@huggingface/transformers 3.0.0-alpha.21 → 3.0.0-alpha.22

package/dist/transformers.js

@@ -3788,64 +3788,64 @@ __webpack_require__.r(__webpack_exports__);
 /* harmony export */   TRACE: () => (/* binding */ Sr),
 /* harmony export */   TRACE_FUNC_BEGIN: () => (/* binding */ Le),
 /* harmony export */   TRACE_FUNC_END: () => (/* binding */ Ve),
-/* harmony export */   Tensor: () => (/* binding */ Be),
+/* harmony export */   Tensor: () => (/* binding */ De),
 /* harmony export */   TrainingSession: () => (/* binding */ Ap),
-/* harmony export */   "default": () => (/* binding */ Lx),
-/* harmony export */   env: () => (/* binding */ we),
+/* harmony export */   "default": () => (/* binding */ Hx),
+/* harmony export */   env: () => (/* binding */ _e),
 /* harmony export */   registerBackend: () => (/* binding */ St)
 /* harmony export */ });
 /*!
- * ONNX Runtime Web v1.20.0-dev.20241013-72cc72cc21
+ * ONNX Runtime Web v1.20.0-dev.20241016-2b8fc5529b
  * Copyright (c) Microsoft Corporation. All rights reserved.
  * Licensed under the MIT License.
  */
-var Un=Object.defineProperty;var _p=Object.getOwnPropertyDescriptor;var $p=Object.getOwnPropertyNames;var xp=Object.prototype.hasOwnProperty;var Vn=(e=>typeof require<"u"?require:typeof Proxy<"u"?new Proxy(e,{get:(t,r)=>(typeof require<"u"?require:t)[r]}):e)(function(e){if(typeof require<"u")return require.apply(this,arguments);throw Error('Dynamic require of "'+e+'" is not supported')});var U=(e,t)=>()=>(e&&(t=e(e=0)),t);var Gt=(e,t)=>{for(var r in t)Un(e,r,{get:t[r],enumerable:!0})},Sp=(e,t,r,n)=>{if(t&&typeof t=="object"||typeof t=="function")for(let o of $p(t))!xp.call(e,o)&&o!==r&&Un(e,o,{get:()=>t[o],enumerable:!(n=_p(t,o))||n.enumerable});return e};var br=e=>Sp(Un({},"__esModule",{value:!0}),e);var wr,xt,St,Tp,vr,_r=U(()=>{"use strict";wr=new Map,xt=[],St=(e,t,r)=>{if(t&&typeof t.init=="function"&&typeof t.createInferenceSessionHandler=="function"){let n=wr.get(e);if(n===void 0)wr.set(e,{backend:t,priority:r});else{if(n.priority>r)return;if(n.priority===r&&n.backend!==t)throw new Error(`cannot register backend "${e}" using priority ${r}`)}if(r>=0){let o=xt.indexOf(e);o!==-1&&xt.splice(o,1);for(let i=0;i<xt.length;i++)if(wr.get(xt[i]).priority<=r){xt.splice(i,0,e);return}xt.push(e)}return}throw new TypeError("not a valid backend")},Tp=async e=>{let t=wr.get(e);if(!t)return"backend not found.";if(t.initialized)return t.backend;if(t.aborted)return t.error;{let r=!!t.initPromise;try{return r||(t.initPromise=t.backend.init(e)),await t.initPromise,t.initialized=!0,t.backend}catch(n){return r||(t.error=`${n}`,t.aborted=!0),t.error}finally{delete t.initPromise}}},vr=async e=>{let t=e.executionProviders||[],r=t.map(l=>typeof l=="string"?l:l.name),n=r.length===0?xt:r,o,i=[],a=new Set;for(let l of n){let c=await Tp(l);typeof c=="string"?i.push({name:l,err:c}):(o||(o=c),o===c&&a.add(l))}if(!o)throw new Error(`no available backend found. ERR: ${i.map(l=>`[${l.name}] ${l.err}`).join(", ")}`);for(let{name:l,err:c}of i)r.includes(l)&&console.warn(`removing requested execution provider "${l}" from session options because it is not available: ${c}`);let d=t.filter(l=>a.has(typeof l=="string"?l:l.name));return[o,new Proxy(e,{get:(l,c)=>c==="executionProviders"?d:Reflect.get(l,c)})]}});var Zi=U(()=>{"use strict";_r()});var Qi,Ji=U(()=>{"use strict";Qi="1.20.0-dev.20240928-1bda91fc57"});var ea,We,Nn=U(()=>{"use strict";Ji();ea="warning",We={wasm:{},webgl:{},webgpu:{},versions:{common:Qi},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}`);ea=e}},get logLevel(){return ea}};Object.defineProperty(We,"logLevel",{enumerable:!0})});var we,ta=U(()=>{"use strict";Nn();we=We});var ra,na,oa=U(()=>{"use strict";ra=(e,t)=>{let r=typeof document<"u"?document.createElement("canvas"):new OffscreenCanvas(1,1);r.width=e.dims[3],r.height=e.dims[2];let n=r.getContext("2d");if(n!=null){let o,i;t?.tensorLayout!==void 0&&t.tensorLayout==="NHWC"?(o=e.dims[2],i=e.dims[3]):(o=e.dims[3],i=e.dims[2]);let a=t?.format!==void 0?t.format:"RGB",d=t?.norm,l,c;d===void 0||d.mean===void 0?l=[255,255,255,255]:typeof d.mean=="number"?l=[d.mean,d.mean,d.mean,d.mean]:(l=[d.mean[0],d.mean[1],d.mean[2],0],d.mean[3]!==void 0&&(l[3]=d.mean[3])),d===void 0||d.bias===void 0?c=[0,0,0,0]:typeof d.bias=="number"?c=[d.bias,d.bias,d.bias,d.bias]:(c=[d.bias[0],d.bias[1],d.bias[2],0],d.bias[3]!==void 0&&(c[3]=d.bias[3]));let m=i*o,u=0,h=m,w=m*2,g=-1;a==="RGBA"?(u=0,h=m,w=m*2,g=m*3):a==="RGB"?(u=0,h=m,w=m*2):a==="RBG"&&(u=0,w=m,h=m*2);for(let y=0;y<i;y++)for(let S=0;S<o;S++){let $=(e.data[u++]-c[0])*l[0],_=(e.data[h++]-c[1])*l[1],x=(e.data[w++]-c[2])*l[2],T=g===-1?255:(e.data[g++]-c[3])*l[3];n.fillStyle="rgba("+$+","+_+","+x+","+T+")",n.fillRect(S,y,1,1)}if("toDataURL"in r)return r.toDataURL();throw new Error("toDataURL is not supported")}else throw new Error("Can not access image data")},na=(e,t)=>{let r=typeof document<"u"?document.createElement("canvas").getContext("2d"):new OffscreenCanvas(1,1).getContext("2d"),n;if(r!=null){let o,i,a;t?.tensorLayout!==void 0&&t.tensorLayout==="NHWC"?(o=e.dims[2],i=e.dims[1],a=e.dims[3]):(o=e.dims[3],i=e.dims[2],a=e.dims[1]);let d=t!==void 0&&t.format!==void 0?t.format:"RGB",l=t?.norm,c,m;l===void 0||l.mean===void 0?c=[255,255,255,255]:typeof l.mean=="number"?c=[l.mean,l.mean,l.mean,l.mean]:(c=[l.mean[0],l.mean[1],l.mean[2],255],l.mean[3]!==void 0&&(c[3]=l.mean[3])),l===void 0||l.bias===void 0?m=[0,0,0,0]:typeof l.bias=="number"?m=[l.bias,l.bias,l.bias,l.bias]:(m=[l.bias[0],l.bias[1],l.bias[2],0],l.bias[3]!==void 0&&(m[3]=l.bias[3]));let u=i*o;if(t!==void 0&&(t.format!==void 0&&a===4&&t.format!=="RGBA"||a===3&&t.format!=="RGB"&&t.format!=="BGR"))throw new Error("Tensor format doesn't match input tensor dims");let h=4,w=0,g=1,y=2,S=3,$=0,_=u,x=u*2,T=-1;d==="RGBA"?($=0,_=u,x=u*2,T=u*3):d==="RGB"?($=0,_=u,x=u*2):d==="RBG"&&($=0,x=u,_=u*2),n=r.createImageData(o,i);for(let C=0;C<i*o;w+=h,g+=h,y+=h,S+=h,C++)n.data[w]=(e.data[$++]-m[0])*c[0],n.data[g]=(e.data[_++]-m[1])*c[1],n.data[y]=(e.data[x++]-m[2])*c[2],n.data[S]=T===-1?255:(e.data[T++]-m[3])*c[3]}else throw new Error("Can not access image data");return n}});var Wn,ia,aa,sa,ua,da,la=U(()=>{"use strict";$r();Wn=(e,t)=>{if(e===void 0)throw new Error("Image buffer must be defined");if(t.height===void 0||t.width===void 0)throw new Error("Image height and width must be defined");if(t.tensorLayout==="NHWC")throw new Error("NHWC Tensor layout is not supported yet");let{height:r,width:n}=t,o=t.norm??{mean:255,bias:0},i,a;typeof o.mean=="number"?i=[o.mean,o.mean,o.mean,o.mean]:i=[o.mean[0],o.mean[1],o.mean[2],o.mean[3]??255],typeof o.bias=="number"?a=[o.bias,o.bias,o.bias,o.bias]:a=[o.bias[0],o.bias[1],o.bias[2],o.bias[3]??0];let d=t.format!==void 0?t.format:"RGBA",l=t.tensorFormat!==void 0&&t.tensorFormat!==void 0?t.tensorFormat:"RGB",c=r*n,m=l==="RGBA"?new Float32Array(c*4):new Float32Array(c*3),u=4,h=0,w=1,g=2,y=3,S=0,$=c,_=c*2,x=-1;d==="RGB"&&(u=3,h=0,w=1,g=2,y=-1),l==="RGBA"?x=c*3:l==="RBG"?(S=0,_=c,$=c*2):l==="BGR"&&(_=0,$=c,S=c*2);for(let C=0;C<c;C++,h+=u,g+=u,w+=u,y+=u)m[S++]=(e[h]+a[0])/i[0],m[$++]=(e[w]+a[1])/i[1],m[_++]=(e[g]+a[2])/i[2],x!==-1&&y!==-1&&(m[x++]=(e[y]+a[3])/i[3]);return l==="RGBA"?new De("float32",m,[1,4,r,n]):new De("float32",m,[1,3,r,n])},ia=async(e,t)=>{let r=typeof HTMLImageElement<"u"&&e instanceof HTMLImageElement,n=typeof ImageData<"u"&&e instanceof ImageData,o=typeof ImageBitmap<"u"&&e instanceof ImageBitmap,i=typeof e=="string",a,d=t??{},l=()=>{if(typeof document<"u")return document.createElement("canvas");if(typeof OffscreenCanvas<"u")return new OffscreenCanvas(1,1);throw new Error("Canvas is not supported")},c=m=>typeof HTMLCanvasElement<"u"&&m instanceof HTMLCanvasElement||m instanceof OffscreenCanvas?m.getContext("2d"):null;if(r){let m=l();m.width=e.width,m.height=e.height;let u=c(m);if(u!=null){let h=e.height,w=e.width;if(t!==void 0&&t.resizedHeight!==void 0&&t.resizedWidth!==void 0&&(h=t.resizedHeight,w=t.resizedWidth),t!==void 0){if(d=t,t.tensorFormat!==void 0)throw new Error("Image input config format must be RGBA for HTMLImageElement");d.tensorFormat="RGBA",d.height=h,d.width=w}else d.tensorFormat="RGBA",d.height=h,d.width=w;u.drawImage(e,0,0),a=u.getImageData(0,0,w,h).data}else throw new Error("Can not access image data")}else if(n){let m,u;if(t!==void 0&&t.resizedWidth!==void 0&&t.resizedHeight!==void 0?(m=t.resizedHeight,u=t.resizedWidth):(m=e.height,u=e.width),t!==void 0&&(d=t),d.format="RGBA",d.height=m,d.width=u,t!==void 0){let h=l();h.width=u,h.height=m;let w=c(h);if(w!=null)w.putImageData(e,0,0),a=w.getImageData(0,0,u,m).data;else throw new Error("Can not access image data")}else a=e.data}else if(o){if(t===void 0)throw new Error("Please provide image config with format for Imagebitmap");let m=l();m.width=e.width,m.height=e.height;let u=c(m);if(u!=null){let h=e.height,w=e.width;return u.drawImage(e,0,0,w,h),a=u.getImageData(0,0,w,h).data,d.height=h,d.width=w,Wn(a,d)}else throw new Error("Can not access image data")}else{if(i)return new Promise((m,u)=>{let h=l(),w=c(h);if(!e||!w)return u();let g=new Image;g.crossOrigin="Anonymous",g.src=e,g.onload=()=>{h.width=g.width,h.height=g.height,w.drawImage(g,0,0,h.width,h.height);let y=w.getImageData(0,0,h.width,h.height);d.height=h.height,d.width=h.width,m(Wn(y.data,d))}});throw new Error("Input data provided is not supported - aborted tensor creation")}if(a!==void 0)return Wn(a,d);throw new Error("Input data provided is not supported - aborted tensor creation")},aa=(e,t)=>{let{width:r,height:n,download:o,dispose:i}=t,a=[1,n,r,4];return new De({location:"texture",type:"float32",texture:e,dims:a,download:o,dispose:i})},sa=(e,t)=>{let{dataType:r,dims:n,download:o,dispose:i}=t;return new De({location:"gpu-buffer",type:r??"float32",gpuBuffer:e,dims:n,download:o,dispose:i})},ua=(e,t)=>{let{dataType:r,dims:n,download:o,dispose:i}=t;return new De({location:"ml-tensor",type:r??"float32",mlTensor:e,dims:n,download:o,dispose:i})},da=(e,t,r)=>new De({location:"cpu-pinned",type:e,data:t,dims:r??[t.length]})});var Tt,Ft,ca,pa,ma=U(()=>{"use strict";Tt=new Map([["float32",Float32Array],["uint8",Uint8Array],["int8",Int8Array],["uint16",Uint16Array],["int16",Int16Array],["int32",Int32Array],["bool",Uint8Array],["float64",Float64Array],["uint32",Uint32Array],["int4",Uint8Array],["uint4",Uint8Array]]),Ft=new Map([[Float32Array,"float32"],[Uint8Array,"uint8"],[Int8Array,"int8"],[Uint16Array,"uint16"],[Int16Array,"int16"],[Int32Array,"int32"],[Float64Array,"float64"],[Uint32Array,"uint32"]]),ca=!1,pa=()=>{if(!ca){ca=!0;let e=typeof BigInt64Array<"u"&&BigInt64Array.from,t=typeof BigUint64Array<"u"&&BigUint64Array.from,r=typeof Float16Array<"u"&&Float16Array.from;e&&(Tt.set("int64",BigInt64Array),Ft.set(BigInt64Array,"int64")),t&&(Tt.set("uint64",BigUint64Array),Ft.set(BigUint64Array,"uint64")),r?(Tt.set("float16",Float16Array),Ft.set(Float16Array,"float16")):Tt.set("float16",Uint16Array)}}});var fa,ha,ga=U(()=>{"use strict";$r();fa=e=>{let t=1;for(let r=0;r<e.length;r++){let n=e[r];if(typeof n!="number"||!Number.isSafeInteger(n))throw new TypeError(`dims[${r}] must be an integer, got: ${n}`);if(n<0)throw new RangeError(`dims[${r}] must be a non-negative integer, got: ${n}`);t*=n}return t},ha=(e,t)=>{switch(e.location){case"cpu":return new De(e.type,e.data,t);case"cpu-pinned":return new De({location:"cpu-pinned",data:e.data,type:e.type,dims:t});case"texture":return new De({location:"texture",texture:e.texture,type:e.type,dims:t});case"gpu-buffer":return new De({location:"gpu-buffer",gpuBuffer:e.gpuBuffer,type:e.type,dims:t});case"ml-tensor":return new De({location:"ml-tensor",mlTensor:e.mlTensor,type:e.type,dims:t});default:throw new Error(`tensorReshape: tensor location ${e.location} is not supported`)}}});var De,$r=U(()=>{"use strict";oa();la();ma();ga();De=class{constructor(t,r,n){pa();let o,i;if(typeof t=="object"&&"location"in t)switch(this.dataLocation=t.location,o=t.type,i=t.dims,t.location){case"cpu-pinned":{let d=Tt.get(o);if(!d)throw new TypeError(`unsupported type "${o}" to create tensor from pinned buffer`);if(!(t.data instanceof d))throw new TypeError(`buffer should be of type ${d.name}`);this.cpuData=t.data;break}case"texture":{if(o!=="float32")throw new TypeError(`unsupported type "${o}" to create tensor from texture`);this.gpuTextureData=t.texture,this.downloader=t.download,this.disposer=t.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 "${o}" to create tensor from gpu buffer`);this.gpuBufferData=t.gpuBuffer,this.downloader=t.download,this.disposer=t.dispose;break}case"ml-tensor":{if(o!=="float32"&&o!=="float16"&&o!=="int32"&&o!=="int64"&&o!=="uint32"&&o!=="uint64"&&o!=="int8"&&o!=="uint8"&&o!=="bool")throw new TypeError(`unsupported type "${o}" to create tensor from MLTensor`);this.mlTensorData=t.mlTensor,this.downloader=t.download,this.disposer=t.dispose;break}default:throw new Error(`Tensor constructor: unsupported location '${this.dataLocation}'`)}else{let d,l;if(typeof t=="string")if(o=t,l=n,t==="string"){if(!Array.isArray(r))throw new TypeError("A string tensor's data must be a string array.");d=r}else{let c=Tt.get(t);if(c===void 0)throw new TypeError(`Unsupported tensor type: ${t}.`);if(Array.isArray(r)){if(t==="float16"&&c===Uint16Array||t==="uint4"||t==="int4")throw new TypeError(`Creating a ${t} tensor from number array is not supported. Please use ${c.name} as data.`);t==="uint64"||t==="int64"?d=c.from(r,BigInt):d=c.from(r)}else if(r instanceof c)d=r;else if(r instanceof Uint8ClampedArray)if(t==="uint8")d=Uint8Array.from(r);else throw new TypeError("A Uint8ClampedArray tensor's data must be type of uint8");else throw new TypeError(`A ${o} tensor's data must be type of ${c}`)}else if(l=r,Array.isArray(t)){if(t.length===0)throw new TypeError("Tensor type cannot be inferred from an empty array.");let c=typeof t[0];if(c==="string")o="string",d=t;else if(c==="boolean")o="bool",d=Uint8Array.from(t);else throw new TypeError(`Invalid element type of data array: ${c}.`)}else if(t instanceof Uint8ClampedArray)o="uint8",d=Uint8Array.from(t);else{let c=Ft.get(t.constructor);if(c===void 0)throw new TypeError(`Unsupported type for tensor data: ${t.constructor}.`);o=c,d=t}if(l===void 0)l=[d.length];else if(!Array.isArray(l))throw new TypeError("A tensor's dims must be a number array");i=l,this.cpuData=d,this.dataLocation="cpu"}let a=fa(i);if(this.cpuData&&a!==this.cpuData.length&&!((o==="uint4"||o==="int4")&&Math.ceil(a/2)===this.cpuData.length))throw new Error(`Tensor's size(${a}) does not match data length(${this.cpuData.length}).`);this.type=o,this.dims=i,this.size=a}static async fromImage(t,r){return ia(t,r)}static fromTexture(t,r){return aa(t,r)}static fromGpuBuffer(t,r){return sa(t,r)}static fromMLTensor(t,r){return ua(t,r)}static fromPinnedBuffer(t,r,n){return da(t,r,n)}toDataURL(t){return ra(this,t)}toImageData(t){return na(this,t)}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(t){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,t&&this.disposer&&(this.disposer(),this.disposer=void 0),r}finally{this.isDownloading=!1}}default:throw new Error(`cannot get data from location: ${this.dataLocation}`)}}dispose(){if(this.isDownloading)throw new Error("The current tensor is being downloaded.");this.disposer&&(this.disposer(),this.disposer=void 0),this.cpuData=void 0,this.gpuTextureData=void 0,this.gpuBufferData=void 0,this.mlTensorData=void 0,this.downloader=void 0,this.isDownloading=void 0,this.dataLocation="none"}ensureValid(){if(this.dataLocation==="none")throw new Error("The tensor is disposed.")}reshape(t){if(this.ensureValid(),this.downloader||this.disposer)throw new Error("Cannot reshape a tensor that owns GPU resource.");return ha(this,t)}}});var Be,xr=U(()=>{"use strict";$r();Be=De});var Sr,ya,Le,Ve,Ln=U(()=>{"use strict";Nn();Sr=(e,t)=>{(typeof We.trace>"u"?!We.wasm.trace:!We.trace)||console.timeStamp(`${e}::ORT::${t}`)},ya=(e,t)=>{let r=new Error().stack?.split(/\r\n|\r|\n/g)||[],n=!1;for(let o=0;o<r.length;o++){if(n&&!r[o].includes("TRACE_FUNC")){let i=`FUNC_${e}::${r[o].trim().split(" ")[1]}`;t&&(i+=`::${t}`),Sr("CPU",i);return}r[o].includes("TRACE_FUNC")&&(n=!0)}},Le=e=>{(typeof We.trace>"u"?!We.wasm.trace:!We.trace)||ya("BEGIN",e)},Ve=e=>{(typeof We.trace>"u"?!We.wasm.trace:!We.trace)||ya("END",e)}});var Tr,ba=U(()=>{"use strict";_r();xr();Ln();Tr=class e{constructor(t){this.handler=t}async run(t,r,n){Le();let o={},i={};if(typeof t!="object"||t===null||t instanceof Be||Array.isArray(t))throw new TypeError("'feeds' must be an object that use input names as keys and OnnxValue as corresponding values.");let a=!0;if(typeof r=="object"){if(r===null)throw new TypeError("Unexpected argument[1]: cannot be null.");if(r instanceof Be)throw new TypeError("'fetches' cannot be a Tensor");if(Array.isArray(r)){if(r.length===0)throw new TypeError("'fetches' cannot be an empty array.");a=!1;for(let c of r){if(typeof c!="string")throw new TypeError("'fetches' must be a string array or an object.");if(this.outputNames.indexOf(c)===-1)throw new RangeError(`'fetches' contains invalid output name: ${c}.`);o[c]=null}if(typeof n=="object"&&n!==null)i=n;else if(typeof n<"u")throw new TypeError("'options' must be an object.")}else{let c=!1,m=Object.getOwnPropertyNames(r);for(let u of this.outputNames)if(m.indexOf(u)!==-1){let h=r[u];(h===null||h instanceof Be)&&(c=!0,a=!1,o[u]=h)}if(c){if(typeof n=="object"&&n!==null)i=n;else if(typeof n<"u")throw new TypeError("'options' must be an object.")}else i=r}}else if(typeof r<"u")throw new TypeError("Unexpected argument[1]: must be 'fetches' or 'options'.");for(let c of this.inputNames)if(typeof t[c]>"u")throw new Error(`input '${c}' is missing in 'feeds'.`);if(a)for(let c of this.outputNames)o[c]=null;let d=await this.handler.run(t,o,i),l={};for(let c in d)if(Object.hasOwnProperty.call(d,c)){let m=d[c];m instanceof Be?l[c]=m:l[c]=new Be(m.type,m.data,m.dims)}return Ve(),l}async release(){return this.handler.dispose()}static async create(t,r,n,o){Le();let i,a={};if(typeof t=="string"){if(i=t,typeof r=="object"&&r!==null)a=r;else if(typeof r<"u")throw new TypeError("'options' must be an object.")}else if(t instanceof Uint8Array){if(i=t,typeof r=="object"&&r!==null)a=r;else if(typeof r<"u")throw new TypeError("'options' must be an object.")}else if(t instanceof ArrayBuffer||typeof SharedArrayBuffer<"u"&&t instanceof SharedArrayBuffer){let m=t,u=0,h=t.byteLength;if(typeof r=="object"&&r!==null)a=r;else if(typeof r=="number"){if(u=r,!Number.isSafeInteger(u))throw new RangeError("'byteOffset' must be an integer.");if(u<0||u>=m.byteLength)throw new RangeError(`'byteOffset' is out of range [0, ${m.byteLength}).`);if(h=t.byteLength-u,typeof n=="number"){if(h=n,!Number.isSafeInteger(h))throw new RangeError("'byteLength' must be an integer.");if(h<=0||u+h>m.byteLength)throw new RangeError(`'byteLength' is out of range (0, ${m.byteLength-u}].`);if(typeof o=="object"&&o!==null)a=o;else if(typeof o<"u")throw new TypeError("'options' must be an object.")}else if(typeof n<"u")throw new TypeError("'byteLength' must be a number.")}else if(typeof r<"u")throw new TypeError("'options' must be an object.");i=new Uint8Array(m,u,h)}else throw new TypeError("Unexpected argument[0]: must be 'path' or 'buffer'.");let[d,l]=await vr(a),c=await d.createInferenceSessionHandler(i,l);return Ve(),new e(c)}startProfiling(){this.handler.startProfiling()}endProfiling(){this.handler.endProfiling()}get inputNames(){return this.handler.inputNames}get outputNames(){return this.handler.outputNames}}});var Ip,wa=U(()=>{"use strict";ba();Ip=Tr});var va=U(()=>{"use strict"});var _a=U(()=>{"use strict"});var $a=U(()=>{"use strict"});var xa=U(()=>{"use strict"});var Cp,Ir,Sa=U(()=>{"use strict";_r();xr();Cp="Training backend could not be resolved. Make sure you're using the correct configuration & WebAssembly files.",Ir=class e{constructor(t,r,n){this.handler=t,this.hasOptimizerModel=r,this.hasEvalModel=n}get trainingInputNames(){return this.handler.inputNames}get trainingOutputNames(){return this.handler.outputNames}get evalInputNames(){if(this.hasEvalModel)return this.handler.evalInputNames;throw new Error("This training session has no evalModel loaded.")}get evalOutputNames(){if(this.hasEvalModel)return this.handler.evalOutputNames;throw new Error("This training session has no evalModel loaded.")}static async create(t,r){let n=t.evalModel||"",o=t.optimizerModel||"",i=r||{},[a,d]=await vr(i);if(a.createTrainingSessionHandler){let l=await a.createTrainingSessionHandler(t.checkpointState,t.trainModel,n,o,d);return new e(l,!!t.optimizerModel,!!t.evalModel)}else throw new Error(Cp)}typeNarrowingForRunStep(t,r,n,o,i){let a={},d={};if(typeof n!="object"||n===null||n instanceof Be||Array.isArray(n))throw new TypeError("'feeds' must be an object that use input names as keys and OnnxValue as corresponding values.");let l=!0;if(typeof o=="object"){if(o===null)throw new TypeError("Unexpected argument[1]: cannot be null.");if(o instanceof Be)throw new TypeError("'fetches' cannot be a Tensor");if(Array.isArray(o)){if(o.length===0)throw new TypeError("'fetches' cannot be an empty array.");l=!1;for(let c of o){if(typeof c!="string")throw new TypeError("'fetches' must be a string array or an object.");if(r.indexOf(c)===-1)throw new RangeError(`'fetches' contains invalid output name: ${c}.`);a[c]=null}if(typeof i=="object"&&i!==null)d=i;else if(typeof i<"u")throw new TypeError("'options' must be an object.")}else{let c=!1,m=Object.getOwnPropertyNames(o);for(let u of r)if(m.indexOf(u)!==-1){let h=o[u];(h===null||h instanceof Be)&&(c=!0,l=!1,a[u]=h)}if(c){if(typeof i=="object"&&i!==null)d=i;else if(typeof i<"u")throw new TypeError("'options' must be an object.")}else d=o}}else if(typeof o<"u")throw new TypeError("Unexpected argument[1]: must be 'fetches' or 'options'.");for(let c of t)if(typeof n[c]>"u")throw new Error(`input '${c}' is missing in 'feeds'.`);if(l)for(let c of r)a[c]=null;return[a,d]}convertHandlerReturnTypeToMapOfTensors(t){let r={};for(let n in t)if(Object.hasOwnProperty.call(t,n)){let o=t[n];o instanceof Be?r[n]=o:r[n]=new Be(o.type,o.data,o.dims)}return r}async lazyResetGrad(){await this.handler.lazyResetGrad()}async runTrainStep(t,r,n){let[o,i]=this.typeNarrowingForRunStep(this.trainingInputNames,this.trainingOutputNames,t,r,n),a=await this.handler.runTrainStep(t,o,i);return this.convertHandlerReturnTypeToMapOfTensors(a)}async runOptimizerStep(t){if(this.hasOptimizerModel)await this.handler.runOptimizerStep(t||{});else throw new Error("This TrainingSession has no OptimizerModel loaded.")}async runEvalStep(t,r,n){if(this.hasEvalModel){let[o,i]=this.typeNarrowingForRunStep(this.evalInputNames,this.evalOutputNames,t,r,n),a=await this.handler.runEvalStep(t,o,i);return this.convertHandlerReturnTypeToMapOfTensors(a)}else throw new Error("This TrainingSession has no EvalModel loaded.")}async getParametersSize(t=!0){return this.handler.getParametersSize(t)}async loadParametersBuffer(t,r=!0){let n=await this.getParametersSize(r);if(t.length!==4*n)throw new Error("Size of the buffer passed into loadParametersBuffer must match the number of parameters in the model. Please use getParametersSize method to check.");return this.handler.loadParametersBuffer(t,r)}async getContiguousParameters(t=!0){return this.handler.getContiguousParameters(t)}async release(){return this.handler.dispose()}}});var Ap,Ta=U(()=>{"use strict";Sa();Ap=Ir});var Hn={};Gt(Hn,{InferenceSession:()=>Ip,TRACE:()=>Sr,TRACE_FUNC_BEGIN:()=>Le,TRACE_FUNC_END:()=>Ve,Tensor:()=>Be,TrainingSession:()=>Ap,env:()=>we,registerBackend:()=>St});var Ke=U(()=>{"use strict";Zi();ta();wa();xr();va();_a();Ln();$a();xa();Ta()});var Cr=U(()=>{"use strict"});var Ea={};Gt(Ea,{default:()=>Ep});var Ca,Aa,Ep,ka=U(()=>{"use strict";Gn();gt();qt();Ca="ort-wasm-proxy-worker",Aa=globalThis.self?.name===Ca;Aa&&(self.onmessage=e=>{let{type:t,in:r}=e.data;try{switch(t){case"init-wasm":Ar(r.wasm).then(()=>{Er(r).then(()=>{postMessage({type:t})},n=>{postMessage({type:t,err:n})})},n=>{postMessage({type:t,err:n})});break;case"init-ep":{let{epName:n,env:o}=r;kr(o,n).then(()=>{postMessage({type:t})},i=>{postMessage({type:t,err:i})});break}case"copy-from":{let{buffer:n}=r,o=jt(n);postMessage({type:t,out:o});break}case"create":{let{model:n,options:o}=r;Pr(n,o).then(i=>{postMessage({type:t,out:i})},i=>{postMessage({type:t,err:i})});break}case"release":Or(r),postMessage({type:t});break;case"run":{let{sessionId:n,inputIndices:o,inputs:i,outputIndices:a,options:d}=r;zr(n,o,i,a,new Array(a.length).fill(null),d).then(l=>{l.some(c=>c[3]!=="cpu")?postMessage({type:t,err:"Proxy does not support non-cpu tensor location."}):postMessage({type:t,out:l},Br([...i,...l]))},l=>{postMessage({type:t,err:l})});break}case"end-profiling":Dr(r),postMessage({type:t});break;default:}}catch(n){postMessage({type:t,err:n})}});Ep=Aa?null:e=>new Worker(e??Ut,{type:"module",name:Ca})});var Oa={};Gt(Oa,{default:()=>kp});var Fn,Pa,kp,za=U(()=>{"use strict";Pa=(Fn=import.meta.url,async function(e={}){function t(){return le.buffer!=pe.buffer&&Ce(),pe}function r(){return le.buffer!=pe.buffer&&Ce(),Q}function n(){return le.buffer!=pe.buffer&&Ce(),be}function o(){return le.buffer!=pe.buffer&&Ce(),ne}function i(){return le.buffer!=pe.buffer&&Ce(),oe}function a(){return le.buffer!=pe.buffer&&Ce(),se}function d(){return le.buffer!=pe.buffer&&Ce(),R}function l(){return le.buffer!=pe.buffer&&Ce(),Re}var c,m,u=Object.assign({},e),h=new Promise((s,p)=>{c=s,m=p}),w=typeof window=="object",g=typeof importScripts=="function",y=g&&self.name=="em-pthread";u.mountExternalData=(s,p)=>{s.startsWith("./")&&(s=s.substring(2)),(u.Fb||(u.Fb=new Map)).set(s,p)},u.unmountExternalData=()=>{delete u.Fb};var S=globalThis.SharedArrayBuffer??new WebAssembly.Memory({initial:0,maximum:0,shared:!0}).buffer.constructor;let $=()=>{let s=(f,b,v)=>(...I)=>{let z=et,B=b?.();I=f(...I);let W=b?.();return B!==W&&(f=W,v(B),b=v=null),et!=z?new Promise((H,j)=>{kn={resolve:H,reject:j}}):I},p=f=>async(...b)=>{try{if(u.Eb)throw Error("Session already started");let v=u.Eb={fc:b[0],errors:[]},I=await f(...b);if(u.Eb!==v)throw Error("Session mismatch");u.Gb?.flush();let z=v.errors;if(0<z.length){let B=await Promise.all(z);if(B=B.filter(W=>W),0<B.length)throw Error(B.join(`
-`))}return I}finally{u.Eb=null}};u._OrtCreateSession=s(u._OrtCreateSession,()=>u._OrtCreateSession,f=>u._OrtCreateSession=f),u._OrtRun=p(s(u._OrtRun,()=>u._OrtRun,f=>u._OrtRun=f)),u._OrtRunWithBinding=p(s(u._OrtRunWithBinding,()=>u._OrtRunWithBinding,f=>u._OrtRunWithBinding=f)),u._OrtBindInput=s(u._OrtBindInput,()=>u._OrtBindInput,f=>u._OrtBindInput=f),$=void 0};u.jsepInit=(s,p)=>{if($?.(),s==="webgpu"){[u.Gb,u.Ub,u.Yb,u.Nb,u.Xb,u.jb,u.Zb,u.bc,u.Vb,u.Wb,u.$b]=p;let f=u.Gb;u.jsepRegisterBuffer=(b,v,I,z)=>f.registerBuffer(b,v,I,z),u.jsepGetBuffer=b=>f.getBuffer(b),u.jsepCreateDownloader=(b,v,I)=>f.createDownloader(b,v,I),u.jsepOnReleaseSession=b=>{f.onReleaseSession(b)},u.jsepOnRunStart=b=>f.onRunStart(b),u.cc=(b,v)=>{f.upload(b,v)}}else if(s==="webnn"){[u.Gb,u.ac,u.Ob,u.jsepEnsureTensor,u.dc,u.jsepDownloadTensor]=p,u.jsepReleaseTensorId=u.Ob;let f=u.Gb;u.jsepOnRunStart=b=>f.onRunStart(b),u.jsepRegisterMLContext=(b,v)=>{f.registerMLContext(b,v)},u.jsepOnReleaseSession=b=>{f.onReleaseSession(b)},u.jsepCreateMLTensorDownloader=(b,v)=>f.createMLTensorDownloader(b,v),u.jsepRegisterMLTensor=(b,v,I)=>f.registerMLTensor(b,v,I)}};var _,x,T=Object.assign({},u),C="./this.program",A=(s,p)=>{throw p},P="";(w||g)&&(g?P=self.location.href:typeof document<"u"&&document.currentScript&&(P=document.currentScript.src),Fn&&(P=Fn),P=P.startsWith("blob:")?"":P.substr(0,P.replace(/[?#].*/,"").lastIndexOf("/")+1),g&&(x=s=>{var p=new XMLHttpRequest;return p.open("GET",s,!1),p.responseType="arraybuffer",p.send(null),new Uint8Array(p.response)}),_=(s,p,f)=>{var b=new XMLHttpRequest;b.open("GET",s,!0),b.responseType="arraybuffer",b.onload=()=>{b.status==200||b.status==0&&b.response?p(b.response):f()},b.onerror=f,b.send(null)});var D,N=console.log.bind(console),L=console.error.bind(console),K=N,X=L;if(Object.assign(u,T),T=null,y){let s=function(p){try{var f=p.data,b=f.cmd;if(b==="load"){let v=[];self.onmessage=I=>v.push(I),self.startWorker=()=>{postMessage({cmd:"loaded"});for(let I of v)s(I);self.onmessage=s};for(let I of f.handlers)u[I]&&!u[I].proxy||(u[I]=(...z)=>{postMessage({Mb:"callHandler",oc:I,args:z})},I=="print"&&(K=u[I]),I=="printErr"&&(X=u[I]));le=f.wasmMemory,Ce(),ie(f.wasmModule)}else if(b==="run"){Dn(f.pthread_ptr,0,0,1,0,0),Cn(f.pthread_ptr),ic(),Lo(),ae||(Vi(),ae=!0);try{ac(f.start_routine,f.arg)}catch(v){if(v!="unwind")throw v}}else b==="cancel"?Rt()&&gr(-1):f.target!=="setimmediate"&&(b==="checkMailbox"?ae&&sr():b&&(X(`worker: received unknown command ${b}`),X(f)))}catch(v){throw Ni(),v}};var qh=s,ie,ae=!1;X=function(...p){p=p.join(" "),console.error(p)},self.alert=function(...p){postMessage({Mb:"alert",text:p.join(" "),qc:Rt()})},u.instantiateWasm=(p,f)=>new Promise(b=>{ie=v=>{v=new WebAssembly.Instance(v,Ro()),f(v),b()}}),self.onunhandledrejection=p=>{throw p.reason||p},self.onmessage=s}u.wasmBinary&&(D=u.wasmBinary);var le,Y,me,pe,Q,be,ne,oe,se,R,G,he,Re,$e=!1;function Ce(){var s=le.buffer;u.HEAP8=pe=new Int8Array(s),u.HEAP16=be=new Int16Array(s),u.HEAPU8=Q=new Uint8Array(s),u.HEAPU16=ne=new Uint16Array(s),u.HEAP32=oe=new Int32Array(s),u.HEAPU32=se=new Uint32Array(s),u.HEAPF32=R=new Float32Array(s),u.HEAPF64=Re=new Float64Array(s),u.HEAP64=G=new BigInt64Array(s),u.HEAPU64=he=new BigUint64Array(s)}if(!y){if(!((le=new WebAssembly.Memory({initial:256,maximum:65536,shared:!0})).buffer instanceof S))throw X("requested a shared WebAssembly.Memory but the returned buffer is not a SharedArrayBuffer, indicating that while the browser has SharedArrayBuffer it does not have WebAssembly threads support - you may need to set a flag"),Error("bad memory");Ce()}var bt=[],Ae=[],Me=[],Ue=0,Ot=null,wt=null;function Oo(){if(--Ue==0&&(Ot!==null&&(clearInterval(Ot),Ot=null),wt)){var s=wt;wt=null,s()}}function zt(s){throw X(s="Aborted("+s+")"),$e=!0,me=1,s=new WebAssembly.RuntimeError(s+". Build with -sASSERTIONS for more info."),m(s),s}var hn,zo=s=>s.startsWith("data:application/octet-stream;base64,"),Do=s=>s.startsWith("file://");function Bo(s){if(s==hn&&D)return new Uint8Array(D);if(x)return x(s);throw"both async and sync fetching of the wasm failed"}function Mo(s,p,f){return function(b){if(!D&&(w||g)){if(typeof fetch=="function"&&!Do(b))return fetch(b,{credentials:"same-origin"}).then(v=>{if(!v.ok)throw`failed to load wasm binary file at '${b}'`;return v.arrayBuffer()}).catch(()=>Bo(b));if(_)return new Promise((v,I)=>{_(b,z=>v(new Uint8Array(z)),I)})}return Promise.resolve().then(()=>Bo(b))}(s).then(b=>WebAssembly.instantiate(b,p)).then(f,b=>{X(`failed to asynchronously prepare wasm: ${b}`),zt(b)})}function Ro(){return{a:{O:oc,Aa:nc,b:uc,aa:qo,B:Yo,qa:Xo,Y:Qo,_:Jo,ra:ei,oa:ti,ha:ri,na:ni,L:oi,Z:ii,W:ai,pa:si,X:ui,wa:dc,F:cc,Q:pc,P:fc,E:gc,u:yc,q:bc,G:wc,A:Ic,R:Cc,ua:Ac,ka:Ec,U:kc,ba:Pc,H:Oc,ja:Cn,ta:zc,t:Dc,x:Rc,o:Uc,l:Nc,c:Tn,n:Wc,j:Gc,w:Fc,p:qc,g:jc,s:Kc,m:Yc,e:Xc,k:Zc,i:Qc,h:Jc,d:ep,ea:tp,fa:rp,ga:np,ca:$i,da:xi,T:op,f:ip,D:ap,I:sp,M:up,y:dp,sa:lp,V:cp,v:Ti,z:pp,N:mp,S:fp,za:hp,ya:gp,la:Ai,ma:Ei,$:vn,C:ki,K:Pi,ia:Oi,J:zi,a:le,xa:wn,va:Mi,r:wp}}}var gn={867476:(s,p,f,b,v)=>{if(u===void 0||!u.Fb)return 1;if((s=Oe(s>>>0)).startsWith("./")&&(s=s.substring(2)),!(s=u.Fb.get(s)))return 2;if(b>>>=0,(p>>>=0)+(f>>>=0)>s.byteLength)return 3;try{let I=s.subarray(p,p+f);switch(v){case 0:r().set(I,b>>>0);break;case 1:u.cc(b,I);break;default:return 4}return 0}catch{return 4}},868159:(s,p,f)=>{u.dc(s,r().subarray(p>>>0,p+f>>>0))},868222:()=>u.ac(),868263:s=>{u.Ob(s)},868299:()=>{u.Vb()},868330:()=>{u.Wb()},868359:()=>{u.$b()},868384:s=>u.Ub(s),868417:s=>u.Yb(s),868449:(s,p,f)=>{u.Nb(s,p,f,!0)},868488:(s,p,f)=>{u.Nb(s,p,f)},868521:()=>typeof wasmOffsetConverter<"u",868578:s=>{u.jb("Abs",s,void 0)},868629:s=>{u.jb("Neg",s,void 0)},868680:s=>{u.jb("Floor",s,void 0)},868733:s=>{u.jb("Ceil",s,void 0)},868785:s=>{u.jb("Reciprocal",s,void 0)},868843:s=>{u.jb("Sqrt",s,void 0)},868895:s=>{u.jb("Exp",s,void 0)},868946:s=>{u.jb("Erf",s,void 0)},868997:s=>{u.jb("Sigmoid",s,void 0)},869052:(s,p,f)=>{u.jb("HardSigmoid",s,{alpha:p,beta:f})},869131:s=>{u.jb("Log",s,void 0)},869182:s=>{u.jb("Sin",s,void 0)},869233:s=>{u.jb("Cos",s,void 0)},869284:s=>{u.jb("Tan",s,void 0)},869335:s=>{u.jb("Asin",s,void 0)},869387:s=>{u.jb("Acos",s,void 0)},869439:s=>{u.jb("Atan",s,void 0)},869491:s=>{u.jb("Sinh",s,void 0)},869543:s=>{u.jb("Cosh",s,void 0)},869595:s=>{u.jb("Asinh",s,void 0)},869648:s=>{u.jb("Acosh",s,void 0)},869701:s=>{u.jb("Atanh",s,void 0)},869754:s=>{u.jb("Tanh",s,void 0)},869806:s=>{u.jb("Not",s,void 0)},869857:(s,p,f)=>{u.jb("Clip",s,{min:p,max:f})},869926:s=>{u.jb("Clip",s,void 0)},869978:(s,p)=>{u.jb("Elu",s,{alpha:p})},870036:s=>{u.jb("Gelu",s,void 0)},870088:s=>{u.jb("Relu",s,void 0)},870140:(s,p)=>{u.jb("LeakyRelu",s,{alpha:p})},870204:(s,p)=>{u.jb("ThresholdedRelu",s,{alpha:p})},870274:(s,p)=>{u.jb("Cast",s,{to:p})},870332:s=>{u.jb("Add",s,void 0)},870383:s=>{u.jb("Sub",s,void 0)},870434:s=>{u.jb("Mul",s,void 0)},870485:s=>{u.jb("Div",s,void 0)},870536:s=>{u.jb("Pow",s,void 0)},870587:s=>{u.jb("Equal",s,void 0)},870640:s=>{u.jb("Greater",s,void 0)},870695:s=>{u.jb("GreaterOrEqual",s,void 0)},870757:s=>{u.jb("Less",s,void 0)},870809:s=>{u.jb("LessOrEqual",s,void 0)},870868:(s,p,f,b,v)=>{u.jb("ReduceMean",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,v>>>0)):[]})},871027:(s,p,f,b,v)=>{u.jb("ReduceMax",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,v>>>0)):[]})},871185:(s,p,f,b,v)=>{u.jb("ReduceMin",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,v>>>0)):[]})},871343:(s,p,f,b,v)=>{u.jb("ReduceProd",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,v>>>0)):[]})},871502:(s,p,f,b,v)=>{u.jb("ReduceSum",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,v>>>0)):[]})},871660:(s,p,f,b,v)=>{u.jb("ReduceL1",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,v>>>0)):[]})},871817:(s,p,f,b,v)=>{u.jb("ReduceL2",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,v>>>0)):[]})},871974:(s,p,f,b,v)=>{u.jb("ReduceLogSum",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,v>>>0)):[]})},872135:(s,p,f,b,v)=>{u.jb("ReduceSumSquare",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,v>>>0)):[]})},872299:(s,p,f,b,v)=>{u.jb("ReduceLogSumExp",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,v>>>0)):[]})},872463:s=>{u.jb("Where",s,void 0)},872516:(s,p,f)=>{u.jb("Transpose",s,{perm:p?Array.from(i().subarray(p>>>0,f>>>0)):[]})},872624:(s,p,f,b)=>{u.jb("DepthToSpace",s,{blocksize:p,mode:Oe(f),format:b?"NHWC":"NCHW"})},872757:(s,p,f,b)=>{u.jb("DepthToSpace",s,{blocksize:p,mode:Oe(f),format:b?"NHWC":"NCHW"})},872890:(s,p,f,b,v,I,z,B,W,H,j,ce,ge,O,ue)=>{u.jb("ConvTranspose",s,{format:W?"NHWC":"NCHW",autoPad:p,dilations:[f],group:b,kernelShape:[v],pads:[I,z],strides:[B],wIsConst:()=>!!t()[H>>>0],outputPadding:j?Array.from(i().subarray(j>>>0,ce>>>0)):[],outputShape:ge?Array.from(i().subarray(ge>>>0,O>>>0)):[],activation:Oe(ue)})},873291:(s,p,f,b,v,I,z,B,W,H,j,ce,ge,O)=>{u.jb("ConvTranspose",s,{format:B?"NHWC":"NCHW",autoPad:p,dilations:Array.from(i().subarray(f>>>0,2+(f>>>0)>>>0)),group:b,kernelShape:Array.from(i().subarray(v>>>0,2+(v>>>0)>>>0)),pads:Array.from(i().subarray(I>>>0,4+(I>>>0)>>>0)),strides:Array.from(i().subarray(z>>>0,2+(z>>>0)>>>0)),wIsConst:()=>!!t()[W>>>0],outputPadding:H?Array.from(i().subarray(H>>>0,j>>>0)):[],outputShape:ce?Array.from(i().subarray(ce>>>0,ge>>>0)):[],activation:Oe(O)})},873856:(s,p,f,b,v,I,z,B,W,H,j,ce,ge,O,ue)=>{u.jb("ConvTranspose",s,{format:W?"NHWC":"NCHW",autoPad:p,dilations:[f],group:b,kernelShape:[v],pads:[I,z],strides:[B],wIsConst:()=>!!t()[H>>>0],outputPadding:j?Array.from(i().subarray(j>>>0,ce>>>0)):[],outputShape:ge?Array.from(i().subarray(ge>>>0,O>>>0)):[],activation:Oe(ue)})},874257:(s,p,f,b,v,I,z,B,W,H,j,ce,ge,O)=>{u.jb("ConvTranspose",s,{format:B?"NHWC":"NCHW",autoPad:p,dilations:Array.from(i().subarray(f>>>0,2+(f>>>0)>>>0)),group:b,kernelShape:Array.from(i().subarray(v>>>0,2+(v>>>0)>>>0)),pads:Array.from(i().subarray(I>>>0,4+(I>>>0)>>>0)),strides:Array.from(i().subarray(z>>>0,2+(z>>>0)>>>0)),wIsConst:()=>!!t()[W>>>0],outputPadding:H?Array.from(i().subarray(H>>>0,j>>>0)):[],outputShape:ce?Array.from(i().subarray(ce>>>0,ge>>>0)):[],activation:Oe(O)})},874822:(s,p)=>{u.jb("GlobalAveragePool",s,{format:p?"NHWC":"NCHW"})},874913:(s,p,f,b,v,I,z,B,W,H,j,ce,ge,O)=>{u.jb("AveragePool",s,{format:O?"NHWC":"NCHW",auto_pad:p,ceil_mode:f,count_include_pad:b,storage_order:v,dilations:I?Array.from(i().subarray(I>>>0,z>>>0)):[],kernel_shape:B?Array.from(i().subarray(B>>>0,W>>>0)):[],pads:H?Array.from(i().subarray(H>>>0,j>>>0)):[],strides:ce?Array.from(i().subarray(ce>>>0,ge>>>0)):[]})},875328:(s,p)=>{u.jb("GlobalAveragePool",s,{format:p?"NHWC":"NCHW"})},875419:(s,p,f,b,v,I,z,B,W,H,j,ce,ge,O)=>{u.jb("AveragePool",s,{format:O?"NHWC":"NCHW",auto_pad:p,ceil_mode:f,count_include_pad:b,storage_order:v,dilations:I?Array.from(i().subarray(I>>>0,z>>>0)):[],kernel_shape:B?Array.from(i().subarray(B>>>0,W>>>0)):[],pads:H?Array.from(i().subarray(H>>>0,j>>>0)):[],strides:ce?Array.from(i().subarray(ce>>>0,ge>>>0)):[]})},875834:(s,p)=>{u.jb("GlobalMaxPool",s,{format:p?"NHWC":"NCHW"})},875921:(s,p,f,b,v,I,z,B,W,H,j,ce,ge,O)=>{u.jb("MaxPool",s,{format:O?"NHWC":"NCHW",auto_pad:p,ceil_mode:f,count_include_pad:b,storage_order:v,dilations:I?Array.from(i().subarray(I>>>0,z>>>0)):[],kernel_shape:B?Array.from(i().subarray(B>>>0,W>>>0)):[],pads:H?Array.from(i().subarray(H>>>0,j>>>0)):[],strides:ce?Array.from(i().subarray(ce>>>0,ge>>>0)):[]})},876332:(s,p)=>{u.jb("GlobalMaxPool",s,{format:p?"NHWC":"NCHW"})},876419:(s,p,f,b,v,I,z,B,W,H,j,ce,ge,O)=>{u.jb("MaxPool",s,{format:O?"NHWC":"NCHW",auto_pad:p,ceil_mode:f,count_include_pad:b,storage_order:v,dilations:I?Array.from(i().subarray(I>>>0,z>>>0)):[],kernel_shape:B?Array.from(i().subarray(B>>>0,W>>>0)):[],pads:H?Array.from(i().subarray(H>>>0,j>>>0)):[],strides:ce?Array.from(i().subarray(ce>>>0,ge>>>0)):[]})},876830:(s,p,f,b,v)=>{u.jb("Gemm",s,{alpha:p,beta:f,transA:b,transB:v})},876934:s=>{u.jb("MatMul",s,void 0)},876988:(s,p,f,b)=>{u.jb("ArgMax",s,{keepDims:!!p,selectLastIndex:!!f,axis:b})},877096:(s,p,f,b)=>{u.jb("ArgMin",s,{keepDims:!!p,selectLastIndex:!!f,axis:b})},877204:(s,p)=>{u.jb("Softmax",s,{axis:p})},877267:(s,p)=>{u.jb("Concat",s,{axis:p})},877327:(s,p,f,b,v)=>{u.jb("Split",s,{axis:p,numOutputs:f,splitSizes:b?Array.from(i().subarray(b>>>0,v>>>0)):[]})},877467:s=>{u.jb("Expand",s,void 0)},877521:(s,p)=>{u.jb("Gather",s,{axis:Number(p)})},877592:(s,p)=>{u.jb("GatherElements",s,{axis:Number(p)})},877671:(s,p,f,b,v,I,z,B,W,H,j)=>{u.jb("Resize",s,{antialias:p,axes:f?Array.from(i().subarray(f>>>0,b>>>0)):[],coordinateTransformMode:Oe(v),cubicCoeffA:I,excludeOutside:z,extrapolationValue:B,keepAspectRatioPolicy:Oe(W),mode:Oe(H),nearestMode:Oe(j)})},878017:(s,p,f,b,v,I,z)=>{u.jb("Slice",s,{starts:p?Array.from(i().subarray(p>>>0,f>>>0)):[],ends:b?Array.from(i().subarray(b>>>0,v>>>0)):[],axes:I?Array.from(i().subarray(I>>>0,z>>>0)):[]})},878233:s=>{u.jb("Tile",s,void 0)},878285:(s,p,f)=>{u.jb("InstanceNormalization",s,{epsilon:p,format:f?"NHWC":"NCHW"})},878399:(s,p,f)=>{u.jb("InstanceNormalization",s,{epsilon:p,format:f?"NHWC":"NCHW"})},878513:s=>{u.jb("Range",s,void 0)},878566:(s,p)=>{u.jb("Einsum",s,{equation:Oe(p)})},878647:(s,p,f,b,v)=>{u.jb("Pad",s,{mode:p,value:f,pads:b?Array.from(i().subarray(b>>>0,v>>>0)):[]})},878774:(s,p,f,b,v,I)=>{u.jb("BatchNormalization",s,{epsilon:p,momentum:f,spatial:!!v,trainingMode:!!b,format:I?"NHWC":"NCHW"})},878943:(s,p,f,b,v,I)=>{u.jb("BatchNormalization",s,{epsilon:p,momentum:f,spatial:!!v,trainingMode:!!b,format:I?"NHWC":"NCHW"})},879112:(s,p,f)=>{u.jb("CumSum",s,{exclusive:Number(p),reverse:Number(f)})},879209:(s,p,f)=>{u.jb("DequantizeLinear",s,{axis:p,blockSize:f})},879299:(s,p,f,b,v,I,z,B,W)=>{u.jb("Attention",s,{numHeads:p,isUnidirectional:f,maskFilterValue:b,scale:v,doRotary:I,qkvHiddenSizes:z?Array.from(i().subarray(Number(B)>>>0,Number(B)+z>>>0)):[],pastPresentShareBuffer:!!W})},879571:s=>{u.jb("BiasAdd",s,void 0)},879626:s=>{u.jb("BiasSplitGelu",s,void 0)},879687:s=>{u.jb("FastGelu",s,void 0)},879743:(s,p,f,b,v,I,z,B,W,H,j,ce,ge,O,ue,Se)=>{u.jb("Conv",s,{format:ce?"NHWC":"NCHW",auto_pad:p,dilations:f?Array.from(i().subarray(f>>>0,b>>>0)):[],group:v,kernel_shape:I?Array.from(i().subarray(I>>>0,z>>>0)):[],pads:B?Array.from(i().subarray(B>>>0,W>>>0)):[],strides:H?Array.from(i().subarray(H>>>0,j>>>0)):[],w_is_const:()=>!!t()[ge>>>0],activation:Oe(O),activation_params:ue?Array.from(d().subarray(ue>>>0,Se>>>0)):[]})},880239:s=>{u.jb("Gelu",s,void 0)},880291:(s,p,f,b)=>{u.jb("GroupQueryAttention",s,{numHeads:p,kvNumHeads:f,scale:b})},880404:(s,p,f,b)=>{u.jb("LayerNormalization",s,{axis:p,epsilon:f,simplified:!!b})},880515:(s,p,f,b)=>{u.jb("LayerNormalization",s,{axis:p,epsilon:f,simplified:!!b})},880626:(s,p,f,b,v,I)=>{u.jb("MatMulNBits",s,{k:p,n:f,accuracyLevel:b,bits:v,blockSize:I})},880753:(s,p,f,b,v,I)=>{u.jb("MultiHeadAttention",s,{numHeads:p,isUnidirectional:f,maskFilterValue:b,scale:v,doRotary:I})},880912:(s,p)=>{u.jb("QuickGelu",s,{alpha:p})},880976:(s,p,f,b,v)=>{u.jb("RotaryEmbedding",s,{interleaved:!!p,numHeads:f,rotaryEmbeddingDim:b,scale:v})},881115:(s,p,f)=>{u.jb("SkipLayerNormalization",s,{epsilon:p,simplified:!!f})},881217:(s,p,f)=>{u.jb("SkipLayerNormalization",s,{epsilon:p,simplified:!!f})},881319:(s,p,f,b)=>{u.jb("GatherBlockQuantized",s,{gatherAxis:p,quantizeAxis:f,blockSize:b})},881440:s=>{u.Zb(s)},881474:(s,p)=>u.bc(s,p,u.Eb.fc,u.Eb.errors)};function nc(s,p,f){return yi(async()=>{await u.Xb(s,p,f)})}function oc(){return typeof wasmOffsetConverter<"u"}function yn(s){this.name="ExitStatus",this.message=`Program terminated with exit(${s})`,this.status=s}var bn=s=>{s.terminate(),s.onmessage=()=>{}},Uo=s=>{pt.length==0&&(Go(),Ho(pt[0]));var p=pt.pop();if(!p)return 6;_t.push(p),Qe[s.Ab]=p,p.Ab=s.Ab;var f={cmd:"run",start_routine:s.hc,arg:s.Qb,pthread_ptr:s.Ab};return p.postMessage(f,s.mc),0},vt=0,xe=(s,p,...f)=>{for(var b=2*f.length,v=Rn(),I=Mn(8*b),z=I>>>3,B=0;B<f.length;B++){var W=f[B];typeof W=="bigint"?(G[z+2*B]=1n,G[z+2*B+1]=W):(G[z+2*B]=0n,l()[z+2*B+1>>>0]=W)}return s=Wi(s,0,b,I,p),yr(v),s};function wn(s){if(y)return xe(0,1,s);if(me=s,!(0<vt)){for(var p of _t)bn(p);for(p of pt)bn(p);pt=[],_t=[],Qe=[],$e=!0}A(s,new yn(s))}function Vo(s){if(y)return xe(1,0,s);vn(s)}var vn=s=>{if(me=s,y)throw Vo(s),"unwind";wn(s)},pt=[],_t=[],No=[],Qe={},Wo=s=>{var p=s.Ab;delete Qe[p],pt.push(s),_t.splice(_t.indexOf(s),1),s.Ab=0,Bn(p)};function Lo(){No.forEach(s=>s())}var Ho=s=>new Promise(p=>{s.onmessage=v=>{var I=(v=v.data).cmd;if(v.targetThread&&v.targetThread!=Rt()){var z=Qe[v.targetThread];z?z.postMessage(v,v.transferList):X(`Internal error! Worker sent a message "${I}" to target pthread ${v.targetThread}, but that thread no longer exists!`)}else I==="checkMailbox"?sr():I==="spawnThread"?Uo(v):I==="cleanupThread"?Wo(Qe[v.thread]):I==="killThread"?(v=v.thread,I=Qe[v],delete Qe[v],bn(I),Bn(v),_t.splice(_t.indexOf(I),1),I.Ab=0):I==="cancelThread"?Qe[v.thread].postMessage({cmd:"cancel"}):I==="loaded"?(s.loaded=!0,p(s)):I==="alert"?alert(`Thread ${v.threadId}: ${v.text}`):v.target==="setimmediate"?s.postMessage(v):I==="callHandler"?u[v.handler](...v.args):I&&X(`worker sent an unknown command ${I}`)},s.onerror=v=>{throw X(`worker sent an error! ${v.filename}:${v.lineno}: ${v.message}`),v};var f,b=[];for(f of[])u.hasOwnProperty(f)&&b.push(f);s.postMessage({cmd:"load",handlers:b,wasmMemory:le,wasmModule:Y})});function Go(){var s=new Worker(new URL(import.meta.url),{type:"module",workerData:"em-pthread",name:"em-pthread"});pt.push(s)}var ar=s=>{for(;0<s.length;)s.shift()(u)},ic=()=>{var s=Rt(),p=a()[s+52>>>2>>>0];s=a()[s+56>>>2>>>0],Hi(p,p-s),yr(p)},ac=(s,p)=>{vt=0,s=Gi(s,p),0<vt?me=s:gr(s)};class sc{constructor(p){this.Jb=p-24}}function uc(s,p,f){var b=new sc(s>>>=0);throw p>>>=0,f>>>=0,a()[b.Jb+16>>>2>>>0]=0,a()[b.Jb+4>>>2>>>0]=p,a()[b.Jb+8>>>2>>>0]=f,s}function Fo(s,p,f,b){return y?xe(2,1,s,p,f,b):qo(s,p,f,b)}function qo(s,p,f,b){if(s>>>=0,p>>>=0,f>>>=0,b>>>=0,S===void 0)return X("Current environment does not support SharedArrayBuffer, pthreads are not available!"),6;var v=[];return y&&v.length===0?Fo(s,p,f,b):(s={hc:f,Ab:s,Qb:b,mc:v},y?(s.Mb="spawnThread",postMessage(s,v),0):Uo(s))}var jo=typeof TextDecoder<"u"?new TextDecoder("utf8"):void 0,Ko=(s,p,f)=>{var b=(p>>>=0)+f;for(f=p;s[f]&&!(f>=b);)++f;if(16<f-p&&s.buffer&&jo)return jo.decode(s.buffer instanceof S?s.slice(p,f):s.subarray(p,f));for(b="";p<f;){var v=s[p++];if(128&v){var I=63&s[p++];if((224&v)==192)b+=String.fromCharCode((31&v)<<6|I);else{var z=63&s[p++];65536>(v=(240&v)==224?(15&v)<<12|I<<6|z:(7&v)<<18|I<<12|z<<6|63&s[p++])?b+=String.fromCharCode(v):(v-=65536,b+=String.fromCharCode(55296|v>>10,56320|1023&v))}}else b+=String.fromCharCode(v)}return b},Oe=(s,p)=>(s>>>=0)?Ko(r(),s,p):"";function Yo(s,p,f){return y?xe(3,1,s,p,f):0}function Xo(s,p){if(y)return xe(4,1,s,p)}var _n=s=>{for(var p=0,f=0;f<s.length;++f){var b=s.charCodeAt(f);127>=b?p++:2047>=b?p+=2:55296<=b&&57343>=b?(p+=4,++f):p+=3}return p},Zo=(s,p,f,b)=>{if(!(0<b))return 0;var v=f>>>=0;b=f+b-1;for(var I=0;I<s.length;++I){var z=s.charCodeAt(I);if(55296<=z&&57343>=z&&(z=65536+((1023&z)<<10)|1023&s.charCodeAt(++I)),127>=z){if(f>=b)break;p[f++>>>0]=z}else{if(2047>=z){if(f+1>=b)break;p[f++>>>0]=192|z>>6}else{if(65535>=z){if(f+2>=b)break;p[f++>>>0]=224|z>>12}else{if(f+3>=b)break;p[f++>>>0]=240|z>>18,p[f++>>>0]=128|z>>12&63}p[f++>>>0]=128|z>>6&63}p[f++>>>0]=128|63&z}}return p[f>>>0]=0,f-v},Dt=(s,p,f)=>Zo(s,r(),p,f);function Qo(s,p){if(y)return xe(5,1,s,p)}function Jo(s,p,f){if(y)return xe(6,1,s,p,f)}function ei(s,p,f){return y?xe(7,1,s,p,f):0}function ti(s,p){if(y)return xe(8,1,s,p)}function ri(s,p,f){if(y)return xe(9,1,s,p,f)}function ni(s,p,f,b){if(y)return xe(10,1,s,p,f,b)}function oi(s,p,f,b){if(y)return xe(11,1,s,p,f,b)}function ii(s,p,f,b){if(y)return xe(12,1,s,p,f,b)}function ai(s){if(y)return xe(13,1,s)}function si(s,p){if(y)return xe(14,1,s,p)}function ui(s,p,f){if(y)return xe(15,1,s,p,f)}var di,mt,dc=()=>{zt("")},Je=s=>{for(var p="";r()[s>>>0];)p+=di[r()[s++>>>0]];return p},$n={},xn={},lc={};function ut(s,p,f={}){if(!("argPackAdvance"in p))throw new TypeError("registerType registeredInstance requires argPackAdvance");return function(b,v,I={}){var z=v.name;if(!b)throw new mt(`type "${z}" must have a positive integer typeid pointer`);if(xn.hasOwnProperty(b)){if(I.Sb)return;throw new mt(`Cannot register type '${z}' twice`)}xn[b]=v,delete lc[b],$n.hasOwnProperty(b)&&(v=$n[b],delete $n[b],v.forEach(B=>B()))}(s,p,f)}var li=(s,p,f)=>{switch(p){case 1:return f?b=>t()[b>>>0]:b=>r()[b>>>0];case 2:return f?b=>n()[b>>>1>>>0]:b=>o()[b>>>1>>>0];case 4:return f?b=>i()[b>>>2>>>0]:b=>a()[b>>>2>>>0];case 8:return f?b=>G[b>>>3]:b=>he[b>>>3];default:throw new TypeError(`invalid integer width (${p}): ${s}`)}};function cc(s,p,f){f>>>=0,ut(s>>>=0,{name:p=Je(p>>>0),fromWireType:b=>b,toWireType:function(b,v){if(typeof v!="bigint"&&typeof v!="number")throw v=v===null?"null":(b=typeof v)=="object"||b==="array"||b==="function"?v.toString():""+v,new TypeError(`Cannot convert "${v}" to ${this.name}`);return typeof v=="number"&&(v=BigInt(v)),v},argPackAdvance:ft,readValueFromPointer:li(p,f,p.indexOf("u")==-1),Db:null})}var ft=8;function pc(s,p,f,b){ut(s>>>=0,{name:p=Je(p>>>0),fromWireType:function(v){return!!v},toWireType:function(v,I){return I?f:b},argPackAdvance:ft,readValueFromPointer:function(v){return this.fromWireType(r()[v>>>0])},Db:null})}var Sn=[],dt=[];function Tn(s){9<(s>>>=0)&&--dt[s+1]==0&&(dt[s]=void 0,Sn.push(s))}var qe=s=>{if(!s)throw new mt("Cannot use deleted val. handle = "+s);return dt[s]},je=s=>{switch(s){case void 0:return 2;case null:return 4;case!0:return 6;case!1:return 8;default:let p=Sn.pop()||dt.length;return dt[p]=s,dt[p+1]=1,p}};function In(s){return this.fromWireType(a()[s>>>2>>>0])}var mc={name:"emscripten::val",fromWireType:s=>{var p=qe(s);return Tn(s),p},toWireType:(s,p)=>je(p),argPackAdvance:ft,readValueFromPointer:In,Db:null};function fc(s){return ut(s>>>0,mc)}var hc=(s,p)=>{switch(p){case 4:return function(f){return this.fromWireType(d()[f>>>2>>>0])};case 8:return function(f){return this.fromWireType(l()[f>>>3>>>0])};default:throw new TypeError(`invalid float width (${p}): ${s}`)}};function gc(s,p,f){f>>>=0,ut(s>>>=0,{name:p=Je(p>>>0),fromWireType:b=>b,toWireType:(b,v)=>v,argPackAdvance:ft,readValueFromPointer:hc(p,f),Db:null})}function yc(s,p,f,b,v){if(s>>>=0,f>>>=0,p=Je(p>>>0),v===-1&&(v=4294967295),v=B=>B,b===0){var I=32-8*f;v=B=>B<<I>>>I}var z=p.includes("unsigned")?function(B,W){return W>>>0}:function(B,W){return W};ut(s,{name:p,fromWireType:v,toWireType:z,argPackAdvance:ft,readValueFromPointer:li(p,f,b!==0),Db:null})}function bc(s,p,f){function b(I){var z=a()[I>>>2>>>0];return I=a()[I+4>>>2>>>0],new v(t().buffer,I,z)}var v=[Int8Array,Uint8Array,Int16Array,Uint16Array,Int32Array,Uint32Array,Float32Array,Float64Array,BigInt64Array,BigUint64Array][p];ut(s>>>=0,{name:f=Je(f>>>0),fromWireType:b,argPackAdvance:ft,readValueFromPointer:b},{Sb:!0})}function wc(s,p){s>>>=0;var f=(p=Je(p>>>0))==="std::string";ut(s,{name:p,fromWireType:function(b){var v=a()[b>>>2>>>0],I=b+4;if(f)for(var z=I,B=0;B<=v;++B){var W=I+B;if(B==v||r()[W>>>0]==0){if(z=Oe(z,W-z),H===void 0)var H=z;else H+=String.fromCharCode(0),H+=z;z=W+1}}else{for(H=Array(v),B=0;B<v;++B)H[B]=String.fromCharCode(r()[I+B>>>0]);H=H.join("")}return tt(b),H},toWireType:function(b,v){v instanceof ArrayBuffer&&(v=new Uint8Array(v));var I=typeof v=="string";if(!(I||v instanceof Uint8Array||v instanceof Uint8ClampedArray||v instanceof Int8Array))throw new mt("Cannot pass non-string to std::string");var z=f&&I?_n(v):v.length,B=hr(4+z+1),W=B+4;if(a()[B>>>2>>>0]=z,f&&I)Dt(v,W,z+1);else if(I)for(I=0;I<z;++I){var H=v.charCodeAt(I);if(255<H)throw tt(W),new mt("String has UTF-16 code units that do not fit in 8 bits");r()[W+I>>>0]=H}else for(I=0;I<z;++I)r()[W+I>>>0]=v[I];return b!==null&&b.push(tt,B),B},argPackAdvance:ft,readValueFromPointer:In,Db(b){tt(b)}})}var ci=typeof TextDecoder<"u"?new TextDecoder("utf-16le"):void 0,vc=(s,p)=>{for(var f=s>>1,b=f+p/2;!(f>=b)&&o()[f>>>0];)++f;if(32<(f<<=1)-s&&ci)return ci.decode(r().slice(s,f));for(f="",b=0;!(b>=p/2);++b){var v=n()[s+2*b>>>1>>>0];if(v==0)break;f+=String.fromCharCode(v)}return f},_c=(s,p,f)=>{if(f??=2147483647,2>f)return 0;var b=p;f=(f-=2)<2*s.length?f/2:s.length;for(var v=0;v<f;++v){var I=s.charCodeAt(v);n()[p>>>1>>>0]=I,p+=2}return n()[p>>>1>>>0]=0,p-b},$c=s=>2*s.length,xc=(s,p)=>{for(var f=0,b="";!(f>=p/4);){var v=i()[s+4*f>>>2>>>0];if(v==0)break;++f,65536<=v?(v-=65536,b+=String.fromCharCode(55296|v>>10,56320|1023&v)):b+=String.fromCharCode(v)}return b},Sc=(s,p,f)=>{if(p>>>=0,f??=2147483647,4>f)return 0;var b=p;f=b+f-4;for(var v=0;v<s.length;++v){var I=s.charCodeAt(v);if(55296<=I&&57343>=I&&(I=65536+((1023&I)<<10)|1023&s.charCodeAt(++v)),i()[p>>>2>>>0]=I,(p+=4)+4>f)break}return i()[p>>>2>>>0]=0,p-b},Tc=s=>{for(var p=0,f=0;f<s.length;++f){var b=s.charCodeAt(f);55296<=b&&57343>=b&&++f,p+=4}return p};function Ic(s,p,f){if(s>>>=0,p>>>=0,f=Je(f>>>=0),p===2)var b=vc,v=_c,I=$c,z=B=>o()[B>>>1>>>0];else p===4&&(b=xc,v=Sc,I=Tc,z=B=>a()[B>>>2>>>0]);ut(s,{name:f,fromWireType:B=>{for(var W,H=a()[B>>>2>>>0],j=B+4,ce=0;ce<=H;++ce){var ge=B+4+ce*p;ce!=H&&z(ge)!=0||(j=b(j,ge-j),W===void 0?W=j:(W+=String.fromCharCode(0),W+=j),j=ge+p)}return tt(B),W},toWireType:(B,W)=>{if(typeof W!="string")throw new mt(`Cannot pass non-string to C++ string type ${f}`);var H=I(W),j=hr(4+H+p);return a()[j>>>2>>>0]=H/p,v(W,j+4,H+p),B!==null&&B.push(tt,j),j},argPackAdvance:ft,readValueFromPointer:In,Db(B){tt(B)}})}function Cc(s,p){ut(s>>>=0,{Tb:!0,name:p=Je(p>>>0),argPackAdvance:0,fromWireType:()=>{},toWireType:()=>{}})}var Ac=()=>1;function Ec(s){Dn(s>>>0,!g,1,!w,131072,!1),Lo()}var pi=s=>{if(!$e)try{if(s(),!(0<vt))try{y?gr(me):vn(me)}catch(p){p instanceof yn||p=="unwind"||A(1,p)}}catch(p){p instanceof yn||p=="unwind"||A(1,p)}};function Cn(s){s>>>=0,typeof Atomics.nc=="function"&&(Atomics.nc(i(),s>>>2,s).value.then(sr),s+=128,Atomics.store(i(),s>>>2,1))}var sr=()=>{var s=Rt();s&&(Cn(s),pi(Li))};function kc(s,p){(s>>>=0)==p>>>0?setTimeout(sr):y?postMessage({targetThread:s,cmd:"checkMailbox"}):(s=Qe[s])&&s.postMessage({cmd:"checkMailbox"})}var An=[];function Pc(s,p,f,b,v){for(p>>>=0,b/=2,An.length=b,f=v>>>0>>>3,v=0;v<b;v++)An[v]=G[f+2*v]?G[f+2*v+1]:l()[f+2*v+1>>>0];return(p?gn[p]:vp[s])(...An)}function Oc(s){s>>>=0,y?postMessage({cmd:"cleanupThread",thread:s}):Wo(Qe[s])}function zc(s){}var En=(s,p)=>{var f=xn[s];if(f===void 0)throw s=Ui(s),f=Je(s),tt(s),new mt(`${p} has unknown type ${f}`);return f},mi=(s,p,f)=>{var b=[];return s=s.toWireType(b,f),b.length&&(a()[p>>>2>>>0]=je(b)),s};function Dc(s,p,f){return p>>>=0,f>>>=0,s=qe(s>>>0),p=En(p,"emval::as"),mi(p,f,s)}var ur=s=>{try{s()}catch(p){zt(p)}},ht=0,et=null,fi=0,dr=[],hi={},gi={},Bc=0,kn=null,Mc=[];function yi(s){return function(p){if(!$e){if(ht===0){var f=!1,b=!1;p((v=0)=>{if(!$e&&(fi=v,f=!0,b)){ht=2,ur(()=>ji(et)),typeof Browser<"u"&&Browser.Kb.Rb&&Browser.Kb.resume(),v=!1;try{var I=function(){var W=i()[et+8>>>2>>>0];return W=q[gi[W]],--vt,W()}()}catch(W){I=W,v=!0}var z=!1;if(!et){var B=kn;B&&(kn=null,(v?B.reject:B.resolve)(I),z=!0)}if(v&&!z)throw I}}),b=!0,f||(ht=1,et=function(){var v=hr(65548),I=v+12;a()[v>>>2>>>0]=I,a()[v+4>>>2>>>0]=I+65536,I=dr[0];var z=hi[I];return z===void 0&&(z=Bc++,hi[I]=z,gi[z]=I),I=z,i()[v+8>>>2>>>0]=I,v}(),typeof Browser<"u"&&Browser.Kb.Rb&&Browser.Kb.pause(),ur(()=>Fi(et)))}else ht===2?(ht=0,ur(Ki),tt(et),et=null,Mc.forEach(pi)):zt(`invalid state: ${ht}`);return fi}}(p=>{s().then(p)})}function Rc(s){return s>>>=0,yi(()=>(s=qe(s)).then(je))}var lr=[];function Uc(s,p,f,b){return f>>>=0,b>>>=0,(s=lr[s>>>0])(null,p=qe(p>>>0),f,b)}var Vc={},cr=s=>{var p=Vc[s];return p===void 0?Je(s):p};function Nc(s,p,f,b,v){return f>>>=0,b>>>=0,v>>>=0,(s=lr[s>>>0])(p=qe(p>>>0),p[f=cr(f)],b,v)}var bi=()=>typeof globalThis=="object"?globalThis:Function("return this")();function Wc(s){return(s>>>=0)==0?je(bi()):(s=cr(s),je(bi()[s]))}var Lc=s=>{var p=lr.length;return lr.push(s),p},Hc=(s,p)=>{for(var f=Array(s),b=0;b<s;++b)f[b]=En(a()[p+4*b>>>2>>>0],"parameter "+b);return f},wi=(s,p)=>Object.defineProperty(p,"name",{value:s});function Gc(s,p,f){var b=(p=Hc(s,p>>>0)).shift();s--;var v=`return function (obj, func, destructorsRef, args) {
-`,I=0,z=[];f===0&&z.push("obj");for(var B=["retType"],W=[b],H=0;H<s;++H)z.push("arg"+H),B.push("argType"+H),W.push(p[H]),v+=`  var arg${H} = argType${H}.readValueFromPointer(args${I?"+"+I:""});
-`,I+=p[H].argPackAdvance;return v+=`  var rv = ${f===1?"new func":"func.call"}(${z.join(", ")});
-`,b.Tb||(B.push("emval_returnValue"),W.push(mi),v+=`  return emval_returnValue(retType, destructorsRef, rv);
-`),B.push(v+`};
-`),s=function(j){var ce=Function;if(!(ce instanceof Function))throw new TypeError(`new_ called with constructor type ${typeof ce} which is not a function`);var ge=wi(ce.name||"unknownFunctionName",function(){});return ge.prototype=ce.prototype,ge=new ge,(j=ce.apply(ge,j))instanceof Object?j:ge}(B)(...W),f=`methodCaller<(${p.map(j=>j.name).join(", ")}) => ${b.name}>`,Lc(wi(f,s))}function Fc(s){return s=cr(s>>>0),je(u[s])}function qc(s,p){return p>>>=0,s=qe(s>>>0),p=qe(p),je(s[p])}function jc(s){9<(s>>>=0)&&(dt[s+1]+=1)}function Kc(){return je([])}function Yc(s){s=qe(s>>>0);for(var p=Array(s.length),f=0;f<s.length;f++)p[f]=s[f];return je(p)}function Xc(s){return je(cr(s>>>0))}function Zc(){return je({})}function Qc(s){for(var p=qe(s>>>=0);p.length;){var f=p.pop();p.pop()(f)}Tn(s)}function Jc(s,p,f){p>>>=0,f>>>=0,s=qe(s>>>0),p=qe(p),f=qe(f),s[p]=f}function ep(s,p){return p>>>=0,s=(s=En(s>>>0,"_emval_take_value")).readValueFromPointer(p),je(s)}function tp(s,p){s=-9007199254740992>s||9007199254740992<s?NaN:Number(s),p>>>=0,s=new Date(1e3*s),i()[p>>>2>>>0]=s.getUTCSeconds(),i()[p+4>>>2>>>0]=s.getUTCMinutes(),i()[p+8>>>2>>>0]=s.getUTCHours(),i()[p+12>>>2>>>0]=s.getUTCDate(),i()[p+16>>>2>>>0]=s.getUTCMonth(),i()[p+20>>>2>>>0]=s.getUTCFullYear()-1900,i()[p+24>>>2>>>0]=s.getUTCDay(),s=(s.getTime()-Date.UTC(s.getUTCFullYear(),0,1,0,0,0,0))/864e5|0,i()[p+28>>>2>>>0]=s}var Bt=s=>s%4==0&&(s%100!=0||s%400==0),vi=[0,31,60,91,121,152,182,213,244,274,305,335],_i=[0,31,59,90,120,151,181,212,243,273,304,334];function rp(s,p){s=-9007199254740992>s||9007199254740992<s?NaN:Number(s),p>>>=0,s=new Date(1e3*s),i()[p>>>2>>>0]=s.getSeconds(),i()[p+4>>>2>>>0]=s.getMinutes(),i()[p+8>>>2>>>0]=s.getHours(),i()[p+12>>>2>>>0]=s.getDate(),i()[p+16>>>2>>>0]=s.getMonth(),i()[p+20>>>2>>>0]=s.getFullYear()-1900,i()[p+24>>>2>>>0]=s.getDay();var f=(Bt(s.getFullYear())?vi:_i)[s.getMonth()]+s.getDate()-1|0;i()[p+28>>>2>>>0]=f,i()[p+36>>>2>>>0]=-60*s.getTimezoneOffset(),f=new Date(s.getFullYear(),6,1).getTimezoneOffset();var b=new Date(s.getFullYear(),0,1).getTimezoneOffset();s=0|(f!=b&&s.getTimezoneOffset()==Math.min(b,f)),i()[p+32>>>2>>>0]=s}function np(s){s>>>=0;var p=new Date(i()[s+20>>>2>>>0]+1900,i()[s+16>>>2>>>0],i()[s+12>>>2>>>0],i()[s+8>>>2>>>0],i()[s+4>>>2>>>0],i()[s>>>2>>>0],0),f=i()[s+32>>>2>>>0],b=p.getTimezoneOffset(),v=new Date(p.getFullYear(),6,1).getTimezoneOffset(),I=new Date(p.getFullYear(),0,1).getTimezoneOffset(),z=Math.min(I,v);return 0>f?i()[s+32>>>2>>>0]=+(v!=I&&z==b):0<f!=(z==b)&&(v=Math.max(I,v),p.setTime(p.getTime()+6e4*((0<f?z:v)-b))),i()[s+24>>>2>>>0]=p.getDay(),f=(Bt(p.getFullYear())?vi:_i)[p.getMonth()]+p.getDate()-1|0,i()[s+28>>>2>>>0]=f,i()[s>>>2>>>0]=p.getSeconds(),i()[s+4>>>2>>>0]=p.getMinutes(),i()[s+8>>>2>>>0]=p.getHours(),i()[s+12>>>2>>>0]=p.getDate(),i()[s+16>>>2>>>0]=p.getMonth(),i()[s+20>>>2>>>0]=p.getYear(),s=p.getTime(),BigInt(isNaN(s)?-1:s/1e3)}function $i(s,p,f,b,v,I,z){return y?xe(16,1,s,p,f,b,v,I,z):-52}function xi(s,p,f,b,v,I){if(y)return xe(17,1,s,p,f,b,v,I)}function op(s,p,f,b){s>>>=0,p>>>=0,f>>>=0,b>>>=0;var v=new Date().getFullYear(),I=new Date(v,0,1),z=new Date(v,6,1);v=I.getTimezoneOffset();var B=z.getTimezoneOffset(),W=Math.max(v,B);a()[s>>>2>>>0]=60*W,i()[p>>>2>>>0]=+(v!=B),I=(s=H=>H.toLocaleTimeString(void 0,{hour12:!1,timeZoneName:"short"}).split(" ")[1])(I),z=s(z),B<v?(Dt(I,f,17),Dt(z,b,17)):(Dt(I,b,17),Dt(z,f,17))}var Pn=[],Si=(s,p)=>{Pn.length=0;for(var f;f=r()[s++>>>0];){var b=f!=105;p+=(b&=f!=112)&&p%8?4:0,Pn.push(f==112?a()[p>>>2>>>0]:f==106?G[p>>>3]:f==105?i()[p>>>2>>>0]:l()[p>>>3>>>0]),p+=b?8:4}return Pn};function ip(s,p,f){return s>>>=0,p=Si(p>>>0,f>>>0),gn[s](...p)}function ap(s,p,f){return s>>>=0,p=Si(p>>>0,f>>>0),gn[s](...p)}var sp=()=>{},up=()=>Date.now();function dp(s,p){return X(Oe(s>>>0,p>>>0))}var Ti,lp=()=>{throw vt+=1,"unwind"};function cp(){return 4294901760}Ti=()=>performance.timeOrigin+performance.now();var pp=()=>navigator.hardwareConcurrency;function mp(){return zt("Cannot use emscripten_pc_get_function without -sUSE_OFFSET_CONVERTER"),0}function fp(s){s>>>=0;var p=r().length;if(s<=p||4294901760<s)return!1;for(var f=1;4>=f;f*=2){var b=p*(1+.2/f);b=Math.min(b,s+100663296);var v=Math;b=Math.max(s,b);e:{v=(v.min.call(v,4294901760,b+(65536-b%65536)%65536)-le.buffer.byteLength+65535)/65536;try{le.grow(v),Ce();var I=1;break e}catch{}I=void 0}if(I)return!0}return!1}var pr=()=>(zt("Cannot use convertFrameToPC (needed by __builtin_return_address) without -sUSE_OFFSET_CONVERTER"),0),Mt={},Ii=s=>{s.forEach(p=>{var f=pr();f&&(Mt[f]=p)})};function hp(){var s=Error().stack.toString().split(`
+var Un=Object.defineProperty;var vp=Object.getOwnPropertyDescriptor;var $p=Object.getOwnPropertyNames;var xp=Object.prototype.hasOwnProperty;var Vn=(e=>typeof require<"u"?require:typeof Proxy<"u"?new Proxy(e,{get:(t,r)=>(typeof require<"u"?require:t)[r]}):e)(function(e){if(typeof require<"u")return require.apply(this,arguments);throw Error('Dynamic require of "'+e+'" is not supported')});var V=(e,t)=>()=>(e&&(t=e(e=0)),t);var Gt=(e,t)=>{for(var r in t)Un(e,r,{get:t[r],enumerable:!0})},Sp=(e,t,r,n)=>{if(t&&typeof t=="object"||typeof t=="function")for(let o of $p(t))!xp.call(e,o)&&o!==r&&Un(e,o,{get:()=>t[o],enumerable:!(n=vp(t,o))||n.enumerable});return e};var br=e=>Sp(Un({},"__esModule",{value:!0}),e);var wr,xt,St,Tp,_r,vr=V(()=>{"use strict";wr=new Map,xt=[],St=(e,t,r)=>{if(t&&typeof t.init=="function"&&typeof t.createInferenceSessionHandler=="function"){let n=wr.get(e);if(n===void 0)wr.set(e,{backend:t,priority:r});else{if(n.priority>r)return;if(n.priority===r&&n.backend!==t)throw new Error(`cannot register backend "${e}" using priority ${r}`)}if(r>=0){let o=xt.indexOf(e);o!==-1&&xt.splice(o,1);for(let i=0;i<xt.length;i++)if(wr.get(xt[i]).priority<=r){xt.splice(i,0,e);return}xt.push(e)}return}throw new TypeError("not a valid backend")},Tp=async e=>{let t=wr.get(e);if(!t)return"backend not found.";if(t.initialized)return t.backend;if(t.aborted)return t.error;{let r=!!t.initPromise;try{return r||(t.initPromise=t.backend.init(e)),await t.initPromise,t.initialized=!0,t.backend}catch(n){return r||(t.error=`${n}`,t.aborted=!0),t.error}finally{delete t.initPromise}}},_r=async e=>{let t=e.executionProviders||[],r=t.map(l=>typeof l=="string"?l:l.name),n=r.length===0?xt:r,o,i=[],a=new Set;for(let l of n){let c=await Tp(l);typeof c=="string"?i.push({name:l,err:c}):(o||(o=c),o===c&&a.add(l))}if(!o)throw new Error(`no available backend found. ERR: ${i.map(l=>`[${l.name}] ${l.err}`).join(", ")}`);for(let{name:l,err:c}of i)r.includes(l)&&console.warn(`removing requested execution provider "${l}" from session options because it is not available: ${c}`);let d=t.filter(l=>a.has(typeof l=="string"?l:l.name));return[o,new Proxy(e,{get:(l,c)=>c==="executionProviders"?d:Reflect.get(l,c)})]}});var Zi=V(()=>{"use strict";vr()});var Qi,Ji=V(()=>{"use strict";Qi="1.20.0-dev.20241016-2b8fc5529b"});var ea,We,Nn=V(()=>{"use strict";Ji();ea="warning",We={wasm:{},webgl:{},webgpu:{},versions:{common:Qi},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}`);ea=e}},get logLevel(){return ea}};Object.defineProperty(We,"logLevel",{enumerable:!0})});var _e,ta=V(()=>{"use strict";Nn();_e=We});var ra,na,oa=V(()=>{"use strict";ra=(e,t)=>{let r=typeof document<"u"?document.createElement("canvas"):new OffscreenCanvas(1,1);r.width=e.dims[3],r.height=e.dims[2];let n=r.getContext("2d");if(n!=null){let o,i;t?.tensorLayout!==void 0&&t.tensorLayout==="NHWC"?(o=e.dims[2],i=e.dims[3]):(o=e.dims[3],i=e.dims[2]);let a=t?.format!==void 0?t.format:"RGB",d=t?.norm,l,c;d===void 0||d.mean===void 0?l=[255,255,255,255]:typeof d.mean=="number"?l=[d.mean,d.mean,d.mean,d.mean]:(l=[d.mean[0],d.mean[1],d.mean[2],0],d.mean[3]!==void 0&&(l[3]=d.mean[3])),d===void 0||d.bias===void 0?c=[0,0,0,0]:typeof d.bias=="number"?c=[d.bias,d.bias,d.bias,d.bias]:(c=[d.bias[0],d.bias[1],d.bias[2],0],d.bias[3]!==void 0&&(c[3]=d.bias[3]));let m=i*o,u=0,h=m,w=m*2,g=-1;a==="RGBA"?(u=0,h=m,w=m*2,g=m*3):a==="RGB"?(u=0,h=m,w=m*2):a==="RBG"&&(u=0,w=m,h=m*2);for(let y=0;y<i;y++)for(let S=0;S<o;S++){let $=(e.data[u++]-c[0])*l[0],v=(e.data[h++]-c[1])*l[1],x=(e.data[w++]-c[2])*l[2],T=g===-1?255:(e.data[g++]-c[3])*l[3];n.fillStyle="rgba("+$+","+v+","+x+","+T+")",n.fillRect(S,y,1,1)}if("toDataURL"in r)return r.toDataURL();throw new Error("toDataURL is not supported")}else throw new Error("Can not access image data")},na=(e,t)=>{let r=typeof document<"u"?document.createElement("canvas").getContext("2d"):new OffscreenCanvas(1,1).getContext("2d"),n;if(r!=null){let o,i,a;t?.tensorLayout!==void 0&&t.tensorLayout==="NHWC"?(o=e.dims[2],i=e.dims[1],a=e.dims[3]):(o=e.dims[3],i=e.dims[2],a=e.dims[1]);let d=t!==void 0&&t.format!==void 0?t.format:"RGB",l=t?.norm,c,m;l===void 0||l.mean===void 0?c=[255,255,255,255]:typeof l.mean=="number"?c=[l.mean,l.mean,l.mean,l.mean]:(c=[l.mean[0],l.mean[1],l.mean[2],255],l.mean[3]!==void 0&&(c[3]=l.mean[3])),l===void 0||l.bias===void 0?m=[0,0,0,0]:typeof l.bias=="number"?m=[l.bias,l.bias,l.bias,l.bias]:(m=[l.bias[0],l.bias[1],l.bias[2],0],l.bias[3]!==void 0&&(m[3]=l.bias[3]));let u=i*o;if(t!==void 0&&(t.format!==void 0&&a===4&&t.format!=="RGBA"||a===3&&t.format!=="RGB"&&t.format!=="BGR"))throw new Error("Tensor format doesn't match input tensor dims");let h=4,w=0,g=1,y=2,S=3,$=0,v=u,x=u*2,T=-1;d==="RGBA"?($=0,v=u,x=u*2,T=u*3):d==="RGB"?($=0,v=u,x=u*2):d==="RBG"&&($=0,x=u,v=u*2),n=r.createImageData(o,i);for(let C=0;C<i*o;w+=h,g+=h,y+=h,S+=h,C++)n.data[w]=(e.data[$++]-m[0])*c[0],n.data[g]=(e.data[v++]-m[1])*c[1],n.data[y]=(e.data[x++]-m[2])*c[2],n.data[S]=T===-1?255:(e.data[T++]-m[3])*c[3]}else throw new Error("Can not access image data");return n}});var Wn,ia,aa,sa,ua,da,la=V(()=>{"use strict";$r();Wn=(e,t)=>{if(e===void 0)throw new Error("Image buffer must be defined");if(t.height===void 0||t.width===void 0)throw new Error("Image height and width must be defined");if(t.tensorLayout==="NHWC")throw new Error("NHWC Tensor layout is not supported yet");let{height:r,width:n}=t,o=t.norm??{mean:255,bias:0},i,a;typeof o.mean=="number"?i=[o.mean,o.mean,o.mean,o.mean]:i=[o.mean[0],o.mean[1],o.mean[2],o.mean[3]??255],typeof o.bias=="number"?a=[o.bias,o.bias,o.bias,o.bias]:a=[o.bias[0],o.bias[1],o.bias[2],o.bias[3]??0];let d=t.format!==void 0?t.format:"RGBA",l=t.tensorFormat!==void 0&&t.tensorFormat!==void 0?t.tensorFormat:"RGB",c=r*n,m=l==="RGBA"?new Float32Array(c*4):new Float32Array(c*3),u=4,h=0,w=1,g=2,y=3,S=0,$=c,v=c*2,x=-1;d==="RGB"&&(u=3,h=0,w=1,g=2,y=-1),l==="RGBA"?x=c*3:l==="RBG"?(S=0,v=c,$=c*2):l==="BGR"&&(v=0,$=c,S=c*2);for(let C=0;C<c;C++,h+=u,g+=u,w+=u,y+=u)m[S++]=(e[h]+a[0])/i[0],m[$++]=(e[w]+a[1])/i[1],m[v++]=(e[g]+a[2])/i[2],x!==-1&&y!==-1&&(m[x++]=(e[y]+a[3])/i[3]);return l==="RGBA"?new Be("float32",m,[1,4,r,n]):new Be("float32",m,[1,3,r,n])},ia=async(e,t)=>{let r=typeof HTMLImageElement<"u"&&e instanceof HTMLImageElement,n=typeof ImageData<"u"&&e instanceof ImageData,o=typeof ImageBitmap<"u"&&e instanceof ImageBitmap,i=typeof e=="string",a,d=t??{},l=()=>{if(typeof document<"u")return document.createElement("canvas");if(typeof OffscreenCanvas<"u")return new OffscreenCanvas(1,1);throw new Error("Canvas is not supported")},c=m=>typeof HTMLCanvasElement<"u"&&m instanceof HTMLCanvasElement||m instanceof OffscreenCanvas?m.getContext("2d"):null;if(r){let m=l();m.width=e.width,m.height=e.height;let u=c(m);if(u!=null){let h=e.height,w=e.width;if(t!==void 0&&t.resizedHeight!==void 0&&t.resizedWidth!==void 0&&(h=t.resizedHeight,w=t.resizedWidth),t!==void 0){if(d=t,t.tensorFormat!==void 0)throw new Error("Image input config format must be RGBA for HTMLImageElement");d.tensorFormat="RGBA",d.height=h,d.width=w}else d.tensorFormat="RGBA",d.height=h,d.width=w;u.drawImage(e,0,0),a=u.getImageData(0,0,w,h).data}else throw new Error("Can not access image data")}else if(n){let m,u;if(t!==void 0&&t.resizedWidth!==void 0&&t.resizedHeight!==void 0?(m=t.resizedHeight,u=t.resizedWidth):(m=e.height,u=e.width),t!==void 0&&(d=t),d.format="RGBA",d.height=m,d.width=u,t!==void 0){let h=l();h.width=u,h.height=m;let w=c(h);if(w!=null)w.putImageData(e,0,0),a=w.getImageData(0,0,u,m).data;else throw new Error("Can not access image data")}else a=e.data}else if(o){if(t===void 0)throw new Error("Please provide image config with format for Imagebitmap");let m=l();m.width=e.width,m.height=e.height;let u=c(m);if(u!=null){let h=e.height,w=e.width;return u.drawImage(e,0,0,w,h),a=u.getImageData(0,0,w,h).data,d.height=h,d.width=w,Wn(a,d)}else throw new Error("Can not access image data")}else{if(i)return new Promise((m,u)=>{let h=l(),w=c(h);if(!e||!w)return u();let g=new Image;g.crossOrigin="Anonymous",g.src=e,g.onload=()=>{h.width=g.width,h.height=g.height,w.drawImage(g,0,0,h.width,h.height);let y=w.getImageData(0,0,h.width,h.height);d.height=h.height,d.width=h.width,m(Wn(y.data,d))}});throw new Error("Input data provided is not supported - aborted tensor creation")}if(a!==void 0)return Wn(a,d);throw new Error("Input data provided is not supported - aborted tensor creation")},aa=(e,t)=>{let{width:r,height:n,download:o,dispose:i}=t,a=[1,n,r,4];return new Be({location:"texture",type:"float32",texture:e,dims:a,download:o,dispose:i})},sa=(e,t)=>{let{dataType:r,dims:n,download:o,dispose:i}=t;return new Be({location:"gpu-buffer",type:r??"float32",gpuBuffer:e,dims:n,download:o,dispose:i})},ua=(e,t)=>{let{dataType:r,dims:n,download:o,dispose:i}=t;return new Be({location:"ml-tensor",type:r??"float32",mlTensor:e,dims:n,download:o,dispose:i})},da=(e,t,r)=>new Be({location:"cpu-pinned",type:e,data:t,dims:r??[t.length]})});var Tt,Ft,ca,pa,ma=V(()=>{"use strict";Tt=new Map([["float32",Float32Array],["uint8",Uint8Array],["int8",Int8Array],["uint16",Uint16Array],["int16",Int16Array],["int32",Int32Array],["bool",Uint8Array],["float64",Float64Array],["uint32",Uint32Array],["int4",Uint8Array],["uint4",Uint8Array]]),Ft=new Map([[Float32Array,"float32"],[Uint8Array,"uint8"],[Int8Array,"int8"],[Uint16Array,"uint16"],[Int16Array,"int16"],[Int32Array,"int32"],[Float64Array,"float64"],[Uint32Array,"uint32"]]),ca=!1,pa=()=>{if(!ca){ca=!0;let e=typeof BigInt64Array<"u"&&BigInt64Array.from,t=typeof BigUint64Array<"u"&&BigUint64Array.from,r=typeof Float16Array<"u"&&Float16Array.from;e&&(Tt.set("int64",BigInt64Array),Ft.set(BigInt64Array,"int64")),t&&(Tt.set("uint64",BigUint64Array),Ft.set(BigUint64Array,"uint64")),r?(Tt.set("float16",Float16Array),Ft.set(Float16Array,"float16")):Tt.set("float16",Uint16Array)}}});var fa,ha,ga=V(()=>{"use strict";$r();fa=e=>{let t=1;for(let r=0;r<e.length;r++){let n=e[r];if(typeof n!="number"||!Number.isSafeInteger(n))throw new TypeError(`dims[${r}] must be an integer, got: ${n}`);if(n<0)throw new RangeError(`dims[${r}] must be a non-negative integer, got: ${n}`);t*=n}return t},ha=(e,t)=>{switch(e.location){case"cpu":return new Be(e.type,e.data,t);case"cpu-pinned":return new Be({location:"cpu-pinned",data:e.data,type:e.type,dims:t});case"texture":return new Be({location:"texture",texture:e.texture,type:e.type,dims:t});case"gpu-buffer":return new Be({location:"gpu-buffer",gpuBuffer:e.gpuBuffer,type:e.type,dims:t});case"ml-tensor":return new Be({location:"ml-tensor",mlTensor:e.mlTensor,type:e.type,dims:t});default:throw new Error(`tensorReshape: tensor location ${e.location} is not supported`)}}});var Be,$r=V(()=>{"use strict";oa();la();ma();ga();Be=class{constructor(t,r,n){pa();let o,i;if(typeof t=="object"&&"location"in t)switch(this.dataLocation=t.location,o=t.type,i=t.dims,t.location){case"cpu-pinned":{let d=Tt.get(o);if(!d)throw new TypeError(`unsupported type "${o}" to create tensor from pinned buffer`);if(!(t.data instanceof d))throw new TypeError(`buffer should be of type ${d.name}`);this.cpuData=t.data;break}case"texture":{if(o!=="float32")throw new TypeError(`unsupported type "${o}" to create tensor from texture`);this.gpuTextureData=t.texture,this.downloader=t.download,this.disposer=t.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 "${o}" to create tensor from gpu buffer`);this.gpuBufferData=t.gpuBuffer,this.downloader=t.download,this.disposer=t.dispose;break}case"ml-tensor":{if(o!=="float32"&&o!=="float16"&&o!=="int32"&&o!=="int64"&&o!=="uint32"&&o!=="uint64"&&o!=="int8"&&o!=="uint8"&&o!=="bool")throw new TypeError(`unsupported type "${o}" to create tensor from MLTensor`);this.mlTensorData=t.mlTensor,this.downloader=t.download,this.disposer=t.dispose;break}default:throw new Error(`Tensor constructor: unsupported location '${this.dataLocation}'`)}else{let d,l;if(typeof t=="string")if(o=t,l=n,t==="string"){if(!Array.isArray(r))throw new TypeError("A string tensor's data must be a string array.");d=r}else{let c=Tt.get(t);if(c===void 0)throw new TypeError(`Unsupported tensor type: ${t}.`);if(Array.isArray(r)){if(t==="float16"&&c===Uint16Array||t==="uint4"||t==="int4")throw new TypeError(`Creating a ${t} tensor from number array is not supported. Please use ${c.name} as data.`);t==="uint64"||t==="int64"?d=c.from(r,BigInt):d=c.from(r)}else if(r instanceof c)d=r;else if(r instanceof Uint8ClampedArray)if(t==="uint8")d=Uint8Array.from(r);else throw new TypeError("A Uint8ClampedArray tensor's data must be type of uint8");else throw new TypeError(`A ${o} tensor's data must be type of ${c}`)}else if(l=r,Array.isArray(t)){if(t.length===0)throw new TypeError("Tensor type cannot be inferred from an empty array.");let c=typeof t[0];if(c==="string")o="string",d=t;else if(c==="boolean")o="bool",d=Uint8Array.from(t);else throw new TypeError(`Invalid element type of data array: ${c}.`)}else if(t instanceof Uint8ClampedArray)o="uint8",d=Uint8Array.from(t);else{let c=Ft.get(t.constructor);if(c===void 0)throw new TypeError(`Unsupported type for tensor data: ${t.constructor}.`);o=c,d=t}if(l===void 0)l=[d.length];else if(!Array.isArray(l))throw new TypeError("A tensor's dims must be a number array");i=l,this.cpuData=d,this.dataLocation="cpu"}let a=fa(i);if(this.cpuData&&a!==this.cpuData.length&&!((o==="uint4"||o==="int4")&&Math.ceil(a/2)===this.cpuData.length))throw new Error(`Tensor's size(${a}) does not match data length(${this.cpuData.length}).`);this.type=o,this.dims=i,this.size=a}static async fromImage(t,r){return ia(t,r)}static fromTexture(t,r){return aa(t,r)}static fromGpuBuffer(t,r){return sa(t,r)}static fromMLTensor(t,r){return ua(t,r)}static fromPinnedBuffer(t,r,n){return da(t,r,n)}toDataURL(t){return ra(this,t)}toImageData(t){return na(this,t)}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(t){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,t&&this.disposer&&(this.disposer(),this.disposer=void 0),r}finally{this.isDownloading=!1}}default:throw new Error(`cannot get data from location: ${this.dataLocation}`)}}dispose(){if(this.isDownloading)throw new Error("The current tensor is being downloaded.");this.disposer&&(this.disposer(),this.disposer=void 0),this.cpuData=void 0,this.gpuTextureData=void 0,this.gpuBufferData=void 0,this.mlTensorData=void 0,this.downloader=void 0,this.isDownloading=void 0,this.dataLocation="none"}ensureValid(){if(this.dataLocation==="none")throw new Error("The tensor is disposed.")}reshape(t){if(this.ensureValid(),this.downloader||this.disposer)throw new Error("Cannot reshape a tensor that owns GPU resource.");return ha(this,t)}}});var De,xr=V(()=>{"use strict";$r();De=Be});var Sr,ya,Le,Ve,Ln=V(()=>{"use strict";Nn();Sr=(e,t)=>{(typeof We.trace>"u"?!We.wasm.trace:!We.trace)||console.timeStamp(`${e}::ORT::${t}`)},ya=(e,t)=>{let r=new Error().stack?.split(/\r\n|\r|\n/g)||[],n=!1;for(let o=0;o<r.length;o++){if(n&&!r[o].includes("TRACE_FUNC")){let i=`FUNC_${e}::${r[o].trim().split(" ")[1]}`;t&&(i+=`::${t}`),Sr("CPU",i);return}r[o].includes("TRACE_FUNC")&&(n=!0)}},Le=e=>{(typeof We.trace>"u"?!We.wasm.trace:!We.trace)||ya("BEGIN",e)},Ve=e=>{(typeof We.trace>"u"?!We.wasm.trace:!We.trace)||ya("END",e)}});var Tr,ba=V(()=>{"use strict";vr();xr();Ln();Tr=class e{constructor(t){this.handler=t}async run(t,r,n){Le();let o={},i={};if(typeof t!="object"||t===null||t instanceof De||Array.isArray(t))throw new TypeError("'feeds' must be an object that use input names as keys and OnnxValue as corresponding values.");let a=!0;if(typeof r=="object"){if(r===null)throw new TypeError("Unexpected argument[1]: cannot be null.");if(r instanceof De)throw new TypeError("'fetches' cannot be a Tensor");if(Array.isArray(r)){if(r.length===0)throw new TypeError("'fetches' cannot be an empty array.");a=!1;for(let c of r){if(typeof c!="string")throw new TypeError("'fetches' must be a string array or an object.");if(this.outputNames.indexOf(c)===-1)throw new RangeError(`'fetches' contains invalid output name: ${c}.`);o[c]=null}if(typeof n=="object"&&n!==null)i=n;else if(typeof n<"u")throw new TypeError("'options' must be an object.")}else{let c=!1,m=Object.getOwnPropertyNames(r);for(let u of this.outputNames)if(m.indexOf(u)!==-1){let h=r[u];(h===null||h instanceof De)&&(c=!0,a=!1,o[u]=h)}if(c){if(typeof n=="object"&&n!==null)i=n;else if(typeof n<"u")throw new TypeError("'options' must be an object.")}else i=r}}else if(typeof r<"u")throw new TypeError("Unexpected argument[1]: must be 'fetches' or 'options'.");for(let c of this.inputNames)if(typeof t[c]>"u")throw new Error(`input '${c}' is missing in 'feeds'.`);if(a)for(let c of this.outputNames)o[c]=null;let d=await this.handler.run(t,o,i),l={};for(let c in d)if(Object.hasOwnProperty.call(d,c)){let m=d[c];m instanceof De?l[c]=m:l[c]=new De(m.type,m.data,m.dims)}return Ve(),l}async release(){return this.handler.dispose()}static async create(t,r,n,o){Le();let i,a={};if(typeof t=="string"){if(i=t,typeof r=="object"&&r!==null)a=r;else if(typeof r<"u")throw new TypeError("'options' must be an object.")}else if(t instanceof Uint8Array){if(i=t,typeof r=="object"&&r!==null)a=r;else if(typeof r<"u")throw new TypeError("'options' must be an object.")}else if(t instanceof ArrayBuffer||typeof SharedArrayBuffer<"u"&&t instanceof SharedArrayBuffer){let m=t,u=0,h=t.byteLength;if(typeof r=="object"&&r!==null)a=r;else if(typeof r=="number"){if(u=r,!Number.isSafeInteger(u))throw new RangeError("'byteOffset' must be an integer.");if(u<0||u>=m.byteLength)throw new RangeError(`'byteOffset' is out of range [0, ${m.byteLength}).`);if(h=t.byteLength-u,typeof n=="number"){if(h=n,!Number.isSafeInteger(h))throw new RangeError("'byteLength' must be an integer.");if(h<=0||u+h>m.byteLength)throw new RangeError(`'byteLength' is out of range (0, ${m.byteLength-u}].`);if(typeof o=="object"&&o!==null)a=o;else if(typeof o<"u")throw new TypeError("'options' must be an object.")}else if(typeof n<"u")throw new TypeError("'byteLength' must be a number.")}else if(typeof r<"u")throw new TypeError("'options' must be an object.");i=new Uint8Array(m,u,h)}else throw new TypeError("Unexpected argument[0]: must be 'path' or 'buffer'.");let[d,l]=await _r(a),c=await d.createInferenceSessionHandler(i,l);return Ve(),new e(c)}startProfiling(){this.handler.startProfiling()}endProfiling(){this.handler.endProfiling()}get inputNames(){return this.handler.inputNames}get outputNames(){return this.handler.outputNames}}});var Ip,wa=V(()=>{"use strict";ba();Ip=Tr});var _a=V(()=>{"use strict"});var va=V(()=>{"use strict"});var $a=V(()=>{"use strict"});var xa=V(()=>{"use strict"});var Cp,Ir,Sa=V(()=>{"use strict";vr();xr();Cp="Training backend could not be resolved. Make sure you're using the correct configuration & WebAssembly files.",Ir=class e{constructor(t,r,n){this.handler=t,this.hasOptimizerModel=r,this.hasEvalModel=n}get trainingInputNames(){return this.handler.inputNames}get trainingOutputNames(){return this.handler.outputNames}get evalInputNames(){if(this.hasEvalModel)return this.handler.evalInputNames;throw new Error("This training session has no evalModel loaded.")}get evalOutputNames(){if(this.hasEvalModel)return this.handler.evalOutputNames;throw new Error("This training session has no evalModel loaded.")}static async create(t,r){let n=t.evalModel||"",o=t.optimizerModel||"",i=r||{},[a,d]=await _r(i);if(a.createTrainingSessionHandler){let l=await a.createTrainingSessionHandler(t.checkpointState,t.trainModel,n,o,d);return new e(l,!!t.optimizerModel,!!t.evalModel)}else throw new Error(Cp)}typeNarrowingForRunStep(t,r,n,o,i){let a={},d={};if(typeof n!="object"||n===null||n instanceof De||Array.isArray(n))throw new TypeError("'feeds' must be an object that use input names as keys and OnnxValue as corresponding values.");let l=!0;if(typeof o=="object"){if(o===null)throw new TypeError("Unexpected argument[1]: cannot be null.");if(o instanceof De)throw new TypeError("'fetches' cannot be a Tensor");if(Array.isArray(o)){if(o.length===0)throw new TypeError("'fetches' cannot be an empty array.");l=!1;for(let c of o){if(typeof c!="string")throw new TypeError("'fetches' must be a string array or an object.");if(r.indexOf(c)===-1)throw new RangeError(`'fetches' contains invalid output name: ${c}.`);a[c]=null}if(typeof i=="object"&&i!==null)d=i;else if(typeof i<"u")throw new TypeError("'options' must be an object.")}else{let c=!1,m=Object.getOwnPropertyNames(o);for(let u of r)if(m.indexOf(u)!==-1){let h=o[u];(h===null||h instanceof De)&&(c=!0,l=!1,a[u]=h)}if(c){if(typeof i=="object"&&i!==null)d=i;else if(typeof i<"u")throw new TypeError("'options' must be an object.")}else d=o}}else if(typeof o<"u")throw new TypeError("Unexpected argument[1]: must be 'fetches' or 'options'.");for(let c of t)if(typeof n[c]>"u")throw new Error(`input '${c}' is missing in 'feeds'.`);if(l)for(let c of r)a[c]=null;return[a,d]}convertHandlerReturnTypeToMapOfTensors(t){let r={};for(let n in t)if(Object.hasOwnProperty.call(t,n)){let o=t[n];o instanceof De?r[n]=o:r[n]=new De(o.type,o.data,o.dims)}return r}async lazyResetGrad(){await this.handler.lazyResetGrad()}async runTrainStep(t,r,n){let[o,i]=this.typeNarrowingForRunStep(this.trainingInputNames,this.trainingOutputNames,t,r,n),a=await this.handler.runTrainStep(t,o,i);return this.convertHandlerReturnTypeToMapOfTensors(a)}async runOptimizerStep(t){if(this.hasOptimizerModel)await this.handler.runOptimizerStep(t||{});else throw new Error("This TrainingSession has no OptimizerModel loaded.")}async runEvalStep(t,r,n){if(this.hasEvalModel){let[o,i]=this.typeNarrowingForRunStep(this.evalInputNames,this.evalOutputNames,t,r,n),a=await this.handler.runEvalStep(t,o,i);return this.convertHandlerReturnTypeToMapOfTensors(a)}else throw new Error("This TrainingSession has no EvalModel loaded.")}async getParametersSize(t=!0){return this.handler.getParametersSize(t)}async loadParametersBuffer(t,r=!0){let n=await this.getParametersSize(r);if(t.length!==4*n)throw new Error("Size of the buffer passed into loadParametersBuffer must match the number of parameters in the model. Please use getParametersSize method to check.");return this.handler.loadParametersBuffer(t,r)}async getContiguousParameters(t=!0){return this.handler.getContiguousParameters(t)}async release(){return this.handler.dispose()}}});var Ap,Ta=V(()=>{"use strict";Sa();Ap=Ir});var Hn={};Gt(Hn,{InferenceSession:()=>Ip,TRACE:()=>Sr,TRACE_FUNC_BEGIN:()=>Le,TRACE_FUNC_END:()=>Ve,Tensor:()=>De,TrainingSession:()=>Ap,env:()=>_e,registerBackend:()=>St});var Ke=V(()=>{"use strict";Zi();ta();wa();xr();_a();va();Ln();$a();xa();Ta()});var Cr=V(()=>{"use strict"});var ka={};Gt(ka,{default:()=>kp});var Ca,Aa,kp,Ea=V(()=>{"use strict";Gn();gt();qt();Ca="ort-wasm-proxy-worker",Aa=globalThis.self?.name===Ca;Aa&&(self.onmessage=e=>{let{type:t,in:r}=e.data;try{switch(t){case"init-wasm":Ar(r.wasm).then(()=>{kr(r).then(()=>{postMessage({type:t})},n=>{postMessage({type:t,err:n})})},n=>{postMessage({type:t,err:n})});break;case"init-ep":{let{epName:n,env:o}=r;Er(o,n).then(()=>{postMessage({type:t})},i=>{postMessage({type:t,err:i})});break}case"copy-from":{let{buffer:n}=r,o=jt(n);postMessage({type:t,out:o});break}case"create":{let{model:n,options:o}=r;Pr(n,o).then(i=>{postMessage({type:t,out:i})},i=>{postMessage({type:t,err:i})});break}case"release":zr(r),postMessage({type:t});break;case"run":{let{sessionId:n,inputIndices:o,inputs:i,outputIndices:a,options:d}=r;Or(n,o,i,a,new Array(a.length).fill(null),d).then(l=>{l.some(c=>c[3]!=="cpu")?postMessage({type:t,err:"Proxy does not support non-cpu tensor location."}):postMessage({type:t,out:l},Dr([...i,...l]))},l=>{postMessage({type:t,err:l})});break}case"end-profiling":Br(r),postMessage({type:t});break;default:}}catch(n){postMessage({type:t,err:n})}});kp=Aa?null:e=>new Worker(e??Ut,{type:"module",name:Ca})});var za={};Gt(za,{default:()=>Ep});var Fn,Pa,Ep,Oa=V(()=>{"use strict";Pa=(Fn=import.meta.url,async function(e={}){function t(){return de.buffer!=ce.buffer&&Ce(),ce}function r(){return de.buffer!=ce.buffer&&Ce(),q}function n(){return de.buffer!=ce.buffer&&Ce(),ue}function o(){return de.buffer!=ce.buffer&&Ce(),re}function i(){return de.buffer!=ce.buffer&&Ce(),ne}function a(){return de.buffer!=ce.buffer&&Ce(),oe}function d(){return de.buffer!=ce.buffer&&Ce(),U}function l(){return de.buffer!=ce.buffer&&Ce(),Re}var c,m,u=Object.assign({},e),h=new Promise((s,p)=>{c=s,m=p}),w=typeof window=="object",g=typeof importScripts=="function",y=g&&self.name=="em-pthread";u.mountExternalData=(s,p)=>{s.startsWith("./")&&(s=s.substring(2)),(u.Fb||(u.Fb=new Map)).set(s,p)},u.unmountExternalData=()=>{delete u.Fb};var S=globalThis.SharedArrayBuffer??new WebAssembly.Memory({initial:0,maximum:0,shared:!0}).buffer.constructor;let $=()=>{let s=(f,b,_)=>(...I)=>{let O=et,D=b?.();I=f(...I);let L=b?.();return D!==L&&(f=L,_(D),b=_=null),et!=O?new Promise((H,X)=>{En={resolve:H,reject:X}}):I},p=f=>async(...b)=>{try{if(u.Eb)throw Error("Session already started");let _=u.Eb={fc:b[0],errors:[]},I=await f(...b);if(u.Eb!==_)throw Error("Session mismatch");u.Gb?.flush();let O=_.errors;if(0<O.length){let D=await Promise.all(O);if(D=D.filter(L=>L),0<D.length)throw Error(D.join(`
+`))}return I}finally{u.Eb=null}};u._OrtCreateSession=s(u._OrtCreateSession,()=>u._OrtCreateSession,f=>u._OrtCreateSession=f),u._OrtRun=p(s(u._OrtRun,()=>u._OrtRun,f=>u._OrtRun=f)),u._OrtRunWithBinding=p(s(u._OrtRunWithBinding,()=>u._OrtRunWithBinding,f=>u._OrtRunWithBinding=f)),u._OrtBindInput=s(u._OrtBindInput,()=>u._OrtBindInput,f=>u._OrtBindInput=f),$=void 0};u.jsepInit=(s,p)=>{if($?.(),s==="webgpu"){[u.Gb,u.Ub,u.Yb,u.Nb,u.Xb,u.jb,u.Zb,u.bc,u.Vb,u.Wb,u.$b]=p;let f=u.Gb;u.jsepRegisterBuffer=(b,_,I,O)=>f.registerBuffer(b,_,I,O),u.jsepGetBuffer=b=>f.getBuffer(b),u.jsepCreateDownloader=(b,_,I)=>f.createDownloader(b,_,I),u.jsepOnReleaseSession=b=>{f.onReleaseSession(b)},u.jsepOnRunStart=b=>f.onRunStart(b),u.cc=(b,_)=>{f.upload(b,_)}}else if(s==="webnn"){[u.Gb,u.ac,u.Ob,u.jsepEnsureTensor,u.dc,u.jsepDownloadTensor]=p,u.jsepReleaseTensorId=u.Ob;let f=u.Gb;u.jsepOnRunStart=b=>f.onRunStart(b),u.jsepRegisterMLContext=(b,_)=>{f.registerMLContext(b,_)},u.jsepOnReleaseSession=b=>{f.onReleaseSession(b)},u.jsepCreateMLTensorDownloader=(b,_)=>f.createMLTensorDownloader(b,_),u.jsepRegisterMLTensor=(b,_,I)=>f.registerMLTensor(b,_,I)}};var v,x,T=Object.assign({},u),C="./this.program",A=(s,p)=>{throw p},P="";(w||g)&&(g?P=self.location.href:typeof document<"u"&&document.currentScript&&(P=document.currentScript.src),Fn&&(P=Fn),P=P.startsWith("blob:")?"":P.substr(0,P.replace(/[?#].*/,"").lastIndexOf("/")+1),g&&(x=s=>{var p=new XMLHttpRequest;return p.open("GET",s,!1),p.responseType="arraybuffer",p.send(null),new Uint8Array(p.response)}),v=(s,p,f)=>{var b=new XMLHttpRequest;b.open("GET",s,!0),b.responseType="arraybuffer",b.onload=()=>{b.status==200||b.status==0&&b.response?p(b.response):f()},b.onerror=f,b.send(null)});var B,N=console.log.bind(console),W=console.error.bind(console),K=N,Z=W;if(Object.assign(u,T),T=null,y){let s=function(p){try{var f=p.data,b=f.cmd;if(b==="load"){let _=[];self.onmessage=I=>_.push(I),self.startWorker=()=>{postMessage({cmd:"loaded"});for(let I of _)s(I);self.onmessage=s};for(let I of f.handlers)u[I]&&!u[I].proxy||(u[I]=(...O)=>{postMessage({Mb:"callHandler",oc:I,args:O})},I=="print"&&(K=u[I]),I=="printErr"&&(Z=u[I]));de=f.wasmMemory,Ce(),ee(f.wasmModule)}else if(b==="run"){Bn(f.pthread_ptr,0,0,1,0,0),Cn(f.pthread_ptr),ic(),Lo(),se||(Vi(),se=!0);try{ac(f.start_routine,f.arg)}catch(_){if(_!="unwind")throw _}}else b==="cancel"?Rt()&&gr(-1):f.target!=="setimmediate"&&(b==="checkMailbox"?se&&sr():b&&(Z(`worker: received unknown command ${b}`),Z(f)))}catch(_){throw Ni(),_}};var jh=s,ee,se=!1;Z=function(...p){p=p.join(" "),console.error(p)},self.alert=function(...p){postMessage({Mb:"alert",text:p.join(" "),qc:Rt()})},u.instantiateWasm=(p,f)=>new Promise(b=>{ee=_=>{_=new WebAssembly.Instance(_,Ro()),f(_),b()}}),self.onunhandledrejection=p=>{throw p.reason||p},self.onmessage=s}u.wasmBinary&&(B=u.wasmBinary);var de,Y,le,ce,q,ue,re,ne,oe,U,G,ye,Re,$e=!1;function Ce(){var s=de.buffer;u.HEAP8=ce=new Int8Array(s),u.HEAP16=ue=new Int16Array(s),u.HEAPU8=q=new Uint8Array(s),u.HEAPU16=re=new Uint16Array(s),u.HEAP32=ne=new Int32Array(s),u.HEAPU32=oe=new Uint32Array(s),u.HEAPF32=U=new Float32Array(s),u.HEAPF64=Re=new Float64Array(s),u.HEAP64=G=new BigInt64Array(s),u.HEAPU64=ye=new BigUint64Array(s)}if(!y){if(!((de=new WebAssembly.Memory({initial:256,maximum:65536,shared:!0})).buffer instanceof S))throw Z("requested a shared WebAssembly.Memory but the returned buffer is not a SharedArrayBuffer, indicating that while the browser has SharedArrayBuffer it does not have WebAssembly threads support - you may need to set a flag"),Error("bad memory");Ce()}var bt=[],Ae=[],Me=[],Ue=0,zt=null,wt=null;function zo(){if(--Ue==0&&(zt!==null&&(clearInterval(zt),zt=null),wt)){var s=wt;wt=null,s()}}function Ot(s){throw Z(s="Aborted("+s+")"),$e=!0,le=1,s=new WebAssembly.RuntimeError(s+". Build with -sASSERTIONS for more info."),m(s),s}var hn,Oo=s=>s.startsWith("data:application/octet-stream;base64,"),Bo=s=>s.startsWith("file://");function Do(s){if(s==hn&&B)return new Uint8Array(B);if(x)return x(s);throw"both async and sync fetching of the wasm failed"}function Mo(s,p,f){return function(b){if(!B&&(w||g)){if(typeof fetch=="function"&&!Bo(b))return fetch(b,{credentials:"same-origin"}).then(_=>{if(!_.ok)throw`failed to load wasm binary file at '${b}'`;return _.arrayBuffer()}).catch(()=>Do(b));if(v)return new Promise((_,I)=>{v(b,O=>_(new Uint8Array(O)),I)})}return Promise.resolve().then(()=>Do(b))}(s).then(b=>WebAssembly.instantiate(b,p)).then(f,b=>{Z(`failed to asynchronously prepare wasm: ${b}`),Ot(b)})}function Ro(){return{a:{O:oc,Aa:nc,b:uc,aa:qo,B:Yo,qa:Xo,Y:Qo,_:Jo,ra:ei,oa:ti,ha:ri,na:ni,L:oi,Z:ii,W:ai,pa:si,X:ui,wa:dc,F:cc,Q:pc,P:fc,E:gc,u:yc,q:bc,G:wc,A:Ic,R:Cc,ua:Ac,ka:kc,U:Ec,ba:Pc,H:zc,ja:Cn,ta:Oc,t:Bc,x:Rc,o:Uc,l:Nc,c:Tn,n:Wc,j:Gc,w:Fc,p:qc,g:jc,s:Kc,m:Yc,e:Xc,k:Zc,i:Qc,h:Jc,d:ep,ea:tp,fa:rp,ga:np,ca:$i,da:xi,T:op,f:ip,D:ap,I:sp,M:up,y:dp,sa:lp,V:cp,v:Ti,z:pp,N:mp,S:fp,za:hp,ya:gp,la:Ai,ma:ki,$:_n,C:Ei,K:Pi,ia:zi,J:Oi,a:de,xa:wn,va:Mi,r:wp}}}var gn={867620:(s,p,f,b,_)=>{if(u===void 0||!u.Fb)return 1;if((s=ze(s>>>0)).startsWith("./")&&(s=s.substring(2)),!(s=u.Fb.get(s)))return 2;if(b>>>=0,(p>>>=0)+(f>>>=0)>s.byteLength)return 3;try{let I=s.subarray(p,p+f);switch(_){case 0:r().set(I,b>>>0);break;case 1:u.cc(b,I);break;default:return 4}return 0}catch{return 4}},868303:(s,p,f)=>{u.dc(s,r().subarray(p>>>0,p+f>>>0))},868366:()=>u.ac(),868407:s=>{u.Ob(s)},868443:()=>{u.Vb()},868474:()=>{u.Wb()},868503:()=>{u.$b()},868528:s=>u.Ub(s),868561:s=>u.Yb(s),868593:(s,p,f)=>{u.Nb(s,p,f,!0)},868632:(s,p,f)=>{u.Nb(s,p,f)},868665:()=>typeof wasmOffsetConverter<"u",868722:s=>{u.jb("Abs",s,void 0)},868773:s=>{u.jb("Neg",s,void 0)},868824:s=>{u.jb("Floor",s,void 0)},868877:s=>{u.jb("Ceil",s,void 0)},868929:s=>{u.jb("Reciprocal",s,void 0)},868987:s=>{u.jb("Sqrt",s,void 0)},869039:s=>{u.jb("Exp",s,void 0)},869090:s=>{u.jb("Erf",s,void 0)},869141:s=>{u.jb("Sigmoid",s,void 0)},869196:(s,p,f)=>{u.jb("HardSigmoid",s,{alpha:p,beta:f})},869275:s=>{u.jb("Log",s,void 0)},869326:s=>{u.jb("Sin",s,void 0)},869377:s=>{u.jb("Cos",s,void 0)},869428:s=>{u.jb("Tan",s,void 0)},869479:s=>{u.jb("Asin",s,void 0)},869531:s=>{u.jb("Acos",s,void 0)},869583:s=>{u.jb("Atan",s,void 0)},869635:s=>{u.jb("Sinh",s,void 0)},869687:s=>{u.jb("Cosh",s,void 0)},869739:s=>{u.jb("Asinh",s,void 0)},869792:s=>{u.jb("Acosh",s,void 0)},869845:s=>{u.jb("Atanh",s,void 0)},869898:s=>{u.jb("Tanh",s,void 0)},869950:s=>{u.jb("Not",s,void 0)},870001:(s,p,f)=>{u.jb("Clip",s,{min:p,max:f})},870070:s=>{u.jb("Clip",s,void 0)},870122:(s,p)=>{u.jb("Elu",s,{alpha:p})},870180:s=>{u.jb("Gelu",s,void 0)},870232:s=>{u.jb("Relu",s,void 0)},870284:(s,p)=>{u.jb("LeakyRelu",s,{alpha:p})},870348:(s,p)=>{u.jb("ThresholdedRelu",s,{alpha:p})},870418:(s,p)=>{u.jb("Cast",s,{to:p})},870476:s=>{u.jb("Add",s,void 0)},870527:s=>{u.jb("Sub",s,void 0)},870578:s=>{u.jb("Mul",s,void 0)},870629:s=>{u.jb("Div",s,void 0)},870680:s=>{u.jb("Pow",s,void 0)},870731:s=>{u.jb("Equal",s,void 0)},870784:s=>{u.jb("Greater",s,void 0)},870839:s=>{u.jb("GreaterOrEqual",s,void 0)},870901:s=>{u.jb("Less",s,void 0)},870953:s=>{u.jb("LessOrEqual",s,void 0)},871012:(s,p,f,b,_)=>{u.jb("ReduceMean",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,_>>>0)):[]})},871171:(s,p,f,b,_)=>{u.jb("ReduceMax",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,_>>>0)):[]})},871329:(s,p,f,b,_)=>{u.jb("ReduceMin",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,_>>>0)):[]})},871487:(s,p,f,b,_)=>{u.jb("ReduceProd",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,_>>>0)):[]})},871646:(s,p,f,b,_)=>{u.jb("ReduceSum",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,_>>>0)):[]})},871804:(s,p,f,b,_)=>{u.jb("ReduceL1",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,_>>>0)):[]})},871961:(s,p,f,b,_)=>{u.jb("ReduceL2",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,_>>>0)):[]})},872118:(s,p,f,b,_)=>{u.jb("ReduceLogSum",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,_>>>0)):[]})},872279:(s,p,f,b,_)=>{u.jb("ReduceSumSquare",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,_>>>0)):[]})},872443:(s,p,f,b,_)=>{u.jb("ReduceLogSumExp",s,{keepDims:!!p,noopWithEmptyAxes:!!f,axes:b?Array.from(i().subarray(b>>>0,_>>>0)):[]})},872607:s=>{u.jb("Where",s,void 0)},872660:(s,p,f)=>{u.jb("Transpose",s,{perm:p?Array.from(i().subarray(p>>>0,f>>>0)):[]})},872768:(s,p,f,b)=>{u.jb("DepthToSpace",s,{blocksize:p,mode:ze(f),format:b?"NHWC":"NCHW"})},872901:(s,p,f,b)=>{u.jb("DepthToSpace",s,{blocksize:p,mode:ze(f),format:b?"NHWC":"NCHW"})},873034:(s,p,f,b,_,I,O,D,L,H,X,fe,be,z,pe)=>{u.jb("ConvTranspose",s,{format:L?"NHWC":"NCHW",autoPad:p,dilations:[f],group:b,kernelShape:[_],pads:[I,O],strides:[D],wIsConst:()=>!!t()[H>>>0],outputPadding:X?Array.from(i().subarray(X>>>0,fe>>>0)):[],outputShape:be?Array.from(i().subarray(be>>>0,z>>>0)):[],activation:ze(pe)})},873435:(s,p,f,b,_,I,O,D,L,H,X,fe,be,z)=>{u.jb("ConvTranspose",s,{format:D?"NHWC":"NCHW",autoPad:p,dilations:Array.from(i().subarray(f>>>0,2+(f>>>0)>>>0)),group:b,kernelShape:Array.from(i().subarray(_>>>0,2+(_>>>0)>>>0)),pads:Array.from(i().subarray(I>>>0,4+(I>>>0)>>>0)),strides:Array.from(i().subarray(O>>>0,2+(O>>>0)>>>0)),wIsConst:()=>!!t()[L>>>0],outputPadding:H?Array.from(i().subarray(H>>>0,X>>>0)):[],outputShape:fe?Array.from(i().subarray(fe>>>0,be>>>0)):[],activation:ze(z)})},874e3:(s,p,f,b,_,I,O,D,L,H,X,fe,be,z,pe)=>{u.jb("ConvTranspose",s,{format:L?"NHWC":"NCHW",autoPad:p,dilations:[f],group:b,kernelShape:[_],pads:[I,O],strides:[D],wIsConst:()=>!!t()[H>>>0],outputPadding:X?Array.from(i().subarray(X>>>0,fe>>>0)):[],outputShape:be?Array.from(i().subarray(be>>>0,z>>>0)):[],activation:ze(pe)})},874401:(s,p,f,b,_,I,O,D,L,H,X,fe,be,z)=>{u.jb("ConvTranspose",s,{format:D?"NHWC":"NCHW",autoPad:p,dilations:Array.from(i().subarray(f>>>0,2+(f>>>0)>>>0)),group:b,kernelShape:Array.from(i().subarray(_>>>0,2+(_>>>0)>>>0)),pads:Array.from(i().subarray(I>>>0,4+(I>>>0)>>>0)),strides:Array.from(i().subarray(O>>>0,2+(O>>>0)>>>0)),wIsConst:()=>!!t()[L>>>0],outputPadding:H?Array.from(i().subarray(H>>>0,X>>>0)):[],outputShape:fe?Array.from(i().subarray(fe>>>0,be>>>0)):[],activation:ze(z)})},874966:(s,p)=>{u.jb("GlobalAveragePool",s,{format:p?"NHWC":"NCHW"})},875057:(s,p,f,b,_,I,O,D,L,H,X,fe,be,z)=>{u.jb("AveragePool",s,{format:z?"NHWC":"NCHW",auto_pad:p,ceil_mode:f,count_include_pad:b,storage_order:_,dilations:I?Array.from(i().subarray(I>>>0,O>>>0)):[],kernel_shape:D?Array.from(i().subarray(D>>>0,L>>>0)):[],pads:H?Array.from(i().subarray(H>>>0,X>>>0)):[],strides:fe?Array.from(i().subarray(fe>>>0,be>>>0)):[]})},875472:(s,p)=>{u.jb("GlobalAveragePool",s,{format:p?"NHWC":"NCHW"})},875563:(s,p,f,b,_,I,O,D,L,H,X,fe,be,z)=>{u.jb("AveragePool",s,{format:z?"NHWC":"NCHW",auto_pad:p,ceil_mode:f,count_include_pad:b,storage_order:_,dilations:I?Array.from(i().subarray(I>>>0,O>>>0)):[],kernel_shape:D?Array.from(i().subarray(D>>>0,L>>>0)):[],pads:H?Array.from(i().subarray(H>>>0,X>>>0)):[],strides:fe?Array.from(i().subarray(fe>>>0,be>>>0)):[]})},875978:(s,p)=>{u.jb("GlobalMaxPool",s,{format:p?"NHWC":"NCHW"})},876065:(s,p,f,b,_,I,O,D,L,H,X,fe,be,z)=>{u.jb("MaxPool",s,{format:z?"NHWC":"NCHW",auto_pad:p,ceil_mode:f,count_include_pad:b,storage_order:_,dilations:I?Array.from(i().subarray(I>>>0,O>>>0)):[],kernel_shape:D?Array.from(i().subarray(D>>>0,L>>>0)):[],pads:H?Array.from(i().subarray(H>>>0,X>>>0)):[],strides:fe?Array.from(i().subarray(fe>>>0,be>>>0)):[]})},876476:(s,p)=>{u.jb("GlobalMaxPool",s,{format:p?"NHWC":"NCHW"})},876563:(s,p,f,b,_,I,O,D,L,H,X,fe,be,z)=>{u.jb("MaxPool",s,{format:z?"NHWC":"NCHW",auto_pad:p,ceil_mode:f,count_include_pad:b,storage_order:_,dilations:I?Array.from(i().subarray(I>>>0,O>>>0)):[],kernel_shape:D?Array.from(i().subarray(D>>>0,L>>>0)):[],pads:H?Array.from(i().subarray(H>>>0,X>>>0)):[],strides:fe?Array.from(i().subarray(fe>>>0,be>>>0)):[]})},876974:(s,p,f,b,_)=>{u.jb("Gemm",s,{alpha:p,beta:f,transA:b,transB:_})},877078:s=>{u.jb("MatMul",s,void 0)},877132:(s,p,f,b)=>{u.jb("ArgMax",s,{keepDims:!!p,selectLastIndex:!!f,axis:b})},877240:(s,p,f,b)=>{u.jb("ArgMin",s,{keepDims:!!p,selectLastIndex:!!f,axis:b})},877348:(s,p)=>{u.jb("Softmax",s,{axis:p})},877411:(s,p)=>{u.jb("Concat",s,{axis:p})},877471:(s,p,f,b,_)=>{u.jb("Split",s,{axis:p,numOutputs:f,splitSizes:b?Array.from(i().subarray(b>>>0,_>>>0)):[]})},877611:s=>{u.jb("Expand",s,void 0)},877665:(s,p)=>{u.jb("Gather",s,{axis:Number(p)})},877736:(s,p)=>{u.jb("GatherElements",s,{axis:Number(p)})},877815:(s,p,f,b,_,I,O,D,L,H,X)=>{u.jb("Resize",s,{antialias:p,axes:f?Array.from(i().subarray(f>>>0,b>>>0)):[],coordinateTransformMode:ze(_),cubicCoeffA:I,excludeOutside:O,extrapolationValue:D,keepAspectRatioPolicy:ze(L),mode:ze(H),nearestMode:ze(X)})},878161:(s,p,f,b,_,I,O)=>{u.jb("Slice",s,{starts:p?Array.from(i().subarray(p>>>0,f>>>0)):[],ends:b?Array.from(i().subarray(b>>>0,_>>>0)):[],axes:I?Array.from(i().subarray(I>>>0,O>>>0)):[]})},878377:s=>{u.jb("Tile",s,void 0)},878429:(s,p,f)=>{u.jb("InstanceNormalization",s,{epsilon:p,format:f?"NHWC":"NCHW"})},878543:(s,p,f)=>{u.jb("InstanceNormalization",s,{epsilon:p,format:f?"NHWC":"NCHW"})},878657:s=>{u.jb("Range",s,void 0)},878710:(s,p)=>{u.jb("Einsum",s,{equation:ze(p)})},878791:(s,p,f,b,_)=>{u.jb("Pad",s,{mode:p,value:f,pads:b?Array.from(i().subarray(b>>>0,_>>>0)):[]})},878918:(s,p,f,b,_,I)=>{u.jb("BatchNormalization",s,{epsilon:p,momentum:f,spatial:!!_,trainingMode:!!b,format:I?"NHWC":"NCHW"})},879087:(s,p,f,b,_,I)=>{u.jb("BatchNormalization",s,{epsilon:p,momentum:f,spatial:!!_,trainingMode:!!b,format:I?"NHWC":"NCHW"})},879256:(s,p,f)=>{u.jb("CumSum",s,{exclusive:Number(p),reverse:Number(f)})},879353:(s,p,f)=>{u.jb("DequantizeLinear",s,{axis:p,blockSize:f})},879443:(s,p,f,b,_,I,O,D,L)=>{u.jb("Attention",s,{numHeads:p,isUnidirectional:f,maskFilterValue:b,scale:_,doRotary:I,qkvHiddenSizes:O?Array.from(i().subarray(Number(D)>>>0,Number(D)+O>>>0)):[],pastPresentShareBuffer:!!L})},879715:s=>{u.jb("BiasAdd",s,void 0)},879770:s=>{u.jb("BiasSplitGelu",s,void 0)},879831:s=>{u.jb("FastGelu",s,void 0)},879887:(s,p,f,b,_,I,O,D,L,H,X,fe,be,z,pe,Se)=>{u.jb("Conv",s,{format:fe?"NHWC":"NCHW",auto_pad:p,dilations:f?Array.from(i().subarray(f>>>0,b>>>0)):[],group:_,kernel_shape:I?Array.from(i().subarray(I>>>0,O>>>0)):[],pads:D?Array.from(i().subarray(D>>>0,L>>>0)):[],strides:H?Array.from(i().subarray(H>>>0,X>>>0)):[],w_is_const:()=>!!t()[be>>>0],activation:ze(z),activation_params:pe?Array.from(d().subarray(pe>>>0,Se>>>0)):[]})},880383:s=>{u.jb("Gelu",s,void 0)},880435:(s,p,f,b)=>{u.jb("GroupQueryAttention",s,{numHeads:p,kvNumHeads:f,scale:b})},880548:(s,p,f,b)=>{u.jb("LayerNormalization",s,{axis:p,epsilon:f,simplified:!!b})},880659:(s,p,f,b)=>{u.jb("LayerNormalization",s,{axis:p,epsilon:f,simplified:!!b})},880770:(s,p,f,b,_,I)=>{u.jb("MatMulNBits",s,{k:p,n:f,accuracyLevel:b,bits:_,blockSize:I})},880897:(s,p,f,b,_,I)=>{u.jb("MultiHeadAttention",s,{numHeads:p,isUnidirectional:f,maskFilterValue:b,scale:_,doRotary:I})},881056:(s,p)=>{u.jb("QuickGelu",s,{alpha:p})},881120:(s,p,f,b,_)=>{u.jb("RotaryEmbedding",s,{interleaved:!!p,numHeads:f,rotaryEmbeddingDim:b,scale:_})},881259:(s,p,f)=>{u.jb("SkipLayerNormalization",s,{epsilon:p,simplified:!!f})},881361:(s,p,f)=>{u.jb("SkipLayerNormalization",s,{epsilon:p,simplified:!!f})},881463:(s,p,f,b)=>{u.jb("GatherBlockQuantized",s,{gatherAxis:p,quantizeAxis:f,blockSize:b})},881584:s=>{u.Zb(s)},881618:(s,p)=>u.bc(s,p,u.Eb.fc,u.Eb.errors)};function nc(s,p,f){return yi(async()=>{await u.Xb(s,p,f)})}function oc(){return typeof wasmOffsetConverter<"u"}function yn(s){this.name="ExitStatus",this.message=`Program terminated with exit(${s})`,this.status=s}var bn=s=>{s.terminate(),s.onmessage=()=>{}},Uo=s=>{pt.length==0&&(Go(),Ho(pt[0]));var p=pt.pop();if(!p)return 6;vt.push(p),Qe[s.Ab]=p,p.Ab=s.Ab;var f={cmd:"run",start_routine:s.hc,arg:s.Qb,pthread_ptr:s.Ab};return p.postMessage(f,s.mc),0},_t=0,xe=(s,p,...f)=>{for(var b=2*f.length,_=Rn(),I=Mn(8*b),O=I>>>3,D=0;D<f.length;D++){var L=f[D];typeof L=="bigint"?(G[O+2*D]=1n,G[O+2*D+1]=L):(G[O+2*D]=0n,l()[O+2*D+1>>>0]=L)}return s=Wi(s,0,b,I,p),yr(_),s};function wn(s){if(y)return xe(0,1,s);if(le=s,!(0<_t)){for(var p of vt)bn(p);for(p of pt)bn(p);pt=[],vt=[],Qe=[],$e=!0}A(s,new yn(s))}function Vo(s){if(y)return xe(1,0,s);_n(s)}var _n=s=>{if(le=s,y)throw Vo(s),"unwind";wn(s)},pt=[],vt=[],No=[],Qe={},Wo=s=>{var p=s.Ab;delete Qe[p],pt.push(s),vt.splice(vt.indexOf(s),1),s.Ab=0,Dn(p)};function Lo(){No.forEach(s=>s())}var Ho=s=>new Promise(p=>{s.onmessage=_=>{var I=(_=_.data).cmd;if(_.targetThread&&_.targetThread!=Rt()){var O=Qe[_.targetThread];O?O.postMessage(_,_.transferList):Z(`Internal error! Worker sent a message "${I}" to target pthread ${_.targetThread}, but that thread no longer exists!`)}else I==="checkMailbox"?sr():I==="spawnThread"?Uo(_):I==="cleanupThread"?Wo(Qe[_.thread]):I==="killThread"?(_=_.thread,I=Qe[_],delete Qe[_],bn(I),Dn(_),vt.splice(vt.indexOf(I),1),I.Ab=0):I==="cancelThread"?Qe[_.thread].postMessage({cmd:"cancel"}):I==="loaded"?(s.loaded=!0,p(s)):I==="alert"?alert(`Thread ${_.threadId}: ${_.text}`):_.target==="setimmediate"?s.postMessage(_):I==="callHandler"?u[_.handler](..._.args):I&&Z(`worker sent an unknown command ${I}`)},s.onerror=_=>{throw Z(`worker sent an error! ${_.filename}:${_.lineno}: ${_.message}`),_};var f,b=[];for(f of[])u.hasOwnProperty(f)&&b.push(f);s.postMessage({cmd:"load",handlers:b,wasmMemory:de,wasmModule:Y})});function Go(){var s=new Worker(new URL(import.meta.url),{type:"module",workerData:"em-pthread",name:"em-pthread"});pt.push(s)}var ar=s=>{for(;0<s.length;)s.shift()(u)},ic=()=>{var s=Rt(),p=a()[s+52>>>2>>>0];s=a()[s+56>>>2>>>0],Hi(p,p-s),yr(p)},ac=(s,p)=>{_t=0,s=Gi(s,p),0<_t?le=s:gr(s)};class sc{constructor(p){this.Jb=p-24}}function uc(s,p,f){var b=new sc(s>>>=0);throw p>>>=0,f>>>=0,a()[b.Jb+16>>>2>>>0]=0,a()[b.Jb+4>>>2>>>0]=p,a()[b.Jb+8>>>2>>>0]=f,s}function Fo(s,p,f,b){return y?xe(2,1,s,p,f,b):qo(s,p,f,b)}function qo(s,p,f,b){if(s>>>=0,p>>>=0,f>>>=0,b>>>=0,S===void 0)return Z("Current environment does not support SharedArrayBuffer, pthreads are not available!"),6;var _=[];return y&&_.length===0?Fo(s,p,f,b):(s={hc:f,Ab:s,Qb:b,mc:_},y?(s.Mb="spawnThread",postMessage(s,_),0):Uo(s))}var jo=typeof TextDecoder<"u"?new TextDecoder("utf8"):void 0,Ko=(s,p,f)=>{var b=(p>>>=0)+f;for(f=p;s[f]&&!(f>=b);)++f;if(16<f-p&&s.buffer&&jo)return jo.decode(s.buffer instanceof S?s.slice(p,f):s.subarray(p,f));for(b="";p<f;){var _=s[p++];if(128&_){var I=63&s[p++];if((224&_)==192)b+=String.fromCharCode((31&_)<<6|I);else{var O=63&s[p++];65536>(_=(240&_)==224?(15&_)<<12|I<<6|O:(7&_)<<18|I<<12|O<<6|63&s[p++])?b+=String.fromCharCode(_):(_-=65536,b+=String.fromCharCode(55296|_>>10,56320|1023&_))}}else b+=String.fromCharCode(_)}return b},ze=(s,p)=>(s>>>=0)?Ko(r(),s,p):"";function Yo(s,p,f){return y?xe(3,1,s,p,f):0}function Xo(s,p){if(y)return xe(4,1,s,p)}var vn=s=>{for(var p=0,f=0;f<s.length;++f){var b=s.charCodeAt(f);127>=b?p++:2047>=b?p+=2:55296<=b&&57343>=b?(p+=4,++f):p+=3}return p},Zo=(s,p,f,b)=>{if(!(0<b))return 0;var _=f>>>=0;b=f+b-1;for(var I=0;I<s.length;++I){var O=s.charCodeAt(I);if(55296<=O&&57343>=O&&(O=65536+((1023&O)<<10)|1023&s.charCodeAt(++I)),127>=O){if(f>=b)break;p[f++>>>0]=O}else{if(2047>=O){if(f+1>=b)break;p[f++>>>0]=192|O>>6}else{if(65535>=O){if(f+2>=b)break;p[f++>>>0]=224|O>>12}else{if(f+3>=b)break;p[f++>>>0]=240|O>>18,p[f++>>>0]=128|O>>12&63}p[f++>>>0]=128|O>>6&63}p[f++>>>0]=128|63&O}}return p[f>>>0]=0,f-_},Bt=(s,p,f)=>Zo(s,r(),p,f);function Qo(s,p){if(y)return xe(5,1,s,p)}function Jo(s,p,f){if(y)return xe(6,1,s,p,f)}function ei(s,p,f){return y?xe(7,1,s,p,f):0}function ti(s,p){if(y)return xe(8,1,s,p)}function ri(s,p,f){if(y)return xe(9,1,s,p,f)}function ni(s,p,f,b){if(y)return xe(10,1,s,p,f,b)}function oi(s,p,f,b){if(y)return xe(11,1,s,p,f,b)}function ii(s,p,f,b){if(y)return xe(12,1,s,p,f,b)}function ai(s){if(y)return xe(13,1,s)}function si(s,p){if(y)return xe(14,1,s,p)}function ui(s,p,f){if(y)return xe(15,1,s,p,f)}var di,mt,dc=()=>{Ot("")},Je=s=>{for(var p="";r()[s>>>0];)p+=di[r()[s++>>>0]];return p},$n={},xn={},lc={};function ut(s,p,f={}){if(!("argPackAdvance"in p))throw new TypeError("registerType registeredInstance requires argPackAdvance");return function(b,_,I={}){var O=_.name;if(!b)throw new mt(`type "${O}" must have a positive integer typeid pointer`);if(xn.hasOwnProperty(b)){if(I.Sb)return;throw new mt(`Cannot register type '${O}' twice`)}xn[b]=_,delete lc[b],$n.hasOwnProperty(b)&&(_=$n[b],delete $n[b],_.forEach(D=>D()))}(s,p,f)}var li=(s,p,f)=>{switch(p){case 1:return f?b=>t()[b>>>0]:b=>r()[b>>>0];case 2:return f?b=>n()[b>>>1>>>0]:b=>o()[b>>>1>>>0];case 4:return f?b=>i()[b>>>2>>>0]:b=>a()[b>>>2>>>0];case 8:return f?b=>G[b>>>3]:b=>ye[b>>>3];default:throw new TypeError(`invalid integer width (${p}): ${s}`)}};function cc(s,p,f){f>>>=0,ut(s>>>=0,{name:p=Je(p>>>0),fromWireType:b=>b,toWireType:function(b,_){if(typeof _!="bigint"&&typeof _!="number")throw _=_===null?"null":(b=typeof _)=="object"||b==="array"||b==="function"?_.toString():""+_,new TypeError(`Cannot convert "${_}" to ${this.name}`);return typeof _=="number"&&(_=BigInt(_)),_},argPackAdvance:ft,readValueFromPointer:li(p,f,p.indexOf("u")==-1),Db:null})}var ft=8;function pc(s,p,f,b){ut(s>>>=0,{name:p=Je(p>>>0),fromWireType:function(_){return!!_},toWireType:function(_,I){return I?f:b},argPackAdvance:ft,readValueFromPointer:function(_){return this.fromWireType(r()[_>>>0])},Db:null})}var Sn=[],dt=[];function Tn(s){9<(s>>>=0)&&--dt[s+1]==0&&(dt[s]=void 0,Sn.push(s))}var qe=s=>{if(!s)throw new mt("Cannot use deleted val. handle = "+s);return dt[s]},je=s=>{switch(s){case void 0:return 2;case null:return 4;case!0:return 6;case!1:return 8;default:let p=Sn.pop()||dt.length;return dt[p]=s,dt[p+1]=1,p}};function In(s){return this.fromWireType(a()[s>>>2>>>0])}var mc={name:"emscripten::val",fromWireType:s=>{var p=qe(s);return Tn(s),p},toWireType:(s,p)=>je(p),argPackAdvance:ft,readValueFromPointer:In,Db:null};function fc(s){return ut(s>>>0,mc)}var hc=(s,p)=>{switch(p){case 4:return function(f){return this.fromWireType(d()[f>>>2>>>0])};case 8:return function(f){return this.fromWireType(l()[f>>>3>>>0])};default:throw new TypeError(`invalid float width (${p}): ${s}`)}};function gc(s,p,f){f>>>=0,ut(s>>>=0,{name:p=Je(p>>>0),fromWireType:b=>b,toWireType:(b,_)=>_,argPackAdvance:ft,readValueFromPointer:hc(p,f),Db:null})}function yc(s,p,f,b,_){if(s>>>=0,f>>>=0,p=Je(p>>>0),_===-1&&(_=4294967295),_=D=>D,b===0){var I=32-8*f;_=D=>D<<I>>>I}var O=p.includes("unsigned")?function(D,L){return L>>>0}:function(D,L){return L};ut(s,{name:p,fromWireType:_,toWireType:O,argPackAdvance:ft,readValueFromPointer:li(p,f,b!==0),Db:null})}function bc(s,p,f){function b(I){var O=a()[I>>>2>>>0];return I=a()[I+4>>>2>>>0],new _(t().buffer,I,O)}var _=[Int8Array,Uint8Array,Int16Array,Uint16Array,Int32Array,Uint32Array,Float32Array,Float64Array,BigInt64Array,BigUint64Array][p];ut(s>>>=0,{name:f=Je(f>>>0),fromWireType:b,argPackAdvance:ft,readValueFromPointer:b},{Sb:!0})}function wc(s,p){s>>>=0;var f=(p=Je(p>>>0))==="std::string";ut(s,{name:p,fromWireType:function(b){var _=a()[b>>>2>>>0],I=b+4;if(f)for(var O=I,D=0;D<=_;++D){var L=I+D;if(D==_||r()[L>>>0]==0){if(O=ze(O,L-O),H===void 0)var H=O;else H+=String.fromCharCode(0),H+=O;O=L+1}}else{for(H=Array(_),D=0;D<_;++D)H[D]=String.fromCharCode(r()[I+D>>>0]);H=H.join("")}return tt(b),H},toWireType:function(b,_){_ instanceof ArrayBuffer&&(_=new Uint8Array(_));var I=typeof _=="string";if(!(I||_ instanceof Uint8Array||_ instanceof Uint8ClampedArray||_ instanceof Int8Array))throw new mt("Cannot pass non-string to std::string");var O=f&&I?vn(_):_.length,D=hr(4+O+1),L=D+4;if(a()[D>>>2>>>0]=O,f&&I)Bt(_,L,O+1);else if(I)for(I=0;I<O;++I){var H=_.charCodeAt(I);if(255<H)throw tt(L),new mt("String has UTF-16 code units that do not fit in 8 bits");r()[L+I>>>0]=H}else for(I=0;I<O;++I)r()[L+I>>>0]=_[I];return b!==null&&b.push(tt,D),D},argPackAdvance:ft,readValueFromPointer:In,Db(b){tt(b)}})}var ci=typeof TextDecoder<"u"?new TextDecoder("utf-16le"):void 0,_c=(s,p)=>{for(var f=s>>1,b=f+p/2;!(f>=b)&&o()[f>>>0];)++f;if(32<(f<<=1)-s&&ci)return ci.decode(r().slice(s,f));for(f="",b=0;!(b>=p/2);++b){var _=n()[s+2*b>>>1>>>0];if(_==0)break;f+=String.fromCharCode(_)}return f},vc=(s,p,f)=>{if(f??=2147483647,2>f)return 0;var b=p;f=(f-=2)<2*s.length?f/2:s.length;for(var _=0;_<f;++_){var I=s.charCodeAt(_);n()[p>>>1>>>0]=I,p+=2}return n()[p>>>1>>>0]=0,p-b},$c=s=>2*s.length,xc=(s,p)=>{for(var f=0,b="";!(f>=p/4);){var _=i()[s+4*f>>>2>>>0];if(_==0)break;++f,65536<=_?(_-=65536,b+=String.fromCharCode(55296|_>>10,56320|1023&_)):b+=String.fromCharCode(_)}return b},Sc=(s,p,f)=>{if(p>>>=0,f??=2147483647,4>f)return 0;var b=p;f=b+f-4;for(var _=0;_<s.length;++_){var I=s.charCodeAt(_);if(55296<=I&&57343>=I&&(I=65536+((1023&I)<<10)|1023&s.charCodeAt(++_)),i()[p>>>2>>>0]=I,(p+=4)+4>f)break}return i()[p>>>2>>>0]=0,p-b},Tc=s=>{for(var p=0,f=0;f<s.length;++f){var b=s.charCodeAt(f);55296<=b&&57343>=b&&++f,p+=4}return p};function Ic(s,p,f){if(s>>>=0,p>>>=0,f=Je(f>>>=0),p===2)var b=_c,_=vc,I=$c,O=D=>o()[D>>>1>>>0];else p===4&&(b=xc,_=Sc,I=Tc,O=D=>a()[D>>>2>>>0]);ut(s,{name:f,fromWireType:D=>{for(var L,H=a()[D>>>2>>>0],X=D+4,fe=0;fe<=H;++fe){var be=D+4+fe*p;fe!=H&&O(be)!=0||(X=b(X,be-X),L===void 0?L=X:(L+=String.fromCharCode(0),L+=X),X=be+p)}return tt(D),L},toWireType:(D,L)=>{if(typeof L!="string")throw new mt(`Cannot pass non-string to C++ string type ${f}`);var H=I(L),X=hr(4+H+p);return a()[X>>>2>>>0]=H/p,_(L,X+4,H+p),D!==null&&D.push(tt,X),X},argPackAdvance:ft,readValueFromPointer:In,Db(D){tt(D)}})}function Cc(s,p){ut(s>>>=0,{Tb:!0,name:p=Je(p>>>0),argPackAdvance:0,fromWireType:()=>{},toWireType:()=>{}})}var Ac=()=>1;function kc(s){Bn(s>>>0,!g,1,!w,131072,!1),Lo()}var pi=s=>{if(!$e)try{if(s(),!(0<_t))try{y?gr(le):_n(le)}catch(p){p instanceof yn||p=="unwind"||A(1,p)}}catch(p){p instanceof yn||p=="unwind"||A(1,p)}};function Cn(s){s>>>=0,typeof Atomics.nc=="function"&&(Atomics.nc(i(),s>>>2,s).value.then(sr),s+=128,Atomics.store(i(),s>>>2,1))}var sr=()=>{var s=Rt();s&&(Cn(s),pi(Li))};function Ec(s,p){(s>>>=0)==p>>>0?setTimeout(sr):y?postMessage({targetThread:s,cmd:"checkMailbox"}):(s=Qe[s])&&s.postMessage({cmd:"checkMailbox"})}var An=[];function Pc(s,p,f,b,_){for(p>>>=0,b/=2,An.length=b,f=_>>>0>>>3,_=0;_<b;_++)An[_]=G[f+2*_]?G[f+2*_+1]:l()[f+2*_+1>>>0];return(p?gn[p]:_p[s])(...An)}function zc(s){s>>>=0,y?postMessage({cmd:"cleanupThread",thread:s}):Wo(Qe[s])}function Oc(s){}var kn=(s,p)=>{var f=xn[s];if(f===void 0)throw s=Ui(s),f=Je(s),tt(s),new mt(`${p} has unknown type ${f}`);return f},mi=(s,p,f)=>{var b=[];return s=s.toWireType(b,f),b.length&&(a()[p>>>2>>>0]=je(b)),s};function Bc(s,p,f){return p>>>=0,f>>>=0,s=qe(s>>>0),p=kn(p,"emval::as"),mi(p,f,s)}var ur=s=>{try{s()}catch(p){Ot(p)}},ht=0,et=null,fi=0,dr=[],hi={},gi={},Dc=0,En=null,Mc=[];function yi(s){return function(p){if(!$e){if(ht===0){var f=!1,b=!1;p((_=0)=>{if(!$e&&(fi=_,f=!0,b)){ht=2,ur(()=>ji(et)),typeof Browser<"u"&&Browser.Kb.Rb&&Browser.Kb.resume(),_=!1;try{var I=function(){var L=i()[et+8>>>2>>>0];return L=j[gi[L]],--_t,L()}()}catch(L){I=L,_=!0}var O=!1;if(!et){var D=En;D&&(En=null,(_?D.reject:D.resolve)(I),O=!0)}if(_&&!O)throw I}}),b=!0,f||(ht=1,et=function(){var _=hr(65548),I=_+12;a()[_>>>2>>>0]=I,a()[_+4>>>2>>>0]=I+65536,I=dr[0];var O=hi[I];return O===void 0&&(O=Dc++,hi[I]=O,gi[O]=I),I=O,i()[_+8>>>2>>>0]=I,_}(),typeof Browser<"u"&&Browser.Kb.Rb&&Browser.Kb.pause(),ur(()=>Fi(et)))}else ht===2?(ht=0,ur(Ki),tt(et),et=null,Mc.forEach(pi)):Ot(`invalid state: ${ht}`);return fi}}(p=>{s().then(p)})}function Rc(s){return s>>>=0,yi(()=>(s=qe(s)).then(je))}var lr=[];function Uc(s,p,f,b){return f>>>=0,b>>>=0,(s=lr[s>>>0])(null,p=qe(p>>>0),f,b)}var Vc={},cr=s=>{var p=Vc[s];return p===void 0?Je(s):p};function Nc(s,p,f,b,_){return f>>>=0,b>>>=0,_>>>=0,(s=lr[s>>>0])(p=qe(p>>>0),p[f=cr(f)],b,_)}var bi=()=>typeof globalThis=="object"?globalThis:Function("return this")();function Wc(s){return(s>>>=0)==0?je(bi()):(s=cr(s),je(bi()[s]))}var Lc=s=>{var p=lr.length;return lr.push(s),p},Hc=(s,p)=>{for(var f=Array(s),b=0;b<s;++b)f[b]=kn(a()[p+4*b>>>2>>>0],"parameter "+b);return f},wi=(s,p)=>Object.defineProperty(p,"name",{value:s});function Gc(s,p,f){var b=(p=Hc(s,p>>>0)).shift();s--;var _=`return function (obj, func, destructorsRef, args) {
+`,I=0,O=[];f===0&&O.push("obj");for(var D=["retType"],L=[b],H=0;H<s;++H)O.push("arg"+H),D.push("argType"+H),L.push(p[H]),_+=`  var arg${H} = argType${H}.readValueFromPointer(args${I?"+"+I:""});
+`,I+=p[H].argPackAdvance;return _+=`  var rv = ${f===1?"new func":"func.call"}(${O.join(", ")});
+`,b.Tb||(D.push("emval_returnValue"),L.push(mi),_+=`  return emval_returnValue(retType, destructorsRef, rv);
+`),D.push(_+`};
+`),s=function(X){var fe=Function;if(!(fe instanceof Function))throw new TypeError(`new_ called with constructor type ${typeof fe} which is not a function`);var be=wi(fe.name||"unknownFunctionName",function(){});return be.prototype=fe.prototype,be=new be,(X=fe.apply(be,X))instanceof Object?X:be}(D)(...L),f=`methodCaller<(${p.map(X=>X.name).join(", ")}) => ${b.name}>`,Lc(wi(f,s))}function Fc(s){return s=cr(s>>>0),je(u[s])}function qc(s,p){return p>>>=0,s=qe(s>>>0),p=qe(p),je(s[p])}function jc(s){9<(s>>>=0)&&(dt[s+1]+=1)}function Kc(){return je([])}function Yc(s){s=qe(s>>>0);for(var p=Array(s.length),f=0;f<s.length;f++)p[f]=s[f];return je(p)}function Xc(s){return je(cr(s>>>0))}function Zc(){return je({})}function Qc(s){for(var p=qe(s>>>=0);p.length;){var f=p.pop();p.pop()(f)}Tn(s)}function Jc(s,p,f){p>>>=0,f>>>=0,s=qe(s>>>0),p=qe(p),f=qe(f),s[p]=f}function ep(s,p){return p>>>=0,s=(s=kn(s>>>0,"_emval_take_value")).readValueFromPointer(p),je(s)}function tp(s,p){s=-9007199254740992>s||9007199254740992<s?NaN:Number(s),p>>>=0,s=new Date(1e3*s),i()[p>>>2>>>0]=s.getUTCSeconds(),i()[p+4>>>2>>>0]=s.getUTCMinutes(),i()[p+8>>>2>>>0]=s.getUTCHours(),i()[p+12>>>2>>>0]=s.getUTCDate(),i()[p+16>>>2>>>0]=s.getUTCMonth(),i()[p+20>>>2>>>0]=s.getUTCFullYear()-1900,i()[p+24>>>2>>>0]=s.getUTCDay(),s=(s.getTime()-Date.UTC(s.getUTCFullYear(),0,1,0,0,0,0))/864e5|0,i()[p+28>>>2>>>0]=s}var Dt=s=>s%4==0&&(s%100!=0||s%400==0),_i=[0,31,60,91,121,152,182,213,244,274,305,335],vi=[0,31,59,90,120,151,181,212,243,273,304,334];function rp(s,p){s=-9007199254740992>s||9007199254740992<s?NaN:Number(s),p>>>=0,s=new Date(1e3*s),i()[p>>>2>>>0]=s.getSeconds(),i()[p+4>>>2>>>0]=s.getMinutes(),i()[p+8>>>2>>>0]=s.getHours(),i()[p+12>>>2>>>0]=s.getDate(),i()[p+16>>>2>>>0]=s.getMonth(),i()[p+20>>>2>>>0]=s.getFullYear()-1900,i()[p+24>>>2>>>0]=s.getDay();var f=(Dt(s.getFullYear())?_i:vi)[s.getMonth()]+s.getDate()-1|0;i()[p+28>>>2>>>0]=f,i()[p+36>>>2>>>0]=-60*s.getTimezoneOffset(),f=new Date(s.getFullYear(),6,1).getTimezoneOffset();var b=new Date(s.getFullYear(),0,1).getTimezoneOffset();s=0|(f!=b&&s.getTimezoneOffset()==Math.min(b,f)),i()[p+32>>>2>>>0]=s}function np(s){s>>>=0;var p=new Date(i()[s+20>>>2>>>0]+1900,i()[s+16>>>2>>>0],i()[s+12>>>2>>>0],i()[s+8>>>2>>>0],i()[s+4>>>2>>>0],i()[s>>>2>>>0],0),f=i()[s+32>>>2>>>0],b=p.getTimezoneOffset(),_=new Date(p.getFullYear(),6,1).getTimezoneOffset(),I=new Date(p.getFullYear(),0,1).getTimezoneOffset(),O=Math.min(I,_);return 0>f?i()[s+32>>>2>>>0]=+(_!=I&&O==b):0<f!=(O==b)&&(_=Math.max(I,_),p.setTime(p.getTime()+6e4*((0<f?O:_)-b))),i()[s+24>>>2>>>0]=p.getDay(),f=(Dt(p.getFullYear())?_i:vi)[p.getMonth()]+p.getDate()-1|0,i()[s+28>>>2>>>0]=f,i()[s>>>2>>>0]=p.getSeconds(),i()[s+4>>>2>>>0]=p.getMinutes(),i()[s+8>>>2>>>0]=p.getHours(),i()[s+12>>>2>>>0]=p.getDate(),i()[s+16>>>2>>>0]=p.getMonth(),i()[s+20>>>2>>>0]=p.getYear(),s=p.getTime(),BigInt(isNaN(s)?-1:s/1e3)}function $i(s,p,f,b,_,I,O){return y?xe(16,1,s,p,f,b,_,I,O):-52}function xi(s,p,f,b,_,I){if(y)return xe(17,1,s,p,f,b,_,I)}function op(s,p,f,b){s>>>=0,p>>>=0,f>>>=0,b>>>=0;var _=new Date().getFullYear(),I=new Date(_,0,1),O=new Date(_,6,1);_=I.getTimezoneOffset();var D=O.getTimezoneOffset(),L=Math.max(_,D);a()[s>>>2>>>0]=60*L,i()[p>>>2>>>0]=+(_!=D),I=(s=H=>H.toLocaleTimeString(void 0,{hour12:!1,timeZoneName:"short"}).split(" ")[1])(I),O=s(O),D<_?(Bt(I,f,17),Bt(O,b,17)):(Bt(I,b,17),Bt(O,f,17))}var Pn=[],Si=(s,p)=>{Pn.length=0;for(var f;f=r()[s++>>>0];){var b=f!=105;p+=(b&=f!=112)&&p%8?4:0,Pn.push(f==112?a()[p>>>2>>>0]:f==106?G[p>>>3]:f==105?i()[p>>>2>>>0]:l()[p>>>3>>>0]),p+=b?8:4}return Pn};function ip(s,p,f){return s>>>=0,p=Si(p>>>0,f>>>0),gn[s](...p)}function ap(s,p,f){return s>>>=0,p=Si(p>>>0,f>>>0),gn[s](...p)}var sp=()=>{},up=()=>Date.now();function dp(s,p){return Z(ze(s>>>0,p>>>0))}var Ti,lp=()=>{throw _t+=1,"unwind"};function cp(){return 4294901760}Ti=()=>performance.timeOrigin+performance.now();var pp=()=>navigator.hardwareConcurrency;function mp(){return Ot("Cannot use emscripten_pc_get_function without -sUSE_OFFSET_CONVERTER"),0}function fp(s){s>>>=0;var p=r().length;if(s<=p||4294901760<s)return!1;for(var f=1;4>=f;f*=2){var b=p*(1+.2/f);b=Math.min(b,s+100663296);var _=Math;b=Math.max(s,b);e:{_=(_.min.call(_,4294901760,b+(65536-b%65536)%65536)-de.buffer.byteLength+65535)/65536;try{de.grow(_),Ce();var I=1;break e}catch{}I=void 0}if(I)return!0}return!1}var pr=()=>(Ot("Cannot use convertFrameToPC (needed by __builtin_return_address) without -sUSE_OFFSET_CONVERTER"),0),Mt={},Ii=s=>{s.forEach(p=>{var f=pr();f&&(Mt[f]=p)})};function hp(){var s=Error().stack.toString().split(`
 `);return s[0]=="Error"&&s.shift(),Ii(s),Mt.Pb=pr(),Mt.ec=s,Mt.Pb}function gp(s,p,f){if(s>>>=0,p>>>=0,Mt.Pb==s)var b=Mt.ec;else(b=Error().stack.toString().split(`
-`))[0]=="Error"&&b.shift(),Ii(b);for(var v=3;b[v]&&pr()!=s;)++v;for(s=0;s<f&&b[s+v];++s)i()[p+4*s>>>2>>>0]=pr();return s}var On,zn={},Ci=()=>{if(!On){var s,p={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",_:C||"./this.program"};for(s in zn)zn[s]===void 0?delete p[s]:p[s]=zn[s];var f=[];for(s in p)f.push(`${s}=${p[s]}`);On=f}return On};function Ai(s,p){if(y)return xe(18,1,s,p);s>>>=0,p>>>=0;var f=0;return Ci().forEach((b,v)=>{var I=p+f;for(v=a()[s+4*v>>>2>>>0]=I,I=0;I<b.length;++I)t()[v++>>>0]=b.charCodeAt(I);t()[v>>>0]=0,f+=b.length+1}),0}function Ei(s,p){if(y)return xe(19,1,s,p);s>>>=0,p>>>=0;var f=Ci();a()[s>>>2>>>0]=f.length;var b=0;return f.forEach(v=>b+=v.length+1),a()[p>>>2>>>0]=b,0}function ki(s){return y?xe(20,1,s):52}function Pi(s,p,f,b){return y?xe(21,1,s,p,f,b):52}function Oi(s,p,f,b){return y?xe(22,1,s,p,f,b):70}var yp=[null,[],[]];function zi(s,p,f,b){if(y)return xe(23,1,s,p,f,b);p>>>=0,f>>>=0,b>>>=0;for(var v=0,I=0;I<f;I++){var z=a()[p>>>2>>>0],B=a()[p+4>>>2>>>0];p+=8;for(var W=0;W<B;W++){var H=r()[z+W>>>0],j=yp[s];H===0||H===10?((s===1?K:X)(Ko(j,0)),j.length=0):j.push(H)}v+=B}return a()[b>>>2>>>0]=v,0}var Di=[31,29,31,30,31,30,31,31,30,31,30,31],Bi=[31,28,31,30,31,30,31,31,30,31,30,31],bp=(s,p)=>{t().set(s,p>>>0)};function Mi(s,p,f,b){function v(O,ue,Se){for(O=typeof O=="number"?O.toString():O||"";O.length<ue;)O=Se[0]+O;return O}function I(O,ue){return v(O,ue,"0")}function z(O,ue){function Se(Xi){return 0>Xi?-1:0<Xi?1:0}var $t;return($t=Se(O.getFullYear()-ue.getFullYear()))===0&&($t=Se(O.getMonth()-ue.getMonth()))===0&&($t=Se(O.getDate()-ue.getDate())),$t}function B(O){switch(O.getDay()){case 0:return new Date(O.getFullYear()-1,11,29);case 1:return O;case 2:return new Date(O.getFullYear(),0,3);case 3:return new Date(O.getFullYear(),0,2);case 4:return new Date(O.getFullYear(),0,1);case 5:return new Date(O.getFullYear()-1,11,31);case 6:return new Date(O.getFullYear()-1,11,30)}}function W(O){var ue=O.Bb;for(O=new Date(new Date(O.Cb+1900,0,1).getTime());0<ue;){var Se=O.getMonth(),$t=(Bt(O.getFullYear())?Di:Bi)[Se];if(!(ue>$t-O.getDate())){O.setDate(O.getDate()+ue);break}ue-=$t-O.getDate()+1,O.setDate(1),11>Se?O.setMonth(Se+1):(O.setMonth(0),O.setFullYear(O.getFullYear()+1))}return Se=new Date(O.getFullYear()+1,0,4),ue=B(new Date(O.getFullYear(),0,4)),Se=B(Se),0>=z(ue,O)?0>=z(Se,O)?O.getFullYear()+1:O.getFullYear():O.getFullYear()-1}s>>>=0,p>>>=0,f>>>=0,b>>>=0;var H=a()[b+40>>>2>>>0];for(var j in b={kc:i()[b>>>2>>>0],jc:i()[b+4>>>2>>>0],Hb:i()[b+8>>>2>>>0],Lb:i()[b+12>>>2>>>0],Ib:i()[b+16>>>2>>>0],Cb:i()[b+20>>>2>>>0],ub:i()[b+24>>>2>>>0],Bb:i()[b+28>>>2>>>0],rc:i()[b+32>>>2>>>0],ic:i()[b+36>>>2>>>0],lc:H?Oe(H):""},f=Oe(f),H={"%c":"%a %b %d %H:%M:%S %Y","%D":"%m/%d/%y","%F":"%Y-%m-%d","%h":"%b","%r":"%I:%M:%S %p","%R":"%H:%M","%T":"%H:%M:%S","%x":"%m/%d/%y","%X":"%H:%M:%S","%Ec":"%c","%EC":"%C","%Ex":"%m/%d/%y","%EX":"%H:%M:%S","%Ey":"%y","%EY":"%Y","%Od":"%d","%Oe":"%e","%OH":"%H","%OI":"%I","%Om":"%m","%OM":"%M","%OS":"%S","%Ou":"%u","%OU":"%U","%OV":"%V","%Ow":"%w","%OW":"%W","%Oy":"%y"})f=f.replace(new RegExp(j,"g"),H[j]);var ce="Sunday Monday Tuesday Wednesday Thursday Friday Saturday".split(" "),ge="January February March April May June July August September October November December".split(" ");for(j in H={"%a":O=>ce[O.ub].substring(0,3),"%A":O=>ce[O.ub],"%b":O=>ge[O.Ib].substring(0,3),"%B":O=>ge[O.Ib],"%C":O=>I((O.Cb+1900)/100|0,2),"%d":O=>I(O.Lb,2),"%e":O=>v(O.Lb,2," "),"%g":O=>W(O).toString().substring(2),"%G":W,"%H":O=>I(O.Hb,2),"%I":O=>((O=O.Hb)==0?O=12:12<O&&(O-=12),I(O,2)),"%j":O=>{for(var ue=0,Se=0;Se<=O.Ib-1;ue+=(Bt(O.Cb+1900)?Di:Bi)[Se++]);return I(O.Lb+ue,3)},"%m":O=>I(O.Ib+1,2),"%M":O=>I(O.jc,2),"%n":()=>`
-`,"%p":O=>0<=O.Hb&&12>O.Hb?"AM":"PM","%S":O=>I(O.kc,2),"%t":()=>"	","%u":O=>O.ub||7,"%U":O=>I(Math.floor((O.Bb+7-O.ub)/7),2),"%V":O=>{var ue=Math.floor((O.Bb+7-(O.ub+6)%7)/7);if(2>=(O.ub+371-O.Bb-2)%7&&ue++,ue)ue==53&&((Se=(O.ub+371-O.Bb)%7)==4||Se==3&&Bt(O.Cb)||(ue=1));else{ue=52;var Se=(O.ub+7-O.Bb-1)%7;(Se==4||Se==5&&Bt(O.Cb%400-1))&&ue++}return I(ue,2)},"%w":O=>O.ub,"%W":O=>I(Math.floor((O.Bb+7-(O.ub+6)%7)/7),2),"%y":O=>(O.Cb+1900).toString().substring(2),"%Y":O=>O.Cb+1900,"%z":O=>{var ue=0<=(O=O.ic);return O=Math.abs(O)/60,(ue?"+":"-")+("0000"+(O/60*100+O%60)).slice(-4)},"%Z":O=>O.lc,"%%":()=>"%"},f=f.replace(/%%/g,"\0\0"),H)f.includes(j)&&(f=f.replace(new RegExp(j,"g"),H[j](b)));return j=function(O){var ue=Array(_n(O)+1);return Zo(O,ue,0,ue.length),ue}(f=f.replace(/\0\0/g,"%")),j.length>p?0:(bp(j,s),j.length-1)}function wp(s,p,f,b){return Mi(s>>>0,p>>>0,f>>>0,b>>>0)}y||function(){for(var s=u.numThreads-1;s--;)Go();bt.unshift(()=>{Ue++,function(p){y?p():Promise.all(pt.map(Ho)).then(p)}(()=>Oo())})}();for(var Ri=Array(256),mr=0;256>mr;++mr)Ri[mr]=String.fromCharCode(mr);di=Ri,mt=u.BindingError=class extends Error{constructor(s){super(s),this.name="BindingError"}},u.InternalError=class extends Error{constructor(s){super(s),this.name="InternalError"}},dt.push(0,1,void 0,1,null,1,!0,1,!1,1),u.count_emval_handles=()=>dt.length/2-5-Sn.length;var vp=[wn,Vo,Fo,Yo,Xo,Qo,Jo,ei,ti,ri,ni,oi,ii,ai,si,ui,$i,xi,Ai,Ei,ki,Pi,Oi,zi],q=function(){function s(f,b){return q=f.exports,q=function(){var v=q,I={};for(let[z,B]of Object.entries(v))I[z]=typeof B=="function"?(...W)=>{dr.push(z);try{return B(...W)}finally{$e||(dr.pop(),et&&ht===1&&dr.length===0&&(ht=0,vt+=1,ur(qi),typeof Fibers<"u"&&Fibers.sc()))}}:B;return I}(),q=function(){var v=q,I=B=>W=>B(W)>>>0,z=B=>()=>B()>>>0;return(v=Object.assign({},v)).Ca=I(v.Ca),v.fb=z(v.fb),v.gb=I(v.gb),v.emscripten_main_runtime_thread_id=z(v.emscripten_main_runtime_thread_id),v.sb=I(v.sb),v.tb=z(v.tb),v}(),No.push(q.ib),Ae.unshift(q.Ba),Y=b,Oo(),q}var p=Ro();if(Ue++,u.instantiateWasm)try{return u.instantiateWasm(p,s)}catch(f){X(`Module.instantiateWasm callback failed with error: ${f}`),m(f)}return hn||=u.locateFile?zo("ort-wasm-simd-threaded.jsep.wasm")?"ort-wasm-simd-threaded.jsep.wasm":u.locateFile?u.locateFile("ort-wasm-simd-threaded.jsep.wasm",P):P+"ort-wasm-simd-threaded.jsep.wasm":new URL(/* asset import */ __webpack_require__(/*! ort-wasm-simd-threaded.jsep.wasm */ "./node_modules/onnxruntime-web/dist/ort-wasm-simd-threaded.jsep.wasm"), __webpack_require__.b).href,function(f,b){var v=hn;return D||typeof WebAssembly.instantiateStreaming!="function"||zo(v)||Do(v)||typeof fetch!="function"?Mo(v,f,b):fetch(v,{credentials:"same-origin"}).then(I=>WebAssembly.instantiateStreaming(I,f).then(b,function(z){return X(`wasm streaming compile failed: ${z}`),X("falling back to ArrayBuffer instantiation"),Mo(v,f,b)}))}(p,function(f){s(f.instance,f.module)}).catch(m),{}}(),Ui=s=>(Ui=q.Ca)(s),Vi=()=>(Vi=q.Da)();u._OrtInit=(s,p)=>(u._OrtInit=q.Ea)(s,p),u._OrtGetLastError=(s,p)=>(u._OrtGetLastError=q.Fa)(s,p),u._OrtCreateSessionOptions=(s,p,f,b,v,I,z,B,W,H)=>(u._OrtCreateSessionOptions=q.Ga)(s,p,f,b,v,I,z,B,W,H),u._OrtAppendExecutionProvider=(s,p)=>(u._OrtAppendExecutionProvider=q.Ha)(s,p),u._OrtAddFreeDimensionOverride=(s,p,f)=>(u._OrtAddFreeDimensionOverride=q.Ia)(s,p,f),u._OrtAddSessionConfigEntry=(s,p,f)=>(u._OrtAddSessionConfigEntry=q.Ja)(s,p,f),u._OrtReleaseSessionOptions=s=>(u._OrtReleaseSessionOptions=q.Ka)(s),u._OrtCreateSession=(s,p,f)=>(u._OrtCreateSession=q.La)(s,p,f),u._OrtReleaseSession=s=>(u._OrtReleaseSession=q.Ma)(s),u._OrtGetInputOutputCount=(s,p,f)=>(u._OrtGetInputOutputCount=q.Na)(s,p,f),u._OrtGetInputName=(s,p)=>(u._OrtGetInputName=q.Oa)(s,p),u._OrtGetOutputName=(s,p)=>(u._OrtGetOutputName=q.Pa)(s,p),u._OrtFree=s=>(u._OrtFree=q.Qa)(s),u._OrtCreateTensor=(s,p,f,b,v,I)=>(u._OrtCreateTensor=q.Ra)(s,p,f,b,v,I),u._OrtGetTensorData=(s,p,f,b,v)=>(u._OrtGetTensorData=q.Sa)(s,p,f,b,v),u._OrtReleaseTensor=s=>(u._OrtReleaseTensor=q.Ta)(s),u._OrtCreateRunOptions=(s,p,f,b)=>(u._OrtCreateRunOptions=q.Ua)(s,p,f,b),u._OrtAddRunConfigEntry=(s,p,f)=>(u._OrtAddRunConfigEntry=q.Va)(s,p,f),u._OrtReleaseRunOptions=s=>(u._OrtReleaseRunOptions=q.Wa)(s),u._OrtCreateBinding=s=>(u._OrtCreateBinding=q.Xa)(s),u._OrtBindInput=(s,p,f)=>(u._OrtBindInput=q.Ya)(s,p,f),u._OrtBindOutput=(s,p,f,b)=>(u._OrtBindOutput=q.Za)(s,p,f,b),u._OrtClearBoundOutputs=s=>(u._OrtClearBoundOutputs=q._a)(s),u._OrtReleaseBinding=s=>(u._OrtReleaseBinding=q.$a)(s),u._OrtRunWithBinding=(s,p,f,b,v)=>(u._OrtRunWithBinding=q.ab)(s,p,f,b,v),u._OrtRun=(s,p,f,b,v,I,z,B)=>(u._OrtRun=q.bb)(s,p,f,b,v,I,z,B),u._OrtEndProfiling=s=>(u._OrtEndProfiling=q.cb)(s),u._JsepOutput=(s,p,f)=>(u._JsepOutput=q.db)(s,p,f),u._JsepGetNodeName=s=>(u._JsepGetNodeName=q.eb)(s);var fr,Rt=()=>(Rt=q.fb)(),hr=u._malloc=s=>(hr=u._malloc=q.gb)(s),tt=u._free=s=>(tt=u._free=q.hb)(s),Dn=(s,p,f,b,v,I)=>(Dn=q.kb)(s,p,f,b,v,I),Ni=()=>(Ni=q.lb)(),Wi=(s,p,f,b,v)=>(Wi=q.mb)(s,p,f,b,v),Bn=s=>(Bn=q.nb)(s),gr=s=>(gr=q.ob)(s),Li=()=>(Li=q.pb)(),Hi=(s,p)=>(Hi=q.qb)(s,p),yr=s=>(yr=q.rb)(s),Mn=s=>(Mn=q.sb)(s),Rn=()=>(Rn=q.tb)(),Gi=u.dynCall_ii=(s,p)=>(Gi=u.dynCall_ii=q.vb)(s,p),Fi=s=>(Fi=q.wb)(s),qi=()=>(qi=q.xb)(),ji=s=>(ji=q.yb)(s),Ki=()=>(Ki=q.zb)();function Yi(){0<Ue||(y?(c(u),y||ar(Ae),startWorker(u)):(ar(bt),0<Ue||fr||(fr=!0,u.calledRun=!0,$e||(y||ar(Ae),c(u),y||ar(Me)))))}return u.___start_em_js=881586,u.___stop_em_js=881808,u.stackSave=()=>Rn(),u.stackRestore=s=>yr(s),u.stackAlloc=s=>Mn(s),u.UTF8ToString=Oe,u.stringToUTF8=Dt,u.lengthBytesUTF8=_n,wt=function s(){fr||Yi(),fr||(wt=s)},Yi(),h}),kp=Pa;globalThis.self?.name==="em-pthread"&&Pa()});var Ut,Pp,Op,zp,Da,Ba,Dp,Ma,qt=U(()=>{"use strict";Cr();Ut= false?0:import.meta.url??(typeof document<"u"?document.currentScript?.src:typeof self<"u"?self.location?.href:void 0),Pp= false||typeof location>"u"?void 0:location.origin,Op=(e,t)=>{try{let r=t??Ut;return(r?new URL(e,r):new URL(e)).origin===Pp}catch{return!1}},zp=async e=>{let r=await(await fetch(e,{credentials:"same-origin"})).blob();return URL.createObjectURL(r)},Da=(ka(),br(Ea)).default,Ba=async()=>{if(!Ut)throw new Error("Failed to load proxy worker: cannot determine the script source URL.");if(Op(Ut))return[void 0,Da()];let e=await zp(Ut);return[e,Da(e)]},Dp=(za(),br(Oa)).default,Ma=async(e,t,r)=>[void 0,Dp]});var qn,jn,Mr,Ra,Bp,Mp,Ar,Te,gt=U(()=>{"use strict";qt();jn=!1,Mr=!1,Ra=!1,Bp=()=>{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}},Mp=()=>{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}},Ar=async e=>{if(jn)return Promise.resolve();if(Mr)throw new Error("multiple calls to 'initializeWebAssembly()' detected.");if(Ra)throw new Error("previous call to 'initializeWebAssembly()' failed.");Mr=!0;let t=e.initTimeout,r=e.numThreads;if(!Mp())throw new Error("WebAssembly SIMD is not supported in the current environment.");let n=Bp();r>1&&!n&&(typeof self<"u"&&!self.crossOriginIsolated&&console.warn("env.wasm.numThreads is set to "+r+", 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=r=1);let o=e.wasmPaths,i=typeof o=="string"?o:void 0,a=o?.mjs,d=a?.href??a,l=o?.wasm,c=l?.href??l,m=e.wasmBinary,[u,h]=await Ma(d,i,r>1),w=!1,g=[];if(t>0&&g.push(new Promise(y=>{setTimeout(()=>{w=!0,y()},t)})),g.push(new Promise((y,S)=>{let $={numThreads:r};m?$.wasmBinary=m:(c||i)&&($.locateFile=(_,x)=>c??(i??x)+_),h($).then(_=>{Mr=!1,jn=!0,qn=_,y(),u&&URL.revokeObjectURL(u)},_=>{Mr=!1,Ra=!0,S(_)})})),await Promise.race(g),w)throw new Error(`WebAssembly backend initializing failed due to timeout: ${t}ms`)},Te=()=>{if(jn&&qn)return qn;throw new Error("WebAssembly is not initialized yet.")}});var Ee,Kt,_e,Rr=U(()=>{"use strict";gt();Ee=(e,t)=>{let r=Te(),n=r.lengthBytesUTF8(e)+1,o=r._malloc(n);return r.stringToUTF8(e,o,n),t.push(o),o},Kt=(e,t,r,n)=>{if(typeof e=="object"&&e!==null){if(r.has(e))throw new Error("Circular reference in options");r.add(e)}Object.entries(e).forEach(([o,i])=>{let a=t?t+o:o;if(typeof i=="object")Kt(i,a+".",r,n);else if(typeof i=="string"||typeof i=="number")n(a,i.toString());else if(typeof i=="boolean")n(a,i?"1":"0");else throw new Error(`Can't handle extra config type: ${typeof i}`)})},_e=e=>{let t=Te(),r=t.stackSave();try{let n=t.stackAlloc(8);t._OrtGetLastError(n,n+4);let o=t.HEAP32[n/4],i=t.HEAPU32[n/4+1],a=i?t.UTF8ToString(i):"";throw new Error(`${e} ERROR_CODE: ${o}, ERROR_MESSAGE: ${a}`)}finally{t.stackRestore(r)}}});var Ua,Va=U(()=>{"use strict";gt();Rr();Ua=e=>{let t=Te(),r=0,n=[],o=e||{};try{if(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?.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?.terminate===void 0&&(o.terminate=!1);let i=0;return e?.tag!==void 0&&(i=Ee(e.tag,n)),r=t._OrtCreateRunOptions(o.logSeverityLevel,o.logVerbosityLevel,!!o.terminate,i),r===0&&_e("Can't create run options."),e?.extra!==void 0&&Kt(e.extra,"",new WeakSet,(a,d)=>{let l=Ee(a,n),c=Ee(d,n);t._OrtAddRunConfigEntry(r,l,c)!==0&&_e(`Can't set a run config entry: ${a} - ${d}.`)}),[r,n]}catch(i){throw r!==0&&t._OrtReleaseRunOptions(r),n.forEach(a=>t._free(a)),i}}});var Rp,Up,Vp,Np,Na,Wa=U(()=>{"use strict";gt();Rr();Rp=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}`)}},Up=e=>{switch(e){case"sequential":return 0;case"parallel":return 1;default:throw new Error(`unsupported execution mode: ${e}`)}},Vp=e=>{e.extra||(e.extra={}),e.extra.session||(e.extra.session={});let t=e.extra.session;t.use_ort_model_bytes_directly||(t.use_ort_model_bytes_directly="1"),e.executionProviders&&e.executionProviders.some(r=>(typeof r=="string"?r:r.name)==="webgpu")&&(e.enableMemPattern=!1)},Np=(e,t,r)=>{for(let n of t){let o=typeof n=="string"?n:n.name;switch(o){case"webnn":if(o="WEBNN",typeof n!="string"){let d=n?.deviceType;if(d){let l=Ee("deviceType",r),c=Ee(d,r);Te()._OrtAddSessionConfigEntry(e,l,c)!==0&&_e(`Can't set a session config entry: 'deviceType' - ${d}.`)}}break;case"webgpu":if(o="JS",typeof n!="string"){let a=n;if(a?.preferredLayout){if(a.preferredLayout!=="NCHW"&&a.preferredLayout!=="NHWC")throw new Error(`preferredLayout must be either 'NCHW' or 'NHWC': ${a.preferredLayout}`);let d=Ee("preferredLayout",r),l=Ee(a.preferredLayout,r);Te()._OrtAddSessionConfigEntry(e,d,l)!==0&&_e(`Can't set a session config entry: 'preferredLayout' - ${a.preferredLayout}.`)}}break;case"wasm":case"cpu":continue;default:throw new Error(`not supported execution provider: ${o}`)}let i=Ee(o,r);Te()._OrtAppendExecutionProvider(e,i)!==0&&_e(`Can't append execution provider: ${o}.`)}},Na=e=>{let t=Te(),r=0,n=[],o=e||{};Vp(o);try{let i=Rp(o.graphOptimizationLevel??"all"),a=Up(o.executionMode??"sequential"),d=typeof o.logId=="string"?Ee(o.logId,n):0,l=o.logSeverityLevel??2;if(!Number.isInteger(l)||l<0||l>4)throw new Error(`log serverity level is not valid: ${l}`);let c=o.logVerbosityLevel??0;if(!Number.isInteger(c)||c<0||c>4)throw new Error(`log verbosity level is not valid: ${c}`);let m=typeof o.optimizedModelFilePath=="string"?Ee(o.optimizedModelFilePath,n):0;if(r=t._OrtCreateSessionOptions(i,!!o.enableCpuMemArena,!!o.enableMemPattern,a,!!o.enableProfiling,0,d,l,c,m),r===0&&_e("Can't create session options."),o.executionProviders&&Np(r,o.executionProviders,n),o.enableGraphCapture!==void 0){if(typeof o.enableGraphCapture!="boolean")throw new Error(`enableGraphCapture must be a boolean value: ${o.enableGraphCapture}`);let u=Ee("enableGraphCapture",n),h=Ee(o.enableGraphCapture.toString(),n);t._OrtAddSessionConfigEntry(r,u,h)!==0&&_e(`Can't set a session config entry: 'enableGraphCapture' - ${o.enableGraphCapture}.`)}if(o.freeDimensionOverrides)for(let[u,h]of Object.entries(o.freeDimensionOverrides)){if(typeof u!="string")throw new Error(`free dimension override name must be a string: ${u}`);if(typeof h!="number"||!Number.isInteger(h)||h<0)throw new Error(`free dimension override value must be a non-negative integer: ${h}`);let w=Ee(u,n);t._OrtAddFreeDimensionOverride(r,w,h)!==0&&_e(`Can't set a free dimension override: ${u} - ${h}.`)}return o.extra!==void 0&&Kt(o.extra,"",new WeakSet,(u,h)=>{let w=Ee(u,n),g=Ee(h,n);t._OrtAddSessionConfigEntry(r,w,g)!==0&&_e(`Can't set a session config entry: ${u} - ${h}.`)}),[r,n]}catch(i){throw r!==0&&t._OrtReleaseSessionOptions(r),n.forEach(a=>t._free(a)),i}}});var Yt,yt,It,Ur,Xt,Vr,Nr,Kn,Z=U(()=>{"use strict";Yt=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}`)}},yt=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}`)}},It=(e,t)=>{let r=[-1,4,1,1,2,2,4,8,-1,1,2,8,4,8,-1,-1,-1,-1,-1,-1,-1,.5,.5][e],n=typeof t=="number"?t:t.reduce((o,i)=>o*i,1);return r>0?Math.ceil(n*r):void 0},Ur=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}`)}},Xt=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}`)}},Vr=e=>e==="float32"||e==="float16"||e==="int32"||e==="int64"||e==="uint32"||e==="uint8"||e==="bool"||e==="uint4"||e==="int4",Nr=e=>e==="float32"||e==="float16"||e==="int32"||e==="int64"||e==="uint32"||e==="uint64"||e==="int8"||e==="uint8"||e==="bool",Kn=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}`)}}});var Zt,Yn=U(()=>{"use strict";Cr();Zt=async e=>{if(typeof e=="string")if(false){}else{let t=await fetch(e);if(!t.ok)throw new Error(`failed to load external data file: ${e}`);let r=t.headers.get("Content-Length"),n=r?parseInt(r,10):0;if(n<1073741824)return new Uint8Array(await t.arrayBuffer());{if(!t.body)throw new Error(`failed to load external data file: ${e}, no response body.`);let o=t.body.getReader(),i;try{i=new ArrayBuffer(n)}catch(d){if(d instanceof RangeError){let l=Math.ceil(n/65536);i=new WebAssembly.Memory({initial:l,maximum:l}).buffer}else throw d}let a=0;for(;;){let{done:d,value:l}=await o.read();if(d)break;let c=l.byteLength;new Uint8Array(i,a,c).set(l),a+=c}return new Uint8Array(i,0,n)}}else return e instanceof Blob?new Uint8Array(await e.arrayBuffer()):e instanceof Uint8Array?e:new Uint8Array(e)}});var Wp,Lp,La,Ha,Wr,Hp,de,Xe=U(()=>{"use strict";Z();Wp=["V","I","W","E","F"],Lp=(e,t)=>{console.log(`[${Wp[e]},${new Date().toISOString()}]${t}`)},Wr=(e,t)=>{La=e,Ha=t},Hp=(e,t)=>{let r=Xt(e),n=Xt(La);r>=n&&Lp(r,typeof t=="function"?t():t)},de=(...e)=>{Ha&&Hp(...e)}});var Lr,Xn=U(()=>{"use strict";Z();Lr=(e,t)=>new(Ur(t))(e)});var Hr=U(()=>{"use strict"});var Ga,Zn,Qn,Gp,Fp,Fa,eo,Jn,ja,Ka=U(()=>{"use strict";Xe();Hr();Ga=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]]),Zn=[],Qn=e=>Math.ceil(e/16)*16,Gp=e=>{for(let t=0;t<Zn.length;t++){let r=Zn[t];if(e<=r)return r}return Math.ceil(e/16)*16},Fp=1,Fa=()=>Fp++,eo=async(e,t,r,n)=>{let o=Qn(r),i=e.device.createBuffer({size:o,usage:GPUBufferUsage.COPY_DST|GPUBufferUsage.MAP_READ});try{let a=e.getCommandEncoder();e.endComputePass(),a.copyBufferToBuffer(t,0,i,0,o),e.flush(),await i.mapAsync(GPUMapMode.READ);let d=i.getMappedRange();if(n){let l=n();return l.set(new Uint8Array(d,0,r)),l}else return new Uint8Array(d.slice(0,r))}finally{i.destroy()}},Jn=class{constructor(t){this.backend=t;this.storageCache=new Map,this.freeBuffers=new Map,this.freeUniformBuffers=new Map,this.buffersForUploadingPending=[],this.buffersPending=[],this.capturedPendingBuffers=new Map;for(let[r]of Ga)Zn.push(r),this.freeBuffers.set(r,[]),this.freeUniformBuffers.set(r,[])}upload(t,r){let n=r.buffer,o=r.byteOffset,i=r.byteLength,a=Qn(i),d=this.storageCache.get(t);if(!d)throw new Error("gpu data for uploading does not exist");if(d.originalSize!==i)throw new Error(`inconsistent data size. gpu data size=${d.originalSize}, data size=${i}`);let l=this.backend.device.createBuffer({mappedAtCreation:!0,size:a,usage:GPUBufferUsage.MAP_WRITE|GPUBufferUsage.COPY_SRC}),c=l.getMappedRange();new Uint8Array(c).set(new Uint8Array(n,o,i)),l.unmap();let m=this.backend.getCommandEncoder();this.backend.endComputePass(),m.copyBufferToBuffer(l,0,d.gpuData.buffer,0,a),de("verbose",()=>`[WebGPU] GpuDataManager.upload(id=${t})`),this.buffersForUploadingPending.push(l)}memcpy(t,r){let n=this.storageCache.get(t);if(!n)throw new Error("source gpu data for memcpy does not exist");let o=this.storageCache.get(r);if(!o)throw new Error("destination gpu data for memcpy does not exist");if(n.originalSize!==o.originalSize)throw new Error("inconsistent source and destination gpu data size");let i=Qn(n.originalSize),a=this.backend.getCommandEncoder();this.backend.endComputePass(),a.copyBufferToBuffer(n.gpuData.buffer,0,o.gpuData.buffer,0,i)}registerExternalBuffer(t,r,n){let o;if(n){if(o=n[0],t===n[1])return de("verbose",()=>`[WebGPU] GpuDataManager.registerExternalBuffer(size=${r}) => id=${o}, buffer is the same, skip.`),o;if(this.backend.capturedCommandList.has(this.backend.currentSessionId))throw new Error(`Registering a different external buffer under graph capture mode is not supported yet.
-             Please use the previous external buffer!`)}else o=Fa();return this.storageCache.set(o,{gpuData:{id:o,type:0,buffer:t},originalSize:r}),de("verbose",()=>`[WebGPU] GpuDataManager.registerExternalBuffer(size=${r}) => id=${o}, registered.`),o}unregisterExternalBuffer(t){t!==void 0&&(this.storageCache.delete(t),de("verbose",()=>`[WebGPU] GpuDataManager.unregisterExternalBuffer() => id=${t}`))}create(t,r=GPUBufferUsage.STORAGE|GPUBufferUsage.COPY_SRC|GPUBufferUsage.COPY_DST){let n=Gp(t),o,i=(r&GPUBufferUsage.STORAGE)===GPUBufferUsage.STORAGE,a=(r&GPUBufferUsage.UNIFORM)===GPUBufferUsage.UNIFORM;if(i||a){let c=(i?this.freeBuffers:this.freeUniformBuffers).get(n);c?c.length>0?o=c.pop():o=this.backend.device.createBuffer({size:n,usage:r}):o=this.backend.device.createBuffer({size:n,usage:r})}else o=this.backend.device.createBuffer({size:n,usage:r});let d={id:Fa(),type:0,buffer:o};return this.storageCache.set(d.id,{gpuData:d,originalSize:t}),de("verbose",()=>`[WebGPU] GpuDataManager.create(size=${t}) => id=${d.id}`),d}get(t){return this.storageCache.get(t)?.gpuData}release(t){let r=this.storageCache.get(t);if(!r)throw new Error("releasing data does not exist");return de("verbose",()=>`[WebGPU] GpuDataManager.release(id=${t}), gpuDataId=${r.gpuData.id}`),this.storageCache.delete(t),this.buffersPending.push(r.gpuData.buffer),r.originalSize}async download(t,r){let n=this.storageCache.get(t);if(!n)throw new Error("data does not exist");await eo(this.backend,n.gpuData.buffer,n.originalSize,r)}refreshPendingBuffers(){for(let t of this.buffersForUploadingPending)t.destroy();if(this.buffersForUploadingPending=[],this.buffersPending.length!==0)if(this.backend.sessionStatus==="default"){for(let t of this.buffersPending){let r=Ga.get(t.size);if((t.usage&GPUBufferUsage.STORAGE)===GPUBufferUsage.STORAGE){let n=this.freeBuffers.get(t.size)||[];r===void 0||n.length>=r?t.destroy():n.push(t)}else if((t.usage&GPUBufferUsage.UNIFORM)===GPUBufferUsage.UNIFORM){let n=this.freeUniformBuffers.get(t.size)||[];r===void 0||n.length>=r?t.destroy():n.push(t)}else t.destroy()}this.buffersPending=[]}else{let t=this.capturedPendingBuffers.get(this.backend.currentSessionId);t||(t=[],this.capturedPendingBuffers.set(this.backend.currentSessionId,t));for(let r of this.buffersPending)t.push(r);this.buffersPending=[]}}dispose(){this.freeBuffers.forEach(t=>{t.forEach(r=>{r.destroy()})}),this.freeUniformBuffers.forEach(t=>{t.forEach(r=>{r.destroy()})}),this.storageCache.forEach(t=>{t.gpuData.buffer.destroy()}),this.capturedPendingBuffers.forEach(t=>{t.forEach(r=>{r.destroy()})}),this.storageCache=new Map,this.freeBuffers=new Map,this.freeUniformBuffers=new Map,this.capturedPendingBuffers=new Map}onReleaseSession(t){let r=this.capturedPendingBuffers.get(t);r&&(r.forEach(n=>{n.destroy()}),this.capturedPendingBuffers.delete(t))}},ja=(...e)=>new Jn(...e)});var to,J,Ie=U(()=>{"use strict";to=class{constructor(t){Object.assign(this,t)}get cacheKey(){return this.key||(this.key=Object.getOwnPropertyNames(this).sort().map(t=>`${this[t]}`).join(";")),this.key}},J=e=>new to(e)});var ro,rt,E,Ct,Gr,Ya,Xa,te=U(()=>{"use strict";ro=class{static calcMatMulShape(t,r){return t[1]!==r[0]?void 0:[t[0],r[1]]}},rt=class{static calcShape(t,r,n=!1){let o=t.length,i=r.length;if(o===0)return r;if(i===0)return t;let a=Math.max(t.length,r.length),d=new Array(a);if(n){if(o<2||i<2)return;let l=ro.calcMatMulShape([t[o-2],t[o-1]],[r[i-2],r[i-1]]);if(l===void 0)return;[d[a-2],d[a-1]]=l}for(let l=n?3:1;l<=a;l++){let c=o-l<0?1:t[o-l],m=i-l<0?1:r[i-l];if(c!==m&&c>1&&m>1)return;let u=Math.max(c,m);if(c&&m)d[a-l]=Math.max(c,m);else{if(u>1)return;d[a-l]=0}}return d}static isValidBroadcast(t,r){let n=t.length,o=r.length;if(n>o)return!1;for(let i=1;i<=n;i++)if(t[n-i]!==1&&t[n-i]!==r[o-i])return!1;return!0}},E=class e{static size(t){return e.getSizeFromDimensionRange(t,0,t.length)}static convertShape(t,r=4){let n=t.length;if(n===0)return[];let o=new Array(n),i=n-1;for(;i>=0;){if(t[i]%r===0){o[i]=t[i]/r;break}if(r%t[i]!==0)throw new Error("cannot convert shape");o[i]=1,r/=t[i],i--}for(i--;i>=0;i--)o[i]=t[i];return o}static sizeFromDimension(t,r){if(r<0||r>t.length)throw new Error(`invalid dimension of ${r} for sizeFromDimension as Tensor has ${t.length} dimensions.`);return e.getSizeFromDimensionRange(t,r,t.length)}static sizeToDimension(t,r){if(r<0||r>t.length)throw new Error(`invalid dimension of ${r} for sizeToDimension as Tensor has ${t.length} dimensions.`);return e.getSizeFromDimensionRange(t,0,r)}static getSizeFromDimensionRange(t,r,n){let o=1;for(let i=r;i<n;i++){if(t[i]<0)throw new Error("cannot get valid size from specified dimension range. Most likely the range contains negative values in them.");o*=t[i]}return o}static computeStrides(t){let r=t.length;if(r===0)return[];if(r===1)return[1];let n=new Array(r);n[r-1]=1,n[r-2]=t[r-1];for(let o=r-3;o>=0;--o)n[o]=n[o+1]*t[o+1];return n}static normalizeAxis(t,r){if(t<-r&&t>=r)throw new Error("unsupported axis for this operation.");return t<0?t+r:t}static normalizeAxes(t,r){return t.map(n=>this.normalizeAxis(n,r??t.length))}static sortBasedOnPerm(t,r){return r?r.map(n=>t[n]):t.slice().reverse()}static padShape(t,r){let n=t.length;return t.map((o,i)=>o+r[i]+r[i+n])}static areEqual(t,r){return t.length!==r.length?!1:t.every((n,o)=>n===r[o])}},Ct=class e{static adjustPoolAttributes(t,r,n,o,i,a){if(!t&&n.length!==r.length-2)throw new Error("length of specified kernel shapes should be 2 less than length of input dimensions");if(t)for(let d=0;d<r.length-2;d++)d>=n.length?n.push(r[d+2]):n[d]=r[d+2];for(let d=0;d<n.length;d++)if(d<o.length){if(o[d]<0)throw new Error("strides should be greater than or equal to 1")}else o.push(1);for(let d=0;d<n.length;d++)if(d<i.length){if(i[d]<0)throw new Error("dilations should be greater than or equal to 1")}else i.push(1);for(let d=0;d<n.length*2;d++)if(d<a.length){if(a[d]<0)throw new Error("pad should be greater than or equal to 1")}else a.push(0);for(let d=0;d<n.length;d++){if(n[d]<=0)throw new Error("kernel shapes need to be greater than 0");if(a[d]>=n[d]||a[d+n.length]>=n[d])throw new Error("pads should be smaller than kernel")}}static adjustPadsBasedOnAutoPad(t,r,n,o,i,a,d){if(d){if(i.length!==2*(t.length-2))throw new Error("length of pads should be twice the length of data dimensions");if(r.length!==t.length-2)throw new Error("length of strides should be the length of data dimensions");if(o.length!==t.length-2)throw new Error("length of kernel shapes should be the length of data dimensions");for(let l=0;l<t.length-2;l++)e.adjustPadAndReturnShape(t[l+(a?1:2)],r[l],n[l],o[l],i,l,l+t.length-2,d)}}static computePoolOutputShape(t,r,n,o,i,a,d){if(r.length<=0)throw new Error("input shape must be of size greater than 0");let l=[r[0],r[1]];return e.computeShapeHelper(t,r,l,n,o,i,a,d),l}static computeConvOutputShape(t,r,n,o,i,a,d){if(t.length<=0||r.length<=0)throw new Error("invalid input tensor dims or invalid filter tensor dims");let l=[t[0],r[0]];return e.computeShapeHelper(!1,t,l,n,o,i,a,d),l}static computeShapeHelper(t,r,n,o,i,a,d,l){if(t)for(let c=0;c<r.length-2;c++)n.push(1);else for(let c=0;c<r.length-2;c++)n.push(e.adjustPadAndReturnShape(r[c+2],o[c],i[c],a[c],d,c,c+r.length-2,l))}static adjustPadAndReturnShape(t,r,n,o,i,a,d,l){let c=n*(o-1)+1;if(l&&l!=="NOTSET")switch(l){case"VALID":return i[a]=0,i[d]=0,Math.floor((t-c)/r+1);case"SAME_LOWER":case"SAME_UPPER":if(n!==1)throw new Error("Dilation not supported for SAME_UPPER or SAME_LOWER");{let u=((t+r-1)/r-1)*r+o-t;return i[a]=Math.floor(l==="SAME_LOWER"?(u+1)/2:u/2),i[d]=u-i[a],Math.floor((t+u-o)/r+1)}default:throw new Error("Unsupported AutoPad type")}else return Math.floor((t+i[a]+i[d]-c)/r+1)}},Gr=class{static getShapeOfGemmResult(t,r,n,o,i){if(t.length!==2||n.length!==2)throw new Error("shape need to be of size 2");let a,d,l;r?(a=t[1],d=t[0]):(a=t[0],d=t[1]);let c=-1;if(o?(l=n[0],c=1):(l=n[1],c=0),n[c]!==d)throw new Error("dimension mismatch");if(a<=0||l<=0||d<=0)throw new Error("invalid shape specified");if(i&&!rt.isValidBroadcast(i,[a,l]))throw new Error("gemm: invalid bias shape for broadcast");return[a,l,d]}},Ya=-34028234663852886e22,Xa=34028234663852886e22});var At,oo,ye,ke,V,ve,io,Et,Ze,F,ao,k,M,Fr,no,Za,Nt,re=U(()=>{"use strict";Z();te();At=64,oo=(e,t)=>{if(t===3)throw new Error("vec3 has same alignment as vec4, use vec4 instead");switch(e){case 10:return t>1?`vec${t}<f16>`:"f16";case 1:return t>1?`vec${t}<f32>`:"f32";case 6:return t>1?`vec${t}<i32>`:"i32";case 12:return t>1?`vec${t}<u32>`:"u32";case 7:if(t>1)throw new Error("currently not supported vecX of uint64 yet");return["vec2<u32>","i32"];case 13:if(t>1)throw new Error("currently not supported vecX of uint64 yet");return["vec2<u32>","u32"];case 9:if(t!==4)throw new Error("bool must be vec4");return["u32","vec4<bool>"];case 22:return"i32";case 21:return"u32";default:throw new Error(`Unknown data type: ${e}`)}},ye=(e,t=1)=>{let r=oo(e,t);return typeof r=="string"?r:r[0]},ke=(e,t=1)=>{let r=oo(e,t);return typeof r=="string"?r:r[1]},V=(...e)=>{let t=[];return e.forEach(r=>{r.length!==0&&t.push({type:12,data:r},{type:12,data:E.computeStrides(r)})}),t},ve=e=>e%4===0?4:e%2===0?2:1,io=(e="f32",t,r="0")=>!t||t===1?`${e}(${r})`:`vec${t}<${e}>(${r})`,Et=(e,t,r)=>e==="f32"?r:t===1?`f32(${r})`:`vec${t}<f32>(${r})`,Ze=(e,t)=>t===4?`(${e}.x + ${e}.y + ${e}.z + ${e}.w)`:t===2?`(${e}.x + ${e}.y)`:t===3?`(${e}.x + ${e}.y + ${e}.z)`:e,F=(e,t,r,n)=>e.startsWith("uniforms.")&&r>4?typeof t=="string"?n==="f16"?`${e}[(${t}) / 8][(${t}) % 8 / 4][(${t}) % 8 % 4]`:`${e}[(${t}) / 4][(${t}) % 4]`:n==="f16"?`${e}[${Math.floor(t/8)}][${Math.floor(t%8/4)}][${t%8%4}]`:`${e}[${Math.floor(t/4)}][${t%4}]`:r>1?`${e}[${t}]`:e,ao=(e,t,r,n,o)=>{let i=typeof r=="number",a=i?r:r.length,d=[...new Array(a).keys()],l=a<2?"u32":a<=4?`vec${a}<u32>`:`array<u32, ${a}>`,c=oo(t,o),m=typeof c=="string"?c:c[1],u=typeof c=="string"?c:c[0],h={indices:l,value:m,storage:u,tensor:t},w=R=>typeof R=="string"?R:`${R}u`,g={offsetToIndices:!1,indicesToOffset:!1,broadcastedIndicesToOffset:!1,set:!1,setByIndices:!1,get:!1,getByIndices:!1},y=i?"uniforms.":"",S=`${y}${e}_shape`,$=`${y}${e}_strides`,_="";for(let R=0;R<a-1;R++)_+=`
-    let dim${R} = current / ${F($,R,a)};
-    let rest${R} = current % ${F($,R,a)};
-    indices[${R}] = dim${R};
-    current = rest${R};
-    `;_+=`indices[${a-1}] = current;`;let x=a<2?"":`
+`))[0]=="Error"&&b.shift(),Ii(b);for(var _=3;b[_]&&pr()!=s;)++_;for(s=0;s<f&&b[s+_];++s)i()[p+4*s>>>2>>>0]=pr();return s}var zn,On={},Ci=()=>{if(!zn){var s,p={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",_:C||"./this.program"};for(s in On)On[s]===void 0?delete p[s]:p[s]=On[s];var f=[];for(s in p)f.push(`${s}=${p[s]}`);zn=f}return zn};function Ai(s,p){if(y)return xe(18,1,s,p);s>>>=0,p>>>=0;var f=0;return Ci().forEach((b,_)=>{var I=p+f;for(_=a()[s+4*_>>>2>>>0]=I,I=0;I<b.length;++I)t()[_++>>>0]=b.charCodeAt(I);t()[_>>>0]=0,f+=b.length+1}),0}function ki(s,p){if(y)return xe(19,1,s,p);s>>>=0,p>>>=0;var f=Ci();a()[s>>>2>>>0]=f.length;var b=0;return f.forEach(_=>b+=_.length+1),a()[p>>>2>>>0]=b,0}function Ei(s){return y?xe(20,1,s):52}function Pi(s,p,f,b){return y?xe(21,1,s,p,f,b):52}function zi(s,p,f,b){return y?xe(22,1,s,p,f,b):70}var yp=[null,[],[]];function Oi(s,p,f,b){if(y)return xe(23,1,s,p,f,b);p>>>=0,f>>>=0,b>>>=0;for(var _=0,I=0;I<f;I++){var O=a()[p>>>2>>>0],D=a()[p+4>>>2>>>0];p+=8;for(var L=0;L<D;L++){var H=r()[O+L>>>0],X=yp[s];H===0||H===10?((s===1?K:Z)(Ko(X,0)),X.length=0):X.push(H)}_+=D}return a()[b>>>2>>>0]=_,0}var Bi=[31,29,31,30,31,30,31,31,30,31,30,31],Di=[31,28,31,30,31,30,31,31,30,31,30,31],bp=(s,p)=>{t().set(s,p>>>0)};function Mi(s,p,f,b){function _(z,pe,Se){for(z=typeof z=="number"?z.toString():z||"";z.length<pe;)z=Se[0]+z;return z}function I(z,pe){return _(z,pe,"0")}function O(z,pe){function Se(Xi){return 0>Xi?-1:0<Xi?1:0}var $t;return($t=Se(z.getFullYear()-pe.getFullYear()))===0&&($t=Se(z.getMonth()-pe.getMonth()))===0&&($t=Se(z.getDate()-pe.getDate())),$t}function D(z){switch(z.getDay()){case 0:return new Date(z.getFullYear()-1,11,29);case 1:return z;case 2:return new Date(z.getFullYear(),0,3);case 3:return new Date(z.getFullYear(),0,2);case 4:return new Date(z.getFullYear(),0,1);case 5:return new Date(z.getFullYear()-1,11,31);case 6:return new Date(z.getFullYear()-1,11,30)}}function L(z){var pe=z.Bb;for(z=new Date(new Date(z.Cb+1900,0,1).getTime());0<pe;){var Se=z.getMonth(),$t=(Dt(z.getFullYear())?Bi:Di)[Se];if(!(pe>$t-z.getDate())){z.setDate(z.getDate()+pe);break}pe-=$t-z.getDate()+1,z.setDate(1),11>Se?z.setMonth(Se+1):(z.setMonth(0),z.setFullYear(z.getFullYear()+1))}return Se=new Date(z.getFullYear()+1,0,4),pe=D(new Date(z.getFullYear(),0,4)),Se=D(Se),0>=O(pe,z)?0>=O(Se,z)?z.getFullYear()+1:z.getFullYear():z.getFullYear()-1}s>>>=0,p>>>=0,f>>>=0,b>>>=0;var H=a()[b+40>>>2>>>0];for(var X in b={kc:i()[b>>>2>>>0],jc:i()[b+4>>>2>>>0],Hb:i()[b+8>>>2>>>0],Lb:i()[b+12>>>2>>>0],Ib:i()[b+16>>>2>>>0],Cb:i()[b+20>>>2>>>0],ub:i()[b+24>>>2>>>0],Bb:i()[b+28>>>2>>>0],rc:i()[b+32>>>2>>>0],ic:i()[b+36>>>2>>>0],lc:H?ze(H):""},f=ze(f),H={"%c":"%a %b %d %H:%M:%S %Y","%D":"%m/%d/%y","%F":"%Y-%m-%d","%h":"%b","%r":"%I:%M:%S %p","%R":"%H:%M","%T":"%H:%M:%S","%x":"%m/%d/%y","%X":"%H:%M:%S","%Ec":"%c","%EC":"%C","%Ex":"%m/%d/%y","%EX":"%H:%M:%S","%Ey":"%y","%EY":"%Y","%Od":"%d","%Oe":"%e","%OH":"%H","%OI":"%I","%Om":"%m","%OM":"%M","%OS":"%S","%Ou":"%u","%OU":"%U","%OV":"%V","%Ow":"%w","%OW":"%W","%Oy":"%y"})f=f.replace(new RegExp(X,"g"),H[X]);var fe="Sunday Monday Tuesday Wednesday Thursday Friday Saturday".split(" "),be="January February March April May June July August September October November December".split(" ");for(X in H={"%a":z=>fe[z.ub].substring(0,3),"%A":z=>fe[z.ub],"%b":z=>be[z.Ib].substring(0,3),"%B":z=>be[z.Ib],"%C":z=>I((z.Cb+1900)/100|0,2),"%d":z=>I(z.Lb,2),"%e":z=>_(z.Lb,2," "),"%g":z=>L(z).toString().substring(2),"%G":L,"%H":z=>I(z.Hb,2),"%I":z=>((z=z.Hb)==0?z=12:12<z&&(z-=12),I(z,2)),"%j":z=>{for(var pe=0,Se=0;Se<=z.Ib-1;pe+=(Dt(z.Cb+1900)?Bi:Di)[Se++]);return I(z.Lb+pe,3)},"%m":z=>I(z.Ib+1,2),"%M":z=>I(z.jc,2),"%n":()=>`
+`,"%p":z=>0<=z.Hb&&12>z.Hb?"AM":"PM","%S":z=>I(z.kc,2),"%t":()=>"	","%u":z=>z.ub||7,"%U":z=>I(Math.floor((z.Bb+7-z.ub)/7),2),"%V":z=>{var pe=Math.floor((z.Bb+7-(z.ub+6)%7)/7);if(2>=(z.ub+371-z.Bb-2)%7&&pe++,pe)pe==53&&((Se=(z.ub+371-z.Bb)%7)==4||Se==3&&Dt(z.Cb)||(pe=1));else{pe=52;var Se=(z.ub+7-z.Bb-1)%7;(Se==4||Se==5&&Dt(z.Cb%400-1))&&pe++}return I(pe,2)},"%w":z=>z.ub,"%W":z=>I(Math.floor((z.Bb+7-(z.ub+6)%7)/7),2),"%y":z=>(z.Cb+1900).toString().substring(2),"%Y":z=>z.Cb+1900,"%z":z=>{var pe=0<=(z=z.ic);return z=Math.abs(z)/60,(pe?"+":"-")+("0000"+(z/60*100+z%60)).slice(-4)},"%Z":z=>z.lc,"%%":()=>"%"},f=f.replace(/%%/g,"\0\0"),H)f.includes(X)&&(f=f.replace(new RegExp(X,"g"),H[X](b)));return X=function(z){var pe=Array(vn(z)+1);return Zo(z,pe,0,pe.length),pe}(f=f.replace(/\0\0/g,"%")),X.length>p?0:(bp(X,s),X.length-1)}function wp(s,p,f,b){return Mi(s>>>0,p>>>0,f>>>0,b>>>0)}y||function(){for(var s=u.numThreads-1;s--;)Go();bt.unshift(()=>{Ue++,function(p){y?p():Promise.all(pt.map(Ho)).then(p)}(()=>zo())})}();for(var Ri=Array(256),mr=0;256>mr;++mr)Ri[mr]=String.fromCharCode(mr);di=Ri,mt=u.BindingError=class extends Error{constructor(s){super(s),this.name="BindingError"}},u.InternalError=class extends Error{constructor(s){super(s),this.name="InternalError"}},dt.push(0,1,void 0,1,null,1,!0,1,!1,1),u.count_emval_handles=()=>dt.length/2-5-Sn.length;var _p=[wn,Vo,Fo,Yo,Xo,Qo,Jo,ei,ti,ri,ni,oi,ii,ai,si,ui,$i,xi,Ai,ki,Ei,Pi,zi,Oi],j=function(){function s(f,b){return j=f.exports,j=function(){var _=j,I={};for(let[O,D]of Object.entries(_))I[O]=typeof D=="function"?(...L)=>{dr.push(O);try{return D(...L)}finally{$e||(dr.pop(),et&&ht===1&&dr.length===0&&(ht=0,_t+=1,ur(qi),typeof Fibers<"u"&&Fibers.sc()))}}:D;return I}(),j=function(){var _=j,I=D=>L=>D(L)>>>0,O=D=>()=>D()>>>0;return(_=Object.assign({},_)).Ca=I(_.Ca),_.fb=O(_.fb),_.gb=I(_.gb),_.emscripten_main_runtime_thread_id=O(_.emscripten_main_runtime_thread_id),_.sb=I(_.sb),_.tb=O(_.tb),_}(),No.push(j.ib),Ae.unshift(j.Ba),Y=b,zo(),j}var p=Ro();if(Ue++,u.instantiateWasm)try{return u.instantiateWasm(p,s)}catch(f){Z(`Module.instantiateWasm callback failed with error: ${f}`),m(f)}return hn||=u.locateFile?Oo("ort-wasm-simd-threaded.jsep.wasm")?"ort-wasm-simd-threaded.jsep.wasm":u.locateFile?u.locateFile("ort-wasm-simd-threaded.jsep.wasm",P):P+"ort-wasm-simd-threaded.jsep.wasm":new URL(/* asset import */ __webpack_require__(/*! ort-wasm-simd-threaded.jsep.wasm */ "./node_modules/onnxruntime-web/dist/ort-wasm-simd-threaded.jsep.wasm"), __webpack_require__.b).href,function(f,b){var _=hn;return B||typeof WebAssembly.instantiateStreaming!="function"||Oo(_)||Bo(_)||typeof fetch!="function"?Mo(_,f,b):fetch(_,{credentials:"same-origin"}).then(I=>WebAssembly.instantiateStreaming(I,f).then(b,function(O){return Z(`wasm streaming compile failed: ${O}`),Z("falling back to ArrayBuffer instantiation"),Mo(_,f,b)}))}(p,function(f){s(f.instance,f.module)}).catch(m),{}}(),Ui=s=>(Ui=j.Ca)(s),Vi=()=>(Vi=j.Da)();u._OrtInit=(s,p)=>(u._OrtInit=j.Ea)(s,p),u._OrtGetLastError=(s,p)=>(u._OrtGetLastError=j.Fa)(s,p),u._OrtCreateSessionOptions=(s,p,f,b,_,I,O,D,L,H)=>(u._OrtCreateSessionOptions=j.Ga)(s,p,f,b,_,I,O,D,L,H),u._OrtAppendExecutionProvider=(s,p)=>(u._OrtAppendExecutionProvider=j.Ha)(s,p),u._OrtAddFreeDimensionOverride=(s,p,f)=>(u._OrtAddFreeDimensionOverride=j.Ia)(s,p,f),u._OrtAddSessionConfigEntry=(s,p,f)=>(u._OrtAddSessionConfigEntry=j.Ja)(s,p,f),u._OrtReleaseSessionOptions=s=>(u._OrtReleaseSessionOptions=j.Ka)(s),u._OrtCreateSession=(s,p,f)=>(u._OrtCreateSession=j.La)(s,p,f),u._OrtReleaseSession=s=>(u._OrtReleaseSession=j.Ma)(s),u._OrtGetInputOutputCount=(s,p,f)=>(u._OrtGetInputOutputCount=j.Na)(s,p,f),u._OrtGetInputName=(s,p)=>(u._OrtGetInputName=j.Oa)(s,p),u._OrtGetOutputName=(s,p)=>(u._OrtGetOutputName=j.Pa)(s,p),u._OrtFree=s=>(u._OrtFree=j.Qa)(s),u._OrtCreateTensor=(s,p,f,b,_,I)=>(u._OrtCreateTensor=j.Ra)(s,p,f,b,_,I),u._OrtGetTensorData=(s,p,f,b,_)=>(u._OrtGetTensorData=j.Sa)(s,p,f,b,_),u._OrtReleaseTensor=s=>(u._OrtReleaseTensor=j.Ta)(s),u._OrtCreateRunOptions=(s,p,f,b)=>(u._OrtCreateRunOptions=j.Ua)(s,p,f,b),u._OrtAddRunConfigEntry=(s,p,f)=>(u._OrtAddRunConfigEntry=j.Va)(s,p,f),u._OrtReleaseRunOptions=s=>(u._OrtReleaseRunOptions=j.Wa)(s),u._OrtCreateBinding=s=>(u._OrtCreateBinding=j.Xa)(s),u._OrtBindInput=(s,p,f)=>(u._OrtBindInput=j.Ya)(s,p,f),u._OrtBindOutput=(s,p,f,b)=>(u._OrtBindOutput=j.Za)(s,p,f,b),u._OrtClearBoundOutputs=s=>(u._OrtClearBoundOutputs=j._a)(s),u._OrtReleaseBinding=s=>(u._OrtReleaseBinding=j.$a)(s),u._OrtRunWithBinding=(s,p,f,b,_)=>(u._OrtRunWithBinding=j.ab)(s,p,f,b,_),u._OrtRun=(s,p,f,b,_,I,O,D)=>(u._OrtRun=j.bb)(s,p,f,b,_,I,O,D),u._OrtEndProfiling=s=>(u._OrtEndProfiling=j.cb)(s),u._JsepOutput=(s,p,f)=>(u._JsepOutput=j.db)(s,p,f),u._JsepGetNodeName=s=>(u._JsepGetNodeName=j.eb)(s);var fr,Rt=()=>(Rt=j.fb)(),hr=u._malloc=s=>(hr=u._malloc=j.gb)(s),tt=u._free=s=>(tt=u._free=j.hb)(s),Bn=(s,p,f,b,_,I)=>(Bn=j.kb)(s,p,f,b,_,I),Ni=()=>(Ni=j.lb)(),Wi=(s,p,f,b,_)=>(Wi=j.mb)(s,p,f,b,_),Dn=s=>(Dn=j.nb)(s),gr=s=>(gr=j.ob)(s),Li=()=>(Li=j.pb)(),Hi=(s,p)=>(Hi=j.qb)(s,p),yr=s=>(yr=j.rb)(s),Mn=s=>(Mn=j.sb)(s),Rn=()=>(Rn=j.tb)(),Gi=u.dynCall_ii=(s,p)=>(Gi=u.dynCall_ii=j.vb)(s,p),Fi=s=>(Fi=j.wb)(s),qi=()=>(qi=j.xb)(),ji=s=>(ji=j.yb)(s),Ki=()=>(Ki=j.zb)();function Yi(){0<Ue||(y?(c(u),y||ar(Ae),startWorker(u)):(ar(bt),0<Ue||fr||(fr=!0,u.calledRun=!0,$e||(y||ar(Ae),c(u),y||ar(Me)))))}return u.___start_em_js=881730,u.___stop_em_js=881952,u.stackSave=()=>Rn(),u.stackRestore=s=>yr(s),u.stackAlloc=s=>Mn(s),u.UTF8ToString=ze,u.stringToUTF8=Bt,u.lengthBytesUTF8=vn,wt=function s(){fr||Yi(),fr||(wt=s)},Yi(),h}),Ep=Pa;globalThis.self?.name==="em-pthread"&&Pa()});var Ut,Pp,zp,Op,Ba,Da,Bp,Ma,qt=V(()=>{"use strict";Cr();Ut= false?0:import.meta.url??(typeof document<"u"?document.currentScript?.src:typeof self<"u"?self.location?.href:void 0),Pp= false||typeof location>"u"?void 0:location.origin,zp=(e,t)=>{try{let r=t??Ut;return(r?new URL(e,r):new URL(e)).origin===Pp}catch{return!1}},Op=async e=>{let r=await(await fetch(e,{credentials:"same-origin"})).blob();return URL.createObjectURL(r)},Ba=(Ea(),br(ka)).default,Da=async()=>{if(!Ut)throw new Error("Failed to load proxy worker: cannot determine the script source URL.");if(zp(Ut))return[void 0,Ba()];let e=await Op(Ut);return[e,Ba(e)]},Bp=(Oa(),br(za)).default,Ma=async(e,t,r)=>[void 0,Bp]});var qn,jn,Mr,Ra,Dp,Mp,Ar,Te,gt=V(()=>{"use strict";qt();jn=!1,Mr=!1,Ra=!1,Dp=()=>{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}},Mp=()=>{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}},Ar=async e=>{if(jn)return Promise.resolve();if(Mr)throw new Error("multiple calls to 'initializeWebAssembly()' detected.");if(Ra)throw new Error("previous call to 'initializeWebAssembly()' failed.");Mr=!0;let t=e.initTimeout,r=e.numThreads;if(!Mp())throw new Error("WebAssembly SIMD is not supported in the current environment.");let n=Dp();r>1&&!n&&(typeof self<"u"&&!self.crossOriginIsolated&&console.warn("env.wasm.numThreads is set to "+r+", 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=r=1);let o=e.wasmPaths,i=typeof o=="string"?o:void 0,a=o?.mjs,d=a?.href??a,l=o?.wasm,c=l?.href??l,m=e.wasmBinary,[u,h]=await Ma(d,i,r>1),w=!1,g=[];if(t>0&&g.push(new Promise(y=>{setTimeout(()=>{w=!0,y()},t)})),g.push(new Promise((y,S)=>{let $={numThreads:r};m?$.wasmBinary=m:(c||i)&&($.locateFile=(v,x)=>c??(i??x)+v),h($).then(v=>{Mr=!1,jn=!0,qn=v,y(),u&&URL.revokeObjectURL(u)},v=>{Mr=!1,Ra=!0,S(v)})})),await Promise.race(g),w)throw new Error(`WebAssembly backend initializing failed due to timeout: ${t}ms`)},Te=()=>{if(jn&&qn)return qn;throw new Error("WebAssembly is not initialized yet.")}});var ke,Kt,ve,Rr=V(()=>{"use strict";gt();ke=(e,t)=>{let r=Te(),n=r.lengthBytesUTF8(e)+1,o=r._malloc(n);return r.stringToUTF8(e,o,n),t.push(o),o},Kt=(e,t,r,n)=>{if(typeof e=="object"&&e!==null){if(r.has(e))throw new Error("Circular reference in options");r.add(e)}Object.entries(e).forEach(([o,i])=>{let a=t?t+o:o;if(typeof i=="object")Kt(i,a+".",r,n);else if(typeof i=="string"||typeof i=="number")n(a,i.toString());else if(typeof i=="boolean")n(a,i?"1":"0");else throw new Error(`Can't handle extra config type: ${typeof i}`)})},ve=e=>{let t=Te(),r=t.stackSave();try{let n=t.stackAlloc(8);t._OrtGetLastError(n,n+4);let o=t.HEAP32[n/4],i=t.HEAPU32[n/4+1],a=i?t.UTF8ToString(i):"";throw new Error(`${e} ERROR_CODE: ${o}, ERROR_MESSAGE: ${a}`)}finally{t.stackRestore(r)}}});var Ua,Va=V(()=>{"use strict";gt();Rr();Ua=e=>{let t=Te(),r=0,n=[],o=e||{};try{if(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?.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?.terminate===void 0&&(o.terminate=!1);let i=0;return e?.tag!==void 0&&(i=ke(e.tag,n)),r=t._OrtCreateRunOptions(o.logSeverityLevel,o.logVerbosityLevel,!!o.terminate,i),r===0&&ve("Can't create run options."),e?.extra!==void 0&&Kt(e.extra,"",new WeakSet,(a,d)=>{let l=ke(a,n),c=ke(d,n);t._OrtAddRunConfigEntry(r,l,c)!==0&&ve(`Can't set a run config entry: ${a} - ${d}.`)}),[r,n]}catch(i){throw r!==0&&t._OrtReleaseRunOptions(r),n.forEach(a=>t._free(a)),i}}});var Rp,Up,Vp,Np,Na,Wa=V(()=>{"use strict";gt();Rr();Rp=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}`)}},Up=e=>{switch(e){case"sequential":return 0;case"parallel":return 1;default:throw new Error(`unsupported execution mode: ${e}`)}},Vp=e=>{e.extra||(e.extra={}),e.extra.session||(e.extra.session={});let t=e.extra.session;t.use_ort_model_bytes_directly||(t.use_ort_model_bytes_directly="1"),e.executionProviders&&e.executionProviders.some(r=>(typeof r=="string"?r:r.name)==="webgpu")&&(e.enableMemPattern=!1)},Np=(e,t,r)=>{for(let n of t){let o=typeof n=="string"?n:n.name;switch(o){case"webnn":if(o="WEBNN",typeof n!="string"){let d=n?.deviceType;if(d){let l=ke("deviceType",r),c=ke(d,r);Te()._OrtAddSessionConfigEntry(e,l,c)!==0&&ve(`Can't set a session config entry: 'deviceType' - ${d}.`)}}break;case"webgpu":if(o="JS",typeof n!="string"){let a=n;if(a?.preferredLayout){if(a.preferredLayout!=="NCHW"&&a.preferredLayout!=="NHWC")throw new Error(`preferredLayout must be either 'NCHW' or 'NHWC': ${a.preferredLayout}`);let d=ke("preferredLayout",r),l=ke(a.preferredLayout,r);Te()._OrtAddSessionConfigEntry(e,d,l)!==0&&ve(`Can't set a session config entry: 'preferredLayout' - ${a.preferredLayout}.`)}}break;case"wasm":case"cpu":continue;default:throw new Error(`not supported execution provider: ${o}`)}let i=ke(o,r);Te()._OrtAppendExecutionProvider(e,i)!==0&&ve(`Can't append execution provider: ${o}.`)}},Na=e=>{let t=Te(),r=0,n=[],o=e||{};Vp(o);try{let i=Rp(o.graphOptimizationLevel??"all"),a=Up(o.executionMode??"sequential"),d=typeof o.logId=="string"?ke(o.logId,n):0,l=o.logSeverityLevel??2;if(!Number.isInteger(l)||l<0||l>4)throw new Error(`log serverity level is not valid: ${l}`);let c=o.logVerbosityLevel??0;if(!Number.isInteger(c)||c<0||c>4)throw new Error(`log verbosity level is not valid: ${c}`);let m=typeof o.optimizedModelFilePath=="string"?ke(o.optimizedModelFilePath,n):0;if(r=t._OrtCreateSessionOptions(i,!!o.enableCpuMemArena,!!o.enableMemPattern,a,!!o.enableProfiling,0,d,l,c,m),r===0&&ve("Can't create session options."),o.executionProviders&&Np(r,o.executionProviders,n),o.enableGraphCapture!==void 0){if(typeof o.enableGraphCapture!="boolean")throw new Error(`enableGraphCapture must be a boolean value: ${o.enableGraphCapture}`);let u=ke("enableGraphCapture",n),h=ke(o.enableGraphCapture.toString(),n);t._OrtAddSessionConfigEntry(r,u,h)!==0&&ve(`Can't set a session config entry: 'enableGraphCapture' - ${o.enableGraphCapture}.`)}if(o.freeDimensionOverrides)for(let[u,h]of Object.entries(o.freeDimensionOverrides)){if(typeof u!="string")throw new Error(`free dimension override name must be a string: ${u}`);if(typeof h!="number"||!Number.isInteger(h)||h<0)throw new Error(`free dimension override value must be a non-negative integer: ${h}`);let w=ke(u,n);t._OrtAddFreeDimensionOverride(r,w,h)!==0&&ve(`Can't set a free dimension override: ${u} - ${h}.`)}return o.extra!==void 0&&Kt(o.extra,"",new WeakSet,(u,h)=>{let w=ke(u,n),g=ke(h,n);t._OrtAddSessionConfigEntry(r,w,g)!==0&&ve(`Can't set a session config entry: ${u} - ${h}.`)}),[r,n]}catch(i){throw r!==0&&t._OrtReleaseSessionOptions(r),n.forEach(a=>t._free(a)),i}}});var Yt,yt,It,Ur,Xt,Vr,Nr,Kn,Q=V(()=>{"use strict";Yt=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}`)}},yt=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}`)}},It=(e,t)=>{let r=[-1,4,1,1,2,2,4,8,-1,1,2,8,4,8,-1,-1,-1,-1,-1,-1,-1,.5,.5][e],n=typeof t=="number"?t:t.reduce((o,i)=>o*i,1);return r>0?Math.ceil(n*r):void 0},Ur=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}`)}},Xt=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}`)}},Vr=e=>e==="float32"||e==="float16"||e==="int32"||e==="int64"||e==="uint32"||e==="uint8"||e==="bool"||e==="uint4"||e==="int4",Nr=e=>e==="float32"||e==="float16"||e==="int32"||e==="int64"||e==="uint32"||e==="uint64"||e==="int8"||e==="uint8"||e==="bool",Kn=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}`)}}});var Zt,Yn=V(()=>{"use strict";Cr();Zt=async e=>{if(typeof e=="string")if(false){}else{let t=await fetch(e);if(!t.ok)throw new Error(`failed to load external data file: ${e}`);let r=t.headers.get("Content-Length"),n=r?parseInt(r,10):0;if(n<1073741824)return new Uint8Array(await t.arrayBuffer());{if(!t.body)throw new Error(`failed to load external data file: ${e}, no response body.`);let o=t.body.getReader(),i;try{i=new ArrayBuffer(n)}catch(d){if(d instanceof RangeError){let l=Math.ceil(n/65536);i=new WebAssembly.Memory({initial:l,maximum:l}).buffer}else throw d}let a=0;for(;;){let{done:d,value:l}=await o.read();if(d)break;let c=l.byteLength;new Uint8Array(i,a,c).set(l),a+=c}return new Uint8Array(i,0,n)}}else return e instanceof Blob?new Uint8Array(await e.arrayBuffer()):e instanceof Uint8Array?e:new Uint8Array(e)}});var Wp,Lp,La,Ha,Wr,Hp,me,Xe=V(()=>{"use strict";Q();Wp=["V","I","W","E","F"],Lp=(e,t)=>{console.log(`[${Wp[e]},${new Date().toISOString()}]${t}`)},Wr=(e,t)=>{La=e,Ha=t},Hp=(e,t)=>{let r=Xt(e),n=Xt(La);r>=n&&Lp(r,typeof t=="function"?t():t)},me=(...e)=>{Ha&&Hp(...e)}});var Lr,Xn=V(()=>{"use strict";Q();Lr=(e,t)=>new(Ur(t))(e)});var Hr=V(()=>{"use strict"});var Ga,Zn,Qn,Gp,Fp,Fa,eo,Jn,ja,Ka=V(()=>{"use strict";Xe();Hr();Ga=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]]),Zn=[],Qn=e=>Math.ceil(e/16)*16,Gp=e=>{for(let t=0;t<Zn.length;t++){let r=Zn[t];if(e<=r)return r}return Math.ceil(e/16)*16},Fp=1,Fa=()=>Fp++,eo=async(e,t,r,n)=>{let o=Qn(r),i=e.device.createBuffer({size:o,usage:GPUBufferUsage.COPY_DST|GPUBufferUsage.MAP_READ});try{let a=e.getCommandEncoder();e.endComputePass(),a.copyBufferToBuffer(t,0,i,0,o),e.flush(),await i.mapAsync(GPUMapMode.READ);let d=i.getMappedRange();if(n){let l=n();return l.set(new Uint8Array(d,0,r)),l}else return new Uint8Array(d.slice(0,r))}finally{i.destroy()}},Jn=class{constructor(t){this.backend=t;this.storageCache=new Map,this.freeBuffers=new Map,this.freeUniformBuffers=new Map,this.buffersForUploadingPending=[],this.buffersPending=[],this.capturedPendingBuffers=new Map;for(let[r]of Ga)Zn.push(r),this.freeBuffers.set(r,[]),this.freeUniformBuffers.set(r,[])}upload(t,r){let n=r.buffer,o=r.byteOffset,i=r.byteLength,a=Qn(i),d=this.storageCache.get(t);if(!d)throw new Error("gpu data for uploading does not exist");if(d.originalSize!==i)throw new Error(`inconsistent data size. gpu data size=${d.originalSize}, data size=${i}`);let l=this.backend.device.createBuffer({mappedAtCreation:!0,size:a,usage:GPUBufferUsage.MAP_WRITE|GPUBufferUsage.COPY_SRC}),c=l.getMappedRange();new Uint8Array(c).set(new Uint8Array(n,o,i)),l.unmap();let m=this.backend.getCommandEncoder();this.backend.endComputePass(),m.copyBufferToBuffer(l,0,d.gpuData.buffer,0,a),me("verbose",()=>`[WebGPU] GpuDataManager.upload(id=${t})`),this.buffersForUploadingPending.push(l)}memcpy(t,r){let n=this.storageCache.get(t);if(!n)throw new Error("source gpu data for memcpy does not exist");let o=this.storageCache.get(r);if(!o)throw new Error("destination gpu data for memcpy does not exist");if(n.originalSize!==o.originalSize)throw new Error("inconsistent source and destination gpu data size");let i=Qn(n.originalSize),a=this.backend.getCommandEncoder();this.backend.endComputePass(),a.copyBufferToBuffer(n.gpuData.buffer,0,o.gpuData.buffer,0,i)}registerExternalBuffer(t,r,n){let o;if(n){if(o=n[0],t===n[1])return me("verbose",()=>`[WebGPU] GpuDataManager.registerExternalBuffer(size=${r}) => id=${o}, buffer is the same, skip.`),o;if(this.backend.capturedCommandList.has(this.backend.currentSessionId))throw new Error(`Registering a different external buffer under graph capture mode is not supported yet.
+             Please use the previous external buffer!`)}else o=Fa();return this.storageCache.set(o,{gpuData:{id:o,type:0,buffer:t},originalSize:r}),me("verbose",()=>`[WebGPU] GpuDataManager.registerExternalBuffer(size=${r}) => id=${o}, registered.`),o}unregisterExternalBuffer(t){t!==void 0&&(this.storageCache.delete(t),me("verbose",()=>`[WebGPU] GpuDataManager.unregisterExternalBuffer() => id=${t}`))}create(t,r=GPUBufferUsage.STORAGE|GPUBufferUsage.COPY_SRC|GPUBufferUsage.COPY_DST){let n=Gp(t),o,i=(r&GPUBufferUsage.STORAGE)===GPUBufferUsage.STORAGE,a=(r&GPUBufferUsage.UNIFORM)===GPUBufferUsage.UNIFORM;if(i||a){let c=(i?this.freeBuffers:this.freeUniformBuffers).get(n);c?c.length>0?o=c.pop():o=this.backend.device.createBuffer({size:n,usage:r}):o=this.backend.device.createBuffer({size:n,usage:r})}else o=this.backend.device.createBuffer({size:n,usage:r});let d={id:Fa(),type:0,buffer:o};return this.storageCache.set(d.id,{gpuData:d,originalSize:t}),me("verbose",()=>`[WebGPU] GpuDataManager.create(size=${t}) => id=${d.id}`),d}get(t){return this.storageCache.get(t)?.gpuData}release(t){let r=this.storageCache.get(t);if(!r)throw new Error("releasing data does not exist");return me("verbose",()=>`[WebGPU] GpuDataManager.release(id=${t}), gpuDataId=${r.gpuData.id}`),this.storageCache.delete(t),this.buffersPending.push(r.gpuData.buffer),r.originalSize}async download(t,r){let n=this.storageCache.get(t);if(!n)throw new Error("data does not exist");await eo(this.backend,n.gpuData.buffer,n.originalSize,r)}refreshPendingBuffers(){for(let t of this.buffersForUploadingPending)t.destroy();if(this.buffersForUploadingPending=[],this.buffersPending.length!==0)if(this.backend.sessionStatus==="default"){for(let t of this.buffersPending){let r=Ga.get(t.size);if((t.usage&GPUBufferUsage.STORAGE)===GPUBufferUsage.STORAGE){let n=this.freeBuffers.get(t.size)||[];r===void 0||n.length>=r?t.destroy():n.push(t)}else if((t.usage&GPUBufferUsage.UNIFORM)===GPUBufferUsage.UNIFORM){let n=this.freeUniformBuffers.get(t.size)||[];r===void 0||n.length>=r?t.destroy():n.push(t)}else t.destroy()}this.buffersPending=[]}else{let t=this.capturedPendingBuffers.get(this.backend.currentSessionId);t||(t=[],this.capturedPendingBuffers.set(this.backend.currentSessionId,t));for(let r of this.buffersPending)t.push(r);this.buffersPending=[]}}dispose(){this.freeBuffers.forEach(t=>{t.forEach(r=>{r.destroy()})}),this.freeUniformBuffers.forEach(t=>{t.forEach(r=>{r.destroy()})}),this.storageCache.forEach(t=>{t.gpuData.buffer.destroy()}),this.capturedPendingBuffers.forEach(t=>{t.forEach(r=>{r.destroy()})}),this.storageCache=new Map,this.freeBuffers=new Map,this.freeUniformBuffers=new Map,this.capturedPendingBuffers=new Map}onReleaseSession(t){let r=this.capturedPendingBuffers.get(t);r&&(r.forEach(n=>{n.destroy()}),this.capturedPendingBuffers.delete(t))}},ja=(...e)=>new Jn(...e)});var to,J,Ie=V(()=>{"use strict";to=class{constructor(t){Object.assign(this,t)}get cacheKey(){return this.key||(this.key=Object.getOwnPropertyNames(this).sort().map(t=>`${this[t]}`).join(";")),this.key}},J=e=>new to(e)});var ro,rt,k,Ct,Gr,Ya,Xa,ie=V(()=>{"use strict";ro=class{static calcMatMulShape(t,r){return t[1]!==r[0]?void 0:[t[0],r[1]]}},rt=class{static calcShape(t,r,n=!1){let o=t.length,i=r.length;if(o===0)return r;if(i===0)return t;let a=Math.max(t.length,r.length),d=new Array(a);if(n){if(o<2||i<2)return;let l=ro.calcMatMulShape([t[o-2],t[o-1]],[r[i-2],r[i-1]]);if(l===void 0)return;[d[a-2],d[a-1]]=l}for(let l=n?3:1;l<=a;l++){let c=o-l<0?1:t[o-l],m=i-l<0?1:r[i-l];if(c!==m&&c>1&&m>1)return;let u=Math.max(c,m);if(c&&m)d[a-l]=Math.max(c,m);else{if(u>1)return;d[a-l]=0}}return d}static isValidBroadcast(t,r){let n=t.length,o=r.length;if(n>o)return!1;for(let i=1;i<=n;i++)if(t[n-i]!==1&&t[n-i]!==r[o-i])return!1;return!0}},k=class e{static size(t){return e.getSizeFromDimensionRange(t,0,t.length)}static convertShape(t,r=4){let n=t.length;if(n===0)return[];let o=new Array(n),i=n-1;for(;i>=0;){if(t[i]%r===0){o[i]=t[i]/r;break}if(r%t[i]!==0)throw new Error("cannot convert shape");o[i]=1,r/=t[i],i--}for(i--;i>=0;i--)o[i]=t[i];return o}static sizeFromDimension(t,r){if(r<0||r>t.length)throw new Error(`invalid dimension of ${r} for sizeFromDimension as Tensor has ${t.length} dimensions.`);return e.getSizeFromDimensionRange(t,r,t.length)}static sizeToDimension(t,r){if(r<0||r>t.length)throw new Error(`invalid dimension of ${r} for sizeToDimension as Tensor has ${t.length} dimensions.`);return e.getSizeFromDimensionRange(t,0,r)}static getSizeFromDimensionRange(t,r,n){let o=1;for(let i=r;i<n;i++){if(t[i]<0)throw new Error("cannot get valid size from specified dimension range. Most likely the range contains negative values in them.");o*=t[i]}return o}static computeStrides(t){let r=t.length;if(r===0)return[];if(r===1)return[1];let n=new Array(r);n[r-1]=1,n[r-2]=t[r-1];for(let o=r-3;o>=0;--o)n[o]=n[o+1]*t[o+1];return n}static normalizeAxis(t,r){if(t<-r&&t>=r)throw new Error("unsupported axis for this operation.");return t<0?t+r:t}static normalizeAxes(t,r){return t.map(n=>this.normalizeAxis(n,r??t.length))}static sortBasedOnPerm(t,r){return r?r.map(n=>t[n]):t.slice().reverse()}static padShape(t,r){let n=t.length;return t.map((o,i)=>o+r[i]+r[i+n])}static areEqual(t,r){return t.length!==r.length?!1:t.every((n,o)=>n===r[o])}},Ct=class e{static adjustPoolAttributes(t,r,n,o,i,a){if(!t&&n.length!==r.length-2)throw new Error("length of specified kernel shapes should be 2 less than length of input dimensions");if(t)for(let d=0;d<r.length-2;d++)d>=n.length?n.push(r[d+2]):n[d]=r[d+2];for(let d=0;d<n.length;d++)if(d<o.length){if(o[d]<0)throw new Error("strides should be greater than or equal to 1")}else o.push(1);for(let d=0;d<n.length;d++)if(d<i.length){if(i[d]<0)throw new Error("dilations should be greater than or equal to 1")}else i.push(1);for(let d=0;d<n.length*2;d++)if(d<a.length){if(a[d]<0)throw new Error("pad should be greater than or equal to 1")}else a.push(0);for(let d=0;d<n.length;d++){if(n[d]<=0)throw new Error("kernel shapes need to be greater than 0");if(a[d]>=n[d]||a[d+n.length]>=n[d])throw new Error("pads should be smaller than kernel")}}static adjustPadsBasedOnAutoPad(t,r,n,o,i,a,d){if(d){if(i.length!==2*(t.length-2))throw new Error("length of pads should be twice the length of data dimensions");if(r.length!==t.length-2)throw new Error("length of strides should be the length of data dimensions");if(o.length!==t.length-2)throw new Error("length of kernel shapes should be the length of data dimensions");for(let l=0;l<t.length-2;l++)e.adjustPadAndReturnShape(t[l+(a?1:2)],r[l],n[l],o[l],i,l,l+t.length-2,d)}}static computePoolOutputShape(t,r,n,o,i,a,d){if(r.length<=0)throw new Error("input shape must be of size greater than 0");let l=[r[0],r[1]];return e.computeShapeHelper(t,r,l,n,o,i,a,d),l}static computeConvOutputShape(t,r,n,o,i,a,d){if(t.length<=0||r.length<=0)throw new Error("invalid input tensor dims or invalid filter tensor dims");let l=[t[0],r[0]];return e.computeShapeHelper(!1,t,l,n,o,i,a,d),l}static computeShapeHelper(t,r,n,o,i,a,d,l){if(t)for(let c=0;c<r.length-2;c++)n.push(1);else for(let c=0;c<r.length-2;c++)n.push(e.adjustPadAndReturnShape(r[c+2],o[c],i[c],a[c],d,c,c+r.length-2,l))}static adjustPadAndReturnShape(t,r,n,o,i,a,d,l){let c=n*(o-1)+1;if(l&&l!=="NOTSET")switch(l){case"VALID":return i[a]=0,i[d]=0,Math.floor((t-c)/r+1);case"SAME_LOWER":case"SAME_UPPER":if(n!==1)throw new Error("Dilation not supported for SAME_UPPER or SAME_LOWER");{let u=((t+r-1)/r-1)*r+o-t;return i[a]=Math.floor(l==="SAME_LOWER"?(u+1)/2:u/2),i[d]=u-i[a],Math.floor((t+u-o)/r+1)}default:throw new Error("Unsupported AutoPad type")}else return Math.floor((t+i[a]+i[d]-c)/r+1)}},Gr=class{static getShapeOfGemmResult(t,r,n,o,i){if(t.length!==2||n.length!==2)throw new Error("shape need to be of size 2");let a,d,l;r?(a=t[1],d=t[0]):(a=t[0],d=t[1]);let c=-1;if(o?(l=n[0],c=1):(l=n[1],c=0),n[c]!==d)throw new Error("dimension mismatch");if(a<=0||l<=0||d<=0)throw new Error("invalid shape specified");if(i&&!rt.isValidBroadcast(i,[a,l]))throw new Error("gemm: invalid bias shape for broadcast");return[a,l,d]}},Ya=-34028234663852886e22,Xa=34028234663852886e22});var At,oo,he,Ee,R,we,io,kt,Ze,F,ao,E,M,Fr,no,Za,Nt,ae=V(()=>{"use strict";Q();ie();At=64,oo=(e,t)=>{if(t===3)throw new Error("vec3 has same alignment as vec4, use vec4 instead");switch(e){case 10:return t>1?`vec${t}<f16>`:"f16";case 1:return t>1?`vec${t}<f32>`:"f32";case 6:return t>1?`vec${t}<i32>`:"i32";case 12:return t>1?`vec${t}<u32>`:"u32";case 7:if(t>1)throw new Error("currently not supported vecX of uint64 yet");return["vec2<u32>","i32"];case 13:if(t>1)throw new Error("currently not supported vecX of uint64 yet");return["vec2<u32>","u32"];case 9:if(t!==4)throw new Error("bool must be vec4");return["u32","vec4<bool>"];case 22:return"i32";case 21:return"u32";default:throw new Error(`Unknown data type: ${e}`)}},he=(e,t=1)=>{let r=oo(e,t);return typeof r=="string"?r:r[0]},Ee=(e,t=1)=>{let r=oo(e,t);return typeof r=="string"?r:r[1]},R=(...e)=>{let t=[];return e.forEach(r=>{r.length!==0&&t.push({type:12,data:r},{type:12,data:k.computeStrides(r)})}),t},we=e=>e%4===0?4:e%2===0?2:1,io=(e="f32",t,r="0")=>!t||t===1?`${e}(${r})`:`vec${t}<${e}>(${r})`,kt=(e,t,r)=>e==="f32"?r:t===1?`f32(${r})`:`vec${t}<f32>(${r})`,Ze=(e,t)=>t===4?`(${e}.x + ${e}.y + ${e}.z + ${e}.w)`:t===2?`(${e}.x + ${e}.y)`:t===3?`(${e}.x + ${e}.y + ${e}.z)`:e,F=(e,t,r,n)=>e.startsWith("uniforms.")&&r>4?typeof t=="string"?n==="f16"?`${e}[(${t}) / 8][(${t}) % 8 / 4][(${t}) % 8 % 4]`:`${e}[(${t}) / 4][(${t}) % 4]`:n==="f16"?`${e}[${Math.floor(t/8)}][${Math.floor(t%8/4)}][${t%8%4}]`:`${e}[${Math.floor(t/4)}][${t%4}]`:r>1?`${e}[${t}]`:e,ao=(e,t,r,n,o)=>{let i=typeof r=="number",a=i?r:r.length,d=[...new Array(a).keys()],l=a<2?"u32":a<=4?`vec${a}<u32>`:`array<u32, ${a}>`,c=oo(t,o),m=typeof c=="string"?c:c[1],u=typeof c=="string"?c:c[0],h={indices:l,value:m,storage:u,tensor:t},w=U=>typeof U=="string"?U:`${U}u`,g={offsetToIndices:!1,indicesToOffset:!1,broadcastedIndicesToOffset:!1,set:!1,setByIndices:!1,get:!1,getByIndices:!1},y=i?"uniforms.":"",S=`${y}${e}_shape`,$=`${y}${e}_strides`,v="";for(let U=0;U<a-1;U++)v+=`
+    let dim${U} = current / ${F($,U,a)};
+    let rest${U} = current % ${F($,U,a)};
+    indices[${U}] = dim${U};
+    current = rest${U};
+    `;v+=`indices[${a-1}] = current;`;let x=a<2?"":`
   fn o2i_${e}(offset: u32) -> ${h.indices} {
     var indices: ${h.indices};
     var current = offset;
-    ${_}
+    ${v}
     return indices;
-  }`,T=R=>(g.offsetToIndices=!0,a<2?R:`o2i_${e}(${R})`),C=[];if(a>=2)for(let R=a-1;R>=0;R--)C.push(`${F($,R,a)} * (indices[${R}])`);let A=a<2?"":`
+  }`,T=U=>(g.offsetToIndices=!0,a<2?U:`o2i_${e}(${U})`),C=[];if(a>=2)for(let U=a-1;U>=0;U--)C.push(`${F($,U,a)} * (indices[${U}])`);let A=a<2?"":`
   fn i2o_${e}(indices: ${h.indices}) -> u32 {
     return ${C.join("+")};
-  }`,P=R=>(g.indicesToOffset=!0,a<2?R:`i2o_${e}(${R})`),D=(...R)=>a===0?"0u":`${h.indices}(${R.map(w).join(",")})`,N=(R,G)=>a<2?`${R}`:`${F(R,G,a)}`,L=(R,G,he)=>a<2?`${R}=${he};`:`${F(R,G,a)}=${he};`,K={},X=(R,G)=>{g.broadcastedIndicesToOffset=!0;let he=`${G.name}broadcastedIndicesTo${e}Offset`;if(he in K)return`${he}(${R})`;let Re=[];for(let $e=a-1;$e>=0;$e--){let Ce=G.indicesGet("outputIndices",$e+G.rank-a);Re.push(`${N($,$e)} * (${Ce} % ${N(S,$e)})`)}return K[he]=`fn ${he}(outputIndices: ${G.type.indices}) -> u32 {
+  }`,P=U=>(g.indicesToOffset=!0,a<2?U:`i2o_${e}(${U})`),B=(...U)=>a===0?"0u":`${h.indices}(${U.map(w).join(",")})`,N=(U,G)=>a<2?`${U}`:`${F(U,G,a)}`,W=(U,G,ye)=>a<2?`${U}=${ye};`:`${F(U,G,a)}=${ye};`,K={},Z=(U,G)=>{g.broadcastedIndicesToOffset=!0;let ye=`${G.name}broadcastedIndicesTo${e}Offset`;if(ye in K)return`${ye}(${U})`;let Re=[];for(let $e=a-1;$e>=0;$e--){let Ce=G.indicesGet("outputIndices",$e+G.rank-a);Re.push(`${N($,$e)} * (${Ce} % ${N(S,$e)})`)}return K[ye]=`fn ${ye}(outputIndices: ${G.type.indices}) -> u32 {
              return ${Re.length>0?Re.join("+"):"0u"};
-           }`,`${he}(${R})`},ie=(R,G)=>(()=>{if(h.storage===h.value)return`${e}[${R}]=${G};`;if(h.storage==="vec2<u32>"&&h.value==="i32")return`${e}[${R}]=vec2<u32>(u32(${G}), select(0u, 0xFFFFFFFFu, ${G} < 0));`;if(h.storage==="vec2<u32>"&&h.value==="u32")return`${e}[${R}]=vec2<u32>(u32(${G}), 0u);`;if(h.storage==="u32"&&h.value==="vec4<bool>")return`${e}[${R}]=dot(vec4<u32>(0x1, 0x100, 0x10000, 0x1000000), vec4<u32>(${G}));`;throw new Error(`not supported combination of storage type ${h.storage} and value type ${h.value} yet`)})(),ae=R=>(()=>{if(h.storage===h.value)return`${e}[${R}]`;if(h.storage==="vec2<u32>"&&h.value==="i32")return`i32(${e}[${R}].x)`;if(h.storage==="vec2<u32>"&&h.value==="u32")return`u32(${e}[${R}].x)`;if(h.storage==="u32"&&h.value==="vec4<bool>")return`vec4<bool>(bool(${e}[${R}] & 0xFFu), bool(${e}[${R}] & 0xFF00u), bool(${e}[${R}] & 0xFF0000u), bool(${e}[${R}] & 0xFF000000u))`;throw new Error(`not supported combination of storage type ${h.storage} and value type ${h.value} yet`)})(),le=a<2?"":`
+           }`,`${ye}(${U})`},ee=(U,G)=>(()=>{if(h.storage===h.value)return`${e}[${U}]=${G};`;if(h.storage==="vec2<u32>"&&h.value==="i32")return`${e}[${U}]=vec2<u32>(u32(${G}), select(0u, 0xFFFFFFFFu, ${G} < 0));`;if(h.storage==="vec2<u32>"&&h.value==="u32")return`${e}[${U}]=vec2<u32>(u32(${G}), 0u);`;if(h.storage==="u32"&&h.value==="vec4<bool>")return`${e}[${U}]=dot(vec4<u32>(0x1, 0x100, 0x10000, 0x1000000), vec4<u32>(${G}));`;throw new Error(`not supported combination of storage type ${h.storage} and value type ${h.value} yet`)})(),se=U=>(()=>{if(h.storage===h.value)return`${e}[${U}]`;if(h.storage==="vec2<u32>"&&h.value==="i32")return`i32(${e}[${U}].x)`;if(h.storage==="vec2<u32>"&&h.value==="u32")return`u32(${e}[${U}].x)`;if(h.storage==="u32"&&h.value==="vec4<bool>")return`vec4<bool>(bool(${e}[${U}] & 0xFFu), bool(${e}[${U}] & 0xFF00u), bool(${e}[${U}] & 0xFF0000u), bool(${e}[${U}] & 0xFF000000u))`;throw new Error(`not supported combination of storage type ${h.storage} and value type ${h.value} yet`)})(),de=a<2?"":`
   fn get_${e}ByIndices(indices: ${h.indices}) -> ${m} {
-    return ${ae(`i2o_${e}(indices)`)};
-  }`,Y=a<2?"":(()=>{let R=d.map(he=>`d${he}: u32`).join(", "),G=d.map(he=>`d${he}`).join(", ");return`
-  fn get_${e}(${R}) -> ${m} {
-    return get_${e}ByIndices(${D(G)});
-  }`})(),me=(...R)=>{if(R.length!==a)throw new Error(`indices length must be ${a}`);let G=R.map(w).join(",");return a===0?ae("0u"):a===1?ae(G[0]):(g.get=!0,g.getByIndices=!0,g.indicesToOffset=!0,`get_${e}(${G})`)},pe=R=>a<2?ae(R):(g.getByIndices=!0,g.indicesToOffset=!0,`get_${e}ByIndices(${R})`),Q=a<2?"":`
+    return ${se(`i2o_${e}(indices)`)};
+  }`,Y=a<2?"":(()=>{let U=d.map(ye=>`d${ye}: u32`).join(", "),G=d.map(ye=>`d${ye}`).join(", ");return`
+  fn get_${e}(${U}) -> ${m} {
+    return get_${e}ByIndices(${B(G)});
+  }`})(),le=(...U)=>{if(U.length!==a)throw new Error(`indices length must be ${a}`);let G=U.map(w).join(",");return a===0?se("0u"):a===1?se(G[0]):(g.get=!0,g.getByIndices=!0,g.indicesToOffset=!0,`get_${e}(${G})`)},ce=U=>a<2?se(U):(g.getByIndices=!0,g.indicesToOffset=!0,`get_${e}ByIndices(${U})`),q=a<2?"":`
   fn set_${e}ByIndices(indices: ${h.indices}, value: ${m}) {
-    ${ie(`i2o_${e}(indices)`,"value")}
-  }`,be=a<2?"":(()=>{let R=d.map(he=>`d${he}: u32`).join(", "),G=d.map(he=>`d${he}`).join(", ");return`
-  fn set_${e}(${R}, value: ${m}) {
-    set_${e}ByIndices(${D(G)}, value);
-  }`})();return{impl:()=>{let R=[],G=!1;return g.offsetToIndices&&(R.push(x),G=!0),g.indicesToOffset&&(R.push(A),G=!0),g.broadcastedIndicesToOffset&&(Object.values(K).forEach(he=>R.push(he)),G=!0),g.set&&(R.push(be),G=!0),g.setByIndices&&(R.push(Q),G=!0),g.get&&(R.push(Y),G=!0),g.getByIndices&&(R.push(le),G=!0),!i&&G&&R.unshift(`const ${S} = ${h.indices}(${r.join(",")});`,`const ${$} = ${h.indices}(${E.computeStrides(r).join(",")});`),R.join(`
-`)},type:h,offsetToIndices:T,indicesToOffset:P,broadcastedIndicesToOffset:X,indices:D,indicesGet:N,indicesSet:L,set:(...R)=>{if(R.length!==a+1)throw new Error(`indices length must be ${a}`);let G=R[a];if(typeof G!="string")throw new Error("value must be string");let he=R.slice(0,a).map(w).join(",");return a===0?ie("0u",G):a===1?ie(he[0],G):(g.set=!0,g.setByIndices=!0,g.indicesToOffset=!0,`set_${e}(${he}, ${G})`)},setByOffset:ie,setByIndices:(R,G)=>a<2?ie(R,G):(g.setByIndices=!0,g.indicesToOffset=!0,`set_${e}ByIndices(${R}, ${G});`),get:me,getByOffset:ae,getByIndices:pe,usage:n,name:e,strides:$,shape:S,rank:a}},k=(e,t,r,n=1)=>ao(e,t,r,"input",n),M=(e,t,r,n=1)=>ao(e,t,r,"output",n),Fr=(e,t,r,n=1)=>ao(e,t,r,"internal",n),no=class{constructor(t,r){this.normalizedDispatchGroup=t;this.limits=r;this.internalVariables=[];this.variables=[];this.uniforms=[];this.variableIndex=0}guardAgainstOutOfBoundsWorkgroupSizes(t){return`if (global_idx >= ${typeof t=="number"?`${t}u`:t}) { return; }`}mainStart(t=At){let r=typeof t=="number"?t:t[0],n=typeof t=="number"?1:t[1],o=typeof t=="number"?1:t[2];if(r>this.limits.maxComputeWorkgroupSizeX||n>this.limits.maxComputeWorkgroupSizeY||o>this.limits.maxComputeWorkgroupSizeZ)throw new Error(`workgroup size [${r}, ${n}, ${o}] exceeds the maximum workgroup size [${this.limits.maxComputeWorkgroupSizeX}, ${this.limits.maxComputeWorkgroupSizeY}, ${this.limits.maxComputeWorkgroupSizeZ}].`);if(r*n*o>this.limits.maxComputeInvocationsPerWorkgroup)throw new Error(`workgroup size [${r}, ${n}, ${o}] exceeds the maximum workgroup invocations ${this.limits.maxComputeInvocationsPerWorkgroup}.`);let i=this.normalizedDispatchGroup[1]===1&&this.normalizedDispatchGroup[2]===1,a=i?`@builtin(global_invocation_id) global_id : vec3<u32>,
+    ${ee(`i2o_${e}(indices)`,"value")}
+  }`,ue=a<2?"":(()=>{let U=d.map(ye=>`d${ye}: u32`).join(", "),G=d.map(ye=>`d${ye}`).join(", ");return`
+  fn set_${e}(${U}, value: ${m}) {
+    set_${e}ByIndices(${B(G)}, value);
+  }`})();return{impl:()=>{let U=[],G=!1;return g.offsetToIndices&&(U.push(x),G=!0),g.indicesToOffset&&(U.push(A),G=!0),g.broadcastedIndicesToOffset&&(Object.values(K).forEach(ye=>U.push(ye)),G=!0),g.set&&(U.push(ue),G=!0),g.setByIndices&&(U.push(q),G=!0),g.get&&(U.push(Y),G=!0),g.getByIndices&&(U.push(de),G=!0),!i&&G&&U.unshift(`const ${S} = ${h.indices}(${r.join(",")});`,`const ${$} = ${h.indices}(${k.computeStrides(r).join(",")});`),U.join(`
+`)},type:h,offsetToIndices:T,indicesToOffset:P,broadcastedIndicesToOffset:Z,indices:B,indicesGet:N,indicesSet:W,set:(...U)=>{if(U.length!==a+1)throw new Error(`indices length must be ${a}`);let G=U[a];if(typeof G!="string")throw new Error("value must be string");let ye=U.slice(0,a).map(w).join(",");return a===0?ee("0u",G):a===1?ee(ye[0],G):(g.set=!0,g.setByIndices=!0,g.indicesToOffset=!0,`set_${e}(${ye}, ${G})`)},setByOffset:ee,setByIndices:(U,G)=>a<2?ee(U,G):(g.setByIndices=!0,g.indicesToOffset=!0,`set_${e}ByIndices(${U}, ${G});`),get:le,getByOffset:se,getByIndices:ce,usage:n,name:e,strides:$,shape:S,rank:a}},E=(e,t,r,n=1)=>ao(e,t,r,"input",n),M=(e,t,r,n=1)=>ao(e,t,r,"output",n),Fr=(e,t,r,n=1)=>ao(e,t,r,"internal",n),no=class{constructor(t,r){this.normalizedDispatchGroup=t;this.limits=r;this.internalVariables=[];this.variables=[];this.uniforms=[];this.variableIndex=0}guardAgainstOutOfBoundsWorkgroupSizes(t){return`if (global_idx >= ${typeof t=="number"?`${t}u`:t}) { return; }`}mainStart(t=At){let r=typeof t=="number"?t:t[0],n=typeof t=="number"?1:t[1],o=typeof t=="number"?1:t[2];if(r>this.limits.maxComputeWorkgroupSizeX||n>this.limits.maxComputeWorkgroupSizeY||o>this.limits.maxComputeWorkgroupSizeZ)throw new Error(`workgroup size [${r}, ${n}, ${o}] exceeds the maximum workgroup size [${this.limits.maxComputeWorkgroupSizeX}, ${this.limits.maxComputeWorkgroupSizeY}, ${this.limits.maxComputeWorkgroupSizeZ}].`);if(r*n*o>this.limits.maxComputeInvocationsPerWorkgroup)throw new Error(`workgroup size [${r}, ${n}, ${o}] exceeds the maximum workgroup invocations ${this.limits.maxComputeInvocationsPerWorkgroup}.`);let i=this.normalizedDispatchGroup[1]===1&&this.normalizedDispatchGroup[2]===1,a=i?`@builtin(global_invocation_id) global_id : vec3<u32>,
     @builtin(workgroup_id) workgroup_id : vec3<u32>,
+    @builtin(local_invocation_index) local_idx : u32,
     @builtin(local_invocation_id) local_id : vec3<u32>`:`@builtin(global_invocation_id) global_id : vec3<u32>,
                                              @builtin(local_invocation_id) local_id : vec3<u32>,
     @builtin(local_invocation_index) local_idx : u32,
     @builtin(workgroup_id) workgroup_id : vec3<u32>,
     @builtin(num_workgroups) num_workgroups : vec3<u32>`,d=i?`let global_idx = global_id.x;
-         let local_idx = local_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 * ${r*n*o}u + local_idx;`;return`@compute @workgroup_size(${r}, ${n}, ${o})
@@ -3856,8 +3856,8 @@ var Un=Object.defineProperty;var _p=Object.getOwnPropertyDescriptor;var $p=Objec
       struct Uniforms { ${t.join(", ")} };
       @group(0) @binding(${this.variableIndex}) var<uniform> uniforms: Uniforms;`}get additionalImplementations(){return this.uniformDeclaration()+this.variables.map(t=>t.impl()).join(`
 `)+this.internalVariables.map(t=>t.impl()).join(`
-`)}get variablesInfo(){if(this.uniforms.length===0)return;let t=r=>[12,10,1,6][["u32","f16","f32","i32"].indexOf(r)];return this.uniforms.map(r=>[t(r.type),r.length??1])}},Za=(e,t)=>new no(e,t),Nt=(e,t)=>{let r=e.length,n=[];for(let o=0;o<r;o++){let i=r-1-o,a=e[i]||1;(t[t.length-1-o]||1)>1&&a===1&&n.unshift(i)}return n}});var qp,Qa,jp,Kp,Yp,Pe,Ja,es,lt=U(()=>{"use strict";Z();te();Ie();re();qp=e=>{if(!e||e.length!==1)throw new Error("Transpose requires 1 input.")},Qa=(e,t)=>t&&t.length!==e?[...new Array(e).keys()].reverse():t,jp=(e,t)=>E.sortBasedOnPerm(e,Qa(e.length,t)),Kp=(e,t,r,n)=>{let o=`fn perm(i: ${n.type.indices}) -> ${r.type.indices} {
-    var a: ${r.type.indices};`;for(let i=0;i<t;++i)o+=r.indicesSet("a",e[i],`i[${i}]`);return o+="return a;}"},Yp=(e,t)=>{let r=[],n=[];for(let o=0;o<e.length;++o)e[o]!==1&&r.push(e[o]),e[t[o]]!==1&&n.push(t[o]);return{newShape:r,newPerm:n}},Pe=(e,t)=>{let r=e.dataType,n=e.dims.length,o=Qa(n,t),i=jp(e.dims,o),{newShape:a,newPerm:d}=Yp(e.dims,o),l=E.areEqual(d,[2,3,1]),c=E.areEqual(d,[3,1,2]),m=a.length===2&&d[0]>d[1]||l||c,u=m?a:e.dims,h=i;m&&(u=l?[a[0],a[1]*a[2]]:c?[a[0]*a[1],a[2]]:a,h=[u[1],u[0]]);let w=k("a",r,u.length),g=M("output",r,h.length),y=16,S;return m?S=$=>`
+`)}get variablesInfo(){if(this.uniforms.length===0)return;let t=r=>[12,10,1,6][["u32","f16","f32","i32"].indexOf(r)];return this.uniforms.map(r=>[t(r.type),r.length??1])}},Za=(e,t)=>new no(e,t),Nt=(e,t)=>{let r=e.length,n=[];for(let o=0;o<r;o++){let i=r-1-o,a=e[i]||1;(t[t.length-1-o]||1)>1&&a===1&&n.unshift(i)}return n}});var qp,Qa,jp,Kp,Yp,Pe,Ja,es,lt=V(()=>{"use strict";Q();ie();Ie();ae();qp=e=>{if(!e||e.length!==1)throw new Error("Transpose requires 1 input.")},Qa=(e,t)=>t&&t.length!==e?[...new Array(e).keys()].reverse():t,jp=(e,t)=>k.sortBasedOnPerm(e,Qa(e.length,t)),Kp=(e,t,r,n)=>{let o=`fn perm(i: ${n.type.indices}) -> ${r.type.indices} {
+    var a: ${r.type.indices};`;for(let i=0;i<t;++i)o+=r.indicesSet("a",e[i],`i[${i}]`);return o+="return a;}"},Yp=(e,t)=>{let r=[],n=[];for(let o=0;o<e.length;++o)e[o]!==1&&r.push(e[o]),e[t[o]]!==1&&n.push(t[o]);return{newShape:r,newPerm:n}},Pe=(e,t)=>{let r=e.dataType,n=e.dims.length,o=Qa(n,t),i=jp(e.dims,o),{newShape:a,newPerm:d}=Yp(e.dims,o),l=k.areEqual(d,[2,3,1]),c=k.areEqual(d,[3,1,2]),m=a.length===2&&d[0]>d[1]||l||c,u=m?a:e.dims,h=i;m&&(u=l?[a[0],a[1]*a[2]]:c?[a[0]*a[1],a[2]]:a,h=[u[1],u[0]]);let w=E("a",r,u.length),g=M("output",r,h.length),y=16,S;return m?S=$=>`
   ${$.registerUniform("output_size","u32").declareVariables(w,g)}
   var<workgroup> tile : array<array<${g.type.value}, ${y+1}>, ${y}>;
   ${$.mainStart([y,y,1])}
@@ -3888,7 +3888,7 @@ var Un=Object.defineProperty;var _p=Object.getOwnPropertyDescriptor;var $p=Objec
     let aIndices = perm(indices);
 
     ${g.setByOffset("global_idx",w.getByIndices("aIndices"))}
-  }`,{name:m?"TransposeShared":"Transpose",shaderCache:{hint:`${t}`,inputDependencies:["rank"]},getRunData:()=>{let $=E.size(i);return{outputs:[{dims:i,dataType:e.dataType}],dispatchGroup:m?{x:Math.ceil(h[1]/y),y:Math.ceil(h[0]/y)}:{x:Math.ceil($/64)},programUniforms:[{type:12,data:$},...V(u,h)]}},getShaderSource:S}},Ja=(e,t)=>{qp(e.inputs),e.compute(Pe(e.inputs[0],t.perm))},es=e=>J({perm:e.perm})});var Xp,Zp,Qp,Jp,em,tm,rm,nm,om,im,nt,ts,rs,ns,os,is,as,ss,us,ds,ls,cs=U(()=>{"use strict";Z();te();re();qr();lt();Xp={max:"select(bestValue, candidate, candidate > bestValue)",min:"select(bestValue, candidate, candidate < bestValue)",mean:"bestValue + candidate",sum:"bestValue + candidate",prod:"bestValue * candidate",sumSquare:"bestValue + candidate * candidate",logSumExp:"bestValue + exp(candidate)",l1:"bestValue + abs(candidate)",l2:"bestValue + candidate * candidate",logSum:"bestValue + candidate"},Zp={max:"select(bestValue, candidate, candidate > bestValue)",min:"select(bestValue, candidate, candidate < bestValue)",mean:"bestValue + candidate",sum:"bestValue + candidate",prod:"bestValue * candidate",sumSquare:"bestValue + candidate",logSumExp:"bestValue + candidate",l1:"bestValue + candidate",l2:"bestValue + candidate",logSum:"bestValue + candidate"},Qp={max:"_A[offset]",min:"_A[offset]",mean:"0",sum:"0",prod:"1",sumSquare:"0",logSumExp:"0",l1:"0",l2:"0",logSum:"0"},Jp={max:"bestValue",min:"bestValue",sum:"bestValue",prod:"bestValue",sumSquare:"bestValue",logSumExp:"log(bestValue)",l1:"bestValue",l2:"sqrt(bestValue)",logSum:"log(bestValue)"},em=(e,t)=>{let r=[];for(let n=t-e;n<t;++n)r.push(n);return r},tm=(e,t)=>{let r=[],n=e.length;for(let i=0;i<n;i++)t.indexOf(i)===-1&&r.push(e[i]);let o=t.map(i=>e[i]);return[r,o]},rm=(e,t)=>{let r=e.length+t.length,n=[],o=0;for(let i=0;i<r;i++)t.indexOf(i)===-1?n.push(e[o++]):n.push(1);return n},nm=(e,t)=>{for(let r=0;r<e.length;++r)if(e[e.length-r-1]!==t-1-r)return!1;return!0},om=(e,t)=>{let r=[];if(!nm(e,t)){for(let n=0;n<t;++n)e.indexOf(n)===-1&&r.push(n);e.forEach(n=>r.push(n))}return r},im=(e,t,r,n,o,i,a)=>{let d=r[0].dims,l=E.size(i),c=E.size(a),m=k("_A",r[0].dataType,d),u=M("output",o,i),h=32,w=`
+  }`,{name:m?"TransposeShared":"Transpose",shaderCache:{hint:`${t}`,inputDependencies:["rank"]},getRunData:()=>{let $=k.size(i);return{outputs:[{dims:i,dataType:e.dataType}],dispatchGroup:m?{x:Math.ceil(h[1]/y),y:Math.ceil(h[0]/y)}:{x:Math.ceil($/64)},programUniforms:[{type:12,data:$},...R(u,h)]}},getShaderSource:S}},Ja=(e,t)=>{qp(e.inputs),e.compute(Pe(e.inputs[0],t.perm))},es=e=>J({perm:e.perm})});var Xp,Zp,Qp,Jp,em,tm,rm,nm,om,im,nt,ts,rs,ns,os,is,as,ss,us,ds,ls,cs=V(()=>{"use strict";Q();ie();ae();qr();lt();Xp={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"},Zp={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"},Qp={max:"_A[offset]",min:"_A[offset]",mean:"0",sum:"0",prod:"1",sumSquare:"0",logSumExp:"0",l1:"0",l2:"0",logSum:"0"},Jp={max:"bestValue",min:"bestValue",sum:"bestValue",prod:"bestValue",sumSquare:"bestValue",logSumExp:"log(bestValue)",l1:"bestValue",l2:"sqrt(bestValue)",logSum:"log(bestValue)"},em=(e,t)=>{let r=[];for(let n=t-e;n<t;++n)r.push(n);return r},tm=(e,t)=>{let r=[],n=e.length;for(let i=0;i<n;i++)t.indexOf(i)===-1&&r.push(e[i]);let o=t.map(i=>e[i]);return[r,o]},rm=(e,t)=>{let r=e.length+t.length,n=[],o=0;for(let i=0;i<r;i++)t.indexOf(i)===-1?n.push(e[o++]):n.push(1);return n},nm=(e,t)=>{for(let r=0;r<e.length;++r)if(e[e.length-r-1]!==t-1-r)return!1;return!0},om=(e,t)=>{let r=[];if(!nm(e,t)){for(let n=0;n<t;++n)e.indexOf(n)===-1&&r.push(n);e.forEach(n=>r.push(n))}return r},im=(e,t,r,n,o,i,a)=>{let d=r[0].dims,l=k.size(i),c=k.size(a),m=E("_A",r[0].dataType,d),u=M("output",o,i),h=32,w=`
           var<workgroup> aBestValues : array<f32, ${h}>;
        `;return{name:e,shaderCache:t,getShaderSource:y=>`
         ${y.registerUniform("reduceSize","u32").declareVariables(m,u)}
@@ -3926,17 +3926,17 @@ var Un=Object.defineProperty;var _p=Object.getOwnPropertyDescriptor;var $p=Objec
          if (local_idx == 0u) {
           ${u.setByOffset("outputIndex",`${n==="mean"?`${u.type.storage}(bestValue / f32(uniforms.reduceSize))`:`${u.type.storage}(${Jp[n]})`}`)};
          }
-        }`,getRunData:()=>({outputs:[{dims:i,dataType:o}],dispatchGroup:{x:l},programUniforms:[{type:12,data:c}]})}},nt=(e,t,r,n)=>{let o=e.inputs.length===1?r:so(e.inputs,r),i=o.axes;i.length===0&&!o.noopWithEmptyAxes&&(i=e.inputs[0].dims.map((w,g)=>g));let a=E.normalizeAxes(i,e.inputs[0].dims.length),d=a,l=e.inputs[0],c=om(d,e.inputs[0].dims.length);c.length>0&&(l=e.compute(Pe(e.inputs[0],c),{inputs:[0],outputs:[-1]})[0],d=em(d.length,l.dims.length));let[m,u]=tm(l.dims,d),h=m;o.keepDims&&(h=rm(m,a)),e.compute(im(t,{hint:o.cacheKey,inputDependencies:["type"]},[l],n,e.inputs[0].dataType,h,u),{inputs:[l]})},ts=(e,t)=>{nt(e,"ReduceMeanShared",t,"mean")},rs=(e,t)=>{nt(e,"ReduceL1Shared",t,"l1")},ns=(e,t)=>{nt(e,"ReduceL2Shared",t,"l2")},os=(e,t)=>{nt(e,"ReduceLogSumExpShared",t,"logSumExp")},is=(e,t)=>{nt(e,"ReduceMaxShared",t,"max")},as=(e,t)=>{nt(e,"ReduceMinShared",t,"min")},ss=(e,t)=>{nt(e,"ReduceProdShared",t,"prod")},us=(e,t)=>{nt(e,"ReduceSumShared",t,"sum")},ds=(e,t)=>{nt(e,"ReduceSumSquareShared",t,"sumSquare")},ls=(e,t)=>{nt(e,"ReduceLogSumShared",t,"logSum")}});var ot,am,jr,so,it,sm,um,dm,lm,cm,pm,mm,fm,hm,gm,at,ps,ms,fs,hs,gs,ys,bs,ws,vs,_s,qr=U(()=>{"use strict";Z();te();Ie();re();cs();ot=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.")},am=e=>["","",`var value = ${e.getByIndices("input_indices")};`,""],jr=(e,t,r,n,o,i,a=!1,d=!1)=>{let l=[],c=r[0].dims,m=c.length,u=E.normalizeAxes(o,m),h=!d&&u.length===0;c.forEach((S,$)=>{h||u.indexOf($)>=0?a&&l.push(1):l.push(S)});let w=l.length,g=E.size(l);return{name:e,shaderCache:t,getShaderSource:S=>{let $=[],_=k("_A",r[0].dataType,m),x=M("output",i,w),T=n(_,x,u),C=T[2];for(let A=0,P=0;A<m;A++)h||u.indexOf(A)>=0?(a&&P++,C=`for(var j${A}: u32 = 0; j${A} < ${c[A]}; j${A}++) {
+        }`,getRunData:()=>({outputs:[{dims:i,dataType:o}],dispatchGroup:{x:l},programUniforms:[{type:12,data:c}]})}},nt=(e,t,r,n)=>{let o=e.inputs.length===1?r:so(e.inputs,r),i=o.axes;i.length===0&&!o.noopWithEmptyAxes&&(i=e.inputs[0].dims.map((w,g)=>g));let a=k.normalizeAxes(i,e.inputs[0].dims.length),d=a,l=e.inputs[0],c=om(d,e.inputs[0].dims.length);c.length>0&&(l=e.compute(Pe(e.inputs[0],c),{inputs:[0],outputs:[-1]})[0],d=em(d.length,l.dims.length));let[m,u]=tm(l.dims,d),h=m;o.keepDims&&(h=rm(m,a)),e.compute(im(t,{hint:o.cacheKey,inputDependencies:["type"]},[l],n,e.inputs[0].dataType,h,u),{inputs:[l]})},ts=(e,t)=>{nt(e,"ReduceMeanShared",t,"mean")},rs=(e,t)=>{nt(e,"ReduceL1Shared",t,"l1")},ns=(e,t)=>{nt(e,"ReduceL2Shared",t,"l2")},os=(e,t)=>{nt(e,"ReduceLogSumExpShared",t,"logSumExp")},is=(e,t)=>{nt(e,"ReduceMaxShared",t,"max")},as=(e,t)=>{nt(e,"ReduceMinShared",t,"min")},ss=(e,t)=>{nt(e,"ReduceProdShared",t,"prod")},us=(e,t)=>{nt(e,"ReduceSumShared",t,"sum")},ds=(e,t)=>{nt(e,"ReduceSumSquareShared",t,"sumSquare")},ls=(e,t)=>{nt(e,"ReduceLogSumShared",t,"logSum")}});var ot,am,jr,so,it,sm,um,dm,lm,cm,pm,mm,fm,hm,gm,at,ps,ms,fs,hs,gs,ys,bs,ws,_s,vs,qr=V(()=>{"use strict";Q();ie();Ie();ae();cs();ot=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.")},am=e=>["","",`var value = ${e.getByIndices("input_indices")};`,""],jr=(e,t,r,n,o,i,a=!1,d=!1)=>{let l=[],c=r[0].dims,m=c.length,u=k.normalizeAxes(o,m),h=!d&&u.length===0;c.forEach((S,$)=>{h||u.indexOf($)>=0?a&&l.push(1):l.push(S)});let w=l.length,g=k.size(l);return{name:e,shaderCache:t,getShaderSource:S=>{let $=[],v=E("_A",r[0].dataType,m),x=M("output",i,w),T=n(v,x,u),C=T[2];for(let A=0,P=0;A<m;A++)h||u.indexOf(A)>=0?(a&&P++,C=`for(var j${A}: u32 = 0; j${A} < ${c[A]}; j${A}++) {
                   ${T[2].includes("last_index")?`let last_index = j${A};`:""}
-                  ${_.indicesSet("input_indices",A,`j${A}`)}
+                  ${v.indicesSet("input_indices",A,`j${A}`)}
                   ${C}
-                }`):($.push(`${_.indicesSet("input_indices",A,x.indicesGet("output_indices",P))};`),P++);return`
+                }`):($.push(`${v.indicesSet("input_indices",A,x.indicesGet("output_indices",P))};`),P++);return`
 
-        ${S.registerUniform("output_size","u32").declareVariables(_,x)}
+        ${S.registerUniform("output_size","u32").declareVariables(v,x)}
 
         ${S.mainStart()}
           ${S.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")}
-          var input_indices: ${_.type.indices};
+          var input_indices: ${v.type.indices};
           let output_indices = ${x.offsetToIndices("global_idx")};
 
           ${$.join(`
@@ -3947,9 +3947,9 @@ var Un=Object.defineProperty;var _p=Object.getOwnPropertyDescriptor;var $p=Objec
           ${T[3]}
           ${T.length===4?x.setByOffset("global_idx","value"):T.slice(4).join(`
 `)}
-        }`},getRunData:()=>({outputs:[{dims:l,dataType:i}],dispatchGroup:{x:Math.ceil(g/64)},programUniforms:[{type:12,data:g},...V(c,l)]})}},so=(e,t)=>{let r=[];return e[1].dims[0]>0&&e[1].getBigInt64Array().forEach(n=>r.push(Number(n))),J({axes:r,keepDims:t.keepDims,noopWithEmptyAxes:t.noopWithEmptyAxes})},it=(e,t,r,n)=>{let o=e.inputs,i=o.length===1?r:so(o,r);e.compute(jr(t,{hint:i.cacheKey,inputDependencies:["rank"]},[o[0]],i.noopWithEmptyAxes&&i.axes.length===0?am:n,i.axes,o[0].dataType,i.keepDims,i.noopWithEmptyAxes),{inputs:[0]})},sm=(e,t)=>{ot(e.inputs),it(e,"ReduceLogSum",t,(n,o)=>[`var value = ${o.type.storage}(0);`,"",`value += ${n.getByIndices("input_indices")};`,"value = log(value);"])},um=(e,t)=>{ot(e.inputs),it(e,"ReduceL1",t,(n,o)=>[`var value = ${o.type.storage}(0);`,"",`value += abs(${n.getByIndices("input_indices")});`,""])},dm=(e,t)=>{ot(e.inputs),it(e,"ReduceL2",t,(n,o)=>[`var t = ${o.type.value}(0); var value = ${o.type.value}(0);`,"",`t = ${n.getByIndices("input_indices")}; value += (t * t);`,"value = sqrt(value);"])},lm=(e,t)=>{ot(e.inputs),it(e,"ReduceLogSumExp",t,(n,o)=>[`var value = ${o.type.storage}(0);`,"",`value += exp(${n.getByIndices("input_indices")});`,"value = log(value);"])},cm=(e,t)=>{ot(e.inputs),it(e,"ReduceMax",t,(n,o,i)=>{let a=[];for(let d=0;d<n.rank;d++)(i.indexOf(d)>=0||i.length===0)&&a.push(n.indicesSet("input_indices",d,0));return[`${a.join(`
+        }`},getRunData:()=>({outputs:[{dims:l,dataType:i}],dispatchGroup:{x:Math.ceil(g/64)},programUniforms:[{type:12,data:g},...R(c,l)]})}},so=(e,t)=>{let r=[];return e[1].dims[0]>0&&e[1].getBigInt64Array().forEach(n=>r.push(Number(n))),J({axes:r,keepDims:t.keepDims,noopWithEmptyAxes:t.noopWithEmptyAxes})},it=(e,t,r,n)=>{let o=e.inputs,i=o.length===1?r:so(o,r);e.compute(jr(t,{hint:i.cacheKey,inputDependencies:["rank"]},[o[0]],i.noopWithEmptyAxes&&i.axes.length===0?am:n,i.axes,o[0].dataType,i.keepDims,i.noopWithEmptyAxes),{inputs:[0]})},sm=(e,t)=>{ot(e.inputs),it(e,"ReduceLogSum",t,(n,o)=>[`var value = ${o.type.storage}(0);`,"",`value += ${n.getByIndices("input_indices")};`,"value = log(value);"])},um=(e,t)=>{ot(e.inputs),it(e,"ReduceL1",t,(n,o)=>[`var value = ${o.type.storage}(0);`,"",`value += abs(${n.getByIndices("input_indices")});`,""])},dm=(e,t)=>{ot(e.inputs),it(e,"ReduceL2",t,(n,o)=>[`var t = ${o.type.value}(0); var value = ${o.type.value}(0);`,"",`t = ${n.getByIndices("input_indices")}; value += (t * t);`,"value = sqrt(value);"])},lm=(e,t)=>{ot(e.inputs),it(e,"ReduceLogSumExp",t,(n,o)=>[`var value = ${o.type.storage}(0);`,"",`value += exp(${n.getByIndices("input_indices")});`,"value = log(value);"])},cm=(e,t)=>{ot(e.inputs),it(e,"ReduceMax",t,(n,o,i)=>{let a=[];for(let d=0;d<n.rank;d++)(i.indexOf(d)>=0||i.length===0)&&a.push(n.indicesSet("input_indices",d,0));return[`${a.join(`
 `)}`,`var value = ${n.getByIndices("input_indices")};`,`value = max(value, ${n.getByIndices("input_indices")});`,""]})},pm=(e,t)=>{ot(e.inputs),it(e,"ReduceMean",t,(n,o,i)=>{let a=1;for(let d=0;d<n.rank;d++)(i.indexOf(d)>=0||i.length===0)&&(a*=e.inputs[0].dims[d]);return["var sum = f32(0);","",`sum += f32(${n.getByIndices("input_indices")});`,`let value = ${o.type.value}(sum / ${a});`]})},mm=(e,t)=>{ot(e.inputs),it(e,"ReduceMin",t,(n,o,i)=>{let a=[];for(let d=0;d<n.rank;d++)(i.indexOf(d)>=0||i.length===0)&&a.push(`input_indices[${d}] = 0;`);return[`${a.join(`
-`)}`,`var value = ${n.getByIndices("input_indices")};`,`value = min(value, ${n.getByIndices("input_indices")});`,""]})},fm=(e,t)=>{ot(e.inputs),it(e,"ReduceProd",t,(n,o)=>[`var value = ${o.type.storage}(1);`,"",`value *= ${n.getByIndices("input_indices")};`,""])},hm=(e,t)=>{ot(e.inputs),it(e,"ReduceSum",t,(n,o)=>[`var value = ${o.type.storage}(0);`,"",`value += ${n.getByIndices("input_indices")};`,""])},gm=(e,t)=>{ot(e.inputs),it(e,"ReduceSumSquare",t,(n,o)=>[`var t = ${o.type.value}(0); var value = ${o.type.value}(0);`,"",`t = ${n.getByIndices("input_indices")}; value += t * t;`,""])},at=(e,t,r)=>{if(t.length===0)return r;let n=1,o=1;for(let i=0;i<t.length;i++)t.indexOf(i)===-1?n*=e[i]:o*=e[i];return o<32&&n>1024},ps=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?pm(e,t):ts(e,t)},ms=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?um(e,t):rs(e,t)},fs=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?dm(e,t):ns(e,t)},hs=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?lm(e,t):os(e,t)},gs=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?cm(e,t):is(e,t)},ys=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?mm(e,t):as(e,t)},bs=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?fm(e,t):ss(e,t)},ws=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?hm(e,t):us(e,t)},vs=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?gm(e,t):ds(e,t)},_s=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?sm(e,t):ls(e,t)}});var $s,xs,Ss,uo,Ts=U(()=>{"use strict";Z();Ie();qr();$s=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.")},xs=(e,t)=>{$s(e.inputs);let r=(n,o,i)=>{let a=[];for(let d=0;d<n.rank;d++)(i.indexOf(d)>=0||i.length===0)&&a.push(`input_indices[${d}] = 0;`);return[`${a.join(`
+`)}`,`var value = ${n.getByIndices("input_indices")};`,`value = min(value, ${n.getByIndices("input_indices")});`,""]})},fm=(e,t)=>{ot(e.inputs),it(e,"ReduceProd",t,(n,o)=>[`var value = ${o.type.storage}(1);`,"",`value *= ${n.getByIndices("input_indices")};`,""])},hm=(e,t)=>{ot(e.inputs),it(e,"ReduceSum",t,(n,o)=>[`var value = ${o.type.storage}(0);`,"",`value += ${n.getByIndices("input_indices")};`,""])},gm=(e,t)=>{ot(e.inputs),it(e,"ReduceSumSquare",t,(n,o)=>[`var t = ${o.type.value}(0); var value = ${o.type.value}(0);`,"",`t = ${n.getByIndices("input_indices")}; value += t * t;`,""])},at=(e,t,r)=>{if(t.length===0)return r;let n=1,o=1;for(let i=0;i<t.length;i++)t.indexOf(i)===-1?n*=e[i]:o*=e[i];return o<32&&n>1024},ps=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?pm(e,t):ts(e,t)},ms=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?um(e,t):rs(e,t)},fs=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?dm(e,t):ns(e,t)},hs=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?lm(e,t):os(e,t)},gs=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?cm(e,t):is(e,t)},ys=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?mm(e,t):as(e,t)},bs=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?fm(e,t):ss(e,t)},ws=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?hm(e,t):us(e,t)},_s=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?gm(e,t):ds(e,t)},vs=(e,t)=>{at(e.inputs[0].dims,t.axes,t.noopWithEmptyAxes)?sm(e,t):ls(e,t)}});var $s,xs,Ss,uo,Ts=V(()=>{"use strict";Q();Ie();qr();$s=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.")},xs=(e,t)=>{$s(e.inputs);let r=(n,o,i)=>{let a=[];for(let d=0;d<n.rank;d++)(i.indexOf(d)>=0||i.length===0)&&a.push(`input_indices[${d}] = 0;`);return[`${a.join(`
 `)}`,`var value = ${n.getByIndices("input_indices")};
 var best_index : i32 = 0;`,`if (${n.getByIndices("input_indices")} ${t.selectLastIndex>0?"<=":"<"} value) {
          value = ${n.getByIndices("input_indices")};
@@ -3959,7 +3959,7 @@ var best_index : i32 = 0;`,`if (${n.getByIndices("input_indices")} ${t.selectLas
 var best_index : i32 = 0;`,`if (${n.getByIndices("input_indices")} ${t.selectLastIndex>0?">=":">"} value) {
          value = ${n.getByIndices("input_indices")};
          best_index = i32(last_index);
-       }`,"",o.setByOffset("global_idx","best_index")]};e.compute(jr("argMax",{hint:t.cacheKey,inputDependencies:["rank"]},[e.inputs[0]],r,[t.axis],7,t.keepDims),{inputs:[0]})},uo=e=>J(e)});var ym,bm,wm,vm,Wt,_m,Is,Kr=U(()=>{"use strict";Z();te();Hr();re();ym=(e,t)=>{let r=e[0],n=e[1],o=e[2],i=e[3],a=e[4],d=e[5];if(a&&d)throw new Error("Attention cannot have both past and attention_bias");if(r.dims.length!==3)throw new Error('Input "input" must have 3 dimensions');let l=r.dims[0],c=r.dims[1],m=r.dims[2];if(o.dims.length!==1)throw new Error('Input "bias" is expected to have 1 dimensions');if(n.dims.length!==2)throw new Error('Input "weights" is expected to have 2 dimensions');if(n.dims[0]!==m)throw new Error("Input 1 dimension 0 should have same length as dimension 2 of input 0");if(o.dims[0]!==n.dims[1])throw new Error('Input "bias" dimension 0 should have same length as dimension 1 of input "weights"');let u=o.dims[0]/3,h=u,w=h;if(t.qkvHiddenSizes.length>0){if(t.qkvHiddenSizes.length!==3)throw new Error("qkv_hidden_sizes attribute should have 3 elements");for(let x of t.qkvHiddenSizes)if(x%t.numHeads!==0)throw new Error("qkv_hidden_sizes should be divisible by num_heads");u=t.qkvHiddenSizes[0],h=t.qkvHiddenSizes[1],w=t.qkvHiddenSizes[2]}let g=c;if(u!==h)throw new Error("qkv_hidden_sizes first element should be same as the second");if(o.dims[0]!==u+h+w)throw new Error('Input "bias" dimension 0 should have same length as sum of Q/K/V hidden sizes');let y=0;if(a){if(h!==w)throw new Error('Input "past" expect k_hidden_size == v_hidden_size');if(a.dims.length!==5)throw new Error('Input "past" must have 5 dimensions');if(a.dims[0]!==2)throw new Error('Input "past" first dimension must be 2');if(a.dims[1]!==l)throw new Error('Input "past" second dimension must be batch_size');if(a.dims[2]!==t.numHeads)throw new Error('Input "past" third dimension must be num_heads');if(a.dims[4]!==h/t.numHeads)throw new Error('Input "past" fifth dimension must be k_hidden_size / num_heads');t.pastPresentShareBuffer||(y=a.dims[3])}let S=g+y,$=-1,_=0;if(i)throw new Error("Mask not supported");if(a)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]!==l||d.dims[1]!==t.numHeads||d.dims[2]!==c||d.dims[3]!==S)throw new Error('Expect "attention_bias" shape (batch_size, num_heads, sequence_length, total_sequence_length)')}return{batchSize:l,sequenceLength:c,pastSequenceLength:y,kvSequenceLength:g,totalSequenceLength:S,maxSequenceLength:$,inputHiddenSize:m,hiddenSize:u,vHiddenSize:w,headSize:Math.floor(u/t.numHeads),vHeadSize:Math.floor(w/t.numHeads),numHeads:t.numHeads,isUnidirectional:!1,pastPresentShareBuffer:!1,maskFilterValue:t.maskFilterValue,maskType:_,scale:t.scale,broadcastResPosBias:!1,passPastInKv:!1,qkvFormat:1}},bm=(e,t,r)=>{let n=ve(r),o=64,i=r/n;i<o&&(o=32);let a=Math.ceil(r/n/o),d=[{type:1,data:1/r},{type:12,data:i},{type:12,data:a}],l=ye(e.dataType,n),c=ke(1,n),m=["type"],u=h=>{let w=M("x",e.dataType,e.dims,n),g=ke(e.dataType),y=[{name:"d_inv",type:"f32"},{name:"d_comp",type:"u32"},{name:"elements_per_thread",type:"u32"}];return`
+       }`,"",o.setByOffset("global_idx","best_index")]};e.compute(jr("argMax",{hint:t.cacheKey,inputDependencies:["rank"]},[e.inputs[0]],r,[t.axis],7,t.keepDims),{inputs:[0]})},uo=e=>J(e)});var ym,bm,wm,_m,Wt,vm,Is,Kr=V(()=>{"use strict";Q();ie();Hr();ae();ym=(e,t)=>{let r=e[0],n=e[1],o=e[2],i=e[3],a=e[4],d=e[5];if(a&&d)throw new Error("Attention cannot have both past and attention_bias");if(r.dims.length!==3)throw new Error('Input "input" must have 3 dimensions');let l=r.dims[0],c=r.dims[1],m=r.dims[2];if(o.dims.length!==1)throw new Error('Input "bias" is expected to have 1 dimensions');if(n.dims.length!==2)throw new Error('Input "weights" is expected to have 2 dimensions');if(n.dims[0]!==m)throw new Error("Input 1 dimension 0 should have same length as dimension 2 of input 0");if(o.dims[0]!==n.dims[1])throw new Error('Input "bias" dimension 0 should have same length as dimension 1 of input "weights"');let u=o.dims[0]/3,h=u,w=h;if(t.qkvHiddenSizes.length>0){if(t.qkvHiddenSizes.length!==3)throw new Error("qkv_hidden_sizes attribute should have 3 elements");for(let x of t.qkvHiddenSizes)if(x%t.numHeads!==0)throw new Error("qkv_hidden_sizes should be divisible by num_heads");u=t.qkvHiddenSizes[0],h=t.qkvHiddenSizes[1],w=t.qkvHiddenSizes[2]}let g=c;if(u!==h)throw new Error("qkv_hidden_sizes first element should be same as the second");if(o.dims[0]!==u+h+w)throw new Error('Input "bias" dimension 0 should have same length as sum of Q/K/V hidden sizes');let y=0;if(a){if(h!==w)throw new Error('Input "past" expect k_hidden_size == v_hidden_size');if(a.dims.length!==5)throw new Error('Input "past" must have 5 dimensions');if(a.dims[0]!==2)throw new Error('Input "past" first dimension must be 2');if(a.dims[1]!==l)throw new Error('Input "past" second dimension must be batch_size');if(a.dims[2]!==t.numHeads)throw new Error('Input "past" third dimension must be num_heads');if(a.dims[4]!==h/t.numHeads)throw new Error('Input "past" fifth dimension must be k_hidden_size / num_heads');t.pastPresentShareBuffer||(y=a.dims[3])}let S=g+y,$=-1,v=0;if(i)throw new Error("Mask not supported");if(a)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]!==l||d.dims[1]!==t.numHeads||d.dims[2]!==c||d.dims[3]!==S)throw new Error('Expect "attention_bias" shape (batch_size, num_heads, sequence_length, total_sequence_length)')}return{batchSize:l,sequenceLength:c,pastSequenceLength:y,kvSequenceLength:g,totalSequenceLength:S,maxSequenceLength:$,inputHiddenSize:m,hiddenSize:u,vHiddenSize:w,headSize:Math.floor(u/t.numHeads),vHeadSize:Math.floor(w/t.numHeads),numHeads:t.numHeads,isUnidirectional:!1,pastPresentShareBuffer:!1,maskFilterValue:t.maskFilterValue,maskType:v,scale:t.scale,broadcastResPosBias:!1,passPastInKv:!1,qkvFormat:1}},bm=(e,t,r)=>{let n=we(r),o=64,i=r/n;i<o&&(o=32);let a=Math.ceil(r/n/o),d=[{type:1,data:1/r},{type:12,data:i},{type:12,data:a}],l=he(e.dataType,n),c=Ee(1,n),m=["type"],u=h=>{let w=M("x",e.dataType,e.dims,n),g=Ee(e.dataType),y=[{name:"d_inv",type:"f32"},{name:"d_comp",type:"u32"},{name:"elements_per_thread",type:"u32"}];return`
   var<workgroup> thread_max: array<f32, ${o}>;
   var<workgroup> thread_sum: array<f32, ${o}>;
   ${h.registerUniforms(y).declareVariables(w)}
@@ -4001,31 +4001,31 @@ var best_index : i32 = 0;`,`if (${n.getByIndices("input_indices")} ${t.selectLas
         x[offset + i] = ${w.type.value}(exp(f32input - max_value) / sum);
       }
     }
-  }`};return{name:"AttentionProbsSoftmax",shaderCache:{hint:`${o};${l};${n}`,inputDependencies:m},getShaderSource:u,getRunData:()=>({outputs:[],dispatchGroup:{x:t},programUniforms:d})}},wm=(e,t,r,n,o,i,a,d)=>{let l=d+i.kvSequenceLength,c=[i.batchSize,i.numHeads,i.sequenceLength,l],m=i.kvNumHeads===void 0&&e>1&&n,u=m?[i.batchSize,i.numHeads,l,i.headSize]:void 0,h=a.scale===0?1/Math.sqrt(i.headSize):a.scale,w=ve(i.headSize),g=i.headSize/w,y=12,S={x:Math.ceil(l/y),y:Math.ceil(i.sequenceLength/y),z:i.batchSize*i.numHeads},$=[{type:12,data:i.sequenceLength},{type:12,data:g},{type:12,data:l},{type:12,data:i.numHeads},{type:1,data:h},{type:12,data:d},{type:12,data:i.kvSequenceLength}],_=m&&n&&E.size(n.dims)>0,x=["type","type"];_&&x.push("type"),o&&x.push("type");let T=[{dims:c,dataType:t.dataType,gpuDataType:0}];m&&T.push({dims:u,dataType:t.dataType,gpuDataType:0});let C=A=>{let P=k("q",t.dataType,t.dims,w),D=k("key",r.dataType,r.dims,w),N=[P,D];if(_){let ae=k("past_key",n.dataType,n.dims,w);N.push(ae)}o&&N.push(k("attention_bias",o.dataType,o.dims));let L=M("output",t.dataType,c),K=[L];m&&K.push(M("present_key",t.dataType,u,w));let X=ke(1,w),ie=[{name:"M",type:"u32"},{name:"K",type:"u32"},{name:"N",type:"u32"},{name:"num_heads",type:"u32"},{name:"alpha",type:"f32"},{name:"past_sequence_length",type:"u32"},{name:"kv_sequence_length",type:"u32"}];return`
+  }`};return{name:"AttentionProbsSoftmax",shaderCache:{hint:`${o};${l};${n}`,inputDependencies:m},getShaderSource:u,getRunData:()=>({outputs:[],dispatchGroup:{x:t},programUniforms:d})}},wm=(e,t,r,n,o,i,a,d)=>{let l=d+i.kvSequenceLength,c=[i.batchSize,i.numHeads,i.sequenceLength,l],m=i.kvNumHeads===void 0&&e>1&&n,u=m?[i.batchSize,i.numHeads,l,i.headSize]:void 0,h=a.scale===0?1/Math.sqrt(i.headSize):a.scale,w=we(i.headSize),g=i.headSize/w,y=12,S={x:Math.ceil(l/y),y:Math.ceil(i.sequenceLength/y),z:i.batchSize*i.numHeads},$=[{type:12,data:i.sequenceLength},{type:12,data:g},{type:12,data:l},{type:12,data:i.numHeads},{type:1,data:h},{type:12,data:d},{type:12,data:i.kvSequenceLength}],v=m&&n&&k.size(n.dims)>0,x=["type","type"];v&&x.push("type"),o&&x.push("type");let T=[{dims:c,dataType:t.dataType,gpuDataType:0}];m&&T.push({dims:u,dataType:t.dataType,gpuDataType:0});let C=A=>{let P=E("q",t.dataType,t.dims,w),B=E("key",r.dataType,r.dims,w),N=[P,B];if(v){let se=E("past_key",n.dataType,n.dims,w);N.push(se)}o&&N.push(E("attention_bias",o.dataType,o.dims));let W=M("output",t.dataType,c),K=[W];m&&K.push(M("present_key",t.dataType,u,w));let Z=Ee(1,w),ee=[{name:"M",type:"u32"},{name:"K",type:"u32"},{name:"N",type:"u32"},{name:"num_heads",type:"u32"},{name:"alpha",type:"f32"},{name:"past_sequence_length",type:"u32"},{name:"kv_sequence_length",type:"u32"}];return`
   const TILE_SIZE = ${y}u;
 
   var<workgroup> tileQ: array<${P.type.storage}, ${y*y}>;
   var<workgroup> tileK: array<${P.type.storage}, ${y*y}>;
-  ${A.registerUniforms(ie).declareVariables(...N,...K)}
+  ${A.registerUniforms(ee).declareVariables(...N,...K)}
   ${A.mainStart([y,y,1])}
     // x holds the N and y holds the M
     let headIdx = workgroup_id.z;
     let m = workgroup_id.y * TILE_SIZE;
     let n = workgroup_id.x * TILE_SIZE;
     let qOffset = uniforms.M * uniforms.K * headIdx + m * uniforms.K;
-    ${(()=>_&&m?`
+    ${(()=>v&&m?`
     let kOffset = uniforms.kv_sequence_length * uniforms.K * headIdx;
     let pastKeyOffset = uniforms.past_sequence_length * uniforms.K * headIdx;`:`
     let kOffset = uniforms.N * uniforms.K * headIdx + n * uniforms.K;`)()}
     ${m?"let presentKeyOffset = headIdx * uniforms.N * uniforms.K;":""}
-    var value = ${X}(0);
+    var value = ${Z}(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?`
+      ${(()=>v&&m?`
               if (n + local_id.y < uniforms.past_sequence_length) {
                 tileK[idx] = past_key[pastKeyOffset + (n + local_id.y) * uniforms.K + w + local_id.x];
               } else {
@@ -4037,7 +4037,7 @@ var best_index : i32 = 0;`,`if (${n.getByIndices("input_indices")} ${t.selectLas
       workgroupBarrier();
 
       for (var k: u32 = 0u; k < TILE_SIZE && w+k < uniforms.K; k++) {
-        value += ${X}(tileQ[TILE_SIZE * local_id.y + k] * tileK[TILE_SIZE * local_id.x + k]);
+        value += ${Z}(tileQ[TILE_SIZE * local_id.y + k] * tileK[TILE_SIZE * local_id.x + k]);
       }
 
       workgroupBarrier();
@@ -4047,13 +4047,13 @@ var best_index : i32 = 0;`,`if (${n.getByIndices("input_indices")} ${t.selectLas
     if (global_id.y < uniforms.M && global_id.x < uniforms.N) {
       let outputIdx = headOffset + global_id.y * uniforms.N + global_id.x;
       var sum: f32 = ${(()=>{switch(w){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: ${w}`)}})()};
-        output[outputIdx] = ${L.type.value} (sum * uniforms.alpha) + ${o?"attention_bias[outputIdx]":"0.0"};
+        output[outputIdx] = ${W.type.value} (sum * uniforms.alpha) + ${o?"attention_bias[outputIdx]":"0.0"};
     }
-  }`};return{name:"AttentionProbs",shaderCache:{hint:`${w};${o!==void 0};${n!==void 0};${e}`,inputDependencies:x},getRunData:()=>({outputs:T,dispatchGroup:S,programUniforms:$}),getShaderSource:C}},vm=(e,t,r,n,o,i)=>{let a=i+o.kvSequenceLength,d=o.nReps?o.nReps:1,l=o.vHiddenSize*d,c=o.kvNumHeads==null&&e>1&&n,m=c?[o.batchSize,o.numHeads,a,o.headSize]:void 0,u=[o.batchSize,o.sequenceLength,l],h=12,w={x:Math.ceil(o.vHeadSize/h),y:Math.ceil(o.sequenceLength/h),z:o.batchSize*o.numHeads},g=[{type:12,data:o.sequenceLength},{type:12,data:a},{type:12,data:o.vHeadSize},{type:12,data:o.numHeads},{type:12,data:l},{type:12,data:i},{type:12,data:o.kvSequenceLength}],y=c&&n&&E.size(n.dims)>0,S=["type","type"];y&&S.push("type");let $=[{dims:u,dataType:t.dataType,gpuDataType:0}];c&&$.push({dims:m,dataType:t.dataType,gpuDataType:0});let _=x=>{let T=k("probs",t.dataType,t.dims),C=k("v",r.dataType,r.dims),A=[T,C];y&&A.push(k("past_value",n.dataType,n.dims));let D=[M("output",t.dataType,u)];c&&D.push(M("present_value",t.dataType,m));let N=[{name:"M",type:"u32"},{name:"K",type:"u32"},{name:"N",type:"u32"},{name:"num_heads",type:"u32"},{name:"v_hidden_size",type:"u32"},{name:"past_sequence_length",type:"u32"},{name:"kv_sequence_length",type:"u32"}];return`
+  }`};return{name:"AttentionProbs",shaderCache:{hint:`${w};${o!==void 0};${n!==void 0};${e}`,inputDependencies:x},getRunData:()=>({outputs:T,dispatchGroup:S,programUniforms:$}),getShaderSource:C}},_m=(e,t,r,n,o,i)=>{let a=i+o.kvSequenceLength,d=o.nReps?o.nReps:1,l=o.vHiddenSize*d,c=o.kvNumHeads==null&&e>1&&n,m=c?[o.batchSize,o.numHeads,a,o.headSize]:void 0,u=[o.batchSize,o.sequenceLength,l],h=12,w={x:Math.ceil(o.vHeadSize/h),y:Math.ceil(o.sequenceLength/h),z:o.batchSize*o.numHeads},g=[{type:12,data:o.sequenceLength},{type:12,data:a},{type:12,data:o.vHeadSize},{type:12,data:o.numHeads},{type:12,data:l},{type:12,data:i},{type:12,data:o.kvSequenceLength}],y=c&&n&&k.size(n.dims)>0,S=["type","type"];y&&S.push("type");let $=[{dims:u,dataType:t.dataType,gpuDataType:0}];c&&$.push({dims:m,dataType:t.dataType,gpuDataType:0});let v=x=>{let T=E("probs",t.dataType,t.dims),C=E("v",r.dataType,r.dims),A=[T,C];y&&A.push(E("past_value",n.dataType,n.dims));let B=[M("output",t.dataType,u)];c&&B.push(M("present_value",t.dataType,m));let N=[{name:"M",type:"u32"},{name:"K",type:"u32"},{name:"N",type:"u32"},{name:"num_heads",type:"u32"},{name:"v_hidden_size",type:"u32"},{name:"past_sequence_length",type:"u32"},{name:"kv_sequence_length",type:"u32"}];return`
   const TILE_SIZE = ${h}u;
   var<workgroup> tileQ: array<${T.type.value}, ${h*h}>;
   var<workgroup> tileK: array<${T.type.value}, ${h*h}>;
-  ${x.registerUniforms(N).declareVariables(...A,...D)}
+  ${x.registerUniforms(N).declareVariables(...A,...B)}
   ${x.mainStart([h,h,1])}
    let headIdx = workgroup_id.z;
    let m = global_id.y;
@@ -4100,12 +4100,12 @@ var best_index : i32 = 0;`,`if (${n.getByIndices("input_indices")} ${t.selectLas
        + currentBatchHeadNumber * uniforms.N + n;
      output[outputIdx] = value;
    }
-  }`};return{name:"AttentionScore",shaderCache:{hint:`${n!==void 0};${e}`,inputDependencies:S},getRunData:()=>({outputs:$,dispatchGroup:w,programUniforms:g}),getShaderSource:_}},Wt=(e,t,r,n,o,i,a,d,l,c,m)=>{let u=Math.min(e.outputCount,1+(a?1:0)+(d?1:0)),h=c.kvNumHeads!==void 0||u>1?c.pastSequenceLength:0,w=h+c.kvSequenceLength,g=l&&E.size(l.dims)>0?l:void 0,y=[t,r];c.kvNumHeads===void 0&&u>1&&a&&E.size(a.dims)>0&&y.push(a),g&&y.push(g);let S=e.compute(wm(u,t,r,a,g,c,m,h),{inputs:y,outputs:c.kvNumHeads===void 0&&u>1?[-1,1]:[-1]})[0];e.compute(bm(S,c.batchSize*c.numHeads*c.sequenceLength,w),{inputs:[S],outputs:[]});let $=[S,n];c.kvNumHeads===void 0&&u>1&&d&&E.size(d.dims)>0&&$.push(d),e.compute(vm(u,S,n,d,c,h),{inputs:$,outputs:c.kvNumHeads===void 0&&u>1?[0,2]:[0]})},_m=(e,t)=>{let r=[t.batchSize,t.numHeads,t.sequenceLength,t.headSize],n=t.sequenceLength,o=t.inputHiddenSize,i=t.headSize,a=12,d={x:Math.ceil(t.headSize/a),y:Math.ceil(t.sequenceLength/a),z:t.batchSize*t.numHeads},l=[e.inputs[0],e.inputs[1],e.inputs[2]],c=[{type:12,data:n},{type:12,data:o},{type:12,data:i},{type:12,data:t.numHeads},{type:12,data:t.headSize},{type:12,data:t.hiddenSize},{type:12,data:t.hiddenSize+t.hiddenSize+t.vHiddenSize}],m=u=>{let h=M("output_q",l[0].dataType,r),w=M("output_k",l[0].dataType,r),g=M("output_v",l[0].dataType,r),y=k("input",l[0].dataType,l[0].dims),S=k("weight",l[1].dataType,l[1].dims),$=k("bias",l[2].dataType,l[2].dims),_=y.type.storage,x=[{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`
+  }`};return{name:"AttentionScore",shaderCache:{hint:`${n!==void 0};${e}`,inputDependencies:S},getRunData:()=>({outputs:$,dispatchGroup:w,programUniforms:g}),getShaderSource:v}},Wt=(e,t,r,n,o,i,a,d,l,c,m)=>{let u=Math.min(e.outputCount,1+(a?1:0)+(d?1:0)),h=c.kvNumHeads!==void 0||u>1?c.pastSequenceLength:0,w=h+c.kvSequenceLength,g=l&&k.size(l.dims)>0?l:void 0,y=[t,r];c.kvNumHeads===void 0&&u>1&&a&&k.size(a.dims)>0&&y.push(a),g&&y.push(g);let S=e.compute(wm(u,t,r,a,g,c,m,h),{inputs:y,outputs:c.kvNumHeads===void 0&&u>1?[-1,1]:[-1]})[0];e.compute(bm(S,c.batchSize*c.numHeads*c.sequenceLength,w),{inputs:[S],outputs:[]});let $=[S,n];c.kvNumHeads===void 0&&u>1&&d&&k.size(d.dims)>0&&$.push(d),e.compute(_m(u,S,n,d,c,h),{inputs:$,outputs:c.kvNumHeads===void 0&&u>1?[0,2]:[0]})},vm=(e,t)=>{let r=[t.batchSize,t.numHeads,t.sequenceLength,t.headSize],n=t.sequenceLength,o=t.inputHiddenSize,i=t.headSize,a=12,d={x:Math.ceil(t.headSize/a),y:Math.ceil(t.sequenceLength/a),z:t.batchSize*t.numHeads},l=[e.inputs[0],e.inputs[1],e.inputs[2]],c=[{type:12,data:n},{type:12,data:o},{type:12,data:i},{type:12,data:t.numHeads},{type:12,data:t.headSize},{type:12,data:t.hiddenSize},{type:12,data:t.hiddenSize+t.hiddenSize+t.vHiddenSize}],m=u=>{let h=M("output_q",l[0].dataType,r),w=M("output_k",l[0].dataType,r),g=M("output_v",l[0].dataType,r),y=E("input",l[0].dataType,l[0].dims),S=E("weight",l[1].dataType,l[1].dims),$=E("bias",l[2].dataType,l[2].dims),v=y.type.storage,x=[{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 = ${a}u;
-  var<workgroup> tileInput: array<${_}, ${a*a}>;
-  var<workgroup> tileWeightQ: array<${_}, ${a*a}>;
-  var<workgroup> tileWeightK: array<${_}, ${a*a}>;
-  var<workgroup> tileWeightV: array<${_}, ${a*a}>;
+  var<workgroup> tileInput: array<${v}, ${a*a}>;
+  var<workgroup> tileWeightQ: array<${v}, ${a*a}>;
+  var<workgroup> tileWeightK: array<${v}, ${a*a}>;
+  var<workgroup> tileWeightV: array<${v}, ${a*a}>;
   ${u.registerUniforms(x).declareVariables(y,S,$,h,w,g)}
   ${u.mainStart([a,a,1])}
     let batchIndex = workgroup_id.z / uniforms.num_heads;
@@ -4118,9 +4118,9 @@ var best_index : i32 = 0;`,`if (${n.getByIndices("input_indices")} ${t.selectLas
     let biasOffsetK = uniforms.hidden_size + biasOffsetQ;
     let biasOffsetV = uniforms.hidden_size + biasOffsetK;
 
-    var valueQ = ${_}(0);
-    var valueK = ${_}(0);
-    var valueV = ${_}(0);
+    var valueQ = ${v}(0);
+    var valueK = ${v}(0);
+    var valueV = ${v}(0);
     for (var w: u32 = 0u; w < uniforms.K; w += TILE_SIZE) {
       if (m < uniforms.M && w + local_id.x < uniforms.K) {
         tileInput[TILE_SIZE * local_id.y + local_id.x] = input[inputOffset + w + local_id.x];
@@ -4155,10 +4155,10 @@ var best_index : i32 = 0;`,`if (${n.getByIndices("input_indices")} ${t.selectLas
       output_k[outputIdx] = valueK;
       output_v[outputIdx] = valueV;
     }
-  }`};return e.compute({name:"AttentionPrepare",shaderCache:{inputDependencies:["type","type","type"]},getRunData:()=>({outputs:[{dims:r,dataType:e.inputs[0].dataType,gpuDataType:0},{dims:r,dataType:e.inputs[0].dataType,gpuDataType:0},{dims:r,dataType:e.inputs[0].dataType,gpuDataType:0}],dispatchGroup:d,programUniforms:c}),getShaderSource:m},{inputs:l,outputs:[-1,-1,-1]})},Is=(e,t)=>{let r=ym(e.inputs,t),[n,o,i]=_m(e,r);return Wt(e,n,o,i,e.inputs[4],void 0,void 0,void 0,e.inputs[5],r,t)}});var $m,xm,Sm,Cs,As=U(()=>{"use strict";Ke();Z();te();Ie();re();$m=(e,t)=>{if(!e||e.length!==5)throw new Error("BatchNormalization requires 5 inputs");let r=(n,o,i)=>{let a=o.length;if(a!==n.length)throw new Error(`${i}: num dimensions != ${a}`);o.forEach((d,l)=>{if(d!==n[l])throw new Error(`${i}: dim[${l}] do not match`)})};if(e[0].dims.length>1){let n=t.format==="NHWC"?t.spatial?e[0].dims.slice(-1):e[0].dims.slice(-1).concat(e[0].dims.slice(1,e[0].dims.length-1)):e[0].dims.slice(1,t.spatial?2:void 0);r(e[1].dims,n,"Invalid input scale"),r(e[2].dims,n,"Invalid input B"),r(e[3].dims,n,"Invalid input mean"),r(e[4].dims,n,"Invalid input var")}else r(e[1].dims,[1],"Invalid input scale"),r(e[2].dims,[1],"Invalid input B"),r(e[3].dims,[1],"Invalid input mean"),r(e[4].dims,[1],"Invalid input var")},xm=(e,t)=>{let{epsilon:r,spatial:n,format:o}=t,i=e[0].dims,a=n?ve(i[i.length-1]):1,d=o==="NHWC"&&i.length>1?a:1,l=E.size(i)/a,c=n,m=c?i.length:i,u=k("x",e[0].dataType,e[0].dims,a),h=k("scale",e[1].dataType,e[1].dims,d),w=k("bias",e[2].dataType,e[2].dims,d),g=k("inputMean",e[3].dataType,e[3].dims,d),y=k("inputVar",e[4].dataType,e[4].dims,d),S=M("y",e[0].dataType,m,a),$=()=>{let x="";if(n)x=`let cOffset = ${i.length===1?"0u":o==="NHWC"?`outputIndices[${i.length-1}] / ${a}`:"outputIndices[1]"};`;else if(o==="NCHW")x=`
+  }`};return e.compute({name:"AttentionPrepare",shaderCache:{inputDependencies:["type","type","type"]},getRunData:()=>({outputs:[{dims:r,dataType:e.inputs[0].dataType,gpuDataType:0},{dims:r,dataType:e.inputs[0].dataType,gpuDataType:0},{dims:r,dataType:e.inputs[0].dataType,gpuDataType:0}],dispatchGroup:d,programUniforms:c}),getShaderSource:m},{inputs:l,outputs:[-1,-1,-1]})},Is=(e,t)=>{let r=ym(e.inputs,t),[n,o,i]=vm(e,r);return Wt(e,n,o,i,e.inputs[4],void 0,void 0,void 0,e.inputs[5],r,t)}});var $m,xm,Sm,Cs,As=V(()=>{"use strict";Ke();Q();ie();Ie();ae();$m=(e,t)=>{if(!e||e.length!==5)throw new Error("BatchNormalization requires 5 inputs");let r=(n,o,i)=>{let a=o.length;if(a!==n.length)throw new Error(`${i}: num dimensions != ${a}`);o.forEach((d,l)=>{if(d!==n[l])throw new Error(`${i}: dim[${l}] do not match`)})};if(e[0].dims.length>1){let n=t.format==="NHWC"?t.spatial?e[0].dims.slice(-1):e[0].dims.slice(-1).concat(e[0].dims.slice(1,e[0].dims.length-1)):e[0].dims.slice(1,t.spatial?2:void 0);r(e[1].dims,n,"Invalid input scale"),r(e[2].dims,n,"Invalid input B"),r(e[3].dims,n,"Invalid input mean"),r(e[4].dims,n,"Invalid input var")}else r(e[1].dims,[1],"Invalid input scale"),r(e[2].dims,[1],"Invalid input B"),r(e[3].dims,[1],"Invalid input mean"),r(e[4].dims,[1],"Invalid input var")},xm=(e,t)=>{let{epsilon:r,spatial:n,format:o}=t,i=e[0].dims,a=n?we(i[i.length-1]):1,d=o==="NHWC"&&i.length>1?a:1,l=k.size(i)/a,c=n,m=c?i.length:i,u=E("x",e[0].dataType,e[0].dims,a),h=E("scale",e[1].dataType,e[1].dims,d),w=E("bias",e[2].dataType,e[2].dims,d),g=E("inputMean",e[3].dataType,e[3].dims,d),y=E("inputVar",e[4].dataType,e[4].dims,d),S=M("y",e[0].dataType,m,a),$=()=>{let x="";if(n)x=`let cOffset = ${i.length===1?"0u":o==="NHWC"?`outputIndices[${i.length-1}] / ${a}`:"outputIndices[1]"};`;else if(o==="NCHW")x=`
             ${S.indicesSet("outputIndices","0","0")}
             let cOffset = ${S.indicesToOffset("outputIndices")};`;else{x=`var cIndices = ${h.type.indices}(0);
-                       cIndices[0] = outputIndices[${i.length-1}];`;for(let T=1;T<h.rank;T++)x+=`cIndices[${T}] = outputIndices[${T}];`;x+=`let cOffset = ${h.indicesToOffset("cIndices")};`}return x},_=x=>`
+                       cIndices[0] = outputIndices[${i.length-1}];`;for(let T=1;T<h.rank;T++)x+=`cIndices[${T}] = outputIndices[${T}];`;x+=`let cOffset = ${h.indicesToOffset("cIndices")};`}return x},v=x=>`
   const epsilon = ${r};
   ${x.registerUniform("outputSize","u32").declareVariables(u,h,w,g,y,S)}
   ${x.mainStart()}
@@ -4172,7 +4172,7 @@ var best_index : i32 = 0;`,`if (${n.getByIndices("input_indices")} ${t.selectLas
     let x = ${u.getByOffset("global_idx")};
     let value = (x - inputMean) * inverseSqrt(inputVar + epsilon) * scale + bias;
     ${S.setByOffset("global_idx","value")}
-  }`;return{name:"BatchNormalization",shaderCache:{hint:`${t.epsilon}_${t.format}_${n}_${a}`,inputDependencies:c?["rank","type","type","type","type"]:void 0},getShaderSource:_,getRunData:()=>({outputs:[{dims:e[0].dims,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(l/64)},programUniforms:c?[{type:12,data:l},...V(i)]:[{type:12,data:l}]})}},Sm=e=>J(e),Cs=(e,t)=>{let{inputs:r,outputCount:n}=e,o=Sm({...t,outputCount:n});if(we.webgpu.validateInputContent&&$m(r,o),t.trainingMode)throw new Error("BatchNormalization trainingMode is not supported yet.");e.compute(xm(r,o))}});var Tm,Im,Es,ks=U(()=>{"use strict";te();re();Tm=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")},Im=e=>{let t=e[0].dims,r=e[0].dims[2],n=E.size(t)/4,o=e[0].dataType,i=k("input",o,t,4),a=k("bias",o,[r],4),d=k("residual",o,t,4),l=M("output",o,t,4);return{name:"BiasAdd",getRunData:()=>({outputs:[{dims:t,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(n/64)}}),getShaderSource:m=>`
+  }`;return{name:"BatchNormalization",shaderCache:{hint:`${t.epsilon}_${t.format}_${n}_${a}`,inputDependencies:c?["rank","type","type","type","type"]:void 0},getShaderSource:v,getRunData:()=>({outputs:[{dims:e[0].dims,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(l/64)},programUniforms:c?[{type:12,data:l},...R(i)]:[{type:12,data:l}]})}},Sm=e=>J(e),Cs=(e,t)=>{let{inputs:r,outputCount:n}=e,o=Sm({...t,outputCount:n});if(_e.webgpu.validateInputContent&&$m(r,o),t.trainingMode)throw new Error("BatchNormalization trainingMode is not supported yet.");e.compute(xm(r,o))}});var Tm,Im,ks,Es=V(()=>{"use strict";ie();ae();Tm=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")},Im=e=>{let t=e[0].dims,r=e[0].dims[2],n=k.size(t)/4,o=e[0].dataType,i=E("input",o,t,4),a=E("bias",o,[r],4),d=E("residual",o,t,4),l=M("output",o,t,4);return{name:"BiasAdd",getRunData:()=>({outputs:[{dims:t,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(n/64)}}),getShaderSource:m=>`
   const channels = ${r}u / 4;
   ${m.declareVariables(i,a,d,l)}
 
@@ -4181,7 +4181,7 @@ var best_index : i32 = 0;`,`if (${n.getByIndices("input_indices")} ${t.selectLas
     let value = ${i.getByOffset("global_idx")}
       + ${a.getByOffset("global_idx % channels")} + ${d.getByOffset("global_idx")};
     ${l.setByOffset("global_idx","value")}
-  }`}},Es=e=>{Tm(e.inputs),e.compute(Im(e.inputs))}});var Cm,fe,Ps,Os,zs,Ds,Bs,Ms,Rs,Us,Vs,Am,Ns,Ws,Ls,Hs,Qt,Gs,Yr,Fs,qs,js,Ks,Ys,Xs,Zs,Qs,Js,eu,tu,ru,nu,ou,iu,au,su,uu,lo,co,du,lu,cu,Em,km,pu,Xr=U(()=>{"use strict";Z();te();Ie();re();Cm=(e,t,r,n,o,i,a)=>{let d=Math.ceil(t/4),l="";typeof o=="string"?l=`${o}(a)`:l=o("a");let c=k("inputData",r,[d],4),m=M("outputData",n,[d],4),u=[{name:"vec_size",type:"u32"}];return a&&u.push(...a),`
+  }`}},ks=e=>{Tm(e.inputs),e.compute(Im(e.inputs))}});var Cm,ge,Ps,zs,Os,Bs,Ds,Ms,Rs,Us,Vs,Am,Ns,Ws,Ls,Hs,Qt,Gs,Yr,Fs,qs,js,Ks,Ys,Xs,Zs,Qs,Js,eu,tu,ru,nu,ou,iu,au,su,uu,lo,co,du,lu,cu,km,Em,pu,Xr=V(()=>{"use strict";Q();ie();Ie();ae();Cm=(e,t,r,n,o,i,a)=>{let d=Math.ceil(t/4),l="";typeof o=="string"?l=`${o}(a)`:l=o("a");let c=E("inputData",r,[d],4),m=M("outputData",n,[d],4),u=[{name:"vec_size",type:"u32"}];return a&&u.push(...a),`
       ${e.registerUniforms(u).declareVariables(c,m)}
 
   ${i??""}
@@ -4191,7 +4191,7 @@ var best_index : i32 = 0;`,`if (${n.getByIndices("input_indices")} ${t.selectLas
 
     let a = ${c.getByOffset("global_idx")};
     ${m.setByOffset("global_idx",l)}
-  }`},fe=(e,t,r,n,o,i=e.dataType,a,d)=>{let l=[{type:12,data:Math.ceil(E.size(e.dims)/4)}];return a&&l.push(...a),{name:t,shaderCache:{hint:o,inputDependencies:["type"]},getShaderSource:c=>Cm(c,E.size(e.dims),e.dataType,i,r,n,d),getRunData:c=>({outputs:[{dims:e.dims,dataType:i}],dispatchGroup:{x:Math.ceil(E.size(c[0].dims)/64/4)},programUniforms:l})}},Ps=e=>{e.compute(fe(e.inputs[0],"Abs","abs"))},Os=e=>{e.compute(fe(e.inputs[0],"Acos","acos"))},zs=e=>{e.compute(fe(e.inputs[0],"Acosh","acosh"))},Ds=e=>{e.compute(fe(e.inputs[0],"Asin","asin"))},Bs=e=>{e.compute(fe(e.inputs[0],"Asinh","asinh"))},Ms=e=>{e.compute(fe(e.inputs[0],"Atan","atan"))},Rs=e=>{e.compute(fe(e.inputs[0],"Atanh","atanh"))},Us=e=>J(e),Vs=(e,t)=>{let r;switch(t.to){case 10:r="vec4<f16>";break;case 1:r="vec4<f32>";break;case 12:r="vec4<u32>";break;case 6:r="vec4<i32>";break;case 9:r="vec4<bool>";break;default:throw new RangeError(`not supported type (specified in attribute 'to' from 'Cast' operator): ${t.to}`)}e.compute(fe(e.inputs[0],"Cast",r,void 0,t.cacheKey,t.to))},Am=e=>{let t,r,n=e.length>=2&&e[1].data!==0,o=e.length>=3&&e[2].data!==0;switch(e[0].dataType){case 1:t=n?e[1].getFloat32Array()[0]:-34028234663852886e22,r=o?e[2].getFloat32Array()[0]:34028234663852886e22;break;case 10:t=n?e[1].getUint16Array()[0]:64511,r=o?e[2].getUint16Array()[0]:31743;break;default:throw new Error("Unsupport data type")}return J({min:t,max:r})},Ns=(e,t)=>{let r=t||Am(e.inputs),n=ke(e.inputs[0].dataType);e.compute(fe(e.inputs[0],"Clip",o=>`clamp(${o}, vec4<${n}>(uniforms.min), vec4<${n}>(uniforms.max))`,void 0,r.cacheKey,void 0,[{type:e.inputs[0].dataType,data:r.min},{type:e.inputs[0].dataType,data:r.max}],[{name:"min",type:n},{name:"max",type:n}]),{inputs:[0]})},Ws=e=>{e.compute(fe(e.inputs[0],"Ceil","ceil"))},Ls=e=>{e.compute(fe(e.inputs[0],"Cos","cos"))},Hs=e=>{e.compute(fe(e.inputs[0],"Cosh","cosh"))},Qt=e=>J(e),Gs=(e,t)=>{let r=ke(e.inputs[0].dataType);e.compute(fe(e.inputs[0],"Elu",n=>`elu_vf32(${n})`,`
+  }`},ge=(e,t,r,n,o,i=e.dataType,a,d)=>{let l=[{type:12,data:Math.ceil(k.size(e.dims)/4)}];return a&&l.push(...a),{name:t,shaderCache:{hint:o,inputDependencies:["type"]},getShaderSource:c=>Cm(c,k.size(e.dims),e.dataType,i,r,n,d),getRunData:c=>({outputs:[{dims:e.dims,dataType:i}],dispatchGroup:{x:Math.ceil(k.size(c[0].dims)/64/4)},programUniforms:l})}},Ps=e=>{e.compute(ge(e.inputs[0],"Abs","abs"))},zs=e=>{e.compute(ge(e.inputs[0],"Acos","acos"))},Os=e=>{e.compute(ge(e.inputs[0],"Acosh","acosh"))},Bs=e=>{e.compute(ge(e.inputs[0],"Asin","asin"))},Ds=e=>{e.compute(ge(e.inputs[0],"Asinh","asinh"))},Ms=e=>{e.compute(ge(e.inputs[0],"Atan","atan"))},Rs=e=>{e.compute(ge(e.inputs[0],"Atanh","atanh"))},Us=e=>J(e),Vs=(e,t)=>{let r;switch(t.to){case 10:r="vec4<f16>";break;case 1:r="vec4<f32>";break;case 12:r="vec4<u32>";break;case 6:r="vec4<i32>";break;case 9:r="vec4<bool>";break;default:throw new RangeError(`not supported type (specified in attribute 'to' from 'Cast' operator): ${t.to}`)}e.compute(ge(e.inputs[0],"Cast",r,void 0,t.cacheKey,t.to))},Am=e=>{let t,r,n=e.length>=2&&e[1].data!==0,o=e.length>=3&&e[2].data!==0;switch(e[0].dataType){case 1:t=n?e[1].getFloat32Array()[0]:-34028234663852886e22,r=o?e[2].getFloat32Array()[0]:34028234663852886e22;break;case 10:t=n?e[1].getUint16Array()[0]:64511,r=o?e[2].getUint16Array()[0]:31743;break;default:throw new Error("Unsupport data type")}return J({min:t,max:r})},Ns=(e,t)=>{let r=t||Am(e.inputs),n=Ee(e.inputs[0].dataType);e.compute(ge(e.inputs[0],"Clip",o=>`clamp(${o}, vec4<${n}>(uniforms.min), vec4<${n}>(uniforms.max))`,void 0,r.cacheKey,void 0,[{type:e.inputs[0].dataType,data:r.min},{type:e.inputs[0].dataType,data:r.max}],[{name:"min",type:n},{name:"max",type:n}]),{inputs:[0]})},Ws=e=>{e.compute(ge(e.inputs[0],"Ceil","ceil"))},Ls=e=>{e.compute(ge(e.inputs[0],"Cos","cos"))},Hs=e=>{e.compute(ge(e.inputs[0],"Cosh","cosh"))},Qt=e=>J(e),Gs=(e,t)=>{let r=Ee(e.inputs[0].dataType);e.compute(ge(e.inputs[0],"Elu",n=>`elu_vf32(${n})`,`
   const elu_alpha_ = ${r}(${t.alpha});
 
   fn elu_f32(a: ${r}) -> ${r} {
@@ -4212,7 +4212,7 @@ fn erf_vf32(v: vec4<${e}>) -> vec4<${e}> {
   let absv = abs(v);
   let x = 1.0 / (1.0 + r0 * absv);
   return sign(v) * (1.0 - ((((r5 * x + r4) * x + r3) * x + r2) * x + r1) * x * exp(-absv * absv));
-}`,Fs=e=>{let t=ke(e.inputs[0].dataType);e.compute(fe(e.inputs[0],"Erf",r=>`erf_vf32(${r})`,Yr(t)))},qs=e=>{e.compute(fe(e.inputs[0],"Exp","exp"))},js=e=>{e.compute(fe(e.inputs[0],"Floor","floor"))},Ks=e=>{let t=ke(e.inputs[0].dataType);e.compute(fe(e.inputs[0],"Gelu",r=>`0.5 * ${r} * (1.0 + erf_vf32(${r} * 0.7071067811865475))`,Yr(t)))},Ys=(e,t)=>{let r=ke(e.inputs[0].dataType);e.compute(fe(e.inputs[0],"LeakyRelu",n=>`select(leaky_relu_alpha_ * ${n}, ${n}, ${n} >= vec4<${r}>(0.0))`,`const leaky_relu_alpha_ = ${r}(${t.alpha});`,t.cacheKey))},Xs=e=>{e.compute(fe(e.inputs[0],"Not",t=>`!${t}`))},Zs=e=>{e.compute(fe(e.inputs[0],"Neg",t=>`-${t}`))},Qs=e=>{e.compute(fe(e.inputs[0],"Reciprocal",t=>`1.0/${t}`))},Js=e=>{let t=ke(e.inputs[0].dataType);e.compute(fe(e.inputs[0],"Relu",r=>`select(vec4<${t}>(0.0), ${r}, ${r} > vec4<${t}>(0.0))`))},eu=e=>{e.compute(fe(e.inputs[0],"Sigmoid",t=>`(1.0 / (1.0 + exp(-${t})))`))},tu=e=>J(e),ru=(e,t)=>{let r=ke(e.inputs[0].dataType);e.compute(fe(e.inputs[0],"HardSigmoid",n=>`max(vec4<${r}>(0.0), min(vec4<${r}>(1.0), ${t.alpha} * ${n} + vec4<${r}>(${t.beta})))`,void 0,t.cacheKey))},nu=e=>{e.compute(fe(e.inputs[0],"Sin","sin"))},ou=e=>{e.compute(fe(e.inputs[0],"Sinh","sinh"))},iu=e=>{e.compute(fe(e.inputs[0],"Sqrt","sqrt"))},au=e=>{e.compute(fe(e.inputs[0],"Tan","tan"))},su=e=>`sign(${e}) * (1 - exp(-2 * abs(${e}))) / (1 + exp(-2 * abs(${e})))`,uu=e=>{e.compute(fe(e.inputs[0],"Tanh",su))},lo=(e="f32")=>`
+}`,Fs=e=>{let t=Ee(e.inputs[0].dataType);e.compute(ge(e.inputs[0],"Erf",r=>`erf_vf32(${r})`,Yr(t)))},qs=e=>{e.compute(ge(e.inputs[0],"Exp","exp"))},js=e=>{e.compute(ge(e.inputs[0],"Floor","floor"))},Ks=e=>{let t=Ee(e.inputs[0].dataType);e.compute(ge(e.inputs[0],"Gelu",r=>`0.5 * ${r} * (1.0 + erf_vf32(${r} * 0.7071067811865475))`,Yr(t)))},Ys=(e,t)=>{let r=Ee(e.inputs[0].dataType);e.compute(ge(e.inputs[0],"LeakyRelu",n=>`select(leaky_relu_alpha_ * ${n}, ${n}, ${n} >= vec4<${r}>(0.0))`,`const leaky_relu_alpha_ = ${r}(${t.alpha});`,t.cacheKey))},Xs=e=>{e.compute(ge(e.inputs[0],"Not",t=>`!${t}`))},Zs=e=>{e.compute(ge(e.inputs[0],"Neg",t=>`-${t}`))},Qs=e=>{e.compute(ge(e.inputs[0],"Reciprocal",t=>`1.0/${t}`))},Js=e=>{let t=Ee(e.inputs[0].dataType);e.compute(ge(e.inputs[0],"Relu",r=>`select(vec4<${t}>(0.0), ${r}, ${r} > vec4<${t}>(0.0))`))},eu=e=>{e.compute(ge(e.inputs[0],"Sigmoid",t=>`(1.0 / (1.0 + exp(-${t})))`))},tu=e=>J(e),ru=(e,t)=>{let r=Ee(e.inputs[0].dataType);e.compute(ge(e.inputs[0],"HardSigmoid",n=>`max(vec4<${r}>(0.0), min(vec4<${r}>(1.0), ${t.alpha} * ${n} + vec4<${r}>(${t.beta})))`,void 0,t.cacheKey))},nu=e=>{e.compute(ge(e.inputs[0],"Sin","sin"))},ou=e=>{e.compute(ge(e.inputs[0],"Sinh","sinh"))},iu=e=>{e.compute(ge(e.inputs[0],"Sqrt","sqrt"))},au=e=>{e.compute(ge(e.inputs[0],"Tan","tan"))},su=e=>`sign(${e}) * (1 - exp(-2 * abs(${e}))) / (1 + exp(-2 * abs(${e})))`,uu=e=>{e.compute(ge(e.inputs[0],"Tanh",su))},lo=(e="f32")=>`
 const fast_gelu_a: ${e} = 0.5;
 const fast_gelu_b: ${e} = 0.7978845608028654;
 const fast_gelu_c: ${e} = 0.035677408136300125;
@@ -4220,7 +4220,7 @@ const fast_gelu_c: ${e} = 0.035677408136300125;
 fn tanh_v(v: vec4<${e}>) -> vec4<${e}> {
   return ${su("v")};
 }
-`,co=e=>`(fast_gelu_a + fast_gelu_a * tanh_v(${e} * (fast_gelu_c * ${e} * ${e} + fast_gelu_b))) * ${e}`,du=e=>{let t=ke(e.inputs[0].dataType);e.compute(fe(e.inputs[0],"FastGelu",co,lo(t),void 0,e.inputs[0].dataType))},lu=(e,t)=>{let r=ke(e.inputs[0].dataType);return e.compute(fe(e.inputs[0],"ThresholdedRelu",n=>`select(vec4<${r}>(0.0), ${n}, ${n} > thresholded_relu_alpha_)`,`const thresholded_relu_alpha_ = vec4<${r}>(${t.alpha});`,t.cacheKey)),0},cu=e=>{e.compute(fe(e.inputs[0],"Log","log"))},Em=(e,t)=>`
+`,co=e=>`(fast_gelu_a + fast_gelu_a * tanh_v(${e} * (fast_gelu_c * ${e} * ${e} + fast_gelu_b))) * ${e}`,du=e=>{let t=Ee(e.inputs[0].dataType);e.compute(ge(e.inputs[0],"FastGelu",co,lo(t),void 0,e.inputs[0].dataType))},lu=(e,t)=>{let r=Ee(e.inputs[0].dataType);return e.compute(ge(e.inputs[0],"ThresholdedRelu",n=>`select(vec4<${r}>(0.0), ${n}, ${n} > thresholded_relu_alpha_)`,`const thresholded_relu_alpha_ = vec4<${r}>(${t.alpha});`,t.cacheKey)),0},cu=e=>{e.compute(ge(e.inputs[0],"Log","log"))},km=(e,t)=>`
 const alpha = vec4<${e}>(${t});
 const one = ${e}(1.0);
 const zero = ${e}(0.0);
@@ -4237,7 +4237,7 @@ fn quick_gelu_impl(x: vec4<${e}>) -> vec4<${e}> {
   }
   return x * x1;
 }
-`,km=e=>`quick_gelu_impl(${e})`,pu=(e,t)=>{let r=ke(e.inputs[0].dataType);e.compute(fe(e.inputs[0],"QuickGelu",km,Em(r,t.alpha),t.cacheKey,e.inputs[0].dataType))}});var Pm,Om,fu,hu=U(()=>{"use strict";te();re();Xr();Pm=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")},Om=e=>{let t=e[0].dims.slice();t[2]=t[2]/2;let r=k("input",e[0].dataType,e[0].dims,4),n=k("bias",e[0].dataType,[e[0].dims[2]],4),o=M("output",e[0].dataType,t,4),i=E.size(t)/4,a=ye(e[0].dataType);return{name:"BiasSplitGelu",getRunData:()=>({outputs:[{dims:t,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(i/64)}}),getShaderSource:l=>`
+`,Em=e=>`quick_gelu_impl(${e})`,pu=(e,t)=>{let r=Ee(e.inputs[0].dataType);e.compute(ge(e.inputs[0],"QuickGelu",Em,km(r,t.alpha),t.cacheKey,e.inputs[0].dataType))}});var Pm,zm,fu,hu=V(()=>{"use strict";ie();ae();Xr();Pm=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")},zm=e=>{let t=e[0].dims.slice();t[2]=t[2]/2;let r=E("input",e[0].dataType,e[0].dims,4),n=E("bias",e[0].dataType,[e[0].dims[2]],4),o=M("output",e[0].dataType,t,4),i=k.size(t)/4,a=he(e[0].dataType);return{name:"BiasSplitGelu",getRunData:()=>({outputs:[{dims:t,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(i/64)}}),getShaderSource:l=>`
   const M_SQRT2 = sqrt(2.0);
   const halfChannels = ${e[0].dims[2]/4/2}u;
 
@@ -4255,12 +4255,12 @@ fn quick_gelu_impl(x: vec4<${e}>) -> vec4<${e}> {
     let geluRight = valueRight * 0.5 * (erf_vf32(valueRight / M_SQRT2) + 1);
 
     ${o.setByOffset("global_idx","valueLeft * geluRight")}
-  }`}},fu=e=>{Pm(e.inputs),e.compute(Om(e.inputs))}});var zm,Dm,st,gu,yu,bu,wu,vu,_u,$u,xu,Su,Tu,Iu=U(()=>{"use strict";Z();te();re();zm=(e,t,r,n,o,i,a,d,l,c,m,u)=>{let h,w;typeof d=="string"?h=w=(_,x)=>`${d}((${_}),(${x}))`:typeof d=="function"?h=w=d:(h=d.scalar,w=d.vector);let g=M("outputData",m,n.length,4),y=k("aData",l,t.length,4),S=k("bData",c,r.length,4),$;if(o)if(i){let _=E.size(t)===1,x=E.size(r)===1,T=t.length>0&&t[t.length-1]%4===0,C=r.length>0&&r[r.length-1]%4===0;_||x?$=g.setByOffset("global_idx",w(_?`${y.type.value}(${y.getByOffset("0")}.x)`:y.getByOffset("global_idx"),x?`${S.type.value}(${S.getByOffset("0")}.x)`:S.getByOffset("global_idx"))):$=`
+  }`}},fu=e=>{Pm(e.inputs),e.compute(zm(e.inputs))}});var Om,Bm,st,gu,yu,bu,wu,_u,vu,$u,xu,Su,Tu,Iu=V(()=>{"use strict";Q();ie();ae();Om=(e,t,r,n,o,i,a,d,l,c,m,u)=>{let h,w;typeof d=="string"?h=w=(v,x)=>`${d}((${v}),(${x}))`:typeof d=="function"?h=w=d:(h=d.scalar,w=d.vector);let g=M("outputData",m,n.length,4),y=E("aData",l,t.length,4),S=E("bData",c,r.length,4),$;if(o)if(i){let v=k.size(t)===1,x=k.size(r)===1,T=t.length>0&&t[t.length-1]%4===0,C=r.length>0&&r[r.length-1]%4===0;v||x?$=g.setByOffset("global_idx",w(v?`${y.type.value}(${y.getByOffset("0")}.x)`:y.getByOffset("global_idx"),x?`${S.type.value}(${S.getByOffset("0")}.x)`:S.getByOffset("global_idx"))):$=`
             let outputIndices = ${g.offsetToIndices("global_idx * 4u")};
             let offsetA = ${y.broadcastedIndicesToOffset("outputIndices",g)};
             let offsetB = ${S.broadcastedIndicesToOffset("outputIndices",g)};
             ${g.setByOffset("global_idx",w(a||T?y.getByOffset("offsetA / 4u"):`${y.type.value}(${y.getByOffset("offsetA / 4u")}[offsetA % 4u])`,a||C?S.getByOffset("offsetB / 4u"):`${S.type.value}(${S.getByOffset("offsetB / 4u")}[offsetB % 4u])`))}
-          `}else $=g.setByOffset("global_idx",w(y.getByOffset("global_idx"),S.getByOffset("global_idx")));else{if(!i)throw new Error("no necessary to use scalar implementation for element-wise binary op implementation.");let _=(x,T,C="")=>{let A=`aData[indexA${T}][componentA${T}]`,P=`bData[indexB${T}][componentB${T}]`;return`
+          `}else $=g.setByOffset("global_idx",w(y.getByOffset("global_idx"),S.getByOffset("global_idx")));else{if(!i)throw new Error("no necessary to use scalar implementation for element-wise binary op implementation.");let v=(x,T,C="")=>{let A=`aData[indexA${T}][componentA${T}]`,P=`bData[indexB${T}][componentB${T}]`;return`
             let outputIndices${T} = ${g.offsetToIndices(`global_idx * 4u + ${T}u`)};
             let offsetA${T} = ${y.broadcastedIndicesToOffset(`outputIndices${T}`,g)};
             let offsetB${T} = ${S.broadcastedIndicesToOffset(`outputIndices${T}`,g)};
@@ -4271,15 +4271,15 @@ fn quick_gelu_impl(x: vec4<${e}>) -> vec4<${e}> {
             ${x}[${T}] = ${C}(${h(A,P)});
           `};m===9?$=`
             var data = vec4<u32>(0);
-            ${_("data",0,"u32")}
-            ${_("data",1,"u32")}
-            ${_("data",2,"u32")}
-            ${_("data",3,"u32")}
+            ${v("data",0,"u32")}
+            ${v("data",1,"u32")}
+            ${v("data",2,"u32")}
+            ${v("data",3,"u32")}
             outputData[global_idx] = dot(vec4<u32>(0x1, 0x100, 0x10000, 0x1000000), vec4<u32>(data));`:$=`
-            ${_("outputData[global_idx]",0)}
-            ${_("outputData[global_idx]",1)}
-            ${_("outputData[global_idx]",2)}
-            ${_("outputData[global_idx]",3)}
+            ${v("outputData[global_idx]",0)}
+            ${v("outputData[global_idx]",1)}
+            ${v("outputData[global_idx]",2)}
+            ${v("outputData[global_idx]",3)}
           `}return`
         ${e.registerUniform("vec_size","u32").declareVariables(y,S,g)}
 
@@ -4288,7 +4288,7 @@ fn quick_gelu_impl(x: vec4<${e}>) -> vec4<${e}> {
         ${e.mainStart()}
         ${e.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.vec_size")}
         ${$}
-      }`},Dm=(e,t,r,n,o,i,a=r.dataType)=>{let d=!E.areEqual(r.dims,n.dims),l=r.dims,c=E.size(r.dims),m=!1,u=!1,h=[d];if(d){let w=rt.calcShape(r.dims,n.dims,!1);if(!w)throw new Error("Can't perform binary op on the given tensors");l=w,c=E.size(l);let g=E.size(r.dims)===1,y=E.size(n.dims)===1,S=r.dims.length>0&&r.dims[r.dims.length-1]%4===0,$=n.dims.length>0&&n.dims[n.dims.length-1]%4===0;h.push(g),h.push(y),h.push(S),h.push($);let _=1;for(let x=1;x<l.length;x++){let T=r.dims[r.dims.length-x]??1,C=n.dims[n.dims.length-x]??1;if(T===C)_*=T;else break}_%4===0?(u=!0,m=!0):(g||y||S||$)&&(m=!0)}else m=!0;return h.push(m),{name:e,shaderCache:{hint:t+h.map(w=>w.toString()).join("_"),inputDependencies:["rank","rank"]},getShaderSource:w=>zm(w,r.dims,n.dims,l,m,d,u,o,r.dataType,n.dataType,a,i),getRunData:()=>({outputs:[{dims:l,dataType:a}],dispatchGroup:{x:Math.ceil(c/64/4)},programUniforms:[{type:12,data:Math.ceil(E.size(l)/4)},...V(r.dims,n.dims,l)]})}},st=(e,t,r,n,o,i)=>{e.compute(Dm(t,o??"",e.inputs[0],e.inputs[1],r,n,i))},gu=e=>{st(e,"Add",(t,r)=>`${t}+${r}`)},yu=e=>{st(e,"Div",(t,r)=>`${t}/${r}`)},bu=e=>{st(e,"Equal",{scalar:(t,r)=>`u32(${t}==${r})`,vector:(t,r)=>`vec4<u32>(${t}==${r})`},void 0,void 0,9)},wu=e=>{st(e,"Mul",(t,r)=>`${t}*${r}`)},vu=e=>{let t=k("input",e.inputs[0].dataType,e.inputs[0].dims).type.value;st(e,"Pow",{scalar:(n,o)=>`pow_custom(${n},${o})`,vector:(n,o)=>`pow_vector_custom(${n},${o})`},`
+      }`},Bm=(e,t,r,n,o,i,a=r.dataType)=>{let d=!k.areEqual(r.dims,n.dims),l=r.dims,c=k.size(r.dims),m=!1,u=!1,h=[d];if(d){let w=rt.calcShape(r.dims,n.dims,!1);if(!w)throw new Error("Can't perform binary op on the given tensors");l=w,c=k.size(l);let g=k.size(r.dims)===1,y=k.size(n.dims)===1,S=r.dims.length>0&&r.dims[r.dims.length-1]%4===0,$=n.dims.length>0&&n.dims[n.dims.length-1]%4===0;h.push(g),h.push(y),h.push(S),h.push($);let v=1;for(let x=1;x<l.length;x++){let T=r.dims[r.dims.length-x]??1,C=n.dims[n.dims.length-x]??1;if(T===C)v*=T;else break}v%4===0?(u=!0,m=!0):(g||y||S||$)&&(m=!0)}else m=!0;return h.push(m),{name:e,shaderCache:{hint:t+h.map(w=>w.toString()).join("_"),inputDependencies:["rank","rank"]},getShaderSource:w=>Om(w,r.dims,n.dims,l,m,d,u,o,r.dataType,n.dataType,a,i),getRunData:()=>({outputs:[{dims:l,dataType:a}],dispatchGroup:{x:Math.ceil(c/64/4)},programUniforms:[{type:12,data:Math.ceil(k.size(l)/4)},...R(r.dims,n.dims,l)]})}},st=(e,t,r,n,o,i)=>{e.compute(Bm(t,o??"",e.inputs[0],e.inputs[1],r,n,i))},gu=e=>{st(e,"Add",(t,r)=>`${t}+${r}`)},yu=e=>{st(e,"Div",(t,r)=>`${t}/${r}`)},bu=e=>{st(e,"Equal",{scalar:(t,r)=>`u32(${t}==${r})`,vector:(t,r)=>`vec4<u32>(${t}==${r})`},void 0,void 0,9)},wu=e=>{st(e,"Mul",(t,r)=>`${t}*${r}`)},_u=e=>{let t=E("input",e.inputs[0].dataType,e.inputs[0].dims).type.value;st(e,"Pow",{scalar:(n,o)=>`pow_custom(${n},${o})`,vector:(n,o)=>`pow_vector_custom(${n},${o})`},`
     fn pow_custom(a : ${t}, b : ${t}) -> ${t} {
       if (b == ${t}(0.0)) {
         return ${t}(1.0);
@@ -4301,7 +4301,7 @@ fn quick_gelu_impl(x: vec4<${e}>) -> vec4<${e}> {
       // TODO: implement vectorized pow
       return vec4<${t}>(pow_custom(a.x, b.x), pow_custom(a.y, b.y), pow_custom(a.z, b.z), pow_custom(a.w, b.w));
     }
-      `)},_u=e=>{st(e,"Sub",(t,r)=>`${t}-${r}`)},$u=e=>{st(e,"Greater",{scalar:(t,r)=>`u32(${t}>${r})`,vector:(t,r)=>`vec4<u32>(${t}>${r})`},void 0,void 0,9)},xu=e=>{st(e,"Less",{scalar:(t,r)=>`u32(${t}<${r})`,vector:(t,r)=>`vec4<u32>(${t}<${r})`},void 0,void 0,9)},Su=e=>{st(e,"GreaterOrEqual",{scalar:(t,r)=>`u32(${t}>=${r})`,vector:(t,r)=>`vec4<u32>(${t}>=${r})`},void 0,void 0,9)},Tu=e=>{st(e,"LessOrEqual",{scalar:(t,r)=>`u32(${t}<=${r})`,vector:(t,r)=>`vec4<u32>(${t}<=${r})`},void 0,void 0,9)}});var Mm,Rm,Um,Vm,Cu,Au,Eu=U(()=>{"use strict";Z();te();Ie();re();Mm=(e,t)=>{if(!e||e.length<1)throw new Error("too few inputs");let r=0,n=e[r],o=n.dataType,i=n.dims.length;e.forEach((a,d)=>{if(d!==r){if(a.dataType!==o)throw new Error("input tensors should be one type");if(a.dims.length!==i)throw new Error("input tensors should have the same shape");a.dims.forEach((l,c)=>{if(c!==t&&l!==n.dims[c])throw new Error("non concat dimensions must match")})}})},Rm=(e,t)=>`
+      `)},vu=e=>{st(e,"Sub",(t,r)=>`${t}-${r}`)},$u=e=>{st(e,"Greater",{scalar:(t,r)=>`u32(${t}>${r})`,vector:(t,r)=>`vec4<u32>(${t}>${r})`},void 0,void 0,9)},xu=e=>{st(e,"Less",{scalar:(t,r)=>`u32(${t}<${r})`,vector:(t,r)=>`vec4<u32>(${t}<${r})`},void 0,void 0,9)},Su=e=>{st(e,"GreaterOrEqual",{scalar:(t,r)=>`u32(${t}>=${r})`,vector:(t,r)=>`vec4<u32>(${t}>=${r})`},void 0,void 0,9)},Tu=e=>{st(e,"LessOrEqual",{scalar:(t,r)=>`u32(${t}<=${r})`,vector:(t,r)=>`vec4<u32>(${t}<=${r})`},void 0,void 0,9)}});var Mm,Rm,Um,Vm,Cu,Au,ku=V(()=>{"use strict";Q();ie();Ie();ae();Mm=(e,t)=>{if(!e||e.length<1)throw new Error("too few inputs");let r=0,n=e[r],o=n.dataType,i=n.dims.length;e.forEach((a,d)=>{if(d!==r){if(a.dataType!==o)throw new Error("input tensors should be one type");if(a.dims.length!==i)throw new Error("input tensors should have the same shape");a.dims.forEach((l,c)=>{if(c!==t&&l!==n.dims[c])throw new Error("non concat dimensions must match")})}})},Rm=(e,t)=>`
   fn calculateInputIndex(index: u32) -> u32 {
     let sizeInConcatAxis = array<u32, ${e}u>(${t});
     for (var i: u32 = 0u; i < ${e}; i += 1u ) {
@@ -4311,7 +4311,7 @@ fn quick_gelu_impl(x: vec4<${e}>) -> vec4<${e}> {
     }
     return ${e}u;
   }`,Um=(e,t)=>{let r=e.length,n=[];for(let o=0;o<r;++o){let i=t.setByOffset("global_idx",e[o].getByIndices("indices"));r===1?n.push(i):o===0?n.push(`if (inputIndex == ${o}u) { ${i} }`):o===r-1?n.push(`else { ${i} }`):n.push(`else if (inputIndex == ${o}) { ${i} }`)}return n.join(`
-`)},Vm=(e,t,r,n)=>{let o=E.size(r),i=new Array(e.length),a=new Array(e.length),d=0,l=[],c=[],m=[{type:12,data:o}];for(let y=0;y<e.length;++y)d+=e[y].dims[t],i[y]=d,c.push(e[y].dims.length),a[y]=k(`input${y}`,n,c[y]),l.push("rank"),m.push({type:12,data:i[y]});for(let y=0;y<e.length;++y)m.push(...V(e[y].dims));m.push(...V(r));let u=M("output",n,r.length),h=u.indicesGet("indices",t),w=Array.from(Array(i.length).keys()).map(y=>`uniforms.sizeInConcatAxis${y}`).join(","),g=y=>`
+`)},Vm=(e,t,r,n)=>{let o=k.size(r),i=new Array(e.length),a=new Array(e.length),d=0,l=[],c=[],m=[{type:12,data:o}];for(let y=0;y<e.length;++y)d+=e[y].dims[t],i[y]=d,c.push(e[y].dims.length),a[y]=E(`input${y}`,n,c[y]),l.push("rank"),m.push({type:12,data:i[y]});for(let y=0;y<e.length;++y)m.push(...R(e[y].dims));m.push(...R(r));let u=M("output",n,r.length),h=u.indicesGet("indices",t),w=Array.from(Array(i.length).keys()).map(y=>`uniforms.sizeInConcatAxis${y}`).join(","),g=y=>`
 
   ${(()=>{y.registerUniform("outputSize","u32");for(let S=0;S<e.length;S++)y.registerUniform(`sizeInConcatAxis${S}`,"u32");return y.declareVariables(...a,u)})()}
 
@@ -4329,11 +4329,11 @@ fn quick_gelu_impl(x: vec4<${e}>) -> vec4<${e}> {
     }
 
     ${Um(a,u)}
-  }`;return{name:"Concat",shaderCache:{hint:`${t}`,inputDependencies:l},getRunData:()=>({outputs:[{dims:r,dataType:n}],dispatchGroup:{x:Math.ceil(o/64)},programUniforms:m}),getShaderSource:g}},Cu=(e,t)=>{let r=e.inputs,n=r[0].dims,o=E.normalizeAxis(t.axis,n.length);Mm(r,o);let i=n.slice();i[o]=r.reduce((d,l)=>d+(l.dims.length>o?l.dims[o]:0),0);let a=r.filter(d=>E.size(d.dims)>0);e.compute(Vm(a,o,i,r[0].dataType),{inputs:a})},Au=e=>J({axis:e.axis})});var He,Ge,Fe,Zr,ct=U(()=>{"use strict";Z();te();He=(e,t,r="f32")=>{switch(e.activation){case"Relu":return`value = max(value, ${t}(0.0));`;case"Sigmoid":return`value = (${t}(1.0) / (${t}(1.0) + exp(-value)));`;case"Clip":return`value = clamp(value, ${t}(${r}(uniforms.clip_min)), ${t}(${r}(uniforms.clip_max)));`;case"HardSigmoid":return`value = max(${t}(0.0), min(${t}(1.0), ${r}(uniforms.alpha) * value + ${r}(uniforms.beta)));`;case"LeakyRelu":return`value = select(${r}(uniforms.alpha) * value, value, value >= ${t}(0.0));`;case"Tanh":return`let e2x = exp(-2.0 * abs(value));
+  }`;return{name:"Concat",shaderCache:{hint:`${t}`,inputDependencies:l},getRunData:()=>({outputs:[{dims:r,dataType:n}],dispatchGroup:{x:Math.ceil(o/64)},programUniforms:m}),getShaderSource:g}},Cu=(e,t)=>{let r=e.inputs,n=r[0].dims,o=k.normalizeAxis(t.axis,n.length);Mm(r,o);let i=n.slice();i[o]=r.reduce((d,l)=>d+(l.dims.length>o?l.dims[o]:0),0);let a=r.filter(d=>k.size(d.dims)>0);e.compute(Vm(a,o,i,r[0].dataType),{inputs:a})},Au=e=>J({axis:e.axis})});var He,Ge,Fe,Zr,ct=V(()=>{"use strict";Q();ie();He=(e,t,r="f32")=>{switch(e.activation){case"Relu":return`value = max(value, ${t}(0.0));`;case"Sigmoid":return`value = (${t}(1.0) / (${t}(1.0) + exp(-value)));`;case"Clip":return`value = clamp(value, ${t}(${r}(uniforms.clip_min)), ${t}(${r}(uniforms.clip_max)));`;case"HardSigmoid":return`value = max(${t}(0.0), min(${t}(1.0), ${r}(uniforms.alpha) * value + ${r}(uniforms.beta)));`;case"LeakyRelu":return`value = select(${r}(uniforms.alpha) * value, value, value >= ${t}(0.0));`;case"Tanh":return`let e2x = exp(-2.0 * abs(value));
               value = sign(value) * (1.0 - e2x) / (1.0 + e2x);
-        `;case"":return"";default:throw new Error(`Unsupported activation ${e.activation}`)}},Ge=(e,t)=>{e.activation==="Clip"?t.push({type:1,data:e.clipMax},{type:1,data:e.clipMin}):e.activation==="HardSigmoid"?t.push({type:1,data:e.alpha},{type:1,data:e.beta}):e.activation==="LeakyRelu"&&t.push({type:1,data:e.alpha})},Fe=(e,t)=>{e.activation==="Clip"?t.push({name:"clip_max",type:"f32"},{name:"clip_min",type:"f32"}):e.activation==="HardSigmoid"?t.push({name:"alpha",type:"f32"},{name:"beta",type:"f32"}):e.activation==="LeakyRelu"&&t.push({name:"alpha",type:"f32"})},Zr=e=>{let t=e?.activation||"";if(t==="HardSigmoid"){let[r,n]=e?.activation_params||[.2,.5];return{activation:t,alpha:r,beta:n}}else if(t==="Clip"){let[r,n]=e?.activation_params||[Ya,Xa];return{activation:t,clipMax:n,clipMin:r}}else if(t==="LeakyRelu"){let[r]=e?.activation_params||[.01];return{activation:t,alpha:r}}return{activation:t}}});var ze,Qr,Jt=U(()=>{"use strict";ze=(e,t)=>{switch(e){case 1:return t;case 2:return`vec2<${t}>`;case 3:return`vec3<${t}>`;case 4:return`vec4<${t}>`;default:throw new Error(`${e}-component is not supported.`)}},Qr=e=>`
+        `;case"":return"";default:throw new Error(`Unsupported activation ${e.activation}`)}},Ge=(e,t)=>{e.activation==="Clip"?t.push({type:1,data:e.clipMax},{type:1,data:e.clipMin}):e.activation==="HardSigmoid"?t.push({type:1,data:e.alpha},{type:1,data:e.beta}):e.activation==="LeakyRelu"&&t.push({type:1,data:e.alpha})},Fe=(e,t)=>{e.activation==="Clip"?t.push({name:"clip_max",type:"f32"},{name:"clip_min",type:"f32"}):e.activation==="HardSigmoid"?t.push({name:"alpha",type:"f32"},{name:"beta",type:"f32"}):e.activation==="LeakyRelu"&&t.push({name:"alpha",type:"f32"})},Zr=e=>{let t=e?.activation||"";if(t==="HardSigmoid"){let[r,n]=e?.activation_params||[.2,.5];return{activation:t,alpha:r,beta:n}}else if(t==="Clip"){let[r,n]=e?.activation_params||[Ya,Xa];return{activation:t,clipMax:n,clipMin:r}}else if(t==="LeakyRelu"){let[r]=e?.activation_params||[.01];return{activation:t,alpha:r}}return{activation:t}}});var Oe,Qr,Jt=V(()=>{"use strict";Oe=(e,t)=>{switch(e){case 1:return t;case 2:return`vec2<${t}>`;case 3:return`vec3<${t}>`;case 4:return`vec4<${t}>`;default:throw new Error(`${e}-component is not supported.`)}},Qr=e=>`
       ${e?"value = value + getBiasByOutputCoords(coords);":""}
-      `});var Jr,po=U(()=>{"use strict";Jr=e=>`
+      `});var Jr,po=V(()=>{"use strict";Jr=e=>`
 fn getIndexFromCoords4D(coords : vec4<i32>, shape : vec4<i32>) -> i32 {
   return dot(coords, vec4<i32>(
       shape.y * shape.z * shape.w, shape.z * shape.w, shape.w, 1));
@@ -4342,7 +4342,7 @@ fn getOutputIndexFromCoords(coords : vec4<i32>) -> i32 {
   return dot(coords, vec4<i32>(
     i32(${e}.x), i32(${e}.y), i32(${e}.z), 1));
 }
-`});var Nm,Wm,er,ku,Lm,tr,Hm,en,rr=U(()=>{"use strict";Z();te();re();ct();Jt();Nm=(e,t)=>e?`
+`});var Nm,Wm,er,Eu,Lm,tr,Hm,en,rr=V(()=>{"use strict";Q();ie();ae();ct();Jt();Nm=(e,t)=>e?`
         mm_Asub[inputRow][inputCol] = mm_readA(batch,
           kStart + inputRow,
           globalRowStart / innerElementSize + inputCol${t?", batchIndices":""});
@@ -4432,7 +4432,7 @@ fn main(@builtin(local_invocation_id) localId : vec3<u32>,
   for (var innerRow = 0; innerRow < rowPerThread; innerRow = innerRow + 1) {
       mm_write(batch, globalRow + innerRow, globalCol, acc[innerRow]);
   }
-}`},ku=(e,t)=>e?`
+}`},Eu=(e,t)=>e?`
             mm_Asub[inputRow][inputCol] = mm_readA(batch,
               kStart + inputRow,
               globalRowStart + inputCol${t?", batchIndices":""});
@@ -4451,7 +4451,7 @@ fn main(@builtin(local_invocation_id) localId : vec3<u32>,
       // Load one tile of A into local memory.
       for (var inputRow = localRow; inputRow < ${h}; inputRow = inputRow + ${t[1]}) {
         for (var inputCol = localCol; inputCol < ${u}; inputCol = inputCol + ${t[0]}) {
-          ${ku(o,n)}
+          ${Eu(o,n)}
         }
       }
       // Load one tile of B into local memory.
@@ -4506,7 +4506,7 @@ for (var t = 0; t < num_tiles; t = t + 1) {
     for (var innerCol = 0; innerCol < ${g}; innerCol = innerCol + 1) {
       let inputRow = tileRowA + innerRow;
       let inputCol = tileColA + innerCol;
-      ${ku(o,n)}
+      ${Eu(o,n)}
     }
   }
 
@@ -4566,7 +4566,7 @@ fn main(@builtin(local_invocation_id) localId : vec3<u32>,
     var acc : array<array<${r}, colPerThread>, rowPerThread>;
     ${S}
   }
-`},Hm=(e,t,r,n,o,i=!1)=>{let[a,d,l]=o,[c,m,u,h]=n,w=Nt(a,l),g=Nt(d,l),y=ye(n[0].type.tensor),S=()=>{let x=m.rank,T=c.rank,C=`var aIndices: ${m.type.indices};`;for(let A=x-2-1,P=T-1;A>=0;A--,P--)C+=`
+`},Hm=(e,t,r,n,o,i=!1)=>{let[a,d,l]=o,[c,m,u,h]=n,w=Nt(a,l),g=Nt(d,l),y=he(n[0].type.tensor),S=()=>{let x=m.rank,T=c.rank,C=`var aIndices: ${m.type.indices};`;for(let A=x-2-1,P=T-1;A>=0;A--,P--)C+=`
 aIndices[${A}] = ${T>1?`batchIndices[${P}]`:"batchIndices"};`;return w.forEach(A=>{C+=`
 aIndices[${A}] = 0;`}),C+=`
 aIndices[${x-2}] = u32(row);
@@ -4575,8 +4575,8 @@ bIndices[${A}] = ${T>1?`batchIndices[${P}]`:"batchIndices"};`;return g.forEach(A
 bIndices[${A}] = 0;`}),C+=`
 bIndices[${x-2}] = u32(row);
                    bIndices[${x-1}] = u32(colIn);`,C};return`
-    fn mm_readA(batch: i32, row: i32, colIn: i32, batchIndices: ${c.type.indices}) -> ${ze(e,y)} {
-      var value = ${ze(e,y)}(0.0);
+    fn mm_readA(batch: i32, row: i32, colIn: i32, batchIndices: ${c.type.indices}) -> ${Oe(e,y)} {
+      var value = ${Oe(e,y)}(0.0);
       let col = colIn * ${e};
       if(row < uniforms.dim_a_outer && col < uniforms.dim_inner)
       {
@@ -4586,8 +4586,8 @@ bIndices[${x-2}] = u32(row);
       return value;
     }
 
-    fn mm_readB(batch: i32, row: i32, colIn: i32, batchIndices: ${c.type.indices}) -> ${ze(e,y)} {
-      var value = ${ze(e,y)}(0.0);
+    fn mm_readB(batch: i32, row: i32, colIn: i32, batchIndices: ${c.type.indices}) -> ${Oe(e,y)} {
+      var value = ${Oe(e,y)}(0.0);
       let col = colIn * ${e};
       if(row < uniforms.dim_inner && col < uniforms.dim_b_outer)
       {
@@ -4597,21 +4597,21 @@ bIndices[${x-2}] = u32(row);
       return value;
     }
 
-    fn mm_write(batch: i32, row: i32, colIn: i32, valueIn: ${ze(e,y)}) {
+    fn mm_write(batch: i32, row: i32, colIn: i32, valueIn: ${Oe(e,y)}) {
       let col = colIn * ${e};
       if (row < uniforms.dim_a_outer && col < uniforms.dim_b_outer) {
         var value = valueIn;
         let coords = vec3<i32>(batch, row, colIn);
-        ${t?`value = value + ${i?"bias[colIn]":`${ze(e,y)}(bias[row])`};`:""}
+        ${t?`value = value + ${i?"bias[colIn]":`${Oe(e,y)}(bias[row])`};`:""}
         ${r}
         ${h.setByIndices("vec3<u32>(coords)","value")}
       }
     }
-    `},en=(e,t,r,n,o=!1,i)=>{let a=e[0].dims,d=e[1].dims,l=a.slice(0,-2),c=d.slice(0,-2),m=n?n.slice(0,-2):r.slice(0,-2),u=E.size(m),h=a[a.length-2],w=a[a.length-1],g=d[d.length-1],y=w%4===0&&g%4===0,S=h<=8?[4,1,1]:[4,4,1],$=[8,8,1],_=[Math.ceil(g/$[0]/S[0]),Math.ceil(h/$[1]/S[1]),Math.ceil(u/$[2]/S[2])],x=y?4:1,T=[...l,h,w/x],C=T.length,A=[...c,w,g/x],P=A.length,D=[u,h,g/x],N=[{type:6,data:h},{type:6,data:g},{type:6,data:w}];Ge(t,N),N.push(...V(m,T,A));let L=["rank","rank"],K=e.length>2;K&&(N.push(...V(e[2].dims)),L.push("rank")),N.push(...V(D));let X=ie=>{let ae=m.length,le=Fr("batchDims",e[0].dataType,ae,1),Y=ye(e[0].dataType),me=k("a",e[0].dataType,C,x),pe=k("b",e[1].dataType,P,x),Q=M("result",e[0].dataType,D.length,x),be=[me,pe];if(K){let G=o?x:1;be.push(k("bias",e[2].dataType,e[2].dims.length,G))}let ne=[{name:"dim_a_outer",type:"i32"},{name:"dim_b_outer",type:"i32"},{name:"dim_inner",type:"i32"}];Fe(t,ne);let oe=ye(Q.type.tensor),se=He(t,Q.type.value,oe),R=Hm(x,K,se,[le,me,pe,Q],[l,c,m],o);return`
-  ${ie.registerUniforms(ne).registerInternalVariables(le).declareVariables(...be,Q)}
-  ${R}
-  ${y?er(S,$,Y,le):tr(S,$,Y,le)}
-                   `};return{name:"MatMul",shaderCache:{hint:`${S};${t.activation};${y};${o}`,inputDependencies:L},getRunData:()=>({outputs:[{dims:i?i(r):r,dataType:e[0].dataType}],dispatchGroup:{x:_[0],y:_[1],z:_[2]},programUniforms:N}),getShaderSource:X}}});var Gm,Pu,Ou=U(()=>{"use strict";Z();Xe();re();ct();Jt();po();rr();Gm=(e,t,r,n,o=!1,i,a=4,d=4,l=4,c="f32")=>{let m=L=>{switch(L){case 1:return"resData = x[xIndex];";case 3:return`resData = vec3<${c}>(x[xIndex], x[xIndex + 1], x[xIndex + 2]);`;case 4:return"resData = x[xIndex / 4];";default:throw new Error(`innerElementSize ${L} is not supported.`)}},u=L=>{switch(L){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 ${L} is not supported.`)}},h=e?`
+    `},en=(e,t,r,n,o=!1,i)=>{let a=e[0].dims,d=e[1].dims,l=a.slice(0,-2),c=d.slice(0,-2),m=n?n.slice(0,-2):r.slice(0,-2),u=k.size(m),h=a[a.length-2],w=a[a.length-1],g=d[d.length-1],y=w%4===0&&g%4===0,S=h<=8?[4,1,1]:[4,4,1],$=[8,8,1],v=[Math.ceil(g/$[0]/S[0]),Math.ceil(h/$[1]/S[1]),Math.ceil(u/$[2]/S[2])],x=y?4:1,T=[...l,h,w/x],C=T.length,A=[...c,w,g/x],P=A.length,B=[u,h,g/x],N=[{type:6,data:h},{type:6,data:g},{type:6,data:w}];Ge(t,N),N.push(...R(m,T,A));let W=["rank","rank"],K=e.length>2;K&&(N.push(...R(e[2].dims)),W.push("rank")),N.push(...R(B));let Z=ee=>{let se=m.length,de=Fr("batchDims",e[0].dataType,se,1),Y=he(e[0].dataType),le=E("a",e[0].dataType,C,x),ce=E("b",e[1].dataType,P,x),q=M("result",e[0].dataType,B.length,x),ue=[le,ce];if(K){let G=o?x:1;ue.push(E("bias",e[2].dataType,e[2].dims.length,G))}let re=[{name:"dim_a_outer",type:"i32"},{name:"dim_b_outer",type:"i32"},{name:"dim_inner",type:"i32"}];Fe(t,re);let ne=he(q.type.tensor),oe=He(t,q.type.value,ne),U=Hm(x,K,oe,[de,le,ce,q],[l,c,m],o);return`
+  ${ee.registerUniforms(re).registerInternalVariables(de).declareVariables(...ue,q)}
+  ${U}
+  ${y?er(S,$,Y,de):tr(S,$,Y,de)}
+                   `};return{name:"MatMul",shaderCache:{hint:`${S};${t.activation};${y};${o}`,inputDependencies:W},getRunData:()=>({outputs:[{dims:i?i(r):r,dataType:e[0].dataType}],dispatchGroup:{x:v[0],y:v[1],z:v[2]},programUniforms:N}),getShaderSource:Z}}});var Gm,Pu,zu=V(()=>{"use strict";Q();Xe();ae();ct();Jt();po();rr();Gm=(e,t,r,n,o=!1,i,a=4,d=4,l=4,c="f32")=>{let m=W=>{switch(W){case 1:return"resData = x[xIndex];";case 3:return`resData = vec3<${c}>(x[xIndex], x[xIndex + 1], x[xIndex + 2]);`;case 4:return"resData = x[xIndex / 4];";default:throw new Error(`innerElementSize ${W} is not supported.`)}},u=W=>{switch(W){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 ${W} is not supported.`)}},h=e?`
     let coord = vec4<i32>(batch, xRow, xCol, xCh);
     `:`
     let coord = vec4<i32>(batch, xCh, xRow, xCol);
@@ -4627,7 +4627,7 @@ bIndices[${x-2}] = u32(row);
       row,
       col / outWidth,
       col % outWidth);
-    `,g=e?"i32(uniforms.x_shape[1])":"i32(uniforms.x_shape[2])",y=e?"i32(uniforms.x_shape[2])":"i32(uniforms.x_shape[3])",S=e?"row":"col",$=e?"col":"row",_=`
+    `,g=e?"i32(uniforms.x_shape[1])":"i32(uniforms.x_shape[2])",y=e?"i32(uniforms.x_shape[2])":"i32(uniforms.x_shape[3])",S=e?"row":"col",$=e?"col":"row",v=`
     let inChannels = i32(uniforms.w_shape[2]);
     let outWidth = ${e?"i32(uniforms.result_shape[2])":"i32(uniforms.result_shape[3])"};
     let outRow = ${S} / outWidth;
@@ -4638,7 +4638,7 @@ bIndices[${x-2}] = u32(row);
     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 = ${$} % inChannels;
-    var resData = ${ze(a,c)}(0.0);
+    var resData = ${Oe(a,c)}(0.0);
     // The bounds checking is always needed since we use it to pad zero for
     // the 'same' padding type.
     if (xRow >= 0 && xRow < ${g} && xCol >= 0 && xCol < ${y}) {
@@ -4648,19 +4648,19 @@ bIndices[${x-2}] = u32(row);
     }
     return resData;`,x=e?t&&n?`
     let col = colIn * ${a};
-    ${_}`:`
+    ${v}`:`
     let col = colIn * ${a};
     if (row < uniforms.dim_a_outer && col < uniforms.dim_inner) {
-      ${_}
+      ${v}
     }
-    return ${ze(a,c)}(0.0);`:n&&r?`
+    return ${Oe(a,c)}(0.0);`:n&&r?`
     let col = colIn * ${a};
-    ${_}`:`
+    ${v}`:`
     let col = colIn * ${a};
     if (row < uniforms.dim_inner && col < uniforms.dim_b_outer) {
-      ${_}
+      ${v}
     }
-    return ${ze(a,c)}(0.0);`,T=`${u(d)}`,C=ze(l,c),A=e?ze(a,c):ze(d,c),P=e?ze(d,c):ze(a,c),D=He(i,C,c);return`
+    return ${Oe(a,c)}(0.0);`,T=`${u(d)}`,C=Oe(l,c),A=e?Oe(a,c):Oe(d,c),P=e?Oe(d,c):Oe(a,c),B=He(i,C,c);return`
     fn mm_readA(batch: i32, row : i32, colIn : i32) -> ${A} {
       ${e?x:T}
     }
@@ -4677,32 +4677,32 @@ bIndices[${x-2}] = u32(row);
       let outWidth = ${e?"i32(uniforms.result_shape[2])":"i32(uniforms.result_shape[3])"};
       ${w}
       ${Qr(o)}
-      ${D}
+      ${B}
       setOutputAtCoords(coords[0], coords[1], coords[2], coords[3], value);
       }
-    }`},Pu=(e,t,r,n,o,i,a,d,l)=>{let c=t.format==="NHWC",m=c?e[0].dims[3]:e[0].dims[1],u=r[0],h=c?r[2]:r[3],w=c?r[1]:r[2],g=c?r[3]:r[1],y=c&&(m%4===0||m%3===0)&&g%4===0,S=c?g:h*w,$=c?h*w:g,_=[8,8,1],x=n<=8?[4,1,1]:[4,4,1],T=[Math.ceil(S/_[0]/x[0]),Math.ceil($/_[1]/x[1]),Math.ceil(u/_[2]/x[2])];de("verbose",()=>`[conv2d_mm_webgpu] dispatch = ${T}`);let C=y?c&&m%4!==0?3:4:1,A=_[1]*x[1],P=_[0]*x[0],D=Math.max(_[0]*C,_[1]),N=n%A===0,L=o%P===0,K=i%D===0,X=y?[C,4,4]:[1,1,1],ie=[{type:6,data:n},{type:6,data:o},{type:6,data:i},{type:6,data:[t.pads[0],t.pads[1]]},{type:6,data:t.strides},{type:6,data:t.dilations}];Ge(t,ie),ie.push(...V(e[0].dims,e[1].dims));let ae=["rank","rank"];a&&(ie.push(...V(e[2].dims)),ae.push("rank")),ie.push(...V(r));let le=Y=>{let me=[{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}];Fe(t,me);let pe=y?4:1,Q=ye(e[0].dataType),be=`
-      fn setOutputAtIndex(flatIndex : i32, value : ${y?`vec4<${Q}>`:Q}) {
-        result[flatIndex] = ${y?`vec4<${Q}>`:Q}(value);
+    }`},Pu=(e,t,r,n,o,i,a,d,l)=>{let c=t.format==="NHWC",m=c?e[0].dims[3]:e[0].dims[1],u=r[0],h=c?r[2]:r[3],w=c?r[1]:r[2],g=c?r[3]:r[1],y=c&&(m%4===0||m%3===0)&&g%4===0,S=c?g:h*w,$=c?h*w:g,v=[8,8,1],x=n<=8?[4,1,1]:[4,4,1],T=[Math.ceil(S/v[0]/x[0]),Math.ceil($/v[1]/x[1]),Math.ceil(u/v[2]/x[2])];me("verbose",()=>`[conv2d_mm_webgpu] dispatch = ${T}`);let C=y?c&&m%4!==0?3:4:1,A=v[1]*x[1],P=v[0]*x[0],B=Math.max(v[0]*C,v[1]),N=n%A===0,W=o%P===0,K=i%B===0,Z=y?[C,4,4]:[1,1,1],ee=[{type:6,data:n},{type:6,data:o},{type:6,data:i},{type:6,data:[t.pads[0],t.pads[1]]},{type:6,data:t.strides},{type:6,data:t.dilations}];Ge(t,ee),ee.push(...R(e[0].dims,e[1].dims));let se=["rank","rank"];a&&(ee.push(...R(e[2].dims)),se.push("rank")),ee.push(...R(r));let de=Y=>{let le=[{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}];Fe(t,le);let ce=y?4:1,q=he(e[0].dataType),ue=`
+      fn setOutputAtIndex(flatIndex : i32, value : ${y?`vec4<${q}>`:q}) {
+        result[flatIndex] = ${y?`vec4<${q}>`:q}(value);
       }
-      fn setOutputAtCoords(d0 : i32, d1 : i32, d2 : i32, d3 : i32, value : ${y?`vec4<${Q}>`:Q}) {
+      fn setOutputAtCoords(d0 : i32, d1 : i32, d2 : i32, d3 : i32, value : ${y?`vec4<${q}>`:q}) {
         let flatIndex = getOutputIndexFromCoords(vec4<i32>(d0, d1, d2, d3));
         setOutputAtIndex(flatIndex ${y?"/ 4":""}, value);
-      }`,ne=k("x",e[0].dataType,e[0].dims.length,C===3?1:C),oe=k("w",e[1].dataType,e[1].dims.length,pe),se=[ne,oe],R=M("result",e[0].dataType,r.length,pe);if(a){let G=k("bias",e[2].dataType,e[2].dims.length,pe);se.push(G),be+=`
-        fn getBiasByOutputCoords(coords : vec4<i32>) -> ${y?`vec4<${Q}>`:Q} {
+      }`,re=E("x",e[0].dataType,e[0].dims.length,C===3?1:C),ne=E("w",e[1].dataType,e[1].dims.length,ce),oe=[re,ne],U=M("result",e[0].dataType,r.length,ce);if(a){let G=E("bias",e[2].dataType,e[2].dims.length,ce);oe.push(G),ue+=`
+        fn getBiasByOutputCoords(coords : vec4<i32>) -> ${y?`vec4<${q}>`:q} {
           return bias[coords.${c?"w":"y"}${y?"/ 4":""}];
         }`}return`
         ${Jr("uniforms.result_strides")}
         //struct Uniforms { xShape : vec4<i32>, wShape : vec4<i32>, outShape : vec4<i32>,
         //  outShapeStrides: vec3<i32>, filterDims : vec2<i32>, pad : vec2<i32>, stride : vec2<i32>,
         //  dilation : vec2<i32>, dimAOuter : i32, dimBOuter : i32, dimInner : i32 };
-        ${Y.registerUniforms(me).declareVariables(...se,R)}
-        ${be}
-        ${Gm(c,N,L,K,a,t,X[0],X[1],X[2],Q)}
-        ${y?er(x,_,Q,void 0,!c,D):tr(x,_,Q,void 0,!c,D,!1,void 0,d)}`};return{name:"Conv2DMatMul",shaderCache:{hint:`${t.cacheKey};${C};${y};${N};${L};${K};${A};${P};${D}`,inputDependencies:ae},getRunData:()=>({outputs:[{dims:l?l(r):r,dataType:e[0].dataType}],dispatchGroup:{x:T[0],y:T[1],z:T[2]},programUniforms:ie}),getShaderSource:le}}});var Fm,zu,tn,qm,Du,jm,Bu,Mu,Ru=U(()=>{"use strict";Z();Xe();te();re();ct();Jt();Fm=e=>{let t=1;for(let r=0;r<e.length;r++)t*=e[r];return t},zu=e=>typeof e=="number"?[e,e,e]:e,tn=(e,t)=>t<=1?e:e+(e-1)*(t-1),qm=(e,t,r,n=1)=>{let o=tn(t,n);return Math.floor((e[0]*(r-1)-r+o)/2)},Du=(e,t,r,n,o)=>{o==null&&(o=qm(e,t[0],n[0]));let i=[0,0,0,r];for(let a=0;a<3;a++)e[a]+2*o>=t[a]&&(i[a]=Math.trunc((e[a]-t[a]+2*o)/n[a]+1));return i},jm=(e,t,r,n,o,i,a,d,l,c)=>{let m,u,h,w;if(e==="VALID"&&(e=0),typeof e=="number"){m={top:e,bottom:e,left:e,right:e,front:e,back:e};let g=Du([t,r,n,1],[d,l,c],1,[o,i,a],e);u=g[0],h=g[1],w=g[2]}else if(Array.isArray(e)){if(!e.every((y,S,$)=>y===$[0]))throw Error(`Unsupported padding parameter: ${e}`);m={top:e[0],bottom:e[1],left:e[2],right:e[3],front:e[4],back:e[5]};let g=Du([t,r,n,1],[d,l,c],1,[o,i,a],e[0]);u=g[0],h=g[1],w=g[2]}else if(e==="SAME_UPPER"){u=Math.ceil(t/o),h=Math.ceil(r/i),w=Math.ceil(n/a);let g=(u-1)*o+d-t,y=(h-1)*i+l-r,S=(w-1)*a+c-n,$=Math.floor(g/2),_=g-$,x=Math.floor(y/2),T=y-x,C=Math.floor(S/2),A=S-C;m={top:x,bottom:T,left:C,right:A,front:$,back:_}}else throw Error(`Unknown padding parameter: ${e}`);return{padInfo:m,outDepth:u,outHeight:h,outWidth:w}},Bu=(e,t,r,n,o,i=!1,a="channelsLast")=>{let d,l,c,m,u;if(a==="channelsLast")[d,l,c,m,u]=e;else if(a==="channelsFirst")[d,u,l,c,m]=e;else throw new Error(`Unknown dataFormat ${a}`);let[h,,w,g,y]=t,[S,$,_]=zu(r),[x,T,C]=zu(n),A=tn(w,x),P=tn(g,T),D=tn(y,C),{padInfo:N,outDepth:L,outHeight:K,outWidth:X}=jm(o,l,c,m,S,$,_,A,P,D),ie=i?h*u:h,ae=[0,0,0,0,0];return a==="channelsFirst"?ae=[d,ie,L,K,X]:a==="channelsLast"&&(ae=[d,L,K,X,ie]),{batchSize:d,dataFormat:a,inDepth:l,inHeight:c,inWidth:m,inChannels:u,outDepth:L,outHeight:K,outWidth:X,outChannels:ie,padInfo:N,strideDepth:S,strideHeight:$,strideWidth:_,filterDepth:w,filterHeight:g,filterWidth:y,effectiveFilterDepth:A,effectiveFilterHeight:P,effectiveFilterWidth:D,dilationDepth:x,dilationHeight:T,dilationWidth:C,inShape:e,outShape:ae,filterShape:t}},Mu=(e,t,r,n,o,i)=>{let a=i==="channelsLast",d=a?e[0].dims[3]:e[0].dims[1],l=!1,c=[64,1,1],m={x:r.map((_,x)=>x)},u=[Math.ceil(Fm(m.x.map(_=>r[_]))/c[0]),1,1];de("verbose",()=>`[conv3d_naive_webgpu] dispatch = ${u}`);let h=l?a&&d%4!==0?3:4:1,w=E.size(r),g=[{type:12,data:w},{type:12,data:n},{type:12,data:o},{type:12,data:t.strides},{type:12,data:t.dilations}];Ge(t,g),g.push(...V(e[0].dims,e[1].dims));let y=["rank","rank"],S=e.length===3;S&&(g.push(...V(e[2].dims)),y.push("rank")),g.push(...V(r));let $=_=>{let x=[{name:"output_size",type:"u32"},{name:"filter_dims",type:"u32",length:n.length},{name:"pads",type:"u32",length:o.length},{name:"strides",type:"u32",length:t.strides.length},{name:"dilations",type:"u32",length:t.dilations.length}];Fe(t,x);let T=l?4:1,C=ye(e[0].dataType),A=k("x",e[0].dataType,e[0].dims.length,h===3?1:h),P=k("W",e[1].dataType,e[1].dims.length,T),D=[A,P],N=M("result",e[0].dataType,r.length,T),L="";if(S){let ie=k("bias",e[2].dataType,e[2].dims.length,T);D.push(ie),L+=`
+        ${Y.registerUniforms(le).declareVariables(...oe,U)}
+        ${ue}
+        ${Gm(c,N,W,K,a,t,Z[0],Z[1],Z[2],q)}
+        ${y?er(x,v,q,void 0,!c,B):tr(x,v,q,void 0,!c,B,!1,void 0,d)}`};return{name:"Conv2DMatMul",shaderCache:{hint:`${t.cacheKey};${C};${y};${N};${W};${K};${A};${P};${B}`,inputDependencies:se},getRunData:()=>({outputs:[{dims:l?l(r):r,dataType:e[0].dataType}],dispatchGroup:{x:T[0],y:T[1],z:T[2]},programUniforms:ee}),getShaderSource:de}}});var Fm,Ou,tn,qm,Bu,jm,Du,Mu,Ru=V(()=>{"use strict";Q();Xe();ie();ae();ct();Jt();Fm=e=>{let t=1;for(let r=0;r<e.length;r++)t*=e[r];return t},Ou=e=>typeof e=="number"?[e,e,e]:e,tn=(e,t)=>t<=1?e:e+(e-1)*(t-1),qm=(e,t,r,n=1)=>{let o=tn(t,n);return Math.floor((e[0]*(r-1)-r+o)/2)},Bu=(e,t,r,n,o)=>{o==null&&(o=qm(e,t[0],n[0]));let i=[0,0,0,r];for(let a=0;a<3;a++)e[a]+2*o>=t[a]&&(i[a]=Math.trunc((e[a]-t[a]+2*o)/n[a]+1));return i},jm=(e,t,r,n,o,i,a,d,l,c)=>{let m,u,h,w;if(e==="VALID"&&(e=0),typeof e=="number"){m={top:e,bottom:e,left:e,right:e,front:e,back:e};let g=Bu([t,r,n,1],[d,l,c],1,[o,i,a],e);u=g[0],h=g[1],w=g[2]}else if(Array.isArray(e)){if(!e.every((y,S,$)=>y===$[0]))throw Error(`Unsupported padding parameter: ${e}`);m={top:e[0],bottom:e[1],left:e[2],right:e[3],front:e[4],back:e[5]};let g=Bu([t,r,n,1],[d,l,c],1,[o,i,a],e[0]);u=g[0],h=g[1],w=g[2]}else if(e==="SAME_UPPER"){u=Math.ceil(t/o),h=Math.ceil(r/i),w=Math.ceil(n/a);let g=(u-1)*o+d-t,y=(h-1)*i+l-r,S=(w-1)*a+c-n,$=Math.floor(g/2),v=g-$,x=Math.floor(y/2),T=y-x,C=Math.floor(S/2),A=S-C;m={top:x,bottom:T,left:C,right:A,front:$,back:v}}else throw Error(`Unknown padding parameter: ${e}`);return{padInfo:m,outDepth:u,outHeight:h,outWidth:w}},Du=(e,t,r,n,o,i=!1,a="channelsLast")=>{let d,l,c,m,u;if(a==="channelsLast")[d,l,c,m,u]=e;else if(a==="channelsFirst")[d,u,l,c,m]=e;else throw new Error(`Unknown dataFormat ${a}`);let[h,,w,g,y]=t,[S,$,v]=Ou(r),[x,T,C]=Ou(n),A=tn(w,x),P=tn(g,T),B=tn(y,C),{padInfo:N,outDepth:W,outHeight:K,outWidth:Z}=jm(o,l,c,m,S,$,v,A,P,B),ee=i?h*u:h,se=[0,0,0,0,0];return a==="channelsFirst"?se=[d,ee,W,K,Z]:a==="channelsLast"&&(se=[d,W,K,Z,ee]),{batchSize:d,dataFormat:a,inDepth:l,inHeight:c,inWidth:m,inChannels:u,outDepth:W,outHeight:K,outWidth:Z,outChannels:ee,padInfo:N,strideDepth:S,strideHeight:$,strideWidth:v,filterDepth:w,filterHeight:g,filterWidth:y,effectiveFilterDepth:A,effectiveFilterHeight:P,effectiveFilterWidth:B,dilationDepth:x,dilationHeight:T,dilationWidth:C,inShape:e,outShape:se,filterShape:t}},Mu=(e,t,r,n,o,i)=>{let a=i==="channelsLast",d=a?e[0].dims[3]:e[0].dims[1],l=!1,c=[64,1,1],m={x:r.map((v,x)=>x)},u=[Math.ceil(Fm(m.x.map(v=>r[v]))/c[0]),1,1];me("verbose",()=>`[conv3d_naive_webgpu] dispatch = ${u}`);let h=l?a&&d%4!==0?3:4:1,w=k.size(r),g=[{type:12,data:w},{type:12,data:n},{type:12,data:o},{type:12,data:t.strides},{type:12,data:t.dilations}];Ge(t,g),g.push(...R(e[0].dims,e[1].dims));let y=["rank","rank"],S=e.length===3;S&&(g.push(...R(e[2].dims)),y.push("rank")),g.push(...R(r));let $=v=>{let x=[{name:"output_size",type:"u32"},{name:"filter_dims",type:"u32",length:n.length},{name:"pads",type:"u32",length:o.length},{name:"strides",type:"u32",length:t.strides.length},{name:"dilations",type:"u32",length:t.dilations.length}];Fe(t,x);let T=l?4:1,C=he(e[0].dataType),A=E("x",e[0].dataType,e[0].dims.length,h===3?1:h),P=E("W",e[1].dataType,e[1].dims.length,T),B=[A,P],N=M("result",e[0].dataType,r.length,T),W="";if(S){let ee=E("bias",e[2].dataType,e[2].dims.length,T);B.push(ee),W+=`
         fn getBiasByOutputCoords(coords : array<u32, 5>) -> ${l?`vec4<${C}>`:C} {
           return bias[${a?F("coords",4,5):F("coords",1,5)}${l?"/ 4":""}];
-        }`}let K=ze(h,C),X=He(t,K,C);return`
-            ${L}
+        }`}let K=Oe(h,C),Z=He(t,K,C);return`
+            ${W}
             fn getX(d0 : u32, d1 : u32, d2 : u32, d3 : u32, d4 : u32) -> f32 {
               let aIndices = array<u32, 5>(d0, d1, d2, d3, d4);
               return ${A.getByIndices("aIndices")};
@@ -4711,9 +4711,9 @@ bIndices[${x-2}] = u32(row);
               let aIndices = array<u32, 5>(d0, d1, d2, d3, d4);
               return ${P.getByIndices("aIndices")};
             }
-          ${_.registerUniforms(x).declareVariables(...D,N)}
-          ${_.mainStart()}
-          ${_.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")}
+          ${v.registerUniforms(x).declareVariables(...B,N)}
+          ${v.mainStart()}
+          ${v.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")}
               let coords = ${N.offsetToIndices("global_idx")};
               let batch = ${F("coords",0,A.rank)};
               let d2 = ${a?F("coords",A.rank-1,A.rank):F("coords",1,A.rank)};
@@ -4804,9 +4804,9 @@ bIndices[${x-2}] = u32(row);
                 }
               }
               ${S?"value = value + getBiasByOutputCoords(coords)":""};
-              ${X}
+              ${Z}
               result[global_idx] = f32(value);
-          }`};return{name:"Conv3DNaive",shaderCache:{hint:`${t.cacheKey};${a};${h};${S}`,inputDependencies:y},getRunData:()=>({outputs:[{dims:r,dataType:e[0].dataType}],dispatchGroup:{x:u[0],y:u[1],z:u[2]},programUniforms:g}),getShaderSource:$}}});var Uu,Vu,Nu=U(()=>{"use strict";Z();te();re();ct();Uu=(e,t,r,n)=>{let o=e.length>2,i=o?"value += b[output_channel];":"",a=e[0].dims,d=e[1].dims,l=t.format==="NHWC",c=l?r[3]:r[1],m=c/t.group,u=l&&m>=4?ve(c):1,h=E.size(r)/u,w=[{type:12,data:h},{type:12,data:t.dilations},{type:12,data:[t.strides[0],t.strides[1]]},{type:12,data:[t.pads[0],t.pads[1]]},{type:12,data:m}];Ge(t,w),w.push(...V(a,[d[0],d[1],d[2],d[3]/u]));let g=o?["rank","rank","rank"]:["rank","rank"];w.push(...V([r[0],r[1],r[2],r[3]/u]));let y=S=>{let $=M("output",e[0].dataType,r.length,u),_=ye($.type.tensor),x=He(t,$.type.value,_),T=k("x",e[0].dataType,a.length),C=k("w",e[1].dataType,d.length,u),A=[T,C];o&&A.push(k("b",e[2].dataType,e[2].dims,u));let P=[{name:"output_size",type:"u32"},{name:"dilations",type:"u32",length:t.dilations.length},{name:"strides",type:"u32",length:2},{name:"pads",type:"u32",length:2},{name:"output_channels_per_group",type:"u32"}];Fe(t,P);let D=l?`
+          }`};return{name:"Conv3DNaive",shaderCache:{hint:`${t.cacheKey};${a};${h};${S}`,inputDependencies:y},getRunData:()=>({outputs:[{dims:r,dataType:e[0].dataType}],dispatchGroup:{x:u[0],y:u[1],z:u[2]},programUniforms:g}),getShaderSource:$}}});var Uu,Vu,Nu=V(()=>{"use strict";Q();ie();ae();ct();Uu=(e,t,r,n)=>{let o=e.length>2,i=o?"value += b[output_channel];":"",a=e[0].dims,d=e[1].dims,l=t.format==="NHWC",c=l?r[3]:r[1],m=c/t.group,u=l&&m>=4?we(c):1,h=k.size(r)/u,w=[{type:12,data:h},{type:12,data:t.dilations},{type:12,data:[t.strides[0],t.strides[1]]},{type:12,data:[t.pads[0],t.pads[1]]},{type:12,data:m}];Ge(t,w),w.push(...R(a,[d[0],d[1],d[2],d[3]/u]));let g=o?["rank","rank","rank"]:["rank","rank"];w.push(...R([r[0],r[1],r[2],r[3]/u]));let y=S=>{let $=M("output",e[0].dataType,r.length,u),v=he($.type.tensor),x=He(t,$.type.value,v),T=E("x",e[0].dataType,a.length),C=E("w",e[1].dataType,d.length,u),A=[T,C];o&&A.push(E("b",e[2].dataType,e[2].dims,u));let P=[{name:"output_size",type:"u32"},{name:"dilations",type:"u32",length:t.dilations.length},{name:"strides",type:"u32",length:2},{name:"pads",type:"u32",length:2},{name:"output_channels_per_group",type:"u32"}];Fe(t,P);let B=l?`
       for (var wHeight: u32 = 0u; wHeight < uniforms.w_shape[0]; wHeight++) {
         let xHeight = xRCCorner.x + wHeight * uniforms.dilations[0];
 
@@ -4864,11 +4864,11 @@ bIndices[${x-2}] = u32(row);
     var in_channel_offset = group_id * uniforms.w_shape[${l?2:1}];
 
     var value: ${$.type.value} = ${$.type.value}(0);
-    ${D}
+    ${B}
     ${i}
     ${x}
     ${$.setByOffset("global_idx","value")}
-  }`};return{name:"GroupedConv",shaderCache:{hint:`${t.cacheKey}_${u}`,inputDependencies:g},getRunData:()=>({outputs:[{dims:n?n(r):r,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(h/64)},programUniforms:w}),getShaderSource:y}},Vu=(e,t,r,n)=>{let o=e.length>2,i=ve(r[3]),a=ve(r[2]),d=E.size(r)/i/a,l=[e[0].dims[0],e[0].dims[1],e[0].dims[2],e[0].dims[3]/i],c=[e[1].dims[0],e[1].dims[1],e[1].dims[2],e[1].dims[3]/i],m=[r[0],r[1],r[2],r[3]/i],u=[{type:12,data:d},{type:6,data:[t.strides[0],t.strides[1]]},{type:6,data:[t.pads[0],t.pads[1]]}];Ge(t,u),u.push(...V(l,c,m));let h=(a-1)*t.strides[1]+c[1],w=g=>{let y=M("output",e[0].dataType,m.length,i),S=ye(y.type.tensor),$=He(t,y.type.value,S),_=k("x",e[0].dataType,l.length,i),x=k("w",e[1].dataType,c.length,i),T=[_,x];o&&T.push(k("b",e[2].dataType,e[2].dims,i));let C=o?"value += b[output_channel];":"",A=[{name:"output_size",type:"u32"},{name:"strides",type:"i32",length:2},{name:"pads",type:"i32",length:2}];return Fe(t,A),`
+  }`};return{name:"GroupedConv",shaderCache:{hint:`${t.cacheKey}_${u}`,inputDependencies:g},getRunData:()=>({outputs:[{dims:n?n(r):r,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(h/64)},programUniforms:w}),getShaderSource:y}},Vu=(e,t,r,n)=>{let o=e.length>2,i=we(r[3]),a=we(r[2]),d=k.size(r)/i/a,l=[e[0].dims[0],e[0].dims[1],e[0].dims[2],e[0].dims[3]/i],c=[e[1].dims[0],e[1].dims[1],e[1].dims[2],e[1].dims[3]/i],m=[r[0],r[1],r[2],r[3]/i],u=[{type:12,data:d},{type:6,data:[t.strides[0],t.strides[1]]},{type:6,data:[t.pads[0],t.pads[1]]}];Ge(t,u),u.push(...R(l,c,m));let h=(a-1)*t.strides[1]+c[1],w=g=>{let y=M("output",e[0].dataType,m.length,i),S=he(y.type.tensor),$=He(t,y.type.value,S),v=E("x",e[0].dataType,l.length,i),x=E("w",e[1].dataType,c.length,i),T=[v,x];o&&T.push(E("b",e[2].dataType,e[2].dims,i));let C=o?"value += b[output_channel];":"",A=[{name:"output_size",type:"u32"},{name:"strides",type:"i32",length:2},{name:"pads",type:"i32",length:2}];return Fe(t,A),`
   ${g.registerUniforms(A).declareVariables(...T,y)}
   ${g.mainStart()}
     ${g.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")}
@@ -4883,7 +4883,7 @@ bIndices[${x-2}] = u32(row);
 
     let x_corner = vec2<i32>(i32(row), i32(col)) * uniforms.strides - uniforms.pads;
 
-    var x_vals: array<${_.type.value}, ${h}>;
+    var x_vals: array<${v.type.value}, ${h}>;
     var values: array<${y.type.value}, ${a}>;
     let input_channel = output_channel;
     // Use constant instead of uniform can give better performance for w's height/width.
@@ -4893,9 +4893,9 @@ bIndices[${x-2}] = u32(row);
         for (var i = 0; i < ${h}; i++) {
           let x_width = x_corner.y + i;
           if (x_width >= 0 && u32(x_width) < uniforms.x_shape[2]) {
-            x_vals[i] = ${_.get("batch","u32(x_height)","u32(x_width)","input_channel")};
+            x_vals[i] = ${v.get("batch","u32(x_height)","u32(x_width)","input_channel")};
           } else {
-            x_vals[i] = ${_.type.value}(0);
+            x_vals[i] = ${v.type.value}(0);
           }
         }
         for (var w_width: u32 = 0u; w_width < ${c[1]}; w_width++) {
@@ -4913,14 +4913,14 @@ bIndices[${x-2}] = u32(row);
       ${$}
       ${y.set("batch","row","col + i","output_channel","value")};
     }
-  }`};return{name:"GroupedConv-Vectorize",shaderCache:{hint:`${t.cacheKey};${i};${a};${h};${c[0]};${c[1]}`,inputDependencies:o?["rank","rank","type"]:["rank","rank"]},getRunData:()=>({outputs:[{dims:n?n(r):r,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(d/64)},programUniforms:u}),getShaderSource:w}}});var mo,Km,Wu,fo=U(()=>{"use strict";Z();te();rr();re();ct();mo=(e,t,r,n,o=!1,i)=>{let a=e[0].dims,d=e[1].dims,l=a[a.length-2],c=d[d.length-1],m=a[a.length-1],u=ve(c),h=ve(m),w=ve(l),g=E.size(r)/u/w,y=e.length>2,S=n?n.slice(0,-2):r.slice(0,-2),_=[E.size(S),l,c],x=[{type:12,data:g},{type:12,data:l},{type:12,data:c},{type:12,data:m}];Ge(t,x),x.push(...V(S,a,d)),y&&x.push(...V(e[2].dims)),x.push(...V(_));let T=C=>{let A=Fr("batch_dims",e[0].dataType,S.length),P=k("a",e[0].dataType,a.length,h),D=k("b",e[1].dataType,d.length,u),N=M("output",e[0].dataType,_.length,u),L=ye(N.type.tensor),K=He(t,N.type.value,L),X=[P,D],ie="";if(y){let ne=o?u:1;X.push(k("bias",e[2].dataType,e[2].dims.length,ne)),ie=`${o?`value += bias[col / ${ne}];`:`value += ${N.type.value}(bias[row + i]);`}`}let ae=a.slice(0,-2),le=d.slice(0,-2),Y=Nt(ae,S),me=Nt(le,S),pe=[{name:"output_size",type:"u32"},{name:"M",type:"u32"},{name:"N",type:"u32"},{name:"K",type:"u32"}];Fe(t,pe);let Q=(ne,oe)=>{let se=ne.rank,R=ne.name;if(se===2)return`var ${R}_indices = ${ne.type.indices}(0u, 0u);`;let G=A.rank,he=`var ${R}_indices: ${ne.type.indices};`;for(let Re=se-2-1,$e=G-1;Re>=0;Re--,$e--)he+=`
-${R}_indices[${Re}] = ${G>1?`batch_indices[${$e}]`:"batch_indices"};`;return oe.forEach(Re=>{he+=`
-${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
-                     ${R}_indices[${se-1}] = 0u;`,he},be=()=>{let ne=`var a_data: ${P.type.value};`;for(let oe=0;oe<h;oe++)ne+=`
-              let b_data${oe} = b[(b_offset + (k + ${oe}) * uniforms.N + col) / ${u}];`;for(let oe=0;oe<w;oe++){ne+=`a_data = a[(a_offset + (row + ${oe}) * uniforms.K + k) / ${h}];`;for(let se=0;se<h;se++)ne+=`
-            values[${oe}] = fma(${D.type.value}(a_data${h===1?"":`[${se}]`}), b_data${se}, values[${oe}]);
-`}return ne};return`
-  ${C.registerUniforms(pe).registerInternalVariables(A).declareVariables(...X,N)}
+  }`};return{name:"GroupedConv-Vectorize",shaderCache:{hint:`${t.cacheKey};${i};${a};${h};${c[0]};${c[1]}`,inputDependencies:o?["rank","rank","type"]:["rank","rank"]},getRunData:()=>({outputs:[{dims:n?n(r):r,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(d/64)},programUniforms:u}),getShaderSource:w}}});var mo,Km,Wu,fo=V(()=>{"use strict";Q();ie();rr();ae();ct();mo=(e,t,r,n,o=!1,i)=>{let a=e[0].dims,d=e[1].dims,l=a[a.length-2],c=d[d.length-1],m=a[a.length-1],u=we(c),h=we(m),w=we(l),g=k.size(r)/u/w,y=e.length>2,S=n?n.slice(0,-2):r.slice(0,-2),v=[k.size(S),l,c],x=[{type:12,data:g},{type:12,data:l},{type:12,data:c},{type:12,data:m}];Ge(t,x),x.push(...R(S,a,d)),y&&x.push(...R(e[2].dims)),x.push(...R(v));let T=C=>{let A=Fr("batch_dims",e[0].dataType,S.length),P=E("a",e[0].dataType,a.length,h),B=E("b",e[1].dataType,d.length,u),N=M("output",e[0].dataType,v.length,u),W=he(N.type.tensor),K=He(t,N.type.value,W),Z=[P,B],ee="";if(y){let re=o?u:1;Z.push(E("bias",e[2].dataType,e[2].dims.length,re)),ee=`${o?`value += bias[col / ${re}];`:`value += ${N.type.value}(bias[row + i]);`}`}let se=a.slice(0,-2),de=d.slice(0,-2),Y=Nt(se,S),le=Nt(de,S),ce=[{name:"output_size",type:"u32"},{name:"M",type:"u32"},{name:"N",type:"u32"},{name:"K",type:"u32"}];Fe(t,ce);let q=(re,ne)=>{let oe=re.rank,U=re.name;if(oe===2)return`var ${U}_indices = ${re.type.indices}(0u, 0u);`;let G=A.rank,ye=`var ${U}_indices: ${re.type.indices};`;for(let Re=oe-2-1,$e=G-1;Re>=0;Re--,$e--)ye+=`
+${U}_indices[${Re}] = ${G>1?`batch_indices[${$e}]`:"batch_indices"};`;return ne.forEach(Re=>{ye+=`
+${U}_indices[${Re}] = 0;`}),ye+=`${U}_indices[${oe-2}] = 0u;
+                     ${U}_indices[${oe-1}] = 0u;`,ye},ue=()=>{let re=`var a_data: ${P.type.value};`;for(let ne=0;ne<h;ne++)re+=`
+              let b_data${ne} = b[(b_offset + (k + ${ne}) * uniforms.N + col) / ${u}];`;for(let ne=0;ne<w;ne++){re+=`a_data = a[(a_offset + (row + ${ne}) * uniforms.K + k) / ${h}];`;for(let oe=0;oe<h;oe++)re+=`
+            values[${ne}] = fma(${B.type.value}(a_data${h===1?"":`[${oe}]`}), b_data${oe}, values[${ne}]);
+`}return re};return`
+  ${C.registerUniforms(ce).registerInternalVariables(A).declareVariables(...Z,N)}
   ${C.mainStart()}
     ${C.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")}
     let col = (global_idx % (uniforms.N / ${u})) * ${u};
@@ -4930,24 +4930,24 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
     let batch = index1 / stride1;
 
     ${r.length===2?"":`let batch_indices = ${A.offsetToIndices("batch")};`}
-    ${Q(P,Y)}
+    ${q(P,Y)}
     let a_offset = ${P.indicesToOffset("a_indices")};
-    ${Q(D,me)}
-    let b_offset = ${D.indicesToOffset("b_indices")};
+    ${q(B,le)}
+    let b_offset = ${B.indicesToOffset("b_indices")};
     var values: array<${N.type.value}, ${w}>;
     for (var k: u32 = 0u; k < uniforms.K; k = k + ${h}) {
-      ${be()}
+      ${ue()}
     }
     for (var i = 0u; i < ${w}u; i++) {
       var value = values[i];
-      ${ie}
+      ${ee}
       ${K}
       let cur_indices = ${N.type.indices}(batch, row + i, col);
       let offset = ${N.indicesToOffset("cur_indices")};
       ${N.setByOffset(`offset / ${u}`,"value")};
     }
   }
-  `};return{name:"MatMulNaive",shaderCache:{hint:`${t.activation};${u};${h};${w};${o}`,inputDependencies:y?["rank","rank","rank"]:["rank","rank"]},getRunData:()=>({outputs:[{dims:i?i(r):r,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(g/64)},programUniforms:x}),getShaderSource:T}},Km=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.")},Wu=e=>{Km(e.inputs);let t=rt.calcShape(e.inputs[0].dims,e.inputs[1].dims,!0);if(!t)throw new Error("Can't use matmul on the given tensors");let r=t[t.length-1],n=e.inputs[0].dims[e.inputs[0].dims.length-1];r<8&&n<8?e.compute(mo(e.inputs,{activation:""},t)):e.compute(en(e.inputs,{activation:""},t))}});var Ym,ho,Xm,go,yo,Lu,Zm,Qm,bo,Hu=U(()=>{"use strict";te();Ou();Ru();rr();Nu();ct();fo();lt();Ym=(e,t,r,n,o,i)=>{let a=e[0],d=e.slice(i?1:2,i?3:4),l=d.length,c=t[0],u=t.slice(2).map((g,y)=>g+(g-1)*(r[y]-1)),w=d.map((g,y)=>g+n[y]+n[y+l]).map((g,y)=>Math.floor((g-u[y]+o[y])/o[y]));return w.splice(0,0,a),w.splice(i?3:1,0,c),w},ho=[2,3,1,0],Xm=(e,t)=>{if(!e||e.length!==2&&e.length!==3)throw new Error("Conv requires 2 or 3 inputs");if(e[0].dims.length>5)throw new Error("greater than 5D is not supported");if(e[0].dims.length!==e[1].dims.length)throw new Error("filter does not have same dimension as input");let r=e[0].dims[t.format==="NHWC"?e[0].dims.length-1:1],n=e[1].dims[1]*t.group;if(r!==n)throw new Error("FILTER_IN_CHANNEL should be equal to DATA_CHANNEL");if(e.length===3&&(e[2].dims.length!==1||e[1].dims[0]!==e[2].dims[0]))throw new Error("invalid bias");let o=e[0].dims.length-2;if(t.dilations.length!==o)throw new Error(`dilations should be ${o}D`);if(t.strides.length!==o)throw new Error(`strides should be ${o}D`);if(t.pads.length!==o*2)throw new Error(`pads should be ${o*2}D`);if(t.kernelShape.length!==0&&t.kernelShape.length!==e[1].dims.length-2)throw new Error("invalid kernel shape")},go=(e,t)=>{let r=e.kernelShape.slice();r.length<t[1].dims.length-2&&r.push(...Array(t[1].dims.length-2-r.length).fill(0));for(let i=2;i<t[1].dims.length;++i)r[i-2]===0&&(r[i-2]=t[1].dims[i]);let n=e.pads.slice();Ct.adjustPadsBasedOnAutoPad(t[0].dims,e.strides,e.dilations,r,n,e.format==="NHWC",e.autoPad);let o=Object.assign({},e);return Object.assign(o,{kernelShape:r,pads:n}),o},yo=e=>{let t=Zr(e),r=e.format,n=["NOTSET","VALID","SAME_UPPER","SAME_LOWER"][e.auto_pad],o=e.dilations,i=e.group,a=e.kernel_shape,d=e.pads,l=e.strides,c=e.w_is_const();return{autoPad:n,format:r,dilations:o,group:i,kernelShape:a,pads:d,strides:l,wIsConst:c,...t,cacheKey:`${e.format};${t.activation};`}},Lu=(e,t,r,n)=>{let o=r.format==="NHWC",i=Ym(t[0].dims,t[1].dims,r.dilations,r.pads,r.strides,o);if(r.group!==1){let A=[t[0]];if(o){let D=e.kernelCustomData.wT??e.compute(Pe(t[1],ho),{inputs:[1],outputs:[r.wIsConst?-2:-1]})[0];r.wIsConst&&!e.kernelCustomData.wT&&(e.kernelCustomData.wT=D),A.push(D)}else A.push(t[1]);t.length===3&&A.push(t[2]),!e.adapterInfo.isArchitecture("ampere")&&o&&t[1].dims[0]===r.group&&t[1].dims[1]===1&&r.dilations[0]===1&&r.dilations[1]===1?e.compute(Vu(A,r,i,n),{inputs:A}):e.compute(Uu(A,r,i,n),{inputs:A});return}let a=t.length===3,d=t[0].dims[o?1:2],l=t[0].dims[o?2:3],c=t[0].dims[o?3:1],m=t[1].dims[2],u=t[1].dims[3],h=i[o?1:2],w=i[o?2:3],g=i[o?3:1],y=o&&m===d&&u===l&&r.pads[0]===0&&r.pads[1]===0;if(y||m===1&&u===1&&r.dilations[0]===1&&r.dilations[1]===1&&r.strides[0]===1&&r.strides[1]===1&&r.pads[0]===0&&r.pads[1]===0){let A=i[0],P,D,N,L=[];if(o){let ie=e.kernelCustomData.wT??e.compute(Pe(t[1],ho),{inputs:[1],outputs:[r.wIsConst?-2:-1]})[0];if(r.wIsConst&&!e.kernelCustomData.wT&&(e.kernelCustomData.wT=ie),y){let ae=d*l*c;P=t[0].reshape([1,A,ae]),D=ie.reshape([1,ae,g]),N=[1,A,g]}else P=t[0].reshape([A,d*l,c]),D=ie.reshape([1,c,g]),N=[A,h*w,g];L.push(P),L.push(D)}else P=t[0].reshape([A,c,d*l]),D=t[1].reshape([1,g,c]),N=[A,g,h*w],L.push(D),L.push(P);a&&L.push(t[2]);let K=N[2],X=L[0].dims[L[0].dims.length-1];K<8&&X<8?e.compute(mo(L,r,i,N,o,n),{inputs:L}):e.compute(en(L,r,i,N,o,n),{inputs:L});return}let S=!0,$=e.kernelCustomData.wT??e.compute(Pe(t[1],ho),{inputs:[1],outputs:[r.wIsConst?-2:-1]})[0];r.wIsConst&&!e.kernelCustomData.wT&&(e.kernelCustomData.wT=$);let _=[t[0],$];a&&_.push(t[2]);let x=o?h*w:g,T=o?g:h*w,C=m*u*c;e.compute(Pu(_,r,i,x,T,C,a,S,n),{inputs:_})},Zm=(e,t)=>{let r=t.format==="NHWC",n=[e.inputs[0].reshape(r?[e.inputs[0].dims[0],1,e.inputs[0].dims[1],e.inputs[0].dims[2]]:[e.inputs[0].dims[0],e.inputs[0].dims[1],1,e.inputs[0].dims[2]]),e.inputs[1].reshape([e.inputs[1].dims[0],e.inputs[1].dims[1],1,e.inputs[1].dims[2]])];e.inputs.length===3&&n.push(e.inputs[2]);let o=[0,t.pads[0],0,t.pads[1]],i=[1].concat(t.strides),a=[1].concat(t.dilations),d=[1].concat(t.kernelShape),l=go({...t,pads:o,strides:i,dilations:a,kernelShape:d},n);Lu(e,n,l,c=>r?[c[0],c[2],c[3]]:[c[0],c[1],c[3]])},Qm=(e,t,r)=>{let n=r.format==="NHWC"?"channelsLast":"channelsFirst",o=go(r,t),i=r.autoPad==="NOTSET"?r.pads:r.autoPad,a=Bu(t[0].dims,t[1].dims,r.strides,r.dilations,i,!1,n);e.compute(Mu(t,o,a.outShape,[a.filterDepth,a.filterHeight,a.filterWidth],[a.padInfo.front,a.padInfo.top,a.padInfo.left],n))},bo=(e,t)=>{if(Xm(e.inputs,t),e.inputs[0].dims.length===3)Zm(e,t);else if(e.inputs[0].dims.length===5)Qm(e,e.inputs,t);else{let r=go(t,e.inputs);Lu(e,e.inputs,r)}}});var Jm,Gu,Fu=U(()=>{"use strict";Z();Xe();re();ct();Jt();po();rr();Jm=(e,t=!1,r,n,o=4)=>{let i=$=>{switch($){case 1:return"return w[getIndexFromCoords4D(coord, vec4<i32>(uniforms.w_shape))];";case 4:return`
+  `};return{name:"MatMulNaive",shaderCache:{hint:`${t.activation};${u};${h};${w};${o}`,inputDependencies:y?["rank","rank","rank"]:["rank","rank"]},getRunData:()=>({outputs:[{dims:i?i(r):r,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(g/64)},programUniforms:x}),getShaderSource:T}},Km=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.")},Wu=e=>{Km(e.inputs);let t=rt.calcShape(e.inputs[0].dims,e.inputs[1].dims,!0);if(!t)throw new Error("Can't use matmul on the given tensors");let r=t[t.length-1],n=e.inputs[0].dims[e.inputs[0].dims.length-1];r<8&&n<8?e.compute(mo(e.inputs,{activation:""},t)):e.compute(en(e.inputs,{activation:""},t))}});var Ym,ho,Xm,go,yo,Lu,Zm,Qm,bo,Hu=V(()=>{"use strict";ie();zu();Ru();rr();Nu();ct();fo();lt();Ym=(e,t,r,n,o,i)=>{let a=e[0],d=e.slice(i?1:2,i?3:4),l=d.length,c=t[0],u=t.slice(2).map((g,y)=>g+(g-1)*(r[y]-1)),w=d.map((g,y)=>g+n[y]+n[y+l]).map((g,y)=>Math.floor((g-u[y]+o[y])/o[y]));return w.splice(0,0,a),w.splice(i?3:1,0,c),w},ho=[2,3,1,0],Xm=(e,t)=>{if(!e||e.length!==2&&e.length!==3)throw new Error("Conv requires 2 or 3 inputs");if(e[0].dims.length>5)throw new Error("greater than 5D is not supported");if(e[0].dims.length!==e[1].dims.length)throw new Error("filter does not have same dimension as input");let r=e[0].dims[t.format==="NHWC"?e[0].dims.length-1:1],n=e[1].dims[1]*t.group;if(r!==n)throw new Error("FILTER_IN_CHANNEL should be equal to DATA_CHANNEL");if(e.length===3&&(e[2].dims.length!==1||e[1].dims[0]!==e[2].dims[0]))throw new Error("invalid bias");let o=e[0].dims.length-2;if(t.dilations.length!==o)throw new Error(`dilations should be ${o}D`);if(t.strides.length!==o)throw new Error(`strides should be ${o}D`);if(t.pads.length!==o*2)throw new Error(`pads should be ${o*2}D`);if(t.kernelShape.length!==0&&t.kernelShape.length!==e[1].dims.length-2)throw new Error("invalid kernel shape")},go=(e,t)=>{let r=e.kernelShape.slice();r.length<t[1].dims.length-2&&r.push(...Array(t[1].dims.length-2-r.length).fill(0));for(let i=2;i<t[1].dims.length;++i)r[i-2]===0&&(r[i-2]=t[1].dims[i]);let n=e.pads.slice();Ct.adjustPadsBasedOnAutoPad(t[0].dims,e.strides,e.dilations,r,n,e.format==="NHWC",e.autoPad);let o=Object.assign({},e);return Object.assign(o,{kernelShape:r,pads:n}),o},yo=e=>{let t=Zr(e),r=e.format,n=["NOTSET","VALID","SAME_UPPER","SAME_LOWER"][e.auto_pad],o=e.dilations,i=e.group,a=e.kernel_shape,d=e.pads,l=e.strides,c=e.w_is_const();return{autoPad:n,format:r,dilations:o,group:i,kernelShape:a,pads:d,strides:l,wIsConst:c,...t,cacheKey:`${e.format};${t.activation};`}},Lu=(e,t,r,n)=>{let o=r.format==="NHWC",i=Ym(t[0].dims,t[1].dims,r.dilations,r.pads,r.strides,o);if(r.group!==1){let A=[t[0]];if(o){let B=e.kernelCustomData.wT??e.compute(Pe(t[1],ho),{inputs:[1],outputs:[r.wIsConst?-2:-1]})[0];r.wIsConst&&!e.kernelCustomData.wT&&(e.kernelCustomData.wT=B),A.push(B)}else A.push(t[1]);t.length===3&&A.push(t[2]),!e.adapterInfo.isArchitecture("ampere")&&o&&t[1].dims[0]===r.group&&t[1].dims[1]===1&&r.dilations[0]===1&&r.dilations[1]===1?e.compute(Vu(A,r,i,n),{inputs:A}):e.compute(Uu(A,r,i,n),{inputs:A});return}let a=t.length===3,d=t[0].dims[o?1:2],l=t[0].dims[o?2:3],c=t[0].dims[o?3:1],m=t[1].dims[2],u=t[1].dims[3],h=i[o?1:2],w=i[o?2:3],g=i[o?3:1],y=o&&m===d&&u===l&&r.pads[0]===0&&r.pads[1]===0;if(y||m===1&&u===1&&r.dilations[0]===1&&r.dilations[1]===1&&r.strides[0]===1&&r.strides[1]===1&&r.pads[0]===0&&r.pads[1]===0){let A=i[0],P,B,N,W=[];if(o){let ee=e.kernelCustomData.wT??e.compute(Pe(t[1],ho),{inputs:[1],outputs:[r.wIsConst?-2:-1]})[0];if(r.wIsConst&&!e.kernelCustomData.wT&&(e.kernelCustomData.wT=ee),y){let se=d*l*c;P=t[0].reshape([1,A,se]),B=ee.reshape([1,se,g]),N=[1,A,g]}else P=t[0].reshape([A,d*l,c]),B=ee.reshape([1,c,g]),N=[A,h*w,g];W.push(P),W.push(B)}else P=t[0].reshape([A,c,d*l]),B=t[1].reshape([1,g,c]),N=[A,g,h*w],W.push(B),W.push(P);a&&W.push(t[2]);let K=N[2],Z=W[0].dims[W[0].dims.length-1];K<8&&Z<8?e.compute(mo(W,r,i,N,o,n),{inputs:W}):e.compute(en(W,r,i,N,o,n),{inputs:W});return}let S=!0,$=e.kernelCustomData.wT??e.compute(Pe(t[1],ho),{inputs:[1],outputs:[r.wIsConst?-2:-1]})[0];r.wIsConst&&!e.kernelCustomData.wT&&(e.kernelCustomData.wT=$);let v=[t[0],$];a&&v.push(t[2]);let x=o?h*w:g,T=o?g:h*w,C=m*u*c;e.compute(Pu(v,r,i,x,T,C,a,S,n),{inputs:v})},Zm=(e,t)=>{let r=t.format==="NHWC",n=[e.inputs[0].reshape(r?[e.inputs[0].dims[0],1,e.inputs[0].dims[1],e.inputs[0].dims[2]]:[e.inputs[0].dims[0],e.inputs[0].dims[1],1,e.inputs[0].dims[2]]),e.inputs[1].reshape([e.inputs[1].dims[0],e.inputs[1].dims[1],1,e.inputs[1].dims[2]])];e.inputs.length===3&&n.push(e.inputs[2]);let o=[0,t.pads[0],0,t.pads[1]],i=[1].concat(t.strides),a=[1].concat(t.dilations),d=[1].concat(t.kernelShape),l=go({...t,pads:o,strides:i,dilations:a,kernelShape:d},n);Lu(e,n,l,c=>r?[c[0],c[2],c[3]]:[c[0],c[1],c[3]])},Qm=(e,t,r)=>{let n=r.format==="NHWC"?"channelsLast":"channelsFirst",o=go(r,t),i=r.autoPad==="NOTSET"?r.pads:r.autoPad,a=Du(t[0].dims,t[1].dims,r.strides,r.dilations,i,!1,n);e.compute(Mu(t,o,a.outShape,[a.filterDepth,a.filterHeight,a.filterWidth],[a.padInfo.front,a.padInfo.top,a.padInfo.left],n))},bo=(e,t)=>{if(Xm(e.inputs,t),e.inputs[0].dims.length===3)Zm(e,t);else if(e.inputs[0].dims.length===5)Qm(e,e.inputs,t);else{let r=go(t,e.inputs);Lu(e,e.inputs,r)}}});var Jm,Gu,Fu=V(()=>{"use strict";Q();Xe();ae();ct();Jt();po();rr();Jm=(e,t=!1,r,n,o=4)=>{let i=$=>{switch($){case 1:return"return w[getIndexFromCoords4D(coord, vec4<i32>(uniforms.w_shape))];";case 4:return`
             let coord1 = vec4<i32>(coordX, coordY, col + 1, rowInner);
             let coord2 = vec4<i32>(coordX, coordY, col + 2, rowInner);
             let coord3 = vec4<i32>(coordX, coordY, col + 3, rowInner);
@@ -5032,21 +5032,21 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
       ${y}
       result[getIndexFromCoords4D(coords, vec4<i32>(uniforms.result_shape))/${o}] = value;
     }
-  }`},Gu=(e,t,r,n,o,i,a,d)=>{let l=t.format==="NHWC",c=l?e[0].dims[3]:e[0].dims[1],m=r[0],u=l?r[2]:r[3],h=l?r[1]:r[2],w=l?r[3]:r[1],g=l&&c%4===0&&c%3&&w%4===0,y=l?w:u*h,S=l?u*h:w,$=[8,8,1],_=n<=8?[4,1,1]:[4,4,1],x=[Math.ceil(y/$[0]/_[0]),Math.ceil(S/$[1]/_[1]),Math.ceil(m/$[2]/_[2])];de("verbose",()=>`[conv_backprop_mm_webgpu] dispatch = ${x}`);let T=g?4:1,C=Math.max($[0]*T,$[1]),A=g?4:1,P=[t.kernelShape[l?1:2],t.kernelShape[l?2:3]],D=[P[0]+(t.dilations[0]<=1?0:(P[0]-1)*(t.dilations[0]-1)),P[1]+(t.dilations[1]<=1?0:(P[1]-1)*(t.dilations[1]-1))],N=[D[0]-1-Math.floor((t.pads[0]+t.pads[2])/2),D[1]-1-Math.floor((t.pads[1]+t.pads[3])/2)],L=[{type:6,data:n},{type:6,data:o},{type:6,data:i},{type:6,data:t.strides},{type:6,data:t.dilations},{type:6,data:P},{type:6,data:N}];Ge(t,L),L.push(...V(e[0].dims,e[1].dims));let K=["rank","rank"];a&&(L.push(...V(e[2].dims)),K.push("rank")),L.push(...V(r));let X=ie=>{let ae=k("x",e[0].dataType,e[0].dims.length,A),le=k("w",e[1].dataType,e[1].dims.length,1),Y=M("result",e[0].dataType,r.length,A),me=[ae,le],pe="";if(a){let ne=k("bias",e[2].dataType,e[2].dims.length,A);me.push(ne),pe+=`
-          fn getBiasByOutputCoords(coords : vec4<i32>) -> ${ne.type.value} {
+  }`},Gu=(e,t,r,n,o,i,a,d)=>{let l=t.format==="NHWC",c=l?e[0].dims[3]:e[0].dims[1],m=r[0],u=l?r[2]:r[3],h=l?r[1]:r[2],w=l?r[3]:r[1],g=l&&c%4===0&&c%3&&w%4===0,y=l?w:u*h,S=l?u*h:w,$=[8,8,1],v=n<=8?[4,1,1]:[4,4,1],x=[Math.ceil(y/$[0]/v[0]),Math.ceil(S/$[1]/v[1]),Math.ceil(m/$[2]/v[2])];me("verbose",()=>`[conv_backprop_mm_webgpu] dispatch = ${x}`);let T=g?4:1,C=Math.max($[0]*T,$[1]),A=g?4:1,P=[t.kernelShape[l?1:2],t.kernelShape[l?2:3]],B=[P[0]+(t.dilations[0]<=1?0:(P[0]-1)*(t.dilations[0]-1)),P[1]+(t.dilations[1]<=1?0:(P[1]-1)*(t.dilations[1]-1))],N=[B[0]-1-Math.floor((t.pads[0]+t.pads[2])/2),B[1]-1-Math.floor((t.pads[1]+t.pads[3])/2)],W=[{type:6,data:n},{type:6,data:o},{type:6,data:i},{type:6,data:t.strides},{type:6,data:t.dilations},{type:6,data:P},{type:6,data:N}];Ge(t,W),W.push(...R(e[0].dims,e[1].dims));let K=["rank","rank"];a&&(W.push(...R(e[2].dims)),K.push("rank")),W.push(...R(r));let Z=ee=>{let se=E("x",e[0].dataType,e[0].dims.length,A),de=E("w",e[1].dataType,e[1].dims.length,1),Y=M("result",e[0].dataType,r.length,A),le=[se,de],ce="";if(a){let re=E("bias",e[2].dataType,e[2].dims.length,A);le.push(re),ce+=`
+          fn getBiasByOutputCoords(coords : vec4<i32>) -> ${re.type.value} {
             return bias[coords.${l?"w":"y"}${g?"/ 4":""}];
-          }`}let Q=[{name:"dim_a_outer",type:"i32"},{name:"dim_b_outer",type:"i32"},{name:"dim_inner",type:"i32"},{name:"strides",type:"i32",length:2},{name:"dilations",type:"i32",length:2},{name:"filter_dims",type:"i32",length:P.length},{name:"pads",type:"i32",length:N.length}];Fe(t,Q);let be=ye(e[0].dataType,1);if(be!=="f16"&&be!=="f32")throw new Error(`elemType ${be} is not supported.`);return`
+          }`}let q=[{name:"dim_a_outer",type:"i32"},{name:"dim_b_outer",type:"i32"},{name:"dim_inner",type:"i32"},{name:"strides",type:"i32",length:2},{name:"dilations",type:"i32",length:2},{name:"filter_dims",type:"i32",length:P.length},{name:"pads",type:"i32",length:N.length}];Fe(t,q);let ue=he(e[0].dataType,1);if(ue!=="f16"&&ue!=="f32")throw new Error(`elemType ${ue} is not supported.`);return`
         ${Jr("uniforms.result_strides")}
-        ${ie.registerUniforms(Q).declareVariables(...me,Y)};
-        ${pe}
-        ${Jm(l,a,t,ae.type.value,T)}
-        ${g?er(_,$,be,void 0,!l,C):tr(_,$,be,void 0,!l,C,!1,void 0,d)}`};return{name:"Conv2DTransposeMatMul",shaderCache:{hint:`${t.cacheKey};${_};${$};${g}`,inputDependencies:K},getRunData:()=>({outputs:[{dims:r,dataType:e[0].dataType}],dispatchGroup:{x:x[0],y:x[1],z:x[2]},programUniforms:L}),getShaderSource:X}}});var ef,wo,qu=U(()=>{"use strict";Z();Xe();te();re();ef=(e,t,r,n,o,i=!1,a,d,l=!1)=>{let c=l?1:2,m=l?2:3,u=l?3:1,h=i?2:1,w=`
+        ${ee.registerUniforms(q).declareVariables(...le,Y)};
+        ${ce}
+        ${Jm(l,a,t,se.type.value,T)}
+        ${g?er(v,$,ue,void 0,!l,C):tr(v,$,ue,void 0,!l,C,!1,void 0,d)}`};return{name:"Conv2DTransposeMatMul",shaderCache:{hint:`${t.cacheKey};${v};${$};${g}`,inputDependencies:K},getRunData:()=>({outputs:[{dims:r,dataType:e[0].dataType}],dispatchGroup:{x:x[0],y:x[1],z:x[2]},programUniforms:W}),getShaderSource:Z}}});var ef,wo,qu=V(()=>{"use strict";Q();Xe();ie();ae();ef=(e,t,r,n,o,i=!1,a,d,l=!1)=>{let c=l?1:2,m=l?2:3,u=l?3:1,h=i?2:1,w=`
   fn setOutputAtIndex(flatIndex : u32, value : ${i?`vec4<${a}>`:a}) {
     result[flatIndex] = ${i?`vec4<${a}>`:a}(value);
   }`;n&&(w+=`
     fn getBiasByOutputCoords(coords : vec4<u32>) -> ${i?`vec4<${a}>`:a} {
       return bias[coords.${l?"w":"y"}${i?"/ 4":""}];
-    }`);let g=i?4:1,y=k("W",t[1].dataType,t[1].dims.length,g),S=k("Dy",t[0].dataType,t[0].dims.length,g),$=[S,y];n&&$.push(k("bias",t[2].dataType,[r[u]].length,g));let _=M("result",t[0].dataType,r.length,g),x=`{
+    }`);let g=i?4:1,y=E("W",t[1].dataType,t[1].dims.length,g),S=E("Dy",t[0].dataType,t[0].dims.length,g),$=[S,y];n&&$.push(E("bias",t[2].dataType,[r[u]].length,g));let v=M("result",t[0].dataType,r.length,g),x=`{
         let batch: u32 = ${o?"global_id.z":"workgroup_id.z"} / uniforms.result_shape[1];
         let r = ${o?"global_id.z":"workgroup_id.z"} % uniforms.result_shape[1];
         let c = ${o?"global_id.y":"workgroup_id.y"} * ${h};
@@ -5147,14 +5147,14 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
 
         for (var i: u32 = 0; i < ${h}; i = i + 1) {
           let value = dotProd[i] + ${n?"bias[c+i]":`vec4<${a}>(0.0)`};
-          ${_.set("batch","r","c + i","d1","value")};
+          ${v.set("batch","r","c + i","d1","value")};
         }
       }`,T=`
-          let outputIndices = ${_.offsetToIndices("global_idx")};
-          let batch = ${_.indicesGet("outputIndices",0)};
-          let d1 = ${_.indicesGet("outputIndices",u)};
-          let r = ${_.indicesGet("outputIndices",c)};
-          let c = ${_.indicesGet("outputIndices",m)};
+          let outputIndices = ${v.offsetToIndices("global_idx")};
+          let batch = ${v.indicesGet("outputIndices",0)};
+          let d1 = ${v.indicesGet("outputIndices",u)};
+          let r = ${v.indicesGet("outputIndices",c)};
+          let c = ${v.indicesGet("outputIndices",m)};
           let dyCorner = vec2<i32>(i32(r), i32(c)) - uniforms.pads;
           let dyRCorner = dyCorner.x;
           let dyCCorner = dyCorner.y;
@@ -5196,14 +5196,14 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
             }
           }
           let value = dotProd + ${n?"bias[d1]":`${a}(0.0)`};
-          ${_.setByOffset("global_idx","value")};
+          ${v.setByOffset("global_idx","value")};
         `;return`
-  ${e.registerUniforms(d).declareVariables(...$,_)}
+  ${e.registerUniforms(d).declareVariables(...$,v)}
   ${w}
 
     ${e.mainStart()}
     ${e.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")};
-  ${i?x:T}}`},wo=(e,t,r)=>{let n=e.length>2,o=t.outputShape,i=E.size(o),a=[Math.ceil(i/64),1,1];de("verbose",()=>`[conv2d_backprop_webgpu] dispatch = ${a}`);let d=t.format==="NHWC",l=["rank","rank"],c=[t.strides[0],t.strides[1]],m=[t.kernelShape[d?1:2],t.kernelShape[d?2:3]],u=[t.dilations[0],t.dilations[1]],h=[m[0]+(t.dilations[0]<=1?0:(t.kernelShape[d?1:2]-1)*(t.dilations[0]-1)),m[1]+(t.dilations[1]<=1?0:(t.kernelShape[d?2:3]-1)*(t.dilations[1]-1))],w=[h[0]-1-Math.floor((t.pads[0]+t.pads[2])/2),h[1]-1-Math.floor(t.pads[1]+t.pads[3])/2],g=!1,y=t.group,S=e[1].dims,$=S[0]/y,_=S[1],x=[{type:12,data:i},{type:12,data:c},{type:12,data:m},{type:12,data:u},{type:12,data:h},{type:6,data:w},{type:12,data:$},{type:12,data:_},...V(e[0].dims,e[1].dims)];n&&(x.push(...V(e[2].dims)),l.push("rank")),x.push(...V(o));let T=a[1]===1&&a[2]===1,C=A=>{let P=[{name:"output_size",type:"u32"},{name:"strides",type:"u32",length:c.length},{name:"filter_dims",type:"u32",length:m.length},{name:"dilations",type:"u32",length:m.length},{name:"effective_filter_dims",type:"u32",length:h.length},{name:"pads",type:"i32",length:w.length},{name:"input_channels_per_group",type:"u32"},{name:"output_channels_per_group",type:"u32"}],D=ye(e[0].dataType);return`${ef(A,e,o,n,T,g,D,P,d)}`};return{name:"ConvTranspose2D",shaderCache:{hint:`${t.cacheKey};`,inputDependencies:l},getRunData:()=>({dispatchGroup:{x:a[0],y:a[1],z:a[2]},outputs:[{dims:r?r(o):o,dataType:e[0].dataType}],programUniforms:x}),getShaderSource:C}}});var tf,rf,nf,ju,Ku,of,af,sf,uf,Yu,Xu=U(()=>{"use strict";Fu();qu();ct();lt();tf=(e,t,r,n,o,i)=>(e-1)*t+r+(n-1)*o+1-i,rf=(e,t,r,n,o)=>{let i=Math.floor(e/2);t==="SAME_UPPER"?(r[n]=i,r[o]=e-i):t==="SAME_LOWER"&&(r[n]=e-i,r[o]=i)},nf=(e,t,r,n,o,i,a,d,l,c)=>{let m=e.length-2,u=c.length===0;l.length<m&&l.push(...Array(m-l.length).fill(0));let h=e[0],w=t[d?3:1]*o;for(let g=0,y=e.length-m-(d?1:0);g<m;++g,++y){let S=e[y],$=u?S*a[g]:c[g],_=tf(S,a[g],i[g],t[y],r[g],$);rf(_,n,i,g,g+m),u&&c.push(a[g]*(S-1)+l[g]+(t[y]-1)*r[g]+1-i[g]-i[g+m])}c.splice(0,0,h),c.splice(d?3:1,0,w)},ju=(e,t)=>{let r=e.kernelShape.slice();if(e.kernelShape.length===0||e.kernelShape.reduce((u,h)=>u*h,1)===0){r.length=0;for(let u=2;u<t[1].dims.length;++u)r.push(t[1].dims[u])}let n=e.format==="NHWC";r.splice(0,0,t[1].dims[0]),r.splice(n?3:1,0,t[1].dims[1]);let o=e.pads.slice(),i=e.outputShape.slice(),a=e.outputPadding.slice(),d=t[0].dims,l=e.dilations.slice();if(l.reduce((u,h)=>u+h,0)===0){let u=t[0].dims.length-2;l=new Array(u).fill(1)}let c=e.strides.slice();if(c.reduce((u,h)=>u+h,0)===0){let u=t[0].dims.length-2;c=new Array(u).fill(1)}nf(d,r,l,e.autoPad,e.group,o,c,n,a,i);let m=Object.assign({},e);return Object.assign(m,{kernelShape:r,pads:o,outputPadding:a,outputShape:i,dilations:l,strides:c}),m},Ku=e=>{let t=Zr(e),r=e.format,n=["NOTSET","VALID","SAME_UPPER","SAME_LOWER"][typeof e.autoPad>"u"?0:e.autoPad],o=e.dilations,i=e.group,a=e.kernelShape,d=e.pads,l=e.strides,c=e.wIsConst(),m=e.outputPadding,u=e.outputShape;return{autoPad:n,format:r,dilations:o,group:i,kernelShape:a,outputPadding:m,outputShape:u,pads:d,strides:l,wIsConst:c,...t,cacheKey:`${e.format};${t.activation};`}},of=(e,t)=>{if(!e||e.length!==2&&e.length!==3)throw new Error("Conv requires 2 or 3 inputs");if(e[0].dims.length!==4&&e[0].dims.length!==3)throw new Error("currently only support 2-dimensional conv");if(e[0].dims.length!==e[1].dims.length)throw new Error("filter does not have same dimension as input");let r=e[0].dims[t.format==="NHWC"?e[0].dims.length-1:1],n=e[1].dims[0];if(r!==n)throw new Error("FILTER_IN_CHANNEL should be equal to DATA_CHANNEL");let o=e[1].dims[1]*t.group;if(e.length===3&&(e[2].dims.length!==1||e[2].dims[0]!==o))throw new Error("invalid bias");let i=e[0].dims.length-2;if(t.dilations.reduce((m,u)=>m+u,0)>0&&t.dilations.length!==i)throw new Error(`dilations should be ${i}D`);if(t.strides.reduce((m,u)=>m+u,0)>0&&t.strides.length!==i)throw new Error(`strides should be ${i}D`);if(t.pads.reduce((m,u)=>m+u,0)>0&&t.pads.length!==i*2)throw new Error(`pads should be ${i*2}D`);if(t.outputPadding.length!==i&&t.outputPadding.length!==0)throw new Error(`output_padding should be ${i}D`);if(t.kernelShape.reduce((m,u)=>m+u,0)>0&&t.kernelShape.length!==0&&t.kernelShape.length!==e[1].dims.length-2)throw new Error("invalid kernel shape");if(t.outputShape.length!==0&&t.outputShape.length!==e[0].dims.length-2)throw new Error("invalid output shape")},af=[2,3,1,0],sf=(e,t,r)=>{let n=ju(r,t),o=r.format==="NHWC",i=n.outputShape,a=i[o?3:1],d=t[0].dims[o?3:1];if(n.group!==1||a===1&&d===1){e.compute(wo(t,n));return}let l=i[o?1:2],c=i[o?2:3],m=t[1].dims[2],u=t[1].dims[3],h=o?l*c:a,w=o?a:l*c,g=m*u*d,y=!0,S=e.kernelCustomData.wT??e.compute(Pe(t[1],af),{inputs:[1],outputs:[r.wIsConst?-2:-1]})[0];r.wIsConst&&!e.kernelCustomData.wT&&(e.kernelCustomData.wT=S);let $=[t[0],S],_=t.length===3;_&&(!o&&t[2].dims.length===1?$.push(t[2].reshape([t[2].dims[0],1,1])):$.push(t[2])),e.compute(Gu($,n,i,h,w,g,_,y),{inputs:$})},uf=(e,t)=>{let r=t.format==="NHWC",n=[e.inputs[0].reshape(r?[e.inputs[0].dims[0],1,e.inputs[0].dims[1],e.inputs[0].dims[2]]:[e.inputs[0].dims[0],e.inputs[0].dims[1],1,e.inputs[0].dims[2]]),e.inputs[1].reshape([e.inputs[1].dims[0],e.inputs[1].dims[1],1,e.inputs[1].dims[2]])];e.inputs.length===3&&n.push(e.inputs[2]);let o=t.kernelShape;(o.length===0||o[0]===0)&&(o=[e.inputs[1].dims[2]]);let i=t.dilations;(i.length===0||i[0]===0)&&(i=[1]);let a=t.strides;(a.length===0||a[0]===0)&&(a=[1]);let d=t.pads;d.length===0&&(d=[0,0]),d=[0,d[0],0,d[1]],a=[1].concat(a),i=[1].concat(i),o=[1].concat(o);let l=ju({...t,pads:d,strides:a,dilations:i,kernelShape:o},n);e.compute(wo(n,l,c=>r?[c[0],c[2],c[3]]:[c[0],c[1],c[3]]))},Yu=(e,t)=>{of(e.inputs,t),e.inputs[0].dims.length===3?uf(e,t):sf(e,e.inputs,t)}});var df,Zu,Qu,Ju=U(()=>{"use strict";Z();te();Ie();re();df=(e,t,r,n)=>{let o=E.size(t),i=t.length,a=k("input",e,i),d=M("output",e,i),l=r.dataType===6?r.getInt32Array()[0]:Number(r.getBigInt64Array()[0]),c=E.normalizeAxis(l,i),m=u=>{let h=` i32(${a.indicesGet("inputIndices","uniforms.axis")}) `,w=F("uniforms.input_shape","uniforms.axis",i),g=n.reverse?h+(n.exclusive?" + 1":""):"0",y=n.reverse?w:h+(n.exclusive?"":" + 1");return`
+  ${i?x:T}}`},wo=(e,t,r)=>{let n=e.length>2,o=t.outputShape,i=k.size(o),a=[Math.ceil(i/64),1,1];me("verbose",()=>`[conv2d_backprop_webgpu] dispatch = ${a}`);let d=t.format==="NHWC",l=["rank","rank"],c=[t.strides[0],t.strides[1]],m=[t.kernelShape[d?1:2],t.kernelShape[d?2:3]],u=[t.dilations[0],t.dilations[1]],h=[m[0]+(t.dilations[0]<=1?0:(t.kernelShape[d?1:2]-1)*(t.dilations[0]-1)),m[1]+(t.dilations[1]<=1?0:(t.kernelShape[d?2:3]-1)*(t.dilations[1]-1))],w=[h[0]-1-Math.floor((t.pads[0]+t.pads[2])/2),h[1]-1-Math.floor(t.pads[1]+t.pads[3])/2],g=!1,y=t.group,S=e[1].dims,$=S[0]/y,v=S[1],x=[{type:12,data:i},{type:12,data:c},{type:12,data:m},{type:12,data:u},{type:12,data:h},{type:6,data:w},{type:12,data:$},{type:12,data:v},...R(e[0].dims,e[1].dims)];n&&(x.push(...R(e[2].dims)),l.push("rank")),x.push(...R(o));let T=a[1]===1&&a[2]===1,C=A=>{let P=[{name:"output_size",type:"u32"},{name:"strides",type:"u32",length:c.length},{name:"filter_dims",type:"u32",length:m.length},{name:"dilations",type:"u32",length:m.length},{name:"effective_filter_dims",type:"u32",length:h.length},{name:"pads",type:"i32",length:w.length},{name:"input_channels_per_group",type:"u32"},{name:"output_channels_per_group",type:"u32"}],B=he(e[0].dataType);return`${ef(A,e,o,n,T,g,B,P,d)}`};return{name:"ConvTranspose2D",shaderCache:{hint:`${t.cacheKey};`,inputDependencies:l},getRunData:()=>({dispatchGroup:{x:a[0],y:a[1],z:a[2]},outputs:[{dims:r?r(o):o,dataType:e[0].dataType}],programUniforms:x}),getShaderSource:C}}});var tf,rf,nf,ju,Ku,of,af,sf,uf,Yu,Xu=V(()=>{"use strict";Fu();qu();ct();lt();tf=(e,t,r,n,o,i)=>(e-1)*t+r+(n-1)*o+1-i,rf=(e,t,r,n,o)=>{let i=Math.floor(e/2);t==="SAME_UPPER"?(r[n]=i,r[o]=e-i):t==="SAME_LOWER"&&(r[n]=e-i,r[o]=i)},nf=(e,t,r,n,o,i,a,d,l,c)=>{let m=e.length-2,u=c.length===0;l.length<m&&l.push(...Array(m-l.length).fill(0));let h=e[0],w=t[d?3:1]*o;for(let g=0,y=e.length-m-(d?1:0);g<m;++g,++y){let S=e[y],$=u?S*a[g]:c[g],v=tf(S,a[g],i[g],t[y],r[g],$);rf(v,n,i,g,g+m),u&&c.push(a[g]*(S-1)+l[g]+(t[y]-1)*r[g]+1-i[g]-i[g+m])}c.splice(0,0,h),c.splice(d?3:1,0,w)},ju=(e,t)=>{let r=e.kernelShape.slice();if(e.kernelShape.length===0||e.kernelShape.reduce((u,h)=>u*h,1)===0){r.length=0;for(let u=2;u<t[1].dims.length;++u)r.push(t[1].dims[u])}let n=e.format==="NHWC";r.splice(0,0,t[1].dims[0]),r.splice(n?3:1,0,t[1].dims[1]);let o=e.pads.slice(),i=e.outputShape.slice(),a=e.outputPadding.slice(),d=t[0].dims,l=e.dilations.slice();if(l.reduce((u,h)=>u+h,0)===0){let u=t[0].dims.length-2;l=new Array(u).fill(1)}let c=e.strides.slice();if(c.reduce((u,h)=>u+h,0)===0){let u=t[0].dims.length-2;c=new Array(u).fill(1)}nf(d,r,l,e.autoPad,e.group,o,c,n,a,i);let m=Object.assign({},e);return Object.assign(m,{kernelShape:r,pads:o,outputPadding:a,outputShape:i,dilations:l,strides:c}),m},Ku=e=>{let t=Zr(e),r=e.format,n=["NOTSET","VALID","SAME_UPPER","SAME_LOWER"][typeof e.autoPad>"u"?0:e.autoPad],o=e.dilations,i=e.group,a=e.kernelShape,d=e.pads,l=e.strides,c=e.wIsConst(),m=e.outputPadding,u=e.outputShape;return{autoPad:n,format:r,dilations:o,group:i,kernelShape:a,outputPadding:m,outputShape:u,pads:d,strides:l,wIsConst:c,...t,cacheKey:`${e.format};${t.activation};`}},of=(e,t)=>{if(!e||e.length!==2&&e.length!==3)throw new Error("Conv requires 2 or 3 inputs");if(e[0].dims.length!==4&&e[0].dims.length!==3)throw new Error("currently only support 2-dimensional conv");if(e[0].dims.length!==e[1].dims.length)throw new Error("filter does not have same dimension as input");let r=e[0].dims[t.format==="NHWC"?e[0].dims.length-1:1],n=e[1].dims[0];if(r!==n)throw new Error("FILTER_IN_CHANNEL should be equal to DATA_CHANNEL");let o=e[1].dims[1]*t.group;if(e.length===3&&(e[2].dims.length!==1||e[2].dims[0]!==o))throw new Error("invalid bias");let i=e[0].dims.length-2;if(t.dilations.reduce((m,u)=>m+u,0)>0&&t.dilations.length!==i)throw new Error(`dilations should be ${i}D`);if(t.strides.reduce((m,u)=>m+u,0)>0&&t.strides.length!==i)throw new Error(`strides should be ${i}D`);if(t.pads.reduce((m,u)=>m+u,0)>0&&t.pads.length!==i*2)throw new Error(`pads should be ${i*2}D`);if(t.outputPadding.length!==i&&t.outputPadding.length!==0)throw new Error(`output_padding should be ${i}D`);if(t.kernelShape.reduce((m,u)=>m+u,0)>0&&t.kernelShape.length!==0&&t.kernelShape.length!==e[1].dims.length-2)throw new Error("invalid kernel shape");if(t.outputShape.length!==0&&t.outputShape.length!==e[0].dims.length-2)throw new Error("invalid output shape")},af=[2,3,1,0],sf=(e,t,r)=>{let n=ju(r,t),o=r.format==="NHWC",i=n.outputShape,a=i[o?3:1],d=t[0].dims[o?3:1];if(n.group!==1||a===1&&d===1){e.compute(wo(t,n));return}let l=i[o?1:2],c=i[o?2:3],m=t[1].dims[2],u=t[1].dims[3],h=o?l*c:a,w=o?a:l*c,g=m*u*d,y=!0,S=e.kernelCustomData.wT??e.compute(Pe(t[1],af),{inputs:[1],outputs:[r.wIsConst?-2:-1]})[0];r.wIsConst&&!e.kernelCustomData.wT&&(e.kernelCustomData.wT=S);let $=[t[0],S],v=t.length===3;v&&(!o&&t[2].dims.length===1?$.push(t[2].reshape([t[2].dims[0],1,1])):$.push(t[2])),e.compute(Gu($,n,i,h,w,g,v,y),{inputs:$})},uf=(e,t)=>{let r=t.format==="NHWC",n=[e.inputs[0].reshape(r?[e.inputs[0].dims[0],1,e.inputs[0].dims[1],e.inputs[0].dims[2]]:[e.inputs[0].dims[0],e.inputs[0].dims[1],1,e.inputs[0].dims[2]]),e.inputs[1].reshape([e.inputs[1].dims[0],e.inputs[1].dims[1],1,e.inputs[1].dims[2]])];e.inputs.length===3&&n.push(e.inputs[2]);let o=t.kernelShape;(o.length===0||o[0]===0)&&(o=[e.inputs[1].dims[2]]);let i=t.dilations;(i.length===0||i[0]===0)&&(i=[1]);let a=t.strides;(a.length===0||a[0]===0)&&(a=[1]);let d=t.pads;d.length===0&&(d=[0,0]),d=[0,d[0],0,d[1]],a=[1].concat(a),i=[1].concat(i),o=[1].concat(o);let l=ju({...t,pads:d,strides:a,dilations:i,kernelShape:o},n);e.compute(wo(n,l,c=>r?[c[0],c[2],c[3]]:[c[0],c[1],c[3]]))},Yu=(e,t)=>{of(e.inputs,t),e.inputs[0].dims.length===3?uf(e,t):sf(e,e.inputs,t)}});var df,Zu,Qu,Ju=V(()=>{"use strict";Q();ie();Ie();ae();df=(e,t,r,n)=>{let o=k.size(t),i=t.length,a=E("input",e,i),d=M("output",e,i),l=r.dataType===6?r.getInt32Array()[0]:Number(r.getBigInt64Array()[0]),c=k.normalizeAxis(l,i),m=u=>{let h=` i32(${a.indicesGet("inputIndices","uniforms.axis")}) `,w=F("uniforms.input_shape","uniforms.axis",i),g=n.reverse?h+(n.exclusive?" + 1":""):"0",y=n.reverse?w:h+(n.exclusive?"":" + 1");return`
                 ${u.registerUniform("outputSize","u32").registerUniform("axis","u32").declareVariables(a,d)}
                 ${u.mainStart()}
                   ${u.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")}
@@ -5216,9 +5216,9 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
                     sum = sum + ${a.getByIndices("inputIndices")};
                   }
                   ${d.setByOffset("global_idx","sum")};
-                }`};return{name:"CumSum",shaderCache:{hint:n.cacheKey,inputDependencies:["rank"]},getRunData:()=>({outputs:[{dims:t,dataType:e}],dispatchGroup:{x:Math.ceil(o/64)},programUniforms:[{type:12,data:o},{type:12,data:c},...V(t,t)]}),getShaderSource:m}},Zu=(e,t)=>{let r=e.inputs[0].dims,n=e.inputs[0].dataType,o=e.inputs[1];e.compute(df(n,r,o,t),{inputs:[0]})},Qu=e=>{let t=e.exclusive===1,r=e.reverse===1;return J({exclusive:t,reverse:r})}});var lf,cf,pf,ed,td,rd=U(()=>{"use strict";Z();te();Ie();re();lf=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.")},cf=(e,t,r,n)=>{let o=[];o.push(`fn perm(i: ${n.type.indices}) -> ${r.type.indices} {
+                }`};return{name:"CumSum",shaderCache:{hint:n.cacheKey,inputDependencies:["rank"]},getRunData:()=>({outputs:[{dims:t,dataType:e}],dispatchGroup:{x:Math.ceil(o/64)},programUniforms:[{type:12,data:o},{type:12,data:c},...R(t,t)]}),getShaderSource:m}},Zu=(e,t)=>{let r=e.inputs[0].dims,n=e.inputs[0].dataType,o=e.inputs[1];e.compute(df(n,r,o,t),{inputs:[0]})},Qu=e=>{let t=e.exclusive===1,r=e.reverse===1;return J({exclusive:t,reverse:r})}});var lf,cf,pf,ed,td,rd=V(()=>{"use strict";Q();ie();Ie();ae();lf=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.")},cf=(e,t,r,n)=>{let o=[];o.push(`fn perm(i: ${n.type.indices}) -> ${r.type.indices} {
     var a: ${r.type.indices};`);for(let i=0;i<t;++i)o.push(r.indicesSet("a",e[i],`i[${i}]`));return o.push("return a;}"),o.join(`
-`)},pf=(e,t)=>{let r,n,o,i,a,d,l=t.format==="NHWC",c=t.blocksize,m=t.mode==="DCR";l?([r,n,o,i]=e.dims,a=m?[r,n,o,c,c,i/c**2]:[r,n,o,i/c**2,c,c],d=m?[0,1,3,2,4,5]:[0,1,4,2,5,3]):([r,n,o,i]=[e.dims[0],e.dims[2],e.dims[3],e.dims[1]],a=m?[r,c,c,i/c**2,n,o]:[r,i/c**2,c,c,n,o],d=m?[0,3,4,1,5,2]:[0,1,4,2,5,3]);let u=e.reshape(a),h=u.dims.length,w=e.dataType,g=k("a",w,h),y=M("output",w,h),S=$=>`
+`)},pf=(e,t)=>{let r,n,o,i,a,d,l=t.format==="NHWC",c=t.blocksize,m=t.mode==="DCR";l?([r,n,o,i]=e.dims,a=m?[r,n,o,c,c,i/c**2]:[r,n,o,i/c**2,c,c],d=m?[0,1,3,2,4,5]:[0,1,4,2,5,3]):([r,n,o,i]=[e.dims[0],e.dims[2],e.dims[3],e.dims[1]],a=m?[r,c,c,i/c**2,n,o]:[r,i/c**2,c,c,n,o],d=m?[0,3,4,1,5,2]:[0,1,4,2,5,3]);let u=e.reshape(a),h=u.dims.length,w=e.dataType,g=E("a",w,h),y=M("output",w,h),S=$=>`
   ${$.registerUniform("output_size","u32").declareVariables(g,y)}
 
   ${cf(d,h,g,y)}
@@ -5230,7 +5230,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
     let aIndices = perm(indices);
 
     ${y.setByOffset("global_idx",g.getByIndices("aIndices"))}
-  }`;return{name:"DepthToSpace",shaderCache:{hint:`${e.dims};${t.blocksize};${t.mode}`,inputDependencies:["rank"]},getRunData:$=>{let _=l?[r,n*c,o*c,i/c**2]:[r,i/c**2,n*c,o*c],x=E.size(_),T=u.dims,C=E.sortBasedOnPerm(T,d);return{outputs:[{dims:_,dataType:$[0].dataType}],dispatchGroup:{x:Math.ceil(x/64)},programUniforms:[{type:12,data:x},...V(T,C)]}},getShaderSource:S}},ed=(e,t)=>{lf(e.inputs),e.compute(pf(e.inputs[0],t))},td=e=>J({blocksize:e.blocksize,mode:e.mode,format:e.format})});var vo,rn,nd,mf,ff,_o,$o,od,hf,id,ad,sd=U(()=>{"use strict";Z();te();Ie();re();vo="[a-zA-Z]|\\.\\.\\.",rn="("+vo+")+",nd="^"+rn+"$",mf="("+rn+",)*"+rn,ff="^"+mf+"$",_o=class{constructor(t=-1){this.symbolToIndices=new Map,this.inputIndex=t}addSymbol(t,r){let n=this.symbolToIndices.get(t);n===void 0?n=[r]:n.push(r),this.symbolToIndices.set(t,n)}},$o=class{constructor(t,r){this.equation=r;this.hasEllipsis=!1,this.symbolToInfo=new Map,this.lhs=new Array,this.outputDims=[];let[n,o]=r.includes("->")?r.split("->",2):[r,""];if(!n.match(RegExp(ff)))throw new Error("Invalid LHS term");if(n.split(",").forEach((d,l)=>{let c=t[l].dims.slice();if(!d.match(RegExp(nd)))throw new Error("Invalid LHS term");let m=this.processTerm(d,!0,c,l);this.lhs.push(m)}),o==="")o+=[...this.symbolToInfo.entries()].filter(([d,l])=>l.count===1||d==="...").map(([d])=>d).join("");else if(!o.match(RegExp(rn)))throw new Error("Invalid RHS");o.match(RegExp(vo,"g"))?.forEach(d=>{if(d==="...")this.outputDims=this.outputDims.concat(this.ellipsisDims);else{let l=this.symbolToInfo.get(d);if(l===void 0)throw new Error("Invalid RHS symbol");this.outputDims.push(l.dimValue)}}),this.rhs=this.processTerm(o,!1,this.outputDims)}addSymbol(t,r,n){let o=this.symbolToInfo.get(t);if(o!==void 0){if(o.dimValue!==r&&o.count!==1)throw new Error("Dimension mismatch");o.count++,o.inputIndices.push(n)}else o={count:1,dimValue:r,inputIndices:[n]};this.symbolToInfo.set(t,o)}processTerm(t,r,n,o=-1){let i=n.length,a=!1,d=[],l=0;if(!t.match(RegExp(nd))&&!r&&t!=="")throw new Error("Invalid LHS term");let c=t.match(RegExp(vo,"g")),m=new _o(o);return c?.forEach((u,h)=>{if(u==="..."){if(a)throw new Error("Only one ellipsis is allowed per input term");a=!0;let w=i-c.length+1;if(w<0)throw new Error("Ellipsis out of bounds");if(d=n.slice(l,l+w),this.hasEllipsis){if(this.ellipsisDims.length!==d.length||this.ellipsisDims.toString()!==d.toString())throw new Error("Ellipsis dimensions mismatch")}else if(r)this.hasEllipsis=!0,this.ellipsisDims=d;else throw new Error("Ellipsis must be specified in the LHS");for(let g=0;g<d.length;g++){let y=String.fromCharCode("0".charCodeAt(0)+g);m.addSymbol(y,h+g),this.addSymbol(y,n[l++],o)}}else m.addSymbol(u,h+(this.hasEllipsis?this.ellipsisDims.length-1:0)),this.addSymbol(u,n[l++],o)}),m}},od=e=>e+"_max",hf=(e,t,r,n)=>{let i=e.map(m=>m.length).map((m,u)=>k(`input${u}`,t,m)),a=E.size(n),d=M("output",t,n.length),l=[...r.symbolToInfo.keys()].filter(m=>!r.rhs.symbolToIndices.has(m)),c=m=>{let u=[],h="var prod = 1.0;",w="var sum = 0.0;",g="sum += prod;",y=[],S=[],$=[],_=[],x=r.symbolToInfo.size===r.rhs.symbolToIndices.size;r.symbolToInfo.forEach((C,A)=>{if(r.rhs.symbolToIndices.has(A)){let P=r.rhs.symbolToIndices.get(A)?.[0];P!==void 0&&r.lhs.forEach((D,N)=>{if(C.inputIndices.includes(N)){let L=D.symbolToIndices.get(A);if(L===void 0)throw new Error("Invalid symbol error");L.forEach(K=>{u.push(`${i[N].indicesSet(`input${N}Indices`,K,d.indicesGet("outputIndices",P))}`)})}})}else r.lhs.forEach((P,D)=>{if(C.inputIndices.includes(D)){let N=P.symbolToIndices.get(A);if(N===void 0)throw new Error("Invalid symbol error");N.forEach(L=>{y.push(`${i[D].indicesSet(`input${D}Indices`,L,`${A}`)}`)}),_.push(`prod *= ${i[D].getByIndices(`input${D}Indices`)};`)}}),S.push(`for(var ${A}: u32 = 0; ${A} < uniforms.${od(A)}; ${A}++) {`),$.push("}")});let T=x?[...u,`let sum = ${i.map((C,A)=>C.getByIndices(`input${A}Indices`)).join(" * ")};`]:[...u,w,...S,...y,h,..._,g,...$];return`
+  }`;return{name:"DepthToSpace",shaderCache:{hint:`${e.dims};${t.blocksize};${t.mode}`,inputDependencies:["rank"]},getRunData:$=>{let v=l?[r,n*c,o*c,i/c**2]:[r,i/c**2,n*c,o*c],x=k.size(v),T=u.dims,C=k.sortBasedOnPerm(T,d);return{outputs:[{dims:v,dataType:$[0].dataType}],dispatchGroup:{x:Math.ceil(x/64)},programUniforms:[{type:12,data:x},...R(T,C)]}},getShaderSource:S}},ed=(e,t)=>{lf(e.inputs),e.compute(pf(e.inputs[0],t))},td=e=>J({blocksize:e.blocksize,mode:e.mode,format:e.format})});var _o,rn,nd,mf,ff,vo,$o,od,hf,id,ad,sd=V(()=>{"use strict";Q();ie();Ie();ae();_o="[a-zA-Z]|\\.\\.\\.",rn="("+_o+")+",nd="^"+rn+"$",mf="("+rn+",)*"+rn,ff="^"+mf+"$",vo=class{constructor(t=-1){this.symbolToIndices=new Map,this.inputIndex=t}addSymbol(t,r){let n=this.symbolToIndices.get(t);n===void 0?n=[r]:n.push(r),this.symbolToIndices.set(t,n)}},$o=class{constructor(t,r){this.equation=r;this.hasEllipsis=!1,this.symbolToInfo=new Map,this.lhs=new Array,this.outputDims=[];let[n,o]=r.includes("->")?r.split("->",2):[r,""];if(!n.match(RegExp(ff)))throw new Error("Invalid LHS term");if(n.split(",").forEach((d,l)=>{let c=t[l].dims.slice();if(!d.match(RegExp(nd)))throw new Error("Invalid LHS term");let m=this.processTerm(d,!0,c,l);this.lhs.push(m)}),o==="")o+=[...this.symbolToInfo.entries()].filter(([d,l])=>l.count===1||d==="...").map(([d])=>d).join("");else if(!o.match(RegExp(rn)))throw new Error("Invalid RHS");o.match(RegExp(_o,"g"))?.forEach(d=>{if(d==="...")this.outputDims=this.outputDims.concat(this.ellipsisDims);else{let l=this.symbolToInfo.get(d);if(l===void 0)throw new Error("Invalid RHS symbol");this.outputDims.push(l.dimValue)}}),this.rhs=this.processTerm(o,!1,this.outputDims)}addSymbol(t,r,n){let o=this.symbolToInfo.get(t);if(o!==void 0){if(o.dimValue!==r&&o.count!==1)throw new Error("Dimension mismatch");o.count++,o.inputIndices.push(n)}else o={count:1,dimValue:r,inputIndices:[n]};this.symbolToInfo.set(t,o)}processTerm(t,r,n,o=-1){let i=n.length,a=!1,d=[],l=0;if(!t.match(RegExp(nd))&&!r&&t!=="")throw new Error("Invalid LHS term");let c=t.match(RegExp(_o,"g")),m=new vo(o);return c?.forEach((u,h)=>{if(u==="..."){if(a)throw new Error("Only one ellipsis is allowed per input term");a=!0;let w=i-c.length+1;if(w<0)throw new Error("Ellipsis out of bounds");if(d=n.slice(l,l+w),this.hasEllipsis){if(this.ellipsisDims.length!==d.length||this.ellipsisDims.toString()!==d.toString())throw new Error("Ellipsis dimensions mismatch")}else if(r)this.hasEllipsis=!0,this.ellipsisDims=d;else throw new Error("Ellipsis must be specified in the LHS");for(let g=0;g<d.length;g++){let y=String.fromCharCode("0".charCodeAt(0)+g);m.addSymbol(y,h+g),this.addSymbol(y,n[l++],o)}}else m.addSymbol(u,h+(this.hasEllipsis?this.ellipsisDims.length-1:0)),this.addSymbol(u,n[l++],o)}),m}},od=e=>e+"_max",hf=(e,t,r,n)=>{let i=e.map(m=>m.length).map((m,u)=>E(`input${u}`,t,m)),a=k.size(n),d=M("output",t,n.length),l=[...r.symbolToInfo.keys()].filter(m=>!r.rhs.symbolToIndices.has(m)),c=m=>{let u=[],h="var prod = 1.0;",w="var sum = 0.0;",g="sum += prod;",y=[],S=[],$=[],v=[],x=r.symbolToInfo.size===r.rhs.symbolToIndices.size;r.symbolToInfo.forEach((C,A)=>{if(r.rhs.symbolToIndices.has(A)){let P=r.rhs.symbolToIndices.get(A)?.[0];P!==void 0&&r.lhs.forEach((B,N)=>{if(C.inputIndices.includes(N)){let W=B.symbolToIndices.get(A);if(W===void 0)throw new Error("Invalid symbol error");W.forEach(K=>{u.push(`${i[N].indicesSet(`input${N}Indices`,K,d.indicesGet("outputIndices",P))}`)})}})}else r.lhs.forEach((P,B)=>{if(C.inputIndices.includes(B)){let N=P.symbolToIndices.get(A);if(N===void 0)throw new Error("Invalid symbol error");N.forEach(W=>{y.push(`${i[B].indicesSet(`input${B}Indices`,W,`${A}`)}`)}),v.push(`prod *= ${i[B].getByIndices(`input${B}Indices`)};`)}}),S.push(`for(var ${A}: u32 = 0; ${A} < uniforms.${od(A)}; ${A}++) {`),$.push("}")});let T=x?[...u,`let sum = ${i.map((C,A)=>C.getByIndices(`input${A}Indices`)).join(" * ")};`]:[...u,w,...S,...y,h,...v,g,...$];return`
             ${m.registerUniforms(l.map(C=>({name:`${od(C)}`,type:"u32"}))).registerUniform("outputSize","u32").declareVariables(...i,d)}
 
             ${m.mainStart()}
@@ -5241,7 +5241,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
             ${T.join(`
 `)};
             ${d.setByOffset("global_idx","sum")};
-          }`};return{name:"Einsum",shaderCache:{hint:r.equation,inputDependencies:e.map(()=>"rank")},getRunData:()=>{let m=l.filter(h=>r.symbolToInfo.has(h)).map(h=>({type:12,data:r.symbolToInfo.get(h)?.dimValue||0}));m.push({type:12,data:a});let u=e.map((h,w)=>[...V(h)]).reduce((h,w)=>h.concat(w),m);return u.push(...V(n)),{outputs:[{dims:n,dataType:t}],dispatchGroup:{x:Math.ceil(a/64)},programUniforms:u}},getShaderSource:c}},id=(e,t)=>{let r=new $o(e.inputs,t.equation),n=r.outputDims,o=e.inputs.map((i,a)=>i.dims);e.compute(hf(o,e.inputs[0].dataType,r,n))},ad=e=>{let t=e.equation.replace(/\s+/g,"");return J({equation:t})}});var gf,ud,yf,bf,dd,ld=U(()=>{"use strict";Z();te();re();gf=e=>{if(!e||e.length!==2)throw new Error("Expand requires 2 input.");let t=e[0].dims,r=Array.from(e[1].getBigInt64Array(),Number),n=r.length<t.length?0:r.length-t.length,o=t.length<r.length?0:t.length-r.length;for(;n<r.length&&o<t.length;++n,++o)if(r[n]!==t[o]&&r[n]!==1&&t[o]!==1)throw new Error("Expand requires shape to be broadcastable to input")},ud=(e,t)=>{let r=e.length-t.length,n=[];for(let o=0;o<r;++o)n.push(e[o]);for(let o=0;o<t.length;++o)n.push(t[o]===1?e[o+r]:t[o]);return n},yf=(e,t)=>e.length>t.length?ud(e,t):ud(t,e),bf=e=>{let t=e[0].dims,r=Array.from(e[1].getBigInt64Array(),Number),n=yf(t,r),o=e[0].dataType,i=o===9?4:1,a=Math.ceil(E.size(n)/i),d=c=>{let m=k("input",o,t.length,i),u=M("output",o,n.length,i),h;if(o===9){let w=(g,y,S="")=>`
+          }`};return{name:"Einsum",shaderCache:{hint:r.equation,inputDependencies:e.map(()=>"rank")},getRunData:()=>{let m=l.filter(h=>r.symbolToInfo.has(h)).map(h=>({type:12,data:r.symbolToInfo.get(h)?.dimValue||0}));m.push({type:12,data:a});let u=e.map((h,w)=>[...R(h)]).reduce((h,w)=>h.concat(w),m);return u.push(...R(n)),{outputs:[{dims:n,dataType:t}],dispatchGroup:{x:Math.ceil(a/64)},programUniforms:u}},getShaderSource:c}},id=(e,t)=>{let r=new $o(e.inputs,t.equation),n=r.outputDims,o=e.inputs.map((i,a)=>i.dims);e.compute(hf(o,e.inputs[0].dataType,r,n))},ad=e=>{let t=e.equation.replace(/\s+/g,"");return J({equation:t})}});var gf,ud,yf,bf,dd,ld=V(()=>{"use strict";Q();ie();ae();gf=e=>{if(!e||e.length!==2)throw new Error("Expand requires 2 input.");let t=e[0].dims,r=Array.from(e[1].getBigInt64Array(),Number),n=r.length<t.length?0:r.length-t.length,o=t.length<r.length?0:t.length-r.length;for(;n<r.length&&o<t.length;++n,++o)if(r[n]!==t[o]&&r[n]!==1&&t[o]!==1)throw new Error("Expand requires shape to be broadcastable to input")},ud=(e,t)=>{let r=e.length-t.length,n=[];for(let o=0;o<r;++o)n.push(e[o]);for(let o=0;o<t.length;++o)n.push(t[o]===1?e[o+r]:t[o]);return n},yf=(e,t)=>e.length>t.length?ud(e,t):ud(t,e),bf=e=>{let t=e[0].dims,r=Array.from(e[1].getBigInt64Array(),Number),n=yf(t,r),o=e[0].dataType,i=o===9?4:1,a=Math.ceil(k.size(n)/i),d=c=>{let m=E("input",o,t.length,i),u=M("output",o,n.length,i),h;if(o===9){let w=(g,y,S="")=>`
           let outputIndices${y} = ${u.offsetToIndices(`outputOffset + ${y}u`)};
           let offset${y} = ${m.broadcastedIndicesToOffset(`outputIndices${y}`,u)};
           let index${y} = offset${y} / 4u;
@@ -5263,13 +5263,13 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
     ${c.registerUniform("vec_size","u32").declareVariables(m,u)}
     ${c.mainStart()}
     ${c.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.vec_size")}
-    ${h}`},l=[{type:12,data:a},...V(t,n)];return{name:"Expand",shaderCache:{hint:`${n.length}`,inputDependencies:["rank"]},getShaderSource:d,getRunData:()=>({outputs:[{dims:n,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(a/64)},programUniforms:l})}},dd=e=>{gf(e.inputs),e.compute(bf(e.inputs),{inputs:[0]})}});var wf,cd,pd=U(()=>{"use strict";Z();te();re();Xr();wf=e=>{let t=e[0].dataType,r=E.size(e[0].dims),n=E.size(e[1].dims),o=n%4===0,i=a=>{let d=k("x",t,[1],4),l=k("bias",t,[1],4),c=M("y",t,[1],4),m=[{name:"output_vec_size",type:"u32"},{name:"bias_size",type:"u32"}],u=w=>`
+    ${h}`},l=[{type:12,data:a},...R(t,n)];return{name:"Expand",shaderCache:{hint:`${n.length}`,inputDependencies:["rank"]},getShaderSource:d,getRunData:()=>({outputs:[{dims:n,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(a/64)},programUniforms:l})}},dd=e=>{gf(e.inputs),e.compute(bf(e.inputs),{inputs:[0]})}});var wf,cd,pd=V(()=>{"use strict";Q();ie();ae();Xr();wf=e=>{let t=e[0].dataType,r=k.size(e[0].dims),n=k.size(e[1].dims),o=n%4===0,i=a=>{let d=E("x",t,[1],4),l=E("bias",t,[1],4),c=M("y",t,[1],4),m=[{name:"output_vec_size",type:"u32"},{name:"bias_size",type:"u32"}],u=w=>`
       let bias${w}_offset: u32 = (global_idx * 4 + ${w}) % uniforms.bias_size;
       let bias${w} = ${l.getByOffset(`bias${w}_offset / 4`)}[bias${w}_offset % 4];`,h=o?`
       let bias = ${l.getByOffset("global_idx % (uniforms.bias_size / 4)")};`:`${u(0)}${u(1)}${u(2)}${u(3)}
       let bias = ${d.type.value}(bias0, bias1, bias2, bias3);`;return`${a.registerUniforms(m).declareVariables(d,l,c)}
 
-    ${lo(ke(t))}
+    ${lo(Ee(t))}
 
     ${a.mainStart(At)}
       ${a.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_vec_size")}
@@ -5278,13 +5278,13 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
       ${h}
       let x_in = x + bias;
       ${c.setByOffset("global_idx",co("x_in"))}
-    }`};return{name:"FastGeluWithBias",shaderCache:{hint:`${o}`,inputDependencies:["type","type"]},getShaderSource:i,getRunData:a=>({outputs:[{dims:a[0].dims,dataType:a[0].dataType}],programUniforms:[{type:12,data:Math.ceil(r/4)},{type:12,data:n}],dispatchGroup:{x:Math.ceil(r/At/4)}})}},cd=e=>{e.inputs.length<2||E.size(e.inputs[1].dims)===0?du(e):e.compute(wf(e.inputs))}});var vf,_f,md,fd,hd=U(()=>{"use strict";Z();te();Ie();re();vf=e=>{if(!e||e.length!==2)throw new Error("Gather requires 2 inputs.")},_f=(e,t)=>{let r=e[0].dims,n=e[1].dims,o=r.length,i=E.normalizeAxis(t.axis,o),a=r.slice(0);a.splice(i,1,...n);let d=r[i],l=e[0].dataType===9?4:1,c=Math.ceil(E.size(a)/l),m=[{type:12,data:c},{type:6,data:d},{type:12,data:i},...V(e[0].dims,e[1].dims,a)],u=h=>{let w=k("data",e[0].dataType,e[0].dims.length,l),g=k("inputIndices",e[1].dataType,e[1].dims.length),y=M("output",e[0].dataType,a.length,l),S=_=>{let x=n.length,T=`var indicesIndices${_}  = ${g.type.indices}(0);`;for(let C=0;C<x;C++)T+=`${x>1?`indicesIndices${_}[${C}]`:`indicesIndices${_}`} = ${a.length>1?`outputIndices${_}[uniforms.axis + ${C}]`:`outputIndices${_}`};`;T+=`
-          var idx${_} = ${g.getByIndices(`indicesIndices${_}`)};
-          if (idx${_} < 0) {
-            idx${_} = idx${_} + uniforms.axisDimLimit;
+    }`};return{name:"FastGeluWithBias",shaderCache:{hint:`${o}`,inputDependencies:["type","type"]},getShaderSource:i,getRunData:a=>({outputs:[{dims:a[0].dims,dataType:a[0].dataType}],programUniforms:[{type:12,data:Math.ceil(r/4)},{type:12,data:n}],dispatchGroup:{x:Math.ceil(r/At/4)}})}},cd=e=>{e.inputs.length<2||k.size(e.inputs[1].dims)===0?du(e):e.compute(wf(e.inputs))}});var _f,vf,md,fd,hd=V(()=>{"use strict";Q();ie();Ie();ae();_f=e=>{if(!e||e.length!==2)throw new Error("Gather requires 2 inputs.")},vf=(e,t)=>{let r=e[0].dims,n=e[1].dims,o=r.length,i=k.normalizeAxis(t.axis,o),a=r.slice(0);a.splice(i,1,...n);let d=r[i],l=e[0].dataType===9?4:1,c=Math.ceil(k.size(a)/l),m=[{type:12,data:c},{type:6,data:d},{type:12,data:i},...R(e[0].dims,e[1].dims,a)],u=h=>{let w=E("data",e[0].dataType,e[0].dims.length,l),g=E("inputIndices",e[1].dataType,e[1].dims.length),y=M("output",e[0].dataType,a.length,l),S=v=>{let x=n.length,T=`var indicesIndices${v}  = ${g.type.indices}(0);`;for(let C=0;C<x;C++)T+=`${x>1?`indicesIndices${v}[${C}]`:`indicesIndices${v}`} = ${a.length>1?`outputIndices${v}[uniforms.axis + ${C}]`:`outputIndices${v}`};`;T+=`
+          var idx${v} = ${g.getByIndices(`indicesIndices${v}`)};
+          if (idx${v} < 0) {
+            idx${v} = idx${v} + uniforms.axisDimLimit;
           }
-          var dataIndices${_} : ${w.type.indices};
-        `;for(let C=0,A=0;C<o;C++)C===i?(T+=`${o>1?`dataIndices${_}[${C}]`:`dataIndices${_}`} = u32(idx${_});`,A+=x):(T+=`${o>1?`dataIndices${_}[${C}]`:`dataIndices${_}`} = ${a.length>1?`outputIndices${_}[${A}]`:`outputIndices${_}`};`,A++);return T},$;if(e[0].dataType===9){let _=(x,T,C="")=>`
+          var dataIndices${v} : ${w.type.indices};
+        `;for(let C=0,A=0;C<o;C++)C===i?(T+=`${o>1?`dataIndices${v}[${C}]`:`dataIndices${v}`} = u32(idx${v});`,A+=x):(T+=`${o>1?`dataIndices${v}[${C}]`:`dataIndices${v}`} = ${a.length>1?`outputIndices${v}[${A}]`:`outputIndices${v}`};`,A++);return T},$;if(e[0].dataType===9){let v=(x,T,C="")=>`
           let outputIndices${T} = ${y.offsetToIndices(`outputOffset + ${T}u`)};
           ${S(T)};
           let offset${T} = ${w.indicesToOffset(`dataIndices${T}`)};
@@ -5294,10 +5294,10 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
         `;$=`
         let outputOffset = global_idx * ${l};
         var value = vec4<u32>(0);
-        ${_("value",0,"u32")}
-        ${_("value",1,"u32")}
-        ${_("value",2,"u32")}
-        ${_("value",3,"u32")}
+        ${v("value",0,"u32")}
+        ${v("value",1,"u32")}
+        ${v("value",2,"u32")}
+        ${v("value",3,"u32")}
         ${y.setByOffset("global_idx","value")}
       `}else $=`
       let outputIndices = ${y.offsetToIndices("global_idx")};
@@ -5309,7 +5309,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
       ${h.mainStart()}
         ${h.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")}
         ${$}
-      }`};return{name:"Gather",shaderCache:{hint:t.cacheKey,inputDependencies:["rank","rank"]},getRunData:()=>({outputs:[{dims:a,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(c/64)},programUniforms:m}),getShaderSource:u}},md=e=>J({axis:e.axis}),fd=(e,t)=>{let r=e.inputs;vf(r),e.compute(_f(e.inputs,t))}});var $f,xf,gd,yd,bd=U(()=>{"use strict";Z();te();Ie();re();$f=(e,t)=>{if(e.length<3||e.length>4)throw new Error("GatherBlockQuantized requires 3 or 4 inputs.");let r=E.normalizeAxis(t.quantizeAxis,e[0].dims.length),n=t.blockSize,o=e[0],i=e[2],a=e.length===4?e[3]:void 0;if(i.dims.length!==o.dims.length||!o.dims.map((d,l)=>l===r?Math.ceil(d/n)===i.dims[l]:d===i.dims[l]).reduce((d,l)=>d&&l,!0))throw new Error("Scales must have the same rank as the input tensor and the dims should match except on gatherAxis.");if(a){if(a.dataType!==o.dataType)throw new Error("Zero point must have the same data type as the input tensor.");if(a.dims.length!==i.dims.length||!a.dims.map((d,l)=>d===i.dims[l]).reduce((d,l)=>d&&l,!0))throw new Error("Zero point must have the same rank as the input tensor and the dims should match except on quantizeAxis.")}},xf=(e,t)=>{let r=e[0].dims,n=e[1].dims,o=r.length,i=E.normalizeAxis(t.gatherAxis,o),a=E.normalizeAxis(t.quantizeAxis,o),d=r.slice(0);d.splice(i,1,...n);let l=E.size(d),c=e[2].dataType,u=e[0].dataType===22,h=[{type:12,data:l},{type:12,data:a},{type:12,data:i},{type:12,data:t.blockSize},...V(...e.map((g,y)=>g.dims),d)],w=g=>{let y=k("data",e[0].dataType,e[0].dims.length),S=k("inputIndices",e[1].dataType,e[1].dims.length),$=k("scales",e[2].dataType,e[2].dims.length),_=e.length>3?k("zeroPoint",e[3].dataType,e[3].dims.length):void 0,x=M("output",c,d.length),T=[y,S,$];_&&T.push(_);let C=[{name:"output_size",type:"u32"},{name:"quantize_axis",type:"u32"},{name:"gather_axis",type:"u32"},{name:"block_size",type:"u32"}];return`
+      }`};return{name:"Gather",shaderCache:{hint:t.cacheKey,inputDependencies:["rank","rank"]},getRunData:()=>({outputs:[{dims:a,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(c/64)},programUniforms:m}),getShaderSource:u}},md=e=>J({axis:e.axis}),fd=(e,t)=>{let r=e.inputs;_f(r),e.compute(vf(e.inputs,t))}});var $f,xf,gd,yd,bd=V(()=>{"use strict";Q();ie();Ie();ae();$f=(e,t)=>{if(e.length<3||e.length>4)throw new Error("GatherBlockQuantized requires 3 or 4 inputs.");let r=k.normalizeAxis(t.quantizeAxis,e[0].dims.length),n=t.blockSize,o=e[0],i=e[2],a=e.length===4?e[3]:void 0;if(i.dims.length!==o.dims.length||!o.dims.map((d,l)=>l===r?Math.ceil(d/n)===i.dims[l]:d===i.dims[l]).reduce((d,l)=>d&&l,!0))throw new Error("Scales must have the same rank as the input tensor and the dims should match except on gatherAxis.");if(a){if(a.dataType!==o.dataType)throw new Error("Zero point must have the same data type as the input tensor.");if(a.dims.length!==i.dims.length||!a.dims.map((d,l)=>d===i.dims[l]).reduce((d,l)=>d&&l,!0))throw new Error("Zero point must have the same rank as the input tensor and the dims should match except on quantizeAxis.")}},xf=(e,t)=>{let r=e[0].dims,n=e[1].dims,o=r.length,i=k.normalizeAxis(t.gatherAxis,o),a=k.normalizeAxis(t.quantizeAxis,o),d=r.slice(0);d.splice(i,1,...n);let l=k.size(d),c=e[2].dataType,u=e[0].dataType===22,h=[{type:12,data:l},{type:12,data:a},{type:12,data:i},{type:12,data:t.blockSize},...R(...e.map((g,y)=>g.dims),d)],w=g=>{let y=E("data",e[0].dataType,e[0].dims.length),S=E("inputIndices",e[1].dataType,e[1].dims.length),$=E("scales",e[2].dataType,e[2].dims.length),v=e.length>3?E("zeroPoint",e[3].dataType,e[3].dims.length):void 0,x=M("output",c,d.length),T=[y,S,$];v&&T.push(v);let C=[{name:"output_size",type:"u32"},{name:"quantize_axis",type:"u32"},{name:"gather_axis",type:"u32"},{name:"block_size",type:"u32"}];return`
         ${g.registerUniforms(C).declareVariables(...T,x)}
         ${g.mainStart()}
         let output_indices = ${x.offsetToIndices("global_idx")};
@@ -5344,18 +5344,18 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
         let quantize_axis_index = ${$.indicesGet("data_indices","uniforms.quantize_axis")} / uniforms.block_size;
         ${$.indicesSet("scale_indices","uniforms.quantize_axis","quantize_axis_index")};
         var scale = ${$.getByIndices("scale_indices")};
-        ${(()=>_?`
+        ${(()=>v?`
               let zero_point_indices = scale_indices;
-              let zero_point_offset = ${_.indicesToOffset("zero_point_indices")};
+              let zero_point_offset = ${v.indicesToOffset("zero_point_indices")};
               let zero_point_index = zero_point_offset % 8;
-              let packed_4bit_zero_points = ${_.getByOffset("zero_point_offset / 8")};
+              let packed_4bit_zero_points = ${v.getByOffset("zero_point_offset / 8")};
               let packed_8bit_zero_points = (packed_4bit_zero_points >> (4 * (zero_point_index % 2))) & 0x0f0f0f0f;
               let zero_point_vec = ${u?"unpack4xI8":"unpack4xU8"}(u32(packed_8bit_zero_points));
               let zero_point = zero_point_vec[zero_point_index / 2];`:"var zero_point = 0")()};
-        let dequantized_data = ${ke(c)}(quantized_data - zero_point) * scale;
+        let dequantized_data = ${Ee(c)}(quantized_data - zero_point) * scale;
         ${x.setByOffset("global_idx","dequantized_data")};
-    }`};return{name:"GatherBlockQuantized",shaderCache:{hint:`${t.cacheKey};${e.filter((g,y)=>y!==1).map(g=>g.dims.join("_")).join(";")}`,inputDependencies:Array.from({length:e.length},(g,y)=>"rank")},getRunData:()=>({outputs:[{dims:d,dataType:c}],dispatchGroup:{x:Math.ceil(l/64)},programUniforms:h}),getShaderSource:w}},gd=(e,t)=>{let r=e.inputs;$f(r,t),e.compute(xf(e.inputs,t))},yd=e=>J({blockSize:e.blockSize,gatherAxis:e.gatherAxis,quantizeAxis:e.quantizeAxis})});var Sf,Tf,wd,vd,_d=U(()=>{"use strict";Z();te();Ie();re();Sf=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.`)},Tf=(e,t)=>{let r=e[0].dims,n=e[0].dataType,o=r.length,i=e[1].dims,a=e[1].dataType,d=E.normalizeAxis(t.axis,o),l=r[d],c=i.slice(0),m=E.size(c),u=k("input",n,o),h=k("indicesInput",a,i.length),w=M("output",n,c.length),g=[{type:12,data:m},{type:6,data:l},{type:12,data:d}];return g.push(...V(r,i,c)),{name:"GatherElements",shaderCache:{inputDependencies:["rank","rank"]},getRunData:()=>({outputs:[{dims:c,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(m/64)},programUniforms:g}),getShaderSource:$=>`
+    }`};return{name:"GatherBlockQuantized",shaderCache:{hint:`${t.cacheKey};${e.filter((g,y)=>y!==1).map(g=>g.dims.join("_")).join(";")}`,inputDependencies:Array.from({length:e.length},(g,y)=>"rank")},getRunData:()=>({outputs:[{dims:d,dataType:c}],dispatchGroup:{x:Math.ceil(l/64)},programUniforms:h}),getShaderSource:w}},gd=(e,t)=>{let r=e.inputs;$f(r,t),e.compute(xf(e.inputs,t))},yd=e=>J({blockSize:e.blockSize,gatherAxis:e.gatherAxis,quantizeAxis:e.quantizeAxis})});var Sf,Tf,wd,_d,vd=V(()=>{"use strict";Q();ie();Ie();ae();Sf=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.`)},Tf=(e,t)=>{let r=e[0].dims,n=e[0].dataType,o=r.length,i=e[1].dims,a=e[1].dataType,d=k.normalizeAxis(t.axis,o),l=r[d],c=i.slice(0),m=k.size(c),u=E("input",n,o),h=E("indicesInput",a,i.length),w=M("output",n,c.length),g=[{type:12,data:m},{type:6,data:l},{type:12,data:d}];return g.push(...R(r,i,c)),{name:"GatherElements",shaderCache:{inputDependencies:["rank","rank"]},getRunData:()=>({outputs:[{dims:c,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(m/64)},programUniforms:g}),getShaderSource:$=>`
       ${$.registerUniform("outputSize","u32").registerUniform("axisDimLimit","i32").registerUniform("axis","u32").declareVariables(u,h,w)}
       ${$.mainStart()}
       ${$.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")}
@@ -5371,7 +5371,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
       let value = ${u.getByIndices("inputIndices")};
 
       ${w.setByOffset("global_idx","value")};
-  }`}},wd=e=>J({axis:e.axis}),vd=(e,t)=>{let r=e.inputs;Sf(r),e.compute(Tf(e.inputs,t))}});var If,Cf,$d,xd,Sd=U(()=>{"use strict";Z();te();re();If=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")},Cf=(e,t)=>{let r=e[0].dims.slice(),n=e[1].dims.slice(),[o,i,a]=Gr.getShapeOfGemmResult(r,t.transA,n,t.transB,e.length===3?e[2].dims:void 0),d=[o,i];if(!d)throw new Error("Can't use gemm on the given tensors");let l=E.size(d),c=[{type:12,data:l},{type:12,data:o},{type:12,data:i},{type:12,data:a},{type:1,data:t.alpha},{type:1,data:t.beta}],m=["type","type"];e.length===3&&(c.push(...V(e[2].dims)),m.push("rank")),c.push(...V(d));let u=h=>{let w="";t.transA&&t.transB?w="value += a[k * uniforms.M + m] * b[n * uniforms.K + k];":t.transA&&!t.transB?w="value += a[k * uniforms.M + m] * b[k * uniforms.N + n];":!t.transA&&t.transB?w="value += a[m * uniforms.K + k] * b[n * uniforms.K + k];":!t.transA&&!t.transB&&(w="value += a[m * uniforms.K + k] * b[k * uniforms.N + n];");let g=t.alpha===1?"":"value *= uniforms.alpha;",y=k("a",e[0].dataType,e[0].dims),S=k("b",e[1].dataType,e[1].dims),$=y.type.value,_=null,x=[y,S];e.length===3&&(_=k("c",e[2].dataType,e[2].dims.length),x.push(_));let T=M("output",e[0].dataType,d.length);x.push(T);let C=[{name:"output_size",type:"u32"},{name:"M",type:"u32"},{name:"N",type:"u32"},{name:"K",type:"u32"},{name:"alpha",type:"f32"},{name:"beta",type:"f32"}];return`
+  }`}},wd=e=>J({axis:e.axis}),_d=(e,t)=>{let r=e.inputs;Sf(r),e.compute(Tf(e.inputs,t))}});var If,Cf,$d,xd,Sd=V(()=>{"use strict";Q();ie();ae();If=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")},Cf=(e,t)=>{let r=e[0].dims.slice(),n=e[1].dims.slice(),[o,i,a]=Gr.getShapeOfGemmResult(r,t.transA,n,t.transB,e.length===3?e[2].dims:void 0),d=[o,i];if(!d)throw new Error("Can't use gemm on the given tensors");let l=k.size(d),c=[{type:12,data:l},{type:12,data:o},{type:12,data:i},{type:12,data:a},{type:1,data:t.alpha},{type:1,data:t.beta}],m=["type","type"];e.length===3&&(c.push(...R(e[2].dims)),m.push("rank")),c.push(...R(d));let u=h=>{let w="";t.transA&&t.transB?w="value += a[k * uniforms.M + m] * b[n * uniforms.K + k];":t.transA&&!t.transB?w="value += a[k * uniforms.M + m] * b[k * uniforms.N + n];":!t.transA&&t.transB?w="value += a[m * uniforms.K + k] * b[n * uniforms.K + k];":!t.transA&&!t.transB&&(w="value += a[m * uniforms.K + k] * b[k * uniforms.N + n];");let g=t.alpha===1?"":"value *= uniforms.alpha;",y=E("a",e[0].dataType,e[0].dims),S=E("b",e[1].dataType,e[1].dims),$=y.type.value,v=null,x=[y,S];e.length===3&&(v=E("c",e[2].dataType,e[2].dims.length),x.push(v));let T=M("output",e[0].dataType,d.length);x.push(T);let C=[{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`
   ${h.registerUniforms(C).declareVariables(...x)}
 
   ${h.mainStart()}
@@ -5386,16 +5386,16 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
     }
 
     ${g}
-    ${(()=>_!=null?`let cOffset = ${_.broadcastedIndicesToOffset("vec2(m, n)",T)}; value += ${$}(uniforms.beta) * ${_.getByOffset("cOffset")};`:"")()}
+    ${(()=>v!=null?`let cOffset = ${v.broadcastedIndicesToOffset("vec2(m, n)",T)}; value += ${$}(uniforms.beta) * ${v.getByOffset("cOffset")};`:"")()}
     output[global_idx] = value;
-  }`};return{name:"Gemm",shaderCache:{hint:`${t.cacheKey}`,inputDependencies:m},getRunData:()=>({outputs:[{dims:d,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(l/64)},programUniforms:c}),getShaderSource:u}},$d=e=>{let t=e.transA,r=e.transB,n=e.alpha,o=e.beta;return{transA:t,transB:r,alpha:n,beta:o,cacheKey:`${e.transA};${e.transB};${e.alpha===1}`}},xd=(e,t)=>{If(e.inputs),e.compute(Cf(e.inputs,t))}});var Ne,kf,Id,Td,Pf,nr,Cd,xo=U(()=>{"use strict";Z();te();Ie();Hr();Kr();re();lt();Ne=(e,t)=>e.length>t&&e[t].dims.length>0?e[t]:void 0,kf=(e,t)=>{let r=e[0],n=Ne(e,1),o=Ne(e,2),i=Ne(e,3),a=Ne(e,4),d=Ne(e,5),l=Ne(e,6),c=Ne(e,7);if(r.dims.length!==3&&r.dims.length!==5)throw new Error("Input query is expected to have 3 or 5 dimensions");let m=r.dims[0],u=r.dims[1],h=r.dims.length===3?r.dims[2]:t.numHeads*r.dims[4],w=u,g=0,y=0,S=Math.floor(h/t.numHeads);if(l&&c&&E.size(l.dims)&&E.size(c.dims)){if(l.dims.length!==4)throw new Error('Input "past_key" is expected to have 4 dimensions');if(l.dims[0]!==m||l.dims[1]!==t.numHeads||l.dims[3]!==S)throw new Error('Input "past_key" shape (batch_size, num_heads, past_sequence_length, head_size)');if(c.dims[0]!==m||c.dims[1]!==t.numHeads||c.dims[3]!==S)throw new Error('Input "past_value" shape (batch_size, num_heads, past_sequence_length, head_size)');if(l.dims[2]!==c.dims[2])throw new Error('Input "past_key" and "past_value" shall have same dim 2 (past_sequence_length)');if(c.dims.length!==4)throw new Error('Input "past_value" is expected to have 4 dimensions');g=l.dims[2],y=l.dims[2]}else if(l&&E.size(l.dims)||c&&E.size(c.dims))throw new Error('Input "past_key" and "past_value" shall be both present or both absent');let $;if(n&&E.size(n.dims)>0){if(r.dims.length!==3)throw new Error('Input "query" is expected to have 3 dimensions when key is given');if(n.dims.length<3||n.dims.length>5)throw new Error('Input "key" is expected to have 3, 4, or 5 dimensions');if(r.dims[0]!==n.dims[0])throw new Error('Input "query" and "key" shall have same dim 0 (batch size)');if(n.dims.length===3){if(n.dims[2]!==r.dims[2])throw new Error('Input "query" and "key" shall have same dim 2 (hidden_size)');$=2,w=n.dims[1]}else if(n.dims.length===5){if(n.dims[2]!==t.numHeads||n.dims[3]!==2||n.dims[4]!==S)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.');$=5,w=n.dims[1]}else{if(n.dims[1]!==t.numHeads||n.dims[3]!==S)throw new Error('Expect "key" shape (batch_size, num_heads, kv_sequence_length, head_size) for past_key');$=0,w=n.dims[2]}}else{if(r.dims.length!==5)throw new Error('Input "query" is expected to have 5 dimensions when key is empty');if(r.dims[2]!==t.numHeads||r.dims[3]!==3)throw new Error('Expect "query" shape (batch_size, kv_sequence_length, num_heads, 3, head_size) for packed kv');$=3}if(i&&E.size(i.dims)>0){if(i.dims.length!==1)throw new Error('Input "bias" is expected to have 1 dimension');if(n&&n.dims.length===5&&n.dims[3]===2)throw new Error("bias is not allowed for packed kv.")}let _=g+w,x=0;if(a&&E.size(a.dims)>0){x=8;let P=a.dims;throw P.length===1?P[0]===m?x=1:P[0]===3*m+2&&(x=3):P.length===2&&P[0]===m&&P[1]===_&&(x=5),x===8?new Error('Input "key_padding_mask" shape shall be (batch_size) or (batch_size, total_sequence_length)'):new Error("Mask not supported")}let T=!1,C=h;if(o&&E.size(o.dims)>0){if(o.dims.length!==3&&o.dims.length!==4)throw new Error('Input "value" is expected to have 3 or 4 dimensions');if(r.dims[0]!==o.dims[0])throw new Error('Input "query" and "value" shall have same dim 0 (batch_size)');if(o.dims.length===3){if(w!==o.dims[1])throw new Error('Input "key" and "value" shall have the same dim 1 (kv_sequence_length)');C=o.dims[2]}else{if(w!==o.dims[2])throw new Error('Input "key" and "value" shall have the same dim 2 (kv_sequence_length)');C=o.dims[1]*o.dims[3],T=!0}}let A=!1;if(a&&E.size(a.dims)>0)throw new Error("Key padding mask is not supported");if(d&&E.size(d.dims)>0){if(d.dims.length!==4)throw new Error('Input "attention_bias" is expected to have 4 dimensions');if(d.dims[0]!==m||d.dims[1]!==t.numHeads||d.dims[2]!==u||d.dims[3]!==_)throw new Error('Expect "attention_bias" shape (batch_size, num_heads, sequence_length, total_sequence_length)')}return{batchSize:m,sequenceLength:u,pastSequenceLength:g,kvSequenceLength:w,totalSequenceLength:_,maxSequenceLength:y,inputHiddenSize:0,hiddenSize:h,vHiddenSize:C,headSize:S,vHeadSize:Math.floor(C/t.numHeads),numHeads:t.numHeads,isUnidirectional:!1,pastPresentShareBuffer:!1,maskFilterValue:t.maskFilterValue,maskType:x,scale:t.scale,broadcastResPosBias:A,passPastInKv:T,qkvFormat:$}},Id=e=>J({...e}),Td=J({perm:[0,2,1,3]}),Pf=(e,t,r,n,o,i,a)=>{let d=[n,o,i],l=E.size(d),c=[{type:12,data:l},{type:12,data:a},{type:12,data:i}],m=u=>{let h=M("qkv_with_bias",t.dataType,d),w=k("qkv",t.dataType,d),g=k("bias",r.dataType,d),y=[{name:"output_size",type:"u32"},{name:"bias_offset",type:"u32"},{name:"hidden_size",type:"u32"}];return`
+  }`};return{name:"Gemm",shaderCache:{hint:`${t.cacheKey}`,inputDependencies:m},getRunData:()=>({outputs:[{dims:d,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(l/64)},programUniforms:c}),getShaderSource:u}},$d=e=>{let t=e.transA,r=e.transB,n=e.alpha,o=e.beta;return{transA:t,transB:r,alpha:n,beta:o,cacheKey:`${e.transA};${e.transB};${e.alpha===1}`}},xd=(e,t)=>{If(e.inputs),e.compute(Cf(e.inputs,t))}});var Ne,Ef,Id,Td,Pf,nr,Cd,xo=V(()=>{"use strict";Q();ie();Ie();Hr();Kr();ae();lt();Ne=(e,t)=>e.length>t&&e[t].dims.length>0?e[t]:void 0,Ef=(e,t)=>{let r=e[0],n=Ne(e,1),o=Ne(e,2),i=Ne(e,3),a=Ne(e,4),d=Ne(e,5),l=Ne(e,6),c=Ne(e,7);if(r.dims.length!==3&&r.dims.length!==5)throw new Error("Input query is expected to have 3 or 5 dimensions");let m=r.dims[0],u=r.dims[1],h=r.dims.length===3?r.dims[2]:t.numHeads*r.dims[4],w=u,g=0,y=0,S=Math.floor(h/t.numHeads);if(l&&c&&k.size(l.dims)&&k.size(c.dims)){if(l.dims.length!==4)throw new Error('Input "past_key" is expected to have 4 dimensions');if(l.dims[0]!==m||l.dims[1]!==t.numHeads||l.dims[3]!==S)throw new Error('Input "past_key" shape (batch_size, num_heads, past_sequence_length, head_size)');if(c.dims[0]!==m||c.dims[1]!==t.numHeads||c.dims[3]!==S)throw new Error('Input "past_value" shape (batch_size, num_heads, past_sequence_length, head_size)');if(l.dims[2]!==c.dims[2])throw new Error('Input "past_key" and "past_value" shall have same dim 2 (past_sequence_length)');if(c.dims.length!==4)throw new Error('Input "past_value" is expected to have 4 dimensions');g=l.dims[2],y=l.dims[2]}else if(l&&k.size(l.dims)||c&&k.size(c.dims))throw new Error('Input "past_key" and "past_value" shall be both present or both absent');let $;if(n&&k.size(n.dims)>0){if(r.dims.length!==3)throw new Error('Input "query" is expected to have 3 dimensions when key is given');if(n.dims.length<3||n.dims.length>5)throw new Error('Input "key" is expected to have 3, 4, or 5 dimensions');if(r.dims[0]!==n.dims[0])throw new Error('Input "query" and "key" shall have same dim 0 (batch size)');if(n.dims.length===3){if(n.dims[2]!==r.dims[2])throw new Error('Input "query" and "key" shall have same dim 2 (hidden_size)');$=2,w=n.dims[1]}else if(n.dims.length===5){if(n.dims[2]!==t.numHeads||n.dims[3]!==2||n.dims[4]!==S)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.');$=5,w=n.dims[1]}else{if(n.dims[1]!==t.numHeads||n.dims[3]!==S)throw new Error('Expect "key" shape (batch_size, num_heads, kv_sequence_length, head_size) for past_key');$=0,w=n.dims[2]}}else{if(r.dims.length!==5)throw new Error('Input "query" is expected to have 5 dimensions when key is empty');if(r.dims[2]!==t.numHeads||r.dims[3]!==3)throw new Error('Expect "query" shape (batch_size, kv_sequence_length, num_heads, 3, head_size) for packed kv');$=3}if(i&&k.size(i.dims)>0){if(i.dims.length!==1)throw new Error('Input "bias" is expected to have 1 dimension');if(n&&n.dims.length===5&&n.dims[3]===2)throw new Error("bias is not allowed for packed kv.")}let v=g+w,x=0;if(a&&k.size(a.dims)>0){x=8;let P=a.dims;throw P.length===1?P[0]===m?x=1:P[0]===3*m+2&&(x=3):P.length===2&&P[0]===m&&P[1]===v&&(x=5),x===8?new Error('Input "key_padding_mask" shape shall be (batch_size) or (batch_size, total_sequence_length)'):new Error("Mask not supported")}let T=!1,C=h;if(o&&k.size(o.dims)>0){if(o.dims.length!==3&&o.dims.length!==4)throw new Error('Input "value" is expected to have 3 or 4 dimensions');if(r.dims[0]!==o.dims[0])throw new Error('Input "query" and "value" shall have same dim 0 (batch_size)');if(o.dims.length===3){if(w!==o.dims[1])throw new Error('Input "key" and "value" shall have the same dim 1 (kv_sequence_length)');C=o.dims[2]}else{if(w!==o.dims[2])throw new Error('Input "key" and "value" shall have the same dim 2 (kv_sequence_length)');C=o.dims[1]*o.dims[3],T=!0}}let A=!1;if(a&&k.size(a.dims)>0)throw new Error("Key padding mask is not supported");if(d&&k.size(d.dims)>0){if(d.dims.length!==4)throw new Error('Input "attention_bias" is expected to have 4 dimensions');if(d.dims[0]!==m||d.dims[1]!==t.numHeads||d.dims[2]!==u||d.dims[3]!==v)throw new Error('Expect "attention_bias" shape (batch_size, num_heads, sequence_length, total_sequence_length)')}return{batchSize:m,sequenceLength:u,pastSequenceLength:g,kvSequenceLength:w,totalSequenceLength:v,maxSequenceLength:y,inputHiddenSize:0,hiddenSize:h,vHiddenSize:C,headSize:S,vHeadSize:Math.floor(C/t.numHeads),numHeads:t.numHeads,isUnidirectional:!1,pastPresentShareBuffer:!1,maskFilterValue:t.maskFilterValue,maskType:x,scale:t.scale,broadcastResPosBias:A,passPastInKv:T,qkvFormat:$}},Id=e=>J({...e}),Td=J({perm:[0,2,1,3]}),Pf=(e,t,r,n,o,i,a)=>{let d=[n,o,i],l=k.size(d),c=[{type:12,data:l},{type:12,data:a},{type:12,data:i}],m=u=>{let h=M("qkv_with_bias",t.dataType,d),w=E("qkv",t.dataType,d),g=E("bias",r.dataType,d),y=[{name:"output_size",type:"u32"},{name:"bias_offset",type:"u32"},{name:"hidden_size",type:"u32"}];return`
   ${u.registerUniforms(y).declareVariables(w,g,h)}
   ${u.mainStart()}
     ${u.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:t.dataType,gpuDataType:0}],dispatchGroup:{x:Math.ceil(l/64)},programUniforms:c}),getShaderSource:m},{inputs:[t,r],outputs:[-1]})[0]},nr=(e,t,r,n,o,i,a,d)=>{let l=i;if(a&&E.size(a.dims)>0){if(n===1)throw new Error("AddBiasReshape is not implemented. Please export your model with packed QKV or KV");return l=Pf(e,i,a,t,n,r*o,d),l=l.reshape([t,n,r,o]),e.compute(Pe(l,Td.perm),{inputs:[l],outputs:[-1]})[0]}else return i.dims.length===3&&(l=i.reshape([t,n,r,o])),e.compute(Pe(l,Td.perm),{inputs:[l],outputs:[-1]})[0]},Cd=(e,t)=>{let r=kf(e.inputs,t),n=e.inputs[0],o=Ne(e.inputs,1),i=Ne(e.inputs,2),a=Ne(e.inputs,3),d=Ne(e.inputs,4),l=Ne(e.inputs,5),c=Ne(e.inputs,6),m=Ne(e.inputs,7);if(n.dims.length===5)throw new Error("Packed QKV is not implemented");if(o?.dims.length===5)throw new Error("Packed KV is not implemented");let u=o&&i&&o.dims.length===4&&i.dims.length===4,h=nr(e,r.batchSize,r.numHeads,r.sequenceLength,r.headSize,n,a,0);if(u)return Wt(e,h,o,i,d,void 0,c,m,l,r,t);if(!o||!i)throw new Error("key and value must be provided");let w=nr(e,r.batchSize,r.numHeads,r.kvSequenceLength,r.headSize,o,a,r.hiddenSize),g=nr(e,r.batchSize,r.numHeads,r.kvSequenceLength,r.vHeadSize,i,a,2*r.hiddenSize);Wt(e,h,w,g,d,void 0,c,m,l,r,t)}});var Ad,Of,zf,So,Ed,To=U(()=>{"use strict";Z();te();re();Ad=e=>Array.from(e.getBigInt64Array(),Number),Of=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(Ad(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")},zf=(e,t)=>{let r=[];for(let n=0;n<e.length;++n)r.push(e[n]*t[n]);return r},So=(e,t)=>{let r=e[0].dims,n=t??Ad(e[1]),o=zf(r,n),i=E.size(o),a=e[0].dataType,d=k("input",a,r.length),l=M("output",a,o.length),c=m=>`
+  }`};return e.compute({name:"MultiHeadAttentionAddBias",shaderCache:{inputDependencies:["type","type"]},getRunData:()=>({outputs:[{dims:d,dataType:t.dataType,gpuDataType:0}],dispatchGroup:{x:Math.ceil(l/64)},programUniforms:c}),getShaderSource:m},{inputs:[t,r],outputs:[-1]})[0]},nr=(e,t,r,n,o,i,a,d)=>{let l=i;if(a&&k.size(a.dims)>0){if(n===1)throw new Error("AddBiasReshape is not implemented. Please export your model with packed QKV or KV");return l=Pf(e,i,a,t,n,r*o,d),l=l.reshape([t,n,r,o]),r===1||n===1?l:e.compute(Pe(l,Td.perm),{inputs:[l],outputs:[-1]})[0]}else return i.dims.length===3&&(l=i.reshape([t,n,r,o])),r===1||n===1?l:e.compute(Pe(l,Td.perm),{inputs:[l],outputs:[-1]})[0]},Cd=(e,t)=>{let r=Ef(e.inputs,t),n=e.inputs[0],o=Ne(e.inputs,1),i=Ne(e.inputs,2),a=Ne(e.inputs,3),d=Ne(e.inputs,4),l=Ne(e.inputs,5),c=Ne(e.inputs,6),m=Ne(e.inputs,7);if(n.dims.length===5)throw new Error("Packed QKV is not implemented");if(o?.dims.length===5)throw new Error("Packed KV is not implemented");let u=o&&i&&o.dims.length===4&&i.dims.length===4,h=nr(e,r.batchSize,r.numHeads,r.sequenceLength,r.headSize,n,a,0);if(u)return Wt(e,h,o,i,d,void 0,c,m,l,r,t);if(!o||!i)throw new Error("key and value must be provided");let w=nr(e,r.batchSize,r.numHeads,r.kvSequenceLength,r.headSize,o,a,r.hiddenSize),g=nr(e,r.batchSize,r.numHeads,r.kvSequenceLength,r.vHeadSize,i,a,2*r.hiddenSize);Wt(e,h,w,g,d,void 0,c,m,l,r,t)}});var Ad,zf,Of,So,kd,To=V(()=>{"use strict";Q();ie();ae();Ad=e=>Array.from(e.getBigInt64Array(),Number),zf=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(Ad(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")},Of=(e,t)=>{let r=[];for(let n=0;n<e.length;++n)r.push(e[n]*t[n]);return r},So=(e,t)=>{let r=e[0].dims,n=t??Ad(e[1]),o=Of(r,n),i=k.size(o),a=e[0].dataType,d=E("input",a,r.length),l=M("output",a,o.length),c=m=>`
       const inputShape = ${d.indices(...r)};
       ${m.registerUniform("output_size","u32").declareVariables(d,l)}
       ${m.mainStart()}
@@ -5409,22 +5409,22 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
         ${d.indicesSet("input_indices","i","input_dim_value")}
       }
       ${l.setByOffset("global_idx",d.getByIndices("input_indices"))}
-    }`;return{name:"Tile",shaderCache:{hint:`${n}`,inputDependencies:["rank"]},getRunData:()=>({outputs:[{dims:o,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(i/64)},programUniforms:[{type:12,data:i},...V(e[0].dims,o)]}),getShaderSource:c}},Ed=e=>{Of(e.inputs),e.compute(So(e.inputs),{inputs:[0]})}});var Df,kd,Od,Bf,Pd,zd,Dd=U(()=>{"use strict";Z();te();Ie();Kr();re();xo();To();lt();Df=(e,t)=>{let r=e[0],n=e[1],o=e[2],i=e[3],a=e[4];if(r.dims.length!==3&&r.dims.length!==5)throw new Error("Input query is expected to have 3 or 5 dimensions");let d=!1,l=r.dims[0],c=r.dims[1],m=r.dims.length===3?d?r.dims[2]/3:r.dims[2]:t.numHeads*r.dims[4],u=c,h=0,w=0,g=Math.floor(m/t.numHeads),y=i&&i.dims.length!==0,S=a&&a.dims.length!==0,$=!0;if(y&&S){if(i.dims.length!==4)throw new Error('Input "past_key" is expected to have 4 dimensions');if(a.dims.length!==4)throw new Error('Input "past_value" is expected to have 4 dimensions');$?(h=i.dims[1],w=i.dims[1]):(h=i.dims[2],w=i.dims[2])}else if(y||S)throw new Error('Input "past_key" and "past_value" shall be both present or both absent');let _;if(n){if(r.dims.length!==3)throw new Error('Input "query" is expected to have 3 dimensions when key is given');if(n.dims.length<3||n.dims.length>5)throw new Error('Input "key" is expected to have 3, 4, or 5 dimensions');if(r.dims[0]!==n.dims[0])throw new Error('Input "query" and "key" shall have same dim 0 (batch size)');if(n.dims.length===3){if(r.dims[2]%n.dims[2]!==0)throw new Error('Dimension 2 of "query" should be a multiple of "key"');_=2,u=n.dims[1]}else if(n.dims.length===5){if(n.dims[2]!==t.numHeads||n.dims[3]!==2||n.dims[4]!==g)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.');_=5,u=n.dims[1]}else{if(n.dims[1]!==t.numHeads||n.dims[3]!==g)throw new Error('Expect "key" shape (batch_size, num_heads, kv_sequence_length, head_size) for past_key');_=0,u=n.dims[2]}}else{if(r.dims.length!==3&&r.dims.length!==5)throw new Error('Input "query" is expected to have 3 or 5 dimensions when key is empty');if(r.dims.length===5&&(r.dims[2]!==t.numHeads||r.dims[3]!==3))throw new Error('Expect "query" shape (batch_size, kv_sequence_length, num_heads, 3, head_size) for packed kv');_=3}let x=0,T=!1,C=m;if(o){if(o.dims.length!==3&&o.dims.length!==4)throw new Error('Input "value" is expected to have 3 or 4 dimensions');if(r.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(u!==o.dims[1])throw new Error('Input "key" and "value" shall have the same dim 1 (kv_sequence_length)');C=o.dims[2]}else{if(u!==o.dims[2])throw new Error('Input "past_key" and "past_value" shall have the same dim 2 (kv_sequence_length)');C=o.dims[1]*o.dims[3],T=!0}}let A=h+u,P=!1;return{batchSize:l,sequenceLength:c,pastSequenceLength:h,kvSequenceLength:u,totalSequenceLength:A,maxSequenceLength:w,inputHiddenSize:0,hiddenSize:m,vHiddenSize:C,headSize:g,vHeadSize:Math.floor(C/t.kvNumHeads),numHeads:t.numHeads,kvNumHeads:t.kvNumHeads,nReps:t.numHeads/t.kvNumHeads,pastPresentShareBuffer:!1,maskType:x,scale:t.scale,broadcastResPosBias:P,passPastInKv:T,qkvFormat:_,isPastkvBSNH:$}},kd=(e,t,r,n)=>{let o=[n.batchSize,n.totalSequenceLength,n.kvNumHeads,n.headSize],i=4,a=E.size(o)/i,d=n.totalSequenceLength,l=M("present_kv",r,o.length,i),c=k("new_kv",e.dataType,e.dims.length,i),m=t?k("past_kv",t.dataType,t.dims.length,i):void 0,u=Math.ceil(n.headSize/i),h={x:d,y:e.dims[0],z:1},w=t?["rank","rank"]:["rank"],g=[{type:12,data:a},{type:12,data:n.pastSequenceLength},{type:12,data:n.kvSequenceLength},{type:12,data:n.totalSequenceLength}],y=[c];m?(g.push(...V(e.dims),...V(t.dims),...V(o)),y.push(m)):g.push(...V(e.dims),...V(o));let S=[{name:"output_size",type:"u32"},{name:"past_seqlen",type:"u32"},{name:"new_seqlen",type:"u32"},{name:"present_seqlen",type:"u32"}],$=`      let past_batch_stride = uniforms.past_seqlen * num_heads * H;
+    }`;return{name:"Tile",shaderCache:{hint:`${n}`,inputDependencies:["rank"]},getRunData:()=>({outputs:[{dims:o,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(i/64)},programUniforms:[{type:12,data:i},...R(e[0].dims,o)]}),getShaderSource:c}},kd=e=>{zf(e.inputs),e.compute(So(e.inputs),{inputs:[0]})}});var Bf,Ed,zd,Df,Pd,Od,Bd=V(()=>{"use strict";Q();ie();Ie();Kr();ae();xo();To();lt();Bf=(e,t)=>{let r=e[0],n=e[1],o=e[2],i=e[3],a=e[4];if(r.dims.length!==3&&r.dims.length!==5)throw new Error("Input query is expected to have 3 or 5 dimensions");let d=!1,l=r.dims[0],c=r.dims[1],m=r.dims.length===3?d?r.dims[2]/3:r.dims[2]:t.numHeads*r.dims[4],u=c,h=0,w=0,g=Math.floor(m/t.numHeads),y=i&&i.dims.length!==0,S=a&&a.dims.length!==0,$=!0;if(y&&S){if(i.dims.length!==4)throw new Error('Input "past_key" is expected to have 4 dimensions');if(a.dims.length!==4)throw new Error('Input "past_value" is expected to have 4 dimensions');$?(h=i.dims[1],w=i.dims[1]):(h=i.dims[2],w=i.dims[2])}else if(y||S)throw new Error('Input "past_key" and "past_value" shall be both present or both absent');let v;if(n){if(r.dims.length!==3)throw new Error('Input "query" is expected to have 3 dimensions when key is given');if(n.dims.length<3||n.dims.length>5)throw new Error('Input "key" is expected to have 3, 4, or 5 dimensions');if(r.dims[0]!==n.dims[0])throw new Error('Input "query" and "key" shall have same dim 0 (batch size)');if(n.dims.length===3){if(r.dims[2]%n.dims[2]!==0)throw new Error('Dimension 2 of "query" should be a multiple of "key"');v=2,u=n.dims[1]}else if(n.dims.length===5){if(n.dims[2]!==t.numHeads||n.dims[3]!==2||n.dims[4]!==g)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.');v=5,u=n.dims[1]}else{if(n.dims[1]!==t.numHeads||n.dims[3]!==g)throw new Error('Expect "key" shape (batch_size, num_heads, kv_sequence_length, head_size) for past_key');v=0,u=n.dims[2]}}else{if(r.dims.length!==3&&r.dims.length!==5)throw new Error('Input "query" is expected to have 3 or 5 dimensions when key is empty');if(r.dims.length===5&&(r.dims[2]!==t.numHeads||r.dims[3]!==3))throw new Error('Expect "query" shape (batch_size, kv_sequence_length, num_heads, 3, head_size) for packed kv');v=3}let x=0,T=!1,C=m;if(o){if(o.dims.length!==3&&o.dims.length!==4)throw new Error('Input "value" is expected to have 3 or 4 dimensions');if(r.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(u!==o.dims[1])throw new Error('Input "key" and "value" shall have the same dim 1 (kv_sequence_length)');C=o.dims[2]}else{if(u!==o.dims[2])throw new Error('Input "past_key" and "past_value" shall have the same dim 2 (kv_sequence_length)');C=o.dims[1]*o.dims[3],T=!0}}let A=h+u,P=!1;return{batchSize:l,sequenceLength:c,pastSequenceLength:h,kvSequenceLength:u,totalSequenceLength:A,maxSequenceLength:w,inputHiddenSize:0,hiddenSize:m,vHiddenSize:C,headSize:g,vHeadSize:Math.floor(C/t.kvNumHeads),numHeads:t.numHeads,kvNumHeads:t.kvNumHeads,nReps:t.numHeads/t.kvNumHeads,pastPresentShareBuffer:!1,maskType:x,scale:t.scale,broadcastResPosBias:P,passPastInKv:T,qkvFormat:v,isPastkvBSNH:$}},Ed=(e,t,r,n)=>{let o=[n.batchSize,n.totalSequenceLength,n.kvNumHeads,n.headSize],i=4,a=k.size(o)/i,d=n.totalSequenceLength,l=M("present_kv",r,o.length,i),c=E("new_kv",e.dataType,e.dims.length,i),m=t?E("past_kv",t.dataType,t.dims.length,i):void 0,u=Math.ceil(n.headSize/i),h={x:d,y:e.dims[0],z:1},w=t?["rank","rank"]:["rank"],g=[{type:12,data:a},{type:12,data:n.pastSequenceLength},{type:12,data:n.kvSequenceLength},{type:12,data:n.totalSequenceLength}],y=[c];m?(g.push(...R(e.dims),...R(t.dims),...R(o)),y.push(m)):g.push(...R(e.dims),...R(o));let S=[{name:"output_size",type:"u32"},{name:"past_seqlen",type:"u32"},{name:"new_seqlen",type:"u32"},{name:"present_seqlen",type:"u32"}],$=`      let past_batch_stride = uniforms.past_seqlen * num_heads * H;
         var past_head_stride = uniforms.past_seqlen * H;
         if (is_bsnh) {
           past_head_stride = H;
         }
         let in_offset = b * past_batch_stride + s * row_stride + n * past_head_stride + h;
-        present_kv[out_offset] = past_kv[in_offset];`,_=`      let new_batch_stride = uniforms.new_seqlen * num_heads * H;
+        present_kv[out_offset] = past_kv[in_offset];`,v=`      let new_batch_stride = uniforms.new_seqlen * num_heads * H;
         let new_row_stride = num_heads * H;
         let new_head_stride = H;
         let in_offset = b * new_batch_stride + (s - past_seqlen) * new_row_stride + n * new_head_stride + h;
         present_kv[out_offset] = new_kv[in_offset];`,x=t?`if (s < past_seqlen) {
         ${$}
         } else if (s < past_seqlen + uniforms.new_seqlen) {
-        ${_}
+        ${v}
         }`:`if (s < past_seqlen + uniforms.new_seqlen) {
-          ${_}
+          ${v}
         }`,T=C=>`
 
   ${C.registerUniforms(S).declareVariables(...y,l)}
@@ -5455,7 +5455,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
 
     let out_offset = b * present_batch_stride + s * row_stride + n * present_head_stride + h;
     ${x}
-  }`;return{name:"ConcatPastNew",shaderCache:{hint:`${n.kvNumHeads}${u}${!!t}`,inputDependencies:w},getRunData:()=>({outputs:[{dims:o,dataType:r}],dispatchGroup:h,programUniforms:g}),getShaderSource:T}},Od=e=>J({...e}),Bf=J({perm:[0,2,1,3]}),Pd=(e,t,r,n,o)=>{let i=t,a=n.kvNumHeads,d=n.nReps;return t.dims.length===3&&n.kvSequenceLength!==0&&(i=t.reshape([n.batchSize,n.kvSequenceLength,a,n.headSize])),r?i=e.compute(kd(i,r,i.dataType,n),{inputs:[i,r],outputs:[n.isPastkvBSNH?o:-1]})[0]:i=e.compute(kd(i,void 0,i.dataType,n),{inputs:[i],outputs:[n.isPastkvBSNH?o:-1]})[0],d!==1&&(i=e.compute(So([i],[1,1,1,d]),{inputs:[i],outputs:[-1]})[0],i=i.reshape([n.batchSize,n.totalSequenceLength,a*d,n.headSize])),e.compute(Pe(i,Bf.perm),{inputs:[i],outputs:[-1]})[0]},zd=(e,t)=>{let r=Df(e.inputs,t);if(e.inputs[0].dims.length===5)throw new Error("Packed QKV is not implemented");if(e.inputs[1]?.dims.length===5)throw new Error("Packed KV is not implemented");let n=nr(e,r.batchSize,r.numHeads,r.sequenceLength,r.headSize,e.inputs[0],void 0,0),o=e.inputs[3]&&e.inputs[3].dims.length!==0?e.inputs[3]:void 0,i=e.inputs[4]&&e.inputs[4].dims.length!==0?e.inputs[4]:void 0,a=Pd(e,e.inputs[1],o,r,1),d=Pd(e,e.inputs[2],i,r,2);Wt(e,n,a,d,void 0,void 0,void 0,void 0,void 0,r,t)}});var Bd,Mf,Rf,Md,Rd=U(()=>{"use strict";Z();te();lt();re();Bd=(e,t,r,n,o,i,a,d)=>{let l=ve(i),c=l===1?"f32":`vec${l}f`,m=l===1?"vec2f":`mat2x${l}f`,u=o*a,h=[o,a,i/l],w=[o,a,2],g=["rank","type","type"],y=[];y.push(...V(h,w));let S=$=>{let _=k("x",t.dataType,3,l),x=k("scale",r.dataType,r.dims),T=k("bias",n.dataType,n.dims),C=M("output",1,3,2),A=[_,x,T,C],P=64;return`
+  }`;return{name:"ConcatPastNew",shaderCache:{hint:`${n.kvNumHeads}${u}${!!t}`,inputDependencies:w},getRunData:()=>({outputs:[{dims:o,dataType:r}],dispatchGroup:h,programUniforms:g}),getShaderSource:T}},zd=e=>J({...e}),Df=J({perm:[0,2,1,3]}),Pd=(e,t,r,n,o)=>{let i=t,a=n.kvNumHeads,d=n.nReps;return t.dims.length===3&&n.kvSequenceLength!==0&&(i=t.reshape([n.batchSize,n.kvSequenceLength,a,n.headSize])),r?i=e.compute(Ed(i,r,i.dataType,n),{inputs:[i,r],outputs:[n.isPastkvBSNH?o:-1]})[0]:i=e.compute(Ed(i,void 0,i.dataType,n),{inputs:[i],outputs:[n.isPastkvBSNH?o:-1]})[0],d!==1&&(i=e.compute(So([i],[1,1,1,d]),{inputs:[i],outputs:[-1]})[0],i=i.reshape([n.batchSize,n.totalSequenceLength,a*d,n.headSize])),e.compute(Pe(i,Df.perm),{inputs:[i],outputs:[-1]})[0]},Od=(e,t)=>{let r=Bf(e.inputs,t);if(e.inputs[0].dims.length===5)throw new Error("Packed QKV is not implemented");if(e.inputs[1]?.dims.length===5)throw new Error("Packed KV is not implemented");let n=nr(e,r.batchSize,r.numHeads,r.sequenceLength,r.headSize,e.inputs[0],void 0,0),o=e.inputs[3]&&e.inputs[3].dims.length!==0?e.inputs[3]:void 0,i=e.inputs[4]&&e.inputs[4].dims.length!==0?e.inputs[4]:void 0,a=Pd(e,e.inputs[1],o,r,1),d=Pd(e,e.inputs[2],i,r,2);Wt(e,n,a,d,void 0,void 0,void 0,void 0,void 0,r,t)}});var Dd,Mf,Rf,Md,Rd=V(()=>{"use strict";Q();ie();lt();ae();Dd=(e,t,r,n,o,i,a,d)=>{let l=we(i),c=l===1?"f32":`vec${l}f`,m=l===1?"vec2f":`mat2x${l}f`,u=o*a,h=[o,a,i/l],w=[o,a,2],g=["rank","type","type"],y=[];y.push(...R(h,w));let S=$=>{let v=E("x",t.dataType,3,l),x=E("scale",r.dataType,r.dims),T=E("bias",n.dataType,n.dims),C=M("output",1,3,2),A=[v,x,T,C],P=64;return`
   var<workgroup> workgroup_shared : array<${m}, ${P}>;
   const workgroup_size = ${P}u;
   ${$.declareVariables(...A)}
@@ -5467,7 +5467,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
     var sum = ${c}(0);
     var squared_sum = ${c}(0);
     for (var h = local_idx; h < hight; h += workgroup_size) {
-      let value = ${c}(${_.get("batch","channel","h")});
+      let value = ${c}(${v.get("batch","channel","h")});
       sum += value;
       squared_sum += value * value;
     }
@@ -5489,19 +5489,19 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
       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:`${l};${d}`,inputDependencies:g},getRunData:()=>({outputs:[{dims:w,dataType:1}],dispatchGroup:{x:u},programUniforms:y}),getShaderSource:S},{inputs:[t,r,n],outputs:[-1]})[0]},Mf=(e,t,r)=>{let n=t[0].dims,o=n,i=2,a=n[0],d=n[1],l=E.sizeFromDimension(n,i),c=ve(l),m=E.size(o)/c,u=Bd(e,t[0],t[1],t[2],a,l,d,r.epsilon),h=[a,d,l/c],w=[a,d],g=["type","none"],y=S=>{let $=k("x",t[0].dataType,h.length,c),_=k("scale_shift",1,w.length,2),x=M("output",t[0].dataType,h.length,c),T=[$,_,x];return`
+  }`};return e.compute({name:"InstanceNormComputeChannelScaleShift",shaderCache:{hint:`${l};${d}`,inputDependencies:g},getRunData:()=>({outputs:[{dims:w,dataType:1}],dispatchGroup:{x:u},programUniforms:y}),getShaderSource:S},{inputs:[t,r,n],outputs:[-1]})[0]},Mf=(e,t,r)=>{let n=t[0].dims,o=n,i=2,a=n[0],d=n[1],l=k.sizeFromDimension(n,i),c=we(l),m=k.size(o)/c,u=Dd(e,t[0],t[1],t[2],a,l,d,r.epsilon),h=[a,d,l/c],w=[a,d],g=["type","none"],y=S=>{let $=E("x",t[0].dataType,h.length,c),v=E("scale_shift",1,w.length,2),x=M("output",t[0].dataType,h.length,c),T=[$,v,x];return`
   ${S.registerUniform("output_size","u32").declareVariables(...T)}
   ${S.mainStart()}
   ${S.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")}
       let outputIndices = ${x.offsetToIndices("global_idx")};
       let batch = outputIndices[0];
       let channel = outputIndices[1];
-      let scale_shift = ${_.getByIndices("vec2<u32>(batch, channel)")};
+      let scale_shift = ${v.getByIndices("vec2<u32>(batch, channel)")};
       let value = ${$.getByOffset("global_idx")} * ${x.type.value}(scale_shift.x) + ${x.type.value}(scale_shift.y);
       ${x.setByOffset("global_idx","value")};
-  }`};e.compute({name:"InstanceNormalization",shaderCache:{hint:`${c}`,inputDependencies:g},getRunData:()=>({outputs:[{dims:o,dataType:t[0].dataType}],dispatchGroup:{x:Math.ceil(m/64)},programUniforms:[{type:12,data:m},...V(h,w,h)]}),getShaderSource:y},{inputs:[t[0],u]})},Rf=(e,t,r)=>{let n=t[0].dims,o=n,i=n[0],a=n[n.length-1],d=E.sizeFromDimension(n,1)/a,l=ve(a),c=E.size(o)/l,m=[{type:12,data:d},{type:12,data:Math.floor(a/l)}],u=["type","type"],h=[0,n.length-1];for(let S=0;S<n.length-2;S++)h.push(S+1);let w=e.compute(Pe(e.inputs[0],h),{inputs:[e.inputs[0]],outputs:[-1]})[0],g=Bd(e,w,t[1],t[2],i,d,a,r.epsilon),y=S=>{let $=ye(t[0].dataType),_=l===1?"vec2f":`mat${l}x2f`,x=A=>{let P=A===0?"x":"y",D=l===1?"f32":`vec${l}f`;switch(l){case 1:return`${$}(${D}(scale.${P}))`;case 2:return`vec2<${$}>(${D}(scale[0].${P}, scale[1].${P}))`;case 4:return`vec4<${$}>(${D}(scale[0].${P}, scale[1].${P}, scale[2].${P}, scale[3].${P}))`;default:throw new Error(`Not supported compoents ${l}`)}},T=k("input",t[0].dataType,t[0].dims,l),C=M("output",t[0].dataType,o,l);return`
+  }`};e.compute({name:"InstanceNormalization",shaderCache:{hint:`${c}`,inputDependencies:g},getRunData:()=>({outputs:[{dims:o,dataType:t[0].dataType}],dispatchGroup:{x:Math.ceil(m/64)},programUniforms:[{type:12,data:m},...R(h,w,h)]}),getShaderSource:y},{inputs:[t[0],u]})},Rf=(e,t,r)=>{let n=t[0].dims,o=n,i=n[0],a=n[n.length-1],d=k.sizeFromDimension(n,1)/a,l=we(a),c=k.size(o)/l,m=[{type:12,data:d},{type:12,data:Math.floor(a/l)}],u=["type","type"],h=[0,n.length-1];for(let S=0;S<n.length-2;S++)h.push(S+1);let w=e.compute(Pe(e.inputs[0],h),{inputs:[e.inputs[0]],outputs:[-1]})[0],g=Dd(e,w,t[1],t[2],i,d,a,r.epsilon),y=S=>{let $=he(t[0].dataType),v=l===1?"vec2f":`mat${l}x2f`,x=A=>{let P=A===0?"x":"y",B=l===1?"f32":`vec${l}f`;switch(l){case 1:return`${$}(${B}(scale.${P}))`;case 2:return`vec2<${$}>(${B}(scale[0].${P}, scale[1].${P}))`;case 4:return`vec4<${$}>(${B}(scale[0].${P}, scale[1].${P}, scale[2].${P}, scale[3].${P}))`;default:throw new Error(`Not supported compoents ${l}`)}},T=E("input",t[0].dataType,t[0].dims,l),C=M("output",t[0].dataType,o,l);return`
   @group(0) @binding(0) var<storage, read> input : array<${T.type.storage}>;
-  @group(0) @binding(1) var<storage, read> scale_input : array<${_}>;
+  @group(0) @binding(1) var<storage, read> scale_input : array<${v}>;
   @group(0) @binding(2) var<storage, read_write> output : array<${C.type.storage}>;
   struct Uniforms {H: u32, C : u32};
   @group(0) @binding(3) var<uniform> uniforms: Uniforms;
@@ -5513,10 +5513,10 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
     let scale_offset = current_image_number * uniforms.C + current_channel_number;
     let scale = scale_input[scale_offset];
     output[global_idx] = fma(input[global_idx], ${x(0)}, ${x(1)});
-  }`};e.compute({name:"InstanceNormalizationNHWC",shaderCache:{hint:`${l}`,inputDependencies:u},getRunData:()=>({outputs:[{dims:o,dataType:t[0].dataType}],dispatchGroup:{x:Math.ceil(c/64)},programUniforms:m}),getShaderSource:y},{inputs:[t[0],g]})},Md=(e,t)=>{t.format==="NHWC"?Rf(e,e.inputs,t):Mf(e,e.inputs,t)}});var Uf,Vf,Ud,Vd=U(()=>{"use strict";Z();te();re();Uf=e=>{if(!e||e.length<2)throw new Error("layerNorm requires at least 2 inputs.")},Vf=(e,t,r)=>{let n=t.simplified,o=e[0].dims,i=e[1],a=!n&&e[2],d=o,l=E.normalizeAxis(t.axis,o.length),c=E.sizeToDimension(o,l),m=E.sizeFromDimension(o,l),u=E.size(i.dims),h=a?E.size(a.dims):0;if(u!==m||a&&h!==m)throw new Error(`Size of X.shape()[axis:] == ${m}.
+  }`};e.compute({name:"InstanceNormalizationNHWC",shaderCache:{hint:`${l}`,inputDependencies:u},getRunData:()=>({outputs:[{dims:o,dataType:t[0].dataType}],dispatchGroup:{x:Math.ceil(c/64)},programUniforms:m}),getShaderSource:y},{inputs:[t[0],g]})},Md=(e,t)=>{t.format==="NHWC"?Rf(e,e.inputs,t):Mf(e,e.inputs,t)}});var Uf,Vf,Ud,Vd=V(()=>{"use strict";Q();ie();ae();Uf=e=>{if(!e||e.length<2)throw new Error("layerNorm requires at least 2 inputs.")},Vf=(e,t,r)=>{let n=t.simplified,o=e[0].dims,i=e[1],a=!n&&e[2],d=o,l=k.normalizeAxis(t.axis,o.length),c=k.sizeToDimension(o,l),m=k.sizeFromDimension(o,l),u=k.size(i.dims),h=a?k.size(a.dims):0;if(u!==m||a&&h!==m)throw new Error(`Size of X.shape()[axis:] == ${m}.
        Size of scale and bias (if provided) must match this.
-       Got scale size of ${u} and bias size of ${h}`);let w=[];for(let C=0;C<o.length;++C)C<l?w.push(o[C]):w.push(1);let g=ve(m),y=["type","type"],S=[{type:12,data:c},{type:1,data:m},{type:12,data:Math.floor(m/g)},{type:1,data:t.epsilon}];a&&y.push("type");let $=r>1,_=r>2,x=C=>{let A=ye(e[0].dataType),P=[k("x",e[0].dataType,e[0].dims,g),k("scale",i.dataType,i.dims,g)];a&&P.push(k("bias",a.dataType,a.dims,g)),P.push(M("output",e[0].dataType,d,g)),$&&P.push(M("mean_data_output",1,w)),_&&P.push(M("inv_std_output",1,w));let D=[{name:"norm_count",type:"u32"},{name:"norm_size",type:"f32"},{name:"norm_size_vectorized",type:"u32"},{name:"epsilon",type:"f32"}];return`
-  ${C.registerUniforms(D).declareVariables(...P)}
+       Got scale size of ${u} and bias size of ${h}`);let w=[];for(let C=0;C<o.length;++C)C<l?w.push(o[C]):w.push(1);let g=we(m),y=["type","type"],S=[{type:12,data:c},{type:1,data:m},{type:12,data:Math.floor(m/g)},{type:1,data:t.epsilon}];a&&y.push("type");let $=r>1,v=r>2,x=C=>{let A=he(e[0].dataType),P=[E("x",e[0].dataType,e[0].dims,g),E("scale",i.dataType,i.dims,g)];a&&P.push(E("bias",a.dataType,a.dims,g)),P.push(M("output",e[0].dataType,d,g)),$&&P.push(M("mean_data_output",1,w)),v&&P.push(M("inv_std_output",1,w));let B=[{name:"norm_count",type:"u32"},{name:"norm_size",type:"f32"},{name:"norm_size_vectorized",type:"u32"},{name:"epsilon",type:"f32"}];return`
+  ${C.registerUniforms(B).declareVariables(...P)}
   ${C.mainStart()}
     ${C.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.norm_count")}
     let offset = global_idx * uniforms.norm_size_vectorized;
@@ -5524,7 +5524,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
     var mean_square_vector = ${io("f32",g)};
 
     for (var h: u32 = 0u; h < uniforms.norm_size_vectorized; h++) {
-      let value = ${Et(A,g,"x[h + offset]")};
+      let value = ${kt(A,g,"x[h + offset]")};
       mean_vector += value;
       mean_square_vector += value * value;
     }
@@ -5532,33 +5532,33 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
     let inv_std_dev = inverseSqrt(${Ze("mean_square_vector",g)} / uniforms.norm_size ${n?"":"- mean * mean"} + uniforms.epsilon);
 
     for (var j: u32 = 0; j < uniforms.norm_size_vectorized; j++) {
-      let f32input = ${Et(A,g,"x[j + offset]")};
-      let f32scale = ${Et(A,g,"scale[j]")};
+      let f32input = ${kt(A,g,"x[j + offset]")};
+      let f32scale = ${kt(A,g,"scale[j]")};
       output[j + offset] = ${P[0].type.value}((f32input ${n?"":"- mean"}) * inv_std_dev * f32scale
-        ${a?`+ ${Et(A,g,"bias[j]")}`:""}
+        ${a?`+ ${kt(A,g,"bias[j]")}`:""}
       );
     }
 
     ${$?"mean_data_output[global_idx] = mean":""};
-    ${_?"inv_std_output[global_idx] = inv_std_dev":""};
-  }`},T=[{dims:d,dataType:e[0].dataType}];return $&&T.push({dims:w,dataType:1}),_&&T.push({dims:w,dataType:1}),{name:"LayerNormalization",shaderCache:{hint:`${g};${r};${n}`,inputDependencies:y},getRunData:()=>({outputs:T,dispatchGroup:{x:Math.ceil(c/64)},programUniforms:S}),getShaderSource:x}},Ud=(e,t)=>{Uf(e.inputs),e.compute(Vf(e.inputs,t,e.outputCount))}});var Nf,Wf,Nd,Wd,Ld=U(()=>{"use strict";Z();te();Ie();re();Nf=(e,t)=>{if(e.length<3||e.length>4)throw new Error("MatMulNBits requires 3 or 4 inputs");let r=e[0],n=r.dims.length;if(r.dims[n-1]!==t.k)throw new Error("The last dim of input shape does not match the k value");let o=Math.floor((t.k+t.blockSize-1)/t.blockSize),i=t.blockSize/8*t.bits,a=e[1];if(!E.areEqual(a.dims,[t.n,o,i]))throw new Error("The second inputs must be 3D tensor with shape N X nBlocksPerCol X blobSize");let l=e[2].dims;if(E.size(l)!==t.n*o)throw new Error("scales input size error.");if(e.length===4){let m=e[3].dims,u=t.bits>4?t.n*o:t.n*Math.floor((o+1)/2);if(E.size(m)!==u)throw new Error("zeroPoints input size error.")}},Wf=(e,t)=>{let r=e[0].dims,n=r.length,o=r[n-2],i=t.k,a=t.n,d=r.slice(0,n-2),l=E.size(d),m=e[1].dims[2]/4,u=e[0].dataType,h=ve(t.k),w=ve(m),g=ve(a),y=d.concat([o,a]),S=o>1&&a/g%2===0?2:1,$=E.size(y)/g/S,_=64,x=[],T=[l,o,i/h],C=E.convertShape(e[1].dims).slice();C.splice(-1,1,m/w),x.push(...V(T)),x.push(...V(C)),x.push(...V(e[2].dims)),e.length===4&&x.push(...V(E.convertShape(e[3].dims)));let A=[l,o,a/g];x.push(...V(A));let P=D=>{let N=T.length,L=k("a",e[0].dataType,N,h),K=k("b",12,C.length,w),X=k("scales",e[2].dataType,e[2].dims.length),ie=[L,K,X],ae=e.length===4?k("zero_points",12,e[3].dims.length):void 0;ae&&ie.push(ae);let le=A.length,Y=M("output",e[0].dataType,le,g),me=ye(e[0].dataType),pe=(()=>{switch(h){case 1:return`array<${me}, 8>`;case 2:return`mat4x2<${me}>`;case 4:return`mat2x4<${me}>`;default:throw new Error(`${h}-component is not supported.`)}})(),Q=()=>{let oe=`
+    ${v?"inv_std_output[global_idx] = inv_std_dev":""};
+  }`},T=[{dims:d,dataType:e[0].dataType}];return $&&T.push({dims:w,dataType:1}),v&&T.push({dims:w,dataType:1}),{name:"LayerNormalization",shaderCache:{hint:`${g};${r};${n}`,inputDependencies:y},getRunData:()=>({outputs:T,dispatchGroup:{x:Math.ceil(c/64)},programUniforms:S}),getShaderSource:x}},Ud=(e,t)=>{Uf(e.inputs),e.compute(Vf(e.inputs,t,e.outputCount))}});var Nf,Wf,Lf,Nd,Wd,Ld=V(()=>{"use strict";Q();ie();Ie();ae();Nf=(e,t)=>{if(e.length<3||e.length>4)throw new Error("MatMulNBits requires 3 or 4 inputs");let r=e[0],n=r.dims.length;if(r.dims[n-1]!==t.k)throw new Error("The last dim of input shape does not match the k value");let o=Math.floor((t.k+t.blockSize-1)/t.blockSize),i=t.blockSize/8*t.bits,a=e[1];if(!k.areEqual(a.dims,[t.n,o,i]))throw new Error("The second inputs must be 3D tensor with shape N X nBlocksPerCol X blobSize");let l=e[2].dims;if(k.size(l)!==t.n*o)throw new Error("scales input size error.");if(e.length===4){let m=e[3].dims,u=t.bits>4?t.n*o:t.n*Math.floor((o+1)/2);if(k.size(m)!==u)throw new Error("zeroPoints input size error.")}},Wf=(e,t)=>{let r=e[0].dims,n=r.length,o=r[n-2],i=t.k,a=t.n,d=r.slice(0,n-2),l=k.size(d),m=e[1].dims[2]/4,u=e[0].dataType,h=we(t.k),w=we(m),g=we(a),y=d.concat([o,a]),S=o>1&&a/g%2===0?2:1,$=k.size(y)/g/S,v=64,x=[],T=[l,o,i/h],C=k.convertShape(e[1].dims).slice();C.splice(-1,1,m/w),x.push(...R(T)),x.push(...R(C)),x.push(...R(e[2].dims)),e.length===4&&x.push(...R(k.convertShape(e[3].dims)));let A=[l,o,a/g];x.push(...R(A));let P=B=>{let N=T.length,W=E("a",e[0].dataType,N,h),K=E("b",12,C.length,w),Z=E("scales",e[2].dataType,e[2].dims.length),ee=[W,K,Z],se=e.length===4?E("zero_points",12,e[3].dims.length):void 0;se&&ee.push(se);let de=A.length,Y=M("output",e[0].dataType,de,g),le=he(e[0].dataType),ce=(()=>{switch(h){case 1:return`array<${le}, 8>`;case 2:return`mat4x2<${le}>`;case 4:return`mat2x4<${le}>`;default:throw new Error(`${h}-component is not supported.`)}})(),q=()=>{let ne=`
           // reuse a data
-            var input_offset = ${L.indicesToOffset(`${L.type.indices}(batch, row, word_offset)`)};
-            var a_data: ${pe};
+            var input_offset = ${W.indicesToOffset(`${W.type.indices}(batch, row, word_offset)`)};
+            var a_data: ${ce};
             for (var j: u32 = 0; j < ${8/h}; j++) {
-              a_data[j] = ${L.getByOffset("input_offset")};
+              a_data[j] = ${W.getByOffset("input_offset")};
               input_offset++;
             }
-          `;for(let se=0;se<g*S;se++)oe+=`
-            b_value = ${w===1?`b${se}_data`:`b${se}_data[i]`};
+          `;for(let oe=0;oe<g*S;oe++)ne+=`
+            b_value = ${w===1?`b${oe}_data`:`b${oe}_data[i]`};
             b_value_lower = unpack4xU8(b_value & b_mask);
             b_value_upper = unpack4xU8((b_value >> 4) & b_mask);
-            b_quantized_values = ${pe}(${Array.from({length:4},(R,G)=>`${me}(b_value_lower[${G}]), ${me}(b_value_upper[${G}])`).join(", ")});
-            b_dequantized_values = ${(()=>h===1?`${pe}(${Array.from({length:8},(R,G)=>`(b_quantized_values[${G}] - ${ae?`zero_point${se}`:"zero_point"}) * scale${se}`).join(", ")});`:`(b_quantized_values - ${pe}(${Array(8).fill(`${ae?`zero_point${se}`:"zero_point"}`).join(",")})) * scale${se};`)()};
-            workgroup_shared[local_id.x * ${S} + ${Math.floor(se/g)}]${g>1?`[${se%g}]`:""} += ${Array.from({length:8/h},(R,G)=>`${h===1?`a_data[${G}] * b_dequantized_values[${G}]`:`dot(a_data[${G}], b_dequantized_values[${G}])`}`).join(" + ")};
-          `;return oe},be=()=>{let oe=`
+            b_quantized_values = ${ce}(${Array.from({length:4},(U,G)=>`${le}(b_value_lower[${G}]), ${le}(b_value_upper[${G}])`).join(", ")});
+            b_dequantized_values = ${(()=>h===1?`${ce}(${Array.from({length:8},(U,G)=>`(b_quantized_values[${G}] - ${se?`zero_point${oe}`:"zero_point"}) * scale${oe}`).join(", ")});`:`(b_quantized_values - ${ce}(${Array(8).fill(`${se?`zero_point${oe}`:"zero_point"}`).join(",")})) * scale${oe};`)()};
+            workgroup_shared[local_id.x * ${S} + ${Math.floor(oe/g)}]${g>1?`[${oe%g}]`:""} += ${Array.from({length:8/h},(U,G)=>`${h===1?`a_data[${G}] * b_dequantized_values[${G}]`:`dot(a_data[${G}], b_dequantized_values[${G}])`}`).join(" + ")};
+          `;return ne},ue=()=>{let ne=`
             var col_index = col * ${g};
-            ${ae?`
+            ${se?`
             let zero_point_bytes_per_col = (nBlocksPerCol + 1) / 2;
             var zero_point_byte_count: u32;
             var zero_point_word_index: u32;
@@ -5567,42 +5567,42 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
             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 = ${me}(8);`}
-            `;for(let se=0;se<g*S;se++)oe+=`
-            let scale${se} = ${X.getByOffset("col_index * nBlocksPerCol + block")};
-            ${ae?`
+            let zero_point = ${le}(8);`}
+            `;for(let oe=0;oe<g*S;oe++)ne+=`
+            let scale${oe} = ${Z.getByOffset("col_index * nBlocksPerCol + block")};
+            ${se?`
             zero_point_byte_count = col_index * zero_point_bytes_per_col + (block >> 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 = ${ae.getByOffset("zero_point_word_index")} >> zero_point_bits_offset;
-            let zero_point${se} = ${me}((zero_point_word) & 0xFu);`:""}
-            col_index += 1;`;return oe},ne=()=>{let oe=`col_index = col * ${g};`;for(let se=0;se<g*S;se++)oe+=`
-            let b${se}_data = ${K.getByIndices(`${K.type.indices}(col_index, block, word)`)};
-            col_index += 1;`;return oe+=`
+            zero_point_word = ${se.getByOffset("zero_point_word_index")} >> zero_point_bits_offset;
+            let zero_point${oe} = ${le}((zero_point_word) & 0xFu);`:""}
+            col_index += 1;`;return ne},re=()=>{let ne=`col_index = col * ${g};`;for(let oe=0;oe<g*S;oe++)ne+=`
+            let b${oe}_data = ${K.getByIndices(`${K.type.indices}(col_index, block, word)`)};
+            col_index += 1;`;return ne+=`
             var b_value: u32;
             let b_mask: u32 = 0x0F0F0F0Fu;
             var b_value_lower: vec4<u32>;
             var b_value_upper: vec4<u32>;
-            var b_quantized_values: ${pe};
-            var b_dequantized_values: ${pe};`,oe};return`
-        var<workgroup> workgroup_shared: array<${Y.type.value}, ${S*_}>;
-        ${D.declareVariables(...ie,Y)}
-        ${D.mainStart([_,1,1])}
-          let output_indices = ${Y.offsetToIndices(`(global_idx / ${_}) * ${S}`)};
+            var b_quantized_values: ${ce};
+            var b_dequantized_values: ${ce};`,ne};return`
+        var<workgroup> workgroup_shared: array<${Y.type.value}, ${S*v}>;
+        ${B.declareVariables(...ee,Y)}
+        ${B.mainStart([v,1,1])}
+          let output_indices = ${Y.offsetToIndices(`(global_idx / ${v}) * ${S}`)};
           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 += ${_}) {
+          for (var block = local_id.x; block < nBlocksPerCol; block += ${v}) {
             //process one block
             var word_offset: u32 = block * ${t.blockSize/h};
-            ${be()}
+            ${ue()}
             for (var word: u32 = 0; word < ${m}; word += ${w}) {
-              ${ne()}
+              ${re()}
               for (var i: u32 = 0; i < ${w}; i++) {
-                ${Q()}
+                ${q()}
                 word_offset += ${8/h};
               }
             }
@@ -5612,13 +5612,87 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
           if (local_id.x < ${S}) {
             var output_value: ${Y.type.value} = ${Y.type.value}(0);
             var workgroup_shared_offset: u32 = local_id.x;
-            for (var b: u32 = 0u; b < ${_}u; b++) {
+            for (var b: u32 = 0u; b < ${v}u; b++) {
               output_value += workgroup_shared[workgroup_shared_offset];
               workgroup_shared_offset += ${S};
             }
             ${Y.setByIndices(`${Y.type.indices}(batch, row, col + local_id.x)`,"output_value")};
           }
-        }`};return{name:"MatMulNBits",shaderCache:{hint:`${t.blockSize};${t.bits};${h};${w};${g};${S};${_}`,inputDependencies:Array(e.length).fill("rank")},getRunData:()=>({outputs:[{dims:y,dataType:u}],dispatchGroup:{x:$},programUniforms:x}),getShaderSource:P}},Nd=(e,t)=>{Nf(e.inputs,t),e.compute(Wf(e.inputs,t))},Wd=e=>J(e)});var Lf,Hf,Gf,Ff,qf,jf,Kf,Yf,Hd,Gd=U(()=>{"use strict";Z();te();re();Lf=e=>{if(!e||e.length<1)throw new Error("Too few inputs");if(e[0].dataType!==1&&e[0].dataType!==10)throw new Error("Input type must be float or float16.");if(e.length>=2){let t=e[0].dims.length*2===e[1].dims[0];if(e.length===4&&(t=e[3].dims[0]*2===e[1].dims[0]),!t)throw new Error("The pads should be a 1D tensor of shape [2 * input_rank] or [2 * num_axes].")}},Hf=(e,t,r)=>{let n="";for(let o=t-1;o>=0;--o)n+=`
+        }`};return{name:"MatMulNBits",shaderCache:{hint:`${t.blockSize};${t.bits};${h};${w};${g};${S};${v}`,inputDependencies:Array(e.length).fill("rank")},getRunData:()=>({outputs:[{dims:y,dataType:u}],dispatchGroup:{x:$},programUniforms:x}),getShaderSource:P}},Lf=(e,t)=>{let r=e[0].dims,n=r.length,o=r[n-2],i=t.k,a=t.n,d=r.slice(0,n-2),l=k.size(d),m=e[1].dims[2]/4,u=e[0].dataType,h=we(t.k),w=we(m),g=d.concat([o,a]),y=128,S=a%8===0?8:a%4===0?4:1,$=y/S,v=$*w*8,x=v/h,T=v/t.blockSize,C=k.size(g)/S,A=[],P=[l,o,i/h],B=k.convertShape(e[1].dims).slice();B.splice(-1,1,m/w),A.push(...R(P)),A.push(...R(B)),A.push(...R(e[2].dims)),e.length===4&&A.push(...R(k.convertShape(e[3].dims)));let N=[l,o,a];A.push(...R(N));let W=K=>{let Z=P.length,ee=E("a",e[0].dataType,Z,h),se=E("b",12,B.length,w),de=E("scales",e[2].dataType,e[2].dims.length),Y=[ee,se,de],le=e.length===4?E("zero_points",12,e[3].dims.length):void 0;le&&Y.push(le);let ce=N.length,q=M("output",e[0].dataType,ce),ue=he(e[0].dataType),re=()=>{switch(h){case 1:return`
+          let a_data0 = vec4<${ue}>(sub_a[word_offset], sub_a[word_offset + 1], sub_a[word_offset + 2], sub_a[word_offset + 3]);
+          let a_data1 = vec4<${ue}>(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<${ue}>(sub_a[word_offset], sub_a[word_offset + 1]);
+          let a_data1 = vec4<${ue}>(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(`${h}-component is not supported.`)}};return`
+        var<workgroup> sub_a: array<${ee.type.value}, ${x}>;
+        var<workgroup> inter_results: array<array<${q.type.value}, ${$}>, ${S}>;
+        ${K.declareVariables(...Y,q)}
+        ${K.mainStart([$,S,1])}
+          let output_indices = ${q.offsetToIndices(`workgroup_index * ${S}`)};
+          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) / ${T} + 1;
+
+          // Loop over shared dimension.
+          for (var tile: u32 = 0; tile < num_tiles; tile += 1) {
+            let a_col_start = tile * ${x};
+            // load one tile A data into shared memory.
+            for (var a_offset = local_idx; a_offset < ${x}; a_offset += ${y})
+            {
+              let a_col = a_col_start + a_offset;
+              if (a_col < uniforms.a_shape[2])
+              {
+                sub_a[a_offset] = ${ee.getByIndices(`${ee.type.indices}(batch, row, a_col)`)};
+              } else {
+                sub_a[a_offset] = ${ee.type.value}(0);
+              }
+            }
+            workgroupBarrier();
+
+            // each thread process one block
+            let b_row = col + local_id.y;
+            let block = tile * ${T} + local_id.x;
+            ${le?`
+            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 = ${le.getByOffset("zero_point_word_index")} >> zero_point_bits_offset;
+            let zero_point = ${ue}((zero_point_word) & 0xFu);`:`
+            // The default zero point is 8 for unsigned 4-bit quantization.
+            let zero_point = ${ue}(8);`}
+            let scale = ${de.getByOffset("b_row * n_blocks_per_col + block")};
+            let b_data = ${se.getByIndices(`${se.type.indices}(b_row, block, 0)`)};
+            var word_offset = local_id.x * ${t.blockSize/h};
+            for (var i: u32 = 0; i < ${w}; i++) {
+              ${re()}
+              let b_value = ${w===1?"b_data":"b_data[i]"};
+              let b_value_lower = unpack4xU8(b_value & 0x0F0F0F0Fu);
+              let b_value_upper = unpack4xU8((b_value >> 4) & 0x0F0F0F0Fu);
+              let b_quantized_values = mat2x4<${ue}>(${Array.from({length:4},(ne,oe)=>`${ue}(b_value_lower[${oe}]), ${ue}(b_value_upper[${oe}])`).join(", ")});
+              let b_dequantized_values = (b_quantized_values - mat2x4<${ue}>(${Array(8).fill("zero_point").join(",")})) * scale;
+              inter_results[local_id.y][local_id.x] += ${Array.from({length:2},(ne,oe)=>`${`dot(a_data${oe}, b_dequantized_values[${oe}])`}`).join(" + ")};
+              word_offset += ${8/h};
+            }
+            workgroupBarrier();
+          }
+
+          if (local_idx < ${S}) {
+            var output_value: ${q.type.value} = ${q.type.value}(0);
+            for (var b = 0u; b < ${$}; b++) {
+              output_value += inter_results[local_idx][b];
+            }
+            if (col + local_idx < uniforms.output_shape[2])
+            {
+              ${q.setByIndices(`${q.type.indices}(batch, row, col + local_idx)`,"output_value")}
+            }
+          }
+        }`};return{name:"BlockwiseMatMulNBits32",shaderCache:{hint:`${t.blockSize};${h};${w};${$};${S}`,inputDependencies:Array(e.length).fill("rank")},getRunData:()=>({outputs:[{dims:g,dataType:u}],dispatchGroup:{x:C},programUniforms:A}),getShaderSource:W}},Nd=(e,t)=>{Nf(e.inputs,t),t.blockSize===32&&e.adapterInfo.isVendor("intel")&&e.adapterInfo.isArchitecture("gen-12lp")?e.compute(Lf(e.inputs,t)):e.compute(Wf(e.inputs,t))},Wd=e=>J(e)});var Hf,Gf,Ff,qf,jf,Kf,Yf,Xf,Hd,Gd=V(()=>{"use strict";Q();ie();ae();Hf=e=>{if(!e||e.length<1)throw new Error("Too few inputs");if(e[0].dataType!==1&&e[0].dataType!==10)throw new Error("Input type must be float or float16.");if(e.length>=2){let t=e[0].dims.length*2===e[1].dims[0];if(e.length===4&&(t=e[3].dims[0]*2===e[1].dims[0]),!t)throw new Error("The pads should be a 1D tensor of shape [2 * input_rank] or [2 * num_axes].")}},Gf=(e,t,r)=>{let n="";for(let o=t-1;o>=0;--o)n+=`
             k = i32(${e.indicesGet("indices",o)}) - ${F("uniforms.pads",o,r)};
             if (k < 0) {
               break;
@@ -5635,7 +5709,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
             ${n}
             value = x[offset];
           }
-      `},Gf=(e,t,r)=>{let n="";for(let o=t-1;o>=0;--o)n+=`
+      `},Ff=(e,t,r)=>{let n="";for(let o=t-1;o>=0;--o)n+=`
                 k = i32(${e.indicesGet("indices",o)}) - ${F("uniforms.pads",o,r)};
                 if (k < 0) {
                   k = -k;
@@ -5653,7 +5727,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
               var k = 0;
               ${n}
               value = x[offset];
-          `},Ff=(e,t,r)=>{let n="";for(let o=t-1;o>=0;--o)n+=`
+          `},qf=(e,t,r)=>{let n="";for(let o=t-1;o>=0;--o)n+=`
                 k = i32(${e.indicesGet("indices",o)}) - ${F("uniforms.pads",o,r)};
                 if (k < 0) {
                   k = 0;
@@ -5667,7 +5741,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
               var k = 0;
               ${n}
               value = x[offset];
-          `},qf=(e,t,r)=>{let n="";for(let o=t-1;o>=0;--o)n+=`
+          `},jf=(e,t,r)=>{let n="";for(let o=t-1;o>=0;--o)n+=`
                 k = i32(${e.indicesGet("indices",o)}) - ${F("uniforms.pads",o,r)};
                 if (k < 0)  {
                   k += i32(${F("uniforms.x_shape",o,t)}]);
@@ -5681,7 +5755,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
               var k = 0;
               ${n}
               value = x[offset];
-          `},jf=(e,t,r)=>{switch(r.mode){case 0:return Hf(e,t,r.pads.length);case 1:return Gf(e,t,r.pads.length);case 2:return Ff(e,t,r.pads.length);case 3:return qf(e,t,r.pads.length);default:throw new Error("Invalid mode")}},Kf=(e,t)=>{let r=E.padShape(e[0].dims.slice(),t.pads),n=e[0].dims,o=E.size(r),i=[{type:12,data:o},{type:6,data:t.pads}],a=e.length>=3&&e[2].data;t.mode===0&&i.push({type:a?e[2].dataType:1,data:t.value}),i.push(...V(e[0].dims,r));let d=["rank"],l=c=>{let m=M("output",e[0].dataType,r.length),u=k("x",e[0].dataType,n.length),h=u.type.value,w=jf(m,n.length,t),g=[{name:"output_size",type:"u32"},{name:"pads",type:"i32",length:t.pads.length}];return t.mode===0&&g.push({name:"constant_value",type:a?h:"f32"}),`
+          `},Kf=(e,t,r)=>{switch(r.mode){case 0:return Gf(e,t,r.pads.length);case 1:return Ff(e,t,r.pads.length);case 2:return qf(e,t,r.pads.length);case 3:return jf(e,t,r.pads.length);default:throw new Error("Invalid mode")}},Yf=(e,t)=>{let r=k.padShape(e[0].dims.slice(),t.pads),n=e[0].dims,o=k.size(r),i=[{type:12,data:o},{type:6,data:t.pads}],a=e.length>=3&&e[2].data;t.mode===0&&i.push({type:a?e[2].dataType:1,data:t.value}),i.push(...R(e[0].dims,r));let d=["rank"],l=c=>{let m=M("output",e[0].dataType,r.length),u=E("x",e[0].dataType,n.length),h=u.type.value,w=Kf(m,n.length,t),g=[{name:"output_size",type:"u32"},{name:"pads",type:"i32",length:t.pads.length}];return t.mode===0&&g.push({name:"constant_value",type:a?h:"f32"}),`
             ${c.registerUniforms(g).declareVariables(u,m)}
             ${c.mainStart()}
             ${c.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")}
@@ -5691,11 +5765,11 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
             var value = ${h}(0);
             ${w}
             output[global_idx] = value;
-        }`};return{name:"Pad",shaderCache:{hint:`${t.mode}${a}`,inputDependencies:d},getRunData:()=>({outputs:[{dims:r,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(E.size(r)/64)},programUniforms:i}),getShaderSource:l}},Yf=(e,t)=>{if(e.length>1){let r=e[1].getBigInt64Array(),n=e.length>=3&&e[2].data?e[2].dataType===10?e[2].getUint16Array()[0]:e[2].getFloat32Array()[0]:0,o=e[0].dims.length,i=new Int32Array(2*o).fill(0);if(e.length>=4){let d=e[3].getBigInt64Array();for(let l=0;l<d.length;l++)i[Number(d[l])]=Number(r[l]),i[Number(d[l])+o]=Number(r[l+d.length])}else r.forEach((d,l)=>i[Number(l)]=Number(d));let a=[];return i.forEach(d=>a.push(d)),{mode:t.mode,value:n,pads:a}}else return t},Hd=(e,t)=>{Lf(e.inputs);let r=Yf(e.inputs,t);e.compute(Kf(e.inputs,r),{inputs:[0]})}});var nn,Fd,qd,jd,Kd,Xf,Zf,Yd,Xd,Zd,Qd,Jd,el,tl,rl,nl,ol,il,al,sl=U(()=>{"use strict";Ke();Z();te();re();nn=e=>{if(we.webgpu.validateInputContent&&(!e||e.length!==1))throw new Error("Pool ops requires 1 input.")},Fd=(e,t,r)=>{let n=t.format==="NHWC",o=e.dims.slice();n&&o.splice(1,0,o.pop());let i=Object.hasOwnProperty.call(t,"dilations"),a=t.kernelShape.slice(),d=t.strides.slice(),l=i?t.dilations.slice():[],c=t.pads.slice();Ct.adjustPoolAttributes(r,o,a,d,l,c);let m=Ct.computePoolOutputShape(r,o,d,l,a,c,t.autoPad),u=Object.assign({},t);i?Object.assign(u,{kernelShape:a,strides:d,pads:c,dilations:l,cacheKey:t.cacheKey}):Object.assign(u,{kernelShape:a,strides:d,pads:c,cacheKey:t.cacheKey});let h=m.slice();return h.push(h.splice(1,1)[0]),[u,n?h:m]},qd=(e,t)=>{let r=t.format==="NHWC",n=E.size(e),o=E.size(t.kernelShape),i=[{type:12,data:n},{type:12,data:o}],a=[{name:"outputSize",type:"u32"},{name:"kernelSize",type:"u32"}];if(t.kernelShape.length<=2){let d=t.kernelShape[t.kernelShape.length-1],l=t.strides[t.strides.length-1],c=t.pads[t.pads.length/2-1],m=t.pads[t.pads.length-1],u=!!(c+m);i.push({type:12,data:d},{type:12,data:l},{type:12,data:c},{type:12,data:m}),a.push({name:"kw",type:"u32"},{name:"sw",type:"u32"},{name:"pwStart",type:"u32"},{name:"pwEnd",type:"u32"});let h=!1;if(t.kernelShape.length===2){let w=t.kernelShape[t.kernelShape.length-2],g=t.strides[t.strides.length-2],y=t.pads[t.pads.length/2-2],S=t.pads[t.pads.length-2];h=!!(y+S),i.push({type:12,data:w},{type:12,data:g},{type:12,data:y},{type:12,data:S}),a.push({name:"kh",type:"u32"},{name:"sh",type:"u32"},{name:"phStart",type:"u32"},{name:"phEnd",type:"u32"})}return[i,a,!0,u,h]}else{if(r)throw new Error("Pooling with kernelShape.length > 2 is not supported for NHWC format.");let d=E.computeStrides(t.kernelShape);i.push({type:12,data:d},{type:12,data:t.pads},{type:12,data:t.strides}),a.push({name:"kernelStrides",type:"u32",length:d.length},{name:"pads",type:"u32",length:t.pads.length},{name:"strides",type:"u32",length:t.strides.length});let l=t.pads.reduce((c,m)=>c+m);return[i,a,!!l,!1,!1]}},jd=(e,t,r,n,o,i,a,d,l,c,m,u)=>{let h=o.format==="NHWC",w=t.type.value,g=M("output",t.type.tensor,n);if(o.kernelShape.length<=2){let y="",S="",$="",_=r-(h?2:1);if(m?y=`
+        }`};return{name:"Pad",shaderCache:{hint:`${t.mode}${a}`,inputDependencies:d},getRunData:()=>({outputs:[{dims:r,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(k.size(r)/64)},programUniforms:i}),getShaderSource:l}},Xf=(e,t)=>{if(e.length>1){let r=e[1].getBigInt64Array(),n=e.length>=3&&e[2].data?e[2].dataType===10?e[2].getUint16Array()[0]:e[2].getFloat32Array()[0]:0,o=e[0].dims.length,i=new Int32Array(2*o).fill(0);if(e.length>=4){let d=e[3].getBigInt64Array();for(let l=0;l<d.length;l++)i[Number(d[l])]=Number(r[l]),i[Number(d[l])+o]=Number(r[l+d.length])}else r.forEach((d,l)=>i[Number(l)]=Number(d));let a=[];return i.forEach(d=>a.push(d)),{mode:t.mode,value:n,pads:a}}else return t},Hd=(e,t)=>{Hf(e.inputs);let r=Xf(e.inputs,t);e.compute(Yf(e.inputs,r),{inputs:[0]})}});var nn,Fd,qd,jd,Kd,Zf,Qf,Yd,Xd,Zd,Qd,Jd,el,tl,rl,nl,ol,il,al,sl=V(()=>{"use strict";Ke();Q();ie();ae();nn=e=>{if(_e.webgpu.validateInputContent&&(!e||e.length!==1))throw new Error("Pool ops requires 1 input.")},Fd=(e,t,r)=>{let n=t.format==="NHWC",o=e.dims.slice();n&&o.splice(1,0,o.pop());let i=Object.hasOwnProperty.call(t,"dilations"),a=t.kernelShape.slice(),d=t.strides.slice(),l=i?t.dilations.slice():[],c=t.pads.slice();Ct.adjustPoolAttributes(r,o,a,d,l,c);let m=Ct.computePoolOutputShape(r,o,d,l,a,c,t.autoPad),u=Object.assign({},t);i?Object.assign(u,{kernelShape:a,strides:d,pads:c,dilations:l,cacheKey:t.cacheKey}):Object.assign(u,{kernelShape:a,strides:d,pads:c,cacheKey:t.cacheKey});let h=m.slice();return h.push(h.splice(1,1)[0]),[u,n?h:m]},qd=(e,t)=>{let r=t.format==="NHWC",n=k.size(e),o=k.size(t.kernelShape),i=[{type:12,data:n},{type:12,data:o}],a=[{name:"outputSize",type:"u32"},{name:"kernelSize",type:"u32"}];if(t.kernelShape.length<=2){let d=t.kernelShape[t.kernelShape.length-1],l=t.strides[t.strides.length-1],c=t.pads[t.pads.length/2-1],m=t.pads[t.pads.length-1],u=!!(c+m);i.push({type:12,data:d},{type:12,data:l},{type:12,data:c},{type:12,data:m}),a.push({name:"kw",type:"u32"},{name:"sw",type:"u32"},{name:"pwStart",type:"u32"},{name:"pwEnd",type:"u32"});let h=!1;if(t.kernelShape.length===2){let w=t.kernelShape[t.kernelShape.length-2],g=t.strides[t.strides.length-2],y=t.pads[t.pads.length/2-2],S=t.pads[t.pads.length-2];h=!!(y+S),i.push({type:12,data:w},{type:12,data:g},{type:12,data:y},{type:12,data:S}),a.push({name:"kh",type:"u32"},{name:"sh",type:"u32"},{name:"phStart",type:"u32"},{name:"phEnd",type:"u32"})}return[i,a,!0,u,h]}else{if(r)throw new Error("Pooling with kernelShape.length > 2 is not supported for NHWC format.");let d=k.computeStrides(t.kernelShape);i.push({type:12,data:d},{type:12,data:t.pads},{type:12,data:t.strides}),a.push({name:"kernelStrides",type:"u32",length:d.length},{name:"pads",type:"u32",length:t.pads.length},{name:"strides",type:"u32",length:t.strides.length});let l=t.pads.reduce((c,m)=>c+m);return[i,a,!!l,!1,!1]}},jd=(e,t,r,n,o,i,a,d,l,c,m,u)=>{let h=o.format==="NHWC",w=t.type.value,g=M("output",t.type.tensor,n);if(o.kernelShape.length<=2){let y="",S="",$="",v=r-(h?2:1);if(m?y=`
                 for (var i: u32 = 0u; i < uniforms.kw; i++) {
-                  xIndices[${_}] = indices[${_}] * uniforms.sw - uniforms.pwStart + i;
-                  if (xIndices[${_}] < 0 || xIndices[${_}]
-                      >= uniforms.x_shape[${_}]) {
+                  xIndices[${v}] = indices[${v}] * uniforms.sw - uniforms.pwStart + i;
+                  if (xIndices[${v}] < 0 || xIndices[${v}]
+                      >= uniforms.x_shape[${v}]) {
                     pad++;
                     continue;
                   }
@@ -5703,7 +5777,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
                   ${i}
                 }`:y=`
                 for (var i: u32 = 0u; i < uniforms.kw; i++) {
-                  xIndices[${_}] = indices[${_}] * uniforms.sw - uniforms.pwStart + i;
+                  xIndices[${v}] = indices[${v}] * uniforms.sw - uniforms.pwStart + i;
                   let x_val = x[${t.indicesToOffset("xIndices")}];
                   ${i}
                 }`,o.kernelShape.length===2){let T=r-(h?3:2);u?S=`
@@ -5780,10 +5854,10 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
               ${a}
 
               output[global_idx] = value;
-            }`}},Kd=e=>`${e.format};${e.ceilMode};${e.autoPad};${e.kernelShape.length}`,Xf=e=>`${Kd(e)};${e.countIncludePad}`,Zf=e=>`${Kd(e)};${e.storageOrder};${e.dilations}`,Yd=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}),Xd=(e,t,r,n)=>{let[o,i]=Fd(t,n,r),a=k("x",t.dataType,t.dims.length),d=a.type.value,l="value += x_val;",c="";o.countIncludePad?c+=`value /= ${d}(uniforms.kernelSize);`:c+=`value /= ${d}(i32(uniforms.kernelSize) - pad);`;let[m,u,h,w,g]=qd(i,o);m.push(...V(t.dims,i));let y=["rank"];return{name:e,shaderCache:{hint:`${n.cacheKey};${h};${w};${g}`,inputDependencies:y},getRunData:()=>({outputs:[{dims:i,dataType:t.dataType}],dispatchGroup:{x:Math.ceil(E.size(i)/64)},programUniforms:m}),getShaderSource:S=>jd(S,a,t.dims.length,i.length,o,l,c,0,u,h,w,g)}},Zd=e=>{let t=e.count_include_pad!==0,r=Yd(e);if(r.ceilMode!==0)throw new Error("using ceil() in shape computation is not yet supported for AveragePool");let n={countIncludePad:t,...r,cacheKey:""};return{...n,cacheKey:Xf(n)}},Qd=(e,t)=>{nn(e.inputs),e.compute(Xd("AveragePool",e.inputs[0],!1,t))},Jd={autoPad:"",ceilMode:0,countIncludePad:!1,kernelShape:[],strides:[],pads:[],storageOrder:0,dilations:[]},el=e=>{let t=e.format;return{format:t,...Jd,cacheKey:t}},tl=(e,t)=>{nn(e.inputs),e.compute(Xd("GlobalAveragePool",e.inputs[0],!0,t))},rl=(e,t,r,n)=>{let[o,i]=Fd(t,n,r),a=`
+            }`}},Kd=e=>`${e.format};${e.ceilMode};${e.autoPad};${e.kernelShape.length}`,Zf=e=>`${Kd(e)};${e.countIncludePad}`,Qf=e=>`${Kd(e)};${e.storageOrder};${e.dilations}`,Yd=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}),Xd=(e,t,r,n)=>{let[o,i]=Fd(t,n,r),a=E("x",t.dataType,t.dims.length),d=a.type.value,l="value += x_val;",c="";o.countIncludePad?c+=`value /= ${d}(uniforms.kernelSize);`:c+=`value /= ${d}(i32(uniforms.kernelSize) - pad);`;let[m,u,h,w,g]=qd(i,o);m.push(...R(t.dims,i));let y=["rank"];return{name:e,shaderCache:{hint:`${n.cacheKey};${h};${w};${g}`,inputDependencies:y},getRunData:()=>({outputs:[{dims:i,dataType:t.dataType}],dispatchGroup:{x:Math.ceil(k.size(i)/64)},programUniforms:m}),getShaderSource:S=>jd(S,a,t.dims.length,i.length,o,l,c,0,u,h,w,g)}},Zd=e=>{let t=e.count_include_pad!==0,r=Yd(e);if(r.ceilMode!==0)throw new Error("using ceil() in shape computation is not yet supported for AveragePool");let n={countIncludePad:t,...r,cacheKey:""};return{...n,cacheKey:Zf(n)}},Qd=(e,t)=>{nn(e.inputs),e.compute(Xd("AveragePool",e.inputs[0],!1,t))},Jd={autoPad:"",ceilMode:0,countIncludePad:!1,kernelShape:[],strides:[],pads:[],storageOrder:0,dilations:[]},el=e=>{let t=e.format;return{format:t,...Jd,cacheKey:t}},tl=(e,t)=>{nn(e.inputs),e.compute(Xd("GlobalAveragePool",e.inputs[0],!0,t))},rl=(e,t,r,n)=>{let[o,i]=Fd(t,n,r),a=`
       value = max(x_val, value);
-    `,d="",l=k("x",t.dataType,t.dims.length),c=["rank"],[m,u,h,w,g]=qd(i,o);return m.push(...V(t.dims,i)),{name:e,shaderCache:{hint:`${n.cacheKey};${h};${w};${g}`,inputDependencies:c},getRunData:()=>({outputs:[{dims:i,dataType:t.dataType}],dispatchGroup:{x:Math.ceil(E.size(i)/64)},programUniforms:m}),getShaderSource:y=>jd(y,l,t.dims.length,i.length,o,a,d,t.dataType===10?-65504:-1e5,u,h,w,g)}},nl=(e,t)=>{nn(e.inputs),e.compute(rl("MaxPool",e.inputs[0],!1,t))},ol=e=>{let t=e.storage_order,r=e.dilations,n=Yd(e);if(t!==0)throw new Error("column major storage order is not yet supported for MaxPool");if(n.ceilMode!==0)throw new Error("using ceil() in shape computation is not yet supported for MaxPool");let o={storageOrder:t,dilations:r,...n,cacheKey:""};return{...o,cacheKey:Zf(o)}},il=e=>{let t=e.format;return{format:t,...Jd,cacheKey:t}},al=(e,t)=>{nn(e.inputs),e.compute(rl("GlobalMaxPool",e.inputs[0],!0,t))}});var Jf,eh,ul,dl,ll=U(()=>{"use strict";Z();te();Ie();re();Jf=(e,t)=>{if(e.length<2||e.length>3)throw new Error("DequantizeLinear requires 2 or 3 inputs.");if(e.length===3&&e[1].dims===e[2].dims)throw new Error("x-scale and x-zero-point must have the same shape.");if(e.length===3&&e[0].dataType!==e[2].dataType)throw new Error("x and x-zero-point must have the same data type.");if(e[0].dataType===6&&e.length>2)throw new Error("In the case of dequantizing int32 there is no zero point.");if(e[1].dims.length!==0&&e[1].dims.length!==1&&e[1].dims.length!==e[0].dims.length)throw new Error("scale input must be a scalar, a 1D tensor, or have the same rank as the input tensor.");if(e.length>2){if(e[0].dataType!==e[2].dataType)throw new Error("x and x-zero-point must have the same data type.");if(e[1].dims.length!==e[2].dims.length)throw new Error("scale and zero-point inputs must have the same rank.");if(!e[1].dims.map((r,n)=>r===e[2].dims[n]).reduce((r,n)=>r&&n,!0))throw new Error("scale and zero-point inputs must have the same shape.")}if(t.blockSize>0){if(e[1].dims.length===0||e[1].dims.length===1&&e[1].dims[0]===1)throw new Error("blockSize must be set only for block quantization.");if(!e[1].dims.map((o,i)=>i===t.axis||o===e[0].dims[i]).reduce((o,i)=>o&&i,!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 r=e[0].dims[t.axis],n=e[1].dims[t.axis];if(t.blockSize<Math.ceil(r/n)||t.blockSize>Math.ceil(r/(n-1)-1))throw new Error("blockSize must be with in the range [ceil(dI / Si), ceil(dI / (Si - 1) - 1)].")}},eh=(e,t)=>{let r=E.normalizeAxis(t.axis,e[0].dims.length),n=e[0].dataType,o=n===3,i=e[0].dims,a=e[1].dataType,d=E.size(i),l=n===3||n===2,c=l?[Math.ceil(E.size(e[0].dims)/4)]:e[0].dims,m=e[1].dims,u=e.length>2?e[2]:void 0,h=u?l?[Math.ceil(E.size(u.dims)/4)]:u.dims:void 0,w=m.length===0||m.length===1&&m[0]===1,g=w===!1&&m.length===1,y=ve(d),S=w&&(!l||y===4),$=S?y:1,_=S&&!l?y:1,x=k("input",l?12:n,c.length,_),T=k("scale",a,m.length),C=u?k("zero_point",l?12:n,h.length):void 0,A=M("output",a,i.length,$),P=[x,T];C&&P.push(C);let D=[c,m];u&&D.push(h);let N=[{type:12,data:d/$},{type:12,data:r},{type:12,data:t.blockSize},...V(...D,i)],L=K=>{let X=[{name:"output_size",type:"u32"},{name:"axis",type:"u32"},{name:"block_size",type:"u32"}];return`
-      ${K.registerUniforms(X).declareVariables(...P,A)}
+    `,d="",l=E("x",t.dataType,t.dims.length),c=["rank"],[m,u,h,w,g]=qd(i,o);return m.push(...R(t.dims,i)),{name:e,shaderCache:{hint:`${n.cacheKey};${h};${w};${g}`,inputDependencies:c},getRunData:()=>({outputs:[{dims:i,dataType:t.dataType}],dispatchGroup:{x:Math.ceil(k.size(i)/64)},programUniforms:m}),getShaderSource:y=>jd(y,l,t.dims.length,i.length,o,a,d,t.dataType===10?-65504:-1e5,u,h,w,g)}},nl=(e,t)=>{nn(e.inputs),e.compute(rl("MaxPool",e.inputs[0],!1,t))},ol=e=>{let t=e.storage_order,r=e.dilations,n=Yd(e);if(t!==0)throw new Error("column major storage order is not yet supported for MaxPool");if(n.ceilMode!==0)throw new Error("using ceil() in shape computation is not yet supported for MaxPool");let o={storageOrder:t,dilations:r,...n,cacheKey:""};return{...o,cacheKey:Qf(o)}},il=e=>{let t=e.format;return{format:t,...Jd,cacheKey:t}},al=(e,t)=>{nn(e.inputs),e.compute(rl("GlobalMaxPool",e.inputs[0],!0,t))}});var eh,th,ul,dl,ll=V(()=>{"use strict";Q();ie();Ie();ae();eh=(e,t)=>{if(e.length<2||e.length>3)throw new Error("DequantizeLinear requires 2 or 3 inputs.");if(e.length===3&&e[1].dims===e[2].dims)throw new Error("x-scale and x-zero-point must have the same shape.");if(e.length===3&&e[0].dataType!==e[2].dataType)throw new Error("x and x-zero-point must have the same data type.");if(e[0].dataType===6&&e.length>2)throw new Error("In the case of dequantizing int32 there is no zero point.");if(e[1].dims.length!==0&&e[1].dims.length!==1&&e[1].dims.length!==e[0].dims.length)throw new Error("scale input must be a scalar, a 1D tensor, or have the same rank as the input tensor.");if(e.length>2){if(e[0].dataType!==e[2].dataType)throw new Error("x and x-zero-point must have the same data type.");if(e[1].dims.length!==e[2].dims.length)throw new Error("scale and zero-point inputs must have the same rank.");if(!e[1].dims.map((r,n)=>r===e[2].dims[n]).reduce((r,n)=>r&&n,!0))throw new Error("scale and zero-point inputs must have the same shape.")}if(t.blockSize>0){if(e[1].dims.length===0||e[1].dims.length===1&&e[1].dims[0]===1)throw new Error("blockSize must be set only for block quantization.");if(!e[1].dims.map((o,i)=>i===t.axis||o===e[0].dims[i]).reduce((o,i)=>o&&i,!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 r=e[0].dims[t.axis],n=e[1].dims[t.axis];if(t.blockSize<Math.ceil(r/n)||t.blockSize>Math.ceil(r/(n-1)-1))throw new Error("blockSize must be with in the range [ceil(dI / Si), ceil(dI / (Si - 1) - 1)].")}},th=(e,t)=>{let r=k.normalizeAxis(t.axis,e[0].dims.length),n=e[0].dataType,o=n===3,i=e[0].dims,a=e[1].dataType,d=k.size(i),l=n===3||n===2,c=l?[Math.ceil(k.size(e[0].dims)/4)]:e[0].dims,m=e[1].dims,u=e.length>2?e[2]:void 0,h=u?l?[Math.ceil(k.size(u.dims)/4)]:u.dims:void 0,w=m.length===0||m.length===1&&m[0]===1,g=w===!1&&m.length===1,y=we(d),S=w&&(!l||y===4),$=S?y:1,v=S&&!l?y:1,x=E("input",l?12:n,c.length,v),T=E("scale",a,m.length),C=u?E("zero_point",l?12:n,h.length):void 0,A=M("output",a,i.length,$),P=[x,T];C&&P.push(C);let B=[c,m];u&&B.push(h);let N=[{type:12,data:d/$},{type:12,data:r},{type:12,data:t.blockSize},...R(...B,i)],W=K=>{let Z=[{name:"output_size",type:"u32"},{name:"axis",type:"u32"},{name:"block_size",type:"u32"}];return`
+      ${K.registerUniforms(Z).declareVariables(...P,A)}
       ${K.mainStart()}
           ${K.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")}
           let output_indices = ${A.offsetToIndices("global_idx")};
@@ -5820,13 +5894,13 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
                 let zero_point_value = zero_point_vec[zero_point_offset % 4];`:`let zero_point_value = ${C.getByIndices("scale_indices")};`:`let zero_point_value = ${l?o?"i32":"u32":x.type.value}(0);`)()};
       // Compute and write output
       ${A.setByOffset("global_idx",`${A.type.value}(x_value - zero_point_value) * scale_value`)};
-      }`};return{name:"DequantizeLinear",shaderCache:{hint:t.cacheKey,inputDependencies:C?["rank","rank","rank"]:["rank","rank"]},getShaderSource:L,getRunData:()=>({outputs:[{dims:i,dataType:a}],dispatchGroup:{x:Math.ceil(d/$/64),y:1,z:1},programUniforms:N})}},ul=(e,t)=>{Jf(e.inputs,t),e.compute(eh(e.inputs,t))},dl=e=>J({axis:e.axis,blockSize:e.blockSize})});var th,rh,cl,pl=U(()=>{"use strict";Ke();Z();re();th=(e,t,r)=>{let n=e===t,o=e<t&&r<0,i=e>t&&r>0;if(n||o||i)throw new Error("Range these inputs' contents are invalid.")},rh=(e,t,r,n)=>{let o=Math.abs(Math.ceil((t-e)/r)),i=[o],a=o,d=[{type:12,data:a},{type:n,data:e},{type:n,data:r},...V(i)],l=c=>{let m=M("output",n,i.length),u=m.type.value,h=[{name:"outputSize",type:"u32"},{name:"start",type:u},{name:"delta",type:u}];return`
+      }`};return{name:"DequantizeLinear",shaderCache:{hint:t.cacheKey,inputDependencies:C?["rank","rank","rank"]:["rank","rank"]},getShaderSource:W,getRunData:()=>({outputs:[{dims:i,dataType:a}],dispatchGroup:{x:Math.ceil(d/$/64),y:1,z:1},programUniforms:N})}},ul=(e,t)=>{eh(e.inputs,t),e.compute(th(e.inputs,t))},dl=e=>J({axis:e.axis,blockSize:e.blockSize})});var rh,nh,cl,pl=V(()=>{"use strict";Ke();Q();ae();rh=(e,t,r)=>{let n=e===t,o=e<t&&r<0,i=e>t&&r>0;if(n||o||i)throw new Error("Range these inputs' contents are invalid.")},nh=(e,t,r,n)=>{let o=Math.abs(Math.ceil((t-e)/r)),i=[o],a=o,d=[{type:12,data:a},{type:n,data:e},{type:n,data:r},...R(i)],l=c=>{let m=M("output",n,i.length),u=m.type.value,h=[{name:"outputSize",type:"u32"},{name:"start",type:u},{name:"delta",type:u}];return`
         ${c.registerUniforms(h).declareVariables(m)}
         ${c.mainStart()}
         ${c.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")}
         output[global_idx] = uniforms.start + ${u}(global_idx) * uniforms.delta;
-      }`};return{name:"Range",shaderCache:{hint:`${n}`},getShaderSource:l,getRunData:()=>({outputs:[{dims:i,dataType:n}],dispatchGroup:{x:Math.ceil(a/64)},programUniforms:d})}},cl=e=>{let t=0,r=0,n=0;e.inputs[0].dataType===6?(t=e.inputs[0].getInt32Array()[0],r=e.inputs[1].getInt32Array()[0],n=e.inputs[2].getInt32Array()[0]):e.inputs[0].dataType===1&&(t=e.inputs[0].getFloat32Array()[0],r=e.inputs[1].getFloat32Array()[0],n=e.inputs[2].getFloat32Array()[0]),we.webgpu.validateInputContent&&th(t,r,n),e.compute(rh(t,r,n,e.inputs[0].dataType),{inputs:[]})}});var nh,oh,ih,ah,sh,uh,dh,lh,ch,ph,mh,ml,fh,hh,gh,yh,bh,fl,hl,gl=U(()=>{"use strict";Z();te();Ie();re();nh=(e,t)=>{if(e.every(r=>r>0||(()=>{throw new Error("Resize requires scales input values to be positive")})),e.length>0){if(t.mode==="linear"){if(!(e.length===2||e.length===3||e.length===4&&e[0]===1&&e[1]===1||e.length===4&&e[0]===1&&e[3]===1||e.length===5&&e[0]===1&&e[1]===1))throw new Error(`For linear mode, Resize requires scales to be 2D, 3D, 4D with either two outermost or one innermost and
-            one outermost scale values equal to 1, or 5D with two outermost scale values equal to 1`)}else if(t.mode==="cubic"&&!(e.length===2||e.length===4&&e[0]===1&&e[1]===1||e.length===4&&e[0]===1&&e[3]===1))throw new Error("Resize requires scales input size to be 2 or 4 for cubic mode")}},oh=(e,t,r)=>{t.every(o=>o>=0&&o<r||(()=>{throw new Error("Resize requires axes input values to be positive and less than rank")}));let n=new Array(r).fill(1);return t.forEach((o,i)=>n[o]=e[i]),n},ih=(e,t,r,n,o,i)=>{let[a,d,l]=r>10?[1,2,3]:[-1,e.length>1?1:-1,-1],c=e[0].dims.length;if(a>0&&e.length>a&&e[a].dims.length>0)e[a].getFloat32Array().forEach(m=>i.push(m));else if(t.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(m=>n.push(m)),n.length!==0&&n.length!==c&&r>=18&&n.length!==t.axes.length)throw new Error("Resize requires scales input size to be same as input rank or axes size for opset 18 and up");nh(n,t),t.axes.length>0&&oh(n,t.axes,c).forEach((m,u)=>n[u]=m)}if(l>0&&e.length>l&&e[l].dims.length===1&&e[l].dims[0]>0&&(e[l].getBigInt64Array().forEach(m=>o.push(Number(m))),o.length!==0&&o.length!==c&&r>=18&&o.length!==t.axes.length))throw new Error("Resize requires sizes input size to be same as input rank or axes size for opset 18 and up");if(t.axes.length>0){if(n.length!==0&&n.length!==t.axes.length)throw new Error('Resize requires "scales" input size to be of axes rank when axes attributes is specified');if(o.length!==0&&o.length!==t.axes.length)throw new Error('Resize requires "sizes" input size to be of rank axes rank when axes attributes is specified')}if(typeof n<"u"&&typeof o<"u"&&n.length>0&&o.length>c)throw new Error("Resize requires only of scales or sizes to be specified")},ah=(e,t)=>`fn getOriginalCoordinateFromResizedCoordinate(xResized: u32, xScale: f32, lengthResized: u32,
+      }`};return{name:"Range",shaderCache:{hint:`${n}`},getShaderSource:l,getRunData:()=>({outputs:[{dims:i,dataType:n}],dispatchGroup:{x:Math.ceil(a/64)},programUniforms:d})}},cl=e=>{let t=0,r=0,n=0;e.inputs[0].dataType===6?(t=e.inputs[0].getInt32Array()[0],r=e.inputs[1].getInt32Array()[0],n=e.inputs[2].getInt32Array()[0]):e.inputs[0].dataType===1&&(t=e.inputs[0].getFloat32Array()[0],r=e.inputs[1].getFloat32Array()[0],n=e.inputs[2].getFloat32Array()[0]),_e.webgpu.validateInputContent&&rh(t,r,n),e.compute(nh(t,r,n,e.inputs[0].dataType),{inputs:[]})}});var oh,ih,ah,sh,uh,dh,lh,ch,ph,mh,fh,ml,hh,gh,yh,bh,wh,fl,hl,gl=V(()=>{"use strict";Q();ie();Ie();ae();oh=(e,t)=>{if(e.every(r=>r>0||(()=>{throw new Error("Resize requires scales input values to be positive")})),e.length>0){if(t.mode==="linear"){if(!(e.length===2||e.length===3||e.length===4&&e[0]===1&&e[1]===1||e.length===4&&e[0]===1&&e[3]===1||e.length===5&&e[0]===1&&e[1]===1))throw new Error(`For linear mode, Resize requires scales to be 2D, 3D, 4D with either two outermost or one innermost and
+            one outermost scale values equal to 1, or 5D with two outermost scale values equal to 1`)}else if(t.mode==="cubic"&&!(e.length===2||e.length===4&&e[0]===1&&e[1]===1||e.length===4&&e[0]===1&&e[3]===1))throw new Error("Resize requires scales input size to be 2 or 4 for cubic mode")}},ih=(e,t,r)=>{t.every(o=>o>=0&&o<r||(()=>{throw new Error("Resize requires axes input values to be positive and less than rank")}));let n=new Array(r).fill(1);return t.forEach((o,i)=>n[o]=e[i]),n},ah=(e,t,r,n,o,i)=>{let[a,d,l]=r>10?[1,2,3]:[-1,e.length>1?1:-1,-1],c=e[0].dims.length;if(a>0&&e.length>a&&e[a].dims.length>0)e[a].getFloat32Array().forEach(m=>i.push(m));else if(t.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(m=>n.push(m)),n.length!==0&&n.length!==c&&r>=18&&n.length!==t.axes.length)throw new Error("Resize requires scales input size to be same as input rank or axes size for opset 18 and up");oh(n,t),t.axes.length>0&&ih(n,t.axes,c).forEach((m,u)=>n[u]=m)}if(l>0&&e.length>l&&e[l].dims.length===1&&e[l].dims[0]>0&&(e[l].getBigInt64Array().forEach(m=>o.push(Number(m))),o.length!==0&&o.length!==c&&r>=18&&o.length!==t.axes.length))throw new Error("Resize requires sizes input size to be same as input rank or axes size for opset 18 and up");if(t.axes.length>0){if(n.length!==0&&n.length!==t.axes.length)throw new Error('Resize requires "scales" input size to be of axes rank when axes attributes is specified');if(o.length!==0&&o.length!==t.axes.length)throw new Error('Resize requires "sizes" input size to be of rank axes rank when axes attributes is specified')}if(typeof n<"u"&&typeof o<"u"&&n.length>0&&o.length>c)throw new Error("Resize requires only of scales or sizes to be specified")},sh=(e,t)=>`fn getOriginalCoordinateFromResizedCoordinate(xResized: u32, xScale: f32, lengthResized: u32,
      lengthOriginal: u32, roiStart: f32, roiEnd: f32) -> ${t} { `+(()=>{switch(e){case"asymmetric":return`return ${t}(xResized) / ${t}(xScale);`;case"pytorch_half_pixel":return`if (lengthResized > 1) {
                     return (${t}(xResized) + 0.5) / ${t}(xScale) - 0.5;
                   } else {
@@ -5851,7 +5925,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
                   const adjustment = ${t}(lengthResized) / outputWidth;
                   const center = ${t}(lengthOriginal) / 2;
                   const offset = center * (1 - adjustment);
-                  return offset + ((${t}(xResized) + 0.5) / ${t}(xScale)) - 0.5;`;case"half_pixel":return`return ((${t}(xResized) + 0.5) / ${t}(xScale)) - 0.5;`;default:throw new Error(`Coordinate transform mode ${e} is not supported`)}})()+"}",sh=(e,t,r)=>`fn getNearestPixelFromOriginal(xOriginal: ${r}, isDownSample: bool) -> ${r} {`+(()=>{switch(e){case"round_prefer_ceil":return"if (fract(xOriginal) == 0.5) {             return ceil(xOriginal);           } else {             return round(xOriginal);           }";case"floor":return"return floor(xOriginal);";case"ceil":return"return ceil(xOriginal);";case"round_prefer_floor":return"if (fract(xOriginal) == 0.5) {                     return floor(xOriginal);                   } else {                     return round(xOriginal);                   }";case"simple":default:if(t<11)return"if (isDownSample)                     {                       return ceil(xOriginal);                     } else {                       return xOriginal;                     }";throw new Error(`Nearest mode ${e} is not supported`)}})()+"}",uh=(e,t,r)=>{let n=new Array(r).fill(0).concat(new Array(r).fill(1)),o=e.length===0?n:e.slice();return t.length>0?(t.forEach((i,a)=>{n[i]=o[a],n[a+r]=o[t.length+a]}),n):o},dh=(e,t,r,n)=>{let o=[];if(r.length>0)if(n.length>0){if(e.forEach(i=>o.push(i)),Math.max(...n)>e.length)throw new Error("axes is out of bound");n.forEach((i,a)=>o[i]=r[a])}else r.forEach(i=>o.push(i));else{if(t.length===0)throw new Error("Resize requires either scales or sizes.");o=e.map((i,a)=>Math.round(i*t[a]))}return o},lh=(e,t,r)=>{let n=(()=>{switch(r.keepAspectRatioPolicy){case"not_larger":return r.axes.length>0?Math.min(...r.axes.map(i=>t[i]),Number.MAX_VALUE):Math.min(...t,Number.MAX_VALUE);case"not_smaller":return r.axes.length>0?Math.max(...r.axes.map(i=>t[i]),Number.MIN_VALUE):Math.max(...t,Number.MIN_VALUE);default:throw new Error(`Keep aspect ratio policy ${r.keepAspectRatioPolicy} is not supported`)}})();t.fill(1,0,t.length);let o=e.slice();return r.axes.length>0?(r.axes.forEach(i=>t[i]=n),r.axes.forEach(i=>o[i]=Math.round(e[i]*t[i]))):(t.fill(n,0,t.length),o.forEach((i,a)=>o[a]=Math.round(i*t[a]))),o},ch=(e,t,r,n,o)=>`
+                  return offset + ((${t}(xResized) + 0.5) / ${t}(xScale)) - 0.5;`;case"half_pixel":return`return ((${t}(xResized) + 0.5) / ${t}(xScale)) - 0.5;`;default:throw new Error(`Coordinate transform mode ${e} is not supported`)}})()+"}",uh=(e,t,r)=>`fn getNearestPixelFromOriginal(xOriginal: ${r}, isDownSample: bool) -> ${r} {`+(()=>{switch(e){case"round_prefer_ceil":return"if (fract(xOriginal) == 0.5) {             return ceil(xOriginal);           } else {             return round(xOriginal);           }";case"floor":return"return floor(xOriginal);";case"ceil":return"return ceil(xOriginal);";case"round_prefer_floor":return"if (fract(xOriginal) == 0.5) {                     return floor(xOriginal);                   } else {                     return round(xOriginal);                   }";case"simple":default:if(t<11)return"if (isDownSample)                     {                       return ceil(xOriginal);                     } else {                       return xOriginal;                     }";throw new Error(`Nearest mode ${e} is not supported`)}})()+"}",dh=(e,t,r)=>{let n=new Array(r).fill(0).concat(new Array(r).fill(1)),o=e.length===0?n:e.slice();return t.length>0?(t.forEach((i,a)=>{n[i]=o[a],n[a+r]=o[t.length+a]}),n):o},lh=(e,t,r,n)=>{let o=[];if(r.length>0)if(n.length>0){if(e.forEach(i=>o.push(i)),Math.max(...n)>e.length)throw new Error("axes is out of bound");n.forEach((i,a)=>o[i]=r[a])}else r.forEach(i=>o.push(i));else{if(t.length===0)throw new Error("Resize requires either scales or sizes.");o=e.map((i,a)=>Math.round(i*t[a]))}return o},ch=(e,t,r)=>{let n=(()=>{switch(r.keepAspectRatioPolicy){case"not_larger":return r.axes.length>0?Math.min(...r.axes.map(i=>t[i]),Number.MAX_VALUE):Math.min(...t,Number.MAX_VALUE);case"not_smaller":return r.axes.length>0?Math.max(...r.axes.map(i=>t[i]),Number.MIN_VALUE):Math.max(...t,Number.MIN_VALUE);default:throw new Error(`Keep aspect ratio policy ${r.keepAspectRatioPolicy} is not supported`)}})();t.fill(1,0,t.length);let o=e.slice();return r.axes.length>0?(r.axes.forEach(i=>t[i]=n),r.axes.forEach(i=>o[i]=Math.round(e[i]*t[i]))):(t.fill(n,0,t.length),o.forEach((i,a)=>o[a]=Math.round(i*t[a]))),o},ph=(e,t,r,n,o)=>`
     fn calculateOriginalIndicesFromOutputIndices(output_indices: ${e.type.indices}) -> array<${e.type.value}, ${r.length}> {
       var original_indices: array<${e.type.value}, ${r.length}>;
       for (var i:u32 = 0; i < ${r.length}; i++) {
@@ -5869,7 +5943,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
         }
       }
       return original_indices;
-    }`,ph=(e,t,r,n,o,i,a)=>`
+    }`,mh=(e,t,r,n,o,i,a)=>`
     fn calculateInputIndicesFromOutputIndices(output_indices: ${t.type.indices}) -> ${e.type.indices} {
       var input_indices: ${e.type.indices};
       for (var i:u32 = 0; i < ${n.length}; i++) {
@@ -5900,7 +5974,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
         ${e.indicesSet("input_indices","i"," input_index")}
       }
       return input_indices;
-    }`,mh=(e,t)=>`
+    }`,fh=(e,t)=>`
     fn checkInputIndices(input_indices: ${e.type.indices}) -> bool {
       for (var i:u32 = 0; i < ${t.length}; i++) {
         var input_index = ${e.indicesGet("input_indices","i")};
@@ -5912,7 +5986,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
     }`,ml=(e,t,r,n)=>e.rank>n?`
     ${e.indicesSet("input_indices",t,"channel")};
     ${e.indicesSet("input_indices",r,"batch")};
-`:"",fh=(e,t,r,n,o)=>{let[a,d,l,c]=r.length===2?[-1,0,1,-1]:[0,2,3,1],m=e.type.value;return`
+`:"",hh=(e,t,r,n,o)=>{let[a,d,l,c]=r.length===2?[-1,0,1,-1]:[0,2,3,1],m=e.type.value;return`
     fn getInputValue(batch: u32, channel: u32, row: u32, col: u32) -> ${m} {
       var input_indices: ${e.type.indices};
       ${e.indicesSet("input_indices",d,`max(0, min(row, ${r[d]} - 1))`)};
@@ -5953,7 +6027,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
         dy2 = 0.5;
       }
       return (x11 * dx2 * dy2 + x12 * dx2 * dy1 + x21 * dx1 * dy2 + x22 * dx1 * dy1);
-    }`},hh=(e,t,r,n,o,i,a,d,l,c)=>{let m=r.length===2,u=!0,[h,w]=m?[0,1]:u?[2,3]:[1,2],g=e.type.value,y=S=>{let $=S===h?"row":"col";return`
+    }`},gh=(e,t,r,n,o,i,a,d,l,c)=>{let m=r.length===2,u=!0,[h,w]=m?[0,1]:u?[2,3]:[1,2],g=e.type.value,y=S=>{let $=S===h?"row":"col";return`
       fn ${$}CubicInterpolation(input_indices: ${e.type.indices}, output_indices: ${t.type.indices}) -> ${g} {
         var output_index = ${t.indicesGet("output_indices",S)};
         var originalIdx: ${g} = getOriginalCoordinateFromResizedCoordinate(output_index, ${o[S]},
@@ -6001,7 +6075,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
     var input_indices: ${e.type.indices} = output_indices;
     return colCubicInterpolation(input_indices, output_indices);
   }
-    `},gh=(e,t,r,n,o)=>{let[a,d,l,c,m]=r.length===3?[-1,0,1,2,-1]:[0,2,3,4,1],u=e.type.value;return`
+    `},yh=(e,t,r,n,o)=>{let[a,d,l,c,m]=r.length===3?[-1,0,1,2,-1]:[0,2,3,4,1],u=e.type.value;return`
     fn getInputValue(batch: u32, channel: u32, depth:u32, height: u32, width: u32) -> ${u} {
       var input_indices: ${e.type.indices};
       ${e.indicesSet("input_indices",d,`max(0, min(depth, ${r[d]} - 1))`)};
@@ -6060,23 +6134,23 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
       }
       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);
-    }`},yh=(e,t,r,n,o,i)=>{let a=e.dims,d=uh(i,t.axes,a.length),l=dh(a,n,o,t.axes),c=n.slice();n.length===0&&(c=a.map((_,x)=>_===0?1:l[x]/_),t.keepAspectRatioPolicy!=="stretch"&&(l=lh(a,c,t)));let m=M("output",e.dataType,l.length),u=k("input",e.dataType,a.length),h=E.size(l),w=a.length===l.length&&a.every((_,x)=>_===l[x]),g=t.coordinateTransformMode==="tf_crop_and_resize",y=t.extrapolationValue,S=u.type.value,$=_=>`
+    }`},bh=(e,t,r,n,o,i)=>{let a=e.dims,d=dh(i,t.axes,a.length),l=lh(a,n,o,t.axes),c=n.slice();n.length===0&&(c=a.map((v,x)=>v===0?1:l[x]/v),t.keepAspectRatioPolicy!=="stretch"&&(l=ch(a,c,t)));let m=M("output",e.dataType,l.length),u=E("input",e.dataType,a.length),h=k.size(l),w=a.length===l.length&&a.every((v,x)=>v===l[x]),g=t.coordinateTransformMode==="tf_crop_and_resize",y=t.extrapolationValue,S=u.type.value,$=v=>`
       ${w?"":`
-      ${ah(t.coordinateTransformMode,S)};
+      ${sh(t.coordinateTransformMode,S)};
       ${(()=>{switch(t.mode){case"nearest":return`
-              ${mh(u,a)};
-              ${sh(t.nearestMode,r,S)};
-              ${ph(u,m,a,l,c.length,d.length,g)};
+              ${fh(u,a)};
+              ${uh(t.nearestMode,r,S)};
+              ${mh(u,m,a,l,c.length,d.length,g)};
               `;case"linear":return`
-              ${ch(m,a,l,c.length,d.length)};
-              ${(()=>{if(a.length===2||a.length===4)return`${fh(u,m,a,g,y)}`;if(a.length===3||a.length===5)return`${gh(u,m,a,g,y)}`;throw Error("Linear mode only supports input dims 2, 3, 4 and 5 are supported in linear mode.")})()};
+              ${ph(m,a,l,c.length,d.length)};
+              ${(()=>{if(a.length===2||a.length===4)return`${hh(u,m,a,g,y)}`;if(a.length===3||a.length===5)return`${yh(u,m,a,g,y)}`;throw Error("Linear mode only supports input dims 2, 3, 4 and 5 are supported in linear mode.")})()};
             `;case"cubic":return`
-            ${(()=>{if(a.length===2||a.length===4)return`${hh(u,m,a,l,c,d,t.cubicCoeffA,g,t.extrapolationValue,t.excludeOutside)}`;throw Error("Cubic mode only supports input dims 2 and 4 are supported in linear mode.")})()};
+            ${(()=>{if(a.length===2||a.length===4)return`${gh(u,m,a,l,c,d,t.cubicCoeffA,g,t.extrapolationValue,t.excludeOutside)}`;throw Error("Cubic mode only supports input dims 2 and 4 are supported in linear mode.")})()};
             `;default:throw Error("Invalid resize mode")}})()};
       `}
-      ${_.registerUniform("output_size","u32").registerUniform("scales","f32",c.length).registerUniform("roi","f32",d.length).declareVariables(u,m)}
-      ${_.mainStart()}
-        ${_.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")}
+      ${v.registerUniform("output_size","u32").registerUniform("scales","f32",c.length).registerUniform("roi","f32",d.length).declareVariables(u,m)}
+      ${v.mainStart()}
+        ${v.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.output_size")}
         ${w?"output[global_idx] = input[global_idx];":`
         let output_indices = ${m.offsetToIndices("global_idx")};
         var input_indices: ${u.type.indices};
@@ -6087,8 +6161,8 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
                   output[global_idx] = ${t.extrapolationValue};
                 }`;case"linear":return`output[global_idx] = ${a.length===2||a.length===4?"bilinearInterpolation":"trilinearInterpolation"}(output_indices);`;case"cubic":return"output[global_idx] = bicubicInterpolation(output_indices);";default:throw Error(`Unsupported resize mode: ${t.mode}`)}})()};
 `}
-      }`;return{name:"Resize",shaderCache:{hint:`${t.cacheKey}|${r}|${c.length>0?c:""}|${o.length>0?o:""}|${d.length>0?d:""}|${w}|${a}`,inputDependencies:["rank"]},getShaderSource:$,getRunData:()=>({outputs:[{dims:l,dataType:e.dataType}],dispatchGroup:{x:Math.ceil(h/64)},programUniforms:[{type:12,data:h},{type:1,data:c},{type:1,data:d},...V(a,l)]})}},bh=e=>{let t=e.customDataBuffer;return new Uint32Array(t,t.byteOffset,1)[0]},fl=(e,t)=>{let r=[],n=[],o=[],i=bh(e);if(t.antialias!==0)throw Error("Only default value (0) for Antialias attribute is supported");ih(e.inputs,t,i,r,n,o),e.compute(yh(e.inputs[0],t,i,r,n,o),{inputs:[0]})},hl=e=>{let t=e.antialias,r=e.axes,n=e.coordinateTransformMode,o=e.cubicCoeffA,i=e.excludeOutside!==0,a=e.extrapolationValue,d=e.keepAspectRatioPolicy,l=e.mode,c=e.nearestMode===""?"simple":e.nearestMode;return J({antialias:t,axes:r,coordinateTransformMode:n,cubicCoeffA:o,excludeOutside:i,extrapolationValue:a,keepAspectRatioPolicy:d,mode:l,nearestMode:c})}});var wh,vh,yl,bl=U(()=>{"use strict";Z();te();Ie();re();wh=(e,t)=>{let[r,n,o,i]=e,{numHeads:a,rotaryEmbeddingDim:d}=t;if(r.dims.length!==3&&r.dims.length!==4)throw new Error(`Input 'x' is expected to have 3 or 4 dimensions, got ${r.dims.length}`);if(!E.areEqual(n.dims,[])&&!E.areEqual(n.dims,[1])&&n.dims.length!==2)throw new Error(`Input 'position_ids' is expected to have 0, 1, or 2 dimensions, got ${n.dims.length}`);if(o.dims.length!==2)throw new Error(`Input 'cos_cache' is expected to have 2 dimensions, got ${o.dims.length}`);if(i.dims.length!==2)throw new Error(`Input 'sin_cache' is expected to have 2 dimensions, got ${i.dims.length}`);if(!E.areEqual(o.dims,i.dims))throw new Error("Inputs 'cos_cache' and 'sin_cache' are expected to have the same shape");if(d>0&&a===0)throw new Error("num_heads must be provided if rotary_embedding_dim is specified");let l=r.dims[0],c=r.dims[r.dims.length-2],m=o.dims[0],u=E.sizeFromDimension(r.dims,1)/c,h=d===0?o.dims[1]*2:u/a;if(d>h)throw new Error("rotary_embedding_dim must be less than or equal to head_size");if(n.dims.length===2){if(l!==n.dims[0])throw new Error(`Input 'position_ids' dimension 0 should be of size batch_size, got ${n.dims[0]}`);if(c!==n.dims[1])throw new Error(`Input 'position_ids' dimension 1 should be of size sequence_length, got ${n.dims[1]}`)}if(h/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(c>m)throw new Error("Updating cos_cache and sin_cache in RotaryEmbedding is not currently supported")},vh=(e,t)=>{let{interleaved:r,numHeads:n,rotaryEmbeddingDim:o,scale:i}=t,a=e[0].dims[0],d=E.sizeFromDimension(e[0].dims,1),l=e[0].dims[e[0].dims.length-2],c=d/l,m=e[2].dims[1],u=o===0?m*2:c/n,h=new Array(a,l,c/u,u-m),w=E.computeStrides(h),g=[{type:1,data:i},{type:12,data:h},{type:12,data:w},...e[0].dims.length===3?new Array({type:12,data:[d,c,u,1]}):[],...e[0].dims.length===4?new Array({type:12,data:[d,u,l*u,1]}):[],...V(e[0].dims,e[1].dims,e[2].dims,e[3].dims,e[0].dims)],y=S=>{let $=k("input",e[0].dataType,e[0].dims.length),_=k("position_ids",e[1].dataType,e[1].dims.length),x=k("cos_cache",e[2].dataType,e[2].dims.length),T=k("sin_cache",e[3].dataType,e[3].dims.length),C=M("output",e[0].dataType,e[0].dims.length);return S.registerUniforms([{name:"scale",type:"f32"},{name:"global_shape",type:"u32",length:h.length},{name:"global_strides",type:"u32",length:w.length},{name:"input_output_strides",type:"u32",length:w.length}]),`
-        ${S.declareVariables($,_,x,T,C)}
+      }`;return{name:"Resize",shaderCache:{hint:`${t.cacheKey}|${r}|${c.length>0?c:""}|${o.length>0?o:""}|${d.length>0?d:""}|${w}|${a}`,inputDependencies:["rank"]},getShaderSource:$,getRunData:()=>({outputs:[{dims:l,dataType:e.dataType}],dispatchGroup:{x:Math.ceil(h/64)},programUniforms:[{type:12,data:h},{type:1,data:c},{type:1,data:d},...R(a,l)]})}},wh=e=>{let t=e.customDataBuffer;return new Uint32Array(t,t.byteOffset,1)[0]},fl=(e,t)=>{let r=[],n=[],o=[],i=wh(e);if(t.antialias!==0)throw Error("Only default value (0) for Antialias attribute is supported");ah(e.inputs,t,i,r,n,o),e.compute(bh(e.inputs[0],t,i,r,n,o),{inputs:[0]})},hl=e=>{let t=e.antialias,r=e.axes,n=e.coordinateTransformMode,o=e.cubicCoeffA,i=e.excludeOutside!==0,a=e.extrapolationValue,d=e.keepAspectRatioPolicy,l=e.mode,c=e.nearestMode===""?"simple":e.nearestMode;return J({antialias:t,axes:r,coordinateTransformMode:n,cubicCoeffA:o,excludeOutside:i,extrapolationValue:a,keepAspectRatioPolicy:d,mode:l,nearestMode:c})}});var _h,vh,yl,bl=V(()=>{"use strict";Q();ie();Ie();ae();_h=(e,t)=>{let[r,n,o,i]=e,{numHeads:a,rotaryEmbeddingDim:d}=t;if(r.dims.length!==3&&r.dims.length!==4)throw new Error(`Input 'x' is expected to have 3 or 4 dimensions, got ${r.dims.length}`);if(!k.areEqual(n.dims,[])&&!k.areEqual(n.dims,[1])&&n.dims.length!==2)throw new Error(`Input 'position_ids' is expected to have 0, 1, or 2 dimensions, got ${n.dims.length}`);if(o.dims.length!==2)throw new Error(`Input 'cos_cache' is expected to have 2 dimensions, got ${o.dims.length}`);if(i.dims.length!==2)throw new Error(`Input 'sin_cache' is expected to have 2 dimensions, got ${i.dims.length}`);if(!k.areEqual(o.dims,i.dims))throw new Error("Inputs 'cos_cache' and 'sin_cache' are expected to have the same shape");if(d>0&&a===0)throw new Error("num_heads must be provided if rotary_embedding_dim is specified");let l=r.dims[0],c=r.dims[r.dims.length-2],m=o.dims[0],u=k.sizeFromDimension(r.dims,1)/c,h=d===0?o.dims[1]*2:u/a;if(d>h)throw new Error("rotary_embedding_dim must be less than or equal to head_size");if(n.dims.length===2){if(l!==n.dims[0])throw new Error(`Input 'position_ids' dimension 0 should be of size batch_size, got ${n.dims[0]}`);if(c!==n.dims[1])throw new Error(`Input 'position_ids' dimension 1 should be of size sequence_length, got ${n.dims[1]}`)}if(h/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(c>m)throw new Error("Updating cos_cache and sin_cache in RotaryEmbedding is not currently supported")},vh=(e,t)=>{let{interleaved:r,numHeads:n,rotaryEmbeddingDim:o,scale:i}=t,a=e[0].dims[0],d=k.sizeFromDimension(e[0].dims,1),l=e[0].dims[e[0].dims.length-2],c=d/l,m=e[2].dims[1],u=o===0?m*2:c/n,h=new Array(a,l,c/u,u-m),w=k.computeStrides(h),g=[{type:1,data:i},{type:12,data:h},{type:12,data:w},...e[0].dims.length===3?new Array({type:12,data:[d,c,u,1]}):[],...e[0].dims.length===4?new Array({type:12,data:[d,u,l*u,1]}):[],...R(e[0].dims,e[1].dims,e[2].dims,e[3].dims,e[0].dims)],y=S=>{let $=E("input",e[0].dataType,e[0].dims.length),v=E("position_ids",e[1].dataType,e[1].dims.length),x=E("cos_cache",e[2].dataType,e[2].dims.length),T=E("sin_cache",e[3].dataType,e[3].dims.length),C=M("output",e[0].dataType,e[0].dims.length);return S.registerUniforms([{name:"scale",type:"f32"},{name:"global_shape",type:"u32",length:h.length},{name:"global_strides",type:"u32",length:w.length},{name:"input_output_strides",type:"u32",length:w.length}]),`
+        ${S.declareVariables($,v,x,T,C)}
 
         ${S.mainStart(At)}
           let half_rotary_emb_dim = uniforms.${x.name}_shape[1];
@@ -6098,9 +6172,9 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
 
           if (bsnh[3] < half_rotary_emb_dim) {
             let position_ids_idx =
-                ${_.broadcastedIndicesToOffset("bsnh.xy",M("",_.type.tensor,2))};
+                ${v.broadcastedIndicesToOffset("bsnh.xy",M("",v.type.tensor,2))};
             let position_id =
-                u32(${_.getByOffset("position_ids_idx")}) + select(0, bsnh[1], position_ids_idx == 0);
+                u32(${v.getByOffset("position_ids_idx")}) + select(0, bsnh[1], position_ids_idx == 0);
             let i = dot(bsnh, uniforms.input_output_strides) + select(0, bsnh[3], ${r});
             let j = i + select(half_rotary_emb_dim, 1, ${r});
             let re = ${$.getByOffset("i")} * ${x.get("position_id","bsnh[3]")} -
@@ -6113,7 +6187,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
             let k = dot(bsnh, uniforms.input_output_strides) + half_rotary_emb_dim;
             ${C.setByOffset("k",$.getByOffset("k"))}
           }
-        }`};return{name:"RotaryEmbedding",shaderCache:{hint:J({interleaved:r}).cacheKey,inputDependencies:["rank","rank","rank","rank"]},getShaderSource:y,getRunData:()=>({outputs:[{dims:e[0].dims,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(E.size(h)/At)},programUniforms:g})}},yl=(e,t)=>{wh(e.inputs,t),e.compute(vh(e.inputs,t))}});var _h,$h,wl,vl=U(()=>{"use strict";Z();te();re();_h=e=>{if(!e||e.length<3)throw new Error("layerNorm requires at least 3 inputs.");let t=e[0],r=e[1],n=e[2];if(t.dataType!==r.dataType||t.dataType!==n.dataType)throw new Error("All inputs must have the same data type");if(t.dims.length!==3&&t.dims.length!==2)throw new Error("Input must be 2D or 3D");if(r.dims.length!==3&&r.dims.length!==2)throw new Error("Skip must be 2D or 3D");let o=t.dims[t.dims.length-1],i=t.dims[t.dims.length-2];if(r.dims[r.dims.length-1]!==o)throw new Error("Skip must have the same hidden size as input");if(r.dims[r.dims.length-2]!==i)throw new Error("Skip must have the same sequence length as input");if(n.dims.length!==1)throw new Error("Gamma must be 1D");if(n.dims[n.dims.length-1]!==o)throw new Error("Gamma must have the same hidden size as input");if(e.length>3){let a=e[3];if(a.dims.length!==1)throw new Error("Beta must be 1D");if(a.dims[a.dims.length-1]!==o)throw new Error("Beta must have the same hidden size as input")}if(e.length>4){let a=e[4];if(a.dims.length!==1)throw new Error("Bias must be 1D");if(a.dims[a.dims.length-1]!==o)throw new Error("Bias must have the same hidden size as input")}},$h=(e,t,r,n)=>{let o=t.simplified,i=e[0].dims,a=E.size(i),d=i,l=a,c=i.slice(-1)[0],m=n?i.slice(0,-1).concat(1):[],u=!o&&e.length>3,h=e.length>4,w=n&&r>1,g=n&&r>2,y=r>3,S=64,$=ve(c),_=[{type:12,data:l},{type:12,data:$},{type:12,data:c},{type:1,data:t.epsilon}],x=C=>{let A=[{name:"output_size",type:"u32"},{name:"components",type:"u32"},{name:"hidden_size",type:"u32"},{name:"epsilon",type:"f32"}],P=[k("x",e[0].dataType,e[0].dims,$),k("skip",e[1].dataType,e[1].dims,$),k("gamma",e[2].dataType,e[2].dims,$)];u&&P.push(k("beta",e[3].dataType,e[3].dims,$)),h&&P.push(k("bias",e[4].dataType,e[4].dims,$)),P.push(M("output",e[0].dataType,d,$)),w&&P.push(M("mean_output",1,m)),g&&P.push(M("inv_std_output",1,m)),y&&P.push(M("input_skip_bias_sum",e[0].dataType,d,$));let D=ye(e[0].dataType),N=ye(1,$);return`
+        }`};return{name:"RotaryEmbedding",shaderCache:{hint:J({interleaved:r}).cacheKey,inputDependencies:["rank","rank","rank","rank"]},getShaderSource:y,getRunData:()=>({outputs:[{dims:e[0].dims,dataType:e[0].dataType}],dispatchGroup:{x:Math.ceil(k.size(h)/At)},programUniforms:g})}},yl=(e,t)=>{_h(e.inputs,t),e.compute(vh(e.inputs,t))}});var $h,xh,wl,_l=V(()=>{"use strict";Q();ie();ae();$h=e=>{if(!e||e.length<3)throw new Error("layerNorm requires at least 3 inputs.");let t=e[0],r=e[1],n=e[2];if(t.dataType!==r.dataType||t.dataType!==n.dataType)throw new Error("All inputs must have the same data type");if(t.dims.length!==3&&t.dims.length!==2)throw new Error("Input must be 2D or 3D");if(r.dims.length!==3&&r.dims.length!==2)throw new Error("Skip must be 2D or 3D");let o=t.dims[t.dims.length-1],i=t.dims[t.dims.length-2];if(r.dims[r.dims.length-1]!==o)throw new Error("Skip must have the same hidden size as input");if(r.dims[r.dims.length-2]!==i)throw new Error("Skip must have the same sequence length as input");if(n.dims.length!==1)throw new Error("Gamma must be 1D");if(n.dims[n.dims.length-1]!==o)throw new Error("Gamma must have the same hidden size as input");if(e.length>3){let a=e[3];if(a.dims.length!==1)throw new Error("Beta must be 1D");if(a.dims[a.dims.length-1]!==o)throw new Error("Beta must have the same hidden size as input")}if(e.length>4){let a=e[4];if(a.dims.length!==1)throw new Error("Bias must be 1D");if(a.dims[a.dims.length-1]!==o)throw new Error("Bias must have the same hidden size as input")}},xh=(e,t,r,n)=>{let o=t.simplified,i=e[0].dims,a=k.size(i),d=i,l=a,c=i.slice(-1)[0],m=n?i.slice(0,-1).concat(1):[],u=!o&&e.length>3,h=e.length>4,w=n&&r>1,g=n&&r>2,y=r>3,S=64,$=we(c),v=[{type:12,data:l},{type:12,data:$},{type:12,data:c},{type:1,data:t.epsilon}],x=C=>{let A=[{name:"output_size",type:"u32"},{name:"components",type:"u32"},{name:"hidden_size",type:"u32"},{name:"epsilon",type:"f32"}],P=[E("x",e[0].dataType,e[0].dims,$),E("skip",e[1].dataType,e[1].dims,$),E("gamma",e[2].dataType,e[2].dims,$)];u&&P.push(E("beta",e[3].dataType,e[3].dims,$)),h&&P.push(E("bias",e[4].dataType,e[4].dims,$)),P.push(M("output",e[0].dataType,d,$)),w&&P.push(M("mean_output",1,m)),g&&P.push(M("inv_std_output",1,m)),y&&P.push(M("input_skip_bias_sum",e[0].dataType,d,$));let B=he(e[0].dataType),N=he(1,$);return`
 
       ${C.registerUniforms(A).declareVariables(...P)}
       var<workgroup> sum_shared : array<${N}, ${S}>;
@@ -6132,12 +6206,12 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
         }
         for (var i: u32 = 0; i < stride; i++) {
           let skip_value = skip[offset + i];
-          let bias_value = ${h?"bias[offset1d + i]":D+"(0.0)"};
+          let bias_value = ${h?"bias[offset1d + i]":B+"(0.0)"};
           let input_value = x[offset + i];
           let value = input_value + skip_value + bias_value;
           ${y?"input_skip_bias_sum[offset + i] = value;":""}
           output[offset + i] = value;
-          let f32_value = ${Et(D,$,"value")};
+          let f32_value = ${kt(B,$,"value")};
           sum_shared[ix] += f32_value;
           sum_squared_shared[ix] += f32_value * f32_value;
         }
@@ -6161,11 +6235,11 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
         ${g?"inv_std_output[global_idx] = inv_std_dev;":""}
 
         for (var i: u32 = 0; i < stride; i++) {
-          output[offset + i] = (output[offset + i] ${o?"":`- ${D}(mean)`}) *
-            ${D}(inv_std_dev) * gamma[offset1d + i]
+          output[offset + i] = (output[offset + i] ${o?"":`- ${B}(mean)`}) *
+            ${B}(inv_std_dev) * gamma[offset1d + i]
             ${u?"+ beta[offset1d + i]":""};
         }
-      }`},T=[{dims:d,dataType:e[0].dataType}];return r>1&&T.push({dims:m,dataType:1}),r>2&&T.push({dims:m,dataType:1}),r>3&&T.push({dims:i,dataType:e[0].dataType}),{name:"SkipLayerNormalization",shaderCache:{hint:`${$};${w};${g};${y}`,inputDependencies:e.map((C,A)=>"type")},getShaderSource:x,getRunData:()=>({outputs:T,dispatchGroup:{x:Math.ceil(l/c)},programUniforms:_})}},wl=(e,t)=>{_h(e.inputs);let n=[0];e.outputCount>1&&n.push(-3),e.outputCount>2&&n.push(-3),e.outputCount>3&&n.push(3),e.compute($h(e.inputs,t,e.outputCount,!1),{outputs:n})}});var xh,on,Sh,_l,Th,Ih,$l,xl,Sl=U(()=>{"use strict";Z();te();Ie();re();xh=(e,t)=>{if(!e||e.length<1)throw new Error("too few inputs");if(t.axes.length!==0){if(t.axes.length!==t.starts.length||t.axes.length!==t.ends.length)throw new Error("axes, starts and ends must have the same length")}else if(t.starts.length!==t.ends.length)throw new Error("starts and ends must have the same length");e.slice(1).forEach((r,n)=>{if(e[n+1].dataType!==6&&e[n+1].dataType!==7)throw new Error(`Input ${n} must be an array of int32 or int64`)})},on=(e,t)=>{let r=[];if(e.length>t)if(e[t].dataType===7)e[t].getBigInt64Array().forEach(n=>r.push(Number(n)));else if(e[t].dataType===6)e[t].getInt32Array().forEach(n=>r.push(Number(n)));else throw new Error(`Input ${t} must be an array of int32 or int64`);return r},Sh=(e,t)=>{if(e.length>1){let r=on(e,1),n=on(e,2),o=on(e,3);return o.length===0&&(o=[...Array(e[0].dims.length).keys()]),J({starts:r,ends:n,axes:o})}else return t},_l=(e,t,r,n,o)=>{let i=e;return e<0&&(i+=r[n[t]]),o[t]<0?Math.max(0,Math.min(i,r[n[t]]-1)):Math.max(0,Math.min(i,r[n[t]]))},Th=(e,t,r)=>`fn calculateInputIndices(output_indices: ${t.type.indices}) -> ${e.type.indices} {
+      }`},T=[{dims:d,dataType:e[0].dataType}];return r>1&&T.push({dims:m,dataType:1}),r>2&&T.push({dims:m,dataType:1}),r>3&&T.push({dims:i,dataType:e[0].dataType}),{name:"SkipLayerNormalization",shaderCache:{hint:`${$};${w};${g};${y}`,inputDependencies:e.map((C,A)=>"type")},getShaderSource:x,getRunData:()=>({outputs:T,dispatchGroup:{x:Math.ceil(l/c)},programUniforms:v})}},wl=(e,t)=>{$h(e.inputs);let n=[0];e.outputCount>1&&n.push(-3),e.outputCount>2&&n.push(-3),e.outputCount>3&&n.push(3),e.compute(xh(e.inputs,t,e.outputCount,!1),{outputs:n})}});var Sh,on,Th,vl,Ih,Ch,$l,xl,Sl=V(()=>{"use strict";Q();ie();Ie();ae();Sh=(e,t)=>{if(!e||e.length<1)throw new Error("too few inputs");if(t.axes.length!==0){if(t.axes.length!==t.starts.length||t.axes.length!==t.ends.length)throw new Error("axes, starts and ends must have the same length")}else if(t.starts.length!==t.ends.length)throw new Error("starts and ends must have the same length");e.slice(1).forEach((r,n)=>{if(e[n+1].dataType!==6&&e[n+1].dataType!==7)throw new Error(`Input ${n} must be an array of int32 or int64`)})},on=(e,t)=>{let r=[];if(e.length>t)if(e[t].dataType===7)e[t].getBigInt64Array().forEach(n=>r.push(Number(n)));else if(e[t].dataType===6)e[t].getInt32Array().forEach(n=>r.push(Number(n)));else throw new Error(`Input ${t} must be an array of int32 or int64`);return r},Th=(e,t)=>{if(e.length>1){let r=on(e,1),n=on(e,2),o=on(e,3);return o.length===0&&(o=[...Array(e[0].dims.length).keys()]),J({starts:r,ends:n,axes:o})}else return t},vl=(e,t,r,n,o)=>{let i=e;return e<0&&(i+=r[n[t]]),o[t]<0?Math.max(0,Math.min(i,r[n[t]]-1)):Math.max(0,Math.min(i,r[n[t]]))},Ih=(e,t,r)=>`fn calculateInputIndices(output_indices: ${t.type.indices}) -> ${e.type.indices} {
           var input_indices: ${e.type.indices};
           var carry = 0u;
           for (var i = ${r.length}; i >= 0; i--) {
@@ -6183,15 +6257,15 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
             ${e.indicesSet("input_indices","i","input_index")};
           }
           return input_indices;
-      }`,Ih=(e,t)=>{let r=e[0].dims,n=E.size(r),o=t.axes.length>0?E.normalizeAxes(t.axes,r.length):[...Array(r.length).keys()],i=on(e,4);i.forEach($=>$!==0||(()=>{throw new Error("step cannot be 0")})),i.length===0&&(i=Array(o.length).fill(1));let a=t.starts.map(($,_)=>_l($,_,r,o,i)),d=t.ends.map(($,_)=>_l($,_,r,o,i));if(o.length!==a.length||o.length!==d.length)throw new Error("start, ends and axes should have the same number of elements");if(o.length!==r.length)for(let $=0;$<r.length;++$)o.includes($)||(a.splice($,0,0),d.splice($,0,r[$]),i.splice($,0,1));let l=i.map($=>Math.sign($));i.forEach(($,_,x)=>{if($<0){let T=(d[_]-a[_])/$,C=a[_],A=C+T*i[_];a[_]=A,d[_]=C,x[_]=-$}});let c=r.slice(0);o.forEach(($,_)=>{c[$]=Math.ceil((d[$]-a[$])/i[$])});let m={dims:c,dataType:e[0].dataType},u=M("output",e[0].dataType,c.length),h=k("input",e[0].dataType,e[0].dims.length),w=E.size(c),g=[{name:"outputSize",type:"u32"},{name:"starts",type:"u32",length:a.length},{name:"signs",type:"i32",length:l.length},{name:"steps",type:"u32",length:i.length}],y=[{type:12,data:w},{type:12,data:a},{type:6,data:l},{type:12,data:i},...V(e[0].dims,c)],S=$=>`
+      }`,Ch=(e,t)=>{let r=e[0].dims,n=k.size(r),o=t.axes.length>0?k.normalizeAxes(t.axes,r.length):[...Array(r.length).keys()],i=on(e,4);i.forEach($=>$!==0||(()=>{throw new Error("step cannot be 0")})),i.length===0&&(i=Array(o.length).fill(1));let a=t.starts.map(($,v)=>vl($,v,r,o,i)),d=t.ends.map(($,v)=>vl($,v,r,o,i));if(o.length!==a.length||o.length!==d.length)throw new Error("start, ends and axes should have the same number of elements");if(o.length!==r.length)for(let $=0;$<r.length;++$)o.includes($)||(a.splice($,0,0),d.splice($,0,r[$]),i.splice($,0,1));let l=i.map($=>Math.sign($));i.forEach(($,v,x)=>{if($<0){let T=(d[v]-a[v])/$,C=a[v],A=C+T*i[v];a[v]=A,d[v]=C,x[v]=-$}});let c=r.slice(0);o.forEach(($,v)=>{c[$]=Math.ceil((d[$]-a[$])/i[$])});let m={dims:c,dataType:e[0].dataType},u=M("output",e[0].dataType,c.length),h=E("input",e[0].dataType,e[0].dims.length),w=k.size(c),g=[{name:"outputSize",type:"u32"},{name:"starts",type:"u32",length:a.length},{name:"signs",type:"i32",length:l.length},{name:"steps",type:"u32",length:i.length}],y=[{type:12,data:w},{type:12,data:a},{type:6,data:l},{type:12,data:i},...R(e[0].dims,c)],S=$=>`
       ${$.registerUniforms(g).declareVariables(h,u)}
-        ${Th(h,u,r)}
+        ${Ih(h,u,r)}
         ${$.mainStart()}
           ${$.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.outputSize")}
           let output_indices = ${u.offsetToIndices("global_idx")};
           let input_indices = calculateInputIndices(output_indices);
           ${u.setByOffset("global_idx",h.getByIndices("input_indices"))}
-      }`;return{name:"Slice",shaderCache:{hint:`${l.length}_${a.length}_${i.length}`,inputDependencies:["rank"]},getShaderSource:S,getRunData:()=>({outputs:[m],dispatchGroup:{x:Math.ceil(n/64)},programUniforms:y})}},$l=(e,t)=>{xh(e.inputs,t);let r=Sh(e.inputs,t);e.compute(Ih(e.inputs,r),{inputs:[0]})},xl=e=>{let t=e.starts,r=e.ends,n=e.axes;return J({starts:t,ends:r,axes:n})}});var Ch,Ah,Tl,Il,Cl=U(()=>{"use strict";Z();te();Ie();lt();re();Ch=e=>{if(!e||e.length!==1)throw new Error("Softmax op requires 1 input.")},Ah=(e,t)=>{let r=e.inputs[0],n=r.dims,o=E.size(n),i=64,a=n.length,d=E.normalizeAxis(t.axis,a),l=d<n.length-1,c,m=[];l?(m=Array.from({length:a},(P,D)=>D),m[d]=a-1,m[a-1]=d,c=e.compute(Pe(r,m),{inputs:[r],outputs:[-1]})[0]):c=r;let u=c.dims,h=u[a-1],w=o/h,g=ve(h),y=h/g,S=(P,D)=>D===4?`max(max(${P}.x, ${P}.y), max(${P}.z, ${P}.w))`:D===2?`max(${P}.x, ${P}.y)`:D===3?`max(max(${P}.x, ${P}.y), ${P}.z)`:P,$=k("x",c.dataType,c.dims,g),_=M("result",c.dataType,c.dims,g),x=$.type.value,T=ye(c.dataType)==="f32"?`var threadMax = ${x}(-3.402823e+38f);`:`var threadMax = ${x}(-65504.0h);`,C=P=>`
+      }`;return{name:"Slice",shaderCache:{hint:`${l.length}_${a.length}_${i.length}`,inputDependencies:["rank"]},getShaderSource:S,getRunData:()=>({outputs:[m],dispatchGroup:{x:Math.ceil(n/64)},programUniforms:y})}},$l=(e,t)=>{Sh(e.inputs,t);let r=Th(e.inputs,t);e.compute(Ch(e.inputs,r),{inputs:[0]})},xl=e=>{let t=e.starts,r=e.ends,n=e.axes;return J({starts:t,ends:r,axes:n})}});var Ah,kh,Tl,Il,Cl=V(()=>{"use strict";Q();ie();Ie();lt();ae();Ah=e=>{if(!e||e.length!==1)throw new Error("Softmax op requires 1 input.")},kh=(e,t)=>{let r=e.inputs[0],n=r.dims,o=k.size(n),i=64,a=n.length,d=k.normalizeAxis(t.axis,a),l=d<n.length-1,c,m=[];l?(m=Array.from({length:a},(P,B)=>B),m[d]=a-1,m[a-1]=d,c=e.compute(Pe(r,m),{inputs:[r],outputs:[-1]})[0]):c=r;let u=c.dims,h=u[a-1],w=o/h,g=we(h),y=h/g,S=(P,B)=>B===4?`max(max(${P}.x, ${P}.y), max(${P}.z, ${P}.w))`:B===2?`max(${P}.x, ${P}.y)`:B===3?`max(max(${P}.x, ${P}.y), ${P}.z)`:P,$=E("x",c.dataType,c.dims,g),v=M("result",c.dataType,c.dims,g),x=$.type.value,T=he(c.dataType)==="f32"?`var threadMax = ${x}(-3.402823e+38f);`:`var threadMax = ${x}(-65504.0h);`,C=P=>`
       var<workgroup> rowMaxShared : ${x};
       var<workgroup> rowSumShared : ${x};
       var<workgroup> threadShared : array<${x}, ${i}>;
@@ -6205,7 +6279,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
         let index = row * row_stride + col;
         result[index] = value;
       }
-      ${P.registerUniform("packedCols","i32").declareVariables($,_)}
+      ${P.registerUniform("packedCols","i32").declareVariables($,v)}
       ${P.mainStart()}
         let gindex = i32(global_idx);
         let lindex = i32(local_idx);
@@ -6263,7 +6337,7 @@ ${R}_indices[${Re}] = 0;`}),he+=`${R}_indices[${se-2}] = 0u;
           let value = exp(getValue(row, col, row_stride) - rowMaxShared) / rowSumShared;
           setValue(row, col, row_stride, value);
         }
-      }`,A=e.compute({name:"Softmax",shaderCache:{hint:`${g}`,inputDependencies:["type"]},getRunData:()=>({outputs:[{dims:u,dataType:c.dataType}],dispatchGroup:{x:w},programUniforms:[{type:6,data:y}]}),getShaderSource:C},{inputs:[c],outputs:[l?-1:0]})[0];l&&e.compute(Pe(A,m),{inputs:[A]})},Tl=(e,t)=>{Ch(e.inputs),Ah(e,t)},Il=e=>J({axis:e.axis})});var Eh,kh,Ph,Oh,zh,Al,El,kl=U(()=>{"use strict";Z();te();Ie();re();Eh=e=>{if(!e||e.length<1)throw new Error("too few inputs")},kh=(e,t)=>{let r=[],n=t.numOutputs;return e[1].dims[0]>0&&(e[1].getBigInt64Array().forEach(o=>r.push(Number(o))),n=r.length),J({numOutputs:n,axis:t.axis,splitSizes:r})},Ph=e=>`
+      }`,A=e.compute({name:"Softmax",shaderCache:{hint:`${g}`,inputDependencies:["type"]},getRunData:()=>({outputs:[{dims:u,dataType:c.dataType}],dispatchGroup:{x:w},programUniforms:[{type:6,data:y}]}),getShaderSource:C},{inputs:[c],outputs:[l?-1:0]})[0];l&&e.compute(Pe(A,m),{inputs:[A]})},Tl=(e,t)=>{Ah(e.inputs),kh(e,t)},Il=e=>J({axis:e.axis})});var Eh,Ph,zh,Oh,Bh,Al,kl,El=V(()=>{"use strict";Q();ie();Ie();ae();Eh=e=>{if(!e||e.length<1)throw new Error("too few inputs")},Ph=(e,t)=>{let r=[],n=t.numOutputs;return e[1].dims[0]>0&&(e[1].getBigInt64Array().forEach(o=>r.push(Number(o))),n=r.length),J({numOutputs:n,axis:t.axis,splitSizes:r})},zh=e=>`
 fn calculateOutputIndex(index: u32) -> u32 {
     for (var i: u32 = 0u; i < ${e}u; i += 1u ) {
     if (index < ${F("uniforms.size_in_split_axis","i",e)}) {
@@ -6275,9 +6349,9 @@ fn calculateOutputIndex(index: u32) -> u32 {
       fn writeBufferData(output_number: u32, indices: ${e[0].type.indices}, global_idx: u32) {
         ${r.join(`
 `)}
-      }`},zh=(e,t)=>{let r=e[0].dims,n=E.size(r),o=e[0].dataType,i=E.normalizeAxis(t.axis,r.length),a=new Array(t.numOutputs),d=k("input",o,r.length),l=new Array(t.numOutputs),c=[],m=[],u=0,h=[{type:12,data:n}];for(let g=0;g<t.numOutputs;g++){u+=t.splitSizes[g],l[g]=u;let y=r.slice();y[i]=t.splitSizes[g],m.push(y),a[g]=M(`output${g}`,o,y.length),c.push({dims:m[g],dataType:e[0].dataType})}h.push({type:12,data:l},...V(r,...m));let w=g=>`
+      }`},Bh=(e,t)=>{let r=e[0].dims,n=k.size(r),o=e[0].dataType,i=k.normalizeAxis(t.axis,r.length),a=new Array(t.numOutputs),d=E("input",o,r.length),l=new Array(t.numOutputs),c=[],m=[],u=0,h=[{type:12,data:n}];for(let g=0;g<t.numOutputs;g++){u+=t.splitSizes[g],l[g]=u;let y=r.slice();y[i]=t.splitSizes[g],m.push(y),a[g]=M(`output${g}`,o,y.length),c.push({dims:m[g],dataType:e[0].dataType})}h.push({type:12,data:l},...R(r,...m));let w=g=>`
   ${g.registerUniform("input_size","u32").registerUniform("size_in_split_axis","u32",l.length).declareVariables(d,...a)}
-  ${Ph(l.length)}
+  ${zh(l.length)}
   ${Oh(a)}
 
   ${g.mainStart()}
@@ -6291,7 +6365,7 @@ fn calculateOutputIndex(index: u32) -> u32 {
       ${d.indicesSet("indices",i,"index")};
     }
     writeBufferData(output_number, indices, global_idx);
-  }`;return{name:"Split",shaderCache:{hint:t.cacheKey,inputDependencies:["rank"]},getShaderSource:w,getRunData:()=>({outputs:c,dispatchGroup:{x:Math.ceil(n/64)},programUniforms:h})}},Al=(e,t)=>{Eh(e.inputs);let r=e.inputs.length===1?t:kh(e.inputs,t);e.compute(zh(e.inputs,r),{inputs:[0]})},El=e=>{let t=e.axis,r=e.splitSizes,n=e.numOutputs<0?r.length:e.numOutputs;if(n!==r.length)throw new Error("numOutputs and splitSizes lengh must be equal");return J({axis:t,numOutputs:n,splitSizes:r})}});var Dh,Bh,Pl,Ol=U(()=>{"use strict";Z();te();re();Dh=(e,t,r,n,o)=>{let i=M("output_data",o,r.length,4),a=k("a_data",t[1].dataType,t[1].dims.length,4),d=k("b_data",t[2].dataType,t[2].dims.length,4),l=k("c_data",t[0].dataType,t[0].dims.length,4),c,m=(u,h,w)=>`select(${h}, ${u}, ${w})`;if(!n)c=i.setByOffset("global_idx",m(a.getByOffset("global_idx"),d.getByOffset("global_idx"),l.getByOffset("global_idx")));else{let u=(h,w,g="")=>{let y=`a_data[index_a${w}][component_a${w}]`,S=`b_data[index_b${w}][component_b${w}]`,$=`bool(c_data[index_c${w}] & (0xffu << (component_c${w} * 8)))`;return`
+  }`;return{name:"Split",shaderCache:{hint:t.cacheKey,inputDependencies:["rank"]},getShaderSource:w,getRunData:()=>({outputs:c,dispatchGroup:{x:Math.ceil(n/64)},programUniforms:h})}},Al=(e,t)=>{Eh(e.inputs);let r=e.inputs.length===1?t:Ph(e.inputs,t);e.compute(Bh(e.inputs,r),{inputs:[0]})},kl=e=>{let t=e.axis,r=e.splitSizes,n=e.numOutputs<0?r.length:e.numOutputs;if(n!==r.length)throw new Error("numOutputs and splitSizes lengh must be equal");return J({axis:t,numOutputs:n,splitSizes:r})}});var Dh,Mh,Pl,zl=V(()=>{"use strict";Q();ie();ae();Dh=(e,t,r,n,o)=>{let i=M("output_data",o,r.length,4),a=E("a_data",t[1].dataType,t[1].dims.length,4),d=E("b_data",t[2].dataType,t[2].dims.length,4),l=E("c_data",t[0].dataType,t[0].dims.length,4),c,m=(u,h,w)=>`select(${h}, ${u}, ${w})`;if(!n)c=i.setByOffset("global_idx",m(a.getByOffset("global_idx"),d.getByOffset("global_idx"),l.getByOffset("global_idx")));else{let u=(h,w,g="")=>{let y=`a_data[index_a${w}][component_a${w}]`,S=`b_data[index_b${w}][component_b${w}]`,$=`bool(c_data[index_c${w}] & (0xffu << (component_c${w} * 8)))`;return`
             let output_indices${w} = ${i.offsetToIndices(`global_idx * 4u + ${w}u`)};
             let offset_a${w} = ${a.broadcastedIndicesToOffset(`output_indices${w}`,i)};
             let offset_b${w} = ${d.broadcastedIndicesToOffset(`output_indices${w}`,i)};
@@ -6319,10 +6393,10 @@ fn calculateOutputIndex(index: u32) -> u32 {
         ${e.mainStart()}
         ${e.guardAgainstOutOfBoundsWorkgroupSizes("uniforms.vec_size")}
         ${c}
-      }`},Bh=e=>{let t=e[1].dims,r=e[2].dims,n=e[0].dims,o=e[1].dataType,i=!(E.areEqual(t,r)&&E.areEqual(r,n)),a=t,d=E.size(t);if(i){let c=rt.calcShape(rt.calcShape(t,r,!1),n,!1);if(!c)throw new Error("Can't perform where op on the given tensors");a=c,d=E.size(a)}let l=Math.ceil(d/4);return{name:"Where",shaderCache:{inputDependencies:["rank","rank","rank"]},getShaderSource:c=>Dh(c,e,a,i,o),getRunData:()=>({outputs:[{dims:a,dataType:o}],dispatchGroup:{x:Math.ceil(d/64/4)},programUniforms:[{type:12,data:l},...V(n,t,r,a)]})}},Pl=e=>{e.compute(Bh(e.inputs))}});var zl,Dl=U(()=>{"use strict";Ts();Kr();As();ks();hu();Iu();Eu();Hu();Xu();Ju();rd();sd();ld();pd();hd();bd();_d();Sd();Dd();Rd();Vd();fo();Ld();xo();Gd();sl();ll();pl();qr();gl();bl();vl();Sl();Cl();kl();To();lt();Xr();Ol();zl=new Map([["Abs",[Ps]],["Acos",[Os]],["Acosh",[zs]],["Add",[gu]],["ArgMax",[Ss,uo]],["ArgMin",[xs,uo]],["Asin",[Ds]],["Asinh",[Bs]],["Atan",[Ms]],["Atanh",[Rs]],["Attention",[Is]],["AveragePool",[Qd,Zd]],["BatchNormalization",[Cs]],["BiasAdd",[Es]],["BiasSplitGelu",[fu]],["Cast",[Vs,Us]],["Ceil",[Ws]],["Clip",[Ns]],["Concat",[Cu,Au]],["Conv",[bo,yo]],["ConvTranspose",[Yu,Ku]],["Cos",[Ls]],["Cosh",[Hs]],["CumSum",[Zu,Qu]],["DepthToSpace",[ed,td]],["DequantizeLinear",[ul,dl]],["Div",[yu]],["Einsum",[id,ad]],["Elu",[Gs,Qt]],["Equal",[bu]],["Erf",[Fs]],["Exp",[qs]],["Expand",[dd]],["FastGelu",[cd]],["Floor",[js]],["FusedConv",[bo,yo]],["Gather",[fd,md]],["GatherElements",[vd,wd]],["GatherBlockQuantized",[gd,yd]],["Gelu",[Ks]],["Gemm",[xd,$d]],["GlobalAveragePool",[tl,el]],["GlobalMaxPool",[al,il]],["Greater",[$u]],["GreaterOrEqual",[Su]],["GroupQueryAttention",[zd,Od]],["HardSigmoid",[ru,tu]],["InstanceNormalization",[Md]],["LayerNormalization",[Ud]],["LeakyRelu",[Ys,Qt]],["Less",[xu]],["LessOrEqual",[Tu]],["Log",[cu]],["MatMul",[Wu]],["MatMulNBits",[Nd,Wd]],["MaxPool",[nl,ol]],["Mul",[wu]],["MultiHeadAttention",[Cd,Id]],["Neg",[Zs]],["Not",[Xs]],["Pad",[Hd]],["Pow",[vu]],["QuickGelu",[pu,Qt]],["Range",[cl]],["Reciprocal",[Qs]],["ReduceMin",[ys]],["ReduceMean",[ps]],["ReduceMax",[gs]],["ReduceSum",[ws]],["ReduceProd",[bs]],["ReduceL1",[ms]],["ReduceL2",[fs]],["ReduceLogSum",[_s]],["ReduceLogSumExp",[hs]],["ReduceSumSquare",[vs]],["Relu",[Js]],["Resize",[fl,hl]],["RotaryEmbedding",[yl]],["Sigmoid",[eu]],["Sin",[nu]],["Sinh",[ou]],["Slice",[$l,xl]],["SkipLayerNormalization",[wl]],["Split",[Al,El]],["Sqrt",[iu]],["Softmax",[Tl,Il]],["Sub",[_u]],["Tan",[au]],["Tanh",[uu]],["ThresholdedRelu",[lu,Qt]],["Tile",[Ed]],["Transpose",[Ja,es]],["Where",[Pl]]])});var an,Bl=U(()=>{"use strict";Ke();Xe();re();an=class{constructor(t){this.backend=t;this.repo=new Map,this.attributesBound=!1}getArtifact(t){return this.repo.get(t)}setArtifact(t,r){this.repo.set(t,r)}run(t,r,n,o,i){Le(t.programInfo.name);let a=this.backend.device,d=this.backend.getComputePassEncoder();this.backend.writeTimestamp(this.backend.pendingDispatchNumber*2);let l=[];for(let m of r)l.push({binding:l.length,resource:{buffer:m.buffer}});for(let m of n)l.push({binding:l.length,resource:{buffer:m.buffer}});i&&l.push({binding:l.length,resource:i});let c=a.createBindGroup({layout:t.computePipeline.getBindGroupLayout(0),entries:l,label:t.programInfo.name});if(this.backend.sessionStatus==="capturing"){let m={kernelId:this.backend.currentKernelId,computePipeline:t.computePipeline,bindGroup:c,dispatchGroup:o};this.backend.capturedCommandList.get(this.backend.currentSessionId).push(m)}d.setPipeline(t.computePipeline),d.setBindGroup(0,c),d.dispatchWorkgroups(...o),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(),Ve(t.programInfo.name)}dispose(){}build(t,r){Le(t.name);let n=this.backend.device,o=[];n.features.has("shader-f16")&&o.push("enable f16;");let i=Za(r,this.backend.device.limits),a=t.getShaderSource(i),d=`${o.join(`
+      }`},Mh=e=>{let t=e[1].dims,r=e[2].dims,n=e[0].dims,o=e[1].dataType,i=!(k.areEqual(t,r)&&k.areEqual(r,n)),a=t,d=k.size(t);if(i){let c=rt.calcShape(rt.calcShape(t,r,!1),n,!1);if(!c)throw new Error("Can't perform where op on the given tensors");a=c,d=k.size(a)}let l=Math.ceil(d/4);return{name:"Where",shaderCache:{inputDependencies:["rank","rank","rank"]},getShaderSource:c=>Dh(c,e,a,i,o),getRunData:()=>({outputs:[{dims:a,dataType:o}],dispatchGroup:{x:Math.ceil(d/64/4)},programUniforms:[{type:12,data:l},...R(n,t,r,a)]})}},Pl=e=>{e.compute(Mh(e.inputs))}});var Ol,Bl=V(()=>{"use strict";Ts();Kr();As();Es();hu();Iu();ku();Hu();Xu();Ju();rd();sd();ld();pd();hd();bd();vd();Sd();Bd();Rd();Vd();fo();Ld();xo();Gd();sl();ll();pl();qr();gl();bl();_l();Sl();Cl();El();To();lt();Xr();zl();Ol=new Map([["Abs",[Ps]],["Acos",[zs]],["Acosh",[Os]],["Add",[gu]],["ArgMax",[Ss,uo]],["ArgMin",[xs,uo]],["Asin",[Bs]],["Asinh",[Ds]],["Atan",[Ms]],["Atanh",[Rs]],["Attention",[Is]],["AveragePool",[Qd,Zd]],["BatchNormalization",[Cs]],["BiasAdd",[ks]],["BiasSplitGelu",[fu]],["Cast",[Vs,Us]],["Ceil",[Ws]],["Clip",[Ns]],["Concat",[Cu,Au]],["Conv",[bo,yo]],["ConvTranspose",[Yu,Ku]],["Cos",[Ls]],["Cosh",[Hs]],["CumSum",[Zu,Qu]],["DepthToSpace",[ed,td]],["DequantizeLinear",[ul,dl]],["Div",[yu]],["Einsum",[id,ad]],["Elu",[Gs,Qt]],["Equal",[bu]],["Erf",[Fs]],["Exp",[qs]],["Expand",[dd]],["FastGelu",[cd]],["Floor",[js]],["FusedConv",[bo,yo]],["Gather",[fd,md]],["GatherElements",[_d,wd]],["GatherBlockQuantized",[gd,yd]],["Gelu",[Ks]],["Gemm",[xd,$d]],["GlobalAveragePool",[tl,el]],["GlobalMaxPool",[al,il]],["Greater",[$u]],["GreaterOrEqual",[Su]],["GroupQueryAttention",[Od,zd]],["HardSigmoid",[ru,tu]],["InstanceNormalization",[Md]],["LayerNormalization",[Ud]],["LeakyRelu",[Ys,Qt]],["Less",[xu]],["LessOrEqual",[Tu]],["Log",[cu]],["MatMul",[Wu]],["MatMulNBits",[Nd,Wd]],["MaxPool",[nl,ol]],["Mul",[wu]],["MultiHeadAttention",[Cd,Id]],["Neg",[Zs]],["Not",[Xs]],["Pad",[Hd]],["Pow",[_u]],["QuickGelu",[pu,Qt]],["Range",[cl]],["Reciprocal",[Qs]],["ReduceMin",[ys]],["ReduceMean",[ps]],["ReduceMax",[gs]],["ReduceSum",[ws]],["ReduceProd",[bs]],["ReduceL1",[ms]],["ReduceL2",[fs]],["ReduceLogSum",[vs]],["ReduceLogSumExp",[hs]],["ReduceSumSquare",[_s]],["Relu",[Js]],["Resize",[fl,hl]],["RotaryEmbedding",[yl]],["Sigmoid",[eu]],["Sin",[nu]],["Sinh",[ou]],["Slice",[$l,xl]],["SkipLayerNormalization",[wl]],["Split",[Al,kl]],["Sqrt",[iu]],["Softmax",[Tl,Il]],["Sub",[vu]],["Tan",[au]],["Tanh",[uu]],["ThresholdedRelu",[lu,Qt]],["Tile",[kd]],["Transpose",[Ja,es]],["Where",[Pl]]])});var an,Dl=V(()=>{"use strict";Ke();Xe();ae();an=class{constructor(t){this.backend=t;this.repo=new Map,this.attributesBound=!1}getArtifact(t){return this.repo.get(t)}setArtifact(t,r){this.repo.set(t,r)}run(t,r,n,o,i){Le(t.programInfo.name);let a=this.backend.device,d=this.backend.getComputePassEncoder();this.backend.writeTimestamp(this.backend.pendingDispatchNumber*2);let l=[];for(let m of r)l.push({binding:l.length,resource:{buffer:m.buffer}});for(let m of n)l.push({binding:l.length,resource:{buffer:m.buffer}});i&&l.push({binding:l.length,resource:i});let c=a.createBindGroup({layout:t.computePipeline.getBindGroupLayout(0),entries:l,label:t.programInfo.name});if(this.backend.sessionStatus==="capturing"){let m={kernelId:this.backend.currentKernelId,computePipeline:t.computePipeline,bindGroup:c,dispatchGroup:o};this.backend.capturedCommandList.get(this.backend.currentSessionId).push(m)}d.setPipeline(t.computePipeline),d.setBindGroup(0,c),d.dispatchWorkgroups(...o),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(),Ve(t.programInfo.name)}dispose(){}build(t,r){Le(t.name);let n=this.backend.device,o=[];n.features.has("shader-f16")&&o.push("enable f16;");let i=Za(r,this.backend.device.limits),a=t.getShaderSource(i),d=`${o.join(`
 `)}
 ${i.additionalImplementations}
-${a}`,l=n.createShaderModule({code:d,label:t.name});de("verbose",()=>`[WebGPU] ${t.name} shader code: ${d}`);let c=n.createComputePipeline({compute:{module:l,entryPoint:"main"},layout:"auto",label:t.name});return Ve(t.name),{programInfo:t,computePipeline:c,uniformVariablesInfo:i.variablesInfo}}normalizeDispatchGroupSize(t){let r=typeof t=="number"?t:t.x,n=typeof t=="number"?1:t.y||1,o=typeof t=="number"?1:t.z||1,i=this.backend.device.limits.maxComputeWorkgroupsPerDimension;if(r<=i&&n<=i&&o<=i)return[r,n,o];let a=r*n*o,d=Math.ceil(Math.sqrt(a));if(d>i){if(d=Math.ceil(Math.cbrt(a)),d>i)throw new Error("Total dispatch size exceeds WebGPU maximum.");return[d,d,d]}else return[d,d,1]}}});var Mh,Rh,Io,sn,Ml=U(()=>{"use strict";Ke();Z();Xe();Xn();Ka();Dl();Bl();Mh=(e,t)=>{if(t.length!==e.length)throw new Error(`inputDependencies length ${t.length} is not equal to inputTensors length ${e.length}.`);let r=[];for(let n=0;n<e.length;++n){let o=e[n].dataType;switch(t[n]){case"none":{r.push("");break}case"type":{r.push(`${o}`);break}case"rank":{let i=e[n].dims.length;r.push(`${o};${i}`);break}case"dims":{let i=e[n].dims.join(",");r.push(`${o};${i}`);break}default:throw new Error(`unsupported input dependency: ${t[n]}`)}}return r.join("|")},Rh=(e,t,r)=>{let n=e.name;return e.shaderCache?.hint&&(n+="["+e.shaderCache.hint+"]"),n+=":"+r+`:${Mh(t,e.shaderCache?.inputDependencies??new Array(t.length).fill("dims"))}`,n},Io=class{constructor(t){t&&(this.architecture=t.architecture,this.vendor=t.vendor)}isArchitecture(t){return this.architecture===t}isVendor(t){return this.vendor===t}},sn=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 t=this.kernelCustomData.get(this.currentKernelId);return t||(t={},this.kernelCustomData.set(this.currentKernelId,t)),t}async initialize(t,r){this.env=t;let n=[],o={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:n};r.features.has("chromium-experimental-timestamp-query-inside-passes")?n.push("chromium-experimental-timestamp-query-inside-passes"):r.features.has("timestamp-query")&&n.push("timestamp-query"),r.features.has("shader-f16")&&n.push("shader-f16"),this.device=await r.requestDevice(o),this.adapterInfo=new Io(r.info||await r.requestAdapterInfo()),this.gpuDataManager=ja(this),this.programManager=new an(this),this.kernels=new Map,this.kernelPersistentData=new Map,this.kernelCustomData=new Map,Wr(t.logLevel,!!t.debug),this.device.onuncapturederror=i=>{i.error instanceof GPUValidationError&&console.error(`An uncaught WebGPU validation error was raised: ${i.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 t=this.getCommandEncoder(),r={};this.queryType==="at-passes"&&(r.timestampWrites={querySet:this.querySet,beginningOfPassWriteIndex:this.pendingDispatchNumber*2,endOfPassWriteIndex:this.pendingDispatchNumber*2+1}),this.computePassEncoder=t.beginComputePass(r)}return this.computePassEncoder}endComputePass(){this.computePassEncoder&&(this.computePassEncoder.end(),this.computePassEncoder=null)}flush(){if(!this.commandEncoder)return;Le(),this.endComputePass();let t;this.queryType!=="none"&&(this.commandEncoder.resolveQuerySet(this.querySet,0,this.pendingDispatchNumber*2,this.queryResolveBuffer,0),t=this.device.createBuffer({size:this.pendingDispatchNumber*2*8,usage:GPUBufferUsage.MAP_READ|GPUBufferUsage.COPY_DST}),this.pendingQueries.set(t,this.pendingKernels),this.pendingKernels=[],this.commandEncoder.copyBufferToBuffer(this.queryResolveBuffer,0,t,0,this.pendingDispatchNumber*2*8)),this.device.queue.submit([this.commandEncoder.finish()]),this.gpuDataManager.refreshPendingBuffers(),this.commandEncoder=null,this.pendingDispatchNumber=0,this.queryType!=="none"&&t.mapAsync(GPUMapMode.READ).then(()=>{let r=new BigUint64Array(t.getMappedRange()),n=this.pendingQueries.get(t);for(let o=0;o<r.length/2;o++){let i=n[o],a=i.kernelId,d=this.kernels.get(a),l=d.kernelType,c=d.kernelName,m=i.programName,u=i.inputTensorViews,h=i.outputTensorViews,w=r[o*2],g=r[o*2+1];typeof this.queryTimeBase>"u"&&(this.queryTimeBase=w);let y=Number(w-this.queryTimeBase),S=Number(g-this.queryTimeBase);if(!Number.isSafeInteger(y)||!Number.isSafeInteger(S))throw new RangeError("incorrect timestamp range");if(this.env.webgpu.profiling?.ondata)this.env.webgpu.profiling.ondata({version:1,inputsMetadata:u.map($=>({dims:$.dims,dataType:yt($.dataType)})),outputsMetadata:h.map($=>({dims:$.dims,dataType:yt($.dataType)})),kernelId:a,kernelType:l,kernelName:c,programName:m,startTime:y,endTime:S});else{let $="";u.forEach((x,T)=>{$+=`input[${T}]: [${x.dims}] | ${yt(x.dataType)}, `});let _="";h.forEach((x,T)=>{_+=`output[${T}]: [${x.dims}] | ${yt(x.dataType)}, `}),console.log(`[profiling] kernel "${a}|${l}|${c}|${m}" ${$}${_}execution time: ${S-y} ns`)}Sr("GPU",`${m}::${w}::${g}`)}t.unmap(),this.pendingQueries.delete(t)}),Ve()}run(t,r,n,o,i,a){Le(t.name);let d=[];for(let x=0;x<r.length;++x){let T=r[x].data;if(T===0)continue;let C=this.gpuDataManager.get(T);if(!C)throw new Error(`no GPU data for input: ${T}`);d.push(C)}let{outputs:l,dispatchGroup:c,programUniforms:m}=t.getRunData(r),u=n.length===0?l.map((x,T)=>T):n;if(u.length!==l.length)throw new Error(`Output size ${u.length} must be equal to ${l.length}.`);let h=[],w=[];for(let x=0;x<l.length;++x){if(!Number.isInteger(u[x])||u[x]<-3||u[x]>=a)throw new Error(`Invalid output index: ${u[x]}`);if(u[x]===-3)continue;let T=u[x]===-1,C=u[x]===-2,A=T||C?i(l[x].dataType,l[x].dims):o(u[x],l[x].dataType,l[x].dims);if(h.push(A),A.data===0)continue;let P=this.gpuDataManager.get(A.data);if(!P)throw new Error(`no GPU data for output: ${A.data}`);if(T&&this.temporaryData.push(P),C){let D=this.kernelPersistentData.get(this.currentKernelId);D||(D=[],this.kernelPersistentData.set(this.currentKernelId,D)),D.push(P)}w.push(P)}if(d.length!==r.length||w.length!==h.length){if(w.length===0)return Ve(t.name),h;throw new Error(`Program ${t.name} has zero-sized tensor(s) in inputs or outputs. This is not supported now.`)}let g;if(m){let x=0,T=[];m.forEach(D=>{let N=typeof D.data=="number"?[D.data]:D.data;if(N.length===0)return;let L=D.type===10?2:4,K,X;D.type===10?(X=N.length>4?16:N.length>2?8:N.length*L,K=N.length>4?16:L*N.length):(X=N.length<=2?N.length*L:16,K=16),x=Math.ceil(x/X)*X,T.push(x);let ie=D.type===10?8:4;x+=N.length>4?Math.ceil(N.length/ie)*K:N.length*L});let C=16;x=Math.ceil(x/C)*C;let A=new ArrayBuffer(x);m.forEach((D,N)=>{let L=T[N],K=typeof D.data=="number"?[D.data]:D.data;if(D.type===6)new Int32Array(A,L,K.length).set(K);else if(D.type===12)new Uint32Array(A,L,K.length).set(K);else if(D.type===10)new Uint16Array(A,L,K.length).set(K);else if(D.type===1)new Float32Array(A,L,K.length).set(K);else throw new Error(`Unsupported uniform type: ${yt(D.type)}`)});let P=this.gpuDataManager.create(x,GPUBufferUsage.COPY_DST|GPUBufferUsage.UNIFORM);this.device.queue.writeBuffer(P.buffer,0,A,0,x),this.gpuDataManager.release(P.id),g={offset:0,size:x,buffer:P.buffer}}let y=this.programManager.normalizeDispatchGroupSize(c),S=y[1]===1&&y[2]===1,$=Rh(t,r,S),_=this.programManager.getArtifact($);if(_||(_=this.programManager.build(t,y),this.programManager.setArtifact($,_),de("info",()=>`[artifact] key: ${$}, programName: ${t.name}`)),m&&_.uniformVariablesInfo){if(m.length!==_.uniformVariablesInfo.length)throw new Error(`Uniform variables count mismatch: expect ${_.uniformVariablesInfo.length}, got ${m.length} in program "${_.programInfo.name}".`);for(let x=0;x<m.length;x++){let T=m[x],C=T.type,A=typeof T.data=="number"?1:T.data.length,[P,D]=_.uniformVariablesInfo[x];if(C!==P||A!==D)throw new Error(`Uniform variable ${x} mismatch: expect type ${P} with size ${D}, got type ${C} with size ${A} in program "${_.programInfo.name}".`)}}if(de("info",()=>`[ProgramManager] run "${t.name}" (key=${$}) with ${y[0]}x${y[1]}x${y[2]}`),this.queryType!=="none"||this.sessionStatus==="capturing"){let x={kernelId:this.currentKernelId,programName:_.programInfo.name,inputTensorViews:r,outputTensorViews:h};this.pendingKernels.push(x),this.sessionStatus==="capturing"&&this.capturedPendingKernels.get(this.currentSessionId).push(x)}return this.programManager.run(_,d,w,y,g),Ve(t.name),h}upload(t,r){this.gpuDataManager.upload(t,r)}memcpy(t,r){this.gpuDataManager.memcpy(t,r)}async download(t,r){await this.gpuDataManager.download(t,r)}alloc(t){return this.gpuDataManager.create(t).id}free(t){return this.gpuDataManager.release(t)}createKernel(t,r,n,o){let i=zl.get(t);if(!i)throw new Error(`kernel not implemented: ${t}`);let a={kernelType:t,kernelName:o,kernelEntry:i[0],attributes:[i[1],n]};this.kernels.set(r,a)}releaseKernel(t){let r=this.kernelPersistentData.get(t);if(r){for(let n of r)this.gpuDataManager.release(n.id);this.kernelPersistentData.delete(t)}this.kernelCustomData.delete(t),this.kernels.delete(t)}computeKernel(t,r,n){let o=this.kernels.get(t);if(!o)throw new Error(`kernel not created: ${t}`);let i=o.kernelType,a=o.kernelName,d=o.kernelEntry,l=o.attributes;if(this.currentKernelId!==null)throw new Error(`kernel "[${i}] ${a}" is not allowed to be called recursively`);this.currentKernelId=t,l[0]&&(l[1]=l[0](l[1]),l[0]=void 0),de("info",()=>`[WebGPU] Start to run kernel "[${i}] ${a}"...`);let c=this.env.debug;this.temporaryData=[];try{return c&&this.device.pushErrorScope("validation"),d(r,l[1]),0}catch(m){return n.push(Promise.resolve(`[WebGPU] Kernel "[${i}] ${a}" failed. ${m}`)),1}finally{c&&n.push(this.device.popErrorScope().then(m=>m?`GPU validation error for kernel "[${i}] ${a}": ${m.message}`:null));for(let m of this.temporaryData)this.gpuDataManager.release(m.id);this.temporaryData=[],this.currentKernelId=null}}registerBuffer(t,r,n,o){let i=this.sessionExternalDataMapping.get(t);i||(i=new Map,this.sessionExternalDataMapping.set(t,i));let a=i.get(r),d=this.gpuDataManager.registerExternalBuffer(n,o,a);return i.set(r,[d,n]),d}unregisterBuffers(t){let r=this.sessionExternalDataMapping.get(t);r&&(r.forEach(n=>this.gpuDataManager.unregisterExternalBuffer(n[0])),this.sessionExternalDataMapping.delete(t))}getBuffer(t){let r=this.gpuDataManager.get(t);if(!r)throw new Error(`no GPU data for buffer: ${t}`);return r.buffer}createDownloader(t,r,n){return async()=>{let o=await eo(this,t,r);return Lr(o.buffer,n)}}writeTimestamp(t){this.queryType==="inside-passes"&&this.computePassEncoder.writeTimestamp(this.querySet,t)}setQueryType(){this.queryType="none",(this.env.webgpu.profiling?.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(){de("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(){de("info","captureEnd"),this.flush(),this.sessionStatus="default"}replay(){de("info","replay"),this.sessionStatus="replaying";let t=this.capturedCommandList.get(this.currentSessionId),r=this.capturedPendingKernels.get(this.currentSessionId),n=t.length;this.pendingKernels=[];for(let o=0;o<n;o++){let i=this.getComputePassEncoder(),a=t[o];this.writeTimestamp(this.pendingDispatchNumber*2),i.setPipeline(a.computePipeline),i.setBindGroup(0,a.bindGroup),i.dispatchWorkgroups(...a.dispatchGroup),this.writeTimestamp(this.pendingDispatchNumber*2+1),this.pendingDispatchNumber++,this.queryType!=="none"&&this.pendingKernels.push(r[o]),(this.pendingDispatchNumber>=this.maxDispatchNumber||this.queryType==="at-passes")&&this.endComputePass(),this.pendingDispatchNumber>=this.maxDispatchNumber&&this.flush()}this.flush(),this.sessionStatus="default"}onReleaseSession(t){this.unregisterBuffers(t),this.capturedCommandList.has(t)&&this.capturedCommandList.delete(t),this.capturedPendingKernels.has(t)&&this.capturedPendingKernels.delete(t),this.gpuDataManager.onReleaseSession(t)}onRunStart(t){this.currentSessionId=t,this.setQueryType()}}});var Uh,Rl,un,Co,Ul,Vl=U(()=>{"use strict";Xe();Uh=1,Rl=()=>Uh++,un=class{constructor(t,r){this.mlContext=t;this.tensorEntry=r,this.tensorCache=r?[r]:[]}get tensor(){return this.tensorEntry?.[0]}get context(){if(!this.mlContext)throw new Error("MLContext has not been set.");return this.mlContext}set context(t){if(this.mlContext&&this.mlContext!==t)throw new Error("MLTensor in use in a different MLContext.");this.mlContext=t}destroy(){for(let[t]of this.tensorCache)t.destroy();this.tensorCache=[],this.tensorEntry=void 0}trySelectTensor(t,r){for(let[n,o,i]of this.tensorCache)if(r===n){if(this.context!==t)throw new Error("MLTensor cannot be registered with a different MLContext.");return this.tensorEntry=[n,o,i],!0}return!1}async ensureTensor(t,r,n){if(this.tensorEntry){let[a,d,l]=this.tensorEntry;if(d===t&&l.every((c,m)=>c===r[m]))return a}for(let[a,d,l]of this.tensorCache)if(d===t&&l.every((c,m)=>c===r[m])){if(n&&this.tensorEntry){de("verbose",()=>`[WebNN] Slowdown may occur, having to copy existing tensor {dataType: ${t}, shape: ${r}}`);let c=await this.context.readTensor(this.tensorEntry[0]);this.context.writeTensor(a,c)}return this.tensorEntry=[a,d,l],a}de("verbose",()=>`[WebNN] MLContext.createTensor {dataType: ${t}, shape: ${r}}`);let o=MLTensorUsage.READ|MLTensorUsage.WRITE,i=await this.context.createTensor({dataType:t,shape:r,dimensions:r,usage:o});return this.tensorEntry=[i,t,r],this.tensorCache.push(this.tensorEntry),this.activeUpload&&(this.mlContext?.writeTensor(i,this.activeUpload),this.activeUpload=void 0),i}upload(t){if(!this.tensorEntry){this.activeUpload=new Uint8Array(t);return}this.mlContext?.writeTensor(this.tensorEntry[0],t)}async download(t){if(this.activeUpload)if(t){t instanceof ArrayBuffer?new Uint8Array(t).set(this.activeUpload):new Uint8Array(t.buffer,t.byteOffset,t.byteLength).set(this.activeUpload);return}else return this.activeUpload.buffer;if(!this.tensorEntry)throw new Error("Tensor has not been created.");return t?this.context.readTensor(this.tensorEntry[0],t):this.context.readTensor(this.tensorEntry[0])}},Co=class{constructor(t){this.backend=t;this.tensorsById=new Map;this.tensorIdsByContext=new Map}reserveTensorId(){let t=Rl();return this.tensorsById.set(t,new un),t}releaseTensorId(t){let r=this.tensorsById.get(t);if(r){r.destroy(),this.tensorsById.delete(t);for(let[n,o]of this.tensorIdsByContext)if(o.has(t)){o.delete(t),o.size===0&&this.tensorIdsByContext.delete(n);break}}}async ensureTensor(t,r,n,o){de("verbose",()=>`[WebNN] TensorManager.ensureTensor {tensorId: ${t}, dataType: ${r}, shape: ${n}, copyOld: ${o}}`);let i=this.tensorsById.get(t);if(!i)throw new Error("Tensor not found.");return i.context=this.backend.currentContext,this.tensorIdsByContext.has(this.backend.currentContext)||this.tensorIdsByContext.set(this.backend.currentContext,new Set),this.tensorIdsByContext.get(this.backend.currentContext)?.add(t),i.ensureTensor(r,n,o)}upload(t,r){this.tensorsById.get(t).upload(r)}async download(t,r){return de("verbose",()=>`[WebNN] TensorManager.download {tensorId: ${t}, dstBuffer: ${r?.byteLength}}`),this.tensorsById.get(t).download(r)}releaseTensorsForContext(t){let r=this.tensorIdsByContext.get(t);if(r){for(let n of r)this.tensorsById.get(n).destroy(),this.tensorsById.delete(n);this.tensorIdsByContext.delete(t)}}registerTensor(t,r,n,o){for(let[d,l]of this.tensorsById)if(l.trySelectTensor(t,r))return d;let i=Rl();this.tensorsById.set(i,new un(t,[r,n,o]));let a=this.tensorIdsByContext.get(t);return a||(a=new Set,this.tensorIdsByContext.set(t,a)),a.add(i),i}},Ul=(...e)=>new Co(...e)});var Nl,dn,Wl=U(()=>{"use strict";Z();gt();Xn();Vl();Xe();Nl=new Map([[1,"float32"],[10,"float16"],[6,"int32"],[12,"uint32"],[7,"int64"],[13,"uint64"],[3,"int8"],[2,"uint8"],[9,"uint8"]]),dn=class{constructor(t){this.tensorManager=Ul(this);this.mlContextBySessionId=new Map;this.sessionIdsByMLContext=new Map;Wr(t.logLevel,!!t.debug)}get currentSessionId(){if(this.activeSessionId===void 0)throw new Error("No active session");return this.activeSessionId}onRunStart(t){this.activeSessionId=t}get currentContext(){let t=this.getMLContext(this.currentSessionId);if(!t)throw new Error(`No MLContext found for session ${this.currentSessionId}`);return t}registerMLContext(t,r){this.mlContextBySessionId.set(t,r);let n=this.sessionIdsByMLContext.get(r);n||(n=new Set,this.sessionIdsByMLContext.set(r,n)),n.add(t)}onReleaseSession(t){let r=this.mlContextBySessionId.get(t);if(!r)return;this.mlContextBySessionId.delete(t);let n=this.sessionIdsByMLContext.get(r);n.delete(t),n.size===0&&(this.sessionIdsByMLContext.delete(r),this.tensorManager.releaseTensorsForContext(r))}getMLContext(t){return this.mlContextBySessionId.get(t)}reserveTensorId(){return this.tensorManager.reserveTensorId()}releaseTensorId(t){de("verbose",()=>`[WebNN] releaseTensorId {tensorId: ${t}}`),this.tensorManager.releaseTensorId(t)}async ensureTensor(t,r,n,o){let i=Nl.get(r);if(!i)throw new Error(`Unsupported ONNX data type: ${r}`);return this.tensorManager.ensureTensor(t,i,n,o)}uploadTensor(t,r){if(!Te().shouldTransferToMLTensor)throw new Error("Trying to upload to a MLTensor while shouldTransferToMLTensor is false");de("verbose",()=>`[WebNN] uploadTensor {tensorId: ${t}, data: ${r.byteLength}}`),this.tensorManager.upload(t,r)}async downloadTensor(t,r){return this.tensorManager.download(t,r)}createMLTensorDownloader(t,r){return async()=>{let n=await this.tensorManager.download(t);return Lr(n,r)}}registerMLTensor(t,r,n){let o=Nl.get(r);if(!o)throw new Error(`Unsupported ONNX data type: ${r}`);let i=this.tensorManager.registerTensor(this.currentContext,t,o,n);return de("verbose",()=>`[WebNN] registerMLTensor {tensor: ${t}, dataType: ${o}, dimensions: ${n}} -> {tensorId: ${i}}`),i}flush(){}}});var Ll={};Gt(Ll,{init:()=>Vh});var or,Ao,Vh,Hl=U(()=>{"use strict";Z();Ml();Xe();te();Wl();or=class e{constructor(t,r,n,o){this.module=t;this.dataType=r;this.data=n;this.dims=o}getFloat32Array(){if(this.dataType!==1)throw new Error("Invalid data type");let t=E.size(this.dims);return t===0?new Float32Array:new Float32Array(this.module.HEAP8.buffer,this.data,t)}getBigInt64Array(){if(this.dataType!==7)throw new Error("Invalid data type");let t=E.size(this.dims);return t===0?new BigInt64Array:new BigInt64Array(this.module.HEAP8.buffer,this.data,t)}getInt32Array(){if(this.dataType!==6)throw new Error("Invalid data type");let t=E.size(this.dims);return t===0?new Int32Array:new Int32Array(this.module.HEAP8.buffer,this.data,t)}getUint16Array(){if(this.dataType!==10&&this.dataType!==4)throw new Error("Invalid data type");let t=E.size(this.dims);return t===0?new Uint16Array:new Uint16Array(this.module.HEAP8.buffer,this.data,t)}reshape(t){if(E.size(t)!==E.size(this.dims))throw new Error("Invalid new shape");return new e(this.module,this.dataType,this.data,t)}},Ao=class{constructor(t,r,n){this.module=t;this.backend=r;this.customDataOffset=0;this.customDataSize=0;this.adapterInfo=r.adapterInfo;let o=t.HEAPU32,i=n>>>2;this.opKernelContext=o[i++];let a=o[i++];this.outputCount=o[i++],this.customDataOffset=o[i++],this.customDataSize=o[i++];let d=[];for(let l=0;l<a;l++){let c=o[i++],m=o[i++],u=o[i++],h=[];for(let w=0;w<u;w++)h.push(o[i++]);d.push(new or(t,c,m,h))}this.inputs=d}get kernelCustomData(){return this.backend.currentKernelCustomData}get customDataBuffer(){return this.module.HEAPU8.subarray(this.customDataOffset,this.customDataOffset+this.customDataSize)}getMaxComputeWorkgroupSizes(){return[this.backend.device.limits.maxComputeWorkgroupSizeX,this.backend.device.limits.maxComputeWorkgroupSizeY,this.backend.device.limits.maxComputeWorkgroupSizeZ]}getMaxComputeWorkgroupStoragesize(){return this.backend.device.limits.maxComputeWorkgroupStorageSize}compute(t,r){let n=r?.inputs?.map(d=>typeof d=="number"?this.inputs[d]:d)??this.inputs,o=r?.outputs??[],i=(d,l,c)=>new or(this.module,l,this.output(d,c),c),a=(d,l)=>{let c=It(d,l);if(!c)throw new Error(`Unsupported data type: ${d}`);let m=c>0?this.backend.gpuDataManager.create(c).id:0;return new or(this.module,d,m,l)};return this.backend.run(t,n,o,i,a,this.outputCount)}output(t,r){let n=this.module.stackSave();try{let o=this.module.stackAlloc((1+r.length)*4),i=o>>2;this.module.HEAPU32[i++]=r.length;for(let a=0;a<r.length;a++)this.module.HEAPU32[i++]=r[a];return this.module._JsepOutput(this.opKernelContext,t,o)}catch(o){throw new Error(`Failed to generate kernel's output[${t}] with dims [${r}]. If you are running with pre-allocated output, please make sure the output type/dims are correct. Error: ${o}`)}finally{this.module.stackRestore(n)}}},Vh=async(e,t,r,n)=>{let o=t.jsepInit;if(!o)throw new Error("Failed to initialize JSEP. The WebAssembly module is not built with JSEP support.");if(e==="webgpu"){let i=new sn;await i.initialize(r,n),o("webgpu",[i,a=>i.alloc(a),a=>i.free(a),(a,d,l,c=!1)=>{if(c)de("verbose",()=>`[WebGPU] jsepCopyGpuToGpu: src=${a}, dst=${d}, size=${l}`),i.memcpy(a,d);else{de("verbose",()=>`[WebGPU] jsepCopyCpuToGpu: dataOffset=${a}, gpuDataId=${d}, size=${l}`);let m=t.HEAPU8.subarray(a>>>0,(a>>>0)+l);i.upload(d,m)}},async(a,d,l)=>{de("verbose",()=>`[WebGPU] jsepCopyGpuToCpu: gpuDataId=${a}, dataOffset=${d}, size=${l}`),await i.download(a,()=>t.HEAPU8.subarray(d>>>0,(d>>>0)+l))},(a,d,l)=>i.createKernel(a,d,l,t.UTF8ToString(t._JsepGetNodeName(d))),a=>i.releaseKernel(a),(a,d,l,c)=>{de("verbose",()=>`[WebGPU] jsepRun: sessionHandle=${l}, kernel=${a}, contextDataOffset=${d}`);let m=new Ao(t,i,d);return i.computeKernel(a,m,c)},()=>i.captureBegin(),()=>i.captureEnd(),()=>i.replay()])}else{let i=new dn(r);o("webnn",[i,()=>i.reserveTensorId(),a=>i.releaseTensorId(a),async(a,d,l,c)=>i.ensureTensor(a,d,l,c),(a,d)=>{i.uploadTensor(a,d)},async(a,d)=>i.downloadTensor(a,d)])}}});var Nh,Er,kr,kt,Wh,jt,Pr,Or,Gl,zr,Dr,Br,Gn=U(()=>{"use strict";Va();Wa();Z();gt();Rr();Yn();Nh=(e,t)=>{Te()._OrtInit(e,t)!==0&&_e("Can't initialize onnxruntime.")},Er=async e=>{Nh(e.wasm.numThreads,Xt(e.logLevel))},kr=async(e,t)=>{{let r=(Hl(),br(Ll)).init;if(t==="webgpu"){if(typeof navigator>"u"||!navigator.gpu)throw new Error("WebGPU is not supported in current environment");let n=e.webgpu.adapter;if(n){if(typeof n.limits!="object"||typeof n.features!="object"||typeof n.requestDevice!="function")throw new Error("Invalid GPU adapter set in `env.webgpu.adapter`. It must be a GPUAdapter object.")}else{let o=e.webgpu.powerPreference;if(o!==void 0&&o!=="low-power"&&o!=="high-performance")throw new Error(`Invalid powerPreference setting: "${o}"`);let i=e.webgpu.forceFallbackAdapter;if(i!==void 0&&typeof i!="boolean")throw new Error(`Invalid forceFallbackAdapter setting: "${i}"`);if(n=await navigator.gpu.requestAdapter({powerPreference:o,forceFallbackAdapter:i}),!n)throw new Error('Failed to get GPU adapter. You may need to enable flag "--enable-unsafe-webgpu" if you are using Chrome.')}await r("webgpu",Te(),e,n)}if(t==="webnn"){if(typeof navigator>"u"||!navigator.ml)throw new Error("WebNN is not supported in current environment");await r("webnn",Te(),e)}}},kt=new Map,Wh=e=>{let t=Te(),r=t.stackSave();try{let n=t.stackAlloc(8);return t._OrtGetInputOutputCount(e,n,n+4)!==0&&_e("Can't get session input/output count."),[t.HEAP32[n/4],t.HEAP32[n/4+1]]}finally{t.stackRestore(r)}},jt=e=>{let t=Te(),r=t._malloc(e.byteLength);if(r===0)throw new Error(`Can't create a session. failed to allocate a buffer of size ${e.byteLength}.`);return t.HEAPU8.set(e,r),[r,e.byteLength]},Pr=async(e,t)=>{let r,n,o=Te();Array.isArray(e)?[r,n]=e:e.buffer===o.HEAPU8.buffer?[r,n]=[e.byteOffset,e.byteLength]:[r,n]=jt(e);let i=0,a=0,d=0,l=[],c=[],m=[];try{if([a,l]=Na(t),t?.externalData&&o.mountExternalData){let _=[];for(let x of t.externalData){let T=typeof x=="string"?x:x.path;_.push(Zt(typeof x=="string"?x:x.data).then(C=>{o.mountExternalData(T,C)}))}await Promise.all(_)}for(let _ of t?.executionProviders??[])if((typeof _=="string"?_:_.name)==="webnn"){if(o.shouldTransferToMLTensor=!1,o.currentContext)throw new Error("WebNN execution provider is already set.");if(typeof _!="string"){let T=_,C=T?.context,A=T?.gpuDevice,P=T?.deviceType,D=T?.numThreads,N=T?.powerPreference;C?o.currentContext=C:A?o.currentContext=await navigator.ml.createContext(A):o.currentContext=await navigator.ml.createContext({deviceType:P,numThreads:D,powerPreference:N})}else o.currentContext=await navigator.ml.createContext();break}i=await o._OrtCreateSession(r,n,a),i===0&&_e("Can't create a session."),o.currentContext&&(o.jsepRegisterMLContext(i,o.currentContext),o.currentContext=void 0,o.shouldTransferToMLTensor=!0);let[u,h]=Wh(i),w=!!t?.enableGraphCapture,g=[],y=[],S=[];for(let _=0;_<u;_++){let x=o._OrtGetInputName(i,_);x===0&&_e("Can't get an input name."),c.push(x),g.push(o.UTF8ToString(x))}for(let _=0;_<h;_++){let x=o._OrtGetOutputName(i,_);x===0&&_e("Can't get an output name."),m.push(x);let T=o.UTF8ToString(x);y.push(T);{if(w&&t?.preferredOutputLocation===void 0){S.push("gpu-buffer");continue}let C=typeof t?.preferredOutputLocation=="string"?t.preferredOutputLocation:t?.preferredOutputLocation?.[T]??"cpu";if(C!=="cpu"&&C!=="cpu-pinned"&&C!=="gpu-buffer"&&C!=="ml-tensor")throw new Error(`Not supported preferred output location: ${C}.`);if(w&&C!=="gpu-buffer")throw new Error(`Not supported preferred output location: ${C}. Only 'gpu-buffer' location is supported when enableGraphCapture is true.`);S.push(C)}}let $=null;return S.some(_=>_==="gpu-buffer"||_==="ml-tensor")&&(d=o._OrtCreateBinding(i),d===0&&_e("Can't create IO binding."),$={handle:d,outputPreferredLocations:S,outputPreferredLocationsEncoded:S.map(_=>Kn(_))}),kt.set(i,[i,c,m,$,w,!1]),[i,g,y]}catch(u){throw c.forEach(h=>o._OrtFree(h)),m.forEach(h=>o._OrtFree(h)),d!==0&&o._OrtReleaseBinding(d),i!==0&&o._OrtReleaseSession(i),u}finally{o._free(r),a!==0&&o._OrtReleaseSessionOptions(a),l.forEach(u=>o._free(u)),o.unmountExternalData?.()}},Or=e=>{let t=Te(),r=kt.get(e);if(!r)throw new Error(`cannot release session. invalid session id: ${e}`);let[n,o,i,a,d]=r;a&&(d&&t._OrtClearBoundOutputs(a.handle),t._OrtReleaseBinding(a.handle)),t.jsepOnReleaseSession?.(e),o.forEach(l=>t._OrtFree(l)),i.forEach(l=>t._OrtFree(l)),t._OrtReleaseSession(n),kt.delete(e)},Gl=(e,t,r,n,o,i=!1)=>{if(!e){t.push(0);return}let a=Te(),d=e[0],l=e[1],c=e[3],m,u;if(d==="string"&&(c==="gpu-buffer"||c==="ml-tensor"))throw new Error("String tensor is not supported on GPU.");if(i&&c!=="gpu-buffer")throw new Error(`External buffer must be provided for input/output index ${o} when enableGraphCapture is true.`);if(c==="gpu-buffer"){let g=e[2].gpuBuffer;u=It(Yt(d),l);let y=a.jsepRegisterBuffer;if(!y)throw new Error('Tensor location "gpu-buffer" is not supported without using WebGPU.');m=y(n,o,g,u)}else if(c==="ml-tensor"){let g=e[2].mlTensor;u=It(Yt(d),l);let y=a.jsepRegisterMLTensor;if(!y)throw new Error('Tensor location "ml-tensor" is not supported without using WebNN.');m=y(g,Yt(d),l)}else{let g=e[2];if(Array.isArray(g)){u=4*g.length,m=a._malloc(u),r.push(m);let y=m/4;for(let S=0;S<g.length;S++){if(typeof g[S]!="string")throw new TypeError(`tensor data at index ${S} is not a string`);a.HEAPU32[y++]=Ee(g[S],r)}}else u=g.byteLength,m=a._malloc(u),r.push(m),a.HEAPU8.set(new Uint8Array(g.buffer,g.byteOffset,u),m)}let h=a.stackSave(),w=a.stackAlloc(4*l.length);try{let g=w/4;l.forEach(S=>a.HEAP32[g++]=S);let y=a._OrtCreateTensor(Yt(d),m,u,w,l.length,Kn(c));y===0&&_e(`Can't create tensor for input/output. session=${n}, index=${o}.`),t.push(y)}finally{a.stackRestore(h)}},zr=async(e,t,r,n,o,i)=>{let a=Te(),d=kt.get(e);if(!d)throw new Error(`cannot run inference. invalid session id: ${e}`);let l=d[0],c=d[1],m=d[2],u=d[3],h=d[4],w=d[5],g=t.length,y=n.length,S=0,$=[],_=[],x=[],T=[],C=a.stackSave(),A=a.stackAlloc(g*4),P=a.stackAlloc(g*4),D=a.stackAlloc(y*4),N=a.stackAlloc(y*4);try{a.jsepOnRunStart?.(l),[S,$]=Ua(i);for(let Y=0;Y<g;Y++)Gl(r[Y],_,T,e,t[Y],h);for(let Y=0;Y<y;Y++)Gl(o[Y],x,T,e,g+n[Y],h);let L=A/4,K=P/4,X=D/4,ie=N/4;for(let Y=0;Y<g;Y++)a.HEAPU32[L++]=_[Y],a.HEAPU32[K++]=c[t[Y]];for(let Y=0;Y<y;Y++)a.HEAPU32[X++]=x[Y],a.HEAPU32[ie++]=m[n[Y]];if(u&&!w){let{handle:Y,outputPreferredLocations:me,outputPreferredLocationsEncoded:pe}=u;if(c.length!==g)throw new Error(`input count from feeds (${g}) is expected to be always equal to model's input count (${c.length}).`);for(let Q=0;Q<g;Q++){let be=t[Q];await a._OrtBindInput(Y,c[be],_[Q])!==0&&_e(`Can't bind input[${Q}] for session=${e}.`)}for(let Q=0;Q<y;Q++){let be=n[Q];o[Q]?.[3]?a._OrtBindOutput(Y,m[be],x[Q],0)!==0&&_e(`Can't bind pre-allocated output[${Q}] for session=${e}.`):a._OrtBindOutput(Y,m[be],0,pe[be])!==0&&_e(`Can't bind output[${Q}] to ${me[Q]} for session=${e}.`)}kt.set(e,[l,c,m,u,h,!0])}let ae;u?ae=await a._OrtRunWithBinding(l,u.handle,y,D,S):ae=await a._OrtRun(l,P,A,g,N,y,D,S),ae!==0&&_e("failed to call OrtRun().");let le=[];for(let Y=0;Y<y;Y++){let me=a.HEAPU32[D/4+Y];if(me===x[Y]){le.push(o[Y]);continue}let pe=a.stackSave(),Q=a.stackAlloc(4*4),be=!1,ne,oe=0;try{a._OrtGetTensorData(me,Q,Q+4,Q+8,Q+12)!==0&&_e(`Can't access output tensor data on index ${Y}.`);let R=Q/4,G=a.HEAPU32[R++];oe=a.HEAPU32[R++];let he=a.HEAPU32[R++],Re=a.HEAPU32[R++],$e=[];for(let Ae=0;Ae<Re;Ae++)$e.push(a.HEAPU32[he/4+Ae]);a._OrtFree(he);let Ce=$e.reduce((Ae,Me)=>Ae*Me,1);ne=yt(G);let bt=u?.outputPreferredLocations[n[Y]];if(ne==="string"){if(bt==="gpu-buffer"||bt==="ml-tensor")throw new Error("String tensor is not supported on GPU.");let Ae=[],Me=oe/4;for(let Ue=0;Ue<Ce;Ue++){let Ot=a.HEAPU32[Me++],wt=Ue===Ce-1?void 0:a.HEAPU32[Me]-Ot;Ae.push(a.UTF8ToString(Ot,wt))}le.push([ne,$e,Ae,"cpu"])}else if(bt==="gpu-buffer"&&Ce>0){let Ae=a.jsepGetBuffer;if(!Ae)throw new Error('preferredLocation "gpu-buffer" is not supported without using WebGPU.');let Me=Ae(oe),Ue=It(G,Ce);if(Ue===void 0||!Vr(ne))throw new Error(`Unsupported data type: ${ne}`);be=!0,le.push([ne,$e,{gpuBuffer:Me,download:a.jsepCreateDownloader(Me,Ue,ne),dispose:()=>{a._OrtReleaseTensor(me)}},"gpu-buffer"])}else if(bt==="ml-tensor"&&Ce>0){let Ae=a.jsepEnsureTensor;if(!Ae)throw new Error('preferredLocation "ml-tensor" is not supported without using WebNN.');if(It(G,Ce)===void 0||!Nr(ne))throw new Error(`Unsupported data type: ${ne}`);let Ue=await Ae(oe,G,$e,!1);be=!0,le.push([ne,$e,{mlTensor:Ue,download:a.jsepCreateMLTensorDownloader(oe,ne),dispose:()=>{a.jsepReleaseTensorId(oe),a._OrtReleaseTensor(me)}},"ml-tensor"])}else{let Ae=Ur(ne),Me=new Ae(Ce);new Uint8Array(Me.buffer,Me.byteOffset,Me.byteLength).set(a.HEAPU8.subarray(oe,oe+Me.byteLength)),le.push([ne,$e,Me,"cpu"])}}finally{a.stackRestore(pe),ne==="string"&&oe&&a._free(oe),be||a._OrtReleaseTensor(me)}}return u&&!h&&(a._OrtClearBoundOutputs(u.handle),kt.set(e,[l,c,m,u,h,!1])),le}finally{a.stackRestore(C),_.forEach(L=>a._OrtReleaseTensor(L)),x.forEach(L=>a._OrtReleaseTensor(L)),T.forEach(L=>a._free(L)),S!==0&&a._OrtReleaseRunOptions(S),$.forEach(L=>a._free(L))}},Dr=e=>{let t=Te(),r=kt.get(e);if(!r)throw new Error("invalid session id");let n=r[0],o=t._OrtEndProfiling(n);o===0&&_e("Can't get an profile file name."),t._OrtFree(o)},Br=e=>{let t=[];for(let r of e){let n=r[2];!Array.isArray(n)&&"buffer"in n&&t.push(n.buffer)}return t}});var Pt,Ye,ir,cn,pn,ln,Eo,ko,Lt,Ht,Hh,Fl,ql,jl,Kl,Yl,Xl,Zl,Po=U(()=>{"use strict";Ke();Gn();gt();qt();Pt=()=>!!we.wasm.proxy&&typeof document<"u",ir=!1,cn=!1,pn=!1,ko=new Map,Lt=(e,t)=>{let r=ko.get(e);r?r.push(t):ko.set(e,[t])},Ht=()=>{if(ir||!cn||pn||!Ye)throw new Error("worker not ready")},Hh=e=>{switch(e.data.type){case"init-wasm":ir=!1,e.data.err?(pn=!0,Eo[1](e.data.err)):(cn=!0,Eo[0]()),ln&&(URL.revokeObjectURL(ln),ln=void 0);break;case"init-ep":case"copy-from":case"create":case"release":case"run":case"end-profiling":{let t=ko.get(e.data.type);e.data.err?t.shift()[1](e.data.err):t.shift()[0](e.data.out);break}default:}},Fl=async()=>{if(!cn){if(ir)throw new Error("multiple calls to 'initWasm()' detected.");if(pn)throw new Error("previous call to 'initWasm()' failed.");if(ir=!0,Pt())return new Promise((e,t)=>{Ye?.terminate(),Ba().then(([r,n])=>{try{Ye=n,Ye.onerror=i=>t(i),Ye.onmessage=Hh,Eo=[e,t];let o={type:"init-wasm",in:we};Ye.postMessage(o),ln=r}catch(o){t(o)}},t)});try{await Ar(we.wasm),await Er(we),cn=!0}catch(e){throw pn=!0,e}finally{ir=!1}}},ql=async e=>{if(Pt())return Ht(),new Promise((t,r)=>{Lt("init-ep",[t,r]);let n={type:"init-ep",in:{epName:e,env:we}};Ye.postMessage(n)});await kr(we,e)},jl=async e=>Pt()?(Ht(),new Promise((t,r)=>{Lt("copy-from",[t,r]);let n={type:"copy-from",in:{buffer:e}};Ye.postMessage(n,[e.buffer])})):jt(e),Kl=async(e,t)=>{if(Pt()){if(t?.preferredOutputLocation)throw new Error('session option "preferredOutputLocation" is not supported for proxy.');return Ht(),new Promise((r,n)=>{Lt("create",[r,n]);let o={type:"create",in:{model:e,options:{...t}}},i=[];e instanceof Uint8Array&&i.push(e.buffer),Ye.postMessage(o,i)})}else return Pr(e,t)},Yl=async e=>{if(Pt())return Ht(),new Promise((t,r)=>{Lt("release",[t,r]);let n={type:"release",in:e};Ye.postMessage(n)});Or(e)},Xl=async(e,t,r,n,o,i)=>{if(Pt()){if(r.some(a=>a[3]!=="cpu"))throw new Error("input tensor on GPU is not supported for proxy.");if(o.some(a=>a))throw new Error("pre-allocated output tensor is not supported for proxy.");return Ht(),new Promise((a,d)=>{Lt("run",[a,d]);let l=r,c={type:"run",in:{sessionId:e,inputIndices:t,inputs:l,outputIndices:n,options:i}};Ye.postMessage(c,Br(l))})}else return zr(e,t,r,n,o,i)},Zl=async e=>{if(Pt())return Ht(),new Promise((t,r)=>{Lt("end-profiling",[t,r]);let n={type:"end-profiling",in:e};Ye.postMessage(n)});Dr(e)}});var Ql,Gh,mn,Jl=U(()=>{"use strict";Ke();Po();Z();Cr();Yn();Ql=(e,t)=>{switch(e.location){case"cpu":return[e.type,e.dims,e.data,"cpu"];case"gpu-buffer":return[e.type,e.dims,{gpuBuffer:e.gpuBuffer},"gpu-buffer"];case"ml-tensor":return[e.type,e.dims,{mlTensor:e.mlTensor},"ml-tensor"];default:throw new Error(`invalid data location: ${e.location} for ${t()}`)}},Gh=e=>{switch(e[3]){case"cpu":return new Be(e[0],e[2],e[1]);case"gpu-buffer":{let t=e[0];if(!Vr(t))throw new Error(`not supported data type: ${t} for deserializing GPU tensor`);let{gpuBuffer:r,download:n,dispose:o}=e[2];return Be.fromGpuBuffer(r,{dataType:t,dims:e[1],download:n,dispose:o})}case"ml-tensor":{let t=e[0];if(!Nr(t))throw new Error(`not supported data type: ${t} for deserializing MLTensor tensor`);let{mlTensor:r,download:n,dispose:o}=e[2];return Be.fromMLTensor(r,{dataType:t,dims:e[1],download:n,dispose:o})}default:throw new Error(`invalid data location: ${e[3]}`)}},mn=class{async fetchModelAndCopyToWasmMemory(t){return jl(await Zt(t))}async loadModel(t,r){Le();let n;typeof t=="string"? false?0:n=await this.fetchModelAndCopyToWasmMemory(t):n=t,[this.sessionId,this.inputNames,this.outputNames]=await Kl(n,r),Ve()}async dispose(){return Yl(this.sessionId)}async run(t,r,n){Le();let o=[],i=[];Object.entries(t).forEach(h=>{let w=h[0],g=h[1],y=this.inputNames.indexOf(w);if(y===-1)throw new Error(`invalid input '${w}'`);o.push(g),i.push(y)});let a=[],d=[];Object.entries(r).forEach(h=>{let w=h[0],g=h[1],y=this.outputNames.indexOf(w);if(y===-1)throw new Error(`invalid output '${w}'`);a.push(g),d.push(y)});let l=o.map((h,w)=>Ql(h,()=>`input "${this.inputNames[i[w]]}"`)),c=a.map((h,w)=>h?Ql(h,()=>`output "${this.outputNames[d[w]]}"`):null),m=await Xl(this.sessionId,i,l,d,c,n),u={};for(let h=0;h<m.length;h++)u[this.outputNames[d[h]]]=a[h]??Gh(m[h]);return Ve(),u}startProfiling(){}endProfiling(){Zl(this.sessionId)}}});var tc={};Gt(tc,{OnnxruntimeWebAssemblyBackend:()=>fn,initializeFlags:()=>ec,wasmBackend:()=>Fh});var ec,fn,Fh,rc=U(()=>{"use strict";Ke();Po();Jl();qt();ec=()=>{if((typeof we.wasm.initTimeout!="number"||we.wasm.initTimeout<0)&&(we.wasm.initTimeout=0),we.wasm.simd===!1&&console.warn('Deprecated property "env.wasm.simd" is set to false. non-SIMD build is no longer provided, and this setting will be ignored.'),typeof we.wasm.proxy!="boolean"&&(we.wasm.proxy=!1),typeof we.wasm.trace!="boolean"&&(we.wasm.trace=!1),typeof we.wasm.numThreads!="number"||!Number.isInteger(we.wasm.numThreads)||we.wasm.numThreads<=0)if(typeof self<"u"&&!self.crossOriginIsolated)we.wasm.numThreads=1;else{let e=typeof navigator>"u"?Vn("node:os").cpus().length:navigator.hardwareConcurrency;we.wasm.numThreads=Math.min(4,Math.ceil((e||1)/2))}},fn=class{async init(t){ec(),await Fl(),await ql(t)}async createInferenceSessionHandler(t,r){let n=new mn;return await n.loadModel(t,r),Promise.resolve(n)}},Fh=new fn});Ke();Ke();Ke();var Ia="1.20.0-dev.20241013-72cc72cc21";var Lx=Hn;{let e=(rc(),br(tc)).wasmBackend;St("webgpu",e,5),St("webnn",e,5),St("cpu",e,10),St("wasm",e,10)}Object.defineProperty(we.versions,"web",{value:Ia,enumerable:!0});
+${a}`,l=n.createShaderModule({code:d,label:t.name});me("verbose",()=>`[WebGPU] ${t.name} shader code: ${d}`);let c=n.createComputePipeline({compute:{module:l,entryPoint:"main"},layout:"auto",label:t.name});return Ve(t.name),{programInfo:t,computePipeline:c,uniformVariablesInfo:i.variablesInfo}}normalizeDispatchGroupSize(t){let r=typeof t=="number"?t:t.x,n=typeof t=="number"?1:t.y||1,o=typeof t=="number"?1:t.z||1,i=this.backend.device.limits.maxComputeWorkgroupsPerDimension;if(r<=i&&n<=i&&o<=i)return[r,n,o];let a=r*n*o,d=Math.ceil(Math.sqrt(a));if(d>i){if(d=Math.ceil(Math.cbrt(a)),d>i)throw new Error("Total dispatch size exceeds WebGPU maximum.");return[d,d,d]}else return[d,d,1]}}});var Rh,Uh,Io,sn,Ml=V(()=>{"use strict";Ke();Q();Xe();Xn();Ka();Bl();Dl();Rh=(e,t)=>{if(t.length!==e.length)throw new Error(`inputDependencies length ${t.length} is not equal to inputTensors length ${e.length}.`);let r=[];for(let n=0;n<e.length;++n){let o=e[n].dataType;switch(t[n]){case"none":{r.push("");break}case"type":{r.push(`${o}`);break}case"rank":{let i=e[n].dims.length;r.push(`${o};${i}`);break}case"dims":{let i=e[n].dims.join(",");r.push(`${o};${i}`);break}default:throw new Error(`unsupported input dependency: ${t[n]}`)}}return r.join("|")},Uh=(e,t,r)=>{let n=e.name;return e.shaderCache?.hint&&(n+="["+e.shaderCache.hint+"]"),n+=":"+r+`:${Rh(t,e.shaderCache?.inputDependencies??new Array(t.length).fill("dims"))}`,n},Io=class{constructor(t){t&&(this.architecture=t.architecture,this.vendor=t.vendor)}isArchitecture(t){return this.architecture===t}isVendor(t){return this.vendor===t}},sn=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 t=this.kernelCustomData.get(this.currentKernelId);return t||(t={},this.kernelCustomData.set(this.currentKernelId,t)),t}async initialize(t,r){this.env=t;let n=[],o={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:n};r.features.has("chromium-experimental-timestamp-query-inside-passes")?n.push("chromium-experimental-timestamp-query-inside-passes"):r.features.has("timestamp-query")&&n.push("timestamp-query"),r.features.has("shader-f16")&&n.push("shader-f16"),this.device=await r.requestDevice(o),this.adapterInfo=new Io(r.info||await r.requestAdapterInfo()),this.gpuDataManager=ja(this),this.programManager=new an(this),this.kernels=new Map,this.kernelPersistentData=new Map,this.kernelCustomData=new Map,Wr(t.logLevel,!!t.debug),this.device.onuncapturederror=i=>{i.error instanceof GPUValidationError&&console.error(`An uncaught WebGPU validation error was raised: ${i.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 t=this.getCommandEncoder(),r={};this.queryType==="at-passes"&&(r.timestampWrites={querySet:this.querySet,beginningOfPassWriteIndex:this.pendingDispatchNumber*2,endOfPassWriteIndex:this.pendingDispatchNumber*2+1}),this.computePassEncoder=t.beginComputePass(r)}return this.computePassEncoder}endComputePass(){this.computePassEncoder&&(this.computePassEncoder.end(),this.computePassEncoder=null)}flush(){if(!this.commandEncoder)return;Le(),this.endComputePass();let t;this.queryType!=="none"&&(this.commandEncoder.resolveQuerySet(this.querySet,0,this.pendingDispatchNumber*2,this.queryResolveBuffer,0),t=this.device.createBuffer({size:this.pendingDispatchNumber*2*8,usage:GPUBufferUsage.MAP_READ|GPUBufferUsage.COPY_DST}),this.pendingQueries.set(t,this.pendingKernels),this.pendingKernels=[],this.commandEncoder.copyBufferToBuffer(this.queryResolveBuffer,0,t,0,this.pendingDispatchNumber*2*8)),this.device.queue.submit([this.commandEncoder.finish()]),this.gpuDataManager.refreshPendingBuffers(),this.commandEncoder=null,this.pendingDispatchNumber=0,this.queryType!=="none"&&t.mapAsync(GPUMapMode.READ).then(()=>{let r=new BigUint64Array(t.getMappedRange()),n=this.pendingQueries.get(t);for(let o=0;o<r.length/2;o++){let i=n[o],a=i.kernelId,d=this.kernels.get(a),l=d.kernelType,c=d.kernelName,m=i.programName,u=i.inputTensorViews,h=i.outputTensorViews,w=r[o*2],g=r[o*2+1];typeof this.queryTimeBase>"u"&&(this.queryTimeBase=w);let y=Number(w-this.queryTimeBase),S=Number(g-this.queryTimeBase);if(!Number.isSafeInteger(y)||!Number.isSafeInteger(S))throw new RangeError("incorrect timestamp range");if(this.env.webgpu.profiling?.ondata)this.env.webgpu.profiling.ondata({version:1,inputsMetadata:u.map($=>({dims:$.dims,dataType:yt($.dataType)})),outputsMetadata:h.map($=>({dims:$.dims,dataType:yt($.dataType)})),kernelId:a,kernelType:l,kernelName:c,programName:m,startTime:y,endTime:S});else{let $="";u.forEach((x,T)=>{$+=`input[${T}]: [${x.dims}] | ${yt(x.dataType)}, `});let v="";h.forEach((x,T)=>{v+=`output[${T}]: [${x.dims}] | ${yt(x.dataType)}, `}),console.log(`[profiling] kernel "${a}|${l}|${c}|${m}" ${$}${v}execution time: ${S-y} ns`)}Sr("GPU",`${m}::${w}::${g}`)}t.unmap(),this.pendingQueries.delete(t)}),Ve()}run(t,r,n,o,i,a){Le(t.name);let d=[];for(let x=0;x<r.length;++x){let T=r[x].data;if(T===0)continue;let C=this.gpuDataManager.get(T);if(!C)throw new Error(`no GPU data for input: ${T}`);d.push(C)}let{outputs:l,dispatchGroup:c,programUniforms:m}=t.getRunData(r),u=n.length===0?l.map((x,T)=>T):n;if(u.length!==l.length)throw new Error(`Output size ${u.length} must be equal to ${l.length}.`);let h=[],w=[];for(let x=0;x<l.length;++x){if(!Number.isInteger(u[x])||u[x]<-3||u[x]>=a)throw new Error(`Invalid output index: ${u[x]}`);if(u[x]===-3)continue;let T=u[x]===-1,C=u[x]===-2,A=T||C?i(l[x].dataType,l[x].dims):o(u[x],l[x].dataType,l[x].dims);if(h.push(A),A.data===0)continue;let P=this.gpuDataManager.get(A.data);if(!P)throw new Error(`no GPU data for output: ${A.data}`);if(T&&this.temporaryData.push(P),C){let B=this.kernelPersistentData.get(this.currentKernelId);B||(B=[],this.kernelPersistentData.set(this.currentKernelId,B)),B.push(P)}w.push(P)}if(d.length!==r.length||w.length!==h.length){if(w.length===0)return Ve(t.name),h;throw new Error(`Program ${t.name} has zero-sized tensor(s) in inputs or outputs. This is not supported now.`)}let g;if(m){let x=0,T=[];m.forEach(B=>{let N=typeof B.data=="number"?[B.data]:B.data;if(N.length===0)return;let W=B.type===10?2:4,K,Z;B.type===10?(Z=N.length>4?16:N.length>2?8:N.length*W,K=N.length>4?16:W*N.length):(Z=N.length<=2?N.length*W:16,K=16),x=Math.ceil(x/Z)*Z,T.push(x);let ee=B.type===10?8:4;x+=N.length>4?Math.ceil(N.length/ee)*K:N.length*W});let C=16;x=Math.ceil(x/C)*C;let A=new ArrayBuffer(x);m.forEach((B,N)=>{let W=T[N],K=typeof B.data=="number"?[B.data]:B.data;if(B.type===6)new Int32Array(A,W,K.length).set(K);else if(B.type===12)new Uint32Array(A,W,K.length).set(K);else if(B.type===10)new Uint16Array(A,W,K.length).set(K);else if(B.type===1)new Float32Array(A,W,K.length).set(K);else throw new Error(`Unsupported uniform type: ${yt(B.type)}`)});let P=this.gpuDataManager.create(x,GPUBufferUsage.COPY_DST|GPUBufferUsage.UNIFORM);this.device.queue.writeBuffer(P.buffer,0,A,0,x),this.gpuDataManager.release(P.id),g={offset:0,size:x,buffer:P.buffer}}let y=this.programManager.normalizeDispatchGroupSize(c),S=y[1]===1&&y[2]===1,$=Uh(t,r,S),v=this.programManager.getArtifact($);if(v||(v=this.programManager.build(t,y),this.programManager.setArtifact($,v),me("info",()=>`[artifact] key: ${$}, programName: ${t.name}`)),m&&v.uniformVariablesInfo){if(m.length!==v.uniformVariablesInfo.length)throw new Error(`Uniform variables count mismatch: expect ${v.uniformVariablesInfo.length}, got ${m.length} in program "${v.programInfo.name}".`);for(let x=0;x<m.length;x++){let T=m[x],C=T.type,A=typeof T.data=="number"?1:T.data.length,[P,B]=v.uniformVariablesInfo[x];if(C!==P||A!==B)throw new Error(`Uniform variable ${x} mismatch: expect type ${P} with size ${B}, got type ${C} with size ${A} in program "${v.programInfo.name}".`)}}if(me("info",()=>`[ProgramManager] run "${t.name}" (key=${$}) with ${y[0]}x${y[1]}x${y[2]}`),this.queryType!=="none"||this.sessionStatus==="capturing"){let x={kernelId:this.currentKernelId,programName:v.programInfo.name,inputTensorViews:r,outputTensorViews:h};this.pendingKernels.push(x),this.sessionStatus==="capturing"&&this.capturedPendingKernels.get(this.currentSessionId).push(x)}return this.programManager.run(v,d,w,y,g),Ve(t.name),h}upload(t,r){this.gpuDataManager.upload(t,r)}memcpy(t,r){this.gpuDataManager.memcpy(t,r)}async download(t,r){await this.gpuDataManager.download(t,r)}alloc(t){return this.gpuDataManager.create(t).id}free(t){return this.gpuDataManager.release(t)}createKernel(t,r,n,o){let i=Ol.get(t);if(!i)throw new Error(`kernel not implemented: ${t}`);let a={kernelType:t,kernelName:o,kernelEntry:i[0],attributes:[i[1],n]};this.kernels.set(r,a)}releaseKernel(t){let r=this.kernelPersistentData.get(t);if(r){for(let n of r)this.gpuDataManager.release(n.id);this.kernelPersistentData.delete(t)}this.kernelCustomData.delete(t),this.kernels.delete(t)}computeKernel(t,r,n){let o=this.kernels.get(t);if(!o)throw new Error(`kernel not created: ${t}`);let i=o.kernelType,a=o.kernelName,d=o.kernelEntry,l=o.attributes;if(this.currentKernelId!==null)throw new Error(`kernel "[${i}] ${a}" is not allowed to be called recursively`);this.currentKernelId=t,l[0]&&(l[1]=l[0](l[1]),l[0]=void 0),me("info",()=>`[WebGPU] Start to run kernel "[${i}] ${a}"...`);let c=this.env.debug;this.temporaryData=[];try{return c&&this.device.pushErrorScope("validation"),d(r,l[1]),0}catch(m){return n.push(Promise.resolve(`[WebGPU] Kernel "[${i}] ${a}" failed. ${m}`)),1}finally{c&&n.push(this.device.popErrorScope().then(m=>m?`GPU validation error for kernel "[${i}] ${a}": ${m.message}`:null));for(let m of this.temporaryData)this.gpuDataManager.release(m.id);this.temporaryData=[],this.currentKernelId=null}}registerBuffer(t,r,n,o){let i=this.sessionExternalDataMapping.get(t);i||(i=new Map,this.sessionExternalDataMapping.set(t,i));let a=i.get(r),d=this.gpuDataManager.registerExternalBuffer(n,o,a);return i.set(r,[d,n]),d}unregisterBuffers(t){let r=this.sessionExternalDataMapping.get(t);r&&(r.forEach(n=>this.gpuDataManager.unregisterExternalBuffer(n[0])),this.sessionExternalDataMapping.delete(t))}getBuffer(t){let r=this.gpuDataManager.get(t);if(!r)throw new Error(`no GPU data for buffer: ${t}`);return r.buffer}createDownloader(t,r,n){return async()=>{let o=await eo(this,t,r);return Lr(o.buffer,n)}}writeTimestamp(t){this.queryType==="inside-passes"&&this.computePassEncoder.writeTimestamp(this.querySet,t)}setQueryType(){this.queryType="none",(this.env.webgpu.profiling?.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(){me("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(){me("info","captureEnd"),this.flush(),this.sessionStatus="default"}replay(){me("info","replay"),this.sessionStatus="replaying";let t=this.capturedCommandList.get(this.currentSessionId),r=this.capturedPendingKernels.get(this.currentSessionId),n=t.length;this.pendingKernels=[];for(let o=0;o<n;o++){let i=this.getComputePassEncoder(),a=t[o];this.writeTimestamp(this.pendingDispatchNumber*2),i.setPipeline(a.computePipeline),i.setBindGroup(0,a.bindGroup),i.dispatchWorkgroups(...a.dispatchGroup),this.writeTimestamp(this.pendingDispatchNumber*2+1),this.pendingDispatchNumber++,this.queryType!=="none"&&this.pendingKernels.push(r[o]),(this.pendingDispatchNumber>=this.maxDispatchNumber||this.queryType==="at-passes")&&this.endComputePass(),this.pendingDispatchNumber>=this.maxDispatchNumber&&this.flush()}this.flush(),this.sessionStatus="default"}onReleaseSession(t){this.unregisterBuffers(t),this.capturedCommandList.has(t)&&this.capturedCommandList.delete(t),this.capturedPendingKernels.has(t)&&this.capturedPendingKernels.delete(t),this.gpuDataManager.onReleaseSession(t)}onRunStart(t){this.currentSessionId=t,this.setQueryType()}}});var Vh,Rl,un,Co,Ul,Vl=V(()=>{"use strict";Xe();Vh=1,Rl=()=>Vh++,un=class{constructor(t,r){this.mlContext=t;this.tensorEntry=r,this.tensorCache=r?[r]:[]}get tensor(){return this.tensorEntry?.[0]}get context(){if(!this.mlContext)throw new Error("MLContext has not been set.");return this.mlContext}set context(t){if(this.mlContext&&this.mlContext!==t)throw new Error("MLTensor in use in a different MLContext.");this.mlContext=t}destroy(){for(let[t]of this.tensorCache)t.destroy();this.tensorCache=[],this.tensorEntry=void 0}trySelectTensor(t,r){for(let[n,o,i]of this.tensorCache)if(r===n){if(this.context!==t)throw new Error("MLTensor cannot be registered with a different MLContext.");return this.tensorEntry=[n,o,i],!0}return!1}async ensureTensor(t,r,n){if(this.tensorEntry){let[a,d,l]=this.tensorEntry;if(d===t&&l.every((c,m)=>c===r[m]))return a}for(let[a,d,l]of this.tensorCache)if(d===t&&l.every((c,m)=>c===r[m])){if(n&&this.tensorEntry){me("verbose",()=>`[WebNN] Slowdown may occur, having to copy existing tensor {dataType: ${t}, shape: ${r}}`);let c=await this.context.readTensor(this.tensorEntry[0]);this.context.writeTensor(a,c)}return this.tensorEntry=[a,d,l],a}me("verbose",()=>`[WebNN] MLContext.createTensor {dataType: ${t}, shape: ${r}}`);let o=MLTensorUsage.READ|MLTensorUsage.WRITE,i=await this.context.createTensor({dataType:t,shape:r,dimensions:r,usage:o});return this.tensorEntry=[i,t,r],this.tensorCache.push(this.tensorEntry),this.activeUpload&&(this.mlContext?.writeTensor(i,this.activeUpload),this.activeUpload=void 0),i}upload(t){if(!this.tensorEntry){this.activeUpload=new Uint8Array(t);return}this.mlContext?.writeTensor(this.tensorEntry[0],t)}async download(t){if(this.activeUpload)if(t){t instanceof ArrayBuffer?new Uint8Array(t).set(this.activeUpload):new Uint8Array(t.buffer,t.byteOffset,t.byteLength).set(this.activeUpload);return}else return this.activeUpload.buffer;if(!this.tensorEntry)throw new Error("Tensor has not been created.");return t?this.context.readTensor(this.tensorEntry[0],t):this.context.readTensor(this.tensorEntry[0])}},Co=class{constructor(t){this.backend=t;this.tensorsById=new Map;this.tensorIdsByContext=new Map}reserveTensorId(){let t=Rl();return this.tensorsById.set(t,new un),t}releaseTensorId(t){let r=this.tensorsById.get(t);if(r){r.destroy(),this.tensorsById.delete(t);for(let[n,o]of this.tensorIdsByContext)if(o.has(t)){o.delete(t),o.size===0&&this.tensorIdsByContext.delete(n);break}}}async ensureTensor(t,r,n,o){me("verbose",()=>`[WebNN] TensorManager.ensureTensor {tensorId: ${t}, dataType: ${r}, shape: ${n}, copyOld: ${o}}`);let i=this.tensorsById.get(t);if(!i)throw new Error("Tensor not found.");return i.context=this.backend.currentContext,this.tensorIdsByContext.has(this.backend.currentContext)||this.tensorIdsByContext.set(this.backend.currentContext,new Set),this.tensorIdsByContext.get(this.backend.currentContext)?.add(t),i.ensureTensor(r,n,o)}upload(t,r){this.tensorsById.get(t).upload(r)}async download(t,r){return me("verbose",()=>`[WebNN] TensorManager.download {tensorId: ${t}, dstBuffer: ${r?.byteLength}}`),this.tensorsById.get(t).download(r)}releaseTensorsForContext(t){let r=this.tensorIdsByContext.get(t);if(r){for(let n of r)this.tensorsById.get(n).destroy(),this.tensorsById.delete(n);this.tensorIdsByContext.delete(t)}}registerTensor(t,r,n,o){for(let[d,l]of this.tensorsById)if(l.trySelectTensor(t,r))return d;let i=Rl();this.tensorsById.set(i,new un(t,[r,n,o]));let a=this.tensorIdsByContext.get(t);return a||(a=new Set,this.tensorIdsByContext.set(t,a)),a.add(i),i}},Ul=(...e)=>new Co(...e)});var Nl,dn,Wl=V(()=>{"use strict";Q();gt();Xn();Vl();Xe();Nl=new Map([[1,"float32"],[10,"float16"],[6,"int32"],[12,"uint32"],[7,"int64"],[13,"uint64"],[3,"int8"],[2,"uint8"],[9,"uint8"]]),dn=class{constructor(t){this.tensorManager=Ul(this);this.mlContextBySessionId=new Map;this.sessionIdsByMLContext=new Map;Wr(t.logLevel,!!t.debug)}get currentSessionId(){if(this.activeSessionId===void 0)throw new Error("No active session");return this.activeSessionId}onRunStart(t){this.activeSessionId=t}get currentContext(){let t=this.getMLContext(this.currentSessionId);if(!t)throw new Error(`No MLContext found for session ${this.currentSessionId}`);return t}registerMLContext(t,r){this.mlContextBySessionId.set(t,r);let n=this.sessionIdsByMLContext.get(r);n||(n=new Set,this.sessionIdsByMLContext.set(r,n)),n.add(t)}onReleaseSession(t){let r=this.mlContextBySessionId.get(t);if(!r)return;this.mlContextBySessionId.delete(t);let n=this.sessionIdsByMLContext.get(r);n.delete(t),n.size===0&&(this.sessionIdsByMLContext.delete(r),this.tensorManager.releaseTensorsForContext(r))}getMLContext(t){return this.mlContextBySessionId.get(t)}reserveTensorId(){return this.tensorManager.reserveTensorId()}releaseTensorId(t){me("verbose",()=>`[WebNN] releaseTensorId {tensorId: ${t}}`),this.tensorManager.releaseTensorId(t)}async ensureTensor(t,r,n,o){let i=Nl.get(r);if(!i)throw new Error(`Unsupported ONNX data type: ${r}`);return this.tensorManager.ensureTensor(t,i,n,o)}uploadTensor(t,r){if(!Te().shouldTransferToMLTensor)throw new Error("Trying to upload to a MLTensor while shouldTransferToMLTensor is false");me("verbose",()=>`[WebNN] uploadTensor {tensorId: ${t}, data: ${r.byteLength}}`),this.tensorManager.upload(t,r)}async downloadTensor(t,r){return this.tensorManager.download(t,r)}createMLTensorDownloader(t,r){return async()=>{let n=await this.tensorManager.download(t);return Lr(n,r)}}registerMLTensor(t,r,n){let o=Nl.get(r);if(!o)throw new Error(`Unsupported ONNX data type: ${r}`);let i=this.tensorManager.registerTensor(this.currentContext,t,o,n);return me("verbose",()=>`[WebNN] registerMLTensor {tensor: ${t}, dataType: ${o}, dimensions: ${n}} -> {tensorId: ${i}}`),i}flush(){}}});var Ll={};Gt(Ll,{init:()=>Nh});var or,Ao,Nh,Hl=V(()=>{"use strict";Q();Ml();Xe();ie();Wl();or=class e{constructor(t,r,n,o){this.module=t;this.dataType=r;this.data=n;this.dims=o}getFloat32Array(){if(this.dataType!==1)throw new Error("Invalid data type");let t=k.size(this.dims);return t===0?new Float32Array:new Float32Array(this.module.HEAP8.buffer,this.data,t)}getBigInt64Array(){if(this.dataType!==7)throw new Error("Invalid data type");let t=k.size(this.dims);return t===0?new BigInt64Array:new BigInt64Array(this.module.HEAP8.buffer,this.data,t)}getInt32Array(){if(this.dataType!==6)throw new Error("Invalid data type");let t=k.size(this.dims);return t===0?new Int32Array:new Int32Array(this.module.HEAP8.buffer,this.data,t)}getUint16Array(){if(this.dataType!==10&&this.dataType!==4)throw new Error("Invalid data type");let t=k.size(this.dims);return t===0?new Uint16Array:new Uint16Array(this.module.HEAP8.buffer,this.data,t)}reshape(t){if(k.size(t)!==k.size(this.dims))throw new Error("Invalid new shape");return new e(this.module,this.dataType,this.data,t)}},Ao=class{constructor(t,r,n){this.module=t;this.backend=r;this.customDataOffset=0;this.customDataSize=0;this.adapterInfo=r.adapterInfo;let o=t.HEAPU32,i=n>>>2;this.opKernelContext=o[i++];let a=o[i++];this.outputCount=o[i++],this.customDataOffset=o[i++],this.customDataSize=o[i++];let d=[];for(let l=0;l<a;l++){let c=o[i++],m=o[i++],u=o[i++],h=[];for(let w=0;w<u;w++)h.push(o[i++]);d.push(new or(t,c,m,h))}this.inputs=d}get kernelCustomData(){return this.backend.currentKernelCustomData}get customDataBuffer(){return this.module.HEAPU8.subarray(this.customDataOffset,this.customDataOffset+this.customDataSize)}getMaxComputeWorkgroupSizes(){return[this.backend.device.limits.maxComputeWorkgroupSizeX,this.backend.device.limits.maxComputeWorkgroupSizeY,this.backend.device.limits.maxComputeWorkgroupSizeZ]}getMaxComputeWorkgroupStoragesize(){return this.backend.device.limits.maxComputeWorkgroupStorageSize}compute(t,r){let n=r?.inputs?.map(d=>typeof d=="number"?this.inputs[d]:d)??this.inputs,o=r?.outputs??[],i=(d,l,c)=>new or(this.module,l,this.output(d,c),c),a=(d,l)=>{let c=It(d,l);if(!c)throw new Error(`Unsupported data type: ${d}`);let m=c>0?this.backend.gpuDataManager.create(c).id:0;return new or(this.module,d,m,l)};return this.backend.run(t,n,o,i,a,this.outputCount)}output(t,r){let n=this.module.stackSave();try{let o=this.module.stackAlloc((1+r.length)*4),i=o>>2;this.module.HEAPU32[i++]=r.length;for(let a=0;a<r.length;a++)this.module.HEAPU32[i++]=r[a];return this.module._JsepOutput(this.opKernelContext,t,o)}catch(o){throw new Error(`Failed to generate kernel's output[${t}] with dims [${r}]. If you are running with pre-allocated output, please make sure the output type/dims are correct. Error: ${o}`)}finally{this.module.stackRestore(n)}}},Nh=async(e,t,r,n)=>{let o=t.jsepInit;if(!o)throw new Error("Failed to initialize JSEP. The WebAssembly module is not built with JSEP support.");if(e==="webgpu"){let i=new sn;await i.initialize(r,n),o("webgpu",[i,a=>i.alloc(a),a=>i.free(a),(a,d,l,c=!1)=>{if(c)me("verbose",()=>`[WebGPU] jsepCopyGpuToGpu: src=${a}, dst=${d}, size=${l}`),i.memcpy(a,d);else{me("verbose",()=>`[WebGPU] jsepCopyCpuToGpu: dataOffset=${a}, gpuDataId=${d}, size=${l}`);let m=t.HEAPU8.subarray(a>>>0,(a>>>0)+l);i.upload(d,m)}},async(a,d,l)=>{me("verbose",()=>`[WebGPU] jsepCopyGpuToCpu: gpuDataId=${a}, dataOffset=${d}, size=${l}`),await i.download(a,()=>t.HEAPU8.subarray(d>>>0,(d>>>0)+l))},(a,d,l)=>i.createKernel(a,d,l,t.UTF8ToString(t._JsepGetNodeName(d))),a=>i.releaseKernel(a),(a,d,l,c)=>{me("verbose",()=>`[WebGPU] jsepRun: sessionHandle=${l}, kernel=${a}, contextDataOffset=${d}`);let m=new Ao(t,i,d);return i.computeKernel(a,m,c)},()=>i.captureBegin(),()=>i.captureEnd(),()=>i.replay()])}else{let i=new dn(r);o("webnn",[i,()=>i.reserveTensorId(),a=>i.releaseTensorId(a),async(a,d,l,c)=>i.ensureTensor(a,d,l,c),(a,d)=>{i.uploadTensor(a,d)},async(a,d)=>i.downloadTensor(a,d)])}}});var Wh,kr,Er,Et,Lh,jt,Pr,zr,Gl,Or,Br,Dr,Gn=V(()=>{"use strict";Va();Wa();Q();gt();Rr();Yn();Wh=(e,t)=>{Te()._OrtInit(e,t)!==0&&ve("Can't initialize onnxruntime.")},kr=async e=>{Wh(e.wasm.numThreads,Xt(e.logLevel))},Er=async(e,t)=>{{let r=(Hl(),br(Ll)).init;if(t==="webgpu"){if(typeof navigator>"u"||!navigator.gpu)throw new Error("WebGPU is not supported in current environment");let n=e.webgpu.adapter;if(n){if(typeof n.limits!="object"||typeof n.features!="object"||typeof n.requestDevice!="function")throw new Error("Invalid GPU adapter set in `env.webgpu.adapter`. It must be a GPUAdapter object.")}else{let o=e.webgpu.powerPreference;if(o!==void 0&&o!=="low-power"&&o!=="high-performance")throw new Error(`Invalid powerPreference setting: "${o}"`);let i=e.webgpu.forceFallbackAdapter;if(i!==void 0&&typeof i!="boolean")throw new Error(`Invalid forceFallbackAdapter setting: "${i}"`);if(n=await navigator.gpu.requestAdapter({powerPreference:o,forceFallbackAdapter:i}),!n)throw new Error('Failed to get GPU adapter. You may need to enable flag "--enable-unsafe-webgpu" if you are using Chrome.')}await r("webgpu",Te(),e,n)}if(t==="webnn"){if(typeof navigator>"u"||!navigator.ml)throw new Error("WebNN is not supported in current environment");await r("webnn",Te(),e)}}},Et=new Map,Lh=e=>{let t=Te(),r=t.stackSave();try{let n=t.stackAlloc(8);return t._OrtGetInputOutputCount(e,n,n+4)!==0&&ve("Can't get session input/output count."),[t.HEAP32[n/4],t.HEAP32[n/4+1]]}finally{t.stackRestore(r)}},jt=e=>{let t=Te(),r=t._malloc(e.byteLength);if(r===0)throw new Error(`Can't create a session. failed to allocate a buffer of size ${e.byteLength}.`);return t.HEAPU8.set(e,r),[r,e.byteLength]},Pr=async(e,t)=>{let r,n,o=Te();Array.isArray(e)?[r,n]=e:e.buffer===o.HEAPU8.buffer?[r,n]=[e.byteOffset,e.byteLength]:[r,n]=jt(e);let i=0,a=0,d=0,l=[],c=[],m=[];try{if([a,l]=Na(t),t?.externalData&&o.mountExternalData){let v=[];for(let x of t.externalData){let T=typeof x=="string"?x:x.path;v.push(Zt(typeof x=="string"?x:x.data).then(C=>{o.mountExternalData(T,C)}))}await Promise.all(v)}for(let v of t?.executionProviders??[])if((typeof v=="string"?v:v.name)==="webnn"){if(o.shouldTransferToMLTensor=!1,o.currentContext)throw new Error("WebNN execution provider is already set.");if(typeof v!="string"){let T=v,C=T?.context,A=T?.gpuDevice,P=T?.deviceType,B=T?.numThreads,N=T?.powerPreference;C?o.currentContext=C:A?o.currentContext=await navigator.ml.createContext(A):o.currentContext=await navigator.ml.createContext({deviceType:P,numThreads:B,powerPreference:N})}else o.currentContext=await navigator.ml.createContext();break}i=await o._OrtCreateSession(r,n,a),i===0&&ve("Can't create a session."),o.currentContext&&(o.jsepRegisterMLContext(i,o.currentContext),o.currentContext=void 0,o.shouldTransferToMLTensor=!0);let[u,h]=Lh(i),w=!!t?.enableGraphCapture,g=[],y=[],S=[];for(let v=0;v<u;v++){let x=o._OrtGetInputName(i,v);x===0&&ve("Can't get an input name."),c.push(x),g.push(o.UTF8ToString(x))}for(let v=0;v<h;v++){let x=o._OrtGetOutputName(i,v);x===0&&ve("Can't get an output name."),m.push(x);let T=o.UTF8ToString(x);y.push(T);{if(w&&t?.preferredOutputLocation===void 0){S.push("gpu-buffer");continue}let C=typeof t?.preferredOutputLocation=="string"?t.preferredOutputLocation:t?.preferredOutputLocation?.[T]??"cpu";if(C!=="cpu"&&C!=="cpu-pinned"&&C!=="gpu-buffer"&&C!=="ml-tensor")throw new Error(`Not supported preferred output location: ${C}.`);if(w&&C!=="gpu-buffer")throw new Error(`Not supported preferred output location: ${C}. Only 'gpu-buffer' location is supported when enableGraphCapture is true.`);S.push(C)}}let $=null;return S.some(v=>v==="gpu-buffer"||v==="ml-tensor")&&(d=o._OrtCreateBinding(i),d===0&&ve("Can't create IO binding."),$={handle:d,outputPreferredLocations:S,outputPreferredLocationsEncoded:S.map(v=>Kn(v))}),Et.set(i,[i,c,m,$,w,!1]),[i,g,y]}catch(u){throw c.forEach(h=>o._OrtFree(h)),m.forEach(h=>o._OrtFree(h)),d!==0&&o._OrtReleaseBinding(d),i!==0&&o._OrtReleaseSession(i),u}finally{o._free(r),a!==0&&o._OrtReleaseSessionOptions(a),l.forEach(u=>o._free(u)),o.unmountExternalData?.()}},zr=e=>{let t=Te(),r=Et.get(e);if(!r)throw new Error(`cannot release session. invalid session id: ${e}`);let[n,o,i,a,d]=r;a&&(d&&t._OrtClearBoundOutputs(a.handle),t._OrtReleaseBinding(a.handle)),t.jsepOnReleaseSession?.(e),o.forEach(l=>t._OrtFree(l)),i.forEach(l=>t._OrtFree(l)),t._OrtReleaseSession(n),Et.delete(e)},Gl=(e,t,r,n,o,i=!1)=>{if(!e){t.push(0);return}let a=Te(),d=e[0],l=e[1],c=e[3],m,u;if(d==="string"&&(c==="gpu-buffer"||c==="ml-tensor"))throw new Error("String tensor is not supported on GPU.");if(i&&c!=="gpu-buffer")throw new Error(`External buffer must be provided for input/output index ${o} when enableGraphCapture is true.`);if(c==="gpu-buffer"){let g=e[2].gpuBuffer;u=It(Yt(d),l);let y=a.jsepRegisterBuffer;if(!y)throw new Error('Tensor location "gpu-buffer" is not supported without using WebGPU.');m=y(n,o,g,u)}else if(c==="ml-tensor"){let g=e[2].mlTensor;u=It(Yt(d),l);let y=a.jsepRegisterMLTensor;if(!y)throw new Error('Tensor location "ml-tensor" is not supported without using WebNN.');m=y(g,Yt(d),l)}else{let g=e[2];if(Array.isArray(g)){u=4*g.length,m=a._malloc(u),r.push(m);let y=m/4;for(let S=0;S<g.length;S++){if(typeof g[S]!="string")throw new TypeError(`tensor data at index ${S} is not a string`);a.HEAPU32[y++]=ke(g[S],r)}}else u=g.byteLength,m=a._malloc(u),r.push(m),a.HEAPU8.set(new Uint8Array(g.buffer,g.byteOffset,u),m)}let h=a.stackSave(),w=a.stackAlloc(4*l.length);try{let g=w/4;l.forEach(S=>a.HEAP32[g++]=S);let y=a._OrtCreateTensor(Yt(d),m,u,w,l.length,Kn(c));y===0&&ve(`Can't create tensor for input/output. session=${n}, index=${o}.`),t.push(y)}finally{a.stackRestore(h)}},Or=async(e,t,r,n,o,i)=>{let a=Te(),d=Et.get(e);if(!d)throw new Error(`cannot run inference. invalid session id: ${e}`);let l=d[0],c=d[1],m=d[2],u=d[3],h=d[4],w=d[5],g=t.length,y=n.length,S=0,$=[],v=[],x=[],T=[],C=a.stackSave(),A=a.stackAlloc(g*4),P=a.stackAlloc(g*4),B=a.stackAlloc(y*4),N=a.stackAlloc(y*4);try{a.jsepOnRunStart?.(l),[S,$]=Ua(i);for(let Y=0;Y<g;Y++)Gl(r[Y],v,T,e,t[Y],h);for(let Y=0;Y<y;Y++)Gl(o[Y],x,T,e,g+n[Y],h);let W=A/4,K=P/4,Z=B/4,ee=N/4;for(let Y=0;Y<g;Y++)a.HEAPU32[W++]=v[Y],a.HEAPU32[K++]=c[t[Y]];for(let Y=0;Y<y;Y++)a.HEAPU32[Z++]=x[Y],a.HEAPU32[ee++]=m[n[Y]];if(u&&!w){let{handle:Y,outputPreferredLocations:le,outputPreferredLocationsEncoded:ce}=u;if(c.length!==g)throw new Error(`input count from feeds (${g}) is expected to be always equal to model's input count (${c.length}).`);for(let q=0;q<g;q++){let ue=t[q];await a._OrtBindInput(Y,c[ue],v[q])!==0&&ve(`Can't bind input[${q}] for session=${e}.`)}for(let q=0;q<y;q++){let ue=n[q];o[q]?.[3]?a._OrtBindOutput(Y,m[ue],x[q],0)!==0&&ve(`Can't bind pre-allocated output[${q}] for session=${e}.`):a._OrtBindOutput(Y,m[ue],0,ce[ue])!==0&&ve(`Can't bind output[${q}] to ${le[q]} for session=${e}.`)}Et.set(e,[l,c,m,u,h,!0])}let se;u?se=await a._OrtRunWithBinding(l,u.handle,y,B,S):se=await a._OrtRun(l,P,A,g,N,y,B,S),se!==0&&ve("failed to call OrtRun().");let de=[];for(let Y=0;Y<y;Y++){let le=a.HEAPU32[B/4+Y];if(le===x[Y]){de.push(o[Y]);continue}let ce=a.stackSave(),q=a.stackAlloc(4*4),ue=!1,re,ne=0;try{a._OrtGetTensorData(le,q,q+4,q+8,q+12)!==0&&ve(`Can't access output tensor data on index ${Y}.`);let U=q/4,G=a.HEAPU32[U++];ne=a.HEAPU32[U++];let ye=a.HEAPU32[U++],Re=a.HEAPU32[U++],$e=[];for(let Ae=0;Ae<Re;Ae++)$e.push(a.HEAPU32[ye/4+Ae]);a._OrtFree(ye);let Ce=$e.reduce((Ae,Me)=>Ae*Me,1);re=yt(G);let bt=u?.outputPreferredLocations[n[Y]];if(re==="string"){if(bt==="gpu-buffer"||bt==="ml-tensor")throw new Error("String tensor is not supported on GPU.");let Ae=[],Me=ne/4;for(let Ue=0;Ue<Ce;Ue++){let zt=a.HEAPU32[Me++],wt=Ue===Ce-1?void 0:a.HEAPU32[Me]-zt;Ae.push(a.UTF8ToString(zt,wt))}de.push([re,$e,Ae,"cpu"])}else if(bt==="gpu-buffer"&&Ce>0){let Ae=a.jsepGetBuffer;if(!Ae)throw new Error('preferredLocation "gpu-buffer" is not supported without using WebGPU.');let Me=Ae(ne),Ue=It(G,Ce);if(Ue===void 0||!Vr(re))throw new Error(`Unsupported data type: ${re}`);ue=!0,de.push([re,$e,{gpuBuffer:Me,download:a.jsepCreateDownloader(Me,Ue,re),dispose:()=>{a._OrtReleaseTensor(le)}},"gpu-buffer"])}else if(bt==="ml-tensor"&&Ce>0){let Ae=a.jsepEnsureTensor;if(!Ae)throw new Error('preferredLocation "ml-tensor" is not supported without using WebNN.');if(It(G,Ce)===void 0||!Nr(re))throw new Error(`Unsupported data type: ${re}`);let Ue=await Ae(ne,G,$e,!1);ue=!0,de.push([re,$e,{mlTensor:Ue,download:a.jsepCreateMLTensorDownloader(ne,re),dispose:()=>{a.jsepReleaseTensorId(ne),a._OrtReleaseTensor(le)}},"ml-tensor"])}else{let Ae=Ur(re),Me=new Ae(Ce);new Uint8Array(Me.buffer,Me.byteOffset,Me.byteLength).set(a.HEAPU8.subarray(ne,ne+Me.byteLength)),de.push([re,$e,Me,"cpu"])}}finally{a.stackRestore(ce),re==="string"&&ne&&a._free(ne),ue||a._OrtReleaseTensor(le)}}return u&&!h&&(a._OrtClearBoundOutputs(u.handle),Et.set(e,[l,c,m,u,h,!1])),de}finally{a.stackRestore(C),v.forEach(W=>a._OrtReleaseTensor(W)),x.forEach(W=>a._OrtReleaseTensor(W)),T.forEach(W=>a._free(W)),S!==0&&a._OrtReleaseRunOptions(S),$.forEach(W=>a._free(W))}},Br=e=>{let t=Te(),r=Et.get(e);if(!r)throw new Error("invalid session id");let n=r[0],o=t._OrtEndProfiling(n);o===0&&ve("Can't get an profile file name."),t._OrtFree(o)},Dr=e=>{let t=[];for(let r of e){let n=r[2];!Array.isArray(n)&&"buffer"in n&&t.push(n.buffer)}return t}});var Pt,Ye,ir,cn,pn,ln,ko,Eo,Lt,Ht,Gh,Fl,ql,jl,Kl,Yl,Xl,Zl,Po=V(()=>{"use strict";Ke();Gn();gt();qt();Pt=()=>!!_e.wasm.proxy&&typeof document<"u",ir=!1,cn=!1,pn=!1,Eo=new Map,Lt=(e,t)=>{let r=Eo.get(e);r?r.push(t):Eo.set(e,[t])},Ht=()=>{if(ir||!cn||pn||!Ye)throw new Error("worker not ready")},Gh=e=>{switch(e.data.type){case"init-wasm":ir=!1,e.data.err?(pn=!0,ko[1](e.data.err)):(cn=!0,ko[0]()),ln&&(URL.revokeObjectURL(ln),ln=void 0);break;case"init-ep":case"copy-from":case"create":case"release":case"run":case"end-profiling":{let t=Eo.get(e.data.type);e.data.err?t.shift()[1](e.data.err):t.shift()[0](e.data.out);break}default:}},Fl=async()=>{if(!cn){if(ir)throw new Error("multiple calls to 'initWasm()' detected.");if(pn)throw new Error("previous call to 'initWasm()' failed.");if(ir=!0,Pt())return new Promise((e,t)=>{Ye?.terminate(),Da().then(([r,n])=>{try{Ye=n,Ye.onerror=i=>t(i),Ye.onmessage=Gh,ko=[e,t];let o={type:"init-wasm",in:_e};Ye.postMessage(o),ln=r}catch(o){t(o)}},t)});try{await Ar(_e.wasm),await kr(_e),cn=!0}catch(e){throw pn=!0,e}finally{ir=!1}}},ql=async e=>{if(Pt())return Ht(),new Promise((t,r)=>{Lt("init-ep",[t,r]);let n={type:"init-ep",in:{epName:e,env:_e}};Ye.postMessage(n)});await Er(_e,e)},jl=async e=>Pt()?(Ht(),new Promise((t,r)=>{Lt("copy-from",[t,r]);let n={type:"copy-from",in:{buffer:e}};Ye.postMessage(n,[e.buffer])})):jt(e),Kl=async(e,t)=>{if(Pt()){if(t?.preferredOutputLocation)throw new Error('session option "preferredOutputLocation" is not supported for proxy.');return Ht(),new Promise((r,n)=>{Lt("create",[r,n]);let o={type:"create",in:{model:e,options:{...t}}},i=[];e instanceof Uint8Array&&i.push(e.buffer),Ye.postMessage(o,i)})}else return Pr(e,t)},Yl=async e=>{if(Pt())return Ht(),new Promise((t,r)=>{Lt("release",[t,r]);let n={type:"release",in:e};Ye.postMessage(n)});zr(e)},Xl=async(e,t,r,n,o,i)=>{if(Pt()){if(r.some(a=>a[3]!=="cpu"))throw new Error("input tensor on GPU is not supported for proxy.");if(o.some(a=>a))throw new Error("pre-allocated output tensor is not supported for proxy.");return Ht(),new Promise((a,d)=>{Lt("run",[a,d]);let l=r,c={type:"run",in:{sessionId:e,inputIndices:t,inputs:l,outputIndices:n,options:i}};Ye.postMessage(c,Dr(l))})}else return Or(e,t,r,n,o,i)},Zl=async e=>{if(Pt())return Ht(),new Promise((t,r)=>{Lt("end-profiling",[t,r]);let n={type:"end-profiling",in:e};Ye.postMessage(n)});Br(e)}});var Ql,Fh,mn,Jl=V(()=>{"use strict";Ke();Po();Q();Cr();Yn();Ql=(e,t)=>{switch(e.location){case"cpu":return[e.type,e.dims,e.data,"cpu"];case"gpu-buffer":return[e.type,e.dims,{gpuBuffer:e.gpuBuffer},"gpu-buffer"];case"ml-tensor":return[e.type,e.dims,{mlTensor:e.mlTensor},"ml-tensor"];default:throw new Error(`invalid data location: ${e.location} for ${t()}`)}},Fh=e=>{switch(e[3]){case"cpu":return new De(e[0],e[2],e[1]);case"gpu-buffer":{let t=e[0];if(!Vr(t))throw new Error(`not supported data type: ${t} for deserializing GPU tensor`);let{gpuBuffer:r,download:n,dispose:o}=e[2];return De.fromGpuBuffer(r,{dataType:t,dims:e[1],download:n,dispose:o})}case"ml-tensor":{let t=e[0];if(!Nr(t))throw new Error(`not supported data type: ${t} for deserializing MLTensor tensor`);let{mlTensor:r,download:n,dispose:o}=e[2];return De.fromMLTensor(r,{dataType:t,dims:e[1],download:n,dispose:o})}default:throw new Error(`invalid data location: ${e[3]}`)}},mn=class{async fetchModelAndCopyToWasmMemory(t){return jl(await Zt(t))}async loadModel(t,r){Le();let n;typeof t=="string"? false?0:n=await this.fetchModelAndCopyToWasmMemory(t):n=t,[this.sessionId,this.inputNames,this.outputNames]=await Kl(n,r),Ve()}async dispose(){return Yl(this.sessionId)}async run(t,r,n){Le();let o=[],i=[];Object.entries(t).forEach(h=>{let w=h[0],g=h[1],y=this.inputNames.indexOf(w);if(y===-1)throw new Error(`invalid input '${w}'`);o.push(g),i.push(y)});let a=[],d=[];Object.entries(r).forEach(h=>{let w=h[0],g=h[1],y=this.outputNames.indexOf(w);if(y===-1)throw new Error(`invalid output '${w}'`);a.push(g),d.push(y)});let l=o.map((h,w)=>Ql(h,()=>`input "${this.inputNames[i[w]]}"`)),c=a.map((h,w)=>h?Ql(h,()=>`output "${this.outputNames[d[w]]}"`):null),m=await Xl(this.sessionId,i,l,d,c,n),u={};for(let h=0;h<m.length;h++)u[this.outputNames[d[h]]]=a[h]??Fh(m[h]);return Ve(),u}startProfiling(){}endProfiling(){Zl(this.sessionId)}}});var tc={};Gt(tc,{OnnxruntimeWebAssemblyBackend:()=>fn,initializeFlags:()=>ec,wasmBackend:()=>qh});var ec,fn,qh,rc=V(()=>{"use strict";Ke();Po();Jl();qt();ec=()=>{if((typeof _e.wasm.initTimeout!="number"||_e.wasm.initTimeout<0)&&(_e.wasm.initTimeout=0),_e.wasm.simd===!1&&console.warn('Deprecated property "env.wasm.simd" is set to false. non-SIMD build is no longer provided, and this setting will be ignored.'),typeof _e.wasm.proxy!="boolean"&&(_e.wasm.proxy=!1),typeof _e.wasm.trace!="boolean"&&(_e.wasm.trace=!1),typeof _e.wasm.numThreads!="number"||!Number.isInteger(_e.wasm.numThreads)||_e.wasm.numThreads<=0)if(typeof self<"u"&&!self.crossOriginIsolated)_e.wasm.numThreads=1;else{let e=typeof navigator>"u"?Vn("node:os").cpus().length:navigator.hardwareConcurrency;_e.wasm.numThreads=Math.min(4,Math.ceil((e||1)/2))}},fn=class{async init(t){ec(),await Fl(),await ql(t)}async createInferenceSessionHandler(t,r){let n=new mn;return await n.loadModel(t,r),Promise.resolve(n)}},qh=new fn});Ke();Ke();Ke();var Ia="1.20.0-dev.20241016-2b8fc5529b";var Hx=Hn;{let e=(rc(),br(tc)).wasmBackend;St("webgpu",e,5),St("webnn",e,5),St("cpu",e,10),St("wasm",e,10)}Object.defineProperty(_e.versions,"web",{value:Ia,enumerable:!0});
 /**
  * @license
  * Copyright 2021 Google LLC. All Rights Reserved.
@@ -6545,9 +6619,10 @@ let wasmInitPromise = null;
  * Create an ONNX inference session.
  * @param {Uint8Array} buffer The ONNX model buffer.
  * @param {import('onnxruntime-common').InferenceSession.SessionOptions} session_options ONNX inference session options.
- * @returns {Promise<import('onnxruntime-common').InferenceSession>} The ONNX inference session.
+ * @param {Object} session_config ONNX inference session configuration.
+ * @returns {Promise<import('onnxruntime-common').InferenceSession & { config: Object}>} The ONNX inference session.
  */
-async function createInferenceSession(buffer, session_options) {
+async function createInferenceSession(buffer, session_options, session_config) {
     if (wasmInitPromise) {
         // A previous session has already initialized the WASM runtime
         // so we wait for it to resolve before creating this new session.
@@ -6556,7 +6631,9 @@ async function createInferenceSession(buffer, session_options) {
 
     const sessionPromise = InferenceSession.create(buffer, session_options);
     wasmInitPromise ??= sessionPromise;
-    return await sessionPromise;
+    const session = await sessionPromise;
+    session.config = session_config;
+    return session;
 }
 
 /**
@@ -6991,7 +7068,7 @@ class AutoConfig {
 /**
  * Transformers.js-specific configuration, possibly present in config.json under the key `transformers.js_config`.
  * @typedef {Object} TransformersJSConfig
- * @property {import('./utils/tensor.js').DataType} [kv_cache_dtype] The data type of the key-value cache.
+ * @property {import('./utils/tensor.js').DataType|Record<import('./utils/dtypes.js').DataType, import('./utils/tensor.js').DataType>} [kv_cache_dtype] The data type of the key-value cache.
  * @property {Record<string, number>} [free_dimension_overrides] Override the free dimensions of the model.
  * See https://onnxruntime.ai/docs/tutorials/web/env-flags-and-session-options.html#freedimensionoverrides
  * for more information.
@@ -7045,7 +7122,7 @@ __webpack_require__.r(__webpack_exports__);
 
 
 
-const VERSION = '3.0.0-alpha.21';
+const VERSION = '3.0.0-alpha.22';
 
 // Check if various APIs are available (depends on environment)
 const IS_BROWSER_ENV = typeof self !== 'undefined';
@@ -9492,7 +9569,7 @@ const MODEL_CLASS_TO_NAME_MAPPING = new Map();
  * @param {string} pretrained_model_name_or_path The path to the directory containing the model file.
  * @param {string} fileName The name of the model file.
  * @param {import('./utils/hub.js').PretrainedModelOptions} options Additional options for loading the model.
- * @returns {Promise<{buffer: Uint8Array, session_options: Object}>} A Promise that resolves to the data needed to create an InferenceSession object.
+ * @returns {Promise<{buffer: Uint8Array, session_options: Object, session_config: Object}>} A Promise that resolves to the data needed to create an InferenceSession object.
  * @private
  */
 async function getSession(pretrained_model_name_or_path, fileName, options) {
@@ -9533,6 +9610,22 @@ async function getSession(pretrained_model_name_or_path, fileName, options) {
         throw new Error(`The device (${selectedDevice}) does not support fp16.`);
     }
 
+    // Only valid for models with a decoder
+    const kv_cache_dtype = custom_config.kv_cache_dtype
+        ? (typeof custom_config.kv_cache_dtype === 'string'
+            ? custom_config.kv_cache_dtype
+            : custom_config.kv_cache_dtype[selectedDtype] ?? 'float32')
+        : undefined;
+
+    if (kv_cache_dtype && !['float32', 'float16'].includes(kv_cache_dtype)) {
+        throw new Error(`Invalid kv_cache_dtype: ${kv_cache_dtype}. Should be one of: float32, float16`);
+    }
+
+    const session_config = {
+        dtype: selectedDtype,
+        kv_cache_dtype,
+    }
+
     // Construct the model file name
     const suffix = _utils_dtypes_js__WEBPACK_IMPORTED_MODULE_2__.DEFAULT_DTYPE_SUFFIX_MAPPING[selectedDtype];
     const modelFileName = `${options.subfolder ?? ''}/${fileName}${suffix}.onnx`;
@@ -9608,7 +9701,8 @@ async function getSession(pretrained_model_name_or_path, fileName, options) {
     }
 
     const buffer = await bufferPromise;
-    return { buffer, session_options };
+
+    return { buffer, session_options, session_config };
 }
 
 /**
@@ -9623,8 +9717,8 @@ async function getSession(pretrained_model_name_or_path, fileName, options) {
 async function constructSessions(pretrained_model_name_or_path, names, options) {
     return Object.fromEntries(await Promise.all(
         Object.keys(names).map(async (name) => {
-            const { buffer, session_options } = await getSession(pretrained_model_name_or_path, names[name], options);
-            const session = await (0,_backends_onnx_js__WEBPACK_IMPORTED_MODULE_1__.createInferenceSession)(buffer, session_options);
+            const { buffer, session_options, session_config } = await getSession(pretrained_model_name_or_path, names[name], options);
+            const session = await (0,_backends_onnx_js__WEBPACK_IMPORTED_MODULE_1__.createInferenceSession)(buffer, session_options, session_config);
             return [name, session];
         })
     ));
@@ -10981,9 +11075,8 @@ class PreTrainedModel extends _utils_generic_js__WEBPACK_IMPORTED_MODULE_3__.Cal
         if (pastKeyValues) {
             Object.assign(decoderFeeds, pastKeyValues)
         } else {
-
-            /** @type {import('./transformers.js').DataType} */
-            const dtype = this.custom_config.kv_cache_dtype ?? 'float32';
+            const session = this.sessions['decoder_model_merged'] ?? this.sessions['model'];
+            const dtype = session?.config?.kv_cache_dtype ?? 'float32';
             const empty = (dtype === 'float16') ? new Uint16Array() : [];
 
             const shapes = (0,_configs_js__WEBPACK_IMPORTED_MODULE_0__.getKeyValueShapes)(this.config);