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

package/dist/transformers.min.cjs

@@ -85,13 +85,13 @@
   \*************************************************************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{version:()=>s});const s="1.19.2"},"./src/backends/onnx.js":
 /*!******************************!*\
   !*** ./src/backends/onnx.js ***!
-  \******************************/(e,t,n)=>{"use strict";var s,r;n.r(t),n.d(t,{Tensor:()=>l.Tensor,createInferenceSession:()=>g,deviceToExecutionProviders:()=>m,isONNXProxy:()=>b,isONNXTensor:()=>w});var o=n(/*! ../env.js */"./src/env.js"),i=n(/*! onnxruntime-node */"onnxruntime-node"),a=n(/*! #onnxruntime-webgpu */"?cb4d"),l=n(/*! onnxruntime-common */"./node_modules/onnxruntime-common/dist/esm/index.js");const c=Object.freeze({auto:null,gpu:null,cpu:"cpu",wasm:"wasm",webgpu:"webgpu",cuda:"cuda",dml:"dml",webnn:{name:"webnn",deviceType:"cpu"},"webnn-npu":{name:"webnn",deviceType:"npu"},"webnn-gpu":{name:"webnn",deviceType:"gpu"},"webnn-cpu":{name:"webnn",deviceType:"cpu"}}),d=[];let u,p;const h=Symbol.for("onnxruntime");if(h in globalThis)p=globalThis[h];else if(o.apis.IS_NODE_ENV){switch(p=i??(s||(s=n.t(i,2))),process.platform){case"win32":d.push("dml");break;case"linux":"x64"===process.arch&&d.push("cuda")}d.push("cpu"),u=["cpu"]}else p=r||(r=n.t(a,2)),o.apis.IS_WEBNN_AVAILABLE&&d.push("webnn-npu","webnn-gpu","webnn-cpu","webnn"),o.apis.IS_WEBGPU_AVAILABLE&&d.push("webgpu"),d.push("wasm"),u=["wasm"];const _=p.InferenceSession;function m(e=null){if(!e)return u;switch(e){case"auto":return d;case"gpu":return d.filter((e=>["webgpu","cuda","dml","webnn-gpu"].includes(e)))}if(d.includes(e))return[c[e]??e];throw new Error(`Unsupported device: "${e}". Should be one of: ${d.join(", ")}.`)}let f=null;async function g(e,t){f&&await f;const n=_.create(e,t);return f??=n,await n}function w(e){return e instanceof p.Tensor}const M=p?.env;function b(){return M?.wasm?.proxy}M?.wasm&&(M.wasm.wasmPaths=`https://cdn.jsdelivr.net/npm/@huggingface/transformers@${o.env.version}/dist/`,M.wasm.proxy=!1,"undefined"!=typeof crossOriginIsolated&&crossOriginIsolated||(M.wasm.numThreads=1)),M?.webgpu&&(M.webgpu.powerPreference="high-performance"),o.env.backends.onnx=M},"./src/configs.js":
+  \******************************/(e,t,n)=>{"use strict";var s,r;n.r(t),n.d(t,{Tensor:()=>l.Tensor,createInferenceSession:()=>g,deviceToExecutionProviders:()=>m,isONNXProxy:()=>b,isONNXTensor:()=>w});var o=n(/*! ../env.js */"./src/env.js"),i=n(/*! onnxruntime-node */"onnxruntime-node"),a=n(/*! #onnxruntime-webgpu */"?cb4d"),l=n(/*! onnxruntime-common */"./node_modules/onnxruntime-common/dist/esm/index.js");const c=Object.freeze({auto:null,gpu:null,cpu:"cpu",wasm:"wasm",webgpu:"webgpu",cuda:"cuda",dml:"dml",webnn:{name:"webnn",deviceType:"cpu"},"webnn-npu":{name:"webnn",deviceType:"npu"},"webnn-gpu":{name:"webnn",deviceType:"gpu"},"webnn-cpu":{name:"webnn",deviceType:"cpu"}}),d=[];let u,p;const h=Symbol.for("onnxruntime");if(h in globalThis)p=globalThis[h];else if(o.apis.IS_NODE_ENV){switch(p=i??(s||(s=n.t(i,2))),process.platform){case"win32":d.push("dml");break;case"linux":"x64"===process.arch&&d.push("cuda")}d.push("cpu"),u=["cpu"]}else p=r||(r=n.t(a,2)),o.apis.IS_WEBNN_AVAILABLE&&d.push("webnn-npu","webnn-gpu","webnn-cpu","webnn"),o.apis.IS_WEBGPU_AVAILABLE&&d.push("webgpu"),d.push("wasm"),u=["wasm"];const _=p.InferenceSession;function m(e=null){if(!e)return u;switch(e){case"auto":return d;case"gpu":return d.filter((e=>["webgpu","cuda","dml","webnn-gpu"].includes(e)))}if(d.includes(e))return[c[e]??e];throw new Error(`Unsupported device: "${e}". Should be one of: ${d.join(", ")}.`)}let f=null;async function g(e,t,n){f&&await f;const s=_.create(e,t);f??=s;const r=await s;return r.config=n,r}function w(e){return e instanceof p.Tensor}const M=p?.env;function b(){return M?.wasm?.proxy}M?.wasm&&(M.wasm.wasmPaths=`https://cdn.jsdelivr.net/npm/@huggingface/transformers@${o.env.version}/dist/`,M.wasm.proxy=!1,"undefined"!=typeof crossOriginIsolated&&crossOriginIsolated||(M.wasm.numThreads=1)),M?.webgpu&&(M.webgpu.powerPreference="high-performance"),o.env.backends.onnx=M},"./src/configs.js":
 /*!************************!*\
   !*** ./src/configs.js ***!
   \************************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{AutoConfig:()=>l,PretrainedConfig:()=>a,getKeyValueShapes:()=>i});var s=n(/*! ./utils/core.js */"./src/utils/core.js"),r=n(/*! ./utils/hub.js */"./src/utils/hub.js");function o(e){const t={};let n={};switch(e.model_type){case"llava":case"paligemma":case"florence2":n=o(e.text_config);break;case"moondream1":n=o(e.phi_config);break;case"musicgen":n=o(e.decoder);break;case"gpt2":case"gptj":case"jais":case"codegen":case"gpt_bigcode":t.num_heads="n_head",t.num_layers="n_layer",t.hidden_size="n_embd";break;case"gpt_neox":case"stablelm":case"opt":case"phi":case"phi3":case"falcon":t.num_heads="num_attention_heads",t.num_layers="num_hidden_layers",t.hidden_size="hidden_size";break;case"llama":case"granite":case"cohere":case"mistral":case"starcoder2":case"qwen2":t.num_heads="num_key_value_heads",t.num_layers="num_hidden_layers",t.hidden_size="hidden_size",t.num_attention_heads="num_attention_heads";break;case"gemma":case"gemma2":t.num_heads="num_key_value_heads",t.num_layers="num_hidden_layers",t.dim_kv="head_dim";break;case"openelm":t.num_heads="num_kv_heads",t.num_layers="num_transformer_layers",t.dim_kv="head_dim";break;case"gpt_neo":case"donut-swin":t.num_heads="num_heads",t.num_layers="num_layers",t.hidden_size="hidden_size";break;case"bloom":t.num_heads="n_head",t.num_layers="n_layer",t.hidden_size="hidden_size";break;case"mpt":t.num_heads="n_heads",t.num_layers="n_layers",t.hidden_size="d_model";break;case"t5":case"mt5":case"longt5":t.num_decoder_layers="num_decoder_layers",t.num_decoder_heads="num_heads",t.decoder_dim_kv="d_kv",t.num_encoder_layers="num_layers",t.num_encoder_heads="num_heads",t.encoder_dim_kv="d_kv";break;case"bart":case"mbart":case"marian":case"whisper":case"m2m_100":case"blenderbot":case"blenderbot-small":case"florence2_language":t.num_decoder_layers="decoder_layers",t.num_decoder_heads="decoder_attention_heads",t.decoder_hidden_size="d_model",t.num_encoder_layers="encoder_layers",t.num_encoder_heads="encoder_attention_heads",t.encoder_hidden_size="d_model";break;case"speecht5":t.num_decoder_layers="decoder_layers",t.num_decoder_heads="decoder_attention_heads",t.decoder_hidden_size="hidden_size",t.num_encoder_layers="encoder_layers",t.num_encoder_heads="encoder_attention_heads",t.encoder_hidden_size="hidden_size";break;case"trocr":t.num_encoder_layers=t.num_decoder_layers="decoder_layers",t.num_encoder_heads=t.num_decoder_heads="decoder_attention_heads",t.encoder_hidden_size=t.decoder_hidden_size="d_model";break;case"musicgen_decoder":t.num_encoder_layers=t.num_decoder_layers="num_hidden_layers",t.num_encoder_heads=t.num_decoder_heads="num_attention_heads",t.encoder_hidden_size=t.decoder_hidden_size="hidden_size";break;case"vision-encoder-decoder":const r=o(e.decoder),i="num_decoder_layers"in r,a=(0,s.pick)(e,["model_type","is_encoder_decoder"]);return i?(a.num_decoder_layers=r.num_decoder_layers,a.num_decoder_heads=r.num_decoder_heads,a.decoder_hidden_size=r.decoder_hidden_size,a.num_encoder_layers=r.num_encoder_layers,a.num_encoder_heads=r.num_encoder_heads,a.encoder_hidden_size=r.encoder_hidden_size):(a.num_layers=r.num_layers,a.num_heads=r.num_heads,a.hidden_size=r.hidden_size),a}const r={...n,...(0,s.pick)(e,["model_type","multi_query","is_encoder_decoder"])};for(const n in t)r[n]=e[t[n]];return r}function i(e,{prefix:t="past_key_values"}={}){const n={},s=e.normalized_config;if(s.is_encoder_decoder&&"num_encoder_heads"in s&&"num_decoder_heads"in s){const e=s.encoder_dim_kv??s.encoder_hidden_size/s.num_encoder_heads,r=s.decoder_dim_kv??s.decoder_hidden_size/s.num_decoder_heads,o=[1,s.num_encoder_heads,0,e],i=[1,s.num_decoder_heads,0,r];for(let e=0;e<s.num_decoder_layers;++e)n[`${t}.${e}.encoder.key`]=o,n[`${t}.${e}.encoder.value`]=o,n[`${t}.${e}.decoder.key`]=i,n[`${t}.${e}.decoder.value`]=i}else{const e=s.num_heads,r=s.num_layers,o=s.dim_kv??s.hidden_size/(s.num_attention_heads??e);if("falcon"===s.model_type){const s=[1*e,0,o];for(let e=0;e<r;++e)n[`${t}.${e}.key`]=s,n[`${t}.${e}.value`]=s}else if(s.multi_query){const s=[1*e,0,2*o];for(let e=0;e<r;++e)n[`${t}.${e}.key_value`]=s}else if("bloom"===s.model_type){const s=[1*e,o,0],i=[1*e,0,o];for(let e=0;e<r;++e)n[`${t}.${e}.key`]=s,n[`${t}.${e}.value`]=i}else if("openelm"===s.model_type)for(let s=0;s<r;++s){const r=[1,e[s],0,o];n[`${t}.${s}.key`]=r,n[`${t}.${s}.value`]=r}else{const s=[1,e,0,o];for(let e=0;e<r;++e)n[`${t}.${e}.key`]=s,n[`${t}.${e}.value`]=s}}return n}class a{model_type=null;is_encoder_decoder=!1;max_position_embeddings;"transformers.js_config";constructor(e){Object.assign(this,e),this.normalized_config=o(this)}static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:s=null,local_files_only:o=!1,revision:i="main"}={}){!n||n instanceof a||(n=new a(n));const l=n??await async function(e,t){return await(0,r.getModelJSON)(e,"config.json",!0,t)}(e,{progress_callback:t,config:n,cache_dir:s,local_files_only:o,revision:i});return new this(l)}}class l{static async from_pretrained(...e){return a.from_pretrained(...e)}}},"./src/env.js":
 /*!********************!*\
   !*** ./src/env.js ***!
-  \********************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{apis:()=>m,env:()=>b});var s=n(/*! fs */"fs"),r=n(/*! path */"path"),o=n(/*! url */"url");const i="undefined"!=typeof self,a=i&&"DedicatedWorkerGlobalScope"===self.constructor.name,l=i&&"caches"in self,c="undefined"!=typeof navigator&&"gpu"in navigator,d="undefined"!=typeof navigator&&"ml"in navigator,u="undefined"!=typeof process,p=u&&"node"===process?.release?.name,h=!y(s),_=!y(r),m=Object.freeze({IS_BROWSER_ENV:i,IS_WEBWORKER_ENV:a,IS_WEB_CACHE_AVAILABLE:l,IS_WEBGPU_AVAILABLE:c,IS_WEBNN_AVAILABLE:d,IS_PROCESS_AVAILABLE:u,IS_NODE_ENV:p,IS_FS_AVAILABLE:h,IS_PATH_AVAILABLE:_}),f=h&&_,g=f?r.dirname(r.dirname(o.fileURLToPath("file:///workspaces/transformers.js/src/env.js"))):"./",w=f?r.join(g,"/.cache/"):null,M="/models/",b={version:"3.0.0-alpha.21",backends:{onnx:{}},allowRemoteModels:!0,remoteHost:"https://huggingface.co/",remotePathTemplate:"{model}/resolve/{revision}/",allowLocalModels:!i,localModelPath:f?r.join(g,M):M,useFS:h,useBrowserCache:l,useFSCache:h,cacheDir:w,useCustomCache:!1,customCache:null};function y(e){return 0===Object.keys(e).length}},"./src/generation/configuration_utils.js":
+  \********************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{apis:()=>m,env:()=>b});var s=n(/*! fs */"fs"),r=n(/*! path */"path"),o=n(/*! url */"url");const i="undefined"!=typeof self,a=i&&"DedicatedWorkerGlobalScope"===self.constructor.name,l=i&&"caches"in self,c="undefined"!=typeof navigator&&"gpu"in navigator,d="undefined"!=typeof navigator&&"ml"in navigator,u="undefined"!=typeof process,p=u&&"node"===process?.release?.name,h=!y(s),_=!y(r),m=Object.freeze({IS_BROWSER_ENV:i,IS_WEBWORKER_ENV:a,IS_WEB_CACHE_AVAILABLE:l,IS_WEBGPU_AVAILABLE:c,IS_WEBNN_AVAILABLE:d,IS_PROCESS_AVAILABLE:u,IS_NODE_ENV:p,IS_FS_AVAILABLE:h,IS_PATH_AVAILABLE:_}),f=h&&_,g=f?r.dirname(r.dirname(o.fileURLToPath("file:///workspaces/transformers.js/src/env.js"))):"./",w=f?r.join(g,"/.cache/"):null,M="/models/",b={version:"3.0.0-alpha.22",backends:{onnx:{}},allowRemoteModels:!0,remoteHost:"https://huggingface.co/",remotePathTemplate:"{model}/resolve/{revision}/",allowLocalModels:!i,localModelPath:f?r.join(g,M):M,useFS:h,useBrowserCache:l,useFSCache:h,cacheDir:w,useCustomCache:!1,customCache:null};function y(e){return 0===Object.keys(e).length}},"./src/generation/configuration_utils.js":
 /*!***********************************************!*\
   !*** ./src/generation/configuration_utils.js ***!
   \***********************************************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{GenerationConfig:()=>r});var s=n(/*! ../utils/core.js */"./src/utils/core.js");class r{max_length=20;max_new_tokens=null;min_length=0;min_new_tokens=null;early_stopping=!1;max_time=null;do_sample=!1;num_beams=1;num_beam_groups=1;penalty_alpha=null;use_cache=!0;temperature=1;top_k=50;top_p=1;typical_p=1;epsilon_cutoff=0;eta_cutoff=0;diversity_penalty=0;repetition_penalty=1;encoder_repetition_penalty=1;length_penalty=1;no_repeat_ngram_size=0;bad_words_ids=null;force_words_ids=null;renormalize_logits=!1;constraints=null;forced_bos_token_id=null;forced_eos_token_id=null;remove_invalid_values=!1;exponential_decay_length_penalty=null;suppress_tokens=null;begin_suppress_tokens=null;forced_decoder_ids=null;guidance_scale=null;num_return_sequences=1;output_attentions=!1;output_hidden_states=!1;output_scores=!1;return_dict_in_generate=!1;pad_token_id=null;bos_token_id=null;eos_token_id=null;encoder_no_repeat_ngram_size=0;decoder_start_token_id=null;generation_kwargs={};constructor(e){Object.assign(this,(0,s.pick)(e,Object.getOwnPropertyNames(this)))}}},"./src/generation/logits_process.js":
@@ -109,7 +109,7 @@
   \*************************************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{BaseStreamer:()=>i,TextStreamer:()=>l,WhisperTextStreamer:()=>c});var s=n(/*! ../utils/core.js */"./src/utils/core.js"),r=n(/*! ../tokenizers.js */"./src/tokenizers.js"),o=n(/*! ../env.js */"./src/env.js");class i{put(e){throw Error("Not implemented")}end(){throw Error("Not implemented")}}const a=o.apis.IS_PROCESS_AVAILABLE?e=>process.stdout.write(e):e=>console.log(e);class l extends i{constructor(e,{skip_prompt:t=!1,callback_function:n=null,token_callback_function:s=null,decode_kwargs:r={},...o}={}){super(),this.tokenizer=e,this.skip_prompt=t,this.callback_function=n??a,this.token_callback_function=s,this.decode_kwargs={...r,...o},this.token_cache=[],this.print_len=0,this.next_tokens_are_prompt=!0}put(e){if(e.length>1)throw Error("TextStreamer only supports batch size of 1");if(this.skip_prompt&&this.next_tokens_are_prompt)return void(this.next_tokens_are_prompt=!1);const t=e[0];this.token_callback_function?.(t),this.token_cache=(0,s.mergeArrays)(this.token_cache,t);const n=this.tokenizer.decode(this.token_cache,this.decode_kwargs);let o;n.endsWith("\n")?(o=n.slice(this.print_len),this.token_cache=[],this.print_len=0):n.length>0&&(0,r.is_chinese_char)(n.charCodeAt(n.length-1))?(o=n.slice(this.print_len),this.print_len+=o.length):(o=n.slice(this.print_len,n.lastIndexOf(" ")+1),this.print_len+=o.length),this.on_finalized_text(o,!1)}end(){let e;if(this.token_cache.length>0){e=this.tokenizer.decode(this.token_cache,this.decode_kwargs).slice(this.print_len),this.token_cache=[],this.print_len=0}else e="";this.next_tokens_are_prompt=!0,this.on_finalized_text(e,!0)}on_finalized_text(e,t){e.length>0&&this.callback_function?.(e),t&&this.callback_function===a&&o.apis.IS_PROCESS_AVAILABLE&&this.callback_function?.("\n")}}class c extends l{constructor(e,{skip_prompt:t=!1,callback_function:n=null,token_callback_function:s=null,on_chunk_start:r=null,on_chunk_end:o=null,on_finalize:i=null,time_precision:a=.02,skip_special_tokens:l=!0,decode_kwargs:c={}}={}){super(e,{skip_prompt:t,callback_function:n,token_callback_function:s,decode_kwargs:{skip_special_tokens:l,...c}}),this.timestamp_begin=e.timestamp_begin,this.on_chunk_start=r,this.on_chunk_end=o,this.on_finalize=i,this.time_precision=a,this.waiting_for_timestamp=!1}put(e){if(e.length>1)throw Error("WhisperTextStreamer only supports batch size of 1");const t=e[0];if(1===t.length){const n=Number(t[0])-this.timestamp_begin;if(n>=0){const t=n*this.time_precision;this.waiting_for_timestamp?this.on_chunk_end?.(t):this.on_chunk_start?.(t),this.waiting_for_timestamp=!this.waiting_for_timestamp,e=[[]]}}return super.put(e)}end(){super.end(),this.on_finalize?.()}}},"./src/models.js":
 /*!***********************!*\
   !*** ./src/models.js ***!
-  \***********************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{ASTForAudioClassification:()=>sn,ASTModel:()=>nn,ASTPreTrainedModel:()=>tn,AlbertForMaskedLM:()=>pt,AlbertForQuestionAnswering:()=>ut,AlbertForSequenceClassification:()=>dt,AlbertModel:()=>ct,AlbertPreTrainedModel:()=>lt,AutoModel:()=>Oa,AutoModelForAudioClassification:()=>nl,AutoModelForAudioFrameClassification:()=>rl,AutoModelForCTC:()=>tl,AutoModelForCausalLM:()=>qa,AutoModelForDepthEstimation:()=>ll,AutoModelForDocumentQuestionAnswering:()=>ol,AutoModelForImageClassification:()=>Qa,AutoModelForImageFeatureExtraction:()=>dl,AutoModelForImageMatting:()=>il,AutoModelForImageSegmentation:()=>Ha,AutoModelForImageToImage:()=>al,AutoModelForMaskGeneration:()=>el,AutoModelForMaskedLM:()=>Ua,AutoModelForNormalEstimation:()=>cl,AutoModelForObjectDetection:()=>Ka,AutoModelForQuestionAnswering:()=>Wa,AutoModelForSemanticSegmentation:()=>Ya,AutoModelForSeq2SeqLM:()=>Va,AutoModelForSequenceClassification:()=>ja,AutoModelForSpeechSeq2Seq:()=>Ra,AutoModelForTextToSpectrogram:()=>Ga,AutoModelForTextToWaveform:()=>$a,AutoModelForTokenClassification:()=>Da,AutoModelForUniversalSegmentation:()=>Ja,AutoModelForVision2Seq:()=>Xa,AutoModelForXVector:()=>sl,AutoModelForZeroShotObjectDetection:()=>Za,BartForConditionalGeneration:()=>Tt,BartForSequenceClassification:()=>vt,BartModel:()=>kt,BartPretrainedModel:()=>xt,BaseModelOutput:()=>q,BeitForImageClassification:()=>dr,BeitModel:()=>cr,BeitPreTrainedModel:()=>lr,BertForMaskedLM:()=>X,BertForQuestionAnswering:()=>Y,BertForSequenceClassification:()=>Q,BertForTokenClassification:()=>H,BertModel:()=>W,BertPreTrainedModel:()=>U,BlenderbotForConditionalGeneration:()=>zt,BlenderbotModel:()=>Lt,BlenderbotPreTrainedModel:()=>Et,BlenderbotSmallForConditionalGeneration:()=>Nt,BlenderbotSmallModel:()=>Bt,BlenderbotSmallPreTrainedModel:()=>It,BloomForCausalLM:()=>vs,BloomModel:()=>Ts,BloomPreTrainedModel:()=>ks,CLIPModel:()=>mn,CLIPPreTrainedModel:()=>_n,CLIPSegForImageSegmentation:()=>Pn,CLIPSegModel:()=>Cn,CLIPSegPreTrainedModel:()=>Fn,CLIPTextModel:()=>fn,CLIPTextModelWithProjection:()=>gn,CLIPVisionModel:()=>wn,CLIPVisionModelWithProjection:()=>Mn,CamembertForMaskedLM:()=>Me,CamembertForQuestionAnswering:()=>xe,CamembertForSequenceClassification:()=>be,CamembertForTokenClassification:()=>ye,CamembertModel:()=>we,CamembertPreTrainedModel:()=>ge,CausalLMOutput:()=>gl,CausalLMOutputWithPast:()=>wl,ChineseCLIPModel:()=>vn,ChineseCLIPPreTrainedModel:()=>Tn,ClapAudioModelWithProjection:()=>Fi,ClapModel:()=>Ti,ClapPreTrainedModel:()=>ki,ClapTextModelWithProjection:()=>vi,CodeGenForCausalLM:()=>Hn,CodeGenModel:()=>Qn,CodeGenPreTrainedModel:()=>Xn,CohereForCausalLM:()=>rs,CohereModel:()=>ss,CoherePreTrainedModel:()=>ns,ConvBertForMaskedLM:()=>ae,ConvBertForQuestionAnswering:()=>de,ConvBertForSequenceClassification:()=>le,ConvBertForTokenClassification:()=>ce,ConvBertModel:()=>ie,ConvBertPreTrainedModel:()=>oe,ConvNextForImageClassification:()=>ao,ConvNextModel:()=>io,ConvNextPreTrainedModel:()=>oo,ConvNextV2ForImageClassification:()=>uo,ConvNextV2Model:()=>co,ConvNextV2PreTrainedModel:()=>lo,DPTForDepthEstimation:()=>Gr,DPTModel:()=>Rr,DPTPreTrainedModel:()=>Vr,DebertaForMaskedLM:()=>ve,DebertaForQuestionAnswering:()=>Pe,DebertaForSequenceClassification:()=>Fe,DebertaForTokenClassification:()=>Ce,DebertaModel:()=>Te,DebertaPreTrainedModel:()=>ke,DebertaV2ForMaskedLM:()=>Ee,DebertaV2ForQuestionAnswering:()=>Ie,DebertaV2ForSequenceClassification:()=>Le,DebertaV2ForTokenClassification:()=>ze,DebertaV2Model:()=>Ae,DebertaV2PreTrainedModel:()=>Se,DecisionTransformerModel:()=>na,DecisionTransformerPreTrainedModel:()=>ta,DeiTForImageClassification:()=>Cr,DeiTModel:()=>Fr,DeiTPreTrainedModel:()=>vr,DepthAnythingForDepthEstimation:()=>qr,DepthAnythingPreTrainedModel:()=>$r,DepthProForDepthEstimation:()=>Yr,DepthProPreTrainedModel:()=>Hr,DetrForObjectDetection:()=>hr,DetrForSegmentation:()=>_r,DetrModel:()=>pr,DetrObjectDetectionOutput:()=>mr,DetrPreTrainedModel:()=>ur,DetrSegmentationOutput:()=>fr,Dinov2ForImageClassification:()=>_o,Dinov2Model:()=>ho,Dinov2PreTrainedModel:()=>po,DistilBertForMaskedLM:()=>Ve,DistilBertForQuestionAnswering:()=>De,DistilBertForSequenceClassification:()=>Oe,DistilBertForTokenClassification:()=>je,DistilBertModel:()=>Ne,DistilBertPreTrainedModel:()=>Be,DonutSwinModel:()=>ro,DonutSwinPreTrainedModel:()=>so,EfficientNetForImageClassification:()=>ji,EfficientNetModel:()=>Oi,EfficientNetPreTrainedModel:()=>Ni,ElectraForMaskedLM:()=>he,ElectraForQuestionAnswering:()=>fe,ElectraForSequenceClassification:()=>_e,ElectraForTokenClassification:()=>me,ElectraModel:()=>pe,ElectraPreTrainedModel:()=>ue,EsmForMaskedLM:()=>$e,EsmForSequenceClassification:()=>qe,EsmForTokenClassification:()=>Ue,EsmModel:()=>Ge,EsmPreTrainedModel:()=>Re,FalconForCausalLM:()=>xi,FalconModel:()=>yi,FalconPreTrainedModel:()=>bi,FastViTForImageClassification:()=>Xs,FastViTModel:()=>Ws,FastViTPreTrainedModel:()=>Us,Florence2ForConditionalGeneration:()=>hn,Florence2PreTrainedModel:()=>pn,GLPNForDepthEstimation:()=>no,GLPNModel:()=>to,GLPNPreTrainedModel:()=>eo,GPT2LMHeadModel:()=>En,GPT2Model:()=>An,GPT2PreTrainedModel:()=>Sn,GPTBigCodeForCausalLM:()=>Wn,GPTBigCodeModel:()=>Un,GPTBigCodePreTrainedModel:()=>qn,GPTJForCausalLM:()=>$n,GPTJModel:()=>Gn,GPTJPreTrainedModel:()=>Rn,GPTNeoForCausalLM:()=>On,GPTNeoModel:()=>Nn,GPTNeoPreTrainedModel:()=>Bn,GPTNeoXForCausalLM:()=>Vn,GPTNeoXModel:()=>Dn,GPTNeoXPreTrainedModel:()=>jn,Gemma2ForCausalLM:()=>ds,Gemma2Model:()=>cs,Gemma2PreTrainedModel:()=>ls,GemmaForCausalLM:()=>as,GemmaModel:()=>is,GemmaPreTrainedModel:()=>os,GraniteForCausalLM:()=>ts,GraniteModel:()=>es,GranitePreTrainedModel:()=>Zn,GroupViTModel:()=>qs,GroupViTPreTrainedModel:()=>$s,HieraForImageClassification:()=>Ar,HieraModel:()=>Sr,HieraPreTrainedModel:()=>Pr,HubertForCTC:()=>Zo,HubertForSequenceClassification:()=>ei,HubertModel:()=>Ko,HubertPreTrainedModel:()=>Jo,ImageMattingOutput:()=>Ml,JAISLMHeadModel:()=>In,JAISModel:()=>zn,JAISPreTrainedModel:()=>Ln,LlamaForCausalLM:()=>Kn,LlamaModel:()=>Jn,LlamaPreTrainedModel:()=>Yn,LlavaForConditionalGeneration:()=>dn,LlavaPreTrainedModel:()=>cn,LongT5ForConditionalGeneration:()=>wt,LongT5Model:()=>gt,LongT5PreTrainedModel:()=>ft,M2M100ForConditionalGeneration:()=>Co,M2M100Model:()=>Fo,M2M100PreTrainedModel:()=>vo,MBartForCausalLM:()=>At,MBartForConditionalGeneration:()=>Pt,MBartForSequenceClassification:()=>St,MBartModel:()=>Ct,MBartPreTrainedModel:()=>Ft,MPNetForMaskedLM:()=>Ze,MPNetForQuestionAnswering:()=>nt,MPNetForSequenceClassification:()=>et,MPNetForTokenClassification:()=>tt,MPNetModel:()=>Ke,MPNetPreTrainedModel:()=>Je,MT5ForConditionalGeneration:()=>yt,MT5Model:()=>bt,MT5PreTrainedModel:()=>Mt,MarianMTModel:()=>To,MarianModel:()=>ko,MarianPreTrainedModel:()=>xo,MaskFormerForInstanceSegmentation:()=>Zr,MaskFormerModel:()=>Kr,MaskFormerPreTrainedModel:()=>Jr,MaskedLMOutput:()=>ml,MistralForCausalLM:()=>fi,MistralModel:()=>mi,MistralPreTrainedModel:()=>_i,MobileBertForMaskedLM:()=>Qe,MobileBertForQuestionAnswering:()=>Ye,MobileBertForSequenceClassification:()=>He,MobileBertModel:()=>Xe,MobileBertPreTrainedModel:()=>We,MobileNetV1ForImageClassification:()=>Ui,MobileNetV1Model:()=>qi,MobileNetV1PreTrainedModel:()=>$i,MobileNetV2ForImageClassification:()=>Qi,MobileNetV2Model:()=>Xi,MobileNetV2PreTrainedModel:()=>Wi,MobileNetV3ForImageClassification:()=>Ji,MobileNetV3Model:()=>Yi,MobileNetV3PreTrainedModel:()=>Hi,MobileNetV4ForImageClassification:()=>ea,MobileNetV4Model:()=>Zi,MobileNetV4PreTrainedModel:()=>Ki,MobileViTForImageClassification:()=>Ks,MobileViTModel:()=>Js,MobileViTPreTrainedModel:()=>Ys,MobileViTV2ForImageClassification:()=>tr,MobileViTV2Model:()=>er,MobileViTV2PreTrainedModel:()=>Zs,ModelOutput:()=>$,Moondream1ForConditionalGeneration:()=>un,MptForCausalLM:()=>Ps,MptModel:()=>Cs,MptPreTrainedModel:()=>Fs,MusicgenForCausalLM:()=>Ri,MusicgenForConditionalGeneration:()=>Gi,MusicgenModel:()=>Vi,MusicgenPreTrainedModel:()=>Di,NomicBertModel:()=>K,NomicBertPreTrainedModel:()=>J,OPTForCausalLM:()=>Es,OPTModel:()=>As,OPTPreTrainedModel:()=>Ss,OpenELMForCausalLM:()=>hs,OpenELMModel:()=>ps,OpenELMPreTrainedModel:()=>us,OwlViTForObjectDetection:()=>rr,OwlViTModel:()=>sr,OwlViTPreTrainedModel:()=>nr,Owlv2ForObjectDetection:()=>ar,Owlv2Model:()=>ir,Owlv2PreTrainedModel:()=>or,Phi3ForCausalLM:()=>xs,Phi3Model:()=>ys,Phi3PreTrainedModel:()=>bs,PhiForCausalLM:()=>Ms,PhiModel:()=>ws,PhiPreTrainedModel:()=>gs,PreTrainedModel:()=>G,PretrainedMixin:()=>sa,PvtForImageClassification:()=>Os,PvtModel:()=>Ns,PvtPreTrainedModel:()=>Bs,PyAnnoteForAudioFrameClassification:()=>Bo,PyAnnoteModel:()=>Io,PyAnnotePreTrainedModel:()=>zo,QuestionAnsweringModelOutput:()=>fl,Qwen2ForCausalLM:()=>fs,Qwen2Model:()=>ms,Qwen2PreTrainedModel:()=>_s,RTDetrForObjectDetection:()=>Mr,RTDetrModel:()=>wr,RTDetrObjectDetectionOutput:()=>br,RTDetrPreTrainedModel:()=>gr,ResNetForImageClassification:()=>zr,ResNetModel:()=>Lr,ResNetPreTrainedModel:()=>Er,RoFormerForMaskedLM:()=>te,RoFormerForQuestionAnswering:()=>re,RoFormerForSequenceClassification:()=>ne,RoFormerForTokenClassification:()=>se,RoFormerModel:()=>ee,RoFormerPreTrainedModel:()=>Z,RobertaForMaskedLM:()=>Dt,RobertaForQuestionAnswering:()=>Gt,RobertaForSequenceClassification:()=>Vt,RobertaForTokenClassification:()=>Rt,RobertaModel:()=>jt,RobertaPreTrainedModel:()=>Ot,SamImageSegmentationOutput:()=>yo,SamModel:()=>bo,SamPreTrainedModel:()=>Mo,SapiensForDepthEstimation:()=>Xr,SapiensForNormalEstimation:()=>Qr,SapiensForSemanticSegmentation:()=>Wr,SapiensPreTrainedModel:()=>Ur,SegformerForImageClassification:()=>Ei,SegformerForSemanticSegmentation:()=>Li,SegformerModel:()=>Ai,SegformerPreTrainedModel:()=>Si,Seq2SeqLMOutput:()=>ul,SequenceClassifierOutput:()=>pl,SiglipModel:()=>yn,SiglipPreTrainedModel:()=>bn,SiglipTextModel:()=>xn,SiglipVisionModel:()=>kn,SpeechT5ForSpeechToText:()=>ci,SpeechT5ForTextToSpeech:()=>di,SpeechT5HifiGan:()=>ui,SpeechT5Model:()=>li,SpeechT5PreTrainedModel:()=>ai,SqueezeBertForMaskedLM:()=>ot,SqueezeBertForQuestionAnswering:()=>at,SqueezeBertForSequenceClassification:()=>it,SqueezeBertModel:()=>rt,SqueezeBertPreTrainedModel:()=>st,StableLmForCausalLM:()=>Bi,StableLmModel:()=>Ii,StableLmPreTrainedModel:()=>zi,Starcoder2ForCausalLM:()=>Mi,Starcoder2Model:()=>wi,Starcoder2PreTrainedModel:()=>gi,Swin2SRForImageSuperResolution:()=>Dr,Swin2SRModel:()=>jr,Swin2SRPreTrainedModel:()=>Or,SwinForImageClassification:()=>Nr,SwinModel:()=>Br,SwinPreTrainedModel:()=>Ir,T5ForConditionalGeneration:()=>mt,T5Model:()=>_t,T5PreTrainedModel:()=>ht,TableTransformerForObjectDetection:()=>kr,TableTransformerModel:()=>xr,TableTransformerObjectDetectionOutput:()=>Tr,TableTransformerPreTrainedModel:()=>yr,TokenClassifierOutput:()=>_l,TrOCRForCausalLM:()=>hi,TrOCRPreTrainedModel:()=>pi,UniSpeechForCTC:()=>Vo,UniSpeechForSequenceClassification:()=>Ro,UniSpeechModel:()=>Do,UniSpeechPreTrainedModel:()=>jo,UniSpeechSatForAudioFrameClassification:()=>Wo,UniSpeechSatForCTC:()=>qo,UniSpeechSatForSequenceClassification:()=>Uo,UniSpeechSatModel:()=>$o,UniSpeechSatPreTrainedModel:()=>Go,ViTForImageClassification:()=>Is,ViTMAEModel:()=>Ds,ViTMAEPreTrainedModel:()=>js,ViTMSNForImageClassification:()=>Gs,ViTMSNModel:()=>Rs,ViTMSNPreTrainedModel:()=>Vs,ViTModel:()=>zs,ViTPreTrainedModel:()=>Ls,VisionEncoderDecoderModel:()=>ln,VitMatteForImageMatting:()=>Hs,VitMattePreTrainedModel:()=>Qs,VitsModel:()=>Pi,VitsModelOutput:()=>bl,VitsPreTrainedModel:()=>Ci,Wav2Vec2BertForCTC:()=>Ho,Wav2Vec2BertForSequenceClassification:()=>Yo,Wav2Vec2BertModel:()=>Qo,Wav2Vec2BertPreTrainedModel:()=>Xo,Wav2Vec2ForAudioFrameClassification:()=>Lo,Wav2Vec2ForCTC:()=>Ao,Wav2Vec2ForSequenceClassification:()=>Eo,Wav2Vec2Model:()=>So,Wav2Vec2PreTrainedModel:()=>Po,WavLMForAudioFrameClassification:()=>ii,WavLMForCTC:()=>si,WavLMForSequenceClassification:()=>ri,WavLMForXVector:()=>oi,WavLMModel:()=>ni,WavLMPreTrainedModel:()=>ti,WeSpeakerResNetModel:()=>Oo,WeSpeakerResNetPreTrainedModel:()=>No,WhisperForConditionalGeneration:()=>an,WhisperModel:()=>on,WhisperPreTrainedModel:()=>rn,XLMForQuestionAnswering:()=>Qt,XLMForSequenceClassification:()=>Wt,XLMForTokenClassification:()=>Xt,XLMModel:()=>qt,XLMPreTrainedModel:()=>$t,XLMRobertaForMaskedLM:()=>Jt,XLMRobertaForQuestionAnswering:()=>en,XLMRobertaForSequenceClassification:()=>Kt,XLMRobertaForTokenClassification:()=>Zt,XLMRobertaModel:()=>Yt,XLMRobertaPreTrainedModel:()=>Ht,XLMWithLMHeadModel:()=>Ut,XVectorOutput:()=>hl,YolosForObjectDetection:()=>go,YolosModel:()=>fo,YolosObjectDetectionOutput:()=>wo,YolosPreTrainedModel:()=>mo});var s=n(/*! ./configs.js */"./src/configs.js"),r=n(/*! ./backends/onnx.js */"./src/backends/onnx.js"),o=n(/*! ./utils/dtypes.js */"./src/utils/dtypes.js"),i=n(/*! ./utils/generic.js */"./src/utils/generic.js"),a=n(/*! ./utils/core.js */"./src/utils/core.js"),l=n(/*! ./utils/hub.js */"./src/utils/hub.js"),c=n(/*! ./generation/logits_process.js */"./src/generation/logits_process.js"),d=n(/*! ./generation/configuration_utils.js */"./src/generation/configuration_utils.js"),u=n(/*! ./utils/tensor.js */"./src/utils/tensor.js"),p=n(/*! ./utils/maths.js */"./src/utils/maths.js"),h=n(/*! ./generation/stopping_criteria.js */"./src/generation/stopping_criteria.js"),_=n(/*! ./generation/logits_sampler.js */"./src/generation/logits_sampler.js"),m=n(/*! ./env.js */"./src/env.js"),f=n(/*! ./models/whisper/generation_whisper.js */"./src/models/whisper/generation_whisper.js"),g=n(/*! ./models/whisper/common_whisper.js */"./src/models/whisper/common_whisper.js");const w=0,M=1,b=2,y=3,x=4,k=5,T=6,v=7,F=new Map,C=new Map,P=new Map;async function S(e,t,n){return Object.fromEntries(await Promise.all(Object.keys(t).map((async i=>{const{buffer:a,session_options:c}=await async function(e,t,n){const i=n.config?.["transformers.js_config"]??{};let a=n.device??i.device;a&&"string"!=typeof a&&(a.hasOwnProperty(t)?a=a[t]:(console.warn(`device not specified for "${t}". Using the default device.`),a=null));const c=a??(m.apis.IS_NODE_ENV?"cpu":"wasm"),d=(0,r.deviceToExecutionProviders)(c);let u=n.dtype??i.dtype;"string"!=typeof u&&(u&&u.hasOwnProperty(t)?u=u[t]:(u=o.DEFAULT_DEVICE_DTYPE_MAPPING[c]??o.DATA_TYPES.fp32,console.warn(`dtype not specified for "${t}". Using the default dtype (${u}) for this device (${c}).`)));const p=u;if(!o.DEFAULT_DTYPE_SUFFIX_MAPPING.hasOwnProperty(p))throw new Error(`Invalid dtype: ${p}. Should be one of: ${Object.keys(o.DATA_TYPES).join(", ")}`);if(p===o.DATA_TYPES.fp16&&"webgpu"===c&&!await(0,o.isWebGpuFp16Supported)())throw new Error(`The device (${c}) does not support fp16.`);const h=o.DEFAULT_DTYPE_SUFFIX_MAPPING[p],_=`${n.subfolder??""}/${t}${h}.onnx`,f={...n.session_options};f.executionProviders??=d;const g=i.free_dimension_overrides;g?f.freeDimensionOverrides??=g:c.startsWith("webnn")&&!f.freeDimensionOverrides&&console.warn('WebNN does not currently support dynamic shapes and requires `free_dimension_overrides` to be set in config.json as a field within "transformers.js_config". When `free_dimension_overrides` is not set, you may experience significant performance degradation.');const w=(0,l.getModelFile)(e,_,!0,n),M=n.use_external_data_format??i.use_external_data_format;let b=[];if(M&&(!0===M||"object"==typeof M&&M.hasOwnProperty(t)&&!0===M[t])){if(m.apis.IS_NODE_ENV)throw new Error("External data format is not yet supported in Node.js");const s=`${t}${h}.onnx_data`,r=`${n.subfolder??""}/${s}`;b.push(new Promise((async(t,o)=>{const i=await(0,l.getModelFile)(e,r,!0,n);t({path:s,data:i})})))}else void 0!==f.externalData&&(b=f.externalData.map((async t=>{if("string"==typeof t.data){const s=await(0,l.getModelFile)(e,t.data,!0,n);return{...t,data:s}}return t})));if(b.length>0&&(f.externalData=await Promise.all(b)),"webgpu"===c){const e=(0,s.getKeyValueShapes)(n.config,{prefix:"present"});if(Object.keys(e).length>0&&!(0,r.isONNXProxy)()){const t={};for(const n in e)t[n]="gpu-buffer";f.preferredOutputLocation=t}}return{buffer:await w,session_options:f}}(e,t[i],n);return[i,await(0,r.createInferenceSession)(a,c)]}))))}async function A(e,t,n){return Object.fromEntries(await Promise.all(Object.keys(t).map((async s=>[s,await(0,l.getModelJSON)(e,t[s],!1,n)]))))}async function E(e,t){const n=function(e,t){const n=Object.create(null),s=[];for(const o of e.inputNames){const e=t[o];e instanceof u.Tensor?n[o]=(0,r.isONNXProxy)()?e.clone():e:s.push(o)}if(s.length>0)throw new Error(`An error occurred during model execution: "Missing the following inputs: ${s.join(", ")}.`);const o=Object.keys(t).length,i=e.inputNames.length;if(o>i){let n=Object.keys(t).filter((t=>!e.inputNames.includes(t)));console.warn(`WARNING: Too many inputs were provided (${o} > ${i}). The following inputs will be ignored: "${n.join(", ")}".`)}return n}(e,t);try{const t=Object.fromEntries(Object.entries(n).map((([e,t])=>[e,t.ort_tensor])));let s=await e.run(t);return s=L(s),s}catch(e){throw console.error(`An error occurred during model execution: "${e}".`),console.error("Inputs given to model:",n),e}}function L(e){for(let t in e)(0,r.isONNXTensor)(e[t])?e[t]=new u.Tensor(e[t]):"object"==typeof e[t]&&L(e[t]);return e}function z(e){if(e instanceof u.Tensor)return e;if(0===e.length)throw Error("items must be non-empty");if(Array.isArray(e[0])){if(e.some((t=>t.length!==e[0].length)))throw Error("Unable to create tensor, you should probably activate truncation and/or padding with 'padding=True' and/or 'truncation=True' to have batched tensors with the same length.");return new u.Tensor("int64",BigInt64Array.from(e.flat().map((e=>BigInt(e)))),[e.length,e[0].length])}return new u.Tensor("int64",BigInt64Array.from(e.map((e=>BigInt(e)))),[1,e.length])}function I(e){return new u.Tensor("bool",[e],[1])}async function B(e,t){let{encoder_outputs:n,input_ids:s,decoder_input_ids:r,...o}=t;if(!n){const s=(0,a.pick)(t,e.sessions.model.inputNames);n=(await N(e,s)).last_hidden_state}o.input_ids=r,o.encoder_hidden_states=n,e.sessions.decoder_model_merged.inputNames.includes("encoder_attention_mask")&&(o.encoder_attention_mask=t.attention_mask);return await O(e,o,!0)}async function N(e,t){const n=e.sessions.model,s=(0,a.pick)(t,n.inputNames);if(n.inputNames.includes("inputs_embeds")&&!s.inputs_embeds){if(!t.input_ids)throw new Error("Both `input_ids` and `inputs_embeds` are missing in the model inputs.");s.inputs_embeds=await e.encode_text({input_ids:t.input_ids})}return n.inputNames.includes("token_type_ids")&&!s.token_type_ids&&(s.token_type_ids=new u.Tensor("int64",new BigInt64Array(s.input_ids.data.length),s.input_ids.dims)),await E(n,s)}async function O(e,t,n=!1){const s=e.sessions[n?"decoder_model_merged":"model"],{past_key_values:r,...o}=t;s.inputNames.includes("use_cache_branch")&&(o.use_cache_branch=I(!!r)),s.inputNames.includes("position_ids")&&o.attention_mask&&!o.position_ids&&(o.position_ids=function(e,t=null){const{input_ids:n,inputs_embeds:s,attention_mask:r}=e,[o,i]=r.dims,a=new BigInt64Array(r.data.length);for(let e=0;e<o;++e){const t=e*i;let n=BigInt(0);for(let e=0;e<i;++e){const s=t+e;0n===r.data[s]?a[s]=BigInt(1):(a[s]=n,n+=r.data[s])}}let l=new u.Tensor("int64",a,r.dims);if(t){const e=-(n??s).dims.at(1);l=l.slice(null,[e,null])}return l}(o,r)),e.addPastKeyValues(o,r);const i=(0,a.pick)(o,s.inputNames);return await E(s,i)}async function j(e,{input_ids:t=null,attention_mask:n=null,pixel_values:s=null,position_ids:r=null,inputs_embeds:o=null,past_key_values:i=null,generation_config:a=null,logits_processor:l=null,...c}){if(!o)if(o=await e.encode_text({input_ids:t}),s&&1!==t.dims[1]){const r=await e.encode_image({pixel_values:s});({inputs_embeds:o,attention_mask:n}=e._merge_input_ids_with_image_features({image_features:r,inputs_embeds:o,input_ids:t,attention_mask:n}))}else if(i&&s&&1===t.dims[1]){const e=t.dims[1],s=Object.values(i)[0].dims.at(-2);n=(0,u.cat)([(0,u.ones)([t.dims[0],s]),n.slice(null,[n.dims[1]-e,n.dims[1]])],1)}return await O(e,{inputs_embeds:o,past_key_values:i,attention_mask:n,position_ids:r,generation_config:a,logits_processor:l},!0)}function D(e,t,n,s){if(n.past_key_values){const t=Object.values(n.past_key_values)[0].dims.at(-2),{input_ids:s,attention_mask:r}=n;if(r&&r.dims[1]>s.dims[1]);else if(t<s.dims[1])n.input_ids=s.slice(null,[t,null]);else if(null!=e.config.image_token_index&&s.data.some((t=>t==e.config.image_token_index))){const r=e.config.num_image_tokens;if(!r)throw new Error("`num_image_tokens` is missing in the model configuration.");const o=s.dims[1]-(t-r);n.input_ids=s.slice(null,[-o,null]),n.attention_mask=(0,u.ones)([1,t+o])}}return n}function V(e,t,n,s){return n.past_key_values&&(t=t.map((e=>[e.at(-1)]))),{...n,decoder_input_ids:z(t)}}function R(e,...t){return e.config.is_encoder_decoder?V(e,...t):D(e,...t)}class G extends i.Callable{main_input_name="input_ids";forward_params=["input_ids","attention_mask"];constructor(e,t,n){super(),this.config=e,this.sessions=t,this.configs=n;const s=P.get(this.constructor),r=F.get(s);switch(this.can_generate=!1,this._forward=null,this._prepare_inputs_for_generation=null,r){case x:this.can_generate=!0,this._forward=O,this._prepare_inputs_for_generation=D;break;case b:case y:case v:this.can_generate=!0,this._forward=B,this._prepare_inputs_for_generation=V;break;case M:this._forward=B;break;case T:this.can_generate=!0,this._forward=j,this._prepare_inputs_for_generation=R;break;default:this._forward=N}this.can_generate&&this.forward_params.push("past_key_values"),this.custom_config=this.config["transformers.js_config"]??{}}async dispose(){const e=[];for(const t of Object.values(this.sessions))t?.handler?.dispose&&e.push(t.handler.dispose());return await Promise.all(e)}static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:r=null,local_files_only:o=!1,revision:i="main",model_file_name:a=null,subfolder:l="onnx",device:c=null,dtype:d=null,use_external_data_format:u=null,session_options:p={}}={}){let h={progress_callback:t,config:n,cache_dir:r,local_files_only:o,revision:i,model_file_name:a,subfolder:l,device:c,dtype:d,use_external_data_format:u,session_options:p};const _=P.get(this),m=F.get(_);let f;if(n=h.config=await s.AutoConfig.from_pretrained(e,h),m===x)f=await Promise.all([S(e,{model:h.model_file_name??"model"},h),A(e,{generation_config:"generation_config.json"},h)]);else if(m===b||m===y)f=await Promise.all([S(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},h),A(e,{generation_config:"generation_config.json"},h)]);else if(m===k)f=await Promise.all([S(e,{model:"vision_encoder",prompt_encoder_mask_decoder:"prompt_encoder_mask_decoder"},h)]);else if(m===M)f=await Promise.all([S(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},h)]);else if(m===T){const t={embed_tokens:"embed_tokens",vision_encoder:"vision_encoder",decoder_model_merged:"decoder_model_merged"};n.is_encoder_decoder&&(t.model="encoder_model"),f=await Promise.all([S(e,t,h),A(e,{generation_config:"generation_config.json"},h)])}else m===v?f=await Promise.all([S(e,{model:"text_encoder",decoder_model_merged:"decoder_model_merged",encodec_decode:"encodec_decode"},h),A(e,{generation_config:"generation_config.json"},h)]):(m!==w&&console.warn(`Model type for '${_??n?.model_type}' not found, assuming encoder-only architecture. Please report this at https://github.com/xenova/transformers.js/issues/new/choose.`),f=await Promise.all([S(e,{model:h.model_file_name??"model"},h)]));return new this(n,...f)}async _call(e){return await this.forward(e)}async forward(e){return await this._forward(this,e)}get generation_config(){return this.configs?.generation_config??null}_get_logits_warper(e){const t=new c.LogitsProcessorList;return null!==e.temperature&&1!==e.temperature&&t.push(new c.TemperatureLogitsWarper(e.temperature)),null!==e.top_k&&0!==e.top_k&&t.push(new c.TopKLogitsWarper(e.top_k)),null!==e.top_p&&e.top_p<1&&t.push(new c.TopPLogitsWarper(e.top_p)),t}_get_logits_processor(e,t,n=null){const s=new c.LogitsProcessorList;if(null!==e.repetition_penalty&&1!==e.repetition_penalty&&s.push(new c.RepetitionPenaltyLogitsProcessor(e.repetition_penalty)),null!==e.no_repeat_ngram_size&&e.no_repeat_ngram_size>0&&s.push(new c.NoRepeatNGramLogitsProcessor(e.no_repeat_ngram_size)),null!==e.bad_words_ids&&s.push(new c.NoBadWordsLogitsProcessor(e.bad_words_ids,e.eos_token_id)),null!==e.min_length&&null!==e.eos_token_id&&e.min_length>0&&s.push(new c.MinLengthLogitsProcessor(e.min_length,e.eos_token_id)),null!==e.min_new_tokens&&null!==e.eos_token_id&&e.min_new_tokens>0&&s.push(new c.MinNewTokensLengthLogitsProcessor(t,e.min_new_tokens,e.eos_token_id)),null!==e.forced_bos_token_id&&s.push(new c.ForcedBOSTokenLogitsProcessor(e.forced_bos_token_id)),null!==e.forced_eos_token_id&&s.push(new c.ForcedEOSTokenLogitsProcessor(e.max_length,e.forced_eos_token_id)),null!==e.begin_suppress_tokens){const n=t>1||null===e.forced_bos_token_id?t:t+1;s.push(new c.SuppressTokensAtBeginLogitsProcessor(e.begin_suppress_tokens,n))}return null!==e.guidance_scale&&e.guidance_scale>1&&s.push(new c.ClassifierFreeGuidanceLogitsProcessor(e.guidance_scale)),null!==n&&s.extend(n),s}_prepare_generation_config(e,t,n=d.GenerationConfig){const s={...this.config};for(const e of["decoder","generator","text_config"])e in s&&Object.assign(s,s[e]);const r=new n(s);return Object.assign(r,this.generation_config??{}),e&&Object.assign(r,e),t&&Object.assign(r,(0,a.pick)(t,Object.getOwnPropertyNames(r))),r}_get_stopping_criteria(e,t=null){const n=new h.StoppingCriteriaList;return null!==e.max_length&&n.push(new h.MaxLengthCriteria(e.max_length,this.config.max_position_embeddings??null)),null!==e.eos_token_id&&n.push(new h.EosTokenCriteria(e.eos_token_id)),t&&n.extend(t),n}_validate_model_class(){if(!this.can_generate){const e=[ha,fa,pa,aa],t=P.get(this.constructor),n=new Set,s=this.config.model_type;for(const t of e){const e=t.get(s);e&&n.add(e[0])}let r=`The current model class (${t}) is not compatible with \`.generate()\`, as it doesn't have a language model head.`;throw n.size>0&&(r+=` Please use the following class instead: ${[...n].join(", ")}`),Error(r)}}prepare_inputs_for_generation(...e){return this._prepare_inputs_for_generation(this,...e)}_update_model_kwargs_for_generation({generated_input_ids:e,outputs:t,model_inputs:n,is_encoder_decoder:s}){return n.past_key_values=this.getPastKeyValues(t,n.past_key_values),n.input_ids=new u.Tensor("int64",e.flat(),[e.length,1]),s||(n.attention_mask=(0,u.cat)([n.attention_mask,(0,u.ones)([n.attention_mask.dims[0],1])],1)),n.position_ids=null,n}_prepare_model_inputs({inputs:e,bos_token_id:t,model_kwargs:n}){const s=(0,a.pick)(n,this.forward_params),r=this.main_input_name;if(r in s){if(e)throw new Error("`inputs`: {inputs}` were passed alongside {input_name} which is not allowed. Make sure to either pass {inputs} or {input_name}=...")}else s[r]=e;return{inputs_tensor:s[r],model_inputs:s,model_input_name:r}}async _prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:e,model_inputs:t,model_input_name:n,generation_config:s}){if(this.sessions.model.inputNames.includes("inputs_embeds")&&!t.inputs_embeds&&"_prepare_inputs_embeds"in this){const{input_ids:e,pixel_values:n,attention_mask:s,...r}=t,o=await this._prepare_inputs_embeds(t);t={...r,...(0,a.pick)(o,["inputs_embeds","attention_mask"])}}let{last_hidden_state:r}=await N(this,t);if(null!==s.guidance_scale&&s.guidance_scale>1)r=(0,u.cat)([r,(0,u.full_like)(r,0)],0),"attention_mask"in t&&(t.attention_mask=(0,u.cat)([t.attention_mask,(0,u.zeros_like)(t.attention_mask)],0));else if(t.decoder_input_ids){const e=z(t.decoder_input_ids).dims[0];if(e!==r.dims[0]){if(1!==r.dims[0])throw new Error(`The encoder outputs have a different batch size (${r.dims[0]}) than the decoder inputs (${e}).`);r=(0,u.cat)(Array.from({length:e},(()=>r)),0)}}return t.encoder_outputs=r,t}_prepare_decoder_input_ids_for_generation({batch_size:e,model_input_name:t,model_kwargs:n,decoder_start_token_id:s,bos_token_id:r,generation_config:o}){let{decoder_input_ids:i,...a}=n;if(i)Array.isArray(i[0])||(i=Array.from({length:e},(()=>i)));else if(s??=r,"musicgen"===this.config.model_type)i=Array.from({length:e*this.config.decoder.num_codebooks},(()=>[s]));else if(Array.isArray(s)){if(s.length!==e)throw new Error(`\`decoder_start_token_id\` expcted to have length ${e} but got ${s.length}`);i=s}else i=Array.from({length:e},(()=>[s]));return i=z(i),n.decoder_attention_mask=(0,u.ones_like)(i),{input_ids:i,model_inputs:a}}async generate({inputs:e=null,generation_config:t=null,logits_processor:n=null,stopping_criteria:s=null,streamer:r=null,...o}){this._validate_model_class(),t=this._prepare_generation_config(t,o);let{inputs_tensor:i,model_inputs:a,model_input_name:l}=this._prepare_model_inputs({inputs:e,model_kwargs:o});const c=this.config.is_encoder_decoder;let d;c&&("encoder_outputs"in a||(a=await this._prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:i,model_inputs:a,model_input_name:l,generation_config:t}))),c?({input_ids:d,model_inputs:a}=this._prepare_decoder_input_ids_for_generation({batch_size:a[l].dims.at(0),model_input_name:l,model_kwargs:a,decoder_start_token_id:t.decoder_start_token_id,bos_token_id:t.bos_token_id,generation_config:t})):d=a[l];let p=d.dims.at(-1);null!==t.max_new_tokens&&(t.max_length=p+t.max_new_tokens);const h=this._get_logits_processor(t,p,n),m=this._get_stopping_criteria(t,s),f=a[l].dims.at(0),g=_.LogitsSampler.getSampler(t),w=new Array(f).fill(0),M=d.tolist();let b;r&&r.put(M);let y={};for(;;){if(a=this.prepare_inputs_for_generation(M,a,t),b=await this.forward(a),t.output_attentions&&t.return_dict_in_generate){const e=this.getAttentions(b);for(const t in e)t in y||(y[t]=[]),y[t].push(e[t])}const e=h(M,b.logits.slice(null,-1,null)),n=[];for(let t=0;t<e.dims.at(0);++t){const s=e[t],r=await g(s);for(const[e,s]of r){const r=BigInt(e);w[t]+=s,M[t].push(r),n.push([r]);break}}r&&r.put(n);if(m(M).every((e=>e)))break;a=this._update_model_kwargs_for_generation({generated_input_ids:n,outputs:b,model_inputs:a,is_encoder_decoder:c})}r&&r.end();const x=this.getPastKeyValues(b,a.past_key_values,!0),k=new u.Tensor("int64",M.flat(),[M.length,M[0].length]);if(t.return_dict_in_generate)return{sequences:k,past_key_values:x,...y};for(const e of Object.values(b))"gpu-buffer"===e.location&&e.dispose();return k}getPastKeyValues(e,t,n=!1){const s=Object.create(null);for(const r in e)if(r.startsWith("present")){const o=r.replace("present","past_key_values"),i=r.includes("encoder");if(s[o]=i&&t?t[o]:e[r],t&&(!i||n)){const e=t[o];"gpu-buffer"===e.location&&e.dispose()}}return s}getAttentions(e){const t={};for(const n of["cross_attentions","encoder_attentions","decoder_attentions"])for(const s in e)s.startsWith(n)&&(n in t||(t[n]=[]),t[n].push(e[s]));return t}addPastKeyValues(e,t){if(t)Object.assign(e,t);else{const t=this.custom_config.kv_cache_dtype??"float32",n="float16"===t?new Uint16Array:[],r=(0,s.getKeyValueShapes)(this.config);for(const s in r)e[s]=new u.Tensor(t,n,r[s])}}async encode_image({pixel_values:e}){const t=(await E(this.sessions.vision_encoder,{pixel_values:e})).image_features;return this.config.num_image_tokens||(console.warn(`The number of image tokens was not set in the model configuration. Setting it to the number of features detected by the vision encoder (${t.dims[1]}).`),this.config.num_image_tokens=t.dims[1]),t}async encode_text({input_ids:e}){return(await E(this.sessions.embed_tokens,{input_ids:e})).inputs_embeds}}class ${}class q extends ${constructor({last_hidden_state:e,hidden_states:t=null,attentions:n=null}){super(),this.last_hidden_state=e,this.hidden_states=t,this.attentions=n}}class U extends G{}class W extends U{}class X extends U{async _call(e){return new ml(await super._call(e))}}class Q extends U{async _call(e){return new pl(await super._call(e))}}class H extends U{async _call(e){return new _l(await super._call(e))}}class Y extends U{async _call(e){return new fl(await super._call(e))}}class J extends G{}class K extends J{}class Z extends G{}class ee extends Z{}class te extends Z{async _call(e){return new ml(await super._call(e))}}class ne extends Z{async _call(e){return new pl(await super._call(e))}}class se extends Z{async _call(e){return new _l(await super._call(e))}}class re extends Z{async _call(e){return new fl(await super._call(e))}}class oe extends G{}class ie extends oe{}class ae extends oe{async _call(e){return new ml(await super._call(e))}}class le extends oe{async _call(e){return new pl(await super._call(e))}}class ce extends oe{async _call(e){return new _l(await super._call(e))}}class de extends oe{async _call(e){return new fl(await super._call(e))}}class ue extends G{}class pe extends ue{}class he extends ue{async _call(e){return new ml(await super._call(e))}}class _e extends ue{async _call(e){return new pl(await super._call(e))}}class me extends ue{async _call(e){return new _l(await super._call(e))}}class fe extends ue{async _call(e){return new fl(await super._call(e))}}class ge extends G{}class we extends ge{}class Me extends ge{async _call(e){return new ml(await super._call(e))}}class be extends ge{async _call(e){return new pl(await super._call(e))}}class ye extends ge{async _call(e){return new _l(await super._call(e))}}class xe extends ge{async _call(e){return new fl(await super._call(e))}}class ke extends G{}class Te extends ke{}class ve extends ke{async _call(e){return new ml(await super._call(e))}}class Fe extends ke{async _call(e){return new pl(await super._call(e))}}class Ce extends ke{async _call(e){return new _l(await super._call(e))}}class Pe extends ke{async _call(e){return new fl(await super._call(e))}}class Se extends G{}class Ae extends Se{}class Ee extends Se{async _call(e){return new ml(await super._call(e))}}class Le extends Se{async _call(e){return new pl(await super._call(e))}}class ze extends Se{async _call(e){return new _l(await super._call(e))}}class Ie extends Se{async _call(e){return new fl(await super._call(e))}}class Be extends G{}class Ne extends Be{}class Oe extends Be{async _call(e){return new pl(await super._call(e))}}class je extends Be{async _call(e){return new _l(await super._call(e))}}class De extends Be{async _call(e){return new fl(await super._call(e))}}class Ve extends Be{async _call(e){return new ml(await super._call(e))}}class Re extends G{}class Ge extends Re{}class $e extends Re{async _call(e){return new ml(await super._call(e))}}class qe extends Re{async _call(e){return new pl(await super._call(e))}}class Ue extends Re{async _call(e){return new _l(await super._call(e))}}class We extends G{}class Xe extends We{}class Qe extends We{async _call(e){return new ml(await super._call(e))}}class He extends We{async _call(e){return new pl(await super._call(e))}}class Ye extends We{async _call(e){return new fl(await super._call(e))}}class Je extends G{}class Ke extends Je{}class Ze extends Je{async _call(e){return new ml(await super._call(e))}}class et extends Je{async _call(e){return new pl(await super._call(e))}}class tt extends Je{async _call(e){return new _l(await super._call(e))}}class nt extends Je{async _call(e){return new fl(await super._call(e))}}class st extends G{}class rt extends st{}class ot extends st{async _call(e){return new ml(await super._call(e))}}class it extends st{async _call(e){return new pl(await super._call(e))}}class at extends st{async _call(e){return new fl(await super._call(e))}}class lt extends G{}class ct extends lt{}class dt extends lt{async _call(e){return new pl(await super._call(e))}}class ut extends lt{async _call(e){return new fl(await super._call(e))}}class pt extends lt{async _call(e){return new ml(await super._call(e))}}class ht extends G{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"]}class _t extends ht{}class mt extends ht{}class ft extends G{}class gt extends ft{}class wt extends ft{}class Mt extends G{}class bt extends Mt{}class yt extends Mt{}class xt extends G{}class kt extends xt{}class Tt extends xt{}class vt extends xt{async _call(e){return new pl(await super._call(e))}}class Ft extends G{}class Ct extends Ft{}class Pt extends Ft{}class St extends Ft{async _call(e){return new pl(await super._call(e))}}class At extends Ft{}class Et extends G{}class Lt extends Et{}class zt extends Et{}class It extends G{}class Bt extends It{}class Nt extends It{}class Ot extends G{}class jt extends Ot{}class Dt extends Ot{async _call(e){return new ml(await super._call(e))}}class Vt extends Ot{async _call(e){return new pl(await super._call(e))}}class Rt extends Ot{async _call(e){return new _l(await super._call(e))}}class Gt extends Ot{async _call(e){return new fl(await super._call(e))}}class $t extends G{}class qt extends $t{}class Ut extends $t{async _call(e){return new ml(await super._call(e))}}class Wt extends $t{async _call(e){return new pl(await super._call(e))}}class Xt extends $t{async _call(e){return new _l(await super._call(e))}}class Qt extends $t{async _call(e){return new fl(await super._call(e))}}class Ht extends G{}class Yt extends Ht{}class Jt extends Ht{async _call(e){return new ml(await super._call(e))}}class Kt extends Ht{async _call(e){return new pl(await super._call(e))}}class Zt extends Ht{async _call(e){return new _l(await super._call(e))}}class en extends Ht{async _call(e){return new fl(await super._call(e))}}class tn extends G{}class nn extends tn{}class sn extends tn{}class rn extends G{requires_attention_mask=!1;main_input_name="input_features";forward_params=["input_features","attention_mask","decoder_input_ids","decoder_attention_mask","past_key_values"]}class on extends rn{}class an extends rn{_prepare_generation_config(e,t){return super._prepare_generation_config(e,t,f.WhisperGenerationConfig)}_retrieve_init_tokens(e){const t=[e.decoder_start_token_id];let n=e.language;const s=e.task;if(e.is_multilingual){n||(console.warn("No language specified - defaulting to English (en)."),n="en");const r=`<|${(0,g.whisper_language_to_code)(n)}|>`;t.push(e.lang_to_id[r]),t.push(e.task_to_id[s??"transcribe"])}else if(n||s)throw new Error("Cannot specify `task` or `language` for an English-only model. If the model is intended to be multilingual, pass `is_multilingual=true` to generate, or update the generation config.");return!e.return_timestamps&&e.no_timestamps_token_id&&t.at(-1)!==e.no_timestamps_token_id?t.push(e.no_timestamps_token_id):e.return_timestamps&&t.at(-1)===e.no_timestamps_token_id&&(console.warn("<|notimestamps|> prompt token is removed from generation_config since `return_timestamps` is set to `true`."),t.pop()),t.filter((e=>null!=e))}async generate({inputs:e=null,generation_config:t=null,logits_processor:n=null,stopping_criteria:s=null,...r}){t=this._prepare_generation_config(t,r);const o=r.decoder_input_ids??this._retrieve_init_tokens(t);if(t.return_timestamps&&(n??=new c.LogitsProcessorList,n.push(new c.WhisperTimeStampLogitsProcessor(t,o))),t.begin_suppress_tokens&&(n??=new c.LogitsProcessorList,n.push(new c.SuppressTokensAtBeginLogitsProcessor(t.begin_suppress_tokens,o.length))),t.return_token_timestamps){if(!t.alignment_heads)throw new Error("Model generation config has no `alignment_heads`, token-level timestamps not available. See https://gist.github.com/hollance/42e32852f24243b748ae6bc1f985b13a on how to add this property to the generation config.");"translate"===t.task&&console.warn("Token-level timestamps may not be reliable for task 'translate'."),t.output_attentions=!0,t.return_dict_in_generate=!0}const i=await super.generate({inputs:e,generation_config:t,logits_processor:n,decoder_input_ids:o,...r});return t.return_token_timestamps&&(i.token_timestamps=this._extract_token_timestamps(i,t.alignment_heads,t.num_frames)),i}_extract_token_timestamps(e,t,n=null,s=.02){if(!e.cross_attentions)throw new Error("Model outputs must contain cross attentions to extract timestamps. This is most likely because the model was not exported with `output_attentions=True`.");null==n&&console.warn("`num_frames` has not been set, meaning the entire audio will be analyzed. This may lead to inaccurate token-level timestamps for short audios (< 30 seconds).");let r=this.config.median_filter_width;void 0===r&&(console.warn("Model config has no `median_filter_width`, using default value of 7."),r=7);const o=e.cross_attentions,i=Array.from({length:this.config.decoder_layers},((e,t)=>(0,u.cat)(o.map((e=>e[t])),2))),l=(0,u.stack)(t.map((([e,t])=>{if(e>=i.length)throw new Error(`Layer index ${e} is out of bounds for cross attentions (length ${i.length}).`);return n?i[e].slice(null,t,null,[0,n]):i[e].slice(null,t)}))).transpose(1,0,2,3),[c,d]=(0,u.std_mean)(l,-2,0,!0),h=l.clone();for(let e=0;e<h.dims[0];++e){const t=h[e];for(let n=0;n<t.dims[0];++n){const s=t[n],o=c[e][n][0].data,i=d[e][n][0].data;for(let e=0;e<s.dims[0];++e){let t=s[e].data;for(let e=0;e<t.length;++e)t[e]=(t[e]-i[e])/o[e];t.set((0,p.medianFilter)(t,r))}}}const _=[(0,u.mean)(h,1)],m=e.sequences.dims,f=new u.Tensor("float32",new Float32Array(m[0]*m[1]),m);for(let e=0;e<m[0];++e){const t=_[e].neg().squeeze_(0),[n,r]=(0,p.dynamic_time_warping)(t.tolist()),o=Array.from({length:n.length-1},((e,t)=>n[t+1]-n[t])),i=(0,a.mergeArrays)([1],o).map((e=>!!e)),l=[];for(let e=0;e<i.length;++e)i[e]&&l.push(r[e]*s);f[e].data.set(l,1)}return f}}class ln extends G{main_input_name="pixel_values";forward_params=["pixel_values","input_ids","encoder_hidden_states","past_key_values"]}class cn extends G{forward_params=["input_ids","pixel_values","attention_mask","position_ids","past_key_values"]}class dn extends cn{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:n,attention_mask:s}){const r=this.config.image_token_index,o=n.tolist().map((e=>e.findIndex((e=>e==r)))),i=o.every((e=>-1===e)),a=o.every((e=>-1!==e));if(!i&&!a)throw new Error("Every input should contain either 0 or 1 image token.");if(i)return{inputs_embeds:e,attention_mask:s};const l=[],c=[];for(let n=0;n<o.length;++n){const r=o[n],i=e[n],a=t[n],d=s[n];l.push((0,u.cat)([i.slice([0,r]),a,i.slice([r+1,i.dims[0]])],0)),c.push((0,u.cat)([d.slice([0,r]),(0,u.ones)([a.dims[0]]),d.slice([r+1,d.dims[0]])],0))}return{inputs_embeds:(0,u.stack)(l,0),attention_mask:(0,u.stack)(c,0)}}}class un extends dn{}class pn extends G{forward_params=["input_ids","inputs_embeds","attention_mask","pixel_values","encoder_outputs","decoder_input_ids","decoder_inputs_embeds","decoder_attention_mask","past_key_values"];main_input_name="inputs_embeds"}class hn extends pn{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:n,attention_mask:s}){return{inputs_embeds:(0,u.cat)([t,e],1),attention_mask:(0,u.cat)([(0,u.ones)(t.dims.slice(0,2)),s],1)}}async _prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:n,attention_mask:s}){if(!e&&!t)throw new Error("Either `input_ids` or `pixel_values` should be provided.");let r,o;return e&&(r=await this.encode_text({input_ids:e})),t&&(o=await this.encode_image({pixel_values:t})),r&&o?({inputs_embeds:n,attention_mask:s}=this._merge_input_ids_with_image_features({inputs_embeds:r,image_features:o,input_ids:e,attention_mask:s})):n=r||o,{inputs_embeds:n,attention_mask:s}}async forward({input_ids:e,pixel_values:t,attention_mask:n,decoder_input_ids:s,decoder_attention_mask:r,encoder_outputs:o,past_key_values:i,inputs_embeds:a,decoder_inputs_embeds:l}){if(a||({inputs_embeds:a,attention_mask:n}=await this._prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:a,attention_mask:n})),!o){let{last_hidden_state:e}=await N(this,{inputs_embeds:a,attention_mask:n});o=e}if(!l){if(!s)throw new Error("Either `decoder_input_ids` or `decoder_inputs_embeds` should be provided.");l=await this.encode_text({input_ids:s})}const c={inputs_embeds:l,attention_mask:r,encoder_attention_mask:n,encoder_hidden_states:o,past_key_values:i};return await O(this,c,!0)}}class _n extends G{}class mn extends _n{}class fn extends _n{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class gn extends _n{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class wn extends _n{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class Mn extends _n{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class bn extends G{}class yn extends bn{}class xn extends bn{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class kn extends _n{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class Tn extends G{}class vn extends Tn{}class Fn extends G{}class Cn extends Fn{}class Pn extends Fn{}class Sn extends G{}class An extends Sn{}class En extends Sn{}class Ln extends G{}class zn extends Ln{}class In extends Ln{}class Bn extends G{}class Nn extends Bn{}class On extends Bn{}class jn extends G{}class Dn extends jn{}class Vn extends jn{}class Rn extends G{}class Gn extends Rn{}class $n extends Rn{}class qn extends G{}class Un extends qn{}class Wn extends qn{}class Xn extends G{}class Qn extends Xn{}class Hn extends Xn{}class Yn extends G{}class Jn extends Yn{}class Kn extends Yn{}class Zn extends G{}class es extends Zn{}class ts extends Zn{}class ns extends G{}class ss extends ns{}class rs extends ns{}class os extends G{}class is extends os{}class as extends os{}class ls extends G{}class cs extends ls{}class ds extends ls{}class us extends G{}class ps extends us{}class hs extends us{}class _s extends G{}class ms extends _s{}class fs extends _s{}class gs extends G{}class ws extends gs{}class Ms extends gs{}class bs extends G{}class ys extends bs{}class xs extends bs{}class ks extends G{}class Ts extends ks{}class vs extends ks{}class Fs extends G{}class Cs extends Fs{}class Ps extends Fs{}class Ss extends G{}class As extends Ss{}class Es extends Ss{}class Ls extends G{}class zs extends Ls{}class Is extends Ls{async _call(e){return new pl(await super._call(e))}}class Bs extends G{}class Ns extends Bs{}class Os extends Bs{async _call(e){return new pl(await super._call(e))}}class js extends G{}class Ds extends js{}class Vs extends G{}class Rs extends Vs{}class Gs extends Vs{async _call(e){return new pl(await super._call(e))}}class $s extends G{}class qs extends $s{}class Us extends G{}class Ws extends Us{}class Xs extends Us{async _call(e){return new pl(await super._call(e))}}class Qs extends G{}class Hs extends Qs{async _call(e){return new Ml(await super._call(e))}}class Ys extends G{}class Js extends Ys{}class Ks extends Ys{async _call(e){return new pl(await super._call(e))}}class Zs extends G{}class er extends Zs{}class tr extends Zs{async _call(e){return new pl(await super._call(e))}}class nr extends G{}class sr extends nr{}class rr extends nr{}class or extends G{}class ir extends or{}class ar extends or{}class lr extends G{}class cr extends lr{}class dr extends lr{async _call(e){return new pl(await super._call(e))}}class ur extends G{}class pr extends ur{}class hr extends ur{async _call(e){return new mr(await super._call(e))}}class _r extends ur{async _call(e){return new fr(await super._call(e))}}class mr extends ${constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class fr extends ${constructor({logits:e,pred_boxes:t,pred_masks:n}){super(),this.logits=e,this.pred_boxes=t,this.pred_masks=n}}class gr extends G{}class wr extends gr{}class Mr extends gr{async _call(e){return new br(await super._call(e))}}class br extends ${constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class yr extends G{}class xr extends yr{}class kr extends yr{async _call(e){return new Tr(await super._call(e))}}class Tr extends mr{}class vr extends G{}class Fr extends vr{}class Cr extends vr{async _call(e){return new pl(await super._call(e))}}class Pr extends G{}class Sr extends Pr{}class Ar extends Pr{async _call(e){return new pl(await super._call(e))}}class Er extends G{}class Lr extends Er{}class zr extends Er{async _call(e){return new pl(await super._call(e))}}class Ir extends G{}class Br extends Ir{}class Nr extends Ir{async _call(e){return new pl(await super._call(e))}}class Or extends G{}class jr extends Or{}class Dr extends Or{}class Vr extends G{}class Rr extends Vr{}class Gr extends Vr{}class $r extends G{}class qr extends $r{}class Ur extends G{}class Wr extends Ur{}class Xr extends Ur{}class Qr extends Ur{}class Hr extends G{}class Yr extends Hr{}class Jr extends G{}class Kr extends Jr{}class Zr extends Jr{}class eo extends G{}class to extends eo{}class no extends eo{}class so extends G{}class ro extends so{}class oo extends G{}class io extends oo{}class ao extends oo{async _call(e){return new pl(await super._call(e))}}class lo extends G{}class co extends lo{}class uo extends lo{async _call(e){return new pl(await super._call(e))}}class po extends G{}class ho extends po{}class _o extends po{async _call(e){return new pl(await super._call(e))}}class mo extends G{}class fo extends mo{}class go extends mo{async _call(e){return new wo(await super._call(e))}}class wo extends ${constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class Mo extends G{}class bo extends Mo{async get_image_embeddings({pixel_values:e}){return await N(this,{pixel_values:e})}async forward(e){if(e.image_embeddings&&e.image_positional_embeddings||(e={...e,...await this.get_image_embeddings(e)}),!e.input_labels&&e.input_points){const t=e.input_points.dims.slice(0,-1),n=t.reduce(((e,t)=>e*t),1);e.input_labels=new u.Tensor("int64",new BigInt64Array(n).fill(1n),t)}const t={image_embeddings:e.image_embeddings,image_positional_embeddings:e.image_positional_embeddings};return e.input_points&&(t.input_points=e.input_points),e.input_labels&&(t.input_labels=e.input_labels),e.input_boxes&&(t.input_boxes=e.input_boxes),await E(this.sessions.prompt_encoder_mask_decoder,t)}async _call(e){return new yo(await super._call(e))}}class yo extends ${constructor({iou_scores:e,pred_masks:t}){super(),this.iou_scores=e,this.pred_masks=t}}class xo extends G{}class ko extends xo{}class To extends xo{}class vo extends G{}class Fo extends vo{}class Co extends vo{}class Po extends G{}class So extends Po{}class Ao extends Po{async _call(e){return new gl(await super._call(e))}}class Eo extends Po{async _call(e){return new pl(await super._call(e))}}class Lo extends Po{async _call(e){return new _l(await super._call(e))}}class zo extends G{}class Io extends zo{}class Bo extends zo{async _call(e){return new _l(await super._call(e))}}class No extends G{}class Oo extends No{}class jo extends G{}class Do extends jo{}class Vo extends jo{async _call(e){return new gl(await super._call(e))}}class Ro extends jo{async _call(e){return new pl(await super._call(e))}}class Go extends G{}class $o extends Go{}class qo extends Go{async _call(e){return new gl(await super._call(e))}}class Uo extends Go{async _call(e){return new pl(await super._call(e))}}class Wo extends Go{async _call(e){return new _l(await super._call(e))}}class Xo extends G{}class Qo extends Xo{}class Ho extends Xo{async _call(e){return new gl(await super._call(e))}}class Yo extends Xo{async _call(e){return new pl(await super._call(e))}}class Jo extends G{}class Ko extends Po{}class Zo extends Po{async _call(e){return new gl(await super._call(e))}}class ei extends Po{async _call(e){return new pl(await super._call(e))}}class ti extends G{}class ni extends ti{}class si extends ti{async _call(e){return new gl(await super._call(e))}}class ri extends ti{async _call(e){return new pl(await super._call(e))}}class oi extends ti{async _call(e){return new hl(await super._call(e))}}class ii extends ti{async _call(e){return new _l(await super._call(e))}}class ai extends G{}class li extends ai{}class ci extends ai{}class di extends ai{async generate_speech(e,t,{threshold:n=.5,minlenratio:s=0,maxlenratio:r=20,vocoder:o=null}={}){const i={input_ids:e},{encoder_outputs:a,encoder_attention_mask:l}=await N(this,i),c=a.dims[1]/this.config.reduction_factor,d=Math.floor(c*r),p=Math.floor(c*s),h=this.config.num_mel_bins;let _=[],m=null,f=null,g=0;for(;;){++g;const e=I(!!f);let s;s=f?f.output_sequence_out:new u.Tensor("float32",new Float32Array(h),[1,1,h]);let r={use_cache_branch:e,output_sequence:s,encoder_attention_mask:l,speaker_embeddings:t,encoder_hidden_states:a};this.addPastKeyValues(r,m),f=await E(this.sessions.decoder_model_merged,r),m=this.getPastKeyValues(f,m);const{prob:o,spectrum:i}=f;if(_.push(i),g>=p&&(Array.from(o.data).filter((e=>e>=n)).length>0||g>=d))break}const w=(0,u.cat)(_),{waveform:M}=await E(o.sessions.model,{spectrogram:w});return{spectrogram:w,waveform:M}}}class ui extends G{main_input_name="spectrogram"}class pi extends G{}class hi extends pi{}class _i extends G{}class mi extends _i{}class fi extends _i{}class gi extends G{}class wi extends gi{}class Mi extends gi{}class bi extends G{}class yi extends bi{}class xi extends bi{}class ki extends G{}class Ti extends ki{}class vi extends ki{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class Fi extends ki{static async from_pretrained(e,t={}){return t.model_file_name??="audio_model",super.from_pretrained(e,t)}}class Ci extends G{}class Pi extends Ci{async _call(e){return new bl(await super._call(e))}}class Si extends G{}class Ai extends Si{}class Ei extends Si{}class Li extends Si{}class zi extends G{}class Ii extends zi{}class Bi extends zi{}class Ni extends G{}class Oi extends Ni{}class ji extends Ni{async _call(e){return new pl(await super._call(e))}}class Di extends G{}class Vi extends Di{}class Ri extends Di{}class Gi extends G{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"];_apply_and_filter_by_delay_pattern_mask(e){const[t,n]=e.dims,s=this.config.decoder.num_codebooks,r=n-s;let o=0;for(let t=0;t<e.size;++t){if(e.data[t]===this.config.decoder.pad_token_id)continue;const i=t%n-Math.floor(t/n)%s;i>0&&i<=r&&(e.data[o++]=e.data[t])}const i=Math.floor(t/s),a=o/(i*s);return new u.Tensor(e.type,e.data.slice(0,o),[i,s,a])}prepare_inputs_for_generation(e,t,n){let s=structuredClone(e);for(let e=0;e<s.length;++e)for(let t=0;t<s[e].length;++t)e%this.config.decoder.num_codebooks>=t&&(s[e][t]=BigInt(this.config.decoder.pad_token_id));null!==n.guidance_scale&&n.guidance_scale>1&&(s=s.concat(s));return super.prepare_inputs_for_generation(s,t,n)}async generate(e){const t=await super.generate(e),n=this._apply_and_filter_by_delay_pattern_mask(t).unsqueeze_(0),{audio_values:s}=await E(this.sessions.encodec_decode,{audio_codes:n});return s}}class $i extends G{}class qi extends $i{}class Ui extends $i{async _call(e){return new pl(await super._call(e))}}class Wi extends G{}class Xi extends Wi{}class Qi extends Wi{async _call(e){return new pl(await super._call(e))}}class Hi extends G{}class Yi extends Hi{}class Ji extends Hi{async _call(e){return new pl(await super._call(e))}}class Ki extends G{}class Zi extends Ki{}class ea extends Ki{async _call(e){return new pl(await super._call(e))}}class ta extends G{}class na extends ta{}class sa{static MODEL_CLASS_MAPPINGS=null;static BASE_IF_FAIL=!1;static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:r=null,local_files_only:o=!1,revision:i="main",model_file_name:a=null,subfolder:l="onnx",device:c=null,dtype:d=null,use_external_data_format:u=null,session_options:p={}}={}){const h={progress_callback:t,config:n,cache_dir:r,local_files_only:o,revision:i,model_file_name:a,subfolder:l,device:c,dtype:d,use_external_data_format:u,session_options:p};if(h.config=await s.AutoConfig.from_pretrained(e,h),!this.MODEL_CLASS_MAPPINGS)throw new Error("`MODEL_CLASS_MAPPINGS` not implemented for this type of `AutoClass`: "+this.name);for(const t of this.MODEL_CLASS_MAPPINGS){const n=t.get(h.config.model_type);if(n)return await n[1].from_pretrained(e,h)}if(this.BASE_IF_FAIL)return console.warn(`Unknown model class "${h.config.model_type}", attempting to construct from base class.`),await G.from_pretrained(e,h);throw Error(`Unsupported model type: ${h.config.model_type}`)}}const ra=new Map([["bert",["BertModel",W]],["nomic_bert",["NomicBertModel",K]],["roformer",["RoFormerModel",ee]],["electra",["ElectraModel",pe]],["esm",["EsmModel",Ge]],["convbert",["ConvBertModel",ie]],["camembert",["CamembertModel",we]],["deberta",["DebertaModel",Te]],["deberta-v2",["DebertaV2Model",Ae]],["mpnet",["MPNetModel",Ke]],["albert",["AlbertModel",ct]],["distilbert",["DistilBertModel",Ne]],["roberta",["RobertaModel",jt]],["xlm",["XLMModel",qt]],["xlm-roberta",["XLMRobertaModel",Yt]],["clap",["ClapModel",Ti]],["clip",["CLIPModel",mn]],["clipseg",["CLIPSegModel",Cn]],["chinese_clip",["ChineseCLIPModel",vn]],["siglip",["SiglipModel",yn]],["mobilebert",["MobileBertModel",Xe]],["squeezebert",["SqueezeBertModel",rt]],["wav2vec2",["Wav2Vec2Model",So]],["wav2vec2-bert",["Wav2Vec2BertModel",Qo]],["unispeech",["UniSpeechModel",Do]],["unispeech-sat",["UniSpeechSatModel",$o]],["hubert",["HubertModel",Ko]],["wavlm",["WavLMModel",ni]],["audio-spectrogram-transformer",["ASTModel",nn]],["vits",["VitsModel",Pi]],["pyannote",["PyAnnoteModel",Io]],["wespeaker-resnet",["WeSpeakerResNetModel",Oo]],["detr",["DetrModel",pr]],["rt_detr",["RTDetrModel",wr]],["table-transformer",["TableTransformerModel",xr]],["vit",["ViTModel",zs]],["pvt",["PvtModel",Ns]],["vit_msn",["ViTMSNModel",Rs]],["vit_mae",["ViTMAEModel",Ds]],["groupvit",["GroupViTModel",qs]],["fastvit",["FastViTModel",Ws]],["mobilevit",["MobileViTModel",Js]],["mobilevitv2",["MobileViTV2Model",er]],["owlvit",["OwlViTModel",sr]],["owlv2",["Owlv2Model",ir]],["beit",["BeitModel",cr]],["deit",["DeiTModel",Fr]],["hiera",["HieraModel",Sr]],["convnext",["ConvNextModel",io]],["convnextv2",["ConvNextV2Model",co]],["dinov2",["Dinov2Model",ho]],["resnet",["ResNetModel",Lr]],["swin",["SwinModel",Br]],["swin2sr",["Swin2SRModel",jr]],["donut-swin",["DonutSwinModel",ro]],["yolos",["YolosModel",fo]],["dpt",["DPTModel",Rr]],["glpn",["GLPNModel",to]],["hifigan",["SpeechT5HifiGan",ui]],["efficientnet",["EfficientNetModel",Oi]],["decision_transformer",["DecisionTransformerModel",na]],["mobilenet_v1",["MobileNetV1Model",qi]],["mobilenet_v2",["MobileNetV2Model",Xi]],["mobilenet_v3",["MobileNetV3Model",Yi]],["mobilenet_v4",["MobileNetV4Model",Zi]],["maskformer",["MaskFormerModel",Kr]]]),oa=new Map([["t5",["T5Model",_t]],["longt5",["LongT5Model",gt]],["mt5",["MT5Model",bt]],["bart",["BartModel",kt]],["mbart",["MBartModel",Ct]],["marian",["MarianModel",ko]],["whisper",["WhisperModel",on]],["m2m_100",["M2M100Model",Fo]],["blenderbot",["BlenderbotModel",Lt]],["blenderbot-small",["BlenderbotSmallModel",Bt]]]),ia=new Map([["bloom",["BloomModel",Ts]],["jais",["JAISModel",zn]],["gpt2",["GPT2Model",An]],["gptj",["GPTJModel",Gn]],["gpt_bigcode",["GPTBigCodeModel",Un]],["gpt_neo",["GPTNeoModel",Nn]],["gpt_neox",["GPTNeoXModel",Dn]],["codegen",["CodeGenModel",Qn]],["llama",["LlamaModel",Jn]],["granite",["GraniteModel",es]],["cohere",["CohereModel",ss]],["gemma",["GemmaModel",is]],["gemma2",["Gemma2Model",cs]],["openelm",["OpenELMModel",ps]],["qwen2",["Qwen2Model",ms]],["phi",["PhiModel",ws]],["phi3",["Phi3Model",ys]],["mpt",["MptModel",Cs]],["opt",["OPTModel",As]],["mistral",["MistralModel",mi]],["starcoder2",["Starcoder2Model",wi]],["falcon",["FalconModel",yi]],["stablelm",["StableLmModel",Ii]]]),aa=new Map([["speecht5",["SpeechT5ForSpeechToText",ci]],["whisper",["WhisperForConditionalGeneration",an]]]),la=new Map([["speecht5",["SpeechT5ForTextToSpeech",di]]]),ca=new Map([["vits",["VitsModel",Pi]],["musicgen",["MusicgenForConditionalGeneration",Gi]]]),da=new Map([["bert",["BertForSequenceClassification",Q]],["roformer",["RoFormerForSequenceClassification",ne]],["electra",["ElectraForSequenceClassification",_e]],["esm",["EsmForSequenceClassification",qe]],["convbert",["ConvBertForSequenceClassification",le]],["camembert",["CamembertForSequenceClassification",be]],["deberta",["DebertaForSequenceClassification",Fe]],["deberta-v2",["DebertaV2ForSequenceClassification",Le]],["mpnet",["MPNetForSequenceClassification",et]],["albert",["AlbertForSequenceClassification",dt]],["distilbert",["DistilBertForSequenceClassification",Oe]],["roberta",["RobertaForSequenceClassification",Vt]],["xlm",["XLMForSequenceClassification",Wt]],["xlm-roberta",["XLMRobertaForSequenceClassification",Kt]],["bart",["BartForSequenceClassification",vt]],["mbart",["MBartForSequenceClassification",St]],["mobilebert",["MobileBertForSequenceClassification",He]],["squeezebert",["SqueezeBertForSequenceClassification",it]]]),ua=new Map([["bert",["BertForTokenClassification",H]],["roformer",["RoFormerForTokenClassification",se]],["electra",["ElectraForTokenClassification",me]],["esm",["EsmForTokenClassification",Ue]],["convbert",["ConvBertForTokenClassification",ce]],["camembert",["CamembertForTokenClassification",ye]],["deberta",["DebertaForTokenClassification",Ce]],["deberta-v2",["DebertaV2ForTokenClassification",ze]],["mpnet",["MPNetForTokenClassification",tt]],["distilbert",["DistilBertForTokenClassification",je]],["roberta",["RobertaForTokenClassification",Rt]],["xlm",["XLMForTokenClassification",Xt]],["xlm-roberta",["XLMRobertaForTokenClassification",Zt]]]),pa=new Map([["t5",["T5ForConditionalGeneration",mt]],["longt5",["LongT5ForConditionalGeneration",wt]],["mt5",["MT5ForConditionalGeneration",yt]],["bart",["BartForConditionalGeneration",Tt]],["mbart",["MBartForConditionalGeneration",Pt]],["marian",["MarianMTModel",To]],["m2m_100",["M2M100ForConditionalGeneration",Co]],["blenderbot",["BlenderbotForConditionalGeneration",zt]],["blenderbot-small",["BlenderbotSmallForConditionalGeneration",Nt]]]),ha=new Map([["bloom",["BloomForCausalLM",vs]],["gpt2",["GPT2LMHeadModel",En]],["jais",["JAISLMHeadModel",In]],["gptj",["GPTJForCausalLM",$n]],["gpt_bigcode",["GPTBigCodeForCausalLM",Wn]],["gpt_neo",["GPTNeoForCausalLM",On]],["gpt_neox",["GPTNeoXForCausalLM",Vn]],["codegen",["CodeGenForCausalLM",Hn]],["llama",["LlamaForCausalLM",Kn]],["granite",["GraniteForCausalLM",ts]],["cohere",["CohereForCausalLM",rs]],["gemma",["GemmaForCausalLM",as]],["gemma2",["Gemma2ForCausalLM",ds]],["openelm",["OpenELMForCausalLM",hs]],["qwen2",["Qwen2ForCausalLM",fs]],["phi",["PhiForCausalLM",Ms]],["phi3",["Phi3ForCausalLM",xs]],["mpt",["MptForCausalLM",Ps]],["opt",["OPTForCausalLM",Es]],["mbart",["MBartForCausalLM",At]],["mistral",["MistralForCausalLM",fi]],["starcoder2",["Starcoder2ForCausalLM",Mi]],["falcon",["FalconForCausalLM",xi]],["trocr",["TrOCRForCausalLM",hi]],["stablelm",["StableLmForCausalLM",Bi]]]),_a=new Map([["bert",["BertForMaskedLM",X]],["roformer",["RoFormerForMaskedLM",te]],["electra",["ElectraForMaskedLM",he]],["esm",["EsmForMaskedLM",$e]],["convbert",["ConvBertForMaskedLM",ae]],["camembert",["CamembertForMaskedLM",Me]],["deberta",["DebertaForMaskedLM",ve]],["deberta-v2",["DebertaV2ForMaskedLM",Ee]],["mpnet",["MPNetForMaskedLM",Ze]],["albert",["AlbertForMaskedLM",pt]],["distilbert",["DistilBertForMaskedLM",Ve]],["roberta",["RobertaForMaskedLM",Dt]],["xlm",["XLMWithLMHeadModel",Ut]],["xlm-roberta",["XLMRobertaForMaskedLM",Jt]],["mobilebert",["MobileBertForMaskedLM",Qe]],["squeezebert",["SqueezeBertForMaskedLM",ot]]]),ma=new Map([["bert",["BertForQuestionAnswering",Y]],["roformer",["RoFormerForQuestionAnswering",re]],["electra",["ElectraForQuestionAnswering",fe]],["convbert",["ConvBertForQuestionAnswering",de]],["camembert",["CamembertForQuestionAnswering",xe]],["deberta",["DebertaForQuestionAnswering",Pe]],["deberta-v2",["DebertaV2ForQuestionAnswering",Ie]],["mpnet",["MPNetForQuestionAnswering",nt]],["albert",["AlbertForQuestionAnswering",ut]],["distilbert",["DistilBertForQuestionAnswering",De]],["roberta",["RobertaForQuestionAnswering",Gt]],["xlm",["XLMForQuestionAnswering",Qt]],["xlm-roberta",["XLMRobertaForQuestionAnswering",en]],["mobilebert",["MobileBertForQuestionAnswering",Ye]],["squeezebert",["SqueezeBertForQuestionAnswering",at]]]),fa=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",ln]]]),ga=new Map([["llava",["LlavaForConditionalGeneration",dn]],["moondream1",["Moondream1ForConditionalGeneration",un]],["florence2",["Florence2ForConditionalGeneration",hn]]]),wa=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",ln]]]),Ma=new Map([["vit",["ViTForImageClassification",Is]],["pvt",["PvtForImageClassification",Os]],["vit_msn",["ViTMSNForImageClassification",Gs]],["fastvit",["FastViTForImageClassification",Xs]],["mobilevit",["MobileViTForImageClassification",Ks]],["mobilevitv2",["MobileViTV2ForImageClassification",tr]],["beit",["BeitForImageClassification",dr]],["deit",["DeiTForImageClassification",Cr]],["hiera",["HieraForImageClassification",Ar]],["convnext",["ConvNextForImageClassification",ao]],["convnextv2",["ConvNextV2ForImageClassification",uo]],["dinov2",["Dinov2ForImageClassification",_o]],["resnet",["ResNetForImageClassification",zr]],["swin",["SwinForImageClassification",Nr]],["segformer",["SegformerForImageClassification",Ei]],["efficientnet",["EfficientNetForImageClassification",ji]],["mobilenet_v1",["MobileNetV1ForImageClassification",Ui]],["mobilenet_v2",["MobileNetV2ForImageClassification",Qi]],["mobilenet_v3",["MobileNetV3ForImageClassification",Ji]],["mobilenet_v4",["MobileNetV4ForImageClassification",ea]]]),ba=new Map([["detr",["DetrForObjectDetection",hr]],["rt_detr",["RTDetrForObjectDetection",Mr]],["table-transformer",["TableTransformerForObjectDetection",kr]],["yolos",["YolosForObjectDetection",go]]]),ya=new Map([["owlvit",["OwlViTForObjectDetection",rr]],["owlv2",["Owlv2ForObjectDetection",ar]]]),xa=new Map([["detr",["DetrForSegmentation",_r]],["clipseg",["CLIPSegForImageSegmentation",Pn]]]),ka=new Map([["segformer",["SegformerForSemanticSegmentation",Li]],["sapiens",["SapiensForSemanticSegmentation",Wr]]]),Ta=new Map([["detr",["DetrForSegmentation",_r]],["maskformer",["MaskFormerForInstanceSegmentation",Zr]]]),va=new Map([["sam",["SamModel",bo]]]),Fa=new Map([["wav2vec2",["Wav2Vec2ForCTC",Ao]],["wav2vec2-bert",["Wav2Vec2BertForCTC",Ho]],["unispeech",["UniSpeechForCTC",Vo]],["unispeech-sat",["UniSpeechSatForCTC",qo]],["wavlm",["WavLMForCTC",si]],["hubert",["HubertForCTC",Zo]]]),Ca=new Map([["wav2vec2",["Wav2Vec2ForSequenceClassification",Eo]],["wav2vec2-bert",["Wav2Vec2BertForSequenceClassification",Yo]],["unispeech",["UniSpeechForSequenceClassification",Ro]],["unispeech-sat",["UniSpeechSatForSequenceClassification",Uo]],["wavlm",["WavLMForSequenceClassification",ri]],["hubert",["HubertForSequenceClassification",ei]],["audio-spectrogram-transformer",["ASTForAudioClassification",sn]]]),Pa=new Map([["wavlm",["WavLMForXVector",oi]]]),Sa=new Map([["unispeech-sat",["UniSpeechSatForAudioFrameClassification",Wo]],["wavlm",["WavLMForAudioFrameClassification",ii]],["wav2vec2",["Wav2Vec2ForAudioFrameClassification",Lo]],["pyannote",["PyAnnoteForAudioFrameClassification",Bo]]]),Aa=new Map([["vitmatte",["VitMatteForImageMatting",Hs]]]),Ea=new Map([["swin2sr",["Swin2SRForImageSuperResolution",Dr]]]),La=new Map([["dpt",["DPTForDepthEstimation",Gr]],["depth_anything",["DepthAnythingForDepthEstimation",qr]],["glpn",["GLPNForDepthEstimation",no]],["sapiens",["SapiensForDepthEstimation",Xr]],["depth_pro",["DepthProForDepthEstimation",Yr]]]),za=new Map([["sapiens",["SapiensForNormalEstimation",Qr]]]),Ia=new Map([["clip",["CLIPVisionModelWithProjection",Mn]],["siglip",["SiglipVisionModel",kn]]]),Ba=[[ra,w],[oa,M],[ia,x],[da,w],[ua,w],[pa,b],[aa,b],[ha,x],[_a,w],[ma,w],[fa,y],[ga,T],[Ma,w],[xa,w],[Ta,w],[ka,w],[Aa,w],[Ea,w],[La,w],[za,w],[ba,w],[ya,w],[va,k],[Fa,w],[Ca,w],[la,b],[ca,w],[Pa,w],[Sa,w],[Ia,w]];for(const[e,t]of Ba)for(const[n,s]of e.values())F.set(n,t),P.set(s,n),C.set(n,s);const Na=[["MusicgenForConditionalGeneration",Gi,v],["CLIPTextModelWithProjection",gn,w],["SiglipTextModel",xn,w],["ClapTextModelWithProjection",vi,w],["ClapAudioModelWithProjection",Fi,w]];for(const[e,t,n]of Na)F.set(e,n),P.set(t,e),C.set(e,t);class Oa extends sa{static MODEL_CLASS_MAPPINGS=Ba.map((e=>e[0]));static BASE_IF_FAIL=!0}class ja extends sa{static MODEL_CLASS_MAPPINGS=[da]}class Da extends sa{static MODEL_CLASS_MAPPINGS=[ua]}class Va extends sa{static MODEL_CLASS_MAPPINGS=[pa]}class Ra extends sa{static MODEL_CLASS_MAPPINGS=[aa]}class Ga extends sa{static MODEL_CLASS_MAPPINGS=[la]}class $a extends sa{static MODEL_CLASS_MAPPINGS=[ca]}class qa extends sa{static MODEL_CLASS_MAPPINGS=[ha]}class Ua extends sa{static MODEL_CLASS_MAPPINGS=[_a]}class Wa extends sa{static MODEL_CLASS_MAPPINGS=[ma]}class Xa extends sa{static MODEL_CLASS_MAPPINGS=[fa]}class Qa extends sa{static MODEL_CLASS_MAPPINGS=[Ma]}class Ha extends sa{static MODEL_CLASS_MAPPINGS=[xa]}class Ya extends sa{static MODEL_CLASS_MAPPINGS=[ka]}class Ja extends sa{static MODEL_CLASS_MAPPINGS=[Ta]}class Ka extends sa{static MODEL_CLASS_MAPPINGS=[ba]}class Za extends sa{static MODEL_CLASS_MAPPINGS=[ya]}class el extends sa{static MODEL_CLASS_MAPPINGS=[va]}class tl extends sa{static MODEL_CLASS_MAPPINGS=[Fa]}class nl extends sa{static MODEL_CLASS_MAPPINGS=[Ca]}class sl extends sa{static MODEL_CLASS_MAPPINGS=[Pa]}class rl extends sa{static MODEL_CLASS_MAPPINGS=[Sa]}class ol extends sa{static MODEL_CLASS_MAPPINGS=[wa]}class il extends sa{static MODEL_CLASS_MAPPINGS=[Aa]}class al extends sa{static MODEL_CLASS_MAPPINGS=[Ea]}class ll extends sa{static MODEL_CLASS_MAPPINGS=[La]}class cl extends sa{static MODEL_CLASS_MAPPINGS=[za]}class dl extends sa{static MODEL_CLASS_MAPPINGS=[Ia]}class ul extends ${constructor({logits:e,past_key_values:t,encoder_outputs:n,decoder_attentions:s=null,cross_attentions:r=null}){super(),this.logits=e,this.past_key_values=t,this.encoder_outputs=n,this.decoder_attentions=s,this.cross_attentions=r}}class pl extends ${constructor({logits:e}){super(),this.logits=e}}class hl extends ${constructor({logits:e,embeddings:t}){super(),this.logits=e,this.embeddings=t}}class _l extends ${constructor({logits:e}){super(),this.logits=e}}class ml extends ${constructor({logits:e}){super(),this.logits=e}}class fl extends ${constructor({start_logits:e,end_logits:t}){super(),this.start_logits=e,this.end_logits=t}}class gl extends ${constructor({logits:e}){super(),this.logits=e}}class wl extends ${constructor({logits:e,past_key_values:t}){super(),this.logits=e,this.past_key_values=t}}class Ml extends ${constructor({alphas:e}){super(),this.alphas=e}}class bl extends ${constructor({waveform:e,spectrogram:t}){super(),this.waveform=e,this.spectrogram=t}}},"./src/models/whisper/common_whisper.js":
+  \***********************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{ASTForAudioClassification:()=>sn,ASTModel:()=>nn,ASTPreTrainedModel:()=>tn,AlbertForMaskedLM:()=>pt,AlbertForQuestionAnswering:()=>ut,AlbertForSequenceClassification:()=>dt,AlbertModel:()=>ct,AlbertPreTrainedModel:()=>lt,AutoModel:()=>Oa,AutoModelForAudioClassification:()=>nl,AutoModelForAudioFrameClassification:()=>rl,AutoModelForCTC:()=>tl,AutoModelForCausalLM:()=>qa,AutoModelForDepthEstimation:()=>ll,AutoModelForDocumentQuestionAnswering:()=>ol,AutoModelForImageClassification:()=>Qa,AutoModelForImageFeatureExtraction:()=>dl,AutoModelForImageMatting:()=>il,AutoModelForImageSegmentation:()=>Ha,AutoModelForImageToImage:()=>al,AutoModelForMaskGeneration:()=>el,AutoModelForMaskedLM:()=>Ua,AutoModelForNormalEstimation:()=>cl,AutoModelForObjectDetection:()=>Ka,AutoModelForQuestionAnswering:()=>Wa,AutoModelForSemanticSegmentation:()=>Ya,AutoModelForSeq2SeqLM:()=>Va,AutoModelForSequenceClassification:()=>ja,AutoModelForSpeechSeq2Seq:()=>Ra,AutoModelForTextToSpectrogram:()=>Ga,AutoModelForTextToWaveform:()=>$a,AutoModelForTokenClassification:()=>Da,AutoModelForUniversalSegmentation:()=>Ja,AutoModelForVision2Seq:()=>Xa,AutoModelForXVector:()=>sl,AutoModelForZeroShotObjectDetection:()=>Za,BartForConditionalGeneration:()=>Tt,BartForSequenceClassification:()=>vt,BartModel:()=>kt,BartPretrainedModel:()=>xt,BaseModelOutput:()=>q,BeitForImageClassification:()=>dr,BeitModel:()=>cr,BeitPreTrainedModel:()=>lr,BertForMaskedLM:()=>X,BertForQuestionAnswering:()=>Y,BertForSequenceClassification:()=>Q,BertForTokenClassification:()=>H,BertModel:()=>W,BertPreTrainedModel:()=>U,BlenderbotForConditionalGeneration:()=>zt,BlenderbotModel:()=>Lt,BlenderbotPreTrainedModel:()=>Et,BlenderbotSmallForConditionalGeneration:()=>Nt,BlenderbotSmallModel:()=>Bt,BlenderbotSmallPreTrainedModel:()=>It,BloomForCausalLM:()=>vs,BloomModel:()=>Ts,BloomPreTrainedModel:()=>ks,CLIPModel:()=>mn,CLIPPreTrainedModel:()=>_n,CLIPSegForImageSegmentation:()=>Pn,CLIPSegModel:()=>Cn,CLIPSegPreTrainedModel:()=>Fn,CLIPTextModel:()=>fn,CLIPTextModelWithProjection:()=>gn,CLIPVisionModel:()=>wn,CLIPVisionModelWithProjection:()=>Mn,CamembertForMaskedLM:()=>Me,CamembertForQuestionAnswering:()=>xe,CamembertForSequenceClassification:()=>be,CamembertForTokenClassification:()=>ye,CamembertModel:()=>we,CamembertPreTrainedModel:()=>ge,CausalLMOutput:()=>gl,CausalLMOutputWithPast:()=>wl,ChineseCLIPModel:()=>vn,ChineseCLIPPreTrainedModel:()=>Tn,ClapAudioModelWithProjection:()=>Fi,ClapModel:()=>Ti,ClapPreTrainedModel:()=>ki,ClapTextModelWithProjection:()=>vi,CodeGenForCausalLM:()=>Hn,CodeGenModel:()=>Qn,CodeGenPreTrainedModel:()=>Xn,CohereForCausalLM:()=>rs,CohereModel:()=>ss,CoherePreTrainedModel:()=>ns,ConvBertForMaskedLM:()=>ae,ConvBertForQuestionAnswering:()=>de,ConvBertForSequenceClassification:()=>le,ConvBertForTokenClassification:()=>ce,ConvBertModel:()=>ie,ConvBertPreTrainedModel:()=>oe,ConvNextForImageClassification:()=>ao,ConvNextModel:()=>io,ConvNextPreTrainedModel:()=>oo,ConvNextV2ForImageClassification:()=>uo,ConvNextV2Model:()=>co,ConvNextV2PreTrainedModel:()=>lo,DPTForDepthEstimation:()=>Gr,DPTModel:()=>Rr,DPTPreTrainedModel:()=>Vr,DebertaForMaskedLM:()=>ve,DebertaForQuestionAnswering:()=>Pe,DebertaForSequenceClassification:()=>Fe,DebertaForTokenClassification:()=>Ce,DebertaModel:()=>Te,DebertaPreTrainedModel:()=>ke,DebertaV2ForMaskedLM:()=>Ee,DebertaV2ForQuestionAnswering:()=>Ie,DebertaV2ForSequenceClassification:()=>Le,DebertaV2ForTokenClassification:()=>ze,DebertaV2Model:()=>Ae,DebertaV2PreTrainedModel:()=>Se,DecisionTransformerModel:()=>na,DecisionTransformerPreTrainedModel:()=>ta,DeiTForImageClassification:()=>Cr,DeiTModel:()=>Fr,DeiTPreTrainedModel:()=>vr,DepthAnythingForDepthEstimation:()=>qr,DepthAnythingPreTrainedModel:()=>$r,DepthProForDepthEstimation:()=>Yr,DepthProPreTrainedModel:()=>Hr,DetrForObjectDetection:()=>hr,DetrForSegmentation:()=>_r,DetrModel:()=>pr,DetrObjectDetectionOutput:()=>mr,DetrPreTrainedModel:()=>ur,DetrSegmentationOutput:()=>fr,Dinov2ForImageClassification:()=>_o,Dinov2Model:()=>ho,Dinov2PreTrainedModel:()=>po,DistilBertForMaskedLM:()=>Ve,DistilBertForQuestionAnswering:()=>De,DistilBertForSequenceClassification:()=>Oe,DistilBertForTokenClassification:()=>je,DistilBertModel:()=>Ne,DistilBertPreTrainedModel:()=>Be,DonutSwinModel:()=>ro,DonutSwinPreTrainedModel:()=>so,EfficientNetForImageClassification:()=>ji,EfficientNetModel:()=>Oi,EfficientNetPreTrainedModel:()=>Ni,ElectraForMaskedLM:()=>he,ElectraForQuestionAnswering:()=>fe,ElectraForSequenceClassification:()=>_e,ElectraForTokenClassification:()=>me,ElectraModel:()=>pe,ElectraPreTrainedModel:()=>ue,EsmForMaskedLM:()=>$e,EsmForSequenceClassification:()=>qe,EsmForTokenClassification:()=>Ue,EsmModel:()=>Ge,EsmPreTrainedModel:()=>Re,FalconForCausalLM:()=>xi,FalconModel:()=>yi,FalconPreTrainedModel:()=>bi,FastViTForImageClassification:()=>Xs,FastViTModel:()=>Ws,FastViTPreTrainedModel:()=>Us,Florence2ForConditionalGeneration:()=>hn,Florence2PreTrainedModel:()=>pn,GLPNForDepthEstimation:()=>no,GLPNModel:()=>to,GLPNPreTrainedModel:()=>eo,GPT2LMHeadModel:()=>En,GPT2Model:()=>An,GPT2PreTrainedModel:()=>Sn,GPTBigCodeForCausalLM:()=>Wn,GPTBigCodeModel:()=>Un,GPTBigCodePreTrainedModel:()=>qn,GPTJForCausalLM:()=>$n,GPTJModel:()=>Gn,GPTJPreTrainedModel:()=>Rn,GPTNeoForCausalLM:()=>On,GPTNeoModel:()=>Nn,GPTNeoPreTrainedModel:()=>Bn,GPTNeoXForCausalLM:()=>Vn,GPTNeoXModel:()=>Dn,GPTNeoXPreTrainedModel:()=>jn,Gemma2ForCausalLM:()=>ds,Gemma2Model:()=>cs,Gemma2PreTrainedModel:()=>ls,GemmaForCausalLM:()=>as,GemmaModel:()=>is,GemmaPreTrainedModel:()=>os,GraniteForCausalLM:()=>ts,GraniteModel:()=>es,GranitePreTrainedModel:()=>Zn,GroupViTModel:()=>qs,GroupViTPreTrainedModel:()=>$s,HieraForImageClassification:()=>Ar,HieraModel:()=>Sr,HieraPreTrainedModel:()=>Pr,HubertForCTC:()=>Zo,HubertForSequenceClassification:()=>ei,HubertModel:()=>Ko,HubertPreTrainedModel:()=>Jo,ImageMattingOutput:()=>Ml,JAISLMHeadModel:()=>In,JAISModel:()=>zn,JAISPreTrainedModel:()=>Ln,LlamaForCausalLM:()=>Kn,LlamaModel:()=>Jn,LlamaPreTrainedModel:()=>Yn,LlavaForConditionalGeneration:()=>dn,LlavaPreTrainedModel:()=>cn,LongT5ForConditionalGeneration:()=>wt,LongT5Model:()=>gt,LongT5PreTrainedModel:()=>ft,M2M100ForConditionalGeneration:()=>Co,M2M100Model:()=>Fo,M2M100PreTrainedModel:()=>vo,MBartForCausalLM:()=>At,MBartForConditionalGeneration:()=>Pt,MBartForSequenceClassification:()=>St,MBartModel:()=>Ct,MBartPreTrainedModel:()=>Ft,MPNetForMaskedLM:()=>Ze,MPNetForQuestionAnswering:()=>nt,MPNetForSequenceClassification:()=>et,MPNetForTokenClassification:()=>tt,MPNetModel:()=>Ke,MPNetPreTrainedModel:()=>Je,MT5ForConditionalGeneration:()=>yt,MT5Model:()=>bt,MT5PreTrainedModel:()=>Mt,MarianMTModel:()=>To,MarianModel:()=>ko,MarianPreTrainedModel:()=>xo,MaskFormerForInstanceSegmentation:()=>Zr,MaskFormerModel:()=>Kr,MaskFormerPreTrainedModel:()=>Jr,MaskedLMOutput:()=>ml,MistralForCausalLM:()=>fi,MistralModel:()=>mi,MistralPreTrainedModel:()=>_i,MobileBertForMaskedLM:()=>Qe,MobileBertForQuestionAnswering:()=>Ye,MobileBertForSequenceClassification:()=>He,MobileBertModel:()=>Xe,MobileBertPreTrainedModel:()=>We,MobileNetV1ForImageClassification:()=>Ui,MobileNetV1Model:()=>qi,MobileNetV1PreTrainedModel:()=>$i,MobileNetV2ForImageClassification:()=>Qi,MobileNetV2Model:()=>Xi,MobileNetV2PreTrainedModel:()=>Wi,MobileNetV3ForImageClassification:()=>Ji,MobileNetV3Model:()=>Yi,MobileNetV3PreTrainedModel:()=>Hi,MobileNetV4ForImageClassification:()=>ea,MobileNetV4Model:()=>Zi,MobileNetV4PreTrainedModel:()=>Ki,MobileViTForImageClassification:()=>Ks,MobileViTModel:()=>Js,MobileViTPreTrainedModel:()=>Ys,MobileViTV2ForImageClassification:()=>tr,MobileViTV2Model:()=>er,MobileViTV2PreTrainedModel:()=>Zs,ModelOutput:()=>$,Moondream1ForConditionalGeneration:()=>un,MptForCausalLM:()=>Ps,MptModel:()=>Cs,MptPreTrainedModel:()=>Fs,MusicgenForCausalLM:()=>Ri,MusicgenForConditionalGeneration:()=>Gi,MusicgenModel:()=>Vi,MusicgenPreTrainedModel:()=>Di,NomicBertModel:()=>K,NomicBertPreTrainedModel:()=>J,OPTForCausalLM:()=>Es,OPTModel:()=>As,OPTPreTrainedModel:()=>Ss,OpenELMForCausalLM:()=>hs,OpenELMModel:()=>ps,OpenELMPreTrainedModel:()=>us,OwlViTForObjectDetection:()=>rr,OwlViTModel:()=>sr,OwlViTPreTrainedModel:()=>nr,Owlv2ForObjectDetection:()=>ar,Owlv2Model:()=>ir,Owlv2PreTrainedModel:()=>or,Phi3ForCausalLM:()=>xs,Phi3Model:()=>ys,Phi3PreTrainedModel:()=>bs,PhiForCausalLM:()=>Ms,PhiModel:()=>ws,PhiPreTrainedModel:()=>gs,PreTrainedModel:()=>G,PretrainedMixin:()=>sa,PvtForImageClassification:()=>Os,PvtModel:()=>Ns,PvtPreTrainedModel:()=>Bs,PyAnnoteForAudioFrameClassification:()=>Bo,PyAnnoteModel:()=>Io,PyAnnotePreTrainedModel:()=>zo,QuestionAnsweringModelOutput:()=>fl,Qwen2ForCausalLM:()=>fs,Qwen2Model:()=>ms,Qwen2PreTrainedModel:()=>_s,RTDetrForObjectDetection:()=>Mr,RTDetrModel:()=>wr,RTDetrObjectDetectionOutput:()=>br,RTDetrPreTrainedModel:()=>gr,ResNetForImageClassification:()=>zr,ResNetModel:()=>Lr,ResNetPreTrainedModel:()=>Er,RoFormerForMaskedLM:()=>te,RoFormerForQuestionAnswering:()=>re,RoFormerForSequenceClassification:()=>ne,RoFormerForTokenClassification:()=>se,RoFormerModel:()=>ee,RoFormerPreTrainedModel:()=>Z,RobertaForMaskedLM:()=>Dt,RobertaForQuestionAnswering:()=>Gt,RobertaForSequenceClassification:()=>Vt,RobertaForTokenClassification:()=>Rt,RobertaModel:()=>jt,RobertaPreTrainedModel:()=>Ot,SamImageSegmentationOutput:()=>yo,SamModel:()=>bo,SamPreTrainedModel:()=>Mo,SapiensForDepthEstimation:()=>Xr,SapiensForNormalEstimation:()=>Qr,SapiensForSemanticSegmentation:()=>Wr,SapiensPreTrainedModel:()=>Ur,SegformerForImageClassification:()=>Ei,SegformerForSemanticSegmentation:()=>Li,SegformerModel:()=>Ai,SegformerPreTrainedModel:()=>Si,Seq2SeqLMOutput:()=>ul,SequenceClassifierOutput:()=>pl,SiglipModel:()=>yn,SiglipPreTrainedModel:()=>bn,SiglipTextModel:()=>xn,SiglipVisionModel:()=>kn,SpeechT5ForSpeechToText:()=>ci,SpeechT5ForTextToSpeech:()=>di,SpeechT5HifiGan:()=>ui,SpeechT5Model:()=>li,SpeechT5PreTrainedModel:()=>ai,SqueezeBertForMaskedLM:()=>ot,SqueezeBertForQuestionAnswering:()=>at,SqueezeBertForSequenceClassification:()=>it,SqueezeBertModel:()=>rt,SqueezeBertPreTrainedModel:()=>st,StableLmForCausalLM:()=>Bi,StableLmModel:()=>Ii,StableLmPreTrainedModel:()=>zi,Starcoder2ForCausalLM:()=>Mi,Starcoder2Model:()=>wi,Starcoder2PreTrainedModel:()=>gi,Swin2SRForImageSuperResolution:()=>Dr,Swin2SRModel:()=>jr,Swin2SRPreTrainedModel:()=>Or,SwinForImageClassification:()=>Nr,SwinModel:()=>Br,SwinPreTrainedModel:()=>Ir,T5ForConditionalGeneration:()=>mt,T5Model:()=>_t,T5PreTrainedModel:()=>ht,TableTransformerForObjectDetection:()=>kr,TableTransformerModel:()=>xr,TableTransformerObjectDetectionOutput:()=>Tr,TableTransformerPreTrainedModel:()=>yr,TokenClassifierOutput:()=>_l,TrOCRForCausalLM:()=>hi,TrOCRPreTrainedModel:()=>pi,UniSpeechForCTC:()=>Vo,UniSpeechForSequenceClassification:()=>Ro,UniSpeechModel:()=>Do,UniSpeechPreTrainedModel:()=>jo,UniSpeechSatForAudioFrameClassification:()=>Wo,UniSpeechSatForCTC:()=>qo,UniSpeechSatForSequenceClassification:()=>Uo,UniSpeechSatModel:()=>$o,UniSpeechSatPreTrainedModel:()=>Go,ViTForImageClassification:()=>Is,ViTMAEModel:()=>Ds,ViTMAEPreTrainedModel:()=>js,ViTMSNForImageClassification:()=>Gs,ViTMSNModel:()=>Rs,ViTMSNPreTrainedModel:()=>Vs,ViTModel:()=>zs,ViTPreTrainedModel:()=>Ls,VisionEncoderDecoderModel:()=>ln,VitMatteForImageMatting:()=>Hs,VitMattePreTrainedModel:()=>Qs,VitsModel:()=>Pi,VitsModelOutput:()=>bl,VitsPreTrainedModel:()=>Ci,Wav2Vec2BertForCTC:()=>Ho,Wav2Vec2BertForSequenceClassification:()=>Yo,Wav2Vec2BertModel:()=>Qo,Wav2Vec2BertPreTrainedModel:()=>Xo,Wav2Vec2ForAudioFrameClassification:()=>Lo,Wav2Vec2ForCTC:()=>Ao,Wav2Vec2ForSequenceClassification:()=>Eo,Wav2Vec2Model:()=>So,Wav2Vec2PreTrainedModel:()=>Po,WavLMForAudioFrameClassification:()=>ii,WavLMForCTC:()=>si,WavLMForSequenceClassification:()=>ri,WavLMForXVector:()=>oi,WavLMModel:()=>ni,WavLMPreTrainedModel:()=>ti,WeSpeakerResNetModel:()=>Oo,WeSpeakerResNetPreTrainedModel:()=>No,WhisperForConditionalGeneration:()=>an,WhisperModel:()=>on,WhisperPreTrainedModel:()=>rn,XLMForQuestionAnswering:()=>Qt,XLMForSequenceClassification:()=>Wt,XLMForTokenClassification:()=>Xt,XLMModel:()=>qt,XLMPreTrainedModel:()=>$t,XLMRobertaForMaskedLM:()=>Jt,XLMRobertaForQuestionAnswering:()=>en,XLMRobertaForSequenceClassification:()=>Kt,XLMRobertaForTokenClassification:()=>Zt,XLMRobertaModel:()=>Yt,XLMRobertaPreTrainedModel:()=>Ht,XLMWithLMHeadModel:()=>Ut,XVectorOutput:()=>hl,YolosForObjectDetection:()=>go,YolosModel:()=>fo,YolosObjectDetectionOutput:()=>wo,YolosPreTrainedModel:()=>mo});var s=n(/*! ./configs.js */"./src/configs.js"),r=n(/*! ./backends/onnx.js */"./src/backends/onnx.js"),o=n(/*! ./utils/dtypes.js */"./src/utils/dtypes.js"),i=n(/*! ./utils/generic.js */"./src/utils/generic.js"),a=n(/*! ./utils/core.js */"./src/utils/core.js"),l=n(/*! ./utils/hub.js */"./src/utils/hub.js"),c=n(/*! ./generation/logits_process.js */"./src/generation/logits_process.js"),d=n(/*! ./generation/configuration_utils.js */"./src/generation/configuration_utils.js"),u=n(/*! ./utils/tensor.js */"./src/utils/tensor.js"),p=n(/*! ./utils/maths.js */"./src/utils/maths.js"),h=n(/*! ./generation/stopping_criteria.js */"./src/generation/stopping_criteria.js"),_=n(/*! ./generation/logits_sampler.js */"./src/generation/logits_sampler.js"),m=n(/*! ./env.js */"./src/env.js"),f=n(/*! ./models/whisper/generation_whisper.js */"./src/models/whisper/generation_whisper.js"),g=n(/*! ./models/whisper/common_whisper.js */"./src/models/whisper/common_whisper.js");const w=0,M=1,b=2,y=3,x=4,k=5,T=6,v=7,F=new Map,C=new Map,P=new Map;async function S(e,t,n){return Object.fromEntries(await Promise.all(Object.keys(t).map((async i=>{const{buffer:a,session_options:c,session_config:d}=await async function(e,t,n){const i=n.config?.["transformers.js_config"]??{};let a=n.device??i.device;a&&"string"!=typeof a&&(a.hasOwnProperty(t)?a=a[t]:(console.warn(`device not specified for "${t}". Using the default device.`),a=null));const c=a??(m.apis.IS_NODE_ENV?"cpu":"wasm"),d=(0,r.deviceToExecutionProviders)(c);let u=n.dtype??i.dtype;"string"!=typeof u&&(u&&u.hasOwnProperty(t)?u=u[t]:(u=o.DEFAULT_DEVICE_DTYPE_MAPPING[c]??o.DATA_TYPES.fp32,console.warn(`dtype not specified for "${t}". Using the default dtype (${u}) for this device (${c}).`)));const p=u;if(!o.DEFAULT_DTYPE_SUFFIX_MAPPING.hasOwnProperty(p))throw new Error(`Invalid dtype: ${p}. Should be one of: ${Object.keys(o.DATA_TYPES).join(", ")}`);if(p===o.DATA_TYPES.fp16&&"webgpu"===c&&!await(0,o.isWebGpuFp16Supported)())throw new Error(`The device (${c}) does not support fp16.`);const h=i.kv_cache_dtype?"string"==typeof i.kv_cache_dtype?i.kv_cache_dtype:i.kv_cache_dtype[p]??"float32":void 0;if(h&&!["float32","float16"].includes(h))throw new Error(`Invalid kv_cache_dtype: ${h}. Should be one of: float32, float16`);const _={dtype:p,kv_cache_dtype:h},f=o.DEFAULT_DTYPE_SUFFIX_MAPPING[p],g=`${n.subfolder??""}/${t}${f}.onnx`,w={...n.session_options};w.executionProviders??=d;const M=i.free_dimension_overrides;M?w.freeDimensionOverrides??=M:c.startsWith("webnn")&&!w.freeDimensionOverrides&&console.warn('WebNN does not currently support dynamic shapes and requires `free_dimension_overrides` to be set in config.json as a field within "transformers.js_config". When `free_dimension_overrides` is not set, you may experience significant performance degradation.');const b=(0,l.getModelFile)(e,g,!0,n),y=n.use_external_data_format??i.use_external_data_format;let x=[];if(y&&(!0===y||"object"==typeof y&&y.hasOwnProperty(t)&&!0===y[t])){if(m.apis.IS_NODE_ENV)throw new Error("External data format is not yet supported in Node.js");const s=`${t}${f}.onnx_data`,r=`${n.subfolder??""}/${s}`;x.push(new Promise((async(t,o)=>{const i=await(0,l.getModelFile)(e,r,!0,n);t({path:s,data:i})})))}else void 0!==w.externalData&&(x=w.externalData.map((async t=>{if("string"==typeof t.data){const s=await(0,l.getModelFile)(e,t.data,!0,n);return{...t,data:s}}return t})));if(x.length>0&&(w.externalData=await Promise.all(x)),"webgpu"===c){const e=(0,s.getKeyValueShapes)(n.config,{prefix:"present"});if(Object.keys(e).length>0&&!(0,r.isONNXProxy)()){const t={};for(const n in e)t[n]="gpu-buffer";w.preferredOutputLocation=t}}return{buffer:await b,session_options:w,session_config:_}}(e,t[i],n);return[i,await(0,r.createInferenceSession)(a,c,d)]}))))}async function A(e,t,n){return Object.fromEntries(await Promise.all(Object.keys(t).map((async s=>[s,await(0,l.getModelJSON)(e,t[s],!1,n)]))))}async function E(e,t){const n=function(e,t){const n=Object.create(null),s=[];for(const o of e.inputNames){const e=t[o];e instanceof u.Tensor?n[o]=(0,r.isONNXProxy)()?e.clone():e:s.push(o)}if(s.length>0)throw new Error(`An error occurred during model execution: "Missing the following inputs: ${s.join(", ")}.`);const o=Object.keys(t).length,i=e.inputNames.length;if(o>i){let n=Object.keys(t).filter((t=>!e.inputNames.includes(t)));console.warn(`WARNING: Too many inputs were provided (${o} > ${i}). The following inputs will be ignored: "${n.join(", ")}".`)}return n}(e,t);try{const t=Object.fromEntries(Object.entries(n).map((([e,t])=>[e,t.ort_tensor])));let s=await e.run(t);return s=L(s),s}catch(e){throw console.error(`An error occurred during model execution: "${e}".`),console.error("Inputs given to model:",n),e}}function L(e){for(let t in e)(0,r.isONNXTensor)(e[t])?e[t]=new u.Tensor(e[t]):"object"==typeof e[t]&&L(e[t]);return e}function z(e){if(e instanceof u.Tensor)return e;if(0===e.length)throw Error("items must be non-empty");if(Array.isArray(e[0])){if(e.some((t=>t.length!==e[0].length)))throw Error("Unable to create tensor, you should probably activate truncation and/or padding with 'padding=True' and/or 'truncation=True' to have batched tensors with the same length.");return new u.Tensor("int64",BigInt64Array.from(e.flat().map((e=>BigInt(e)))),[e.length,e[0].length])}return new u.Tensor("int64",BigInt64Array.from(e.map((e=>BigInt(e)))),[1,e.length])}function I(e){return new u.Tensor("bool",[e],[1])}async function B(e,t){let{encoder_outputs:n,input_ids:s,decoder_input_ids:r,...o}=t;if(!n){const s=(0,a.pick)(t,e.sessions.model.inputNames);n=(await N(e,s)).last_hidden_state}o.input_ids=r,o.encoder_hidden_states=n,e.sessions.decoder_model_merged.inputNames.includes("encoder_attention_mask")&&(o.encoder_attention_mask=t.attention_mask);return await O(e,o,!0)}async function N(e,t){const n=e.sessions.model,s=(0,a.pick)(t,n.inputNames);if(n.inputNames.includes("inputs_embeds")&&!s.inputs_embeds){if(!t.input_ids)throw new Error("Both `input_ids` and `inputs_embeds` are missing in the model inputs.");s.inputs_embeds=await e.encode_text({input_ids:t.input_ids})}return n.inputNames.includes("token_type_ids")&&!s.token_type_ids&&(s.token_type_ids=new u.Tensor("int64",new BigInt64Array(s.input_ids.data.length),s.input_ids.dims)),await E(n,s)}async function O(e,t,n=!1){const s=e.sessions[n?"decoder_model_merged":"model"],{past_key_values:r,...o}=t;s.inputNames.includes("use_cache_branch")&&(o.use_cache_branch=I(!!r)),s.inputNames.includes("position_ids")&&o.attention_mask&&!o.position_ids&&(o.position_ids=function(e,t=null){const{input_ids:n,inputs_embeds:s,attention_mask:r}=e,[o,i]=r.dims,a=new BigInt64Array(r.data.length);for(let e=0;e<o;++e){const t=e*i;let n=BigInt(0);for(let e=0;e<i;++e){const s=t+e;0n===r.data[s]?a[s]=BigInt(1):(a[s]=n,n+=r.data[s])}}let l=new u.Tensor("int64",a,r.dims);if(t){const e=-(n??s).dims.at(1);l=l.slice(null,[e,null])}return l}(o,r)),e.addPastKeyValues(o,r);const i=(0,a.pick)(o,s.inputNames);return await E(s,i)}async function j(e,{input_ids:t=null,attention_mask:n=null,pixel_values:s=null,position_ids:r=null,inputs_embeds:o=null,past_key_values:i=null,generation_config:a=null,logits_processor:l=null,...c}){if(!o)if(o=await e.encode_text({input_ids:t}),s&&1!==t.dims[1]){const r=await e.encode_image({pixel_values:s});({inputs_embeds:o,attention_mask:n}=e._merge_input_ids_with_image_features({image_features:r,inputs_embeds:o,input_ids:t,attention_mask:n}))}else if(i&&s&&1===t.dims[1]){const e=t.dims[1],s=Object.values(i)[0].dims.at(-2);n=(0,u.cat)([(0,u.ones)([t.dims[0],s]),n.slice(null,[n.dims[1]-e,n.dims[1]])],1)}return await O(e,{inputs_embeds:o,past_key_values:i,attention_mask:n,position_ids:r,generation_config:a,logits_processor:l},!0)}function D(e,t,n,s){if(n.past_key_values){const t=Object.values(n.past_key_values)[0].dims.at(-2),{input_ids:s,attention_mask:r}=n;if(r&&r.dims[1]>s.dims[1]);else if(t<s.dims[1])n.input_ids=s.slice(null,[t,null]);else if(null!=e.config.image_token_index&&s.data.some((t=>t==e.config.image_token_index))){const r=e.config.num_image_tokens;if(!r)throw new Error("`num_image_tokens` is missing in the model configuration.");const o=s.dims[1]-(t-r);n.input_ids=s.slice(null,[-o,null]),n.attention_mask=(0,u.ones)([1,t+o])}}return n}function V(e,t,n,s){return n.past_key_values&&(t=t.map((e=>[e.at(-1)]))),{...n,decoder_input_ids:z(t)}}function R(e,...t){return e.config.is_encoder_decoder?V(e,...t):D(e,...t)}class G extends i.Callable{main_input_name="input_ids";forward_params=["input_ids","attention_mask"];constructor(e,t,n){super(),this.config=e,this.sessions=t,this.configs=n;const s=P.get(this.constructor),r=F.get(s);switch(this.can_generate=!1,this._forward=null,this._prepare_inputs_for_generation=null,r){case x:this.can_generate=!0,this._forward=O,this._prepare_inputs_for_generation=D;break;case b:case y:case v:this.can_generate=!0,this._forward=B,this._prepare_inputs_for_generation=V;break;case M:this._forward=B;break;case T:this.can_generate=!0,this._forward=j,this._prepare_inputs_for_generation=R;break;default:this._forward=N}this.can_generate&&this.forward_params.push("past_key_values"),this.custom_config=this.config["transformers.js_config"]??{}}async dispose(){const e=[];for(const t of Object.values(this.sessions))t?.handler?.dispose&&e.push(t.handler.dispose());return await Promise.all(e)}static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:r=null,local_files_only:o=!1,revision:i="main",model_file_name:a=null,subfolder:l="onnx",device:c=null,dtype:d=null,use_external_data_format:u=null,session_options:p={}}={}){let h={progress_callback:t,config:n,cache_dir:r,local_files_only:o,revision:i,model_file_name:a,subfolder:l,device:c,dtype:d,use_external_data_format:u,session_options:p};const _=P.get(this),m=F.get(_);let f;if(n=h.config=await s.AutoConfig.from_pretrained(e,h),m===x)f=await Promise.all([S(e,{model:h.model_file_name??"model"},h),A(e,{generation_config:"generation_config.json"},h)]);else if(m===b||m===y)f=await Promise.all([S(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},h),A(e,{generation_config:"generation_config.json"},h)]);else if(m===k)f=await Promise.all([S(e,{model:"vision_encoder",prompt_encoder_mask_decoder:"prompt_encoder_mask_decoder"},h)]);else if(m===M)f=await Promise.all([S(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},h)]);else if(m===T){const t={embed_tokens:"embed_tokens",vision_encoder:"vision_encoder",decoder_model_merged:"decoder_model_merged"};n.is_encoder_decoder&&(t.model="encoder_model"),f=await Promise.all([S(e,t,h),A(e,{generation_config:"generation_config.json"},h)])}else m===v?f=await Promise.all([S(e,{model:"text_encoder",decoder_model_merged:"decoder_model_merged",encodec_decode:"encodec_decode"},h),A(e,{generation_config:"generation_config.json"},h)]):(m!==w&&console.warn(`Model type for '${_??n?.model_type}' not found, assuming encoder-only architecture. Please report this at https://github.com/xenova/transformers.js/issues/new/choose.`),f=await Promise.all([S(e,{model:h.model_file_name??"model"},h)]));return new this(n,...f)}async _call(e){return await this.forward(e)}async forward(e){return await this._forward(this,e)}get generation_config(){return this.configs?.generation_config??null}_get_logits_warper(e){const t=new c.LogitsProcessorList;return null!==e.temperature&&1!==e.temperature&&t.push(new c.TemperatureLogitsWarper(e.temperature)),null!==e.top_k&&0!==e.top_k&&t.push(new c.TopKLogitsWarper(e.top_k)),null!==e.top_p&&e.top_p<1&&t.push(new c.TopPLogitsWarper(e.top_p)),t}_get_logits_processor(e,t,n=null){const s=new c.LogitsProcessorList;if(null!==e.repetition_penalty&&1!==e.repetition_penalty&&s.push(new c.RepetitionPenaltyLogitsProcessor(e.repetition_penalty)),null!==e.no_repeat_ngram_size&&e.no_repeat_ngram_size>0&&s.push(new c.NoRepeatNGramLogitsProcessor(e.no_repeat_ngram_size)),null!==e.bad_words_ids&&s.push(new c.NoBadWordsLogitsProcessor(e.bad_words_ids,e.eos_token_id)),null!==e.min_length&&null!==e.eos_token_id&&e.min_length>0&&s.push(new c.MinLengthLogitsProcessor(e.min_length,e.eos_token_id)),null!==e.min_new_tokens&&null!==e.eos_token_id&&e.min_new_tokens>0&&s.push(new c.MinNewTokensLengthLogitsProcessor(t,e.min_new_tokens,e.eos_token_id)),null!==e.forced_bos_token_id&&s.push(new c.ForcedBOSTokenLogitsProcessor(e.forced_bos_token_id)),null!==e.forced_eos_token_id&&s.push(new c.ForcedEOSTokenLogitsProcessor(e.max_length,e.forced_eos_token_id)),null!==e.begin_suppress_tokens){const n=t>1||null===e.forced_bos_token_id?t:t+1;s.push(new c.SuppressTokensAtBeginLogitsProcessor(e.begin_suppress_tokens,n))}return null!==e.guidance_scale&&e.guidance_scale>1&&s.push(new c.ClassifierFreeGuidanceLogitsProcessor(e.guidance_scale)),null!==n&&s.extend(n),s}_prepare_generation_config(e,t,n=d.GenerationConfig){const s={...this.config};for(const e of["decoder","generator","text_config"])e in s&&Object.assign(s,s[e]);const r=new n(s);return Object.assign(r,this.generation_config??{}),e&&Object.assign(r,e),t&&Object.assign(r,(0,a.pick)(t,Object.getOwnPropertyNames(r))),r}_get_stopping_criteria(e,t=null){const n=new h.StoppingCriteriaList;return null!==e.max_length&&n.push(new h.MaxLengthCriteria(e.max_length,this.config.max_position_embeddings??null)),null!==e.eos_token_id&&n.push(new h.EosTokenCriteria(e.eos_token_id)),t&&n.extend(t),n}_validate_model_class(){if(!this.can_generate){const e=[ha,fa,pa,aa],t=P.get(this.constructor),n=new Set,s=this.config.model_type;for(const t of e){const e=t.get(s);e&&n.add(e[0])}let r=`The current model class (${t}) is not compatible with \`.generate()\`, as it doesn't have a language model head.`;throw n.size>0&&(r+=` Please use the following class instead: ${[...n].join(", ")}`),Error(r)}}prepare_inputs_for_generation(...e){return this._prepare_inputs_for_generation(this,...e)}_update_model_kwargs_for_generation({generated_input_ids:e,outputs:t,model_inputs:n,is_encoder_decoder:s}){return n.past_key_values=this.getPastKeyValues(t,n.past_key_values),n.input_ids=new u.Tensor("int64",e.flat(),[e.length,1]),s||(n.attention_mask=(0,u.cat)([n.attention_mask,(0,u.ones)([n.attention_mask.dims[0],1])],1)),n.position_ids=null,n}_prepare_model_inputs({inputs:e,bos_token_id:t,model_kwargs:n}){const s=(0,a.pick)(n,this.forward_params),r=this.main_input_name;if(r in s){if(e)throw new Error("`inputs`: {inputs}` were passed alongside {input_name} which is not allowed. Make sure to either pass {inputs} or {input_name}=...")}else s[r]=e;return{inputs_tensor:s[r],model_inputs:s,model_input_name:r}}async _prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:e,model_inputs:t,model_input_name:n,generation_config:s}){if(this.sessions.model.inputNames.includes("inputs_embeds")&&!t.inputs_embeds&&"_prepare_inputs_embeds"in this){const{input_ids:e,pixel_values:n,attention_mask:s,...r}=t,o=await this._prepare_inputs_embeds(t);t={...r,...(0,a.pick)(o,["inputs_embeds","attention_mask"])}}let{last_hidden_state:r}=await N(this,t);if(null!==s.guidance_scale&&s.guidance_scale>1)r=(0,u.cat)([r,(0,u.full_like)(r,0)],0),"attention_mask"in t&&(t.attention_mask=(0,u.cat)([t.attention_mask,(0,u.zeros_like)(t.attention_mask)],0));else if(t.decoder_input_ids){const e=z(t.decoder_input_ids).dims[0];if(e!==r.dims[0]){if(1!==r.dims[0])throw new Error(`The encoder outputs have a different batch size (${r.dims[0]}) than the decoder inputs (${e}).`);r=(0,u.cat)(Array.from({length:e},(()=>r)),0)}}return t.encoder_outputs=r,t}_prepare_decoder_input_ids_for_generation({batch_size:e,model_input_name:t,model_kwargs:n,decoder_start_token_id:s,bos_token_id:r,generation_config:o}){let{decoder_input_ids:i,...a}=n;if(i)Array.isArray(i[0])||(i=Array.from({length:e},(()=>i)));else if(s??=r,"musicgen"===this.config.model_type)i=Array.from({length:e*this.config.decoder.num_codebooks},(()=>[s]));else if(Array.isArray(s)){if(s.length!==e)throw new Error(`\`decoder_start_token_id\` expcted to have length ${e} but got ${s.length}`);i=s}else i=Array.from({length:e},(()=>[s]));return i=z(i),n.decoder_attention_mask=(0,u.ones_like)(i),{input_ids:i,model_inputs:a}}async generate({inputs:e=null,generation_config:t=null,logits_processor:n=null,stopping_criteria:s=null,streamer:r=null,...o}){this._validate_model_class(),t=this._prepare_generation_config(t,o);let{inputs_tensor:i,model_inputs:a,model_input_name:l}=this._prepare_model_inputs({inputs:e,model_kwargs:o});const c=this.config.is_encoder_decoder;let d;c&&("encoder_outputs"in a||(a=await this._prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:i,model_inputs:a,model_input_name:l,generation_config:t}))),c?({input_ids:d,model_inputs:a}=this._prepare_decoder_input_ids_for_generation({batch_size:a[l].dims.at(0),model_input_name:l,model_kwargs:a,decoder_start_token_id:t.decoder_start_token_id,bos_token_id:t.bos_token_id,generation_config:t})):d=a[l];let p=d.dims.at(-1);null!==t.max_new_tokens&&(t.max_length=p+t.max_new_tokens);const h=this._get_logits_processor(t,p,n),m=this._get_stopping_criteria(t,s),f=a[l].dims.at(0),g=_.LogitsSampler.getSampler(t),w=new Array(f).fill(0),M=d.tolist();let b;r&&r.put(M);let y={};for(;;){if(a=this.prepare_inputs_for_generation(M,a,t),b=await this.forward(a),t.output_attentions&&t.return_dict_in_generate){const e=this.getAttentions(b);for(const t in e)t in y||(y[t]=[]),y[t].push(e[t])}const e=h(M,b.logits.slice(null,-1,null)),n=[];for(let t=0;t<e.dims.at(0);++t){const s=e[t],r=await g(s);for(const[e,s]of r){const r=BigInt(e);w[t]+=s,M[t].push(r),n.push([r]);break}}r&&r.put(n);if(m(M).every((e=>e)))break;a=this._update_model_kwargs_for_generation({generated_input_ids:n,outputs:b,model_inputs:a,is_encoder_decoder:c})}r&&r.end();const x=this.getPastKeyValues(b,a.past_key_values,!0),k=new u.Tensor("int64",M.flat(),[M.length,M[0].length]);if(t.return_dict_in_generate)return{sequences:k,past_key_values:x,...y};for(const e of Object.values(b))"gpu-buffer"===e.location&&e.dispose();return k}getPastKeyValues(e,t,n=!1){const s=Object.create(null);for(const r in e)if(r.startsWith("present")){const o=r.replace("present","past_key_values"),i=r.includes("encoder");if(s[o]=i&&t?t[o]:e[r],t&&(!i||n)){const e=t[o];"gpu-buffer"===e.location&&e.dispose()}}return s}getAttentions(e){const t={};for(const n of["cross_attentions","encoder_attentions","decoder_attentions"])for(const s in e)s.startsWith(n)&&(n in t||(t[n]=[]),t[n].push(e[s]));return t}addPastKeyValues(e,t){if(t)Object.assign(e,t);else{const t=this.sessions.decoder_model_merged??this.sessions.model,n=t?.config?.kv_cache_dtype??"float32",r="float16"===n?new Uint16Array:[],o=(0,s.getKeyValueShapes)(this.config);for(const t in o)e[t]=new u.Tensor(n,r,o[t])}}async encode_image({pixel_values:e}){const t=(await E(this.sessions.vision_encoder,{pixel_values:e})).image_features;return this.config.num_image_tokens||(console.warn(`The number of image tokens was not set in the model configuration. Setting it to the number of features detected by the vision encoder (${t.dims[1]}).`),this.config.num_image_tokens=t.dims[1]),t}async encode_text({input_ids:e}){return(await E(this.sessions.embed_tokens,{input_ids:e})).inputs_embeds}}class ${}class q extends ${constructor({last_hidden_state:e,hidden_states:t=null,attentions:n=null}){super(),this.last_hidden_state=e,this.hidden_states=t,this.attentions=n}}class U extends G{}class W extends U{}class X extends U{async _call(e){return new ml(await super._call(e))}}class Q extends U{async _call(e){return new pl(await super._call(e))}}class H extends U{async _call(e){return new _l(await super._call(e))}}class Y extends U{async _call(e){return new fl(await super._call(e))}}class J extends G{}class K extends J{}class Z extends G{}class ee extends Z{}class te extends Z{async _call(e){return new ml(await super._call(e))}}class ne extends Z{async _call(e){return new pl(await super._call(e))}}class se extends Z{async _call(e){return new _l(await super._call(e))}}class re extends Z{async _call(e){return new fl(await super._call(e))}}class oe extends G{}class ie extends oe{}class ae extends oe{async _call(e){return new ml(await super._call(e))}}class le extends oe{async _call(e){return new pl(await super._call(e))}}class ce extends oe{async _call(e){return new _l(await super._call(e))}}class de extends oe{async _call(e){return new fl(await super._call(e))}}class ue extends G{}class pe extends ue{}class he extends ue{async _call(e){return new ml(await super._call(e))}}class _e extends ue{async _call(e){return new pl(await super._call(e))}}class me extends ue{async _call(e){return new _l(await super._call(e))}}class fe extends ue{async _call(e){return new fl(await super._call(e))}}class ge extends G{}class we extends ge{}class Me extends ge{async _call(e){return new ml(await super._call(e))}}class be extends ge{async _call(e){return new pl(await super._call(e))}}class ye extends ge{async _call(e){return new _l(await super._call(e))}}class xe extends ge{async _call(e){return new fl(await super._call(e))}}class ke extends G{}class Te extends ke{}class ve extends ke{async _call(e){return new ml(await super._call(e))}}class Fe extends ke{async _call(e){return new pl(await super._call(e))}}class Ce extends ke{async _call(e){return new _l(await super._call(e))}}class Pe extends ke{async _call(e){return new fl(await super._call(e))}}class Se extends G{}class Ae extends Se{}class Ee extends Se{async _call(e){return new ml(await super._call(e))}}class Le extends Se{async _call(e){return new pl(await super._call(e))}}class ze extends Se{async _call(e){return new _l(await super._call(e))}}class Ie extends Se{async _call(e){return new fl(await super._call(e))}}class Be extends G{}class Ne extends Be{}class Oe extends Be{async _call(e){return new pl(await super._call(e))}}class je extends Be{async _call(e){return new _l(await super._call(e))}}class De extends Be{async _call(e){return new fl(await super._call(e))}}class Ve extends Be{async _call(e){return new ml(await super._call(e))}}class Re extends G{}class Ge extends Re{}class $e extends Re{async _call(e){return new ml(await super._call(e))}}class qe extends Re{async _call(e){return new pl(await super._call(e))}}class Ue extends Re{async _call(e){return new _l(await super._call(e))}}class We extends G{}class Xe extends We{}class Qe extends We{async _call(e){return new ml(await super._call(e))}}class He extends We{async _call(e){return new pl(await super._call(e))}}class Ye extends We{async _call(e){return new fl(await super._call(e))}}class Je extends G{}class Ke extends Je{}class Ze extends Je{async _call(e){return new ml(await super._call(e))}}class et extends Je{async _call(e){return new pl(await super._call(e))}}class tt extends Je{async _call(e){return new _l(await super._call(e))}}class nt extends Je{async _call(e){return new fl(await super._call(e))}}class st extends G{}class rt extends st{}class ot extends st{async _call(e){return new ml(await super._call(e))}}class it extends st{async _call(e){return new pl(await super._call(e))}}class at extends st{async _call(e){return new fl(await super._call(e))}}class lt extends G{}class ct extends lt{}class dt extends lt{async _call(e){return new pl(await super._call(e))}}class ut extends lt{async _call(e){return new fl(await super._call(e))}}class pt extends lt{async _call(e){return new ml(await super._call(e))}}class ht extends G{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"]}class _t extends ht{}class mt extends ht{}class ft extends G{}class gt extends ft{}class wt extends ft{}class Mt extends G{}class bt extends Mt{}class yt extends Mt{}class xt extends G{}class kt extends xt{}class Tt extends xt{}class vt extends xt{async _call(e){return new pl(await super._call(e))}}class Ft extends G{}class Ct extends Ft{}class Pt extends Ft{}class St extends Ft{async _call(e){return new pl(await super._call(e))}}class At extends Ft{}class Et extends G{}class Lt extends Et{}class zt extends Et{}class It extends G{}class Bt extends It{}class Nt extends It{}class Ot extends G{}class jt extends Ot{}class Dt extends Ot{async _call(e){return new ml(await super._call(e))}}class Vt extends Ot{async _call(e){return new pl(await super._call(e))}}class Rt extends Ot{async _call(e){return new _l(await super._call(e))}}class Gt extends Ot{async _call(e){return new fl(await super._call(e))}}class $t extends G{}class qt extends $t{}class Ut extends $t{async _call(e){return new ml(await super._call(e))}}class Wt extends $t{async _call(e){return new pl(await super._call(e))}}class Xt extends $t{async _call(e){return new _l(await super._call(e))}}class Qt extends $t{async _call(e){return new fl(await super._call(e))}}class Ht extends G{}class Yt extends Ht{}class Jt extends Ht{async _call(e){return new ml(await super._call(e))}}class Kt extends Ht{async _call(e){return new pl(await super._call(e))}}class Zt extends Ht{async _call(e){return new _l(await super._call(e))}}class en extends Ht{async _call(e){return new fl(await super._call(e))}}class tn extends G{}class nn extends tn{}class sn extends tn{}class rn extends G{requires_attention_mask=!1;main_input_name="input_features";forward_params=["input_features","attention_mask","decoder_input_ids","decoder_attention_mask","past_key_values"]}class on extends rn{}class an extends rn{_prepare_generation_config(e,t){return super._prepare_generation_config(e,t,f.WhisperGenerationConfig)}_retrieve_init_tokens(e){const t=[e.decoder_start_token_id];let n=e.language;const s=e.task;if(e.is_multilingual){n||(console.warn("No language specified - defaulting to English (en)."),n="en");const r=`<|${(0,g.whisper_language_to_code)(n)}|>`;t.push(e.lang_to_id[r]),t.push(e.task_to_id[s??"transcribe"])}else if(n||s)throw new Error("Cannot specify `task` or `language` for an English-only model. If the model is intended to be multilingual, pass `is_multilingual=true` to generate, or update the generation config.");return!e.return_timestamps&&e.no_timestamps_token_id&&t.at(-1)!==e.no_timestamps_token_id?t.push(e.no_timestamps_token_id):e.return_timestamps&&t.at(-1)===e.no_timestamps_token_id&&(console.warn("<|notimestamps|> prompt token is removed from generation_config since `return_timestamps` is set to `true`."),t.pop()),t.filter((e=>null!=e))}async generate({inputs:e=null,generation_config:t=null,logits_processor:n=null,stopping_criteria:s=null,...r}){t=this._prepare_generation_config(t,r);const o=r.decoder_input_ids??this._retrieve_init_tokens(t);if(t.return_timestamps&&(n??=new c.LogitsProcessorList,n.push(new c.WhisperTimeStampLogitsProcessor(t,o))),t.begin_suppress_tokens&&(n??=new c.LogitsProcessorList,n.push(new c.SuppressTokensAtBeginLogitsProcessor(t.begin_suppress_tokens,o.length))),t.return_token_timestamps){if(!t.alignment_heads)throw new Error("Model generation config has no `alignment_heads`, token-level timestamps not available. See https://gist.github.com/hollance/42e32852f24243b748ae6bc1f985b13a on how to add this property to the generation config.");"translate"===t.task&&console.warn("Token-level timestamps may not be reliable for task 'translate'."),t.output_attentions=!0,t.return_dict_in_generate=!0}const i=await super.generate({inputs:e,generation_config:t,logits_processor:n,decoder_input_ids:o,...r});return t.return_token_timestamps&&(i.token_timestamps=this._extract_token_timestamps(i,t.alignment_heads,t.num_frames)),i}_extract_token_timestamps(e,t,n=null,s=.02){if(!e.cross_attentions)throw new Error("Model outputs must contain cross attentions to extract timestamps. This is most likely because the model was not exported with `output_attentions=True`.");null==n&&console.warn("`num_frames` has not been set, meaning the entire audio will be analyzed. This may lead to inaccurate token-level timestamps for short audios (< 30 seconds).");let r=this.config.median_filter_width;void 0===r&&(console.warn("Model config has no `median_filter_width`, using default value of 7."),r=7);const o=e.cross_attentions,i=Array.from({length:this.config.decoder_layers},((e,t)=>(0,u.cat)(o.map((e=>e[t])),2))),l=(0,u.stack)(t.map((([e,t])=>{if(e>=i.length)throw new Error(`Layer index ${e} is out of bounds for cross attentions (length ${i.length}).`);return n?i[e].slice(null,t,null,[0,n]):i[e].slice(null,t)}))).transpose(1,0,2,3),[c,d]=(0,u.std_mean)(l,-2,0,!0),h=l.clone();for(let e=0;e<h.dims[0];++e){const t=h[e];for(let n=0;n<t.dims[0];++n){const s=t[n],o=c[e][n][0].data,i=d[e][n][0].data;for(let e=0;e<s.dims[0];++e){let t=s[e].data;for(let e=0;e<t.length;++e)t[e]=(t[e]-i[e])/o[e];t.set((0,p.medianFilter)(t,r))}}}const _=[(0,u.mean)(h,1)],m=e.sequences.dims,f=new u.Tensor("float32",new Float32Array(m[0]*m[1]),m);for(let e=0;e<m[0];++e){const t=_[e].neg().squeeze_(0),[n,r]=(0,p.dynamic_time_warping)(t.tolist()),o=Array.from({length:n.length-1},((e,t)=>n[t+1]-n[t])),i=(0,a.mergeArrays)([1],o).map((e=>!!e)),l=[];for(let e=0;e<i.length;++e)i[e]&&l.push(r[e]*s);f[e].data.set(l,1)}return f}}class ln extends G{main_input_name="pixel_values";forward_params=["pixel_values","input_ids","encoder_hidden_states","past_key_values"]}class cn extends G{forward_params=["input_ids","pixel_values","attention_mask","position_ids","past_key_values"]}class dn extends cn{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:n,attention_mask:s}){const r=this.config.image_token_index,o=n.tolist().map((e=>e.findIndex((e=>e==r)))),i=o.every((e=>-1===e)),a=o.every((e=>-1!==e));if(!i&&!a)throw new Error("Every input should contain either 0 or 1 image token.");if(i)return{inputs_embeds:e,attention_mask:s};const l=[],c=[];for(let n=0;n<o.length;++n){const r=o[n],i=e[n],a=t[n],d=s[n];l.push((0,u.cat)([i.slice([0,r]),a,i.slice([r+1,i.dims[0]])],0)),c.push((0,u.cat)([d.slice([0,r]),(0,u.ones)([a.dims[0]]),d.slice([r+1,d.dims[0]])],0))}return{inputs_embeds:(0,u.stack)(l,0),attention_mask:(0,u.stack)(c,0)}}}class un extends dn{}class pn extends G{forward_params=["input_ids","inputs_embeds","attention_mask","pixel_values","encoder_outputs","decoder_input_ids","decoder_inputs_embeds","decoder_attention_mask","past_key_values"];main_input_name="inputs_embeds"}class hn extends pn{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:n,attention_mask:s}){return{inputs_embeds:(0,u.cat)([t,e],1),attention_mask:(0,u.cat)([(0,u.ones)(t.dims.slice(0,2)),s],1)}}async _prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:n,attention_mask:s}){if(!e&&!t)throw new Error("Either `input_ids` or `pixel_values` should be provided.");let r,o;return e&&(r=await this.encode_text({input_ids:e})),t&&(o=await this.encode_image({pixel_values:t})),r&&o?({inputs_embeds:n,attention_mask:s}=this._merge_input_ids_with_image_features({inputs_embeds:r,image_features:o,input_ids:e,attention_mask:s})):n=r||o,{inputs_embeds:n,attention_mask:s}}async forward({input_ids:e,pixel_values:t,attention_mask:n,decoder_input_ids:s,decoder_attention_mask:r,encoder_outputs:o,past_key_values:i,inputs_embeds:a,decoder_inputs_embeds:l}){if(a||({inputs_embeds:a,attention_mask:n}=await this._prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:a,attention_mask:n})),!o){let{last_hidden_state:e}=await N(this,{inputs_embeds:a,attention_mask:n});o=e}if(!l){if(!s)throw new Error("Either `decoder_input_ids` or `decoder_inputs_embeds` should be provided.");l=await this.encode_text({input_ids:s})}const c={inputs_embeds:l,attention_mask:r,encoder_attention_mask:n,encoder_hidden_states:o,past_key_values:i};return await O(this,c,!0)}}class _n extends G{}class mn extends _n{}class fn extends _n{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class gn extends _n{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class wn extends _n{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class Mn extends _n{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class bn extends G{}class yn extends bn{}class xn extends bn{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class kn extends _n{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class Tn extends G{}class vn extends Tn{}class Fn extends G{}class Cn extends Fn{}class Pn extends Fn{}class Sn extends G{}class An extends Sn{}class En extends Sn{}class Ln extends G{}class zn extends Ln{}class In extends Ln{}class Bn extends G{}class Nn extends Bn{}class On extends Bn{}class jn extends G{}class Dn extends jn{}class Vn extends jn{}class Rn extends G{}class Gn extends Rn{}class $n extends Rn{}class qn extends G{}class Un extends qn{}class Wn extends qn{}class Xn extends G{}class Qn extends Xn{}class Hn extends Xn{}class Yn extends G{}class Jn extends Yn{}class Kn extends Yn{}class Zn extends G{}class es extends Zn{}class ts extends Zn{}class ns extends G{}class ss extends ns{}class rs extends ns{}class os extends G{}class is extends os{}class as extends os{}class ls extends G{}class cs extends ls{}class ds extends ls{}class us extends G{}class ps extends us{}class hs extends us{}class _s extends G{}class ms extends _s{}class fs extends _s{}class gs extends G{}class ws extends gs{}class Ms extends gs{}class bs extends G{}class ys extends bs{}class xs extends bs{}class ks extends G{}class Ts extends ks{}class vs extends ks{}class Fs extends G{}class Cs extends Fs{}class Ps extends Fs{}class Ss extends G{}class As extends Ss{}class Es extends Ss{}class Ls extends G{}class zs extends Ls{}class Is extends Ls{async _call(e){return new pl(await super._call(e))}}class Bs extends G{}class Ns extends Bs{}class Os extends Bs{async _call(e){return new pl(await super._call(e))}}class js extends G{}class Ds extends js{}class Vs extends G{}class Rs extends Vs{}class Gs extends Vs{async _call(e){return new pl(await super._call(e))}}class $s extends G{}class qs extends $s{}class Us extends G{}class Ws extends Us{}class Xs extends Us{async _call(e){return new pl(await super._call(e))}}class Qs extends G{}class Hs extends Qs{async _call(e){return new Ml(await super._call(e))}}class Ys extends G{}class Js extends Ys{}class Ks extends Ys{async _call(e){return new pl(await super._call(e))}}class Zs extends G{}class er extends Zs{}class tr extends Zs{async _call(e){return new pl(await super._call(e))}}class nr extends G{}class sr extends nr{}class rr extends nr{}class or extends G{}class ir extends or{}class ar extends or{}class lr extends G{}class cr extends lr{}class dr extends lr{async _call(e){return new pl(await super._call(e))}}class ur extends G{}class pr extends ur{}class hr extends ur{async _call(e){return new mr(await super._call(e))}}class _r extends ur{async _call(e){return new fr(await super._call(e))}}class mr extends ${constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class fr extends ${constructor({logits:e,pred_boxes:t,pred_masks:n}){super(),this.logits=e,this.pred_boxes=t,this.pred_masks=n}}class gr extends G{}class wr extends gr{}class Mr extends gr{async _call(e){return new br(await super._call(e))}}class br extends ${constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class yr extends G{}class xr extends yr{}class kr extends yr{async _call(e){return new Tr(await super._call(e))}}class Tr extends mr{}class vr extends G{}class Fr extends vr{}class Cr extends vr{async _call(e){return new pl(await super._call(e))}}class Pr extends G{}class Sr extends Pr{}class Ar extends Pr{async _call(e){return new pl(await super._call(e))}}class Er extends G{}class Lr extends Er{}class zr extends Er{async _call(e){return new pl(await super._call(e))}}class Ir extends G{}class Br extends Ir{}class Nr extends Ir{async _call(e){return new pl(await super._call(e))}}class Or extends G{}class jr extends Or{}class Dr extends Or{}class Vr extends G{}class Rr extends Vr{}class Gr extends Vr{}class $r extends G{}class qr extends $r{}class Ur extends G{}class Wr extends Ur{}class Xr extends Ur{}class Qr extends Ur{}class Hr extends G{}class Yr extends Hr{}class Jr extends G{}class Kr extends Jr{}class Zr extends Jr{}class eo extends G{}class to extends eo{}class no extends eo{}class so extends G{}class ro extends so{}class oo extends G{}class io extends oo{}class ao extends oo{async _call(e){return new pl(await super._call(e))}}class lo extends G{}class co extends lo{}class uo extends lo{async _call(e){return new pl(await super._call(e))}}class po extends G{}class ho extends po{}class _o extends po{async _call(e){return new pl(await super._call(e))}}class mo extends G{}class fo extends mo{}class go extends mo{async _call(e){return new wo(await super._call(e))}}class wo extends ${constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class Mo extends G{}class bo extends Mo{async get_image_embeddings({pixel_values:e}){return await N(this,{pixel_values:e})}async forward(e){if(e.image_embeddings&&e.image_positional_embeddings||(e={...e,...await this.get_image_embeddings(e)}),!e.input_labels&&e.input_points){const t=e.input_points.dims.slice(0,-1),n=t.reduce(((e,t)=>e*t),1);e.input_labels=new u.Tensor("int64",new BigInt64Array(n).fill(1n),t)}const t={image_embeddings:e.image_embeddings,image_positional_embeddings:e.image_positional_embeddings};return e.input_points&&(t.input_points=e.input_points),e.input_labels&&(t.input_labels=e.input_labels),e.input_boxes&&(t.input_boxes=e.input_boxes),await E(this.sessions.prompt_encoder_mask_decoder,t)}async _call(e){return new yo(await super._call(e))}}class yo extends ${constructor({iou_scores:e,pred_masks:t}){super(),this.iou_scores=e,this.pred_masks=t}}class xo extends G{}class ko extends xo{}class To extends xo{}class vo extends G{}class Fo extends vo{}class Co extends vo{}class Po extends G{}class So extends Po{}class Ao extends Po{async _call(e){return new gl(await super._call(e))}}class Eo extends Po{async _call(e){return new pl(await super._call(e))}}class Lo extends Po{async _call(e){return new _l(await super._call(e))}}class zo extends G{}class Io extends zo{}class Bo extends zo{async _call(e){return new _l(await super._call(e))}}class No extends G{}class Oo extends No{}class jo extends G{}class Do extends jo{}class Vo extends jo{async _call(e){return new gl(await super._call(e))}}class Ro extends jo{async _call(e){return new pl(await super._call(e))}}class Go extends G{}class $o extends Go{}class qo extends Go{async _call(e){return new gl(await super._call(e))}}class Uo extends Go{async _call(e){return new pl(await super._call(e))}}class Wo extends Go{async _call(e){return new _l(await super._call(e))}}class Xo extends G{}class Qo extends Xo{}class Ho extends Xo{async _call(e){return new gl(await super._call(e))}}class Yo extends Xo{async _call(e){return new pl(await super._call(e))}}class Jo extends G{}class Ko extends Po{}class Zo extends Po{async _call(e){return new gl(await super._call(e))}}class ei extends Po{async _call(e){return new pl(await super._call(e))}}class ti extends G{}class ni extends ti{}class si extends ti{async _call(e){return new gl(await super._call(e))}}class ri extends ti{async _call(e){return new pl(await super._call(e))}}class oi extends ti{async _call(e){return new hl(await super._call(e))}}class ii extends ti{async _call(e){return new _l(await super._call(e))}}class ai extends G{}class li extends ai{}class ci extends ai{}class di extends ai{async generate_speech(e,t,{threshold:n=.5,minlenratio:s=0,maxlenratio:r=20,vocoder:o=null}={}){const i={input_ids:e},{encoder_outputs:a,encoder_attention_mask:l}=await N(this,i),c=a.dims[1]/this.config.reduction_factor,d=Math.floor(c*r),p=Math.floor(c*s),h=this.config.num_mel_bins;let _=[],m=null,f=null,g=0;for(;;){++g;const e=I(!!f);let s;s=f?f.output_sequence_out:new u.Tensor("float32",new Float32Array(h),[1,1,h]);let r={use_cache_branch:e,output_sequence:s,encoder_attention_mask:l,speaker_embeddings:t,encoder_hidden_states:a};this.addPastKeyValues(r,m),f=await E(this.sessions.decoder_model_merged,r),m=this.getPastKeyValues(f,m);const{prob:o,spectrum:i}=f;if(_.push(i),g>=p&&(Array.from(o.data).filter((e=>e>=n)).length>0||g>=d))break}const w=(0,u.cat)(_),{waveform:M}=await E(o.sessions.model,{spectrogram:w});return{spectrogram:w,waveform:M}}}class ui extends G{main_input_name="spectrogram"}class pi extends G{}class hi extends pi{}class _i extends G{}class mi extends _i{}class fi extends _i{}class gi extends G{}class wi extends gi{}class Mi extends gi{}class bi extends G{}class yi extends bi{}class xi extends bi{}class ki extends G{}class Ti extends ki{}class vi extends ki{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class Fi extends ki{static async from_pretrained(e,t={}){return t.model_file_name??="audio_model",super.from_pretrained(e,t)}}class Ci extends G{}class Pi extends Ci{async _call(e){return new bl(await super._call(e))}}class Si extends G{}class Ai extends Si{}class Ei extends Si{}class Li extends Si{}class zi extends G{}class Ii extends zi{}class Bi extends zi{}class Ni extends G{}class Oi extends Ni{}class ji extends Ni{async _call(e){return new pl(await super._call(e))}}class Di extends G{}class Vi extends Di{}class Ri extends Di{}class Gi extends G{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"];_apply_and_filter_by_delay_pattern_mask(e){const[t,n]=e.dims,s=this.config.decoder.num_codebooks,r=n-s;let o=0;for(let t=0;t<e.size;++t){if(e.data[t]===this.config.decoder.pad_token_id)continue;const i=t%n-Math.floor(t/n)%s;i>0&&i<=r&&(e.data[o++]=e.data[t])}const i=Math.floor(t/s),a=o/(i*s);return new u.Tensor(e.type,e.data.slice(0,o),[i,s,a])}prepare_inputs_for_generation(e,t,n){let s=structuredClone(e);for(let e=0;e<s.length;++e)for(let t=0;t<s[e].length;++t)e%this.config.decoder.num_codebooks>=t&&(s[e][t]=BigInt(this.config.decoder.pad_token_id));null!==n.guidance_scale&&n.guidance_scale>1&&(s=s.concat(s));return super.prepare_inputs_for_generation(s,t,n)}async generate(e){const t=await super.generate(e),n=this._apply_and_filter_by_delay_pattern_mask(t).unsqueeze_(0),{audio_values:s}=await E(this.sessions.encodec_decode,{audio_codes:n});return s}}class $i extends G{}class qi extends $i{}class Ui extends $i{async _call(e){return new pl(await super._call(e))}}class Wi extends G{}class Xi extends Wi{}class Qi extends Wi{async _call(e){return new pl(await super._call(e))}}class Hi extends G{}class Yi extends Hi{}class Ji extends Hi{async _call(e){return new pl(await super._call(e))}}class Ki extends G{}class Zi extends Ki{}class ea extends Ki{async _call(e){return new pl(await super._call(e))}}class ta extends G{}class na extends ta{}class sa{static MODEL_CLASS_MAPPINGS=null;static BASE_IF_FAIL=!1;static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:r=null,local_files_only:o=!1,revision:i="main",model_file_name:a=null,subfolder:l="onnx",device:c=null,dtype:d=null,use_external_data_format:u=null,session_options:p={}}={}){const h={progress_callback:t,config:n,cache_dir:r,local_files_only:o,revision:i,model_file_name:a,subfolder:l,device:c,dtype:d,use_external_data_format:u,session_options:p};if(h.config=await s.AutoConfig.from_pretrained(e,h),!this.MODEL_CLASS_MAPPINGS)throw new Error("`MODEL_CLASS_MAPPINGS` not implemented for this type of `AutoClass`: "+this.name);for(const t of this.MODEL_CLASS_MAPPINGS){const n=t.get(h.config.model_type);if(n)return await n[1].from_pretrained(e,h)}if(this.BASE_IF_FAIL)return console.warn(`Unknown model class "${h.config.model_type}", attempting to construct from base class.`),await G.from_pretrained(e,h);throw Error(`Unsupported model type: ${h.config.model_type}`)}}const ra=new Map([["bert",["BertModel",W]],["nomic_bert",["NomicBertModel",K]],["roformer",["RoFormerModel",ee]],["electra",["ElectraModel",pe]],["esm",["EsmModel",Ge]],["convbert",["ConvBertModel",ie]],["camembert",["CamembertModel",we]],["deberta",["DebertaModel",Te]],["deberta-v2",["DebertaV2Model",Ae]],["mpnet",["MPNetModel",Ke]],["albert",["AlbertModel",ct]],["distilbert",["DistilBertModel",Ne]],["roberta",["RobertaModel",jt]],["xlm",["XLMModel",qt]],["xlm-roberta",["XLMRobertaModel",Yt]],["clap",["ClapModel",Ti]],["clip",["CLIPModel",mn]],["clipseg",["CLIPSegModel",Cn]],["chinese_clip",["ChineseCLIPModel",vn]],["siglip",["SiglipModel",yn]],["mobilebert",["MobileBertModel",Xe]],["squeezebert",["SqueezeBertModel",rt]],["wav2vec2",["Wav2Vec2Model",So]],["wav2vec2-bert",["Wav2Vec2BertModel",Qo]],["unispeech",["UniSpeechModel",Do]],["unispeech-sat",["UniSpeechSatModel",$o]],["hubert",["HubertModel",Ko]],["wavlm",["WavLMModel",ni]],["audio-spectrogram-transformer",["ASTModel",nn]],["vits",["VitsModel",Pi]],["pyannote",["PyAnnoteModel",Io]],["wespeaker-resnet",["WeSpeakerResNetModel",Oo]],["detr",["DetrModel",pr]],["rt_detr",["RTDetrModel",wr]],["table-transformer",["TableTransformerModel",xr]],["vit",["ViTModel",zs]],["pvt",["PvtModel",Ns]],["vit_msn",["ViTMSNModel",Rs]],["vit_mae",["ViTMAEModel",Ds]],["groupvit",["GroupViTModel",qs]],["fastvit",["FastViTModel",Ws]],["mobilevit",["MobileViTModel",Js]],["mobilevitv2",["MobileViTV2Model",er]],["owlvit",["OwlViTModel",sr]],["owlv2",["Owlv2Model",ir]],["beit",["BeitModel",cr]],["deit",["DeiTModel",Fr]],["hiera",["HieraModel",Sr]],["convnext",["ConvNextModel",io]],["convnextv2",["ConvNextV2Model",co]],["dinov2",["Dinov2Model",ho]],["resnet",["ResNetModel",Lr]],["swin",["SwinModel",Br]],["swin2sr",["Swin2SRModel",jr]],["donut-swin",["DonutSwinModel",ro]],["yolos",["YolosModel",fo]],["dpt",["DPTModel",Rr]],["glpn",["GLPNModel",to]],["hifigan",["SpeechT5HifiGan",ui]],["efficientnet",["EfficientNetModel",Oi]],["decision_transformer",["DecisionTransformerModel",na]],["mobilenet_v1",["MobileNetV1Model",qi]],["mobilenet_v2",["MobileNetV2Model",Xi]],["mobilenet_v3",["MobileNetV3Model",Yi]],["mobilenet_v4",["MobileNetV4Model",Zi]],["maskformer",["MaskFormerModel",Kr]]]),oa=new Map([["t5",["T5Model",_t]],["longt5",["LongT5Model",gt]],["mt5",["MT5Model",bt]],["bart",["BartModel",kt]],["mbart",["MBartModel",Ct]],["marian",["MarianModel",ko]],["whisper",["WhisperModel",on]],["m2m_100",["M2M100Model",Fo]],["blenderbot",["BlenderbotModel",Lt]],["blenderbot-small",["BlenderbotSmallModel",Bt]]]),ia=new Map([["bloom",["BloomModel",Ts]],["jais",["JAISModel",zn]],["gpt2",["GPT2Model",An]],["gptj",["GPTJModel",Gn]],["gpt_bigcode",["GPTBigCodeModel",Un]],["gpt_neo",["GPTNeoModel",Nn]],["gpt_neox",["GPTNeoXModel",Dn]],["codegen",["CodeGenModel",Qn]],["llama",["LlamaModel",Jn]],["granite",["GraniteModel",es]],["cohere",["CohereModel",ss]],["gemma",["GemmaModel",is]],["gemma2",["Gemma2Model",cs]],["openelm",["OpenELMModel",ps]],["qwen2",["Qwen2Model",ms]],["phi",["PhiModel",ws]],["phi3",["Phi3Model",ys]],["mpt",["MptModel",Cs]],["opt",["OPTModel",As]],["mistral",["MistralModel",mi]],["starcoder2",["Starcoder2Model",wi]],["falcon",["FalconModel",yi]],["stablelm",["StableLmModel",Ii]]]),aa=new Map([["speecht5",["SpeechT5ForSpeechToText",ci]],["whisper",["WhisperForConditionalGeneration",an]]]),la=new Map([["speecht5",["SpeechT5ForTextToSpeech",di]]]),ca=new Map([["vits",["VitsModel",Pi]],["musicgen",["MusicgenForConditionalGeneration",Gi]]]),da=new Map([["bert",["BertForSequenceClassification",Q]],["roformer",["RoFormerForSequenceClassification",ne]],["electra",["ElectraForSequenceClassification",_e]],["esm",["EsmForSequenceClassification",qe]],["convbert",["ConvBertForSequenceClassification",le]],["camembert",["CamembertForSequenceClassification",be]],["deberta",["DebertaForSequenceClassification",Fe]],["deberta-v2",["DebertaV2ForSequenceClassification",Le]],["mpnet",["MPNetForSequenceClassification",et]],["albert",["AlbertForSequenceClassification",dt]],["distilbert",["DistilBertForSequenceClassification",Oe]],["roberta",["RobertaForSequenceClassification",Vt]],["xlm",["XLMForSequenceClassification",Wt]],["xlm-roberta",["XLMRobertaForSequenceClassification",Kt]],["bart",["BartForSequenceClassification",vt]],["mbart",["MBartForSequenceClassification",St]],["mobilebert",["MobileBertForSequenceClassification",He]],["squeezebert",["SqueezeBertForSequenceClassification",it]]]),ua=new Map([["bert",["BertForTokenClassification",H]],["roformer",["RoFormerForTokenClassification",se]],["electra",["ElectraForTokenClassification",me]],["esm",["EsmForTokenClassification",Ue]],["convbert",["ConvBertForTokenClassification",ce]],["camembert",["CamembertForTokenClassification",ye]],["deberta",["DebertaForTokenClassification",Ce]],["deberta-v2",["DebertaV2ForTokenClassification",ze]],["mpnet",["MPNetForTokenClassification",tt]],["distilbert",["DistilBertForTokenClassification",je]],["roberta",["RobertaForTokenClassification",Rt]],["xlm",["XLMForTokenClassification",Xt]],["xlm-roberta",["XLMRobertaForTokenClassification",Zt]]]),pa=new Map([["t5",["T5ForConditionalGeneration",mt]],["longt5",["LongT5ForConditionalGeneration",wt]],["mt5",["MT5ForConditionalGeneration",yt]],["bart",["BartForConditionalGeneration",Tt]],["mbart",["MBartForConditionalGeneration",Pt]],["marian",["MarianMTModel",To]],["m2m_100",["M2M100ForConditionalGeneration",Co]],["blenderbot",["BlenderbotForConditionalGeneration",zt]],["blenderbot-small",["BlenderbotSmallForConditionalGeneration",Nt]]]),ha=new Map([["bloom",["BloomForCausalLM",vs]],["gpt2",["GPT2LMHeadModel",En]],["jais",["JAISLMHeadModel",In]],["gptj",["GPTJForCausalLM",$n]],["gpt_bigcode",["GPTBigCodeForCausalLM",Wn]],["gpt_neo",["GPTNeoForCausalLM",On]],["gpt_neox",["GPTNeoXForCausalLM",Vn]],["codegen",["CodeGenForCausalLM",Hn]],["llama",["LlamaForCausalLM",Kn]],["granite",["GraniteForCausalLM",ts]],["cohere",["CohereForCausalLM",rs]],["gemma",["GemmaForCausalLM",as]],["gemma2",["Gemma2ForCausalLM",ds]],["openelm",["OpenELMForCausalLM",hs]],["qwen2",["Qwen2ForCausalLM",fs]],["phi",["PhiForCausalLM",Ms]],["phi3",["Phi3ForCausalLM",xs]],["mpt",["MptForCausalLM",Ps]],["opt",["OPTForCausalLM",Es]],["mbart",["MBartForCausalLM",At]],["mistral",["MistralForCausalLM",fi]],["starcoder2",["Starcoder2ForCausalLM",Mi]],["falcon",["FalconForCausalLM",xi]],["trocr",["TrOCRForCausalLM",hi]],["stablelm",["StableLmForCausalLM",Bi]]]),_a=new Map([["bert",["BertForMaskedLM",X]],["roformer",["RoFormerForMaskedLM",te]],["electra",["ElectraForMaskedLM",he]],["esm",["EsmForMaskedLM",$e]],["convbert",["ConvBertForMaskedLM",ae]],["camembert",["CamembertForMaskedLM",Me]],["deberta",["DebertaForMaskedLM",ve]],["deberta-v2",["DebertaV2ForMaskedLM",Ee]],["mpnet",["MPNetForMaskedLM",Ze]],["albert",["AlbertForMaskedLM",pt]],["distilbert",["DistilBertForMaskedLM",Ve]],["roberta",["RobertaForMaskedLM",Dt]],["xlm",["XLMWithLMHeadModel",Ut]],["xlm-roberta",["XLMRobertaForMaskedLM",Jt]],["mobilebert",["MobileBertForMaskedLM",Qe]],["squeezebert",["SqueezeBertForMaskedLM",ot]]]),ma=new Map([["bert",["BertForQuestionAnswering",Y]],["roformer",["RoFormerForQuestionAnswering",re]],["electra",["ElectraForQuestionAnswering",fe]],["convbert",["ConvBertForQuestionAnswering",de]],["camembert",["CamembertForQuestionAnswering",xe]],["deberta",["DebertaForQuestionAnswering",Pe]],["deberta-v2",["DebertaV2ForQuestionAnswering",Ie]],["mpnet",["MPNetForQuestionAnswering",nt]],["albert",["AlbertForQuestionAnswering",ut]],["distilbert",["DistilBertForQuestionAnswering",De]],["roberta",["RobertaForQuestionAnswering",Gt]],["xlm",["XLMForQuestionAnswering",Qt]],["xlm-roberta",["XLMRobertaForQuestionAnswering",en]],["mobilebert",["MobileBertForQuestionAnswering",Ye]],["squeezebert",["SqueezeBertForQuestionAnswering",at]]]),fa=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",ln]]]),ga=new Map([["llava",["LlavaForConditionalGeneration",dn]],["moondream1",["Moondream1ForConditionalGeneration",un]],["florence2",["Florence2ForConditionalGeneration",hn]]]),wa=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",ln]]]),Ma=new Map([["vit",["ViTForImageClassification",Is]],["pvt",["PvtForImageClassification",Os]],["vit_msn",["ViTMSNForImageClassification",Gs]],["fastvit",["FastViTForImageClassification",Xs]],["mobilevit",["MobileViTForImageClassification",Ks]],["mobilevitv2",["MobileViTV2ForImageClassification",tr]],["beit",["BeitForImageClassification",dr]],["deit",["DeiTForImageClassification",Cr]],["hiera",["HieraForImageClassification",Ar]],["convnext",["ConvNextForImageClassification",ao]],["convnextv2",["ConvNextV2ForImageClassification",uo]],["dinov2",["Dinov2ForImageClassification",_o]],["resnet",["ResNetForImageClassification",zr]],["swin",["SwinForImageClassification",Nr]],["segformer",["SegformerForImageClassification",Ei]],["efficientnet",["EfficientNetForImageClassification",ji]],["mobilenet_v1",["MobileNetV1ForImageClassification",Ui]],["mobilenet_v2",["MobileNetV2ForImageClassification",Qi]],["mobilenet_v3",["MobileNetV3ForImageClassification",Ji]],["mobilenet_v4",["MobileNetV4ForImageClassification",ea]]]),ba=new Map([["detr",["DetrForObjectDetection",hr]],["rt_detr",["RTDetrForObjectDetection",Mr]],["table-transformer",["TableTransformerForObjectDetection",kr]],["yolos",["YolosForObjectDetection",go]]]),ya=new Map([["owlvit",["OwlViTForObjectDetection",rr]],["owlv2",["Owlv2ForObjectDetection",ar]]]),xa=new Map([["detr",["DetrForSegmentation",_r]],["clipseg",["CLIPSegForImageSegmentation",Pn]]]),ka=new Map([["segformer",["SegformerForSemanticSegmentation",Li]],["sapiens",["SapiensForSemanticSegmentation",Wr]]]),Ta=new Map([["detr",["DetrForSegmentation",_r]],["maskformer",["MaskFormerForInstanceSegmentation",Zr]]]),va=new Map([["sam",["SamModel",bo]]]),Fa=new Map([["wav2vec2",["Wav2Vec2ForCTC",Ao]],["wav2vec2-bert",["Wav2Vec2BertForCTC",Ho]],["unispeech",["UniSpeechForCTC",Vo]],["unispeech-sat",["UniSpeechSatForCTC",qo]],["wavlm",["WavLMForCTC",si]],["hubert",["HubertForCTC",Zo]]]),Ca=new Map([["wav2vec2",["Wav2Vec2ForSequenceClassification",Eo]],["wav2vec2-bert",["Wav2Vec2BertForSequenceClassification",Yo]],["unispeech",["UniSpeechForSequenceClassification",Ro]],["unispeech-sat",["UniSpeechSatForSequenceClassification",Uo]],["wavlm",["WavLMForSequenceClassification",ri]],["hubert",["HubertForSequenceClassification",ei]],["audio-spectrogram-transformer",["ASTForAudioClassification",sn]]]),Pa=new Map([["wavlm",["WavLMForXVector",oi]]]),Sa=new Map([["unispeech-sat",["UniSpeechSatForAudioFrameClassification",Wo]],["wavlm",["WavLMForAudioFrameClassification",ii]],["wav2vec2",["Wav2Vec2ForAudioFrameClassification",Lo]],["pyannote",["PyAnnoteForAudioFrameClassification",Bo]]]),Aa=new Map([["vitmatte",["VitMatteForImageMatting",Hs]]]),Ea=new Map([["swin2sr",["Swin2SRForImageSuperResolution",Dr]]]),La=new Map([["dpt",["DPTForDepthEstimation",Gr]],["depth_anything",["DepthAnythingForDepthEstimation",qr]],["glpn",["GLPNForDepthEstimation",no]],["sapiens",["SapiensForDepthEstimation",Xr]],["depth_pro",["DepthProForDepthEstimation",Yr]]]),za=new Map([["sapiens",["SapiensForNormalEstimation",Qr]]]),Ia=new Map([["clip",["CLIPVisionModelWithProjection",Mn]],["siglip",["SiglipVisionModel",kn]]]),Ba=[[ra,w],[oa,M],[ia,x],[da,w],[ua,w],[pa,b],[aa,b],[ha,x],[_a,w],[ma,w],[fa,y],[ga,T],[Ma,w],[xa,w],[Ta,w],[ka,w],[Aa,w],[Ea,w],[La,w],[za,w],[ba,w],[ya,w],[va,k],[Fa,w],[Ca,w],[la,b],[ca,w],[Pa,w],[Sa,w],[Ia,w]];for(const[e,t]of Ba)for(const[n,s]of e.values())F.set(n,t),P.set(s,n),C.set(n,s);const Na=[["MusicgenForConditionalGeneration",Gi,v],["CLIPTextModelWithProjection",gn,w],["SiglipTextModel",xn,w],["ClapTextModelWithProjection",vi,w],["ClapAudioModelWithProjection",Fi,w]];for(const[e,t,n]of Na)F.set(e,n),P.set(t,e),C.set(e,t);class Oa extends sa{static MODEL_CLASS_MAPPINGS=Ba.map((e=>e[0]));static BASE_IF_FAIL=!0}class ja extends sa{static MODEL_CLASS_MAPPINGS=[da]}class Da extends sa{static MODEL_CLASS_MAPPINGS=[ua]}class Va extends sa{static MODEL_CLASS_MAPPINGS=[pa]}class Ra extends sa{static MODEL_CLASS_MAPPINGS=[aa]}class Ga extends sa{static MODEL_CLASS_MAPPINGS=[la]}class $a extends sa{static MODEL_CLASS_MAPPINGS=[ca]}class qa extends sa{static MODEL_CLASS_MAPPINGS=[ha]}class Ua extends sa{static MODEL_CLASS_MAPPINGS=[_a]}class Wa extends sa{static MODEL_CLASS_MAPPINGS=[ma]}class Xa extends sa{static MODEL_CLASS_MAPPINGS=[fa]}class Qa extends sa{static MODEL_CLASS_MAPPINGS=[Ma]}class Ha extends sa{static MODEL_CLASS_MAPPINGS=[xa]}class Ya extends sa{static MODEL_CLASS_MAPPINGS=[ka]}class Ja extends sa{static MODEL_CLASS_MAPPINGS=[Ta]}class Ka extends sa{static MODEL_CLASS_MAPPINGS=[ba]}class Za extends sa{static MODEL_CLASS_MAPPINGS=[ya]}class el extends sa{static MODEL_CLASS_MAPPINGS=[va]}class tl extends sa{static MODEL_CLASS_MAPPINGS=[Fa]}class nl extends sa{static MODEL_CLASS_MAPPINGS=[Ca]}class sl extends sa{static MODEL_CLASS_MAPPINGS=[Pa]}class rl extends sa{static MODEL_CLASS_MAPPINGS=[Sa]}class ol extends sa{static MODEL_CLASS_MAPPINGS=[wa]}class il extends sa{static MODEL_CLASS_MAPPINGS=[Aa]}class al extends sa{static MODEL_CLASS_MAPPINGS=[Ea]}class ll extends sa{static MODEL_CLASS_MAPPINGS=[La]}class cl extends sa{static MODEL_CLASS_MAPPINGS=[za]}class dl extends sa{static MODEL_CLASS_MAPPINGS=[Ia]}class ul extends ${constructor({logits:e,past_key_values:t,encoder_outputs:n,decoder_attentions:s=null,cross_attentions:r=null}){super(),this.logits=e,this.past_key_values=t,this.encoder_outputs=n,this.decoder_attentions=s,this.cross_attentions=r}}class pl extends ${constructor({logits:e}){super(),this.logits=e}}class hl extends ${constructor({logits:e,embeddings:t}){super(),this.logits=e,this.embeddings=t}}class _l extends ${constructor({logits:e}){super(),this.logits=e}}class ml extends ${constructor({logits:e}){super(),this.logits=e}}class fl extends ${constructor({start_logits:e,end_logits:t}){super(),this.start_logits=e,this.end_logits=t}}class gl extends ${constructor({logits:e}){super(),this.logits=e}}class wl extends ${constructor({logits:e,past_key_values:t}){super(),this.logits=e,this.past_key_values=t}}class Ml extends ${constructor({alphas:e}){super(),this.alphas=e}}class bl extends ${constructor({waveform:e,spectrogram:t}){super(),this.waveform=e,this.spectrogram=t}}},"./src/models/whisper/common_whisper.js":
 /*!**********************************************!*\
   !*** ./src/models/whisper/common_whisper.js ***!
   \**********************************************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{WHISPER_LANGUAGE_MAPPING:()=>r,WHISPER_TO_LANGUAGE_CODE_MAPPING:()=>o,whisper_language_to_code:()=>i});const s=[["en","english"],["zh","chinese"],["de","german"],["es","spanish"],["ru","russian"],["ko","korean"],["fr","french"],["ja","japanese"],["pt","portuguese"],["tr","turkish"],["pl","polish"],["ca","catalan"],["nl","dutch"],["ar","arabic"],["sv","swedish"],["it","italian"],["id","indonesian"],["hi","hindi"],["fi","finnish"],["vi","vietnamese"],["he","hebrew"],["uk","ukrainian"],["el","greek"],["ms","malay"],["cs","czech"],["ro","romanian"],["da","danish"],["hu","hungarian"],["ta","tamil"],["no","norwegian"],["th","thai"],["ur","urdu"],["hr","croatian"],["bg","bulgarian"],["lt","lithuanian"],["la","latin"],["mi","maori"],["ml","malayalam"],["cy","welsh"],["sk","slovak"],["te","telugu"],["fa","persian"],["lv","latvian"],["bn","bengali"],["sr","serbian"],["az","azerbaijani"],["sl","slovenian"],["kn","kannada"],["et","estonian"],["mk","macedonian"],["br","breton"],["eu","basque"],["is","icelandic"],["hy","armenian"],["ne","nepali"],["mn","mongolian"],["bs","bosnian"],["kk","kazakh"],["sq","albanian"],["sw","swahili"],["gl","galician"],["mr","marathi"],["pa","punjabi"],["si","sinhala"],["km","khmer"],["sn","shona"],["yo","yoruba"],["so","somali"],["af","afrikaans"],["oc","occitan"],["ka","georgian"],["be","belarusian"],["tg","tajik"],["sd","sindhi"],["gu","gujarati"],["am","amharic"],["yi","yiddish"],["lo","lao"],["uz","uzbek"],["fo","faroese"],["ht","haitian creole"],["ps","pashto"],["tk","turkmen"],["nn","nynorsk"],["mt","maltese"],["sa","sanskrit"],["lb","luxembourgish"],["my","myanmar"],["bo","tibetan"],["tl","tagalog"],["mg","malagasy"],["as","assamese"],["tt","tatar"],["haw","hawaiian"],["ln","lingala"],["ha","hausa"],["ba","bashkir"],["jw","javanese"],["su","sundanese"]],r=new Map(s),o=new Map([...s.map((([e,t])=>[t,e])),["burmese","my"],["valencian","ca"],["flemish","nl"],["haitian","ht"],["letzeburgesch","lb"],["pushto","ps"],["panjabi","pa"],["moldavian","ro"],["moldovan","ro"],["sinhalese","si"],["castilian","es"]]);function i(e){e=e.toLowerCase();let t=o.get(e);if(void 0===t){if(!r.has(e)){const t=2===e.length?r.keys():r.values();throw new Error(`Language "${e}" is not supported. Must be one of: ${JSON.stringify(t)}`)}t=e}return t}},"./src/models/whisper/generation_whisper.js":