web-llm-runner 0.1.14 → 0.1.17

package/lib/index.js

@@ -12814,7 +12814,7 @@ class EnvImpl {
 /**
  * Represent a set of flags as a global singleton.
  */
-const env$2 = new EnvImpl();
+const env$3 = new EnvImpl();
 
 // Copyright (c) Microsoft Corporation. All rights reserved.
 // Licensed under the MIT License.
@@ -13513,7 +13513,7 @@ var lib = /*#__PURE__*/Object.freeze({
 	__proto__: null,
 	InferenceSession: InferenceSession$1,
 	Tensor: Tensor$1,
-	env: env$2,
+	env: env$3,
 	registerBackend: registerBackend
 });
 
@@ -13673,7 +13673,7 @@ if (onnx_env?.wasm) {
  * @property {Object} customCache The custom cache to use. Defaults to `null`. Note: this must be an object which
  * implements the `match` and `put` functions of the Web Cache API. For more information, see https://developer.mozilla.org/en-US/docs/Web/API/Cache
  */
-const env$1 = {
+const env$2 = {
     version: VERSION,
 
     remoteHost: 'https://huggingface.co/',
@@ -13862,12 +13862,12 @@ function isValidUrl(string, protocols = null, validHosts = null) {
  */
 async function getFile(urlOrPath) {
 
-    if (env$1.useFS && !isValidUrl(urlOrPath, ['http:', 'https:', 'blob:'])) {
+    if (env$2.useFS && !isValidUrl(urlOrPath, ['http:', 'https:', 'blob:'])) {
         return new FileResponse(urlOrPath);
 
     } else if (typeof process !== 'undefined' && process?.release?.name === 'node') {
         const IS_CI = !!process.env?.TESTING_REMOTELY;
-        const version = env$1.version;
+        const version = env$2.version;
 
         const headers = new Headers();
         headers.set('User-Agent', `transformers.js/${version}; is_ci/${IS_CI};`);
@@ -14014,7 +14014,7 @@ async function tryCache(cache, ...names) {
  */
 async function getModelFile(path_or_repo_id, filename, fatal = true, options = {}) {
 
-    if (!env$1.allowLocalModels) {
+    if (!env$2.allowLocalModels) {
         // User has disabled local models, so we just make sure other settings are correct.
 
         if (options.local_files_only) {
@@ -14032,7 +14032,7 @@ async function getModelFile(path_or_repo_id, filename, fatal = true, options = {
     // First, check if the a caching backend is available
     // If no caching mechanism available, will download the file every time
     let cache;
-    if (!cache && env$1.useBrowserCache) {
+    if (!cache && env$2.useBrowserCache) {
         if (typeof caches === 'undefined') {
             throw Error('Browser cache is not available in this environment.')
         }
@@ -14048,21 +14048,21 @@ async function getModelFile(path_or_repo_id, filename, fatal = true, options = {
         }
     }
 
-    if (!cache && env$1.useFSCache) {
+    if (!cache && env$2.useFSCache) {
         // TODO throw error if not available
 
         // If `cache_dir` is not specified, use the default cache directory
-        cache = new FileCache(options.cache_dir ?? env$1.cacheDir);
+        cache = new FileCache(options.cache_dir ?? env$2.cacheDir);
     }
 
     const revision = options.revision ?? 'main';
 
     let requestURL = pathJoin(path_or_repo_id, filename);
-    let localPath = pathJoin(env$1.localModelPath, requestURL);
+    let localPath = pathJoin(env$2.localModelPath, requestURL);
 
     let remoteURL = pathJoin(
-        env$1.remoteHost,
-        env$1.remotePathTemplate
+        env$2.remoteHost,
+        env$2.remotePathTemplate
             .replaceAll('{model}', path_or_repo_id)
             .replaceAll('{revision}', encodeURIComponent(revision)),
         filename
@@ -14096,7 +14096,7 @@ async function getModelFile(path_or_repo_id, filename, fatal = true, options = {
     if (response === undefined) {
         // Caching not available, or file is not cached, so we perform the request
 
-        if (env$1.allowLocalModels) {
+        if (env$2.allowLocalModels) {
             // Accessing local models is enabled, so we try to get the file locally.
             // If request is a valid HTTP URL, we skip the local file check. Otherwise, we try to get the file locally.
             const isURL = isValidUrl(requestURL, ['http:', 'https:']);
@@ -23790,7 +23790,7 @@ class BeamSearchSampler extends Sampler {
     }
 }
 
-const { InferenceSession, Tensor: ONNXTensor, env } = ONNX;
+const { InferenceSession, Tensor: ONNXTensor, env: env$1 } = ONNX;
 
 /** @typedef {import('onnxruntime-web').InferenceSession} InferenceSession */
 
@@ -23877,7 +23877,7 @@ function validateInputs(session, inputs) {
         // NOTE: When `env.wasm.proxy is true` the tensor is moved across the Worker
         // boundary, transferring ownership to the worker and invalidating the tensor.
         // So, in this case, we simply sacrifice a clone for it.
-        checkedInputs[inputName] = env.wasm.proxy ? tensor.clone() : tensor;
+        checkedInputs[inputName] = env$1.wasm.proxy ? tensor.clone() : tensor;
     }
     if (missingInputs.length > 0) {
         throw new Error(
@@ -30562,7 +30562,7 @@ class RawImage {
             // Clean up: remove the anchor element from the DOM
             downloadLink.remove();
 
-        } else if (!env$1.useFS) {
+        } else if (!env$2.useFS) {
             throw new Error('Unable to save the image because filesystem is disabled in this environment.')
 
         } else {
@@ -36561,7 +36561,7 @@ const TASK_ALIASES = Object.freeze({
  * @returns {Promise<AllTasks[T]>} A Pipeline object for the specified task.
  * @throws {Error} If an unsupported pipeline is requested.
  */
-async function pipeline(
+async function pipeline$1(
     task,
     model = null,
     {
@@ -36679,6 +36679,511 @@ async function loadItems(mapping, model, pretrainedOptions) {
     return result;
 }
 
+/**
+ * @file Entry point for the Transformers.js library. Only the exports from this file
+ * are available to the end user, and are grouped as follows:
+ * 
+ * 1. [Pipelines](./pipelines)
+ * 2. [Environment variables](./env)
+ * 3. [Models](./models)
+ * 4. [Tokenizers](./tokenizers)
+ * 5. [Processors](./processors)
+ * 
+ * @module transformers
+ */
+
+var Transformers = /*#__PURE__*/Object.freeze({
+	__proto__: null,
+	ASTFeatureExtractor: ASTFeatureExtractor,
+	ASTForAudioClassification: ASTForAudioClassification,
+	ASTModel: ASTModel,
+	ASTPreTrainedModel: ASTPreTrainedModel,
+	AlbertForMaskedLM: AlbertForMaskedLM,
+	AlbertForQuestionAnswering: AlbertForQuestionAnswering,
+	AlbertForSequenceClassification: AlbertForSequenceClassification,
+	AlbertModel: AlbertModel,
+	AlbertPreTrainedModel: AlbertPreTrainedModel,
+	AlbertTokenizer: AlbertTokenizer,
+	AudioClassificationPipeline: AudioClassificationPipeline,
+	AutoConfig: AutoConfig,
+	AutoModel: AutoModel,
+	AutoModelForAudioClassification: AutoModelForAudioClassification,
+	AutoModelForCTC: AutoModelForCTC,
+	AutoModelForCausalLM: AutoModelForCausalLM,
+	AutoModelForDepthEstimation: AutoModelForDepthEstimation,
+	AutoModelForDocumentQuestionAnswering: AutoModelForDocumentQuestionAnswering,
+	AutoModelForImageClassification: AutoModelForImageClassification,
+	AutoModelForImageFeatureExtraction: AutoModelForImageFeatureExtraction,
+	AutoModelForImageSegmentation: AutoModelForImageSegmentation,
+	AutoModelForImageToImage: AutoModelForImageToImage,
+	AutoModelForMaskedLM: AutoModelForMaskedLM,
+	AutoModelForObjectDetection: AutoModelForObjectDetection,
+	AutoModelForQuestionAnswering: AutoModelForQuestionAnswering,
+	AutoModelForSemanticSegmentation: AutoModelForSemanticSegmentation,
+	AutoModelForSeq2SeqLM: AutoModelForSeq2SeqLM,
+	AutoModelForSequenceClassification: AutoModelForSequenceClassification,
+	AutoModelForSpeechSeq2Seq: AutoModelForSpeechSeq2Seq,
+	AutoModelForTextToSpectrogram: AutoModelForTextToSpectrogram,
+	AutoModelForTextToWaveform: AutoModelForTextToWaveform,
+	AutoModelForTokenClassification: AutoModelForTokenClassification,
+	AutoModelForVision2Seq: AutoModelForVision2Seq,
+	AutoModelForZeroShotObjectDetection: AutoModelForZeroShotObjectDetection,
+	AutoProcessor: AutoProcessor,
+	AutoTokenizer: AutoTokenizer,
+	AutomaticSpeechRecognitionPipeline: AutomaticSpeechRecognitionPipeline,
+	BartForConditionalGeneration: BartForConditionalGeneration,
+	BartForSequenceClassification: BartForSequenceClassification,
+	BartModel: BartModel,
+	BartPretrainedModel: BartPretrainedModel,
+	BartTokenizer: BartTokenizer,
+	BeitFeatureExtractor: BeitFeatureExtractor,
+	BeitForImageClassification: BeitForImageClassification,
+	BeitModel: BeitModel,
+	BeitPreTrainedModel: BeitPreTrainedModel,
+	BertForMaskedLM: BertForMaskedLM,
+	BertForQuestionAnswering: BertForQuestionAnswering,
+	BertForSequenceClassification: BertForSequenceClassification,
+	BertForTokenClassification: BertForTokenClassification,
+	BertModel: BertModel,
+	BertPreTrainedModel: BertPreTrainedModel,
+	BertTokenizer: BertTokenizer,
+	BitImageProcessor: BitImageProcessor,
+	BlenderbotForConditionalGeneration: BlenderbotForConditionalGeneration,
+	BlenderbotModel: BlenderbotModel,
+	BlenderbotPreTrainedModel: BlenderbotPreTrainedModel,
+	BlenderbotSmallForConditionalGeneration: BlenderbotSmallForConditionalGeneration,
+	BlenderbotSmallModel: BlenderbotSmallModel,
+	BlenderbotSmallPreTrainedModel: BlenderbotSmallPreTrainedModel,
+	BlenderbotSmallTokenizer: BlenderbotSmallTokenizer,
+	BlenderbotTokenizer: BlenderbotTokenizer,
+	BloomForCausalLM: BloomForCausalLM,
+	BloomModel: BloomModel,
+	BloomPreTrainedModel: BloomPreTrainedModel,
+	BloomTokenizer: BloomTokenizer,
+	CLIPFeatureExtractor: CLIPFeatureExtractor,
+	CLIPModel: CLIPModel,
+	CLIPPreTrainedModel: CLIPPreTrainedModel,
+	CLIPSegForImageSegmentation: CLIPSegForImageSegmentation,
+	CLIPSegModel: CLIPSegModel,
+	CLIPSegPreTrainedModel: CLIPSegPreTrainedModel,
+	CLIPTextModelWithProjection: CLIPTextModelWithProjection,
+	CLIPTokenizer: CLIPTokenizer,
+	CLIPVisionModelWithProjection: CLIPVisionModelWithProjection,
+	CamembertForMaskedLM: CamembertForMaskedLM,
+	CamembertForQuestionAnswering: CamembertForQuestionAnswering,
+	CamembertForSequenceClassification: CamembertForSequenceClassification,
+	CamembertForTokenClassification: CamembertForTokenClassification,
+	CamembertModel: CamembertModel,
+	CamembertPreTrainedModel: CamembertPreTrainedModel,
+	CamembertTokenizer: CamembertTokenizer,
+	CausalLMOutput: CausalLMOutput,
+	ChineseCLIPFeatureExtractor: ChineseCLIPFeatureExtractor,
+	ChineseCLIPModel: ChineseCLIPModel,
+	ChineseCLIPPreTrainedModel: ChineseCLIPPreTrainedModel,
+	ClapAudioModelWithProjection: ClapAudioModelWithProjection,
+	ClapFeatureExtractor: ClapFeatureExtractor,
+	ClapModel: ClapModel,
+	ClapPreTrainedModel: ClapPreTrainedModel,
+	ClapTextModelWithProjection: ClapTextModelWithProjection,
+	CodeGenForCausalLM: CodeGenForCausalLM,
+	CodeGenModel: CodeGenModel,
+	CodeGenPreTrainedModel: CodeGenPreTrainedModel,
+	CodeGenTokenizer: CodeGenTokenizer,
+	CodeLlamaTokenizer: CodeLlamaTokenizer,
+	CohereTokenizer: CohereTokenizer,
+	ConvBertForMaskedLM: ConvBertForMaskedLM,
+	ConvBertForQuestionAnswering: ConvBertForQuestionAnswering,
+	ConvBertForSequenceClassification: ConvBertForSequenceClassification,
+	ConvBertForTokenClassification: ConvBertForTokenClassification,
+	ConvBertModel: ConvBertModel,
+	ConvBertPreTrainedModel: ConvBertPreTrainedModel,
+	ConvBertTokenizer: ConvBertTokenizer,
+	ConvNextFeatureExtractor: ConvNextFeatureExtractor,
+	ConvNextForImageClassification: ConvNextForImageClassification,
+	ConvNextImageProcessor: ConvNextImageProcessor,
+	ConvNextModel: ConvNextModel,
+	ConvNextPreTrainedModel: ConvNextPreTrainedModel,
+	ConvNextV2ForImageClassification: ConvNextV2ForImageClassification,
+	ConvNextV2Model: ConvNextV2Model,
+	ConvNextV2PreTrainedModel: ConvNextV2PreTrainedModel,
+	DPTFeatureExtractor: DPTFeatureExtractor,
+	DPTForDepthEstimation: DPTForDepthEstimation,
+	DPTImageProcessor: DPTImageProcessor,
+	DPTModel: DPTModel,
+	DPTPreTrainedModel: DPTPreTrainedModel,
+	DebertaForMaskedLM: DebertaForMaskedLM,
+	DebertaForQuestionAnswering: DebertaForQuestionAnswering,
+	DebertaForSequenceClassification: DebertaForSequenceClassification,
+	DebertaForTokenClassification: DebertaForTokenClassification,
+	DebertaModel: DebertaModel,
+	DebertaPreTrainedModel: DebertaPreTrainedModel,
+	DebertaTokenizer: DebertaTokenizer,
+	DebertaV2ForMaskedLM: DebertaV2ForMaskedLM,
+	DebertaV2ForQuestionAnswering: DebertaV2ForQuestionAnswering,
+	DebertaV2ForSequenceClassification: DebertaV2ForSequenceClassification,
+	DebertaV2ForTokenClassification: DebertaV2ForTokenClassification,
+	DebertaV2Model: DebertaV2Model,
+	DebertaV2PreTrainedModel: DebertaV2PreTrainedModel,
+	DebertaV2Tokenizer: DebertaV2Tokenizer,
+	DeiTFeatureExtractor: DeiTFeatureExtractor,
+	DeiTForImageClassification: DeiTForImageClassification,
+	DeiTModel: DeiTModel,
+	DeiTPreTrainedModel: DeiTPreTrainedModel,
+	DepthAnythingForDepthEstimation: DepthAnythingForDepthEstimation,
+	DepthAnythingPreTrainedModel: DepthAnythingPreTrainedModel,
+	DepthEstimationPipeline: DepthEstimationPipeline,
+	DetrFeatureExtractor: DetrFeatureExtractor,
+	DetrForObjectDetection: DetrForObjectDetection,
+	DetrForSegmentation: DetrForSegmentation,
+	DetrModel: DetrModel,
+	DetrObjectDetectionOutput: DetrObjectDetectionOutput,
+	DetrPreTrainedModel: DetrPreTrainedModel,
+	DetrSegmentationOutput: DetrSegmentationOutput,
+	Dinov2ForImageClassification: Dinov2ForImageClassification,
+	Dinov2Model: Dinov2Model,
+	Dinov2PreTrainedModel: Dinov2PreTrainedModel,
+	DistilBertForMaskedLM: DistilBertForMaskedLM,
+	DistilBertForQuestionAnswering: DistilBertForQuestionAnswering,
+	DistilBertForSequenceClassification: DistilBertForSequenceClassification,
+	DistilBertForTokenClassification: DistilBertForTokenClassification,
+	DistilBertModel: DistilBertModel,
+	DistilBertPreTrainedModel: DistilBertPreTrainedModel,
+	DistilBertTokenizer: DistilBertTokenizer,
+	DocumentQuestionAnsweringPipeline: DocumentQuestionAnsweringPipeline,
+	DonutFeatureExtractor: DonutFeatureExtractor,
+	DonutSwinModel: DonutSwinModel,
+	DonutSwinPreTrainedModel: DonutSwinPreTrainedModel,
+	EfficientNetForImageClassification: EfficientNetForImageClassification,
+	EfficientNetImageProcessor: EfficientNetImageProcessor,
+	EfficientNetModel: EfficientNetModel,
+	EfficientNetPreTrainedModel: EfficientNetPreTrainedModel,
+	ElectraForMaskedLM: ElectraForMaskedLM,
+	ElectraForQuestionAnswering: ElectraForQuestionAnswering,
+	ElectraForSequenceClassification: ElectraForSequenceClassification,
+	ElectraForTokenClassification: ElectraForTokenClassification,
+	ElectraModel: ElectraModel,
+	ElectraPreTrainedModel: ElectraPreTrainedModel,
+	ElectraTokenizer: ElectraTokenizer,
+	EsmForMaskedLM: EsmForMaskedLM,
+	EsmForSequenceClassification: EsmForSequenceClassification,
+	EsmForTokenClassification: EsmForTokenClassification,
+	EsmModel: EsmModel,
+	EsmPreTrainedModel: EsmPreTrainedModel,
+	EsmTokenizer: EsmTokenizer,
+	FFT: FFT,
+	FalconForCausalLM: FalconForCausalLM,
+	FalconModel: FalconModel,
+	FalconPreTrainedModel: FalconPreTrainedModel,
+	FalconTokenizer: FalconTokenizer,
+	FastViTForImageClassification: FastViTForImageClassification,
+	FastViTModel: FastViTModel,
+	FastViTPreTrainedModel: FastViTPreTrainedModel,
+	FeatureExtractionPipeline: FeatureExtractionPipeline,
+	FeatureExtractor: FeatureExtractor,
+	FillMaskPipeline: FillMaskPipeline,
+	GLPNFeatureExtractor: GLPNFeatureExtractor,
+	GLPNForDepthEstimation: GLPNForDepthEstimation,
+	GLPNModel: GLPNModel,
+	GLPNPreTrainedModel: GLPNPreTrainedModel,
+	GPT2LMHeadModel: GPT2LMHeadModel,
+	GPT2Model: GPT2Model,
+	GPT2PreTrainedModel: GPT2PreTrainedModel,
+	GPT2Tokenizer: GPT2Tokenizer,
+	GPTBigCodeForCausalLM: GPTBigCodeForCausalLM,
+	GPTBigCodeModel: GPTBigCodeModel,
+	GPTBigCodePreTrainedModel: GPTBigCodePreTrainedModel,
+	GPTJForCausalLM: GPTJForCausalLM,
+	GPTJModel: GPTJModel,
+	GPTJPreTrainedModel: GPTJPreTrainedModel,
+	GPTNeoForCausalLM: GPTNeoForCausalLM,
+	GPTNeoModel: GPTNeoModel,
+	GPTNeoPreTrainedModel: GPTNeoPreTrainedModel,
+	GPTNeoXForCausalLM: GPTNeoXForCausalLM,
+	GPTNeoXModel: GPTNeoXModel,
+	GPTNeoXPreTrainedModel: GPTNeoXPreTrainedModel,
+	GPTNeoXTokenizer: GPTNeoXTokenizer,
+	GemmaTokenizer: GemmaTokenizer,
+	Grok1Tokenizer: Grok1Tokenizer,
+	HerbertTokenizer: HerbertTokenizer,
+	HubertForCTC: HubertForCTC,
+	HubertForSequenceClassification: HubertForSequenceClassification,
+	HubertModel: HubertModel,
+	ImageClassificationPipeline: ImageClassificationPipeline,
+	ImageFeatureExtractionPipeline: ImageFeatureExtractionPipeline,
+	ImageFeatureExtractor: ImageFeatureExtractor,
+	ImageMattingOutput: ImageMattingOutput,
+	ImageSegmentationPipeline: ImageSegmentationPipeline,
+	ImageToImagePipeline: ImageToImagePipeline,
+	ImageToTextPipeline: ImageToTextPipeline,
+	LlamaForCausalLM: LlamaForCausalLM,
+	LlamaModel: LlamaModel,
+	LlamaPreTrainedModel: LlamaPreTrainedModel,
+	LlamaTokenizer: LlamaTokenizer,
+	LongT5ForConditionalGeneration: LongT5ForConditionalGeneration,
+	LongT5Model: LongT5Model,
+	LongT5PreTrainedModel: LongT5PreTrainedModel,
+	M2M100ForConditionalGeneration: M2M100ForConditionalGeneration,
+	M2M100Model: M2M100Model,
+	M2M100PreTrainedModel: M2M100PreTrainedModel,
+	M2M100Tokenizer: M2M100Tokenizer,
+	MBart50Tokenizer: MBart50Tokenizer,
+	MBartForCausalLM: MBartForCausalLM,
+	MBartForConditionalGeneration: MBartForConditionalGeneration,
+	MBartForSequenceClassification: MBartForSequenceClassification,
+	MBartModel: MBartModel,
+	MBartPreTrainedModel: MBartPreTrainedModel,
+	MBartTokenizer: MBartTokenizer,
+	MPNetForMaskedLM: MPNetForMaskedLM,
+	MPNetForQuestionAnswering: MPNetForQuestionAnswering,
+	MPNetForSequenceClassification: MPNetForSequenceClassification,
+	MPNetForTokenClassification: MPNetForTokenClassification,
+	MPNetModel: MPNetModel,
+	MPNetPreTrainedModel: MPNetPreTrainedModel,
+	MPNetTokenizer: MPNetTokenizer,
+	MT5ForConditionalGeneration: MT5ForConditionalGeneration,
+	MT5Model: MT5Model,
+	MT5PreTrainedModel: MT5PreTrainedModel,
+	MarianMTModel: MarianMTModel,
+	MarianModel: MarianModel,
+	MarianPreTrainedModel: MarianPreTrainedModel,
+	MarianTokenizer: MarianTokenizer,
+	MaskedLMOutput: MaskedLMOutput,
+	MistralForCausalLM: MistralForCausalLM,
+	MistralModel: MistralModel,
+	MistralPreTrainedModel: MistralPreTrainedModel,
+	MobileBertForMaskedLM: MobileBertForMaskedLM,
+	MobileBertForQuestionAnswering: MobileBertForQuestionAnswering,
+	MobileBertForSequenceClassification: MobileBertForSequenceClassification,
+	MobileBertModel: MobileBertModel,
+	MobileBertPreTrainedModel: MobileBertPreTrainedModel,
+	MobileBertTokenizer: MobileBertTokenizer,
+	MobileViTFeatureExtractor: MobileViTFeatureExtractor,
+	MobileViTForImageClassification: MobileViTForImageClassification,
+	MobileViTImageProcessor: MobileViTImageProcessor,
+	MobileViTModel: MobileViTModel,
+	MobileViTPreTrainedModel: MobileViTPreTrainedModel,
+	MobileViTV2ForImageClassification: MobileViTV2ForImageClassification,
+	MobileViTV2Model: MobileViTV2Model,
+	MobileViTV2PreTrainedModel: MobileViTV2PreTrainedModel,
+	ModelOutput: ModelOutput,
+	MptForCausalLM: MptForCausalLM,
+	MptModel: MptModel,
+	MptPreTrainedModel: MptPreTrainedModel,
+	NllbTokenizer: NllbTokenizer,
+	NomicBertModel: NomicBertModel,
+	NomicBertPreTrainedModel: NomicBertPreTrainedModel,
+	NougatImageProcessor: NougatImageProcessor,
+	NougatTokenizer: NougatTokenizer,
+	OPTForCausalLM: OPTForCausalLM,
+	OPTModel: OPTModel,
+	OPTPreTrainedModel: OPTPreTrainedModel,
+	ObjectDetectionPipeline: ObjectDetectionPipeline,
+	OwlViTFeatureExtractor: OwlViTFeatureExtractor,
+	OwlViTForObjectDetection: OwlViTForObjectDetection,
+	OwlViTModel: OwlViTModel,
+	OwlViTPreTrainedModel: OwlViTPreTrainedModel,
+	OwlViTProcessor: OwlViTProcessor,
+	Owlv2ForObjectDetection: Owlv2ForObjectDetection,
+	Owlv2ImageProcessor: Owlv2ImageProcessor,
+	Owlv2Model: Owlv2Model,
+	Owlv2PreTrainedModel: Owlv2PreTrainedModel,
+	PhiForCausalLM: PhiForCausalLM,
+	PhiModel: PhiModel,
+	PhiPreTrainedModel: PhiPreTrainedModel,
+	Pipeline: Pipeline,
+	PreTrainedModel: PreTrainedModel,
+	PreTrainedTokenizer: PreTrainedTokenizer,
+	PretrainedConfig: PretrainedConfig,
+	PretrainedMixin: PretrainedMixin,
+	Processor: Processor,
+	QuestionAnsweringModelOutput: QuestionAnsweringModelOutput,
+	QuestionAnsweringPipeline: QuestionAnsweringPipeline,
+	Qwen2ForCausalLM: Qwen2ForCausalLM,
+	Qwen2Model: Qwen2Model,
+	Qwen2PreTrainedModel: Qwen2PreTrainedModel,
+	Qwen2Tokenizer: Qwen2Tokenizer,
+	RawImage: RawImage,
+	ResNetForImageClassification: ResNetForImageClassification,
+	ResNetModel: ResNetModel,
+	ResNetPreTrainedModel: ResNetPreTrainedModel,
+	RoFormerForMaskedLM: RoFormerForMaskedLM,
+	RoFormerForQuestionAnswering: RoFormerForQuestionAnswering,
+	RoFormerForSequenceClassification: RoFormerForSequenceClassification,
+	RoFormerForTokenClassification: RoFormerForTokenClassification,
+	RoFormerModel: RoFormerModel,
+	RoFormerPreTrainedModel: RoFormerPreTrainedModel,
+	RoFormerTokenizer: RoFormerTokenizer,
+	RobertaForMaskedLM: RobertaForMaskedLM,
+	RobertaForQuestionAnswering: RobertaForQuestionAnswering,
+	RobertaForSequenceClassification: RobertaForSequenceClassification,
+	RobertaForTokenClassification: RobertaForTokenClassification,
+	RobertaModel: RobertaModel,
+	RobertaPreTrainedModel: RobertaPreTrainedModel,
+	RobertaTokenizer: RobertaTokenizer,
+	SamImageProcessor: SamImageProcessor,
+	SamImageSegmentationOutput: SamImageSegmentationOutput,
+	SamModel: SamModel,
+	SamPreTrainedModel: SamPreTrainedModel,
+	SamProcessor: SamProcessor,
+	SeamlessM4TFeatureExtractor: SeamlessM4TFeatureExtractor,
+	SegformerFeatureExtractor: SegformerFeatureExtractor,
+	SegformerForImageClassification: SegformerForImageClassification,
+	SegformerForSemanticSegmentation: SegformerForSemanticSegmentation,
+	SegformerPreTrainedModel: SegformerPreTrainedModel,
+	Seq2SeqLMOutput: Seq2SeqLMOutput,
+	SequenceClassifierOutput: SequenceClassifierOutput,
+	SiglipImageProcessor: SiglipImageProcessor,
+	SiglipModel: SiglipModel,
+	SiglipPreTrainedModel: SiglipPreTrainedModel,
+	SiglipTextModel: SiglipTextModel,
+	SiglipTokenizer: SiglipTokenizer,
+	SiglipVisionModel: SiglipVisionModel,
+	SpeechT5FeatureExtractor: SpeechT5FeatureExtractor,
+	SpeechT5ForSpeechToText: SpeechT5ForSpeechToText,
+	SpeechT5ForTextToSpeech: SpeechT5ForTextToSpeech,
+	SpeechT5HifiGan: SpeechT5HifiGan,
+	SpeechT5PreTrainedModel: SpeechT5PreTrainedModel,
+	SpeechT5Processor: SpeechT5Processor,
+	SpeechT5Tokenizer: SpeechT5Tokenizer,
+	SqueezeBertForMaskedLM: SqueezeBertForMaskedLM,
+	SqueezeBertForQuestionAnswering: SqueezeBertForQuestionAnswering,
+	SqueezeBertForSequenceClassification: SqueezeBertForSequenceClassification,
+	SqueezeBertModel: SqueezeBertModel,
+	SqueezeBertPreTrainedModel: SqueezeBertPreTrainedModel,
+	SqueezeBertTokenizer: SqueezeBertTokenizer,
+	StableLmForCausalLM: StableLmForCausalLM,
+	StableLmPreTrainedModel: StableLmPreTrainedModel,
+	Starcoder2ForCausalLM: Starcoder2ForCausalLM,
+	Starcoder2Model: Starcoder2Model,
+	Starcoder2PreTrainedModel: Starcoder2PreTrainedModel,
+	SummarizationPipeline: SummarizationPipeline,
+	Swin2SRForImageSuperResolution: Swin2SRForImageSuperResolution,
+	Swin2SRImageProcessor: Swin2SRImageProcessor,
+	Swin2SRModel: Swin2SRModel,
+	Swin2SRPreTrainedModel: Swin2SRPreTrainedModel,
+	SwinForImageClassification: SwinForImageClassification,
+	SwinModel: SwinModel,
+	SwinPreTrainedModel: SwinPreTrainedModel,
+	T5ForConditionalGeneration: T5ForConditionalGeneration,
+	T5Model: T5Model,
+	T5PreTrainedModel: T5PreTrainedModel,
+	T5Tokenizer: T5Tokenizer,
+	TableTransformerForObjectDetection: TableTransformerForObjectDetection,
+	TableTransformerModel: TableTransformerModel,
+	TableTransformerObjectDetectionOutput: TableTransformerObjectDetectionOutput,
+	TableTransformerPreTrainedModel: TableTransformerPreTrainedModel,
+	Tensor: Tensor,
+	Text2TextGenerationPipeline: Text2TextGenerationPipeline,
+	TextClassificationPipeline: TextClassificationPipeline,
+	TextGenerationPipeline: TextGenerationPipeline,
+	TextToAudioPipeline: TextToAudioPipeline,
+	TokenClassificationPipeline: TokenClassificationPipeline,
+	TokenClassifierOutput: TokenClassifierOutput,
+	TokenizerModel: TokenizerModel,
+	TrOCRForCausalLM: TrOCRForCausalLM,
+	TrOCRPreTrainedModel: TrOCRPreTrainedModel,
+	TranslationPipeline: TranslationPipeline,
+	UniSpeechForCTC: UniSpeechForCTC,
+	UniSpeechForSequenceClassification: UniSpeechForSequenceClassification,
+	UniSpeechModel: UniSpeechModel,
+	UniSpeechPreTrainedModel: UniSpeechPreTrainedModel,
+	UniSpeechSatForAudioFrameClassification: UniSpeechSatForAudioFrameClassification,
+	UniSpeechSatForCTC: UniSpeechSatForCTC,
+	UniSpeechSatForSequenceClassification: UniSpeechSatForSequenceClassification,
+	UniSpeechSatModel: UniSpeechSatModel,
+	UniSpeechSatPreTrainedModel: UniSpeechSatPreTrainedModel,
+	ViTFeatureExtractor: ViTFeatureExtractor,
+	ViTForImageClassification: ViTForImageClassification,
+	ViTImageProcessor: ViTImageProcessor,
+	ViTModel: ViTModel,
+	ViTPreTrainedModel: ViTPreTrainedModel,
+	VisionEncoderDecoderModel: VisionEncoderDecoderModel,
+	VitMatteForImageMatting: VitMatteForImageMatting,
+	VitMatteImageProcessor: VitMatteImageProcessor,
+	VitMattePreTrainedModel: VitMattePreTrainedModel,
+	VitsModel: VitsModel,
+	VitsModelOutput: VitsModelOutput,
+	VitsPreTrainedModel: VitsPreTrainedModel,
+	VitsTokenizer: VitsTokenizer,
+	Wav2Vec2BertForCTC: Wav2Vec2BertForCTC,
+	Wav2Vec2BertForSequenceClassification: Wav2Vec2BertForSequenceClassification,
+	Wav2Vec2BertModel: Wav2Vec2BertModel,
+	Wav2Vec2BertPreTrainedModel: Wav2Vec2BertPreTrainedModel,
+	Wav2Vec2CTCTokenizer: Wav2Vec2CTCTokenizer,
+	Wav2Vec2FeatureExtractor: Wav2Vec2FeatureExtractor,
+	Wav2Vec2ForAudioFrameClassification: Wav2Vec2ForAudioFrameClassification,
+	Wav2Vec2ForCTC: Wav2Vec2ForCTC,
+	Wav2Vec2ForSequenceClassification: Wav2Vec2ForSequenceClassification,
+	Wav2Vec2Model: Wav2Vec2Model,
+	Wav2Vec2PreTrainedModel: Wav2Vec2PreTrainedModel,
+	Wav2Vec2ProcessorWithLM: Wav2Vec2ProcessorWithLM,
+	WavLMForAudioFrameClassification: WavLMForAudioFrameClassification,
+	WavLMForCTC: WavLMForCTC,
+	WavLMForSequenceClassification: WavLMForSequenceClassification,
+	WavLMForXVector: WavLMForXVector,
+	WavLMModel: WavLMModel,
+	WavLMPreTrainedModel: WavLMPreTrainedModel,
+	WhisperFeatureExtractor: WhisperFeatureExtractor,
+	WhisperForConditionalGeneration: WhisperForConditionalGeneration,
+	WhisperModel: WhisperModel,
+	WhisperPreTrainedModel: WhisperPreTrainedModel,
+	WhisperProcessor: WhisperProcessor,
+	WhisperTokenizer: WhisperTokenizer,
+	XLMForQuestionAnswering: XLMForQuestionAnswering,
+	XLMForSequenceClassification: XLMForSequenceClassification,
+	XLMForTokenClassification: XLMForTokenClassification,
+	XLMModel: XLMModel,
+	XLMPreTrainedModel: XLMPreTrainedModel,
+	XLMRobertaForMaskedLM: XLMRobertaForMaskedLM,
+	XLMRobertaForQuestionAnswering: XLMRobertaForQuestionAnswering,
+	XLMRobertaForSequenceClassification: XLMRobertaForSequenceClassification,
+	XLMRobertaForTokenClassification: XLMRobertaForTokenClassification,
+	XLMRobertaModel: XLMRobertaModel,
+	XLMRobertaPreTrainedModel: XLMRobertaPreTrainedModel,
+	XLMRobertaTokenizer: XLMRobertaTokenizer,
+	XLMTokenizer: XLMTokenizer,
+	XLMWithLMHeadModel: XLMWithLMHeadModel,
+	XVectorOutput: XVectorOutput,
+	YolosFeatureExtractor: YolosFeatureExtractor,
+	YolosForObjectDetection: YolosForObjectDetection,
+	YolosModel: YolosModel,
+	YolosObjectDetectionOutput: YolosObjectDetectionOutput,
+	YolosPreTrainedModel: YolosPreTrainedModel,
+	ZeroShotAudioClassificationPipeline: ZeroShotAudioClassificationPipeline,
+	ZeroShotClassificationPipeline: ZeroShotClassificationPipeline,
+	ZeroShotImageClassificationPipeline: ZeroShotImageClassificationPipeline,
+	ZeroShotObjectDetectionPipeline: ZeroShotObjectDetectionPipeline,
+	bankers_round: bankers_round,
+	cat: cat,
+	dynamicTimeWarping: dynamicTimeWarping,
+	env: env$2,
+	getTopItems: getTopItems,
+	hanning: hanning,
+	interpolate: interpolate,
+	interpolate_data: interpolate_data,
+	log_softmax: log_softmax,
+	max: max,
+	mean: mean,
+	mean_pooling: mean_pooling,
+	medianFilter: medianFilter,
+	mel_filter_bank: mel_filter_bank,
+	min: min,
+	ones: ones,
+	ones_like: ones_like,
+	permute: permute,
+	permute_data: permute_data,
+	pipeline: pipeline$1,
+	quantize_embeddings: quantize_embeddings,
+	read_audio: read_audio,
+	round: round,
+	softmax: softmax,
+	spectrogram: spectrogram,
+	stack: stack,
+	std_mean: std_mean,
+	window_function: window_function
+});
+
+const { pipeline, env } = Transformers;
 /**
  * ONNXEngine implements MLCEngineInterface using Transformers.js (ONNX Runtime Web).
  * This provides a CPU/WASM fallback for browsers that do not support WebGPU.
@@ -36688,6 +37193,7 @@ class ONNXEngine {
     modelId = null;
     appConfig;
     initProgressCallback;
+    repoId = null;
     // APIs
     chat;
     completions;
@@ -36698,8 +37204,8 @@ class ONNXEngine {
         this.embeddings = new Embeddings(this);
         this.appConfig = prebuiltAppConfig;
         // Default env settings for browser
-        env$1.allowLocalModels = false;
-        env$1.useBrowserCache = true;
+        env.allowLocalModels = false;
+        env.useBrowserCache = true;
     }
     setInitProgressCallback(initProgressCallback) {
         this.initProgressCallback = initProgressCallback;
@@ -36719,6 +37225,7 @@ class ONNXEngine {
             const { findModelRecord } = await Promise.resolve().then(function () { return support; });
             const record = findModelRecord(id, this.appConfig);
             repoId = record.onnx_id || id;
+            this.repoId = repoId;
         }
         catch (e) {
             log.warn(`Model record not found for ${id}, using raw ID for ONNX.`);
@@ -36739,8 +37246,9 @@ class ONNXEngine {
         }
         try {
             // For T5 models, text2text-generation is the standard task in transformers.js
-            const task = repoId.toLowerCase().includes("t5") ? "text2text-generation" : "text-generation";
-            this.generator = await pipeline(task, repoId, {
+            const currentRepoId = this.repoId || id;
+            const task = currentRepoId.toLowerCase().includes("t5") ? "text2text-generation" : "text-generation";
+            this.generator = await pipeline(task, currentRepoId, {
                 progress_callback: (p) => {
                     if (this.initProgressCallback && (p.status === 'progress' || p.status === 'downloading')) {
                         const pctValue = (typeof p.progress === 'number') ? p.progress : 0;
@@ -36812,28 +37320,138 @@ class ONNXEngine {
         };
     }
     async *asyncGenerateStreaming(prompt, request) {
-        // Current simple implementation yields only a single chunk.
-        // In future iterations, we can integrate the Transformers.js TextStreamer
-        const result = await this.generateNonStreaming(prompt, request);
-        const content = result.choices[0].message.content;
+        if (!this.generator)
+            throw new Error("ONNX model not loaded.");
+        const model = this.modelId;
+        const created = Math.floor(Date.now() / 1000);
+        const id = crypto.randomUUID();
+        const queue = [];
+        let isDone = false;
+        const streamer = new undefined(this.generator.tokenizer, {
+            skip_prompt: true,
+            callback_function: (text) => {
+                queue.push(text);
+            },
+        });
+        // Run generation in the background
+        (this.repoId || "").toLowerCase().includes("t5") ? "text2text-generation" : "text-generation";
+        this.generator(prompt, {
+            max_new_tokens: request.max_tokens || 256,
+            temperature: request.temperature || 0.7,
+            top_p: request.top_p || 1.0,
+            do_sample: (request.temperature ?? 1.0) > 0,
+            repetition_penalty: request.repetition_penalty || 1.1,
+            streamer,
+        }).finally(() => {
+            isDone = true;
+        });
+        while (!isDone || queue.length > 0) {
+            if (queue.length > 0) {
+                const content = queue.shift();
+                yield {
+                    id,
+                    choices: [{
+                            delta: { content },
+                            finish_reason: null,
+                            index: 0,
+                        }],
+                    model,
+                    object: 'chat.completion.chunk',
+                    created,
+                };
+            }
+            else {
+                await new Promise(r => setTimeout(r, 10));
+            }
+        }
         yield {
-            id: result.id,
+            id,
             choices: [{
-                    delta: { role: 'assistant', content: content },
+                    delta: {},
                     finish_reason: 'stop',
                     index: 0,
-                    logprobs: null
                 }],
-            model: result.model,
+            model,
             object: 'chat.completion.chunk',
-            created: result.created
+            created,
         };
     }
-    async completion(_request) {
-        throw new Error("Generic completion not yet implemented in ONNXEngine fallback.");
+    async completion(request) {
+        if (!this.generator)
+            throw new Error("ONNX model not loaded.");
+        const prompt = typeof request.prompt === 'string' ? request.prompt : (Array.isArray(request.prompt) ? request.prompt[0] : "");
+        if (request.stream) {
+            return this.asyncGenerateStreamingCompletion(prompt, request);
+        }
+        else {
+            const result = await this.generator(prompt, {
+                max_new_tokens: request.max_tokens || 256,
+                temperature: request.temperature || 0.7,
+                top_p: request.top_p || 1.0,
+                do_sample: (request.temperature ?? 1.0) > 0,
+                repetition_penalty: request.repetition_penalty || 1.1,
+            });
+            const fullText = result[0].generated_text;
+            const text = fullText.startsWith(prompt) ? fullText.slice(prompt.length) : fullText;
+            return {
+                id: crypto.randomUUID(),
+                choices: [{
+                        text,
+                        finish_reason: 'stop',
+                        index: 0,
+                        logprobs: null
+                    }],
+                model: this.modelId,
+                object: 'text_completion',
+                created: Math.floor(Date.now() / 1000),
+                usage: { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0 }
+            };
+        }
+    }
+    async *asyncGenerateStreamingCompletion(prompt, request) {
+        const id = crypto.randomUUID();
+        const created = Math.floor(Date.now() / 1000);
+        const model = this.modelId;
+        const queue = [];
+        let isDone = false;
+        const streamer = new undefined(this.generator.tokenizer, {
+            skip_prompt: true,
+            callback_function: (text) => { queue.push(text); },
+        });
+        this.generator(prompt, {
+            max_new_tokens: request.max_tokens || 256,
+            temperature: request.temperature || 0.7,
+            streamer,
+        }).finally(() => { isDone = true; });
+        while (!isDone || queue.length > 0) {
+            if (queue.length > 0) {
+                yield {
+                    id,
+                    choices: [{ text: queue.shift(), finish_reason: null, index: 0 }],
+                    model,
+                    object: 'text_completion',
+                    created,
+                };
+            }
+            else {
+                await new Promise(r => setTimeout(r, 10));
+            }
+        }
     }
-    async embedding(_request) {
-        throw new Error("Embeddings not yet implemented in ONNXEngine fallback.");
+    async embedding(request) {
+        const input = Array.isArray(request.input) ? request.input : [request.input];
+        const extractor = await pipeline('feature-extraction', this.modelId);
+        const results = await Promise.all(input.map(text => extractor(text, { pooling: 'mean', normalize: true })));
+        return {
+            object: 'list',
+            data: results.map((res, i) => ({
+                object: 'embedding',
+                index: i,
+                embedding: Array.from(res.data)
+            })),
+            model: this.modelId,
+            usage: { prompt_tokens: 0, total_tokens: 0, extra: {} }
+        };
     }
     async runtimeStatsText() {
         return "Backend: ONNX Runtime (WASM/CPU Falback)";
@@ -38034,7 +38652,28 @@ class WebLLM {
         return list.map((m) => m.model_id);
     }
     async local_model_available(model_id) {
-        return await hasModelInCache(model_id);
+        const isMLCCached = await hasModelInCache(model_id);
+        if (isMLCCached)
+            return true;
+        // Check ONNX cache fallback
+        const record = prebuiltAppConfig.model_list.find(m => m.model_id === model_id);
+        if (record && record.onnx_id) {
+            return await this.hasONNXInCache(record.onnx_id);
+        }
+        return false;
+    }
+    async hasONNXInCache(onnx_id) {
+        if (typeof caches === 'undefined')
+            return false;
+        try {
+            const cache = await caches.open('transformers-cache');
+            const url = `https://huggingface.co/${onnx_id}/resolve/main/config.json`;
+            const match = await cache.match(url);
+            return !!match;
+        }
+        catch (e) {
+            return false;
+        }
     }
     async download_model(model_id, progressCallback) {
         // Initial feedback
@@ -38055,6 +38694,14 @@ class WebLLM {
         return this.downloadProgress[model_id] || "No progress available.";
     }
     async delete_model(model_id) {
+        const record = prebuiltAppConfig.model_list.find(m => m.model_id === model_id);
+        if (record && record.onnx_id) {
+            // For ONNX, we currently clear the whole transformers-cache for simplicity
+            // as individual file deletion is complex without a full manifest.
+            if (typeof caches !== 'undefined') {
+                await caches.delete('transformers-cache');
+            }
+        }
         await deleteModelAllInfoInCache(model_id);
     }
     // chat endpoints (Stateful)