T,SuppressTokensAtBeginLogitsProcessor:()=>y,TemperatureLogitsWarper:()=>C,TopKLogitsWarper:()=>A,TopPLogitsWarper:()=>D,WhisperTimeStampLogitsProcessor:()=>m});var i=t("./src/utils/generic.js");t("./src/utils/tensor.js");var o=t("./src/utils/maths.js");class n extends i.Callable{_call(P,k){throw Error("`_call` should be implemented in a subclass")}}class u extends i.Callable{_call(P,k){throw Error("`_call` should be implemented in a subclass")}}class d extends i.Callable{constructor(){super(),this.processors=[]}push(P){this.processors.push(P)}extend(P){this.processors.push(...P)}_call(P,k){let O=k;for(const R of this.processors)O=R(P,O);return O}[Symbol.iterator](){return this.processors.values()}}class h extends n{constructor(P){super(),this.bos_token_id=P}_call(P,k){for(let O=0;O=1&&U[U.length-1]>=this.timestamp_begin,se=U.length<2||U[U.length-2]>=this.timestamp_begin;if(te&&(se?R.subarray(this.timestamp_begin).fill(-1/0):R.subarray(0,this.eos_token_id).fill(-1/0)),P[O].length===this.begin_index&&this.max_initial_timestamp_index!==null){const oe=this.timestamp_begin+this.max_initial_timestamp_index;R.subarray(oe+1).fill(-1/0)}const K=(0,o.log_softmax)(R),pe=Math.log(K.subarray(this.timestamp_begin).map(Math.exp).reduce((oe,ge)=>oe+ge)),re=(0,o.max)(K.subarray(0,this.timestamp_begin))[0];pe>re&&R.subarray(0,this.timestamp_begin).fill(-1/0)}return k}}class g extends n{constructor(P){super(),this.no_repeat_ngram_size=P}getNgrams(P){const k=P.length,O=[];for(let U=0;U1 to use the classifier free guidance processor, got guidance scale ${P}.`);this.guidance_scale=P}_call(P,k){if(k.dims[0]!==2*P.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 ${k.dims[0]} for the logits and ${P.length} for the input ids.`);const O=P.length,R=k.slice([0,O],null),U=k.slice([O,k.dims[0]],null);for(let te=0;te1)throw new Error(`\`top_p\` must be a float > 0 and < 1, but is ${P}`);if(!Number.isInteger(O)||O<1)throw new Error(`\`min_tokens_to_keep\` must be a positive integer, but is ${O}`);this.top_p=P,this.filter_value=k,this.min_tokens_to_keep=O}}class A extends u{constructor(P,{filter_value:k=-1/0,min_tokens_to_keep:O=1}={}){if(super(),!Number.isInteger(P)||P<0)throw new Error(`\`top_k\` must be a positive integer, but is ${P}`);this.top_k=Math.max(P,O),this.filter_value=k}}},"./src/generation/logits_sampler.js":(e,r,t)=>{t.r(r),t.d(r,{LogitsSampler:()=>u});var i=t("./src/utils/generic.js"),o=t("./src/utils/tensor.js"),n=t("./src/utils/maths.js");t("./src/generation/configuration_utils.js");class u extends i.Callable{constructor(m){super(),this.generation_config=m}async _call(m){return this.sample(m)}async sample(m){throw Error("sample should be implemented in subclasses.")}getLogits(m,g){let T=m.dims.at(-1),M=m.data;if(g===-1)M=M.slice(-T);else{let I=g*T;M=M.slice(I,I+T)}return M}randomSelect(m){let g=0;for(let M=0;M1)return new f(m);if(m.num_return_sequences>1)throw Error(`num_return_sequences has to be 1 when doing greedy search, but is ${m.num_return_sequences}.`);return new d(m)}}class d extends u{async sample(m){const g=(0,n.max)(m.data)[1];return[[BigInt(g),0]]}}class h extends u{async sample(m){let g=m.dims.at(-1);this.generation_config.top_k>0&&(g=Math.min(this.generation_config.top_k,g));const[T,M]=await(0,o.topk)(m,g),I=(0,n.softmax)(T.data);return Array.from({length:this.generation_config.num_beams},()=>{const z=this.randomSelect(I);return[M.data[z],Math.log(I[z])]})}}class f extends u{async sample(m){let g=m.dims.at(-1);this.generation_config.top_k>0&&(g=Math.min(this.generation_config.top_k,g));const[T,M]=await(0,o.topk)(m,g),I=(0,n.softmax)(T.data);return Array.from({length:this.generation_config.num_beams},(z,E)=>[M.data[E],Math.log(I[E])])}}},"./src/generation/stopping_criteria.js":(e,r,t)=>{t.r(r),t.d(r,{EosTokenCriteria:()=>d,InterruptableStoppingCriteria:()=>h,MaxLengthCriteria:()=>u,StoppingCriteria:()=>o,StoppingCriteriaList:()=>n});var i=t("./src/utils/generic.js");class o extends i.Callable{_call(y,m){throw Error("StoppingCriteria needs to be subclassed")}}class n extends i.Callable{constructor(){super(),this.criteria=[]}push(y){this.criteria.push(y)}extend(y){y instanceof n?y=y.criteria:y instanceof o&&(y=[y]),this.criteria.push(...y)}_call(y,m){const g=new Array(y.length).fill(!1);for(const T of this.criteria){const M=T(y,m);for(let I=0;Im.length>=this.max_length)}}class d extends o{constructor(y){super(),Array.isArray(y)||(y=[y]),this.eos_token_id=y}_call(y,m){return y.map(g=>{const T=g.at(-1);return this.eos_token_id.some(M=>T==M)})}}class h extends o{constructor(){super(),this.interrupted=!1}interrupt(){this.interrupted=!0}reset(){this.interrupted=!1}_call(y,m){return new Array(y.length).fill(this.interrupted)}}},"./src/generation/streamers.js":(e,r,t)=>{t.r(r),t.d(r,{BaseStreamer:()=>u,TextStreamer:()=>h,WhisperTextStreamer:()=>f});var i=t("./src/utils/core.js"),o=t("./src/tokenizers.js"),n=t("./src/env.js");class u{put(m){throw Error("Not implemented")}end(){throw Error("Not implemented")}}const d=n.apis.IS_PROCESS_AVAILABLE?y=>process.stdout.write(y):y=>console.log(y);class h extends u{constructor(m,{skip_prompt:g=!1,callback_function:T=null,token_callback_function:M=null,skip_special_tokens:I=!0,decode_kwargs:z={},...E}={}){super(),this.tokenizer=m,this.skip_prompt=g,this.callback_function=T??d,this.token_callback_function=M,this.decode_kwargs={skip_special_tokens:I,...z,...E},this.token_cache=[],this.print_len=0,this.next_tokens_are_prompt=!0}put(m){var z;if(m.length>1)throw Error("TextStreamer only supports batch size of 1");const g=this.next_tokens_are_prompt;if(g&&(this.next_tokens_are_prompt=!1,this.skip_prompt))return;const T=m[0];(z=this.token_callback_function)==null||z.call(this,T),this.token_cache=(0,i.mergeArrays)(this.token_cache,T);const M=this.tokenizer.decode(this.token_cache,this.decode_kwargs);let I;g||M.endsWith(` +`)?(I=M.slice(this.print_len),this.token_cache=[],this.print_len=0):M.length>0&&(0,o.is_chinese_char)(M.charCodeAt(M.length-1))?(I=M.slice(this.print_len),this.print_len+=I.length):(I=M.slice(this.print_len,M.lastIndexOf(" ")+1),this.print_len+=I.length),this.on_finalized_text(I,!1)}end(){let m;this.token_cache.length>0?(m=this.tokenizer.decode(this.token_cache,this.decode_kwargs).slice(this.print_len),this.token_cache=[],this.print_len=0):m="",this.next_tokens_are_prompt=!0,this.on_finalized_text(m,!0)}on_finalized_text(m,g){var T,M;m.length>0&&((T=this.callback_function)==null||T.call(this,m)),g&&this.callback_function===d&&n.apis.IS_PROCESS_AVAILABLE&&((M=this.callback_function)==null||M.call(this,` +`))}}class f extends h{constructor(m,{skip_prompt:g=!1,callback_function:T=null,token_callback_function:M=null,on_chunk_start:I=null,on_chunk_end:z=null,on_finalize:E=null,time_precision:C=.02,skip_special_tokens:D=!0,decode_kwargs:A={}}={}){super(m,{skip_prompt:g,skip_special_tokens:D,callback_function:T,token_callback_function:M,decode_kwargs:A}),this.timestamp_begin=m.timestamp_begin,this.on_chunk_start=I,this.on_chunk_end=z,this.on_finalize=E,this.time_precision=C,this.waiting_for_timestamp=!1}put(m){var T,M;if(m.length>1)throw Error("WhisperTextStreamer only supports batch size of 1");const g=m[0];if(g.length===1){const I=Number(g[0])-this.timestamp_begin;if(I>=0){const z=I*this.time_precision;this.waiting_for_timestamp?(T=this.on_chunk_end)==null||T.call(this,z):(M=this.on_chunk_start)==null||M.call(this,z),this.waiting_for_timestamp=!this.waiting_for_timestamp,m=[[]]}}return super.put(m)}end(){var m;super.end(),(m=this.on_finalize)==null||m.call(this)}}},"./src/models.js":(e,r,t)=>{t.r(r),t.d(r,{ASTForAudioClassification:()=>Ka,ASTModel:()=>Ed,ASTPreTrainedModel:()=>Ga,AlbertForMaskedLM:()=>ie,AlbertForQuestionAnswering:()=>J,AlbertForSequenceClassification:()=>L,AlbertModel:()=>Te,AlbertPreTrainedModel:()=>qs,AutoModel:()=>Qu,AutoModelForAudioClassification:()=>yh,AutoModelForAudioFrameClassification:()=>bh,AutoModelForAudioTextToText:()=>kh,AutoModelForCTC:()=>gh,AutoModelForCausalLM:()=>ah,AutoModelForDepthEstimation:()=>Th,AutoModelForDocumentQuestionAnswering:()=>Mh,AutoModelForImageClassification:()=>uh,AutoModelForImageFeatureExtraction:()=>Ph,AutoModelForImageMatting:()=>vh,AutoModelForImageSegmentation:()=>dh,AutoModelForImageTextToText:()=>Sh,AutoModelForImageToImage:()=>xh,AutoModelForMaskGeneration:()=>_h,AutoModelForMaskedLM:()=>oh,AutoModelForNormalEstimation:()=>Ch,AutoModelForObjectDetection:()=>fh,AutoModelForPoseEstimation:()=>Eh,AutoModelForQuestionAnswering:()=>lh,AutoModelForSemanticSegmentation:()=>ph,AutoModelForSeq2SeqLM:()=>rh,AutoModelForSequenceClassification:()=>eh,AutoModelForSpeechSeq2Seq:()=>sh,AutoModelForTextToSpectrogram:()=>nh,AutoModelForTextToWaveform:()=>ih,AutoModelForTokenClassification:()=>th,AutoModelForUniversalSegmentation:()=>hh,AutoModelForVision2Seq:()=>ch,AutoModelForXVector:()=>wh,AutoModelForZeroShotObjectDetection:()=>mh,BartForConditionalGeneration:()=>sr,BartForSequenceClassification:()=>Pr,BartModel:()=>fr,BartPretrainedModel:()=>Ht,BaseModelOutput:()=>Fe,BeitForImageClassification:()=>el,BeitModel:()=>Zo,BeitPreTrainedModel:()=>Ji,BertForMaskedLM:()=>Ue,BertForQuestionAnswering:()=>at,BertForSequenceClassification:()=>Ge,BertForTokenClassification:()=>Je,BertModel:()=>Re,BertPreTrainedModel:()=>Ve,BlenderbotForConditionalGeneration:()=>zr,BlenderbotModel:()=>Ir,BlenderbotPreTrainedModel:()=>ur,BlenderbotSmallForConditionalGeneration:()=>Br,BlenderbotSmallModel:()=>ss,BlenderbotSmallPreTrainedModel:()=>Yr,BloomForCausalLM:()=>Oi,BloomModel:()=>Fo,BloomPreTrainedModel:()=>Ai,CLIPModel:()=>to,CLIPPreTrainedModel:()=>sn,CLIPSegForImageSegmentation:()=>lo,CLIPSegModel:()=>Vd,CLIPSegPreTrainedModel:()=>oo,CLIPTextModel:()=>Ld,CLIPTextModelWithProjection:()=>wn,CLIPVisionModel:()=>ro,CLIPVisionModelWithProjection:()=>so,CamembertForMaskedLM:()=>mt,CamembertForQuestionAnswering:()=>Fr,CamembertForSequenceClassification:()=>tr,CamembertForTokenClassification:()=>rr,CamembertModel:()=>Nt,CamembertPreTrainedModel:()=>wt,CausalLMOutput:()=>on,CausalLMOutputWithPast:()=>bf,ChineseCLIPModel:()=>ao,ChineseCLIPPreTrainedModel:()=>xi,ClapAudioModelWithProjection:()=>Zc,ClapModel:()=>Jc,ClapPreTrainedModel:()=>Jn,ClapTextModelWithProjection:()=>Yc,CodeGenForCausalLM:()=>Jd,CodeGenModel:()=>Xd,CodeGenPreTrainedModel:()=>_o,CohereForCausalLM:()=>cp,CohereModel:()=>lp,CoherePreTrainedModel:()=>Eo,ConvBertForMaskedLM:()=>as,ConvBertForQuestionAnswering:()=>G,ConvBertForSequenceClassification:()=>W,ConvBertForTokenClassification:()=>ye,ConvBertModel:()=>ht,ConvBertPreTrainedModel:()=>er,ConvNextForImageClassification:()=>Hl,ConvNextModel:()=>Kl,ConvNextPreTrainedModel:()=>pa,ConvNextV2ForImageClassification:()=>Ql,ConvNextV2Model:()=>ql,ConvNextV2PreTrainedModel:()=>ha,DFineForObjectDetection:()=>pl,DFineModel:()=>dl,DFinePreTrainedModel:()=>sa,DPTForDepthEstimation:()=>Sl,DPTModel:()=>Pl,DPTPreTrainedModel:()=>ca,DacDecoderModel:()=>Nu,DacDecoderOutput:()=>zu,DacEncoderModel:()=>Ru,DacEncoderOutput:()=>Lu,DacModel:()=>Bu,DacPreTrainedModel:()=>ni,DebertaForMaskedLM:()=>xs,DebertaForQuestionAnswering:()=>en,DebertaForSequenceClassification:()=>Ws,DebertaForTokenClassification:()=>Ts,DebertaModel:()=>Kr,DebertaPreTrainedModel:()=>pr,DebertaV2ForMaskedLM:()=>os,DebertaV2ForQuestionAnswering:()=>Ks,DebertaV2ForSequenceClassification:()=>Us,DebertaV2ForTokenClassification:()=>Gs,DebertaV2Model:()=>tn,DebertaV2PreTrainedModel:()=>rs,DecisionTransformerModel:()=>bu,DecisionTransformerPreTrainedModel:()=>wu,DeiTForImageClassification:()=>gl,DeiTModel:()=>_l,DeiTPreTrainedModel:()=>ia,DepthAnythingForDepthEstimation:()=>$l,DepthAnythingPreTrainedModel:()=>kl,DepthProForDepthEstimation:()=>Dl,DepthProPreTrainedModel:()=>Fl,DetrForObjectDetection:()=>rl,DetrForSegmentation:()=>Gn,DetrModel:()=>tl,DetrObjectDetectionOutput:()=>Yi,DetrPreTrainedModel:()=>Un,DetrSegmentationOutput:()=>sl,Dinov2ForImageClassification:()=>Jl,Dinov2Model:()=>Xl,Dinov2PreTrainedModel:()=>fa,Dinov2WithRegistersForImageClassification:()=>Zl,Dinov2WithRegistersModel:()=>Yl,Dinov2WithRegistersPreTrainedModel:()=>ma,DistilBertForMaskedLM:()=>qe,DistilBertForQuestionAnswering:()=>Er,DistilBertForSequenceClassification:()=>Cs,DistilBertForTokenClassification:()=>Hs,DistilBertModel:()=>Jt,DistilBertPreTrainedModel:()=>Wr,DonutSwinModel:()=>Gl,DonutSwinPreTrainedModel:()=>Ul,EfficientNetForImageClassification:()=>au,EfficientNetModel:()=>iu,EfficientNetPreTrainedModel:()=>Ca,ElectraForMaskedLM:()=>Xe,ElectraForQuestionAnswering:()=>Qt,ElectraForSequenceClassification:()=>Le,ElectraForTokenClassification:()=>Ft,ElectraModel:()=>$e,ElectraPreTrainedModel:()=>ne,EsmForMaskedLM:()=>cr,EsmForSequenceClassification:()=>ls,EsmForTokenClassification:()=>Es,EsmModel:()=>bt,EsmPreTrainedModel:()=>ut,ExaoneForCausalLM:()=>bo,ExaoneModel:()=>wo,ExaonePreTrainedModel:()=>Si,FalconForCausalLM:()=>Xc,FalconModel:()=>Qc,FalconPreTrainedModel:()=>va,FastViTForImageClassification:()=>Ui,FastViTModel:()=>Go,FastViTPreTrainedModel:()=>Wn,Florence2ForConditionalGeneration:()=>Od,Florence2PreTrainedModel:()=>Ad,GLPNForDepthEstimation:()=>Wl,GLPNModel:()=>Vl,GLPNPreTrainedModel:()=>da,GPT2LMHeadModel:()=>Ud,GPT2Model:()=>Wd,GPT2PreTrainedModel:()=>co,GPTBigCodeForCausalLM:()=>Qd,GPTBigCodeModel:()=>qd,GPTBigCodePreTrainedModel:()=>mo,GPTJForCausalLM:()=>Pe,GPTJModel:()=>ps,GPTJPreTrainedModel:()=>fo,GPTNeoForCausalLM:()=>bn,GPTNeoModel:()=>Bn,GPTNeoPreTrainedModel:()=>po,GPTNeoXForCausalLM:()=>Hd,GPTNeoXModel:()=>ho,GPTNeoXPreTrainedModel:()=>Ci,Gemma2ForCausalLM:()=>hp,Gemma2Model:()=>pp,Gemma2PreTrainedModel:()=>So,Gemma3ForCausalLM:()=>mp,Gemma3Model:()=>fp,Gemma3PreTrainedModel:()=>ko,GemmaForCausalLM:()=>dp,GemmaModel:()=>up,GemmaPreTrainedModel:()=>Po,GlmForCausalLM:()=>Mn,GlmModel:()=>Pi,GlmPreTrainedModel:()=>Ei,GraniteForCausalLM:()=>op,GraniteModel:()=>ap,GranitePreTrainedModel:()=>Co,GroundingDinoForObjectDetection:()=>tc,GroundingDinoPreTrainedModel:()=>ec,GroupViTModel:()=>xn,GroupViTPreTrainedModel:()=>Wi,HeliumForCausalLM:()=>tp,HeliumModel:()=>ep,HeliumPreTrainedModel:()=>yo,HieraForImageClassification:()=>wl,HieraModel:()=>yl,HieraPreTrainedModel:()=>aa,HubertForCTC:()=>Ic,HubertForSequenceClassification:()=>Ac,HubertModel:()=>$c,HubertPreTrainedModel:()=>xp,IJepaForImageClassification:()=>Bo,IJepaModel:()=>zo,IJepaPreTrainedModel:()=>Li,Idefics3ForConditionalGeneration:()=>vi,Idefics3PreTrainedModel:()=>jd,ImageMattingOutput:()=>Ih,JAISLMHeadModel:()=>Kd,JAISModel:()=>Gd,JAISPreTrainedModel:()=>uo,JinaCLIPModel:()=>Rd,JinaCLIPPreTrainedModel:()=>Ti,JinaCLIPTextModel:()=>Js,JinaCLIPVisionModel:()=>Nd,LiteWhisperForConditionalGeneration:()=>Qa,LlamaForCausalLM:()=>Zd,LlamaModel:()=>Yd,LlamaPreTrainedModel:()=>go,LlavaForConditionalGeneration:()=>Mi,LlavaOnevisionForConditionalGeneration:()=>$d,LlavaPreTrainedModel:()=>kd,LongT5ForConditionalGeneration:()=>it,LongT5Model:()=>ct,LongT5PreTrainedModel:()=>tt,M2M100ForConditionalGeneration:()=>dc,M2M100Model:()=>uc,M2M100PreTrainedModel:()=>ya,MBartForCausalLM:()=>Qs,MBartForConditionalGeneration:()=>ws,MBartForSequenceClassification:()=>Ur,MBartModel:()=>qr,MBartPreTrainedModel:()=>$r,MPNetForMaskedLM:()=>gn,MPNetForQuestionAnswering:()=>Hr,MPNetForSequenceClassification:()=>Fn,MPNetForTokenClassification:()=>Dn,MPNetModel:()=>_n,MPNetPreTrainedModel:()=>gs,MT5ForConditionalGeneration:()=>qt,MT5Model:()=>At,MT5PreTrainedModel:()=>dt,MarianMTModel:()=>cc,MarianModel:()=>lc,MarianPreTrainedModel:()=>ga,MaskFormerForInstanceSegmentation:()=>Nl,MaskFormerModel:()=>Rl,MaskFormerPreTrainedModel:()=>ua,MaskedLMOutput:()=>Or,Metric3DForDepthEstimation:()=>Ll,Metric3DPreTrainedModel:()=>jl,Metric3Dv2ForDepthEstimation:()=>Bl,Metric3Dv2PreTrainedModel:()=>zl,MgpstrForSceneTextRecognition:()=>Cu,MgpstrModelOutput:()=>xu,MgpstrPreTrainedModel:()=>Tu,MimiDecoderModel:()=>ju,MimiDecoderOutput:()=>Ou,MimiEncoderModel:()=>Du,MimiEncoderOutput:()=>Au,MimiModel:()=>Fu,MimiPreTrainedModel:()=>si,MistralForCausalLM:()=>Kc,MistralModel:()=>Gc,MistralPreTrainedModel:()=>ba,MobileBertForMaskedLM:()=>Ss,MobileBertForQuestionAnswering:()=>$s,MobileBertForSequenceClassification:()=>ks,MobileBertModel:()=>Ps,MobileBertPreTrainedModel:()=>Xr,MobileLLMForCausalLM:()=>rp,MobileLLMModel:()=>Mo,MobileLLMPreTrainedModel:()=>ki,MobileNetV1ForImageClassification:()=>lu,MobileNetV1ForSemanticSegmentation:()=>cu,MobileNetV1Model:()=>ou,MobileNetV1PreTrainedModel:()=>Zn,MobileNetV2ForImageClassification:()=>du,MobileNetV2ForSemanticSegmentation:()=>pu,MobileNetV2Model:()=>uu,MobileNetV2PreTrainedModel:()=>ei,MobileNetV3ForImageClassification:()=>fu,MobileNetV3ForSemanticSegmentation:()=>mu,MobileNetV3Model:()=>hu,MobileNetV3PreTrainedModel:()=>ti,MobileNetV4ForImageClassification:()=>gu,MobileNetV4ForSemanticSegmentation:()=>yu,MobileNetV4Model:()=>_u,MobileNetV4PreTrainedModel:()=>ri,MobileViTForImageClassification:()=>qo,MobileViTModel:()=>Ki,MobileViTPreTrainedModel:()=>Gi,MobileViTV2ForImageClassification:()=>Me,MobileViTV2Model:()=>qi,MobileViTV2PreTrainedModel:()=>Hi,ModelOutput:()=>Oe,ModernBertForMaskedLM:()=>st,ModernBertForSequenceClassification:()=>ft,ModernBertForTokenClassification:()=>yt,ModernBertModel:()=>Ke,ModernBertPreTrainedModel:()=>ot,Moondream1ForConditionalGeneration:()=>Id,MoonshineForConditionalGeneration:()=>Sd,MoonshineModel:()=>uf,MoonshinePreTrainedModel:()=>Xa,MptForCausalLM:()=>Di,MptModel:()=>Fi,MptPreTrainedModel:()=>Rn,MultiModalityCausalLM:()=>vu,MultiModalityPreTrainedModel:()=>Mu,MusicgenForCausalLM:()=>Pp,MusicgenForConditionalGeneration:()=>Pa,MusicgenModel:()=>Ep,MusicgenPreTrainedModel:()=>Ea,NomicBertModel:()=>vt,NomicBertPreTrainedModel:()=>St,OPTForCausalLM:()=>Do,OPTModel:()=>ji,OPTPreTrainedModel:()=>Nn,Olmo2ForCausalLM:()=>ip,Olmo2Model:()=>np,Olmo2PreTrainedModel:()=>To,OlmoForCausalLM:()=>sp,OlmoModel:()=>xo,OlmoPreTrainedModel:()=>vo,OpenELMForCausalLM:()=>gp,OpenELMModel:()=>_p,OpenELMPreTrainedModel:()=>$o,OwlViTForObjectDetection:()=>Xo,OwlViTModel:()=>Qo,OwlViTPreTrainedModel:()=>Qi,Owlv2ForObjectDetection:()=>Yo,Owlv2Model:()=>Jo,Owlv2PreTrainedModel:()=>Xi,PaliGemmaForConditionalGeneration:()=>Dd,PaliGemmaPreTrainedModel:()=>Fd,PatchTSMixerForPrediction:()=>ku,PatchTSMixerModel:()=>Su,PatchTSMixerPreTrainedModel:()=>ka,PatchTSTForPrediction:()=>Pu,PatchTSTModel:()=>Eu,PatchTSTPreTrainedModel:()=>Sa,Phi3ForCausalLM:()=>nn,Phi3Model:()=>Zr,Phi3PreTrainedModel:()=>vn,Phi3VForCausalLM:()=>eo,Phi3VPreTrainedModel:()=>Za,PhiForCausalLM:()=>Ii,PhiModel:()=>Oo,PhiPreTrainedModel:()=>$i,PreTrainedModel:()=>X,PretrainedMixin:()=>Ut,PvtForImageClassification:()=>Wo,PvtModel:()=>Vo,PvtPreTrainedModel:()=>zi,PyAnnoteForAudioFrameClassification:()=>gc,PyAnnoteModel:()=>_c,PyAnnotePreTrainedModel:()=>wa,QuestionAnsweringModelOutput:()=>Nr,Qwen2ForCausalLM:()=>wp,Qwen2Model:()=>yp,Qwen2PreTrainedModel:()=>Io,Qwen2VLForConditionalGeneration:()=>Y,Qwen2VLPreTrainedModel:()=>vp,Qwen3ForCausalLM:()=>Mp,Qwen3Model:()=>bp,Qwen3PreTrainedModel:()=>Ao,RFDetrForObjectDetection:()=>cl,RFDetrModel:()=>ll,RFDetrObjectDetectionOutput:()=>ul,RFDetrPreTrainedModel:()=>ra,RTDetrForObjectDetection:()=>ea,RTDetrModel:()=>nl,RTDetrObjectDetectionOutput:()=>Tn,RTDetrPreTrainedModel:()=>Zi,RTDetrV2ForObjectDetection:()=>al,RTDetrV2Model:()=>il,RTDetrV2ObjectDetectionOutput:()=>ol,RTDetrV2PreTrainedModel:()=>ta,ResNetForImageClassification:()=>Ml,ResNetModel:()=>bl,ResNetPreTrainedModel:()=>oa,RoFormerForMaskedLM:()=>Vt,RoFormerForQuestionAnswering:()=>Cr,RoFormerForSequenceClassification:()=>he,RoFormerForTokenClassification:()=>ce,RoFormerModel:()=>xt,RoFormerPreTrainedModel:()=>He,RobertaForMaskedLM:()=>yr,RobertaForQuestionAnswering:()=>Rr,RobertaForSequenceClassification:()=>Mr,RobertaForTokenClassification:()=>vr,RobertaModel:()=>Ar,RobertaPreTrainedModel:()=>Dr,SamImageSegmentationOutput:()=>oc,SamModel:()=>ac,SamPreTrainedModel:()=>ic,SapiensForDepthEstimation:()=>Al,SapiensForNormalEstimation:()=>Ol,SapiensForSemanticSegmentation:()=>Il,SapiensPreTrainedModel:()=>Hn,SegformerForImageClassification:()=>tu,SegformerForSemanticSegmentation:()=>ru,SegformerModel:()=>Cp,SegformerPreTrainedModel:()=>Yn,Seq2SeqLMOutput:()=>wf,SequenceClassifierOutput:()=>Dt,SiglipModel:()=>io,SiglipPreTrainedModel:()=>no,SiglipTextModel:()=>zd,SiglipVisionModel:()=>Bd,SmolVLMForConditionalGeneration:()=>Ya,SnacDecoderModel:()=>Uu,SnacEncoderModel:()=>Wu,SnacModel:()=>Vu,SnacPreTrainedModel:()=>ii,SpeechT5ForSpeechToText:()=>Rc,SpeechT5ForTextToSpeech:()=>Nc,SpeechT5HifiGan:()=>Vc,SpeechT5Model:()=>Tp,SpeechT5PreTrainedModel:()=>Xn,SqueezeBertForMaskedLM:()=>cs,SqueezeBertForQuestionAnswering:()=>yn,SqueezeBertForSequenceClassification:()=>Jr,SqueezeBertModel:()=>rn,SqueezeBertPreTrainedModel:()=>ys,StableLmForCausalLM:()=>nu,StableLmModel:()=>su,StableLmPreTrainedModel:()=>Ta,Starcoder2ForCausalLM:()=>qc,Starcoder2Model:()=>Hc,Starcoder2PreTrainedModel:()=>Ma,StyleTextToSpeech2Model:()=>Bc,StyleTextToSpeech2PreTrainedModel:()=>zc,Swin2SRForImageSuperResolution:()=>El,Swin2SRModel:()=>Cl,Swin2SRPreTrainedModel:()=>la,SwinForImageClassification:()=>xl,SwinForSemanticSegmentation:()=>Tl,SwinModel:()=>vl,SwinPreTrainedModel:()=>Kn,T5ForConditionalGeneration:()=>ze,T5Model:()=>be,T5PreTrainedModel:()=>ve,TableTransformerForObjectDetection:()=>fl,TableTransformerModel:()=>hl,TableTransformerObjectDetectionOutput:()=>ml,TableTransformerPreTrainedModel:()=>na,TokenClassifierOutput:()=>Sr,TrOCRForCausalLM:()=>Uc,TrOCRPreTrainedModel:()=>Wc,UltravoxModel:()=>Iu,UltravoxPreTrainedModel:()=>$u,UniSpeechForCTC:()=>Mc,UniSpeechForSequenceClassification:()=>vc,UniSpeechModel:()=>bc,UniSpeechPreTrainedModel:()=>qn,UniSpeechSatForAudioFrameClassification:()=>Ec,UniSpeechSatForCTC:()=>Tc,UniSpeechSatForSequenceClassification:()=>Cc,UniSpeechSatModel:()=>xc,UniSpeechSatPreTrainedModel:()=>Cn,ViTForImageClassification:()=>Lo,ViTMAEModel:()=>Bi,ViTMAEPreTrainedModel:()=>Uo,ViTMSNForImageClassification:()=>Vi,ViTMSNModel:()=>Ni,ViTMSNPreTrainedModel:()=>Ri,ViTModel:()=>jo,ViTPreTrainedModel:()=>Vn,VisionEncoderDecoderModel:()=>Ja,VitMatteForImageMatting:()=>Ho,VitMattePreTrainedModel:()=>Ko,VitPoseForPoseEstimation:()=>No,VitPosePreTrainedModel:()=>Ro,VitsModel:()=>xa,VitsModelOutput:()=>Ah,VitsPreTrainedModel:()=>eu,Wav2Vec2BertForCTC:()=>Sc,Wav2Vec2BertForSequenceClassification:()=>kc,Wav2Vec2BertModel:()=>Pc,Wav2Vec2BertPreTrainedModel:()=>Qn,Wav2Vec2ForAudioFrameClassification:()=>mc,Wav2Vec2ForCTC:()=>hc,Wav2Vec2ForSequenceClassification:()=>fc,Wav2Vec2Model:()=>pc,Wav2Vec2PreTrainedModel:()=>Is,WavLMForAudioFrameClassification:()=>Lc,WavLMForCTC:()=>Fc,WavLMForSequenceClassification:()=>Dc,WavLMForXVector:()=>jc,WavLMModel:()=>Oc,WavLMPreTrainedModel:()=>an,WeSpeakerResNetModel:()=>wc,WeSpeakerResNetPreTrainedModel:()=>yc,WhisperForConditionalGeneration:()=>qa,WhisperModel:()=>Pd,WhisperPreTrainedModel:()=>Ha,XLMForQuestionAnswering:()=>ds,XLMForSequenceClassification:()=>jn,XLMForTokenClassification:()=>Xs,XLMModel:()=>bi,XLMPreTrainedModel:()=>us,XLMRobertaForMaskedLM:()=>Ln,XLMRobertaForQuestionAnswering:()=>zn,XLMRobertaForSequenceClassification:()=>Td,XLMRobertaForTokenClassification:()=>Cd,XLMRobertaModel:()=>Ua,XLMRobertaPreTrainedModel:()=>Tt,XLMWithLMHeadModel:()=>xd,XVectorOutput:()=>$h,YolosForObjectDetection:()=>sc,YolosModel:()=>rc,YolosObjectDetectionOutput:()=>nc,YolosPreTrainedModel:()=>_a});var i=t("./src/configs.js"),o=t("./src/backends/onnx.js"),n=t("./src/utils/dtypes.js"),u=t("./src/utils/generic.js"),d=t("./src/utils/core.js"),h=t("./src/utils/hub.js"),f=t("./src/utils/constants.js"),y=t("./src/generation/logits_process.js"),m=t("./src/generation/configuration_utils.js"),g=t("./src/utils/tensor.js"),T=t("./src/utils/image.js"),M=t("./src/utils/maths.js"),I=t("./src/generation/stopping_criteria.js"),z=t("./src/generation/logits_sampler.js"),E=t("./src/env.js"),C=t("./src/models/whisper/generation_whisper.js"),D=t("./src/models/whisper/common_whisper.js");const A={EncoderOnly:0,EncoderDecoder:1,Seq2Seq:2,Vision2Seq:3,DecoderOnly:4,MaskGeneration:5,ImageTextToText:6,Musicgen:7,MultiModality:8,Phi3V:9,AudioTextToText:10,AutoEncoder:11},$=new Map,P=new Map,k=new Map;async function O(S,F,V){var kr;let _e=((kr=V.config)==null?void 0:kr["transformers.js_config"])??{},Ie=V.device??_e.device;Ie&&typeof Ie!="string"&&(Ie.hasOwnProperty(F)?Ie=Ie[F]:(console.warn(`device not specified for "${F}". Using the default device.`),Ie=null));const ke=Ie??(E.apis.IS_NODE_ENV?"cpu":"wasm"),Be=(0,o.deviceToExecutionProviders)(ke),Qe=_e.device_config??{};Qe.hasOwnProperty(ke)&&(_e={..._e,...Qe[ke]});let rt=V.dtype??_e.dtype;if(typeof rt!="string"&&(rt&&rt.hasOwnProperty(F)?rt=rt[F]:(rt=n.DEFAULT_DEVICE_DTYPE_MAPPING[ke]??n.DATA_TYPES.fp32,console.warn(`dtype not specified for "${F}". Using the default dtype (${rt}) for this device (${ke}).`))),rt===n.DATA_TYPES.auto){let Rt=_e.dtype;typeof Rt!="string"&&(Rt=Rt==null?void 0:Rt[F]),Rt&&Rt!==n.DATA_TYPES.auto&&n.DATA_TYPES.hasOwnProperty(Rt)?rt=Rt:rt=n.DEFAULT_DEVICE_DTYPE_MAPPING[ke]??n.DATA_TYPES.fp32}const lt=rt;if(n.DEFAULT_DTYPE_SUFFIX_MAPPING.hasOwnProperty(lt)){if(lt===n.DATA_TYPES.fp16&&ke==="webgpu"&&!await(0,n.isWebGpuFp16Supported)())throw new Error(`The device (${ke}) does not support fp16.`)}else throw new Error(`Invalid dtype: ${lt}. Should be one of: ${Object.keys(n.DATA_TYPES).join(", ")}`);const Pt=_e.kv_cache_dtype,It=Pt?typeof Pt=="string"?Pt:Pt[lt]??"float32":void 0;if(It&&!["float32","float16"].includes(It))throw new Error(`Invalid kv_cache_dtype: ${It}. Should be one of: float32, float16`);const Ct={dtype:lt,kv_cache_dtype:It,device:ke},Bt=n.DEFAULT_DTYPE_SUFFIX_MAPPING[lt],kt=`${F}${Bt}.onnx`,jt=`${V.subfolder??""}/${kt}`,_t={...V.session_options};_t.executionProviders??(_t.executionProviders=Be);const Wt=_e.free_dimension_overrides;Wt?_t.freeDimensionOverrides??(_t.freeDimensionOverrides=Wt):ke.startsWith("webnn")&&!_t.freeDimensionOverrides&&console.warn(`WebNN does not currently support dynamic shapes and requires 'free_dimension_overrides' to be set in config.json, preferably as a field within config["transformers.js_config"]["device_config"]["${ke}"]. When 'free_dimension_overrides' is not set, you may experience significant performance degradation.`);const nr=E.apis.IS_NODE_ENV&&E.env.useFSCache,ar=(0,h.getModelFile)(S,jt,!0,V,nr),mr=V.use_external_data_format??_e.use_external_data_format;let hr=[];if(mr){let Rt;typeof mr=="object"?mr.hasOwnProperty(kt)?Rt=mr[kt]:mr.hasOwnProperty(F)?Rt=mr[F]:Rt=!1:Rt=mr;const xr=+Rt;if(xr>h.MAX_EXTERNAL_DATA_CHUNKS)throw new Error(`The number of external data chunks (${xr}) exceeds the maximum allowed value (${h.MAX_EXTERNAL_DATA_CHUNKS}).`);for(let es=0;es{const cn=await(0,h.getModelFile)(S,Gr,!0,V,nr);ns(cn instanceof Uint8Array?{path:ln,data:cn}:ln)}))}}else _t.externalData!==void 0&&(hr=_t.externalData.map(async Rt=>{if(typeof Rt.data=="string"){const xr=await(0,h.getModelFile)(S,Rt.data,!0,V);return{...Rt,data:xr}}return Rt}));if(hr.length>0){const Rt=await Promise.all(hr);E.apis.IS_NODE_ENV||(_t.externalData=Rt)}if(ke==="webgpu"){const Rt=(0,i.getKeyValueShapes)(V.config,{prefix:"present"});if(Object.keys(Rt).length>0&&!(0,o.isONNXProxy)()){const xr={};for(const es in Rt)xr[es]="gpu-buffer";_t.preferredOutputLocation=xr}}return{buffer_or_path:await ar,session_options:_t,session_config:Ct}}async function R(S,F,V){return Object.fromEntries(await Promise.all(Object.keys(F).map(async _e=>{const{buffer_or_path:Ie,session_options:ke,session_config:Be}=await O(S,F[_e],V),Qe=await(0,o.createInferenceSession)(Ie,ke,Be);return[_e,Qe]})))}async function U(S,F,V){return Object.fromEntries(await Promise.all(Object.keys(F).map(async _e=>{const Ie=await(0,h.getModelJSON)(S,F[_e],!1,V);return[_e,Ie]})))}function te(S,F){const V=Object.create(null),_e=[];for(const Be of S.inputNames){const Qe=F[Be];if(!(Qe instanceof g.Tensor)){_e.push(Be);continue}V[Be]=(0,o.isONNXProxy)()?Qe.clone():Qe}if(_e.length>0)throw new Error(`An error occurred during model execution: "Missing the following inputs: ${_e.join(", ")}.`);const Ie=Object.keys(F).length,ke=S.inputNames.length;if(Ie>ke){let Be=Object.keys(F).filter(Qe=>!S.inputNames.includes(Qe));console.warn(`WARNING: Too many inputs were provided (${Ie} > ${ke}). The following inputs will be ignored: "${Be.join(", ")}".`)}return V}let se=Promise.resolve();async function K(S,F){const V=te(S,F);try{const _e=Object.fromEntries(Object.entries(V).map(([Be,Qe])=>[Be,Qe.ort_tensor])),Ie=()=>S.run(_e),ke=await(E.apis.IS_BROWSER_ENV||E.apis.IS_WEBWORKER_ENV?se=se.then(Ie):Ie());return pe(ke)}catch(_e){const Ie=Object.fromEntries(Object.entries(V).map(([ke,Be])=>{const Qe={type:Be.type,dims:Be.dims,location:Be.location};return Qe.location!=="gpu-buffer"&&(Qe.data=Be.data),[ke,Qe]}));throw console.error(`An error occurred during model execution: "${_e}".`),console.error("Inputs given to model:",Ie),_e}}function pe(S){for(let F in S)(0,o.isONNXTensor)(S[F])?S[F]=new g.Tensor(S[F]):typeof S[F]=="object"&&pe(S[F]);return S}function re(S){if(S instanceof g.Tensor)return S;if(S.length===0)throw Error("items must be non-empty");if(Array.isArray(S[0])){if(S.some(F=>F.length!==S[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 g.Tensor("int64",BigInt64Array.from(S.flat().map(F=>BigInt(F))),[S.length,S[0].length])}else return new g.Tensor("int64",BigInt64Array.from(S.map(F=>BigInt(F))),[1,S.length])}function oe(S){return new g.Tensor("bool",[S],[1])}async function ge(S,F){let{encoder_outputs:V,input_ids:_e,decoder_input_ids:Ie,...ke}=F;if(!V){const Qe=(0,d.pick)(F,S.sessions.model.inputNames);V=(await le(S,Qe)).last_hidden_state}return ke.input_ids=Ie,ke.encoder_hidden_states=V,S.sessions.decoder_model_merged.inputNames.includes("encoder_attention_mask")&&(ke.encoder_attention_mask=F.attention_mask),await Ce(S,ke,!0)}async function le(S,F){const V=S.sessions.model,_e=(0,d.pick)(F,V.inputNames);if(V.inputNames.includes("inputs_embeds")&&!_e.inputs_embeds){if(!F.input_ids)throw new Error("Both `input_ids` and `inputs_embeds` are missing in the model inputs.");_e.inputs_embeds=await S.encode_text({input_ids:F.input_ids})}if(V.inputNames.includes("token_type_ids")&&!_e.token_type_ids){if(!_e.input_ids)throw new Error("Both `input_ids` and `token_type_ids` are missing in the model inputs.");_e.token_type_ids=(0,g.zeros_like)(_e.input_ids)}if(V.inputNames.includes("pixel_mask")&&!_e.pixel_mask){if(!_e.pixel_values)throw new Error("Both `pixel_values` and `pixel_mask` are missing in the model inputs.");const Ie=_e.pixel_values.dims;_e.pixel_mask=(0,g.ones)([Ie[0],Ie[2],Ie[3]])}return await K(V,_e)}async function Se(S,F){const V=await S.encode(F);return await S.decode(V)}async function Ce(S,F,V=!1){const _e=S.sessions[V?"decoder_model_merged":"model"],{past_key_values:Ie,...ke}=F;if(_e.inputNames.includes("use_cache_branch")&&(ke.use_cache_branch=oe(!!Ie)),_e.inputNames.includes("position_ids")&&ke.attention_mask&&!ke.position_ids){const Qe=["paligemma","gemma3_text","gemma3"].includes(S.config.model_type)?1:0;ke.position_ids=Ze(ke,Ie,Qe)}S.addPastKeyValues(ke,Ie);const Be=(0,d.pick)(ke,_e.inputNames);return await K(_e,Be)}function q({modality_token_id:S,inputs_embeds:F,modality_features:V,input_ids:_e,attention_mask:Ie}){const ke=_e.tolist().map(lt=>lt.reduce((Pt,It,Ct)=>(It==S&&Pt.push(Ct),Pt),[])),Be=ke.reduce((lt,Pt)=>lt+Pt.length,0),Qe=V.dims[0];if(Be!==Qe)throw new Error(`Number of tokens and features do not match: tokens: ${Be}, features ${Qe}`);let rt=0;for(let lt=0;ltke.dims[1])){if(IeQe==S.config.image_token_index)){const Qe=S.config.num_image_tokens;if(!Qe)throw new Error("`num_image_tokens` is missing in the model configuration.");const rt=ke.dims[1]-(Ie-Qe);V.input_ids=ke.slice(null,[-rt,null]),V.attention_mask=(0,g.ones)([1,Ie+rt])}}}return V}function Ee(S,F,V,_e){return V.past_key_values&&(F=F.map(Ie=>[Ie.at(-1)])),{...V,decoder_input_ids:re(F)}}function Z(S,...F){return S.config.is_encoder_decoder?Ee(S,...F):et(S,...F)}function me(S,F,V,_e){const Ie=!!V.past_key_values;return _e.guidance_scale!==null&&_e.guidance_scale>1&&(Ie?V.input_ids=(0,g.cat)([V.input_ids,V.input_ids],0):(V.input_ids=(0,g.cat)([V.input_ids,(0,g.full_like)(V.input_ids,BigInt(_e.pad_token_id))],0),V.attention_mask=(0,g.cat)([V.attention_mask,(0,g.full_like)(V.attention_mask,0n)],0))),(Ie||!V.pixel_values)&&(V.pixel_values=(0,g.full)([0,0,3,384,384],1)),Ie&&(V.images_seq_mask=new g.Tensor("bool",new Array(1).fill(!0).fill(!1,0,1),[1,1]),V.images_emb_mask=new g.Tensor("bool",new Array(0).fill(!1),[1,1,0])),V}class X extends u.Callable{constructor(V,_e,Ie){super();de(this,"main_input_name","input_ids");de(this,"forward_params",["input_ids","attention_mask"]);this.config=V,this.sessions=_e,this.configs=Ie;const ke=k.get(this.constructor),Be=$.get(ke);switch(this.can_generate=!1,this._forward=null,this._prepare_inputs_for_generation=null,Be){case A.DecoderOnly:this.can_generate=!0,this._forward=Ce,this._prepare_inputs_for_generation=et;break;case A.Seq2Seq:case A.Vision2Seq:case A.Musicgen:this.can_generate=!0,this._forward=ge,this._prepare_inputs_for_generation=Ee;break;case A.EncoderDecoder:this._forward=ge;break;case A.ImageTextToText:this.can_generate=!0,this._forward=we,this._prepare_inputs_for_generation=Z;break;case A.AudioTextToText:this.can_generate=!0,this._forward=Ae,this._prepare_inputs_for_generation=Z;break;case A.Phi3V:this.can_generate=!0,this._prepare_inputs_for_generation=Z;break;case A.MultiModality:this.can_generate=!0,this._prepare_inputs_for_generation=me;break;case A.AutoEncoder:this._forward=Se;break;default:this._forward=le;break}this.can_generate&&this.forward_params.push("past_key_values"),this.custom_config=this.config["transformers.js_config"]??{}}async dispose(){var _e;const V=[];for(const Ie of Object.values(this.sessions))(_e=Ie==null?void 0:Ie.handler)!=null&&_e.dispose&&V.push(Ie.handler.dispose());return await Promise.all(V)}static async from_pretrained(V,{progress_callback:_e=null,config:Ie=null,cache_dir:ke=null,local_files_only:Be=!1,revision:Qe="main",model_file_name:rt=null,subfolder:lt="onnx",device:Pt=null,dtype:It=null,use_external_data_format:Ct=null,session_options:Bt={}}={}){let kt={progress_callback:_e,config:Ie,cache_dir:ke,local_files_only:Be,revision:Qe,model_file_name:rt,subfolder:lt,device:Pt,dtype:It,use_external_data_format:Ct,session_options:Bt};const jt=k.get(this),_t=$.get(jt);Ie=kt.config=await i.AutoConfig.from_pretrained(V,kt);let Wt;if(_t===A.DecoderOnly)Wt=await Promise.all([R(V,{model:kt.model_file_name??"model"},kt),U(V,{generation_config:"generation_config.json"},kt)]);else if(_t===A.Seq2Seq||_t===A.Vision2Seq)Wt=await Promise.all([R(V,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},kt),U(V,{generation_config:"generation_config.json"},kt)]);else if(_t===A.MaskGeneration)Wt=await Promise.all([R(V,{model:"vision_encoder",prompt_encoder_mask_decoder:"prompt_encoder_mask_decoder"},kt)]);else if(_t===A.EncoderDecoder)Wt=await Promise.all([R(V,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},kt)]);else if(_t===A.ImageTextToText){const nr={embed_tokens:"embed_tokens",vision_encoder:"vision_encoder",decoder_model_merged:"decoder_model_merged"};Ie.is_encoder_decoder&&(nr.model="encoder_model"),Wt=await Promise.all([R(V,nr,kt),U(V,{generation_config:"generation_config.json"},kt)])}else if(_t===A.AudioTextToText){const nr={embed_tokens:"embed_tokens",audio_encoder:"audio_encoder",decoder_model_merged:"decoder_model_merged"};Wt=await Promise.all([R(V,nr,kt),U(V,{generation_config:"generation_config.json"},kt)])}else if(_t===A.Musicgen)Wt=await Promise.all([R(V,{model:"text_encoder",decoder_model_merged:"decoder_model_merged",encodec_decode:"encodec_decode"},kt),U(V,{generation_config:"generation_config.json"},kt)]);else if(_t===A.MultiModality)Wt=await Promise.all([R(V,{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"},kt),U(V,{generation_config:"generation_config.json"},kt)]);else if(_t===A.Phi3V)Wt=await Promise.all([R(V,{prepare_inputs_embeds:"prepare_inputs_embeds",model:"model",vision_encoder:"vision_encoder"},kt),U(V,{generation_config:"generation_config.json"},kt)]);else if(_t===A.AutoEncoder)Wt=await Promise.all([R(V,{encoder_model:"encoder_model",decoder_model:"decoder_model"},kt)]);else{if(_t!==A.EncoderOnly){const nr=jt??(Ie==null?void 0:Ie.model_type);nr!=="custom"&&console.warn(`Model type for '${nr}' not found, assuming encoder-only architecture. Please report this at ${f.GITHUB_ISSUE_URL}.`)}Wt=await Promise.all([R(V,{model:kt.model_file_name??"model"},kt)])}return new this(Ie,...Wt)}async _call(V){return await this.forward(V)}async forward(V){return await this._forward(this,V)}get generation_config(){var V;return((V=this.configs)==null?void 0:V.generation_config)??null}_get_logits_warper(V){const _e=new y.LogitsProcessorList;return V.temperature!==null&&V.temperature!==1&&_e.push(new y.TemperatureLogitsWarper(V.temperature)),V.top_k!==null&&V.top_k!==0&&_e.push(new y.TopKLogitsWarper(V.top_k)),V.top_p!==null&&V.top_p<1&&_e.push(new y.TopPLogitsWarper(V.top_p)),_e}_get_logits_processor(V,_e,Ie=null){const ke=new y.LogitsProcessorList;if(V.repetition_penalty!==null&&V.repetition_penalty!==1&&ke.push(new y.RepetitionPenaltyLogitsProcessor(V.repetition_penalty)),V.no_repeat_ngram_size!==null&&V.no_repeat_ngram_size>0&&ke.push(new y.NoRepeatNGramLogitsProcessor(V.no_repeat_ngram_size)),V.bad_words_ids!==null&&ke.push(new y.NoBadWordsLogitsProcessor(V.bad_words_ids,V.eos_token_id)),V.min_length!==null&&V.eos_token_id!==null&&V.min_length>0&&ke.push(new y.MinLengthLogitsProcessor(V.min_length,V.eos_token_id)),V.min_new_tokens!==null&&V.eos_token_id!==null&&V.min_new_tokens>0&&ke.push(new y.MinNewTokensLengthLogitsProcessor(_e,V.min_new_tokens,V.eos_token_id)),V.forced_bos_token_id!==null&&ke.push(new y.ForcedBOSTokenLogitsProcessor(V.forced_bos_token_id)),V.forced_eos_token_id!==null&&ke.push(new y.ForcedEOSTokenLogitsProcessor(V.max_length,V.forced_eos_token_id)),V.begin_suppress_tokens!==null){const Be=_e>1||V.forced_bos_token_id===null?_e:_e+1;ke.push(new y.SuppressTokensAtBeginLogitsProcessor(V.begin_suppress_tokens,Be))}return V.guidance_scale!==null&&V.guidance_scale>1&&ke.push(new y.ClassifierFreeGuidanceLogitsProcessor(V.guidance_scale)),Ie!==null&&ke.extend(Ie),ke}_prepare_generation_config(V,_e,Ie=m.GenerationConfig){const ke={...this.config};for(const Qe of["decoder","generator","text_config"])Qe in ke&&Object.assign(ke,ke[Qe]);const Be=new Ie(ke);return Object.assign(Be,this.generation_config??{}),V&&Object.assign(Be,V),_e&&Object.assign(Be,(0,d.pick)(_e,Object.getOwnPropertyNames(Be))),Be}_get_stopping_criteria(V,_e=null){const Ie=new I.StoppingCriteriaList;return V.max_length!==null&&Ie.push(new I.MaxLengthCriteria(V.max_length,this.config.max_position_embeddings??null)),V.eos_token_id!==null&&Ie.push(new I.EosTokenCriteria(V.eos_token_id)),_e&&Ie.extend(_e),Ie}_validate_model_class(){if(!this.can_generate){const V=[Hu,qu,Ku,Gu],_e=k.get(this.constructor),Ie=new Set,ke=this.config.model_type;for(const Qe of V){const rt=Qe.get(ke);rt&&Ie.add(rt[0])}let Be=`The current model class (${_e}) is not compatible with \`.generate()\`, as it doesn't have a language model head.`;throw Ie.size>0&&(Be+=` Please use the following class instead: ${[...Ie].join(", ")}`),Error(Be)}}prepare_inputs_for_generation(...V){return this._prepare_inputs_for_generation(this,...V)}_update_model_kwargs_for_generation({generated_input_ids:V,outputs:_e,model_inputs:Ie,is_encoder_decoder:ke}){return Ie.past_key_values=this.getPastKeyValues(_e,Ie.past_key_values),Ie.input_ids=new g.Tensor("int64",V.flat(),[V.length,1]),ke||(Ie.attention_mask=(0,g.cat)([Ie.attention_mask,(0,g.ones)([Ie.attention_mask.dims[0],1])],1)),Ie.position_ids=null,Ie}_prepare_model_inputs({inputs:V,bos_token_id:_e,model_kwargs:Ie}){const ke=(0,d.pick)(Ie,this.forward_params),Be=this.main_input_name;if(Be in ke){if(V)throw new Error("`inputs`: {inputs}` were passed alongside {input_name} which is not allowed. Make sure to either pass {inputs} or {input_name}=...")}else ke[Be]=V;return{inputs_tensor:ke[Be],model_inputs:ke,model_input_name:Be}}async _prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:V,model_inputs:_e,model_input_name:Ie,generation_config:ke}){if(this.sessions.model.inputNames.includes("inputs_embeds")&&!_e.inputs_embeds&&"_prepare_inputs_embeds"in this){const{input_ids:Qe,pixel_values:rt,attention_mask:lt,...Pt}=_e,It=await this._prepare_inputs_embeds(_e);_e={...Pt,...(0,d.pick)(It,["inputs_embeds","attention_mask"])}}let{last_hidden_state:Be}=await le(this,_e);if(ke.guidance_scale!==null&&ke.guidance_scale>1)Be=(0,g.cat)([Be,(0,g.full_like)(Be,0)],0),"attention_mask"in _e&&(_e.attention_mask=(0,g.cat)([_e.attention_mask,(0,g.zeros_like)(_e.attention_mask)],0));else if(_e.decoder_input_ids){const Qe=re(_e.decoder_input_ids).dims[0];if(Qe!==Be.dims[0]){if(Be.dims[0]!==1)throw new Error(`The encoder outputs have a different batch size (${Be.dims[0]}) than the decoder inputs (${Qe}).`);Be=(0,g.cat)(Array.from({length:Qe},()=>Be),0)}}return _e.encoder_outputs=Be,_e}_prepare_decoder_input_ids_for_generation({batch_size:V,model_input_name:_e,model_kwargs:Ie,decoder_start_token_id:ke,bos_token_id:Be,generation_config:Qe}){let{decoder_input_ids:rt,...lt}=Ie;if(!(rt instanceof g.Tensor)){if(rt)Array.isArray(rt[0])||(rt=Array.from({length:V},()=>rt));else if(ke??(ke=Be),this.config.model_type==="musicgen")rt=Array.from({length:V*this.config.decoder.num_codebooks},()=>[ke]);else if(Array.isArray(ke)){if(ke.length!==V)throw new Error(`\`decoder_start_token_id\` expcted to have length ${V} but got ${ke.length}`);rt=ke}else rt=Array.from({length:V},()=>[ke]);rt=re(rt)}return Ie.decoder_attention_mask=(0,g.ones_like)(rt),{input_ids:rt,model_inputs:lt}}async generate({inputs:V=null,generation_config:_e=null,logits_processor:Ie=null,stopping_criteria:ke=null,streamer:Be=null,...Qe}){this._validate_model_class(),_e=this._prepare_generation_config(_e,Qe);let{inputs_tensor:rt,model_inputs:lt,model_input_name:Pt}=this._prepare_model_inputs({inputs:V,model_kwargs:Qe});const It=this.config.is_encoder_decoder;It&&("encoder_outputs"in lt||(lt=await this._prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:rt,model_inputs:lt,model_input_name:Pt,generation_config:_e})));let Ct;It?{input_ids:Ct,model_inputs:lt}=this._prepare_decoder_input_ids_for_generation({batch_size:lt[Pt].dims.at(0),model_input_name:Pt,model_kwargs:lt,decoder_start_token_id:_e.decoder_start_token_id,bos_token_id:_e.bos_token_id,generation_config:_e}):Ct=lt[Pt];let Bt=Ct.dims.at(-1);_e.max_new_tokens!==null&&(_e.max_length=Bt+_e.max_new_tokens);const kt=this._get_logits_processor(_e,Bt,Ie),jt=this._get_stopping_criteria(_e,ke),_t=lt[Pt].dims.at(0),Wt=z.LogitsSampler.getSampler(_e),nr=new Array(_t).fill(0),ar=Ct.tolist();Be&&Be.put(ar);let mr,hr={};for(;;){if(lt=this.prepare_inputs_for_generation(ar,lt,_e),mr=await this.forward(lt),_e.output_attentions&&_e.return_dict_in_generate){const Gr=this.getAttentions(mr);for(const ns in Gr)ns in hr||(hr[ns]=[]),hr[ns].push(Gr[ns])}const Rt=mr.logits.slice(null,-1,null),xr=kt(ar,Rt),es=[];for(let Gr=0;GrGr))break;lt=this._update_model_kwargs_for_generation({generated_input_ids:es,outputs:mr,model_inputs:lt,is_encoder_decoder:It})}Be&&Be.end();const wr=this.getPastKeyValues(mr,lt.past_key_values,!0),kr=new g.Tensor("int64",ar.flat(),[ar.length,ar[0].length]);if(_e.return_dict_in_generate)return{sequences:kr,past_key_values:wr,...hr};for(const Rt of Object.values(mr))Rt.location==="gpu-buffer"&&Rt.dispose();return kr}getPastKeyValues(V,_e,Ie=!1){const ke=Object.create(null);for(const Be in V)if(Be.startsWith("present")){const Qe=Be.replace("present","past_key_values"),rt=Be.includes("encoder");if(rt&&_e?ke[Qe]=_e[Qe]:ke[Qe]=V[Be],_e&&(!rt||Ie)){const lt=_e[Qe];lt.location==="gpu-buffer"&<.dispose()}}return ke}getAttentions(V){const _e={};for(const Ie of["cross_attentions","encoder_attentions","decoder_attentions"])for(const ke in V)ke.startsWith(Ie)&&(Ie in _e||(_e[Ie]=[]),_e[Ie].push(V[ke]));return _e}addPastKeyValues(V,_e){var Ie,ke,Be;if(_e)Object.assign(V,_e);else{const Qe=this.sessions.decoder_model_merged??this.sessions.model,rt=((Ie=Qe==null?void 0:Qe.config)==null?void 0:Ie.kv_cache_dtype)??"float32",lt=rt==="float16"?new g.DataTypeMap.float16:[],Pt=((Be=(ke=V[this.main_input_name]??V.attention_mask)==null?void 0:ke.dims)==null?void 0:Be[0])??1,It=(0,i.getKeyValueShapes)(this.config,{batch_size:Pt});for(const Ct in It)V[Ct]=new g.Tensor(rt,lt,It[Ct])}}async encode_image({pixel_values:V}){const _e=(await K(this.sessions.vision_encoder,{pixel_values:V})).image_features;return this.config.num_image_tokens||(console.warn(`The number of image tokens was not set in the model configuration. Setting it to the number of features detected by the vision encoder (${_e.dims[1]}).`),this.config.num_image_tokens=_e.dims[1]),_e}async encode_text({input_ids:V}){return(await K(this.sessions.embed_tokens,{input_ids:V})).inputs_embeds}async encode_audio({audio_values:V}){return(await K(this.sessions.audio_encoder,{audio_values:V})).audio_features}}class Oe{}class Fe extends Oe{constructor({last_hidden_state:F,hidden_states:V=null,attentions:_e=null}){super(),this.last_hidden_state=F,this.hidden_states=V,this.attentions=_e}}class Ve extends X{}class Re extends Ve{}class Ue extends Ve{async _call(F){return new Or(await super._call(F))}}class Ge extends Ve{async _call(F){return new Dt(await super._call(F))}}class Je extends Ve{async _call(F){return new Sr(await super._call(F))}}class at extends Ve{async _call(F){return new Nr(await super._call(F))}}class ot extends X{}class Ke extends ot{}class st extends ot{async _call(F){return new Or(await super._call(F))}}class ft extends ot{async _call(F){return new Dt(await super._call(F))}}class yt extends ot{async _call(F){return new Sr(await super._call(F))}}class St extends X{}class vt extends St{}class He extends X{}class xt extends He{}class Vt extends He{async _call(F){return new Or(await super._call(F))}}class he extends He{async _call(F){return new Dt(await super._call(F))}}class ce extends He{async _call(F){return new Sr(await super._call(F))}}class Cr extends He{async _call(F){return new Nr(await super._call(F))}}class er extends X{}class ht extends er{}class as extends er{async _call(F){return new Or(await super._call(F))}}class W extends er{async _call(F){return new Dt(await super._call(F))}}class ye extends er{async _call(F){return new Sr(await super._call(F))}}class G extends er{async _call(F){return new Nr(await super._call(F))}}class ne extends X{}class $e extends ne{}class Xe extends ne{async _call(F){return new Or(await super._call(F))}}class Le extends ne{async _call(F){return new Dt(await super._call(F))}}class Ft extends ne{async _call(F){return new Sr(await super._call(F))}}class Qt extends ne{async _call(F){return new Nr(await super._call(F))}}class wt extends X{}class Nt extends wt{}class mt extends wt{async _call(F){return new Or(await super._call(F))}}class tr extends wt{async _call(F){return new Dt(await super._call(F))}}class rr extends wt{async _call(F){return new Sr(await super._call(F))}}class Fr extends wt{async _call(F){return new Nr(await super._call(F))}}class pr extends X{}class Kr extends pr{}class xs extends pr{async _call(F){return new Or(await super._call(F))}}class Ws extends pr{async _call(F){return new Dt(await super._call(F))}}class Ts extends pr{async _call(F){return new Sr(await super._call(F))}}class en extends pr{async _call(F){return new Nr(await super._call(F))}}class rs extends X{}class tn extends rs{}class os extends rs{async _call(F){return new Or(await super._call(F))}}class Us extends rs{async _call(F){return new Dt(await super._call(F))}}class Gs extends rs{async _call(F){return new Sr(await super._call(F))}}class Ks extends rs{async _call(F){return new Nr(await super._call(F))}}class Wr extends X{}class Jt extends Wr{}class Cs extends Wr{async _call(F){return new Dt(await super._call(F))}}class Hs extends Wr{async _call(F){return new Sr(await super._call(F))}}class Er extends Wr{async _call(F){return new Nr(await super._call(F))}}class qe extends Wr{async _call(F){return new Or(await super._call(F))}}class ut extends X{}class bt extends ut{}class cr extends ut{async _call(F){return new Or(await super._call(F))}}class ls extends ut{async _call(F){return new Dt(await super._call(F))}}class Es extends ut{async _call(F){return new Sr(await super._call(F))}}class Xr extends X{}class Ps extends Xr{}class Ss extends Xr{async _call(F){return new Or(await super._call(F))}}class ks extends Xr{async _call(F){return new Dt(await super._call(F))}}class $s extends Xr{async _call(F){return new Nr(await super._call(F))}}class gs extends X{}class _n extends gs{}class gn extends gs{async _call(F){return new Or(await super._call(F))}}class Fn extends gs{async _call(F){return new Dt(await super._call(F))}}class Dn extends gs{async _call(F){return new Sr(await super._call(F))}}class Hr extends gs{async _call(F){return new Nr(await super._call(F))}}class ys extends X{}class rn extends ys{}class cs extends ys{async _call(F){return new Or(await super._call(F))}}class Jr extends ys{async _call(F){return new Dt(await super._call(F))}}class yn extends ys{async _call(F){return new Nr(await super._call(F))}}class qs extends X{}class Te extends qs{}class L extends qs{async _call(F){return new Dt(await super._call(F))}}class J extends qs{async _call(F){return new Nr(await super._call(F))}}class ie extends qs{async _call(F){return new Or(await super._call(F))}}class ve extends X{constructor(){super(...arguments);de(this,"forward_params",["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"])}}class be extends ve{}class ze extends ve{}class tt extends X{}class ct extends tt{}class it extends tt{}class dt extends X{}class At extends dt{}class qt extends dt{}class Ht extends X{}class fr extends Ht{}class sr extends Ht{}class Pr extends Ht{async _call(F){return new Dt(await super._call(F))}}class $r extends X{}class qr extends $r{}class ws extends $r{}class Ur extends $r{async _call(F){return new Dt(await super._call(F))}}class Qs extends $r{}class ur extends X{}class Ir extends ur{}class zr extends ur{}class Yr extends X{}class ss extends Yr{}class Br extends Yr{}class Dr extends X{}class Ar extends Dr{}class yr extends Dr{async _call(F){return new Or(await super._call(F))}}class Mr extends Dr{async _call(F){return new Dt(await super._call(F))}}class vr extends Dr{async _call(F){return new Sr(await super._call(F))}}class Rr extends Dr{async _call(F){return new Nr(await super._call(F))}}class us extends X{}class bi extends us{}class xd extends us{async _call(F){return new Or(await super._call(F))}}class jn extends us{async _call(F){return new Dt(await super._call(F))}}class Xs extends us{async _call(F){return new Sr(await super._call(F))}}class ds extends us{async _call(F){return new Nr(await super._call(F))}}class Tt extends X{}class Ua extends Tt{}class Ln extends Tt{async _call(F){return new Or(await super._call(F))}}class Td extends Tt{async _call(F){return new Dt(await super._call(F))}}class Cd extends Tt{async _call(F){return new Sr(await super._call(F))}}class zn extends Tt{async _call(F){return new Nr(await super._call(F))}}class Ga extends X{}class Ed extends Ga{}class Ka extends Ga{}class Ha extends X{constructor(){super(...arguments);de(this,"requires_attention_mask",!1);de(this,"main_input_name","input_features");de(this,"forward_params",["input_features","attention_mask","decoder_input_ids","decoder_attention_mask","past_key_values"])}}class Pd extends Ha{}class qa extends Ha{_prepare_generation_config(F,V){return super._prepare_generation_config(F,V,C.WhisperGenerationConfig)}_retrieve_init_tokens(F){const V=[F.decoder_start_token_id];let _e=F.language;const Ie=F.task;if(F.is_multilingual){_e||(console.warn("No language specified - defaulting to English (en)."),_e="en");const Be=`<|${(0,D.whisper_language_to_code)(_e)}|>`;V.push(F.lang_to_id[Be]),V.push(F.task_to_id[Ie??"transcribe"])}else if(_e||Ie)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!F.return_timestamps&&F.no_timestamps_token_id&&V.at(-1)!==F.no_timestamps_token_id?V.push(F.no_timestamps_token_id):F.return_timestamps&&V.at(-1)===F.no_timestamps_token_id&&(console.warn("<|notimestamps|> prompt token is removed from generation_config since `return_timestamps` is set to `true`."),V.pop()),V.filter(ke=>ke!=null)}async generate({inputs:F=null,generation_config:V=null,logits_processor:_e=null,stopping_criteria:Ie=null,...ke}){V=this._prepare_generation_config(V,ke);const Be=ke.decoder_input_ids??this._retrieve_init_tokens(V);if(V.return_timestamps&&(_e??(_e=new y.LogitsProcessorList),_e.push(new y.WhisperTimeStampLogitsProcessor(V,Be))),V.begin_suppress_tokens&&(_e??(_e=new y.LogitsProcessorList),_e.push(new y.SuppressTokensAtBeginLogitsProcessor(V.begin_suppress_tokens,Be.length))),V.return_token_timestamps){if(!V.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.");V.task==="translate"&&console.warn("Token-level timestamps may not be reliable for task 'translate'."),V.output_attentions=!0,V.return_dict_in_generate=!0}const Qe=await super.generate({inputs:F,generation_config:V,logits_processor:_e,decoder_input_ids:Be,...ke});return V.return_token_timestamps&&(Qe.token_timestamps=this._extract_token_timestamps(Qe,V.alignment_heads,V.num_frames)),Qe}_extract_token_timestamps(F,V,_e=null,Ie=.02){if(!F.cross_attentions)throw new Error("Model outputs must contain cross attentions to extract timestamps. This is most likely because the model was not exported with `output_attentions=True`.");_e==null&&console.warn("`num_frames` has not been set, meaning the entire audio will be analyzed. This may lead to inaccurate token-level timestamps for short audios (< 30 seconds).");let ke=this.config.median_filter_width;ke===void 0&&(console.warn("Model config has no `median_filter_width`, using default value of 7."),ke=7);const Be=F.cross_attentions,Qe=Array.from({length:this.config.decoder_layers},(jt,_t)=>(0,g.cat)(Be.map(Wt=>Wt[_t]),2)),rt=(0,g.stack)(V.map(([jt,_t])=>{if(jt>=Qe.length)throw new Error(`Layer index ${jt} is out of bounds for cross attentions (length ${Qe.length}).`);return _e?Qe[jt].slice(null,_t,null,[0,_e]):Qe[jt].slice(null,_t)})).transpose(1,0,2,3),[lt,Pt]=(0,g.std_mean)(rt,-2,0,!0),It=rt.clone();for(let jt=0;jtWt[kr+1]-Wt[kr]),mr=(0,d.mergeArrays)([1],ar).map(wr=>!!wr),hr=[];for(let wr=0;wrCt.findIndex(Bt=>Bt==ke)),rt=Qe.every(Ct=>Ct===-1),lt=Qe.every(Ct=>Ct!==-1);if(!rt&&!lt)throw new Error("Every input should contain either 0 or 1 image token.");if(rt)return{inputs_embeds:F,attention_mask:Ie};const Pt=[],It=[];for(let Ct=0;CtArray.from({length:F.dims[0]},ar=>Array.from({length:F.dims[1]},mr=>1))),kt=V?V.tolist():[],jt=_e?_e.tolist():[];let _t=0,Wt=0;for(let nr=0;nrCt[nr][Tr]==1),hr=ar.reduce((or,Tr,Ys)=>(Tr==rt&&or.push(Ys),or),[]).map(or=>ar[or+1]),wr=hr.filter(or=>or==Be).length,kr=hr.filter(or=>or==Qe).length;let Rt=[],xr=0,es=wr,ln=kr;for(let or=0;orhs>xr&&dn==Be),Ys=ar.findIndex((dn,hs)=>hs>xr&&dn==Qe),un=es>0&&Tr!==-1?Tr:ar.length+1,Pn=ln>0&&Ys!==-1?Ys:ar.length+1;let Oa,Xu,Ju,Yu;un0?(0,M.max)(Rt.at(-1))[0]+1:0;Rt.push(Array.from({length:3*ed},(dn,hs)=>Oh+hs%ed));const td=ed+Oh,Da=Mf*Zu*Fa,vf=Array.from({length:Da},(dn,hs)=>td+Math.floor(hs/(Zu*Fa))),xf=Array.from({length:Da},(dn,hs)=>td+Math.floor(hs/Fa)%Zu),Tf=Array.from({length:Da},(dn,hs)=>td+hs%Fa);Rt.push([vf,xf,Tf].flat()),xr=Oa+Da}if(xr0?(0,M.max)(Rt.at(-1))[0]+1:0,Tr=ar.length-xr;Rt.push(Array.from({length:3*Tr},(Ys,un)=>or+un%Tr))}const Gr=Rt.reduce((or,Tr)=>or+Tr.length,0),ns=new Array(Gr);let $a=0;for(let or=0;or<3;++or)for(let Tr=0;TrIt[_t%It.length]),kt=Array.from({length:Ct[0]},(jt,_t)=>(0,M.max)(It.subarray(Ct[1]*_t,Ct[1]*(_t+1)))[0]+1n+BigInt(Ct[1]));return[new g.Tensor("int64",Bt,[3,...Ct]),new g.Tensor("int64",kt,[kt.length,1])]}else{const[It,Ct]=F.dims,Bt=BigInt64Array.from({length:3*It*Ct},(kt,jt)=>BigInt(Math.floor(jt%Ct/It)));return[new g.Tensor("int64",Bt,[3,...F.dims]),(0,g.zeros)([It,1])]}}async encode_image({pixel_values:F,image_grid_thw:V}){return(await K(this.sessions.vision_encoder,{pixel_values:F,grid_thw:V})).image_features}_merge_input_ids_with_image_features(F){return N({image_token_id:this.config.image_token_id,...F})}prepare_inputs_for_generation(F,V,_e){if(V.attention_mask&&!V.position_ids)if(!V.past_key_values)[V.position_ids,V.rope_deltas]=this.get_rope_index(V.input_ids,V.image_grid_thw,V.video_grid_thw,V.attention_mask);else{V.pixel_values=null;const Ie=BigInt(Object.values(V.past_key_values)[0].dims.at(-2)),ke=V.rope_deltas.map(Be=>Ie+Be);V.position_ids=(0,g.stack)([ke,ke,ke],0)}return V}}class $i extends X{}class Oo extends $i{}class Ii extends $i{}class vn extends X{}class Zr extends vn{}class nn extends vn{}class Ai extends X{}class Fo extends Ai{}class Oi extends Ai{}class Rn extends X{}class Fi extends Rn{}class Di extends Rn{}class Nn extends X{}class ji extends Nn{}class Do extends Nn{}class Vn extends X{}class jo extends Vn{}class Lo extends Vn{async _call(F){return new Dt(await super._call(F))}}class Li extends X{}class zo extends Li{}class Bo extends Li{async _call(F){return new Dt(await super._call(F))}}class Ro extends X{}class No extends Ro{}class zi extends X{}class Vo extends zi{}class Wo extends zi{async _call(F){return new Dt(await super._call(F))}}class Uo extends X{}class Bi extends Uo{}class Ri extends X{}class Ni extends Ri{}class Vi extends Ri{async _call(F){return new Dt(await super._call(F))}}class Wi extends X{}class xn extends Wi{}class Wn extends X{}class Go extends Wn{}class Ui extends Wn{async _call(F){return new Dt(await super._call(F))}}class Ko extends X{}class Ho extends Ko{async _call(F){return new Ih(await super._call(F))}}class Gi extends X{}class Ki extends Gi{}class qo extends Gi{async _call(F){return new Dt(await super._call(F))}}class Hi extends X{}class qi extends Hi{}class Me extends Hi{async _call(F){return new Dt(await super._call(F))}}class Qi extends X{}class Qo extends Qi{}class Xo extends Qi{}class Xi extends X{}class Jo extends Xi{}class Yo extends Xi{}class Ji extends X{}class Zo extends Ji{}class el extends Ji{async _call(F){return new Dt(await super._call(F))}}class Un extends X{}class tl extends Un{}class rl extends Un{async _call(F){return new Yi(await super._call(F))}}class Gn extends Un{async _call(F){return new sl(await super._call(F))}}class Yi extends Oe{constructor({logits:F,pred_boxes:V}){super(),this.logits=F,this.pred_boxes=V}}class sl extends Oe{constructor({logits:F,pred_boxes:V,pred_masks:_e}){super(),this.logits=F,this.pred_boxes=V,this.pred_masks=_e}}class Zi extends X{}class nl extends Zi{}class ea extends Zi{async _call(F){return new Tn(await super._call(F))}}class Tn extends Oe{constructor({logits:F,pred_boxes:V}){super(),this.logits=F,this.pred_boxes=V}}class ta extends X{}class il extends ta{}class al extends ta{async _call(F){return new ol(await super._call(F))}}class ol extends Tn{}class ra extends X{}class ll extends ra{}class cl extends ra{async _call(F){return new ul(await super._call(F))}}class ul extends Tn{}class sa extends X{}class dl extends sa{}class pl extends sa{async _call(F){return new Tn(await super._call(F))}}class na extends X{}class hl extends na{}class fl extends na{async _call(F){return new ml(await super._call(F))}}class ml extends Yi{}class ia extends X{}class _l extends ia{}class gl extends ia{async _call(F){return new Dt(await super._call(F))}}class aa extends X{}class yl extends aa{}class wl extends aa{async _call(F){return new Dt(await super._call(F))}}class oa extends X{}class bl extends oa{}class Ml extends oa{async _call(F){return new Dt(await super._call(F))}}class Kn extends X{}class vl extends Kn{}class xl extends Kn{async _call(F){return new Dt(await super._call(F))}}class Tl extends Kn{}class la extends X{}class Cl extends la{}class El extends la{}class ca extends X{}class Pl extends ca{}class Sl extends ca{}class kl extends X{}class $l extends kl{}class Hn extends X{}class Il extends Hn{}class Al extends Hn{}class Ol extends Hn{}class Fl extends X{}class Dl extends Fl{}class jl extends X{}class Ll extends jl{}class zl extends X{}class Bl extends zl{}class ua extends X{}class Rl extends ua{}class Nl extends ua{}class da extends X{}class Vl extends da{}class Wl extends da{}class Ul extends X{}class Gl extends Ul{}class pa extends X{}class Kl extends pa{}class Hl extends pa{async _call(F){return new Dt(await super._call(F))}}class ha extends X{}class ql extends ha{}class Ql extends ha{async _call(F){return new Dt(await super._call(F))}}class fa extends X{}class Xl extends fa{}class Jl extends fa{async _call(F){return new Dt(await super._call(F))}}class ma extends X{}class Yl extends ma{}class Zl extends ma{async _call(F){return new Dt(await super._call(F))}}class ec extends X{}class tc extends ec{}class _a extends X{}class rc extends _a{}class sc extends _a{async _call(F){return new nc(await super._call(F))}}class nc extends Oe{constructor({logits:F,pred_boxes:V}){super(),this.logits=F,this.pred_boxes=V}}class ic extends X{}class ac extends ic{async get_image_embeddings({pixel_values:F}){return await le(this,{pixel_values:F})}async forward(F){if((!F.image_embeddings||!F.image_positional_embeddings)&&(F={...F,...await this.get_image_embeddings(F)}),!F.input_labels&&F.input_points){const _e=F.input_points.dims.slice(0,-1),Ie=_e.reduce((ke,Be)=>ke*Be,1);F.input_labels=new g.Tensor("int64",new BigInt64Array(Ie).fill(1n),_e)}const V={image_embeddings:F.image_embeddings,image_positional_embeddings:F.image_positional_embeddings};return F.input_points&&(V.input_points=F.input_points),F.input_labels&&(V.input_labels=F.input_labels),F.input_boxes&&(V.input_boxes=F.input_boxes),await K(this.sessions.prompt_encoder_mask_decoder,V)}async _call(F){return new oc(await super._call(F))}}class oc extends Oe{constructor({iou_scores:F,pred_masks:V}){super(),this.iou_scores=F,this.pred_masks=V}}class ga extends X{}class lc extends ga{}class cc extends ga{}class ya extends X{}class uc extends ya{}class dc extends ya{}class Is extends X{}class pc extends Is{}class hc extends Is{async _call(F){return new on(await super._call(F))}}class fc extends Is{async _call(F){return new Dt(await super._call(F))}}class mc extends Is{async _call(F){return new Sr(await super._call(F))}}class wa extends X{}class _c extends wa{}class gc extends wa{async _call(F){return new Sr(await super._call(F))}}class yc extends X{}class wc extends yc{}class qn extends X{}class bc extends qn{}class Mc extends qn{async _call(F){return new on(await super._call(F))}}class vc extends qn{async _call(F){return new Dt(await super._call(F))}}class Cn extends X{}class xc extends Cn{}class Tc extends Cn{async _call(F){return new on(await super._call(F))}}class Cc extends Cn{async _call(F){return new Dt(await super._call(F))}}class Ec extends Cn{async _call(F){return new Sr(await super._call(F))}}class Qn extends X{}class Pc extends Qn{}class Sc extends Qn{async _call(F){return new on(await super._call(F))}}class kc extends Qn{async _call(F){return new Dt(await super._call(F))}}class xp extends X{}class $c extends Is{}class Ic extends Is{async _call(F){return new on(await super._call(F))}}class Ac extends Is{async _call(F){return new Dt(await super._call(F))}}class an extends X{}class Oc extends an{}class Fc extends an{async _call(F){return new on(await super._call(F))}}class Dc extends an{async _call(F){return new Dt(await super._call(F))}}class jc extends an{async _call(F){return new $h(await super._call(F))}}class Lc extends an{async _call(F){return new Sr(await super._call(F))}}class zc extends X{}class Bc extends zc{}class Xn extends X{}class Tp extends Xn{}class Rc extends Xn{}class Nc extends Xn{async generate_speech(F,V,{threshold:_e=.5,minlenratio:Ie=0,maxlenratio:ke=20,vocoder:Be=null}={}){const Qe={input_ids:F},{encoder_outputs:rt,encoder_attention_mask:lt}=await le(this,Qe),Pt=rt.dims[1]/this.config.reduction_factor,It=Math.floor(Pt*ke),Ct=Math.floor(Pt*Ie),Bt=this.config.num_mel_bins;let kt=[],jt=null,_t=null,Wt=0;for(;;){++Wt;const mr=oe(!!_t);let hr;_t?hr=_t.output_sequence_out:hr=new g.Tensor("float32",new Float32Array(Bt),[1,1,Bt]);let wr={use_cache_branch:mr,output_sequence:hr,encoder_attention_mask:lt,speaker_embeddings:V,encoder_hidden_states:rt};this.addPastKeyValues(wr,jt),_t=await K(this.sessions.decoder_model_merged,wr),jt=this.getPastKeyValues(_t,jt);const{prob:kr,spectrum:Rt}=_t;if(kt.push(Rt),Wt>=Ct&&(Array.from(kr.data).filter(xr=>xr>=_e).length>0||Wt>=It))break}const nr=(0,g.cat)(kt),{waveform:ar}=await K(Be.sessions.model,{spectrogram:nr});return{spectrogram:nr,waveform:ar}}}class Vc extends X{constructor(){super(...arguments);de(this,"main_input_name","spectrogram")}}class Wc extends X{}class Uc extends Wc{}class ba extends X{}class Gc extends ba{}class Kc extends ba{}class Ma extends X{}class Hc extends Ma{}class qc extends Ma{}class va extends X{}class Qc extends va{}class Xc extends va{}class Jn extends X{}class Jc extends Jn{}class Yc extends Jn{static async from_pretrained(F,V={}){return super.from_pretrained(F,{...V,model_file_name:V.model_file_name??"text_model"})}}class Zc extends Jn{static async from_pretrained(F,V={}){return super.from_pretrained(F,{...V,model_file_name:V.model_file_name??"audio_model"})}}class eu extends X{}class xa extends eu{async _call(F){return new Ah(await super._call(F))}}class Yn extends X{}class Cp extends Yn{}class tu extends Yn{}class ru extends Yn{}class Ta extends X{}class su extends Ta{}class nu extends Ta{}class Ca extends X{}class iu extends Ca{}class au extends Ca{async _call(F){return new Dt(await super._call(F))}}class Ea extends X{}class Ep extends Ea{}class Pp extends Ea{}class Pa extends X{constructor(){super(...arguments);de(this,"forward_params",["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"])}_apply_and_filter_by_delay_pattern_mask(V){const[_e,Ie]=V.dims,ke=this.config.decoder.num_codebooks,Be=Ie-ke;let Qe=0;for(let Pt=0;Pt0&&Bt<=Be&&(V.data[Qe++]=V.data[Pt])}const rt=Math.floor(_e/ke),lt=Qe/(rt*ke);return new g.Tensor(V.type,V.data.slice(0,Qe),[rt,ke,lt])}prepare_inputs_for_generation(V,_e,Ie){let ke=structuredClone(V);for(let Qe=0;Qe=rt&&(ke[Qe][rt]=BigInt(this.config.decoder.pad_token_id));return Ie.guidance_scale!==null&&Ie.guidance_scale>1&&(ke=ke.concat(ke)),super.prepare_inputs_for_generation(ke,_e,Ie)}async generate(V){const _e=await super.generate(V),Ie=this._apply_and_filter_by_delay_pattern_mask(_e).unsqueeze_(0),{audio_values:ke}=await K(this.sessions.encodec_decode,{audio_codes:Ie});return ke}}class Zn extends X{}class ou extends Zn{}class lu extends Zn{async _call(F){return new Dt(await super._call(F))}}class cu extends Zn{}class ei extends X{}class uu extends ei{}class du extends ei{async _call(F){return new Dt(await super._call(F))}}class pu extends ei{}class ti extends X{}class hu extends ti{}class fu extends ti{async _call(F){return new Dt(await super._call(F))}}class mu extends ti{}class ri extends X{}class _u extends ri{}class gu extends ri{async _call(F){return new Dt(await super._call(F))}}class yu extends ri{}class wu extends X{}class bu extends wu{}class Mu extends X{}class vu extends Mu{constructor(...V){super(...V);de(this,"forward_params",["input_ids","pixel_values","images_seq_mask","images_emb_mask","attention_mask","position_ids","past_key_values"]);this._generation_mode="text"}async forward(V){const _e=this._generation_mode??"text";let Ie;if(_e==="text"||!V.past_key_values){const lt=this.sessions.prepare_inputs_embeds,Pt=(0,d.pick)(V,lt.inputNames);Ie=await K(lt,Pt)}else{const lt=this.sessions.gen_img_embeds,Pt=(0,d.pick)({image_ids:V.input_ids},lt.inputNames);Ie=await K(lt,Pt)}const ke={...V,...Ie},Be=await Ce(this,ke),Qe=this.sessions[_e==="text"?"lm_head":"gen_head"];if(!Qe)throw new Error(`Unable to find "${Qe}" generation head`);const rt=await K(Qe,(0,d.pick)(Be,Qe.inputNames));return{...Ie,...Be,...rt}}async generate(V){return this._generation_mode="text",super.generate(V)}async generate_images(V){this._generation_mode="image";const _e=(V.inputs??V[this.main_input_name]).dims[1],ke=(await super.generate(V)).slice(null,[_e,null]),Be=this.sessions.image_decode,{decoded_image:Qe}=await K(Be,{generated_tokens:ke}),rt=Qe.add_(1).mul_(255/2).clamp_(0,255).to("uint8"),lt=[];for(const Pt of rt){const It=T.RawImage.fromTensor(Pt);lt.push(It)}return lt}}class xu extends Oe{constructor({char_logits:F,bpe_logits:V,wp_logits:_e}){super(),this.char_logits=F,this.bpe_logits=V,this.wp_logits=_e}get logits(){return[this.char_logits,this.bpe_logits,this.wp_logits]}}class Tu extends X{}class Cu extends Tu{async _call(F){return new xu(await super._call(F))}}class Sa extends X{}class Eu extends Sa{}class Pu extends Sa{}class ka extends X{}class Su extends ka{}class ku extends ka{}class $u extends X{constructor(){super(...arguments);de(this,"forward_params",["input_ids","attention_mask","position_ids","audio_values","past_key_values"])}}class Iu extends $u{_merge_input_ids_with_audio_features(F){const V=F.audio_features.dims.at(-1),_e=F.audio_features.view(-1,V);return Q({audio_token_id:this.config.ignore_index,...F,audio_features:_e})}}class si extends X{constructor(){super(...arguments);de(this,"main_input_name","input_values");de(this,"forward_params",["input_values"])}}class Au extends Oe{constructor({audio_codes:F}){super(),this.audio_codes=F}}class Ou extends Oe{constructor({audio_values:F}){super(),this.audio_values=F}}class Fu extends si{async encode(F){return new Au(await K(this.sessions.encoder_model,F))}async decode(F){return new Ou(await K(this.sessions.decoder_model,F))}}class Du extends si{static async from_pretrained(F,V={}){return super.from_pretrained(F,{...V,model_file_name:V.model_file_name??"encoder_model"})}}class ju extends si{static async from_pretrained(F,V={}){return super.from_pretrained(F,{...V,model_file_name:V.model_file_name??"decoder_model"})}}class ni extends X{constructor(){super(...arguments);de(this,"main_input_name","input_values");de(this,"forward_params",["input_values"])}}class Lu extends Oe{constructor({audio_codes:F}){super(),this.audio_codes=F}}class zu extends Oe{constructor({audio_values:F}){super(),this.audio_values=F}}class Bu extends ni{async encode(F){return new Lu(await K(this.sessions.encoder_model,F))}async decode(F){return new zu(await K(this.sessions.decoder_model,F))}}class Ru extends ni{static async from_pretrained(F,V={}){return super.from_pretrained(F,{...V,model_file_name:V.model_file_name??"encoder_model"})}}class Nu extends ni{static async from_pretrained(F,V={}){return super.from_pretrained(F,{...V,model_file_name:V.model_file_name??"decoder_model"})}}class ii extends X{constructor(){super(...arguments);de(this,"main_input_name","input_values");de(this,"forward_params",["input_values"])}}class Vu extends ii{async encode(F){return await K(this.sessions.encoder_model,F)}async decode(F){return await K(this.sessions.decoder_model,F)}}class Wu extends ii{static async from_pretrained(F,V={}){return super.from_pretrained(F,{...V,model_file_name:V.model_file_name??"encoder_model"})}}class Uu extends ii{static async from_pretrained(F,V={}){return super.from_pretrained(F,{...V,model_file_name:V.model_file_name??"decoder_model"})}}class Ut{static async from_pretrained(F,{progress_callback:V=null,config:_e=null,cache_dir:Ie=null,local_files_only:ke=!1,revision:Be="main",model_file_name:Qe=null,subfolder:rt="onnx",device:lt=null,dtype:Pt=null,use_external_data_format:It=null,session_options:Ct={}}={}){const Bt={progress_callback:V,config:_e,cache_dir:Ie,local_files_only:ke,revision:Be,model_file_name:Qe,subfolder:rt,device:lt,dtype:Pt,use_external_data_format:It,session_options:Ct};if(Bt.config=await i.AutoConfig.from_pretrained(F,Bt),!this.MODEL_CLASS_MAPPINGS)throw new Error("`MODEL_CLASS_MAPPINGS` not implemented for this type of `AutoClass`: "+this.name);const kt=Bt.config.model_type;for(const jt of this.MODEL_CLASS_MAPPINGS){let _t=jt.get(kt);if(!_t){for(const Wt of jt.values())if(Wt[0]===kt){_t=Wt;break}if(!_t)continue}return await _t[1].from_pretrained(F,Bt)}if(this.BASE_IF_FAIL)return Zp.has(kt)||console.warn(`Unknown model class "${kt}", attempting to construct from base class.`),await X.from_pretrained(F,Bt);throw Error(`Unsupported model type: ${kt}`)}}de(Ut,"MODEL_CLASS_MAPPINGS",null),de(Ut,"BASE_IF_FAIL",!1);const df=new Map([["bert",["BertModel",Re]],["modernbert",["ModernBertModel",Ke]],["nomic_bert",["NomicBertModel",vt]],["roformer",["RoFormerModel",xt]],["electra",["ElectraModel",$e]],["esm",["EsmModel",bt]],["convbert",["ConvBertModel",ht]],["camembert",["CamembertModel",Nt]],["deberta",["DebertaModel",Kr]],["deberta-v2",["DebertaV2Model",tn]],["mpnet",["MPNetModel",_n]],["albert",["AlbertModel",Te]],["distilbert",["DistilBertModel",Jt]],["roberta",["RobertaModel",Ar]],["xlm",["XLMModel",bi]],["xlm-roberta",["XLMRobertaModel",Ua]],["clap",["ClapModel",Jc]],["clip",["CLIPModel",to]],["clipseg",["CLIPSegModel",Vd]],["chinese_clip",["ChineseCLIPModel",ao]],["siglip",["SiglipModel",io]],["jina_clip",["JinaCLIPModel",Rd]],["mobilebert",["MobileBertModel",Ps]],["squeezebert",["SqueezeBertModel",rn]],["wav2vec2",["Wav2Vec2Model",pc]],["wav2vec2-bert",["Wav2Vec2BertModel",Pc]],["unispeech",["UniSpeechModel",bc]],["unispeech-sat",["UniSpeechSatModel",xc]],["hubert",["HubertModel",$c]],["wavlm",["WavLMModel",Oc]],["audio-spectrogram-transformer",["ASTModel",Ed]],["vits",["VitsModel",xa]],["pyannote",["PyAnnoteModel",_c]],["wespeaker-resnet",["WeSpeakerResNetModel",wc]],["detr",["DetrModel",tl]],["rt_detr",["RTDetrModel",nl]],["rt_detr_v2",["RTDetrV2Model",il]],["rf_detr",["RFDetrModel",ll]],["d_fine",["DFineModel",dl]],["table-transformer",["TableTransformerModel",hl]],["vit",["ViTModel",jo]],["ijepa",["IJepaModel",zo]],["pvt",["PvtModel",Vo]],["vit_msn",["ViTMSNModel",Ni]],["vit_mae",["ViTMAEModel",Bi]],["groupvit",["GroupViTModel",xn]],["fastvit",["FastViTModel",Go]],["mobilevit",["MobileViTModel",Ki]],["mobilevitv2",["MobileViTV2Model",qi]],["owlvit",["OwlViTModel",Qo]],["owlv2",["Owlv2Model",Jo]],["beit",["BeitModel",Zo]],["deit",["DeiTModel",_l]],["hiera",["HieraModel",yl]],["convnext",["ConvNextModel",Kl]],["convnextv2",["ConvNextV2Model",ql]],["dinov2",["Dinov2Model",Xl]],["dinov2_with_registers",["Dinov2WithRegistersModel",Yl]],["resnet",["ResNetModel",bl]],["swin",["SwinModel",vl]],["swin2sr",["Swin2SRModel",Cl]],["donut-swin",["DonutSwinModel",Gl]],["yolos",["YolosModel",rc]],["dpt",["DPTModel",Pl]],["glpn",["GLPNModel",Vl]],["hifigan",["SpeechT5HifiGan",Vc]],["efficientnet",["EfficientNetModel",iu]],["decision_transformer",["DecisionTransformerModel",bu]],["patchtst",["PatchTSTForPrediction",Eu]],["patchtsmixer",["PatchTSMixerForPrediction",Su]],["mobilenet_v1",["MobileNetV1Model",ou]],["mobilenet_v2",["MobileNetV2Model",uu]],["mobilenet_v3",["MobileNetV3Model",hu]],["mobilenet_v4",["MobileNetV4Model",_u]],["maskformer",["MaskFormerModel",Rl]],["mgp-str",["MgpstrForSceneTextRecognition",Cu]],["style_text_to_speech_2",["StyleTextToSpeech2Model",Bc]]]),pf=new Map([["t5",["T5Model",be]],["longt5",["LongT5Model",ct]],["mt5",["MT5Model",At]],["bart",["BartModel",fr]],["mbart",["MBartModel",qr]],["marian",["MarianModel",lc]],["whisper",["WhisperModel",Pd]],["m2m_100",["M2M100Model",uc]],["blenderbot",["BlenderbotModel",Ir]],["blenderbot-small",["BlenderbotSmallModel",ss]]]),hf=new Map([["mimi",["MimiModel",Fu]],["dac",["DacModel",Bu]],["snac",["SnacModel",Vu]]]),ff=new Map([["bloom",["BloomModel",Fo]],["jais",["JAISModel",Gd]],["gpt2",["GPT2Model",Wd]],["gptj",["GPTJModel",ps]],["gpt_bigcode",["GPTBigCodeModel",qd]],["gpt_neo",["GPTNeoModel",Bn]],["gpt_neox",["GPTNeoXModel",ho]],["codegen",["CodeGenModel",Xd]],["llama",["LlamaModel",Yd]],["exaone",["ExaoneModel",wo]],["olmo",["OlmoModel",xo]],["olmo2",["Olmo2Model",np]],["mobilellm",["MobileLLMModel",Mo]],["granite",["GraniteModel",ap]],["cohere",["CohereModel",lp]],["gemma",["GemmaModel",up]],["gemma2",["Gemma2Model",pp]],["gemma3_text",["Gemma3Model",fp]],["helium",["HeliumModel",ep]],["glm",["GlmModel",Pi]],["openelm",["OpenELMModel",_p]],["qwen2",["Qwen2Model",yp]],["qwen3",["Qwen3Model",bp]],["phi",["PhiModel",Oo]],["phi3",["Phi3Model",Zr]],["mpt",["MptModel",Fi]],["opt",["OPTModel",ji]],["mistral",["MistralModel",Gc]],["starcoder2",["Starcoder2Model",Hc]],["falcon",["FalconModel",Qc]],["stablelm",["StableLmModel",su]]]),Gu=new Map([["speecht5",["SpeechT5ForSpeechToText",Rc]],["whisper",["WhisperForConditionalGeneration",qa]],["lite-whisper",["LiteWhisperForConditionalGeneration",Qa]],["moonshine",["MoonshineForConditionalGeneration",Sd]]]),Sp=new Map([["speecht5",["SpeechT5ForTextToSpeech",Nc]]]),kp=new Map([["vits",["VitsModel",xa]],["musicgen",["MusicgenForConditionalGeneration",Pa]]]),$p=new Map([["bert",["BertForSequenceClassification",Ge]],["modernbert",["ModernBertForSequenceClassification",ft]],["roformer",["RoFormerForSequenceClassification",he]],["electra",["ElectraForSequenceClassification",Le]],["esm",["EsmForSequenceClassification",ls]],["convbert",["ConvBertForSequenceClassification",W]],["camembert",["CamembertForSequenceClassification",tr]],["deberta",["DebertaForSequenceClassification",Ws]],["deberta-v2",["DebertaV2ForSequenceClassification",Us]],["mpnet",["MPNetForSequenceClassification",Fn]],["albert",["AlbertForSequenceClassification",L]],["distilbert",["DistilBertForSequenceClassification",Cs]],["roberta",["RobertaForSequenceClassification",Mr]],["xlm",["XLMForSequenceClassification",jn]],["xlm-roberta",["XLMRobertaForSequenceClassification",Td]],["bart",["BartForSequenceClassification",Pr]],["mbart",["MBartForSequenceClassification",Ur]],["mobilebert",["MobileBertForSequenceClassification",ks]],["squeezebert",["SqueezeBertForSequenceClassification",Jr]]]),Ip=new Map([["bert",["BertForTokenClassification",Je]],["modernbert",["ModernBertForTokenClassification",yt]],["roformer",["RoFormerForTokenClassification",ce]],["electra",["ElectraForTokenClassification",Ft]],["esm",["EsmForTokenClassification",Es]],["convbert",["ConvBertForTokenClassification",ye]],["camembert",["CamembertForTokenClassification",rr]],["deberta",["DebertaForTokenClassification",Ts]],["deberta-v2",["DebertaV2ForTokenClassification",Gs]],["mpnet",["MPNetForTokenClassification",Dn]],["distilbert",["DistilBertForTokenClassification",Hs]],["roberta",["RobertaForTokenClassification",vr]],["xlm",["XLMForTokenClassification",Xs]],["xlm-roberta",["XLMRobertaForTokenClassification",Cd]]]),Ku=new Map([["t5",["T5ForConditionalGeneration",ze]],["longt5",["LongT5ForConditionalGeneration",it]],["mt5",["MT5ForConditionalGeneration",qt]],["bart",["BartForConditionalGeneration",sr]],["mbart",["MBartForConditionalGeneration",ws]],["marian",["MarianMTModel",cc]],["m2m_100",["M2M100ForConditionalGeneration",dc]],["blenderbot",["BlenderbotForConditionalGeneration",zr]],["blenderbot-small",["BlenderbotSmallForConditionalGeneration",Br]]]),Hu=new Map([["bloom",["BloomForCausalLM",Oi]],["gpt2",["GPT2LMHeadModel",Ud]],["jais",["JAISLMHeadModel",Kd]],["gptj",["GPTJForCausalLM",Pe]],["gpt_bigcode",["GPTBigCodeForCausalLM",Qd]],["gpt_neo",["GPTNeoForCausalLM",bn]],["gpt_neox",["GPTNeoXForCausalLM",Hd]],["codegen",["CodeGenForCausalLM",Jd]],["llama",["LlamaForCausalLM",Zd]],["exaone",["ExaoneForCausalLM",bo]],["olmo",["OlmoForCausalLM",sp]],["olmo2",["Olmo2ForCausalLM",ip]],["mobilellm",["MobileLLMForCausalLM",rp]],["granite",["GraniteForCausalLM",op]],["cohere",["CohereForCausalLM",cp]],["gemma",["GemmaForCausalLM",dp]],["gemma2",["Gemma2ForCausalLM",hp]],["gemma3_text",["Gemma3ForCausalLM",mp]],["helium",["HeliumForCausalLM",tp]],["glm",["GlmForCausalLM",Mn]],["openelm",["OpenELMForCausalLM",gp]],["qwen2",["Qwen2ForCausalLM",wp]],["qwen3",["Qwen3ForCausalLM",Mp]],["phi",["PhiForCausalLM",Ii]],["phi3",["Phi3ForCausalLM",nn]],["mpt",["MptForCausalLM",Di]],["opt",["OPTForCausalLM",Do]],["mbart",["MBartForCausalLM",Qs]],["mistral",["MistralForCausalLM",Kc]],["starcoder2",["Starcoder2ForCausalLM",qc]],["falcon",["FalconForCausalLM",Xc]],["trocr",["TrOCRForCausalLM",Uc]],["stablelm",["StableLmForCausalLM",nu]],["phi3_v",["Phi3VForCausalLM",eo]]]),mf=new Map([["multi_modality",["MultiModalityCausalLM",vu]]]),Ap=new Map([["bert",["BertForMaskedLM",Ue]],["modernbert",["ModernBertForMaskedLM",st]],["roformer",["RoFormerForMaskedLM",Vt]],["electra",["ElectraForMaskedLM",Xe]],["esm",["EsmForMaskedLM",cr]],["convbert",["ConvBertForMaskedLM",as]],["camembert",["CamembertForMaskedLM",mt]],["deberta",["DebertaForMaskedLM",xs]],["deberta-v2",["DebertaV2ForMaskedLM",os]],["mpnet",["MPNetForMaskedLM",gn]],["albert",["AlbertForMaskedLM",ie]],["distilbert",["DistilBertForMaskedLM",qe]],["roberta",["RobertaForMaskedLM",yr]],["xlm",["XLMWithLMHeadModel",xd]],["xlm-roberta",["XLMRobertaForMaskedLM",Ln]],["mobilebert",["MobileBertForMaskedLM",Ss]],["squeezebert",["SqueezeBertForMaskedLM",cs]]]),Op=new Map([["bert",["BertForQuestionAnswering",at]],["roformer",["RoFormerForQuestionAnswering",Cr]],["electra",["ElectraForQuestionAnswering",Qt]],["convbert",["ConvBertForQuestionAnswering",G]],["camembert",["CamembertForQuestionAnswering",Fr]],["deberta",["DebertaForQuestionAnswering",en]],["deberta-v2",["DebertaV2ForQuestionAnswering",Ks]],["mpnet",["MPNetForQuestionAnswering",Hr]],["albert",["AlbertForQuestionAnswering",J]],["distilbert",["DistilBertForQuestionAnswering",Er]],["roberta",["RobertaForQuestionAnswering",Rr]],["xlm",["XLMForQuestionAnswering",ds]],["xlm-roberta",["XLMRobertaForQuestionAnswering",zn]],["mobilebert",["MobileBertForQuestionAnswering",$s]],["squeezebert",["SqueezeBertForQuestionAnswering",yn]]]),qu=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",Ja]],["idefics3",["Idefics3ForConditionalGeneration",vi]],["smolvlm",["SmolVLMForConditionalGeneration",Ya]]]),Fp=new Map([["llava",["LlavaForConditionalGeneration",Mi]],["llava_onevision",["LlavaOnevisionForConditionalGeneration",$d]],["moondream1",["Moondream1ForConditionalGeneration",Id]],["florence2",["Florence2ForConditionalGeneration",Od]],["qwen2-vl",["Qwen2VLForConditionalGeneration",Y]],["idefics3",["Idefics3ForConditionalGeneration",vi]],["smolvlm",["SmolVLMForConditionalGeneration",Ya]],["paligemma",["PaliGemmaForConditionalGeneration",Dd]]]),Dp=new Map([["ultravox",["UltravoxModel",Iu]]]),_f=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",Ja]]]),jp=new Map([["vit",["ViTForImageClassification",Lo]],["ijepa",["IJepaForImageClassification",Bo]],["pvt",["PvtForImageClassification",Wo]],["vit_msn",["ViTMSNForImageClassification",Vi]],["fastvit",["FastViTForImageClassification",Ui]],["mobilevit",["MobileViTForImageClassification",qo]],["mobilevitv2",["MobileViTV2ForImageClassification",Me]],["beit",["BeitForImageClassification",el]],["deit",["DeiTForImageClassification",gl]],["hiera",["HieraForImageClassification",wl]],["convnext",["ConvNextForImageClassification",Hl]],["convnextv2",["ConvNextV2ForImageClassification",Ql]],["dinov2",["Dinov2ForImageClassification",Jl]],["dinov2_with_registers",["Dinov2WithRegistersForImageClassification",Zl]],["resnet",["ResNetForImageClassification",Ml]],["swin",["SwinForImageClassification",xl]],["segformer",["SegformerForImageClassification",tu]],["efficientnet",["EfficientNetForImageClassification",au]],["mobilenet_v1",["MobileNetV1ForImageClassification",lu]],["mobilenet_v2",["MobileNetV2ForImageClassification",du]],["mobilenet_v3",["MobileNetV3ForImageClassification",fu]],["mobilenet_v4",["MobileNetV4ForImageClassification",gu]]]),Lp=new Map([["detr",["DetrForObjectDetection",rl]],["rt_detr",["RTDetrForObjectDetection",ea]],["rt_detr_v2",["RTDetrV2ForObjectDetection",al]],["rf_detr",["RFDetrForObjectDetection",cl]],["d_fine",["DFineForObjectDetection",pl]],["table-transformer",["TableTransformerForObjectDetection",fl]],["yolos",["YolosForObjectDetection",sc]]]),zp=new Map([["owlvit",["OwlViTForObjectDetection",Xo]],["owlv2",["Owlv2ForObjectDetection",Yo]],["grounding-dino",["GroundingDinoForObjectDetection",tc]]]),En=new Map([["detr",["DetrForSegmentation",Gn]],["clipseg",["CLIPSegForImageSegmentation",lo]]]),Bp=new Map([["segformer",["SegformerForSemanticSegmentation",ru]],["sapiens",["SapiensForSemanticSegmentation",Il]],["swin",["SwinForSemanticSegmentation",Tl]],["mobilenet_v1",["MobileNetV1ForSemanticSegmentation",cu]],["mobilenet_v2",["MobileNetV2ForSemanticSegmentation",pu]],["mobilenet_v3",["MobileNetV3ForSemanticSegmentation",mu]],["mobilenet_v4",["MobileNetV4ForSemanticSegmentation",yu]]]),Rp=new Map([["detr",["DetrForSegmentation",Gn]],["maskformer",["MaskFormerForInstanceSegmentation",Nl]]]),Np=new Map([["sam",["SamModel",ac]]]),Vp=new Map([["wav2vec2",["Wav2Vec2ForCTC",hc]],["wav2vec2-bert",["Wav2Vec2BertForCTC",Sc]],["unispeech",["UniSpeechForCTC",Mc]],["unispeech-sat",["UniSpeechSatForCTC",Tc]],["wavlm",["WavLMForCTC",Fc]],["hubert",["HubertForCTC",Ic]]]),Wp=new Map([["wav2vec2",["Wav2Vec2ForSequenceClassification",fc]],["wav2vec2-bert",["Wav2Vec2BertForSequenceClassification",kc]],["unispeech",["UniSpeechForSequenceClassification",vc]],["unispeech-sat",["UniSpeechSatForSequenceClassification",Cc]],["wavlm",["WavLMForSequenceClassification",Dc]],["hubert",["HubertForSequenceClassification",Ac]],["audio-spectrogram-transformer",["ASTForAudioClassification",Ka]]]),Up=new Map([["wavlm",["WavLMForXVector",jc]]]),Gp=new Map([["unispeech-sat",["UniSpeechSatForAudioFrameClassification",Ec]],["wavlm",["WavLMForAudioFrameClassification",Lc]],["wav2vec2",["Wav2Vec2ForAudioFrameClassification",mc]],["pyannote",["PyAnnoteForAudioFrameClassification",gc]]]),Kp=new Map([["vitmatte",["VitMatteForImageMatting",Ho]]]),gf=new Map([["patchtst",["PatchTSTForPrediction",Pu]],["patchtsmixer",["PatchTSMixerForPrediction",ku]]]),Hp=new Map([["swin2sr",["Swin2SRForImageSuperResolution",El]]]),qp=new Map([["dpt",["DPTForDepthEstimation",Sl]],["depth_anything",["DepthAnythingForDepthEstimation",$l]],["glpn",["GLPNForDepthEstimation",Wl]],["sapiens",["SapiensForDepthEstimation",Al]],["depth_pro",["DepthProForDepthEstimation",Dl]],["metric3d",["Metric3DForDepthEstimation",Ll]],["metric3dv2",["Metric3Dv2ForDepthEstimation",Bl]]]),Qp=new Map([["sapiens",["SapiensForNormalEstimation",Ol]]]),Xp=new Map([["vitpose",["VitPoseForPoseEstimation",No]]]),Jp=new Map([["clip",["CLIPVisionModelWithProjection",so]],["siglip",["SiglipVisionModel",Bd]],["jina_clip",["JinaCLIPVisionModel",Nd]]]),Yp=[[df,A.EncoderOnly],[pf,A.EncoderDecoder],[ff,A.DecoderOnly],[hf,A.AutoEncoder],[$p,A.EncoderOnly],[Ip,A.EncoderOnly],[Ku,A.Seq2Seq],[Gu,A.Seq2Seq],[Hu,A.DecoderOnly],[mf,A.MultiModality],[Ap,A.EncoderOnly],[Op,A.EncoderOnly],[qu,A.Vision2Seq],[Fp,A.ImageTextToText],[Dp,A.AudioTextToText],[jp,A.EncoderOnly],[En,A.EncoderOnly],[Rp,A.EncoderOnly],[Bp,A.EncoderOnly],[Kp,A.EncoderOnly],[gf,A.EncoderOnly],[Hp,A.EncoderOnly],[qp,A.EncoderOnly],[Qp,A.EncoderOnly],[Xp,A.EncoderOnly],[Lp,A.EncoderOnly],[zp,A.EncoderOnly],[Np,A.MaskGeneration],[Vp,A.EncoderOnly],[Wp,A.EncoderOnly],[Sp,A.Seq2Seq],[kp,A.EncoderOnly],[Up,A.EncoderOnly],[Gp,A.EncoderOnly],[Jp,A.EncoderOnly]];for(const[S,F]of Yp)for(const[V,_e]of S.values())$.set(V,F),k.set(_e,V),P.set(V,_e);const yf=[["MusicgenForConditionalGeneration",Pa,A.Musicgen],["Phi3VForCausalLM",eo,A.Phi3V],["CLIPTextModelWithProjection",wn,A.EncoderOnly],["SiglipTextModel",zd,A.EncoderOnly],["JinaCLIPTextModel",Js,A.EncoderOnly],["ClapTextModelWithProjection",Yc,A.EncoderOnly],["ClapAudioModelWithProjection",Zc,A.EncoderOnly],["DacEncoderModel",Ru,A.EncoderOnly],["DacDecoderModel",Nu,A.EncoderOnly],["MimiEncoderModel",Du,A.EncoderOnly],["MimiDecoderModel",ju,A.EncoderOnly],["SnacEncoderModel",Wu,A.EncoderOnly],["SnacDecoderModel",Uu,A.EncoderOnly]];for(const[S,F,V]of yf)$.set(S,V),k.set(F,S),P.set(S,F);const Zp=new Map([["modnet",En],["birefnet",En],["isnet",En],["ben",En]]);for(const[S,F]of Zp.entries())F.set(S,["PreTrainedModel",X]),$.set(S,A.EncoderOnly),k.set(X,S),P.set(S,X);class Qu extends Ut{}de(Qu,"MODEL_CLASS_MAPPINGS",Yp.map(F=>F[0])),de(Qu,"BASE_IF_FAIL",!0);class eh extends Ut{}de(eh,"MODEL_CLASS_MAPPINGS",[$p]);class th extends Ut{}de(th,"MODEL_CLASS_MAPPINGS",[Ip]);class rh extends Ut{}de(rh,"MODEL_CLASS_MAPPINGS",[Ku]);class sh extends Ut{}de(sh,"MODEL_CLASS_MAPPINGS",[Gu]);class nh extends Ut{}de(nh,"MODEL_CLASS_MAPPINGS",[Sp]);class ih extends Ut{}de(ih,"MODEL_CLASS_MAPPINGS",[kp]);class ah extends Ut{}de(ah,"MODEL_CLASS_MAPPINGS",[Hu]);class oh extends Ut{}de(oh,"MODEL_CLASS_MAPPINGS",[Ap]);class lh extends Ut{}de(lh,"MODEL_CLASS_MAPPINGS",[Op]);class ch extends Ut{}de(ch,"MODEL_CLASS_MAPPINGS",[qu]);class uh extends Ut{}de(uh,"MODEL_CLASS_MAPPINGS",[jp]);class dh extends Ut{}de(dh,"MODEL_CLASS_MAPPINGS",[En]);class ph extends Ut{}de(ph,"MODEL_CLASS_MAPPINGS",[Bp]);class hh extends Ut{}de(hh,"MODEL_CLASS_MAPPINGS",[Rp]);class fh extends Ut{}de(fh,"MODEL_CLASS_MAPPINGS",[Lp]);class mh extends Ut{}de(mh,"MODEL_CLASS_MAPPINGS",[zp]);class _h extends Ut{}de(_h,"MODEL_CLASS_MAPPINGS",[Np]);class gh extends Ut{}de(gh,"MODEL_CLASS_MAPPINGS",[Vp]);class yh extends Ut{}de(yh,"MODEL_CLASS_MAPPINGS",[Wp]);class wh extends Ut{}de(wh,"MODEL_CLASS_MAPPINGS",[Up]);class bh extends Ut{}de(bh,"MODEL_CLASS_MAPPINGS",[Gp]);class Mh extends Ut{}de(Mh,"MODEL_CLASS_MAPPINGS",[_f]);class vh extends Ut{}de(vh,"MODEL_CLASS_MAPPINGS",[Kp]);class xh extends Ut{}de(xh,"MODEL_CLASS_MAPPINGS",[Hp]);class Th extends Ut{}de(Th,"MODEL_CLASS_MAPPINGS",[qp]);class Ch extends Ut{}de(Ch,"MODEL_CLASS_MAPPINGS",[Qp]);class Eh extends Ut{}de(Eh,"MODEL_CLASS_MAPPINGS",[Xp]);class Ph extends Ut{}de(Ph,"MODEL_CLASS_MAPPINGS",[Jp]);class Sh extends Ut{}de(Sh,"MODEL_CLASS_MAPPINGS",[Fp]);class kh extends Ut{}de(kh,"MODEL_CLASS_MAPPINGS",[Dp]);class wf extends Oe{constructor({logits:F,past_key_values:V,encoder_outputs:_e,decoder_attentions:Ie=null,cross_attentions:ke=null}){super(),this.logits=F,this.past_key_values=V,this.encoder_outputs=_e,this.decoder_attentions=Ie,this.cross_attentions=ke}}class Dt extends Oe{constructor({logits:F,...V}){super(),this.logits=F;const _e=Object.values(V);_e.length>0&&(this.attentions=_e)}}class $h extends Oe{constructor({logits:F,embeddings:V}){super(),this.logits=F,this.embeddings=V}}class Sr extends Oe{constructor({logits:F}){super(),this.logits=F}}class Or extends Oe{constructor({logits:F}){super(),this.logits=F}}class Nr extends Oe{constructor({start_logits:F,end_logits:V}){super(),this.start_logits=F,this.end_logits=V}}class on extends Oe{constructor({logits:F}){super(),this.logits=F}}class bf extends Oe{constructor({logits:F,past_key_values:V}){super(),this.logits=F,this.past_key_values=V}}class Ih extends Oe{constructor({alphas:F}){super(),this.alphas=F}}class Ah extends Oe{constructor({waveform:F,spectrogram:V}){super(),this.waveform=F,this.spectrogram=V}}},"./src/models/audio_spectrogram_transformer/feature_extraction_audio_spectrogram_transformer.js":(e,r,t)=>{t.r(r),t.d(r,{ASTFeatureExtractor:()=>n});var i=t("./src/base/feature_extraction_utils.js");t("./src/utils/tensor.js");var o=t("./src/utils/audio.js");class n extends i.FeatureExtractor{constructor(d){super(d);const h=this.config.sampling_rate,f=(0,o.mel_filter_bank)(257,this.config.num_mel_bins,20,Math.floor(h/2),h,null,"kaldi",!0);this.mel_filters=f,this.window=(0,o.window_function)(400,"hann",{periodic:!1}),this.mean=this.config.mean,this.std=this.config.std}async _extract_fbank_features(d,h){return(0,o.spectrogram)(d,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1192092955078125e-22,remove_dc_offset:!0,max_num_frames:h,transpose:!0})}async _call(d){(0,i.validate_audio_inputs)(d,"ASTFeatureExtractor");const h=await this._extract_fbank_features(d,this.config.max_length);if(this.config.do_normalize){const f=this.std*2,y=h.data;for(let m=0;m{t.r(r),t.d(r,{AutoFeatureExtractor:()=>u});var i=t("./src/utils/constants.js"),o=t("./src/utils/hub.js");t("./src/base/feature_extraction_utils.js");var n=t("./src/models/feature_extractors.js");class u{static async from_pretrained(h,f={}){const y=await(0,o.getModelJSON)(h,i.FEATURE_EXTRACTOR_NAME,!0,f),m=y.feature_extractor_type,g=n[m];if(!g)throw new Error(`Unknown feature_extractor_type: '${m}'. Please report this at ${i.GITHUB_ISSUE_URL}.`);return new g(y)}}},"./src/models/auto/image_processing_auto.js":(e,r,t)=>{t.r(r),t.d(r,{AutoImageProcessor:()=>d});var i=t("./src/utils/constants.js"),o=t("./src/utils/hub.js"),n=t("./src/base/image_processors_utils.js"),u=t("./src/models/image_processors.js");class d{static async from_pretrained(f,y={}){const m=await(0,o.getModelJSON)(f,i.IMAGE_PROCESSOR_NAME,!0,y),g=m.image_processor_type??m.feature_extractor_type;let T=u[g];return T||(g!==void 0&&console.warn(`Image processor type '${g}' not found, assuming base ImageProcessor. Please report this at ${i.GITHUB_ISSUE_URL}.`),T=n.ImageProcessor),new T(m)}}},"./src/models/auto/processing_auto.js":(e,r,t)=>{t.r(r),t.d(r,{AutoProcessor:()=>f});var i=t("./src/utils/constants.js"),o=t("./src/utils/hub.js"),n=t("./src/base/processing_utils.js"),u=t("./src/models/processors.js"),d=t("./src/models/image_processors.js"),h=t("./src/models/feature_extractors.js");class f{static async from_pretrained(m,g={}){const T=await(0,o.getModelJSON)(m,i.IMAGE_PROCESSOR_NAME,!0,g),{image_processor_type:M,feature_extractor_type:I,processor_class:z}=T;if(z&&u[z])return u[z].from_pretrained(m,g);if(!M&&!I)throw new Error("No `image_processor_type` or `feature_extractor_type` found in the config.");const E={};if(M){const D=d[M];if(!D)throw new Error(`Unknown image_processor_type: '${M}'.`);E.image_processor=new D(T)}if(I){const D=d[I];if(D)E.image_processor=new D(T);else{const A=h[I];if(!A)throw new Error(`Unknown feature_extractor_type: '${I}'.`);E.feature_extractor=new A(T)}}const C={};return new n.Processor(C,E)}}},"./src/models/beit/image_processing_beit.js":(e,r,t)=>{t.r(r),t.d(r,{BeitFeatureExtractor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{}},"./src/models/bit/image_processing_bit.js":(e,r,t)=>{t.r(r),t.d(r,{BitImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{}},"./src/models/chinese_clip/image_processing_chinese_clip.js":(e,r,t)=>{t.r(r),t.d(r,{ChineseCLIPFeatureExtractor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{}},"./src/models/clap/feature_extraction_clap.js":(e,r,t)=>{t.r(r),t.d(r,{ClapFeatureExtractor:()=>n});var i=t("./src/base/feature_extraction_utils.js");t("./src/utils/tensor.js");var o=t("./src/utils/audio.js");class n extends i.FeatureExtractor{constructor(d){super(d),this.mel_filters=(0,o.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,null,"htk"),this.mel_filters_slaney=(0,o.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,"slaney","slaney"),this.window=(0,o.window_function)(this.config.fft_window_size,"hann")}async _get_input_mel(d,h,f,y){let m;const g=d.length-h;if(g>0)if(f==="rand_trunc"){const T=Math.floor(Math.random()*(g+1));d=d.subarray(T,T+h),m=await this._extract_fbank_features(d,this.mel_filters_slaney,this.config.nb_max_samples)}else throw new Error(`Truncation strategy "${f}" not implemented`);else{if(g<0){let T=new Float64Array(h);if(T.set(d),y==="repeat")for(let M=d.length;M{t.r(r),t.d(r,{CLIPFeatureExtractor:()=>n,CLIPImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{}class n extends o{}},"./src/models/convnext/image_processing_convnext.js":(e,r,t)=>{t.r(r),t.d(r,{ConvNextFeatureExtractor:()=>n,ConvNextImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{constructor(d){super(d),this.crop_pct=this.config.crop_pct??224/256}async resize(d){var f;const h=(f=this.size)==null?void 0:f.shortest_edge;if(h===void 0)throw new Error("Size dictionary must contain 'shortest_edge' key.");if(h<384){const y=Math.floor(h/this.crop_pct),[m,g]=this.get_resize_output_image_size(d,{shortest_edge:y});d=await d.resize(m,g,{resample:this.resample}),d=await d.center_crop(h,h)}else d=await d.resize(h,h,{resample:this.resample});return d}}class n extends o{}},"./src/models/dac/feature_extraction_dac.js":(e,r,t)=>{t.r(r),t.d(r,{DacFeatureExtractor:()=>o});var i=t("./src/models/encodec/feature_extraction_encodec.js");class o extends i.EncodecFeatureExtractor{}},"./src/models/deit/image_processing_deit.js":(e,r,t)=>{t.r(r),t.d(r,{DeiTFeatureExtractor:()=>n,DeiTImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{}class n extends o{}},"./src/models/detr/image_processing_detr.js":(e,r,t)=>{t.r(r),t.d(r,{DetrFeatureExtractor:()=>u,DetrImageProcessor:()=>n});var i=t("./src/base/image_processors_utils.js"),o=t("./src/utils/tensor.js");class n extends i.ImageProcessor{async _call(h){const f=await super._call(h),y=[f.pixel_values.dims[0],64,64],m=(0,o.full)(y,1n);return{...f,pixel_mask:m}}post_process_object_detection(...h){return(0,i.post_process_object_detection)(...h)}post_process_panoptic_segmentation(...h){return(0,i.post_process_panoptic_segmentation)(...h)}post_process_instance_segmentation(...h){return(0,i.post_process_instance_segmentation)(...h)}}class u extends n{}},"./src/models/donut/image_processing_donut.js":(e,r,t)=>{t.r(r),t.d(r,{DonutFeatureExtractor:()=>n,DonutImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{pad_image(d,h,f,y={}){const[m,g,T]=h;let M=this.image_mean;Array.isArray(this.image_mean)||(M=new Array(T).fill(M));let I=this.image_std;Array.isArray(I)||(I=new Array(T).fill(M));const z=M.map((E,C)=>-E/I[C]);return super.pad_image(d,h,f,{center:!0,constant_values:z,...y})}}class n extends o{}},"./src/models/dpt/image_processing_dpt.js":(e,r,t)=>{t.r(r),t.d(r,{DPTFeatureExtractor:()=>n,DPTImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{}class n extends o{}},"./src/models/efficientnet/image_processing_efficientnet.js":(e,r,t)=>{t.r(r),t.d(r,{EfficientNetImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{constructor(u){super(u),this.include_top=this.config.include_top??!0,this.include_top&&(this.image_std=this.image_std.map(d=>d*d))}}},"./src/models/encodec/feature_extraction_encodec.js":(e,r,t)=>{t.r(r),t.d(r,{EncodecFeatureExtractor:()=>n});var i=t("./src/base/feature_extraction_utils.js"),o=t("./src/utils/tensor.js");class n extends i.FeatureExtractor{async _call(d){(0,i.validate_audio_inputs)(d,"EncodecFeatureExtractor"),d instanceof Float64Array&&(d=new Float32Array(d));const h=this.config.feature_size;if(d.length%h!==0)throw new Error(`The length of the audio data must be a multiple of the number of channels (${h}).`);const f=[1,h,d.length/h];return{input_values:new o.Tensor("float32",d,f)}}}},"./src/models/feature_extractors.js":(e,r,t)=>{t.r(r),t.d(r,{ASTFeatureExtractor:()=>i.ASTFeatureExtractor,ClapFeatureExtractor:()=>n.ClapFeatureExtractor,DacFeatureExtractor:()=>u.DacFeatureExtractor,EncodecFeatureExtractor:()=>o.EncodecFeatureExtractor,ImageFeatureExtractor:()=>I.ImageProcessor,MoonshineFeatureExtractor:()=>d.MoonshineFeatureExtractor,PyAnnoteFeatureExtractor:()=>h.PyAnnoteFeatureExtractor,SeamlessM4TFeatureExtractor:()=>f.SeamlessM4TFeatureExtractor,SnacFeatureExtractor:()=>y.SnacFeatureExtractor,SpeechT5FeatureExtractor:()=>m.SpeechT5FeatureExtractor,Wav2Vec2FeatureExtractor:()=>g.Wav2Vec2FeatureExtractor,WeSpeakerFeatureExtractor:()=>T.WeSpeakerFeatureExtractor,WhisperFeatureExtractor:()=>M.WhisperFeatureExtractor});var i=t("./src/models/audio_spectrogram_transformer/feature_extraction_audio_spectrogram_transformer.js"),o=t("./src/models/encodec/feature_extraction_encodec.js"),n=t("./src/models/clap/feature_extraction_clap.js"),u=t("./src/models/dac/feature_extraction_dac.js"),d=t("./src/models/moonshine/feature_extraction_moonshine.js"),h=t("./src/models/pyannote/feature_extraction_pyannote.js"),f=t("./src/models/seamless_m4t/feature_extraction_seamless_m4t.js"),y=t("./src/models/snac/feature_extraction_snac.js"),m=t("./src/models/speecht5/feature_extraction_speecht5.js"),g=t("./src/models/wav2vec2/feature_extraction_wav2vec2.js"),T=t("./src/models/wespeaker/feature_extraction_wespeaker.js"),M=t("./src/models/whisper/feature_extraction_whisper.js"),I=t("./src/base/image_processors_utils.js")},"./src/models/florence2/processing_florence2.js":(e,r,t)=>{t.r(r),t.d(r,{Florence2Processor:()=>u});var i=t("./src/base/processing_utils.js"),o=t("./src/models/auto/image_processing_auto.js"),n=t("./src/tokenizers.js");class u extends i.Processor{constructor(h,f){super(h,f);const{tasks_answer_post_processing_type:y,task_prompts_without_inputs:m,task_prompts_with_input:g}=this.image_processor.config;this.tasks_answer_post_processing_type=new Map(Object.entries(y??{})),this.task_prompts_without_inputs=new Map(Object.entries(m??{})),this.task_prompts_with_input=new Map(Object.entries(g??{})),this.regexes={quad_boxes:/(.+?)/gm,bboxes:/([^<]+)?/gm},this.size_per_bin=1e3}construct_prompts(h){typeof h=="string"&&(h=[h]);const f=[];for(const y of h)if(this.task_prompts_without_inputs.has(y))f.push(this.task_prompts_without_inputs.get(y));else{for(const[m,g]of this.task_prompts_with_input)if(y.includes(m)){f.push(g.replaceAll("{input}",y).replaceAll(m,""));break}f.length!==h.length&&f.push(y)}return f}post_process_generation(h,f,y){const m=this.tasks_answer_post_processing_type.get(f)??"pure_text";h=h.replaceAll("","").replaceAll("","");let g;switch(m){case"pure_text":g=h;break;case"description_with_bboxes":case"bboxes":case"phrase_grounding":case"ocr":const T=m==="ocr"?"quad_boxes":"bboxes",M=h.matchAll(this.regexes[T]),I=[],z=[];for(const[E,C,...D]of M)I.push(C?C.trim():I.at(-1)??""),z.push(D.map((A,$)=>(Number(A)+.5)/this.size_per_bin*y[$%2]));g={labels:I,[T]:z};break;default:throw new Error(`Task "${f}" (of type "${m}") not yet implemented.`)}return{[f]:g}}async _call(h,f=null,y={}){if(!h&&!f)throw new Error("Either text or images must be provided");const m=await this.image_processor(h,y),g=f?this.tokenizer(f,y):{};return{...m,...g}}}de(u,"tokenizer_class",n.AutoTokenizer),de(u,"image_processor_class",o.AutoImageProcessor)},"./src/models/glpn/image_processing_glpn.js":(e,r,t)=>{t.r(r),t.d(r,{GLPNFeatureExtractor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{}},"./src/models/grounding_dino/image_processing_grounding_dino.js":(e,r,t)=>{t.r(r),t.d(r,{GroundingDinoImageProcessor:()=>n});var i=t("./src/base/image_processors_utils.js"),o=t("./src/utils/tensor.js");class n extends i.ImageProcessor{async _call(d){const h=await super._call(d),f=h.pixel_values.dims,y=(0,o.ones)([f[0],f[2],f[3]]);return{...h,pixel_mask:y}}}},"./src/models/grounding_dino/processing_grounding_dino.js":(e,r,t)=>{t.r(r),t.d(r,{GroundingDinoProcessor:()=>h});var i=t("./src/base/processing_utils.js"),o=t("./src/models/auto/image_processing_auto.js"),n=t("./src/tokenizers.js"),u=t("./src/base/image_processors_utils.js");function d(f,y){const g=f.dims.at(-1)-1,T=f.tolist();T.fill(!1,0,1),T.fill(!1,g);const M=y.tolist();return T.map((I,z)=>I?z:null).filter(I=>I!==null).map(I=>M[I])}class h extends i.Processor{async _call(y,m,g={}){const T=y?await this.image_processor(y,g):{};return{...m?this.tokenizer(m,g):{},...T}}post_process_grounded_object_detection(y,m,{box_threshold:g=.25,text_threshold:T=.25,target_sizes:M=null}={}){const{logits:I,pred_boxes:z}=y,E=I.dims[0];if(M!==null&&M.length!==E)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");const C=I.dims.at(1),D=I.sigmoid(),A=D.max(-1).tolist(),$=z.tolist().map(k=>k.map(O=>(0,u.center_to_corners_format)(O))),P=[];for(let k=0;kK.map((pe,re)=>pe*O[(re+1)%2])));const R=A[k],U=[],te=[],se=[];for(let K=0;K{t.r(r),t.d(r,{Idefics3ImageProcessor:()=>n});var i=t("./src/base/image_processors_utils.js"),o=t("./src/utils/tensor.js");class n extends i.ImageProcessor{constructor(d){super(d),this.do_image_splitting=d.do_image_splitting??!0,this.max_image_size=d.max_image_size}get_resize_for_vision_encoder(d,h){let[f,y]=d.dims.slice(-2);const m=y/f;return y>=f?(y=Math.ceil(y/h)*h,f=Math.floor(y/m),f=Math.ceil(f/h)*h):(f=Math.ceil(f/h)*h,y=Math.floor(f*m),y=Math.ceil(y/h)*h),{height:f,width:y}}async _call(d,{do_image_splitting:h=null,return_row_col_info:f=!1}={}){let y;if(!Array.isArray(d))y=[[d]];else{if(d.length===0||!d[0])throw new Error("No images provided.");Array.isArray(d[0])?y=d:y=[d]}let m=[],g=[],T=[];const M=[],I=[];for(const k of y){let O=await Promise.all(k.map(te=>this.preprocess(te)));M.push(...O.map(te=>te.original_size)),I.push(...O.map(te=>te.reshaped_input_size)),O.forEach(te=>te.pixel_values.unsqueeze_(0));const{longest_edge:R}=this.max_image_size;let U;if(h??this.do_image_splitting){let te=new Array(O.length),se=new Array(O.length);U=await Promise.all(O.map(async(K,pe)=>{const re=this.get_resize_for_vision_encoder(K.pixel_values,R),oe=await(0,o.interpolate_4d)(K.pixel_values,{size:[re.height,re.width]}),{frames:ge,num_splits_h:le,num_splits_w:Se}=await this.split_image(oe,this.max_image_size);return te[pe]=le,se[pe]=Se,(0,o.cat)(ge,0)})),g.push(te),T.push(se)}else{const te=[R,R];U=await Promise.all(O.map(se=>(0,o.interpolate_4d)(se.pixel_values,{size:te}))),g.push(new Array(O.length).fill(0)),T.push(new Array(O.length).fill(0))}m.push((0,o.cat)(U,0))}const z=m.length,[E,C,D,A]=m[0].dims;let $,P;if(z===1)$=m[0].unsqueeze_(0),P=(0,o.full)([z,E,D,A],!0);else{const k=Math.max(...m.map(U=>U.dims.at(0)));P=(0,o.full)([z,k,D,A],!0);const O=P.data,R=k*D*A;for(let U=0;Uf||T>y){M=Math.ceil(g/f),I=Math.ceil(T/y);const z=Math.ceil(g/M),E=Math.ceil(T/I);for(let A=0;A{t.r(r),t.d(r,{Idefics3Processor:()=>y});var i=t("./src/base/processing_utils.js"),o=t("./src/models/auto/image_processing_auto.js"),n=t("./src/tokenizers.js");t("./src/utils/image.js");var u=t("./src/utils/core.js");function d(m,g,T,M,I,z){let E="";for(let C=0;C`+I.repeat(m);E+=` +`}return E+=` +${M}${z}`+I.repeat(m)+`${M}`,E}function h(m,g,T,M){return`${g}${M}`+T.repeat(m)+`${g}`}function f(m,g,T,M,I,z){return m===0&&g===0?h(T,M,I,z):d(T,m,g,M,I,z)}class y extends i.Processor{constructor(){super(...arguments);de(this,"fake_image_token","");de(this,"image_token","");de(this,"global_img_token","")}async _call(T,M=null,I={}){I.return_row_col_info??(I.return_row_col_info=!0);let z;M&&(z=await this.image_processor(M,I)),Array.isArray(T)||(T=[T]);const E=z.rows??[new Array(T.length).fill(0)],C=z.cols??[new Array(T.length).fill(0)],D=this.config.image_seq_len,A=[],$=[];for(let k=0;kf(pe,U[re],D,this.fake_image_token,this.image_token,this.global_img_token)),se=O.split(this.image_token);if(se.length===0)throw new Error("The image token should be present in the text.");let K=se[0];for(let pe=0;pe{t.r(r),t.d(r,{BeitFeatureExtractor:()=>i.BeitFeatureExtractor,BitImageProcessor:()=>o.BitImageProcessor,CLIPFeatureExtractor:()=>u.CLIPFeatureExtractor,CLIPImageProcessor:()=>u.CLIPImageProcessor,ChineseCLIPFeatureExtractor:()=>n.ChineseCLIPFeatureExtractor,ConvNextFeatureExtractor:()=>d.ConvNextFeatureExtractor,ConvNextImageProcessor:()=>d.ConvNextImageProcessor,DPTFeatureExtractor:()=>m.DPTFeatureExtractor,DPTImageProcessor:()=>m.DPTImageProcessor,DeiTFeatureExtractor:()=>h.DeiTFeatureExtractor,DeiTImageProcessor:()=>h.DeiTImageProcessor,DetrFeatureExtractor:()=>f.DetrFeatureExtractor,DetrImageProcessor:()=>f.DetrImageProcessor,DonutFeatureExtractor:()=>y.DonutFeatureExtractor,DonutImageProcessor:()=>y.DonutImageProcessor,EfficientNetImageProcessor:()=>g.EfficientNetImageProcessor,GLPNFeatureExtractor:()=>T.GLPNFeatureExtractor,GroundingDinoImageProcessor:()=>M.GroundingDinoImageProcessor,Idefics3ImageProcessor:()=>I.Idefics3ImageProcessor,JinaCLIPImageProcessor:()=>E.JinaCLIPImageProcessor,LlavaOnevisionImageProcessor:()=>C.LlavaOnevisionImageProcessor,Mask2FormerImageProcessor:()=>D.Mask2FormerImageProcessor,MaskFormerFeatureExtractor:()=>A.MaskFormerFeatureExtractor,MaskFormerImageProcessor:()=>A.MaskFormerImageProcessor,MobileNetV1FeatureExtractor:()=>$.MobileNetV1FeatureExtractor,MobileNetV1ImageProcessor:()=>$.MobileNetV1ImageProcessor,MobileNetV2FeatureExtractor:()=>P.MobileNetV2FeatureExtractor,MobileNetV2ImageProcessor:()=>P.MobileNetV2ImageProcessor,MobileNetV3FeatureExtractor:()=>k.MobileNetV3FeatureExtractor,MobileNetV3ImageProcessor:()=>k.MobileNetV3ImageProcessor,MobileNetV4FeatureExtractor:()=>O.MobileNetV4FeatureExtractor,MobileNetV4ImageProcessor:()=>O.MobileNetV4ImageProcessor,MobileViTFeatureExtractor:()=>R.MobileViTFeatureExtractor,MobileViTImageProcessor:()=>R.MobileViTImageProcessor,NougatImageProcessor:()=>U.NougatImageProcessor,OwlViTFeatureExtractor:()=>se.OwlViTFeatureExtractor,OwlViTImageProcessor:()=>se.OwlViTImageProcessor,Owlv2ImageProcessor:()=>te.Owlv2ImageProcessor,Phi3VImageProcessor:()=>K.Phi3VImageProcessor,PvtImageProcessor:()=>pe.PvtImageProcessor,Qwen2VLImageProcessor:()=>re.Qwen2VLImageProcessor,RTDetrImageProcessor:()=>oe.RTDetrImageProcessor,SamImageProcessor:()=>ge.SamImageProcessor,SegformerFeatureExtractor:()=>le.SegformerFeatureExtractor,SegformerImageProcessor:()=>le.SegformerImageProcessor,SiglipImageProcessor:()=>Se.SiglipImageProcessor,SmolVLMImageProcessor:()=>Ce.SmolVLMImageProcessor,Swin2SRImageProcessor:()=>q.Swin2SRImageProcessor,VLMImageProcessor:()=>z.VLMImageProcessor,ViTFeatureExtractor:()=>N.ViTFeatureExtractor,ViTImageProcessor:()=>N.ViTImageProcessor,VitMatteImageProcessor:()=>Q.VitMatteImageProcessor,VitPoseImageProcessor:()=>ue.VitPoseImageProcessor,YolosFeatureExtractor:()=>Ae.YolosFeatureExtractor,YolosImageProcessor:()=>Ae.YolosImageProcessor});var i=t("./src/models/beit/image_processing_beit.js"),o=t("./src/models/bit/image_processing_bit.js"),n=t("./src/models/chinese_clip/image_processing_chinese_clip.js"),u=t("./src/models/clip/image_processing_clip.js"),d=t("./src/models/convnext/image_processing_convnext.js"),h=t("./src/models/deit/image_processing_deit.js"),f=t("./src/models/detr/image_processing_detr.js"),y=t("./src/models/donut/image_processing_donut.js"),m=t("./src/models/dpt/image_processing_dpt.js"),g=t("./src/models/efficientnet/image_processing_efficientnet.js"),T=t("./src/models/glpn/image_processing_glpn.js"),M=t("./src/models/grounding_dino/image_processing_grounding_dino.js"),I=t("./src/models/idefics3/image_processing_idefics3.js"),z=t("./src/models/janus/image_processing_janus.js"),E=t("./src/models/jina_clip/image_processing_jina_clip.js"),C=t("./src/models/llava_onevision/image_processing_llava_onevision.js"),D=t("./src/models/mask2former/image_processing_mask2former.js"),A=t("./src/models/maskformer/image_processing_maskformer.js"),$=t("./src/models/mobilenet_v1/image_processing_mobilenet_v1.js"),P=t("./src/models/mobilenet_v2/image_processing_mobilenet_v2.js"),k=t("./src/models/mobilenet_v3/image_processing_mobilenet_v3.js"),O=t("./src/models/mobilenet_v4/image_processing_mobilenet_v4.js"),R=t("./src/models/mobilevit/image_processing_mobilevit.js"),U=t("./src/models/nougat/image_processing_nougat.js"),te=t("./src/models/owlv2/image_processing_owlv2.js"),se=t("./src/models/owlvit/image_processing_owlvit.js"),K=t("./src/models/phi3_v/image_processing_phi3_v.js"),pe=t("./src/models/pvt/image_processing_pvt.js"),re=t("./src/models/qwen2_vl/image_processing_qwen2_vl.js"),oe=t("./src/models/rt_detr/image_processing_rt_detr.js"),ge=t("./src/models/sam/image_processing_sam.js"),le=t("./src/models/segformer/image_processing_segformer.js"),Se=t("./src/models/siglip/image_processing_siglip.js"),Ce=t("./src/models/smolvlm/image_processing_smolvlm.js"),q=t("./src/models/swin2sr/image_processing_swin2sr.js"),N=t("./src/models/vit/image_processing_vit.js"),Q=t("./src/models/vitmatte/image_processing_vitmatte.js"),ue=t("./src/models/vitpose/image_processing_vitpose.js"),Ae=t("./src/models/yolos/image_processing_yolos.js")},"./src/models/janus/image_processing_janus.js":(e,r,t)=>{t.r(r),t.d(r,{VLMImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{constructor(u){super({do_pad:!0,pad_size:{width:u.image_size,height:u.image_size},...u}),this.constant_values=this.config.background_color.map(d=>d*this.rescale_factor)}pad_image(u,d,h,f){return super.pad_image(u,d,h,{constant_values:this.constant_values,center:!0,...f})}}},"./src/models/janus/processing_janus.js":(e,r,t)=>{t.r(r),t.d(r,{VLChatProcessor:()=>f});var i=t("./src/base/processing_utils.js"),o=t("./src/models/auto/image_processing_auto.js"),n=t("./src/tokenizers.js"),u=t("./src/utils/core.js"),d=t("./src/utils/tensor.js"),h=t("./src/utils/image.js");class f extends i.Processor{constructor(m,g){super(m,g),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(m,{images:g=null,chat_template:T="default"}={}){g?Array.isArray(g)||(g=[g]):g=await Promise.all(m.filter(U=>U.images).flatMap(U=>U.images).map(U=>h.RawImage.read(U)));const M=this.tokenizer,I=M.apply_chat_template(m,{tokenize:!1,add_generation_prompt:!0,chat_template:T}),z=U=>M.encode(U,{add_special_tokens:!1}),E=I.split(this.image_tag),C=E.length-1;if(g.length!==C)throw new Error(`Number of images provided (${g.length}) does not match number of "${this.image_tag}" image tags (${C})`);const[D,A,$]=M.model.convert_tokens_to_ids([this.image_tag,this.image_start_tag,this.image_end_tag]);let P=z(E[0]),k=new Array(P.length).fill(!1);for(let U=1;U0){const U=await this.image_processor(g);return U.pixel_values.unsqueeze_(0),{...R,...U}}return R}}de(f,"image_processor_class",o.AutoImageProcessor),de(f,"tokenizer_class",n.AutoTokenizer),de(f,"uses_processor_config",!0)},"./src/models/jina_clip/image_processing_jina_clip.js":(e,r,t)=>{t.r(r),t.d(r,{JinaCLIPImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{constructor(u){const{resize_mode:d,fill_color:h,interpolation:f,size:y,...m}=u,g=d==="squash"?{width:y,height:y}:d==="shortest"?{shortest_edge:y}:{longest_edge:y},T=f==="bicubic"?3:2;super({...m,size:g,resample:T,do_center_crop:!0,crop_size:y,do_normalize:!0})}}},"./src/models/jina_clip/processing_jina_clip.js":(e,r,t)=>{t.r(r),t.d(r,{JinaCLIPProcessor:()=>u});var i=t("./src/base/processing_utils.js"),o=t("./src/models/auto/image_processing_auto.js"),n=t("./src/tokenizers.js");class u extends i.Processor{async _call(h=null,f=null,y={}){if(!h&&!f)throw new Error("Either text or images must be provided");const m=h?this.tokenizer(h,y):{},g=f?await this.image_processor(f,y):{};return{...m,...g}}}de(u,"tokenizer_class",n.AutoTokenizer),de(u,"image_processor_class",o.AutoImageProcessor)},"./src/models/llava_onevision/image_processing_llava_onevision.js":(e,r,t)=>{t.r(r),t.d(r,{LlavaOnevisionImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{}},"./src/models/mask2former/image_processing_mask2former.js":(e,r,t)=>{t.r(r),t.d(r,{Mask2FormerImageProcessor:()=>o});var i=t("./src/models/maskformer/image_processing_maskformer.js");class o extends i.MaskFormerImageProcessor{}},"./src/models/maskformer/image_processing_maskformer.js":(e,r,t)=>{t.r(r),t.d(r,{MaskFormerFeatureExtractor:()=>n,MaskFormerImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{post_process_panoptic_segmentation(...d){return(0,i.post_process_panoptic_segmentation)(...d)}post_process_instance_segmentation(...d){return(0,i.post_process_instance_segmentation)(...d)}}class n extends o{}},"./src/models/mgp_str/processing_mgp_str.js":(e,r,t)=>{t.r(r),t.d(r,{MgpstrProcessor:()=>h});var i=t("./src/base/processing_utils.js"),o=t("./src/models/auto/image_processing_auto.js"),n=t("./src/tokenizers.js"),u=t("./src/utils/maths.js");const d={char:["char_decode",1],bpe:["bpe_decode",2],wp:["wp_decode",102]};class h extends i.Processor{get char_tokenizer(){return this.components.char_tokenizer}get bpe_tokenizer(){return this.components.bpe_tokenizer}get wp_tokenizer(){return this.components.wp_tokenizer}_decode_helper(y,m){if(!d.hasOwnProperty(m))throw new Error(`Format ${m} is not supported.`);const[g,T]=d[m],M=this[g].bind(this),[I,z]=y.dims,E=[],C=[],D=y.tolist();for(let $=0;$0?O.reduce((U,te)=>U*te,1):0;C.push(k),E.push(R)}return[M(C),E]}char_decode(y){return this.char_tokenizer.batch_decode(y).map(m=>m.replaceAll(" ",""))}bpe_decode(y){return this.bpe_tokenizer.batch_decode(y)}wp_decode(y){return this.wp_tokenizer.batch_decode(y).map(m=>m.replaceAll(" ",""))}batch_decode([y,m,g]){const[T,M]=this._decode_helper(y,"char"),[I,z]=this._decode_helper(m,"bpe"),[E,C]=this._decode_helper(g,"wp"),D=[],A=[];for(let $=0;${t.r(r),t.d(r,{MobileNetV1FeatureExtractor:()=>n,MobileNetV1ImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{}class n extends o{}},"./src/models/mobilenet_v2/image_processing_mobilenet_v2.js":(e,r,t)=>{t.r(r),t.d(r,{MobileNetV2FeatureExtractor:()=>n,MobileNetV2ImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{}class n extends o{}},"./src/models/mobilenet_v3/image_processing_mobilenet_v3.js":(e,r,t)=>{t.r(r),t.d(r,{MobileNetV3FeatureExtractor:()=>n,MobileNetV3ImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{}class n extends o{}},"./src/models/mobilenet_v4/image_processing_mobilenet_v4.js":(e,r,t)=>{t.r(r),t.d(r,{MobileNetV4FeatureExtractor:()=>n,MobileNetV4ImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{}class n extends o{}},"./src/models/mobilevit/image_processing_mobilevit.js":(e,r,t)=>{t.r(r),t.d(r,{MobileViTFeatureExtractor:()=>n,MobileViTImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{}class n extends o{}},"./src/models/moonshine/feature_extraction_moonshine.js":(e,r,t)=>{t.r(r),t.d(r,{MoonshineFeatureExtractor:()=>n});var i=t("./src/base/feature_extraction_utils.js"),o=t("./src/utils/tensor.js");class n extends i.FeatureExtractor{async _call(d){(0,i.validate_audio_inputs)(d,"MoonshineFeatureExtractor"),d instanceof Float64Array&&(d=new Float32Array(d));const h=[1,d.length];return{input_values:new o.Tensor("float32",d,h)}}}},"./src/models/moonshine/processing_moonshine.js":(e,r,t)=>{t.r(r),t.d(r,{MoonshineProcessor:()=>u});var i=t("./src/models/auto/feature_extraction_auto.js"),o=t("./src/tokenizers.js"),n=t("./src/base/processing_utils.js");class u extends n.Processor{async _call(h){return await this.feature_extractor(h)}}de(u,"tokenizer_class",o.AutoTokenizer),de(u,"feature_extractor_class",i.AutoFeatureExtractor)},"./src/models/nougat/image_processing_nougat.js":(e,r,t)=>{t.r(r),t.d(r,{NougatImageProcessor:()=>o});var i=t("./src/models/donut/image_processing_donut.js");class o extends i.DonutImageProcessor{}},"./src/models/owlv2/image_processing_owlv2.js":(e,r,t)=>{t.r(r),t.d(r,{Owlv2ImageProcessor:()=>o});var i=t("./src/models/owlvit/image_processing_owlvit.js");class o extends i.OwlViTImageProcessor{}},"./src/models/owlvit/image_processing_owlvit.js":(e,r,t)=>{t.r(r),t.d(r,{OwlViTFeatureExtractor:()=>n,OwlViTImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{post_process_object_detection(...d){return(0,i.post_process_object_detection)(...d)}}class n extends o{}},"./src/models/owlvit/processing_owlvit.js":(e,r,t)=>{t.r(r),t.d(r,{OwlViTProcessor:()=>u});var i=t("./src/base/processing_utils.js"),o=t("./src/models/auto/image_processing_auto.js"),n=t("./src/tokenizers.js");class u extends i.Processor{}de(u,"tokenizer_class",n.AutoTokenizer),de(u,"image_processor_class",o.AutoImageProcessor)},"./src/models/paligemma/processing_paligemma.js":(e,r,t)=>{t.r(r),t.d(r,{PaliGemmaProcessor:()=>h});var i=t("./src/base/processing_utils.js"),o=t("./src/models/auto/image_processing_auto.js"),n=t("./src/tokenizers.js");const u="";function d(f,y,m,g,T){return`${g.repeat(m*T)}${y}${f} +`}class h extends i.Processor{async _call(y,m=null,g={}){m||(console.warn("You are using PaliGemma without a text prefix. It will perform as a picture-captioning model."),m=""),Array.isArray(y)||(y=[y]),Array.isArray(m)||(m=[m]);const T=this.tokenizer.bos_token,M=this.image_processor.config.image_seq_length;let I;m.some(C=>C.includes(u))?I=m.map(C=>{const D=C.replaceAll(u,u.repeat(M)),A=D.lastIndexOf(u),$=A===-1?0:A+u.length;return D.slice(0,$)+T+D.slice($)+` +`}):(console.warn("You are passing both `text` and `images` to `PaliGemmaProcessor`. The processor expects special image tokens in the text, as many tokens as there are images per each text. It is recommended to add `` tokens in the very beginning of your text. For this call, we will infer how many images each text has and add special tokens."),I=m.map(C=>d(C,T,M,u,y.length)));const z=this.tokenizer(I,g);return{...await this.image_processor(y,g),...z}}}de(h,"tokenizer_class",n.AutoTokenizer),de(h,"image_processor_class",o.AutoImageProcessor),de(h,"uses_processor_config",!1)},"./src/models/phi3_v/image_processing_phi3_v.js":(e,r,t)=>{t.r(r),t.d(r,{Phi3VImageProcessor:()=>y});var i=t("./src/base/image_processors_utils.js"),o=t("./src/utils/tensor.js");const n=336,u=[2,3],{ceil:d,floor:h,sqrt:f}=Math;class y extends i.ImageProcessor{constructor(g){super({...g,do_normalize:!0,do_pad:!0,pad_size:"custom",do_convert_rgb:!0,do_resize:!0}),this._num_crops=g.num_crops}calc_num_image_tokens_from_image_size(g,T){const{num_img_tokens:M}=this.config;return h((h(T/n)*h(g/n)+1)*M+1+(h(T/n)+1)*f(M))}get_resize_output_image_size(g,T){const M=this._num_crops,[I,z]=g.size;let E=I/z,C=1;for(;C*Math.ceil(C/E)<=M;)C+=1;C-=1;const D=Math.floor(C*336),A=Math.floor(D/E);return[D,A]}pad_image(g,T,M,I={}){const[z,E]=T,C=n*d(z/n),D=n*d(E/n),A=[1,1,1].map(($,P)=>($-this.image_mean[P])/this.image_std[P]);return super.pad_image(g,T,{width:D,height:C},{center:!0,constant_values:A,...I})}async _call(g,{num_crops:T=null}={}){if(this._num_crops=T??(T=this.config.num_crops),T<4||f(T)%1!==0)throw new Error("num_crops must be a square number >= 4");Array.isArray(g)||(g=[g]);const M=g.length,I=await Promise.all(g.map(k=>this.preprocess(k))),z=I.map(k=>k.original_size),E=I.map(k=>k.reshaped_input_size),C=[];for(const{pixel_values:k}of I){k.unsqueeze_(0);const[O,R]=k.dims.slice(-2),U=await(0,o.interpolate_4d)(k,{size:[n,n],mode:"bicubic"});if(T>0){const te=[],se=f(T),K=h(R/se),pe=h(O/se);for(let oe=0;oek.map(O=>n*d(O/n))),$=new o.Tensor("int64",A.flat(),[M,2]),P=A.map(([k,O])=>this.calc_num_image_tokens_from_image_size(O,k));return{pixel_values:D,original_sizes:z,reshaped_input_sizes:E,image_sizes:$,num_img_tokens:P}}}},"./src/models/phi3_v/processing_phi3_v.js":(e,r,t)=>{t.r(r),t.d(r,{Phi3VProcessor:()=>h});var i=t("./src/base/processing_utils.js"),o=t("./src/models/auto/image_processing_auto.js"),n=t("./src/tokenizers.js");t("./src/utils/image.js");const u="<|image|>",d=/<\|image_\d+\|>/g;class h extends i.Processor{async _call(y,m=null,{padding:g=!0,truncation:T=!0,num_crops:M=null}={}){Array.isArray(y)||(y=[y]);let I,z;if(m){z=await this.image_processor(m,{num_crops:M});const{num_img_tokens:E}=z,C=y.map((A,$)=>A.split(d).join(u.repeat(E[$])));I=this.tokenizer(C,{padding:g,truncation:T});const D=this.tokenizer.model.convert_tokens_to_ids([u])[0];I.input_ids.map_(A=>A==D?-A:A)}else I=this.tokenizer(y);return{...I,...z}}}de(h,"image_processor_class",o.AutoImageProcessor),de(h,"tokenizer_class",n.AutoTokenizer)},"./src/models/processors.js":(e,r,t)=>{t.r(r),t.d(r,{Florence2Processor:()=>i.Florence2Processor,GroundingDinoProcessor:()=>o.GroundingDinoProcessor,Idefics3Processor:()=>n.Idefics3Processor,JinaCLIPProcessor:()=>d.JinaCLIPProcessor,MgpstrProcessor:()=>h.MgpstrProcessor,MoonshineProcessor:()=>f.MoonshineProcessor,OwlViTProcessor:()=>y.OwlViTProcessor,PaliGemmaProcessor:()=>g.PaliGemmaProcessor,Phi3VProcessor:()=>m.Phi3VProcessor,PyAnnoteProcessor:()=>T.PyAnnoteProcessor,Qwen2VLProcessor:()=>M.Qwen2VLProcessor,SamProcessor:()=>I.SamProcessor,SmolVLMProcessor:()=>z.SmolVLMProcessor,SpeechT5Processor:()=>E.SpeechT5Processor,UltravoxProcessor:()=>C.UltravoxProcessor,VLChatProcessor:()=>u.VLChatProcessor,Wav2Vec2Processor:()=>D.Wav2Vec2Processor,Wav2Vec2ProcessorWithLM:()=>A.Wav2Vec2ProcessorWithLM,WhisperProcessor:()=>$.WhisperProcessor});var i=t("./src/models/florence2/processing_florence2.js"),o=t("./src/models/grounding_dino/processing_grounding_dino.js"),n=t("./src/models/idefics3/processing_idefics3.js"),u=t("./src/models/janus/processing_janus.js"),d=t("./src/models/jina_clip/processing_jina_clip.js"),h=t("./src/models/mgp_str/processing_mgp_str.js"),f=t("./src/models/moonshine/processing_moonshine.js"),y=t("./src/models/owlvit/processing_owlvit.js"),m=t("./src/models/phi3_v/processing_phi3_v.js"),g=t("./src/models/paligemma/processing_paligemma.js"),T=t("./src/models/pyannote/processing_pyannote.js"),M=t("./src/models/qwen2_vl/processing_qwen2_vl.js"),I=t("./src/models/sam/processing_sam.js"),z=t("./src/models/smolvlm/processing_smolvlm.js"),E=t("./src/models/speecht5/processing_speecht5.js"),C=t("./src/models/ultravox/processing_ultravox.js"),D=t("./src/models/wav2vec2/processing_wav2vec2.js"),A=t("./src/models/wav2vec2_with_lm/processing_wav2vec2_with_lm.js"),$=t("./src/models/whisper/processing_whisper.js")},"./src/models/pvt/image_processing_pvt.js":(e,r,t)=>{t.r(r),t.d(r,{PvtImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{}},"./src/models/pyannote/feature_extraction_pyannote.js":(e,r,t)=>{t.r(r),t.d(r,{PyAnnoteFeatureExtractor:()=>u});var i=t("./src/base/feature_extraction_utils.js"),o=t("./src/utils/tensor.js"),n=t("./src/utils/maths.js");class u extends i.FeatureExtractor{async _call(h){(0,i.validate_audio_inputs)(h,"PyAnnoteFeatureExtractor"),h instanceof Float64Array&&(h=new Float32Array(h));const f=[1,1,h.length];return{input_values:new o.Tensor("float32",h,f)}}samples_to_frames(h){return(h-this.config.offset)/this.config.step}post_process_speaker_diarization(h,f){const y=f/this.samples_to_frames(f)/this.config.sampling_rate,m=[];for(const g of h.tolist()){const T=[];let M=-1;for(let I=0;I({id:I,start:z*y,end:E*y,confidence:C/(E-z)})))}return m}}},"./src/models/pyannote/processing_pyannote.js":(e,r,t)=>{t.r(r),t.d(r,{PyAnnoteProcessor:()=>n});var i=t("./src/base/processing_utils.js"),o=t("./src/models/pyannote/feature_extraction_pyannote.js");class n extends i.Processor{async _call(d){return await this.feature_extractor(d)}post_process_speaker_diarization(...d){return this.feature_extractor.post_process_speaker_diarization(...d)}get sampling_rate(){return this.feature_extractor.config.sampling_rate}}de(n,"feature_extractor_class",o.PyAnnoteFeatureExtractor)},"./src/models/qwen2_vl/image_processing_qwen2_vl.js":(e,r,t)=>{t.r(r),t.d(r,{Qwen2VLImageProcessor:()=>n});var i=t("./src/base/image_processors_utils.js"),o=t("./src/utils/tensor.js");class n extends i.ImageProcessor{async _call(d,...h){const{pixel_values:f,original_sizes:y,reshaped_input_sizes:m}=await super._call(d,...h);let g=f;const{temporal_patch_size:T,merge_size:M,patch_size:I}=this.config;g.dims[0]===1&&(g=(0,o.cat)(Array.from({length:T},()=>g),0));const z=g.dims[0]/T,E=g.dims[1],C=Math.floor(g.dims[2]/I),D=Math.floor(g.dims[3]/I),A=g.view(z,T,E,Math.floor(C/M),M,I,Math.floor(D/M),M,I).permute(0,3,6,4,7,2,1,5,8).view(z*C*D,E*T*I*I),$=new o.Tensor("int64",[z,C,D],[1,3]);return{pixel_values:A,image_grid_thw:$,original_sizes:y,reshaped_input_sizes:m}}}},"./src/models/qwen2_vl/processing_qwen2_vl.js":(e,r,t)=>{t.r(r),t.d(r,{Qwen2VLProcessor:()=>u});var i=t("./src/base/processing_utils.js"),o=t("./src/models/auto/image_processing_auto.js"),n=t("./src/tokenizers.js");t("./src/utils/image.js");class u extends i.Processor{async _call(h,f=null,...y){Array.isArray(h)||(h=[h]);let m,g;if(f&&(m=await this.image_processor(f),g=m.image_grid_thw),g){let M=this.image_processor.config.merge_size**2,I=0;const z=g.tolist();h=h.map(E=>{for(;E.includes("<|image_pad|>");){const C=Number(z[I++].reduce((D,A)=>D*A,1n));E=E.replace("<|image_pad|>","<|placeholder|>".repeat(Math.floor(C/M)))}return E.replaceAll("<|placeholder|>","<|image_pad|>")})}return{...this.tokenizer(h),...m}}}de(u,"image_processor_class",o.AutoImageProcessor),de(u,"tokenizer_class",n.AutoTokenizer)},"./src/models/rt_detr/image_processing_rt_detr.js":(e,r,t)=>{t.r(r),t.d(r,{RTDetrImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{post_process_object_detection(...u){return(0,i.post_process_object_detection)(...u)}}},"./src/models/sam/image_processing_sam.js":(e,r,t)=>{t.r(r),t.d(r,{SamImageProcessor:()=>u});var i=t("./src/base/image_processors_utils.js"),o=t("./src/utils/core.js"),n=t("./src/utils/tensor.js");class u extends i.ImageProcessor{reshape_input_points(h,f,y,m=!1){h=structuredClone(h);let g=(0,o.calculateDimensions)(h);if(g.length===3)m||(g=[1,...g]),h=[h];else if(g.length!==4)throw Error("The input_points must be a 4D tensor of shape `batch_size`, `point_batch_size`, `nb_points_per_image`, `2`.");for(let T=0;Tm!==f.dims[g]))throw Error(`The first ${y.length} dimensions of 'input_points' and 'input_labels' must be the same.`);return new n.Tensor("int64",h.flat(1/0).map(BigInt),y)}async _call(h,{input_points:f=null,input_labels:y=null,input_boxes:m=null}={}){const g=await super._call(h);if(f&&(g.input_points=this.reshape_input_points(f,g.original_sizes,g.reshaped_input_sizes)),y){if(!g.input_points)throw Error("`input_points` must be provided if `input_labels` are provided.");g.input_labels=this.add_input_labels(y,g.input_points)}return m&&(g.input_boxes=this.reshape_input_points(m,g.original_sizes,g.reshaped_input_sizes,!0)),g}async post_process_masks(h,f,y,{mask_threshold:m=0,binarize:g=!0,pad_size:T=null}={}){const M=[];T=T??this.pad_size;const I=[T.height,T.width];for(let z=0;zm&&($[P]=1);D=new n.Tensor("bool",$,D.dims)}M.push(D)}return M}generate_crop_boxes(h,f,{crop_n_layers:y=0,overlap_ratio:m=512/1500,points_per_crop:g=32,crop_n_points_downscale_factor:T=1}={}){}}},"./src/models/sam/processing_sam.js":(e,r,t)=>{t.r(r),t.d(r,{SamProcessor:()=>n});var i=t("./src/base/processing_utils.js"),o=t("./src/models/auto/image_processing_auto.js");class n extends i.Processor{async _call(...d){return await this.image_processor(...d)}post_process_masks(...d){return this.image_processor.post_process_masks(...d)}reshape_input_points(...d){return this.image_processor.reshape_input_points(...d)}}de(n,"image_processor_class",o.AutoImageProcessor)},"./src/models/seamless_m4t/feature_extraction_seamless_m4t.js":(e,r,t)=>{t.r(r),t.d(r,{SeamlessM4TFeatureExtractor:()=>u});var i=t("./src/base/feature_extraction_utils.js"),o=t("./src/utils/tensor.js"),n=t("./src/utils/audio.js");class u extends i.FeatureExtractor{constructor(h){super(h);const f=this.config.sampling_rate,y=(0,n.mel_filter_bank)(257,this.config.num_mel_bins,20,Math.floor(f/2),f,null,"kaldi",!0);this.mel_filters=y,this.window=(0,n.window_function)(400,"povey",{periodic:!1})}async _extract_fbank_features(h,f){return h=h.map(y=>y*32768),(0,n.spectrogram)(h,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1192092955078125e-22,remove_dc_offset:!0,max_num_frames:f,transpose:!0})}async _call(h,{padding:f=!0,pad_to_multiple_of:y=2,do_normalize_per_mel_bins:m=!0,return_attention_mask:g=!0}={}){(0,i.validate_audio_inputs)(h,"SeamlessM4TFeatureExtractor");let T=await this._extract_fbank_features(h,this.config.max_length);if(m){const[$,P]=T.dims,k=T.data;for(let O=0;O0){const R=new Float32Array(P*($+O));R.set(k),R.fill(this.config.padding_value,k.length);const U=$+O;T=new o.Tensor(T.type,R,[U,P]),g&&(M=new o.Tensor("int64",new BigInt64Array(U),[1,U]),M.data.fill(1n,0,$))}}const[I,z]=T.dims,E=this.config.stride;if(I%E!==0)throw new Error(`The number of frames (${I}) must be a multiple of the stride (${E}).`);const D=T.view(1,Math.floor(I/E),z*E),A={input_features:D};if(g){const $=D.dims[1],P=new BigInt64Array($);if(M){const k=M.data;for(let O=1,R=0;O{t.r(r),t.d(r,{SegformerFeatureExtractor:()=>n,SegformerImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{post_process_semantic_segmentation(...d){return(0,i.post_process_semantic_segmentation)(...d)}}class n extends o{}},"./src/models/siglip/image_processing_siglip.js":(e,r,t)=>{t.r(r),t.d(r,{SiglipImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{}},"./src/models/smolvlm/image_processing_smolvlm.js":(e,r,t)=>{t.r(r),t.d(r,{SmolVLMImageProcessor:()=>i.Idefics3ImageProcessor});var i=t("./src/models/idefics3/image_processing_idefics3.js")},"./src/models/smolvlm/processing_smolvlm.js":(e,r,t)=>{t.r(r),t.d(r,{SmolVLMProcessor:()=>i.Idefics3Processor});var i=t("./src/models/idefics3/processing_idefics3.js")},"./src/models/snac/feature_extraction_snac.js":(e,r,t)=>{t.r(r),t.d(r,{SnacFeatureExtractor:()=>o});var i=t("./src/models/dac/feature_extraction_dac.js");class o extends i.DacFeatureExtractor{}},"./src/models/speecht5/feature_extraction_speecht5.js":(e,r,t)=>{t.r(r),t.d(r,{SpeechT5FeatureExtractor:()=>o});var i=t("./src/base/feature_extraction_utils.js");class o extends i.FeatureExtractor{}},"./src/models/speecht5/processing_speecht5.js":(e,r,t)=>{t.r(r),t.d(r,{SpeechT5Processor:()=>u});var i=t("./src/base/processing_utils.js"),o=t("./src/tokenizers.js"),n=t("./src/models/auto/feature_extraction_auto.js");class u extends i.Processor{async _call(h){return await this.feature_extractor(h)}}de(u,"tokenizer_class",o.AutoTokenizer),de(u,"feature_extractor_class",n.AutoFeatureExtractor)},"./src/models/swin2sr/image_processing_swin2sr.js":(e,r,t)=>{t.r(r),t.d(r,{Swin2SRImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{pad_image(u,d,h,f={}){const[y,m,g]=d;return super.pad_image(u,d,{width:m+(h-m%h)%h,height:y+(h-y%h)%h},{mode:"symmetric",center:!1,constant_values:-1,...f})}}},"./src/models/ultravox/processing_ultravox.js":(e,r,t)=>{t.r(r),t.d(r,{UltravoxProcessor:()=>u});var i=t("./src/models/auto/feature_extraction_auto.js"),o=t("./src/tokenizers.js"),n=t("./src/base/processing_utils.js");class u extends n.Processor{async _call(h,f=null,y={}){if(Array.isArray(h))throw new Error("Batched inputs are not supported yet.");let m={};if(f){const T=f.length,{input_features:M}=await this.feature_extractor(f,{...y,max_length:T}),I=Math.round(T/this.config.encoder_ds_factor+1e-4),z=1+Math.ceil(I/this.config.stack_factor);m.audio_token_len=[z],m.audio_values=M;const E=this.config.audio_placeholder;if(!h.includes(E))throw new Error(`The input text does not contain the image token ${E}.`);h=h.replaceAll(E,E.repeat(z))}return{...this.tokenizer(h,{add_special_tokens:!1,...y}),...m}}}de(u,"tokenizer_class",o.AutoTokenizer),de(u,"feature_extractor_class",i.AutoFeatureExtractor),de(u,"uses_processor_config",!0)},"./src/models/vit/image_processing_vit.js":(e,r,t)=>{t.r(r),t.d(r,{ViTFeatureExtractor:()=>n,ViTImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{}class n extends o{}},"./src/models/vitmatte/image_processing_vitmatte.js":(e,r,t)=>{t.r(r),t.d(r,{VitMatteImageProcessor:()=>n});var i=t("./src/base/image_processors_utils.js"),o=t("./src/utils/tensor.js");class n extends i.ImageProcessor{async _call(d,h){Array.isArray(d)||(d=[d]),Array.isArray(h)||(h=[h]);const f=await Promise.all(d.map(g=>this.preprocess(g))),y=await Promise.all(h.map(g=>this.preprocess(g,{do_normalize:!1,do_convert_rgb:!1,do_convert_grayscale:!0})));return{pixel_values:(0,o.stack)(f.map((g,T)=>(0,o.cat)([g.pixel_values,y[T].pixel_values],0)),0),original_sizes:f.map(g=>g.original_size),reshaped_input_sizes:f.map(g=>g.reshaped_input_size)}}}},"./src/models/vitpose/image_processing_vitpose.js":(e,r,t)=>{t.r(r),t.d(r,{VitPoseImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{post_process_pose_estimation(u,d,{threshold:h=null}={}){const f=u.tolist(),[y,m,g,T]=u.dims,M=[];for(let I=0;I{t.r(r),t.d(r,{Wav2Vec2FeatureExtractor:()=>n});var i=t("./src/base/feature_extraction_utils.js"),o=t("./src/utils/tensor.js");class n extends i.FeatureExtractor{_zero_mean_unit_var_norm(d){const f=d.reduce((m,g)=>m+g,0)/d.length,y=d.reduce((m,g)=>m+(g-f)**2,0)/d.length;return d.map(m=>(m-f)/Math.sqrt(y+1e-7))}async _call(d){(0,i.validate_audio_inputs)(d,"Wav2Vec2FeatureExtractor"),d instanceof Float64Array&&(d=new Float32Array(d));let h=d;this.config.do_normalize&&(h=this._zero_mean_unit_var_norm(h));const f=[1,h.length];return{input_values:new o.Tensor("float32",h,f),attention_mask:new o.Tensor("int64",new BigInt64Array(h.length).fill(1n),f)}}}},"./src/models/wav2vec2/processing_wav2vec2.js":(e,r,t)=>{t.r(r),t.d(r,{Wav2Vec2Processor:()=>u});var i=t("./src/tokenizers.js"),o=t("./src/models/auto/feature_extraction_auto.js"),n=t("./src/base/processing_utils.js");class u extends n.Processor{async _call(h){return await this.feature_extractor(h)}}de(u,"tokenizer_class",i.AutoTokenizer),de(u,"feature_extractor_class",o.AutoFeatureExtractor)},"./src/models/wav2vec2_with_lm/processing_wav2vec2_with_lm.js":(e,r,t)=>{t.r(r),t.d(r,{Wav2Vec2ProcessorWithLM:()=>u});var i=t("./src/tokenizers.js"),o=t("./src/models/auto/feature_extraction_auto.js"),n=t("./src/base/processing_utils.js");class u extends n.Processor{async _call(h){return await this.feature_extractor(h)}}de(u,"tokenizer_class",i.AutoTokenizer),de(u,"feature_extractor_class",o.AutoFeatureExtractor)},"./src/models/wespeaker/feature_extraction_wespeaker.js":(e,r,t)=>{t.r(r),t.d(r,{WeSpeakerFeatureExtractor:()=>n});var i=t("./src/base/feature_extraction_utils.js");t("./src/utils/tensor.js");var o=t("./src/utils/audio.js");class n extends i.FeatureExtractor{constructor(d){super(d);const h=this.config.sampling_rate,f=(0,o.mel_filter_bank)(257,this.config.num_mel_bins,20,Math.floor(h/2),h,null,"kaldi",!0);this.mel_filters=f,this.window=(0,o.window_function)(400,"hamming",{periodic:!1}),this.min_num_frames=this.config.min_num_frames}async _extract_fbank_features(d){return d=d.map(h=>h*32768),(0,o.spectrogram)(d,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1192092955078125e-22,remove_dc_offset:!0,transpose:!0,min_num_frames:this.min_num_frames})}async _call(d){(0,i.validate_audio_inputs)(d,"WeSpeakerFeatureExtractor");const h=(await this._extract_fbank_features(d)).unsqueeze_(0);if(this.config.fbank_centering_span===null){const f=h.mean(1).data,y=h.data,[m,g,T]=h.dims;for(let M=0;M{t.r(r),t.d(r,{WHISPER_LANGUAGE_MAPPING:()=>o,WHISPER_TO_LANGUAGE_CODE_MAPPING:()=>n,whisper_language_to_code:()=>u});const i=[["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(i),n=new Map([...i.map(([d,h])=>[h,d]),["burmese","my"],["valencian","ca"],["flemish","nl"],["haitian","ht"],["letzeburgesch","lb"],["pushto","ps"],["panjabi","pa"],["moldavian","ro"],["moldovan","ro"],["sinhalese","si"],["castilian","es"]]);function u(d){d=d.toLowerCase();let h=n.get(d);if(h===void 0){const f=d.match(/^<\|([a-z]{2})\|>$/);if(f&&(d=f[1]),o.has(d))h=d;else{const m=d.length===2?o.keys():o.values();throw new Error(`Language "${d}" is not supported. Must be one of: ${JSON.stringify(Array.from(m))}`)}}return h}},"./src/models/whisper/feature_extraction_whisper.js":(e,r,t)=>{t.r(r),t.d(r,{WhisperFeatureExtractor:()=>u});var i=t("./src/base/feature_extraction_utils.js");t("./src/utils/tensor.js");var o=t("./src/utils/audio.js"),n=t("./src/utils/maths.js");class u extends i.FeatureExtractor{constructor(h){var f;super(h),(f=this.config).mel_filters??(f.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(h){const f=await(0,o.spectrogram)(h,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(h.length/this.config.hop_length),this.config.nb_max_frames)}),y=f.data,m=(0,n.max)(y)[0];for(let g=0;gm?(h.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`."),y=h.slice(0,m)):(y=new Float32Array(m),y.set(h)),{input_features:(await this._extract_fbank_features(y)).unsqueeze_(0)}}}},"./src/models/whisper/generation_whisper.js":(e,r,t)=>{t.r(r),t.d(r,{WhisperGenerationConfig:()=>o});var i=t("./src/generation/configuration_utils.js");class o extends i.GenerationConfig{constructor(){super(...arguments);de(this,"return_timestamps",null);de(this,"return_token_timestamps",null);de(this,"num_frames",null);de(this,"alignment_heads",null);de(this,"task",null);de(this,"language",null);de(this,"no_timestamps_token_id",null);de(this,"prompt_ids",null);de(this,"is_multilingual",null);de(this,"lang_to_id",null);de(this,"task_to_id",null);de(this,"max_initial_timestamp_index",1)}}},"./src/models/whisper/processing_whisper.js":(e,r,t)=>{t.r(r),t.d(r,{WhisperProcessor:()=>u});var i=t("./src/models/auto/feature_extraction_auto.js"),o=t("./src/tokenizers.js"),n=t("./src/base/processing_utils.js");class u extends n.Processor{async _call(h){return await this.feature_extractor(h)}}de(u,"tokenizer_class",o.AutoTokenizer),de(u,"feature_extractor_class",i.AutoFeatureExtractor)},"./src/models/yolos/image_processing_yolos.js":(e,r,t)=>{t.r(r),t.d(r,{YolosFeatureExtractor:()=>n,YolosImageProcessor:()=>o});var i=t("./src/base/image_processors_utils.js");class o extends i.ImageProcessor{post_process_object_detection(...d){return(0,i.post_process_object_detection)(...d)}}class n extends o{}},"./src/ops/registry.js":(e,r,t)=>{t.r(r),t.d(r,{TensorOpRegistry:()=>h});var i=t("./src/backends/onnx.js"),o=t("./src/utils/tensor.js"),n=t("./src/env.js");const u=n.apis.IS_BROWSER_ENV||n.apis.IS_WEBWORKER_ENV,d=async(f,y,m)=>{const g=await(0,i.createInferenceSession)(new Uint8Array(f),y);let T=Promise.resolve();return async M=>{const I=(0,i.isONNXProxy)(),z=Object.fromEntries(Object.entries(M).map(([C,D])=>[C,(I?D.clone():D).ort_tensor])),E=await(T=u?T.then(()=>g.run(z)):g.run(z));return Array.isArray(m)?m.map(C=>new o.Tensor(E[C])):new o.Tensor(E[m])}};class h{static get nearest_interpolate_4d(){return this._nearest_interpolate_4d||(this._nearest_interpolate_4d=d([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=d([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=d([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=d([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=d([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=d([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=d([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=d([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}}de(h,"session_options",{})},"./src/pipelines.js":(e,r,t)=>{t.r(r),t.d(r,{AudioClassificationPipeline:()=>se,AutomaticSpeechRecognitionPipeline:()=>pe,BackgroundRemovalPipeline:()=>le,DepthEstimationPipeline:()=>Ae,DocumentQuestionAnsweringPipeline:()=>N,FeatureExtractionPipeline:()=>U,FillMaskPipeline:()=>D,ImageClassificationPipeline:()=>oe,ImageFeatureExtractionPipeline:()=>te,ImageSegmentationPipeline:()=>ge,ImageToImagePipeline:()=>ue,ImageToTextPipeline:()=>re,ObjectDetectionPipeline:()=>Ce,Pipeline:()=>I,QuestionAnsweringPipeline:()=>C,SummarizationPipeline:()=>$,Text2TextGenerationPipeline:()=>A,TextClassificationPipeline:()=>z,TextGenerationPipeline:()=>O,TextToAudioPipeline:()=>Q,TokenClassificationPipeline:()=>E,TranslationPipeline:()=>P,ZeroShotAudioClassificationPipeline:()=>K,ZeroShotClassificationPipeline:()=>R,ZeroShotImageClassificationPipeline:()=>Se,ZeroShotObjectDetectionPipeline:()=>q,pipeline:()=>Ze});var i=t("./src/tokenizers.js"),o=t("./src/models.js"),n=t("./src/models/auto/processing_auto.js");t("./src/base/processing_utils.js");var u=t("./src/utils/generic.js"),d=t("./src/utils/core.js"),h=t("./src/utils/maths.js"),f=t("./src/utils/audio.js"),y=t("./src/utils/tensor.js"),m=t("./src/utils/image.js");async function g(Ee){return Array.isArray(Ee)||(Ee=[Ee]),await Promise.all(Ee.map(Z=>m.RawImage.read(Z)))}async function T(Ee,Z){return Array.isArray(Ee)||(Ee=[Ee]),await Promise.all(Ee.map(me=>typeof me=="string"||me instanceof URL?(0,f.read_audio)(me,Z):me instanceof Float64Array?new Float32Array(me):me))}function M(Ee,Z){Z&&(Ee=Ee.map(Ve=>Ve|0));const[me,X,Oe,Fe]=Ee;return{xmin:me,ymin:X,xmax:Oe,ymax:Fe}}class I extends u.Callable{constructor({task:Z,model:me,tokenizer:X=null,processor:Oe=null}){super(),this.task=Z,this.model=me,this.tokenizer=X,this.processor=Oe}async dispose(){await this.model.dispose()}}class z extends I{constructor(Z){super(Z)}async _call(Z,{top_k:me=1}={}){const X=this.tokenizer(Z,{padding:!0,truncation:!0}),Oe=await this.model(X),Fe=this.model.config.problem_type==="multi_label_classification"?Ue=>Ue.sigmoid():Ue=>new y.Tensor("float32",(0,h.softmax)(Ue.data),Ue.dims),Ve=this.model.config.id2label,Re=[];for(const Ue of Oe.logits){const Ge=Fe(Ue),Je=await(0,y.topk)(Ge,me),at=Je[0].tolist(),Ke=Je[1].tolist().map((st,ft)=>({label:Ve?Ve[st]:`LABEL_${st}`,score:at[ft]}));me===1?Re.push(...Ke):Re.push(Ke)}return Array.isArray(Z)||me===1?Re:Re[0]}}class E extends I{constructor(Z){super(Z)}async _call(Z,{ignore_labels:me=["O"]}={}){const X=Array.isArray(Z),Oe=this.tokenizer(X?Z:[Z],{padding:!0,truncation:!0}),Ve=(await this.model(Oe)).logits,Re=this.model.config.id2label,Ue=[];for(let Ge=0;GeHe==this.tokenizer.sep_token_id);Ue[at].map((He,xt)=>He==1&&(xt===0||xt>Ke&&Ge.findIndex(Vt=>Vt==ot[xt])===-1));const st=Fe[at].tolist(),ft=Ve[at].tolist();for(let He=1;Hext==ot[He])!==-1)&&(st[He]=-1/0,ft[He]=-1/0);const yt=(0,h.softmax)(st).map((He,xt)=>[He,xt]),St=(0,h.softmax)(ft).map((He,xt)=>[He,xt]);yt[0][0]=0,St[0][0]=0;const vt=(0,d.product)(yt,St).filter(He=>He[0][1]<=He[1][1]).map(He=>[He[0][1],He[1][1],He[0][0]*He[1][0]]).sort((He,xt)=>xt[2]-He[2]);for(let He=0;Hest==this.tokenizer.mask_token_id);if(Ge===-1)throw Error(`Mask token (${this.tokenizer.mask_token}) not found in text.`);const Je=Oe[Re][Ge],at=await(0,y.topk)(new y.Tensor("float32",(0,h.softmax)(Je.data),Je.dims),me),ot=at[0].tolist(),Ke=at[1].tolist();Fe.push(Ke.map((st,ft)=>{const yt=Ue.slice();return yt[Ge]=st,{score:ot[ft],token:Number(st),token_str:this.tokenizer.decode([st]),sequence:this.tokenizer.decode(yt,{skip_special_tokens:!0})}}))}return Array.isArray(Z)?Fe:Fe[0]}}class A extends I{constructor(me){super(me);de(this,"_key","generated_text")}async _call(me,X={}){Array.isArray(me)||(me=[me]),this.model.config.prefix&&(me=me.map(Ge=>this.model.config.prefix+Ge));const Oe=this.model.config.task_specific_params;Oe&&Oe[this.task]&&Oe[this.task].prefix&&(me=me.map(Ge=>Oe[this.task].prefix+Ge));const Fe=this.tokenizer,Ve={padding:!0,truncation:!0};let Re;this instanceof P&&"_build_translation_inputs"in Fe?Re=Fe._build_translation_inputs(me,Ve,X):Re=Fe(me,Ve);const Ue=await this.model.generate({...Re,...X});return Fe.batch_decode(Ue,{skip_special_tokens:!0}).map(Ge=>({[this._key]:Ge}))}}class $ extends A{constructor(me){super(me);de(this,"_key","summary_text")}}class P extends A{constructor(me){super(me);de(this,"_key","translation_text")}}function k(Ee){return Array.isArray(Ee)&&Ee.every(Z=>"role"in Z&&"content"in Z)}class O extends I{constructor(Z){super(Z)}async _call(Z,me={}){let X=!1,Oe=!1,Fe;if(typeof Z=="string")Fe=Z=[Z];else if(Array.isArray(Z)&&Z.every(Ke=>typeof Ke=="string"))X=!0,Fe=Z;else{if(k(Z))Z=[Z];else if(Array.isArray(Z)&&Z.every(k))X=!0;else throw new Error("Input must be a string, an array of strings, a Chat, or an array of Chats");Oe=!0,Fe=Z.map(Ke=>this.tokenizer.apply_chat_template(Ke,{tokenize:!1,add_generation_prompt:!0}))}const Ve=me.add_special_tokens??!1,Re=Oe?!1:me.return_full_text??!0;this.tokenizer.padding_side="left";const Ue=this.tokenizer(Fe,{add_special_tokens:Ve,padding:!0,truncation:!0}),Ge=await this.model.generate({...Ue,...me}),Je=this.tokenizer.batch_decode(Ge,{skip_special_tokens:!0});let at;!Re&&Ue.input_ids.dims.at(-1)>0&&(at=this.tokenizer.batch_decode(Ue.input_ids,{skip_special_tokens:!0}).map(Ke=>Ke.length));const ot=Array.from({length:Z.length},Ke=>[]);for(let Ke=0;Ke[me.toLowerCase(),X])),this.entailment_id=this.label2id.entailment,this.entailment_id===void 0&&(console.warn("Could not find 'entailment' in label2id mapping. Using 2 as entailment_id."),this.entailment_id=2),this.contradiction_id=this.label2id.contradiction??this.label2id.not_entailment,this.contradiction_id===void 0&&(console.warn("Could not find 'contradiction' in label2id mapping. Using 0 as contradiction_id."),this.contradiction_id=0)}async _call(Z,me,{hypothesis_template:X="This example is {}.",multi_label:Oe=!1}={}){const Fe=Array.isArray(Z);Fe||(Z=[Z]),Array.isArray(me)||(me=[me]);const Ve=me.map(Ge=>X.replace("{}",Ge)),Re=Oe||me.length===1,Ue=[];for(const Ge of Z){const Je=[];for(const Ke of Ve){const st=this.tokenizer(Ge,{text_pair:Ke,padding:!0,truncation:!0}),ft=await this.model(st);Re?Je.push([ft.logits.data[this.contradiction_id],ft.logits.data[this.entailment_id]]):Je.push(ft.logits.data[this.entailment_id])}const ot=(Re?Je.map(Ke=>(0,h.softmax)(Ke)[1]):(0,h.softmax)(Je)).map((Ke,st)=>[Ke,st]).sort((Ke,st)=>st[0]-Ke[0]);Ue.push({sequence:Ge,labels:ot.map(Ke=>me[Ke[1]]),scores:ot.map(Ke=>Ke[0])})}return Fe?Ue:Ue[0]}}class U extends I{constructor(Z){super(Z)}async _call(Z,{pooling:me="none",normalize:X=!1,quantize:Oe=!1,precision:Fe="binary"}={}){const Ve=this.tokenizer(Z,{padding:!0,truncation:!0}),Re=await this.model(Ve);let Ue=Re.last_hidden_state??Re.logits??Re.token_embeddings;if(me!=="none")if(me==="mean")Ue=(0,y.mean_pooling)(Ue,Ve.attention_mask);else if(me==="cls")Ue=Ue.slice(null,0);else throw Error(`Pooling method '${me}' not supported.`);return X&&(Ue=Ue.normalize(2,-1)),Oe&&(Ue=(0,y.quantize_embeddings)(Ue,Fe)),Ue}}class te extends I{constructor(Z){super(Z)}async _call(Z,{pool:me=null}={}){const X=await g(Z),{pixel_values:Oe}=await this.processor(X),Fe=await this.model({pixel_values:Oe});let Ve;if(me){if(!("pooler_output"in Fe))throw Error("No pooled output was returned. Make sure the model has a 'pooler' layer when using the 'pool' option.");Ve=Fe.pooler_output}else Ve=Fe.last_hidden_state??Fe.logits??Fe.image_embeds;return Ve}}class se extends I{constructor(Z){super(Z)}async _call(Z,{top_k:me=5}={}){const X=this.processor.feature_extractor.config.sampling_rate,Oe=await T(Z,X),Fe=this.model.config.id2label,Ve=[];for(const Re of Oe){const Ue=await this.processor(Re),Je=(await this.model(Ue)).logits[0],at=await(0,y.topk)(new y.Tensor("float32",(0,h.softmax)(Je.data),Je.dims),me),ot=at[0].tolist(),st=at[1].tolist().map((ft,yt)=>({label:Fe?Fe[ft]:`LABEL_${ft}`,score:ot[yt]}));Ve.push(st)}return Array.isArray(Z)?Ve:Ve[0]}}class K extends I{constructor(Z){super(Z)}async _call(Z,me,{hypothesis_template:X="This is a sound of {}."}={}){const Oe=!Array.isArray(Z);Oe&&(Z=[Z]);const Fe=me.map(Je=>X.replace("{}",Je)),Ve=this.tokenizer(Fe,{padding:!0,truncation:!0}),Re=this.processor.feature_extractor.config.sampling_rate,Ue=await T(Z,Re),Ge=[];for(const Je of Ue){const at=await this.processor(Je),ot=await this.model({...Ve,...at}),Ke=(0,h.softmax)(ot.logits_per_audio.data);Ge.push([...Ke].map((st,ft)=>({score:st,label:me[ft]})))}return Oe?Ge[0]:Ge}}class pe extends I{constructor(Z){super(Z)}async _call(Z,me={}){switch(this.model.config.model_type){case"whisper":case"lite-whisper":return this._call_whisper(Z,me);case"wav2vec2":case"wav2vec2-bert":case"unispeech":case"unispeech-sat":case"hubert":return this._call_wav2vec2(Z,me);case"moonshine":return this._call_moonshine(Z,me);default:throw new Error(`AutomaticSpeechRecognitionPipeline does not support model type '${this.model.config.model_type}'.`)}}async _call_wav2vec2(Z,me){me.language&&console.warn('`language` parameter is not yet supported for `wav2vec2` models, defaulting to "English".'),me.task&&console.warn('`task` parameter is not yet supported for `wav2vec2` models, defaulting to "transcribe".');const X=!Array.isArray(Z);X&&(Z=[Z]);const Oe=this.processor.feature_extractor.config.sampling_rate,Fe=await T(Z,Oe),Ve=[];for(const Re of Fe){const Ue=await this.processor(Re),Je=(await this.model(Ue)).logits[0],at=[];for(const Ke of Je)at.push((0,h.max)(Ke.data)[1]);const ot=this.tokenizer.decode(at);Ve.push({text:ot})}return X?Ve[0]:Ve}async _call_whisper(Z,me){const X=me.return_timestamps??!1,Oe=me.chunk_length_s??0,Fe=me.force_full_sequences??!1;let Ve=me.stride_length_s??null;const Re={...me};X==="word"&&(Re.return_token_timestamps=!0,Re.return_timestamps=!1);const Ue=!Array.isArray(Z);Ue&&(Z=[Z]);const Ge=this.processor.feature_extractor.config.chunk_length/this.model.config.max_source_positions,Je=this.processor.feature_extractor.config.hop_length,at=this.processor.feature_extractor.config.sampling_rate,ot=await T(Z,at),Ke=[];for(const st of ot){let ft=[];if(Oe>0){if(Ve===null)Ve=Oe/6;else if(Oe<=Ve)throw Error("`chunk_length_s` must be larger than `stride_length_s`.");const vt=at*Oe,He=at*Ve,xt=vt-2*He;let Vt=0;for(;;){const he=Vt+vt,ce=st.subarray(Vt,he),Cr=await this.processor(ce),er=Vt===0,ht=he>=st.length;if(ft.push({stride:[ce.length,er?0:He,ht?0:He],input_features:Cr.input_features,is_last:ht}),ht)break;Vt+=xt}}else ft=[{stride:[st.length,0,0],input_features:(await this.processor(st)).input_features,is_last:!0}];for(const vt of ft){Re.num_frames=Math.floor(vt.stride[0]/Je);const He=await this.model.generate({inputs:vt.input_features,...Re});X==="word"?(vt.tokens=He.sequences.tolist()[0],vt.token_timestamps=He.token_timestamps.tolist()[0].map(xt=>(0,h.round)(xt,2))):vt.tokens=He[0].tolist(),vt.stride=vt.stride.map(xt=>xt/at)}const[yt,St]=this.tokenizer._decode_asr(ft,{time_precision:Ge,return_timestamps:X,force_full_sequences:Fe});Ke.push({text:yt,...St})}return Ue?Ke[0]:Ke}async _call_moonshine(Z,me){const X=!Array.isArray(Z);X&&(Z=[Z]);const Oe=this.processor.feature_extractor.config.sampling_rate,Fe=await T(Z,Oe),Ve=[];for(const Re of Fe){const Ue=await this.processor(Re),Ge=Math.floor(Re.length/Oe)*6,Je=await this.model.generate({max_new_tokens:Ge,...me,...Ue}),at=this.processor.batch_decode(Je,{skip_special_tokens:!0})[0];Ve.push({text:at})}return X?Ve[0]:Ve}}class re extends I{constructor(Z){super(Z)}async _call(Z,me={}){const X=Array.isArray(Z),Oe=await g(Z),{pixel_values:Fe}=await this.processor(Oe),Ve=[];for(const Re of Fe){Re.dims=[1,...Re.dims];const Ue=await this.model.generate({inputs:Re,...me}),Ge=this.tokenizer.batch_decode(Ue,{skip_special_tokens:!0}).map(Je=>({generated_text:Je.trim()}));Ve.push(Ge)}return X?Ve:Ve[0]}}class oe extends I{constructor(Z){super(Z)}async _call(Z,{top_k:me=5}={}){const X=await g(Z),{pixel_values:Oe}=await this.processor(X),Fe=await this.model({pixel_values:Oe}),Ve=this.model.config.id2label,Re=[];for(const Ue of Fe.logits){const Ge=await(0,y.topk)(new y.Tensor("float32",(0,h.softmax)(Ue.data),Ue.dims),me),Je=Ge[0].tolist(),ot=Ge[1].tolist().map((Ke,st)=>({label:Ve?Ve[Ke]:`LABEL_${Ke}`,score:Je[st]}));Re.push(ot)}return Array.isArray(Z)?Re:Re[0]}}class ge extends I{constructor(Z){super(Z),this.subtasks_mapping={panoptic:"post_process_panoptic_segmentation",instance:"post_process_instance_segmentation",semantic:"post_process_semantic_segmentation"}}async _call(Z,{threshold:me=.5,mask_threshold:X=.5,overlap_mask_area_threshold:Oe=.8,label_ids_to_fuse:Fe=null,target_sizes:Ve=null,subtask:Re=null}={}){if(Array.isArray(Z)&&Z.length!==1)throw Error("Image segmentation pipeline currently only supports a batch size of 1.");const Ge=await g(Z),Je=Ge.map(vt=>[vt.height,vt.width]),at=await this.processor(Ge),{inputNames:ot,outputNames:Ke}=this.model.sessions.model;if(!ot.includes("pixel_values")){if(ot.length!==1)throw Error(`Expected a single input name, but got ${ot.length} inputs: ${ot}.`);const vt=ot[0];if(vt in at)throw Error(`Input name ${vt} already exists in the inputs.`);at[vt]=at.pixel_values}const st=await this.model(at);let ft=null;if(Re!==null)ft=this.subtasks_mapping[Re];else if(this.processor.image_processor){for(const[vt,He]of Object.entries(this.subtasks_mapping))if(He in this.processor.image_processor){ft=this.processor.image_processor[He].bind(this.processor.image_processor),Re=vt;break}}const yt=this.model.config.id2label,St=[];if(Re)if(Re==="panoptic"||Re==="instance"){const vt=ft(st,me,X,Oe,Fe,Ve??Je)[0],He=vt.segmentation;for(const xt of vt.segments_info){const Vt=new Uint8ClampedArray(He.data.length);for(let ce=0;ceCr<-1e-5||Cr>1+1e-5)&&he.sigmoid_();const ce=await m.RawImage.fromTensor(he.mul_(255).to("uint8")).resize(Vt[1],Vt[0]);St.push({label:null,score:null,mask:ce})}}return St}}class le extends ge{constructor(Z){super(Z)}async _call(Z,me={}){if(Array.isArray(Z)&&Z.length!==1)throw Error("Background removal pipeline currently only supports a batch size of 1.");const Oe=await g(Z),Fe=await super._call(Z,me);return Oe.map((Re,Ue)=>{const Ge=Re.clone();return Ge.putAlpha(Fe[Ue].mask),Ge})}}class Se extends I{constructor(Z){super(Z)}async _call(Z,me,{hypothesis_template:X="This is a photo of {}"}={}){const Oe=Array.isArray(Z),Fe=await g(Z),Ve=me.map(ot=>X.replace("{}",ot)),Re=this.tokenizer(Ve,{padding:this.model.config.model_type==="siglip"?"max_length":!0,truncation:!0}),{pixel_values:Ue}=await this.processor(Fe),Ge=await this.model({...Re,pixel_values:Ue}),Je=this.model.config.model_type==="siglip"?ot=>ot.sigmoid().data:ot=>(0,h.softmax)(ot.data),at=[];for(const ot of Ge.logits_per_image){const st=[...Je(ot)].map((ft,yt)=>({score:ft,label:me[yt]}));st.sort((ft,yt)=>yt.score-ft.score),at.push(st)}return Oe?at:at[0]}}class Ce extends I{constructor(Z){super(Z)}async _call(Z,{threshold:me=.9,percentage:X=!1}={}){const Oe=Array.isArray(Z);if(Oe&&Z.length!==1)throw Error("Object detection pipeline currently only supports a batch size of 1.");const Fe=await g(Z),Ve=X?null:Fe.map(Ke=>[Ke.height,Ke.width]),{pixel_values:Re,pixel_mask:Ue}=await this.processor(Fe),Ge=await this.model({pixel_values:Re,pixel_mask:Ue}),Je=this.processor.image_processor.post_process_object_detection(Ge,me,Ve),at=this.model.config.id2label,ot=Je.map(Ke=>Ke.boxes.map((st,ft)=>({score:Ke.scores[ft],label:at[Ke.classes[ft]],box:M(st,!X)})));return Oe?ot:ot[0]}}class q extends I{constructor(Z){super(Z)}async _call(Z,me,{threshold:X=.1,top_k:Oe=null,percentage:Fe=!1}={}){const Ve=Array.isArray(Z),Re=await g(Z),Ue=this.tokenizer(me,{padding:!0,truncation:!0}),Ge=await this.processor(Re),Je=[];for(let at=0;at({score:St.scores[He],label:St.labels[He],box:M(vt,!Fe)}))}else{const St=this.processor.image_processor.post_process_object_detection(ft,X,Ke,!0)[0];yt=St.boxes.map((vt,He)=>({score:St.scores[He],label:me[St.classes[He]],box:M(vt,!Fe)}))}yt.sort((St,vt)=>vt.score-St.score),Oe!==null&&(yt=yt.slice(0,Oe)),Je.push(yt)}return Ve?Je:Je[0]}}class N extends I{constructor(Z){super(Z)}async _call(Z,me,X={}){const Oe=(await g(Z))[0],{pixel_values:Fe}=await this.processor(Oe),Ve=`${me}`,Re=this.tokenizer(Ve,{add_special_tokens:!1,padding:!0,truncation:!0}).input_ids,Ue=await this.model.generate({inputs:Fe,max_length:this.model.config.decoder.max_position_embeddings,decoder_input_ids:Re,...X}),Je=this.tokenizer.batch_decode(Ue)[0].match(/(.*?)<\/s_answer>/);let at=null;return Je&&Je.length>=2&&(at=Je[1].trim()),[{answer:at}]}}class Q extends I{constructor(me){super(me);de(this,"DEFAULT_VOCODER_ID","Xenova/speecht5_hifigan");this.vocoder=me.vocoder??null}async _call(me,{speaker_embeddings:X=null}={}){return this.processor?this._call_text_to_spectrogram(me,{speaker_embeddings:X}):this._call_text_to_waveform(me)}async _call_text_to_waveform(me){const X=this.tokenizer(me,{padding:!0,truncation:!0}),{waveform:Oe}=await this.model(X),Fe=this.model.config.sampling_rate;return new f.RawAudio(Oe.data,Fe)}async _call_text_to_spectrogram(me,{speaker_embeddings:X}){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"})),(typeof X=="string"||X instanceof URL)&&(X=new Float32Array(await(await fetch(X)).arrayBuffer())),X instanceof Float32Array)X=new y.Tensor("float32",X,[1,X.length]);else if(!(X instanceof y.Tensor))throw new Error("Speaker embeddings must be a `Tensor`, `Float32Array`, `string`, or `URL`.");const{input_ids:Oe}=this.tokenizer(me,{padding:!0,truncation:!0}),{waveform:Fe}=await this.model.generate_speech(Oe,X,{vocoder:this.vocoder}),Ve=this.processor.feature_extractor.config.sampling_rate;return new f.RawAudio(Fe.data,Ve)}}class ue extends I{constructor(Z){super(Z)}async _call(Z){const me=await g(Z),X=await this.processor(me),Oe=await this.model(X),Fe=[];for(const Ve of Oe.reconstruction){const Re=Ve.squeeze().clamp_(0,1).mul_(255).round_().to("uint8");Fe.push(m.RawImage.fromTensor(Re))}return Fe.length>1?Fe:Fe[0]}}class Ae extends I{constructor(Z){super(Z)}async _call(Z){const me=await g(Z),X=await this.processor(me),{predicted_depth:Oe}=await this.model(X),Fe=[];for(let Ve=0;Ve1?Fe:Fe[0]}}const we=Object.freeze({"text-classification":{tokenizer:i.AutoTokenizer,pipeline:z,model:o.AutoModelForSequenceClassification,default:{model:"Xenova/distilbert-base-uncased-finetuned-sst-2-english"},type:"text"},"token-classification":{tokenizer:i.AutoTokenizer,pipeline:E,model:o.AutoModelForTokenClassification,default:{model:"Xenova/bert-base-multilingual-cased-ner-hrl"},type:"text"},"question-answering":{tokenizer:i.AutoTokenizer,pipeline:C,model:o.AutoModelForQuestionAnswering,default:{model:"Xenova/distilbert-base-cased-distilled-squad"},type:"text"},"fill-mask":{tokenizer:i.AutoTokenizer,pipeline:D,model:o.AutoModelForMaskedLM,default:{model:"Xenova/bert-base-uncased"},type:"text"},summarization:{tokenizer:i.AutoTokenizer,pipeline:$,model:o.AutoModelForSeq2SeqLM,default:{model:"Xenova/distilbart-cnn-6-6"},type:"text"},translation:{tokenizer:i.AutoTokenizer,pipeline:P,model:o.AutoModelForSeq2SeqLM,default:{model:"Xenova/t5-small"},type:"text"},"text2text-generation":{tokenizer:i.AutoTokenizer,pipeline:A,model:o.AutoModelForSeq2SeqLM,default:{model:"Xenova/flan-t5-small"},type:"text"},"text-generation":{tokenizer:i.AutoTokenizer,pipeline:O,model:o.AutoModelForCausalLM,default:{model:"Xenova/gpt2"},type:"text"},"zero-shot-classification":{tokenizer:i.AutoTokenizer,pipeline:R,model:o.AutoModelForSequenceClassification,default:{model:"Xenova/distilbert-base-uncased-mnli"},type:"text"},"audio-classification":{pipeline:se,model:o.AutoModelForAudioClassification,processor:n.AutoProcessor,default:{model:"Xenova/wav2vec2-base-superb-ks"},type:"audio"},"zero-shot-audio-classification":{tokenizer:i.AutoTokenizer,pipeline:K,model:o.AutoModel,processor:n.AutoProcessor,default:{model:"Xenova/clap-htsat-unfused"},type:"multimodal"},"automatic-speech-recognition":{tokenizer:i.AutoTokenizer,pipeline:pe,model:[o.AutoModelForSpeechSeq2Seq,o.AutoModelForCTC],processor:n.AutoProcessor,default:{model:"Xenova/whisper-tiny.en"},type:"multimodal"},"text-to-audio":{tokenizer:i.AutoTokenizer,pipeline:Q,model:[o.AutoModelForTextToWaveform,o.AutoModelForTextToSpectrogram],processor:[n.AutoProcessor,null],default:{model:"Xenova/speecht5_tts"},type:"text"},"image-to-text":{tokenizer:i.AutoTokenizer,pipeline:re,model:o.AutoModelForVision2Seq,processor:n.AutoProcessor,default:{model:"Xenova/vit-gpt2-image-captioning"},type:"multimodal"},"image-classification":{pipeline:oe,model:o.AutoModelForImageClassification,processor:n.AutoProcessor,default:{model:"Xenova/vit-base-patch16-224"},type:"multimodal"},"image-segmentation":{pipeline:ge,model:[o.AutoModelForImageSegmentation,o.AutoModelForSemanticSegmentation,o.AutoModelForUniversalSegmentation],processor:n.AutoProcessor,default:{model:"Xenova/detr-resnet-50-panoptic"},type:"multimodal"},"background-removal":{pipeline:le,model:[o.AutoModelForImageSegmentation,o.AutoModelForSemanticSegmentation,o.AutoModelForUniversalSegmentation],processor:n.AutoProcessor,default:{model:"Xenova/modnet"},type:"image"},"zero-shot-image-classification":{tokenizer:i.AutoTokenizer,pipeline:Se,model:o.AutoModel,processor:n.AutoProcessor,default:{model:"Xenova/clip-vit-base-patch32"},type:"multimodal"},"object-detection":{pipeline:Ce,model:o.AutoModelForObjectDetection,processor:n.AutoProcessor,default:{model:"Xenova/detr-resnet-50"},type:"multimodal"},"zero-shot-object-detection":{tokenizer:i.AutoTokenizer,pipeline:q,model:o.AutoModelForZeroShotObjectDetection,processor:n.AutoProcessor,default:{model:"Xenova/owlvit-base-patch32"},type:"multimodal"},"document-question-answering":{tokenizer:i.AutoTokenizer,pipeline:N,model:o.AutoModelForDocumentQuestionAnswering,processor:n.AutoProcessor,default:{model:"Xenova/donut-base-finetuned-docvqa"},type:"multimodal"},"image-to-image":{pipeline:ue,model:o.AutoModelForImageToImage,processor:n.AutoProcessor,default:{model:"Xenova/swin2SR-classical-sr-x2-64"},type:"image"},"depth-estimation":{pipeline:Ae,model:o.AutoModelForDepthEstimation,processor:n.AutoProcessor,default:{model:"Xenova/dpt-large"},type:"image"},"feature-extraction":{tokenizer:i.AutoTokenizer,pipeline:U,model:o.AutoModel,default:{model:"Xenova/all-MiniLM-L6-v2"},type:"text"},"image-feature-extraction":{processor:n.AutoProcessor,pipeline:te,model:[o.AutoModelForImageFeatureExtraction,o.AutoModel],default:{model:"Xenova/vit-base-patch16-224-in21k"},type:"image"}}),ae=Object.freeze({"sentiment-analysis":"text-classification",ner:"token-classification",asr:"automatic-speech-recognition","text-to-speech":"text-to-audio",embeddings:"feature-extraction"});async function Ze(Ee,Z=null,{progress_callback:me=null,config:X=null,cache_dir:Oe=null,local_files_only:Fe=!1,revision:Ve="main",device:Re=null,dtype:Ue=null,subfolder:Ge="onnx",use_external_data_format:Je=null,model_file_name:at=null,session_options:ot={}}={}){Ee=ae[Ee]??Ee;const Ke=we[Ee.split("_",1)[0]];if(!Ke)throw Error(`Unsupported pipeline: ${Ee}. Must be one of [${Object.keys(we)}]`);Z||(Z=Ke.default.model,console.log(`No model specified. Using default model: "${Z}".`));const st={progress_callback:me,config:X,cache_dir:Oe,local_files_only:Fe,revision:Ve,device:Re,dtype:Ue,subfolder:Ge,use_external_data_format:Je,model_file_name:at,session_options:ot},ft=new Map([["tokenizer",Ke.tokenizer],["model",Ke.model],["processor",Ke.processor]]),yt=await et(ft,Z,st);yt.task=Ee,(0,d.dispatchCallback)(me,{status:"ready",task:Ee,model:Z});const St=Ke.pipeline;return new St(yt)}async function et(Ee,Z,me){const X=Object.create(null),Oe=[];for(const[Fe,Ve]of Ee.entries()){if(!Ve)continue;let Re;Array.isArray(Ve)?Re=new Promise(async(Ue,Ge)=>{var at,ot;let Je;for(const Ke of Ve){if(Ke===null){Ue(null);return}try{Ue(await Ke.from_pretrained(Z,me));return}catch(st){if((at=st.message)!=null&&at.includes("Unsupported model type"))Je=st;else if((ot=st.message)!=null&&ot.includes("Could not locate file"))Je=st;else{Ge(st);return}}}Ge(Je)}):Re=Ve.from_pretrained(Z,me),X[Fe]=Re,Oe.push(Re)}await Promise.all(Oe);for(const[Fe,Ve]of Object.entries(X))X[Fe]=await Ve;return X}},"./src/tokenizers.js":(e,r,t)=>{t.r(r),t.d(r,{AlbertTokenizer:()=>Qt,AutoTokenizer:()=>qs,BartTokenizer:()=>tn,BertTokenizer:()=>Ft,BlenderbotSmallTokenizer:()=>Hr,BlenderbotTokenizer:()=>Dn,BloomTokenizer:()=>Ks,CLIPTokenizer:()=>gs,CamembertTokenizer:()=>xs,CodeGenTokenizer:()=>$s,CodeLlamaTokenizer:()=>Cs,CohereTokenizer:()=>Jr,ConvBertTokenizer:()=>Fr,DebertaTokenizer:()=>mt,DebertaV2Tokenizer:()=>tr,DistilBertTokenizer:()=>Kr,ElectraTokenizer:()=>Ts,EsmTokenizer:()=>bt,FalconTokenizer:()=>qe,GPT2Tokenizer:()=>rs,GPTNeoXTokenizer:()=>ut,GemmaTokenizer:()=>ls,Grok1Tokenizer:()=>Es,HerbertTokenizer:()=>rr,LlamaTokenizer:()=>Jt,M2M100Tokenizer:()=>Ss,MBart50Tokenizer:()=>Us,MBartTokenizer:()=>os,MPNetTokenizer:()=>Er,MarianTokenizer:()=>gn,MgpstrTokenizer:()=>yn,MobileBertTokenizer:()=>wt,NllbTokenizer:()=>Ps,NougatTokenizer:()=>rn,PreTrainedTokenizer:()=>Le,Qwen2Tokenizer:()=>cr,RoFormerTokenizer:()=>pr,RobertaTokenizer:()=>Gs,SiglipTokenizer:()=>_n,SpeechT5Tokenizer:()=>ys,SqueezeBertTokenizer:()=>Nt,T5Tokenizer:()=>en,TokenizerModel:()=>te,VitsTokenizer:()=>cs,Wav2Vec2CTCTokenizer:()=>Fn,WhisperTokenizer:()=>ks,XLMRobertaTokenizer:()=>Hs,XLMTokenizer:()=>Ws,is_chinese_char:()=>D});var i=t("./src/utils/generic.js"),o=t("./src/utils/core.js"),n=t("./src/utils/hub.js"),u=t("./src/utils/maths.js"),d=t("./src/utils/tensor.js"),h=t("./src/utils/data-structures.js"),f=t("./node_modules/@huggingface/jinja/dist/index.js"),y=t("./src/models/whisper/common_whisper.js");async function m(Te,L){const J=await Promise.all([(0,n.getModelJSON)(Te,"tokenizer.json",!0,L),(0,n.getModelJSON)(Te,"tokenizer_config.json",!0,L)]);return L.legacy!==null&&(J[1].legacy=L.legacy),J}function g(Te,L){const J=[];let ie=0;for(const ve of Te.matchAll(L)){const be=ve[0];ie0&&J.push(be),ie=ve.index+be.length}return ie=19968&&Te<=40959||Te>=13312&&Te<=19903||Te>=131072&&Te<=173791||Te>=173824&&Te<=177983||Te>=177984&&Te<=178207||Te>=178208&&Te<=183983||Te>=63744&&Te<=64255||Te>=194560&&Te<=195103}function A(Te,L,J){const ie=[];let ve=0;for(;vethis.tokens_to_ids.get(J)??this.unk_token_id)}convert_ids_to_tokens(L){return L.map(J=>this.vocab[J]??this.unk_token)}}class se extends te{constructor(L){super(L),this.tokens_to_ids=M(L.vocab),this.unk_token_id=this.tokens_to_ids.get(L.unk_token),this.unk_token=L.unk_token,this.max_input_chars_per_word=L.max_input_chars_per_word??100,this.vocab=new Array(this.tokens_to_ids.size);for(const[J,ie]of this.tokens_to_ids)this.vocab[ie]=J}encode(L){const J=[];for(const ie of L){const ve=[...ie];if(ve.length>this.max_input_chars_per_word){J.push(this.unk_token);continue}let be=!1,ze=0;const tt=[];for(;ze0&&(dt=this.config.continuing_subword_prefix+dt),this.tokens_to_ids.has(dt)){it=dt;break}--ct}if(it===null){be=!0;break}tt.push(it),ze=ct}be?J.push(this.unk_token):J.push(...tt)}return J}}class K extends te{constructor(L,J){super(L);const ie=L.vocab.length;this.vocab=new Array(ie),this.scores=new Array(ie);for(let ve=0;ve[ve,be])),this.bos_token=" ",this.bos_token_id=this.tokens_to_ids.get(this.bos_token),this.eos_token=J.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,u.min)(this.scores)[0],this.unk_score=this.minScore-10,this.scores[this.unk_token_id]=this.unk_score,this.trie=new h.CharTrie,this.trie.extend(this.vocab),this.fuse_unk=!0}populateNodes(L){const J=L.chars,ie=1;let ve=0;for(;ve{const Te=[...Array.from({length:94},(ve,be)=>be+33),...Array.from({length:12},(ve,be)=>be+161),...Array.from({length:82},(ve,be)=>be+174)],L=Te.slice();let J=0;for(let ve=0;ve<256;++ve)Te.includes(ve)||(Te.push(ve),L.push(256+J),J+=1);const ie=L.map(ve=>String.fromCharCode(ve));return Object.fromEntries(Te.map((ve,be)=>[ve,ie[be]]))})(),re=(0,o.reverseDictionary)(pe);class oe extends te{constructor(L){super(L),this.tokens_to_ids=M(L.vocab),this.unk_token_id=this.tokens_to_ids.get(L.unk_token),this.unk_token=L.unk_token,this.vocab=new Array(this.tokens_to_ids.size);for(const[ie,ve]of this.tokens_to_ids)this.vocab[ve]=ie;const J=Array.isArray(L.merges[0]);this.merges=J?L.merges:L.merges.map(ie=>ie.split(" ",2)),this.bpe_ranks=new Map(this.merges.map((ie,ve)=>[JSON.stringify(ie),ve])),this.end_of_word_suffix=L.end_of_word_suffix,this.continuing_subword_suffix=L.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 h.LRUCache(this.cache_capacity)}clear_cache(){this.cache.clear()}bpe(L){if(L.length===0)return[];const J=this.cache.get(L);if(J!==void 0)return J;const ie=Array.from(L);this.end_of_word_suffix&&(ie[ie.length-1]+=this.end_of_word_suffix);let ve=[];if(ie.length>1){const be=new h.PriorityQueue((ct,it)=>ct.score`<0x${tt.toString(16).toUpperCase().padStart(2,"0")}>`);ze.every(tt=>this.tokens_to_ids.has(tt))?J.push(...ze):J.push(this.unk_token)}else J.push(this.unk_token)}return J}}class ge extends te{constructor(L,J){super(L),this.tokens_to_ids=M(J.target_lang?L.vocab[J.target_lang]:L.vocab),this.bos_token=J.bos_token,this.bos_token_id=this.tokens_to_ids.get(this.bos_token),this.eos_token=J.eos_token,this.eos_token_id=this.tokens_to_ids.get(this.eos_token),this.pad_token=J.pad_token,this.pad_token_id=this.tokens_to_ids.get(this.pad_token),this.unk_token=J.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[ie,ve]of this.tokens_to_ids)this.vocab[ve]=ie}encode(L){return L}}class le extends i.Callable{constructor(L){super(),this.config=L}static fromConfig(L){if(L===null)return null;switch(L.type){case"BertNormalizer":return new Ee(L);case"Precompiled":return new ht(L);case"Sequence":return new et(L);case"Replace":return new Se(L);case"NFC":return new q(L);case"NFD":return new N(L);case"NFKC":return new Q(L);case"NFKD":return new ue(L);case"Strip":return new Ae(L);case"StripAccents":return new we(L);case"Lowercase":return new ae(L);case"Prepend":return new Ze(L);default:throw new Error(`Unknown Normalizer type: ${L.type}`)}}normalize(L){throw Error("normalize should be implemented in subclass.")}_call(L){return this.normalize(L)}}class Se extends le{normalize(L){const J=T(this.config.pattern);return J===null?L:L.replaceAll(J,this.config.content)}}class Ce extends le{constructor(){super(...arguments);de(this,"form")}normalize(J){return J=J.normalize(this.form),J}}class q extends Ce{constructor(){super(...arguments);de(this,"form","NFC")}}class N extends Ce{constructor(){super(...arguments);de(this,"form","NFD")}}class Q extends Ce{constructor(){super(...arguments);de(this,"form","NFKC")}}class ue extends Ce{constructor(){super(...arguments);de(this,"form","NFKD")}}class Ae extends le{normalize(L){return this.config.strip_left&&this.config.strip_right?L=L.trim():(this.config.strip_left&&(L=L.trimStart()),this.config.strip_right&&(L=L.trimEnd())),L}}class we extends le{normalize(L){return L=E(L),L}}class ae extends le{normalize(L){return L=L.toLowerCase(),L}}class Ze extends le{normalize(L){return L=this.config.prepend+L,L}}class et extends le{constructor(L){super(L),this.normalizers=L.normalizers.map(J=>le.fromConfig(J))}normalize(L){return this.normalizers.reduce((J,ie)=>ie.normalize(J),L)}}class Ee extends le{_tokenize_chinese_chars(L){const J=[];for(let ie=0;iethis.pre_tokenize_text(ie,J)):this.pre_tokenize_text(L,J)).flat()}_call(L,J){return this.pre_tokenize(L,J)}}class me extends Z{constructor(L){super(),this.pattern=new RegExp(`[^\\s${P}]+|[${P}]`,"gu")}pre_tokenize_text(L,J){return L.trim().match(this.pattern)||[]}}class X extends Z{constructor(L){super(),this.config=L,this.add_prefix_space=this.config.add_prefix_space,this.trim_offsets=this.config.trim_offsets,this.use_regex=this.config.use_regex??!0,this.pattern=new RegExp("'s|'t|'re|'ve|'m|'ll|'d| ?\\p{L}+| ?\\p{N}+| ?[^\\s\\p{L}\\p{N}]+|\\s+(?!\\S)|\\s+","gu"),this.byte_encoder=pe,this.text_encoder=new TextEncoder}pre_tokenize_text(L,J){return this.add_prefix_space&&!L.startsWith(" ")&&(L=" "+L),(this.use_regex?L.match(this.pattern)||[]:[L]).map(ve=>Array.from(this.text_encoder.encode(ve),be=>this.byte_encoder[be]).join(""))}}class Oe extends Z{constructor(L){super(),this.config=L,this.pattern=T(this.config.pattern,this.config.invert)}pre_tokenize_text(L,J){var ie;return this.pattern===null?[]:this.config.invert?L.match(this.pattern)||[]:((ie=this.config.behavior)==null?void 0:ie.toLowerCase())==="removed"?L.split(this.pattern).filter(ve=>ve):g(L,this.pattern)}}class Fe extends Z{constructor(L){super(),this.config=L,this.pattern=new RegExp(`[^${P}]+|[${P}]+`,"gu")}pre_tokenize_text(L,J){return L.match(this.pattern)||[]}}class Ve extends Z{constructor(L){super(),this.config=L;const J=`[^\\d]+|\\d${this.config.individual_digits?"":"+"}`;this.pattern=new RegExp(J,"gu")}pre_tokenize_text(L,J){return L.match(this.pattern)||[]}}class Re extends i.Callable{constructor(L){super(),this.config=L}static fromConfig(L){if(L===null)return null;switch(L.type){case"TemplateProcessing":return new Je(L);case"ByteLevel":return new at(L);case"RobertaProcessing":return new Ge(L);case"BertProcessing":return new Ue(L);case"Sequence":return new ot(L);default:throw new Error(`Unknown PostProcessor type: ${L.type}`)}}post_process(L,...J){throw Error("post_process should be implemented in subclass.")}_call(L,...J){return this.post_process(L,...J)}}class Ue extends Re{constructor(L){super(L),this.cls=L.cls[0],this.sep=L.sep[0]}post_process(L,J=null,{add_special_tokens:ie=!0}={}){ie&&(L=(0,o.mergeArrays)([this.cls],L,[this.sep]));let ve=new Array(L.length).fill(0);if(J!==null){const be=ie&&this instanceof Ge?[this.sep]:[],ze=ie?[this.sep]:[];L=(0,o.mergeArrays)(L,be,J,ze),ve=(0,o.mergeArrays)(ve,new Array(J.length+be.length+ze.length).fill(1))}return{tokens:L,token_type_ids:ve}}}class Ge extends Ue{}class Je extends Re{constructor(L){super(L),this.single=L.single,this.pair=L.pair}post_process(L,J=null,{add_special_tokens:ie=!0}={}){const ve=J===null?this.single:this.pair;let be=[],ze=[];for(const tt of ve)"SpecialToken"in tt?ie&&(be.push(tt.SpecialToken.id),ze.push(tt.SpecialToken.type_id)):"Sequence"in tt&&(tt.Sequence.id==="A"?(be=(0,o.mergeArrays)(be,L),ze=(0,o.mergeArrays)(ze,new Array(L.length).fill(tt.Sequence.type_id))):tt.Sequence.id==="B"&&(be=(0,o.mergeArrays)(be,J),ze=(0,o.mergeArrays)(ze,new Array(J.length).fill(tt.Sequence.type_id))));return{tokens:be,token_type_ids:ze}}}class at extends Re{post_process(L,J=null){return J&&(L=(0,o.mergeArrays)(L,J)),{tokens:L}}}class ot extends Re{constructor(L){super(L),this.processors=L.processors.map(J=>Re.fromConfig(J))}post_process(L,J=null,ie={}){let ve;for(const be of this.processors)if(be instanceof at)L=be.post_process(L).tokens,J&&(J=be.post_process(J).tokens);else{const ze=be.post_process(L,J,ie);L=ze.tokens,ve=ze.token_type_ids}return{tokens:L,token_type_ids:ve}}}class Ke extends i.Callable{constructor(L){super(),this.config=L,this.added_tokens=[],this.end_of_word_suffix=null,this.trim_offsets=L.trim_offsets}static fromConfig(L){if(L===null)return null;switch(L.type){case"WordPiece":return new vt(L);case"Metaspace":return new er(L);case"ByteLevel":return new He(L);case"Replace":return new st(L);case"ByteFallback":return new ft(L);case"Fuse":return new yt(L);case"Strip":return new St(L);case"Sequence":return new Vt(L);case"CTC":return new xt(L);case"BPEDecoder":return new he(L);default:throw new Error(`Unknown Decoder type: ${L.type}`)}}_call(L){return this.decode(L)}decode(L){return this.decode_chain(L).join("")}decode_chain(L){throw Error("`decode_chain` should be implemented in subclass.")}}class st extends Ke{decode_chain(L){const J=T(this.config.pattern);return J===null?L:L.map(ie=>ie.replaceAll(J,this.config.content))}}class ft extends Ke{constructor(L){super(L),this.text_decoder=new TextDecoder}decode_chain(L){const J=[];let ie=[];for(const ve of L){let be=null;if(ve.length===6&&ve.startsWith("<0x")&&ve.endsWith(">")){const ze=parseInt(ve.slice(3,5),16);isNaN(ze)||(be=ze)}if(be!==null)ie.push(be);else{if(ie.length>0){const ze=this.text_decoder.decode(Uint8Array.from(ie));J.push(ze),ie=[]}J.push(ve)}}if(ie.length>0){const ve=this.text_decoder.decode(Uint8Array.from(ie));J.push(ve),ie=[]}return J}}class yt extends Ke{decode_chain(L){return[L.join("")]}}class St extends Ke{constructor(L){super(L),this.content=this.config.content,this.start=this.config.start,this.stop=this.config.stop}decode_chain(L){return L.map(J=>{let ie=0;for(let be=0;be(ie!==0&&(J.startsWith(this.config.prefix)?J=J.replace(this.config.prefix,""):J=" "+J),this.cleanup&&(J=z(J)),J))}}class He extends Ke{constructor(L){super(L),this.byte_decoder=re,this.text_decoder=new TextDecoder("utf-8",{fatal:!1,ignoreBOM:!0}),this.end_of_word_suffix=null}convert_tokens_to_string(L){const J=L.join(""),ie=new Uint8Array([...J].map(be=>this.byte_decoder[be]));return this.text_decoder.decode(ie)}decode_chain(L){const J=[];let ie=[];for(const ve of L)this.added_tokens.find(be=>be.content===ve)!==void 0?(ie.length>0&&(J.push(this.convert_tokens_to_string(ie)),ie=[]),J.push(ve)):ie.push(ve);return ie.length>0&&J.push(this.convert_tokens_to_string(ie)),J}}class xt extends Ke{constructor(L){super(L),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(L){if(L.length===0)return"";const J=[L[0]];for(let be=1;bebe!==this.pad_token).join("");return this.cleanup&&(ve=z(ve).replaceAll(this.word_delimiter_token," ").trim()),ve}decode_chain(L){return[this.convert_tokens_to_string(L)]}}class Vt extends Ke{constructor(L){super(L),this.decoders=L.decoders.map(J=>Ke.fromConfig(J))}decode_chain(L){return this.decoders.reduce((J,ie)=>ie.decode_chain(J),L)}}class he extends Ke{constructor(L){super(L),this.suffix=this.config.suffix}decode_chain(L){return L.map((J,ie)=>J.replaceAll(this.suffix,ie===L.length-1?"":" "))}}class ce extends Ke{decode_chain(L){let J="";for(let ie=1;ieie.normalize("NFKC")).join("～"):L=L.normalize("NFKC"),L}}class as extends Z{constructor(L){super(),this.tokenizers=L.pretokenizers.map(J=>Z.fromConfig(J))}pre_tokenize_text(L,J){return this.tokenizers.reduce((ie,ve)=>ve.pre_tokenize(ie,J),[L])}}class W extends Z{constructor(L){super()}pre_tokenize_text(L,J){return L.match(/\w+|[^\w\s]+/g)||[]}}class ye extends Z{constructor(L){super()}pre_tokenize_text(L,J){return $(L)}}class G extends Z{constructor(L){super(),this.config=L,this.pattern=T(this.config.pattern),this.content=this.config.content}pre_tokenize_text(L,J){return this.pattern===null?[L]:[L.replaceAll(this.pattern,this.config.content)]}}const ne=["bos_token","eos_token","unk_token","sep_token","pad_token","cls_token","mask_token"];function $e(Te,L,J,ie){for(const ve of Object.keys(Te)){const be=L-Te[ve].length,ze=J(ve),tt=new Array(be).fill(ze);Te[ve]=ie==="right"?(0,o.mergeArrays)(Te[ve],tt):(0,o.mergeArrays)(tt,Te[ve])}}function Xe(Te,L){for(const J of Object.keys(Te))Te[J].length=L}class Le extends i.Callable{constructor(J,ie){super();de(this,"return_token_type_ids",!1);de(this,"padding_side","right");this._tokenizer_config=ie,this.normalizer=le.fromConfig(J.normalizer),this.pre_tokenizer=Z.fromConfig(J.pre_tokenizer),this.model=te.fromConfig(J.model,ie),this.post_processor=Re.fromConfig(J.post_processor),this.decoder=Ke.fromConfig(J.decoder),this.special_tokens=[],this.all_special_ids=[],this.added_tokens=[];for(const ve of J.added_tokens){const be=new U(ve);this.added_tokens.push(be),this.model.tokens_to_ids.set(be.content,be.id),this.model.vocab[be.id]=be.content,be.special&&(this.special_tokens.push(be.content),this.all_special_ids.push(be.id))}if(this.additional_special_tokens=ie.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 h.DictionarySplitter(this.added_tokens.map(ve=>ve.content)),this.added_tokens_map=new Map(this.added_tokens.map(ve=>[ve.content,ve])),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=ie.model_max_length,this.remove_space=ie.remove_space,this.clean_up_tokenization_spaces=ie.clean_up_tokenization_spaces??!0,this.do_lowercase_and_remove_accent=ie.do_lowercase_and_remove_accent??!1,ie.padding_side&&(this.padding_side=ie.padding_side),this.legacy=!1,this.chat_template=ie.chat_template??null,Array.isArray(this.chat_template)){const ve=Object.create(null);for(const{name:be,template:ze}of this.chat_template){if(typeof be!="string"||typeof ze!="string")throw new Error('Chat template must be a list of objects with "name" and "template" properties');ve[be]=ze}this.chat_template=ve}this._compiled_template_cache=new Map}getToken(...J){for(const ie of J){const ve=this._tokenizer_config[ie];if(ve)if(typeof ve=="object"){if(ve.__type==="AddedToken")return ve.content;throw Error(`Unknown token: ${ve}`)}else return ve}return null}static async from_pretrained(J,{progress_callback:ie=null,config:ve=null,cache_dir:be=null,local_files_only:ze=!1,revision:tt="main",legacy:ct=null}={}){const it=await m(J,{progress_callback:ie,config:ve,cache_dir:be,local_files_only:ze,revision:tt,legacy:ct});return new this(...it)}_call(J,{text_pair:ie=null,add_special_tokens:ve=!0,padding:be=!1,truncation:ze=null,max_length:tt=null,return_tensor:ct=!0,return_token_type_ids:it=null}={}){const dt=Array.isArray(J);let At;if(dt){if(J.length===0)throw Error("text array must be non-empty");if(ie!==null){if(Array.isArray(ie)){if(J.length!==ie.length)throw Error("text and text_pair must have the same length")}else throw Error("text_pair must also be an array");At=J.map((Ht,fr)=>this._encode_plus(Ht,{text_pair:ie[fr],add_special_tokens:ve,return_token_type_ids:it}))}else At=J.map(Ht=>this._encode_plus(Ht,{add_special_tokens:ve,return_token_type_ids:it}))}else{if(J==null)throw Error("text may not be null or undefined");if(Array.isArray(ie))throw Error("When specifying `text_pair`, since `text` is a string, `text_pair` must also be a string (i.e., not an array).");At=[this._encode_plus(J,{text_pair:ie,add_special_tokens:ve,return_token_type_ids:it})]}if(tt===null?tt=this.model_max_length:ze===null&&(be===!0?(console.warn("`max_length` is ignored when `padding: true` and there is no truncation strategy. To pad to max length, use `padding: 'max_length'`."),tt=this.model_max_length):be===!1&&(console.warn("Truncation was not explicitly activated but `max_length` is provided a specific value, please use `truncation: true` to explicitly truncate examples to max length."),ze=!0)),be===!0&&(tt=Math.min((0,u.max)(At.map(Ht=>Ht.input_ids.length))[0],tt??1/0)),tt=Math.min(tt,this.model_max_length??1/0),be||ze)for(let Ht=0;Httt?ze&&Xe(At[Ht],tt):be&&$e(At[Ht],tt,fr=>fr==="input_ids"?this.pad_token_id:0,this.padding_side));const qt={};if(ct){if(!(be&&ze)&&At.some(fr=>{var sr;for(const Pr of Object.keys(fr))if(fr[Pr].length!==((sr=At[0][Pr])==null?void 0:sr.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 Ht=[At.length,At[0].input_ids.length];for(const fr of Object.keys(At[0]))qt[fr]=new d.Tensor("int64",BigInt64Array.from(At.flatMap(sr=>sr[fr]).map(BigInt)),Ht)}else{for(const Ht of Object.keys(At[0]))qt[Ht]=At.map(fr=>fr[Ht]);if(!dt)for(const Ht of Object.keys(qt))qt[Ht]=qt[Ht][0]}return qt}_encode_text(J){if(J===null)return null;const ie=this.added_tokens_splitter.split(J);for(let be=0;be0&&(ie[be-1]=ie[be-1].trimEnd()),ze.rstrip&&be{if(be.length===0)return[];if(this.added_tokens_map.has(be))return[be];if(this.remove_space===!0&&(be=be.trim().split(/\s+/).join(" ")),this.do_lowercase_and_remove_accent&&(be=C(be)),this.normalizer!==null&&(be=this.normalizer(be)),be.length===0)return[];const tt=this.pre_tokenizer!==null?this.pre_tokenizer(be,{section_index:ze}):[be];return this.model(tt)})}_encode_plus(J,{text_pair:ie=null,add_special_tokens:ve=!0,return_token_type_ids:be=null}={}){const{tokens:ze,token_type_ids:tt}=this._tokenize_helper(J,{pair:ie,add_special_tokens:ve}),ct=this.model.convert_tokens_to_ids(ze),it={input_ids:ct,attention_mask:new Array(ct.length).fill(1)};return(be??this.return_token_type_ids)&&tt&&(it.token_type_ids=tt),it}_tokenize_helper(J,{pair:ie=null,add_special_tokens:ve=!1}={}){const be=this._encode_text(J),ze=this._encode_text(ie);return this.post_processor?this.post_processor(be,ze,{add_special_tokens:ve}):{tokens:(0,o.mergeArrays)(be??[],ze??[])}}tokenize(J,{pair:ie=null,add_special_tokens:ve=!1}={}){return this._tokenize_helper(J,{pair:ie,add_special_tokens:ve}).tokens}encode(J,{text_pair:ie=null,add_special_tokens:ve=!0,return_token_type_ids:be=null}={}){return this._encode_plus(J,{text_pair:ie,add_special_tokens:ve,return_token_type_ids:be}).input_ids}batch_decode(J,ie={}){return J instanceof d.Tensor&&(J=J.tolist()),J.map(ve=>this.decode(ve,ie))}decode(J,ie={}){if(J instanceof d.Tensor&&(J=I(J)),!Array.isArray(J)||J.length===0||!(0,o.isIntegralNumber)(J[0]))throw Error("token_ids must be a non-empty array of integers.");return this.decode_single(J,ie)}decode_single(J,{skip_special_tokens:ie=!1,clean_up_tokenization_spaces:ve=null}){let be=this.model.convert_ids_to_tokens(J);ie&&(be=be.filter(tt=>!this.special_tokens.includes(tt)));let ze=this.decoder?this.decoder(be):be.join(" ");return this.decoder&&this.decoder.end_of_word_suffix&&(ze=ze.replaceAll(this.decoder.end_of_word_suffix," "),ie&&(ze=ze.trim())),(ve??this.clean_up_tokenization_spaces)&&(ze=z(ze)),ze}get_chat_template({chat_template:J=null,tools:ie=null}={}){if(this.chat_template&&typeof this.chat_template=="object"){const ve=this.chat_template;if(J!==null&&Object.hasOwn(ve,J))J=ve[J];else if(J===null)if(ie!==null&&"tool_use"in ve)J=ve.tool_use;else if("default"in ve)J=ve.default;else throw Error(`This model has multiple chat templates with no default specified! Please either pass a chat template or the name of the template you wish to use to the 'chat_template' argument. Available template names are ${Object.keys(ve).sort()}.`)}else if(J===null)if(this.chat_template)J=this.chat_template;else throw Error("Cannot use apply_chat_template() because tokenizer.chat_template is not set and no template argument was passed! For information about writing templates and setting the tokenizer.chat_template attribute, please see the documentation at https://huggingface.co/docs/transformers/main/en/chat_templating");return J}apply_chat_template(J,{tools:ie=null,documents:ve=null,chat_template:be=null,add_generation_prompt:ze=!1,tokenize:tt=!0,padding:ct=!1,truncation:it=!1,max_length:dt=null,return_tensor:At=!0,return_dict:qt=!1,tokenizer_kwargs:Ht={},...fr}={}){if(be=this.get_chat_template({chat_template:be,tools:ie}),typeof be!="string")throw Error(`chat_template must be a string, but got ${typeof be}`);let sr=this._compiled_template_cache.get(be);sr===void 0&&(sr=new f.Template(be),this._compiled_template_cache.set(be,sr));const Pr=Object.create(null);for(const qr of ne){const ws=this.getToken(qr);ws&&(Pr[qr]=ws)}const $r=sr.render({messages:J,add_generation_prompt:ze,tools:ie,documents:ve,...Pr,...fr});if(tt){const qr=this._call($r,{add_special_tokens:!1,padding:ct,truncation:it,max_length:dt,return_tensor:At,...Ht});return qt?qr:qr.input_ids}return $r}}class Ft extends Le{constructor(){super(...arguments);de(this,"return_token_type_ids",!0)}}class Qt extends Le{constructor(){super(...arguments);de(this,"return_token_type_ids",!0)}}class wt extends Le{constructor(){super(...arguments);de(this,"return_token_type_ids",!0)}}class Nt extends Le{constructor(){super(...arguments);de(this,"return_token_type_ids",!0)}}class mt extends Le{constructor(){super(...arguments);de(this,"return_token_type_ids",!0)}}class tr extends Le{constructor(){super(...arguments);de(this,"return_token_type_ids",!0)}}class rr extends Le{constructor(){super(...arguments);de(this,"return_token_type_ids",!0)}}class Fr extends Le{constructor(){super(...arguments);de(this,"return_token_type_ids",!0)}}class pr extends Le{constructor(){super(...arguments);de(this,"return_token_type_ids",!0)}}class Kr extends Le{}class xs extends Le{}class Ws extends Le{constructor(J,ie){super(J,ie);de(this,"return_token_type_ids",!0);console.warn('WARNING: `XLMTokenizer` is not yet supported by Hugging Face\'s "fast" tokenizers library. Therefore, you may experience slightly inaccurate results.')}}class Ts extends Le{constructor(){super(...arguments);de(this,"return_token_type_ids",!0)}}class en extends Le{}class rs extends Le{}class tn extends Le{}class os extends Le{constructor(L,J){super(L,J),this.languageRegex=/^[a-z]{2}_[A-Z]{2}$/,this.language_codes=this.special_tokens.filter(ie=>this.languageRegex.test(ie)),this.lang_to_token=ie=>ie}_build_translation_inputs(L,J,ie){return Xr(this,L,J,ie)}}class Us extends os{}class Gs extends Le{}class Ks extends Le{}const Wr="▁";class Jt extends Le{constructor(J,ie){super(J,ie);de(this,"padding_side","left");this.legacy=ie.legacy??!0,this.legacy||(this.normalizer=null,this.pre_tokenizer=new Cr({replacement:Wr,add_prefix_space:!0,prepend_scheme:"first"}))}_encode_text(J){if(J===null)return null;if(this.legacy||J.length===0)return super._encode_text(J);let ie=super._encode_text(Wr+J.replaceAll(Wr," "));return ie.length>1&&ie[0]===Wr&&this.special_tokens.includes(ie[1])&&(ie=ie.slice(1)),ie}}class Cs extends Le{}class Hs extends Le{}class Er extends Le{}class qe extends Le{}class ut extends Le{}class bt extends Le{}class cr extends Le{}class ls extends Le{}class Es extends Le{}function Xr(Te,L,J,ie){if(!("language_codes"in Te)||!Array.isArray(Te.language_codes))throw new Error("Tokenizer must have `language_codes` attribute set and it should be an array of language ids.");if(!("languageRegex"in Te)||!(Te.languageRegex instanceof RegExp))throw new Error("Tokenizer must have `languageRegex` attribute set and it should be a regular expression.");if(!("lang_to_token"in Te)||typeof Te.lang_to_token!="function")throw new Error("Tokenizer must have `lang_to_token` attribute set and it should be a function.");const ve=ie.src_lang,be=ie.tgt_lang;if(!Te.language_codes.includes(be))throw new Error(`Target language code "${be}" is not valid. Must be one of: {${Te.language_codes.join(", ")}}`);if(ve!==void 0){if(!Te.language_codes.includes(ve))throw new Error(`Source language code "${ve}" is not valid. Must be one of: {${Te.language_codes.join(", ")}}`);for(const ze of Te.post_processor.config.single)if("SpecialToken"in ze&&Te.languageRegex.test(ze.SpecialToken.id)){ze.SpecialToken.id=Te.lang_to_token(ve);break}}return ie.forced_bos_token_id=Te.model.convert_tokens_to_ids([Te.lang_to_token(be)])[0],Te._call(L,J)}class Ps extends Le{constructor(L,J){super(L,J),this.languageRegex=/^[a-z]{3}_[A-Z][a-z]{3}$/,this.language_codes=this.special_tokens.filter(ie=>this.languageRegex.test(ie)),this.lang_to_token=ie=>ie}_build_translation_inputs(L,J,ie){return Xr(this,L,J,ie)}}class Ss extends Le{constructor(L,J){super(L,J),this.languageRegex=/^__[a-z]{2,3}__$/,this.language_codes=this.special_tokens.filter(ie=>this.languageRegex.test(ie)).map(ie=>ie.slice(2,-2)),this.lang_to_token=ie=>`__${ie}__`}_build_translation_inputs(L,J,ie){return Xr(this,L,J,ie)}}class ks extends Le{get timestamp_begin(){return this.model.convert_tokens_to_ids(["<|notimestamps|>"])[0]+1}_decode_asr(L,{return_timestamps:J=!1,return_language:ie=!1,time_precision:ve=null,force_full_sequences:be=!0}={}){if(ve===null)throw Error("Must specify time_precision");let ze=null;const tt=J==="word";function ct(){return{language:ze,timestamp:[null,null],text:""}}const it=[];let dt=ct(),At=0;const qt=this.timestamp_begin,fr=qt+1500;let sr=[],Pr=[],$r=!1,qr=null;const ws=new Set(this.all_special_ids);for(const ur of L){const Ir=ur.tokens,zr=tt?ur.token_timestamps:null;let Yr=null,ss=qt;if("stride"in ur){const[Ar,yr,Mr]=ur.stride;if(At-=yr,qr=Ar-Mr,yr&&(ss=yr/ve+qt),Mr)for(let vr=Ir.length-1;vr>=0;--vr){const Rr=Number(Ir[vr]);if(Rr>=qt){if(Yr!==null&&(Rr-qt)*ve=qt&&yr<=fr){const Mr=(yr-qt)*ve+At,vr=(0,u.round)(Mr,2);if(Yr!==null&&yr>=Yr)$r=!0;else if($r||sr.length>0&&yr0?(sr.push(Br),tt&&Pr.push(Dr)):sr.every(Ar=>Ar.length===0)&&(dt=ct(),sr=[],Br=[],Pr=[],Dr=[])}if(sr.length>0){if(be&&J)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[ur,Ir]=this.findLongestCommonSequence(sr,Pr),zr=this.decode(ur);dt.text=zr,tt&&(dt.words=this.collateWordTimestamps(ur,Ir,ze)),it.push(dt)}let Ur=Object.create(null);const Qs=it.map(ur=>ur.text).join("");if(J||ie){for(let ur=0;ur0;let tt=ze?[]:null,ct=ze?J[0]:null;for(let it=1;ityr===ss[Mr]&&ct[Qs+Mr]<=J[it][zr+Mr]).length:Br=Ir.filter((yr,Mr)=>yr===ss[Mr]).length;const Dr=Ur/1e4,Ar=Br/Ur+Dr;Br>1&&Ar>At&&(At=Ar,qt=[Qs,ur,zr,Yr])}const[fr,sr,Pr,$r]=qt,qr=Math.floor((sr+fr)/2),ws=Math.floor(($r+Pr)/2);be.push(...ie.slice(0,qr)),ie=dt.slice(ws),ve=ie.length,ze&&(tt.push(...ct.slice(0,qr)),ct=J[it].slice(ws))}return be.push(...ie),ze?(tt.push(...ct),[be,tt]):[be,[]]}collateWordTimestamps(L,J,ie){const[ve,be,ze]=this.combineTokensIntoWords(L,ie),tt=[];for(let ct=0;ct=ve){const tt=((ze-ve)*ie).toFixed(2);be.push(`<|${tt}|>`),be.push([])}else be[be.length-1].push(ze);return be=be.map(ze=>typeof ze=="string"?ze:super.decode(ze,J)),be.join("")}splitTokensOnUnicode(L){const J=this.decode(L,{decode_with_timestamps:!0}),ie="�",ve=[],be=[],ze=[];let tt=[],ct=[],it=0;for(let dt=0;dt=this.model.tokens_to_ids.get("<|endoftext|>"),fr=dt.startsWith(" "),sr=dt.trim(),Pr=ct.test(sr);if(Ht||fr||Pr||be.length===0)be.push(dt),ze.push(At),tt.push(qt);else{const $r=be.length-1;be[$r]+=dt,ze[$r].push(...At),tt[$r].push(...qt)}}return[be,ze,tt]}mergePunctuations(L,J,ie,ve,be){const ze=structuredClone(L),tt=structuredClone(J),ct=structuredClone(ie);let it=ze.length-2,dt=ze.length-1;for(;it>=0;)ze[it].startsWith(" ")&&ve.includes(ze[it].trim())?(ze[dt]=ze[it]+ze[dt],tt[dt]=(0,o.mergeArrays)(tt[it],tt[dt]),ct[dt]=(0,o.mergeArrays)(ct[it],ct[dt]),ze[it]="",tt[it]=[],ct[it]=[]):dt=it,--it;for(it=0,dt=1;dtAt),tt.filter(At=>At.length>0),ct.filter(At=>At.length>0)]}}class $s extends Le{}class gs extends Le{}class _n extends Le{}class gn extends Le{constructor(L,J){super(L,J),this.languageRegex=/^(>>\w+<<)\s*/g,this.supported_language_codes=this.model.vocab.filter(ie=>this.languageRegex.test(ie)),console.warn('WARNING: `MarianTokenizer` is not yet supported by Hugging Face\'s "fast" tokenizers library. Therefore, you may experience slightly inaccurate results.')}_encode_text(L){if(L===null)return null;const[J,...ie]=L.trim().split(this.languageRegex);if(ie.length===0)return super._encode_text(J);if(ie.length===2){const[ve,be]=ie;return this.supported_language_codes.includes(ve)||console.warn(`Unsupported language code "${ve}" detected, which may lead to unexpected behavior. Should be one of: ${JSON.stringify(this.supported_language_codes)}`),(0,o.mergeArrays)([ve],super._encode_text(be))}}}class Fn extends Le{}class Dn extends Le{}class Hr extends Le{}class ys extends Le{}class rn extends Le{}class cs extends Le{constructor(L,J){super(L,J),this.decoder=new ce({})}}class Jr extends Le{}class yn extends Le{}class qs{static async from_pretrained(L,{progress_callback:J=null,config:ie=null,cache_dir:ve=null,local_files_only:be=!1,revision:ze="main",legacy:tt=null}={}){var qt;const[ct,it]=await m(L,{progress_callback:J,config:ie,cache_dir:ve,local_files_only:be,revision:ze,legacy:tt}),dt=((qt=it.tokenizer_class)==null?void 0:qt.replace(/Fast$/,""))??"PreTrainedTokenizer";let At=this.TOKENIZER_CLASS_MAPPING[dt];return At||(console.warn(`Unknown tokenizer class "${dt}", attempting to construct from base class.`),At=Le),new At(ct,it)}}de(qs,"TOKENIZER_CLASS_MAPPING",{T5Tokenizer:en,DistilBertTokenizer:Kr,CamembertTokenizer:xs,DebertaTokenizer:mt,DebertaV2Tokenizer:tr,BertTokenizer:Ft,HerbertTokenizer:rr,ConvBertTokenizer:Fr,RoFormerTokenizer:pr,XLMTokenizer:Ws,ElectraTokenizer:Ts,MobileBertTokenizer:wt,SqueezeBertTokenizer:Nt,AlbertTokenizer:Qt,GPT2Tokenizer:rs,BartTokenizer:tn,MBartTokenizer:os,MBart50Tokenizer:Us,RobertaTokenizer:Gs,WhisperTokenizer:ks,CodeGenTokenizer:$s,CLIPTokenizer:gs,SiglipTokenizer:_n,MarianTokenizer:gn,BloomTokenizer:Ks,NllbTokenizer:Ps,M2M100Tokenizer:Ss,LlamaTokenizer:Jt,CodeLlamaTokenizer:Cs,XLMRobertaTokenizer:Hs,MPNetTokenizer:Er,FalconTokenizer:qe,GPTNeoXTokenizer:ut,EsmTokenizer:bt,Wav2Vec2CTCTokenizer:Fn,BlenderbotTokenizer:Dn,BlenderbotSmallTokenizer:Hr,SpeechT5Tokenizer:ys,NougatTokenizer:rn,VitsTokenizer:cs,Qwen2Tokenizer:cr,GemmaTokenizer:ls,Grok1Tokenizer:Es,CohereTokenizer:Jr,MgpstrTokenizer:yn,PreTrainedTokenizer:Le})},"./src/utils/audio.js":(e,r,t)=>{t.r(r),t.d(r,{RawAudio:()=>se,hamming:()=>g,hanning:()=>m,mel_filter_bank:()=>D,read_audio:()=>f,spectrogram:()=>O,window_function:()=>R});var i=t("./src/utils/hub.js"),o=t("./src/utils/maths.js"),n=t("./src/utils/core.js"),u=t("./src/env.js"),d=t("?7a2c"),h=t("./src/utils/tensor.js");async function f(K,pe){if(typeof AudioContext>"u")throw Error("Unable to load audio from path/URL since `AudioContext` is not available in your environment. Instead, audio data should be passed directly to the pipeline/processor. For more information and some example code, see https://huggingface.co/docs/transformers.js/guides/node-audio-processing.");const re=await(await(0,i.getFile)(K)).arrayBuffer(),oe=new AudioContext({sampleRate:pe});typeof pe>"u"&&console.warn(`No sampling rate provided, using default of ${oe.sampleRate}Hz.`);const ge=await oe.decodeAudioData(re);let le;if(ge.numberOfChannels===2){const Se=Math.sqrt(2),Ce=ge.getChannelData(0),q=ge.getChannelData(1);le=new Float32Array(Ce.length);for(let N=0;N2595*Math.log10(1+K/700),kaldi:K=>1127*Math.log(1+K/700),slaney:(K,pe=1e3,re=15,oe=27/Math.log(6.4))=>K>=pe?re+Math.log(K/pe)*oe:3*K/200};function M(K,pe="htk"){const re=T[pe];if(!re)throw new Error('mel_scale should be one of "htk", "slaney" or "kaldi".');return typeof K=="number"?re(K):K.map(oe=>re(oe))}const I={htk:K=>700*(10**(K/2595)-1),kaldi:K=>700*(Math.exp(K/1127)-1),slaney:(K,pe=1e3,re=15,oe=Math.log(6.4)/27)=>K>=re?pe*Math.exp(oe*(K-re)):200*K/3};function z(K,pe="htk"){const re=I[pe];if(!re)throw new Error('mel_scale should be one of "htk", "slaney" or "kaldi".');return typeof K=="number"?re(K):K.map(oe=>re(oe))}function E(K,pe){const re=Float64Array.from({length:pe.length-1},(Se,Ce)=>pe[Ce+1]-pe[Ce]),oe=Array.from({length:K.length},()=>new Array(pe.length));for(let Se=0;Senew Array(K.length));for(let Se=0;SeK+oe*le)}function D(K,pe,re,oe,ge,le=null,Se="htk",Ce=!1){if(le!==null&&le!=="slaney")throw new Error('norm must be one of null or "slaney"');if(K<2)throw new Error(`Require num_frequency_bins: ${K} >= 2`);if(re>oe)throw new Error(`Require min_frequency: ${re} <= max_frequency: ${oe}`);const q=M(re,Se),N=M(oe,Se),Q=C(q,N,pe+2);let ue=z(Q,Se),Ae;if(Ce){const ae=ge/((K-1)*2);Ae=M(Float64Array.from({length:K},(Ze,et)=>et*ae),Se),ue=Q}else Ae=C(0,Math.floor(ge/2),K);const we=E(Ae,ue);if(le!==null&&le==="slaney")for(let ae=0;aege)throw Error(`frame_length (${re}) may not be larger than fft_length (${ge})`);if(Oe!==re)throw new Error(`Length of the window (${Oe}) must equal frame_length (${re})`);if(oe<=0)throw new Error("hop_length must be greater than zero");if(le===null&&Q!==null)throw new Error("You have provided `mel_filters` but `power` is `None`. Mel spectrogram computation is not yet supported for complex-valued spectrogram. Specify `power` to fix this issue.");if(Se){if(Ce!=="reflect")throw new Error(`pad_mode="${Ce}" not implemented yet.`);const yt=Math.floor((ge-1)/2)+1;K=A(K,yt,yt)}let Fe=Math.floor(1+Math.floor((K.length-re)/oe));Ee!==null&&FeFe?me&&(Ue=Z):Ue=Re=Z);const Ge=new o.FFT(ge),Je=new Float64Array(ge),at=new Float64Array(Ge.outputBufferSize),ot=new Float32Array(Ve*Ue);for(let yt=0;yt=1;--He)Je[He]-=N*Je[He-1];Je[0]*=1-N}for(let He=0;HeMath.pow(Ce,.85));break;default:throw new Error(`Unknown window type ${pe}.`)}if(re&&(Se=Se.subarray(0,K)),oe===null)return Se;if(K>oe)throw new Error(`Length of the window (${K}) may not be larger than frame_length (${oe})`);return Se}function U(K,pe){let re=44;const oe=new ArrayBuffer(re+K.length*4),ge=new DataView(oe);te(ge,0,"RIFF"),ge.setUint32(4,36+K.length*4,!0),te(ge,8,"WAVE"),te(ge,12,"fmt "),ge.setUint32(16,16,!0),ge.setUint16(20,3,!0),ge.setUint16(22,1,!0),ge.setUint32(24,pe,!0),ge.setUint32(28,pe*4,!0),ge.setUint16(32,4,!0),ge.setUint16(34,32,!0),te(ge,36,"data"),ge.setUint32(40,K.length*4,!0);for(let le=0;le{let le=await ge.arrayBuffer();d.writeFileSync(oe,Buffer.from(le))};else throw new Error("Unable to save because filesystem is disabled in this environment.");await re(pe,this.toBlob())}}},"./src/utils/constants.js":(e,r,t)=>{t.r(r),t.d(r,{CHAT_TEMPLATE_NAME:()=>h,CONFIG_NAME:()=>o,FEATURE_EXTRACTOR_NAME:()=>n,GENERATION_CONFIG_NAME:()=>f,GITHUB_ISSUE_URL:()=>i,IMAGE_PROCESSOR_NAME:()=>u,PROCESSOR_NAME:()=>d});const i="https://github.com/huggingface/transformers.js/issues/new/choose",o="config.json",n="preprocessor_config.json",u=n,d="processor_config.json",h="chat_template.json",f="generation_config.json"},"./src/utils/core.js":(e,r,t)=>{t.r(r),t.d(r,{calculateDimensions:()=>f,calculateReflectOffset:()=>T,count:()=>E,dispatchCallback:()=>i,escapeRegExp:()=>n,isIntegralNumber:()=>d,isNullishDimension:()=>h,isTypedArray:()=>u,len:()=>z,mergeArrays:()=>m,pick:()=>I,pop:()=>y,product:()=>g,reverseDictionary:()=>o,saveBlob:()=>M});function i(C,D){C&&C(D)}function o(C){return Object.fromEntries(Object.entries(C).map(([D,A])=>[A,D]))}function n(C){return C.replace(/[.*+?^${}()|[\]\\]/g,"\\$&")}function u(C){var D,A,$;return(($=(A=(D=C==null?void 0:C.prototype)==null?void 0:D.__proto__)==null?void 0:A.constructor)==null?void 0:$.name)==="TypedArray"}function d(C){return Number.isInteger(C)||typeof C=="bigint"}function h(C){return C==null||C===-1}function f(C){const D=[];let A=C;for(;Array.isArray(A);)D.push(A.length),A=A[0];return D}function y(C,D,A=void 0){const $=C[D];if($!==void 0)return delete C[D],$;if(A===void 0)throw Error(`Key ${D} does not exist in object.`);return A}function m(...C){return Array.prototype.concat.apply([],C)}function g(...C){return C.reduce((D,A)=>D.flatMap($=>A.map(P=>[$,P])))}function T(C,D){return Math.abs((C+D)%(2*D)-D)}function M(C,D){const A=URL.createObjectURL(D),$=document.createElement("a");$.href=A,$.download=C,$.click(),$.remove(),URL.revokeObjectURL(A)}function I(C,D){return Object.assign({},...D.map(A=>{if(C[A]!==void 0)return{[A]:C[A]}}))}function z(C){let D=0;for(const A of C)++D;return D}function E(C,D){let A=0;for(const $ of C)$===D&&++A;return A}},"./src/utils/data-structures.js":(e,r,t)=>{t.r(r),t.d(r,{CharTrie:()=>o,DictionarySplitter:()=>h,LRUCache:()=>f,PriorityQueue:()=>i,TokenLattice:()=>u});class i{constructor(m=(T,M)=>T>M,g=1/0){this._heap=[],this._comparator=m,this._maxSize=g}get size(){return this._heap.length}isEmpty(){return this.size===0}peek(){return this._heap[0]}push(...m){return this.extend(m)}extend(m){for(const g of m)if(this.size0&&this._swap(0,g),this._heap.pop(),this._siftDown(),m}replace(m){const g=this.peek();return this._heap[0]=m,this._siftDown(),g}_parent(m){return(m+1>>>1)-1}_left(m){return(m<<1)+1}_right(m){return m+1<<1}_greater(m,g){return this._comparator(this._heap[m],this._heap[g])}_swap(m,g){const T=this._heap[m];this._heap[m]=this._heap[g],this._heap[g]=T}_siftUp(){this._siftUpFrom(this.size-1)}_siftUpFrom(m){for(;m>0&&this._greater(m,this._parent(m));)this._swap(m,this._parent(m)),m=this._parent(m)}_siftDown(){let m=0;for(;this._left(m)[]),this.endNodes=Array.from({length:this.len+1},()=>[]);const M=new d(this.bosTokenId,0,0,0,0),I=new d(this.eosTokenId,1,this.len,0,0);this.nodes.push(M.clone()),this.nodes.push(I.clone()),this.beginNodes[this.len].push(I),this.endNodes[0].push(M)}insert(m,g,T,M){const I=this.nodes.length,z=new d(M,I,m,g,T);this.beginNodes[m].push(z),this.endNodes[m+g].push(z),this.nodes.push(z)}viterbi(){const m=this.len;let g=0;for(;g<=m;){if(this.beginNodes[g].length==0)return[];for(let E of this.beginNodes[g]){E.prev=null;let C=0,D=null;for(let A of this.endNodes[g]){const $=A.backtraceScore+E.score;(D===null||$>C)&&(D=A.clone(),C=$)}if(D!==null)E.prev=D,E.backtraceScore=C;else return[]}++g}const T=[],I=this.beginNodes[m][0].prev;if(I===null)return[];let z=I.clone();for(;z.prev!==null;)T.push(z.clone()),z=z.clone().prev.clone();return T.reverse(),T}piece(m){return this.chars.slice(m.pos,m.pos+m.length).join("")}tokens(){return this.viterbi().map(g=>this.piece(g))}tokenIds(){return this.viterbi().map(g=>g.tokenId)}}class d{constructor(m,g,T,M,I){this.tokenId=m,this.nodeId=g,this.pos=T,this.length=M,this.score=I,this.prev=null,this.backtraceScore=0}clone(){const m=new d(this.tokenId,this.nodeId,this.pos,this.length,this.score);return m.prev=this.prev,m.backtraceScore=this.backtraceScore,m}}class h{constructor(m){this.trie=this._buildTrie(m)}_buildTrie(m){var T;const g=Object.create(null);for(const M of m){let I=g;for(let z=0;zM&&g.push(m.slice(M,I)),g.push(E),I+=E.length,M=I):++I}return Mthis.capacity&&this.cache.delete(this.cache.keys().next().value)}clear(){this.cache.clear()}}},"./src/utils/devices.js":(e,r,t)=>{t.r(r),t.d(r,{DEVICE_TYPES:()=>i});const i=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,r,t)=>{t.r(r),t.d(r,{DATA_TYPES:()=>u,DEFAULT_DEVICE_DTYPE_MAPPING:()=>d,DEFAULT_DTYPE_SUFFIX_MAPPING:()=>h,isWebGpuFp16Supported:()=>n});var i=t("./src/env.js"),o=t("./src/utils/devices.js");const n=function(){let f;return async function(){if(f===void 0)if(!i.apis.IS_WEBGPU_AVAILABLE)f=!1;else try{f=(await navigator.gpu.requestAdapter()).features.has("shader-f16")}catch{f=!1}return f}}(),u=Object.freeze({auto:"auto",fp32:"fp32",fp16:"fp16",q8:"q8",int8:"int8",uint8:"uint8",q4:"q4",bnb4:"bnb4",q4f16:"q4f16"}),d=Object.freeze({[o.DEVICE_TYPES.wasm]:u.q8}),h=Object.freeze({[u.fp32]:"",[u.fp16]:"_fp16",[u.int8]:"_int8",[u.uint8]:"_uint8",[u.q8]:"_quantized",[u.q4]:"_q4",[u.q4f16]:"_q4f16",[u.bnb4]:"_bnb4"})},"./src/utils/generic.js":(e,r,t)=>{t.r(r),t.d(r,{Callable:()=>i});const i=class{constructor(){let o=function(...n){return o._call(...n)};return Object.setPrototypeOf(o,new.target.prototype)}_call(...o){throw Error("Must implement _call method in subclass")}}},"./src/utils/hub.js":(e,r,t)=>{t.r(r),t.d(r,{MAX_EXTERNAL_DATA_CHUNKS:()=>d,getFile:()=>T,getModelFile:()=>C,getModelJSON:()=>D});var i=t("?7a2c"),o=t("?a42a"),n=t("./src/env.js"),u=t("./src/utils/core.js");const d=100,h={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 f{constructor(k){if(this.filePath=k,this.headers=new Headers,this.exists=i.existsSync(k),this.exists){this.status=200,this.statusText="OK";let O=i.statSync(k);this.headers.set("content-length",O.size.toString()),this.updateContentType();const R=i.createReadStream(k);this.body=new ReadableStream({start(U){R.on("data",te=>U.enqueue(te)),R.on("end",()=>U.close()),R.on("error",te=>U.error(te))},cancel(){R.destroy()}})}else this.status=404,this.statusText="Not Found",this.body=null}updateContentType(){const k=this.filePath.toString().split(".").pop().toLowerCase();this.headers.set("content-type",h[k]??"application/octet-stream")}clone(){let k=new f(this.filePath);return k.exists=this.exists,k.status=this.status,k.statusText=this.statusText,k.headers=new Headers(this.headers),k}async arrayBuffer(){return(await i.promises.readFile(this.filePath)).buffer}async blob(){const k=await i.promises.readFile(this.filePath);return new Blob([k],{type:this.headers.get("content-type")})}async text(){return await i.promises.readFile(this.filePath,"utf8")}async json(){return JSON.parse(await this.text())}}function y(P,k=null,O=null){let R;try{R=new URL(P)}catch{return!1}return!(k&&!k.includes(R.protocol)||O&&!O.includes(R.hostname))}const m=/^(\b[\w\-.]+\b\/)?\b[\w\-.]{1,96}\b$/;function g(P){return!(!m.test(P)||P.includes("..")||P.includes("--")||P.endsWith(".git")||P.endsWith(".ipynb"))}async function T(P){var k;if(n.env.useFS&&!y(P,["http:","https:","blob:"]))return new f(P instanceof URL?P.protocol==="file:"?P.pathname:P.toString():P);if(typeof process<"u"&&((k=process==null?void 0:process.release)==null?void 0:k.name)==="node"){const O=!!(hn!=null&&hn.TESTING_REMOTELY),R=n.env.version,U=new Headers;if(U.set("User-Agent",`transformers.js/${R}; is_ci/${O};`),y(P,["http:","https:"],["huggingface.co","hf.co"])){const se=(hn==null?void 0:hn.HF_TOKEN)??(hn==null?void 0:hn.HF_ACCESS_TOKEN);se&&U.set("Authorization",`Bearer ${se}`)}return fetch(P,{headers:U})}else return fetch(P)}const M={400:"Bad request error occurred while trying to load file",401:"Unauthorized access to file",403:"Forbidden access to file",404:"Could not locate file",408:"Request timeout error occurred while trying to load file",500:"Internal server error error occurred while trying to load file",502:"Bad gateway error occurred while trying to load file",503:"Service unavailable error occurred while trying to load file",504:"Gateway timeout error occurred while trying to load file"};function I(P,k,O){if(!O)return null;const R=M[P]??`Error (${P}) occurred while trying to load file`;throw Error(`${R}: "${k}".`)}class z{constructor(k){this.path=k}async match(k){let O=o.join(this.path,k),R=new f(O);if(R.exists)return R}async put(k,O,R=void 0){let U=o.join(this.path,k);try{const te=O.headers.get("Content-Length"),se=parseInt(te??"0");let K=0;await i.promises.mkdir(o.dirname(U),{recursive:!0});const pe=i.createWriteStream(U),re=O.body.getReader();for(;;){const{done:oe,value:ge}=await re.read();if(oe)break;await new Promise((Se,Ce)=>{pe.write(ge,q=>{if(q){Ce(q);return}Se()})}),K+=ge.length;const le=se?K/se*100:0;R==null||R({progress:le,loaded:K,total:se})}pe.close()}catch(te){try{await i.promises.unlink(U)}catch{}throw te}}}async function E(P,...k){for(let O of k)try{let R=await P.match(O);if(R)return R}catch{continue}}async function C(P,k,O=!0,R={},U=!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`.")}(0,u.dispatchCallback)(R.progress_callback,{status:"initiate",name:P,file:k});let te;if(!te&&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");te=n.env.customCache}if(!te&&n.env.useBrowserCache){if(typeof caches>"u")throw Error("Browser cache is not available in this environment.");try{te=await caches.open("transformers-cache")}catch(ue){console.warn("An error occurred while opening the browser cache:",ue)}}if(!te&&n.env.useFSCache){if(!n.apis.IS_FS_AVAILABLE)throw Error("File System Cache is not available in this environment.");te=new z(R.cache_dir??n.env.cacheDir)}const se=R.revision??"main",K=$(P,k),pe=g(P),re=pe?$(n.env.localModelPath,K):K,oe=$(n.env.remoteHost,n.env.remotePathTemplate.replaceAll("{model}",P).replaceAll("{revision}",encodeURIComponent(se)),k);let ge;const le=te instanceof z?se==="main"?K:$(P,se,k):oe;let Se=!1,Ce;te&&(Ce=await E(te,re,le));const q=Ce!==void 0;if(Ce===void 0){if(n.env.allowLocalModels)if(y(K,["http:","https:"])){if(R.local_files_only)throw new Error(`\`local_files_only=true\`, but attempted to load a remote file from: ${K}.`);if(!n.env.allowRemoteModels)throw new Error(`\`env.allowRemoteModels=false\`, but attempted to load a remote file from: ${K}.`)}else try{Ce=await T(re),ge=re}catch(Ae){console.warn(`Unable to load from local path "${re}": "${Ae}"`)}if(Ce===void 0||Ce.status===404){if(R.local_files_only||!n.env.allowRemoteModels){if(O)throw Error(`\`local_files_only=true\` or \`env.allowRemoteModels=false\` and file was not found locally at "${re}".`);return null}if(!pe)throw Error(`Local file missing at "${re}" and download aborted due to invalid model ID "${P}".`);if(Ce=await T(oe),Ce.status!==200)return I(Ce.status,oe,O);ge=le}Se=te&&typeof Response<"u"&&Ce instanceof Response&&Ce.status===200}(0,u.dispatchCallback)(R.progress_callback,{status:"download",name:P,file:k});let N;if(!(n.apis.IS_NODE_ENV&&U)){let ue;R.progress_callback?q&&typeof navigator<"u"&&/firefox/i.test(navigator.userAgent)?(ue=new Uint8Array(await Ce.arrayBuffer()),(0,u.dispatchCallback)(R.progress_callback,{status:"progress",name:P,file:k,progress:100,loaded:ue.length,total:ue.length})):ue=await A(Ce,Ae=>{(0,u.dispatchCallback)(R.progress_callback,{status:"progress",name:P,file:k,...Ae})}):ue=new Uint8Array(await Ce.arrayBuffer()),N=ue}if(Se&&ge&&await te.match(ge)===void 0&&(N?await te.put(ge,new Response(N,{headers:Ce.headers})).catch(ue=>{console.warn(`Unable to add response to browser cache: ${ue}.`)}):await te.put(ge,Ce,R.progress_callback)),(0,u.dispatchCallback)(R.progress_callback,{status:"done",name:P,file:k}),N){if(!n.apis.IS_NODE_ENV&&U)throw new Error("Cannot return path in a browser environment.");return N}if(Ce instanceof f)return Ce.filePath;const Q=await(te==null?void 0:te.match(ge));if(Q instanceof f)return Q.filePath;if(Q instanceof Response)return new Uint8Array(await Q.arrayBuffer());if(typeof Q=="string")return Q;throw new Error("Unable to get model file path or buffer.")}async function D(P,k,O=!0,R={}){const U=await C(P,k,O,R,!1);if(U===null)return{};const se=new TextDecoder("utf-8").decode(U);return JSON.parse(se)}async function A(P,k){const O=P.headers.get("Content-Length");O===null&&console.warn("Unable to determine content-length from response headers. Will expand buffer when needed.");let R=parseInt(O??"0"),U=new Uint8Array(R),te=0;const se=P.body.getReader();async function K(){const{done:pe,value:re}=await se.read();if(pe)return;const oe=te+re.length;if(oe>R){R=oe;const le=new Uint8Array(R);le.set(U),U=le}U.set(re,te),te=oe;const ge=te/R*100;return k({progress:ge,loaded:te,total:R}),K()}return await K(),U}function $(...P){return P=P.map((k,O)=>(O&&(k=k.replace(new RegExp("^/"),"")),O!==P.length-1&&(k=k.replace(new RegExp("/$"),"")),k)),P.join("/")}},"./src/utils/image.js":(e,r,t)=>{t.r(r),t.d(r,{RawImage:()=>M,load_image:()=>I});var i=t("./src/utils/core.js"),o=t("./src/utils/hub.js"),n=t("./src/env.js"),u=t("./src/utils/tensor.js"),d=t("?2b25");let h,f,y;const m=n.apis.IS_BROWSER_ENV||n.apis.IS_WEBWORKER_ENV;if(m)h=(z,E)=>{if(!self.OffscreenCanvas)throw new Error("OffscreenCanvas not supported by this browser.");return new self.OffscreenCanvas(z,E)},y=self.createImageBitmap,f=self.ImageData;else if(d)y=async z=>{const C=(await z.metadata()).channels,{data:D,info:A}=await z.rotate().raw().toBuffer({resolveWithObject:!0}),$=new M(new Uint8ClampedArray(D),A.width,A.height,A.channels);return C!==void 0&&C!==A.channels&&$.convert(C),$};else throw new Error("Unable to load image processing library.");const g={0:"nearest",1:"lanczos",2:"bilinear",3:"bicubic",4:"box",5:"hamming"},T=new Map([["png","image/png"],["jpg","image/jpeg"],["jpeg","image/jpeg"],["gif","image/gif"]]);class M{constructor(E,C,D,A){this.data=E,this.width=C,this.height=D,this.channels=A}get size(){return[this.width,this.height]}static async read(E){if(E instanceof M)return E;if(typeof E=="string"||E instanceof URL)return await this.fromURL(E);if(E instanceof Blob)return await this.fromBlob(E);if(typeof HTMLCanvasElement<"u"&&E instanceof HTMLCanvasElement||typeof OffscreenCanvas<"u"&&E instanceof OffscreenCanvas)return this.fromCanvas(E);throw new Error(`Unsupported input type: ${typeof E}`)}static fromCanvas(E){if(!m)throw new Error("fromCanvas() is only supported in browser environments.");const D=E.getContext("2d").getImageData(0,0,E.width,E.height).data;return new M(D,E.width,E.height,4)}static async fromURL(E){const C=await(0,o.getFile)(E);if(C.status!==200)throw new Error(`Unable to read image from "${E}" (${C.status} ${C.statusText})`);const D=await C.blob();return this.fromBlob(D)}static async fromBlob(E){if(m){const C=await y(E),D=h(C.width,C.height).getContext("2d");return D.drawImage(C,0,0),new this(D.getImageData(0,0,C.width,C.height).data,C.width,C.height,4)}else{const C=d(await E.arrayBuffer());return await y(C)}}static fromTensor(E,C="CHW"){if(E.dims.length!==3)throw new Error(`Tensor should have 3 dimensions, but has ${E.dims.length} dimensions.`);if(C==="CHW")E=E.transpose(1,2,0);else if(C!=="HWC")throw new Error(`Unsupported channel format: ${C}`);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: ${E.dims[2]}`)}}grayscale(){if(this.channels===1)return this;const E=new Uint8ClampedArray(this.width*this.height*1);switch(this.channels){case 3:case 4:for(let C=0,D=0;C=0?O=D:U=-D,A>=0?R=A:te=-A,k.drawImage(P,O,R,E,C,U,te,E,C),new M(k.getImageData(0,0,E,C).data,E,C,4).convert($)}else{let $=this.toSharp();if(D>=0&&A>=0)$=$.extract({left:Math.floor(D),top:Math.floor(A),width:E,height:C});else if(D<=0&&A<=0){const P=Math.floor(-A),k=Math.floor(-D);$=$.extend({top:P,left:k,right:E-this.width-k,bottom:C-this.height-P})}else{let P=[0,0],k=0;A<0?(P[0]=Math.floor(-A),P[1]=C-this.height-P[0]):k=Math.floor(A);let O=[0,0],R=0;D<0?(O[0]=Math.floor(-D),O[1]=E-this.width-O[0]):R=Math.floor(D),$=$.extend({top:P[0],bottom:P[1],left:O[0],right:O[1]}).extract({left:R,top:k,width:E,height:C})}return await y($)}}async toBlob(E="image/png",C=1){if(!m)throw new Error("toBlob() is only supported in browser environments.");return await this.toCanvas().convertToBlob({type:E,quality:C})}toTensor(E="CHW"){let C=new u.Tensor("uint8",new Uint8Array(this.data),[this.height,this.width,this.channels]);if(E!=="HWC")if(E==="CHW")C=C.permute(2,0,1);else throw new Error(`Unsupported channel format: ${E}`);return C}toCanvas(){if(!m)throw new Error("toCanvas() is only supported in browser environments.");const E=this.clone().rgba(),C=h(E.width,E.height),D=new f(E.data,E.width,E.height);return C.getContext("2d").putImageData(D,0,0),C}split(){const{data:E,width:C,height:D,channels:A}=this,$=E.constructor,P=E.length/A,k=Array.from({length:A},()=>new $(P));for(let O=0;Onew M(O,C,D,1))}_update(E,C,D,A=null){return this.data=E,this.width=C,this.height=D,A!==null&&(this.channels=A),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: ${this.channels}`)}return this}async save(E){if(m){if(n.apis.IS_WEBWORKER_ENV)throw new Error("Unable to save an image from a Web Worker.");const C=E.split(".").pop().toLowerCase(),D=T.get(C)??"image/png",A=await this.toBlob(D);(0,i.saveBlob)(E,A)}else{if(n.apis.IS_FS_AVAILABLE)return await this.toSharp().toFile(E);throw new Error("Unable to save the image because filesystem is disabled in this environment.")}}toSharp(){if(m)throw new Error("toSharp() is only supported in server-side environments.");return d(this.data,{raw:{width:this.width,height:this.height,channels:this.channels}})}}const I=M.read.bind(M)},"./src/utils/maths.js":(e,r,t)=>{t.r(r),t.d(r,{FFT:()=>I,bankers_round:()=>C,cos_sim:()=>h,dot:()=>d,dynamic_time_warping:()=>D,interpolate_data:()=>i,log_softmax:()=>u,magnitude:()=>f,max:()=>m,medianFilter:()=>z,min:()=>y,permute_data:()=>o,round:()=>E,softmax:()=>n});function i(A,[$,P,k],[O,R],U="bilinear",te=!1){const se=R/k,K=O/P,pe=new A.constructor(O*R*$),re=P*k,oe=O*R;for(let ge=0;ge=0;--te)O[te]=se,k[te]=$[P[te]],se*=k[te];const R=P.map((te,se)=>O[P.indexOf(se)]),U=new A.constructor(A.length);for(let te=0;te=0;--K)se+=pe%$[K]*R[K],pe=Math.floor(pe/$[K]);U[se]=A[te]}return[U,k]}function n(A){const $=m(A)[0],P=A.map(R=>Math.exp(R-$)),k=P.reduce((R,U)=>R+U,0);return P.map(R=>R/k)}function u(A){const $=m(A)[0];let P=0;for(let R=0;RR-$-k)}function d(A,$){let P=0;for(let k=0;k$+P*P,0))}function y(A){if(A.length===0)throw Error("Array must not be empty");let $=A[0],P=0;for(let k=1;k$&&($=A[k],P=k);return[$,P]}function g(A){return A>0&&(A&A-1)===0}class T{constructor($){if(this.size=$|0,this.size<=1||!g(this.size))throw new Error("FFT size must be a power of two larger than 1");this._csize=$<<1,this.table=new Float64Array(this.size*2);for(let k=0;kk;k<<=1)++P;this._width=P%2===0?P-1:P,this._bitrev=new Int32Array(1<>>O&3)<>>1);for(let O=0;O<$.length;O+=2)k[O>>>1]=$[O];return k}toComplexArray($,P){const k=P||this.createComplexArray();for(let O=0;O>>1],k[O+1]=0;return k}transform($,P){if($===P)throw new Error("Input and output buffers must be different");this._transform4($,P,1)}realTransform($,P){if($===P)throw new Error("Input and output buffers must be different");this._realTransform4($,P,1)}inverseTransform($,P){if($===P)throw new Error("Input and output buffers must be different");this._transform4($,P,-1);for(let k=0;k<$.length;++k)$[k]/=this.size}_transform4($,P,k){const O=this._csize;let U=1<>=2;U>=2;U>>=2){te=O/U<<1;const oe=te>>>2;for(se=0;se>>1,U>>>1)}else for(se=0,K=0;se>>1,U>>>1,k)}const re=this.table;for(U>>=2;U>=2;U>>=2){te=O/U<<1;const ge=te>>>1,le=ge>>>1,Se=le>>>1;for(se=0;se>>1;for(let ge=2;ge>1;++pe){const re=(pe+1-$)**2/2,oe=Math.sqrt(se**2+K**2)**re,ge=re*Math.atan2(K,se),le=2*pe;R[le]=oe*Math.cos(ge),R[le+1]=oe*Math.sin(ge),U[le]=R[le],U[le+1]=-R[le+1]}this._slicedChirpBuffer=R.subarray(P,k),this._f=new T(O>>1),this._f.transform(this._chirpBuffer,U)}_transform($,P,k){const O=this._buffer1,R=this._buffer2,U=this._outBuffer1,te=this._outBuffer2,se=this._chirpBuffer,K=this._slicedChirpBuffer,pe=this._a;if(k)for(let re=0;re>1,le=P[ge];O[re]=le*K[re],O[oe]=le*K[oe]}else for(let re=0;re=A.length&&(se=2*(A.length-1)-se),k[U++]=A[se]}k.sort(),P[R]=k[O]}return P}function E(A,$){const P=Math.pow(10,$);return Math.round(A*P)/P}function C(A){const $=Math.round(A);return Math.abs(A)%1===.5?$%2===0?$:$-1:$}function D(A){const $=A.length,P=A[0].length,k=[$+1,P+1],O=Array.from({length:k[0]},()=>Array(k[1]).fill(1/0));O[0][0]=0;const R=Array.from({length:k[0]},()=>Array(k[1]).fill(-1));for(let pe=1;pe0||te>0;)switch(se.push(U-1),K.push(te-1),R[U][te]){case 0:--U,--te;break;case 1:--U;break;case 2:--te;break;default:throw new Error(`Internal error in dynamic time warping. Unexpected trace[${U}, ${te}]. Please file a bug report.`)}return se.reverse(),K.reverse(),[se,K]}},"./src/utils/tensor.js":(e,r,t)=>{t.r(r),t.d(r,{DataTypeMap:()=>u,Tensor:()=>d,cat:()=>P,full:()=>K,full_like:()=>pe,interpolate:()=>y,interpolate_4d:()=>m,layer_norm:()=>C,matmul:()=>g,mean:()=>U,mean_pooling:()=>E,ones:()=>re,ones_like:()=>oe,permute:()=>f,quantize_embeddings:()=>Ce,rand:()=>Se,rfft:()=>T,slice:()=>z,stack:()=>k,std_mean:()=>R,topk:()=>M,zeros:()=>ge,zeros_like:()=>le});var i=t("./src/utils/maths.js"),o=t("./src/backends/onnx.js"),n=t("./src/ops/registry.js");const u=Object.freeze({float32:Float32Array,float16:typeof Float16Array<"u"?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 d{constructor(...N){de(this,"ort_tensor");return(0,o.isONNXTensor)(N[0])?this.ort_tensor=N[0]:this.ort_tensor=new o.Tensor(N[0],N[1],N[2]),new Proxy(this,{get:(Q,ue)=>{if(typeof ue=="string"){let Ae=Number(ue);if(Number.isInteger(Ae))return Q._getitem(Ae)}return Q[ue]},set:(Q,ue,Ae)=>Q[ue]=Ae})}get dims(){return this.ort_tensor.dims}set dims(N){this.ort_tensor.dims=N}get type(){return this.ort_tensor.type}get data(){return this.ort_tensor.data}get size(){return this.ort_tensor.size}get location(){return this.ort_tensor.location}dispose(){this.ort_tensor.dispose()}*[Symbol.iterator](){const[N,...Q]=this.dims;if(Q.length>0){const ue=Q.reduce((Ae,we)=>Ae*we);for(let Ae=0;Ae0){const Ae=ue.reduce((we,ae)=>we*ae);return this._subarray(N,Ae,ue)}else return new d(this.type,[this.data[N]],ue)}indexOf(N){const Q=this.data;for(let ue=0;ueX)throw new Error(`Invalid slice: ${Z}`);const Oe=[Math.max(me,0),Math.min(X,this.dims[Ee])];ue.push(Oe),Q.push(Oe[1]-Oe[0])}else throw new Error(`Invalid slice: ${Z}`)}const Ae=ue.map(([Ee,Z])=>Z-Ee),we=Ae.reduce((Ee,Z)=>Ee*Z),ae=this.data,Ze=new ae.constructor(we),et=this.stride();for(let Ee=0;Ee=0;--me){const Oe=Ae[me];Z+=(X%Oe+ue[me][0])*et[me],X=Math.floor(X/Oe)}Ze[Ee]=ae[Z]}return new d(this.type,Ze,Q)}permute(...N){return f(this,N)}transpose(...N){return this.permute(...N)}sum(N=null,Q=!1){return this.norm(1,N,Q)}norm(N="fro",Q=null,ue=!1){if(N==="fro")N=2;else if(typeof N=="string")throw Error(`Unsupported norm: ${N}`);const Ae=this.data,we=(Ee,Z)=>Ee+Z**N;if(Q===null){const Ee=Ae.reduce(we,0)**(1/N);return new d(this.type,[Ee],[])}const[ae,Ze,et]=O(we,this,Q,ue);if(N!==1)for(let Ee=0;Ee=0;--et){const me=this.dims[et];if(et!==Q){const X=Ee%me;Ze+=X*Z,Z*=this.dims[et]}Ee=Math.floor(Ee/me)}Ae[ae]/=we[Ze]}return this}normalize(N=2,Q=1){return this.clone().normalize_(N,Q)}stride(){return te(this.dims)}squeeze(N=null){return new d(this.type,this.data,D(this.dims,N))}squeeze_(N=null){return this.dims=D(this.dims,N),this}unsqueeze(N=null){return new d(this.type,this.data,A(this.dims,N))}unsqueeze_(N=null){return this.dims=A(this.dims,N),this}flatten_(N=0,Q=-1){Q=(Q+this.dims.length)%this.dims.length;let ue=this.dims.slice(0,N),Ae=this.dims.slice(N,Q+1),we=this.dims.slice(Q+1);return this.dims=[...ue,Ae.reduce((ae,Ze)=>ae*Ze,1),...we],this}flatten(N=0,Q=-1){return this.clone().flatten_(N,Q)}view(...N){let Q=-1;for(let Ae=0;AeZe!==Q?we*ae:we,1);N[Q]=ue.length/Ae}return new d(this.type,ue,N)}neg_(){const N=this.data;for(let Q=0;QN?1:0;return new d("bool",Q,this.dims)}lt(N){const Q=new Uint8Array(this.data.length),ue=this.data;for(let Ae=0;AeMath.min(ae,Ze),this,N,Q,1/0);return new d(ue,Ae,we)}max(N=null,Q=!1){if(N===null){const ae=(0,i.max)(this.data)[0];return new d(this.type,[ae],[])}const[ue,Ae,we]=O((ae,Ze)=>Math.max(ae,Ze),this,N,Q,-1/0);return new d(ue,Ae,we)}argmin(N=null,Q=!1){if(N!==null)throw new Error("`dim !== null` not yet implemented.");const ue=(0,i.min)(this.data)[1];return new d("int64",[BigInt(ue)],[])}argmax(N=null,Q=!1){if(N!==null)throw new Error("`dim !== null` not yet implemented.");const ue=(0,i.max)(this.data)[1];return new d("int64",[BigInt(ue)],[])}to(N){if(this.type===N)return this;if(!u.hasOwnProperty(N))throw new Error(`Unsupported type: ${N}`);let Q;const ue=["int64","uint64"].includes(this.type),Ae=["int64","uint64"].includes(N);return ue&&!Ae?Q=Number:!ue&&Ae&&(Q=BigInt),new d(N,u[N].from(this.data,Q),this.dims)}}function h(q,N){const Q=q.length,ue=N.reduce((we,ae)=>we*ae);if(Q!==ue)throw Error(`cannot reshape array of size ${Q} into shape (${N})`);let Ae=q;for(let we=N.length-1;we>=0;we--)Ae=Ae.reduce((ae,Ze)=>{let et=ae[ae.length-1];return et.lengthnew d("int64",q,[q.length]);async function z(q,N,Q,ue,Ae){return await(await n.TensorOpRegistry.slice)({x:q,s:I(N),e:I(Q),a:I(ue),t:I(Ae??new Array(ue.length).fill(1))})}function E(q,N){const Q=q.data,ue=N.data,Ae=[q.dims[0],q.dims[2]],we=new Q.constructor(Ae[0]*Ae[1]),[ae,Ze,et]=q.dims;let Ee=0;for(let Z=0;ZQ!==1):typeof N=="number"?q[N]===1&&q.splice(N,1):Array.isArray(N)&&(q=q.filter((Q,ue)=>Q!==1||!N.includes(ue))),q}function A(q,N){return N=$(N,q.length+1),q=q.slice(),q.splice(N,0,1),q}function $(q,N,Q=null,ue=!0){if(q<-N||q>=N){if(ue)throw new Error(`IndexError: index ${q} is out of bounds for dimension${Q===null?"":" "+Q} with size ${N}`);return q<-N?0:N}return q<0&&(q=(q%N+N)%N),q}function P(q,N=0){N=$(N,q[0].dims.length);const Q=q[0].dims.slice();Q[N]=q.reduce((ae,Ze)=>ae+Ze.dims[N],0);const ue=Q.reduce((ae,Ze)=>ae*Ze,1),Ae=new q[0].data.constructor(ue),we=q[0].type;if(N===0){let ae=0;for(const Ze of q){const et=Ze.data;Ae.set(et,ae),ae+=et.length}}else{let ae=0;for(let Ze=0;Ze=0;--X){const Ve=Ee[X];let Re=Oe%Ve;X===N&&(Re+=ae),me+=Re*Fe,Fe*=Q[X],Oe=Math.floor(Oe/Ve)}Ae[me]=et[Z]}ae+=Ee[N]}}return new d(we,Ae,Q)}function k(q,N=0){return P(q.map(Q=>Q.unsqueeze(N)),N)}function O(q,N,Q=null,ue=!1,Ae=null){const we=N.data,ae=N.dims;Q=$(Q,ae.length);const Ze=ae.slice();Ze[Q]=1;const et=new we.constructor(we.length/ae[Q]);Ae!==null&&et.fill(Ae);for(let Ee=0;Ee=0;--me){const Fe=ae[me];if(me!==Q){const Ve=X%Fe;Z+=Ve*Oe,Oe*=Ze[me]}X=Math.floor(X/Fe)}et[Z]=q(et[Z],we[Ee],Ee,Z)}return ue||Ze.splice(Q,1),[N.type,et,Ze]}function R(q,N=null,Q=1,ue=!1){const Ae=q.data,we=q.dims;if(N===null){const Oe=Ae.reduce((Ue,Ge)=>Ue+Ge,0)/Ae.length,Fe=Math.sqrt(Ae.reduce((Ue,Ge)=>Ue+(Ge-Oe)**2,0)/(Ae.length-Q)),Ve=new d(q.type,[Oe],[]);return[new d(q.type,[Fe],[]),Ve]}N=$(N,we.length);const ae=U(q,N,ue),Ze=ae.data,[et,Ee,Z]=O((X,Oe,Fe,Ve)=>X+(Oe-Ze[Ve])**2,q,N,ue);for(let X=0;XEe+Z,0);return new d(q.type,[et/Ae.length],[])}N=$(N,ue.length);const[we,ae,Ze]=O((et,Ee)=>et+Ee,q,N,Q);if(ue[N]!==1)for(let et=0;et=0;--Q)N[Q]=ue,ue*=q[Q];return N}function se(q,N,Q,ue){const Ae=q.reduce((we,ae)=>we*ae,1);return new d(Q,new ue(Ae).fill(N),q)}function K(q,N){let Q,ue;if(typeof N=="number")Q="float32",ue=Float32Array;else if(typeof N=="bigint")Q="int64",ue=BigInt64Array;else if(typeof N=="boolean")Q="bool",ue=Uint8Array;else throw new Error(`Unsupported data type: ${typeof N}`);return se(q,N,Q,ue)}function pe(q,N){return K(q.dims,N)}function re(q){return se(q,1n,"int64",BigInt64Array)}function oe(q){return re(q.dims)}function ge(q){return se(q,0n,"int64",BigInt64Array)}function le(q){return ge(q.dims)}function Se(q){const N=q.reduce((Q,ue)=>Q*ue,1);return new d("float32",Float32Array.from({length:N},()=>Math.random()),q)}function Ce(q,N){if(q.dims.length!==2)throw new Error("The tensor must have 2 dimensions");if(q.dims.at(-1)%8!==0)throw new Error("The last dimension of the tensor must be a multiple of 8");if(!["binary","ubinary"].includes(N))throw new Error("The precision must be either 'binary' or 'ubinary'");const Q=N==="binary",ue=Q?"int8":"uint8",Ae=Q?Int8Array:Uint8Array,we=q.data,ae=new Ae(we.length/8);for(let Ze=0;Ze0?1:0,Ee=Math.floor(Ze/8),Z=Ze%8;ae[Ee]|=et<<7-Z,Q&&Z===0&&(ae[Ee]-=128)}return new d(ue,ae,[q.dims[0],q.dims[1]/8])}},"./src/utils/video.js":(e,r,t)=>{t.r(r),t.d(r,{RawVideo:()=>u,RawVideoFrame:()=>n,load_video:()=>d});var i=t("./src/utils/image.js"),o=t("./src/env.js");class n{constructor(f,y){this.image=f,this.timestamp=y}}class u{constructor(f,y){f.length>0&&f[0]instanceof i.RawImage&&(f=f.map((m,g)=>new n(m,(g+1)/(f.length+1)*y))),this.frames=f,this.duration=y}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 d(h,{num_frames:f=null,fps:y=null}={}){if(!o.apis.IS_BROWSER_ENV)throw new Error("`load_video` is currently only supported in browser environments.");if(f==null&&y==null)throw new Error("Either num_frames or fps must be provided.");const m=[],g=document.createElement("video");if(g.crossOrigin="anonymous",g.muted=!0,typeof h=="string")g.src=h;else if(h instanceof Blob)g.src=URL.createObjectURL(h);else if(h instanceof HTMLVideoElement)g.src=h.src;else throw new Error("Invalid URL or video element provided.");if(await new Promise(D=>g.onloadedmetadata=D),g.seekable.start(0)===g.seekable.end(0)){const A=await(await fetch(g.src)).blob();g.src=URL.createObjectURL(A),await new Promise($=>g.onloadedmetadata=$)}const T=g.duration;let M,I;f!=null?(M=f,I=f===1?0:T/(f-1)):(I=1/y,M=Math.floor(T/I));let z=[];for(let D=0;D{g.onseeked=k}),C.drawImage(g,0,0,E.width,E.height);const A=C.getImageData(0,0,E.width,E.height),$=new i.RawImage(A.data,E.width,E.height,4),P=new n($,D);m.push(P)}return g.remove(),new u(m,T)}}},t0={};function Yt(e){var r=t0[e];if(r!==void 0)return r.exports;var t=t0[e]={exports:{}};return YT[e](t,t.exports,Yt),t.exports}(()=>{var e=Object.getPrototypeOf?t=>Object.getPrototypeOf(t):t=>t.__proto__,r;Yt.t=function(t,i){if(i&1&&(t=this(t)),i&8||typeof t=="object"&&t&&(i&4&&t.__esModule||i&16&&typeof t.then=="function"))return t;var o=Object.create(null);Yt.r(o);var n={};r=r||[null,e({}),e([]),e(e)];for(var u=i&2&&t;typeof u=="object"&&!~r.indexOf(u);u=e(u))Object.getOwnPropertyNames(u).forEach(d=>n[d]=()=>t[d]);return n.default=()=>t,Yt.d(o,n),o}})();Yt.d=(e,r)=>{for(var t in r)Yt.o(r,t)&&!Yt.o(e,t)&&Object.defineProperty(e,t,{enumerable:!0,get:r[t]})};Yt.o=(e,r)=>Object.prototype.hasOwnProperty.call(e,r);Yt.r=e=>{typeof Symbol<"u"&&Symbol.toStringTag&&Object.defineProperty(e,Symbol.toStringTag,{value:"Module"}),Object.defineProperty(e,"__esModule",{value:!0})};var b={};(()=>{/*!*****************************!*\ + !*** ./src/transformers.js ***! + \*****************************/Yt.r(b),Yt.d(b,{ASTFeatureExtractor:()=>m.ASTFeatureExtractor,ASTForAudioClassification:()=>t.ASTForAudioClassification,ASTModel:()=>t.ASTModel,ASTPreTrainedModel:()=>t.ASTPreTrainedModel,AlbertForMaskedLM:()=>t.AlbertForMaskedLM,AlbertForQuestionAnswering:()=>t.AlbertForQuestionAnswering,AlbertForSequenceClassification:()=>t.AlbertForSequenceClassification,AlbertModel:()=>t.AlbertModel,AlbertPreTrainedModel:()=>t.AlbertPreTrainedModel,AlbertTokenizer:()=>i.AlbertTokenizer,AudioClassificationPipeline:()=>r.AudioClassificationPipeline,AutoConfig:()=>o.AutoConfig,AutoFeatureExtractor:()=>g.AutoFeatureExtractor,AutoImageProcessor:()=>I.AutoImageProcessor,AutoModel:()=>t.AutoModel,AutoModelForAudioClassification:()=>t.AutoModelForAudioClassification,AutoModelForAudioFrameClassification:()=>t.AutoModelForAudioFrameClassification,AutoModelForAudioTextToText:()=>t.AutoModelForAudioTextToText,AutoModelForCTC:()=>t.AutoModelForCTC,AutoModelForCausalLM:()=>t.AutoModelForCausalLM,AutoModelForDepthEstimation:()=>t.AutoModelForDepthEstimation,AutoModelForDocumentQuestionAnswering:()=>t.AutoModelForDocumentQuestionAnswering,AutoModelForImageClassification:()=>t.AutoModelForImageClassification,AutoModelForImageFeatureExtraction:()=>t.AutoModelForImageFeatureExtraction,AutoModelForImageMatting:()=>t.AutoModelForImageMatting,AutoModelForImageSegmentation:()=>t.AutoModelForImageSegmentation,AutoModelForImageTextToText:()=>t.AutoModelForImageTextToText,AutoModelForImageToImage:()=>t.AutoModelForImageToImage,AutoModelForMaskGeneration:()=>t.AutoModelForMaskGeneration,AutoModelForMaskedLM:()=>t.AutoModelForMaskedLM,AutoModelForNormalEstimation:()=>t.AutoModelForNormalEstimation,AutoModelForObjectDetection:()=>t.AutoModelForObjectDetection,AutoModelForPoseEstimation:()=>t.AutoModelForPoseEstimation,AutoModelForQuestionAnswering:()=>t.AutoModelForQuestionAnswering,AutoModelForSemanticSegmentation:()=>t.AutoModelForSemanticSegmentation,AutoModelForSeq2SeqLM:()=>t.AutoModelForSeq2SeqLM,AutoModelForSequenceClassification:()=>t.AutoModelForSequenceClassification,AutoModelForSpeechSeq2Seq:()=>t.AutoModelForSpeechSeq2Seq,AutoModelForTextToSpectrogram:()=>t.AutoModelForTextToSpectrogram,AutoModelForTextToWaveform:()=>t.AutoModelForTextToWaveform,AutoModelForTokenClassification:()=>t.AutoModelForTokenClassification,AutoModelForUniversalSegmentation:()=>t.AutoModelForUniversalSegmentation,AutoModelForVision2Seq:()=>t.AutoModelForVision2Seq,AutoModelForXVector:()=>t.AutoModelForXVector,AutoModelForZeroShotObjectDetection:()=>t.AutoModelForZeroShotObjectDetection,AutoProcessor:()=>C.AutoProcessor,AutoTokenizer:()=>i.AutoTokenizer,AutomaticSpeechRecognitionPipeline:()=>r.AutomaticSpeechRecognitionPipeline,BackgroundRemovalPipeline:()=>r.BackgroundRemovalPipeline,BartForConditionalGeneration:()=>t.BartForConditionalGeneration,BartForSequenceClassification:()=>t.BartForSequenceClassification,BartModel:()=>t.BartModel,BartPretrainedModel:()=>t.BartPretrainedModel,BartTokenizer:()=>i.BartTokenizer,BaseModelOutput:()=>t.BaseModelOutput,BaseStreamer:()=>D.BaseStreamer,BeitFeatureExtractor:()=>M.BeitFeatureExtractor,BeitForImageClassification:()=>t.BeitForImageClassification,BeitModel:()=>t.BeitModel,BeitPreTrainedModel:()=>t.BeitPreTrainedModel,BertForMaskedLM:()=>t.BertForMaskedLM,BertForQuestionAnswering:()=>t.BertForQuestionAnswering,BertForSequenceClassification:()=>t.BertForSequenceClassification,BertForTokenClassification:()=>t.BertForTokenClassification,BertModel:()=>t.BertModel,BertPreTrainedModel:()=>t.BertPreTrainedModel,BertTokenizer:()=>i.BertTokenizer,BitImageProcessor:()=>M.BitImageProcessor,BlenderbotForConditionalGeneration:()=>t.BlenderbotForConditionalGeneration,BlenderbotModel:()=>t.BlenderbotModel,BlenderbotPreTrainedModel:()=>t.BlenderbotPreTrainedModel,BlenderbotSmallForConditionalGeneration:()=>t.BlenderbotSmallForConditionalGeneration,BlenderbotSmallModel:()=>t.BlenderbotSmallModel,BlenderbotSmallPreTrainedModel:()=>t.BlenderbotSmallPreTrainedModel,BlenderbotSmallTokenizer:()=>i.BlenderbotSmallTokenizer,BlenderbotTokenizer:()=>i.BlenderbotTokenizer,BloomForCausalLM:()=>t.BloomForCausalLM,BloomModel:()=>t.BloomModel,BloomPreTrainedModel:()=>t.BloomPreTrainedModel,BloomTokenizer:()=>i.BloomTokenizer,CLIPFeatureExtractor:()=>M.CLIPFeatureExtractor,CLIPImageProcessor:()=>M.CLIPImageProcessor,CLIPModel:()=>t.CLIPModel,CLIPPreTrainedModel:()=>t.CLIPPreTrainedModel,CLIPSegForImageSegmentation:()=>t.CLIPSegForImageSegmentation,CLIPSegModel:()=>t.CLIPSegModel,CLIPSegPreTrainedModel:()=>t.CLIPSegPreTrainedModel,CLIPTextModel:()=>t.CLIPTextModel,CLIPTextModelWithProjection:()=>t.CLIPTextModelWithProjection,CLIPTokenizer:()=>i.CLIPTokenizer,CLIPVisionModel:()=>t.CLIPVisionModel,CLIPVisionModelWithProjection:()=>t.CLIPVisionModelWithProjection,CamembertForMaskedLM:()=>t.CamembertForMaskedLM,CamembertForQuestionAnswering:()=>t.CamembertForQuestionAnswering,CamembertForSequenceClassification:()=>t.CamembertForSequenceClassification,CamembertForTokenClassification:()=>t.CamembertForTokenClassification,CamembertModel:()=>t.CamembertModel,CamembertPreTrainedModel:()=>t.CamembertPreTrainedModel,CamembertTokenizer:()=>i.CamembertTokenizer,CausalLMOutput:()=>t.CausalLMOutput,CausalLMOutputWithPast:()=>t.CausalLMOutputWithPast,ChineseCLIPFeatureExtractor:()=>M.ChineseCLIPFeatureExtractor,ChineseCLIPModel:()=>t.ChineseCLIPModel,ChineseCLIPPreTrainedModel:()=>t.ChineseCLIPPreTrainedModel,ClapAudioModelWithProjection:()=>t.ClapAudioModelWithProjection,ClapFeatureExtractor:()=>m.ClapFeatureExtractor,ClapModel:()=>t.ClapModel,ClapPreTrainedModel:()=>t.ClapPreTrainedModel,ClapTextModelWithProjection:()=>t.ClapTextModelWithProjection,ClassifierFreeGuidanceLogitsProcessor:()=>$.ClassifierFreeGuidanceLogitsProcessor,CodeGenForCausalLM:()=>t.CodeGenForCausalLM,CodeGenModel:()=>t.CodeGenModel,CodeGenPreTrainedModel:()=>t.CodeGenPreTrainedModel,CodeGenTokenizer:()=>i.CodeGenTokenizer,CodeLlamaTokenizer:()=>i.CodeLlamaTokenizer,CohereForCausalLM:()=>t.CohereForCausalLM,CohereModel:()=>t.CohereModel,CoherePreTrainedModel:()=>t.CoherePreTrainedModel,CohereTokenizer:()=>i.CohereTokenizer,ConvBertForMaskedLM:()=>t.ConvBertForMaskedLM,ConvBertForQuestionAnswering:()=>t.ConvBertForQuestionAnswering,ConvBertForSequenceClassification:()=>t.ConvBertForSequenceClassification,ConvBertForTokenClassification:()=>t.ConvBertForTokenClassification,ConvBertModel:()=>t.ConvBertModel,ConvBertPreTrainedModel:()=>t.ConvBertPreTrainedModel,ConvBertTokenizer:()=>i.ConvBertTokenizer,ConvNextFeatureExtractor:()=>M.ConvNextFeatureExtractor,ConvNextForImageClassification:()=>t.ConvNextForImageClassification,ConvNextImageProcessor:()=>M.ConvNextImageProcessor,ConvNextModel:()=>t.ConvNextModel,ConvNextPreTrainedModel:()=>t.ConvNextPreTrainedModel,ConvNextV2ForImageClassification:()=>t.ConvNextV2ForImageClassification,ConvNextV2Model:()=>t.ConvNextV2Model,ConvNextV2PreTrainedModel:()=>t.ConvNextV2PreTrainedModel,DFineForObjectDetection:()=>t.DFineForObjectDetection,DFineModel:()=>t.DFineModel,DFinePreTrainedModel:()=>t.DFinePreTrainedModel,DPTFeatureExtractor:()=>M.DPTFeatureExtractor,DPTForDepthEstimation:()=>t.DPTForDepthEstimation,DPTImageProcessor:()=>M.DPTImageProcessor,DPTModel:()=>t.DPTModel,DPTPreTrainedModel:()=>t.DPTPreTrainedModel,DacDecoderModel:()=>t.DacDecoderModel,DacDecoderOutput:()=>t.DacDecoderOutput,DacEncoderModel:()=>t.DacEncoderModel,DacEncoderOutput:()=>t.DacEncoderOutput,DacFeatureExtractor:()=>m.DacFeatureExtractor,DacModel:()=>t.DacModel,DacPreTrainedModel:()=>t.DacPreTrainedModel,DataTypeMap:()=>h.DataTypeMap,DebertaForMaskedLM:()=>t.DebertaForMaskedLM,DebertaForQuestionAnswering:()=>t.DebertaForQuestionAnswering,DebertaForSequenceClassification:()=>t.DebertaForSequenceClassification,DebertaForTokenClassification:()=>t.DebertaForTokenClassification,DebertaModel:()=>t.DebertaModel,DebertaPreTrainedModel:()=>t.DebertaPreTrainedModel,DebertaTokenizer:()=>i.DebertaTokenizer,DebertaV2ForMaskedLM:()=>t.DebertaV2ForMaskedLM,DebertaV2ForQuestionAnswering:()=>t.DebertaV2ForQuestionAnswering,DebertaV2ForSequenceClassification:()=>t.DebertaV2ForSequenceClassification,DebertaV2ForTokenClassification:()=>t.DebertaV2ForTokenClassification,DebertaV2Model:()=>t.DebertaV2Model,DebertaV2PreTrainedModel:()=>t.DebertaV2PreTrainedModel,DebertaV2Tokenizer:()=>i.DebertaV2Tokenizer,DecisionTransformerModel:()=>t.DecisionTransformerModel,DecisionTransformerPreTrainedModel:()=>t.DecisionTransformerPreTrainedModel,DeiTFeatureExtractor:()=>M.DeiTFeatureExtractor,DeiTForImageClassification:()=>t.DeiTForImageClassification,DeiTImageProcessor:()=>M.DeiTImageProcessor,DeiTModel:()=>t.DeiTModel,DeiTPreTrainedModel:()=>t.DeiTPreTrainedModel,DepthAnythingForDepthEstimation:()=>t.DepthAnythingForDepthEstimation,DepthAnythingPreTrainedModel:()=>t.DepthAnythingPreTrainedModel,DepthEstimationPipeline:()=>r.DepthEstimationPipeline,DepthProForDepthEstimation:()=>t.DepthProForDepthEstimation,DepthProPreTrainedModel:()=>t.DepthProPreTrainedModel,DetrFeatureExtractor:()=>M.DetrFeatureExtractor,DetrForObjectDetection:()=>t.DetrForObjectDetection,DetrForSegmentation:()=>t.DetrForSegmentation,DetrImageProcessor:()=>M.DetrImageProcessor,DetrModel:()=>t.DetrModel,DetrObjectDetectionOutput:()=>t.DetrObjectDetectionOutput,DetrPreTrainedModel:()=>t.DetrPreTrainedModel,DetrSegmentationOutput:()=>t.DetrSegmentationOutput,Dinov2ForImageClassification:()=>t.Dinov2ForImageClassification,Dinov2Model:()=>t.Dinov2Model,Dinov2PreTrainedModel:()=>t.Dinov2PreTrainedModel,Dinov2WithRegistersForImageClassification:()=>t.Dinov2WithRegistersForImageClassification,Dinov2WithRegistersModel:()=>t.Dinov2WithRegistersModel,Dinov2WithRegistersPreTrainedModel:()=>t.Dinov2WithRegistersPreTrainedModel,DistilBertForMaskedLM:()=>t.DistilBertForMaskedLM,DistilBertForQuestionAnswering:()=>t.DistilBertForQuestionAnswering,DistilBertForSequenceClassification:()=>t.DistilBertForSequenceClassification,DistilBertForTokenClassification:()=>t.DistilBertForTokenClassification,DistilBertModel:()=>t.DistilBertModel,DistilBertPreTrainedModel:()=>t.DistilBertPreTrainedModel,DistilBertTokenizer:()=>i.DistilBertTokenizer,DocumentQuestionAnsweringPipeline:()=>r.DocumentQuestionAnsweringPipeline,DonutFeatureExtractor:()=>M.DonutFeatureExtractor,DonutImageProcessor:()=>M.DonutImageProcessor,DonutSwinModel:()=>t.DonutSwinModel,DonutSwinPreTrainedModel:()=>t.DonutSwinPreTrainedModel,EfficientNetForImageClassification:()=>t.EfficientNetForImageClassification,EfficientNetImageProcessor:()=>M.EfficientNetImageProcessor,EfficientNetModel:()=>t.EfficientNetModel,EfficientNetPreTrainedModel:()=>t.EfficientNetPreTrainedModel,ElectraForMaskedLM:()=>t.ElectraForMaskedLM,ElectraForQuestionAnswering:()=>t.ElectraForQuestionAnswering,ElectraForSequenceClassification:()=>t.ElectraForSequenceClassification,ElectraForTokenClassification:()=>t.ElectraForTokenClassification,ElectraModel:()=>t.ElectraModel,ElectraPreTrainedModel:()=>t.ElectraPreTrainedModel,ElectraTokenizer:()=>i.ElectraTokenizer,EncodecFeatureExtractor:()=>m.EncodecFeatureExtractor,EosTokenCriteria:()=>A.EosTokenCriteria,EsmForMaskedLM:()=>t.EsmForMaskedLM,EsmForSequenceClassification:()=>t.EsmForSequenceClassification,EsmForTokenClassification:()=>t.EsmForTokenClassification,EsmModel:()=>t.EsmModel,EsmPreTrainedModel:()=>t.EsmPreTrainedModel,EsmTokenizer:()=>i.EsmTokenizer,ExaoneForCausalLM:()=>t.ExaoneForCausalLM,ExaoneModel:()=>t.ExaoneModel,ExaonePreTrainedModel:()=>t.ExaonePreTrainedModel,FFT:()=>f.FFT,FalconForCausalLM:()=>t.FalconForCausalLM,FalconModel:()=>t.FalconModel,FalconPreTrainedModel:()=>t.FalconPreTrainedModel,FalconTokenizer:()=>i.FalconTokenizer,FastViTForImageClassification:()=>t.FastViTForImageClassification,FastViTModel:()=>t.FastViTModel,FastViTPreTrainedModel:()=>t.FastViTPreTrainedModel,FeatureExtractionPipeline:()=>r.FeatureExtractionPipeline,FeatureExtractor:()=>y.FeatureExtractor,FillMaskPipeline:()=>r.FillMaskPipeline,Florence2ForConditionalGeneration:()=>t.Florence2ForConditionalGeneration,Florence2PreTrainedModel:()=>t.Florence2PreTrainedModel,Florence2Processor:()=>E.Florence2Processor,ForcedBOSTokenLogitsProcessor:()=>$.ForcedBOSTokenLogitsProcessor,ForcedEOSTokenLogitsProcessor:()=>$.ForcedEOSTokenLogitsProcessor,GLPNFeatureExtractor:()=>M.GLPNFeatureExtractor,GLPNForDepthEstimation:()=>t.GLPNForDepthEstimation,GLPNModel:()=>t.GLPNModel,GLPNPreTrainedModel:()=>t.GLPNPreTrainedModel,GPT2LMHeadModel:()=>t.GPT2LMHeadModel,GPT2Model:()=>t.GPT2Model,GPT2PreTrainedModel:()=>t.GPT2PreTrainedModel,GPT2Tokenizer:()=>i.GPT2Tokenizer,GPTBigCodeForCausalLM:()=>t.GPTBigCodeForCausalLM,GPTBigCodeModel:()=>t.GPTBigCodeModel,GPTBigCodePreTrainedModel:()=>t.GPTBigCodePreTrainedModel,GPTJForCausalLM:()=>t.GPTJForCausalLM,GPTJModel:()=>t.GPTJModel,GPTJPreTrainedModel:()=>t.GPTJPreTrainedModel,GPTNeoForCausalLM:()=>t.GPTNeoForCausalLM,GPTNeoModel:()=>t.GPTNeoModel,GPTNeoPreTrainedModel:()=>t.GPTNeoPreTrainedModel,GPTNeoXForCausalLM:()=>t.GPTNeoXForCausalLM,GPTNeoXModel:()=>t.GPTNeoXModel,GPTNeoXPreTrainedModel:()=>t.GPTNeoXPreTrainedModel,GPTNeoXTokenizer:()=>i.GPTNeoXTokenizer,Gemma2ForCausalLM:()=>t.Gemma2ForCausalLM,Gemma2Model:()=>t.Gemma2Model,Gemma2PreTrainedModel:()=>t.Gemma2PreTrainedModel,Gemma3ForCausalLM:()=>t.Gemma3ForCausalLM,Gemma3Model:()=>t.Gemma3Model,Gemma3PreTrainedModel:()=>t.Gemma3PreTrainedModel,GemmaForCausalLM:()=>t.GemmaForCausalLM,GemmaModel:()=>t.GemmaModel,GemmaPreTrainedModel:()=>t.GemmaPreTrainedModel,GemmaTokenizer:()=>i.GemmaTokenizer,GlmForCausalLM:()=>t.GlmForCausalLM,GlmModel:()=>t.GlmModel,GlmPreTrainedModel:()=>t.GlmPreTrainedModel,GraniteForCausalLM:()=>t.GraniteForCausalLM,GraniteModel:()=>t.GraniteModel,GranitePreTrainedModel:()=>t.GranitePreTrainedModel,Grok1Tokenizer:()=>i.Grok1Tokenizer,GroundingDinoForObjectDetection:()=>t.GroundingDinoForObjectDetection,GroundingDinoImageProcessor:()=>M.GroundingDinoImageProcessor,GroundingDinoPreTrainedModel:()=>t.GroundingDinoPreTrainedModel,GroundingDinoProcessor:()=>E.GroundingDinoProcessor,GroupViTModel:()=>t.GroupViTModel,GroupViTPreTrainedModel:()=>t.GroupViTPreTrainedModel,HeliumForCausalLM:()=>t.HeliumForCausalLM,HeliumModel:()=>t.HeliumModel,HeliumPreTrainedModel:()=>t.HeliumPreTrainedModel,HerbertTokenizer:()=>i.HerbertTokenizer,HieraForImageClassification:()=>t.HieraForImageClassification,HieraModel:()=>t.HieraModel,HieraPreTrainedModel:()=>t.HieraPreTrainedModel,HubertForCTC:()=>t.HubertForCTC,HubertForSequenceClassification:()=>t.HubertForSequenceClassification,HubertModel:()=>t.HubertModel,HubertPreTrainedModel:()=>t.HubertPreTrainedModel,IJepaForImageClassification:()=>t.IJepaForImageClassification,IJepaModel:()=>t.IJepaModel,IJepaPreTrainedModel:()=>t.IJepaPreTrainedModel,Idefics3ForConditionalGeneration:()=>t.Idefics3ForConditionalGeneration,Idefics3ImageProcessor:()=>M.Idefics3ImageProcessor,Idefics3PreTrainedModel:()=>t.Idefics3PreTrainedModel,Idefics3Processor:()=>E.Idefics3Processor,ImageClassificationPipeline:()=>r.ImageClassificationPipeline,ImageFeatureExtractionPipeline:()=>r.ImageFeatureExtractionPipeline,ImageFeatureExtractor:()=>m.ImageFeatureExtractor,ImageMattingOutput:()=>t.ImageMattingOutput,ImageProcessor:()=>T.ImageProcessor,ImageSegmentationPipeline:()=>r.ImageSegmentationPipeline,ImageToImagePipeline:()=>r.ImageToImagePipeline,ImageToTextPipeline:()=>r.ImageToTextPipeline,InterruptableStoppingCriteria:()=>A.InterruptableStoppingCriteria,JAISLMHeadModel:()=>t.JAISLMHeadModel,JAISModel:()=>t.JAISModel,JAISPreTrainedModel:()=>t.JAISPreTrainedModel,JinaCLIPImageProcessor:()=>M.JinaCLIPImageProcessor,JinaCLIPModel:()=>t.JinaCLIPModel,JinaCLIPPreTrainedModel:()=>t.JinaCLIPPreTrainedModel,JinaCLIPProcessor:()=>E.JinaCLIPProcessor,JinaCLIPTextModel:()=>t.JinaCLIPTextModel,JinaCLIPVisionModel:()=>t.JinaCLIPVisionModel,LiteWhisperForConditionalGeneration:()=>t.LiteWhisperForConditionalGeneration,LlamaForCausalLM:()=>t.LlamaForCausalLM,LlamaModel:()=>t.LlamaModel,LlamaPreTrainedModel:()=>t.LlamaPreTrainedModel,LlamaTokenizer:()=>i.LlamaTokenizer,LlavaForConditionalGeneration:()=>t.LlavaForConditionalGeneration,LlavaOnevisionForConditionalGeneration:()=>t.LlavaOnevisionForConditionalGeneration,LlavaOnevisionImageProcessor:()=>M.LlavaOnevisionImageProcessor,LlavaPreTrainedModel:()=>t.LlavaPreTrainedModel,LogitsProcessor:()=>$.LogitsProcessor,LogitsProcessorList:()=>$.LogitsProcessorList,LogitsWarper:()=>$.LogitsWarper,LongT5ForConditionalGeneration:()=>t.LongT5ForConditionalGeneration,LongT5Model:()=>t.LongT5Model,LongT5PreTrainedModel:()=>t.LongT5PreTrainedModel,M2M100ForConditionalGeneration:()=>t.M2M100ForConditionalGeneration,M2M100Model:()=>t.M2M100Model,M2M100PreTrainedModel:()=>t.M2M100PreTrainedModel,M2M100Tokenizer:()=>i.M2M100Tokenizer,MBart50Tokenizer:()=>i.MBart50Tokenizer,MBartForCausalLM:()=>t.MBartForCausalLM,MBartForConditionalGeneration:()=>t.MBartForConditionalGeneration,MBartForSequenceClassification:()=>t.MBartForSequenceClassification,MBartModel:()=>t.MBartModel,MBartPreTrainedModel:()=>t.MBartPreTrainedModel,MBartTokenizer:()=>i.MBartTokenizer,MPNetForMaskedLM:()=>t.MPNetForMaskedLM,MPNetForQuestionAnswering:()=>t.MPNetForQuestionAnswering,MPNetForSequenceClassification:()=>t.MPNetForSequenceClassification,MPNetForTokenClassification:()=>t.MPNetForTokenClassification,MPNetModel:()=>t.MPNetModel,MPNetPreTrainedModel:()=>t.MPNetPreTrainedModel,MPNetTokenizer:()=>i.MPNetTokenizer,MT5ForConditionalGeneration:()=>t.MT5ForConditionalGeneration,MT5Model:()=>t.MT5Model,MT5PreTrainedModel:()=>t.MT5PreTrainedModel,MarianMTModel:()=>t.MarianMTModel,MarianModel:()=>t.MarianModel,MarianPreTrainedModel:()=>t.MarianPreTrainedModel,MarianTokenizer:()=>i.MarianTokenizer,Mask2FormerImageProcessor:()=>M.Mask2FormerImageProcessor,MaskFormerFeatureExtractor:()=>M.MaskFormerFeatureExtractor,MaskFormerForInstanceSegmentation:()=>t.MaskFormerForInstanceSegmentation,MaskFormerImageProcessor:()=>M.MaskFormerImageProcessor,MaskFormerModel:()=>t.MaskFormerModel,MaskFormerPreTrainedModel:()=>t.MaskFormerPreTrainedModel,MaskedLMOutput:()=>t.MaskedLMOutput,MaxLengthCriteria:()=>A.MaxLengthCriteria,Metric3DForDepthEstimation:()=>t.Metric3DForDepthEstimation,Metric3DPreTrainedModel:()=>t.Metric3DPreTrainedModel,Metric3Dv2ForDepthEstimation:()=>t.Metric3Dv2ForDepthEstimation,Metric3Dv2PreTrainedModel:()=>t.Metric3Dv2PreTrainedModel,MgpstrForSceneTextRecognition:()=>t.MgpstrForSceneTextRecognition,MgpstrModelOutput:()=>t.MgpstrModelOutput,MgpstrPreTrainedModel:()=>t.MgpstrPreTrainedModel,MgpstrProcessor:()=>E.MgpstrProcessor,MgpstrTokenizer:()=>i.MgpstrTokenizer,MimiDecoderModel:()=>t.MimiDecoderModel,MimiDecoderOutput:()=>t.MimiDecoderOutput,MimiEncoderModel:()=>t.MimiEncoderModel,MimiEncoderOutput:()=>t.MimiEncoderOutput,MimiModel:()=>t.MimiModel,MimiPreTrainedModel:()=>t.MimiPreTrainedModel,MinLengthLogitsProcessor:()=>$.MinLengthLogitsProcessor,MinNewTokensLengthLogitsProcessor:()=>$.MinNewTokensLengthLogitsProcessor,MistralForCausalLM:()=>t.MistralForCausalLM,MistralModel:()=>t.MistralModel,MistralPreTrainedModel:()=>t.MistralPreTrainedModel,MobileBertForMaskedLM:()=>t.MobileBertForMaskedLM,MobileBertForQuestionAnswering:()=>t.MobileBertForQuestionAnswering,MobileBertForSequenceClassification:()=>t.MobileBertForSequenceClassification,MobileBertModel:()=>t.MobileBertModel,MobileBertPreTrainedModel:()=>t.MobileBertPreTrainedModel,MobileBertTokenizer:()=>i.MobileBertTokenizer,MobileLLMForCausalLM:()=>t.MobileLLMForCausalLM,MobileLLMModel:()=>t.MobileLLMModel,MobileLLMPreTrainedModel:()=>t.MobileLLMPreTrainedModel,MobileNetV1FeatureExtractor:()=>M.MobileNetV1FeatureExtractor,MobileNetV1ForImageClassification:()=>t.MobileNetV1ForImageClassification,MobileNetV1ForSemanticSegmentation:()=>t.MobileNetV1ForSemanticSegmentation,MobileNetV1ImageProcessor:()=>M.MobileNetV1ImageProcessor,MobileNetV1Model:()=>t.MobileNetV1Model,MobileNetV1PreTrainedModel:()=>t.MobileNetV1PreTrainedModel,MobileNetV2FeatureExtractor:()=>M.MobileNetV2FeatureExtractor,MobileNetV2ForImageClassification:()=>t.MobileNetV2ForImageClassification,MobileNetV2ForSemanticSegmentation:()=>t.MobileNetV2ForSemanticSegmentation,MobileNetV2ImageProcessor:()=>M.MobileNetV2ImageProcessor,MobileNetV2Model:()=>t.MobileNetV2Model,MobileNetV2PreTrainedModel:()=>t.MobileNetV2PreTrainedModel,MobileNetV3FeatureExtractor:()=>M.MobileNetV3FeatureExtractor,MobileNetV3ForImageClassification:()=>t.MobileNetV3ForImageClassification,MobileNetV3ForSemanticSegmentation:()=>t.MobileNetV3ForSemanticSegmentation,MobileNetV3ImageProcessor:()=>M.MobileNetV3ImageProcessor,MobileNetV3Model:()=>t.MobileNetV3Model,MobileNetV3PreTrainedModel:()=>t.MobileNetV3PreTrainedModel,MobileNetV4FeatureExtractor:()=>M.MobileNetV4FeatureExtractor,MobileNetV4ForImageClassification:()=>t.MobileNetV4ForImageClassification,MobileNetV4ForSemanticSegmentation:()=>t.MobileNetV4ForSemanticSegmentation,MobileNetV4ImageProcessor:()=>M.MobileNetV4ImageProcessor,MobileNetV4Model:()=>t.MobileNetV4Model,MobileNetV4PreTrainedModel:()=>t.MobileNetV4PreTrainedModel,MobileViTFeatureExtractor:()=>M.MobileViTFeatureExtractor,MobileViTForImageClassification:()=>t.MobileViTForImageClassification,MobileViTImageProcessor:()=>M.MobileViTImageProcessor,MobileViTModel:()=>t.MobileViTModel,MobileViTPreTrainedModel:()=>t.MobileViTPreTrainedModel,MobileViTV2ForImageClassification:()=>t.MobileViTV2ForImageClassification,MobileViTV2Model:()=>t.MobileViTV2Model,MobileViTV2PreTrainedModel:()=>t.MobileViTV2PreTrainedModel,ModelOutput:()=>t.ModelOutput,ModernBertForMaskedLM:()=>t.ModernBertForMaskedLM,ModernBertForSequenceClassification:()=>t.ModernBertForSequenceClassification,ModernBertForTokenClassification:()=>t.ModernBertForTokenClassification,ModernBertModel:()=>t.ModernBertModel,ModernBertPreTrainedModel:()=>t.ModernBertPreTrainedModel,Moondream1ForConditionalGeneration:()=>t.Moondream1ForConditionalGeneration,MoonshineFeatureExtractor:()=>m.MoonshineFeatureExtractor,MoonshineForConditionalGeneration:()=>t.MoonshineForConditionalGeneration,MoonshineModel:()=>t.MoonshineModel,MoonshinePreTrainedModel:()=>t.MoonshinePreTrainedModel,MoonshineProcessor:()=>E.MoonshineProcessor,MptForCausalLM:()=>t.MptForCausalLM,MptModel:()=>t.MptModel,MptPreTrainedModel:()=>t.MptPreTrainedModel,MultiModalityCausalLM:()=>t.MultiModalityCausalLM,MultiModalityPreTrainedModel:()=>t.MultiModalityPreTrainedModel,MusicgenForCausalLM:()=>t.MusicgenForCausalLM,MusicgenForConditionalGeneration:()=>t.MusicgenForConditionalGeneration,MusicgenModel:()=>t.MusicgenModel,MusicgenPreTrainedModel:()=>t.MusicgenPreTrainedModel,NllbTokenizer:()=>i.NllbTokenizer,NoBadWordsLogitsProcessor:()=>$.NoBadWordsLogitsProcessor,NoRepeatNGramLogitsProcessor:()=>$.NoRepeatNGramLogitsProcessor,NomicBertModel:()=>t.NomicBertModel,NomicBertPreTrainedModel:()=>t.NomicBertPreTrainedModel,NougatImageProcessor:()=>M.NougatImageProcessor,NougatTokenizer:()=>i.NougatTokenizer,OPTForCausalLM:()=>t.OPTForCausalLM,OPTModel:()=>t.OPTModel,OPTPreTrainedModel:()=>t.OPTPreTrainedModel,ObjectDetectionPipeline:()=>r.ObjectDetectionPipeline,Olmo2ForCausalLM:()=>t.Olmo2ForCausalLM,Olmo2Model:()=>t.Olmo2Model,Olmo2PreTrainedModel:()=>t.Olmo2PreTrainedModel,OlmoForCausalLM:()=>t.OlmoForCausalLM,OlmoModel:()=>t.OlmoModel,OlmoPreTrainedModel:()=>t.OlmoPreTrainedModel,OpenELMForCausalLM:()=>t.OpenELMForCausalLM,OpenELMModel:()=>t.OpenELMModel,OpenELMPreTrainedModel:()=>t.OpenELMPreTrainedModel,OwlViTFeatureExtractor:()=>M.OwlViTFeatureExtractor,OwlViTForObjectDetection:()=>t.OwlViTForObjectDetection,OwlViTImageProcessor:()=>M.OwlViTImageProcessor,OwlViTModel:()=>t.OwlViTModel,OwlViTPreTrainedModel:()=>t.OwlViTPreTrainedModel,OwlViTProcessor:()=>E.OwlViTProcessor,Owlv2ForObjectDetection:()=>t.Owlv2ForObjectDetection,Owlv2ImageProcessor:()=>M.Owlv2ImageProcessor,Owlv2Model:()=>t.Owlv2Model,Owlv2PreTrainedModel:()=>t.Owlv2PreTrainedModel,PaliGemmaForConditionalGeneration:()=>t.PaliGemmaForConditionalGeneration,PaliGemmaPreTrainedModel:()=>t.PaliGemmaPreTrainedModel,PaliGemmaProcessor:()=>E.PaliGemmaProcessor,PatchTSMixerForPrediction:()=>t.PatchTSMixerForPrediction,PatchTSMixerModel:()=>t.PatchTSMixerModel,PatchTSMixerPreTrainedModel:()=>t.PatchTSMixerPreTrainedModel,PatchTSTForPrediction:()=>t.PatchTSTForPrediction,PatchTSTModel:()=>t.PatchTSTModel,PatchTSTPreTrainedModel:()=>t.PatchTSTPreTrainedModel,Phi3ForCausalLM:()=>t.Phi3ForCausalLM,Phi3Model:()=>t.Phi3Model,Phi3PreTrainedModel:()=>t.Phi3PreTrainedModel,Phi3VForCausalLM:()=>t.Phi3VForCausalLM,Phi3VImageProcessor:()=>M.Phi3VImageProcessor,Phi3VPreTrainedModel:()=>t.Phi3VPreTrainedModel,Phi3VProcessor:()=>E.Phi3VProcessor,PhiForCausalLM:()=>t.PhiForCausalLM,PhiModel:()=>t.PhiModel,PhiPreTrainedModel:()=>t.PhiPreTrainedModel,Pipeline:()=>r.Pipeline,PreTrainedModel:()=>t.PreTrainedModel,PreTrainedTokenizer:()=>i.PreTrainedTokenizer,PretrainedConfig:()=>o.PretrainedConfig,PretrainedMixin:()=>t.PretrainedMixin,Processor:()=>z.Processor,PvtForImageClassification:()=>t.PvtForImageClassification,PvtImageProcessor:()=>M.PvtImageProcessor,PvtModel:()=>t.PvtModel,PvtPreTrainedModel:()=>t.PvtPreTrainedModel,PyAnnoteFeatureExtractor:()=>m.PyAnnoteFeatureExtractor,PyAnnoteForAudioFrameClassification:()=>t.PyAnnoteForAudioFrameClassification,PyAnnoteModel:()=>t.PyAnnoteModel,PyAnnotePreTrainedModel:()=>t.PyAnnotePreTrainedModel,PyAnnoteProcessor:()=>E.PyAnnoteProcessor,QuestionAnsweringModelOutput:()=>t.QuestionAnsweringModelOutput,QuestionAnsweringPipeline:()=>r.QuestionAnsweringPipeline,Qwen2ForCausalLM:()=>t.Qwen2ForCausalLM,Qwen2Model:()=>t.Qwen2Model,Qwen2PreTrainedModel:()=>t.Qwen2PreTrainedModel,Qwen2Tokenizer:()=>i.Qwen2Tokenizer,Qwen2VLForConditionalGeneration:()=>t.Qwen2VLForConditionalGeneration,Qwen2VLImageProcessor:()=>M.Qwen2VLImageProcessor,Qwen2VLPreTrainedModel:()=>t.Qwen2VLPreTrainedModel,Qwen2VLProcessor:()=>E.Qwen2VLProcessor,Qwen3ForCausalLM:()=>t.Qwen3ForCausalLM,Qwen3Model:()=>t.Qwen3Model,Qwen3PreTrainedModel:()=>t.Qwen3PreTrainedModel,RFDetrForObjectDetection:()=>t.RFDetrForObjectDetection,RFDetrModel:()=>t.RFDetrModel,RFDetrObjectDetectionOutput:()=>t.RFDetrObjectDetectionOutput,RFDetrPreTrainedModel:()=>t.RFDetrPreTrainedModel,RTDetrForObjectDetection:()=>t.RTDetrForObjectDetection,RTDetrImageProcessor:()=>M.RTDetrImageProcessor,RTDetrModel:()=>t.RTDetrModel,RTDetrObjectDetectionOutput:()=>t.RTDetrObjectDetectionOutput,RTDetrPreTrainedModel:()=>t.RTDetrPreTrainedModel,RTDetrV2ForObjectDetection:()=>t.RTDetrV2ForObjectDetection,RTDetrV2Model:()=>t.RTDetrV2Model,RTDetrV2ObjectDetectionOutput:()=>t.RTDetrV2ObjectDetectionOutput,RTDetrV2PreTrainedModel:()=>t.RTDetrV2PreTrainedModel,RawAudio:()=>n.RawAudio,RawImage:()=>u.RawImage,RawVideo:()=>d.RawVideo,RawVideoFrame:()=>d.RawVideoFrame,RepetitionPenaltyLogitsProcessor:()=>$.RepetitionPenaltyLogitsProcessor,ResNetForImageClassification:()=>t.ResNetForImageClassification,ResNetModel:()=>t.ResNetModel,ResNetPreTrainedModel:()=>t.ResNetPreTrainedModel,RoFormerForMaskedLM:()=>t.RoFormerForMaskedLM,RoFormerForQuestionAnswering:()=>t.RoFormerForQuestionAnswering,RoFormerForSequenceClassification:()=>t.RoFormerForSequenceClassification,RoFormerForTokenClassification:()=>t.RoFormerForTokenClassification,RoFormerModel:()=>t.RoFormerModel,RoFormerPreTrainedModel:()=>t.RoFormerPreTrainedModel,RoFormerTokenizer:()=>i.RoFormerTokenizer,RobertaForMaskedLM:()=>t.RobertaForMaskedLM,RobertaForQuestionAnswering:()=>t.RobertaForQuestionAnswering,RobertaForSequenceClassification:()=>t.RobertaForSequenceClassification,RobertaForTokenClassification:()=>t.RobertaForTokenClassification,RobertaModel:()=>t.RobertaModel,RobertaPreTrainedModel:()=>t.RobertaPreTrainedModel,RobertaTokenizer:()=>i.RobertaTokenizer,SamImageProcessor:()=>M.SamImageProcessor,SamImageSegmentationOutput:()=>t.SamImageSegmentationOutput,SamModel:()=>t.SamModel,SamPreTrainedModel:()=>t.SamPreTrainedModel,SamProcessor:()=>E.SamProcessor,SapiensForDepthEstimation:()=>t.SapiensForDepthEstimation,SapiensForNormalEstimation:()=>t.SapiensForNormalEstimation,SapiensForSemanticSegmentation:()=>t.SapiensForSemanticSegmentation,SapiensPreTrainedModel:()=>t.SapiensPreTrainedModel,SeamlessM4TFeatureExtractor:()=>m.SeamlessM4TFeatureExtractor,SegformerFeatureExtractor:()=>M.SegformerFeatureExtractor,SegformerForImageClassification:()=>t.SegformerForImageClassification,SegformerForSemanticSegmentation:()=>t.SegformerForSemanticSegmentation,SegformerImageProcessor:()=>M.SegformerImageProcessor,SegformerModel:()=>t.SegformerModel,SegformerPreTrainedModel:()=>t.SegformerPreTrainedModel,Seq2SeqLMOutput:()=>t.Seq2SeqLMOutput,SequenceClassifierOutput:()=>t.SequenceClassifierOutput,SiglipImageProcessor:()=>M.SiglipImageProcessor,SiglipModel:()=>t.SiglipModel,SiglipPreTrainedModel:()=>t.SiglipPreTrainedModel,SiglipTextModel:()=>t.SiglipTextModel,SiglipTokenizer:()=>i.SiglipTokenizer,SiglipVisionModel:()=>t.SiglipVisionModel,SmolVLMForConditionalGeneration:()=>t.SmolVLMForConditionalGeneration,SmolVLMImageProcessor:()=>M.SmolVLMImageProcessor,SmolVLMProcessor:()=>E.SmolVLMProcessor,SnacDecoderModel:()=>t.SnacDecoderModel,SnacEncoderModel:()=>t.SnacEncoderModel,SnacFeatureExtractor:()=>m.SnacFeatureExtractor,SnacModel:()=>t.SnacModel,SnacPreTrainedModel:()=>t.SnacPreTrainedModel,SpeechT5FeatureExtractor:()=>m.SpeechT5FeatureExtractor,SpeechT5ForSpeechToText:()=>t.SpeechT5ForSpeechToText,SpeechT5ForTextToSpeech:()=>t.SpeechT5ForTextToSpeech,SpeechT5HifiGan:()=>t.SpeechT5HifiGan,SpeechT5Model:()=>t.SpeechT5Model,SpeechT5PreTrainedModel:()=>t.SpeechT5PreTrainedModel,SpeechT5Processor:()=>E.SpeechT5Processor,SpeechT5Tokenizer:()=>i.SpeechT5Tokenizer,SqueezeBertForMaskedLM:()=>t.SqueezeBertForMaskedLM,SqueezeBertForQuestionAnswering:()=>t.SqueezeBertForQuestionAnswering,SqueezeBertForSequenceClassification:()=>t.SqueezeBertForSequenceClassification,SqueezeBertModel:()=>t.SqueezeBertModel,SqueezeBertPreTrainedModel:()=>t.SqueezeBertPreTrainedModel,SqueezeBertTokenizer:()=>i.SqueezeBertTokenizer,StableLmForCausalLM:()=>t.StableLmForCausalLM,StableLmModel:()=>t.StableLmModel,StableLmPreTrainedModel:()=>t.StableLmPreTrainedModel,Starcoder2ForCausalLM:()=>t.Starcoder2ForCausalLM,Starcoder2Model:()=>t.Starcoder2Model,Starcoder2PreTrainedModel:()=>t.Starcoder2PreTrainedModel,StoppingCriteria:()=>A.StoppingCriteria,StoppingCriteriaList:()=>A.StoppingCriteriaList,StyleTextToSpeech2Model:()=>t.StyleTextToSpeech2Model,StyleTextToSpeech2PreTrainedModel:()=>t.StyleTextToSpeech2PreTrainedModel,SummarizationPipeline:()=>r.SummarizationPipeline,SuppressTokensAtBeginLogitsProcessor:()=>$.SuppressTokensAtBeginLogitsProcessor,Swin2SRForImageSuperResolution:()=>t.Swin2SRForImageSuperResolution,Swin2SRImageProcessor:()=>M.Swin2SRImageProcessor,Swin2SRModel:()=>t.Swin2SRModel,Swin2SRPreTrainedModel:()=>t.Swin2SRPreTrainedModel,SwinForImageClassification:()=>t.SwinForImageClassification,SwinForSemanticSegmentation:()=>t.SwinForSemanticSegmentation,SwinModel:()=>t.SwinModel,SwinPreTrainedModel:()=>t.SwinPreTrainedModel,T5ForConditionalGeneration:()=>t.T5ForConditionalGeneration,T5Model:()=>t.T5Model,T5PreTrainedModel:()=>t.T5PreTrainedModel,T5Tokenizer:()=>i.T5Tokenizer,TableTransformerForObjectDetection:()=>t.TableTransformerForObjectDetection,TableTransformerModel:()=>t.TableTransformerModel,TableTransformerObjectDetectionOutput:()=>t.TableTransformerObjectDetectionOutput,TableTransformerPreTrainedModel:()=>t.TableTransformerPreTrainedModel,TemperatureLogitsWarper:()=>$.TemperatureLogitsWarper,Tensor:()=>h.Tensor,Text2TextGenerationPipeline:()=>r.Text2TextGenerationPipeline,TextClassificationPipeline:()=>r.TextClassificationPipeline,TextGenerationPipeline:()=>r.TextGenerationPipeline,TextStreamer:()=>D.TextStreamer,TextToAudioPipeline:()=>r.TextToAudioPipeline,TokenClassificationPipeline:()=>r.TokenClassificationPipeline,TokenClassifierOutput:()=>t.TokenClassifierOutput,TokenizerModel:()=>i.TokenizerModel,TopKLogitsWarper:()=>$.TopKLogitsWarper,TopPLogitsWarper:()=>$.TopPLogitsWarper,TrOCRForCausalLM:()=>t.TrOCRForCausalLM,TrOCRPreTrainedModel:()=>t.TrOCRPreTrainedModel,TranslationPipeline:()=>r.TranslationPipeline,UltravoxModel:()=>t.UltravoxModel,UltravoxPreTrainedModel:()=>t.UltravoxPreTrainedModel,UltravoxProcessor:()=>E.UltravoxProcessor,UniSpeechForCTC:()=>t.UniSpeechForCTC,UniSpeechForSequenceClassification:()=>t.UniSpeechForSequenceClassification,UniSpeechModel:()=>t.UniSpeechModel,UniSpeechPreTrainedModel:()=>t.UniSpeechPreTrainedModel,UniSpeechSatForAudioFrameClassification:()=>t.UniSpeechSatForAudioFrameClassification,UniSpeechSatForCTC:()=>t.UniSpeechSatForCTC,UniSpeechSatForSequenceClassification:()=>t.UniSpeechSatForSequenceClassification,UniSpeechSatModel:()=>t.UniSpeechSatModel,UniSpeechSatPreTrainedModel:()=>t.UniSpeechSatPreTrainedModel,VLChatProcessor:()=>E.VLChatProcessor,VLMImageProcessor:()=>M.VLMImageProcessor,ViTFeatureExtractor:()=>M.ViTFeatureExtractor,ViTForImageClassification:()=>t.ViTForImageClassification,ViTImageProcessor:()=>M.ViTImageProcessor,ViTMAEModel:()=>t.ViTMAEModel,ViTMAEPreTrainedModel:()=>t.ViTMAEPreTrainedModel,ViTMSNForImageClassification:()=>t.ViTMSNForImageClassification,ViTMSNModel:()=>t.ViTMSNModel,ViTMSNPreTrainedModel:()=>t.ViTMSNPreTrainedModel,ViTModel:()=>t.ViTModel,ViTPreTrainedModel:()=>t.ViTPreTrainedModel,VisionEncoderDecoderModel:()=>t.VisionEncoderDecoderModel,VitMatteForImageMatting:()=>t.VitMatteForImageMatting,VitMatteImageProcessor:()=>M.VitMatteImageProcessor,VitMattePreTrainedModel:()=>t.VitMattePreTrainedModel,VitPoseForPoseEstimation:()=>t.VitPoseForPoseEstimation,VitPoseImageProcessor:()=>M.VitPoseImageProcessor,VitPosePreTrainedModel:()=>t.VitPosePreTrainedModel,VitsModel:()=>t.VitsModel,VitsModelOutput:()=>t.VitsModelOutput,VitsPreTrainedModel:()=>t.VitsPreTrainedModel,VitsTokenizer:()=>i.VitsTokenizer,Wav2Vec2BertForCTC:()=>t.Wav2Vec2BertForCTC,Wav2Vec2BertForSequenceClassification:()=>t.Wav2Vec2BertForSequenceClassification,Wav2Vec2BertModel:()=>t.Wav2Vec2BertModel,Wav2Vec2BertPreTrainedModel:()=>t.Wav2Vec2BertPreTrainedModel,Wav2Vec2CTCTokenizer:()=>i.Wav2Vec2CTCTokenizer,Wav2Vec2FeatureExtractor:()=>m.Wav2Vec2FeatureExtractor,Wav2Vec2ForAudioFrameClassification:()=>t.Wav2Vec2ForAudioFrameClassification,Wav2Vec2ForCTC:()=>t.Wav2Vec2ForCTC,Wav2Vec2ForSequenceClassification:()=>t.Wav2Vec2ForSequenceClassification,Wav2Vec2Model:()=>t.Wav2Vec2Model,Wav2Vec2PreTrainedModel:()=>t.Wav2Vec2PreTrainedModel,Wav2Vec2Processor:()=>E.Wav2Vec2Processor,Wav2Vec2ProcessorWithLM:()=>E.Wav2Vec2ProcessorWithLM,WavLMForAudioFrameClassification:()=>t.WavLMForAudioFrameClassification,WavLMForCTC:()=>t.WavLMForCTC,WavLMForSequenceClassification:()=>t.WavLMForSequenceClassification,WavLMForXVector:()=>t.WavLMForXVector,WavLMModel:()=>t.WavLMModel,WavLMPreTrainedModel:()=>t.WavLMPreTrainedModel,WeSpeakerFeatureExtractor:()=>m.WeSpeakerFeatureExtractor,WeSpeakerResNetModel:()=>t.WeSpeakerResNetModel,WeSpeakerResNetPreTrainedModel:()=>t.WeSpeakerResNetPreTrainedModel,WhisperFeatureExtractor:()=>m.WhisperFeatureExtractor,WhisperForConditionalGeneration:()=>t.WhisperForConditionalGeneration,WhisperModel:()=>t.WhisperModel,WhisperPreTrainedModel:()=>t.WhisperPreTrainedModel,WhisperProcessor:()=>E.WhisperProcessor,WhisperTextStreamer:()=>D.WhisperTextStreamer,WhisperTimeStampLogitsProcessor:()=>$.WhisperTimeStampLogitsProcessor,WhisperTokenizer:()=>i.WhisperTokenizer,XLMForQuestionAnswering:()=>t.XLMForQuestionAnswering,XLMForSequenceClassification:()=>t.XLMForSequenceClassification,XLMForTokenClassification:()=>t.XLMForTokenClassification,XLMModel:()=>t.XLMModel,XLMPreTrainedModel:()=>t.XLMPreTrainedModel,XLMRobertaForMaskedLM:()=>t.XLMRobertaForMaskedLM,XLMRobertaForQuestionAnswering:()=>t.XLMRobertaForQuestionAnswering,XLMRobertaForSequenceClassification:()=>t.XLMRobertaForSequenceClassification,XLMRobertaForTokenClassification:()=>t.XLMRobertaForTokenClassification,XLMRobertaModel:()=>t.XLMRobertaModel,XLMRobertaPreTrainedModel:()=>t.XLMRobertaPreTrainedModel,XLMRobertaTokenizer:()=>i.XLMRobertaTokenizer,XLMTokenizer:()=>i.XLMTokenizer,XLMWithLMHeadModel:()=>t.XLMWithLMHeadModel,XVectorOutput:()=>t.XVectorOutput,YolosFeatureExtractor:()=>M.YolosFeatureExtractor,YolosForObjectDetection:()=>t.YolosForObjectDetection,YolosImageProcessor:()=>M.YolosImageProcessor,YolosModel:()=>t.YolosModel,YolosObjectDetectionOutput:()=>t.YolosObjectDetectionOutput,YolosPreTrainedModel:()=>t.YolosPreTrainedModel,ZeroShotAudioClassificationPipeline:()=>r.ZeroShotAudioClassificationPipeline,ZeroShotClassificationPipeline:()=>r.ZeroShotClassificationPipeline,ZeroShotImageClassificationPipeline:()=>r.ZeroShotImageClassificationPipeline,ZeroShotObjectDetectionPipeline:()=>r.ZeroShotObjectDetectionPipeline,bankers_round:()=>f.bankers_round,cat:()=>h.cat,cos_sim:()=>f.cos_sim,dot:()=>f.dot,dynamic_time_warping:()=>f.dynamic_time_warping,env:()=>e.env,full:()=>h.full,full_like:()=>h.full_like,getKeyValueShapes:()=>o.getKeyValueShapes,hamming:()=>n.hamming,hanning:()=>n.hanning,interpolate:()=>h.interpolate,interpolate_4d:()=>h.interpolate_4d,interpolate_data:()=>f.interpolate_data,is_chinese_char:()=>i.is_chinese_char,layer_norm:()=>h.layer_norm,load_image:()=>u.load_image,load_video:()=>d.load_video,log_softmax:()=>f.log_softmax,magnitude:()=>f.magnitude,matmul:()=>h.matmul,max:()=>f.max,mean:()=>h.mean,mean_pooling:()=>h.mean_pooling,medianFilter:()=>f.medianFilter,mel_filter_bank:()=>n.mel_filter_bank,min:()=>f.min,ones:()=>h.ones,ones_like:()=>h.ones_like,permute:()=>h.permute,permute_data:()=>f.permute_data,pipeline:()=>r.pipeline,quantize_embeddings:()=>h.quantize_embeddings,rand:()=>h.rand,read_audio:()=>n.read_audio,rfft:()=>h.rfft,round:()=>f.round,slice:()=>h.slice,softmax:()=>f.softmax,spectrogram:()=>n.spectrogram,stack:()=>h.stack,std_mean:()=>h.std_mean,topk:()=>h.topk,window_function:()=>n.window_function,zeros:()=>h.zeros,zeros_like:()=>h.zeros_like});var e=Yt("./src/env.js"),r=Yt("./src/pipelines.js"),t=Yt("./src/models.js"),i=Yt("./src/tokenizers.js"),o=Yt("./src/configs.js"),n=Yt("./src/utils/audio.js"),u=Yt("./src/utils/image.js"),d=Yt("./src/utils/video.js"),h=Yt("./src/utils/tensor.js"),f=Yt("./src/utils/maths.js"),y=Yt("./src/base/feature_extraction_utils.js"),m=Yt("./src/models/feature_extractors.js"),g=Yt("./src/models/auto/feature_extraction_auto.js"),T=Yt("./src/base/image_processors_utils.js"),M=Yt("./src/models/image_processors.js"),I=Yt("./src/models/auto/image_processing_auto.js"),z=Yt("./src/base/processing_utils.js"),E=Yt("./src/models/processors.js"),C=Yt("./src/models/auto/processing_auto.js"),D=Yt("./src/generation/streamers.js"),A=Yt("./src/generation/stopping_criteria.js"),$=Yt("./src/generation/logits_process.js")})();b.ASTFeatureExtractor;b.ASTForAudioClassification;b.ASTModel;b.ASTPreTrainedModel;b.AlbertForMaskedLM;b.AlbertForQuestionAnswering;b.AlbertForSequenceClassification;b.AlbertModel;b.AlbertPreTrainedModel;b.AlbertTokenizer;b.AudioClassificationPipeline;b.AutoConfig;b.AutoFeatureExtractor;b.AutoImageProcessor;var ZT=b.AutoModel;b.AutoModelForAudioClassification;b.AutoModelForAudioFrameClassification;b.AutoModelForAudioTextToText;b.AutoModelForCTC;b.AutoModelForCausalLM;b.AutoModelForDepthEstimation;b.AutoModelForDocumentQuestionAnswering;b.AutoModelForImageClassification;b.AutoModelForImageFeatureExtraction;b.AutoModelForImageMatting;b.AutoModelForImageSegmentation;b.AutoModelForImageTextToText;b.AutoModelForImageToImage;b.AutoModelForMaskGeneration;b.AutoModelForMaskedLM;b.AutoModelForNormalEstimation;b.AutoModelForObjectDetection;b.AutoModelForPoseEstimation;b.AutoModelForQuestionAnswering;b.AutoModelForSemanticSegmentation;b.AutoModelForSeq2SeqLM;b.AutoModelForSequenceClassification;b.AutoModelForSpeechSeq2Seq;b.AutoModelForTextToSpectrogram;b.AutoModelForTextToWaveform;b.AutoModelForTokenClassification;b.AutoModelForUniversalSegmentation;b.AutoModelForVision2Seq;b.AutoModelForXVector;b.AutoModelForZeroShotObjectDetection;b.AutoProcessor;b.AutoTokenizer;b.AutomaticSpeechRecognitionPipeline;b.BackgroundRemovalPipeline;b.BartForConditionalGeneration;b.BartForSequenceClassification;b.BartModel;b.BartPretrainedModel;b.BartTokenizer;b.BaseModelOutput;b.BaseStreamer;b.BeitFeatureExtractor;b.BeitForImageClassification;b.BeitModel;b.BeitPreTrainedModel;b.BertForMaskedLM;b.BertForQuestionAnswering;b.BertForSequenceClassification;b.BertForTokenClassification;b.BertModel;b.BertPreTrainedModel;b.BertTokenizer;b.BitImageProcessor;b.BlenderbotForConditionalGeneration;b.BlenderbotModel;b.BlenderbotPreTrainedModel;b.BlenderbotSmallForConditionalGeneration;b.BlenderbotSmallModel;b.BlenderbotSmallPreTrainedModel;b.BlenderbotSmallTokenizer;b.BlenderbotTokenizer;b.BloomForCausalLM;b.BloomModel;b.BloomPreTrainedModel;b.BloomTokenizer;b.CLIPFeatureExtractor;b.CLIPImageProcessor;b.CLIPModel;b.CLIPPreTrainedModel;b.CLIPSegForImageSegmentation;b.CLIPSegModel;b.CLIPSegPreTrainedModel;b.CLIPTextModel;b.CLIPTextModelWithProjection;b.CLIPTokenizer;b.CLIPVisionModel;b.CLIPVisionModelWithProjection;b.CamembertForMaskedLM;b.CamembertForQuestionAnswering;b.CamembertForSequenceClassification;b.CamembertForTokenClassification;b.CamembertModel;b.CamembertPreTrainedModel;b.CamembertTokenizer;b.CausalLMOutput;b.CausalLMOutputWithPast;b.ChineseCLIPFeatureExtractor;b.ChineseCLIPModel;b.ChineseCLIPPreTrainedModel;b.ClapAudioModelWithProjection;b.ClapFeatureExtractor;b.ClapModel;b.ClapPreTrainedModel;b.ClapTextModelWithProjection;b.ClassifierFreeGuidanceLogitsProcessor;b.CodeGenForCausalLM;b.CodeGenModel;b.CodeGenPreTrainedModel;b.CodeGenTokenizer;b.CodeLlamaTokenizer;b.CohereForCausalLM;b.CohereModel;b.CoherePreTrainedModel;b.CohereTokenizer;b.ConvBertForMaskedLM;b.ConvBertForQuestionAnswering;b.ConvBertForSequenceClassification;b.ConvBertForTokenClassification;b.ConvBertModel;b.ConvBertPreTrainedModel;b.ConvBertTokenizer;b.ConvNextFeatureExtractor;b.ConvNextForImageClassification;b.ConvNextImageProcessor;b.ConvNextModel;b.ConvNextPreTrainedModel;b.ConvNextV2ForImageClassification;b.ConvNextV2Model;b.ConvNextV2PreTrainedModel;b.DFineForObjectDetection;b.DFineModel;b.DFinePreTrainedModel;b.DPTFeatureExtractor;b.DPTForDepthEstimation;b.DPTImageProcessor;b.DPTModel;b.DPTPreTrainedModel;b.DacDecoderModel;b.DacDecoderOutput;b.DacEncoderModel;b.DacEncoderOutput;b.DacFeatureExtractor;b.DacModel;b.DacPreTrainedModel;b.DataTypeMap;b.DebertaForMaskedLM;b.DebertaForQuestionAnswering;b.DebertaForSequenceClassification;b.DebertaForTokenClassification;b.DebertaModel;b.DebertaPreTrainedModel;b.DebertaTokenizer;b.DebertaV2ForMaskedLM;b.DebertaV2ForQuestionAnswering;b.DebertaV2ForSequenceClassification;b.DebertaV2ForTokenClassification;b.DebertaV2Model;b.DebertaV2PreTrainedModel;b.DebertaV2Tokenizer;b.DecisionTransformerModel;b.DecisionTransformerPreTrainedModel;b.DeiTFeatureExtractor;b.DeiTForImageClassification;b.DeiTImageProcessor;b.DeiTModel;b.DeiTPreTrainedModel;b.DepthAnythingForDepthEstimation;b.DepthAnythingPreTrainedModel;b.DepthEstimationPipeline;b.DepthProForDepthEstimation;b.DepthProPreTrainedModel;b.DetrFeatureExtractor;b.DetrForObjectDetection;b.DetrForSegmentation;b.DetrImageProcessor;b.DetrModel;b.DetrObjectDetectionOutput;b.DetrPreTrainedModel;b.DetrSegmentationOutput;b.Dinov2ForImageClassification;b.Dinov2Model;b.Dinov2PreTrainedModel;b.Dinov2WithRegistersForImageClassification;b.Dinov2WithRegistersModel;b.Dinov2WithRegistersPreTrainedModel;b.DistilBertForMaskedLM;b.DistilBertForQuestionAnswering;b.DistilBertForSequenceClassification;b.DistilBertForTokenClassification;b.DistilBertModel;b.DistilBertPreTrainedModel;b.DistilBertTokenizer;b.DocumentQuestionAnsweringPipeline;b.DonutFeatureExtractor;b.DonutImageProcessor;b.DonutSwinModel;b.DonutSwinPreTrainedModel;b.EfficientNetForImageClassification;b.EfficientNetImageProcessor;b.EfficientNetModel;b.EfficientNetPreTrainedModel;b.ElectraForMaskedLM;b.ElectraForQuestionAnswering;b.ElectraForSequenceClassification;b.ElectraForTokenClassification;b.ElectraModel;b.ElectraPreTrainedModel;b.ElectraTokenizer;b.EncodecFeatureExtractor;b.EosTokenCriteria;b.EsmForMaskedLM;b.EsmForSequenceClassification;b.EsmForTokenClassification;b.EsmModel;b.EsmPreTrainedModel;b.EsmTokenizer;b.ExaoneForCausalLM;b.ExaoneModel;b.ExaonePreTrainedModel;b.FFT;b.FalconForCausalLM;b.FalconModel;b.FalconPreTrainedModel;b.FalconTokenizer;b.FastViTForImageClassification;b.FastViTModel;b.FastViTPreTrainedModel;b.FeatureExtractionPipeline;b.FeatureExtractor;b.FillMaskPipeline;b.Florence2ForConditionalGeneration;b.Florence2PreTrainedModel;b.Florence2Processor;b.ForcedBOSTokenLogitsProcessor;b.ForcedEOSTokenLogitsProcessor;b.GLPNFeatureExtractor;b.GLPNForDepthEstimation;b.GLPNModel;b.GLPNPreTrainedModel;b.GPT2LMHeadModel;b.GPT2Model;b.GPT2PreTrainedModel;b.GPT2Tokenizer;b.GPTBigCodeForCausalLM;b.GPTBigCodeModel;b.GPTBigCodePreTrainedModel;b.GPTJForCausalLM;b.GPTJModel;b.GPTJPreTrainedModel;b.GPTNeoForCausalLM;b.GPTNeoModel;b.GPTNeoPreTrainedModel;b.GPTNeoXForCausalLM;b.GPTNeoXModel;b.GPTNeoXPreTrainedModel;b.GPTNeoXTokenizer;b.Gemma2ForCausalLM;b.Gemma2Model;b.Gemma2PreTrainedModel;b.Gemma3ForCausalLM;b.Gemma3Model;b.Gemma3PreTrainedModel;b.GemmaForCausalLM;b.GemmaModel;b.GemmaPreTrainedModel;b.GemmaTokenizer;b.GlmForCausalLM;b.GlmModel;b.GlmPreTrainedModel;b.GraniteForCausalLM;b.GraniteModel;b.GranitePreTrainedModel;b.Grok1Tokenizer;b.GroundingDinoForObjectDetection;b.GroundingDinoImageProcessor;b.GroundingDinoPreTrainedModel;b.GroundingDinoProcessor;b.GroupViTModel;b.GroupViTPreTrainedModel;b.HeliumForCausalLM;b.HeliumModel;b.HeliumPreTrainedModel;b.HerbertTokenizer;b.HieraForImageClassification;b.HieraModel;b.HieraPreTrainedModel;b.HubertForCTC;b.HubertForSequenceClassification;b.HubertModel;b.HubertPreTrainedModel;b.IJepaForImageClassification;b.IJepaModel;b.IJepaPreTrainedModel;b.Idefics3ForConditionalGeneration;b.Idefics3ImageProcessor;b.Idefics3PreTrainedModel;b.Idefics3Processor;b.ImageClassificationPipeline;b.ImageFeatureExtractionPipeline;b.ImageFeatureExtractor;b.ImageMattingOutput;b.ImageProcessor;b.ImageSegmentationPipeline;b.ImageToImagePipeline;b.ImageToTextPipeline;b.InterruptableStoppingCriteria;b.JAISLMHeadModel;b.JAISModel;b.JAISPreTrainedModel;b.JinaCLIPImageProcessor;b.JinaCLIPModel;b.JinaCLIPPreTrainedModel;b.JinaCLIPProcessor;b.JinaCLIPTextModel;b.JinaCLIPVisionModel;b.LiteWhisperForConditionalGeneration;b.LlamaForCausalLM;b.LlamaModel;b.LlamaPreTrainedModel;b.LlamaTokenizer;b.LlavaForConditionalGeneration;b.LlavaOnevisionForConditionalGeneration;b.LlavaOnevisionImageProcessor;b.LlavaPreTrainedModel;b.LogitsProcessor;b.LogitsProcessorList;b.LogitsWarper;b.LongT5ForConditionalGeneration;b.LongT5Model;b.LongT5PreTrainedModel;b.M2M100ForConditionalGeneration;b.M2M100Model;b.M2M100PreTrainedModel;b.M2M100Tokenizer;b.MBart50Tokenizer;b.MBartForCausalLM;b.MBartForConditionalGeneration;b.MBartForSequenceClassification;b.MBartModel;b.MBartPreTrainedModel;b.MBartTokenizer;b.MPNetForMaskedLM;b.MPNetForQuestionAnswering;b.MPNetForSequenceClassification;b.MPNetForTokenClassification;b.MPNetModel;b.MPNetPreTrainedModel;b.MPNetTokenizer;b.MT5ForConditionalGeneration;b.MT5Model;b.MT5PreTrainedModel;b.MarianMTModel;b.MarianModel;b.MarianPreTrainedModel;b.MarianTokenizer;b.Mask2FormerImageProcessor;b.MaskFormerFeatureExtractor;b.MaskFormerForInstanceSegmentation;b.MaskFormerImageProcessor;b.MaskFormerModel;b.MaskFormerPreTrainedModel;b.MaskedLMOutput;b.MaxLengthCriteria;b.Metric3DForDepthEstimation;b.Metric3DPreTrainedModel;b.Metric3Dv2ForDepthEstimation;b.Metric3Dv2PreTrainedModel;b.MgpstrForSceneTextRecognition;b.MgpstrModelOutput;b.MgpstrPreTrainedModel;b.MgpstrProcessor;b.MgpstrTokenizer;b.MimiDecoderModel;b.MimiDecoderOutput;b.MimiEncoderModel;b.MimiEncoderOutput;b.MimiModel;b.MimiPreTrainedModel;b.MinLengthLogitsProcessor;b.MinNewTokensLengthLogitsProcessor;b.MistralForCausalLM;b.MistralModel;b.MistralPreTrainedModel;b.MobileBertForMaskedLM;b.MobileBertForQuestionAnswering;b.MobileBertForSequenceClassification;b.MobileBertModel;b.MobileBertPreTrainedModel;b.MobileBertTokenizer;b.MobileLLMForCausalLM;b.MobileLLMModel;b.MobileLLMPreTrainedModel;b.MobileNetV1FeatureExtractor;b.MobileNetV1ForImageClassification;b.MobileNetV1ForSemanticSegmentation;b.MobileNetV1ImageProcessor;b.MobileNetV1Model;b.MobileNetV1PreTrainedModel;b.MobileNetV2FeatureExtractor;b.MobileNetV2ForImageClassification;b.MobileNetV2ForSemanticSegmentation;b.MobileNetV2ImageProcessor;b.MobileNetV2Model;b.MobileNetV2PreTrainedModel;b.MobileNetV3FeatureExtractor;b.MobileNetV3ForImageClassification;b.MobileNetV3ForSemanticSegmentation;b.MobileNetV3ImageProcessor;b.MobileNetV3Model;b.MobileNetV3PreTrainedModel;b.MobileNetV4FeatureExtractor;b.MobileNetV4ForImageClassification;b.MobileNetV4ForSemanticSegmentation;b.MobileNetV4ImageProcessor;b.MobileNetV4Model;b.MobileNetV4PreTrainedModel;b.MobileViTFeatureExtractor;b.MobileViTForImageClassification;b.MobileViTImageProcessor;b.MobileViTModel;b.MobileViTPreTrainedModel;b.MobileViTV2ForImageClassification;b.MobileViTV2Model;b.MobileViTV2PreTrainedModel;b.ModelOutput;b.ModernBertForMaskedLM;b.ModernBertForSequenceClassification;b.ModernBertForTokenClassification;b.ModernBertModel;b.ModernBertPreTrainedModel;b.Moondream1ForConditionalGeneration;b.MoonshineFeatureExtractor;b.MoonshineForConditionalGeneration;b.MoonshineModel;b.MoonshinePreTrainedModel;b.MoonshineProcessor;b.MptForCausalLM;b.MptModel;b.MptPreTrainedModel;b.MultiModalityCausalLM;b.MultiModalityPreTrainedModel;b.MusicgenForCausalLM;b.MusicgenForConditionalGeneration;b.MusicgenModel;b.MusicgenPreTrainedModel;b.NllbTokenizer;b.NoBadWordsLogitsProcessor;b.NoRepeatNGramLogitsProcessor;b.NomicBertModel;b.NomicBertPreTrainedModel;b.NougatImageProcessor;b.NougatTokenizer;b.OPTForCausalLM;b.OPTModel;b.OPTPreTrainedModel;b.ObjectDetectionPipeline;b.Olmo2ForCausalLM;b.Olmo2Model;b.Olmo2PreTrainedModel;b.OlmoForCausalLM;b.OlmoModel;b.OlmoPreTrainedModel;b.OpenELMForCausalLM;b.OpenELMModel;b.OpenELMPreTrainedModel;b.OwlViTFeatureExtractor;b.OwlViTForObjectDetection;b.OwlViTImageProcessor;b.OwlViTModel;b.OwlViTPreTrainedModel;b.OwlViTProcessor;b.Owlv2ForObjectDetection;b.Owlv2ImageProcessor;b.Owlv2Model;b.Owlv2PreTrainedModel;b.PaliGemmaForConditionalGeneration;b.PaliGemmaPreTrainedModel;b.PaliGemmaProcessor;b.PatchTSMixerForPrediction;b.PatchTSMixerModel;b.PatchTSMixerPreTrainedModel;b.PatchTSTForPrediction;b.PatchTSTModel;b.PatchTSTPreTrainedModel;b.Phi3ForCausalLM;b.Phi3Model;b.Phi3PreTrainedModel;b.Phi3VForCausalLM;b.Phi3VImageProcessor;b.Phi3VPreTrainedModel;b.Phi3VProcessor;b.PhiForCausalLM;b.PhiModel;b.PhiPreTrainedModel;b.Pipeline;b.PreTrainedModel;b.PreTrainedTokenizer;b.PretrainedConfig;b.PretrainedMixin;b.Processor;b.PvtForImageClassification;b.PvtImageProcessor;b.PvtModel;b.PvtPreTrainedModel;b.PyAnnoteFeatureExtractor;b.PyAnnoteForAudioFrameClassification;b.PyAnnoteModel;b.PyAnnotePreTrainedModel;b.PyAnnoteProcessor;b.QuestionAnsweringModelOutput;b.QuestionAnsweringPipeline;b.Qwen2ForCausalLM;b.Qwen2Model;b.Qwen2PreTrainedModel;b.Qwen2Tokenizer;b.Qwen2VLForConditionalGeneration;b.Qwen2VLImageProcessor;b.Qwen2VLPreTrainedModel;b.Qwen2VLProcessor;b.Qwen3ForCausalLM;b.Qwen3Model;b.Qwen3PreTrainedModel;b.RFDetrForObjectDetection;b.RFDetrModel;b.RFDetrObjectDetectionOutput;b.RFDetrPreTrainedModel;b.RTDetrForObjectDetection;b.RTDetrImageProcessor;b.RTDetrModel;b.RTDetrObjectDetectionOutput;b.RTDetrPreTrainedModel;b.RTDetrV2ForObjectDetection;b.RTDetrV2Model;b.RTDetrV2ObjectDetectionOutput;b.RTDetrV2PreTrainedModel;b.RawAudio;b.RawImage;b.RawVideo;b.RawVideoFrame;b.RepetitionPenaltyLogitsProcessor;b.ResNetForImageClassification;b.ResNetModel;b.ResNetPreTrainedModel;b.RoFormerForMaskedLM;b.RoFormerForQuestionAnswering;b.RoFormerForSequenceClassification;b.RoFormerForTokenClassification;b.RoFormerModel;b.RoFormerPreTrainedModel;b.RoFormerTokenizer;b.RobertaForMaskedLM;b.RobertaForQuestionAnswering;b.RobertaForSequenceClassification;b.RobertaForTokenClassification;b.RobertaModel;b.RobertaPreTrainedModel;b.RobertaTokenizer;b.SamImageProcessor;b.SamImageSegmentationOutput;b.SamModel;b.SamPreTrainedModel;b.SamProcessor;b.SapiensForDepthEstimation;b.SapiensForNormalEstimation;b.SapiensForSemanticSegmentation;b.SapiensPreTrainedModel;b.SeamlessM4TFeatureExtractor;b.SegformerFeatureExtractor;b.SegformerForImageClassification;b.SegformerForSemanticSegmentation;b.SegformerImageProcessor;b.SegformerModel;b.SegformerPreTrainedModel;b.Seq2SeqLMOutput;b.SequenceClassifierOutput;b.SiglipImageProcessor;b.SiglipModel;b.SiglipPreTrainedModel;b.SiglipTextModel;b.SiglipTokenizer;b.SiglipVisionModel;b.SmolVLMForConditionalGeneration;b.SmolVLMImageProcessor;b.SmolVLMProcessor;b.SnacDecoderModel;b.SnacEncoderModel;b.SnacFeatureExtractor;b.SnacModel;b.SnacPreTrainedModel;b.SpeechT5FeatureExtractor;b.SpeechT5ForSpeechToText;b.SpeechT5ForTextToSpeech;b.SpeechT5HifiGan;b.SpeechT5Model;b.SpeechT5PreTrainedModel;b.SpeechT5Processor;b.SpeechT5Tokenizer;b.SqueezeBertForMaskedLM;b.SqueezeBertForQuestionAnswering;b.SqueezeBertForSequenceClassification;b.SqueezeBertModel;b.SqueezeBertPreTrainedModel;b.SqueezeBertTokenizer;b.StableLmForCausalLM;b.StableLmModel;b.StableLmPreTrainedModel;b.Starcoder2ForCausalLM;b.Starcoder2Model;b.Starcoder2PreTrainedModel;b.StoppingCriteria;b.StoppingCriteriaList;b.StyleTextToSpeech2Model;b.StyleTextToSpeech2PreTrainedModel;b.SummarizationPipeline;b.SuppressTokensAtBeginLogitsProcessor;b.Swin2SRForImageSuperResolution;b.Swin2SRImageProcessor;b.Swin2SRModel;b.Swin2SRPreTrainedModel;b.SwinForImageClassification;b.SwinForSemanticSegmentation;b.SwinModel;b.SwinPreTrainedModel;b.T5ForConditionalGeneration;b.T5Model;b.T5PreTrainedModel;b.T5Tokenizer;b.TableTransformerForObjectDetection;b.TableTransformerModel;b.TableTransformerObjectDetectionOutput;b.TableTransformerPreTrainedModel;b.TemperatureLogitsWarper;var y_=b.Tensor;b.Text2TextGenerationPipeline;b.TextClassificationPipeline;b.TextGenerationPipeline;b.TextStreamer;b.TextToAudioPipeline;b.TokenClassificationPipeline;b.TokenClassifierOutput;b.TokenizerModel;b.TopKLogitsWarper;b.TopPLogitsWarper;b.TrOCRForCausalLM;b.TrOCRPreTrainedModel;b.TranslationPipeline;b.UltravoxModel;b.UltravoxPreTrainedModel;b.UltravoxProcessor;b.UniSpeechForCTC;b.UniSpeechForSequenceClassification;b.UniSpeechModel;b.UniSpeechPreTrainedModel;b.UniSpeechSatForAudioFrameClassification;b.UniSpeechSatForCTC;b.UniSpeechSatForSequenceClassification;b.UniSpeechSatModel;b.UniSpeechSatPreTrainedModel;b.VLChatProcessor;b.VLMImageProcessor;b.ViTFeatureExtractor;b.ViTForImageClassification;b.ViTImageProcessor;b.ViTMAEModel;b.ViTMAEPreTrainedModel;b.ViTMSNForImageClassification;b.ViTMSNModel;b.ViTMSNPreTrainedModel;b.ViTModel;b.ViTPreTrainedModel;b.VisionEncoderDecoderModel;b.VitMatteForImageMatting;b.VitMatteImageProcessor;b.VitMattePreTrainedModel;b.VitPoseForPoseEstimation;b.VitPoseImageProcessor;b.VitPosePreTrainedModel;b.VitsModel;b.VitsModelOutput;b.VitsPreTrainedModel;b.VitsTokenizer;b.Wav2Vec2BertForCTC;b.Wav2Vec2BertForSequenceClassification;b.Wav2Vec2BertModel;b.Wav2Vec2BertPreTrainedModel;b.Wav2Vec2CTCTokenizer;b.Wav2Vec2FeatureExtractor;b.Wav2Vec2ForAudioFrameClassification;b.Wav2Vec2ForCTC;b.Wav2Vec2ForSequenceClassification;b.Wav2Vec2Model;b.Wav2Vec2PreTrainedModel;b.Wav2Vec2Processor;b.Wav2Vec2ProcessorWithLM;b.WavLMForAudioFrameClassification;b.WavLMForCTC;b.WavLMForSequenceClassification;b.WavLMForXVector;b.WavLMModel;b.WavLMPreTrainedModel;b.WeSpeakerFeatureExtractor;b.WeSpeakerResNetModel;b.WeSpeakerResNetPreTrainedModel;b.WhisperFeatureExtractor;b.WhisperForConditionalGeneration;b.WhisperModel;b.WhisperPreTrainedModel;b.WhisperProcessor;b.WhisperTextStreamer;b.WhisperTimeStampLogitsProcessor;b.WhisperTokenizer;b.XLMForQuestionAnswering;b.XLMForSequenceClassification;b.XLMForTokenClassification;b.XLMModel;b.XLMPreTrainedModel;b.XLMRobertaForMaskedLM;b.XLMRobertaForQuestionAnswering;b.XLMRobertaForSequenceClassification;b.XLMRobertaForTokenClassification;b.XLMRobertaModel;b.XLMRobertaPreTrainedModel;b.XLMRobertaTokenizer;b.XLMTokenizer;b.XLMWithLMHeadModel;b.XVectorOutput;b.YolosFeatureExtractor;b.YolosForObjectDetection;b.YolosImageProcessor;b.YolosModel;b.YolosObjectDetectionOutput;b.YolosPreTrainedModel;b.ZeroShotAudioClassificationPipeline;b.ZeroShotClassificationPipeline;b.ZeroShotImageClassificationPipeline;b.ZeroShotObjectDetectionPipeline;b.bankers_round;b.cat;b.cos_sim;b.dot;b.dynamic_time_warping;var lf=b.env;b.full;b.full_like;b.getKeyValueShapes;b.hamming;b.hanning;b.interpolate;b.interpolate_4d;b.interpolate_data;b.is_chinese_char;b.layer_norm;b.load_image;b.load_video;b.log_softmax;b.magnitude;b.matmul;b.max;b.mean;b.mean_pooling;b.medianFilter;b.mel_filter_bank;b.min;b.ones;b.ones_like;b.permute;b.permute_data;var eC=b.pipeline;b.quantize_embeddings;b.rand;b.read_audio;b.rfft;b.round;b.slice;b.softmax;b.spectrogram;b.stack;b.std_mean;b.topk;b.window_function;b.zeros;b.zeros_like;async function tC(){try{if(!await navigator.gpu.requestAdapter())throw new Error("WebGPU is not supported (no adapter found)");return!0}catch(e){self.postMessage({status:"error",data:e.toString()})}}const rC="/models/",sC="onnx-community/whisper-base",nC="onnx-community/silero-vad",cf=16e3,w_=cf/1e3,iC=.3,aC=.1,oC=1500*w_,lC=80,Uv=lC*w_,cC=500*w_,uC=30,dC=512,pC=Math.ceil(Uv/dC);lf.localModelPath=rC;lf.allowRemoteModels=!0;lf.allowLocalModels=!0;lf.backends.onnx.wasm.proxy=!1;const hC=new y_("int64",[cf],[]);let r0=new y_("float32",new Float32Array(2*1*128),[2,1,128]);const Ra=new Float32Array(uC*cf);let In=0,yd=[],Jh=0,af=!1,Gv=!1;class vd{static async Init(r=null){if(this.silero_vad||(this.silero_vad=await ZT.from_pretrained(nC,{config:{model_type:"custom"},dtype:"fp32"})),!this.transcriber){const t="webgpu",i={webgpu:{encoder_model:"fp32",decoder_model_merged:"fp32"},wasm:{encoder_model:"fp32",decoder_model_merged:"q8"}};this.transcriber=await eC("automatic-speech-recognition",sC,{device:t,dtype:i[t]}),await this.transcriber(new Float32Array(cf))}}}de(vd,"silero_vad",null),de(vd,"transcriber",null);async function fC(){await vd.Init(e=>{e.file=e.name+"/"+e.file,self.postMessage(e)}),self.postMessage({status:"ready"}),Gv=!0}async function mC(e){const r=new y_("float32",e,[1,e.length]),{stateN:t,output:i}=await vd.silero_vad({input:r,sr:hC,state:r0});r0=t;const o=i.data[0];return o>iC||af&&o>=aC}async function _C(e){const r=await vd.transcriber(e).then(({text:t})=>t.trim());["","[BLANK_AUDIO]"].includes(r)||(console.log("Transcribed:",r),self.postMessage({status:"from_stt",text:r}))}function Kv(e=0){Ra.fill(0,e),In=e,af=!1,Jh=0}function s0(e){const r=(e==null?void 0:e.length)??0,t=Ra.slice(0,In+Uv),i=yd.reduce((u,d)=>u+d.length,0),o=new Float32Array(i+t.length);let n=0;for(const u of yd)o.set(u,n),n+=u.length;o.set(t,n),_C(o),e&&Ra.set(e,0),Kv(r)}async function gC(e){if(!Gv)return;const r=af,t=await mC(e);if(t&&!r&&console.log("Speech detected, starting recording..."),!r&&!t){yd.length>=pC&&yd.shift(),yd.push(e);return}const i=Ra.length-In;if(e.length>=i){Ra.set(e.subarray(0,i),In),In+=i;const o=e.subarray(i);s0(o);return}else Ra.set(e,In),In+=e.length;if(t){af=!0,Jh=0;return}if(Jh+=e.length,!(Jh{const{type:r,data:t}=e.data;switch(r){case"check":tC();break;case"load_stt":fC();break;case"audio":gC(t);break}});