@huggingface/transformers 3.0.0-alpha.15 → 3.0.0-alpha.16

package/dist/transformers.min.js

@@ -151,10 +151,10 @@ var r,i,a,s,o,l,u,d,c,p,h,m,f,g,_,w,y,b,v,x,M,T,k,$,S,C,P,E,F,A,I,z,O,B=Object.d
   \******************************/(e,t,n)=>{var r;n.r(t),n.d(t,{Tensor:()=>o.Tensor,createInferenceSession:()=>f,deviceToExecutionProviders:()=>h,isONNXProxy:()=>w,isONNXTensor:()=>g});var i=n(/*! ../env.js */"./src/env.js"),a=n(/*! onnxruntime-node */"?2ce3"),s=n(/*! #onnxruntime-webgpu */"./node_modules/onnxruntime-web/dist/ort.webgpu.bundle.min.mjs"),o=n(/*! onnxruntime-common */"./node_modules/onnxruntime-common/dist/esm/index.js");const l=Object.freeze({auto:null,gpu:null,cpu:"cpu",wasm:"wasm",webgpu:"webgpu",cuda:"cuda",dml:"dml",webnn:{name:"webnn",deviceType:"cpu"},"webnn-npu":{name:"webnn",deviceType:"npu"},"webnn-gpu":{name:"webnn",deviceType:"gpu"},"webnn-cpu":{name:"webnn",deviceType:"cpu"}}),u=[];let d,c;if(i.apis.IS_NODE_ENV){switch(c=a??(r||(r=n.t(a,2))),process.platform){case"win32":u.push("dml");break;case"linux":"x64"===process.arch&&u.push("cuda")}u.push("cpu"),d=["cpu"]}else c=s,i.apis.IS_WEBNN_AVAILABLE&&u.push("webnn-npu","webnn-gpu","webnn-cpu","webnn"),i.apis.IS_WEBGPU_AVAILABLE&&u.push("webgpu"),u.push("wasm"),d=["wasm"];const p=c.InferenceSession;function h(e=null){if(!e)return d;switch(e){case"auto":return u;case"gpu":return u.filter((e=>["webgpu","cuda","dml","webnn-gpu"].includes(e)))}if(u.includes(e))return[l[e]??e];throw new Error(`Unsupported device: "${e}". Should be one of: ${u.join(", ")}.`)}let m=null;async function f(e,t){m&&await m;const n=p.create(e,t);return m??=n,await n}function g(e){return e instanceof c.Tensor}const _=c?.env;function w(){return _?.wasm?.proxy}_?.wasm&&(_.wasm.wasmPaths=`https://cdn.jsdelivr.net/npm/@huggingface/transformers@${i.env.version}/dist/`,_.wasm.proxy=!1,"undefined"!=typeof crossOriginIsolated&&crossOriginIsolated||(_.wasm.numThreads=1)),_?.webgpu&&(_.webgpu.powerPreference="high-performance"),i.env.backends.onnx=_},"./src/configs.js":
 /*!************************!*\
   !*** ./src/configs.js ***!
-  \************************/(e,t,n)=>{n.r(t),n.d(t,{AutoConfig:()=>l,PretrainedConfig:()=>o,getKeyValueShapes:()=>s});var r=n(/*! ./utils/core.js */"./src/utils/core.js"),i=n(/*! ./utils/hub.js */"./src/utils/hub.js");function a(e){const t={};let n={};switch(e.model_type){case"llava":case"paligemma":case"florence2":n=a(e.text_config);break;case"moondream1":n=a(e.phi_config);break;case"musicgen":n=a(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"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 i=a(e.decoder),s="num_decoder_layers"in i,o=(0,r.pick)(e,["model_type","is_encoder_decoder"]);return s?(o.num_decoder_layers=i.num_decoder_layers,o.num_decoder_heads=i.num_decoder_heads,o.decoder_hidden_size=i.decoder_hidden_size,o.num_encoder_layers=i.num_encoder_layers,o.num_encoder_heads=i.num_encoder_heads,o.encoder_hidden_size=i.encoder_hidden_size):(o.num_layers=i.num_layers,o.num_heads=i.num_heads,o.hidden_size=i.hidden_size),o}const i={...n,...(0,r.pick)(e,["model_type","multi_query","is_encoder_decoder"])};for(const n in t)i[n]=e[t[n]];return i}function s(e,{prefix:t="past_key_values"}={}){const n={},r=e.normalized_config;if(r.is_encoder_decoder&&"num_encoder_heads"in r&&"num_decoder_heads"in r){const e=r.encoder_dim_kv??r.encoder_hidden_size/r.num_encoder_heads,i=r.decoder_dim_kv??r.decoder_hidden_size/r.num_decoder_heads,a=[1,r.num_encoder_heads,0,e],s=[1,r.num_decoder_heads,0,i];for(let e=0;e<r.num_decoder_layers;++e)n[`${t}.${e}.encoder.key`]=a,n[`${t}.${e}.encoder.value`]=a,n[`${t}.${e}.decoder.key`]=s,n[`${t}.${e}.decoder.value`]=s}else{const e=r.num_heads,i=r.num_layers,a=r.dim_kv??r.hidden_size/(r.num_attention_heads??e);if("falcon"===r.model_type){const r=[1*e,0,a];for(let e=0;e<i;++e)n[`${t}.${e}.key`]=r,n[`${t}.${e}.value`]=r}else if(r.multi_query){const r=[1*e,0,2*a];for(let e=0;e<i;++e)n[`${t}.${e}.key_value`]=r}else if("bloom"===r.model_type){const r=[1*e,a,0],s=[1*e,0,a];for(let e=0;e<i;++e)n[`${t}.${e}.key`]=r,n[`${t}.${e}.value`]=s}else if("openelm"===r.model_type)for(let r=0;r<i;++r){const i=[1,e[r],0,a];n[`${t}.${r}.key`]=i,n[`${t}.${r}.value`]=i}else{const r=[1,e,0,a];for(let e=0;e<i;++e)n[`${t}.${e}.key`]=r,n[`${t}.${e}.value`]=r}}return n}class o{max_position_embeddings;constructor(e){this.model_type=null,this.is_encoder_decoder=!1,Object.assign(this,e),this.normalized_config=a(this)}static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:r=null,local_files_only:a=!1,revision:s="main"}={}){!n||n instanceof o||(n=new o(n));const l=n??await async function(e,t){return await(0,i.getModelJSON)(e,"config.json",!0,t)}(e,{progress_callback:t,config:n,cache_dir:r,local_files_only:a,revision:s});return new this(l)}}class l{static async from_pretrained(...e){return o.from_pretrained(...e)}}},"./src/env.js":
+  \************************/(e,t,n)=>{n.r(t),n.d(t,{AutoConfig:()=>l,PretrainedConfig:()=>o,getKeyValueShapes:()=>s});var r=n(/*! ./utils/core.js */"./src/utils/core.js"),i=n(/*! ./utils/hub.js */"./src/utils/hub.js");function a(e){const t={};let n={};switch(e.model_type){case"llava":case"paligemma":case"florence2":n=a(e.text_config);break;case"moondream1":n=a(e.phi_config);break;case"musicgen":n=a(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"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 i=a(e.decoder),s="num_decoder_layers"in i,o=(0,r.pick)(e,["model_type","is_encoder_decoder"]);return s?(o.num_decoder_layers=i.num_decoder_layers,o.num_decoder_heads=i.num_decoder_heads,o.decoder_hidden_size=i.decoder_hidden_size,o.num_encoder_layers=i.num_encoder_layers,o.num_encoder_heads=i.num_encoder_heads,o.encoder_hidden_size=i.encoder_hidden_size):(o.num_layers=i.num_layers,o.num_heads=i.num_heads,o.hidden_size=i.hidden_size),o}const i={...n,...(0,r.pick)(e,["model_type","multi_query","is_encoder_decoder"])};for(const n in t)i[n]=e[t[n]];return i}function s(e,{prefix:t="past_key_values"}={}){const n={},r=e.normalized_config;if(r.is_encoder_decoder&&"num_encoder_heads"in r&&"num_decoder_heads"in r){const e=r.encoder_dim_kv??r.encoder_hidden_size/r.num_encoder_heads,i=r.decoder_dim_kv??r.decoder_hidden_size/r.num_decoder_heads,a=[1,r.num_encoder_heads,0,e],s=[1,r.num_decoder_heads,0,i];for(let e=0;e<r.num_decoder_layers;++e)n[`${t}.${e}.encoder.key`]=a,n[`${t}.${e}.encoder.value`]=a,n[`${t}.${e}.decoder.key`]=s,n[`${t}.${e}.decoder.value`]=s}else{const e=r.num_heads,i=r.num_layers,a=r.dim_kv??r.hidden_size/(r.num_attention_heads??e);if("falcon"===r.model_type){const r=[1*e,0,a];for(let e=0;e<i;++e)n[`${t}.${e}.key`]=r,n[`${t}.${e}.value`]=r}else if(r.multi_query){const r=[1*e,0,2*a];for(let e=0;e<i;++e)n[`${t}.${e}.key_value`]=r}else if("bloom"===r.model_type){const r=[1*e,a,0],s=[1*e,0,a];for(let e=0;e<i;++e)n[`${t}.${e}.key`]=r,n[`${t}.${e}.value`]=s}else if("openelm"===r.model_type)for(let r=0;r<i;++r){const i=[1,e[r],0,a];n[`${t}.${r}.key`]=i,n[`${t}.${r}.value`]=i}else{const r=[1,e,0,a];for(let e=0;e<i;++e)n[`${t}.${e}.key`]=r,n[`${t}.${e}.value`]=r}}return n}class o{model_type=null;is_encoder_decoder=!1;max_position_embeddings;"transformers.js_config";constructor(e){Object.assign(this,e),this.normalized_config=a(this)}static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:r=null,local_files_only:a=!1,revision:s="main"}={}){!n||n instanceof o||(n=new o(n));const l=n??await async function(e,t){return await(0,i.getModelJSON)(e,"config.json",!0,t)}(e,{progress_callback:t,config:n,cache_dir:r,local_files_only:a,revision:s});return new this(l)}}class l{static async from_pretrained(...e){return o.from_pretrained(...e)}}},"./src/env.js":
 /*!********************!*\
   !*** ./src/env.js ***!
-  \********************/(e,t,n)=>{n.r(t),n.d(t,{apis:()=>f,env:()=>b});var r=n(/*! fs */"?569f"),i=n(/*! path */"?3f59"),a=n(/*! url */"?154a");const s="undefined"!=typeof self,o=s&&"DedicatedWorkerGlobalScope"===self.constructor.name,l=s&&"caches"in self,u="undefined"!=typeof navigator&&"gpu"in navigator,d="undefined"!=typeof navigator&&"ml"in navigator,c="undefined"!=typeof process,p=c&&"node"===process?.release?.name,h=!v(r),m=!v(i),f=Object.freeze({IS_BROWSER_ENV:s,IS_WEBWORKER_ENV:o,IS_WEB_CACHE_AVAILABLE:l,IS_WEBGPU_AVAILABLE:u,IS_WEBNN_AVAILABLE:d,IS_PROCESS_AVAILABLE:c,IS_NODE_ENV:p,IS_FS_AVAILABLE:h,IS_PATH_AVAILABLE:m}),g=h&&m,_=g?i.dirname(i.dirname(a.fileURLToPath(import.meta.url))):"./",w=g?i.join(_,"/.cache/"):null,y="/models/",b={version:"3.0.0-alpha.15",backends:{onnx:{}},allowRemoteModels:!0,remoteHost:"https://huggingface.co/",remotePathTemplate:"{model}/resolve/{revision}/",allowLocalModels:!s,localModelPath:g?i.join(_,y):y,useFS:h,useBrowserCache:l,useFSCache:h,cacheDir:w,useCustomCache:!1,customCache:null};function v(e){return 0===Object.keys(e).length}},"./src/generation/configuration_utils.js":
+  \********************/(e,t,n)=>{n.r(t),n.d(t,{apis:()=>f,env:()=>b});var r=n(/*! fs */"?569f"),i=n(/*! path */"?3f59"),a=n(/*! url */"?154a");const s="undefined"!=typeof self,o=s&&"DedicatedWorkerGlobalScope"===self.constructor.name,l=s&&"caches"in self,u="undefined"!=typeof navigator&&"gpu"in navigator,d="undefined"!=typeof navigator&&"ml"in navigator,c="undefined"!=typeof process,p=c&&"node"===process?.release?.name,h=!v(r),m=!v(i),f=Object.freeze({IS_BROWSER_ENV:s,IS_WEBWORKER_ENV:o,IS_WEB_CACHE_AVAILABLE:l,IS_WEBGPU_AVAILABLE:u,IS_WEBNN_AVAILABLE:d,IS_PROCESS_AVAILABLE:c,IS_NODE_ENV:p,IS_FS_AVAILABLE:h,IS_PATH_AVAILABLE:m}),g=h&&m,_=g?i.dirname(i.dirname(a.fileURLToPath(import.meta.url))):"./",w=g?i.join(_,"/.cache/"):null,y="/models/",b={version:"3.0.0-alpha.16",backends:{onnx:{}},allowRemoteModels:!0,remoteHost:"https://huggingface.co/",remotePathTemplate:"{model}/resolve/{revision}/",allowLocalModels:!s,localModelPath:g?i.join(_,y):y,useFS:h,useBrowserCache:l,useFSCache:h,cacheDir:w,useCustomCache:!1,customCache:null};function v(e){return 0===Object.keys(e).length}},"./src/generation/configuration_utils.js":
 /*!***********************************************!*\
   !*** ./src/generation/configuration_utils.js ***!
   \***********************************************/(e,t,n)=>{n.r(t),n.d(t,{GenerationConfig:()=>i});var r=n(/*! ../utils/core.js */"./src/utils/core.js");class i{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,r.pick)(e,Object.getOwnPropertyNames(this)))}}},"./src/generation/logits_process.js":
@@ -172,7 +172,7 @@ var r,i,a,s,o,l,u,d,c,p,h,m,f,g,_,w,y,b,v,x,M,T,k,$,S,C,P,E,F,A,I,z,O,B=Object.d
   \*************************************/(e,t,n)=>{n.r(t),n.d(t,{BaseStreamer:()=>s,TextStreamer:()=>l,WhisperTextStreamer:()=>u});var r=n(/*! ../utils/core.js */"./src/utils/core.js"),i=n(/*! ../tokenizers.js */"./src/tokenizers.js"),a=n(/*! ../env.js */"./src/env.js");class s{put(e){throw Error("Not implemented")}end(){throw Error("Not implemented")}}const o=a.apis.IS_PROCESS_AVAILABLE?e=>process.stdout.write(e):e=>console.log(e);class l extends s{constructor(e,{skip_prompt:t=!1,callback_function:n=null,token_callback_function:r=null,decode_kwargs:i={},...a}={}){super(),this.tokenizer=e,this.skip_prompt=t,this.callback_function=n??o,this.token_callback_function=r,this.decode_kwargs={...i,...a},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,r.mergeArrays)(this.token_cache,t);const n=this.tokenizer.decode(this.token_cache,this.decode_kwargs);let a;n.endsWith("\n")?(a=n.slice(this.print_len),this.token_cache=[],this.print_len=0):n.length>0&&(0,i.is_chinese_char)(n.charCodeAt(n.length-1))?(a=n.slice(this.print_len),this.print_len+=a.length):(a=n.slice(this.print_len,n.lastIndexOf(" ")+1),this.print_len+=a.length),this.on_finalized_text(a,!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===o&&a.apis.IS_PROCESS_AVAILABLE&&this.callback_function?.("\n")}}class u extends l{constructor(e,{skip_prompt:t=!1,callback_function:n=null,token_callback_function:r=null,on_chunk_start:i=null,on_chunk_end:a=null,on_finalize:s=null,time_precision:o=.02,skip_special_tokens:l=!0,decode_kwargs:u={}}={}){super(e,{skip_prompt:t,callback_function:n,token_callback_function:r,decode_kwargs:{skip_special_tokens:l,...u}}),this.timestamp_begin=e.timestamp_begin,this.on_chunk_start=i,this.on_chunk_end=a,this.on_finalize=s,this.time_precision=o,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)=>{n.r(t),n.d(t,{ASTForAudioClassification:()=>nn,ASTModel:()=>tn,ASTPreTrainedModel:()=>en,AlbertForMaskedLM:()=>ct,AlbertForQuestionAnswering:()=>dt,AlbertForSequenceClassification:()=>ut,AlbertModel:()=>lt,AlbertPreTrainedModel:()=>ot,AutoModel:()=>Fo,AutoModelForAudioClassification:()=>Qo,AutoModelForAudioFrameClassification:()=>Jo,AutoModelForCTC:()=>Ko,AutoModelForCausalLM:()=>Do,AutoModelForDepthEstimation:()=>nl,AutoModelForDocumentQuestionAnswering:()=>Zo,AutoModelForImageClassification:()=>jo,AutoModelForImageFeatureExtraction:()=>il,AutoModelForImageMatting:()=>el,AutoModelForImageSegmentation:()=>Go,AutoModelForImageToImage:()=>tl,AutoModelForMaskGeneration:()=>Xo,AutoModelForMaskedLM:()=>Ro,AutoModelForNormalEstimation:()=>rl,AutoModelForObjectDetection:()=>Wo,AutoModelForQuestionAnswering:()=>No,AutoModelForSemanticSegmentation:()=>qo,AutoModelForSeq2SeqLM:()=>zo,AutoModelForSequenceClassification:()=>Ao,AutoModelForSpeechSeq2Seq:()=>Oo,AutoModelForTextToSpectrogram:()=>Bo,AutoModelForTextToWaveform:()=>Lo,AutoModelForTokenClassification:()=>Io,AutoModelForUniversalSegmentation:()=>Uo,AutoModelForVision2Seq:()=>Vo,AutoModelForXVector:()=>Yo,AutoModelForZeroShotObjectDetection:()=>Ho,BartForConditionalGeneration:()=>Mt,BartForSequenceClassification:()=>Tt,BartModel:()=>xt,BartPretrainedModel:()=>vt,BaseModelOutput:()=>G,BeitForImageClassification:()=>si,BeitModel:()=>ai,BeitPreTrainedModel:()=>ii,BertForMaskedLM:()=>W,BertForQuestionAnswering:()=>K,BertForSequenceClassification:()=>H,BertForTokenClassification:()=>X,BertModel:()=>U,BertPreTrainedModel:()=>q,BlenderbotForConditionalGeneration:()=>At,BlenderbotModel:()=>Ft,BlenderbotPreTrainedModel:()=>Et,BlenderbotSmallForConditionalGeneration:()=>Ot,BlenderbotSmallModel:()=>zt,BlenderbotSmallPreTrainedModel:()=>It,BloomForCausalLM:()=>vr,BloomModel:()=>br,BloomPreTrainedModel:()=>yr,CLIPModel:()=>mn,CLIPPreTrainedModel:()=>hn,CLIPSegForImageSegmentation:()=>Sn,CLIPSegModel:()=>$n,CLIPSegPreTrainedModel:()=>kn,CLIPTextModel:()=>fn,CLIPTextModelWithProjection:()=>gn,CLIPVisionModel:()=>_n,CLIPVisionModelWithProjection:()=>wn,CamembertForMaskedLM:()=>we,CamembertForQuestionAnswering:()=>ve,CamembertForSequenceClassification:()=>ye,CamembertForTokenClassification:()=>be,CamembertModel:()=>_e,CamembertPreTrainedModel:()=>ge,CausalLMOutput:()=>cl,CausalLMOutputWithPast:()=>pl,ChineseCLIPModel:()=>Tn,ChineseCLIPPreTrainedModel:()=>Mn,ClapAudioModelWithProjection:()=>ys,ClapModel:()=>_s,ClapPreTrainedModel:()=>gs,ClapTextModelWithProjection:()=>ws,CodeGenForCausalLM:()=>Xn,CodeGenModel:()=>Hn,CodeGenPreTrainedModel:()=>Wn,CohereForCausalLM:()=>er,CohereModel:()=>Zn,CoherePreTrainedModel:()=>Jn,ConvBertForMaskedLM:()=>se,ConvBertForQuestionAnswering:()=>ue,ConvBertForSequenceClassification:()=>oe,ConvBertForTokenClassification:()=>le,ConvBertModel:()=>ae,ConvBertPreTrainedModel:()=>ie,ConvNextForImageClassification:()=>ta,ConvNextModel:()=>ea,ConvNextPreTrainedModel:()=>Zi,ConvNextV2ForImageClassification:()=>ia,ConvNextV2Model:()=>ra,ConvNextV2PreTrainedModel:()=>na,DPTForDepthEstimation:()=>Di,DPTModel:()=>Li,DPTPreTrainedModel:()=>Bi,DebertaForMaskedLM:()=>Te,DebertaForQuestionAnswering:()=>Se,DebertaForSequenceClassification:()=>ke,DebertaForTokenClassification:()=>$e,DebertaModel:()=>Me,DebertaPreTrainedModel:()=>xe,DebertaV2ForMaskedLM:()=>Ee,DebertaV2ForQuestionAnswering:()=>Ie,DebertaV2ForSequenceClassification:()=>Fe,DebertaV2ForTokenClassification:()=>Ae,DebertaV2Model:()=>Pe,DebertaV2PreTrainedModel:()=>Ce,DecisionTransformerModel:()=>Ks,DecisionTransformerPreTrainedModel:()=>Xs,DeiTForImageClassification:()=>Mi,DeiTModel:()=>xi,DeiTPreTrainedModel:()=>vi,DepthAnythingForDepthEstimation:()=>Ni,DepthAnythingPreTrainedModel:()=>Ri,DetrForObjectDetection:()=>ui,DetrForSegmentation:()=>di,DetrModel:()=>li,DetrObjectDetectionOutput:()=>ci,DetrPreTrainedModel:()=>oi,DetrSegmentationOutput:()=>pi,Dinov2ForImageClassification:()=>oa,Dinov2Model:()=>sa,Dinov2PreTrainedModel:()=>aa,DistilBertForMaskedLM:()=>Re,DistilBertForQuestionAnswering:()=>De,DistilBertForSequenceClassification:()=>Be,DistilBertForTokenClassification:()=>Le,DistilBertModel:()=>Oe,DistilBertPreTrainedModel:()=>ze,DonutSwinModel:()=>Ji,DonutSwinPreTrainedModel:()=>Yi,EfficientNetForImageClassification:()=>Fs,EfficientNetModel:()=>Es,EfficientNetPreTrainedModel:()=>Ps,ElectraForMaskedLM:()=>pe,ElectraForQuestionAnswering:()=>fe,ElectraForSequenceClassification:()=>he,ElectraForTokenClassification:()=>me,ElectraModel:()=>ce,ElectraPreTrainedModel:()=>de,EsmForMaskedLM:()=>je,EsmForSequenceClassification:()=>Ge,EsmForTokenClassification:()=>qe,EsmModel:()=>Ve,EsmPreTrainedModel:()=>Ne,FalconForCausalLM:()=>fs,FalconModel:()=>ms,FalconPreTrainedModel:()=>hs,FastViTForImageClassification:()=>Gr,FastViTModel:()=>jr,FastViTPreTrainedModel:()=>Vr,Florence2ForConditionalGeneration:()=>pn,Florence2PreTrainedModel:()=>cn,GLPNForDepthEstimation:()=>Qi,GLPNModel:()=>Ki,GLPNPreTrainedModel:()=>Xi,GPT2LMHeadModel:()=>En,GPT2Model:()=>Pn,GPT2PreTrainedModel:()=>Cn,GPTBigCodeForCausalLM:()=>Un,GPTBigCodeModel:()=>qn,GPTBigCodePreTrainedModel:()=>Gn,GPTJForCausalLM:()=>jn,GPTJModel:()=>Vn,GPTJPreTrainedModel:()=>Nn,GPTNeoForCausalLM:()=>Bn,GPTNeoModel:()=>On,GPTNeoPreTrainedModel:()=>zn,GPTNeoXForCausalLM:()=>Rn,GPTNeoXModel:()=>Dn,GPTNeoXPreTrainedModel:()=>Ln,Gemma2ForCausalLM:()=>sr,Gemma2Model:()=>ar,Gemma2PreTrainedModel:()=>ir,GemmaForCausalLM:()=>rr,GemmaModel:()=>nr,GemmaPreTrainedModel:()=>tr,GroupViTModel:()=>Nr,GroupViTPreTrainedModel:()=>Rr,HieraForImageClassification:()=>$i,HieraModel:()=>ki,HieraPreTrainedModel:()=>Ti,HubertForCTC:()=>Wa,HubertForSequenceClassification:()=>Ha,HubertModel:()=>Ua,HubertPreTrainedModel:()=>qa,ImageMattingOutput:()=>hl,JAISLMHeadModel:()=>In,JAISModel:()=>An,JAISPreTrainedModel:()=>Fn,LlamaForCausalLM:()=>Yn,LlamaModel:()=>Qn,LlamaPreTrainedModel:()=>Kn,LlavaForConditionalGeneration:()=>un,LlavaPreTrainedModel:()=>ln,LongT5ForConditionalGeneration:()=>_t,LongT5Model:()=>gt,LongT5PreTrainedModel:()=>ft,M2M100ForConditionalGeneration:()=>ba,M2M100Model:()=>ya,M2M100PreTrainedModel:()=>wa,MBartForCausalLM:()=>Pt,MBartForConditionalGeneration:()=>St,MBartForSequenceClassification:()=>Ct,MBartModel:()=>$t,MBartPreTrainedModel:()=>kt,MPNetForMaskedLM:()=>Je,MPNetForQuestionAnswering:()=>tt,MPNetForSequenceClassification:()=>Ze,MPNetForTokenClassification:()=>et,MPNetModel:()=>Ye,MPNetPreTrainedModel:()=>Qe,MT5ForConditionalGeneration:()=>bt,MT5Model:()=>yt,MT5PreTrainedModel:()=>wt,MarianMTModel:()=>_a,MarianModel:()=>ga,MarianPreTrainedModel:()=>fa,MaskFormerForInstanceSegmentation:()=>Hi,MaskFormerModel:()=>Wi,MaskFormerPreTrainedModel:()=>Ui,MaskedLMOutput:()=>ul,MistralForCausalLM:()=>us,MistralModel:()=>ls,MistralPreTrainedModel:()=>os,MobileBertForMaskedLM:()=>He,MobileBertForQuestionAnswering:()=>Ke,MobileBertForSequenceClassification:()=>Xe,MobileBertModel:()=>We,MobileBertPreTrainedModel:()=>Ue,MobileNetV1ForImageClassification:()=>Ds,MobileNetV1Model:()=>Ls,MobileNetV1PreTrainedModel:()=>Bs,MobileNetV2ForImageClassification:()=>Vs,MobileNetV2Model:()=>Ns,MobileNetV2PreTrainedModel:()=>Rs,MobileNetV3ForImageClassification:()=>qs,MobileNetV3Model:()=>Gs,MobileNetV3PreTrainedModel:()=>js,MobileNetV4ForImageClassification:()=>Hs,MobileNetV4Model:()=>Ws,MobileNetV4PreTrainedModel:()=>Us,MobileViTForImageClassification:()=>Xr,MobileViTModel:()=>Hr,MobileViTPreTrainedModel:()=>Wr,MobileViTV2ForImageClassification:()=>Yr,MobileViTV2Model:()=>Qr,MobileViTV2PreTrainedModel:()=>Kr,ModelOutput:()=>j,Moondream1ForConditionalGeneration:()=>dn,MptForCausalLM:()=>Tr,MptModel:()=>Mr,MptPreTrainedModel:()=>xr,MusicgenForCausalLM:()=>zs,MusicgenForConditionalGeneration:()=>Os,MusicgenModel:()=>Is,MusicgenPreTrainedModel:()=>As,NomicBertModel:()=>Y,NomicBertPreTrainedModel:()=>Q,OPTForCausalLM:()=>Sr,OPTModel:()=>$r,OPTPreTrainedModel:()=>kr,OpenELMForCausalLM:()=>ur,OpenELMModel:()=>lr,OpenELMPreTrainedModel:()=>or,OwlViTForObjectDetection:()=>ei,OwlViTModel:()=>Zr,OwlViTPreTrainedModel:()=>Jr,Owlv2ForObjectDetection:()=>ri,Owlv2Model:()=>ni,Owlv2PreTrainedModel:()=>ti,Phi3ForCausalLM:()=>wr,Phi3Model:()=>_r,Phi3PreTrainedModel:()=>gr,PhiForCausalLM:()=>fr,PhiModel:()=>mr,PhiPreTrainedModel:()=>hr,PreTrainedModel:()=>V,PretrainedMixin:()=>Qs,PvtForImageClassification:()=>Ir,PvtModel:()=>Ar,PvtPreTrainedModel:()=>Fr,PyAnnoteForAudioFrameClassification:()=>Ca,PyAnnoteModel:()=>Sa,PyAnnotePreTrainedModel:()=>$a,QuestionAnsweringModelOutput:()=>dl,Qwen2ForCausalLM:()=>pr,Qwen2Model:()=>cr,Qwen2PreTrainedModel:()=>dr,RTDetrForObjectDetection:()=>fi,RTDetrModel:()=>mi,RTDetrObjectDetectionOutput:()=>gi,RTDetrPreTrainedModel:()=>hi,ResNetForImageClassification:()=>Pi,ResNetModel:()=>Ci,ResNetPreTrainedModel:()=>Si,RoFormerForMaskedLM:()=>ee,RoFormerForQuestionAnswering:()=>re,RoFormerForSequenceClassification:()=>te,RoFormerForTokenClassification:()=>ne,RoFormerModel:()=>Z,RoFormerPreTrainedModel:()=>J,RobertaForMaskedLM:()=>Dt,RobertaForQuestionAnswering:()=>Vt,RobertaForSequenceClassification:()=>Rt,RobertaForTokenClassification:()=>Nt,RobertaModel:()=>Lt,RobertaPreTrainedModel:()=>Bt,SamImageSegmentationOutput:()=>ma,SamModel:()=>ha,SamPreTrainedModel:()=>pa,SapiensForDepthEstimation:()=>Gi,SapiensForNormalEstimation:()=>qi,SapiensForSemanticSegmentation:()=>ji,SapiensPreTrainedModel:()=>Vi,SegformerForImageClassification:()=>Ts,SegformerForSemanticSegmentation:()=>ks,SegformerModel:()=>Ms,SegformerPreTrainedModel:()=>xs,Seq2SeqLMOutput:()=>al,SequenceClassifierOutput:()=>sl,SiglipModel:()=>bn,SiglipPreTrainedModel:()=>yn,SiglipTextModel:()=>vn,SiglipVisionModel:()=>xn,SpeechT5ForSpeechToText:()=>ns,SpeechT5ForTextToSpeech:()=>rs,SpeechT5HifiGan:()=>is,SpeechT5Model:()=>ts,SpeechT5PreTrainedModel:()=>es,SqueezeBertForMaskedLM:()=>it,SqueezeBertForQuestionAnswering:()=>st,SqueezeBertForSequenceClassification:()=>at,SqueezeBertModel:()=>rt,SqueezeBertPreTrainedModel:()=>nt,StableLmForCausalLM:()=>Cs,StableLmModel:()=>Ss,StableLmPreTrainedModel:()=>$s,Starcoder2ForCausalLM:()=>ps,Starcoder2Model:()=>cs,Starcoder2PreTrainedModel:()=>ds,Swin2SRForImageSuperResolution:()=>Oi,Swin2SRModel:()=>zi,Swin2SRPreTrainedModel:()=>Ii,SwinForImageClassification:()=>Ai,SwinModel:()=>Fi,SwinPreTrainedModel:()=>Ei,T5ForConditionalGeneration:()=>mt,T5Model:()=>ht,T5PreTrainedModel:()=>pt,TableTransformerForObjectDetection:()=>yi,TableTransformerModel:()=>wi,TableTransformerObjectDetectionOutput:()=>bi,TableTransformerPreTrainedModel:()=>_i,TokenClassifierOutput:()=>ll,TrOCRForCausalLM:()=>ss,TrOCRPreTrainedModel:()=>as,UniSpeechForCTC:()=>Ia,UniSpeechForSequenceClassification:()=>za,UniSpeechModel:()=>Aa,UniSpeechPreTrainedModel:()=>Fa,UniSpeechSatForAudioFrameClassification:()=>Ra,UniSpeechSatForCTC:()=>La,UniSpeechSatForSequenceClassification:()=>Da,UniSpeechSatModel:()=>Ba,UniSpeechSatPreTrainedModel:()=>Oa,ViTForImageClassification:()=>Er,ViTMAEModel:()=>Or,ViTMAEPreTrainedModel:()=>zr,ViTMSNForImageClassification:()=>Dr,ViTMSNModel:()=>Lr,ViTMSNPreTrainedModel:()=>Br,ViTModel:()=>Pr,ViTPreTrainedModel:()=>Cr,VisionEncoderDecoderModel:()=>on,VitMatteForImageMatting:()=>Ur,VitMattePreTrainedModel:()=>qr,VitsModel:()=>vs,VitsModelOutput:()=>ml,VitsPreTrainedModel:()=>bs,Wav2Vec2BertForCTC:()=>ja,Wav2Vec2BertForSequenceClassification:()=>Ga,Wav2Vec2BertModel:()=>Va,Wav2Vec2BertPreTrainedModel:()=>Na,Wav2Vec2ForAudioFrameClassification:()=>ka,Wav2Vec2ForCTC:()=>Ma,Wav2Vec2ForSequenceClassification:()=>Ta,Wav2Vec2Model:()=>xa,Wav2Vec2PreTrainedModel:()=>va,WavLMForAudioFrameClassification:()=>Za,WavLMForCTC:()=>Qa,WavLMForSequenceClassification:()=>Ya,WavLMForXVector:()=>Ja,WavLMModel:()=>Ka,WavLMPreTrainedModel:()=>Xa,WeSpeakerResNetModel:()=>Ea,WeSpeakerResNetPreTrainedModel:()=>Pa,WhisperForConditionalGeneration:()=>sn,WhisperModel:()=>an,WhisperPreTrainedModel:()=>rn,XLMForQuestionAnswering:()=>Ht,XLMForSequenceClassification:()=>Ut,XLMForTokenClassification:()=>Wt,XLMModel:()=>Gt,XLMPreTrainedModel:()=>jt,XLMRobertaForMaskedLM:()=>Qt,XLMRobertaForQuestionAnswering:()=>Zt,XLMRobertaForSequenceClassification:()=>Yt,XLMRobertaForTokenClassification:()=>Jt,XLMRobertaModel:()=>Kt,XLMRobertaPreTrainedModel:()=>Xt,XLMWithLMHeadModel:()=>qt,XVectorOutput:()=>ol,YolosForObjectDetection:()=>da,YolosModel:()=>ua,YolosObjectDetectionOutput:()=>ca,YolosPreTrainedModel:()=>la});var r=n(/*! ./configs.js */"./src/configs.js"),i=n(/*! ./backends/onnx.js */"./src/backends/onnx.js"),a=n(/*! ./utils/dtypes.js */"./src/utils/dtypes.js"),s=n(/*! ./utils/generic.js */"./src/utils/generic.js"),o=n(/*! ./utils/core.js */"./src/utils/core.js"),l=n(/*! ./utils/hub.js */"./src/utils/hub.js"),u=n(/*! ./generation/logits_process.js */"./src/generation/logits_process.js"),d=n(/*! ./generation/configuration_utils.js */"./src/generation/configuration_utils.js"),c=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"),m=n(/*! ./generation/logits_sampler.js */"./src/generation/logits_sampler.js"),f=n(/*! ./env.js */"./src/env.js"),g=n(/*! ./models/whisper/generation_whisper.js */"./src/models/whisper/generation_whisper.js"),_=n(/*! ./models/whisper/common_whisper.js */"./src/models/whisper/common_whisper.js");const w=0,y=1,b=2,v=3,x=4,M=5,T=6,k=7,$=new Map,S=new Map,C=new Map;async function P(e,t,n){return Object.fromEntries(await Promise.all(Object.keys(t).map((async s=>{const{buffer:o,session_options:u}=await async function(e,t,n){let s=n.device;s&&"string"!=typeof s&&(s.hasOwnProperty(t)?s=s[t]:(console.warn(`device not specified for "${t}". Using the default device.`),s=null));const o=s??(f.apis.IS_NODE_ENV?"cpu":"wasm"),u=(0,i.deviceToExecutionProviders)(o);let d=n.dtype;"string"!=typeof d&&(d&&d.hasOwnProperty(t)?d=d[t]:(d=a.DEFAULT_DEVICE_DTYPE_MAPPING[o]??a.DATA_TYPES.fp32,console.warn(`dtype not specified for "${t}". Using the default dtype (${d}) for this device (${o}).`)));const c=d;if(!a.DEFAULT_DTYPE_SUFFIX_MAPPING.hasOwnProperty(c))throw new Error(`Invalid dtype: ${c}. Should be one of: ${Object.keys(a.DATA_TYPES).join(", ")}`);if(c===a.DATA_TYPES.fp16&&"webgpu"===o&&!await(0,a.isWebGpuFp16Supported)())throw new Error(`The device (${o}) does not support fp16.`);const p=a.DEFAULT_DTYPE_SUFFIX_MAPPING[c],h=`${n.subfolder??""}/${t}${p}.onnx`,m={...n.session_options}??{};m.executionProviders??=u;const g=(0,l.getModelFile)(e,h,!0,n);let _=[];if(n.use_external_data_format&&(!0===n.use_external_data_format||"object"==typeof n.use_external_data_format&&n.use_external_data_format.hasOwnProperty(t)&&!0===n.use_external_data_format[t])){if(f.apis.IS_NODE_ENV)throw new Error("External data format is not yet supported in Node.js");const r=`${t}${p}.onnx_data`,i=`${n.subfolder??""}/${r}`;_.push(new Promise((async(t,a)=>{const s=await(0,l.getModelFile)(e,i,!0,n);t({path:r,data:s})})))}else void 0!==m.externalData&&(_=m.externalData.map((async t=>{if("string"==typeof t.data){const r=await(0,l.getModelFile)(e,t.data,!0,n);return{...t,data:r}}return t})));if(_.length>0&&(m.externalData=await Promise.all(_)),"webgpu"===o){const e=(0,r.getKeyValueShapes)(n.config,{prefix:"present"});if(Object.keys(e).length>0&&!(0,i.isONNXProxy)()){const t={};for(const n in e)t[n]="gpu-buffer";m.preferredOutputLocation=t}}return{buffer:await g,session_options:m}}(e,t[s],n);return[s,await(0,i.createInferenceSession)(o,u)]}))))}async function E(e,t){const n=function(e,t){const n=Object.create(null),r=[];for(const a of e.inputNames){const e=t[a];e instanceof c.Tensor?n[a]=(0,i.isONNXProxy)()?e.clone():e:r.push(a)}if(r.length>0)throw new Error(`An error occurred during model execution: "Missing the following inputs: ${r.join(", ")}.`);const a=Object.keys(t).length,s=e.inputNames.length;if(a>s){let n=Object.keys(t).filter((t=>!e.inputNames.includes(t)));console.warn(`WARNING: Too many inputs were provided (${a} > ${s}). The following inputs will be ignored: "${n.join(", ")}".`)}return n}(e,t);try{const t=Object.fromEntries(Object.entries(n).map((([e,t])=>[e,t.ort_tensor])));let r=await e.run(t);return r=F(r),r}catch(e){throw console.error(`An error occurred during model execution: "${e}".`),console.error("Inputs given to model:",n),e}}function F(e){for(let t in e)(0,i.isONNXTensor)(e[t])?e[t]=new c.Tensor(e[t]):"object"==typeof e[t]&&F(e[t]);return e}function A(e){if(e instanceof c.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 c.Tensor("int64",BigInt64Array.from(e.flat().map((e=>BigInt(e)))),[e.length,e[0].length])}return new c.Tensor("int64",BigInt64Array.from(e.map((e=>BigInt(e)))),[1,e.length])}function I(e){return new c.Tensor("bool",[e],[1])}async function z(e,t){let{encoder_outputs:n,input_ids:r,decoder_input_ids:i,...a}=t;if(!n){const r=(0,o.pick)(t,e.sessions.model.inputNames);n=(await O(e,r)).last_hidden_state}a.input_ids=i,a.encoder_hidden_states=n,e.sessions.decoder_model_merged.inputNames.includes("encoder_attention_mask")&&(a.encoder_attention_mask=t.attention_mask);return await B(e,a,!0)}async function O(e,t){const n=e.sessions.model,r=(0,o.pick)(t,n.inputNames);if(n.inputNames.includes("inputs_embeds")&&!r.inputs_embeds){if(!t.input_ids)throw new Error("Both `input_ids` and `inputs_embeds` are missing in the model inputs.");r.inputs_embeds=await e.encode_text({input_ids:t.input_ids})}return n.inputNames.includes("token_type_ids")&&!r.token_type_ids&&(r.token_type_ids=new c.Tensor("int64",new BigInt64Array(r.input_ids.data.length),r.input_ids.dims)),await E(n,r)}async function B(e,t,n=!1){const r=e.sessions[n?"decoder_model_merged":"model"],{past_key_values:i,...a}=t;r.inputNames.includes("use_cache_branch")&&(a.use_cache_branch=I(!!i)),r.inputNames.includes("position_ids")&&a.attention_mask&&!a.position_ids&&(a.position_ids=function(e,t=null){const{input_ids:n,inputs_embeds:r,attention_mask:i}=e,[a,s]=i.dims,o=new BigInt64Array(i.data.length);for(let e=0;e<a;++e){const t=e*s;let n=BigInt(0);for(let e=0;e<s;++e){const r=t+e;0n===i.data[r]?o[r]=BigInt(1):(o[r]=n,n+=i.data[r])}}let l=new c.Tensor("int64",o,i.dims);if(t){const e=-(n??r).dims.at(1);l=l.slice(null,[e,null])}return l}(a,i)),e.addPastKeyValues(a,i);const s=(0,o.pick)(a,r.inputNames);return await E(r,s)}async function L(e,{input_ids:t=null,attention_mask:n=null,pixel_values:r=null,position_ids:i=null,inputs_embeds:a=null,past_key_values:s=null,generation_config:o=null,logits_processor:l=null,...u}){if(!a)if(a=await e.encode_text({input_ids:t}),r&&1!==t.dims[1]){const i=await e.encode_image({pixel_values:r});({inputs_embeds:a,attention_mask:n}=e._merge_input_ids_with_image_features({image_features:i,inputs_embeds:a,input_ids:t,attention_mask:n}))}else if(s&&r&&1===t.dims[1]){const e=t.dims[1],r=Object.values(s)[0].dims.at(-2);n=(0,c.cat)([(0,c.ones)([t.dims[0],r]),n.slice(null,[n.dims[1]-e,n.dims[1]])],1)}return await B(e,{inputs_embeds:a,past_key_values:s,attention_mask:n,position_ids:i,generation_config:o,logits_processor:l},!0)}function D(e,t,n,r){if(n.past_key_values){const t=Object.values(n.past_key_values)[0].dims.at(-2),{input_ids:r,attention_mask:i}=n;if(i&&i.dims[1]>r.dims[1]);else if(t<r.dims[1])n.input_ids=r.slice(null,[t,null]);else if(null!=e.config.image_token_index&&r.data.some((t=>t==e.config.image_token_index))){const i=e.config.num_image_tokens;if(!i)throw new Error("`num_image_tokens` is missing in the model configuration.");const a=r.dims[1]-(t-i);n.input_ids=r.slice(null,[-a,null]),n.attention_mask=(0,c.ones)([1,t+a])}}return n}function R(e,t,n,r){return n.past_key_values&&(t=t.map((e=>[e.at(-1)]))),{...n,decoder_input_ids:A(t)}}function N(e,...t){return e.config.is_encoder_decoder?R(e,...t):D(e,...t)}class V extends s.Callable{main_input_name="input_ids";forward_params=["input_ids","attention_mask"];constructor(e,t){super(),this.config=e,this.sessions=t;const n=C.get(this.constructor),r=$.get(n);switch(this.can_generate=!1,this._forward=null,this._prepare_inputs_for_generation=null,r){case x:this.can_generate=!0,this._forward=B,this._prepare_inputs_for_generation=D;break;case b:case v:case k:this.can_generate=!0,this._forward=z,this._prepare_inputs_for_generation=R;break;case y:this._forward=z;break;case T:this.can_generate=!0,this._forward=L,this._prepare_inputs_for_generation=N;break;default:this._forward=O}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:i=null,local_files_only:a=!1,revision:s="main",model_file_name:o=null,subfolder:u="onnx",device:d=null,dtype:c=null,use_external_data_format:p=null,session_options:h={}}={}){let m={progress_callback:t,config:n,cache_dir:i,local_files_only:a,revision:s,model_file_name:o,subfolder:u,device:d,dtype:c,use_external_data_format:p,session_options:h};const f=C.get(this),g=$.get(f);let _;if(n=m.config=await r.AutoConfig.from_pretrained(e,m),g===x)_=await Promise.all([P(e,{model:m.model_file_name??"model"},m),(0,l.getModelJSON)(e,"generation_config.json",!1,m)]);else if(g===b||g===v)_=await Promise.all([P(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},m),(0,l.getModelJSON)(e,"generation_config.json",!1,m)]);else if(g===M)_=await Promise.all([P(e,{model:"vision_encoder",prompt_encoder_mask_decoder:"prompt_encoder_mask_decoder"},m)]);else if(g===y)_=await Promise.all([P(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},m)]);else if(g===T){const t={embed_tokens:"embed_tokens",vision_encoder:"vision_encoder",decoder_model_merged:"decoder_model_merged"};n.is_encoder_decoder&&(t.model="encoder_model"),_=await Promise.all([P(e,t,m),(0,l.getModelJSON)(e,"generation_config.json",!1,m)])}else g===k?_=await Promise.all([P(e,{model:"text_encoder",decoder_model_merged:"decoder_model_merged",encodec_decode:"encodec_decode"},m),(0,l.getModelJSON)(e,"generation_config.json",!1,m)]):(g!==w&&console.warn(`Model type for '${f??n?.model_type}' not found, assuming encoder-only architecture. Please report this at https://github.com/xenova/transformers.js/issues/new/choose.`),_=await Promise.all([P(e,{model:m.model_file_name??"model"},m)]));return new this(n,..._)}async _call(e){return await this.forward(e)}async forward(e){return await this._forward(this,e)}_get_logits_warper(e){const t=new u.LogitsProcessorList;return null!==e.temperature&&1!==e.temperature&&t.push(new u.TemperatureLogitsWarper(e.temperature)),null!==e.top_k&&0!==e.top_k&&t.push(new u.TopKLogitsWarper(e.top_k)),null!==e.top_p&&e.top_p<1&&t.push(new u.TopPLogitsWarper(e.top_p)),t}_get_logits_processor(e,t,n=null){const r=new u.LogitsProcessorList;if(null!==e.repetition_penalty&&1!==e.repetition_penalty&&r.push(new u.RepetitionPenaltyLogitsProcessor(e.repetition_penalty)),null!==e.no_repeat_ngram_size&&e.no_repeat_ngram_size>0&&r.push(new u.NoRepeatNGramLogitsProcessor(e.no_repeat_ngram_size)),null!==e.bad_words_ids&&r.push(new u.NoBadWordsLogitsProcessor(e.bad_words_ids,e.eos_token_id)),null!==e.min_length&&null!==e.eos_token_id&&e.min_length>0&&r.push(new u.MinLengthLogitsProcessor(e.min_length,e.eos_token_id)),null!==e.min_new_tokens&&null!==e.eos_token_id&&e.min_new_tokens>0&&r.push(new u.MinNewTokensLengthLogitsProcessor(t,e.min_new_tokens,e.eos_token_id)),null!==e.forced_bos_token_id&&r.push(new u.ForcedBOSTokenLogitsProcessor(e.forced_bos_token_id)),null!==e.forced_eos_token_id&&r.push(new u.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;r.push(new u.SuppressTokensAtBeginLogitsProcessor(e.begin_suppress_tokens,n))}return null!==e.guidance_scale&&e.guidance_scale>1&&r.push(new u.ClassifierFreeGuidanceLogitsProcessor(e.guidance_scale)),null!==n&&r.extend(n),r}_prepare_generation_config(e,t,n=d.GenerationConfig){const r={...this.config};for(const e of["decoder","generator","text_config"])e in r&&Object.assign(r,r[e]);const i=new n(r);return"generation_config"in this&&Object.assign(i,this.generation_config),e&&Object.assign(i,e),t&&Object.assign(i,(0,o.pick)(t,Object.getOwnPropertyNames(i))),i}_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=[so,uo,ao,eo],t=C.get(this.constructor),n=new Set,r=this.config.model_type;for(const t of e){const e=t.get(r);e&&n.add(e[0])}let i=`The current model class (${t}) is not compatible with \`.generate()\`, as it doesn't have a language model head.`;throw n.size>0&&(i+=` Please use the following class instead: ${[...n].join(", ")}`),Error(i)}}prepare_inputs_for_generation(...e){return this._prepare_inputs_for_generation(this,...e)}_update_model_kwargs_for_generation({generated_input_ids:e,outputs:t,model_inputs:n,is_encoder_decoder:r}){return n.past_key_values=this.getPastKeyValues(t,n.past_key_values),n.input_ids=new c.Tensor("int64",e.flat(),[e.length,1]),r||(n.attention_mask=(0,c.cat)([n.attention_mask,(0,c.ones)([n.attention_mask.dims[0],1])],1)),n.position_ids=null,n}_prepare_model_inputs({inputs:e,bos_token_id:t,model_kwargs:n}){const r=(0,o.pick)(n,this.forward_params),i=this.main_input_name;if(i in r){if(e)throw new Error("`inputs`: {inputs}` were passed alongside {input_name} which is not allowed. Make sure to either pass {inputs} or {input_name}=...")}else r[i]=e;return{inputs_tensor:r[i],model_inputs:r,model_input_name:i}}async _prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:e,model_inputs:t,model_input_name:n,generation_config:r}){if(this.sessions.model.inputNames.includes("inputs_embeds")&&!t.inputs_embeds&&"_prepare_inputs_embeds"in this){const{input_ids:e,pixel_values:n,attention_mask:r,...i}=t,a=await this._prepare_inputs_embeds(t);t={...i,...(0,o.pick)(a,["inputs_embeds","attention_mask"])}}let{last_hidden_state:i}=await O(this,t);if(null!==r.guidance_scale&&r.guidance_scale>1)i=(0,c.cat)([i,(0,c.full_like)(i,0)],0),"attention_mask"in t&&(t.attention_mask=(0,c.cat)([t.attention_mask,(0,c.zeros_like)(t.attention_mask)],0));else if(t.decoder_input_ids){const e=A(t.decoder_input_ids).dims[0];if(e!==i.dims[0]){if(1!==i.dims[0])throw new Error(`The encoder outputs have a different batch size (${i.dims[0]}) than the decoder inputs (${e}).`);i=(0,c.cat)(Array.from({length:e},(()=>i)),0)}}return t.encoder_outputs=i,t}_prepare_decoder_input_ids_for_generation({batch_size:e,model_input_name:t,model_kwargs:n,decoder_start_token_id:r,bos_token_id:i,generation_config:a}){let{decoder_input_ids:s,...o}=n;if(s)Array.isArray(s[0])||(s=Array.from({length:e},(()=>s)));else if(r??=i,"musicgen"===this.config.model_type)s=Array.from({length:e*this.config.decoder.num_codebooks},(()=>[r]));else if(Array.isArray(r)){if(r.length!==e)throw new Error(`\`decoder_start_token_id\` expcted to have length ${e} but got ${r.length}`);s=r}else s=Array.from({length:e},(()=>[r]));return s=A(s),n.decoder_attention_mask=(0,c.ones_like)(s),{input_ids:s,model_inputs:o}}async generate({inputs:e=null,generation_config:t=null,logits_processor:n=null,stopping_criteria:r=null,streamer:i=null,...a}){this._validate_model_class(),t=this._prepare_generation_config(t,a);let{inputs_tensor:s,model_inputs:o,model_input_name:l}=this._prepare_model_inputs({inputs:e,model_kwargs:a});const u=this.config.is_encoder_decoder;let d;u&&("encoder_outputs"in o||(o=await this._prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:s,model_inputs:o,model_input_name:l,generation_config:t}))),u?({input_ids:d,model_inputs:o}=this._prepare_decoder_input_ids_for_generation({batch_size:o[l].dims.at(0),model_input_name:l,model_kwargs:o,decoder_start_token_id:t.decoder_start_token_id,bos_token_id:t.bos_token_id,generation_config:t})):d=o[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),f=this._get_stopping_criteria(t,r),g=o[l].dims.at(0),_=m.LogitsSampler.getSampler(t),w=new Array(g).fill(0),y=d.tolist();i&&i.put(y);let b=null,v={};for(;;){o=this.prepare_inputs_for_generation(y,o,t);const e=await this.forward(o);if(t.output_attentions&&t.return_dict_in_generate){const t=this.getAttentions(e);for(const e in t)e in v||(v[e]=[]),v[e].push(t[e])}const n=h(y,e.logits.slice(null,-1,null)),r=[];for(let e=0;e<n.dims.at(0);++e){const t=n[e],i=await _(t);for(const[t,n]of i){const i=BigInt(t);w[e]+=n,y[e].push(i),r.push([i]);break}}i&&i.put(r);if(f(y).every((e=>e))){t.return_dict_in_generate&&(b=this.getPastKeyValues(e,o.past_key_values,!1));break}o=this._update_model_kwargs_for_generation({generated_input_ids:r,outputs:e,model_inputs:o,is_encoder_decoder:u})}i&&i.end();const x=new c.Tensor("int64",y.flat(),[y.length,y[0].length]);return t.return_dict_in_generate?{sequences:x,past_key_values:b,...v}:x}getPastKeyValues(e,t,n=!0){const r=Object.create(null);for(const i in e)if(i.startsWith("present")){const a=i.replace("present","past_key_values");if(t&&i.includes("encoder"))r[a]=t[a];else{if(n&&t){const e=t[a];"gpu-buffer"===e.location&&e.dispose()}r[a]=e[i]}}return r}getAttentions(e){const t={};for(const n of["cross_attentions","encoder_attentions","decoder_attentions"])for(const r in e)r.startsWith(n)&&(n in t||(t[n]=[]),t[n].push(e[r]));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:[],i=(0,r.getKeyValueShapes)(this.config);for(const r in i)e[r]=new c.Tensor(t,n,i[r])}}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 j{}class G extends j{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 q extends V{}class U extends q{}class W extends q{async _call(e){return new ul(await super._call(e))}}class H extends q{async _call(e){return new sl(await super._call(e))}}class X extends q{async _call(e){return new ll(await super._call(e))}}class K extends q{async _call(e){return new dl(await super._call(e))}}class Q extends V{}class Y extends Q{}class J extends V{}class Z extends J{}class ee extends J{async _call(e){return new ul(await super._call(e))}}class te extends J{async _call(e){return new sl(await super._call(e))}}class ne extends J{async _call(e){return new ll(await super._call(e))}}class re extends J{async _call(e){return new dl(await super._call(e))}}class ie extends V{}class ae extends ie{}class se extends ie{async _call(e){return new ul(await super._call(e))}}class oe extends ie{async _call(e){return new sl(await super._call(e))}}class le extends ie{async _call(e){return new ll(await super._call(e))}}class ue extends ie{async _call(e){return new dl(await super._call(e))}}class de extends V{}class ce extends de{}class pe extends de{async _call(e){return new ul(await super._call(e))}}class he extends de{async _call(e){return new sl(await super._call(e))}}class me extends de{async _call(e){return new ll(await super._call(e))}}class fe extends de{async _call(e){return new dl(await super._call(e))}}class ge extends V{}class _e extends ge{}class we extends ge{async _call(e){return new ul(await super._call(e))}}class ye extends ge{async _call(e){return new sl(await super._call(e))}}class be extends ge{async _call(e){return new ll(await super._call(e))}}class ve extends ge{async _call(e){return new dl(await super._call(e))}}class xe extends V{}class Me extends xe{}class Te extends xe{async _call(e){return new ul(await super._call(e))}}class ke extends xe{async _call(e){return new sl(await super._call(e))}}class $e extends xe{async _call(e){return new ll(await super._call(e))}}class Se extends xe{async _call(e){return new dl(await super._call(e))}}class Ce extends V{}class Pe extends Ce{}class Ee extends Ce{async _call(e){return new ul(await super._call(e))}}class Fe extends Ce{async _call(e){return new sl(await super._call(e))}}class Ae extends Ce{async _call(e){return new ll(await super._call(e))}}class Ie extends Ce{async _call(e){return new dl(await super._call(e))}}class ze extends V{}class Oe extends ze{}class Be extends ze{async _call(e){return new sl(await super._call(e))}}class Le extends ze{async _call(e){return new ll(await super._call(e))}}class De extends ze{async _call(e){return new dl(await super._call(e))}}class Re extends ze{async _call(e){return new ul(await super._call(e))}}class Ne extends V{}class Ve extends Ne{}class je extends Ne{async _call(e){return new ul(await super._call(e))}}class Ge extends Ne{async _call(e){return new sl(await super._call(e))}}class qe extends Ne{async _call(e){return new ll(await super._call(e))}}class Ue extends V{}class We extends Ue{}class He extends Ue{async _call(e){return new ul(await super._call(e))}}class Xe extends Ue{async _call(e){return new sl(await super._call(e))}}class Ke extends Ue{async _call(e){return new dl(await super._call(e))}}class Qe extends V{}class Ye extends Qe{}class Je extends Qe{async _call(e){return new ul(await super._call(e))}}class Ze extends Qe{async _call(e){return new sl(await super._call(e))}}class et extends Qe{async _call(e){return new ll(await super._call(e))}}class tt extends Qe{async _call(e){return new dl(await super._call(e))}}class nt extends V{}class rt extends nt{}class it extends nt{async _call(e){return new ul(await super._call(e))}}class at extends nt{async _call(e){return new sl(await super._call(e))}}class st extends nt{async _call(e){return new dl(await super._call(e))}}class ot extends V{}class lt extends ot{}class ut extends ot{async _call(e){return new sl(await super._call(e))}}class dt extends ot{async _call(e){return new dl(await super._call(e))}}class ct extends ot{async _call(e){return new ul(await super._call(e))}}class pt extends V{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"];constructor(e,t,n){super(e,t),this.generation_config=n}}class ht extends pt{}class mt extends pt{}class ft extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class gt extends ft{}class _t extends ft{}class wt extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class yt extends wt{}class bt extends wt{}class vt extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class xt extends vt{}class Mt extends vt{}class Tt extends vt{async _call(e){return new sl(await super._call(e))}}class kt extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class $t extends kt{}class St extends kt{}class Ct extends kt{async _call(e){return new sl(await super._call(e))}}class Pt extends kt{}class Et extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class Ft extends Et{}class At extends Et{}class It extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class zt extends It{}class Ot extends It{}class Bt extends V{}class Lt extends Bt{}class Dt extends Bt{async _call(e){return new ul(await super._call(e))}}class Rt extends Bt{async _call(e){return new sl(await super._call(e))}}class Nt extends Bt{async _call(e){return new ll(await super._call(e))}}class Vt extends Bt{async _call(e){return new dl(await super._call(e))}}class jt extends V{}class Gt extends jt{}class qt extends jt{async _call(e){return new ul(await super._call(e))}}class Ut extends jt{async _call(e){return new sl(await super._call(e))}}class Wt extends jt{async _call(e){return new ll(await super._call(e))}}class Ht extends jt{async _call(e){return new dl(await super._call(e))}}class Xt extends V{}class Kt extends Xt{}class Qt extends Xt{async _call(e){return new ul(await super._call(e))}}class Yt extends Xt{async _call(e){return new sl(await super._call(e))}}class Jt extends Xt{async _call(e){return new ll(await super._call(e))}}class Zt extends Xt{async _call(e){return new dl(await super._call(e))}}class en extends V{}class tn extends en{}class nn extends en{}class rn extends V{requires_attention_mask=!1;main_input_name="input_features";forward_params=["input_features","attention_mask","decoder_input_ids","decoder_attention_mask","past_key_values"];constructor(e,t,n){super(e,t),this.generation_config=n}}class an extends rn{}class sn extends rn{_prepare_generation_config(e,t){return super._prepare_generation_config(e,t,g.WhisperGenerationConfig)}_retrieve_init_tokens(e){const t=[e.decoder_start_token_id];let n=e.language;const r=e.task;if(e.is_multilingual){n||(console.warn("No language specified - defaulting to English (en)."),n="en");const i=`<|${(0,_.whisper_language_to_code)(n)}|>`;t.push(e.lang_to_id[i]),t.push(e.task_to_id[r??"transcribe"])}else if(n||r)throw new Error("Cannot specify `task` or `language` for an English-only model. If the model is intended to be multilingual, pass `is_multilingual=true` to generate, or update the generation config.");return!e.return_timestamps&&e.no_timestamps_token_id&&t.at(-1)!==e.no_timestamps_token_id?t.push(e.no_timestamps_token_id):e.return_timestamps&&t.at(-1)===e.no_timestamps_token_id&&(console.warn("<|notimestamps|> prompt token is removed from generation_config since `return_timestamps` is set to `true`."),t.pop()),t.filter((e=>null!=e))}async generate({inputs:e=null,generation_config:t=null,logits_processor:n=null,stopping_criteria:r=null,...i}){t=this._prepare_generation_config(t,i);const a=i.decoder_input_ids??this._retrieve_init_tokens(t);if(t.return_timestamps&&(n??=new u.LogitsProcessorList,n.push(new u.WhisperTimeStampLogitsProcessor(t,a))),t.begin_suppress_tokens&&(n??=new u.LogitsProcessorList,n.push(new u.SuppressTokensAtBeginLogitsProcessor(t.begin_suppress_tokens,a.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 s=await super.generate({inputs:e,generation_config:t,logits_processor:n,decoder_input_ids:a,...i});return t.return_token_timestamps&&(s.token_timestamps=this._extract_token_timestamps(s,t.alignment_heads,t.num_frames)),s}_extract_token_timestamps(e,t,n=null,r=.02){if(!e.cross_attentions)throw new Error("Model outputs must contain cross attentions to extract timestamps. This is most likely because the model was not exported with `output_attentions=True`.");null==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 i=this.config.median_filter_width;void 0===i&&(console.warn("Model config has no `median_filter_width`, using default value of 7."),i=7);const a=e.cross_attentions,s=Array.from({length:this.config.decoder_layers},((e,t)=>(0,c.cat)(a.map((e=>e[t])),2))),l=(0,c.stack)(t.map((([e,t])=>{if(e>=s.length)throw new Error(`Layer index ${e} is out of bounds for cross attentions (length ${s.length}).`);return n?s[e].slice(null,t,null,[0,n]):s[e].slice(null,t)}))).transpose(1,0,2,3),[u,d]=(0,c.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 r=t[n],a=u[e][n][0].data,s=d[e][n][0].data;for(let e=0;e<r.dims[0];++e){let t=r[e].data;for(let e=0;e<t.length;++e)t[e]=(t[e]-s[e])/a[e];t.set((0,p.medianFilter)(t,i))}}}const m=[(0,c.mean)(h,1)],f=e.sequences.dims,g=new c.Tensor("float32",new Float32Array(f[0]*f[1]),f);for(let e=0;e<f[0];++e){const t=m[e].neg().squeeze_(0),[n,i]=(0,p.dynamic_time_warping)(t.tolist()),a=Array.from({length:n.length-1},((e,t)=>n[t+1]-n[t])),s=(0,o.mergeArrays)([1],a).map((e=>!!e)),l=[];for(let e=0;e<s.length;++e)s[e]&&l.push(i[e]*r);g[e].data.set(l,1)}return g}}class on extends V{main_input_name="pixel_values";forward_params=["pixel_values","input_ids","encoder_hidden_states","past_key_values"];constructor(e,t,n){super(e,t),this.generation_config=n}}class ln extends V{forward_params=["input_ids","pixel_values","attention_mask","position_ids","past_key_values"];constructor(e,t,n){super(e,t),this.generation_config=n}}class un extends ln{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:n,attention_mask:r}){const i=this.config.image_token_index,a=n.tolist().map((e=>e.findIndex((e=>e==i)))),s=a.every((e=>-1===e)),o=a.every((e=>-1!==e));if(!s&&!o)throw new Error("Every input should contain either 0 or 1 image token.");if(s)return{inputs_embeds:e,attention_mask:r};const l=[],u=[];for(let n=0;n<a.length;++n){const i=a[n],s=e[n],o=t[n],d=r[n];l.push((0,c.cat)([s.slice([0,i]),o,s.slice([i+1,s.dims[0]])],0)),u.push((0,c.cat)([d.slice([0,i]),(0,c.ones)([o.dims[0]]),d.slice([i+1,d.dims[0]])],0))}return{inputs_embeds:(0,c.stack)(l,0),attention_mask:(0,c.stack)(u,0)}}}class dn extends un{}class cn extends V{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";constructor(e,t,n){super(e,t),this.generation_config=n}}class pn extends cn{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:n,attention_mask:r}){return{inputs_embeds:(0,c.cat)([t,e],1),attention_mask:(0,c.cat)([(0,c.ones)(t.dims.slice(0,2)),r],1)}}async _prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:n,attention_mask:r}){if(!e&&!t)throw new Error("Either `input_ids` or `pixel_values` should be provided.");let i,a;return e&&(i=await this.encode_text({input_ids:e})),t&&(a=await this.encode_image({pixel_values:t})),i&&a?({inputs_embeds:n,attention_mask:r}=this._merge_input_ids_with_image_features({inputs_embeds:i,image_features:a,input_ids:e,attention_mask:r})):n=i||a,{inputs_embeds:n,attention_mask:r}}async forward({input_ids:e,pixel_values:t,attention_mask:n,decoder_input_ids:r,decoder_attention_mask:i,encoder_outputs:a,past_key_values:s,inputs_embeds:o,decoder_inputs_embeds:l}){if(o||({inputs_embeds:o,attention_mask:n}=await this._prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:o,attention_mask:n})),!a){let{last_hidden_state:e}=await O(this,{inputs_embeds:o,attention_mask:n});a=e}if(!l){if(!r)throw new Error("Either `decoder_input_ids` or `decoder_inputs_embeds` should be provided.");l=await this.encode_text({input_ids:r})}const u={inputs_embeds:l,attention_mask:i,encoder_attention_mask:n,encoder_hidden_states:a,past_key_values:s};return await B(this,u,!0)}}class hn extends V{}class mn extends hn{}class fn extends hn{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class gn extends hn{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class _n extends hn{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class wn extends hn{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class yn extends V{}class bn extends yn{}class vn extends yn{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class xn extends hn{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class Mn extends V{}class Tn extends Mn{}class kn extends V{}class $n extends kn{}class Sn extends kn{}class Cn extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class Pn extends Cn{}class En extends Cn{}class Fn extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class An extends Fn{}class In extends Fn{}class zn extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class On extends zn{}class Bn extends zn{}class Ln extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class Dn extends Ln{}class Rn extends Ln{}class Nn extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class Vn extends Nn{}class jn extends Nn{}class Gn extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class qn extends Gn{}class Un extends Gn{}class Wn extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class Hn extends Wn{}class Xn extends Wn{}class Kn extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class Qn extends Kn{}class Yn extends Kn{}class Jn extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class Zn extends Jn{}class er extends Jn{}class tr extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class nr extends tr{}class rr extends tr{}class ir extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class ar extends ir{}class sr extends ir{}class or extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class lr extends or{}class ur extends or{}class dr extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class cr extends dr{}class pr extends dr{}class hr extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class mr extends hr{}class fr extends hr{}class gr extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class _r extends gr{}class wr extends gr{}class yr extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class br extends yr{}class vr extends yr{}class xr extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class Mr extends xr{}class Tr extends xr{}class kr extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class $r extends kr{}class Sr extends kr{}class Cr extends V{}class Pr extends Cr{}class Er extends Cr{async _call(e){return new sl(await super._call(e))}}class Fr extends V{}class Ar extends Fr{}class Ir extends Fr{async _call(e){return new sl(await super._call(e))}}class zr extends V{}class Or extends zr{}class Br extends V{}class Lr extends Br{}class Dr extends Br{async _call(e){return new sl(await super._call(e))}}class Rr extends V{}class Nr extends Rr{}class Vr extends V{}class jr extends Vr{}class Gr extends Vr{async _call(e){return new sl(await super._call(e))}}class qr extends V{}class Ur extends qr{async _call(e){return new hl(await super._call(e))}}class Wr extends V{}class Hr extends Wr{}class Xr extends Wr{async _call(e){return new sl(await super._call(e))}}class Kr extends V{}class Qr extends Kr{}class Yr extends Kr{async _call(e){return new sl(await super._call(e))}}class Jr extends V{}class Zr extends Jr{}class ei extends Jr{}class ti extends V{}class ni extends ti{}class ri extends ti{}class ii extends V{}class ai extends ii{}class si extends ii{async _call(e){return new sl(await super._call(e))}}class oi extends V{}class li extends oi{}class ui extends oi{async _call(e){return new ci(await super._call(e))}}class di extends oi{async _call(e){return new pi(await super._call(e))}}class ci extends j{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class pi extends j{constructor({logits:e,pred_boxes:t,pred_masks:n}){super(),this.logits=e,this.pred_boxes=t,this.pred_masks=n}}class hi extends V{}class mi extends hi{}class fi extends hi{async _call(e){return new gi(await super._call(e))}}class gi extends j{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class _i extends V{}class wi extends _i{}class yi extends _i{async _call(e){return new bi(await super._call(e))}}class bi extends ci{}class vi extends V{}class xi extends vi{}class Mi extends vi{async _call(e){return new sl(await super._call(e))}}class Ti extends V{}class ki extends Ti{}class $i extends Ti{async _call(e){return new sl(await super._call(e))}}class Si extends V{}class Ci extends Si{}class Pi extends Si{async _call(e){return new sl(await super._call(e))}}class Ei extends V{}class Fi extends Ei{}class Ai extends Ei{async _call(e){return new sl(await super._call(e))}}class Ii extends V{}class zi extends Ii{}class Oi extends Ii{}class Bi extends V{}class Li extends Bi{}class Di extends Bi{}class Ri extends V{}class Ni extends Ri{}class Vi extends V{}class ji extends Vi{}class Gi extends Vi{}class qi extends Vi{}class Ui extends V{}class Wi extends Ui{}class Hi extends Ui{}class Xi extends V{}class Ki extends Xi{}class Qi extends Xi{}class Yi extends V{}class Ji extends Yi{}class Zi extends V{}class ea extends Zi{}class ta extends Zi{async _call(e){return new sl(await super._call(e))}}class na extends V{}class ra extends na{}class ia extends na{async _call(e){return new sl(await super._call(e))}}class aa extends V{}class sa extends aa{}class oa extends aa{async _call(e){return new sl(await super._call(e))}}class la extends V{}class ua extends la{}class da extends la{async _call(e){return new ca(await super._call(e))}}class ca extends j{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class pa extends V{}class ha extends pa{async get_image_embeddings({pixel_values:e}){return await O(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 c.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 ma(await super._call(e))}}class ma extends j{constructor({iou_scores:e,pred_masks:t}){super(),this.iou_scores=e,this.pred_masks=t}}class fa extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class ga extends fa{}class _a extends fa{}class wa extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class ya extends wa{}class ba extends wa{}class va extends V{}class xa extends va{}class Ma extends va{async _call(e){return new cl(await super._call(e))}}class Ta extends va{async _call(e){return new sl(await super._call(e))}}class ka extends va{async _call(e){return new ll(await super._call(e))}}class $a extends V{}class Sa extends $a{}class Ca extends $a{async _call(e){return new ll(await super._call(e))}}class Pa extends V{}class Ea extends Pa{}class Fa extends V{}class Aa extends Fa{}class Ia extends Fa{async _call(e){return new cl(await super._call(e))}}class za extends Fa{async _call(e){return new sl(await super._call(e))}}class Oa extends V{}class Ba extends Oa{}class La extends Oa{async _call(e){return new cl(await super._call(e))}}class Da extends Oa{async _call(e){return new sl(await super._call(e))}}class Ra extends Oa{async _call(e){return new ll(await super._call(e))}}class Na extends V{}class Va extends Na{}class ja extends Na{async _call(e){return new cl(await super._call(e))}}class Ga extends Na{async _call(e){return new sl(await super._call(e))}}class qa extends V{}class Ua extends va{}class Wa extends va{async _call(e){return new cl(await super._call(e))}}class Ha extends va{async _call(e){return new sl(await super._call(e))}}class Xa extends V{}class Ka extends Xa{}class Qa extends Xa{async _call(e){return new cl(await super._call(e))}}class Ya extends Xa{async _call(e){return new sl(await super._call(e))}}class Ja extends Xa{async _call(e){return new ol(await super._call(e))}}class Za extends Xa{async _call(e){return new ll(await super._call(e))}}class es extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class ts extends es{}class ns extends es{}class rs extends es{async generate_speech(e,t,{threshold:n=.5,minlenratio:r=0,maxlenratio:i=20,vocoder:a=null}={}){const s={input_ids:e},{encoder_outputs:o,encoder_attention_mask:l}=await O(this,s),u=o.dims[1]/this.config.reduction_factor,d=Math.floor(u*i),p=Math.floor(u*r),h=this.config.num_mel_bins;let m=[],f=null,g=null,_=0;for(;;){++_;const e=I(!!g);let r;r=g?g.output_sequence_out:new c.Tensor("float32",new Float32Array(h),[1,1,h]);let i={use_cache_branch:e,output_sequence:r,encoder_attention_mask:l,speaker_embeddings:t,encoder_hidden_states:o};this.addPastKeyValues(i,f),g=await E(this.sessions.decoder_model_merged,i),f=this.getPastKeyValues(g,f);const{prob:a,spectrum:s}=g;if(m.push(s),_>=p&&(Array.from(a.data).filter((e=>e>=n)).length>0||_>=d))break}const w=(0,c.cat)(m),{waveform:y}=await E(a.sessions.model,{spectrogram:w});return{spectrogram:w,waveform:y}}}class is extends V{main_input_name="spectrogram"}class as extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class ss extends as{}class os extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class ls extends os{}class us extends os{}class ds extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class cs extends ds{}class ps extends ds{}class hs extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class ms extends hs{}class fs extends hs{}class gs extends V{}class _s extends gs{}class ws extends gs{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class ys extends gs{static async from_pretrained(e,t={}){return t.model_file_name??="audio_model",super.from_pretrained(e,t)}}class bs extends V{}class vs extends bs{async _call(e){return new ml(await super._call(e))}}class xs extends V{}class Ms extends xs{}class Ts extends xs{}class ks extends xs{}class $s extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class Ss extends $s{}class Cs extends $s{}class Ps extends V{}class Es extends Ps{}class Fs extends Ps{async _call(e){return new sl(await super._call(e))}}class As extends V{}class Is extends As{}class zs extends As{}class Os extends V{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"];constructor(e,t,n){super(e,t),this.generation_config=n}_apply_and_filter_by_delay_pattern_mask(e){const[t,n]=e.dims,r=this.config.decoder.num_codebooks,i=n-r;let a=0;for(let t=0;t<e.size;++t){if(e.data[t]===this.config.decoder.pad_token_id)continue;const s=t%n-Math.floor(t/n)%r;s>0&&s<=i&&(e.data[a++]=e.data[t])}const s=Math.floor(t/r),o=a/(s*r);return new c.Tensor(e.type,e.data.slice(0,a),[s,r,o])}prepare_inputs_for_generation(e,t,n){let r=structuredClone(e);for(let e=0;e<r.length;++e)for(let t=0;t<r[e].length;++t)e%this.config.decoder.num_codebooks>=t&&(r[e][t]=BigInt(this.config.decoder.pad_token_id));null!==n.guidance_scale&&n.guidance_scale>1&&(r=r.concat(r));return super.prepare_inputs_for_generation(r,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:r}=await E(this.sessions.encodec_decode,{audio_codes:n});return r}}class Bs extends V{}class Ls extends Bs{}class Ds extends Bs{async _call(e){return new sl(await super._call(e))}}class Rs extends V{}class Ns extends Rs{}class Vs extends Rs{async _call(e){return new sl(await super._call(e))}}class js extends V{}class Gs extends js{}class qs extends js{async _call(e){return new sl(await super._call(e))}}class Us extends V{}class Ws extends Us{}class Hs extends Us{async _call(e){return new sl(await super._call(e))}}class Xs extends V{}class Ks extends Xs{}class Qs{static MODEL_CLASS_MAPPINGS=null;static BASE_IF_FAIL=!1;static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:i=null,local_files_only:a=!1,revision:s="main",model_file_name:o=null,subfolder:l="onnx",device:u=null,dtype:d=null,use_external_data_format:c=null,session_options:p={}}={}){const h={progress_callback:t,config:n,cache_dir:i,local_files_only:a,revision:s,model_file_name:o,subfolder:l,device:u,dtype:d,use_external_data_format:c,session_options:p};if(h.config=await r.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 V.from_pretrained(e,h);throw Error(`Unsupported model type: ${h.config.model_type}`)}}const Ys=new Map([["bert",["BertModel",U]],["nomic_bert",["NomicBertModel",Y]],["roformer",["RoFormerModel",Z]],["electra",["ElectraModel",ce]],["esm",["EsmModel",Ve]],["convbert",["ConvBertModel",ae]],["camembert",["CamembertModel",_e]],["deberta",["DebertaModel",Me]],["deberta-v2",["DebertaV2Model",Pe]],["mpnet",["MPNetModel",Ye]],["albert",["AlbertModel",lt]],["distilbert",["DistilBertModel",Oe]],["roberta",["RobertaModel",Lt]],["xlm",["XLMModel",Gt]],["xlm-roberta",["XLMRobertaModel",Kt]],["clap",["ClapModel",_s]],["clip",["CLIPModel",mn]],["clipseg",["CLIPSegModel",$n]],["chinese_clip",["ChineseCLIPModel",Tn]],["siglip",["SiglipModel",bn]],["mobilebert",["MobileBertModel",We]],["squeezebert",["SqueezeBertModel",rt]],["wav2vec2",["Wav2Vec2Model",xa]],["wav2vec2-bert",["Wav2Vec2BertModel",Va]],["unispeech",["UniSpeechModel",Aa]],["unispeech-sat",["UniSpeechSatModel",Ba]],["hubert",["HubertModel",Ua]],["wavlm",["WavLMModel",Ka]],["audio-spectrogram-transformer",["ASTModel",tn]],["vits",["VitsModel",vs]],["pyannote",["PyAnnoteModel",Sa]],["wespeaker-resnet",["WeSpeakerResNetModel",Ea]],["detr",["DetrModel",li]],["rt_detr",["RTDetrModel",mi]],["table-transformer",["TableTransformerModel",wi]],["vit",["ViTModel",Pr]],["pvt",["PvtModel",Ar]],["vit_msn",["ViTMSNModel",Lr]],["vit_mae",["ViTMAEModel",Or]],["groupvit",["GroupViTModel",Nr]],["fastvit",["FastViTModel",jr]],["mobilevit",["MobileViTModel",Hr]],["mobilevitv2",["MobileViTV2Model",Qr]],["owlvit",["OwlViTModel",Zr]],["owlv2",["Owlv2Model",ni]],["beit",["BeitModel",ai]],["deit",["DeiTModel",xi]],["hiera",["HieraModel",ki]],["convnext",["ConvNextModel",ea]],["convnextv2",["ConvNextV2Model",ra]],["dinov2",["Dinov2Model",sa]],["resnet",["ResNetModel",Ci]],["swin",["SwinModel",Fi]],["swin2sr",["Swin2SRModel",zi]],["donut-swin",["DonutSwinModel",Ji]],["yolos",["YolosModel",ua]],["dpt",["DPTModel",Li]],["glpn",["GLPNModel",Ki]],["hifigan",["SpeechT5HifiGan",is]],["efficientnet",["EfficientNetModel",Es]],["decision_transformer",["DecisionTransformerModel",Ks]],["mobilenet_v1",["MobileNetV1Model",Ls]],["mobilenet_v2",["MobileNetV2Model",Ns]],["mobilenet_v3",["MobileNetV3Model",Gs]],["mobilenet_v4",["MobileNetV4Model",Ws]],["maskformer",["MaskFormerModel",Wi]]]),Js=new Map([["t5",["T5Model",ht]],["longt5",["LongT5Model",gt]],["mt5",["MT5Model",yt]],["bart",["BartModel",xt]],["mbart",["MBartModel",$t]],["marian",["MarianModel",ga]],["whisper",["WhisperModel",an]],["m2m_100",["M2M100Model",ya]],["blenderbot",["BlenderbotModel",Ft]],["blenderbot-small",["BlenderbotSmallModel",zt]]]),Zs=new Map([["bloom",["BloomModel",br]],["jais",["JAISModel",An]],["gpt2",["GPT2Model",Pn]],["gptj",["GPTJModel",Vn]],["gpt_bigcode",["GPTBigCodeModel",qn]],["gpt_neo",["GPTNeoModel",On]],["gpt_neox",["GPTNeoXModel",Dn]],["codegen",["CodeGenModel",Hn]],["llama",["LlamaModel",Qn]],["cohere",["CohereModel",Zn]],["gemma",["GemmaModel",nr]],["gemma2",["Gemma2Model",ar]],["openelm",["OpenELMModel",lr]],["qwen2",["Qwen2Model",cr]],["phi",["PhiModel",mr]],["phi3",["Phi3Model",_r]],["mpt",["MptModel",Mr]],["opt",["OPTModel",$r]],["mistral",["MistralModel",ls]],["starcoder2",["Starcoder2Model",cs]],["falcon",["FalconModel",ms]],["stablelm",["StableLmModel",Ss]]]),eo=new Map([["speecht5",["SpeechT5ForSpeechToText",ns]],["whisper",["WhisperForConditionalGeneration",sn]]]),to=new Map([["speecht5",["SpeechT5ForTextToSpeech",rs]]]),no=new Map([["vits",["VitsModel",vs]],["musicgen",["MusicgenForConditionalGeneration",Os]]]),ro=new Map([["bert",["BertForSequenceClassification",H]],["roformer",["RoFormerForSequenceClassification",te]],["electra",["ElectraForSequenceClassification",he]],["esm",["EsmForSequenceClassification",Ge]],["convbert",["ConvBertForSequenceClassification",oe]],["camembert",["CamembertForSequenceClassification",ye]],["deberta",["DebertaForSequenceClassification",ke]],["deberta-v2",["DebertaV2ForSequenceClassification",Fe]],["mpnet",["MPNetForSequenceClassification",Ze]],["albert",["AlbertForSequenceClassification",ut]],["distilbert",["DistilBertForSequenceClassification",Be]],["roberta",["RobertaForSequenceClassification",Rt]],["xlm",["XLMForSequenceClassification",Ut]],["xlm-roberta",["XLMRobertaForSequenceClassification",Yt]],["bart",["BartForSequenceClassification",Tt]],["mbart",["MBartForSequenceClassification",Ct]],["mobilebert",["MobileBertForSequenceClassification",Xe]],["squeezebert",["SqueezeBertForSequenceClassification",at]]]),io=new Map([["bert",["BertForTokenClassification",X]],["roformer",["RoFormerForTokenClassification",ne]],["electra",["ElectraForTokenClassification",me]],["esm",["EsmForTokenClassification",qe]],["convbert",["ConvBertForTokenClassification",le]],["camembert",["CamembertForTokenClassification",be]],["deberta",["DebertaForTokenClassification",$e]],["deberta-v2",["DebertaV2ForTokenClassification",Ae]],["mpnet",["MPNetForTokenClassification",et]],["distilbert",["DistilBertForTokenClassification",Le]],["roberta",["RobertaForTokenClassification",Nt]],["xlm",["XLMForTokenClassification",Wt]],["xlm-roberta",["XLMRobertaForTokenClassification",Jt]]]),ao=new Map([["t5",["T5ForConditionalGeneration",mt]],["longt5",["LongT5ForConditionalGeneration",_t]],["mt5",["MT5ForConditionalGeneration",bt]],["bart",["BartForConditionalGeneration",Mt]],["mbart",["MBartForConditionalGeneration",St]],["marian",["MarianMTModel",_a]],["m2m_100",["M2M100ForConditionalGeneration",ba]],["blenderbot",["BlenderbotForConditionalGeneration",At]],["blenderbot-small",["BlenderbotSmallForConditionalGeneration",Ot]]]),so=new Map([["bloom",["BloomForCausalLM",vr]],["gpt2",["GPT2LMHeadModel",En]],["jais",["JAISLMHeadModel",In]],["gptj",["GPTJForCausalLM",jn]],["gpt_bigcode",["GPTBigCodeForCausalLM",Un]],["gpt_neo",["GPTNeoForCausalLM",Bn]],["gpt_neox",["GPTNeoXForCausalLM",Rn]],["codegen",["CodeGenForCausalLM",Xn]],["llama",["LlamaForCausalLM",Yn]],["cohere",["CohereForCausalLM",er]],["gemma",["GemmaForCausalLM",rr]],["gemma2",["Gemma2ForCausalLM",sr]],["openelm",["OpenELMForCausalLM",ur]],["qwen2",["Qwen2ForCausalLM",pr]],["phi",["PhiForCausalLM",fr]],["phi3",["Phi3ForCausalLM",wr]],["mpt",["MptForCausalLM",Tr]],["opt",["OPTForCausalLM",Sr]],["mbart",["MBartForCausalLM",Pt]],["mistral",["MistralForCausalLM",us]],["starcoder2",["Starcoder2ForCausalLM",ps]],["falcon",["FalconForCausalLM",fs]],["trocr",["TrOCRForCausalLM",ss]],["stablelm",["StableLmForCausalLM",Cs]]]),oo=new Map([["bert",["BertForMaskedLM",W]],["roformer",["RoFormerForMaskedLM",ee]],["electra",["ElectraForMaskedLM",pe]],["esm",["EsmForMaskedLM",je]],["convbert",["ConvBertForMaskedLM",se]],["camembert",["CamembertForMaskedLM",we]],["deberta",["DebertaForMaskedLM",Te]],["deberta-v2",["DebertaV2ForMaskedLM",Ee]],["mpnet",["MPNetForMaskedLM",Je]],["albert",["AlbertForMaskedLM",ct]],["distilbert",["DistilBertForMaskedLM",Re]],["roberta",["RobertaForMaskedLM",Dt]],["xlm",["XLMWithLMHeadModel",qt]],["xlm-roberta",["XLMRobertaForMaskedLM",Qt]],["mobilebert",["MobileBertForMaskedLM",He]],["squeezebert",["SqueezeBertForMaskedLM",it]]]),lo=new Map([["bert",["BertForQuestionAnswering",K]],["roformer",["RoFormerForQuestionAnswering",re]],["electra",["ElectraForQuestionAnswering",fe]],["convbert",["ConvBertForQuestionAnswering",ue]],["camembert",["CamembertForQuestionAnswering",ve]],["deberta",["DebertaForQuestionAnswering",Se]],["deberta-v2",["DebertaV2ForQuestionAnswering",Ie]],["mpnet",["MPNetForQuestionAnswering",tt]],["albert",["AlbertForQuestionAnswering",dt]],["distilbert",["DistilBertForQuestionAnswering",De]],["roberta",["RobertaForQuestionAnswering",Vt]],["xlm",["XLMForQuestionAnswering",Ht]],["xlm-roberta",["XLMRobertaForQuestionAnswering",Zt]],["mobilebert",["MobileBertForQuestionAnswering",Ke]],["squeezebert",["SqueezeBertForQuestionAnswering",st]]]),uo=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",on]]]),co=new Map([["llava",["LlavaForConditionalGeneration",un]],["moondream1",["Moondream1ForConditionalGeneration",dn]],["florence2",["Florence2ForConditionalGeneration",pn]]]),po=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",on]]]),ho=new Map([["vit",["ViTForImageClassification",Er]],["pvt",["PvtForImageClassification",Ir]],["vit_msn",["ViTMSNForImageClassification",Dr]],["fastvit",["FastViTForImageClassification",Gr]],["mobilevit",["MobileViTForImageClassification",Xr]],["mobilevitv2",["MobileViTV2ForImageClassification",Yr]],["beit",["BeitForImageClassification",si]],["deit",["DeiTForImageClassification",Mi]],["hiera",["HieraForImageClassification",$i]],["convnext",["ConvNextForImageClassification",ta]],["convnextv2",["ConvNextV2ForImageClassification",ia]],["dinov2",["Dinov2ForImageClassification",oa]],["resnet",["ResNetForImageClassification",Pi]],["swin",["SwinForImageClassification",Ai]],["segformer",["SegformerForImageClassification",Ts]],["efficientnet",["EfficientNetForImageClassification",Fs]],["mobilenet_v1",["MobileNetV1ForImageClassification",Ds]],["mobilenet_v2",["MobileNetV2ForImageClassification",Vs]],["mobilenet_v3",["MobileNetV3ForImageClassification",qs]],["mobilenet_v4",["MobileNetV4ForImageClassification",Hs]]]),mo=new Map([["detr",["DetrForObjectDetection",ui]],["rt_detr",["RTDetrForObjectDetection",fi]],["table-transformer",["TableTransformerForObjectDetection",yi]],["yolos",["YolosForObjectDetection",da]]]),fo=new Map([["owlvit",["OwlViTForObjectDetection",ei]],["owlv2",["Owlv2ForObjectDetection",ri]]]),go=new Map([["detr",["DetrForSegmentation",di]],["clipseg",["CLIPSegForImageSegmentation",Sn]]]),_o=new Map([["segformer",["SegformerForSemanticSegmentation",ks]],["sapiens",["SapiensForSemanticSegmentation",ji]]]),wo=new Map([["detr",["DetrForSegmentation",di]],["maskformer",["MaskFormerForInstanceSegmentation",Hi]]]),yo=new Map([["sam",["SamModel",ha]]]),bo=new Map([["wav2vec2",["Wav2Vec2ForCTC",Ma]],["wav2vec2-bert",["Wav2Vec2BertForCTC",ja]],["unispeech",["UniSpeechForCTC",Ia]],["unispeech-sat",["UniSpeechSatForCTC",La]],["wavlm",["WavLMForCTC",Qa]],["hubert",["HubertForCTC",Wa]]]),vo=new Map([["wav2vec2",["Wav2Vec2ForSequenceClassification",Ta]],["wav2vec2-bert",["Wav2Vec2BertForSequenceClassification",Ga]],["unispeech",["UniSpeechForSequenceClassification",za]],["unispeech-sat",["UniSpeechSatForSequenceClassification",Da]],["wavlm",["WavLMForSequenceClassification",Ya]],["hubert",["HubertForSequenceClassification",Ha]],["audio-spectrogram-transformer",["ASTForAudioClassification",nn]]]),xo=new Map([["wavlm",["WavLMForXVector",Ja]]]),Mo=new Map([["unispeech-sat",["UniSpeechSatForAudioFrameClassification",Ra]],["wavlm",["WavLMForAudioFrameClassification",Za]],["wav2vec2",["Wav2Vec2ForAudioFrameClassification",ka]],["pyannote",["PyAnnoteForAudioFrameClassification",Ca]]]),To=new Map([["vitmatte",["VitMatteForImageMatting",Ur]]]),ko=new Map([["swin2sr",["Swin2SRForImageSuperResolution",Oi]]]),$o=new Map([["dpt",["DPTForDepthEstimation",Di]],["depth_anything",["DepthAnythingForDepthEstimation",Ni]],["glpn",["GLPNForDepthEstimation",Qi]],["sapiens",["SapiensForDepthEstimation",Gi]]]),So=new Map([["sapiens",["SapiensForNormalEstimation",qi]]]),Co=new Map([["clip",["CLIPVisionModelWithProjection",wn]],["siglip",["SiglipVisionModel",xn]]]),Po=[[Ys,w],[Js,y],[Zs,x],[ro,w],[io,w],[ao,b],[eo,b],[so,x],[oo,w],[lo,w],[uo,v],[co,T],[ho,w],[go,w],[wo,w],[_o,w],[To,w],[ko,w],[$o,w],[So,w],[mo,w],[fo,w],[yo,M],[bo,w],[vo,w],[to,b],[no,w],[xo,w],[Mo,w],[Co,w]];for(const[e,t]of Po)for(const[n,r]of e.values())$.set(n,t),C.set(r,n),S.set(n,r);const Eo=[["MusicgenForConditionalGeneration",Os,k],["CLIPTextModelWithProjection",gn,w],["SiglipTextModel",vn,w],["ClapTextModelWithProjection",ws,w],["ClapAudioModelWithProjection",ys,w]];for(const[e,t,n]of Eo)$.set(e,n),C.set(t,e),S.set(e,t);class Fo extends Qs{static MODEL_CLASS_MAPPINGS=Po.map((e=>e[0]));static BASE_IF_FAIL=!0}class Ao extends Qs{static MODEL_CLASS_MAPPINGS=[ro]}class Io extends Qs{static MODEL_CLASS_MAPPINGS=[io]}class zo extends Qs{static MODEL_CLASS_MAPPINGS=[ao]}class Oo extends Qs{static MODEL_CLASS_MAPPINGS=[eo]}class Bo extends Qs{static MODEL_CLASS_MAPPINGS=[to]}class Lo extends Qs{static MODEL_CLASS_MAPPINGS=[no]}class Do extends Qs{static MODEL_CLASS_MAPPINGS=[so]}class Ro extends Qs{static MODEL_CLASS_MAPPINGS=[oo]}class No extends Qs{static MODEL_CLASS_MAPPINGS=[lo]}class Vo extends Qs{static MODEL_CLASS_MAPPINGS=[uo]}class jo extends Qs{static MODEL_CLASS_MAPPINGS=[ho]}class Go extends Qs{static MODEL_CLASS_MAPPINGS=[go]}class qo extends Qs{static MODEL_CLASS_MAPPINGS=[_o]}class Uo extends Qs{static MODEL_CLASS_MAPPINGS=[wo]}class Wo extends Qs{static MODEL_CLASS_MAPPINGS=[mo]}class Ho extends Qs{static MODEL_CLASS_MAPPINGS=[fo]}class Xo extends Qs{static MODEL_CLASS_MAPPINGS=[yo]}class Ko extends Qs{static MODEL_CLASS_MAPPINGS=[bo]}class Qo extends Qs{static MODEL_CLASS_MAPPINGS=[vo]}class Yo extends Qs{static MODEL_CLASS_MAPPINGS=[xo]}class Jo extends Qs{static MODEL_CLASS_MAPPINGS=[Mo]}class Zo extends Qs{static MODEL_CLASS_MAPPINGS=[po]}class el extends Qs{static MODEL_CLASS_MAPPINGS=[To]}class tl extends Qs{static MODEL_CLASS_MAPPINGS=[ko]}class nl extends Qs{static MODEL_CLASS_MAPPINGS=[$o]}class rl extends Qs{static MODEL_CLASS_MAPPINGS=[So]}class il extends Qs{static MODEL_CLASS_MAPPINGS=[Co]}class al extends j{constructor({logits:e,past_key_values:t,encoder_outputs:n,decoder_attentions:r=null,cross_attentions:i=null}){super(),this.logits=e,this.past_key_values=t,this.encoder_outputs=n,this.decoder_attentions=r,this.cross_attentions=i}}class sl extends j{constructor({logits:e}){super(),this.logits=e}}class ol extends j{constructor({logits:e,embeddings:t}){super(),this.logits=e,this.embeddings=t}}class ll extends j{constructor({logits:e}){super(),this.logits=e}}class ul extends j{constructor({logits:e}){super(),this.logits=e}}class dl extends j{constructor({start_logits:e,end_logits:t}){super(),this.start_logits=e,this.end_logits=t}}class cl extends j{constructor({logits:e}){super(),this.logits=e}}class pl extends j{constructor({logits:e,past_key_values:t}){super(),this.logits=e,this.past_key_values=t}}class hl extends j{constructor({alphas:e}){super(),this.alphas=e}}class ml extends j{constructor({waveform:e,spectrogram:t}){super(),this.waveform=e,this.spectrogram=t}}},"./src/models/whisper/common_whisper.js":
+  \***********************/(e,t,n)=>{n.r(t),n.d(t,{ASTForAudioClassification:()=>nn,ASTModel:()=>tn,ASTPreTrainedModel:()=>en,AlbertForMaskedLM:()=>ct,AlbertForQuestionAnswering:()=>dt,AlbertForSequenceClassification:()=>ut,AlbertModel:()=>lt,AlbertPreTrainedModel:()=>ot,AutoModel:()=>Fo,AutoModelForAudioClassification:()=>Qo,AutoModelForAudioFrameClassification:()=>Jo,AutoModelForCTC:()=>Ko,AutoModelForCausalLM:()=>Do,AutoModelForDepthEstimation:()=>nl,AutoModelForDocumentQuestionAnswering:()=>Zo,AutoModelForImageClassification:()=>jo,AutoModelForImageFeatureExtraction:()=>il,AutoModelForImageMatting:()=>el,AutoModelForImageSegmentation:()=>Go,AutoModelForImageToImage:()=>tl,AutoModelForMaskGeneration:()=>Xo,AutoModelForMaskedLM:()=>Ro,AutoModelForNormalEstimation:()=>rl,AutoModelForObjectDetection:()=>Wo,AutoModelForQuestionAnswering:()=>No,AutoModelForSemanticSegmentation:()=>qo,AutoModelForSeq2SeqLM:()=>zo,AutoModelForSequenceClassification:()=>Ao,AutoModelForSpeechSeq2Seq:()=>Oo,AutoModelForTextToSpectrogram:()=>Bo,AutoModelForTextToWaveform:()=>Lo,AutoModelForTokenClassification:()=>Io,AutoModelForUniversalSegmentation:()=>Uo,AutoModelForVision2Seq:()=>Vo,AutoModelForXVector:()=>Yo,AutoModelForZeroShotObjectDetection:()=>Ho,BartForConditionalGeneration:()=>Mt,BartForSequenceClassification:()=>Tt,BartModel:()=>xt,BartPretrainedModel:()=>vt,BaseModelOutput:()=>G,BeitForImageClassification:()=>si,BeitModel:()=>ai,BeitPreTrainedModel:()=>ii,BertForMaskedLM:()=>W,BertForQuestionAnswering:()=>K,BertForSequenceClassification:()=>H,BertForTokenClassification:()=>X,BertModel:()=>U,BertPreTrainedModel:()=>q,BlenderbotForConditionalGeneration:()=>At,BlenderbotModel:()=>Ft,BlenderbotPreTrainedModel:()=>Et,BlenderbotSmallForConditionalGeneration:()=>Ot,BlenderbotSmallModel:()=>zt,BlenderbotSmallPreTrainedModel:()=>It,BloomForCausalLM:()=>vr,BloomModel:()=>br,BloomPreTrainedModel:()=>yr,CLIPModel:()=>mn,CLIPPreTrainedModel:()=>hn,CLIPSegForImageSegmentation:()=>Sn,CLIPSegModel:()=>$n,CLIPSegPreTrainedModel:()=>kn,CLIPTextModel:()=>fn,CLIPTextModelWithProjection:()=>gn,CLIPVisionModel:()=>_n,CLIPVisionModelWithProjection:()=>wn,CamembertForMaskedLM:()=>we,CamembertForQuestionAnswering:()=>ve,CamembertForSequenceClassification:()=>ye,CamembertForTokenClassification:()=>be,CamembertModel:()=>_e,CamembertPreTrainedModel:()=>ge,CausalLMOutput:()=>cl,CausalLMOutputWithPast:()=>pl,ChineseCLIPModel:()=>Tn,ChineseCLIPPreTrainedModel:()=>Mn,ClapAudioModelWithProjection:()=>ys,ClapModel:()=>_s,ClapPreTrainedModel:()=>gs,ClapTextModelWithProjection:()=>ws,CodeGenForCausalLM:()=>Xn,CodeGenModel:()=>Hn,CodeGenPreTrainedModel:()=>Wn,CohereForCausalLM:()=>er,CohereModel:()=>Zn,CoherePreTrainedModel:()=>Jn,ConvBertForMaskedLM:()=>se,ConvBertForQuestionAnswering:()=>ue,ConvBertForSequenceClassification:()=>oe,ConvBertForTokenClassification:()=>le,ConvBertModel:()=>ae,ConvBertPreTrainedModel:()=>ie,ConvNextForImageClassification:()=>ta,ConvNextModel:()=>ea,ConvNextPreTrainedModel:()=>Zi,ConvNextV2ForImageClassification:()=>ia,ConvNextV2Model:()=>ra,ConvNextV2PreTrainedModel:()=>na,DPTForDepthEstimation:()=>Di,DPTModel:()=>Li,DPTPreTrainedModel:()=>Bi,DebertaForMaskedLM:()=>Te,DebertaForQuestionAnswering:()=>Se,DebertaForSequenceClassification:()=>ke,DebertaForTokenClassification:()=>$e,DebertaModel:()=>Me,DebertaPreTrainedModel:()=>xe,DebertaV2ForMaskedLM:()=>Ee,DebertaV2ForQuestionAnswering:()=>Ie,DebertaV2ForSequenceClassification:()=>Fe,DebertaV2ForTokenClassification:()=>Ae,DebertaV2Model:()=>Pe,DebertaV2PreTrainedModel:()=>Ce,DecisionTransformerModel:()=>Ks,DecisionTransformerPreTrainedModel:()=>Xs,DeiTForImageClassification:()=>Mi,DeiTModel:()=>xi,DeiTPreTrainedModel:()=>vi,DepthAnythingForDepthEstimation:()=>Ni,DepthAnythingPreTrainedModel:()=>Ri,DetrForObjectDetection:()=>ui,DetrForSegmentation:()=>di,DetrModel:()=>li,DetrObjectDetectionOutput:()=>ci,DetrPreTrainedModel:()=>oi,DetrSegmentationOutput:()=>pi,Dinov2ForImageClassification:()=>oa,Dinov2Model:()=>sa,Dinov2PreTrainedModel:()=>aa,DistilBertForMaskedLM:()=>Re,DistilBertForQuestionAnswering:()=>De,DistilBertForSequenceClassification:()=>Be,DistilBertForTokenClassification:()=>Le,DistilBertModel:()=>Oe,DistilBertPreTrainedModel:()=>ze,DonutSwinModel:()=>Ji,DonutSwinPreTrainedModel:()=>Yi,EfficientNetForImageClassification:()=>Fs,EfficientNetModel:()=>Es,EfficientNetPreTrainedModel:()=>Ps,ElectraForMaskedLM:()=>pe,ElectraForQuestionAnswering:()=>fe,ElectraForSequenceClassification:()=>he,ElectraForTokenClassification:()=>me,ElectraModel:()=>ce,ElectraPreTrainedModel:()=>de,EsmForMaskedLM:()=>je,EsmForSequenceClassification:()=>Ge,EsmForTokenClassification:()=>qe,EsmModel:()=>Ve,EsmPreTrainedModel:()=>Ne,FalconForCausalLM:()=>fs,FalconModel:()=>ms,FalconPreTrainedModel:()=>hs,FastViTForImageClassification:()=>Gr,FastViTModel:()=>jr,FastViTPreTrainedModel:()=>Vr,Florence2ForConditionalGeneration:()=>pn,Florence2PreTrainedModel:()=>cn,GLPNForDepthEstimation:()=>Qi,GLPNModel:()=>Ki,GLPNPreTrainedModel:()=>Xi,GPT2LMHeadModel:()=>En,GPT2Model:()=>Pn,GPT2PreTrainedModel:()=>Cn,GPTBigCodeForCausalLM:()=>Un,GPTBigCodeModel:()=>qn,GPTBigCodePreTrainedModel:()=>Gn,GPTJForCausalLM:()=>jn,GPTJModel:()=>Vn,GPTJPreTrainedModel:()=>Nn,GPTNeoForCausalLM:()=>Bn,GPTNeoModel:()=>On,GPTNeoPreTrainedModel:()=>zn,GPTNeoXForCausalLM:()=>Rn,GPTNeoXModel:()=>Dn,GPTNeoXPreTrainedModel:()=>Ln,Gemma2ForCausalLM:()=>sr,Gemma2Model:()=>ar,Gemma2PreTrainedModel:()=>ir,GemmaForCausalLM:()=>rr,GemmaModel:()=>nr,GemmaPreTrainedModel:()=>tr,GroupViTModel:()=>Nr,GroupViTPreTrainedModel:()=>Rr,HieraForImageClassification:()=>$i,HieraModel:()=>ki,HieraPreTrainedModel:()=>Ti,HubertForCTC:()=>Wa,HubertForSequenceClassification:()=>Ha,HubertModel:()=>Ua,HubertPreTrainedModel:()=>qa,ImageMattingOutput:()=>hl,JAISLMHeadModel:()=>In,JAISModel:()=>An,JAISPreTrainedModel:()=>Fn,LlamaForCausalLM:()=>Yn,LlamaModel:()=>Qn,LlamaPreTrainedModel:()=>Kn,LlavaForConditionalGeneration:()=>un,LlavaPreTrainedModel:()=>ln,LongT5ForConditionalGeneration:()=>_t,LongT5Model:()=>gt,LongT5PreTrainedModel:()=>ft,M2M100ForConditionalGeneration:()=>ba,M2M100Model:()=>ya,M2M100PreTrainedModel:()=>wa,MBartForCausalLM:()=>Pt,MBartForConditionalGeneration:()=>St,MBartForSequenceClassification:()=>Ct,MBartModel:()=>$t,MBartPreTrainedModel:()=>kt,MPNetForMaskedLM:()=>Je,MPNetForQuestionAnswering:()=>tt,MPNetForSequenceClassification:()=>Ze,MPNetForTokenClassification:()=>et,MPNetModel:()=>Ye,MPNetPreTrainedModel:()=>Qe,MT5ForConditionalGeneration:()=>bt,MT5Model:()=>yt,MT5PreTrainedModel:()=>wt,MarianMTModel:()=>_a,MarianModel:()=>ga,MarianPreTrainedModel:()=>fa,MaskFormerForInstanceSegmentation:()=>Hi,MaskFormerModel:()=>Wi,MaskFormerPreTrainedModel:()=>Ui,MaskedLMOutput:()=>ul,MistralForCausalLM:()=>us,MistralModel:()=>ls,MistralPreTrainedModel:()=>os,MobileBertForMaskedLM:()=>He,MobileBertForQuestionAnswering:()=>Ke,MobileBertForSequenceClassification:()=>Xe,MobileBertModel:()=>We,MobileBertPreTrainedModel:()=>Ue,MobileNetV1ForImageClassification:()=>Ds,MobileNetV1Model:()=>Ls,MobileNetV1PreTrainedModel:()=>Bs,MobileNetV2ForImageClassification:()=>Vs,MobileNetV2Model:()=>Ns,MobileNetV2PreTrainedModel:()=>Rs,MobileNetV3ForImageClassification:()=>qs,MobileNetV3Model:()=>Gs,MobileNetV3PreTrainedModel:()=>js,MobileNetV4ForImageClassification:()=>Hs,MobileNetV4Model:()=>Ws,MobileNetV4PreTrainedModel:()=>Us,MobileViTForImageClassification:()=>Xr,MobileViTModel:()=>Hr,MobileViTPreTrainedModel:()=>Wr,MobileViTV2ForImageClassification:()=>Yr,MobileViTV2Model:()=>Qr,MobileViTV2PreTrainedModel:()=>Kr,ModelOutput:()=>j,Moondream1ForConditionalGeneration:()=>dn,MptForCausalLM:()=>Tr,MptModel:()=>Mr,MptPreTrainedModel:()=>xr,MusicgenForCausalLM:()=>zs,MusicgenForConditionalGeneration:()=>Os,MusicgenModel:()=>Is,MusicgenPreTrainedModel:()=>As,NomicBertModel:()=>Y,NomicBertPreTrainedModel:()=>Q,OPTForCausalLM:()=>Sr,OPTModel:()=>$r,OPTPreTrainedModel:()=>kr,OpenELMForCausalLM:()=>ur,OpenELMModel:()=>lr,OpenELMPreTrainedModel:()=>or,OwlViTForObjectDetection:()=>ei,OwlViTModel:()=>Zr,OwlViTPreTrainedModel:()=>Jr,Owlv2ForObjectDetection:()=>ri,Owlv2Model:()=>ni,Owlv2PreTrainedModel:()=>ti,Phi3ForCausalLM:()=>wr,Phi3Model:()=>_r,Phi3PreTrainedModel:()=>gr,PhiForCausalLM:()=>fr,PhiModel:()=>mr,PhiPreTrainedModel:()=>hr,PreTrainedModel:()=>V,PretrainedMixin:()=>Qs,PvtForImageClassification:()=>Ir,PvtModel:()=>Ar,PvtPreTrainedModel:()=>Fr,PyAnnoteForAudioFrameClassification:()=>Ca,PyAnnoteModel:()=>Sa,PyAnnotePreTrainedModel:()=>$a,QuestionAnsweringModelOutput:()=>dl,Qwen2ForCausalLM:()=>pr,Qwen2Model:()=>cr,Qwen2PreTrainedModel:()=>dr,RTDetrForObjectDetection:()=>fi,RTDetrModel:()=>mi,RTDetrObjectDetectionOutput:()=>gi,RTDetrPreTrainedModel:()=>hi,ResNetForImageClassification:()=>Pi,ResNetModel:()=>Ci,ResNetPreTrainedModel:()=>Si,RoFormerForMaskedLM:()=>ee,RoFormerForQuestionAnswering:()=>re,RoFormerForSequenceClassification:()=>te,RoFormerForTokenClassification:()=>ne,RoFormerModel:()=>Z,RoFormerPreTrainedModel:()=>J,RobertaForMaskedLM:()=>Dt,RobertaForQuestionAnswering:()=>Vt,RobertaForSequenceClassification:()=>Rt,RobertaForTokenClassification:()=>Nt,RobertaModel:()=>Lt,RobertaPreTrainedModel:()=>Bt,SamImageSegmentationOutput:()=>ma,SamModel:()=>ha,SamPreTrainedModel:()=>pa,SapiensForDepthEstimation:()=>Gi,SapiensForNormalEstimation:()=>qi,SapiensForSemanticSegmentation:()=>ji,SapiensPreTrainedModel:()=>Vi,SegformerForImageClassification:()=>Ts,SegformerForSemanticSegmentation:()=>ks,SegformerModel:()=>Ms,SegformerPreTrainedModel:()=>xs,Seq2SeqLMOutput:()=>al,SequenceClassifierOutput:()=>sl,SiglipModel:()=>bn,SiglipPreTrainedModel:()=>yn,SiglipTextModel:()=>vn,SiglipVisionModel:()=>xn,SpeechT5ForSpeechToText:()=>ns,SpeechT5ForTextToSpeech:()=>rs,SpeechT5HifiGan:()=>is,SpeechT5Model:()=>ts,SpeechT5PreTrainedModel:()=>es,SqueezeBertForMaskedLM:()=>it,SqueezeBertForQuestionAnswering:()=>st,SqueezeBertForSequenceClassification:()=>at,SqueezeBertModel:()=>rt,SqueezeBertPreTrainedModel:()=>nt,StableLmForCausalLM:()=>Cs,StableLmModel:()=>Ss,StableLmPreTrainedModel:()=>$s,Starcoder2ForCausalLM:()=>ps,Starcoder2Model:()=>cs,Starcoder2PreTrainedModel:()=>ds,Swin2SRForImageSuperResolution:()=>Oi,Swin2SRModel:()=>zi,Swin2SRPreTrainedModel:()=>Ii,SwinForImageClassification:()=>Ai,SwinModel:()=>Fi,SwinPreTrainedModel:()=>Ei,T5ForConditionalGeneration:()=>mt,T5Model:()=>ht,T5PreTrainedModel:()=>pt,TableTransformerForObjectDetection:()=>yi,TableTransformerModel:()=>wi,TableTransformerObjectDetectionOutput:()=>bi,TableTransformerPreTrainedModel:()=>_i,TokenClassifierOutput:()=>ll,TrOCRForCausalLM:()=>ss,TrOCRPreTrainedModel:()=>as,UniSpeechForCTC:()=>Ia,UniSpeechForSequenceClassification:()=>za,UniSpeechModel:()=>Aa,UniSpeechPreTrainedModel:()=>Fa,UniSpeechSatForAudioFrameClassification:()=>Ra,UniSpeechSatForCTC:()=>La,UniSpeechSatForSequenceClassification:()=>Da,UniSpeechSatModel:()=>Ba,UniSpeechSatPreTrainedModel:()=>Oa,ViTForImageClassification:()=>Er,ViTMAEModel:()=>Or,ViTMAEPreTrainedModel:()=>zr,ViTMSNForImageClassification:()=>Dr,ViTMSNModel:()=>Lr,ViTMSNPreTrainedModel:()=>Br,ViTModel:()=>Pr,ViTPreTrainedModel:()=>Cr,VisionEncoderDecoderModel:()=>on,VitMatteForImageMatting:()=>Ur,VitMattePreTrainedModel:()=>qr,VitsModel:()=>vs,VitsModelOutput:()=>ml,VitsPreTrainedModel:()=>bs,Wav2Vec2BertForCTC:()=>ja,Wav2Vec2BertForSequenceClassification:()=>Ga,Wav2Vec2BertModel:()=>Va,Wav2Vec2BertPreTrainedModel:()=>Na,Wav2Vec2ForAudioFrameClassification:()=>ka,Wav2Vec2ForCTC:()=>Ma,Wav2Vec2ForSequenceClassification:()=>Ta,Wav2Vec2Model:()=>xa,Wav2Vec2PreTrainedModel:()=>va,WavLMForAudioFrameClassification:()=>Za,WavLMForCTC:()=>Qa,WavLMForSequenceClassification:()=>Ya,WavLMForXVector:()=>Ja,WavLMModel:()=>Ka,WavLMPreTrainedModel:()=>Xa,WeSpeakerResNetModel:()=>Ea,WeSpeakerResNetPreTrainedModel:()=>Pa,WhisperForConditionalGeneration:()=>sn,WhisperModel:()=>an,WhisperPreTrainedModel:()=>rn,XLMForQuestionAnswering:()=>Ht,XLMForSequenceClassification:()=>Ut,XLMForTokenClassification:()=>Wt,XLMModel:()=>Gt,XLMPreTrainedModel:()=>jt,XLMRobertaForMaskedLM:()=>Qt,XLMRobertaForQuestionAnswering:()=>Zt,XLMRobertaForSequenceClassification:()=>Yt,XLMRobertaForTokenClassification:()=>Jt,XLMRobertaModel:()=>Kt,XLMRobertaPreTrainedModel:()=>Xt,XLMWithLMHeadModel:()=>qt,XVectorOutput:()=>ol,YolosForObjectDetection:()=>da,YolosModel:()=>ua,YolosObjectDetectionOutput:()=>ca,YolosPreTrainedModel:()=>la});var r=n(/*! ./configs.js */"./src/configs.js"),i=n(/*! ./backends/onnx.js */"./src/backends/onnx.js"),a=n(/*! ./utils/dtypes.js */"./src/utils/dtypes.js"),s=n(/*! ./utils/generic.js */"./src/utils/generic.js"),o=n(/*! ./utils/core.js */"./src/utils/core.js"),l=n(/*! ./utils/hub.js */"./src/utils/hub.js"),u=n(/*! ./generation/logits_process.js */"./src/generation/logits_process.js"),d=n(/*! ./generation/configuration_utils.js */"./src/generation/configuration_utils.js"),c=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"),m=n(/*! ./generation/logits_sampler.js */"./src/generation/logits_sampler.js"),f=n(/*! ./env.js */"./src/env.js"),g=n(/*! ./models/whisper/generation_whisper.js */"./src/models/whisper/generation_whisper.js"),_=n(/*! ./models/whisper/common_whisper.js */"./src/models/whisper/common_whisper.js");const w=0,y=1,b=2,v=3,x=4,M=5,T=6,k=7,$=new Map,S=new Map,C=new Map;async function P(e,t,n){return Object.fromEntries(await Promise.all(Object.keys(t).map((async s=>{const{buffer:o,session_options:u}=await async function(e,t,n){const s=n.config?.["transformers.js_config"]??{};let o=n.device??s.device;o&&"string"!=typeof o&&(o.hasOwnProperty(t)?o=o[t]:(console.warn(`device not specified for "${t}". Using the default device.`),o=null));const u=o??(f.apis.IS_NODE_ENV?"cpu":"wasm"),d=(0,i.deviceToExecutionProviders)(u);let c=n.dtype??s.dtype;"string"!=typeof c&&(c&&c.hasOwnProperty(t)?c=c[t]:(c=a.DEFAULT_DEVICE_DTYPE_MAPPING[u]??a.DATA_TYPES.fp32,console.warn(`dtype not specified for "${t}". Using the default dtype (${c}) for this device (${u}).`)));const p=c;if(!a.DEFAULT_DTYPE_SUFFIX_MAPPING.hasOwnProperty(p))throw new Error(`Invalid dtype: ${p}. Should be one of: ${Object.keys(a.DATA_TYPES).join(", ")}`);if(p===a.DATA_TYPES.fp16&&"webgpu"===u&&!await(0,a.isWebGpuFp16Supported)())throw new Error(`The device (${u}) does not support fp16.`);const h=a.DEFAULT_DTYPE_SUFFIX_MAPPING[p],m=`${n.subfolder??""}/${t}${h}.onnx`,g={...n.session_options}??{};g.executionProviders??=d;const _=s.free_dimension_overrides;_?g.freeDimensionOverrides??=_:u.startsWith("webnn")&&!g.freeDimensionOverrides&&console.warn('WebNN does not currently support dynamic shapes and requires `free_dimension_overrides` to be set in config.json as a field within "transformers.js_config". When `free_dimension_overrides` is not set, you may experience significant performance degradation.');const w=(0,l.getModelFile)(e,m,!0,n);let y=[];if(n.use_external_data_format&&(!0===n.use_external_data_format||"object"==typeof n.use_external_data_format&&n.use_external_data_format.hasOwnProperty(t)&&!0===n.use_external_data_format[t])){if(f.apis.IS_NODE_ENV)throw new Error("External data format is not yet supported in Node.js");const r=`${t}${h}.onnx_data`,i=`${n.subfolder??""}/${r}`;y.push(new Promise((async(t,a)=>{const s=await(0,l.getModelFile)(e,i,!0,n);t({path:r,data:s})})))}else void 0!==g.externalData&&(y=g.externalData.map((async t=>{if("string"==typeof t.data){const r=await(0,l.getModelFile)(e,t.data,!0,n);return{...t,data:r}}return t})));if(y.length>0&&(g.externalData=await Promise.all(y)),"webgpu"===u){const e=(0,r.getKeyValueShapes)(n.config,{prefix:"present"});if(Object.keys(e).length>0&&!(0,i.isONNXProxy)()){const t={};for(const n in e)n.includes("encoder")||(t[n]="gpu-buffer");g.preferredOutputLocation=t}}return{buffer:await w,session_options:g}}(e,t[s],n);return[s,await(0,i.createInferenceSession)(o,u)]}))))}async function E(e,t){const n=function(e,t){const n=Object.create(null),r=[];for(const a of e.inputNames){const e=t[a];e instanceof c.Tensor?n[a]=(0,i.isONNXProxy)()?e.clone():e:r.push(a)}if(r.length>0)throw new Error(`An error occurred during model execution: "Missing the following inputs: ${r.join(", ")}.`);const a=Object.keys(t).length,s=e.inputNames.length;if(a>s){let n=Object.keys(t).filter((t=>!e.inputNames.includes(t)));console.warn(`WARNING: Too many inputs were provided (${a} > ${s}). The following inputs will be ignored: "${n.join(", ")}".`)}return n}(e,t);try{const t=Object.fromEntries(Object.entries(n).map((([e,t])=>[e,t.ort_tensor])));let r=await e.run(t);return r=F(r),r}catch(e){throw console.error(`An error occurred during model execution: "${e}".`),console.error("Inputs given to model:",n),e}}function F(e){for(let t in e)(0,i.isONNXTensor)(e[t])?e[t]=new c.Tensor(e[t]):"object"==typeof e[t]&&F(e[t]);return e}function A(e){if(e instanceof c.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 c.Tensor("int64",BigInt64Array.from(e.flat().map((e=>BigInt(e)))),[e.length,e[0].length])}return new c.Tensor("int64",BigInt64Array.from(e.map((e=>BigInt(e)))),[1,e.length])}function I(e){return new c.Tensor("bool",[e],[1])}async function z(e,t){let{encoder_outputs:n,input_ids:r,decoder_input_ids:i,...a}=t;if(!n){const r=(0,o.pick)(t,e.sessions.model.inputNames);n=(await O(e,r)).last_hidden_state}a.input_ids=i,a.encoder_hidden_states=n,e.sessions.decoder_model_merged.inputNames.includes("encoder_attention_mask")&&(a.encoder_attention_mask=t.attention_mask);return await B(e,a,!0)}async function O(e,t){const n=e.sessions.model,r=(0,o.pick)(t,n.inputNames);if(n.inputNames.includes("inputs_embeds")&&!r.inputs_embeds){if(!t.input_ids)throw new Error("Both `input_ids` and `inputs_embeds` are missing in the model inputs.");r.inputs_embeds=await e.encode_text({input_ids:t.input_ids})}return n.inputNames.includes("token_type_ids")&&!r.token_type_ids&&(r.token_type_ids=new c.Tensor("int64",new BigInt64Array(r.input_ids.data.length),r.input_ids.dims)),await E(n,r)}async function B(e,t,n=!1){const r=e.sessions[n?"decoder_model_merged":"model"],{past_key_values:i,...a}=t;r.inputNames.includes("use_cache_branch")&&(a.use_cache_branch=I(!!i)),r.inputNames.includes("position_ids")&&a.attention_mask&&!a.position_ids&&(a.position_ids=function(e,t=null){const{input_ids:n,inputs_embeds:r,attention_mask:i}=e,[a,s]=i.dims,o=new BigInt64Array(i.data.length);for(let e=0;e<a;++e){const t=e*s;let n=BigInt(0);for(let e=0;e<s;++e){const r=t+e;0n===i.data[r]?o[r]=BigInt(1):(o[r]=n,n+=i.data[r])}}let l=new c.Tensor("int64",o,i.dims);if(t){const e=-(n??r).dims.at(1);l=l.slice(null,[e,null])}return l}(a,i)),e.addPastKeyValues(a,i);const s=(0,o.pick)(a,r.inputNames);return await E(r,s)}async function L(e,{input_ids:t=null,attention_mask:n=null,pixel_values:r=null,position_ids:i=null,inputs_embeds:a=null,past_key_values:s=null,generation_config:o=null,logits_processor:l=null,...u}){if(!a)if(a=await e.encode_text({input_ids:t}),r&&1!==t.dims[1]){const i=await e.encode_image({pixel_values:r});({inputs_embeds:a,attention_mask:n}=e._merge_input_ids_with_image_features({image_features:i,inputs_embeds:a,input_ids:t,attention_mask:n}))}else if(s&&r&&1===t.dims[1]){const e=t.dims[1],r=Object.values(s)[0].dims.at(-2);n=(0,c.cat)([(0,c.ones)([t.dims[0],r]),n.slice(null,[n.dims[1]-e,n.dims[1]])],1)}return await B(e,{inputs_embeds:a,past_key_values:s,attention_mask:n,position_ids:i,generation_config:o,logits_processor:l},!0)}function D(e,t,n,r){if(n.past_key_values){const t=Object.values(n.past_key_values)[0].dims.at(-2),{input_ids:r,attention_mask:i}=n;if(i&&i.dims[1]>r.dims[1]);else if(t<r.dims[1])n.input_ids=r.slice(null,[t,null]);else if(null!=e.config.image_token_index&&r.data.some((t=>t==e.config.image_token_index))){const i=e.config.num_image_tokens;if(!i)throw new Error("`num_image_tokens` is missing in the model configuration.");const a=r.dims[1]-(t-i);n.input_ids=r.slice(null,[-a,null]),n.attention_mask=(0,c.ones)([1,t+a])}}return n}function R(e,t,n,r){return n.past_key_values&&(t=t.map((e=>[e.at(-1)]))),{...n,decoder_input_ids:A(t)}}function N(e,...t){return e.config.is_encoder_decoder?R(e,...t):D(e,...t)}class V extends s.Callable{main_input_name="input_ids";forward_params=["input_ids","attention_mask"];constructor(e,t){super(),this.config=e,this.sessions=t;const n=C.get(this.constructor),r=$.get(n);switch(this.can_generate=!1,this._forward=null,this._prepare_inputs_for_generation=null,r){case x:this.can_generate=!0,this._forward=B,this._prepare_inputs_for_generation=D;break;case b:case v:case k:this.can_generate=!0,this._forward=z,this._prepare_inputs_for_generation=R;break;case y:this._forward=z;break;case T:this.can_generate=!0,this._forward=L,this._prepare_inputs_for_generation=N;break;default:this._forward=O}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:i=null,local_files_only:a=!1,revision:s="main",model_file_name:o=null,subfolder:u="onnx",device:d=null,dtype:c=null,use_external_data_format:p=null,session_options:h={}}={}){let m={progress_callback:t,config:n,cache_dir:i,local_files_only:a,revision:s,model_file_name:o,subfolder:u,device:d,dtype:c,use_external_data_format:p,session_options:h};const f=C.get(this),g=$.get(f);let _;if(n=m.config=await r.AutoConfig.from_pretrained(e,m),g===x)_=await Promise.all([P(e,{model:m.model_file_name??"model"},m),(0,l.getModelJSON)(e,"generation_config.json",!1,m)]);else if(g===b||g===v)_=await Promise.all([P(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},m),(0,l.getModelJSON)(e,"generation_config.json",!1,m)]);else if(g===M)_=await Promise.all([P(e,{model:"vision_encoder",prompt_encoder_mask_decoder:"prompt_encoder_mask_decoder"},m)]);else if(g===y)_=await Promise.all([P(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},m)]);else if(g===T){const t={embed_tokens:"embed_tokens",vision_encoder:"vision_encoder",decoder_model_merged:"decoder_model_merged"};n.is_encoder_decoder&&(t.model="encoder_model"),_=await Promise.all([P(e,t,m),(0,l.getModelJSON)(e,"generation_config.json",!1,m)])}else g===k?_=await Promise.all([P(e,{model:"text_encoder",decoder_model_merged:"decoder_model_merged",encodec_decode:"encodec_decode"},m),(0,l.getModelJSON)(e,"generation_config.json",!1,m)]):(g!==w&&console.warn(`Model type for '${f??n?.model_type}' not found, assuming encoder-only architecture. Please report this at https://github.com/xenova/transformers.js/issues/new/choose.`),_=await Promise.all([P(e,{model:m.model_file_name??"model"},m)]));return new this(n,..._)}async _call(e){return await this.forward(e)}async forward(e){return await this._forward(this,e)}_get_logits_warper(e){const t=new u.LogitsProcessorList;return null!==e.temperature&&1!==e.temperature&&t.push(new u.TemperatureLogitsWarper(e.temperature)),null!==e.top_k&&0!==e.top_k&&t.push(new u.TopKLogitsWarper(e.top_k)),null!==e.top_p&&e.top_p<1&&t.push(new u.TopPLogitsWarper(e.top_p)),t}_get_logits_processor(e,t,n=null){const r=new u.LogitsProcessorList;if(null!==e.repetition_penalty&&1!==e.repetition_penalty&&r.push(new u.RepetitionPenaltyLogitsProcessor(e.repetition_penalty)),null!==e.no_repeat_ngram_size&&e.no_repeat_ngram_size>0&&r.push(new u.NoRepeatNGramLogitsProcessor(e.no_repeat_ngram_size)),null!==e.bad_words_ids&&r.push(new u.NoBadWordsLogitsProcessor(e.bad_words_ids,e.eos_token_id)),null!==e.min_length&&null!==e.eos_token_id&&e.min_length>0&&r.push(new u.MinLengthLogitsProcessor(e.min_length,e.eos_token_id)),null!==e.min_new_tokens&&null!==e.eos_token_id&&e.min_new_tokens>0&&r.push(new u.MinNewTokensLengthLogitsProcessor(t,e.min_new_tokens,e.eos_token_id)),null!==e.forced_bos_token_id&&r.push(new u.ForcedBOSTokenLogitsProcessor(e.forced_bos_token_id)),null!==e.forced_eos_token_id&&r.push(new u.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;r.push(new u.SuppressTokensAtBeginLogitsProcessor(e.begin_suppress_tokens,n))}return null!==e.guidance_scale&&e.guidance_scale>1&&r.push(new u.ClassifierFreeGuidanceLogitsProcessor(e.guidance_scale)),null!==n&&r.extend(n),r}_prepare_generation_config(e,t,n=d.GenerationConfig){const r={...this.config};for(const e of["decoder","generator","text_config"])e in r&&Object.assign(r,r[e]);const i=new n(r);return"generation_config"in this&&Object.assign(i,this.generation_config),e&&Object.assign(i,e),t&&Object.assign(i,(0,o.pick)(t,Object.getOwnPropertyNames(i))),i}_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=[so,uo,ao,eo],t=C.get(this.constructor),n=new Set,r=this.config.model_type;for(const t of e){const e=t.get(r);e&&n.add(e[0])}let i=`The current model class (${t}) is not compatible with \`.generate()\`, as it doesn't have a language model head.`;throw n.size>0&&(i+=` Please use the following class instead: ${[...n].join(", ")}`),Error(i)}}prepare_inputs_for_generation(...e){return this._prepare_inputs_for_generation(this,...e)}_update_model_kwargs_for_generation({generated_input_ids:e,outputs:t,model_inputs:n,is_encoder_decoder:r}){return n.past_key_values=this.getPastKeyValues(t,n.past_key_values),n.input_ids=new c.Tensor("int64",e.flat(),[e.length,1]),r||(n.attention_mask=(0,c.cat)([n.attention_mask,(0,c.ones)([n.attention_mask.dims[0],1])],1)),n.position_ids=null,n}_prepare_model_inputs({inputs:e,bos_token_id:t,model_kwargs:n}){const r=(0,o.pick)(n,this.forward_params),i=this.main_input_name;if(i in r){if(e)throw new Error("`inputs`: {inputs}` were passed alongside {input_name} which is not allowed. Make sure to either pass {inputs} or {input_name}=...")}else r[i]=e;return{inputs_tensor:r[i],model_inputs:r,model_input_name:i}}async _prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:e,model_inputs:t,model_input_name:n,generation_config:r}){if(this.sessions.model.inputNames.includes("inputs_embeds")&&!t.inputs_embeds&&"_prepare_inputs_embeds"in this){const{input_ids:e,pixel_values:n,attention_mask:r,...i}=t,a=await this._prepare_inputs_embeds(t);t={...i,...(0,o.pick)(a,["inputs_embeds","attention_mask"])}}let{last_hidden_state:i}=await O(this,t);if(null!==r.guidance_scale&&r.guidance_scale>1)i=(0,c.cat)([i,(0,c.full_like)(i,0)],0),"attention_mask"in t&&(t.attention_mask=(0,c.cat)([t.attention_mask,(0,c.zeros_like)(t.attention_mask)],0));else if(t.decoder_input_ids){const e=A(t.decoder_input_ids).dims[0];if(e!==i.dims[0]){if(1!==i.dims[0])throw new Error(`The encoder outputs have a different batch size (${i.dims[0]}) than the decoder inputs (${e}).`);i=(0,c.cat)(Array.from({length:e},(()=>i)),0)}}return t.encoder_outputs=i,t}_prepare_decoder_input_ids_for_generation({batch_size:e,model_input_name:t,model_kwargs:n,decoder_start_token_id:r,bos_token_id:i,generation_config:a}){let{decoder_input_ids:s,...o}=n;if(s)Array.isArray(s[0])||(s=Array.from({length:e},(()=>s)));else if(r??=i,"musicgen"===this.config.model_type)s=Array.from({length:e*this.config.decoder.num_codebooks},(()=>[r]));else if(Array.isArray(r)){if(r.length!==e)throw new Error(`\`decoder_start_token_id\` expcted to have length ${e} but got ${r.length}`);s=r}else s=Array.from({length:e},(()=>[r]));return s=A(s),n.decoder_attention_mask=(0,c.ones_like)(s),{input_ids:s,model_inputs:o}}async generate({inputs:e=null,generation_config:t=null,logits_processor:n=null,stopping_criteria:r=null,streamer:i=null,...a}){this._validate_model_class(),t=this._prepare_generation_config(t,a);let{inputs_tensor:s,model_inputs:o,model_input_name:l}=this._prepare_model_inputs({inputs:e,model_kwargs:a});const u=this.config.is_encoder_decoder;let d;u&&("encoder_outputs"in o||(o=await this._prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:s,model_inputs:o,model_input_name:l,generation_config:t}))),u?({input_ids:d,model_inputs:o}=this._prepare_decoder_input_ids_for_generation({batch_size:o[l].dims.at(0),model_input_name:l,model_kwargs:o,decoder_start_token_id:t.decoder_start_token_id,bos_token_id:t.bos_token_id,generation_config:t})):d=o[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),f=this._get_stopping_criteria(t,r),g=o[l].dims.at(0),_=m.LogitsSampler.getSampler(t),w=new Array(g).fill(0),y=d.tolist();let b;i&&i.put(y);let v={};for(;;){if(o=this.prepare_inputs_for_generation(y,o,t),b=await this.forward(o),t.output_attentions&&t.return_dict_in_generate){const e=this.getAttentions(b);for(const t in e)t in v||(v[t]=[]),v[t].push(e[t])}const e=h(y,b.logits.slice(null,-1,null)),n=[];for(let t=0;t<e.dims.at(0);++t){const r=e[t],i=await _(r);for(const[e,r]of i){const i=BigInt(e);w[t]+=r,y[t].push(i),n.push([i]);break}}i&&i.put(n);if(f(y).every((e=>e)))break;o=this._update_model_kwargs_for_generation({generated_input_ids:n,outputs:b,model_inputs:o,is_encoder_decoder:u})}i&&i.end();const x=this.getPastKeyValues(b,o.past_key_values,!0),M=new c.Tensor("int64",y.flat(),[y.length,y[0].length]);if(t.return_dict_in_generate)return{sequences:M,past_key_values:x,...v};for(const e of Object.values(b))"gpu-buffer"===e.location&&e.dispose();return M}getPastKeyValues(e,t,n=!1){const r=Object.create(null);for(const i in e)if(i.startsWith("present")){const a=i.replace("present","past_key_values"),s=i.includes("encoder");if(r[a]=s&&t?t[a]:e[i],t&&(!s||n)){const e=t[a];"gpu-buffer"===e.location&&e.dispose()}}return r}getAttentions(e){const t={};for(const n of["cross_attentions","encoder_attentions","decoder_attentions"])for(const r in e)r.startsWith(n)&&(n in t||(t[n]=[]),t[n].push(e[r]));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:[],i=(0,r.getKeyValueShapes)(this.config);for(const r in i)e[r]=new c.Tensor(t,n,i[r])}}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 j{}class G extends j{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 q extends V{}class U extends q{}class W extends q{async _call(e){return new ul(await super._call(e))}}class H extends q{async _call(e){return new sl(await super._call(e))}}class X extends q{async _call(e){return new ll(await super._call(e))}}class K extends q{async _call(e){return new dl(await super._call(e))}}class Q extends V{}class Y extends Q{}class J extends V{}class Z extends J{}class ee extends J{async _call(e){return new ul(await super._call(e))}}class te extends J{async _call(e){return new sl(await super._call(e))}}class ne extends J{async _call(e){return new ll(await super._call(e))}}class re extends J{async _call(e){return new dl(await super._call(e))}}class ie extends V{}class ae extends ie{}class se extends ie{async _call(e){return new ul(await super._call(e))}}class oe extends ie{async _call(e){return new sl(await super._call(e))}}class le extends ie{async _call(e){return new ll(await super._call(e))}}class ue extends ie{async _call(e){return new dl(await super._call(e))}}class de extends V{}class ce extends de{}class pe extends de{async _call(e){return new ul(await super._call(e))}}class he extends de{async _call(e){return new sl(await super._call(e))}}class me extends de{async _call(e){return new ll(await super._call(e))}}class fe extends de{async _call(e){return new dl(await super._call(e))}}class ge extends V{}class _e extends ge{}class we extends ge{async _call(e){return new ul(await super._call(e))}}class ye extends ge{async _call(e){return new sl(await super._call(e))}}class be extends ge{async _call(e){return new ll(await super._call(e))}}class ve extends ge{async _call(e){return new dl(await super._call(e))}}class xe extends V{}class Me extends xe{}class Te extends xe{async _call(e){return new ul(await super._call(e))}}class ke extends xe{async _call(e){return new sl(await super._call(e))}}class $e extends xe{async _call(e){return new ll(await super._call(e))}}class Se extends xe{async _call(e){return new dl(await super._call(e))}}class Ce extends V{}class Pe extends Ce{}class Ee extends Ce{async _call(e){return new ul(await super._call(e))}}class Fe extends Ce{async _call(e){return new sl(await super._call(e))}}class Ae extends Ce{async _call(e){return new ll(await super._call(e))}}class Ie extends Ce{async _call(e){return new dl(await super._call(e))}}class ze extends V{}class Oe extends ze{}class Be extends ze{async _call(e){return new sl(await super._call(e))}}class Le extends ze{async _call(e){return new ll(await super._call(e))}}class De extends ze{async _call(e){return new dl(await super._call(e))}}class Re extends ze{async _call(e){return new ul(await super._call(e))}}class Ne extends V{}class Ve extends Ne{}class je extends Ne{async _call(e){return new ul(await super._call(e))}}class Ge extends Ne{async _call(e){return new sl(await super._call(e))}}class qe extends Ne{async _call(e){return new ll(await super._call(e))}}class Ue extends V{}class We extends Ue{}class He extends Ue{async _call(e){return new ul(await super._call(e))}}class Xe extends Ue{async _call(e){return new sl(await super._call(e))}}class Ke extends Ue{async _call(e){return new dl(await super._call(e))}}class Qe extends V{}class Ye extends Qe{}class Je extends Qe{async _call(e){return new ul(await super._call(e))}}class Ze extends Qe{async _call(e){return new sl(await super._call(e))}}class et extends Qe{async _call(e){return new ll(await super._call(e))}}class tt extends Qe{async _call(e){return new dl(await super._call(e))}}class nt extends V{}class rt extends nt{}class it extends nt{async _call(e){return new ul(await super._call(e))}}class at extends nt{async _call(e){return new sl(await super._call(e))}}class st extends nt{async _call(e){return new dl(await super._call(e))}}class ot extends V{}class lt extends ot{}class ut extends ot{async _call(e){return new sl(await super._call(e))}}class dt extends ot{async _call(e){return new dl(await super._call(e))}}class ct extends ot{async _call(e){return new ul(await super._call(e))}}class pt extends V{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"];constructor(e,t,n){super(e,t),this.generation_config=n}}class ht extends pt{}class mt extends pt{}class ft extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class gt extends ft{}class _t extends ft{}class wt extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class yt extends wt{}class bt extends wt{}class vt extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class xt extends vt{}class Mt extends vt{}class Tt extends vt{async _call(e){return new sl(await super._call(e))}}class kt extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class $t extends kt{}class St extends kt{}class Ct extends kt{async _call(e){return new sl(await super._call(e))}}class Pt extends kt{}class Et extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class Ft extends Et{}class At extends Et{}class It extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class zt extends It{}class Ot extends It{}class Bt extends V{}class Lt extends Bt{}class Dt extends Bt{async _call(e){return new ul(await super._call(e))}}class Rt extends Bt{async _call(e){return new sl(await super._call(e))}}class Nt extends Bt{async _call(e){return new ll(await super._call(e))}}class Vt extends Bt{async _call(e){return new dl(await super._call(e))}}class jt extends V{}class Gt extends jt{}class qt extends jt{async _call(e){return new ul(await super._call(e))}}class Ut extends jt{async _call(e){return new sl(await super._call(e))}}class Wt extends jt{async _call(e){return new ll(await super._call(e))}}class Ht extends jt{async _call(e){return new dl(await super._call(e))}}class Xt extends V{}class Kt extends Xt{}class Qt extends Xt{async _call(e){return new ul(await super._call(e))}}class Yt extends Xt{async _call(e){return new sl(await super._call(e))}}class Jt extends Xt{async _call(e){return new ll(await super._call(e))}}class Zt extends Xt{async _call(e){return new dl(await super._call(e))}}class en extends V{}class tn extends en{}class nn extends en{}class rn extends V{requires_attention_mask=!1;main_input_name="input_features";forward_params=["input_features","attention_mask","decoder_input_ids","decoder_attention_mask","past_key_values"];constructor(e,t,n){super(e,t),this.generation_config=n}}class an extends rn{}class sn extends rn{_prepare_generation_config(e,t){return super._prepare_generation_config(e,t,g.WhisperGenerationConfig)}_retrieve_init_tokens(e){const t=[e.decoder_start_token_id];let n=e.language;const r=e.task;if(e.is_multilingual){n||(console.warn("No language specified - defaulting to English (en)."),n="en");const i=`<|${(0,_.whisper_language_to_code)(n)}|>`;t.push(e.lang_to_id[i]),t.push(e.task_to_id[r??"transcribe"])}else if(n||r)throw new Error("Cannot specify `task` or `language` for an English-only model. If the model is intended to be multilingual, pass `is_multilingual=true` to generate, or update the generation config.");return!e.return_timestamps&&e.no_timestamps_token_id&&t.at(-1)!==e.no_timestamps_token_id?t.push(e.no_timestamps_token_id):e.return_timestamps&&t.at(-1)===e.no_timestamps_token_id&&(console.warn("<|notimestamps|> prompt token is removed from generation_config since `return_timestamps` is set to `true`."),t.pop()),t.filter((e=>null!=e))}async generate({inputs:e=null,generation_config:t=null,logits_processor:n=null,stopping_criteria:r=null,...i}){t=this._prepare_generation_config(t,i);const a=i.decoder_input_ids??this._retrieve_init_tokens(t);if(t.return_timestamps&&(n??=new u.LogitsProcessorList,n.push(new u.WhisperTimeStampLogitsProcessor(t,a))),t.begin_suppress_tokens&&(n??=new u.LogitsProcessorList,n.push(new u.SuppressTokensAtBeginLogitsProcessor(t.begin_suppress_tokens,a.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 s=await super.generate({inputs:e,generation_config:t,logits_processor:n,decoder_input_ids:a,...i});return t.return_token_timestamps&&(s.token_timestamps=this._extract_token_timestamps(s,t.alignment_heads,t.num_frames)),s}_extract_token_timestamps(e,t,n=null,r=.02){if(!e.cross_attentions)throw new Error("Model outputs must contain cross attentions to extract timestamps. This is most likely because the model was not exported with `output_attentions=True`.");null==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 i=this.config.median_filter_width;void 0===i&&(console.warn("Model config has no `median_filter_width`, using default value of 7."),i=7);const a=e.cross_attentions,s=Array.from({length:this.config.decoder_layers},((e,t)=>(0,c.cat)(a.map((e=>e[t])),2))),l=(0,c.stack)(t.map((([e,t])=>{if(e>=s.length)throw new Error(`Layer index ${e} is out of bounds for cross attentions (length ${s.length}).`);return n?s[e].slice(null,t,null,[0,n]):s[e].slice(null,t)}))).transpose(1,0,2,3),[u,d]=(0,c.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 r=t[n],a=u[e][n][0].data,s=d[e][n][0].data;for(let e=0;e<r.dims[0];++e){let t=r[e].data;for(let e=0;e<t.length;++e)t[e]=(t[e]-s[e])/a[e];t.set((0,p.medianFilter)(t,i))}}}const m=[(0,c.mean)(h,1)],f=e.sequences.dims,g=new c.Tensor("float32",new Float32Array(f[0]*f[1]),f);for(let e=0;e<f[0];++e){const t=m[e].neg().squeeze_(0),[n,i]=(0,p.dynamic_time_warping)(t.tolist()),a=Array.from({length:n.length-1},((e,t)=>n[t+1]-n[t])),s=(0,o.mergeArrays)([1],a).map((e=>!!e)),l=[];for(let e=0;e<s.length;++e)s[e]&&l.push(i[e]*r);g[e].data.set(l,1)}return g}}class on extends V{main_input_name="pixel_values";forward_params=["pixel_values","input_ids","encoder_hidden_states","past_key_values"];constructor(e,t,n){super(e,t),this.generation_config=n}}class ln extends V{forward_params=["input_ids","pixel_values","attention_mask","position_ids","past_key_values"];constructor(e,t,n){super(e,t),this.generation_config=n}}class un extends ln{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:n,attention_mask:r}){const i=this.config.image_token_index,a=n.tolist().map((e=>e.findIndex((e=>e==i)))),s=a.every((e=>-1===e)),o=a.every((e=>-1!==e));if(!s&&!o)throw new Error("Every input should contain either 0 or 1 image token.");if(s)return{inputs_embeds:e,attention_mask:r};const l=[],u=[];for(let n=0;n<a.length;++n){const i=a[n],s=e[n],o=t[n],d=r[n];l.push((0,c.cat)([s.slice([0,i]),o,s.slice([i+1,s.dims[0]])],0)),u.push((0,c.cat)([d.slice([0,i]),(0,c.ones)([o.dims[0]]),d.slice([i+1,d.dims[0]])],0))}return{inputs_embeds:(0,c.stack)(l,0),attention_mask:(0,c.stack)(u,0)}}}class dn extends un{}class cn extends V{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";constructor(e,t,n){super(e,t),this.generation_config=n}}class pn extends cn{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:n,attention_mask:r}){return{inputs_embeds:(0,c.cat)([t,e],1),attention_mask:(0,c.cat)([(0,c.ones)(t.dims.slice(0,2)),r],1)}}async _prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:n,attention_mask:r}){if(!e&&!t)throw new Error("Either `input_ids` or `pixel_values` should be provided.");let i,a;return e&&(i=await this.encode_text({input_ids:e})),t&&(a=await this.encode_image({pixel_values:t})),i&&a?({inputs_embeds:n,attention_mask:r}=this._merge_input_ids_with_image_features({inputs_embeds:i,image_features:a,input_ids:e,attention_mask:r})):n=i||a,{inputs_embeds:n,attention_mask:r}}async forward({input_ids:e,pixel_values:t,attention_mask:n,decoder_input_ids:r,decoder_attention_mask:i,encoder_outputs:a,past_key_values:s,inputs_embeds:o,decoder_inputs_embeds:l}){if(o||({inputs_embeds:o,attention_mask:n}=await this._prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:o,attention_mask:n})),!a){let{last_hidden_state:e}=await O(this,{inputs_embeds:o,attention_mask:n});a=e}if(!l){if(!r)throw new Error("Either `decoder_input_ids` or `decoder_inputs_embeds` should be provided.");l=await this.encode_text({input_ids:r})}const u={inputs_embeds:l,attention_mask:i,encoder_attention_mask:n,encoder_hidden_states:a,past_key_values:s};return await B(this,u,!0)}}class hn extends V{}class mn extends hn{}class fn extends hn{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class gn extends hn{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class _n extends hn{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class wn extends hn{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class yn extends V{}class bn extends yn{}class vn extends yn{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class xn extends hn{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class Mn extends V{}class Tn extends Mn{}class kn extends V{}class $n extends kn{}class Sn extends kn{}class Cn extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class Pn extends Cn{}class En extends Cn{}class Fn extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class An extends Fn{}class In extends Fn{}class zn extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class On extends zn{}class Bn extends zn{}class Ln extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class Dn extends Ln{}class Rn extends Ln{}class Nn extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class Vn extends Nn{}class jn extends Nn{}class Gn extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class qn extends Gn{}class Un extends Gn{}class Wn extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class Hn extends Wn{}class Xn extends Wn{}class Kn extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class Qn extends Kn{}class Yn extends Kn{}class Jn extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class Zn extends Jn{}class er extends Jn{}class tr extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class nr extends tr{}class rr extends tr{}class ir extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class ar extends ir{}class sr extends ir{}class or extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class lr extends or{}class ur extends or{}class dr extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class cr extends dr{}class pr extends dr{}class hr extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class mr extends hr{}class fr extends hr{}class gr extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class _r extends gr{}class wr extends gr{}class yr extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class br extends yr{}class vr extends yr{}class xr extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class Mr extends xr{}class Tr extends xr{}class kr extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class $r extends kr{}class Sr extends kr{}class Cr extends V{}class Pr extends Cr{}class Er extends Cr{async _call(e){return new sl(await super._call(e))}}class Fr extends V{}class Ar extends Fr{}class Ir extends Fr{async _call(e){return new sl(await super._call(e))}}class zr extends V{}class Or extends zr{}class Br extends V{}class Lr extends Br{}class Dr extends Br{async _call(e){return new sl(await super._call(e))}}class Rr extends V{}class Nr extends Rr{}class Vr extends V{}class jr extends Vr{}class Gr extends Vr{async _call(e){return new sl(await super._call(e))}}class qr extends V{}class Ur extends qr{async _call(e){return new hl(await super._call(e))}}class Wr extends V{}class Hr extends Wr{}class Xr extends Wr{async _call(e){return new sl(await super._call(e))}}class Kr extends V{}class Qr extends Kr{}class Yr extends Kr{async _call(e){return new sl(await super._call(e))}}class Jr extends V{}class Zr extends Jr{}class ei extends Jr{}class ti extends V{}class ni extends ti{}class ri extends ti{}class ii extends V{}class ai extends ii{}class si extends ii{async _call(e){return new sl(await super._call(e))}}class oi extends V{}class li extends oi{}class ui extends oi{async _call(e){return new ci(await super._call(e))}}class di extends oi{async _call(e){return new pi(await super._call(e))}}class ci extends j{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class pi extends j{constructor({logits:e,pred_boxes:t,pred_masks:n}){super(),this.logits=e,this.pred_boxes=t,this.pred_masks=n}}class hi extends V{}class mi extends hi{}class fi extends hi{async _call(e){return new gi(await super._call(e))}}class gi extends j{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class _i extends V{}class wi extends _i{}class yi extends _i{async _call(e){return new bi(await super._call(e))}}class bi extends ci{}class vi extends V{}class xi extends vi{}class Mi extends vi{async _call(e){return new sl(await super._call(e))}}class Ti extends V{}class ki extends Ti{}class $i extends Ti{async _call(e){return new sl(await super._call(e))}}class Si extends V{}class Ci extends Si{}class Pi extends Si{async _call(e){return new sl(await super._call(e))}}class Ei extends V{}class Fi extends Ei{}class Ai extends Ei{async _call(e){return new sl(await super._call(e))}}class Ii extends V{}class zi extends Ii{}class Oi extends Ii{}class Bi extends V{}class Li extends Bi{}class Di extends Bi{}class Ri extends V{}class Ni extends Ri{}class Vi extends V{}class ji extends Vi{}class Gi extends Vi{}class qi extends Vi{}class Ui extends V{}class Wi extends Ui{}class Hi extends Ui{}class Xi extends V{}class Ki extends Xi{}class Qi extends Xi{}class Yi extends V{}class Ji extends Yi{}class Zi extends V{}class ea extends Zi{}class ta extends Zi{async _call(e){return new sl(await super._call(e))}}class na extends V{}class ra extends na{}class ia extends na{async _call(e){return new sl(await super._call(e))}}class aa extends V{}class sa extends aa{}class oa extends aa{async _call(e){return new sl(await super._call(e))}}class la extends V{}class ua extends la{}class da extends la{async _call(e){return new ca(await super._call(e))}}class ca extends j{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class pa extends V{}class ha extends pa{async get_image_embeddings({pixel_values:e}){return await O(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 c.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 ma(await super._call(e))}}class ma extends j{constructor({iou_scores:e,pred_masks:t}){super(),this.iou_scores=e,this.pred_masks=t}}class fa extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class ga extends fa{}class _a extends fa{}class wa extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class ya extends wa{}class ba extends wa{}class va extends V{}class xa extends va{}class Ma extends va{async _call(e){return new cl(await super._call(e))}}class Ta extends va{async _call(e){return new sl(await super._call(e))}}class ka extends va{async _call(e){return new ll(await super._call(e))}}class $a extends V{}class Sa extends $a{}class Ca extends $a{async _call(e){return new ll(await super._call(e))}}class Pa extends V{}class Ea extends Pa{}class Fa extends V{}class Aa extends Fa{}class Ia extends Fa{async _call(e){return new cl(await super._call(e))}}class za extends Fa{async _call(e){return new sl(await super._call(e))}}class Oa extends V{}class Ba extends Oa{}class La extends Oa{async _call(e){return new cl(await super._call(e))}}class Da extends Oa{async _call(e){return new sl(await super._call(e))}}class Ra extends Oa{async _call(e){return new ll(await super._call(e))}}class Na extends V{}class Va extends Na{}class ja extends Na{async _call(e){return new cl(await super._call(e))}}class Ga extends Na{async _call(e){return new sl(await super._call(e))}}class qa extends V{}class Ua extends va{}class Wa extends va{async _call(e){return new cl(await super._call(e))}}class Ha extends va{async _call(e){return new sl(await super._call(e))}}class Xa extends V{}class Ka extends Xa{}class Qa extends Xa{async _call(e){return new cl(await super._call(e))}}class Ya extends Xa{async _call(e){return new sl(await super._call(e))}}class Ja extends Xa{async _call(e){return new ol(await super._call(e))}}class Za extends Xa{async _call(e){return new ll(await super._call(e))}}class es extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class ts extends es{}class ns extends es{}class rs extends es{async generate_speech(e,t,{threshold:n=.5,minlenratio:r=0,maxlenratio:i=20,vocoder:a=null}={}){const s={input_ids:e},{encoder_outputs:o,encoder_attention_mask:l}=await O(this,s),u=o.dims[1]/this.config.reduction_factor,d=Math.floor(u*i),p=Math.floor(u*r),h=this.config.num_mel_bins;let m=[],f=null,g=null,_=0;for(;;){++_;const e=I(!!g);let r;r=g?g.output_sequence_out:new c.Tensor("float32",new Float32Array(h),[1,1,h]);let i={use_cache_branch:e,output_sequence:r,encoder_attention_mask:l,speaker_embeddings:t,encoder_hidden_states:o};this.addPastKeyValues(i,f),g=await E(this.sessions.decoder_model_merged,i),f=this.getPastKeyValues(g,f);const{prob:a,spectrum:s}=g;if(m.push(s),_>=p&&(Array.from(a.data).filter((e=>e>=n)).length>0||_>=d))break}const w=(0,c.cat)(m),{waveform:y}=await E(a.sessions.model,{spectrogram:w});return{spectrogram:w,waveform:y}}}class is extends V{main_input_name="spectrogram"}class as extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class ss extends as{}class os extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class ls extends os{}class us extends os{}class ds extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class cs extends ds{}class ps extends ds{}class hs extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class ms extends hs{}class fs extends hs{}class gs extends V{}class _s extends gs{}class ws extends gs{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class ys extends gs{static async from_pretrained(e,t={}){return t.model_file_name??="audio_model",super.from_pretrained(e,t)}}class bs extends V{}class vs extends bs{async _call(e){return new ml(await super._call(e))}}class xs extends V{}class Ms extends xs{}class Ts extends xs{}class ks extends xs{}class $s extends V{constructor(e,t,n){super(e,t),this.generation_config=n}}class Ss extends $s{}class Cs extends $s{}class Ps extends V{}class Es extends Ps{}class Fs extends Ps{async _call(e){return new sl(await super._call(e))}}class As extends V{}class Is extends As{}class zs extends As{}class Os extends V{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"];constructor(e,t,n){super(e,t),this.generation_config=n}_apply_and_filter_by_delay_pattern_mask(e){const[t,n]=e.dims,r=this.config.decoder.num_codebooks,i=n-r;let a=0;for(let t=0;t<e.size;++t){if(e.data[t]===this.config.decoder.pad_token_id)continue;const s=t%n-Math.floor(t/n)%r;s>0&&s<=i&&(e.data[a++]=e.data[t])}const s=Math.floor(t/r),o=a/(s*r);return new c.Tensor(e.type,e.data.slice(0,a),[s,r,o])}prepare_inputs_for_generation(e,t,n){let r=structuredClone(e);for(let e=0;e<r.length;++e)for(let t=0;t<r[e].length;++t)e%this.config.decoder.num_codebooks>=t&&(r[e][t]=BigInt(this.config.decoder.pad_token_id));null!==n.guidance_scale&&n.guidance_scale>1&&(r=r.concat(r));return super.prepare_inputs_for_generation(r,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:r}=await E(this.sessions.encodec_decode,{audio_codes:n});return r}}class Bs extends V{}class Ls extends Bs{}class Ds extends Bs{async _call(e){return new sl(await super._call(e))}}class Rs extends V{}class Ns extends Rs{}class Vs extends Rs{async _call(e){return new sl(await super._call(e))}}class js extends V{}class Gs extends js{}class qs extends js{async _call(e){return new sl(await super._call(e))}}class Us extends V{}class Ws extends Us{}class Hs extends Us{async _call(e){return new sl(await super._call(e))}}class Xs extends V{}class Ks extends Xs{}class Qs{static MODEL_CLASS_MAPPINGS=null;static BASE_IF_FAIL=!1;static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:i=null,local_files_only:a=!1,revision:s="main",model_file_name:o=null,subfolder:l="onnx",device:u=null,dtype:d=null,use_external_data_format:c=null,session_options:p={}}={}){const h={progress_callback:t,config:n,cache_dir:i,local_files_only:a,revision:s,model_file_name:o,subfolder:l,device:u,dtype:d,use_external_data_format:c,session_options:p};if(h.config=await r.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 V.from_pretrained(e,h);throw Error(`Unsupported model type: ${h.config.model_type}`)}}const Ys=new Map([["bert",["BertModel",U]],["nomic_bert",["NomicBertModel",Y]],["roformer",["RoFormerModel",Z]],["electra",["ElectraModel",ce]],["esm",["EsmModel",Ve]],["convbert",["ConvBertModel",ae]],["camembert",["CamembertModel",_e]],["deberta",["DebertaModel",Me]],["deberta-v2",["DebertaV2Model",Pe]],["mpnet",["MPNetModel",Ye]],["albert",["AlbertModel",lt]],["distilbert",["DistilBertModel",Oe]],["roberta",["RobertaModel",Lt]],["xlm",["XLMModel",Gt]],["xlm-roberta",["XLMRobertaModel",Kt]],["clap",["ClapModel",_s]],["clip",["CLIPModel",mn]],["clipseg",["CLIPSegModel",$n]],["chinese_clip",["ChineseCLIPModel",Tn]],["siglip",["SiglipModel",bn]],["mobilebert",["MobileBertModel",We]],["squeezebert",["SqueezeBertModel",rt]],["wav2vec2",["Wav2Vec2Model",xa]],["wav2vec2-bert",["Wav2Vec2BertModel",Va]],["unispeech",["UniSpeechModel",Aa]],["unispeech-sat",["UniSpeechSatModel",Ba]],["hubert",["HubertModel",Ua]],["wavlm",["WavLMModel",Ka]],["audio-spectrogram-transformer",["ASTModel",tn]],["vits",["VitsModel",vs]],["pyannote",["PyAnnoteModel",Sa]],["wespeaker-resnet",["WeSpeakerResNetModel",Ea]],["detr",["DetrModel",li]],["rt_detr",["RTDetrModel",mi]],["table-transformer",["TableTransformerModel",wi]],["vit",["ViTModel",Pr]],["pvt",["PvtModel",Ar]],["vit_msn",["ViTMSNModel",Lr]],["vit_mae",["ViTMAEModel",Or]],["groupvit",["GroupViTModel",Nr]],["fastvit",["FastViTModel",jr]],["mobilevit",["MobileViTModel",Hr]],["mobilevitv2",["MobileViTV2Model",Qr]],["owlvit",["OwlViTModel",Zr]],["owlv2",["Owlv2Model",ni]],["beit",["BeitModel",ai]],["deit",["DeiTModel",xi]],["hiera",["HieraModel",ki]],["convnext",["ConvNextModel",ea]],["convnextv2",["ConvNextV2Model",ra]],["dinov2",["Dinov2Model",sa]],["resnet",["ResNetModel",Ci]],["swin",["SwinModel",Fi]],["swin2sr",["Swin2SRModel",zi]],["donut-swin",["DonutSwinModel",Ji]],["yolos",["YolosModel",ua]],["dpt",["DPTModel",Li]],["glpn",["GLPNModel",Ki]],["hifigan",["SpeechT5HifiGan",is]],["efficientnet",["EfficientNetModel",Es]],["decision_transformer",["DecisionTransformerModel",Ks]],["mobilenet_v1",["MobileNetV1Model",Ls]],["mobilenet_v2",["MobileNetV2Model",Ns]],["mobilenet_v3",["MobileNetV3Model",Gs]],["mobilenet_v4",["MobileNetV4Model",Ws]],["maskformer",["MaskFormerModel",Wi]]]),Js=new Map([["t5",["T5Model",ht]],["longt5",["LongT5Model",gt]],["mt5",["MT5Model",yt]],["bart",["BartModel",xt]],["mbart",["MBartModel",$t]],["marian",["MarianModel",ga]],["whisper",["WhisperModel",an]],["m2m_100",["M2M100Model",ya]],["blenderbot",["BlenderbotModel",Ft]],["blenderbot-small",["BlenderbotSmallModel",zt]]]),Zs=new Map([["bloom",["BloomModel",br]],["jais",["JAISModel",An]],["gpt2",["GPT2Model",Pn]],["gptj",["GPTJModel",Vn]],["gpt_bigcode",["GPTBigCodeModel",qn]],["gpt_neo",["GPTNeoModel",On]],["gpt_neox",["GPTNeoXModel",Dn]],["codegen",["CodeGenModel",Hn]],["llama",["LlamaModel",Qn]],["cohere",["CohereModel",Zn]],["gemma",["GemmaModel",nr]],["gemma2",["Gemma2Model",ar]],["openelm",["OpenELMModel",lr]],["qwen2",["Qwen2Model",cr]],["phi",["PhiModel",mr]],["phi3",["Phi3Model",_r]],["mpt",["MptModel",Mr]],["opt",["OPTModel",$r]],["mistral",["MistralModel",ls]],["starcoder2",["Starcoder2Model",cs]],["falcon",["FalconModel",ms]],["stablelm",["StableLmModel",Ss]]]),eo=new Map([["speecht5",["SpeechT5ForSpeechToText",ns]],["whisper",["WhisperForConditionalGeneration",sn]]]),to=new Map([["speecht5",["SpeechT5ForTextToSpeech",rs]]]),no=new Map([["vits",["VitsModel",vs]],["musicgen",["MusicgenForConditionalGeneration",Os]]]),ro=new Map([["bert",["BertForSequenceClassification",H]],["roformer",["RoFormerForSequenceClassification",te]],["electra",["ElectraForSequenceClassification",he]],["esm",["EsmForSequenceClassification",Ge]],["convbert",["ConvBertForSequenceClassification",oe]],["camembert",["CamembertForSequenceClassification",ye]],["deberta",["DebertaForSequenceClassification",ke]],["deberta-v2",["DebertaV2ForSequenceClassification",Fe]],["mpnet",["MPNetForSequenceClassification",Ze]],["albert",["AlbertForSequenceClassification",ut]],["distilbert",["DistilBertForSequenceClassification",Be]],["roberta",["RobertaForSequenceClassification",Rt]],["xlm",["XLMForSequenceClassification",Ut]],["xlm-roberta",["XLMRobertaForSequenceClassification",Yt]],["bart",["BartForSequenceClassification",Tt]],["mbart",["MBartForSequenceClassification",Ct]],["mobilebert",["MobileBertForSequenceClassification",Xe]],["squeezebert",["SqueezeBertForSequenceClassification",at]]]),io=new Map([["bert",["BertForTokenClassification",X]],["roformer",["RoFormerForTokenClassification",ne]],["electra",["ElectraForTokenClassification",me]],["esm",["EsmForTokenClassification",qe]],["convbert",["ConvBertForTokenClassification",le]],["camembert",["CamembertForTokenClassification",be]],["deberta",["DebertaForTokenClassification",$e]],["deberta-v2",["DebertaV2ForTokenClassification",Ae]],["mpnet",["MPNetForTokenClassification",et]],["distilbert",["DistilBertForTokenClassification",Le]],["roberta",["RobertaForTokenClassification",Nt]],["xlm",["XLMForTokenClassification",Wt]],["xlm-roberta",["XLMRobertaForTokenClassification",Jt]]]),ao=new Map([["t5",["T5ForConditionalGeneration",mt]],["longt5",["LongT5ForConditionalGeneration",_t]],["mt5",["MT5ForConditionalGeneration",bt]],["bart",["BartForConditionalGeneration",Mt]],["mbart",["MBartForConditionalGeneration",St]],["marian",["MarianMTModel",_a]],["m2m_100",["M2M100ForConditionalGeneration",ba]],["blenderbot",["BlenderbotForConditionalGeneration",At]],["blenderbot-small",["BlenderbotSmallForConditionalGeneration",Ot]]]),so=new Map([["bloom",["BloomForCausalLM",vr]],["gpt2",["GPT2LMHeadModel",En]],["jais",["JAISLMHeadModel",In]],["gptj",["GPTJForCausalLM",jn]],["gpt_bigcode",["GPTBigCodeForCausalLM",Un]],["gpt_neo",["GPTNeoForCausalLM",Bn]],["gpt_neox",["GPTNeoXForCausalLM",Rn]],["codegen",["CodeGenForCausalLM",Xn]],["llama",["LlamaForCausalLM",Yn]],["cohere",["CohereForCausalLM",er]],["gemma",["GemmaForCausalLM",rr]],["gemma2",["Gemma2ForCausalLM",sr]],["openelm",["OpenELMForCausalLM",ur]],["qwen2",["Qwen2ForCausalLM",pr]],["phi",["PhiForCausalLM",fr]],["phi3",["Phi3ForCausalLM",wr]],["mpt",["MptForCausalLM",Tr]],["opt",["OPTForCausalLM",Sr]],["mbart",["MBartForCausalLM",Pt]],["mistral",["MistralForCausalLM",us]],["starcoder2",["Starcoder2ForCausalLM",ps]],["falcon",["FalconForCausalLM",fs]],["trocr",["TrOCRForCausalLM",ss]],["stablelm",["StableLmForCausalLM",Cs]]]),oo=new Map([["bert",["BertForMaskedLM",W]],["roformer",["RoFormerForMaskedLM",ee]],["electra",["ElectraForMaskedLM",pe]],["esm",["EsmForMaskedLM",je]],["convbert",["ConvBertForMaskedLM",se]],["camembert",["CamembertForMaskedLM",we]],["deberta",["DebertaForMaskedLM",Te]],["deberta-v2",["DebertaV2ForMaskedLM",Ee]],["mpnet",["MPNetForMaskedLM",Je]],["albert",["AlbertForMaskedLM",ct]],["distilbert",["DistilBertForMaskedLM",Re]],["roberta",["RobertaForMaskedLM",Dt]],["xlm",["XLMWithLMHeadModel",qt]],["xlm-roberta",["XLMRobertaForMaskedLM",Qt]],["mobilebert",["MobileBertForMaskedLM",He]],["squeezebert",["SqueezeBertForMaskedLM",it]]]),lo=new Map([["bert",["BertForQuestionAnswering",K]],["roformer",["RoFormerForQuestionAnswering",re]],["electra",["ElectraForQuestionAnswering",fe]],["convbert",["ConvBertForQuestionAnswering",ue]],["camembert",["CamembertForQuestionAnswering",ve]],["deberta",["DebertaForQuestionAnswering",Se]],["deberta-v2",["DebertaV2ForQuestionAnswering",Ie]],["mpnet",["MPNetForQuestionAnswering",tt]],["albert",["AlbertForQuestionAnswering",dt]],["distilbert",["DistilBertForQuestionAnswering",De]],["roberta",["RobertaForQuestionAnswering",Vt]],["xlm",["XLMForQuestionAnswering",Ht]],["xlm-roberta",["XLMRobertaForQuestionAnswering",Zt]],["mobilebert",["MobileBertForQuestionAnswering",Ke]],["squeezebert",["SqueezeBertForQuestionAnswering",st]]]),uo=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",on]]]),co=new Map([["llava",["LlavaForConditionalGeneration",un]],["moondream1",["Moondream1ForConditionalGeneration",dn]],["florence2",["Florence2ForConditionalGeneration",pn]]]),po=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",on]]]),ho=new Map([["vit",["ViTForImageClassification",Er]],["pvt",["PvtForImageClassification",Ir]],["vit_msn",["ViTMSNForImageClassification",Dr]],["fastvit",["FastViTForImageClassification",Gr]],["mobilevit",["MobileViTForImageClassification",Xr]],["mobilevitv2",["MobileViTV2ForImageClassification",Yr]],["beit",["BeitForImageClassification",si]],["deit",["DeiTForImageClassification",Mi]],["hiera",["HieraForImageClassification",$i]],["convnext",["ConvNextForImageClassification",ta]],["convnextv2",["ConvNextV2ForImageClassification",ia]],["dinov2",["Dinov2ForImageClassification",oa]],["resnet",["ResNetForImageClassification",Pi]],["swin",["SwinForImageClassification",Ai]],["segformer",["SegformerForImageClassification",Ts]],["efficientnet",["EfficientNetForImageClassification",Fs]],["mobilenet_v1",["MobileNetV1ForImageClassification",Ds]],["mobilenet_v2",["MobileNetV2ForImageClassification",Vs]],["mobilenet_v3",["MobileNetV3ForImageClassification",qs]],["mobilenet_v4",["MobileNetV4ForImageClassification",Hs]]]),mo=new Map([["detr",["DetrForObjectDetection",ui]],["rt_detr",["RTDetrForObjectDetection",fi]],["table-transformer",["TableTransformerForObjectDetection",yi]],["yolos",["YolosForObjectDetection",da]]]),fo=new Map([["owlvit",["OwlViTForObjectDetection",ei]],["owlv2",["Owlv2ForObjectDetection",ri]]]),go=new Map([["detr",["DetrForSegmentation",di]],["clipseg",["CLIPSegForImageSegmentation",Sn]]]),_o=new Map([["segformer",["SegformerForSemanticSegmentation",ks]],["sapiens",["SapiensForSemanticSegmentation",ji]]]),wo=new Map([["detr",["DetrForSegmentation",di]],["maskformer",["MaskFormerForInstanceSegmentation",Hi]]]),yo=new Map([["sam",["SamModel",ha]]]),bo=new Map([["wav2vec2",["Wav2Vec2ForCTC",Ma]],["wav2vec2-bert",["Wav2Vec2BertForCTC",ja]],["unispeech",["UniSpeechForCTC",Ia]],["unispeech-sat",["UniSpeechSatForCTC",La]],["wavlm",["WavLMForCTC",Qa]],["hubert",["HubertForCTC",Wa]]]),vo=new Map([["wav2vec2",["Wav2Vec2ForSequenceClassification",Ta]],["wav2vec2-bert",["Wav2Vec2BertForSequenceClassification",Ga]],["unispeech",["UniSpeechForSequenceClassification",za]],["unispeech-sat",["UniSpeechSatForSequenceClassification",Da]],["wavlm",["WavLMForSequenceClassification",Ya]],["hubert",["HubertForSequenceClassification",Ha]],["audio-spectrogram-transformer",["ASTForAudioClassification",nn]]]),xo=new Map([["wavlm",["WavLMForXVector",Ja]]]),Mo=new Map([["unispeech-sat",["UniSpeechSatForAudioFrameClassification",Ra]],["wavlm",["WavLMForAudioFrameClassification",Za]],["wav2vec2",["Wav2Vec2ForAudioFrameClassification",ka]],["pyannote",["PyAnnoteForAudioFrameClassification",Ca]]]),To=new Map([["vitmatte",["VitMatteForImageMatting",Ur]]]),ko=new Map([["swin2sr",["Swin2SRForImageSuperResolution",Oi]]]),$o=new Map([["dpt",["DPTForDepthEstimation",Di]],["depth_anything",["DepthAnythingForDepthEstimation",Ni]],["glpn",["GLPNForDepthEstimation",Qi]],["sapiens",["SapiensForDepthEstimation",Gi]]]),So=new Map([["sapiens",["SapiensForNormalEstimation",qi]]]),Co=new Map([["clip",["CLIPVisionModelWithProjection",wn]],["siglip",["SiglipVisionModel",xn]]]),Po=[[Ys,w],[Js,y],[Zs,x],[ro,w],[io,w],[ao,b],[eo,b],[so,x],[oo,w],[lo,w],[uo,v],[co,T],[ho,w],[go,w],[wo,w],[_o,w],[To,w],[ko,w],[$o,w],[So,w],[mo,w],[fo,w],[yo,M],[bo,w],[vo,w],[to,b],[no,w],[xo,w],[Mo,w],[Co,w]];for(const[e,t]of Po)for(const[n,r]of e.values())$.set(n,t),C.set(r,n),S.set(n,r);const Eo=[["MusicgenForConditionalGeneration",Os,k],["CLIPTextModelWithProjection",gn,w],["SiglipTextModel",vn,w],["ClapTextModelWithProjection",ws,w],["ClapAudioModelWithProjection",ys,w]];for(const[e,t,n]of Eo)$.set(e,n),C.set(t,e),S.set(e,t);class Fo extends Qs{static MODEL_CLASS_MAPPINGS=Po.map((e=>e[0]));static BASE_IF_FAIL=!0}class Ao extends Qs{static MODEL_CLASS_MAPPINGS=[ro]}class Io extends Qs{static MODEL_CLASS_MAPPINGS=[io]}class zo extends Qs{static MODEL_CLASS_MAPPINGS=[ao]}class Oo extends Qs{static MODEL_CLASS_MAPPINGS=[eo]}class Bo extends Qs{static MODEL_CLASS_MAPPINGS=[to]}class Lo extends Qs{static MODEL_CLASS_MAPPINGS=[no]}class Do extends Qs{static MODEL_CLASS_MAPPINGS=[so]}class Ro extends Qs{static MODEL_CLASS_MAPPINGS=[oo]}class No extends Qs{static MODEL_CLASS_MAPPINGS=[lo]}class Vo extends Qs{static MODEL_CLASS_MAPPINGS=[uo]}class jo extends Qs{static MODEL_CLASS_MAPPINGS=[ho]}class Go extends Qs{static MODEL_CLASS_MAPPINGS=[go]}class qo extends Qs{static MODEL_CLASS_MAPPINGS=[_o]}class Uo extends Qs{static MODEL_CLASS_MAPPINGS=[wo]}class Wo extends Qs{static MODEL_CLASS_MAPPINGS=[mo]}class Ho extends Qs{static MODEL_CLASS_MAPPINGS=[fo]}class Xo extends Qs{static MODEL_CLASS_MAPPINGS=[yo]}class Ko extends Qs{static MODEL_CLASS_MAPPINGS=[bo]}class Qo extends Qs{static MODEL_CLASS_MAPPINGS=[vo]}class Yo extends Qs{static MODEL_CLASS_MAPPINGS=[xo]}class Jo extends Qs{static MODEL_CLASS_MAPPINGS=[Mo]}class Zo extends Qs{static MODEL_CLASS_MAPPINGS=[po]}class el extends Qs{static MODEL_CLASS_MAPPINGS=[To]}class tl extends Qs{static MODEL_CLASS_MAPPINGS=[ko]}class nl extends Qs{static MODEL_CLASS_MAPPINGS=[$o]}class rl extends Qs{static MODEL_CLASS_MAPPINGS=[So]}class il extends Qs{static MODEL_CLASS_MAPPINGS=[Co]}class al extends j{constructor({logits:e,past_key_values:t,encoder_outputs:n,decoder_attentions:r=null,cross_attentions:i=null}){super(),this.logits=e,this.past_key_values=t,this.encoder_outputs=n,this.decoder_attentions=r,this.cross_attentions=i}}class sl extends j{constructor({logits:e}){super(),this.logits=e}}class ol extends j{constructor({logits:e,embeddings:t}){super(),this.logits=e,this.embeddings=t}}class ll extends j{constructor({logits:e}){super(),this.logits=e}}class ul extends j{constructor({logits:e}){super(),this.logits=e}}class dl extends j{constructor({start_logits:e,end_logits:t}){super(),this.start_logits=e,this.end_logits=t}}class cl extends j{constructor({logits:e}){super(),this.logits=e}}class pl extends j{constructor({logits:e,past_key_values:t}){super(),this.logits=e,this.past_key_values=t}}class hl extends j{constructor({alphas:e}){super(),this.alphas=e}}class ml extends j{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)=>{n.r(t),n.d(t,{WHISPER_LANGUAGE_MAPPING:()=>i,WHISPER_TO_LANGUAGE_CODE_MAPPING:()=>a,whisper_language_to_code:()=>s});const r=[["en","english"],["zh","chinese"],["de","german"],["es","spanish"],["ru","russian"],["ko","korean"],["fr","french"],["ja","japanese"],["pt","portuguese"],["tr","turkish"],["pl","polish"],["ca","catalan"],["nl","dutch"],["ar","arabic"],["sv","swedish"],["it","italian"],["id","indonesian"],["hi","hindi"],["fi","finnish"],["vi","vietnamese"],["he","hebrew"],["uk","ukrainian"],["el","greek"],["ms","malay"],["cs","czech"],["ro","romanian"],["da","danish"],["hu","hungarian"],["ta","tamil"],["no","norwegian"],["th","thai"],["ur","urdu"],["hr","croatian"],["bg","bulgarian"],["lt","lithuanian"],["la","latin"],["mi","maori"],["ml","malayalam"],["cy","welsh"],["sk","slovak"],["te","telugu"],["fa","persian"],["lv","latvian"],["bn","bengali"],["sr","serbian"],["az","azerbaijani"],["sl","slovenian"],["kn","kannada"],["et","estonian"],["mk","macedonian"],["br","breton"],["eu","basque"],["is","icelandic"],["hy","armenian"],["ne","nepali"],["mn","mongolian"],["bs","bosnian"],["kk","kazakh"],["sq","albanian"],["sw","swahili"],["gl","galician"],["mr","marathi"],["pa","punjabi"],["si","sinhala"],["km","khmer"],["sn","shona"],["yo","yoruba"],["so","somali"],["af","afrikaans"],["oc","occitan"],["ka","georgian"],["be","belarusian"],["tg","tajik"],["sd","sindhi"],["gu","gujarati"],["am","amharic"],["yi","yiddish"],["lo","lao"],["uz","uzbek"],["fo","faroese"],["ht","haitian creole"],["ps","pashto"],["tk","turkmen"],["nn","nynorsk"],["mt","maltese"],["sa","sanskrit"],["lb","luxembourgish"],["my","myanmar"],["bo","tibetan"],["tl","tagalog"],["mg","malagasy"],["as","assamese"],["tt","tatar"],["haw","hawaiian"],["ln","lingala"],["ha","hausa"],["ba","bashkir"],["jw","javanese"],["su","sundanese"]],i=new Map(r),a=new Map([...r.map((([e,t])=>[t,e])),["burmese","my"],["valencian","ca"],["flemish","nl"],["haitian","ht"],["letzeburgesch","lb"],["pushto","ps"],["panjabi","pa"],["moldavian","ro"],["moldovan","ro"],["sinhalese","si"],["castilian","es"]]);function s(e){e=e.toLowerCase();let t=a.get(e);if(void 0===t){if(!i.has(e)){const t=2===e.length?i.keys():i.values();throw new Error(`Language "${e}" is not supported. Must be one of: ${JSON.stringify(t)}`)}t=e}return t}},"./src/models/whisper/generation_whisper.js":
@@ -190,7 +190,7 @@ var r,i,a,s,o,l,u,d,c,p,h,m,f,g,_,w,y,b,v,x,M,T,k,$,S,C,P,E,F,A,I,z,O,B=Object.d
   \***************************/(e,t,n)=>{n.r(t),n.d(t,{ASTFeatureExtractor:()=>ae,AutoProcessor:()=>we,BeitFeatureExtractor:()=>H,BitImageProcessor:()=>$,CLIPFeatureExtractor:()=>C,CLIPImageProcessor:()=>P,ChineseCLIPFeatureExtractor:()=>E,ClapFeatureExtractor:()=>se,ConvNextFeatureExtractor:()=>A,ConvNextImageProcessor:()=>I,DPTFeatureExtractor:()=>T,DPTImageProcessor:()=>k,DeiTFeatureExtractor:()=>W,DetrFeatureExtractor:()=>Q,DonutFeatureExtractor:()=>X,EfficientNetImageProcessor:()=>B,FeatureExtractor:()=>y,Florence2Processor:()=>_e,GLPNFeatureExtractor:()=>S,ImageFeatureExtractor:()=>b,MaskFormerFeatureExtractor:()=>Y,MobileNetV1FeatureExtractor:()=>L,MobileNetV2FeatureExtractor:()=>D,MobileNetV3FeatureExtractor:()=>R,MobileNetV4FeatureExtractor:()=>N,MobileViTFeatureExtractor:()=>V,MobileViTImageProcessor:()=>j,NougatImageProcessor:()=>K,OwlViTFeatureExtractor:()=>G,OwlViTProcessor:()=>ge,Owlv2ImageProcessor:()=>q,Processor:()=>de,PvtImageProcessor:()=>M,PyAnnoteFeatureExtractor:()=>oe,PyAnnoteProcessor:()=>me,RTDetrImageProcessor:()=>U,SamImageProcessor:()=>Z,SamProcessor:()=>ce,SapiensFeatureExtractor:()=>v,SeamlessM4TFeatureExtractor:()=>ie,SegformerFeatureExtractor:()=>x,SiglipImageProcessor:()=>F,SpeechT5FeatureExtractor:()=>ue,SpeechT5Processor:()=>fe,Swin2SRImageProcessor:()=>ee,ViTFeatureExtractor:()=>z,ViTImageProcessor:()=>O,VitMatteImageProcessor:()=>te,Wav2Vec2FeatureExtractor:()=>re,Wav2Vec2ProcessorWithLM:()=>he,WeSpeakerFeatureExtractor:()=>le,WhisperFeatureExtractor:()=>ne,WhisperProcessor:()=>pe,YolosFeatureExtractor:()=>J});var r=n(/*! ./utils/generic.js */"./src/utils/generic.js"),i=n(/*! ./utils/core.js */"./src/utils/core.js"),a=n(/*! ./utils/hub.js */"./src/utils/hub.js"),s=n(/*! ./utils/maths.js */"./src/utils/maths.js"),o=n(/*! ./utils/tensor.js */"./src/utils/tensor.js"),l=(n(/*! ./utils/image.js */"./src/utils/image.js"),n(/*! ./utils/audio.js */"./src/utils/audio.js"));function u([e,t,n,r]){return[e-n/2,t-r/2,e+n/2,t+r/2]}function d(e,t=.5,n=null,r=!1){const i=e.logits,a=e.pred_boxes,[o,l,d]=i.dims;if(null!==n&&n.length!==o)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");let c=[];for(let e=0;e<o;++e){let o=null!==n?n[e]:null,p={boxes:[],classes:[],scores:[]},h=i[e],m=a[e];for(let e=0;e<l;++e){let n,i=h[e],a=[];if(r){n=i.sigmoid().data;for(let e=0;e<n.length;++e)n[e]>t&&a.push(e)}else{let e=(0,s.max)(i.data)[1];if(e===d-1)continue;if(n=(0,s.softmax)(i.data),n[e]<t)continue;a.push(e)}for(const t of a){let r=m[e].data;r=u(r),null!==o&&(r=r.map(((e,t)=>e*o[(t+1)%2]))),p.boxes.push(r),p.classes.push(t),p.scores.push(n[t])}}c.push(p)}return c}function c(e,t=null){const n=e.logits,r=n.dims[0];if(null!==t&&t.length!==r)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");const i=[];for(let e=0;e<r;++e){const r=null!==t?t[e]:null;let a=n[e];null!==r&&(a=(0,o.interpolate)(a,r,"bilinear",!1));const[s,l]=r??a.dims.slice(-2),u=new o.Tensor("int32",new Int32Array(s*l),[s,l]),d=a[0].data,c=u.data;for(let e=1;e<a.dims[0];++e){const t=a[e].data;for(let n=0;n<t.length;++n)t[n]>d[n]&&(d[n]=t[n],c[n]=e)}const p=new Array(a.dims[0]);for(let e=0;e<c.length;++e){const t=c[e];p[t]=t}const h=p.filter((e=>void 0!==e));i.push({segmentation:u,labels:h})}return i}function p(e,t,n,r){const i=[],a=[],o=[];for(let l=0;l<e.dims[0];++l){const u=e[l],d=t[l],c=(0,s.max)(u.data)[1];if(c===r)continue;const p=(0,s.softmax)(u.data)[c];p>n&&(i.push(d),a.push(p),o.push(c))}return[i,a,o]}function h(e,t,n,r=.5,i=.8){const a=[];let s=0,o=0;const l=t[n].data;for(let t=0;t<e.length;++t)e[t]===n&&(a.push(t),++s),l[t]>=r&&++o;let u=s>0&&o>0;if(u){u=s/o>i}return[u,a]}function m(e,t,n,r,i,a=null,s=null){const[l,u]=s??e[0].dims,d=new o.Tensor("int32",new Int32Array(l*u),[l,u]),c=[];if(null!==s)for(let t=0;t<e.length;++t)e[t]=(0,o.interpolate)(e[t],s,"bilinear",!1);const p=new Int32Array(e[0].data.length),m=new Float32Array(e[0].data.length);for(let n=0;n<e.length;++n){let r=t[n];const i=e[n].data;for(let e=0;e<i.length;++e)i[e]*=r,i[e]>m[e]&&(p[e]=n,m[e]=i[e])}let f=0;const g=d.data;for(let a=0;a<n.length;++a){const s=n[a],[o,l]=h(p,e,a,r,i);if(o){++f;for(const e of l)g[e]=f;c.push({id:f,label_id:s,score:t[a]})}}return[d,c]}function f(e,t=.5,n=.5,r=.8,i=null,a=null){null===i&&(console.warn("`label_ids_to_fuse` unset. No instance will be fused."),i=new Set);const s=e.class_queries_logits??e.logits,l=(e.masks_queries_logits??e.pred_masks).sigmoid();let[u,d,c]=s.dims;if(c-=1,null!==a&&a.length!==u)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");let h=[];for(let e=0;e<u;++e){let u=null!==a?a[e]:null,d=s[e],f=l[e],[g,_,w]=p(d,f,t,c);if(0===w.length){let[e,t]=u??f.dims.slice(-2),n=new o.Tensor("int32",new Int32Array(e*t).fill(-1),[e,t]);h.push({segmentation:n,segments_info:[]});continue}let[y,b]=m(g,_,w,n,r,i,u);h.push({segmentation:y,segments_info:b})}return h}function g(e,t){if(!(e instanceof Float32Array||e instanceof Float64Array))throw new Error(`${t} expects input to be a Float32Array or a Float64Array, but got ${e?.constructor?.name??typeof e} instead. If using the feature extractor directly, remember to use \`read_audio(url, sampling_rate)\` to obtain the raw audio data of the file/url.`)}function _(e,t,n=0,r=null){const i=e/t;let a=(0,s.bankers_round)(i)*t;return null!==r&&a>r&&(a=Math.floor(i)*t),a<n&&(a=Math.ceil(i)*t),a}function w([e,t],n){return[Math.max(Math.floor(e/n),1)*n,Math.max(Math.floor(t/n),1)*n]}class y extends r.Callable{constructor(e){super(),this.config=e}}class b extends y{constructor(e){super(e),this.image_mean=this.config.image_mean??this.config.mean,this.image_std=this.config.image_std??this.config.std,this.resample=this.config.resample??2,this.do_rescale=this.config.do_rescale??!0,this.rescale_factor=this.config.rescale_factor??1/255,this.do_normalize=this.config.do_normalize,this.do_resize=this.config.do_resize,this.do_thumbnail=this.config.do_thumbnail,this.size=this.config.size,this.size_divisibility=this.config.size_divisibility??this.config.size_divisor,this.do_center_crop=this.config.do_center_crop,this.crop_size=this.config.crop_size,this.do_convert_rgb=this.config.do_convert_rgb??!0,this.do_crop_margin=this.config.do_crop_margin,this.pad_size=this.config.pad_size,this.do_pad=this.config.do_pad,this.do_pad&&!this.pad_size&&this.size&&void 0!==this.size.width&&void 0!==this.size.height&&(this.pad_size=this.size),this.do_flip_channel_order=this.config.do_flip_channel_order??!1}async thumbnail(e,t,n=2){const r=e.height,i=e.width,a=t.height,s=t.width;let o=Math.min(r,a),l=Math.min(i,s);return o===r&&l===i?e:(r>i?l=Math.floor(i*o/r):i>r&&(o=Math.floor(r*l/i)),await e.resize(l,o,{resample:n}))}async crop_margin(e,t=200){const n=e.clone().grayscale(),r=(0,s.min)(n.data)[0],i=(0,s.max)(n.data)[0]-r;if(0===i)return e;const a=t/255;let o=n.width,l=n.height,u=0,d=0;const c=n.data;for(let e=0;e<n.height;++e){const t=e*n.width;for(let s=0;s<n.width;++s)(c[t+s]-r)/i<a&&(o=Math.min(o,s),l=Math.min(l,e),u=Math.max(u,s),d=Math.max(d,e))}return e=await e.crop([o,l,u,d])}pad_image(e,t,n,{mode:r="constant",center:a=!1,constant_values:s=0}={}){const[o,l,u]=t;let d,c;if("number"==typeof n?(d=n,c=n):(d=n.width,c=n.height),d!==l||c!==o){const n=new Float32Array(d*c*u);if(Array.isArray(s))for(let e=0;e<n.length;++e)n[e]=s[e%u];else 0!==s&&n.fill(s);const[p,h]=a?[Math.floor((d-l)/2),Math.floor((c-o)/2)]:[0,0];for(let t=0;t<o;++t){const r=(t+h)*d,i=t*l;for(let t=0;t<l;++t){const a=(r+t+p)*u,s=(i+t)*u;for(let t=0;t<u;++t)n[a+t]=e[s+t]}}if("symmetric"===r){if(a)throw new Error("`center` padding is not supported when `mode` is set to `symmetric`.");const t=o-1,r=l-1;for(let a=0;a<c;++a){const s=a*d,c=(0,i.calculateReflectOffset)(a,t)*l;for(let t=0;t<d;++t){if(a<o&&t<l)continue;const d=(s+t)*u,p=(c+(0,i.calculateReflectOffset)(t,r))*u;for(let t=0;t<u;++t)n[d+t]=e[p+t]}}}e=n,t=[c,d,u]}return[e,t]}rescale(e){for(let t=0;t<e.length;++t)e[t]=this.rescale_factor*e[t]}get_resize_output_image_size(e,t){const[n,r]=e.size;let i,a;if(this.do_thumbnail){const{height:e,width:n}=t;i=Math.min(e,n)}else Number.isInteger(t)?(i=t,a=this.config.max_size??i):void 0!==t&&(i=t.shortest_edge,a=t.longest_edge);if(void 0!==i||void 0!==a){const e=void 0===i?1:Math.max(i/n,i/r),t=n*e,s=r*e,o=void 0===a?1:Math.min(a/t,a/s);let l=Math.floor(Number((t*o).toFixed(2))),u=Math.floor(Number((s*o).toFixed(2)));return void 0!==this.size_divisibility&&([l,u]=w([l,u],this.size_divisibility)),[l,u]}if(void 0!==t&&void 0!==t.width&&void 0!==t.height){let e=t.width,i=t.height;if(this.config.keep_aspect_ratio&&this.config.ensure_multiple_of){let t=i/r,a=e/n;Math.abs(1-a)<Math.abs(1-t)?t=a:a=t,i=_(t*r,this.config.ensure_multiple_of),e=_(a*n,this.config.ensure_multiple_of)}return[e,i]}if(void 0!==this.size_divisibility)return w([n,r],this.size_divisibility);throw new Error(`Could not resize image due to unsupported \`this.size\` option in config: ${JSON.stringify(t)}`)}async resize(e){const[t,n]=this.get_resize_output_image_size(e,this.size);return await e.resize(t,n,{resample:this.resample})}async preprocess(e,{do_normalize:t=null,do_pad:n=null,do_convert_rgb:r=null,do_convert_grayscale:i=null,do_flip_channel_order:a=null}={}){this.do_crop_margin&&(e=await this.crop_margin(e));const[s,l]=e.size;if(r??this.do_convert_rgb?e=e.rgb():i&&(e=e.grayscale()),this.do_resize&&(e=await this.resize(e)),this.do_thumbnail&&(e=await this.thumbnail(e,this.size,this.resample)),this.do_center_crop){let t,n;Number.isInteger(this.crop_size)?(t=this.crop_size,n=this.crop_size):(t=this.crop_size.width,n=this.crop_size.height),e=await e.center_crop(t,n)}const u=[e.height,e.width];let d=Float32Array.from(e.data),c=[e.height,e.width,e.channels];if(this.do_rescale&&this.rescale(d),t??this.do_normalize){let t=this.image_mean;Array.isArray(this.image_mean)||(t=new Array(e.channels).fill(t));let n=this.image_std;if(Array.isArray(this.image_std)||(n=new Array(e.channels).fill(t)),t.length!==e.channels||n.length!==e.channels)throw new Error(`When set to arrays, the length of \`image_mean\` (${t.length}) and \`image_std\` (${n.length}) must match the number of channels in the image (${e.channels}).`);for(let r=0;r<d.length;r+=e.channels)for(let i=0;i<e.channels;++i)d[r+i]=(d[r+i]-t[i])/n[i]}if(n??this.do_pad)if(this.pad_size){const t=this.pad_image(d,[e.height,e.width,e.channels],this.pad_size);[d,c]=t}else if(this.size_divisibility){const[e,t]=w([c[1],c[0]],this.size_divisibility);[d,c]=this.pad_image(d,c,{width:e,height:t})}if(a??this.do_flip_channel_order){if(3!==c[2])throw new Error("Flipping channel order is only supported for RGB images.");for(let e=0;e<d.length;e+=3){const t=d[e];d[e]=d[e+2],d[e+2]=t}}return{original_size:[l,s],reshaped_input_size:u,pixel_values:new o.Tensor("float32",d,c).permute(2,0,1)}}async _call(e,...t){Array.isArray(e)||(e=[e]);const n=await Promise.all(e.map((e=>this.preprocess(e))));return{pixel_values:(0,o.stack)(n.map((e=>e.pixel_values)),0),original_sizes:n.map((e=>e.original_size)),reshaped_input_sizes:n.map((e=>e.reshaped_input_size))}}}class v extends b{post_process_semantic_segmentation(...e){return c(...e)}}class x extends b{post_process_semantic_segmentation(...e){return c(...e)}}class M extends b{}class T extends b{}class k extends T{}class $ extends b{}class S extends b{}class C extends b{}class P extends C{}class E extends b{}class F extends b{}class A extends b{constructor(e){super(e),this.crop_pct=this.config.crop_pct??.875}async resize(e){const t=this.size?.shortest_edge;if(void 0===t)throw new Error("Size dictionary must contain 'shortest_edge' key.");if(t<384){const n=Math.floor(t/this.crop_pct),[r,i]=this.get_resize_output_image_size(e,{shortest_edge:n});e=await e.resize(r,i,{resample:this.resample}),e=await e.center_crop(t,t)}else e=await e.resize(t,t,{resample:this.resample});return e}}class I extends A{}class z extends b{}class O extends b{}class B extends b{constructor(e){super(e),this.include_top=this.config.include_top??!0,this.include_top&&(this.image_std=this.image_std.map((e=>e*e)))}}class L extends b{}class D extends b{}class R extends b{}class N extends b{}class V extends b{}class j extends V{}class G extends b{post_process_object_detection(...e){return d(...e)}}class q extends G{}class U extends b{post_process_object_detection(...e){return d(...e)}}class W extends b{}class H extends b{}class X extends b{pad_image(e,t,n,r={}){const[i,a,s]=t;let o=this.image_mean;Array.isArray(this.image_mean)||(o=new Array(s).fill(o));let l=this.image_std;Array.isArray(l)||(l=new Array(s).fill(o));const u=o.map(((e,t)=>-e/l[t]));return super.pad_image(e,t,n,{center:!0,constant_values:u,...r})}}class K extends X{}class Q extends b{async _call(e){const t=await super._call(e),n=[t.pixel_values.dims[0],64,64],r=(0,o.full)(n,1n);return{...t,pixel_mask:r}}post_process_object_detection(...e){return d(...e)}post_process_panoptic_segmentation(...e){return f(...e)}post_process_instance_segmentation(){throw Error("Not implemented yet")}}class Y extends b{post_process_panoptic_segmentation(...e){return f(...e)}post_process_instance_segmentation(){throw Error("Not implemented yet")}}class J extends b{post_process_object_detection(...e){return d(...e)}}class Z extends b{reshape_input_points(e,t,n,r=!1){e=structuredClone(e);let a=(0,i.calculateDimensions)(e);if(3===a.length)r||(a=[1,...a]),e=[e];else if(4!==a.length)throw Error("The input_points must be a 4D tensor of shape `batch_size`, `point_batch_size`, `nb_points_per_image`, `2`.");for(let r=0;r<e.length;++r){let i=t[r],a=n[r],s=[a[0]/i[0],a[1]/i[1]];for(let t=0;t<e[r].length;++t)for(let n=0;n<e[r][t].length;++n)for(let i=0;i<e[r][t][n].length;++i)e[r][t][n][i]*=s[i%2]}return new o.Tensor("float32",Float32Array.from(e.flat(1/0)),a)}add_input_labels(e,t){let n=(0,i.calculateDimensions)(e);if(2===n.length)n=[1,...n],e=[e];else if(3!==n.length)throw Error("The input_points must be a 4D tensor of shape `batch_size`, `point_batch_size`, `nb_points_per_image`, `2`.");if(n.some(((e,n)=>e!==t.dims[n])))throw Error(`The first ${n.length} dimensions of 'input_points' and 'input_labels' must be the same.`);return new o.Tensor("int64",e.flat(1/0).map(BigInt),n)}async _call(e,{input_points:t=null,input_labels:n=null,input_boxes:r=null}={}){const i=await super._call(e);if(t&&(i.input_points=this.reshape_input_points(t,i.original_sizes,i.reshaped_input_sizes)),n){if(!i.input_points)throw Error("`input_points` must be provided if `input_labels` are provided.");i.input_labels=this.add_input_labels(n,i.input_points)}return r&&(i.input_boxes=this.reshape_input_points(r,i.original_sizes,i.reshaped_input_sizes,!0)),i}async post_process_masks(e,t,n,{mask_threshold:r=0,binarize:i=!0,pad_size:a=null}={}){const s=[],l=[(a=a??this.pad_size).height,a.width];for(let a=0;a<t.length;++a){const u=t[a],d=n[a];let c=await(0,o.interpolate_4d)(e[a],{mode:"bilinear",size:l});if(c=c.slice(null,null,[0,d[0]],[0,d[1]]),c=await(0,o.interpolate_4d)(c,{mode:"bilinear",size:u}),i){const e=c.data,t=new Uint8Array(e.length);for(let n=0;n<e.length;++n)e[n]>r&&(t[n]=1);c=new o.Tensor("bool",t,c.dims)}s.push(c)}return s}generate_crop_boxes(e,t,{crop_n_layers:n=0,overlap_ratio:r=512/1500,points_per_crop:i=32,crop_n_points_downscale_factor:a=1}={}){}}class ee extends b{pad_image(e,t,n,r={}){const[i,a,s]=t;return super.pad_image(e,t,{width:a+(n-a%n)%n,height:i+(n-i%n)%n},{mode:"symmetric",center:!1,constant_values:-1,...r})}}class te extends b{async _call(e,t){Array.isArray(e)||(e=[e]),Array.isArray(t)||(t=[t]);const n=await Promise.all(e.map((e=>this.preprocess(e)))),r=await Promise.all(t.map((e=>this.preprocess(e,{do_normalize:!1,do_convert_rgb:!1,do_convert_grayscale:!0}))));return{pixel_values:(0,o.stack)(n.map(((e,t)=>(0,o.cat)([e.pixel_values,r[t].pixel_values],0))),0),original_sizes:n.map((e=>e.original_size)),reshaped_input_sizes:n.map((e=>e.reshaped_input_size))}}}class ne extends y{constructor(e){super(e),this.config.mel_filters??=(0,l.mel_filter_bank)(Math.floor(1+this.config.n_fft/2),this.config.feature_size,0,8e3,this.config.sampling_rate,"slaney","slaney"),this.window=(0,l.window_function)(this.config.n_fft,"hann")}async _extract_fbank_features(e){const t=await(0,l.spectrogram)(e,this.window,this.config.n_fft,this.config.hop_length,{power:2,mel_filters:this.config.mel_filters,log_mel:"log10",max_num_frames:this.config.nb_max_frames}),n=t.data,r=(0,s.max)(n)[0];for(let e=0;e<n.length;++e)n[e]=(Math.max(n[e],r-8)+4)/4;return t}async _call(e){let t;g(e,"WhisperFeatureExtractor"),e.length>this.config.n_samples?(console.warn("Attempting to extract features for audio longer than 30 seconds. If using a pipeline to extract transcript from a long audio clip, remember to specify `chunk_length_s` and/or `stride_length_s`."),t=e.slice(0,this.config.n_samples)):(t=new Float32Array(this.config.n_samples),t.set(e));return{input_features:(await this._extract_fbank_features(t)).unsqueeze_(0)}}}class re extends y{_zero_mean_unit_var_norm(e){const t=e.reduce(((e,t)=>e+t),0)/e.length,n=e.reduce(((e,n)=>e+(n-t)**2),0)/e.length;return e.map((e=>(e-t)/Math.sqrt(n+1e-7)))}async _call(e){g(e,"Wav2Vec2FeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));let t=e;this.config.do_normalize&&(t=this._zero_mean_unit_var_norm(t));const n=[1,t.length];return{input_values:new o.Tensor("float32",t,n),attention_mask:new o.Tensor("int64",new BigInt64Array(t.length).fill(1n),n)}}}class ie extends y{constructor(e){super(e);const t=this.config.sampling_rate,n=(0,l.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<n.length;++e)n[e].push(0);this.mel_filters=n,this.window=(0,l.window_function)(400,"povey",{periodic:!1})}async _extract_fbank_features(e,t){return e=e.map((e=>32768*e)),(0,l.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,max_num_frames:t,transpose:!0})}async _call(e,{padding:t=!0,pad_to_multiple_of:n=2,do_normalize_per_mel_bins:r=!0,return_attention_mask:i=!0}={}){g(e,"SeamlessM4TFeatureExtractor");let a,s=await this._extract_fbank_features(e,this.config.max_length);if(r){const[e,t]=s.dims,n=s.data;for(let r=0;r<t;++r){let i=0;for(let a=0;a<e;++a)i+=n[a*t+r];const a=i/e;let s=0;for(let i=0;i<e;++i)s+=(n[i*t+r]-a)**2;s/=e-1;const o=Math.sqrt(s+1e-7);for(let i=0;i<e;++i){const e=i*t+r;n[e]=(n[e]-a)/o}}}if(t){const[e,t]=s.dims,r=s.data,l=e%n;if(l>0){const n=new Float32Array(t*(e+l));n.set(r),n.fill(this.config.padding_value,r.length);const u=e+l;s=new o.Tensor(s.type,n,[u,t]),i&&(a=new o.Tensor("int64",new BigInt64Array(u),[1,u]),a.data.fill(1n,0,e))}}const[l,u]=s.dims,d=this.config.stride;if(0!==l%d)throw new Error(`The number of frames (${l}) must be a multiple of the stride (${d}).`);const c=s.view(1,Math.floor(l/d),u*d),p={input_features:c};if(i){const e=c.dims[1],t=new BigInt64Array(e);if(a){const e=a.data;for(let n=1,r=0;n<l;n+=d,++r)t[r]=e[n]}else t.fill(1n);p.attention_mask=new o.Tensor("int64",t,[1,e])}return p}}class ae extends y{constructor(e){super(e);const t=this.config.sampling_rate,n=(0,l.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<n.length;++e)n[e].push(0);this.mel_filters=n,this.window=(0,l.window_function)(400,"hann",{periodic:!1}),this.mean=this.config.mean,this.std=this.config.std}async _extract_fbank_features(e,t){return(0,l.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,max_num_frames:t,transpose:!0})}async _call(e){g(e,"ASTFeatureExtractor");const t=await this._extract_fbank_features(e,this.config.max_length);if(this.config.do_normalize){const e=2*this.std,n=t.data;for(let t=0;t<n.length;++t)n[t]=(n[t]-this.mean)/e}return{input_values:t.unsqueeze_(0)}}}class se extends y{constructor(e){super(e),this.mel_filters=(0,l.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,null,"htk"),this.mel_filters_slaney=(0,l.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,"slaney","slaney"),this.window=(0,l.window_function)(this.config.fft_window_size,"hann")}async _get_input_mel(e,t,n,r){let i,a=!1;const s=e.length-t;if(s>0){if("rand_trunc"!==n)throw new Error(`Truncation strategy "${n}" not implemented`);{a=!0;const n=Math.floor(Math.random()*(s+1));e=e.subarray(n,n+t),i=await this._extract_fbank_features(e,this.mel_filters_slaney,this.config.nb_max_samples)}}else{if(s<0){let n=new Float64Array(t);if(n.set(e),"repeat"===r)for(let r=e.length;r<t;r+=e.length)n.set(e.subarray(0,Math.min(e.length,t-r)),r);else if("repeatpad"===r)for(let t=e.length;t<-s;t+=e.length)n.set(e,t);e=n}if("fusion"===n)throw new Error(`Truncation strategy "${n}" not implemented`);i=await this._extract_fbank_features(e,this.mel_filters_slaney,this.config.nb_max_samples)}return i.unsqueeze_(0)}async _extract_fbank_features(e,t,n=null){return(0,l.spectrogram)(e,this.window,this.config.fft_window_size,this.config.hop_length,{power:2,mel_filters:t,log_mel:"dB",max_num_frames:n,do_pad:!1,transpose:!0})}async _call(e,{max_length:t=null}={}){g(e,"ClapFeatureExtractor");return{input_features:(await this._get_input_mel(e,t??this.config.nb_max_samples,this.config.truncation,this.config.padding)).unsqueeze_(0)}}}class oe extends y{async _call(e){g(e,"PyAnnoteFeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));const t=[1,1,e.length];return{input_values:new o.Tensor("float32",e,t)}}samples_to_frames(e){return(e-this.config.offset)/this.config.step}post_process_speaker_diarization(e,t){const n=t/this.samples_to_frames(t)/this.config.sampling_rate,r=[];for(const t of e.tolist()){const e=[];let i=-1;for(let n=0;n<t.length;++n){const r=(0,s.softmax)(t[n]),[a,o]=(0,s.max)(r),[l,u]=[n,n+1];o!==i?(i=o,e.push({id:o,start:l,end:u,score:a})):(e.at(-1).end=u,e.at(-1).score+=a)}r.push(e.map((({id:e,start:t,end:r,score:i})=>({id:e,start:t*n,end:r*n,confidence:i/(r-t)}))))}return r}}class le extends y{constructor(e){super(e);const t=this.config.sampling_rate,n=(0,l.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<n.length;++e)n[e].push(0);this.mel_filters=n,this.window=(0,l.window_function)(400,"hamming",{periodic:!1}),this.min_num_frames=this.config.min_num_frames}async _extract_fbank_features(e){return e=e.map((e=>32768*e)),(0,l.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,transpose:!0,min_num_frames:this.min_num_frames})}async _call(e){g(e,"WeSpeakerFeatureExtractor");const t=(await this._extract_fbank_features(e)).unsqueeze_(0);if(null===this.config.fbank_centering_span){const e=t.mean(1).data,n=t.data,[r,i,a]=t.dims;for(let t=0;t<r;++t){const r=t*i*a,s=t*a;for(let t=0;t<i;++t){const i=r+t*a;for(let t=0;t<a;++t)n[i+t]-=e[s+t]}}}return{input_features:t}}}class ue extends y{}class de extends r.Callable{constructor(e){super(),this.feature_extractor=e}async _call(e,...t){return await this.feature_extractor(e,...t)}}class ce extends de{async _call(...e){return await this.feature_extractor(...e)}post_process_masks(...e){return this.feature_extractor.post_process_masks(...e)}reshape_input_points(...e){return this.feature_extractor.reshape_input_points(...e)}}class pe extends de{async _call(e){return await this.feature_extractor(e)}}class he extends de{async _call(e){return await this.feature_extractor(e)}}class me extends de{async _call(e){return await this.feature_extractor(e)}post_process_speaker_diarization(...e){return this.feature_extractor.post_process_speaker_diarization(...e)}}class fe extends de{async _call(e){return await this.feature_extractor(e)}}class ge extends de{}class _e extends de{constructor(e){super(e);const{tasks_answer_post_processing_type:t,task_prompts_without_inputs:n,task_prompts_with_input:r}=e.config;this.tasks_answer_post_processing_type=new Map(Object.entries(t??{})),this.task_prompts_without_inputs=new Map(Object.entries(n??{})),this.task_prompts_with_input=new Map(Object.entries(r??{})),this.regexes={quad_boxes:/(.+?)<loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)>/gm,bboxes:/([^<]+)?<loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)>/gm},this.size_per_bin=1e3}construct_prompts(e){"string"==typeof e&&(e=[e]);const t=[];for(const n of e)if(this.task_prompts_without_inputs.has(n))t.push(this.task_prompts_without_inputs.get(n));else{for(const[e,r]of this.task_prompts_with_input)if(n.includes(e)){t.push(r.replaceAll("{input}",n).replaceAll(e,""));break}t.length!==e.length&&t.push(n)}return t}post_process_generation(e,t,n){const r=this.tasks_answer_post_processing_type.get(t)??"pure_text";let i;switch(e=e.replaceAll("<s>","").replaceAll("</s>",""),r){case"pure_text":i=e;break;case"description_with_bboxes":case"bboxes":case"phrase_grounding":case"ocr":const a="ocr"===r?"quad_boxes":"bboxes",s=e.matchAll(this.regexes[a]),o=[],l=[];for(const[e,t,...r]of s)o.push(t?t.trim():o.at(-1)??""),l.push(r.map(((e,t)=>(Number(e)+.5)/this.size_per_bin*n[t%2])));i={labels:o,[a]:l};break;default:throw new Error(`Task "${t}" (of type "${r}") not yet implemented.`)}return{[t]:i}}}class we{static FEATURE_EXTRACTOR_CLASS_MAPPING={ImageFeatureExtractor:b,WhisperFeatureExtractor:ne,ViTFeatureExtractor:z,MobileViTFeatureExtractor:V,MobileViTImageProcessor:j,MobileNetV1FeatureExtractor:L,MobileNetV2FeatureExtractor:D,MobileNetV3FeatureExtractor:R,MobileNetV4FeatureExtractor:N,OwlViTFeatureExtractor:G,Owlv2ImageProcessor:q,CLIPFeatureExtractor:C,CLIPImageProcessor:P,Florence2Processor:_e,ChineseCLIPFeatureExtractor:E,SiglipImageProcessor:F,ConvNextFeatureExtractor:A,ConvNextImageProcessor:I,SegformerFeatureExtractor:x,SapiensFeatureExtractor:v,BitImageProcessor:$,DPTImageProcessor:k,DPTFeatureExtractor:T,PvtImageProcessor:M,GLPNFeatureExtractor:S,BeitFeatureExtractor:H,DeiTFeatureExtractor:W,DetrFeatureExtractor:Q,RTDetrImageProcessor:U,MaskFormerFeatureExtractor:Y,YolosFeatureExtractor:J,DonutFeatureExtractor:X,NougatImageProcessor:K,EfficientNetImageProcessor:B,ViTImageProcessor:O,VitMatteImageProcessor:te,SamImageProcessor:Z,Swin2SRImageProcessor:ee,Wav2Vec2FeatureExtractor:re,SeamlessM4TFeatureExtractor:ie,SpeechT5FeatureExtractor:ue,ASTFeatureExtractor:ae,ClapFeatureExtractor:se,PyAnnoteFeatureExtractor:oe,WeSpeakerFeatureExtractor:le};static PROCESSOR_CLASS_MAPPING={WhisperProcessor:pe,Wav2Vec2ProcessorWithLM:he,PyAnnoteProcessor:me,SamProcessor:ce,SpeechT5Processor:fe,OwlViTProcessor:ge,Florence2Processor:_e};static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:r=null,local_files_only:i=!1,revision:s="main"}={}){let o=n??await(0,a.getModelJSON)(e,"preprocessor_config.json",!0,{progress_callback:t,config:n,cache_dir:r,local_files_only:i,revision:s}),l=o.feature_extractor_type??o.image_processor_type,u=this.FEATURE_EXTRACTOR_CLASS_MAPPING[l];if(!u){if(void 0===o.size)throw new Error(`Unknown Feature Extractor type: ${l}`);console.warn(`Feature extractor type "${l}" not found, assuming ImageFeatureExtractor due to size parameter in config.`),u=b}return new(this.PROCESSOR_CLASS_MAPPING[o.processor_class]??de)(new u(o))}}},"./src/tokenizers.js":
 /*!***************************!*\
   !*** ./src/tokenizers.js ***!
-  \***************************/(e,t,n)=>{n.r(t),n.d(t,{AlbertTokenizer:()=>xe,AutoTokenizer:()=>ht,BartTokenizer:()=>Be,BertTokenizer:()=>ve,BlenderbotSmallTokenizer:()=>lt,BlenderbotTokenizer:()=>ot,BloomTokenizer:()=>Ne,CLIPTokenizer:()=>rt,CamembertTokenizer:()=>Fe,CodeGenTokenizer:()=>nt,CodeLlamaTokenizer:()=>Ge,CohereTokenizer:()=>pt,ConvBertTokenizer:()=>Ce,DebertaTokenizer:()=>ke,DebertaV2Tokenizer:()=>$e,DistilBertTokenizer:()=>Ee,ElectraTokenizer:()=>Ie,EsmTokenizer:()=>Xe,FalconTokenizer:()=>We,GPT2Tokenizer:()=>Oe,GPTNeoXTokenizer:()=>He,GemmaTokenizer:()=>Qe,Grok1Tokenizer:()=>Ye,HerbertTokenizer:()=>Se,LlamaTokenizer:()=>je,M2M100Tokenizer:()=>et,MBart50Tokenizer:()=>De,MBartTokenizer:()=>Le,MPNetTokenizer:()=>Ue,MarianTokenizer:()=>at,MobileBertTokenizer:()=>Me,NllbTokenizer:()=>Ze,NougatTokenizer:()=>dt,PreTrainedTokenizer:()=>be,Qwen2Tokenizer:()=>Ke,RoFormerTokenizer:()=>Pe,RobertaTokenizer:()=>Re,SiglipTokenizer:()=>it,SpeechT5Tokenizer:()=>ut,SqueezeBertTokenizer:()=>Te,T5Tokenizer:()=>ze,TokenizerModel:()=>M,VitsTokenizer:()=>ct,Wav2Vec2CTCTokenizer:()=>st,WhisperTokenizer:()=>tt,XLMRobertaTokenizer:()=>qe,XLMTokenizer:()=>Ae,is_chinese_char:()=>w});var r=n(/*! ./utils/generic.js */"./src/utils/generic.js"),i=n(/*! ./utils/core.js */"./src/utils/core.js"),a=n(/*! ./utils/hub.js */"./src/utils/hub.js"),s=n(/*! ./utils/maths.js */"./src/utils/maths.js"),o=n(/*! ./utils/tensor.js */"./src/utils/tensor.js"),l=n(/*! ./utils/data-structures.js */"./src/utils/data-structures.js"),u=n(/*! @huggingface/jinja */"./node_modules/@huggingface/jinja/dist/index.js"),d=n(/*! ./models/whisper/common_whisper.js */"./src/models/whisper/common_whisper.js"),c=n(/*! ./utils/constants.js */"./src/utils/constants.js");async function p(e,t){const n=await Promise.all([(0,a.getModelJSON)(e,"tokenizer.json",!0,t),(0,a.getModelJSON)(e,"tokenizer_config.json",!0,t)]);return null!==t.legacy&&(n[1].legacy=t.legacy),n}function h(e,t=!0){if(void 0!==e.Regex){let t=e.Regex.replace(/\\([#&~])/g,"$1");for(const[e,n]of v)t=t.replaceAll(e,n);return new RegExp(t,"gu")}if(void 0!==e.String){const n=(0,i.escapeRegExp)(e.String);return new RegExp(t?n:`(${n})`,"gu")}return console.warn("Unknown pattern type:",e),null}function m(e){return new Map(Object.entries(e))}function f(e){const t=e.dims;switch(t.length){case 1:return e.tolist();case 2:if(1!==t[0])throw new Error("Unable to decode tensor with `batch size !== 1`. Use `tokenizer.batch_decode(...)` for batched inputs.");return e.tolist()[0];default:throw new Error(`Expected tensor to have 1-2 dimensions, got ${t.length}.`)}}function g(e){return e.replace(/ \./g,".").replace(/ \?/g,"?").replace(/ \!/g,"!").replace(/ ,/g,",").replace(/ \' /g,"'").replace(/ n\'t/g,"n't").replace(/ \'m/g,"'m").replace(/ \'s/g,"'s").replace(/ \'ve/g,"'ve").replace(/ \'re/g,"'re")}function _(e){return e.replace(/[\u0300-\u036f]/g,"")}function w(e){return e>=19968&&e<=40959||e>=13312&&e<=19903||e>=131072&&e<=173791||e>=173824&&e<=177983||e>=177984&&e<=178207||e>=178208&&e<=183983||e>=63744&&e<=64255||e>=194560&&e<=195103}const y="\\p{P}\\u0021-\\u002F\\u003A-\\u0040\\u005B-\\u0060\\u007B-\\u007E",b=new RegExp(`^[${y}]+$`,"gu"),v=new Map([["(?i:'s|'t|'re|'ve|'m|'ll|'d)","(?:'([sS]|[tT]|[rR][eE]|[vV][eE]|[mM]|[lL][lL]|[dD]))"]]);class x{constructor(e){this.content=e.content,this.id=e.id,this.single_word=e.single_word??!1,this.lstrip=e.lstrip??!1,this.rstrip=e.rstrip??!1,this.special=e.special??!1,this.normalized=e.normalized??null}}class M extends r.Callable{constructor(e){super(),this.config=e,this.vocab=[],this.tokens_to_ids=new Map,this.unk_token_id=void 0,this.unk_token=void 0,this.end_of_word_suffix=void 0,this.fuse_unk=this.config.fuse_unk??!1}static fromConfig(e,...t){switch(e.type){case"WordPiece":return new T(e);case"Unigram":return new k(e,...t);case"BPE":return new C(e);default:if(e.vocab)return new P(e,...t);throw new Error(`Unknown TokenizerModel type: ${e.type}`)}}_call(e){return e=this.encode(e),this.fuse_unk&&(e=function(e,t,n){const r=[];let i=0;for(;i<e.length;)if(r.push(e[i]),(t.get(e[i])??n)===n)for(;++i<e.length&&(t.get(e[i])??n)===n;)t.get(r.at(-1))!==n&&(r[r.length-1]+=e[i]);else++i;return r}(e,this.tokens_to_ids,this.unk_token_id)),e}encode(e){throw Error("encode should be implemented in subclass.")}convert_tokens_to_ids(e){return e.map((e=>this.tokens_to_ids.get(e)??this.unk_token_id))}convert_ids_to_tokens(e){return e.map((e=>this.vocab[e]??this.unk_token))}}class T extends M{constructor(e){super(e),this.tokens_to_ids=m(e.vocab),this.unk_token_id=this.tokens_to_ids.get(e.unk_token),this.unk_token=e.unk_token,this.max_input_chars_per_word=e.max_input_chars_per_word??100,this.vocab=new Array(this.tokens_to_ids.size);for(const[e,t]of this.tokens_to_ids)this.vocab[t]=e}encode(e){const t=[];for(const n of e){const e=[...n];if(e.length>this.max_input_chars_per_word){t.push(this.unk_token);continue}let r=!1,i=0;const a=[];for(;i<e.length;){let t=e.length,n=null;for(;i<t;){let r=e.slice(i,t).join("");if(i>0&&(r=this.config.continuing_subword_prefix+r),this.tokens_to_ids.has(r)){n=r;break}--t}if(null===n){r=!0;break}a.push(n),i=t}r?t.push(this.unk_token):t.push(...a)}return t}}class k extends M{constructor(e,t){super(e);const n=e.vocab.length;this.vocab=new Array(n),this.scores=new Array(n);for(let t=0;t<n;++t){const n=e.vocab[t];this.vocab[t]=n[0],this.scores[t]=n[1]}this.unk_token_id=e.unk_id,this.unk_token=this.vocab[e.unk_id],this.tokens_to_ids=new Map(this.vocab.map(((e,t)=>[e,t]))),this.bosToken=" ",this.bosTokenId=this.tokens_to_ids.get(this.bosToken),this.eosToken=t.eos_token,this.eosTokenId=this.tokens_to_ids.get(this.eosToken),this.unkToken=this.vocab[this.unk_token_id],this.minScore=(0,s.min)(this.scores)[0],this.unkScore=this.minScore-10,this.scores[this.unk_token_id]=this.unkScore,this.trie=new l.CharTrie,this.trie.extend(this.vocab),this.fuse_unk=!0}populateNodes(e){const t=e.sentence,n=t.length;let r=0;for(;r<n;){const n=1;let i=!1;const a=[];for(let s of this.trie.commonPrefixSearch(t.slice(r))){a.push(s);const t=this.tokens_to_ids.get(s),o=this.scores[t],l=s.length;e.insert(r,l,o,t),i||l!==n||(i=!0)}i||e.insert(r,n,this.unkScore,this.unk_token_id),r+=n}}tokenize(e){const t=new l.TokenLattice(e,this.bosTokenId,this.eosTokenId);return this.populateNodes(t),t.tokens()}encode(e){const t=[];for(const n of e){const e=this.tokenize(n);t.push(...e)}return t}}const $=(()=>{const e=[...Array.from({length:"~".charCodeAt(0)-"!".charCodeAt(0)+1},((e,t)=>t+"!".charCodeAt(0))),...Array.from({length:"¬".charCodeAt(0)-"¡".charCodeAt(0)+1},((e,t)=>t+"¡".charCodeAt(0))),...Array.from({length:"ÿ".charCodeAt(0)-"®".charCodeAt(0)+1},((e,t)=>t+"®".charCodeAt(0)))],t=e.slice();let n=0;for(let r=0;r<256;++r)e.includes(r)||(e.push(r),t.push(256+n),n+=1);const r=t.map((e=>String.fromCharCode(e)));return Object.fromEntries(e.map(((e,t)=>[e,r[t]])))})(),S=(0,i.reverseDictionary)($);class C extends M{constructor(e){super(e),this.BPE_SPLIT_TOKEN=" ",this.tokens_to_ids=m(e.vocab),this.unk_token_id=this.tokens_to_ids.get(e.unk_token),this.unk_token=e.unk_token,this.vocab=new Array(this.tokens_to_ids.size);for(const[e,t]of this.tokens_to_ids)this.vocab[t]=e;this.bpe_ranks=new Map(e.merges.map(((e,t)=>[e,t]))),this.merges=e.merges.map((e=>e.split(this.BPE_SPLIT_TOKEN))),this.end_of_word_suffix=e.end_of_word_suffix,this.continuing_subword_suffix=e.continuing_subword_suffix??null,this.byte_fallback=this.config.byte_fallback??!1,this.byte_fallback&&(this.text_encoder=new TextEncoder),this.ignore_merges=this.config.ignore_merges??!1,this.cache=new Map}bpe(e){if(0===e.length)return[];const t=this.cache.get(e);if(void 0!==t)return t;const n=Array.from(e);this.end_of_word_suffix&&(n[n.length-1]+=this.end_of_word_suffix);let r=[];if(n.length>1){const e=new l.PriorityQueue(((e,t)=>e.score<t.score));let t={token:n[0],bias:0,prev:null,next:null},i=t;for(let t=1;t<n.length;++t){const r={bias:t/n.length,token:n[t],prev:i,next:null};i.next=r,this._add_node(e,i),i=r}for(;!e.isEmpty();){const n=e.pop();if(n.deleted||!n.next||n.next.deleted)continue;if(n.deleted=!0,n.next.deleted=!0,n.prev){const e={...n.prev};n.prev.deleted=!0,n.prev=e,e.prev?e.prev.next=e:t=e}const r={token:n.token+n.next.token,bias:n.bias,prev:n.prev,next:n.next.next};r.prev?(r.prev.next=r,this._add_node(e,r.prev)):t=r,r.next&&(r.next.prev=r,this._add_node(e,r))}for(let e=t;null!==e;e=e.next)r.push(e.token)}else r=n;if(this.continuing_subword_suffix)for(let e=0;e<r.length-1;++e)r[e]+=this.continuing_subword_suffix;return this.cache.set(e,r),r}_add_node(e,t){const n=this.bpe_ranks.get(t.token+this.BPE_SPLIT_TOKEN+t.next.token);void 0!==n&&(t.score=n+t.bias,e.push(t))}encode(e){const t=[];for(const n of e){if(this.ignore_merges&&this.tokens_to_ids.has(n)){t.push(n);continue}const e=this.bpe(n);for(const n of e)if(this.tokens_to_ids.has(n))t.push(n);else if(this.byte_fallback){const e=Array.from(this.text_encoder.encode(n)).map((e=>`<0x${e.toString(16).toUpperCase().padStart(2,"0")}>`));e.every((e=>this.tokens_to_ids.has(e)))?t.push(...e):t.push(this.unk_token)}else t.push(this.unk_token)}return t}}class P extends M{constructor(e,t){super(e),this.tokens_to_ids=m(t.target_lang?e.vocab[t.target_lang]:e.vocab),this.bos_token=t.bos_token,this.bos_token_id=this.tokens_to_ids.get(this.bos_token),this.eos_token=t.eos_token,this.eos_token_id=this.tokens_to_ids.get(this.eos_token),this.pad_token=t.pad_token,this.pad_token_id=this.tokens_to_ids.get(this.pad_token),this.unk_token=t.unk_token,this.unk_token_id=this.tokens_to_ids.get(this.unk_token),this.vocab=new Array(this.tokens_to_ids.size);for(const[e,t]of this.tokens_to_ids)this.vocab[t]=e}encode(e){return e}}class E extends r.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"BertNormalizer":return new N(e);case"Precompiled":return new pe(e);case"Sequence":return new R(e);case"Replace":return new F(e);case"NFC":return new A(e);case"NFKC":return new I(e);case"NFKD":return new z(e);case"Strip":return new O(e);case"StripAccents":return new B(e);case"Lowercase":return new L(e);case"Prepend":return new D(e);default:throw new Error(`Unknown Normalizer type: ${e.type}`)}}normalize(e){throw Error("normalize should be implemented in subclass.")}_call(e){return this.normalize(e)}}class F extends E{normalize(e){const t=h(this.config.pattern);return null===t?e:e.replaceAll(t,this.config.content)}}class A extends E{normalize(e){return e=e.normalize("NFC")}}class I extends E{normalize(e){return e=e.normalize("NFKC")}}class z extends E{normalize(e){return e=e.normalize("NFKD")}}class O extends E{normalize(e){return this.config.strip_left&&this.config.strip_right?e=e.trim():(this.config.strip_left&&(e=e.trimStart()),this.config.strip_right&&(e=e.trimEnd())),e}}class B extends E{normalize(e){return e=_(e)}}class L extends E{normalize(e){return e=e.toLowerCase()}}class D extends E{normalize(e){return e=this.config.prepend+e}}class R extends E{constructor(e){super(e),this.normalizers=e.normalizers.map((e=>E.fromConfig(e)))}normalize(e){return this.normalizers.reduce(((e,t)=>t.normalize(e)),e)}}class N extends E{_tokenize_chinese_chars(e){const t=[];for(let n=0;n<e.length;++n){const r=e[n];w(r.charCodeAt(0))?(t.push(" "),t.push(r),t.push(" ")):t.push(r)}return t.join("")}stripAccents(e){return e.normalize("NFD").replace(/[\u0300-\u036f]/g,"")}_is_control(e){switch(e){case"\t":case"\n":case"\r":return!1;default:return/^\p{Cc}|\p{Cf}|\p{Co}|\p{Cs}$/u.test(e)}}_clean_text(e){const t=[];for(const n of e){const e=n.charCodeAt(0);0===e||65533===e||this._is_control(n)||(/^\s$/.test(n)?t.push(" "):t.push(n))}return t.join("")}normalize(e){return this.config.clean_text&&(e=this._clean_text(e)),this.config.handle_chinese_chars&&(e=this._tokenize_chinese_chars(e)),this.config.lowercase?(e=e.toLowerCase(),!1!==this.config.strip_accents&&(e=this.stripAccents(e))):this.config.strip_accents&&(e=this.stripAccents(e)),e}}class V extends r.Callable{static fromConfig(e){if(null===e)return null;switch(e.type){case"BertPreTokenizer":return new j(e);case"Sequence":return new he(e);case"Whitespace":return new me(e);case"WhitespaceSplit":return new fe(e);case"Metaspace":return new de(e);case"ByteLevel":return new G(e);case"Split":return new q(e);case"Punctuation":return new U(e);case"Digits":return new W(e);case"Replace":return new ge(e);default:throw new Error(`Unknown PreTokenizer type: ${e.type}`)}}pre_tokenize_text(e,t){throw Error("pre_tokenize_text should be implemented in subclass.")}pre_tokenize(e,t){return(Array.isArray(e)?e.map((e=>this.pre_tokenize_text(e,t))):this.pre_tokenize_text(e,t)).flat()}_call(e,t){return this.pre_tokenize(e,t)}}class j extends V{constructor(e){super(),this.pattern=new RegExp(`[^\\s${y}]+|[${y}]`,"gu")}pre_tokenize_text(e,t){return e.trim().match(this.pattern)||[]}}class G extends V{constructor(e){super(),this.config=e,this.add_prefix_space=this.config.add_prefix_space,this.trim_offsets=this.config.trim_offsets,this.use_regex=this.config.use_regex??!0,this.pattern=/'s|'t|'re|'ve|'m|'ll|'d| ?\p{L}+| ?\p{N}+| ?[^\s\p{L}\p{N}]+|\s+(?!\S)|\s+/gu,this.byte_encoder=$,this.text_encoder=new TextEncoder}pre_tokenize_text(e,t){this.add_prefix_space&&!e.startsWith(" ")&&(e=" "+e);return(this.use_regex?e.match(this.pattern)||[]:[e]).map((e=>Array.from(this.text_encoder.encode(e),(e=>this.byte_encoder[e])).join("")))}}class q extends V{constructor(e){super(),this.config=e,this.pattern=h(this.config.pattern,this.config.invert)}pre_tokenize_text(e,t){return null===this.pattern?[]:this.config.invert?e.match(this.pattern)||[]:function(e,t){const n=[];let r=0;for(const i of e.matchAll(t)){const t=i[0];r<i.index&&n.push(e.slice(r,i.index)),t.length>0&&n.push(t),r=i.index+t.length}return r<e.length&&n.push(e.slice(r)),n}(e,this.pattern)}}class U extends V{constructor(e){super(),this.config=e,this.pattern=new RegExp(`[^${y}]+|[${y}]+`,"gu")}pre_tokenize_text(e,t){return e.match(this.pattern)||[]}}class W extends V{constructor(e){super(),this.config=e;const t="[^\\d]+|\\d"+(this.config.individual_digits?"":"+");this.pattern=new RegExp(t,"gu")}pre_tokenize_text(e,t){return e.match(this.pattern)||[]}}class H extends r.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"TemplateProcessing":return new Q(e);case"ByteLevel":return new Y(e);case"RobertaProcessing":return new K(e);case"BertProcessing":return new X(e);case"Sequence":return new J(e);default:throw new Error(`Unknown PostProcessor type: ${e.type}`)}}post_process(e,...t){throw Error("post_process should be implemented in subclass.")}_call(e,...t){return this.post_process(e,...t)}}class X extends H{constructor(e){super(e),this.cls=e.cls[0],this.sep=e.sep[0]}post_process(e,t=null,{add_special_tokens:n=!0}={}){n&&(e=(0,i.mergeArrays)([this.cls],e,[this.sep]));let r=new Array(e.length).fill(0);if(null!==t){const a=n&&this instanceof K?[this.sep]:[],s=n?[this.sep]:[];e=(0,i.mergeArrays)(e,a,t,s),r=(0,i.mergeArrays)(r,new Array(t.length+a.length+s.length).fill(1))}return{tokens:e,token_type_ids:r}}}class K extends X{}class Q extends H{constructor(e){super(e),this.single=e.single,this.pair=e.pair}post_process(e,t=null,{add_special_tokens:n=!0}={}){const r=null===t?this.single:this.pair;let a=[],s=[];for(const o of r)"SpecialToken"in o?n&&(a.push(o.SpecialToken.id),s.push(o.SpecialToken.type_id)):"Sequence"in o&&("A"===o.Sequence.id?(a=(0,i.mergeArrays)(a,e),s=(0,i.mergeArrays)(s,new Array(e.length).fill(o.Sequence.type_id))):"B"===o.Sequence.id&&(a=(0,i.mergeArrays)(a,t),s=(0,i.mergeArrays)(s,new Array(t.length).fill(o.Sequence.type_id))));return{tokens:a,token_type_ids:s}}}class Y extends H{post_process(e,t=null){return t&&(e=(0,i.mergeArrays)(e,t)),{tokens:e}}}class J extends H{constructor(e){super(e),this.processors=e.processors.map((e=>H.fromConfig(e)))}post_process(e,t=null,n={}){let r;for(const i of this.processors)if(i instanceof Y){if(e=i.post_process(e).tokens,t){t=i.post_process(t).tokens}}else{const a=i.post_process(e,t,n);e=a.tokens,r=a.token_type_ids}return{tokens:e,token_type_ids:r}}}class Z extends r.Callable{constructor(e){super(),this.config=e,this.added_tokens=[],this.end_of_word_suffix=null,this.trim_offsets=e.trim_offsets}static fromConfig(e){if(null===e)return null;switch(e.type){case"WordPiece":return new ie(e);case"Metaspace":return new ce(e);case"ByteLevel":return new ae(e);case"Replace":return new ee(e);case"ByteFallback":return new te(e);case"Fuse":return new ne(e);case"Strip":return new re(e);case"Sequence":return new oe(e);case"CTC":return new se(e);case"BPEDecoder":return new le(e);default:throw new Error(`Unknown Decoder type: ${e.type}`)}}_call(e){return this.decode(e)}decode(e){return this.decode_chain(e).join("")}decode_chain(e){throw Error("`decode_chain` should be implemented in subclass.")}}class ee extends Z{decode_chain(e){const t=h(this.config.pattern);return null===t?e:e.map((e=>e.replaceAll(t,this.config.content)))}}class te extends Z{constructor(e){super(e),this.text_decoder=new TextDecoder}decode_chain(e){const t=[];let n=[];for(const r of e){let e=null;if(6===r.length&&r.startsWith("<0x")&&r.endsWith(">")){const t=parseInt(r.slice(3,5),16);isNaN(t)||(e=t)}if(null!==e)n.push(e);else{if(n.length>0){const e=this.text_decoder.decode(Uint8Array.from(n));t.push(e),n=[]}t.push(r)}}if(n.length>0){const e=this.text_decoder.decode(Uint8Array.from(n));t.push(e),n=[]}return t}}class ne extends Z{decode_chain(e){return[e.join("")]}}class re extends Z{constructor(e){super(e),this.content=this.config.content,this.start=this.config.start,this.stop=this.config.stop}decode_chain(e){return e.map((e=>{let t=0;for(let n=0;n<this.start&&e[n]===this.content;++n)t=n+1;let n=e.length;for(let t=0;t<this.stop;++t){const r=e.length-t-1;if(e[r]!==this.content)break;n=r}return e.slice(t,n)}))}}class ie extends Z{constructor(e){super(e),this.cleanup=e.cleanup}decode_chain(e){return e.map(((e,t)=>(0!==t&&(e=e.startsWith(this.config.prefix)?e.replace(this.config.prefix,""):" "+e),this.cleanup&&(e=g(e)),e)))}}class ae extends Z{constructor(e){super(e),this.byte_decoder=S,this.text_decoder=new TextDecoder("utf-8",{fatal:!1,ignoreBOM:!0}),this.end_of_word_suffix=null}convert_tokens_to_string(e){const t=e.join(""),n=new Uint8Array([...t].map((e=>this.byte_decoder[e])));return this.text_decoder.decode(n)}decode_chain(e){const t=[];let n=[];for(const r of e)void 0!==this.added_tokens.find((e=>e.content===r))?(n.length>0&&(t.push(this.convert_tokens_to_string(n)),n=[]),t.push(r)):n.push(r);return n.length>0&&t.push(this.convert_tokens_to_string(n)),t}}class se extends Z{constructor(e){super(e),this.pad_token=this.config.pad_token,this.word_delimiter_token=this.config.word_delimiter_token,this.cleanup=this.config.cleanup}convert_tokens_to_string(e){if(0===e.length)return"";const t=[e[0]];for(let n=1;n<e.length;++n)e[n]!==t.at(-1)&&t.push(e[n]);let n=t.filter((e=>e!==this.pad_token)).join("");return this.cleanup&&(n=g(n).replaceAll(this.word_delimiter_token," ").trim()),n}decode_chain(e){return[this.convert_tokens_to_string(e)]}}class oe extends Z{constructor(e){super(e),this.decoders=e.decoders.map((e=>Z.fromConfig(e)))}decode_chain(e){return this.decoders.reduce(((e,t)=>t.decode_chain(e)),e)}}class le extends Z{constructor(e){super(e),this.suffix=this.config.suffix}decode_chain(e){return e.map(((t,n)=>t.replaceAll(this.suffix,n===e.length-1?"":" ")))}}class ue extends Z{decode_chain(e){let t="";for(let n=1;n<e.length;n+=2)t+=e[n];return[t]}}class de extends V{constructor(e){super(),this.addPrefixSpace=e.add_prefix_space,this.replacement=e.replacement,this.strRep=e.str_rep||this.replacement,this.prepend_scheme=e.prepend_scheme??"always"}pre_tokenize_text(e,{section_index:t}={}){let n=e.replaceAll(" ",this.strRep);return this.addPrefixSpace&&!n.startsWith(this.replacement)&&("always"===this.prepend_scheme||"first"===this.prepend_scheme&&0===t)&&(n=this.strRep+n),[n]}}class ce extends Z{constructor(e){super(e),this.addPrefixSpace=e.add_prefix_space,this.replacement=e.replacement}decode_chain(e){const t=[];for(let n=0;n<e.length;++n){let r=e[n].replaceAll(this.replacement," ");this.addPrefixSpace&&0==n&&r.startsWith(" ")&&(r=r.substring(1)),t.push(r)}return t}}class pe extends E{constructor(e){super(e),this.charsmap=e.precompiled_charsmap}normalize(e){if((e=(e=e.replace(/[\u0001-\u0008\u000B\u000E-\u001F\u007F\u008F\u009F]/gm,"")).replace(/[\u0009\u000A\u000C\u000D\u1680\u200B\u200C\u200E\u200F\u2028\u2029\u2581\uFEFF\uFFFD]/gm," ")).includes("～")){const t=e.split("～");e=t.map((e=>e.normalize("NFKC"))).join("～")}else e=e.normalize("NFKC");return e}}class he extends V{constructor(e){super(),this.tokenizers=e.pretokenizers.map((e=>V.fromConfig(e)))}pre_tokenize_text(e,t){return this.tokenizers.reduce(((e,n)=>n.pre_tokenize(e,t)),[e])}}class me extends V{constructor(e){super()}pre_tokenize_text(e,t){return e.match(/\w+|[^\w\s]+/g)||[]}}class fe extends V{constructor(e){super()}pre_tokenize_text(e,t){return function(e){return e.match(/\S+/g)||[]}(e)}}class ge extends V{constructor(e){super(),this.config=e,this.pattern=h(this.config.pattern),this.content=this.config.content}pre_tokenize_text(e,t){return null===this.pattern?[e]:[e.replaceAll(this.pattern,this.config.content)]}}const _e=["bos_token","eos_token","unk_token","sep_token","pad_token","cls_token","mask_token"];function we(e,t,n,r){for(const a of Object.keys(e)){const s=t-e[a].length,o=n(a),l=new Array(s).fill(o);e[a]="right"===r?(0,i.mergeArrays)(e[a],l):(0,i.mergeArrays)(l,e[a])}}function ye(e,t){for(const n of Object.keys(e))e[n].length=t}class be extends r.Callable{return_token_type_ids=!1;padding_side="right";constructor(e,t){super(),this._tokenizer_config=t,this.normalizer=E.fromConfig(e.normalizer),this.pre_tokenizer=V.fromConfig(e.pre_tokenizer),this.model=M.fromConfig(e.model,t),this.post_processor=H.fromConfig(e.post_processor),this.decoder=Z.fromConfig(e.decoder),this.special_tokens=[],this.all_special_ids=[],this.added_tokens=[];for(const t of e.added_tokens){const e=new x(t);this.added_tokens.push(e),this.model.tokens_to_ids.set(e.content,e.id),this.model.vocab[e.id]=e.content,e.special&&(this.special_tokens.push(e.content),this.all_special_ids.push(e.id))}if(this.additional_special_tokens=t.additional_special_tokens??[],this.special_tokens.push(...this.additional_special_tokens),this.special_tokens=[...new Set(this.special_tokens)],this.decoder&&(this.decoder.added_tokens=this.added_tokens,this.decoder.end_of_word_suffix=this.model.end_of_word_suffix),this.added_tokens_regex=this.added_tokens.length>0?new RegExp(this.added_tokens.slice().sort(((e,t)=>t.content.length-e.content.length)).map((e=>`${e.lstrip?"\\s*":""}(${(0,i.escapeRegExp)(e.content)})${e.rstrip?"\\s*":""}`)).join("|")):null,this.mask_token=this.getToken("mask_token"),this.mask_token_id=this.model.tokens_to_ids.get(this.mask_token),this.pad_token=this.getToken("pad_token","eos_token"),this.pad_token_id=this.model.tokens_to_ids.get(this.pad_token),this.sep_token=this.getToken("sep_token"),this.sep_token_id=this.model.tokens_to_ids.get(this.sep_token),this.unk_token=this.getToken("unk_token"),this.unk_token_id=this.model.tokens_to_ids.get(this.unk_token),this.model_max_length=t.model_max_length,this.remove_space=t.remove_space,this.clean_up_tokenization_spaces=t.clean_up_tokenization_spaces??!0,this.do_lowercase_and_remove_accent=t.do_lowercase_and_remove_accent??!1,t.padding_side&&(this.padding_side=t.padding_side),this.legacy=!1,this.chat_template=t.chat_template??null,Array.isArray(this.chat_template)){const e=Object.create(null);for(const{name:t,template:n}of this.chat_template){if("string"!=typeof t||"string"!=typeof n)throw new Error('Chat template must be a list of objects with "name" and "template" properties');e[t]=n}this.chat_template=e}this._compiled_template_cache=new Map}getToken(...e){for(const t of e){const e=this._tokenizer_config[t];if(e){if("object"==typeof e){if("AddedToken"===e.__type)return e.content;throw Error(`Unknown token: ${e}`)}return e}}return null}static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:r=null,local_files_only:i=!1,revision:a="main",legacy:s=null}={}){return new this(...await p(e,{progress_callback:t,config:n,cache_dir:r,local_files_only:i,revision:a,legacy:s}))}_call(e,{text_pair:t=null,add_special_tokens:n=!0,padding:r=!1,truncation:i=null,max_length:a=null,return_tensor:l=!0,return_token_type_ids:u=null}={}){const d=Array.isArray(e);let c;if(d){if(0===e.length)throw Error("text array must be non-empty");if(null!==t){if(!Array.isArray(t))throw Error("text_pair must also be an array");if(e.length!==t.length)throw Error("text and text_pair must have the same length");c=e.map(((e,r)=>this._encode_plus(e,{text_pair:t[r],add_special_tokens:n,return_token_type_ids:u})))}else c=e.map((e=>this._encode_plus(e,{add_special_tokens:n,return_token_type_ids:u})))}else{if(null==e)throw Error("text may not be null or undefined");if(Array.isArray(t))throw Error("When specifying `text_pair`, since `text` is a string, `text_pair` must also be a string (i.e., not an array).");c=[this._encode_plus(e,{text_pair:t,add_special_tokens:n,return_token_type_ids:u})]}if(null===a?a="max_length"===r?this.model_max_length:(0,s.max)(c.map((e=>e.input_ids.length)))[0]:i||console.warn("Truncation was not explicitly activated but `max_length` is provided a specific value, please use `truncation=true` to explicitly truncate examples to max length."),a=Math.min(a,this.model_max_length??1/0),r||i)for(let e=0;e<c.length;++e)c[e].input_ids.length!==a&&(c[e].input_ids.length>a?i&&ye(c[e],a):r&&we(c[e],a,(e=>"input_ids"===e?this.pad_token_id:0),this.padding_side));const p={};if(l){if((!r||!i)&&c.some((e=>{for(const t of Object.keys(e))if(e[t].length!==c[0][t]?.length)return!0;return!1})))throw Error("Unable to create tensor, you should probably activate truncation and/or padding with 'padding=true' and 'truncation=true' to have batched tensors with the same length.");const e=[c.length,c[0].input_ids.length];for(const t of Object.keys(c[0]))p[t]=new o.Tensor("int64",BigInt64Array.from(c.flatMap((e=>e[t])).map(BigInt)),e)}else{for(const e of Object.keys(c[0]))p[e]=c.map((t=>t[e]));if(!d)for(const e of Object.keys(p))p[e]=p[e][0]}return p}_encode_text(e){if(null===e)return null;const t=(this.added_tokens_regex?e.split(this.added_tokens_regex).filter((e=>e)):[e]).map(((e,t)=>{if(void 0!==this.added_tokens.find((t=>t.content===e)))return e;{if(!0===this.remove_space&&(e=e.trim().split(/\s+/).join(" ")),this.do_lowercase_and_remove_accent&&(e=function(e){return _(e.toLowerCase())}(e)),null!==this.normalizer&&(e=this.normalizer(e)),0===e.length)return[];const n=null!==this.pre_tokenizer?this.pre_tokenizer(e,{section_index:t}):[e];return this.model(n)}})).flat();return t}_encode_plus(e,{text_pair:t=null,add_special_tokens:n=!0,return_token_type_ids:r=null}={}){const{tokens:i,token_type_ids:a}=this._tokenize_helper(e,{pair:t,add_special_tokens:n}),s=this.model.convert_tokens_to_ids(i),o={input_ids:s,attention_mask:new Array(s.length).fill(1)};return(r??this.return_token_type_ids)&&a&&(o.token_type_ids=a),o}_tokenize_helper(e,{pair:t=null,add_special_tokens:n=!1}={}){const r=this._encode_text(e),a=this._encode_text(t);return this.post_processor?this.post_processor(r,a,{add_special_tokens:n}):{tokens:(0,i.mergeArrays)(r??[],a??[])}}tokenize(e,{pair:t=null,add_special_tokens:n=!1}={}){return this._tokenize_helper(e,{pair:t,add_special_tokens:n}).tokens}encode(e,{text_pair:t=null,add_special_tokens:n=!0,return_token_type_ids:r=null}={}){return this._encode_plus(e,{text_pair:t,add_special_tokens:n,return_token_type_ids:r}).input_ids}batch_decode(e,t={}){return e instanceof o.Tensor&&(e=e.tolist()),e.map((e=>this.decode(e,t)))}decode(e,t={}){if(e instanceof o.Tensor&&(e=f(e)),!Array.isArray(e)||0===e.length||!(0,i.isIntegralNumber)(e[0]))throw Error("token_ids must be a non-empty array of integers.");return this.decode_single(e,t)}decode_single(e,{skip_special_tokens:t=!1,clean_up_tokenization_spaces:n=null}){let r=this.model.convert_ids_to_tokens(e);t&&(r=r.filter((e=>!this.special_tokens.includes(e))));let i=this.decoder?this.decoder(r):r.join(" ");return this.decoder&&this.decoder.end_of_word_suffix&&(i=i.replaceAll(this.decoder.end_of_word_suffix," "),t&&(i=i.trim())),(n??this.clean_up_tokenization_spaces)&&(i=g(i)),i}get_chat_template({chat_template:e=null,tools:t=null}={}){if(this.chat_template&&"object"==typeof this.chat_template){const n=this.chat_template;if(null!==e&&Object.hasOwn(n,e))e=n[e];else if(null===e)if(null!==t&&"tool_use"in n)e=n.tool_use;else{if(!("default"in n))throw Error(`This model has multiple chat templates with no default specified! Please either pass a chat template or the name of the template you wish to use to the 'chat_template' argument. Available template names are ${Object.keys(n).sort()}.`);e=n.default}}else if(null===e){if(!this.chat_template)throw Error("Cannot use apply_chat_template() because tokenizer.chat_template is not set and no template argument was passed! For information about writing templates and setting the tokenizer.chat_template attribute, please see the documentation at https://huggingface.co/docs/transformers/main/en/chat_templating");e=this.chat_template}return e}apply_chat_template(e,{tools:t=null,documents:n=null,chat_template:r=null,add_generation_prompt:i=!1,tokenize:a=!0,padding:s=!1,truncation:o=!1,max_length:l=null,return_tensor:d=!0,return_dict:c=!1,tokenizer_kwargs:p={},...h}={}){if("string"!=typeof(r=this.get_chat_template({chat_template:r,tools:t})))throw Error("chat_template must be a string, but got "+typeof r);let m=this._compiled_template_cache.get(r);void 0===m&&(m=new u.Template(r),this._compiled_template_cache.set(r,m));const f=Object.create(null);for(const e of _e){const t=this.getToken(e);t&&(f[e]=t)}const g=m.render({messages:e,add_generation_prompt:i,tools:t,documents:n,...f,...h});if(a){const e=this._call(g,{add_special_tokens:!1,padding:s,truncation:o,max_length:l,return_tensor:d,...p});return c?e:e.input_ids}return g}}class ve extends be{return_token_type_ids=!0}class xe extends be{return_token_type_ids=!0}class Me extends be{return_token_type_ids=!0}class Te extends be{return_token_type_ids=!0}class ke extends be{return_token_type_ids=!0}class $e extends be{return_token_type_ids=!0}class Se extends be{return_token_type_ids=!0}class Ce extends be{return_token_type_ids=!0}class Pe extends be{return_token_type_ids=!0}class Ee extends be{}class Fe extends be{}class Ae extends be{return_token_type_ids=!0;constructor(e,t){super(e,t),console.warn('WARNING: `XLMTokenizer` is not yet supported by Hugging Face\'s "fast" tokenizers library. Therefore, you may experience slightly inaccurate results.')}}class Ie extends be{return_token_type_ids=!0}class ze extends be{}class Oe extends be{}class Be extends be{}class Le extends be{constructor(e,t){super(e,t),this.languageRegex=/^[a-z]{2}_[A-Z]{2}$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))),this.lang_to_token=e=>e}_build_translation_inputs(e,t,n){return Je(this,e,t,n)}}class De extends Le{}class Re extends be{}class Ne extends be{constructor(e,t){const n=".,!?…。，、।۔،",r=e.pre_tokenizer?.pretokenizers[0]?.pattern;r&&r.Regex===` ?[^(\\s|[${n}])]+`&&(r.Regex=` ?[^\\s${n}]+`),super(e,t)}}const Ve="▁";class je extends be{padding_side="left";constructor(e,t){super(e,t),this.legacy=t.legacy??!0,this.legacy||(this.normalizer=null,this.pre_tokenizer=new de({replacement:Ve,add_prefix_space:!0,prepend_scheme:"first"}))}_encode_text(e){if(null===e)return null;if(this.legacy||0===e.length)return super._encode_text(e);let t=super._encode_text(Ve+e.replaceAll(Ve," "));return t.length>1&&t[0]===Ve&&this.special_tokens.includes(t[1])&&(t=t.slice(1)),t}}class Ge extends be{}class qe extends be{}class Ue extends be{}class We extends be{}class He extends be{}class Xe extends be{}class Ke extends be{}class Qe extends be{}class Ye extends be{}function Je(e,t,n,r){if(!("language_codes"in e)||!Array.isArray(e.language_codes))throw new Error("Tokenizer must have `language_codes` attribute set and it should be an array of language ids.");if(!("languageRegex"in e&&e.languageRegex instanceof RegExp))throw new Error("Tokenizer must have `languageRegex` attribute set and it should be a regular expression.");if(!("lang_to_token"in e)||"function"!=typeof e.lang_to_token)throw new Error("Tokenizer must have `lang_to_token` attribute set and it should be a function.");const i=r.src_lang,a=r.tgt_lang;if(!e.language_codes.includes(a))throw new Error(`Target language code "${a}" is not valid. Must be one of: {${e.language_codes.join(", ")}}`);if(void 0!==i){if(!e.language_codes.includes(i))throw new Error(`Source language code "${i}" is not valid. Must be one of: {${e.language_codes.join(", ")}}`);for(const t of e.post_processor.config.single)if("SpecialToken"in t&&e.languageRegex.test(t.SpecialToken.id)){t.SpecialToken.id=e.lang_to_token(i);break}}return r.forced_bos_token_id=e.model.convert_tokens_to_ids([e.lang_to_token(a)])[0],e._call(t,n)}class Ze extends be{constructor(e,t){super(e,t),this.languageRegex=/^[a-z]{3}_[A-Z][a-z]{3}$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))),this.lang_to_token=e=>e}_build_translation_inputs(e,t,n){return Je(this,e,t,n)}}class et extends be{constructor(e,t){super(e,t),this.languageRegex=/^__[a-z]{2,3}__$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))).map((e=>e.slice(2,-2))),this.lang_to_token=e=>`__${e}__`}_build_translation_inputs(e,t,n){return Je(this,e,t,n)}}class tt extends be{get timestamp_begin(){return this.model.convert_tokens_to_ids(["<|notimestamps|>"])[0]+1}_decode_asr(e,{return_timestamps:t=!1,return_language:n=!1,time_precision:r=null,force_full_sequences:i=!0}={}){if(null===r)throw Error("Must specify time_precision");let a=null;const o="word"===t;function l(){return{language:a,timestamp:[null,null],text:""}}const u=[];let c=l(),p=0;const h=this.timestamp_begin;let m=[],f=[],g=!1,_=null;const w=new Set(this.all_special_ids);for(const n of e){const e=n.tokens,i=o?n.token_timestamps:null;let y=null,v=h;if("stride"in n){const[t,i,a]=n.stride;if(p-=i,_=t-a,i&&(v=i/r+h),a)for(let t=e.length-1;t>=0;--t){const n=Number(e[t]);if(n>=h){if(null!==y&&(n-h)*r<_)break;y=n}}}let x=[],M=[];for(let n=0;n<e.length;++n){const _=Number(e[n]);if(w.has(_)){const e=this.decode([_]),n=d.WHISPER_LANGUAGE_MAPPING.get(e.slice(2,-2));if(void 0!==n){if(null!==a&&n!==a&&!t){m.push(x);const e=this.findLongestCommonSequence(m)[0],t=this.decode(e);c.text=t,u.push(c),m=[],x=[],c=l()}a=c.language=n}}else if(_>=h){const e=(_-h)*r+p,t=(0,s.round)(e,2);if(null!==y&&_>=y)g=!0;else if(g||m.length>0&&_<v)g=!1;else if(null===c.timestamp[0])c.timestamp[0]=t;else if(t===c.timestamp[0]);else{c.timestamp[1]=t,m.push(x),o&&f.push(M);const[e,n]=this.findLongestCommonSequence(m,f),r=this.decode(e);c.text=r,o&&(c.words=this.collateWordTimestamps(e,n,a)),u.push(c),m=[],x=[],f=[],M=[],c=l()}}else if(x.push(_),o){let e,t=(0,s.round)(i[n]+p,2);if(n+1<i.length){e=(0,s.round)(i[n+1]+p,2);const a=this.decode([_]);b.test(a)&&(e=(0,s.round)(Math.min(t+r,e),2))}else e=null;M.push([t,e])}}if("stride"in n){const[e,t,r]=n.stride;p+=e-r}x.length>0?(m.push(x),o&&f.push(M)):m.every((e=>0===e.length))&&(c=l(),m=[],x=[],f=[],M=[])}if(m.length>0){if(i&&t)throw new Error("Whisper did not predict an ending timestamp, which can happen if audio is cut off in the middle of a word. Also make sure WhisperTimeStampLogitsProcessor was used during generation.");const[e,n]=this.findLongestCommonSequence(m,f),r=this.decode(e);c.text=r,o&&(c.words=this.collateWordTimestamps(e,n,a)),u.push(c)}let y=Object.create(null);const v=u.map((e=>e.text)).join("");if(t||n){for(let e=0;e<u.length;++e){const r=u[e];t||delete r.timestamp,n||delete r.language}if(o){const e=[];for(const t of u)for(const n of t.words)e.push(n);y={chunks:e}}else y={chunks:u}}return[v,y]}findLongestCommonSequence(e,t=null){let n=e[0],r=n.length,i=[];const a=Array.isArray(t)&&t.length>0;let s=a?[]:null,o=a?t[0]:null;for(let l=1;l<e.length;++l){const u=e[l];let d=0,c=[r,r,0,0];const p=u.length;for(let e=1;e<r+p;++e){const i=Math.max(0,r-e),s=Math.min(r,r+p-e),h=n.slice(i,s),m=Math.max(0,e-r),f=Math.min(p,e),g=u.slice(m,f);if(h.length!==g.length)throw new Error("There is a bug within whisper `decode_asr` function, please report it. Dropping to prevent bad inference.");let _;_=a?h.filter(((e,n)=>e===g[n]&&o[i+n]<=t[l][m+n])).length:h.filter(((e,t)=>e===g[t])).length;const w=_/e+e/1e4;_>1&&w>d&&(d=w,c=[i,s,m,f])}const[h,m,f,g]=c,_=Math.floor((m+h)/2),w=Math.floor((g+f)/2);i.push(...n.slice(0,_)),n=u.slice(w),r=n.length,a&&(s.push(...o.slice(0,_)),o=t[l].slice(w))}return i.push(...n),a?(s.push(...o),[i,s]):[i,[]]}collateWordTimestamps(e,t,n){const[r,i,a]=this.combineTokensIntoWords(e,n),s=[];for(let e=0;e<r.length;++e){const n=a[e];s.push({text:r[e],timestamp:[t[n.at(0)][0],t[n.at(-1)][1]]})}return s}combineTokensIntoWords(e,t,n="\"'“¡¿([{-",r="\"'.。,，!！?？:：”)]}、"){let i,a,s;return["chinese","japanese","thai","lao","myanmar"].includes(t=t??"english")?[i,a,s]=this.splitTokensOnUnicode(e):[i,a,s]=this.splitTokensOnSpaces(e),this.mergePunctuations(i,a,s,n,r)}decode(e,t){let n;return t?.decode_with_timestamps?(e instanceof o.Tensor&&(e=f(e)),n=this.decodeWithTimestamps(e,t)):n=super.decode(e,t),n}decodeWithTimestamps(e,t){const n=t?.time_precision??.02,r=Array.from(this.all_special_ids).at(-1)+1;let i=[[]];for(let t of e)if(t=Number(t),t>=r){const e=((t-r)*n).toFixed(2);i.push(`<|${e}|>`),i.push([])}else i[i.length-1].push(t);return i=i.map((e=>"string"==typeof e?e:super.decode(e,t))),i.join("")}splitTokensOnUnicode(e){const t=this.decode(e,{decode_with_timestamps:!0}),n=[],r=[],i=[];let a=[],s=[],o=0;for(let l=0;l<e.length;++l){const u=e[l];a.push(u),s.push(l);const d=this.decode(a,{decode_with_timestamps:!0});d.includes("�")&&"�"!==t[o+d.indexOf("�")]||(n.push(d),r.push(a),i.push(s),a=[],s=[],o+=d.length)}return[n,r,i]}splitTokensOnSpaces(e){const[t,n,r]=this.splitTokensOnUnicode(e),i=[],a=[],s=[],o=new RegExp(`^[${y}]$`,"gu");for(let e=0;e<t.length;++e){const l=t[e],u=n[e],d=r[e],c=u[0]>=this.model.tokens_to_ids.get("<|endoftext|>"),p=l.startsWith(" "),h=l.trim(),m=o.test(h);if(c||p||m||0===i.length)i.push(l),a.push(u),s.push(d);else{const e=i.length-1;i[e]+=l,a[e].push(...u),s[e].push(...d)}}return[i,a,s]}mergePunctuations(e,t,n,r,a){const s=structuredClone(e),o=structuredClone(t),l=structuredClone(n);let u=s.length-2,d=s.length-1;for(;u>=0;)s[u].startsWith(" ")&&r.includes(s[u].trim())?(s[d]=s[u]+s[d],o[d]=(0,i.mergeArrays)(o[u],o[d]),l[d]=(0,i.mergeArrays)(l[u],l[d]),s[u]="",o[u]=[],l[u]=[]):d=u,--u;for(u=0,d=1;d<s.length;)!s[u].endsWith(" ")&&a.includes(s[d])?(s[u]+=s[d],o[u]=(0,i.mergeArrays)(o[u],o[d]),l[u]=(0,i.mergeArrays)(l[u],l[d]),s[d]="",o[d]=[],l[d]=[]):u=d,++d;return[s.filter((e=>e)),o.filter((e=>e.length>0)),l.filter((e=>e.length>0))]}get_decoder_prompt_ids({language:e=null,task:t=null,no_timestamps:n=!0}={}){const r=[];if(e){const t=(0,d.whisper_language_to_code)(e),n=this.model.tokens_to_ids.get(`<|${t}|>`);if(void 0===n)throw new Error(`Unable to find language "${t}" in model vocabulary. Please report this issue at ${c.GITHUB_ISSUE_URL}.`);r.push(n)}else r.push(null);if(t){if("transcribe"!==(t=t.toLowerCase())&&"translate"!==t)throw new Error(`Task "${t}" is not supported. Must be one of: ["transcribe", "translate"]`);const e=this.model.tokens_to_ids.get(`<|${t}|>`);if(void 0===e)throw new Error(`Unable to find task "${t}" in model vocabulary. Please report this issue at ${c.GITHUB_ISSUE_URL}.`);r.push(e)}else r.push(null);if(n){const e=this.model.tokens_to_ids.get("<|notimestamps|>");if(void 0===e)throw new Error(`Unable to find "<|notimestamps|>" in model vocabulary. Please report this issue at ${c.GITHUB_ISSUE_URL}.`);r.push(e)}return r.map(((e,t)=>[t+1,e])).filter((e=>null!==e[1]))}}class nt extends be{}class rt extends be{}class it extends be{}class at extends be{constructor(e,t){super(e,t),this.languageRegex=/^(>>\w+<<)\s*/g,this.supported_language_codes=this.model.vocab.filter((e=>this.languageRegex.test(e))),console.warn('WARNING: `MarianTokenizer` is not yet supported by Hugging Face\'s "fast" tokenizers library. Therefore, you may experience slightly inaccurate results.')}_encode_text(e){if(null===e)return null;const[t,...n]=e.trim().split(this.languageRegex);if(0===n.length)return super._encode_text(t);if(2===n.length){const[e,t]=n;return this.supported_language_codes.includes(e)||console.warn(`Unsupported language code "${e}" detected, which may lead to unexpected behavior. Should be one of: ${JSON.stringify(this.supported_language_codes)}`),(0,i.mergeArrays)([e],super._encode_text(t))}}}class st extends be{}class ot extends be{}class lt extends be{}class ut extends be{}class dt extends be{}class ct extends be{constructor(e,t){super(e,t),this.decoder=new ue({})}}class pt extends be{}class ht{static TOKENIZER_CLASS_MAPPING={T5Tokenizer:ze,DistilBertTokenizer:Ee,CamembertTokenizer:Fe,DebertaTokenizer:ke,DebertaV2Tokenizer:$e,BertTokenizer:ve,HerbertTokenizer:Se,ConvBertTokenizer:Ce,RoFormerTokenizer:Pe,XLMTokenizer:Ae,ElectraTokenizer:Ie,MobileBertTokenizer:Me,SqueezeBertTokenizer:Te,AlbertTokenizer:xe,GPT2Tokenizer:Oe,BartTokenizer:Be,MBartTokenizer:Le,MBart50Tokenizer:De,RobertaTokenizer:Re,WhisperTokenizer:tt,CodeGenTokenizer:nt,CLIPTokenizer:rt,SiglipTokenizer:it,MarianTokenizer:at,BloomTokenizer:Ne,NllbTokenizer:Ze,M2M100Tokenizer:et,LlamaTokenizer:je,CodeLlamaTokenizer:Ge,XLMRobertaTokenizer:qe,MPNetTokenizer:Ue,FalconTokenizer:We,GPTNeoXTokenizer:He,EsmTokenizer:Xe,Wav2Vec2CTCTokenizer:st,BlenderbotTokenizer:ot,BlenderbotSmallTokenizer:lt,SpeechT5Tokenizer:ut,NougatTokenizer:dt,VitsTokenizer:ct,Qwen2Tokenizer:Ke,GemmaTokenizer:Qe,Grok1Tokenizer:Ye,CohereTokenizer:pt,PreTrainedTokenizer:be};static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:r=null,local_files_only:i=!1,revision:a="main",legacy:s=null}={}){const[o,l]=await p(e,{progress_callback:t,config:n,cache_dir:r,local_files_only:i,revision:a,legacy:s}),u=l.tokenizer_class?.replace(/Fast$/,"")??"PreTrainedTokenizer";let d=this.TOKENIZER_CLASS_MAPPING[u];return d||(console.warn(`Unknown tokenizer class "${u}", attempting to construct from base class.`),d=be),new d(o,l)}}},"./src/utils/audio.js":
+  \***************************/(e,t,n)=>{n.r(t),n.d(t,{AlbertTokenizer:()=>xe,AutoTokenizer:()=>ht,BartTokenizer:()=>Be,BertTokenizer:()=>ve,BlenderbotSmallTokenizer:()=>lt,BlenderbotTokenizer:()=>ot,BloomTokenizer:()=>Ne,CLIPTokenizer:()=>rt,CamembertTokenizer:()=>Fe,CodeGenTokenizer:()=>nt,CodeLlamaTokenizer:()=>Ge,CohereTokenizer:()=>pt,ConvBertTokenizer:()=>Ce,DebertaTokenizer:()=>ke,DebertaV2Tokenizer:()=>$e,DistilBertTokenizer:()=>Ee,ElectraTokenizer:()=>Ie,EsmTokenizer:()=>Xe,FalconTokenizer:()=>We,GPT2Tokenizer:()=>Oe,GPTNeoXTokenizer:()=>He,GemmaTokenizer:()=>Qe,Grok1Tokenizer:()=>Ye,HerbertTokenizer:()=>Se,LlamaTokenizer:()=>je,M2M100Tokenizer:()=>et,MBart50Tokenizer:()=>De,MBartTokenizer:()=>Le,MPNetTokenizer:()=>Ue,MarianTokenizer:()=>at,MobileBertTokenizer:()=>Me,NllbTokenizer:()=>Ze,NougatTokenizer:()=>dt,PreTrainedTokenizer:()=>be,Qwen2Tokenizer:()=>Ke,RoFormerTokenizer:()=>Pe,RobertaTokenizer:()=>Re,SiglipTokenizer:()=>it,SpeechT5Tokenizer:()=>ut,SqueezeBertTokenizer:()=>Te,T5Tokenizer:()=>ze,TokenizerModel:()=>M,VitsTokenizer:()=>ct,Wav2Vec2CTCTokenizer:()=>st,WhisperTokenizer:()=>tt,XLMRobertaTokenizer:()=>qe,XLMTokenizer:()=>Ae,is_chinese_char:()=>w});var r=n(/*! ./utils/generic.js */"./src/utils/generic.js"),i=n(/*! ./utils/core.js */"./src/utils/core.js"),a=n(/*! ./utils/hub.js */"./src/utils/hub.js"),s=n(/*! ./utils/maths.js */"./src/utils/maths.js"),o=n(/*! ./utils/tensor.js */"./src/utils/tensor.js"),l=n(/*! ./utils/data-structures.js */"./src/utils/data-structures.js"),u=n(/*! @huggingface/jinja */"./node_modules/@huggingface/jinja/dist/index.js"),d=n(/*! ./models/whisper/common_whisper.js */"./src/models/whisper/common_whisper.js"),c=n(/*! ./utils/constants.js */"./src/utils/constants.js");async function p(e,t){const n=await Promise.all([(0,a.getModelJSON)(e,"tokenizer.json",!0,t),(0,a.getModelJSON)(e,"tokenizer_config.json",!0,t)]);return null!==t.legacy&&(n[1].legacy=t.legacy),n}function h(e,t=!0){if(void 0!==e.Regex){let t=e.Regex.replace(/\\([#&~])/g,"$1");for(const[e,n]of v)t=t.replaceAll(e,n);return new RegExp(t,"gu")}if(void 0!==e.String){const n=(0,i.escapeRegExp)(e.String);return new RegExp(t?n:`(${n})`,"gu")}return console.warn("Unknown pattern type:",e),null}function m(e){return new Map(Object.entries(e))}function f(e){const t=e.dims;switch(t.length){case 1:return e.tolist();case 2:if(1!==t[0])throw new Error("Unable to decode tensor with `batch size !== 1`. Use `tokenizer.batch_decode(...)` for batched inputs.");return e.tolist()[0];default:throw new Error(`Expected tensor to have 1-2 dimensions, got ${t.length}.`)}}function g(e){return e.replace(/ \./g,".").replace(/ \?/g,"?").replace(/ \!/g,"!").replace(/ ,/g,",").replace(/ \' /g,"'").replace(/ n\'t/g,"n't").replace(/ \'m/g,"'m").replace(/ \'s/g,"'s").replace(/ \'ve/g,"'ve").replace(/ \'re/g,"'re")}function _(e){return e.replace(/\p{M}/gu,"")}function w(e){return e>=19968&&e<=40959||e>=13312&&e<=19903||e>=131072&&e<=173791||e>=173824&&e<=177983||e>=177984&&e<=178207||e>=178208&&e<=183983||e>=63744&&e<=64255||e>=194560&&e<=195103}const y="\\p{P}\\u0021-\\u002F\\u003A-\\u0040\\u005B-\\u0060\\u007B-\\u007E",b=new RegExp(`^[${y}]+$`,"gu"),v=new Map([["(?i:'s|'t|'re|'ve|'m|'ll|'d)","(?:'([sS]|[tT]|[rR][eE]|[vV][eE]|[mM]|[lL][lL]|[dD]))"]]);class x{constructor(e){this.content=e.content,this.id=e.id,this.single_word=e.single_word??!1,this.lstrip=e.lstrip??!1,this.rstrip=e.rstrip??!1,this.special=e.special??!1,this.normalized=e.normalized??null}}class M extends r.Callable{constructor(e){super(),this.config=e,this.vocab=[],this.tokens_to_ids=new Map,this.unk_token_id=void 0,this.unk_token=void 0,this.end_of_word_suffix=void 0,this.fuse_unk=this.config.fuse_unk??!1}static fromConfig(e,...t){switch(e.type){case"WordPiece":return new T(e);case"Unigram":return new k(e,...t);case"BPE":return new C(e);default:if(e.vocab)return new P(e,...t);throw new Error(`Unknown TokenizerModel type: ${e.type}`)}}_call(e){return e=this.encode(e),this.fuse_unk&&(e=function(e,t,n){const r=[];let i=0;for(;i<e.length;)if(r.push(e[i]),(t.get(e[i])??n)===n)for(;++i<e.length&&(t.get(e[i])??n)===n;)t.get(r.at(-1))!==n&&(r[r.length-1]+=e[i]);else++i;return r}(e,this.tokens_to_ids,this.unk_token_id)),e}encode(e){throw Error("encode should be implemented in subclass.")}convert_tokens_to_ids(e){return e.map((e=>this.tokens_to_ids.get(e)??this.unk_token_id))}convert_ids_to_tokens(e){return e.map((e=>this.vocab[e]??this.unk_token))}}class T extends M{constructor(e){super(e),this.tokens_to_ids=m(e.vocab),this.unk_token_id=this.tokens_to_ids.get(e.unk_token),this.unk_token=e.unk_token,this.max_input_chars_per_word=e.max_input_chars_per_word??100,this.vocab=new Array(this.tokens_to_ids.size);for(const[e,t]of this.tokens_to_ids)this.vocab[t]=e}encode(e){const t=[];for(const n of e){const e=[...n];if(e.length>this.max_input_chars_per_word){t.push(this.unk_token);continue}let r=!1,i=0;const a=[];for(;i<e.length;){let t=e.length,n=null;for(;i<t;){let r=e.slice(i,t).join("");if(i>0&&(r=this.config.continuing_subword_prefix+r),this.tokens_to_ids.has(r)){n=r;break}--t}if(null===n){r=!0;break}a.push(n),i=t}r?t.push(this.unk_token):t.push(...a)}return t}}class k extends M{constructor(e,t){super(e);const n=e.vocab.length;this.vocab=new Array(n),this.scores=new Array(n);for(let t=0;t<n;++t){const n=e.vocab[t];this.vocab[t]=n[0],this.scores[t]=n[1]}this.unk_token_id=e.unk_id,this.unk_token=this.vocab[e.unk_id],this.tokens_to_ids=new Map(this.vocab.map(((e,t)=>[e,t]))),this.bos_token=" ",this.bos_token_id=this.tokens_to_ids.get(this.bos_token),this.eos_token=t.eos_token,this.eos_token_id=this.tokens_to_ids.get(this.eos_token),this.unk_token=this.vocab[this.unk_token_id],this.minScore=(0,s.min)(this.scores)[0],this.unk_score=this.minScore-10,this.scores[this.unk_token_id]=this.unk_score,this.trie=new l.CharTrie,this.trie.extend(this.vocab),this.fuse_unk=!0}populateNodes(e){const t=e.chars;let n=0;for(;n<t.length;){let r=!1;const a=[],s=t.slice(n).join(""),o=this.trie.commonPrefixSearch(s);for(const t of o){a.push(t);const s=this.tokens_to_ids.get(t),o=this.scores[s],l=(0,i.len)(t);e.insert(n,l,o,s),r||1!==l||(r=!0)}r||e.insert(n,1,this.unk_score,this.unk_token_id),n+=1}}tokenize(e){const t=new l.TokenLattice(e,this.bos_token_id,this.eos_token_id);return this.populateNodes(t),t.tokens()}encode(e){const t=[];for(const n of e){const e=this.tokenize(n);t.push(...e)}return t}}const $=(()=>{const e=[...Array.from({length:"~".charCodeAt(0)-"!".charCodeAt(0)+1},((e,t)=>t+"!".charCodeAt(0))),...Array.from({length:"¬".charCodeAt(0)-"¡".charCodeAt(0)+1},((e,t)=>t+"¡".charCodeAt(0))),...Array.from({length:"ÿ".charCodeAt(0)-"®".charCodeAt(0)+1},((e,t)=>t+"®".charCodeAt(0)))],t=e.slice();let n=0;for(let r=0;r<256;++r)e.includes(r)||(e.push(r),t.push(256+n),n+=1);const r=t.map((e=>String.fromCharCode(e)));return Object.fromEntries(e.map(((e,t)=>[e,r[t]])))})(),S=(0,i.reverseDictionary)($);class C extends M{constructor(e){super(e),this.BPE_SPLIT_TOKEN=" ",this.tokens_to_ids=m(e.vocab),this.unk_token_id=this.tokens_to_ids.get(e.unk_token),this.unk_token=e.unk_token,this.vocab=new Array(this.tokens_to_ids.size);for(const[e,t]of this.tokens_to_ids)this.vocab[t]=e;this.bpe_ranks=new Map(e.merges.map(((e,t)=>[e,t]))),this.merges=e.merges.map((e=>e.split(this.BPE_SPLIT_TOKEN))),this.end_of_word_suffix=e.end_of_word_suffix,this.continuing_subword_suffix=e.continuing_subword_suffix??null,this.byte_fallback=this.config.byte_fallback??!1,this.byte_fallback&&(this.text_encoder=new TextEncoder),this.ignore_merges=this.config.ignore_merges??!1,this.cache=new Map}bpe(e){if(0===e.length)return[];const t=this.cache.get(e);if(void 0!==t)return t;const n=Array.from(e);this.end_of_word_suffix&&(n[n.length-1]+=this.end_of_word_suffix);let r=[];if(n.length>1){const e=new l.PriorityQueue(((e,t)=>e.score<t.score));let t={token:n[0],bias:0,prev:null,next:null},i=t;for(let t=1;t<n.length;++t){const r={bias:t/n.length,token:n[t],prev:i,next:null};i.next=r,this._add_node(e,i),i=r}for(;!e.isEmpty();){const n=e.pop();if(n.deleted||!n.next||n.next.deleted)continue;if(n.deleted=!0,n.next.deleted=!0,n.prev){const e={...n.prev};n.prev.deleted=!0,n.prev=e,e.prev?e.prev.next=e:t=e}const r={token:n.token+n.next.token,bias:n.bias,prev:n.prev,next:n.next.next};r.prev?(r.prev.next=r,this._add_node(e,r.prev)):t=r,r.next&&(r.next.prev=r,this._add_node(e,r))}for(let e=t;null!==e;e=e.next)r.push(e.token)}else r=n;if(this.continuing_subword_suffix)for(let e=0;e<r.length-1;++e)r[e]+=this.continuing_subword_suffix;return this.cache.set(e,r),r}_add_node(e,t){const n=this.bpe_ranks.get(t.token+this.BPE_SPLIT_TOKEN+t.next.token);void 0!==n&&(t.score=n+t.bias,e.push(t))}encode(e){const t=[];for(const n of e){if(this.ignore_merges&&this.tokens_to_ids.has(n)){t.push(n);continue}const e=this.bpe(n);for(const n of e)if(this.tokens_to_ids.has(n))t.push(n);else if(this.byte_fallback){const e=Array.from(this.text_encoder.encode(n)).map((e=>`<0x${e.toString(16).toUpperCase().padStart(2,"0")}>`));e.every((e=>this.tokens_to_ids.has(e)))?t.push(...e):t.push(this.unk_token)}else t.push(this.unk_token)}return t}}class P extends M{constructor(e,t){super(e),this.tokens_to_ids=m(t.target_lang?e.vocab[t.target_lang]:e.vocab),this.bos_token=t.bos_token,this.bos_token_id=this.tokens_to_ids.get(this.bos_token),this.eos_token=t.eos_token,this.eos_token_id=this.tokens_to_ids.get(this.eos_token),this.pad_token=t.pad_token,this.pad_token_id=this.tokens_to_ids.get(this.pad_token),this.unk_token=t.unk_token,this.unk_token_id=this.tokens_to_ids.get(this.unk_token),this.vocab=new Array(this.tokens_to_ids.size);for(const[e,t]of this.tokens_to_ids)this.vocab[t]=e}encode(e){return e}}class E extends r.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"BertNormalizer":return new N(e);case"Precompiled":return new pe(e);case"Sequence":return new R(e);case"Replace":return new F(e);case"NFC":return new A(e);case"NFKC":return new I(e);case"NFKD":return new z(e);case"Strip":return new O(e);case"StripAccents":return new B(e);case"Lowercase":return new L(e);case"Prepend":return new D(e);default:throw new Error(`Unknown Normalizer type: ${e.type}`)}}normalize(e){throw Error("normalize should be implemented in subclass.")}_call(e){return this.normalize(e)}}class F extends E{normalize(e){const t=h(this.config.pattern);return null===t?e:e.replaceAll(t,this.config.content)}}class A extends E{normalize(e){return e=e.normalize("NFC")}}class I extends E{normalize(e){return e=e.normalize("NFKC")}}class z extends E{normalize(e){return e=e.normalize("NFKD")}}class O extends E{normalize(e){return this.config.strip_left&&this.config.strip_right?e=e.trim():(this.config.strip_left&&(e=e.trimStart()),this.config.strip_right&&(e=e.trimEnd())),e}}class B extends E{normalize(e){return e=_(e)}}class L extends E{normalize(e){return e=e.toLowerCase()}}class D extends E{normalize(e){return e=this.config.prepend+e}}class R extends E{constructor(e){super(e),this.normalizers=e.normalizers.map((e=>E.fromConfig(e)))}normalize(e){return this.normalizers.reduce(((e,t)=>t.normalize(e)),e)}}class N extends E{_tokenize_chinese_chars(e){const t=[];for(let n=0;n<e.length;++n){const r=e[n];w(r.charCodeAt(0))?(t.push(" "),t.push(r),t.push(" ")):t.push(r)}return t.join("")}stripAccents(e){return e.normalize("NFD").replace(/\p{Mn}/gu,"")}_is_control(e){switch(e){case"\t":case"\n":case"\r":return!1;default:return/^\p{Cc}|\p{Cf}|\p{Co}|\p{Cs}$/u.test(e)}}_clean_text(e){const t=[];for(const n of e){const e=n.charCodeAt(0);0===e||65533===e||this._is_control(n)||(/^\s$/.test(n)?t.push(" "):t.push(n))}return t.join("")}normalize(e){return this.config.clean_text&&(e=this._clean_text(e)),this.config.handle_chinese_chars&&(e=this._tokenize_chinese_chars(e)),this.config.lowercase?(e=e.toLowerCase(),!1!==this.config.strip_accents&&(e=this.stripAccents(e))):this.config.strip_accents&&(e=this.stripAccents(e)),e}}class V extends r.Callable{static fromConfig(e){if(null===e)return null;switch(e.type){case"BertPreTokenizer":return new j(e);case"Sequence":return new he(e);case"Whitespace":return new me(e);case"WhitespaceSplit":return new fe(e);case"Metaspace":return new de(e);case"ByteLevel":return new G(e);case"Split":return new q(e);case"Punctuation":return new U(e);case"Digits":return new W(e);case"Replace":return new ge(e);default:throw new Error(`Unknown PreTokenizer type: ${e.type}`)}}pre_tokenize_text(e,t){throw Error("pre_tokenize_text should be implemented in subclass.")}pre_tokenize(e,t){return(Array.isArray(e)?e.map((e=>this.pre_tokenize_text(e,t))):this.pre_tokenize_text(e,t)).flat()}_call(e,t){return this.pre_tokenize(e,t)}}class j extends V{constructor(e){super(),this.pattern=new RegExp(`[^\\s${y}]+|[${y}]`,"gu")}pre_tokenize_text(e,t){return e.trim().match(this.pattern)||[]}}class G extends V{constructor(e){super(),this.config=e,this.add_prefix_space=this.config.add_prefix_space,this.trim_offsets=this.config.trim_offsets,this.use_regex=this.config.use_regex??!0,this.pattern=/'s|'t|'re|'ve|'m|'ll|'d| ?\p{L}+| ?\p{N}+| ?[^\s\p{L}\p{N}]+|\s+(?!\S)|\s+/gu,this.byte_encoder=$,this.text_encoder=new TextEncoder}pre_tokenize_text(e,t){this.add_prefix_space&&!e.startsWith(" ")&&(e=" "+e);return(this.use_regex?e.match(this.pattern)||[]:[e]).map((e=>Array.from(this.text_encoder.encode(e),(e=>this.byte_encoder[e])).join("")))}}class q extends V{constructor(e){super(),this.config=e,this.pattern=h(this.config.pattern,this.config.invert)}pre_tokenize_text(e,t){return null===this.pattern?[]:this.config.invert?e.match(this.pattern)||[]:function(e,t){const n=[];let r=0;for(const i of e.matchAll(t)){const t=i[0];r<i.index&&n.push(e.slice(r,i.index)),t.length>0&&n.push(t),r=i.index+t.length}return r<e.length&&n.push(e.slice(r)),n}(e,this.pattern)}}class U extends V{constructor(e){super(),this.config=e,this.pattern=new RegExp(`[^${y}]+|[${y}]+`,"gu")}pre_tokenize_text(e,t){return e.match(this.pattern)||[]}}class W extends V{constructor(e){super(),this.config=e;const t="[^\\d]+|\\d"+(this.config.individual_digits?"":"+");this.pattern=new RegExp(t,"gu")}pre_tokenize_text(e,t){return e.match(this.pattern)||[]}}class H extends r.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"TemplateProcessing":return new Q(e);case"ByteLevel":return new Y(e);case"RobertaProcessing":return new K(e);case"BertProcessing":return new X(e);case"Sequence":return new J(e);default:throw new Error(`Unknown PostProcessor type: ${e.type}`)}}post_process(e,...t){throw Error("post_process should be implemented in subclass.")}_call(e,...t){return this.post_process(e,...t)}}class X extends H{constructor(e){super(e),this.cls=e.cls[0],this.sep=e.sep[0]}post_process(e,t=null,{add_special_tokens:n=!0}={}){n&&(e=(0,i.mergeArrays)([this.cls],e,[this.sep]));let r=new Array(e.length).fill(0);if(null!==t){const a=n&&this instanceof K?[this.sep]:[],s=n?[this.sep]:[];e=(0,i.mergeArrays)(e,a,t,s),r=(0,i.mergeArrays)(r,new Array(t.length+a.length+s.length).fill(1))}return{tokens:e,token_type_ids:r}}}class K extends X{}class Q extends H{constructor(e){super(e),this.single=e.single,this.pair=e.pair}post_process(e,t=null,{add_special_tokens:n=!0}={}){const r=null===t?this.single:this.pair;let a=[],s=[];for(const o of r)"SpecialToken"in o?n&&(a.push(o.SpecialToken.id),s.push(o.SpecialToken.type_id)):"Sequence"in o&&("A"===o.Sequence.id?(a=(0,i.mergeArrays)(a,e),s=(0,i.mergeArrays)(s,new Array(e.length).fill(o.Sequence.type_id))):"B"===o.Sequence.id&&(a=(0,i.mergeArrays)(a,t),s=(0,i.mergeArrays)(s,new Array(t.length).fill(o.Sequence.type_id))));return{tokens:a,token_type_ids:s}}}class Y extends H{post_process(e,t=null){return t&&(e=(0,i.mergeArrays)(e,t)),{tokens:e}}}class J extends H{constructor(e){super(e),this.processors=e.processors.map((e=>H.fromConfig(e)))}post_process(e,t=null,n={}){let r;for(const i of this.processors)if(i instanceof Y){if(e=i.post_process(e).tokens,t){t=i.post_process(t).tokens}}else{const a=i.post_process(e,t,n);e=a.tokens,r=a.token_type_ids}return{tokens:e,token_type_ids:r}}}class Z extends r.Callable{constructor(e){super(),this.config=e,this.added_tokens=[],this.end_of_word_suffix=null,this.trim_offsets=e.trim_offsets}static fromConfig(e){if(null===e)return null;switch(e.type){case"WordPiece":return new ie(e);case"Metaspace":return new ce(e);case"ByteLevel":return new ae(e);case"Replace":return new ee(e);case"ByteFallback":return new te(e);case"Fuse":return new ne(e);case"Strip":return new re(e);case"Sequence":return new oe(e);case"CTC":return new se(e);case"BPEDecoder":return new le(e);default:throw new Error(`Unknown Decoder type: ${e.type}`)}}_call(e){return this.decode(e)}decode(e){return this.decode_chain(e).join("")}decode_chain(e){throw Error("`decode_chain` should be implemented in subclass.")}}class ee extends Z{decode_chain(e){const t=h(this.config.pattern);return null===t?e:e.map((e=>e.replaceAll(t,this.config.content)))}}class te extends Z{constructor(e){super(e),this.text_decoder=new TextDecoder}decode_chain(e){const t=[];let n=[];for(const r of e){let e=null;if(6===r.length&&r.startsWith("<0x")&&r.endsWith(">")){const t=parseInt(r.slice(3,5),16);isNaN(t)||(e=t)}if(null!==e)n.push(e);else{if(n.length>0){const e=this.text_decoder.decode(Uint8Array.from(n));t.push(e),n=[]}t.push(r)}}if(n.length>0){const e=this.text_decoder.decode(Uint8Array.from(n));t.push(e),n=[]}return t}}class ne extends Z{decode_chain(e){return[e.join("")]}}class re extends Z{constructor(e){super(e),this.content=this.config.content,this.start=this.config.start,this.stop=this.config.stop}decode_chain(e){return e.map((e=>{let t=0;for(let n=0;n<this.start&&e[n]===this.content;++n)t=n+1;let n=e.length;for(let t=0;t<this.stop;++t){const r=e.length-t-1;if(e[r]!==this.content)break;n=r}return e.slice(t,n)}))}}class ie extends Z{constructor(e){super(e),this.cleanup=e.cleanup}decode_chain(e){return e.map(((e,t)=>(0!==t&&(e=e.startsWith(this.config.prefix)?e.replace(this.config.prefix,""):" "+e),this.cleanup&&(e=g(e)),e)))}}class ae extends Z{constructor(e){super(e),this.byte_decoder=S,this.text_decoder=new TextDecoder("utf-8",{fatal:!1,ignoreBOM:!0}),this.end_of_word_suffix=null}convert_tokens_to_string(e){const t=e.join(""),n=new Uint8Array([...t].map((e=>this.byte_decoder[e])));return this.text_decoder.decode(n)}decode_chain(e){const t=[];let n=[];for(const r of e)void 0!==this.added_tokens.find((e=>e.content===r))?(n.length>0&&(t.push(this.convert_tokens_to_string(n)),n=[]),t.push(r)):n.push(r);return n.length>0&&t.push(this.convert_tokens_to_string(n)),t}}class se extends Z{constructor(e){super(e),this.pad_token=this.config.pad_token,this.word_delimiter_token=this.config.word_delimiter_token,this.cleanup=this.config.cleanup}convert_tokens_to_string(e){if(0===e.length)return"";const t=[e[0]];for(let n=1;n<e.length;++n)e[n]!==t.at(-1)&&t.push(e[n]);let n=t.filter((e=>e!==this.pad_token)).join("");return this.cleanup&&(n=g(n).replaceAll(this.word_delimiter_token," ").trim()),n}decode_chain(e){return[this.convert_tokens_to_string(e)]}}class oe extends Z{constructor(e){super(e),this.decoders=e.decoders.map((e=>Z.fromConfig(e)))}decode_chain(e){return this.decoders.reduce(((e,t)=>t.decode_chain(e)),e)}}class le extends Z{constructor(e){super(e),this.suffix=this.config.suffix}decode_chain(e){return e.map(((t,n)=>t.replaceAll(this.suffix,n===e.length-1?"":" ")))}}class ue extends Z{decode_chain(e){let t="";for(let n=1;n<e.length;n+=2)t+=e[n];return[t]}}class de extends V{constructor(e){super(),this.addPrefixSpace=e.add_prefix_space,this.replacement=e.replacement,this.strRep=e.str_rep||this.replacement,this.prepend_scheme=e.prepend_scheme??"always"}pre_tokenize_text(e,{section_index:t}={}){let n=e.replaceAll(" ",this.strRep);return this.addPrefixSpace&&!n.startsWith(this.replacement)&&("always"===this.prepend_scheme||"first"===this.prepend_scheme&&0===t)&&(n=this.strRep+n),[n]}}class ce extends Z{constructor(e){super(e),this.addPrefixSpace=e.add_prefix_space,this.replacement=e.replacement}decode_chain(e){const t=[];for(let n=0;n<e.length;++n){let r=e[n].replaceAll(this.replacement," ");this.addPrefixSpace&&0==n&&r.startsWith(" ")&&(r=r.substring(1)),t.push(r)}return t}}class pe extends E{constructor(e){super(e),this.charsmap=e.precompiled_charsmap}normalize(e){if((e=(e=e.replace(/[\u0001-\u0008\u000B\u000E-\u001F\u007F\u008F\u009F]/gm,"")).replace(/[\u0009\u000A\u000C\u000D\u00A0\u1680\u2000-\u200F\u2028\u2029\u202F\u205F\u2581\u3000\uFEFF\uFFFD]/gm," ")).includes("～")){const t=e.split("～");e=t.map((e=>e.normalize("NFKC"))).join("～")}else e=e.normalize("NFKC");return e}}class he extends V{constructor(e){super(),this.tokenizers=e.pretokenizers.map((e=>V.fromConfig(e)))}pre_tokenize_text(e,t){return this.tokenizers.reduce(((e,n)=>n.pre_tokenize(e,t)),[e])}}class me extends V{constructor(e){super()}pre_tokenize_text(e,t){return e.match(/\w+|[^\w\s]+/g)||[]}}class fe extends V{constructor(e){super()}pre_tokenize_text(e,t){return function(e){return e.match(/\S+/g)||[]}(e)}}class ge extends V{constructor(e){super(),this.config=e,this.pattern=h(this.config.pattern),this.content=this.config.content}pre_tokenize_text(e,t){return null===this.pattern?[e]:[e.replaceAll(this.pattern,this.config.content)]}}const _e=["bos_token","eos_token","unk_token","sep_token","pad_token","cls_token","mask_token"];function we(e,t,n,r){for(const a of Object.keys(e)){const s=t-e[a].length,o=n(a),l=new Array(s).fill(o);e[a]="right"===r?(0,i.mergeArrays)(e[a],l):(0,i.mergeArrays)(l,e[a])}}function ye(e,t){for(const n of Object.keys(e))e[n].length=t}class be extends r.Callable{return_token_type_ids=!1;padding_side="right";constructor(e,t){super(),this._tokenizer_config=t,this.normalizer=E.fromConfig(e.normalizer),this.pre_tokenizer=V.fromConfig(e.pre_tokenizer),this.model=M.fromConfig(e.model,t),this.post_processor=H.fromConfig(e.post_processor),this.decoder=Z.fromConfig(e.decoder),this.special_tokens=[],this.all_special_ids=[],this.added_tokens=[];for(const t of e.added_tokens){const e=new x(t);this.added_tokens.push(e),this.model.tokens_to_ids.set(e.content,e.id),this.model.vocab[e.id]=e.content,e.special&&(this.special_tokens.push(e.content),this.all_special_ids.push(e.id))}if(this.additional_special_tokens=t.additional_special_tokens??[],this.special_tokens.push(...this.additional_special_tokens),this.special_tokens=[...new Set(this.special_tokens)],this.decoder&&(this.decoder.added_tokens=this.added_tokens,this.decoder.end_of_word_suffix=this.model.end_of_word_suffix),this.added_tokens_regex=this.added_tokens.length>0?new RegExp(this.added_tokens.slice().sort(((e,t)=>t.content.length-e.content.length)).map((e=>`${e.lstrip?"\\s*":""}(${(0,i.escapeRegExp)(e.content)})${e.rstrip?"\\s*":""}`)).join("|")):null,this.mask_token=this.getToken("mask_token"),this.mask_token_id=this.model.tokens_to_ids.get(this.mask_token),this.pad_token=this.getToken("pad_token","eos_token"),this.pad_token_id=this.model.tokens_to_ids.get(this.pad_token),this.sep_token=this.getToken("sep_token"),this.sep_token_id=this.model.tokens_to_ids.get(this.sep_token),this.unk_token=this.getToken("unk_token"),this.unk_token_id=this.model.tokens_to_ids.get(this.unk_token),this.model_max_length=t.model_max_length,this.remove_space=t.remove_space,this.clean_up_tokenization_spaces=t.clean_up_tokenization_spaces??!0,this.do_lowercase_and_remove_accent=t.do_lowercase_and_remove_accent??!1,t.padding_side&&(this.padding_side=t.padding_side),this.legacy=!1,this.chat_template=t.chat_template??null,Array.isArray(this.chat_template)){const e=Object.create(null);for(const{name:t,template:n}of this.chat_template){if("string"!=typeof t||"string"!=typeof n)throw new Error('Chat template must be a list of objects with "name" and "template" properties');e[t]=n}this.chat_template=e}this._compiled_template_cache=new Map}getToken(...e){for(const t of e){const e=this._tokenizer_config[t];if(e){if("object"==typeof e){if("AddedToken"===e.__type)return e.content;throw Error(`Unknown token: ${e}`)}return e}}return null}static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:r=null,local_files_only:i=!1,revision:a="main",legacy:s=null}={}){return new this(...await p(e,{progress_callback:t,config:n,cache_dir:r,local_files_only:i,revision:a,legacy:s}))}_call(e,{text_pair:t=null,add_special_tokens:n=!0,padding:r=!1,truncation:i=null,max_length:a=null,return_tensor:l=!0,return_token_type_ids:u=null}={}){const d=Array.isArray(e);let c;if(d){if(0===e.length)throw Error("text array must be non-empty");if(null!==t){if(!Array.isArray(t))throw Error("text_pair must also be an array");if(e.length!==t.length)throw Error("text and text_pair must have the same length");c=e.map(((e,r)=>this._encode_plus(e,{text_pair:t[r],add_special_tokens:n,return_token_type_ids:u})))}else c=e.map((e=>this._encode_plus(e,{add_special_tokens:n,return_token_type_ids:u})))}else{if(null==e)throw Error("text may not be null or undefined");if(Array.isArray(t))throw Error("When specifying `text_pair`, since `text` is a string, `text_pair` must also be a string (i.e., not an array).");c=[this._encode_plus(e,{text_pair:t,add_special_tokens:n,return_token_type_ids:u})]}if(null===a?a="max_length"===r?this.model_max_length:(0,s.max)(c.map((e=>e.input_ids.length)))[0]:i||console.warn("Truncation was not explicitly activated but `max_length` is provided a specific value, please use `truncation=true` to explicitly truncate examples to max length."),a=Math.min(a,this.model_max_length??1/0),r||i)for(let e=0;e<c.length;++e)c[e].input_ids.length!==a&&(c[e].input_ids.length>a?i&&ye(c[e],a):r&&we(c[e],a,(e=>"input_ids"===e?this.pad_token_id:0),this.padding_side));const p={};if(l){if((!r||!i)&&c.some((e=>{for(const t of Object.keys(e))if(e[t].length!==c[0][t]?.length)return!0;return!1})))throw Error("Unable to create tensor, you should probably activate truncation and/or padding with 'padding=true' and 'truncation=true' to have batched tensors with the same length.");const e=[c.length,c[0].input_ids.length];for(const t of Object.keys(c[0]))p[t]=new o.Tensor("int64",BigInt64Array.from(c.flatMap((e=>e[t])).map(BigInt)),e)}else{for(const e of Object.keys(c[0]))p[e]=c.map((t=>t[e]));if(!d)for(const e of Object.keys(p))p[e]=p[e][0]}return p}_encode_text(e){if(null===e)return null;const t=(this.added_tokens_regex?e.split(this.added_tokens_regex).filter((e=>e)):[e]).map(((e,t)=>{if(void 0!==this.added_tokens.find((t=>t.content===e)))return e;{if(!0===this.remove_space&&(e=e.trim().split(/\s+/).join(" ")),this.do_lowercase_and_remove_accent&&(e=function(e){return _(e.toLowerCase())}(e)),null!==this.normalizer&&(e=this.normalizer(e)),0===e.length)return[];const n=null!==this.pre_tokenizer?this.pre_tokenizer(e,{section_index:t}):[e];return this.model(n)}})).flat();return t}_encode_plus(e,{text_pair:t=null,add_special_tokens:n=!0,return_token_type_ids:r=null}={}){const{tokens:i,token_type_ids:a}=this._tokenize_helper(e,{pair:t,add_special_tokens:n}),s=this.model.convert_tokens_to_ids(i),o={input_ids:s,attention_mask:new Array(s.length).fill(1)};return(r??this.return_token_type_ids)&&a&&(o.token_type_ids=a),o}_tokenize_helper(e,{pair:t=null,add_special_tokens:n=!1}={}){const r=this._encode_text(e),a=this._encode_text(t);return this.post_processor?this.post_processor(r,a,{add_special_tokens:n}):{tokens:(0,i.mergeArrays)(r??[],a??[])}}tokenize(e,{pair:t=null,add_special_tokens:n=!1}={}){return this._tokenize_helper(e,{pair:t,add_special_tokens:n}).tokens}encode(e,{text_pair:t=null,add_special_tokens:n=!0,return_token_type_ids:r=null}={}){return this._encode_plus(e,{text_pair:t,add_special_tokens:n,return_token_type_ids:r}).input_ids}batch_decode(e,t={}){return e instanceof o.Tensor&&(e=e.tolist()),e.map((e=>this.decode(e,t)))}decode(e,t={}){if(e instanceof o.Tensor&&(e=f(e)),!Array.isArray(e)||0===e.length||!(0,i.isIntegralNumber)(e[0]))throw Error("token_ids must be a non-empty array of integers.");return this.decode_single(e,t)}decode_single(e,{skip_special_tokens:t=!1,clean_up_tokenization_spaces:n=null}){let r=this.model.convert_ids_to_tokens(e);t&&(r=r.filter((e=>!this.special_tokens.includes(e))));let i=this.decoder?this.decoder(r):r.join(" ");return this.decoder&&this.decoder.end_of_word_suffix&&(i=i.replaceAll(this.decoder.end_of_word_suffix," "),t&&(i=i.trim())),(n??this.clean_up_tokenization_spaces)&&(i=g(i)),i}get_chat_template({chat_template:e=null,tools:t=null}={}){if(this.chat_template&&"object"==typeof this.chat_template){const n=this.chat_template;if(null!==e&&Object.hasOwn(n,e))e=n[e];else if(null===e)if(null!==t&&"tool_use"in n)e=n.tool_use;else{if(!("default"in n))throw Error(`This model has multiple chat templates with no default specified! Please either pass a chat template or the name of the template you wish to use to the 'chat_template' argument. Available template names are ${Object.keys(n).sort()}.`);e=n.default}}else if(null===e){if(!this.chat_template)throw Error("Cannot use apply_chat_template() because tokenizer.chat_template is not set and no template argument was passed! For information about writing templates and setting the tokenizer.chat_template attribute, please see the documentation at https://huggingface.co/docs/transformers/main/en/chat_templating");e=this.chat_template}return e}apply_chat_template(e,{tools:t=null,documents:n=null,chat_template:r=null,add_generation_prompt:i=!1,tokenize:a=!0,padding:s=!1,truncation:o=!1,max_length:l=null,return_tensor:d=!0,return_dict:c=!1,tokenizer_kwargs:p={},...h}={}){if("string"!=typeof(r=this.get_chat_template({chat_template:r,tools:t})))throw Error("chat_template must be a string, but got "+typeof r);let m=this._compiled_template_cache.get(r);void 0===m&&(m=new u.Template(r),this._compiled_template_cache.set(r,m));const f=Object.create(null);for(const e of _e){const t=this.getToken(e);t&&(f[e]=t)}const g=m.render({messages:e,add_generation_prompt:i,tools:t,documents:n,...f,...h});if(a){const e=this._call(g,{add_special_tokens:!1,padding:s,truncation:o,max_length:l,return_tensor:d,...p});return c?e:e.input_ids}return g}}class ve extends be{return_token_type_ids=!0}class xe extends be{return_token_type_ids=!0}class Me extends be{return_token_type_ids=!0}class Te extends be{return_token_type_ids=!0}class ke extends be{return_token_type_ids=!0}class $e extends be{return_token_type_ids=!0}class Se extends be{return_token_type_ids=!0}class Ce extends be{return_token_type_ids=!0}class Pe extends be{return_token_type_ids=!0}class Ee extends be{}class Fe extends be{}class Ae extends be{return_token_type_ids=!0;constructor(e,t){super(e,t),console.warn('WARNING: `XLMTokenizer` is not yet supported by Hugging Face\'s "fast" tokenizers library. Therefore, you may experience slightly inaccurate results.')}}class Ie extends be{return_token_type_ids=!0}class ze extends be{}class Oe extends be{}class Be extends be{}class Le extends be{constructor(e,t){super(e,t),this.languageRegex=/^[a-z]{2}_[A-Z]{2}$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))),this.lang_to_token=e=>e}_build_translation_inputs(e,t,n){return Je(this,e,t,n)}}class De extends Le{}class Re extends be{}class Ne extends be{constructor(e,t){const n=".,!?…。，、।۔،",r=e.pre_tokenizer?.pretokenizers[0]?.pattern;r&&r.Regex===` ?[^(\\s|[${n}])]+`&&(r.Regex=` ?[^\\s${n}]+`),super(e,t)}}const Ve="▁";class je extends be{padding_side="left";constructor(e,t){super(e,t),this.legacy=t.legacy??!0,this.legacy||(this.normalizer=null,this.pre_tokenizer=new de({replacement:Ve,add_prefix_space:!0,prepend_scheme:"first"}))}_encode_text(e){if(null===e)return null;if(this.legacy||0===e.length)return super._encode_text(e);let t=super._encode_text(Ve+e.replaceAll(Ve," "));return t.length>1&&t[0]===Ve&&this.special_tokens.includes(t[1])&&(t=t.slice(1)),t}}class Ge extends be{}class qe extends be{}class Ue extends be{}class We extends be{}class He extends be{}class Xe extends be{}class Ke extends be{}class Qe extends be{}class Ye extends be{}function Je(e,t,n,r){if(!("language_codes"in e)||!Array.isArray(e.language_codes))throw new Error("Tokenizer must have `language_codes` attribute set and it should be an array of language ids.");if(!("languageRegex"in e&&e.languageRegex instanceof RegExp))throw new Error("Tokenizer must have `languageRegex` attribute set and it should be a regular expression.");if(!("lang_to_token"in e)||"function"!=typeof e.lang_to_token)throw new Error("Tokenizer must have `lang_to_token` attribute set and it should be a function.");const i=r.src_lang,a=r.tgt_lang;if(!e.language_codes.includes(a))throw new Error(`Target language code "${a}" is not valid. Must be one of: {${e.language_codes.join(", ")}}`);if(void 0!==i){if(!e.language_codes.includes(i))throw new Error(`Source language code "${i}" is not valid. Must be one of: {${e.language_codes.join(", ")}}`);for(const t of e.post_processor.config.single)if("SpecialToken"in t&&e.languageRegex.test(t.SpecialToken.id)){t.SpecialToken.id=e.lang_to_token(i);break}}return r.forced_bos_token_id=e.model.convert_tokens_to_ids([e.lang_to_token(a)])[0],e._call(t,n)}class Ze extends be{constructor(e,t){super(e,t),this.languageRegex=/^[a-z]{3}_[A-Z][a-z]{3}$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))),this.lang_to_token=e=>e}_build_translation_inputs(e,t,n){return Je(this,e,t,n)}}class et extends be{constructor(e,t){super(e,t),this.languageRegex=/^__[a-z]{2,3}__$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))).map((e=>e.slice(2,-2))),this.lang_to_token=e=>`__${e}__`}_build_translation_inputs(e,t,n){return Je(this,e,t,n)}}class tt extends be{get timestamp_begin(){return this.model.convert_tokens_to_ids(["<|notimestamps|>"])[0]+1}_decode_asr(e,{return_timestamps:t=!1,return_language:n=!1,time_precision:r=null,force_full_sequences:i=!0}={}){if(null===r)throw Error("Must specify time_precision");let a=null;const o="word"===t;function l(){return{language:a,timestamp:[null,null],text:""}}const u=[];let c=l(),p=0;const h=this.timestamp_begin;let m=[],f=[],g=!1,_=null;const w=new Set(this.all_special_ids);for(const n of e){const e=n.tokens,i=o?n.token_timestamps:null;let y=null,v=h;if("stride"in n){const[t,i,a]=n.stride;if(p-=i,_=t-a,i&&(v=i/r+h),a)for(let t=e.length-1;t>=0;--t){const n=Number(e[t]);if(n>=h){if(null!==y&&(n-h)*r<_)break;y=n}}}let x=[],M=[];for(let n=0;n<e.length;++n){const _=Number(e[n]);if(w.has(_)){const e=this.decode([_]),n=d.WHISPER_LANGUAGE_MAPPING.get(e.slice(2,-2));if(void 0!==n){if(null!==a&&n!==a&&!t){m.push(x);const e=this.findLongestCommonSequence(m)[0],t=this.decode(e);c.text=t,u.push(c),m=[],x=[],c=l()}a=c.language=n}}else if(_>=h){const e=(_-h)*r+p,t=(0,s.round)(e,2);if(null!==y&&_>=y)g=!0;else if(g||m.length>0&&_<v)g=!1;else if(null===c.timestamp[0])c.timestamp[0]=t;else if(t===c.timestamp[0]);else{c.timestamp[1]=t,m.push(x),o&&f.push(M);const[e,n]=this.findLongestCommonSequence(m,f),r=this.decode(e);c.text=r,o&&(c.words=this.collateWordTimestamps(e,n,a)),u.push(c),m=[],x=[],f=[],M=[],c=l()}}else if(x.push(_),o){let e,t=(0,s.round)(i[n]+p,2);if(n+1<i.length){e=(0,s.round)(i[n+1]+p,2);const a=this.decode([_]);b.test(a)&&(e=(0,s.round)(Math.min(t+r,e),2))}else e=null;M.push([t,e])}}if("stride"in n){const[e,t,r]=n.stride;p+=e-r}x.length>0?(m.push(x),o&&f.push(M)):m.every((e=>0===e.length))&&(c=l(),m=[],x=[],f=[],M=[])}if(m.length>0){if(i&&t)throw new Error("Whisper did not predict an ending timestamp, which can happen if audio is cut off in the middle of a word. Also make sure WhisperTimeStampLogitsProcessor was used during generation.");const[e,n]=this.findLongestCommonSequence(m,f),r=this.decode(e);c.text=r,o&&(c.words=this.collateWordTimestamps(e,n,a)),u.push(c)}let y=Object.create(null);const v=u.map((e=>e.text)).join("");if(t||n){for(let e=0;e<u.length;++e){const r=u[e];t||delete r.timestamp,n||delete r.language}if(o){const e=[];for(const t of u)for(const n of t.words)e.push(n);y={chunks:e}}else y={chunks:u}}return[v,y]}findLongestCommonSequence(e,t=null){let n=e[0],r=n.length,i=[];const a=Array.isArray(t)&&t.length>0;let s=a?[]:null,o=a?t[0]:null;for(let l=1;l<e.length;++l){const u=e[l];let d=0,c=[r,r,0,0];const p=u.length;for(let e=1;e<r+p;++e){const i=Math.max(0,r-e),s=Math.min(r,r+p-e),h=n.slice(i,s),m=Math.max(0,e-r),f=Math.min(p,e),g=u.slice(m,f);if(h.length!==g.length)throw new Error("There is a bug within whisper `decode_asr` function, please report it. Dropping to prevent bad inference.");let _;_=a?h.filter(((e,n)=>e===g[n]&&o[i+n]<=t[l][m+n])).length:h.filter(((e,t)=>e===g[t])).length;const w=_/e+e/1e4;_>1&&w>d&&(d=w,c=[i,s,m,f])}const[h,m,f,g]=c,_=Math.floor((m+h)/2),w=Math.floor((g+f)/2);i.push(...n.slice(0,_)),n=u.slice(w),r=n.length,a&&(s.push(...o.slice(0,_)),o=t[l].slice(w))}return i.push(...n),a?(s.push(...o),[i,s]):[i,[]]}collateWordTimestamps(e,t,n){const[r,i,a]=this.combineTokensIntoWords(e,n),s=[];for(let e=0;e<r.length;++e){const n=a[e];s.push({text:r[e],timestamp:[t[n.at(0)][0],t[n.at(-1)][1]]})}return s}combineTokensIntoWords(e,t,n="\"'“¡¿([{-",r="\"'.。,，!！?？:：”)]}、"){let i,a,s;return["chinese","japanese","thai","lao","myanmar"].includes(t=t??"english")?[i,a,s]=this.splitTokensOnUnicode(e):[i,a,s]=this.splitTokensOnSpaces(e),this.mergePunctuations(i,a,s,n,r)}decode(e,t){let n;return t?.decode_with_timestamps?(e instanceof o.Tensor&&(e=f(e)),n=this.decodeWithTimestamps(e,t)):n=super.decode(e,t),n}decodeWithTimestamps(e,t){const n=t?.time_precision??.02,r=Array.from(this.all_special_ids).at(-1)+1;let i=[[]];for(let t of e)if(t=Number(t),t>=r){const e=((t-r)*n).toFixed(2);i.push(`<|${e}|>`),i.push([])}else i[i.length-1].push(t);return i=i.map((e=>"string"==typeof e?e:super.decode(e,t))),i.join("")}splitTokensOnUnicode(e){const t=this.decode(e,{decode_with_timestamps:!0}),n=[],r=[],i=[];let a=[],s=[],o=0;for(let l=0;l<e.length;++l){const u=e[l];a.push(u),s.push(l);const d=this.decode(a,{decode_with_timestamps:!0});d.includes("�")&&"�"!==t[o+d.indexOf("�")]||(n.push(d),r.push(a),i.push(s),a=[],s=[],o+=d.length)}return[n,r,i]}splitTokensOnSpaces(e){const[t,n,r]=this.splitTokensOnUnicode(e),i=[],a=[],s=[],o=new RegExp(`^[${y}]$`,"gu");for(let e=0;e<t.length;++e){const l=t[e],u=n[e],d=r[e],c=u[0]>=this.model.tokens_to_ids.get("<|endoftext|>"),p=l.startsWith(" "),h=l.trim(),m=o.test(h);if(c||p||m||0===i.length)i.push(l),a.push(u),s.push(d);else{const e=i.length-1;i[e]+=l,a[e].push(...u),s[e].push(...d)}}return[i,a,s]}mergePunctuations(e,t,n,r,a){const s=structuredClone(e),o=structuredClone(t),l=structuredClone(n);let u=s.length-2,d=s.length-1;for(;u>=0;)s[u].startsWith(" ")&&r.includes(s[u].trim())?(s[d]=s[u]+s[d],o[d]=(0,i.mergeArrays)(o[u],o[d]),l[d]=(0,i.mergeArrays)(l[u],l[d]),s[u]="",o[u]=[],l[u]=[]):d=u,--u;for(u=0,d=1;d<s.length;)!s[u].endsWith(" ")&&a.includes(s[d])?(s[u]+=s[d],o[u]=(0,i.mergeArrays)(o[u],o[d]),l[u]=(0,i.mergeArrays)(l[u],l[d]),s[d]="",o[d]=[],l[d]=[]):u=d,++d;return[s.filter((e=>e)),o.filter((e=>e.length>0)),l.filter((e=>e.length>0))]}get_decoder_prompt_ids({language:e=null,task:t=null,no_timestamps:n=!0}={}){const r=[];if(e){const t=(0,d.whisper_language_to_code)(e),n=this.model.tokens_to_ids.get(`<|${t}|>`);if(void 0===n)throw new Error(`Unable to find language "${t}" in model vocabulary. Please report this issue at ${c.GITHUB_ISSUE_URL}.`);r.push(n)}else r.push(null);if(t){if("transcribe"!==(t=t.toLowerCase())&&"translate"!==t)throw new Error(`Task "${t}" is not supported. Must be one of: ["transcribe", "translate"]`);const e=this.model.tokens_to_ids.get(`<|${t}|>`);if(void 0===e)throw new Error(`Unable to find task "${t}" in model vocabulary. Please report this issue at ${c.GITHUB_ISSUE_URL}.`);r.push(e)}else r.push(null);if(n){const e=this.model.tokens_to_ids.get("<|notimestamps|>");if(void 0===e)throw new Error(`Unable to find "<|notimestamps|>" in model vocabulary. Please report this issue at ${c.GITHUB_ISSUE_URL}.`);r.push(e)}return r.map(((e,t)=>[t+1,e])).filter((e=>null!==e[1]))}}class nt extends be{}class rt extends be{}class it extends be{}class at extends be{constructor(e,t){super(e,t),this.languageRegex=/^(>>\w+<<)\s*/g,this.supported_language_codes=this.model.vocab.filter((e=>this.languageRegex.test(e))),console.warn('WARNING: `MarianTokenizer` is not yet supported by Hugging Face\'s "fast" tokenizers library. Therefore, you may experience slightly inaccurate results.')}_encode_text(e){if(null===e)return null;const[t,...n]=e.trim().split(this.languageRegex);if(0===n.length)return super._encode_text(t);if(2===n.length){const[e,t]=n;return this.supported_language_codes.includes(e)||console.warn(`Unsupported language code "${e}" detected, which may lead to unexpected behavior. Should be one of: ${JSON.stringify(this.supported_language_codes)}`),(0,i.mergeArrays)([e],super._encode_text(t))}}}class st extends be{}class ot extends be{}class lt extends be{}class ut extends be{}class dt extends be{}class ct extends be{constructor(e,t){super(e,t),this.decoder=new ue({})}}class pt extends be{}class ht{static TOKENIZER_CLASS_MAPPING={T5Tokenizer:ze,DistilBertTokenizer:Ee,CamembertTokenizer:Fe,DebertaTokenizer:ke,DebertaV2Tokenizer:$e,BertTokenizer:ve,HerbertTokenizer:Se,ConvBertTokenizer:Ce,RoFormerTokenizer:Pe,XLMTokenizer:Ae,ElectraTokenizer:Ie,MobileBertTokenizer:Me,SqueezeBertTokenizer:Te,AlbertTokenizer:xe,GPT2Tokenizer:Oe,BartTokenizer:Be,MBartTokenizer:Le,MBart50Tokenizer:De,RobertaTokenizer:Re,WhisperTokenizer:tt,CodeGenTokenizer:nt,CLIPTokenizer:rt,SiglipTokenizer:it,MarianTokenizer:at,BloomTokenizer:Ne,NllbTokenizer:Ze,M2M100Tokenizer:et,LlamaTokenizer:je,CodeLlamaTokenizer:Ge,XLMRobertaTokenizer:qe,MPNetTokenizer:Ue,FalconTokenizer:We,GPTNeoXTokenizer:He,EsmTokenizer:Xe,Wav2Vec2CTCTokenizer:st,BlenderbotTokenizer:ot,BlenderbotSmallTokenizer:lt,SpeechT5Tokenizer:ut,NougatTokenizer:dt,VitsTokenizer:ct,Qwen2Tokenizer:Ke,GemmaTokenizer:Qe,Grok1Tokenizer:Ye,CohereTokenizer:pt,PreTrainedTokenizer:be};static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:r=null,local_files_only:i=!1,revision:a="main",legacy:s=null}={}){const[o,l]=await p(e,{progress_callback:t,config:n,cache_dir:r,local_files_only:i,revision:a,legacy:s}),u=l.tokenizer_class?.replace(/Fast$/,"")??"PreTrainedTokenizer";let d=this.TOKENIZER_CLASS_MAPPING[u];return d||(console.warn(`Unknown tokenizer class "${u}", attempting to construct from base class.`),d=be),new d(o,l)}}},"./src/utils/audio.js":
 /*!****************************!*\
   !*** ./src/utils/audio.js ***!
   \****************************/(e,t,n)=>{n.r(t),n.d(t,{hamming:()=>d,hanning:()=>u,mel_filter_bank:()=>f,read_audio:()=>o,spectrogram:()=>_,window_function:()=>w});var r=n(/*! ./hub.js */"./src/utils/hub.js"),i=n(/*! ./maths.js */"./src/utils/maths.js"),a=n(/*! ./core.js */"./src/utils/core.js"),s=n(/*! ./tensor.js */"./src/utils/tensor.js");async function o(e,t){if("undefined"==typeof AudioContext)throw Error("Unable to load audio from path/URL since `AudioContext` is not available in your environment. Instead, audio data should be passed directly to the pipeline/processor. For more information and some example code, see https://huggingface.co/docs/transformers.js/guides/node-audio-processing.");const n=await(await(0,r.getFile)(e)).arrayBuffer(),i=new AudioContext({sampleRate:t});void 0===t&&console.warn(`No sampling rate provided, using default of ${i.sampleRate}Hz.`);const a=await i.decodeAudioData(n);let s;if(2===a.numberOfChannels){const e=Math.sqrt(2),t=a.getChannelData(0),n=a.getChannelData(1);s=new Float32Array(t.length);for(let r=0;r<a.length;++r)s[r]=e*(t[r]+n[r])/2}else s=a.getChannelData(0);return s}function l(e,t){if(e<1)return new Float64Array;if(1===e)return new Float64Array([1]);const n=1-t,r=2*Math.PI/(e-1),i=new Float64Array(e);for(let a=0;a<e;++a)i[a]=t-n*Math.cos(a*r);return i}function u(e){return l(e,.5)}function d(e){return l(e,.54)}const c={htk:e=>2595*Math.log10(1+e/700),kaldi:e=>1127*Math.log(1+e/700),slaney:(e,t=1e3,n=15,r=27/Math.log(6.4))=>e>=t?n+Math.log(e/t)*r:3*e/200};function p(e,t="htk"){const n=c[t];if(!n)throw new Error('mel_scale should be one of "htk", "slaney" or "kaldi".');return"number"==typeof e?n(e):e.map((e=>n(e)))}const h={htk:e=>700*(10**(e/2595)-1),kaldi:e=>700*(Math.exp(e/1127)-1),slaney:(e,t=1e3,n=15,r=Math.log(6.4)/27)=>e>=n?t*Math.exp(r*(e-n)):200*e/3};function m(e,t,n){const r=(t-e)/(n-1);return Float64Array.from({length:n},((t,n)=>e+r*n))}function f(e,t,n,r,i,a=null,s="htk",o=!1){if(null!==a&&"slaney"!==a)throw new Error('norm must be one of null or "slaney"');const l=m(p(n,s),p(r,s),t+2);let u,d=function(e,t="htk"){const n=h[t];if(!n)throw new Error('mel_scale should be one of "htk", "slaney" or "kaldi".');return"number"==typeof e?n(e):e.map((e=>n(e)))}(l,s);if(o){const t=i/(2*e);u=p(Float64Array.from({length:e},((e,n)=>n*t)),s),d=l}else u=m(0,Math.floor(i/2),e);const c=function(e,t){const n=Float64Array.from({length:t.length-1},((e,n)=>t[n+1]-t[n])),r=Array.from({length:e.length},(()=>new Array(t.length)));for(let n=0;n<e.length;++n){const i=r[n];for(let r=0;r<t.length;++r)i[r]=t[r]-e[n]}const i=t.length-2,a=Array.from({length:i},(()=>new Array(e.length)));for(let t=0;t<e.length;++t){const e=r[t];for(let r=0;r<i;++r){const i=-e[r]/n[r],s=e[r+2]/n[r+1];a[r][t]=Math.max(0,Math.min(i,s))}}return a}(u,d);if(null!==a&&"slaney"===a)for(let n=0;n<t;++n){const t=c[n],r=2/(d[n+2]-d[n]);for(let n=0;n<e;++n)t[n]*=r}return c}function g(e,t,n,r,a){if(n<=0)throw new Error("reference must be greater than zero");if(r<=0)throw new Error("min_value must be greater than zero");n=Math.max(r,n);const s=Math.log10(n);for(let n=0;n<e.length;++n)e[n]=t*Math.log10(Math.max(r,e[n])-s);if(null!==a){if(a<=0)throw new Error("db_range must be greater than zero");const t=(0,i.max)(e)[0]-a;for(let n=0;n<e.length;++n)e[n]=Math.max(e[n],t)}return e}async function _(e,t,n,r,{fft_length:o=null,power:l=1,center:u=!0,pad_mode:d="reflect",onesided:c=!0,preemphasis:p=null,mel_filters:h=null,mel_floor:m=1e-10,log_mel:f=null,reference:_=1,min_value:w=1e-10,db_range:y=null,remove_dc_offset:b=null,min_num_frames:v=null,max_num_frames:x=null,do_pad:M=!0,transpose:T=!1}={}){const k=t.length;if(null===o&&(o=n),n>o)throw Error(`frame_length (${n}) may not be larger than fft_length (${o})`);if(k!==n)throw new Error(`Length of the window (${k}) must equal frame_length (${n})`);if(r<=0)throw new Error("hop_length must be greater than zero");if(null===l&&null!==h)throw new Error("You have provided `mel_filters` but `power` is `None`. Mel spectrogram computation is not yet supported for complex-valued spectrogram. Specify `power` to fix this issue.");if(u){if("reflect"!==d)throw new Error(`pad_mode="${d}" not implemented yet.`);const t=Math.floor((o-1)/2)+1;e=function(e,t,n){const r=new e.constructor(e.length+t+n),i=e.length-1;for(let n=0;n<e.length;++n)r[t+n]=e[n];for(let n=1;n<=t;++n)r[t-n]=e[(0,a.calculateReflectOffset)(n,i)];for(let s=1;s<=n;++s)r[i+t+s]=e[(0,a.calculateReflectOffset)(i-s,i)];return r}(e,t,t)}let $=Math.floor(1+Math.floor((e.length-n)/r));null!==v&&$<v&&($=v);const S=c?Math.floor(o/2)+1:o;let C=$,P=$;null!==x&&(x>$?M&&(P=x):P=C=x);const E=new i.FFT(o),F=new Float64Array(o),A=new Float64Array(E.outputBufferSize),I=new Float32Array(S*P);for(let i=0;i<C;++i){const a=i*r,s=Math.min(e.length-a,n);s!==n&&F.fill(0,0,n);for(let t=0;t<s;++t)F[t]=e[a+t];if(b){let e=0;for(let t=0;t<s;++t)e+=F[t];const t=e/s;for(let e=0;e<s;++e)F[e]-=t}if(null!==p){for(let e=s-1;e>=1;--e)F[e]-=p*F[e-1];F[0]*=1-p}for(let e=0;e<t.length;++e)F[e]*=t[e];E.realTransform(A,F);for(let e=0;e<S;++e){const t=e<<1;I[e*P+i]=A[t]**2+A[t+1]**2}}if(null!==l&&2!==l){const e=2/l;for(let t=0;t<I.length;++t)I[t]**=e}const z=h.length;let O=await(0,s.matmul)(new s.Tensor("float32",h.flat(),[z,S]),new s.Tensor("float32",I,[S,P]));T&&(O=O.transpose(1,0));const B=O.data;for(let e=0;e<B.length;++e)B[e]=Math.max(m,B[e]);if(null!==l&&null!==f){const e=Math.min(B.length,C*z);switch(f){case"log":for(let t=0;t<e;++t)B[t]=Math.log(B[t]);break;case"log10":for(let t=0;t<e;++t)B[t]=Math.log10(B[t]);break;case"dB":if(1===l)!function(e,t=1,n=1e-5,r=null){g(e,20,t,n,r)}(B,_,w,y);else{if(2!==l)throw new Error(`Cannot use log_mel option '${f}' with power ${l}`);!function(e,t=1,n=1e-10,r=null){g(e,10,t,n,r)}(B,_,w,y)}break;default:throw new Error(`log_mel must be one of null, 'log', 'log10' or 'dB'. Got '${f}'`)}}return O}function w(e,t,{periodic:n=!0,frame_length:r=null,center:i=!0}={}){const a=n?e+1:e;let s;switch(t){case"boxcar":s=new Float64Array(a).fill(1);break;case"hann":case"hann_window":s=u(a);break;case"hamming":s=d(a);break;case"povey":s=u(a).map((e=>Math.pow(e,.85)));break;default:throw new Error(`Unknown window type ${t}.`)}if(n&&(s=s.subarray(0,e)),null===r)return s;if(e>r)throw new Error(`Length of the window (${e}) may not be larger than frame_length (${r})`);return s}},"./src/utils/constants.js":
@@ -199,10 +199,10 @@ var r,i,a,s,o,l,u,d,c,p,h,m,f,g,_,w,y,b,v,x,M,T,k,$,S,C,P,E,F,A,I,z,O,B=Object.d
   \********************************/(e,t,n)=>{n.r(t),n.d(t,{GITHUB_ISSUE_URL:()=>r});const r="https://github.com/xenova/transformers.js/issues/new/choose"},"./src/utils/core.js":
 /*!***************************!*\
   !*** ./src/utils/core.js ***!
-  \***************************/(e,t,n)=>{function r(e,t){e&&e(t)}function i(e){return Object.fromEntries(Object.entries(e).map((([e,t])=>[t,e])))}function a(e){return e.replace(/[.*+?^${}()|[\]\\]/g,"\\$&")}function s(e){return"TypedArray"===e?.prototype?.__proto__?.constructor?.name}function o(e){return Number.isInteger(e)||"bigint"==typeof e}function l(e){const t=[];let n=e;for(;Array.isArray(n);)t.push(n.length),n=n[0];return t}function u(e,t,n=void 0){const r=e[t];if(void 0!==r)return delete e[t],r;if(void 0===n)throw Error(`Key ${t} does not exist in object.`);return n}function d(...e){return Array.prototype.concat.apply([],e)}function c(...e){return e.reduce(((e,t)=>e.flatMap((e=>t.map((t=>[e,t]))))))}function p(e,t){return Math.abs((e+t)%(2*t)-t)}function h(e,t){return Object.assign({},...t.map((t=>{if(void 0!==e[t])return{[t]:e[t]}})))}n.r(t),n.d(t,{calculateDimensions:()=>l,calculateReflectOffset:()=>p,dispatchCallback:()=>r,escapeRegExp:()=>a,isIntegralNumber:()=>o,isTypedArray:()=>s,mergeArrays:()=>d,pick:()=>h,pop:()=>u,product:()=>c,reverseDictionary:()=>i})},"./src/utils/data-structures.js":
+  \***************************/(e,t,n)=>{function r(e,t){e&&e(t)}function i(e){return Object.fromEntries(Object.entries(e).map((([e,t])=>[t,e])))}function a(e){return e.replace(/[.*+?^${}()|[\]\\]/g,"\\$&")}function s(e){return"TypedArray"===e?.prototype?.__proto__?.constructor?.name}function o(e){return Number.isInteger(e)||"bigint"==typeof e}function l(e){const t=[];let n=e;for(;Array.isArray(n);)t.push(n.length),n=n[0];return t}function u(e,t,n=void 0){const r=e[t];if(void 0!==r)return delete e[t],r;if(void 0===n)throw Error(`Key ${t} does not exist in object.`);return n}function d(...e){return Array.prototype.concat.apply([],e)}function c(...e){return e.reduce(((e,t)=>e.flatMap((e=>t.map((t=>[e,t]))))))}function p(e,t){return Math.abs((e+t)%(2*t)-t)}function h(e,t){return Object.assign({},...t.map((t=>{if(void 0!==e[t])return{[t]:e[t]}})))}function m(e){let t=0;for(const n of e)++t;return t}n.r(t),n.d(t,{calculateDimensions:()=>l,calculateReflectOffset:()=>p,dispatchCallback:()=>r,escapeRegExp:()=>a,isIntegralNumber:()=>o,isTypedArray:()=>s,len:()=>m,mergeArrays:()=>d,pick:()=>h,pop:()=>u,product:()=>c,reverseDictionary:()=>i})},"./src/utils/data-structures.js":
 /*!**************************************!*\
   !*** ./src/utils/data-structures.js ***!
-  \**************************************/(e,t,n)=>{n.r(t),n.d(t,{CharTrie:()=>i,PriorityQueue:()=>r,TokenLattice:()=>s});class r{constructor(e=(e,t)=>e>t,t=1/0){this._heap=[],this._comparator=e,this._maxSize=t}get size(){return this._heap.length}isEmpty(){return 0===this.size}peek(){return this._heap[0]}push(...e){return this.extend(e)}extend(e){for(const t of e)if(this.size<this._maxSize)this._heap.push(t),this._siftUp();else{const e=this._smallest();this._comparator(t,this._heap[e])&&(this._heap[e]=t,this._siftUpFrom(e))}return this.size}pop(){const e=this.peek(),t=this.size-1;return t>0&&this._swap(0,t),this._heap.pop(),this._siftDown(),e}replace(e){const t=this.peek();return this._heap[0]=e,this._siftDown(),t}_parent(e){return(e+1>>>1)-1}_left(e){return 1+(e<<1)}_right(e){return e+1<<1}_greater(e,t){return this._comparator(this._heap[e],this._heap[t])}_swap(e,t){const n=this._heap[e];this._heap[e]=this._heap[t],this._heap[t]=n}_siftUp(){this._siftUpFrom(this.size-1)}_siftUpFrom(e){for(;e>0&&this._greater(e,this._parent(e));)this._swap(e,this._parent(e)),e=this._parent(e)}_siftDown(){let e=0;for(;this._left(e)<this.size&&this._greater(this._left(e),e)||this._right(e)<this.size&&this._greater(this._right(e),e);){const t=this._right(e)<this.size&&this._greater(this._right(e),this._left(e))?this._right(e):this._left(e);this._swap(e,t),e=t}}_smallest(){return 2**Math.floor(Math.log2(this.size))-1}}class i{constructor(){this.root=a.default()}extend(e){for(let t of e)this.push(t)}push(e){let t=this.root;for(let n of e){let e=t.children.get(n);void 0===e&&(e=a.default(),t.children.set(n,e)),t=e}t.isLeaf=!0}*commonPrefixSearch(e){let t=this.root,n="";for(let r=0;r<e.length&&void 0!==t;++r){const i=e[r];n+=i,t=t.children.get(i),void 0!==t&&t.isLeaf&&(yield n)}}}class a{constructor(e,t){this.isLeaf=e,this.children=t}static default(){return new a(!1,new Map)}}class s{constructor(e,t,n){this.sentence=e,this.len=e.length,this.bosTokenId=t,this.eosTokenId=n,this.nodes=[],this.beginNodes=Array.from({length:this.len+1},(()=>[])),this.endNodes=Array.from({length:this.len+1},(()=>[]));const r=new o(this.bosTokenId,0,0,0,0),i=new o(this.eosTokenId,1,this.len,0,0);this.nodes.push(r.clone()),this.nodes.push(i.clone()),this.beginNodes[this.len].push(i),this.endNodes[0].push(r)}insert(e,t,n,r){const i=this.nodes.length,a=new o(r,i,e,t,n);this.beginNodes[e].push(a),this.endNodes[e+t].push(a),this.nodes.push(a)}viterbi(){const e=this.len;let t=0;for(;t<=e;){if(0==this.beginNodes[t].length)return[];for(let e of this.beginNodes[t]){e.prev=null;let n=0,r=null;for(let i of this.endNodes[t]){const t=i.backtraceScore+e.score;(null===r||t>n)&&(r=i.clone(),n=t)}if(null===r)return[];e.prev=r,e.backtraceScore=n}++t}const n=[],r=this.beginNodes[e][0].prev;if(null===r)return[];let i=r.clone();for(;null!==i.prev;){n.push(i.clone());const e=i.clone();i=e.prev.clone()}return n.reverse(),n}piece(e){return this.sentence.slice(e.pos,e.pos+e.length)}tokens(){return this.viterbi().map((e=>this.piece(e)))}tokenIds(){return this.viterbi().map((e=>e.tokenId))}}class o{constructor(e,t,n,r,i){this.tokenId=e,this.nodeId=t,this.pos=n,this.length=r,this.score=i,this.prev=null,this.backtraceScore=0}clone(){const e=new o(this.tokenId,this.nodeId,this.pos,this.length,this.score);return e.prev=this.prev,e.backtraceScore=this.backtraceScore,e}}},"./src/utils/devices.js":
+  \**************************************/(e,t,n)=>{n.r(t),n.d(t,{CharTrie:()=>i,PriorityQueue:()=>r,TokenLattice:()=>s});class r{constructor(e=(e,t)=>e>t,t=1/0){this._heap=[],this._comparator=e,this._maxSize=t}get size(){return this._heap.length}isEmpty(){return 0===this.size}peek(){return this._heap[0]}push(...e){return this.extend(e)}extend(e){for(const t of e)if(this.size<this._maxSize)this._heap.push(t),this._siftUp();else{const e=this._smallest();this._comparator(t,this._heap[e])&&(this._heap[e]=t,this._siftUpFrom(e))}return this.size}pop(){const e=this.peek(),t=this.size-1;return t>0&&this._swap(0,t),this._heap.pop(),this._siftDown(),e}replace(e){const t=this.peek();return this._heap[0]=e,this._siftDown(),t}_parent(e){return(e+1>>>1)-1}_left(e){return 1+(e<<1)}_right(e){return e+1<<1}_greater(e,t){return this._comparator(this._heap[e],this._heap[t])}_swap(e,t){const n=this._heap[e];this._heap[e]=this._heap[t],this._heap[t]=n}_siftUp(){this._siftUpFrom(this.size-1)}_siftUpFrom(e){for(;e>0&&this._greater(e,this._parent(e));)this._swap(e,this._parent(e)),e=this._parent(e)}_siftDown(){let e=0;for(;this._left(e)<this.size&&this._greater(this._left(e),e)||this._right(e)<this.size&&this._greater(this._right(e),e);){const t=this._right(e)<this.size&&this._greater(this._right(e),this._left(e))?this._right(e):this._left(e);this._swap(e,t),e=t}}_smallest(){return 2**Math.floor(Math.log2(this.size))-1}}class i{constructor(){this.root=a.default()}extend(e){for(const t of e)this.push(t)}push(e){let t=this.root;for(const n of e){let e=t.children.get(n);void 0===e&&(e=a.default(),t.children.set(n,e)),t=e}t.isLeaf=!0}*commonPrefixSearch(e){let t=this.root;if(void 0===t)return;let n="";for(const r of e){if(n+=r,t=t.children.get(r),void 0===t)return;t.isLeaf&&(yield n)}}}class a{constructor(e,t){this.isLeaf=e,this.children=t}static default(){return new a(!1,new Map)}}class s{constructor(e,t,n){this.chars=Array.from(e),this.len=this.chars.length,this.bosTokenId=t,this.eosTokenId=n,this.nodes=[],this.beginNodes=Array.from({length:this.len+1},(()=>[])),this.endNodes=Array.from({length:this.len+1},(()=>[]));const r=new o(this.bosTokenId,0,0,0,0),i=new o(this.eosTokenId,1,this.len,0,0);this.nodes.push(r.clone()),this.nodes.push(i.clone()),this.beginNodes[this.len].push(i),this.endNodes[0].push(r)}insert(e,t,n,r){const i=this.nodes.length,a=new o(r,i,e,t,n);this.beginNodes[e].push(a),this.endNodes[e+t].push(a),this.nodes.push(a)}viterbi(){const e=this.len;let t=0;for(;t<=e;){if(0==this.beginNodes[t].length)return[];for(let e of this.beginNodes[t]){e.prev=null;let n=0,r=null;for(let i of this.endNodes[t]){const t=i.backtraceScore+e.score;(null===r||t>n)&&(r=i.clone(),n=t)}if(null===r)return[];e.prev=r,e.backtraceScore=n}++t}const n=[],r=this.beginNodes[e][0].prev;if(null===r)return[];let i=r.clone();for(;null!==i.prev;){n.push(i.clone());const e=i.clone();i=e.prev.clone()}return n.reverse(),n}piece(e){return this.chars.slice(e.pos,e.pos+e.length).join("")}tokens(){return this.viterbi().map((e=>this.piece(e)))}tokenIds(){return this.viterbi().map((e=>e.tokenId))}}class o{constructor(e,t,n,r,i){this.tokenId=e,this.nodeId=t,this.pos=n,this.length=r,this.score=i,this.prev=null,this.backtraceScore=0}clone(){const e=new o(this.tokenId,this.nodeId,this.pos,this.length,this.score);return e.prev=this.prev,e.backtraceScore=this.backtraceScore,e}}},"./src/utils/devices.js":
 /*!******************************!*\
   !*** ./src/utils/devices.js ***!
   \******************************/(e,t,n)=>{n.r(t),n.d(t,{DEVICE_TYPES:()=>r});const r=Object.freeze({auto:"auto",gpu:"gpu",cpu:"cpu",wasm:"wasm",webgpu:"webgpu",cuda:"cuda",dml:"dml",webnn:"webnn","webnn-npu":"webnn-npu","webnn-gpu":"webnn-gpu","webnn-cpu":"webnn-cpu"})},"./src/utils/dtypes.js":