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

package/dist/transformers.min.mjs

@@ -88,10 +88,10 @@ import*as e from"fs";import*as t from"onnxruntime-node";import*as r from"path";i
   \******************************/(e,t,r)=>{var n;r.r(t),r.d(t,{Tensor:()=>i.Tensor,createInferenceSession:()=>_,deviceToExecutionProviders:()=>h,isONNXProxy:()=>M,isONNXTensor:()=>f});var o=r(/*! ../env.js */"./src/env.js"),s=r(/*! onnxruntime-node */"onnxruntime-node"),a=r(/*! #onnxruntime-webgpu */"?cb4d"),i=r(/*! 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"}}),c=[];let d,u;if(o.apis.IS_NODE_ENV){switch(u=s.default??s,process.platform){case"win32":c.push("dml");break;case"linux":"x64"===process.arch&&c.push("cuda")}c.push("cpu"),d=["cpu"]}else u=n||(n=r.t(a,2)),o.apis.IS_WEBNN_AVAILABLE&&c.push("webnn-npu","webnn-gpu","webnn-cpu","webnn"),o.apis.IS_WEBGPU_AVAILABLE&&c.push("webgpu"),c.push("wasm"),d=["wasm"];const p=u.InferenceSession;function h(e=null){if(!e)return d;switch(e){case"auto":return c;case"gpu":return c.filter((e=>["webgpu","cuda","dml","webnn-gpu"].includes(e)))}if(c.includes(e))return[l[e]??e];throw new Error(`Unsupported device: "${e}". Should be one of: ${c.join(", ")}.`)}let m=null;async function _(e,t){m&&await m;const r=p.create(e,t);return m??=r,await r}function f(e){return e instanceof u.Tensor}const g=u?.env;function M(){return g?.wasm?.proxy}g?.wasm&&(g.wasm.wasmPaths=`https://cdn.jsdelivr.net/npm/@huggingface/transformers@${o.env.version}/dist/`,g.wasm.proxy=!1,"undefined"!=typeof crossOriginIsolated&&crossOriginIsolated||(g.wasm.numThreads=1)),g?.webgpu&&(g.webgpu.powerPreference="high-performance"),o.env.backends.onnx=g},"./src/configs.js":
 /*!************************!*\
   !*** ./src/configs.js ***!
-  \************************/(e,t,r)=>{r.r(t),r.d(t,{AutoConfig:()=>l,PretrainedConfig:()=>i,getKeyValueShapes:()=>a});var n=r(/*! ./utils/core.js */"./src/utils/core.js"),o=r(/*! ./utils/hub.js */"./src/utils/hub.js");function s(e){const t={};let r={};switch(e.model_type){case"llava":case"paligemma":case"florence2":r=s(e.text_config);break;case"moondream1":r=s(e.phi_config);break;case"musicgen":r=s(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 o=s(e.decoder),a="num_decoder_layers"in o,i=(0,n.pick)(e,["model_type","is_encoder_decoder"]);return a?(i.num_decoder_layers=o.num_decoder_layers,i.num_decoder_heads=o.num_decoder_heads,i.decoder_hidden_size=o.decoder_hidden_size,i.num_encoder_layers=o.num_encoder_layers,i.num_encoder_heads=o.num_encoder_heads,i.encoder_hidden_size=o.encoder_hidden_size):(i.num_layers=o.num_layers,i.num_heads=o.num_heads,i.hidden_size=o.hidden_size),i}const o={...r,...(0,n.pick)(e,["model_type","multi_query","is_encoder_decoder"])};for(const r in t)o[r]=e[t[r]];return o}function a(e,{prefix:t="past_key_values"}={}){const r={},n=e.normalized_config;if(n.is_encoder_decoder&&"num_encoder_heads"in n&&"num_decoder_heads"in n){const e=n.encoder_dim_kv??n.encoder_hidden_size/n.num_encoder_heads,o=n.decoder_dim_kv??n.decoder_hidden_size/n.num_decoder_heads,s=[1,n.num_encoder_heads,0,e],a=[1,n.num_decoder_heads,0,o];for(let e=0;e<n.num_decoder_layers;++e)r[`${t}.${e}.encoder.key`]=s,r[`${t}.${e}.encoder.value`]=s,r[`${t}.${e}.decoder.key`]=a,r[`${t}.${e}.decoder.value`]=a}else{const e=n.num_heads,o=n.num_layers,s=n.dim_kv??n.hidden_size/(n.num_attention_heads??e);if("falcon"===n.model_type){const n=[1*e,0,s];for(let e=0;e<o;++e)r[`${t}.${e}.key`]=n,r[`${t}.${e}.value`]=n}else if(n.multi_query){const n=[1*e,0,2*s];for(let e=0;e<o;++e)r[`${t}.${e}.key_value`]=n}else if("bloom"===n.model_type){const n=[1*e,s,0],a=[1*e,0,s];for(let e=0;e<o;++e)r[`${t}.${e}.key`]=n,r[`${t}.${e}.value`]=a}else if("openelm"===n.model_type)for(let n=0;n<o;++n){const o=[1,e[n],0,s];r[`${t}.${n}.key`]=o,r[`${t}.${n}.value`]=o}else{const n=[1,e,0,s];for(let e=0;e<o;++e)r[`${t}.${e}.key`]=n,r[`${t}.${e}.value`]=n}}return r}class i{max_position_embeddings;constructor(e){this.model_type=null,this.is_encoder_decoder=!1,Object.assign(this,e),this.normalized_config=s(this)}static async from_pretrained(e,{progress_callback:t=null,config:r=null,cache_dir:n=null,local_files_only:s=!1,revision:a="main"}={}){!r||r instanceof i||(r=new i(r));const l=r??await async function(e,t){return await(0,o.getModelJSON)(e,"config.json",!0,t)}(e,{progress_callback:t,config:r,cache_dir:n,local_files_only:s,revision:a});return new this(l)}}class l{static async from_pretrained(...e){return i.from_pretrained(...e)}}},"./src/env.js":
+  \************************/(e,t,r)=>{r.r(t),r.d(t,{AutoConfig:()=>l,PretrainedConfig:()=>i,getKeyValueShapes:()=>a});var n=r(/*! ./utils/core.js */"./src/utils/core.js"),o=r(/*! ./utils/hub.js */"./src/utils/hub.js");function s(e){const t={};let r={};switch(e.model_type){case"llava":case"paligemma":case"florence2":r=s(e.text_config);break;case"moondream1":r=s(e.phi_config);break;case"musicgen":r=s(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 o=s(e.decoder),a="num_decoder_layers"in o,i=(0,n.pick)(e,["model_type","is_encoder_decoder"]);return a?(i.num_decoder_layers=o.num_decoder_layers,i.num_decoder_heads=o.num_decoder_heads,i.decoder_hidden_size=o.decoder_hidden_size,i.num_encoder_layers=o.num_encoder_layers,i.num_encoder_heads=o.num_encoder_heads,i.encoder_hidden_size=o.encoder_hidden_size):(i.num_layers=o.num_layers,i.num_heads=o.num_heads,i.hidden_size=o.hidden_size),i}const o={...r,...(0,n.pick)(e,["model_type","multi_query","is_encoder_decoder"])};for(const r in t)o[r]=e[t[r]];return o}function a(e,{prefix:t="past_key_values"}={}){const r={},n=e.normalized_config;if(n.is_encoder_decoder&&"num_encoder_heads"in n&&"num_decoder_heads"in n){const e=n.encoder_dim_kv??n.encoder_hidden_size/n.num_encoder_heads,o=n.decoder_dim_kv??n.decoder_hidden_size/n.num_decoder_heads,s=[1,n.num_encoder_heads,0,e],a=[1,n.num_decoder_heads,0,o];for(let e=0;e<n.num_decoder_layers;++e)r[`${t}.${e}.encoder.key`]=s,r[`${t}.${e}.encoder.value`]=s,r[`${t}.${e}.decoder.key`]=a,r[`${t}.${e}.decoder.value`]=a}else{const e=n.num_heads,o=n.num_layers,s=n.dim_kv??n.hidden_size/(n.num_attention_heads??e);if("falcon"===n.model_type){const n=[1*e,0,s];for(let e=0;e<o;++e)r[`${t}.${e}.key`]=n,r[`${t}.${e}.value`]=n}else if(n.multi_query){const n=[1*e,0,2*s];for(let e=0;e<o;++e)r[`${t}.${e}.key_value`]=n}else if("bloom"===n.model_type){const n=[1*e,s,0],a=[1*e,0,s];for(let e=0;e<o;++e)r[`${t}.${e}.key`]=n,r[`${t}.${e}.value`]=a}else if("openelm"===n.model_type)for(let n=0;n<o;++n){const o=[1,e[n],0,s];r[`${t}.${n}.key`]=o,r[`${t}.${n}.value`]=o}else{const n=[1,e,0,s];for(let e=0;e<o;++e)r[`${t}.${e}.key`]=n,r[`${t}.${e}.value`]=n}}return r}class i{model_type=null;is_encoder_decoder=!1;max_position_embeddings;"transformers.js_config";constructor(e){Object.assign(this,e),this.normalized_config=s(this)}static async from_pretrained(e,{progress_callback:t=null,config:r=null,cache_dir:n=null,local_files_only:s=!1,revision:a="main"}={}){!r||r instanceof i||(r=new i(r));const l=r??await async function(e,t){return await(0,o.getModelJSON)(e,"config.json",!0,t)}(e,{progress_callback:t,config:r,cache_dir:n,local_files_only:s,revision:a});return new this(l)}}class l{static async from_pretrained(...e){return i.from_pretrained(...e)}}},"./src/env.js":
 /*!********************!*\
   !*** ./src/env.js ***!
-  \********************/(e,t,r)=>{r.r(t),r.d(t,{apis:()=>_,env:()=>T});var n=r(/*! fs */"fs"),o=r(/*! path */"path"),s=r(/*! url */"url");const a="undefined"!=typeof self,i=a&&"DedicatedWorkerGlobalScope"===self.constructor.name,l=a&&"caches"in self,c="undefined"!=typeof navigator&&"gpu"in navigator,d="undefined"!=typeof navigator&&"ml"in navigator,u="undefined"!=typeof process,p=u&&"node"===process?.release?.name,h=!b(n.default),m=!b(o.default),_=Object.freeze({IS_BROWSER_ENV:a,IS_WEBWORKER_ENV:i,IS_WEB_CACHE_AVAILABLE:l,IS_WEBGPU_AVAILABLE:c,IS_WEBNN_AVAILABLE:d,IS_PROCESS_AVAILABLE:u,IS_NODE_ENV:p,IS_FS_AVAILABLE:h,IS_PATH_AVAILABLE:m}),f=h&&m,g=f?o.default.dirname(o.default.dirname(s.default.fileURLToPath(import.meta.url))):"./",M=f?o.default.join(g,"/.cache/"):null,w="/models/",T={version:"3.0.0-alpha.15",backends:{onnx:{}},allowRemoteModels:!0,remoteHost:"https://huggingface.co/",remotePathTemplate:"{model}/resolve/{revision}/",allowLocalModels:!a,localModelPath:f?o.default.join(g,w):w,useFS:h,useBrowserCache:l,useFSCache:h,cacheDir:M,useCustomCache:!1,customCache:null};function b(e){return 0===Object.keys(e).length}},"./src/generation/configuration_utils.js":
+  \********************/(e,t,r)=>{r.r(t),r.d(t,{apis:()=>_,env:()=>T});var n=r(/*! fs */"fs"),o=r(/*! path */"path"),s=r(/*! url */"url");const a="undefined"!=typeof self,i=a&&"DedicatedWorkerGlobalScope"===self.constructor.name,l=a&&"caches"in self,c="undefined"!=typeof navigator&&"gpu"in navigator,d="undefined"!=typeof navigator&&"ml"in navigator,u="undefined"!=typeof process,p=u&&"node"===process?.release?.name,h=!b(n.default),m=!b(o.default),_=Object.freeze({IS_BROWSER_ENV:a,IS_WEBWORKER_ENV:i,IS_WEB_CACHE_AVAILABLE:l,IS_WEBGPU_AVAILABLE:c,IS_WEBNN_AVAILABLE:d,IS_PROCESS_AVAILABLE:u,IS_NODE_ENV:p,IS_FS_AVAILABLE:h,IS_PATH_AVAILABLE:m}),f=h&&m,g=f?o.default.dirname(o.default.dirname(s.default.fileURLToPath(import.meta.url))):"./",M=f?o.default.join(g,"/.cache/"):null,w="/models/",T={version:"3.0.0-alpha.16",backends:{onnx:{}},allowRemoteModels:!0,remoteHost:"https://huggingface.co/",remotePathTemplate:"{model}/resolve/{revision}/",allowLocalModels:!a,localModelPath:f?o.default.join(g,w):w,useFS:h,useBrowserCache:l,useFSCache:h,cacheDir:M,useCustomCache:!1,customCache:null};function b(e){return 0===Object.keys(e).length}},"./src/generation/configuration_utils.js":
 /*!***********************************************!*\
   !*** ./src/generation/configuration_utils.js ***!
   \***********************************************/(e,t,r)=>{r.r(t),r.d(t,{GenerationConfig:()=>o});var n=r(/*! ../utils/core.js */"./src/utils/core.js");class o{max_length=20;max_new_tokens=null;min_length=0;min_new_tokens=null;early_stopping=!1;max_time=null;do_sample=!1;num_beams=1;num_beam_groups=1;penalty_alpha=null;use_cache=!0;temperature=1;top_k=50;top_p=1;typical_p=1;epsilon_cutoff=0;eta_cutoff=0;diversity_penalty=0;repetition_penalty=1;encoder_repetition_penalty=1;length_penalty=1;no_repeat_ngram_size=0;bad_words_ids=null;force_words_ids=null;renormalize_logits=!1;constraints=null;forced_bos_token_id=null;forced_eos_token_id=null;remove_invalid_values=!1;exponential_decay_length_penalty=null;suppress_tokens=null;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,n.pick)(e,Object.getOwnPropertyNames(this)))}}},"./src/generation/logits_process.js":
@@ -109,7 +109,7 @@ import*as e from"fs";import*as t from"onnxruntime-node";import*as r from"path";i
   \*************************************/(e,t,r)=>{r.r(t),r.d(t,{BaseStreamer:()=>a,TextStreamer:()=>l,WhisperTextStreamer:()=>c});var n=r(/*! ../utils/core.js */"./src/utils/core.js"),o=r(/*! ../tokenizers.js */"./src/tokenizers.js"),s=r(/*! ../env.js */"./src/env.js");class a{put(e){throw Error("Not implemented")}end(){throw Error("Not implemented")}}const i=s.apis.IS_PROCESS_AVAILABLE?e=>process.stdout.write(e):e=>console.log(e);class l extends a{constructor(e,{skip_prompt:t=!1,callback_function:r=null,token_callback_function:n=null,decode_kwargs:o={},...s}={}){super(),this.tokenizer=e,this.skip_prompt=t,this.callback_function=r??i,this.token_callback_function=n,this.decode_kwargs={...o,...s},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,n.mergeArrays)(this.token_cache,t);const r=this.tokenizer.decode(this.token_cache,this.decode_kwargs);let s;r.endsWith("\n")?(s=r.slice(this.print_len),this.token_cache=[],this.print_len=0):r.length>0&&(0,o.is_chinese_char)(r.charCodeAt(r.length-1))?(s=r.slice(this.print_len),this.print_len+=s.length):(s=r.slice(this.print_len,r.lastIndexOf(" ")+1),this.print_len+=s.length),this.on_finalized_text(s,!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===i&&s.apis.IS_PROCESS_AVAILABLE&&this.callback_function?.("\n")}}class c extends l{constructor(e,{skip_prompt:t=!1,callback_function:r=null,token_callback_function:n=null,on_chunk_start:o=null,on_chunk_end:s=null,on_finalize:a=null,time_precision:i=.02,skip_special_tokens:l=!0,decode_kwargs:c={}}={}){super(e,{skip_prompt:t,callback_function:r,token_callback_function:n,decode_kwargs:{skip_special_tokens:l,...c}}),this.timestamp_begin=e.timestamp_begin,this.on_chunk_start=o,this.on_chunk_end=s,this.on_finalize=a,this.time_precision=i,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 r=Number(t[0])-this.timestamp_begin;if(r>=0){const t=r*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,r)=>{r.r(t),r.d(t,{ASTForAudioClassification:()=>rr,ASTModel:()=>tr,ASTPreTrainedModel:()=>er,AlbertForMaskedLM:()=>ut,AlbertForQuestionAnswering:()=>dt,AlbertForSequenceClassification:()=>ct,AlbertModel:()=>lt,AlbertPreTrainedModel:()=>it,AutoModel:()=>Ei,AutoModelForAudioClassification:()=>Yi,AutoModelForAudioFrameClassification:()=>Ki,AutoModelForCTC:()=>Hi,AutoModelForCausalLM:()=>Di,AutoModelForDepthEstimation:()=>rl,AutoModelForDocumentQuestionAnswering:()=>Zi,AutoModelForImageClassification:()=>Gi,AutoModelForImageFeatureExtraction:()=>ol,AutoModelForImageMatting:()=>el,AutoModelForImageSegmentation:()=>qi,AutoModelForImageToImage:()=>tl,AutoModelForMaskGeneration:()=>Qi,AutoModelForMaskedLM:()=>Vi,AutoModelForNormalEstimation:()=>nl,AutoModelForObjectDetection:()=>Ui,AutoModelForQuestionAnswering:()=>ji,AutoModelForSemanticSegmentation:()=>$i,AutoModelForSeq2SeqLM:()=>Ii,AutoModelForSequenceClassification:()=>Li,AutoModelForSpeechSeq2Seq:()=>Bi,AutoModelForTextToSpectrogram:()=>Ni,AutoModelForTextToWaveform:()=>Oi,AutoModelForTokenClassification:()=>zi,AutoModelForUniversalSegmentation:()=>Wi,AutoModelForVision2Seq:()=>Ri,AutoModelForXVector:()=>Ji,AutoModelForZeroShotObjectDetection:()=>Xi,BartForConditionalGeneration:()=>kt,BartForSequenceClassification:()=>yt,BartModel:()=>xt,BartPretrainedModel:()=>bt,BaseModelOutput:()=>q,BeitForImageClassification:()=>ao,BeitModel:()=>so,BeitPreTrainedModel:()=>oo,BertForMaskedLM:()=>U,BertForQuestionAnswering:()=>H,BertForSequenceClassification:()=>X,BertForTokenClassification:()=>Q,BertModel:()=>W,BertPreTrainedModel:()=>$,BlenderbotForConditionalGeneration:()=>Lt,BlenderbotModel:()=>Et,BlenderbotPreTrainedModel:()=>At,BlenderbotSmallForConditionalGeneration:()=>Bt,BlenderbotSmallModel:()=>It,BlenderbotSmallPreTrainedModel:()=>zt,BloomForCausalLM:()=>bn,BloomModel:()=>Tn,BloomPreTrainedModel:()=>wn,CLIPModel:()=>hr,CLIPPreTrainedModel:()=>pr,CLIPSegForImageSegmentation:()=>Pr,CLIPSegModel:()=>Fr,CLIPSegPreTrainedModel:()=>yr,CLIPTextModel:()=>mr,CLIPTextModelWithProjection:()=>_r,CLIPVisionModel:()=>fr,CLIPVisionModelWithProjection:()=>gr,CamembertForMaskedLM:()=>Me,CamembertForQuestionAnswering:()=>be,CamembertForSequenceClassification:()=>we,CamembertForTokenClassification:()=>Te,CamembertModel:()=>ge,CamembertPreTrainedModel:()=>fe,CausalLMOutput:()=>ul,CausalLMOutputWithPast:()=>pl,ChineseCLIPModel:()=>kr,ChineseCLIPPreTrainedModel:()=>xr,ClapAudioModelWithProjection:()=>Ta,ClapModel:()=>Ma,ClapPreTrainedModel:()=>ga,ClapTextModelWithProjection:()=>wa,CodeGenForCausalLM:()=>Xr,CodeGenModel:()=>Ur,CodeGenPreTrainedModel:()=>Wr,CohereForCausalLM:()=>Zr,CohereModel:()=>Kr,CoherePreTrainedModel:()=>Jr,ConvBertForMaskedLM:()=>ae,ConvBertForQuestionAnswering:()=>ce,ConvBertForSequenceClassification:()=>ie,ConvBertForTokenClassification:()=>le,ConvBertModel:()=>se,ConvBertPreTrainedModel:()=>oe,ConvNextForImageClassification:()=>rs,ConvNextModel:()=>ts,ConvNextPreTrainedModel:()=>es,ConvNextV2ForImageClassification:()=>ss,ConvNextV2Model:()=>os,ConvNextV2PreTrainedModel:()=>ns,DPTForDepthEstimation:()=>Vo,DPTModel:()=>Do,DPTPreTrainedModel:()=>Oo,DebertaForMaskedLM:()=>ye,DebertaForQuestionAnswering:()=>Ce,DebertaForSequenceClassification:()=>Fe,DebertaForTokenClassification:()=>Pe,DebertaModel:()=>ke,DebertaPreTrainedModel:()=>xe,DebertaV2ForMaskedLM:()=>Ae,DebertaV2ForQuestionAnswering:()=>ze,DebertaV2ForSequenceClassification:()=>Ee,DebertaV2ForTokenClassification:()=>Le,DebertaV2Model:()=>Se,DebertaV2PreTrainedModel:()=>ve,DecisionTransformerModel:()=>Ya,DecisionTransformerPreTrainedModel:()=>Ha,DeiTForImageClassification:()=>yo,DeiTModel:()=>ko,DeiTPreTrainedModel:()=>xo,DepthAnythingForDepthEstimation:()=>Ro,DepthAnythingPreTrainedModel:()=>jo,DetrForObjectDetection:()=>co,DetrForSegmentation:()=>uo,DetrModel:()=>lo,DetrObjectDetectionOutput:()=>po,DetrPreTrainedModel:()=>io,DetrSegmentationOutput:()=>ho,Dinov2ForImageClassification:()=>ls,Dinov2Model:()=>is,Dinov2PreTrainedModel:()=>as,DistilBertForMaskedLM:()=>Ve,DistilBertForQuestionAnswering:()=>De,DistilBertForSequenceClassification:()=>Ne,DistilBertForTokenClassification:()=>Oe,DistilBertModel:()=>Be,DistilBertPreTrainedModel:()=>Ie,DonutSwinModel:()=>Zo,DonutSwinPreTrainedModel:()=>Ko,EfficientNetForImageClassification:()=>La,EfficientNetModel:()=>Ea,EfficientNetPreTrainedModel:()=>Aa,ElectraForMaskedLM:()=>pe,ElectraForQuestionAnswering:()=>_e,ElectraForSequenceClassification:()=>he,ElectraForTokenClassification:()=>me,ElectraModel:()=>ue,ElectraPreTrainedModel:()=>de,EsmForMaskedLM:()=>Ge,EsmForSequenceClassification:()=>qe,EsmForTokenClassification:()=>$e,EsmModel:()=>Re,EsmPreTrainedModel:()=>je,FalconForCausalLM:()=>fa,FalconModel:()=>_a,FalconPreTrainedModel:()=>ma,FastViTForImageClassification:()=>qn,FastViTModel:()=>Gn,FastViTPreTrainedModel:()=>Rn,Florence2ForConditionalGeneration:()=>ur,Florence2PreTrainedModel:()=>dr,GLPNForDepthEstimation:()=>Jo,GLPNModel:()=>Yo,GLPNPreTrainedModel:()=>Ho,GPT2LMHeadModel:()=>Sr,GPT2Model:()=>vr,GPT2PreTrainedModel:()=>Cr,GPTBigCodeForCausalLM:()=>$r,GPTBigCodeModel:()=>qr,GPTBigCodePreTrainedModel:()=>Gr,GPTJForCausalLM:()=>Rr,GPTJModel:()=>jr,GPTJPreTrainedModel:()=>Vr,GPTNeoForCausalLM:()=>Br,GPTNeoModel:()=>Ir,GPTNeoPreTrainedModel:()=>zr,GPTNeoXForCausalLM:()=>Dr,GPTNeoXModel:()=>Or,GPTNeoXPreTrainedModel:()=>Nr,Gemma2ForCausalLM:()=>sn,Gemma2Model:()=>on,Gemma2PreTrainedModel:()=>nn,GemmaForCausalLM:()=>rn,GemmaModel:()=>tn,GemmaPreTrainedModel:()=>en,GroupViTModel:()=>jn,GroupViTPreTrainedModel:()=>Vn,HieraForImageClassification:()=>Co,HieraModel:()=>Po,HieraPreTrainedModel:()=>Fo,HubertForCTC:()=>Xs,HubertForSequenceClassification:()=>Qs,HubertModel:()=>Us,HubertPreTrainedModel:()=>Ws,ImageMattingOutput:()=>hl,JAISLMHeadModel:()=>Lr,JAISModel:()=>Er,JAISPreTrainedModel:()=>Ar,LlamaForCausalLM:()=>Yr,LlamaModel:()=>Hr,LlamaPreTrainedModel:()=>Qr,LlavaForConditionalGeneration:()=>lr,LlavaPreTrainedModel:()=>ir,LongT5ForConditionalGeneration:()=>gt,LongT5Model:()=>ft,LongT5PreTrainedModel:()=>_t,M2M100ForConditionalGeneration:()=>bs,M2M100Model:()=>Ts,M2M100PreTrainedModel:()=>ws,MBartForCausalLM:()=>St,MBartForConditionalGeneration:()=>Ct,MBartForSequenceClassification:()=>vt,MBartModel:()=>Pt,MBartPreTrainedModel:()=>Ft,MPNetForMaskedLM:()=>Ke,MPNetForQuestionAnswering:()=>tt,MPNetForSequenceClassification:()=>Ze,MPNetForTokenClassification:()=>et,MPNetModel:()=>Je,MPNetPreTrainedModel:()=>Ye,MT5ForConditionalGeneration:()=>Tt,MT5Model:()=>wt,MT5PreTrainedModel:()=>Mt,MarianMTModel:()=>Ms,MarianModel:()=>gs,MarianPreTrainedModel:()=>fs,MaskFormerForInstanceSegmentation:()=>Qo,MaskFormerModel:()=>Xo,MaskFormerPreTrainedModel:()=>Uo,MaskedLMOutput:()=>cl,MistralForCausalLM:()=>da,MistralModel:()=>ca,MistralPreTrainedModel:()=>la,MobileBertForMaskedLM:()=>Xe,MobileBertForQuestionAnswering:()=>He,MobileBertForSequenceClassification:()=>Qe,MobileBertModel:()=>Ue,MobileBertPreTrainedModel:()=>We,MobileNetV1ForImageClassification:()=>Va,MobileNetV1Model:()=>Da,MobileNetV1PreTrainedModel:()=>Oa,MobileNetV2ForImageClassification:()=>Ga,MobileNetV2Model:()=>Ra,MobileNetV2PreTrainedModel:()=>ja,MobileNetV3ForImageClassification:()=>Wa,MobileNetV3Model:()=>$a,MobileNetV3PreTrainedModel:()=>qa,MobileNetV4ForImageClassification:()=>Qa,MobileNetV4Model:()=>Xa,MobileNetV4PreTrainedModel:()=>Ua,MobileViTForImageClassification:()=>Qn,MobileViTModel:()=>Xn,MobileViTPreTrainedModel:()=>Un,MobileViTV2ForImageClassification:()=>Jn,MobileViTV2Model:()=>Yn,MobileViTV2PreTrainedModel:()=>Hn,ModelOutput:()=>G,Moondream1ForConditionalGeneration:()=>cr,MptForCausalLM:()=>yn,MptModel:()=>kn,MptPreTrainedModel:()=>xn,MusicgenForCausalLM:()=>Ba,MusicgenForConditionalGeneration:()=>Na,MusicgenModel:()=>Ia,MusicgenPreTrainedModel:()=>za,NomicBertModel:()=>J,NomicBertPreTrainedModel:()=>Y,OPTForCausalLM:()=>Cn,OPTModel:()=>Pn,OPTPreTrainedModel:()=>Fn,OpenELMForCausalLM:()=>cn,OpenELMModel:()=>ln,OpenELMPreTrainedModel:()=>an,OwlViTForObjectDetection:()=>eo,OwlViTModel:()=>Zn,OwlViTPreTrainedModel:()=>Kn,Owlv2ForObjectDetection:()=>no,Owlv2Model:()=>ro,Owlv2PreTrainedModel:()=>to,Phi3ForCausalLM:()=>Mn,Phi3Model:()=>gn,Phi3PreTrainedModel:()=>fn,PhiForCausalLM:()=>_n,PhiModel:()=>mn,PhiPreTrainedModel:()=>hn,PreTrainedModel:()=>R,PretrainedMixin:()=>Ja,PvtForImageClassification:()=>zn,PvtModel:()=>Ln,PvtPreTrainedModel:()=>En,PyAnnoteForAudioFrameClassification:()=>Ss,PyAnnoteModel:()=>vs,PyAnnotePreTrainedModel:()=>Cs,QuestionAnsweringModelOutput:()=>dl,Qwen2ForCausalLM:()=>pn,Qwen2Model:()=>un,Qwen2PreTrainedModel:()=>dn,RTDetrForObjectDetection:()=>fo,RTDetrModel:()=>_o,RTDetrObjectDetectionOutput:()=>go,RTDetrPreTrainedModel:()=>mo,ResNetForImageClassification:()=>Ao,ResNetModel:()=>So,ResNetPreTrainedModel:()=>vo,RoFormerForMaskedLM:()=>ee,RoFormerForQuestionAnswering:()=>ne,RoFormerForSequenceClassification:()=>te,RoFormerForTokenClassification:()=>re,RoFormerModel:()=>Z,RoFormerPreTrainedModel:()=>K,RobertaForMaskedLM:()=>Dt,RobertaForQuestionAnswering:()=>Rt,RobertaForSequenceClassification:()=>Vt,RobertaForTokenClassification:()=>jt,RobertaModel:()=>Ot,RobertaPreTrainedModel:()=>Nt,SamImageSegmentationOutput:()=>_s,SamModel:()=>ms,SamPreTrainedModel:()=>hs,SapiensForDepthEstimation:()=>$o,SapiensForNormalEstimation:()=>Wo,SapiensForSemanticSegmentation:()=>qo,SapiensPreTrainedModel:()=>Go,SegformerForImageClassification:()=>Fa,SegformerForSemanticSegmentation:()=>Pa,SegformerModel:()=>ya,SegformerPreTrainedModel:()=>ka,Seq2SeqLMOutput:()=>sl,SequenceClassifierOutput:()=>al,SiglipModel:()=>wr,SiglipPreTrainedModel:()=>Mr,SiglipTextModel:()=>Tr,SiglipVisionModel:()=>br,SpeechT5ForSpeechToText:()=>na,SpeechT5ForTextToSpeech:()=>oa,SpeechT5HifiGan:()=>sa,SpeechT5Model:()=>ra,SpeechT5PreTrainedModel:()=>ta,SqueezeBertForMaskedLM:()=>ot,SqueezeBertForQuestionAnswering:()=>at,SqueezeBertForSequenceClassification:()=>st,SqueezeBertModel:()=>nt,SqueezeBertPreTrainedModel:()=>rt,StableLmForCausalLM:()=>Sa,StableLmModel:()=>va,StableLmPreTrainedModel:()=>Ca,Starcoder2ForCausalLM:()=>ha,Starcoder2Model:()=>pa,Starcoder2PreTrainedModel:()=>ua,Swin2SRForImageSuperResolution:()=>No,Swin2SRModel:()=>Bo,Swin2SRPreTrainedModel:()=>Io,SwinForImageClassification:()=>zo,SwinModel:()=>Lo,SwinPreTrainedModel:()=>Eo,T5ForConditionalGeneration:()=>mt,T5Model:()=>ht,T5PreTrainedModel:()=>pt,TableTransformerForObjectDetection:()=>To,TableTransformerModel:()=>wo,TableTransformerObjectDetectionOutput:()=>bo,TableTransformerPreTrainedModel:()=>Mo,TokenClassifierOutput:()=>ll,TrOCRForCausalLM:()=>ia,TrOCRPreTrainedModel:()=>aa,UniSpeechForCTC:()=>Is,UniSpeechForSequenceClassification:()=>Bs,UniSpeechModel:()=>zs,UniSpeechPreTrainedModel:()=>Ls,UniSpeechSatForAudioFrameClassification:()=>js,UniSpeechSatForCTC:()=>Ds,UniSpeechSatForSequenceClassification:()=>Vs,UniSpeechSatModel:()=>Os,UniSpeechSatPreTrainedModel:()=>Ns,ViTForImageClassification:()=>An,ViTMAEModel:()=>Bn,ViTMAEPreTrainedModel:()=>In,ViTMSNForImageClassification:()=>Dn,ViTMSNModel:()=>On,ViTMSNPreTrainedModel:()=>Nn,ViTModel:()=>Sn,ViTPreTrainedModel:()=>vn,VisionEncoderDecoderModel:()=>ar,VitMatteForImageMatting:()=>Wn,VitMattePreTrainedModel:()=>$n,VitsModel:()=>xa,VitsModelOutput:()=>ml,VitsPreTrainedModel:()=>ba,Wav2Vec2BertForCTC:()=>qs,Wav2Vec2BertForSequenceClassification:()=>$s,Wav2Vec2BertModel:()=>Gs,Wav2Vec2BertPreTrainedModel:()=>Rs,Wav2Vec2ForAudioFrameClassification:()=>Ps,Wav2Vec2ForCTC:()=>ys,Wav2Vec2ForSequenceClassification:()=>Fs,Wav2Vec2Model:()=>ks,Wav2Vec2PreTrainedModel:()=>xs,WavLMForAudioFrameClassification:()=>ea,WavLMForCTC:()=>Js,WavLMForSequenceClassification:()=>Ks,WavLMForXVector:()=>Zs,WavLMModel:()=>Ys,WavLMPreTrainedModel:()=>Hs,WeSpeakerResNetModel:()=>Es,WeSpeakerResNetPreTrainedModel:()=>As,WhisperForConditionalGeneration:()=>sr,WhisperModel:()=>or,WhisperPreTrainedModel:()=>nr,XLMForQuestionAnswering:()=>Xt,XLMForSequenceClassification:()=>Wt,XLMForTokenClassification:()=>Ut,XLMModel:()=>qt,XLMPreTrainedModel:()=>Gt,XLMRobertaForMaskedLM:()=>Yt,XLMRobertaForQuestionAnswering:()=>Zt,XLMRobertaForSequenceClassification:()=>Jt,XLMRobertaForTokenClassification:()=>Kt,XLMRobertaModel:()=>Ht,XLMRobertaPreTrainedModel:()=>Qt,XLMWithLMHeadModel:()=>$t,XVectorOutput:()=>il,YolosForObjectDetection:()=>us,YolosModel:()=>ds,YolosObjectDetectionOutput:()=>ps,YolosPreTrainedModel:()=>cs});var n=r(/*! ./configs.js */"./src/configs.js"),o=r(/*! ./backends/onnx.js */"./src/backends/onnx.js"),s=r(/*! ./utils/dtypes.js */"./src/utils/dtypes.js"),a=r(/*! ./utils/generic.js */"./src/utils/generic.js"),i=r(/*! ./utils/core.js */"./src/utils/core.js"),l=r(/*! ./utils/hub.js */"./src/utils/hub.js"),c=r(/*! ./generation/logits_process.js */"./src/generation/logits_process.js"),d=r(/*! ./generation/configuration_utils.js */"./src/generation/configuration_utils.js"),u=r(/*! ./utils/tensor.js */"./src/utils/tensor.js"),p=r(/*! ./utils/maths.js */"./src/utils/maths.js"),h=r(/*! ./generation/stopping_criteria.js */"./src/generation/stopping_criteria.js"),m=r(/*! ./generation/logits_sampler.js */"./src/generation/logits_sampler.js"),_=r(/*! ./env.js */"./src/env.js"),f=r(/*! ./models/whisper/generation_whisper.js */"./src/models/whisper/generation_whisper.js"),g=r(/*! ./models/whisper/common_whisper.js */"./src/models/whisper/common_whisper.js");const M=0,w=1,T=2,b=3,x=4,k=5,y=6,F=7,P=new Map,C=new Map,v=new Map;async function S(e,t,r){return Object.fromEntries(await Promise.all(Object.keys(t).map((async a=>{const{buffer:i,session_options:c}=await async function(e,t,r){let a=r.device;a&&"string"!=typeof a&&(a.hasOwnProperty(t)?a=a[t]:(console.warn(`device not specified for "${t}". Using the default device.`),a=null));const i=a??(_.apis.IS_NODE_ENV?"cpu":"wasm"),c=(0,o.deviceToExecutionProviders)(i);let d=r.dtype;"string"!=typeof d&&(d&&d.hasOwnProperty(t)?d=d[t]:(d=s.DEFAULT_DEVICE_DTYPE_MAPPING[i]??s.DATA_TYPES.fp32,console.warn(`dtype not specified for "${t}". Using the default dtype (${d}) for this device (${i}).`)));const u=d;if(!s.DEFAULT_DTYPE_SUFFIX_MAPPING.hasOwnProperty(u))throw new Error(`Invalid dtype: ${u}. Should be one of: ${Object.keys(s.DATA_TYPES).join(", ")}`);if(u===s.DATA_TYPES.fp16&&"webgpu"===i&&!await(0,s.isWebGpuFp16Supported)())throw new Error(`The device (${i}) does not support fp16.`);const p=s.DEFAULT_DTYPE_SUFFIX_MAPPING[u],h=`${r.subfolder??""}/${t}${p}.onnx`,m={...r.session_options}??{};m.executionProviders??=c;const f=(0,l.getModelFile)(e,h,!0,r);let g=[];if(r.use_external_data_format&&(!0===r.use_external_data_format||"object"==typeof r.use_external_data_format&&r.use_external_data_format.hasOwnProperty(t)&&!0===r.use_external_data_format[t])){if(_.apis.IS_NODE_ENV)throw new Error("External data format is not yet supported in Node.js");const n=`${t}${p}.onnx_data`,o=`${r.subfolder??""}/${n}`;g.push(new Promise((async(t,s)=>{const a=await(0,l.getModelFile)(e,o,!0,r);t({path:n,data:a})})))}else void 0!==m.externalData&&(g=m.externalData.map((async t=>{if("string"==typeof t.data){const n=await(0,l.getModelFile)(e,t.data,!0,r);return{...t,data:n}}return t})));if(g.length>0&&(m.externalData=await Promise.all(g)),"webgpu"===i){const e=(0,n.getKeyValueShapes)(r.config,{prefix:"present"});if(Object.keys(e).length>0&&!(0,o.isONNXProxy)()){const t={};for(const r in e)t[r]="gpu-buffer";m.preferredOutputLocation=t}}return{buffer:await f,session_options:m}}(e,t[a],r);return[a,await(0,o.createInferenceSession)(i,c)]}))))}async function A(e,t){const r=function(e,t){const r=Object.create(null),n=[];for(const s of e.inputNames){const e=t[s];e instanceof u.Tensor?r[s]=(0,o.isONNXProxy)()?e.clone():e:n.push(s)}if(n.length>0)throw new Error(`An error occurred during model execution: "Missing the following inputs: ${n.join(", ")}.`);const s=Object.keys(t).length,a=e.inputNames.length;if(s>a){let r=Object.keys(t).filter((t=>!e.inputNames.includes(t)));console.warn(`WARNING: Too many inputs were provided (${s} > ${a}). The following inputs will be ignored: "${r.join(", ")}".`)}return r}(e,t);try{const t=Object.fromEntries(Object.entries(r).map((([e,t])=>[e,t.ort_tensor])));let n=await e.run(t);return n=E(n),n}catch(e){throw console.error(`An error occurred during model execution: "${e}".`),console.error("Inputs given to model:",r),e}}function E(e){for(let t in e)(0,o.isONNXTensor)(e[t])?e[t]=new u.Tensor(e[t]):"object"==typeof e[t]&&E(e[t]);return e}function L(e){if(e instanceof u.Tensor)return e;if(0===e.length)throw Error("items must be non-empty");if(Array.isArray(e[0])){if(e.some((t=>t.length!==e[0].length)))throw Error("Unable to create tensor, you should probably activate truncation and/or padding with 'padding=True' and/or 'truncation=True' to have batched tensors with the same length.");return new u.Tensor("int64",BigInt64Array.from(e.flat().map((e=>BigInt(e)))),[e.length,e[0].length])}return new u.Tensor("int64",BigInt64Array.from(e.map((e=>BigInt(e)))),[1,e.length])}function z(e){return new u.Tensor("bool",[e],[1])}async function I(e,t){let{encoder_outputs:r,input_ids:n,decoder_input_ids:o,...s}=t;if(!r){const n=(0,i.pick)(t,e.sessions.model.inputNames);r=(await B(e,n)).last_hidden_state}s.input_ids=o,s.encoder_hidden_states=r,e.sessions.decoder_model_merged.inputNames.includes("encoder_attention_mask")&&(s.encoder_attention_mask=t.attention_mask);return await N(e,s,!0)}async function B(e,t){const r=e.sessions.model,n=(0,i.pick)(t,r.inputNames);if(r.inputNames.includes("inputs_embeds")&&!n.inputs_embeds){if(!t.input_ids)throw new Error("Both `input_ids` and `inputs_embeds` are missing in the model inputs.");n.inputs_embeds=await e.encode_text({input_ids:t.input_ids})}return r.inputNames.includes("token_type_ids")&&!n.token_type_ids&&(n.token_type_ids=new u.Tensor("int64",new BigInt64Array(n.input_ids.data.length),n.input_ids.dims)),await A(r,n)}async function N(e,t,r=!1){const n=e.sessions[r?"decoder_model_merged":"model"],{past_key_values:o,...s}=t;n.inputNames.includes("use_cache_branch")&&(s.use_cache_branch=z(!!o)),n.inputNames.includes("position_ids")&&s.attention_mask&&!s.position_ids&&(s.position_ids=function(e,t=null){const{input_ids:r,inputs_embeds:n,attention_mask:o}=e,[s,a]=o.dims,i=new BigInt64Array(o.data.length);for(let e=0;e<s;++e){const t=e*a;let r=BigInt(0);for(let e=0;e<a;++e){const n=t+e;0n===o.data[n]?i[n]=BigInt(1):(i[n]=r,r+=o.data[n])}}let l=new u.Tensor("int64",i,o.dims);if(t){const e=-(r??n).dims.at(1);l=l.slice(null,[e,null])}return l}(s,o)),e.addPastKeyValues(s,o);const a=(0,i.pick)(s,n.inputNames);return await A(n,a)}async function O(e,{input_ids:t=null,attention_mask:r=null,pixel_values:n=null,position_ids:o=null,inputs_embeds:s=null,past_key_values:a=null,generation_config:i=null,logits_processor:l=null,...c}){if(!s)if(s=await e.encode_text({input_ids:t}),n&&1!==t.dims[1]){const o=await e.encode_image({pixel_values:n});({inputs_embeds:s,attention_mask:r}=e._merge_input_ids_with_image_features({image_features:o,inputs_embeds:s,input_ids:t,attention_mask:r}))}else if(a&&n&&1===t.dims[1]){const e=t.dims[1],n=Object.values(a)[0].dims.at(-2);r=(0,u.cat)([(0,u.ones)([t.dims[0],n]),r.slice(null,[r.dims[1]-e,r.dims[1]])],1)}return await N(e,{inputs_embeds:s,past_key_values:a,attention_mask:r,position_ids:o,generation_config:i,logits_processor:l},!0)}function D(e,t,r,n){if(r.past_key_values){const t=Object.values(r.past_key_values)[0].dims.at(-2),{input_ids:n,attention_mask:o}=r;if(o&&o.dims[1]>n.dims[1]);else if(t<n.dims[1])r.input_ids=n.slice(null,[t,null]);else if(null!=e.config.image_token_index&&n.data.some((t=>t==e.config.image_token_index))){const o=e.config.num_image_tokens;if(!o)throw new Error("`num_image_tokens` is missing in the model configuration.");const s=n.dims[1]-(t-o);r.input_ids=n.slice(null,[-s,null]),r.attention_mask=(0,u.ones)([1,t+s])}}return r}function V(e,t,r,n){return r.past_key_values&&(t=t.map((e=>[e.at(-1)]))),{...r,decoder_input_ids:L(t)}}function j(e,...t){return e.config.is_encoder_decoder?V(e,...t):D(e,...t)}class R extends a.Callable{main_input_name="input_ids";forward_params=["input_ids","attention_mask"];constructor(e,t){super(),this.config=e,this.sessions=t;const r=v.get(this.constructor),n=P.get(r);switch(this.can_generate=!1,this._forward=null,this._prepare_inputs_for_generation=null,n){case x:this.can_generate=!0,this._forward=N,this._prepare_inputs_for_generation=D;break;case T:case b:case F:this.can_generate=!0,this._forward=I,this._prepare_inputs_for_generation=V;break;case w:this._forward=I;break;case y:this.can_generate=!0,this._forward=O,this._prepare_inputs_for_generation=j;break;default:this._forward=B}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:r=null,cache_dir:o=null,local_files_only:s=!1,revision:a="main",model_file_name:i=null,subfolder:c="onnx",device:d=null,dtype:u=null,use_external_data_format:p=null,session_options:h={}}={}){let m={progress_callback:t,config:r,cache_dir:o,local_files_only:s,revision:a,model_file_name:i,subfolder:c,device:d,dtype:u,use_external_data_format:p,session_options:h};const _=v.get(this),f=P.get(_);let g;if(r=m.config=await n.AutoConfig.from_pretrained(e,m),f===x)g=await Promise.all([S(e,{model:m.model_file_name??"model"},m),(0,l.getModelJSON)(e,"generation_config.json",!1,m)]);else if(f===T||f===b)g=await Promise.all([S(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},m),(0,l.getModelJSON)(e,"generation_config.json",!1,m)]);else if(f===k)g=await Promise.all([S(e,{model:"vision_encoder",prompt_encoder_mask_decoder:"prompt_encoder_mask_decoder"},m)]);else if(f===w)g=await Promise.all([S(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},m)]);else if(f===y){const t={embed_tokens:"embed_tokens",vision_encoder:"vision_encoder",decoder_model_merged:"decoder_model_merged"};r.is_encoder_decoder&&(t.model="encoder_model"),g=await Promise.all([S(e,t,m),(0,l.getModelJSON)(e,"generation_config.json",!1,m)])}else f===F?g=await Promise.all([S(e,{model:"text_encoder",decoder_model_merged:"decoder_model_merged",encodec_decode:"encodec_decode"},m),(0,l.getModelJSON)(e,"generation_config.json",!1,m)]):(f!==M&&console.warn(`Model type for '${_??r?.model_type}' not found, assuming encoder-only architecture. Please report this at https://github.com/xenova/transformers.js/issues/new/choose.`),g=await Promise.all([S(e,{model:m.model_file_name??"model"},m)]));return new this(r,...g)}async _call(e){return await this.forward(e)}async forward(e){return await this._forward(this,e)}_get_logits_warper(e){const t=new c.LogitsProcessorList;return null!==e.temperature&&1!==e.temperature&&t.push(new c.TemperatureLogitsWarper(e.temperature)),null!==e.top_k&&0!==e.top_k&&t.push(new c.TopKLogitsWarper(e.top_k)),null!==e.top_p&&e.top_p<1&&t.push(new c.TopPLogitsWarper(e.top_p)),t}_get_logits_processor(e,t,r=null){const n=new c.LogitsProcessorList;if(null!==e.repetition_penalty&&1!==e.repetition_penalty&&n.push(new c.RepetitionPenaltyLogitsProcessor(e.repetition_penalty)),null!==e.no_repeat_ngram_size&&e.no_repeat_ngram_size>0&&n.push(new c.NoRepeatNGramLogitsProcessor(e.no_repeat_ngram_size)),null!==e.bad_words_ids&&n.push(new c.NoBadWordsLogitsProcessor(e.bad_words_ids,e.eos_token_id)),null!==e.min_length&&null!==e.eos_token_id&&e.min_length>0&&n.push(new c.MinLengthLogitsProcessor(e.min_length,e.eos_token_id)),null!==e.min_new_tokens&&null!==e.eos_token_id&&e.min_new_tokens>0&&n.push(new c.MinNewTokensLengthLogitsProcessor(t,e.min_new_tokens,e.eos_token_id)),null!==e.forced_bos_token_id&&n.push(new c.ForcedBOSTokenLogitsProcessor(e.forced_bos_token_id)),null!==e.forced_eos_token_id&&n.push(new c.ForcedEOSTokenLogitsProcessor(e.max_length,e.forced_eos_token_id)),null!==e.begin_suppress_tokens){const r=t>1||null===e.forced_bos_token_id?t:t+1;n.push(new c.SuppressTokensAtBeginLogitsProcessor(e.begin_suppress_tokens,r))}return null!==e.guidance_scale&&e.guidance_scale>1&&n.push(new c.ClassifierFreeGuidanceLogitsProcessor(e.guidance_scale)),null!==r&&n.extend(r),n}_prepare_generation_config(e,t,r=d.GenerationConfig){const n={...this.config};for(const e of["decoder","generator","text_config"])e in n&&Object.assign(n,n[e]);const o=new r(n);return"generation_config"in this&&Object.assign(o,this.generation_config),e&&Object.assign(o,e),t&&Object.assign(o,(0,i.pick)(t,Object.getOwnPropertyNames(o))),o}_get_stopping_criteria(e,t=null){const r=new h.StoppingCriteriaList;return null!==e.max_length&&r.push(new h.MaxLengthCriteria(e.max_length,this.config.max_position_embeddings??null)),null!==e.eos_token_id&&r.push(new h.EosTokenCriteria(e.eos_token_id)),t&&r.extend(t),r}_validate_model_class(){if(!this.can_generate){const e=[ii,di,ai,ti],t=v.get(this.constructor),r=new Set,n=this.config.model_type;for(const t of e){const e=t.get(n);e&&r.add(e[0])}let o=`The current model class (${t}) is not compatible with \`.generate()\`, as it doesn't have a language model head.`;throw r.size>0&&(o+=` Please use the following class instead: ${[...r].join(", ")}`),Error(o)}}prepare_inputs_for_generation(...e){return this._prepare_inputs_for_generation(this,...e)}_update_model_kwargs_for_generation({generated_input_ids:e,outputs:t,model_inputs:r,is_encoder_decoder:n}){return r.past_key_values=this.getPastKeyValues(t,r.past_key_values),r.input_ids=new u.Tensor("int64",e.flat(),[e.length,1]),n||(r.attention_mask=(0,u.cat)([r.attention_mask,(0,u.ones)([r.attention_mask.dims[0],1])],1)),r.position_ids=null,r}_prepare_model_inputs({inputs:e,bos_token_id:t,model_kwargs:r}){const n=(0,i.pick)(r,this.forward_params),o=this.main_input_name;if(o in n){if(e)throw new Error("`inputs`: {inputs}` were passed alongside {input_name} which is not allowed. Make sure to either pass {inputs} or {input_name}=...")}else n[o]=e;return{inputs_tensor:n[o],model_inputs:n,model_input_name:o}}async _prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:e,model_inputs:t,model_input_name:r,generation_config:n}){if(this.sessions.model.inputNames.includes("inputs_embeds")&&!t.inputs_embeds&&"_prepare_inputs_embeds"in this){const{input_ids:e,pixel_values:r,attention_mask:n,...o}=t,s=await this._prepare_inputs_embeds(t);t={...o,...(0,i.pick)(s,["inputs_embeds","attention_mask"])}}let{last_hidden_state:o}=await B(this,t);if(null!==n.guidance_scale&&n.guidance_scale>1)o=(0,u.cat)([o,(0,u.full_like)(o,0)],0),"attention_mask"in t&&(t.attention_mask=(0,u.cat)([t.attention_mask,(0,u.zeros_like)(t.attention_mask)],0));else if(t.decoder_input_ids){const e=L(t.decoder_input_ids).dims[0];if(e!==o.dims[0]){if(1!==o.dims[0])throw new Error(`The encoder outputs have a different batch size (${o.dims[0]}) than the decoder inputs (${e}).`);o=(0,u.cat)(Array.from({length:e},(()=>o)),0)}}return t.encoder_outputs=o,t}_prepare_decoder_input_ids_for_generation({batch_size:e,model_input_name:t,model_kwargs:r,decoder_start_token_id:n,bos_token_id:o,generation_config:s}){let{decoder_input_ids:a,...i}=r;if(a)Array.isArray(a[0])||(a=Array.from({length:e},(()=>a)));else if(n??=o,"musicgen"===this.config.model_type)a=Array.from({length:e*this.config.decoder.num_codebooks},(()=>[n]));else if(Array.isArray(n)){if(n.length!==e)throw new Error(`\`decoder_start_token_id\` expcted to have length ${e} but got ${n.length}`);a=n}else a=Array.from({length:e},(()=>[n]));return a=L(a),r.decoder_attention_mask=(0,u.ones_like)(a),{input_ids:a,model_inputs:i}}async generate({inputs:e=null,generation_config:t=null,logits_processor:r=null,stopping_criteria:n=null,streamer:o=null,...s}){this._validate_model_class(),t=this._prepare_generation_config(t,s);let{inputs_tensor:a,model_inputs:i,model_input_name:l}=this._prepare_model_inputs({inputs:e,model_kwargs:s});const c=this.config.is_encoder_decoder;let d;c&&("encoder_outputs"in i||(i=await this._prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:a,model_inputs:i,model_input_name:l,generation_config:t}))),c?({input_ids:d,model_inputs:i}=this._prepare_decoder_input_ids_for_generation({batch_size:i[l].dims.at(0),model_input_name:l,model_kwargs:i,decoder_start_token_id:t.decoder_start_token_id,bos_token_id:t.bos_token_id,generation_config:t})):d=i[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,r),_=this._get_stopping_criteria(t,n),f=i[l].dims.at(0),g=m.LogitsSampler.getSampler(t),M=new Array(f).fill(0),w=d.tolist();o&&o.put(w);let T=null,b={};for(;;){i=this.prepare_inputs_for_generation(w,i,t);const e=await this.forward(i);if(t.output_attentions&&t.return_dict_in_generate){const t=this.getAttentions(e);for(const e in t)e in b||(b[e]=[]),b[e].push(t[e])}const r=h(w,e.logits.slice(null,-1,null)),n=[];for(let e=0;e<r.dims.at(0);++e){const t=r[e],o=await g(t);for(const[t,r]of o){const o=BigInt(t);M[e]+=r,w[e].push(o),n.push([o]);break}}o&&o.put(n);if(_(w).every((e=>e))){t.return_dict_in_generate&&(T=this.getPastKeyValues(e,i.past_key_values,!1));break}i=this._update_model_kwargs_for_generation({generated_input_ids:n,outputs:e,model_inputs:i,is_encoder_decoder:c})}o&&o.end();const x=new u.Tensor("int64",w.flat(),[w.length,w[0].length]);return t.return_dict_in_generate?{sequences:x,past_key_values:T,...b}:x}getPastKeyValues(e,t,r=!0){const n=Object.create(null);for(const o in e)if(o.startsWith("present")){const s=o.replace("present","past_key_values");if(t&&o.includes("encoder"))n[s]=t[s];else{if(r&&t){const e=t[s];"gpu-buffer"===e.location&&e.dispose()}n[s]=e[o]}}return n}getAttentions(e){const t={};for(const r of["cross_attentions","encoder_attentions","decoder_attentions"])for(const n in e)n.startsWith(r)&&(r in t||(t[r]=[]),t[r].push(e[n]));return t}addPastKeyValues(e,t){if(t)Object.assign(e,t);else{const t=this.custom_config.kv_cache_dtype??"float32",r="float16"===t?new Uint16Array:[],o=(0,n.getKeyValueShapes)(this.config);for(const n in o)e[n]=new u.Tensor(t,r,o[n])}}async encode_image({pixel_values:e}){const t=(await A(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 A(this.sessions.embed_tokens,{input_ids:e})).inputs_embeds}}class G{}class q extends G{constructor({last_hidden_state:e,hidden_states:t=null,attentions:r=null}){super(),this.last_hidden_state=e,this.hidden_states=t,this.attentions=r}}class $ extends R{}class W extends ${}class U extends ${async _call(e){return new cl(await super._call(e))}}class X extends ${async _call(e){return new al(await super._call(e))}}class Q extends ${async _call(e){return new ll(await super._call(e))}}class H extends ${async _call(e){return new dl(await super._call(e))}}class Y extends R{}class J extends Y{}class K extends R{}class Z extends K{}class ee extends K{async _call(e){return new cl(await super._call(e))}}class te extends K{async _call(e){return new al(await super._call(e))}}class re extends K{async _call(e){return new ll(await super._call(e))}}class ne extends K{async _call(e){return new dl(await super._call(e))}}class oe extends R{}class se extends oe{}class ae extends oe{async _call(e){return new cl(await super._call(e))}}class ie extends oe{async _call(e){return new al(await super._call(e))}}class le extends oe{async _call(e){return new ll(await super._call(e))}}class ce extends oe{async _call(e){return new dl(await super._call(e))}}class de extends R{}class ue extends de{}class pe extends de{async _call(e){return new cl(await super._call(e))}}class he extends de{async _call(e){return new al(await super._call(e))}}class me extends de{async _call(e){return new ll(await super._call(e))}}class _e extends de{async _call(e){return new dl(await super._call(e))}}class fe extends R{}class ge extends fe{}class Me extends fe{async _call(e){return new cl(await super._call(e))}}class we extends fe{async _call(e){return new al(await super._call(e))}}class Te extends fe{async _call(e){return new ll(await super._call(e))}}class be extends fe{async _call(e){return new dl(await super._call(e))}}class xe extends R{}class ke extends xe{}class ye extends xe{async _call(e){return new cl(await super._call(e))}}class Fe extends xe{async _call(e){return new al(await super._call(e))}}class Pe extends xe{async _call(e){return new ll(await super._call(e))}}class Ce extends xe{async _call(e){return new dl(await super._call(e))}}class ve extends R{}class Se extends ve{}class Ae extends ve{async _call(e){return new cl(await super._call(e))}}class Ee extends ve{async _call(e){return new al(await super._call(e))}}class Le extends ve{async _call(e){return new ll(await super._call(e))}}class ze extends ve{async _call(e){return new dl(await super._call(e))}}class Ie extends R{}class Be extends Ie{}class Ne extends Ie{async _call(e){return new al(await super._call(e))}}class Oe extends Ie{async _call(e){return new ll(await super._call(e))}}class De extends Ie{async _call(e){return new dl(await super._call(e))}}class Ve extends Ie{async _call(e){return new cl(await super._call(e))}}class je extends R{}class Re extends je{}class Ge extends je{async _call(e){return new cl(await super._call(e))}}class qe extends je{async _call(e){return new al(await super._call(e))}}class $e extends je{async _call(e){return new ll(await super._call(e))}}class We extends R{}class Ue extends We{}class Xe extends We{async _call(e){return new cl(await super._call(e))}}class Qe extends We{async _call(e){return new al(await super._call(e))}}class He extends We{async _call(e){return new dl(await super._call(e))}}class Ye extends R{}class Je extends Ye{}class Ke extends Ye{async _call(e){return new cl(await super._call(e))}}class Ze extends Ye{async _call(e){return new al(await super._call(e))}}class et extends Ye{async _call(e){return new ll(await super._call(e))}}class tt extends Ye{async _call(e){return new dl(await super._call(e))}}class rt extends R{}class nt extends rt{}class ot extends rt{async _call(e){return new cl(await super._call(e))}}class st extends rt{async _call(e){return new al(await super._call(e))}}class at extends rt{async _call(e){return new dl(await super._call(e))}}class it extends R{}class lt extends it{}class ct extends it{async _call(e){return new al(await super._call(e))}}class dt extends it{async _call(e){return new dl(await super._call(e))}}class ut extends it{async _call(e){return new cl(await super._call(e))}}class pt extends R{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"];constructor(e,t,r){super(e,t),this.generation_config=r}}class ht extends pt{}class mt extends pt{}class _t extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class ft extends _t{}class gt extends _t{}class Mt extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class wt extends Mt{}class Tt extends Mt{}class bt extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class xt extends bt{}class kt extends bt{}class yt extends bt{async _call(e){return new al(await super._call(e))}}class Ft extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Pt extends Ft{}class Ct extends Ft{}class vt extends Ft{async _call(e){return new al(await super._call(e))}}class St extends Ft{}class At extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Et extends At{}class Lt extends At{}class zt extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class It extends zt{}class Bt extends zt{}class Nt extends R{}class Ot extends Nt{}class Dt extends Nt{async _call(e){return new cl(await super._call(e))}}class Vt extends Nt{async _call(e){return new al(await super._call(e))}}class jt extends Nt{async _call(e){return new ll(await super._call(e))}}class Rt extends Nt{async _call(e){return new dl(await super._call(e))}}class Gt extends R{}class qt extends Gt{}class $t extends Gt{async _call(e){return new cl(await super._call(e))}}class Wt extends Gt{async _call(e){return new al(await super._call(e))}}class Ut extends Gt{async _call(e){return new ll(await super._call(e))}}class Xt extends Gt{async _call(e){return new dl(await super._call(e))}}class Qt extends R{}class Ht extends Qt{}class Yt extends Qt{async _call(e){return new cl(await super._call(e))}}class Jt extends Qt{async _call(e){return new al(await super._call(e))}}class Kt extends Qt{async _call(e){return new ll(await super._call(e))}}class Zt extends Qt{async _call(e){return new dl(await super._call(e))}}class er extends R{}class tr extends er{}class rr extends er{}class nr extends R{requires_attention_mask=!1;main_input_name="input_features";forward_params=["input_features","attention_mask","decoder_input_ids","decoder_attention_mask","past_key_values"];constructor(e,t,r){super(e,t),this.generation_config=r}}class or extends nr{}class sr extends nr{_prepare_generation_config(e,t){return super._prepare_generation_config(e,t,f.WhisperGenerationConfig)}_retrieve_init_tokens(e){const t=[e.decoder_start_token_id];let r=e.language;const n=e.task;if(e.is_multilingual){r||(console.warn("No language specified - defaulting to English (en)."),r="en");const o=`<|${(0,g.whisper_language_to_code)(r)}|>`;t.push(e.lang_to_id[o]),t.push(e.task_to_id[n??"transcribe"])}else if(r||n)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:r=null,stopping_criteria:n=null,...o}){t=this._prepare_generation_config(t,o);const s=o.decoder_input_ids??this._retrieve_init_tokens(t);if(t.return_timestamps&&(r??=new c.LogitsProcessorList,r.push(new c.WhisperTimeStampLogitsProcessor(t,s))),t.begin_suppress_tokens&&(r??=new c.LogitsProcessorList,r.push(new c.SuppressTokensAtBeginLogitsProcessor(t.begin_suppress_tokens,s.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 a=await super.generate({inputs:e,generation_config:t,logits_processor:r,decoder_input_ids:s,...o});return t.return_token_timestamps&&(a.token_timestamps=this._extract_token_timestamps(a,t.alignment_heads,t.num_frames)),a}_extract_token_timestamps(e,t,r=null,n=.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==r&&console.warn("`num_frames` has not been set, meaning the entire audio will be analyzed. This may lead to inaccurate token-level timestamps for short audios (< 30 seconds).");let o=this.config.median_filter_width;void 0===o&&(console.warn("Model config has no `median_filter_width`, using default value of 7."),o=7);const s=e.cross_attentions,a=Array.from({length:this.config.decoder_layers},((e,t)=>(0,u.cat)(s.map((e=>e[t])),2))),l=(0,u.stack)(t.map((([e,t])=>{if(e>=a.length)throw new Error(`Layer index ${e} is out of bounds for cross attentions (length ${a.length}).`);return r?a[e].slice(null,t,null,[0,r]):a[e].slice(null,t)}))).transpose(1,0,2,3),[c,d]=(0,u.std_mean)(l,-2,0,!0),h=l.clone();for(let e=0;e<h.dims[0];++e){const t=h[e];for(let r=0;r<t.dims[0];++r){const n=t[r],s=c[e][r][0].data,a=d[e][r][0].data;for(let e=0;e<n.dims[0];++e){let t=n[e].data;for(let e=0;e<t.length;++e)t[e]=(t[e]-a[e])/s[e];t.set((0,p.medianFilter)(t,o))}}}const m=[(0,u.mean)(h,1)],_=e.sequences.dims,f=new u.Tensor("float32",new Float32Array(_[0]*_[1]),_);for(let e=0;e<_[0];++e){const t=m[e].neg().squeeze_(0),[r,o]=(0,p.dynamic_time_warping)(t.tolist()),s=Array.from({length:r.length-1},((e,t)=>r[t+1]-r[t])),a=(0,i.mergeArrays)([1],s).map((e=>!!e)),l=[];for(let e=0;e<a.length;++e)a[e]&&l.push(o[e]*n);f[e].data.set(l,1)}return f}}class ar extends R{main_input_name="pixel_values";forward_params=["pixel_values","input_ids","encoder_hidden_states","past_key_values"];constructor(e,t,r){super(e,t),this.generation_config=r}}class ir extends R{forward_params=["input_ids","pixel_values","attention_mask","position_ids","past_key_values"];constructor(e,t,r){super(e,t),this.generation_config=r}}class lr extends ir{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:r,attention_mask:n}){const o=this.config.image_token_index,s=r.tolist().map((e=>e.findIndex((e=>e==o)))),a=s.every((e=>-1===e)),i=s.every((e=>-1!==e));if(!a&&!i)throw new Error("Every input should contain either 0 or 1 image token.");if(a)return{inputs_embeds:e,attention_mask:n};const l=[],c=[];for(let r=0;r<s.length;++r){const o=s[r],a=e[r],i=t[r],d=n[r];l.push((0,u.cat)([a.slice([0,o]),i,a.slice([o+1,a.dims[0]])],0)),c.push((0,u.cat)([d.slice([0,o]),(0,u.ones)([i.dims[0]]),d.slice([o+1,d.dims[0]])],0))}return{inputs_embeds:(0,u.stack)(l,0),attention_mask:(0,u.stack)(c,0)}}}class cr extends lr{}class dr extends R{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,r){super(e,t),this.generation_config=r}}class ur extends dr{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:r,attention_mask:n}){return{inputs_embeds:(0,u.cat)([t,e],1),attention_mask:(0,u.cat)([(0,u.ones)(t.dims.slice(0,2)),n],1)}}async _prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:r,attention_mask:n}){if(!e&&!t)throw new Error("Either `input_ids` or `pixel_values` should be provided.");let o,s;return e&&(o=await this.encode_text({input_ids:e})),t&&(s=await this.encode_image({pixel_values:t})),o&&s?({inputs_embeds:r,attention_mask:n}=this._merge_input_ids_with_image_features({inputs_embeds:o,image_features:s,input_ids:e,attention_mask:n})):r=o||s,{inputs_embeds:r,attention_mask:n}}async forward({input_ids:e,pixel_values:t,attention_mask:r,decoder_input_ids:n,decoder_attention_mask:o,encoder_outputs:s,past_key_values:a,inputs_embeds:i,decoder_inputs_embeds:l}){if(i||({inputs_embeds:i,attention_mask:r}=await this._prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:i,attention_mask:r})),!s){let{last_hidden_state:e}=await B(this,{inputs_embeds:i,attention_mask:r});s=e}if(!l){if(!n)throw new Error("Either `decoder_input_ids` or `decoder_inputs_embeds` should be provided.");l=await this.encode_text({input_ids:n})}const c={inputs_embeds:l,attention_mask:o,encoder_attention_mask:r,encoder_hidden_states:s,past_key_values:a};return await N(this,c,!0)}}class pr extends R{}class hr extends pr{}class mr extends pr{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class _r extends pr{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class fr extends pr{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class gr extends pr{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class Mr extends R{}class wr extends Mr{}class Tr extends Mr{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class br extends pr{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class xr extends R{}class kr extends xr{}class yr extends R{}class Fr extends yr{}class Pr extends yr{}class Cr extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class vr extends Cr{}class Sr extends Cr{}class Ar extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Er extends Ar{}class Lr extends Ar{}class zr extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Ir extends zr{}class Br extends zr{}class Nr extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Or extends Nr{}class Dr extends Nr{}class Vr extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class jr extends Vr{}class Rr extends Vr{}class Gr extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class qr extends Gr{}class $r extends Gr{}class Wr extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Ur extends Wr{}class Xr extends Wr{}class Qr extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Hr extends Qr{}class Yr extends Qr{}class Jr extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Kr extends Jr{}class Zr extends Jr{}class en extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class tn extends en{}class rn extends en{}class nn extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class on extends nn{}class sn extends nn{}class an extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class ln extends an{}class cn extends an{}class dn extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class un extends dn{}class pn extends dn{}class hn extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class mn extends hn{}class _n extends hn{}class fn extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class gn extends fn{}class Mn extends fn{}class wn extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Tn extends wn{}class bn extends wn{}class xn extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class kn extends xn{}class yn extends xn{}class Fn extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Pn extends Fn{}class Cn extends Fn{}class vn extends R{}class Sn extends vn{}class An extends vn{async _call(e){return new al(await super._call(e))}}class En extends R{}class Ln extends En{}class zn extends En{async _call(e){return new al(await super._call(e))}}class In extends R{}class Bn extends In{}class Nn extends R{}class On extends Nn{}class Dn extends Nn{async _call(e){return new al(await super._call(e))}}class Vn extends R{}class jn extends Vn{}class Rn extends R{}class Gn extends Rn{}class qn extends Rn{async _call(e){return new al(await super._call(e))}}class $n extends R{}class Wn extends $n{async _call(e){return new hl(await super._call(e))}}class Un extends R{}class Xn extends Un{}class Qn extends Un{async _call(e){return new al(await super._call(e))}}class Hn extends R{}class Yn extends Hn{}class Jn extends Hn{async _call(e){return new al(await super._call(e))}}class Kn extends R{}class Zn extends Kn{}class eo extends Kn{}class to extends R{}class ro extends to{}class no extends to{}class oo extends R{}class so extends oo{}class ao extends oo{async _call(e){return new al(await super._call(e))}}class io extends R{}class lo extends io{}class co extends io{async _call(e){return new po(await super._call(e))}}class uo extends io{async _call(e){return new ho(await super._call(e))}}class po extends G{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class ho extends G{constructor({logits:e,pred_boxes:t,pred_masks:r}){super(),this.logits=e,this.pred_boxes=t,this.pred_masks=r}}class mo extends R{}class _o extends mo{}class fo extends mo{async _call(e){return new go(await super._call(e))}}class go extends G{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class Mo extends R{}class wo extends Mo{}class To extends Mo{async _call(e){return new bo(await super._call(e))}}class bo extends po{}class xo extends R{}class ko extends xo{}class yo extends xo{async _call(e){return new al(await super._call(e))}}class Fo extends R{}class Po extends Fo{}class Co extends Fo{async _call(e){return new al(await super._call(e))}}class vo extends R{}class So extends vo{}class Ao extends vo{async _call(e){return new al(await super._call(e))}}class Eo extends R{}class Lo extends Eo{}class zo extends Eo{async _call(e){return new al(await super._call(e))}}class Io extends R{}class Bo extends Io{}class No extends Io{}class Oo extends R{}class Do extends Oo{}class Vo extends Oo{}class jo extends R{}class Ro extends jo{}class Go extends R{}class qo extends Go{}class $o extends Go{}class Wo extends Go{}class Uo extends R{}class Xo extends Uo{}class Qo extends Uo{}class Ho extends R{}class Yo extends Ho{}class Jo extends Ho{}class Ko extends R{}class Zo extends Ko{}class es extends R{}class ts extends es{}class rs extends es{async _call(e){return new al(await super._call(e))}}class ns extends R{}class os extends ns{}class ss extends ns{async _call(e){return new al(await super._call(e))}}class as extends R{}class is extends as{}class ls extends as{async _call(e){return new al(await super._call(e))}}class cs extends R{}class ds extends cs{}class us extends cs{async _call(e){return new ps(await super._call(e))}}class ps extends G{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class hs extends R{}class ms extends hs{async get_image_embeddings({pixel_values:e}){return await B(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),r=t.reduce(((e,t)=>e*t),1);e.input_labels=new u.Tensor("int64",new BigInt64Array(r).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 A(this.sessions.prompt_encoder_mask_decoder,t)}async _call(e){return new _s(await super._call(e))}}class _s extends G{constructor({iou_scores:e,pred_masks:t}){super(),this.iou_scores=e,this.pred_masks=t}}class fs extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class gs extends fs{}class Ms extends fs{}class ws extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Ts extends ws{}class bs extends ws{}class xs extends R{}class ks extends xs{}class ys extends xs{async _call(e){return new ul(await super._call(e))}}class Fs extends xs{async _call(e){return new al(await super._call(e))}}class Ps extends xs{async _call(e){return new ll(await super._call(e))}}class Cs extends R{}class vs extends Cs{}class Ss extends Cs{async _call(e){return new ll(await super._call(e))}}class As extends R{}class Es extends As{}class Ls extends R{}class zs extends Ls{}class Is extends Ls{async _call(e){return new ul(await super._call(e))}}class Bs extends Ls{async _call(e){return new al(await super._call(e))}}class Ns extends R{}class Os extends Ns{}class Ds extends Ns{async _call(e){return new ul(await super._call(e))}}class Vs extends Ns{async _call(e){return new al(await super._call(e))}}class js extends Ns{async _call(e){return new ll(await super._call(e))}}class Rs extends R{}class Gs extends Rs{}class qs extends Rs{async _call(e){return new ul(await super._call(e))}}class $s extends Rs{async _call(e){return new al(await super._call(e))}}class Ws extends R{}class Us extends xs{}class Xs extends xs{async _call(e){return new ul(await super._call(e))}}class Qs extends xs{async _call(e){return new al(await super._call(e))}}class Hs extends R{}class Ys extends Hs{}class Js extends Hs{async _call(e){return new ul(await super._call(e))}}class Ks extends Hs{async _call(e){return new al(await super._call(e))}}class Zs extends Hs{async _call(e){return new il(await super._call(e))}}class ea extends Hs{async _call(e){return new ll(await super._call(e))}}class ta extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class ra extends ta{}class na extends ta{}class oa extends ta{async generate_speech(e,t,{threshold:r=.5,minlenratio:n=0,maxlenratio:o=20,vocoder:s=null}={}){const a={input_ids:e},{encoder_outputs:i,encoder_attention_mask:l}=await B(this,a),c=i.dims[1]/this.config.reduction_factor,d=Math.floor(c*o),p=Math.floor(c*n),h=this.config.num_mel_bins;let m=[],_=null,f=null,g=0;for(;;){++g;const e=z(!!f);let n;n=f?f.output_sequence_out:new u.Tensor("float32",new Float32Array(h),[1,1,h]);let o={use_cache_branch:e,output_sequence:n,encoder_attention_mask:l,speaker_embeddings:t,encoder_hidden_states:i};this.addPastKeyValues(o,_),f=await A(this.sessions.decoder_model_merged,o),_=this.getPastKeyValues(f,_);const{prob:s,spectrum:a}=f;if(m.push(a),g>=p&&(Array.from(s.data).filter((e=>e>=r)).length>0||g>=d))break}const M=(0,u.cat)(m),{waveform:w}=await A(s.sessions.model,{spectrogram:M});return{spectrogram:M,waveform:w}}}class sa extends R{main_input_name="spectrogram"}class aa extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class ia extends aa{}class la extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class ca extends la{}class da extends la{}class ua extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class pa extends ua{}class ha extends ua{}class ma extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class _a extends ma{}class fa extends ma{}class ga extends R{}class Ma extends ga{}class wa extends ga{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class Ta extends ga{static async from_pretrained(e,t={}){return t.model_file_name??="audio_model",super.from_pretrained(e,t)}}class ba extends R{}class xa extends ba{async _call(e){return new ml(await super._call(e))}}class ka extends R{}class ya extends ka{}class Fa extends ka{}class Pa extends ka{}class Ca extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class va extends Ca{}class Sa extends Ca{}class Aa extends R{}class Ea extends Aa{}class La extends Aa{async _call(e){return new al(await super._call(e))}}class za extends R{}class Ia extends za{}class Ba extends za{}class Na extends R{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"];constructor(e,t,r){super(e,t),this.generation_config=r}_apply_and_filter_by_delay_pattern_mask(e){const[t,r]=e.dims,n=this.config.decoder.num_codebooks,o=r-n;let s=0;for(let t=0;t<e.size;++t){if(e.data[t]===this.config.decoder.pad_token_id)continue;const a=t%r-Math.floor(t/r)%n;a>0&&a<=o&&(e.data[s++]=e.data[t])}const a=Math.floor(t/n),i=s/(a*n);return new u.Tensor(e.type,e.data.slice(0,s),[a,n,i])}prepare_inputs_for_generation(e,t,r){let n=structuredClone(e);for(let e=0;e<n.length;++e)for(let t=0;t<n[e].length;++t)e%this.config.decoder.num_codebooks>=t&&(n[e][t]=BigInt(this.config.decoder.pad_token_id));null!==r.guidance_scale&&r.guidance_scale>1&&(n=n.concat(n));return super.prepare_inputs_for_generation(n,t,r)}async generate(e){const t=await super.generate(e),r=this._apply_and_filter_by_delay_pattern_mask(t).unsqueeze_(0),{audio_values:n}=await A(this.sessions.encodec_decode,{audio_codes:r});return n}}class Oa extends R{}class Da extends Oa{}class Va extends Oa{async _call(e){return new al(await super._call(e))}}class ja extends R{}class Ra extends ja{}class Ga extends ja{async _call(e){return new al(await super._call(e))}}class qa extends R{}class $a extends qa{}class Wa extends qa{async _call(e){return new al(await super._call(e))}}class Ua extends R{}class Xa extends Ua{}class Qa extends Ua{async _call(e){return new al(await super._call(e))}}class Ha extends R{}class Ya extends Ha{}class Ja{static MODEL_CLASS_MAPPINGS=null;static BASE_IF_FAIL=!1;static async from_pretrained(e,{progress_callback:t=null,config:r=null,cache_dir:o=null,local_files_only:s=!1,revision:a="main",model_file_name:i=null,subfolder:l="onnx",device:c=null,dtype:d=null,use_external_data_format:u=null,session_options:p={}}={}){const h={progress_callback:t,config:r,cache_dir:o,local_files_only:s,revision:a,model_file_name:i,subfolder:l,device:c,dtype:d,use_external_data_format:u,session_options:p};if(h.config=await n.AutoConfig.from_pretrained(e,h),!this.MODEL_CLASS_MAPPINGS)throw new Error("`MODEL_CLASS_MAPPINGS` not implemented for this type of `AutoClass`: "+this.name);for(const t of this.MODEL_CLASS_MAPPINGS){const r=t.get(h.config.model_type);if(r)return await r[1].from_pretrained(e,h)}if(this.BASE_IF_FAIL)return console.warn(`Unknown model class "${h.config.model_type}", attempting to construct from base class.`),await R.from_pretrained(e,h);throw Error(`Unsupported model type: ${h.config.model_type}`)}}const Ka=new Map([["bert",["BertModel",W]],["nomic_bert",["NomicBertModel",J]],["roformer",["RoFormerModel",Z]],["electra",["ElectraModel",ue]],["esm",["EsmModel",Re]],["convbert",["ConvBertModel",se]],["camembert",["CamembertModel",ge]],["deberta",["DebertaModel",ke]],["deberta-v2",["DebertaV2Model",Se]],["mpnet",["MPNetModel",Je]],["albert",["AlbertModel",lt]],["distilbert",["DistilBertModel",Be]],["roberta",["RobertaModel",Ot]],["xlm",["XLMModel",qt]],["xlm-roberta",["XLMRobertaModel",Ht]],["clap",["ClapModel",Ma]],["clip",["CLIPModel",hr]],["clipseg",["CLIPSegModel",Fr]],["chinese_clip",["ChineseCLIPModel",kr]],["siglip",["SiglipModel",wr]],["mobilebert",["MobileBertModel",Ue]],["squeezebert",["SqueezeBertModel",nt]],["wav2vec2",["Wav2Vec2Model",ks]],["wav2vec2-bert",["Wav2Vec2BertModel",Gs]],["unispeech",["UniSpeechModel",zs]],["unispeech-sat",["UniSpeechSatModel",Os]],["hubert",["HubertModel",Us]],["wavlm",["WavLMModel",Ys]],["audio-spectrogram-transformer",["ASTModel",tr]],["vits",["VitsModel",xa]],["pyannote",["PyAnnoteModel",vs]],["wespeaker-resnet",["WeSpeakerResNetModel",Es]],["detr",["DetrModel",lo]],["rt_detr",["RTDetrModel",_o]],["table-transformer",["TableTransformerModel",wo]],["vit",["ViTModel",Sn]],["pvt",["PvtModel",Ln]],["vit_msn",["ViTMSNModel",On]],["vit_mae",["ViTMAEModel",Bn]],["groupvit",["GroupViTModel",jn]],["fastvit",["FastViTModel",Gn]],["mobilevit",["MobileViTModel",Xn]],["mobilevitv2",["MobileViTV2Model",Yn]],["owlvit",["OwlViTModel",Zn]],["owlv2",["Owlv2Model",ro]],["beit",["BeitModel",so]],["deit",["DeiTModel",ko]],["hiera",["HieraModel",Po]],["convnext",["ConvNextModel",ts]],["convnextv2",["ConvNextV2Model",os]],["dinov2",["Dinov2Model",is]],["resnet",["ResNetModel",So]],["swin",["SwinModel",Lo]],["swin2sr",["Swin2SRModel",Bo]],["donut-swin",["DonutSwinModel",Zo]],["yolos",["YolosModel",ds]],["dpt",["DPTModel",Do]],["glpn",["GLPNModel",Yo]],["hifigan",["SpeechT5HifiGan",sa]],["efficientnet",["EfficientNetModel",Ea]],["decision_transformer",["DecisionTransformerModel",Ya]],["mobilenet_v1",["MobileNetV1Model",Da]],["mobilenet_v2",["MobileNetV2Model",Ra]],["mobilenet_v3",["MobileNetV3Model",$a]],["mobilenet_v4",["MobileNetV4Model",Xa]],["maskformer",["MaskFormerModel",Xo]]]),Za=new Map([["t5",["T5Model",ht]],["longt5",["LongT5Model",ft]],["mt5",["MT5Model",wt]],["bart",["BartModel",xt]],["mbart",["MBartModel",Pt]],["marian",["MarianModel",gs]],["whisper",["WhisperModel",or]],["m2m_100",["M2M100Model",Ts]],["blenderbot",["BlenderbotModel",Et]],["blenderbot-small",["BlenderbotSmallModel",It]]]),ei=new Map([["bloom",["BloomModel",Tn]],["jais",["JAISModel",Er]],["gpt2",["GPT2Model",vr]],["gptj",["GPTJModel",jr]],["gpt_bigcode",["GPTBigCodeModel",qr]],["gpt_neo",["GPTNeoModel",Ir]],["gpt_neox",["GPTNeoXModel",Or]],["codegen",["CodeGenModel",Ur]],["llama",["LlamaModel",Hr]],["cohere",["CohereModel",Kr]],["gemma",["GemmaModel",tn]],["gemma2",["Gemma2Model",on]],["openelm",["OpenELMModel",ln]],["qwen2",["Qwen2Model",un]],["phi",["PhiModel",mn]],["phi3",["Phi3Model",gn]],["mpt",["MptModel",kn]],["opt",["OPTModel",Pn]],["mistral",["MistralModel",ca]],["starcoder2",["Starcoder2Model",pa]],["falcon",["FalconModel",_a]],["stablelm",["StableLmModel",va]]]),ti=new Map([["speecht5",["SpeechT5ForSpeechToText",na]],["whisper",["WhisperForConditionalGeneration",sr]]]),ri=new Map([["speecht5",["SpeechT5ForTextToSpeech",oa]]]),ni=new Map([["vits",["VitsModel",xa]],["musicgen",["MusicgenForConditionalGeneration",Na]]]),oi=new Map([["bert",["BertForSequenceClassification",X]],["roformer",["RoFormerForSequenceClassification",te]],["electra",["ElectraForSequenceClassification",he]],["esm",["EsmForSequenceClassification",qe]],["convbert",["ConvBertForSequenceClassification",ie]],["camembert",["CamembertForSequenceClassification",we]],["deberta",["DebertaForSequenceClassification",Fe]],["deberta-v2",["DebertaV2ForSequenceClassification",Ee]],["mpnet",["MPNetForSequenceClassification",Ze]],["albert",["AlbertForSequenceClassification",ct]],["distilbert",["DistilBertForSequenceClassification",Ne]],["roberta",["RobertaForSequenceClassification",Vt]],["xlm",["XLMForSequenceClassification",Wt]],["xlm-roberta",["XLMRobertaForSequenceClassification",Jt]],["bart",["BartForSequenceClassification",yt]],["mbart",["MBartForSequenceClassification",vt]],["mobilebert",["MobileBertForSequenceClassification",Qe]],["squeezebert",["SqueezeBertForSequenceClassification",st]]]),si=new Map([["bert",["BertForTokenClassification",Q]],["roformer",["RoFormerForTokenClassification",re]],["electra",["ElectraForTokenClassification",me]],["esm",["EsmForTokenClassification",$e]],["convbert",["ConvBertForTokenClassification",le]],["camembert",["CamembertForTokenClassification",Te]],["deberta",["DebertaForTokenClassification",Pe]],["deberta-v2",["DebertaV2ForTokenClassification",Le]],["mpnet",["MPNetForTokenClassification",et]],["distilbert",["DistilBertForTokenClassification",Oe]],["roberta",["RobertaForTokenClassification",jt]],["xlm",["XLMForTokenClassification",Ut]],["xlm-roberta",["XLMRobertaForTokenClassification",Kt]]]),ai=new Map([["t5",["T5ForConditionalGeneration",mt]],["longt5",["LongT5ForConditionalGeneration",gt]],["mt5",["MT5ForConditionalGeneration",Tt]],["bart",["BartForConditionalGeneration",kt]],["mbart",["MBartForConditionalGeneration",Ct]],["marian",["MarianMTModel",Ms]],["m2m_100",["M2M100ForConditionalGeneration",bs]],["blenderbot",["BlenderbotForConditionalGeneration",Lt]],["blenderbot-small",["BlenderbotSmallForConditionalGeneration",Bt]]]),ii=new Map([["bloom",["BloomForCausalLM",bn]],["gpt2",["GPT2LMHeadModel",Sr]],["jais",["JAISLMHeadModel",Lr]],["gptj",["GPTJForCausalLM",Rr]],["gpt_bigcode",["GPTBigCodeForCausalLM",$r]],["gpt_neo",["GPTNeoForCausalLM",Br]],["gpt_neox",["GPTNeoXForCausalLM",Dr]],["codegen",["CodeGenForCausalLM",Xr]],["llama",["LlamaForCausalLM",Yr]],["cohere",["CohereForCausalLM",Zr]],["gemma",["GemmaForCausalLM",rn]],["gemma2",["Gemma2ForCausalLM",sn]],["openelm",["OpenELMForCausalLM",cn]],["qwen2",["Qwen2ForCausalLM",pn]],["phi",["PhiForCausalLM",_n]],["phi3",["Phi3ForCausalLM",Mn]],["mpt",["MptForCausalLM",yn]],["opt",["OPTForCausalLM",Cn]],["mbart",["MBartForCausalLM",St]],["mistral",["MistralForCausalLM",da]],["starcoder2",["Starcoder2ForCausalLM",ha]],["falcon",["FalconForCausalLM",fa]],["trocr",["TrOCRForCausalLM",ia]],["stablelm",["StableLmForCausalLM",Sa]]]),li=new Map([["bert",["BertForMaskedLM",U]],["roformer",["RoFormerForMaskedLM",ee]],["electra",["ElectraForMaskedLM",pe]],["esm",["EsmForMaskedLM",Ge]],["convbert",["ConvBertForMaskedLM",ae]],["camembert",["CamembertForMaskedLM",Me]],["deberta",["DebertaForMaskedLM",ye]],["deberta-v2",["DebertaV2ForMaskedLM",Ae]],["mpnet",["MPNetForMaskedLM",Ke]],["albert",["AlbertForMaskedLM",ut]],["distilbert",["DistilBertForMaskedLM",Ve]],["roberta",["RobertaForMaskedLM",Dt]],["xlm",["XLMWithLMHeadModel",$t]],["xlm-roberta",["XLMRobertaForMaskedLM",Yt]],["mobilebert",["MobileBertForMaskedLM",Xe]],["squeezebert",["SqueezeBertForMaskedLM",ot]]]),ci=new Map([["bert",["BertForQuestionAnswering",H]],["roformer",["RoFormerForQuestionAnswering",ne]],["electra",["ElectraForQuestionAnswering",_e]],["convbert",["ConvBertForQuestionAnswering",ce]],["camembert",["CamembertForQuestionAnswering",be]],["deberta",["DebertaForQuestionAnswering",Ce]],["deberta-v2",["DebertaV2ForQuestionAnswering",ze]],["mpnet",["MPNetForQuestionAnswering",tt]],["albert",["AlbertForQuestionAnswering",dt]],["distilbert",["DistilBertForQuestionAnswering",De]],["roberta",["RobertaForQuestionAnswering",Rt]],["xlm",["XLMForQuestionAnswering",Xt]],["xlm-roberta",["XLMRobertaForQuestionAnswering",Zt]],["mobilebert",["MobileBertForQuestionAnswering",He]],["squeezebert",["SqueezeBertForQuestionAnswering",at]]]),di=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",ar]]]),ui=new Map([["llava",["LlavaForConditionalGeneration",lr]],["moondream1",["Moondream1ForConditionalGeneration",cr]],["florence2",["Florence2ForConditionalGeneration",ur]]]),pi=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",ar]]]),hi=new Map([["vit",["ViTForImageClassification",An]],["pvt",["PvtForImageClassification",zn]],["vit_msn",["ViTMSNForImageClassification",Dn]],["fastvit",["FastViTForImageClassification",qn]],["mobilevit",["MobileViTForImageClassification",Qn]],["mobilevitv2",["MobileViTV2ForImageClassification",Jn]],["beit",["BeitForImageClassification",ao]],["deit",["DeiTForImageClassification",yo]],["hiera",["HieraForImageClassification",Co]],["convnext",["ConvNextForImageClassification",rs]],["convnextv2",["ConvNextV2ForImageClassification",ss]],["dinov2",["Dinov2ForImageClassification",ls]],["resnet",["ResNetForImageClassification",Ao]],["swin",["SwinForImageClassification",zo]],["segformer",["SegformerForImageClassification",Fa]],["efficientnet",["EfficientNetForImageClassification",La]],["mobilenet_v1",["MobileNetV1ForImageClassification",Va]],["mobilenet_v2",["MobileNetV2ForImageClassification",Ga]],["mobilenet_v3",["MobileNetV3ForImageClassification",Wa]],["mobilenet_v4",["MobileNetV4ForImageClassification",Qa]]]),mi=new Map([["detr",["DetrForObjectDetection",co]],["rt_detr",["RTDetrForObjectDetection",fo]],["table-transformer",["TableTransformerForObjectDetection",To]],["yolos",["YolosForObjectDetection",us]]]),_i=new Map([["owlvit",["OwlViTForObjectDetection",eo]],["owlv2",["Owlv2ForObjectDetection",no]]]),fi=new Map([["detr",["DetrForSegmentation",uo]],["clipseg",["CLIPSegForImageSegmentation",Pr]]]),gi=new Map([["segformer",["SegformerForSemanticSegmentation",Pa]],["sapiens",["SapiensForSemanticSegmentation",qo]]]),Mi=new Map([["detr",["DetrForSegmentation",uo]],["maskformer",["MaskFormerForInstanceSegmentation",Qo]]]),wi=new Map([["sam",["SamModel",ms]]]),Ti=new Map([["wav2vec2",["Wav2Vec2ForCTC",ys]],["wav2vec2-bert",["Wav2Vec2BertForCTC",qs]],["unispeech",["UniSpeechForCTC",Is]],["unispeech-sat",["UniSpeechSatForCTC",Ds]],["wavlm",["WavLMForCTC",Js]],["hubert",["HubertForCTC",Xs]]]),bi=new Map([["wav2vec2",["Wav2Vec2ForSequenceClassification",Fs]],["wav2vec2-bert",["Wav2Vec2BertForSequenceClassification",$s]],["unispeech",["UniSpeechForSequenceClassification",Bs]],["unispeech-sat",["UniSpeechSatForSequenceClassification",Vs]],["wavlm",["WavLMForSequenceClassification",Ks]],["hubert",["HubertForSequenceClassification",Qs]],["audio-spectrogram-transformer",["ASTForAudioClassification",rr]]]),xi=new Map([["wavlm",["WavLMForXVector",Zs]]]),ki=new Map([["unispeech-sat",["UniSpeechSatForAudioFrameClassification",js]],["wavlm",["WavLMForAudioFrameClassification",ea]],["wav2vec2",["Wav2Vec2ForAudioFrameClassification",Ps]],["pyannote",["PyAnnoteForAudioFrameClassification",Ss]]]),yi=new Map([["vitmatte",["VitMatteForImageMatting",Wn]]]),Fi=new Map([["swin2sr",["Swin2SRForImageSuperResolution",No]]]),Pi=new Map([["dpt",["DPTForDepthEstimation",Vo]],["depth_anything",["DepthAnythingForDepthEstimation",Ro]],["glpn",["GLPNForDepthEstimation",Jo]],["sapiens",["SapiensForDepthEstimation",$o]]]),Ci=new Map([["sapiens",["SapiensForNormalEstimation",Wo]]]),vi=new Map([["clip",["CLIPVisionModelWithProjection",gr]],["siglip",["SiglipVisionModel",br]]]),Si=[[Ka,M],[Za,w],[ei,x],[oi,M],[si,M],[ai,T],[ti,T],[ii,x],[li,M],[ci,M],[di,b],[ui,y],[hi,M],[fi,M],[Mi,M],[gi,M],[yi,M],[Fi,M],[Pi,M],[Ci,M],[mi,M],[_i,M],[wi,k],[Ti,M],[bi,M],[ri,T],[ni,M],[xi,M],[ki,M],[vi,M]];for(const[e,t]of Si)for(const[r,n]of e.values())P.set(r,t),v.set(n,r),C.set(r,n);const Ai=[["MusicgenForConditionalGeneration",Na,F],["CLIPTextModelWithProjection",_r,M],["SiglipTextModel",Tr,M],["ClapTextModelWithProjection",wa,M],["ClapAudioModelWithProjection",Ta,M]];for(const[e,t,r]of Ai)P.set(e,r),v.set(t,e),C.set(e,t);class Ei extends Ja{static MODEL_CLASS_MAPPINGS=Si.map((e=>e[0]));static BASE_IF_FAIL=!0}class Li extends Ja{static MODEL_CLASS_MAPPINGS=[oi]}class zi extends Ja{static MODEL_CLASS_MAPPINGS=[si]}class Ii extends Ja{static MODEL_CLASS_MAPPINGS=[ai]}class Bi extends Ja{static MODEL_CLASS_MAPPINGS=[ti]}class Ni extends Ja{static MODEL_CLASS_MAPPINGS=[ri]}class Oi extends Ja{static MODEL_CLASS_MAPPINGS=[ni]}class Di extends Ja{static MODEL_CLASS_MAPPINGS=[ii]}class Vi extends Ja{static MODEL_CLASS_MAPPINGS=[li]}class ji extends Ja{static MODEL_CLASS_MAPPINGS=[ci]}class Ri extends Ja{static MODEL_CLASS_MAPPINGS=[di]}class Gi extends Ja{static MODEL_CLASS_MAPPINGS=[hi]}class qi extends Ja{static MODEL_CLASS_MAPPINGS=[fi]}class $i extends Ja{static MODEL_CLASS_MAPPINGS=[gi]}class Wi extends Ja{static MODEL_CLASS_MAPPINGS=[Mi]}class Ui extends Ja{static MODEL_CLASS_MAPPINGS=[mi]}class Xi extends Ja{static MODEL_CLASS_MAPPINGS=[_i]}class Qi extends Ja{static MODEL_CLASS_MAPPINGS=[wi]}class Hi extends Ja{static MODEL_CLASS_MAPPINGS=[Ti]}class Yi extends Ja{static MODEL_CLASS_MAPPINGS=[bi]}class Ji extends Ja{static MODEL_CLASS_MAPPINGS=[xi]}class Ki extends Ja{static MODEL_CLASS_MAPPINGS=[ki]}class Zi extends Ja{static MODEL_CLASS_MAPPINGS=[pi]}class el extends Ja{static MODEL_CLASS_MAPPINGS=[yi]}class tl extends Ja{static MODEL_CLASS_MAPPINGS=[Fi]}class rl extends Ja{static MODEL_CLASS_MAPPINGS=[Pi]}class nl extends Ja{static MODEL_CLASS_MAPPINGS=[Ci]}class ol extends Ja{static MODEL_CLASS_MAPPINGS=[vi]}class sl extends G{constructor({logits:e,past_key_values:t,encoder_outputs:r,decoder_attentions:n=null,cross_attentions:o=null}){super(),this.logits=e,this.past_key_values=t,this.encoder_outputs=r,this.decoder_attentions=n,this.cross_attentions=o}}class al extends G{constructor({logits:e}){super(),this.logits=e}}class il extends G{constructor({logits:e,embeddings:t}){super(),this.logits=e,this.embeddings=t}}class ll extends G{constructor({logits:e}){super(),this.logits=e}}class cl extends G{constructor({logits:e}){super(),this.logits=e}}class dl extends G{constructor({start_logits:e,end_logits:t}){super(),this.start_logits=e,this.end_logits=t}}class ul extends G{constructor({logits:e}){super(),this.logits=e}}class pl extends G{constructor({logits:e,past_key_values:t}){super(),this.logits=e,this.past_key_values=t}}class hl extends G{constructor({alphas:e}){super(),this.alphas=e}}class ml extends G{constructor({waveform:e,spectrogram:t}){super(),this.waveform=e,this.spectrogram=t}}},"./src/models/whisper/common_whisper.js":
+  \***********************/(e,t,r)=>{r.r(t),r.d(t,{ASTForAudioClassification:()=>rr,ASTModel:()=>tr,ASTPreTrainedModel:()=>er,AlbertForMaskedLM:()=>ut,AlbertForQuestionAnswering:()=>dt,AlbertForSequenceClassification:()=>ct,AlbertModel:()=>lt,AlbertPreTrainedModel:()=>it,AutoModel:()=>Ei,AutoModelForAudioClassification:()=>Yi,AutoModelForAudioFrameClassification:()=>Ki,AutoModelForCTC:()=>Hi,AutoModelForCausalLM:()=>Di,AutoModelForDepthEstimation:()=>rl,AutoModelForDocumentQuestionAnswering:()=>Zi,AutoModelForImageClassification:()=>Gi,AutoModelForImageFeatureExtraction:()=>ol,AutoModelForImageMatting:()=>el,AutoModelForImageSegmentation:()=>qi,AutoModelForImageToImage:()=>tl,AutoModelForMaskGeneration:()=>Qi,AutoModelForMaskedLM:()=>Vi,AutoModelForNormalEstimation:()=>nl,AutoModelForObjectDetection:()=>Ui,AutoModelForQuestionAnswering:()=>ji,AutoModelForSemanticSegmentation:()=>$i,AutoModelForSeq2SeqLM:()=>Ii,AutoModelForSequenceClassification:()=>Li,AutoModelForSpeechSeq2Seq:()=>Bi,AutoModelForTextToSpectrogram:()=>Ni,AutoModelForTextToWaveform:()=>Oi,AutoModelForTokenClassification:()=>zi,AutoModelForUniversalSegmentation:()=>Wi,AutoModelForVision2Seq:()=>Ri,AutoModelForXVector:()=>Ji,AutoModelForZeroShotObjectDetection:()=>Xi,BartForConditionalGeneration:()=>kt,BartForSequenceClassification:()=>yt,BartModel:()=>xt,BartPretrainedModel:()=>bt,BaseModelOutput:()=>q,BeitForImageClassification:()=>ao,BeitModel:()=>so,BeitPreTrainedModel:()=>oo,BertForMaskedLM:()=>U,BertForQuestionAnswering:()=>H,BertForSequenceClassification:()=>X,BertForTokenClassification:()=>Q,BertModel:()=>W,BertPreTrainedModel:()=>$,BlenderbotForConditionalGeneration:()=>Lt,BlenderbotModel:()=>Et,BlenderbotPreTrainedModel:()=>At,BlenderbotSmallForConditionalGeneration:()=>Bt,BlenderbotSmallModel:()=>It,BlenderbotSmallPreTrainedModel:()=>zt,BloomForCausalLM:()=>bn,BloomModel:()=>Tn,BloomPreTrainedModel:()=>wn,CLIPModel:()=>hr,CLIPPreTrainedModel:()=>pr,CLIPSegForImageSegmentation:()=>Pr,CLIPSegModel:()=>Fr,CLIPSegPreTrainedModel:()=>yr,CLIPTextModel:()=>mr,CLIPTextModelWithProjection:()=>_r,CLIPVisionModel:()=>fr,CLIPVisionModelWithProjection:()=>gr,CamembertForMaskedLM:()=>Me,CamembertForQuestionAnswering:()=>be,CamembertForSequenceClassification:()=>we,CamembertForTokenClassification:()=>Te,CamembertModel:()=>ge,CamembertPreTrainedModel:()=>fe,CausalLMOutput:()=>ul,CausalLMOutputWithPast:()=>pl,ChineseCLIPModel:()=>kr,ChineseCLIPPreTrainedModel:()=>xr,ClapAudioModelWithProjection:()=>Ta,ClapModel:()=>Ma,ClapPreTrainedModel:()=>ga,ClapTextModelWithProjection:()=>wa,CodeGenForCausalLM:()=>Xr,CodeGenModel:()=>Ur,CodeGenPreTrainedModel:()=>Wr,CohereForCausalLM:()=>Zr,CohereModel:()=>Kr,CoherePreTrainedModel:()=>Jr,ConvBertForMaskedLM:()=>ae,ConvBertForQuestionAnswering:()=>ce,ConvBertForSequenceClassification:()=>ie,ConvBertForTokenClassification:()=>le,ConvBertModel:()=>se,ConvBertPreTrainedModel:()=>oe,ConvNextForImageClassification:()=>rs,ConvNextModel:()=>ts,ConvNextPreTrainedModel:()=>es,ConvNextV2ForImageClassification:()=>ss,ConvNextV2Model:()=>os,ConvNextV2PreTrainedModel:()=>ns,DPTForDepthEstimation:()=>Vo,DPTModel:()=>Do,DPTPreTrainedModel:()=>Oo,DebertaForMaskedLM:()=>ye,DebertaForQuestionAnswering:()=>Ce,DebertaForSequenceClassification:()=>Fe,DebertaForTokenClassification:()=>Pe,DebertaModel:()=>ke,DebertaPreTrainedModel:()=>xe,DebertaV2ForMaskedLM:()=>Ae,DebertaV2ForQuestionAnswering:()=>ze,DebertaV2ForSequenceClassification:()=>Ee,DebertaV2ForTokenClassification:()=>Le,DebertaV2Model:()=>Se,DebertaV2PreTrainedModel:()=>ve,DecisionTransformerModel:()=>Ya,DecisionTransformerPreTrainedModel:()=>Ha,DeiTForImageClassification:()=>yo,DeiTModel:()=>ko,DeiTPreTrainedModel:()=>xo,DepthAnythingForDepthEstimation:()=>Ro,DepthAnythingPreTrainedModel:()=>jo,DetrForObjectDetection:()=>co,DetrForSegmentation:()=>uo,DetrModel:()=>lo,DetrObjectDetectionOutput:()=>po,DetrPreTrainedModel:()=>io,DetrSegmentationOutput:()=>ho,Dinov2ForImageClassification:()=>ls,Dinov2Model:()=>is,Dinov2PreTrainedModel:()=>as,DistilBertForMaskedLM:()=>Ve,DistilBertForQuestionAnswering:()=>De,DistilBertForSequenceClassification:()=>Ne,DistilBertForTokenClassification:()=>Oe,DistilBertModel:()=>Be,DistilBertPreTrainedModel:()=>Ie,DonutSwinModel:()=>Zo,DonutSwinPreTrainedModel:()=>Ko,EfficientNetForImageClassification:()=>La,EfficientNetModel:()=>Ea,EfficientNetPreTrainedModel:()=>Aa,ElectraForMaskedLM:()=>pe,ElectraForQuestionAnswering:()=>_e,ElectraForSequenceClassification:()=>he,ElectraForTokenClassification:()=>me,ElectraModel:()=>ue,ElectraPreTrainedModel:()=>de,EsmForMaskedLM:()=>Ge,EsmForSequenceClassification:()=>qe,EsmForTokenClassification:()=>$e,EsmModel:()=>Re,EsmPreTrainedModel:()=>je,FalconForCausalLM:()=>fa,FalconModel:()=>_a,FalconPreTrainedModel:()=>ma,FastViTForImageClassification:()=>qn,FastViTModel:()=>Gn,FastViTPreTrainedModel:()=>Rn,Florence2ForConditionalGeneration:()=>ur,Florence2PreTrainedModel:()=>dr,GLPNForDepthEstimation:()=>Jo,GLPNModel:()=>Yo,GLPNPreTrainedModel:()=>Ho,GPT2LMHeadModel:()=>Sr,GPT2Model:()=>vr,GPT2PreTrainedModel:()=>Cr,GPTBigCodeForCausalLM:()=>$r,GPTBigCodeModel:()=>qr,GPTBigCodePreTrainedModel:()=>Gr,GPTJForCausalLM:()=>Rr,GPTJModel:()=>jr,GPTJPreTrainedModel:()=>Vr,GPTNeoForCausalLM:()=>Br,GPTNeoModel:()=>Ir,GPTNeoPreTrainedModel:()=>zr,GPTNeoXForCausalLM:()=>Dr,GPTNeoXModel:()=>Or,GPTNeoXPreTrainedModel:()=>Nr,Gemma2ForCausalLM:()=>sn,Gemma2Model:()=>on,Gemma2PreTrainedModel:()=>nn,GemmaForCausalLM:()=>rn,GemmaModel:()=>tn,GemmaPreTrainedModel:()=>en,GroupViTModel:()=>jn,GroupViTPreTrainedModel:()=>Vn,HieraForImageClassification:()=>Co,HieraModel:()=>Po,HieraPreTrainedModel:()=>Fo,HubertForCTC:()=>Xs,HubertForSequenceClassification:()=>Qs,HubertModel:()=>Us,HubertPreTrainedModel:()=>Ws,ImageMattingOutput:()=>hl,JAISLMHeadModel:()=>Lr,JAISModel:()=>Er,JAISPreTrainedModel:()=>Ar,LlamaForCausalLM:()=>Yr,LlamaModel:()=>Hr,LlamaPreTrainedModel:()=>Qr,LlavaForConditionalGeneration:()=>lr,LlavaPreTrainedModel:()=>ir,LongT5ForConditionalGeneration:()=>gt,LongT5Model:()=>ft,LongT5PreTrainedModel:()=>_t,M2M100ForConditionalGeneration:()=>bs,M2M100Model:()=>Ts,M2M100PreTrainedModel:()=>ws,MBartForCausalLM:()=>St,MBartForConditionalGeneration:()=>Ct,MBartForSequenceClassification:()=>vt,MBartModel:()=>Pt,MBartPreTrainedModel:()=>Ft,MPNetForMaskedLM:()=>Ke,MPNetForQuestionAnswering:()=>tt,MPNetForSequenceClassification:()=>Ze,MPNetForTokenClassification:()=>et,MPNetModel:()=>Je,MPNetPreTrainedModel:()=>Ye,MT5ForConditionalGeneration:()=>Tt,MT5Model:()=>wt,MT5PreTrainedModel:()=>Mt,MarianMTModel:()=>Ms,MarianModel:()=>gs,MarianPreTrainedModel:()=>fs,MaskFormerForInstanceSegmentation:()=>Qo,MaskFormerModel:()=>Xo,MaskFormerPreTrainedModel:()=>Uo,MaskedLMOutput:()=>cl,MistralForCausalLM:()=>da,MistralModel:()=>ca,MistralPreTrainedModel:()=>la,MobileBertForMaskedLM:()=>Xe,MobileBertForQuestionAnswering:()=>He,MobileBertForSequenceClassification:()=>Qe,MobileBertModel:()=>Ue,MobileBertPreTrainedModel:()=>We,MobileNetV1ForImageClassification:()=>Va,MobileNetV1Model:()=>Da,MobileNetV1PreTrainedModel:()=>Oa,MobileNetV2ForImageClassification:()=>Ga,MobileNetV2Model:()=>Ra,MobileNetV2PreTrainedModel:()=>ja,MobileNetV3ForImageClassification:()=>Wa,MobileNetV3Model:()=>$a,MobileNetV3PreTrainedModel:()=>qa,MobileNetV4ForImageClassification:()=>Qa,MobileNetV4Model:()=>Xa,MobileNetV4PreTrainedModel:()=>Ua,MobileViTForImageClassification:()=>Qn,MobileViTModel:()=>Xn,MobileViTPreTrainedModel:()=>Un,MobileViTV2ForImageClassification:()=>Jn,MobileViTV2Model:()=>Yn,MobileViTV2PreTrainedModel:()=>Hn,ModelOutput:()=>G,Moondream1ForConditionalGeneration:()=>cr,MptForCausalLM:()=>yn,MptModel:()=>kn,MptPreTrainedModel:()=>xn,MusicgenForCausalLM:()=>Ba,MusicgenForConditionalGeneration:()=>Na,MusicgenModel:()=>Ia,MusicgenPreTrainedModel:()=>za,NomicBertModel:()=>J,NomicBertPreTrainedModel:()=>Y,OPTForCausalLM:()=>Cn,OPTModel:()=>Pn,OPTPreTrainedModel:()=>Fn,OpenELMForCausalLM:()=>cn,OpenELMModel:()=>ln,OpenELMPreTrainedModel:()=>an,OwlViTForObjectDetection:()=>eo,OwlViTModel:()=>Zn,OwlViTPreTrainedModel:()=>Kn,Owlv2ForObjectDetection:()=>no,Owlv2Model:()=>ro,Owlv2PreTrainedModel:()=>to,Phi3ForCausalLM:()=>Mn,Phi3Model:()=>gn,Phi3PreTrainedModel:()=>fn,PhiForCausalLM:()=>_n,PhiModel:()=>mn,PhiPreTrainedModel:()=>hn,PreTrainedModel:()=>R,PretrainedMixin:()=>Ja,PvtForImageClassification:()=>zn,PvtModel:()=>Ln,PvtPreTrainedModel:()=>En,PyAnnoteForAudioFrameClassification:()=>Ss,PyAnnoteModel:()=>vs,PyAnnotePreTrainedModel:()=>Cs,QuestionAnsweringModelOutput:()=>dl,Qwen2ForCausalLM:()=>pn,Qwen2Model:()=>un,Qwen2PreTrainedModel:()=>dn,RTDetrForObjectDetection:()=>fo,RTDetrModel:()=>_o,RTDetrObjectDetectionOutput:()=>go,RTDetrPreTrainedModel:()=>mo,ResNetForImageClassification:()=>Ao,ResNetModel:()=>So,ResNetPreTrainedModel:()=>vo,RoFormerForMaskedLM:()=>ee,RoFormerForQuestionAnswering:()=>ne,RoFormerForSequenceClassification:()=>te,RoFormerForTokenClassification:()=>re,RoFormerModel:()=>Z,RoFormerPreTrainedModel:()=>K,RobertaForMaskedLM:()=>Dt,RobertaForQuestionAnswering:()=>Rt,RobertaForSequenceClassification:()=>Vt,RobertaForTokenClassification:()=>jt,RobertaModel:()=>Ot,RobertaPreTrainedModel:()=>Nt,SamImageSegmentationOutput:()=>_s,SamModel:()=>ms,SamPreTrainedModel:()=>hs,SapiensForDepthEstimation:()=>$o,SapiensForNormalEstimation:()=>Wo,SapiensForSemanticSegmentation:()=>qo,SapiensPreTrainedModel:()=>Go,SegformerForImageClassification:()=>Fa,SegformerForSemanticSegmentation:()=>Pa,SegformerModel:()=>ya,SegformerPreTrainedModel:()=>ka,Seq2SeqLMOutput:()=>sl,SequenceClassifierOutput:()=>al,SiglipModel:()=>wr,SiglipPreTrainedModel:()=>Mr,SiglipTextModel:()=>Tr,SiglipVisionModel:()=>br,SpeechT5ForSpeechToText:()=>na,SpeechT5ForTextToSpeech:()=>oa,SpeechT5HifiGan:()=>sa,SpeechT5Model:()=>ra,SpeechT5PreTrainedModel:()=>ta,SqueezeBertForMaskedLM:()=>ot,SqueezeBertForQuestionAnswering:()=>at,SqueezeBertForSequenceClassification:()=>st,SqueezeBertModel:()=>nt,SqueezeBertPreTrainedModel:()=>rt,StableLmForCausalLM:()=>Sa,StableLmModel:()=>va,StableLmPreTrainedModel:()=>Ca,Starcoder2ForCausalLM:()=>ha,Starcoder2Model:()=>pa,Starcoder2PreTrainedModel:()=>ua,Swin2SRForImageSuperResolution:()=>No,Swin2SRModel:()=>Bo,Swin2SRPreTrainedModel:()=>Io,SwinForImageClassification:()=>zo,SwinModel:()=>Lo,SwinPreTrainedModel:()=>Eo,T5ForConditionalGeneration:()=>mt,T5Model:()=>ht,T5PreTrainedModel:()=>pt,TableTransformerForObjectDetection:()=>To,TableTransformerModel:()=>wo,TableTransformerObjectDetectionOutput:()=>bo,TableTransformerPreTrainedModel:()=>Mo,TokenClassifierOutput:()=>ll,TrOCRForCausalLM:()=>ia,TrOCRPreTrainedModel:()=>aa,UniSpeechForCTC:()=>Is,UniSpeechForSequenceClassification:()=>Bs,UniSpeechModel:()=>zs,UniSpeechPreTrainedModel:()=>Ls,UniSpeechSatForAudioFrameClassification:()=>js,UniSpeechSatForCTC:()=>Ds,UniSpeechSatForSequenceClassification:()=>Vs,UniSpeechSatModel:()=>Os,UniSpeechSatPreTrainedModel:()=>Ns,ViTForImageClassification:()=>An,ViTMAEModel:()=>Bn,ViTMAEPreTrainedModel:()=>In,ViTMSNForImageClassification:()=>Dn,ViTMSNModel:()=>On,ViTMSNPreTrainedModel:()=>Nn,ViTModel:()=>Sn,ViTPreTrainedModel:()=>vn,VisionEncoderDecoderModel:()=>ar,VitMatteForImageMatting:()=>Wn,VitMattePreTrainedModel:()=>$n,VitsModel:()=>xa,VitsModelOutput:()=>ml,VitsPreTrainedModel:()=>ba,Wav2Vec2BertForCTC:()=>qs,Wav2Vec2BertForSequenceClassification:()=>$s,Wav2Vec2BertModel:()=>Gs,Wav2Vec2BertPreTrainedModel:()=>Rs,Wav2Vec2ForAudioFrameClassification:()=>Ps,Wav2Vec2ForCTC:()=>ys,Wav2Vec2ForSequenceClassification:()=>Fs,Wav2Vec2Model:()=>ks,Wav2Vec2PreTrainedModel:()=>xs,WavLMForAudioFrameClassification:()=>ea,WavLMForCTC:()=>Js,WavLMForSequenceClassification:()=>Ks,WavLMForXVector:()=>Zs,WavLMModel:()=>Ys,WavLMPreTrainedModel:()=>Hs,WeSpeakerResNetModel:()=>Es,WeSpeakerResNetPreTrainedModel:()=>As,WhisperForConditionalGeneration:()=>sr,WhisperModel:()=>or,WhisperPreTrainedModel:()=>nr,XLMForQuestionAnswering:()=>Xt,XLMForSequenceClassification:()=>Wt,XLMForTokenClassification:()=>Ut,XLMModel:()=>qt,XLMPreTrainedModel:()=>Gt,XLMRobertaForMaskedLM:()=>Yt,XLMRobertaForQuestionAnswering:()=>Zt,XLMRobertaForSequenceClassification:()=>Jt,XLMRobertaForTokenClassification:()=>Kt,XLMRobertaModel:()=>Ht,XLMRobertaPreTrainedModel:()=>Qt,XLMWithLMHeadModel:()=>$t,XVectorOutput:()=>il,YolosForObjectDetection:()=>us,YolosModel:()=>ds,YolosObjectDetectionOutput:()=>ps,YolosPreTrainedModel:()=>cs});var n=r(/*! ./configs.js */"./src/configs.js"),o=r(/*! ./backends/onnx.js */"./src/backends/onnx.js"),s=r(/*! ./utils/dtypes.js */"./src/utils/dtypes.js"),a=r(/*! ./utils/generic.js */"./src/utils/generic.js"),i=r(/*! ./utils/core.js */"./src/utils/core.js"),l=r(/*! ./utils/hub.js */"./src/utils/hub.js"),c=r(/*! ./generation/logits_process.js */"./src/generation/logits_process.js"),d=r(/*! ./generation/configuration_utils.js */"./src/generation/configuration_utils.js"),u=r(/*! ./utils/tensor.js */"./src/utils/tensor.js"),p=r(/*! ./utils/maths.js */"./src/utils/maths.js"),h=r(/*! ./generation/stopping_criteria.js */"./src/generation/stopping_criteria.js"),m=r(/*! ./generation/logits_sampler.js */"./src/generation/logits_sampler.js"),_=r(/*! ./env.js */"./src/env.js"),f=r(/*! ./models/whisper/generation_whisper.js */"./src/models/whisper/generation_whisper.js"),g=r(/*! ./models/whisper/common_whisper.js */"./src/models/whisper/common_whisper.js");const M=0,w=1,T=2,b=3,x=4,k=5,y=6,F=7,P=new Map,C=new Map,v=new Map;async function S(e,t,r){return Object.fromEntries(await Promise.all(Object.keys(t).map((async a=>{const{buffer:i,session_options:c}=await async function(e,t,r){const a=r.config?.["transformers.js_config"]??{};let i=r.device??a.device;i&&"string"!=typeof i&&(i.hasOwnProperty(t)?i=i[t]:(console.warn(`device not specified for "${t}". Using the default device.`),i=null));const c=i??(_.apis.IS_NODE_ENV?"cpu":"wasm"),d=(0,o.deviceToExecutionProviders)(c);let u=r.dtype??a.dtype;"string"!=typeof u&&(u&&u.hasOwnProperty(t)?u=u[t]:(u=s.DEFAULT_DEVICE_DTYPE_MAPPING[c]??s.DATA_TYPES.fp32,console.warn(`dtype not specified for "${t}". Using the default dtype (${u}) for this device (${c}).`)));const p=u;if(!s.DEFAULT_DTYPE_SUFFIX_MAPPING.hasOwnProperty(p))throw new Error(`Invalid dtype: ${p}. Should be one of: ${Object.keys(s.DATA_TYPES).join(", ")}`);if(p===s.DATA_TYPES.fp16&&"webgpu"===c&&!await(0,s.isWebGpuFp16Supported)())throw new Error(`The device (${c}) does not support fp16.`);const h=s.DEFAULT_DTYPE_SUFFIX_MAPPING[p],m=`${r.subfolder??""}/${t}${h}.onnx`,f={...r.session_options}??{};f.executionProviders??=d;const g=a.free_dimension_overrides;g?f.freeDimensionOverrides??=g:c.startsWith("webnn")&&!f.freeDimensionOverrides&&console.warn('WebNN does not currently support dynamic shapes and requires `free_dimension_overrides` to be set in config.json as a field within "transformers.js_config". When `free_dimension_overrides` is not set, you may experience significant performance degradation.');const M=(0,l.getModelFile)(e,m,!0,r);let w=[];if(r.use_external_data_format&&(!0===r.use_external_data_format||"object"==typeof r.use_external_data_format&&r.use_external_data_format.hasOwnProperty(t)&&!0===r.use_external_data_format[t])){if(_.apis.IS_NODE_ENV)throw new Error("External data format is not yet supported in Node.js");const n=`${t}${h}.onnx_data`,o=`${r.subfolder??""}/${n}`;w.push(new Promise((async(t,s)=>{const a=await(0,l.getModelFile)(e,o,!0,r);t({path:n,data:a})})))}else void 0!==f.externalData&&(w=f.externalData.map((async t=>{if("string"==typeof t.data){const n=await(0,l.getModelFile)(e,t.data,!0,r);return{...t,data:n}}return t})));if(w.length>0&&(f.externalData=await Promise.all(w)),"webgpu"===c){const e=(0,n.getKeyValueShapes)(r.config,{prefix:"present"});if(Object.keys(e).length>0&&!(0,o.isONNXProxy)()){const t={};for(const r in e)r.includes("encoder")||(t[r]="gpu-buffer");f.preferredOutputLocation=t}}return{buffer:await M,session_options:f}}(e,t[a],r);return[a,await(0,o.createInferenceSession)(i,c)]}))))}async function A(e,t){const r=function(e,t){const r=Object.create(null),n=[];for(const s of e.inputNames){const e=t[s];e instanceof u.Tensor?r[s]=(0,o.isONNXProxy)()?e.clone():e:n.push(s)}if(n.length>0)throw new Error(`An error occurred during model execution: "Missing the following inputs: ${n.join(", ")}.`);const s=Object.keys(t).length,a=e.inputNames.length;if(s>a){let r=Object.keys(t).filter((t=>!e.inputNames.includes(t)));console.warn(`WARNING: Too many inputs were provided (${s} > ${a}). The following inputs will be ignored: "${r.join(", ")}".`)}return r}(e,t);try{const t=Object.fromEntries(Object.entries(r).map((([e,t])=>[e,t.ort_tensor])));let n=await e.run(t);return n=E(n),n}catch(e){throw console.error(`An error occurred during model execution: "${e}".`),console.error("Inputs given to model:",r),e}}function E(e){for(let t in e)(0,o.isONNXTensor)(e[t])?e[t]=new u.Tensor(e[t]):"object"==typeof e[t]&&E(e[t]);return e}function L(e){if(e instanceof u.Tensor)return e;if(0===e.length)throw Error("items must be non-empty");if(Array.isArray(e[0])){if(e.some((t=>t.length!==e[0].length)))throw Error("Unable to create tensor, you should probably activate truncation and/or padding with 'padding=True' and/or 'truncation=True' to have batched tensors with the same length.");return new u.Tensor("int64",BigInt64Array.from(e.flat().map((e=>BigInt(e)))),[e.length,e[0].length])}return new u.Tensor("int64",BigInt64Array.from(e.map((e=>BigInt(e)))),[1,e.length])}function z(e){return new u.Tensor("bool",[e],[1])}async function I(e,t){let{encoder_outputs:r,input_ids:n,decoder_input_ids:o,...s}=t;if(!r){const n=(0,i.pick)(t,e.sessions.model.inputNames);r=(await B(e,n)).last_hidden_state}s.input_ids=o,s.encoder_hidden_states=r,e.sessions.decoder_model_merged.inputNames.includes("encoder_attention_mask")&&(s.encoder_attention_mask=t.attention_mask);return await N(e,s,!0)}async function B(e,t){const r=e.sessions.model,n=(0,i.pick)(t,r.inputNames);if(r.inputNames.includes("inputs_embeds")&&!n.inputs_embeds){if(!t.input_ids)throw new Error("Both `input_ids` and `inputs_embeds` are missing in the model inputs.");n.inputs_embeds=await e.encode_text({input_ids:t.input_ids})}return r.inputNames.includes("token_type_ids")&&!n.token_type_ids&&(n.token_type_ids=new u.Tensor("int64",new BigInt64Array(n.input_ids.data.length),n.input_ids.dims)),await A(r,n)}async function N(e,t,r=!1){const n=e.sessions[r?"decoder_model_merged":"model"],{past_key_values:o,...s}=t;n.inputNames.includes("use_cache_branch")&&(s.use_cache_branch=z(!!o)),n.inputNames.includes("position_ids")&&s.attention_mask&&!s.position_ids&&(s.position_ids=function(e,t=null){const{input_ids:r,inputs_embeds:n,attention_mask:o}=e,[s,a]=o.dims,i=new BigInt64Array(o.data.length);for(let e=0;e<s;++e){const t=e*a;let r=BigInt(0);for(let e=0;e<a;++e){const n=t+e;0n===o.data[n]?i[n]=BigInt(1):(i[n]=r,r+=o.data[n])}}let l=new u.Tensor("int64",i,o.dims);if(t){const e=-(r??n).dims.at(1);l=l.slice(null,[e,null])}return l}(s,o)),e.addPastKeyValues(s,o);const a=(0,i.pick)(s,n.inputNames);return await A(n,a)}async function O(e,{input_ids:t=null,attention_mask:r=null,pixel_values:n=null,position_ids:o=null,inputs_embeds:s=null,past_key_values:a=null,generation_config:i=null,logits_processor:l=null,...c}){if(!s)if(s=await e.encode_text({input_ids:t}),n&&1!==t.dims[1]){const o=await e.encode_image({pixel_values:n});({inputs_embeds:s,attention_mask:r}=e._merge_input_ids_with_image_features({image_features:o,inputs_embeds:s,input_ids:t,attention_mask:r}))}else if(a&&n&&1===t.dims[1]){const e=t.dims[1],n=Object.values(a)[0].dims.at(-2);r=(0,u.cat)([(0,u.ones)([t.dims[0],n]),r.slice(null,[r.dims[1]-e,r.dims[1]])],1)}return await N(e,{inputs_embeds:s,past_key_values:a,attention_mask:r,position_ids:o,generation_config:i,logits_processor:l},!0)}function D(e,t,r,n){if(r.past_key_values){const t=Object.values(r.past_key_values)[0].dims.at(-2),{input_ids:n,attention_mask:o}=r;if(o&&o.dims[1]>n.dims[1]);else if(t<n.dims[1])r.input_ids=n.slice(null,[t,null]);else if(null!=e.config.image_token_index&&n.data.some((t=>t==e.config.image_token_index))){const o=e.config.num_image_tokens;if(!o)throw new Error("`num_image_tokens` is missing in the model configuration.");const s=n.dims[1]-(t-o);r.input_ids=n.slice(null,[-s,null]),r.attention_mask=(0,u.ones)([1,t+s])}}return r}function V(e,t,r,n){return r.past_key_values&&(t=t.map((e=>[e.at(-1)]))),{...r,decoder_input_ids:L(t)}}function j(e,...t){return e.config.is_encoder_decoder?V(e,...t):D(e,...t)}class R extends a.Callable{main_input_name="input_ids";forward_params=["input_ids","attention_mask"];constructor(e,t){super(),this.config=e,this.sessions=t;const r=v.get(this.constructor),n=P.get(r);switch(this.can_generate=!1,this._forward=null,this._prepare_inputs_for_generation=null,n){case x:this.can_generate=!0,this._forward=N,this._prepare_inputs_for_generation=D;break;case T:case b:case F:this.can_generate=!0,this._forward=I,this._prepare_inputs_for_generation=V;break;case w:this._forward=I;break;case y:this.can_generate=!0,this._forward=O,this._prepare_inputs_for_generation=j;break;default:this._forward=B}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:r=null,cache_dir:o=null,local_files_only:s=!1,revision:a="main",model_file_name:i=null,subfolder:c="onnx",device:d=null,dtype:u=null,use_external_data_format:p=null,session_options:h={}}={}){let m={progress_callback:t,config:r,cache_dir:o,local_files_only:s,revision:a,model_file_name:i,subfolder:c,device:d,dtype:u,use_external_data_format:p,session_options:h};const _=v.get(this),f=P.get(_);let g;if(r=m.config=await n.AutoConfig.from_pretrained(e,m),f===x)g=await Promise.all([S(e,{model:m.model_file_name??"model"},m),(0,l.getModelJSON)(e,"generation_config.json",!1,m)]);else if(f===T||f===b)g=await Promise.all([S(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},m),(0,l.getModelJSON)(e,"generation_config.json",!1,m)]);else if(f===k)g=await Promise.all([S(e,{model:"vision_encoder",prompt_encoder_mask_decoder:"prompt_encoder_mask_decoder"},m)]);else if(f===w)g=await Promise.all([S(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},m)]);else if(f===y){const t={embed_tokens:"embed_tokens",vision_encoder:"vision_encoder",decoder_model_merged:"decoder_model_merged"};r.is_encoder_decoder&&(t.model="encoder_model"),g=await Promise.all([S(e,t,m),(0,l.getModelJSON)(e,"generation_config.json",!1,m)])}else f===F?g=await Promise.all([S(e,{model:"text_encoder",decoder_model_merged:"decoder_model_merged",encodec_decode:"encodec_decode"},m),(0,l.getModelJSON)(e,"generation_config.json",!1,m)]):(f!==M&&console.warn(`Model type for '${_??r?.model_type}' not found, assuming encoder-only architecture. Please report this at https://github.com/xenova/transformers.js/issues/new/choose.`),g=await Promise.all([S(e,{model:m.model_file_name??"model"},m)]));return new this(r,...g)}async _call(e){return await this.forward(e)}async forward(e){return await this._forward(this,e)}_get_logits_warper(e){const t=new c.LogitsProcessorList;return null!==e.temperature&&1!==e.temperature&&t.push(new c.TemperatureLogitsWarper(e.temperature)),null!==e.top_k&&0!==e.top_k&&t.push(new c.TopKLogitsWarper(e.top_k)),null!==e.top_p&&e.top_p<1&&t.push(new c.TopPLogitsWarper(e.top_p)),t}_get_logits_processor(e,t,r=null){const n=new c.LogitsProcessorList;if(null!==e.repetition_penalty&&1!==e.repetition_penalty&&n.push(new c.RepetitionPenaltyLogitsProcessor(e.repetition_penalty)),null!==e.no_repeat_ngram_size&&e.no_repeat_ngram_size>0&&n.push(new c.NoRepeatNGramLogitsProcessor(e.no_repeat_ngram_size)),null!==e.bad_words_ids&&n.push(new c.NoBadWordsLogitsProcessor(e.bad_words_ids,e.eos_token_id)),null!==e.min_length&&null!==e.eos_token_id&&e.min_length>0&&n.push(new c.MinLengthLogitsProcessor(e.min_length,e.eos_token_id)),null!==e.min_new_tokens&&null!==e.eos_token_id&&e.min_new_tokens>0&&n.push(new c.MinNewTokensLengthLogitsProcessor(t,e.min_new_tokens,e.eos_token_id)),null!==e.forced_bos_token_id&&n.push(new c.ForcedBOSTokenLogitsProcessor(e.forced_bos_token_id)),null!==e.forced_eos_token_id&&n.push(new c.ForcedEOSTokenLogitsProcessor(e.max_length,e.forced_eos_token_id)),null!==e.begin_suppress_tokens){const r=t>1||null===e.forced_bos_token_id?t:t+1;n.push(new c.SuppressTokensAtBeginLogitsProcessor(e.begin_suppress_tokens,r))}return null!==e.guidance_scale&&e.guidance_scale>1&&n.push(new c.ClassifierFreeGuidanceLogitsProcessor(e.guidance_scale)),null!==r&&n.extend(r),n}_prepare_generation_config(e,t,r=d.GenerationConfig){const n={...this.config};for(const e of["decoder","generator","text_config"])e in n&&Object.assign(n,n[e]);const o=new r(n);return"generation_config"in this&&Object.assign(o,this.generation_config),e&&Object.assign(o,e),t&&Object.assign(o,(0,i.pick)(t,Object.getOwnPropertyNames(o))),o}_get_stopping_criteria(e,t=null){const r=new h.StoppingCriteriaList;return null!==e.max_length&&r.push(new h.MaxLengthCriteria(e.max_length,this.config.max_position_embeddings??null)),null!==e.eos_token_id&&r.push(new h.EosTokenCriteria(e.eos_token_id)),t&&r.extend(t),r}_validate_model_class(){if(!this.can_generate){const e=[ii,di,ai,ti],t=v.get(this.constructor),r=new Set,n=this.config.model_type;for(const t of e){const e=t.get(n);e&&r.add(e[0])}let o=`The current model class (${t}) is not compatible with \`.generate()\`, as it doesn't have a language model head.`;throw r.size>0&&(o+=` Please use the following class instead: ${[...r].join(", ")}`),Error(o)}}prepare_inputs_for_generation(...e){return this._prepare_inputs_for_generation(this,...e)}_update_model_kwargs_for_generation({generated_input_ids:e,outputs:t,model_inputs:r,is_encoder_decoder:n}){return r.past_key_values=this.getPastKeyValues(t,r.past_key_values),r.input_ids=new u.Tensor("int64",e.flat(),[e.length,1]),n||(r.attention_mask=(0,u.cat)([r.attention_mask,(0,u.ones)([r.attention_mask.dims[0],1])],1)),r.position_ids=null,r}_prepare_model_inputs({inputs:e,bos_token_id:t,model_kwargs:r}){const n=(0,i.pick)(r,this.forward_params),o=this.main_input_name;if(o in n){if(e)throw new Error("`inputs`: {inputs}` were passed alongside {input_name} which is not allowed. Make sure to either pass {inputs} or {input_name}=...")}else n[o]=e;return{inputs_tensor:n[o],model_inputs:n,model_input_name:o}}async _prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:e,model_inputs:t,model_input_name:r,generation_config:n}){if(this.sessions.model.inputNames.includes("inputs_embeds")&&!t.inputs_embeds&&"_prepare_inputs_embeds"in this){const{input_ids:e,pixel_values:r,attention_mask:n,...o}=t,s=await this._prepare_inputs_embeds(t);t={...o,...(0,i.pick)(s,["inputs_embeds","attention_mask"])}}let{last_hidden_state:o}=await B(this,t);if(null!==n.guidance_scale&&n.guidance_scale>1)o=(0,u.cat)([o,(0,u.full_like)(o,0)],0),"attention_mask"in t&&(t.attention_mask=(0,u.cat)([t.attention_mask,(0,u.zeros_like)(t.attention_mask)],0));else if(t.decoder_input_ids){const e=L(t.decoder_input_ids).dims[0];if(e!==o.dims[0]){if(1!==o.dims[0])throw new Error(`The encoder outputs have a different batch size (${o.dims[0]}) than the decoder inputs (${e}).`);o=(0,u.cat)(Array.from({length:e},(()=>o)),0)}}return t.encoder_outputs=o,t}_prepare_decoder_input_ids_for_generation({batch_size:e,model_input_name:t,model_kwargs:r,decoder_start_token_id:n,bos_token_id:o,generation_config:s}){let{decoder_input_ids:a,...i}=r;if(a)Array.isArray(a[0])||(a=Array.from({length:e},(()=>a)));else if(n??=o,"musicgen"===this.config.model_type)a=Array.from({length:e*this.config.decoder.num_codebooks},(()=>[n]));else if(Array.isArray(n)){if(n.length!==e)throw new Error(`\`decoder_start_token_id\` expcted to have length ${e} but got ${n.length}`);a=n}else a=Array.from({length:e},(()=>[n]));return a=L(a),r.decoder_attention_mask=(0,u.ones_like)(a),{input_ids:a,model_inputs:i}}async generate({inputs:e=null,generation_config:t=null,logits_processor:r=null,stopping_criteria:n=null,streamer:o=null,...s}){this._validate_model_class(),t=this._prepare_generation_config(t,s);let{inputs_tensor:a,model_inputs:i,model_input_name:l}=this._prepare_model_inputs({inputs:e,model_kwargs:s});const c=this.config.is_encoder_decoder;let d;c&&("encoder_outputs"in i||(i=await this._prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:a,model_inputs:i,model_input_name:l,generation_config:t}))),c?({input_ids:d,model_inputs:i}=this._prepare_decoder_input_ids_for_generation({batch_size:i[l].dims.at(0),model_input_name:l,model_kwargs:i,decoder_start_token_id:t.decoder_start_token_id,bos_token_id:t.bos_token_id,generation_config:t})):d=i[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,r),_=this._get_stopping_criteria(t,n),f=i[l].dims.at(0),g=m.LogitsSampler.getSampler(t),M=new Array(f).fill(0),w=d.tolist();let T;o&&o.put(w);let b={};for(;;){if(i=this.prepare_inputs_for_generation(w,i,t),T=await this.forward(i),t.output_attentions&&t.return_dict_in_generate){const e=this.getAttentions(T);for(const t in e)t in b||(b[t]=[]),b[t].push(e[t])}const e=h(w,T.logits.slice(null,-1,null)),r=[];for(let t=0;t<e.dims.at(0);++t){const n=e[t],o=await g(n);for(const[e,n]of o){const o=BigInt(e);M[t]+=n,w[t].push(o),r.push([o]);break}}o&&o.put(r);if(_(w).every((e=>e)))break;i=this._update_model_kwargs_for_generation({generated_input_ids:r,outputs:T,model_inputs:i,is_encoder_decoder:c})}o&&o.end();const x=this.getPastKeyValues(T,i.past_key_values,!0),k=new u.Tensor("int64",w.flat(),[w.length,w[0].length]);if(t.return_dict_in_generate)return{sequences:k,past_key_values:x,...b};for(const e of Object.values(T))"gpu-buffer"===e.location&&e.dispose();return k}getPastKeyValues(e,t,r=!1){const n=Object.create(null);for(const o in e)if(o.startsWith("present")){const s=o.replace("present","past_key_values"),a=o.includes("encoder");if(n[s]=a&&t?t[s]:e[o],t&&(!a||r)){const e=t[s];"gpu-buffer"===e.location&&e.dispose()}}return n}getAttentions(e){const t={};for(const r of["cross_attentions","encoder_attentions","decoder_attentions"])for(const n in e)n.startsWith(r)&&(r in t||(t[r]=[]),t[r].push(e[n]));return t}addPastKeyValues(e,t){if(t)Object.assign(e,t);else{const t=this.custom_config.kv_cache_dtype??"float32",r="float16"===t?new Uint16Array:[],o=(0,n.getKeyValueShapes)(this.config);for(const n in o)e[n]=new u.Tensor(t,r,o[n])}}async encode_image({pixel_values:e}){const t=(await A(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 A(this.sessions.embed_tokens,{input_ids:e})).inputs_embeds}}class G{}class q extends G{constructor({last_hidden_state:e,hidden_states:t=null,attentions:r=null}){super(),this.last_hidden_state=e,this.hidden_states=t,this.attentions=r}}class $ extends R{}class W extends ${}class U extends ${async _call(e){return new cl(await super._call(e))}}class X extends ${async _call(e){return new al(await super._call(e))}}class Q extends ${async _call(e){return new ll(await super._call(e))}}class H extends ${async _call(e){return new dl(await super._call(e))}}class Y extends R{}class J extends Y{}class K extends R{}class Z extends K{}class ee extends K{async _call(e){return new cl(await super._call(e))}}class te extends K{async _call(e){return new al(await super._call(e))}}class re extends K{async _call(e){return new ll(await super._call(e))}}class ne extends K{async _call(e){return new dl(await super._call(e))}}class oe extends R{}class se extends oe{}class ae extends oe{async _call(e){return new cl(await super._call(e))}}class ie extends oe{async _call(e){return new al(await super._call(e))}}class le extends oe{async _call(e){return new ll(await super._call(e))}}class ce extends oe{async _call(e){return new dl(await super._call(e))}}class de extends R{}class ue extends de{}class pe extends de{async _call(e){return new cl(await super._call(e))}}class he extends de{async _call(e){return new al(await super._call(e))}}class me extends de{async _call(e){return new ll(await super._call(e))}}class _e extends de{async _call(e){return new dl(await super._call(e))}}class fe extends R{}class ge extends fe{}class Me extends fe{async _call(e){return new cl(await super._call(e))}}class we extends fe{async _call(e){return new al(await super._call(e))}}class Te extends fe{async _call(e){return new ll(await super._call(e))}}class be extends fe{async _call(e){return new dl(await super._call(e))}}class xe extends R{}class ke extends xe{}class ye extends xe{async _call(e){return new cl(await super._call(e))}}class Fe extends xe{async _call(e){return new al(await super._call(e))}}class Pe extends xe{async _call(e){return new ll(await super._call(e))}}class Ce extends xe{async _call(e){return new dl(await super._call(e))}}class ve extends R{}class Se extends ve{}class Ae extends ve{async _call(e){return new cl(await super._call(e))}}class Ee extends ve{async _call(e){return new al(await super._call(e))}}class Le extends ve{async _call(e){return new ll(await super._call(e))}}class ze extends ve{async _call(e){return new dl(await super._call(e))}}class Ie extends R{}class Be extends Ie{}class Ne extends Ie{async _call(e){return new al(await super._call(e))}}class Oe extends Ie{async _call(e){return new ll(await super._call(e))}}class De extends Ie{async _call(e){return new dl(await super._call(e))}}class Ve extends Ie{async _call(e){return new cl(await super._call(e))}}class je extends R{}class Re extends je{}class Ge extends je{async _call(e){return new cl(await super._call(e))}}class qe extends je{async _call(e){return new al(await super._call(e))}}class $e extends je{async _call(e){return new ll(await super._call(e))}}class We extends R{}class Ue extends We{}class Xe extends We{async _call(e){return new cl(await super._call(e))}}class Qe extends We{async _call(e){return new al(await super._call(e))}}class He extends We{async _call(e){return new dl(await super._call(e))}}class Ye extends R{}class Je extends Ye{}class Ke extends Ye{async _call(e){return new cl(await super._call(e))}}class Ze extends Ye{async _call(e){return new al(await super._call(e))}}class et extends Ye{async _call(e){return new ll(await super._call(e))}}class tt extends Ye{async _call(e){return new dl(await super._call(e))}}class rt extends R{}class nt extends rt{}class ot extends rt{async _call(e){return new cl(await super._call(e))}}class st extends rt{async _call(e){return new al(await super._call(e))}}class at extends rt{async _call(e){return new dl(await super._call(e))}}class it extends R{}class lt extends it{}class ct extends it{async _call(e){return new al(await super._call(e))}}class dt extends it{async _call(e){return new dl(await super._call(e))}}class ut extends it{async _call(e){return new cl(await super._call(e))}}class pt extends R{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"];constructor(e,t,r){super(e,t),this.generation_config=r}}class ht extends pt{}class mt extends pt{}class _t extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class ft extends _t{}class gt extends _t{}class Mt extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class wt extends Mt{}class Tt extends Mt{}class bt extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class xt extends bt{}class kt extends bt{}class yt extends bt{async _call(e){return new al(await super._call(e))}}class Ft extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Pt extends Ft{}class Ct extends Ft{}class vt extends Ft{async _call(e){return new al(await super._call(e))}}class St extends Ft{}class At extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Et extends At{}class Lt extends At{}class zt extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class It extends zt{}class Bt extends zt{}class Nt extends R{}class Ot extends Nt{}class Dt extends Nt{async _call(e){return new cl(await super._call(e))}}class Vt extends Nt{async _call(e){return new al(await super._call(e))}}class jt extends Nt{async _call(e){return new ll(await super._call(e))}}class Rt extends Nt{async _call(e){return new dl(await super._call(e))}}class Gt extends R{}class qt extends Gt{}class $t extends Gt{async _call(e){return new cl(await super._call(e))}}class Wt extends Gt{async _call(e){return new al(await super._call(e))}}class Ut extends Gt{async _call(e){return new ll(await super._call(e))}}class Xt extends Gt{async _call(e){return new dl(await super._call(e))}}class Qt extends R{}class Ht extends Qt{}class Yt extends Qt{async _call(e){return new cl(await super._call(e))}}class Jt extends Qt{async _call(e){return new al(await super._call(e))}}class Kt extends Qt{async _call(e){return new ll(await super._call(e))}}class Zt extends Qt{async _call(e){return new dl(await super._call(e))}}class er extends R{}class tr extends er{}class rr extends er{}class nr extends R{requires_attention_mask=!1;main_input_name="input_features";forward_params=["input_features","attention_mask","decoder_input_ids","decoder_attention_mask","past_key_values"];constructor(e,t,r){super(e,t),this.generation_config=r}}class or extends nr{}class sr extends nr{_prepare_generation_config(e,t){return super._prepare_generation_config(e,t,f.WhisperGenerationConfig)}_retrieve_init_tokens(e){const t=[e.decoder_start_token_id];let r=e.language;const n=e.task;if(e.is_multilingual){r||(console.warn("No language specified - defaulting to English (en)."),r="en");const o=`<|${(0,g.whisper_language_to_code)(r)}|>`;t.push(e.lang_to_id[o]),t.push(e.task_to_id[n??"transcribe"])}else if(r||n)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:r=null,stopping_criteria:n=null,...o}){t=this._prepare_generation_config(t,o);const s=o.decoder_input_ids??this._retrieve_init_tokens(t);if(t.return_timestamps&&(r??=new c.LogitsProcessorList,r.push(new c.WhisperTimeStampLogitsProcessor(t,s))),t.begin_suppress_tokens&&(r??=new c.LogitsProcessorList,r.push(new c.SuppressTokensAtBeginLogitsProcessor(t.begin_suppress_tokens,s.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 a=await super.generate({inputs:e,generation_config:t,logits_processor:r,decoder_input_ids:s,...o});return t.return_token_timestamps&&(a.token_timestamps=this._extract_token_timestamps(a,t.alignment_heads,t.num_frames)),a}_extract_token_timestamps(e,t,r=null,n=.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==r&&console.warn("`num_frames` has not been set, meaning the entire audio will be analyzed. This may lead to inaccurate token-level timestamps for short audios (< 30 seconds).");let o=this.config.median_filter_width;void 0===o&&(console.warn("Model config has no `median_filter_width`, using default value of 7."),o=7);const s=e.cross_attentions,a=Array.from({length:this.config.decoder_layers},((e,t)=>(0,u.cat)(s.map((e=>e[t])),2))),l=(0,u.stack)(t.map((([e,t])=>{if(e>=a.length)throw new Error(`Layer index ${e} is out of bounds for cross attentions (length ${a.length}).`);return r?a[e].slice(null,t,null,[0,r]):a[e].slice(null,t)}))).transpose(1,0,2,3),[c,d]=(0,u.std_mean)(l,-2,0,!0),h=l.clone();for(let e=0;e<h.dims[0];++e){const t=h[e];for(let r=0;r<t.dims[0];++r){const n=t[r],s=c[e][r][0].data,a=d[e][r][0].data;for(let e=0;e<n.dims[0];++e){let t=n[e].data;for(let e=0;e<t.length;++e)t[e]=(t[e]-a[e])/s[e];t.set((0,p.medianFilter)(t,o))}}}const m=[(0,u.mean)(h,1)],_=e.sequences.dims,f=new u.Tensor("float32",new Float32Array(_[0]*_[1]),_);for(let e=0;e<_[0];++e){const t=m[e].neg().squeeze_(0),[r,o]=(0,p.dynamic_time_warping)(t.tolist()),s=Array.from({length:r.length-1},((e,t)=>r[t+1]-r[t])),a=(0,i.mergeArrays)([1],s).map((e=>!!e)),l=[];for(let e=0;e<a.length;++e)a[e]&&l.push(o[e]*n);f[e].data.set(l,1)}return f}}class ar extends R{main_input_name="pixel_values";forward_params=["pixel_values","input_ids","encoder_hidden_states","past_key_values"];constructor(e,t,r){super(e,t),this.generation_config=r}}class ir extends R{forward_params=["input_ids","pixel_values","attention_mask","position_ids","past_key_values"];constructor(e,t,r){super(e,t),this.generation_config=r}}class lr extends ir{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:r,attention_mask:n}){const o=this.config.image_token_index,s=r.tolist().map((e=>e.findIndex((e=>e==o)))),a=s.every((e=>-1===e)),i=s.every((e=>-1!==e));if(!a&&!i)throw new Error("Every input should contain either 0 or 1 image token.");if(a)return{inputs_embeds:e,attention_mask:n};const l=[],c=[];for(let r=0;r<s.length;++r){const o=s[r],a=e[r],i=t[r],d=n[r];l.push((0,u.cat)([a.slice([0,o]),i,a.slice([o+1,a.dims[0]])],0)),c.push((0,u.cat)([d.slice([0,o]),(0,u.ones)([i.dims[0]]),d.slice([o+1,d.dims[0]])],0))}return{inputs_embeds:(0,u.stack)(l,0),attention_mask:(0,u.stack)(c,0)}}}class cr extends lr{}class dr extends R{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,r){super(e,t),this.generation_config=r}}class ur extends dr{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:r,attention_mask:n}){return{inputs_embeds:(0,u.cat)([t,e],1),attention_mask:(0,u.cat)([(0,u.ones)(t.dims.slice(0,2)),n],1)}}async _prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:r,attention_mask:n}){if(!e&&!t)throw new Error("Either `input_ids` or `pixel_values` should be provided.");let o,s;return e&&(o=await this.encode_text({input_ids:e})),t&&(s=await this.encode_image({pixel_values:t})),o&&s?({inputs_embeds:r,attention_mask:n}=this._merge_input_ids_with_image_features({inputs_embeds:o,image_features:s,input_ids:e,attention_mask:n})):r=o||s,{inputs_embeds:r,attention_mask:n}}async forward({input_ids:e,pixel_values:t,attention_mask:r,decoder_input_ids:n,decoder_attention_mask:o,encoder_outputs:s,past_key_values:a,inputs_embeds:i,decoder_inputs_embeds:l}){if(i||({inputs_embeds:i,attention_mask:r}=await this._prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:i,attention_mask:r})),!s){let{last_hidden_state:e}=await B(this,{inputs_embeds:i,attention_mask:r});s=e}if(!l){if(!n)throw new Error("Either `decoder_input_ids` or `decoder_inputs_embeds` should be provided.");l=await this.encode_text({input_ids:n})}const c={inputs_embeds:l,attention_mask:o,encoder_attention_mask:r,encoder_hidden_states:s,past_key_values:a};return await N(this,c,!0)}}class pr extends R{}class hr extends pr{}class mr extends pr{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class _r extends pr{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class fr extends pr{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class gr extends pr{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class Mr extends R{}class wr extends Mr{}class Tr extends Mr{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class br extends pr{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class xr extends R{}class kr extends xr{}class yr extends R{}class Fr extends yr{}class Pr extends yr{}class Cr extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class vr extends Cr{}class Sr extends Cr{}class Ar extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Er extends Ar{}class Lr extends Ar{}class zr extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Ir extends zr{}class Br extends zr{}class Nr extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Or extends Nr{}class Dr extends Nr{}class Vr extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class jr extends Vr{}class Rr extends Vr{}class Gr extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class qr extends Gr{}class $r extends Gr{}class Wr extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Ur extends Wr{}class Xr extends Wr{}class Qr extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Hr extends Qr{}class Yr extends Qr{}class Jr extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Kr extends Jr{}class Zr extends Jr{}class en extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class tn extends en{}class rn extends en{}class nn extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class on extends nn{}class sn extends nn{}class an extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class ln extends an{}class cn extends an{}class dn extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class un extends dn{}class pn extends dn{}class hn extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class mn extends hn{}class _n extends hn{}class fn extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class gn extends fn{}class Mn extends fn{}class wn extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Tn extends wn{}class bn extends wn{}class xn extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class kn extends xn{}class yn extends xn{}class Fn extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Pn extends Fn{}class Cn extends Fn{}class vn extends R{}class Sn extends vn{}class An extends vn{async _call(e){return new al(await super._call(e))}}class En extends R{}class Ln extends En{}class zn extends En{async _call(e){return new al(await super._call(e))}}class In extends R{}class Bn extends In{}class Nn extends R{}class On extends Nn{}class Dn extends Nn{async _call(e){return new al(await super._call(e))}}class Vn extends R{}class jn extends Vn{}class Rn extends R{}class Gn extends Rn{}class qn extends Rn{async _call(e){return new al(await super._call(e))}}class $n extends R{}class Wn extends $n{async _call(e){return new hl(await super._call(e))}}class Un extends R{}class Xn extends Un{}class Qn extends Un{async _call(e){return new al(await super._call(e))}}class Hn extends R{}class Yn extends Hn{}class Jn extends Hn{async _call(e){return new al(await super._call(e))}}class Kn extends R{}class Zn extends Kn{}class eo extends Kn{}class to extends R{}class ro extends to{}class no extends to{}class oo extends R{}class so extends oo{}class ao extends oo{async _call(e){return new al(await super._call(e))}}class io extends R{}class lo extends io{}class co extends io{async _call(e){return new po(await super._call(e))}}class uo extends io{async _call(e){return new ho(await super._call(e))}}class po extends G{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class ho extends G{constructor({logits:e,pred_boxes:t,pred_masks:r}){super(),this.logits=e,this.pred_boxes=t,this.pred_masks=r}}class mo extends R{}class _o extends mo{}class fo extends mo{async _call(e){return new go(await super._call(e))}}class go extends G{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class Mo extends R{}class wo extends Mo{}class To extends Mo{async _call(e){return new bo(await super._call(e))}}class bo extends po{}class xo extends R{}class ko extends xo{}class yo extends xo{async _call(e){return new al(await super._call(e))}}class Fo extends R{}class Po extends Fo{}class Co extends Fo{async _call(e){return new al(await super._call(e))}}class vo extends R{}class So extends vo{}class Ao extends vo{async _call(e){return new al(await super._call(e))}}class Eo extends R{}class Lo extends Eo{}class zo extends Eo{async _call(e){return new al(await super._call(e))}}class Io extends R{}class Bo extends Io{}class No extends Io{}class Oo extends R{}class Do extends Oo{}class Vo extends Oo{}class jo extends R{}class Ro extends jo{}class Go extends R{}class qo extends Go{}class $o extends Go{}class Wo extends Go{}class Uo extends R{}class Xo extends Uo{}class Qo extends Uo{}class Ho extends R{}class Yo extends Ho{}class Jo extends Ho{}class Ko extends R{}class Zo extends Ko{}class es extends R{}class ts extends es{}class rs extends es{async _call(e){return new al(await super._call(e))}}class ns extends R{}class os extends ns{}class ss extends ns{async _call(e){return new al(await super._call(e))}}class as extends R{}class is extends as{}class ls extends as{async _call(e){return new al(await super._call(e))}}class cs extends R{}class ds extends cs{}class us extends cs{async _call(e){return new ps(await super._call(e))}}class ps extends G{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class hs extends R{}class ms extends hs{async get_image_embeddings({pixel_values:e}){return await B(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),r=t.reduce(((e,t)=>e*t),1);e.input_labels=new u.Tensor("int64",new BigInt64Array(r).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 A(this.sessions.prompt_encoder_mask_decoder,t)}async _call(e){return new _s(await super._call(e))}}class _s extends G{constructor({iou_scores:e,pred_masks:t}){super(),this.iou_scores=e,this.pred_masks=t}}class fs extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class gs extends fs{}class Ms extends fs{}class ws extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class Ts extends ws{}class bs extends ws{}class xs extends R{}class ks extends xs{}class ys extends xs{async _call(e){return new ul(await super._call(e))}}class Fs extends xs{async _call(e){return new al(await super._call(e))}}class Ps extends xs{async _call(e){return new ll(await super._call(e))}}class Cs extends R{}class vs extends Cs{}class Ss extends Cs{async _call(e){return new ll(await super._call(e))}}class As extends R{}class Es extends As{}class Ls extends R{}class zs extends Ls{}class Is extends Ls{async _call(e){return new ul(await super._call(e))}}class Bs extends Ls{async _call(e){return new al(await super._call(e))}}class Ns extends R{}class Os extends Ns{}class Ds extends Ns{async _call(e){return new ul(await super._call(e))}}class Vs extends Ns{async _call(e){return new al(await super._call(e))}}class js extends Ns{async _call(e){return new ll(await super._call(e))}}class Rs extends R{}class Gs extends Rs{}class qs extends Rs{async _call(e){return new ul(await super._call(e))}}class $s extends Rs{async _call(e){return new al(await super._call(e))}}class Ws extends R{}class Us extends xs{}class Xs extends xs{async _call(e){return new ul(await super._call(e))}}class Qs extends xs{async _call(e){return new al(await super._call(e))}}class Hs extends R{}class Ys extends Hs{}class Js extends Hs{async _call(e){return new ul(await super._call(e))}}class Ks extends Hs{async _call(e){return new al(await super._call(e))}}class Zs extends Hs{async _call(e){return new il(await super._call(e))}}class ea extends Hs{async _call(e){return new ll(await super._call(e))}}class ta extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class ra extends ta{}class na extends ta{}class oa extends ta{async generate_speech(e,t,{threshold:r=.5,minlenratio:n=0,maxlenratio:o=20,vocoder:s=null}={}){const a={input_ids:e},{encoder_outputs:i,encoder_attention_mask:l}=await B(this,a),c=i.dims[1]/this.config.reduction_factor,d=Math.floor(c*o),p=Math.floor(c*n),h=this.config.num_mel_bins;let m=[],_=null,f=null,g=0;for(;;){++g;const e=z(!!f);let n;n=f?f.output_sequence_out:new u.Tensor("float32",new Float32Array(h),[1,1,h]);let o={use_cache_branch:e,output_sequence:n,encoder_attention_mask:l,speaker_embeddings:t,encoder_hidden_states:i};this.addPastKeyValues(o,_),f=await A(this.sessions.decoder_model_merged,o),_=this.getPastKeyValues(f,_);const{prob:s,spectrum:a}=f;if(m.push(a),g>=p&&(Array.from(s.data).filter((e=>e>=r)).length>0||g>=d))break}const M=(0,u.cat)(m),{waveform:w}=await A(s.sessions.model,{spectrogram:M});return{spectrogram:M,waveform:w}}}class sa extends R{main_input_name="spectrogram"}class aa extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class ia extends aa{}class la extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class ca extends la{}class da extends la{}class ua extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class pa extends ua{}class ha extends ua{}class ma extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class _a extends ma{}class fa extends ma{}class ga extends R{}class Ma extends ga{}class wa extends ga{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class Ta extends ga{static async from_pretrained(e,t={}){return t.model_file_name??="audio_model",super.from_pretrained(e,t)}}class ba extends R{}class xa extends ba{async _call(e){return new ml(await super._call(e))}}class ka extends R{}class ya extends ka{}class Fa extends ka{}class Pa extends ka{}class Ca extends R{constructor(e,t,r){super(e,t),this.generation_config=r}}class va extends Ca{}class Sa extends Ca{}class Aa extends R{}class Ea extends Aa{}class La extends Aa{async _call(e){return new al(await super._call(e))}}class za extends R{}class Ia extends za{}class Ba extends za{}class Na extends R{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"];constructor(e,t,r){super(e,t),this.generation_config=r}_apply_and_filter_by_delay_pattern_mask(e){const[t,r]=e.dims,n=this.config.decoder.num_codebooks,o=r-n;let s=0;for(let t=0;t<e.size;++t){if(e.data[t]===this.config.decoder.pad_token_id)continue;const a=t%r-Math.floor(t/r)%n;a>0&&a<=o&&(e.data[s++]=e.data[t])}const a=Math.floor(t/n),i=s/(a*n);return new u.Tensor(e.type,e.data.slice(0,s),[a,n,i])}prepare_inputs_for_generation(e,t,r){let n=structuredClone(e);for(let e=0;e<n.length;++e)for(let t=0;t<n[e].length;++t)e%this.config.decoder.num_codebooks>=t&&(n[e][t]=BigInt(this.config.decoder.pad_token_id));null!==r.guidance_scale&&r.guidance_scale>1&&(n=n.concat(n));return super.prepare_inputs_for_generation(n,t,r)}async generate(e){const t=await super.generate(e),r=this._apply_and_filter_by_delay_pattern_mask(t).unsqueeze_(0),{audio_values:n}=await A(this.sessions.encodec_decode,{audio_codes:r});return n}}class Oa extends R{}class Da extends Oa{}class Va extends Oa{async _call(e){return new al(await super._call(e))}}class ja extends R{}class Ra extends ja{}class Ga extends ja{async _call(e){return new al(await super._call(e))}}class qa extends R{}class $a extends qa{}class Wa extends qa{async _call(e){return new al(await super._call(e))}}class Ua extends R{}class Xa extends Ua{}class Qa extends Ua{async _call(e){return new al(await super._call(e))}}class Ha extends R{}class Ya extends Ha{}class Ja{static MODEL_CLASS_MAPPINGS=null;static BASE_IF_FAIL=!1;static async from_pretrained(e,{progress_callback:t=null,config:r=null,cache_dir:o=null,local_files_only:s=!1,revision:a="main",model_file_name:i=null,subfolder:l="onnx",device:c=null,dtype:d=null,use_external_data_format:u=null,session_options:p={}}={}){const h={progress_callback:t,config:r,cache_dir:o,local_files_only:s,revision:a,model_file_name:i,subfolder:l,device:c,dtype:d,use_external_data_format:u,session_options:p};if(h.config=await n.AutoConfig.from_pretrained(e,h),!this.MODEL_CLASS_MAPPINGS)throw new Error("`MODEL_CLASS_MAPPINGS` not implemented for this type of `AutoClass`: "+this.name);for(const t of this.MODEL_CLASS_MAPPINGS){const r=t.get(h.config.model_type);if(r)return await r[1].from_pretrained(e,h)}if(this.BASE_IF_FAIL)return console.warn(`Unknown model class "${h.config.model_type}", attempting to construct from base class.`),await R.from_pretrained(e,h);throw Error(`Unsupported model type: ${h.config.model_type}`)}}const Ka=new Map([["bert",["BertModel",W]],["nomic_bert",["NomicBertModel",J]],["roformer",["RoFormerModel",Z]],["electra",["ElectraModel",ue]],["esm",["EsmModel",Re]],["convbert",["ConvBertModel",se]],["camembert",["CamembertModel",ge]],["deberta",["DebertaModel",ke]],["deberta-v2",["DebertaV2Model",Se]],["mpnet",["MPNetModel",Je]],["albert",["AlbertModel",lt]],["distilbert",["DistilBertModel",Be]],["roberta",["RobertaModel",Ot]],["xlm",["XLMModel",qt]],["xlm-roberta",["XLMRobertaModel",Ht]],["clap",["ClapModel",Ma]],["clip",["CLIPModel",hr]],["clipseg",["CLIPSegModel",Fr]],["chinese_clip",["ChineseCLIPModel",kr]],["siglip",["SiglipModel",wr]],["mobilebert",["MobileBertModel",Ue]],["squeezebert",["SqueezeBertModel",nt]],["wav2vec2",["Wav2Vec2Model",ks]],["wav2vec2-bert",["Wav2Vec2BertModel",Gs]],["unispeech",["UniSpeechModel",zs]],["unispeech-sat",["UniSpeechSatModel",Os]],["hubert",["HubertModel",Us]],["wavlm",["WavLMModel",Ys]],["audio-spectrogram-transformer",["ASTModel",tr]],["vits",["VitsModel",xa]],["pyannote",["PyAnnoteModel",vs]],["wespeaker-resnet",["WeSpeakerResNetModel",Es]],["detr",["DetrModel",lo]],["rt_detr",["RTDetrModel",_o]],["table-transformer",["TableTransformerModel",wo]],["vit",["ViTModel",Sn]],["pvt",["PvtModel",Ln]],["vit_msn",["ViTMSNModel",On]],["vit_mae",["ViTMAEModel",Bn]],["groupvit",["GroupViTModel",jn]],["fastvit",["FastViTModel",Gn]],["mobilevit",["MobileViTModel",Xn]],["mobilevitv2",["MobileViTV2Model",Yn]],["owlvit",["OwlViTModel",Zn]],["owlv2",["Owlv2Model",ro]],["beit",["BeitModel",so]],["deit",["DeiTModel",ko]],["hiera",["HieraModel",Po]],["convnext",["ConvNextModel",ts]],["convnextv2",["ConvNextV2Model",os]],["dinov2",["Dinov2Model",is]],["resnet",["ResNetModel",So]],["swin",["SwinModel",Lo]],["swin2sr",["Swin2SRModel",Bo]],["donut-swin",["DonutSwinModel",Zo]],["yolos",["YolosModel",ds]],["dpt",["DPTModel",Do]],["glpn",["GLPNModel",Yo]],["hifigan",["SpeechT5HifiGan",sa]],["efficientnet",["EfficientNetModel",Ea]],["decision_transformer",["DecisionTransformerModel",Ya]],["mobilenet_v1",["MobileNetV1Model",Da]],["mobilenet_v2",["MobileNetV2Model",Ra]],["mobilenet_v3",["MobileNetV3Model",$a]],["mobilenet_v4",["MobileNetV4Model",Xa]],["maskformer",["MaskFormerModel",Xo]]]),Za=new Map([["t5",["T5Model",ht]],["longt5",["LongT5Model",ft]],["mt5",["MT5Model",wt]],["bart",["BartModel",xt]],["mbart",["MBartModel",Pt]],["marian",["MarianModel",gs]],["whisper",["WhisperModel",or]],["m2m_100",["M2M100Model",Ts]],["blenderbot",["BlenderbotModel",Et]],["blenderbot-small",["BlenderbotSmallModel",It]]]),ei=new Map([["bloom",["BloomModel",Tn]],["jais",["JAISModel",Er]],["gpt2",["GPT2Model",vr]],["gptj",["GPTJModel",jr]],["gpt_bigcode",["GPTBigCodeModel",qr]],["gpt_neo",["GPTNeoModel",Ir]],["gpt_neox",["GPTNeoXModel",Or]],["codegen",["CodeGenModel",Ur]],["llama",["LlamaModel",Hr]],["cohere",["CohereModel",Kr]],["gemma",["GemmaModel",tn]],["gemma2",["Gemma2Model",on]],["openelm",["OpenELMModel",ln]],["qwen2",["Qwen2Model",un]],["phi",["PhiModel",mn]],["phi3",["Phi3Model",gn]],["mpt",["MptModel",kn]],["opt",["OPTModel",Pn]],["mistral",["MistralModel",ca]],["starcoder2",["Starcoder2Model",pa]],["falcon",["FalconModel",_a]],["stablelm",["StableLmModel",va]]]),ti=new Map([["speecht5",["SpeechT5ForSpeechToText",na]],["whisper",["WhisperForConditionalGeneration",sr]]]),ri=new Map([["speecht5",["SpeechT5ForTextToSpeech",oa]]]),ni=new Map([["vits",["VitsModel",xa]],["musicgen",["MusicgenForConditionalGeneration",Na]]]),oi=new Map([["bert",["BertForSequenceClassification",X]],["roformer",["RoFormerForSequenceClassification",te]],["electra",["ElectraForSequenceClassification",he]],["esm",["EsmForSequenceClassification",qe]],["convbert",["ConvBertForSequenceClassification",ie]],["camembert",["CamembertForSequenceClassification",we]],["deberta",["DebertaForSequenceClassification",Fe]],["deberta-v2",["DebertaV2ForSequenceClassification",Ee]],["mpnet",["MPNetForSequenceClassification",Ze]],["albert",["AlbertForSequenceClassification",ct]],["distilbert",["DistilBertForSequenceClassification",Ne]],["roberta",["RobertaForSequenceClassification",Vt]],["xlm",["XLMForSequenceClassification",Wt]],["xlm-roberta",["XLMRobertaForSequenceClassification",Jt]],["bart",["BartForSequenceClassification",yt]],["mbart",["MBartForSequenceClassification",vt]],["mobilebert",["MobileBertForSequenceClassification",Qe]],["squeezebert",["SqueezeBertForSequenceClassification",st]]]),si=new Map([["bert",["BertForTokenClassification",Q]],["roformer",["RoFormerForTokenClassification",re]],["electra",["ElectraForTokenClassification",me]],["esm",["EsmForTokenClassification",$e]],["convbert",["ConvBertForTokenClassification",le]],["camembert",["CamembertForTokenClassification",Te]],["deberta",["DebertaForTokenClassification",Pe]],["deberta-v2",["DebertaV2ForTokenClassification",Le]],["mpnet",["MPNetForTokenClassification",et]],["distilbert",["DistilBertForTokenClassification",Oe]],["roberta",["RobertaForTokenClassification",jt]],["xlm",["XLMForTokenClassification",Ut]],["xlm-roberta",["XLMRobertaForTokenClassification",Kt]]]),ai=new Map([["t5",["T5ForConditionalGeneration",mt]],["longt5",["LongT5ForConditionalGeneration",gt]],["mt5",["MT5ForConditionalGeneration",Tt]],["bart",["BartForConditionalGeneration",kt]],["mbart",["MBartForConditionalGeneration",Ct]],["marian",["MarianMTModel",Ms]],["m2m_100",["M2M100ForConditionalGeneration",bs]],["blenderbot",["BlenderbotForConditionalGeneration",Lt]],["blenderbot-small",["BlenderbotSmallForConditionalGeneration",Bt]]]),ii=new Map([["bloom",["BloomForCausalLM",bn]],["gpt2",["GPT2LMHeadModel",Sr]],["jais",["JAISLMHeadModel",Lr]],["gptj",["GPTJForCausalLM",Rr]],["gpt_bigcode",["GPTBigCodeForCausalLM",$r]],["gpt_neo",["GPTNeoForCausalLM",Br]],["gpt_neox",["GPTNeoXForCausalLM",Dr]],["codegen",["CodeGenForCausalLM",Xr]],["llama",["LlamaForCausalLM",Yr]],["cohere",["CohereForCausalLM",Zr]],["gemma",["GemmaForCausalLM",rn]],["gemma2",["Gemma2ForCausalLM",sn]],["openelm",["OpenELMForCausalLM",cn]],["qwen2",["Qwen2ForCausalLM",pn]],["phi",["PhiForCausalLM",_n]],["phi3",["Phi3ForCausalLM",Mn]],["mpt",["MptForCausalLM",yn]],["opt",["OPTForCausalLM",Cn]],["mbart",["MBartForCausalLM",St]],["mistral",["MistralForCausalLM",da]],["starcoder2",["Starcoder2ForCausalLM",ha]],["falcon",["FalconForCausalLM",fa]],["trocr",["TrOCRForCausalLM",ia]],["stablelm",["StableLmForCausalLM",Sa]]]),li=new Map([["bert",["BertForMaskedLM",U]],["roformer",["RoFormerForMaskedLM",ee]],["electra",["ElectraForMaskedLM",pe]],["esm",["EsmForMaskedLM",Ge]],["convbert",["ConvBertForMaskedLM",ae]],["camembert",["CamembertForMaskedLM",Me]],["deberta",["DebertaForMaskedLM",ye]],["deberta-v2",["DebertaV2ForMaskedLM",Ae]],["mpnet",["MPNetForMaskedLM",Ke]],["albert",["AlbertForMaskedLM",ut]],["distilbert",["DistilBertForMaskedLM",Ve]],["roberta",["RobertaForMaskedLM",Dt]],["xlm",["XLMWithLMHeadModel",$t]],["xlm-roberta",["XLMRobertaForMaskedLM",Yt]],["mobilebert",["MobileBertForMaskedLM",Xe]],["squeezebert",["SqueezeBertForMaskedLM",ot]]]),ci=new Map([["bert",["BertForQuestionAnswering",H]],["roformer",["RoFormerForQuestionAnswering",ne]],["electra",["ElectraForQuestionAnswering",_e]],["convbert",["ConvBertForQuestionAnswering",ce]],["camembert",["CamembertForQuestionAnswering",be]],["deberta",["DebertaForQuestionAnswering",Ce]],["deberta-v2",["DebertaV2ForQuestionAnswering",ze]],["mpnet",["MPNetForQuestionAnswering",tt]],["albert",["AlbertForQuestionAnswering",dt]],["distilbert",["DistilBertForQuestionAnswering",De]],["roberta",["RobertaForQuestionAnswering",Rt]],["xlm",["XLMForQuestionAnswering",Xt]],["xlm-roberta",["XLMRobertaForQuestionAnswering",Zt]],["mobilebert",["MobileBertForQuestionAnswering",He]],["squeezebert",["SqueezeBertForQuestionAnswering",at]]]),di=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",ar]]]),ui=new Map([["llava",["LlavaForConditionalGeneration",lr]],["moondream1",["Moondream1ForConditionalGeneration",cr]],["florence2",["Florence2ForConditionalGeneration",ur]]]),pi=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",ar]]]),hi=new Map([["vit",["ViTForImageClassification",An]],["pvt",["PvtForImageClassification",zn]],["vit_msn",["ViTMSNForImageClassification",Dn]],["fastvit",["FastViTForImageClassification",qn]],["mobilevit",["MobileViTForImageClassification",Qn]],["mobilevitv2",["MobileViTV2ForImageClassification",Jn]],["beit",["BeitForImageClassification",ao]],["deit",["DeiTForImageClassification",yo]],["hiera",["HieraForImageClassification",Co]],["convnext",["ConvNextForImageClassification",rs]],["convnextv2",["ConvNextV2ForImageClassification",ss]],["dinov2",["Dinov2ForImageClassification",ls]],["resnet",["ResNetForImageClassification",Ao]],["swin",["SwinForImageClassification",zo]],["segformer",["SegformerForImageClassification",Fa]],["efficientnet",["EfficientNetForImageClassification",La]],["mobilenet_v1",["MobileNetV1ForImageClassification",Va]],["mobilenet_v2",["MobileNetV2ForImageClassification",Ga]],["mobilenet_v3",["MobileNetV3ForImageClassification",Wa]],["mobilenet_v4",["MobileNetV4ForImageClassification",Qa]]]),mi=new Map([["detr",["DetrForObjectDetection",co]],["rt_detr",["RTDetrForObjectDetection",fo]],["table-transformer",["TableTransformerForObjectDetection",To]],["yolos",["YolosForObjectDetection",us]]]),_i=new Map([["owlvit",["OwlViTForObjectDetection",eo]],["owlv2",["Owlv2ForObjectDetection",no]]]),fi=new Map([["detr",["DetrForSegmentation",uo]],["clipseg",["CLIPSegForImageSegmentation",Pr]]]),gi=new Map([["segformer",["SegformerForSemanticSegmentation",Pa]],["sapiens",["SapiensForSemanticSegmentation",qo]]]),Mi=new Map([["detr",["DetrForSegmentation",uo]],["maskformer",["MaskFormerForInstanceSegmentation",Qo]]]),wi=new Map([["sam",["SamModel",ms]]]),Ti=new Map([["wav2vec2",["Wav2Vec2ForCTC",ys]],["wav2vec2-bert",["Wav2Vec2BertForCTC",qs]],["unispeech",["UniSpeechForCTC",Is]],["unispeech-sat",["UniSpeechSatForCTC",Ds]],["wavlm",["WavLMForCTC",Js]],["hubert",["HubertForCTC",Xs]]]),bi=new Map([["wav2vec2",["Wav2Vec2ForSequenceClassification",Fs]],["wav2vec2-bert",["Wav2Vec2BertForSequenceClassification",$s]],["unispeech",["UniSpeechForSequenceClassification",Bs]],["unispeech-sat",["UniSpeechSatForSequenceClassification",Vs]],["wavlm",["WavLMForSequenceClassification",Ks]],["hubert",["HubertForSequenceClassification",Qs]],["audio-spectrogram-transformer",["ASTForAudioClassification",rr]]]),xi=new Map([["wavlm",["WavLMForXVector",Zs]]]),ki=new Map([["unispeech-sat",["UniSpeechSatForAudioFrameClassification",js]],["wavlm",["WavLMForAudioFrameClassification",ea]],["wav2vec2",["Wav2Vec2ForAudioFrameClassification",Ps]],["pyannote",["PyAnnoteForAudioFrameClassification",Ss]]]),yi=new Map([["vitmatte",["VitMatteForImageMatting",Wn]]]),Fi=new Map([["swin2sr",["Swin2SRForImageSuperResolution",No]]]),Pi=new Map([["dpt",["DPTForDepthEstimation",Vo]],["depth_anything",["DepthAnythingForDepthEstimation",Ro]],["glpn",["GLPNForDepthEstimation",Jo]],["sapiens",["SapiensForDepthEstimation",$o]]]),Ci=new Map([["sapiens",["SapiensForNormalEstimation",Wo]]]),vi=new Map([["clip",["CLIPVisionModelWithProjection",gr]],["siglip",["SiglipVisionModel",br]]]),Si=[[Ka,M],[Za,w],[ei,x],[oi,M],[si,M],[ai,T],[ti,T],[ii,x],[li,M],[ci,M],[di,b],[ui,y],[hi,M],[fi,M],[Mi,M],[gi,M],[yi,M],[Fi,M],[Pi,M],[Ci,M],[mi,M],[_i,M],[wi,k],[Ti,M],[bi,M],[ri,T],[ni,M],[xi,M],[ki,M],[vi,M]];for(const[e,t]of Si)for(const[r,n]of e.values())P.set(r,t),v.set(n,r),C.set(r,n);const Ai=[["MusicgenForConditionalGeneration",Na,F],["CLIPTextModelWithProjection",_r,M],["SiglipTextModel",Tr,M],["ClapTextModelWithProjection",wa,M],["ClapAudioModelWithProjection",Ta,M]];for(const[e,t,r]of Ai)P.set(e,r),v.set(t,e),C.set(e,t);class Ei extends Ja{static MODEL_CLASS_MAPPINGS=Si.map((e=>e[0]));static BASE_IF_FAIL=!0}class Li extends Ja{static MODEL_CLASS_MAPPINGS=[oi]}class zi extends Ja{static MODEL_CLASS_MAPPINGS=[si]}class Ii extends Ja{static MODEL_CLASS_MAPPINGS=[ai]}class Bi extends Ja{static MODEL_CLASS_MAPPINGS=[ti]}class Ni extends Ja{static MODEL_CLASS_MAPPINGS=[ri]}class Oi extends Ja{static MODEL_CLASS_MAPPINGS=[ni]}class Di extends Ja{static MODEL_CLASS_MAPPINGS=[ii]}class Vi extends Ja{static MODEL_CLASS_MAPPINGS=[li]}class ji extends Ja{static MODEL_CLASS_MAPPINGS=[ci]}class Ri extends Ja{static MODEL_CLASS_MAPPINGS=[di]}class Gi extends Ja{static MODEL_CLASS_MAPPINGS=[hi]}class qi extends Ja{static MODEL_CLASS_MAPPINGS=[fi]}class $i extends Ja{static MODEL_CLASS_MAPPINGS=[gi]}class Wi extends Ja{static MODEL_CLASS_MAPPINGS=[Mi]}class Ui extends Ja{static MODEL_CLASS_MAPPINGS=[mi]}class Xi extends Ja{static MODEL_CLASS_MAPPINGS=[_i]}class Qi extends Ja{static MODEL_CLASS_MAPPINGS=[wi]}class Hi extends Ja{static MODEL_CLASS_MAPPINGS=[Ti]}class Yi extends Ja{static MODEL_CLASS_MAPPINGS=[bi]}class Ji extends Ja{static MODEL_CLASS_MAPPINGS=[xi]}class Ki extends Ja{static MODEL_CLASS_MAPPINGS=[ki]}class Zi extends Ja{static MODEL_CLASS_MAPPINGS=[pi]}class el extends Ja{static MODEL_CLASS_MAPPINGS=[yi]}class tl extends Ja{static MODEL_CLASS_MAPPINGS=[Fi]}class rl extends Ja{static MODEL_CLASS_MAPPINGS=[Pi]}class nl extends Ja{static MODEL_CLASS_MAPPINGS=[Ci]}class ol extends Ja{static MODEL_CLASS_MAPPINGS=[vi]}class sl extends G{constructor({logits:e,past_key_values:t,encoder_outputs:r,decoder_attentions:n=null,cross_attentions:o=null}){super(),this.logits=e,this.past_key_values=t,this.encoder_outputs=r,this.decoder_attentions=n,this.cross_attentions=o}}class al extends G{constructor({logits:e}){super(),this.logits=e}}class il extends G{constructor({logits:e,embeddings:t}){super(),this.logits=e,this.embeddings=t}}class ll extends G{constructor({logits:e}){super(),this.logits=e}}class cl extends G{constructor({logits:e}){super(),this.logits=e}}class dl extends G{constructor({start_logits:e,end_logits:t}){super(),this.start_logits=e,this.end_logits=t}}class ul extends G{constructor({logits:e}){super(),this.logits=e}}class pl extends G{constructor({logits:e,past_key_values:t}){super(),this.logits=e,this.past_key_values=t}}class hl extends G{constructor({alphas:e}){super(),this.alphas=e}}class ml extends G{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,r)=>{r.r(t),r.d(t,{WHISPER_LANGUAGE_MAPPING:()=>o,WHISPER_TO_LANGUAGE_CODE_MAPPING:()=>s,whisper_language_to_code:()=>a});const n=[["en","english"],["zh","chinese"],["de","german"],["es","spanish"],["ru","russian"],["ko","korean"],["fr","french"],["ja","japanese"],["pt","portuguese"],["tr","turkish"],["pl","polish"],["ca","catalan"],["nl","dutch"],["ar","arabic"],["sv","swedish"],["it","italian"],["id","indonesian"],["hi","hindi"],["fi","finnish"],["vi","vietnamese"],["he","hebrew"],["uk","ukrainian"],["el","greek"],["ms","malay"],["cs","czech"],["ro","romanian"],["da","danish"],["hu","hungarian"],["ta","tamil"],["no","norwegian"],["th","thai"],["ur","urdu"],["hr","croatian"],["bg","bulgarian"],["lt","lithuanian"],["la","latin"],["mi","maori"],["ml","malayalam"],["cy","welsh"],["sk","slovak"],["te","telugu"],["fa","persian"],["lv","latvian"],["bn","bengali"],["sr","serbian"],["az","azerbaijani"],["sl","slovenian"],["kn","kannada"],["et","estonian"],["mk","macedonian"],["br","breton"],["eu","basque"],["is","icelandic"],["hy","armenian"],["ne","nepali"],["mn","mongolian"],["bs","bosnian"],["kk","kazakh"],["sq","albanian"],["sw","swahili"],["gl","galician"],["mr","marathi"],["pa","punjabi"],["si","sinhala"],["km","khmer"],["sn","shona"],["yo","yoruba"],["so","somali"],["af","afrikaans"],["oc","occitan"],["ka","georgian"],["be","belarusian"],["tg","tajik"],["sd","sindhi"],["gu","gujarati"],["am","amharic"],["yi","yiddish"],["lo","lao"],["uz","uzbek"],["fo","faroese"],["ht","haitian creole"],["ps","pashto"],["tk","turkmen"],["nn","nynorsk"],["mt","maltese"],["sa","sanskrit"],["lb","luxembourgish"],["my","myanmar"],["bo","tibetan"],["tl","tagalog"],["mg","malagasy"],["as","assamese"],["tt","tatar"],["haw","hawaiian"],["ln","lingala"],["ha","hausa"],["ba","bashkir"],["jw","javanese"],["su","sundanese"]],o=new Map(n),s=new Map([...n.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 a(e){e=e.toLowerCase();let t=s.get(e);if(void 0===t){if(!o.has(e)){const t=2===e.length?o.keys():o.values();throw new Error(`Language "${e}" is not supported. Must be one of: ${JSON.stringify(t)}`)}t=e}return t}},"./src/models/whisper/generation_whisper.js":
@@ -127,7 +127,7 @@ import*as e from"fs";import*as t from"onnxruntime-node";import*as r from"path";i
   \***************************/(e,t,r)=>{r.r(t),r.d(t,{ASTFeatureExtractor:()=>se,AutoProcessor:()=>Me,BeitFeatureExtractor:()=>X,BitImageProcessor:()=>P,CLIPFeatureExtractor:()=>v,CLIPImageProcessor:()=>S,ChineseCLIPFeatureExtractor:()=>A,ClapFeatureExtractor:()=>ae,ConvNextFeatureExtractor:()=>L,ConvNextImageProcessor:()=>z,DPTFeatureExtractor:()=>y,DPTImageProcessor:()=>F,DeiTFeatureExtractor:()=>U,DetrFeatureExtractor:()=>Y,DonutFeatureExtractor:()=>Q,EfficientNetImageProcessor:()=>N,FeatureExtractor:()=>w,Florence2Processor:()=>ge,GLPNFeatureExtractor:()=>C,ImageFeatureExtractor:()=>T,MaskFormerFeatureExtractor:()=>J,MobileNetV1FeatureExtractor:()=>O,MobileNetV2FeatureExtractor:()=>D,MobileNetV3FeatureExtractor:()=>V,MobileNetV4FeatureExtractor:()=>j,MobileViTFeatureExtractor:()=>R,MobileViTImageProcessor:()=>G,NougatImageProcessor:()=>H,OwlViTFeatureExtractor:()=>q,OwlViTProcessor:()=>fe,Owlv2ImageProcessor:()=>$,Processor:()=>de,PvtImageProcessor:()=>k,PyAnnoteFeatureExtractor:()=>ie,PyAnnoteProcessor:()=>me,RTDetrImageProcessor:()=>W,SamImageProcessor:()=>Z,SamProcessor:()=>ue,SapiensFeatureExtractor:()=>b,SeamlessM4TFeatureExtractor:()=>oe,SegformerFeatureExtractor:()=>x,SiglipImageProcessor:()=>E,SpeechT5FeatureExtractor:()=>ce,SpeechT5Processor:()=>_e,Swin2SRImageProcessor:()=>ee,ViTFeatureExtractor:()=>I,ViTImageProcessor:()=>B,VitMatteImageProcessor:()=>te,Wav2Vec2FeatureExtractor:()=>ne,Wav2Vec2ProcessorWithLM:()=>he,WeSpeakerFeatureExtractor:()=>le,WhisperFeatureExtractor:()=>re,WhisperProcessor:()=>pe,YolosFeatureExtractor:()=>K});var n=r(/*! ./utils/generic.js */"./src/utils/generic.js"),o=r(/*! ./utils/core.js */"./src/utils/core.js"),s=r(/*! ./utils/hub.js */"./src/utils/hub.js"),a=r(/*! ./utils/maths.js */"./src/utils/maths.js"),i=r(/*! ./utils/tensor.js */"./src/utils/tensor.js"),l=(r(/*! ./utils/image.js */"./src/utils/image.js"),r(/*! ./utils/audio.js */"./src/utils/audio.js"));function c([e,t,r,n]){return[e-r/2,t-n/2,e+r/2,t+n/2]}function d(e,t=.5,r=null,n=!1){const o=e.logits,s=e.pred_boxes,[i,l,d]=o.dims;if(null!==r&&r.length!==i)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");let u=[];for(let e=0;e<i;++e){let i=null!==r?r[e]:null,p={boxes:[],classes:[],scores:[]},h=o[e],m=s[e];for(let e=0;e<l;++e){let r,o=h[e],s=[];if(n){r=o.sigmoid().data;for(let e=0;e<r.length;++e)r[e]>t&&s.push(e)}else{let e=(0,a.max)(o.data)[1];if(e===d-1)continue;if(r=(0,a.softmax)(o.data),r[e]<t)continue;s.push(e)}for(const t of s){let n=m[e].data;n=c(n),null!==i&&(n=n.map(((e,t)=>e*i[(t+1)%2]))),p.boxes.push(n),p.classes.push(t),p.scores.push(r[t])}}u.push(p)}return u}function u(e,t=null){const r=e.logits,n=r.dims[0];if(null!==t&&t.length!==n)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");const o=[];for(let e=0;e<n;++e){const n=null!==t?t[e]:null;let s=r[e];null!==n&&(s=(0,i.interpolate)(s,n,"bilinear",!1));const[a,l]=n??s.dims.slice(-2),c=new i.Tensor("int32",new Int32Array(a*l),[a,l]),d=s[0].data,u=c.data;for(let e=1;e<s.dims[0];++e){const t=s[e].data;for(let r=0;r<t.length;++r)t[r]>d[r]&&(d[r]=t[r],u[r]=e)}const p=new Array(s.dims[0]);for(let e=0;e<u.length;++e){const t=u[e];p[t]=t}const h=p.filter((e=>void 0!==e));o.push({segmentation:c,labels:h})}return o}function p(e,t,r,n){const o=[],s=[],i=[];for(let l=0;l<e.dims[0];++l){const c=e[l],d=t[l],u=(0,a.max)(c.data)[1];if(u===n)continue;const p=(0,a.softmax)(c.data)[u];p>r&&(o.push(d),s.push(p),i.push(u))}return[o,s,i]}function h(e,t,r,n=.5,o=.8){const s=[];let a=0,i=0;const l=t[r].data;for(let t=0;t<e.length;++t)e[t]===r&&(s.push(t),++a),l[t]>=n&&++i;let c=a>0&&i>0;if(c){c=a/i>o}return[c,s]}function m(e,t,r,n,o,s=null,a=null){const[l,c]=a??e[0].dims,d=new i.Tensor("int32",new Int32Array(l*c),[l,c]),u=[];if(null!==a)for(let t=0;t<e.length;++t)e[t]=(0,i.interpolate)(e[t],a,"bilinear",!1);const p=new Int32Array(e[0].data.length),m=new Float32Array(e[0].data.length);for(let r=0;r<e.length;++r){let n=t[r];const o=e[r].data;for(let e=0;e<o.length;++e)o[e]*=n,o[e]>m[e]&&(p[e]=r,m[e]=o[e])}let _=0;const f=d.data;for(let s=0;s<r.length;++s){const a=r[s],[i,l]=h(p,e,s,n,o);if(i){++_;for(const e of l)f[e]=_;u.push({id:_,label_id:a,score:t[s]})}}return[d,u]}function _(e,t=.5,r=.5,n=.8,o=null,s=null){null===o&&(console.warn("`label_ids_to_fuse` unset. No instance will be fused."),o=new Set);const a=e.class_queries_logits??e.logits,l=(e.masks_queries_logits??e.pred_masks).sigmoid();let[c,d,u]=a.dims;if(u-=1,null!==s&&s.length!==c)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<c;++e){let c=null!==s?s[e]:null,d=a[e],_=l[e],[f,g,M]=p(d,_,t,u);if(0===M.length){let[e,t]=c??_.dims.slice(-2),r=new i.Tensor("int32",new Int32Array(e*t).fill(-1),[e,t]);h.push({segmentation:r,segments_info:[]});continue}let[w,T]=m(f,g,M,r,n,o,c);h.push({segmentation:w,segments_info:T})}return h}function f(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 g(e,t,r=0,n=null){const o=e/t;let s=(0,a.bankers_round)(o)*t;return null!==n&&s>n&&(s=Math.floor(o)*t),s<r&&(s=Math.ceil(o)*t),s}function M([e,t],r){return[Math.max(Math.floor(e/r),1)*r,Math.max(Math.floor(t/r),1)*r]}class w extends n.Callable{constructor(e){super(),this.config=e}}class T extends w{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,r=2){const n=e.height,o=e.width,s=t.height,a=t.width;let i=Math.min(n,s),l=Math.min(o,a);return i===n&&l===o?e:(n>o?l=Math.floor(o*i/n):o>n&&(i=Math.floor(n*l/o)),await e.resize(l,i,{resample:r}))}async crop_margin(e,t=200){const r=e.clone().grayscale(),n=(0,a.min)(r.data)[0],o=(0,a.max)(r.data)[0]-n;if(0===o)return e;const s=t/255;let i=r.width,l=r.height,c=0,d=0;const u=r.data;for(let e=0;e<r.height;++e){const t=e*r.width;for(let a=0;a<r.width;++a)(u[t+a]-n)/o<s&&(i=Math.min(i,a),l=Math.min(l,e),c=Math.max(c,a),d=Math.max(d,e))}return e=await e.crop([i,l,c,d])}pad_image(e,t,r,{mode:n="constant",center:s=!1,constant_values:a=0}={}){const[i,l,c]=t;let d,u;if("number"==typeof r?(d=r,u=r):(d=r.width,u=r.height),d!==l||u!==i){const r=new Float32Array(d*u*c);if(Array.isArray(a))for(let e=0;e<r.length;++e)r[e]=a[e%c];else 0!==a&&r.fill(a);const[p,h]=s?[Math.floor((d-l)/2),Math.floor((u-i)/2)]:[0,0];for(let t=0;t<i;++t){const n=(t+h)*d,o=t*l;for(let t=0;t<l;++t){const s=(n+t+p)*c,a=(o+t)*c;for(let t=0;t<c;++t)r[s+t]=e[a+t]}}if("symmetric"===n){if(s)throw new Error("`center` padding is not supported when `mode` is set to `symmetric`.");const t=i-1,n=l-1;for(let s=0;s<u;++s){const a=s*d,u=(0,o.calculateReflectOffset)(s,t)*l;for(let t=0;t<d;++t){if(s<i&&t<l)continue;const d=(a+t)*c,p=(u+(0,o.calculateReflectOffset)(t,n))*c;for(let t=0;t<c;++t)r[d+t]=e[p+t]}}}e=r,t=[u,d,c]}return[e,t]}rescale(e){for(let t=0;t<e.length;++t)e[t]=this.rescale_factor*e[t]}get_resize_output_image_size(e,t){const[r,n]=e.size;let o,s;if(this.do_thumbnail){const{height:e,width:r}=t;o=Math.min(e,r)}else Number.isInteger(t)?(o=t,s=this.config.max_size??o):void 0!==t&&(o=t.shortest_edge,s=t.longest_edge);if(void 0!==o||void 0!==s){const e=void 0===o?1:Math.max(o/r,o/n),t=r*e,a=n*e,i=void 0===s?1:Math.min(s/t,s/a);let l=Math.floor(Number((t*i).toFixed(2))),c=Math.floor(Number((a*i).toFixed(2)));return void 0!==this.size_divisibility&&([l,c]=M([l,c],this.size_divisibility)),[l,c]}if(void 0!==t&&void 0!==t.width&&void 0!==t.height){let e=t.width,o=t.height;if(this.config.keep_aspect_ratio&&this.config.ensure_multiple_of){let t=o/n,s=e/r;Math.abs(1-s)<Math.abs(1-t)?t=s:s=t,o=g(t*n,this.config.ensure_multiple_of),e=g(s*r,this.config.ensure_multiple_of)}return[e,o]}if(void 0!==this.size_divisibility)return M([r,n],this.size_divisibility);throw new Error(`Could not resize image due to unsupported \`this.size\` option in config: ${JSON.stringify(t)}`)}async resize(e){const[t,r]=this.get_resize_output_image_size(e,this.size);return await e.resize(t,r,{resample:this.resample})}async preprocess(e,{do_normalize:t=null,do_pad:r=null,do_convert_rgb:n=null,do_convert_grayscale:o=null,do_flip_channel_order:s=null}={}){this.do_crop_margin&&(e=await this.crop_margin(e));const[a,l]=e.size;if(n??this.do_convert_rgb?e=e.rgb():o&&(e=e.grayscale()),this.do_resize&&(e=await this.resize(e)),this.do_thumbnail&&(e=await this.thumbnail(e,this.size,this.resample)),this.do_center_crop){let t,r;Number.isInteger(this.crop_size)?(t=this.crop_size,r=this.crop_size):(t=this.crop_size.width,r=this.crop_size.height),e=await e.center_crop(t,r)}const c=[e.height,e.width];let d=Float32Array.from(e.data),u=[e.height,e.width,e.channels];if(this.do_rescale&&this.rescale(d),t??this.do_normalize){let t=this.image_mean;Array.isArray(this.image_mean)||(t=new Array(e.channels).fill(t));let r=this.image_std;if(Array.isArray(this.image_std)||(r=new Array(e.channels).fill(t)),t.length!==e.channels||r.length!==e.channels)throw new Error(`When set to arrays, the length of \`image_mean\` (${t.length}) and \`image_std\` (${r.length}) must match the number of channels in the image (${e.channels}).`);for(let n=0;n<d.length;n+=e.channels)for(let o=0;o<e.channels;++o)d[n+o]=(d[n+o]-t[o])/r[o]}if(r??this.do_pad)if(this.pad_size){const t=this.pad_image(d,[e.height,e.width,e.channels],this.pad_size);[d,u]=t}else if(this.size_divisibility){const[e,t]=M([u[1],u[0]],this.size_divisibility);[d,u]=this.pad_image(d,u,{width:e,height:t})}if(s??this.do_flip_channel_order){if(3!==u[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,a],reshaped_input_size:c,pixel_values:new i.Tensor("float32",d,u).permute(2,0,1)}}async _call(e,...t){Array.isArray(e)||(e=[e]);const r=await Promise.all(e.map((e=>this.preprocess(e))));return{pixel_values:(0,i.stack)(r.map((e=>e.pixel_values)),0),original_sizes:r.map((e=>e.original_size)),reshaped_input_sizes:r.map((e=>e.reshaped_input_size))}}}class b extends T{post_process_semantic_segmentation(...e){return u(...e)}}class x extends T{post_process_semantic_segmentation(...e){return u(...e)}}class k extends T{}class y extends T{}class F extends y{}class P extends T{}class C extends T{}class v extends T{}class S extends v{}class A extends T{}class E extends T{}class L extends T{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 r=Math.floor(t/this.crop_pct),[n,o]=this.get_resize_output_image_size(e,{shortest_edge:r});e=await e.resize(n,o,{resample:this.resample}),e=await e.center_crop(t,t)}else e=await e.resize(t,t,{resample:this.resample});return e}}class z extends L{}class I extends T{}class B extends T{}class N extends T{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 O extends T{}class D extends T{}class V extends T{}class j extends T{}class R extends T{}class G extends R{}class q extends T{post_process_object_detection(...e){return d(...e)}}class $ extends q{}class W extends T{post_process_object_detection(...e){return d(...e)}}class U extends T{}class X extends T{}class Q extends T{pad_image(e,t,r,n={}){const[o,s,a]=t;let i=this.image_mean;Array.isArray(this.image_mean)||(i=new Array(a).fill(i));let l=this.image_std;Array.isArray(l)||(l=new Array(a).fill(i));const c=i.map(((e,t)=>-e/l[t]));return super.pad_image(e,t,r,{center:!0,constant_values:c,...n})}}class H extends Q{}class Y extends T{async _call(e){const t=await super._call(e),r=[t.pixel_values.dims[0],64,64],n=(0,i.full)(r,1n);return{...t,pixel_mask:n}}post_process_object_detection(...e){return d(...e)}post_process_panoptic_segmentation(...e){return _(...e)}post_process_instance_segmentation(){throw Error("Not implemented yet")}}class J extends T{post_process_panoptic_segmentation(...e){return _(...e)}post_process_instance_segmentation(){throw Error("Not implemented yet")}}class K extends T{post_process_object_detection(...e){return d(...e)}}class Z extends T{reshape_input_points(e,t,r,n=!1){e=structuredClone(e);let s=(0,o.calculateDimensions)(e);if(3===s.length)n||(s=[1,...s]),e=[e];else if(4!==s.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 n=0;n<e.length;++n){let o=t[n],s=r[n],a=[s[0]/o[0],s[1]/o[1]];for(let t=0;t<e[n].length;++t)for(let r=0;r<e[n][t].length;++r)for(let o=0;o<e[n][t][r].length;++o)e[n][t][r][o]*=a[o%2]}return new i.Tensor("float32",Float32Array.from(e.flat(1/0)),s)}add_input_labels(e,t){let r=(0,o.calculateDimensions)(e);if(2===r.length)r=[1,...r],e=[e];else if(3!==r.length)throw Error("The input_points must be a 4D tensor of shape `batch_size`, `point_batch_size`, `nb_points_per_image`, `2`.");if(r.some(((e,r)=>e!==t.dims[r])))throw Error(`The first ${r.length} dimensions of 'input_points' and 'input_labels' must be the same.`);return new i.Tensor("int64",e.flat(1/0).map(BigInt),r)}async _call(e,{input_points:t=null,input_labels:r=null,input_boxes:n=null}={}){const o=await super._call(e);if(t&&(o.input_points=this.reshape_input_points(t,o.original_sizes,o.reshaped_input_sizes)),r){if(!o.input_points)throw Error("`input_points` must be provided if `input_labels` are provided.");o.input_labels=this.add_input_labels(r,o.input_points)}return n&&(o.input_boxes=this.reshape_input_points(n,o.original_sizes,o.reshaped_input_sizes,!0)),o}async post_process_masks(e,t,r,{mask_threshold:n=0,binarize:o=!0,pad_size:s=null}={}){const a=[],l=[(s=s??this.pad_size).height,s.width];for(let s=0;s<t.length;++s){const c=t[s],d=r[s];let u=await(0,i.interpolate_4d)(e[s],{mode:"bilinear",size:l});if(u=u.slice(null,null,[0,d[0]],[0,d[1]]),u=await(0,i.interpolate_4d)(u,{mode:"bilinear",size:c}),o){const e=u.data,t=new Uint8Array(e.length);for(let r=0;r<e.length;++r)e[r]>n&&(t[r]=1);u=new i.Tensor("bool",t,u.dims)}a.push(u)}return a}generate_crop_boxes(e,t,{crop_n_layers:r=0,overlap_ratio:n=512/1500,points_per_crop:o=32,crop_n_points_downscale_factor:s=1}={}){}}class ee extends T{pad_image(e,t,r,n={}){const[o,s,a]=t;return super.pad_image(e,t,{width:s+(r-s%r)%r,height:o+(r-o%r)%r},{mode:"symmetric",center:!1,constant_values:-1,...n})}}class te extends T{async _call(e,t){Array.isArray(e)||(e=[e]),Array.isArray(t)||(t=[t]);const r=await Promise.all(e.map((e=>this.preprocess(e)))),n=await Promise.all(t.map((e=>this.preprocess(e,{do_normalize:!1,do_convert_rgb:!1,do_convert_grayscale:!0}))));return{pixel_values:(0,i.stack)(r.map(((e,t)=>(0,i.cat)([e.pixel_values,n[t].pixel_values],0))),0),original_sizes:r.map((e=>e.original_size)),reshaped_input_sizes:r.map((e=>e.reshaped_input_size))}}}class re extends w{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}),r=t.data,n=(0,a.max)(r)[0];for(let e=0;e<r.length;++e)r[e]=(Math.max(r[e],n-8)+4)/4;return t}async _call(e){let t;f(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 ne extends w{_zero_mean_unit_var_norm(e){const t=e.reduce(((e,t)=>e+t),0)/e.length,r=e.reduce(((e,r)=>e+(r-t)**2),0)/e.length;return e.map((e=>(e-t)/Math.sqrt(r+1e-7)))}async _call(e){f(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 r=[1,t.length];return{input_values:new i.Tensor("float32",t,r),attention_mask:new i.Tensor("int64",new BigInt64Array(t.length).fill(1n),r)}}}class oe extends w{constructor(e){super(e);const t=this.config.sampling_rate,r=(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<r.length;++e)r[e].push(0);this.mel_filters=r,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:r=2,do_normalize_per_mel_bins:n=!0,return_attention_mask:o=!0}={}){f(e,"SeamlessM4TFeatureExtractor");let s,a=await this._extract_fbank_features(e,this.config.max_length);if(n){const[e,t]=a.dims,r=a.data;for(let n=0;n<t;++n){let o=0;for(let s=0;s<e;++s)o+=r[s*t+n];const s=o/e;let a=0;for(let o=0;o<e;++o)a+=(r[o*t+n]-s)**2;a/=e-1;const i=Math.sqrt(a+1e-7);for(let o=0;o<e;++o){const e=o*t+n;r[e]=(r[e]-s)/i}}}if(t){const[e,t]=a.dims,n=a.data,l=e%r;if(l>0){const r=new Float32Array(t*(e+l));r.set(n),r.fill(this.config.padding_value,n.length);const c=e+l;a=new i.Tensor(a.type,r,[c,t]),o&&(s=new i.Tensor("int64",new BigInt64Array(c),[1,c]),s.data.fill(1n,0,e))}}const[l,c]=a.dims,d=this.config.stride;if(0!==l%d)throw new Error(`The number of frames (${l}) must be a multiple of the stride (${d}).`);const u=a.view(1,Math.floor(l/d),c*d),p={input_features:u};if(o){const e=u.dims[1],t=new BigInt64Array(e);if(s){const e=s.data;for(let r=1,n=0;r<l;r+=d,++n)t[n]=e[r]}else t.fill(1n);p.attention_mask=new i.Tensor("int64",t,[1,e])}return p}}class se extends w{constructor(e){super(e);const t=this.config.sampling_rate,r=(0,l.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<r.length;++e)r[e].push(0);this.mel_filters=r,this.window=(0,l.window_function)(400,"hann",{periodic:!1}),this.mean=this.config.mean,this.std=this.config.std}async _extract_fbank_features(e,t){return(0,l.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,max_num_frames:t,transpose:!0})}async _call(e){f(e,"ASTFeatureExtractor");const t=await this._extract_fbank_features(e,this.config.max_length);if(this.config.do_normalize){const e=2*this.std,r=t.data;for(let t=0;t<r.length;++t)r[t]=(r[t]-this.mean)/e}return{input_values:t.unsqueeze_(0)}}}class ae extends w{constructor(e){super(e),this.mel_filters=(0,l.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,null,"htk"),this.mel_filters_slaney=(0,l.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,"slaney","slaney"),this.window=(0,l.window_function)(this.config.fft_window_size,"hann")}async _get_input_mel(e,t,r,n){let o,s=!1;const a=e.length-t;if(a>0){if("rand_trunc"!==r)throw new Error(`Truncation strategy "${r}" not implemented`);{s=!0;const r=Math.floor(Math.random()*(a+1));e=e.subarray(r,r+t),o=await this._extract_fbank_features(e,this.mel_filters_slaney,this.config.nb_max_samples)}}else{if(a<0){let r=new Float64Array(t);if(r.set(e),"repeat"===n)for(let n=e.length;n<t;n+=e.length)r.set(e.subarray(0,Math.min(e.length,t-n)),n);else if("repeatpad"===n)for(let t=e.length;t<-a;t+=e.length)r.set(e,t);e=r}if("fusion"===r)throw new Error(`Truncation strategy "${r}" not implemented`);o=await this._extract_fbank_features(e,this.mel_filters_slaney,this.config.nb_max_samples)}return o.unsqueeze_(0)}async _extract_fbank_features(e,t,r=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:r,do_pad:!1,transpose:!0})}async _call(e,{max_length:t=null}={}){f(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 ie extends w{async _call(e){f(e,"PyAnnoteFeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));const t=[1,1,e.length];return{input_values:new i.Tensor("float32",e,t)}}samples_to_frames(e){return(e-this.config.offset)/this.config.step}post_process_speaker_diarization(e,t){const r=t/this.samples_to_frames(t)/this.config.sampling_rate,n=[];for(const t of e.tolist()){const e=[];let o=-1;for(let r=0;r<t.length;++r){const n=(0,a.softmax)(t[r]),[s,i]=(0,a.max)(n),[l,c]=[r,r+1];i!==o?(o=i,e.push({id:i,start:l,end:c,score:s})):(e.at(-1).end=c,e.at(-1).score+=s)}n.push(e.map((({id:e,start:t,end:n,score:o})=>({id:e,start:t*r,end:n*r,confidence:o/(n-t)}))))}return n}}class le extends w{constructor(e){super(e);const t=this.config.sampling_rate,r=(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<r.length;++e)r[e].push(0);this.mel_filters=r,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){f(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,r=t.data,[n,o,s]=t.dims;for(let t=0;t<n;++t){const n=t*o*s,a=t*s;for(let t=0;t<o;++t){const o=n+t*s;for(let t=0;t<s;++t)r[o+t]-=e[a+t]}}}return{input_features:t}}}class ce extends w{}class de extends n.Callable{constructor(e){super(),this.feature_extractor=e}async _call(e,...t){return await this.feature_extractor(e,...t)}}class ue 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 _e extends de{async _call(e){return await this.feature_extractor(e)}}class fe extends de{}class ge extends de{constructor(e){super(e);const{tasks_answer_post_processing_type:t,task_prompts_without_inputs:r,task_prompts_with_input:n}=e.config;this.tasks_answer_post_processing_type=new Map(Object.entries(t??{})),this.task_prompts_without_inputs=new Map(Object.entries(r??{})),this.task_prompts_with_input=new Map(Object.entries(n??{})),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 r of e)if(this.task_prompts_without_inputs.has(r))t.push(this.task_prompts_without_inputs.get(r));else{for(const[e,n]of this.task_prompts_with_input)if(r.includes(e)){t.push(n.replaceAll("{input}",r).replaceAll(e,""));break}t.length!==e.length&&t.push(r)}return t}post_process_generation(e,t,r){const n=this.tasks_answer_post_processing_type.get(t)??"pure_text";let o;switch(e=e.replaceAll("<s>","").replaceAll("</s>",""),n){case"pure_text":o=e;break;case"description_with_bboxes":case"bboxes":case"phrase_grounding":case"ocr":const s="ocr"===n?"quad_boxes":"bboxes",a=e.matchAll(this.regexes[s]),i=[],l=[];for(const[e,t,...n]of a)i.push(t?t.trim():i.at(-1)??""),l.push(n.map(((e,t)=>(Number(e)+.5)/this.size_per_bin*r[t%2])));o={labels:i,[s]:l};break;default:throw new Error(`Task "${t}" (of type "${n}") not yet implemented.`)}return{[t]:o}}}class Me{static FEATURE_EXTRACTOR_CLASS_MAPPING={ImageFeatureExtractor:T,WhisperFeatureExtractor:re,ViTFeatureExtractor:I,MobileViTFeatureExtractor:R,MobileViTImageProcessor:G,MobileNetV1FeatureExtractor:O,MobileNetV2FeatureExtractor:D,MobileNetV3FeatureExtractor:V,MobileNetV4FeatureExtractor:j,OwlViTFeatureExtractor:q,Owlv2ImageProcessor:$,CLIPFeatureExtractor:v,CLIPImageProcessor:S,Florence2Processor:ge,ChineseCLIPFeatureExtractor:A,SiglipImageProcessor:E,ConvNextFeatureExtractor:L,ConvNextImageProcessor:z,SegformerFeatureExtractor:x,SapiensFeatureExtractor:b,BitImageProcessor:P,DPTImageProcessor:F,DPTFeatureExtractor:y,PvtImageProcessor:k,GLPNFeatureExtractor:C,BeitFeatureExtractor:X,DeiTFeatureExtractor:U,DetrFeatureExtractor:Y,RTDetrImageProcessor:W,MaskFormerFeatureExtractor:J,YolosFeatureExtractor:K,DonutFeatureExtractor:Q,NougatImageProcessor:H,EfficientNetImageProcessor:N,ViTImageProcessor:B,VitMatteImageProcessor:te,SamImageProcessor:Z,Swin2SRImageProcessor:ee,Wav2Vec2FeatureExtractor:ne,SeamlessM4TFeatureExtractor:oe,SpeechT5FeatureExtractor:ce,ASTFeatureExtractor:se,ClapFeatureExtractor:ae,PyAnnoteFeatureExtractor:ie,WeSpeakerFeatureExtractor:le};static PROCESSOR_CLASS_MAPPING={WhisperProcessor:pe,Wav2Vec2ProcessorWithLM:he,PyAnnoteProcessor:me,SamProcessor:ue,SpeechT5Processor:_e,OwlViTProcessor:fe,Florence2Processor:ge};static async from_pretrained(e,{progress_callback:t=null,config:r=null,cache_dir:n=null,local_files_only:o=!1,revision:a="main"}={}){let i=r??await(0,s.getModelJSON)(e,"preprocessor_config.json",!0,{progress_callback:t,config:r,cache_dir:n,local_files_only:o,revision:a}),l=i.feature_extractor_type??i.image_processor_type,c=this.FEATURE_EXTRACTOR_CLASS_MAPPING[l];if(!c){if(void 0===i.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.`),c=T}return new(this.PROCESSOR_CLASS_MAPPING[i.processor_class]??de)(new c(i))}}},"./src/tokenizers.js":
 /*!***************************!*\
   !*** ./src/tokenizers.js ***!
-  \***************************/(e,t,r)=>{r.r(t),r.d(t,{AlbertTokenizer:()=>xe,AutoTokenizer:()=>ht,BartTokenizer:()=>Ne,BertTokenizer:()=>be,BlenderbotSmallTokenizer:()=>lt,BlenderbotTokenizer:()=>it,BloomTokenizer:()=>je,CLIPTokenizer:()=>nt,CamembertTokenizer:()=>Ee,CodeGenTokenizer:()=>rt,CodeLlamaTokenizer:()=>qe,CohereTokenizer:()=>pt,ConvBertTokenizer:()=>ve,DebertaTokenizer:()=>Fe,DebertaV2Tokenizer:()=>Pe,DistilBertTokenizer:()=>Ae,ElectraTokenizer:()=>ze,EsmTokenizer:()=>Qe,FalconTokenizer:()=>Ue,GPT2Tokenizer:()=>Be,GPTNeoXTokenizer:()=>Xe,GemmaTokenizer:()=>Ye,Grok1Tokenizer:()=>Je,HerbertTokenizer:()=>Ce,LlamaTokenizer:()=>Ge,M2M100Tokenizer:()=>et,MBart50Tokenizer:()=>De,MBartTokenizer:()=>Oe,MPNetTokenizer:()=>We,MarianTokenizer:()=>st,MobileBertTokenizer:()=>ke,NllbTokenizer:()=>Ze,NougatTokenizer:()=>dt,PreTrainedTokenizer:()=>Te,Qwen2Tokenizer:()=>He,RoFormerTokenizer:()=>Se,RobertaTokenizer:()=>Ve,SiglipTokenizer:()=>ot,SpeechT5Tokenizer:()=>ct,SqueezeBertTokenizer:()=>ye,T5Tokenizer:()=>Ie,TokenizerModel:()=>k,VitsTokenizer:()=>ut,Wav2Vec2CTCTokenizer:()=>at,WhisperTokenizer:()=>tt,XLMRobertaTokenizer:()=>$e,XLMTokenizer:()=>Le,is_chinese_char:()=>M});var n=r(/*! ./utils/generic.js */"./src/utils/generic.js"),o=r(/*! ./utils/core.js */"./src/utils/core.js"),s=r(/*! ./utils/hub.js */"./src/utils/hub.js"),a=r(/*! ./utils/maths.js */"./src/utils/maths.js"),i=r(/*! ./utils/tensor.js */"./src/utils/tensor.js"),l=r(/*! ./utils/data-structures.js */"./src/utils/data-structures.js"),c=r(/*! @huggingface/jinja */"./node_modules/@huggingface/jinja/dist/index.js"),d=r(/*! ./models/whisper/common_whisper.js */"./src/models/whisper/common_whisper.js"),u=r(/*! ./utils/constants.js */"./src/utils/constants.js");async function p(e,t){const r=await Promise.all([(0,s.getModelJSON)(e,"tokenizer.json",!0,t),(0,s.getModelJSON)(e,"tokenizer_config.json",!0,t)]);return null!==t.legacy&&(r[1].legacy=t.legacy),r}function h(e,t=!0){if(void 0!==e.Regex){let t=e.Regex.replace(/\\([#&~])/g,"$1");for(const[e,r]of b)t=t.replaceAll(e,r);return new RegExp(t,"gu")}if(void 0!==e.String){const r=(0,o.escapeRegExp)(e.String);return new RegExp(t?r:`(${r})`,"gu")}return console.warn("Unknown pattern type:",e),null}function m(e){return new Map(Object.entries(e))}function _(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 ${t.length}.`)}}function f(e){return e.replace(/ \./g,".").replace(/ \?/g,"?").replace(/ \!/g,"!").replace(/ ,/g,",").replace(/ \' /g,"'").replace(/ n\'t/g,"n't").replace(/ \'m/g,"'m").replace(/ \'s/g,"'s").replace(/ \'ve/g,"'ve").replace(/ \'re/g,"'re")}function g(e){return e.replace(/[\u0300-\u036f]/g,"")}function M(e){return e>=19968&&e<=40959||e>=13312&&e<=19903||e>=131072&&e<=173791||e>=173824&&e<=177983||e>=177984&&e<=178207||e>=178208&&e<=183983||e>=63744&&e<=64255||e>=194560&&e<=195103}const w="\\p{P}\\u0021-\\u002F\\u003A-\\u0040\\u005B-\\u0060\\u007B-\\u007E",T=new RegExp(`^[${w}]+$`,"gu"),b=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 k extends n.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 y(e);case"Unigram":return new F(e,...t);case"BPE":return new v(e);default:if(e.vocab)return new S(e,...t);throw new Error(`Unknown TokenizerModel type: ${e.type}`)}}_call(e){return e=this.encode(e),this.fuse_unk&&(e=function(e,t,r){const n=[];let o=0;for(;o<e.length;)if(n.push(e[o]),(t.get(e[o])??r)===r)for(;++o<e.length&&(t.get(e[o])??r)===r;)t.get(n.at(-1))!==r&&(n[n.length-1]+=e[o]);else++o;return n}(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 y extends k{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 r of e){const e=[...r];if(e.length>this.max_input_chars_per_word){t.push(this.unk_token);continue}let n=!1,o=0;const s=[];for(;o<e.length;){let t=e.length,r=null;for(;o<t;){let n=e.slice(o,t).join("");if(o>0&&(n=this.config.continuing_subword_prefix+n),this.tokens_to_ids.has(n)){r=n;break}--t}if(null===r){n=!0;break}s.push(r),o=t}n?t.push(this.unk_token):t.push(...s)}return t}}class F extends k{constructor(e,t){super(e);const r=e.vocab.length;this.vocab=new Array(r),this.scores=new Array(r);for(let t=0;t<r;++t){const r=e.vocab[t];this.vocab[t]=r[0],this.scores[t]=r[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,a.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,r=t.length;let n=0;for(;n<r;){const r=1;let o=!1;const s=[];for(let a of this.trie.commonPrefixSearch(t.slice(n))){s.push(a);const t=this.tokens_to_ids.get(a),i=this.scores[t],l=a.length;e.insert(n,l,i,t),o||l!==r||(o=!0)}o||e.insert(n,r,this.unkScore,this.unk_token_id),n+=r}}tokenize(e){const t=new l.TokenLattice(e,this.bosTokenId,this.eosTokenId);return this.populateNodes(t),t.tokens()}encode(e){const t=[];for(const r of e){const e=this.tokenize(r);t.push(...e)}return t}}const P=(()=>{const e=[...Array.from({length:"~".charCodeAt(0)-"!".charCodeAt(0)+1},((e,t)=>t+"!".charCodeAt(0))),...Array.from({length:"¬".charCodeAt(0)-"¡".charCodeAt(0)+1},((e,t)=>t+"¡".charCodeAt(0))),...Array.from({length:"ÿ".charCodeAt(0)-"®".charCodeAt(0)+1},((e,t)=>t+"®".charCodeAt(0)))],t=e.slice();let r=0;for(let n=0;n<256;++n)e.includes(n)||(e.push(n),t.push(256+r),r+=1);const n=t.map((e=>String.fromCharCode(e)));return Object.fromEntries(e.map(((e,t)=>[e,n[t]])))})(),C=(0,o.reverseDictionary)(P);class v extends k{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 r=Array.from(e);this.end_of_word_suffix&&(r[r.length-1]+=this.end_of_word_suffix);let n=[];if(r.length>1){const e=new l.PriorityQueue(((e,t)=>e.score<t.score));let t={token:r[0],bias:0,prev:null,next:null},o=t;for(let t=1;t<r.length;++t){const n={bias:t/r.length,token:r[t],prev:o,next:null};o.next=n,this._add_node(e,o),o=n}for(;!e.isEmpty();){const r=e.pop();if(r.deleted||!r.next||r.next.deleted)continue;if(r.deleted=!0,r.next.deleted=!0,r.prev){const e={...r.prev};r.prev.deleted=!0,r.prev=e,e.prev?e.prev.next=e:t=e}const n={token:r.token+r.next.token,bias:r.bias,prev:r.prev,next:r.next.next};n.prev?(n.prev.next=n,this._add_node(e,n.prev)):t=n,n.next&&(n.next.prev=n,this._add_node(e,n))}for(let e=t;null!==e;e=e.next)n.push(e.token)}else n=r;if(this.continuing_subword_suffix)for(let e=0;e<n.length-1;++e)n[e]+=this.continuing_subword_suffix;return this.cache.set(e,n),n}_add_node(e,t){const r=this.bpe_ranks.get(t.token+this.BPE_SPLIT_TOKEN+t.next.token);void 0!==r&&(t.score=r+t.bias,e.push(t))}encode(e){const t=[];for(const r of e){if(this.ignore_merges&&this.tokens_to_ids.has(r)){t.push(r);continue}const e=this.bpe(r);for(const r of e)if(this.tokens_to_ids.has(r))t.push(r);else if(this.byte_fallback){const e=Array.from(this.text_encoder.encode(r)).map((e=>`<0x${e.toString(16).toUpperCase().padStart(2,"0")}>`));e.every((e=>this.tokens_to_ids.has(e)))?t.push(...e):t.push(this.unk_token)}else t.push(this.unk_token)}return t}}class S extends k{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 A extends n.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"BertNormalizer":return new j(e);case"Precompiled":return new pe(e);case"Sequence":return new V(e);case"Replace":return new E(e);case"NFC":return new L(e);case"NFKC":return new z(e);case"NFKD":return new I(e);case"Strip":return new B(e);case"StripAccents":return new N(e);case"Lowercase":return new O(e);case"Prepend":return new D(e);default:throw new Error(`Unknown Normalizer type: ${e.type}`)}}normalize(e){throw Error("normalize should be implemented in subclass.")}_call(e){return this.normalize(e)}}class E extends A{normalize(e){const t=h(this.config.pattern);return null===t?e:e.replaceAll(t,this.config.content)}}class L extends A{normalize(e){return e=e.normalize("NFC")}}class z extends A{normalize(e){return e=e.normalize("NFKC")}}class I extends A{normalize(e){return e=e.normalize("NFKD")}}class B extends A{normalize(e){return this.config.strip_left&&this.config.strip_right?e=e.trim():(this.config.strip_left&&(e=e.trimStart()),this.config.strip_right&&(e=e.trimEnd())),e}}class N extends A{normalize(e){return e=g(e)}}class O extends A{normalize(e){return e=e.toLowerCase()}}class D extends A{normalize(e){return e=this.config.prepend+e}}class V extends A{constructor(e){super(e),this.normalizers=e.normalizers.map((e=>A.fromConfig(e)))}normalize(e){return this.normalizers.reduce(((e,t)=>t.normalize(e)),e)}}class j extends A{_tokenize_chinese_chars(e){const t=[];for(let r=0;r<e.length;++r){const n=e[r];M(n.charCodeAt(0))?(t.push(" "),t.push(n),t.push(" ")):t.push(n)}return t.join("")}stripAccents(e){return e.normalize("NFD").replace(/[\u0300-\u036f]/g,"")}_is_control(e){switch(e){case"\t":case"\n":case"\r":return!1;default:return/^\p{Cc}|\p{Cf}|\p{Co}|\p{Cs}$/u.test(e)}}_clean_text(e){const t=[];for(const r of e){const e=r.charCodeAt(0);0===e||65533===e||this._is_control(r)||(/^\s$/.test(r)?t.push(" "):t.push(r))}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 R extends n.Callable{static fromConfig(e){if(null===e)return null;switch(e.type){case"BertPreTokenizer":return new G(e);case"Sequence":return new he(e);case"Whitespace":return new me(e);case"WhitespaceSplit":return new _e(e);case"Metaspace":return new de(e);case"ByteLevel":return new q(e);case"Split":return new $(e);case"Punctuation":return new W(e);case"Digits":return new U(e);case"Replace":return new fe(e);default:throw new Error(`Unknown PreTokenizer type: ${e.type}`)}}pre_tokenize_text(e,t){throw Error("pre_tokenize_text should be implemented in subclass.")}pre_tokenize(e,t){return(Array.isArray(e)?e.map((e=>this.pre_tokenize_text(e,t))):this.pre_tokenize_text(e,t)).flat()}_call(e,t){return this.pre_tokenize(e,t)}}class G extends R{constructor(e){super(),this.pattern=new RegExp(`[^\\s${w}]+|[${w}]`,"gu")}pre_tokenize_text(e,t){return e.trim().match(this.pattern)||[]}}class q extends R{constructor(e){super(),this.config=e,this.add_prefix_space=this.config.add_prefix_space,this.trim_offsets=this.config.trim_offsets,this.use_regex=this.config.use_regex??!0,this.pattern=/'s|'t|'re|'ve|'m|'ll|'d| ?\p{L}+| ?\p{N}+| ?[^\s\p{L}\p{N}]+|\s+(?!\S)|\s+/gu,this.byte_encoder=P,this.text_encoder=new TextEncoder}pre_tokenize_text(e,t){this.add_prefix_space&&!e.startsWith(" ")&&(e=" "+e);return(this.use_regex?e.match(this.pattern)||[]:[e]).map((e=>Array.from(this.text_encoder.encode(e),(e=>this.byte_encoder[e])).join("")))}}class $ extends R{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 r=[];let n=0;for(const o of e.matchAll(t)){const t=o[0];n<o.index&&r.push(e.slice(n,o.index)),t.length>0&&r.push(t),n=o.index+t.length}return n<e.length&&r.push(e.slice(n)),r}(e,this.pattern)}}class W extends R{constructor(e){super(),this.config=e,this.pattern=new RegExp(`[^${w}]+|[${w}]+`,"gu")}pre_tokenize_text(e,t){return e.match(this.pattern)||[]}}class U extends R{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 X extends n.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"TemplateProcessing":return new Y(e);case"ByteLevel":return new J(e);case"RobertaProcessing":return new H(e);case"BertProcessing":return new Q(e);case"Sequence":return new K(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 Q extends X{constructor(e){super(e),this.cls=e.cls[0],this.sep=e.sep[0]}post_process(e,t=null,{add_special_tokens:r=!0}={}){r&&(e=(0,o.mergeArrays)([this.cls],e,[this.sep]));let n=new Array(e.length).fill(0);if(null!==t){const s=r&&this instanceof H?[this.sep]:[],a=r?[this.sep]:[];e=(0,o.mergeArrays)(e,s,t,a),n=(0,o.mergeArrays)(n,new Array(t.length+s.length+a.length).fill(1))}return{tokens:e,token_type_ids:n}}}class H extends Q{}class Y extends X{constructor(e){super(e),this.single=e.single,this.pair=e.pair}post_process(e,t=null,{add_special_tokens:r=!0}={}){const n=null===t?this.single:this.pair;let s=[],a=[];for(const i of n)"SpecialToken"in i?r&&(s.push(i.SpecialToken.id),a.push(i.SpecialToken.type_id)):"Sequence"in i&&("A"===i.Sequence.id?(s=(0,o.mergeArrays)(s,e),a=(0,o.mergeArrays)(a,new Array(e.length).fill(i.Sequence.type_id))):"B"===i.Sequence.id&&(s=(0,o.mergeArrays)(s,t),a=(0,o.mergeArrays)(a,new Array(t.length).fill(i.Sequence.type_id))));return{tokens:s,token_type_ids:a}}}class J extends X{post_process(e,t=null){return t&&(e=(0,o.mergeArrays)(e,t)),{tokens:e}}}class K extends X{constructor(e){super(e),this.processors=e.processors.map((e=>X.fromConfig(e)))}post_process(e,t=null,r={}){let n;for(const o of this.processors)if(o instanceof J){if(e=o.post_process(e).tokens,t){t=o.post_process(t).tokens}}else{const s=o.post_process(e,t,r);e=s.tokens,n=s.token_type_ids}return{tokens:e,token_type_ids:n}}}class Z extends n.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 oe(e);case"Metaspace":return new ue(e);case"ByteLevel":return new se(e);case"Replace":return new ee(e);case"ByteFallback":return new te(e);case"Fuse":return new re(e);case"Strip":return new ne(e);case"Sequence":return new ie(e);case"CTC":return new ae(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 r=[];for(const n of e){let e=null;if(6===n.length&&n.startsWith("<0x")&&n.endsWith(">")){const t=parseInt(n.slice(3,5),16);isNaN(t)||(e=t)}if(null!==e)r.push(e);else{if(r.length>0){const e=this.text_decoder.decode(Uint8Array.from(r));t.push(e),r=[]}t.push(n)}}if(r.length>0){const e=this.text_decoder.decode(Uint8Array.from(r));t.push(e),r=[]}return t}}class re extends Z{decode_chain(e){return[e.join("")]}}class ne 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 r=0;r<this.start&&e[r]===this.content;++r)t=r+1;let r=e.length;for(let t=0;t<this.stop;++t){const n=e.length-t-1;if(e[n]!==this.content)break;r=n}return e.slice(t,r)}))}}class oe 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=f(e)),e)))}}class se extends Z{constructor(e){super(e),this.byte_decoder=C,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(""),r=new Uint8Array([...t].map((e=>this.byte_decoder[e])));return this.text_decoder.decode(r)}decode_chain(e){const t=[];let r=[];for(const n of e)void 0!==this.added_tokens.find((e=>e.content===n))?(r.length>0&&(t.push(this.convert_tokens_to_string(r)),r=[]),t.push(n)):r.push(n);return r.length>0&&t.push(this.convert_tokens_to_string(r)),t}}class ae 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 r=1;r<e.length;++r)e[r]!==t.at(-1)&&t.push(e[r]);let r=t.filter((e=>e!==this.pad_token)).join("");return this.cleanup&&(r=f(r).replaceAll(this.word_delimiter_token," ").trim()),r}decode_chain(e){return[this.convert_tokens_to_string(e)]}}class ie 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,r)=>t.replaceAll(this.suffix,r===e.length-1?"":" ")))}}class ce extends Z{decode_chain(e){let t="";for(let r=1;r<e.length;r+=2)t+=e[r];return[t]}}class de extends R{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 r=e.replaceAll(" ",this.strRep);return this.addPrefixSpace&&!r.startsWith(this.replacement)&&("always"===this.prepend_scheme||"first"===this.prepend_scheme&&0===t)&&(r=this.strRep+r),[r]}}class ue extends Z{constructor(e){super(e),this.addPrefixSpace=e.add_prefix_space,this.replacement=e.replacement}decode_chain(e){const t=[];for(let r=0;r<e.length;++r){let n=e[r].replaceAll(this.replacement," ");this.addPrefixSpace&&0==r&&n.startsWith(" ")&&(n=n.substring(1)),t.push(n)}return t}}class pe extends A{constructor(e){super(e),this.charsmap=e.precompiled_charsmap}normalize(e){if((e=(e=e.replace(/[\u0001-\u0008\u000B\u000E-\u001F\u007F\u008F\u009F]/gm,"")).replace(/[\u0009\u000A\u000C\u000D\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 R{constructor(e){super(),this.tokenizers=e.pretokenizers.map((e=>R.fromConfig(e)))}pre_tokenize_text(e,t){return this.tokenizers.reduce(((e,r)=>r.pre_tokenize(e,t)),[e])}}class me extends R{constructor(e){super()}pre_tokenize_text(e,t){return e.match(/\w+|[^\w\s]+/g)||[]}}class _e extends R{constructor(e){super()}pre_tokenize_text(e,t){return function(e){return e.match(/\S+/g)||[]}(e)}}class fe extends R{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 ge=["bos_token","eos_token","unk_token","sep_token","pad_token","cls_token","mask_token"];function Me(e,t,r,n){for(const s of Object.keys(e)){const a=t-e[s].length,i=r(s),l=new Array(a).fill(i);e[s]="right"===n?(0,o.mergeArrays)(e[s],l):(0,o.mergeArrays)(l,e[s])}}function we(e,t){for(const r of Object.keys(e))e[r].length=t}class Te extends n.Callable{return_token_type_ids=!1;padding_side="right";constructor(e,t){super(),this._tokenizer_config=t,this.normalizer=A.fromConfig(e.normalizer),this.pre_tokenizer=R.fromConfig(e.pre_tokenizer),this.model=k.fromConfig(e.model,t),this.post_processor=X.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*":""}(${(0,o.escapeRegExp)(e.content)})${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:r}of this.chat_template){if("string"!=typeof t||"string"!=typeof r)throw new Error('Chat template must be a list of objects with "name" and "template" properties');e[t]=r}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:r=null,cache_dir:n=null,local_files_only:o=!1,revision:s="main",legacy:a=null}={}){return new this(...await p(e,{progress_callback:t,config:r,cache_dir:n,local_files_only:o,revision:s,legacy:a}))}_call(e,{text_pair:t=null,add_special_tokens:r=!0,padding:n=!1,truncation:o=null,max_length:s=null,return_tensor:l=!0,return_token_type_ids:c=null}={}){const d=Array.isArray(e);let u;if(d){if(0===e.length)throw Error("text array must be non-empty");if(null!==t){if(!Array.isArray(t))throw Error("text_pair must also be an array");if(e.length!==t.length)throw Error("text and text_pair must have the same length");u=e.map(((e,n)=>this._encode_plus(e,{text_pair:t[n],add_special_tokens:r,return_token_type_ids:c})))}else u=e.map((e=>this._encode_plus(e,{add_special_tokens:r,return_token_type_ids:c})))}else{if(null==e)throw Error("text may not be null or undefined");if(Array.isArray(t))throw Error("When specifying `text_pair`, since `text` is a string, `text_pair` must also be a string (i.e., not an array).");u=[this._encode_plus(e,{text_pair:t,add_special_tokens:r,return_token_type_ids:c})]}if(null===s?s="max_length"===n?this.model_max_length:(0,a.max)(u.map((e=>e.input_ids.length)))[0]:o||console.warn("Truncation was not explicitly activated but `max_length` is provided a specific value, please use `truncation=true` to explicitly truncate examples to max length."),s=Math.min(s,this.model_max_length??1/0),n||o)for(let e=0;e<u.length;++e)u[e].input_ids.length!==s&&(u[e].input_ids.length>s?o&&we(u[e],s):n&&Me(u[e],s,(e=>"input_ids"===e?this.pad_token_id:0),this.padding_side));const p={};if(l){if((!n||!o)&&u.some((e=>{for(const t of Object.keys(e))if(e[t].length!==u[0][t]?.length)return!0;return!1})))throw Error("Unable to create tensor, you should probably activate truncation and/or padding with 'padding=true' and 'truncation=true' to have batched tensors with the same length.");const e=[u.length,u[0].input_ids.length];for(const t of Object.keys(u[0]))p[t]=new i.Tensor("int64",BigInt64Array.from(u.flatMap((e=>e[t])).map(BigInt)),e)}else{for(const e of Object.keys(u[0]))p[e]=u.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 g(e.toLowerCase())}(e)),null!==this.normalizer&&(e=this.normalizer(e)),0===e.length)return[];const r=null!==this.pre_tokenizer?this.pre_tokenizer(e,{section_index:t}):[e];return this.model(r)}})).flat();return t}_encode_plus(e,{text_pair:t=null,add_special_tokens:r=!0,return_token_type_ids:n=null}={}){const{tokens:o,token_type_ids:s}=this._tokenize_helper(e,{pair:t,add_special_tokens:r}),a=this.model.convert_tokens_to_ids(o),i={input_ids:a,attention_mask:new Array(a.length).fill(1)};return(n??this.return_token_type_ids)&&s&&(i.token_type_ids=s),i}_tokenize_helper(e,{pair:t=null,add_special_tokens:r=!1}={}){const n=this._encode_text(e),s=this._encode_text(t);return this.post_processor?this.post_processor(n,s,{add_special_tokens:r}):{tokens:(0,o.mergeArrays)(n??[],s??[])}}tokenize(e,{pair:t=null,add_special_tokens:r=!1}={}){return this._tokenize_helper(e,{pair:t,add_special_tokens:r}).tokens}encode(e,{text_pair:t=null,add_special_tokens:r=!0,return_token_type_ids:n=null}={}){return this._encode_plus(e,{text_pair:t,add_special_tokens:r,return_token_type_ids:n}).input_ids}batch_decode(e,t={}){return e instanceof i.Tensor&&(e=e.tolist()),e.map((e=>this.decode(e,t)))}decode(e,t={}){if(e instanceof i.Tensor&&(e=_(e)),!Array.isArray(e)||0===e.length||!(0,o.isIntegralNumber)(e[0]))throw Error("token_ids must be a non-empty array of integers.");return this.decode_single(e,t)}decode_single(e,{skip_special_tokens:t=!1,clean_up_tokenization_spaces:r=null}){let n=this.model.convert_ids_to_tokens(e);t&&(n=n.filter((e=>!this.special_tokens.includes(e))));let o=this.decoder?this.decoder(n):n.join(" ");return this.decoder&&this.decoder.end_of_word_suffix&&(o=o.replaceAll(this.decoder.end_of_word_suffix," "),t&&(o=o.trim())),(r??this.clean_up_tokenization_spaces)&&(o=f(o)),o}get_chat_template({chat_template:e=null,tools:t=null}={}){if(this.chat_template&&"object"==typeof this.chat_template){const r=this.chat_template;if(null!==e&&Object.hasOwn(r,e))e=r[e];else if(null===e)if(null!==t&&"tool_use"in r)e=r.tool_use;else{if(!("default"in r))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(r).sort()}.`);e=r.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:r=null,chat_template:n=null,add_generation_prompt:o=!1,tokenize:s=!0,padding:a=!1,truncation:i=!1,max_length:l=null,return_tensor:d=!0,return_dict:u=!1,tokenizer_kwargs:p={},...h}={}){if("string"!=typeof(n=this.get_chat_template({chat_template:n,tools:t})))throw Error("chat_template must be a string, but got "+typeof n);let m=this._compiled_template_cache.get(n);void 0===m&&(m=new c.Template(n),this._compiled_template_cache.set(n,m));const _=Object.create(null);for(const e of ge){const t=this.getToken(e);t&&(_[e]=t)}const f=m.render({messages:e,add_generation_prompt:o,tools:t,documents:r,..._,...h});if(s){const e=this._call(f,{add_special_tokens:!1,padding:a,truncation:i,max_length:l,return_tensor:d,...p});return u?e:e.input_ids}return f}}class be extends Te{return_token_type_ids=!0}class xe extends Te{return_token_type_ids=!0}class ke extends Te{return_token_type_ids=!0}class ye extends Te{return_token_type_ids=!0}class Fe extends Te{return_token_type_ids=!0}class Pe extends Te{return_token_type_ids=!0}class Ce extends Te{return_token_type_ids=!0}class ve extends Te{return_token_type_ids=!0}class Se extends Te{return_token_type_ids=!0}class Ae extends Te{}class Ee extends Te{}class Le extends Te{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 ze extends Te{return_token_type_ids=!0}class Ie extends Te{}class Be extends Te{}class Ne extends Te{}class Oe extends Te{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,r){return Ke(this,e,t,r)}}class De extends Oe{}class Ve extends Te{}class je extends Te{constructor(e,t){const r=".,!?…。，、।۔،",n=e.pre_tokenizer?.pretokenizers[0]?.pattern;n&&n.Regex===` ?[^(\\s|[${r}])]+`&&(n.Regex=` ?[^\\s${r}]+`),super(e,t)}}const Re="▁";class Ge extends Te{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:Re,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(Re+e.replaceAll(Re," "));return t.length>1&&t[0]===Re&&this.special_tokens.includes(t[1])&&(t=t.slice(1)),t}}class qe extends Te{}class $e extends Te{}class We extends Te{}class Ue extends Te{}class Xe extends Te{}class Qe extends Te{}class He extends Te{}class Ye extends Te{}class Je extends Te{}function Ke(e,t,r,n){if(!("language_codes"in e)||!Array.isArray(e.language_codes))throw new Error("Tokenizer must have `language_codes` attribute set and it should be an array of language ids.");if(!("languageRegex"in e&&e.languageRegex instanceof RegExp))throw new Error("Tokenizer must have `languageRegex` attribute set and it should be a regular expression.");if(!("lang_to_token"in e)||"function"!=typeof e.lang_to_token)throw new Error("Tokenizer must have `lang_to_token` attribute set and it should be a function.");const o=n.src_lang,s=n.tgt_lang;if(!e.language_codes.includes(s))throw new Error(`Target language code "${s}" is not valid. Must be one of: {${e.language_codes.join(", ")}}`);if(void 0!==o){if(!e.language_codes.includes(o))throw new Error(`Source language code "${o}" is not valid. Must be one of: {${e.language_codes.join(", ")}}`);for(const t of e.post_processor.config.single)if("SpecialToken"in t&&e.languageRegex.test(t.SpecialToken.id)){t.SpecialToken.id=e.lang_to_token(o);break}}return n.forced_bos_token_id=e.model.convert_tokens_to_ids([e.lang_to_token(s)])[0],e._call(t,r)}class Ze extends Te{constructor(e,t){super(e,t),this.languageRegex=/^[a-z]{3}_[A-Z][a-z]{3}$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))),this.lang_to_token=e=>e}_build_translation_inputs(e,t,r){return Ke(this,e,t,r)}}class et extends Te{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,r){return Ke(this,e,t,r)}}class tt extends Te{get timestamp_begin(){return this.model.convert_tokens_to_ids(["<|notimestamps|>"])[0]+1}_decode_asr(e,{return_timestamps:t=!1,return_language:r=!1,time_precision:n=null,force_full_sequences:o=!0}={}){if(null===n)throw Error("Must specify time_precision");let s=null;const i="word"===t;function l(){return{language:s,timestamp:[null,null],text:""}}const c=[];let u=l(),p=0;const h=this.timestamp_begin;let m=[],_=[],f=!1,g=null;const M=new Set(this.all_special_ids);for(const r of e){const e=r.tokens,o=i?r.token_timestamps:null;let w=null,b=h;if("stride"in r){const[t,o,s]=r.stride;if(p-=o,g=t-s,o&&(b=o/n+h),s)for(let t=e.length-1;t>=0;--t){const r=Number(e[t]);if(r>=h){if(null!==w&&(r-h)*n<g)break;w=r}}}let x=[],k=[];for(let r=0;r<e.length;++r){const g=Number(e[r]);if(M.has(g)){const e=this.decode([g]),r=d.WHISPER_LANGUAGE_MAPPING.get(e.slice(2,-2));if(void 0!==r){if(null!==s&&r!==s&&!t){m.push(x);const e=this.findLongestCommonSequence(m)[0],t=this.decode(e);u.text=t,c.push(u),m=[],x=[],u=l()}s=u.language=r}}else if(g>=h){const e=(g-h)*n+p,t=(0,a.round)(e,2);if(null!==w&&g>=w)f=!0;else if(f||m.length>0&&g<b)f=!1;else if(null===u.timestamp[0])u.timestamp[0]=t;else if(t===u.timestamp[0]);else{u.timestamp[1]=t,m.push(x),i&&_.push(k);const[e,r]=this.findLongestCommonSequence(m,_),n=this.decode(e);u.text=n,i&&(u.words=this.collateWordTimestamps(e,r,s)),c.push(u),m=[],x=[],_=[],k=[],u=l()}}else if(x.push(g),i){let e,t=(0,a.round)(o[r]+p,2);if(r+1<o.length){e=(0,a.round)(o[r+1]+p,2);const s=this.decode([g]);T.test(s)&&(e=(0,a.round)(Math.min(t+n,e),2))}else e=null;k.push([t,e])}}if("stride"in r){const[e,t,n]=r.stride;p+=e-n}x.length>0?(m.push(x),i&&_.push(k)):m.every((e=>0===e.length))&&(u=l(),m=[],x=[],_=[],k=[])}if(m.length>0){if(o&&t)throw new Error("Whisper did not predict an ending timestamp, which can happen if audio is cut off in the middle of a word. Also make sure WhisperTimeStampLogitsProcessor was used during generation.");const[e,r]=this.findLongestCommonSequence(m,_),n=this.decode(e);u.text=n,i&&(u.words=this.collateWordTimestamps(e,r,s)),c.push(u)}let w=Object.create(null);const b=c.map((e=>e.text)).join("");if(t||r){for(let e=0;e<c.length;++e){const n=c[e];t||delete n.timestamp,r||delete n.language}if(i){const e=[];for(const t of c)for(const r of t.words)e.push(r);w={chunks:e}}else w={chunks:c}}return[b,w]}findLongestCommonSequence(e,t=null){let r=e[0],n=r.length,o=[];const s=Array.isArray(t)&&t.length>0;let a=s?[]:null,i=s?t[0]:null;for(let l=1;l<e.length;++l){const c=e[l];let d=0,u=[n,n,0,0];const p=c.length;for(let e=1;e<n+p;++e){const o=Math.max(0,n-e),a=Math.min(n,n+p-e),h=r.slice(o,a),m=Math.max(0,e-n),_=Math.min(p,e),f=c.slice(m,_);if(h.length!==f.length)throw new Error("There is a bug within whisper `decode_asr` function, please report it. Dropping to prevent bad inference.");let g;g=s?h.filter(((e,r)=>e===f[r]&&i[o+r]<=t[l][m+r])).length:h.filter(((e,t)=>e===f[t])).length;const M=g/e+e/1e4;g>1&&M>d&&(d=M,u=[o,a,m,_])}const[h,m,_,f]=u,g=Math.floor((m+h)/2),M=Math.floor((f+_)/2);o.push(...r.slice(0,g)),r=c.slice(M),n=r.length,s&&(a.push(...i.slice(0,g)),i=t[l].slice(M))}return o.push(...r),s?(a.push(...i),[o,a]):[o,[]]}collateWordTimestamps(e,t,r){const[n,o,s]=this.combineTokensIntoWords(e,r),a=[];for(let e=0;e<n.length;++e){const r=s[e];a.push({text:n[e],timestamp:[t[r.at(0)][0],t[r.at(-1)][1]]})}return a}combineTokensIntoWords(e,t,r="\"'“¡¿([{-",n="\"'.。,，!！?？:：”)]}、"){let o,s,a;return["chinese","japanese","thai","lao","myanmar"].includes(t=t??"english")?[o,s,a]=this.splitTokensOnUnicode(e):[o,s,a]=this.splitTokensOnSpaces(e),this.mergePunctuations(o,s,a,r,n)}decode(e,t){let r;return t?.decode_with_timestamps?(e instanceof i.Tensor&&(e=_(e)),r=this.decodeWithTimestamps(e,t)):r=super.decode(e,t),r}decodeWithTimestamps(e,t){const r=t?.time_precision??.02,n=Array.from(this.all_special_ids).at(-1)+1;let o=[[]];for(let t of e)if(t=Number(t),t>=n){const e=((t-n)*r).toFixed(2);o.push(`<|${e}|>`),o.push([])}else o[o.length-1].push(t);return o=o.map((e=>"string"==typeof e?e:super.decode(e,t))),o.join("")}splitTokensOnUnicode(e){const t=this.decode(e,{decode_with_timestamps:!0}),r=[],n=[],o=[];let s=[],a=[],i=0;for(let l=0;l<e.length;++l){const c=e[l];s.push(c),a.push(l);const d=this.decode(s,{decode_with_timestamps:!0});d.includes("�")&&"�"!==t[i+d.indexOf("�")]||(r.push(d),n.push(s),o.push(a),s=[],a=[],i+=d.length)}return[r,n,o]}splitTokensOnSpaces(e){const[t,r,n]=this.splitTokensOnUnicode(e),o=[],s=[],a=[],i=new RegExp(`^[${w}]$`,"gu");for(let e=0;e<t.length;++e){const l=t[e],c=r[e],d=n[e],u=c[0]>=this.model.tokens_to_ids.get("<|endoftext|>"),p=l.startsWith(" "),h=l.trim(),m=i.test(h);if(u||p||m||0===o.length)o.push(l),s.push(c),a.push(d);else{const e=o.length-1;o[e]+=l,s[e].push(...c),a[e].push(...d)}}return[o,s,a]}mergePunctuations(e,t,r,n,s){const a=structuredClone(e),i=structuredClone(t),l=structuredClone(r);let c=a.length-2,d=a.length-1;for(;c>=0;)a[c].startsWith(" ")&&n.includes(a[c].trim())?(a[d]=a[c]+a[d],i[d]=(0,o.mergeArrays)(i[c],i[d]),l[d]=(0,o.mergeArrays)(l[c],l[d]),a[c]="",i[c]=[],l[c]=[]):d=c,--c;for(c=0,d=1;d<a.length;)!a[c].endsWith(" ")&&s.includes(a[d])?(a[c]+=a[d],i[c]=(0,o.mergeArrays)(i[c],i[d]),l[c]=(0,o.mergeArrays)(l[c],l[d]),a[d]="",i[d]=[],l[d]=[]):c=d,++d;return[a.filter((e=>e)),i.filter((e=>e.length>0)),l.filter((e=>e.length>0))]}get_decoder_prompt_ids({language:e=null,task:t=null,no_timestamps:r=!0}={}){const n=[];if(e){const t=(0,d.whisper_language_to_code)(e),r=this.model.tokens_to_ids.get(`<|${t}|>`);if(void 0===r)throw new Error(`Unable to find language "${t}" in model vocabulary. Please report this issue at ${u.GITHUB_ISSUE_URL}.`);n.push(r)}else n.push(null);if(t){if("transcribe"!==(t=t.toLowerCase())&&"translate"!==t)throw new Error(`Task "${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 ${u.GITHUB_ISSUE_URL}.`);n.push(e)}else n.push(null);if(r){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 ${u.GITHUB_ISSUE_URL}.`);n.push(e)}return n.map(((e,t)=>[t+1,e])).filter((e=>null!==e[1]))}}class rt extends Te{}class nt extends Te{}class ot extends Te{}class st extends Te{constructor(e,t){super(e,t),this.languageRegex=/^(>>\w+<<)\s*/g,this.supported_language_codes=this.model.vocab.filter((e=>this.languageRegex.test(e))),console.warn('WARNING: `MarianTokenizer` is not yet supported by Hugging Face\'s "fast" tokenizers library. Therefore, you may experience slightly inaccurate results.')}_encode_text(e){if(null===e)return null;const[t,...r]=e.trim().split(this.languageRegex);if(0===r.length)return super._encode_text(t);if(2===r.length){const[e,t]=r;return this.supported_language_codes.includes(e)||console.warn(`Unsupported language code "${e}" detected, which may lead to unexpected behavior. Should be one of: ${JSON.stringify(this.supported_language_codes)}`),(0,o.mergeArrays)([e],super._encode_text(t))}}}class at extends Te{}class it extends Te{}class lt extends Te{}class ct extends Te{}class dt extends Te{}class ut extends Te{constructor(e,t){super(e,t),this.decoder=new ce({})}}class pt extends Te{}class ht{static TOKENIZER_CLASS_MAPPING={T5Tokenizer:Ie,DistilBertTokenizer:Ae,CamembertTokenizer:Ee,DebertaTokenizer:Fe,DebertaV2Tokenizer:Pe,BertTokenizer:be,HerbertTokenizer:Ce,ConvBertTokenizer:ve,RoFormerTokenizer:Se,XLMTokenizer:Le,ElectraTokenizer:ze,MobileBertTokenizer:ke,SqueezeBertTokenizer:ye,AlbertTokenizer:xe,GPT2Tokenizer:Be,BartTokenizer:Ne,MBartTokenizer:Oe,MBart50Tokenizer:De,RobertaTokenizer:Ve,WhisperTokenizer:tt,CodeGenTokenizer:rt,CLIPTokenizer:nt,SiglipTokenizer:ot,MarianTokenizer:st,BloomTokenizer:je,NllbTokenizer:Ze,M2M100Tokenizer:et,LlamaTokenizer:Ge,CodeLlamaTokenizer:qe,XLMRobertaTokenizer:$e,MPNetTokenizer:We,FalconTokenizer:Ue,GPTNeoXTokenizer:Xe,EsmTokenizer:Qe,Wav2Vec2CTCTokenizer:at,BlenderbotTokenizer:it,BlenderbotSmallTokenizer:lt,SpeechT5Tokenizer:ct,NougatTokenizer:dt,VitsTokenizer:ut,Qwen2Tokenizer:He,GemmaTokenizer:Ye,Grok1Tokenizer:Je,CohereTokenizer:pt,PreTrainedTokenizer:Te};static async from_pretrained(e,{progress_callback:t=null,config:r=null,cache_dir:n=null,local_files_only:o=!1,revision:s="main",legacy:a=null}={}){const[i,l]=await p(e,{progress_callback:t,config:r,cache_dir:n,local_files_only:o,revision:s,legacy:a}),c=l.tokenizer_class?.replace(/Fast$/,"")??"PreTrainedTokenizer";let d=this.TOKENIZER_CLASS_MAPPING[c];return d||(console.warn(`Unknown tokenizer class "${c}", attempting to construct from base class.`),d=Te),new d(i,l)}}},"./src/utils/audio.js":
+  \***************************/(e,t,r)=>{r.r(t),r.d(t,{AlbertTokenizer:()=>xe,AutoTokenizer:()=>ht,BartTokenizer:()=>Ne,BertTokenizer:()=>be,BlenderbotSmallTokenizer:()=>lt,BlenderbotTokenizer:()=>it,BloomTokenizer:()=>je,CLIPTokenizer:()=>nt,CamembertTokenizer:()=>Ee,CodeGenTokenizer:()=>rt,CodeLlamaTokenizer:()=>qe,CohereTokenizer:()=>pt,ConvBertTokenizer:()=>ve,DebertaTokenizer:()=>Fe,DebertaV2Tokenizer:()=>Pe,DistilBertTokenizer:()=>Ae,ElectraTokenizer:()=>ze,EsmTokenizer:()=>Qe,FalconTokenizer:()=>Ue,GPT2Tokenizer:()=>Be,GPTNeoXTokenizer:()=>Xe,GemmaTokenizer:()=>Ye,Grok1Tokenizer:()=>Je,HerbertTokenizer:()=>Ce,LlamaTokenizer:()=>Ge,M2M100Tokenizer:()=>et,MBart50Tokenizer:()=>De,MBartTokenizer:()=>Oe,MPNetTokenizer:()=>We,MarianTokenizer:()=>st,MobileBertTokenizer:()=>ke,NllbTokenizer:()=>Ze,NougatTokenizer:()=>dt,PreTrainedTokenizer:()=>Te,Qwen2Tokenizer:()=>He,RoFormerTokenizer:()=>Se,RobertaTokenizer:()=>Ve,SiglipTokenizer:()=>ot,SpeechT5Tokenizer:()=>ct,SqueezeBertTokenizer:()=>ye,T5Tokenizer:()=>Ie,TokenizerModel:()=>k,VitsTokenizer:()=>ut,Wav2Vec2CTCTokenizer:()=>at,WhisperTokenizer:()=>tt,XLMRobertaTokenizer:()=>$e,XLMTokenizer:()=>Le,is_chinese_char:()=>M});var n=r(/*! ./utils/generic.js */"./src/utils/generic.js"),o=r(/*! ./utils/core.js */"./src/utils/core.js"),s=r(/*! ./utils/hub.js */"./src/utils/hub.js"),a=r(/*! ./utils/maths.js */"./src/utils/maths.js"),i=r(/*! ./utils/tensor.js */"./src/utils/tensor.js"),l=r(/*! ./utils/data-structures.js */"./src/utils/data-structures.js"),c=r(/*! @huggingface/jinja */"./node_modules/@huggingface/jinja/dist/index.js"),d=r(/*! ./models/whisper/common_whisper.js */"./src/models/whisper/common_whisper.js"),u=r(/*! ./utils/constants.js */"./src/utils/constants.js");async function p(e,t){const r=await Promise.all([(0,s.getModelJSON)(e,"tokenizer.json",!0,t),(0,s.getModelJSON)(e,"tokenizer_config.json",!0,t)]);return null!==t.legacy&&(r[1].legacy=t.legacy),r}function h(e,t=!0){if(void 0!==e.Regex){let t=e.Regex.replace(/\\([#&~])/g,"$1");for(const[e,r]of b)t=t.replaceAll(e,r);return new RegExp(t,"gu")}if(void 0!==e.String){const r=(0,o.escapeRegExp)(e.String);return new RegExp(t?r:`(${r})`,"gu")}return console.warn("Unknown pattern type:",e),null}function m(e){return new Map(Object.entries(e))}function _(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 ${t.length}.`)}}function f(e){return e.replace(/ \./g,".").replace(/ \?/g,"?").replace(/ \!/g,"!").replace(/ ,/g,",").replace(/ \' /g,"'").replace(/ n\'t/g,"n't").replace(/ \'m/g,"'m").replace(/ \'s/g,"'s").replace(/ \'ve/g,"'ve").replace(/ \'re/g,"'re")}function g(e){return e.replace(/\p{M}/gu,"")}function M(e){return e>=19968&&e<=40959||e>=13312&&e<=19903||e>=131072&&e<=173791||e>=173824&&e<=177983||e>=177984&&e<=178207||e>=178208&&e<=183983||e>=63744&&e<=64255||e>=194560&&e<=195103}const w="\\p{P}\\u0021-\\u002F\\u003A-\\u0040\\u005B-\\u0060\\u007B-\\u007E",T=new RegExp(`^[${w}]+$`,"gu"),b=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 k extends n.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 y(e);case"Unigram":return new F(e,...t);case"BPE":return new v(e);default:if(e.vocab)return new S(e,...t);throw new Error(`Unknown TokenizerModel type: ${e.type}`)}}_call(e){return e=this.encode(e),this.fuse_unk&&(e=function(e,t,r){const n=[];let o=0;for(;o<e.length;)if(n.push(e[o]),(t.get(e[o])??r)===r)for(;++o<e.length&&(t.get(e[o])??r)===r;)t.get(n.at(-1))!==r&&(n[n.length-1]+=e[o]);else++o;return n}(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 y extends k{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 r of e){const e=[...r];if(e.length>this.max_input_chars_per_word){t.push(this.unk_token);continue}let n=!1,o=0;const s=[];for(;o<e.length;){let t=e.length,r=null;for(;o<t;){let n=e.slice(o,t).join("");if(o>0&&(n=this.config.continuing_subword_prefix+n),this.tokens_to_ids.has(n)){r=n;break}--t}if(null===r){n=!0;break}s.push(r),o=t}n?t.push(this.unk_token):t.push(...s)}return t}}class F extends k{constructor(e,t){super(e);const r=e.vocab.length;this.vocab=new Array(r),this.scores=new Array(r);for(let t=0;t<r;++t){const r=e.vocab[t];this.vocab[t]=r[0],this.scores[t]=r[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,a.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 r=0;for(;r<t.length;){let n=!1;const s=[],a=t.slice(r).join(""),i=this.trie.commonPrefixSearch(a);for(const t of i){s.push(t);const a=this.tokens_to_ids.get(t),i=this.scores[a],l=(0,o.len)(t);e.insert(r,l,i,a),n||1!==l||(n=!0)}n||e.insert(r,1,this.unk_score,this.unk_token_id),r+=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 r of e){const e=this.tokenize(r);t.push(...e)}return t}}const P=(()=>{const e=[...Array.from({length:"~".charCodeAt(0)-"!".charCodeAt(0)+1},((e,t)=>t+"!".charCodeAt(0))),...Array.from({length:"¬".charCodeAt(0)-"¡".charCodeAt(0)+1},((e,t)=>t+"¡".charCodeAt(0))),...Array.from({length:"ÿ".charCodeAt(0)-"®".charCodeAt(0)+1},((e,t)=>t+"®".charCodeAt(0)))],t=e.slice();let r=0;for(let n=0;n<256;++n)e.includes(n)||(e.push(n),t.push(256+r),r+=1);const n=t.map((e=>String.fromCharCode(e)));return Object.fromEntries(e.map(((e,t)=>[e,n[t]])))})(),C=(0,o.reverseDictionary)(P);class v extends k{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 r=Array.from(e);this.end_of_word_suffix&&(r[r.length-1]+=this.end_of_word_suffix);let n=[];if(r.length>1){const e=new l.PriorityQueue(((e,t)=>e.score<t.score));let t={token:r[0],bias:0,prev:null,next:null},o=t;for(let t=1;t<r.length;++t){const n={bias:t/r.length,token:r[t],prev:o,next:null};o.next=n,this._add_node(e,o),o=n}for(;!e.isEmpty();){const r=e.pop();if(r.deleted||!r.next||r.next.deleted)continue;if(r.deleted=!0,r.next.deleted=!0,r.prev){const e={...r.prev};r.prev.deleted=!0,r.prev=e,e.prev?e.prev.next=e:t=e}const n={token:r.token+r.next.token,bias:r.bias,prev:r.prev,next:r.next.next};n.prev?(n.prev.next=n,this._add_node(e,n.prev)):t=n,n.next&&(n.next.prev=n,this._add_node(e,n))}for(let e=t;null!==e;e=e.next)n.push(e.token)}else n=r;if(this.continuing_subword_suffix)for(let e=0;e<n.length-1;++e)n[e]+=this.continuing_subword_suffix;return this.cache.set(e,n),n}_add_node(e,t){const r=this.bpe_ranks.get(t.token+this.BPE_SPLIT_TOKEN+t.next.token);void 0!==r&&(t.score=r+t.bias,e.push(t))}encode(e){const t=[];for(const r of e){if(this.ignore_merges&&this.tokens_to_ids.has(r)){t.push(r);continue}const e=this.bpe(r);for(const r of e)if(this.tokens_to_ids.has(r))t.push(r);else if(this.byte_fallback){const e=Array.from(this.text_encoder.encode(r)).map((e=>`<0x${e.toString(16).toUpperCase().padStart(2,"0")}>`));e.every((e=>this.tokens_to_ids.has(e)))?t.push(...e):t.push(this.unk_token)}else t.push(this.unk_token)}return t}}class S extends k{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 A extends n.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"BertNormalizer":return new j(e);case"Precompiled":return new pe(e);case"Sequence":return new V(e);case"Replace":return new E(e);case"NFC":return new L(e);case"NFKC":return new z(e);case"NFKD":return new I(e);case"Strip":return new B(e);case"StripAccents":return new N(e);case"Lowercase":return new O(e);case"Prepend":return new D(e);default:throw new Error(`Unknown Normalizer type: ${e.type}`)}}normalize(e){throw Error("normalize should be implemented in subclass.")}_call(e){return this.normalize(e)}}class E extends A{normalize(e){const t=h(this.config.pattern);return null===t?e:e.replaceAll(t,this.config.content)}}class L extends A{normalize(e){return e=e.normalize("NFC")}}class z extends A{normalize(e){return e=e.normalize("NFKC")}}class I extends A{normalize(e){return e=e.normalize("NFKD")}}class B extends A{normalize(e){return this.config.strip_left&&this.config.strip_right?e=e.trim():(this.config.strip_left&&(e=e.trimStart()),this.config.strip_right&&(e=e.trimEnd())),e}}class N extends A{normalize(e){return e=g(e)}}class O extends A{normalize(e){return e=e.toLowerCase()}}class D extends A{normalize(e){return e=this.config.prepend+e}}class V extends A{constructor(e){super(e),this.normalizers=e.normalizers.map((e=>A.fromConfig(e)))}normalize(e){return this.normalizers.reduce(((e,t)=>t.normalize(e)),e)}}class j extends A{_tokenize_chinese_chars(e){const t=[];for(let r=0;r<e.length;++r){const n=e[r];M(n.charCodeAt(0))?(t.push(" "),t.push(n),t.push(" ")):t.push(n)}return t.join("")}stripAccents(e){return e.normalize("NFD").replace(/\p{Mn}/gu,"")}_is_control(e){switch(e){case"\t":case"\n":case"\r":return!1;default:return/^\p{Cc}|\p{Cf}|\p{Co}|\p{Cs}$/u.test(e)}}_clean_text(e){const t=[];for(const r of e){const e=r.charCodeAt(0);0===e||65533===e||this._is_control(r)||(/^\s$/.test(r)?t.push(" "):t.push(r))}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 R extends n.Callable{static fromConfig(e){if(null===e)return null;switch(e.type){case"BertPreTokenizer":return new G(e);case"Sequence":return new he(e);case"Whitespace":return new me(e);case"WhitespaceSplit":return new _e(e);case"Metaspace":return new de(e);case"ByteLevel":return new q(e);case"Split":return new $(e);case"Punctuation":return new W(e);case"Digits":return new U(e);case"Replace":return new fe(e);default:throw new Error(`Unknown PreTokenizer type: ${e.type}`)}}pre_tokenize_text(e,t){throw Error("pre_tokenize_text should be implemented in subclass.")}pre_tokenize(e,t){return(Array.isArray(e)?e.map((e=>this.pre_tokenize_text(e,t))):this.pre_tokenize_text(e,t)).flat()}_call(e,t){return this.pre_tokenize(e,t)}}class G extends R{constructor(e){super(),this.pattern=new RegExp(`[^\\s${w}]+|[${w}]`,"gu")}pre_tokenize_text(e,t){return e.trim().match(this.pattern)||[]}}class q extends R{constructor(e){super(),this.config=e,this.add_prefix_space=this.config.add_prefix_space,this.trim_offsets=this.config.trim_offsets,this.use_regex=this.config.use_regex??!0,this.pattern=/'s|'t|'re|'ve|'m|'ll|'d| ?\p{L}+| ?\p{N}+| ?[^\s\p{L}\p{N}]+|\s+(?!\S)|\s+/gu,this.byte_encoder=P,this.text_encoder=new TextEncoder}pre_tokenize_text(e,t){this.add_prefix_space&&!e.startsWith(" ")&&(e=" "+e);return(this.use_regex?e.match(this.pattern)||[]:[e]).map((e=>Array.from(this.text_encoder.encode(e),(e=>this.byte_encoder[e])).join("")))}}class $ extends R{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 r=[];let n=0;for(const o of e.matchAll(t)){const t=o[0];n<o.index&&r.push(e.slice(n,o.index)),t.length>0&&r.push(t),n=o.index+t.length}return n<e.length&&r.push(e.slice(n)),r}(e,this.pattern)}}class W extends R{constructor(e){super(),this.config=e,this.pattern=new RegExp(`[^${w}]+|[${w}]+`,"gu")}pre_tokenize_text(e,t){return e.match(this.pattern)||[]}}class U extends R{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 X extends n.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"TemplateProcessing":return new Y(e);case"ByteLevel":return new J(e);case"RobertaProcessing":return new H(e);case"BertProcessing":return new Q(e);case"Sequence":return new K(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 Q extends X{constructor(e){super(e),this.cls=e.cls[0],this.sep=e.sep[0]}post_process(e,t=null,{add_special_tokens:r=!0}={}){r&&(e=(0,o.mergeArrays)([this.cls],e,[this.sep]));let n=new Array(e.length).fill(0);if(null!==t){const s=r&&this instanceof H?[this.sep]:[],a=r?[this.sep]:[];e=(0,o.mergeArrays)(e,s,t,a),n=(0,o.mergeArrays)(n,new Array(t.length+s.length+a.length).fill(1))}return{tokens:e,token_type_ids:n}}}class H extends Q{}class Y extends X{constructor(e){super(e),this.single=e.single,this.pair=e.pair}post_process(e,t=null,{add_special_tokens:r=!0}={}){const n=null===t?this.single:this.pair;let s=[],a=[];for(const i of n)"SpecialToken"in i?r&&(s.push(i.SpecialToken.id),a.push(i.SpecialToken.type_id)):"Sequence"in i&&("A"===i.Sequence.id?(s=(0,o.mergeArrays)(s,e),a=(0,o.mergeArrays)(a,new Array(e.length).fill(i.Sequence.type_id))):"B"===i.Sequence.id&&(s=(0,o.mergeArrays)(s,t),a=(0,o.mergeArrays)(a,new Array(t.length).fill(i.Sequence.type_id))));return{tokens:s,token_type_ids:a}}}class J extends X{post_process(e,t=null){return t&&(e=(0,o.mergeArrays)(e,t)),{tokens:e}}}class K extends X{constructor(e){super(e),this.processors=e.processors.map((e=>X.fromConfig(e)))}post_process(e,t=null,r={}){let n;for(const o of this.processors)if(o instanceof J){if(e=o.post_process(e).tokens,t){t=o.post_process(t).tokens}}else{const s=o.post_process(e,t,r);e=s.tokens,n=s.token_type_ids}return{tokens:e,token_type_ids:n}}}class Z extends n.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 oe(e);case"Metaspace":return new ue(e);case"ByteLevel":return new se(e);case"Replace":return new ee(e);case"ByteFallback":return new te(e);case"Fuse":return new re(e);case"Strip":return new ne(e);case"Sequence":return new ie(e);case"CTC":return new ae(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 r=[];for(const n of e){let e=null;if(6===n.length&&n.startsWith("<0x")&&n.endsWith(">")){const t=parseInt(n.slice(3,5),16);isNaN(t)||(e=t)}if(null!==e)r.push(e);else{if(r.length>0){const e=this.text_decoder.decode(Uint8Array.from(r));t.push(e),r=[]}t.push(n)}}if(r.length>0){const e=this.text_decoder.decode(Uint8Array.from(r));t.push(e),r=[]}return t}}class re extends Z{decode_chain(e){return[e.join("")]}}class ne 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 r=0;r<this.start&&e[r]===this.content;++r)t=r+1;let r=e.length;for(let t=0;t<this.stop;++t){const n=e.length-t-1;if(e[n]!==this.content)break;r=n}return e.slice(t,r)}))}}class oe 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=f(e)),e)))}}class se extends Z{constructor(e){super(e),this.byte_decoder=C,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(""),r=new Uint8Array([...t].map((e=>this.byte_decoder[e])));return this.text_decoder.decode(r)}decode_chain(e){const t=[];let r=[];for(const n of e)void 0!==this.added_tokens.find((e=>e.content===n))?(r.length>0&&(t.push(this.convert_tokens_to_string(r)),r=[]),t.push(n)):r.push(n);return r.length>0&&t.push(this.convert_tokens_to_string(r)),t}}class ae 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 r=1;r<e.length;++r)e[r]!==t.at(-1)&&t.push(e[r]);let r=t.filter((e=>e!==this.pad_token)).join("");return this.cleanup&&(r=f(r).replaceAll(this.word_delimiter_token," ").trim()),r}decode_chain(e){return[this.convert_tokens_to_string(e)]}}class ie 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,r)=>t.replaceAll(this.suffix,r===e.length-1?"":" ")))}}class ce extends Z{decode_chain(e){let t="";for(let r=1;r<e.length;r+=2)t+=e[r];return[t]}}class de extends R{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 r=e.replaceAll(" ",this.strRep);return this.addPrefixSpace&&!r.startsWith(this.replacement)&&("always"===this.prepend_scheme||"first"===this.prepend_scheme&&0===t)&&(r=this.strRep+r),[r]}}class ue extends Z{constructor(e){super(e),this.addPrefixSpace=e.add_prefix_space,this.replacement=e.replacement}decode_chain(e){const t=[];for(let r=0;r<e.length;++r){let n=e[r].replaceAll(this.replacement," ");this.addPrefixSpace&&0==r&&n.startsWith(" ")&&(n=n.substring(1)),t.push(n)}return t}}class pe extends A{constructor(e){super(e),this.charsmap=e.precompiled_charsmap}normalize(e){if((e=(e=e.replace(/[\u0001-\u0008\u000B\u000E-\u001F\u007F\u008F\u009F]/gm,"")).replace(/[\u0009\u000A\u000C\u000D\u00A0\u1680\u2000-\u200F\u2028\u2029\u202F\u205F\u2581\u3000\uFEFF\uFFFD]/gm," ")).includes("～")){const t=e.split("～");e=t.map((e=>e.normalize("NFKC"))).join("～")}else e=e.normalize("NFKC");return e}}class he extends R{constructor(e){super(),this.tokenizers=e.pretokenizers.map((e=>R.fromConfig(e)))}pre_tokenize_text(e,t){return this.tokenizers.reduce(((e,r)=>r.pre_tokenize(e,t)),[e])}}class me extends R{constructor(e){super()}pre_tokenize_text(e,t){return e.match(/\w+|[^\w\s]+/g)||[]}}class _e extends R{constructor(e){super()}pre_tokenize_text(e,t){return function(e){return e.match(/\S+/g)||[]}(e)}}class fe extends R{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 ge=["bos_token","eos_token","unk_token","sep_token","pad_token","cls_token","mask_token"];function Me(e,t,r,n){for(const s of Object.keys(e)){const a=t-e[s].length,i=r(s),l=new Array(a).fill(i);e[s]="right"===n?(0,o.mergeArrays)(e[s],l):(0,o.mergeArrays)(l,e[s])}}function we(e,t){for(const r of Object.keys(e))e[r].length=t}class Te extends n.Callable{return_token_type_ids=!1;padding_side="right";constructor(e,t){super(),this._tokenizer_config=t,this.normalizer=A.fromConfig(e.normalizer),this.pre_tokenizer=R.fromConfig(e.pre_tokenizer),this.model=k.fromConfig(e.model,t),this.post_processor=X.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*":""}(${(0,o.escapeRegExp)(e.content)})${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:r}of this.chat_template){if("string"!=typeof t||"string"!=typeof r)throw new Error('Chat template must be a list of objects with "name" and "template" properties');e[t]=r}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:r=null,cache_dir:n=null,local_files_only:o=!1,revision:s="main",legacy:a=null}={}){return new this(...await p(e,{progress_callback:t,config:r,cache_dir:n,local_files_only:o,revision:s,legacy:a}))}_call(e,{text_pair:t=null,add_special_tokens:r=!0,padding:n=!1,truncation:o=null,max_length:s=null,return_tensor:l=!0,return_token_type_ids:c=null}={}){const d=Array.isArray(e);let u;if(d){if(0===e.length)throw Error("text array must be non-empty");if(null!==t){if(!Array.isArray(t))throw Error("text_pair must also be an array");if(e.length!==t.length)throw Error("text and text_pair must have the same length");u=e.map(((e,n)=>this._encode_plus(e,{text_pair:t[n],add_special_tokens:r,return_token_type_ids:c})))}else u=e.map((e=>this._encode_plus(e,{add_special_tokens:r,return_token_type_ids:c})))}else{if(null==e)throw Error("text may not be null or undefined");if(Array.isArray(t))throw Error("When specifying `text_pair`, since `text` is a string, `text_pair` must also be a string (i.e., not an array).");u=[this._encode_plus(e,{text_pair:t,add_special_tokens:r,return_token_type_ids:c})]}if(null===s?s="max_length"===n?this.model_max_length:(0,a.max)(u.map((e=>e.input_ids.length)))[0]:o||console.warn("Truncation was not explicitly activated but `max_length` is provided a specific value, please use `truncation=true` to explicitly truncate examples to max length."),s=Math.min(s,this.model_max_length??1/0),n||o)for(let e=0;e<u.length;++e)u[e].input_ids.length!==s&&(u[e].input_ids.length>s?o&&we(u[e],s):n&&Me(u[e],s,(e=>"input_ids"===e?this.pad_token_id:0),this.padding_side));const p={};if(l){if((!n||!o)&&u.some((e=>{for(const t of Object.keys(e))if(e[t].length!==u[0][t]?.length)return!0;return!1})))throw Error("Unable to create tensor, you should probably activate truncation and/or padding with 'padding=true' and 'truncation=true' to have batched tensors with the same length.");const e=[u.length,u[0].input_ids.length];for(const t of Object.keys(u[0]))p[t]=new i.Tensor("int64",BigInt64Array.from(u.flatMap((e=>e[t])).map(BigInt)),e)}else{for(const e of Object.keys(u[0]))p[e]=u.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 g(e.toLowerCase())}(e)),null!==this.normalizer&&(e=this.normalizer(e)),0===e.length)return[];const r=null!==this.pre_tokenizer?this.pre_tokenizer(e,{section_index:t}):[e];return this.model(r)}})).flat();return t}_encode_plus(e,{text_pair:t=null,add_special_tokens:r=!0,return_token_type_ids:n=null}={}){const{tokens:o,token_type_ids:s}=this._tokenize_helper(e,{pair:t,add_special_tokens:r}),a=this.model.convert_tokens_to_ids(o),i={input_ids:a,attention_mask:new Array(a.length).fill(1)};return(n??this.return_token_type_ids)&&s&&(i.token_type_ids=s),i}_tokenize_helper(e,{pair:t=null,add_special_tokens:r=!1}={}){const n=this._encode_text(e),s=this._encode_text(t);return this.post_processor?this.post_processor(n,s,{add_special_tokens:r}):{tokens:(0,o.mergeArrays)(n??[],s??[])}}tokenize(e,{pair:t=null,add_special_tokens:r=!1}={}){return this._tokenize_helper(e,{pair:t,add_special_tokens:r}).tokens}encode(e,{text_pair:t=null,add_special_tokens:r=!0,return_token_type_ids:n=null}={}){return this._encode_plus(e,{text_pair:t,add_special_tokens:r,return_token_type_ids:n}).input_ids}batch_decode(e,t={}){return e instanceof i.Tensor&&(e=e.tolist()),e.map((e=>this.decode(e,t)))}decode(e,t={}){if(e instanceof i.Tensor&&(e=_(e)),!Array.isArray(e)||0===e.length||!(0,o.isIntegralNumber)(e[0]))throw Error("token_ids must be a non-empty array of integers.");return this.decode_single(e,t)}decode_single(e,{skip_special_tokens:t=!1,clean_up_tokenization_spaces:r=null}){let n=this.model.convert_ids_to_tokens(e);t&&(n=n.filter((e=>!this.special_tokens.includes(e))));let o=this.decoder?this.decoder(n):n.join(" ");return this.decoder&&this.decoder.end_of_word_suffix&&(o=o.replaceAll(this.decoder.end_of_word_suffix," "),t&&(o=o.trim())),(r??this.clean_up_tokenization_spaces)&&(o=f(o)),o}get_chat_template({chat_template:e=null,tools:t=null}={}){if(this.chat_template&&"object"==typeof this.chat_template){const r=this.chat_template;if(null!==e&&Object.hasOwn(r,e))e=r[e];else if(null===e)if(null!==t&&"tool_use"in r)e=r.tool_use;else{if(!("default"in r))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(r).sort()}.`);e=r.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:r=null,chat_template:n=null,add_generation_prompt:o=!1,tokenize:s=!0,padding:a=!1,truncation:i=!1,max_length:l=null,return_tensor:d=!0,return_dict:u=!1,tokenizer_kwargs:p={},...h}={}){if("string"!=typeof(n=this.get_chat_template({chat_template:n,tools:t})))throw Error("chat_template must be a string, but got "+typeof n);let m=this._compiled_template_cache.get(n);void 0===m&&(m=new c.Template(n),this._compiled_template_cache.set(n,m));const _=Object.create(null);for(const e of ge){const t=this.getToken(e);t&&(_[e]=t)}const f=m.render({messages:e,add_generation_prompt:o,tools:t,documents:r,..._,...h});if(s){const e=this._call(f,{add_special_tokens:!1,padding:a,truncation:i,max_length:l,return_tensor:d,...p});return u?e:e.input_ids}return f}}class be extends Te{return_token_type_ids=!0}class xe extends Te{return_token_type_ids=!0}class ke extends Te{return_token_type_ids=!0}class ye extends Te{return_token_type_ids=!0}class Fe extends Te{return_token_type_ids=!0}class Pe extends Te{return_token_type_ids=!0}class Ce extends Te{return_token_type_ids=!0}class ve extends Te{return_token_type_ids=!0}class Se extends Te{return_token_type_ids=!0}class Ae extends Te{}class Ee extends Te{}class Le extends Te{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 ze extends Te{return_token_type_ids=!0}class Ie extends Te{}class Be extends Te{}class Ne extends Te{}class Oe extends Te{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,r){return Ke(this,e,t,r)}}class De extends Oe{}class Ve extends Te{}class je extends Te{constructor(e,t){const r=".,!?…。，、।۔،",n=e.pre_tokenizer?.pretokenizers[0]?.pattern;n&&n.Regex===` ?[^(\\s|[${r}])]+`&&(n.Regex=` ?[^\\s${r}]+`),super(e,t)}}const Re="▁";class Ge extends Te{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:Re,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(Re+e.replaceAll(Re," "));return t.length>1&&t[0]===Re&&this.special_tokens.includes(t[1])&&(t=t.slice(1)),t}}class qe extends Te{}class $e extends Te{}class We extends Te{}class Ue extends Te{}class Xe extends Te{}class Qe extends Te{}class He extends Te{}class Ye extends Te{}class Je extends Te{}function Ke(e,t,r,n){if(!("language_codes"in e)||!Array.isArray(e.language_codes))throw new Error("Tokenizer must have `language_codes` attribute set and it should be an array of language ids.");if(!("languageRegex"in e&&e.languageRegex instanceof RegExp))throw new Error("Tokenizer must have `languageRegex` attribute set and it should be a regular expression.");if(!("lang_to_token"in e)||"function"!=typeof e.lang_to_token)throw new Error("Tokenizer must have `lang_to_token` attribute set and it should be a function.");const o=n.src_lang,s=n.tgt_lang;if(!e.language_codes.includes(s))throw new Error(`Target language code "${s}" is not valid. Must be one of: {${e.language_codes.join(", ")}}`);if(void 0!==o){if(!e.language_codes.includes(o))throw new Error(`Source language code "${o}" is not valid. Must be one of: {${e.language_codes.join(", ")}}`);for(const t of e.post_processor.config.single)if("SpecialToken"in t&&e.languageRegex.test(t.SpecialToken.id)){t.SpecialToken.id=e.lang_to_token(o);break}}return n.forced_bos_token_id=e.model.convert_tokens_to_ids([e.lang_to_token(s)])[0],e._call(t,r)}class Ze extends Te{constructor(e,t){super(e,t),this.languageRegex=/^[a-z]{3}_[A-Z][a-z]{3}$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))),this.lang_to_token=e=>e}_build_translation_inputs(e,t,r){return Ke(this,e,t,r)}}class et extends Te{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,r){return Ke(this,e,t,r)}}class tt extends Te{get timestamp_begin(){return this.model.convert_tokens_to_ids(["<|notimestamps|>"])[0]+1}_decode_asr(e,{return_timestamps:t=!1,return_language:r=!1,time_precision:n=null,force_full_sequences:o=!0}={}){if(null===n)throw Error("Must specify time_precision");let s=null;const i="word"===t;function l(){return{language:s,timestamp:[null,null],text:""}}const c=[];let u=l(),p=0;const h=this.timestamp_begin;let m=[],_=[],f=!1,g=null;const M=new Set(this.all_special_ids);for(const r of e){const e=r.tokens,o=i?r.token_timestamps:null;let w=null,b=h;if("stride"in r){const[t,o,s]=r.stride;if(p-=o,g=t-s,o&&(b=o/n+h),s)for(let t=e.length-1;t>=0;--t){const r=Number(e[t]);if(r>=h){if(null!==w&&(r-h)*n<g)break;w=r}}}let x=[],k=[];for(let r=0;r<e.length;++r){const g=Number(e[r]);if(M.has(g)){const e=this.decode([g]),r=d.WHISPER_LANGUAGE_MAPPING.get(e.slice(2,-2));if(void 0!==r){if(null!==s&&r!==s&&!t){m.push(x);const e=this.findLongestCommonSequence(m)[0],t=this.decode(e);u.text=t,c.push(u),m=[],x=[],u=l()}s=u.language=r}}else if(g>=h){const e=(g-h)*n+p,t=(0,a.round)(e,2);if(null!==w&&g>=w)f=!0;else if(f||m.length>0&&g<b)f=!1;else if(null===u.timestamp[0])u.timestamp[0]=t;else if(t===u.timestamp[0]);else{u.timestamp[1]=t,m.push(x),i&&_.push(k);const[e,r]=this.findLongestCommonSequence(m,_),n=this.decode(e);u.text=n,i&&(u.words=this.collateWordTimestamps(e,r,s)),c.push(u),m=[],x=[],_=[],k=[],u=l()}}else if(x.push(g),i){let e,t=(0,a.round)(o[r]+p,2);if(r+1<o.length){e=(0,a.round)(o[r+1]+p,2);const s=this.decode([g]);T.test(s)&&(e=(0,a.round)(Math.min(t+n,e),2))}else e=null;k.push([t,e])}}if("stride"in r){const[e,t,n]=r.stride;p+=e-n}x.length>0?(m.push(x),i&&_.push(k)):m.every((e=>0===e.length))&&(u=l(),m=[],x=[],_=[],k=[])}if(m.length>0){if(o&&t)throw new Error("Whisper did not predict an ending timestamp, which can happen if audio is cut off in the middle of a word. Also make sure WhisperTimeStampLogitsProcessor was used during generation.");const[e,r]=this.findLongestCommonSequence(m,_),n=this.decode(e);u.text=n,i&&(u.words=this.collateWordTimestamps(e,r,s)),c.push(u)}let w=Object.create(null);const b=c.map((e=>e.text)).join("");if(t||r){for(let e=0;e<c.length;++e){const n=c[e];t||delete n.timestamp,r||delete n.language}if(i){const e=[];for(const t of c)for(const r of t.words)e.push(r);w={chunks:e}}else w={chunks:c}}return[b,w]}findLongestCommonSequence(e,t=null){let r=e[0],n=r.length,o=[];const s=Array.isArray(t)&&t.length>0;let a=s?[]:null,i=s?t[0]:null;for(let l=1;l<e.length;++l){const c=e[l];let d=0,u=[n,n,0,0];const p=c.length;for(let e=1;e<n+p;++e){const o=Math.max(0,n-e),a=Math.min(n,n+p-e),h=r.slice(o,a),m=Math.max(0,e-n),_=Math.min(p,e),f=c.slice(m,_);if(h.length!==f.length)throw new Error("There is a bug within whisper `decode_asr` function, please report it. Dropping to prevent bad inference.");let g;g=s?h.filter(((e,r)=>e===f[r]&&i[o+r]<=t[l][m+r])).length:h.filter(((e,t)=>e===f[t])).length;const M=g/e+e/1e4;g>1&&M>d&&(d=M,u=[o,a,m,_])}const[h,m,_,f]=u,g=Math.floor((m+h)/2),M=Math.floor((f+_)/2);o.push(...r.slice(0,g)),r=c.slice(M),n=r.length,s&&(a.push(...i.slice(0,g)),i=t[l].slice(M))}return o.push(...r),s?(a.push(...i),[o,a]):[o,[]]}collateWordTimestamps(e,t,r){const[n,o,s]=this.combineTokensIntoWords(e,r),a=[];for(let e=0;e<n.length;++e){const r=s[e];a.push({text:n[e],timestamp:[t[r.at(0)][0],t[r.at(-1)][1]]})}return a}combineTokensIntoWords(e,t,r="\"'“¡¿([{-",n="\"'.。,，!！?？:：”)]}、"){let o,s,a;return["chinese","japanese","thai","lao","myanmar"].includes(t=t??"english")?[o,s,a]=this.splitTokensOnUnicode(e):[o,s,a]=this.splitTokensOnSpaces(e),this.mergePunctuations(o,s,a,r,n)}decode(e,t){let r;return t?.decode_with_timestamps?(e instanceof i.Tensor&&(e=_(e)),r=this.decodeWithTimestamps(e,t)):r=super.decode(e,t),r}decodeWithTimestamps(e,t){const r=t?.time_precision??.02,n=Array.from(this.all_special_ids).at(-1)+1;let o=[[]];for(let t of e)if(t=Number(t),t>=n){const e=((t-n)*r).toFixed(2);o.push(`<|${e}|>`),o.push([])}else o[o.length-1].push(t);return o=o.map((e=>"string"==typeof e?e:super.decode(e,t))),o.join("")}splitTokensOnUnicode(e){const t=this.decode(e,{decode_with_timestamps:!0}),r=[],n=[],o=[];let s=[],a=[],i=0;for(let l=0;l<e.length;++l){const c=e[l];s.push(c),a.push(l);const d=this.decode(s,{decode_with_timestamps:!0});d.includes("�")&&"�"!==t[i+d.indexOf("�")]||(r.push(d),n.push(s),o.push(a),s=[],a=[],i+=d.length)}return[r,n,o]}splitTokensOnSpaces(e){const[t,r,n]=this.splitTokensOnUnicode(e),o=[],s=[],a=[],i=new RegExp(`^[${w}]$`,"gu");for(let e=0;e<t.length;++e){const l=t[e],c=r[e],d=n[e],u=c[0]>=this.model.tokens_to_ids.get("<|endoftext|>"),p=l.startsWith(" "),h=l.trim(),m=i.test(h);if(u||p||m||0===o.length)o.push(l),s.push(c),a.push(d);else{const e=o.length-1;o[e]+=l,s[e].push(...c),a[e].push(...d)}}return[o,s,a]}mergePunctuations(e,t,r,n,s){const a=structuredClone(e),i=structuredClone(t),l=structuredClone(r);let c=a.length-2,d=a.length-1;for(;c>=0;)a[c].startsWith(" ")&&n.includes(a[c].trim())?(a[d]=a[c]+a[d],i[d]=(0,o.mergeArrays)(i[c],i[d]),l[d]=(0,o.mergeArrays)(l[c],l[d]),a[c]="",i[c]=[],l[c]=[]):d=c,--c;for(c=0,d=1;d<a.length;)!a[c].endsWith(" ")&&s.includes(a[d])?(a[c]+=a[d],i[c]=(0,o.mergeArrays)(i[c],i[d]),l[c]=(0,o.mergeArrays)(l[c],l[d]),a[d]="",i[d]=[],l[d]=[]):c=d,++d;return[a.filter((e=>e)),i.filter((e=>e.length>0)),l.filter((e=>e.length>0))]}get_decoder_prompt_ids({language:e=null,task:t=null,no_timestamps:r=!0}={}){const n=[];if(e){const t=(0,d.whisper_language_to_code)(e),r=this.model.tokens_to_ids.get(`<|${t}|>`);if(void 0===r)throw new Error(`Unable to find language "${t}" in model vocabulary. Please report this issue at ${u.GITHUB_ISSUE_URL}.`);n.push(r)}else n.push(null);if(t){if("transcribe"!==(t=t.toLowerCase())&&"translate"!==t)throw new Error(`Task "${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 ${u.GITHUB_ISSUE_URL}.`);n.push(e)}else n.push(null);if(r){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 ${u.GITHUB_ISSUE_URL}.`);n.push(e)}return n.map(((e,t)=>[t+1,e])).filter((e=>null!==e[1]))}}class rt extends Te{}class nt extends Te{}class ot extends Te{}class st extends Te{constructor(e,t){super(e,t),this.languageRegex=/^(>>\w+<<)\s*/g,this.supported_language_codes=this.model.vocab.filter((e=>this.languageRegex.test(e))),console.warn('WARNING: `MarianTokenizer` is not yet supported by Hugging Face\'s "fast" tokenizers library. Therefore, you may experience slightly inaccurate results.')}_encode_text(e){if(null===e)return null;const[t,...r]=e.trim().split(this.languageRegex);if(0===r.length)return super._encode_text(t);if(2===r.length){const[e,t]=r;return this.supported_language_codes.includes(e)||console.warn(`Unsupported language code "${e}" detected, which may lead to unexpected behavior. Should be one of: ${JSON.stringify(this.supported_language_codes)}`),(0,o.mergeArrays)([e],super._encode_text(t))}}}class at extends Te{}class it extends Te{}class lt extends Te{}class ct extends Te{}class dt extends Te{}class ut extends Te{constructor(e,t){super(e,t),this.decoder=new ce({})}}class pt extends Te{}class ht{static TOKENIZER_CLASS_MAPPING={T5Tokenizer:Ie,DistilBertTokenizer:Ae,CamembertTokenizer:Ee,DebertaTokenizer:Fe,DebertaV2Tokenizer:Pe,BertTokenizer:be,HerbertTokenizer:Ce,ConvBertTokenizer:ve,RoFormerTokenizer:Se,XLMTokenizer:Le,ElectraTokenizer:ze,MobileBertTokenizer:ke,SqueezeBertTokenizer:ye,AlbertTokenizer:xe,GPT2Tokenizer:Be,BartTokenizer:Ne,MBartTokenizer:Oe,MBart50Tokenizer:De,RobertaTokenizer:Ve,WhisperTokenizer:tt,CodeGenTokenizer:rt,CLIPTokenizer:nt,SiglipTokenizer:ot,MarianTokenizer:st,BloomTokenizer:je,NllbTokenizer:Ze,M2M100Tokenizer:et,LlamaTokenizer:Ge,CodeLlamaTokenizer:qe,XLMRobertaTokenizer:$e,MPNetTokenizer:We,FalconTokenizer:Ue,GPTNeoXTokenizer:Xe,EsmTokenizer:Qe,Wav2Vec2CTCTokenizer:at,BlenderbotTokenizer:it,BlenderbotSmallTokenizer:lt,SpeechT5Tokenizer:ct,NougatTokenizer:dt,VitsTokenizer:ut,Qwen2Tokenizer:He,GemmaTokenizer:Ye,Grok1Tokenizer:Je,CohereTokenizer:pt,PreTrainedTokenizer:Te};static async from_pretrained(e,{progress_callback:t=null,config:r=null,cache_dir:n=null,local_files_only:o=!1,revision:s="main",legacy:a=null}={}){const[i,l]=await p(e,{progress_callback:t,config:r,cache_dir:n,local_files_only:o,revision:s,legacy:a}),c=l.tokenizer_class?.replace(/Fast$/,"")??"PreTrainedTokenizer";let d=this.TOKENIZER_CLASS_MAPPING[c];return d||(console.warn(`Unknown tokenizer class "${c}", attempting to construct from base class.`),d=Te),new d(i,l)}}},"./src/utils/audio.js":
 /*!****************************!*\
   !*** ./src/utils/audio.js ***!
   \****************************/(e,t,r)=>{r.r(t),r.d(t,{hamming:()=>d,hanning:()=>c,mel_filter_bank:()=>_,read_audio:()=>i,spectrogram:()=>g,window_function:()=>M});var n=r(/*! ./hub.js */"./src/utils/hub.js"),o=r(/*! ./maths.js */"./src/utils/maths.js"),s=r(/*! ./core.js */"./src/utils/core.js"),a=r(/*! ./tensor.js */"./src/utils/tensor.js");async function i(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 r=await(await(0,n.getFile)(e)).arrayBuffer(),o=new AudioContext({sampleRate:t});void 0===t&&console.warn(`No sampling rate provided, using default of ${o.sampleRate}Hz.`);const s=await o.decodeAudioData(r);let a;if(2===s.numberOfChannels){const e=Math.sqrt(2),t=s.getChannelData(0),r=s.getChannelData(1);a=new Float32Array(t.length);for(let n=0;n<s.length;++n)a[n]=e*(t[n]+r[n])/2}else a=s.getChannelData(0);return a}function l(e,t){if(e<1)return new Float64Array;if(1===e)return new Float64Array([1]);const r=1-t,n=2*Math.PI/(e-1),o=new Float64Array(e);for(let s=0;s<e;++s)o[s]=t-r*Math.cos(s*n);return o}function c(e){return l(e,.5)}function d(e){return l(e,.54)}const u={htk:e=>2595*Math.log10(1+e/700),kaldi:e=>1127*Math.log(1+e/700),slaney:(e,t=1e3,r=15,n=27/Math.log(6.4))=>e>=t?r+Math.log(e/t)*n:3*e/200};function p(e,t="htk"){const r=u[t];if(!r)throw new Error('mel_scale should be one of "htk", "slaney" or "kaldi".');return"number"==typeof e?r(e):e.map((e=>r(e)))}const h={htk:e=>700*(10**(e/2595)-1),kaldi:e=>700*(Math.exp(e/1127)-1),slaney:(e,t=1e3,r=15,n=Math.log(6.4)/27)=>e>=r?t*Math.exp(n*(e-r)):200*e/3};function m(e,t,r){const n=(t-e)/(r-1);return Float64Array.from({length:r},((t,r)=>e+n*r))}function _(e,t,r,n,o,s=null,a="htk",i=!1){if(null!==s&&"slaney"!==s)throw new Error('norm must be one of null or "slaney"');const l=m(p(r,a),p(n,a),t+2);let c,d=function(e,t="htk"){const r=h[t];if(!r)throw new Error('mel_scale should be one of "htk", "slaney" or "kaldi".');return"number"==typeof e?r(e):e.map((e=>r(e)))}(l,a);if(i){const t=o/(2*e);c=p(Float64Array.from({length:e},((e,r)=>r*t)),a),d=l}else c=m(0,Math.floor(o/2),e);const u=function(e,t){const r=Float64Array.from({length:t.length-1},((e,r)=>t[r+1]-t[r])),n=Array.from({length:e.length},(()=>new Array(t.length)));for(let r=0;r<e.length;++r){const o=n[r];for(let n=0;n<t.length;++n)o[n]=t[n]-e[r]}const o=t.length-2,s=Array.from({length:o},(()=>new Array(e.length)));for(let t=0;t<e.length;++t){const e=n[t];for(let n=0;n<o;++n){const o=-e[n]/r[n],a=e[n+2]/r[n+1];s[n][t]=Math.max(0,Math.min(o,a))}}return s}(c,d);if(null!==s&&"slaney"===s)for(let r=0;r<t;++r){const t=u[r],n=2/(d[r+2]-d[r]);for(let r=0;r<e;++r)t[r]*=n}return u}function f(e,t,r,n,s){if(r<=0)throw new Error("reference must be greater than zero");if(n<=0)throw new Error("min_value must be greater than zero");r=Math.max(n,r);const a=Math.log10(r);for(let r=0;r<e.length;++r)e[r]=t*Math.log10(Math.max(n,e[r])-a);if(null!==s){if(s<=0)throw new Error("db_range must be greater than zero");const t=(0,o.max)(e)[0]-s;for(let r=0;r<e.length;++r)e[r]=Math.max(e[r],t)}return e}async function g(e,t,r,n,{fft_length:i=null,power:l=1,center:c=!0,pad_mode:d="reflect",onesided:u=!0,preemphasis:p=null,mel_filters:h=null,mel_floor:m=1e-10,log_mel:_=null,reference:g=1,min_value:M=1e-10,db_range:w=null,remove_dc_offset:T=null,min_num_frames:b=null,max_num_frames:x=null,do_pad:k=!0,transpose:y=!1}={}){const F=t.length;if(null===i&&(i=r),r>i)throw Error(`frame_length (${r}) may not be larger than fft_length (${i})`);if(F!==r)throw new Error(`Length of the window (${F}) must equal frame_length (${r})`);if(n<=0)throw new Error("hop_length must be greater than zero");if(null===l&&null!==h)throw new Error("You have provided `mel_filters` but `power` is `None`. Mel spectrogram computation is not yet supported for complex-valued spectrogram. Specify `power` to fix this issue.");if(c){if("reflect"!==d)throw new Error(`pad_mode="${d}" not implemented yet.`);const t=Math.floor((i-1)/2)+1;e=function(e,t,r){const n=new e.constructor(e.length+t+r),o=e.length-1;for(let r=0;r<e.length;++r)n[t+r]=e[r];for(let r=1;r<=t;++r)n[t-r]=e[(0,s.calculateReflectOffset)(r,o)];for(let a=1;a<=r;++a)n[o+t+a]=e[(0,s.calculateReflectOffset)(o-a,o)];return n}(e,t,t)}let P=Math.floor(1+Math.floor((e.length-r)/n));null!==b&&P<b&&(P=b);const C=u?Math.floor(i/2)+1:i;let v=P,S=P;null!==x&&(x>P?k&&(S=x):S=v=x);const A=new o.FFT(i),E=new Float64Array(i),L=new Float64Array(A.outputBufferSize),z=new Float32Array(C*S);for(let o=0;o<v;++o){const s=o*n,a=Math.min(e.length-s,r);a!==r&&E.fill(0,0,r);for(let t=0;t<a;++t)E[t]=e[s+t];if(T){let e=0;for(let t=0;t<a;++t)e+=E[t];const t=e/a;for(let e=0;e<a;++e)E[e]-=t}if(null!==p){for(let e=a-1;e>=1;--e)E[e]-=p*E[e-1];E[0]*=1-p}for(let e=0;e<t.length;++e)E[e]*=t[e];A.realTransform(L,E);for(let e=0;e<C;++e){const t=e<<1;z[e*S+o]=L[t]**2+L[t+1]**2}}if(null!==l&&2!==l){const e=2/l;for(let t=0;t<z.length;++t)z[t]**=e}const I=h.length;let B=await(0,a.matmul)(new a.Tensor("float32",h.flat(),[I,C]),new a.Tensor("float32",z,[C,S]));y&&(B=B.transpose(1,0));const N=B.data;for(let e=0;e<N.length;++e)N[e]=Math.max(m,N[e]);if(null!==l&&null!==_){const e=Math.min(N.length,v*I);switch(_){case"log":for(let t=0;t<e;++t)N[t]=Math.log(N[t]);break;case"log10":for(let t=0;t<e;++t)N[t]=Math.log10(N[t]);break;case"dB":if(1===l)!function(e,t=1,r=1e-5,n=null){f(e,20,t,r,n)}(N,g,M,w);else{if(2!==l)throw new Error(`Cannot use log_mel option '${_}' with power ${l}`);!function(e,t=1,r=1e-10,n=null){f(e,10,t,r,n)}(N,g,M,w)}break;default:throw new Error(`log_mel must be one of null, 'log', 'log10' or 'dB'. Got '${_}'`)}}return B}function M(e,t,{periodic:r=!0,frame_length:n=null,center:o=!0}={}){const s=r?e+1:e;let a;switch(t){case"boxcar":a=new Float64Array(s).fill(1);break;case"hann":case"hann_window":a=c(s);break;case"hamming":a=d(s);break;case"povey":a=c(s).map((e=>Math.pow(e,.85)));break;default:throw new Error(`Unknown window type ${t}.`)}if(r&&(a=a.subarray(0,e)),null===n)return a;if(e>n)throw new Error(`Length of the window (${e}) may not be larger than frame_length (${n})`);return a}},"./src/utils/constants.js":
@@ -136,10 +136,10 @@ import*as e from"fs";import*as t from"onnxruntime-node";import*as r from"path";i
   \********************************/(e,t,r)=>{r.r(t),r.d(t,{GITHUB_ISSUE_URL:()=>n});const n="https://github.com/xenova/transformers.js/issues/new/choose"},"./src/utils/core.js":
 /*!***************************!*\
   !*** ./src/utils/core.js ***!
-  \***************************/(e,t,r)=>{function n(e,t){e&&e(t)}function o(e){return Object.fromEntries(Object.entries(e).map((([e,t])=>[t,e])))}function s(e){return e.replace(/[.*+?^${}()|[\]\\]/g,"\\$&")}function a(e){return"TypedArray"===e?.prototype?.__proto__?.constructor?.name}function i(e){return Number.isInteger(e)||"bigint"==typeof e}function l(e){const t=[];let r=e;for(;Array.isArray(r);)t.push(r.length),r=r[0];return t}function c(e,t,r=void 0){const n=e[t];if(void 0!==n)return delete e[t],n;if(void 0===r)throw Error(`Key ${t} does not exist in object.`);return r}function d(...e){return Array.prototype.concat.apply([],e)}function u(...e){return e.reduce(((e,t)=>e.flatMap((e=>t.map((t=>[e,t]))))))}function p(e,t){return Math.abs((e+t)%(2*t)-t)}function h(e,t){return Object.assign({},...t.map((t=>{if(void 0!==e[t])return{[t]:e[t]}})))}r.r(t),r.d(t,{calculateDimensions:()=>l,calculateReflectOffset:()=>p,dispatchCallback:()=>n,escapeRegExp:()=>s,isIntegralNumber:()=>i,isTypedArray:()=>a,mergeArrays:()=>d,pick:()=>h,pop:()=>c,product:()=>u,reverseDictionary:()=>o})},"./src/utils/data-structures.js":
+  \***************************/(e,t,r)=>{function n(e,t){e&&e(t)}function o(e){return Object.fromEntries(Object.entries(e).map((([e,t])=>[t,e])))}function s(e){return e.replace(/[.*+?^${}()|[\]\\]/g,"\\$&")}function a(e){return"TypedArray"===e?.prototype?.__proto__?.constructor?.name}function i(e){return Number.isInteger(e)||"bigint"==typeof e}function l(e){const t=[];let r=e;for(;Array.isArray(r);)t.push(r.length),r=r[0];return t}function c(e,t,r=void 0){const n=e[t];if(void 0!==n)return delete e[t],n;if(void 0===r)throw Error(`Key ${t} does not exist in object.`);return r}function d(...e){return Array.prototype.concat.apply([],e)}function u(...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 r of e)++t;return t}r.r(t),r.d(t,{calculateDimensions:()=>l,calculateReflectOffset:()=>p,dispatchCallback:()=>n,escapeRegExp:()=>s,isIntegralNumber:()=>i,isTypedArray:()=>a,len:()=>m,mergeArrays:()=>d,pick:()=>h,pop:()=>c,product:()=>u,reverseDictionary:()=>o})},"./src/utils/data-structures.js":
 /*!**************************************!*\
   !*** ./src/utils/data-structures.js ***!
-  \**************************************/(e,t,r)=>{r.r(t),r.d(t,{CharTrie:()=>o,PriorityQueue:()=>n,TokenLattice:()=>a});class n{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 r=this._heap[e];this._heap[e]=this._heap[t],this._heap[t]=r}_siftUp(){this._siftUpFrom(this.size-1)}_siftUpFrom(e){for(;e>0&&this._greater(e,this._parent(e));)this._swap(e,this._parent(e)),e=this._parent(e)}_siftDown(){let e=0;for(;this._left(e)<this.size&&this._greater(this._left(e),e)||this._right(e)<this.size&&this._greater(this._right(e),e);){const t=this._right(e)<this.size&&this._greater(this._right(e),this._left(e))?this._right(e):this._left(e);this._swap(e,t),e=t}}_smallest(){return 2**Math.floor(Math.log2(this.size))-1}}class o{constructor(){this.root=s.default()}extend(e){for(let t of e)this.push(t)}push(e){let t=this.root;for(let r of e){let e=t.children.get(r);void 0===e&&(e=s.default(),t.children.set(r,e)),t=e}t.isLeaf=!0}*commonPrefixSearch(e){let t=this.root,r="";for(let n=0;n<e.length&&void 0!==t;++n){const o=e[n];r+=o,t=t.children.get(o),void 0!==t&&t.isLeaf&&(yield r)}}}class s{constructor(e,t){this.isLeaf=e,this.children=t}static default(){return new s(!1,new Map)}}class a{constructor(e,t,r){this.sentence=e,this.len=e.length,this.bosTokenId=t,this.eosTokenId=r,this.nodes=[],this.beginNodes=Array.from({length:this.len+1},(()=>[])),this.endNodes=Array.from({length:this.len+1},(()=>[]));const n=new i(this.bosTokenId,0,0,0,0),o=new i(this.eosTokenId,1,this.len,0,0);this.nodes.push(n.clone()),this.nodes.push(o.clone()),this.beginNodes[this.len].push(o),this.endNodes[0].push(n)}insert(e,t,r,n){const o=this.nodes.length,s=new i(n,o,e,t,r);this.beginNodes[e].push(s),this.endNodes[e+t].push(s),this.nodes.push(s)}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 r=0,n=null;for(let o of this.endNodes[t]){const t=o.backtraceScore+e.score;(null===n||t>r)&&(n=o.clone(),r=t)}if(null===n)return[];e.prev=n,e.backtraceScore=r}++t}const r=[],n=this.beginNodes[e][0].prev;if(null===n)return[];let o=n.clone();for(;null!==o.prev;){r.push(o.clone());const e=o.clone();o=e.prev.clone()}return r.reverse(),r}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 i{constructor(e,t,r,n,o){this.tokenId=e,this.nodeId=t,this.pos=r,this.length=n,this.score=o,this.prev=null,this.backtraceScore=0}clone(){const e=new i(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,r)=>{r.r(t),r.d(t,{CharTrie:()=>o,PriorityQueue:()=>n,TokenLattice:()=>a});class n{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 r=this._heap[e];this._heap[e]=this._heap[t],this._heap[t]=r}_siftUp(){this._siftUpFrom(this.size-1)}_siftUpFrom(e){for(;e>0&&this._greater(e,this._parent(e));)this._swap(e,this._parent(e)),e=this._parent(e)}_siftDown(){let e=0;for(;this._left(e)<this.size&&this._greater(this._left(e),e)||this._right(e)<this.size&&this._greater(this._right(e),e);){const t=this._right(e)<this.size&&this._greater(this._right(e),this._left(e))?this._right(e):this._left(e);this._swap(e,t),e=t}}_smallest(){return 2**Math.floor(Math.log2(this.size))-1}}class o{constructor(){this.root=s.default()}extend(e){for(const t of e)this.push(t)}push(e){let t=this.root;for(const r of e){let e=t.children.get(r);void 0===e&&(e=s.default(),t.children.set(r,e)),t=e}t.isLeaf=!0}*commonPrefixSearch(e){let t=this.root;if(void 0===t)return;let r="";for(const n of e){if(r+=n,t=t.children.get(n),void 0===t)return;t.isLeaf&&(yield r)}}}class s{constructor(e,t){this.isLeaf=e,this.children=t}static default(){return new s(!1,new Map)}}class a{constructor(e,t,r){this.chars=Array.from(e),this.len=this.chars.length,this.bosTokenId=t,this.eosTokenId=r,this.nodes=[],this.beginNodes=Array.from({length:this.len+1},(()=>[])),this.endNodes=Array.from({length:this.len+1},(()=>[]));const n=new i(this.bosTokenId,0,0,0,0),o=new i(this.eosTokenId,1,this.len,0,0);this.nodes.push(n.clone()),this.nodes.push(o.clone()),this.beginNodes[this.len].push(o),this.endNodes[0].push(n)}insert(e,t,r,n){const o=this.nodes.length,s=new i(n,o,e,t,r);this.beginNodes[e].push(s),this.endNodes[e+t].push(s),this.nodes.push(s)}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 r=0,n=null;for(let o of this.endNodes[t]){const t=o.backtraceScore+e.score;(null===n||t>r)&&(n=o.clone(),r=t)}if(null===n)return[];e.prev=n,e.backtraceScore=r}++t}const r=[],n=this.beginNodes[e][0].prev;if(null===n)return[];let o=n.clone();for(;null!==o.prev;){r.push(o.clone());const e=o.clone();o=e.prev.clone()}return r.reverse(),r}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 i{constructor(e,t,r,n,o){this.tokenId=e,this.nodeId=t,this.pos=r,this.length=n,this.score=o,this.prev=null,this.backtraceScore=0}clone(){const e=new i(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,r)=>{r.r(t),r.d(t,{DEVICE_TYPES:()=>n});const n=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":