@huggingface/transformers 3.0.0-alpha.13 → 3.0.0-alpha.14

package/dist/transformers.min.cjs

@@ -91,13 +91,13 @@
   \************************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{AutoConfig:()=>l,PretrainedConfig:()=>a,getKeyValueShapes:()=>i});var s=n(/*! ./utils/core.js */"./src/utils/core.js"),r=n(/*! ./utils/hub.js */"./src/utils/hub.js");function o(e){const t={};let n={};switch(e.model_type){case"llava":case"paligemma":case"florence2":n=o(e.text_config);break;case"moondream1":n=o(e.phi_config);break;case"musicgen":n=o(e.decoder);break;case"gpt2":case"gptj":case"jais":case"codegen":case"gpt_bigcode":t.num_heads="n_head",t.num_layers="n_layer",t.hidden_size="n_embd";break;case"gpt_neox":case"stablelm":case"opt":case"phi":case"phi3":case"falcon":t.num_heads="num_attention_heads",t.num_layers="num_hidden_layers",t.hidden_size="hidden_size";break;case"llama":case"cohere":case"mistral":case"starcoder2":case"qwen2":t.num_heads="num_key_value_heads",t.num_layers="num_hidden_layers",t.hidden_size="hidden_size",t.num_attention_heads="num_attention_heads";break;case"gemma":case"gemma2":t.num_heads="num_key_value_heads",t.num_layers="num_hidden_layers",t.dim_kv="head_dim";break;case"openelm":t.num_heads="num_kv_heads",t.num_layers="num_transformer_layers",t.dim_kv="head_dim";break;case"gpt_neo":case"donut-swin":t.num_heads="num_heads",t.num_layers="num_layers",t.hidden_size="hidden_size";break;case"bloom":t.num_heads="n_head",t.num_layers="n_layer",t.hidden_size="hidden_size";break;case"mpt":t.num_heads="n_heads",t.num_layers="n_layers",t.hidden_size="d_model";break;case"t5":case"mt5":case"longt5":t.num_decoder_layers="num_decoder_layers",t.num_decoder_heads="num_heads",t.decoder_dim_kv="d_kv",t.num_encoder_layers="num_layers",t.num_encoder_heads="num_heads",t.encoder_dim_kv="d_kv";break;case"bart":case"mbart":case"marian":case"whisper":case"m2m_100":case"blenderbot":case"blenderbot-small":case"florence2_language":t.num_decoder_layers="decoder_layers",t.num_decoder_heads="decoder_attention_heads",t.decoder_hidden_size="d_model",t.num_encoder_layers="encoder_layers",t.num_encoder_heads="encoder_attention_heads",t.encoder_hidden_size="d_model";break;case"speecht5":t.num_decoder_layers="decoder_layers",t.num_decoder_heads="decoder_attention_heads",t.decoder_hidden_size="hidden_size",t.num_encoder_layers="encoder_layers",t.num_encoder_heads="encoder_attention_heads",t.encoder_hidden_size="hidden_size";break;case"trocr":t.num_encoder_layers=t.num_decoder_layers="decoder_layers",t.num_encoder_heads=t.num_decoder_heads="decoder_attention_heads",t.encoder_hidden_size=t.decoder_hidden_size="d_model";break;case"musicgen_decoder":t.num_encoder_layers=t.num_decoder_layers="num_hidden_layers",t.num_encoder_heads=t.num_decoder_heads="num_attention_heads",t.encoder_hidden_size=t.decoder_hidden_size="hidden_size";break;case"vision-encoder-decoder":const r=o(e.decoder),i="num_decoder_layers"in r,a=(0,s.pick)(e,["model_type","is_encoder_decoder"]);return i?(a.num_decoder_layers=r.num_decoder_layers,a.num_decoder_heads=r.num_decoder_heads,a.decoder_hidden_size=r.decoder_hidden_size,a.num_encoder_layers=r.num_encoder_layers,a.num_encoder_heads=r.num_encoder_heads,a.encoder_hidden_size=r.encoder_hidden_size):(a.num_layers=r.num_layers,a.num_heads=r.num_heads,a.hidden_size=r.hidden_size),a}const r={...n,...(0,s.pick)(e,["model_type","multi_query","is_encoder_decoder"])};for(const n in t)r[n]=e[t[n]];return r}function i(e,{prefix:t="past_key_values"}={}){const n={},s=e.normalized_config;if(s.is_encoder_decoder&&"num_encoder_heads"in s&&"num_decoder_heads"in s){const e=s.encoder_dim_kv??s.encoder_hidden_size/s.num_encoder_heads,r=s.decoder_dim_kv??s.decoder_hidden_size/s.num_decoder_heads,o=[1,s.num_encoder_heads,0,e],i=[1,s.num_decoder_heads,0,r];for(let e=0;e<s.num_decoder_layers;++e)n[`${t}.${e}.encoder.key`]=o,n[`${t}.${e}.encoder.value`]=o,n[`${t}.${e}.decoder.key`]=i,n[`${t}.${e}.decoder.value`]=i}else{const e=s.num_heads,r=s.num_layers,o=s.dim_kv??s.hidden_size/(s.num_attention_heads??e);if("falcon"===s.model_type){const s=[1*e,0,o];for(let e=0;e<r;++e)n[`${t}.${e}.key`]=s,n[`${t}.${e}.value`]=s}else if(s.multi_query){const s=[1*e,0,2*o];for(let e=0;e<r;++e)n[`${t}.${e}.key_value`]=s}else if("bloom"===s.model_type){const s=[1*e,o,0],i=[1*e,0,o];for(let e=0;e<r;++e)n[`${t}.${e}.key`]=s,n[`${t}.${e}.value`]=i}else if("openelm"===s.model_type)for(let s=0;s<r;++s){const r=[1,e[s],0,o];n[`${t}.${s}.key`]=r,n[`${t}.${s}.value`]=r}else{const s=[1,e,0,o];for(let e=0;e<r;++e)n[`${t}.${e}.key`]=s,n[`${t}.${e}.value`]=s}}return n}class a{max_position_embeddings;constructor(e){this.model_type=null,this.is_encoder_decoder=!1,Object.assign(this,e),this.normalized_config=o(this)}static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:s=null,local_files_only:o=!1,revision:i="main"}={}){!n||n instanceof a||(n=new a(n));const l=n??await async function(e,t){return await(0,r.getModelJSON)(e,"config.json",!0,t)}(e,{progress_callback:t,config:n,cache_dir:s,local_files_only:o,revision:i});return new this(l)}}class l{static async from_pretrained(...e){return a.from_pretrained(...e)}}},"./src/env.js":
 /*!********************!*\
   !*** ./src/env.js ***!
-  \********************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{apis:()=>m,env:()=>b});var s=n(/*! fs */"fs"),r=n(/*! path */"path"),o=n(/*! url */"url");const i="undefined"!=typeof self,a=i&&"DedicatedWorkerGlobalScope"===self.constructor.name,l=i&&"caches"in self,c="undefined"!=typeof navigator&&"gpu"in navigator,d="undefined"!=typeof navigator&&"ml"in navigator,u="undefined"!=typeof process,p=u&&"node"===process?.release?.name,h=!y(s),_=!y(r),m=Object.freeze({IS_BROWSER_ENV:i,IS_WEBWORKER_ENV:a,IS_WEB_CACHE_AVAILABLE:l,IS_WEBGPU_AVAILABLE:c,IS_WEBNN_AVAILABLE:d,IS_PROCESS_AVAILABLE:u,IS_NODE_ENV:p,IS_FS_AVAILABLE:h,IS_PATH_AVAILABLE:_}),f=h&&_,g=f?r.dirname(r.dirname(o.fileURLToPath("file:///workspaces/transformers.js/src/env.js"))):"./",w=f?r.join(g,"/.cache/"):null,M="/models/",b={version:"3.0.0-alpha.13",backends:{onnx:{}},allowRemoteModels:!0,remoteHost:"https://huggingface.co/",remotePathTemplate:"{model}/resolve/{revision}/",allowLocalModels:!i,localModelPath:f?r.join(g,M):M,useFS:h,useBrowserCache:l,useFSCache:h,cacheDir:w,useCustomCache:!1,customCache:null};function y(e){return 0===Object.keys(e).length}},"./src/generation/configuration_utils.js":
+  \********************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{apis:()=>m,env:()=>b});var s=n(/*! fs */"fs"),r=n(/*! path */"path"),o=n(/*! url */"url");const i="undefined"!=typeof self,a=i&&"DedicatedWorkerGlobalScope"===self.constructor.name,l=i&&"caches"in self,c="undefined"!=typeof navigator&&"gpu"in navigator,d="undefined"!=typeof navigator&&"ml"in navigator,u="undefined"!=typeof process,p=u&&"node"===process?.release?.name,h=!y(s),_=!y(r),m=Object.freeze({IS_BROWSER_ENV:i,IS_WEBWORKER_ENV:a,IS_WEB_CACHE_AVAILABLE:l,IS_WEBGPU_AVAILABLE:c,IS_WEBNN_AVAILABLE:d,IS_PROCESS_AVAILABLE:u,IS_NODE_ENV:p,IS_FS_AVAILABLE:h,IS_PATH_AVAILABLE:_}),f=h&&_,g=f?r.dirname(r.dirname(o.fileURLToPath("file:///workspaces/transformers.js/src/env.js"))):"./",w=f?r.join(g,"/.cache/"):null,M="/models/",b={version:"3.0.0-alpha.14",backends:{onnx:{}},allowRemoteModels:!0,remoteHost:"https://huggingface.co/",remotePathTemplate:"{model}/resolve/{revision}/",allowLocalModels:!i,localModelPath:f?r.join(g,M):M,useFS:h,useBrowserCache:l,useFSCache:h,cacheDir:w,useCustomCache:!1,customCache:null};function y(e){return 0===Object.keys(e).length}},"./src/generation/configuration_utils.js":
 /*!***********************************************!*\
   !*** ./src/generation/configuration_utils.js ***!
   \***********************************************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{GenerationConfig:()=>r});var s=n(/*! ../utils/core.js */"./src/utils/core.js");class r{max_length=20;max_new_tokens=null;min_length=0;min_new_tokens=null;early_stopping=!1;max_time=null;do_sample=!1;num_beams=1;num_beam_groups=1;penalty_alpha=null;use_cache=!0;temperature=1;top_k=50;top_p=1;typical_p=1;epsilon_cutoff=0;eta_cutoff=0;diversity_penalty=0;repetition_penalty=1;encoder_repetition_penalty=1;length_penalty=1;no_repeat_ngram_size=0;bad_words_ids=null;force_words_ids=null;renormalize_logits=!1;constraints=null;forced_bos_token_id=null;forced_eos_token_id=null;remove_invalid_values=!1;exponential_decay_length_penalty=null;suppress_tokens=null;begin_suppress_tokens=null;forced_decoder_ids=null;guidance_scale=null;num_return_sequences=1;output_attentions=!1;output_hidden_states=!1;output_scores=!1;return_dict_in_generate=!1;pad_token_id=null;bos_token_id=null;eos_token_id=null;encoder_no_repeat_ngram_size=0;decoder_start_token_id=null;generation_kwargs={};constructor(e){Object.assign(this,(0,s.pick)(e,Object.getOwnPropertyNames(this)))}}},"./src/generation/logits_process.js":
 /*!******************************************!*\
   !*** ./src/generation/logits_process.js ***!
-  \******************************************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{ClassifierFreeGuidanceLogitsProcessor:()=>g,ForcedBOSTokenLogitsProcessor:()=>l,ForcedEOSTokenLogitsProcessor:()=>c,LogitsProcessor:()=>o,LogitsProcessorList:()=>a,LogitsWarper:()=>i,MinLengthLogitsProcessor:()=>_,MinNewTokensLengthLogitsProcessor:()=>m,NoBadWordsLogitsProcessor:()=>f,NoRepeatNGramLogitsProcessor:()=>p,RepetitionPenaltyLogitsProcessor:()=>h,SuppressTokensAtBeginLogitsProcessor:()=>d,TemperatureLogitsWarper:()=>w,TopKLogitsWarper:()=>b,TopPLogitsWarper:()=>M,WhisperTimeStampLogitsProcessor:()=>u});var s=n(/*! ../utils/generic.js */"./src/utils/generic.js"),r=(n(/*! ../utils/tensor.js */"./src/utils/tensor.js"),n(/*! ../utils/maths.js */"./src/utils/maths.js"));class o extends s.Callable{_call(e,t){throw Error("`_call` should be implemented in a subclass")}}class i extends s.Callable{_call(e,t){throw Error("`_call` should be implemented in a subclass")}}class a extends s.Callable{constructor(){super(),this.processors=[]}push(e){this.processors.push(e)}extend(e){this.processors.push(...e)}_call(e,t){let n=t;for(const t of this.processors)n=t(e,n);return n}[Symbol.iterator](){return this.processors.values()}}class l extends o{constructor(e){super(),this.bos_token_id=e}_call(e,t){for(let n=0;n<e.length;++n)if(1===e[n].length){const e=t[n].data;e.fill(-1/0),e[this.bos_token_id]=0}return t}}class c extends o{constructor(e,t){super(),this.max_length=e,this.eos_token_id=Array.isArray(t)?t:[t]}_call(e,t){for(let n=0;n<e.length;++n)if(e[n].length===this.max_length-1){const e=t[n].data;e.fill(-1/0);for(const t of this.eos_token_id)e[t]=0}return t}}class d extends o{constructor(e,t){super(),this.begin_suppress_tokens=e,this.begin_index=t}_call(e,t){for(let n=0;n<e.length;++n)if(e[n].length===this.begin_index){const e=t[n].data;for(const t of this.begin_suppress_tokens)e[t]=-1/0}return t}}class u extends o{constructor(e,t){super(),this.eos_token_id=Array.isArray(e.eos_token_id)?e.eos_token_id[0]:e.eos_token_id,this.no_timestamps_token_id=e.no_timestamps_token_id,this.timestamp_begin=this.no_timestamps_token_id+1,this.begin_index=t.length,t.at(-1)===this.no_timestamps_token_id&&(this.begin_index-=1),this.max_initial_timestamp_index=e.max_initial_timestamp_index}_call(e,t){for(let n=0;n<e.length;++n){const s=t[n].data;if(s[this.no_timestamps_token_id]=-1/0,e[n].length===this.begin_index-1){s.fill(-1/0),s[this.timestamp_begin]=0;continue}const o=e[n].slice(this.begin_index),i=o.length>=1&&o[o.length-1]>=this.timestamp_begin,a=o.length<2||o[o.length-2]>=this.timestamp_begin;if(i&&(a?s.subarray(this.timestamp_begin).fill(-1/0):s.subarray(0,this.eos_token_id).fill(-1/0)),e[n].length===this.begin_index&&null!==this.max_initial_timestamp_index){const e=this.timestamp_begin+this.max_initial_timestamp_index;s.subarray(e+1).fill(-1/0)}const l=(0,r.log_softmax)(s);Math.log(l.subarray(this.timestamp_begin).map(Math.exp).reduce(((e,t)=>e+t)))>(0,r.max)(l.subarray(0,this.timestamp_begin))[0]&&s.subarray(0,this.timestamp_begin).fill(-1/0)}return t}}class p extends o{constructor(e){super(),this.no_repeat_ngram_size=e}getNgrams(e){const t=e.length,n=[];for(let s=0;s<t+1-this.no_repeat_ngram_size;++s){const t=[];for(let n=0;n<this.no_repeat_ngram_size;++n)t.push(e[s+n]);n.push(t.map(Number))}const s=new Map;for(const e of n){const t=e.slice(0,e.length-1),n=JSON.stringify(t),r=s.get(n)??[];r.push(e[e.length-1]),s.set(n,r)}return s}getGeneratedNgrams(e,t){const n=t.slice(t.length+1-this.no_repeat_ngram_size,t.length);return e.get(JSON.stringify(n.map(Number)))??[]}calcBannedNgramTokens(e){const t=[];if(e.length+1<this.no_repeat_ngram_size)return t;{const t=this.getNgrams(e);return this.getGeneratedNgrams(t,e)}}_call(e,t){for(let n=0;n<e.length;++n){const s=t[n].data,r=this.calcBannedNgramTokens(e[n]);for(const e of r)s[e]=-1/0}return t}}class h extends o{constructor(e){super(),this.penalty=e}_call(e,t){for(let n=0;n<e.length;++n){const s=t[n].data;for(const t of e[n]){const e=Number(t);s[e]<0?s[e]*=this.penalty:s[e]/=this.penalty}}return t}}class _ extends o{constructor(e,t){super(),this.min_length=e,this.eos_token_id=Array.isArray(t)?t:[t]}_call(e,t){for(let n=0;n<e.length;++n)if(e[n].length<this.min_length){const e=t[n].data;for(const t of this.eos_token_id)e[t]=-1/0}return t}}class m extends o{constructor(e,t,n){super(),this.prompt_length_to_skip=e,this.min_new_tokens=t,this.eos_token_id=Array.isArray(n)?n:[n]}_call(e,t){for(let n=0;n<e.length;++n){if(e[n].length-this.prompt_length_to_skip<this.min_new_tokens){const e=t[n].data;for(const t of this.eos_token_id)e[t]=-1/0}}return t}}class f extends o{constructor(e,t){super(),this.bad_words_ids=e,this.eos_token_id=Array.isArray(t)?t:[t]}_call(e,t){for(let n=0;n<e.length;++n){const s=t[n].data;for(const t of this.bad_words_ids){let n=!0;for(let s=1;s<=t.length-1&&t.length<e[s].length;++s)if(t.at(-s-1)!=e[s].at(-s)){n=!1;break}n&&(s[t.at(-1)]=-1/0)}}return t}}class g extends o{constructor(e){if(super(),e<=1)throw new Error(`Require guidance scale >1 to use the classifier free guidance processor, got guidance scale ${e}.`);this.guidance_scale=e}_call(e,t){if(t.dims[0]!==2*e.length)throw new Error(`Logits should have twice the batch size of the input ids, the first half of batches corresponding to the conditional inputs, and the second half of batches corresponding to the unconditional inputs. Got batch size ${t.dims[0]} for the logits and ${e.length} for the input ids.`);const n=e.length,s=t.slice([0,n],null),r=t.slice([n,t.dims[0]],null);for(let e=0;e<r.data.length;++e)r.data[e]+=(s.data[e]-r.data[e])*this.guidance_scale;return r}}class w extends i{constructor(e){if(super(),"number"!=typeof e||e<=0){let t=`\`temperature\` (=${e}) must be a strictly positive float, otherwise your next token scores will be invalid.`;0===e&&(t+=" If you're looking for greedy decoding strategies, set `do_sample=false`.")}this.temperature=e}_call(e,t){const n=t.data;for(let e=0;e<n.length;++e)n[e]/=this.temperature;return t}}class M extends i{constructor(e,{filter_value:t=-1/0,min_tokens_to_keep:n=1}={}){if(super(),e<0||e>1)throw new Error(`\`top_p\` must be a float > 0 and < 1, but is ${e}`);if(!Number.isInteger(n)||n<1)throw new Error(`\`min_tokens_to_keep\` must be a positive integer, but is ${n}`);this.top_p=e,this.filter_value=t,this.min_tokens_to_keep=n}}class b extends i{constructor(e,{filter_value:t=-1/0,min_tokens_to_keep:n=1}={}){if(super(),!Number.isInteger(e)||e<0)throw new Error(`\`top_k\` must be a positive integer, but is ${e}`);this.top_k=Math.max(e,n),this.filter_value=t}}},"./src/generation/logits_sampler.js":
+  \******************************************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{ClassifierFreeGuidanceLogitsProcessor:()=>g,ForcedBOSTokenLogitsProcessor:()=>l,ForcedEOSTokenLogitsProcessor:()=>c,LogitsProcessor:()=>o,LogitsProcessorList:()=>a,LogitsWarper:()=>i,MinLengthLogitsProcessor:()=>_,MinNewTokensLengthLogitsProcessor:()=>m,NoBadWordsLogitsProcessor:()=>f,NoRepeatNGramLogitsProcessor:()=>p,RepetitionPenaltyLogitsProcessor:()=>h,SuppressTokensAtBeginLogitsProcessor:()=>d,TemperatureLogitsWarper:()=>w,TopKLogitsWarper:()=>b,TopPLogitsWarper:()=>M,WhisperTimeStampLogitsProcessor:()=>u});var s=n(/*! ../utils/generic.js */"./src/utils/generic.js"),r=(n(/*! ../utils/tensor.js */"./src/utils/tensor.js"),n(/*! ../utils/maths.js */"./src/utils/maths.js"));class o extends s.Callable{_call(e,t){throw Error("`_call` should be implemented in a subclass")}}class i extends s.Callable{_call(e,t){throw Error("`_call` should be implemented in a subclass")}}class a extends s.Callable{constructor(){super(),this.processors=[]}push(e){this.processors.push(e)}extend(e){this.processors.push(...e)}_call(e,t){let n=t;for(const t of this.processors)n=t(e,n);return n}[Symbol.iterator](){return this.processors.values()}}class l extends o{constructor(e){super(),this.bos_token_id=e}_call(e,t){for(let n=0;n<e.length;++n)if(1===e[n].length){const e=t[n].data;e.fill(-1/0),e[this.bos_token_id]=0}return t}}class c extends o{constructor(e,t){super(),this.max_length=e,this.eos_token_id=Array.isArray(t)?t:[t]}_call(e,t){for(let n=0;n<e.length;++n)if(e[n].length===this.max_length-1){const e=t[n].data;e.fill(-1/0);for(const t of this.eos_token_id)e[t]=0}return t}}class d extends o{constructor(e,t){super(),this.begin_suppress_tokens=e,this.begin_index=t}_call(e,t){for(let n=0;n<e.length;++n)if(e[n].length===this.begin_index){const e=t[n].data;for(const t of this.begin_suppress_tokens)e[t]=-1/0}return t}}class u extends o{constructor(e,t){super(),this.eos_token_id=Array.isArray(e.eos_token_id)?e.eos_token_id[0]:e.eos_token_id,this.no_timestamps_token_id=e.no_timestamps_token_id,this.timestamp_begin=this.no_timestamps_token_id+1,this.begin_index=t.length,t.at(-1)===this.no_timestamps_token_id&&(this.begin_index-=1),this.max_initial_timestamp_index=e.max_initial_timestamp_index}_call(e,t){for(let n=0;n<e.length;++n){const s=t[n].data;if(s[this.no_timestamps_token_id]=-1/0,e[n].length===this.begin_index-1){s.fill(-1/0),s[this.timestamp_begin]=0;continue}const o=e[n].slice(this.begin_index),i=o.length>=1&&o[o.length-1]>=this.timestamp_begin,a=o.length<2||o[o.length-2]>=this.timestamp_begin;if(i&&(a?s.subarray(this.timestamp_begin).fill(-1/0):s.subarray(0,this.eos_token_id).fill(-1/0)),e[n].length===this.begin_index&&null!==this.max_initial_timestamp_index){const e=this.timestamp_begin+this.max_initial_timestamp_index;s.subarray(e+1).fill(-1/0)}const l=(0,r.log_softmax)(s);Math.log(l.subarray(this.timestamp_begin).map(Math.exp).reduce(((e,t)=>e+t)))>(0,r.max)(l.subarray(0,this.timestamp_begin))[0]&&s.subarray(0,this.timestamp_begin).fill(-1/0)}return t}}class p extends o{constructor(e){super(),this.no_repeat_ngram_size=e}getNgrams(e){const t=e.length,n=[];for(let s=0;s<t+1-this.no_repeat_ngram_size;++s){const t=[];for(let n=0;n<this.no_repeat_ngram_size;++n)t.push(e[s+n]);n.push(t.map(Number))}const s=new Map;for(const e of n){const t=e.slice(0,e.length-1),n=JSON.stringify(t),r=s.get(n)??[];r.push(e[e.length-1]),s.set(n,r)}return s}getGeneratedNgrams(e,t){const n=t.slice(t.length+1-this.no_repeat_ngram_size,t.length);return e.get(JSON.stringify(n.map(Number)))??[]}calcBannedNgramTokens(e){const t=[];if(e.length+1<this.no_repeat_ngram_size)return t;{const t=this.getNgrams(e);return this.getGeneratedNgrams(t,e)}}_call(e,t){for(let n=0;n<e.length;++n){const s=t[n].data,r=this.calcBannedNgramTokens(e[n]);for(const e of r)s[e]=-1/0}return t}}class h extends o{constructor(e){super(),this.penalty=e}_call(e,t){for(let n=0;n<e.length;++n){const s=t[n].data;for(const t of e[n]){const e=Number(t);s[e]<0?s[e]*=this.penalty:s[e]/=this.penalty}}return t}}class _ extends o{constructor(e,t){super(),this.min_length=e,this.eos_token_id=Array.isArray(t)?t:[t]}_call(e,t){for(let n=0;n<e.length;++n)if(e[n].length<this.min_length){const e=t[n].data;for(const t of this.eos_token_id)e[t]=-1/0}return t}}class m extends o{constructor(e,t,n){super(),this.prompt_length_to_skip=e,this.min_new_tokens=t,this.eos_token_id=Array.isArray(n)?n:[n]}_call(e,t){for(let n=0;n<e.length;++n){if(e[n].length-this.prompt_length_to_skip<this.min_new_tokens){const e=t[n].data;for(const t of this.eos_token_id)e[t]=-1/0}}return t}}class f extends o{constructor(e,t){super(),this.bad_words_ids=e,this.eos_token_id=Array.isArray(t)?t:[t]}_call(e,t){for(let n=0;n<e.length;++n){const s=t[n].data,r=e[n];for(const e of this.bad_words_ids){let t=!0;for(let n=1;n<=e.length-1&&e.length<r.length;++n)if(e.at(-n-1)!=r.at(-n)){t=!1;break}t&&(s[e.at(-1)]=-1/0)}}return t}}class g extends o{constructor(e){if(super(),e<=1)throw new Error(`Require guidance scale >1 to use the classifier free guidance processor, got guidance scale ${e}.`);this.guidance_scale=e}_call(e,t){if(t.dims[0]!==2*e.length)throw new Error(`Logits should have twice the batch size of the input ids, the first half of batches corresponding to the conditional inputs, and the second half of batches corresponding to the unconditional inputs. Got batch size ${t.dims[0]} for the logits and ${e.length} for the input ids.`);const n=e.length,s=t.slice([0,n],null),r=t.slice([n,t.dims[0]],null);for(let e=0;e<r.data.length;++e)r.data[e]+=(s.data[e]-r.data[e])*this.guidance_scale;return r}}class w extends i{constructor(e){if(super(),"number"!=typeof e||e<=0){let t=`\`temperature\` (=${e}) must be a strictly positive float, otherwise your next token scores will be invalid.`;0===e&&(t+=" If you're looking for greedy decoding strategies, set `do_sample=false`.")}this.temperature=e}_call(e,t){const n=t.data;for(let e=0;e<n.length;++e)n[e]/=this.temperature;return t}}class M extends i{constructor(e,{filter_value:t=-1/0,min_tokens_to_keep:n=1}={}){if(super(),e<0||e>1)throw new Error(`\`top_p\` must be a float > 0 and < 1, but is ${e}`);if(!Number.isInteger(n)||n<1)throw new Error(`\`min_tokens_to_keep\` must be a positive integer, but is ${n}`);this.top_p=e,this.filter_value=t,this.min_tokens_to_keep=n}}class b extends i{constructor(e,{filter_value:t=-1/0,min_tokens_to_keep:n=1}={}){if(super(),!Number.isInteger(e)||e<0)throw new Error(`\`top_k\` must be a positive integer, but is ${e}`);this.top_k=Math.max(e,n),this.filter_value=t}}},"./src/generation/logits_sampler.js":
 /*!******************************************!*\
   !*** ./src/generation/logits_sampler.js ***!
   \******************************************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{LogitsSampler:()=>i});var s=n(/*! ../utils/generic.js */"./src/utils/generic.js"),r=n(/*! ../utils/tensor.js */"./src/utils/tensor.js"),o=n(/*! ../utils/maths.js */"./src/utils/maths.js");n(/*! ../generation/configuration_utils.js */"./src/generation/configuration_utils.js");class i extends s.Callable{constructor(e){super(),this.generation_config=e}async _call(e){return this.sample(e)}async sample(e){throw Error("sample should be implemented in subclasses.")}getLogits(e,t){let n=e.dims.at(-1),s=e.data;if(-1===t)s=s.slice(-n);else{let e=t*n;s=s.slice(e,e+n)}return s}randomSelect(e){let t=0;for(let n=0;n<e.length;++n)t+=e[n];let n=Math.random()*t;for(let t=0;t<e.length;++t)if(n-=e[t],n<=0)return t;return 0}static getSampler(e){if(e.do_sample)return new l(e);if(e.num_beams>1)return new c(e);if(e.num_return_sequences>1)throw Error(`num_return_sequences has to be 1 when doing greedy search, but is ${e.num_return_sequences}.`);return new a(e)}}class a extends i{async sample(e){const t=(0,o.max)(e.data)[1];return[[BigInt(t),0]]}}class l extends i{async sample(e){let t=e.dims.at(-1);this.generation_config.top_k>0&&(t=Math.min(this.generation_config.top_k,t));const[n,s]=await(0,r.topk)(e,t),i=(0,o.softmax)(n.data);return Array.from({length:this.generation_config.num_beams},(()=>{const e=this.randomSelect(i);return[s.data[e],Math.log(i[e])]}))}}class c extends i{async sample(e){let t=e.dims.at(-1);this.generation_config.top_k>0&&(t=Math.min(this.generation_config.top_k,t));const[n,s]=await(0,r.topk)(e,t),i=(0,o.softmax)(n.data);return Array.from({length:this.generation_config.num_beams},((e,t)=>[s.data[t],Math.log(i[t])]))}}},"./src/generation/stopping_criteria.js":
@@ -109,7 +109,7 @@
   \*************************************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{BaseStreamer:()=>i,TextStreamer:()=>l,WhisperTextStreamer:()=>c});var s=n(/*! ../utils/core.js */"./src/utils/core.js"),r=n(/*! ../tokenizers.js */"./src/tokenizers.js"),o=n(/*! ../env.js */"./src/env.js");class i{put(e){throw Error("Not implemented")}end(){throw Error("Not implemented")}}const a=o.apis.IS_PROCESS_AVAILABLE?e=>process.stdout.write(e):e=>console.log(e);class l extends i{constructor(e,{skip_prompt:t=!1,callback_function:n=null,token_callback_function:s=null,decode_kwargs:r={},...o}={}){super(),this.tokenizer=e,this.skip_prompt=t,this.callback_function=n??a,this.token_callback_function=s,this.decode_kwargs={...r,...o},this.token_cache=[],this.print_len=0,this.next_tokens_are_prompt=!0}put(e){if(e.length>1)throw Error("TextStreamer only supports batch size of 1");if(this.skip_prompt&&this.next_tokens_are_prompt)return void(this.next_tokens_are_prompt=!1);const t=e[0];this.token_callback_function?.(t),this.token_cache=(0,s.mergeArrays)(this.token_cache,t);const n=this.tokenizer.decode(this.token_cache,this.decode_kwargs);let o;n.endsWith("\n")?(o=n.slice(this.print_len),this.token_cache=[],this.print_len=0):n.length>0&&(0,r.is_chinese_char)(n.charCodeAt(n.length-1))?(o=n.slice(this.print_len),this.print_len+=o.length):(o=n.slice(this.print_len,n.lastIndexOf(" ")+1),this.print_len+=o.length),this.on_finalized_text(o,!1)}end(){let e;if(this.token_cache.length>0){e=this.tokenizer.decode(this.token_cache,this.decode_kwargs).slice(this.print_len),this.token_cache=[],this.print_len=0}else e="";this.next_tokens_are_prompt=!0,this.on_finalized_text(e,!0)}on_finalized_text(e,t){e.length>0&&this.callback_function?.(e),t&&this.callback_function===a&&o.apis.IS_PROCESS_AVAILABLE&&this.callback_function?.("\n")}}class c extends l{constructor(e,{skip_prompt:t=!1,callback_function:n=null,token_callback_function:s=null,on_chunk_start:r=null,on_chunk_end:o=null,on_finalize:i=null,time_precision:a=.02,skip_special_tokens:l=!0,decode_kwargs:c={}}={}){super(e,{skip_prompt:t,callback_function:n,token_callback_function:s,decode_kwargs:{skip_special_tokens:l,...c}}),this.timestamp_begin=e.timestamp_begin,this.on_chunk_start=r,this.on_chunk_end=o,this.on_finalize=i,this.time_precision=a,this.waiting_for_timestamp=!1}put(e){if(e.length>1)throw Error("WhisperTextStreamer only supports batch size of 1");const t=e[0];if(1===t.length){const n=Number(t[0])-this.timestamp_begin;if(n>=0){const t=n*this.time_precision;this.waiting_for_timestamp?this.on_chunk_end?.(t):this.on_chunk_start?.(t),this.waiting_for_timestamp=!this.waiting_for_timestamp,e=[[]]}}return super.put(e)}end(){super.end(),this.on_finalize?.()}}},"./src/models.js":
 /*!***********************!*\
   !*** ./src/models.js ***!
-  \***********************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{ASTForAudioClassification:()=>nn,ASTModel:()=>tn,ASTPreTrainedModel:()=>en,AlbertForMaskedLM:()=>ut,AlbertForQuestionAnswering:()=>dt,AlbertForSequenceClassification:()=>ct,AlbertModel:()=>lt,AlbertPreTrainedModel:()=>at,AutoModel:()=>da,AutoModelForAudioClassification:()=>Pa,AutoModelForAudioFrameClassification:()=>Aa,AutoModelForCTC:()=>Fa,AutoModelForCausalLM:()=>ga,AutoModelForDepthEstimation:()=>Ia,AutoModelForDocumentQuestionAnswering:()=>Ea,AutoModelForImageClassification:()=>ya,AutoModelForImageFeatureExtraction:()=>Na,AutoModelForImageMatting:()=>za,AutoModelForImageSegmentation:()=>xa,AutoModelForImageToImage:()=>La,AutoModelForMaskGeneration:()=>Ca,AutoModelForMaskedLM:()=>wa,AutoModelForNormalEstimation:()=>Ba,AutoModelForObjectDetection:()=>Ta,AutoModelForQuestionAnswering:()=>Ma,AutoModelForSemanticSegmentation:()=>ka,AutoModelForSeq2SeqLM:()=>ha,AutoModelForSequenceClassification:()=>ua,AutoModelForSpeechSeq2Seq:()=>_a,AutoModelForTextToSpectrogram:()=>ma,AutoModelForTextToWaveform:()=>fa,AutoModelForTokenClassification:()=>pa,AutoModelForVision2Seq:()=>ba,AutoModelForXVector:()=>Sa,AutoModelForZeroShotObjectDetection:()=>va,BartForConditionalGeneration:()=>kt,BartForSequenceClassification:()=>Tt,BartModel:()=>xt,BartPretrainedModel:()=>yt,BaseModelOutput:()=>$,BeitForImageClassification:()=>Qs,BeitModel:()=>Xs,BeitPreTrainedModel:()=>Ws,BertForMaskedLM:()=>W,BertForQuestionAnswering:()=>H,BertForSequenceClassification:()=>X,BertForTokenClassification:()=>Q,BertModel:()=>U,BertPreTrainedModel:()=>q,BlenderbotForConditionalGeneration:()=>zt,BlenderbotModel:()=>Et,BlenderbotPreTrainedModel:()=>At,BlenderbotSmallForConditionalGeneration:()=>Bt,BlenderbotSmallModel:()=>It,BlenderbotSmallPreTrainedModel:()=>Lt,BloomForCausalLM:()=>Ms,BloomModel:()=>ws,BloomPreTrainedModel:()=>gs,CLIPModel:()=>_n,CLIPPreTrainedModel:()=>hn,CLIPSegForImageSegmentation:()=>vn,CLIPSegModel:()=>Tn,CLIPSegPreTrainedModel:()=>kn,CLIPTextModelWithProjection:()=>mn,CLIPVisionModelWithProjection:()=>fn,CamembertForMaskedLM:()=>we,CamembertForQuestionAnswering:()=>ye,CamembertForSequenceClassification:()=>Me,CamembertForTokenClassification:()=>be,CamembertModel:()=>ge,CamembertPreTrainedModel:()=>fe,CausalLMOutput:()=>$a,CausalLMOutputWithPast:()=>qa,ChineseCLIPModel:()=>xn,ChineseCLIPPreTrainedModel:()=>yn,ClapAudioModelWithProjection:()=>ni,ClapModel:()=>ei,ClapPreTrainedModel:()=>Zo,ClapTextModelWithProjection:()=>ti,CodeGenForCausalLM:()=>Wn,CodeGenModel:()=>Un,CodeGenPreTrainedModel:()=>qn,CohereForCausalLM:()=>Kn,CohereModel:()=>Jn,CoherePreTrainedModel:()=>Yn,ConvBertForMaskedLM:()=>ie,ConvBertForQuestionAnswering:()=>ce,ConvBertForSequenceClassification:()=>ae,ConvBertForTokenClassification:()=>le,ConvBertModel:()=>oe,ConvBertPreTrainedModel:()=>re,ConvNextForImageClassification:()=>Or,ConvNextModel:()=>Nr,ConvNextPreTrainedModel:()=>Br,ConvNextV2ForImageClassification:()=>Vr,ConvNextV2Model:()=>Dr,ConvNextV2PreTrainedModel:()=>jr,DPTForDepthEstimation:()=>kr,DPTModel:()=>xr,DPTPreTrainedModel:()=>yr,DebertaForMaskedLM:()=>Te,DebertaForQuestionAnswering:()=>Fe,DebertaForSequenceClassification:()=>ve,DebertaForTokenClassification:()=>Ce,DebertaModel:()=>ke,DebertaPreTrainedModel:()=>xe,DebertaV2ForMaskedLM:()=>Ae,DebertaV2ForQuestionAnswering:()=>Le,DebertaV2ForSequenceClassification:()=>Ee,DebertaV2ForTokenClassification:()=>ze,DebertaV2Model:()=>Se,DebertaV2PreTrainedModel:()=>Pe,DeiTForImageClassification:()=>ur,DeiTModel:()=>dr,DeiTPreTrainedModel:()=>cr,DepthAnythingForDepthEstimation:()=>vr,DepthAnythingPreTrainedModel:()=>Tr,DetrForObjectDetection:()=>Js,DetrForSegmentation:()=>Ks,DetrModel:()=>Ys,DetrObjectDetectionOutput:()=>Zs,DetrPreTrainedModel:()=>Hs,DetrSegmentationOutput:()=>er,Dinov2ForImageClassification:()=>$r,Dinov2Model:()=>Gr,Dinov2PreTrainedModel:()=>Rr,DistilBertForMaskedLM:()=>De,DistilBertForQuestionAnswering:()=>je,DistilBertForSequenceClassification:()=>Ne,DistilBertForTokenClassification:()=>Oe,DistilBertModel:()=>Be,DistilBertPreTrainedModel:()=>Ie,DonutSwinModel:()=>Ir,DonutSwinPreTrainedModel:()=>Lr,EfficientNetForImageClassification:()=>_i,EfficientNetModel:()=>hi,EfficientNetPreTrainedModel:()=>pi,ElectraForMaskedLM:()=>pe,ElectraForQuestionAnswering:()=>me,ElectraForSequenceClassification:()=>he,ElectraForTokenClassification:()=>_e,ElectraModel:()=>ue,ElectraPreTrainedModel:()=>de,EsmForMaskedLM:()=>Ge,EsmForSequenceClassification:()=>$e,EsmForTokenClassification:()=>qe,EsmModel:()=>Re,EsmPreTrainedModel:()=>Ve,FalconForCausalLM:()=>Ko,FalconModel:()=>Jo,FalconPreTrainedModel:()=>Yo,FastViTForImageClassification:()=>Es,FastViTModel:()=>As,FastViTPreTrainedModel:()=>Ss,Florence2ForConditionalGeneration:()=>pn,Florence2PreTrainedModel:()=>un,GLPNForDepthEstimation:()=>zr,GLPNModel:()=>Er,GLPNPreTrainedModel:()=>Ar,GPT2LMHeadModel:()=>Pn,GPT2Model:()=>Fn,GPT2PreTrainedModel:()=>Cn,GPTBigCodeForCausalLM:()=>$n,GPTBigCodeModel:()=>Gn,GPTBigCodePreTrainedModel:()=>Rn,GPTJForCausalLM:()=>Vn,GPTJModel:()=>Dn,GPTJPreTrainedModel:()=>jn,GPTNeoForCausalLM:()=>In,GPTNeoModel:()=>Ln,GPTNeoPreTrainedModel:()=>zn,GPTNeoXForCausalLM:()=>On,GPTNeoXModel:()=>Nn,GPTNeoXPreTrainedModel:()=>Bn,Gemma2ForCausalLM:()=>rs,Gemma2Model:()=>ss,Gemma2PreTrainedModel:()=>ns,GemmaForCausalLM:()=>ts,GemmaModel:()=>es,GemmaPreTrainedModel:()=>Zn,HubertForCTC:()=>So,HubertForSequenceClassification:()=>Ao,HubertModel:()=>Po,HubertPreTrainedModel:()=>Fo,ImageMattingOutput:()=>Ua,JAISLMHeadModel:()=>En,JAISModel:()=>An,JAISPreTrainedModel:()=>Sn,LlamaForCausalLM:()=>Hn,LlamaModel:()=>Qn,LlamaPreTrainedModel:()=>Xn,LlavaForConditionalGeneration:()=>cn,LlavaPreTrainedModel:()=>ln,LongT5ForConditionalGeneration:()=>gt,LongT5Model:()=>ft,LongT5PreTrainedModel:()=>mt,M2M100ForConditionalGeneration:()=>no,M2M100Model:()=>to,M2M100PreTrainedModel:()=>eo,MBartForCausalLM:()=>St,MBartForConditionalGeneration:()=>Ft,MBartForSequenceClassification:()=>Pt,MBartModel:()=>Ct,MBartPreTrainedModel:()=>vt,MPNetForMaskedLM:()=>Ke,MPNetForQuestionAnswering:()=>tt,MPNetForSequenceClassification:()=>Ze,MPNetForTokenClassification:()=>et,MPNetModel:()=>Je,MPNetPreTrainedModel:()=>Ye,MT5ForConditionalGeneration:()=>bt,MT5Model:()=>Mt,MT5PreTrainedModel:()=>wt,MarianMTModel:()=>Zr,MarianModel:()=>Kr,MarianPreTrainedModel:()=>Jr,MaskedLMOutput:()=>Ra,MistralForCausalLM:()=>Wo,MistralModel:()=>Uo,MistralPreTrainedModel:()=>qo,MobileBertForMaskedLM:()=>Xe,MobileBertForQuestionAnswering:()=>He,MobileBertForSequenceClassification:()=>Qe,MobileBertModel:()=>We,MobileBertPreTrainedModel:()=>Ue,MobileNetV1ForImageClassification:()=>yi,MobileNetV1Model:()=>bi,MobileNetV1PreTrainedModel:()=>Mi,MobileNetV2ForImageClassification:()=>Ti,MobileNetV2Model:()=>ki,MobileNetV2PreTrainedModel:()=>xi,MobileNetV3ForImageClassification:()=>Fi,MobileNetV3Model:()=>Ci,MobileNetV3PreTrainedModel:()=>vi,MobileNetV4ForImageClassification:()=>Ai,MobileNetV4Model:()=>Si,MobileNetV4PreTrainedModel:()=>Pi,MobileViTForImageClassification:()=>Ns,MobileViTModel:()=>Bs,MobileViTPreTrainedModel:()=>Is,MobileViTV2ForImageClassification:()=>Ds,MobileViTV2Model:()=>js,MobileViTV2PreTrainedModel:()=>Os,ModelOutput:()=>G,Moondream1ForConditionalGeneration:()=>dn,MptForCausalLM:()=>xs,MptModel:()=>ys,MptPreTrainedModel:()=>bs,MusicgenForCausalLM:()=>gi,MusicgenForConditionalGeneration:()=>wi,MusicgenModel:()=>fi,MusicgenPreTrainedModel:()=>mi,NomicBertModel:()=>J,NomicBertPreTrainedModel:()=>Y,OPTForCausalLM:()=>vs,OPTModel:()=>Ts,OPTPreTrainedModel:()=>ks,OpenELMForCausalLM:()=>as,OpenELMModel:()=>is,OpenELMPreTrainedModel:()=>os,OwlViTForObjectDetection:()=>Gs,OwlViTModel:()=>Rs,OwlViTPreTrainedModel:()=>Vs,Owlv2ForObjectDetection:()=>Us,Owlv2Model:()=>qs,Owlv2PreTrainedModel:()=>$s,Phi3ForCausalLM:()=>fs,Phi3Model:()=>ms,Phi3PreTrainedModel:()=>_s,PhiForCausalLM:()=>hs,PhiModel:()=>ps,PhiPreTrainedModel:()=>us,PreTrainedModel:()=>R,PretrainedMixin:()=>Ei,PyAnnoteForAudioFrameClassification:()=>uo,PyAnnoteModel:()=>co,PyAnnotePreTrainedModel:()=>lo,QuestionAnsweringModelOutput:()=>Ga,Qwen2ForCausalLM:()=>ds,Qwen2Model:()=>cs,Qwen2PreTrainedModel:()=>ls,RTDetrForObjectDetection:()=>sr,RTDetrModel:()=>nr,RTDetrObjectDetectionOutput:()=>rr,RTDetrPreTrainedModel:()=>tr,ResNetForImageClassification:()=>_r,ResNetModel:()=>hr,ResNetPreTrainedModel:()=>pr,RoFormerForMaskedLM:()=>ee,RoFormerForQuestionAnswering:()=>se,RoFormerForSequenceClassification:()=>te,RoFormerForTokenClassification:()=>ne,RoFormerModel:()=>Z,RoFormerPreTrainedModel:()=>K,RobertaForMaskedLM:()=>jt,RobertaForQuestionAnswering:()=>Rt,RobertaForSequenceClassification:()=>Dt,RobertaForTokenClassification:()=>Vt,RobertaModel:()=>Ot,RobertaPreTrainedModel:()=>Nt,SamImageSegmentationOutput:()=>Yr,SamModel:()=>Hr,SamPreTrainedModel:()=>Qr,SapiensForDepthEstimation:()=>Pr,SapiensForNormalEstimation:()=>Sr,SapiensForSemanticSegmentation:()=>Fr,SapiensPreTrainedModel:()=>Cr,SegformerForImageClassification:()=>ai,SegformerForSemanticSegmentation:()=>li,SegformerModel:()=>ii,SegformerPreTrainedModel:()=>oi,Seq2SeqLMOutput:()=>Oa,SequenceClassifierOutput:()=>ja,SiglipModel:()=>wn,SiglipPreTrainedModel:()=>gn,SiglipTextModel:()=>Mn,SiglipVisionModel:()=>bn,SpeechT5ForSpeechToText:()=>Do,SpeechT5ForTextToSpeech:()=>Vo,SpeechT5HifiGan:()=>Ro,SpeechT5Model:()=>jo,SpeechT5PreTrainedModel:()=>Oo,SqueezeBertForMaskedLM:()=>rt,SqueezeBertForQuestionAnswering:()=>it,SqueezeBertForSequenceClassification:()=>ot,SqueezeBertModel:()=>st,SqueezeBertPreTrainedModel:()=>nt,StableLmForCausalLM:()=>ui,StableLmModel:()=>di,StableLmPreTrainedModel:()=>ci,Starcoder2ForCausalLM:()=>Ho,Starcoder2Model:()=>Qo,Starcoder2PreTrainedModel:()=>Xo,Swin2SRForImageSuperResolution:()=>br,Swin2SRModel:()=>Mr,Swin2SRPreTrainedModel:()=>wr,SwinForImageClassification:()=>gr,SwinModel:()=>fr,SwinPreTrainedModel:()=>mr,T5ForConditionalGeneration:()=>_t,T5Model:()=>ht,T5PreTrainedModel:()=>pt,TableTransformerForObjectDetection:()=>ar,TableTransformerModel:()=>ir,TableTransformerObjectDetectionOutput:()=>lr,TableTransformerPreTrainedModel:()=>or,TokenClassifierOutput:()=>Va,TrOCRForCausalLM:()=>$o,TrOCRPreTrainedModel:()=>Go,UniSpeechForCTC:()=>fo,UniSpeechForSequenceClassification:()=>go,UniSpeechModel:()=>mo,UniSpeechPreTrainedModel:()=>_o,UniSpeechSatForAudioFrameClassification:()=>xo,UniSpeechSatForCTC:()=>bo,UniSpeechSatForSequenceClassification:()=>yo,UniSpeechSatModel:()=>Mo,UniSpeechSatPreTrainedModel:()=>wo,ViTForImageClassification:()=>Ps,ViTModel:()=>Fs,ViTPreTrainedModel:()=>Cs,VisionEncoderDecoderModel:()=>an,VitMatteForImageMatting:()=>Ls,VitMattePreTrainedModel:()=>zs,VitsModel:()=>ri,VitsModelOutput:()=>Wa,VitsPreTrainedModel:()=>si,Wav2Vec2BertForCTC:()=>vo,Wav2Vec2BertForSequenceClassification:()=>Co,Wav2Vec2BertModel:()=>To,Wav2Vec2BertPreTrainedModel:()=>ko,Wav2Vec2ForAudioFrameClassification:()=>ao,Wav2Vec2ForCTC:()=>oo,Wav2Vec2ForSequenceClassification:()=>io,Wav2Vec2Model:()=>ro,Wav2Vec2PreTrainedModel:()=>so,WavLMForAudioFrameClassification:()=>No,WavLMForCTC:()=>Lo,WavLMForSequenceClassification:()=>Io,WavLMForXVector:()=>Bo,WavLMModel:()=>zo,WavLMPreTrainedModel:()=>Eo,WeSpeakerResNetModel:()=>ho,WeSpeakerResNetPreTrainedModel:()=>po,WhisperForConditionalGeneration:()=>on,WhisperModel:()=>rn,WhisperPreTrainedModel:()=>sn,XLMForQuestionAnswering:()=>Xt,XLMForSequenceClassification:()=>Ut,XLMForTokenClassification:()=>Wt,XLMModel:()=>$t,XLMPreTrainedModel:()=>Gt,XLMRobertaForMaskedLM:()=>Yt,XLMRobertaForQuestionAnswering:()=>Zt,XLMRobertaForSequenceClassification:()=>Jt,XLMRobertaForTokenClassification:()=>Kt,XLMRobertaModel:()=>Ht,XLMRobertaPreTrainedModel:()=>Qt,XLMWithLMHeadModel:()=>qt,XVectorOutput:()=>Da,YolosForObjectDetection:()=>Wr,YolosModel:()=>Ur,YolosObjectDetectionOutput:()=>Xr,YolosPreTrainedModel:()=>qr});var s=n(/*! ./configs.js */"./src/configs.js"),r=n(/*! ./backends/onnx.js */"./src/backends/onnx.js"),o=n(/*! ./utils/dtypes.js */"./src/utils/dtypes.js"),i=n(/*! ./utils/generic.js */"./src/utils/generic.js"),a=n(/*! ./utils/core.js */"./src/utils/core.js"),l=n(/*! ./utils/hub.js */"./src/utils/hub.js"),c=n(/*! ./generation/logits_process.js */"./src/generation/logits_process.js"),d=n(/*! ./generation/configuration_utils.js */"./src/generation/configuration_utils.js"),u=n(/*! ./utils/tensor.js */"./src/utils/tensor.js"),p=n(/*! ./utils/maths.js */"./src/utils/maths.js"),h=n(/*! ./generation/stopping_criteria.js */"./src/generation/stopping_criteria.js"),_=n(/*! ./generation/logits_sampler.js */"./src/generation/logits_sampler.js"),m=n(/*! ./env.js */"./src/env.js"),f=n(/*! ./models/whisper/generation_whisper.js */"./src/models/whisper/generation_whisper.js"),g=n(/*! ./models/whisper/common_whisper.js */"./src/models/whisper/common_whisper.js");const w=0,M=1,b=2,y=3,x=4,k=5,T=6,v=7,C=new Map,F=new Map,P=new Map;async function S(e,t,n){return Object.fromEntries(await Promise.all(Object.keys(t).map((async i=>{const{buffer:a,session_options:c}=await async function(e,t,n){let i=n.device;i&&"string"!=typeof i&&(i.hasOwnProperty(t)?i=i[t]:(console.warn(`device not specified for "${t}". Using the default device.`),i=null));const a=i??(m.apis.IS_NODE_ENV?"cpu":"wasm"),c=(0,r.deviceToExecutionProviders)(a);let d=n.dtype;"string"!=typeof d&&(d&&d.hasOwnProperty(t)?d=d[t]:(d=o.DEFAULT_DEVICE_DTYPE_MAPPING[a]??o.DATA_TYPES.fp32,console.warn(`dtype not specified for "${t}". Using the default dtype (${d}) for this device (${a}).`)));const u=d;if(!o.DEFAULT_DTYPE_SUFFIX_MAPPING.hasOwnProperty(u))throw new Error(`Invalid dtype: ${u}. Should be one of: ${Object.keys(o.DATA_TYPES).join(", ")}`);if(u===o.DATA_TYPES.fp16&&"webgpu"===a&&!await(0,o.isWebGpuFp16Supported)())throw new Error(`The device (${a}) does not support fp16.`);const p=o.DEFAULT_DTYPE_SUFFIX_MAPPING[u],h=`${n.subfolder??""}/${t}${p}.onnx`,_={...n.session_options}??{};_.executionProviders??=c;const f=(0,l.getModelFile)(e,h,!0,n);let g=[];if(n.use_external_data_format&&(!0===n.use_external_data_format||"object"==typeof n.use_external_data_format&&n.use_external_data_format.hasOwnProperty(t)&&!0===n.use_external_data_format[t])){if(m.apis.IS_NODE_ENV)throw new Error("External data format is not yet supported in Node.js");const s=`${t}${p}.onnx_data`,r=`${n.subfolder??""}/${s}`;g.push(new Promise((async(t,o)=>{const i=await(0,l.getModelFile)(e,r,!0,n);t({path:s,data:i})})))}else void 0!==_.externalData&&(g=_.externalData.map((async t=>{if("string"==typeof t.data){const s=await(0,l.getModelFile)(e,t.data,!0,n);return{...t,data:s}}return t})));if(g.length>0&&(_.externalData=await Promise.all(g)),"webgpu"===a){const e=(0,s.getKeyValueShapes)(n.config,{prefix:"present"});if(Object.keys(e).length>0&&!(0,r.isONNXProxy)()){const t={};for(const n in e)t[n]="gpu-buffer";_.preferredOutputLocation=t}}return{buffer:await f,session_options:_}}(e,t[i],n);return[i,await(0,r.createInferenceSession)(a,c)]}))))}async function A(e,t){const n=function(e,t){const n=Object.create(null),s=[];for(const o of e.inputNames){const e=t[o];e instanceof u.Tensor?n[o]=(0,r.isONNXProxy)()?e.clone():e:s.push(o)}if(s.length>0)throw new Error(`An error occurred during model execution: "Missing the following inputs: ${s.join(", ")}.`);const o=Object.keys(t).length,i=e.inputNames.length;if(o>i){let n=Object.keys(t).filter((t=>!e.inputNames.includes(t)));console.warn(`WARNING: Too many inputs were provided (${o} > ${i}). The following inputs will be ignored: "${n.join(", ")}".`)}return n}(e,t);try{const t=Object.fromEntries(Object.entries(n).map((([e,t])=>[e,t.ort_tensor])));let s=await e.run(t);return s=E(s),s}catch(e){throw console.error(`An error occurred during model execution: "${e}".`),console.error("Inputs given to model:",n),e}}function E(e){for(let t in e)(0,r.isONNXTensor)(e[t])?e[t]=new u.Tensor(e[t]):"object"==typeof e[t]&&E(e[t]);return e}function z(e){if(e instanceof u.Tensor)return e;if(0===e.length)throw Error("items must be non-empty");if(Array.isArray(e[0])){if(e.some((t=>t.length!==e[0].length)))throw Error("Unable to create tensor, you should probably activate truncation and/or padding with 'padding=True' and/or 'truncation=True' to have batched tensors with the same length.");return new u.Tensor("int64",BigInt64Array.from(e.flat().map((e=>BigInt(e)))),[e.length,e[0].length])}return new u.Tensor("int64",BigInt64Array.from(e.map((e=>BigInt(e)))),[1,e.length])}function L(e){return new u.Tensor("bool",[e],[1])}async function I(e,t){let{encoder_outputs:n,input_ids:s,decoder_input_ids:r,...o}=t;if(!n){const s=(0,a.pick)(t,e.sessions.model.inputNames);n=(await B(e,s)).last_hidden_state}o.input_ids=r,o.encoder_hidden_states=n,e.sessions.decoder_model_merged.inputNames.includes("encoder_attention_mask")&&(o.encoder_attention_mask=t.attention_mask);return await N(e,o,!0)}async function B(e,t){const n=e.sessions.model,s=(0,a.pick)(t,n.inputNames);if(n.inputNames.includes("inputs_embeds")&&!s.inputs_embeds){if(!t.input_ids)throw new Error("Both `input_ids` and `inputs_embeds` are missing in the model inputs.");s.inputs_embeds=await e.encode_text({input_ids:t.input_ids})}return n.inputNames.includes("token_type_ids")&&!s.token_type_ids&&(s.token_type_ids=new u.Tensor("int64",new BigInt64Array(s.input_ids.data.length),s.input_ids.dims)),await A(n,s)}async function N(e,t,n=!1){const s=e.sessions[n?"decoder_model_merged":"model"],{past_key_values:r,...o}=t;s.inputNames.includes("use_cache_branch")&&(o.use_cache_branch=L(!!r)),s.inputNames.includes("position_ids")&&o.attention_mask&&!o.position_ids&&(o.position_ids=function(e,t=null){const{input_ids:n,inputs_embeds:s,attention_mask:r}=e,[o,i]=r.dims,a=new BigInt64Array(r.data.length);for(let e=0;e<o;++e){const t=e*i;let n=BigInt(0);for(let e=0;e<i;++e){const s=t+e;0n===r.data[s]?a[s]=BigInt(1):(a[s]=n,n+=r.data[s])}}let l=new u.Tensor("int64",a,r.dims);if(t){const e=-(n??s).dims.at(1);l=l.slice(null,[e,null])}return l}(o,r)),e.addPastKeyValues(o,r);const i=(0,a.pick)(o,s.inputNames);return await A(s,i)}async function O(e,{input_ids:t=null,attention_mask:n=null,pixel_values:s=null,position_ids:r=null,inputs_embeds:o=null,past_key_values:i=null,generation_config:a=null,logits_processor:l=null,...c}){if(!o)if(o=await e.encode_text({input_ids:t}),s&&1!==t.dims[1]){const r=await e.encode_image({pixel_values:s});({inputs_embeds:o,attention_mask:n}=e._merge_input_ids_with_image_features({image_features:r,inputs_embeds:o,input_ids:t,attention_mask:n}))}else if(i&&s&&1===t.dims[1]){const e=t.dims[1],s=Object.values(i)[0].dims.at(-2);n=(0,u.cat)([(0,u.ones)([t.dims[0],s]),n.slice(null,[n.dims[1]-e,n.dims[1]])],1)}return await N(e,{inputs_embeds:o,past_key_values:i,attention_mask:n,position_ids:r,generation_config:a,logits_processor:l},!0)}function j(e,t,n,s){if(n.past_key_values){const t=Object.values(n.past_key_values)[0].dims.at(-2),{input_ids:s,attention_mask:r}=n;if(r&&r.dims[1]>s.dims[1]);else if(t<s.dims[1])n.input_ids=s.slice(null,[t,null]);else if(null!=e.config.image_token_index&&s.data.some((t=>t==e.config.image_token_index))){const r=e.config.num_image_tokens;if(!r)throw new Error("`num_image_tokens` is missing in the model configuration.");const o=s.dims[1]-(t-r);n.input_ids=s.slice(null,[-o,null]),n.attention_mask=(0,u.ones)([1,t+o])}}return n}function D(e,t,n,s){return n.past_key_values&&(t=t.map((e=>[e.at(-1)]))),{...n,decoder_input_ids:z(t)}}function V(e,...t){return e.config.is_encoder_decoder?D(e,...t):j(e,...t)}class R extends i.Callable{main_input_name="input_ids";forward_params=["input_ids","attention_mask"];constructor(e,t){super(),this.config=e,this.sessions=t;const n=P.get(this.constructor),s=C.get(n);switch(this.can_generate=!1,this._forward=null,this._prepare_inputs_for_generation=null,s){case x:this.can_generate=!0,this._forward=N,this._prepare_inputs_for_generation=j;break;case b:case y:case v:this.can_generate=!0,this._forward=I,this._prepare_inputs_for_generation=D;break;case M:this._forward=I;break;case T:this.can_generate=!0,this._forward=O,this._prepare_inputs_for_generation=V;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:n=null,cache_dir:r=null,local_files_only:o=!1,revision:i="main",model_file_name:a=null,subfolder:c="onnx",device:d=null,dtype:u=null,use_external_data_format:p=null,session_options:h={}}={}){let _={progress_callback:t,config:n,cache_dir:r,local_files_only:o,revision:i,model_file_name:a,subfolder:c,device:d,dtype:u,use_external_data_format:p,session_options:h};const m=P.get(this),f=C.get(m);let g;if(n=_.config=await s.AutoConfig.from_pretrained(e,_),f===x)g=await Promise.all([S(e,{model:_.model_file_name??"model"},_),(0,l.getModelJSON)(e,"generation_config.json",!1,_)]);else if(f===b||f===y)g=await Promise.all([S(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},_),(0,l.getModelJSON)(e,"generation_config.json",!1,_)]);else if(f===k)g=await Promise.all([S(e,{model:"vision_encoder",prompt_encoder_mask_decoder:"prompt_encoder_mask_decoder"},_)]);else if(f===M)g=await Promise.all([S(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},_)]);else if(f===T){const t={embed_tokens:"embed_tokens",vision_encoder:"vision_encoder",decoder_model_merged:"decoder_model_merged"};n.is_encoder_decoder&&(t.model="encoder_model"),g=await Promise.all([S(e,t,_),(0,l.getModelJSON)(e,"generation_config.json",!1,_)])}else f===v?g=await Promise.all([S(e,{model:"text_encoder",decoder_model_merged:"decoder_model_merged",encodec_decode:"encodec_decode"},_),(0,l.getModelJSON)(e,"generation_config.json",!1,_)]):(f!==w&&console.warn(`Model type for '${m??n?.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:_.model_file_name??"model"},_)]));return new this(n,...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,n=null){const s=new c.LogitsProcessorList;if(null!==e.repetition_penalty&&1!==e.repetition_penalty&&s.push(new c.RepetitionPenaltyLogitsProcessor(e.repetition_penalty)),null!==e.no_repeat_ngram_size&&e.no_repeat_ngram_size>0&&s.push(new c.NoRepeatNGramLogitsProcessor(e.no_repeat_ngram_size)),null!==e.bad_words_ids&&s.push(new c.NoBadWordsLogitsProcessor(e.bad_words_ids,e.eos_token_id)),null!==e.min_length&&null!==e.eos_token_id&&e.min_length>0&&s.push(new c.MinLengthLogitsProcessor(e.min_length,e.eos_token_id)),null!==e.min_new_tokens&&null!==e.eos_token_id&&e.min_new_tokens>0&&s.push(new c.MinNewTokensLengthLogitsProcessor(t,e.min_new_tokens,e.eos_token_id)),null!==e.forced_bos_token_id&&s.push(new c.ForcedBOSTokenLogitsProcessor(e.forced_bos_token_id)),null!==e.forced_eos_token_id&&s.push(new c.ForcedEOSTokenLogitsProcessor(e.max_length,e.forced_eos_token_id)),null!==e.begin_suppress_tokens){const n=t>1||null===e.forced_bos_token_id?t:t+1;s.push(new c.SuppressTokensAtBeginLogitsProcessor(e.begin_suppress_tokens,n))}return null!==e.guidance_scale&&e.guidance_scale>1&&s.push(new c.ClassifierFreeGuidanceLogitsProcessor(e.guidance_scale)),null!==n&&s.extend(n),s}_prepare_generation_config(e,t,n=d.GenerationConfig){const s={...this.config};for(const e of["decoder","generator","text_config"])e in s&&Object.assign(s,s[e]);const r=new n(s);return"generation_config"in this&&Object.assign(r,this.generation_config),e&&Object.assign(r,e),t&&Object.assign(r,(0,a.pick)(t,Object.getOwnPropertyNames(r))),r}_get_stopping_criteria(e,t=null){const n=new h.StoppingCriteriaList;return null!==e.max_length&&n.push(new h.MaxLengthCriteria(e.max_length,this.config.max_position_embeddings??null)),null!==e.eos_token_id&&n.push(new h.EosTokenCriteria(e.eos_token_id)),t&&n.extend(t),n}_validate_model_class(){if(!this.can_generate){const e=[Ri,qi,Vi,Bi],t=P.get(this.constructor),n=new Set,s=this.config.model_type;for(const t of e){const e=t.get(s);e&&n.add(e[0])}let r=`The current model class (${t}) is not compatible with \`.generate()\`, as it doesn't have a language model head.`;throw n.size>0&&(r+=` Please use the following class instead: ${[...n].join(", ")}`),Error(r)}}prepare_inputs_for_generation(...e){return this._prepare_inputs_for_generation(this,...e)}_update_model_kwargs_for_generation({generated_input_ids:e,outputs:t,model_inputs:n,is_encoder_decoder:s}){return n.past_key_values=this.getPastKeyValues(t,n.past_key_values),n.input_ids=new u.Tensor("int64",e.flat(),[e.length,1]),s||(n.attention_mask=(0,u.cat)([n.attention_mask,(0,u.ones)([n.attention_mask.dims[0],1])],1)),n.position_ids=null,n}_prepare_model_inputs({inputs:e,bos_token_id:t,model_kwargs:n}){const s=(0,a.pick)(n,this.forward_params),r=this.main_input_name;if(r in s){if(e)throw new Error("`inputs`: {inputs}` were passed alongside {input_name} which is not allowed. Make sure to either pass {inputs} or {input_name}=...")}else s[r]=e;return{inputs_tensor:s[r],model_inputs:s,model_input_name:r}}async _prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:e,model_inputs:t,model_input_name:n,generation_config:s}){if(this.sessions.model.inputNames.includes("inputs_embeds")&&!t.inputs_embeds&&"_prepare_inputs_embeds"in this){const{input_ids:e,pixel_values:n,attention_mask:s,...r}=t,o=await this._prepare_inputs_embeds(t);t={...r,...(0,a.pick)(o,["inputs_embeds","attention_mask"])}}let{last_hidden_state:r}=await B(this,t);if(null!==s.guidance_scale&&s.guidance_scale>1)r=(0,u.cat)([r,(0,u.full_like)(r,0)],0),"attention_mask"in t&&(t.attention_mask=(0,u.cat)([t.attention_mask,(0,u.zeros_like)(t.attention_mask)],0));else if(t.decoder_input_ids){const e=z(t.decoder_input_ids).dims[0];if(e!==r.dims[0]){if(1!==r.dims[0])throw new Error(`The encoder outputs have a different batch size (${r.dims[0]}) than the decoder inputs (${e}).`);r=(0,u.cat)(Array.from({length:e},(()=>r)),0)}}return t.encoder_outputs=r,t}_prepare_decoder_input_ids_for_generation({batch_size:e,model_input_name:t,model_kwargs:n,decoder_start_token_id:s,bos_token_id:r,generation_config:o}){let{decoder_input_ids:i,...a}=n;if(i)Array.isArray(i[0])||(i=Array.from({length:e},(()=>i)));else if(s??=r,"musicgen"===this.config.model_type)i=Array.from({length:e*this.config.decoder.num_codebooks},(()=>[s]));else if(Array.isArray(s)){if(s.length!==e)throw new Error(`\`decoder_start_token_id\` expcted to have length ${e} but got ${s.length}`);i=s}else i=Array.from({length:e},(()=>[s]));return i=z(i),n.decoder_attention_mask=(0,u.ones_like)(i),{input_ids:i,model_inputs:a}}async generate({inputs:e=null,generation_config:t=null,logits_processor:n=null,stopping_criteria:s=null,streamer:r=null,...o}){this._validate_model_class(),t=this._prepare_generation_config(t,o);let{inputs_tensor:i,model_inputs:a,model_input_name:l}=this._prepare_model_inputs({inputs:e,model_kwargs:o});const c=this.config.is_encoder_decoder;let d;c&&("encoder_outputs"in a||(a=await this._prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:i,model_inputs:a,model_input_name:l,generation_config:t}))),c?({input_ids:d,model_inputs:a}=this._prepare_decoder_input_ids_for_generation({batch_size:a[l].dims.at(0),model_input_name:l,model_kwargs:a,decoder_start_token_id:t.decoder_start_token_id,bos_token_id:t.bos_token_id,generation_config:t})):d=a[l];let p=d.dims.at(-1);null!==t.max_new_tokens&&(t.max_length=p+t.max_new_tokens);const h=this._get_logits_processor(t,p,n),m=this._get_stopping_criteria(t,s),f=a[l].dims.at(0),g=_.LogitsSampler.getSampler(t),w=new Array(f).fill(0),M=d.tolist();r&&r.put(M);let b=null,y={};for(;;){a=this.prepare_inputs_for_generation(M,a,t);const e=await this.forward(a);if(t.output_attentions&&t.return_dict_in_generate){const t=this.getAttentions(e);for(const e in t)e in y||(y[e]=[]),y[e].push(t[e])}const n=h(M,e.logits.slice(null,-1,null)),s=[];for(let e=0;e<n.dims.at(0);++e){const t=n[e],r=await g(t);for(const[t,n]of r){const r=BigInt(t);w[e]+=n,M[e].push(r),s.push([r]);break}}r&&r.put(s);if(m(M).every((e=>e))){t.return_dict_in_generate&&(b=this.getPastKeyValues(e,a.past_key_values,!1));break}a=this._update_model_kwargs_for_generation({generated_input_ids:s,outputs:e,model_inputs:a,is_encoder_decoder:c})}r&&r.end();const x=new u.Tensor("int64",M.flat(),[M.length,M[0].length]);return t.return_dict_in_generate?{sequences:x,past_key_values:b,...y}:x}getPastKeyValues(e,t,n=!0){const s=Object.create(null);for(const r in e)if(r.startsWith("present")){const o=r.replace("present","past_key_values");if(t&&r.includes("encoder"))s[o]=t[o];else{if(n&&t){const e=t[o];"gpu-buffer"===e.location&&e.dispose()}s[o]=e[r]}}return s}getAttentions(e){const t={};for(const n of["cross_attentions","encoder_attentions","decoder_attentions"])for(const s in e)s.startsWith(n)&&(n in t||(t[n]=[]),t[n].push(e[s]));return t}addPastKeyValues(e,t){if(t)Object.assign(e,t);else{const t=this.custom_config.kv_cache_dtype??"float32",n="float16"===t?new Uint16Array:[],r=(0,s.getKeyValueShapes)(this.config);for(const s in r)e[s]=new u.Tensor(t,n,r[s])}}async encode_image({pixel_values:e}){const t=(await 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 $ extends G{constructor({last_hidden_state:e,hidden_states:t=null,attentions:n=null}){super(),this.last_hidden_state=e,this.hidden_states=t,this.attentions=n}}class q extends R{}class U extends q{}class W extends q{async _call(e){return new Ra(await super._call(e))}}class X extends q{async _call(e){return new ja(await super._call(e))}}class Q extends q{async _call(e){return new Va(await super._call(e))}}class H extends q{async _call(e){return new Ga(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 Ra(await super._call(e))}}class te extends K{async _call(e){return new ja(await super._call(e))}}class ne extends K{async _call(e){return new Va(await super._call(e))}}class se extends K{async _call(e){return new Ga(await super._call(e))}}class re extends R{}class oe extends re{}class ie extends re{async _call(e){return new Ra(await super._call(e))}}class ae extends re{async _call(e){return new ja(await super._call(e))}}class le extends re{async _call(e){return new Va(await super._call(e))}}class ce extends re{async _call(e){return new Ga(await super._call(e))}}class de extends R{}class ue extends de{}class pe extends de{async _call(e){return new Ra(await super._call(e))}}class he extends de{async _call(e){return new ja(await super._call(e))}}class _e extends de{async _call(e){return new Va(await super._call(e))}}class me extends de{async _call(e){return new Ga(await super._call(e))}}class fe extends R{}class ge extends fe{}class we extends fe{async _call(e){return new Ra(await super._call(e))}}class Me extends fe{async _call(e){return new ja(await super._call(e))}}class be extends fe{async _call(e){return new Va(await super._call(e))}}class ye extends fe{async _call(e){return new Ga(await super._call(e))}}class xe extends R{}class ke extends xe{}class Te extends xe{async _call(e){return new Ra(await super._call(e))}}class ve extends xe{async _call(e){return new ja(await super._call(e))}}class Ce extends xe{async _call(e){return new Va(await super._call(e))}}class Fe extends xe{async _call(e){return new Ga(await super._call(e))}}class Pe extends R{}class Se extends Pe{}class Ae extends Pe{async _call(e){return new Ra(await super._call(e))}}class Ee extends Pe{async _call(e){return new ja(await super._call(e))}}class ze extends Pe{async _call(e){return new Va(await super._call(e))}}class Le extends Pe{async _call(e){return new Ga(await super._call(e))}}class Ie extends R{}class Be extends Ie{}class Ne extends Ie{async _call(e){return new ja(await super._call(e))}}class Oe extends Ie{async _call(e){return new Va(await super._call(e))}}class je extends Ie{async _call(e){return new Ga(await super._call(e))}}class De extends Ie{async _call(e){return new Ra(await super._call(e))}}class Ve extends R{}class Re extends Ve{}class Ge extends Ve{async _call(e){return new Ra(await super._call(e))}}class $e extends Ve{async _call(e){return new ja(await super._call(e))}}class qe extends Ve{async _call(e){return new Va(await super._call(e))}}class Ue extends R{}class We extends Ue{}class Xe extends Ue{async _call(e){return new Ra(await super._call(e))}}class Qe extends Ue{async _call(e){return new ja(await super._call(e))}}class He extends Ue{async _call(e){return new Ga(await super._call(e))}}class Ye extends R{}class Je extends Ye{}class Ke extends Ye{async _call(e){return new Ra(await super._call(e))}}class Ze extends Ye{async _call(e){return new ja(await super._call(e))}}class et extends Ye{async _call(e){return new Va(await super._call(e))}}class tt extends Ye{async _call(e){return new Ga(await super._call(e))}}class nt extends R{}class st extends nt{}class rt extends nt{async _call(e){return new Ra(await super._call(e))}}class ot extends nt{async _call(e){return new ja(await super._call(e))}}class it extends nt{async _call(e){return new Ga(await super._call(e))}}class at extends R{}class lt extends at{}class ct extends at{async _call(e){return new ja(await super._call(e))}}class dt extends at{async _call(e){return new Ga(await super._call(e))}}class ut extends at{async _call(e){return new Ra(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,n){super(e,t),this.generation_config=n}}class ht extends pt{}class _t extends pt{}class mt extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class ft extends mt{}class gt extends mt{}class wt extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Mt extends wt{}class bt extends wt{}class yt extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class xt extends yt{}class kt extends yt{}class Tt extends yt{async _call(e){return new ja(await super._call(e))}}class vt extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Ct extends vt{}class Ft extends vt{}class Pt extends vt{async _call(e){return new ja(await super._call(e))}}class St extends vt{}class At extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Et extends At{}class zt extends At{}class Lt extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class It extends Lt{}class Bt extends Lt{}class Nt extends R{}class Ot extends Nt{}class jt extends Nt{async _call(e){return new Ra(await super._call(e))}}class Dt extends Nt{async _call(e){return new ja(await super._call(e))}}class Vt extends Nt{async _call(e){return new Va(await super._call(e))}}class Rt extends Nt{async _call(e){return new Ga(await super._call(e))}}class Gt extends R{}class $t extends Gt{}class qt extends Gt{async _call(e){return new Ra(await super._call(e))}}class Ut extends Gt{async _call(e){return new ja(await super._call(e))}}class Wt extends Gt{async _call(e){return new Va(await super._call(e))}}class Xt extends Gt{async _call(e){return new Ga(await super._call(e))}}class Qt extends R{}class Ht extends Qt{}class Yt extends Qt{async _call(e){return new Ra(await super._call(e))}}class Jt extends Qt{async _call(e){return new ja(await super._call(e))}}class Kt extends Qt{async _call(e){return new Va(await super._call(e))}}class Zt extends Qt{async _call(e){return new Ga(await super._call(e))}}class en extends R{}class tn extends en{}class nn extends en{}class sn extends R{requires_attention_mask=!1;main_input_name="input_features";forward_params=["input_features","attention_mask","decoder_input_ids","decoder_attention_mask","past_key_values"];constructor(e,t,n){super(e,t),this.generation_config=n}}class rn extends sn{}class on extends sn{_prepare_generation_config(e,t){return super._prepare_generation_config(e,t,f.WhisperGenerationConfig)}_retrieve_init_tokens(e){const t=[e.decoder_start_token_id];let n=e.language;const s=e.task;if(e.is_multilingual){n||(console.warn("No language specified - defaulting to English (en)."),n="en");const r=`<|${(0,g.whisper_language_to_code)(n)}|>`;t.push(e.lang_to_id[r]),t.push(e.task_to_id[s??"transcribe"])}else if(n||s)throw new Error("Cannot specify `task` or `language` for an English-only model. If the model is intended to be multilingual, pass `is_multilingual=true` to generate, or update the generation config.");return!e.return_timestamps&&e.no_timestamps_token_id&&t.at(-1)!==e.no_timestamps_token_id?t.push(e.no_timestamps_token_id):e.return_timestamps&&t.at(-1)===e.no_timestamps_token_id&&(console.warn("<|notimestamps|> prompt token is removed from generation_config since `return_timestamps` is set to `true`."),t.pop()),t.filter((e=>null!=e))}async generate({inputs:e=null,generation_config:t=null,logits_processor:n=null,stopping_criteria:s=null,...r}){t=this._prepare_generation_config(t,r);const o=r.decoder_input_ids??this._retrieve_init_tokens(t);if(t.return_timestamps&&(n??=new c.LogitsProcessorList,n.push(new c.WhisperTimeStampLogitsProcessor(t,o))),t.begin_suppress_tokens&&(n??=new c.LogitsProcessorList,n.push(new c.SuppressTokensAtBeginLogitsProcessor(t.begin_suppress_tokens,o.length))),t.return_token_timestamps){if(!t.alignment_heads)throw new Error("Model generation config has no `alignment_heads`, token-level timestamps not available. See https://gist.github.com/hollance/42e32852f24243b748ae6bc1f985b13a on how to add this property to the generation config.");"translate"===t.task&&console.warn("Token-level timestamps may not be reliable for task 'translate'."),t.output_attentions=!0,t.return_dict_in_generate=!0}const i=await super.generate({inputs:e,generation_config:t,logits_processor:n,decoder_input_ids:o,...r});return t.return_token_timestamps&&(i.token_timestamps=this._extract_token_timestamps(i,t.alignment_heads,t.num_frames)),i}_extract_token_timestamps(e,t,n=null,s=.02){if(!e.cross_attentions)throw new Error("Model outputs must contain cross attentions to extract timestamps. This is most likely because the model was not exported with `output_attentions=True`.");null==n&&console.warn("`num_frames` has not been set, meaning the entire audio will be analyzed. This may lead to inaccurate token-level timestamps for short audios (< 30 seconds).");let r=this.config.median_filter_width;void 0===r&&(console.warn("Model config has no `median_filter_width`, using default value of 7."),r=7);const o=e.cross_attentions,i=Array.from({length:this.config.decoder_layers},((e,t)=>(0,u.cat)(o.map((e=>e[t])),2))),l=(0,u.stack)(t.map((([e,t])=>{if(e>=i.length)throw new Error(`Layer index ${e} is out of bounds for cross attentions (length ${i.length}).`);return n?i[e].slice(null,t,null,[0,n]):i[e].slice(null,t)}))).transpose(1,0,2,3),[c,d]=(0,u.std_mean)(l,-2,0,!0),h=l.clone();for(let e=0;e<h.dims[0];++e){const t=h[e];for(let n=0;n<t.dims[0];++n){const s=t[n],o=c[e][n][0].data,i=d[e][n][0].data;for(let e=0;e<s.dims[0];++e){let t=s[e].data;for(let e=0;e<t.length;++e)t[e]=(t[e]-i[e])/o[e];t.set((0,p.medianFilter)(t,r))}}}const _=[(0,u.mean)(h,1)],m=e.sequences.dims,f=new u.Tensor("float32",new Float32Array(m[0]*m[1]),m);for(let e=0;e<m[0];++e){const t=_[e].neg().squeeze_(0),[n,r]=(0,p.dynamic_time_warping)(t.tolist()),o=Array.from({length:n.length-1},((e,t)=>n[t+1]-n[t])),i=(0,a.mergeArrays)([1],o).map((e=>!!e)),l=[];for(let e=0;e<i.length;++e)i[e]&&l.push(r[e]*s);f[e].data.set(l,1)}return f}}class an extends R{main_input_name="pixel_values";forward_params=["pixel_values","input_ids","encoder_hidden_states","past_key_values"];constructor(e,t,n){super(e,t),this.generation_config=n}}class ln extends R{forward_params=["input_ids","pixel_values","attention_mask","position_ids","past_key_values"];constructor(e,t,n){super(e,t),this.generation_config=n}}class cn extends ln{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:n,attention_mask:s}){const r=this.config.image_token_index,o=n.tolist().map((e=>e.findIndex((e=>e==r)))),i=o.every((e=>-1===e)),a=o.every((e=>-1!==e));if(!i&&!a)throw new Error("Every input should contain either 0 or 1 image token.");if(i)return{inputs_embeds:e,attention_mask:s};const l=[],c=[];for(let n=0;n<o.length;++n){const r=o[n],i=e[n],a=t[n],d=s[n];l.push((0,u.cat)([i.slice([0,r]),a,i.slice([r+1,i.dims[0]])],0)),c.push((0,u.cat)([d.slice([0,r]),(0,u.ones)([a.dims[0]]),d.slice([r+1,d.dims[0]])],0))}return{inputs_embeds:(0,u.stack)(l,0),attention_mask:(0,u.stack)(c,0)}}}class dn extends cn{}class un 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,n){super(e,t),this.generation_config=n}}class pn extends un{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:n,attention_mask:s}){return{inputs_embeds:(0,u.cat)([t,e],1),attention_mask:(0,u.cat)([(0,u.ones)(t.dims.slice(0,2)),s],1)}}async _prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:n,attention_mask:s}){if(!e&&!t)throw new Error("Either `input_ids` or `pixel_values` should be provided.");let r,o;return e&&(r=await this.encode_text({input_ids:e})),t&&(o=await this.encode_image({pixel_values:t})),r&&o?({inputs_embeds:n,attention_mask:s}=this._merge_input_ids_with_image_features({inputs_embeds:r,image_features:o,input_ids:e,attention_mask:s})):n=r||o,{inputs_embeds:n,attention_mask:s}}async forward({input_ids:e,pixel_values:t,attention_mask:n,decoder_input_ids:s,decoder_attention_mask:r,encoder_outputs:o,past_key_values:i,inputs_embeds:a,decoder_inputs_embeds:l}){if(a||({inputs_embeds:a,attention_mask:n}=await this._prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:a,attention_mask:n})),!o){let{last_hidden_state:e}=await B(this,{inputs_embeds:a,attention_mask:n});o=e}if(!l){if(!s)throw new Error("Either `decoder_input_ids` or `decoder_inputs_embeds` should be provided.");l=await this.encode_text({input_ids:s})}const c={inputs_embeds:l,attention_mask:r,encoder_attention_mask:n,encoder_hidden_states:o,past_key_values:i};return await N(this,c,!0)}}class hn extends R{}class _n extends hn{}class mn extends hn{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class fn extends hn{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class gn extends R{}class wn extends gn{}class Mn extends gn{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class bn extends hn{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class yn extends R{}class xn extends yn{}class kn extends R{}class Tn extends kn{}class vn extends kn{}class Cn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Fn extends Cn{}class Pn extends Cn{}class Sn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class An extends Sn{}class En extends Sn{}class zn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Ln extends zn{}class In extends zn{}class Bn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Nn extends Bn{}class On extends Bn{}class jn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Dn extends jn{}class Vn extends jn{}class Rn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Gn extends Rn{}class $n extends Rn{}class qn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Un extends qn{}class Wn extends qn{}class Xn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Qn extends Xn{}class Hn extends Xn{}class Yn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Jn extends Yn{}class Kn extends Yn{}class Zn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class es extends Zn{}class ts extends Zn{}class ns extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class ss extends ns{}class rs extends ns{}class os extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class is extends os{}class as extends os{}class ls extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class cs extends ls{}class ds extends ls{}class us extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class ps extends us{}class hs extends us{}class _s extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class ms extends _s{}class fs extends _s{}class gs extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class ws extends gs{}class Ms extends gs{}class bs extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class ys extends bs{}class xs extends bs{}class ks extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Ts extends ks{}class vs extends ks{}class Cs extends R{}class Fs extends Cs{}class Ps extends Cs{async _call(e){return new ja(await super._call(e))}}class Ss extends R{}class As extends Ss{}class Es extends Ss{async _call(e){return new ja(await super._call(e))}}class zs extends R{}class Ls extends zs{async _call(e){return new Ua(await super._call(e))}}class Is extends R{}class Bs extends Is{}class Ns extends Is{async _call(e){return new ja(await super._call(e))}}class Os extends R{}class js extends Os{}class Ds extends Os{async _call(e){return new ja(await super._call(e))}}class Vs extends R{}class Rs extends Vs{}class Gs extends Vs{}class $s extends R{}class qs extends $s{}class Us extends $s{}class Ws extends R{}class Xs extends Ws{}class Qs extends Ws{async _call(e){return new ja(await super._call(e))}}class Hs extends R{}class Ys extends Hs{}class Js extends Hs{async _call(e){return new Zs(await super._call(e))}}class Ks extends Hs{async _call(e){return new er(await super._call(e))}}class Zs extends G{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class er extends G{constructor({logits:e,pred_boxes:t,pred_masks:n}){super(),this.logits=e,this.pred_boxes=t,this.pred_masks=n}}class tr extends R{}class nr extends tr{}class sr extends tr{async _call(e){return new rr(await super._call(e))}}class rr extends G{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class or extends R{}class ir extends or{}class ar extends or{async _call(e){return new lr(await super._call(e))}}class lr extends Zs{}class cr extends R{}class dr extends cr{}class ur extends cr{async _call(e){return new ja(await super._call(e))}}class pr extends R{}class hr extends pr{}class _r extends pr{async _call(e){return new ja(await super._call(e))}}class mr extends R{}class fr extends mr{}class gr extends mr{async _call(e){return new ja(await super._call(e))}}class wr extends R{}class Mr extends wr{}class br extends wr{}class yr extends R{}class xr extends yr{}class kr extends yr{}class Tr extends R{}class vr extends Tr{}class Cr extends R{}class Fr extends Cr{}class Pr extends Cr{}class Sr extends Cr{}class Ar extends R{}class Er extends Ar{}class zr extends Ar{}class Lr extends R{}class Ir extends Lr{}class Br extends R{}class Nr extends Br{}class Or extends Br{async _call(e){return new ja(await super._call(e))}}class jr extends R{}class Dr extends jr{}class Vr extends jr{async _call(e){return new ja(await super._call(e))}}class Rr extends R{}class Gr extends Rr{}class $r extends Rr{async _call(e){return new ja(await super._call(e))}}class qr extends R{}class Ur extends qr{}class Wr extends qr{async _call(e){return new Xr(await super._call(e))}}class Xr extends G{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class Qr extends R{}class Hr extends Qr{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),n=t.reduce(((e,t)=>e*t),1);e.input_labels=new u.Tensor("int64",new BigInt64Array(n).fill(1n),t)}const t={image_embeddings:e.image_embeddings,image_positional_embeddings:e.image_positional_embeddings};return e.input_points&&(t.input_points=e.input_points),e.input_labels&&(t.input_labels=e.input_labels),e.input_boxes&&(t.input_boxes=e.input_boxes),await A(this.sessions.prompt_encoder_mask_decoder,t)}async _call(e){return new Yr(await super._call(e))}}class Yr extends G{constructor({iou_scores:e,pred_masks:t}){super(),this.iou_scores=e,this.pred_masks=t}}class Jr extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Kr extends Jr{}class Zr extends Jr{}class eo extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class to extends eo{}class no extends eo{}class so extends R{}class ro extends so{}class oo extends so{async _call(e){return new $a(await super._call(e))}}class io extends so{async _call(e){return new ja(await super._call(e))}}class ao extends so{async _call(e){return new Va(await super._call(e))}}class lo extends R{}class co extends lo{}class uo extends lo{async _call(e){return new Va(await super._call(e))}}class po extends R{}class ho extends po{}class _o extends R{}class mo extends _o{}class fo extends _o{async _call(e){return new $a(await super._call(e))}}class go extends _o{async _call(e){return new ja(await super._call(e))}}class wo extends R{}class Mo extends wo{}class bo extends wo{async _call(e){return new $a(await super._call(e))}}class yo extends wo{async _call(e){return new ja(await super._call(e))}}class xo extends wo{async _call(e){return new Va(await super._call(e))}}class ko extends R{}class To extends ko{}class vo extends ko{async _call(e){return new $a(await super._call(e))}}class Co extends ko{async _call(e){return new ja(await super._call(e))}}class Fo extends R{}class Po extends so{}class So extends so{async _call(e){return new $a(await super._call(e))}}class Ao extends so{async _call(e){return new ja(await super._call(e))}}class Eo extends R{}class zo extends Eo{}class Lo extends Eo{async _call(e){return new $a(await super._call(e))}}class Io extends Eo{async _call(e){return new ja(await super._call(e))}}class Bo extends Eo{async _call(e){return new Da(await super._call(e))}}class No extends Eo{async _call(e){return new Va(await super._call(e))}}class Oo extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class jo extends Oo{}class Do extends Oo{}class Vo extends Oo{async generate_speech(e,t,{threshold:n=.5,minlenratio:s=0,maxlenratio:r=20,vocoder:o=null}={}){const i={input_ids:e},{encoder_outputs:a,encoder_attention_mask:l}=await B(this,i),c=a.dims[1]/this.config.reduction_factor,d=Math.floor(c*r),p=Math.floor(c*s),h=this.config.num_mel_bins;let _=[],m=null,f=null,g=0;for(;;){++g;const e=L(!!f);let s;s=f?f.output_sequence_out:new u.Tensor("float32",new Float32Array(h),[1,1,h]);let r={use_cache_branch:e,output_sequence:s,encoder_attention_mask:l,speaker_embeddings:t,encoder_hidden_states:a};this.addPastKeyValues(r,m),f=await A(this.sessions.decoder_model_merged,r),m=this.getPastKeyValues(f,m);const{prob:o,spectrum:i}=f;if(_.push(i),g>=p&&(Array.from(o.data).filter((e=>e>=n)).length>0||g>=d))break}const w=(0,u.cat)(_),{waveform:M}=await A(o.sessions.model,{spectrogram:w});return{spectrogram:w,waveform:M}}}class Ro extends R{main_input_name="spectrogram"}class Go extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class $o extends Go{}class qo extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Uo extends qo{}class Wo extends qo{}class Xo extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Qo extends Xo{}class Ho extends Xo{}class Yo extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Jo extends Yo{}class Ko extends Yo{}class Zo extends R{}class ei extends Zo{}class ti extends Zo{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class ni extends Zo{static async from_pretrained(e,t={}){return t.model_file_name??="audio_model",super.from_pretrained(e,t)}}class si extends R{}class ri extends si{async _call(e){return new Wa(await super._call(e))}}class oi extends R{}class ii extends oi{}class ai extends oi{}class li extends oi{}class ci extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class di extends ci{}class ui extends ci{}class pi extends R{}class hi extends pi{}class _i extends pi{async _call(e){return new ja(await super._call(e))}}class mi extends R{}class fi extends mi{}class gi extends mi{}class wi extends R{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"];constructor(e,t,n){super(e,t),this.generation_config=n}_apply_and_filter_by_delay_pattern_mask(e){const[t,n]=e.dims,s=this.config.decoder.num_codebooks,r=n-s;let o=0;for(let t=0;t<e.size;++t){if(e.data[t]===this.config.decoder.pad_token_id)continue;const i=t%n-Math.floor(t/n)%s;i>0&&i<=r&&(e.data[o++]=e.data[t])}const i=Math.floor(t/s),a=o/(i*s);return new u.Tensor(e.type,e.data.slice(0,o),[i,s,a])}prepare_inputs_for_generation(e,t,n){let s=structuredClone(e);for(let e=0;e<s.length;++e)for(let t=0;t<s[e].length;++t)e%this.config.decoder.num_codebooks>=t&&(s[e][t]=BigInt(this.config.decoder.pad_token_id));null!==n.guidance_scale&&n.guidance_scale>1&&(s=s.concat(s));return super.prepare_inputs_for_generation(s,t,n)}async generate(e){const t=await super.generate(e),n=this._apply_and_filter_by_delay_pattern_mask(t).unsqueeze_(0),{audio_values:s}=await A(this.sessions.encodec_decode,{audio_codes:n});return s}}class Mi extends R{}class bi extends Mi{}class yi extends Mi{async _call(e){return new ja(await super._call(e))}}class xi extends R{}class ki extends xi{}class Ti extends xi{async _call(e){return new ja(await super._call(e))}}class vi extends R{}class Ci extends vi{}class Fi extends vi{async _call(e){return new ja(await super._call(e))}}class Pi extends R{}class Si extends Pi{}class Ai extends Pi{async _call(e){return new ja(await super._call(e))}}class Ei{static MODEL_CLASS_MAPPINGS=null;static BASE_IF_FAIL=!1;static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:r=null,local_files_only:o=!1,revision:i="main",model_file_name:a=null,subfolder:l="onnx",device:c=null,dtype:d=null,use_external_data_format:u=null,session_options:p={}}={}){let h={progress_callback:t,config:n,cache_dir:r,local_files_only:o,revision:i,model_file_name:a,subfolder:l,device:c,dtype:d,use_external_data_format:u,session_options:p};if(h.config=await s.AutoConfig.from_pretrained(e,h),!this.MODEL_CLASS_MAPPINGS)throw new Error("`MODEL_CLASS_MAPPINGS` not implemented for this type of `AutoClass`: "+this.name);for(let t of this.MODEL_CLASS_MAPPINGS){const n=t.get(h.config.model_type);if(n)return await n[1].from_pretrained(e,h)}if(this.BASE_IF_FAIL)return console.warn(`Unknown model class "${h.config.model_type}", attempting to construct from base class.`),await R.from_pretrained(e,h);throw Error(`Unsupported model type: ${h.config.model_type}`)}}const zi=new Map([["bert",["BertModel",U]],["nomic_bert",["NomicBertModel",J]],["roformer",["RoFormerModel",Z]],["electra",["ElectraModel",ue]],["esm",["EsmModel",Re]],["convbert",["ConvBertModel",oe]],["camembert",["CamembertModel",ge]],["deberta",["DebertaModel",ke]],["deberta-v2",["DebertaV2Model",Se]],["mpnet",["MPNetModel",Je]],["albert",["AlbertModel",lt]],["distilbert",["DistilBertModel",Be]],["roberta",["RobertaModel",Ot]],["xlm",["XLMModel",$t]],["xlm-roberta",["XLMRobertaModel",Ht]],["clap",["ClapModel",ei]],["clip",["CLIPModel",_n]],["clipseg",["CLIPSegModel",Tn]],["chinese_clip",["ChineseCLIPModel",xn]],["siglip",["SiglipModel",wn]],["mobilebert",["MobileBertModel",We]],["squeezebert",["SqueezeBertModel",st]],["wav2vec2",["Wav2Vec2Model",ro]],["wav2vec2-bert",["Wav2Vec2BertModel",To]],["unispeech",["UniSpeechModel",mo]],["unispeech-sat",["UniSpeechSatModel",Mo]],["hubert",["HubertModel",Po]],["wavlm",["WavLMModel",zo]],["audio-spectrogram-transformer",["ASTModel",tn]],["vits",["VitsModel",ri]],["pyannote",["PyAnnoteModel",co]],["wespeaker-resnet",["WeSpeakerResNetModel",ho]],["detr",["DetrModel",Ys]],["rt_detr",["RTDetrModel",nr]],["table-transformer",["TableTransformerModel",ir]],["vit",["ViTModel",Fs]],["fastvit",["FastViTModel",As]],["mobilevit",["MobileViTModel",Bs]],["mobilevitv2",["MobileViTV2Model",js]],["owlvit",["OwlViTModel",Rs]],["owlv2",["Owlv2Model",qs]],["beit",["BeitModel",Xs]],["deit",["DeiTModel",dr]],["convnext",["ConvNextModel",Nr]],["convnextv2",["ConvNextV2Model",Dr]],["dinov2",["Dinov2Model",Gr]],["resnet",["ResNetModel",hr]],["swin",["SwinModel",fr]],["swin2sr",["Swin2SRModel",Mr]],["donut-swin",["DonutSwinModel",Ir]],["yolos",["YolosModel",Ur]],["dpt",["DPTModel",xr]],["glpn",["GLPNModel",Er]],["hifigan",["SpeechT5HifiGan",Ro]],["efficientnet",["EfficientNetModel",hi]],["mobilenet_v1",["MobileNetV1Model",bi]],["mobilenet_v2",["MobileNetV2Model",ki]],["mobilenet_v3",["MobileNetV3Model",Ci]],["mobilenet_v4",["MobileNetV4Model",Si]]]),Li=new Map([["t5",["T5Model",ht]],["longt5",["LongT5Model",ft]],["mt5",["MT5Model",Mt]],["bart",["BartModel",xt]],["mbart",["MBartModel",Ct]],["marian",["MarianModel",Kr]],["whisper",["WhisperModel",rn]],["m2m_100",["M2M100Model",to]],["blenderbot",["BlenderbotModel",Et]],["blenderbot-small",["BlenderbotSmallModel",It]]]),Ii=new Map([["bloom",["BloomModel",ws]],["jais",["JAISModel",An]],["gpt2",["GPT2Model",Fn]],["gptj",["GPTJModel",Dn]],["gpt_bigcode",["GPTBigCodeModel",Gn]],["gpt_neo",["GPTNeoModel",Ln]],["gpt_neox",["GPTNeoXModel",Nn]],["codegen",["CodeGenModel",Un]],["llama",["LlamaModel",Qn]],["cohere",["CohereModel",Jn]],["gemma",["GemmaModel",es]],["gemma2",["Gemma2Model",ss]],["openelm",["OpenELMModel",is]],["qwen2",["Qwen2Model",cs]],["phi",["PhiModel",ps]],["phi3",["Phi3Model",ms]],["mpt",["MptModel",ys]],["opt",["OPTModel",Ts]],["mistral",["MistralModel",Uo]],["starcoder2",["Starcoder2Model",Qo]],["falcon",["FalconModel",Jo]],["stablelm",["StableLmModel",di]]]),Bi=new Map([["speecht5",["SpeechT5ForSpeechToText",Do]],["whisper",["WhisperForConditionalGeneration",on]]]),Ni=new Map([["speecht5",["SpeechT5ForTextToSpeech",Vo]]]),Oi=new Map([["vits",["VitsModel",ri]],["musicgen",["MusicgenForConditionalGeneration",wi]]]),ji=new Map([["bert",["BertForSequenceClassification",X]],["roformer",["RoFormerForSequenceClassification",te]],["electra",["ElectraForSequenceClassification",he]],["esm",["EsmForSequenceClassification",$e]],["convbert",["ConvBertForSequenceClassification",ae]],["camembert",["CamembertForSequenceClassification",Me]],["deberta",["DebertaForSequenceClassification",ve]],["deberta-v2",["DebertaV2ForSequenceClassification",Ee]],["mpnet",["MPNetForSequenceClassification",Ze]],["albert",["AlbertForSequenceClassification",ct]],["distilbert",["DistilBertForSequenceClassification",Ne]],["roberta",["RobertaForSequenceClassification",Dt]],["xlm",["XLMForSequenceClassification",Ut]],["xlm-roberta",["XLMRobertaForSequenceClassification",Jt]],["bart",["BartForSequenceClassification",Tt]],["mbart",["MBartForSequenceClassification",Pt]],["mobilebert",["MobileBertForSequenceClassification",Qe]],["squeezebert",["SqueezeBertForSequenceClassification",ot]]]),Di=new Map([["bert",["BertForTokenClassification",Q]],["roformer",["RoFormerForTokenClassification",ne]],["electra",["ElectraForTokenClassification",_e]],["esm",["EsmForTokenClassification",qe]],["convbert",["ConvBertForTokenClassification",le]],["camembert",["CamembertForTokenClassification",be]],["deberta",["DebertaForTokenClassification",Ce]],["deberta-v2",["DebertaV2ForTokenClassification",ze]],["mpnet",["MPNetForTokenClassification",et]],["distilbert",["DistilBertForTokenClassification",Oe]],["roberta",["RobertaForTokenClassification",Vt]],["xlm",["XLMForTokenClassification",Wt]],["xlm-roberta",["XLMRobertaForTokenClassification",Kt]]]),Vi=new Map([["t5",["T5ForConditionalGeneration",_t]],["longt5",["LongT5ForConditionalGeneration",gt]],["mt5",["MT5ForConditionalGeneration",bt]],["bart",["BartForConditionalGeneration",kt]],["mbart",["MBartForConditionalGeneration",Ft]],["marian",["MarianMTModel",Zr]],["m2m_100",["M2M100ForConditionalGeneration",no]],["blenderbot",["BlenderbotForConditionalGeneration",zt]],["blenderbot-small",["BlenderbotSmallForConditionalGeneration",Bt]]]),Ri=new Map([["bloom",["BloomForCausalLM",Ms]],["gpt2",["GPT2LMHeadModel",Pn]],["jais",["JAISLMHeadModel",En]],["gptj",["GPTJForCausalLM",Vn]],["gpt_bigcode",["GPTBigCodeForCausalLM",$n]],["gpt_neo",["GPTNeoForCausalLM",In]],["gpt_neox",["GPTNeoXForCausalLM",On]],["codegen",["CodeGenForCausalLM",Wn]],["llama",["LlamaForCausalLM",Hn]],["cohere",["CohereForCausalLM",Kn]],["gemma",["GemmaForCausalLM",ts]],["gemma2",["Gemma2ForCausalLM",rs]],["openelm",["OpenELMForCausalLM",as]],["qwen2",["Qwen2ForCausalLM",ds]],["phi",["PhiForCausalLM",hs]],["phi3",["Phi3ForCausalLM",fs]],["mpt",["MptForCausalLM",xs]],["opt",["OPTForCausalLM",vs]],["mbart",["MBartForCausalLM",St]],["mistral",["MistralForCausalLM",Wo]],["starcoder2",["Starcoder2ForCausalLM",Ho]],["falcon",["FalconForCausalLM",Ko]],["trocr",["TrOCRForCausalLM",$o]],["stablelm",["StableLmForCausalLM",ui]]]),Gi=new Map([["bert",["BertForMaskedLM",W]],["roformer",["RoFormerForMaskedLM",ee]],["electra",["ElectraForMaskedLM",pe]],["esm",["EsmForMaskedLM",Ge]],["convbert",["ConvBertForMaskedLM",ie]],["camembert",["CamembertForMaskedLM",we]],["deberta",["DebertaForMaskedLM",Te]],["deberta-v2",["DebertaV2ForMaskedLM",Ae]],["mpnet",["MPNetForMaskedLM",Ke]],["albert",["AlbertForMaskedLM",ut]],["distilbert",["DistilBertForMaskedLM",De]],["roberta",["RobertaForMaskedLM",jt]],["xlm",["XLMWithLMHeadModel",qt]],["xlm-roberta",["XLMRobertaForMaskedLM",Yt]],["mobilebert",["MobileBertForMaskedLM",Xe]],["squeezebert",["SqueezeBertForMaskedLM",rt]]]),$i=new Map([["bert",["BertForQuestionAnswering",H]],["roformer",["RoFormerForQuestionAnswering",se]],["electra",["ElectraForQuestionAnswering",me]],["convbert",["ConvBertForQuestionAnswering",ce]],["camembert",["CamembertForQuestionAnswering",ye]],["deberta",["DebertaForQuestionAnswering",Fe]],["deberta-v2",["DebertaV2ForQuestionAnswering",Le]],["mpnet",["MPNetForQuestionAnswering",tt]],["albert",["AlbertForQuestionAnswering",dt]],["distilbert",["DistilBertForQuestionAnswering",je]],["roberta",["RobertaForQuestionAnswering",Rt]],["xlm",["XLMForQuestionAnswering",Xt]],["xlm-roberta",["XLMRobertaForQuestionAnswering",Zt]],["mobilebert",["MobileBertForQuestionAnswering",He]],["squeezebert",["SqueezeBertForQuestionAnswering",it]]]),qi=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",an]]]),Ui=new Map([["llava",["LlavaForConditionalGeneration",cn]],["moondream1",["Moondream1ForConditionalGeneration",dn]],["florence2",["Florence2ForConditionalGeneration",pn]]]),Wi=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",an]]]),Xi=new Map([["vit",["ViTForImageClassification",Ps]],["fastvit",["FastViTForImageClassification",Es]],["mobilevit",["MobileViTForImageClassification",Ns]],["mobilevitv2",["MobileViTV2ForImageClassification",Ds]],["beit",["BeitForImageClassification",Qs]],["deit",["DeiTForImageClassification",ur]],["convnext",["ConvNextForImageClassification",Or]],["convnextv2",["ConvNextV2ForImageClassification",Vr]],["dinov2",["Dinov2ForImageClassification",$r]],["resnet",["ResNetForImageClassification",_r]],["swin",["SwinForImageClassification",gr]],["segformer",["SegformerForImageClassification",ai]],["efficientnet",["EfficientNetForImageClassification",_i]],["mobilenet_v1",["MobileNetV1ForImageClassification",yi]],["mobilenet_v2",["MobileNetV2ForImageClassification",Ti]],["mobilenet_v3",["MobileNetV3ForImageClassification",Fi]],["mobilenet_v4",["MobileNetV4ForImageClassification",Ai]]]),Qi=new Map([["detr",["DetrForObjectDetection",Js]],["rt_detr",["RTDetrForObjectDetection",sr]],["table-transformer",["TableTransformerForObjectDetection",ar]],["yolos",["YolosForObjectDetection",Wr]]]),Hi=new Map([["owlvit",["OwlViTForObjectDetection",Gs]],["owlv2",["Owlv2ForObjectDetection",Us]]]),Yi=new Map([["detr",["DetrForSegmentation",Ks]],["clipseg",["CLIPSegForImageSegmentation",vn]]]),Ji=new Map([["segformer",["SegformerForSemanticSegmentation",li]],["sapiens",["SapiensForSemanticSegmentation",Fr]]]),Ki=new Map([["sam",["SamModel",Hr]]]),Zi=new Map([["wav2vec2",["Wav2Vec2ForCTC",oo]],["wav2vec2-bert",["Wav2Vec2BertForCTC",vo]],["unispeech",["UniSpeechForCTC",fo]],["unispeech-sat",["UniSpeechSatForCTC",bo]],["wavlm",["WavLMForCTC",Lo]],["hubert",["HubertForCTC",So]]]),ea=new Map([["wav2vec2",["Wav2Vec2ForSequenceClassification",io]],["wav2vec2-bert",["Wav2Vec2BertForSequenceClassification",Co]],["unispeech",["UniSpeechForSequenceClassification",go]],["unispeech-sat",["UniSpeechSatForSequenceClassification",yo]],["wavlm",["WavLMForSequenceClassification",Io]],["hubert",["HubertForSequenceClassification",Ao]],["audio-spectrogram-transformer",["ASTForAudioClassification",nn]]]),ta=new Map([["wavlm",["WavLMForXVector",Bo]]]),na=new Map([["unispeech-sat",["UniSpeechSatForAudioFrameClassification",xo]],["wavlm",["WavLMForAudioFrameClassification",No]],["wav2vec2",["Wav2Vec2ForAudioFrameClassification",ao]],["pyannote",["PyAnnoteForAudioFrameClassification",uo]]]),sa=new Map([["vitmatte",["VitMatteForImageMatting",Ls]]]),ra=new Map([["swin2sr",["Swin2SRForImageSuperResolution",br]]]),oa=new Map([["dpt",["DPTForDepthEstimation",kr]],["depth_anything",["DepthAnythingForDepthEstimation",vr]],["glpn",["GLPNForDepthEstimation",zr]],["sapiens",["SapiensForDepthEstimation",Pr]]]),ia=new Map([["sapiens",["SapiensForNormalEstimation",Sr]]]),aa=new Map([["clip",["CLIPVisionModelWithProjection",fn]],["siglip",["SiglipVisionModel",bn]]]),la=[[zi,w],[Li,M],[Ii,x],[ji,w],[Di,w],[Vi,b],[Bi,b],[Ri,x],[Gi,w],[$i,w],[qi,y],[Ui,T],[Xi,w],[Yi,w],[Ji,w],[sa,w],[ra,w],[oa,w],[ia,w],[Qi,w],[Hi,w],[Ki,k],[Zi,w],[ea,w],[Ni,b],[Oi,w],[ta,w],[na,w],[aa,w]];for(const[e,t]of la)for(const[n,s]of e.values())C.set(n,t),P.set(s,n),F.set(n,s);const ca=[["MusicgenForConditionalGeneration",wi,v],["CLIPTextModelWithProjection",mn,w],["SiglipTextModel",Mn,w],["ClapTextModelWithProjection",ti,w],["ClapAudioModelWithProjection",ni,w]];for(const[e,t,n]of ca)C.set(e,n),P.set(t,e),F.set(e,t);class da extends Ei{static MODEL_CLASS_MAPPINGS=la.map((e=>e[0]));static BASE_IF_FAIL=!0}class ua extends Ei{static MODEL_CLASS_MAPPINGS=[ji]}class pa extends Ei{static MODEL_CLASS_MAPPINGS=[Di]}class ha extends Ei{static MODEL_CLASS_MAPPINGS=[Vi]}class _a extends Ei{static MODEL_CLASS_MAPPINGS=[Bi]}class ma extends Ei{static MODEL_CLASS_MAPPINGS=[Ni]}class fa extends Ei{static MODEL_CLASS_MAPPINGS=[Oi]}class ga extends Ei{static MODEL_CLASS_MAPPINGS=[Ri]}class wa extends Ei{static MODEL_CLASS_MAPPINGS=[Gi]}class Ma extends Ei{static MODEL_CLASS_MAPPINGS=[$i]}class ba extends Ei{static MODEL_CLASS_MAPPINGS=[qi]}class ya extends Ei{static MODEL_CLASS_MAPPINGS=[Xi]}class xa extends Ei{static MODEL_CLASS_MAPPINGS=[Yi]}class ka extends Ei{static MODEL_CLASS_MAPPINGS=[Ji]}class Ta extends Ei{static MODEL_CLASS_MAPPINGS=[Qi]}class va extends Ei{static MODEL_CLASS_MAPPINGS=[Hi]}class Ca extends Ei{static MODEL_CLASS_MAPPINGS=[Ki]}class Fa extends Ei{static MODEL_CLASS_MAPPINGS=[Zi]}class Pa extends Ei{static MODEL_CLASS_MAPPINGS=[ea]}class Sa extends Ei{static MODEL_CLASS_MAPPINGS=[ta]}class Aa extends Ei{static MODEL_CLASS_MAPPINGS=[na]}class Ea extends Ei{static MODEL_CLASS_MAPPINGS=[Wi]}class za extends Ei{static MODEL_CLASS_MAPPINGS=[sa]}class La extends Ei{static MODEL_CLASS_MAPPINGS=[ra]}class Ia extends Ei{static MODEL_CLASS_MAPPINGS=[oa]}class Ba extends Ei{static MODEL_CLASS_MAPPINGS=[ia]}class Na extends Ei{static MODEL_CLASS_MAPPINGS=[aa]}class Oa extends G{constructor({logits:e,past_key_values:t,encoder_outputs:n,decoder_attentions:s=null,cross_attentions:r=null}){super(),this.logits=e,this.past_key_values=t,this.encoder_outputs=n,this.decoder_attentions=s,this.cross_attentions=r}}class ja extends G{constructor({logits:e}){super(),this.logits=e}}class Da extends G{constructor({logits:e,embeddings:t}){super(),this.logits=e,this.embeddings=t}}class Va extends G{constructor({logits:e}){super(),this.logits=e}}class Ra extends G{constructor({logits:e}){super(),this.logits=e}}class Ga extends G{constructor({start_logits:e,end_logits:t}){super(),this.start_logits=e,this.end_logits=t}}class $a extends G{constructor({logits:e}){super(),this.logits=e}}class qa extends G{constructor({logits:e,past_key_values:t}){super(),this.logits=e,this.past_key_values=t}}class Ua extends G{constructor({alphas:e}){super(),this.alphas=e}}class Wa extends G{constructor({waveform:e,spectrogram:t}){super(),this.waveform=e,this.spectrogram=t}}},"./src/models/whisper/common_whisper.js":
+  \***********************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{ASTForAudioClassification:()=>nn,ASTModel:()=>tn,ASTPreTrainedModel:()=>en,AlbertForMaskedLM:()=>ut,AlbertForQuestionAnswering:()=>dt,AlbertForSequenceClassification:()=>ct,AlbertModel:()=>lt,AlbertPreTrainedModel:()=>at,AutoModel:()=>ha,AutoModelForAudioClassification:()=>Ea,AutoModelForAudioFrameClassification:()=>La,AutoModelForCTC:()=>Aa,AutoModelForCausalLM:()=>ba,AutoModelForDepthEstimation:()=>Oa,AutoModelForDocumentQuestionAnswering:()=>Ia,AutoModelForImageClassification:()=>Ta,AutoModelForImageFeatureExtraction:()=>Da,AutoModelForImageMatting:()=>Ba,AutoModelForImageSegmentation:()=>va,AutoModelForImageToImage:()=>Na,AutoModelForMaskGeneration:()=>Sa,AutoModelForMaskedLM:()=>ya,AutoModelForNormalEstimation:()=>ja,AutoModelForObjectDetection:()=>Fa,AutoModelForQuestionAnswering:()=>xa,AutoModelForSemanticSegmentation:()=>Ca,AutoModelForSeq2SeqLM:()=>fa,AutoModelForSequenceClassification:()=>_a,AutoModelForSpeechSeq2Seq:()=>ga,AutoModelForTextToSpectrogram:()=>wa,AutoModelForTextToWaveform:()=>Ma,AutoModelForTokenClassification:()=>ma,AutoModelForVision2Seq:()=>ka,AutoModelForXVector:()=>za,AutoModelForZeroShotObjectDetection:()=>Pa,BartForConditionalGeneration:()=>kt,BartForSequenceClassification:()=>Tt,BartModel:()=>xt,BartPretrainedModel:()=>yt,BaseModelOutput:()=>$,BeitForImageClassification:()=>Qs,BeitModel:()=>Xs,BeitPreTrainedModel:()=>Ws,BertForMaskedLM:()=>W,BertForQuestionAnswering:()=>H,BertForSequenceClassification:()=>X,BertForTokenClassification:()=>Q,BertModel:()=>U,BertPreTrainedModel:()=>q,BlenderbotForConditionalGeneration:()=>zt,BlenderbotModel:()=>Et,BlenderbotPreTrainedModel:()=>At,BlenderbotSmallForConditionalGeneration:()=>Bt,BlenderbotSmallModel:()=>It,BlenderbotSmallPreTrainedModel:()=>Lt,BloomForCausalLM:()=>Ms,BloomModel:()=>ws,BloomPreTrainedModel:()=>gs,CLIPModel:()=>_n,CLIPPreTrainedModel:()=>hn,CLIPSegForImageSegmentation:()=>vn,CLIPSegModel:()=>Tn,CLIPSegPreTrainedModel:()=>kn,CLIPTextModelWithProjection:()=>mn,CLIPVisionModelWithProjection:()=>fn,CamembertForMaskedLM:()=>we,CamembertForQuestionAnswering:()=>ye,CamembertForSequenceClassification:()=>Me,CamembertForTokenClassification:()=>be,CamembertModel:()=>ge,CamembertPreTrainedModel:()=>fe,CausalLMOutput:()=>Wa,CausalLMOutputWithPast:()=>Xa,ChineseCLIPModel:()=>xn,ChineseCLIPPreTrainedModel:()=>yn,ClapAudioModelWithProjection:()=>oi,ClapModel:()=>si,ClapPreTrainedModel:()=>ni,ClapTextModelWithProjection:()=>ri,CodeGenForCausalLM:()=>Wn,CodeGenModel:()=>Un,CodeGenPreTrainedModel:()=>qn,CohereForCausalLM:()=>Kn,CohereModel:()=>Jn,CoherePreTrainedModel:()=>Yn,ConvBertForMaskedLM:()=>ie,ConvBertForQuestionAnswering:()=>ce,ConvBertForSequenceClassification:()=>ae,ConvBertForTokenClassification:()=>le,ConvBertModel:()=>oe,ConvBertPreTrainedModel:()=>re,ConvNextForImageClassification:()=>Vr,ConvNextModel:()=>Dr,ConvNextPreTrainedModel:()=>jr,ConvNextV2ForImageClassification:()=>$r,ConvNextV2Model:()=>Gr,ConvNextV2PreTrainedModel:()=>Rr,DPTForDepthEstimation:()=>Cr,DPTModel:()=>vr,DPTPreTrainedModel:()=>Tr,DebertaForMaskedLM:()=>Te,DebertaForQuestionAnswering:()=>Fe,DebertaForSequenceClassification:()=>ve,DebertaForTokenClassification:()=>Ce,DebertaModel:()=>ke,DebertaPreTrainedModel:()=>xe,DebertaV2ForMaskedLM:()=>Ae,DebertaV2ForQuestionAnswering:()=>Le,DebertaV2ForSequenceClassification:()=>Ee,DebertaV2ForTokenClassification:()=>ze,DebertaV2Model:()=>Se,DebertaV2PreTrainedModel:()=>Pe,DeiTForImageClassification:()=>ur,DeiTModel:()=>dr,DeiTPreTrainedModel:()=>cr,DepthAnythingForDepthEstimation:()=>Pr,DepthAnythingPreTrainedModel:()=>Fr,DetrForObjectDetection:()=>Js,DetrForSegmentation:()=>Ks,DetrModel:()=>Ys,DetrObjectDetectionOutput:()=>Zs,DetrPreTrainedModel:()=>Hs,DetrSegmentationOutput:()=>er,Dinov2ForImageClassification:()=>Wr,Dinov2Model:()=>Ur,Dinov2PreTrainedModel:()=>qr,DistilBertForMaskedLM:()=>De,DistilBertForQuestionAnswering:()=>je,DistilBertForSequenceClassification:()=>Ne,DistilBertForTokenClassification:()=>Oe,DistilBertModel:()=>Be,DistilBertPreTrainedModel:()=>Ie,DonutSwinModel:()=>Or,DonutSwinPreTrainedModel:()=>Nr,EfficientNetForImageClassification:()=>gi,EfficientNetModel:()=>fi,EfficientNetPreTrainedModel:()=>mi,ElectraForMaskedLM:()=>pe,ElectraForQuestionAnswering:()=>me,ElectraForSequenceClassification:()=>he,ElectraForTokenClassification:()=>_e,ElectraModel:()=>ue,ElectraPreTrainedModel:()=>de,EsmForMaskedLM:()=>Ge,EsmForSequenceClassification:()=>$e,EsmForTokenClassification:()=>qe,EsmModel:()=>Re,EsmPreTrainedModel:()=>Ve,FalconForCausalLM:()=>ti,FalconModel:()=>ei,FalconPreTrainedModel:()=>Zo,FastViTForImageClassification:()=>Es,FastViTModel:()=>As,FastViTPreTrainedModel:()=>Ss,Florence2ForConditionalGeneration:()=>pn,Florence2PreTrainedModel:()=>un,GLPNForDepthEstimation:()=>Br,GLPNModel:()=>Ir,GLPNPreTrainedModel:()=>Lr,GPT2LMHeadModel:()=>Pn,GPT2Model:()=>Fn,GPT2PreTrainedModel:()=>Cn,GPTBigCodeForCausalLM:()=>$n,GPTBigCodeModel:()=>Gn,GPTBigCodePreTrainedModel:()=>Rn,GPTJForCausalLM:()=>Vn,GPTJModel:()=>Dn,GPTJPreTrainedModel:()=>jn,GPTNeoForCausalLM:()=>In,GPTNeoModel:()=>Ln,GPTNeoPreTrainedModel:()=>zn,GPTNeoXForCausalLM:()=>On,GPTNeoXModel:()=>Nn,GPTNeoXPreTrainedModel:()=>Bn,Gemma2ForCausalLM:()=>rs,Gemma2Model:()=>ss,Gemma2PreTrainedModel:()=>ns,GemmaForCausalLM:()=>ts,GemmaModel:()=>es,GemmaPreTrainedModel:()=>Zn,HieraForImageClassification:()=>_r,HieraModel:()=>hr,HieraPreTrainedModel:()=>pr,HubertForCTC:()=>zo,HubertForSequenceClassification:()=>Lo,HubertModel:()=>Eo,HubertPreTrainedModel:()=>Ao,ImageMattingOutput:()=>Qa,JAISLMHeadModel:()=>En,JAISModel:()=>An,JAISPreTrainedModel:()=>Sn,LlamaForCausalLM:()=>Hn,LlamaModel:()=>Qn,LlamaPreTrainedModel:()=>Xn,LlavaForConditionalGeneration:()=>cn,LlavaPreTrainedModel:()=>ln,LongT5ForConditionalGeneration:()=>gt,LongT5Model:()=>ft,LongT5PreTrainedModel:()=>mt,M2M100ForConditionalGeneration:()=>oo,M2M100Model:()=>ro,M2M100PreTrainedModel:()=>so,MBartForCausalLM:()=>St,MBartForConditionalGeneration:()=>Ft,MBartForSequenceClassification:()=>Pt,MBartModel:()=>Ct,MBartPreTrainedModel:()=>vt,MPNetForMaskedLM:()=>Ke,MPNetForQuestionAnswering:()=>tt,MPNetForSequenceClassification:()=>Ze,MPNetForTokenClassification:()=>et,MPNetModel:()=>Je,MPNetPreTrainedModel:()=>Ye,MT5ForConditionalGeneration:()=>bt,MT5Model:()=>Mt,MT5PreTrainedModel:()=>wt,MarianMTModel:()=>no,MarianModel:()=>to,MarianPreTrainedModel:()=>eo,MaskedLMOutput:()=>qa,MistralForCausalLM:()=>Ho,MistralModel:()=>Qo,MistralPreTrainedModel:()=>Xo,MobileBertForMaskedLM:()=>Xe,MobileBertForQuestionAnswering:()=>He,MobileBertForSequenceClassification:()=>Qe,MobileBertModel:()=>We,MobileBertPreTrainedModel:()=>Ue,MobileNetV1ForImageClassification:()=>Ti,MobileNetV1Model:()=>ki,MobileNetV1PreTrainedModel:()=>xi,MobileNetV2ForImageClassification:()=>Fi,MobileNetV2Model:()=>Ci,MobileNetV2PreTrainedModel:()=>vi,MobileNetV3ForImageClassification:()=>Ai,MobileNetV3Model:()=>Si,MobileNetV3PreTrainedModel:()=>Pi,MobileNetV4ForImageClassification:()=>Li,MobileNetV4Model:()=>zi,MobileNetV4PreTrainedModel:()=>Ei,MobileViTForImageClassification:()=>Ns,MobileViTModel:()=>Bs,MobileViTPreTrainedModel:()=>Is,MobileViTV2ForImageClassification:()=>Ds,MobileViTV2Model:()=>js,MobileViTV2PreTrainedModel:()=>Os,ModelOutput:()=>G,Moondream1ForConditionalGeneration:()=>dn,MptForCausalLM:()=>xs,MptModel:()=>ys,MptPreTrainedModel:()=>bs,MusicgenForCausalLM:()=>bi,MusicgenForConditionalGeneration:()=>yi,MusicgenModel:()=>Mi,MusicgenPreTrainedModel:()=>wi,NomicBertModel:()=>J,NomicBertPreTrainedModel:()=>Y,OPTForCausalLM:()=>vs,OPTModel:()=>Ts,OPTPreTrainedModel:()=>ks,OpenELMForCausalLM:()=>as,OpenELMModel:()=>is,OpenELMPreTrainedModel:()=>os,OwlViTForObjectDetection:()=>Gs,OwlViTModel:()=>Rs,OwlViTPreTrainedModel:()=>Vs,Owlv2ForObjectDetection:()=>Us,Owlv2Model:()=>qs,Owlv2PreTrainedModel:()=>$s,Phi3ForCausalLM:()=>fs,Phi3Model:()=>ms,Phi3PreTrainedModel:()=>_s,PhiForCausalLM:()=>hs,PhiModel:()=>ps,PhiPreTrainedModel:()=>us,PreTrainedModel:()=>R,PretrainedMixin:()=>Ii,PyAnnoteForAudioFrameClassification:()=>_o,PyAnnoteModel:()=>ho,PyAnnotePreTrainedModel:()=>po,QuestionAnsweringModelOutput:()=>Ua,Qwen2ForCausalLM:()=>ds,Qwen2Model:()=>cs,Qwen2PreTrainedModel:()=>ls,RTDetrForObjectDetection:()=>sr,RTDetrModel:()=>nr,RTDetrObjectDetectionOutput:()=>rr,RTDetrPreTrainedModel:()=>tr,ResNetForImageClassification:()=>gr,ResNetModel:()=>fr,ResNetPreTrainedModel:()=>mr,RoFormerForMaskedLM:()=>ee,RoFormerForQuestionAnswering:()=>se,RoFormerForSequenceClassification:()=>te,RoFormerForTokenClassification:()=>ne,RoFormerModel:()=>Z,RoFormerPreTrainedModel:()=>K,RobertaForMaskedLM:()=>jt,RobertaForQuestionAnswering:()=>Rt,RobertaForSequenceClassification:()=>Dt,RobertaForTokenClassification:()=>Vt,RobertaModel:()=>Ot,RobertaPreTrainedModel:()=>Nt,SamImageSegmentationOutput:()=>Zr,SamModel:()=>Kr,SamPreTrainedModel:()=>Jr,SapiensForDepthEstimation:()=>Er,SapiensForNormalEstimation:()=>zr,SapiensForSemanticSegmentation:()=>Ar,SapiensPreTrainedModel:()=>Sr,SegformerForImageClassification:()=>di,SegformerForSemanticSegmentation:()=>ui,SegformerModel:()=>ci,SegformerPreTrainedModel:()=>li,Seq2SeqLMOutput:()=>Va,SequenceClassifierOutput:()=>Ra,SiglipModel:()=>wn,SiglipPreTrainedModel:()=>gn,SiglipTextModel:()=>Mn,SiglipVisionModel:()=>bn,SpeechT5ForSpeechToText:()=>Go,SpeechT5ForTextToSpeech:()=>$o,SpeechT5HifiGan:()=>qo,SpeechT5Model:()=>Ro,SpeechT5PreTrainedModel:()=>Vo,SqueezeBertForMaskedLM:()=>rt,SqueezeBertForQuestionAnswering:()=>it,SqueezeBertForSequenceClassification:()=>ot,SqueezeBertModel:()=>st,SqueezeBertPreTrainedModel:()=>nt,StableLmForCausalLM:()=>_i,StableLmModel:()=>hi,StableLmPreTrainedModel:()=>pi,Starcoder2ForCausalLM:()=>Ko,Starcoder2Model:()=>Jo,Starcoder2PreTrainedModel:()=>Yo,Swin2SRForImageSuperResolution:()=>kr,Swin2SRModel:()=>xr,Swin2SRPreTrainedModel:()=>yr,SwinForImageClassification:()=>br,SwinModel:()=>Mr,SwinPreTrainedModel:()=>wr,T5ForConditionalGeneration:()=>_t,T5Model:()=>ht,T5PreTrainedModel:()=>pt,TableTransformerForObjectDetection:()=>ar,TableTransformerModel:()=>ir,TableTransformerObjectDetectionOutput:()=>lr,TableTransformerPreTrainedModel:()=>or,TokenClassifierOutput:()=>$a,TrOCRForCausalLM:()=>Wo,TrOCRPreTrainedModel:()=>Uo,UniSpeechForCTC:()=>Mo,UniSpeechForSequenceClassification:()=>bo,UniSpeechModel:()=>wo,UniSpeechPreTrainedModel:()=>go,UniSpeechSatForAudioFrameClassification:()=>vo,UniSpeechSatForCTC:()=>ko,UniSpeechSatForSequenceClassification:()=>To,UniSpeechSatModel:()=>xo,UniSpeechSatPreTrainedModel:()=>yo,ViTForImageClassification:()=>Ps,ViTModel:()=>Fs,ViTPreTrainedModel:()=>Cs,VisionEncoderDecoderModel:()=>an,VitMatteForImageMatting:()=>Ls,VitMattePreTrainedModel:()=>zs,VitsModel:()=>ai,VitsModelOutput:()=>Ha,VitsPreTrainedModel:()=>ii,Wav2Vec2BertForCTC:()=>Po,Wav2Vec2BertForSequenceClassification:()=>So,Wav2Vec2BertModel:()=>Fo,Wav2Vec2BertPreTrainedModel:()=>Co,Wav2Vec2ForAudioFrameClassification:()=>uo,Wav2Vec2ForCTC:()=>lo,Wav2Vec2ForSequenceClassification:()=>co,Wav2Vec2Model:()=>ao,Wav2Vec2PreTrainedModel:()=>io,WavLMForAudioFrameClassification:()=>Do,WavLMForCTC:()=>No,WavLMForSequenceClassification:()=>Oo,WavLMForXVector:()=>jo,WavLMModel:()=>Bo,WavLMPreTrainedModel:()=>Io,WeSpeakerResNetModel:()=>fo,WeSpeakerResNetPreTrainedModel:()=>mo,WhisperForConditionalGeneration:()=>on,WhisperModel:()=>rn,WhisperPreTrainedModel:()=>sn,XLMForQuestionAnswering:()=>Xt,XLMForSequenceClassification:()=>Ut,XLMForTokenClassification:()=>Wt,XLMModel:()=>$t,XLMPreTrainedModel:()=>Gt,XLMRobertaForMaskedLM:()=>Yt,XLMRobertaForQuestionAnswering:()=>Zt,XLMRobertaForSequenceClassification:()=>Jt,XLMRobertaForTokenClassification:()=>Kt,XLMRobertaModel:()=>Ht,XLMRobertaPreTrainedModel:()=>Qt,XLMWithLMHeadModel:()=>qt,XVectorOutput:()=>Ga,YolosForObjectDetection:()=>Hr,YolosModel:()=>Qr,YolosObjectDetectionOutput:()=>Yr,YolosPreTrainedModel:()=>Xr});var s=n(/*! ./configs.js */"./src/configs.js"),r=n(/*! ./backends/onnx.js */"./src/backends/onnx.js"),o=n(/*! ./utils/dtypes.js */"./src/utils/dtypes.js"),i=n(/*! ./utils/generic.js */"./src/utils/generic.js"),a=n(/*! ./utils/core.js */"./src/utils/core.js"),l=n(/*! ./utils/hub.js */"./src/utils/hub.js"),c=n(/*! ./generation/logits_process.js */"./src/generation/logits_process.js"),d=n(/*! ./generation/configuration_utils.js */"./src/generation/configuration_utils.js"),u=n(/*! ./utils/tensor.js */"./src/utils/tensor.js"),p=n(/*! ./utils/maths.js */"./src/utils/maths.js"),h=n(/*! ./generation/stopping_criteria.js */"./src/generation/stopping_criteria.js"),_=n(/*! ./generation/logits_sampler.js */"./src/generation/logits_sampler.js"),m=n(/*! ./env.js */"./src/env.js"),f=n(/*! ./models/whisper/generation_whisper.js */"./src/models/whisper/generation_whisper.js"),g=n(/*! ./models/whisper/common_whisper.js */"./src/models/whisper/common_whisper.js");const w=0,M=1,b=2,y=3,x=4,k=5,T=6,v=7,C=new Map,F=new Map,P=new Map;async function S(e,t,n){return Object.fromEntries(await Promise.all(Object.keys(t).map((async i=>{const{buffer:a,session_options:c}=await async function(e,t,n){let i=n.device;i&&"string"!=typeof i&&(i.hasOwnProperty(t)?i=i[t]:(console.warn(`device not specified for "${t}". Using the default device.`),i=null));const a=i??(m.apis.IS_NODE_ENV?"cpu":"wasm"),c=(0,r.deviceToExecutionProviders)(a);let d=n.dtype;"string"!=typeof d&&(d&&d.hasOwnProperty(t)?d=d[t]:(d=o.DEFAULT_DEVICE_DTYPE_MAPPING[a]??o.DATA_TYPES.fp32,console.warn(`dtype not specified for "${t}". Using the default dtype (${d}) for this device (${a}).`)));const u=d;if(!o.DEFAULT_DTYPE_SUFFIX_MAPPING.hasOwnProperty(u))throw new Error(`Invalid dtype: ${u}. Should be one of: ${Object.keys(o.DATA_TYPES).join(", ")}`);if(u===o.DATA_TYPES.fp16&&"webgpu"===a&&!await(0,o.isWebGpuFp16Supported)())throw new Error(`The device (${a}) does not support fp16.`);const p=o.DEFAULT_DTYPE_SUFFIX_MAPPING[u],h=`${n.subfolder??""}/${t}${p}.onnx`,_={...n.session_options}??{};_.executionProviders??=c;const f=(0,l.getModelFile)(e,h,!0,n);let g=[];if(n.use_external_data_format&&(!0===n.use_external_data_format||"object"==typeof n.use_external_data_format&&n.use_external_data_format.hasOwnProperty(t)&&!0===n.use_external_data_format[t])){if(m.apis.IS_NODE_ENV)throw new Error("External data format is not yet supported in Node.js");const s=`${t}${p}.onnx_data`,r=`${n.subfolder??""}/${s}`;g.push(new Promise((async(t,o)=>{const i=await(0,l.getModelFile)(e,r,!0,n);t({path:s,data:i})})))}else void 0!==_.externalData&&(g=_.externalData.map((async t=>{if("string"==typeof t.data){const s=await(0,l.getModelFile)(e,t.data,!0,n);return{...t,data:s}}return t})));if(g.length>0&&(_.externalData=await Promise.all(g)),"webgpu"===a){const e=(0,s.getKeyValueShapes)(n.config,{prefix:"present"});if(Object.keys(e).length>0&&!(0,r.isONNXProxy)()){const t={};for(const n in e)t[n]="gpu-buffer";_.preferredOutputLocation=t}}return{buffer:await f,session_options:_}}(e,t[i],n);return[i,await(0,r.createInferenceSession)(a,c)]}))))}async function A(e,t){const n=function(e,t){const n=Object.create(null),s=[];for(const o of e.inputNames){const e=t[o];e instanceof u.Tensor?n[o]=(0,r.isONNXProxy)()?e.clone():e:s.push(o)}if(s.length>0)throw new Error(`An error occurred during model execution: "Missing the following inputs: ${s.join(", ")}.`);const o=Object.keys(t).length,i=e.inputNames.length;if(o>i){let n=Object.keys(t).filter((t=>!e.inputNames.includes(t)));console.warn(`WARNING: Too many inputs were provided (${o} > ${i}). The following inputs will be ignored: "${n.join(", ")}".`)}return n}(e,t);try{const t=Object.fromEntries(Object.entries(n).map((([e,t])=>[e,t.ort_tensor])));let s=await e.run(t);return s=E(s),s}catch(e){throw console.error(`An error occurred during model execution: "${e}".`),console.error("Inputs given to model:",n),e}}function E(e){for(let t in e)(0,r.isONNXTensor)(e[t])?e[t]=new u.Tensor(e[t]):"object"==typeof e[t]&&E(e[t]);return e}function z(e){if(e instanceof u.Tensor)return e;if(0===e.length)throw Error("items must be non-empty");if(Array.isArray(e[0])){if(e.some((t=>t.length!==e[0].length)))throw Error("Unable to create tensor, you should probably activate truncation and/or padding with 'padding=True' and/or 'truncation=True' to have batched tensors with the same length.");return new u.Tensor("int64",BigInt64Array.from(e.flat().map((e=>BigInt(e)))),[e.length,e[0].length])}return new u.Tensor("int64",BigInt64Array.from(e.map((e=>BigInt(e)))),[1,e.length])}function L(e){return new u.Tensor("bool",[e],[1])}async function I(e,t){let{encoder_outputs:n,input_ids:s,decoder_input_ids:r,...o}=t;if(!n){const s=(0,a.pick)(t,e.sessions.model.inputNames);n=(await B(e,s)).last_hidden_state}o.input_ids=r,o.encoder_hidden_states=n,e.sessions.decoder_model_merged.inputNames.includes("encoder_attention_mask")&&(o.encoder_attention_mask=t.attention_mask);return await N(e,o,!0)}async function B(e,t){const n=e.sessions.model,s=(0,a.pick)(t,n.inputNames);if(n.inputNames.includes("inputs_embeds")&&!s.inputs_embeds){if(!t.input_ids)throw new Error("Both `input_ids` and `inputs_embeds` are missing in the model inputs.");s.inputs_embeds=await e.encode_text({input_ids:t.input_ids})}return n.inputNames.includes("token_type_ids")&&!s.token_type_ids&&(s.token_type_ids=new u.Tensor("int64",new BigInt64Array(s.input_ids.data.length),s.input_ids.dims)),await A(n,s)}async function N(e,t,n=!1){const s=e.sessions[n?"decoder_model_merged":"model"],{past_key_values:r,...o}=t;s.inputNames.includes("use_cache_branch")&&(o.use_cache_branch=L(!!r)),s.inputNames.includes("position_ids")&&o.attention_mask&&!o.position_ids&&(o.position_ids=function(e,t=null){const{input_ids:n,inputs_embeds:s,attention_mask:r}=e,[o,i]=r.dims,a=new BigInt64Array(r.data.length);for(let e=0;e<o;++e){const t=e*i;let n=BigInt(0);for(let e=0;e<i;++e){const s=t+e;0n===r.data[s]?a[s]=BigInt(1):(a[s]=n,n+=r.data[s])}}let l=new u.Tensor("int64",a,r.dims);if(t){const e=-(n??s).dims.at(1);l=l.slice(null,[e,null])}return l}(o,r)),e.addPastKeyValues(o,r);const i=(0,a.pick)(o,s.inputNames);return await A(s,i)}async function O(e,{input_ids:t=null,attention_mask:n=null,pixel_values:s=null,position_ids:r=null,inputs_embeds:o=null,past_key_values:i=null,generation_config:a=null,logits_processor:l=null,...c}){if(!o)if(o=await e.encode_text({input_ids:t}),s&&1!==t.dims[1]){const r=await e.encode_image({pixel_values:s});({inputs_embeds:o,attention_mask:n}=e._merge_input_ids_with_image_features({image_features:r,inputs_embeds:o,input_ids:t,attention_mask:n}))}else if(i&&s&&1===t.dims[1]){const e=t.dims[1],s=Object.values(i)[0].dims.at(-2);n=(0,u.cat)([(0,u.ones)([t.dims[0],s]),n.slice(null,[n.dims[1]-e,n.dims[1]])],1)}return await N(e,{inputs_embeds:o,past_key_values:i,attention_mask:n,position_ids:r,generation_config:a,logits_processor:l},!0)}function j(e,t,n,s){if(n.past_key_values){const t=Object.values(n.past_key_values)[0].dims.at(-2),{input_ids:s,attention_mask:r}=n;if(r&&r.dims[1]>s.dims[1]);else if(t<s.dims[1])n.input_ids=s.slice(null,[t,null]);else if(null!=e.config.image_token_index&&s.data.some((t=>t==e.config.image_token_index))){const r=e.config.num_image_tokens;if(!r)throw new Error("`num_image_tokens` is missing in the model configuration.");const o=s.dims[1]-(t-r);n.input_ids=s.slice(null,[-o,null]),n.attention_mask=(0,u.ones)([1,t+o])}}return n}function D(e,t,n,s){return n.past_key_values&&(t=t.map((e=>[e.at(-1)]))),{...n,decoder_input_ids:z(t)}}function V(e,...t){return e.config.is_encoder_decoder?D(e,...t):j(e,...t)}class R extends i.Callable{main_input_name="input_ids";forward_params=["input_ids","attention_mask"];constructor(e,t){super(),this.config=e,this.sessions=t;const n=P.get(this.constructor),s=C.get(n);switch(this.can_generate=!1,this._forward=null,this._prepare_inputs_for_generation=null,s){case x:this.can_generate=!0,this._forward=N,this._prepare_inputs_for_generation=j;break;case b:case y:case v:this.can_generate=!0,this._forward=I,this._prepare_inputs_for_generation=D;break;case M:this._forward=I;break;case T:this.can_generate=!0,this._forward=O,this._prepare_inputs_for_generation=V;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:n=null,cache_dir:r=null,local_files_only:o=!1,revision:i="main",model_file_name:a=null,subfolder:c="onnx",device:d=null,dtype:u=null,use_external_data_format:p=null,session_options:h={}}={}){let _={progress_callback:t,config:n,cache_dir:r,local_files_only:o,revision:i,model_file_name:a,subfolder:c,device:d,dtype:u,use_external_data_format:p,session_options:h};const m=P.get(this),f=C.get(m);let g;if(n=_.config=await s.AutoConfig.from_pretrained(e,_),f===x)g=await Promise.all([S(e,{model:_.model_file_name??"model"},_),(0,l.getModelJSON)(e,"generation_config.json",!1,_)]);else if(f===b||f===y)g=await Promise.all([S(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},_),(0,l.getModelJSON)(e,"generation_config.json",!1,_)]);else if(f===k)g=await Promise.all([S(e,{model:"vision_encoder",prompt_encoder_mask_decoder:"prompt_encoder_mask_decoder"},_)]);else if(f===M)g=await Promise.all([S(e,{model:"encoder_model",decoder_model_merged:"decoder_model_merged"},_)]);else if(f===T){const t={embed_tokens:"embed_tokens",vision_encoder:"vision_encoder",decoder_model_merged:"decoder_model_merged"};n.is_encoder_decoder&&(t.model="encoder_model"),g=await Promise.all([S(e,t,_),(0,l.getModelJSON)(e,"generation_config.json",!1,_)])}else f===v?g=await Promise.all([S(e,{model:"text_encoder",decoder_model_merged:"decoder_model_merged",encodec_decode:"encodec_decode"},_),(0,l.getModelJSON)(e,"generation_config.json",!1,_)]):(f!==w&&console.warn(`Model type for '${m??n?.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:_.model_file_name??"model"},_)]));return new this(n,...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,n=null){const s=new c.LogitsProcessorList;if(null!==e.repetition_penalty&&1!==e.repetition_penalty&&s.push(new c.RepetitionPenaltyLogitsProcessor(e.repetition_penalty)),null!==e.no_repeat_ngram_size&&e.no_repeat_ngram_size>0&&s.push(new c.NoRepeatNGramLogitsProcessor(e.no_repeat_ngram_size)),null!==e.bad_words_ids&&s.push(new c.NoBadWordsLogitsProcessor(e.bad_words_ids,e.eos_token_id)),null!==e.min_length&&null!==e.eos_token_id&&e.min_length>0&&s.push(new c.MinLengthLogitsProcessor(e.min_length,e.eos_token_id)),null!==e.min_new_tokens&&null!==e.eos_token_id&&e.min_new_tokens>0&&s.push(new c.MinNewTokensLengthLogitsProcessor(t,e.min_new_tokens,e.eos_token_id)),null!==e.forced_bos_token_id&&s.push(new c.ForcedBOSTokenLogitsProcessor(e.forced_bos_token_id)),null!==e.forced_eos_token_id&&s.push(new c.ForcedEOSTokenLogitsProcessor(e.max_length,e.forced_eos_token_id)),null!==e.begin_suppress_tokens){const n=t>1||null===e.forced_bos_token_id?t:t+1;s.push(new c.SuppressTokensAtBeginLogitsProcessor(e.begin_suppress_tokens,n))}return null!==e.guidance_scale&&e.guidance_scale>1&&s.push(new c.ClassifierFreeGuidanceLogitsProcessor(e.guidance_scale)),null!==n&&s.extend(n),s}_prepare_generation_config(e,t,n=d.GenerationConfig){const s={...this.config};for(const e of["decoder","generator","text_config"])e in s&&Object.assign(s,s[e]);const r=new n(s);return"generation_config"in this&&Object.assign(r,this.generation_config),e&&Object.assign(r,e),t&&Object.assign(r,(0,a.pick)(t,Object.getOwnPropertyNames(r))),r}_get_stopping_criteria(e,t=null){const n=new h.StoppingCriteriaList;return null!==e.max_length&&n.push(new h.MaxLengthCriteria(e.max_length,this.config.max_position_embeddings??null)),null!==e.eos_token_id&&n.push(new h.EosTokenCriteria(e.eos_token_id)),t&&n.extend(t),n}_validate_model_class(){if(!this.can_generate){const e=[qi,Xi,$i,ji],t=P.get(this.constructor),n=new Set,s=this.config.model_type;for(const t of e){const e=t.get(s);e&&n.add(e[0])}let r=`The current model class (${t}) is not compatible with \`.generate()\`, as it doesn't have a language model head.`;throw n.size>0&&(r+=` Please use the following class instead: ${[...n].join(", ")}`),Error(r)}}prepare_inputs_for_generation(...e){return this._prepare_inputs_for_generation(this,...e)}_update_model_kwargs_for_generation({generated_input_ids:e,outputs:t,model_inputs:n,is_encoder_decoder:s}){return n.past_key_values=this.getPastKeyValues(t,n.past_key_values),n.input_ids=new u.Tensor("int64",e.flat(),[e.length,1]),s||(n.attention_mask=(0,u.cat)([n.attention_mask,(0,u.ones)([n.attention_mask.dims[0],1])],1)),n.position_ids=null,n}_prepare_model_inputs({inputs:e,bos_token_id:t,model_kwargs:n}){const s=(0,a.pick)(n,this.forward_params),r=this.main_input_name;if(r in s){if(e)throw new Error("`inputs`: {inputs}` were passed alongside {input_name} which is not allowed. Make sure to either pass {inputs} or {input_name}=...")}else s[r]=e;return{inputs_tensor:s[r],model_inputs:s,model_input_name:r}}async _prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:e,model_inputs:t,model_input_name:n,generation_config:s}){if(this.sessions.model.inputNames.includes("inputs_embeds")&&!t.inputs_embeds&&"_prepare_inputs_embeds"in this){const{input_ids:e,pixel_values:n,attention_mask:s,...r}=t,o=await this._prepare_inputs_embeds(t);t={...r,...(0,a.pick)(o,["inputs_embeds","attention_mask"])}}let{last_hidden_state:r}=await B(this,t);if(null!==s.guidance_scale&&s.guidance_scale>1)r=(0,u.cat)([r,(0,u.full_like)(r,0)],0),"attention_mask"in t&&(t.attention_mask=(0,u.cat)([t.attention_mask,(0,u.zeros_like)(t.attention_mask)],0));else if(t.decoder_input_ids){const e=z(t.decoder_input_ids).dims[0];if(e!==r.dims[0]){if(1!==r.dims[0])throw new Error(`The encoder outputs have a different batch size (${r.dims[0]}) than the decoder inputs (${e}).`);r=(0,u.cat)(Array.from({length:e},(()=>r)),0)}}return t.encoder_outputs=r,t}_prepare_decoder_input_ids_for_generation({batch_size:e,model_input_name:t,model_kwargs:n,decoder_start_token_id:s,bos_token_id:r,generation_config:o}){let{decoder_input_ids:i,...a}=n;if(i)Array.isArray(i[0])||(i=Array.from({length:e},(()=>i)));else if(s??=r,"musicgen"===this.config.model_type)i=Array.from({length:e*this.config.decoder.num_codebooks},(()=>[s]));else if(Array.isArray(s)){if(s.length!==e)throw new Error(`\`decoder_start_token_id\` expcted to have length ${e} but got ${s.length}`);i=s}else i=Array.from({length:e},(()=>[s]));return i=z(i),n.decoder_attention_mask=(0,u.ones_like)(i),{input_ids:i,model_inputs:a}}async generate({inputs:e=null,generation_config:t=null,logits_processor:n=null,stopping_criteria:s=null,streamer:r=null,...o}){this._validate_model_class(),t=this._prepare_generation_config(t,o);let{inputs_tensor:i,model_inputs:a,model_input_name:l}=this._prepare_model_inputs({inputs:e,model_kwargs:o});const c=this.config.is_encoder_decoder;let d;c&&("encoder_outputs"in a||(a=await this._prepare_encoder_decoder_kwargs_for_generation({inputs_tensor:i,model_inputs:a,model_input_name:l,generation_config:t}))),c?({input_ids:d,model_inputs:a}=this._prepare_decoder_input_ids_for_generation({batch_size:a[l].dims.at(0),model_input_name:l,model_kwargs:a,decoder_start_token_id:t.decoder_start_token_id,bos_token_id:t.bos_token_id,generation_config:t})):d=a[l];let p=d.dims.at(-1);null!==t.max_new_tokens&&(t.max_length=p+t.max_new_tokens);const h=this._get_logits_processor(t,p,n),m=this._get_stopping_criteria(t,s),f=a[l].dims.at(0),g=_.LogitsSampler.getSampler(t),w=new Array(f).fill(0),M=d.tolist();r&&r.put(M);let b=null,y={};for(;;){a=this.prepare_inputs_for_generation(M,a,t);const e=await this.forward(a);if(t.output_attentions&&t.return_dict_in_generate){const t=this.getAttentions(e);for(const e in t)e in y||(y[e]=[]),y[e].push(t[e])}const n=h(M,e.logits.slice(null,-1,null)),s=[];for(let e=0;e<n.dims.at(0);++e){const t=n[e],r=await g(t);for(const[t,n]of r){const r=BigInt(t);w[e]+=n,M[e].push(r),s.push([r]);break}}r&&r.put(s);if(m(M).every((e=>e))){t.return_dict_in_generate&&(b=this.getPastKeyValues(e,a.past_key_values,!1));break}a=this._update_model_kwargs_for_generation({generated_input_ids:s,outputs:e,model_inputs:a,is_encoder_decoder:c})}r&&r.end();const x=new u.Tensor("int64",M.flat(),[M.length,M[0].length]);return t.return_dict_in_generate?{sequences:x,past_key_values:b,...y}:x}getPastKeyValues(e,t,n=!0){const s=Object.create(null);for(const r in e)if(r.startsWith("present")){const o=r.replace("present","past_key_values");if(t&&r.includes("encoder"))s[o]=t[o];else{if(n&&t){const e=t[o];"gpu-buffer"===e.location&&e.dispose()}s[o]=e[r]}}return s}getAttentions(e){const t={};for(const n of["cross_attentions","encoder_attentions","decoder_attentions"])for(const s in e)s.startsWith(n)&&(n in t||(t[n]=[]),t[n].push(e[s]));return t}addPastKeyValues(e,t){if(t)Object.assign(e,t);else{const t=this.custom_config.kv_cache_dtype??"float32",n="float16"===t?new Uint16Array:[],r=(0,s.getKeyValueShapes)(this.config);for(const s in r)e[s]=new u.Tensor(t,n,r[s])}}async encode_image({pixel_values:e}){const t=(await 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 $ extends G{constructor({last_hidden_state:e,hidden_states:t=null,attentions:n=null}){super(),this.last_hidden_state=e,this.hidden_states=t,this.attentions=n}}class q extends R{}class U extends q{}class W extends q{async _call(e){return new qa(await super._call(e))}}class X extends q{async _call(e){return new Ra(await super._call(e))}}class Q extends q{async _call(e){return new $a(await super._call(e))}}class H extends q{async _call(e){return new Ua(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 qa(await super._call(e))}}class te extends K{async _call(e){return new Ra(await super._call(e))}}class ne extends K{async _call(e){return new $a(await super._call(e))}}class se extends K{async _call(e){return new Ua(await super._call(e))}}class re extends R{}class oe extends re{}class ie extends re{async _call(e){return new qa(await super._call(e))}}class ae extends re{async _call(e){return new Ra(await super._call(e))}}class le extends re{async _call(e){return new $a(await super._call(e))}}class ce extends re{async _call(e){return new Ua(await super._call(e))}}class de extends R{}class ue extends de{}class pe extends de{async _call(e){return new qa(await super._call(e))}}class he extends de{async _call(e){return new Ra(await super._call(e))}}class _e extends de{async _call(e){return new $a(await super._call(e))}}class me extends de{async _call(e){return new Ua(await super._call(e))}}class fe extends R{}class ge extends fe{}class we extends fe{async _call(e){return new qa(await super._call(e))}}class Me extends fe{async _call(e){return new Ra(await super._call(e))}}class be extends fe{async _call(e){return new $a(await super._call(e))}}class ye extends fe{async _call(e){return new Ua(await super._call(e))}}class xe extends R{}class ke extends xe{}class Te extends xe{async _call(e){return new qa(await super._call(e))}}class ve extends xe{async _call(e){return new Ra(await super._call(e))}}class Ce extends xe{async _call(e){return new $a(await super._call(e))}}class Fe extends xe{async _call(e){return new Ua(await super._call(e))}}class Pe extends R{}class Se extends Pe{}class Ae extends Pe{async _call(e){return new qa(await super._call(e))}}class Ee extends Pe{async _call(e){return new Ra(await super._call(e))}}class ze extends Pe{async _call(e){return new $a(await super._call(e))}}class Le extends Pe{async _call(e){return new Ua(await super._call(e))}}class Ie extends R{}class Be extends Ie{}class Ne extends Ie{async _call(e){return new Ra(await super._call(e))}}class Oe extends Ie{async _call(e){return new $a(await super._call(e))}}class je extends Ie{async _call(e){return new Ua(await super._call(e))}}class De extends Ie{async _call(e){return new qa(await super._call(e))}}class Ve extends R{}class Re extends Ve{}class Ge extends Ve{async _call(e){return new qa(await super._call(e))}}class $e extends Ve{async _call(e){return new Ra(await super._call(e))}}class qe extends Ve{async _call(e){return new $a(await super._call(e))}}class Ue extends R{}class We extends Ue{}class Xe extends Ue{async _call(e){return new qa(await super._call(e))}}class Qe extends Ue{async _call(e){return new Ra(await super._call(e))}}class He extends Ue{async _call(e){return new Ua(await super._call(e))}}class Ye extends R{}class Je extends Ye{}class Ke extends Ye{async _call(e){return new qa(await super._call(e))}}class Ze extends Ye{async _call(e){return new Ra(await super._call(e))}}class et extends Ye{async _call(e){return new $a(await super._call(e))}}class tt extends Ye{async _call(e){return new Ua(await super._call(e))}}class nt extends R{}class st extends nt{}class rt extends nt{async _call(e){return new qa(await super._call(e))}}class ot extends nt{async _call(e){return new Ra(await super._call(e))}}class it extends nt{async _call(e){return new Ua(await super._call(e))}}class at extends R{}class lt extends at{}class ct extends at{async _call(e){return new Ra(await super._call(e))}}class dt extends at{async _call(e){return new Ua(await super._call(e))}}class ut extends at{async _call(e){return new qa(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,n){super(e,t),this.generation_config=n}}class ht extends pt{}class _t extends pt{}class mt extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class ft extends mt{}class gt extends mt{}class wt extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Mt extends wt{}class bt extends wt{}class yt extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class xt extends yt{}class kt extends yt{}class Tt extends yt{async _call(e){return new Ra(await super._call(e))}}class vt extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Ct extends vt{}class Ft extends vt{}class Pt extends vt{async _call(e){return new Ra(await super._call(e))}}class St extends vt{}class At extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Et extends At{}class zt extends At{}class Lt extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class It extends Lt{}class Bt extends Lt{}class Nt extends R{}class Ot extends Nt{}class jt extends Nt{async _call(e){return new qa(await super._call(e))}}class Dt extends Nt{async _call(e){return new Ra(await super._call(e))}}class Vt extends Nt{async _call(e){return new $a(await super._call(e))}}class Rt extends Nt{async _call(e){return new Ua(await super._call(e))}}class Gt extends R{}class $t extends Gt{}class qt extends Gt{async _call(e){return new qa(await super._call(e))}}class Ut extends Gt{async _call(e){return new Ra(await super._call(e))}}class Wt extends Gt{async _call(e){return new $a(await super._call(e))}}class Xt extends Gt{async _call(e){return new Ua(await super._call(e))}}class Qt extends R{}class Ht extends Qt{}class Yt extends Qt{async _call(e){return new qa(await super._call(e))}}class Jt extends Qt{async _call(e){return new Ra(await super._call(e))}}class Kt extends Qt{async _call(e){return new $a(await super._call(e))}}class Zt extends Qt{async _call(e){return new Ua(await super._call(e))}}class en extends R{}class tn extends en{}class nn extends en{}class sn extends R{requires_attention_mask=!1;main_input_name="input_features";forward_params=["input_features","attention_mask","decoder_input_ids","decoder_attention_mask","past_key_values"];constructor(e,t,n){super(e,t),this.generation_config=n}}class rn extends sn{}class on extends sn{_prepare_generation_config(e,t){return super._prepare_generation_config(e,t,f.WhisperGenerationConfig)}_retrieve_init_tokens(e){const t=[e.decoder_start_token_id];let n=e.language;const s=e.task;if(e.is_multilingual){n||(console.warn("No language specified - defaulting to English (en)."),n="en");const r=`<|${(0,g.whisper_language_to_code)(n)}|>`;t.push(e.lang_to_id[r]),t.push(e.task_to_id[s??"transcribe"])}else if(n||s)throw new Error("Cannot specify `task` or `language` for an English-only model. If the model is intended to be multilingual, pass `is_multilingual=true` to generate, or update the generation config.");return!e.return_timestamps&&e.no_timestamps_token_id&&t.at(-1)!==e.no_timestamps_token_id?t.push(e.no_timestamps_token_id):e.return_timestamps&&t.at(-1)===e.no_timestamps_token_id&&(console.warn("<|notimestamps|> prompt token is removed from generation_config since `return_timestamps` is set to `true`."),t.pop()),t.filter((e=>null!=e))}async generate({inputs:e=null,generation_config:t=null,logits_processor:n=null,stopping_criteria:s=null,...r}){t=this._prepare_generation_config(t,r);const o=r.decoder_input_ids??this._retrieve_init_tokens(t);if(t.return_timestamps&&(n??=new c.LogitsProcessorList,n.push(new c.WhisperTimeStampLogitsProcessor(t,o))),t.begin_suppress_tokens&&(n??=new c.LogitsProcessorList,n.push(new c.SuppressTokensAtBeginLogitsProcessor(t.begin_suppress_tokens,o.length))),t.return_token_timestamps){if(!t.alignment_heads)throw new Error("Model generation config has no `alignment_heads`, token-level timestamps not available. See https://gist.github.com/hollance/42e32852f24243b748ae6bc1f985b13a on how to add this property to the generation config.");"translate"===t.task&&console.warn("Token-level timestamps may not be reliable for task 'translate'."),t.output_attentions=!0,t.return_dict_in_generate=!0}const i=await super.generate({inputs:e,generation_config:t,logits_processor:n,decoder_input_ids:o,...r});return t.return_token_timestamps&&(i.token_timestamps=this._extract_token_timestamps(i,t.alignment_heads,t.num_frames)),i}_extract_token_timestamps(e,t,n=null,s=.02){if(!e.cross_attentions)throw new Error("Model outputs must contain cross attentions to extract timestamps. This is most likely because the model was not exported with `output_attentions=True`.");null==n&&console.warn("`num_frames` has not been set, meaning the entire audio will be analyzed. This may lead to inaccurate token-level timestamps for short audios (< 30 seconds).");let r=this.config.median_filter_width;void 0===r&&(console.warn("Model config has no `median_filter_width`, using default value of 7."),r=7);const o=e.cross_attentions,i=Array.from({length:this.config.decoder_layers},((e,t)=>(0,u.cat)(o.map((e=>e[t])),2))),l=(0,u.stack)(t.map((([e,t])=>{if(e>=i.length)throw new Error(`Layer index ${e} is out of bounds for cross attentions (length ${i.length}).`);return n?i[e].slice(null,t,null,[0,n]):i[e].slice(null,t)}))).transpose(1,0,2,3),[c,d]=(0,u.std_mean)(l,-2,0,!0),h=l.clone();for(let e=0;e<h.dims[0];++e){const t=h[e];for(let n=0;n<t.dims[0];++n){const s=t[n],o=c[e][n][0].data,i=d[e][n][0].data;for(let e=0;e<s.dims[0];++e){let t=s[e].data;for(let e=0;e<t.length;++e)t[e]=(t[e]-i[e])/o[e];t.set((0,p.medianFilter)(t,r))}}}const _=[(0,u.mean)(h,1)],m=e.sequences.dims,f=new u.Tensor("float32",new Float32Array(m[0]*m[1]),m);for(let e=0;e<m[0];++e){const t=_[e].neg().squeeze_(0),[n,r]=(0,p.dynamic_time_warping)(t.tolist()),o=Array.from({length:n.length-1},((e,t)=>n[t+1]-n[t])),i=(0,a.mergeArrays)([1],o).map((e=>!!e)),l=[];for(let e=0;e<i.length;++e)i[e]&&l.push(r[e]*s);f[e].data.set(l,1)}return f}}class an extends R{main_input_name="pixel_values";forward_params=["pixel_values","input_ids","encoder_hidden_states","past_key_values"];constructor(e,t,n){super(e,t),this.generation_config=n}}class ln extends R{forward_params=["input_ids","pixel_values","attention_mask","position_ids","past_key_values"];constructor(e,t,n){super(e,t),this.generation_config=n}}class cn extends ln{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:n,attention_mask:s}){const r=this.config.image_token_index,o=n.tolist().map((e=>e.findIndex((e=>e==r)))),i=o.every((e=>-1===e)),a=o.every((e=>-1!==e));if(!i&&!a)throw new Error("Every input should contain either 0 or 1 image token.");if(i)return{inputs_embeds:e,attention_mask:s};const l=[],c=[];for(let n=0;n<o.length;++n){const r=o[n],i=e[n],a=t[n],d=s[n];l.push((0,u.cat)([i.slice([0,r]),a,i.slice([r+1,i.dims[0]])],0)),c.push((0,u.cat)([d.slice([0,r]),(0,u.ones)([a.dims[0]]),d.slice([r+1,d.dims[0]])],0))}return{inputs_embeds:(0,u.stack)(l,0),attention_mask:(0,u.stack)(c,0)}}}class dn extends cn{}class un 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,n){super(e,t),this.generation_config=n}}class pn extends un{_merge_input_ids_with_image_features({inputs_embeds:e,image_features:t,input_ids:n,attention_mask:s}){return{inputs_embeds:(0,u.cat)([t,e],1),attention_mask:(0,u.cat)([(0,u.ones)(t.dims.slice(0,2)),s],1)}}async _prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:n,attention_mask:s}){if(!e&&!t)throw new Error("Either `input_ids` or `pixel_values` should be provided.");let r,o;return e&&(r=await this.encode_text({input_ids:e})),t&&(o=await this.encode_image({pixel_values:t})),r&&o?({inputs_embeds:n,attention_mask:s}=this._merge_input_ids_with_image_features({inputs_embeds:r,image_features:o,input_ids:e,attention_mask:s})):n=r||o,{inputs_embeds:n,attention_mask:s}}async forward({input_ids:e,pixel_values:t,attention_mask:n,decoder_input_ids:s,decoder_attention_mask:r,encoder_outputs:o,past_key_values:i,inputs_embeds:a,decoder_inputs_embeds:l}){if(a||({inputs_embeds:a,attention_mask:n}=await this._prepare_inputs_embeds({input_ids:e,pixel_values:t,inputs_embeds:a,attention_mask:n})),!o){let{last_hidden_state:e}=await B(this,{inputs_embeds:a,attention_mask:n});o=e}if(!l){if(!s)throw new Error("Either `decoder_input_ids` or `decoder_inputs_embeds` should be provided.");l=await this.encode_text({input_ids:s})}const c={inputs_embeds:l,attention_mask:r,encoder_attention_mask:n,encoder_hidden_states:o,past_key_values:i};return await N(this,c,!0)}}class hn extends R{}class _n extends hn{}class mn extends hn{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class fn extends hn{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class gn extends R{}class wn extends gn{}class Mn extends gn{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class bn extends hn{static async from_pretrained(e,t={}){return t.model_file_name??="vision_model",super.from_pretrained(e,t)}}class yn extends R{}class xn extends yn{}class kn extends R{}class Tn extends kn{}class vn extends kn{}class Cn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Fn extends Cn{}class Pn extends Cn{}class Sn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class An extends Sn{}class En extends Sn{}class zn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Ln extends zn{}class In extends zn{}class Bn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Nn extends Bn{}class On extends Bn{}class jn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Dn extends jn{}class Vn extends jn{}class Rn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Gn extends Rn{}class $n extends Rn{}class qn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Un extends qn{}class Wn extends qn{}class Xn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Qn extends Xn{}class Hn extends Xn{}class Yn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Jn extends Yn{}class Kn extends Yn{}class Zn extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class es extends Zn{}class ts extends Zn{}class ns extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class ss extends ns{}class rs extends ns{}class os extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class is extends os{}class as extends os{}class ls extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class cs extends ls{}class ds extends ls{}class us extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class ps extends us{}class hs extends us{}class _s extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class ms extends _s{}class fs extends _s{}class gs extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class ws extends gs{}class Ms extends gs{}class bs extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class ys extends bs{}class xs extends bs{}class ks extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Ts extends ks{}class vs extends ks{}class Cs extends R{}class Fs extends Cs{}class Ps extends Cs{async _call(e){return new Ra(await super._call(e))}}class Ss extends R{}class As extends Ss{}class Es extends Ss{async _call(e){return new Ra(await super._call(e))}}class zs extends R{}class Ls extends zs{async _call(e){return new Qa(await super._call(e))}}class Is extends R{}class Bs extends Is{}class Ns extends Is{async _call(e){return new Ra(await super._call(e))}}class Os extends R{}class js extends Os{}class Ds extends Os{async _call(e){return new Ra(await super._call(e))}}class Vs extends R{}class Rs extends Vs{}class Gs extends Vs{}class $s extends R{}class qs extends $s{}class Us extends $s{}class Ws extends R{}class Xs extends Ws{}class Qs extends Ws{async _call(e){return new Ra(await super._call(e))}}class Hs extends R{}class Ys extends Hs{}class Js extends Hs{async _call(e){return new Zs(await super._call(e))}}class Ks extends Hs{async _call(e){return new er(await super._call(e))}}class Zs extends G{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class er extends G{constructor({logits:e,pred_boxes:t,pred_masks:n}){super(),this.logits=e,this.pred_boxes=t,this.pred_masks=n}}class tr extends R{}class nr extends tr{}class sr extends tr{async _call(e){return new rr(await super._call(e))}}class rr extends G{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class or extends R{}class ir extends or{}class ar extends or{async _call(e){return new lr(await super._call(e))}}class lr extends Zs{}class cr extends R{}class dr extends cr{}class ur extends cr{async _call(e){return new Ra(await super._call(e))}}class pr extends R{}class hr extends pr{}class _r extends pr{async _call(e){return new Ra(await super._call(e))}}class mr extends R{}class fr extends mr{}class gr extends mr{async _call(e){return new Ra(await super._call(e))}}class wr extends R{}class Mr extends wr{}class br extends wr{async _call(e){return new Ra(await super._call(e))}}class yr extends R{}class xr extends yr{}class kr extends yr{}class Tr extends R{}class vr extends Tr{}class Cr extends Tr{}class Fr extends R{}class Pr extends Fr{}class Sr extends R{}class Ar extends Sr{}class Er extends Sr{}class zr extends Sr{}class Lr extends R{}class Ir extends Lr{}class Br extends Lr{}class Nr extends R{}class Or extends Nr{}class jr extends R{}class Dr extends jr{}class Vr extends jr{async _call(e){return new Ra(await super._call(e))}}class Rr extends R{}class Gr extends Rr{}class $r extends Rr{async _call(e){return new Ra(await super._call(e))}}class qr extends R{}class Ur extends qr{}class Wr extends qr{async _call(e){return new Ra(await super._call(e))}}class Xr extends R{}class Qr extends Xr{}class Hr extends Xr{async _call(e){return new Yr(await super._call(e))}}class Yr extends G{constructor({logits:e,pred_boxes:t}){super(),this.logits=e,this.pred_boxes=t}}class Jr extends R{}class Kr extends Jr{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),n=t.reduce(((e,t)=>e*t),1);e.input_labels=new u.Tensor("int64",new BigInt64Array(n).fill(1n),t)}const t={image_embeddings:e.image_embeddings,image_positional_embeddings:e.image_positional_embeddings};return e.input_points&&(t.input_points=e.input_points),e.input_labels&&(t.input_labels=e.input_labels),e.input_boxes&&(t.input_boxes=e.input_boxes),await A(this.sessions.prompt_encoder_mask_decoder,t)}async _call(e){return new Zr(await super._call(e))}}class Zr extends G{constructor({iou_scores:e,pred_masks:t}){super(),this.iou_scores=e,this.pred_masks=t}}class eo extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class to extends eo{}class no extends eo{}class so extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class ro extends so{}class oo extends so{}class io extends R{}class ao extends io{}class lo extends io{async _call(e){return new Wa(await super._call(e))}}class co extends io{async _call(e){return new Ra(await super._call(e))}}class uo extends io{async _call(e){return new $a(await super._call(e))}}class po extends R{}class ho extends po{}class _o extends po{async _call(e){return new $a(await super._call(e))}}class mo extends R{}class fo extends mo{}class go extends R{}class wo extends go{}class Mo extends go{async _call(e){return new Wa(await super._call(e))}}class bo extends go{async _call(e){return new Ra(await super._call(e))}}class yo extends R{}class xo extends yo{}class ko extends yo{async _call(e){return new Wa(await super._call(e))}}class To extends yo{async _call(e){return new Ra(await super._call(e))}}class vo extends yo{async _call(e){return new $a(await super._call(e))}}class Co extends R{}class Fo extends Co{}class Po extends Co{async _call(e){return new Wa(await super._call(e))}}class So extends Co{async _call(e){return new Ra(await super._call(e))}}class Ao extends R{}class Eo extends io{}class zo extends io{async _call(e){return new Wa(await super._call(e))}}class Lo extends io{async _call(e){return new Ra(await super._call(e))}}class Io extends R{}class Bo extends Io{}class No extends Io{async _call(e){return new Wa(await super._call(e))}}class Oo extends Io{async _call(e){return new Ra(await super._call(e))}}class jo extends Io{async _call(e){return new Ga(await super._call(e))}}class Do extends Io{async _call(e){return new $a(await super._call(e))}}class Vo extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Ro extends Vo{}class Go extends Vo{}class $o extends Vo{async generate_speech(e,t,{threshold:n=.5,minlenratio:s=0,maxlenratio:r=20,vocoder:o=null}={}){const i={input_ids:e},{encoder_outputs:a,encoder_attention_mask:l}=await B(this,i),c=a.dims[1]/this.config.reduction_factor,d=Math.floor(c*r),p=Math.floor(c*s),h=this.config.num_mel_bins;let _=[],m=null,f=null,g=0;for(;;){++g;const e=L(!!f);let s;s=f?f.output_sequence_out:new u.Tensor("float32",new Float32Array(h),[1,1,h]);let r={use_cache_branch:e,output_sequence:s,encoder_attention_mask:l,speaker_embeddings:t,encoder_hidden_states:a};this.addPastKeyValues(r,m),f=await A(this.sessions.decoder_model_merged,r),m=this.getPastKeyValues(f,m);const{prob:o,spectrum:i}=f;if(_.push(i),g>=p&&(Array.from(o.data).filter((e=>e>=n)).length>0||g>=d))break}const w=(0,u.cat)(_),{waveform:M}=await A(o.sessions.model,{spectrogram:w});return{spectrogram:w,waveform:M}}}class qo extends R{main_input_name="spectrogram"}class Uo extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Wo extends Uo{}class Xo extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Qo extends Xo{}class Ho extends Xo{}class Yo extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class Jo extends Yo{}class Ko extends Yo{}class Zo extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class ei extends Zo{}class ti extends Zo{}class ni extends R{}class si extends ni{}class ri extends ni{static async from_pretrained(e,t={}){return t.model_file_name??="text_model",super.from_pretrained(e,t)}}class oi extends ni{static async from_pretrained(e,t={}){return t.model_file_name??="audio_model",super.from_pretrained(e,t)}}class ii extends R{}class ai extends ii{async _call(e){return new Ha(await super._call(e))}}class li extends R{}class ci extends li{}class di extends li{}class ui extends li{}class pi extends R{constructor(e,t,n){super(e,t),this.generation_config=n}}class hi extends pi{}class _i extends pi{}class mi extends R{}class fi extends mi{}class gi extends mi{async _call(e){return new Ra(await super._call(e))}}class wi extends R{}class Mi extends wi{}class bi extends wi{}class yi extends R{forward_params=["input_ids","attention_mask","encoder_outputs","decoder_input_ids","decoder_attention_mask","past_key_values"];constructor(e,t,n){super(e,t),this.generation_config=n}_apply_and_filter_by_delay_pattern_mask(e){const[t,n]=e.dims,s=this.config.decoder.num_codebooks,r=n-s;let o=0;for(let t=0;t<e.size;++t){if(e.data[t]===this.config.decoder.pad_token_id)continue;const i=t%n-Math.floor(t/n)%s;i>0&&i<=r&&(e.data[o++]=e.data[t])}const i=Math.floor(t/s),a=o/(i*s);return new u.Tensor(e.type,e.data.slice(0,o),[i,s,a])}prepare_inputs_for_generation(e,t,n){let s=structuredClone(e);for(let e=0;e<s.length;++e)for(let t=0;t<s[e].length;++t)e%this.config.decoder.num_codebooks>=t&&(s[e][t]=BigInt(this.config.decoder.pad_token_id));null!==n.guidance_scale&&n.guidance_scale>1&&(s=s.concat(s));return super.prepare_inputs_for_generation(s,t,n)}async generate(e){const t=await super.generate(e),n=this._apply_and_filter_by_delay_pattern_mask(t).unsqueeze_(0),{audio_values:s}=await A(this.sessions.encodec_decode,{audio_codes:n});return s}}class xi extends R{}class ki extends xi{}class Ti extends xi{async _call(e){return new Ra(await super._call(e))}}class vi extends R{}class Ci extends vi{}class Fi extends vi{async _call(e){return new Ra(await super._call(e))}}class Pi extends R{}class Si extends Pi{}class Ai extends Pi{async _call(e){return new Ra(await super._call(e))}}class Ei extends R{}class zi extends Ei{}class Li extends Ei{async _call(e){return new Ra(await super._call(e))}}class Ii{static MODEL_CLASS_MAPPINGS=null;static BASE_IF_FAIL=!1;static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:r=null,local_files_only:o=!1,revision:i="main",model_file_name:a=null,subfolder:l="onnx",device:c=null,dtype:d=null,use_external_data_format:u=null,session_options:p={}}={}){let h={progress_callback:t,config:n,cache_dir:r,local_files_only:o,revision:i,model_file_name:a,subfolder:l,device:c,dtype:d,use_external_data_format:u,session_options:p};if(h.config=await s.AutoConfig.from_pretrained(e,h),!this.MODEL_CLASS_MAPPINGS)throw new Error("`MODEL_CLASS_MAPPINGS` not implemented for this type of `AutoClass`: "+this.name);for(let t of this.MODEL_CLASS_MAPPINGS){const n=t.get(h.config.model_type);if(n)return await n[1].from_pretrained(e,h)}if(this.BASE_IF_FAIL)return console.warn(`Unknown model class "${h.config.model_type}", attempting to construct from base class.`),await R.from_pretrained(e,h);throw Error(`Unsupported model type: ${h.config.model_type}`)}}const Bi=new Map([["bert",["BertModel",U]],["nomic_bert",["NomicBertModel",J]],["roformer",["RoFormerModel",Z]],["electra",["ElectraModel",ue]],["esm",["EsmModel",Re]],["convbert",["ConvBertModel",oe]],["camembert",["CamembertModel",ge]],["deberta",["DebertaModel",ke]],["deberta-v2",["DebertaV2Model",Se]],["mpnet",["MPNetModel",Je]],["albert",["AlbertModel",lt]],["distilbert",["DistilBertModel",Be]],["roberta",["RobertaModel",Ot]],["xlm",["XLMModel",$t]],["xlm-roberta",["XLMRobertaModel",Ht]],["clap",["ClapModel",si]],["clip",["CLIPModel",_n]],["clipseg",["CLIPSegModel",Tn]],["chinese_clip",["ChineseCLIPModel",xn]],["siglip",["SiglipModel",wn]],["mobilebert",["MobileBertModel",We]],["squeezebert",["SqueezeBertModel",st]],["wav2vec2",["Wav2Vec2Model",ao]],["wav2vec2-bert",["Wav2Vec2BertModel",Fo]],["unispeech",["UniSpeechModel",wo]],["unispeech-sat",["UniSpeechSatModel",xo]],["hubert",["HubertModel",Eo]],["wavlm",["WavLMModel",Bo]],["audio-spectrogram-transformer",["ASTModel",tn]],["vits",["VitsModel",ai]],["pyannote",["PyAnnoteModel",ho]],["wespeaker-resnet",["WeSpeakerResNetModel",fo]],["detr",["DetrModel",Ys]],["rt_detr",["RTDetrModel",nr]],["table-transformer",["TableTransformerModel",ir]],["vit",["ViTModel",Fs]],["fastvit",["FastViTModel",As]],["mobilevit",["MobileViTModel",Bs]],["mobilevitv2",["MobileViTV2Model",js]],["owlvit",["OwlViTModel",Rs]],["owlv2",["Owlv2Model",qs]],["beit",["BeitModel",Xs]],["deit",["DeiTModel",dr]],["hiera",["HieraModel",hr]],["convnext",["ConvNextModel",Dr]],["convnextv2",["ConvNextV2Model",Gr]],["dinov2",["Dinov2Model",Ur]],["resnet",["ResNetModel",fr]],["swin",["SwinModel",Mr]],["swin2sr",["Swin2SRModel",xr]],["donut-swin",["DonutSwinModel",Or]],["yolos",["YolosModel",Qr]],["dpt",["DPTModel",vr]],["glpn",["GLPNModel",Ir]],["hifigan",["SpeechT5HifiGan",qo]],["efficientnet",["EfficientNetModel",fi]],["mobilenet_v1",["MobileNetV1Model",ki]],["mobilenet_v2",["MobileNetV2Model",Ci]],["mobilenet_v3",["MobileNetV3Model",Si]],["mobilenet_v4",["MobileNetV4Model",zi]]]),Ni=new Map([["t5",["T5Model",ht]],["longt5",["LongT5Model",ft]],["mt5",["MT5Model",Mt]],["bart",["BartModel",xt]],["mbart",["MBartModel",Ct]],["marian",["MarianModel",to]],["whisper",["WhisperModel",rn]],["m2m_100",["M2M100Model",ro]],["blenderbot",["BlenderbotModel",Et]],["blenderbot-small",["BlenderbotSmallModel",It]]]),Oi=new Map([["bloom",["BloomModel",ws]],["jais",["JAISModel",An]],["gpt2",["GPT2Model",Fn]],["gptj",["GPTJModel",Dn]],["gpt_bigcode",["GPTBigCodeModel",Gn]],["gpt_neo",["GPTNeoModel",Ln]],["gpt_neox",["GPTNeoXModel",Nn]],["codegen",["CodeGenModel",Un]],["llama",["LlamaModel",Qn]],["cohere",["CohereModel",Jn]],["gemma",["GemmaModel",es]],["gemma2",["Gemma2Model",ss]],["openelm",["OpenELMModel",is]],["qwen2",["Qwen2Model",cs]],["phi",["PhiModel",ps]],["phi3",["Phi3Model",ms]],["mpt",["MptModel",ys]],["opt",["OPTModel",Ts]],["mistral",["MistralModel",Qo]],["starcoder2",["Starcoder2Model",Jo]],["falcon",["FalconModel",ei]],["stablelm",["StableLmModel",hi]]]),ji=new Map([["speecht5",["SpeechT5ForSpeechToText",Go]],["whisper",["WhisperForConditionalGeneration",on]]]),Di=new Map([["speecht5",["SpeechT5ForTextToSpeech",$o]]]),Vi=new Map([["vits",["VitsModel",ai]],["musicgen",["MusicgenForConditionalGeneration",yi]]]),Ri=new Map([["bert",["BertForSequenceClassification",X]],["roformer",["RoFormerForSequenceClassification",te]],["electra",["ElectraForSequenceClassification",he]],["esm",["EsmForSequenceClassification",$e]],["convbert",["ConvBertForSequenceClassification",ae]],["camembert",["CamembertForSequenceClassification",Me]],["deberta",["DebertaForSequenceClassification",ve]],["deberta-v2",["DebertaV2ForSequenceClassification",Ee]],["mpnet",["MPNetForSequenceClassification",Ze]],["albert",["AlbertForSequenceClassification",ct]],["distilbert",["DistilBertForSequenceClassification",Ne]],["roberta",["RobertaForSequenceClassification",Dt]],["xlm",["XLMForSequenceClassification",Ut]],["xlm-roberta",["XLMRobertaForSequenceClassification",Jt]],["bart",["BartForSequenceClassification",Tt]],["mbart",["MBartForSequenceClassification",Pt]],["mobilebert",["MobileBertForSequenceClassification",Qe]],["squeezebert",["SqueezeBertForSequenceClassification",ot]]]),Gi=new Map([["bert",["BertForTokenClassification",Q]],["roformer",["RoFormerForTokenClassification",ne]],["electra",["ElectraForTokenClassification",_e]],["esm",["EsmForTokenClassification",qe]],["convbert",["ConvBertForTokenClassification",le]],["camembert",["CamembertForTokenClassification",be]],["deberta",["DebertaForTokenClassification",Ce]],["deberta-v2",["DebertaV2ForTokenClassification",ze]],["mpnet",["MPNetForTokenClassification",et]],["distilbert",["DistilBertForTokenClassification",Oe]],["roberta",["RobertaForTokenClassification",Vt]],["xlm",["XLMForTokenClassification",Wt]],["xlm-roberta",["XLMRobertaForTokenClassification",Kt]]]),$i=new Map([["t5",["T5ForConditionalGeneration",_t]],["longt5",["LongT5ForConditionalGeneration",gt]],["mt5",["MT5ForConditionalGeneration",bt]],["bart",["BartForConditionalGeneration",kt]],["mbart",["MBartForConditionalGeneration",Ft]],["marian",["MarianMTModel",no]],["m2m_100",["M2M100ForConditionalGeneration",oo]],["blenderbot",["BlenderbotForConditionalGeneration",zt]],["blenderbot-small",["BlenderbotSmallForConditionalGeneration",Bt]]]),qi=new Map([["bloom",["BloomForCausalLM",Ms]],["gpt2",["GPT2LMHeadModel",Pn]],["jais",["JAISLMHeadModel",En]],["gptj",["GPTJForCausalLM",Vn]],["gpt_bigcode",["GPTBigCodeForCausalLM",$n]],["gpt_neo",["GPTNeoForCausalLM",In]],["gpt_neox",["GPTNeoXForCausalLM",On]],["codegen",["CodeGenForCausalLM",Wn]],["llama",["LlamaForCausalLM",Hn]],["cohere",["CohereForCausalLM",Kn]],["gemma",["GemmaForCausalLM",ts]],["gemma2",["Gemma2ForCausalLM",rs]],["openelm",["OpenELMForCausalLM",as]],["qwen2",["Qwen2ForCausalLM",ds]],["phi",["PhiForCausalLM",hs]],["phi3",["Phi3ForCausalLM",fs]],["mpt",["MptForCausalLM",xs]],["opt",["OPTForCausalLM",vs]],["mbart",["MBartForCausalLM",St]],["mistral",["MistralForCausalLM",Ho]],["starcoder2",["Starcoder2ForCausalLM",Ko]],["falcon",["FalconForCausalLM",ti]],["trocr",["TrOCRForCausalLM",Wo]],["stablelm",["StableLmForCausalLM",_i]]]),Ui=new Map([["bert",["BertForMaskedLM",W]],["roformer",["RoFormerForMaskedLM",ee]],["electra",["ElectraForMaskedLM",pe]],["esm",["EsmForMaskedLM",Ge]],["convbert",["ConvBertForMaskedLM",ie]],["camembert",["CamembertForMaskedLM",we]],["deberta",["DebertaForMaskedLM",Te]],["deberta-v2",["DebertaV2ForMaskedLM",Ae]],["mpnet",["MPNetForMaskedLM",Ke]],["albert",["AlbertForMaskedLM",ut]],["distilbert",["DistilBertForMaskedLM",De]],["roberta",["RobertaForMaskedLM",jt]],["xlm",["XLMWithLMHeadModel",qt]],["xlm-roberta",["XLMRobertaForMaskedLM",Yt]],["mobilebert",["MobileBertForMaskedLM",Xe]],["squeezebert",["SqueezeBertForMaskedLM",rt]]]),Wi=new Map([["bert",["BertForQuestionAnswering",H]],["roformer",["RoFormerForQuestionAnswering",se]],["electra",["ElectraForQuestionAnswering",me]],["convbert",["ConvBertForQuestionAnswering",ce]],["camembert",["CamembertForQuestionAnswering",ye]],["deberta",["DebertaForQuestionAnswering",Fe]],["deberta-v2",["DebertaV2ForQuestionAnswering",Le]],["mpnet",["MPNetForQuestionAnswering",tt]],["albert",["AlbertForQuestionAnswering",dt]],["distilbert",["DistilBertForQuestionAnswering",je]],["roberta",["RobertaForQuestionAnswering",Rt]],["xlm",["XLMForQuestionAnswering",Xt]],["xlm-roberta",["XLMRobertaForQuestionAnswering",Zt]],["mobilebert",["MobileBertForQuestionAnswering",He]],["squeezebert",["SqueezeBertForQuestionAnswering",it]]]),Xi=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",an]]]),Qi=new Map([["llava",["LlavaForConditionalGeneration",cn]],["moondream1",["Moondream1ForConditionalGeneration",dn]],["florence2",["Florence2ForConditionalGeneration",pn]]]),Hi=new Map([["vision-encoder-decoder",["VisionEncoderDecoderModel",an]]]),Yi=new Map([["vit",["ViTForImageClassification",Ps]],["fastvit",["FastViTForImageClassification",Es]],["mobilevit",["MobileViTForImageClassification",Ns]],["mobilevitv2",["MobileViTV2ForImageClassification",Ds]],["beit",["BeitForImageClassification",Qs]],["deit",["DeiTForImageClassification",ur]],["hiera",["HieraForImageClassification",_r]],["convnext",["ConvNextForImageClassification",Vr]],["convnextv2",["ConvNextV2ForImageClassification",$r]],["dinov2",["Dinov2ForImageClassification",Wr]],["resnet",["ResNetForImageClassification",gr]],["swin",["SwinForImageClassification",br]],["segformer",["SegformerForImageClassification",di]],["efficientnet",["EfficientNetForImageClassification",gi]],["mobilenet_v1",["MobileNetV1ForImageClassification",Ti]],["mobilenet_v2",["MobileNetV2ForImageClassification",Fi]],["mobilenet_v3",["MobileNetV3ForImageClassification",Ai]],["mobilenet_v4",["MobileNetV4ForImageClassification",Li]]]),Ji=new Map([["detr",["DetrForObjectDetection",Js]],["rt_detr",["RTDetrForObjectDetection",sr]],["table-transformer",["TableTransformerForObjectDetection",ar]],["yolos",["YolosForObjectDetection",Hr]]]),Ki=new Map([["owlvit",["OwlViTForObjectDetection",Gs]],["owlv2",["Owlv2ForObjectDetection",Us]]]),Zi=new Map([["detr",["DetrForSegmentation",Ks]],["clipseg",["CLIPSegForImageSegmentation",vn]]]),ea=new Map([["segformer",["SegformerForSemanticSegmentation",ui]],["sapiens",["SapiensForSemanticSegmentation",Ar]]]),ta=new Map([["sam",["SamModel",Kr]]]),na=new Map([["wav2vec2",["Wav2Vec2ForCTC",lo]],["wav2vec2-bert",["Wav2Vec2BertForCTC",Po]],["unispeech",["UniSpeechForCTC",Mo]],["unispeech-sat",["UniSpeechSatForCTC",ko]],["wavlm",["WavLMForCTC",No]],["hubert",["HubertForCTC",zo]]]),sa=new Map([["wav2vec2",["Wav2Vec2ForSequenceClassification",co]],["wav2vec2-bert",["Wav2Vec2BertForSequenceClassification",So]],["unispeech",["UniSpeechForSequenceClassification",bo]],["unispeech-sat",["UniSpeechSatForSequenceClassification",To]],["wavlm",["WavLMForSequenceClassification",Oo]],["hubert",["HubertForSequenceClassification",Lo]],["audio-spectrogram-transformer",["ASTForAudioClassification",nn]]]),ra=new Map([["wavlm",["WavLMForXVector",jo]]]),oa=new Map([["unispeech-sat",["UniSpeechSatForAudioFrameClassification",vo]],["wavlm",["WavLMForAudioFrameClassification",Do]],["wav2vec2",["Wav2Vec2ForAudioFrameClassification",uo]],["pyannote",["PyAnnoteForAudioFrameClassification",_o]]]),ia=new Map([["vitmatte",["VitMatteForImageMatting",Ls]]]),aa=new Map([["swin2sr",["Swin2SRForImageSuperResolution",kr]]]),la=new Map([["dpt",["DPTForDepthEstimation",Cr]],["depth_anything",["DepthAnythingForDepthEstimation",Pr]],["glpn",["GLPNForDepthEstimation",Br]],["sapiens",["SapiensForDepthEstimation",Er]]]),ca=new Map([["sapiens",["SapiensForNormalEstimation",zr]]]),da=new Map([["clip",["CLIPVisionModelWithProjection",fn]],["siglip",["SiglipVisionModel",bn]]]),ua=[[Bi,w],[Ni,M],[Oi,x],[Ri,w],[Gi,w],[$i,b],[ji,b],[qi,x],[Ui,w],[Wi,w],[Xi,y],[Qi,T],[Yi,w],[Zi,w],[ea,w],[ia,w],[aa,w],[la,w],[ca,w],[Ji,w],[Ki,w],[ta,k],[na,w],[sa,w],[Di,b],[Vi,w],[ra,w],[oa,w],[da,w]];for(const[e,t]of ua)for(const[n,s]of e.values())C.set(n,t),P.set(s,n),F.set(n,s);const pa=[["MusicgenForConditionalGeneration",yi,v],["CLIPTextModelWithProjection",mn,w],["SiglipTextModel",Mn,w],["ClapTextModelWithProjection",ri,w],["ClapAudioModelWithProjection",oi,w]];for(const[e,t,n]of pa)C.set(e,n),P.set(t,e),F.set(e,t);class ha extends Ii{static MODEL_CLASS_MAPPINGS=ua.map((e=>e[0]));static BASE_IF_FAIL=!0}class _a extends Ii{static MODEL_CLASS_MAPPINGS=[Ri]}class ma extends Ii{static MODEL_CLASS_MAPPINGS=[Gi]}class fa extends Ii{static MODEL_CLASS_MAPPINGS=[$i]}class ga extends Ii{static MODEL_CLASS_MAPPINGS=[ji]}class wa extends Ii{static MODEL_CLASS_MAPPINGS=[Di]}class Ma extends Ii{static MODEL_CLASS_MAPPINGS=[Vi]}class ba extends Ii{static MODEL_CLASS_MAPPINGS=[qi]}class ya extends Ii{static MODEL_CLASS_MAPPINGS=[Ui]}class xa extends Ii{static MODEL_CLASS_MAPPINGS=[Wi]}class ka extends Ii{static MODEL_CLASS_MAPPINGS=[Xi]}class Ta extends Ii{static MODEL_CLASS_MAPPINGS=[Yi]}class va extends Ii{static MODEL_CLASS_MAPPINGS=[Zi]}class Ca extends Ii{static MODEL_CLASS_MAPPINGS=[ea]}class Fa extends Ii{static MODEL_CLASS_MAPPINGS=[Ji]}class Pa extends Ii{static MODEL_CLASS_MAPPINGS=[Ki]}class Sa extends Ii{static MODEL_CLASS_MAPPINGS=[ta]}class Aa extends Ii{static MODEL_CLASS_MAPPINGS=[na]}class Ea extends Ii{static MODEL_CLASS_MAPPINGS=[sa]}class za extends Ii{static MODEL_CLASS_MAPPINGS=[ra]}class La extends Ii{static MODEL_CLASS_MAPPINGS=[oa]}class Ia extends Ii{static MODEL_CLASS_MAPPINGS=[Hi]}class Ba extends Ii{static MODEL_CLASS_MAPPINGS=[ia]}class Na extends Ii{static MODEL_CLASS_MAPPINGS=[aa]}class Oa extends Ii{static MODEL_CLASS_MAPPINGS=[la]}class ja extends Ii{static MODEL_CLASS_MAPPINGS=[ca]}class Da extends Ii{static MODEL_CLASS_MAPPINGS=[da]}class Va extends G{constructor({logits:e,past_key_values:t,encoder_outputs:n,decoder_attentions:s=null,cross_attentions:r=null}){super(),this.logits=e,this.past_key_values=t,this.encoder_outputs=n,this.decoder_attentions=s,this.cross_attentions=r}}class Ra extends G{constructor({logits:e}){super(),this.logits=e}}class Ga extends G{constructor({logits:e,embeddings:t}){super(),this.logits=e,this.embeddings=t}}class $a extends G{constructor({logits:e}){super(),this.logits=e}}class qa extends G{constructor({logits:e}){super(),this.logits=e}}class Ua extends G{constructor({start_logits:e,end_logits:t}){super(),this.start_logits=e,this.end_logits=t}}class Wa extends G{constructor({logits:e}){super(),this.logits=e}}class Xa extends G{constructor({logits:e,past_key_values:t}){super(),this.logits=e,this.past_key_values=t}}class Qa extends G{constructor({alphas:e}){super(),this.alphas=e}}class Ha 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,n)=>{"use strict";n.r(t),n.d(t,{WHISPER_LANGUAGE_MAPPING:()=>r,WHISPER_TO_LANGUAGE_CODE_MAPPING:()=>o,whisper_language_to_code:()=>i});const s=[["en","english"],["zh","chinese"],["de","german"],["es","spanish"],["ru","russian"],["ko","korean"],["fr","french"],["ja","japanese"],["pt","portuguese"],["tr","turkish"],["pl","polish"],["ca","catalan"],["nl","dutch"],["ar","arabic"],["sv","swedish"],["it","italian"],["id","indonesian"],["hi","hindi"],["fi","finnish"],["vi","vietnamese"],["he","hebrew"],["uk","ukrainian"],["el","greek"],["ms","malay"],["cs","czech"],["ro","romanian"],["da","danish"],["hu","hungarian"],["ta","tamil"],["no","norwegian"],["th","thai"],["ur","urdu"],["hr","croatian"],["bg","bulgarian"],["lt","lithuanian"],["la","latin"],["mi","maori"],["ml","malayalam"],["cy","welsh"],["sk","slovak"],["te","telugu"],["fa","persian"],["lv","latvian"],["bn","bengali"],["sr","serbian"],["az","azerbaijani"],["sl","slovenian"],["kn","kannada"],["et","estonian"],["mk","macedonian"],["br","breton"],["eu","basque"],["is","icelandic"],["hy","armenian"],["ne","nepali"],["mn","mongolian"],["bs","bosnian"],["kk","kazakh"],["sq","albanian"],["sw","swahili"],["gl","galician"],["mr","marathi"],["pa","punjabi"],["si","sinhala"],["km","khmer"],["sn","shona"],["yo","yoruba"],["so","somali"],["af","afrikaans"],["oc","occitan"],["ka","georgian"],["be","belarusian"],["tg","tajik"],["sd","sindhi"],["gu","gujarati"],["am","amharic"],["yi","yiddish"],["lo","lao"],["uz","uzbek"],["fo","faroese"],["ht","haitian creole"],["ps","pashto"],["tk","turkmen"],["nn","nynorsk"],["mt","maltese"],["sa","sanskrit"],["lb","luxembourgish"],["my","myanmar"],["bo","tibetan"],["tl","tagalog"],["mg","malagasy"],["as","assamese"],["tt","tatar"],["haw","hawaiian"],["ln","lingala"],["ha","hausa"],["ba","bashkir"],["jw","javanese"],["su","sundanese"]],r=new Map(s),o=new Map([...s.map((([e,t])=>[t,e])),["burmese","my"],["valencian","ca"],["flemish","nl"],["haitian","ht"],["letzeburgesch","lb"],["pushto","ps"],["panjabi","pa"],["moldavian","ro"],["moldovan","ro"],["sinhalese","si"],["castilian","es"]]);function i(e){e=e.toLowerCase();let t=o.get(e);if(void 0===t){if(!r.has(e)){const t=2===e.length?r.keys():r.values();throw new Error(`Language "${e}" is not supported. Must be one of: ${JSON.stringify(t)}`)}t=e}return t}},"./src/models/whisper/generation_whisper.js":
@@ -124,7 +124,7 @@
   \**************************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{AudioClassificationPipeline:()=>P,AutomaticSpeechRecognitionPipeline:()=>A,DepthEstimationPipeline:()=>V,DocumentQuestionAnsweringPipeline:()=>O,FeatureExtractionPipeline:()=>C,FillMaskPipeline:()=>M,ImageClassificationPipeline:()=>z,ImageFeatureExtractionPipeline:()=>F,ImageSegmentationPipeline:()=>L,ImageToImagePipeline:()=>D,ImageToTextPipeline:()=>E,ObjectDetectionPipeline:()=>B,Pipeline:()=>m,QuestionAnsweringPipeline:()=>w,SummarizationPipeline:()=>y,Text2TextGenerationPipeline:()=>b,TextClassificationPipeline:()=>f,TextGenerationPipeline:()=>T,TextToAudioPipeline:()=>j,TokenClassificationPipeline:()=>g,TranslationPipeline:()=>x,ZeroShotAudioClassificationPipeline:()=>S,ZeroShotClassificationPipeline:()=>v,ZeroShotImageClassificationPipeline:()=>I,ZeroShotObjectDetectionPipeline:()=>N,pipeline:()=>$});var s=n(/*! ./tokenizers.js */"./src/tokenizers.js"),r=n(/*! ./models.js */"./src/models.js"),o=n(/*! ./processors.js */"./src/processors.js"),i=n(/*! ./utils/generic.js */"./src/utils/generic.js"),a=n(/*! ./utils/core.js */"./src/utils/core.js"),l=n(/*! ./utils/maths.js */"./src/utils/maths.js"),c=n(/*! ./utils/audio.js */"./src/utils/audio.js"),d=n(/*! ./utils/tensor.js */"./src/utils/tensor.js"),u=n(/*! ./utils/image.js */"./src/utils/image.js");async function p(e){return Array.isArray(e)||(e=[e]),await Promise.all(e.map((e=>u.RawImage.read(e))))}async function h(e,t){return Array.isArray(e)||(e=[e]),await Promise.all(e.map((e=>"string"==typeof e||e instanceof URL?(0,c.read_audio)(e,t):e instanceof Float64Array?new Float32Array(e):e)))}function _(e,t){t&&(e=e.map((e=>0|e)));const[n,s,r,o]=e;return{xmin:n,ymin:s,xmax:r,ymax:o}}class m extends i.Callable{constructor({task:e,model:t,tokenizer:n=null,processor:s=null}){super(),this.task=e,this.model=t,this.tokenizer=n,this.processor=s}async dispose(){await this.model.dispose()}}class f extends m{constructor(e){super(e)}async _call(e,{top_k:t=1}={}){const n=this.tokenizer(e,{padding:!0,truncation:!0}),s=await this.model(n),r="multi_label_classification"===this.model.config.problem_type?e=>e.sigmoid():e=>new d.Tensor("float32",(0,l.softmax)(e.data),e.dims),o=this.model.config.id2label,i=[];for(const e of s.logits){const n=r(e),s=await(0,d.topk)(n,t),a=s[0].tolist(),l=s[1].tolist().map(((e,t)=>({label:o?o[e]:`LABEL_${e}`,score:a[t]})));1===t?i.push(...l):i.push(l)}return Array.isArray(e)||1===t?i:i[0]}}class g extends m{constructor(e){super(e)}async _call(e,{ignore_labels:t=["O"]}={}){const n=Array.isArray(e),s=this.tokenizer(n?e:[e],{padding:!0,truncation:!0}),r=(await this.model(s)).logits,o=this.model.config.id2label,i=[];for(let e=0;e<r.dims[0];++e){const n=s.input_ids[e],a=r[e],c=[];for(let e=0;e<a.dims[0];++e){const s=a[e],r=(0,l.max)(s.data)[1],i=o?o[r]:`LABEL_${r}`;if(t.includes(i))continue;const d=this.tokenizer.decode([n[e].item()],{skip_special_tokens:!0});if(""===d)continue;const u=(0,l.softmax)(s.data);c.push({entity:i,score:u[r],index:e,word:d})}i.push(c)}return n?i:i[0]}}class w extends m{constructor(e){super(e)}async _call(e,t,{top_k:n=1}={}){const s=this.tokenizer(e,{text_pair:t,padding:!0,truncation:!0}),{start_logits:r,end_logits:o}=await this.model(s),i=s.input_ids.tolist(),c=s.attention_mask.tolist(),d=this.tokenizer.all_special_ids,u=[];for(let e=0;e<r.dims[0];++e){const t=i[e],s=t.findIndex((e=>e==this.tokenizer.sep_token_id)),p=(c[e].map(((e,n)=>1==e&&(0===n||n>s&&-1===d.findIndex((e=>e==t[n]))))),r[e].tolist()),h=o[e].tolist();for(let n=1;n<p.length;++n)(0==c[e]||n<=s||-1!==d.findIndex((e=>e==t[n])))&&(p[n]=-1/0,h[n]=-1/0);const _=(0,l.softmax)(p).map(((e,t)=>[e,t])),m=(0,l.softmax)(h).map(((e,t)=>[e,t]));_[0][0]=0,m[0][0]=0;const f=(0,a.product)(_,m).filter((e=>e[0][1]<=e[1][1])).map((e=>[e[0][1],e[1][1],e[0][0]*e[1][0]])).sort(((e,t)=>t[2]-e[2]));for(let e=0;e<Math.min(f.length,n);++e){const[n,s,r]=f[e],o=t.slice(n,s+1),i=this.tokenizer.decode(o,{skip_special_tokens:!0});u.push({answer:i,score:r})}}return 1===n?u[0]:u}}class M extends m{constructor(e){super(e)}async _call(e,{top_k:t=5}={}){const n=this.tokenizer(e,{padding:!0,truncation:!0}),{logits:s}=await this.model(n),r=[],o=n.input_ids.tolist();for(let e=0;e<o.length;++e){const n=o[e],i=n.findIndex((e=>e==this.tokenizer.mask_token_id));if(-1===i)throw Error(`Mask token (${this.tokenizer.mask_token}) not found in text.`);const a=s[e][i],c=await(0,d.topk)(new d.Tensor("float32",(0,l.softmax)(a.data),a.dims),t),u=c[0].tolist(),p=c[1].tolist();r.push(p.map(((e,t)=>{const s=n.slice();return s[i]=e,{score:u[t],token:Number(e),token_str:this.tokenizer.model.vocab[e],sequence:this.tokenizer.decode(s,{skip_special_tokens:!0})}})))}return Array.isArray(e)?r:r[0]}}class b extends m{_key="generated_text";constructor(e){super(e)}async _call(e,t={}){Array.isArray(e)||(e=[e]),this.model.config.prefix&&(e=e.map((e=>this.model.config.prefix+e)));const n=this.model.config.task_specific_params;n&&n[this.task]&&n[this.task].prefix&&(e=e.map((e=>n[this.task].prefix+e)));const s=this.tokenizer,r={padding:!0,truncation:!0};let o;o=this instanceof x&&"_build_translation_inputs"in s?s._build_translation_inputs(e,r,t):s(e,r);const i=await this.model.generate({...o,...t});return s.batch_decode(i,{skip_special_tokens:!0}).map((e=>({[this._key]:e})))}}class y extends b{_key="summary_text";constructor(e){super(e)}}class x extends b{_key="translation_text";constructor(e){super(e)}}function k(e){return Array.isArray(e)&&e.every((e=>"role"in e&&"content"in e))}class T extends m{constructor(e){super(e)}async _call(e,t={}){let n,s=!1,r=!1;if("string"==typeof e)n=e=[e];else if(Array.isArray(e)&&e.every((e=>"string"==typeof e)))s=!0,n=e;else{if(k(e))e=[e];else{if(!Array.isArray(e)||!e.every(k))throw new Error("Input must be a string, an array of strings, a Chat, or an array of Chats");s=!0}r=!0,n=e.map((e=>this.tokenizer.apply_chat_template(e,{tokenize:!1,add_generation_prompt:!0})))}const o=t.add_special_tokens??!1,i=!r&&(t.return_full_text??!0);this.tokenizer.padding_side="left";const a=this.tokenizer(n,{add_special_tokens:o,padding:!0,truncation:!0}),l=await this.model.generate({...a,...t}),c=this.tokenizer.batch_decode(l,{skip_special_tokens:!0});let d;!i&&a.input_ids.dims.at(-1)>0&&(d=this.tokenizer.batch_decode(a.input_ids,{skip_special_tokens:!0}).map((e=>e.length)));const u=Array.from({length:e.length},(e=>[]));for(let t=0;t<c.length;++t){const n=Math.floor(t/l.dims[0]*e.length);d&&(c[t]=c[t].slice(d[n])),u[n].push({generated_text:r?[...e[n],{role:"assistant",content:c[t]}]:c[t]})}return s||1!==u.length?u:u[0]}}class v extends m{constructor(e){super(e),this.label2id=Object.fromEntries(Object.entries(this.model.config.label2id).map((([e,t])=>[e.toLowerCase(),t]))),this.entailment_id=this.label2id.entailment,void 0===this.entailment_id&&(console.warn("Could not find 'entailment' in label2id mapping. Using 2 as entailment_id."),this.entailment_id=2),this.contradiction_id=this.label2id.contradiction??this.label2id.not_entailment,void 0===this.contradiction_id&&(console.warn("Could not find 'contradiction' in label2id mapping. Using 0 as contradiction_id."),this.contradiction_id=0)}async _call(e,t,{hypothesis_template:n="This example is {}.",multi_label:s=!1}={}){const r=Array.isArray(e);r||(e=[e]),Array.isArray(t)||(t=[t]);const o=t.map((e=>n.replace("{}",e))),i=s||1===t.length,a=[];for(const n of e){const e=[];for(const t of o){const s=this.tokenizer(n,{text_pair:t,padding:!0,truncation:!0}),r=await this.model(s);i?e.push([r.logits.data[this.contradiction_id],r.logits.data[this.entailment_id]]):e.push(r.logits.data[this.entailment_id])}const s=(i?e.map((e=>(0,l.softmax)(e)[1])):(0,l.softmax)(e)).map(((e,t)=>[e,t])).sort(((e,t)=>t[0]-e[0]));a.push({sequence:n,labels:s.map((e=>t[e[1]])),scores:s.map((e=>e[0]))})}return r?a:a[0]}}class C extends m{constructor(e){super(e)}async _call(e,{pooling:t="none",normalize:n=!1,quantize:s=!1,precision:r="binary"}={}){const o=this.tokenizer(e,{padding:!0,truncation:!0}),i=await this.model(o);let a=i.last_hidden_state??i.logits??i.token_embeddings;if("none"===t);else if("mean"===t)a=(0,d.mean_pooling)(a,o.attention_mask);else{if("cls"!==t)throw Error(`Pooling method '${t}' not supported.`);a=a.slice(null,0)}return n&&(a=a.normalize(2,-1)),s&&(a=(0,d.quantize_embeddings)(a,r)),a}}class F extends m{constructor(e){super(e)}async _call(e,{pool:t=null}={}){const n=await p(e),{pixel_values:s}=await this.processor(n),r=await this.model({pixel_values:s});let o;if(t){if(!("pooler_output"in r))throw Error("No pooled output was returned. Make sure the model has a 'pooler' layer when using the 'pool' option.");o=r.pooler_output}else o=r.last_hidden_state??r.logits??r.image_embeds;return o}}class P extends m{constructor(e){super(e)}async _call(e,{top_k:t=5}={}){const n=this.processor.feature_extractor.config.sampling_rate,s=await h(e,n),r=this.model.config.id2label,o=[];for(const e of s){const n=await this.processor(e),s=(await this.model(n)).logits[0],i=await(0,d.topk)(new d.Tensor("float32",(0,l.softmax)(s.data),s.dims),t),a=i[0].tolist(),c=i[1].tolist().map(((e,t)=>({label:r?r[e]:`LABEL_${e}`,score:a[t]})));o.push(c)}return Array.isArray(e)?o:o[0]}}class S extends m{constructor(e){super(e)}async _call(e,t,{hypothesis_template:n="This is a sound of {}."}={}){const s=!Array.isArray(e);s&&(e=[e]);const r=t.map((e=>n.replace("{}",e))),o=this.tokenizer(r,{padding:!0,truncation:!0}),i=this.processor.feature_extractor.config.sampling_rate,a=await h(e,i),c=[];for(const e of a){const n=await this.processor(e),s=await this.model({...o,...n}),r=(0,l.softmax)(s.logits_per_audio.data);c.push([...r].map(((e,n)=>({score:e,label:t[n]}))))}return s?c[0]:c}}class A extends m{constructor(e){super(e)}async _call(e,t={}){switch(this.model.config.model_type){case"whisper":return this._call_whisper(e,t);case"wav2vec2":case"wav2vec2-bert":case"unispeech":case"unispeech-sat":case"hubert":return this._call_wav2vec2(e,t);default:throw new Error(`AutomaticSpeechRecognitionPipeline does not support model type '${this.model.config.model_type}'.`)}}async _call_wav2vec2(e,t){t.language&&console.warn('`language` parameter is not yet supported for `wav2vec2` models, defaulting to "English".'),t.task&&console.warn('`task` parameter is not yet supported for `wav2vec2` models, defaulting to "transcribe".');const n=!Array.isArray(e);n&&(e=[e]);const s=this.processor.feature_extractor.config.sampling_rate,r=await h(e,s),o=[];for(const e of r){const t=await this.processor(e),n=(await this.model(t)).logits[0],s=[];for(const e of n)s.push((0,l.max)(e.data)[1]);const r=this.tokenizer.decode(s);o.push({text:r})}return n?o[0]:o}async _call_whisper(e,t){const n=t.return_timestamps??!1,s=t.chunk_length_s??0,r=t.force_full_sequences??!1;let o=t.stride_length_s??null;const i={...t};"word"===n&&(i.return_token_timestamps=!0,i.return_timestamps=!1);const a=!Array.isArray(e);a&&(e=[e]);const c=this.processor.feature_extractor.config.chunk_length/this.model.config.max_source_positions,d=this.processor.feature_extractor.config.hop_length,u=this.processor.feature_extractor.config.sampling_rate,p=await h(e,u),_=[];for(const e of p){let t=[];if(s>0){if(null===o)o=s/6;else if(s<=o)throw Error("`chunk_length_s` must be larger than `stride_length_s`.");const n=u*s,r=u*o,i=n-2*r;let a=0;for(;;){const s=a+n,o=e.subarray(a,s),l=await this.processor(o),c=0===a,d=s>=e.length;if(t.push({stride:[o.length,c?0:r,d?0:r],input_features:l.input_features,is_last:d}),d)break;a+=i}}else t=[{stride:[e.length,0,0],input_features:(await this.processor(e)).input_features,is_last:!0}];for(const e of t){i.num_frames=Math.floor(e.stride[0]/d);const t=await this.model.generate({inputs:e.input_features,...i});"word"===n?(e.tokens=t.sequences.tolist()[0],e.token_timestamps=t.token_timestamps.tolist()[0].map((e=>(0,l.round)(e,2)))):e.tokens=t[0].tolist(),e.stride=e.stride.map((e=>e/u))}const[a,p]=this.tokenizer._decode_asr(t,{time_precision:c,return_timestamps:n,force_full_sequences:r});_.push({text:a,...p})}return a?_[0]:_}}class E extends m{constructor(e){super(e)}async _call(e,t={}){const n=Array.isArray(e),s=await p(e),{pixel_values:r}=await this.processor(s),o=[];for(const e of r){e.dims=[1,...e.dims];const n=await this.model.generate({inputs:e,...t}),s=this.tokenizer.batch_decode(n,{skip_special_tokens:!0}).map((e=>({generated_text:e.trim()})));o.push(s)}return n?o:o[0]}}class z extends m{constructor(e){super(e)}async _call(e,{top_k:t=5}={}){const n=await p(e),{pixel_values:s}=await this.processor(n),r=await this.model({pixel_values:s}),o=this.model.config.id2label,i=[];for(const e of r.logits){const n=await(0,d.topk)(new d.Tensor("float32",(0,l.softmax)(e.data),e.dims),t),s=n[0].tolist(),r=n[1].tolist().map(((e,t)=>({label:o?o[e]:`LABEL_${e}`,score:s[t]})));i.push(r)}return Array.isArray(e)?i:i[0]}}class L extends m{constructor(e){super(e),this.subtasks_mapping={panoptic:"post_process_panoptic_segmentation",instance:"post_process_instance_segmentation",semantic:"post_process_semantic_segmentation"}}async _call(e,{threshold:t=.5,mask_threshold:n=.5,overlap_mask_area_threshold:s=.8,label_ids_to_fuse:r=null,target_sizes:o=null,subtask:i=null}={}){if(Array.isArray(e)&&1!==e.length)throw Error("Image segmentation pipeline currently only supports a batch size of 1.");const a=await p(e),l=a.map((e=>[e.height,e.width])),{pixel_values:c,pixel_mask:d}=await this.processor(a),h=await this.model({pixel_values:c,pixel_mask:d});let _=null;if(null!==i)_=this.subtasks_mapping[i];else for(let[e,t]of Object.entries(this.subtasks_mapping))if(t in this.processor.feature_extractor){_=this.processor.feature_extractor[t].bind(this.processor.feature_extractor),i=e;break}const m=this.model.config.id2label,f=[];if("panoptic"===i||"instance"===i){const e=_(h,t,n,s,r,o??l)[0],i=e.segmentation;for(const t of e.segments_info){const e=new Uint8ClampedArray(i.data.length);for(let n=0;n<i.data.length;++n)i.data[n]===t.id&&(e[n]=255);const n=new u.RawImage(e,i.dims[1],i.dims[0],1);f.push({score:t.score,label:m[t.label_id],mask:n})}}else{if("semantic"!==i)throw Error(`Subtask ${i} not supported.`);{const{segmentation:e,labels:t}=_(h,o??l)[0];for(const n of t){const t=new Uint8ClampedArray(e.data.length);for(let s=0;s<e.data.length;++s)e.data[s]===n&&(t[s]=255);const s=new u.RawImage(t,e.dims[1],e.dims[0],1);f.push({score:null,label:m[n],mask:s})}}}return f}}class I extends m{constructor(e){super(e)}async _call(e,t,{hypothesis_template:n="This is a photo of {}"}={}){const s=Array.isArray(e),r=await p(e),o=t.map((e=>n.replace("{}",e))),i=this.tokenizer(o,{padding:"siglip"!==this.model.config.model_type||"max_length",truncation:!0}),{pixel_values:a}=await this.processor(r),c=await this.model({...i,pixel_values:a}),d="siglip"===this.model.config.model_type?e=>e.sigmoid().data:e=>(0,l.softmax)(e.data),u=[];for(const e of c.logits_per_image){const n=[...d(e)].map(((e,n)=>({score:e,label:t[n]})));n.sort(((e,t)=>t.score-e.score)),u.push(n)}return s?u:u[0]}}class B extends m{constructor(e){super(e)}async _call(e,{threshold:t=.9,percentage:n=!1}={}){const s=Array.isArray(e);if(s&&1!==e.length)throw Error("Object detection pipeline currently only supports a batch size of 1.");const r=await p(e),o=n?null:r.map((e=>[e.height,e.width])),{pixel_values:i,pixel_mask:a}=await this.processor(r),l=await this.model({pixel_values:i,pixel_mask:a}),c=this.processor.feature_extractor.post_process_object_detection(l,t,o),d=this.model.config.id2label,u=c.map((e=>e.boxes.map(((t,s)=>({score:e.scores[s],label:d[e.classes[s]],box:_(t,!n)})))));return s?u:u[0]}}class N extends m{constructor(e){super(e)}async _call(e,t,{threshold:n=.1,top_k:s=null,percentage:r=!1}={}){const o=Array.isArray(e),i=await p(e),a=this.tokenizer(t,{padding:!0,truncation:!0}),l=await this.processor(i),c=[];for(let e=0;e<i.length;++e){const o=i[e],d=r?null:[[o.height,o.width]],u=l.pixel_values[e].unsqueeze_(0),p=await this.model({...a,pixel_values:u}),h=this.processor.feature_extractor.post_process_object_detection(p,n,d,!0)[0];let m=h.boxes.map(((e,n)=>({score:h.scores[n],label:t[h.classes[n]],box:_(e,!r)}))).sort(((e,t)=>t.score-e.score));null!==s&&(m=m.slice(0,s)),c.push(m)}return o?c:c[0]}}class O extends m{constructor(e){super(e)}async _call(e,t,n={}){throw new Error("This pipeline is not yet supported in Transformers.js v3.")}}class j extends m{DEFAULT_VOCODER_ID="Xenova/speecht5_hifigan";constructor(e){super(e),this.vocoder=e.vocoder??null}async _call(e,{speaker_embeddings:t=null}={}){return this.processor?this._call_text_to_spectrogram(e,{speaker_embeddings:t}):this._call_text_to_waveform(e)}async _call_text_to_waveform(e){const t=this.tokenizer(e,{padding:!0,truncation:!0}),{waveform:n}=await this.model(t),s=this.model.config.sampling_rate;return{audio:n.data,sampling_rate:s}}async _call_text_to_spectrogram(e,{speaker_embeddings:t}){if(this.vocoder||(console.log("No vocoder specified, using default HifiGan vocoder."),this.vocoder=await r.AutoModel.from_pretrained(this.DEFAULT_VOCODER_ID,{dtype:"fp32"})),("string"==typeof t||t instanceof URL)&&(t=new Float32Array(await(await fetch(t)).arrayBuffer())),t instanceof Float32Array)t=new d.Tensor("float32",t,[1,t.length]);else if(!(t instanceof d.Tensor))throw new Error("Speaker embeddings must be a `Tensor`, `Float32Array`, `string`, or `URL`.");const{input_ids:n}=this.tokenizer(e,{padding:!0,truncation:!0}),{waveform:s}=await this.model.generate_speech(n,t,{vocoder:this.vocoder}),o=this.processor.feature_extractor.config.sampling_rate;return{audio:s.data,sampling_rate:o}}}class D extends m{constructor(e){super(e)}async _call(e){const t=await p(e),n=await this.processor(t),s=await this.model(n),r=[];for(const e of s.reconstruction){const t=e.squeeze().clamp_(0,1).mul_(255).round_().to("uint8");r.push(u.RawImage.fromTensor(t))}return r.length>1?r:r[0]}}class V extends m{constructor(e){super(e)}async _call(e){const t=await p(e),n=await this.processor(t),{predicted_depth:s}=await this.model(n),r=[];for(let e=0;e<t.length;++e){const n=(0,d.interpolate)(s[e],t[e].size.reverse(),"bilinear",!1),o=n.mul_(255/(0,l.max)(n.data)[0]).to("uint8");r.push({predicted_depth:s[e],depth:u.RawImage.fromTensor(o)})}return r.length>1?r:r[0]}}const R=Object.freeze({"text-classification":{tokenizer:s.AutoTokenizer,pipeline:f,model:r.AutoModelForSequenceClassification,default:{model:"Xenova/distilbert-base-uncased-finetuned-sst-2-english"},type:"text"},"token-classification":{tokenizer:s.AutoTokenizer,pipeline:g,model:r.AutoModelForTokenClassification,default:{model:"Xenova/bert-base-multilingual-cased-ner-hrl"},type:"text"},"question-answering":{tokenizer:s.AutoTokenizer,pipeline:w,model:r.AutoModelForQuestionAnswering,default:{model:"Xenova/distilbert-base-cased-distilled-squad"},type:"text"},"fill-mask":{tokenizer:s.AutoTokenizer,pipeline:M,model:r.AutoModelForMaskedLM,default:{model:"Xenova/bert-base-uncased"},type:"text"},summarization:{tokenizer:s.AutoTokenizer,pipeline:y,model:r.AutoModelForSeq2SeqLM,default:{model:"Xenova/distilbart-cnn-6-6"},type:"text"},translation:{tokenizer:s.AutoTokenizer,pipeline:x,model:r.AutoModelForSeq2SeqLM,default:{model:"Xenova/t5-small"},type:"text"},"text2text-generation":{tokenizer:s.AutoTokenizer,pipeline:b,model:r.AutoModelForSeq2SeqLM,default:{model:"Xenova/flan-t5-small"},type:"text"},"text-generation":{tokenizer:s.AutoTokenizer,pipeline:T,model:r.AutoModelForCausalLM,default:{model:"Xenova/gpt2"},type:"text"},"zero-shot-classification":{tokenizer:s.AutoTokenizer,pipeline:v,model:r.AutoModelForSequenceClassification,default:{model:"Xenova/distilbert-base-uncased-mnli"},type:"text"},"audio-classification":{pipeline:P,model:r.AutoModelForAudioClassification,processor:o.AutoProcessor,default:{model:"Xenova/wav2vec2-base-superb-ks"},type:"audio"},"zero-shot-audio-classification":{tokenizer:s.AutoTokenizer,pipeline:S,model:r.AutoModel,processor:o.AutoProcessor,default:{model:"Xenova/clap-htsat-unfused"},type:"multimodal"},"automatic-speech-recognition":{tokenizer:s.AutoTokenizer,pipeline:A,model:[r.AutoModelForSpeechSeq2Seq,r.AutoModelForCTC],processor:o.AutoProcessor,default:{model:"Xenova/whisper-tiny.en"},type:"multimodal"},"text-to-audio":{tokenizer:s.AutoTokenizer,pipeline:j,model:[r.AutoModelForTextToWaveform,r.AutoModelForTextToSpectrogram],processor:[o.AutoProcessor,null],default:{model:"Xenova/speecht5_tts"},type:"text"},"image-to-text":{tokenizer:s.AutoTokenizer,pipeline:E,model:r.AutoModelForVision2Seq,processor:o.AutoProcessor,default:{model:"Xenova/vit-gpt2-image-captioning"},type:"multimodal"},"image-classification":{pipeline:z,model:r.AutoModelForImageClassification,processor:o.AutoProcessor,default:{model:"Xenova/vit-base-patch16-224"},type:"multimodal"},"image-segmentation":{pipeline:L,model:[r.AutoModelForImageSegmentation,r.AutoModelForSemanticSegmentation],processor:o.AutoProcessor,default:{model:"Xenova/detr-resnet-50-panoptic"},type:"multimodal"},"zero-shot-image-classification":{tokenizer:s.AutoTokenizer,pipeline:I,model:r.AutoModel,processor:o.AutoProcessor,default:{model:"Xenova/clip-vit-base-patch32"},type:"multimodal"},"object-detection":{pipeline:B,model:r.AutoModelForObjectDetection,processor:o.AutoProcessor,default:{model:"Xenova/detr-resnet-50"},type:"multimodal"},"zero-shot-object-detection":{tokenizer:s.AutoTokenizer,pipeline:N,model:r.AutoModelForZeroShotObjectDetection,processor:o.AutoProcessor,default:{model:"Xenova/owlvit-base-patch32"},type:"multimodal"},"document-question-answering":{tokenizer:s.AutoTokenizer,pipeline:O,model:r.AutoModelForDocumentQuestionAnswering,processor:o.AutoProcessor,default:{model:"Xenova/donut-base-finetuned-docvqa"},type:"multimodal"},"image-to-image":{pipeline:D,model:r.AutoModelForImageToImage,processor:o.AutoProcessor,default:{model:"Xenova/swin2SR-classical-sr-x2-64"},type:"image"},"depth-estimation":{pipeline:V,model:r.AutoModelForDepthEstimation,processor:o.AutoProcessor,default:{model:"Xenova/dpt-large"},type:"image"},"feature-extraction":{tokenizer:s.AutoTokenizer,pipeline:C,model:r.AutoModel,default:{model:"Xenova/all-MiniLM-L6-v2"},type:"text"},"image-feature-extraction":{processor:o.AutoProcessor,pipeline:F,model:[r.AutoModelForImageFeatureExtraction,r.AutoModel],default:{model:"Xenova/vit-base-patch16-224-in21k"},type:"image"}}),G=Object.freeze({"sentiment-analysis":"text-classification",ner:"token-classification",asr:"automatic-speech-recognition","text-to-speech":"text-to-audio",embeddings:"feature-extraction"});async function $(e,t=null,{progress_callback:n=null,config:s=null,cache_dir:r=null,local_files_only:o=!1,revision:i="main",device:l=null,dtype:c=null,model_file_name:d=null,session_options:u={}}={}){e=G[e]??e;const p=R[e.split("_",1)[0]];if(!p)throw Error(`Unsupported pipeline: ${e}. Must be one of [${Object.keys(R)}]`);t||(t=p.default.model,console.log(`No model specified. Using default model: "${t}".`));const h={progress_callback:n,config:s,cache_dir:r,local_files_only:o,revision:i,device:l,dtype:c,model_file_name:d,session_options:u},_=new Map([["tokenizer",p.tokenizer],["model",p.model],["processor",p.processor]]),m=await async function(e,t,n){const s=Object.create(null),r=[];for(let[o,i]of e.entries()){if(!i)continue;let e;e=Array.isArray(i)?new Promise((async(e,s)=>{let r;for(let o of i){if(null===o)return void e(null);try{return void e(await o.from_pretrained(t,n))}catch(e){if(e.message?.includes("Unsupported model type"))r=e;else{if(!e.message?.includes("Could not locate file"))return void s(e);r=e}}}s(r)})):i.from_pretrained(t,n),s[o]=e,r.push(e)}await Promise.all(r);for(let[e,t]of Object.entries(s))s[e]=await t;return s}(_,t,h);m.task=e,(0,a.dispatchCallback)(n,{status:"ready",task:e,model:t});return new(0,p.pipeline)(m)}},"./src/processors.js":
 /*!***************************!*\
   !*** ./src/processors.js ***!
-  \***************************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{ASTFeatureExtractor:()=>Z,AutoProcessor:()=>pe,BeitFeatureExtractor:()=>G,BitImageProcessor:()=>y,CLIPFeatureExtractor:()=>k,CLIPImageProcessor:()=>T,ChineseCLIPFeatureExtractor:()=>v,ClapFeatureExtractor:()=>ee,ConvNextFeatureExtractor:()=>F,ConvNextImageProcessor:()=>P,DPTFeatureExtractor:()=>M,DPTImageProcessor:()=>b,DeiTFeatureExtractor:()=>R,DetrFeatureExtractor:()=>U,DonutFeatureExtractor:()=>$,EfficientNetImageProcessor:()=>E,FeatureExtractor:()=>m,Florence2Processor:()=>ue,GLPNFeatureExtractor:()=>x,ImageFeatureExtractor:()=>f,MobileNetV1FeatureExtractor:()=>z,MobileNetV2FeatureExtractor:()=>L,MobileNetV3FeatureExtractor:()=>I,MobileNetV4FeatureExtractor:()=>B,MobileViTFeatureExtractor:()=>N,MobileViTImageProcessor:()=>O,NougatImageProcessor:()=>q,OwlViTFeatureExtractor:()=>j,OwlViTProcessor:()=>de,Owlv2ImageProcessor:()=>D,Processor:()=>re,PyAnnoteFeatureExtractor:()=>te,PyAnnoteProcessor:()=>le,RTDetrImageProcessor:()=>V,SamImageProcessor:()=>X,SamProcessor:()=>oe,SapiensFeatureExtractor:()=>g,SeamlessM4TFeatureExtractor:()=>K,SegformerFeatureExtractor:()=>w,SiglipImageProcessor:()=>C,SpeechT5FeatureExtractor:()=>se,SpeechT5Processor:()=>ce,Swin2SRImageProcessor:()=>Q,ViTFeatureExtractor:()=>S,ViTImageProcessor:()=>A,VitMatteImageProcessor:()=>H,Wav2Vec2FeatureExtractor:()=>J,Wav2Vec2ProcessorWithLM:()=>ae,WeSpeakerFeatureExtractor:()=>ne,WhisperFeatureExtractor:()=>Y,WhisperProcessor:()=>ie,YolosFeatureExtractor:()=>W});var s=n(/*! ./utils/generic.js */"./src/utils/generic.js"),r=n(/*! ./utils/core.js */"./src/utils/core.js"),o=n(/*! ./utils/hub.js */"./src/utils/hub.js"),i=n(/*! ./utils/maths.js */"./src/utils/maths.js"),a=n(/*! ./utils/tensor.js */"./src/utils/tensor.js"),l=(n(/*! ./utils/image.js */"./src/utils/image.js"),n(/*! ./utils/audio.js */"./src/utils/audio.js"));function c([e,t,n,s]){return[e-n/2,t-s/2,e+n/2,t+s/2]}function d(e,t=.5,n=null,s=!1){const r=e.logits,o=e.pred_boxes,[a,l,d]=r.dims;if(null!==n&&n.length!==a)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<a;++e){let a=null!==n?n[e]:null,p={boxes:[],classes:[],scores:[]},h=r[e],_=o[e];for(let e=0;e<l;++e){let n,r=h[e],o=[];if(s){n=r.sigmoid().data;for(let e=0;e<n.length;++e)n[e]>t&&o.push(e)}else{let e=(0,i.max)(r.data)[1];if(e===d-1)continue;if(n=(0,i.softmax)(r.data),n[e]<t)continue;o.push(e)}for(const t of o){let s=_[e].data;s=c(s),null!==a&&(s=s.map(((e,t)=>e*a[(t+1)%2]))),p.boxes.push(s),p.classes.push(t),p.scores.push(n[t])}}u.push(p)}return u}function u(e,t=null){const n=e.logits,s=n.dims[0];if(null!==t&&t.length!==s)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");const r=[];for(let e=0;e<s;++e){const s=null!==t?t[e]:null;let o=n[e];null!==s&&(o=(0,a.interpolate)(o,s,"bilinear",!1));const[i,l]=s??o.dims.slice(-2),c=new a.Tensor("int32",new Int32Array(i*l),[i,l]),d=o[0].data,u=c.data;for(let e=1;e<o.dims[0];++e){const t=o[e].data;for(let n=0;n<t.length;++n)t[n]>d[n]&&(d[n]=t[n],u[n]=e)}const p=new Array(o.dims[0]),h=c.data;for(let e=0;e<h.length;++e){const t=h[e];p[t]=t}const _=p.filter((e=>void 0!==e));r.push({segmentation:c,labels:_})}return r}function p(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 h(e,t,n=0,s=null){const r=e/t;let o=(0,i.bankers_round)(r)*t;return null!==s&&o>s&&(o=Math.floor(r)*t),o<n&&(o=Math.ceil(r)*t),o}function _([e,t],n){return[Math.max(Math.floor(e/n),1)*n,Math.max(Math.floor(t/n),1)*n]}class m extends s.Callable{constructor(e){super(),this.config=e}}class f extends m{constructor(e){super(e),this.image_mean=this.config.image_mean??this.config.mean,this.image_std=this.config.image_std??this.config.std,this.resample=this.config.resample??2,this.do_rescale=this.config.do_rescale??!0,this.rescale_factor=this.config.rescale_factor??1/255,this.do_normalize=this.config.do_normalize,this.do_resize=this.config.do_resize,this.do_thumbnail=this.config.do_thumbnail,this.size=this.config.size,this.size_divisibility=this.config.size_divisibility??this.config.size_divisor,this.do_center_crop=this.config.do_center_crop,this.crop_size=this.config.crop_size,this.do_convert_rgb=this.config.do_convert_rgb??!0,this.do_crop_margin=this.config.do_crop_margin,this.pad_size=this.config.pad_size,this.do_pad=this.config.do_pad,this.do_pad&&!this.pad_size&&this.size&&void 0!==this.size.width&&void 0!==this.size.height&&(this.pad_size=this.size),this.do_flip_channel_order=this.config.do_flip_channel_order??!1}async thumbnail(e,t,n=2){const s=e.height,r=e.width,o=t.height,i=t.width;let a=Math.min(s,o),l=Math.min(r,i);return a===s&&l===r?e:(s>r?l=Math.floor(r*a/s):r>s&&(a=Math.floor(s*l/r)),await e.resize(l,a,{resample:n}))}async crop_margin(e,t=200){const n=e.clone().grayscale(),s=(0,i.min)(n.data)[0],r=(0,i.max)(n.data)[0]-s;if(0===r)return e;const o=t/255;let a=n.width,l=n.height,c=0,d=0;const u=n.data;for(let e=0;e<n.height;++e){const t=e*n.width;for(let i=0;i<n.width;++i)(u[t+i]-s)/r<o&&(a=Math.min(a,i),l=Math.min(l,e),c=Math.max(c,i),d=Math.max(d,e))}return e=await e.crop([a,l,c,d])}pad_image(e,t,n,{mode:s="constant",center:o=!1,constant_values:i=0}={}){const[a,l,c]=t;let d,u;if("number"==typeof n?(d=n,u=n):(d=n.width,u=n.height),d!==l||u!==a){const n=new Float32Array(d*u*c);if(Array.isArray(i))for(let e=0;e<n.length;++e)n[e]=i[e%c];else 0!==i&&n.fill(i);const[p,h]=o?[Math.floor((d-l)/2),Math.floor((u-a)/2)]:[0,0];for(let t=0;t<a;++t){const s=(t+h)*d,r=t*l;for(let t=0;t<l;++t){const o=(s+t+p)*c,i=(r+t)*c;for(let t=0;t<c;++t)n[o+t]=e[i+t]}}if("symmetric"===s){if(o)throw new Error("`center` padding is not supported when `mode` is set to `symmetric`.");const t=a-1,s=l-1;for(let o=0;o<u;++o){const i=o*d,u=(0,r.calculateReflectOffset)(o,t)*l;for(let t=0;t<d;++t){if(o<a&&t<l)continue;const d=(i+t)*c,p=(u+(0,r.calculateReflectOffset)(t,s))*c;for(let t=0;t<c;++t)n[d+t]=e[p+t]}}}e=n,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[n,s]=e.size;let r,o;if(this.do_thumbnail){const{height:e,width:n}=t;r=Math.min(e,n)}else Number.isInteger(t)?(r=t,o=this.config.max_size??r):void 0!==t&&(r=t.shortest_edge,o=t.longest_edge);if(void 0!==r||void 0!==o){const e=void 0===r?1:Math.max(r/n,r/s),t=n*e,i=s*e,a=void 0===o?1:Math.min(o/t,o/i);let l=Math.floor(Number((t*a).toFixed(2))),c=Math.floor(Number((i*a).toFixed(2)));return void 0!==this.size_divisibility&&([l,c]=_([l,c],this.size_divisibility)),[l,c]}if(void 0!==t&&void 0!==t.width&&void 0!==t.height){let e=t.width,r=t.height;if(this.config.keep_aspect_ratio&&this.config.ensure_multiple_of){let t=r/s,o=e/n;Math.abs(1-o)<Math.abs(1-t)?t=o:o=t,r=h(t*s,this.config.ensure_multiple_of),e=h(o*n,this.config.ensure_multiple_of)}return[e,r]}if(void 0!==this.size_divisibility)return _([n,s],this.size_divisibility);throw new Error(`Could not resize image due to unsupported \`this.size\` option in config: ${JSON.stringify(t)}`)}async resize(e){const[t,n]=this.get_resize_output_image_size(e,this.size);return await e.resize(t,n,{resample:this.resample})}async preprocess(e,{do_normalize:t=null,do_pad:n=null,do_convert_rgb:s=null,do_convert_grayscale:r=null,do_flip_channel_order:o=null}={}){this.do_crop_margin&&(e=await this.crop_margin(e));const[i,l]=e.size;if(s??this.do_convert_rgb?e=e.rgb():r&&(e=e.grayscale()),this.do_resize&&(e=await this.resize(e)),this.do_thumbnail&&(e=await this.thumbnail(e,this.size,this.resample)),this.do_center_crop){let t,n;Number.isInteger(this.crop_size)?(t=this.crop_size,n=this.crop_size):(t=this.crop_size.width,n=this.crop_size.height),e=await e.center_crop(t,n)}const c=[e.height,e.width];let d=Float32Array.from(e.data),u=[e.height,e.width,e.channels];if(this.do_rescale&&this.rescale(d),t??this.do_normalize){let t=this.image_mean;Array.isArray(this.image_mean)||(t=new Array(e.channels).fill(t));let n=this.image_std;if(Array.isArray(this.image_std)||(n=new Array(e.channels).fill(t)),t.length!==e.channels||n.length!==e.channels)throw new Error(`When set to arrays, the length of \`image_mean\` (${t.length}) and \`image_std\` (${n.length}) must match the number of channels in the image (${e.channels}).`);for(let s=0;s<d.length;s+=e.channels)for(let r=0;r<e.channels;++r)d[s+r]=(d[s+r]-t[r])/n[r]}if(n??this.do_pad)if(this.pad_size){const t=this.pad_image(d,[e.height,e.width,e.channels],this.pad_size);[d,u]=t}else if(this.size_divisibility){const[e,t]=_([u[1],u[0]],this.size_divisibility);[d,u]=this.pad_image(d,u,{width:e,height:t})}if(o??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,i],reshaped_input_size:c,pixel_values:new a.Tensor("float32",d,u).permute(2,0,1)}}async _call(e,...t){Array.isArray(e)||(e=[e]);const n=await Promise.all(e.map((e=>this.preprocess(e))));return{pixel_values:(0,a.stack)(n.map((e=>e.pixel_values)),0),original_sizes:n.map((e=>e.original_size)),reshaped_input_sizes:n.map((e=>e.reshaped_input_size))}}}class g extends f{post_process_semantic_segmentation(...e){return u(...e)}}class w extends f{post_process_semantic_segmentation(...e){return u(...e)}}class M extends f{}class b extends M{}class y extends f{}class x extends f{}class k extends f{}class T extends k{}class v extends f{}class C extends f{}class F extends f{constructor(e){super(e),this.crop_pct=this.config.crop_pct??.875}async resize(e){const t=this.size?.shortest_edge;if(void 0===t)throw new Error("Size dictionary must contain 'shortest_edge' key.");if(t<384){const n=Math.floor(t/this.crop_pct),[s,r]=this.get_resize_output_image_size(e,{shortest_edge:n});e=await e.resize(s,r,{resample:this.resample}),e=await e.center_crop(t,t)}else e=await e.resize(t,t,{resample:this.resample});return e}}class P extends F{}class S extends f{}class A extends f{}class E extends f{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 z extends f{}class L extends f{}class I extends f{}class B extends f{}class N extends f{}class O extends N{}class j extends f{post_process_object_detection(...e){return d(...e)}}class D extends j{}class V extends f{post_process_object_detection(...e){return d(...e)}}class R extends f{}class G extends f{}class $ extends f{pad_image(e,t,n,s={}){const[r,o,i]=t;let a=this.image_mean;Array.isArray(this.image_mean)||(a=new Array(i).fill(a));let l=this.image_std;Array.isArray(l)||(l=new Array(i).fill(a));const c=a.map(((e,t)=>-e/l[t]));return super.pad_image(e,t,n,{center:!0,constant_values:c,...s})}}class q extends ${}class U extends f{async _call(e){const t=await super._call(e),n=[t.pixel_values.dims[0],64,64],s=new a.Tensor("int64",new BigInt64Array(n.reduce(((e,t)=>e*t))).fill(1n),n);return{...t,pixel_mask:s}}post_process_object_detection(...e){return d(...e)}remove_low_and_no_objects(e,t,n,s){let r=[],o=[],a=[];for(let l=0;l<e.dims[0];++l){let c=e[l],d=t[l],u=(0,i.max)(c.data)[1];if(u===s)continue;let p=(0,i.softmax)(c.data)[u];p>n&&(r.push(d),o.push(p),a.push(u))}return[r,o,a]}check_segment_validity(e,t,n,s=.5,r=.8){let o=[],i=0,a=0;const l=t[n].data;for(let t=0;t<e.length;++t)e[t]===n&&(o.push(t),++i),l[t]>=s&&++a;let c=i>0&&a>0;if(c){c=i/a>r}return[c,o]}compute_segments(e,t,n,s,r,o=null,i=null){let[l,c]=i??e[0].dims,d=new a.Tensor("int32",new Int32Array(l*c),[l,c]),u=[];if(null!==i)for(let t=0;t<e.length;++t)e[t]=(0,a.interpolate)(e[t],i,"bilinear",!1);let p=new Int32Array(e[0].data.length),h=new Float32Array(e[0].data.length);for(let n=0;n<e.length;++n){let s=t[n];const r=e[n].data;for(let e=0;e<r.length;++e)r[e]*=s,r[e]>h[e]&&(p[e]=n,h[e]=r[e])}let _=0;const m=d.data;for(let o=0;o<n.length;++o){let i=n[o],[a,l]=this.check_segment_validity(p,e,o,s,r);if(a){++_;for(let e of l)m[e]=_;u.push({id:_,label_id:i,score:t[o]})}}return[d,u]}post_process_panoptic_segmentation(e,t=.5,n=.5,s=.8,r=null,o=null){null===r&&(console.warn("`label_ids_to_fuse` unset. No instance will be fused."),r=new Set);const i=e.logits,l=e.pred_masks.sigmoid();let[c,d,u]=i.dims;if(u-=1,null!==o&&o.length!==c)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");let p=[];for(let e=0;e<c;++e){let c=null!==o?o[e]:null,d=i[e],h=l[e],[_,m,f]=this.remove_low_and_no_objects(d,h,t,u);if(0===f.length){let[e,t]=c??h.dims.slice(-2),n=new a.Tensor("int32",new Int32Array(e*t).fill(-1),[e,t]);p.push({segmentation:n,segments_info:[]});continue}let[g,w]=this.compute_segments(_,m,f,n,s,r,c);p.push({segmentation:g,segments_info:w})}return p}post_process_instance_segmentation(){throw Error("Not implemented yet")}}class W extends f{post_process_object_detection(...e){return d(...e)}}class X extends f{reshape_input_points(e,t,n,s=!1){e=structuredClone(e);let o=(0,r.calculateDimensions)(e);if(3===o.length)s||(o=[1,...o]),e=[e];else if(4!==o.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 s=0;s<e.length;++s){let r=t[s],o=n[s],i=[o[0]/r[0],o[1]/r[1]];for(let t=0;t<e[s].length;++t)for(let n=0;n<e[s][t].length;++n)for(let r=0;r<e[s][t][n].length;++r)e[s][t][n][r]*=i[r%2]}return new a.Tensor("float32",Float32Array.from(e.flat(1/0)),o)}add_input_labels(e,t){let n=(0,r.calculateDimensions)(e);if(2===n.length)n=[1,...n],e=[e];else if(3!==n.length)throw Error("The input_points must be a 4D tensor of shape `batch_size`, `point_batch_size`, `nb_points_per_image`, `2`.");if(n.some(((e,n)=>e!==t.dims[n])))throw Error(`The first ${n.length} dimensions of 'input_points' and 'input_labels' must be the same.`);return new a.Tensor("int64",e.flat(1/0).map(BigInt),n)}async _call(e,{input_points:t=null,input_labels:n=null,input_boxes:s=null}={}){const r=await super._call(e);if(t&&(r.input_points=this.reshape_input_points(t,r.original_sizes,r.reshaped_input_sizes)),n){if(!r.input_points)throw Error("`input_points` must be provided if `input_labels` are provided.");r.input_labels=this.add_input_labels(n,r.input_points)}return s&&(r.input_boxes=this.reshape_input_points(s,r.original_sizes,r.reshaped_input_sizes,!0)),r}async post_process_masks(e,t,n,{mask_threshold:s=0,binarize:r=!0,pad_size:o=null}={}){const i=[],l=[(o=o??this.pad_size).height,o.width];for(let o=0;o<t.length;++o){const c=t[o],d=n[o];let u=await(0,a.interpolate_4d)(e[o],{mode:"bilinear",size:l});if(u=u.slice(null,null,[0,d[0]],[0,d[1]]),u=await(0,a.interpolate_4d)(u,{mode:"bilinear",size:c}),r){const e=u.data,t=new Uint8Array(e.length);for(let n=0;n<e.length;++n)e[n]>s&&(t[n]=1);u=new a.Tensor("bool",t,u.dims)}i.push(u)}return i}generate_crop_boxes(e,t,{crop_n_layers:n=0,overlap_ratio:s=512/1500,points_per_crop:r=32,crop_n_points_downscale_factor:o=1}={}){}}class Q extends f{pad_image(e,t,n,s={}){const[r,o,i]=t;return super.pad_image(e,t,{width:o+(n-o%n)%n,height:r+(n-r%n)%n},{mode:"symmetric",center:!1,constant_values:-1,...s})}}class H extends f{async _call(e,t){Array.isArray(e)||(e=[e]),Array.isArray(t)||(t=[t]);const n=await Promise.all(e.map((e=>this.preprocess(e)))),s=await Promise.all(t.map((e=>this.preprocess(e,{do_normalize:!1,do_convert_rgb:!1,do_convert_grayscale:!0})))),r=(0,a.stack)(n.map(((e,t)=>(0,a.cat)([e.pixel_values,s[t].pixel_values],0))),0);return{pixel_values:r,original_sizes:n.map((e=>e.original_size)),reshaped_input_sizes:n.map((e=>e.reshaped_input_size))}}}class Y extends m{constructor(e){super(e),this.config.mel_filters??=(0,l.mel_filter_bank)(Math.floor(1+this.config.n_fft/2),this.config.feature_size,0,8e3,this.config.sampling_rate,"slaney","slaney"),this.window=(0,l.window_function)(this.config.n_fft,"hann")}async _extract_fbank_features(e){const t=await(0,l.spectrogram)(e,this.window,this.config.n_fft,this.config.hop_length,{power:2,mel_filters:this.config.mel_filters,log_mel:"log10",max_num_frames:this.config.nb_max_frames}),n=t.data,s=(0,i.max)(n)[0];for(let e=0;e<n.length;++e)n[e]=(Math.max(n[e],s-8)+4)/4;return t}async _call(e){let t;p(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 J extends m{_zero_mean_unit_var_norm(e){const t=e.reduce(((e,t)=>e+t),0)/e.length,n=e.reduce(((e,n)=>e+(n-t)**2),0)/e.length;return e.map((e=>(e-t)/Math.sqrt(n+1e-7)))}async _call(e){p(e,"Wav2Vec2FeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));let t=e;this.config.do_normalize&&(t=this._zero_mean_unit_var_norm(t));const n=[1,t.length];return{input_values:new a.Tensor("float32",t,n),attention_mask:new a.Tensor("int64",new BigInt64Array(t.length).fill(1n),n)}}}class K extends m{constructor(e){super(e);const t=this.config.sampling_rate,n=(0,l.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<n.length;++e)n[e].push(0);this.mel_filters=n,this.window=(0,l.window_function)(400,"povey",{periodic:!1})}async _extract_fbank_features(e,t){return e=e.map((e=>32768*e)),(0,l.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,max_num_frames:t,transpose:!0})}async _call(e,{padding:t=!0,pad_to_multiple_of:n=2,do_normalize_per_mel_bins:s=!0,return_attention_mask:r=!0}={}){p(e,"SeamlessM4TFeatureExtractor");let o,i=await this._extract_fbank_features(e,this.config.max_length);if(s){const[e,t]=i.dims,n=i.data;for(let s=0;s<t;++s){let r=0;for(let o=0;o<e;++o)r+=n[o*t+s];const o=r/e;let i=0;for(let r=0;r<e;++r)i+=(n[r*t+s]-o)**2;i/=e-1;const a=Math.sqrt(i+1e-7);for(let r=0;r<e;++r){const e=r*t+s;n[e]=(n[e]-o)/a}}}if(t){const[e,t]=i.dims,s=i.data,l=e%n;if(l>0){const n=new Float32Array(t*(e+l));n.set(s),n.fill(this.config.padding_value,s.length);const c=e+l;i=new a.Tensor(i.type,n,[c,t]),r&&(o=new a.Tensor("int64",new BigInt64Array(c),[1,c]),o.data.fill(1n,0,e))}}const[l,c]=i.dims,d=this.config.stride;if(0!==l%d)throw new Error(`The number of frames (${l}) must be a multiple of the stride (${d}).`);const u=i.view(1,Math.floor(l/d),c*d),h={input_features:u};if(r){const e=u.dims[1],t=new BigInt64Array(e);if(o){const e=o.data;for(let n=1,s=0;n<l;n+=d,++s)t[s]=e[n]}else t.fill(1n);h.attention_mask=new a.Tensor("int64",t,[1,e])}return h}}class Z extends m{constructor(e){super(e);const t=this.config.sampling_rate,n=(0,l.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<n.length;++e)n[e].push(0);this.mel_filters=n,this.window=(0,l.window_function)(400,"hann",{periodic:!1}),this.mean=this.config.mean,this.std=this.config.std}async _extract_fbank_features(e,t){return(0,l.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,max_num_frames:t,transpose:!0})}async _call(e){p(e,"ASTFeatureExtractor");const t=await this._extract_fbank_features(e,this.config.max_length);if(this.config.do_normalize){const e=2*this.std,n=t.data;for(let t=0;t<n.length;++t)n[t]=(n[t]-this.mean)/e}return{input_values:t.unsqueeze_(0)}}}class ee extends m{constructor(e){super(e),this.mel_filters=(0,l.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,null,"htk"),this.mel_filters_slaney=(0,l.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,"slaney","slaney"),this.window=(0,l.window_function)(this.config.fft_window_size,"hann")}async _get_input_mel(e,t,n,s){let r,o=!1;const i=e.length-t;if(i>0){if("rand_trunc"!==n)throw new Error(`Truncation strategy "${n}" not implemented`);{o=!0;const n=Math.floor(Math.random()*(i+1));e=e.subarray(n,n+t),r=await this._extract_fbank_features(e,this.mel_filters_slaney,this.config.nb_max_samples)}}else{if(i<0){let n=new Float64Array(t);if(n.set(e),"repeat"===s)for(let s=e.length;s<t;s+=e.length)n.set(e.subarray(0,Math.min(e.length,t-s)),s);else if("repeatpad"===s)for(let t=e.length;t<-i;t+=e.length)n.set(e,t);e=n}if("fusion"===n)throw new Error(`Truncation strategy "${n}" not implemented`);r=await this._extract_fbank_features(e,this.mel_filters_slaney,this.config.nb_max_samples)}return r.unsqueeze_(0)}async _extract_fbank_features(e,t,n=null){return(0,l.spectrogram)(e,this.window,this.config.fft_window_size,this.config.hop_length,{power:2,mel_filters:t,log_mel:"dB",max_num_frames:n,do_pad:!1,transpose:!0})}async _call(e,{max_length:t=null}={}){p(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 te extends m{async _call(e){p(e,"PyAnnoteFeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));const t=[1,1,e.length];return{input_values:new a.Tensor("float32",e,t)}}samples_to_frames(e){return(e-this.config.offset)/this.config.step}post_process_speaker_diarization(e,t){const n=t/this.samples_to_frames(t)/this.config.sampling_rate,s=[];for(const t of e.tolist()){const e=[];let r=-1;for(let n=0;n<t.length;++n){const s=(0,i.softmax)(t[n]),[o,a]=(0,i.max)(s),[l,c]=[n,n+1];a!==r?(r=a,e.push({id:a,start:l,end:c,score:o})):(e.at(-1).end=c,e.at(-1).score+=o)}s.push(e.map((({id:e,start:t,end:s,score:r})=>({id:e,start:t*n,end:s*n,confidence:r/(s-t)}))))}return s}}class ne extends m{constructor(e){super(e);const t=this.config.sampling_rate,n=(0,l.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<n.length;++e)n[e].push(0);this.mel_filters=n,this.window=(0,l.window_function)(400,"hamming",{periodic:!1}),this.min_num_frames=this.config.min_num_frames}async _extract_fbank_features(e){return e=e.map((e=>32768*e)),(0,l.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,transpose:!0,min_num_frames:this.min_num_frames})}async _call(e){p(e,"WeSpeakerFeatureExtractor");const t=(await this._extract_fbank_features(e)).unsqueeze_(0);if(null===this.config.fbank_centering_span){const e=t.mean(1).data,n=t.data,[s,r,o]=t.dims;for(let t=0;t<s;++t){const s=t*r*o,i=t*o;for(let t=0;t<r;++t){const r=s+t*o;for(let t=0;t<o;++t)n[r+t]-=e[i+t]}}}return{input_features:t}}}class se extends m{}class re extends s.Callable{constructor(e){super(),this.feature_extractor=e}async _call(e,...t){return await this.feature_extractor(e,...t)}}class oe extends re{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 ie extends re{async _call(e){return await this.feature_extractor(e)}}class ae extends re{async _call(e){return await this.feature_extractor(e)}}class le extends re{async _call(e){return await this.feature_extractor(e)}post_process_speaker_diarization(...e){return this.feature_extractor.post_process_speaker_diarization(...e)}}class ce extends re{async _call(e){return await this.feature_extractor(e)}}class de extends re{}class ue extends re{constructor(e){super(e);const{tasks_answer_post_processing_type:t,task_prompts_without_inputs:n,task_prompts_with_input:s}=e.config;this.tasks_answer_post_processing_type=new Map(Object.entries(t??{})),this.task_prompts_without_inputs=new Map(Object.entries(n??{})),this.task_prompts_with_input=new Map(Object.entries(s??{})),this.regexes={quad_boxes:/(.+?)<loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)>/gm,bboxes:/([^<]+)?<loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)>/gm},this.size_per_bin=1e3}construct_prompts(e){"string"==typeof e&&(e=[e]);const t=[];for(const n of e)if(this.task_prompts_without_inputs.has(n))t.push(this.task_prompts_without_inputs.get(n));else{for(const[e,s]of this.task_prompts_with_input)if(n.includes(e)){t.push(s.replaceAll("{input}",n).replaceAll(e,""));break}t.length!==e.length&&t.push(n)}return t}post_process_generation(e,t,n){const s=this.tasks_answer_post_processing_type.get(t)??"pure_text";let r;switch(e=e.replaceAll("<s>","").replaceAll("</s>",""),s){case"pure_text":r=e;break;case"description_with_bboxes":case"bboxes":case"phrase_grounding":case"ocr":const o="ocr"===s?"quad_boxes":"bboxes",i=e.matchAll(this.regexes[o]),a=[],l=[];for(const[e,t,...s]of i)a.push(t?t.trim():a.at(-1)??""),l.push(s.map(((e,t)=>(Number(e)+.5)/this.size_per_bin*n[t%2])));r={labels:a,[o]:l};break;default:throw new Error(`Task "${t}" (of type "${s}") not yet implemented.`)}return{[t]:r}}}class pe{static FEATURE_EXTRACTOR_CLASS_MAPPING={ImageFeatureExtractor:f,WhisperFeatureExtractor:Y,ViTFeatureExtractor:S,MobileViTFeatureExtractor:N,MobileViTImageProcessor:O,MobileNetV1FeatureExtractor:z,MobileNetV2FeatureExtractor:L,MobileNetV3FeatureExtractor:I,MobileNetV4FeatureExtractor:B,OwlViTFeatureExtractor:j,Owlv2ImageProcessor:D,CLIPFeatureExtractor:k,CLIPImageProcessor:T,Florence2Processor:ue,ChineseCLIPFeatureExtractor:v,SiglipImageProcessor:C,ConvNextFeatureExtractor:F,ConvNextImageProcessor:P,SegformerFeatureExtractor:w,SapiensFeatureExtractor:g,BitImageProcessor:y,DPTImageProcessor:b,DPTFeatureExtractor:M,GLPNFeatureExtractor:x,BeitFeatureExtractor:G,DeiTFeatureExtractor:R,DetrFeatureExtractor:U,RTDetrImageProcessor:V,YolosFeatureExtractor:W,DonutFeatureExtractor:$,NougatImageProcessor:q,EfficientNetImageProcessor:E,ViTImageProcessor:A,VitMatteImageProcessor:H,SamImageProcessor:X,Swin2SRImageProcessor:Q,Wav2Vec2FeatureExtractor:J,SeamlessM4TFeatureExtractor:K,SpeechT5FeatureExtractor:se,ASTFeatureExtractor:Z,ClapFeatureExtractor:ee,PyAnnoteFeatureExtractor:te,WeSpeakerFeatureExtractor:ne};static PROCESSOR_CLASS_MAPPING={WhisperProcessor:ie,Wav2Vec2ProcessorWithLM:ae,PyAnnoteProcessor:le,SamProcessor:oe,SpeechT5Processor:ce,OwlViTProcessor:de,Florence2Processor:ue};static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:s=null,local_files_only:r=!1,revision:i="main"}={}){let a=n??await(0,o.getModelJSON)(e,"preprocessor_config.json",!0,{progress_callback:t,config:n,cache_dir:s,local_files_only:r,revision:i}),l=a.feature_extractor_type??a.image_processor_type,c=this.FEATURE_EXTRACTOR_CLASS_MAPPING[l];if(!c){if(void 0===a.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=f}return new(this.PROCESSOR_CLASS_MAPPING[a.processor_class]??re)(new c(a))}}},"./src/tokenizers.js":
+  \***************************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{ASTFeatureExtractor:()=>Z,AutoProcessor:()=>pe,BeitFeatureExtractor:()=>G,BitImageProcessor:()=>y,CLIPFeatureExtractor:()=>k,CLIPImageProcessor:()=>T,ChineseCLIPFeatureExtractor:()=>v,ClapFeatureExtractor:()=>ee,ConvNextFeatureExtractor:()=>F,ConvNextImageProcessor:()=>P,DPTFeatureExtractor:()=>M,DPTImageProcessor:()=>b,DeiTFeatureExtractor:()=>R,DetrFeatureExtractor:()=>U,DonutFeatureExtractor:()=>$,EfficientNetImageProcessor:()=>E,FeatureExtractor:()=>m,Florence2Processor:()=>ue,GLPNFeatureExtractor:()=>x,ImageFeatureExtractor:()=>f,MobileNetV1FeatureExtractor:()=>z,MobileNetV2FeatureExtractor:()=>L,MobileNetV3FeatureExtractor:()=>I,MobileNetV4FeatureExtractor:()=>B,MobileViTFeatureExtractor:()=>N,MobileViTImageProcessor:()=>O,NougatImageProcessor:()=>q,OwlViTFeatureExtractor:()=>j,OwlViTProcessor:()=>de,Owlv2ImageProcessor:()=>D,Processor:()=>re,PyAnnoteFeatureExtractor:()=>te,PyAnnoteProcessor:()=>le,RTDetrImageProcessor:()=>V,SamImageProcessor:()=>X,SamProcessor:()=>oe,SapiensFeatureExtractor:()=>g,SeamlessM4TFeatureExtractor:()=>K,SegformerFeatureExtractor:()=>w,SiglipImageProcessor:()=>C,SpeechT5FeatureExtractor:()=>se,SpeechT5Processor:()=>ce,Swin2SRImageProcessor:()=>Q,ViTFeatureExtractor:()=>S,ViTImageProcessor:()=>A,VitMatteImageProcessor:()=>H,Wav2Vec2FeatureExtractor:()=>J,Wav2Vec2ProcessorWithLM:()=>ae,WeSpeakerFeatureExtractor:()=>ne,WhisperFeatureExtractor:()=>Y,WhisperProcessor:()=>ie,YolosFeatureExtractor:()=>W});var s=n(/*! ./utils/generic.js */"./src/utils/generic.js"),r=n(/*! ./utils/core.js */"./src/utils/core.js"),o=n(/*! ./utils/hub.js */"./src/utils/hub.js"),i=n(/*! ./utils/maths.js */"./src/utils/maths.js"),a=n(/*! ./utils/tensor.js */"./src/utils/tensor.js"),l=(n(/*! ./utils/image.js */"./src/utils/image.js"),n(/*! ./utils/audio.js */"./src/utils/audio.js"));function c([e,t,n,s]){return[e-n/2,t-s/2,e+n/2,t+s/2]}function d(e,t=.5,n=null,s=!1){const r=e.logits,o=e.pred_boxes,[a,l,d]=r.dims;if(null!==n&&n.length!==a)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<a;++e){let a=null!==n?n[e]:null,p={boxes:[],classes:[],scores:[]},h=r[e],_=o[e];for(let e=0;e<l;++e){let n,r=h[e],o=[];if(s){n=r.sigmoid().data;for(let e=0;e<n.length;++e)n[e]>t&&o.push(e)}else{let e=(0,i.max)(r.data)[1];if(e===d-1)continue;if(n=(0,i.softmax)(r.data),n[e]<t)continue;o.push(e)}for(const t of o){let s=_[e].data;s=c(s),null!==a&&(s=s.map(((e,t)=>e*a[(t+1)%2]))),p.boxes.push(s),p.classes.push(t),p.scores.push(n[t])}}u.push(p)}return u}function u(e,t=null){const n=e.logits,s=n.dims[0];if(null!==t&&t.length!==s)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");const r=[];for(let e=0;e<s;++e){const s=null!==t?t[e]:null;let o=n[e];null!==s&&(o=(0,a.interpolate)(o,s,"bilinear",!1));const[i,l]=s??o.dims.slice(-2),c=new a.Tensor("int32",new Int32Array(i*l),[i,l]),d=o[0].data,u=c.data;for(let e=1;e<o.dims[0];++e){const t=o[e].data;for(let n=0;n<t.length;++n)t[n]>d[n]&&(d[n]=t[n],u[n]=e)}const p=new Array(o.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));r.push({segmentation:c,labels:h})}return r}function p(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 h(e,t,n=0,s=null){const r=e/t;let o=(0,i.bankers_round)(r)*t;return null!==s&&o>s&&(o=Math.floor(r)*t),o<n&&(o=Math.ceil(r)*t),o}function _([e,t],n){return[Math.max(Math.floor(e/n),1)*n,Math.max(Math.floor(t/n),1)*n]}class m extends s.Callable{constructor(e){super(),this.config=e}}class f extends m{constructor(e){super(e),this.image_mean=this.config.image_mean??this.config.mean,this.image_std=this.config.image_std??this.config.std,this.resample=this.config.resample??2,this.do_rescale=this.config.do_rescale??!0,this.rescale_factor=this.config.rescale_factor??1/255,this.do_normalize=this.config.do_normalize,this.do_resize=this.config.do_resize,this.do_thumbnail=this.config.do_thumbnail,this.size=this.config.size,this.size_divisibility=this.config.size_divisibility??this.config.size_divisor,this.do_center_crop=this.config.do_center_crop,this.crop_size=this.config.crop_size,this.do_convert_rgb=this.config.do_convert_rgb??!0,this.do_crop_margin=this.config.do_crop_margin,this.pad_size=this.config.pad_size,this.do_pad=this.config.do_pad,this.do_pad&&!this.pad_size&&this.size&&void 0!==this.size.width&&void 0!==this.size.height&&(this.pad_size=this.size),this.do_flip_channel_order=this.config.do_flip_channel_order??!1}async thumbnail(e,t,n=2){const s=e.height,r=e.width,o=t.height,i=t.width;let a=Math.min(s,o),l=Math.min(r,i);return a===s&&l===r?e:(s>r?l=Math.floor(r*a/s):r>s&&(a=Math.floor(s*l/r)),await e.resize(l,a,{resample:n}))}async crop_margin(e,t=200){const n=e.clone().grayscale(),s=(0,i.min)(n.data)[0],r=(0,i.max)(n.data)[0]-s;if(0===r)return e;const o=t/255;let a=n.width,l=n.height,c=0,d=0;const u=n.data;for(let e=0;e<n.height;++e){const t=e*n.width;for(let i=0;i<n.width;++i)(u[t+i]-s)/r<o&&(a=Math.min(a,i),l=Math.min(l,e),c=Math.max(c,i),d=Math.max(d,e))}return e=await e.crop([a,l,c,d])}pad_image(e,t,n,{mode:s="constant",center:o=!1,constant_values:i=0}={}){const[a,l,c]=t;let d,u;if("number"==typeof n?(d=n,u=n):(d=n.width,u=n.height),d!==l||u!==a){const n=new Float32Array(d*u*c);if(Array.isArray(i))for(let e=0;e<n.length;++e)n[e]=i[e%c];else 0!==i&&n.fill(i);const[p,h]=o?[Math.floor((d-l)/2),Math.floor((u-a)/2)]:[0,0];for(let t=0;t<a;++t){const s=(t+h)*d,r=t*l;for(let t=0;t<l;++t){const o=(s+t+p)*c,i=(r+t)*c;for(let t=0;t<c;++t)n[o+t]=e[i+t]}}if("symmetric"===s){if(o)throw new Error("`center` padding is not supported when `mode` is set to `symmetric`.");const t=a-1,s=l-1;for(let o=0;o<u;++o){const i=o*d,u=(0,r.calculateReflectOffset)(o,t)*l;for(let t=0;t<d;++t){if(o<a&&t<l)continue;const d=(i+t)*c,p=(u+(0,r.calculateReflectOffset)(t,s))*c;for(let t=0;t<c;++t)n[d+t]=e[p+t]}}}e=n,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[n,s]=e.size;let r,o;if(this.do_thumbnail){const{height:e,width:n}=t;r=Math.min(e,n)}else Number.isInteger(t)?(r=t,o=this.config.max_size??r):void 0!==t&&(r=t.shortest_edge,o=t.longest_edge);if(void 0!==r||void 0!==o){const e=void 0===r?1:Math.max(r/n,r/s),t=n*e,i=s*e,a=void 0===o?1:Math.min(o/t,o/i);let l=Math.floor(Number((t*a).toFixed(2))),c=Math.floor(Number((i*a).toFixed(2)));return void 0!==this.size_divisibility&&([l,c]=_([l,c],this.size_divisibility)),[l,c]}if(void 0!==t&&void 0!==t.width&&void 0!==t.height){let e=t.width,r=t.height;if(this.config.keep_aspect_ratio&&this.config.ensure_multiple_of){let t=r/s,o=e/n;Math.abs(1-o)<Math.abs(1-t)?t=o:o=t,r=h(t*s,this.config.ensure_multiple_of),e=h(o*n,this.config.ensure_multiple_of)}return[e,r]}if(void 0!==this.size_divisibility)return _([n,s],this.size_divisibility);throw new Error(`Could not resize image due to unsupported \`this.size\` option in config: ${JSON.stringify(t)}`)}async resize(e){const[t,n]=this.get_resize_output_image_size(e,this.size);return await e.resize(t,n,{resample:this.resample})}async preprocess(e,{do_normalize:t=null,do_pad:n=null,do_convert_rgb:s=null,do_convert_grayscale:r=null,do_flip_channel_order:o=null}={}){this.do_crop_margin&&(e=await this.crop_margin(e));const[i,l]=e.size;if(s??this.do_convert_rgb?e=e.rgb():r&&(e=e.grayscale()),this.do_resize&&(e=await this.resize(e)),this.do_thumbnail&&(e=await this.thumbnail(e,this.size,this.resample)),this.do_center_crop){let t,n;Number.isInteger(this.crop_size)?(t=this.crop_size,n=this.crop_size):(t=this.crop_size.width,n=this.crop_size.height),e=await e.center_crop(t,n)}const c=[e.height,e.width];let d=Float32Array.from(e.data),u=[e.height,e.width,e.channels];if(this.do_rescale&&this.rescale(d),t??this.do_normalize){let t=this.image_mean;Array.isArray(this.image_mean)||(t=new Array(e.channels).fill(t));let n=this.image_std;if(Array.isArray(this.image_std)||(n=new Array(e.channels).fill(t)),t.length!==e.channels||n.length!==e.channels)throw new Error(`When set to arrays, the length of \`image_mean\` (${t.length}) and \`image_std\` (${n.length}) must match the number of channels in the image (${e.channels}).`);for(let s=0;s<d.length;s+=e.channels)for(let r=0;r<e.channels;++r)d[s+r]=(d[s+r]-t[r])/n[r]}if(n??this.do_pad)if(this.pad_size){const t=this.pad_image(d,[e.height,e.width,e.channels],this.pad_size);[d,u]=t}else if(this.size_divisibility){const[e,t]=_([u[1],u[0]],this.size_divisibility);[d,u]=this.pad_image(d,u,{width:e,height:t})}if(o??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,i],reshaped_input_size:c,pixel_values:new a.Tensor("float32",d,u).permute(2,0,1)}}async _call(e,...t){Array.isArray(e)||(e=[e]);const n=await Promise.all(e.map((e=>this.preprocess(e))));return{pixel_values:(0,a.stack)(n.map((e=>e.pixel_values)),0),original_sizes:n.map((e=>e.original_size)),reshaped_input_sizes:n.map((e=>e.reshaped_input_size))}}}class g extends f{post_process_semantic_segmentation(...e){return u(...e)}}class w extends f{post_process_semantic_segmentation(...e){return u(...e)}}class M extends f{}class b extends M{}class y extends f{}class x extends f{}class k extends f{}class T extends k{}class v extends f{}class C extends f{}class F extends f{constructor(e){super(e),this.crop_pct=this.config.crop_pct??.875}async resize(e){const t=this.size?.shortest_edge;if(void 0===t)throw new Error("Size dictionary must contain 'shortest_edge' key.");if(t<384){const n=Math.floor(t/this.crop_pct),[s,r]=this.get_resize_output_image_size(e,{shortest_edge:n});e=await e.resize(s,r,{resample:this.resample}),e=await e.center_crop(t,t)}else e=await e.resize(t,t,{resample:this.resample});return e}}class P extends F{}class S extends f{}class A extends f{}class E extends f{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 z extends f{}class L extends f{}class I extends f{}class B extends f{}class N extends f{}class O extends N{}class j extends f{post_process_object_detection(...e){return d(...e)}}class D extends j{}class V extends f{post_process_object_detection(...e){return d(...e)}}class R extends f{}class G extends f{}class $ extends f{pad_image(e,t,n,s={}){const[r,o,i]=t;let a=this.image_mean;Array.isArray(this.image_mean)||(a=new Array(i).fill(a));let l=this.image_std;Array.isArray(l)||(l=new Array(i).fill(a));const c=a.map(((e,t)=>-e/l[t]));return super.pad_image(e,t,n,{center:!0,constant_values:c,...s})}}class q extends ${}class U extends f{async _call(e){const t=await super._call(e),n=[t.pixel_values.dims[0],64,64],s=new a.Tensor("int64",new BigInt64Array(n.reduce(((e,t)=>e*t))).fill(1n),n);return{...t,pixel_mask:s}}post_process_object_detection(...e){return d(...e)}remove_low_and_no_objects(e,t,n,s){let r=[],o=[],a=[];for(let l=0;l<e.dims[0];++l){let c=e[l],d=t[l],u=(0,i.max)(c.data)[1];if(u===s)continue;let p=(0,i.softmax)(c.data)[u];p>n&&(r.push(d),o.push(p),a.push(u))}return[r,o,a]}check_segment_validity(e,t,n,s=.5,r=.8){let o=[],i=0,a=0;const l=t[n].data;for(let t=0;t<e.length;++t)e[t]===n&&(o.push(t),++i),l[t]>=s&&++a;let c=i>0&&a>0;if(c){c=i/a>r}return[c,o]}compute_segments(e,t,n,s,r,o=null,i=null){let[l,c]=i??e[0].dims,d=new a.Tensor("int32",new Int32Array(l*c),[l,c]),u=[];if(null!==i)for(let t=0;t<e.length;++t)e[t]=(0,a.interpolate)(e[t],i,"bilinear",!1);let p=new Int32Array(e[0].data.length),h=new Float32Array(e[0].data.length);for(let n=0;n<e.length;++n){let s=t[n];const r=e[n].data;for(let e=0;e<r.length;++e)r[e]*=s,r[e]>h[e]&&(p[e]=n,h[e]=r[e])}let _=0;const m=d.data;for(let o=0;o<n.length;++o){let i=n[o],[a,l]=this.check_segment_validity(p,e,o,s,r);if(a){++_;for(let e of l)m[e]=_;u.push({id:_,label_id:i,score:t[o]})}}return[d,u]}post_process_panoptic_segmentation(e,t=.5,n=.5,s=.8,r=null,o=null){null===r&&(console.warn("`label_ids_to_fuse` unset. No instance will be fused."),r=new Set);const i=e.logits,l=e.pred_masks.sigmoid();let[c,d,u]=i.dims;if(u-=1,null!==o&&o.length!==c)throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits");let p=[];for(let e=0;e<c;++e){let c=null!==o?o[e]:null,d=i[e],h=l[e],[_,m,f]=this.remove_low_and_no_objects(d,h,t,u);if(0===f.length){let[e,t]=c??h.dims.slice(-2),n=new a.Tensor("int32",new Int32Array(e*t).fill(-1),[e,t]);p.push({segmentation:n,segments_info:[]});continue}let[g,w]=this.compute_segments(_,m,f,n,s,r,c);p.push({segmentation:g,segments_info:w})}return p}post_process_instance_segmentation(){throw Error("Not implemented yet")}}class W extends f{post_process_object_detection(...e){return d(...e)}}class X extends f{reshape_input_points(e,t,n,s=!1){e=structuredClone(e);let o=(0,r.calculateDimensions)(e);if(3===o.length)s||(o=[1,...o]),e=[e];else if(4!==o.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 s=0;s<e.length;++s){let r=t[s],o=n[s],i=[o[0]/r[0],o[1]/r[1]];for(let t=0;t<e[s].length;++t)for(let n=0;n<e[s][t].length;++n)for(let r=0;r<e[s][t][n].length;++r)e[s][t][n][r]*=i[r%2]}return new a.Tensor("float32",Float32Array.from(e.flat(1/0)),o)}add_input_labels(e,t){let n=(0,r.calculateDimensions)(e);if(2===n.length)n=[1,...n],e=[e];else if(3!==n.length)throw Error("The input_points must be a 4D tensor of shape `batch_size`, `point_batch_size`, `nb_points_per_image`, `2`.");if(n.some(((e,n)=>e!==t.dims[n])))throw Error(`The first ${n.length} dimensions of 'input_points' and 'input_labels' must be the same.`);return new a.Tensor("int64",e.flat(1/0).map(BigInt),n)}async _call(e,{input_points:t=null,input_labels:n=null,input_boxes:s=null}={}){const r=await super._call(e);if(t&&(r.input_points=this.reshape_input_points(t,r.original_sizes,r.reshaped_input_sizes)),n){if(!r.input_points)throw Error("`input_points` must be provided if `input_labels` are provided.");r.input_labels=this.add_input_labels(n,r.input_points)}return s&&(r.input_boxes=this.reshape_input_points(s,r.original_sizes,r.reshaped_input_sizes,!0)),r}async post_process_masks(e,t,n,{mask_threshold:s=0,binarize:r=!0,pad_size:o=null}={}){const i=[],l=[(o=o??this.pad_size).height,o.width];for(let o=0;o<t.length;++o){const c=t[o],d=n[o];let u=await(0,a.interpolate_4d)(e[o],{mode:"bilinear",size:l});if(u=u.slice(null,null,[0,d[0]],[0,d[1]]),u=await(0,a.interpolate_4d)(u,{mode:"bilinear",size:c}),r){const e=u.data,t=new Uint8Array(e.length);for(let n=0;n<e.length;++n)e[n]>s&&(t[n]=1);u=new a.Tensor("bool",t,u.dims)}i.push(u)}return i}generate_crop_boxes(e,t,{crop_n_layers:n=0,overlap_ratio:s=512/1500,points_per_crop:r=32,crop_n_points_downscale_factor:o=1}={}){}}class Q extends f{pad_image(e,t,n,s={}){const[r,o,i]=t;return super.pad_image(e,t,{width:o+(n-o%n)%n,height:r+(n-r%n)%n},{mode:"symmetric",center:!1,constant_values:-1,...s})}}class H extends f{async _call(e,t){Array.isArray(e)||(e=[e]),Array.isArray(t)||(t=[t]);const n=await Promise.all(e.map((e=>this.preprocess(e)))),s=await Promise.all(t.map((e=>this.preprocess(e,{do_normalize:!1,do_convert_rgb:!1,do_convert_grayscale:!0})))),r=(0,a.stack)(n.map(((e,t)=>(0,a.cat)([e.pixel_values,s[t].pixel_values],0))),0);return{pixel_values:r,original_sizes:n.map((e=>e.original_size)),reshaped_input_sizes:n.map((e=>e.reshaped_input_size))}}}class Y extends m{constructor(e){super(e),this.config.mel_filters??=(0,l.mel_filter_bank)(Math.floor(1+this.config.n_fft/2),this.config.feature_size,0,8e3,this.config.sampling_rate,"slaney","slaney"),this.window=(0,l.window_function)(this.config.n_fft,"hann")}async _extract_fbank_features(e){const t=await(0,l.spectrogram)(e,this.window,this.config.n_fft,this.config.hop_length,{power:2,mel_filters:this.config.mel_filters,log_mel:"log10",max_num_frames:this.config.nb_max_frames}),n=t.data,s=(0,i.max)(n)[0];for(let e=0;e<n.length;++e)n[e]=(Math.max(n[e],s-8)+4)/4;return t}async _call(e){let t;p(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 J extends m{_zero_mean_unit_var_norm(e){const t=e.reduce(((e,t)=>e+t),0)/e.length,n=e.reduce(((e,n)=>e+(n-t)**2),0)/e.length;return e.map((e=>(e-t)/Math.sqrt(n+1e-7)))}async _call(e){p(e,"Wav2Vec2FeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));let t=e;this.config.do_normalize&&(t=this._zero_mean_unit_var_norm(t));const n=[1,t.length];return{input_values:new a.Tensor("float32",t,n),attention_mask:new a.Tensor("int64",new BigInt64Array(t.length).fill(1n),n)}}}class K extends m{constructor(e){super(e);const t=this.config.sampling_rate,n=(0,l.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<n.length;++e)n[e].push(0);this.mel_filters=n,this.window=(0,l.window_function)(400,"povey",{periodic:!1})}async _extract_fbank_features(e,t){return e=e.map((e=>32768*e)),(0,l.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,max_num_frames:t,transpose:!0})}async _call(e,{padding:t=!0,pad_to_multiple_of:n=2,do_normalize_per_mel_bins:s=!0,return_attention_mask:r=!0}={}){p(e,"SeamlessM4TFeatureExtractor");let o,i=await this._extract_fbank_features(e,this.config.max_length);if(s){const[e,t]=i.dims,n=i.data;for(let s=0;s<t;++s){let r=0;for(let o=0;o<e;++o)r+=n[o*t+s];const o=r/e;let i=0;for(let r=0;r<e;++r)i+=(n[r*t+s]-o)**2;i/=e-1;const a=Math.sqrt(i+1e-7);for(let r=0;r<e;++r){const e=r*t+s;n[e]=(n[e]-o)/a}}}if(t){const[e,t]=i.dims,s=i.data,l=e%n;if(l>0){const n=new Float32Array(t*(e+l));n.set(s),n.fill(this.config.padding_value,s.length);const c=e+l;i=new a.Tensor(i.type,n,[c,t]),r&&(o=new a.Tensor("int64",new BigInt64Array(c),[1,c]),o.data.fill(1n,0,e))}}const[l,c]=i.dims,d=this.config.stride;if(0!==l%d)throw new Error(`The number of frames (${l}) must be a multiple of the stride (${d}).`);const u=i.view(1,Math.floor(l/d),c*d),h={input_features:u};if(r){const e=u.dims[1],t=new BigInt64Array(e);if(o){const e=o.data;for(let n=1,s=0;n<l;n+=d,++s)t[s]=e[n]}else t.fill(1n);h.attention_mask=new a.Tensor("int64",t,[1,e])}return h}}class Z extends m{constructor(e){super(e);const t=this.config.sampling_rate,n=(0,l.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<n.length;++e)n[e].push(0);this.mel_filters=n,this.window=(0,l.window_function)(400,"hann",{periodic:!1}),this.mean=this.config.mean,this.std=this.config.std}async _extract_fbank_features(e,t){return(0,l.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,max_num_frames:t,transpose:!0})}async _call(e){p(e,"ASTFeatureExtractor");const t=await this._extract_fbank_features(e,this.config.max_length);if(this.config.do_normalize){const e=2*this.std,n=t.data;for(let t=0;t<n.length;++t)n[t]=(n[t]-this.mean)/e}return{input_values:t.unsqueeze_(0)}}}class ee extends m{constructor(e){super(e),this.mel_filters=(0,l.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,null,"htk"),this.mel_filters_slaney=(0,l.mel_filter_bank)(this.config.nb_frequency_bins,this.config.feature_size,this.config.frequency_min,this.config.frequency_max,this.config.sampling_rate,"slaney","slaney"),this.window=(0,l.window_function)(this.config.fft_window_size,"hann")}async _get_input_mel(e,t,n,s){let r,o=!1;const i=e.length-t;if(i>0){if("rand_trunc"!==n)throw new Error(`Truncation strategy "${n}" not implemented`);{o=!0;const n=Math.floor(Math.random()*(i+1));e=e.subarray(n,n+t),r=await this._extract_fbank_features(e,this.mel_filters_slaney,this.config.nb_max_samples)}}else{if(i<0){let n=new Float64Array(t);if(n.set(e),"repeat"===s)for(let s=e.length;s<t;s+=e.length)n.set(e.subarray(0,Math.min(e.length,t-s)),s);else if("repeatpad"===s)for(let t=e.length;t<-i;t+=e.length)n.set(e,t);e=n}if("fusion"===n)throw new Error(`Truncation strategy "${n}" not implemented`);r=await this._extract_fbank_features(e,this.mel_filters_slaney,this.config.nb_max_samples)}return r.unsqueeze_(0)}async _extract_fbank_features(e,t,n=null){return(0,l.spectrogram)(e,this.window,this.config.fft_window_size,this.config.hop_length,{power:2,mel_filters:t,log_mel:"dB",max_num_frames:n,do_pad:!1,transpose:!0})}async _call(e,{max_length:t=null}={}){p(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 te extends m{async _call(e){p(e,"PyAnnoteFeatureExtractor"),e instanceof Float64Array&&(e=new Float32Array(e));const t=[1,1,e.length];return{input_values:new a.Tensor("float32",e,t)}}samples_to_frames(e){return(e-this.config.offset)/this.config.step}post_process_speaker_diarization(e,t){const n=t/this.samples_to_frames(t)/this.config.sampling_rate,s=[];for(const t of e.tolist()){const e=[];let r=-1;for(let n=0;n<t.length;++n){const s=(0,i.softmax)(t[n]),[o,a]=(0,i.max)(s),[l,c]=[n,n+1];a!==r?(r=a,e.push({id:a,start:l,end:c,score:o})):(e.at(-1).end=c,e.at(-1).score+=o)}s.push(e.map((({id:e,start:t,end:s,score:r})=>({id:e,start:t*n,end:s*n,confidence:r/(s-t)}))))}return s}}class ne extends m{constructor(e){super(e);const t=this.config.sampling_rate,n=(0,l.mel_filter_bank)(256,this.config.num_mel_bins,20,Math.floor(t/2),t,null,"kaldi",!0);for(let e=0;e<n.length;++e)n[e].push(0);this.mel_filters=n,this.window=(0,l.window_function)(400,"hamming",{periodic:!1}),this.min_num_frames=this.config.min_num_frames}async _extract_fbank_features(e){return e=e.map((e=>32768*e)),(0,l.spectrogram)(e,this.window,400,160,{fft_length:512,power:2,center:!1,preemphasis:.97,mel_filters:this.mel_filters,log_mel:"log",mel_floor:1.192092955078125e-7,remove_dc_offset:!0,transpose:!0,min_num_frames:this.min_num_frames})}async _call(e){p(e,"WeSpeakerFeatureExtractor");const t=(await this._extract_fbank_features(e)).unsqueeze_(0);if(null===this.config.fbank_centering_span){const e=t.mean(1).data,n=t.data,[s,r,o]=t.dims;for(let t=0;t<s;++t){const s=t*r*o,i=t*o;for(let t=0;t<r;++t){const r=s+t*o;for(let t=0;t<o;++t)n[r+t]-=e[i+t]}}}return{input_features:t}}}class se extends m{}class re extends s.Callable{constructor(e){super(),this.feature_extractor=e}async _call(e,...t){return await this.feature_extractor(e,...t)}}class oe extends re{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 ie extends re{async _call(e){return await this.feature_extractor(e)}}class ae extends re{async _call(e){return await this.feature_extractor(e)}}class le extends re{async _call(e){return await this.feature_extractor(e)}post_process_speaker_diarization(...e){return this.feature_extractor.post_process_speaker_diarization(...e)}}class ce extends re{async _call(e){return await this.feature_extractor(e)}}class de extends re{}class ue extends re{constructor(e){super(e);const{tasks_answer_post_processing_type:t,task_prompts_without_inputs:n,task_prompts_with_input:s}=e.config;this.tasks_answer_post_processing_type=new Map(Object.entries(t??{})),this.task_prompts_without_inputs=new Map(Object.entries(n??{})),this.task_prompts_with_input=new Map(Object.entries(s??{})),this.regexes={quad_boxes:/(.+?)<loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)>/gm,bboxes:/([^<]+)?<loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)>/gm},this.size_per_bin=1e3}construct_prompts(e){"string"==typeof e&&(e=[e]);const t=[];for(const n of e)if(this.task_prompts_without_inputs.has(n))t.push(this.task_prompts_without_inputs.get(n));else{for(const[e,s]of this.task_prompts_with_input)if(n.includes(e)){t.push(s.replaceAll("{input}",n).replaceAll(e,""));break}t.length!==e.length&&t.push(n)}return t}post_process_generation(e,t,n){const s=this.tasks_answer_post_processing_type.get(t)??"pure_text";let r;switch(e=e.replaceAll("<s>","").replaceAll("</s>",""),s){case"pure_text":r=e;break;case"description_with_bboxes":case"bboxes":case"phrase_grounding":case"ocr":const o="ocr"===s?"quad_boxes":"bboxes",i=e.matchAll(this.regexes[o]),a=[],l=[];for(const[e,t,...s]of i)a.push(t?t.trim():a.at(-1)??""),l.push(s.map(((e,t)=>(Number(e)+.5)/this.size_per_bin*n[t%2])));r={labels:a,[o]:l};break;default:throw new Error(`Task "${t}" (of type "${s}") not yet implemented.`)}return{[t]:r}}}class pe{static FEATURE_EXTRACTOR_CLASS_MAPPING={ImageFeatureExtractor:f,WhisperFeatureExtractor:Y,ViTFeatureExtractor:S,MobileViTFeatureExtractor:N,MobileViTImageProcessor:O,MobileNetV1FeatureExtractor:z,MobileNetV2FeatureExtractor:L,MobileNetV3FeatureExtractor:I,MobileNetV4FeatureExtractor:B,OwlViTFeatureExtractor:j,Owlv2ImageProcessor:D,CLIPFeatureExtractor:k,CLIPImageProcessor:T,Florence2Processor:ue,ChineseCLIPFeatureExtractor:v,SiglipImageProcessor:C,ConvNextFeatureExtractor:F,ConvNextImageProcessor:P,SegformerFeatureExtractor:w,SapiensFeatureExtractor:g,BitImageProcessor:y,DPTImageProcessor:b,DPTFeatureExtractor:M,GLPNFeatureExtractor:x,BeitFeatureExtractor:G,DeiTFeatureExtractor:R,DetrFeatureExtractor:U,RTDetrImageProcessor:V,YolosFeatureExtractor:W,DonutFeatureExtractor:$,NougatImageProcessor:q,EfficientNetImageProcessor:E,ViTImageProcessor:A,VitMatteImageProcessor:H,SamImageProcessor:X,Swin2SRImageProcessor:Q,Wav2Vec2FeatureExtractor:J,SeamlessM4TFeatureExtractor:K,SpeechT5FeatureExtractor:se,ASTFeatureExtractor:Z,ClapFeatureExtractor:ee,PyAnnoteFeatureExtractor:te,WeSpeakerFeatureExtractor:ne};static PROCESSOR_CLASS_MAPPING={WhisperProcessor:ie,Wav2Vec2ProcessorWithLM:ae,PyAnnoteProcessor:le,SamProcessor:oe,SpeechT5Processor:ce,OwlViTProcessor:de,Florence2Processor:ue};static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:s=null,local_files_only:r=!1,revision:i="main"}={}){let a=n??await(0,o.getModelJSON)(e,"preprocessor_config.json",!0,{progress_callback:t,config:n,cache_dir:s,local_files_only:r,revision:i}),l=a.feature_extractor_type??a.image_processor_type,c=this.FEATURE_EXTRACTOR_CLASS_MAPPING[l];if(!c){if(void 0===a.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=f}return new(this.PROCESSOR_CLASS_MAPPING[a.processor_class]??re)(new c(a))}}},"./src/tokenizers.js":
 /*!***************************!*\
   !*** ./src/tokenizers.js ***!
   \***************************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{AlbertTokenizer:()=>xe,AutoTokenizer:()=>ht,BartTokenizer:()=>Ne,BertTokenizer:()=>ye,BlenderbotSmallTokenizer:()=>lt,BlenderbotTokenizer:()=>at,BloomTokenizer:()=>Ve,CLIPTokenizer:()=>st,CamembertTokenizer:()=>Ee,CodeGenTokenizer:()=>nt,CodeLlamaTokenizer:()=>$e,CohereTokenizer:()=>pt,ConvBertTokenizer:()=>Pe,DebertaTokenizer:()=>ve,DebertaV2Tokenizer:()=>Ce,DistilBertTokenizer:()=>Ae,ElectraTokenizer:()=>Le,EsmTokenizer:()=>Qe,FalconTokenizer:()=>We,GPT2Tokenizer:()=>Be,GPTNeoXTokenizer:()=>Xe,GemmaTokenizer:()=>Ye,Grok1Tokenizer:()=>Je,HerbertTokenizer:()=>Fe,LlamaTokenizer:()=>Ge,M2M100Tokenizer:()=>et,MBart50Tokenizer:()=>je,MBartTokenizer:()=>Oe,MPNetTokenizer:()=>Ue,MarianTokenizer:()=>ot,MobileBertTokenizer:()=>ke,NllbTokenizer:()=>Ze,NougatTokenizer:()=>dt,PreTrainedTokenizer:()=>be,Qwen2Tokenizer:()=>He,RoFormerTokenizer:()=>Se,RobertaTokenizer:()=>De,SiglipTokenizer:()=>rt,SpeechT5Tokenizer:()=>ct,SqueezeBertTokenizer:()=>Te,T5Tokenizer:()=>Ie,TokenizerModel:()=>k,VitsTokenizer:()=>ut,Wav2Vec2CTCTokenizer:()=>it,WhisperTokenizer:()=>tt,XLMRobertaTokenizer:()=>qe,XLMTokenizer:()=>ze,is_chinese_char:()=>w});var s=n(/*! ./utils/generic.js */"./src/utils/generic.js"),r=n(/*! ./utils/core.js */"./src/utils/core.js"),o=n(/*! ./utils/hub.js */"./src/utils/hub.js"),i=n(/*! ./utils/maths.js */"./src/utils/maths.js"),a=n(/*! ./utils/tensor.js */"./src/utils/tensor.js"),l=n(/*! ./utils/data-structures.js */"./src/utils/data-structures.js"),c=n(/*! @huggingface/jinja */"./node_modules/@huggingface/jinja/dist/index.js"),d=n(/*! ./models/whisper/common_whisper.js */"./src/models/whisper/common_whisper.js"),u=n(/*! ./utils/constants.js */"./src/utils/constants.js");async function p(e,t){const n=await Promise.all([(0,o.getModelJSON)(e,"tokenizer.json",!0,t),(0,o.getModelJSON)(e,"tokenizer_config.json",!0,t)]);return null!==t.legacy&&(n[1].legacy=t.legacy),n}function h(e,t=!0){if(void 0!==e.Regex){let t=e.Regex.replace(/\\([#&~])/g,"$1");for(const[e,n]of y)t=t.replaceAll(e,n);return new RegExp(t,"gu")}if(void 0!==e.String){const n=(0,r.escapeRegExp)(e.String);return new RegExp(t?n:`(${n})`,"gu")}return console.warn("Unknown pattern type:",e),null}function _(e){return new Map(Object.entries(e))}function m(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 ${t.length}.`)}}function f(e){return e.replace(/ \./g,".").replace(/ \?/g,"?").replace(/ \!/g,"!").replace(/ ,/g,",").replace(/ \' /g,"'").replace(/ n\'t/g,"n't").replace(/ \'m/g,"'m").replace(/ \'s/g,"'s").replace(/ \'ve/g,"'ve").replace(/ \'re/g,"'re")}function g(e){return e.replace(/[\u0300-\u036f]/g,"")}function w(e){return e>=19968&&e<=40959||e>=13312&&e<=19903||e>=131072&&e<=173791||e>=173824&&e<=177983||e>=177984&&e<=178207||e>=178208&&e<=183983||e>=63744&&e<=64255||e>=194560&&e<=195103}const M="\\p{P}\\u0021-\\u002F\\u003A-\\u0040\\u005B-\\u0060\\u007B-\\u007E",b=new RegExp(`^[${M}]+$`,"gu"),y=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 s.Callable{constructor(e){super(),this.config=e,this.vocab=[],this.tokens_to_ids=new Map,this.unk_token_id=void 0,this.unk_token=void 0,this.end_of_word_suffix=void 0,this.fuse_unk=this.config.fuse_unk??!1}static fromConfig(e,...t){switch(e.type){case"WordPiece":return new T(e);case"Unigram":return new v(e,...t);case"BPE":return new P(e);default:if(e.vocab)return new S(e,...t);throw new Error(`Unknown TokenizerModel type: ${e.type}`)}}_call(e){let t=this.encode(e);return this.fuse_unk&&(t=function(e,t,n){const s=[];let r=0;for(;r<e.length;)if(s.push(e[r]),(n.get(e[r])??t)===t)for(;r<e.length&&(n.get(e[r])??t)===t;)++r;else++r;return s}(t,this.unk_token_id,this.tokens_to_ids)),t}encode(e){throw Error("encode should be implemented in subclass.")}convert_tokens_to_ids(e){return e.map((e=>this.tokens_to_ids.get(e)??this.unk_token_id))}convert_ids_to_tokens(e){return e.map((e=>this.vocab[e]??this.unk_token))}}class T extends k{constructor(e){super(e),this.tokens_to_ids=_(e.vocab),this.unk_token_id=this.tokens_to_ids.get(e.unk_token),this.unk_token=e.unk_token,this.max_input_chars_per_word=e.max_input_chars_per_word??100,this.vocab=new Array(this.tokens_to_ids.size);for(const[e,t]of this.tokens_to_ids)this.vocab[t]=e}encode(e){const t=[];for(const n of e){const e=[...n];if(e.length>this.max_input_chars_per_word){t.push(this.unk_token);continue}let s=!1,r=0;const o=[];for(;r<e.length;){let t=e.length,n=null;for(;r<t;){let s=e.slice(r,t).join("");if(r>0&&(s=this.config.continuing_subword_prefix+s),this.tokens_to_ids.has(s)){n=s;break}--t}if(null===n){s=!0;break}o.push(n),r=t}s?t.push(this.unk_token):t.push(...o)}return t}}class v extends k{constructor(e,t){super(e);const n=e.vocab.length;this.vocab=new Array(n),this.scores=new Array(n);for(let t=0;t<n;++t){const n=e.vocab[t];this.vocab[t]=n[0],this.scores[t]=n[1]}this.unk_token_id=e.unk_id,this.unk_token=this.vocab[e.unk_id],this.tokens_to_ids=new Map(this.vocab.map(((e,t)=>[e,t]))),this.bosToken=" ",this.bosTokenId=this.tokens_to_ids.get(this.bosToken),this.eosToken=t.eos_token,this.eosTokenId=this.tokens_to_ids.get(this.eosToken),this.unkToken=this.vocab[this.unk_token_id],this.minScore=(0,i.min)(this.scores)[0],this.unkScore=this.minScore-10,this.scores[this.unk_token_id]=this.unkScore,this.trie=new l.CharTrie,this.trie.extend(this.vocab),this.fuse_unk=!0}populateNodes(e){const t=e.sentence,n=t.length;let s=0;for(;s<n;){const n=1;let r=!1;const o=[];for(let i of this.trie.commonPrefixSearch(t.slice(s))){o.push(i);const t=this.tokens_to_ids.get(i),a=this.scores[t],l=i.length;e.insert(s,l,a,t),r||l!==n||(r=!0)}r||e.insert(s,n,this.unkScore,this.unk_token_id),s+=n}}tokenize(e){const t=new l.TokenLattice(e,this.bosTokenId,this.eosTokenId);return this.populateNodes(t),t.tokens()}encode(e){const t=[];for(const n of e){const e=this.tokenize(n);t.push(...e)}return t}}const C=(()=>{const e=[...Array.from({length:"~".charCodeAt(0)-"!".charCodeAt(0)+1},((e,t)=>t+"!".charCodeAt(0))),...Array.from({length:"¬".charCodeAt(0)-"¡".charCodeAt(0)+1},((e,t)=>t+"¡".charCodeAt(0))),...Array.from({length:"ÿ".charCodeAt(0)-"®".charCodeAt(0)+1},((e,t)=>t+"®".charCodeAt(0)))],t=e.slice();let n=0;for(let s=0;s<256;++s)e.includes(s)||(e.push(s),t.push(256+n),n+=1);const s=t.map((e=>String.fromCharCode(e)));return Object.fromEntries(e.map(((e,t)=>[e,s[t]])))})(),F=(0,r.reverseDictionary)(C);class P extends k{constructor(e){super(e),this.BPE_SPLIT_TOKEN=" ",this.tokens_to_ids=_(e.vocab),this.unk_token_id=this.tokens_to_ids.get(e.unk_token),this.unk_token=e.unk_token,this.vocab=new Array(this.tokens_to_ids.size);for(const[e,t]of this.tokens_to_ids)this.vocab[t]=e;this.bpe_ranks=new Map(e.merges.map(((e,t)=>[e,t]))),this.merges=e.merges.map((e=>e.split(this.BPE_SPLIT_TOKEN))),this.end_of_word_suffix=e.end_of_word_suffix,this.continuing_subword_suffix=e.continuing_subword_suffix??null,this.byte_fallback=this.config.byte_fallback??!1,this.byte_fallback&&(this.text_encoder=new TextEncoder),this.ignore_merges=this.config.ignore_merges??!1,this.cache=new Map}bpe(e){if(0===e.length)return[];const t=this.cache.get(e);if(void 0!==t)return t;const n=Array.from(e);this.end_of_word_suffix&&(n[n.length-1]+=this.end_of_word_suffix);let s=[];if(n.length>1){const e=new l.PriorityQueue(((e,t)=>e.score<t.score));let t={token:n[0],bias:0,prev:null,next:null},r=t;for(let t=1;t<n.length;++t){const s={bias:t/n.length,token:n[t],prev:r,next:null};r.next=s,this._add_node(e,r),r=s}for(;!e.isEmpty();){const n=e.pop();if(n.deleted||!n.next||n.next.deleted)continue;if(n.deleted=!0,n.next.deleted=!0,n.prev){const e={...n.prev};n.prev.deleted=!0,n.prev=e,e.prev?e.prev.next=e:t=e}const s={token:n.token+n.next.token,bias:n.bias,prev:n.prev,next:n.next.next};s.prev?(s.prev.next=s,this._add_node(e,s.prev)):t=s,s.next&&(s.next.prev=s,this._add_node(e,s))}for(let e=t;null!==e;e=e.next)s.push(e.token)}else s=n;if(this.continuing_subword_suffix)for(let e=0;e<s.length-1;++e)s[e]+=this.continuing_subword_suffix;return this.cache.set(e,s),s}_add_node(e,t){const n=this.bpe_ranks.get(t.token+this.BPE_SPLIT_TOKEN+t.next.token);void 0!==n&&(t.score=n+t.bias,e.push(t))}encode(e){const t=[];for(const n of e){if(this.ignore_merges&&this.tokens_to_ids.has(n)){t.push(n);continue}const e=this.bpe(n);for(const n of e)this.tokens_to_ids.has(n)?t.push(n):this.byte_fallback?t.push(...Array.from(this.text_encoder.encode(n)).map((e=>`<0x${e.toString(16).toUpperCase().padStart(2,"0")}>`))):t.push(this.unk_token)}return t}}class S extends k{constructor(e,t){super(e),this.tokens_to_ids=_(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 s.Callable{constructor(e){super(),this.config=e}static fromConfig(e){if(null===e)return null;switch(e.type){case"BertNormalizer":return new V(e);case"Precompiled":return new pe(e);case"Sequence":return new D(e);case"Replace":return new E(e);case"NFC":return new z(e);case"NFKC":return new L(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 j(e);default:throw new Error(`Unknown Normalizer type: ${e.type}`)}}normalize(e){throw Error("normalize should be implemented in subclass.")}_call(e){return this.normalize(e)}}class E extends A{normalize(e){const t=h(this.config.pattern);return null===t?e:e.replaceAll(t,this.config.content)}}class z extends A{normalize(e){return e=e.normalize("NFC")}}class L extends A{normalize(e){return e=e.normalize("NFKC")}}class I extends A{normalize(e){return e=e.normalize("NFKD")}}class B extends A{normalize(e){return this.config.strip_left&&this.config.strip_right?e=e.trim():(this.config.strip_left&&(e=e.trimStart()),this.config.strip_right&&(e=e.trimEnd())),e}}class N extends A{normalize(e){return e=g(e)}}class O extends A{normalize(e){return e=e.toLowerCase()}}class j extends A{normalize(e){return e=this.config.prepend+e}}class D extends A{constructor(e){super(e),this.normalizers=e.normalizers.map((e=>A.fromConfig(e)))}normalize(e){return this.normalizers.reduce(((e,t)=>t.normalize(e)),e)}}class V extends A{_tokenize_chinese_chars(e){const t=[];for(let n=0;n<e.length;++n){const s=e[n];w(s.charCodeAt(0))?(t.push(" "),t.push(s),t.push(" ")):t.push(s)}return t.join("")}stripAccents(e){return e.normalize("NFD").replace(/[\u0300-\u036f]/g,"")}_is_control(e){switch(e){case"\t":case"\n":case"\r":return!1;default:return/^\p{Cc}|\p{Cf}|\p{Co}|\p{Cs}$/u.test(e)}}_clean_text(e){const t=[];for(const n of e){const e=n.charCodeAt(0);0===e||65533===e||this._is_control(n)||(/^\s$/.test(n)?t.push(" "):t.push(n))}return t.join("")}normalize(e){return this.config.clean_text&&(e=this._clean_text(e)),this.config.handle_chinese_chars&&(e=this._tokenize_chinese_chars(e)),this.config.lowercase?(e=e.toLowerCase(),!1!==this.config.strip_accents&&(e=this.stripAccents(e))):this.config.strip_accents&&(e=this.stripAccents(e)),e}}class R extends s.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 _e(e);case"WhitespaceSplit":return new me(e);case"Metaspace":return new de(e);case"ByteLevel":return new $(e);case"Split":return new q(e);case"Punctuation":return new U(e);case"Digits":return new W(e);case"Replace":return new fe(e);default:throw new Error(`Unknown PreTokenizer type: ${e.type}`)}}pre_tokenize_text(e,t){throw Error("pre_tokenize_text should be implemented in subclass.")}pre_tokenize(e,t){return(Array.isArray(e)?e.map((e=>this.pre_tokenize_text(e,t))):this.pre_tokenize_text(e,t)).flat()}_call(e,t){return this.pre_tokenize(e,t)}}class G extends R{constructor(e){super(),this.pattern=new RegExp(`[^\\s${M}]+|[${M}]`,"gu")}pre_tokenize_text(e,t){return e.trim().match(this.pattern)||[]}}class $ 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=C,this.text_encoder=new TextEncoder}pre_tokenize_text(e,t){this.add_prefix_space&&!e.startsWith(" ")&&(e=" "+e);return(this.use_regex?e.match(this.pattern)||[]:[e]).map((e=>Array.from(this.text_encoder.encode(e),(e=>this.byte_encoder[e])).join("")))}}class q extends 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 n=[];let s=0;for(const r of e.matchAll(t)){const t=r[0];s<r.index&&n.push(e.slice(s,r.index)),t.length>0&&n.push(t),s=r.index+t.length}return s<e.length&&n.push(e.slice(s)),n}(e,this.pattern)}}class U extends R{constructor(e){super(),this.config=e,this.pattern=new RegExp(`[^${M}]+|[${M}]+`,"gu")}pre_tokenize_text(e,t){return e.match(this.pattern)||[]}}class W 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 s.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:n=!0}={}){n&&(e=(0,r.mergeArrays)([this.cls],e,[this.sep]));let s=new Array(e.length).fill(0);if(null!==t){const o=n&&this instanceof H?[this.sep]:[],i=n?[this.sep]:[];e=(0,r.mergeArrays)(e,o,t,i),s=(0,r.mergeArrays)(s,new Array(t.length+o.length+i.length).fill(1))}return{tokens:e,token_type_ids:s}}}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:n=!0}={}){const s=null===t?this.single:this.pair;let o=[],i=[];for(const a of s)"SpecialToken"in a?n&&(o.push(a.SpecialToken.id),i.push(a.SpecialToken.type_id)):"Sequence"in a&&("A"===a.Sequence.id?(o=(0,r.mergeArrays)(o,e),i=(0,r.mergeArrays)(i,new Array(e.length).fill(a.Sequence.type_id))):"B"===a.Sequence.id&&(o=(0,r.mergeArrays)(o,t),i=(0,r.mergeArrays)(i,new Array(t.length).fill(a.Sequence.type_id))));return{tokens:o,token_type_ids:i}}}class J extends X{post_process(e,t=null){return t&&(e=(0,r.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,n={}){let s;for(const r of this.processors)if(r instanceof J){if(e=r.post_process(e).tokens,t){t=r.post_process(t).tokens}}else{const o=r.post_process(e,t,n);e=o.tokens,s=o.token_type_ids}return{tokens:e,token_type_ids:s}}}class Z extends s.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 re(e);case"Metaspace":return new ue(e);case"ByteLevel":return new oe(e);case"Replace":return new ee(e);case"ByteFallback":return new te(e);case"Fuse":return new ne(e);case"Strip":return new se(e);case"Sequence":return new ae(e);case"CTC":return new ie(e);case"BPEDecoder":return new le(e);default:throw new Error(`Unknown Decoder type: ${e.type}`)}}_call(e){return this.decode(e)}decode(e){return this.decode_chain(e).join("")}decode_chain(e){throw Error("`decode_chain` should be implemented in subclass.")}}class ee extends Z{decode_chain(e){const t=h(this.config.pattern);return null===t?e:e.map((e=>e.replaceAll(t,this.config.content)))}}class te extends Z{constructor(e){super(e),this.text_decoder=new TextDecoder}decode_chain(e){const t=[];let n=[];for(const s of e){let e=null;if(6===s.length&&s.startsWith("<0x")&&s.endsWith(">")){const t=parseInt(s.slice(3,5),16);isNaN(t)||(e=t)}if(null!==e)n.push(e);else{if(n.length>0){const e=this.text_decoder.decode(Uint8Array.from(n));t.push(e),n=[]}t.push(s)}}if(n.length>0){const e=this.text_decoder.decode(Uint8Array.from(n));t.push(e),n=[]}return t}}class ne extends Z{decode_chain(e){return[e.join("")]}}class se extends Z{constructor(e){super(e),this.content=this.config.content,this.start=this.config.start,this.stop=this.config.stop}decode_chain(e){return e.map((e=>{let t=0;for(let n=0;n<this.start&&e[n]===this.content;++n)t=n+1;let n=e.length;for(let t=0;t<this.stop;++t){const s=e.length-t-1;if(e[s]!==this.content)break;n=s}return e.slice(t,n)}))}}class re 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 oe extends Z{constructor(e){super(e),this.byte_decoder=F,this.text_decoder=new TextDecoder("utf-8",{fatal:!1,ignoreBOM:!0}),this.end_of_word_suffix=null}convert_tokens_to_string(e){const t=e.join(""),n=new Uint8Array([...t].map((e=>this.byte_decoder[e])));return this.text_decoder.decode(n)}decode_chain(e){const t=[];let n=[];for(const s of e)void 0!==this.added_tokens.find((e=>e.content===s))?(n.length>0&&(t.push(this.convert_tokens_to_string(n)),n=[]),t.push(s)):n.push(s);return n.length>0&&t.push(this.convert_tokens_to_string(n)),t}}class ie extends Z{constructor(e){super(e),this.pad_token=this.config.pad_token,this.word_delimiter_token=this.config.word_delimiter_token,this.cleanup=this.config.cleanup}convert_tokens_to_string(e){if(0===e.length)return"";const t=[e[0]];for(let n=1;n<e.length;++n)e[n]!==t.at(-1)&&t.push(e[n]);let n=t.filter((e=>e!==this.pad_token)).join("");return this.cleanup&&(n=f(n).replaceAll(this.word_delimiter_token," ").trim()),n}decode_chain(e){return[this.convert_tokens_to_string(e)]}}class ae extends Z{constructor(e){super(e),this.decoders=e.decoders.map((e=>Z.fromConfig(e)))}decode_chain(e){return this.decoders.reduce(((e,t)=>t.decode_chain(e)),e)}}class le extends Z{constructor(e){super(e),this.suffix=this.config.suffix}decode_chain(e){return e.map(((t,n)=>t.replaceAll(this.suffix,n===e.length-1?"":" ")))}}class ce extends Z{decode_chain(e){let t="";for(let n=1;n<e.length;n+=2)t+=e[n];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 n=e.replaceAll(" ",this.strRep);return this.addPrefixSpace&&!n.startsWith(this.replacement)&&("always"===this.prepend_scheme||"first"===this.prepend_scheme&&0===t)&&(n=this.strRep+n),[n]}}class ue extends Z{constructor(e){super(e),this.addPrefixSpace=e.add_prefix_space,this.replacement=e.replacement}decode_chain(e){const t=[];for(let n=0;n<e.length;++n){let s=e[n].replaceAll(this.replacement," ");this.addPrefixSpace&&0==n&&s.startsWith(" ")&&(s=s.substring(1)),t.push(s)}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,n)=>n.pre_tokenize(e,t)),[e])}}class _e extends R{constructor(e){super()}pre_tokenize_text(e,t){return e.match(/\w+|[^\w\s]+/g)||[]}}class me 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 we(e,t,n,s){for(const o of Object.keys(e)){const i=t-e[o].length,a=n(o),l=new Array(i).fill(a);e[o]="right"===s?(0,r.mergeArrays)(e[o],l):(0,r.mergeArrays)(l,e[o])}}function Me(e,t){for(const n of Object.keys(e))e[n].length=t}class be extends s.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.toSorted(((e,t)=>t.content.length-e.content.length)).map((e=>`${e.lstrip?"\\s*":""}(${(0,r.escapeRegExp)(e.content)})${e.rstrip?"\\s*":""}`)).join("|")):null,this.mask_token=this.getToken("mask_token"),this.mask_token_id=this.model.tokens_to_ids.get(this.mask_token),this.pad_token=this.getToken("pad_token","eos_token"),this.pad_token_id=this.model.tokens_to_ids.get(this.pad_token),this.sep_token=this.getToken("sep_token"),this.sep_token_id=this.model.tokens_to_ids.get(this.sep_token),this.unk_token=this.getToken("unk_token"),this.unk_token_id=this.model.tokens_to_ids.get(this.unk_token),this.model_max_length=t.model_max_length,this.remove_space=t.remove_space,this.clean_up_tokenization_spaces=t.clean_up_tokenization_spaces??!0,this.do_lowercase_and_remove_accent=t.do_lowercase_and_remove_accent??!1,t.padding_side&&(this.padding_side=t.padding_side),this.legacy=!1,this.chat_template=t.chat_template??null,Array.isArray(this.chat_template)){const e=Object.create(null);for(const{name:t,template:n}of this.chat_template){if("string"!=typeof t||"string"!=typeof n)throw new Error('Chat template must be a list of objects with "name" and "template" properties');e[t]=n}this.chat_template=e}this._compiled_template_cache=new Map}getToken(...e){for(const t of e){const e=this._tokenizer_config[t];if(e){if("object"==typeof e){if("AddedToken"===e.__type)return e.content;throw Error(`Unknown token: ${e}`)}return e}}return null}static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:s=null,local_files_only:r=!1,revision:o="main",legacy:i=null}={}){return new this(...await p(e,{progress_callback:t,config:n,cache_dir:s,local_files_only:r,revision:o,legacy:i}))}_call(e,{text_pair:t=null,add_special_tokens:n=!0,padding:s=!1,truncation:r=null,max_length:o=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,s)=>this._encode_plus(e,{text_pair:t[s],add_special_tokens:n,return_token_type_ids:c})))}else u=e.map((e=>this._encode_plus(e,{add_special_tokens:n,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:n,return_token_type_ids:c})]}if(null===o?o="max_length"===s?this.model_max_length:(0,i.max)(u.map((e=>e.input_ids.length)))[0]:r||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."),o=Math.min(o,this.model_max_length??1/0),s||r)for(let e=0;e<u.length;++e)u[e].input_ids.length!==o&&(u[e].input_ids.length>o?r&&Me(u[e],o):s&&we(u[e],o,(e=>"input_ids"===e?this.pad_token_id:0),this.padding_side));const p={};if(l){if((!s||!r)&&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 a.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 n=null!==this.pre_tokenizer?this.pre_tokenizer(e,{section_index:t}):[e];return this.model(n)}})).flat();return t}_encode_plus(e,{text_pair:t=null,add_special_tokens:n=!0,return_token_type_ids:s=null}={}){const{tokens:r,token_type_ids:o}=this._tokenize_helper(e,{pair:t,add_special_tokens:n}),i=this.model.convert_tokens_to_ids(r),a={input_ids:i,attention_mask:new Array(i.length).fill(1)};return(s??this.return_token_type_ids)&&o&&(a.token_type_ids=o),a}_tokenize_helper(e,{pair:t=null,add_special_tokens:n=!1}={}){const s=this._encode_text(e),o=this._encode_text(t);return this.post_processor?this.post_processor(s,o,{add_special_tokens:n}):{tokens:(0,r.mergeArrays)(s??[],o??[])}}tokenize(e,{pair:t=null,add_special_tokens:n=!1}={}){return this._tokenize_helper(e,{pair:t,add_special_tokens:n}).tokens}encode(e,{text_pair:t=null,add_special_tokens:n=!0,return_token_type_ids:s=null}={}){return this._encode_plus(e,{text_pair:t,add_special_tokens:n,return_token_type_ids:s}).input_ids}batch_decode(e,t={}){return e instanceof a.Tensor&&(e=e.tolist()),e.map((e=>this.decode(e,t)))}decode(e,t={}){if(e instanceof a.Tensor&&(e=m(e)),!Array.isArray(e)||0===e.length||!(0,r.isIntegralNumber)(e[0]))throw Error("token_ids must be a non-empty array of integers.");return this.decode_single(e,t)}decode_single(e,{skip_special_tokens:t=!1,clean_up_tokenization_spaces:n=null}){let s=this.model.convert_ids_to_tokens(e);t&&(s=s.filter((e=>!this.special_tokens.includes(e))));let r=this.decoder?this.decoder(s):s.join(" ");return this.decoder&&this.decoder.end_of_word_suffix&&(r=r.replaceAll(this.decoder.end_of_word_suffix," "),t&&(r=r.trim())),(n??this.clean_up_tokenization_spaces)&&(r=f(r)),r}apply_chat_template(e,{tools:t=null,documents:n=null,chat_template:s=null,add_generation_prompt:r=!1,tokenize:o=!0,padding:i=!1,truncation:a=!1,max_length:l=null,return_tensor:d=!0,return_dict:u=!1,tokenizer_kwargs:p={},...h}={}){if(this.chat_template&&"object"==typeof this.chat_template||null===this.chat_template){const e=this.chat_template;if(null!==s&&Object.hasOwn(e,s))s=e[s];else if(null===s&&"default"in e)s=e.default;else if(null===s)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 ${Object.keys(e).sort()}.`)}else{if(!this.chat_template)throw Error("Cannot use apply_chat_template() because tokenizer.chat_template is not set and no template argument was passed! For information about writing templates and setting the tokenizer.chat_template attribute, please see the documentation at https://huggingface.co/docs/transformers/main/en/chat_templating");s=this.chat_template}if("string"!=typeof s)throw Error("chat_template must be a string, but got "+typeof s);let _=this._compiled_template_cache.get(s);void 0===_&&(_=new c.Template(s),this._compiled_template_cache.set(s,_));const m=Object.create(null);for(const e of ge){const t=this.getToken(e);t&&(m[e]=t)}const f=_.render({messages:e,add_generation_prompt:r,tools:t,documents:n,...m,...h});if(o){const e=this._call(f,{add_special_tokens:!1,padding:i,truncation:a,max_length:l,return_tensor:d,...p});return u?e:e.input_ids}return f}}class ye extends be{return_token_type_ids=!0}class xe extends be{return_token_type_ids=!0}class ke extends be{return_token_type_ids=!0}class Te extends be{return_token_type_ids=!0}class ve extends be{return_token_type_ids=!0}class Ce extends be{return_token_type_ids=!0}class Fe extends be{return_token_type_ids=!0}class Pe extends be{return_token_type_ids=!0}class Se extends be{return_token_type_ids=!0}class Ae extends be{}class Ee extends be{}class ze extends be{return_token_type_ids=!0;constructor(e,t){super(e,t),console.warn('WARNING: `XLMTokenizer` is not yet supported by Hugging Face\'s "fast" tokenizers library. Therefore, you may experience slightly inaccurate results.')}}class Le extends be{return_token_type_ids=!0}class Ie extends be{}class Be extends be{}class Ne extends be{}class Oe extends be{constructor(e,t){super(e,t),this.languageRegex=/^[a-z]{2}_[A-Z]{2}$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))),this.lang_to_token=e=>e}_build_translation_inputs(e,t,n){return Ke(this,e,t,n)}}class je extends Oe{}class De extends be{}class Ve extends be{constructor(e,t){const n=".,!?…。，、।۔،",s=e.pre_tokenizer?.pretokenizers[0]?.pattern;s&&s.Regex===` ?[^(\\s|[${n}])]+`&&(s.Regex=` ?[^\\s${n}]+`),super(e,t)}}const Re="▁";class Ge extends be{padding_side="left";constructor(e,t){super(e,t),this.legacy=t.legacy??!0,this.legacy||(this.normalizer=null,this.pre_tokenizer=new de({replacement: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 $e extends be{}class qe extends be{}class Ue extends be{}class We extends be{}class Xe extends be{}class Qe extends be{}class He extends be{}class Ye extends be{}class Je extends be{}function Ke(e,t,n,s){if(!("language_codes"in e)||!Array.isArray(e.language_codes))throw new Error("Tokenizer must have `language_codes` attribute set and it should be an array of language ids.");if(!("languageRegex"in e&&e.languageRegex instanceof RegExp))throw new Error("Tokenizer must have `languageRegex` attribute set and it should be a regular expression.");if(!("lang_to_token"in e)||"function"!=typeof e.lang_to_token)throw new Error("Tokenizer must have `lang_to_token` attribute set and it should be a function.");const r=s.src_lang,o=s.tgt_lang;if(!e.language_codes.includes(o))throw new Error(`Target language code "${o}" is not valid. Must be one of: {${e.language_codes.join(", ")}}`);if(void 0!==r){if(!e.language_codes.includes(r))throw new Error(`Source language code "${r}" is not valid. Must be one of: {${e.language_codes.join(", ")}}`);for(const t of e.post_processor.config.single)if("SpecialToken"in t&&e.languageRegex.test(t.SpecialToken.id)){t.SpecialToken.id=e.lang_to_token(r);break}}return s.forced_bos_token_id=e.model.convert_tokens_to_ids([e.lang_to_token(o)])[0],e._call(t,n)}class Ze extends be{constructor(e,t){super(e,t),this.languageRegex=/^[a-z]{3}_[A-Z][a-z]{3}$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))),this.lang_to_token=e=>e}_build_translation_inputs(e,t,n){return Ke(this,e,t,n)}}class et extends be{constructor(e,t){super(e,t),this.languageRegex=/^__[a-z]{2,3}__$/,this.language_codes=this.special_tokens.filter((e=>this.languageRegex.test(e))).map((e=>e.slice(2,-2))),this.lang_to_token=e=>`__${e}__`}_build_translation_inputs(e,t,n){return Ke(this,e,t,n)}}class tt extends be{get timestamp_begin(){return this.model.convert_tokens_to_ids(["<|notimestamps|>"])[0]+1}_decode_asr(e,{return_timestamps:t=!1,return_language:n=!1,time_precision:s=null,force_full_sequences:r=!0}={}){if(null===s)throw Error("Must specify time_precision");let o=null;const a="word"===t;function l(){return{language:o,timestamp:[null,null],text:""}}const c=[];let u=l(),p=0;const h=this.timestamp_begin;let _=[],m=[],f=!1,g=null;const w=new Set(this.all_special_ids);for(const n of e){const e=n.tokens,r=a?n.token_timestamps:null;let M=null,y=h;if("stride"in n){const[t,r,o]=n.stride;if(p-=r,g=t-o,r&&(y=r/s+h),o)for(let t=e.length-1;t>=0;--t){const n=Number(e[t]);if(n>=h){if(null!==M&&(n-h)*s<g)break;M=n}}}let x=[],k=[];for(let n=0;n<e.length;++n){const g=Number(e[n]);if(w.has(g)){const e=this.decode([g]),n=d.WHISPER_LANGUAGE_MAPPING.get(e.slice(2,-2));if(void 0!==n){if(null!==o&&n!==o&&!t){_.push(x);const e=this.findLongestCommonSequence(_)[0],t=this.decode(e);u.text=t,c.push(u),_=[],x=[],u=l()}o=u.language=n}}else if(g>=h){const e=(g-h)*s+p,t=(0,i.round)(e,2);if(null!==M&&g>=M)f=!0;else if(f||_.length>0&&g<y)f=!1;else if(null===u.timestamp[0])u.timestamp[0]=t;else if(t===u.timestamp[0]);else{u.timestamp[1]=t,_.push(x),a&&m.push(k);const[e,n]=this.findLongestCommonSequence(_,m),s=this.decode(e);u.text=s,a&&(u.words=this.collateWordTimestamps(e,n,o)),c.push(u),_=[],x=[],m=[],k=[],u=l()}}else if(x.push(g),a){let e,t=(0,i.round)(r[n]+p,2);if(n+1<r.length){e=(0,i.round)(r[n+1]+p,2);const o=this.decode([g]);b.test(o)&&(e=(0,i.round)(Math.min(t+s,e),2))}else e=null;k.push([t,e])}}if("stride"in n){const[e,t,s]=n.stride;p+=e-s}x.length>0?(_.push(x),a&&m.push(k)):_.every((e=>0===e.length))&&(u=l(),_=[],x=[],m=[],k=[])}if(_.length>0){if(r&&t)throw new Error("Whisper did not predict an ending timestamp, which can happen if audio is cut off in the middle of a word. Also make sure WhisperTimeStampLogitsProcessor was used during generation.");const[e,n]=this.findLongestCommonSequence(_,m),s=this.decode(e);u.text=s,a&&(u.words=this.collateWordTimestamps(e,n,o)),c.push(u)}let M=Object.create(null);const y=c.map((e=>e.text)).join("");if(t||n){for(let e=0;e<c.length;++e){const s=c[e];t||delete s.timestamp,n||delete s.language}if(a){const e=[];for(const t of c)for(const n of t.words)e.push(n);M={chunks:e}}else M={chunks:c}}return[y,M]}findLongestCommonSequence(e,t=null){let n=e[0],s=n.length,r=[];const o=Array.isArray(t)&&t.length>0;let i=o?[]:null,a=o?t[0]:null;for(let l=1;l<e.length;++l){const c=e[l];let d=0,u=[s,s,0,0];const p=c.length;for(let e=1;e<s+p;++e){const r=Math.max(0,s-e),i=Math.min(s,s+p-e),h=n.slice(r,i),_=Math.max(0,e-s),m=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=o?h.filter(((e,n)=>e===f[n]&&a[r+n]<=t[l][_+n])).length:h.filter(((e,t)=>e===f[t])).length;const w=g/e+e/1e4;g>1&&w>d&&(d=w,u=[r,i,_,m])}const[h,_,m,f]=u,g=Math.floor((_+h)/2),w=Math.floor((f+m)/2);r.push(...n.slice(0,g)),n=c.slice(w),s=n.length,o&&(i.push(...a.slice(0,g)),a=t[l].slice(w))}return r.push(...n),o?(i.push(...a),[r,i]):[r,[]]}collateWordTimestamps(e,t,n){const[s,r,o]=this.combineTokensIntoWords(e,n),i=[];for(let e=0;e<s.length;++e){const n=o[e];i.push({text:s[e],timestamp:[t[n.at(0)][0],t[n.at(-1)][1]]})}return i}combineTokensIntoWords(e,t,n="\"'“¡¿([{-",s="\"'.。,，!！?？:：”)]}、"){let r,o,i;return["chinese","japanese","thai","lao","myanmar"].includes(t=t??"english")?[r,o,i]=this.splitTokensOnUnicode(e):[r,o,i]=this.splitTokensOnSpaces(e),this.mergePunctuations(r,o,i,n,s)}decode(e,t){let n;return t?.decode_with_timestamps?(e instanceof a.Tensor&&(e=m(e)),n=this.decodeWithTimestamps(e,t)):n=super.decode(e,t),n}decodeWithTimestamps(e,t){const n=t?.time_precision??.02,s=Array.from(this.all_special_ids).at(-1)+1;let r=[[]];for(let t of e)if(t=Number(t),t>=s){const e=((t-s)*n).toFixed(2);r.push(`<|${e}|>`),r.push([])}else r[r.length-1].push(t);return r=r.map((e=>"string"==typeof e?e:super.decode(e,t))),r.join("")}splitTokensOnUnicode(e){const t=this.decode(e,{decode_with_timestamps:!0}),n=[],s=[],r=[];let o=[],i=[],a=0;for(let l=0;l<e.length;++l){const c=e[l];o.push(c),i.push(l);const d=this.decode(o,{decode_with_timestamps:!0});d.includes("�")&&"�"!==t[a+d.indexOf("�")]||(n.push(d),s.push(o),r.push(i),o=[],i=[],a+=d.length)}return[n,s,r]}splitTokensOnSpaces(e){const[t,n,s]=this.splitTokensOnUnicode(e),r=[],o=[],i=[],a=new RegExp(`^[${M}]$`,"gu");for(let e=0;e<t.length;++e){const l=t[e],c=n[e],d=s[e],u=c[0]>=this.model.tokens_to_ids.get("<|endoftext|>"),p=l.startsWith(" "),h=l.trim(),_=a.test(h);if(u||p||_||0===r.length)r.push(l),o.push(c),i.push(d);else{const e=r.length-1;r[e]+=l,o[e].push(...c),i[e].push(...d)}}return[r,o,i]}mergePunctuations(e,t,n,s,o){const i=structuredClone(e),a=structuredClone(t),l=structuredClone(n);let c=i.length-2,d=i.length-1;for(;c>=0;)i[c].startsWith(" ")&&s.includes(i[c].trim())?(i[d]=i[c]+i[d],a[d]=(0,r.mergeArrays)(a[c],a[d]),l[d]=(0,r.mergeArrays)(l[c],l[d]),i[c]="",a[c]=[],l[c]=[]):d=c,--c;for(c=0,d=1;d<i.length;)!i[c].endsWith(" ")&&o.includes(i[d])?(i[c]+=i[d],a[c]=(0,r.mergeArrays)(a[c],a[d]),l[c]=(0,r.mergeArrays)(l[c],l[d]),i[d]="",a[d]=[],l[d]=[]):c=d,++d;return[i.filter((e=>e)),a.filter((e=>e.length>0)),l.filter((e=>e.length>0))]}get_decoder_prompt_ids({language:e=null,task:t=null,no_timestamps:n=!0}={}){const s=[];if(e){const t=(0,d.whisper_language_to_code)(e),n=this.model.tokens_to_ids.get(`<|${t}|>`);if(void 0===n)throw new Error(`Unable to find language "${t}" in model vocabulary. Please report this issue at ${u.GITHUB_ISSUE_URL}.`);s.push(n)}else s.push(null);if(t){if("transcribe"!==(t=t.toLowerCase())&&"translate"!==t)throw new Error(`Task "${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}.`);s.push(e)}else s.push(null);if(n){const e=this.model.tokens_to_ids.get("<|notimestamps|>");if(void 0===e)throw new Error(`Unable to find "<|notimestamps|>" in model vocabulary. Please report this issue at ${u.GITHUB_ISSUE_URL}.`);s.push(e)}return s.map(((e,t)=>[t+1,e])).filter((e=>null!==e[1]))}}class nt extends be{}class st extends be{}class rt extends be{}class ot extends be{constructor(e,t){super(e,t),this.languageRegex=/^(>>\w+<<)\s*/g,this.supported_language_codes=this.model.vocab.filter((e=>this.languageRegex.test(e))),console.warn('WARNING: `MarianTokenizer` is not yet supported by Hugging Face\'s "fast" tokenizers library. Therefore, you may experience slightly inaccurate results.')}_encode_text(e){if(null===e)return null;const[t,...n]=e.trim().split(this.languageRegex);if(0===n.length)return super._encode_text(t);if(2===n.length){const[e,t]=n;return this.supported_language_codes.includes(e)||console.warn(`Unsupported language code "${e}" detected, which may lead to unexpected behavior. Should be one of: ${JSON.stringify(this.supported_language_codes)}`),(0,r.mergeArrays)([e],super._encode_text(t))}}}class it extends be{}class at extends be{}class lt extends be{}class ct extends be{}class dt extends be{}class ut extends be{constructor(e,t){super(e,t),this.decoder=new ce({})}}class pt extends be{}class ht{static TOKENIZER_CLASS_MAPPING={T5Tokenizer:Ie,DistilBertTokenizer:Ae,CamembertTokenizer:Ee,DebertaTokenizer:ve,DebertaV2Tokenizer:Ce,BertTokenizer:ye,HerbertTokenizer:Fe,ConvBertTokenizer:Pe,RoFormerTokenizer:Se,XLMTokenizer:ze,ElectraTokenizer:Le,MobileBertTokenizer:ke,SqueezeBertTokenizer:Te,AlbertTokenizer:xe,GPT2Tokenizer:Be,BartTokenizer:Ne,MBartTokenizer:Oe,MBart50Tokenizer:je,RobertaTokenizer:De,WhisperTokenizer:tt,CodeGenTokenizer:nt,CLIPTokenizer:st,SiglipTokenizer:rt,MarianTokenizer:ot,BloomTokenizer:Ve,NllbTokenizer:Ze,M2M100Tokenizer:et,LlamaTokenizer:Ge,CodeLlamaTokenizer:$e,XLMRobertaTokenizer:qe,MPNetTokenizer:Ue,FalconTokenizer:We,GPTNeoXTokenizer:Xe,EsmTokenizer:Qe,Wav2Vec2CTCTokenizer:it,BlenderbotTokenizer:at,BlenderbotSmallTokenizer:lt,SpeechT5Tokenizer:ct,NougatTokenizer:dt,VitsTokenizer:ut,Qwen2Tokenizer:He,GemmaTokenizer:Ye,Grok1Tokenizer:Je,CohereTokenizer:pt,PreTrainedTokenizer:be};static async from_pretrained(e,{progress_callback:t=null,config:n=null,cache_dir:s=null,local_files_only:r=!1,revision:o="main",legacy:i=null}={}){const[a,l]=await p(e,{progress_callback:t,config:n,cache_dir:s,local_files_only:r,revision:o,legacy:i}),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=be),new d(a,l)}}},"./src/utils/audio.js":
@@ -163,5 +163,5 @@
   \*****************************/(e,t,n)=>{"use strict";n.r(t),n.d(t,{Tensor:()=>a,cat:()=>M,full:()=>T,full_like:()=>v,interpolate:()=>c,interpolate_4d:()=>d,layer_norm:()=>m,matmul:()=>u,mean:()=>x,mean_pooling:()=>_,ones:()=>C,ones_like:()=>F,permute:()=>l,quantize_embeddings:()=>A,rfft:()=>p,stack:()=>b,std_mean:()=>y,topk:()=>h,zeros:()=>P,zeros_like:()=>S});var s=n(/*! ./maths.js */"./src/utils/maths.js"),r=n(/*! ../backends/onnx.js */"./src/backends/onnx.js"),o=n(/*! ../ops/registry.js */"./src/ops/registry.js");const i=Object.freeze({float32:Float32Array,float16:Uint16Array,float64:Float64Array,string:Array,int8:Int8Array,uint8:Uint8Array,int16:Int16Array,uint16:Uint16Array,int32:Int32Array,uint32:Uint32Array,int64:BigInt64Array,uint64:BigUint64Array,bool:Uint8Array});class a{get dims(){return this.ort_tensor.dims}set dims(e){this.ort_tensor.dims=e}get type(){return this.ort_tensor.type}get data(){return this.ort_tensor.data}get size(){return this.ort_tensor.size}get location(){return this.ort_tensor.location}ort_tensor;constructor(...e){return(0,r.isONNXTensor)(e[0])?this.ort_tensor=e[0]:this.ort_tensor=new r.Tensor(e[0],e[1],e[2]),new Proxy(this,{get:(e,t)=>{if("string"==typeof t){let n=Number(t);if(Number.isInteger(n))return e._getitem(n)}return e[t]},set:(e,t,n)=>e[t]=n})}dispose(){this.ort_tensor.dispose()}*[Symbol.iterator](){const[e,...t]=this.dims;if(t.length>0){const n=t.reduce(((e,t)=>e*t));for(let s=0;s<e;++s)yield this._subarray(s,n,t)}else yield*this.data}_getitem(e){const[t,...n]=this.dims;if(e=w(e,t),n.length>0){const t=n.reduce(((e,t)=>e*t));return this._subarray(e,t,n)}return new a(this.type,[this.data[e]],n)}indexOf(e){const t=this.data;for(let n=0;n<t.length;++n)if(t[n]==e)return n;return-1}_subarray(e,t,n){const s=e*t,r=(e+1)*t,o="subarray"in this.data?this.data.subarray(s,r):this.data.slice(s,r);return new a(this.type,o,n)}item(){const e=this.data;if(1!==e.length)throw new Error(`a Tensor with ${e.length} elements cannot be converted to Scalar`);return e[0]}tolist(){return function(e,t){const n=e.length,s=t.reduce(((e,t)=>e*t));if(n!==s)throw Error(`cannot reshape array of size ${n} into shape (${t})`);let r=e;for(let e=t.length-1;e>=0;e--)r=r.reduce(((n,s)=>{let r=n[n.length-1];return r.length<t[e]?r.push(s):n.push([s]),n}),[[]]);return r[0]}(this.data,this.dims)}sigmoid(){return this.clone().sigmoid_()}sigmoid_(){const e=this.data;for(let t=0;t<e.length;++t)e[t]=1/(1+Math.exp(-e[t]));return this}mul(e){return this.clone().mul_(e)}mul_(e){const t=this.data;for(let n=0;n<t.length;++n)t[n]*=e;return this}div(e){return this.clone().div_(e)}div_(e){const t=this.data;for(let n=0;n<t.length;++n)t[n]/=e;return this}add(e){return this.clone().add_(e)}add_(e){const t=this.data;for(let n=0;n<t.length;++n)t[n]+=e;return this}sub(e){return this.clone().sub_(e)}sub_(e){const t=this.data;for(let n=0;n<t.length;++n)t[n]-=e;return this}clone(){return new a(this.type,this.data.slice(),this.dims.slice())}slice(...e){const t=[],n=[];for(let s=0;s<this.dims.length;++s){let r=e[s];if(null==r)n.push([0,this.dims[s]]),t.push(this.dims[s]);else if("number"==typeof r)r=w(r,this.dims[s],s),n.push([r,r+1]);else{if(!Array.isArray(r)||2!==r.length)throw new Error(`Invalid slice: ${r}`);{let[e,o]=r;if(e=null===e?0:w(e,this.dims[s],s,!1),o=null===o?this.dims[s]:w(o,this.dims[s],s,!1),e>o)throw new Error(`Invalid slice: ${r}`);const i=[Math.max(e,0),Math.min(o,this.dims[s])];n.push(i),t.push(i[1]-i[0])}}}const s=n.map((([e,t])=>t-e)),r=s.reduce(((e,t)=>e*t)),o=this.data,i=new o.constructor(r),l=this.stride();for(let e=0;e<r;++e){let t=0;for(let r=s.length-1,o=e;r>=0;--r){const e=s[r];t+=(o%e+n[r][0])*l[r],o=Math.floor(o/e)}i[e]=o[t]}return new a(this.type,i,t)}permute(...e){return l(this,e)}transpose(...e){return this.permute(...e)}sum(e=null,t=!1){return this.norm(1,e,t)}norm(e="fro",t=null,n=!1){if("fro"===e)e=2;else if("string"==typeof e)throw Error(`Unsupported norm: ${e}`);const s=this.data;if(null===t){let t=s.reduce(((t,n)=>t+n**e),0)**(1/e);return new a(this.type,[t],[])}t=w(t,this.dims.length);const r=this.dims.slice();r[t]=1;const o=new s.constructor(s.length/this.dims[t]);for(let n=0;n<s.length;++n){let i=0;for(let e=this.dims.length-1,s=n,o=1;e>=0;--e){const n=this.dims[e];if(e!==t){i+=s%n*o,o*=r[e]}s=Math.floor(s/n)}o[i]+=s[n]**e}if(1!==e)for(let t=0;t<o.length;++t)o[t]=o[t]**(1/e);return n||r.splice(t,1),new a(this.type,o,r)}normalize_(e=2,t=1){t=w(t,this.dims.length);const n=this.norm(e,t,!0),s=this.data,r=n.data;for(let e=0;e<s.length;++e){let n=0;for(let s=this.dims.length-1,r=e,o=1;s>=0;--s){const e=this.dims[s];if(s!==t){n+=r%e*o,o*=this.dims[s]}r=Math.floor(r/e)}s[e]/=r[n]}return this}normalize(e=2,t=1){return this.clone().normalize_(e,t)}stride(){return function(e){const t=new Array(e.length);for(let n=e.length-1,s=1;n>=0;--n)t[n]=s,s*=e[n];return t}(this.dims)}squeeze(e=null){return new a(this.type,this.data,f(this.dims,e))}squeeze_(e=null){return this.dims=f(this.dims,e),this}unsqueeze(e=null){return new a(this.type,this.data,g(this.dims,e))}unsqueeze_(e=null){return this.dims=g(this.dims,e),this}flatten_(e=0,t=-1){t=(t+this.dims.length)%this.dims.length;let n=this.dims.slice(0,e),s=this.dims.slice(e,t+1),r=this.dims.slice(t+1);return this.dims=[...n,s.reduce(((e,t)=>e*t),1),...r],this}flatten(e=0,t=-1){return this.clone().flatten_(e,t)}view(...e){let t=-1;for(let n=0;n<e.length;++n)if(-1===e[n]){if(-1!==t)throw new Error("Only one dimension can be inferred");t=n}const n=this.data;if(-1!==t){const s=e.reduce(((e,n,s)=>s!==t?e*n:e),1);e[t]=n.length/s}return new a(this.type,n,e)}neg_(){const e=this.data;for(let t=0;t<e.length;++t)e[t]=-e[t];return this}neg(){return this.clone().neg_()}clamp_(e,t){const n=this.data;for(let s=0;s<n.length;++s)n[s]=Math.min(Math.max(n[s],e),t);return this}clamp(e,t){return this.clone().clamp_(e,t)}round_(){const e=this.data;for(let t=0;t<e.length;++t)e[t]=Math.round(e[t]);return this}round(){return this.clone().round_()}mean(e=null,t=!1){return x(this,e,t)}to(e){if(this.type===e)return this;if(!i.hasOwnProperty(e))throw new Error(`Unsupported type: ${e}`);return new a(e,i[e].from(this.data),this.dims)}}function l(e,t){const[n,r]=(0,s.permute_data)(e.data,e.dims,t);return new a(e.type,n,r)}function c(e,[t,n],r="bilinear",o=!1){const i=e.dims.at(-3)??1,l=e.dims.at(-2),c=e.dims.at(-1);let d=(0,s.interpolate_data)(e.data,[i,l,c],[t,n],r,o);return new a(e.type,d,[i,t,n])}async function d(e,{size:t=null,mode:n="bilinear"}={}){if(4!==e.dims.length)throw new Error("`interpolate_4d` currently only supports 4D input.");if(!t)throw new Error("`interpolate_4d` requires a `size` argument.");let s,r;if(2===t.length)s=[...e.dims.slice(0,2),...t];else if(3===t.length)s=[e.dims[0],...t];else{if(4!==t.length)throw new Error("`size` must be of length 2, 3, or 4.");s=t}if("bilinear"===n)r=await o.TensorOpRegistry.bilinear_interpolate_4d;else{if("bicubic"!==n)throw new Error(`Unsupported mode: ${n}`);r=await o.TensorOpRegistry.bicubic_interpolate_4d}const i=new a("int64",new BigInt64Array(s.map(BigInt)),[s.length]);return await r({x:e,s:i})}async function u(e,t){const n=await o.TensorOpRegistry.matmul;return await n({a:e,b:t})}async function p(e,t){const n=await o.TensorOpRegistry.rfft;return await n({x:e,a:t})}async function h(e,t){const n=await o.TensorOpRegistry.top_k;return t=null===t?e.dims.at(-1):Math.min(t,e.dims.at(-1)),await n({x:e,k:new a("int64",[BigInt(t)],[1])})}function _(e,t){const n=e.data,s=t.data,r=[e.dims[0],e.dims[2]],o=new n.constructor(r[0]*r[1]),[i,l,c]=e.dims;let d=0;for(let e=0;e<i;++e){const t=e*c*l;for(let r=0;r<c;++r){let i=0,a=0;const u=e*l,p=t+r;for(let e=0;e<l;++e){const t=Number(s[u+e]);a+=t,i+=n[p+e*c]*t}const h=i/a;o[d++]=h}}return new a(e.type,o,r)}function m(e,t,{eps:n=1e-5}={}){if(2!==e.dims.length)throw new Error("`layer_norm` currently only supports 2D input.");const[s,r]=e.dims;if(1!==t.length&&t[0]!==r)throw new Error("`normalized_shape` must be a 1D array with shape `[input.dims[1]]`.");const[o,i]=y(e,1,0,!0),l=o.data,c=i.data,d=e.data,u=new d.constructor(d.length);for(let e=0;e<s;++e){const t=e*r;for(let s=0;s<r;++s){const r=t+s;u[r]=(d[r]-c[e])/(l[e]+n)}}return new a(e.type,u,e.dims)}function f(e,t){return e=e.slice(),null===t?e=e.filter((e=>1!==e)):"number"==typeof t?1===e[t]&&e.splice(t,1):Array.isArray(t)&&(e=e.filter(((e,n)=>1!==e||!t.includes(n)))),e}function g(e,t){return t=w(t,e.length+1),(e=e.slice()).splice(t,0,1),e}function w(e,t,n=null,s=!0){if(s&&(e<-t||e>=t))throw new Error(`IndexError: index ${e} is out of bounds for dimension${null===n?"":" "+n} with size ${t}`);return e<0&&(e=(e%t+t)%t),e}function M(e,t=0){t=w(t,e[0].dims.length);const n=e[0].dims.slice();n[t]=e.reduce(((e,n)=>e+n.dims[t]),0);const s=n.reduce(((e,t)=>e*t),1),r=new e[0].data.constructor(s),o=e[0].type;if(0===t){let t=0;for(const n of e){const e=n.data;r.set(e,t),t+=e.length}}else{let s=0;for(let o=0;o<e.length;++o){const{data:i,dims:a}=e[o];for(let e=0;e<i.length;++e){let o=0;for(let r=a.length-1,i=e,l=1;r>=0;--r){const e=a[r];let c=i%e;r===t&&(c+=s),o+=c*l,l*=n[r],i=Math.floor(i/e)}r[o]=i[e]}s+=a[t]}}return new a(o,r,n)}function b(e,t=0){return M(e.map((e=>e.unsqueeze(t))),t)}function y(e,t=null,n=1,s=!1){const r=e.data,o=e.dims;if(null===t){const t=r.reduce(((e,t)=>e+t),0)/r.length,s=Math.sqrt(r.reduce(((e,n)=>e+(n-t)**2),0)/(r.length-n)),o=new a(e.type,[t],[]);return[new a(e.type,[s],[]),o]}const i=x(e,t=w(t,o.length),s),l=i.data,c=o.slice();c[t]=1;const d=new r.constructor(r.length/o[t]);for(let e=0;e<r.length;++e){let n=0;for(let s=o.length-1,r=e,i=1;s>=0;--s){const e=o[s];if(s!==t){n+=r%e*i,i*=c[s]}r=Math.floor(r/e)}d[n]+=(r[e]-l[n])**2}for(let e=0;e<d.length;++e)d[e]=Math.sqrt(d[e]/(o[t]-n));s||c.splice(t,1);return[new a(e.type,d,c),i]}function x(e,t=null,n=!1){const s=e.data;if(null===t){const t=s.reduce(((e,t)=>e+t),0);return new a(e.type,[t/s.length],[])}const r=e.dims;t=w(t,r.length);const o=r.slice();o[t]=1;const i=new s.constructor(s.length/r[t]);for(let e=0;e<s.length;++e){let n=0;for(let s=r.length-1,i=e,a=1;s>=0;--s){const e=r[s];if(s!==t){n+=i%e*a,a*=o[s]}i=Math.floor(i/e)}i[n]+=s[e]}if(1!==r[t])for(let e=0;e<i.length;++e)i[e]=i[e]/r[t];return n||o.splice(t,1),new a(e.type,i,o)}function k(e,t,n,s){const r=e.reduce(((e,t)=>e*t),1);return new a(n,new s(r).fill(t),e)}function T(e,t){let n,s;if("number"==typeof t)n="float32",s=Float32Array;else{if("bigint"!=typeof t)throw new Error("Unsupported data type: "+typeof t);n="int64",s=BigInt64Array}return k(e,t,n,s)}function v(e,t){return T(e.dims,t)}function C(e){return k(e,1n,"int64",BigInt64Array)}function F(e){return C(e.dims)}function P(e){return k(e,0n,"int64",BigInt64Array)}function S(e){return P(e.dims)}function A(e,t){if(2!==e.dims.length)throw new Error("The tensor must have 2 dimensions");if(e.dims.at(-1)%8!=0)throw new Error("The last dimension of the tensor must be a multiple of 8");if(!["binary","ubinary"].includes(t))throw new Error("The precision must be either 'binary' or 'ubinary'");const n="binary"===t,s=n?"int8":"uint8",r=n?Int8Array:Uint8Array,o=e.data,i=new r(o.length/8);for(let e=0;e<o.length;++e){const t=o[e]>0?1:0,s=Math.floor(e/8),r=e%8;i[s]|=t<<7-r,n&&0===r&&(i[s]-=128)}return new a(s,i,[e.dims[0],e.dims[1]/8])}}},s={};function r(e){var t=s[e];if(void 0!==t)return t.exports;var o=s[e]={exports:{}};return n[e](o,o.exports,r),o.exports}t=Object.getPrototypeOf?e=>Object.getPrototypeOf(e):e=>e.__proto__,r.t=function(n,s){if(1&s&&(n=this(n)),8&s)return n;if("object"==typeof n&&n){if(4&s&&n.__esModule)return n;if(16&s&&"function"==typeof n.then)return n}var o=Object.create(null);r.r(o);var i={};e=e||[null,t({}),t([]),t(t)];for(var a=2&s&&n;"object"==typeof a&&!~e.indexOf(a);a=t(a))Object.getOwnPropertyNames(a).forEach((e=>i[e]=()=>n[e]));return i.default=()=>n,r.d(o,i),o},r.d=(e,t)=>{for(var n in t)r.o(t,n)&&!r.o(e,n)&&Object.defineProperty(e,n,{enumerable:!0,get:t[n]})},r.o=(e,t)=>Object.prototype.hasOwnProperty.call(e,t),r.r=e=>{"undefined"!=typeof Symbol&&Symbol.toStringTag&&Object.defineProperty(e,Symbol.toStringTag,{value:"Module"}),Object.defineProperty(e,"__esModule",{value:!0})};var o={};(()=>{"use strict";
 /*!*****************************!*\
   !*** ./src/transformers.js ***!
-  \*****************************/r.r(o),r.d(o,{ASTFeatureExtractor:()=>i.ASTFeatureExtractor,ASTForAudioClassification:()=>n.ASTForAudioClassification,ASTModel:()=>n.ASTModel,ASTPreTrainedModel:()=>n.ASTPreTrainedModel,AlbertForMaskedLM:()=>n.AlbertForMaskedLM,AlbertForQuestionAnswering:()=>n.AlbertForQuestionAnswering,AlbertForSequenceClassification:()=>n.AlbertForSequenceClassification,AlbertModel:()=>n.AlbertModel,AlbertPreTrainedModel:()=>n.AlbertPreTrainedModel,AlbertTokenizer:()=>s.AlbertTokenizer,AudioClassificationPipeline:()=>t.AudioClassificationPipeline,AutoConfig:()=>a.AutoConfig,AutoModel:()=>n.AutoModel,AutoModelForAudioClassification:()=>n.AutoModelForAudioClassification,AutoModelForAudioFrameClassification:()=>n.AutoModelForAudioFrameClassification,AutoModelForCTC:()=>n.AutoModelForCTC,AutoModelForCausalLM:()=>n.AutoModelForCausalLM,AutoModelForDepthEstimation:()=>n.AutoModelForDepthEstimation,AutoModelForDocumentQuestionAnswering:()=>n.AutoModelForDocumentQuestionAnswering,AutoModelForImageClassification:()=>n.AutoModelForImageClassification,AutoModelForImageFeatureExtraction:()=>n.AutoModelForImageFeatureExtraction,AutoModelForImageMatting:()=>n.AutoModelForImageMatting,AutoModelForImageSegmentation:()=>n.AutoModelForImageSegmentation,AutoModelForImageToImage:()=>n.AutoModelForImageToImage,AutoModelForMaskGeneration:()=>n.AutoModelForMaskGeneration,AutoModelForMaskedLM:()=>n.AutoModelForMaskedLM,AutoModelForNormalEstimation:()=>n.AutoModelForNormalEstimation,AutoModelForObjectDetection:()=>n.AutoModelForObjectDetection,AutoModelForQuestionAnswering:()=>n.AutoModelForQuestionAnswering,AutoModelForSemanticSegmentation:()=>n.AutoModelForSemanticSegmentation,AutoModelForSeq2SeqLM:()=>n.AutoModelForSeq2SeqLM,AutoModelForSequenceClassification:()=>n.AutoModelForSequenceClassification,AutoModelForSpeechSeq2Seq:()=>n.AutoModelForSpeechSeq2Seq,AutoModelForTextToSpectrogram:()=>n.AutoModelForTextToSpectrogram,AutoModelForTextToWaveform:()=>n.AutoModelForTextToWaveform,AutoModelForTokenClassification:()=>n.AutoModelForTokenClassification,AutoModelForVision2Seq:()=>n.AutoModelForVision2Seq,AutoModelForXVector:()=>n.AutoModelForXVector,AutoModelForZeroShotObjectDetection:()=>n.AutoModelForZeroShotObjectDetection,AutoProcessor:()=>i.AutoProcessor,AutoTokenizer:()=>s.AutoTokenizer,AutomaticSpeechRecognitionPipeline:()=>t.AutomaticSpeechRecognitionPipeline,BartForConditionalGeneration:()=>n.BartForConditionalGeneration,BartForSequenceClassification:()=>n.BartForSequenceClassification,BartModel:()=>n.BartModel,BartPretrainedModel:()=>n.BartPretrainedModel,BartTokenizer:()=>s.BartTokenizer,BaseModelOutput:()=>n.BaseModelOutput,BaseStreamer:()=>p.BaseStreamer,BeitFeatureExtractor:()=>i.BeitFeatureExtractor,BeitForImageClassification:()=>n.BeitForImageClassification,BeitModel:()=>n.BeitModel,BeitPreTrainedModel:()=>n.BeitPreTrainedModel,BertForMaskedLM:()=>n.BertForMaskedLM,BertForQuestionAnswering:()=>n.BertForQuestionAnswering,BertForSequenceClassification:()=>n.BertForSequenceClassification,BertForTokenClassification:()=>n.BertForTokenClassification,BertModel:()=>n.BertModel,BertPreTrainedModel:()=>n.BertPreTrainedModel,BertTokenizer:()=>s.BertTokenizer,BitImageProcessor:()=>i.BitImageProcessor,BlenderbotForConditionalGeneration:()=>n.BlenderbotForConditionalGeneration,BlenderbotModel:()=>n.BlenderbotModel,BlenderbotPreTrainedModel:()=>n.BlenderbotPreTrainedModel,BlenderbotSmallForConditionalGeneration:()=>n.BlenderbotSmallForConditionalGeneration,BlenderbotSmallModel:()=>n.BlenderbotSmallModel,BlenderbotSmallPreTrainedModel:()=>n.BlenderbotSmallPreTrainedModel,BlenderbotSmallTokenizer:()=>s.BlenderbotSmallTokenizer,BlenderbotTokenizer:()=>s.BlenderbotTokenizer,BloomForCausalLM:()=>n.BloomForCausalLM,BloomModel:()=>n.BloomModel,BloomPreTrainedModel:()=>n.BloomPreTrainedModel,BloomTokenizer:()=>s.BloomTokenizer,CLIPFeatureExtractor:()=>i.CLIPFeatureExtractor,CLIPImageProcessor:()=>i.CLIPImageProcessor,CLIPModel:()=>n.CLIPModel,CLIPPreTrainedModel:()=>n.CLIPPreTrainedModel,CLIPSegForImageSegmentation:()=>n.CLIPSegForImageSegmentation,CLIPSegModel:()=>n.CLIPSegModel,CLIPSegPreTrainedModel:()=>n.CLIPSegPreTrainedModel,CLIPTextModelWithProjection:()=>n.CLIPTextModelWithProjection,CLIPTokenizer:()=>s.CLIPTokenizer,CLIPVisionModelWithProjection:()=>n.CLIPVisionModelWithProjection,CamembertForMaskedLM:()=>n.CamembertForMaskedLM,CamembertForQuestionAnswering:()=>n.CamembertForQuestionAnswering,CamembertForSequenceClassification:()=>n.CamembertForSequenceClassification,CamembertForTokenClassification:()=>n.CamembertForTokenClassification,CamembertModel:()=>n.CamembertModel,CamembertPreTrainedModel:()=>n.CamembertPreTrainedModel,CamembertTokenizer:()=>s.CamembertTokenizer,CausalLMOutput:()=>n.CausalLMOutput,CausalLMOutputWithPast:()=>n.CausalLMOutputWithPast,ChineseCLIPFeatureExtractor:()=>i.ChineseCLIPFeatureExtractor,ChineseCLIPModel:()=>n.ChineseCLIPModel,ChineseCLIPPreTrainedModel:()=>n.ChineseCLIPPreTrainedModel,ClapAudioModelWithProjection:()=>n.ClapAudioModelWithProjection,ClapFeatureExtractor:()=>i.ClapFeatureExtractor,ClapModel:()=>n.ClapModel,ClapPreTrainedModel:()=>n.ClapPreTrainedModel,ClapTextModelWithProjection:()=>n.ClapTextModelWithProjection,CodeGenForCausalLM:()=>n.CodeGenForCausalLM,CodeGenModel:()=>n.CodeGenModel,CodeGenPreTrainedModel:()=>n.CodeGenPreTrainedModel,CodeGenTokenizer:()=>s.CodeGenTokenizer,CodeLlamaTokenizer:()=>s.CodeLlamaTokenizer,CohereForCausalLM:()=>n.CohereForCausalLM,CohereModel:()=>n.CohereModel,CoherePreTrainedModel:()=>n.CoherePreTrainedModel,CohereTokenizer:()=>s.CohereTokenizer,ConvBertForMaskedLM:()=>n.ConvBertForMaskedLM,ConvBertForQuestionAnswering:()=>n.ConvBertForQuestionAnswering,ConvBertForSequenceClassification:()=>n.ConvBertForSequenceClassification,ConvBertForTokenClassification:()=>n.ConvBertForTokenClassification,ConvBertModel:()=>n.ConvBertModel,ConvBertPreTrainedModel:()=>n.ConvBertPreTrainedModel,ConvBertTokenizer:()=>s.ConvBertTokenizer,ConvNextFeatureExtractor:()=>i.ConvNextFeatureExtractor,ConvNextForImageClassification:()=>n.ConvNextForImageClassification,ConvNextImageProcessor:()=>i.ConvNextImageProcessor,ConvNextModel:()=>n.ConvNextModel,ConvNextPreTrainedModel:()=>n.ConvNextPreTrainedModel,ConvNextV2ForImageClassification:()=>n.ConvNextV2ForImageClassification,ConvNextV2Model:()=>n.ConvNextV2Model,ConvNextV2PreTrainedModel:()=>n.ConvNextV2PreTrainedModel,DPTFeatureExtractor:()=>i.DPTFeatureExtractor,DPTForDepthEstimation:()=>n.DPTForDepthEstimation,DPTImageProcessor:()=>i.DPTImageProcessor,DPTModel:()=>n.DPTModel,DPTPreTrainedModel:()=>n.DPTPreTrainedModel,DebertaForMaskedLM:()=>n.DebertaForMaskedLM,DebertaForQuestionAnswering:()=>n.DebertaForQuestionAnswering,DebertaForSequenceClassification:()=>n.DebertaForSequenceClassification,DebertaForTokenClassification:()=>n.DebertaForTokenClassification,DebertaModel:()=>n.DebertaModel,DebertaPreTrainedModel:()=>n.DebertaPreTrainedModel,DebertaTokenizer:()=>s.DebertaTokenizer,DebertaV2ForMaskedLM:()=>n.DebertaV2ForMaskedLM,DebertaV2ForQuestionAnswering:()=>n.DebertaV2ForQuestionAnswering,DebertaV2ForSequenceClassification:()=>n.DebertaV2ForSequenceClassification,DebertaV2ForTokenClassification:()=>n.DebertaV2ForTokenClassification,DebertaV2Model:()=>n.DebertaV2Model,DebertaV2PreTrainedModel:()=>n.DebertaV2PreTrainedModel,DebertaV2Tokenizer:()=>s.DebertaV2Tokenizer,DeiTFeatureExtractor:()=>i.DeiTFeatureExtractor,DeiTForImageClassification:()=>n.DeiTForImageClassification,DeiTModel:()=>n.DeiTModel,DeiTPreTrainedModel:()=>n.DeiTPreTrainedModel,DepthAnythingForDepthEstimation:()=>n.DepthAnythingForDepthEstimation,DepthAnythingPreTrainedModel:()=>n.DepthAnythingPreTrainedModel,DepthEstimationPipeline:()=>t.DepthEstimationPipeline,DetrFeatureExtractor:()=>i.DetrFeatureExtractor,DetrForObjectDetection:()=>n.DetrForObjectDetection,DetrForSegmentation:()=>n.DetrForSegmentation,DetrModel:()=>n.DetrModel,DetrObjectDetectionOutput:()=>n.DetrObjectDetectionOutput,DetrPreTrainedModel:()=>n.DetrPreTrainedModel,DetrSegmentationOutput:()=>n.DetrSegmentationOutput,Dinov2ForImageClassification:()=>n.Dinov2ForImageClassification,Dinov2Model:()=>n.Dinov2Model,Dinov2PreTrainedModel:()=>n.Dinov2PreTrainedModel,DistilBertForMaskedLM:()=>n.DistilBertForMaskedLM,DistilBertForQuestionAnswering:()=>n.DistilBertForQuestionAnswering,DistilBertForSequenceClassification:()=>n.DistilBertForSequenceClassification,DistilBertForTokenClassification:()=>n.DistilBertForTokenClassification,DistilBertModel:()=>n.DistilBertModel,DistilBertPreTrainedModel:()=>n.DistilBertPreTrainedModel,DistilBertTokenizer:()=>s.DistilBertTokenizer,DocumentQuestionAnsweringPipeline:()=>t.DocumentQuestionAnsweringPipeline,DonutFeatureExtractor:()=>i.DonutFeatureExtractor,DonutSwinModel:()=>n.DonutSwinModel,DonutSwinPreTrainedModel:()=>n.DonutSwinPreTrainedModel,EfficientNetForImageClassification:()=>n.EfficientNetForImageClassification,EfficientNetImageProcessor:()=>i.EfficientNetImageProcessor,EfficientNetModel:()=>n.EfficientNetModel,EfficientNetPreTrainedModel:()=>n.EfficientNetPreTrainedModel,ElectraForMaskedLM:()=>n.ElectraForMaskedLM,ElectraForQuestionAnswering:()=>n.ElectraForQuestionAnswering,ElectraForSequenceClassification:()=>n.ElectraForSequenceClassification,ElectraForTokenClassification:()=>n.ElectraForTokenClassification,ElectraModel:()=>n.ElectraModel,ElectraPreTrainedModel:()=>n.ElectraPreTrainedModel,ElectraTokenizer:()=>s.ElectraTokenizer,EosTokenCriteria:()=>h.EosTokenCriteria,EsmForMaskedLM:()=>n.EsmForMaskedLM,EsmForSequenceClassification:()=>n.EsmForSequenceClassification,EsmForTokenClassification:()=>n.EsmForTokenClassification,EsmModel:()=>n.EsmModel,EsmPreTrainedModel:()=>n.EsmPreTrainedModel,EsmTokenizer:()=>s.EsmTokenizer,FFT:()=>u.FFT,FalconForCausalLM:()=>n.FalconForCausalLM,FalconModel:()=>n.FalconModel,FalconPreTrainedModel:()=>n.FalconPreTrainedModel,FalconTokenizer:()=>s.FalconTokenizer,FastViTForImageClassification:()=>n.FastViTForImageClassification,FastViTModel:()=>n.FastViTModel,FastViTPreTrainedModel:()=>n.FastViTPreTrainedModel,FeatureExtractionPipeline:()=>t.FeatureExtractionPipeline,FeatureExtractor:()=>i.FeatureExtractor,FillMaskPipeline:()=>t.FillMaskPipeline,Florence2ForConditionalGeneration:()=>n.Florence2ForConditionalGeneration,Florence2PreTrainedModel:()=>n.Florence2PreTrainedModel,Florence2Processor:()=>i.Florence2Processor,GLPNFeatureExtractor:()=>i.GLPNFeatureExtractor,GLPNForDepthEstimation:()=>n.GLPNForDepthEstimation,GLPNModel:()=>n.GLPNModel,GLPNPreTrainedModel:()=>n.GLPNPreTrainedModel,GPT2LMHeadModel:()=>n.GPT2LMHeadModel,GPT2Model:()=>n.GPT2Model,GPT2PreTrainedModel:()=>n.GPT2PreTrainedModel,GPT2Tokenizer:()=>s.GPT2Tokenizer,GPTBigCodeForCausalLM:()=>n.GPTBigCodeForCausalLM,GPTBigCodeModel:()=>n.GPTBigCodeModel,GPTBigCodePreTrainedModel:()=>n.GPTBigCodePreTrainedModel,GPTJForCausalLM:()=>n.GPTJForCausalLM,GPTJModel:()=>n.GPTJModel,GPTJPreTrainedModel:()=>n.GPTJPreTrainedModel,GPTNeoForCausalLM:()=>n.GPTNeoForCausalLM,GPTNeoModel:()=>n.GPTNeoModel,GPTNeoPreTrainedModel:()=>n.GPTNeoPreTrainedModel,GPTNeoXForCausalLM:()=>n.GPTNeoXForCausalLM,GPTNeoXModel:()=>n.GPTNeoXModel,GPTNeoXPreTrainedModel:()=>n.GPTNeoXPreTrainedModel,GPTNeoXTokenizer:()=>s.GPTNeoXTokenizer,Gemma2ForCausalLM:()=>n.Gemma2ForCausalLM,Gemma2Model:()=>n.Gemma2Model,Gemma2PreTrainedModel:()=>n.Gemma2PreTrainedModel,GemmaForCausalLM:()=>n.GemmaForCausalLM,GemmaModel:()=>n.GemmaModel,GemmaPreTrainedModel:()=>n.GemmaPreTrainedModel,GemmaTokenizer:()=>s.GemmaTokenizer,Grok1Tokenizer:()=>s.Grok1Tokenizer,HerbertTokenizer:()=>s.HerbertTokenizer,HubertForCTC:()=>n.HubertForCTC,HubertForSequenceClassification:()=>n.HubertForSequenceClassification,HubertModel:()=>n.HubertModel,HubertPreTrainedModel:()=>n.HubertPreTrainedModel,ImageClassificationPipeline:()=>t.ImageClassificationPipeline,ImageFeatureExtractionPipeline:()=>t.ImageFeatureExtractionPipeline,ImageFeatureExtractor:()=>i.ImageFeatureExtractor,ImageMattingOutput:()=>n.ImageMattingOutput,ImageSegmentationPipeline:()=>t.ImageSegmentationPipeline,ImageToImagePipeline:()=>t.ImageToImagePipeline,ImageToTextPipeline:()=>t.ImageToTextPipeline,InterruptableStoppingCriteria:()=>h.InterruptableStoppingCriteria,JAISLMHeadModel:()=>n.JAISLMHeadModel,JAISModel:()=>n.JAISModel,JAISPreTrainedModel:()=>n.JAISPreTrainedModel,LlamaForCausalLM:()=>n.LlamaForCausalLM,LlamaModel:()=>n.LlamaModel,LlamaPreTrainedModel:()=>n.LlamaPreTrainedModel,LlamaTokenizer:()=>s.LlamaTokenizer,LlavaForConditionalGeneration:()=>n.LlavaForConditionalGeneration,LlavaPreTrainedModel:()=>n.LlavaPreTrainedModel,LongT5ForConditionalGeneration:()=>n.LongT5ForConditionalGeneration,LongT5Model:()=>n.LongT5Model,LongT5PreTrainedModel:()=>n.LongT5PreTrainedModel,M2M100ForConditionalGeneration:()=>n.M2M100ForConditionalGeneration,M2M100Model:()=>n.M2M100Model,M2M100PreTrainedModel:()=>n.M2M100PreTrainedModel,M2M100Tokenizer:()=>s.M2M100Tokenizer,MBart50Tokenizer:()=>s.MBart50Tokenizer,MBartForCausalLM:()=>n.MBartForCausalLM,MBartForConditionalGeneration:()=>n.MBartForConditionalGeneration,MBartForSequenceClassification:()=>n.MBartForSequenceClassification,MBartModel:()=>n.MBartModel,MBartPreTrainedModel:()=>n.MBartPreTrainedModel,MBartTokenizer:()=>s.MBartTokenizer,MPNetForMaskedLM:()=>n.MPNetForMaskedLM,MPNetForQuestionAnswering:()=>n.MPNetForQuestionAnswering,MPNetForSequenceClassification:()=>n.MPNetForSequenceClassification,MPNetForTokenClassification:()=>n.MPNetForTokenClassification,MPNetModel:()=>n.MPNetModel,MPNetPreTrainedModel:()=>n.MPNetPreTrainedModel,MPNetTokenizer:()=>s.MPNetTokenizer,MT5ForConditionalGeneration:()=>n.MT5ForConditionalGeneration,MT5Model:()=>n.MT5Model,MT5PreTrainedModel:()=>n.MT5PreTrainedModel,MarianMTModel:()=>n.MarianMTModel,MarianModel:()=>n.MarianModel,MarianPreTrainedModel:()=>n.MarianPreTrainedModel,MarianTokenizer:()=>s.MarianTokenizer,MaskedLMOutput:()=>n.MaskedLMOutput,MaxLengthCriteria:()=>h.MaxLengthCriteria,MistralForCausalLM:()=>n.MistralForCausalLM,MistralModel:()=>n.MistralModel,MistralPreTrainedModel:()=>n.MistralPreTrainedModel,MobileBertForMaskedLM:()=>n.MobileBertForMaskedLM,MobileBertForQuestionAnswering:()=>n.MobileBertForQuestionAnswering,MobileBertForSequenceClassification:()=>n.MobileBertForSequenceClassification,MobileBertModel:()=>n.MobileBertModel,MobileBertPreTrainedModel:()=>n.MobileBertPreTrainedModel,MobileBertTokenizer:()=>s.MobileBertTokenizer,MobileNetV1FeatureExtractor:()=>i.MobileNetV1FeatureExtractor,MobileNetV1ForImageClassification:()=>n.MobileNetV1ForImageClassification,MobileNetV1Model:()=>n.MobileNetV1Model,MobileNetV1PreTrainedModel:()=>n.MobileNetV1PreTrainedModel,MobileNetV2FeatureExtractor:()=>i.MobileNetV2FeatureExtractor,MobileNetV2ForImageClassification:()=>n.MobileNetV2ForImageClassification,MobileNetV2Model:()=>n.MobileNetV2Model,MobileNetV2PreTrainedModel:()=>n.MobileNetV2PreTrainedModel,MobileNetV3FeatureExtractor:()=>i.MobileNetV3FeatureExtractor,MobileNetV3ForImageClassification:()=>n.MobileNetV3ForImageClassification,MobileNetV3Model:()=>n.MobileNetV3Model,MobileNetV3PreTrainedModel:()=>n.MobileNetV3PreTrainedModel,MobileNetV4FeatureExtractor:()=>i.MobileNetV4FeatureExtractor,MobileNetV4ForImageClassification:()=>n.MobileNetV4ForImageClassification,MobileNetV4Model:()=>n.MobileNetV4Model,MobileNetV4PreTrainedModel:()=>n.MobileNetV4PreTrainedModel,MobileViTFeatureExtractor:()=>i.MobileViTFeatureExtractor,MobileViTForImageClassification:()=>n.MobileViTForImageClassification,MobileViTImageProcessor:()=>i.MobileViTImageProcessor,MobileViTModel:()=>n.MobileViTModel,MobileViTPreTrainedModel:()=>n.MobileViTPreTrainedModel,MobileViTV2ForImageClassification:()=>n.MobileViTV2ForImageClassification,MobileViTV2Model:()=>n.MobileViTV2Model,MobileViTV2PreTrainedModel:()=>n.MobileViTV2PreTrainedModel,ModelOutput:()=>n.ModelOutput,Moondream1ForConditionalGeneration:()=>n.Moondream1ForConditionalGeneration,MptForCausalLM:()=>n.MptForCausalLM,MptModel:()=>n.MptModel,MptPreTrainedModel:()=>n.MptPreTrainedModel,MusicgenForCausalLM:()=>n.MusicgenForCausalLM,MusicgenForConditionalGeneration:()=>n.MusicgenForConditionalGeneration,MusicgenModel:()=>n.MusicgenModel,MusicgenPreTrainedModel:()=>n.MusicgenPreTrainedModel,NllbTokenizer:()=>s.NllbTokenizer,NomicBertModel:()=>n.NomicBertModel,NomicBertPreTrainedModel:()=>n.NomicBertPreTrainedModel,NougatImageProcessor:()=>i.NougatImageProcessor,NougatTokenizer:()=>s.NougatTokenizer,OPTForCausalLM:()=>n.OPTForCausalLM,OPTModel:()=>n.OPTModel,OPTPreTrainedModel:()=>n.OPTPreTrainedModel,ObjectDetectionPipeline:()=>t.ObjectDetectionPipeline,OpenELMForCausalLM:()=>n.OpenELMForCausalLM,OpenELMModel:()=>n.OpenELMModel,OpenELMPreTrainedModel:()=>n.OpenELMPreTrainedModel,OwlViTFeatureExtractor:()=>i.OwlViTFeatureExtractor,OwlViTForObjectDetection:()=>n.OwlViTForObjectDetection,OwlViTModel:()=>n.OwlViTModel,OwlViTPreTrainedModel:()=>n.OwlViTPreTrainedModel,OwlViTProcessor:()=>i.OwlViTProcessor,Owlv2ForObjectDetection:()=>n.Owlv2ForObjectDetection,Owlv2ImageProcessor:()=>i.Owlv2ImageProcessor,Owlv2Model:()=>n.Owlv2Model,Owlv2PreTrainedModel:()=>n.Owlv2PreTrainedModel,Phi3ForCausalLM:()=>n.Phi3ForCausalLM,Phi3Model:()=>n.Phi3Model,Phi3PreTrainedModel:()=>n.Phi3PreTrainedModel,PhiForCausalLM:()=>n.PhiForCausalLM,PhiModel:()=>n.PhiModel,PhiPreTrainedModel:()=>n.PhiPreTrainedModel,Pipeline:()=>t.Pipeline,PreTrainedModel:()=>n.PreTrainedModel,PreTrainedTokenizer:()=>s.PreTrainedTokenizer,PretrainedConfig:()=>a.PretrainedConfig,PretrainedMixin:()=>n.PretrainedMixin,Processor:()=>i.Processor,PyAnnoteFeatureExtractor:()=>i.PyAnnoteFeatureExtractor,PyAnnoteForAudioFrameClassification:()=>n.PyAnnoteForAudioFrameClassification,PyAnnoteModel:()=>n.PyAnnoteModel,PyAnnotePreTrainedModel:()=>n.PyAnnotePreTrainedModel,PyAnnoteProcessor:()=>i.PyAnnoteProcessor,QuestionAnsweringModelOutput:()=>n.QuestionAnsweringModelOutput,QuestionAnsweringPipeline:()=>t.QuestionAnsweringPipeline,Qwen2ForCausalLM:()=>n.Qwen2ForCausalLM,Qwen2Model:()=>n.Qwen2Model,Qwen2PreTrainedModel:()=>n.Qwen2PreTrainedModel,Qwen2Tokenizer:()=>s.Qwen2Tokenizer,RTDetrForObjectDetection:()=>n.RTDetrForObjectDetection,RTDetrImageProcessor:()=>i.RTDetrImageProcessor,RTDetrModel:()=>n.RTDetrModel,RTDetrObjectDetectionOutput:()=>n.RTDetrObjectDetectionOutput,RTDetrPreTrainedModel:()=>n.RTDetrPreTrainedModel,RawImage:()=>c.RawImage,ResNetForImageClassification:()=>n.ResNetForImageClassification,ResNetModel:()=>n.ResNetModel,ResNetPreTrainedModel:()=>n.ResNetPreTrainedModel,RoFormerForMaskedLM:()=>n.RoFormerForMaskedLM,RoFormerForQuestionAnswering:()=>n.RoFormerForQuestionAnswering,RoFormerForSequenceClassification:()=>n.RoFormerForSequenceClassification,RoFormerForTokenClassification:()=>n.RoFormerForTokenClassification,RoFormerModel:()=>n.RoFormerModel,RoFormerPreTrainedModel:()=>n.RoFormerPreTrainedModel,RoFormerTokenizer:()=>s.RoFormerTokenizer,RobertaForMaskedLM:()=>n.RobertaForMaskedLM,RobertaForQuestionAnswering:()=>n.RobertaForQuestionAnswering,RobertaForSequenceClassification:()=>n.RobertaForSequenceClassification,RobertaForTokenClassification:()=>n.RobertaForTokenClassification,RobertaModel:()=>n.RobertaModel,RobertaPreTrainedModel:()=>n.RobertaPreTrainedModel,RobertaTokenizer:()=>s.RobertaTokenizer,SamImageProcessor:()=>i.SamImageProcessor,SamImageSegmentationOutput:()=>n.SamImageSegmentationOutput,SamModel:()=>n.SamModel,SamPreTrainedModel:()=>n.SamPreTrainedModel,SamProcessor:()=>i.SamProcessor,SapiensFeatureExtractor:()=>i.SapiensFeatureExtractor,SapiensForDepthEstimation:()=>n.SapiensForDepthEstimation,SapiensForNormalEstimation:()=>n.SapiensForNormalEstimation,SapiensForSemanticSegmentation:()=>n.SapiensForSemanticSegmentation,SapiensPreTrainedModel:()=>n.SapiensPreTrainedModel,SeamlessM4TFeatureExtractor:()=>i.SeamlessM4TFeatureExtractor,SegformerFeatureExtractor:()=>i.SegformerFeatureExtractor,SegformerForImageClassification:()=>n.SegformerForImageClassification,SegformerForSemanticSegmentation:()=>n.SegformerForSemanticSegmentation,SegformerModel:()=>n.SegformerModel,SegformerPreTrainedModel:()=>n.SegformerPreTrainedModel,Seq2SeqLMOutput:()=>n.Seq2SeqLMOutput,SequenceClassifierOutput:()=>n.SequenceClassifierOutput,SiglipImageProcessor:()=>i.SiglipImageProcessor,SiglipModel:()=>n.SiglipModel,SiglipPreTrainedModel:()=>n.SiglipPreTrainedModel,SiglipTextModel:()=>n.SiglipTextModel,SiglipTokenizer:()=>s.SiglipTokenizer,SiglipVisionModel:()=>n.SiglipVisionModel,SpeechT5FeatureExtractor:()=>i.SpeechT5FeatureExtractor,SpeechT5ForSpeechToText:()=>n.SpeechT5ForSpeechToText,SpeechT5ForTextToSpeech:()=>n.SpeechT5ForTextToSpeech,SpeechT5HifiGan:()=>n.SpeechT5HifiGan,SpeechT5Model:()=>n.SpeechT5Model,SpeechT5PreTrainedModel:()=>n.SpeechT5PreTrainedModel,SpeechT5Processor:()=>i.SpeechT5Processor,SpeechT5Tokenizer:()=>s.SpeechT5Tokenizer,SqueezeBertForMaskedLM:()=>n.SqueezeBertForMaskedLM,SqueezeBertForQuestionAnswering:()=>n.SqueezeBertForQuestionAnswering,SqueezeBertForSequenceClassification:()=>n.SqueezeBertForSequenceClassification,SqueezeBertModel:()=>n.SqueezeBertModel,SqueezeBertPreTrainedModel:()=>n.SqueezeBertPreTrainedModel,SqueezeBertTokenizer:()=>s.SqueezeBertTokenizer,StableLmForCausalLM:()=>n.StableLmForCausalLM,StableLmModel:()=>n.StableLmModel,StableLmPreTrainedModel:()=>n.StableLmPreTrainedModel,Starcoder2ForCausalLM:()=>n.Starcoder2ForCausalLM,Starcoder2Model:()=>n.Starcoder2Model,Starcoder2PreTrainedModel:()=>n.Starcoder2PreTrainedModel,StoppingCriteria:()=>h.StoppingCriteria,StoppingCriteriaList:()=>h.StoppingCriteriaList,SummarizationPipeline:()=>t.SummarizationPipeline,Swin2SRForImageSuperResolution:()=>n.Swin2SRForImageSuperResolution,Swin2SRImageProcessor:()=>i.Swin2SRImageProcessor,Swin2SRModel:()=>n.Swin2SRModel,Swin2SRPreTrainedModel:()=>n.Swin2SRPreTrainedModel,SwinForImageClassification:()=>n.SwinForImageClassification,SwinModel:()=>n.SwinModel,SwinPreTrainedModel:()=>n.SwinPreTrainedModel,T5ForConditionalGeneration:()=>n.T5ForConditionalGeneration,T5Model:()=>n.T5Model,T5PreTrainedModel:()=>n.T5PreTrainedModel,T5Tokenizer:()=>s.T5Tokenizer,TableTransformerForObjectDetection:()=>n.TableTransformerForObjectDetection,TableTransformerModel:()=>n.TableTransformerModel,TableTransformerObjectDetectionOutput:()=>n.TableTransformerObjectDetectionOutput,TableTransformerPreTrainedModel:()=>n.TableTransformerPreTrainedModel,Tensor:()=>d.Tensor,Text2TextGenerationPipeline:()=>t.Text2TextGenerationPipeline,TextClassificationPipeline:()=>t.TextClassificationPipeline,TextGenerationPipeline:()=>t.TextGenerationPipeline,TextStreamer:()=>p.TextStreamer,TextToAudioPipeline:()=>t.TextToAudioPipeline,TokenClassificationPipeline:()=>t.TokenClassificationPipeline,TokenClassifierOutput:()=>n.TokenClassifierOutput,TokenizerModel:()=>s.TokenizerModel,TrOCRForCausalLM:()=>n.TrOCRForCausalLM,TrOCRPreTrainedModel:()=>n.TrOCRPreTrainedModel,TranslationPipeline:()=>t.TranslationPipeline,UniSpeechForCTC:()=>n.UniSpeechForCTC,UniSpeechForSequenceClassification:()=>n.UniSpeechForSequenceClassification,UniSpeechModel:()=>n.UniSpeechModel,UniSpeechPreTrainedModel:()=>n.UniSpeechPreTrainedModel,UniSpeechSatForAudioFrameClassification:()=>n.UniSpeechSatForAudioFrameClassification,UniSpeechSatForCTC:()=>n.UniSpeechSatForCTC,UniSpeechSatForSequenceClassification:()=>n.UniSpeechSatForSequenceClassification,UniSpeechSatModel:()=>n.UniSpeechSatModel,UniSpeechSatPreTrainedModel:()=>n.UniSpeechSatPreTrainedModel,ViTFeatureExtractor:()=>i.ViTFeatureExtractor,ViTForImageClassification:()=>n.ViTForImageClassification,ViTImageProcessor:()=>i.ViTImageProcessor,ViTModel:()=>n.ViTModel,ViTPreTrainedModel:()=>n.ViTPreTrainedModel,VisionEncoderDecoderModel:()=>n.VisionEncoderDecoderModel,VitMatteForImageMatting:()=>n.VitMatteForImageMatting,VitMatteImageProcessor:()=>i.VitMatteImageProcessor,VitMattePreTrainedModel:()=>n.VitMattePreTrainedModel,VitsModel:()=>n.VitsModel,VitsModelOutput:()=>n.VitsModelOutput,VitsPreTrainedModel:()=>n.VitsPreTrainedModel,VitsTokenizer:()=>s.VitsTokenizer,Wav2Vec2BertForCTC:()=>n.Wav2Vec2BertForCTC,Wav2Vec2BertForSequenceClassification:()=>n.Wav2Vec2BertForSequenceClassification,Wav2Vec2BertModel:()=>n.Wav2Vec2BertModel,Wav2Vec2BertPreTrainedModel:()=>n.Wav2Vec2BertPreTrainedModel,Wav2Vec2CTCTokenizer:()=>s.Wav2Vec2CTCTokenizer,Wav2Vec2FeatureExtractor:()=>i.Wav2Vec2FeatureExtractor,Wav2Vec2ForAudioFrameClassification:()=>n.Wav2Vec2ForAudioFrameClassification,Wav2Vec2ForCTC:()=>n.Wav2Vec2ForCTC,Wav2Vec2ForSequenceClassification:()=>n.Wav2Vec2ForSequenceClassification,Wav2Vec2Model:()=>n.Wav2Vec2Model,Wav2Vec2PreTrainedModel:()=>n.Wav2Vec2PreTrainedModel,Wav2Vec2ProcessorWithLM:()=>i.Wav2Vec2ProcessorWithLM,WavLMForAudioFrameClassification:()=>n.WavLMForAudioFrameClassification,WavLMForCTC:()=>n.WavLMForCTC,WavLMForSequenceClassification:()=>n.WavLMForSequenceClassification,WavLMForXVector:()=>n.WavLMForXVector,WavLMModel:()=>n.WavLMModel,WavLMPreTrainedModel:()=>n.WavLMPreTrainedModel,WeSpeakerFeatureExtractor:()=>i.WeSpeakerFeatureExtractor,WeSpeakerResNetModel:()=>n.WeSpeakerResNetModel,WeSpeakerResNetPreTrainedModel:()=>n.WeSpeakerResNetPreTrainedModel,WhisperFeatureExtractor:()=>i.WhisperFeatureExtractor,WhisperForConditionalGeneration:()=>n.WhisperForConditionalGeneration,WhisperModel:()=>n.WhisperModel,WhisperPreTrainedModel:()=>n.WhisperPreTrainedModel,WhisperProcessor:()=>i.WhisperProcessor,WhisperTextStreamer:()=>p.WhisperTextStreamer,WhisperTokenizer:()=>s.WhisperTokenizer,XLMForQuestionAnswering:()=>n.XLMForQuestionAnswering,XLMForSequenceClassification:()=>n.XLMForSequenceClassification,XLMForTokenClassification:()=>n.XLMForTokenClassification,XLMModel:()=>n.XLMModel,XLMPreTrainedModel:()=>n.XLMPreTrainedModel,XLMRobertaForMaskedLM:()=>n.XLMRobertaForMaskedLM,XLMRobertaForQuestionAnswering:()=>n.XLMRobertaForQuestionAnswering,XLMRobertaForSequenceClassification:()=>n.XLMRobertaForSequenceClassification,XLMRobertaForTokenClassification:()=>n.XLMRobertaForTokenClassification,XLMRobertaModel:()=>n.XLMRobertaModel,XLMRobertaPreTrainedModel:()=>n.XLMRobertaPreTrainedModel,XLMRobertaTokenizer:()=>s.XLMRobertaTokenizer,XLMTokenizer:()=>s.XLMTokenizer,XLMWithLMHeadModel:()=>n.XLMWithLMHeadModel,XVectorOutput:()=>n.XVectorOutput,YolosFeatureExtractor:()=>i.YolosFeatureExtractor,YolosForObjectDetection:()=>n.YolosForObjectDetection,YolosModel:()=>n.YolosModel,YolosObjectDetectionOutput:()=>n.YolosObjectDetectionOutput,YolosPreTrainedModel:()=>n.YolosPreTrainedModel,ZeroShotAudioClassificationPipeline:()=>t.ZeroShotAudioClassificationPipeline,ZeroShotClassificationPipeline:()=>t.ZeroShotClassificationPipeline,ZeroShotImageClassificationPipeline:()=>t.ZeroShotImageClassificationPipeline,ZeroShotObjectDetectionPipeline:()=>t.ZeroShotObjectDetectionPipeline,bankers_round:()=>u.bankers_round,cat:()=>d.cat,cos_sim:()=>u.cos_sim,dot:()=>u.dot,dynamic_time_warping:()=>u.dynamic_time_warping,env:()=>e.env,full:()=>d.full,full_like:()=>d.full_like,getKeyValueShapes:()=>a.getKeyValueShapes,hamming:()=>l.hamming,hanning:()=>l.hanning,interpolate:()=>d.interpolate,interpolate_4d:()=>d.interpolate_4d,interpolate_data:()=>u.interpolate_data,is_chinese_char:()=>s.is_chinese_char,layer_norm:()=>d.layer_norm,log_softmax:()=>u.log_softmax,magnitude:()=>u.magnitude,matmul:()=>d.matmul,max:()=>u.max,mean:()=>d.mean,mean_pooling:()=>d.mean_pooling,medianFilter:()=>u.medianFilter,mel_filter_bank:()=>l.mel_filter_bank,min:()=>u.min,ones:()=>d.ones,ones_like:()=>d.ones_like,permute:()=>d.permute,permute_data:()=>u.permute_data,pipeline:()=>t.pipeline,quantize_embeddings:()=>d.quantize_embeddings,read_audio:()=>l.read_audio,rfft:()=>d.rfft,round:()=>u.round,softmax:()=>u.softmax,spectrogram:()=>l.spectrogram,stack:()=>d.stack,std_mean:()=>d.std_mean,topk:()=>d.topk,window_function:()=>l.window_function,zeros:()=>d.zeros,zeros_like:()=>d.zeros_like});var e=r(/*! ./env.js */"./src/env.js"),t=r(/*! ./pipelines.js */"./src/pipelines.js"),n=r(/*! ./models.js */"./src/models.js"),s=r(/*! ./tokenizers.js */"./src/tokenizers.js"),i=r(/*! ./processors.js */"./src/processors.js"),a=r(/*! ./configs.js */"./src/configs.js"),l=r(/*! ./utils/audio.js */"./src/utils/audio.js"),c=r(/*! ./utils/image.js */"./src/utils/image.js"),d=r(/*! ./utils/tensor.js */"./src/utils/tensor.js"),u=r(/*! ./utils/maths.js */"./src/utils/maths.js"),p=r(/*! ./generation/streamers.js */"./src/generation/streamers.js"),h=r(/*! ./generation/stopping_criteria.js */"./src/generation/stopping_criteria.js")})();var i=exports;for(var a in o)i[a]=o[a];o.__esModule&&Object.defineProperty(i,"__esModule",{value:!0})})();
+  \*****************************/r.r(o),r.d(o,{ASTFeatureExtractor:()=>i.ASTFeatureExtractor,ASTForAudioClassification:()=>n.ASTForAudioClassification,ASTModel:()=>n.ASTModel,ASTPreTrainedModel:()=>n.ASTPreTrainedModel,AlbertForMaskedLM:()=>n.AlbertForMaskedLM,AlbertForQuestionAnswering:()=>n.AlbertForQuestionAnswering,AlbertForSequenceClassification:()=>n.AlbertForSequenceClassification,AlbertModel:()=>n.AlbertModel,AlbertPreTrainedModel:()=>n.AlbertPreTrainedModel,AlbertTokenizer:()=>s.AlbertTokenizer,AudioClassificationPipeline:()=>t.AudioClassificationPipeline,AutoConfig:()=>a.AutoConfig,AutoModel:()=>n.AutoModel,AutoModelForAudioClassification:()=>n.AutoModelForAudioClassification,AutoModelForAudioFrameClassification:()=>n.AutoModelForAudioFrameClassification,AutoModelForCTC:()=>n.AutoModelForCTC,AutoModelForCausalLM:()=>n.AutoModelForCausalLM,AutoModelForDepthEstimation:()=>n.AutoModelForDepthEstimation,AutoModelForDocumentQuestionAnswering:()=>n.AutoModelForDocumentQuestionAnswering,AutoModelForImageClassification:()=>n.AutoModelForImageClassification,AutoModelForImageFeatureExtraction:()=>n.AutoModelForImageFeatureExtraction,AutoModelForImageMatting:()=>n.AutoModelForImageMatting,AutoModelForImageSegmentation:()=>n.AutoModelForImageSegmentation,AutoModelForImageToImage:()=>n.AutoModelForImageToImage,AutoModelForMaskGeneration:()=>n.AutoModelForMaskGeneration,AutoModelForMaskedLM:()=>n.AutoModelForMaskedLM,AutoModelForNormalEstimation:()=>n.AutoModelForNormalEstimation,AutoModelForObjectDetection:()=>n.AutoModelForObjectDetection,AutoModelForQuestionAnswering:()=>n.AutoModelForQuestionAnswering,AutoModelForSemanticSegmentation:()=>n.AutoModelForSemanticSegmentation,AutoModelForSeq2SeqLM:()=>n.AutoModelForSeq2SeqLM,AutoModelForSequenceClassification:()=>n.AutoModelForSequenceClassification,AutoModelForSpeechSeq2Seq:()=>n.AutoModelForSpeechSeq2Seq,AutoModelForTextToSpectrogram:()=>n.AutoModelForTextToSpectrogram,AutoModelForTextToWaveform:()=>n.AutoModelForTextToWaveform,AutoModelForTokenClassification:()=>n.AutoModelForTokenClassification,AutoModelForVision2Seq:()=>n.AutoModelForVision2Seq,AutoModelForXVector:()=>n.AutoModelForXVector,AutoModelForZeroShotObjectDetection:()=>n.AutoModelForZeroShotObjectDetection,AutoProcessor:()=>i.AutoProcessor,AutoTokenizer:()=>s.AutoTokenizer,AutomaticSpeechRecognitionPipeline:()=>t.AutomaticSpeechRecognitionPipeline,BartForConditionalGeneration:()=>n.BartForConditionalGeneration,BartForSequenceClassification:()=>n.BartForSequenceClassification,BartModel:()=>n.BartModel,BartPretrainedModel:()=>n.BartPretrainedModel,BartTokenizer:()=>s.BartTokenizer,BaseModelOutput:()=>n.BaseModelOutput,BaseStreamer:()=>p.BaseStreamer,BeitFeatureExtractor:()=>i.BeitFeatureExtractor,BeitForImageClassification:()=>n.BeitForImageClassification,BeitModel:()=>n.BeitModel,BeitPreTrainedModel:()=>n.BeitPreTrainedModel,BertForMaskedLM:()=>n.BertForMaskedLM,BertForQuestionAnswering:()=>n.BertForQuestionAnswering,BertForSequenceClassification:()=>n.BertForSequenceClassification,BertForTokenClassification:()=>n.BertForTokenClassification,BertModel:()=>n.BertModel,BertPreTrainedModel:()=>n.BertPreTrainedModel,BertTokenizer:()=>s.BertTokenizer,BitImageProcessor:()=>i.BitImageProcessor,BlenderbotForConditionalGeneration:()=>n.BlenderbotForConditionalGeneration,BlenderbotModel:()=>n.BlenderbotModel,BlenderbotPreTrainedModel:()=>n.BlenderbotPreTrainedModel,BlenderbotSmallForConditionalGeneration:()=>n.BlenderbotSmallForConditionalGeneration,BlenderbotSmallModel:()=>n.BlenderbotSmallModel,BlenderbotSmallPreTrainedModel:()=>n.BlenderbotSmallPreTrainedModel,BlenderbotSmallTokenizer:()=>s.BlenderbotSmallTokenizer,BlenderbotTokenizer:()=>s.BlenderbotTokenizer,BloomForCausalLM:()=>n.BloomForCausalLM,BloomModel:()=>n.BloomModel,BloomPreTrainedModel:()=>n.BloomPreTrainedModel,BloomTokenizer:()=>s.BloomTokenizer,CLIPFeatureExtractor:()=>i.CLIPFeatureExtractor,CLIPImageProcessor:()=>i.CLIPImageProcessor,CLIPModel:()=>n.CLIPModel,CLIPPreTrainedModel:()=>n.CLIPPreTrainedModel,CLIPSegForImageSegmentation:()=>n.CLIPSegForImageSegmentation,CLIPSegModel:()=>n.CLIPSegModel,CLIPSegPreTrainedModel:()=>n.CLIPSegPreTrainedModel,CLIPTextModelWithProjection:()=>n.CLIPTextModelWithProjection,CLIPTokenizer:()=>s.CLIPTokenizer,CLIPVisionModelWithProjection:()=>n.CLIPVisionModelWithProjection,CamembertForMaskedLM:()=>n.CamembertForMaskedLM,CamembertForQuestionAnswering:()=>n.CamembertForQuestionAnswering,CamembertForSequenceClassification:()=>n.CamembertForSequenceClassification,CamembertForTokenClassification:()=>n.CamembertForTokenClassification,CamembertModel:()=>n.CamembertModel,CamembertPreTrainedModel:()=>n.CamembertPreTrainedModel,CamembertTokenizer:()=>s.CamembertTokenizer,CausalLMOutput:()=>n.CausalLMOutput,CausalLMOutputWithPast:()=>n.CausalLMOutputWithPast,ChineseCLIPFeatureExtractor:()=>i.ChineseCLIPFeatureExtractor,ChineseCLIPModel:()=>n.ChineseCLIPModel,ChineseCLIPPreTrainedModel:()=>n.ChineseCLIPPreTrainedModel,ClapAudioModelWithProjection:()=>n.ClapAudioModelWithProjection,ClapFeatureExtractor:()=>i.ClapFeatureExtractor,ClapModel:()=>n.ClapModel,ClapPreTrainedModel:()=>n.ClapPreTrainedModel,ClapTextModelWithProjection:()=>n.ClapTextModelWithProjection,CodeGenForCausalLM:()=>n.CodeGenForCausalLM,CodeGenModel:()=>n.CodeGenModel,CodeGenPreTrainedModel:()=>n.CodeGenPreTrainedModel,CodeGenTokenizer:()=>s.CodeGenTokenizer,CodeLlamaTokenizer:()=>s.CodeLlamaTokenizer,CohereForCausalLM:()=>n.CohereForCausalLM,CohereModel:()=>n.CohereModel,CoherePreTrainedModel:()=>n.CoherePreTrainedModel,CohereTokenizer:()=>s.CohereTokenizer,ConvBertForMaskedLM:()=>n.ConvBertForMaskedLM,ConvBertForQuestionAnswering:()=>n.ConvBertForQuestionAnswering,ConvBertForSequenceClassification:()=>n.ConvBertForSequenceClassification,ConvBertForTokenClassification:()=>n.ConvBertForTokenClassification,ConvBertModel:()=>n.ConvBertModel,ConvBertPreTrainedModel:()=>n.ConvBertPreTrainedModel,ConvBertTokenizer:()=>s.ConvBertTokenizer,ConvNextFeatureExtractor:()=>i.ConvNextFeatureExtractor,ConvNextForImageClassification:()=>n.ConvNextForImageClassification,ConvNextImageProcessor:()=>i.ConvNextImageProcessor,ConvNextModel:()=>n.ConvNextModel,ConvNextPreTrainedModel:()=>n.ConvNextPreTrainedModel,ConvNextV2ForImageClassification:()=>n.ConvNextV2ForImageClassification,ConvNextV2Model:()=>n.ConvNextV2Model,ConvNextV2PreTrainedModel:()=>n.ConvNextV2PreTrainedModel,DPTFeatureExtractor:()=>i.DPTFeatureExtractor,DPTForDepthEstimation:()=>n.DPTForDepthEstimation,DPTImageProcessor:()=>i.DPTImageProcessor,DPTModel:()=>n.DPTModel,DPTPreTrainedModel:()=>n.DPTPreTrainedModel,DebertaForMaskedLM:()=>n.DebertaForMaskedLM,DebertaForQuestionAnswering:()=>n.DebertaForQuestionAnswering,DebertaForSequenceClassification:()=>n.DebertaForSequenceClassification,DebertaForTokenClassification:()=>n.DebertaForTokenClassification,DebertaModel:()=>n.DebertaModel,DebertaPreTrainedModel:()=>n.DebertaPreTrainedModel,DebertaTokenizer:()=>s.DebertaTokenizer,DebertaV2ForMaskedLM:()=>n.DebertaV2ForMaskedLM,DebertaV2ForQuestionAnswering:()=>n.DebertaV2ForQuestionAnswering,DebertaV2ForSequenceClassification:()=>n.DebertaV2ForSequenceClassification,DebertaV2ForTokenClassification:()=>n.DebertaV2ForTokenClassification,DebertaV2Model:()=>n.DebertaV2Model,DebertaV2PreTrainedModel:()=>n.DebertaV2PreTrainedModel,DebertaV2Tokenizer:()=>s.DebertaV2Tokenizer,DeiTFeatureExtractor:()=>i.DeiTFeatureExtractor,DeiTForImageClassification:()=>n.DeiTForImageClassification,DeiTModel:()=>n.DeiTModel,DeiTPreTrainedModel:()=>n.DeiTPreTrainedModel,DepthAnythingForDepthEstimation:()=>n.DepthAnythingForDepthEstimation,DepthAnythingPreTrainedModel:()=>n.DepthAnythingPreTrainedModel,DepthEstimationPipeline:()=>t.DepthEstimationPipeline,DetrFeatureExtractor:()=>i.DetrFeatureExtractor,DetrForObjectDetection:()=>n.DetrForObjectDetection,DetrForSegmentation:()=>n.DetrForSegmentation,DetrModel:()=>n.DetrModel,DetrObjectDetectionOutput:()=>n.DetrObjectDetectionOutput,DetrPreTrainedModel:()=>n.DetrPreTrainedModel,DetrSegmentationOutput:()=>n.DetrSegmentationOutput,Dinov2ForImageClassification:()=>n.Dinov2ForImageClassification,Dinov2Model:()=>n.Dinov2Model,Dinov2PreTrainedModel:()=>n.Dinov2PreTrainedModel,DistilBertForMaskedLM:()=>n.DistilBertForMaskedLM,DistilBertForQuestionAnswering:()=>n.DistilBertForQuestionAnswering,DistilBertForSequenceClassification:()=>n.DistilBertForSequenceClassification,DistilBertForTokenClassification:()=>n.DistilBertForTokenClassification,DistilBertModel:()=>n.DistilBertModel,DistilBertPreTrainedModel:()=>n.DistilBertPreTrainedModel,DistilBertTokenizer:()=>s.DistilBertTokenizer,DocumentQuestionAnsweringPipeline:()=>t.DocumentQuestionAnsweringPipeline,DonutFeatureExtractor:()=>i.DonutFeatureExtractor,DonutSwinModel:()=>n.DonutSwinModel,DonutSwinPreTrainedModel:()=>n.DonutSwinPreTrainedModel,EfficientNetForImageClassification:()=>n.EfficientNetForImageClassification,EfficientNetImageProcessor:()=>i.EfficientNetImageProcessor,EfficientNetModel:()=>n.EfficientNetModel,EfficientNetPreTrainedModel:()=>n.EfficientNetPreTrainedModel,ElectraForMaskedLM:()=>n.ElectraForMaskedLM,ElectraForQuestionAnswering:()=>n.ElectraForQuestionAnswering,ElectraForSequenceClassification:()=>n.ElectraForSequenceClassification,ElectraForTokenClassification:()=>n.ElectraForTokenClassification,ElectraModel:()=>n.ElectraModel,ElectraPreTrainedModel:()=>n.ElectraPreTrainedModel,ElectraTokenizer:()=>s.ElectraTokenizer,EosTokenCriteria:()=>h.EosTokenCriteria,EsmForMaskedLM:()=>n.EsmForMaskedLM,EsmForSequenceClassification:()=>n.EsmForSequenceClassification,EsmForTokenClassification:()=>n.EsmForTokenClassification,EsmModel:()=>n.EsmModel,EsmPreTrainedModel:()=>n.EsmPreTrainedModel,EsmTokenizer:()=>s.EsmTokenizer,FFT:()=>u.FFT,FalconForCausalLM:()=>n.FalconForCausalLM,FalconModel:()=>n.FalconModel,FalconPreTrainedModel:()=>n.FalconPreTrainedModel,FalconTokenizer:()=>s.FalconTokenizer,FastViTForImageClassification:()=>n.FastViTForImageClassification,FastViTModel:()=>n.FastViTModel,FastViTPreTrainedModel:()=>n.FastViTPreTrainedModel,FeatureExtractionPipeline:()=>t.FeatureExtractionPipeline,FeatureExtractor:()=>i.FeatureExtractor,FillMaskPipeline:()=>t.FillMaskPipeline,Florence2ForConditionalGeneration:()=>n.Florence2ForConditionalGeneration,Florence2PreTrainedModel:()=>n.Florence2PreTrainedModel,Florence2Processor:()=>i.Florence2Processor,GLPNFeatureExtractor:()=>i.GLPNFeatureExtractor,GLPNForDepthEstimation:()=>n.GLPNForDepthEstimation,GLPNModel:()=>n.GLPNModel,GLPNPreTrainedModel:()=>n.GLPNPreTrainedModel,GPT2LMHeadModel:()=>n.GPT2LMHeadModel,GPT2Model:()=>n.GPT2Model,GPT2PreTrainedModel:()=>n.GPT2PreTrainedModel,GPT2Tokenizer:()=>s.GPT2Tokenizer,GPTBigCodeForCausalLM:()=>n.GPTBigCodeForCausalLM,GPTBigCodeModel:()=>n.GPTBigCodeModel,GPTBigCodePreTrainedModel:()=>n.GPTBigCodePreTrainedModel,GPTJForCausalLM:()=>n.GPTJForCausalLM,GPTJModel:()=>n.GPTJModel,GPTJPreTrainedModel:()=>n.GPTJPreTrainedModel,GPTNeoForCausalLM:()=>n.GPTNeoForCausalLM,GPTNeoModel:()=>n.GPTNeoModel,GPTNeoPreTrainedModel:()=>n.GPTNeoPreTrainedModel,GPTNeoXForCausalLM:()=>n.GPTNeoXForCausalLM,GPTNeoXModel:()=>n.GPTNeoXModel,GPTNeoXPreTrainedModel:()=>n.GPTNeoXPreTrainedModel,GPTNeoXTokenizer:()=>s.GPTNeoXTokenizer,Gemma2ForCausalLM:()=>n.Gemma2ForCausalLM,Gemma2Model:()=>n.Gemma2Model,Gemma2PreTrainedModel:()=>n.Gemma2PreTrainedModel,GemmaForCausalLM:()=>n.GemmaForCausalLM,GemmaModel:()=>n.GemmaModel,GemmaPreTrainedModel:()=>n.GemmaPreTrainedModel,GemmaTokenizer:()=>s.GemmaTokenizer,Grok1Tokenizer:()=>s.Grok1Tokenizer,HerbertTokenizer:()=>s.HerbertTokenizer,HieraForImageClassification:()=>n.HieraForImageClassification,HieraModel:()=>n.HieraModel,HieraPreTrainedModel:()=>n.HieraPreTrainedModel,HubertForCTC:()=>n.HubertForCTC,HubertForSequenceClassification:()=>n.HubertForSequenceClassification,HubertModel:()=>n.HubertModel,HubertPreTrainedModel:()=>n.HubertPreTrainedModel,ImageClassificationPipeline:()=>t.ImageClassificationPipeline,ImageFeatureExtractionPipeline:()=>t.ImageFeatureExtractionPipeline,ImageFeatureExtractor:()=>i.ImageFeatureExtractor,ImageMattingOutput:()=>n.ImageMattingOutput,ImageSegmentationPipeline:()=>t.ImageSegmentationPipeline,ImageToImagePipeline:()=>t.ImageToImagePipeline,ImageToTextPipeline:()=>t.ImageToTextPipeline,InterruptableStoppingCriteria:()=>h.InterruptableStoppingCriteria,JAISLMHeadModel:()=>n.JAISLMHeadModel,JAISModel:()=>n.JAISModel,JAISPreTrainedModel:()=>n.JAISPreTrainedModel,LlamaForCausalLM:()=>n.LlamaForCausalLM,LlamaModel:()=>n.LlamaModel,LlamaPreTrainedModel:()=>n.LlamaPreTrainedModel,LlamaTokenizer:()=>s.LlamaTokenizer,LlavaForConditionalGeneration:()=>n.LlavaForConditionalGeneration,LlavaPreTrainedModel:()=>n.LlavaPreTrainedModel,LongT5ForConditionalGeneration:()=>n.LongT5ForConditionalGeneration,LongT5Model:()=>n.LongT5Model,LongT5PreTrainedModel:()=>n.LongT5PreTrainedModel,M2M100ForConditionalGeneration:()=>n.M2M100ForConditionalGeneration,M2M100Model:()=>n.M2M100Model,M2M100PreTrainedModel:()=>n.M2M100PreTrainedModel,M2M100Tokenizer:()=>s.M2M100Tokenizer,MBart50Tokenizer:()=>s.MBart50Tokenizer,MBartForCausalLM:()=>n.MBartForCausalLM,MBartForConditionalGeneration:()=>n.MBartForConditionalGeneration,MBartForSequenceClassification:()=>n.MBartForSequenceClassification,MBartModel:()=>n.MBartModel,MBartPreTrainedModel:()=>n.MBartPreTrainedModel,MBartTokenizer:()=>s.MBartTokenizer,MPNetForMaskedLM:()=>n.MPNetForMaskedLM,MPNetForQuestionAnswering:()=>n.MPNetForQuestionAnswering,MPNetForSequenceClassification:()=>n.MPNetForSequenceClassification,MPNetForTokenClassification:()=>n.MPNetForTokenClassification,MPNetModel:()=>n.MPNetModel,MPNetPreTrainedModel:()=>n.MPNetPreTrainedModel,MPNetTokenizer:()=>s.MPNetTokenizer,MT5ForConditionalGeneration:()=>n.MT5ForConditionalGeneration,MT5Model:()=>n.MT5Model,MT5PreTrainedModel:()=>n.MT5PreTrainedModel,MarianMTModel:()=>n.MarianMTModel,MarianModel:()=>n.MarianModel,MarianPreTrainedModel:()=>n.MarianPreTrainedModel,MarianTokenizer:()=>s.MarianTokenizer,MaskedLMOutput:()=>n.MaskedLMOutput,MaxLengthCriteria:()=>h.MaxLengthCriteria,MistralForCausalLM:()=>n.MistralForCausalLM,MistralModel:()=>n.MistralModel,MistralPreTrainedModel:()=>n.MistralPreTrainedModel,MobileBertForMaskedLM:()=>n.MobileBertForMaskedLM,MobileBertForQuestionAnswering:()=>n.MobileBertForQuestionAnswering,MobileBertForSequenceClassification:()=>n.MobileBertForSequenceClassification,MobileBertModel:()=>n.MobileBertModel,MobileBertPreTrainedModel:()=>n.MobileBertPreTrainedModel,MobileBertTokenizer:()=>s.MobileBertTokenizer,MobileNetV1FeatureExtractor:()=>i.MobileNetV1FeatureExtractor,MobileNetV1ForImageClassification:()=>n.MobileNetV1ForImageClassification,MobileNetV1Model:()=>n.MobileNetV1Model,MobileNetV1PreTrainedModel:()=>n.MobileNetV1PreTrainedModel,MobileNetV2FeatureExtractor:()=>i.MobileNetV2FeatureExtractor,MobileNetV2ForImageClassification:()=>n.MobileNetV2ForImageClassification,MobileNetV2Model:()=>n.MobileNetV2Model,MobileNetV2PreTrainedModel:()=>n.MobileNetV2PreTrainedModel,MobileNetV3FeatureExtractor:()=>i.MobileNetV3FeatureExtractor,MobileNetV3ForImageClassification:()=>n.MobileNetV3ForImageClassification,MobileNetV3Model:()=>n.MobileNetV3Model,MobileNetV3PreTrainedModel:()=>n.MobileNetV3PreTrainedModel,MobileNetV4FeatureExtractor:()=>i.MobileNetV4FeatureExtractor,MobileNetV4ForImageClassification:()=>n.MobileNetV4ForImageClassification,MobileNetV4Model:()=>n.MobileNetV4Model,MobileNetV4PreTrainedModel:()=>n.MobileNetV4PreTrainedModel,MobileViTFeatureExtractor:()=>i.MobileViTFeatureExtractor,MobileViTForImageClassification:()=>n.MobileViTForImageClassification,MobileViTImageProcessor:()=>i.MobileViTImageProcessor,MobileViTModel:()=>n.MobileViTModel,MobileViTPreTrainedModel:()=>n.MobileViTPreTrainedModel,MobileViTV2ForImageClassification:()=>n.MobileViTV2ForImageClassification,MobileViTV2Model:()=>n.MobileViTV2Model,MobileViTV2PreTrainedModel:()=>n.MobileViTV2PreTrainedModel,ModelOutput:()=>n.ModelOutput,Moondream1ForConditionalGeneration:()=>n.Moondream1ForConditionalGeneration,MptForCausalLM:()=>n.MptForCausalLM,MptModel:()=>n.MptModel,MptPreTrainedModel:()=>n.MptPreTrainedModel,MusicgenForCausalLM:()=>n.MusicgenForCausalLM,MusicgenForConditionalGeneration:()=>n.MusicgenForConditionalGeneration,MusicgenModel:()=>n.MusicgenModel,MusicgenPreTrainedModel:()=>n.MusicgenPreTrainedModel,NllbTokenizer:()=>s.NllbTokenizer,NomicBertModel:()=>n.NomicBertModel,NomicBertPreTrainedModel:()=>n.NomicBertPreTrainedModel,NougatImageProcessor:()=>i.NougatImageProcessor,NougatTokenizer:()=>s.NougatTokenizer,OPTForCausalLM:()=>n.OPTForCausalLM,OPTModel:()=>n.OPTModel,OPTPreTrainedModel:()=>n.OPTPreTrainedModel,ObjectDetectionPipeline:()=>t.ObjectDetectionPipeline,OpenELMForCausalLM:()=>n.OpenELMForCausalLM,OpenELMModel:()=>n.OpenELMModel,OpenELMPreTrainedModel:()=>n.OpenELMPreTrainedModel,OwlViTFeatureExtractor:()=>i.OwlViTFeatureExtractor,OwlViTForObjectDetection:()=>n.OwlViTForObjectDetection,OwlViTModel:()=>n.OwlViTModel,OwlViTPreTrainedModel:()=>n.OwlViTPreTrainedModel,OwlViTProcessor:()=>i.OwlViTProcessor,Owlv2ForObjectDetection:()=>n.Owlv2ForObjectDetection,Owlv2ImageProcessor:()=>i.Owlv2ImageProcessor,Owlv2Model:()=>n.Owlv2Model,Owlv2PreTrainedModel:()=>n.Owlv2PreTrainedModel,Phi3ForCausalLM:()=>n.Phi3ForCausalLM,Phi3Model:()=>n.Phi3Model,Phi3PreTrainedModel:()=>n.Phi3PreTrainedModel,PhiForCausalLM:()=>n.PhiForCausalLM,PhiModel:()=>n.PhiModel,PhiPreTrainedModel:()=>n.PhiPreTrainedModel,Pipeline:()=>t.Pipeline,PreTrainedModel:()=>n.PreTrainedModel,PreTrainedTokenizer:()=>s.PreTrainedTokenizer,PretrainedConfig:()=>a.PretrainedConfig,PretrainedMixin:()=>n.PretrainedMixin,Processor:()=>i.Processor,PyAnnoteFeatureExtractor:()=>i.PyAnnoteFeatureExtractor,PyAnnoteForAudioFrameClassification:()=>n.PyAnnoteForAudioFrameClassification,PyAnnoteModel:()=>n.PyAnnoteModel,PyAnnotePreTrainedModel:()=>n.PyAnnotePreTrainedModel,PyAnnoteProcessor:()=>i.PyAnnoteProcessor,QuestionAnsweringModelOutput:()=>n.QuestionAnsweringModelOutput,QuestionAnsweringPipeline:()=>t.QuestionAnsweringPipeline,Qwen2ForCausalLM:()=>n.Qwen2ForCausalLM,Qwen2Model:()=>n.Qwen2Model,Qwen2PreTrainedModel:()=>n.Qwen2PreTrainedModel,Qwen2Tokenizer:()=>s.Qwen2Tokenizer,RTDetrForObjectDetection:()=>n.RTDetrForObjectDetection,RTDetrImageProcessor:()=>i.RTDetrImageProcessor,RTDetrModel:()=>n.RTDetrModel,RTDetrObjectDetectionOutput:()=>n.RTDetrObjectDetectionOutput,RTDetrPreTrainedModel:()=>n.RTDetrPreTrainedModel,RawImage:()=>c.RawImage,ResNetForImageClassification:()=>n.ResNetForImageClassification,ResNetModel:()=>n.ResNetModel,ResNetPreTrainedModel:()=>n.ResNetPreTrainedModel,RoFormerForMaskedLM:()=>n.RoFormerForMaskedLM,RoFormerForQuestionAnswering:()=>n.RoFormerForQuestionAnswering,RoFormerForSequenceClassification:()=>n.RoFormerForSequenceClassification,RoFormerForTokenClassification:()=>n.RoFormerForTokenClassification,RoFormerModel:()=>n.RoFormerModel,RoFormerPreTrainedModel:()=>n.RoFormerPreTrainedModel,RoFormerTokenizer:()=>s.RoFormerTokenizer,RobertaForMaskedLM:()=>n.RobertaForMaskedLM,RobertaForQuestionAnswering:()=>n.RobertaForQuestionAnswering,RobertaForSequenceClassification:()=>n.RobertaForSequenceClassification,RobertaForTokenClassification:()=>n.RobertaForTokenClassification,RobertaModel:()=>n.RobertaModel,RobertaPreTrainedModel:()=>n.RobertaPreTrainedModel,RobertaTokenizer:()=>s.RobertaTokenizer,SamImageProcessor:()=>i.SamImageProcessor,SamImageSegmentationOutput:()=>n.SamImageSegmentationOutput,SamModel:()=>n.SamModel,SamPreTrainedModel:()=>n.SamPreTrainedModel,SamProcessor:()=>i.SamProcessor,SapiensFeatureExtractor:()=>i.SapiensFeatureExtractor,SapiensForDepthEstimation:()=>n.SapiensForDepthEstimation,SapiensForNormalEstimation:()=>n.SapiensForNormalEstimation,SapiensForSemanticSegmentation:()=>n.SapiensForSemanticSegmentation,SapiensPreTrainedModel:()=>n.SapiensPreTrainedModel,SeamlessM4TFeatureExtractor:()=>i.SeamlessM4TFeatureExtractor,SegformerFeatureExtractor:()=>i.SegformerFeatureExtractor,SegformerForImageClassification:()=>n.SegformerForImageClassification,SegformerForSemanticSegmentation:()=>n.SegformerForSemanticSegmentation,SegformerModel:()=>n.SegformerModel,SegformerPreTrainedModel:()=>n.SegformerPreTrainedModel,Seq2SeqLMOutput:()=>n.Seq2SeqLMOutput,SequenceClassifierOutput:()=>n.SequenceClassifierOutput,SiglipImageProcessor:()=>i.SiglipImageProcessor,SiglipModel:()=>n.SiglipModel,SiglipPreTrainedModel:()=>n.SiglipPreTrainedModel,SiglipTextModel:()=>n.SiglipTextModel,SiglipTokenizer:()=>s.SiglipTokenizer,SiglipVisionModel:()=>n.SiglipVisionModel,SpeechT5FeatureExtractor:()=>i.SpeechT5FeatureExtractor,SpeechT5ForSpeechToText:()=>n.SpeechT5ForSpeechToText,SpeechT5ForTextToSpeech:()=>n.SpeechT5ForTextToSpeech,SpeechT5HifiGan:()=>n.SpeechT5HifiGan,SpeechT5Model:()=>n.SpeechT5Model,SpeechT5PreTrainedModel:()=>n.SpeechT5PreTrainedModel,SpeechT5Processor:()=>i.SpeechT5Processor,SpeechT5Tokenizer:()=>s.SpeechT5Tokenizer,SqueezeBertForMaskedLM:()=>n.SqueezeBertForMaskedLM,SqueezeBertForQuestionAnswering:()=>n.SqueezeBertForQuestionAnswering,SqueezeBertForSequenceClassification:()=>n.SqueezeBertForSequenceClassification,SqueezeBertModel:()=>n.SqueezeBertModel,SqueezeBertPreTrainedModel:()=>n.SqueezeBertPreTrainedModel,SqueezeBertTokenizer:()=>s.SqueezeBertTokenizer,StableLmForCausalLM:()=>n.StableLmForCausalLM,StableLmModel:()=>n.StableLmModel,StableLmPreTrainedModel:()=>n.StableLmPreTrainedModel,Starcoder2ForCausalLM:()=>n.Starcoder2ForCausalLM,Starcoder2Model:()=>n.Starcoder2Model,Starcoder2PreTrainedModel:()=>n.Starcoder2PreTrainedModel,StoppingCriteria:()=>h.StoppingCriteria,StoppingCriteriaList:()=>h.StoppingCriteriaList,SummarizationPipeline:()=>t.SummarizationPipeline,Swin2SRForImageSuperResolution:()=>n.Swin2SRForImageSuperResolution,Swin2SRImageProcessor:()=>i.Swin2SRImageProcessor,Swin2SRModel:()=>n.Swin2SRModel,Swin2SRPreTrainedModel:()=>n.Swin2SRPreTrainedModel,SwinForImageClassification:()=>n.SwinForImageClassification,SwinModel:()=>n.SwinModel,SwinPreTrainedModel:()=>n.SwinPreTrainedModel,T5ForConditionalGeneration:()=>n.T5ForConditionalGeneration,T5Model:()=>n.T5Model,T5PreTrainedModel:()=>n.T5PreTrainedModel,T5Tokenizer:()=>s.T5Tokenizer,TableTransformerForObjectDetection:()=>n.TableTransformerForObjectDetection,TableTransformerModel:()=>n.TableTransformerModel,TableTransformerObjectDetectionOutput:()=>n.TableTransformerObjectDetectionOutput,TableTransformerPreTrainedModel:()=>n.TableTransformerPreTrainedModel,Tensor:()=>d.Tensor,Text2TextGenerationPipeline:()=>t.Text2TextGenerationPipeline,TextClassificationPipeline:()=>t.TextClassificationPipeline,TextGenerationPipeline:()=>t.TextGenerationPipeline,TextStreamer:()=>p.TextStreamer,TextToAudioPipeline:()=>t.TextToAudioPipeline,TokenClassificationPipeline:()=>t.TokenClassificationPipeline,TokenClassifierOutput:()=>n.TokenClassifierOutput,TokenizerModel:()=>s.TokenizerModel,TrOCRForCausalLM:()=>n.TrOCRForCausalLM,TrOCRPreTrainedModel:()=>n.TrOCRPreTrainedModel,TranslationPipeline:()=>t.TranslationPipeline,UniSpeechForCTC:()=>n.UniSpeechForCTC,UniSpeechForSequenceClassification:()=>n.UniSpeechForSequenceClassification,UniSpeechModel:()=>n.UniSpeechModel,UniSpeechPreTrainedModel:()=>n.UniSpeechPreTrainedModel,UniSpeechSatForAudioFrameClassification:()=>n.UniSpeechSatForAudioFrameClassification,UniSpeechSatForCTC:()=>n.UniSpeechSatForCTC,UniSpeechSatForSequenceClassification:()=>n.UniSpeechSatForSequenceClassification,UniSpeechSatModel:()=>n.UniSpeechSatModel,UniSpeechSatPreTrainedModel:()=>n.UniSpeechSatPreTrainedModel,ViTFeatureExtractor:()=>i.ViTFeatureExtractor,ViTForImageClassification:()=>n.ViTForImageClassification,ViTImageProcessor:()=>i.ViTImageProcessor,ViTModel:()=>n.ViTModel,ViTPreTrainedModel:()=>n.ViTPreTrainedModel,VisionEncoderDecoderModel:()=>n.VisionEncoderDecoderModel,VitMatteForImageMatting:()=>n.VitMatteForImageMatting,VitMatteImageProcessor:()=>i.VitMatteImageProcessor,VitMattePreTrainedModel:()=>n.VitMattePreTrainedModel,VitsModel:()=>n.VitsModel,VitsModelOutput:()=>n.VitsModelOutput,VitsPreTrainedModel:()=>n.VitsPreTrainedModel,VitsTokenizer:()=>s.VitsTokenizer,Wav2Vec2BertForCTC:()=>n.Wav2Vec2BertForCTC,Wav2Vec2BertForSequenceClassification:()=>n.Wav2Vec2BertForSequenceClassification,Wav2Vec2BertModel:()=>n.Wav2Vec2BertModel,Wav2Vec2BertPreTrainedModel:()=>n.Wav2Vec2BertPreTrainedModel,Wav2Vec2CTCTokenizer:()=>s.Wav2Vec2CTCTokenizer,Wav2Vec2FeatureExtractor:()=>i.Wav2Vec2FeatureExtractor,Wav2Vec2ForAudioFrameClassification:()=>n.Wav2Vec2ForAudioFrameClassification,Wav2Vec2ForCTC:()=>n.Wav2Vec2ForCTC,Wav2Vec2ForSequenceClassification:()=>n.Wav2Vec2ForSequenceClassification,Wav2Vec2Model:()=>n.Wav2Vec2Model,Wav2Vec2PreTrainedModel:()=>n.Wav2Vec2PreTrainedModel,Wav2Vec2ProcessorWithLM:()=>i.Wav2Vec2ProcessorWithLM,WavLMForAudioFrameClassification:()=>n.WavLMForAudioFrameClassification,WavLMForCTC:()=>n.WavLMForCTC,WavLMForSequenceClassification:()=>n.WavLMForSequenceClassification,WavLMForXVector:()=>n.WavLMForXVector,WavLMModel:()=>n.WavLMModel,WavLMPreTrainedModel:()=>n.WavLMPreTrainedModel,WeSpeakerFeatureExtractor:()=>i.WeSpeakerFeatureExtractor,WeSpeakerResNetModel:()=>n.WeSpeakerResNetModel,WeSpeakerResNetPreTrainedModel:()=>n.WeSpeakerResNetPreTrainedModel,WhisperFeatureExtractor:()=>i.WhisperFeatureExtractor,WhisperForConditionalGeneration:()=>n.WhisperForConditionalGeneration,WhisperModel:()=>n.WhisperModel,WhisperPreTrainedModel:()=>n.WhisperPreTrainedModel,WhisperProcessor:()=>i.WhisperProcessor,WhisperTextStreamer:()=>p.WhisperTextStreamer,WhisperTokenizer:()=>s.WhisperTokenizer,XLMForQuestionAnswering:()=>n.XLMForQuestionAnswering,XLMForSequenceClassification:()=>n.XLMForSequenceClassification,XLMForTokenClassification:()=>n.XLMForTokenClassification,XLMModel:()=>n.XLMModel,XLMPreTrainedModel:()=>n.XLMPreTrainedModel,XLMRobertaForMaskedLM:()=>n.XLMRobertaForMaskedLM,XLMRobertaForQuestionAnswering:()=>n.XLMRobertaForQuestionAnswering,XLMRobertaForSequenceClassification:()=>n.XLMRobertaForSequenceClassification,XLMRobertaForTokenClassification:()=>n.XLMRobertaForTokenClassification,XLMRobertaModel:()=>n.XLMRobertaModel,XLMRobertaPreTrainedModel:()=>n.XLMRobertaPreTrainedModel,XLMRobertaTokenizer:()=>s.XLMRobertaTokenizer,XLMTokenizer:()=>s.XLMTokenizer,XLMWithLMHeadModel:()=>n.XLMWithLMHeadModel,XVectorOutput:()=>n.XVectorOutput,YolosFeatureExtractor:()=>i.YolosFeatureExtractor,YolosForObjectDetection:()=>n.YolosForObjectDetection,YolosModel:()=>n.YolosModel,YolosObjectDetectionOutput:()=>n.YolosObjectDetectionOutput,YolosPreTrainedModel:()=>n.YolosPreTrainedModel,ZeroShotAudioClassificationPipeline:()=>t.ZeroShotAudioClassificationPipeline,ZeroShotClassificationPipeline:()=>t.ZeroShotClassificationPipeline,ZeroShotImageClassificationPipeline:()=>t.ZeroShotImageClassificationPipeline,ZeroShotObjectDetectionPipeline:()=>t.ZeroShotObjectDetectionPipeline,bankers_round:()=>u.bankers_round,cat:()=>d.cat,cos_sim:()=>u.cos_sim,dot:()=>u.dot,dynamic_time_warping:()=>u.dynamic_time_warping,env:()=>e.env,full:()=>d.full,full_like:()=>d.full_like,getKeyValueShapes:()=>a.getKeyValueShapes,hamming:()=>l.hamming,hanning:()=>l.hanning,interpolate:()=>d.interpolate,interpolate_4d:()=>d.interpolate_4d,interpolate_data:()=>u.interpolate_data,is_chinese_char:()=>s.is_chinese_char,layer_norm:()=>d.layer_norm,log_softmax:()=>u.log_softmax,magnitude:()=>u.magnitude,matmul:()=>d.matmul,max:()=>u.max,mean:()=>d.mean,mean_pooling:()=>d.mean_pooling,medianFilter:()=>u.medianFilter,mel_filter_bank:()=>l.mel_filter_bank,min:()=>u.min,ones:()=>d.ones,ones_like:()=>d.ones_like,permute:()=>d.permute,permute_data:()=>u.permute_data,pipeline:()=>t.pipeline,quantize_embeddings:()=>d.quantize_embeddings,read_audio:()=>l.read_audio,rfft:()=>d.rfft,round:()=>u.round,softmax:()=>u.softmax,spectrogram:()=>l.spectrogram,stack:()=>d.stack,std_mean:()=>d.std_mean,topk:()=>d.topk,window_function:()=>l.window_function,zeros:()=>d.zeros,zeros_like:()=>d.zeros_like});var e=r(/*! ./env.js */"./src/env.js"),t=r(/*! ./pipelines.js */"./src/pipelines.js"),n=r(/*! ./models.js */"./src/models.js"),s=r(/*! ./tokenizers.js */"./src/tokenizers.js"),i=r(/*! ./processors.js */"./src/processors.js"),a=r(/*! ./configs.js */"./src/configs.js"),l=r(/*! ./utils/audio.js */"./src/utils/audio.js"),c=r(/*! ./utils/image.js */"./src/utils/image.js"),d=r(/*! ./utils/tensor.js */"./src/utils/tensor.js"),u=r(/*! ./utils/maths.js */"./src/utils/maths.js"),p=r(/*! ./generation/streamers.js */"./src/generation/streamers.js"),h=r(/*! ./generation/stopping_criteria.js */"./src/generation/stopping_criteria.js")})();var i=exports;for(var a in o)i[a]=o[a];o.__esModule&&Object.defineProperty(i,"__esModule",{value:!0})})();
 //# sourceMappingURL=transformers.min.cjs.map